From 2cee0669299cd2b980aa9ae253c24107a4813b26 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Thu, 26 Sep 2019 06:09:32 +0000 Subject: enable ring0 to support arm64 This patch enabled the basic framework for arm64 guest. Serveral jobs were finished in this patch: 1, ring0.Vectors() 2, switchToUser() 3, basic framwork for Arm64 guest. Signed-off-by: Bin Lu --- pkg/sentry/platform/ring0/BUILD | 33 +- pkg/sentry/platform/ring0/aarch64.go | 109 ++++++ pkg/sentry/platform/ring0/defs.go | 11 - pkg/sentry/platform/ring0/defs_amd64.go | 11 + pkg/sentry/platform/ring0/defs_arm64.go | 133 +++++++ pkg/sentry/platform/ring0/entry_arm64.go | 60 +++ pkg/sentry/platform/ring0/entry_arm64.s | 565 +++++++++++++++++++++++++++++ pkg/sentry/platform/ring0/kernel_arm64.go | 58 +++ pkg/sentry/platform/ring0/lib_arm64.go | 25 ++ pkg/sentry/platform/ring0/offsets_arm64.go | 124 +++++++ 10 files changed, 1105 insertions(+), 24 deletions(-) create mode 100644 pkg/sentry/platform/ring0/aarch64.go create mode 100644 pkg/sentry/platform/ring0/defs_arm64.go create mode 100644 pkg/sentry/platform/ring0/entry_arm64.go create mode 100644 pkg/sentry/platform/ring0/entry_arm64.s create mode 100644 pkg/sentry/platform/ring0/kernel_arm64.go create mode 100644 pkg/sentry/platform/ring0/lib_arm64.go create mode 100644 pkg/sentry/platform/ring0/offsets_arm64.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/ring0/BUILD b/pkg/sentry/platform/ring0/BUILD index 8ed6c7652..939a0033a 100644 --- a/pkg/sentry/platform/ring0/BUILD +++ b/pkg/sentry/platform/ring0/BUILD @@ -1,17 +1,16 @@ load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools/go_generics:defs.bzl", "go_template", "go_template_instance") package(licenses = ["notice"]) -load("//tools/go_generics:defs.bzl", "go_template", "go_template_instance") - go_template( name = "defs", - srcs = [ - "defs.go", - "defs_amd64.go", - "offsets_amd64.go", - "x86.go", - ], + srcs = select( + { + "@bazel_tools//src/conditions:linux_aarch64": ["defs.go", "defs_arm64.go", "offsets_arm64.go", "aarch64.go",], + "//conditions:default": ["defs.go", "defs_amd64.go", "offsets_amd64.go", "x86.go",], + }, + ), visibility = [":__subpackages__"], ) @@ -23,10 +22,15 @@ go_template_instance( ) genrule( - name = "entry_impl_amd64", - srcs = ["entry_amd64.s"], - outs = ["entry_impl_amd64.s"], - cmd = "(echo -e '// build +amd64\\n' && $(location //pkg/sentry/platform/ring0/gen_offsets) && cat $(SRCS)) > $@", + name = "entry_impl", + srcs = ["entry_amd64.s", "entry_arm64.s"], + outs = ["entry_impl.s"], + cmd = select( + { + "@bazel_tools//src/conditions:linux_aarch64": "(echo -e '// build +arm64\\n' && $(location //pkg/sentry/platform/ring0/gen_offsets) && cat $(location entry_arm64.s)) > $@", + "//conditions:default": "(echo -e '// build +amd64\\n' && $(location //pkg/sentry/platform/ring0/gen_offsets) && cat $(location entry_amd64.s)) > $@", + }, + ), tools = ["//pkg/sentry/platform/ring0/gen_offsets"], ) @@ -35,12 +39,15 @@ go_library( srcs = [ "defs_impl.go", "entry_amd64.go", - "entry_impl_amd64.s", + "entry_arm64.go", + "entry_impl.s", "kernel.go", "kernel_amd64.go", + "kernel_arm64.go", "kernel_unsafe.go", "lib_amd64.go", "lib_amd64.s", + "lib_arm64.go", "ring0.go", ], importpath = "gvisor.dev/gvisor/pkg/sentry/platform/ring0", diff --git a/pkg/sentry/platform/ring0/aarch64.go b/pkg/sentry/platform/ring0/aarch64.go new file mode 100644 index 000000000..6b078cd1e --- /dev/null +++ b/pkg/sentry/platform/ring0/aarch64.go @@ -0,0 +1,109 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package ring0 + +// Useful bits. +const ( + _PGD_PGT_BASE = 0x1000 + _PGD_PGT_SIZE = 0x1000 + _PUD_PGT_BASE = 0x2000 + _PUD_PGT_SIZE = 0x1000 + _PMD_PGT_BASE = 0x3000 + _PMD_PGT_SIZE = 0x4000 + _PTE_PGT_BASE = 0x7000 + _PTE_PGT_SIZE = 0x1000 + + _PSR_MODE_EL0t = 0x0 + _PSR_MODE_EL1t = 0x4 + _PSR_MODE_EL1h = 0x5 + _PSR_EL_MASK = 0xf + + _PSR_D_BIT = 0x200 + _PSR_A_BIT = 0x100 + _PSR_I_BIT = 0x80 + _PSR_F_BIT = 0x40 +) + +const ( + // KernelFlagsSet should always be set in the kernel. + KernelFlagsSet = _PSR_MODE_EL1h + + // UserFlagsSet are always set in userspace. + UserFlagsSet = _PSR_MODE_EL0t + + KernelFlagsClear = _PSR_EL_MASK + UserFlagsClear = _PSR_EL_MASK + + PsrDefaultSet = _PSR_D_BIT | _PSR_A_BIT | _PSR_I_BIT | _PSR_F_BIT +) + +// Vector is an exception vector. +type Vector uintptr + +// Exception vectors. +const ( + El1SyncInvalid = iota + El1IrqInvalid + El1FiqInvalid + El1ErrorInvalid + El1Sync + El1Irq + El1Fiq + El1Error + El0Sync + El0Irq + El0Fiq + El0Error + El0Sync_invalid + El0Irq_invalid + El0Fiq_invalid + El0Error_invalid + El1Sync_da + El1Sync_ia + El1Sync_sp_pc + El1Sync_undef + El1Sync_dbg + El1Sync_inv + El0Sync_svc + El0Sync_da + El0Sync_ia + El0Sync_fpsimd_acc + El0Sync_sve_acc + El0Sync_sys + El0Sync_sp_pc + El0Sync_undef + El0Sync_dbg + El0Sync_inv + VirtualizationException + _NR_INTERRUPTS +) + +// System call vectors. +const ( + Syscall Vector = El0Sync_svc + PageFault Vector = El0Sync_da +) + +// VirtualAddressBits returns the number bits available for virtual addresses. +func VirtualAddressBits() uint32 { + return 48 +} + +// PhysicalAddressBits returns the number of bits available for physical addresses. +func PhysicalAddressBits() uint32 { + return 40 +} diff --git a/pkg/sentry/platform/ring0/defs.go b/pkg/sentry/platform/ring0/defs.go index 076063f85..3f094c2a7 100644 --- a/pkg/sentry/platform/ring0/defs.go +++ b/pkg/sentry/platform/ring0/defs.go @@ -20,17 +20,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/usermem" ) -var ( - // UserspaceSize is the total size of userspace. - UserspaceSize = uintptr(1) << (VirtualAddressBits() - 1) - - // MaximumUserAddress is the largest possible user address. - MaximumUserAddress = (UserspaceSize - 1) & ^uintptr(usermem.PageSize-1) - - // KernelStartAddress is the starting kernel address. - KernelStartAddress = ^uintptr(0) - (UserspaceSize - 1) -) - // Kernel is a global kernel object. // // This contains global state, shared by multiple CPUs. diff --git a/pkg/sentry/platform/ring0/defs_amd64.go b/pkg/sentry/platform/ring0/defs_amd64.go index 7206322b1..10dbd381f 100644 --- a/pkg/sentry/platform/ring0/defs_amd64.go +++ b/pkg/sentry/platform/ring0/defs_amd64.go @@ -20,6 +20,17 @@ import ( "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" ) +var ( + // UserspaceSize is the total size of userspace. + UserspaceSize = uintptr(1) << (VirtualAddressBits() - 1) + + // MaximumUserAddress is the largest possible user address. + MaximumUserAddress = (UserspaceSize - 1) & ^uintptr(usermem.PageSize-1) + + // KernelStartAddress is the starting kernel address. + KernelStartAddress = ^uintptr(0) - (UserspaceSize - 1) +) + // Segment indices and Selectors. const ( // Index into GDT array. diff --git a/pkg/sentry/platform/ring0/defs_arm64.go b/pkg/sentry/platform/ring0/defs_arm64.go new file mode 100644 index 000000000..fbfbd9bab --- /dev/null +++ b/pkg/sentry/platform/ring0/defs_arm64.go @@ -0,0 +1,133 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package ring0 + +import ( + "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" +) + +var ( + // UserspaceSize is the total size of userspace. + UserspaceSize = uintptr(1) << (VirtualAddressBits()) + + // MaximumUserAddress is the largest possible user address. + MaximumUserAddress = (UserspaceSize - 1) & ^uintptr(usermem.PageSize-1) + + // KernelStartAddress is the starting kernel address. + KernelStartAddress = ^uintptr(0) - (UserspaceSize - 1) +) + +// KernelOpts has initialization options for the kernel. +type KernelOpts struct { + // PageTables are the kernel pagetables; this must be provided. + PageTables *pagetables.PageTables +} + +// KernelArchState contains architecture-specific state. +type KernelArchState struct { + KernelOpts +} + +// CPUArchState contains CPU-specific arch state. +type CPUArchState struct { + // stack is the stack used for interrupts on this CPU. + stack [512]byte + + // errorCode is the error code from the last exception. + errorCode uintptr + + // errorType indicates the type of error code here, it is always set + // along with the errorCode value above. + // + // It will either by 1, which indicates a user error, or 0 indicating a + // kernel error. If the error code below returns false (kernel error), + // then it cannot provide relevant information about the last + // exception. + errorType uintptr + + // faultAddr is the value of far_el1. + faultAddr uintptr + + // ttbr0Kvm is the value of ttbr0_el1 for sentry. + ttbr0Kvm uintptr + + // ttbr0App is the value of ttbr0_el1 for applicaton. + ttbr0App uintptr + + // exception vector. + vecCode Vector + + // application context pointer. + appAddr uintptr +} + +// ErrorCode returns the last error code. +// +// The returned boolean indicates whether the error code corresponds to the +// last user error or not. If it does not, then fault information must be +// ignored. This is generally the result of a kernel fault while servicing a +// user fault. +// +//go:nosplit +func (c *CPU) ErrorCode() (value uintptr, user bool) { + return c.errorCode, c.errorType != 0 +} + +// ClearErrorCode resets the error code. +// +//go:nosplit +func (c *CPU) ClearErrorCode() { + c.errorCode = 0 // No code. + c.errorType = 1 // User mode. +} + +//go:nosplit +func (c *CPU) GetFaultAddr() (value uintptr) { + return c.faultAddr +} + +//go:nosplit +func (c *CPU) SetTtbr0Kvm(value uintptr) { + c.ttbr0Kvm = value +} + +//go:nosplit +func (c *CPU) SetTtbr0App(value uintptr) { + c.ttbr0App = value +} + +//go:nosplit +func (c *CPU) GetVector() (value Vector) { + return c.vecCode +} + +//go:nosplit +func (c *CPU) SetAppAddr(value uintptr) { + c.appAddr = value +} + +// SwitchArchOpts are embedded in SwitchOpts. +type SwitchArchOpts struct { + // UserASID indicates that the application ASID to be used on switch, + UserASID uint16 + + // KernelASID indicates that the kernel ASID to be used on return, + KernelASID uint16 +} + +func init() { +} diff --git a/pkg/sentry/platform/ring0/entry_arm64.go b/pkg/sentry/platform/ring0/entry_arm64.go new file mode 100644 index 000000000..0dfa42c36 --- /dev/null +++ b/pkg/sentry/platform/ring0/entry_arm64.go @@ -0,0 +1,60 @@ +// Copyright 2019 Google Inc. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package ring0 + +// This is an assembly function. +// +// The sysenter function is invoked in two situations: +// +// (1) The guest kernel has executed a system call. +// (2) The guest application has executed a system call. +// +// The interrupt flag is examined to determine whether the system call was +// executed from kernel mode or not and the appropriate stub is called. + +func El1_sync_invalid() +func El1_irq_invalid() +func El1_fiq_invalid() +func El1_error_invalid() + +func El1_sync() +func El1_irq() +func El1_fiq() +func El1_error() + +func El0_sync() +func El0_irq() +func El0_fiq() +func El0_error() + +func El0_sync_invalid() +func El0_irq_invalid() +func El0_fiq_invalid() +func El0_error_invalid() + +func Vectors() + +// Start is the CPU entrypoint. +// +// The CPU state will be set to c.Registers(). +func Start() +func kernelExitToEl1() + +func kernelExitToEl0() + +// Shutdown execution +func Shutdown() diff --git a/pkg/sentry/platform/ring0/entry_arm64.s b/pkg/sentry/platform/ring0/entry_arm64.s new file mode 100644 index 000000000..29c475882 --- /dev/null +++ b/pkg/sentry/platform/ring0/entry_arm64.s @@ -0,0 +1,565 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "funcdata.h" +#include "textflag.h" + +// NB: Offsets are programatically generated (see BUILD). +// +// This file is concatenated with the definitions. + +// Saves a register set. +// +// This is a macro because it may need to executed in contents where a stack is +// not available for calls. +// + +#define ERET() \ + WORD $0xd69f03e0 + +#define RSV_REG R18_PLATFORM +#define RSV_REG_APP R9 + +#define REGISTERS_SAVE(reg, offset) \ + MOVD R0, offset+PTRACE_R0(reg); \ + MOVD R1, offset+PTRACE_R1(reg); \ + MOVD R2, offset+PTRACE_R2(reg); \ + MOVD R3, offset+PTRACE_R3(reg); \ + MOVD R4, offset+PTRACE_R4(reg); \ + MOVD R5, offset+PTRACE_R5(reg); \ + MOVD R6, offset+PTRACE_R6(reg); \ + MOVD R7, offset+PTRACE_R7(reg); \ + MOVD R8, offset+PTRACE_R8(reg); \ + MOVD R10, offset+PTRACE_R10(reg); \ + MOVD R11, offset+PTRACE_R11(reg); \ + MOVD R12, offset+PTRACE_R12(reg); \ + MOVD R13, offset+PTRACE_R13(reg); \ + MOVD R14, offset+PTRACE_R14(reg); \ + MOVD R15, offset+PTRACE_R15(reg); \ + MOVD R16, offset+PTRACE_R16(reg); \ + MOVD R17, offset+PTRACE_R17(reg); \ + MOVD R19, offset+PTRACE_R19(reg); \ + MOVD R20, offset+PTRACE_R20(reg); \ + MOVD R21, offset+PTRACE_R21(reg); \ + MOVD R22, offset+PTRACE_R22(reg); \ + MOVD R23, offset+PTRACE_R23(reg); \ + MOVD R24, offset+PTRACE_R24(reg); \ + MOVD R25, offset+PTRACE_R25(reg); \ + MOVD R26, offset+PTRACE_R26(reg); \ + MOVD R27, offset+PTRACE_R27(reg); \ + MOVD g, offset+PTRACE_R28(reg); \ + MOVD R29, offset+PTRACE_R29(reg); \ + MOVD R30, offset+PTRACE_R30(reg); + +#define REGISTERS_LOAD(reg, offset) \ + MOVD offset+PTRACE_R0(reg), R0; \ + MOVD offset+PTRACE_R1(reg), R1; \ + MOVD offset+PTRACE_R2(reg), R2; \ + MOVD offset+PTRACE_R3(reg), R3; \ + MOVD offset+PTRACE_R4(reg), R4; \ + MOVD offset+PTRACE_R5(reg), R5; \ + MOVD offset+PTRACE_R6(reg), R6; \ + MOVD offset+PTRACE_R7(reg), R7; \ + MOVD offset+PTRACE_R8(reg), R8; \ + MOVD offset+PTRACE_R10(reg), R10; \ + MOVD offset+PTRACE_R11(reg), R11; \ + MOVD offset+PTRACE_R12(reg), R12; \ + MOVD offset+PTRACE_R13(reg), R13; \ + MOVD offset+PTRACE_R14(reg), R14; \ + MOVD offset+PTRACE_R15(reg), R15; \ + MOVD offset+PTRACE_R16(reg), R16; \ + MOVD offset+PTRACE_R17(reg), R17; \ + MOVD offset+PTRACE_R19(reg), R19; \ + MOVD offset+PTRACE_R20(reg), R20; \ + MOVD offset+PTRACE_R21(reg), R21; \ + MOVD offset+PTRACE_R22(reg), R22; \ + MOVD offset+PTRACE_R23(reg), R23; \ + MOVD offset+PTRACE_R24(reg), R24; \ + MOVD offset+PTRACE_R25(reg), R25; \ + MOVD offset+PTRACE_R26(reg), R26; \ + MOVD offset+PTRACE_R27(reg), R27; \ + MOVD offset+PTRACE_R28(reg), g; \ + MOVD offset+PTRACE_R29(reg), R29; \ + MOVD offset+PTRACE_R30(reg), R30; + +//NOP +#define nop31Instructions() \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; \ + WORD $0xd503201f; + +#define ESR_ELx_EC_UNKNOWN (0x00) +#define ESR_ELx_EC_WFx (0x01) +/* Unallocated EC: 0x02 */ +#define ESR_ELx_EC_CP15_32 (0x03) +#define ESR_ELx_EC_CP15_64 (0x04) +#define ESR_ELx_EC_CP14_MR (0x05) +#define ESR_ELx_EC_CP14_LS (0x06) +#define ESR_ELx_EC_FP_ASIMD (0x07) +#define ESR_ELx_EC_CP10_ID (0x08) /* EL2 only */ +#define ESR_ELx_EC_PAC (0x09) /* EL2 and above */ +/* Unallocated EC: 0x0A - 0x0B */ +#define ESR_ELx_EC_CP14_64 (0x0C) +/* Unallocated EC: 0x0d */ +#define ESR_ELx_EC_ILL (0x0E) +/* Unallocated EC: 0x0F - 0x10 */ +#define ESR_ELx_EC_SVC32 (0x11) +#define ESR_ELx_EC_HVC32 (0x12) /* EL2 only */ +#define ESR_ELx_EC_SMC32 (0x13) /* EL2 and above */ +/* Unallocated EC: 0x14 */ +#define ESR_ELx_EC_SVC64 (0x15) +#define ESR_ELx_EC_HVC64 (0x16) /* EL2 and above */ +#define ESR_ELx_EC_SMC64 (0x17) /* EL2 and above */ +#define ESR_ELx_EC_SYS64 (0x18) +#define ESR_ELx_EC_SVE (0x19) +/* Unallocated EC: 0x1A - 0x1E */ +#define ESR_ELx_EC_IMP_DEF (0x1f) /* EL3 only */ +#define ESR_ELx_EC_IABT_LOW (0x20) +#define ESR_ELx_EC_IABT_CUR (0x21) +#define ESR_ELx_EC_PC_ALIGN (0x22) +/* Unallocated EC: 0x23 */ +#define ESR_ELx_EC_DABT_LOW (0x24) +#define ESR_ELx_EC_DABT_CUR (0x25) +#define ESR_ELx_EC_SP_ALIGN (0x26) +/* Unallocated EC: 0x27 */ +#define ESR_ELx_EC_FP_EXC32 (0x28) +/* Unallocated EC: 0x29 - 0x2B */ +#define ESR_ELx_EC_FP_EXC64 (0x2C) +/* Unallocated EC: 0x2D - 0x2E */ +#define ESR_ELx_EC_SERROR (0x2F) +#define ESR_ELx_EC_BREAKPT_LOW (0x30) +#define ESR_ELx_EC_BREAKPT_CUR (0x31) +#define ESR_ELx_EC_SOFTSTP_LOW (0x32) +#define ESR_ELx_EC_SOFTSTP_CUR (0x33) +#define ESR_ELx_EC_WATCHPT_LOW (0x34) +#define ESR_ELx_EC_WATCHPT_CUR (0x35) +/* Unallocated EC: 0x36 - 0x37 */ +#define ESR_ELx_EC_BKPT32 (0x38) +/* Unallocated EC: 0x39 */ +#define ESR_ELx_EC_VECTOR32 (0x3A) /* EL2 only */ +/* Unallocted EC: 0x3B */ +#define ESR_ELx_EC_BRK64 (0x3C) +/* Unallocated EC: 0x3D - 0x3F */ +#define ESR_ELx_EC_MAX (0x3F) + +#define ESR_ELx_EC_SHIFT (26) +#define ESR_ELx_EC_MASK (UL(0x3F) << ESR_ELx_EC_SHIFT) +#define ESR_ELx_EC(esr) (((esr) & ESR_ELx_EC_MASK) >> ESR_ELx_EC_SHIFT) + +#define ESR_ELx_IL_SHIFT (25) +#define ESR_ELx_IL (UL(1) << ESR_ELx_IL_SHIFT) +#define ESR_ELx_ISS_MASK (ESR_ELx_IL - 1) + +/* ISS field definitions shared by different classes */ +#define ESR_ELx_WNR_SHIFT (6) +#define ESR_ELx_WNR (UL(1) << ESR_ELx_WNR_SHIFT) + +/* Asynchronous Error Type */ +#define ESR_ELx_IDS_SHIFT (24) +#define ESR_ELx_IDS (UL(1) << ESR_ELx_IDS_SHIFT) +#define ESR_ELx_AET_SHIFT (10) +#define ESR_ELx_AET (UL(0x7) << ESR_ELx_AET_SHIFT) + +#define ESR_ELx_AET_UC (UL(0) << ESR_ELx_AET_SHIFT) +#define ESR_ELx_AET_UEU (UL(1) << ESR_ELx_AET_SHIFT) +#define ESR_ELx_AET_UEO (UL(2) << ESR_ELx_AET_SHIFT) +#define ESR_ELx_AET_UER (UL(3) << ESR_ELx_AET_SHIFT) +#define ESR_ELx_AET_CE (UL(6) << ESR_ELx_AET_SHIFT) + +/* Shared ISS field definitions for Data/Instruction aborts */ +#define ESR_ELx_SET_SHIFT (11) +#define ESR_ELx_SET_MASK (UL(3) << ESR_ELx_SET_SHIFT) +#define ESR_ELx_FnV_SHIFT (10) +#define ESR_ELx_FnV (UL(1) << ESR_ELx_FnV_SHIFT) +#define ESR_ELx_EA_SHIFT (9) +#define ESR_ELx_EA (UL(1) << ESR_ELx_EA_SHIFT) +#define ESR_ELx_S1PTW_SHIFT (7) +#define ESR_ELx_S1PTW (UL(1) << ESR_ELx_S1PTW_SHIFT) + +/* Shared ISS fault status code(IFSC/DFSC) for Data/Instruction aborts */ +#define ESR_ELx_FSC (0x3F) +#define ESR_ELx_FSC_TYPE (0x3C) +#define ESR_ELx_FSC_EXTABT (0x10) +#define ESR_ELx_FSC_SERROR (0x11) +#define ESR_ELx_FSC_ACCESS (0x08) +#define ESR_ELx_FSC_FAULT (0x04) +#define ESR_ELx_FSC_PERM (0x0C) + +/* ISS field definitions for Data Aborts */ +#define ESR_ELx_ISV_SHIFT (24) +#define ESR_ELx_ISV (UL(1) << ESR_ELx_ISV_SHIFT) +#define ESR_ELx_SAS_SHIFT (22) +#define ESR_ELx_SAS (UL(3) << ESR_ELx_SAS_SHIFT) +#define ESR_ELx_SSE_SHIFT (21) +#define ESR_ELx_SSE (UL(1) << ESR_ELx_SSE_SHIFT) +#define ESR_ELx_SRT_SHIFT (16) +#define ESR_ELx_SRT_MASK (UL(0x1F) << ESR_ELx_SRT_SHIFT) +#define ESR_ELx_SF_SHIFT (15) +#define ESR_ELx_SF (UL(1) << ESR_ELx_SF_SHIFT) +#define ESR_ELx_AR_SHIFT (14) +#define ESR_ELx_AR (UL(1) << ESR_ELx_AR_SHIFT) +#define ESR_ELx_CM_SHIFT (8) +#define ESR_ELx_CM (UL(1) << ESR_ELx_CM_SHIFT) + +/* ISS field definitions for exceptions taken in to Hyp */ +#define ESR_ELx_CV (UL(1) << 24) +#define ESR_ELx_COND_SHIFT (20) +#define ESR_ELx_COND_MASK (UL(0xF) << ESR_ELx_COND_SHIFT) +#define ESR_ELx_WFx_ISS_TI (UL(1) << 0) +#define ESR_ELx_WFx_ISS_WFI (UL(0) << 0) +#define ESR_ELx_WFx_ISS_WFE (UL(1) << 0) +#define ESR_ELx_xVC_IMM_MASK ((1UL << 16) - 1) + +#define LOAD_KERNEL_ADDRESS(from, to) \ + MOVD from, to; \ + ORR $0xffff000000000000, to, to; + +// LOAD_KERNEL_STACK loads the kernel temporary stack. +#define LOAD_KERNEL_STACK(from) \ + LOAD_KERNEL_ADDRESS(CPU_SELF(from), RSV_REG); \ + MOVD $CPU_STACK_TOP(RSV_REG), RSV_REG; \ + MOVD RSV_REG, RSP; \ + ISB $15; \ + DSB $15; + +#define SWITCH_TO_APP_PAGETABLE(from) \ + MOVD CPU_TTBR0_APP(from), RSV_REG; \ + WORD $0xd5182012; \ // MSR R18, TTBR0_EL1 + ISB $15; \ + DSB $15; + +#define SWITCH_TO_KVM_PAGETABLE(from) \ + MOVD CPU_TTBR0_KVM(from), RSV_REG; \ + WORD $0xd5182012; \ // MSR R18, TTBR0_EL1 + ISB $15; \ + DSB $15; + +#define IRQ_ENABLE \ + MSR $2, DAIFSet; + +#define IRQ_DISABLE \ + MSR $2, DAIFClr; + +#define KERNEL_ENTRY_FROM_EL0 \ + SUB $16, RSP, RSP; \ // step1, save r18, r9 into kernel temporary stack. + STP (RSV_REG, RSV_REG_APP), 16*0(RSP); \ + WORD $0xd538d092; \ //MRS TPIDR_EL1, R18, step2, switch user pagetable. + SWITCH_TO_KVM_PAGETABLE(RSV_REG); \ + WORD $0xd538d092; \ //MRS TPIDR_EL1, R18 + MOVD CPU_APP_ADDR(RSV_REG), RSV_REG_APP; \ // step3, load app context pointer. + REGISTERS_SAVE(RSV_REG_APP, 0); \ // step4, save app context. + MOVD RSV_REG_APP, R20; \ + LDP 16*0(RSP), (RSV_REG, RSV_REG_APP); \ + ADD $16, RSP, RSP; \ + MOVD RSV_REG, PTRACE_R18(R20); \ + MOVD RSV_REG_APP, PTRACE_R9(R20); \ + MOVD R20, RSV_REG_APP; \ + WORD $0xd5384003; \ // MRS SPSR_EL1, R3 + MOVD R3, PTRACE_PSTATE(RSV_REG_APP); \ + MRS ELR_EL1, R3; \ + MOVD R3, PTRACE_PC(RSV_REG_APP); \ + WORD $0xd5384103; \ // MRS SP_EL0, R3 + MOVD R3, PTRACE_SP(RSV_REG_APP); + +#define KERNEL_ENTRY_FROM_EL1 \ + WORD $0xd538d092; \ //MRS TPIDR_EL1, R18 + REGISTERS_SAVE(RSV_REG, CPU_REGISTERS); \ // save sentry context + MOVD RSV_REG_APP, CPU_REGISTERS+PTRACE_R9(RSV_REG); \ + WORD $0xd5384004; \ // MRS SPSR_EL1, R4 + MOVD R4, CPU_REGISTERS+PTRACE_PSTATE(RSV_REG); \ + MRS ELR_EL1, R4; \ + MOVD R4, CPU_REGISTERS+PTRACE_PC(RSV_REG); \ + MOVD RSP, R4; \ + MOVD R4, CPU_REGISTERS+PTRACE_SP(RSV_REG); + +TEXT ·Halt(SB),NOSPLIT,$0 + // clear bluepill. + WORD $0xd538d092 //MRS TPIDR_EL1, R18 + CMP RSV_REG, R9 + BNE mmio_exit + MOVD $0, CPU_REGISTERS+PTRACE_R9(RSV_REG) +mmio_exit: + // MMIO_EXIT. + MOVD $0, R9 + MOVD R0, 0xffff000000001000(R9) + B ·kernelExitToEl1(SB) + +TEXT ·Shutdown(SB),NOSPLIT,$0 + // PSCI EVENT. + MOVD $0x84000009, R0 + HVC $0 + +// See kernel.go. +TEXT ·Current(SB),NOSPLIT,$0-8 + MOVD CPU_SELF(RSV_REG), R8 + MOVD R8, ret+0(FP) + RET + +#define STACK_FRAME_SIZE 16 + +TEXT ·kernelExitToEl0(SB),NOSPLIT,$0 + ERET() + +TEXT ·kernelExitToEl1(SB),NOSPLIT,$0 + ERET() + +TEXT ·Start(SB),NOSPLIT,$0 + IRQ_DISABLE + MOVD R8, RSV_REG + ORR $0xffff000000000000, RSV_REG, RSV_REG + WORD $0xd518d092 //MSR R18, TPIDR_EL1 + + B ·kernelExitToEl1(SB) + +TEXT ·El1_sync_invalid(SB),NOSPLIT,$0 + B ·Shutdown(SB) + +TEXT ·El1_irq_invalid(SB),NOSPLIT,$0 + B ·Shutdown(SB) + +TEXT ·El1_fiq_invalid(SB),NOSPLIT,$0 + B ·Shutdown(SB) + +TEXT ·El1_error_invalid(SB),NOSPLIT,$0 + B ·Shutdown(SB) + +TEXT ·El1_sync(SB),NOSPLIT,$0 + KERNEL_ENTRY_FROM_EL1 + WORD $0xd5385219 // MRS ESR_EL1, R25 + LSR $ESR_ELx_EC_SHIFT, R25, R24 + CMP $ESR_ELx_EC_DABT_CUR, R24 + BEQ el1_da + CMP $ESR_ELx_EC_IABT_CUR, R24 + BEQ el1_ia + CMP $ESR_ELx_EC_SYS64, R24 + BEQ el1_undef + CMP $ESR_ELx_EC_SP_ALIGN, R24 + BEQ el1_sp_pc + CMP $ESR_ELx_EC_PC_ALIGN, R24 + BEQ el1_sp_pc + CMP $ESR_ELx_EC_UNKNOWN, R24 + BEQ el1_undef + CMP $ESR_ELx_EC_SVC64, R24 + BEQ el1_svc + CMP $ESR_ELx_EC_BREAKPT_CUR, R24 + BGE el1_dbg + B el1_invalid + +el1_da: + B ·Halt(SB) + +el1_ia: + B ·Halt(SB) + +el1_sp_pc: + B ·Shutdown(SB) + +el1_undef: + B ·Shutdown(SB) + +el1_svc: + B ·Halt(SB) + +el1_dbg: + B ·Shutdown(SB) + +el1_invalid: + B ·Shutdown(SB) + +TEXT ·El1_irq(SB),NOSPLIT,$0 + B ·Shutdown(SB) + +TEXT ·El1_fiq(SB),NOSPLIT,$0 + B ·Shutdown(SB) + +TEXT ·El1_error(SB),NOSPLIT,$0 + B ·Shutdown(SB) + +TEXT ·El0_sync(SB),NOSPLIT,$0 + KERNEL_ENTRY_FROM_EL0 + WORD $0xd5385219 // MRS ESR_EL1, R25 + LSR $ESR_ELx_EC_SHIFT, R25, R24 + CMP $ESR_ELx_EC_SVC64, R24 + BEQ el0_svc + CMP $ESR_ELx_EC_DABT_LOW, R24 + BEQ el0_da + CMP $ESR_ELx_EC_IABT_LOW, R24 + BEQ el0_ia + CMP $ESR_ELx_EC_FP_ASIMD, R24 + BEQ el0_fpsimd_acc + CMP $ESR_ELx_EC_SVE, R24 + BEQ el0_sve_acc + CMP $ESR_ELx_EC_FP_EXC64, R24 + BEQ el0_fpsimd_exc + CMP $ESR_ELx_EC_SP_ALIGN, R24 + BEQ el0_sp_pc + CMP $ESR_ELx_EC_PC_ALIGN, R24 + BEQ el0_sp_pc + CMP $ESR_ELx_EC_UNKNOWN, R24 + BEQ el0_undef + CMP $ESR_ELx_EC_BREAKPT_LOW, R24 + BGE el0_dbg + B el0_invalid + +el0_svc: + B ·Halt(SB) + +el0_da: + B ·Halt(SB) + +el0_ia: + B ·Shutdown(SB) + +el0_fpsimd_acc: + B ·Shutdown(SB) + +el0_sve_acc: + B ·Shutdown(SB) + +el0_fpsimd_exc: + B ·Shutdown(SB) + +el0_sp_pc: + B ·Shutdown(SB) + +el0_undef: + B ·Shutdown(SB) + +el0_dbg: + B ·Shutdown(SB) + +el0_invalid: + B ·Shutdown(SB) + +TEXT ·El0_irq(SB),NOSPLIT,$0 + B ·Shutdown(SB) + +TEXT ·El0_fiq(SB),NOSPLIT,$0 + B ·Shutdown(SB) + +TEXT ·El0_error(SB),NOSPLIT,$0 + B ·Shutdown(SB) + +TEXT ·El0_sync_invalid(SB),NOSPLIT,$0 + B ·Shutdown(SB) + +TEXT ·El0_irq_invalid(SB),NOSPLIT,$0 + B ·Shutdown(SB) + +TEXT ·El0_fiq_invalid(SB),NOSPLIT,$0 + B ·Shutdown(SB) + +TEXT ·El0_error_invalid(SB),NOSPLIT,$0 + B ·Shutdown(SB) + +TEXT ·Vectors(SB),NOSPLIT,$0 + B ·El1_sync_invalid(SB) + nop31Instructions() + B ·El1_irq_invalid(SB) + nop31Instructions() + B ·El1_fiq_invalid(SB) + nop31Instructions() + B ·El1_error_invalid(SB) + nop31Instructions() + + B ·El1_sync(SB) + nop31Instructions() + B ·El1_irq(SB) + nop31Instructions() + B ·El1_fiq(SB) + nop31Instructions() + B ·El1_error(SB) + nop31Instructions() + + B ·El0_sync(SB) + nop31Instructions() + B ·El0_irq(SB) + nop31Instructions() + B ·El0_fiq(SB) + nop31Instructions() + B ·El0_error(SB) + nop31Instructions() + + B ·El0_sync_invalid(SB) + nop31Instructions() + B ·El0_irq_invalid(SB) + nop31Instructions() + B ·El0_fiq_invalid(SB) + nop31Instructions() + B ·El0_error_invalid(SB) + nop31Instructions() + + WORD $0xd503201f //nop + nop31Instructions() + WORD $0xd503201f + nop31Instructions() + WORD $0xd503201f + nop31Instructions() + WORD $0xd503201f + nop31Instructions() + + WORD $0xd503201f + nop31Instructions() + WORD $0xd503201f + nop31Instructions() + WORD $0xd503201f + nop31Instructions() + WORD $0xd503201f + nop31Instructions() + + WORD $0xd503201f + nop31Instructions() + WORD $0xd503201f + nop31Instructions() + WORD $0xd503201f + nop31Instructions() + WORD $0xd503201f + nop31Instructions() + + WORD $0xd503201f + nop31Instructions() + WORD $0xd503201f + nop31Instructions() + WORD $0xd503201f + nop31Instructions() + WORD $0xd503201f + nop31Instructions() diff --git a/pkg/sentry/platform/ring0/kernel_arm64.go b/pkg/sentry/platform/ring0/kernel_arm64.go new file mode 100644 index 000000000..ed82a131e --- /dev/null +++ b/pkg/sentry/platform/ring0/kernel_arm64.go @@ -0,0 +1,58 @@ +// Copyright 2019 Google Inc. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package ring0 + +// init initializes architecture-specific state. +func (k *Kernel) init(opts KernelOpts) { + // Save the root page tables. + k.PageTables = opts.PageTables +} + +// init initializes architecture-specific state. +func (c *CPU) init() { + // Set the kernel stack pointer(virtual address). + c.registers.Sp = uint64(c.StackTop()) + +} + +// StackTop returns the kernel's stack address. +// +//go:nosplit +func (c *CPU) StackTop() uint64 { + return uint64(kernelAddr(&c.stack[0])) + uint64(len(c.stack)) +} + +// IsCanonical indicates whether addr is canonical per the arm64 spec. +// +//go:nosplit +func IsCanonical(addr uint64) bool { + return addr <= 0x0000ffffffffffff || addr > 0xffff000000000000 +} + +//go:nosplit +func (c *CPU) SwitchToUser(switchOpts SwitchOpts) (vector Vector) { + // Sanitize registers. + regs := switchOpts.Registers + + regs.Pstate &= ^uint64(UserFlagsClear) + regs.Pstate |= UserFlagsSet + kernelExitToEl0() + vector = c.vecCode + + // Perform the switch. + return +} diff --git a/pkg/sentry/platform/ring0/lib_arm64.go b/pkg/sentry/platform/ring0/lib_arm64.go new file mode 100644 index 000000000..900ee6380 --- /dev/null +++ b/pkg/sentry/platform/ring0/lib_arm64.go @@ -0,0 +1,25 @@ +// Copyright 2019 Google Inc. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package ring0 + +// LoadFloatingPoint loads floating point state by the most efficient mechanism +// available (set by Init). +var LoadFloatingPoint func(*byte) + +// SaveFloatingPoint saves floating point state by the most efficient mechanism +// available (set by Init). +var SaveFloatingPoint func(*byte) diff --git a/pkg/sentry/platform/ring0/offsets_arm64.go b/pkg/sentry/platform/ring0/offsets_arm64.go new file mode 100644 index 000000000..d7aa1c7cc --- /dev/null +++ b/pkg/sentry/platform/ring0/offsets_arm64.go @@ -0,0 +1,124 @@ +// Copyright 2019 Google Inc. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package ring0 + +import ( + "fmt" + "io" + "reflect" + "syscall" +) + +// Emit prints architecture-specific offsets. +func Emit(w io.Writer) { + fmt.Fprintf(w, "// Automatically generated, do not edit.\n") + + c := &CPU{} + fmt.Fprintf(w, "\n// CPU offsets.\n") + fmt.Fprintf(w, "#define CPU_SELF 0x%02x\n", reflect.ValueOf(&c.self).Pointer()-reflect.ValueOf(c).Pointer()) + fmt.Fprintf(w, "#define CPU_REGISTERS 0x%02x\n", reflect.ValueOf(&c.registers).Pointer()-reflect.ValueOf(c).Pointer()) + fmt.Fprintf(w, "#define CPU_STACK_TOP 0x%02x\n", reflect.ValueOf(&c.stack[0]).Pointer()-reflect.ValueOf(c).Pointer()+uintptr(len(c.stack))) + fmt.Fprintf(w, "#define CPU_ERROR_CODE 0x%02x\n", reflect.ValueOf(&c.errorCode).Pointer()-reflect.ValueOf(c).Pointer()) + fmt.Fprintf(w, "#define CPU_ERROR_TYPE 0x%02x\n", reflect.ValueOf(&c.errorType).Pointer()-reflect.ValueOf(c).Pointer()) + fmt.Fprintf(w, "#define CPU_FAULT_ADDR 0x%02x\n", reflect.ValueOf(&c.faultAddr).Pointer()-reflect.ValueOf(c).Pointer()) + fmt.Fprintf(w, "#define CPU_TTBR0_KVM 0x%02x\n", reflect.ValueOf(&c.ttbr0Kvm).Pointer()-reflect.ValueOf(c).Pointer()) + fmt.Fprintf(w, "#define CPU_TTBR0_APP 0x%02x\n", reflect.ValueOf(&c.ttbr0App).Pointer()-reflect.ValueOf(c).Pointer()) + fmt.Fprintf(w, "#define CPU_VECTOR_CODE 0x%02x\n", reflect.ValueOf(&c.vecCode).Pointer()-reflect.ValueOf(c).Pointer()) + fmt.Fprintf(w, "#define CPU_APP_ADDR 0x%02x\n", reflect.ValueOf(&c.appAddr).Pointer()-reflect.ValueOf(c).Pointer()) + + fmt.Fprintf(w, "\n// Bits.\n") + fmt.Fprintf(w, "#define _KERNEL_FLAGS 0x%02x\n", KernelFlagsSet) + + fmt.Fprintf(w, "\n// Vectors.\n") + fmt.Fprintf(w, "#define El1SyncInvalid 0x%02x\n", El1SyncInvalid) + fmt.Fprintf(w, "#define El1IrqInvalid 0x%02x\n", El1IrqInvalid) + fmt.Fprintf(w, "#define El1FiqInvalid 0x%02x\n", El1FiqInvalid) + fmt.Fprintf(w, "#define El1ErrorInvalid 0x%02x\n", El1ErrorInvalid) + + fmt.Fprintf(w, "#define El1Sync 0x%02x\n", El1Sync) + fmt.Fprintf(w, "#define El1Irq 0x%02x\n", El1Irq) + fmt.Fprintf(w, "#define El1Fiq 0x%02x\n", El1Fiq) + fmt.Fprintf(w, "#define El1Error 0x%02x\n", El1Error) + + fmt.Fprintf(w, "#define El0Sync 0x%02x\n", El0Sync) + fmt.Fprintf(w, "#define El0Irq 0x%02x\n", El0Irq) + fmt.Fprintf(w, "#define El0Fiq 0x%02x\n", El0Fiq) + fmt.Fprintf(w, "#define El0Error 0x%02x\n", El0Error) + + fmt.Fprintf(w, "#define El0Sync_invalid 0x%02x\n", El0Sync_invalid) + fmt.Fprintf(w, "#define El0Irq_invalid 0x%02x\n", El0Irq_invalid) + fmt.Fprintf(w, "#define El0Fiq_invalid 0x%02x\n", El0Fiq_invalid) + fmt.Fprintf(w, "#define El0Error_invalid 0x%02x\n", El0Error_invalid) + + fmt.Fprintf(w, "#define El1Sync_da 0x%02x\n", El1Sync_da) + fmt.Fprintf(w, "#define El1Sync_ia 0x%02x\n", El1Sync_ia) + fmt.Fprintf(w, "#define El1Sync_sp_pc 0x%02x\n", El1Sync_sp_pc) + fmt.Fprintf(w, "#define El1Sync_undef 0x%02x\n", El1Sync_undef) + fmt.Fprintf(w, "#define El1Sync_dbg 0x%02x\n", El1Sync_dbg) + fmt.Fprintf(w, "#define El1Sync_inv 0x%02x\n", El1Sync_inv) + + fmt.Fprintf(w, "#define El0Sync_svc 0x%02x\n", El0Sync_svc) + fmt.Fprintf(w, "#define El0Sync_da 0x%02x\n", El0Sync_da) + fmt.Fprintf(w, "#define El0Sync_ia 0x%02x\n", El0Sync_ia) + fmt.Fprintf(w, "#define El0Sync_fpsimd_acc 0x%02x\n", El0Sync_fpsimd_acc) + fmt.Fprintf(w, "#define El0Sync_sve_acc 0x%02x\n", El0Sync_sve_acc) + fmt.Fprintf(w, "#define El0Sync_sys 0x%02x\n", El0Sync_sys) + fmt.Fprintf(w, "#define El0Sync_sp_pc 0x%02x\n", El0Sync_sp_pc) + fmt.Fprintf(w, "#define El0Sync_undef 0x%02x\n", El0Sync_undef) + fmt.Fprintf(w, "#define El0Sync_dbg 0x%02x\n", El0Sync_dbg) + fmt.Fprintf(w, "#define El0Sync_inv 0x%02x\n", El0Sync_inv) + + fmt.Fprintf(w, "#define PageFault 0x%02x\n", PageFault) + fmt.Fprintf(w, "#define Syscall 0x%02x\n", Syscall) + + p := &syscall.PtraceRegs{} + fmt.Fprintf(w, "\n// Ptrace registers.\n") + fmt.Fprintf(w, "#define PTRACE_R0 0x%02x\n", reflect.ValueOf(&p.Regs[0]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R1 0x%02x\n", reflect.ValueOf(&p.Regs[1]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R2 0x%02x\n", reflect.ValueOf(&p.Regs[2]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R3 0x%02x\n", reflect.ValueOf(&p.Regs[3]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R4 0x%02x\n", reflect.ValueOf(&p.Regs[4]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R5 0x%02x\n", reflect.ValueOf(&p.Regs[5]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R6 0x%02x\n", reflect.ValueOf(&p.Regs[6]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R7 0x%02x\n", reflect.ValueOf(&p.Regs[7]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R8 0x%02x\n", reflect.ValueOf(&p.Regs[8]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R9 0x%02x\n", reflect.ValueOf(&p.Regs[9]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R10 0x%02x\n", reflect.ValueOf(&p.Regs[10]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R11 0x%02x\n", reflect.ValueOf(&p.Regs[11]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R12 0x%02x\n", reflect.ValueOf(&p.Regs[12]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R13 0x%02x\n", reflect.ValueOf(&p.Regs[13]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R14 0x%02x\n", reflect.ValueOf(&p.Regs[14]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R15 0x%02x\n", reflect.ValueOf(&p.Regs[15]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R16 0x%02x\n", reflect.ValueOf(&p.Regs[16]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R17 0x%02x\n", reflect.ValueOf(&p.Regs[17]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R18 0x%02x\n", reflect.ValueOf(&p.Regs[18]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R19 0x%02x\n", reflect.ValueOf(&p.Regs[19]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R20 0x%02x\n", reflect.ValueOf(&p.Regs[20]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R21 0x%02x\n", reflect.ValueOf(&p.Regs[21]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R22 0x%02x\n", reflect.ValueOf(&p.Regs[22]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R23 0x%02x\n", reflect.ValueOf(&p.Regs[23]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R24 0x%02x\n", reflect.ValueOf(&p.Regs[24]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R25 0x%02x\n", reflect.ValueOf(&p.Regs[25]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R26 0x%02x\n", reflect.ValueOf(&p.Regs[26]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R27 0x%02x\n", reflect.ValueOf(&p.Regs[27]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R28 0x%02x\n", reflect.ValueOf(&p.Regs[28]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R29 0x%02x\n", reflect.ValueOf(&p.Regs[29]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R30 0x%02x\n", reflect.ValueOf(&p.Regs[30]).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_SP 0x%02x\n", reflect.ValueOf(&p.Sp).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_PC 0x%02x\n", reflect.ValueOf(&p.Pc).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_PSTATE 0x%02x\n", reflect.ValueOf(&p.Pstate).Pointer()-reflect.ValueOf(p).Pointer()) +} -- cgit v1.2.3 From 345f140169dc59bfc73bc522f4877f1683b2558a Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Fri, 20 Sep 2019 14:37:33 +0000 Subject: Optimize kvm/physical_map.go on Arm platform Signed-off-by: Bin Lu --- pkg/sentry/platform/kvm/BUILD | 2 ++ pkg/sentry/platform/kvm/physical_map.go | 12 +----------- pkg/sentry/platform/kvm/physical_map_amd64.go | 22 ++++++++++++++++++++++ pkg/sentry/platform/kvm/physical_map_arm64.go | 19 +++++++++++++++++++ 4 files changed, 44 insertions(+), 11 deletions(-) create mode 100644 pkg/sentry/platform/kvm/physical_map_amd64.go create mode 100644 pkg/sentry/platform/kvm/physical_map_arm64.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/kvm/BUILD b/pkg/sentry/platform/kvm/BUILD index fe979dccf..2046df525 100644 --- a/pkg/sentry/platform/kvm/BUILD +++ b/pkg/sentry/platform/kvm/BUILD @@ -24,6 +24,8 @@ go_library( "machine_amd64_unsafe.go", "machine_unsafe.go", "physical_map.go", + "physical_map_amd64.go", + "physical_map_arm64.go", "virtual_map.go", ], importpath = "gvisor.dev/gvisor/pkg/sentry/platform/kvm", diff --git a/pkg/sentry/platform/kvm/physical_map.go b/pkg/sentry/platform/kvm/physical_map.go index 586e91bb2..91de5dab1 100644 --- a/pkg/sentry/platform/kvm/physical_map.go +++ b/pkg/sentry/platform/kvm/physical_map.go @@ -24,15 +24,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/usermem" ) -const ( - // reservedMemory is a chunk of physical memory reserved starting at - // physical address zero. There are some special pages in this region, - // so we just call the whole thing off. - // - // Other architectures may define this to be zero. - reservedMemory = 0x100000000 -) - type region struct { virtual uintptr length uintptr @@ -59,8 +50,7 @@ func fillAddressSpace() (excludedRegions []region) { // We can cut vSize in half, because the kernel will be using the top // half and we ignore it while constructing mappings. It's as if we've // already excluded half the possible addresses. - vSize := uintptr(1) << ring0.VirtualAddressBits() - vSize = vSize >> 1 + vSize := ring0.UserspaceSize // We exclude reservedMemory below from our physical memory size, so it // needs to be dropped here as well. Otherwise, we could end up with diff --git a/pkg/sentry/platform/kvm/physical_map_amd64.go b/pkg/sentry/platform/kvm/physical_map_amd64.go new file mode 100644 index 000000000..c5adfb577 --- /dev/null +++ b/pkg/sentry/platform/kvm/physical_map_amd64.go @@ -0,0 +1,22 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package kvm + +const ( + // reservedMemory is a chunk of physical memory reserved starting at + // physical address zero. There are some special pages in this region, + // so we just call the whole thing off. + reservedMemory = 0x100000000 +) diff --git a/pkg/sentry/platform/kvm/physical_map_arm64.go b/pkg/sentry/platform/kvm/physical_map_arm64.go new file mode 100644 index 000000000..4d8561453 --- /dev/null +++ b/pkg/sentry/platform/kvm/physical_map_arm64.go @@ -0,0 +1,19 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package kvm + +const ( + reservedMemory = 0 +) -- cgit v1.2.3 From 198f1cddb82d46570ae63cb704b4a1b88cf0de1f Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Mon, 28 Oct 2019 10:18:55 -0700 Subject: Update comment FDTable.GetFile doesn't exist. PiperOrigin-RevId: 277089842 --- pkg/sentry/kernel/task.go | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index c82ef5486..11a8c6c87 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -709,9 +709,9 @@ func (t *Task) FDTable() *FDTable { return t.fdTable } -// GetFile is a convenience wrapper t.FDTable().GetFile. +// GetFile is a convenience wrapper t.FDTable().Get. // -// Precondition: same as FDTable. +// Precondition: same as FDTable.Get. func (t *Task) GetFile(fd int32) *fs.File { f, _ := t.fdTable.Get(fd) return f -- cgit v1.2.3 From 29273b03842a85bce8314799348231520ceb6e9c Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 29 Oct 2019 10:03:18 -0700 Subject: Disallow execveat on interpreter scripts with fd opened with O_CLOEXEC. When an interpreter script is opened with O_CLOEXEC and the resulting fd is passed into execveat, an ENOENT error should occur (the script would otherwise be inaccessible to the interpreter). This matches the actual behavior of Linux's execveat. PiperOrigin-RevId: 277306680 --- pkg/sentry/kernel/kernel.go | 1 + pkg/sentry/loader/loader.go | 9 +++++++++ pkg/sentry/syscalls/linux/sys_thread.go | 5 ++++- test/syscalls/linux/exec.cc | 33 +++++++++++++++++++++++++++++++++ 4 files changed, 47 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index fcfe7a16d..e64d648e2 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -812,6 +812,7 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, ResolveFinal: true, Filename: args.Filename, File: args.File, + CloseOnExec: false, Argv: args.Argv, Envv: args.Envv, Features: k.featureSet, diff --git a/pkg/sentry/loader/loader.go b/pkg/sentry/loader/loader.go index 818941762..f75ebe08a 100644 --- a/pkg/sentry/loader/loader.go +++ b/pkg/sentry/loader/loader.go @@ -66,6 +66,12 @@ type LoadArgs struct { // nil, then File will be loaded and Filename will be ignored. File *fs.File + // CloseOnExec indicates that the executable (or one of its parent + // directories) was opened with O_CLOEXEC. If the executable is an + // interpreter script, then cause an ENOENT error to occur, since the + // script would otherwise be inaccessible to the interpreter. + CloseOnExec bool + // Argv is the vector of arguments to pass to the executable. Argv []string @@ -279,6 +285,9 @@ func loadExecutable(ctx context.Context, args LoadArgs) (loadedELF, arch.Context d.IncRef() return loaded, ac, d, args.Argv, err case bytes.Equal(hdr[:2], []byte(interpreterScriptMagic)): + if args.CloseOnExec { + return loadedELF{}, nil, nil, nil, syserror.ENOENT + } args.Filename, args.Argv, err = parseInterpreterScript(ctx, args.Filename, args.File, args.Argv) if err != nil { ctx.Infof("Error loading interpreter script: %v", err) diff --git a/pkg/sentry/syscalls/linux/sys_thread.go b/pkg/sentry/syscalls/linux/sys_thread.go index 2476f8858..4115116ff 100644 --- a/pkg/sentry/syscalls/linux/sys_thread.go +++ b/pkg/sentry/syscalls/linux/sys_thread.go @@ -120,6 +120,7 @@ func execveat(t *kernel.Task, dirFD int32, pathnameAddr, argvAddr, envvAddr user var wd *fs.Dirent var executable *fs.File + var closeOnExec bool if dirFD == linux.AT_FDCWD || path.IsAbs(pathname) { // Even if the pathname is absolute, we may still need the wd // for interpreter scripts if the path of the interpreter is @@ -127,11 +128,12 @@ func execveat(t *kernel.Task, dirFD int32, pathnameAddr, argvAddr, envvAddr user wd = t.FSContext().WorkingDirectory() } else { // Need to extract the given FD. - f := t.GetFile(dirFD) + f, fdFlags := t.FDTable().Get(dirFD) if f == nil { return 0, nil, syserror.EBADF } defer f.DecRef() + closeOnExec = fdFlags.CloseOnExec if atEmptyPath && len(pathname) == 0 { executable = f @@ -157,6 +159,7 @@ func execveat(t *kernel.Task, dirFD int32, pathnameAddr, argvAddr, envvAddr user ResolveFinal: resolveFinal, Filename: pathname, File: executable, + CloseOnExec: closeOnExec, Argv: argv, Envv: envv, Features: t.Arch().FeatureSet(), diff --git a/test/syscalls/linux/exec.cc b/test/syscalls/linux/exec.cc index 21a5ffd40..a9067df2a 100644 --- a/test/syscalls/linux/exec.cc +++ b/test/syscalls/linux/exec.cc @@ -681,6 +681,39 @@ TEST(ExecveatTest, SymlinkNoFollowWithNormalFile) { ArgEnvExitStatus(0, 0), ""); } +TEST(ExecveatTest, BasicWithCloexecFD) { + std::string path = WorkloadPath(kBasicWorkload); + const FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(path, O_CLOEXEC)); + + CheckExecveat(fd.get(), "", {path}, {}, AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH, + ArgEnvExitStatus(0, 0), absl::StrCat(path, "\n")); +} + +TEST(ExecveatTest, InterpreterScriptWithCloexecFD) { + std::string path = WorkloadPath(kExitScript); + const FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(path, O_CLOEXEC)); + + int execve_errno; + ASSERT_NO_ERRNO_AND_VALUE(ForkAndExecveat(fd.get(), "", {path}, {}, + AT_EMPTY_PATH, /*child=*/nullptr, + &execve_errno)); + EXPECT_EQ(execve_errno, ENOENT); +} + +TEST(ExecveatTest, InterpreterScriptWithCloexecDirFD) { + std::string absolute_path = WorkloadPath(kExitScript); + std::string parent_dir = std::string(Dirname(absolute_path)); + std::string base = std::string(Basename(absolute_path)); + const FileDescriptor dirfd = + ASSERT_NO_ERRNO_AND_VALUE(Open(parent_dir, O_CLOEXEC | O_DIRECTORY)); + + int execve_errno; + ASSERT_NO_ERRNO_AND_VALUE(ForkAndExecveat(dirfd.get(), base, {base}, {}, + /*flags=*/0, /*child=*/nullptr, + &execve_errno)); + EXPECT_EQ(execve_errno, ENOENT); +} + TEST(ExecveatTest, InvalidFlags) { int execve_errno; ASSERT_NO_ERRNO_AND_VALUE(ForkAndExecveat( -- cgit v1.2.3 From 2e00771d5abb3d821703965953c2b21ef7c20911 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 29 Oct 2019 12:50:03 -0700 Subject: Refactor logic for loadExecutable. Separate the handling of filenames and *fs.File objects in a more explicit way for the sake of clarity. PiperOrigin-RevId: 277344203 --- pkg/sentry/loader/loader.go | 134 ++++++++++++++++++++++---------------------- 1 file changed, 67 insertions(+), 67 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/loader/loader.go b/pkg/sentry/loader/loader.go index f75ebe08a..803e7d41e 100644 --- a/pkg/sentry/loader/loader.go +++ b/pkg/sentry/loader/loader.go @@ -20,6 +20,7 @@ import ( "fmt" "io" "path" + "strings" "gvisor.dev/gvisor/pkg/abi" "gvisor.dev/gvisor/pkg/abi/linux" @@ -99,20 +100,20 @@ func readFull(ctx context.Context, f *fs.File, dst usermem.IOSequence, offset in return total, nil } -// openPath opens args.Filename for loading. +// openPath opens args.Filename and checks that it is valid for loading. // -// openPath returns the fs.Dirent and an *fs.File for args.Filename, which is -// not installed in the Task FDTable. The caller takes ownership of both. +// openPath returns an *fs.Dirent and *fs.File for args.Filename, which is not +// installed in the Task FDTable. The caller takes ownership of both. // // args.Filename must be a readable, executable, regular file. func openPath(ctx context.Context, args LoadArgs) (*fs.Dirent, *fs.File, error) { - var err error if args.Filename == "" { ctx.Infof("cannot open empty name") return nil, nil, syserror.ENOENT } var d *fs.Dirent + var err error if args.ResolveFinal { d, err = args.Mounts.FindInode(ctx, args.Root, args.WorkingDirectory, args.Filename, args.RemainingTraversals) } else { @@ -121,67 +122,60 @@ func openPath(ctx context.Context, args LoadArgs) (*fs.Dirent, *fs.File, error) if err != nil { return nil, nil, err } - - // Open file will take a reference to Dirent, so destroy this one. + // Defer a DecRef for the sake of failure cases. defer d.DecRef() if !args.ResolveFinal && fs.IsSymlink(d.Inode.StableAttr) { return nil, nil, syserror.ELOOP } - return openFile(ctx, nil, d, args.Filename) -} + if err := checkPermission(ctx, d); err != nil { + return nil, nil, err + } -// openFile takes that file's Dirent and performs checks on it. If provided a -// *fs.Dirent and not a *fs.File, it creates a *fs.File object from the Dirent's -// Inode and performs checks on that. -// -// openFile returns an *fs.File and *fs.Dirent, and the caller takes ownership -// of both. -// -// "dirent" and "file" must not both be nil and point to a readable, executable, regular file. -func openFile(ctx context.Context, file *fs.File, dirent *fs.Dirent, name string) (*fs.Dirent, *fs.File, error) { - // file and dirent must not be nil. - if dirent == nil && file == nil { - ctx.Infof("dirent and file cannot both be nil.") - return nil, nil, syserror.ENOENT + // If they claim it's a directory, then make sure. + // + // N.B. we reject directories below, but we must first reject + // non-directories passed as directories. + if strings.HasSuffix(args.Filename, "/") && !fs.IsDir(d.Inode.StableAttr) { + return nil, nil, syserror.ENOTDIR + } + + if err := checkIsRegularFile(ctx, d, args.Filename); err != nil { + return nil, nil, err } - if file != nil { - dirent = file.Dirent + f, err := d.Inode.GetFile(ctx, d, fs.FileFlags{Read: true}) + if err != nil { + return nil, nil, err } + // Defer a DecRef for the sake of failure cases. + defer f.DecRef() - // Perform permissions checks on the file. - if err := checkFile(ctx, dirent, name); err != nil { + if err := checkPread(ctx, f, args.Filename); err != nil { return nil, nil, err } - if file == nil { - var ferr error - if file, ferr = dirent.Inode.GetFile(ctx, dirent, fs.FileFlags{Read: true}); ferr != nil { - return nil, nil, ferr - } - } else { - // GetFile takes a reference to the created file, so make one in the case - // that the file reference already existed. - file.IncRef() + d.IncRef() + f.IncRef() + return d, f, err +} + +// checkFile performs checks on a file to be executed. +func checkFile(ctx context.Context, f *fs.File, filename string) error { + if err := checkPermission(ctx, f.Dirent); err != nil { + return err } - // We must be able to read at arbitrary offsets. - if !file.Flags().Pread { - file.DecRef() - ctx.Infof("%s cannot be read at an offset: %+v", file.MappedName(ctx), file.Flags()) - return nil, nil, syserror.EACCES + if err := checkIsRegularFile(ctx, f.Dirent, filename); err != nil { + return err } - // Grab reference for caller. - dirent.IncRef() - return dirent, file, nil + return checkPread(ctx, f, filename) } -// checkFile performs file permissions checks for binaries called in openPath -// and openFile -func checkFile(ctx context.Context, d *fs.Dirent, name string) error { +// checkPermission checks whether the file is readable and executable. +func checkPermission(ctx context.Context, d *fs.Dirent) error { perms := fs.PermMask{ // TODO(gvisor.dev/issue/160): Linux requires only execute // permission, not read. However, our backing filesystems may @@ -192,26 +186,26 @@ func checkFile(ctx context.Context, d *fs.Dirent, name string) error { Read: true, Execute: true, } - if err := d.Inode.CheckPermission(ctx, perms); err != nil { - return err - } + return d.Inode.CheckPermission(ctx, perms) +} - // If they claim it's a directory, then make sure. - // - // N.B. we reject directories below, but we must first reject - // non-directories passed as directories. - if len(name) > 0 && name[len(name)-1] == '/' && !fs.IsDir(d.Inode.StableAttr) { - return syserror.ENOTDIR +// checkIsRegularFile prevents us from trying to execute a directory, pipe, etc. +func checkIsRegularFile(ctx context.Context, d *fs.Dirent, filename string) error { + attr := d.Inode.StableAttr + if !fs.IsRegular(attr) { + ctx.Infof("%s is not regular: %v", filename, attr) + return syserror.EACCES } + return nil +} - // No exec-ing directories, pipes, etc! - if !fs.IsRegular(d.Inode.StableAttr) { - ctx.Infof("%s is not regular: %v", name, d.Inode.StableAttr) +// checkPread checks whether we can read the file at arbitrary offsets. +func checkPread(ctx context.Context, f *fs.File, filename string) error { + if !f.Flags().Pread { + ctx.Infof("%s cannot be read at an offset: %+v", filename, f.Flags()) return syserror.EACCES } - return nil - } // allocStack allocates and maps a stack in to any available part of the address space. @@ -248,25 +242,31 @@ func loadExecutable(ctx context.Context, args LoadArgs) (loadedELF, arch.Context ) if args.File == nil { d, args.File, err = openPath(ctx, args) + // We will return d in the successful case, but defer a DecRef for the + // sake of intermediate loops and failure cases. + if d != nil { + defer d.DecRef() + } + if args.File != nil { + defer args.File.DecRef() + } } else { - d, args.File, err = openFile(ctx, args.File, nil, "") + d = args.File.Dirent + d.IncRef() + defer d.DecRef() + err = checkFile(ctx, args.File, args.Filename) } - if err != nil { ctx.Infof("Error opening %s: %v", args.Filename, err) return loadedELF{}, nil, nil, nil, err } - defer args.File.DecRef() - // We will return d in the successful case, but defer a DecRef - // for intermediate loops and failure cases. - defer d.DecRef() // Check the header. Is this an ELF or interpreter script? var hdr [4]uint8 // N.B. We assume that reading from a regular file cannot block. _, err = readFull(ctx, args.File, usermem.BytesIOSequence(hdr[:]), 0) - // Allow unexpected EOF, as a valid executable could be only - // three bytes (e.g., #!a). + // Allow unexpected EOF, as a valid executable could be only three bytes + // (e.g., #!a). if err != nil && err != io.ErrUnexpectedEOF { if err == io.EOF { err = syserror.ENOEXEC -- cgit v1.2.3 From c0b8fd4b6a9fcb595f3200577b93d07737cfaacd Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Tue, 29 Oct 2019 13:17:01 -0700 Subject: Update build tags to allow Go 1.14 Currently there are no ABI changes. We should check again closer to release. PiperOrigin-RevId: 277349744 --- pkg/procid/procid_amd64.s | 2 +- pkg/procid/procid_arm64.s | 2 +- pkg/sentry/platform/kvm/bluepill_unsafe.go | 2 +- pkg/sentry/platform/kvm/machine_unsafe.go | 2 +- pkg/sentry/platform/ptrace/subprocess_unsafe.go | 2 +- pkg/sentry/vfs/mount_unsafe.go | 2 +- pkg/sleep/sleep_unsafe.go | 2 +- pkg/tcpip/link/rawfile/blockingpoll_yield_unsafe.go | 2 +- pkg/tcpip/time_unsafe.go | 2 +- third_party/gvsync/downgradable_rwmutex_1_13_unsafe.go | 2 +- third_party/gvsync/downgradable_rwmutex_unsafe.go | 2 +- third_party/gvsync/memmove_unsafe.go | 2 +- 12 files changed, 12 insertions(+), 12 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/procid/procid_amd64.s b/pkg/procid/procid_amd64.s index 30ec8e6e2..38cea9be3 100644 --- a/pkg/procid/procid_amd64.s +++ b/pkg/procid/procid_amd64.s @@ -14,7 +14,7 @@ // +build amd64 // +build go1.8 -// +build !go1.14 +// +build !go1.15 #include "textflag.h" diff --git a/pkg/procid/procid_arm64.s b/pkg/procid/procid_arm64.s index e340d9f98..4f4b70fef 100644 --- a/pkg/procid/procid_arm64.s +++ b/pkg/procid/procid_arm64.s @@ -14,7 +14,7 @@ // +build arm64 // +build go1.8 -// +build !go1.14 +// +build !go1.15 #include "textflag.h" diff --git a/pkg/sentry/platform/kvm/bluepill_unsafe.go b/pkg/sentry/platform/kvm/bluepill_unsafe.go index 7e8e9f42a..ee730ad70 100644 --- a/pkg/sentry/platform/kvm/bluepill_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_unsafe.go @@ -13,7 +13,7 @@ // limitations under the License. // +build go1.12 -// +build !go1.14 +// +build !go1.15 // Check go:linkname function signatures when updating Go version. diff --git a/pkg/sentry/platform/kvm/machine_unsafe.go b/pkg/sentry/platform/kvm/machine_unsafe.go index 405e00292..e00c7ae40 100644 --- a/pkg/sentry/platform/kvm/machine_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_unsafe.go @@ -13,7 +13,7 @@ // limitations under the License. // +build go1.12 -// +build !go1.14 +// +build !go1.15 // Check go:linkname function signatures when updating Go version. diff --git a/pkg/sentry/platform/ptrace/subprocess_unsafe.go b/pkg/sentry/platform/ptrace/subprocess_unsafe.go index b80a3604d..2ae6b9f9d 100644 --- a/pkg/sentry/platform/ptrace/subprocess_unsafe.go +++ b/pkg/sentry/platform/ptrace/subprocess_unsafe.go @@ -13,7 +13,7 @@ // limitations under the License. // +build go1.12 -// +build !go1.14 +// +build !go1.15 // Check go:linkname function signatures when updating Go version. diff --git a/pkg/sentry/vfs/mount_unsafe.go b/pkg/sentry/vfs/mount_unsafe.go index b0511aa40..75e6c7dfa 100644 --- a/pkg/sentry/vfs/mount_unsafe.go +++ b/pkg/sentry/vfs/mount_unsafe.go @@ -13,7 +13,7 @@ // limitations under the License. // +build go1.12 -// +build !go1.14 +// +build !go1.15 // Check go:linkname function signatures when updating Go version. diff --git a/pkg/sleep/sleep_unsafe.go b/pkg/sleep/sleep_unsafe.go index 8f5e60a25..acbf0229b 100644 --- a/pkg/sleep/sleep_unsafe.go +++ b/pkg/sleep/sleep_unsafe.go @@ -13,7 +13,7 @@ // limitations under the License. // +build go1.11 -// +build !go1.14 +// +build !go1.15 // Check go:linkname function signatures when updating Go version. diff --git a/pkg/tcpip/link/rawfile/blockingpoll_yield_unsafe.go b/pkg/tcpip/link/rawfile/blockingpoll_yield_unsafe.go index dda3b10a6..0b5a6cf49 100644 --- a/pkg/tcpip/link/rawfile/blockingpoll_yield_unsafe.go +++ b/pkg/tcpip/link/rawfile/blockingpoll_yield_unsafe.go @@ -14,7 +14,7 @@ // +build linux,amd64 linux,arm64 // +build go1.12 -// +build !go1.14 +// +build !go1.15 // Check go:linkname function signatures when updating Go version. diff --git a/pkg/tcpip/time_unsafe.go b/pkg/tcpip/time_unsafe.go index a52262e87..48764b978 100644 --- a/pkg/tcpip/time_unsafe.go +++ b/pkg/tcpip/time_unsafe.go @@ -13,7 +13,7 @@ // limitations under the License. // +build go1.9 -// +build !go1.14 +// +build !go1.15 // Check go:linkname function signatures when updating Go version. diff --git a/third_party/gvsync/downgradable_rwmutex_1_13_unsafe.go b/third_party/gvsync/downgradable_rwmutex_1_13_unsafe.go index 8baec5458..3b9346843 100644 --- a/third_party/gvsync/downgradable_rwmutex_1_13_unsafe.go +++ b/third_party/gvsync/downgradable_rwmutex_1_13_unsafe.go @@ -4,7 +4,7 @@ // license that can be found in the LICENSE file. // +build go1.13 -// +build !go1.14 +// +build !go1.15 // Check go:linkname function signatures when updating Go version. diff --git a/third_party/gvsync/downgradable_rwmutex_unsafe.go b/third_party/gvsync/downgradable_rwmutex_unsafe.go index 1f6007aa1..b7862d185 100644 --- a/third_party/gvsync/downgradable_rwmutex_unsafe.go +++ b/third_party/gvsync/downgradable_rwmutex_unsafe.go @@ -4,7 +4,7 @@ // license that can be found in the LICENSE file. // +build go1.12 -// +build !go1.14 +// +build !go1.15 // Check go:linkname function signatures when updating Go version. diff --git a/third_party/gvsync/memmove_unsafe.go b/third_party/gvsync/memmove_unsafe.go index 84b69f215..9dd1d6142 100644 --- a/third_party/gvsync/memmove_unsafe.go +++ b/third_party/gvsync/memmove_unsafe.go @@ -4,7 +4,7 @@ // license that can be found in the LICENSE file. // +build go1.12 -// +build !go1.14 +// +build !go1.15 // Check go:linkname function signatures when updating Go version. -- cgit v1.2.3 From 38330e93774e68324d8f43adb27178453dee18b6 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 29 Oct 2019 13:58:09 -0700 Subject: Update symlink traversal limit when resolving interpreter path. When execveat is called on an interpreter script, the symlink count for resolving the script path should be separate from the count for resolving the the corresponding interpreter. An ELOOP error should not occur if we do not hit the symlink limit along any individual path, even if the total number of symlinks encountered exceeds the limit. Closes #574 PiperOrigin-RevId: 277358474 --- pkg/sentry/loader/elf.go | 2 ++ pkg/sentry/loader/loader.go | 2 ++ test/syscalls/linux/exec.cc | 41 +++++++++++++++++++++++++++++++++++++++++ 3 files changed, 45 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/loader/elf.go b/pkg/sentry/loader/elf.go index 3ea037e4d..c2c3ec06e 100644 --- a/pkg/sentry/loader/elf.go +++ b/pkg/sentry/loader/elf.go @@ -644,6 +644,8 @@ func loadELF(ctx context.Context, args LoadArgs) (loadedELF, arch.Context, error // resolved, the interpreter should still be resolved if it is // a symlink. args.ResolveFinal = true + // Refresh the traversal limit. + *args.RemainingTraversals = linux.MaxSymlinkTraversals args.Filename = bin.interpreter d, i, err := openPath(ctx, args) if err != nil { diff --git a/pkg/sentry/loader/loader.go b/pkg/sentry/loader/loader.go index 803e7d41e..b03eeb005 100644 --- a/pkg/sentry/loader/loader.go +++ b/pkg/sentry/loader/loader.go @@ -293,6 +293,8 @@ func loadExecutable(ctx context.Context, args LoadArgs) (loadedELF, arch.Context ctx.Infof("Error loading interpreter script: %v", err) return loadedELF{}, nil, nil, nil, err } + // Refresh the traversal limit for the interpreter. + *args.RemainingTraversals = linux.MaxSymlinkTraversals default: ctx.Infof("Unknown magic: %v", hdr) return loadedELF{}, nil, nil, nil, syserror.ENOEXEC diff --git a/test/syscalls/linux/exec.cc b/test/syscalls/linux/exec.cc index a9067df2a..581f03533 100644 --- a/test/syscalls/linux/exec.cc +++ b/test/syscalls/linux/exec.cc @@ -533,6 +533,47 @@ TEST(ExecTest, CloexecEventfd) { W_EXITCODE(0, 0), ""); } +constexpr int kLinuxMaxSymlinks = 40; + +TEST(ExecTest, SymlinkLimitExceeded) { + std::string path = WorkloadPath(kBasicWorkload); + + // Hold onto TempPath objects so they are not destructed prematurely. + std::vector symlinks; + for (int i = 0; i < kLinuxMaxSymlinks + 1; i++) { + symlinks.push_back( + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateSymlinkTo("/tmp", path))); + path = symlinks[i].path(); + } + + int execve_errno; + ASSERT_NO_ERRNO_AND_VALUE( + ForkAndExec(path, {path}, {}, /*child=*/nullptr, &execve_errno)); + EXPECT_EQ(execve_errno, ELOOP); +} + +TEST(ExecTest, SymlinkLimitRefreshedForInterpreter) { + std::string tmp_dir = "/tmp"; + std::string interpreter_path = "/bin/echo"; + TempPath script = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileWith( + tmp_dir, absl::StrCat("#!", interpreter_path), 0755)); + std::string script_path = script.path(); + + // Hold onto TempPath objects so they are not destructed prematurely. + std::vector interpreter_symlinks; + std::vector script_symlinks; + for (int i = 0; i < kLinuxMaxSymlinks; i++) { + interpreter_symlinks.push_back(ASSERT_NO_ERRNO_AND_VALUE( + TempPath::CreateSymlinkTo(tmp_dir, interpreter_path))); + interpreter_path = interpreter_symlinks[i].path(); + script_symlinks.push_back(ASSERT_NO_ERRNO_AND_VALUE( + TempPath::CreateSymlinkTo(tmp_dir, script_path))); + script_path = script_symlinks[i].path(); + } + + CheckExec(script_path, {script_path}, {}, ArgEnvExitStatus(0, 0), ""); +} + TEST(ExecveatTest, BasicWithFDCWD) { std::string path = WorkloadPath(kBasicWorkload); CheckExecveat(AT_FDCWD, path, {path}, {}, /*flags=*/0, ArgEnvExitStatus(0, 0), -- cgit v1.2.3 From d7f5e823e24501c33a377ee6c73210b00bf3d89f Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 29 Oct 2019 13:58:20 -0700 Subject: Fix grammar in comment. Missing "for". PiperOrigin-RevId: 277358513 --- pkg/sentry/kernel/task.go | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index 11a8c6c87..9be3dae3c 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -709,7 +709,7 @@ func (t *Task) FDTable() *FDTable { return t.fdTable } -// GetFile is a convenience wrapper t.FDTable().Get. +// GetFile is a convenience wrapper for t.FDTable().Get. // // Precondition: same as FDTable.Get. func (t *Task) GetFile(fd int32) *fs.File { -- cgit v1.2.3 From a2c51efe3669f0380042b2375eae79e403d3680c Mon Sep 17 00:00:00 2001 From: Ian Gudger Date: Tue, 29 Oct 2019 16:13:43 -0700 Subject: Add endpoint tracking to the stack. In the future this will replace DanglingEndpoints. DanglingEndpoints must be kept for now due to issues with save/restore. This is arguably a cleaner design and allows the stack to know which transport endpoints might still be using its link endpoints. Updates #837 PiperOrigin-RevId: 277386633 --- pkg/sentry/inet/BUILD | 5 ++- pkg/sentry/inet/inet.go | 12 ++++++ pkg/sentry/inet/test_stack.go | 16 +++++++- pkg/sentry/socket/hostinet/BUILD | 1 + pkg/sentry/socket/hostinet/stack.go | 10 +++++ pkg/sentry/socket/netstack/stack.go | 15 +++++++ pkg/sentry/socket/rpcinet/BUILD | 1 + pkg/sentry/socket/rpcinet/stack.go | 10 +++++ pkg/tcpip/stack/stack.go | 59 ++++++++++++++++++++++++++-- pkg/tcpip/stack/transport_demuxer.go | 65 +++++++++++++++++++++++-------- pkg/tcpip/stack/transport_test.go | 3 +- pkg/tcpip/transport/tcp/endpoint.go | 5 ++- pkg/tcpip/transport/tcp/endpoint_state.go | 7 +++- 13 files changed, 182 insertions(+), 27 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/inet/BUILD b/pkg/sentry/inet/BUILD index d5284f0d9..8d60ad4ad 100644 --- a/pkg/sentry/inet/BUILD +++ b/pkg/sentry/inet/BUILD @@ -13,5 +13,8 @@ go_library( "test_stack.go", ], importpath = "gvisor.dev/gvisor/pkg/sentry/inet", - deps = ["//pkg/sentry/context"], + deps = [ + "//pkg/sentry/context", + "//pkg/tcpip/stack", + ], ) diff --git a/pkg/sentry/inet/inet.go b/pkg/sentry/inet/inet.go index bc6cb1095..a7dfb78a7 100644 --- a/pkg/sentry/inet/inet.go +++ b/pkg/sentry/inet/inet.go @@ -15,6 +15,8 @@ // Package inet defines semantics for IP stacks. package inet +import "gvisor.dev/gvisor/pkg/tcpip/stack" + // Stack represents a TCP/IP stack. type Stack interface { // Interfaces returns all network interfaces as a mapping from interface @@ -58,6 +60,16 @@ type Stack interface { // Resume restarts the network stack after restore. Resume() + + // RegisteredEndpoints returns all endpoints which are currently registered. + RegisteredEndpoints() []stack.TransportEndpoint + + // CleanupEndpoints returns endpoints currently in the cleanup state. + CleanupEndpoints() []stack.TransportEndpoint + + // RestoreCleanupEndpoints adds endpoints to cleanup tracking. This is useful + // for restoring a stack after a save. + RestoreCleanupEndpoints([]stack.TransportEndpoint) } // Interface contains information about a network interface. diff --git a/pkg/sentry/inet/test_stack.go b/pkg/sentry/inet/test_stack.go index b9eed7c3a..dcfcbd97e 100644 --- a/pkg/sentry/inet/test_stack.go +++ b/pkg/sentry/inet/test_stack.go @@ -14,6 +14,8 @@ package inet +import "gvisor.dev/gvisor/pkg/tcpip/stack" + // TestStack is a dummy implementation of Stack for tests. type TestStack struct { InterfacesMap map[int32]Interface @@ -94,5 +96,17 @@ func (s *TestStack) RouteTable() []Route { } // Resume implements Stack.Resume. -func (s *TestStack) Resume() { +func (s *TestStack) Resume() {} + +// RegisteredEndpoints implements inet.Stack.RegisteredEndpoints. +func (s *TestStack) RegisteredEndpoints() []stack.TransportEndpoint { + return nil } + +// CleanupEndpoints implements inet.Stack.CleanupEndpoints. +func (s *TestStack) CleanupEndpoints() []stack.TransportEndpoint { + return nil +} + +// RestoreCleanupEndpoints implements inet.Stack.RestoreCleanupEndpoints. +func (s *TestStack) RestoreCleanupEndpoints([]stack.TransportEndpoint) {} diff --git a/pkg/sentry/socket/hostinet/BUILD b/pkg/sentry/socket/hostinet/BUILD index 4d174dda4..8b66a719d 100644 --- a/pkg/sentry/socket/hostinet/BUILD +++ b/pkg/sentry/socket/hostinet/BUILD @@ -32,6 +32,7 @@ go_library( "//pkg/sentry/usermem", "//pkg/syserr", "//pkg/syserror", + "//pkg/tcpip/stack", "//pkg/waiter", ], ) diff --git a/pkg/sentry/socket/hostinet/stack.go b/pkg/sentry/socket/hostinet/stack.go index d4387f5d4..e67b46c9e 100644 --- a/pkg/sentry/socket/hostinet/stack.go +++ b/pkg/sentry/socket/hostinet/stack.go @@ -31,6 +31,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/tcpip/stack" ) var defaultRecvBufSize = inet.TCPBufferSize{ @@ -442,3 +443,12 @@ func (s *Stack) RouteTable() []inet.Route { // Resume implements inet.Stack.Resume. func (s *Stack) Resume() {} + +// RegisteredEndpoints implements inet.Stack.RegisteredEndpoints. +func (s *Stack) RegisteredEndpoints() []stack.TransportEndpoint { return nil } + +// CleanupEndpoints implements inet.Stack.CleanupEndpoints. +func (s *Stack) CleanupEndpoints() []stack.TransportEndpoint { return nil } + +// RestoreCleanupEndpoints implements inet.Stack.RestoreCleanupEndpoints. +func (s *Stack) RestoreCleanupEndpoints([]stack.TransportEndpoint) {} diff --git a/pkg/sentry/socket/netstack/stack.go b/pkg/sentry/socket/netstack/stack.go index d5db8c17c..a0db2d4fd 100644 --- a/pkg/sentry/socket/netstack/stack.go +++ b/pkg/sentry/socket/netstack/stack.go @@ -291,3 +291,18 @@ func (s *Stack) FillDefaultIPTables() { func (s *Stack) Resume() { s.Stack.Resume() } + +// RegisteredEndpoints implements inet.Stack.RegisteredEndpoints. +func (s *Stack) RegisteredEndpoints() []stack.TransportEndpoint { + return s.Stack.RegisteredEndpoints() +} + +// CleanupEndpoints implements inet.Stack.CleanupEndpoints. +func (s *Stack) CleanupEndpoints() []stack.TransportEndpoint { + return s.Stack.CleanupEndpoints() +} + +// RestoreCleanupEndpoints implements inet.Stack.RestoreCleanupEndpoints. +func (s *Stack) RestoreCleanupEndpoints(es []stack.TransportEndpoint) { + s.Stack.RestoreCleanupEndpoints(es) +} diff --git a/pkg/sentry/socket/rpcinet/BUILD b/pkg/sentry/socket/rpcinet/BUILD index 3a6baa308..4668b87d1 100644 --- a/pkg/sentry/socket/rpcinet/BUILD +++ b/pkg/sentry/socket/rpcinet/BUILD @@ -37,6 +37,7 @@ go_library( "//pkg/syserror", "//pkg/tcpip", "//pkg/tcpip/buffer", + "//pkg/tcpip/stack", "//pkg/unet", "//pkg/waiter", ], diff --git a/pkg/sentry/socket/rpcinet/stack.go b/pkg/sentry/socket/rpcinet/stack.go index 5dcb6b455..f7878a760 100644 --- a/pkg/sentry/socket/rpcinet/stack.go +++ b/pkg/sentry/socket/rpcinet/stack.go @@ -23,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/conn" "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/notifier" "gvisor.dev/gvisor/pkg/syserr" + "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/unet" ) @@ -165,3 +166,12 @@ func (s *Stack) RouteTable() []inet.Route { // Resume implements inet.Stack.Resume. func (s *Stack) Resume() {} + +// RegisteredEndpoints implements inet.Stack.RegisteredEndpoints. +func (s *Stack) RegisteredEndpoints() []stack.TransportEndpoint { return nil } + +// CleanupEndpoints implements inet.Stack.CleanupEndpoints. +func (s *Stack) CleanupEndpoints() []stack.TransportEndpoint { return nil } + +// RestoreCleanupEndpoints implements inet.Stack.RestoreCleanupEndpoints. +func (s *Stack) RestoreCleanupEndpoints([]stack.TransportEndpoint) {} diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index 242d2150c..360c54b2d 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -361,9 +361,10 @@ type Stack struct { linkAddrCache *linkAddrCache - mu sync.RWMutex - nics map[tcpip.NICID]*NIC - forwarding bool + mu sync.RWMutex + nics map[tcpip.NICID]*NIC + forwarding bool + cleanupEndpoints map[TransportEndpoint]struct{} // route is the route table passed in by the user via SetRouteTable(), // it is used by FindRoute() to build a route for a specific @@ -513,6 +514,7 @@ func New(opts Options) *Stack { networkProtocols: make(map[tcpip.NetworkProtocolNumber]NetworkProtocol), linkAddrResolvers: make(map[tcpip.NetworkProtocolNumber]LinkAddressResolver), nics: make(map[tcpip.NICID]*NIC), + cleanupEndpoints: make(map[TransportEndpoint]struct{}), linkAddrCache: newLinkAddrCache(ageLimit, resolutionTimeout, resolutionAttempts), PortManager: ports.NewPortManager(), clock: clock, @@ -1136,6 +1138,25 @@ func (s *Stack) UnregisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip s.demux.unregisterEndpoint(netProtos, protocol, id, ep, bindToDevice) } +// StartTransportEndpointCleanup removes the endpoint with the given id from +// the stack transport dispatcher. It also transitions it to the cleanup stage. +func (s *Stack) StartTransportEndpointCleanup(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, bindToDevice tcpip.NICID) { + s.mu.Lock() + defer s.mu.Unlock() + + s.cleanupEndpoints[ep] = struct{}{} + + s.demux.unregisterEndpoint(netProtos, protocol, id, ep, bindToDevice) +} + +// CompleteTransportEndpointCleanup removes the endpoint from the cleanup +// stage. +func (s *Stack) CompleteTransportEndpointCleanup(ep TransportEndpoint) { + s.mu.Lock() + delete(s.cleanupEndpoints, ep) + s.mu.Unlock() +} + // RegisterRawTransportEndpoint registers the given endpoint with the stack // transport dispatcher. Received packets that match the provided transport // protocol will be delivered to the given endpoint. @@ -1157,6 +1178,38 @@ func (s *Stack) RegisterRestoredEndpoint(e ResumableEndpoint) { s.mu.Unlock() } +// RegisteredEndpoints returns all endpoints which are currently registered. +func (s *Stack) RegisteredEndpoints() []TransportEndpoint { + s.mu.Lock() + defer s.mu.Unlock() + var es []TransportEndpoint + for _, e := range s.demux.protocol { + es = append(es, e.transportEndpoints()...) + } + return es +} + +// CleanupEndpoints returns endpoints currently in the cleanup state. +func (s *Stack) CleanupEndpoints() []TransportEndpoint { + s.mu.Lock() + es := make([]TransportEndpoint, 0, len(s.cleanupEndpoints)) + for e := range s.cleanupEndpoints { + es = append(es, e) + } + s.mu.Unlock() + return es +} + +// RestoreCleanupEndpoints adds endpoints to cleanup tracking. This is useful +// for restoring a stack after a save. +func (s *Stack) RestoreCleanupEndpoints(es []TransportEndpoint) { + s.mu.Lock() + for _, e := range es { + s.cleanupEndpoints[e] = struct{}{} + } + s.mu.Unlock() +} + // Resume restarts the stack after a restore. This must be called after the // entire system has been restored. func (s *Stack) Resume() { diff --git a/pkg/tcpip/stack/transport_demuxer.go b/pkg/tcpip/stack/transport_demuxer.go index 9aff90a3d..f633632f0 100644 --- a/pkg/tcpip/stack/transport_demuxer.go +++ b/pkg/tcpip/stack/transport_demuxer.go @@ -41,6 +41,31 @@ type transportEndpoints struct { rawEndpoints []RawTransportEndpoint } +// unregisterEndpoint unregisters the endpoint with the given id such that it +// won't receive any more packets. +func (eps *transportEndpoints) unregisterEndpoint(id TransportEndpointID, ep TransportEndpoint, bindToDevice tcpip.NICID) { + eps.mu.Lock() + defer eps.mu.Unlock() + epsByNic, ok := eps.endpoints[id] + if !ok { + return + } + if !epsByNic.unregisterEndpoint(bindToDevice, ep) { + return + } + delete(eps.endpoints, id) +} + +func (eps *transportEndpoints) transportEndpoints() []TransportEndpoint { + eps.mu.RLock() + defer eps.mu.RUnlock() + es := make([]TransportEndpoint, 0, len(eps.endpoints)) + for _, e := range eps.endpoints { + es = append(es, e.transportEndpoints()...) + } + return es +} + type endpointsByNic struct { mu sync.RWMutex endpoints map[tcpip.NICID]*multiPortEndpoint @@ -48,6 +73,16 @@ type endpointsByNic struct { seed uint32 } +func (epsByNic *endpointsByNic) transportEndpoints() []TransportEndpoint { + epsByNic.mu.RLock() + defer epsByNic.mu.RUnlock() + var eps []TransportEndpoint + for _, ep := range epsByNic.endpoints { + eps = append(eps, ep.transportEndpoints()...) + } + return eps +} + // HandlePacket is called by the stack when new packets arrive to this transport // endpoint. func (epsByNic *endpointsByNic) handlePacket(r *Route, id TransportEndpointID, vv buffer.VectorisedView) { @@ -127,21 +162,6 @@ func (epsByNic *endpointsByNic) unregisterEndpoint(bindToDevice tcpip.NICID, t T return len(epsByNic.endpoints) == 0 } -// unregisterEndpoint unregisters the endpoint with the given id such that it -// won't receive any more packets. -func (eps *transportEndpoints) unregisterEndpoint(id TransportEndpointID, ep TransportEndpoint, bindToDevice tcpip.NICID) { - eps.mu.Lock() - defer eps.mu.Unlock() - epsByNic, ok := eps.endpoints[id] - if !ok { - return - } - if !epsByNic.unregisterEndpoint(bindToDevice, ep) { - return - } - delete(eps.endpoints, id) -} - // transportDemuxer demultiplexes packets targeted at a transport endpoint // (i.e., after they've been parsed by the network layer). It does two levels // of demultiplexing: first based on the network and transport protocols, then @@ -183,14 +203,27 @@ func (d *transportDemuxer) registerEndpoint(netProtos []tcpip.NetworkProtocolNum // multiPortEndpoint is a container for TransportEndpoints which are bound to // the same pair of address and port. endpointsArr always has at least one // element. +// +// FIXME(gvisor.dev/issue/873): Restore this properly. Currently, we just save +// this to ensure that the underlying endpoints get saved/restored, but not not +// use the restored copy. +// +// +stateify savable type multiPortEndpoint struct { - mu sync.RWMutex + mu sync.RWMutex `state:"nosave"` endpointsArr []TransportEndpoint endpointsMap map[TransportEndpoint]int // reuse indicates if more than one endpoint is allowed. reuse bool } +func (ep *multiPortEndpoint) transportEndpoints() []TransportEndpoint { + ep.mu.RLock() + eps := append([]TransportEndpoint(nil), ep.endpointsArr...) + ep.mu.RUnlock() + return eps +} + // reciprocalScale scales a value into range [0, n). // // This is similar to val % n, but faster. diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go index db951c9ce..ae6fda3a9 100644 --- a/pkg/tcpip/stack/transport_test.go +++ b/pkg/tcpip/stack/transport_test.go @@ -218,8 +218,7 @@ func (f *fakeTransportEndpoint) State() uint32 { return 0 } -func (f *fakeTransportEndpoint) ModerateRecvBuf(copied int) { -} +func (f *fakeTransportEndpoint) ModerateRecvBuf(copied int) {} func (f *fakeTransportEndpoint) IPTables() (iptables.IPTables, error) { return iptables.IPTables{}, nil diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index ce8307cee..8a3ca0f1b 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -686,7 +686,7 @@ func (e *endpoint) Close() { // in Listen() when trying to register. if e.state == StateListen && e.isPortReserved { if e.isRegistered { - e.stack.UnregisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.bindToDevice) + e.stack.StartTransportEndpointCleanup(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.bindToDevice) e.isRegistered = false } @@ -747,7 +747,7 @@ func (e *endpoint) cleanupLocked() { e.workerCleanup = false if e.isRegistered { - e.stack.UnregisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.bindToDevice) + e.stack.StartTransportEndpointCleanup(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.bindToDevice) e.isRegistered = false } @@ -757,6 +757,7 @@ func (e *endpoint) cleanupLocked() { } e.route.Release() + e.stack.CompleteTransportEndpointCleanup(e) tcpip.DeleteDanglingEndpoint(e) } diff --git a/pkg/tcpip/transport/tcp/endpoint_state.go b/pkg/tcpip/transport/tcp/endpoint_state.go index eae17237e..19f003b6b 100644 --- a/pkg/tcpip/transport/tcp/endpoint_state.go +++ b/pkg/tcpip/transport/tcp/endpoint_state.go @@ -193,8 +193,10 @@ func (e *endpoint) Resume(s *stack.Stack) { if len(e.BindAddr) == 0 { e.BindAddr = e.ID.LocalAddress } - if err := e.Bind(tcpip.FullAddress{Addr: e.BindAddr, Port: e.ID.LocalPort}); err != nil { - panic("endpoint binding failed: " + err.String()) + addr := e.BindAddr + port := e.ID.LocalPort + if err := e.Bind(tcpip.FullAddress{Addr: addr, Port: port}); err != nil { + panic(fmt.Sprintf("endpoint binding [%v]:%d failed: %v", addr, port, err)) } } @@ -265,6 +267,7 @@ func (e *endpoint) Resume(s *stack.Stack) { } fallthrough case StateError: + e.stack.CompleteTransportEndpointCleanup(e) tcpip.DeleteDanglingEndpoint(e) } } -- cgit v1.2.3 From ca933329fa46ce219b39f4cf8cba1754b36cc2c2 Mon Sep 17 00:00:00 2001 From: lubinszARM <34124929+lubinszARM@users.noreply.github.com> Date: Wed, 30 Oct 2019 15:51:42 -0700 Subject: support using KVM_MEM_READONLY for arm64 regions On Arm platform, "setMemoryRegion" has extra permission checks. In virt/kvm/arm/mmu.c: kvm_arch_prepare_memory_region() .... if (writable && !(vma->vm_flags & VM_WRITE)) { ret = -EPERM; break; } .... So, for Arm platform, the "flags" for kvm_memory_region is required. And on x86 platform, the "flags" can be always set as '0'. Signed-off-by: Bin Lu COPYBARA_INTEGRATE_REVIEW=https://github.com/google/gvisor/pull/810 from lubinszARM:pr_setregion 8c99b19cfb0c859c6630a1cfff951db65fcf87ac PiperOrigin-RevId: 277602603 --- pkg/sentry/platform/kvm/BUILD | 1 + pkg/sentry/platform/kvm/address_space.go | 2 +- pkg/sentry/platform/kvm/allocator.go | 2 +- pkg/sentry/platform/kvm/bluepill_fault.go | 10 ++-- pkg/sentry/platform/kvm/bluepill_unsafe.go | 2 +- pkg/sentry/platform/kvm/kvm_const.go | 7 +++ pkg/sentry/platform/kvm/machine.go | 26 +++++++++-- pkg/sentry/platform/kvm/machine_amd64.go | 10 ++++ pkg/sentry/platform/kvm/machine_amd64_unsafe.go | 24 ---------- pkg/sentry/platform/kvm/machine_arm64.go | 61 +++++++++++++++++++++++++ pkg/sentry/platform/kvm/machine_unsafe.go | 24 ++++++++++ 11 files changed, 133 insertions(+), 36 deletions(-) create mode 100644 pkg/sentry/platform/kvm/machine_arm64.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/kvm/BUILD b/pkg/sentry/platform/kvm/BUILD index 31fa48ec5..6803d488c 100644 --- a/pkg/sentry/platform/kvm/BUILD +++ b/pkg/sentry/platform/kvm/BUILD @@ -23,6 +23,7 @@ go_library( "machine.go", "machine_amd64.go", "machine_amd64_unsafe.go", + "machine_arm64.go", "machine_unsafe.go", "physical_map.go", "virtual_map.go", diff --git a/pkg/sentry/platform/kvm/address_space.go b/pkg/sentry/platform/kvm/address_space.go index acd41f73d..ea8b9632e 100644 --- a/pkg/sentry/platform/kvm/address_space.go +++ b/pkg/sentry/platform/kvm/address_space.go @@ -127,7 +127,7 @@ func (as *addressSpace) mapHost(addr usermem.Addr, m hostMapEntry, at usermem.Ac // not have physical mappings, the KVM module may inject // spurious exceptions when emulation fails (i.e. it tries to // emulate because the RIP is pointed at those pages). - as.machine.mapPhysical(physical, length) + as.machine.mapPhysical(physical, length, physicalRegions, _KVM_MEM_FLAGS_NONE) // Install the page table mappings. Note that the ordering is // important; if the pagetable mappings were installed before diff --git a/pkg/sentry/platform/kvm/allocator.go b/pkg/sentry/platform/kvm/allocator.go index 80942e9c9..3f35414bb 100644 --- a/pkg/sentry/platform/kvm/allocator.go +++ b/pkg/sentry/platform/kvm/allocator.go @@ -54,7 +54,7 @@ func (a allocator) PhysicalFor(ptes *pagetables.PTEs) uintptr { // //go:nosplit func (a allocator) LookupPTEs(physical uintptr) *pagetables.PTEs { - virtualStart, physicalStart, _, ok := calculateBluepillFault(physical) + virtualStart, physicalStart, _, ok := calculateBluepillFault(physical, physicalRegions) if !ok { panic(fmt.Sprintf("LookupPTEs failed for 0x%x", physical)) } diff --git a/pkg/sentry/platform/kvm/bluepill_fault.go b/pkg/sentry/platform/kvm/bluepill_fault.go index b97476053..f6459cda9 100644 --- a/pkg/sentry/platform/kvm/bluepill_fault.go +++ b/pkg/sentry/platform/kvm/bluepill_fault.go @@ -46,9 +46,9 @@ func yield() { // calculateBluepillFault calculates the fault address range. // //go:nosplit -func calculateBluepillFault(physical uintptr) (virtualStart, physicalStart, length uintptr, ok bool) { +func calculateBluepillFault(physical uintptr, phyRegions []physicalRegion) (virtualStart, physicalStart, length uintptr, ok bool) { alignedPhysical := physical &^ uintptr(usermem.PageSize-1) - for _, pr := range physicalRegions { + for _, pr := range phyRegions { end := pr.physical + pr.length if physical < pr.physical || physical >= end { continue @@ -77,12 +77,12 @@ func calculateBluepillFault(physical uintptr) (virtualStart, physicalStart, leng // The corresponding virtual address is returned. This may throw on error. // //go:nosplit -func handleBluepillFault(m *machine, physical uintptr) (uintptr, bool) { +func handleBluepillFault(m *machine, physical uintptr, phyRegions []physicalRegion, flags uint32) (uintptr, bool) { // Paging fault: we need to map the underlying physical pages for this // fault. This all has to be done in this function because we're in a // signal handler context. (We can't call any functions that might // split the stack.) - virtualStart, physicalStart, length, ok := calculateBluepillFault(physical) + virtualStart, physicalStart, length, ok := calculateBluepillFault(physical, phyRegions) if !ok { return 0, false } @@ -96,7 +96,7 @@ func handleBluepillFault(m *machine, physical uintptr) (uintptr, bool) { yield() // Race with another call. slot = atomic.SwapUint32(&m.nextSlot, ^uint32(0)) } - errno := m.setMemoryRegion(int(slot), physicalStart, length, virtualStart) + errno := m.setMemoryRegion(int(slot), physicalStart, length, virtualStart, flags) if errno == 0 { // Successfully added region; we can increment nextSlot and // allow another set to proceed here. diff --git a/pkg/sentry/platform/kvm/bluepill_unsafe.go b/pkg/sentry/platform/kvm/bluepill_unsafe.go index ee730ad70..3734bfb7a 100644 --- a/pkg/sentry/platform/kvm/bluepill_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_unsafe.go @@ -162,7 +162,7 @@ func bluepillHandler(context unsafe.Pointer) { // For MMIO, the physical address is the first data item. physical := uintptr(c.runData.data[0]) - virtual, ok := handleBluepillFault(c.machine, physical) + virtual, ok := handleBluepillFault(c.machine, physical, physicalRegions, _KVM_MEM_FLAGS_NONE) if !ok { c.die(bluepillArchContext(context), "invalid physical address") return diff --git a/pkg/sentry/platform/kvm/kvm_const.go b/pkg/sentry/platform/kvm/kvm_const.go index d05f05c29..766131d60 100644 --- a/pkg/sentry/platform/kvm/kvm_const.go +++ b/pkg/sentry/platform/kvm/kvm_const.go @@ -62,3 +62,10 @@ const ( _KVM_NR_INTERRUPTS = 0x100 _KVM_NR_CPUID_ENTRIES = 0x100 ) + +// KVM kvm_memory_region::flags. +const ( + _KVM_MEM_LOG_DIRTY_PAGES = uint32(1) << 0 + _KVM_MEM_READONLY = uint32(1) << 1 + _KVM_MEM_FLAGS_NONE = 0 +) diff --git a/pkg/sentry/platform/kvm/machine.go b/pkg/sentry/platform/kvm/machine.go index cc6c138b2..7d02ebf19 100644 --- a/pkg/sentry/platform/kvm/machine.go +++ b/pkg/sentry/platform/kvm/machine.go @@ -215,6 +215,17 @@ func newMachine(vm int) (*machine, error) { return true // Keep iterating. }) + var physicalRegionsReadOnly []physicalRegion + var physicalRegionsAvailable []physicalRegion + + physicalRegionsReadOnly = rdonlyRegionsForSetMem() + physicalRegionsAvailable = availableRegionsForSetMem() + + // Map all read-only regions. + for _, r := range physicalRegionsReadOnly { + m.mapPhysical(r.physical, r.length, physicalRegionsReadOnly, _KVM_MEM_READONLY) + } + // Ensure that the currently mapped virtual regions are actually // available in the VM. Note that this doesn't guarantee no future // faults, however it should guarantee that everything is available to @@ -223,6 +234,13 @@ func newMachine(vm int) (*machine, error) { if excludeVirtualRegion(vr) { return // skip region. } + + for _, r := range physicalRegionsReadOnly { + if vr.virtual == r.virtual { + return + } + } + for virtual := vr.virtual; virtual < vr.virtual+vr.length; { physical, length, ok := translateToPhysical(virtual) if !ok { @@ -236,7 +254,7 @@ func newMachine(vm int) (*machine, error) { } // Ensure the physical range is mapped. - m.mapPhysical(physical, length) + m.mapPhysical(physical, length, physicalRegionsAvailable, _KVM_MEM_FLAGS_NONE) virtual += length } }) @@ -256,9 +274,9 @@ func newMachine(vm int) (*machine, error) { // not available. This attempts to be efficient for calls in the hot path. // // This panics on error. -func (m *machine) mapPhysical(physical, length uintptr) { +func (m *machine) mapPhysical(physical, length uintptr, phyRegions []physicalRegion, flags uint32) { for end := physical + length; physical < end; { - _, physicalStart, length, ok := calculateBluepillFault(physical) + _, physicalStart, length, ok := calculateBluepillFault(physical, phyRegions) if !ok { // Should never happen. panic("mapPhysical on unknown physical address") @@ -266,7 +284,7 @@ func (m *machine) mapPhysical(physical, length uintptr) { if _, ok := m.mappingCache.LoadOrStore(physicalStart, true); !ok { // Not present in the cache; requires setting the slot. - if _, ok := handleBluepillFault(m, physical); !ok { + if _, ok := handleBluepillFault(m, physical, phyRegions, flags); !ok { panic("handleBluepillFault failed") } } diff --git a/pkg/sentry/platform/kvm/machine_amd64.go b/pkg/sentry/platform/kvm/machine_amd64.go index c1cbe33be..b99fe425e 100644 --- a/pkg/sentry/platform/kvm/machine_amd64.go +++ b/pkg/sentry/platform/kvm/machine_amd64.go @@ -355,3 +355,13 @@ func (m *machine) retryInGuest(fn func()) { } } } + +// On x86 platform, the flags for "setMemoryRegion" can always be set as 0. +// There is no need to return read-only physicalRegions. +func rdonlyRegionsForSetMem() (phyRegions []physicalRegion) { + return nil +} + +func availableRegionsForSetMem() (phyRegions []physicalRegion) { + return physicalRegions +} diff --git a/pkg/sentry/platform/kvm/machine_amd64_unsafe.go b/pkg/sentry/platform/kvm/machine_amd64_unsafe.go index 506ec9af1..61227cafb 100644 --- a/pkg/sentry/platform/kvm/machine_amd64_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_amd64_unsafe.go @@ -26,30 +26,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/time" ) -// setMemoryRegion initializes a region. -// -// This may be called from bluepillHandler, and therefore returns an errno -// directly (instead of wrapping in an error) to avoid allocations. -// -//go:nosplit -func (m *machine) setMemoryRegion(slot int, physical, length, virtual uintptr) syscall.Errno { - userRegion := userMemoryRegion{ - slot: uint32(slot), - flags: 0, - guestPhysAddr: uint64(physical), - memorySize: uint64(length), - userspaceAddr: uint64(virtual), - } - - // Set the region. - _, _, errno := syscall.RawSyscall( - syscall.SYS_IOCTL, - uintptr(m.fd), - _KVM_SET_USER_MEMORY_REGION, - uintptr(unsafe.Pointer(&userRegion))) - return errno -} - // loadSegments copies the current segments. // // This may be called from within the signal context and throws on error. diff --git a/pkg/sentry/platform/kvm/machine_arm64.go b/pkg/sentry/platform/kvm/machine_arm64.go new file mode 100644 index 000000000..b7e2cfb9d --- /dev/null +++ b/pkg/sentry/platform/kvm/machine_arm64.go @@ -0,0 +1,61 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package kvm + +// Get all read-only physicalRegions. +func rdonlyRegionsForSetMem() (phyRegions []physicalRegion) { + var rdonlyRegions []region + + applyVirtualRegions(func(vr virtualRegion) { + if excludeVirtualRegion(vr) { + return + } + + if !vr.accessType.Write && vr.accessType.Read { + rdonlyRegions = append(rdonlyRegions, vr.region) + } + }) + + for _, r := range rdonlyRegions { + physical, _, ok := translateToPhysical(r.virtual) + if !ok { + continue + } + + phyRegions = append(phyRegions, physicalRegion{ + region: region{ + virtual: r.virtual, + length: r.length, + }, + physical: physical, + }) + } + + return phyRegions +} + +// Get all available physicalRegions. +func availableRegionsForSetMem() (phyRegions []physicalRegion) { + var excludeRegions []region + applyVirtualRegions(func(vr virtualRegion) { + if !vr.accessType.Write { + excludeRegions = append(excludeRegions, vr.region) + } + }) + + phyRegions = computePhysicalRegions(excludeRegions) + + return phyRegions +} diff --git a/pkg/sentry/platform/kvm/machine_unsafe.go b/pkg/sentry/platform/kvm/machine_unsafe.go index e00c7ae40..ed9433311 100644 --- a/pkg/sentry/platform/kvm/machine_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_unsafe.go @@ -35,6 +35,30 @@ func entersyscall() //go:linkname exitsyscall runtime.exitsyscall func exitsyscall() +// setMemoryRegion initializes a region. +// +// This may be called from bluepillHandler, and therefore returns an errno +// directly (instead of wrapping in an error) to avoid allocations. +// +//go:nosplit +func (m *machine) setMemoryRegion(slot int, physical, length, virtual uintptr, flags uint32) syscall.Errno { + userRegion := userMemoryRegion{ + slot: uint32(slot), + flags: uint32(flags), + guestPhysAddr: uint64(physical), + memorySize: uint64(length), + userspaceAddr: uint64(virtual), + } + + // Set the region. + _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(m.fd), + _KVM_SET_USER_MEMORY_REGION, + uintptr(unsafe.Pointer(&userRegion))) + return errno +} + // mapRunData maps the vCPU run data. func mapRunData(fd int) (*runData, error) { r, _, errno := syscall.RawSyscall6( -- cgit v1.2.3 From f7dbddaf77a6059c2f5a441d068a39219fe593bd Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Thu, 31 Oct 2019 12:27:46 -0700 Subject: platform/kvm: calll sigtimedwait with zero timeout sigtimedwait is used to check pending signals and it should not block. PiperOrigin-RevId: 277777269 --- pkg/sentry/platform/kvm/bluepill_unsafe.go | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/kvm/bluepill_unsafe.go b/pkg/sentry/platform/kvm/bluepill_unsafe.go index 3734bfb7a..ca011ef78 100644 --- a/pkg/sentry/platform/kvm/bluepill_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_unsafe.go @@ -80,13 +80,17 @@ func bluepillHandler(context unsafe.Pointer) { // interrupted KVM. Since we're in a signal handler // currently, all signals are masked and the signal // must have been delivered directly to this thread. + timeout := syscall.Timespec{} sig, _, errno := syscall.RawSyscall6( syscall.SYS_RT_SIGTIMEDWAIT, uintptr(unsafe.Pointer(&bounceSignalMask)), - 0, // siginfo. - 0, // timeout. - 8, // sigset size. + 0, // siginfo. + uintptr(unsafe.Pointer(&timeout)), // timeout. + 8, // sigset size. 0, 0) + if errno == syscall.EAGAIN { + continue + } if errno != 0 { throw("error waiting for pending signal") } -- cgit v1.2.3 From a99d3479a84ca86843e500dbdf58db0af389b536 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Thu, 31 Oct 2019 18:02:04 -0700 Subject: Add context to state. PiperOrigin-RevId: 277840416 --- pkg/sentry/context/context.go | 63 +++++++++++++++++++++++--------------- pkg/sentry/kernel/context.go | 32 +++++++++++++++++++ pkg/sentry/kernel/kernel.go | 13 ++++---- pkg/sentry/pgalloc/save_restore.go | 13 ++++---- pkg/state/decode.go | 4 +++ pkg/state/encode.go | 4 +++ pkg/state/map.go | 11 +++++++ pkg/state/state.go | 7 +++-- pkg/state/state_test.go | 11 ++++--- 9 files changed, 115 insertions(+), 43 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/context/context.go b/pkg/sentry/context/context.go index dfd62cbdb..23e009ef3 100644 --- a/pkg/sentry/context/context.go +++ b/pkg/sentry/context/context.go @@ -12,10 +12,20 @@ // See the License for the specific language governing permissions and // limitations under the License. -// Package context defines the sentry's Context type. +// Package context defines an internal context type. +// +// The given Context conforms to the standard Go context, but mandates +// additional methods that are specific to the kernel internals. Note however, +// that the Context described by this package carries additional constraints +// regarding concurrent access and retaining beyond the scope of a call. +// +// See the Context type for complete details. package context import ( + "context" + "time" + "gvisor.dev/gvisor/pkg/amutex" "gvisor.dev/gvisor/pkg/log" ) @@ -59,6 +69,7 @@ func ThreadGroupIDFromContext(ctx Context) (tgid int32, ok bool) { type Context interface { log.Logger amutex.Sleeper + context.Context // UninterruptibleSleepStart indicates the beginning of an uninterruptible // sleep state (equivalent to Linux's TASK_UNINTERRUPTIBLE). If deactivate @@ -72,19 +83,36 @@ type Context interface { // AddressSpace is activated. Normally activate is the same value as the // deactivate parameter passed to UninterruptibleSleepStart. UninterruptibleSleepFinish(activate bool) +} + +// NoopSleeper is a noop implementation of amutex.Sleeper and UninterruptibleSleep +// methods for anonymous embedding in other types that do not implement sleeps. +type NoopSleeper struct { + amutex.NoopSleeper +} + +// UninterruptibleSleepStart does nothing. +func (NoopSleeper) UninterruptibleSleepStart(bool) {} + +// UninterruptibleSleepFinish does nothing. +func (NoopSleeper) UninterruptibleSleepFinish(bool) {} + +// Deadline returns zero values, meaning no deadline. +func (NoopSleeper) Deadline() (time.Time, bool) { + return time.Time{}, false +} + +// Done returns nil. +func (NoopSleeper) Done() <-chan struct{} { + return nil +} - // Value returns the value associated with this Context for key, or nil if - // no value is associated with key. Successive calls to Value with the same - // key returns the same result. - // - // A key identifies a specific value in a Context. Functions that wish to - // retrieve values from Context typically allocate a key in a global - // variable then use that key as the argument to Context.Value. A key can - // be any type that supports equality; packages should define keys as an - // unexported type to avoid collisions. - Value(key interface{}) interface{} +// Err returns nil. +func (NoopSleeper) Err() error { + return nil } +// logContext implements basic logging. type logContext struct { log.Logger NoopSleeper @@ -95,19 +123,6 @@ func (logContext) Value(key interface{}) interface{} { return nil } -// NoopSleeper is a noop implementation of amutex.Sleeper and -// Context.UninterruptibleSleep* methods for anonymous embedding in other types -// that do not want to notify kernel.Task about sleeps. -type NoopSleeper struct { - amutex.NoopSleeper -} - -// UninterruptibleSleepStart does nothing. -func (NoopSleeper) UninterruptibleSleepStart(bool) {} - -// UninterruptibleSleepFinish does nothing. -func (NoopSleeper) UninterruptibleSleepFinish(bool) {} - // bgContext is the context returned by context.Background. var bgContext = &logContext{Logger: log.Log()} diff --git a/pkg/sentry/kernel/context.go b/pkg/sentry/kernel/context.go index e3f5b0d83..3c9dceaba 100644 --- a/pkg/sentry/kernel/context.go +++ b/pkg/sentry/kernel/context.go @@ -15,6 +15,8 @@ package kernel import ( + "time" + "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/context" ) @@ -97,6 +99,21 @@ func TaskFromContext(ctx context.Context) *Task { return nil } +// Deadline implements context.Context.Deadline. +func (*Task) Deadline() (time.Time, bool) { + return time.Time{}, false +} + +// Done implements context.Context.Done. +func (*Task) Done() <-chan struct{} { + return nil +} + +// Err implements context.Context.Err. +func (*Task) Err() error { + return nil +} + // AsyncContext returns a context.Context that may be used by goroutines that // do work on behalf of t and therefore share its contextual values, but are // not t's task goroutine (e.g. asynchronous I/O). @@ -129,6 +146,21 @@ func (ctx taskAsyncContext) IsLogging(level log.Level) bool { return ctx.t.IsLogging(level) } +// Deadline implements context.Context.Deadline. +func (ctx taskAsyncContext) Deadline() (time.Time, bool) { + return ctx.t.Deadline() +} + +// Done implements context.Context.Done. +func (ctx taskAsyncContext) Done() <-chan struct{} { + return ctx.t.Done() +} + +// Err implements context.Context.Err. +func (ctx taskAsyncContext) Err() error { + return ctx.t.Err() +} + // Value implements context.Context.Value. func (ctx taskAsyncContext) Value(key interface{}) interface{} { return ctx.t.Value(key) diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index e64d648e2..28ba950bd 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -391,7 +391,7 @@ func (k *Kernel) SaveTo(w io.Writer) error { // // N.B. This will also be saved along with the full kernel save below. cpuidStart := time.Now() - if err := state.Save(w, k.FeatureSet(), nil); err != nil { + if err := state.Save(k.SupervisorContext(), w, k.FeatureSet(), nil); err != nil { return err } log.Infof("CPUID save took [%s].", time.Since(cpuidStart)) @@ -399,7 +399,7 @@ func (k *Kernel) SaveTo(w io.Writer) error { // Save the kernel state. kernelStart := time.Now() var stats state.Stats - if err := state.Save(w, k, &stats); err != nil { + if err := state.Save(k.SupervisorContext(), w, k, &stats); err != nil { return err } log.Infof("Kernel save stats: %s", &stats) @@ -407,7 +407,7 @@ func (k *Kernel) SaveTo(w io.Writer) error { // Save the memory file's state. memoryStart := time.Now() - if err := k.mf.SaveTo(w); err != nil { + if err := k.mf.SaveTo(k.SupervisorContext(), w); err != nil { return err } log.Infof("Memory save took [%s].", time.Since(memoryStart)) @@ -542,7 +542,7 @@ func (k *Kernel) LoadFrom(r io.Reader, net inet.Stack, clocks sentrytime.Clocks) // don't need to explicitly install it in the Kernel. cpuidStart := time.Now() var features cpuid.FeatureSet - if err := state.Load(r, &features, nil); err != nil { + if err := state.Load(k.SupervisorContext(), r, &features, nil); err != nil { return err } log.Infof("CPUID load took [%s].", time.Since(cpuidStart)) @@ -558,7 +558,7 @@ func (k *Kernel) LoadFrom(r io.Reader, net inet.Stack, clocks sentrytime.Clocks) // Load the kernel state. kernelStart := time.Now() var stats state.Stats - if err := state.Load(r, k, &stats); err != nil { + if err := state.Load(k.SupervisorContext(), r, k, &stats); err != nil { return err } log.Infof("Kernel load stats: %s", &stats) @@ -566,7 +566,7 @@ func (k *Kernel) LoadFrom(r io.Reader, net inet.Stack, clocks sentrytime.Clocks) // Load the memory file's state. memoryStart := time.Now() - if err := k.mf.LoadFrom(r); err != nil { + if err := k.mf.LoadFrom(k.SupervisorContext(), r); err != nil { return err } log.Infof("Memory load took [%s].", time.Since(memoryStart)) @@ -1322,6 +1322,7 @@ func (k *Kernel) ListSockets() []*SocketEntry { return socks } +// supervisorContext is a privileged context. type supervisorContext struct { context.NoopSleeper log.Logger diff --git a/pkg/sentry/pgalloc/save_restore.go b/pkg/sentry/pgalloc/save_restore.go index 1effc7735..aafce1d00 100644 --- a/pkg/sentry/pgalloc/save_restore.go +++ b/pkg/sentry/pgalloc/save_restore.go @@ -16,6 +16,7 @@ package pgalloc import ( "bytes" + "context" "fmt" "io" "runtime" @@ -29,7 +30,7 @@ import ( ) // SaveTo writes f's state to the given stream. -func (f *MemoryFile) SaveTo(w io.Writer) error { +func (f *MemoryFile) SaveTo(ctx context.Context, w io.Writer) error { // Wait for reclaim. f.mu.Lock() defer f.mu.Unlock() @@ -78,10 +79,10 @@ func (f *MemoryFile) SaveTo(w io.Writer) error { } // Save metadata. - if err := state.Save(w, &f.fileSize, nil); err != nil { + if err := state.Save(ctx, w, &f.fileSize, nil); err != nil { return err } - if err := state.Save(w, &f.usage, nil); err != nil { + if err := state.Save(ctx, w, &f.usage, nil); err != nil { return err } @@ -114,9 +115,9 @@ func (f *MemoryFile) SaveTo(w io.Writer) error { } // LoadFrom loads MemoryFile state from the given stream. -func (f *MemoryFile) LoadFrom(r io.Reader) error { +func (f *MemoryFile) LoadFrom(ctx context.Context, r io.Reader) error { // Load metadata. - if err := state.Load(r, &f.fileSize, nil); err != nil { + if err := state.Load(ctx, r, &f.fileSize, nil); err != nil { return err } if err := f.file.Truncate(f.fileSize); err != nil { @@ -124,7 +125,7 @@ func (f *MemoryFile) LoadFrom(r io.Reader) error { } newMappings := make([]uintptr, f.fileSize>>chunkShift) f.mappings.Store(newMappings) - if err := state.Load(r, &f.usage, nil); err != nil { + if err := state.Load(ctx, r, &f.usage, nil); err != nil { return err } diff --git a/pkg/state/decode.go b/pkg/state/decode.go index 47e6b878a..590c241a3 100644 --- a/pkg/state/decode.go +++ b/pkg/state/decode.go @@ -16,6 +16,7 @@ package state import ( "bytes" + "context" "encoding/binary" "errors" "fmt" @@ -133,6 +134,9 @@ func (os *objectState) findCycle() []*objectState { // to ensure that all callbacks are executed, otherwise the callback graph was // not acyclic. type decodeState struct { + // ctx is the decode context. + ctx context.Context + // objectByID is the set of objects in progress. objectsByID map[uint64]*objectState diff --git a/pkg/state/encode.go b/pkg/state/encode.go index 5d9409a45..c5118d3a9 100644 --- a/pkg/state/encode.go +++ b/pkg/state/encode.go @@ -16,6 +16,7 @@ package state import ( "container/list" + "context" "encoding/binary" "fmt" "io" @@ -38,6 +39,9 @@ type queuedObject struct { // The encoding process is a breadth-first traversal of the object graph. The // inherent races and dependencies are much simpler than the decode case. type encodeState struct { + // ctx is the encode context. + ctx context.Context + // lastID is the last object ID. // // See idsByObject for context. Because of the special zero encoding diff --git a/pkg/state/map.go b/pkg/state/map.go index 7e6fefed4..4f3ebb0da 100644 --- a/pkg/state/map.go +++ b/pkg/state/map.go @@ -15,6 +15,7 @@ package state import ( + "context" "fmt" "reflect" "sort" @@ -219,3 +220,13 @@ func (m Map) AfterLoad(fn func()) { // data dependencies have been cleared. m.os.callbacks = append(m.os.callbacks, fn) } + +// Context returns the current context object. +func (m Map) Context() context.Context { + if m.es != nil { + return m.es.ctx + } else if m.ds != nil { + return m.ds.ctx + } + return context.Background() // No context. +} diff --git a/pkg/state/state.go b/pkg/state/state.go index d408ff84a..dbe507ab4 100644 --- a/pkg/state/state.go +++ b/pkg/state/state.go @@ -50,6 +50,7 @@ package state import ( + "context" "fmt" "io" "reflect" @@ -86,9 +87,10 @@ func UnwrapErrState(err error) error { } // Save saves the given object state. -func Save(w io.Writer, rootPtr interface{}, stats *Stats) error { +func Save(ctx context.Context, w io.Writer, rootPtr interface{}, stats *Stats) error { // Create the encoding state. es := &encodeState{ + ctx: ctx, idsByObject: make(map[uintptr]uint64), w: w, stats: stats, @@ -101,9 +103,10 @@ func Save(w io.Writer, rootPtr interface{}, stats *Stats) error { } // Load loads a checkpoint. -func Load(r io.Reader, rootPtr interface{}, stats *Stats) error { +func Load(ctx context.Context, r io.Reader, rootPtr interface{}, stats *Stats) error { // Create the decoding state. ds := &decodeState{ + ctx: ctx, objectsByID: make(map[uint64]*objectState), deferred: make(map[uint64]*pb.Object), r: r, diff --git a/pkg/state/state_test.go b/pkg/state/state_test.go index 7c24bbcda..d7221e9e8 100644 --- a/pkg/state/state_test.go +++ b/pkg/state/state_test.go @@ -16,6 +16,7 @@ package state import ( "bytes" + "context" "io/ioutil" "math" "reflect" @@ -46,7 +47,7 @@ func runTest(t *testing.T, tests []TestCase) { saveBuffer := &bytes.Buffer{} saveObjectPtr := reflect.New(reflect.TypeOf(root)) saveObjectPtr.Elem().Set(reflect.ValueOf(root)) - if err := Save(saveBuffer, saveObjectPtr.Interface(), nil); err != nil && !test.Fail { + if err := Save(context.Background(), saveBuffer, saveObjectPtr.Interface(), nil); err != nil && !test.Fail { t.Errorf(" FAIL: Save failed unexpectedly: %v", err) continue } else if err != nil { @@ -56,7 +57,7 @@ func runTest(t *testing.T, tests []TestCase) { // Load a new copy of the object. loadObjectPtr := reflect.New(reflect.TypeOf(root)) - if err := Load(bytes.NewReader(saveBuffer.Bytes()), loadObjectPtr.Interface(), nil); err != nil && !test.Fail { + if err := Load(context.Background(), bytes.NewReader(saveBuffer.Bytes()), loadObjectPtr.Interface(), nil); err != nil && !test.Fail { t.Errorf(" FAIL: Load failed unexpectedly: %v", err) continue } else if err != nil { @@ -624,7 +625,7 @@ func BenchmarkEncoding(b *testing.B) { bs := buildObject(b.N) var stats Stats b.StartTimer() - if err := Save(ioutil.Discard, bs, &stats); err != nil { + if err := Save(context.Background(), ioutil.Discard, bs, &stats); err != nil { b.Errorf("save failed: %v", err) } b.StopTimer() @@ -638,12 +639,12 @@ func BenchmarkDecoding(b *testing.B) { bs := buildObject(b.N) var newBS benchStruct buf := &bytes.Buffer{} - if err := Save(buf, bs, nil); err != nil { + if err := Save(context.Background(), buf, bs, nil); err != nil { b.Errorf("save failed: %v", err) } var stats Stats b.StartTimer() - if err := Load(buf, &newBS, &stats); err != nil { + if err := Load(context.Background(), buf, &newBS, &stats); err != nil { b.Errorf("load failed: %v", err) } b.StopTimer() -- cgit v1.2.3 From e70f28664af53b0428405c695c90a91b9bb43f67 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Fri, 1 Nov 2019 11:44:07 -0700 Subject: Allow the watchdog to detect when the sandbox is stuck during setup. The watchdog currently can find stuck tasks, but has no way to tell if the sandbox is stuck before the application starts executing. This CL adds a startup timeout and action to the watchdog. If Start() is not called before the given timeout (if non-zero), then the watchdog will take the action. PiperOrigin-RevId: 277970577 --- pkg/sentry/watchdog/watchdog.go | 152 ++++++++++++++++++++++++++++------------ runsc/boot/controller.go | 4 +- runsc/boot/loader.go | 4 +- 3 files changed, 112 insertions(+), 48 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/watchdog/watchdog.go b/pkg/sentry/watchdog/watchdog.go index 145102c0d..ecce6c69f 100644 --- a/pkg/sentry/watchdog/watchdog.go +++ b/pkg/sentry/watchdog/watchdog.go @@ -42,8 +42,35 @@ import ( ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" ) -// DefaultTimeout is a resonable timeout value for most applications. -const DefaultTimeout = 3 * time.Minute +// Opts configures the watchdog. +type Opts struct { + // TaskTimeout is the amount of time to allow a task to execute the + // same syscall without blocking before it's declared stuck. + TaskTimeout time.Duration + + // TaskTimeoutAction indicates what action to take when a stuck tasks + // is detected. + TaskTimeoutAction Action + + // StartupTimeout is the amount of time to allow between watchdog + // creation and calling watchdog.Start. + StartupTimeout time.Duration + + // StartupTimeoutAction indicates what action to take when + // watchdog.Start is not called within the timeout. + StartupTimeoutAction Action +} + +// DefaultOpts is a default set of options for the watchdog. +var DefaultOpts = Opts{ + // Task timeout. + TaskTimeout: 3 * time.Minute, + TaskTimeoutAction: LogWarning, + + // Startup timeout. + StartupTimeout: 30 * time.Second, + StartupTimeoutAction: LogWarning, +} // descheduleThreshold is the amount of time scheduling needs to be off before the entire wait period // is discounted from task's last update time. It's set high enough that small scheduling delays won't @@ -61,6 +88,7 @@ type Action int const ( // LogWarning logs warning message followed by stack trace. LogWarning Action = iota + // Panic will do the same logging as LogWarning and panic(). Panic ) @@ -80,17 +108,13 @@ func (a Action) String() string { // Watchdog is the main watchdog class. It controls a goroutine that periodically // analyses all tasks and reports if any of them appear to be stuck. type Watchdog struct { + // Configuration options are embedded. + Opts + // period indicates how often to check all tasks. It's calculated based on - // 'taskTimeout'. + // opts.TaskTimeout. period time.Duration - // taskTimeout is the amount of time to allow a task to execute the same syscall - // without blocking before it's declared stuck. - taskTimeout time.Duration - - // timeoutAction indicates what action to take when a stuck tasks is detected. - timeoutAction Action - // k is where the tasks come from. k *kernel.Kernel @@ -113,8 +137,12 @@ type Watchdog struct { // mu protects the fields below. mu sync.Mutex - // started is true if the watchdog has been started before. - started bool + // running is true if the watchdog is running. + running bool + + // startCalled is true if Start has ever been called. It remains true + // even if Stop is called. + startCalled bool } type offender struct { @@ -122,58 +150,81 @@ type offender struct { } // New creates a new watchdog. -func New(k *kernel.Kernel, taskTimeout time.Duration, a Action) *Watchdog { - // 4 is arbitrary, just don't want to prolong 'taskTimeout' too much. - period := taskTimeout / 4 - return &Watchdog{ - k: k, - period: period, - taskTimeout: taskTimeout, - timeoutAction: a, - offenders: make(map[*kernel.Task]*offender), - stop: make(chan struct{}), - done: make(chan struct{}), +func New(k *kernel.Kernel, opts Opts) *Watchdog { + // 4 is arbitrary, just don't want to prolong 'TaskTimeout' too much. + period := opts.TaskTimeout / 4 + w := &Watchdog{ + Opts: opts, + k: k, + period: period, + offenders: make(map[*kernel.Task]*offender), + stop: make(chan struct{}), + done: make(chan struct{}), + } + + // Handle StartupTimeout if it exists. + if w.StartupTimeout > 0 { + log.Infof("Watchdog waiting %v for startup", w.StartupTimeout) + go w.waitForStart() // S/R-SAFE: watchdog is stopped buring save and restarted after restore. } + + return w } // Start starts the watchdog. func (w *Watchdog) Start() { - if w.taskTimeout == 0 { - log.Infof("Watchdog disabled") - return - } - w.mu.Lock() defer w.mu.Unlock() - if w.started { + w.startCalled = true + + if w.running { return } + if w.TaskTimeout == 0 { + log.Infof("Watchdog task timeout disabled") + return + } w.lastRun = w.k.MonotonicClock().Now() - log.Infof("Starting watchdog, period: %v, timeout: %v, action: %v", w.period, w.taskTimeout, w.timeoutAction) + log.Infof("Starting watchdog, period: %v, timeout: %v, action: %v", w.period, w.TaskTimeout, w.TaskTimeoutAction) go w.loop() // S/R-SAFE: watchdog is stopped during save and restarted after restore. - w.started = true + w.running = true } // Stop requests the watchdog to stop and wait for it. func (w *Watchdog) Stop() { - if w.taskTimeout == 0 { + if w.TaskTimeout == 0 { return } w.mu.Lock() defer w.mu.Unlock() - if !w.started { + if !w.running { return } log.Infof("Stopping watchdog") w.stop <- struct{}{} <-w.done - w.started = false + w.running = false log.Infof("Watchdog stopped") } +// waitForStart waits for Start to be called and takes action if it does not +// happen within the startup timeout. +func (w *Watchdog) waitForStart() { + <-time.After(w.StartupTimeout) + w.mu.Lock() + defer w.mu.Unlock() + if w.startCalled { + // We are fine. + return + } + var buf bytes.Buffer + buf.WriteString("Watchdog.Start() not called within %s:\n") + w.doAction(w.StartupTimeoutAction, false, &buf) +} + // loop is the main watchdog routine. It only returns when 'Stop()' is called. func (w *Watchdog) loop() { // Loop until someone stops it. @@ -202,7 +253,7 @@ func (w *Watchdog) runTurn() { select { case <-done: - case <-time.After(w.taskTimeout): + case <-time.After(w.TaskTimeout): // Report if the watchdog is not making progress. // No one is wathching the watchdog watcher though. w.reportStuckWatchdog() @@ -231,7 +282,7 @@ func (w *Watchdog) runTurn() { if tsched.State == kernel.TaskGoroutineRunningSys { lastUpdateTime := ktime.FromNanoseconds(int64(tsched.Timestamp * uint64(linux.ClockTick))) elapsed := now.Sub(lastUpdateTime) - discount - if elapsed > w.taskTimeout { + if elapsed > w.TaskTimeout { tc, ok := w.offenders[t] if !ok { // New stuck task detected. @@ -261,28 +312,34 @@ func (w *Watchdog) report(offenders map[*kernel.Task]*offender, newTaskFound boo tid := w.k.TaskSet().Root.IDOfTask(t) buf.WriteString(fmt.Sprintf("\tTask tid: %v (%#x), entered RunSys state %v ago.\n", tid, uint64(tid), now.Sub(o.lastUpdateTime))) } + buf.WriteString("Search for '(*Task).run(0x..., 0x)' in the stack dump to find the offending goroutine") - w.onStuckTask(newTaskFound, &buf) + + // Dump stack only if a new task is detected or if it sometime has + // passed since the last time a stack dump was generated. + skipStack := newTaskFound || time.Since(w.lastStackDump) >= stackDumpSameTaskPeriod + w.doAction(w.TaskTimeoutAction, skipStack, &buf) } func (w *Watchdog) reportStuckWatchdog() { var buf bytes.Buffer buf.WriteString("Watchdog goroutine is stuck:\n") - w.onStuckTask(true, &buf) + w.doAction(w.TaskTimeoutAction, false, &buf) } -func (w *Watchdog) onStuckTask(newTaskFound bool, msg *bytes.Buffer) { - switch w.timeoutAction { +// doAction will take the given action. If the action is LogWarnind and +// skipStack is true, then the stack printing will be skipped. +func (w *Watchdog) doAction(action Action, skipStack bool, msg *bytes.Buffer) { + switch action { case LogWarning: - // Dump stack only if a new task is detected or if it sometime has passed since - // the last time a stack dump was generated. - if !newTaskFound && time.Since(w.lastStackDump) < stackDumpSameTaskPeriod { + if skipStack { msg.WriteString("\n...[stack dump skipped]...") log.Warningf(msg.String()) - } else { - log.TracebackAll(msg.String()) - w.lastStackDump = time.Now() + return + } + log.TracebackAll(msg.String()) + w.lastStackDump = time.Now() case Panic: // Panic will skip over running tasks, which is likely the culprit here. So manually @@ -301,5 +358,8 @@ func (w *Watchdog) onStuckTask(newTaskFound bool, msg *bytes.Buffer) { case <-time.After(1 * time.Second): } panic(fmt.Sprintf("Stack for running G's are skipped while panicking.\n%s", msg.String())) + default: + panic(fmt.Sprintf("Unknown watchdog action %v", action)) + } } diff --git a/runsc/boot/controller.go b/runsc/boot/controller.go index 928285683..f62be4c59 100644 --- a/runsc/boot/controller.go +++ b/runsc/boot/controller.go @@ -380,7 +380,9 @@ func (cm *containerManager) Restore(o *RestoreOpts, _ *struct{}) error { } // Since we have a new kernel we also must make a new watchdog. - dog := watchdog.New(k, watchdog.DefaultTimeout, cm.l.conf.WatchdogAction) + dogOpts := watchdog.DefaultOpts + dogOpts.TaskTimeoutAction = cm.l.conf.WatchdogAction + dog := watchdog.New(k, dogOpts) // Change the loader fields to reflect the changes made when restoring. cm.l.k = k diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index 86df384f8..4d1bd2d08 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -300,7 +300,9 @@ func New(args Args) (*Loader, error) { } // Create a watchdog. - dog := watchdog.New(k, watchdog.DefaultTimeout, args.Conf.WatchdogAction) + dogOpts := watchdog.DefaultOpts + dogOpts.TaskTimeoutAction = args.Conf.WatchdogAction + dog := watchdog.New(k, dogOpts) procArgs, err := newProcess(args.ID, args.Spec, creds, k, k.RootPIDNamespace()) if err != nil { -- cgit v1.2.3 From 515fee5b6d4f3270c951f72283aef79a28d463dd Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Fri, 1 Nov 2019 12:42:04 -0700 Subject: Add SO_PASSCRED support to netlink sockets Since we only supporting sending messages from the kernel, the peer is always the kernel, simplifying handling. There are currently no known users of SO_PASSCRED that would actually receive messages from gVisor, but adding full support is barely more work than stubbing out fake support. Updates #1117 Fixes #1119 PiperOrigin-RevId: 277981465 --- pkg/sentry/socket/netlink/BUILD | 1 + pkg/sentry/socket/netlink/socket.go | 76 ++++++++++++++++++- test/syscalls/linux/socket_netlink_route.cc | 110 +++++++++++++++++++++++++++- 3 files changed, 183 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netlink/BUILD b/pkg/sentry/socket/netlink/BUILD index f95803f91..79589e3c8 100644 --- a/pkg/sentry/socket/netlink/BUILD +++ b/pkg/sentry/socket/netlink/BUILD @@ -20,6 +20,7 @@ go_library( "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", "//pkg/sentry/kernel", + "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/time", "//pkg/sentry/safemem", "//pkg/sentry/socket", diff --git a/pkg/sentry/socket/netlink/socket.go b/pkg/sentry/socket/netlink/socket.go index b2732ca29..05dac4f0a 100644 --- a/pkg/sentry/socket/netlink/socket.go +++ b/pkg/sentry/socket/netlink/socket.go @@ -27,6 +27,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/socket" @@ -61,7 +62,7 @@ var netlinkSocketDevice = device.NewAnonDevice() // This implementation only supports userspace sending and receiving messages // to/from the kernel. // -// Socket implements socket.Socket. +// Socket implements socket.Socket and transport.Credentialer. // // +stateify savable type Socket struct { @@ -104,9 +105,13 @@ type Socket struct { // sendBufferSize is the send buffer "size". We don't actually have a // fixed buffer but only consume this many bytes. sendBufferSize uint32 + + // passcred indicates if this socket wants SCM credentials. + passcred bool } var _ socket.Socket = (*Socket)(nil) +var _ transport.Credentialer = (*Socket)(nil) // NewSocket creates a new Socket. func NewSocket(t *kernel.Task, skType linux.SockType, protocol Protocol) (*Socket, *syserr.Error) { @@ -172,6 +177,22 @@ func (s *Socket) EventUnregister(e *waiter.Entry) { s.ep.EventUnregister(e) } +// Passcred implements transport.Credentialer.Passcred. +func (s *Socket) Passcred() bool { + s.mu.Lock() + passcred := s.passcred + s.mu.Unlock() + return passcred +} + +// ConnectedPasscred implements transport.Credentialer.ConnectedPasscred. +func (s *Socket) ConnectedPasscred() bool { + // This socket is connected to the kernel, which doesn't need creds. + // + // This is arbitrary, as ConnectedPasscred on this type has no callers. + return false +} + // Ioctl implements fs.FileOperations.Ioctl. func (*Socket) Ioctl(context.Context, *fs.File, usermem.IO, arch.SyscallArguments) (uintptr, error) { // TODO(b/68878065): no ioctls supported. @@ -309,9 +330,20 @@ func (s *Socket) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem. // We don't have limit on receiving size. return int32(math.MaxInt32), nil + case linux.SO_PASSCRED: + if outLen < sizeOfInt32 { + return nil, syserr.ErrInvalidArgument + } + var passcred int32 + if s.Passcred() { + passcred = 1 + } + return passcred, nil + default: socket.GetSockOptEmitUnimplementedEvent(t, name) } + case linux.SOL_NETLINK: switch name { case linux.NETLINK_BROADCAST_ERROR, @@ -348,6 +380,7 @@ func (s *Socket) SetSockOpt(t *kernel.Task, level int, name int, opt []byte) *sy s.sendBufferSize = size s.mu.Unlock() return nil + case linux.SO_RCVBUF: if len(opt) < sizeOfInt32 { return syserr.ErrInvalidArgument @@ -355,6 +388,18 @@ func (s *Socket) SetSockOpt(t *kernel.Task, level int, name int, opt []byte) *sy // We don't have limit on receiving size. So just accept anything as // valid for compatibility. return nil + + case linux.SO_PASSCRED: + if len(opt) < sizeOfInt32 { + return syserr.ErrInvalidArgument + } + passcred := usermem.ByteOrder.Uint32(opt) + + s.mu.Lock() + s.passcred = passcred != 0 + s.mu.Unlock() + return nil + default: socket.SetSockOptEmitUnimplementedEvent(t, name) } @@ -483,6 +528,26 @@ func (s *Socket) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, _ }) } +// kernelSCM implements control.SCMCredentials with credentials that represent +// the kernel itself rather than a Task. +// +// +stateify savable +type kernelSCM struct{} + +// Equals implements transport.CredentialsControlMessage.Equals. +func (kernelSCM) Equals(oc transport.CredentialsControlMessage) bool { + _, ok := oc.(kernelSCM) + return ok +} + +// Credentials implements control.SCMCredentials.Credentials. +func (kernelSCM) Credentials(*kernel.Task) (kernel.ThreadID, auth.UID, auth.GID) { + return 0, auth.RootUID, auth.RootGID +} + +// kernelCreds is the concrete version of kernelSCM used in all creds. +var kernelCreds = &kernelSCM{} + // sendResponse sends the response messages in ms back to userspace. func (s *Socket) sendResponse(ctx context.Context, ms *MessageSet) *syserr.Error { // Linux combines multiple netlink messages into a single datagram. @@ -491,10 +556,15 @@ func (s *Socket) sendResponse(ctx context.Context, ms *MessageSet) *syserr.Error bufs = append(bufs, m.Finalize()) } + // All messages are from the kernel. + cms := transport.ControlMessages{ + Credentials: kernelCreds, + } + if len(bufs) > 0 { // RecvMsg never receives the address, so we don't need to send // one. - _, notify, err := s.connection.Send(bufs, transport.ControlMessages{}, tcpip.FullAddress{}) + _, notify, err := s.connection.Send(bufs, cms, tcpip.FullAddress{}) // If the buffer is full, we simply drop messages, just like // Linux. if err != nil && err != syserr.ErrWouldBlock { @@ -521,7 +591,7 @@ func (s *Socket) sendResponse(ctx context.Context, ms *MessageSet) *syserr.Error // Add the dump_done_errno payload. m.Put(int64(0)) - _, notify, err := s.connection.Send([][]byte{m.Finalize()}, transport.ControlMessages{}, tcpip.FullAddress{}) + _, notify, err := s.connection.Send([][]byte{m.Finalize()}, cms, tcpip.FullAddress{}) if err != nil && err != syserr.ErrWouldBlock { return err } diff --git a/test/syscalls/linux/socket_netlink_route.cc b/test/syscalls/linux/socket_netlink_route.cc index dd4a11655..be0dadcd6 100644 --- a/test/syscalls/linux/socket_netlink_route.cc +++ b/test/syscalls/linux/socket_netlink_route.cc @@ -195,7 +195,8 @@ INSTANTIATE_TEST_SUITE_P( std::make_tuple(SO_DOMAIN, IsEqual(AF_NETLINK), absl::StrFormat("AF_NETLINK (%d)", AF_NETLINK)), std::make_tuple(SO_PROTOCOL, IsEqual(NETLINK_ROUTE), - absl::StrFormat("NETLINK_ROUTE (%d)", NETLINK_ROUTE)))); + absl::StrFormat("NETLINK_ROUTE (%d)", NETLINK_ROUTE)), + std::make_tuple(SO_PASSCRED, IsEqual(0), "0"))); // Validates the reponses to RTM_GETLINK + NLM_F_DUMP. void CheckGetLinkResponse(const struct nlmsghdr* hdr, int seq, int port) { @@ -692,6 +693,113 @@ TEST(NetlinkRouteTest, RecvmsgTruncPeek) { } while (type != NLMSG_DONE && type != NLMSG_ERROR); } +// No SCM_CREDENTIALS are received without SO_PASSCRED set. +TEST(NetlinkRouteTest, NoPasscredNoCreds) { + FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket()); + + ASSERT_THAT(setsockopt(fd.get(), SOL_SOCKET, SO_PASSCRED, &kSockOptOff, + sizeof(kSockOptOff)), + SyscallSucceeds()); + + struct request { + struct nlmsghdr hdr; + struct rtgenmsg rgm; + }; + + constexpr uint32_t kSeq = 12345; + + struct request req; + req.hdr.nlmsg_len = sizeof(req); + req.hdr.nlmsg_type = RTM_GETADDR; + req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP; + req.hdr.nlmsg_seq = kSeq; + req.rgm.rtgen_family = AF_UNSPEC; + + struct iovec iov = {}; + iov.iov_base = &req; + iov.iov_len = sizeof(req); + + struct msghdr msg = {}; + msg.msg_iov = &iov; + msg.msg_iovlen = 1; + + ASSERT_THAT(RetryEINTR(sendmsg)(fd.get(), &msg, 0), SyscallSucceeds()); + + iov.iov_base = NULL; + iov.iov_len = 0; + + char control[CMSG_SPACE(sizeof(struct ucred))] = {}; + msg.msg_control = control; + msg.msg_controllen = sizeof(control); + + // Note: This test assumes at least one message is returned by the + // RTM_GETADDR request. + ASSERT_THAT(RetryEINTR(recvmsg)(fd.get(), &msg, 0), SyscallSucceeds()); + + // No control messages. + EXPECT_EQ(CMSG_FIRSTHDR(&msg), nullptr); +} + +// SCM_CREDENTIALS are received with SO_PASSCRED set. +TEST(NetlinkRouteTest, PasscredCreds) { + FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket()); + + ASSERT_THAT(setsockopt(fd.get(), SOL_SOCKET, SO_PASSCRED, &kSockOptOn, + sizeof(kSockOptOn)), + SyscallSucceeds()); + + struct request { + struct nlmsghdr hdr; + struct rtgenmsg rgm; + }; + + constexpr uint32_t kSeq = 12345; + + struct request req; + req.hdr.nlmsg_len = sizeof(req); + req.hdr.nlmsg_type = RTM_GETADDR; + req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP; + req.hdr.nlmsg_seq = kSeq; + req.rgm.rtgen_family = AF_UNSPEC; + + struct iovec iov = {}; + iov.iov_base = &req; + iov.iov_len = sizeof(req); + + struct msghdr msg = {}; + msg.msg_iov = &iov; + msg.msg_iovlen = 1; + + ASSERT_THAT(RetryEINTR(sendmsg)(fd.get(), &msg, 0), SyscallSucceeds()); + + iov.iov_base = NULL; + iov.iov_len = 0; + + char control[CMSG_SPACE(sizeof(struct ucred))] = {}; + msg.msg_control = control; + msg.msg_controllen = sizeof(control); + + // Note: This test assumes at least one message is returned by the + // RTM_GETADDR request. + ASSERT_THAT(RetryEINTR(recvmsg)(fd.get(), &msg, 0), SyscallSucceeds()); + + struct ucred creds; + struct cmsghdr* cmsg = CMSG_FIRSTHDR(&msg); + ASSERT_NE(cmsg, nullptr); + ASSERT_EQ(cmsg->cmsg_len, CMSG_LEN(sizeof(creds))); + ASSERT_EQ(cmsg->cmsg_level, SOL_SOCKET); + ASSERT_EQ(cmsg->cmsg_type, SCM_CREDENTIALS); + + memcpy(&creds, CMSG_DATA(cmsg), sizeof(creds)); + + // The peer is the kernel, which is "PID" 0. + EXPECT_EQ(creds.pid, 0); + // The kernel identifies as root. Also allow nobody in case this test is + // running in a userns without root mapped. + EXPECT_THAT(creds.uid, AnyOf(Eq(0), Eq(65534))); + EXPECT_THAT(creds.gid, AnyOf(Eq(0), Eq(65534))); +} + } // namespace } // namespace testing -- cgit v1.2.3 From 3b4f5445d03f7d2f170d68a8a4969b8acbad773e Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Mon, 4 Nov 2019 09:54:14 -0800 Subject: Update membarrier bug Updates #267 PiperOrigin-RevId: 278402684 --- pkg/sentry/syscalls/linux/linux64_amd64.go | 2 +- pkg/sentry/syscalls/linux/linux64_arm64.go | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/linux64_amd64.go b/pkg/sentry/syscalls/linux/linux64_amd64.go index 3021440ed..81e4f93a6 100644 --- a/pkg/sentry/syscalls/linux/linux64_amd64.go +++ b/pkg/sentry/syscalls/linux/linux64_amd64.go @@ -364,7 +364,7 @@ var AMD64 = &kernel.SyscallTable{ 321: syscalls.CapError("bpf", linux.CAP_SYS_ADMIN, "", nil), 322: syscalls.Supported("execveat", Execveat), 323: syscalls.ErrorWithEvent("userfaultfd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/266"}), // TODO(b/118906345) - 324: syscalls.ErrorWithEvent("membarrier", syserror.ENOSYS, "", []string{"gvisor.dev/issue/267"}), // TODO(b/118904897) + 324: syscalls.ErrorWithEvent("membarrier", syserror.ENOSYS, "", []string{"gvisor.dev/issue/267"}), // TODO(gvisor.dev/issue/267) 325: syscalls.PartiallySupported("mlock2", Mlock2, "Stub implementation. The sandbox lacks appropriate permissions.", nil), // Syscalls after 325 are "backports" from versions of Linux after 4.4. diff --git a/pkg/sentry/syscalls/linux/linux64_arm64.go b/pkg/sentry/syscalls/linux/linux64_arm64.go index 4cf7f836a..a809115e0 100644 --- a/pkg/sentry/syscalls/linux/linux64_arm64.go +++ b/pkg/sentry/syscalls/linux/linux64_arm64.go @@ -297,7 +297,7 @@ var ARM64 = &kernel.SyscallTable{ 280: syscalls.CapError("bpf", linux.CAP_SYS_ADMIN, "", nil), 281: syscalls.ErrorWithEvent("execveat", syserror.ENOSYS, "", []string{"gvisor.dev/issue/265"}), // TODO(b/118901836) 282: syscalls.ErrorWithEvent("userfaultfd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/266"}), // TODO(b/118906345) - 283: syscalls.ErrorWithEvent("membarrier", syserror.ENOSYS, "", []string{"gvisor.dev/issue/267"}), // TODO(b/118904897) + 283: syscalls.ErrorWithEvent("membarrier", syserror.ENOSYS, "", []string{"gvisor.dev/issue/267"}), // TODO(gvisor.dev/issue/267) 284: syscalls.PartiallySupported("mlock2", Mlock2, "Stub implementation. The sandbox lacks appropriate permissions.", nil), 285: syscalls.ErrorWithEvent("copy_file_range", syserror.ENOSYS, "", nil), 286: syscalls.Supported("preadv2", Preadv2), -- cgit v1.2.3 From b23b36e701c40827065217f4652a51eebc5f9913 Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Mon, 4 Nov 2019 10:06:00 -0800 Subject: Add NETLINK_KOBJECT_UEVENT socket support NETLINK_KOBJECT_UEVENT sockets send udev-style messages for device events. gVisor doesn't have any device events, so our sockets don't need to do anything once created. systemd's device manager needs to be able to create one of these sockets. It also wants to install a BPF filter on the socket. Since we'll never send any messages, the filter would never be invoked, thus we just fake it out. Fixes #1117 Updates #1119 PiperOrigin-RevId: 278405893 --- pkg/sentry/socket/netlink/provider.go | 7 ++ pkg/sentry/socket/netlink/route/protocol.go | 5 + pkg/sentry/socket/netlink/socket.go | 42 ++++++++ pkg/sentry/socket/netlink/uevent/BUILD | 17 +++ pkg/sentry/socket/netlink/uevent/protocol.go | 60 +++++++++++ runsc/boot/BUILD | 1 + runsc/boot/loader.go | 1 + test/syscalls/BUILD | 4 + test/syscalls/linux/BUILD | 29 +++++ test/syscalls/linux/socket_netdevice.cc | 3 +- test/syscalls/linux/socket_netlink.cc | 153 +++++++++++++++++++++++++++ test/syscalls/linux/socket_netlink_route.cc | 140 ++++-------------------- test/syscalls/linux/socket_netlink_uevent.cc | 83 +++++++++++++++ test/syscalls/linux/socket_netlink_util.cc | 5 +- test/syscalls/linux/socket_netlink_util.h | 5 +- 15 files changed, 431 insertions(+), 124 deletions(-) create mode 100644 pkg/sentry/socket/netlink/uevent/BUILD create mode 100644 pkg/sentry/socket/netlink/uevent/protocol.go create mode 100644 test/syscalls/linux/socket_netlink.cc create mode 100644 test/syscalls/linux/socket_netlink_uevent.cc (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netlink/provider.go b/pkg/sentry/socket/netlink/provider.go index 689cad997..be005df24 100644 --- a/pkg/sentry/socket/netlink/provider.go +++ b/pkg/sentry/socket/netlink/provider.go @@ -30,6 +30,13 @@ type Protocol interface { // Protocol returns the Linux netlink protocol value. Protocol() int + // CanSend returns true if this protocol may ever send messages. + // + // TODO(gvisor.dev/issue/1119): This is a workaround to allow + // advertising support for otherwise unimplemented features on sockets + // that will never send messages, thus making those features no-ops. + CanSend() bool + // ProcessMessage processes a single message from userspace. // // If err == nil, any messages added to ms will be sent back to the diff --git a/pkg/sentry/socket/netlink/route/protocol.go b/pkg/sentry/socket/netlink/route/protocol.go index cc70ac237..6b4a0ecf4 100644 --- a/pkg/sentry/socket/netlink/route/protocol.go +++ b/pkg/sentry/socket/netlink/route/protocol.go @@ -61,6 +61,11 @@ func (p *Protocol) Protocol() int { return linux.NETLINK_ROUTE } +// CanSend implements netlink.Protocol.CanSend. +func (p *Protocol) CanSend() bool { + return true +} + // dumpLinks handles RTM_GETLINK + NLM_F_DUMP requests. func (p *Protocol) dumpLinks(ctx context.Context, hdr linux.NetlinkMessageHeader, data []byte, ms *netlink.MessageSet) *syserr.Error { // NLM_F_DUMP + RTM_GETLINK messages are supposed to include an diff --git a/pkg/sentry/socket/netlink/socket.go b/pkg/sentry/socket/netlink/socket.go index 05dac4f0a..4a1b87a9a 100644 --- a/pkg/sentry/socket/netlink/socket.go +++ b/pkg/sentry/socket/netlink/socket.go @@ -54,6 +54,8 @@ const ( maxSendBufferSize = 4 << 20 // 4MB ) +var errNoFilter = syserr.New("no filter attached", linux.ENOENT) + // netlinkSocketDevice is the netlink socket virtual device. var netlinkSocketDevice = device.NewAnonDevice() @@ -108,6 +110,12 @@ type Socket struct { // passcred indicates if this socket wants SCM credentials. passcred bool + + // filter indicates that this socket has a BPF filter "installed". + // + // TODO(gvisor.dev/issue/1119): We don't actually support filtering, + // this is just bookkeeping for tracking add/remove. + filter bool } var _ socket.Socket = (*Socket)(nil) @@ -400,6 +408,40 @@ func (s *Socket) SetSockOpt(t *kernel.Task, level int, name int, opt []byte) *sy s.mu.Unlock() return nil + case linux.SO_ATTACH_FILTER: + // TODO(gvisor.dev/issue/1119): We don't actually + // support filtering. If this socket can't ever send + // messages, then there is nothing to filter and we can + // advertise support. Otherwise, be conservative and + // return an error. + if s.protocol.CanSend() { + socket.SetSockOptEmitUnimplementedEvent(t, name) + return syserr.ErrProtocolNotAvailable + } + + s.mu.Lock() + s.filter = true + s.mu.Unlock() + return nil + + case linux.SO_DETACH_FILTER: + // TODO(gvisor.dev/issue/1119): See above. + if s.protocol.CanSend() { + socket.SetSockOptEmitUnimplementedEvent(t, name) + return syserr.ErrProtocolNotAvailable + } + + s.mu.Lock() + filter := s.filter + s.filter = false + s.mu.Unlock() + + if !filter { + return errNoFilter + } + + return nil + default: socket.SetSockOptEmitUnimplementedEvent(t, name) } diff --git a/pkg/sentry/socket/netlink/uevent/BUILD b/pkg/sentry/socket/netlink/uevent/BUILD new file mode 100644 index 000000000..0777f3baf --- /dev/null +++ b/pkg/sentry/socket/netlink/uevent/BUILD @@ -0,0 +1,17 @@ +load("//tools/go_stateify:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_library( + name = "uevent", + srcs = ["protocol.go"], + importpath = "gvisor.dev/gvisor/pkg/sentry/socket/netlink/uevent", + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/abi/linux", + "//pkg/sentry/context", + "//pkg/sentry/kernel", + "//pkg/sentry/socket/netlink", + "//pkg/syserr", + ], +) diff --git a/pkg/sentry/socket/netlink/uevent/protocol.go b/pkg/sentry/socket/netlink/uevent/protocol.go new file mode 100644 index 000000000..b5d7808d7 --- /dev/null +++ b/pkg/sentry/socket/netlink/uevent/protocol.go @@ -0,0 +1,60 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package uevent provides a NETLINK_KOBJECT_UEVENT socket protocol. +// +// NETLINK_KOBJECT_UEVENT sockets send udev-style device events. gVisor does +// not support any device events, so these sockets never send any messages. +package uevent + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/socket/netlink" + "gvisor.dev/gvisor/pkg/syserr" +) + +// Protocol implements netlink.Protocol. +// +// +stateify savable +type Protocol struct{} + +var _ netlink.Protocol = (*Protocol)(nil) + +// NewProtocol creates a NETLINK_KOBJECT_UEVENT netlink.Protocol. +func NewProtocol(t *kernel.Task) (netlink.Protocol, *syserr.Error) { + return &Protocol{}, nil +} + +// Protocol implements netlink.Protocol.Protocol. +func (p *Protocol) Protocol() int { + return linux.NETLINK_KOBJECT_UEVENT +} + +// CanSend implements netlink.Protocol.CanSend. +func (p *Protocol) CanSend() bool { + return false +} + +// ProcessMessage implements netlink.Protocol.ProcessMessage. +func (p *Protocol) ProcessMessage(ctx context.Context, hdr linux.NetlinkMessageHeader, data []byte, ms *netlink.MessageSet) *syserr.Error { + // Silently ignore all messages. + return nil +} + +// init registers the NETLINK_KOBJECT_UEVENT provider. +func init() { + netlink.RegisterProvider(linux.NETLINK_KOBJECT_UEVENT, NewProtocol) +} diff --git a/runsc/boot/BUILD b/runsc/boot/BUILD index 6fe2b57de..58e86ae7f 100644 --- a/runsc/boot/BUILD +++ b/runsc/boot/BUILD @@ -60,6 +60,7 @@ go_library( "//pkg/sentry/socket/hostinet", "//pkg/sentry/socket/netlink", "//pkg/sentry/socket/netlink/route", + "//pkg/sentry/socket/netlink/uevent", "//pkg/sentry/socket/netstack", "//pkg/sentry/socket/unix", "//pkg/sentry/state", diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index 4d1bd2d08..f05d5973f 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -65,6 +65,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/socket/hostinet" _ "gvisor.dev/gvisor/pkg/sentry/socket/netlink" _ "gvisor.dev/gvisor/pkg/sentry/socket/netlink/route" + _ "gvisor.dev/gvisor/pkg/sentry/socket/netlink/uevent" "gvisor.dev/gvisor/pkg/sentry/socket/netstack" _ "gvisor.dev/gvisor/pkg/sentry/socket/unix" ) diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index a53a23afd..3e5b6b3c3 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -511,8 +511,12 @@ syscall_test(test = "//test/syscalls/linux:socket_ip_unbound_test") syscall_test(test = "//test/syscalls/linux:socket_netdevice_test") +syscall_test(test = "//test/syscalls/linux:socket_netlink_test") + syscall_test(test = "//test/syscalls/linux:socket_netlink_route_test") +syscall_test(test = "//test/syscalls/linux:socket_netlink_uevent_test") + syscall_test(test = "//test/syscalls/linux:socket_blocking_local_test") syscall_test(test = "//test/syscalls/linux:socket_blocking_ip_test") diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 833fbaa09..93bff8299 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -2675,6 +2675,20 @@ cc_binary( ], ) +cc_binary( + name = "socket_netlink_test", + testonly = 1, + srcs = ["socket_netlink.cc"], + linkstatic = 1, + deps = [ + ":socket_test_util", + "//test/util:file_descriptor", + "//test/util:test_main", + "//test/util:test_util", + "@com_google_googletest//:gtest", + ], +) + cc_binary( name = "socket_netlink_route_test", testonly = 1, @@ -2692,6 +2706,21 @@ cc_binary( ], ) +cc_binary( + name = "socket_netlink_uevent_test", + testonly = 1, + srcs = ["socket_netlink_uevent.cc"], + linkstatic = 1, + deps = [ + ":socket_netlink_util", + ":socket_test_util", + "//test/util:file_descriptor", + "//test/util:test_main", + "//test/util:test_util", + "@com_google_googletest//:gtest", + ], +) + # These socket tests are in a library because the test cases are shared # across several test build targets. cc_library( diff --git a/test/syscalls/linux/socket_netdevice.cc b/test/syscalls/linux/socket_netdevice.cc index 765f8e0e4..405dbbd73 100644 --- a/test/syscalls/linux/socket_netdevice.cc +++ b/test/syscalls/linux/socket_netdevice.cc @@ -68,7 +68,8 @@ TEST(NetdeviceTest, Netmask) { // Use a netlink socket to get the netmask, which we'll then compare to the // netmask obtained via ioctl. - FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket()); + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); uint32_t port = ASSERT_NO_ERRNO_AND_VALUE(NetlinkPortID(fd.get())); struct request { diff --git a/test/syscalls/linux/socket_netlink.cc b/test/syscalls/linux/socket_netlink.cc new file mode 100644 index 000000000..4ec0fd4fa --- /dev/null +++ b/test/syscalls/linux/socket_netlink.cc @@ -0,0 +1,153 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include +#include +#include +#include + +#include "gtest/gtest.h" +#include "test/syscalls/linux/socket_test_util.h" +#include "test/util/file_descriptor.h" +#include "test/util/test_util.h" + +// Tests for all netlink socket protocols. + +namespace gvisor { +namespace testing { + +namespace { + +// NetlinkTest parameter is the protocol to test. +using NetlinkTest = ::testing::TestWithParam; + +// Netlink sockets must be SOCK_DGRAM or SOCK_RAW. +TEST_P(NetlinkTest, Types) { + const int protocol = GetParam(); + + EXPECT_THAT(socket(AF_NETLINK, SOCK_STREAM, protocol), + SyscallFailsWithErrno(ESOCKTNOSUPPORT)); + EXPECT_THAT(socket(AF_NETLINK, SOCK_SEQPACKET, protocol), + SyscallFailsWithErrno(ESOCKTNOSUPPORT)); + EXPECT_THAT(socket(AF_NETLINK, SOCK_RDM, protocol), + SyscallFailsWithErrno(ESOCKTNOSUPPORT)); + EXPECT_THAT(socket(AF_NETLINK, SOCK_DCCP, protocol), + SyscallFailsWithErrno(ESOCKTNOSUPPORT)); + EXPECT_THAT(socket(AF_NETLINK, SOCK_PACKET, protocol), + SyscallFailsWithErrno(ESOCKTNOSUPPORT)); + + int fd; + EXPECT_THAT(fd = socket(AF_NETLINK, SOCK_DGRAM, protocol), SyscallSucceeds()); + EXPECT_THAT(close(fd), SyscallSucceeds()); + + EXPECT_THAT(fd = socket(AF_NETLINK, SOCK_RAW, protocol), SyscallSucceeds()); + EXPECT_THAT(close(fd), SyscallSucceeds()); +} + +TEST_P(NetlinkTest, AutomaticPort) { + const int protocol = GetParam(); + + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_NETLINK, SOCK_RAW, protocol)); + + struct sockaddr_nl addr = {}; + addr.nl_family = AF_NETLINK; + + EXPECT_THAT( + bind(fd.get(), reinterpret_cast(&addr), sizeof(addr)), + SyscallSucceeds()); + + socklen_t addrlen = sizeof(addr); + EXPECT_THAT(getsockname(fd.get(), reinterpret_cast(&addr), + &addrlen), + SyscallSucceeds()); + EXPECT_EQ(addrlen, sizeof(addr)); + // This is the only netlink socket in the process, so it should get the PID as + // the port id. + // + // N.B. Another process could theoretically have explicitly reserved our pid + // as a port ID, but that is very unlikely. + EXPECT_EQ(addr.nl_pid, getpid()); +} + +// Calling connect automatically binds to an automatic port. +TEST_P(NetlinkTest, ConnectBinds) { + const int protocol = GetParam(); + + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_NETLINK, SOCK_RAW, protocol)); + + struct sockaddr_nl addr = {}; + addr.nl_family = AF_NETLINK; + + EXPECT_THAT(connect(fd.get(), reinterpret_cast(&addr), + sizeof(addr)), + SyscallSucceeds()); + + socklen_t addrlen = sizeof(addr); + EXPECT_THAT(getsockname(fd.get(), reinterpret_cast(&addr), + &addrlen), + SyscallSucceeds()); + EXPECT_EQ(addrlen, sizeof(addr)); + + // Each test is running in a pid namespace, so another process can explicitly + // reserve our pid as a port ID. In this case, a negative portid value will be + // set. + if (static_cast(addr.nl_pid) > 0) { + EXPECT_EQ(addr.nl_pid, getpid()); + } + + memset(&addr, 0, sizeof(addr)); + addr.nl_family = AF_NETLINK; + + // Connecting again is allowed, but keeps the same port. + EXPECT_THAT(connect(fd.get(), reinterpret_cast(&addr), + sizeof(addr)), + SyscallSucceeds()); + + addrlen = sizeof(addr); + EXPECT_THAT(getsockname(fd.get(), reinterpret_cast(&addr), + &addrlen), + SyscallSucceeds()); + EXPECT_EQ(addrlen, sizeof(addr)); + EXPECT_EQ(addr.nl_pid, getpid()); +} + +TEST_P(NetlinkTest, GetPeerName) { + const int protocol = GetParam(); + + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_NETLINK, SOCK_RAW, protocol)); + + struct sockaddr_nl addr = {}; + socklen_t addrlen = sizeof(addr); + + EXPECT_THAT(getpeername(fd.get(), reinterpret_cast(&addr), + &addrlen), + SyscallSucceeds()); + + EXPECT_EQ(addrlen, sizeof(addr)); + EXPECT_EQ(addr.nl_family, AF_NETLINK); + // Peer is the kernel if we didn't connect elsewhere. + EXPECT_EQ(addr.nl_pid, 0); +} + +INSTANTIATE_TEST_SUITE_P(ProtocolTest, NetlinkTest, + ::testing::Values(NETLINK_ROUTE, + NETLINK_KOBJECT_UEVENT)); + +} // namespace + +} // namespace testing +} // namespace gvisor diff --git a/test/syscalls/linux/socket_netlink_route.cc b/test/syscalls/linux/socket_netlink_route.cc index be0dadcd6..ef567f512 100644 --- a/test/syscalls/linux/socket_netlink_route.cc +++ b/test/syscalls/linux/socket_netlink_route.cc @@ -41,112 +41,7 @@ namespace { using ::testing::AnyOf; using ::testing::Eq; -// Netlink sockets must be SOCK_DGRAM or SOCK_RAW. -TEST(NetlinkRouteTest, Types) { - EXPECT_THAT(socket(AF_NETLINK, SOCK_STREAM, NETLINK_ROUTE), - SyscallFailsWithErrno(ESOCKTNOSUPPORT)); - EXPECT_THAT(socket(AF_NETLINK, SOCK_SEQPACKET, NETLINK_ROUTE), - SyscallFailsWithErrno(ESOCKTNOSUPPORT)); - EXPECT_THAT(socket(AF_NETLINK, SOCK_RDM, NETLINK_ROUTE), - SyscallFailsWithErrno(ESOCKTNOSUPPORT)); - EXPECT_THAT(socket(AF_NETLINK, SOCK_DCCP, NETLINK_ROUTE), - SyscallFailsWithErrno(ESOCKTNOSUPPORT)); - EXPECT_THAT(socket(AF_NETLINK, SOCK_PACKET, NETLINK_ROUTE), - SyscallFailsWithErrno(ESOCKTNOSUPPORT)); - - int fd; - EXPECT_THAT(fd = socket(AF_NETLINK, SOCK_DGRAM, NETLINK_ROUTE), - SyscallSucceeds()); - EXPECT_THAT(close(fd), SyscallSucceeds()); - - EXPECT_THAT(fd = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE), - SyscallSucceeds()); - EXPECT_THAT(close(fd), SyscallSucceeds()); -} - -TEST(NetlinkRouteTest, AutomaticPort) { - FileDescriptor fd = - ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE)); - - struct sockaddr_nl addr = {}; - addr.nl_family = AF_NETLINK; - - EXPECT_THAT( - bind(fd.get(), reinterpret_cast(&addr), sizeof(addr)), - SyscallSucceeds()); - - socklen_t addrlen = sizeof(addr); - EXPECT_THAT(getsockname(fd.get(), reinterpret_cast(&addr), - &addrlen), - SyscallSucceeds()); - EXPECT_EQ(addrlen, sizeof(addr)); - // This is the only netlink socket in the process, so it should get the PID as - // the port id. - // - // N.B. Another process could theoretically have explicitly reserved our pid - // as a port ID, but that is very unlikely. - EXPECT_EQ(addr.nl_pid, getpid()); -} - -// Calling connect automatically binds to an automatic port. -TEST(NetlinkRouteTest, ConnectBinds) { - FileDescriptor fd = - ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE)); - - struct sockaddr_nl addr = {}; - addr.nl_family = AF_NETLINK; - - EXPECT_THAT(connect(fd.get(), reinterpret_cast(&addr), - sizeof(addr)), - SyscallSucceeds()); - - socklen_t addrlen = sizeof(addr); - EXPECT_THAT(getsockname(fd.get(), reinterpret_cast(&addr), - &addrlen), - SyscallSucceeds()); - EXPECT_EQ(addrlen, sizeof(addr)); - - // Each test is running in a pid namespace, so another process can explicitly - // reserve our pid as a port ID. In this case, a negative portid value will be - // set. - if (static_cast(addr.nl_pid) > 0) { - EXPECT_EQ(addr.nl_pid, getpid()); - } - - memset(&addr, 0, sizeof(addr)); - addr.nl_family = AF_NETLINK; - - // Connecting again is allowed, but keeps the same port. - EXPECT_THAT(connect(fd.get(), reinterpret_cast(&addr), - sizeof(addr)), - SyscallSucceeds()); - - addrlen = sizeof(addr); - EXPECT_THAT(getsockname(fd.get(), reinterpret_cast(&addr), - &addrlen), - SyscallSucceeds()); - EXPECT_EQ(addrlen, sizeof(addr)); - EXPECT_EQ(addr.nl_pid, getpid()); -} - -TEST(NetlinkRouteTest, GetPeerName) { - FileDescriptor fd = - ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE)); - - struct sockaddr_nl addr = {}; - socklen_t addrlen = sizeof(addr); - - EXPECT_THAT(getpeername(fd.get(), reinterpret_cast(&addr), - &addrlen), - SyscallSucceeds()); - - EXPECT_EQ(addrlen, sizeof(addr)); - EXPECT_EQ(addr.nl_family, AF_NETLINK); - // Peer is the kernel if we didn't connect elsewhere. - EXPECT_EQ(addr.nl_pid, 0); -} - -// Parameters for GetSockOpt test. They are: +// Parameters for SockOptTest. They are: // 0: Socket option to query. // 1: A predicate to run on the returned sockopt value. Should return true if // the value is considered ok. @@ -219,7 +114,8 @@ void CheckGetLinkResponse(const struct nlmsghdr* hdr, int seq, int port) { } TEST(NetlinkRouteTest, GetLinkDump) { - FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket()); + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); uint32_t port = ASSERT_NO_ERRNO_AND_VALUE(NetlinkPortID(fd.get())); struct request { @@ -260,7 +156,8 @@ TEST(NetlinkRouteTest, GetLinkDump) { } TEST(NetlinkRouteTest, MsgHdrMsgUnsuppType) { - FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket()); + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); struct request { struct nlmsghdr hdr; @@ -293,7 +190,8 @@ TEST(NetlinkRouteTest, MsgHdrMsgUnsuppType) { } TEST(NetlinkRouteTest, MsgHdrMsgTrunc) { - FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket()); + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); struct request { struct nlmsghdr hdr; @@ -332,7 +230,8 @@ TEST(NetlinkRouteTest, MsgHdrMsgTrunc) { } TEST(NetlinkRouteTest, MsgTruncMsgHdrMsgTrunc) { - FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket()); + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); struct request { struct nlmsghdr hdr; @@ -373,7 +272,8 @@ TEST(NetlinkRouteTest, MsgTruncMsgHdrMsgTrunc) { } TEST(NetlinkRouteTest, ControlMessageIgnored) { - FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket()); + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); uint32_t port = ASSERT_NO_ERRNO_AND_VALUE(NetlinkPortID(fd.get())); struct request { @@ -408,7 +308,8 @@ TEST(NetlinkRouteTest, ControlMessageIgnored) { } TEST(NetlinkRouteTest, GetAddrDump) { - FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket()); + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); uint32_t port = ASSERT_NO_ERRNO_AND_VALUE(NetlinkPortID(fd.get())); struct request { @@ -468,7 +369,8 @@ TEST(NetlinkRouteTest, LookupAll) { // GetRouteDump tests a RTM_GETROUTE + NLM_F_DUMP request. TEST(NetlinkRouteTest, GetRouteDump) { - FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket()); + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); uint32_t port = ASSERT_NO_ERRNO_AND_VALUE(NetlinkPortID(fd.get())); struct request { @@ -544,7 +446,8 @@ TEST(NetlinkRouteTest, GetRouteDump) { // buffer. MSG_TRUNC with a zero length buffer should consume subsequent // messages off the socket. TEST(NetlinkRouteTest, RecvmsgTrunc) { - FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket()); + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); struct request { struct nlmsghdr hdr; @@ -620,7 +523,8 @@ TEST(NetlinkRouteTest, RecvmsgTrunc) { // it, so a properly sized buffer can be allocated to store the message. This // test tests that scenario. TEST(NetlinkRouteTest, RecvmsgTruncPeek) { - FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket()); + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); struct request { struct nlmsghdr hdr; @@ -695,7 +599,8 @@ TEST(NetlinkRouteTest, RecvmsgTruncPeek) { // No SCM_CREDENTIALS are received without SO_PASSCRED set. TEST(NetlinkRouteTest, NoPasscredNoCreds) { - FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket()); + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); ASSERT_THAT(setsockopt(fd.get(), SOL_SOCKET, SO_PASSCRED, &kSockOptOff, sizeof(kSockOptOff)), @@ -742,7 +647,8 @@ TEST(NetlinkRouteTest, NoPasscredNoCreds) { // SCM_CREDENTIALS are received with SO_PASSCRED set. TEST(NetlinkRouteTest, PasscredCreds) { - FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket()); + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); ASSERT_THAT(setsockopt(fd.get(), SOL_SOCKET, SO_PASSCRED, &kSockOptOn, sizeof(kSockOptOn)), diff --git a/test/syscalls/linux/socket_netlink_uevent.cc b/test/syscalls/linux/socket_netlink_uevent.cc new file mode 100644 index 000000000..da425bed4 --- /dev/null +++ b/test/syscalls/linux/socket_netlink_uevent.cc @@ -0,0 +1,83 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include +#include +#include +#include +#include + +#include "gtest/gtest.h" +#include "test/syscalls/linux/socket_netlink_util.h" +#include "test/syscalls/linux/socket_test_util.h" +#include "test/util/file_descriptor.h" +#include "test/util/test_util.h" + +// Tests for NETLINK_KOBJECT_UEVENT sockets. +// +// gVisor never sends any messages on these sockets, so we don't test the events +// themselves. + +namespace gvisor { +namespace testing { + +namespace { + +// SO_PASSCRED can be enabled. Since no messages are sent in gVisor, we don't +// actually test receiving credentials. +TEST(NetlinkUeventTest, PassCred) { + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_KOBJECT_UEVENT)); + + EXPECT_THAT(setsockopt(fd.get(), SOL_SOCKET, SO_PASSCRED, &kSockOptOn, + sizeof(kSockOptOn)), + SyscallSucceeds()); +} + +// SO_DETACH_FILTER fails without a filter already installed. +TEST(NetlinkUeventTest, DetachNoFilter) { + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_KOBJECT_UEVENT)); + + int opt; + EXPECT_THAT( + setsockopt(fd.get(), SOL_SOCKET, SO_DETACH_FILTER, &opt, sizeof(opt)), + SyscallFailsWithErrno(ENOENT)); +} + +// We can attach a BPF filter. +TEST(NetlinkUeventTest, AttachFilter) { + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_KOBJECT_UEVENT)); + + // Minimal BPF program: a single ret. + struct sock_filter filter = {0x6, 0, 0, 0}; + struct sock_fprog prog = {}; + prog.len = 1; + prog.filter = &filter; + + EXPECT_THAT( + setsockopt(fd.get(), SOL_SOCKET, SO_ATTACH_FILTER, &prog, sizeof(prog)), + SyscallSucceeds()); + + int opt; + EXPECT_THAT( + setsockopt(fd.get(), SOL_SOCKET, SO_DETACH_FILTER, &opt, sizeof(opt)), + SyscallSucceeds()); +} + +} // namespace + +} // namespace testing +} // namespace gvisor diff --git a/test/syscalls/linux/socket_netlink_util.cc b/test/syscalls/linux/socket_netlink_util.cc index fcb8f8a88..5f05bab10 100644 --- a/test/syscalls/linux/socket_netlink_util.cc +++ b/test/syscalls/linux/socket_netlink_util.cc @@ -16,7 +16,6 @@ #include #include -#include #include @@ -27,9 +26,9 @@ namespace gvisor { namespace testing { -PosixErrorOr NetlinkBoundSocket() { +PosixErrorOr NetlinkBoundSocket(int protocol) { FileDescriptor fd; - ASSIGN_OR_RETURN_ERRNO(fd, Socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE)); + ASSIGN_OR_RETURN_ERRNO(fd, Socket(AF_NETLINK, SOCK_RAW, protocol)); struct sockaddr_nl addr = {}; addr.nl_family = AF_NETLINK; diff --git a/test/syscalls/linux/socket_netlink_util.h b/test/syscalls/linux/socket_netlink_util.h index db8639a2f..da99f0d60 100644 --- a/test/syscalls/linux/socket_netlink_util.h +++ b/test/syscalls/linux/socket_netlink_util.h @@ -17,7 +17,6 @@ #include #include -#include #include "test/util/file_descriptor.h" #include "test/util/posix_error.h" @@ -25,8 +24,8 @@ namespace gvisor { namespace testing { -// Returns a bound NETLINK_ROUTE socket. -PosixErrorOr NetlinkBoundSocket(); +// Returns a bound netlink socket. +PosixErrorOr NetlinkBoundSocket(int protocol); // Returns the port ID of the passed socket. PosixErrorOr NetlinkPortID(int fd); -- cgit v1.2.3 From 4fdd69d681bb3abb68a043377a2fb0ec8a031d54 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Mon, 4 Nov 2019 10:56:13 -0800 Subject: Check that a file is a regular file with open(O_TRUNC). It was possible to panic the sentry by opening a cache revalidating folder with O_TRUNC|O_CREAT. PiperOrigin-RevId: 278417533 --- pkg/sentry/fs/inode_operations.go | 2 ++ pkg/sentry/syscalls/linux/sys_file.go | 8 ++++---- 2 files changed, 6 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/inode_operations.go b/pkg/sentry/fs/inode_operations.go index 5cde9d215..d6c35c2dc 100644 --- a/pkg/sentry/fs/inode_operations.go +++ b/pkg/sentry/fs/inode_operations.go @@ -221,6 +221,8 @@ type InodeOperations interface { // sys_ftruncate. // // Implementations need not check that length >= 0. + // + // Truncate must only be called on regular files. Truncate(ctx context.Context, inode *Inode, size int64) error // Allocate allows the caller to reserve disk space for the inode. diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go index b9a8e3e21..c9f57fe27 100644 --- a/pkg/sentry/syscalls/linux/sys_file.go +++ b/pkg/sentry/syscalls/linux/sys_file.go @@ -169,7 +169,7 @@ func openAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint) (fd uint if dirPath { return syserror.ENOTDIR } - if flags&linux.O_TRUNC != 0 { + if flags&linux.O_TRUNC != 0 && fs.IsRegular(d.Inode.StableAttr) { if err := d.Inode.Truncate(t, d, 0); err != nil { return err } @@ -397,7 +397,7 @@ func createAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint, mode l } // Should we truncate the file? - if flags&linux.O_TRUNC != 0 { + if flags&linux.O_TRUNC != 0 && fs.IsRegular(found.Inode.StableAttr) { if err := found.Inode.Truncate(t, found, 0); err != nil { return err } @@ -1483,7 +1483,7 @@ func Truncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if fs.IsDir(d.Inode.StableAttr) { return syserror.EISDIR } - if !fs.IsFile(d.Inode.StableAttr) { + if !fs.IsRegular(d.Inode.StableAttr) { return syserror.EINVAL } @@ -1523,7 +1523,7 @@ func Ftruncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys // Note that this is different from truncate(2) above, where a // directory returns EISDIR. - if !fs.IsFile(file.Dirent.Inode.StableAttr) { + if !fs.IsRegular(file.Dirent.Inode.StableAttr) { return 0, nil, syserror.EINVAL } -- cgit v1.2.3 From 66ebb6575f929a389d3c929977ed5e31d706fcfe Mon Sep 17 00:00:00 2001 From: Bhasker Hariharan Date: Thu, 7 Nov 2019 09:45:26 -0800 Subject: Add support for TIME_WAIT timeout. This change adds explicit support for honoring the 2MSL timeout for sockets in TIME_WAIT state. It also adds support for the TCP_LINGER2 option that allows modification of the FIN_WAIT2 state timeout duration for a given socket. It also adds an option to modify the Stack wide TIME_WAIT timeout but this is only for testing. On Linux this is fixed at 60s. Further, we also now correctly process RST's in CLOSE_WAIT and close the socket similar to linux without moving it to error state. We also now handle SYN in ESTABLISHED state as per RFC5961#section-4.1. Earlier we would just drop these SYNs. Which can result in some tests that pass on linux to fail on gVisor. Netstack now honors TIME_WAIT correctly as well as handles the following cases correctly. - TCP RSTs in TIME_WAIT are ignored. - A duplicate TCP FIN during TIME_WAIT extends the TIME_WAIT and a dup ACK is sent in response to the FIN as the dup FIN indicates potential loss of the original final ACK. - An out of order segment during TIME_WAIT generates a dup ACK. - A new SYN w/ a sequence number > the highest sequence number in the previous connection closes the TIME_WAIT early and opens a new connection. Further to make the SYN case work correctly the ISN (Initial Sequence Number) generation for Netstack has been updated to be as per RFC. Its not a pure random number anymore and follows the recommendation in https://tools.ietf.org/html/rfc6528#page-3. The current hash used is not a cryptographically secure hash function. A separate change will update the hash function used to Siphash similar to what is used in Linux. PiperOrigin-RevId: 279106406 --- pkg/sentry/socket/netstack/netstack.go | 20 + pkg/tcpip/adapters/gonet/gonet_test.go | 12 +- pkg/tcpip/stack/stack.go | 20 +- pkg/tcpip/stack/transport_demuxer.go | 33 +- pkg/tcpip/tcpip.go | 12 +- pkg/tcpip/transport/tcp/BUILD | 2 +- pkg/tcpip/transport/tcp/accept.go | 17 +- pkg/tcpip/transport/tcp/connect.go | 322 ++++++++++++-- pkg/tcpip/transport/tcp/endpoint.go | 101 ++++- pkg/tcpip/transport/tcp/endpoint_state.go | 26 +- pkg/tcpip/transport/tcp/protocol.go | 43 ++ pkg/tcpip/transport/tcp/rcv.go | 167 ++++++- pkg/tcpip/transport/tcp/tcp_test.go | 622 ++++++++++++++++++++++++++- test/syscalls/BUILD | 22 +- test/syscalls/linux/BUILD | 1 + test/syscalls/linux/socket_inet_loopback.cc | 336 +++++++++++++++ test/syscalls/linux/socket_ip_tcp_generic.cc | 93 ++++ 17 files changed, 1736 insertions(+), 113 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 27c6692c4..d92399efd 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -1173,6 +1173,18 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa copy(b, v) return b, nil + case linux.TCP_LINGER2: + if outLen < sizeOfInt32 { + return nil, syserr.ErrInvalidArgument + } + + var v tcpip.TCPLingerTimeoutOption + if err := ep.GetSockOpt(&v); err != nil { + return nil, syserr.TranslateNetstackError(err) + } + + return int32(time.Duration(v) / time.Second), nil + default: emitUnimplementedEventTCP(t, name) } @@ -1556,6 +1568,14 @@ func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) * } return nil + case linux.TCP_LINGER2: + if len(optVal) < sizeOfInt32 { + return syserr.ErrInvalidArgument + } + + v := usermem.ByteOrder.Uint32(optVal) + return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.TCPLingerTimeoutOption(time.Second * time.Duration(v)))) + case linux.TCP_REPAIR_OPTIONS: t.Kernel().EmitUnimplementedEvent(t) diff --git a/pkg/tcpip/adapters/gonet/gonet_test.go b/pkg/tcpip/adapters/gonet/gonet_test.go index 8ced960bb..ee077ae83 100644 --- a/pkg/tcpip/adapters/gonet/gonet_test.go +++ b/pkg/tcpip/adapters/gonet/gonet_test.go @@ -151,10 +151,8 @@ func TestCloseReader(t *testing.T) { buf := make([]byte, 256) n, err := c.Read(buf) - got, ok := err.(*net.OpError) - want := tcpip.ErrConnectionAborted - if n != 0 || !ok || got.Err.Error() != want.String() { - t.Errorf("c.Read() = (%d, %v), want (0, OpError(%v))", n, err, want) + if n != 0 || err != io.EOF { + t.Errorf("c.Read() = (%d, %v), want (0, EOF)", n, err) } }() sender, err := connect(s, addr) @@ -203,10 +201,8 @@ func TestCloseReaderWithForwarder(t *testing.T) { buf := make([]byte, 256) n, e := c.Read(buf) - got, ok := e.(*net.OpError) - want := tcpip.ErrConnectionAborted - if n != 0 || !ok || got.Err.Error() != want.String() { - t.Errorf("c.Read() = (%d, %v), want (0, OpError(%v))", n, e, want) + if n != 0 || e != io.EOF { + t.Errorf("c.Read() = (%d, %v), want (0, EOF)", n, e) } }) s.SetTransportProtocolHandler(tcp.ProtocolNumber, fwd.HandlePacket) diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index 99809df75..2f8d8e822 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -402,11 +402,11 @@ type Stack struct { // by the stack. icmpRateLimiter *ICMPRateLimiter - // portSeed is a one-time random value initialized at stack startup + // seed is a one-time random value initialized at stack startup // and is used to seed the TCP port picking on active connections // // TODO(gvisor.dev/issue/940): S/R this field. - portSeed uint32 + seed uint32 // ndpConfigs is the default NDP configurations used by interfaces. ndpConfigs NDPConfigurations @@ -544,7 +544,7 @@ func New(opts Options) *Stack { stats: opts.Stats.FillIn(), handleLocal: opts.HandleLocal, icmpRateLimiter: NewICMPRateLimiter(), - portSeed: generateRandUint32(), + seed: generateRandUint32(), ndpConfigs: opts.NDPConfigs, autoGenIPv6LinkLocal: opts.AutoGenIPv6LinkLocal, uniqueIDGenerator: opts.UniqueID, @@ -1186,6 +1186,12 @@ func (s *Stack) CompleteTransportEndpointCleanup(ep TransportEndpoint) { s.mu.Unlock() } +// FindTransportEndpoint finds an endpoint that most closely matches the provided +// id. If no endpoint is found it returns nil. +func (s *Stack) FindTransportEndpoint(netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, id TransportEndpointID, r *Route) TransportEndpoint { + return s.demux.findTransportEndpoint(netProto, transProto, id, r) +} + // RegisterRawTransportEndpoint registers the given endpoint with the stack // transport dispatcher. Received packets that match the provided transport // protocol will be delivered to the given endpoint. @@ -1573,12 +1579,12 @@ func (s *Stack) HandleNDPRA(id tcpip.NICID, ip tcpip.Address, ra header.NDPRoute return nil } -// PortSeed returns a 32 bit value that can be used as a seed value for port -// picking. +// Seed returns a 32 bit value that can be used as a seed value for port +// picking, ISN generation etc. // // NOTE: The seed is generated once during stack initialization only. -func (s *Stack) PortSeed() uint32 { - return s.portSeed +func (s *Stack) Seed() uint32 { + return s.seed } func generateRandUint32() uint32 { diff --git a/pkg/tcpip/stack/transport_demuxer.go b/pkg/tcpip/stack/transport_demuxer.go index 594570216..cb805522b 100644 --- a/pkg/tcpip/stack/transport_demuxer.go +++ b/pkg/tcpip/stack/transport_demuxer.go @@ -103,7 +103,6 @@ func (epsByNic *endpointsByNic) handlePacket(r *Route, id TransportEndpointID, p epsByNic.mu.RUnlock() // Don't use defer for performance reasons. return } - // multiPortEndpoints are guaranteed to have at least one element. selectEndpoint(id, mpep, epsByNic.seed).HandlePacket(r, id, pkt) epsByNic.mu.RUnlock() // Don't use defer for performance reasons. @@ -507,10 +506,40 @@ func (d *transportDemuxer) findAllEndpointsLocked(eps *transportEndpoints, id Tr if ep, ok := eps.endpoints[nid]; ok { matchedEPs = append(matchedEPs, ep) } - return matchedEPs } +// findTransportEndpoint find a single endpoint that most closely matches the provided id. +func (d *transportDemuxer) findTransportEndpoint(netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, id TransportEndpointID, r *Route) TransportEndpoint { + eps, ok := d.protocol[protocolIDs{netProto, transProto}] + if !ok { + return nil + } + // Try to find the endpoint. + eps.mu.RLock() + epsByNic := d.findEndpointLocked(eps, id) + // Fail if we didn't find one. + if epsByNic == nil { + eps.mu.RUnlock() + return nil + } + + epsByNic.mu.RLock() + eps.mu.RUnlock() + + mpep, ok := epsByNic.endpoints[r.ref.nic.ID()] + if !ok { + if mpep, ok = epsByNic.endpoints[0]; !ok { + epsByNic.mu.RUnlock() // Don't use defer for performance reasons. + return nil + } + } + + ep := selectEndpoint(id, mpep, epsByNic.seed) + epsByNic.mu.RUnlock() + return ep +} + // findEndpointLocked returns the endpoint that most closely matches the given // id. func (d *transportDemuxer) findEndpointLocked(eps *transportEndpoints, id TransportEndpointID) *endpointsByNic { diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 3edb513d4..bd5eb89ca 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -586,6 +586,16 @@ type MaxSegOption int // A zero value indicates the default. type TTLOption uint8 +// TCPLingerTimeoutOption is used by SetSockOpt/GetSockOpt to set/get the +// maximum duration for which a socket lingers in the TCP_FIN_WAIT_2 state +// before being marked closed. +type TCPLingerTimeoutOption time.Duration + +// TCPTimeWaitTimeoutOption is used by SetSockOpt/GetSockOpt to set/get the +// maximum duration for which a socket lingers in the TIME_WAIT state +// before being marked closed. +type TCPTimeWaitTimeoutOption time.Duration + // MulticastTTLOption is used by SetSockOpt/GetSockOpt to control the default // TTL value for multicast messages. The default is 1. type MulticastTTLOption uint8 @@ -1329,8 +1339,8 @@ var ( // GetDanglingEndpoints returns all dangling endpoints. func GetDanglingEndpoints() []Endpoint { - es := make([]Endpoint, 0, len(danglingEndpoints)) danglingEndpointsMu.Lock() + es := make([]Endpoint, 0, len(danglingEndpoints)) for e := range danglingEndpoints { es = append(es, e) } diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD index f1dbc6f91..3f47b328d 100644 --- a/pkg/tcpip/transport/tcp/BUILD +++ b/pkg/tcpip/transport/tcp/BUILD @@ -71,7 +71,7 @@ filegroup( go_test( name = "tcp_test", - size = "small", + size = "medium", srcs = [ "dual_stack_test.go", "sack_scoreboard_test.go", diff --git a/pkg/tcpip/transport/tcp/accept.go b/pkg/tcpip/transport/tcp/accept.go index cb0e13ebc..0e8e0a2b4 100644 --- a/pkg/tcpip/transport/tcp/accept.go +++ b/pkg/tcpip/transport/tcp/accept.go @@ -269,8 +269,8 @@ func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, i func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *header.TCPSynOptions) (*endpoint, *tcpip.Error) { // Create new endpoint. irs := s.sequenceNumber - cookie := l.createCookie(s.id, irs, encodeMSS(opts.MSS)) - ep, err := l.createConnectingEndpoint(s, cookie, irs, opts) + isn := generateSecureISN(s.id, l.stack.Seed()) + ep, err := l.createConnectingEndpoint(s, isn, irs, opts) if err != nil { return nil, err } @@ -289,7 +289,7 @@ func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *head // Perform the 3-way handshake. h := newHandshake(ep, seqnum.Size(ep.initialReceiveWindow())) - h.resetToSynRcvd(cookie, irs, opts) + h.resetToSynRcvd(isn, irs, opts) if err := h.execute(); err != nil { ep.Close() if l.listenEP != nil { @@ -361,6 +361,7 @@ func (e *endpoint) handleSynSegment(ctx *listenContext, s *segment, opts *header defer decSynRcvdCount() defer e.decSynRcvdCount() defer s.decRef() + n, err := ctx.createEndpointAndPerformHandshake(s, opts) if err != nil { e.stack.Stats().TCP.FailedConnectionAttempts.Increment() @@ -368,6 +369,11 @@ func (e *endpoint) handleSynSegment(ctx *listenContext, s *segment, opts *header return } ctx.removePendingEndpoint(n) + // Start the protocol goroutine. + wq := &waiter.Queue{} + n.startAcceptedLoop(wq) + e.stack.Stats().TCP.PassiveConnectionOpenings.Increment() + e.deliverAccepted(n) } @@ -543,6 +549,11 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { // number of goroutines as we do check before // entering here that there was at least some // space available in the backlog. + + // Start the protocol goroutine. + wq := &waiter.Queue{} + n.startAcceptedLoop(wq) + e.stack.Stats().TCP.PassiveConnectionOpenings.Increment() go e.deliverAccepted(n) } } diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index ca982c451..a114c06c1 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -15,6 +15,7 @@ package tcp import ( + "encoding/binary" "sync" "time" @@ -22,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/hash/jenkins" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/seqnum" "gvisor.dev/gvisor/pkg/tcpip/stack" @@ -139,7 +141,32 @@ func (h *handshake) resetState() { h.flags = header.TCPFlagSyn h.ackNum = 0 h.mss = 0 - h.iss = seqnum.Value(uint32(b[0]) | uint32(b[1])<<8 | uint32(b[2])<<16 | uint32(b[3])<<24) + h.iss = generateSecureISN(h.ep.ID, h.ep.stack.Seed()) +} + +// generateSecureISN generates a secure Initial Sequence number based on the +// recommendation here https://tools.ietf.org/html/rfc6528#page-3. +func generateSecureISN(id stack.TransportEndpointID, seed uint32) seqnum.Value { + isnHasher := jenkins.Sum32(seed) + isnHasher.Write([]byte(id.LocalAddress)) + isnHasher.Write([]byte(id.RemoteAddress)) + portBuf := make([]byte, 2) + binary.LittleEndian.PutUint16(portBuf, id.LocalPort) + isnHasher.Write(portBuf) + binary.LittleEndian.PutUint16(portBuf, id.RemotePort) + isnHasher.Write(portBuf) + // The time period here is 64ns. This is similar to what linux uses + // generate a sequence number that overlaps less than one + // time per MSL (2 minutes). + // + // A 64ns clock ticks 10^9/64 = 15625000) times in a second. + // To wrap the whole 32 bit space would require + // 2^32/1562500 ~ 274 seconds. + // + // Which sort of guarantees that we won't reuse the ISN for a new + // connection for the same tuple for at least 274s. + isn := isnHasher.Sum32() + uint32(time.Now().UnixNano()>>6) + return seqnum.Value(isn) } // effectiveRcvWndScale returns the effective receive window scale to be used. @@ -809,7 +836,19 @@ func (e *endpoint) resetConnectionLocked(err *tcpip.Error) { e.state = StateError e.HardError = err if err != tcpip.ErrConnectionReset { - e.sendRaw(buffer.VectorisedView{}, header.TCPFlagAck|header.TCPFlagRst, e.snd.sndUna, e.rcv.rcvNxt, 0) + // The exact sequence number to be used for the RST is the same as the + // one used by Linux. We need to handle the case of window being shrunk + // which can cause sndNxt to be outside the acceptable window on the + // receiver. + // + // See: https://www.snellman.net/blog/archive/2016-02-01-tcp-rst/ for more + // information. + sndWndEnd := e.snd.sndUna.Add(e.snd.sndWnd) + resetSeqNum := sndWndEnd + if !sndWndEnd.LessThan(e.snd.sndNxt) || e.snd.sndNxt.Size(sndWndEnd) < (1< + // + // After sending the acknowledgment, TCP MUST drop the unacceptable + // segment and stop processing further. + // + // By sending an ACK, the remote peer is challenged to confirm the loss + // of the previous connection and the request to start a new connection. + // A legitimate peer, after restart, would not have a TCB in the + // synchronized state. Thus, when the ACK arrives, the peer should send + // a RST segment back with the sequence number derived from the ACK + // field that caused the RST. + + // This RST will confirm that the remote peer has indeed closed the + // previous connection. Upon receipt of a valid RST, the local TCP + // endpoint MUST terminate its connection. The local TCP endpoint + // should then rely on SYN retransmission from the remote end to + // re-establish the connection. + + e.snd.sendAck() } else if s.flagIsSet(header.TCPFlagAck) { // Patch the window size in the segment according to the // send window scale. @@ -856,7 +960,15 @@ func (e *endpoint) handleSegments() *tcpip.Error { // RFC 793, page 41 states that "once in the ESTABLISHED // state all segments must carry current acknowledgment // information." - e.rcv.handleRcvdSegment(s) + drop, err := e.rcv.handleRcvdSegment(s) + if err != nil { + s.decRef() + return err + } + if drop { + s.decRef() + continue + } e.snd.handleRcvdSegment(s) } s.decRef() @@ -955,7 +1067,6 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { } e.mu.Unlock() - // When the protocol loop exits we should wake up our waiters. e.waiterQueue.Notify(waiter.EventHUp | waiter.EventErr | waiter.EventIn | waiter.EventOut) } @@ -1001,6 +1112,10 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { // RTT itself. e.rcvAutoParams.prevCopied = initialRcvWnd e.rcvListMu.Unlock() + e.stack.Stats().TCP.CurrentEstablished.Increment() + e.mu.Lock() + e.state = StateEstablished + e.mu.Unlock() } e.keepalive.timer.init(&e.keepalive.waker) @@ -1008,10 +1123,6 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { // Tell waiters that the endpoint is connected and writable. e.mu.Lock() - if e.state != StateEstablished { - e.stack.Stats().TCP.CurrentEstablished.Increment() - e.state = StateEstablished - } drained := e.drainDone != nil e.mu.Unlock() if drained { @@ -1042,7 +1153,13 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { { w: &closeWaker, f: func() *tcpip.Error { - return tcpip.ErrConnectionAborted + // This means the socket is being closed due + // to the TCP_FIN_WAIT2 timeout was hit. Just + // mark the socket as closed. + e.mu.Lock() + e.state = StateClose + e.mu.Unlock() + return nil }, }, { @@ -1085,17 +1202,18 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { e.resetConnectionLocked(tcpip.ErrConnectionAborted) e.mu.Unlock() } + if n¬ifyClose != 0 && closeTimer == nil { - // Reset the connection 3 seconds after - // the endpoint has been closed. - // - // The timer could fire in background - // when the endpoint is drained. That's - // OK as the loop here will not honor - // the firing until the undrain arrives. - closeTimer = time.AfterFunc(3*time.Second, func() { - closeWaker.Assert() - }) + e.mu.Lock() + if e.state == StateFinWait2 && e.closed { + // The socket has been closed and we are in FIN_WAIT2 + // so start the FIN_WAIT2 timer. + closeTimer = time.AfterFunc(e.tcpLingerTimeout, func() { + closeWaker.Assert() + }) + e.waiterQueue.Notify(waiter.EventHUp | waiter.EventErr | waiter.EventIn | waiter.EventOut) + } + e.mu.Unlock() } if n¬ifyKeepaliveChanged != 0 { @@ -1117,6 +1235,12 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { } } + if n¬ifyTickleWorker != 0 { + // Just a tickle notification. No need to do + // anything. + return nil + } + return nil }, }, @@ -1143,15 +1267,16 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { } e.rcvListMu.Unlock() - e.mu.RLock() + e.mu.Lock() if e.workerCleanup { e.notifyProtocolGoroutine(notifyClose) } - e.mu.RUnlock() // Main loop. Handle segments until both send and receive ends of the // connection have completed. - for !e.rcv.closed || !e.snd.closed || e.snd.sndUna != e.snd.sndNxtList { + + for e.state != StateTimeWait && e.state != StateClose && e.state != StateError { + e.mu.Unlock() e.workMu.Unlock() v, _ := s.Fetch(true) e.workMu.Lock() @@ -1167,6 +1292,23 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { return nil } + e.mu.Lock() + } + + state := e.state + e.mu.Unlock() + var reuseTW func() + if state == StateTimeWait { + // Disable close timer as we now entering real TIME_WAIT. + if closeTimer != nil { + closeTimer.Stop() + } + // Mark the current sleeper done so as to free all associated + // wakers. + s.Done() + // Wake up any waiters before we enter TIME_WAIT. + e.waiterQueue.Notify(waiter.EventHUp | waiter.EventErr | waiter.EventIn | waiter.EventOut) + reuseTW = e.doTimeWait() } // Mark endpoint as closed. @@ -1176,8 +1318,130 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { e.stack.Stats().TCP.CurrentEstablished.Decrement() e.state = StateClose } + // Lock released below. epilogue() + // A new SYN was received during TIME_WAIT and we need to abort + // the timewait and redirect the segment to the listener queue + if reuseTW != nil { + reuseTW() + } + return nil } + +// handleTimeWaitSegments processes segments received during TIME_WAIT +// state. +func (e *endpoint) handleTimeWaitSegments() (extendTimeWait bool, reuseTW func()) { + checkRequeue := true + for i := 0; i < maxSegmentsPerWake; i++ { + s := e.segmentQueue.dequeue() + if s == nil { + checkRequeue = false + break + } + extTW, newSyn := e.rcv.handleTimeWaitSegment(s) + if newSyn { + info := e.EndpointInfo.TransportEndpointInfo + newID := info.ID + newID.RemoteAddress = "" + newID.RemotePort = 0 + netProtos := []tcpip.NetworkProtocolNumber{info.NetProto} + // If the local address is an IPv4 address then also + // look for IPv6 dual stack endpoints that might be + // listening on the local address. + if newID.LocalAddress.To4() != "" { + netProtos = []tcpip.NetworkProtocolNumber{header.IPv4ProtocolNumber, header.IPv6ProtocolNumber} + } + for _, netProto := range netProtos { + if listenEP := e.stack.FindTransportEndpoint(netProto, info.TransProto, newID, &s.route); listenEP != nil { + tcpEP := listenEP.(*endpoint) + if EndpointState(tcpEP.State()) == StateListen { + reuseTW = func() { + tcpEP.enqueueSegment(s) + } + // We explicitly do not decRef + // the segment as it's still + // valid and being reflected to + // a listening endpoint. + return false, reuseTW + } + } + } + } + if extTW { + extendTimeWait = true + } + s.decRef() + } + if checkRequeue && !e.segmentQueue.empty() { + e.newSegmentWaker.Assert() + } + return extendTimeWait, nil +} + +// doTimeWait is responsible for handling the TCP behaviour once a socket +// enters the TIME_WAIT state. Optionally it can return a closure that +// should be executed after releasing the endpoint registrations. This is +// done in cases where a new SYN is received during TIME_WAIT that carries +// a sequence number larger than one see on the connection. +func (e *endpoint) doTimeWait() (twReuse func()) { + // Trigger a 2 * MSL time wait state. During this period + // we will drop all incoming segments. + // NOTE: On Linux this is not configurable and is fixed at 60 seconds. + timeWaitDuration := DefaultTCPTimeWaitTimeout + + // Get the stack wide configuration. + var tcpTW tcpip.TCPTimeWaitTimeoutOption + if err := e.stack.TransportProtocolOption(ProtocolNumber, &tcpTW); err == nil { + timeWaitDuration = time.Duration(tcpTW) + } + + const newSegment = 1 + const notification = 2 + const timeWaitDone = 3 + + s := sleep.Sleeper{} + s.AddWaker(&e.newSegmentWaker, newSegment) + s.AddWaker(&e.notificationWaker, notification) + + var timeWaitWaker sleep.Waker + s.AddWaker(&timeWaitWaker, timeWaitDone) + timeWaitTimer := time.AfterFunc(timeWaitDuration, timeWaitWaker.Assert) + defer timeWaitTimer.Stop() + + for { + e.workMu.Unlock() + v, _ := s.Fetch(true) + e.workMu.Lock() + switch v { + case newSegment: + extendTimeWait, reuseTW := e.handleTimeWaitSegments() + if reuseTW != nil { + return reuseTW + } + if extendTimeWait { + timeWaitTimer.Reset(timeWaitDuration) + } + case notification: + n := e.fetchNotifications() + if n¬ifyClose != 0 { + return nil + } + if n¬ifyDrain != 0 { + for !e.segmentQueue.empty() { + // Ignore extending TIME_WAIT during a + // save. For sockets in TIME_WAIT we just + // terminate the TIME_WAIT early. + e.handleTimeWaitSegments() + } + close(e.drainDone) + <-e.undrain + return nil + } + case timeWaitDone: + return nil + } + } +} diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 79fec6b77..04c92c04c 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -121,6 +121,11 @@ const ( notifyReset notifyKeepaliveChanged notifyMSSChanged + // notifyTickleWorker is used to tickle the protocol main loop during a + // restore after we update the endpoint state to the correct one. This + // ensures the loop terminates if the final state of the endpoint is + // say TIME_WAIT. + notifyTickleWorker ) // SACKInfo holds TCP SACK related information for a given endpoint. @@ -320,6 +325,11 @@ type endpoint struct { state EndpointState `state:".(EndpointState)"` + // origEndpointState is only used during a restore phase to save the + // endpoint state at restore time as the socket is moved to it's correct + // state. + origEndpointState EndpointState `state:"nosave"` + isPortReserved bool `state:"manual"` isRegistered bool boundNICID tcpip.NICID `state:"manual"` @@ -503,6 +513,16 @@ type endpoint struct { // TODO(b/142022063): Add ability to save and restore per endpoint stats. stats Stats `state:"nosave"` + + // tcpLingerTimeout is the maximum amount of a time a socket + // a socket stays in TIME_WAIT state before being marked + // closed. + tcpLingerTimeout time.Duration + + // closed indicates that the user has called closed on the + // endpoint and at this point the endpoint is only around + // to complete the TCP shutdown. + closed bool } // UniqueID implements stack.TransportEndpoint.UniqueID. @@ -599,6 +619,11 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue e.SetSockOptInt(tcpip.DelayOption, 1) } + var tcpLT tcpip.TCPLingerTimeoutOption + if err := s.TransportProtocolOption(ProtocolNumber, &tcpLT); err == nil { + e.tcpLingerTimeout = time.Duration(tcpLT) + } + if p := s.GetTCPProbe(); p != nil { e.probe = p } @@ -686,6 +711,13 @@ func (e *endpoint) notifyProtocolGoroutine(n uint32) { // with it. It must be called only once and with no other concurrent calls to // the endpoint. func (e *endpoint) Close() { + e.mu.Lock() + closed := e.closed + e.mu.Unlock() + if closed { + return + } + // Issue a shutdown so that the peer knows we won't send any more data // if we're connected, or stop accepting if we're listening. e.Shutdown(tcpip.ShutdownWrite | tcpip.ShutdownRead) @@ -706,6 +738,8 @@ func (e *endpoint) Close() { e.isPortReserved = false } + // Mark endpoint as closed. + e.closed = true // Either perform the local cleanup or kick the worker to make sure it // knows it needs to cleanup. tcpip.AddDanglingEndpoint(e) @@ -731,9 +765,7 @@ func (e *endpoint) closePendingAcceptableConnectionsLocked() { go func() { defer close(done) for n := range e.acceptedChan { - n.mu.Lock() - n.resetConnectionLocked(tcpip.ErrConnectionAborted) - n.mu.Unlock() + n.notifyProtocolGoroutine(notifyReset) n.Close() } }() @@ -1349,6 +1381,28 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.mu.Unlock() return nil + case tcpip.TCPLingerTimeoutOption: + e.mu.Lock() + if v < 0 { + // Same as effectively disabling TCPLinger timeout. + v = 0 + } + var stkTCPLingerTimeout tcpip.TCPLingerTimeoutOption + if err := e.stack.TransportProtocolOption(header.TCPProtocolNumber, &stkTCPLingerTimeout); err != nil { + // We were unable to retrieve a stack config, just use + // the DefaultTCPLingerTimeout. + if v > tcpip.TCPLingerTimeoutOption(DefaultTCPLingerTimeout) { + stkTCPLingerTimeout = tcpip.TCPLingerTimeoutOption(DefaultTCPLingerTimeout) + } + } + // Cap it to the stack wide TCPLinger timeout. + if v > stkTCPLingerTimeout { + v = stkTCPLingerTimeout + } + e.tcpLingerTimeout = time.Duration(v) + e.mu.Unlock() + return nil + default: return nil } @@ -1562,6 +1616,12 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { e.mu.RUnlock() return nil + case *tcpip.TCPLingerTimeoutOption: + e.mu.Lock() + *o = tcpip.TCPLingerTimeoutOption(e.tcpLingerTimeout) + e.mu.Unlock() + return nil + default: return tcpip.ErrUnknownProtocolOption } @@ -1696,7 +1756,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) *tc // src IP to ensure that for a given tuple (srcIP, destIP, // destPort) the offset used as a starting point is the same to // ensure that we can cycle through the port space effectively. - h := jenkins.Sum32(e.stack.PortSeed()) + h := jenkins.Sum32(e.stack.Seed()) h.Write([]byte(e.ID.LocalAddress)) h.Write([]byte(e.ID.RemoteAddress)) portBuf := make([]byte, 2) @@ -1782,9 +1842,8 @@ func (*endpoint) ConnectEndpoint(tcpip.Endpoint) *tcpip.Error { // peer. func (e *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error { e.mu.Lock() - defer e.mu.Unlock() e.shutdownFlags |= flags - + finQueued := false switch { case e.state.connected(): // Close for read. @@ -1799,6 +1858,7 @@ func (e *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error { // the connection with a RST. if (e.shutdownFlags&tcpip.ShutdownWrite) != 0 && rcvBufUsed > 0 { e.notifyProtocolGoroutine(notifyReset) + e.mu.Unlock() return nil } } @@ -1817,14 +1877,11 @@ func (e *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error { s := newSegmentFromView(&e.route, e.ID, nil) e.sndQueue.PushBack(s) e.sndBufInQueue++ - + finQueued = true // Mark endpoint as closed. e.sndClosed = true e.sndBufMu.Unlock() - - // Tell protocol goroutine to close. - e.sndCloseWaker.Assert() } case e.state == StateListen: @@ -1832,11 +1889,20 @@ func (e *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error { if flags&tcpip.ShutdownRead != 0 { e.notifyProtocolGoroutine(notifyClose) } - default: + e.mu.Unlock() return tcpip.ErrNotConnected } - + e.mu.Unlock() + if finQueued { + if e.workMu.TryLock() { + e.handleClose() + e.workMu.Unlock() + } else { + // Tell protocol goroutine to close. + e.sndCloseWaker.Assert() + } + } return nil } @@ -1928,12 +1994,7 @@ func (e *endpoint) Accept() (tcpip.Endpoint, *waiter.Queue, *tcpip.Error) { return nil, nil, tcpip.ErrWouldBlock } - // Start the protocol goroutine. - wq := &waiter.Queue{} - n.startAcceptedLoop(wq) - e.stack.Stats().TCP.PassiveConnectionOpenings.Increment() - - return n, wq, nil + return n, n.waiterQueue, nil } // Bind binds the endpoint to a specific local port and optionally address. @@ -2058,6 +2119,10 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk e.stack.Stats().TCP.ResetsReceived.Increment() } + e.enqueueSegment(s) +} + +func (e *endpoint) enqueueSegment(s *segment) { // Send packet to worker goroutine. if e.segmentQueue.enqueue(s) { e.newSegmentWaker.Assert() diff --git a/pkg/tcpip/transport/tcp/endpoint_state.go b/pkg/tcpip/transport/tcp/endpoint_state.go index 19f003b6b..7aa4c3f0e 100644 --- a/pkg/tcpip/transport/tcp/endpoint_state.go +++ b/pkg/tcpip/transport/tcp/endpoint_state.go @@ -78,7 +78,7 @@ func (e *endpoint) beforeSave() { } fallthrough case StateError, StateClose: - for e.state == StateError && e.workerRunning { + for (e.state == StateError || e.state == StateClose) && e.workerRunning { e.mu.Unlock() time.Sleep(100 * time.Millisecond) e.mu.Lock() @@ -165,6 +165,12 @@ func (e *endpoint) loadState(state EndpointState) { // afterLoad is invoked by stateify. func (e *endpoint) afterLoad() { + // Freeze segment queue before registering to prevent any segments + // from being delivered while it is being restored. + e.origEndpointState = e.state + // Restore the endpoint to InitialState as it will be moved to + // its origEndpointState during Resume. + e.state = StateInitial stack.StackFromEnv.RegisterRestoredEndpoint(e) } @@ -173,8 +179,8 @@ func (e *endpoint) Resume(s *stack.Stack) { e.stack = s e.segmentQueue.setLimit(MaxUnprocessedSegments) e.workMu.Init() + state := e.origEndpointState - state := e.state switch state { case StateInitial, StateBound, StateListen, StateConnecting, StateEstablished: var ss SendBufferSizeOption @@ -189,7 +195,6 @@ func (e *endpoint) Resume(s *stack.Stack) { } bind := func() { - e.state = StateInitial if len(e.BindAddr) == 0 { e.BindAddr = e.ID.LocalAddress } @@ -219,6 +224,16 @@ func (e *endpoint) Resume(s *stack.Stack) { if err := e.connect(tcpip.FullAddress{NIC: e.boundNICID, Addr: e.connectingAddress, Port: e.ID.RemotePort}, false, e.workerRunning); err != tcpip.ErrConnectStarted { panic("endpoint connecting failed: " + err.String()) } + e.mu.Lock() + e.state = e.origEndpointState + closed := e.closed + e.mu.Unlock() + e.notifyProtocolGoroutine(notifyTickleWorker) + if state == StateFinWait2 && closed { + // If the endpoint has been closed then make sure we notify so + // that the FIN_WAIT2 timer is started after a restore. + e.notifyProtocolGoroutine(notifyClose) + } connectedLoading.Done() case StateListen: tcpip.AsyncLoading.Add(1) @@ -265,8 +280,11 @@ func (e *endpoint) Resume(s *stack.Stack) { tcpip.AsyncLoading.Done() }() } - fallthrough + e.state = StateClose + e.stack.CompleteTransportEndpointCleanup(e) + tcpip.DeleteDanglingEndpoint(e) case StateError: + e.state = StateError e.stack.CompleteTransportEndpointCleanup(e) tcpip.DeleteDanglingEndpoint(e) } diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go index c8e4a0d7e..89b965c23 100644 --- a/pkg/tcpip/transport/tcp/protocol.go +++ b/pkg/tcpip/transport/tcp/protocol.go @@ -23,6 +23,7 @@ package tcp import ( "strings" "sync" + "time" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" @@ -54,6 +55,14 @@ const ( // MaxUnprocessedSegments is the maximum number of unprocessed segments // that can be queued for a given endpoint. MaxUnprocessedSegments = 300 + + // DefaultTCPLingerTimeout is the amount of time that sockets linger in + // FIN_WAIT_2 state before being marked closed. + DefaultTCPLingerTimeout = 60 * time.Second + + // DefaultTCPTimeWaitTimeout is the amount of time that sockets linger + // in TIME_WAIT state before being marked closed. + DefaultTCPTimeWaitTimeout = 60 * time.Second ) // SACKEnabled option can be used to enable SACK support in the TCP @@ -93,6 +102,8 @@ type protocol struct { congestionControl string availableCongestionControl []string moderateReceiveBuffer bool + tcpLingerTimeout time.Duration + tcpTimeWaitTimeout time.Duration } // Number returns the tcp protocol number. @@ -212,6 +223,24 @@ func (p *protocol) SetOption(option interface{}) *tcpip.Error { p.mu.Unlock() return nil + case tcpip.TCPLingerTimeoutOption: + if v < 0 { + v = 0 + } + p.mu.Lock() + p.tcpLingerTimeout = time.Duration(v) + p.mu.Unlock() + return nil + + case tcpip.TCPTimeWaitTimeoutOption: + if v < 0 { + v = 0 + } + p.mu.Lock() + p.tcpTimeWaitTimeout = time.Duration(v) + p.mu.Unlock() + return nil + default: return tcpip.ErrUnknownProtocolOption } @@ -262,6 +291,18 @@ func (p *protocol) Option(option interface{}) *tcpip.Error { p.mu.Unlock() return nil + case *tcpip.TCPLingerTimeoutOption: + p.mu.Lock() + *v = tcpip.TCPLingerTimeoutOption(p.tcpLingerTimeout) + p.mu.Unlock() + return nil + + case *tcpip.TCPTimeWaitTimeoutOption: + p.mu.Lock() + *v = tcpip.TCPTimeWaitTimeoutOption(p.tcpTimeWaitTimeout) + p.mu.Unlock() + return nil + default: return tcpip.ErrUnknownProtocolOption } @@ -274,5 +315,7 @@ func NewProtocol() stack.TransportProtocol { recvBufferSize: ReceiveBufferSizeOption{MinBufferSize, DefaultReceiveBufferSize, MaxBufferSize}, congestionControl: ccReno, availableCongestionControl: []string{ccReno, ccCubic}, + tcpLingerTimeout: DefaultTCPLingerTimeout, + tcpTimeWaitTimeout: DefaultTCPTimeWaitTimeout, } } diff --git a/pkg/tcpip/transport/tcp/rcv.go b/pkg/tcpip/transport/tcp/rcv.go index e90f9a7d9..068b90fb6 100644 --- a/pkg/tcpip/transport/tcp/rcv.go +++ b/pkg/tcpip/transport/tcp/rcv.go @@ -18,6 +18,7 @@ import ( "container/heap" "time" + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/seqnum" ) @@ -209,6 +210,11 @@ func (r *receiver) consumeSegment(s *segment, segSeq seqnum.Value, segLen seqnum switch r.ep.state { case StateFinWait1: r.ep.state = StateFinWait2 + // Notify protocol goroutine that we have received an + // ACK to our FIN so that it can start the FIN_WAIT2 + // timer to abort connection if the other side does + // not close within 2MSL. + r.ep.notifyProtocolGoroutine(notifyClose) case StateClosing: r.ep.state = StateTimeWait case StateLastAck: @@ -253,23 +259,105 @@ func (r *receiver) updateRTT() { r.ep.rcvListMu.Unlock() } -// handleRcvdSegment handles TCP segments directed at the connection managed by -// r as they arrive. It is called by the protocol main loop. -func (r *receiver) handleRcvdSegment(s *segment) { +func (r *receiver) handleRcvdSegmentClosing(s *segment, state EndpointState, closed bool) (drop bool, err *tcpip.Error) { + r.ep.rcvListMu.Lock() + rcvClosed := r.ep.rcvClosed || r.closed + r.ep.rcvListMu.Unlock() + + // If we are in one of the shutdown states then we need to do + // additional checks before we try and process the segment. + switch state { + case StateCloseWait, StateClosing, StateLastAck: + if !s.sequenceNumber.LessThanEq(r.rcvNxt) { + s.decRef() + // Just drop the segment as we have + // already received a FIN and this + // segment is after the sequence number + // for the FIN. + return true, nil + } + fallthrough + case StateFinWait1: + fallthrough + case StateFinWait2: + // If we are closed for reads (either due to an + // incoming FIN or the user calling shutdown(.., + // SHUT_RD) then any data past the rcvNxt should + // trigger a RST. + endDataSeq := s.sequenceNumber.Add(seqnum.Size(s.data.Size())) + if rcvClosed && r.rcvNxt.LessThan(endDataSeq) { + s.decRef() + return true, tcpip.ErrConnectionAborted + } + if state == StateFinWait1 { + break + } + + // If it's a retransmission of an old data segment + // or a pure ACK then allow it. + if s.sequenceNumber.Add(s.logicalLen()).LessThanEq(r.rcvNxt) || + s.logicalLen() == 0 { + break + } + + // In FIN-WAIT2 if the socket is fully + // closed(not owned by application on our end + // then the only acceptable segment is a + // FIN. Since FIN can technically also carry + // data we verify that the segment carrying a + // FIN ends at exactly e.rcvNxt+1. + // + // From RFC793 page 25. + // + // For sequence number purposes, the SYN is + // considered to occur before the first actual + // data octet of the segment in which it occurs, + // while the FIN is considered to occur after + // the last actual data octet in a segment in + // which it occurs. + if closed && (!s.flagIsSet(header.TCPFlagFin) || s.sequenceNumber.Add(s.logicalLen()) != r.rcvNxt+1) { + s.decRef() + return true, tcpip.ErrConnectionAborted + } + } + // We don't care about receive processing anymore if the receive side // is closed. - if r.closed { - return + // + // NOTE: We still want to permit a FIN as it's possible only our + // end has closed and the peer is yet to send a FIN. Hence we + // compare only the payload. + segEnd := s.sequenceNumber.Add(seqnum.Size(s.data.Size())) + if rcvClosed && !segEnd.LessThanEq(r.rcvNxt) { + return true, nil + } + return false, nil +} + +// handleRcvdSegment handles TCP segments directed at the connection managed by +// r as they arrive. It is called by the protocol main loop. +func (r *receiver) handleRcvdSegment(s *segment) (drop bool, err *tcpip.Error) { + r.ep.mu.RLock() + state := r.ep.state + closed := r.ep.closed + r.ep.mu.RUnlock() + + if state != StateEstablished { + drop, err := r.handleRcvdSegmentClosing(s, state, closed) + if drop || err != nil { + return drop, err + } } segLen := seqnum.Size(s.data.Size()) segSeq := s.sequenceNumber // If the sequence number range is outside the acceptable range, just - // send an ACK. This is according to RFC 793, page 37. + // send an ACK and stop further processing of the segment. + // This is according to RFC 793, page 68. if !r.acceptable(segSeq, segLen) { r.ep.snd.sendAck() - return + return true, nil } // Defer segment processing if it can't be consumed now. @@ -288,7 +376,7 @@ func (r *receiver) handleRcvdSegment(s *segment) { // have to retransmit. r.ep.snd.sendAck() } - return + return false, nil } // Since we consumed a segment update the receiver's RTT estimate @@ -315,4 +403,67 @@ func (r *receiver) handleRcvdSegment(s *segment) { r.pendingBufUsed -= s.logicalLen() s.decRef() } + return false, nil +} + +// handleTimeWaitSegment handles inbound segments received when the endpoint +// has entered the TIME_WAIT state. +func (r *receiver) handleTimeWaitSegment(s *segment) (resetTimeWait bool, newSyn bool) { + segSeq := s.sequenceNumber + segLen := seqnum.Size(s.data.Size()) + + // Just silently drop any RST packets in TIME_WAIT. We do not support + // TIME_WAIT assasination as a result we confirm w/ fix 1 as described + // in https://tools.ietf.org/html/rfc1337#section-3. + if s.flagIsSet(header.TCPFlagRst) { + return false, false + } + + // If it's a SYN and the sequence number is higher than any seen before + // for this connection then try and redirect it to a listening endpoint + // if available. + // + // RFC 1122: + // "When a connection is [...] on TIME-WAIT state [...] + // [a TCP] MAY accept a new SYN from the remote TCP to + // reopen the connection directly, if it: + + // (1) assigns its initial sequence number for the new + // connection to be larger than the largest sequence + // number it used on the previous connection incarnation, + // and + + // (2) returns to TIME-WAIT state if the SYN turns out + // to be an old duplicate". + if s.flagIsSet(header.TCPFlagSyn) && r.rcvNxt.LessThan(segSeq) { + + return false, true + } + + // Drop the segment if it does not contain an ACK. + if !s.flagIsSet(header.TCPFlagAck) { + return false, false + } + + // Update Timestamp if required. See RFC7323, section-4.3. + if r.ep.sendTSOk && s.parsedOptions.TS { + r.ep.updateRecentTimestamp(s.parsedOptions.TSVal, r.ep.snd.maxSentAck, segSeq) + } + + if segSeq.Add(1) == r.rcvNxt && s.flagIsSet(header.TCPFlagFin) { + // If it's a FIN-ACK then resetTimeWait and send an ACK, as it + // indicates our final ACK could have been lost. + r.ep.snd.sendAck() + return true, false + } + + // If the sequence number range is outside the acceptable range or + // carries data then just send an ACK. This is according to RFC 793, + // page 37. + // + // NOTE: In TIME_WAIT the only acceptable sequence number is rcvNxt. + if segSeq != r.rcvNxt || segLen != 0 { + r.ep.snd.sendAck() + } + return false, false } diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index f4ea5f091..0c1704d74 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -206,17 +206,18 @@ func TestTCPResetSentForACKWhenNotUsingSynCookies(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() + // Set TCPLingerTimeout to 5 seconds so that sockets are marked closed wq := &waiter.Queue{} ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, wq) if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) + t.Fatalf("NewEndpoint failed: %s", err) } if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { - t.Fatalf("Bind failed: %v", err) + t.Fatalf("Bind failed: %s", err) } if err := ep.Listen(10); err != nil { - t.Fatalf("Listen failed: %v", err) + t.Fatalf("Listen failed: %s", err) } // Send a SYN request. @@ -256,7 +257,7 @@ func TestTCPResetSentForACKWhenNotUsingSynCookies(t *testing.T) { case <-ch: c.EP, _, err = ep.Accept() if err != nil { - t.Fatalf("Accept failed: %v", err) + t.Fatalf("Accept failed: %s", err) } case <-time.After(1 * time.Second): @@ -264,6 +265,13 @@ func TestTCPResetSentForACKWhenNotUsingSynCookies(t *testing.T) { } } + // Lower stackwide TIME_WAIT timeout so that the reservations + // are released instantly on Close. + tcpTW := tcpip.TCPTimeWaitTimeoutOption(1 * time.Millisecond) + if err := c.Stack().SetTransportProtocolOption(tcp.ProtocolNumber, tcpTW); err != nil { + t.Fatalf("e.stack.SetTransportProtocolOption(%d, %s) = %s", tcp.ProtocolNumber, tcpTW, err) + } + c.EP.Close() checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), @@ -285,6 +293,11 @@ func TestTCPResetSentForACKWhenNotUsingSynCookies(t *testing.T) { // Get the ACK to the FIN we just sent. c.GetPacket() + // Since an active close was done we need to wait for a little more than + // tcpLingerTimeout for the port reservations to be released and the + // socket to move to a CLOSED state. + time.Sleep(20 * time.Millisecond) + // Now resend the same ACK, this ACK should generate a RST as there // should be no endpoint in SYN-RCVD state and we are not using // syn-cookies yet. The reason we send the same ACK is we need a valid @@ -376,6 +389,13 @@ func TestConnectResetAfterClose(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() + // Set TCPLinger to 3 seconds so that sockets are marked closed + // after 3 second in FIN_WAIT2 state. + tcpLingerTimeout := 3 * time.Second + if err := c.Stack().SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.TCPLingerTimeoutOption(tcpLingerTimeout)); err != nil { + t.Fatalf("c.stack.SetTransportProtocolOption(tcp, tcpip.TCPLingerTimeoutOption(%d) failed: %s", tcpLingerTimeout, err) + } + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) ep := c.EP c.EP = nil @@ -396,12 +416,24 @@ func TestConnectResetAfterClose(t *testing.T) { DstPort: c.Port, Flags: header.TCPFlagAck, SeqNum: 790, - AckNum: c.IRS.Add(1), + AckNum: c.IRS.Add(2), + RcvWnd: 30000, + }) + + // Wait for the ep to give up waiting for a FIN. + time.Sleep(tcpLingerTimeout + 1*time.Second) + + // Now send an ACK and it should trigger a RST as the endpoint should + // not exist anymore. + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: c.Port, + Flags: header.TCPFlagAck, + SeqNum: 790, + AckNum: c.IRS.Add(2), RcvWnd: 30000, }) - // Wait for the ep to give up waiting for a FIN, and send a RST. - time.Sleep(3 * time.Second) for { b := c.GetPacket() tcpHdr := header.TCP(header.IPv4(b).Payload()) @@ -413,7 +445,7 @@ func TestConnectResetAfterClose(t *testing.T) { checker.IPv4(t, b, checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), + checker.SeqNum(uint32(c.IRS)+2), checker.AckNum(790), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagRst), ), @@ -1110,8 +1142,7 @@ func TestRstOnCloseWithUnreadDataFinConvertRst(t *testing.T) { checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagRst), - // We shouldn't consume a sequence number on RST. - checker.SeqNum(uint32(c.IRS)+1), + checker.SeqNum(uint32(c.IRS)+2), )) // The RST puts the endpoint into an error state. if got, want := tcp.EndpointState(c.EP.State()), tcp.StateError; got != want { @@ -3085,6 +3116,13 @@ func TestReadAfterClosedState(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() + // Set TCPTimeWaitTimeout to 1 seconds so that sockets are marked closed + // after 1 second in TIME_WAIT state. + tcpTimeWaitTimeout := 1 * time.Second + if err := c.Stack().SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.TCPTimeWaitTimeoutOption(tcpTimeWaitTimeout)); err != nil { + t.Fatalf("c.stack.SetTransportProtocolOption(tcp, tcpip.TCPTimeWaitTimeout(%d) failed: %s", tcpTimeWaitTimeout, err) + } + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) we, ch := waiter.NewChannelEntry(nil) @@ -3092,12 +3130,12 @@ func TestReadAfterClosedState(t *testing.T) { defer c.WQ.EventUnregister(&we) if _, _, err := c.EP.Read(nil); err != tcpip.ErrWouldBlock { - t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrWouldBlock) + t.Fatalf("got c.EP.Read(nil) = %v, want = %s", err, tcpip.ErrWouldBlock) } // Shutdown immediately for write, check that we get a FIN. if err := c.EP.Shutdown(tcpip.ShutdownWrite); err != nil { - t.Fatalf("Shutdown failed: %v", err) + t.Fatalf("Shutdown failed: %s", err) } checker.IPv4(t, c.GetPacket(), @@ -3135,10 +3173,9 @@ func TestReadAfterClosedState(t *testing.T) { ), ) - // Give the stack the chance to transition to closed state. Note that since - // both the sender and receiver are now closed, we effectively skip the - // TIME-WAIT state. - time.Sleep(1 * time.Second) + // Give the stack the chance to transition to closed state from + // TIME_WAIT. + time.Sleep(tcpTimeWaitTimeout * 2) if got, want := tcp.EndpointState(c.EP.State()), tcp.StateClose; got != want { t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) @@ -3155,7 +3192,7 @@ func TestReadAfterClosedState(t *testing.T) { peekBuf := make([]byte, 10) n, _, err := c.EP.Peek([][]byte{peekBuf}) if err != nil { - t.Fatalf("Peek failed: %v", err) + t.Fatalf("Peek failed: %s", err) } peekBuf = peekBuf[:n] @@ -3166,7 +3203,7 @@ func TestReadAfterClosedState(t *testing.T) { // Receive data. v, _, err := c.EP.Read(nil) if err != nil { - t.Fatalf("Read failed: %v", err) + t.Fatalf("Read failed: %s", err) } if !bytes.Equal(data, v) { @@ -3176,11 +3213,11 @@ func TestReadAfterClosedState(t *testing.T) { // Now that we drained the queue, check that functions fail with the // right error code. if _, _, err := c.EP.Read(nil); err != tcpip.ErrClosedForReceive { - t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrClosedForReceive) + t.Fatalf("got c.EP.Read(nil) = %v, want = %s", err, tcpip.ErrClosedForReceive) } if _, _, err := c.EP.Peek([][]byte{peekBuf}); err != tcpip.ErrClosedForReceive { - t.Fatalf("got c.EP.Peek(...) = %v, want = %v", err, tcpip.ErrClosedForReceive) + t.Fatalf("got c.EP.Peek(...) = %v, want = %s", err, tcpip.ErrClosedForReceive) } } @@ -4347,7 +4384,8 @@ func TestListenBacklogFullSynCookieInUse(t *testing.T) { // Send a SYN request. irs := seqnum.Value(789) c.SendPacket(nil, &context.Headers{ - SrcPort: context.TestPort, + // pick a different src port for new SYN. + SrcPort: context.TestPort + 1, DstPort: context.StackPort, Flags: header.TCPFlagSyn, SeqNum: irs, @@ -4893,3 +4931,545 @@ func checkDelayOption(t *testing.T, c *context.Context, wantDelayEnabled tcp.Del t.Errorf("ep.GetSockOptInt(tcpip.DelayOption) got: %d, want: %d", gotDelayOption, wantDelayOption) } } + +func TestTCPLingerTimeout(t *testing.T) { + c := context.New(t, 1500 /* mtu */) + defer c.Cleanup() + + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) + + testCases := []struct { + name string + tcpLingerTimeout time.Duration + want time.Duration + }{ + {"NegativeLingerTimeout", -123123, 0}, + {"ZeroLingerTimeout", 0, 0}, + {"InRangeLingerTimeout", 10 * time.Second, 10 * time.Second}, + // Values > stack's TCPLingerTimeout are capped to the stack's + // value. Defaults to tcp.DefaultTCPLingerTimeout(60 seconds) + {"AboveMaxLingerTimeout", 65 * time.Second, 60 * time.Second}, + } + for _, tc := range testCases { + t.Run(tc.name, func(t *testing.T) { + if err := c.EP.SetSockOpt(tcpip.TCPLingerTimeoutOption(tc.tcpLingerTimeout)); err != nil { + t.Fatalf("SetSockOpt(%s) = %s", tc.tcpLingerTimeout, err) + } + var v tcpip.TCPLingerTimeoutOption + if err := c.EP.GetSockOpt(&v); err != nil { + t.Fatalf("GetSockOpt(tcpip.TCPLingerTimeoutOption) = %s", err) + } + if got, want := time.Duration(v), tc.want; got != want { + t.Fatalf("unexpected linger timeout got: %s, want: %s", got, want) + } + }) + } +} + +func TestTCPTimeWaitRSTIgnored(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + wq := &waiter.Queue{} + ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, wq) + if err != nil { + t.Fatalf("NewEndpoint failed: %s", err) + } + if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatalf("Bind failed: %s", err) + } + + if err := ep.Listen(10); err != nil { + t.Fatalf("Listen failed: %s", err) + } + + // Send a SYN request. + iss := seqnum.Value(789) + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn, + SeqNum: iss, + }) + + // Receive the SYN-ACK reply. + b := c.GetPacket() + tcpHdr := header.TCP(header.IPv4(b).Payload()) + c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) + + ackHeaders := &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: iss + 1, + AckNum: c.IRS + 1, + } + + // Send ACK. + c.SendPacket(nil, ackHeaders) + + // Try to accept the connection. + we, ch := waiter.NewChannelEntry(nil) + wq.EventRegister(&we, waiter.EventIn) + defer wq.EventUnregister(&we) + + c.EP, _, err = ep.Accept() + if err == tcpip.ErrWouldBlock { + // Wait for connection to be established. + select { + case <-ch: + c.EP, _, err = ep.Accept() + if err != nil { + t.Fatalf("Accept failed: %s", err) + } + + case <-time.After(1 * time.Second): + t.Fatalf("Timed out waiting for accept") + } + } + + c.EP.Close() + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+1)), + checker.AckNum(uint32(iss)+1), + checker.TCPFlags(header.TCPFlagFin|header.TCPFlagAck))) + + finHeaders := &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck | header.TCPFlagFin, + SeqNum: iss + 1, + AckNum: c.IRS + 2, + } + + c.SendPacket(nil, finHeaders) + + // Get the ACK to the FIN we just sent. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+2)), + checker.AckNum(uint32(iss)+2), + checker.TCPFlags(header.TCPFlagAck))) + + // Now send a RST and this should be ignored and not + // generate an ACK. + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagRst, + SeqNum: iss + 1, + AckNum: c.IRS + 2, + }) + + c.CheckNoPacketTimeout("unexpected packet received in TIME_WAIT state", 1*time.Second) + + // Out of order ACK should generate an immediate ACK in + // TIME_WAIT. + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: iss + 1, + AckNum: c.IRS + 3, + }) + + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+2)), + checker.AckNum(uint32(iss)+2), + checker.TCPFlags(header.TCPFlagAck))) +} + +func TestTCPTimeWaitOutOfOrder(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + wq := &waiter.Queue{} + ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, wq) + if err != nil { + t.Fatalf("NewEndpoint failed: %s", err) + } + if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatalf("Bind failed: %s", err) + } + + if err := ep.Listen(10); err != nil { + t.Fatalf("Listen failed: %s", err) + } + + // Send a SYN request. + iss := seqnum.Value(789) + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn, + SeqNum: iss, + }) + + // Receive the SYN-ACK reply. + b := c.GetPacket() + tcpHdr := header.TCP(header.IPv4(b).Payload()) + c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) + + ackHeaders := &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: iss + 1, + AckNum: c.IRS + 1, + } + + // Send ACK. + c.SendPacket(nil, ackHeaders) + + // Try to accept the connection. + we, ch := waiter.NewChannelEntry(nil) + wq.EventRegister(&we, waiter.EventIn) + defer wq.EventUnregister(&we) + + c.EP, _, err = ep.Accept() + if err == tcpip.ErrWouldBlock { + // Wait for connection to be established. + select { + case <-ch: + c.EP, _, err = ep.Accept() + if err != nil { + t.Fatalf("Accept failed: %s", err) + } + + case <-time.After(1 * time.Second): + t.Fatalf("Timed out waiting for accept") + } + } + + c.EP.Close() + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+1)), + checker.AckNum(uint32(iss)+1), + checker.TCPFlags(header.TCPFlagFin|header.TCPFlagAck))) + + finHeaders := &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck | header.TCPFlagFin, + SeqNum: iss + 1, + AckNum: c.IRS + 2, + } + + c.SendPacket(nil, finHeaders) + + // Get the ACK to the FIN we just sent. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+2)), + checker.AckNum(uint32(iss)+2), + checker.TCPFlags(header.TCPFlagAck))) + + // Out of order ACK should generate an immediate ACK in + // TIME_WAIT. + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: iss + 1, + AckNum: c.IRS + 3, + }) + + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+2)), + checker.AckNum(uint32(iss)+2), + checker.TCPFlags(header.TCPFlagAck))) +} + +func TestTCPTimeWaitNewSyn(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + wq := &waiter.Queue{} + ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, wq) + if err != nil { + t.Fatalf("NewEndpoint failed: %s", err) + } + if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatalf("Bind failed: %s", err) + } + + if err := ep.Listen(10); err != nil { + t.Fatalf("Listen failed: %s", err) + } + + // Send a SYN request. + iss := seqnum.Value(789) + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn, + SeqNum: iss, + }) + + // Receive the SYN-ACK reply. + b := c.GetPacket() + tcpHdr := header.TCP(header.IPv4(b).Payload()) + c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) + + ackHeaders := &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: iss + 1, + AckNum: c.IRS + 1, + } + + // Send ACK. + c.SendPacket(nil, ackHeaders) + + // Try to accept the connection. + we, ch := waiter.NewChannelEntry(nil) + wq.EventRegister(&we, waiter.EventIn) + defer wq.EventUnregister(&we) + + c.EP, _, err = ep.Accept() + if err == tcpip.ErrWouldBlock { + // Wait for connection to be established. + select { + case <-ch: + c.EP, _, err = ep.Accept() + if err != nil { + t.Fatalf("Accept failed: %s", err) + } + + case <-time.After(1 * time.Second): + t.Fatalf("Timed out waiting for accept") + } + } + + c.EP.Close() + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+1)), + checker.AckNum(uint32(iss)+1), + checker.TCPFlags(header.TCPFlagFin|header.TCPFlagAck))) + + finHeaders := &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck | header.TCPFlagFin, + SeqNum: iss + 1, + AckNum: c.IRS + 2, + } + + c.SendPacket(nil, finHeaders) + + // Get the ACK to the FIN we just sent. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+2)), + checker.AckNum(uint32(iss)+2), + checker.TCPFlags(header.TCPFlagAck))) + + // Send a SYN request w/ sequence number lower than + // the highest sequence number sent. We just reuse + // the same number. + iss = seqnum.Value(789) + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn, + SeqNum: iss, + }) + + c.CheckNoPacketTimeout("unexpected packet received in response to SYN", 1*time.Second) + + // Send a SYN request w/ sequence number higher than + // the highest sequence number sent. + iss = seqnum.Value(792) + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn, + SeqNum: iss, + }) + + // Receive the SYN-ACK reply. + b = c.GetPacket() + tcpHdr = header.TCP(header.IPv4(b).Payload()) + c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) + + ackHeaders = &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: iss + 1, + AckNum: c.IRS + 1, + } + + // Send ACK. + c.SendPacket(nil, ackHeaders) + + // Try to accept the connection. + c.EP, _, err = ep.Accept() + if err == tcpip.ErrWouldBlock { + // Wait for connection to be established. + select { + case <-ch: + c.EP, _, err = ep.Accept() + if err != nil { + t.Fatalf("Accept failed: %s", err) + } + + case <-time.After(1 * time.Second): + t.Fatalf("Timed out waiting for accept") + } + } +} + +func TestTCPTimeWaitDuplicateFINExtendsTimeWait(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + // Set TCPTimeWaitTimeout to 5 seconds so that sockets are marked closed + // after 5 seconds in TIME_WAIT state. + tcpTimeWaitTimeout := 5 * time.Second + if err := c.Stack().SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.TCPTimeWaitTimeoutOption(tcpTimeWaitTimeout)); err != nil { + t.Fatalf("c.stack.SetTransportProtocolOption(tcp, tcpip.TCPLingerTimeoutOption(%d) failed: %s", tcpTimeWaitTimeout, err) + } + + wq := &waiter.Queue{} + ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, wq) + if err != nil { + t.Fatalf("NewEndpoint failed: %s", err) + } + if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatalf("Bind failed: %s", err) + } + + if err := ep.Listen(10); err != nil { + t.Fatalf("Listen failed: %s", err) + } + + // Send a SYN request. + iss := seqnum.Value(789) + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn, + SeqNum: iss, + }) + + // Receive the SYN-ACK reply. + b := c.GetPacket() + tcpHdr := header.TCP(header.IPv4(b).Payload()) + c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) + + ackHeaders := &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: iss + 1, + AckNum: c.IRS + 1, + } + + // Send ACK. + c.SendPacket(nil, ackHeaders) + + // Try to accept the connection. + we, ch := waiter.NewChannelEntry(nil) + wq.EventRegister(&we, waiter.EventIn) + defer wq.EventUnregister(&we) + + c.EP, _, err = ep.Accept() + if err == tcpip.ErrWouldBlock { + // Wait for connection to be established. + select { + case <-ch: + c.EP, _, err = ep.Accept() + if err != nil { + t.Fatalf("Accept failed: %s", err) + } + + case <-time.After(1 * time.Second): + t.Fatalf("Timed out waiting for accept") + } + } + + c.EP.Close() + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+1)), + checker.AckNum(uint32(iss)+1), + checker.TCPFlags(header.TCPFlagFin|header.TCPFlagAck))) + + finHeaders := &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck | header.TCPFlagFin, + SeqNum: iss + 1, + AckNum: c.IRS + 2, + } + + c.SendPacket(nil, finHeaders) + + // Get the ACK to the FIN we just sent. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+2)), + checker.AckNum(uint32(iss)+2), + checker.TCPFlags(header.TCPFlagAck))) + + time.Sleep(2 * time.Second) + + // Now send a duplicate FIN. This should cause the TIME_WAIT to extend + // by another 5 seconds and also send us a duplicate ACK as it should + // indicate that the final ACK was potentially lost. + c.SendPacket(nil, finHeaders) + + // Get the ACK to the FIN we just sent. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+2)), + checker.AckNum(uint32(iss)+2), + checker.TCPFlags(header.TCPFlagAck))) + + // Sleep for 4 seconds so at this point we are 1 second past the + // original tcpLingerTimeout of 5 seconds. + time.Sleep(4 * time.Second) + + // Send an ACK and it should not generate any packet as the socket + // should still be in TIME_WAIT for another another 5 seconds due + // to the duplicate FIN we sent earlier. + *ackHeaders = *finHeaders + ackHeaders.SeqNum = ackHeaders.SeqNum + 1 + ackHeaders.Flags = header.TCPFlagAck + c.SendPacket(nil, ackHeaders) + + c.CheckNoPacketTimeout("unexpected packet received from endpoint in TIME_WAIT", 1*time.Second) + // Now sleep for another 2 seconds so that we are past the + // extended TIME_WAIT of 7 seconds (2 + 5). + time.Sleep(2 * time.Second) + + // Resend the same ACK. + c.SendPacket(nil, ackHeaders) + + // Receive the RST that should be generated as there is no valid + // endpoint. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(ackHeaders.AckNum)), + checker.AckNum(uint32(ackHeaders.SeqNum)), + checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck))) +} diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 3e5b6b3c3..722d14b53 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -9,7 +9,7 @@ syscall_test(test = "//test/syscalls/linux:accept_bind_stream_test") syscall_test( size = "large", - shard_count = 10, + shard_count = 50, test = "//test/syscalls/linux:accept_bind_test", ) @@ -434,7 +434,7 @@ syscall_test( syscall_test( size = "large", - shard_count = 10, + shard_count = 50, test = "//test/syscalls/linux:socket_abstract_test", ) @@ -445,7 +445,7 @@ syscall_test( syscall_test( size = "large", - shard_count = 10, + shard_count = 50, test = "//test/syscalls/linux:socket_domain_test", ) @@ -458,19 +458,19 @@ syscall_test( syscall_test( size = "large", add_overlay = True, - shard_count = 10, + shard_count = 50, test = "//test/syscalls/linux:socket_filesystem_test", ) syscall_test( size = "large", - shard_count = 10, + shard_count = 50, test = "//test/syscalls/linux:socket_inet_loopback_test", ) syscall_test( size = "large", - shard_count = 10, + shard_count = 50, test = "//test/syscalls/linux:socket_ip_tcp_generic_loopback_test", ) @@ -481,13 +481,13 @@ syscall_test( syscall_test( size = "large", - shard_count = 10, + shard_count = 50, test = "//test/syscalls/linux:socket_ip_tcp_loopback_test", ) syscall_test( size = "medium", - shard_count = 10, + shard_count = 50, test = "//test/syscalls/linux:socket_ip_tcp_udp_generic_loopback_test", ) @@ -498,7 +498,7 @@ syscall_test( syscall_test( size = "large", - shard_count = 10, + shard_count = 50, test = "//test/syscalls/linux:socket_ip_udp_loopback_test", ) @@ -560,7 +560,7 @@ syscall_test( syscall_test( size = "large", add_overlay = True, - shard_count = 10, + shard_count = 50, test = "//test/syscalls/linux:socket_unix_pair_test", ) @@ -599,7 +599,7 @@ syscall_test( syscall_test( size = "large", - shard_count = 10, + shard_count = 50, test = "//test/syscalls/linux:socket_unix_unbound_stream_test", ) diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 93bff8299..f8b8cb724 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -2141,6 +2141,7 @@ cc_library( deps = [ ":socket_test_util", "//test/util:test_util", + "//test/util:thread_util", "@com_google_googletest//:gtest", ], alwayslink = 1, diff --git a/test/syscalls/linux/socket_inet_loopback.cc b/test/syscalls/linux/socket_inet_loopback.cc index ab375aaaf..2eeee352e 100644 --- a/test/syscalls/linux/socket_inet_loopback.cc +++ b/test/syscalls/linux/socket_inet_loopback.cc @@ -13,6 +13,7 @@ // limitations under the License. #include +#include #include #include #include @@ -31,6 +32,7 @@ #include "gtest/gtest.h" #include "absl/memory/memory.h" #include "absl/strings/str_cat.h" +#include "absl/time/clock.h" #include "absl/time/time.h" #include "test/syscalls/linux/socket_test_util.h" #include "test/util/file_descriptor.h" @@ -267,6 +269,340 @@ TEST_P(SocketInetLoopbackTest, TCPbacklog) { } } +// TCPFinWait2Test creates a pair of connected sockets then closes one end to +// trigger FIN_WAIT2 state for the closed endpoint. Then it binds the same local +// IP/port on a new socket and tries to connect. The connect should fail w/ +// an EADDRINUSE. Then we wait till the FIN_WAIT2 timeout is over and try the +// connect again with a new socket and this time it should succeed. +// +// TCP timers are not S/R today, this can cause this test to be flaky when run +// under random S/R due to timer being reset on a restore. +TEST_P(SocketInetLoopbackTest, TCPFinWait2Test_NoRandomSave) { + auto const& param = GetParam(); + TestAddress const& listener = param.listener; + TestAddress const& connector = param.connector; + + // Create the listening socket. + const FileDescriptor listen_fd = ASSERT_NO_ERRNO_AND_VALUE( + Socket(listener.family(), SOCK_STREAM, IPPROTO_TCP)); + sockaddr_storage listen_addr = listener.addr; + ASSERT_THAT(bind(listen_fd.get(), reinterpret_cast(&listen_addr), + listener.addr_len), + SyscallSucceeds()); + ASSERT_THAT(listen(listen_fd.get(), SOMAXCONN), SyscallSucceeds()); + + // Get the port bound by the listening socket. + socklen_t addrlen = listener.addr_len; + ASSERT_THAT(getsockname(listen_fd.get(), + reinterpret_cast(&listen_addr), &addrlen), + SyscallSucceeds()); + + uint16_t const port = + ASSERT_NO_ERRNO_AND_VALUE(AddrPort(listener.family(), listen_addr)); + + // Connect to the listening socket. + FileDescriptor conn_fd = ASSERT_NO_ERRNO_AND_VALUE( + Socket(connector.family(), SOCK_STREAM, IPPROTO_TCP)); + + // Lower FIN_WAIT2 state to 5 seconds for test. + constexpr int kTCPLingerTimeout = 5; + EXPECT_THAT(setsockopt(conn_fd.get(), IPPROTO_TCP, TCP_LINGER2, + &kTCPLingerTimeout, sizeof(kTCPLingerTimeout)), + SyscallSucceedsWithValue(0)); + + sockaddr_storage conn_addr = connector.addr; + ASSERT_NO_ERRNO(SetAddrPort(connector.family(), &conn_addr, port)); + ASSERT_THAT(RetryEINTR(connect)(conn_fd.get(), + reinterpret_cast(&conn_addr), + connector.addr_len), + SyscallSucceeds()); + + // Accept the connection. + auto accepted = + ASSERT_NO_ERRNO_AND_VALUE(Accept(listen_fd.get(), nullptr, nullptr)); + + // Get the address/port bound by the connecting socket. + sockaddr_storage conn_bound_addr; + socklen_t conn_addrlen = connector.addr_len; + ASSERT_THAT( + getsockname(conn_fd.get(), reinterpret_cast(&conn_bound_addr), + &conn_addrlen), + SyscallSucceeds()); + + // close the connecting FD to trigger FIN_WAIT2 on the connected fd. + conn_fd.reset(); + + // Now bind and connect a new socket. + const FileDescriptor conn_fd2 = ASSERT_NO_ERRNO_AND_VALUE( + Socket(connector.family(), SOCK_STREAM, IPPROTO_TCP)); + + // Disable cooperative saves after this point. As a save between the first + // bind/connect and the second one can cause the linger timeout timer to + // be restarted causing the final bind/connect to fail. + DisableSave ds; + + // TODO(gvisor.dev/issue/1030): Portmanager does not track all 5 tuple + // reservations which causes the bind() to succeed on gVisor but connect + // correctly fails. + if (IsRunningOnGvisor()) { + ASSERT_THAT( + bind(conn_fd2.get(), reinterpret_cast(&conn_bound_addr), + conn_addrlen), + SyscallSucceeds()); + ASSERT_THAT(RetryEINTR(connect)(conn_fd2.get(), + reinterpret_cast(&conn_addr), + conn_addrlen), + SyscallFailsWithErrno(EADDRINUSE)); + } else { + ASSERT_THAT( + bind(conn_fd2.get(), reinterpret_cast(&conn_bound_addr), + conn_addrlen), + SyscallFailsWithErrno(EADDRINUSE)); + } + + // Sleep for a little over the linger timeout to reduce flakiness in + // save/restore tests. + absl::SleepFor(absl::Seconds(kTCPLingerTimeout + 1)); + + ds.reset(); + + if (!IsRunningOnGvisor()) { + ASSERT_THAT( + bind(conn_fd2.get(), reinterpret_cast(&conn_bound_addr), + conn_addrlen), + SyscallSucceeds()); + } + ASSERT_THAT(RetryEINTR(connect)(conn_fd2.get(), + reinterpret_cast(&conn_addr), + conn_addrlen), + SyscallSucceeds()); +} + +// TCPLinger2TimeoutAfterClose creates a pair of connected sockets +// then closes one end to trigger FIN_WAIT2 state for the closed endpont. +// It then sleeps for the TCP_LINGER2 timeout and verifies that bind/ +// connecting the same address succeeds. +// +// TCP timers are not S/R today, this can cause this test to be flaky when run +// under random S/R due to timer being reset on a restore. +TEST_P(SocketInetLoopbackTest, TCPLinger2TimeoutAfterClose_NoRandomSave) { + auto const& param = GetParam(); + TestAddress const& listener = param.listener; + TestAddress const& connector = param.connector; + + // Create the listening socket. + const FileDescriptor listen_fd = ASSERT_NO_ERRNO_AND_VALUE( + Socket(listener.family(), SOCK_STREAM, IPPROTO_TCP)); + sockaddr_storage listen_addr = listener.addr; + ASSERT_THAT(bind(listen_fd.get(), reinterpret_cast(&listen_addr), + listener.addr_len), + SyscallSucceeds()); + ASSERT_THAT(listen(listen_fd.get(), SOMAXCONN), SyscallSucceeds()); + + // Get the port bound by the listening socket. + socklen_t addrlen = listener.addr_len; + ASSERT_THAT(getsockname(listen_fd.get(), + reinterpret_cast(&listen_addr), &addrlen), + SyscallSucceeds()); + + uint16_t const port = + ASSERT_NO_ERRNO_AND_VALUE(AddrPort(listener.family(), listen_addr)); + + // Connect to the listening socket. + FileDescriptor conn_fd = ASSERT_NO_ERRNO_AND_VALUE( + Socket(connector.family(), SOCK_STREAM, IPPROTO_TCP)); + + sockaddr_storage conn_addr = connector.addr; + ASSERT_NO_ERRNO(SetAddrPort(connector.family(), &conn_addr, port)); + ASSERT_THAT(RetryEINTR(connect)(conn_fd.get(), + reinterpret_cast(&conn_addr), + connector.addr_len), + SyscallSucceeds()); + + // Accept the connection. + auto accepted = + ASSERT_NO_ERRNO_AND_VALUE(Accept(listen_fd.get(), nullptr, nullptr)); + + // Get the address/port bound by the connecting socket. + sockaddr_storage conn_bound_addr; + socklen_t conn_addrlen = connector.addr_len; + ASSERT_THAT( + getsockname(conn_fd.get(), reinterpret_cast(&conn_bound_addr), + &conn_addrlen), + SyscallSucceeds()); + + constexpr int kTCPLingerTimeout = 5; + EXPECT_THAT(setsockopt(conn_fd.get(), IPPROTO_TCP, TCP_LINGER2, + &kTCPLingerTimeout, sizeof(kTCPLingerTimeout)), + SyscallSucceedsWithValue(0)); + + // close the connecting FD to trigger FIN_WAIT2 on the connected fd. + conn_fd.reset(); + + absl::SleepFor(absl::Seconds(kTCPLingerTimeout + 1)); + + // Now bind and connect a new socket and verify that we can immediately + // rebind the address bound by the conn_fd as it never entered TIME_WAIT. + const FileDescriptor conn_fd2 = ASSERT_NO_ERRNO_AND_VALUE( + Socket(connector.family(), SOCK_STREAM, IPPROTO_TCP)); + + ASSERT_THAT(bind(conn_fd2.get(), + reinterpret_cast(&conn_bound_addr), conn_addrlen), + SyscallSucceeds()); + ASSERT_THAT(RetryEINTR(connect)(conn_fd2.get(), + reinterpret_cast(&conn_addr), + conn_addrlen), + SyscallSucceeds()); +} + +// TCPResetAfterClose creates a pair of connected sockets then closes +// one end to trigger FIN_WAIT2 state for the closed endpoint verifies +// that we generate RSTs for any new data after the socket is fully +// closed. +TEST_P(SocketInetLoopbackTest, TCPResetAfterClose) { + auto const& param = GetParam(); + TestAddress const& listener = param.listener; + TestAddress const& connector = param.connector; + + // Create the listening socket. + const FileDescriptor listen_fd = ASSERT_NO_ERRNO_AND_VALUE( + Socket(listener.family(), SOCK_STREAM, IPPROTO_TCP)); + sockaddr_storage listen_addr = listener.addr; + ASSERT_THAT(bind(listen_fd.get(), reinterpret_cast(&listen_addr), + listener.addr_len), + SyscallSucceeds()); + ASSERT_THAT(listen(listen_fd.get(), SOMAXCONN), SyscallSucceeds()); + + // Get the port bound by the listening socket. + socklen_t addrlen = listener.addr_len; + ASSERT_THAT(getsockname(listen_fd.get(), + reinterpret_cast(&listen_addr), &addrlen), + SyscallSucceeds()); + + uint16_t const port = + ASSERT_NO_ERRNO_AND_VALUE(AddrPort(listener.family(), listen_addr)); + + // Connect to the listening socket. + FileDescriptor conn_fd = ASSERT_NO_ERRNO_AND_VALUE( + Socket(connector.family(), SOCK_STREAM, IPPROTO_TCP)); + + sockaddr_storage conn_addr = connector.addr; + ASSERT_NO_ERRNO(SetAddrPort(connector.family(), &conn_addr, port)); + ASSERT_THAT(RetryEINTR(connect)(conn_fd.get(), + reinterpret_cast(&conn_addr), + connector.addr_len), + SyscallSucceeds()); + + // Accept the connection. + auto accepted = + ASSERT_NO_ERRNO_AND_VALUE(Accept(listen_fd.get(), nullptr, nullptr)); + + // close the connecting FD to trigger FIN_WAIT2 on the connected fd. + conn_fd.reset(); + + int data = 1234; + + // Now send data which should trigger a RST as the other end should + // have timed out and closed the socket. + EXPECT_THAT(RetryEINTR(send)(accepted.get(), &data, sizeof(data), 0), + SyscallSucceeds()); + // Sleep for a shortwhile to get a RST back. + absl::SleepFor(absl::Seconds(1)); + + // Try writing again and we should get an EPIPE back. + EXPECT_THAT(RetryEINTR(send)(accepted.get(), &data, sizeof(data), 0), + SyscallFailsWithErrno(EPIPE)); + + // Trying to read should return zero as the other end did send + // us a FIN. We do it twice to verify that the RST does not cause an + // ECONNRESET on the read after EOF has been read by applicaiton. + EXPECT_THAT(RetryEINTR(recv)(accepted.get(), &data, sizeof(data), 0), + SyscallSucceedsWithValue(0)); + EXPECT_THAT(RetryEINTR(recv)(accepted.get(), &data, sizeof(data), 0), + SyscallSucceedsWithValue(0)); +} + +// This test is disabled under random save as the the restore run +// results in the stack.Seed() being different which can cause +// sequence number of final connect to be one that is considered +// old and can cause the test to be flaky. +TEST_P(SocketInetLoopbackTest, TCPTimeWaitTest_NoRandomSave) { + auto const& param = GetParam(); + TestAddress const& listener = param.listener; + TestAddress const& connector = param.connector; + + // Create the listening socket. + const FileDescriptor listen_fd = ASSERT_NO_ERRNO_AND_VALUE( + Socket(listener.family(), SOCK_STREAM, IPPROTO_TCP)); + sockaddr_storage listen_addr = listener.addr; + ASSERT_THAT(bind(listen_fd.get(), reinterpret_cast(&listen_addr), + listener.addr_len), + SyscallSucceeds()); + ASSERT_THAT(listen(listen_fd.get(), SOMAXCONN), SyscallSucceeds()); + + // Get the port bound by the listening socket. + socklen_t addrlen = listener.addr_len; + ASSERT_THAT(getsockname(listen_fd.get(), + reinterpret_cast(&listen_addr), &addrlen), + SyscallSucceeds()); + + uint16_t const port = + ASSERT_NO_ERRNO_AND_VALUE(AddrPort(listener.family(), listen_addr)); + + // Connect to the listening socket. + FileDescriptor conn_fd = ASSERT_NO_ERRNO_AND_VALUE( + Socket(connector.family(), SOCK_STREAM, IPPROTO_TCP)); + + // We disable saves after this point as a S/R causes the netstack seed + // to be regenerated which changes what ports/ISN is picked for a given + // tuple (src ip,src port, dst ip, dst port). This can cause the final + // SYN to use a sequence number that looks like one from the current + // connection in TIME_WAIT and will not be accepted causing the test + // to timeout. + // + // TODO(gvisor.dev/issue/940): S/R portSeed/portHint + DisableSave ds; + sockaddr_storage conn_addr = connector.addr; + ASSERT_NO_ERRNO(SetAddrPort(connector.family(), &conn_addr, port)); + ASSERT_THAT(RetryEINTR(connect)(conn_fd.get(), + reinterpret_cast(&conn_addr), + connector.addr_len), + SyscallSucceeds()); + + // Accept the connection. + auto accepted = + ASSERT_NO_ERRNO_AND_VALUE(Accept(listen_fd.get(), nullptr, nullptr)); + + // Get the address/port bound by the connecting socket. + sockaddr_storage conn_bound_addr; + socklen_t conn_addrlen = connector.addr_len; + ASSERT_THAT( + getsockname(conn_fd.get(), reinterpret_cast(&conn_bound_addr), + &conn_addrlen), + SyscallSucceeds()); + + // close the accept FD to trigger TIME_WAIT on the accepted socket which + // should cause the conn_fd to follow CLOSE_WAIT->LAST_ACK->CLOSED instead of + // TIME_WAIT. + accepted.reset(); + absl::SleepFor(absl::Seconds(1)); + conn_fd.reset(); + absl::SleepFor(absl::Seconds(1)); + + // Now bind and connect a new socket and verify that we can immediately + // rebind the address bound by the conn_fd as it never entered TIME_WAIT. + const FileDescriptor conn_fd2 = ASSERT_NO_ERRNO_AND_VALUE( + Socket(connector.family(), SOCK_STREAM, IPPROTO_TCP)); + + ASSERT_THAT(bind(conn_fd2.get(), + reinterpret_cast(&conn_bound_addr), conn_addrlen), + SyscallSucceeds()); + ASSERT_THAT(RetryEINTR(connect)(conn_fd2.get(), + reinterpret_cast(&conn_addr), + conn_addrlen), + SyscallSucceeds()); +} + INSTANTIATE_TEST_SUITE_P( All, SocketInetLoopbackTest, ::testing::Values( diff --git a/test/syscalls/linux/socket_ip_tcp_generic.cc b/test/syscalls/linux/socket_ip_tcp_generic.cc index 592448289..a37b49447 100644 --- a/test/syscalls/linux/socket_ip_tcp_generic.cc +++ b/test/syscalls/linux/socket_ip_tcp_generic.cc @@ -26,6 +26,7 @@ #include "gtest/gtest.h" #include "test/syscalls/linux/socket_test_util.h" #include "test/util/test_util.h" +#include "test/util/thread_util.h" namespace gvisor { namespace testing { @@ -243,6 +244,31 @@ TEST_P(TCPSocketPairTest, ShutdownRdAllowsReadOfReceivedDataBeforeEOF) { SyscallSucceedsWithValue(0)); } +// This test verifies that a shutdown(wr) by the server after sending +// data allows the client to still read() the queued data and a client +// close after sending response allows server to read the incoming +// response. +TEST_P(TCPSocketPairTest, ShutdownWrServerClientClose) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + char buf[10] = {}; + ScopedThread t([&]() { + ASSERT_THAT(RetryEINTR(read)(sockets->first_fd(), buf, sizeof(buf)), + SyscallSucceedsWithValue(sizeof(buf))); + ASSERT_THAT(RetryEINTR(write)(sockets->first_fd(), buf, sizeof(buf)), + SyscallSucceedsWithValue(sizeof(buf))); + ASSERT_THAT(close(sockets->release_first_fd()), + SyscallSucceedsWithValue(0)); + }); + ASSERT_THAT(RetryEINTR(write)(sockets->second_fd(), buf, sizeof(buf)), + SyscallSucceedsWithValue(sizeof(buf))); + ASSERT_THAT(RetryEINTR(shutdown)(sockets->second_fd(), SHUT_WR), + SyscallSucceedsWithValue(0)); + t.Join(); + + ASSERT_THAT(RetryEINTR(read)(sockets->second_fd(), buf, sizeof(buf)), + SyscallSucceedsWithValue(sizeof(buf))); +} + TEST_P(TCPSocketPairTest, ClosedReadNonBlockingSocket) { auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); @@ -696,5 +722,72 @@ TEST_P(TCPSocketPairTest, SetCongestionControlFailsForUnsupported) { EXPECT_EQ(0, memcmp(got_cc, old_cc, sizeof(old_cc))); } +// Linux and Netstack both default to a 60s TCP_LINGER2 timeout. +constexpr int kDefaultTCPLingerTimeout = 60; + +TEST_P(TCPSocketPairTest, TCPLingerTimeoutDefault) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + int get = -1; + socklen_t get_len = sizeof(get); + EXPECT_THAT( + getsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_LINGER2, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kDefaultTCPLingerTimeout); +} + +TEST_P(TCPSocketPairTest, SetTCPLingerTimeoutZeroOrLess) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + constexpr int kZero = 0; + EXPECT_THAT(setsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_LINGER2, &kZero, + sizeof(kZero)), + SyscallSucceedsWithValue(0)); + + constexpr int kNegative = -1234; + EXPECT_THAT(setsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_LINGER2, + &kNegative, sizeof(kNegative)), + SyscallSucceedsWithValue(0)); +} + +TEST_P(TCPSocketPairTest, SetTCPLingerTimeoutAboveDefault) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + // Values above the net.ipv4.tcp_fin_timeout are capped to tcp_fin_timeout + // on linux (defaults to 60 seconds on linux). + constexpr int kAboveDefault = kDefaultTCPLingerTimeout + 1; + EXPECT_THAT(setsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_LINGER2, + &kAboveDefault, sizeof(kAboveDefault)), + SyscallSucceedsWithValue(0)); + + int get = -1; + socklen_t get_len = sizeof(get); + EXPECT_THAT( + getsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_LINGER2, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kDefaultTCPLingerTimeout); +} + +TEST_P(TCPSocketPairTest, SetTCPLingerTimeout) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + // Values above the net.ipv4.tcp_fin_timeout are capped to tcp_fin_timeout + // on linux (defaults to 60 seconds on linux). + constexpr int kTCPLingerTimeout = kDefaultTCPLingerTimeout - 1; + EXPECT_THAT(setsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_LINGER2, + &kTCPLingerTimeout, sizeof(kTCPLingerTimeout)), + SyscallSucceedsWithValue(0)); + + int get = -1; + socklen_t get_len = sizeof(get); + EXPECT_THAT( + getsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_LINGER2, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kTCPLingerTimeout); +} + } // namespace testing } // namespace gvisor -- cgit v1.2.3 From af58a4e3bb0ba81c103429317eb0c2735450136c Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Fri, 8 Nov 2019 12:18:40 -0800 Subject: Automated rollback of changelist 278417533 PiperOrigin-RevId: 279365629 --- pkg/sentry/fs/inode_operations.go | 2 -- pkg/sentry/syscalls/linux/sys_file.go | 8 ++++---- 2 files changed, 4 insertions(+), 6 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/inode_operations.go b/pkg/sentry/fs/inode_operations.go index d6c35c2dc..5cde9d215 100644 --- a/pkg/sentry/fs/inode_operations.go +++ b/pkg/sentry/fs/inode_operations.go @@ -221,8 +221,6 @@ type InodeOperations interface { // sys_ftruncate. // // Implementations need not check that length >= 0. - // - // Truncate must only be called on regular files. Truncate(ctx context.Context, inode *Inode, size int64) error // Allocate allows the caller to reserve disk space for the inode. diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go index c9f57fe27..b9a8e3e21 100644 --- a/pkg/sentry/syscalls/linux/sys_file.go +++ b/pkg/sentry/syscalls/linux/sys_file.go @@ -169,7 +169,7 @@ func openAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint) (fd uint if dirPath { return syserror.ENOTDIR } - if flags&linux.O_TRUNC != 0 && fs.IsRegular(d.Inode.StableAttr) { + if flags&linux.O_TRUNC != 0 { if err := d.Inode.Truncate(t, d, 0); err != nil { return err } @@ -397,7 +397,7 @@ func createAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint, mode l } // Should we truncate the file? - if flags&linux.O_TRUNC != 0 && fs.IsRegular(found.Inode.StableAttr) { + if flags&linux.O_TRUNC != 0 { if err := found.Inode.Truncate(t, found, 0); err != nil { return err } @@ -1483,7 +1483,7 @@ func Truncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if fs.IsDir(d.Inode.StableAttr) { return syserror.EISDIR } - if !fs.IsRegular(d.Inode.StableAttr) { + if !fs.IsFile(d.Inode.StableAttr) { return syserror.EINVAL } @@ -1523,7 +1523,7 @@ func Ftruncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys // Note that this is different from truncate(2) above, where a // directory returns EISDIR. - if !fs.IsRegular(file.Dirent.Inode.StableAttr) { + if !fs.IsFile(file.Dirent.Inode.StableAttr) { return 0, nil, syserror.EINVAL } -- cgit v1.2.3 From 05871a1cdc73e98df58f56841be23a4eac27225c Mon Sep 17 00:00:00 2001 From: Haibo Xu Date: Mon, 11 Nov 2019 08:20:18 +0000 Subject: Enable runsc/boot support on arm64. This patch also include a minor change to replace syscall.Dup2 with syscall.Dup3 which was missed in a previous commit(ref a25a976). Signed-off-by: Haibo Xu Change-Id: I00beb9cc492e44c762ebaa3750201c63c1f7c2f3 --- pkg/seccomp/seccomp_test_victim.go | 2 +- pkg/sentry/fs/gofer/inode.go | 2 +- runsc/boot/BUILD | 2 ++ runsc/boot/filter/BUILD | 2 ++ runsc/boot/filter/config.go | 6 +----- runsc/boot/filter/config_amd64.go | 31 +++++++++++++++++++++++++++++++ runsc/boot/filter/config_arm64.go | 21 +++++++++++++++++++++ runsc/boot/loader.go | 4 ---- runsc/boot/loader_amd64.go | 28 ++++++++++++++++++++++++++++ runsc/boot/loader_arm64.go | 28 ++++++++++++++++++++++++++++ 10 files changed, 115 insertions(+), 11 deletions(-) create mode 100644 runsc/boot/filter/config_amd64.go create mode 100644 runsc/boot/filter/config_arm64.go create mode 100644 runsc/boot/loader_amd64.go create mode 100644 runsc/boot/loader_arm64.go (limited to 'pkg/sentry') diff --git a/pkg/seccomp/seccomp_test_victim.go b/pkg/seccomp/seccomp_test_victim.go index 48413f1fb..da6b9eaaf 100644 --- a/pkg/seccomp/seccomp_test_victim.go +++ b/pkg/seccomp/seccomp_test_victim.go @@ -38,7 +38,7 @@ func main() { syscall.SYS_CLONE: {}, syscall.SYS_CLOSE: {}, syscall.SYS_DUP: {}, - syscall.SYS_DUP2: {}, + syscall.SYS_DUP3: {}, syscall.SYS_EPOLL_CREATE1: {}, syscall.SYS_EPOLL_CTL: {}, syscall.SYS_EPOLL_WAIT: {}, diff --git a/pkg/sentry/fs/gofer/inode.go b/pkg/sentry/fs/gofer/inode.go index 99910388f..54a8ceef8 100644 --- a/pkg/sentry/fs/gofer/inode.go +++ b/pkg/sentry/fs/gofer/inode.go @@ -268,7 +268,7 @@ func (i *inodeFileState) recreateReadHandles(ctx context.Context, writer *handle // operations on the old will see the new data. Then, make the new handle take // ownereship of the old FD and mark the old readHandle to not close the FD // when done. - if err := syscall.Dup2(h.Host.FD(), i.readHandles.Host.FD()); err != nil { + if err := syscall.Dup3(h.Host.FD(), i.readHandles.Host.FD(), 0); err != nil { return err } diff --git a/runsc/boot/BUILD b/runsc/boot/BUILD index 58e86ae7f..847d2f91c 100644 --- a/runsc/boot/BUILD +++ b/runsc/boot/BUILD @@ -15,6 +15,8 @@ go_library( "fs.go", "limits.go", "loader.go", + "loader_amd64.go", + "loader_arm64.go", "network.go", "pprof.go", "strace.go", diff --git a/runsc/boot/filter/BUILD b/runsc/boot/filter/BUILD index f5509b6b7..3a9dcfc04 100644 --- a/runsc/boot/filter/BUILD +++ b/runsc/boot/filter/BUILD @@ -6,6 +6,8 @@ go_library( name = "filter", srcs = [ "config.go", + "config_amd64.go", + "config_arm64.go", "extra_filters.go", "extra_filters_msan.go", "extra_filters_race.go", diff --git a/runsc/boot/filter/config.go b/runsc/boot/filter/config.go index 5ad108261..b5bd61a3a 100644 --- a/runsc/boot/filter/config.go +++ b/runsc/boot/filter/config.go @@ -26,10 +26,6 @@ import ( // allowedSyscalls is the set of syscalls executed by the Sentry to the host OS. var allowedSyscalls = seccomp.SyscallRules{ - syscall.SYS_ARCH_PRCTL: []seccomp.Rule{ - {seccomp.AllowValue(linux.ARCH_GET_FS)}, - {seccomp.AllowValue(linux.ARCH_SET_FS)}, - }, syscall.SYS_CLOCK_GETTIME: {}, syscall.SYS_CLONE: []seccomp.Rule{ { @@ -44,7 +40,7 @@ var allowedSyscalls = seccomp.SyscallRules{ }, syscall.SYS_CLOSE: {}, syscall.SYS_DUP: {}, - syscall.SYS_DUP2: {}, + syscall.SYS_DUP3: {}, syscall.SYS_EPOLL_CREATE1: {}, syscall.SYS_EPOLL_CTL: {}, syscall.SYS_EPOLL_PWAIT: []seccomp.Rule{ diff --git a/runsc/boot/filter/config_amd64.go b/runsc/boot/filter/config_amd64.go new file mode 100644 index 000000000..058d9c264 --- /dev/null +++ b/runsc/boot/filter/config_amd64.go @@ -0,0 +1,31 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build amd64 + +package filter + +import ( + "syscall" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/seccomp" +) + +func init() { + allowedSyscalls[syscall.SYS_ARCH_PRCTL] = []seccomp.Rule{ + {seccomp.AllowValue(linux.ARCH_GET_FS)}, + {seccomp.AllowValue(linux.ARCH_SET_FS)}, + } +} diff --git a/runsc/boot/filter/config_arm64.go b/runsc/boot/filter/config_arm64.go new file mode 100644 index 000000000..7fa9bbda3 --- /dev/null +++ b/runsc/boot/filter/config_arm64.go @@ -0,0 +1,21 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package filter + +// Reserve for future customization. +func init() { +} diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index f05d5973f..df6052c88 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -43,7 +43,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/sighandling" - slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" "gvisor.dev/gvisor/pkg/sentry/time" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/sentry/watchdog" @@ -147,9 +146,6 @@ type execProcess struct { func init() { // Initialize the random number generator. mrand.Seed(gtime.Now().UnixNano()) - - // Register the global syscall table. - kernel.RegisterSyscallTable(slinux.AMD64) } // Args are the arguments for New(). diff --git a/runsc/boot/loader_amd64.go b/runsc/boot/loader_amd64.go new file mode 100644 index 000000000..d16d20d89 --- /dev/null +++ b/runsc/boot/loader_amd64.go @@ -0,0 +1,28 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build amd64 + +// Package boot loads the kernel and runs a container. +package boot + +import ( + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" +) + +func init() { + // Register the global syscall table. + kernel.RegisterSyscallTable(linux.AMD64) +} diff --git a/runsc/boot/loader_arm64.go b/runsc/boot/loader_arm64.go new file mode 100644 index 000000000..8712e764a --- /dev/null +++ b/runsc/boot/loader_arm64.go @@ -0,0 +1,28 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +// Package boot loads the kernel and runs a container. +package boot + +import ( + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" +) + +func init() { + // Register the global syscall table. + kernel.RegisterSyscallTable(linux.ARM64) +} -- cgit v1.2.3 From c5d9b5b8816e99507661e1d39ec51033fb69e212 Mon Sep 17 00:00:00 2001 From: Haibo Xu Date: Mon, 11 Nov 2019 09:15:45 +0000 Subject: Enable sentry/fs/host support on arm64. newfstatat() syscall is not supported on arm64, so we resort to use the fstatat() syscall. Signed-off-by: Haibo Xu Change-Id: Iea95550ea53bcf85c01f7b3b95da70ad0952177d --- pkg/sentry/fs/host/BUILD | 2 ++ pkg/sentry/fs/host/util.go | 2 +- pkg/sentry/fs/host/util_amd64_unsafe.go | 41 +++++++++++++++++++++++++++++++++ pkg/sentry/fs/host/util_arm64_unsafe.go | 41 +++++++++++++++++++++++++++++++++ pkg/sentry/fs/host/util_unsafe.go | 19 --------------- 5 files changed, 85 insertions(+), 20 deletions(-) create mode 100644 pkg/sentry/fs/host/util_amd64_unsafe.go create mode 100644 pkg/sentry/fs/host/util_arm64_unsafe.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/host/BUILD b/pkg/sentry/fs/host/BUILD index 1cbed07ae..23daeb528 100644 --- a/pkg/sentry/fs/host/BUILD +++ b/pkg/sentry/fs/host/BUILD @@ -21,6 +21,8 @@ go_library( "socket_unsafe.go", "tty.go", "util.go", + "util_amd64_unsafe.go", + "util_arm64_unsafe.go", "util_unsafe.go", ], importpath = "gvisor.dev/gvisor/pkg/sentry/fs/host", diff --git a/pkg/sentry/fs/host/util.go b/pkg/sentry/fs/host/util.go index bad61a9a1..e37e687c6 100644 --- a/pkg/sentry/fs/host/util.go +++ b/pkg/sentry/fs/host/util.go @@ -155,7 +155,7 @@ func unstableAttr(mo *superOperations, s *syscall.Stat_t) fs.UnstableAttr { AccessTime: ktime.FromUnix(s.Atim.Sec, s.Atim.Nsec), ModificationTime: ktime.FromUnix(s.Mtim.Sec, s.Mtim.Nsec), StatusChangeTime: ktime.FromUnix(s.Ctim.Sec, s.Ctim.Nsec), - Links: s.Nlink, + Links: uint64(s.Nlink), } } diff --git a/pkg/sentry/fs/host/util_amd64_unsafe.go b/pkg/sentry/fs/host/util_amd64_unsafe.go new file mode 100644 index 000000000..66da6e9f5 --- /dev/null +++ b/pkg/sentry/fs/host/util_amd64_unsafe.go @@ -0,0 +1,41 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build amd64 + +package host + +import ( + "syscall" + "unsafe" +) + +func fstatat(fd int, name string, flags int) (syscall.Stat_t, error) { + var stat syscall.Stat_t + namePtr, err := syscall.BytePtrFromString(name) + if err != nil { + return stat, err + } + _, _, errno := syscall.Syscall6( + syscall.SYS_NEWFSTATAT, + uintptr(fd), + uintptr(unsafe.Pointer(namePtr)), + uintptr(unsafe.Pointer(&stat)), + uintptr(flags), + 0, 0) + if errno != 0 { + return stat, errno + } + return stat, nil +} diff --git a/pkg/sentry/fs/host/util_arm64_unsafe.go b/pkg/sentry/fs/host/util_arm64_unsafe.go new file mode 100644 index 000000000..e8cb94aeb --- /dev/null +++ b/pkg/sentry/fs/host/util_arm64_unsafe.go @@ -0,0 +1,41 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package host + +import ( + "syscall" + "unsafe" +) + +func fstatat(fd int, name string, flags int) (syscall.Stat_t, error) { + var stat syscall.Stat_t + namePtr, err := syscall.BytePtrFromString(name) + if err != nil { + return stat, err + } + _, _, errno := syscall.Syscall6( + syscall.SYS_FSTATAT, + uintptr(fd), + uintptr(unsafe.Pointer(namePtr)), + uintptr(unsafe.Pointer(&stat)), + uintptr(flags), + 0, 0) + if errno != 0 { + return stat, errno + } + return stat, nil +} diff --git a/pkg/sentry/fs/host/util_unsafe.go b/pkg/sentry/fs/host/util_unsafe.go index 2b76f1065..3ab36b088 100644 --- a/pkg/sentry/fs/host/util_unsafe.go +++ b/pkg/sentry/fs/host/util_unsafe.go @@ -116,22 +116,3 @@ func setTimestamps(fd int, ts fs.TimeSpec) error { } return nil } - -func fstatat(fd int, name string, flags int) (syscall.Stat_t, error) { - var stat syscall.Stat_t - namePtr, err := syscall.BytePtrFromString(name) - if err != nil { - return stat, err - } - _, _, errno := syscall.Syscall6( - syscall.SYS_NEWFSTATAT, - uintptr(fd), - uintptr(unsafe.Pointer(namePtr)), - uintptr(unsafe.Pointer(&stat)), - uintptr(flags), - 0, 0) - if errno != 0 { - return stat, errno - } - return stat, nil -} -- cgit v1.2.3 From 1d8b7292d72ce93d465e4ded19237fb92c08bc56 Mon Sep 17 00:00:00 2001 From: Haibo Xu Date: Mon, 11 Nov 2019 09:42:04 +0000 Subject: Fix some build errors on arm64. Initialize the VDSO "os" and "arch" fields explicitly, or the VDSO load process would failed on arm64 platform. Signed-off-by: Haibo Xu Change-Id: Ic6768df88e43cd7c7956eb630511672ae11ac52f --- pkg/sentry/kernel/ptrace_arm64.go | 1 - pkg/sentry/loader/vdso.go | 2 ++ 2 files changed, 2 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/ptrace_arm64.go b/pkg/sentry/kernel/ptrace_arm64.go index 0acdf769d..61e412911 100644 --- a/pkg/sentry/kernel/ptrace_arm64.go +++ b/pkg/sentry/kernel/ptrace_arm64.go @@ -17,7 +17,6 @@ package kernel import ( - "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/loader/vdso.go b/pkg/sentry/loader/vdso.go index ada28aea3..df8a81907 100644 --- a/pkg/sentry/loader/vdso.go +++ b/pkg/sentry/loader/vdso.go @@ -268,6 +268,8 @@ func PrepareVDSO(ctx context.Context, mfp pgalloc.MemoryFileProvider) (*VDSO, er // some applications may not be able to handle multiple [vdso] // hints. vdso: mm.NewSpecialMappable("", mfp, vdso), + os: info.os, + arch: info.arch, phdrs: info.phdrs, }, nil } -- cgit v1.2.3 From c2d3dc0c13e1adfc182c33d57e410c46fe12415f Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Tue, 12 Nov 2019 22:49:22 -0800 Subject: Use overlay MountSource when binding socket in overlay. PiperOrigin-RevId: 280131840 --- pkg/sentry/fs/inode_overlay.go | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/inode_overlay.go b/pkg/sentry/fs/inode_overlay.go index 5a388dad1..a09147080 100644 --- a/pkg/sentry/fs/inode_overlay.go +++ b/pkg/sentry/fs/inode_overlay.go @@ -462,7 +462,9 @@ func overlayBind(ctx context.Context, o *overlayEntry, parent *Dirent, name stri inode.DecRef() return nil, err } - return NewDirent(ctx, newOverlayInode(ctx, entry, inode.MountSource), name), nil + // Use the parent's MountSource, since that corresponds to the overlay, + // and not the upper filesystem. + return NewDirent(ctx, newOverlayInode(ctx, entry, parent.Inode.MountSource), name), nil } func overlayBoundEndpoint(o *overlayEntry, path string) transport.BoundEndpoint { -- cgit v1.2.3 From 9ca15dbf14ac6318a34540354020b8a71d789077 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Thu, 14 Nov 2019 14:03:24 -0800 Subject: Avoid unnecessary slice allocation in usermem.BytesIO.blocksFromAddrRanges(). PiperOrigin-RevId: 280507239 --- pkg/sentry/usermem/bytes_io.go | 37 ++++++++++++++++++++++++++----------- 1 file changed, 26 insertions(+), 11 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/usermem/bytes_io.go b/pkg/sentry/usermem/bytes_io.go index 8d88396ba..7898851b3 100644 --- a/pkg/sentry/usermem/bytes_io.go +++ b/pkg/sentry/usermem/bytes_io.go @@ -102,19 +102,34 @@ func (b *BytesIO) rangeCheck(addr Addr, length int) (int, error) { } func (b *BytesIO) blocksFromAddrRanges(ars AddrRangeSeq) (safemem.BlockSeq, error) { - blocks := make([]safemem.Block, 0, ars.NumRanges()) - for !ars.IsEmpty() { - ar := ars.Head() - n, err := b.rangeCheck(ar.Start, int(ar.Length())) - if n != 0 { - blocks = append(blocks, safemem.BlockFromSafeSlice(b.Bytes[int(ar.Start):int(ar.Start)+n])) + switch ars.NumRanges() { + case 0: + return safemem.BlockSeq{}, nil + case 1: + block, err := b.blockFromAddrRange(ars.Head()) + return safemem.BlockSeqOf(block), err + default: + blocks := make([]safemem.Block, 0, ars.NumRanges()) + for !ars.IsEmpty() { + block, err := b.blockFromAddrRange(ars.Head()) + if block.Len() != 0 { + blocks = append(blocks, block) + } + if err != nil { + return safemem.BlockSeqFromSlice(blocks), err + } + ars = ars.Tail() } - if err != nil { - return safemem.BlockSeqFromSlice(blocks), err - } - ars = ars.Tail() + return safemem.BlockSeqFromSlice(blocks), nil + } +} + +func (b *BytesIO) blockFromAddrRange(ar AddrRange) (safemem.Block, error) { + n, err := b.rangeCheck(ar.Start, int(ar.Length())) + if n == 0 { + return safemem.Block{}, err } - return safemem.BlockSeqFromSlice(blocks), nil + return safemem.BlockFromSafeSlice(b.Bytes[int(ar.Start) : int(ar.Start)+n]), err } // BytesIOSequence returns an IOSequence representing the given byte slice. -- cgit v1.2.3 From 339536de5eefe782813aabae4aeeb312b3c4dde7 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Thu, 14 Nov 2019 15:55:07 -0800 Subject: Check that a file is a regular file with open(O_TRUNC). It was possible to panic the sentry by opening a cache revalidating folder with O_TRUNC|O_CREAT. Avoids breaking php tests. PiperOrigin-RevId: 280533213 --- pkg/sentry/fs/inode.go | 4 ++++ pkg/sentry/fs/tty/master.go | 1 + pkg/sentry/fs/tty/slave.go | 1 + pkg/sentry/syscalls/linux/sys_file.go | 9 +++++---- test/syscalls/linux/open.cc | 22 ++++++++++++++++++++++ test/syscalls/linux/open_create.cc | 24 ++++++++++++++++++++++++ test/syscalls/linux/pty.cc | 20 +++++++++++++++++++- test/util/pty_util.cc | 10 +++++++++- test/util/pty_util.h | 3 +++ 9 files changed, 88 insertions(+), 6 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/inode.go b/pkg/sentry/fs/inode.go index f4ddfa406..2d43dff1d 100644 --- a/pkg/sentry/fs/inode.go +++ b/pkg/sentry/fs/inode.go @@ -344,6 +344,10 @@ func (i *Inode) SetTimestamps(ctx context.Context, d *Dirent, ts TimeSpec) error // Truncate calls i.InodeOperations.Truncate with i as the Inode. func (i *Inode) Truncate(ctx context.Context, d *Dirent, size int64) error { + if IsDir(i.StableAttr) { + return syserror.EISDIR + } + if i.overlay != nil { return overlayTruncate(ctx, i.overlay, d, size) } diff --git a/pkg/sentry/fs/tty/master.go b/pkg/sentry/fs/tty/master.go index 19b7557d5..bc56be696 100644 --- a/pkg/sentry/fs/tty/master.go +++ b/pkg/sentry/fs/tty/master.go @@ -32,6 +32,7 @@ import ( // +stateify savable type masterInodeOperations struct { fsutil.SimpleFileInode + fsutil.InodeNoopTruncate // d is the containing dir. d *dirInodeOperations diff --git a/pkg/sentry/fs/tty/slave.go b/pkg/sentry/fs/tty/slave.go index 944c4ada1..4cbea0367 100644 --- a/pkg/sentry/fs/tty/slave.go +++ b/pkg/sentry/fs/tty/slave.go @@ -31,6 +31,7 @@ import ( // +stateify savable type slaveInodeOperations struct { fsutil.SimpleFileInode + fsutil.InodeNoopTruncate // d is the containing dir. d *dirInodeOperations diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go index b9a8e3e21..167c2b60b 100644 --- a/pkg/sentry/syscalls/linux/sys_file.go +++ b/pkg/sentry/syscalls/linux/sys_file.go @@ -169,10 +169,11 @@ func openAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint) (fd uint if dirPath { return syserror.ENOTDIR } - if flags&linux.O_TRUNC != 0 { - if err := d.Inode.Truncate(t, d, 0); err != nil { - return err - } + } + + if flags&linux.O_TRUNC != 0 { + if err := d.Inode.Truncate(t, d, 0); err != nil { + return err } } diff --git a/test/syscalls/linux/open.cc b/test/syscalls/linux/open.cc index 2b1df52ce..267ae19f6 100644 --- a/test/syscalls/linux/open.cc +++ b/test/syscalls/linux/open.cc @@ -73,6 +73,28 @@ class OpenTest : public FileTest { const std::string test_data_ = "hello world\n"; }; +TEST_F(OpenTest, OTrunc) { + auto dirpath = JoinPath(GetAbsoluteTestTmpdir(), "truncd"); + ASSERT_THAT(mkdir(dirpath.c_str(), 0777), SyscallSucceeds()); + ASSERT_THAT(open(dirpath.c_str(), O_TRUNC, 0666), + SyscallFailsWithErrno(EISDIR)); +} + +TEST_F(OpenTest, OTruncAndReadOnlyDir) { + auto dirpath = JoinPath(GetAbsoluteTestTmpdir(), "truncd"); + ASSERT_THAT(mkdir(dirpath.c_str(), 0777), SyscallSucceeds()); + ASSERT_THAT(open(dirpath.c_str(), O_TRUNC | O_RDONLY, 0666), + SyscallFailsWithErrno(EISDIR)); +} + +TEST_F(OpenTest, OTruncAndReadOnlyFile) { + auto dirpath = JoinPath(GetAbsoluteTestTmpdir(), "truncfile"); + const FileDescriptor existing = + ASSERT_NO_ERRNO_AND_VALUE(Open(dirpath.c_str(), O_RDWR | O_CREAT, 0666)); + const FileDescriptor otrunc = ASSERT_NO_ERRNO_AND_VALUE( + Open(dirpath.c_str(), O_TRUNC | O_RDONLY, 0666)); +} + TEST_F(OpenTest, ReadOnly) { char buf; const FileDescriptor ro_file = diff --git a/test/syscalls/linux/open_create.cc b/test/syscalls/linux/open_create.cc index e5a85ef9d..431733dbe 100644 --- a/test/syscalls/linux/open_create.cc +++ b/test/syscalls/linux/open_create.cc @@ -88,6 +88,30 @@ TEST(CreateTest, CreateExclusively) { SyscallFailsWithErrno(EEXIST)); } +TEST(CreateTeast, CreatWithOTrunc) { + std::string dirpath = JoinPath(GetAbsoluteTestTmpdir(), "truncd"); + ASSERT_THAT(mkdir(dirpath.c_str(), 0777), SyscallSucceeds()); + ASSERT_THAT(open(dirpath.c_str(), O_CREAT | O_TRUNC, 0666), + SyscallFailsWithErrno(EISDIR)); +} + +TEST(CreateTeast, CreatDirWithOTruncAndReadOnly) { + std::string dirpath = JoinPath(GetAbsoluteTestTmpdir(), "truncd"); + ASSERT_THAT(mkdir(dirpath.c_str(), 0777), SyscallSucceeds()); + ASSERT_THAT(open(dirpath.c_str(), O_CREAT | O_TRUNC | O_RDONLY, 0666), + SyscallFailsWithErrno(EISDIR)); +} + +TEST(CreateTeast, CreatFileWithOTruncAndReadOnly) { + std::string dirpath = JoinPath(GetAbsoluteTestTmpdir(), "truncfile"); + int dirfd; + ASSERT_THAT(dirfd = open(dirpath.c_str(), O_RDWR | O_CREAT, 0666), + SyscallSucceeds()); + ASSERT_THAT(open(dirpath.c_str(), O_CREAT | O_TRUNC | O_RDONLY, 0666), + SyscallSucceeds()); + ASSERT_THAT(close(dirfd), SyscallSucceeds()); +} + TEST(CreateTest, CreateFailsOnUnpermittedDir) { // Make sure we don't have CAP_DAC_OVERRIDE, since that allows the user to // always override directory permissions. diff --git a/test/syscalls/linux/pty.cc b/test/syscalls/linux/pty.cc index 99a0df235..dafe64d20 100644 --- a/test/syscalls/linux/pty.cc +++ b/test/syscalls/linux/pty.cc @@ -70,6 +70,8 @@ constexpr absl::Duration kTimeout = absl::Seconds(20); // The maximum line size in bytes returned per read from a pty file. constexpr int kMaxLineSize = 4096; +constexpr char kMasterPath[] = "/dev/ptmx"; + // glibc defines its own, different, version of struct termios. We care about // what the kernel does, not glibc. #define KERNEL_NCCS 19 @@ -376,9 +378,25 @@ PosixErrorOr PollAndReadFd(int fd, void* buf, size_t count, return PosixError(ETIMEDOUT, "Poll timed out"); } +TEST(PtyTrunc, Truncate) { + // Opening PTYs with O_TRUNC shouldn't cause an error, but calls to + // (f)truncate should. + FileDescriptor master = + ASSERT_NO_ERRNO_AND_VALUE(Open(kMasterPath, O_RDWR | O_TRUNC)); + int n = ASSERT_NO_ERRNO_AND_VALUE(SlaveID(master)); + std::string spath = absl::StrCat("/dev/pts/", n); + FileDescriptor slave = + ASSERT_NO_ERRNO_AND_VALUE(Open(spath, O_RDWR | O_NONBLOCK | O_TRUNC)); + + EXPECT_THAT(truncate(kMasterPath, 0), SyscallFailsWithErrno(EINVAL)); + EXPECT_THAT(truncate(spath.c_str(), 0), SyscallFailsWithErrno(EINVAL)); + EXPECT_THAT(ftruncate(master.get(), 0), SyscallFailsWithErrno(EINVAL)); + EXPECT_THAT(ftruncate(slave.get(), 0), SyscallFailsWithErrno(EINVAL)); +} + TEST(BasicPtyTest, StatUnopenedMaster) { struct stat s; - ASSERT_THAT(stat("/dev/ptmx", &s), SyscallSucceeds()); + ASSERT_THAT(stat(kMasterPath, &s), SyscallSucceeds()); EXPECT_EQ(s.st_rdev, makedev(TTYAUX_MAJOR, kPtmxMinor)); EXPECT_EQ(s.st_size, 0); diff --git a/test/util/pty_util.cc b/test/util/pty_util.cc index c0fd9a095..c01f916aa 100644 --- a/test/util/pty_util.cc +++ b/test/util/pty_util.cc @@ -24,6 +24,14 @@ namespace gvisor { namespace testing { PosixErrorOr OpenSlave(const FileDescriptor& master) { + PosixErrorOr n = SlaveID(master); + if (!n.ok()) { + return PosixErrorOr(n.error()); + } + return Open(absl::StrCat("/dev/pts/", n.ValueOrDie()), O_RDWR | O_NONBLOCK); +} + +PosixErrorOr SlaveID(const FileDescriptor& master) { // Get pty index. int n; int ret = ioctl(master.get(), TIOCGPTN, &n); @@ -38,7 +46,7 @@ PosixErrorOr OpenSlave(const FileDescriptor& master) { return PosixError(errno, "ioctl(TIOSPTLCK) failed"); } - return Open(absl::StrCat("/dev/pts/", n), O_RDWR | O_NONBLOCK); + return n; } } // namespace testing diff --git a/test/util/pty_util.h b/test/util/pty_util.h index 367b14f15..0722da379 100644 --- a/test/util/pty_util.h +++ b/test/util/pty_util.h @@ -24,6 +24,9 @@ namespace testing { // Opens the slave end of the passed master as R/W and nonblocking. PosixErrorOr OpenSlave(const FileDescriptor& master); +// Get the number of the slave end of the master. +PosixErrorOr SlaveID(const FileDescriptor& master); + } // namespace testing } // namespace gvisor -- cgit v1.2.3 From af323eb7c1830053627de6161f8ce73ac5f06d4e Mon Sep 17 00:00:00 2001 From: Ting-Yu Wang Date: Thu, 14 Nov 2019 17:02:59 -0800 Subject: Fix return codes for {get,set}sockopt for some nullptr cases. Updates #1092 PiperOrigin-RevId: 280547239 --- pkg/sentry/syscalls/linux/sys_socket.go | 23 +++++++++-------------- test/syscalls/linux/socket_ip_unbound.cc | 32 ++++++++++++++++++++++++++++++++ 2 files changed, 41 insertions(+), 14 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/sys_socket.go b/pkg/sentry/syscalls/linux/sys_socket.go index b5a72ce63..ab1001f16 100644 --- a/pkg/sentry/syscalls/linux/sys_socket.go +++ b/pkg/sentry/syscalls/linux/sys_socket.go @@ -447,16 +447,13 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy return 0, nil, syserror.ENOTSOCK } - // Read the length if present. Reject negative values. + // Read the length. Reject negative values. optLen := int32(0) - if optLenAddr != 0 { - if _, err := t.CopyIn(optLenAddr, &optLen); err != nil { - return 0, nil, err - } - - if optLen < 0 { - return 0, nil, syserror.EINVAL - } + if _, err := t.CopyIn(optLenAddr, &optLen); err != nil { + return 0, nil, err + } + if optLen < 0 { + return 0, nil, syserror.EINVAL } // Call syscall implementation then copy both value and value len out. @@ -465,11 +462,9 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy return 0, nil, e.ToError() } - if optLenAddr != 0 { - vLen := int32(binary.Size(v)) - if _, err := t.CopyOut(optLenAddr, vLen); err != nil { - return 0, nil, err - } + vLen := int32(binary.Size(v)) + if _, err := t.CopyOut(optLenAddr, vLen); err != nil { + return 0, nil, err } if v != nil { diff --git a/test/syscalls/linux/socket_ip_unbound.cc b/test/syscalls/linux/socket_ip_unbound.cc index b02872308..b6754111f 100644 --- a/test/syscalls/linux/socket_ip_unbound.cc +++ b/test/syscalls/linux/socket_ip_unbound.cc @@ -354,6 +354,38 @@ TEST_P(IPUnboundSocketTest, InvalidNegativeTOS) { EXPECT_EQ(get, expect); } +TEST_P(IPUnboundSocketTest, NullTOS) { + auto socket = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + TOSOption t = GetTOSOption(GetParam().domain); + int set_sz = sizeof(int); + if (GetParam().domain == AF_INET) { + EXPECT_THAT(setsockopt(socket->get(), t.level, t.option, nullptr, set_sz), + SyscallFailsWithErrno(EFAULT)); + } else { // AF_INET6 + // The AF_INET6 behavior is not yet compatible. gVisor will try to read + // optval from user memory at syscall handler, it needs substantial + // refactoring to implement this behavior just for IPv6. + if (IsRunningOnGvisor()) { + EXPECT_THAT(setsockopt(socket->get(), t.level, t.option, nullptr, set_sz), + SyscallFailsWithErrno(EFAULT)); + } else { + // Linux's IPv6 stack treats nullptr optval as input of 0, so the call + // succeeds. (net/ipv6/ipv6_sockglue.c, do_ipv6_setsockopt()) + // + // Linux's implementation would need fixing as passing a nullptr as optval + // and non-zero optlen may not be valid. + EXPECT_THAT(setsockopt(socket->get(), t.level, t.option, nullptr, set_sz), + SyscallSucceedsWithValue(0)); + } + } + socklen_t get_sz = sizeof(int); + EXPECT_THAT(getsockopt(socket->get(), t.level, t.option, nullptr, &get_sz), + SyscallFailsWithErrno(EFAULT)); + int get = -1; + EXPECT_THAT(getsockopt(socket->get(), t.level, t.option, &get, nullptr), + SyscallFailsWithErrno(EFAULT)); +} + INSTANTIATE_TEST_SUITE_P( IPUnboundSockets, IPUnboundSocketTest, ::testing::ValuesIn(VecCat(VecCat( -- cgit v1.2.3 From 26b3341b9ae08bb72971d5465c77e6c8db82c996 Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Mon, 18 Nov 2019 14:55:30 -0800 Subject: platform/ptrace: use host.GetCPU instead of the getcpu syscall This should save ~200ns from switchToApp (on ptrace too). // mpratt PiperOrigin-RevId: 281159895 --- pkg/sentry/platform/ptrace/BUILD | 1 + .../platform/ptrace/subprocess_linux_unsafe.go | 21 +++------------------ 2 files changed, 4 insertions(+), 18 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/ptrace/BUILD b/pkg/sentry/platform/ptrace/BUILD index ebcc8c098..0df8cfa0f 100644 --- a/pkg/sentry/platform/ptrace/BUILD +++ b/pkg/sentry/platform/ptrace/BUILD @@ -28,6 +28,7 @@ go_library( "//pkg/procid", "//pkg/seccomp", "//pkg/sentry/arch", + "//pkg/sentry/hostcpu", "//pkg/sentry/platform", "//pkg/sentry/platform/interrupt", "//pkg/sentry/platform/safecopy", diff --git a/pkg/sentry/platform/ptrace/subprocess_linux_unsafe.go b/pkg/sentry/platform/ptrace/subprocess_linux_unsafe.go index de6783fb0..2e6fbe488 100644 --- a/pkg/sentry/platform/ptrace/subprocess_linux_unsafe.go +++ b/pkg/sentry/platform/ptrace/subprocess_linux_unsafe.go @@ -25,6 +25,7 @@ import ( "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/hostcpu" ) // maskPool contains reusable CPU masks for setting affinity. Unfortunately, @@ -49,20 +50,6 @@ func unmaskAllSignals() syscall.Errno { return errno } -// getCPU gets the current CPU. -// -// Precondition: the current runtime thread should be locked. -func getCPU() (uint32, error) { - var cpu uintptr - if _, _, errno := syscall.RawSyscall( - unix.SYS_GETCPU, - uintptr(unsafe.Pointer(&cpu)), - 0, 0); errno != 0 { - return 0, errno - } - return uint32(cpu), nil -} - // setCPU sets the CPU affinity. func (t *thread) setCPU(cpu uint32) error { mask := maskPool.Get().([]uintptr) @@ -93,10 +80,8 @@ func (t *thread) setCPU(cpu uint32) error { // // Precondition: the current runtime thread should be locked. func (t *thread) bind() { - currentCPU, err := getCPU() - if err != nil { - return - } + currentCPU := hostcpu.GetCPU() + if oldCPU := atomic.SwapUint32(&t.cpu, currentCPU); oldCPU != currentCPU { // Set the affinity on the thread and save the CPU for next // round; we don't expect CPUs to bounce around too frequently. -- cgit v1.2.3 From ef6f93625457c166628fc9de57c15d986ae83159 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Mon, 18 Nov 2019 16:25:03 -0800 Subject: Add vfs.GenericParseMountOptions(). Equivalent to fs.GenericMountSourceOptions(). PiperOrigin-RevId: 281179287 --- pkg/sentry/vfs/BUILD | 1 + pkg/sentry/vfs/filesystem_impl_util.go | 43 ++++++++++++++++++++++++++++++++++ 2 files changed, 44 insertions(+) create mode 100644 pkg/sentry/vfs/filesystem_impl_util.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index eff4b44f6..4f2c2de9f 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -12,6 +12,7 @@ go_library( "file_description.go", "file_description_impl_util.go", "filesystem.go", + "filesystem_impl_util.go", "filesystem_type.go", "mount.go", "mount_unsafe.go", diff --git a/pkg/sentry/vfs/filesystem_impl_util.go b/pkg/sentry/vfs/filesystem_impl_util.go new file mode 100644 index 000000000..465e610e0 --- /dev/null +++ b/pkg/sentry/vfs/filesystem_impl_util.go @@ -0,0 +1,43 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs + +import ( + "strings" +) + +// GenericParseMountOptions parses a comma-separated list of options of the +// form "key" or "key=value", where neither key nor value contain commas, and +// returns it as a map. If str contains duplicate keys, then the last value +// wins. For example: +// +// str = "key0=value0,key1,key2=value2,key0=value3" -> map{'key0':'value3','key1':'','key2':'value2'} +// +// GenericParseMountOptions is not appropriate if values may contain commas, +// e.g. in the case of the mpol mount option for tmpfs(5). +func GenericParseMountOptions(str string) map[string]string { + m := make(map[string]string) + for _, opt := range strings.Split(str, ",") { + if len(opt) > 0 { + res := strings.SplitN(opt, "=", 2) + if len(res) == 2 { + m[res[0]] = res[1] + } else { + m[opt] = "" + } + } + } + return m +} -- cgit v1.2.3 From 012102eefd2b145ddee774cba28e4fa889fadd49 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Wed, 20 Nov 2019 14:54:03 -0800 Subject: Pass OpenTruncate to gofer in Open call when opening file with O_TRUNC. Note that the Sentry still calls Truncate() on the file before calling Open. A new p9 version check was added to ensure that the p9 server can handle the the OpenTruncate flag. If not, then the flag is stripped before sending. PiperOrigin-RevId: 281609112 --- pkg/p9/handlers.go | 18 ++++----- pkg/p9/p9.go | 4 -- pkg/p9/p9test/client_test.go | 78 +++++++++++++++++++++++--------------- pkg/p9/version.go | 8 +++- pkg/sentry/fs/flags.go | 7 ++++ pkg/sentry/fs/gofer/file_state.go | 8 +++- pkg/sentry/fs/gofer/handles.go | 5 ++- pkg/sentry/fs/gofer/inode.go | 18 ++++++--- pkg/sentry/syscalls/linux/flags.go | 1 + 9 files changed, 95 insertions(+), 52 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/p9/handlers.go b/pkg/p9/handlers.go index ba9a55d6d..51869c7d6 100644 --- a/pkg/p9/handlers.go +++ b/pkg/p9/handlers.go @@ -272,15 +272,15 @@ func (t *Tlopen) handle(cs *connState) message { return newErr(syscall.EINVAL) } - // Are flags valid? - flags := t.Flags &^ OpenFlagsIgnoreMask - if flags&^OpenFlagsModeMask != 0 { - return newErr(syscall.EINVAL) - } - - // Is this an attempt to open a directory as writable? Don't accept. - if ref.mode.IsDir() && flags != ReadOnly { - return newErr(syscall.EINVAL) + if ref.mode.IsDir() { + // Directory must be opened ReadOnly. + if t.Flags&OpenFlagsModeMask != ReadOnly { + return newErr(syscall.EISDIR) + } + // Directory not truncatable. + if t.Flags&OpenTruncate != 0 { + return newErr(syscall.EISDIR) + } } var ( diff --git a/pkg/p9/p9.go b/pkg/p9/p9.go index 415200d60..d3090535a 100644 --- a/pkg/p9/p9.go +++ b/pkg/p9/p9.go @@ -47,10 +47,6 @@ const ( // OpenTruncate is a Tlopen flag indicating that the opened file should be // truncated. OpenTruncate OpenFlags = 01000 - - // OpenFlagsIgnoreMask is a list of OpenFlags mode bits that are ignored for Tlopen. - // Note that syscall.O_LARGEFILE is set to zero, use value from Linux fcntl.h. - OpenFlagsIgnoreMask OpenFlags = syscall.O_DIRECTORY | syscall.O_NOATIME | 0100000 ) // ConnectFlags is the mode passed to Connect operations. diff --git a/pkg/p9/p9test/client_test.go b/pkg/p9/p9test/client_test.go index 8bbdb2488..6e758148d 100644 --- a/pkg/p9/p9test/client_test.go +++ b/pkg/p9/p9test/client_test.go @@ -1044,11 +1044,11 @@ func TestReaddir(t *testing.T) { if _, err := f.Readdir(0, 1); err != syscall.EINVAL { t.Errorf("readdir got %v, wanted EINVAL", err) } - if _, _, _, err := f.Open(p9.ReadWrite); err != syscall.EINVAL { - t.Errorf("readdir got %v, wanted EINVAL", err) + if _, _, _, err := f.Open(p9.ReadWrite); err != syscall.EISDIR { + t.Errorf("readdir got %v, wanted EISDIR", err) } - if _, _, _, err := f.Open(p9.WriteOnly); err != syscall.EINVAL { - t.Errorf("readdir got %v, wanted EINVAL", err) + if _, _, _, err := f.Open(p9.WriteOnly); err != syscall.EISDIR { + t.Errorf("readdir got %v, wanted EISDIR", err) } backend.EXPECT().Open(p9.ReadOnly).Times(1) if _, _, _, err := f.Open(p9.ReadOnly); err != nil { @@ -1065,75 +1065,93 @@ func TestReaddir(t *testing.T) { func TestOpen(t *testing.T) { type openTest struct { name string - mode p9.OpenFlags + flags p9.OpenFlags err error match func(p9.FileMode) bool } cases := []openTest{ - { - name: "invalid", - mode: ^p9.OpenFlagsModeMask, - err: syscall.EINVAL, - match: func(p9.FileMode) bool { return true }, - }, { name: "not-openable-read-only", - mode: p9.ReadOnly, + flags: p9.ReadOnly, err: syscall.EINVAL, match: func(mode p9.FileMode) bool { return !p9.CanOpen(mode) }, }, { name: "not-openable-write-only", - mode: p9.WriteOnly, + flags: p9.WriteOnly, err: syscall.EINVAL, match: func(mode p9.FileMode) bool { return !p9.CanOpen(mode) }, }, { name: "not-openable-read-write", - mode: p9.ReadWrite, + flags: p9.ReadWrite, err: syscall.EINVAL, match: func(mode p9.FileMode) bool { return !p9.CanOpen(mode) }, }, { name: "directory-read-only", - mode: p9.ReadOnly, + flags: p9.ReadOnly, err: nil, match: func(mode p9.FileMode) bool { return mode.IsDir() }, }, { name: "directory-read-write", - mode: p9.ReadWrite, - err: syscall.EINVAL, + flags: p9.ReadWrite, + err: syscall.EISDIR, match: func(mode p9.FileMode) bool { return mode.IsDir() }, }, { name: "directory-write-only", - mode: p9.WriteOnly, - err: syscall.EINVAL, + flags: p9.WriteOnly, + err: syscall.EISDIR, match: func(mode p9.FileMode) bool { return mode.IsDir() }, }, { name: "read-only", - mode: p9.ReadOnly, + flags: p9.ReadOnly, err: nil, match: func(mode p9.FileMode) bool { return p9.CanOpen(mode) }, }, { name: "write-only", - mode: p9.WriteOnly, + flags: p9.WriteOnly, err: nil, match: func(mode p9.FileMode) bool { return p9.CanOpen(mode) && !mode.IsDir() }, }, { name: "read-write", - mode: p9.ReadWrite, + flags: p9.ReadWrite, + err: nil, + match: func(mode p9.FileMode) bool { return p9.CanOpen(mode) && !mode.IsDir() }, + }, + { + name: "directory-read-only-truncate", + flags: p9.ReadOnly | p9.OpenTruncate, + err: syscall.EISDIR, + match: func(mode p9.FileMode) bool { return mode.IsDir() }, + }, + { + name: "read-only-truncate", + flags: p9.ReadOnly | p9.OpenTruncate, + err: nil, + match: func(mode p9.FileMode) bool { return p9.CanOpen(mode) && !mode.IsDir() }, + }, + { + name: "write-only-truncate", + flags: p9.WriteOnly | p9.OpenTruncate, + err: nil, + match: func(mode p9.FileMode) bool { return p9.CanOpen(mode) && !mode.IsDir() }, + }, + { + name: "read-write-truncate", + flags: p9.ReadWrite | p9.OpenTruncate, err: nil, match: func(mode p9.FileMode) bool { return p9.CanOpen(mode) && !mode.IsDir() }, }, } - // Open(mode OpenFlags) (*fd.FD, QID, uint32, error) + // Open(flags OpenFlags) (*fd.FD, QID, uint32, error) // - only works on Regular, NamedPipe, BLockDevice, CharacterDevice // - returning a file works as expected for name := range newTypeMap(nil) { @@ -1171,25 +1189,25 @@ func TestOpen(t *testing.T) { // Attempt the given open. if tc.err != nil { // We expect an error, just test and return. - if _, _, _, err := f.Open(tc.mode); err != tc.err { - t.Fatalf("open with mode %v got %v, want %v", tc.mode, err, tc.err) + if _, _, _, err := f.Open(tc.flags); err != tc.err { + t.Fatalf("open with flags %v got %v, want %v", tc.flags, err, tc.err) } return } // Run an FD test, since we expect success. fdTest(t, func(send *fd.FD) *fd.FD { - backend.EXPECT().Open(tc.mode).Return(send, p9.QID{}, uint32(0), nil).Times(1) - recv, _, _, err := f.Open(tc.mode) + backend.EXPECT().Open(tc.flags).Return(send, p9.QID{}, uint32(0), nil).Times(1) + recv, _, _, err := f.Open(tc.flags) if err != tc.err { - t.Fatalf("open with mode %v got %v, want %v", tc.mode, err, tc.err) + t.Fatalf("open with flags %v got %v, want %v", tc.flags, err, tc.err) } return recv }) // If the open was successful, attempt another one. - if _, _, _, err := f.Open(tc.mode); err != syscall.EINVAL { - t.Errorf("second open with mode %v got %v, want EINVAL", tc.mode, err) + if _, _, _, err := f.Open(tc.flags); err != syscall.EINVAL { + t.Errorf("second open with flags %v got %v, want EINVAL", tc.flags, err) } // Ensure that all illegal operations fail. diff --git a/pkg/p9/version.go b/pkg/p9/version.go index f1ffdd23a..36a694c58 100644 --- a/pkg/p9/version.go +++ b/pkg/p9/version.go @@ -26,7 +26,7 @@ const ( // // Clients are expected to start requesting this version number and // to continuously decrement it until a Tversion request succeeds. - highestSupportedVersion uint32 = 8 + highestSupportedVersion uint32 = 9 // lowestSupportedVersion is the lowest supported version X in a // version string of the format 9P2000.L.Google.X. @@ -155,3 +155,9 @@ func versionSupportsTallocate(v uint32) bool { func versionSupportsFlipcall(v uint32) bool { return v >= 8 } + +// VersionSupportsOpenTruncateFlag returns true if version v supports +// passing the OpenTruncate flag to Tlopen. +func VersionSupportsOpenTruncateFlag(v uint32) bool { + return v >= 9 +} diff --git a/pkg/sentry/fs/flags.go b/pkg/sentry/fs/flags.go index 0fab876a9..4338ae1fa 100644 --- a/pkg/sentry/fs/flags.go +++ b/pkg/sentry/fs/flags.go @@ -64,6 +64,10 @@ type FileFlags struct { // NonSeekable indicates that file.offset isn't used. NonSeekable bool + + // Truncate indicates that the file should be truncated before opened. + // This is only applicable if the file is regular. + Truncate bool } // SettableFileFlags is a subset of FileFlags above that can be changed @@ -118,6 +122,9 @@ func (f FileFlags) ToLinux() (mask uint) { if f.LargeFile { mask |= linux.O_LARGEFILE } + if f.Truncate { + mask |= linux.O_TRUNC + } switch { case f.Read && f.Write: diff --git a/pkg/sentry/fs/gofer/file_state.go b/pkg/sentry/fs/gofer/file_state.go index c2fbb4be9..bb8312849 100644 --- a/pkg/sentry/fs/gofer/file_state.go +++ b/pkg/sentry/fs/gofer/file_state.go @@ -28,8 +28,14 @@ func (f *fileOperations) afterLoad() { // Manually load the open handles. var err error + + // The file may have been opened with Truncate, but we don't + // want to re-open it with Truncate or we will lose data. + flags := f.flags + flags.Truncate = false + // TODO(b/38173783): Context is not plumbed to save/restore. - f.handles, err = f.inodeOperations.fileState.getHandles(context.Background(), f.flags, f.inodeOperations.cachingInodeOps) + f.handles, err = f.inodeOperations.fileState.getHandles(context.Background(), flags, f.inodeOperations.cachingInodeOps) if err != nil { return fmt.Errorf("failed to re-open handle: %v", err) } diff --git a/pkg/sentry/fs/gofer/handles.go b/pkg/sentry/fs/gofer/handles.go index 39c8ec33d..b86c49b39 100644 --- a/pkg/sentry/fs/gofer/handles.go +++ b/pkg/sentry/fs/gofer/handles.go @@ -64,7 +64,7 @@ func (h *handles) DecRef() { }) } -func newHandles(ctx context.Context, file contextFile, flags fs.FileFlags) (*handles, error) { +func newHandles(ctx context.Context, client *p9.Client, file contextFile, flags fs.FileFlags) (*handles, error) { _, newFile, err := file.walk(ctx, nil) if err != nil { return nil, err @@ -81,6 +81,9 @@ func newHandles(ctx context.Context, file contextFile, flags fs.FileFlags) (*han default: panic("impossible fs.FileFlags") } + if flags.Truncate && p9.VersionSupportsOpenTruncateFlag(client.Version()) { + p9flags |= p9.OpenTruncate + } hostFile, _, _, err := newFile.open(ctx, p9flags) if err != nil { diff --git a/pkg/sentry/fs/gofer/inode.go b/pkg/sentry/fs/gofer/inode.go index 99910388f..4237bf353 100644 --- a/pkg/sentry/fs/gofer/inode.go +++ b/pkg/sentry/fs/gofer/inode.go @@ -180,7 +180,7 @@ func (i *inodeFileState) setSharedHandlesLocked(flags fs.FileFlags, h *handles) // given flags. func (i *inodeFileState) getHandles(ctx context.Context, flags fs.FileFlags, cache *fsutil.CachingInodeOperations) (*handles, error) { if !i.canShareHandles() { - return newHandles(ctx, i.file, flags) + return newHandles(ctx, i.s.client, i.file, flags) } i.handlesMu.Lock() @@ -201,19 +201,25 @@ func (i *inodeFileState) getHandles(ctx context.Context, flags fs.FileFlags, cac // whether previously open read handle was recreated. Host mappings must be // invalidated if so. func (i *inodeFileState) getHandlesLocked(ctx context.Context, flags fs.FileFlags) (*handles, bool, error) { - // Do we already have usable shared handles? - if flags.Write { + // Check if we are able to use cached handles. + if flags.Truncate && p9.VersionSupportsOpenTruncateFlag(i.s.client.Version()) { + // If we are truncating (and the gofer supports it), then we + // always need a new handle. Don't return one from the cache. + } else if flags.Write { if i.writeHandles != nil && (i.writeHandlesRW || !flags.Read) { + // File is opened for writing, and we have cached write + // handles that we can use. i.writeHandles.IncRef() return i.writeHandles, false, nil } } else if i.readHandles != nil { + // File is opened for reading and we have cached handles. i.readHandles.IncRef() return i.readHandles, false, nil } - // No; get new handles and cache them for future sharing. - h, err := newHandles(ctx, i.file, flags) + // Get new handles and cache them for future sharing. + h, err := newHandles(ctx, i.s.client, i.file, flags) if err != nil { return nil, false, err } @@ -239,7 +245,7 @@ func (i *inodeFileState) recreateReadHandles(ctx context.Context, writer *handle if !flags.Read { // Writer can't be used for read, must create a new handle. var err error - h, err = newHandles(ctx, i.file, fs.FileFlags{Read: true}) + h, err = newHandles(ctx, i.s.client, i.file, fs.FileFlags{Read: true}) if err != nil { return err } diff --git a/pkg/sentry/syscalls/linux/flags.go b/pkg/sentry/syscalls/linux/flags.go index 444f2b004..07961dad9 100644 --- a/pkg/sentry/syscalls/linux/flags.go +++ b/pkg/sentry/syscalls/linux/flags.go @@ -50,5 +50,6 @@ func linuxToFlags(mask uint) fs.FileFlags { Directory: mask&linux.O_DIRECTORY != 0, Async: mask&linux.O_ASYNC != 0, LargeFile: mask&linux.O_LARGEFILE != 0, + Truncate: mask&linux.O_TRUNC != 0, } } -- cgit v1.2.3 From c0f89eba6ebdec08460bd796fc62d6aef674d141 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Thu, 21 Nov 2019 11:29:49 -0800 Subject: Import and structure cleanup. PiperOrigin-RevId: 281795269 --- pkg/eventchannel/BUILD | 1 + pkg/flipcall/BUILD | 2 +- pkg/flipcall/flipcall_unsafe.go | 10 +- pkg/sentry/BUILD | 3 + pkg/sentry/fs/BUILD | 2 +- pkg/sentry/fs/fdpipe/pipe_opener_test.go | 1 + pkg/sentry/fs/overlay.go | 4 +- pkg/sentry/fsimpl/memfs/BUILD | 3 +- pkg/sentry/kernel/BUILD | 4 +- pkg/sentry/kernel/auth/BUILD | 2 +- pkg/sentry/kernel/futex/BUILD | 2 +- pkg/sentry/kernel/signalfd/BUILD | 4 +- pkg/sentry/kernel/task.go | 4 +- pkg/sentry/mm/BUILD | 2 +- pkg/sentry/mm/mm.go | 6 +- pkg/sentry/strace/strace.proto | 3 +- pkg/sentry/time/BUILD | 4 +- pkg/sentry/vfs/BUILD | 2 +- pkg/sentry/vfs/mount_unsafe.go | 4 +- pkg/state/object.proto | 56 ++++---- pkg/syncutil/BUILD | 54 ++++++++ pkg/syncutil/LICENSE | 27 ++++ pkg/syncutil/README.md | 5 + pkg/syncutil/atomicptr_unsafe.go | 47 +++++++ pkg/syncutil/atomicptrtest/BUILD | 29 ++++ pkg/syncutil/atomicptrtest/atomicptr_test.go | 31 +++++ pkg/syncutil/downgradable_rwmutex_1_12_unsafe.go | 21 +++ pkg/syncutil/downgradable_rwmutex_1_13_unsafe.go | 16 +++ pkg/syncutil/downgradable_rwmutex_test.go | 150 ++++++++++++++++++++ pkg/syncutil/downgradable_rwmutex_unsafe.go | 143 +++++++++++++++++++ pkg/syncutil/memmove_unsafe.go | 28 ++++ pkg/syncutil/norace_unsafe.go | 35 +++++ pkg/syncutil/race_unsafe.go | 41 ++++++ pkg/syncutil/seqatomic_unsafe.go | 72 ++++++++++ pkg/syncutil/seqatomictest/BUILD | 35 +++++ pkg/syncutil/seqatomictest/seqatomic_test.go | 132 ++++++++++++++++++ pkg/syncutil/seqcount.go | 149 ++++++++++++++++++++ pkg/syncutil/seqcount_test.go | 153 +++++++++++++++++++++ pkg/syncutil/syncutil.go | 7 + test/syscalls/linux/accept_bind.cc | 2 + test/syscalls/linux/accept_bind_stream.cc | 2 + test/syscalls/linux/chmod.cc | 1 + test/syscalls/linux/chroot.cc | 1 + test/syscalls/linux/clock_gettime.cc | 1 + test/syscalls/linux/concurrency.cc | 1 + test/syscalls/linux/exec_binary.cc | 1 + test/syscalls/linux/file_base.h | 1 + test/syscalls/linux/flock.cc | 1 + test/syscalls/linux/fork.cc | 1 + test/syscalls/linux/getdents.cc | 1 + test/syscalls/linux/ip_socket_test_util.cc | 5 +- test/syscalls/linux/memory_accounting.cc | 1 + test/syscalls/linux/mlock.cc | 1 + test/syscalls/linux/mmap.cc | 1 + test/syscalls/linux/mount.cc | 1 + test/syscalls/linux/read.cc | 1 + test/syscalls/linux/rename.cc | 1 + test/syscalls/linux/seccomp.cc | 1 + test/syscalls/linux/select.cc | 1 + test/syscalls/linux/shm.cc | 1 - test/syscalls/linux/socket_blocking.cc | 1 + test/syscalls/linux/socket_ip_loopback_blocking.cc | 1 + .../linux/socket_ip_tcp_generic_loopback.cc | 1 + .../linux/socket_ip_tcp_loopback_blocking.cc | 1 + .../linux/socket_ip_tcp_loopback_nonblock.cc | 1 + .../socket_ipv4_tcp_unbound_external_networking.cc | 1 + ...et_ipv4_tcp_unbound_external_networking_test.cc | 3 +- ...et_ipv4_udp_unbound_external_networking_test.cc | 3 +- test/syscalls/linux/socket_netlink_util.cc | 4 +- test/syscalls/linux/socket_unix_blocking_local.cc | 3 +- test/syscalls/linux/socket_unix_dgram.cc | 1 + .../linux/socket_unix_dgram_non_blocking.cc | 1 + .../linux/socket_unix_non_stream_blocking_local.cc | 3 +- test/syscalls/linux/socket_unix_seqpacket.cc | 1 + .../linux/socket_unix_stream_blocking_local.cc | 3 +- .../linux/socket_unix_stream_nonblock_local.cc | 3 +- .../syscalls/linux/socket_unix_unbound_abstract.cc | 1 + .../linux/socket_unix_unbound_filesystem.cc | 1 + .../linux/socket_unix_unbound_seqpacket.cc | 1 + test/syscalls/linux/socket_unix_unbound_stream.cc | 1 + test/syscalls/linux/sync.cc | 3 +- test/syscalls/linux/truncate.cc | 1 + .../syscalls/linux/unix_domain_socket_test_util.cc | 1 + test/syscalls/linux/unix_domain_socket_test_util.h | 1 + test/syscalls/linux/utimes.cc | 1 + test/syscalls/linux/vdso_clock_gettime.cc | 1 + test/util/fs_util_test.cc | 4 +- test/util/mount_util.h | 1 + test/util/posix_error_test.cc | 1 + test/util/rlimit_util.cc | 1 + test/util/signal_util.cc | 1 + test/util/signal_util.h | 1 + test/util/temp_path.h | 1 + test/util/test_util_test.cc | 1 + third_party/gvsync/BUILD | 53 ------- third_party/gvsync/LICENSE | 27 ---- third_party/gvsync/README.md | 3 - third_party/gvsync/atomicptr_unsafe.go | 47 ------- third_party/gvsync/atomicptrtest/BUILD | 28 ---- third_party/gvsync/atomicptrtest/atomicptr_test.go | 31 ----- .../gvsync/downgradable_rwmutex_1_12_unsafe.go | 21 --- .../gvsync/downgradable_rwmutex_1_13_unsafe.go | 16 --- third_party/gvsync/downgradable_rwmutex_test.go | 150 -------------------- third_party/gvsync/downgradable_rwmutex_unsafe.go | 143 ------------------- third_party/gvsync/gvsync.go | 7 - third_party/gvsync/memmove_unsafe.go | 28 ---- third_party/gvsync/norace_unsafe.go | 35 ----- third_party/gvsync/race_unsafe.go | 41 ------ third_party/gvsync/seqatomic_unsafe.go | 72 ---------- third_party/gvsync/seqatomictest/BUILD | 34 ----- third_party/gvsync/seqatomictest/seqatomic_test.go | 132 ------------------ third_party/gvsync/seqcount.go | 149 -------------------- third_party/gvsync/seqcount_test.go | 153 --------------------- tools/go_marshal/test/BUILD | 3 +- tools/go_marshal/test/external/BUILD | 4 +- 115 files changed, 1302 insertions(+), 1250 deletions(-) create mode 100644 pkg/syncutil/BUILD create mode 100644 pkg/syncutil/LICENSE create mode 100644 pkg/syncutil/README.md create mode 100644 pkg/syncutil/atomicptr_unsafe.go create mode 100644 pkg/syncutil/atomicptrtest/BUILD create mode 100644 pkg/syncutil/atomicptrtest/atomicptr_test.go create mode 100644 pkg/syncutil/downgradable_rwmutex_1_12_unsafe.go create mode 100644 pkg/syncutil/downgradable_rwmutex_1_13_unsafe.go create mode 100644 pkg/syncutil/downgradable_rwmutex_test.go create mode 100644 pkg/syncutil/downgradable_rwmutex_unsafe.go create mode 100644 pkg/syncutil/memmove_unsafe.go create mode 100644 pkg/syncutil/norace_unsafe.go create mode 100644 pkg/syncutil/race_unsafe.go create mode 100644 pkg/syncutil/seqatomic_unsafe.go create mode 100644 pkg/syncutil/seqatomictest/BUILD create mode 100644 pkg/syncutil/seqatomictest/seqatomic_test.go create mode 100644 pkg/syncutil/seqcount.go create mode 100644 pkg/syncutil/seqcount_test.go create mode 100644 pkg/syncutil/syncutil.go delete mode 100644 third_party/gvsync/BUILD delete mode 100644 third_party/gvsync/LICENSE delete mode 100644 third_party/gvsync/README.md delete mode 100644 third_party/gvsync/atomicptr_unsafe.go delete mode 100644 third_party/gvsync/atomicptrtest/BUILD delete mode 100644 third_party/gvsync/atomicptrtest/atomicptr_test.go delete mode 100644 third_party/gvsync/downgradable_rwmutex_1_12_unsafe.go delete mode 100644 third_party/gvsync/downgradable_rwmutex_1_13_unsafe.go delete mode 100644 third_party/gvsync/downgradable_rwmutex_test.go delete mode 100644 third_party/gvsync/downgradable_rwmutex_unsafe.go delete mode 100644 third_party/gvsync/gvsync.go delete mode 100644 third_party/gvsync/memmove_unsafe.go delete mode 100644 third_party/gvsync/norace_unsafe.go delete mode 100644 third_party/gvsync/race_unsafe.go delete mode 100644 third_party/gvsync/seqatomic_unsafe.go delete mode 100644 third_party/gvsync/seqatomictest/BUILD delete mode 100644 third_party/gvsync/seqatomictest/seqatomic_test.go delete mode 100644 third_party/gvsync/seqcount.go delete mode 100644 third_party/gvsync/seqcount_test.go (limited to 'pkg/sentry') diff --git a/pkg/eventchannel/BUILD b/pkg/eventchannel/BUILD index 71f2abc83..0b4b7cc44 100644 --- a/pkg/eventchannel/BUILD +++ b/pkg/eventchannel/BUILD @@ -25,6 +25,7 @@ go_library( proto_library( name = "eventchannel_proto", srcs = ["event.proto"], + visibility = ["//:sandbox"], ) go_proto_library( diff --git a/pkg/flipcall/BUILD b/pkg/flipcall/BUILD index 5643d5f26..e590a71ba 100644 --- a/pkg/flipcall/BUILD +++ b/pkg/flipcall/BUILD @@ -19,7 +19,7 @@ go_library( "//pkg/abi/linux", "//pkg/log", "//pkg/memutil", - "//third_party/gvsync", + "//pkg/syncutil", ], ) diff --git a/pkg/flipcall/flipcall_unsafe.go b/pkg/flipcall/flipcall_unsafe.go index a37952637..27b8939fc 100644 --- a/pkg/flipcall/flipcall_unsafe.go +++ b/pkg/flipcall/flipcall_unsafe.go @@ -18,7 +18,7 @@ import ( "reflect" "unsafe" - "gvisor.dev/gvisor/third_party/gvsync" + "gvisor.dev/gvisor/pkg/syncutil" ) // Packets consist of a 16-byte header followed by an arbitrarily-sized @@ -75,13 +75,13 @@ func (ep *Endpoint) Data() []byte { var ioSync int64 func raceBecomeActive() { - if gvsync.RaceEnabled { - gvsync.RaceAcquire((unsafe.Pointer)(&ioSync)) + if syncutil.RaceEnabled { + syncutil.RaceAcquire((unsafe.Pointer)(&ioSync)) } } func raceBecomeInactive() { - if gvsync.RaceEnabled { - gvsync.RaceReleaseMerge((unsafe.Pointer)(&ioSync)) + if syncutil.RaceEnabled { + syncutil.RaceReleaseMerge((unsafe.Pointer)(&ioSync)) } } diff --git a/pkg/sentry/BUILD b/pkg/sentry/BUILD index 2d6379c86..2a7122957 100644 --- a/pkg/sentry/BUILD +++ b/pkg/sentry/BUILD @@ -10,5 +10,8 @@ package_group( "//runsc/...", # Code generated by go_marshal relies on go_marshal libraries. "//tools/go_marshal/...", + + # Keep the old paths as a temporary measure. + "//third_party/golang/gvisor/pkg/sentry/...", ], ) diff --git a/pkg/sentry/fs/BUILD b/pkg/sentry/fs/BUILD index 378602cc9..c035ffff7 100644 --- a/pkg/sentry/fs/BUILD +++ b/pkg/sentry/fs/BUILD @@ -68,9 +68,9 @@ go_library( "//pkg/sentry/usage", "//pkg/sentry/usermem", "//pkg/state", + "//pkg/syncutil", "//pkg/syserror", "//pkg/waiter", - "//third_party/gvsync", ], ) diff --git a/pkg/sentry/fs/fdpipe/pipe_opener_test.go b/pkg/sentry/fs/fdpipe/pipe_opener_test.go index 8e4d839e1..577445148 100644 --- a/pkg/sentry/fs/fdpipe/pipe_opener_test.go +++ b/pkg/sentry/fs/fdpipe/pipe_opener_test.go @@ -25,6 +25,7 @@ import ( "time" "github.com/google/uuid" + "gvisor.dev/gvisor/pkg/fd" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/context/contexttest" diff --git a/pkg/sentry/fs/overlay.go b/pkg/sentry/fs/overlay.go index 1d3ff39e0..25573e986 100644 --- a/pkg/sentry/fs/overlay.go +++ b/pkg/sentry/fs/overlay.go @@ -23,8 +23,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/syncutil" "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/third_party/gvsync" ) // The virtual filesystem implements an overlay configuration. For a high-level @@ -199,7 +199,7 @@ type overlayEntry struct { upper *Inode // dirCacheMu protects dirCache. - dirCacheMu gvsync.DowngradableRWMutex `state:"nosave"` + dirCacheMu syncutil.DowngradableRWMutex `state:"nosave"` // dirCache is cache of DentAttrs from upper and lower Inodes. dirCache *SortedDentryMap diff --git a/pkg/sentry/fsimpl/memfs/BUILD b/pkg/sentry/fsimpl/memfs/BUILD index 04d667273..952b20c51 100644 --- a/pkg/sentry/fsimpl/memfs/BUILD +++ b/pkg/sentry/fsimpl/memfs/BUILD @@ -1,10 +1,9 @@ load("//tools/go_stateify:defs.bzl", "go_library") load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools/go_generics:defs.bzl", "go_template_instance") package(licenses = ["notice"]) -load("//tools/go_generics:defs.bzl", "go_template_instance") - go_template_instance( name = "dentry_list", out = "dentry_list.go", diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD index e041c51b3..2706927ff 100644 --- a/pkg/sentry/kernel/BUILD +++ b/pkg/sentry/kernel/BUILD @@ -35,7 +35,7 @@ go_template_instance( out = "seqatomic_taskgoroutineschedinfo_unsafe.go", package = "kernel", suffix = "TaskGoroutineSchedInfo", - template = "//third_party/gvsync:generic_seqatomic", + template = "//pkg/syncutil:generic_seqatomic", types = { "Value": "TaskGoroutineSchedInfo", }, @@ -209,12 +209,12 @@ go_library( "//pkg/sentry/usermem", "//pkg/state", "//pkg/state/statefile", + "//pkg/syncutil", "//pkg/syserr", "//pkg/syserror", "//pkg/tcpip", "//pkg/tcpip/stack", "//pkg/waiter", - "//third_party/gvsync", ], ) diff --git a/pkg/sentry/kernel/auth/BUILD b/pkg/sentry/kernel/auth/BUILD index 51de4568a..04c244447 100644 --- a/pkg/sentry/kernel/auth/BUILD +++ b/pkg/sentry/kernel/auth/BUILD @@ -8,7 +8,7 @@ go_template_instance( out = "atomicptr_credentials_unsafe.go", package = "auth", suffix = "Credentials", - template = "//third_party/gvsync:generic_atomicptr", + template = "//pkg/syncutil:generic_atomicptr", types = { "Value": "Credentials", }, diff --git a/pkg/sentry/kernel/futex/BUILD b/pkg/sentry/kernel/futex/BUILD index 34286c7a8..75ec31761 100644 --- a/pkg/sentry/kernel/futex/BUILD +++ b/pkg/sentry/kernel/futex/BUILD @@ -9,7 +9,7 @@ go_template_instance( out = "atomicptr_bucket_unsafe.go", package = "futex", suffix = "Bucket", - template = "//third_party/gvsync:generic_atomicptr", + template = "//pkg/syncutil:generic_atomicptr", types = { "Value": "bucket", }, diff --git a/pkg/sentry/kernel/signalfd/BUILD b/pkg/sentry/kernel/signalfd/BUILD index 50b69d154..9f7e19b4d 100644 --- a/pkg/sentry/kernel/signalfd/BUILD +++ b/pkg/sentry/kernel/signalfd/BUILD @@ -1,7 +1,7 @@ -package(licenses = ["notice"]) - load("//tools/go_stateify:defs.bzl", "go_library") +package(licenses = ["notice"]) + go_library( name = "signalfd", srcs = ["signalfd.go"], diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index 9be3dae3c..80c8e5464 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -35,8 +35,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/uniqueid" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/syncutil" "gvisor.dev/gvisor/pkg/waiter" - "gvisor.dev/gvisor/third_party/gvsync" ) // Task represents a thread of execution in the untrusted app. It @@ -83,7 +83,7 @@ type Task struct { // // gosched is protected by goschedSeq. gosched is owned by the task // goroutine. - goschedSeq gvsync.SeqCount `state:"nosave"` + goschedSeq syncutil.SeqCount `state:"nosave"` gosched TaskGoroutineSchedInfo // yieldCount is the number of times the task goroutine has called diff --git a/pkg/sentry/mm/BUILD b/pkg/sentry/mm/BUILD index a804b8b5c..839931f67 100644 --- a/pkg/sentry/mm/BUILD +++ b/pkg/sentry/mm/BUILD @@ -118,9 +118,9 @@ go_library( "//pkg/sentry/safemem", "//pkg/sentry/usage", "//pkg/sentry/usermem", + "//pkg/syncutil", "//pkg/syserror", "//pkg/tcpip/buffer", - "//third_party/gvsync", ], ) diff --git a/pkg/sentry/mm/mm.go b/pkg/sentry/mm/mm.go index f350e0109..58a5c186d 100644 --- a/pkg/sentry/mm/mm.go +++ b/pkg/sentry/mm/mm.go @@ -44,7 +44,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/usermem" - "gvisor.dev/gvisor/third_party/gvsync" + "gvisor.dev/gvisor/pkg/syncutil" ) // MemoryManager implements a virtual address space. @@ -82,7 +82,7 @@ type MemoryManager struct { users int32 // mappingMu is analogous to Linux's struct mm_struct::mmap_sem. - mappingMu gvsync.DowngradableRWMutex `state:"nosave"` + mappingMu syncutil.DowngradableRWMutex `state:"nosave"` // vmas stores virtual memory areas. Since vmas are stored by value, // clients should usually use vmaIterator.ValuePtr() instead of @@ -125,7 +125,7 @@ type MemoryManager struct { // activeMu is loosely analogous to Linux's struct // mm_struct::page_table_lock. - activeMu gvsync.DowngradableRWMutex `state:"nosave"` + activeMu syncutil.DowngradableRWMutex `state:"nosave"` // pmas stores platform mapping areas used to implement vmas. Since pmas // are stored by value, clients should usually use pmaIterator.ValuePtr() diff --git a/pkg/sentry/strace/strace.proto b/pkg/sentry/strace/strace.proto index 4b2f73a5f..906c52c51 100644 --- a/pkg/sentry/strace/strace.proto +++ b/pkg/sentry/strace/strace.proto @@ -32,8 +32,7 @@ message Strace { } } -message StraceEnter { -} +message StraceEnter {} message StraceExit { // Return value formatted as string. diff --git a/pkg/sentry/time/BUILD b/pkg/sentry/time/BUILD index d3a4cd943..18e212dff 100644 --- a/pkg/sentry/time/BUILD +++ b/pkg/sentry/time/BUILD @@ -9,7 +9,7 @@ go_template_instance( out = "seqatomic_parameters_unsafe.go", package = "time", suffix = "Parameters", - template = "//third_party/gvsync:generic_seqatomic", + template = "//pkg/syncutil:generic_seqatomic", types = { "Value": "Parameters", }, @@ -36,8 +36,8 @@ go_library( deps = [ "//pkg/log", "//pkg/metric", + "//pkg/syncutil", "//pkg/syserror", - "//third_party/gvsync", ], ) diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index 4f2c2de9f..74a325309 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -33,9 +33,9 @@ go_library( "//pkg/sentry/kernel/auth", "//pkg/sentry/memmap", "//pkg/sentry/usermem", + "//pkg/syncutil", "//pkg/syserror", "//pkg/waiter", - "//third_party/gvsync", ], ) diff --git a/pkg/sentry/vfs/mount_unsafe.go b/pkg/sentry/vfs/mount_unsafe.go index 75e6c7dfa..c98b42f91 100644 --- a/pkg/sentry/vfs/mount_unsafe.go +++ b/pkg/sentry/vfs/mount_unsafe.go @@ -26,7 +26,7 @@ import ( "sync/atomic" "unsafe" - "gvisor.dev/gvisor/third_party/gvsync" + "gvisor.dev/gvisor/pkg/syncutil" ) // mountKey represents the location at which a Mount is mounted. It is @@ -72,7 +72,7 @@ type mountTable struct { // intrinsics and inline assembly, limiting the performance of this // approach.) - seq gvsync.SeqCount + seq syncutil.SeqCount seed uint32 // for hashing keys // size holds both length (number of elements) and capacity (number of diff --git a/pkg/state/object.proto b/pkg/state/object.proto index 952289069..5ebcfb151 100644 --- a/pkg/state/object.proto +++ b/pkg/state/object.proto @@ -18,8 +18,8 @@ package gvisor.state.statefile; // Slice is a slice value. message Slice { - uint32 length = 1; - uint32 capacity = 2; + uint32 length = 1; + uint32 capacity = 2; uint64 ref_value = 3; } @@ -30,13 +30,13 @@ message Array { // Map is a map value. message Map { - repeated Object keys = 1; + repeated Object keys = 1; repeated Object values = 2; } // Interface is an interface value. message Interface { - string type = 1; + string type = 1; Object value = 2; } @@ -47,7 +47,7 @@ message Struct { // Field encodes a single field. message Field { - string name = 1; + string name = 1; Object value = 2; } @@ -113,28 +113,28 @@ message Float32s { // Note that ref_value references an Object.id, below. message Object { oneof value { - bool bool_value = 1; - bytes string_value = 2; - int64 int64_value = 3; - uint64 uint64_value = 4; - double double_value = 5; - uint64 ref_value = 6; - Slice slice_value = 7; - Array array_value = 8; - Interface interface_value = 9; - Struct struct_value = 10; - Map map_value = 11; - bytes byte_array_value = 12; - Uint16s uint16_array_value = 13; - Uint32s uint32_array_value = 14; - Uint64s uint64_array_value = 15; - Uintptrs uintptr_array_value = 16; - Int8s int8_array_value = 17; - Int16s int16_array_value = 18; - Int32s int32_array_value = 19; - Int64s int64_array_value = 20; - Bools bool_array_value = 21; - Float64s float64_array_value = 22; - Float32s float32_array_value = 23; + bool bool_value = 1; + bytes string_value = 2; + int64 int64_value = 3; + uint64 uint64_value = 4; + double double_value = 5; + uint64 ref_value = 6; + Slice slice_value = 7; + Array array_value = 8; + Interface interface_value = 9; + Struct struct_value = 10; + Map map_value = 11; + bytes byte_array_value = 12; + Uint16s uint16_array_value = 13; + Uint32s uint32_array_value = 14; + Uint64s uint64_array_value = 15; + Uintptrs uintptr_array_value = 16; + Int8s int8_array_value = 17; + Int16s int16_array_value = 18; + Int32s int32_array_value = 19; + Int64s int64_array_value = 20; + Bools bool_array_value = 21; + Float64s float64_array_value = 22; + Float32s float32_array_value = 23; } } diff --git a/pkg/syncutil/BUILD b/pkg/syncutil/BUILD new file mode 100644 index 000000000..b06a90bef --- /dev/null +++ b/pkg/syncutil/BUILD @@ -0,0 +1,54 @@ +load("//tools/go_stateify:defs.bzl", "go_library") +load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools/go_generics:defs.bzl", "go_template") + +package( + default_visibility = ["//:sandbox"], + licenses = ["notice"], +) + +exports_files(["LICENSE"]) + +go_template( + name = "generic_atomicptr", + srcs = ["atomicptr_unsafe.go"], + types = [ + "Value", + ], +) + +go_template( + name = "generic_seqatomic", + srcs = ["seqatomic_unsafe.go"], + types = [ + "Value", + ], + deps = [ + ":sync", + ], +) + +go_library( + name = "syncutil", + srcs = [ + "downgradable_rwmutex_1_12_unsafe.go", + "downgradable_rwmutex_1_13_unsafe.go", + "downgradable_rwmutex_unsafe.go", + "memmove_unsafe.go", + "norace_unsafe.go", + "race_unsafe.go", + "seqcount.go", + "syncutil.go", + ], + importpath = "gvisor.dev/gvisor/pkg/syncutil", +) + +go_test( + name = "syncutil_test", + size = "small", + srcs = [ + "downgradable_rwmutex_test.go", + "seqcount_test.go", + ], + embed = [":syncutil"], +) diff --git a/pkg/syncutil/LICENSE b/pkg/syncutil/LICENSE new file mode 100644 index 000000000..6a66aea5e --- /dev/null +++ b/pkg/syncutil/LICENSE @@ -0,0 +1,27 @@ +Copyright (c) 2009 The Go Authors. All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are +met: + + * Redistributions of source code must retain the above copyright +notice, this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above +copyright notice, this list of conditions and the following disclaimer +in the documentation and/or other materials provided with the +distribution. + * Neither the name of Google Inc. nor the names of its +contributors may be used to endorse or promote products derived from +this software without specific prior written permission. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/pkg/syncutil/README.md b/pkg/syncutil/README.md new file mode 100644 index 000000000..2183c4e20 --- /dev/null +++ b/pkg/syncutil/README.md @@ -0,0 +1,5 @@ +# Syncutil + +This package provides additional synchronization primitives not provided by the +Go stdlib 'sync' package. It is partially derived from the upstream 'sync' +package from go1.10. diff --git a/pkg/syncutil/atomicptr_unsafe.go b/pkg/syncutil/atomicptr_unsafe.go new file mode 100644 index 000000000..525c4beed --- /dev/null +++ b/pkg/syncutil/atomicptr_unsafe.go @@ -0,0 +1,47 @@ +// Copyright 2019 The gVisor Authors. +// +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// Package template doesn't exist. This file must be instantiated using the +// go_template_instance rule in tools/go_generics/defs.bzl. +package template + +import ( + "sync/atomic" + "unsafe" +) + +// Value is a required type parameter. +type Value struct{} + +// An AtomicPtr is a pointer to a value of type Value that can be atomically +// loaded and stored. The zero value of an AtomicPtr represents nil. +// +// Note that copying AtomicPtr by value performs a non-atomic read of the +// stored pointer, which is unsafe if Store() can be called concurrently; in +// this case, do `dst.Store(src.Load())` instead. +// +// +stateify savable +type AtomicPtr struct { + ptr unsafe.Pointer `state:".(*Value)"` +} + +func (p *AtomicPtr) savePtr() *Value { + return p.Load() +} + +func (p *AtomicPtr) loadPtr(v *Value) { + p.Store(v) +} + +// Load returns the value set by the most recent Store. It returns nil if there +// has been no previous call to Store. +func (p *AtomicPtr) Load() *Value { + return (*Value)(atomic.LoadPointer(&p.ptr)) +} + +// Store sets the value returned by Load to x. +func (p *AtomicPtr) Store(x *Value) { + atomic.StorePointer(&p.ptr, (unsafe.Pointer)(x)) +} diff --git a/pkg/syncutil/atomicptrtest/BUILD b/pkg/syncutil/atomicptrtest/BUILD new file mode 100644 index 000000000..63f411a90 --- /dev/null +++ b/pkg/syncutil/atomicptrtest/BUILD @@ -0,0 +1,29 @@ +load("//tools/go_stateify:defs.bzl", "go_library") +load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools/go_generics:defs.bzl", "go_template_instance") + +package(licenses = ["notice"]) + +go_template_instance( + name = "atomicptr_int", + out = "atomicptr_int_unsafe.go", + package = "atomicptr", + suffix = "Int", + template = "//pkg/syncutil:generic_atomicptr", + types = { + "Value": "int", + }, +) + +go_library( + name = "atomicptr", + srcs = ["atomicptr_int_unsafe.go"], + importpath = "gvisor.dev/gvisor/pkg/syncutil/atomicptr", +) + +go_test( + name = "atomicptr_test", + size = "small", + srcs = ["atomicptr_test.go"], + embed = [":atomicptr"], +) diff --git a/pkg/syncutil/atomicptrtest/atomicptr_test.go b/pkg/syncutil/atomicptrtest/atomicptr_test.go new file mode 100644 index 000000000..8fdc5112e --- /dev/null +++ b/pkg/syncutil/atomicptrtest/atomicptr_test.go @@ -0,0 +1,31 @@ +// Copyright 2019 The gVisor Authors. +// +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package atomicptr + +import ( + "testing" +) + +func newInt(val int) *int { + return &val +} + +func TestAtomicPtr(t *testing.T) { + var p AtomicPtrInt + if got := p.Load(); got != nil { + t.Errorf("initial value is %p (%v), wanted nil", got, got) + } + want := newInt(42) + p.Store(want) + if got := p.Load(); got != want { + t.Errorf("wrong value: got %p (%v), wanted %p (%v)", got, got, want, want) + } + want = newInt(100) + p.Store(want) + if got := p.Load(); got != want { + t.Errorf("wrong value: got %p (%v), wanted %p (%v)", got, got, want, want) + } +} diff --git a/pkg/syncutil/downgradable_rwmutex_1_12_unsafe.go b/pkg/syncutil/downgradable_rwmutex_1_12_unsafe.go new file mode 100644 index 000000000..7c6336e62 --- /dev/null +++ b/pkg/syncutil/downgradable_rwmutex_1_12_unsafe.go @@ -0,0 +1,21 @@ +// Copyright 2009 The Go Authors. All rights reserved. +// Copyright 2019 The gVisor Authors. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// +build go1.12 +// +build !go1.13 + +// TODO(b/133868570): Delete once Go 1.12 is no longer supported. + +package syncutil + +import _ "unsafe" + +//go:linkname runtimeSemrelease112 sync.runtime_Semrelease +func runtimeSemrelease112(s *uint32, handoff bool) + +func runtimeSemrelease(s *uint32, handoff bool, skipframes int) { + // 'skipframes' is only available starting from 1.13. + runtimeSemrelease112(s, handoff) +} diff --git a/pkg/syncutil/downgradable_rwmutex_1_13_unsafe.go b/pkg/syncutil/downgradable_rwmutex_1_13_unsafe.go new file mode 100644 index 000000000..3c3673119 --- /dev/null +++ b/pkg/syncutil/downgradable_rwmutex_1_13_unsafe.go @@ -0,0 +1,16 @@ +// Copyright 2009 The Go Authors. All rights reserved. +// Copyright 2019 The gVisor Authors. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// +build go1.13 +// +build !go1.15 + +// Check go:linkname function signatures when updating Go version. + +package syncutil + +import _ "unsafe" + +//go:linkname runtimeSemrelease sync.runtime_Semrelease +func runtimeSemrelease(s *uint32, handoff bool, skipframes int) diff --git a/pkg/syncutil/downgradable_rwmutex_test.go b/pkg/syncutil/downgradable_rwmutex_test.go new file mode 100644 index 000000000..ffaf7ecc7 --- /dev/null +++ b/pkg/syncutil/downgradable_rwmutex_test.go @@ -0,0 +1,150 @@ +// Copyright 2009 The Go Authors. All rights reserved. +// Copyright 2019 The gVisor Authors. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// GOMAXPROCS=10 go test + +// Copy/pasted from the standard library's sync/rwmutex_test.go, except for the +// addition of downgradingWriter and the renaming of num_iterations to +// numIterations to shut up Golint. + +package syncutil + +import ( + "fmt" + "runtime" + "sync/atomic" + "testing" +) + +func parallelReader(m *DowngradableRWMutex, clocked, cunlock, cdone chan bool) { + m.RLock() + clocked <- true + <-cunlock + m.RUnlock() + cdone <- true +} + +func doTestParallelReaders(numReaders, gomaxprocs int) { + runtime.GOMAXPROCS(gomaxprocs) + var m DowngradableRWMutex + clocked := make(chan bool) + cunlock := make(chan bool) + cdone := make(chan bool) + for i := 0; i < numReaders; i++ { + go parallelReader(&m, clocked, cunlock, cdone) + } + // Wait for all parallel RLock()s to succeed. + for i := 0; i < numReaders; i++ { + <-clocked + } + for i := 0; i < numReaders; i++ { + cunlock <- true + } + // Wait for the goroutines to finish. + for i := 0; i < numReaders; i++ { + <-cdone + } +} + +func TestParallelReaders(t *testing.T) { + defer runtime.GOMAXPROCS(runtime.GOMAXPROCS(-1)) + doTestParallelReaders(1, 4) + doTestParallelReaders(3, 4) + doTestParallelReaders(4, 2) +} + +func reader(rwm *DowngradableRWMutex, numIterations int, activity *int32, cdone chan bool) { + for i := 0; i < numIterations; i++ { + rwm.RLock() + n := atomic.AddInt32(activity, 1) + if n < 1 || n >= 10000 { + panic(fmt.Sprintf("wlock(%d)\n", n)) + } + for i := 0; i < 100; i++ { + } + atomic.AddInt32(activity, -1) + rwm.RUnlock() + } + cdone <- true +} + +func writer(rwm *DowngradableRWMutex, numIterations int, activity *int32, cdone chan bool) { + for i := 0; i < numIterations; i++ { + rwm.Lock() + n := atomic.AddInt32(activity, 10000) + if n != 10000 { + panic(fmt.Sprintf("wlock(%d)\n", n)) + } + for i := 0; i < 100; i++ { + } + atomic.AddInt32(activity, -10000) + rwm.Unlock() + } + cdone <- true +} + +func downgradingWriter(rwm *DowngradableRWMutex, numIterations int, activity *int32, cdone chan bool) { + for i := 0; i < numIterations; i++ { + rwm.Lock() + n := atomic.AddInt32(activity, 10000) + if n != 10000 { + panic(fmt.Sprintf("wlock(%d)\n", n)) + } + for i := 0; i < 100; i++ { + } + atomic.AddInt32(activity, -10000) + rwm.DowngradeLock() + n = atomic.AddInt32(activity, 1) + if n < 1 || n >= 10000 { + panic(fmt.Sprintf("wlock(%d)\n", n)) + } + for i := 0; i < 100; i++ { + } + n = atomic.AddInt32(activity, -1) + rwm.RUnlock() + } + cdone <- true +} + +func HammerDowngradableRWMutex(gomaxprocs, numReaders, numIterations int) { + runtime.GOMAXPROCS(gomaxprocs) + // Number of active readers + 10000 * number of active writers. + var activity int32 + var rwm DowngradableRWMutex + cdone := make(chan bool) + go writer(&rwm, numIterations, &activity, cdone) + go downgradingWriter(&rwm, numIterations, &activity, cdone) + var i int + for i = 0; i < numReaders/2; i++ { + go reader(&rwm, numIterations, &activity, cdone) + } + go writer(&rwm, numIterations, &activity, cdone) + go downgradingWriter(&rwm, numIterations, &activity, cdone) + for ; i < numReaders; i++ { + go reader(&rwm, numIterations, &activity, cdone) + } + // Wait for the 4 writers and all readers to finish. + for i := 0; i < 4+numReaders; i++ { + <-cdone + } +} + +func TestDowngradableRWMutex(t *testing.T) { + defer runtime.GOMAXPROCS(runtime.GOMAXPROCS(-1)) + n := 1000 + if testing.Short() { + n = 5 + } + HammerDowngradableRWMutex(1, 1, n) + HammerDowngradableRWMutex(1, 3, n) + HammerDowngradableRWMutex(1, 10, n) + HammerDowngradableRWMutex(4, 1, n) + HammerDowngradableRWMutex(4, 3, n) + HammerDowngradableRWMutex(4, 10, n) + HammerDowngradableRWMutex(10, 1, n) + HammerDowngradableRWMutex(10, 3, n) + HammerDowngradableRWMutex(10, 10, n) + HammerDowngradableRWMutex(10, 5, n) +} diff --git a/pkg/syncutil/downgradable_rwmutex_unsafe.go b/pkg/syncutil/downgradable_rwmutex_unsafe.go new file mode 100644 index 000000000..07feca402 --- /dev/null +++ b/pkg/syncutil/downgradable_rwmutex_unsafe.go @@ -0,0 +1,143 @@ +// Copyright 2009 The Go Authors. All rights reserved. +// Copyright 2019 The gVisor Authors. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// +build go1.12 +// +build !go1.15 + +// Check go:linkname function signatures when updating Go version. + +// This is mostly copied from the standard library's sync/rwmutex.go. +// +// Happens-before relationships indicated to the race detector: +// - Unlock -> Lock (via writerSem) +// - Unlock -> RLock (via readerSem) +// - RUnlock -> Lock (via writerSem) +// - DowngradeLock -> RLock (via readerSem) + +package syncutil + +import ( + "sync" + "sync/atomic" + "unsafe" +) + +//go:linkname runtimeSemacquire sync.runtime_Semacquire +func runtimeSemacquire(s *uint32) + +// DowngradableRWMutex is identical to sync.RWMutex, but adds the DowngradeLock +// method. +type DowngradableRWMutex struct { + w sync.Mutex // held if there are pending writers + writerSem uint32 // semaphore for writers to wait for completing readers + readerSem uint32 // semaphore for readers to wait for completing writers + readerCount int32 // number of pending readers + readerWait int32 // number of departing readers +} + +const rwmutexMaxReaders = 1 << 30 + +// RLock locks rw for reading. +func (rw *DowngradableRWMutex) RLock() { + if RaceEnabled { + RaceDisable() + } + if atomic.AddInt32(&rw.readerCount, 1) < 0 { + // A writer is pending, wait for it. + runtimeSemacquire(&rw.readerSem) + } + if RaceEnabled { + RaceEnable() + RaceAcquire(unsafe.Pointer(&rw.readerSem)) + } +} + +// RUnlock undoes a single RLock call. +func (rw *DowngradableRWMutex) RUnlock() { + if RaceEnabled { + RaceReleaseMerge(unsafe.Pointer(&rw.writerSem)) + RaceDisable() + } + if r := atomic.AddInt32(&rw.readerCount, -1); r < 0 { + if r+1 == 0 || r+1 == -rwmutexMaxReaders { + panic("RUnlock of unlocked DowngradableRWMutex") + } + // A writer is pending. + if atomic.AddInt32(&rw.readerWait, -1) == 0 { + // The last reader unblocks the writer. + runtimeSemrelease(&rw.writerSem, false, 0) + } + } + if RaceEnabled { + RaceEnable() + } +} + +// Lock locks rw for writing. +func (rw *DowngradableRWMutex) Lock() { + if RaceEnabled { + RaceDisable() + } + // First, resolve competition with other writers. + rw.w.Lock() + // Announce to readers there is a pending writer. + r := atomic.AddInt32(&rw.readerCount, -rwmutexMaxReaders) + rwmutexMaxReaders + // Wait for active readers. + if r != 0 && atomic.AddInt32(&rw.readerWait, r) != 0 { + runtimeSemacquire(&rw.writerSem) + } + if RaceEnabled { + RaceEnable() + RaceAcquire(unsafe.Pointer(&rw.writerSem)) + } +} + +// Unlock unlocks rw for writing. +func (rw *DowngradableRWMutex) Unlock() { + if RaceEnabled { + RaceRelease(unsafe.Pointer(&rw.writerSem)) + RaceRelease(unsafe.Pointer(&rw.readerSem)) + RaceDisable() + } + // Announce to readers there is no active writer. + r := atomic.AddInt32(&rw.readerCount, rwmutexMaxReaders) + if r >= rwmutexMaxReaders { + panic("Unlock of unlocked DowngradableRWMutex") + } + // Unblock blocked readers, if any. + for i := 0; i < int(r); i++ { + runtimeSemrelease(&rw.readerSem, false, 0) + } + // Allow other writers to proceed. + rw.w.Unlock() + if RaceEnabled { + RaceEnable() + } +} + +// DowngradeLock atomically unlocks rw for writing and locks it for reading. +func (rw *DowngradableRWMutex) DowngradeLock() { + if RaceEnabled { + RaceRelease(unsafe.Pointer(&rw.readerSem)) + RaceDisable() + } + // Announce to readers there is no active writer and one additional reader. + r := atomic.AddInt32(&rw.readerCount, rwmutexMaxReaders+1) + if r >= rwmutexMaxReaders+1 { + panic("DowngradeLock of unlocked DowngradableRWMutex") + } + // Unblock blocked readers, if any. Note that this loop starts as 1 since r + // includes this goroutine. + for i := 1; i < int(r); i++ { + runtimeSemrelease(&rw.readerSem, false, 0) + } + // Allow other writers to proceed to rw.w.Lock(). Note that they will still + // block on rw.writerSem since at least this reader exists, such that + // DowngradeLock() is atomic with the previous write lock. + rw.w.Unlock() + if RaceEnabled { + RaceEnable() + } +} diff --git a/pkg/syncutil/memmove_unsafe.go b/pkg/syncutil/memmove_unsafe.go new file mode 100644 index 000000000..348675baa --- /dev/null +++ b/pkg/syncutil/memmove_unsafe.go @@ -0,0 +1,28 @@ +// Copyright 2019 The gVisor Authors. +// +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// +build go1.12 +// +build !go1.15 + +// Check go:linkname function signatures when updating Go version. + +package syncutil + +import ( + "unsafe" +) + +//go:linkname memmove runtime.memmove +//go:noescape +func memmove(to, from unsafe.Pointer, n uintptr) + +// Memmove is exported for SeqAtomicLoad/SeqAtomicTryLoad, which can't +// define it because go_generics can't update the go:linkname annotation. +// Furthermore, go:linkname silently doesn't work if the local name is exported +// (this is of course undocumented), which is why this indirection is +// necessary. +func Memmove(to, from unsafe.Pointer, n uintptr) { + memmove(to, from, n) +} diff --git a/pkg/syncutil/norace_unsafe.go b/pkg/syncutil/norace_unsafe.go new file mode 100644 index 000000000..0a0a9deda --- /dev/null +++ b/pkg/syncutil/norace_unsafe.go @@ -0,0 +1,35 @@ +// Copyright 2019 The gVisor Authors. +// +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// +build !race + +package syncutil + +import ( + "unsafe" +) + +// RaceEnabled is true if the Go data race detector is enabled. +const RaceEnabled = false + +// RaceDisable has the same semantics as runtime.RaceDisable. +func RaceDisable() { +} + +// RaceEnable has the same semantics as runtime.RaceEnable. +func RaceEnable() { +} + +// RaceAcquire has the same semantics as runtime.RaceAcquire. +func RaceAcquire(addr unsafe.Pointer) { +} + +// RaceRelease has the same semantics as runtime.RaceRelease. +func RaceRelease(addr unsafe.Pointer) { +} + +// RaceReleaseMerge has the same semantics as runtime.RaceReleaseMerge. +func RaceReleaseMerge(addr unsafe.Pointer) { +} diff --git a/pkg/syncutil/race_unsafe.go b/pkg/syncutil/race_unsafe.go new file mode 100644 index 000000000..206067ec1 --- /dev/null +++ b/pkg/syncutil/race_unsafe.go @@ -0,0 +1,41 @@ +// Copyright 2019 The gVisor Authors. +// +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// +build race + +package syncutil + +import ( + "runtime" + "unsafe" +) + +// RaceEnabled is true if the Go data race detector is enabled. +const RaceEnabled = true + +// RaceDisable has the same semantics as runtime.RaceDisable. +func RaceDisable() { + runtime.RaceDisable() +} + +// RaceEnable has the same semantics as runtime.RaceEnable. +func RaceEnable() { + runtime.RaceEnable() +} + +// RaceAcquire has the same semantics as runtime.RaceAcquire. +func RaceAcquire(addr unsafe.Pointer) { + runtime.RaceAcquire(addr) +} + +// RaceRelease has the same semantics as runtime.RaceRelease. +func RaceRelease(addr unsafe.Pointer) { + runtime.RaceRelease(addr) +} + +// RaceReleaseMerge has the same semantics as runtime.RaceReleaseMerge. +func RaceReleaseMerge(addr unsafe.Pointer) { + runtime.RaceReleaseMerge(addr) +} diff --git a/pkg/syncutil/seqatomic_unsafe.go b/pkg/syncutil/seqatomic_unsafe.go new file mode 100644 index 000000000..cb6d2eb22 --- /dev/null +++ b/pkg/syncutil/seqatomic_unsafe.go @@ -0,0 +1,72 @@ +// Copyright 2019 The gVisor Authors. +// +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// Package template doesn't exist. This file must be instantiated using the +// go_template_instance rule in tools/go_generics/defs.bzl. +package template + +import ( + "fmt" + "reflect" + "strings" + "unsafe" + + "gvisor.dev/gvisor/pkg/syncutil" +) + +// Value is a required type parameter. +// +// Value must not contain any pointers, including interface objects, function +// objects, slices, maps, channels, unsafe.Pointer, and arrays or structs +// containing any of the above. An init() function will panic if this property +// does not hold. +type Value struct{} + +// SeqAtomicLoad returns a copy of *ptr, ensuring that the read does not race +// with any writer critical sections in sc. +func SeqAtomicLoad(sc *syncutil.SeqCount, ptr *Value) Value { + // This function doesn't use SeqAtomicTryLoad because doing so is + // measurably, significantly (~20%) slower; Go is awful at inlining. + var val Value + for { + epoch := sc.BeginRead() + if syncutil.RaceEnabled { + // runtime.RaceDisable() doesn't actually stop the race detector, + // so it can't help us here. Instead, call runtime.memmove + // directly, which is not instrumented by the race detector. + syncutil.Memmove(unsafe.Pointer(&val), unsafe.Pointer(ptr), unsafe.Sizeof(val)) + } else { + // This is ~40% faster for short reads than going through memmove. + val = *ptr + } + if sc.ReadOk(epoch) { + break + } + } + return val +} + +// SeqAtomicTryLoad returns a copy of *ptr while in a reader critical section +// in sc initiated by a call to sc.BeginRead() that returned epoch. If the read +// would race with a writer critical section, SeqAtomicTryLoad returns +// (unspecified, false). +func SeqAtomicTryLoad(sc *syncutil.SeqCount, epoch syncutil.SeqCountEpoch, ptr *Value) (Value, bool) { + var val Value + if syncutil.RaceEnabled { + syncutil.Memmove(unsafe.Pointer(&val), unsafe.Pointer(ptr), unsafe.Sizeof(val)) + } else { + val = *ptr + } + return val, sc.ReadOk(epoch) +} + +func init() { + var val Value + typ := reflect.TypeOf(val) + name := typ.Name() + if ptrs := syncutil.PointersInType(typ, name); len(ptrs) != 0 { + panic(fmt.Sprintf("SeqAtomicLoad<%s> is invalid since values %s of type %s contain pointers:\n%s", typ, name, typ, strings.Join(ptrs, "\n"))) + } +} diff --git a/pkg/syncutil/seqatomictest/BUILD b/pkg/syncutil/seqatomictest/BUILD new file mode 100644 index 000000000..ba18f3238 --- /dev/null +++ b/pkg/syncutil/seqatomictest/BUILD @@ -0,0 +1,35 @@ +load("//tools/go_stateify:defs.bzl", "go_library") +load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools/go_generics:defs.bzl", "go_template_instance") + +package(licenses = ["notice"]) + +go_template_instance( + name = "seqatomic_int", + out = "seqatomic_int_unsafe.go", + package = "seqatomic", + suffix = "Int", + template = "//pkg/syncutil:generic_seqatomic", + types = { + "Value": "int", + }, +) + +go_library( + name = "seqatomic", + srcs = ["seqatomic_int_unsafe.go"], + importpath = "gvisor.dev/gvisor/pkg/syncutil/seqatomic", + deps = [ + "//pkg/syncutil", + ], +) + +go_test( + name = "seqatomic_test", + size = "small", + srcs = ["seqatomic_test.go"], + embed = [":seqatomic"], + deps = [ + "//pkg/syncutil", + ], +) diff --git a/pkg/syncutil/seqatomictest/seqatomic_test.go b/pkg/syncutil/seqatomictest/seqatomic_test.go new file mode 100644 index 000000000..b0db44999 --- /dev/null +++ b/pkg/syncutil/seqatomictest/seqatomic_test.go @@ -0,0 +1,132 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package seqatomic + +import ( + "sync/atomic" + "testing" + "time" + + "gvisor.dev/gvisor/pkg/syncutil" +) + +func TestSeqAtomicLoadUncontended(t *testing.T) { + var seq syncutil.SeqCount + const want = 1 + data := want + if got := SeqAtomicLoadInt(&seq, &data); got != want { + t.Errorf("SeqAtomicLoadInt: got %v, wanted %v", got, want) + } +} + +func TestSeqAtomicLoadAfterWrite(t *testing.T) { + var seq syncutil.SeqCount + var data int + const want = 1 + seq.BeginWrite() + data = want + seq.EndWrite() + if got := SeqAtomicLoadInt(&seq, &data); got != want { + t.Errorf("SeqAtomicLoadInt: got %v, wanted %v", got, want) + } +} + +func TestSeqAtomicLoadDuringWrite(t *testing.T) { + var seq syncutil.SeqCount + var data int + const want = 1 + seq.BeginWrite() + go func() { + time.Sleep(time.Second) + data = want + seq.EndWrite() + }() + if got := SeqAtomicLoadInt(&seq, &data); got != want { + t.Errorf("SeqAtomicLoadInt: got %v, wanted %v", got, want) + } +} + +func TestSeqAtomicTryLoadUncontended(t *testing.T) { + var seq syncutil.SeqCount + const want = 1 + data := want + epoch := seq.BeginRead() + if got, ok := SeqAtomicTryLoadInt(&seq, epoch, &data); !ok || got != want { + t.Errorf("SeqAtomicTryLoadInt: got (%v, %v), wanted (%v, true)", got, ok, want) + } +} + +func TestSeqAtomicTryLoadDuringWrite(t *testing.T) { + var seq syncutil.SeqCount + var data int + epoch := seq.BeginRead() + seq.BeginWrite() + if got, ok := SeqAtomicTryLoadInt(&seq, epoch, &data); ok { + t.Errorf("SeqAtomicTryLoadInt: got (%v, true), wanted (_, false)", got) + } + seq.EndWrite() +} + +func TestSeqAtomicTryLoadAfterWrite(t *testing.T) { + var seq syncutil.SeqCount + var data int + epoch := seq.BeginRead() + seq.BeginWrite() + seq.EndWrite() + if got, ok := SeqAtomicTryLoadInt(&seq, epoch, &data); ok { + t.Errorf("SeqAtomicTryLoadInt: got (%v, true), wanted (_, false)", got) + } +} + +func BenchmarkSeqAtomicLoadIntUncontended(b *testing.B) { + var seq syncutil.SeqCount + const want = 42 + data := want + b.RunParallel(func(pb *testing.PB) { + for pb.Next() { + if got := SeqAtomicLoadInt(&seq, &data); got != want { + b.Fatalf("SeqAtomicLoadInt: got %v, wanted %v", got, want) + } + } + }) +} + +func BenchmarkSeqAtomicTryLoadIntUncontended(b *testing.B) { + var seq syncutil.SeqCount + const want = 42 + data := want + b.RunParallel(func(pb *testing.PB) { + epoch := seq.BeginRead() + for pb.Next() { + if got, ok := SeqAtomicTryLoadInt(&seq, epoch, &data); !ok || got != want { + b.Fatalf("SeqAtomicTryLoadInt: got (%v, %v), wanted (%v, true)", got, ok, want) + } + } + }) +} + +// For comparison: +func BenchmarkAtomicValueLoadIntUncontended(b *testing.B) { + var a atomic.Value + const want = 42 + a.Store(int(want)) + b.RunParallel(func(pb *testing.PB) { + for pb.Next() { + if got := a.Load().(int); got != want { + b.Fatalf("atomic.Value.Load: got %v, wanted %v", got, want) + } + } + }) +} diff --git a/pkg/syncutil/seqcount.go b/pkg/syncutil/seqcount.go new file mode 100644 index 000000000..11d8dbfaa --- /dev/null +++ b/pkg/syncutil/seqcount.go @@ -0,0 +1,149 @@ +// Copyright 2019 The gVisor Authors. +// +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package syncutil + +import ( + "fmt" + "reflect" + "runtime" + "sync/atomic" +) + +// SeqCount is a synchronization primitive for optimistic reader/writer +// synchronization in cases where readers can work with stale data and +// therefore do not need to block writers. +// +// Compared to sync/atomic.Value: +// +// - Mutation of SeqCount-protected data does not require memory allocation, +// whereas atomic.Value generally does. This is a significant advantage when +// writes are common. +// +// - Atomic reads of SeqCount-protected data require copying. This is a +// disadvantage when atomic reads are common. +// +// - SeqCount may be more flexible: correct use of SeqCount.ReadOk allows other +// operations to be made atomic with reads of SeqCount-protected data. +// +// - SeqCount may be less flexible: as of this writing, SeqCount-protected data +// cannot include pointers. +// +// - SeqCount is more cumbersome to use; atomic reads of SeqCount-protected +// data require instantiating function templates using go_generics (see +// seqatomic.go). +type SeqCount struct { + // epoch is incremented by BeginWrite and EndWrite, such that epoch is odd + // if a writer critical section is active, and a read from data protected + // by this SeqCount is atomic iff epoch is the same even value before and + // after the read. + epoch uint32 +} + +// SeqCountEpoch tracks writer critical sections in a SeqCount. +type SeqCountEpoch struct { + val uint32 +} + +// We assume that: +// +// - All functions in sync/atomic that perform a memory read are at least a +// read fence: memory reads before calls to such functions cannot be reordered +// after the call, and memory reads after calls to such functions cannot be +// reordered before the call, even if those reads do not use sync/atomic. +// +// - All functions in sync/atomic that perform a memory write are at least a +// write fence: memory writes before calls to such functions cannot be +// reordered after the call, and memory writes after calls to such functions +// cannot be reordered before the call, even if those writes do not use +// sync/atomic. +// +// As of this writing, the Go memory model completely fails to describe +// sync/atomic, but these properties are implied by +// https://groups.google.com/forum/#!topic/golang-nuts/7EnEhM3U7B8. + +// BeginRead indicates the beginning of a reader critical section. Reader +// critical sections DO NOT BLOCK writer critical sections, so operations in a +// reader critical section MAY RACE with writer critical sections. Races are +// detected by ReadOk at the end of the reader critical section. Thus, the +// low-level structure of readers is generally: +// +// for { +// epoch := seq.BeginRead() +// // do something idempotent with seq-protected data +// if seq.ReadOk(epoch) { +// break +// } +// } +// +// However, since reader critical sections may race with writer critical +// sections, the Go race detector will (accurately) flag data races in readers +// using this pattern. Most users of SeqCount will need to use the +// SeqAtomicLoad function template in seqatomic.go. +func (s *SeqCount) BeginRead() SeqCountEpoch { + epoch := atomic.LoadUint32(&s.epoch) + for epoch&1 != 0 { + runtime.Gosched() + epoch = atomic.LoadUint32(&s.epoch) + } + return SeqCountEpoch{epoch} +} + +// ReadOk returns true if the reader critical section initiated by a previous +// call to BeginRead() that returned epoch did not race with any writer critical +// sections. +// +// ReadOk may be called any number of times during a reader critical section. +// Reader critical sections do not need to be explicitly terminated; the last +// call to ReadOk is implicitly the end of the reader critical section. +func (s *SeqCount) ReadOk(epoch SeqCountEpoch) bool { + return atomic.LoadUint32(&s.epoch) == epoch.val +} + +// BeginWrite indicates the beginning of a writer critical section. +// +// SeqCount does not support concurrent writer critical sections; clients with +// concurrent writers must synchronize them using e.g. sync.Mutex. +func (s *SeqCount) BeginWrite() { + if epoch := atomic.AddUint32(&s.epoch, 1); epoch&1 == 0 { + panic("SeqCount.BeginWrite during writer critical section") + } +} + +// EndWrite ends the effect of a preceding BeginWrite. +func (s *SeqCount) EndWrite() { + if epoch := atomic.AddUint32(&s.epoch, 1); epoch&1 != 0 { + panic("SeqCount.EndWrite outside writer critical section") + } +} + +// PointersInType returns a list of pointers reachable from values named +// valName of the given type. +// +// PointersInType is not exhaustive, but it is guaranteed that if typ contains +// at least one pointer, then PointersInTypeOf returns a non-empty list. +func PointersInType(typ reflect.Type, valName string) []string { + switch kind := typ.Kind(); kind { + case reflect.Bool, reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64, reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr, reflect.Float32, reflect.Float64, reflect.Complex64, reflect.Complex128: + return nil + + case reflect.Chan, reflect.Func, reflect.Interface, reflect.Map, reflect.Ptr, reflect.Slice, reflect.String, reflect.UnsafePointer: + return []string{valName} + + case reflect.Array: + return PointersInType(typ.Elem(), valName+"[]") + + case reflect.Struct: + var ptrs []string + for i, n := 0, typ.NumField(); i < n; i++ { + field := typ.Field(i) + ptrs = append(ptrs, PointersInType(field.Type, fmt.Sprintf("%s.%s", valName, field.Name))...) + } + return ptrs + + default: + return []string{fmt.Sprintf("%s (of type %s with unknown kind %s)", valName, typ, kind)} + } +} diff --git a/pkg/syncutil/seqcount_test.go b/pkg/syncutil/seqcount_test.go new file mode 100644 index 000000000..14d6aedea --- /dev/null +++ b/pkg/syncutil/seqcount_test.go @@ -0,0 +1,153 @@ +// Copyright 2019 The gVisor Authors. +// +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package syncutil + +import ( + "reflect" + "testing" + "time" +) + +func TestSeqCountWriteUncontended(t *testing.T) { + var seq SeqCount + seq.BeginWrite() + seq.EndWrite() +} + +func TestSeqCountReadUncontended(t *testing.T) { + var seq SeqCount + epoch := seq.BeginRead() + if !seq.ReadOk(epoch) { + t.Errorf("ReadOk: got false, wanted true") + } +} + +func TestSeqCountBeginReadAfterWrite(t *testing.T) { + var seq SeqCount + var data int32 + const want = 1 + seq.BeginWrite() + data = want + seq.EndWrite() + epoch := seq.BeginRead() + if data != want { + t.Errorf("Reader: got %v, wanted %v", data, want) + } + if !seq.ReadOk(epoch) { + t.Errorf("ReadOk: got false, wanted true") + } +} + +func TestSeqCountBeginReadDuringWrite(t *testing.T) { + var seq SeqCount + var data int + const want = 1 + seq.BeginWrite() + go func() { + time.Sleep(time.Second) + data = want + seq.EndWrite() + }() + epoch := seq.BeginRead() + if data != want { + t.Errorf("Reader: got %v, wanted %v", data, want) + } + if !seq.ReadOk(epoch) { + t.Errorf("ReadOk: got false, wanted true") + } +} + +func TestSeqCountReadOkAfterWrite(t *testing.T) { + var seq SeqCount + epoch := seq.BeginRead() + seq.BeginWrite() + seq.EndWrite() + if seq.ReadOk(epoch) { + t.Errorf("ReadOk: got true, wanted false") + } +} + +func TestSeqCountReadOkDuringWrite(t *testing.T) { + var seq SeqCount + epoch := seq.BeginRead() + seq.BeginWrite() + if seq.ReadOk(epoch) { + t.Errorf("ReadOk: got true, wanted false") + } + seq.EndWrite() +} + +func BenchmarkSeqCountWriteUncontended(b *testing.B) { + var seq SeqCount + for i := 0; i < b.N; i++ { + seq.BeginWrite() + seq.EndWrite() + } +} + +func BenchmarkSeqCountReadUncontended(b *testing.B) { + var seq SeqCount + b.RunParallel(func(pb *testing.PB) { + for pb.Next() { + epoch := seq.BeginRead() + if !seq.ReadOk(epoch) { + b.Fatalf("ReadOk: got false, wanted true") + } + } + }) +} + +func TestPointersInType(t *testing.T) { + for _, test := range []struct { + name string // used for both test and value name + val interface{} + ptrs []string + }{ + { + name: "EmptyStruct", + val: struct{}{}, + }, + { + name: "Int", + val: int(0), + }, + { + name: "MixedStruct", + val: struct { + b bool + I int + ExportedPtr *struct{} + unexportedPtr *struct{} + arr [2]int + ptrArr [2]*int + nestedStruct struct { + nestedNonptr int + nestedPtr *int + } + structArr [1]struct { + nonptr int + ptr *int + } + }{}, + ptrs: []string{ + "MixedStruct.ExportedPtr", + "MixedStruct.unexportedPtr", + "MixedStruct.ptrArr[]", + "MixedStruct.nestedStruct.nestedPtr", + "MixedStruct.structArr[].ptr", + }, + }, + } { + t.Run(test.name, func(t *testing.T) { + typ := reflect.TypeOf(test.val) + ptrs := PointersInType(typ, test.name) + t.Logf("Found pointers: %v", ptrs) + if (len(ptrs) != 0 || len(test.ptrs) != 0) && !reflect.DeepEqual(ptrs, test.ptrs) { + t.Errorf("Got %v, wanted %v", ptrs, test.ptrs) + } + }) + } +} diff --git a/pkg/syncutil/syncutil.go b/pkg/syncutil/syncutil.go new file mode 100644 index 000000000..66e750d06 --- /dev/null +++ b/pkg/syncutil/syncutil.go @@ -0,0 +1,7 @@ +// Copyright 2019 The gVisor Authors. +// +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// Package syncutil provides synchronization primitives. +package syncutil diff --git a/test/syscalls/linux/accept_bind.cc b/test/syscalls/linux/accept_bind.cc index 427c42ede..e08c578f0 100644 --- a/test/syscalls/linux/accept_bind.cc +++ b/test/syscalls/linux/accept_bind.cc @@ -14,8 +14,10 @@ #include #include + #include #include + #include "gtest/gtest.h" #include "test/syscalls/linux/socket_test_util.h" #include "test/syscalls/linux/unix_domain_socket_test_util.h" diff --git a/test/syscalls/linux/accept_bind_stream.cc b/test/syscalls/linux/accept_bind_stream.cc index 7bcd91e9e..4857f160b 100644 --- a/test/syscalls/linux/accept_bind_stream.cc +++ b/test/syscalls/linux/accept_bind_stream.cc @@ -14,8 +14,10 @@ #include #include + #include #include + #include "gtest/gtest.h" #include "test/syscalls/linux/socket_test_util.h" #include "test/syscalls/linux/unix_domain_socket_test_util.h" diff --git a/test/syscalls/linux/chmod.cc b/test/syscalls/linux/chmod.cc index 7e918b9b2..a06b5cfd6 100644 --- a/test/syscalls/linux/chmod.cc +++ b/test/syscalls/linux/chmod.cc @@ -16,6 +16,7 @@ #include #include #include + #include #include "gtest/gtest.h" diff --git a/test/syscalls/linux/chroot.cc b/test/syscalls/linux/chroot.cc index de1611c21..04bc2d7b9 100644 --- a/test/syscalls/linux/chroot.cc +++ b/test/syscalls/linux/chroot.cc @@ -19,6 +19,7 @@ #include #include #include + #include #include diff --git a/test/syscalls/linux/clock_gettime.cc b/test/syscalls/linux/clock_gettime.cc index c9e3ed6b2..2aa91691e 100644 --- a/test/syscalls/linux/clock_gettime.cc +++ b/test/syscalls/linux/clock_gettime.cc @@ -14,6 +14,7 @@ #include #include + #include #include #include diff --git a/test/syscalls/linux/concurrency.cc b/test/syscalls/linux/concurrency.cc index 4e0a13f8b..00b96b34a 100644 --- a/test/syscalls/linux/concurrency.cc +++ b/test/syscalls/linux/concurrency.cc @@ -13,6 +13,7 @@ // limitations under the License. #include + #include #include "gtest/gtest.h" diff --git a/test/syscalls/linux/exec_binary.cc b/test/syscalls/linux/exec_binary.cc index 0a3931e5a..736452b0c 100644 --- a/test/syscalls/linux/exec_binary.cc +++ b/test/syscalls/linux/exec_binary.cc @@ -20,6 +20,7 @@ #include #include #include + #include #include #include diff --git a/test/syscalls/linux/file_base.h b/test/syscalls/linux/file_base.h index 4d155b618..4e048320e 100644 --- a/test/syscalls/linux/file_base.h +++ b/test/syscalls/linux/file_base.h @@ -27,6 +27,7 @@ #include #include #include + #include #include diff --git a/test/syscalls/linux/flock.cc b/test/syscalls/linux/flock.cc index b4a91455d..3ecb8db8e 100644 --- a/test/syscalls/linux/flock.cc +++ b/test/syscalls/linux/flock.cc @@ -14,6 +14,7 @@ #include #include + #include #include "gtest/gtest.h" diff --git a/test/syscalls/linux/fork.cc b/test/syscalls/linux/fork.cc index dd6e1a422..371890110 100644 --- a/test/syscalls/linux/fork.cc +++ b/test/syscalls/linux/fork.cc @@ -20,6 +20,7 @@ #include #include #include + #include #include diff --git a/test/syscalls/linux/getdents.cc b/test/syscalls/linux/getdents.cc index fe9cfafe8..ad2dbacb8 100644 --- a/test/syscalls/linux/getdents.cc +++ b/test/syscalls/linux/getdents.cc @@ -23,6 +23,7 @@ #include #include #include + #include #include #include diff --git a/test/syscalls/linux/ip_socket_test_util.cc b/test/syscalls/linux/ip_socket_test_util.cc index 57e99596f..8398fc95f 100644 --- a/test/syscalls/linux/ip_socket_test_util.cc +++ b/test/syscalls/linux/ip_socket_test_util.cc @@ -12,13 +12,14 @@ // See the License for the specific language governing permissions and // limitations under the License. +#include "test/syscalls/linux/ip_socket_test_util.h" + #include #include #include #include -#include -#include "test/syscalls/linux/ip_socket_test_util.h" +#include namespace gvisor { namespace testing { diff --git a/test/syscalls/linux/memory_accounting.cc b/test/syscalls/linux/memory_accounting.cc index ff2f49863..94aea4077 100644 --- a/test/syscalls/linux/memory_accounting.cc +++ b/test/syscalls/linux/memory_accounting.cc @@ -13,6 +13,7 @@ // limitations under the License. #include + #include #include "gtest/gtest.h" diff --git a/test/syscalls/linux/mlock.cc b/test/syscalls/linux/mlock.cc index 283c21ed3..620b4f8b4 100644 --- a/test/syscalls/linux/mlock.cc +++ b/test/syscalls/linux/mlock.cc @@ -16,6 +16,7 @@ #include #include #include + #include #include diff --git a/test/syscalls/linux/mmap.cc b/test/syscalls/linux/mmap.cc index a112316e9..6f2639d8a 100644 --- a/test/syscalls/linux/mmap.cc +++ b/test/syscalls/linux/mmap.cc @@ -28,6 +28,7 @@ #include #include #include + #include #include "gmock/gmock.h" diff --git a/test/syscalls/linux/mount.cc b/test/syscalls/linux/mount.cc index e35be3cab..a3e9745cf 100644 --- a/test/syscalls/linux/mount.cc +++ b/test/syscalls/linux/mount.cc @@ -18,6 +18,7 @@ #include #include #include + #include #include #include diff --git a/test/syscalls/linux/read.cc b/test/syscalls/linux/read.cc index 4430fa3c2..2633ba31b 100644 --- a/test/syscalls/linux/read.cc +++ b/test/syscalls/linux/read.cc @@ -14,6 +14,7 @@ #include #include + #include #include "gtest/gtest.h" diff --git a/test/syscalls/linux/rename.cc b/test/syscalls/linux/rename.cc index 5b474ff32..833c0dc4f 100644 --- a/test/syscalls/linux/rename.cc +++ b/test/syscalls/linux/rename.cc @@ -14,6 +14,7 @@ #include #include + #include #include "gtest/gtest.h" diff --git a/test/syscalls/linux/seccomp.cc b/test/syscalls/linux/seccomp.cc index e77586852..7e41fe7d8 100644 --- a/test/syscalls/linux/seccomp.cc +++ b/test/syscalls/linux/seccomp.cc @@ -25,6 +25,7 @@ #include #include #include + #include #include "gmock/gmock.h" diff --git a/test/syscalls/linux/select.cc b/test/syscalls/linux/select.cc index e06a2666d..424e2a67f 100644 --- a/test/syscalls/linux/select.cc +++ b/test/syscalls/linux/select.cc @@ -16,6 +16,7 @@ #include #include #include + #include #include #include diff --git a/test/syscalls/linux/shm.cc b/test/syscalls/linux/shm.cc index eb7a3966f..7ba752599 100644 --- a/test/syscalls/linux/shm.cc +++ b/test/syscalls/linux/shm.cc @@ -13,7 +13,6 @@ // limitations under the License. #include - #include #include #include diff --git a/test/syscalls/linux/socket_blocking.cc b/test/syscalls/linux/socket_blocking.cc index d7ce57566..7e88aa2d9 100644 --- a/test/syscalls/linux/socket_blocking.cc +++ b/test/syscalls/linux/socket_blocking.cc @@ -17,6 +17,7 @@ #include #include #include + #include #include "gtest/gtest.h" diff --git a/test/syscalls/linux/socket_ip_loopback_blocking.cc b/test/syscalls/linux/socket_ip_loopback_blocking.cc index d7fc9715b..e58eedaba 100644 --- a/test/syscalls/linux/socket_ip_loopback_blocking.cc +++ b/test/syscalls/linux/socket_ip_loopback_blocking.cc @@ -13,6 +13,7 @@ // limitations under the License. #include + #include #include "test/syscalls/linux/ip_socket_test_util.h" diff --git a/test/syscalls/linux/socket_ip_tcp_generic_loopback.cc b/test/syscalls/linux/socket_ip_tcp_generic_loopback.cc index 0dc274e2d..d11f7cc23 100644 --- a/test/syscalls/linux/socket_ip_tcp_generic_loopback.cc +++ b/test/syscalls/linux/socket_ip_tcp_generic_loopback.cc @@ -13,6 +13,7 @@ // limitations under the License. #include + #include #include "test/syscalls/linux/ip_socket_test_util.h" diff --git a/test/syscalls/linux/socket_ip_tcp_loopback_blocking.cc b/test/syscalls/linux/socket_ip_tcp_loopback_blocking.cc index cd3ad97d0..fcd20102f 100644 --- a/test/syscalls/linux/socket_ip_tcp_loopback_blocking.cc +++ b/test/syscalls/linux/socket_ip_tcp_loopback_blocking.cc @@ -13,6 +13,7 @@ // limitations under the License. #include + #include #include "test/syscalls/linux/ip_socket_test_util.h" diff --git a/test/syscalls/linux/socket_ip_tcp_loopback_nonblock.cc b/test/syscalls/linux/socket_ip_tcp_loopback_nonblock.cc index 1acdecc17..63a05b799 100644 --- a/test/syscalls/linux/socket_ip_tcp_loopback_nonblock.cc +++ b/test/syscalls/linux/socket_ip_tcp_loopback_nonblock.cc @@ -13,6 +13,7 @@ // limitations under the License. #include + #include #include "test/syscalls/linux/ip_socket_test_util.h" diff --git a/test/syscalls/linux/socket_ipv4_tcp_unbound_external_networking.cc b/test/syscalls/linux/socket_ipv4_tcp_unbound_external_networking.cc index 3c3712b50..80f12b0a9 100644 --- a/test/syscalls/linux/socket_ipv4_tcp_unbound_external_networking.cc +++ b/test/syscalls/linux/socket_ipv4_tcp_unbound_external_networking.cc @@ -18,6 +18,7 @@ #include #include #include + #include #include diff --git a/test/syscalls/linux/socket_ipv4_tcp_unbound_external_networking_test.cc b/test/syscalls/linux/socket_ipv4_tcp_unbound_external_networking_test.cc index 92f03e045..3ac790873 100644 --- a/test/syscalls/linux/socket_ipv4_tcp_unbound_external_networking_test.cc +++ b/test/syscalls/linux/socket_ipv4_tcp_unbound_external_networking_test.cc @@ -12,10 +12,11 @@ // See the License for the specific language governing permissions and // limitations under the License. +#include "test/syscalls/linux/socket_ipv4_tcp_unbound_external_networking.h" + #include #include "test/syscalls/linux/ip_socket_test_util.h" -#include "test/syscalls/linux/socket_ipv4_tcp_unbound_external_networking.h" #include "test/syscalls/linux/socket_test_util.h" #include "test/util/test_util.h" diff --git a/test/syscalls/linux/socket_ipv4_udp_unbound_external_networking_test.cc b/test/syscalls/linux/socket_ipv4_udp_unbound_external_networking_test.cc index 9d4e1ab97..8f47952b0 100644 --- a/test/syscalls/linux/socket_ipv4_udp_unbound_external_networking_test.cc +++ b/test/syscalls/linux/socket_ipv4_udp_unbound_external_networking_test.cc @@ -12,10 +12,11 @@ // See the License for the specific language governing permissions and // limitations under the License. +#include "test/syscalls/linux/socket_ipv4_udp_unbound_external_networking.h" + #include #include "test/syscalls/linux/ip_socket_test_util.h" -#include "test/syscalls/linux/socket_ipv4_udp_unbound_external_networking.h" #include "test/syscalls/linux/socket_test_util.h" #include "test/util/test_util.h" diff --git a/test/syscalls/linux/socket_netlink_util.cc b/test/syscalls/linux/socket_netlink_util.cc index 5f05bab10..723f5d728 100644 --- a/test/syscalls/linux/socket_netlink_util.cc +++ b/test/syscalls/linux/socket_netlink_util.cc @@ -12,15 +12,15 @@ // See the License for the specific language governing permissions and // limitations under the License. -#include +#include "test/syscalls/linux/socket_netlink_util.h" #include #include +#include #include #include "absl/strings/str_cat.h" -#include "test/syscalls/linux/socket_netlink_util.h" #include "test/syscalls/linux/socket_test_util.h" namespace gvisor { diff --git a/test/syscalls/linux/socket_unix_blocking_local.cc b/test/syscalls/linux/socket_unix_blocking_local.cc index 1994139e6..6f84221b2 100644 --- a/test/syscalls/linux/socket_unix_blocking_local.cc +++ b/test/syscalls/linux/socket_unix_blocking_local.cc @@ -12,10 +12,9 @@ // See the License for the specific language governing permissions and // limitations under the License. -#include "test/syscalls/linux/socket_blocking.h" - #include +#include "test/syscalls/linux/socket_blocking.h" #include "test/syscalls/linux/socket_test_util.h" #include "test/syscalls/linux/unix_domain_socket_test_util.h" #include "test/util/test_util.h" diff --git a/test/syscalls/linux/socket_unix_dgram.cc b/test/syscalls/linux/socket_unix_dgram.cc index 3245cf7c9..af0df4fb4 100644 --- a/test/syscalls/linux/socket_unix_dgram.cc +++ b/test/syscalls/linux/socket_unix_dgram.cc @@ -16,6 +16,7 @@ #include #include + #include "gtest/gtest.h" #include "test/syscalls/linux/socket_test_util.h" #include "test/syscalls/linux/unix_domain_socket_test_util.h" diff --git a/test/syscalls/linux/socket_unix_dgram_non_blocking.cc b/test/syscalls/linux/socket_unix_dgram_non_blocking.cc index cd4fba25c..2db8b68d3 100644 --- a/test/syscalls/linux/socket_unix_dgram_non_blocking.cc +++ b/test/syscalls/linux/socket_unix_dgram_non_blocking.cc @@ -14,6 +14,7 @@ #include #include + #include "gtest/gtest.h" #include "test/syscalls/linux/socket_test_util.h" #include "test/syscalls/linux/unix_domain_socket_test_util.h" diff --git a/test/syscalls/linux/socket_unix_non_stream_blocking_local.cc b/test/syscalls/linux/socket_unix_non_stream_blocking_local.cc index da762cd83..8855d5001 100644 --- a/test/syscalls/linux/socket_unix_non_stream_blocking_local.cc +++ b/test/syscalls/linux/socket_unix_non_stream_blocking_local.cc @@ -12,10 +12,9 @@ // See the License for the specific language governing permissions and // limitations under the License. -#include "test/syscalls/linux/socket_non_stream_blocking.h" - #include +#include "test/syscalls/linux/socket_non_stream_blocking.h" #include "test/syscalls/linux/socket_test_util.h" #include "test/syscalls/linux/unix_domain_socket_test_util.h" #include "test/util/test_util.h" diff --git a/test/syscalls/linux/socket_unix_seqpacket.cc b/test/syscalls/linux/socket_unix_seqpacket.cc index 60fa9e38a..84d3a569e 100644 --- a/test/syscalls/linux/socket_unix_seqpacket.cc +++ b/test/syscalls/linux/socket_unix_seqpacket.cc @@ -16,6 +16,7 @@ #include #include + #include "gtest/gtest.h" #include "test/syscalls/linux/socket_test_util.h" #include "test/syscalls/linux/unix_domain_socket_test_util.h" diff --git a/test/syscalls/linux/socket_unix_stream_blocking_local.cc b/test/syscalls/linux/socket_unix_stream_blocking_local.cc index fa0a9d367..08e579ba7 100644 --- a/test/syscalls/linux/socket_unix_stream_blocking_local.cc +++ b/test/syscalls/linux/socket_unix_stream_blocking_local.cc @@ -12,10 +12,9 @@ // See the License for the specific language governing permissions and // limitations under the License. -#include "test/syscalls/linux/socket_stream_blocking.h" - #include +#include "test/syscalls/linux/socket_stream_blocking.h" #include "test/syscalls/linux/socket_test_util.h" #include "test/syscalls/linux/unix_domain_socket_test_util.h" #include "test/util/test_util.h" diff --git a/test/syscalls/linux/socket_unix_stream_nonblock_local.cc b/test/syscalls/linux/socket_unix_stream_nonblock_local.cc index ec777c59f..1936aa135 100644 --- a/test/syscalls/linux/socket_unix_stream_nonblock_local.cc +++ b/test/syscalls/linux/socket_unix_stream_nonblock_local.cc @@ -11,10 +11,9 @@ // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. // See the License for the specific language governing permissions and // limitations under the License. -#include "test/syscalls/linux/socket_stream_nonblock.h" - #include +#include "test/syscalls/linux/socket_stream_nonblock.h" #include "test/syscalls/linux/socket_test_util.h" #include "test/syscalls/linux/unix_domain_socket_test_util.h" #include "test/util/test_util.h" diff --git a/test/syscalls/linux/socket_unix_unbound_abstract.cc b/test/syscalls/linux/socket_unix_unbound_abstract.cc index 7f5816ace..8b1762000 100644 --- a/test/syscalls/linux/socket_unix_unbound_abstract.cc +++ b/test/syscalls/linux/socket_unix_unbound_abstract.cc @@ -14,6 +14,7 @@ #include #include + #include "gtest/gtest.h" #include "test/syscalls/linux/socket_test_util.h" #include "test/syscalls/linux/unix_domain_socket_test_util.h" diff --git a/test/syscalls/linux/socket_unix_unbound_filesystem.cc b/test/syscalls/linux/socket_unix_unbound_filesystem.cc index b14f24086..cab912152 100644 --- a/test/syscalls/linux/socket_unix_unbound_filesystem.cc +++ b/test/syscalls/linux/socket_unix_unbound_filesystem.cc @@ -14,6 +14,7 @@ #include #include + #include "gtest/gtest.h" #include "test/syscalls/linux/socket_test_util.h" #include "test/syscalls/linux/unix_domain_socket_test_util.h" diff --git a/test/syscalls/linux/socket_unix_unbound_seqpacket.cc b/test/syscalls/linux/socket_unix_unbound_seqpacket.cc index 50ffa1d04..cb99030f5 100644 --- a/test/syscalls/linux/socket_unix_unbound_seqpacket.cc +++ b/test/syscalls/linux/socket_unix_unbound_seqpacket.cc @@ -14,6 +14,7 @@ #include #include + #include "gtest/gtest.h" #include "test/syscalls/linux/socket_test_util.h" #include "test/syscalls/linux/unix_domain_socket_test_util.h" diff --git a/test/syscalls/linux/socket_unix_unbound_stream.cc b/test/syscalls/linux/socket_unix_unbound_stream.cc index 344918c34..f185dded3 100644 --- a/test/syscalls/linux/socket_unix_unbound_stream.cc +++ b/test/syscalls/linux/socket_unix_unbound_stream.cc @@ -14,6 +14,7 @@ #include #include + #include "gtest/gtest.h" #include "test/syscalls/linux/socket_test_util.h" #include "test/syscalls/linux/unix_domain_socket_test_util.h" diff --git a/test/syscalls/linux/sync.cc b/test/syscalls/linux/sync.cc index fe479390d..8aa2525a9 100644 --- a/test/syscalls/linux/sync.cc +++ b/test/syscalls/linux/sync.cc @@ -14,10 +14,9 @@ #include #include -#include - #include #include + #include #include "gtest/gtest.h" diff --git a/test/syscalls/linux/truncate.cc b/test/syscalls/linux/truncate.cc index e5cc5d97c..c988c6380 100644 --- a/test/syscalls/linux/truncate.cc +++ b/test/syscalls/linux/truncate.cc @@ -19,6 +19,7 @@ #include #include #include + #include #include diff --git a/test/syscalls/linux/unix_domain_socket_test_util.cc b/test/syscalls/linux/unix_domain_socket_test_util.cc index 7fb9eed8d..b05ab2900 100644 --- a/test/syscalls/linux/unix_domain_socket_test_util.cc +++ b/test/syscalls/linux/unix_domain_socket_test_util.cc @@ -15,6 +15,7 @@ #include "test/syscalls/linux/unix_domain_socket_test_util.h" #include + #include #include "gtest/gtest.h" diff --git a/test/syscalls/linux/unix_domain_socket_test_util.h b/test/syscalls/linux/unix_domain_socket_test_util.h index 5eca0b7f0..b8073db17 100644 --- a/test/syscalls/linux/unix_domain_socket_test_util.h +++ b/test/syscalls/linux/unix_domain_socket_test_util.h @@ -16,6 +16,7 @@ #define GVISOR_TEST_SYSCALLS_UNIX_DOMAIN_SOCKET_TEST_UTIL_H_ #include + #include "test/syscalls/linux/socket_test_util.h" namespace gvisor { diff --git a/test/syscalls/linux/utimes.cc b/test/syscalls/linux/utimes.cc index 80716859a..12b925a51 100644 --- a/test/syscalls/linux/utimes.cc +++ b/test/syscalls/linux/utimes.cc @@ -20,6 +20,7 @@ #include #include #include + #include #include "absl/time/time.h" diff --git a/test/syscalls/linux/vdso_clock_gettime.cc b/test/syscalls/linux/vdso_clock_gettime.cc index 40c0014b9..ce1899f45 100644 --- a/test/syscalls/linux/vdso_clock_gettime.cc +++ b/test/syscalls/linux/vdso_clock_gettime.cc @@ -17,6 +17,7 @@ #include #include #include + #include #include #include diff --git a/test/util/fs_util_test.cc b/test/util/fs_util_test.cc index 2a200320a..657b6a46e 100644 --- a/test/util/fs_util_test.cc +++ b/test/util/fs_util_test.cc @@ -12,12 +12,14 @@ // See the License for the specific language governing permissions and // limitations under the License. +#include "test/util/fs_util.h" + #include + #include #include "gmock/gmock.h" #include "gtest/gtest.h" -#include "test/util/fs_util.h" #include "test/util/posix_error.h" #include "test/util/temp_path.h" #include "test/util/test_util.h" diff --git a/test/util/mount_util.h b/test/util/mount_util.h index 38ec6c8a1..484de560e 100644 --- a/test/util/mount_util.h +++ b/test/util/mount_util.h @@ -17,6 +17,7 @@ #include #include + #include #include diff --git a/test/util/posix_error_test.cc b/test/util/posix_error_test.cc index d67270842..bf9465abb 100644 --- a/test/util/posix_error_test.cc +++ b/test/util/posix_error_test.cc @@ -15,6 +15,7 @@ #include "test/util/posix_error.h" #include + #include "gmock/gmock.h" #include "gtest/gtest.h" diff --git a/test/util/rlimit_util.cc b/test/util/rlimit_util.cc index 684253f78..d7bfc1606 100644 --- a/test/util/rlimit_util.cc +++ b/test/util/rlimit_util.cc @@ -15,6 +15,7 @@ #include "test/util/rlimit_util.h" #include + #include #include "test/util/cleanup.h" diff --git a/test/util/signal_util.cc b/test/util/signal_util.cc index 26738864f..5ee95ee80 100644 --- a/test/util/signal_util.cc +++ b/test/util/signal_util.cc @@ -15,6 +15,7 @@ #include "test/util/signal_util.h" #include + #include #include "gtest/gtest.h" diff --git a/test/util/signal_util.h b/test/util/signal_util.h index 7fd2af015..bcf85c337 100644 --- a/test/util/signal_util.h +++ b/test/util/signal_util.h @@ -18,6 +18,7 @@ #include #include #include + #include #include "gmock/gmock.h" diff --git a/test/util/temp_path.h b/test/util/temp_path.h index 92d669503..9e5ac11f4 100644 --- a/test/util/temp_path.h +++ b/test/util/temp_path.h @@ -16,6 +16,7 @@ #define GVISOR_TEST_UTIL_TEMP_PATH_H_ #include + #include #include diff --git a/test/util/test_util_test.cc b/test/util/test_util_test.cc index b7300d9e5..f42100374 100644 --- a/test/util/test_util_test.cc +++ b/test/util/test_util_test.cc @@ -15,6 +15,7 @@ #include "test/util/test_util.h" #include + #include #include "gmock/gmock.h" diff --git a/third_party/gvsync/BUILD b/third_party/gvsync/BUILD deleted file mode 100644 index 7d6d59c48..000000000 --- a/third_party/gvsync/BUILD +++ /dev/null @@ -1,53 +0,0 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library", "go_test") -load("//tools/go_generics:defs.bzl", "go_template") - -package( - default_visibility = ["//:sandbox"], - licenses = ["notice"], -) - -exports_files(["LICENSE"]) - -go_template( - name = "generic_atomicptr", - srcs = ["atomicptr_unsafe.go"], - types = [ - "Value", - ], -) - -go_template( - name = "generic_seqatomic", - srcs = ["seqatomic_unsafe.go"], - types = [ - "Value", - ], - deps = [ - ":sync", - ], -) - -go_library( - name = "gvsync", - srcs = [ - "downgradable_rwmutex_1_12_unsafe.go", - "downgradable_rwmutex_1_13_unsafe.go", - "downgradable_rwmutex_unsafe.go", - "gvsync.go", - "memmove_unsafe.go", - "norace_unsafe.go", - "race_unsafe.go", - "seqcount.go", - ], - importpath = "gvisor.dev/gvisor/third_party/gvsync", -) - -go_test( - name = "gvsync_test", - size = "small", - srcs = [ - "downgradable_rwmutex_test.go", - "seqcount_test.go", - ], - embed = [":gvsync"], -) diff --git a/third_party/gvsync/LICENSE b/third_party/gvsync/LICENSE deleted file mode 100644 index 6a66aea5e..000000000 --- a/third_party/gvsync/LICENSE +++ /dev/null @@ -1,27 +0,0 @@ -Copyright (c) 2009 The Go Authors. All rights reserved. - -Redistribution and use in source and binary forms, with or without -modification, are permitted provided that the following conditions are -met: - - * Redistributions of source code must retain the above copyright -notice, this list of conditions and the following disclaimer. - * Redistributions in binary form must reproduce the above -copyright notice, this list of conditions and the following disclaimer -in the documentation and/or other materials provided with the -distribution. - * Neither the name of Google Inc. nor the names of its -contributors may be used to endorse or promote products derived from -this software without specific prior written permission. - -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/third_party/gvsync/README.md b/third_party/gvsync/README.md deleted file mode 100644 index fcc7e6f44..000000000 --- a/third_party/gvsync/README.md +++ /dev/null @@ -1,3 +0,0 @@ -This package provides additional synchronization primitives not provided by the -Go stdlib 'sync' package. It is partially derived from the upstream 'sync' -package. diff --git a/third_party/gvsync/atomicptr_unsafe.go b/third_party/gvsync/atomicptr_unsafe.go deleted file mode 100644 index 525c4beed..000000000 --- a/third_party/gvsync/atomicptr_unsafe.go +++ /dev/null @@ -1,47 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -// Package template doesn't exist. This file must be instantiated using the -// go_template_instance rule in tools/go_generics/defs.bzl. -package template - -import ( - "sync/atomic" - "unsafe" -) - -// Value is a required type parameter. -type Value struct{} - -// An AtomicPtr is a pointer to a value of type Value that can be atomically -// loaded and stored. The zero value of an AtomicPtr represents nil. -// -// Note that copying AtomicPtr by value performs a non-atomic read of the -// stored pointer, which is unsafe if Store() can be called concurrently; in -// this case, do `dst.Store(src.Load())` instead. -// -// +stateify savable -type AtomicPtr struct { - ptr unsafe.Pointer `state:".(*Value)"` -} - -func (p *AtomicPtr) savePtr() *Value { - return p.Load() -} - -func (p *AtomicPtr) loadPtr(v *Value) { - p.Store(v) -} - -// Load returns the value set by the most recent Store. It returns nil if there -// has been no previous call to Store. -func (p *AtomicPtr) Load() *Value { - return (*Value)(atomic.LoadPointer(&p.ptr)) -} - -// Store sets the value returned by Load to x. -func (p *AtomicPtr) Store(x *Value) { - atomic.StorePointer(&p.ptr, (unsafe.Pointer)(x)) -} diff --git a/third_party/gvsync/atomicptrtest/BUILD b/third_party/gvsync/atomicptrtest/BUILD deleted file mode 100644 index 447ecf96a..000000000 --- a/third_party/gvsync/atomicptrtest/BUILD +++ /dev/null @@ -1,28 +0,0 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -package(licenses = ["notice"]) - -go_template_instance( - name = "atomicptr_int", - out = "atomicptr_int_unsafe.go", - package = "atomicptr", - suffix = "Int", - template = "//third_party/gvsync:generic_atomicptr", - types = { - "Value": "int", - }, -) - -go_library( - name = "atomicptr", - srcs = ["atomicptr_int_unsafe.go"], - importpath = "gvisor.dev/gvisor/third_party/gvsync/atomicptr", -) - -go_test( - name = "atomicptr_test", - size = "small", - srcs = ["atomicptr_test.go"], - embed = [":atomicptr"], -) diff --git a/third_party/gvsync/atomicptrtest/atomicptr_test.go b/third_party/gvsync/atomicptrtest/atomicptr_test.go deleted file mode 100644 index 8fdc5112e..000000000 --- a/third_party/gvsync/atomicptrtest/atomicptr_test.go +++ /dev/null @@ -1,31 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -package atomicptr - -import ( - "testing" -) - -func newInt(val int) *int { - return &val -} - -func TestAtomicPtr(t *testing.T) { - var p AtomicPtrInt - if got := p.Load(); got != nil { - t.Errorf("initial value is %p (%v), wanted nil", got, got) - } - want := newInt(42) - p.Store(want) - if got := p.Load(); got != want { - t.Errorf("wrong value: got %p (%v), wanted %p (%v)", got, got, want, want) - } - want = newInt(100) - p.Store(want) - if got := p.Load(); got != want { - t.Errorf("wrong value: got %p (%v), wanted %p (%v)", got, got, want, want) - } -} diff --git a/third_party/gvsync/downgradable_rwmutex_1_12_unsafe.go b/third_party/gvsync/downgradable_rwmutex_1_12_unsafe.go deleted file mode 100644 index 855b2a2b1..000000000 --- a/third_party/gvsync/downgradable_rwmutex_1_12_unsafe.go +++ /dev/null @@ -1,21 +0,0 @@ -// Copyright 2009 The Go Authors. All rights reserved. -// Copyright 2019 The gVisor Authors. -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -// +build go1.12 -// +build !go1.13 - -// TODO(b/133868570): Delete once Go 1.12 is no longer supported. - -package gvsync - -import _ "unsafe" - -//go:linkname runtimeSemrelease112 sync.runtime_Semrelease -func runtimeSemrelease112(s *uint32, handoff bool) - -func runtimeSemrelease(s *uint32, handoff bool, skipframes int) { - // 'skipframes' is only available starting from 1.13. - runtimeSemrelease112(s, handoff) -} diff --git a/third_party/gvsync/downgradable_rwmutex_1_13_unsafe.go b/third_party/gvsync/downgradable_rwmutex_1_13_unsafe.go deleted file mode 100644 index 3b9346843..000000000 --- a/third_party/gvsync/downgradable_rwmutex_1_13_unsafe.go +++ /dev/null @@ -1,16 +0,0 @@ -// Copyright 2009 The Go Authors. All rights reserved. -// Copyright 2019 The gVisor Authors. -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -// +build go1.13 -// +build !go1.15 - -// Check go:linkname function signatures when updating Go version. - -package gvsync - -import _ "unsafe" - -//go:linkname runtimeSemrelease sync.runtime_Semrelease -func runtimeSemrelease(s *uint32, handoff bool, skipframes int) diff --git a/third_party/gvsync/downgradable_rwmutex_test.go b/third_party/gvsync/downgradable_rwmutex_test.go deleted file mode 100644 index 40c384b8b..000000000 --- a/third_party/gvsync/downgradable_rwmutex_test.go +++ /dev/null @@ -1,150 +0,0 @@ -// Copyright 2009 The Go Authors. All rights reserved. -// Copyright 2019 The gVisor Authors. -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -// GOMAXPROCS=10 go test - -// Copy/pasted from the standard library's sync/rwmutex_test.go, except for the -// addition of downgradingWriter and the renaming of num_iterations to -// numIterations to shut up Golint. - -package gvsync - -import ( - "fmt" - "runtime" - "sync/atomic" - "testing" -) - -func parallelReader(m *DowngradableRWMutex, clocked, cunlock, cdone chan bool) { - m.RLock() - clocked <- true - <-cunlock - m.RUnlock() - cdone <- true -} - -func doTestParallelReaders(numReaders, gomaxprocs int) { - runtime.GOMAXPROCS(gomaxprocs) - var m DowngradableRWMutex - clocked := make(chan bool) - cunlock := make(chan bool) - cdone := make(chan bool) - for i := 0; i < numReaders; i++ { - go parallelReader(&m, clocked, cunlock, cdone) - } - // Wait for all parallel RLock()s to succeed. - for i := 0; i < numReaders; i++ { - <-clocked - } - for i := 0; i < numReaders; i++ { - cunlock <- true - } - // Wait for the goroutines to finish. - for i := 0; i < numReaders; i++ { - <-cdone - } -} - -func TestParallelReaders(t *testing.T) { - defer runtime.GOMAXPROCS(runtime.GOMAXPROCS(-1)) - doTestParallelReaders(1, 4) - doTestParallelReaders(3, 4) - doTestParallelReaders(4, 2) -} - -func reader(rwm *DowngradableRWMutex, numIterations int, activity *int32, cdone chan bool) { - for i := 0; i < numIterations; i++ { - rwm.RLock() - n := atomic.AddInt32(activity, 1) - if n < 1 || n >= 10000 { - panic(fmt.Sprintf("wlock(%d)\n", n)) - } - for i := 0; i < 100; i++ { - } - atomic.AddInt32(activity, -1) - rwm.RUnlock() - } - cdone <- true -} - -func writer(rwm *DowngradableRWMutex, numIterations int, activity *int32, cdone chan bool) { - for i := 0; i < numIterations; i++ { - rwm.Lock() - n := atomic.AddInt32(activity, 10000) - if n != 10000 { - panic(fmt.Sprintf("wlock(%d)\n", n)) - } - for i := 0; i < 100; i++ { - } - atomic.AddInt32(activity, -10000) - rwm.Unlock() - } - cdone <- true -} - -func downgradingWriter(rwm *DowngradableRWMutex, numIterations int, activity *int32, cdone chan bool) { - for i := 0; i < numIterations; i++ { - rwm.Lock() - n := atomic.AddInt32(activity, 10000) - if n != 10000 { - panic(fmt.Sprintf("wlock(%d)\n", n)) - } - for i := 0; i < 100; i++ { - } - atomic.AddInt32(activity, -10000) - rwm.DowngradeLock() - n = atomic.AddInt32(activity, 1) - if n < 1 || n >= 10000 { - panic(fmt.Sprintf("wlock(%d)\n", n)) - } - for i := 0; i < 100; i++ { - } - n = atomic.AddInt32(activity, -1) - rwm.RUnlock() - } - cdone <- true -} - -func HammerDowngradableRWMutex(gomaxprocs, numReaders, numIterations int) { - runtime.GOMAXPROCS(gomaxprocs) - // Number of active readers + 10000 * number of active writers. - var activity int32 - var rwm DowngradableRWMutex - cdone := make(chan bool) - go writer(&rwm, numIterations, &activity, cdone) - go downgradingWriter(&rwm, numIterations, &activity, cdone) - var i int - for i = 0; i < numReaders/2; i++ { - go reader(&rwm, numIterations, &activity, cdone) - } - go writer(&rwm, numIterations, &activity, cdone) - go downgradingWriter(&rwm, numIterations, &activity, cdone) - for ; i < numReaders; i++ { - go reader(&rwm, numIterations, &activity, cdone) - } - // Wait for the 4 writers and all readers to finish. - for i := 0; i < 4+numReaders; i++ { - <-cdone - } -} - -func TestDowngradableRWMutex(t *testing.T) { - defer runtime.GOMAXPROCS(runtime.GOMAXPROCS(-1)) - n := 1000 - if testing.Short() { - n = 5 - } - HammerDowngradableRWMutex(1, 1, n) - HammerDowngradableRWMutex(1, 3, n) - HammerDowngradableRWMutex(1, 10, n) - HammerDowngradableRWMutex(4, 1, n) - HammerDowngradableRWMutex(4, 3, n) - HammerDowngradableRWMutex(4, 10, n) - HammerDowngradableRWMutex(10, 1, n) - HammerDowngradableRWMutex(10, 3, n) - HammerDowngradableRWMutex(10, 10, n) - HammerDowngradableRWMutex(10, 5, n) -} diff --git a/third_party/gvsync/downgradable_rwmutex_unsafe.go b/third_party/gvsync/downgradable_rwmutex_unsafe.go deleted file mode 100644 index b7862d185..000000000 --- a/third_party/gvsync/downgradable_rwmutex_unsafe.go +++ /dev/null @@ -1,143 +0,0 @@ -// Copyright 2009 The Go Authors. All rights reserved. -// Copyright 2019 The gVisor Authors. -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -// +build go1.12 -// +build !go1.15 - -// Check go:linkname function signatures when updating Go version. - -// This is mostly copied from the standard library's sync/rwmutex.go. -// -// Happens-before relationships indicated to the race detector: -// - Unlock -> Lock (via writerSem) -// - Unlock -> RLock (via readerSem) -// - RUnlock -> Lock (via writerSem) -// - DowngradeLock -> RLock (via readerSem) - -package gvsync - -import ( - "sync" - "sync/atomic" - "unsafe" -) - -//go:linkname runtimeSemacquire sync.runtime_Semacquire -func runtimeSemacquire(s *uint32) - -// DowngradableRWMutex is identical to sync.RWMutex, but adds the DowngradeLock -// method. -type DowngradableRWMutex struct { - w sync.Mutex // held if there are pending writers - writerSem uint32 // semaphore for writers to wait for completing readers - readerSem uint32 // semaphore for readers to wait for completing writers - readerCount int32 // number of pending readers - readerWait int32 // number of departing readers -} - -const rwmutexMaxReaders = 1 << 30 - -// RLock locks rw for reading. -func (rw *DowngradableRWMutex) RLock() { - if RaceEnabled { - RaceDisable() - } - if atomic.AddInt32(&rw.readerCount, 1) < 0 { - // A writer is pending, wait for it. - runtimeSemacquire(&rw.readerSem) - } - if RaceEnabled { - RaceEnable() - RaceAcquire(unsafe.Pointer(&rw.readerSem)) - } -} - -// RUnlock undoes a single RLock call. -func (rw *DowngradableRWMutex) RUnlock() { - if RaceEnabled { - RaceReleaseMerge(unsafe.Pointer(&rw.writerSem)) - RaceDisable() - } - if r := atomic.AddInt32(&rw.readerCount, -1); r < 0 { - if r+1 == 0 || r+1 == -rwmutexMaxReaders { - panic("RUnlock of unlocked DowngradableRWMutex") - } - // A writer is pending. - if atomic.AddInt32(&rw.readerWait, -1) == 0 { - // The last reader unblocks the writer. - runtimeSemrelease(&rw.writerSem, false, 0) - } - } - if RaceEnabled { - RaceEnable() - } -} - -// Lock locks rw for writing. -func (rw *DowngradableRWMutex) Lock() { - if RaceEnabled { - RaceDisable() - } - // First, resolve competition with other writers. - rw.w.Lock() - // Announce to readers there is a pending writer. - r := atomic.AddInt32(&rw.readerCount, -rwmutexMaxReaders) + rwmutexMaxReaders - // Wait for active readers. - if r != 0 && atomic.AddInt32(&rw.readerWait, r) != 0 { - runtimeSemacquire(&rw.writerSem) - } - if RaceEnabled { - RaceEnable() - RaceAcquire(unsafe.Pointer(&rw.writerSem)) - } -} - -// Unlock unlocks rw for writing. -func (rw *DowngradableRWMutex) Unlock() { - if RaceEnabled { - RaceRelease(unsafe.Pointer(&rw.writerSem)) - RaceRelease(unsafe.Pointer(&rw.readerSem)) - RaceDisable() - } - // Announce to readers there is no active writer. - r := atomic.AddInt32(&rw.readerCount, rwmutexMaxReaders) - if r >= rwmutexMaxReaders { - panic("Unlock of unlocked DowngradableRWMutex") - } - // Unblock blocked readers, if any. - for i := 0; i < int(r); i++ { - runtimeSemrelease(&rw.readerSem, false, 0) - } - // Allow other writers to proceed. - rw.w.Unlock() - if RaceEnabled { - RaceEnable() - } -} - -// DowngradeLock atomically unlocks rw for writing and locks it for reading. -func (rw *DowngradableRWMutex) DowngradeLock() { - if RaceEnabled { - RaceRelease(unsafe.Pointer(&rw.readerSem)) - RaceDisable() - } - // Announce to readers there is no active writer and one additional reader. - r := atomic.AddInt32(&rw.readerCount, rwmutexMaxReaders+1) - if r >= rwmutexMaxReaders+1 { - panic("DowngradeLock of unlocked DowngradableRWMutex") - } - // Unblock blocked readers, if any. Note that this loop starts as 1 since r - // includes this goroutine. - for i := 1; i < int(r); i++ { - runtimeSemrelease(&rw.readerSem, false, 0) - } - // Allow other writers to proceed to rw.w.Lock(). Note that they will still - // block on rw.writerSem since at least this reader exists, such that - // DowngradeLock() is atomic with the previous write lock. - rw.w.Unlock() - if RaceEnabled { - RaceEnable() - } -} diff --git a/third_party/gvsync/gvsync.go b/third_party/gvsync/gvsync.go deleted file mode 100644 index 3bbef13c3..000000000 --- a/third_party/gvsync/gvsync.go +++ /dev/null @@ -1,7 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -// Package gvsync provides synchronization primitives. -package gvsync diff --git a/third_party/gvsync/memmove_unsafe.go b/third_party/gvsync/memmove_unsafe.go deleted file mode 100644 index 9dd1d6142..000000000 --- a/third_party/gvsync/memmove_unsafe.go +++ /dev/null @@ -1,28 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -// +build go1.12 -// +build !go1.15 - -// Check go:linkname function signatures when updating Go version. - -package gvsync - -import ( - "unsafe" -) - -//go:linkname memmove runtime.memmove -//go:noescape -func memmove(to, from unsafe.Pointer, n uintptr) - -// Memmove is exported for SeqAtomicLoad/SeqAtomicTryLoad, which can't -// define it because go_generics can't update the go:linkname annotation. -// Furthermore, go:linkname silently doesn't work if the local name is exported -// (this is of course undocumented), which is why this indirection is -// necessary. -func Memmove(to, from unsafe.Pointer, n uintptr) { - memmove(to, from, n) -} diff --git a/third_party/gvsync/norace_unsafe.go b/third_party/gvsync/norace_unsafe.go deleted file mode 100644 index e3852db8c..000000000 --- a/third_party/gvsync/norace_unsafe.go +++ /dev/null @@ -1,35 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -// +build !race - -package gvsync - -import ( - "unsafe" -) - -// RaceEnabled is true if the Go data race detector is enabled. -const RaceEnabled = false - -// RaceDisable has the same semantics as runtime.RaceDisable. -func RaceDisable() { -} - -// RaceEnable has the same semantics as runtime.RaceEnable. -func RaceEnable() { -} - -// RaceAcquire has the same semantics as runtime.RaceAcquire. -func RaceAcquire(addr unsafe.Pointer) { -} - -// RaceRelease has the same semantics as runtime.RaceRelease. -func RaceRelease(addr unsafe.Pointer) { -} - -// RaceReleaseMerge has the same semantics as runtime.RaceReleaseMerge. -func RaceReleaseMerge(addr unsafe.Pointer) { -} diff --git a/third_party/gvsync/race_unsafe.go b/third_party/gvsync/race_unsafe.go deleted file mode 100644 index 13c02a830..000000000 --- a/third_party/gvsync/race_unsafe.go +++ /dev/null @@ -1,41 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -// +build race - -package gvsync - -import ( - "runtime" - "unsafe" -) - -// RaceEnabled is true if the Go data race detector is enabled. -const RaceEnabled = true - -// RaceDisable has the same semantics as runtime.RaceDisable. -func RaceDisable() { - runtime.RaceDisable() -} - -// RaceEnable has the same semantics as runtime.RaceEnable. -func RaceEnable() { - runtime.RaceEnable() -} - -// RaceAcquire has the same semantics as runtime.RaceAcquire. -func RaceAcquire(addr unsafe.Pointer) { - runtime.RaceAcquire(addr) -} - -// RaceRelease has the same semantics as runtime.RaceRelease. -func RaceRelease(addr unsafe.Pointer) { - runtime.RaceRelease(addr) -} - -// RaceReleaseMerge has the same semantics as runtime.RaceReleaseMerge. -func RaceReleaseMerge(addr unsafe.Pointer) { - runtime.RaceReleaseMerge(addr) -} diff --git a/third_party/gvsync/seqatomic_unsafe.go b/third_party/gvsync/seqatomic_unsafe.go deleted file mode 100644 index 382eeed43..000000000 --- a/third_party/gvsync/seqatomic_unsafe.go +++ /dev/null @@ -1,72 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -// Package template doesn't exist. This file must be instantiated using the -// go_template_instance rule in tools/go_generics/defs.bzl. -package template - -import ( - "fmt" - "reflect" - "strings" - "unsafe" - - "gvisor.dev/gvisor/third_party/gvsync" -) - -// Value is a required type parameter. -// -// Value must not contain any pointers, including interface objects, function -// objects, slices, maps, channels, unsafe.Pointer, and arrays or structs -// containing any of the above. An init() function will panic if this property -// does not hold. -type Value struct{} - -// SeqAtomicLoad returns a copy of *ptr, ensuring that the read does not race -// with any writer critical sections in sc. -func SeqAtomicLoad(sc *gvsync.SeqCount, ptr *Value) Value { - // This function doesn't use SeqAtomicTryLoad because doing so is - // measurably, significantly (~20%) slower; Go is awful at inlining. - var val Value - for { - epoch := sc.BeginRead() - if gvsync.RaceEnabled { - // runtime.RaceDisable() doesn't actually stop the race detector, - // so it can't help us here. Instead, call runtime.memmove - // directly, which is not instrumented by the race detector. - gvsync.Memmove(unsafe.Pointer(&val), unsafe.Pointer(ptr), unsafe.Sizeof(val)) - } else { - // This is ~40% faster for short reads than going through memmove. - val = *ptr - } - if sc.ReadOk(epoch) { - break - } - } - return val -} - -// SeqAtomicTryLoad returns a copy of *ptr while in a reader critical section -// in sc initiated by a call to sc.BeginRead() that returned epoch. If the read -// would race with a writer critical section, SeqAtomicTryLoad returns -// (unspecified, false). -func SeqAtomicTryLoad(sc *gvsync.SeqCount, epoch gvsync.SeqCountEpoch, ptr *Value) (Value, bool) { - var val Value - if gvsync.RaceEnabled { - gvsync.Memmove(unsafe.Pointer(&val), unsafe.Pointer(ptr), unsafe.Sizeof(val)) - } else { - val = *ptr - } - return val, sc.ReadOk(epoch) -} - -func init() { - var val Value - typ := reflect.TypeOf(val) - name := typ.Name() - if ptrs := gvsync.PointersInType(typ, name); len(ptrs) != 0 { - panic(fmt.Sprintf("SeqAtomicLoad<%s> is invalid since values %s of type %s contain pointers:\n%s", typ, name, typ, strings.Join(ptrs, "\n"))) - } -} diff --git a/third_party/gvsync/seqatomictest/BUILD b/third_party/gvsync/seqatomictest/BUILD deleted file mode 100644 index c858c20c4..000000000 --- a/third_party/gvsync/seqatomictest/BUILD +++ /dev/null @@ -1,34 +0,0 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -package(licenses = ["notice"]) - -go_template_instance( - name = "seqatomic_int", - out = "seqatomic_int_unsafe.go", - package = "seqatomic", - suffix = "Int", - template = "//third_party/gvsync:generic_seqatomic", - types = { - "Value": "int", - }, -) - -go_library( - name = "seqatomic", - srcs = ["seqatomic_int_unsafe.go"], - importpath = "gvisor.dev/gvisor/third_party/gvsync/seqatomic", - deps = [ - "//third_party/gvsync", - ], -) - -go_test( - name = "seqatomic_test", - size = "small", - srcs = ["seqatomic_test.go"], - embed = [":seqatomic"], - deps = [ - "//third_party/gvsync", - ], -) diff --git a/third_party/gvsync/seqatomictest/seqatomic_test.go b/third_party/gvsync/seqatomictest/seqatomic_test.go deleted file mode 100644 index a5447f589..000000000 --- a/third_party/gvsync/seqatomictest/seqatomic_test.go +++ /dev/null @@ -1,132 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package seqatomic - -import ( - "sync/atomic" - "testing" - "time" - - "gvisor.dev/gvisor/third_party/gvsync" -) - -func TestSeqAtomicLoadUncontended(t *testing.T) { - var seq gvsync.SeqCount - const want = 1 - data := want - if got := SeqAtomicLoadInt(&seq, &data); got != want { - t.Errorf("SeqAtomicLoadInt: got %v, wanted %v", got, want) - } -} - -func TestSeqAtomicLoadAfterWrite(t *testing.T) { - var seq gvsync.SeqCount - var data int - const want = 1 - seq.BeginWrite() - data = want - seq.EndWrite() - if got := SeqAtomicLoadInt(&seq, &data); got != want { - t.Errorf("SeqAtomicLoadInt: got %v, wanted %v", got, want) - } -} - -func TestSeqAtomicLoadDuringWrite(t *testing.T) { - var seq gvsync.SeqCount - var data int - const want = 1 - seq.BeginWrite() - go func() { - time.Sleep(time.Second) - data = want - seq.EndWrite() - }() - if got := SeqAtomicLoadInt(&seq, &data); got != want { - t.Errorf("SeqAtomicLoadInt: got %v, wanted %v", got, want) - } -} - -func TestSeqAtomicTryLoadUncontended(t *testing.T) { - var seq gvsync.SeqCount - const want = 1 - data := want - epoch := seq.BeginRead() - if got, ok := SeqAtomicTryLoadInt(&seq, epoch, &data); !ok || got != want { - t.Errorf("SeqAtomicTryLoadInt: got (%v, %v), wanted (%v, true)", got, ok, want) - } -} - -func TestSeqAtomicTryLoadDuringWrite(t *testing.T) { - var seq gvsync.SeqCount - var data int - epoch := seq.BeginRead() - seq.BeginWrite() - if got, ok := SeqAtomicTryLoadInt(&seq, epoch, &data); ok { - t.Errorf("SeqAtomicTryLoadInt: got (%v, true), wanted (_, false)", got) - } - seq.EndWrite() -} - -func TestSeqAtomicTryLoadAfterWrite(t *testing.T) { - var seq gvsync.SeqCount - var data int - epoch := seq.BeginRead() - seq.BeginWrite() - seq.EndWrite() - if got, ok := SeqAtomicTryLoadInt(&seq, epoch, &data); ok { - t.Errorf("SeqAtomicTryLoadInt: got (%v, true), wanted (_, false)", got) - } -} - -func BenchmarkSeqAtomicLoadIntUncontended(b *testing.B) { - var seq gvsync.SeqCount - const want = 42 - data := want - b.RunParallel(func(pb *testing.PB) { - for pb.Next() { - if got := SeqAtomicLoadInt(&seq, &data); got != want { - b.Fatalf("SeqAtomicLoadInt: got %v, wanted %v", got, want) - } - } - }) -} - -func BenchmarkSeqAtomicTryLoadIntUncontended(b *testing.B) { - var seq gvsync.SeqCount - const want = 42 - data := want - b.RunParallel(func(pb *testing.PB) { - epoch := seq.BeginRead() - for pb.Next() { - if got, ok := SeqAtomicTryLoadInt(&seq, epoch, &data); !ok || got != want { - b.Fatalf("SeqAtomicTryLoadInt: got (%v, %v), wanted (%v, true)", got, ok, want) - } - } - }) -} - -// For comparison: -func BenchmarkAtomicValueLoadIntUncontended(b *testing.B) { - var a atomic.Value - const want = 42 - a.Store(int(want)) - b.RunParallel(func(pb *testing.PB) { - for pb.Next() { - if got := a.Load().(int); got != want { - b.Fatalf("atomic.Value.Load: got %v, wanted %v", got, want) - } - } - }) -} diff --git a/third_party/gvsync/seqcount.go b/third_party/gvsync/seqcount.go deleted file mode 100644 index 2c9c2c3d6..000000000 --- a/third_party/gvsync/seqcount.go +++ /dev/null @@ -1,149 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -package gvsync - -import ( - "fmt" - "reflect" - "runtime" - "sync/atomic" -) - -// SeqCount is a synchronization primitive for optimistic reader/writer -// synchronization in cases where readers can work with stale data and -// therefore do not need to block writers. -// -// Compared to sync/atomic.Value: -// -// - Mutation of SeqCount-protected data does not require memory allocation, -// whereas atomic.Value generally does. This is a significant advantage when -// writes are common. -// -// - Atomic reads of SeqCount-protected data require copying. This is a -// disadvantage when atomic reads are common. -// -// - SeqCount may be more flexible: correct use of SeqCount.ReadOk allows other -// operations to be made atomic with reads of SeqCount-protected data. -// -// - SeqCount may be less flexible: as of this writing, SeqCount-protected data -// cannot include pointers. -// -// - SeqCount is more cumbersome to use; atomic reads of SeqCount-protected -// data require instantiating function templates using go_generics (see -// seqatomic.go). -type SeqCount struct { - // epoch is incremented by BeginWrite and EndWrite, such that epoch is odd - // if a writer critical section is active, and a read from data protected - // by this SeqCount is atomic iff epoch is the same even value before and - // after the read. - epoch uint32 -} - -// SeqCountEpoch tracks writer critical sections in a SeqCount. -type SeqCountEpoch struct { - val uint32 -} - -// We assume that: -// -// - All functions in sync/atomic that perform a memory read are at least a -// read fence: memory reads before calls to such functions cannot be reordered -// after the call, and memory reads after calls to such functions cannot be -// reordered before the call, even if those reads do not use sync/atomic. -// -// - All functions in sync/atomic that perform a memory write are at least a -// write fence: memory writes before calls to such functions cannot be -// reordered after the call, and memory writes after calls to such functions -// cannot be reordered before the call, even if those writes do not use -// sync/atomic. -// -// As of this writing, the Go memory model completely fails to describe -// sync/atomic, but these properties are implied by -// https://groups.google.com/forum/#!topic/golang-nuts/7EnEhM3U7B8. - -// BeginRead indicates the beginning of a reader critical section. Reader -// critical sections DO NOT BLOCK writer critical sections, so operations in a -// reader critical section MAY RACE with writer critical sections. Races are -// detected by ReadOk at the end of the reader critical section. Thus, the -// low-level structure of readers is generally: -// -// for { -// epoch := seq.BeginRead() -// // do something idempotent with seq-protected data -// if seq.ReadOk(epoch) { -// break -// } -// } -// -// However, since reader critical sections may race with writer critical -// sections, the Go race detector will (accurately) flag data races in readers -// using this pattern. Most users of SeqCount will need to use the -// SeqAtomicLoad function template in seqatomic.go. -func (s *SeqCount) BeginRead() SeqCountEpoch { - epoch := atomic.LoadUint32(&s.epoch) - for epoch&1 != 0 { - runtime.Gosched() - epoch = atomic.LoadUint32(&s.epoch) - } - return SeqCountEpoch{epoch} -} - -// ReadOk returns true if the reader critical section initiated by a previous -// call to BeginRead() that returned epoch did not race with any writer critical -// sections. -// -// ReadOk may be called any number of times during a reader critical section. -// Reader critical sections do not need to be explicitly terminated; the last -// call to ReadOk is implicitly the end of the reader critical section. -func (s *SeqCount) ReadOk(epoch SeqCountEpoch) bool { - return atomic.LoadUint32(&s.epoch) == epoch.val -} - -// BeginWrite indicates the beginning of a writer critical section. -// -// SeqCount does not support concurrent writer critical sections; clients with -// concurrent writers must synchronize them using e.g. sync.Mutex. -func (s *SeqCount) BeginWrite() { - if epoch := atomic.AddUint32(&s.epoch, 1); epoch&1 == 0 { - panic("SeqCount.BeginWrite during writer critical section") - } -} - -// EndWrite ends the effect of a preceding BeginWrite. -func (s *SeqCount) EndWrite() { - if epoch := atomic.AddUint32(&s.epoch, 1); epoch&1 != 0 { - panic("SeqCount.EndWrite outside writer critical section") - } -} - -// PointersInType returns a list of pointers reachable from values named -// valName of the given type. -// -// PointersInType is not exhaustive, but it is guaranteed that if typ contains -// at least one pointer, then PointersInTypeOf returns a non-empty list. -func PointersInType(typ reflect.Type, valName string) []string { - switch kind := typ.Kind(); kind { - case reflect.Bool, reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64, reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr, reflect.Float32, reflect.Float64, reflect.Complex64, reflect.Complex128: - return nil - - case reflect.Chan, reflect.Func, reflect.Interface, reflect.Map, reflect.Ptr, reflect.Slice, reflect.String, reflect.UnsafePointer: - return []string{valName} - - case reflect.Array: - return PointersInType(typ.Elem(), valName+"[]") - - case reflect.Struct: - var ptrs []string - for i, n := 0, typ.NumField(); i < n; i++ { - field := typ.Field(i) - ptrs = append(ptrs, PointersInType(field.Type, fmt.Sprintf("%s.%s", valName, field.Name))...) - } - return ptrs - - default: - return []string{fmt.Sprintf("%s (of type %s with unknown kind %s)", valName, typ, kind)} - } -} diff --git a/third_party/gvsync/seqcount_test.go b/third_party/gvsync/seqcount_test.go deleted file mode 100644 index 085e574b3..000000000 --- a/third_party/gvsync/seqcount_test.go +++ /dev/null @@ -1,153 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -package gvsync - -import ( - "reflect" - "testing" - "time" -) - -func TestSeqCountWriteUncontended(t *testing.T) { - var seq SeqCount - seq.BeginWrite() - seq.EndWrite() -} - -func TestSeqCountReadUncontended(t *testing.T) { - var seq SeqCount - epoch := seq.BeginRead() - if !seq.ReadOk(epoch) { - t.Errorf("ReadOk: got false, wanted true") - } -} - -func TestSeqCountBeginReadAfterWrite(t *testing.T) { - var seq SeqCount - var data int32 - const want = 1 - seq.BeginWrite() - data = want - seq.EndWrite() - epoch := seq.BeginRead() - if data != want { - t.Errorf("Reader: got %v, wanted %v", data, want) - } - if !seq.ReadOk(epoch) { - t.Errorf("ReadOk: got false, wanted true") - } -} - -func TestSeqCountBeginReadDuringWrite(t *testing.T) { - var seq SeqCount - var data int - const want = 1 - seq.BeginWrite() - go func() { - time.Sleep(time.Second) - data = want - seq.EndWrite() - }() - epoch := seq.BeginRead() - if data != want { - t.Errorf("Reader: got %v, wanted %v", data, want) - } - if !seq.ReadOk(epoch) { - t.Errorf("ReadOk: got false, wanted true") - } -} - -func TestSeqCountReadOkAfterWrite(t *testing.T) { - var seq SeqCount - epoch := seq.BeginRead() - seq.BeginWrite() - seq.EndWrite() - if seq.ReadOk(epoch) { - t.Errorf("ReadOk: got true, wanted false") - } -} - -func TestSeqCountReadOkDuringWrite(t *testing.T) { - var seq SeqCount - epoch := seq.BeginRead() - seq.BeginWrite() - if seq.ReadOk(epoch) { - t.Errorf("ReadOk: got true, wanted false") - } - seq.EndWrite() -} - -func BenchmarkSeqCountWriteUncontended(b *testing.B) { - var seq SeqCount - for i := 0; i < b.N; i++ { - seq.BeginWrite() - seq.EndWrite() - } -} - -func BenchmarkSeqCountReadUncontended(b *testing.B) { - var seq SeqCount - b.RunParallel(func(pb *testing.PB) { - for pb.Next() { - epoch := seq.BeginRead() - if !seq.ReadOk(epoch) { - b.Fatalf("ReadOk: got false, wanted true") - } - } - }) -} - -func TestPointersInType(t *testing.T) { - for _, test := range []struct { - name string // used for both test and value name - val interface{} - ptrs []string - }{ - { - name: "EmptyStruct", - val: struct{}{}, - }, - { - name: "Int", - val: int(0), - }, - { - name: "MixedStruct", - val: struct { - b bool - I int - ExportedPtr *struct{} - unexportedPtr *struct{} - arr [2]int - ptrArr [2]*int - nestedStruct struct { - nestedNonptr int - nestedPtr *int - } - structArr [1]struct { - nonptr int - ptr *int - } - }{}, - ptrs: []string{ - "MixedStruct.ExportedPtr", - "MixedStruct.unexportedPtr", - "MixedStruct.ptrArr[]", - "MixedStruct.nestedStruct.nestedPtr", - "MixedStruct.structArr[].ptr", - }, - }, - } { - t.Run(test.name, func(t *testing.T) { - typ := reflect.TypeOf(test.val) - ptrs := PointersInType(typ, test.name) - t.Logf("Found pointers: %v", ptrs) - if (len(ptrs) != 0 || len(test.ptrs) != 0) && !reflect.DeepEqual(ptrs, test.ptrs) { - t.Errorf("Got %v, wanted %v", ptrs, test.ptrs) - } - }) - } -} diff --git a/tools/go_marshal/test/BUILD b/tools/go_marshal/test/BUILD index fa82f8e9b..d412e1ccf 100644 --- a/tools/go_marshal/test/BUILD +++ b/tools/go_marshal/test/BUILD @@ -1,9 +1,8 @@ load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools/go_marshal:defs.bzl", "go_library") package(licenses = ["notice"]) -load("//tools/go_marshal:defs.bzl", "go_library") - package_group( name = "gomarshal_test", packages = [ diff --git a/tools/go_marshal/test/external/BUILD b/tools/go_marshal/test/external/BUILD index 8fb43179b..9bb89e1da 100644 --- a/tools/go_marshal/test/external/BUILD +++ b/tools/go_marshal/test/external/BUILD @@ -1,7 +1,7 @@ -package(licenses = ["notice"]) - load("//tools/go_marshal:defs.bzl", "go_library") +package(licenses = ["notice"]) + go_library( name = "external", testonly = 1, -- cgit v1.2.3 From 07635d20d40e1a279c4b063abaaad51048400ed7 Mon Sep 17 00:00:00 2001 From: lubinszARM <34124929+lubinszARM@users.noreply.github.com> Date: Fri, 22 Nov 2019 11:54:04 -0800 Subject: enable ring0/pagetables to support arm64 Signed-off-by: Bin Lu COPYBARA_INTEGRATE_REVIEW=https://github.com/google/gvisor/pull/891 from lubinszARM:pr_pagetable 2385de75a8662af3ab1ae289dd74dd0e5dcfaf66 PiperOrigin-RevId: 282013224 --- pkg/sentry/platform/ring0/pagetables/BUILD | 16 +- pkg/sentry/platform/ring0/pagetables/pagetables.go | 9 - .../ring0/pagetables/pagetables_aarch64.go | 212 ++++++++++++++ .../platform/ring0/pagetables/pagetables_amd64.go | 9 + .../platform/ring0/pagetables/pagetables_arm64.go | 57 ++++ .../ring0/pagetables/pagetables_arm64_test.go | 80 ++++++ .../platform/ring0/pagetables/walker_arm64.go | 314 +++++++++++++++++++++ 7 files changed, 684 insertions(+), 13 deletions(-) create mode 100644 pkg/sentry/platform/ring0/pagetables/pagetables_aarch64.go create mode 100644 pkg/sentry/platform/ring0/pagetables/pagetables_arm64.go create mode 100644 pkg/sentry/platform/ring0/pagetables/pagetables_arm64_test.go create mode 100644 pkg/sentry/platform/ring0/pagetables/walker_arm64.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/ring0/pagetables/BUILD b/pkg/sentry/platform/ring0/pagetables/BUILD index 934a90378..e2e15ba5c 100644 --- a/pkg/sentry/platform/ring0/pagetables/BUILD +++ b/pkg/sentry/platform/ring0/pagetables/BUILD @@ -1,14 +1,17 @@ -load("//tools/go_stateify:defs.bzl", "go_library") load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools/go_stateify:defs.bzl", "go_library") load("//tools/go_generics:defs.bzl", "go_template", "go_template_instance") package(licenses = ["notice"]) +config_setting( + name = "aarch64", + constraint_values = ["@bazel_tools//platforms:aarch64"], +) + go_template( name = "generic_walker", - srcs = [ - "walker_amd64.go", - ], + srcs = ["walker_amd64.go"], opt_types = [ "Visitor", ], @@ -76,9 +79,13 @@ go_library( "allocator.go", "allocator_unsafe.go", "pagetables.go", + "pagetables_aarch64.go", "pagetables_amd64.go", + "pagetables_arm64.go", "pagetables_x86.go", "pcids_x86.go", + "walker_amd64.go", + "walker_arm64.go", "walker_empty.go", "walker_lookup.go", "walker_map.go", @@ -97,6 +104,7 @@ go_test( size = "small", srcs = [ "pagetables_amd64_test.go", + "pagetables_arm64_test.go", "pagetables_test.go", "walker_check.go", ], diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables.go b/pkg/sentry/platform/ring0/pagetables/pagetables.go index 904f1a6de..30c64a372 100644 --- a/pkg/sentry/platform/ring0/pagetables/pagetables.go +++ b/pkg/sentry/platform/ring0/pagetables/pagetables.go @@ -48,15 +48,6 @@ func New(a Allocator) *PageTables { return p } -// Init initializes a set of PageTables. -// -//go:nosplit -func (p *PageTables) Init(allocator Allocator) { - p.Allocator = allocator - p.root = p.Allocator.NewPTEs() - p.rootPhysical = p.Allocator.PhysicalFor(p.root) -} - // mapVisitor is used for map. type mapVisitor struct { target uintptr // Input. diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64.go b/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64.go new file mode 100644 index 000000000..e78424766 --- /dev/null +++ b/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64.go @@ -0,0 +1,212 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package pagetables + +import ( + "sync/atomic" + + "gvisor.dev/gvisor/pkg/sentry/usermem" +) + +// archPageTables is architecture-specific data. +type archPageTables struct { + // root is the pagetable root for kernel space. + root *PTEs + + // rootPhysical is the cached physical address of the root. + // + // This is saved only to prevent constant translation. + rootPhysical uintptr + + asid uint16 +} + +// TTBR0_EL1 returns the translation table base register 0. +// +//go:nosplit +func (p *PageTables) TTBR0_EL1(noFlush bool, asid uint16) uint64 { + return uint64(p.rootPhysical) | (uint64(asid)&ttbrASIDMask)< end") + } + if start < lowerTop { + if end <= lowerTop { + w.iterateRangeCanonical(start, end) + } else if end > lowerTop && end <= upperBottom { + if w.visitor.requiresAlloc() { + panic("alloc spans non-canonical range") + } + w.iterateRangeCanonical(start, lowerTop) + } else { + if w.visitor.requiresAlloc() { + panic("alloc spans non-canonical range") + } + w.iterateRangeCanonical(start, lowerTop) + w.iterateRangeCanonical(upperBottom, end) + } + } else if start < upperBottom { + if end <= upperBottom { + if w.visitor.requiresAlloc() { + panic("alloc spans non-canonical range") + } + } else { + if w.visitor.requiresAlloc() { + panic("alloc spans non-canonical range") + } + w.iterateRangeCanonical(upperBottom, end) + } + } else { + w.iterateRangeCanonical(start, end) + } +} + +// next returns the next address quantized by the given size. +// +//go:nosplit +func next(start uintptr, size uintptr) uintptr { + start &= ^(size - 1) + start += size + return start +} + +// iterateRangeCanonical walks a canonical range. +// +//go:nosplit +func (w *Walker) iterateRangeCanonical(start, end uintptr) { + pgdEntryIndex := w.pageTables.root + if start >= upperBottom { + pgdEntryIndex = w.pageTables.archPageTables.root + } + + for pgdIndex := (uint16((start & pgdMask) >> pgdShift)); start < end && pgdIndex < entriesPerPage; pgdIndex++ { + var ( + pgdEntry = &pgdEntryIndex[pgdIndex] + pudEntries *PTEs + ) + if !pgdEntry.Valid() { + if !w.visitor.requiresAlloc() { + // Skip over this entry. + start = next(start, pgdSize) + continue + } + + // Allocate a new pgd. + pudEntries = w.pageTables.Allocator.NewPTEs() + pgdEntry.setPageTable(w.pageTables, pudEntries) + } else { + pudEntries = w.pageTables.Allocator.LookupPTEs(pgdEntry.Address()) + } + + // Map the next level. + clearPUDEntries := uint16(0) + + for pudIndex := uint16((start & pudMask) >> pudShift); start < end && pudIndex < entriesPerPage; pudIndex++ { + var ( + pudEntry = &pudEntries[pudIndex] + pmdEntries *PTEs + ) + if !pudEntry.Valid() { + if !w.visitor.requiresAlloc() { + // Skip over this entry. + clearPUDEntries++ + start = next(start, pudSize) + continue + } + + // This level has 1-GB sect pages. Is this + // entire region at least as large as a single + // PUD entry? If so, we can skip allocating a + // new page for the pmd. + if start&(pudSize-1) == 0 && end-start >= pudSize { + pudEntry.SetSect() + w.visitor.visit(uintptr(start), pudEntry, pudSize-1) + if pudEntry.Valid() { + start = next(start, pudSize) + continue + } + } + + // Allocate a new pud. + pmdEntries = w.pageTables.Allocator.NewPTEs() + pudEntry.setPageTable(w.pageTables, pmdEntries) + + } else if pudEntry.IsSect() { + // Does this page need to be split? + if w.visitor.requiresSplit() && (start&(pudSize-1) != 0 || end < next(start, pudSize)) { + // Install the relevant entries. + pmdEntries = w.pageTables.Allocator.NewPTEs() + for index := uint16(0); index < entriesPerPage; index++ { + pmdEntries[index].SetSect() + pmdEntries[index].Set( + pudEntry.Address()+(pmdSize*uintptr(index)), + pudEntry.Opts()) + } + pudEntry.setPageTable(w.pageTables, pmdEntries) + } else { + // A sect page to be checked directly. + w.visitor.visit(uintptr(start), pudEntry, pudSize-1) + + // Might have been cleared. + if !pudEntry.Valid() { + clearPUDEntries++ + } + + // Note that the sect page was changed. + start = next(start, pudSize) + continue + } + + } else { + pmdEntries = w.pageTables.Allocator.LookupPTEs(pudEntry.Address()) + } + + // Map the next level, since this is valid. + clearPMDEntries := uint16(0) + + for pmdIndex := uint16((start & pmdMask) >> pmdShift); start < end && pmdIndex < entriesPerPage; pmdIndex++ { + var ( + pmdEntry = &pmdEntries[pmdIndex] + pteEntries *PTEs + ) + if !pmdEntry.Valid() { + if !w.visitor.requiresAlloc() { + // Skip over this entry. + clearPMDEntries++ + start = next(start, pmdSize) + continue + } + + // This level has 2-MB huge pages. If this + // region is contined in a single PMD entry? + // As above, we can skip allocating a new page. + if start&(pmdSize-1) == 0 && end-start >= pmdSize { + pmdEntry.SetSect() + w.visitor.visit(uintptr(start), pmdEntry, pmdSize-1) + if pmdEntry.Valid() { + start = next(start, pmdSize) + continue + } + } + + // Allocate a new pmd. + pteEntries = w.pageTables.Allocator.NewPTEs() + pmdEntry.setPageTable(w.pageTables, pteEntries) + + } else if pmdEntry.IsSect() { + // Does this page need to be split? + if w.visitor.requiresSplit() && (start&(pmdSize-1) != 0 || end < next(start, pmdSize)) { + // Install the relevant entries. + pteEntries = w.pageTables.Allocator.NewPTEs() + for index := uint16(0); index < entriesPerPage; index++ { + pteEntries[index].Set( + pmdEntry.Address()+(pteSize*uintptr(index)), + pmdEntry.Opts()) + } + pmdEntry.setPageTable(w.pageTables, pteEntries) + } else { + // A huge page to be checked directly. + w.visitor.visit(uintptr(start), pmdEntry, pmdSize-1) + + // Might have been cleared. + if !pmdEntry.Valid() { + clearPMDEntries++ + } + + // Note that the huge page was changed. + start = next(start, pmdSize) + continue + } + + } else { + pteEntries = w.pageTables.Allocator.LookupPTEs(pmdEntry.Address()) + } + + // Map the next level, since this is valid. + clearPTEEntries := uint16(0) + + for pteIndex := uint16((start & pteMask) >> pteShift); start < end && pteIndex < entriesPerPage; pteIndex++ { + var ( + pteEntry = &pteEntries[pteIndex] + ) + if !pteEntry.Valid() && !w.visitor.requiresAlloc() { + clearPTEEntries++ + start += pteSize + continue + } + + // At this point, we are guaranteed that start%pteSize == 0. + w.visitor.visit(uintptr(start), pteEntry, pteSize-1) + if !pteEntry.Valid() { + if w.visitor.requiresAlloc() { + panic("PTE not set after iteration with requiresAlloc!") + } + clearPTEEntries++ + } + + // Note that the pte was changed. + start += pteSize + continue + } + + // Check if we no longer need this page. + if clearPTEEntries == entriesPerPage { + pmdEntry.Clear() + w.pageTables.Allocator.FreePTEs(pteEntries) + clearPMDEntries++ + } + } + + // Check if we no longer need this page. + if clearPMDEntries == entriesPerPage { + pudEntry.Clear() + w.pageTables.Allocator.FreePTEs(pmdEntries) + clearPUDEntries++ + } + } + + // Check if we no longer need this page. + if clearPUDEntries == entriesPerPage { + pgdEntry.Clear() + w.pageTables.Allocator.FreePTEs(pudEntries) + } + } +} -- cgit v1.2.3 From a5f7b82036f4a062183f1fed9c27227636c8eed5 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Mon, 25 Nov 2019 11:19:15 -0800 Subject: Internal change. PiperOrigin-RevId: 282396322 --- pkg/sentry/BUILD | 3 --- 1 file changed, 3 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/BUILD b/pkg/sentry/BUILD index 2a7122957..2d6379c86 100644 --- a/pkg/sentry/BUILD +++ b/pkg/sentry/BUILD @@ -10,8 +10,5 @@ package_group( "//runsc/...", # Code generated by go_marshal relies on go_marshal libraries. "//tools/go_marshal/...", - - # Keep the old paths as a temporary measure. - "//third_party/golang/gvisor/pkg/sentry/...", ], ) -- cgit v1.2.3 From 128948d6ae94009c6ad13a0bd96e03e45a560477 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Mon, 25 Nov 2019 15:20:25 -0800 Subject: Implement basic umounting for vfs2. This is required to test filesystems with a non-trivial implementation of FilesystemImpl.Release(). Propagation isn't handled yet, and umount isn't yet plumbed out to VirtualFilesystem.UmountAt(), but otherwise the implementation of umount is believed to be correct. - Move entering mountTable.seq writer critical sections to callers of mountTable.{insert,remove}Seqed. This is required since umount(2) must ensure that no new references are taken on the candidate mount after checking that it isn't busy, which is only possible by entering a vfs.mountTable.seq writer critical section before the check and remaining in it until after VFS.umountRecursiveLocked() is complete. (Linux does the same thing: fs/namespace.c:do_umount() => lock_mount_hash(), fs/pnode.c:propagate_mount_busy(), umount_tree(), unlock_mount_hash().) - It's not possible for dentry deletion to umount while only holding VFS.mountMu for reading, but it's also very unappealing to hold VFS.mountMu exclusively around e.g. gofer unlink RPCs. Introduce dentry.mu to avoid these problems. This means that VFS.mountMu is never acquired for reading, so change it to a sync.Mutex. PiperOrigin-RevId: 282444343 --- pkg/sentry/fsimpl/memfs/BUILD | 4 +- pkg/sentry/fsimpl/memfs/benchmark_test.go | 22 ++- pkg/sentry/vfs/README.md | 4 +- pkg/sentry/vfs/dentry.go | 128 +++++++++--- pkg/sentry/vfs/mount.go | 319 +++++++++++++++++++++--------- pkg/sentry/vfs/mount_test.go | 34 ++-- pkg/sentry/vfs/mount_unsafe.go | 60 +++--- pkg/sentry/vfs/resolving_path.go | 8 +- pkg/sentry/vfs/syscalls.go | 2 + pkg/sentry/vfs/vfs.go | 11 +- 10 files changed, 423 insertions(+), 169 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/memfs/BUILD b/pkg/sentry/fsimpl/memfs/BUILD index 952b20c51..bc5c0b591 100644 --- a/pkg/sentry/fsimpl/memfs/BUILD +++ b/pkg/sentry/fsimpl/memfs/BUILD @@ -1,9 +1,10 @@ load("//tools/go_stateify:defs.bzl", "go_library") load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") package(licenses = ["notice"]) +load("//tools/go_generics:defs.bzl", "go_template_instance") + go_template_instance( name = "dentry_list", out = "dentry_list.go", @@ -48,6 +49,7 @@ go_test( deps = [ ":memfs", "//pkg/abi/linux", + "//pkg/refs", "//pkg/sentry/context", "//pkg/sentry/context/contexttest", "//pkg/sentry/fs", diff --git a/pkg/sentry/fsimpl/memfs/benchmark_test.go b/pkg/sentry/fsimpl/memfs/benchmark_test.go index a94b17db6..23a846c08 100644 --- a/pkg/sentry/fsimpl/memfs/benchmark_test.go +++ b/pkg/sentry/fsimpl/memfs/benchmark_test.go @@ -21,6 +21,7 @@ import ( "testing" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/context/contexttest" "gvisor.dev/gvisor/pkg/sentry/fs" @@ -160,6 +161,8 @@ func BenchmarkVFS1TmpfsStat(b *testing.B) { b.Fatalf("stat(%q) failed: %v", filePath, err) } } + // Don't include deferred cleanup in benchmark time. + b.StopTimer() }) } } @@ -177,6 +180,7 @@ func BenchmarkVFS2MemfsStat(b *testing.B) { if err != nil { b.Fatalf("failed to create tmpfs root mount: %v", err) } + defer mntns.DecRef(vfsObj) var filePathBuilder strings.Builder filePathBuilder.WriteByte('/') @@ -186,7 +190,6 @@ func BenchmarkVFS2MemfsStat(b *testing.B) { defer root.DecRef() vd := root vd.IncRef() - defer vd.DecRef() for i := depth; i > 0; i-- { name := fmt.Sprintf("%d", i) pop := vfs.PathOperation{ @@ -219,6 +222,8 @@ func BenchmarkVFS2MemfsStat(b *testing.B) { Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL, Mode: 0644, }) + vd.DecRef() + vd = vfs.VirtualDentry{} if err != nil { b.Fatalf("failed to create file %q: %v", filename, err) } @@ -243,6 +248,8 @@ func BenchmarkVFS2MemfsStat(b *testing.B) { b.Fatalf("got wrong permissions (%0o)", stat.Mode) } } + // Don't include deferred cleanup in benchmark time. + b.StopTimer() }) } } @@ -343,6 +350,8 @@ func BenchmarkVFS1TmpfsMountStat(b *testing.B) { b.Fatalf("stat(%q) failed: %v", filePath, err) } } + // Don't include deferred cleanup in benchmark time. + b.StopTimer() }) } } @@ -360,6 +369,7 @@ func BenchmarkVFS2MemfsMountStat(b *testing.B) { if err != nil { b.Fatalf("failed to create tmpfs root mount: %v", err) } + defer mntns.DecRef(vfsObj) var filePathBuilder strings.Builder filePathBuilder.WriteByte('/') @@ -395,7 +405,6 @@ func BenchmarkVFS2MemfsMountStat(b *testing.B) { if err != nil { b.Fatalf("failed to walk to mount root: %v", err) } - defer vd.DecRef() for i := depth; i > 0; i-- { name := fmt.Sprintf("%d", i) pop := vfs.PathOperation{ @@ -435,6 +444,7 @@ func BenchmarkVFS2MemfsMountStat(b *testing.B) { Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL, Mode: 0644, }) + vd.DecRef() if err != nil { b.Fatalf("failed to create file %q: %v", filename, err) } @@ -459,6 +469,14 @@ func BenchmarkVFS2MemfsMountStat(b *testing.B) { b.Fatalf("got wrong permissions (%0o)", stat.Mode) } } + // Don't include deferred cleanup in benchmark time. + b.StopTimer() }) } } + +func init() { + // Turn off reference leak checking for a fair comparison between vfs1 and + // vfs2. + refs.SetLeakMode(refs.NoLeakChecking) +} diff --git a/pkg/sentry/vfs/README.md b/pkg/sentry/vfs/README.md index 7847854bc..9aa133bcb 100644 --- a/pkg/sentry/vfs/README.md +++ b/pkg/sentry/vfs/README.md @@ -39,8 +39,8 @@ Mount references are held by: - Mount: Each referenced Mount holds a reference on its parent, which is the mount containing its mount point. -- VirtualFilesystem: A reference is held on all Mounts that are attached - (reachable by Mount traversal). +- VirtualFilesystem: A reference is held on each Mount that has not been + umounted. MountNamespace and FileDescription references are held by users of VFS. The expectation is that each `kernel.Task` holds a reference on its corresponding diff --git a/pkg/sentry/vfs/dentry.go b/pkg/sentry/vfs/dentry.go index 45912fc58..09ed5a70e 100644 --- a/pkg/sentry/vfs/dentry.go +++ b/pkg/sentry/vfs/dentry.go @@ -16,6 +16,7 @@ package vfs import ( "fmt" + "sync" "sync/atomic" "gvisor.dev/gvisor/pkg/syserror" @@ -50,7 +51,7 @@ import ( // and not inodes. Furthermore, when parties outside the scope of VFS can // rename inodes on such filesystems, VFS generally cannot "follow" the rename, // both due to synchronization issues and because it may not even be able to -// name the destination path; this implies that it would in fact be *incorrect* +// name the destination path; this implies that it would in fact be incorrect // for Dentries to be associated with inodes on such filesystems. Consequently, // operations that are inode operations in Linux are FilesystemImpl methods // and/or FileDescriptionImpl methods in gVisor's VFS. Filesystems that do @@ -84,6 +85,9 @@ type Dentry struct { // mounts is accessed using atomic memory operations. mounts uint32 + // mu synchronizes disowning and mounting over this Dentry. + mu sync.Mutex + // children are child Dentries. children map[string]*Dentry @@ -228,36 +232,48 @@ func (vfs *VirtualFilesystem) PrepareDeleteDentry(mntns *MountNamespace, d *Dent panic("d is already disowned") } } - vfs.mountMu.RLock() - if _, ok := mntns.mountpoints[d]; ok { - vfs.mountMu.RUnlock() + vfs.mountMu.Lock() + if mntns.mountpoints[d] != 0 { + vfs.mountMu.Unlock() return syserror.EBUSY } - // Return with vfs.mountMu locked, which will be unlocked by - // AbortDeleteDentry or CommitDeleteDentry. + d.mu.Lock() + vfs.mountMu.Unlock() + // Return with d.mu locked to block attempts to mount over it; it will be + // unlocked by AbortDeleteDentry or CommitDeleteDentry. return nil } // AbortDeleteDentry must be called after PrepareDeleteDentry if the deletion // fails. -func (vfs *VirtualFilesystem) AbortDeleteDentry() { - vfs.mountMu.RUnlock() +func (vfs *VirtualFilesystem) AbortDeleteDentry(d *Dentry) { + d.mu.Unlock() } // CommitDeleteDentry must be called after the file represented by d is // deleted, and causes d to become disowned. // +// CommitDeleteDentry is a mutator of d and d.Parent(). +// // Preconditions: PrepareDeleteDentry was previously called on d. func (vfs *VirtualFilesystem) CommitDeleteDentry(d *Dentry) { - delete(d.parent.children, d.name) + if d.parent != nil { + delete(d.parent.children, d.name) + } d.setDisowned() - // TODO: lazily unmount mounts at d - vfs.mountMu.RUnlock() + d.mu.Unlock() + if d.isMounted() { + vfs.forgetDisownedMountpoint(d) + } } // DeleteDentry combines PrepareDeleteDentry and CommitDeleteDentry, as // appropriate for in-memory filesystems that don't need to ensure that some // external state change succeeds before committing the deletion. +// +// DeleteDentry is a mutator of d and d.Parent(). +// +// Preconditions: d is a child Dentry. func (vfs *VirtualFilesystem) DeleteDentry(mntns *MountNamespace, d *Dentry) error { if err := vfs.PrepareDeleteDentry(mntns, d); err != nil { return err @@ -266,6 +282,27 @@ func (vfs *VirtualFilesystem) DeleteDentry(mntns *MountNamespace, d *Dentry) err return nil } +// ForceDeleteDentry causes d to become disowned. It should only be used in +// cases where VFS has no ability to stop the deletion (e.g. d represents the +// local state of a file on a remote filesystem on which the file has already +// been deleted). +// +// ForceDeleteDentry is a mutator of d and d.Parent(). +// +// Preconditions: d is a child Dentry. +func (vfs *VirtualFilesystem) ForceDeleteDentry(d *Dentry) { + if checkInvariants { + if d.parent == nil { + panic("d is independent") + } + if d.IsDisowned() { + panic("d is already disowned") + } + } + d.mu.Lock() + vfs.CommitDeleteDentry(d) +} + // PrepareRenameDentry must be called before attempting to rename the file // represented by from. If to is not nil, it represents the file that will be // replaced or exchanged by the rename. If PrepareRenameDentry succeeds, the @@ -291,18 +328,21 @@ func (vfs *VirtualFilesystem) PrepareRenameDentry(mntns *MountNamespace, from, t } } } - vfs.mountMu.RLock() - if _, ok := mntns.mountpoints[from]; ok { - vfs.mountMu.RUnlock() + vfs.mountMu.Lock() + if mntns.mountpoints[from] != 0 { + vfs.mountMu.Unlock() return syserror.EBUSY } if to != nil { - if _, ok := mntns.mountpoints[to]; ok { - vfs.mountMu.RUnlock() + if mntns.mountpoints[to] != 0 { + vfs.mountMu.Unlock() return syserror.EBUSY } + to.mu.Lock() } - // Return with vfs.mountMu locked, which will be unlocked by + from.mu.Lock() + vfs.mountMu.Unlock() + // Return with from.mu and to.mu locked, which will be unlocked by // AbortRenameDentry, CommitRenameReplaceDentry, or // CommitRenameExchangeDentry. return nil @@ -310,38 +350,76 @@ func (vfs *VirtualFilesystem) PrepareRenameDentry(mntns *MountNamespace, from, t // AbortRenameDentry must be called after PrepareRenameDentry if the rename // fails. -func (vfs *VirtualFilesystem) AbortRenameDentry() { - vfs.mountMu.RUnlock() +func (vfs *VirtualFilesystem) AbortRenameDentry(from, to *Dentry) { + from.mu.Unlock() + if to != nil { + to.mu.Unlock() + } } // CommitRenameReplaceDentry must be called after the file represented by from // is renamed without RENAME_EXCHANGE. If to is not nil, it represents the file // that was replaced by from. // +// CommitRenameReplaceDentry is a mutator of from, to, from.Parent(), and +// to.Parent(). +// // Preconditions: PrepareRenameDentry was previously called on from and to. // newParent.Child(newName) == to. func (vfs *VirtualFilesystem) CommitRenameReplaceDentry(from, newParent *Dentry, newName string, to *Dentry) { - if to != nil { - to.setDisowned() - // TODO: lazily unmount mounts at d - } if newParent.children == nil { newParent.children = make(map[string]*Dentry) } newParent.children[newName] = from from.parent = newParent from.name = newName - vfs.mountMu.RUnlock() + from.mu.Unlock() + if to != nil { + to.setDisowned() + to.mu.Unlock() + if to.isMounted() { + vfs.forgetDisownedMountpoint(to) + } + } } // CommitRenameExchangeDentry must be called after the files represented by // from and to are exchanged by rename(RENAME_EXCHANGE). // +// CommitRenameExchangeDentry is a mutator of from, to, from.Parent(), and +// to.Parent(). +// // Preconditions: PrepareRenameDentry was previously called on from and to. func (vfs *VirtualFilesystem) CommitRenameExchangeDentry(from, to *Dentry) { from.parent, to.parent = to.parent, from.parent from.name, to.name = to.name, from.name from.parent.children[from.name] = from to.parent.children[to.name] = to - vfs.mountMu.RUnlock() + from.mu.Unlock() + to.mu.Unlock() +} + +// forgetDisownedMountpoint is called when a mount point is deleted to umount +// all mounts using it in all other mount namespaces. +// +// forgetDisownedMountpoint is analogous to Linux's +// fs/namespace.c:__detach_mounts(). +func (vfs *VirtualFilesystem) forgetDisownedMountpoint(d *Dentry) { + var ( + vdsToDecRef []VirtualDentry + mountsToDecRef []*Mount + ) + vfs.mountMu.Lock() + vfs.mounts.seq.BeginWrite() + for mnt := range vfs.mountpoints[d] { + vdsToDecRef, mountsToDecRef = vfs.umountRecursiveLocked(mnt, &umountRecursiveOptions{}, vdsToDecRef, mountsToDecRef) + } + vfs.mounts.seq.EndWrite() + vfs.mountMu.Unlock() + for _, vd := range vdsToDecRef { + vd.DecRef() + } + for _, mnt := range mountsToDecRef { + mnt.decRef() + } } diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index 11702f720..198fb8067 100644 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go @@ -38,16 +38,12 @@ import ( // Mount is analogous to Linux's struct mount. (gVisor does not distinguish // between struct mount and struct vfsmount.) type Mount struct { - // The lower 63 bits of refs are a reference count. The MSB of refs is set - // if the Mount has been eagerly unmounted, as by umount(2) without the - // MNT_DETACH flag. refs is accessed using atomic memory operations. - refs int64 - - // The lower 63 bits of writers is the number of calls to - // Mount.CheckBeginWrite() that have not yet been paired with a call to - // Mount.EndWrite(). The MSB of writers is set if MS_RDONLY is in effect. - // writers is accessed using atomic memory operations. - writers int64 + // vfs, fs, and root are immutable. References are held on fs and root. + // + // Invariant: root belongs to fs. + vfs *VirtualFilesystem + fs *Filesystem + root *Dentry // key is protected by VirtualFilesystem.mountMu and // VirtualFilesystem.mounts.seq, and may be nil. References are held on @@ -57,13 +53,29 @@ type Mount struct { // key.parent.fs. key mountKey - // fs, root, and ns are immutable. References are held on fs and root (but - // not ns). - // - // Invariant: root belongs to fs. - fs *Filesystem - root *Dentry - ns *MountNamespace + // ns is the namespace in which this Mount was mounted. ns is protected by + // VirtualFilesystem.mountMu. + ns *MountNamespace + + // The lower 63 bits of refs are a reference count. The MSB of refs is set + // if the Mount has been eagerly umounted, as by umount(2) without the + // MNT_DETACH flag. refs is accessed using atomic memory operations. + refs int64 + + // children is the set of all Mounts for which Mount.key.parent is this + // Mount. children is protected by VirtualFilesystem.mountMu. + children map[*Mount]struct{} + + // umounted is true if VFS.umountRecursiveLocked() has been called on this + // Mount. VirtualFilesystem does not hold a reference on Mounts for which + // umounted is true. umounted is protected by VirtualFilesystem.mountMu. + umounted bool + + // The lower 63 bits of writers is the number of calls to + // Mount.CheckBeginWrite() that have not yet been paired with a call to + // Mount.EndWrite(). The MSB of writers is set if MS_RDONLY is in effect. + // writers is accessed using atomic memory operations. + writers int64 } // A MountNamespace is a collection of Mounts. @@ -73,13 +85,16 @@ type Mount struct { // // MountNamespace is analogous to Linux's struct mnt_namespace. type MountNamespace struct { - refs int64 // accessed using atomic memory operations - // root is the MountNamespace's root mount. root is immutable. root *Mount - // mountpoints contains all Dentries which are mount points in this - // namespace. mountpoints is protected by VirtualFilesystem.mountMu. + // refs is the reference count. refs is accessed using atomic memory + // operations. + refs int64 + + // mountpoints maps all Dentries which are mount points in this namespace + // to the number of Mounts for which they are mount points. mountpoints is + // protected by VirtualFilesystem.mountMu. // // mountpoints is used to determine if a Dentry can be moved or removed // (which requires that the Dentry is not a mount point in the calling @@ -89,7 +104,7 @@ type MountNamespace struct { // MountNamespace; this is required to ensure that // VFS.PrepareDeleteDentry() and VFS.PrepareRemoveDentry() operate // correctly on unreferenced MountNamespaces. - mountpoints map[*Dentry]struct{} + mountpoints map[*Dentry]uint32 } // NewMountNamespace returns a new mount namespace with a root filesystem @@ -106,9 +121,10 @@ func (vfs *VirtualFilesystem) NewMountNamespace(ctx context.Context, creds *auth } mntns := &MountNamespace{ refs: 1, - mountpoints: make(map[*Dentry]struct{}), + mountpoints: make(map[*Dentry]uint32), } mntns.root = &Mount{ + vfs: vfs, fs: fs, root: root, ns: mntns, @@ -136,8 +152,10 @@ func (vfs *VirtualFilesystem) NewMount(ctx context.Context, creds *auth.Credenti return err } vfs.mountMu.Lock() + vd.dentry.mu.Lock() for { if vd.dentry.IsDisowned() { + vd.dentry.mu.Unlock() vfs.mountMu.Unlock() vd.DecRef() root.decRef(fs) @@ -153,36 +171,208 @@ func (vfs *VirtualFilesystem) NewMount(ctx context.Context, creds *auth.Credenti if nextmnt == nil { break } - nextmnt.incRef() + // It's possible that nextmnt has been umounted but not disconnected, + // in which case vfs no longer holds a reference on it, and the last + // reference may be concurrently dropped even though we're holding + // vfs.mountMu. + if !nextmnt.tryIncMountedRef() { + break + } + // This can't fail since we're holding vfs.mountMu. nextmnt.root.incRef(nextmnt.fs) + vd.dentry.mu.Unlock() vd.DecRef() vd = VirtualDentry{ mount: nextmnt, dentry: nextmnt.root, } + vd.dentry.mu.Lock() } // TODO: Linux requires that either both the mount point and the mount root // are directories, or neither are, and returns ENOTDIR if this is not the // case. mntns := vd.mount.ns mnt := &Mount{ + vfs: vfs, fs: fs, root: root, ns: mntns, refs: 1, } - mnt.storeKey(vd.mount, vd.dentry) + vfs.mounts.seq.BeginWrite() + vfs.connectLocked(mnt, vd, mntns) + vfs.mounts.seq.EndWrite() + vd.dentry.mu.Unlock() + vfs.mountMu.Unlock() + return nil +} + +type umountRecursiveOptions struct { + // If eager is true, ensure that future calls to Mount.tryIncMountedRef() + // on umounted mounts fail. + // + // eager is analogous to Linux's UMOUNT_SYNC. + eager bool + + // If disconnectHierarchy is true, Mounts that are umounted hierarchically + // should be disconnected from their parents. (Mounts whose parents are not + // umounted, which in most cases means the Mount passed to the initial call + // to umountRecursiveLocked, are unconditionally disconnected for + // consistency with Linux.) + // + // disconnectHierarchy is analogous to Linux's !UMOUNT_CONNECTED. + disconnectHierarchy bool +} + +// umountRecursiveLocked marks mnt and its descendants as umounted. It does not +// release mount or dentry references; instead, it appends VirtualDentries and +// Mounts on which references must be dropped to vdsToDecRef and mountsToDecRef +// respectively, and returns updated slices. (This is necessary because +// filesystem locks possibly taken by DentryImpl.DecRef() may precede +// vfs.mountMu in the lock order, and Mount.decRef() may lock vfs.mountMu.) +// +// umountRecursiveLocked is analogous to Linux's fs/namespace.c:umount_tree(). +// +// Preconditions: vfs.mountMu must be locked. vfs.mounts.seq must be in a +// writer critical section. +func (vfs *VirtualFilesystem) umountRecursiveLocked(mnt *Mount, opts *umountRecursiveOptions, vdsToDecRef []VirtualDentry, mountsToDecRef []*Mount) ([]VirtualDentry, []*Mount) { + if !mnt.umounted { + mnt.umounted = true + mountsToDecRef = append(mountsToDecRef, mnt) + if parent := mnt.parent(); parent != nil && (opts.disconnectHierarchy || !parent.umounted) { + vdsToDecRef = append(vdsToDecRef, vfs.disconnectLocked(mnt)) + } + } + if opts.eager { + for { + refs := atomic.LoadInt64(&mnt.refs) + if refs < 0 { + break + } + if atomic.CompareAndSwapInt64(&mnt.refs, refs, refs|math.MinInt64) { + break + } + } + } + for child := range mnt.children { + vdsToDecRef, mountsToDecRef = vfs.umountRecursiveLocked(child, opts, vdsToDecRef, mountsToDecRef) + } + return vdsToDecRef, mountsToDecRef +} + +// connectLocked makes vd the mount parent/point for mnt. It consumes +// references held by vd. +// +// Preconditions: vfs.mountMu must be locked. vfs.mounts.seq must be in a +// writer critical section. d.mu must be locked. mnt.parent() == nil. +func (vfs *VirtualFilesystem) connectLocked(mnt *Mount, vd VirtualDentry, mntns *MountNamespace) { + mnt.storeKey(vd) + if vd.mount.children == nil { + vd.mount.children = make(map[*Mount]struct{}) + } + vd.mount.children[mnt] = struct{}{} atomic.AddUint32(&vd.dentry.mounts, 1) - mntns.mountpoints[vd.dentry] = struct{}{} + mntns.mountpoints[vd.dentry]++ + vfs.mounts.insertSeqed(mnt) vfsmpmounts, ok := vfs.mountpoints[vd.dentry] if !ok { vfsmpmounts = make(map[*Mount]struct{}) vfs.mountpoints[vd.dentry] = vfsmpmounts } vfsmpmounts[mnt] = struct{}{} - vfs.mounts.Insert(mnt) - vfs.mountMu.Unlock() - return nil +} + +// disconnectLocked makes vd have no mount parent/point and returns its old +// mount parent/point with a reference held. +// +// Preconditions: vfs.mountMu must be locked. vfs.mounts.seq must be in a +// writer critical section. mnt.parent() != nil. +func (vfs *VirtualFilesystem) disconnectLocked(mnt *Mount) VirtualDentry { + vd := mnt.loadKey() + mnt.storeKey(VirtualDentry{}) + delete(vd.mount.children, mnt) + atomic.AddUint32(&vd.dentry.mounts, math.MaxUint32) // -1 + mnt.ns.mountpoints[vd.dentry]-- + if mnt.ns.mountpoints[vd.dentry] == 0 { + delete(mnt.ns.mountpoints, vd.dentry) + } + vfs.mounts.removeSeqed(mnt) + vfsmpmounts := vfs.mountpoints[vd.dentry] + delete(vfsmpmounts, mnt) + if len(vfsmpmounts) == 0 { + delete(vfs.mountpoints, vd.dentry) + } + return vd +} + +// tryIncMountedRef increments mnt's reference count and returns true. If mnt's +// reference count is already zero, or has been eagerly umounted, +// tryIncMountedRef does nothing and returns false. +// +// tryIncMountedRef does not require that a reference is held on mnt. +func (mnt *Mount) tryIncMountedRef() bool { + for { + refs := atomic.LoadInt64(&mnt.refs) + if refs <= 0 { // refs < 0 => MSB set => eagerly unmounted + return false + } + if atomic.CompareAndSwapInt64(&mnt.refs, refs, refs+1) { + return true + } + } +} + +func (mnt *Mount) incRef() { + // In general, negative values for mnt.refs are valid because the MSB is + // the eager-unmount bit. + atomic.AddInt64(&mnt.refs, 1) +} + +func (mnt *Mount) decRef() { + refs := atomic.AddInt64(&mnt.refs, -1) + if refs&^math.MinInt64 == 0 { // mask out MSB + var vd VirtualDentry + if mnt.parent() != nil { + mnt.vfs.mountMu.Lock() + mnt.vfs.mounts.seq.BeginWrite() + vd = mnt.vfs.disconnectLocked(mnt) + mnt.vfs.mounts.seq.EndWrite() + mnt.vfs.mountMu.Unlock() + } + mnt.root.decRef(mnt.fs) + mnt.fs.decRef() + if vd.Ok() { + vd.DecRef() + } + } +} + +// IncRef increments mntns' reference count. +func (mntns *MountNamespace) IncRef() { + if atomic.AddInt64(&mntns.refs, 1) <= 1 { + panic("MountNamespace.IncRef() called without holding a reference") + } +} + +// DecRef decrements mntns' reference count. +func (mntns *MountNamespace) DecRef(vfs *VirtualFilesystem) { + if refs := atomic.AddInt64(&mntns.refs, -1); refs == 0 { + vfs.mountMu.Lock() + vfs.mounts.seq.BeginWrite() + vdsToDecRef, mountsToDecRef := vfs.umountRecursiveLocked(mntns.root, &umountRecursiveOptions{ + disconnectHierarchy: true, + }, nil, nil) + vfs.mounts.seq.EndWrite() + vfs.mountMu.Unlock() + for _, vd := range vdsToDecRef { + vd.DecRef() + } + for _, mnt := range mountsToDecRef { + mnt.decRef() + } + } else if refs < 0 { + panic("MountNamespace.DecRef() called without holding a reference") + } } // getMountAt returns the last Mount in the stack mounted at (mnt, d). It takes @@ -231,12 +421,12 @@ retryFirst: } // getMountpointAt returns the mount point for the stack of Mounts including -// mnt. It takes a reference on the returned Mount and Dentry. If no such mount +// mnt. It takes a reference on the returned VirtualDentry. If no such mount // point exists (i.e. mnt is a root mount), getMountpointAt returns (nil, nil). // // Preconditions: References are held on mnt and root. vfsroot is not (mnt, // mnt.root). -func (vfs *VirtualFilesystem) getMountpointAt(mnt *Mount, vfsroot VirtualDentry) (*Mount, *Dentry) { +func (vfs *VirtualFilesystem) getMountpointAt(mnt *Mount, vfsroot VirtualDentry) VirtualDentry { // The first mount is special-cased: // // - The caller must have already checked mnt against vfsroot. @@ -246,12 +436,12 @@ func (vfs *VirtualFilesystem) getMountpointAt(mnt *Mount, vfsroot VirtualDentry) // - We don't drop the caller's reference on mnt. retryFirst: epoch := vfs.mounts.seq.BeginRead() - parent, point := mnt.loadKey() + parent, point := mnt.parent(), mnt.point() if !vfs.mounts.seq.ReadOk(epoch) { goto retryFirst } if parent == nil { - return nil, nil + return VirtualDentry{} } if !parent.tryIncMountedRef() { // Raced with umount. @@ -263,6 +453,11 @@ retryFirst: parent.decRef() goto retryFirst } + if !vfs.mounts.seq.ReadOk(epoch) { + point.decRef(parent.fs) + parent.decRef() + goto retryFirst + } mnt = parent d := point for { @@ -274,7 +469,7 @@ retryFirst: } retryNotFirst: epoch := vfs.mounts.seq.BeginRead() - parent, point := mnt.loadKey() + parent, point := mnt.parent(), mnt.point() if !vfs.mounts.seq.ReadOk(epoch) { goto retryNotFirst } @@ -301,43 +496,7 @@ retryFirst: mnt = parent d = point } - return mnt, d -} - -// tryIncMountedRef increments mnt's reference count and returns true. If mnt's -// reference count is already zero, or has been eagerly unmounted, -// tryIncMountedRef does nothing and returns false. -// -// tryIncMountedRef does not require that a reference is held on mnt. -func (mnt *Mount) tryIncMountedRef() bool { - for { - refs := atomic.LoadInt64(&mnt.refs) - if refs <= 0 { // refs < 0 => MSB set => eagerly unmounted - return false - } - if atomic.CompareAndSwapInt64(&mnt.refs, refs, refs+1) { - return true - } - } -} - -func (mnt *Mount) incRef() { - // In general, negative values for mnt.refs are valid because the MSB is - // the eager-unmount bit. - atomic.AddInt64(&mnt.refs, 1) -} - -func (mnt *Mount) decRef() { - refs := atomic.AddInt64(&mnt.refs, -1) - if refs&^math.MinInt64 == 0 { // mask out MSB - parent, point := mnt.loadKey() - if point != nil { - point.decRef(parent.fs) - parent.decRef() - } - mnt.root.decRef(mnt.fs) - mnt.fs.decRef() - } + return VirtualDentry{mnt, d} } // CheckBeginWrite increments the counter of in-progress write operations on @@ -360,7 +519,7 @@ func (mnt *Mount) EndWrite() { atomic.AddInt64(&mnt.writers, -1) } -// Preconditions: VirtualFilesystem.mountMu must be locked for writing. +// Preconditions: VirtualFilesystem.mountMu must be locked. func (mnt *Mount) setReadOnlyLocked(ro bool) error { if oldRO := atomic.LoadInt64(&mnt.writers) < 0; oldRO == ro { return nil @@ -383,22 +542,6 @@ func (mnt *Mount) Filesystem() *Filesystem { return mnt.fs } -// IncRef increments mntns' reference count. -func (mntns *MountNamespace) IncRef() { - if atomic.AddInt64(&mntns.refs, 1) <= 1 { - panic("MountNamespace.IncRef() called without holding a reference") - } -} - -// DecRef decrements mntns' reference count. -func (mntns *MountNamespace) DecRef() { - if refs := atomic.AddInt64(&mntns.refs, 0); refs == 0 { - // TODO: unmount mntns.root - } else if refs < 0 { - panic("MountNamespace.DecRef() called without holding a reference") - } -} - // Root returns mntns' root. A reference is taken on the returned // VirtualDentry. func (mntns *MountNamespace) Root() VirtualDentry { diff --git a/pkg/sentry/vfs/mount_test.go b/pkg/sentry/vfs/mount_test.go index f394d7483..adff0b94b 100644 --- a/pkg/sentry/vfs/mount_test.go +++ b/pkg/sentry/vfs/mount_test.go @@ -37,7 +37,7 @@ func TestMountTableInsertLookup(t *testing.T) { mt.Init() mount := &Mount{} - mount.storeKey(&Mount{}, &Dentry{}) + mount.storeKey(VirtualDentry{&Mount{}, &Dentry{}}) mt.Insert(mount) if m := mt.Lookup(mount.parent(), mount.point()); m != mount { @@ -78,18 +78,10 @@ const enableComparativeBenchmarks = false func newBenchMount() *Mount { mount := &Mount{} - mount.storeKey(&Mount{}, &Dentry{}) + mount.storeKey(VirtualDentry{&Mount{}, &Dentry{}}) return mount } -func vdkey(mnt *Mount) VirtualDentry { - parent, point := mnt.loadKey() - return VirtualDentry{ - mount: parent, - dentry: point, - } -} - func BenchmarkMountTableParallelLookup(b *testing.B) { for numG, maxG := 1, runtime.GOMAXPROCS(0); numG >= 0 && numG <= maxG; numG *= 2 { for _, numMounts := range benchNumMounts { @@ -101,7 +93,7 @@ func BenchmarkMountTableParallelLookup(b *testing.B) { for i := 0; i < numMounts; i++ { mount := newBenchMount() mt.Insert(mount) - keys = append(keys, vdkey(mount)) + keys = append(keys, mount.loadKey()) } var ready sync.WaitGroup @@ -153,7 +145,7 @@ func BenchmarkMountMapParallelLookup(b *testing.B) { keys := make([]VirtualDentry, 0, numMounts) for i := 0; i < numMounts; i++ { mount := newBenchMount() - key := vdkey(mount) + key := mount.loadKey() ms[key] = mount keys = append(keys, key) } @@ -208,7 +200,7 @@ func BenchmarkMountSyncMapParallelLookup(b *testing.B) { keys := make([]VirtualDentry, 0, numMounts) for i := 0; i < numMounts; i++ { mount := newBenchMount() - key := vdkey(mount) + key := mount.loadKey() ms.Store(key, mount) keys = append(keys, key) } @@ -290,7 +282,7 @@ func BenchmarkMountMapNegativeLookup(b *testing.B) { ms := make(map[VirtualDentry]*Mount) for i := 0; i < numMounts; i++ { mount := newBenchMount() - ms[vdkey(mount)] = mount + ms[mount.loadKey()] = mount } negkeys := make([]VirtualDentry, 0, numMounts) for i := 0; i < numMounts; i++ { @@ -325,7 +317,7 @@ func BenchmarkMountSyncMapNegativeLookup(b *testing.B) { var ms sync.Map for i := 0; i < numMounts; i++ { mount := newBenchMount() - ms.Store(vdkey(mount), mount) + ms.Store(mount.loadKey(), mount) } negkeys := make([]VirtualDentry, 0, numMounts) for i := 0; i < numMounts; i++ { @@ -379,7 +371,7 @@ func BenchmarkMountMapInsert(b *testing.B) { b.ResetTimer() for i := range mounts { mount := mounts[i] - ms[vdkey(mount)] = mount + ms[mount.loadKey()] = mount } } @@ -399,7 +391,7 @@ func BenchmarkMountSyncMapInsert(b *testing.B) { b.ResetTimer() for i := range mounts { mount := mounts[i] - ms.Store(vdkey(mount), mount) + ms.Store(mount.loadKey(), mount) } } @@ -432,13 +424,13 @@ func BenchmarkMountMapRemove(b *testing.B) { ms := make(map[VirtualDentry]*Mount) for i := range mounts { mount := mounts[i] - ms[vdkey(mount)] = mount + ms[mount.loadKey()] = mount } b.ResetTimer() for i := range mounts { mount := mounts[i] - delete(ms, vdkey(mount)) + delete(ms, mount.loadKey()) } } @@ -454,12 +446,12 @@ func BenchmarkMountSyncMapRemove(b *testing.B) { var ms sync.Map for i := range mounts { mount := mounts[i] - ms.Store(vdkey(mount), mount) + ms.Store(mount.loadKey(), mount) } b.ResetTimer() for i := range mounts { mount := mounts[i] - ms.Delete(vdkey(mount)) + ms.Delete(mount.loadKey()) } } diff --git a/pkg/sentry/vfs/mount_unsafe.go b/pkg/sentry/vfs/mount_unsafe.go index c98b42f91..ab13fa461 100644 --- a/pkg/sentry/vfs/mount_unsafe.go +++ b/pkg/sentry/vfs/mount_unsafe.go @@ -38,16 +38,6 @@ type mountKey struct { point unsafe.Pointer // *Dentry } -// Invariant: mnt.key's fields are nil. parent and point are non-nil. -func (mnt *Mount) storeKey(parent *Mount, point *Dentry) { - atomic.StorePointer(&mnt.key.parent, unsafe.Pointer(parent)) - atomic.StorePointer(&mnt.key.point, unsafe.Pointer(point)) -} - -func (mnt *Mount) loadKey() (*Mount, *Dentry) { - return (*Mount)(atomic.LoadPointer(&mnt.key.parent)), (*Dentry)(atomic.LoadPointer(&mnt.key.point)) -} - func (mnt *Mount) parent() *Mount { return (*Mount)(atomic.LoadPointer(&mnt.key.parent)) } @@ -56,6 +46,19 @@ func (mnt *Mount) point() *Dentry { return (*Dentry)(atomic.LoadPointer(&mnt.key.point)) } +func (mnt *Mount) loadKey() VirtualDentry { + return VirtualDentry{ + mount: mnt.parent(), + dentry: mnt.point(), + } +} + +// Invariant: mnt.key.parent == nil. vd.Ok(). +func (mnt *Mount) storeKey(vd VirtualDentry) { + atomic.StorePointer(&mnt.key.parent, unsafe.Pointer(vd.mount)) + atomic.StorePointer(&mnt.key.point, unsafe.Pointer(vd.dentry)) +} + // mountTable maps (mount parent, mount point) pairs to mounts. It supports // efficient concurrent lookup, even in the presence of concurrent mutators // (provided mutation is sufficiently uncommon). @@ -201,9 +204,19 @@ loop: // Insert inserts the given mount into mt. // -// Preconditions: There are no concurrent mutators of mt. mt must not already -// contain a Mount with the same mount point and parent. +// Preconditions: mt must not already contain a Mount with the same mount point +// and parent. func (mt *mountTable) Insert(mount *Mount) { + mt.seq.BeginWrite() + mt.insertSeqed(mount) + mt.seq.EndWrite() +} + +// insertSeqed inserts the given mount into mt. +// +// Preconditions: mt.seq must be in a writer critical section. mt must not +// already contain a Mount with the same mount point and parent. +func (mt *mountTable) insertSeqed(mount *Mount) { hash := memhash(unsafe.Pointer(&mount.key), uintptr(mt.seed), mountKeyBytes) // We're under the maximum load factor if: @@ -215,10 +228,8 @@ func (mt *mountTable) Insert(mount *Mount) { tcap := uintptr(1) << order if ((tlen + 1) * mtMaxLoadDen) <= (uint64(mtMaxLoadNum) << order) { // Atomically insert the new element into the table. - mt.seq.BeginWrite() atomic.AddUint64(&mt.size, mtSizeLenOne) mtInsertLocked(mt.slots, tcap, unsafe.Pointer(mount), hash) - mt.seq.EndWrite() return } @@ -241,8 +252,6 @@ func (mt *mountTable) Insert(mount *Mount) { for { oldSlot := (*mountSlot)(oldCur) if oldSlot.value != nil { - // Don't need to lock mt.seq yet since newSlots isn't visible - // to readers. mtInsertLocked(newSlots, newCap, oldSlot.value, oldSlot.hash) } if oldCur == oldLast { @@ -252,11 +261,9 @@ func (mt *mountTable) Insert(mount *Mount) { } // Insert the new element into the new table. mtInsertLocked(newSlots, newCap, unsafe.Pointer(mount), hash) - // Atomically switch to the new table. - mt.seq.BeginWrite() + // Switch to the new table. atomic.AddUint64(&mt.size, mtSizeLenOne|mtSizeOrderOne) atomic.StorePointer(&mt.slots, newSlots) - mt.seq.EndWrite() } // Preconditions: There are no concurrent mutators of the table (slots, cap). @@ -294,9 +301,18 @@ func mtInsertLocked(slots unsafe.Pointer, cap uintptr, value unsafe.Pointer, has // Remove removes the given mount from mt. // -// Preconditions: There are no concurrent mutators of mt. mt must contain -// mount. +// Preconditions: mt must contain mount. func (mt *mountTable) Remove(mount *Mount) { + mt.seq.BeginWrite() + mt.removeSeqed(mount) + mt.seq.EndWrite() +} + +// removeSeqed removes the given mount from mt. +// +// Preconditions: mt.seq must be in a writer critical section. mt must contain +// mount. +func (mt *mountTable) removeSeqed(mount *Mount) { hash := memhash(unsafe.Pointer(&mount.key), uintptr(mt.seed), mountKeyBytes) tcap := uintptr(1) << (mt.size & mtSizeOrderMask) mask := tcap - 1 @@ -311,7 +327,6 @@ func (mt *mountTable) Remove(mount *Mount) { // backward until we either find an empty slot, or an element that // is already in its first-probed slot. (This is backward shift // deletion.) - mt.seq.BeginWrite() for { nextOff := (off + mountSlotBytes) & offmask nextSlot := (*mountSlot)(unsafe.Pointer(uintptr(slots) + nextOff)) @@ -330,7 +345,6 @@ func (mt *mountTable) Remove(mount *Mount) { } atomic.StorePointer(&slot.value, nil) atomic.AddUint64(&mt.size, mtSizeLenNegOne) - mt.seq.EndWrite() return } if checkInvariants && slotValue == nil { diff --git a/pkg/sentry/vfs/resolving_path.go b/pkg/sentry/vfs/resolving_path.go index 8d05c8583..61bce6426 100644 --- a/pkg/sentry/vfs/resolving_path.go +++ b/pkg/sentry/vfs/resolving_path.go @@ -269,11 +269,11 @@ func (rp *ResolvingPath) ResolveParent(d *Dentry) (*Dentry, error) { parent = d } else if d == rp.mount.root { // At mount root ... - mnt, mntpt := rp.vfs.getMountpointAt(rp.mount, rp.root) - if mnt != nil { + vd := rp.vfs.getMountpointAt(rp.mount, rp.root) + if vd.Ok() { // ... of non-root mount. - rp.nextMount = mnt - rp.nextStart = mntpt + rp.nextMount = vd.mount + rp.nextStart = vd.dentry return nil, resolveMountRootError{} } // ... of root mount. diff --git a/pkg/sentry/vfs/syscalls.go b/pkg/sentry/vfs/syscalls.go index abde0feaa..49952b2cc 100644 --- a/pkg/sentry/vfs/syscalls.go +++ b/pkg/sentry/vfs/syscalls.go @@ -230,6 +230,8 @@ func (fd *FileDescription) SetStatusFlags(ctx context.Context, flags uint32) err // // - VFS.SymlinkAt() // +// - VFS.UmountAt() +// // - VFS.UnlinkAt() // // - FileDescription.(almost everything) diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 4a8a69540..88e865d86 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -16,9 +16,14 @@ // // Lock order: // -// Filesystem implementation locks +// FilesystemImpl/FileDescriptionImpl locks // VirtualFilesystem.mountMu +// Dentry.mu +// Locks acquired by FilesystemImpls between Prepare{Delete,Rename}Dentry and Commit{Delete,Rename*}Dentry // VirtualFilesystem.fsTypesMu +// +// Locking Dentry.mu in multiple Dentries requires holding +// VirtualFilesystem.mountMu. package vfs import ( @@ -33,7 +38,7 @@ type VirtualFilesystem struct { // mountMu serializes mount mutations. // // mountMu is analogous to Linux's namespace_sem. - mountMu sync.RWMutex + mountMu sync.Mutex // mounts maps (mount parent, mount point) pairs to mounts. (Since mounts // are uniquely namespaced, including mount parent in the key correctly @@ -52,7 +57,7 @@ type VirtualFilesystem struct { // mountpoints maps mount points to mounts at those points in all // namespaces. mountpoints is protected by mountMu. // - // mountpoints is used to find mounts that must be unmounted due to + // mountpoints is used to find mounts that must be umounted due to // removal of a mount point Dentry from another mount namespace. ("A file // or directory that is a mount point in one namespace that is not a mount // point in another namespace, may be renamed, unlinked, or removed -- cgit v1.2.3 From b72e1b3c0873ea29d031db42e39ca053923eecff Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Mon, 25 Nov 2019 18:09:15 -0800 Subject: Minor VFS2 interface changes. - Remove the Filesystem argument from DentryImpl.*Ref(); in general DentryImpls that need the Filesystem for reference counting will probably also need it for other interface methods that don't plumb Filesystem, so it's easier to just store a pointer to the filesystem in the DentryImpl. - Add a pointer to the VirtualFilesystem to Filesystem, which is needed by the gofer client to disown dentries for cache eviction triggered by dentry reference count changes. - Rename FilesystemType.NewFilesystem to GetFilesystem; in some cases (e.g. sysfs, cgroupfs) it's much cleaner for there to be only one Filesystem that is used by all mounts, and in at least one case (devtmpfs) it's visibly incorrect not to do so, so NewFilesystem doesn't always actually create and return a *new* Filesystem. - Require callers of FileDescription.Init() to increment Mount/Dentry references. This is because the gofer client may, in the OpenAt() path, take a reference on a dentry with 0 references, which is safe due to synchronization that is outside the scope of this CL, and it would be safer to still have its implementation of DentryImpl.IncRef() check for an increment for 0 references in other cases. - Add FileDescription.TryIncRef. This is used by the gofer client to take references on "special file descriptions" (FDs for files such as pipes, sockets, and devices), which use per-FD handles (fids) instead of dentry-shared handles, for sync() and syncfs(). PiperOrigin-RevId: 282473364 --- pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go | 4 +- pkg/sentry/fsimpl/ext/block_map_file.go | 4 +- pkg/sentry/fsimpl/ext/block_map_test.go | 4 +- pkg/sentry/fsimpl/ext/dentry.go | 10 ++-- pkg/sentry/fsimpl/ext/ext.go | 10 ++-- pkg/sentry/fsimpl/ext/ext_test.go | 14 +++--- pkg/sentry/fsimpl/ext/extent_file.go | 6 +-- pkg/sentry/fsimpl/ext/extent_test.go | 4 +- pkg/sentry/fsimpl/ext/file_description.go | 4 +- pkg/sentry/fsimpl/ext/inode.go | 28 +++++++----- pkg/sentry/fsimpl/memfs/benchmark_test.go | 6 +-- pkg/sentry/fsimpl/memfs/filesystem.go | 15 ++++-- pkg/sentry/fsimpl/memfs/memfs.go | 16 +++---- pkg/sentry/fsimpl/memfs/named_pipe.go | 5 +- pkg/sentry/fsimpl/memfs/pipe_test.go | 2 +- pkg/sentry/vfs/dentry.go | 24 ++++++---- pkg/sentry/vfs/file_description.go | 33 +++++++++++-- pkg/sentry/vfs/file_description_impl_util_test.go | 2 +- pkg/sentry/vfs/filesystem.go | 20 ++++++-- pkg/sentry/vfs/filesystem_type.go | 10 ++-- pkg/sentry/vfs/mount.go | 56 ++++++++++++----------- pkg/sentry/vfs/resolving_path.go | 8 ++-- pkg/sentry/vfs/syscalls.go | 2 +- pkg/sentry/vfs/testutil.go | 12 ++--- pkg/sentry/vfs/vfs.go | 8 ++-- 25 files changed, 186 insertions(+), 121 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go index 10a8083a0..94cd74095 100644 --- a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go +++ b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go @@ -50,7 +50,7 @@ func setUp(b *testing.B, imagePath string) (context.Context, *vfs.VirtualFilesys // Create VFS. vfsObj := vfs.New() vfsObj.MustRegisterFilesystemType("extfs", ext.FilesystemType{}) - mntns, err := vfsObj.NewMountNamespace(ctx, creds, imagePath, "extfs", &vfs.NewFilesystemOptions{InternalData: int(f.Fd())}) + mntns, err := vfsObj.NewMountNamespace(ctx, creds, imagePath, "extfs", &vfs.GetFilesystemOptions{InternalData: int(f.Fd())}) if err != nil { f.Close() return nil, nil, nil, nil, err @@ -81,7 +81,7 @@ func mount(b *testing.B, imagePath string, vfsfs *vfs.VirtualFilesystem, pop *vf ctx := contexttest.Context(b) creds := auth.CredentialsFromContext(ctx) - if err := vfsfs.NewMount(ctx, creds, imagePath, pop, "extfs", &vfs.NewFilesystemOptions{InternalData: int(f.Fd())}); err != nil { + if err := vfsfs.NewMount(ctx, creds, imagePath, pop, "extfs", &vfs.GetFilesystemOptions{InternalData: int(f.Fd())}); err != nil { b.Fatalf("failed to mount tmpfs submount: %v", err) } return func() { diff --git a/pkg/sentry/fsimpl/ext/block_map_file.go b/pkg/sentry/fsimpl/ext/block_map_file.go index cea89bcd9..a2d8c3ad6 100644 --- a/pkg/sentry/fsimpl/ext/block_map_file.go +++ b/pkg/sentry/fsimpl/ext/block_map_file.go @@ -154,7 +154,7 @@ func (f *blockMapFile) read(curPhyBlk uint32, relFileOff uint64, height uint, ds toRead = len(dst) } - n, _ := f.regFile.inode.dev.ReadAt(dst[:toRead], curPhyBlkOff+int64(relFileOff)) + n, _ := f.regFile.inode.fs.dev.ReadAt(dst[:toRead], curPhyBlkOff+int64(relFileOff)) if n < toRead { return n, syserror.EIO } @@ -174,7 +174,7 @@ func (f *blockMapFile) read(curPhyBlk uint32, relFileOff uint64, height uint, ds curChildOff := relFileOff % childCov for i := startIdx; i < endIdx; i++ { var childPhyBlk uint32 - err := readFromDisk(f.regFile.inode.dev, curPhyBlkOff+int64(i*4), &childPhyBlk) + err := readFromDisk(f.regFile.inode.fs.dev, curPhyBlkOff+int64(i*4), &childPhyBlk) if err != nil { return read, err } diff --git a/pkg/sentry/fsimpl/ext/block_map_test.go b/pkg/sentry/fsimpl/ext/block_map_test.go index 213aa3919..181727ef7 100644 --- a/pkg/sentry/fsimpl/ext/block_map_test.go +++ b/pkg/sentry/fsimpl/ext/block_map_test.go @@ -87,12 +87,14 @@ func blockMapSetUp(t *testing.T) (*blockMapFile, []byte) { mockDisk := make([]byte, mockBMDiskSize) regFile := regularFile{ inode: inode{ + fs: &filesystem{ + dev: bytes.NewReader(mockDisk), + }, diskInode: &disklayout.InodeNew{ InodeOld: disklayout.InodeOld{ SizeLo: getMockBMFileFize(), }, }, - dev: bytes.NewReader(mockDisk), blkSize: uint64(mockBMBlkSize), }, } diff --git a/pkg/sentry/fsimpl/ext/dentry.go b/pkg/sentry/fsimpl/ext/dentry.go index 054fb42b6..a080cb189 100644 --- a/pkg/sentry/fsimpl/ext/dentry.go +++ b/pkg/sentry/fsimpl/ext/dentry.go @@ -41,16 +41,18 @@ func newDentry(in *inode) *dentry { } // IncRef implements vfs.DentryImpl.IncRef. -func (d *dentry) IncRef(vfsfs *vfs.Filesystem) { +func (d *dentry) IncRef() { d.inode.incRef() } // TryIncRef implements vfs.DentryImpl.TryIncRef. -func (d *dentry) TryIncRef(vfsfs *vfs.Filesystem) bool { +func (d *dentry) TryIncRef() bool { return d.inode.tryIncRef() } // DecRef implements vfs.DentryImpl.DecRef. -func (d *dentry) DecRef(vfsfs *vfs.Filesystem) { - d.inode.decRef(vfsfs.Impl().(*filesystem)) +func (d *dentry) DecRef() { + // FIXME(b/134676337): filesystem.mu may not be locked as required by + // inode.decRef(). + d.inode.decRef() } diff --git a/pkg/sentry/fsimpl/ext/ext.go b/pkg/sentry/fsimpl/ext/ext.go index f10accafc..4b7d17dc6 100644 --- a/pkg/sentry/fsimpl/ext/ext.go +++ b/pkg/sentry/fsimpl/ext/ext.go @@ -40,14 +40,14 @@ var _ vfs.FilesystemType = (*FilesystemType)(nil) // Currently there are two ways of mounting an ext(2/3/4) fs: // 1. Specify a mount with our internal special MountType in the OCI spec. // 2. Expose the device to the container and mount it from application layer. -func getDeviceFd(source string, opts vfs.NewFilesystemOptions) (io.ReaderAt, error) { +func getDeviceFd(source string, opts vfs.GetFilesystemOptions) (io.ReaderAt, error) { if opts.InternalData == nil { // User mount call. // TODO(b/134676337): Open the device specified by `source` and return that. panic("unimplemented") } - // NewFilesystem call originated from within the sentry. + // GetFilesystem call originated from within the sentry. devFd, ok := opts.InternalData.(int) if !ok { return nil, errors.New("internal data for ext fs must be an int containing the file descriptor to device") @@ -91,8 +91,8 @@ func isCompatible(sb disklayout.SuperBlock) bool { return true } -// NewFilesystem implements vfs.FilesystemType.NewFilesystem. -func (FilesystemType) NewFilesystem(ctx context.Context, creds *auth.Credentials, source string, opts vfs.NewFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { +// GetFilesystem implements vfs.FilesystemType.GetFilesystem. +func (FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { // TODO(b/134676337): Ensure that the user is mounting readonly. If not, // EACCESS should be returned according to mount(2). Filesystem independent // flags (like readonly) are currently not available in pkg/sentry/vfs. @@ -103,7 +103,7 @@ func (FilesystemType) NewFilesystem(ctx context.Context, creds *auth.Credentials } fs := filesystem{dev: dev, inodeCache: make(map[uint32]*inode)} - fs.vfsfs.Init(&fs) + fs.vfsfs.Init(vfsObj, &fs) fs.sb, err = readSuperBlock(dev) if err != nil { return nil, nil, err diff --git a/pkg/sentry/fsimpl/ext/ext_test.go b/pkg/sentry/fsimpl/ext/ext_test.go index 1aa2bd6a4..307e4d68c 100644 --- a/pkg/sentry/fsimpl/ext/ext_test.go +++ b/pkg/sentry/fsimpl/ext/ext_test.go @@ -66,7 +66,7 @@ func setUp(t *testing.T, imagePath string) (context.Context, *vfs.VirtualFilesys // Create VFS. vfsObj := vfs.New() vfsObj.MustRegisterFilesystemType("extfs", FilesystemType{}) - mntns, err := vfsObj.NewMountNamespace(ctx, creds, localImagePath, "extfs", &vfs.NewFilesystemOptions{InternalData: int(f.Fd())}) + mntns, err := vfsObj.NewMountNamespace(ctx, creds, localImagePath, "extfs", &vfs.GetFilesystemOptions{InternalData: int(f.Fd())}) if err != nil { f.Close() return nil, nil, nil, nil, err @@ -509,27 +509,27 @@ func TestIterDirents(t *testing.T) { } wantDirents := []vfs.Dirent{ - vfs.Dirent{ + { Name: ".", Type: linux.DT_DIR, }, - vfs.Dirent{ + { Name: "..", Type: linux.DT_DIR, }, - vfs.Dirent{ + { Name: "lost+found", Type: linux.DT_DIR, }, - vfs.Dirent{ + { Name: "file.txt", Type: linux.DT_REG, }, - vfs.Dirent{ + { Name: "bigfile.txt", Type: linux.DT_REG, }, - vfs.Dirent{ + { Name: "symlink.txt", Type: linux.DT_LNK, }, diff --git a/pkg/sentry/fsimpl/ext/extent_file.go b/pkg/sentry/fsimpl/ext/extent_file.go index 38b68a2d3..3d3ebaca6 100644 --- a/pkg/sentry/fsimpl/ext/extent_file.go +++ b/pkg/sentry/fsimpl/ext/extent_file.go @@ -99,7 +99,7 @@ func (f *extentFile) buildExtTree() error { func (f *extentFile) buildExtTreeFromDisk(entry disklayout.ExtentEntry) (*disklayout.ExtentNode, error) { var header disklayout.ExtentHeader off := entry.PhysicalBlock() * f.regFile.inode.blkSize - err := readFromDisk(f.regFile.inode.dev, int64(off), &header) + err := readFromDisk(f.regFile.inode.fs.dev, int64(off), &header) if err != nil { return nil, err } @@ -115,7 +115,7 @@ func (f *extentFile) buildExtTreeFromDisk(entry disklayout.ExtentEntry) (*diskla curEntry = &disklayout.ExtentIdx{} } - err := readFromDisk(f.regFile.inode.dev, int64(off), curEntry) + err := readFromDisk(f.regFile.inode.fs.dev, int64(off), curEntry) if err != nil { return nil, err } @@ -229,7 +229,7 @@ func (f *extentFile) readFromExtent(ex *disklayout.Extent, off uint64, dst []byt toRead = len(dst) } - n, _ := f.regFile.inode.dev.ReadAt(dst[:toRead], int64(readStart)) + n, _ := f.regFile.inode.fs.dev.ReadAt(dst[:toRead], int64(readStart)) if n < toRead { return n, syserror.EIO } diff --git a/pkg/sentry/fsimpl/ext/extent_test.go b/pkg/sentry/fsimpl/ext/extent_test.go index 42d0a484b..a2382daa3 100644 --- a/pkg/sentry/fsimpl/ext/extent_test.go +++ b/pkg/sentry/fsimpl/ext/extent_test.go @@ -180,13 +180,15 @@ func extentTreeSetUp(t *testing.T, root *disklayout.ExtentNode) (*extentFile, [] mockExtentFile := &extentFile{ regFile: regularFile{ inode: inode{ + fs: &filesystem{ + dev: bytes.NewReader(mockDisk), + }, diskInode: &disklayout.InodeNew{ InodeOld: disklayout.InodeOld{ SizeLo: uint32(mockExtentBlkSize) * getNumPhyBlks(root), }, }, blkSize: mockExtentBlkSize, - dev: bytes.NewReader(mockDisk), }, }, } diff --git a/pkg/sentry/fsimpl/ext/file_description.go b/pkg/sentry/fsimpl/ext/file_description.go index 4d18b28cb..5eca2b83f 100644 --- a/pkg/sentry/fsimpl/ext/file_description.go +++ b/pkg/sentry/fsimpl/ext/file_description.go @@ -36,11 +36,11 @@ type fileDescription struct { } func (fd *fileDescription) filesystem() *filesystem { - return fd.vfsfd.VirtualDentry().Mount().Filesystem().Impl().(*filesystem) + return fd.vfsfd.Mount().Filesystem().Impl().(*filesystem) } func (fd *fileDescription) inode() *inode { - return fd.vfsfd.VirtualDentry().Dentry().Impl().(*dentry).inode + return fd.vfsfd.Dentry().Impl().(*dentry).inode } // StatusFlags implements vfs.FileDescriptionImpl.StatusFlags. diff --git a/pkg/sentry/fsimpl/ext/inode.go b/pkg/sentry/fsimpl/ext/inode.go index e6c847a71..24249525c 100644 --- a/pkg/sentry/fsimpl/ext/inode.go +++ b/pkg/sentry/fsimpl/ext/inode.go @@ -16,7 +16,6 @@ package ext import ( "fmt" - "io" "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" @@ -42,13 +41,13 @@ type inode struct { // refs is a reference count. refs is accessed using atomic memory operations. refs int64 + // fs is the containing filesystem. + fs *filesystem + // inodeNum is the inode number of this inode on disk. This is used to // identify inodes within the ext filesystem. inodeNum uint32 - // dev represents the underlying device. Same as filesystem.dev. - dev io.ReaderAt - // blkSize is the fs data block size. Same as filesystem.sb.BlockSize(). blkSize uint64 @@ -81,10 +80,10 @@ func (in *inode) tryIncRef() bool { // decRef decrements the inode ref count and releases the inode resources if // the ref count hits 0. // -// Precondition: Must have locked fs.mu. -func (in *inode) decRef(fs *filesystem) { +// Precondition: Must have locked filesystem.mu. +func (in *inode) decRef() { if refs := atomic.AddInt64(&in.refs, -1); refs == 0 { - delete(fs.inodeCache, in.inodeNum) + delete(in.fs.inodeCache, in.inodeNum) } else if refs < 0 { panic("ext.inode.decRef() called without holding a reference") } @@ -117,8 +116,8 @@ func newInode(fs *filesystem, inodeNum uint32) (*inode, error) { // Build the inode based on its type. inode := inode{ + fs: fs, inodeNum: inodeNum, - dev: fs.dev, blkSize: blkSize, diskInode: diskInode, } @@ -154,11 +153,14 @@ func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*v if err := in.checkPermissions(rp.Credentials(), ats); err != nil { return nil, err } + mnt := rp.Mount() switch in.impl.(type) { case *regularFile: var fd regularFileFD fd.flags = flags - fd.vfsfd.Init(&fd, rp.Mount(), vfsd) + mnt.IncRef() + vfsd.IncRef() + fd.vfsfd.Init(&fd, mnt, vfsd) return &fd.vfsfd, nil case *directory: // Can't open directories writably. This check is not necessary for a read @@ -167,8 +169,10 @@ func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*v return nil, syserror.EISDIR } var fd directoryFD - fd.vfsfd.Init(&fd, rp.Mount(), vfsd) fd.flags = flags + mnt.IncRef() + vfsd.IncRef() + fd.vfsfd.Init(&fd, mnt, vfsd) return &fd.vfsfd, nil case *symlink: if flags&linux.O_PATH == 0 { @@ -177,7 +181,9 @@ func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*v } var fd symlinkFD fd.flags = flags - fd.vfsfd.Init(&fd, rp.Mount(), vfsd) + mnt.IncRef() + vfsd.IncRef() + fd.vfsfd.Init(&fd, mnt, vfsd) return &fd.vfsfd, nil default: panic(fmt.Sprintf("unknown inode type: %T", in.impl)) diff --git a/pkg/sentry/fsimpl/memfs/benchmark_test.go b/pkg/sentry/fsimpl/memfs/benchmark_test.go index 23a846c08..ea6417ce7 100644 --- a/pkg/sentry/fsimpl/memfs/benchmark_test.go +++ b/pkg/sentry/fsimpl/memfs/benchmark_test.go @@ -176,7 +176,7 @@ func BenchmarkVFS2MemfsStat(b *testing.B) { // Create VFS. vfsObj := vfs.New() vfsObj.MustRegisterFilesystemType("memfs", memfs.FilesystemType{}) - mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "memfs", &vfs.NewFilesystemOptions{}) + mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "memfs", &vfs.GetFilesystemOptions{}) if err != nil { b.Fatalf("failed to create tmpfs root mount: %v", err) } @@ -365,7 +365,7 @@ func BenchmarkVFS2MemfsMountStat(b *testing.B) { // Create VFS. vfsObj := vfs.New() vfsObj.MustRegisterFilesystemType("memfs", memfs.FilesystemType{}) - mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "memfs", &vfs.NewFilesystemOptions{}) + mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "memfs", &vfs.GetFilesystemOptions{}) if err != nil { b.Fatalf("failed to create tmpfs root mount: %v", err) } @@ -394,7 +394,7 @@ func BenchmarkVFS2MemfsMountStat(b *testing.B) { } defer mountPoint.DecRef() // Create and mount the submount. - if err := vfsObj.NewMount(ctx, creds, "", &pop, "memfs", &vfs.NewFilesystemOptions{}); err != nil { + if err := vfsObj.NewMount(ctx, creds, "", &pop, "memfs", &vfs.GetFilesystemOptions{}); err != nil { b.Fatalf("failed to mount tmpfs submount: %v", err) } filePathBuilder.WriteString(mountPointName) diff --git a/pkg/sentry/fsimpl/memfs/filesystem.go b/pkg/sentry/fsimpl/memfs/filesystem.go index f006c40cd..08a9cb8ef 100644 --- a/pkg/sentry/fsimpl/memfs/filesystem.go +++ b/pkg/sentry/fsimpl/memfs/filesystem.go @@ -159,7 +159,7 @@ func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, op return nil, err } } - inode.incRef() // vfsd.IncRef(&fs.vfsfs) + inode.incRef() return vfsd, nil } @@ -379,6 +379,7 @@ func (i *inode) open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentr return nil, err } } + mnt := rp.Mount() switch impl := i.impl.(type) { case *regularFile: var fd regularFileFD @@ -386,12 +387,14 @@ func (i *inode) open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentr fd.readable = vfs.MayReadFileWithOpenFlags(flags) fd.writable = vfs.MayWriteFileWithOpenFlags(flags) if fd.writable { - if err := rp.Mount().CheckBeginWrite(); err != nil { + if err := mnt.CheckBeginWrite(); err != nil { return nil, err } - // Mount.EndWrite() is called by regularFileFD.Release(). + // mnt.EndWrite() is called by regularFileFD.Release(). } - fd.vfsfd.Init(&fd, rp.Mount(), vfsd) + mnt.IncRef() + vfsd.IncRef() + fd.vfsfd.Init(&fd, mnt, vfsd) if flags&linux.O_TRUNC != 0 { impl.mu.Lock() impl.data = impl.data[:0] @@ -405,7 +408,9 @@ func (i *inode) open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentr return nil, syserror.EISDIR } var fd directoryFD - fd.vfsfd.Init(&fd, rp.Mount(), vfsd) + mnt.IncRef() + vfsd.IncRef() + fd.vfsfd.Init(&fd, mnt, vfsd) fd.flags = flags return &fd.vfsfd, nil case *symlink: diff --git a/pkg/sentry/fsimpl/memfs/memfs.go b/pkg/sentry/fsimpl/memfs/memfs.go index 64c851c1a..4cb2a4e0f 100644 --- a/pkg/sentry/fsimpl/memfs/memfs.go +++ b/pkg/sentry/fsimpl/memfs/memfs.go @@ -52,10 +52,10 @@ type filesystem struct { nextInoMinusOne uint64 // accessed using atomic memory operations } -// NewFilesystem implements vfs.FilesystemType.NewFilesystem. -func (fstype FilesystemType) NewFilesystem(ctx context.Context, creds *auth.Credentials, source string, opts vfs.NewFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { +// GetFilesystem implements vfs.FilesystemType.GetFilesystem. +func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { var fs filesystem - fs.vfsfs.Init(&fs) + fs.vfsfs.Init(vfsObj, &fs) root := fs.newDentry(fs.newDirectory(creds, 01777)) return &fs.vfsfs, &root.vfsd, nil } @@ -99,17 +99,17 @@ func (fs *filesystem) newDentry(inode *inode) *dentry { } // IncRef implements vfs.DentryImpl.IncRef. -func (d *dentry) IncRef(vfsfs *vfs.Filesystem) { +func (d *dentry) IncRef() { d.inode.incRef() } // TryIncRef implements vfs.DentryImpl.TryIncRef. -func (d *dentry) TryIncRef(vfsfs *vfs.Filesystem) bool { +func (d *dentry) TryIncRef() bool { return d.inode.tryIncRef() } // DecRef implements vfs.DentryImpl.DecRef. -func (d *dentry) DecRef(vfsfs *vfs.Filesystem) { +func (d *dentry) DecRef() { d.inode.decRef() } @@ -266,11 +266,11 @@ type fileDescription struct { } func (fd *fileDescription) filesystem() *filesystem { - return fd.vfsfd.VirtualDentry().Mount().Filesystem().Impl().(*filesystem) + return fd.vfsfd.Mount().Filesystem().Impl().(*filesystem) } func (fd *fileDescription) inode() *inode { - return fd.vfsfd.VirtualDentry().Dentry().Impl().(*dentry).inode + return fd.vfsfd.Dentry().Impl().(*dentry).inode } // StatusFlags implements vfs.FileDescriptionImpl.StatusFlags. diff --git a/pkg/sentry/fsimpl/memfs/named_pipe.go b/pkg/sentry/fsimpl/memfs/named_pipe.go index 732ed7c58..91cb4b1fc 100644 --- a/pkg/sentry/fsimpl/memfs/named_pipe.go +++ b/pkg/sentry/fsimpl/memfs/named_pipe.go @@ -54,6 +54,9 @@ func newNamedPipeFD(ctx context.Context, np *namedPipe, rp *vfs.ResolvingPath, v if err != nil { return nil, err } - fd.vfsfd.Init(&fd, rp.Mount(), vfsd) + mnt := rp.Mount() + mnt.IncRef() + vfsd.IncRef() + fd.vfsfd.Init(&fd, mnt, vfsd) return &fd.vfsfd, nil } diff --git a/pkg/sentry/fsimpl/memfs/pipe_test.go b/pkg/sentry/fsimpl/memfs/pipe_test.go index 0674b81a3..a3a870571 100644 --- a/pkg/sentry/fsimpl/memfs/pipe_test.go +++ b/pkg/sentry/fsimpl/memfs/pipe_test.go @@ -152,7 +152,7 @@ func setup(t *testing.T) (context.Context, *auth.Credentials, *vfs.VirtualFilesy // Create VFS. vfsObj := vfs.New() vfsObj.MustRegisterFilesystemType("memfs", FilesystemType{}) - mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "memfs", &vfs.NewFilesystemOptions{}) + mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "memfs", &vfs.GetFilesystemOptions{}) if err != nil { t.Fatalf("failed to create tmpfs root mount: %v", err) } diff --git a/pkg/sentry/vfs/dentry.go b/pkg/sentry/vfs/dentry.go index 09ed5a70e..40f4c1d09 100644 --- a/pkg/sentry/vfs/dentry.go +++ b/pkg/sentry/vfs/dentry.go @@ -118,7 +118,7 @@ func (d *Dentry) Impl() DentryImpl { type DentryImpl interface { // IncRef increments the Dentry's reference count. A Dentry with a non-zero // reference count must remain coherent with the state of the filesystem. - IncRef(fs *Filesystem) + IncRef() // TryIncRef increments the Dentry's reference count and returns true. If // the Dentry's reference count is zero, TryIncRef may do nothing and @@ -126,10 +126,10 @@ type DentryImpl interface { // guarantee that the Dentry is coherent with the state of the filesystem.) // // TryIncRef does not require that a reference is held on the Dentry. - TryIncRef(fs *Filesystem) bool + TryIncRef() bool // DecRef decrements the Dentry's reference count. - DecRef(fs *Filesystem) + DecRef() } // IsDisowned returns true if d is disowned. @@ -146,16 +146,20 @@ func (d *Dentry) isMounted() bool { return atomic.LoadUint32(&d.mounts) != 0 } -func (d *Dentry) incRef(fs *Filesystem) { - d.impl.IncRef(fs) +// IncRef increments d's reference count. +func (d *Dentry) IncRef() { + d.impl.IncRef() } -func (d *Dentry) tryIncRef(fs *Filesystem) bool { - return d.impl.TryIncRef(fs) +// TryIncRef increments d's reference count and returns true. If d's reference +// count is zero, TryIncRef may instead do nothing and return false. +func (d *Dentry) TryIncRef() bool { + return d.impl.TryIncRef() } -func (d *Dentry) decRef(fs *Filesystem) { - d.impl.DecRef(fs) +// DecRef decrements d's reference count. +func (d *Dentry) DecRef() { + d.impl.DecRef() } // These functions are exported so that filesystem implementations can use @@ -420,6 +424,6 @@ func (vfs *VirtualFilesystem) forgetDisownedMountpoint(d *Dentry) { vd.DecRef() } for _, mnt := range mountsToDecRef { - mnt.decRef() + mnt.DecRef() } } diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 3a9665800..34007eb57 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -47,15 +47,14 @@ type FileDescription struct { impl FileDescriptionImpl } -// Init must be called before first use of fd. It takes references on mnt and -// d. +// Init must be called before first use of fd. It takes ownership of references +// on mnt and d held by the caller. func (fd *FileDescription) Init(impl FileDescriptionImpl, mnt *Mount, d *Dentry) { fd.refs = 1 fd.vd = VirtualDentry{ mount: mnt, dentry: d, } - fd.vd.IncRef() fd.impl = impl } @@ -64,6 +63,18 @@ func (fd *FileDescription) Impl() FileDescriptionImpl { return fd.impl } +// Mount returns the mount on which fd was opened. It does not take a reference +// on the returned Mount. +func (fd *FileDescription) Mount() *Mount { + return fd.vd.mount +} + +// Dentry returns the dentry at which fd was opened. It does not take a +// reference on the returned Dentry. +func (fd *FileDescription) Dentry() *Dentry { + return fd.vd.dentry +} + // VirtualDentry returns the location at which fd was opened. It does not take // a reference on the returned VirtualDentry. func (fd *FileDescription) VirtualDentry() VirtualDentry { @@ -75,6 +86,22 @@ func (fd *FileDescription) IncRef() { atomic.AddInt64(&fd.refs, 1) } +// TryIncRef increments fd's reference count and returns true. If fd's +// reference count is already zero, TryIncRef does nothing and returns false. +// +// TryIncRef does not require that a reference is held on fd. +func (fd *FileDescription) TryIncRef() bool { + for { + refs := atomic.LoadInt64(&fd.refs) + if refs <= 0 { + return false + } + if atomic.CompareAndSwapInt64(&fd.refs, refs, refs+1) { + return true + } + } +} + // DecRef decrements fd's reference count. func (fd *FileDescription) DecRef() { if refs := atomic.AddInt64(&fd.refs, -1); refs == 0 { diff --git a/pkg/sentry/vfs/file_description_impl_util_test.go b/pkg/sentry/vfs/file_description_impl_util_test.go index 511b829fc..a5561dcbe 100644 --- a/pkg/sentry/vfs/file_description_impl_util_test.go +++ b/pkg/sentry/vfs/file_description_impl_util_test.go @@ -90,7 +90,7 @@ func TestGenCountFD(t *testing.T) { vfsObj := New() // vfs.New() vfsObj.MustRegisterFilesystemType("testfs", FDTestFilesystemType{}) - mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "testfs", &NewFilesystemOptions{}) + mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "testfs", &GetFilesystemOptions{}) if err != nil { t.Fatalf("failed to create testfs root mount: %v", err) } diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index 7a074b718..76ff8cf51 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -33,29 +33,41 @@ type Filesystem struct { // operations. refs int64 + // vfs is the VirtualFilesystem that uses this Filesystem. vfs is + // immutable. + vfs *VirtualFilesystem + // impl is the FilesystemImpl associated with this Filesystem. impl is // immutable. This should be the last field in Dentry. impl FilesystemImpl } // Init must be called before first use of fs. -func (fs *Filesystem) Init(impl FilesystemImpl) { +func (fs *Filesystem) Init(vfsObj *VirtualFilesystem, impl FilesystemImpl) { fs.refs = 1 + fs.vfs = vfsObj fs.impl = impl } +// VirtualFilesystem returns the containing VirtualFilesystem. +func (fs *Filesystem) VirtualFilesystem() *VirtualFilesystem { + return fs.vfs +} + // Impl returns the FilesystemImpl associated with fs. func (fs *Filesystem) Impl() FilesystemImpl { return fs.impl } -func (fs *Filesystem) incRef() { +// IncRef increments fs' reference count. +func (fs *Filesystem) IncRef() { if atomic.AddInt64(&fs.refs, 1) <= 1 { - panic("Filesystem.incRef() called without holding a reference") + panic("Filesystem.IncRef() called without holding a reference") } } -func (fs *Filesystem) decRef() { +// DecRef decrements fs' reference count. +func (fs *Filesystem) DecRef() { if refs := atomic.AddInt64(&fs.refs, -1); refs == 0 { fs.impl.Release() } else if refs < 0 { diff --git a/pkg/sentry/vfs/filesystem_type.go b/pkg/sentry/vfs/filesystem_type.go index f401ad7f3..c335e206d 100644 --- a/pkg/sentry/vfs/filesystem_type.go +++ b/pkg/sentry/vfs/filesystem_type.go @@ -25,21 +25,21 @@ import ( // // FilesystemType is analogous to Linux's struct file_system_type. type FilesystemType interface { - // NewFilesystem returns a Filesystem configured by the given options, + // GetFilesystem returns a Filesystem configured by the given options, // along with its mount root. A reference is taken on the returned // Filesystem and Dentry. - NewFilesystem(ctx context.Context, creds *auth.Credentials, source string, opts NewFilesystemOptions) (*Filesystem, *Dentry, error) + GetFilesystem(ctx context.Context, vfsObj *VirtualFilesystem, creds *auth.Credentials, source string, opts GetFilesystemOptions) (*Filesystem, *Dentry, error) } -// NewFilesystemOptions contains options to FilesystemType.NewFilesystem. -type NewFilesystemOptions struct { +// GetFilesystemOptions contains options to FilesystemType.GetFilesystem. +type GetFilesystemOptions struct { // Data is the string passed as the 5th argument to mount(2), which is // usually a comma-separated list of filesystem-specific mount options. Data string // InternalData holds opaque FilesystemType-specific data. There is // intentionally no way for applications to specify InternalData; if it is - // not nil, the call to NewFilesystem originates from within the sentry. + // not nil, the call to GetFilesystem originates from within the sentry. InternalData interface{} } diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index 198fb8067..1c3b2e987 100644 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go @@ -110,12 +110,12 @@ type MountNamespace struct { // NewMountNamespace returns a new mount namespace with a root filesystem // configured by the given arguments. A reference is taken on the returned // MountNamespace. -func (vfs *VirtualFilesystem) NewMountNamespace(ctx context.Context, creds *auth.Credentials, source, fsTypeName string, opts *NewFilesystemOptions) (*MountNamespace, error) { +func (vfs *VirtualFilesystem) NewMountNamespace(ctx context.Context, creds *auth.Credentials, source, fsTypeName string, opts *GetFilesystemOptions) (*MountNamespace, error) { fsType := vfs.getFilesystemType(fsTypeName) if fsType == nil { return nil, syserror.ENODEV } - fs, root, err := fsType.NewFilesystem(ctx, creds, source, *opts) + fs, root, err := fsType.GetFilesystem(ctx, vfs, creds, source, *opts) if err != nil { return nil, err } @@ -133,13 +133,13 @@ func (vfs *VirtualFilesystem) NewMountNamespace(ctx context.Context, creds *auth return mntns, nil } -// NewMount creates and mounts a new Filesystem. -func (vfs *VirtualFilesystem) NewMount(ctx context.Context, creds *auth.Credentials, source string, target *PathOperation, fsTypeName string, opts *NewFilesystemOptions) error { +// NewMount creates and mounts a Filesystem configured by the given arguments. +func (vfs *VirtualFilesystem) NewMount(ctx context.Context, creds *auth.Credentials, source string, target *PathOperation, fsTypeName string, opts *GetFilesystemOptions) error { fsType := vfs.getFilesystemType(fsTypeName) if fsType == nil { return syserror.ENODEV } - fs, root, err := fsType.NewFilesystem(ctx, creds, source, *opts) + fs, root, err := fsType.GetFilesystem(ctx, vfs, creds, source, *opts) if err != nil { return err } @@ -147,8 +147,8 @@ func (vfs *VirtualFilesystem) NewMount(ctx context.Context, creds *auth.Credenti // lock ordering. vd, err := vfs.GetDentryAt(ctx, creds, target, &GetDentryOptions{}) if err != nil { - root.decRef(fs) - fs.decRef() + root.DecRef() + fs.DecRef() return err } vfs.mountMu.Lock() @@ -158,8 +158,8 @@ func (vfs *VirtualFilesystem) NewMount(ctx context.Context, creds *auth.Credenti vd.dentry.mu.Unlock() vfs.mountMu.Unlock() vd.DecRef() - root.decRef(fs) - fs.decRef() + root.DecRef() + fs.DecRef() return syserror.ENOENT } // vd might have been mounted over between vfs.GetDentryAt() and @@ -179,7 +179,7 @@ func (vfs *VirtualFilesystem) NewMount(ctx context.Context, creds *auth.Credenti break } // This can't fail since we're holding vfs.mountMu. - nextmnt.root.incRef(nextmnt.fs) + nextmnt.root.IncRef() vd.dentry.mu.Unlock() vd.DecRef() vd = VirtualDentry{ @@ -229,7 +229,7 @@ type umountRecursiveOptions struct { // Mounts on which references must be dropped to vdsToDecRef and mountsToDecRef // respectively, and returns updated slices. (This is necessary because // filesystem locks possibly taken by DentryImpl.DecRef() may precede -// vfs.mountMu in the lock order, and Mount.decRef() may lock vfs.mountMu.) +// vfs.mountMu in the lock order, and Mount.DecRef() may lock vfs.mountMu.) // // umountRecursiveLocked is analogous to Linux's fs/namespace.c:umount_tree(). // @@ -322,13 +322,15 @@ func (mnt *Mount) tryIncMountedRef() bool { } } -func (mnt *Mount) incRef() { +// IncRef increments mnt's reference count. +func (mnt *Mount) IncRef() { // In general, negative values for mnt.refs are valid because the MSB is // the eager-unmount bit. atomic.AddInt64(&mnt.refs, 1) } -func (mnt *Mount) decRef() { +// DecRef decrements mnt's reference count. +func (mnt *Mount) DecRef() { refs := atomic.AddInt64(&mnt.refs, -1) if refs&^math.MinInt64 == 0 { // mask out MSB var vd VirtualDentry @@ -339,8 +341,8 @@ func (mnt *Mount) decRef() { mnt.vfs.mounts.seq.EndWrite() mnt.vfs.mountMu.Unlock() } - mnt.root.decRef(mnt.fs) - mnt.fs.decRef() + mnt.root.DecRef() + mnt.fs.DecRef() if vd.Ok() { vd.DecRef() } @@ -368,7 +370,7 @@ func (mntns *MountNamespace) DecRef(vfs *VirtualFilesystem) { vd.DecRef() } for _, mnt := range mountsToDecRef { - mnt.decRef() + mnt.DecRef() } } else if refs < 0 { panic("MountNamespace.DecRef() called without holding a reference") @@ -413,7 +415,7 @@ retryFirst: // Raced with umount. continue } - mnt.decRef() + mnt.DecRef() mnt = next d = next.root } @@ -447,15 +449,15 @@ retryFirst: // Raced with umount. goto retryFirst } - if !point.tryIncRef(parent.fs) { + if !point.TryIncRef() { // Since Mount holds a reference on Mount.key.point, this can only // happen due to a racing change to Mount.key. - parent.decRef() + parent.DecRef() goto retryFirst } if !vfs.mounts.seq.ReadOk(epoch) { - point.decRef(parent.fs) - parent.decRef() + point.DecRef() + parent.DecRef() goto retryFirst } mnt = parent @@ -480,19 +482,19 @@ retryFirst: // Raced with umount. goto retryNotFirst } - if !point.tryIncRef(parent.fs) { + if !point.TryIncRef() { // Since Mount holds a reference on Mount.key.point, this can // only happen due to a racing change to Mount.key. - parent.decRef() + parent.DecRef() goto retryNotFirst } if !vfs.mounts.seq.ReadOk(epoch) { - point.decRef(parent.fs) - parent.decRef() + point.DecRef() + parent.DecRef() goto retryNotFirst } - d.decRef(mnt.fs) - mnt.decRef() + d.DecRef() + mnt.DecRef() mnt = parent d = point } diff --git a/pkg/sentry/vfs/resolving_path.go b/pkg/sentry/vfs/resolving_path.go index 61bce6426..621f5a6f8 100644 --- a/pkg/sentry/vfs/resolving_path.go +++ b/pkg/sentry/vfs/resolving_path.go @@ -149,20 +149,20 @@ func (vfs *VirtualFilesystem) putResolvingPath(rp *ResolvingPath) { func (rp *ResolvingPath) decRefStartAndMount() { if rp.flags&rpflagsHaveStartRef != 0 { - rp.start.decRef(rp.mount.fs) + rp.start.DecRef() } if rp.flags&rpflagsHaveMountRef != 0 { - rp.mount.decRef() + rp.mount.DecRef() } } func (rp *ResolvingPath) releaseErrorState() { if rp.nextStart != nil { - rp.nextStart.decRef(rp.nextMount.fs) + rp.nextStart.DecRef() rp.nextStart = nil } if rp.nextMount != nil { - rp.nextMount.decRef() + rp.nextMount.DecRef() rp.nextMount = nil } } diff --git a/pkg/sentry/vfs/syscalls.go b/pkg/sentry/vfs/syscalls.go index 49952b2cc..436151afa 100644 --- a/pkg/sentry/vfs/syscalls.go +++ b/pkg/sentry/vfs/syscalls.go @@ -63,7 +63,7 @@ func (vfs *VirtualFilesystem) GetDentryAt(ctx context.Context, creds *auth.Crede mount: rp.mount, dentry: d, } - rp.mount.incRef() + rp.mount.IncRef() vfs.putResolvingPath(rp) return vd, nil } diff --git a/pkg/sentry/vfs/testutil.go b/pkg/sentry/vfs/testutil.go index 70b192ece..593144cb7 100644 --- a/pkg/sentry/vfs/testutil.go +++ b/pkg/sentry/vfs/testutil.go @@ -33,10 +33,10 @@ type FDTestFilesystem struct { vfsfs Filesystem } -// NewFilesystem implements FilesystemType.NewFilesystem. -func (fstype FDTestFilesystemType) NewFilesystem(ctx context.Context, creds *auth.Credentials, source string, opts NewFilesystemOptions) (*Filesystem, *Dentry, error) { +// GetFilesystem implements FilesystemType.GetFilesystem. +func (fstype FDTestFilesystemType) GetFilesystem(ctx context.Context, vfsObj *VirtualFilesystem, creds *auth.Credentials, source string, opts GetFilesystemOptions) (*Filesystem, *Dentry, error) { var fs FDTestFilesystem - fs.vfsfs.Init(&fs) + fs.vfsfs.Init(vfsObj, &fs) return &fs.vfsfs, fs.NewDentry(), nil } @@ -126,14 +126,14 @@ func (fs *FDTestFilesystem) NewDentry() *Dentry { } // IncRef implements DentryImpl.IncRef. -func (d *fdTestDentry) IncRef(vfsfs *Filesystem) { +func (d *fdTestDentry) IncRef() { } // TryIncRef implements DentryImpl.TryIncRef. -func (d *fdTestDentry) TryIncRef(vfsfs *Filesystem) bool { +func (d *fdTestDentry) TryIncRef() bool { return true } // DecRef implements DentryImpl.DecRef. -func (d *fdTestDentry) DecRef(vfsfs *Filesystem) { +func (d *fdTestDentry) DecRef() { } diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 88e865d86..f0cd3ffe5 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -116,15 +116,15 @@ func (vd VirtualDentry) Ok() bool { // IncRef increments the reference counts on the Mount and Dentry represented // by vd. func (vd VirtualDentry) IncRef() { - vd.mount.incRef() - vd.dentry.incRef(vd.mount.fs) + vd.mount.IncRef() + vd.dentry.IncRef() } // DecRef decrements the reference counts on the Mount and Dentry represented // by vd. func (vd VirtualDentry) DecRef() { - vd.dentry.decRef(vd.mount.fs) - vd.mount.decRef() + vd.dentry.DecRef() + vd.mount.DecRef() } // Mount returns the Mount associated with vd. It does not take a reference on -- cgit v1.2.3 From e91c1675cd49254936d04f01b814a0cd802ff6de Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Tue, 26 Nov 2019 15:32:45 +0800 Subject: passed the kvm test case of "TestKernelFault" on Arm64 platform Signed-off-by: Bin Lu --- pkg/sentry/platform/ring0/entry_arm64.s | 10 ++++++++++ 1 file changed, 10 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/ring0/entry_arm64.s b/pkg/sentry/platform/ring0/entry_arm64.s index 29c475882..0ba4c6b73 100644 --- a/pkg/sentry/platform/ring0/entry_arm64.s +++ b/pkg/sentry/platform/ring0/entry_arm64.s @@ -385,6 +385,16 @@ TEXT ·El1_sync(SB),NOSPLIT,$0 B el1_invalid el1_da: + WORD $0xd538d092 //MRS TPIDR_EL1, R18 + WORD $0xd538601a //MRS FAR_EL1, R26 + + MOVD R26, CPU_FAULT_ADDR(RSV_REG) + + MOVD $0, CPU_ERROR_TYPE(RSV_REG) + + MOVD $PageFault, R3 + MOVD R3, CPU_VECTOR_CODE(RSV_REG) + B ·Halt(SB) el1_ia: -- cgit v1.2.3 From e710f654015d2d35d8a8df07ef54ddbc442cbc9d Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Tue, 26 Nov 2019 15:53:26 +0800 Subject: Prepare the vcpu environment for a container application Mainly 2 jobs were finished in this patch: 1, context switching for a container application: a, R0-R30 b, pc\pstate\sp_el0 c, pagetable_el0 for container application This job can help us to pass the following test cases: "TestApplicationSyscall", "TestApplicationFault" 2, checking pagetable_el0 is empty This job can help us to pass the following test case: "TestInvalidate" Signed-off-by: Bin Lu --- pkg/sentry/platform/ring0/entry_arm64.s | 67 +++++++++++++++++++++++++++++++++ 1 file changed, 67 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/ring0/entry_arm64.s b/pkg/sentry/platform/ring0/entry_arm64.s index 29c475882..674c569e1 100644 --- a/pkg/sentry/platform/ring0/entry_arm64.s +++ b/pkg/sentry/platform/ring0/entry_arm64.s @@ -337,6 +337,73 @@ TEXT ·Current(SB),NOSPLIT,$0-8 #define STACK_FRAME_SIZE 16 TEXT ·kernelExitToEl0(SB),NOSPLIT,$0 + // Step1, save sentry context into memory. + REGISTERS_SAVE(RSV_REG, CPU_REGISTERS) + MOVD RSV_REG_APP, CPU_REGISTERS+PTRACE_R9(RSV_REG) + + WORD $0xd5384003 // MRS SPSR_EL1, R3 + MOVD R3, CPU_REGISTERS+PTRACE_PSTATE(RSV_REG) + MOVD R30, CPU_REGISTERS+PTRACE_PC(RSV_REG) + MOVD RSP, R3 + MOVD R3, CPU_REGISTERS+PTRACE_SP(RSV_REG) + + MOVD CPU_REGISTERS+PTRACE_R3(RSV_REG), R3 + + // Step2, save SP_EL1, PSTATE into kernel temporary stack. + // switch to temporary stack. + LOAD_KERNEL_STACK(RSV_REG) + WORD $0xd538d092 //MRS TPIDR_EL1, R18 + + SUB $STACK_FRAME_SIZE, RSP, RSP + MOVD CPU_REGISTERS+PTRACE_SP(RSV_REG), R11 + MOVD CPU_REGISTERS+PTRACE_PSTATE(RSV_REG), R12 + STP (R11, R12), 16*0(RSP) + + MOVD CPU_REGISTERS+PTRACE_R11(RSV_REG), R11 + MOVD CPU_REGISTERS+PTRACE_R12(RSV_REG), R12 + + // Step3, test user pagetable. + // If user pagetable is empty, trapped in el1_ia. + WORD $0xd538d092 //MRS TPIDR_EL1, R18 + SWITCH_TO_APP_PAGETABLE(RSV_REG) + WORD $0xd538d092 //MRS TPIDR_EL1, R18 + SWITCH_TO_KVM_PAGETABLE(RSV_REG) + WORD $0xd538d092 //MRS TPIDR_EL1, R18 + + // If pagetable is not empty, recovery kernel temporary stack. + ADD $STACK_FRAME_SIZE, RSP, RSP + + // Step4, load app context pointer. + MOVD CPU_APP_ADDR(RSV_REG), RSV_REG_APP + + // Step5, prepare the environment for container application. + // set sp_el0. + MOVD PTRACE_SP(RSV_REG_APP), R1 + WORD $0xd5184101 //MSR R1, SP_EL0 + // set pc. + MOVD PTRACE_PC(RSV_REG_APP), R1 + MSR R1, ELR_EL1 + // set pstate. + MOVD PTRACE_PSTATE(RSV_REG_APP), R1 + WORD $0xd5184001 //MSR R1, SPSR_EL1 + + // RSV_REG & RSV_REG_APP will be loaded at the end. + REGISTERS_LOAD(RSV_REG_APP, 0) + + // switch to user pagetable. + MOVD PTRACE_R18(RSV_REG_APP), RSV_REG + MOVD PTRACE_R9(RSV_REG_APP), RSV_REG_APP + + SUB $STACK_FRAME_SIZE, RSP, RSP + STP (RSV_REG, RSV_REG_APP), 16*0(RSP) + + WORD $0xd538d092 //MRS TPIDR_EL1, R18 + + SWITCH_TO_APP_PAGETABLE(RSV_REG) + + LDP 16*0(RSP), (RSV_REG, RSV_REG_APP) + ADD $STACK_FRAME_SIZE, RSP, RSP + ERET() TEXT ·kernelExitToEl1(SB),NOSPLIT,$0 -- cgit v1.2.3 From 3e0062480650ded910ef6bc80883668da89e2ef8 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Tue, 26 Nov 2019 15:57:07 +0800 Subject: passed the kvm test case of "TestApplicationSyscall" on Arm64 platform For test case "TestApplicationSyscall", Syscall in guest user level will be trapped in el0_svc. And in el0_svc, we use mmio_exit to leave the KVM guest for now. Signed-off-by: Bin Lu --- pkg/sentry/platform/ring0/entry_arm64.s | 10 ++++++++++ 1 file changed, 10 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/ring0/entry_arm64.s b/pkg/sentry/platform/ring0/entry_arm64.s index 29c475882..bd9f09469 100644 --- a/pkg/sentry/platform/ring0/entry_arm64.s +++ b/pkg/sentry/platform/ring0/entry_arm64.s @@ -441,6 +441,16 @@ TEXT ·El0_sync(SB),NOSPLIT,$0 B el0_invalid el0_svc: + WORD $0xd538d092 //MRS TPIDR_EL1, R18 + + MOVD $0, CPU_ERROR_CODE(RSV_REG) // Clear error code. + + MOVD $1, R3 + MOVD R3, CPU_ERROR_TYPE(RSV_REG) // Set error type to user. + + MOVD $Syscall, R3 + MOVD R3, CPU_VECTOR_CODE(RSV_REG) + B ·Halt(SB) el0_da: -- cgit v1.2.3 From 3f0e91b00450f926d8378ebd518b557c1f273712 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Tue, 26 Nov 2019 15:59:51 +0800 Subject: passed the kvm test case of "TestApplicationFault" on Arm64 platform For test case "TestApplicationFault", Memory-fault in guest user level will be trapped in el0_da. And in el0_da, we use mmio_exit to leave the KVM guest. Signed-off-by: Bin Lu --- pkg/sentry/platform/ring0/entry_arm64.s | 11 +++++++++++ 1 file changed, 11 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/ring0/entry_arm64.s b/pkg/sentry/platform/ring0/entry_arm64.s index 29c475882..7b7420785 100644 --- a/pkg/sentry/platform/ring0/entry_arm64.s +++ b/pkg/sentry/platform/ring0/entry_arm64.s @@ -444,6 +444,17 @@ el0_svc: B ·Halt(SB) el0_da: + WORD $0xd538d092 //MRS TPIDR_EL1, R18 + WORD $0xd538601a //MRS FAR_EL1, R26 + + MOVD R26, CPU_FAULT_ADDR(RSV_REG) + + MOVD $1, R3 + MOVD R3, CPU_ERROR_TYPE(RSV_REG) // Set error type to user. + + MOVD $PageFault, R3 + MOVD R3, CPU_VECTOR_CODE(RSV_REG) + B ·Halt(SB) el0_ia: -- cgit v1.2.3 From 519ceabdf90129664fa1f70f49d0472a9106910f Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 26 Nov 2019 17:01:56 -0800 Subject: Mark execveat as supported for linux64_arm64. PiperOrigin-RevId: 282667122 --- pkg/sentry/syscalls/linux/linux64_arm64.go | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/linux64_arm64.go b/pkg/sentry/syscalls/linux/linux64_arm64.go index a809115e0..f1dd4b0c0 100644 --- a/pkg/sentry/syscalls/linux/linux64_arm64.go +++ b/pkg/sentry/syscalls/linux/linux64_arm64.go @@ -295,7 +295,7 @@ var ARM64 = &kernel.SyscallTable{ 278: syscalls.Supported("getrandom", GetRandom), 279: syscalls.Supported("memfd_create", MemfdCreate), 280: syscalls.CapError("bpf", linux.CAP_SYS_ADMIN, "", nil), - 281: syscalls.ErrorWithEvent("execveat", syserror.ENOSYS, "", []string{"gvisor.dev/issue/265"}), // TODO(b/118901836) + 281: syscalls.Supported("execveat", Execveat), 282: syscalls.ErrorWithEvent("userfaultfd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/266"}), // TODO(b/118906345) 283: syscalls.ErrorWithEvent("membarrier", syserror.ENOSYS, "", []string{"gvisor.dev/issue/267"}), // TODO(gvisor.dev/issue/267) 284: syscalls.PartiallySupported("mlock2", Mlock2, "Stub implementation. The sandbox lacks appropriate permissions.", nil), -- cgit v1.2.3 From 20279c305ece6a458006999c8dafc5672ca92803 Mon Sep 17 00:00:00 2001 From: Ian Lewis Date: Tue, 26 Nov 2019 18:19:47 -0800 Subject: Allow open(O_TRUNC) and (f)truncate for proc files. This allows writable proc and devices files to be opened with O_CREAT|O_TRUNC. This is encountered most frequently when interacting with proc or devices files via the command line. e.g. $ echo 8192 1048576 4194304 > /proc/sys/net/ipv4/tcp_rmem Also adds a test to test the behavior of open(O_TRUNC), truncate, and ftruncate on named pipes. Fixes #1116 PiperOrigin-RevId: 282677425 --- pkg/sentry/fs/proc/sys_net.go | 17 ++++++++++++++--- pkg/sentry/fs/tty/master.go | 6 +++++- pkg/sentry/fs/tty/slave.go | 6 +++++- pkg/sentry/syscalls/linux/sys_file.go | 12 ++++++++++-- test/syscalls/linux/pipe.cc | 14 ++++++++++++++ 5 files changed, 48 insertions(+), 7 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/proc/sys_net.go b/pkg/sentry/fs/proc/sys_net.go index f3b63dfc2..bd93f83fa 100644 --- a/pkg/sentry/fs/proc/sys_net.go +++ b/pkg/sentry/fs/proc/sys_net.go @@ -64,7 +64,7 @@ var _ fs.InodeOperations = (*tcpMemInode)(nil) func newTCPMemInode(ctx context.Context, msrc *fs.MountSource, s inet.Stack, dir tcpMemDir) *fs.Inode { tm := &tcpMemInode{ - SimpleFileInode: *fsutil.NewSimpleFileInode(ctx, fs.RootOwner, fs.FilePermsFromMode(0444), linux.PROC_SUPER_MAGIC), + SimpleFileInode: *fsutil.NewSimpleFileInode(ctx, fs.RootOwner, fs.FilePermsFromMode(0644), linux.PROC_SUPER_MAGIC), s: s, dir: dir, } @@ -77,6 +77,11 @@ func newTCPMemInode(ctx context.Context, msrc *fs.MountSource, s inet.Stack, dir return fs.NewInode(ctx, tm, msrc, sattr) } +// Truncate implements fs.InodeOperations.Truncate. +func (tcpMemInode) Truncate(context.Context, *fs.Inode, int64) error { + return nil +} + // GetFile implements fs.InodeOperations.GetFile. func (m *tcpMemInode) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { flags.Pread = true @@ -168,14 +173,15 @@ func writeSize(dirType tcpMemDir, s inet.Stack, size inet.TCPBufferSize) error { // +stateify savable type tcpSack struct { + fsutil.SimpleFileInode + stack inet.Stack `state:"wait"` enabled *bool - fsutil.SimpleFileInode } func newTCPSackInode(ctx context.Context, msrc *fs.MountSource, s inet.Stack) *fs.Inode { ts := &tcpSack{ - SimpleFileInode: *fsutil.NewSimpleFileInode(ctx, fs.RootOwner, fs.FilePermsFromMode(0444), linux.PROC_SUPER_MAGIC), + SimpleFileInode: *fsutil.NewSimpleFileInode(ctx, fs.RootOwner, fs.FilePermsFromMode(0644), linux.PROC_SUPER_MAGIC), stack: s, } sattr := fs.StableAttr{ @@ -187,6 +193,11 @@ func newTCPSackInode(ctx context.Context, msrc *fs.MountSource, s inet.Stack) *f return fs.NewInode(ctx, ts, msrc, sattr) } +// Truncate implements fs.InodeOperations.Truncate. +func (tcpSack) Truncate(context.Context, *fs.Inode, int64) error { + return nil +} + // GetFile implements fs.InodeOperations.GetFile. func (s *tcpSack) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { flags.Pread = true diff --git a/pkg/sentry/fs/tty/master.go b/pkg/sentry/fs/tty/master.go index bc56be696..934828c12 100644 --- a/pkg/sentry/fs/tty/master.go +++ b/pkg/sentry/fs/tty/master.go @@ -32,7 +32,6 @@ import ( // +stateify savable type masterInodeOperations struct { fsutil.SimpleFileInode - fsutil.InodeNoopTruncate // d is the containing dir. d *dirInodeOperations @@ -77,6 +76,11 @@ func newMasterInode(ctx context.Context, d *dirInodeOperations, owner fs.FileOwn func (mi *masterInodeOperations) Release(ctx context.Context) { } +// Truncate implements fs.InodeOperations.Truncate. +func (masterInodeOperations) Truncate(context.Context, *fs.Inode, int64) error { + return nil +} + // GetFile implements fs.InodeOperations.GetFile. // // It allocates a new terminal. diff --git a/pkg/sentry/fs/tty/slave.go b/pkg/sentry/fs/tty/slave.go index 4cbea0367..2a51e6bab 100644 --- a/pkg/sentry/fs/tty/slave.go +++ b/pkg/sentry/fs/tty/slave.go @@ -31,7 +31,6 @@ import ( // +stateify savable type slaveInodeOperations struct { fsutil.SimpleFileInode - fsutil.InodeNoopTruncate // d is the containing dir. d *dirInodeOperations @@ -73,6 +72,11 @@ func (si *slaveInodeOperations) Release(ctx context.Context) { si.t.DecRef() } +// Truncate implements fs.InodeOperations.Truncate. +func (slaveInodeOperations) Truncate(context.Context, *fs.Inode, int64) error { + return nil +} + // GetFile implements fs.InodeOperations.GetFile. // // This may race with destruction of the terminal. If the terminal is gone, it diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go index 167c2b60b..3b9181002 100644 --- a/pkg/sentry/syscalls/linux/sys_file.go +++ b/pkg/sentry/syscalls/linux/sys_file.go @@ -171,6 +171,9 @@ func openAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint) (fd uint } } + // Truncate is called when O_TRUNC is specified for any kind of + // existing Dirent. Behavior is delegated to the entry's Truncate + // implementation. if flags&linux.O_TRUNC != 0 { if err := d.Inode.Truncate(t, d, 0); err != nil { return err @@ -397,7 +400,9 @@ func createAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint, mode l return err } - // Should we truncate the file? + // Truncate is called when O_TRUNC is specified for any kind of + // existing Dirent. Behavior is delegated to the entry's Truncate + // implementation. if flags&linux.O_TRUNC != 0 { if err := found.Inode.Truncate(t, found, 0); err != nil { return err @@ -1484,6 +1489,8 @@ func Truncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if fs.IsDir(d.Inode.StableAttr) { return syserror.EISDIR } + // In contrast to open(O_TRUNC), truncate(2) is only valid for file + // types. if !fs.IsFile(d.Inode.StableAttr) { return syserror.EINVAL } @@ -1522,7 +1529,8 @@ func Ftruncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys return 0, nil, syserror.EINVAL } - // Note that this is different from truncate(2) above, where a + // In contrast to open(O_TRUNC), truncate(2) is only valid for file + // types. Note that this is different from truncate(2) above, where a // directory returns EISDIR. if !fs.IsFile(file.Dirent.Inode.StableAttr) { return 0, nil, syserror.EINVAL diff --git a/test/syscalls/linux/pipe.cc b/test/syscalls/linux/pipe.cc index c0b354e65..ac9b21b24 100644 --- a/test/syscalls/linux/pipe.cc +++ b/test/syscalls/linux/pipe.cc @@ -212,6 +212,20 @@ TEST(Pipe2Test, BadOptions) { EXPECT_THAT(pipe2(fds, 0xDEAD), SyscallFailsWithErrno(EINVAL)); } +// Tests that opening named pipes with O_TRUNC shouldn't cause an error, but +// calls to (f)truncate should. +TEST(NamedPipeTest, Truncate) { + const std::string tmp_path = NewTempAbsPath(); + SKIP_IF(mkfifo(tmp_path.c_str(), 0644) != 0); + + ASSERT_THAT(open(tmp_path.c_str(), O_NONBLOCK | O_RDONLY), SyscallSucceeds()); + FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE( + Open(tmp_path.c_str(), O_RDWR | O_NONBLOCK | O_TRUNC)); + + ASSERT_THAT(truncate(tmp_path.c_str(), 0), SyscallFailsWithErrno(EINVAL)); + ASSERT_THAT(ftruncate(fd.get(), 0), SyscallFailsWithErrno(EINVAL)); +} + TEST_P(PipeTest, Seek) { SKIP_IF(!CreateBlocking()); -- cgit v1.2.3 From 684f757a228f88e5fabe6ebe6ed54f0db20fd63d Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Wed, 27 Nov 2019 16:19:35 -0800 Subject: Add support for receiving TOS and TCLASS control messages in hostinet. This involves allowing getsockopt/setsockopt for the corresponding socket options, as well as allowing hostinet to process control messages received from the actual recvmsg syscall. PiperOrigin-RevId: 282851425 --- pkg/abi/linux/socket.go | 9 ++++ pkg/sentry/socket/control/control.go | 24 ++++++++- pkg/sentry/socket/hostinet/BUILD | 1 + pkg/sentry/socket/hostinet/socket.go | 92 ++++++++++++++++++++++++--------- pkg/sentry/syscalls/linux/sys_socket.go | 8 +++ pkg/tcpip/tcpip.go | 14 ++++- runsc/boot/filter/config.go | 29 +++++++++-- 7 files changed, 145 insertions(+), 32 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/socket.go b/pkg/abi/linux/socket.go index 2e2cc6be7..766ee4014 100644 --- a/pkg/abi/linux/socket.go +++ b/pkg/abi/linux/socket.go @@ -422,6 +422,15 @@ type ControlMessageRights []int32 // ControlMessageRights. const SizeOfControlMessageRight = 4 +// SizeOfControlMessageInq is the size of a TCP_INQ control message. +const SizeOfControlMessageInq = 4 + +// SizeOfControlMessageTOS is the size of an IP_TOS control message. +const SizeOfControlMessageTOS = 1 + +// SizeOfControlMessageTClass is the size of an IPV6_TCLASS control message. +const SizeOfControlMessageTClass = 4 + // SCM_MAX_FD is the maximum number of FDs accepted in a single sendmsg call. // From net/scm.h. const SCM_MAX_FD = 253 diff --git a/pkg/sentry/socket/control/control.go b/pkg/sentry/socket/control/control.go index 4e95101b7..0371acede 100644 --- a/pkg/sentry/socket/control/control.go +++ b/pkg/sentry/socket/control/control.go @@ -320,11 +320,33 @@ func PackInq(t *kernel.Task, inq int32, buf []byte) []byte { buf, linux.SOL_TCP, linux.TCP_INQ, - 4, + t.Arch().Width(), inq, ) } +// PackTOS packs an IP_TOS socket control message. +func PackTOS(t *kernel.Task, tos int8, buf []byte) []byte { + return putCmsgStruct( + buf, + linux.SOL_IP, + linux.IP_TOS, + t.Arch().Width(), + tos, + ) +} + +// PackTClass packs an IPV6_TCLASS socket control message. +func PackTClass(t *kernel.Task, tClass int32, buf []byte) []byte { + return putCmsgStruct( + buf, + linux.SOL_IPV6, + linux.IPV6_TCLASS, + t.Arch().Width(), + tClass, + ) +} + // Parse parses a raw socket control message into portable objects. func Parse(t *kernel.Task, socketOrEndpoint interface{}, buf []byte) (transport.ControlMessages, error) { var ( diff --git a/pkg/sentry/socket/hostinet/BUILD b/pkg/sentry/socket/hostinet/BUILD index 8b66a719d..b1cf1126f 100644 --- a/pkg/sentry/socket/hostinet/BUILD +++ b/pkg/sentry/socket/hostinet/BUILD @@ -34,5 +34,6 @@ go_library( "//pkg/syserror", "//pkg/tcpip/stack", "//pkg/waiter", + "@org_golang_x_sys//unix:go_default_library", ], ) diff --git a/pkg/sentry/socket/hostinet/socket.go b/pkg/sentry/socket/hostinet/socket.go index 92beb1bcf..aa234f760 100644 --- a/pkg/sentry/socket/hostinet/socket.go +++ b/pkg/sentry/socket/hostinet/socket.go @@ -18,6 +18,7 @@ import ( "fmt" "syscall" + "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/fdnotifier" @@ -41,6 +42,10 @@ const ( // sizeofSockaddr is the size in bytes of the largest sockaddr type // supported by this package. sizeofSockaddr = syscall.SizeofSockaddrInet6 // sizeof(sockaddr_in6) > sizeof(sockaddr_in) + + // maxControlLen is the maximum size of a control message buffer used in a + // recvmsg syscall. + maxControlLen = 1024 ) // socketOperations implements fs.FileOperations and socket.Socket for a socket @@ -281,26 +286,32 @@ func (s *socketOperations) GetSockOpt(t *kernel.Task, level int, name int, outPt // Whitelist options and constrain option length. var optlen int switch level { - case syscall.SOL_IPV6: + case linux.SOL_IP: + switch name { + case linux.IP_RECVTOS: + optlen = sizeofInt32 + } + case linux.SOL_IPV6: switch name { - case syscall.IPV6_V6ONLY: + case linux.IPV6_RECVTCLASS, linux.IPV6_V6ONLY: optlen = sizeofInt32 } - case syscall.SOL_SOCKET: + case linux.SOL_SOCKET: switch name { - case syscall.SO_ERROR, syscall.SO_KEEPALIVE, syscall.SO_SNDBUF, syscall.SO_RCVBUF, syscall.SO_REUSEADDR: + case linux.SO_ERROR, linux.SO_KEEPALIVE, linux.SO_SNDBUF, linux.SO_RCVBUF, linux.SO_REUSEADDR: optlen = sizeofInt32 - case syscall.SO_LINGER: + case linux.SO_LINGER: optlen = syscall.SizeofLinger } - case syscall.SOL_TCP: + case linux.SOL_TCP: switch name { - case syscall.TCP_NODELAY: + case linux.TCP_NODELAY: optlen = sizeofInt32 - case syscall.TCP_INFO: + case linux.TCP_INFO: optlen = int(linux.SizeOfTCPInfo) } } + if optlen == 0 { return nil, syserr.ErrProtocolNotAvailable // ENOPROTOOPT } @@ -320,19 +331,24 @@ func (s *socketOperations) SetSockOpt(t *kernel.Task, level int, name int, opt [ // Whitelist options and constrain option length. var optlen int switch level { - case syscall.SOL_IPV6: + case linux.SOL_IP: switch name { - case syscall.IPV6_V6ONLY: + case linux.IP_RECVTOS: optlen = sizeofInt32 } - case syscall.SOL_SOCKET: + case linux.SOL_IPV6: switch name { - case syscall.SO_SNDBUF, syscall.SO_RCVBUF, syscall.SO_REUSEADDR: + case linux.IPV6_RECVTCLASS, linux.IPV6_V6ONLY: optlen = sizeofInt32 } - case syscall.SOL_TCP: + case linux.SOL_SOCKET: switch name { - case syscall.TCP_NODELAY: + case linux.SO_SNDBUF, linux.SO_RCVBUF, linux.SO_REUSEADDR: + optlen = sizeofInt32 + } + case linux.SOL_TCP: + switch name { + case linux.TCP_NODELAY: optlen = sizeofInt32 } } @@ -354,11 +370,11 @@ func (s *socketOperations) SetSockOpt(t *kernel.Task, level int, name int, opt [ } // RecvMsg implements socket.Socket.RecvMsg. -func (s *socketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, haveDeadline bool, deadline ktime.Time, senderRequested bool, controlDataLen uint64) (int, int, linux.SockAddr, uint32, socket.ControlMessages, *syserr.Error) { +func (s *socketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, haveDeadline bool, deadline ktime.Time, senderRequested bool, controlLen uint64) (int, int, linux.SockAddr, uint32, socket.ControlMessages, *syserr.Error) { // Whitelist flags. // // FIXME(jamieliu): We can't support MSG_ERRQUEUE because it uses ancillary - // messages that netstack/tcpip/transport/unix doesn't understand. Kill the + // messages that gvisor/pkg/tcpip/transport/unix doesn't understand. Kill the // Socket interface's dependence on netstack. if flags&^(syscall.MSG_DONTWAIT|syscall.MSG_PEEK|syscall.MSG_TRUNC) != 0 { return 0, 0, nil, 0, socket.ControlMessages{}, syserr.ErrInvalidArgument @@ -370,6 +386,7 @@ func (s *socketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags senderAddrBuf = make([]byte, sizeofSockaddr) } + var controlBuf []byte var msgFlags int recvmsgToBlocks := safemem.ReaderFunc(func(dsts safemem.BlockSeq) (uint64, error) { @@ -384,11 +401,6 @@ func (s *socketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags // We always do a non-blocking recv*(). sysflags := flags | syscall.MSG_DONTWAIT - if dsts.NumBlocks() == 1 { - // Skip allocating []syscall.Iovec. - return recvfrom(s.fd, dsts.Head().ToSlice(), sysflags, &senderAddrBuf) - } - iovs := iovecsFromBlockSeq(dsts) msg := syscall.Msghdr{ Iov: &iovs[0], @@ -398,12 +410,18 @@ func (s *socketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags msg.Name = &senderAddrBuf[0] msg.Namelen = uint32(len(senderAddrBuf)) } + if controlLen > 0 { + controlBuf = make([]byte, maxControlLen) + msg.Control = &controlBuf[0] + msg.Controllen = maxControlLen + } n, err := recvmsg(s.fd, &msg, sysflags) if err != nil { return 0, err } senderAddrBuf = senderAddrBuf[:msg.Namelen] msgFlags = int(msg.Flags) + controlLen = uint64(msg.Controllen) return n, nil }) @@ -429,14 +447,38 @@ func (s *socketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags n, err = dst.CopyOutFrom(t, recvmsgToBlocks) } } - - // We don't allow control messages. - msgFlags &^= linux.MSG_CTRUNC + if err != nil { + return 0, 0, nil, 0, socket.ControlMessages{}, syserr.FromError(err) + } if senderRequested { senderAddr = socket.UnmarshalSockAddr(s.family, senderAddrBuf) } - return int(n), msgFlags, senderAddr, uint32(len(senderAddrBuf)), socket.ControlMessages{}, syserr.FromError(err) + + unixControlMessages, err := unix.ParseSocketControlMessage(controlBuf[:controlLen]) + if err != nil { + return 0, 0, nil, 0, socket.ControlMessages{}, syserr.FromError(err) + } + + controlMessages := socket.ControlMessages{} + for _, unixCmsg := range unixControlMessages { + switch unixCmsg.Header.Level { + case syscall.SOL_IP: + switch unixCmsg.Header.Type { + case syscall.IP_TOS: + controlMessages.IP.HasTOS = true + binary.Unmarshal(unixCmsg.Data[:linux.SizeOfControlMessageTOS], usermem.ByteOrder, &controlMessages.IP.TOS) + } + case syscall.SOL_IPV6: + switch unixCmsg.Header.Type { + case syscall.IPV6_TCLASS: + controlMessages.IP.HasTClass = true + binary.Unmarshal(unixCmsg.Data[:linux.SizeOfControlMessageTClass], usermem.ByteOrder, &controlMessages.IP.TClass) + } + } + } + + return int(n), msgFlags, senderAddr, uint32(len(senderAddrBuf)), controlMessages, nil } // SendMsg implements socket.Socket.SendMsg. diff --git a/pkg/sentry/syscalls/linux/sys_socket.go b/pkg/sentry/syscalls/linux/sys_socket.go index ab1001f16..13f77565f 100644 --- a/pkg/sentry/syscalls/linux/sys_socket.go +++ b/pkg/sentry/syscalls/linux/sys_socket.go @@ -802,6 +802,14 @@ func recvSingleMsg(t *kernel.Task, s socket.Socket, msgPtr usermem.Addr, flags i controlData = control.PackInq(t, cms.IP.Inq, controlData) } + if cms.IP.HasTOS { + controlData = control.PackTOS(t, cms.IP.TOS, controlData) + } + + if cms.IP.HasTClass { + controlData = control.PackTClass(t, cms.IP.TClass, controlData) + } + if cms.Unix.Rights != nil { controlData, mflags = control.PackRights(t, cms.Unix.Rights.(control.SCMRights), flags&linux.MSG_CMSG_CLOEXEC != 0, controlData, mflags) } diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index bd5eb89ca..5746043cc 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -308,7 +308,7 @@ type ControlMessages struct { // HasTimestamp indicates whether Timestamp is valid/set. HasTimestamp bool - // Timestamp is the time (in ns) that the last packed used to create + // Timestamp is the time (in ns) that the last packet used to create // the read data was received. Timestamp int64 @@ -317,6 +317,18 @@ type ControlMessages struct { // Inq is the number of bytes ready to be received. Inq int32 + + // HasTOS indicates whether Tos is valid/set. + HasTOS bool + + // TOS is the IPv4 type of service of the associated packet. + TOS int8 + + // HasTClass indicates whether Tclass is valid/set. + HasTClass bool + + // Tclass is the IPv6 traffic class of the associated packet. + TClass int32 } // Endpoint is the interface implemented by transport protocols (e.g., tcp, udp) diff --git a/runsc/boot/filter/config.go b/runsc/boot/filter/config.go index 677356193..bf690160c 100644 --- a/runsc/boot/filter/config.go +++ b/runsc/boot/filter/config.go @@ -134,11 +134,6 @@ var allowedSyscalls = seccomp.SyscallRules{ seccomp.AllowValue(syscall.SOL_SOCKET), seccomp.AllowValue(syscall.SO_SNDBUF), }, - { - seccomp.AllowAny{}, - seccomp.AllowValue(syscall.SOL_SOCKET), - seccomp.AllowValue(syscall.SO_REUSEADDR), - }, }, syscall.SYS_GETTID: {}, syscall.SYS_GETTIMEOFDAY: {}, @@ -315,6 +310,16 @@ func hostInetFilters() seccomp.SyscallRules { syscall.SYS_GETPEERNAME: {}, syscall.SYS_GETSOCKNAME: {}, syscall.SYS_GETSOCKOPT: []seccomp.Rule{ + { + seccomp.AllowAny{}, + seccomp.AllowValue(syscall.SOL_IP), + seccomp.AllowValue(syscall.IP_RECVTOS), + }, + { + seccomp.AllowAny{}, + seccomp.AllowValue(syscall.SOL_IPV6), + seccomp.AllowValue(syscall.IPV6_RECVTCLASS), + }, { seccomp.AllowAny{}, seccomp.AllowValue(syscall.SOL_IPV6), @@ -418,6 +423,20 @@ func hostInetFilters() seccomp.SyscallRules { seccomp.AllowAny{}, seccomp.AllowValue(4), }, + { + seccomp.AllowAny{}, + seccomp.AllowValue(syscall.SOL_IP), + seccomp.AllowValue(syscall.IP_RECVTOS), + seccomp.AllowAny{}, + seccomp.AllowValue(4), + }, + { + seccomp.AllowAny{}, + seccomp.AllowValue(syscall.SOL_IPV6), + seccomp.AllowValue(syscall.IPV6_RECVTCLASS), + seccomp.AllowAny{}, + seccomp.AllowValue(4), + }, }, syscall.SYS_SHUTDOWN: []seccomp.Rule{ { -- cgit v1.2.3 From 9194aab2aaada137b377fdfcb812a7c015857d5d Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Mon, 2 Dec 2019 08:38:45 -0800 Subject: Support sending IP_TOS and IPV6_TCLASS control messages with hostinet sockets. There are two potential ways of sending a TOS byte with outgoing packets: including a control message in sendmsg, or setting the IP_TOS/IPV6_TCLASS socket options (for IPV4 and IPV6 respectively). This change lets hostinet support the former. PiperOrigin-RevId: 283346737 --- pkg/sentry/socket/control/BUILD | 1 + pkg/sentry/socket/control/control.go | 154 +++++++++++++++++++++----------- pkg/sentry/socket/hostinet/BUILD | 1 + pkg/sentry/socket/hostinet/socket.go | 8 +- pkg/sentry/syscalls/linux/sys_socket.go | 4 +- 5 files changed, 115 insertions(+), 53 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/control/BUILD b/pkg/sentry/socket/control/BUILD index 4a6e83a8b..357517ed4 100644 --- a/pkg/sentry/socket/control/BUILD +++ b/pkg/sentry/socket/control/BUILD @@ -17,6 +17,7 @@ go_library( "//pkg/sentry/fs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", + "//pkg/sentry/socket", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usermem", "//pkg/syserror", diff --git a/pkg/sentry/socket/control/control.go b/pkg/sentry/socket/control/control.go index 0371acede..782a3cb92 100644 --- a/pkg/sentry/socket/control/control.go +++ b/pkg/sentry/socket/control/control.go @@ -23,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" @@ -347,30 +348,63 @@ func PackTClass(t *kernel.Task, tClass int32, buf []byte) []byte { ) } +func addSpaceForCmsg(cmsgDataLen int, buf []byte) []byte { + newBuf := make([]byte, 0, len(buf)+linux.SizeOfControlMessageHeader+cmsgDataLen) + return append(newBuf, buf...) +} + +// PackControlMessages converts the given ControlMessages struct into a buffer. +// We skip control messages specific to Unix domain sockets. +func PackControlMessages(t *kernel.Task, cmsgs socket.ControlMessages) []byte { + var buf []byte + // The use of t.Arch().Width() is analogous to Linux's use of sizeof(long) in + // CMSG_ALIGN. + width := t.Arch().Width() + + if cmsgs.IP.HasTimestamp { + buf = addSpaceForCmsg(int(width), buf) + buf = PackTimestamp(t, cmsgs.IP.Timestamp, buf) + } + + if cmsgs.IP.HasInq { + // In Linux, TCP_CM_INQ is added after SO_TIMESTAMP. + buf = addSpaceForCmsg(AlignUp(linux.SizeOfControlMessageInq, width), buf) + buf = PackInq(t, cmsgs.IP.Inq, buf) + } + + if cmsgs.IP.HasTOS { + buf = addSpaceForCmsg(AlignUp(linux.SizeOfControlMessageTOS, width), buf) + buf = PackTOS(t, cmsgs.IP.TOS, buf) + } + + if cmsgs.IP.HasTClass { + buf = addSpaceForCmsg(AlignUp(linux.SizeOfControlMessageTClass, width), buf) + buf = PackTClass(t, cmsgs.IP.TClass, buf) + } + + return buf +} + // Parse parses a raw socket control message into portable objects. -func Parse(t *kernel.Task, socketOrEndpoint interface{}, buf []byte) (transport.ControlMessages, error) { +func Parse(t *kernel.Task, socketOrEndpoint interface{}, buf []byte) (socket.ControlMessages, error) { var ( - fds linux.ControlMessageRights - haveCreds bool - creds linux.ControlMessageCredentials + cmsgs socket.ControlMessages + fds linux.ControlMessageRights ) for i := 0; i < len(buf); { if i+linux.SizeOfControlMessageHeader > len(buf) { - return transport.ControlMessages{}, syserror.EINVAL + return cmsgs, syserror.EINVAL } var h linux.ControlMessageHeader binary.Unmarshal(buf[i:i+linux.SizeOfControlMessageHeader], usermem.ByteOrder, &h) if h.Length < uint64(linux.SizeOfControlMessageHeader) { - return transport.ControlMessages{}, syserror.EINVAL + return socket.ControlMessages{}, syserror.EINVAL } if h.Length > uint64(len(buf)-i) { - return transport.ControlMessages{}, syserror.EINVAL - } - if h.Level != linux.SOL_SOCKET { - return transport.ControlMessages{}, syserror.EINVAL + return socket.ControlMessages{}, syserror.EINVAL } i += linux.SizeOfControlMessageHeader @@ -380,59 +414,79 @@ func Parse(t *kernel.Task, socketOrEndpoint interface{}, buf []byte) (transport. // sizeof(long) in CMSG_ALIGN. width := t.Arch().Width() - switch h.Type { - case linux.SCM_RIGHTS: - rightsSize := AlignDown(length, linux.SizeOfControlMessageRight) - numRights := rightsSize / linux.SizeOfControlMessageRight - - if len(fds)+numRights > linux.SCM_MAX_FD { - return transport.ControlMessages{}, syserror.EINVAL + switch h.Level { + case linux.SOL_SOCKET: + switch h.Type { + case linux.SCM_RIGHTS: + rightsSize := AlignDown(length, linux.SizeOfControlMessageRight) + numRights := rightsSize / linux.SizeOfControlMessageRight + + if len(fds)+numRights > linux.SCM_MAX_FD { + return socket.ControlMessages{}, syserror.EINVAL + } + + for j := i; j < i+rightsSize; j += linux.SizeOfControlMessageRight { + fds = append(fds, int32(usermem.ByteOrder.Uint32(buf[j:j+linux.SizeOfControlMessageRight]))) + } + + i += AlignUp(length, width) + + case linux.SCM_CREDENTIALS: + if length < linux.SizeOfControlMessageCredentials { + return socket.ControlMessages{}, syserror.EINVAL + } + + var creds linux.ControlMessageCredentials + binary.Unmarshal(buf[i:i+linux.SizeOfControlMessageCredentials], usermem.ByteOrder, &creds) + scmCreds, err := NewSCMCredentials(t, creds) + if err != nil { + return socket.ControlMessages{}, err + } + cmsgs.Unix.Credentials = scmCreds + i += AlignUp(length, width) + + default: + // Unknown message type. + return socket.ControlMessages{}, syserror.EINVAL } - - for j := i; j < i+rightsSize; j += linux.SizeOfControlMessageRight { - fds = append(fds, int32(usermem.ByteOrder.Uint32(buf[j:j+linux.SizeOfControlMessageRight]))) + case linux.SOL_IP: + switch h.Type { + case linux.IP_TOS: + cmsgs.IP.HasTOS = true + binary.Unmarshal(buf[i:i+linux.SizeOfControlMessageTOS], usermem.ByteOrder, &cmsgs.IP.TOS) + i += AlignUp(length, width) + + default: + return socket.ControlMessages{}, syserror.EINVAL } - - i += AlignUp(length, width) - - case linux.SCM_CREDENTIALS: - if length < linux.SizeOfControlMessageCredentials { - return transport.ControlMessages{}, syserror.EINVAL + case linux.SOL_IPV6: + switch h.Type { + case linux.IPV6_TCLASS: + cmsgs.IP.HasTClass = true + binary.Unmarshal(buf[i:i+linux.SizeOfControlMessageTClass], usermem.ByteOrder, &cmsgs.IP.TClass) + i += AlignUp(length, width) + + default: + return socket.ControlMessages{}, syserror.EINVAL } - - binary.Unmarshal(buf[i:i+linux.SizeOfControlMessageCredentials], usermem.ByteOrder, &creds) - haveCreds = true - i += AlignUp(length, width) - default: - // Unknown message type. - return transport.ControlMessages{}, syserror.EINVAL + return socket.ControlMessages{}, syserror.EINVAL } } - var credentials SCMCredentials - if haveCreds { - var err error - if credentials, err = NewSCMCredentials(t, creds); err != nil { - return transport.ControlMessages{}, err - } - } else { - credentials = makeCreds(t, socketOrEndpoint) + if cmsgs.Unix.Credentials == nil { + cmsgs.Unix.Credentials = makeCreds(t, socketOrEndpoint) } - var rights SCMRights if len(fds) > 0 { - var err error - if rights, err = NewSCMRights(t, fds); err != nil { - return transport.ControlMessages{}, err + rights, err := NewSCMRights(t, fds) + if err != nil { + return socket.ControlMessages{}, err } + cmsgs.Unix.Rights = rights } - if credentials == nil && rights == nil { - return transport.ControlMessages{}, nil - } - - return transport.ControlMessages{Credentials: credentials, Rights: rights}, nil + return cmsgs, nil } func makeCreds(t *kernel.Task, socketOrEndpoint interface{}) SCMCredentials { diff --git a/pkg/sentry/socket/hostinet/BUILD b/pkg/sentry/socket/hostinet/BUILD index b1cf1126f..4c44c7c0f 100644 --- a/pkg/sentry/socket/hostinet/BUILD +++ b/pkg/sentry/socket/hostinet/BUILD @@ -29,6 +29,7 @@ go_library( "//pkg/sentry/kernel/time", "//pkg/sentry/safemem", "//pkg/sentry/socket", + "//pkg/sentry/socket/control", "//pkg/sentry/usermem", "//pkg/syserr", "//pkg/syserror", diff --git a/pkg/sentry/socket/hostinet/socket.go b/pkg/sentry/socket/hostinet/socket.go index aa234f760..8d9363aac 100644 --- a/pkg/sentry/socket/hostinet/socket.go +++ b/pkg/sentry/socket/hostinet/socket.go @@ -30,6 +30,7 @@ import ( ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/socket" + "gvisor.dev/gvisor/pkg/sentry/socket/control" "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" @@ -488,6 +489,7 @@ func (s *socketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to [] return 0, syserr.ErrInvalidArgument } + controlBuf := control.PackControlMessages(t, controlMessages) sendmsgFromBlocks := safemem.WriterFunc(func(srcs safemem.BlockSeq) (uint64, error) { // Refuse to do anything if any part of src.Addrs was unusable. if uint64(src.NumBytes()) != srcs.NumBytes() { @@ -500,7 +502,7 @@ func (s *socketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to [] // We always do a non-blocking send*(). sysflags := flags | syscall.MSG_DONTWAIT - if srcs.NumBlocks() == 1 { + if srcs.NumBlocks() == 1 && len(controlBuf) == 0 { // Skip allocating []syscall.Iovec. src := srcs.Head() n, _, errno := syscall.Syscall6(syscall.SYS_SENDTO, uintptr(s.fd), src.Addr(), uintptr(src.Len()), uintptr(sysflags), uintptr(firstBytePtr(to)), uintptr(len(to))) @@ -519,6 +521,10 @@ func (s *socketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to [] msg.Name = &to[0] msg.Namelen = uint32(len(to)) } + if len(controlBuf) != 0 { + msg.Control = &controlBuf[0] + msg.Controllen = uint64(len(controlBuf)) + } return sendmsg(s.fd, &msg, sysflags) }) diff --git a/pkg/sentry/syscalls/linux/sys_socket.go b/pkg/sentry/syscalls/linux/sys_socket.go index 13f77565f..d8acae063 100644 --- a/pkg/sentry/syscalls/linux/sys_socket.go +++ b/pkg/sentry/syscalls/linux/sys_socket.go @@ -1068,10 +1068,10 @@ func sendSingleMsg(t *kernel.Task, s socket.Socket, file *fs.File, msgPtr userme } // Call the syscall implementation. - n, e := s.SendMsg(t, src, to, int(flags), haveDeadline, deadline, socket.ControlMessages{Unix: controlMessages}) + n, e := s.SendMsg(t, src, to, int(flags), haveDeadline, deadline, controlMessages) err = handleIOError(t, n != 0, e.ToError(), kernel.ERESTARTSYS, "sendmsg", file) if err != nil { - controlMessages.Release() + controlMessages.Unix.Release() } return uintptr(n), err } -- cgit v1.2.3 From 03760e5623f3ee736252fda7da033fd51144af1e Mon Sep 17 00:00:00 2001 From: Haibo Xu Date: Wed, 20 Nov 2019 09:24:41 +0000 Subject: platform/ptrace: make some operations arch specific Make the patchSignalInfo/cpuid faulting/initial thread seccomp rules operations architecture dependent. Signed-off-by: Haibo Xu Change-Id: Iaf692dbe3700d2e01168ec2f1b4beeda9136fd62 --- pkg/sentry/platform/ptrace/subprocess_amd64.go | 42 ++++++++++++++++++++++++++ pkg/sentry/platform/ptrace/subprocess_arm64.go | 37 ++++++++++++++++++++++- pkg/sentry/platform/ptrace/subprocess_linux.go | 36 +++++----------------- 3 files changed, 85 insertions(+), 30 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/ptrace/subprocess_amd64.go b/pkg/sentry/platform/ptrace/subprocess_amd64.go index 4649a94a7..a55cff507 100644 --- a/pkg/sentry/platform/ptrace/subprocess_amd64.go +++ b/pkg/sentry/platform/ptrace/subprocess_amd64.go @@ -21,6 +21,8 @@ import ( "strings" "syscall" + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/seccomp" "gvisor.dev/gvisor/pkg/sentry/arch" ) @@ -143,3 +145,43 @@ func (t *thread) adjustInitRegsRip() { func initChildProcessPPID(initregs *syscall.PtraceRegs, ppid int32) { initregs.R15 = uint64(ppid) } + +// patchSignalInfo patches the signal info to account for hitting the seccomp +// filters from vsyscall emulation, specified below. We allow for SIGSYS as a +// synchronous trap, but patch the structure to appear like a SIGSEGV with the +// Rip as the faulting address. +// +// Note that this should only be called after verifying that the signalInfo has +// been generated by the kernel. +func patchSignalInfo(regs *syscall.PtraceRegs, signalInfo *arch.SignalInfo) { + if linux.Signal(signalInfo.Signo) == linux.SIGSYS { + signalInfo.Signo = int32(linux.SIGSEGV) + + // Unwind the kernel emulation, if any has occurred. A SIGSYS is delivered + // with the si_call_addr field pointing to the current RIP. This field + // aligns with the si_addr field for a SIGSEGV, so we don't need to touch + // anything there. We do need to unwind emulation however, so we set the + // instruction pointer to the faulting value, and "unpop" the stack. + regs.Rip = signalInfo.Addr() + regs.Rsp -= 8 + } +} + +// enableCpuidFault enable cpuid-faulting; this may fail on older kernels or hardware, +// so we just disregard the result. Host CPUID will be enabled. +func enableCpuidFault() { + syscall.RawSyscall6(syscall.SYS_ARCH_PRCTL, linux.ARCH_SET_CPUID, 0, 0, 0, 0, 0) +} + +// appendArchSeccompRules append architecture specific seccomp rules when creating BPF program. +// Ref attachedThread() for more detail. +func appendArchSeccompRules(rules []seccomp.RuleSet) []seccomp.RuleSet { + return append(rules, seccomp.RuleSet{ + Rules: seccomp.SyscallRules{ + syscall.SYS_ARCH_PRCTL: []seccomp.Rule{ + {seccomp.AllowValue(linux.ARCH_SET_CPUID), seccomp.AllowValue(0)}, + }, + }, + Action: linux.SECCOMP_RET_ALLOW, + }) +} diff --git a/pkg/sentry/platform/ptrace/subprocess_arm64.go b/pkg/sentry/platform/ptrace/subprocess_arm64.go index bec884ba5..aed34e7ee 100644 --- a/pkg/sentry/platform/ptrace/subprocess_arm64.go +++ b/pkg/sentry/platform/ptrace/subprocess_arm64.go @@ -17,8 +17,12 @@ package ptrace import ( + "fmt" + "strings" "syscall" + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/seccomp" "gvisor.dev/gvisor/pkg/sentry/arch" ) @@ -37,7 +41,7 @@ const ( // resetSysemuRegs sets up emulation registers. // // This should be called prior to calling sysemu. -func (s *subprocess) resetSysemuRegs(regs *syscall.PtraceRegs) { +func (t *thread) resetSysemuRegs(regs *syscall.PtraceRegs) { } // createSyscallRegs sets up syscall registers. @@ -124,3 +128,34 @@ func (t *thread) adjustInitRegsRip() { func initChildProcessPPID(initregs *syscall.PtraceRegs, ppid int32) { initregs.Regs[7] = uint64(ppid) } + +// patchSignalInfo patches the signal info to account for hitting the seccomp +// filters from vsyscall emulation, specified below. We allow for SIGSYS as a +// synchronous trap, but patch the structure to appear like a SIGSEGV with the +// Rip as the faulting address. +// +// Note that this should only be called after verifying that the signalInfo has +// been generated by the kernel. +func patchSignalInfo(regs *syscall.PtraceRegs, signalInfo *arch.SignalInfo) { + if linux.Signal(signalInfo.Signo) == linux.SIGSYS { + signalInfo.Signo = int32(linux.SIGSEGV) + + // Unwind the kernel emulation, if any has occurred. A SIGSYS is delivered + // with the si_call_addr field pointing to the current RIP. This field + // aligns with the si_addr field for a SIGSEGV, so we don't need to touch + // anything there. We do need to unwind emulation however, so we set the + // instruction pointer to the faulting value, and "unpop" the stack. + regs.Pc = signalInfo.Addr() + regs.Sp -= 8 + } +} + +// Noop on arm64. +func enableCpuidFault() { +} + +// appendArchSeccompRules append architecture specific seccomp rules when creating BPF program. +// Ref attachedThread() for more detail. +func appendArchSeccompRules(rules []seccomp.RuleSet) []seccomp.RuleSet { + return rules +} diff --git a/pkg/sentry/platform/ptrace/subprocess_linux.go b/pkg/sentry/platform/ptrace/subprocess_linux.go index 3782d4332..cf13ea5e4 100644 --- a/pkg/sentry/platform/ptrace/subprocess_linux.go +++ b/pkg/sentry/platform/ptrace/subprocess_linux.go @@ -20,6 +20,7 @@ import ( "fmt" "syscall" + "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/procid" @@ -77,27 +78,6 @@ func probeSeccomp() bool { } } -// patchSignalInfo patches the signal info to account for hitting the seccomp -// filters from vsyscall emulation, specified below. We allow for SIGSYS as a -// synchronous trap, but patch the structure to appear like a SIGSEGV with the -// Rip as the faulting address. -// -// Note that this should only be called after verifying that the signalInfo has -// been generated by the kernel. -func patchSignalInfo(regs *syscall.PtraceRegs, signalInfo *arch.SignalInfo) { - if linux.Signal(signalInfo.Signo) == linux.SIGSYS { - signalInfo.Signo = int32(linux.SIGSEGV) - - // Unwind the kernel emulation, if any has occurred. A SIGSYS is delivered - // with the si_call_addr field pointing to the current RIP. This field - // aligns with the si_addr field for a SIGSEGV, so we don't need to touch - // anything there. We do need to unwind emulation however, so we set the - // instruction pointer to the faulting value, and "unpop" the stack. - regs.Rip = signalInfo.Addr() - regs.Rsp -= 8 - } -} - // createStub creates a fresh stub processes. // // Precondition: the runtime OS thread must be locked. @@ -149,7 +129,7 @@ func attachedThread(flags uintptr, defaultAction linux.BPFAction) (*thread, erro Rules: seccomp.SyscallRules{ syscall.SYS_GETTIMEOFDAY: {}, syscall.SYS_TIME: {}, - 309: {}, // SYS_GETCPU. + unix.SYS_GETCPU: {}, // SYS_GETCPU was not defined in package syscall on amd64. }, Action: linux.SECCOMP_RET_TRAP, Vsyscall: true, @@ -173,10 +153,7 @@ func attachedThread(flags uintptr, defaultAction linux.BPFAction) (*thread, erro // For the initial process creation. syscall.SYS_WAIT4: {}, - syscall.SYS_ARCH_PRCTL: []seccomp.Rule{ - {seccomp.AllowValue(linux.ARCH_SET_CPUID), seccomp.AllowValue(0)}, - }, - syscall.SYS_EXIT: {}, + syscall.SYS_EXIT: {}, // For the stub prctl dance (all). syscall.SYS_PRCTL: []seccomp.Rule{ @@ -196,6 +173,8 @@ func attachedThread(flags uintptr, defaultAction linux.BPFAction) (*thread, erro }, Action: linux.SECCOMP_RET_ALLOW, }) + + rules = appendArchSeccompRules(rules) } instrs, err := seccomp.BuildProgram(rules, defaultAction) if err != nil { @@ -267,9 +246,8 @@ func attachedThread(flags uintptr, defaultAction linux.BPFAction) (*thread, erro syscall.RawSyscall(syscall.SYS_EXIT, uintptr(errno), 0, 0) } - // Enable cpuid-faulting; this may fail on older kernels or hardware, - // so we just disregard the result. Host CPUID will be enabled. - syscall.RawSyscall6(syscall.SYS_ARCH_PRCTL, linux.ARCH_SET_CPUID, 0, 0, 0, 0, 0) + // Enable cpuid-faulting. + enableCpuidFault() // Call the stub; should not return. stubCall(stubStart, ppid) -- cgit v1.2.3 From 19b2d997ec702e559bdb5f5e60634a7c5d7d288e Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 3 Dec 2019 08:32:03 -0800 Subject: Support IP_TOS and IPV6_TCLASS socket options for hostinet sockets. There are two potential ways of sending a TOS byte with outgoing packets: including a control message in sendmsg, or setting the IP_TOS/IPV6_TCLASS socket options (for IPV4 and IPV6 respectively). This change lets hostinet support the latter. Fixes #1188 PiperOrigin-RevId: 283550925 --- pkg/sentry/socket/hostinet/socket.go | 8 ++++---- runsc/boot/filter/config.go | 24 ++++++++++++++++++++++++ 2 files changed, 28 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/hostinet/socket.go b/pkg/sentry/socket/hostinet/socket.go index 8d9363aac..a8c152b54 100644 --- a/pkg/sentry/socket/hostinet/socket.go +++ b/pkg/sentry/socket/hostinet/socket.go @@ -289,12 +289,12 @@ func (s *socketOperations) GetSockOpt(t *kernel.Task, level int, name int, outPt switch level { case linux.SOL_IP: switch name { - case linux.IP_RECVTOS: + case linux.IP_TOS, linux.IP_RECVTOS: optlen = sizeofInt32 } case linux.SOL_IPV6: switch name { - case linux.IPV6_RECVTCLASS, linux.IPV6_V6ONLY: + case linux.IPV6_TCLASS, linux.IPV6_RECVTCLASS, linux.IPV6_V6ONLY: optlen = sizeofInt32 } case linux.SOL_SOCKET: @@ -334,12 +334,12 @@ func (s *socketOperations) SetSockOpt(t *kernel.Task, level int, name int, opt [ switch level { case linux.SOL_IP: switch name { - case linux.IP_RECVTOS: + case linux.IP_TOS, linux.IP_RECVTOS: optlen = sizeofInt32 } case linux.SOL_IPV6: switch name { - case linux.IPV6_RECVTCLASS, linux.IPV6_V6ONLY: + case linux.IPV6_TCLASS, linux.IPV6_RECVTCLASS, linux.IPV6_V6ONLY: optlen = sizeofInt32 } case linux.SOL_SOCKET: diff --git a/runsc/boot/filter/config.go b/runsc/boot/filter/config.go index bf690160c..4fb9adca6 100644 --- a/runsc/boot/filter/config.go +++ b/runsc/boot/filter/config.go @@ -310,11 +310,21 @@ func hostInetFilters() seccomp.SyscallRules { syscall.SYS_GETPEERNAME: {}, syscall.SYS_GETSOCKNAME: {}, syscall.SYS_GETSOCKOPT: []seccomp.Rule{ + { + seccomp.AllowAny{}, + seccomp.AllowValue(syscall.SOL_IP), + seccomp.AllowValue(syscall.IP_TOS), + }, { seccomp.AllowAny{}, seccomp.AllowValue(syscall.SOL_IP), seccomp.AllowValue(syscall.IP_RECVTOS), }, + { + seccomp.AllowAny{}, + seccomp.AllowValue(syscall.SOL_IPV6), + seccomp.AllowValue(syscall.IPV6_TCLASS), + }, { seccomp.AllowAny{}, seccomp.AllowValue(syscall.SOL_IPV6), @@ -423,6 +433,13 @@ func hostInetFilters() seccomp.SyscallRules { seccomp.AllowAny{}, seccomp.AllowValue(4), }, + { + seccomp.AllowAny{}, + seccomp.AllowValue(syscall.SOL_IP), + seccomp.AllowValue(syscall.IP_TOS), + seccomp.AllowAny{}, + seccomp.AllowValue(4), + }, { seccomp.AllowAny{}, seccomp.AllowValue(syscall.SOL_IP), @@ -430,6 +447,13 @@ func hostInetFilters() seccomp.SyscallRules { seccomp.AllowAny{}, seccomp.AllowValue(4), }, + { + seccomp.AllowAny{}, + seccomp.AllowValue(syscall.SOL_IPV6), + seccomp.AllowValue(syscall.IPV6_TCLASS), + seccomp.AllowAny{}, + seccomp.AllowValue(4), + }, { seccomp.AllowAny{}, seccomp.AllowValue(syscall.SOL_IPV6), -- cgit v1.2.3 From 812189664cab0a17ae29095e4029e2f8762a6779 Mon Sep 17 00:00:00 2001 From: Zach Koopmans Date: Tue, 3 Dec 2019 10:22:01 -0800 Subject: Remove TODO for obsolete bug. PiperOrigin-RevId: 283571456 --- pkg/sentry/socket/rpcinet/syscall_rpc.proto | 1 - 1 file changed, 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/rpcinet/syscall_rpc.proto b/pkg/sentry/socket/rpcinet/syscall_rpc.proto index 9586f5923..b677e9eb3 100644 --- a/pkg/sentry/socket/rpcinet/syscall_rpc.proto +++ b/pkg/sentry/socket/rpcinet/syscall_rpc.proto @@ -3,7 +3,6 @@ syntax = "proto3"; // package syscall_rpc is a set of networking related system calls that can be // forwarded to a socket gofer. // -// TODO(b/77963526): Document individual RPCs. package syscall_rpc; message SendmsgRequest { -- cgit v1.2.3 From 3e832bec1b48b95951c3b83eb5a7b70f29b1f10f Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Tue, 3 Dec 2019 13:31:07 -0800 Subject: Point TODOs to gvisor.dev PiperOrigin-RevId: 283610781 --- pkg/sentry/fsimpl/proc/filesystems.go | 2 +- pkg/sentry/fsimpl/proc/mounts.go | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/proc/filesystems.go b/pkg/sentry/fsimpl/proc/filesystems.go index c36c4aff5..0e016bca5 100644 --- a/pkg/sentry/fsimpl/proc/filesystems.go +++ b/pkg/sentry/fsimpl/proc/filesystems.go @@ -19,7 +19,7 @@ package proc // +stateify savable type filesystemsData struct{} -// TODO(b/138862512): Implement vfs.DynamicBytesSource.Generate for +// TODO(gvisor.dev/issue/1195): Implement vfs.DynamicBytesSource.Generate for // filesystemsData. We would need to retrive filesystem names from // vfs.VirtualFilesystem. Also needs vfs replacement for // fs.Filesystem.AllowUserList() and fs.FilesystemRequiresDev. diff --git a/pkg/sentry/fsimpl/proc/mounts.go b/pkg/sentry/fsimpl/proc/mounts.go index e81b1e910..8683cf677 100644 --- a/pkg/sentry/fsimpl/proc/mounts.go +++ b/pkg/sentry/fsimpl/proc/mounts.go @@ -16,7 +16,7 @@ package proc import "gvisor.dev/gvisor/pkg/sentry/kernel" -// TODO(b/138862512): Implement mountInfoFile and mountsFile. +// TODO(gvisor.dev/issue/1195): Implement mountInfoFile and mountsFile. // mountInfoFile implements vfs.DynamicBytesSource for /proc/[pid]/mountinfo. // -- cgit v1.2.3 From 154dcdec072ddad9e1c96b56e023d7f77fecf2ad Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Tue, 3 Dec 2019 13:42:30 -0800 Subject: Remove watchdog TODO I have not seen a false positive stuck task yet. Biggest offender was whitelistfs which is going away. PiperOrigin-RevId: 283613064 --- pkg/sentry/watchdog/watchdog.go | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/watchdog/watchdog.go b/pkg/sentry/watchdog/watchdog.go index ecce6c69f..5e4611333 100644 --- a/pkg/sentry/watchdog/watchdog.go +++ b/pkg/sentry/watchdog/watchdog.go @@ -287,7 +287,9 @@ func (w *Watchdog) runTurn() { if !ok { // New stuck task detected. // - // TODO(b/65849403): Tasks blocked doing IO may be considered stuck in kernel. + // Note that tasks blocked doing IO may be considered stuck in kernel, + // unless they are surrounded b + // Task.UninterruptibleSleepStart/Finish. tc = &offender{lastUpdateTime: lastUpdateTime} stuckTasks.Increment() newTaskFound = true -- cgit v1.2.3 From 43643752f05a0b25259b116558ccd870a539cc05 Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Tue, 3 Dec 2019 13:46:09 -0800 Subject: strace: don't create a slice with a negative value PiperOrigin-RevId: 283613824 --- pkg/sentry/strace/socket.go | 9 +++++++++ test/syscalls/linux/socket_unix_cmsg.cc | 29 +++++++++++++++++++++++++++++ 2 files changed, 38 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/strace/socket.go b/pkg/sentry/strace/socket.go index 94334f6d2..51f2efb39 100644 --- a/pkg/sentry/strace/socket.go +++ b/pkg/sentry/strace/socket.go @@ -208,6 +208,15 @@ func cmsghdr(t *kernel.Task, addr usermem.Addr, length uint64, maxBytes uint64) i += linux.SizeOfControlMessageHeader width := t.Arch().Width() length := int(h.Length) - linux.SizeOfControlMessageHeader + if length < 0 { + strs = append(strs, fmt.Sprintf( + "{level=%s, type=%s, length=%d, content too short}", + level, + typ, + h.Length, + )) + break + } if skipData { strs = append(strs, fmt.Sprintf("{level=%s, type=%s, length=%d}", level, typ, h.Length)) diff --git a/test/syscalls/linux/socket_unix_cmsg.cc b/test/syscalls/linux/socket_unix_cmsg.cc index 1159c5229..a16899493 100644 --- a/test/syscalls/linux/socket_unix_cmsg.cc +++ b/test/syscalls/linux/socket_unix_cmsg.cc @@ -149,6 +149,35 @@ TEST_P(UnixSocketPairCmsgTest, BadFDPass) { SyscallFailsWithErrno(EBADF)); } +TEST_P(UnixSocketPairCmsgTest, ShortCmsg) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + char sent_data[20]; + RandomizeBuffer(sent_data, sizeof(sent_data)); + + int sent_fd = -1; + + struct msghdr msg = {}; + char control[CMSG_SPACE(sizeof(sent_fd))]; + msg.msg_control = control; + msg.msg_controllen = sizeof(control); + + struct cmsghdr* cmsg = CMSG_FIRSTHDR(&msg); + cmsg->cmsg_len = 1; + cmsg->cmsg_level = SOL_SOCKET; + cmsg->cmsg_type = SCM_RIGHTS; + memcpy(CMSG_DATA(cmsg), &sent_fd, sizeof(sent_fd)); + + struct iovec iov; + iov.iov_base = sent_data; + iov.iov_len = sizeof(sent_data); + msg.msg_iov = &iov; + msg.msg_iovlen = 1; + + ASSERT_THAT(RetryEINTR(sendmsg)(sockets->first_fd(), &msg, 0), + SyscallFailsWithErrno(EINVAL)); +} + // BasicFDPassNoSpace starts off by sending a single FD just like BasicFDPass. // The difference is that when calling recvmsg, no space for FDs is provided, // only space for the cmsg header. -- cgit v1.2.3 From 035407153989b189a3ce42df43d6f528fa95444f Mon Sep 17 00:00:00 2001 From: Zach Koopmans Date: Tue, 3 Dec 2019 15:06:18 -0800 Subject: Fix printing /proc/[pid]/io for /proc/[pid]/task/[tid]/io. PiperOrigin-RevId: 283630669 --- pkg/sentry/fs/proc/task.go | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go index 87184ec67..2a598149d 100644 --- a/pkg/sentry/fs/proc/task.go +++ b/pkg/sentry/fs/proc/task.go @@ -639,7 +639,7 @@ func (i *ioData) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]se io.Accumulate(i.IOUsage()) var buf bytes.Buffer - fmt.Fprintf(&buf, "char: %d\n", io.CharsRead) + fmt.Fprintf(&buf, "rchar: %d\n", io.CharsRead) fmt.Fprintf(&buf, "wchar: %d\n", io.CharsWritten) fmt.Fprintf(&buf, "syscr: %d\n", io.ReadSyscalls) fmt.Fprintf(&buf, "syscw: %d\n", io.WriteSyscalls) -- cgit v1.2.3 From 70b68bb058e2f3281ba0e245ca404f264cfc547b Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Wed, 4 Dec 2019 16:28:16 +0800 Subject: Add a floating test case for Arm64 Signed-off-by: Bin Lu --- pkg/sentry/platform/kvm/testutil/testutil_arm64.s | 15 +++++++++++++++ 1 file changed, 15 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/kvm/testutil/testutil_arm64.s b/pkg/sentry/platform/kvm/testutil/testutil_arm64.s index 2cd28b2d2..0bebee852 100644 --- a/pkg/sentry/platform/kvm/testutil/testutil_arm64.s +++ b/pkg/sentry/platform/kvm/testutil/testutil_arm64.s @@ -50,6 +50,21 @@ TEXT ·SpinLoop(SB),NOSPLIT,$0 start: B start +TEXT ·FloatingPointWorks(SB),NOSPLIT,$0-8 + NO_LOCAL_POINTERS + FMOVD $(9.9), F0 + MOVD $SYS_GETPID, R8 // getpid + SVC + FMOVD $(9.9), F1 + FCMPD F0, F1 + BNE isNaN + MOVD $1, R0 + MOVD R0, ret+0(FP) + RET +isNaN: + MOVD $0, ret+0(FP) + RET + // MVN: bitwise logical NOT // This case simulates an application that modified R0-R30. #define TWIDDLE_REGS() \ -- cgit v1.2.3 From 1eda90d0848658f330e5f37ce18209bd3d069766 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Wed, 4 Dec 2019 13:53:08 -0800 Subject: Remove TODO since we don't plan to support debug registers PiperOrigin-RevId: 283828423 --- pkg/sentry/arch/arch_amd64.go | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/arch/arch_amd64.go b/pkg/sentry/arch/arch_amd64.go index 9e7db8b30..67daa6c24 100644 --- a/pkg/sentry/arch/arch_amd64.go +++ b/pkg/sentry/arch/arch_amd64.go @@ -305,7 +305,7 @@ func (c *context64) PtracePeekUser(addr uintptr) (interface{}, error) { buf := binary.Marshal(nil, usermem.ByteOrder, c.ptraceGetRegs()) return c.Native(uintptr(usermem.ByteOrder.Uint64(buf[addr:]))), nil } - // TODO(b/34088053): debug registers + // Note: x86 debug registers are missing. return c.Native(0), nil } @@ -320,6 +320,6 @@ func (c *context64) PtracePokeUser(addr, data uintptr) error { _, err := c.PtraceSetRegs(bytes.NewBuffer(buf)) return err } - // TODO(b/34088053): debug registers + // Note: x86 debug registers are missing. return nil } -- cgit v1.2.3 From 02258607f97353932d56bfde9274d50dda18e374 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Thu, 5 Dec 2019 12:56:31 -0800 Subject: Add vfs.CheckSetStat() and its dependencies. PiperOrigin-RevId: 284033820 --- pkg/sentry/vfs/permissions.go | 62 +++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 62 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/vfs/permissions.go b/pkg/sentry/vfs/permissions.go index f8e74355c..f1edb0680 100644 --- a/pkg/sentry/vfs/permissions.go +++ b/pkg/sentry/vfs/permissions.go @@ -119,3 +119,65 @@ func MayWriteFileWithOpenFlags(flags uint32) bool { return false } } + +// CheckSetStat checks that creds has permission to change the metadata of a +// file with the given permissions, UID, and GID as specified by stat, subject +// to the rules of Linux's fs/attr.c:setattr_prepare(). +func CheckSetStat(creds *auth.Credentials, stat *linux.Statx, mode uint16, kuid auth.KUID, kgid auth.KGID) error { + if stat.Mask&linux.STATX_MODE != 0 { + if !CanActAsOwner(creds, kuid) { + return syserror.EPERM + } + // TODO(b/30815691): "If the calling process is not privileged (Linux: + // does not have the CAP_FSETID capability), and the group of the file + // does not match the effective group ID of the process or one of its + // supplementary group IDs, the S_ISGID bit will be turned off, but + // this will not cause an error to be returned." - chmod(2) + } + if stat.Mask&linux.STATX_UID != 0 { + if !((creds.EffectiveKUID == kuid && auth.KUID(stat.UID) == kuid) || + HasCapabilityOnFile(creds, linux.CAP_CHOWN, kuid, kgid)) { + return syserror.EPERM + } + } + if stat.Mask&linux.STATX_GID != 0 { + if !((creds.EffectiveKUID == kuid && creds.InGroup(auth.KGID(stat.GID))) || + HasCapabilityOnFile(creds, linux.CAP_CHOWN, kuid, kgid)) { + return syserror.EPERM + } + } + if stat.Mask&(linux.STATX_ATIME|linux.STATX_MTIME|linux.STATX_CTIME) != 0 { + if !CanActAsOwner(creds, kuid) { + if (stat.Mask&linux.STATX_ATIME != 0 && stat.Atime.Nsec != linux.UTIME_NOW) || + (stat.Mask&linux.STATX_MTIME != 0 && stat.Mtime.Nsec != linux.UTIME_NOW) || + (stat.Mask&linux.STATX_CTIME != 0 && stat.Ctime.Nsec != linux.UTIME_NOW) { + return syserror.EPERM + } + // isDir is irrelevant in the following call to + // GenericCheckPermissions since ats == MayWrite means that + // CAP_DAC_READ_SEARCH does not apply, and CAP_DAC_OVERRIDE + // applies, regardless of isDir. + if err := GenericCheckPermissions(creds, MayWrite, false /* isDir */, mode, kuid, kgid); err != nil { + return err + } + } + } + return nil +} + +// CanActAsOwner returns true if creds can act as the owner of a file with the +// given owning UID, consistent with Linux's +// fs/inode.c:inode_owner_or_capable(). +func CanActAsOwner(creds *auth.Credentials, kuid auth.KUID) bool { + if creds.EffectiveKUID == kuid { + return true + } + return creds.HasCapability(linux.CAP_FOWNER) && creds.UserNamespace.MapFromKUID(kuid).Ok() +} + +// HasCapabilityOnFile returns true if creds has the given capability with +// respect to a file with the given owning UID and GID, consistent with Linux's +// kernel/capability.c:capable_wrt_inode_uidgid(). +func HasCapabilityOnFile(creds *auth.Credentials, cp linux.Capability, kuid auth.KUID, kgid auth.KGID) bool { + return creds.HasCapability(cp) && creds.UserNamespace.MapFromKUID(kuid).Ok() && creds.UserNamespace.MapFromKGID(kgid).Ok() +} -- cgit v1.2.3 From 0a32c0235744191947a6bf890031026e06788837 Mon Sep 17 00:00:00 2001 From: Zach Koopmans Date: Thu, 5 Dec 2019 13:22:31 -0800 Subject: Create correct file for /proc/[pid]/task/[tid]/io PiperOrigin-RevId: 284038840 --- pkg/sentry/fs/proc/task.go | 32 +++++++++++++++++--------------- test/syscalls/linux/proc.cc | 40 ++++++++++++++++++++++++++++++++++++++++ 2 files changed, 57 insertions(+), 15 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go index 2a598149d..0e46c5fb7 100644 --- a/pkg/sentry/fs/proc/task.go +++ b/pkg/sentry/fs/proc/task.go @@ -67,29 +67,28 @@ type taskDir struct { var _ fs.InodeOperations = (*taskDir)(nil) // newTaskDir creates a new proc task entry. -func (p *proc) newTaskDir(t *kernel.Task, msrc *fs.MountSource, showSubtasks bool) *fs.Inode { +func (p *proc) newTaskDir(t *kernel.Task, msrc *fs.MountSource, isThreadGroup bool) *fs.Inode { contents := map[string]*fs.Inode{ - "auxv": newAuxvec(t, msrc), - "cmdline": newExecArgInode(t, msrc, cmdlineExecArg), - "comm": newComm(t, msrc), - "environ": newExecArgInode(t, msrc, environExecArg), - "exe": newExe(t, msrc), - "fd": newFdDir(t, msrc), - "fdinfo": newFdInfoDir(t, msrc), - "gid_map": newGIDMap(t, msrc), - // FIXME(b/123511468): create the correct io file for threads. - "io": newIO(t, msrc), + "auxv": newAuxvec(t, msrc), + "cmdline": newExecArgInode(t, msrc, cmdlineExecArg), + "comm": newComm(t, msrc), + "environ": newExecArgInode(t, msrc, environExecArg), + "exe": newExe(t, msrc), + "fd": newFdDir(t, msrc), + "fdinfo": newFdInfoDir(t, msrc), + "gid_map": newGIDMap(t, msrc), + "io": newIO(t, msrc, isThreadGroup), "maps": newMaps(t, msrc), "mountinfo": seqfile.NewSeqFileInode(t, &mountInfoFile{t: t}, msrc), "mounts": seqfile.NewSeqFileInode(t, &mountsFile{t: t}, msrc), "ns": newNamespaceDir(t, msrc), "smaps": newSmaps(t, msrc), - "stat": newTaskStat(t, msrc, showSubtasks, p.pidns), + "stat": newTaskStat(t, msrc, isThreadGroup, p.pidns), "statm": newStatm(t, msrc), "status": newStatus(t, msrc, p.pidns), "uid_map": newUIDMap(t, msrc), } - if showSubtasks { + if isThreadGroup { contents["task"] = p.newSubtasks(t, msrc) } if len(p.cgroupControllers) > 0 { @@ -619,8 +618,11 @@ type ioData struct { ioUsage } -func newIO(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { - return newProcInode(t, seqfile.NewSeqFile(t, &ioData{t.ThreadGroup()}), msrc, fs.SpecialFile, t) +func newIO(t *kernel.Task, msrc *fs.MountSource, isThreadGroup bool) *fs.Inode { + if isThreadGroup { + return newProcInode(t, seqfile.NewSeqFile(t, &ioData{t.ThreadGroup()}), msrc, fs.SpecialFile, t) + } + return newProcInode(t, seqfile.NewSeqFile(t, &ioData{t}), msrc, fs.SpecialFile, t) } // NeedsUpdate returns whether the generation is old or not. diff --git a/test/syscalls/linux/proc.cc b/test/syscalls/linux/proc.cc index 512de5ee0..8cf08991b 100644 --- a/test/syscalls/linux/proc.cc +++ b/test/syscalls/linux/proc.cc @@ -37,6 +37,7 @@ #include #include #include +#include #include #include #include @@ -51,6 +52,7 @@ #include "absl/strings/str_split.h" #include "absl/strings/string_view.h" #include "absl/synchronization/mutex.h" +#include "absl/synchronization/notification.h" #include "absl/time/clock.h" #include "absl/time/time.h" #include "test/util/capability_util.h" @@ -1988,6 +1990,44 @@ TEST(Proc, GetdentsEnoent) { SyscallFailsWithErrno(ENOENT)); } +void CheckSyscwFromIOFile(const std::string& path, const std::string& regex) { + std::string output; + ASSERT_NO_ERRNO(GetContents(path, &output)); + ASSERT_THAT(output, ContainsRegex(absl::StrCat("syscw:\\s+", regex, "\n"))); +} + +// Checks that there is variable accounting of IO between threads/tasks. +TEST(Proc, PidTidIOAccounting) { + absl::Notification notification; + + // Run a thread with a bunch of writes. Check that io account records exactly + // the number of write calls. File open/close is there to prevent buffering. + ScopedThread writer([¬ification] { + const int num_writes = 100; + for (int i = 0; i < num_writes; i++) { + auto path = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); + ASSERT_NO_ERRNO(SetContents(path.path(), "a")); + } + notification.Notify(); + const std::string& writer_dir = + absl::StrCat("/proc/", getpid(), "/task/", gettid(), "/io"); + + CheckSyscwFromIOFile(writer_dir, std::to_string(num_writes)); + }); + + // Run a thread and do no writes. Check that no writes are recorded. + ScopedThread noop([¬ification] { + notification.WaitForNotification(); + const std::string& noop_dir = + absl::StrCat("/proc/", getpid(), "/task/", gettid(), "/io"); + + CheckSyscwFromIOFile(noop_dir, "0"); + }); + + writer.Join(); + noop.Join(); +} + } // namespace } // namespace testing } // namespace gvisor -- cgit v1.2.3 From 13f0f6069af4d49e236cbee4f0284c190784db37 Mon Sep 17 00:00:00 2001 From: Ian Gudger Date: Thu, 5 Dec 2019 17:27:15 -0800 Subject: Implement F_GETOWN_EX and F_SETOWN_EX. Some versions of glibc will convert F_GETOWN fcntl(2) calls into F_GETOWN_EX in some cases. PiperOrigin-RevId: 284089373 --- pkg/abi/linux/fcntl.go | 41 ++++++--- pkg/sentry/syscalls/linux/sys_file.go | 70 +++++++++++++-- test/syscalls/linux/BUILD | 1 + test/syscalls/linux/fcntl.cc | 162 +++++++++++++++++++++++++++++++++- test/syscalls/linux/ioctl.cc | 3 +- 5 files changed, 255 insertions(+), 22 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/fcntl.go b/pkg/abi/linux/fcntl.go index f78315ebf..6663a199c 100644 --- a/pkg/abi/linux/fcntl.go +++ b/pkg/abi/linux/fcntl.go @@ -16,15 +16,17 @@ package linux // Commands from linux/fcntl.h. const ( - F_DUPFD = 0x0 - F_GETFD = 0x1 - F_SETFD = 0x2 - F_GETFL = 0x3 - F_SETFL = 0x4 - F_SETLK = 0x6 - F_SETLKW = 0x7 - F_SETOWN = 0x8 - F_GETOWN = 0x9 + F_DUPFD = 0 + F_GETFD = 1 + F_SETFD = 2 + F_GETFL = 3 + F_SETFL = 4 + F_SETLK = 6 + F_SETLKW = 7 + F_SETOWN = 8 + F_GETOWN = 9 + F_SETOWN_EX = 15 + F_GETOWN_EX = 16 F_DUPFD_CLOEXEC = 1024 + 6 F_SETPIPE_SZ = 1024 + 7 F_GETPIPE_SZ = 1024 + 8 @@ -32,9 +34,9 @@ const ( // Commands for F_SETLK. const ( - F_RDLCK = 0x0 - F_WRLCK = 0x1 - F_UNLCK = 0x2 + F_RDLCK = 0 + F_WRLCK = 1 + F_UNLCK = 2 ) // Flags for fcntl. @@ -42,7 +44,7 @@ const ( FD_CLOEXEC = 00000001 ) -// Lock structure for F_SETLK. +// Flock is the lock structure for F_SETLK. type Flock struct { Type int16 Whence int16 @@ -52,3 +54,16 @@ type Flock struct { Pid int32 _ [4]byte } + +// Flags for F_SETOWN_EX and F_GETOWN_EX. +const ( + F_OWNER_TID = 0 + F_OWNER_PID = 1 + F_OWNER_PGRP = 2 +) + +// FOwnerEx is the owner structure for F_SETOWN_EX and F_GETOWN_EX. +type FOwnerEx struct { + Type int32 + PID int32 +} diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go index 3b9181002..9bc2445a5 100644 --- a/pkg/sentry/syscalls/linux/sys_file.go +++ b/pkg/sentry/syscalls/linux/sys_file.go @@ -840,25 +840,42 @@ func Dup3(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC return uintptr(newfd), nil, nil } -func fGetOwn(t *kernel.Task, file *fs.File) int32 { +func fGetOwnEx(t *kernel.Task, file *fs.File) linux.FOwnerEx { ma := file.Async(nil) if ma == nil { - return 0 + return linux.FOwnerEx{} } a := ma.(*fasync.FileAsync) ot, otg, opg := a.Owner() switch { case ot != nil: - return int32(t.PIDNamespace().IDOfTask(ot)) + return linux.FOwnerEx{ + Type: linux.F_OWNER_TID, + PID: int32(t.PIDNamespace().IDOfTask(ot)), + } case otg != nil: - return int32(t.PIDNamespace().IDOfThreadGroup(otg)) + return linux.FOwnerEx{ + Type: linux.F_OWNER_PID, + PID: int32(t.PIDNamespace().IDOfThreadGroup(otg)), + } case opg != nil: - return int32(-t.PIDNamespace().IDOfProcessGroup(opg)) + return linux.FOwnerEx{ + Type: linux.F_OWNER_PGRP, + PID: int32(t.PIDNamespace().IDOfProcessGroup(opg)), + } default: - return 0 + return linux.FOwnerEx{} } } +func fGetOwn(t *kernel.Task, file *fs.File) int32 { + owner := fGetOwnEx(t, file) + if owner.Type == linux.F_OWNER_PGRP { + return -owner.PID + } + return owner.PID +} + // fSetOwn sets the file's owner with the semantics of F_SETOWN in Linux. // // If who is positive, it represents a PID. If negative, it represents a PGID. @@ -901,11 +918,13 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall t.FDTable().SetFlags(fd, kernel.FDFlags{ CloseOnExec: flags&linux.FD_CLOEXEC != 0, }) + return 0, nil, nil case linux.F_GETFL: return uintptr(file.Flags().ToLinux()), nil, nil case linux.F_SETFL: flags := uint(args[2].Uint()) file.SetFlags(linuxToFlags(flags).Settable()) + return 0, nil, nil case linux.F_SETLK, linux.F_SETLKW: // In Linux the file system can choose to provide lock operations for an inode. // Normally pipe and socket types lack lock operations. We diverge and use a heavy @@ -1008,6 +1027,44 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall case linux.F_SETOWN: fSetOwn(t, file, args[2].Int()) return 0, nil, nil + case linux.F_GETOWN_EX: + addr := args[2].Pointer() + owner := fGetOwnEx(t, file) + _, err := t.CopyOut(addr, &owner) + return 0, nil, err + case linux.F_SETOWN_EX: + addr := args[2].Pointer() + var owner linux.FOwnerEx + n, err := t.CopyIn(addr, &owner) + if err != nil { + return 0, nil, err + } + a := file.Async(fasync.New).(*fasync.FileAsync) + switch owner.Type { + case linux.F_OWNER_TID: + task := t.PIDNamespace().TaskWithID(kernel.ThreadID(owner.PID)) + if task == nil { + return 0, nil, syserror.ESRCH + } + a.SetOwnerTask(t, task) + return uintptr(n), nil, nil + case linux.F_OWNER_PID: + tg := t.PIDNamespace().ThreadGroupWithID(kernel.ThreadID(owner.PID)) + if tg == nil { + return 0, nil, syserror.ESRCH + } + a.SetOwnerThreadGroup(t, tg) + return uintptr(n), nil, nil + case linux.F_OWNER_PGRP: + pg := t.PIDNamespace().ProcessGroupWithID(kernel.ProcessGroupID(owner.PID)) + if pg == nil { + return 0, nil, syserror.ESRCH + } + a.SetOwnerProcessGroup(t, pg) + return uintptr(n), nil, nil + default: + return 0, nil, syserror.EINVAL + } case linux.F_GET_SEALS: val, err := tmpfs.GetSeals(file.Dirent.Inode) return uintptr(val), nil, err @@ -1035,7 +1092,6 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // Everything else is not yet supported. return 0, nil, syserror.EINVAL } - return 0, nil, nil } const ( diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 7ce2e6270..61f310db9 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -753,6 +753,7 @@ cc_binary( "//test/util:eventfd_util", "//test/util:multiprocess_util", "//test/util:posix_error", + "//test/util:save_util", "//test/util:temp_path", "//test/util:test_util", "//test/util:timer_util", diff --git a/test/syscalls/linux/fcntl.cc b/test/syscalls/linux/fcntl.cc index 8a45be12a..4f3aa81d6 100644 --- a/test/syscalls/linux/fcntl.cc +++ b/test/syscalls/linux/fcntl.cc @@ -14,6 +14,7 @@ #include #include +#include #include #include @@ -32,6 +33,7 @@ #include "test/util/eventfd_util.h" #include "test/util/multiprocess_util.h" #include "test/util/posix_error.h" +#include "test/util/save_util.h" #include "test/util/temp_path.h" #include "test/util/test_util.h" #include "test/util/timer_util.h" @@ -910,8 +912,166 @@ TEST(FcntlTest, GetOwn) { FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); - ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN), + EXPECT_EQ(syscall(__NR_fcntl, s.get(), F_GETOWN), 0); + MaybeSave(); +} + +TEST(FcntlTest, GetOwnEx) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + f_owner_ex owner = {}; + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN_EX, &owner), + SyscallSucceedsWithValue(0)); +} + +TEST(FcntlTest, SetOwnExInvalidType) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + f_owner_ex owner = {}; + owner.type = __pid_type(-1); + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), + SyscallFailsWithErrno(EINVAL)); +} + +TEST(FcntlTest, SetOwnExInvalidTid) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + f_owner_ex owner = {}; + owner.type = F_OWNER_TID; + owner.pid = -1; + + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), + SyscallFailsWithErrno(ESRCH)); +} + +TEST(FcntlTest, SetOwnExInvalidPid) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + f_owner_ex owner = {}; + owner.type = F_OWNER_PID; + owner.pid = -1; + + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), + SyscallFailsWithErrno(ESRCH)); +} + +TEST(FcntlTest, SetOwnExInvalidPgrp) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + f_owner_ex owner = {}; + owner.type = F_OWNER_PGRP; + owner.pid = -1; + + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), + SyscallFailsWithErrno(ESRCH)); +} + +TEST(FcntlTest, SetOwnExTid) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + f_owner_ex owner = {}; + owner.type = F_OWNER_TID; + EXPECT_THAT(owner.pid = syscall(__NR_gettid), SyscallSucceeds()); + + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), + SyscallSucceeds()); + + EXPECT_EQ(syscall(__NR_fcntl, s.get(), F_GETOWN), owner.pid); + MaybeSave(); +} + +TEST(FcntlTest, SetOwnExPid) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + f_owner_ex owner = {}; + owner.type = F_OWNER_PID; + EXPECT_THAT(owner.pid = getpid(), SyscallSucceeds()); + + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), + SyscallSucceeds()); + + EXPECT_EQ(syscall(__NR_fcntl, s.get(), F_GETOWN), owner.pid); + MaybeSave(); +} + +TEST(FcntlTest, SetOwnExPgrp) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + f_owner_ex owner = {}; + owner.type = F_OWNER_PGRP; + EXPECT_THAT(owner.pid = getpgrp(), SyscallSucceeds()); + + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), + SyscallSucceeds()); + + // NOTE(igudger): I don't understand why, but this is flaky on Linux. + // GetOwnExPgrp (below) does not have this issue. + SKIP_IF(!IsRunningOnGvisor()); + + EXPECT_EQ(syscall(__NR_fcntl, s.get(), F_GETOWN), -owner.pid); + MaybeSave(); +} + +TEST(FcntlTest, GetOwnExTid) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + f_owner_ex set_owner = {}; + set_owner.type = F_OWNER_TID; + EXPECT_THAT(set_owner.pid = syscall(__NR_gettid), SyscallSucceeds()); + + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &set_owner), + SyscallSucceeds()); + + f_owner_ex got_owner = {}; + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN_EX, &got_owner), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(got_owner.type, set_owner.type); + EXPECT_EQ(got_owner.pid, set_owner.pid); +} + +TEST(FcntlTest, GetOwnExPid) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + f_owner_ex set_owner = {}; + set_owner.type = F_OWNER_PID; + EXPECT_THAT(set_owner.pid = getpid(), SyscallSucceeds()); + + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &set_owner), + SyscallSucceeds()); + + f_owner_ex got_owner = {}; + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN_EX, &got_owner), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(got_owner.type, set_owner.type); + EXPECT_EQ(got_owner.pid, set_owner.pid); +} + +TEST(FcntlTest, GetOwnExPgrp) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + f_owner_ex set_owner = {}; + set_owner.type = F_OWNER_PGRP; + EXPECT_THAT(set_owner.pid = getpgrp(), SyscallSucceeds()); + + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &set_owner), + SyscallSucceeds()); + + f_owner_ex got_owner = {}; + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN_EX, &got_owner), SyscallSucceedsWithValue(0)); + EXPECT_EQ(got_owner.type, set_owner.type); + EXPECT_EQ(got_owner.pid, set_owner.pid); } } // namespace diff --git a/test/syscalls/linux/ioctl.cc b/test/syscalls/linux/ioctl.cc index c4f8bff08..b0a07a064 100644 --- a/test/syscalls/linux/ioctl.cc +++ b/test/syscalls/linux/ioctl.cc @@ -215,7 +215,8 @@ TEST_F(IoctlTest, FIOASYNCSelfTarget2) { auto mask_cleanup = ASSERT_NO_ERRNO_AND_VALUE(ScopedSignalMask(SIG_UNBLOCK, SIGIO)); - pid_t pid = getpid(); + pid_t pid = -1; + EXPECT_THAT(pid = getpid(), SyscallSucceeds()); EXPECT_THAT(ioctl(pair->second_fd(), FIOSETOWN, &pid), SyscallSucceeds()); int set = 1; -- cgit v1.2.3 From f8bb3f79041bf819cdf803c1009a442154692301 Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Fri, 6 Dec 2019 08:35:18 -0800 Subject: Document ELF PT_LOAD difference from Linux PiperOrigin-RevId: 284191345 --- pkg/sentry/loader/elf.go | 2 ++ 1 file changed, 2 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/loader/elf.go b/pkg/sentry/loader/elf.go index c2c3ec06e..6299a3e2f 100644 --- a/pkg/sentry/loader/elf.go +++ b/pkg/sentry/loader/elf.go @@ -408,6 +408,8 @@ func loadParsedELF(ctx context.Context, m *mm.MemoryManager, f *fs.File, info el start = vaddr } if vaddr < end { + // NOTE(b/37474556): Linux allows out-of-order + // segments, in violation of the spec. ctx.Infof("PT_LOAD headers out-of-order. %#x < %#x", vaddr, end) return loadedELF{}, syserror.ENOEXEC } -- cgit v1.2.3 From 663fe840f79ac3d8e2ce1a1f1409d84cf2a9d37e Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Fri, 6 Dec 2019 14:32:53 -0800 Subject: Implement TTY field in control.Processes(). Threadgroups already know their TTY (if they have one), which now contains the TTY Index, and is returned in the Processes() call. PiperOrigin-RevId: 284263850 --- pkg/sentry/control/proc.go | 26 +++++++--- pkg/sentry/control/proc_test.go | 10 ++-- pkg/sentry/fs/tty/terminal.go | 4 +- pkg/sentry/kernel/tty.go | 11 +++++ runsc/container/container_test.go | 95 +++++++++++++++++++++++++++++++++++- runsc/container/test_app/BUILD | 1 + runsc/container/test_app/test_app.go | 40 +++++++++++++++ 7 files changed, 173 insertions(+), 14 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/control/proc.go b/pkg/sentry/control/proc.go index c35faeb4c..a6f90b2bb 100644 --- a/pkg/sentry/control/proc.go +++ b/pkg/sentry/control/proc.go @@ -268,7 +268,6 @@ func (proc *Proc) Ps(args *PsArgs, out *string) error { } // Process contains information about a single process in a Sandbox. -// TODO(b/117881927): Implement TTY field. type Process struct { UID auth.KUID `json:"uid"` PID kernel.ThreadID `json:"pid"` @@ -276,6 +275,9 @@ type Process struct { PPID kernel.ThreadID `json:"ppid"` // Processor utilization C int32 `json:"c"` + // TTY name of the process. Will be of the form "pts/N" if there is a + // TTY, or "?" if there is not. + TTY string `json:"tty"` // Start time STime string `json:"stime"` // CPU time @@ -285,18 +287,19 @@ type Process struct { } // ProcessListToTable prints a table with the following format: -// UID PID PPID C STIME TIME CMD -// 0 1 0 0 14:04 505262ns tail +// UID PID PPID C TTY STIME TIME CMD +// 0 1 0 0 pty/4 14:04 505262ns tail func ProcessListToTable(pl []*Process) string { var buf bytes.Buffer tw := tabwriter.NewWriter(&buf, 10, 1, 3, ' ', 0) - fmt.Fprint(tw, "UID\tPID\tPPID\tC\tSTIME\tTIME\tCMD") + fmt.Fprint(tw, "UID\tPID\tPPID\tC\tTTY\tSTIME\tTIME\tCMD") for _, d := range pl { - fmt.Fprintf(tw, "\n%d\t%d\t%d\t%d\t%s\t%s\t%s", + fmt.Fprintf(tw, "\n%d\t%d\t%d\t%d\t%s\t%s\t%s\t%s", d.UID, d.PID, d.PPID, d.C, + d.TTY, d.STime, d.Time, d.Cmd) @@ -347,7 +350,7 @@ func Processes(k *kernel.Kernel, containerID string, out *[]*Process) error { if p := tg.Leader().Parent(); p != nil { ppid = p.PIDNamespace().IDOfThreadGroup(p.ThreadGroup()) } - *out = append(*out, &Process{ + p := Process{ UID: tg.Leader().Credentials().EffectiveKUID, PID: pid, PPID: ppid, @@ -355,7 +358,9 @@ func Processes(k *kernel.Kernel, containerID string, out *[]*Process) error { C: percentCPU(tg.CPUStats(), tg.Leader().StartTime(), now), Time: tg.CPUStats().SysTime.String(), Cmd: tg.Leader().Name(), - }) + TTY: ttyName(tg.TTY()), + } + *out = append(*out, &p) } sort.Slice(*out, func(i, j int) bool { return (*out)[i].PID < (*out)[j].PID }) return nil @@ -395,3 +400,10 @@ func percentCPU(stats usage.CPUStats, startTime, now ktime.Time) int32 { } return int32(percentCPU) } + +func ttyName(tty *kernel.TTY) string { + if tty == nil { + return "?" + } + return fmt.Sprintf("pts/%d", tty.Index) +} diff --git a/pkg/sentry/control/proc_test.go b/pkg/sentry/control/proc_test.go index d8ada2694..0a88459b2 100644 --- a/pkg/sentry/control/proc_test.go +++ b/pkg/sentry/control/proc_test.go @@ -34,7 +34,7 @@ func TestProcessListTable(t *testing.T) { }{ { pl: []*Process{}, - expected: "UID PID PPID C STIME TIME CMD", + expected: "UID PID PPID C TTY STIME TIME CMD", }, { pl: []*Process{ @@ -43,6 +43,7 @@ func TestProcessListTable(t *testing.T) { PID: 0, PPID: 0, C: 0, + TTY: "?", STime: "0", Time: "0", Cmd: "zero", @@ -52,14 +53,15 @@ func TestProcessListTable(t *testing.T) { PID: 1, PPID: 1, C: 1, + TTY: "pts/4", STime: "1", Time: "1", Cmd: "one", }, }, - expected: `UID PID PPID C STIME TIME CMD -0 0 0 0 0 0 zero -1 1 1 1 1 1 one`, + expected: `UID PID PPID C TTY STIME TIME CMD +0 0 0 0 ? 0 0 zero +1 1 1 1 pts/4 1 1 one`, }, } diff --git a/pkg/sentry/fs/tty/terminal.go b/pkg/sentry/fs/tty/terminal.go index ff8138820..917f90cc0 100644 --- a/pkg/sentry/fs/tty/terminal.go +++ b/pkg/sentry/fs/tty/terminal.go @@ -53,8 +53,8 @@ func newTerminal(ctx context.Context, d *dirInodeOperations, n uint32) *Terminal d: d, n: n, ld: newLineDiscipline(termios), - masterKTTY: &kernel.TTY{}, - slaveKTTY: &kernel.TTY{}, + masterKTTY: &kernel.TTY{Index: n}, + slaveKTTY: &kernel.TTY{Index: n}, } t.EnableLeakCheck("tty.Terminal") return &t diff --git a/pkg/sentry/kernel/tty.go b/pkg/sentry/kernel/tty.go index 34f84487a..048de26dc 100644 --- a/pkg/sentry/kernel/tty.go +++ b/pkg/sentry/kernel/tty.go @@ -21,8 +21,19 @@ import "sync" // // +stateify savable type TTY struct { + // Index is the terminal index. It is immutable. + Index uint32 + mu sync.Mutex `state:"nosave"` // tg is protected by mu. tg *ThreadGroup } + +// TTY returns the thread group's controlling terminal. If nil, there is no +// controlling terminal. +func (tg *ThreadGroup) TTY() *TTY { + tg.signalHandlers.mu.Lock() + defer tg.signalHandlers.mu.Unlock() + return tg.tty +} diff --git a/runsc/container/container_test.go b/runsc/container/container_test.go index 07eacaac0..1d06f2780 100644 --- a/runsc/container/container_test.go +++ b/runsc/container/container_test.go @@ -98,10 +98,14 @@ func procListsEqual(got, want []*control.Process) bool { for i := range got { pd1 := got[i] pd2 := want[i] - // Zero out unimplemented and timing dependant fields. + // Zero out timing dependant fields. pd1.Time = "" pd1.STime = "" pd1.C = 0 + // Ignore TTY field too, since it's not relevant in the cases + // where we use this method. Tests that care about the TTY + // field should check for it themselves. + pd1.TTY = "" if *pd1 != *pd2 { return false } @@ -2112,6 +2116,95 @@ func TestOverlayfsStaleRead(t *testing.T) { } } +// TestTTYField checks TTY field returned by container.Processes(). +func TestTTYField(t *testing.T) { + stop := testutil.StartReaper() + defer stop() + + testApp, err := testutil.FindFile("runsc/container/test_app/test_app") + if err != nil { + t.Fatal("error finding test_app:", err) + } + + testCases := []struct { + name string + useTTY bool + wantTTYField string + }{ + { + name: "no tty", + useTTY: false, + wantTTYField: "?", + }, + { + name: "tty used", + useTTY: true, + wantTTYField: "pts/0", + }, + } + + for _, test := range testCases { + t.Run(test.name, func(t *testing.T) { + conf := testutil.TestConfig() + + // We will run /bin/sleep, possibly with an open TTY. + cmd := []string{"/bin/sleep", "10000"} + if test.useTTY { + // Run inside the "pty-runner". + cmd = append([]string{testApp, "pty-runner"}, cmd...) + } + + spec := testutil.NewSpecWithArgs(cmd...) + rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer os.RemoveAll(rootDir) + defer os.RemoveAll(bundleDir) + + // Create and start the container. + args := Args{ + ID: testutil.UniqueContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + c, err := New(conf, args) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer c.Destroy() + if err := c.Start(conf); err != nil { + t.Fatalf("error starting container: %v", err) + } + + // Wait for sleep to be running, and check the TTY + // field. + var gotTTYField string + cb := func() error { + ps, err := c.Processes() + if err != nil { + err = fmt.Errorf("error getting process data from container: %v", err) + return &backoff.PermanentError{Err: err} + } + for _, p := range ps { + if strings.Contains(p.Cmd, "sleep") { + gotTTYField = p.TTY + return nil + } + } + return fmt.Errorf("sleep not running") + } + if err := testutil.Poll(cb, 30*time.Second); err != nil { + t.Fatalf("error waiting for sleep process: %v", err) + } + + if gotTTYField != test.wantTTYField { + t.Errorf("tty field got %q, want %q", gotTTYField, test.wantTTYField) + } + }) + } +} + // executeSync synchronously executes a new process. func (cont *Container) executeSync(args *control.ExecArgs) (syscall.WaitStatus, error) { pid, err := cont.Execute(args) diff --git a/runsc/container/test_app/BUILD b/runsc/container/test_app/BUILD index 9bf9e6e9d..bfd338bb6 100644 --- a/runsc/container/test_app/BUILD +++ b/runsc/container/test_app/BUILD @@ -15,5 +15,6 @@ go_binary( "//pkg/unet", "//runsc/testutil", "@com_github_google_subcommands//:go_default_library", + "@com_github_kr_pty//:go_default_library", ], ) diff --git a/runsc/container/test_app/test_app.go b/runsc/container/test_app/test_app.go index 913d781c6..a1c8a741a 100644 --- a/runsc/container/test_app/test_app.go +++ b/runsc/container/test_app/test_app.go @@ -19,6 +19,7 @@ package main import ( "context" "fmt" + "io" "io/ioutil" "log" "net" @@ -31,6 +32,7 @@ import ( "flag" "github.com/google/subcommands" + "github.com/kr/pty" "gvisor.dev/gvisor/runsc/testutil" ) @@ -41,6 +43,7 @@ func main() { subcommands.Register(new(fdReceiver), "") subcommands.Register(new(fdSender), "") subcommands.Register(new(forkBomb), "") + subcommands.Register(new(ptyRunner), "") subcommands.Register(new(reaper), "") subcommands.Register(new(syscall), "") subcommands.Register(new(taskTree), "") @@ -352,3 +355,40 @@ func (c *capability) Execute(ctx context.Context, f *flag.FlagSet, args ...inter return subcommands.ExitSuccess } + +type ptyRunner struct{} + +// Name implements subcommands.Command. +func (*ptyRunner) Name() string { + return "pty-runner" +} + +// Synopsis implements subcommands.Command. +func (*ptyRunner) Synopsis() string { + return "runs the given command with an open pty terminal" +} + +// Usage implements subcommands.Command. +func (*ptyRunner) Usage() string { + return "pty-runner [command]" +} + +// SetFlags implements subcommands.Command.SetFlags. +func (*ptyRunner) SetFlags(f *flag.FlagSet) {} + +// Execute implements subcommands.Command. +func (*ptyRunner) Execute(_ context.Context, fs *flag.FlagSet, _ ...interface{}) subcommands.ExitStatus { + c := exec.Command(fs.Args()[0], fs.Args()[1:]...) + f, err := pty.Start(c) + if err != nil { + fmt.Printf("pty.Start failed: %v", err) + return subcommands.ExitFailure + } + defer f.Close() + + // Copy stdout from the command to keep this process alive until the + // subprocess exits. + io.Copy(os.Stdout, f) + + return subcommands.ExitSuccess +} -- cgit v1.2.3 From 371e210b83c244d8828ad2fa1b3d7cef15fbf463 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Fri, 6 Dec 2019 16:58:28 -0800 Subject: Add runtime tracing. This adds meaningful annotations to the trace generated by the runtime/trace package. PiperOrigin-RevId: 284290115 --- pkg/sentry/control/pprof.go | 15 ++++++- pkg/sentry/kernel/kernel.go | 20 ++++++++- pkg/sentry/kernel/syscalls.go | 8 ++++ pkg/sentry/kernel/task.go | 20 +++++---- pkg/sentry/kernel/task_block.go | 8 +++- pkg/sentry/kernel/task_clone.go | 1 + pkg/sentry/kernel/task_exec.go | 3 +- pkg/sentry/kernel/task_exit.go | 1 + pkg/sentry/kernel/task_log.go | 86 +++++++++++++++++++++++++++++++++++++-- pkg/sentry/kernel/task_run.go | 14 +++++++ pkg/sentry/kernel/task_start.go | 8 ++-- pkg/sentry/kernel/task_syscall.go | 8 ++++ runsc/boot/controller.go | 4 +- runsc/cmd/debug.go | 29 +++++++------ scripts/dev.sh | 3 +- 15 files changed, 190 insertions(+), 38 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/control/pprof.go b/pkg/sentry/control/pprof.go index 1f78d54a2..e1f2fea60 100644 --- a/pkg/sentry/control/pprof.go +++ b/pkg/sentry/control/pprof.go @@ -22,6 +22,7 @@ import ( "sync" "gvisor.dev/gvisor/pkg/fd" + "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/urpc" ) @@ -56,6 +57,9 @@ type Profile struct { // traceFile is the current execution trace output file. traceFile *fd.FD + + // Kernel is the kernel under profile. + Kernel *kernel.Kernel } // StartCPUProfile is an RPC stub which starts recording the CPU profile in a @@ -147,6 +151,9 @@ func (p *Profile) StartTrace(o *ProfileOpts, _ *struct{}) error { return err } + // Ensure all trace contexts are registered. + p.Kernel.RebuildTraceContexts() + p.traceFile = output return nil } @@ -158,9 +165,15 @@ func (p *Profile) StopTrace(_, _ *struct{}) error { defer p.mu.Unlock() if p.traceFile == nil { - return errors.New("Execution tracing not start") + return errors.New("Execution tracing not started") } + // Similarly to the case above, if tasks have not ended traces, we will + // lose information. Thus we need to rebuild the tasks in order to have + // complete information. This will not lose information if multiple + // traces are overlapping. + p.Kernel.RebuildTraceContexts() + trace.Stop() p.traceFile.Close() p.traceFile = nil diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 28ba950bd..bd3fb4c03 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -841,9 +841,11 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, AbstractSocketNamespace: args.AbstractSocketNamespace, ContainerID: args.ContainerID, } - if _, err := k.tasks.NewTask(config); err != nil { + t, err := k.tasks.NewTask(config) + if err != nil { return nil, 0, err } + t.traceExecEvent(tc) // Simulate exec for tracing. // Success. tgid := k.tasks.Root.IDOfThreadGroup(tg) @@ -1118,6 +1120,22 @@ func (k *Kernel) SendContainerSignal(cid string, info *arch.SignalInfo) error { return lastErr } +// RebuildTraceContexts rebuilds the trace context for all tasks. +// +// Unfortunately, if these are built while tracing is not enabled, then we will +// not have meaningful trace data. Rebuilding here ensures that we can do so +// after tracing has been enabled. +func (k *Kernel) RebuildTraceContexts() { + k.extMu.Lock() + defer k.extMu.Unlock() + k.tasks.mu.RLock() + defer k.tasks.mu.RUnlock() + + for t, tid := range k.tasks.Root.tids { + t.rebuildTraceContext(tid) + } +} + // FeatureSet returns the FeatureSet. func (k *Kernel) FeatureSet() *cpuid.FeatureSet { return k.featureSet diff --git a/pkg/sentry/kernel/syscalls.go b/pkg/sentry/kernel/syscalls.go index 220fa73a2..2fdee0282 100644 --- a/pkg/sentry/kernel/syscalls.go +++ b/pkg/sentry/kernel/syscalls.go @@ -339,6 +339,14 @@ func (s *SyscallTable) Lookup(sysno uintptr) SyscallFn { return nil } +// LookupName looks up a syscall name. +func (s *SyscallTable) LookupName(sysno uintptr) string { + if sc, ok := s.Table[sysno]; ok { + return sc.Name + } + return fmt.Sprintf("sys_%d", sysno) // Unlikely. +} + // LookupEmulate looks up an emulation syscall number. func (s *SyscallTable) LookupEmulate(addr usermem.Addr) (uintptr, bool) { sysno, ok := s.Emulate[addr] diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index 80c8e5464..ab0c6c4aa 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -15,6 +15,8 @@ package kernel import ( + gocontext "context" + "runtime/trace" "sync" "sync/atomic" @@ -390,7 +392,14 @@ type Task struct { // logPrefix is a string containing the task's thread ID in the root PID // namespace, and is prepended to log messages emitted by Task.Infof etc. - logPrefix atomic.Value `state:".(string)"` + logPrefix atomic.Value `state:"nosave"` + + // traceContext and traceTask are both used for tracing, and are + // updated along with the logPrefix in updateInfoLocked. + // + // These are exclusive to the task goroutine. + traceContext gocontext.Context `state:"nosave"` + traceTask *trace.Task `state:"nosave"` // creds is the task's credentials. // @@ -528,14 +537,6 @@ func (t *Task) loadPtraceTracer(tracer *Task) { t.ptraceTracer.Store(tracer) } -func (t *Task) saveLogPrefix() string { - return t.logPrefix.Load().(string) -} - -func (t *Task) loadLogPrefix(prefix string) { - t.logPrefix.Store(prefix) -} - func (t *Task) saveSyscallFilters() []bpf.Program { if f := t.syscallFilters.Load(); f != nil { return f.([]bpf.Program) @@ -549,6 +550,7 @@ func (t *Task) loadSyscallFilters(filters []bpf.Program) { // afterLoad is invoked by stateify. func (t *Task) afterLoad() { + t.updateInfoLocked() t.interruptChan = make(chan struct{}, 1) t.gosched.State = TaskGoroutineNonexistent if t.stop != nil { diff --git a/pkg/sentry/kernel/task_block.go b/pkg/sentry/kernel/task_block.go index dd69939f9..4a4a69ee2 100644 --- a/pkg/sentry/kernel/task_block.go +++ b/pkg/sentry/kernel/task_block.go @@ -16,6 +16,7 @@ package kernel import ( "runtime" + "runtime/trace" "time" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" @@ -133,19 +134,24 @@ func (t *Task) block(C <-chan struct{}, timerChan <-chan struct{}) error { runtime.Gosched() } + region := trace.StartRegion(t.traceContext, blockRegion) select { case <-C: + region.End() t.SleepFinish(true) + // Woken by event. return nil case <-interrupt: + region.End() t.SleepFinish(false) // Return the indicated error on interrupt. return syserror.ErrInterrupted case <-timerChan: - // We've timed out. + region.End() t.SleepFinish(true) + // We've timed out. return syserror.ETIMEDOUT } } diff --git a/pkg/sentry/kernel/task_clone.go b/pkg/sentry/kernel/task_clone.go index 0916fd658..3eadfedb4 100644 --- a/pkg/sentry/kernel/task_clone.go +++ b/pkg/sentry/kernel/task_clone.go @@ -299,6 +299,7 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { // nt that it must receive before its task goroutine starts running. tid := nt.k.tasks.Root.IDOfTask(nt) defer nt.Start(tid) + t.traceCloneEvent(tid) // "If fork/clone and execve are allowed by @prog, any child processes will // be constrained to the same filters and system call ABI as the parent." - diff --git a/pkg/sentry/kernel/task_exec.go b/pkg/sentry/kernel/task_exec.go index 17a089b90..90a6190f1 100644 --- a/pkg/sentry/kernel/task_exec.go +++ b/pkg/sentry/kernel/task_exec.go @@ -129,6 +129,7 @@ type runSyscallAfterExecStop struct { } func (r *runSyscallAfterExecStop) execute(t *Task) taskRunState { + t.traceExecEvent(r.tc) t.tg.pidns.owner.mu.Lock() t.tg.execing = nil if t.killed() { @@ -253,7 +254,7 @@ func (t *Task) promoteLocked() { t.tg.leader = t t.Infof("Becoming TID %d (in root PID namespace)", t.tg.pidns.owner.Root.tids[t]) - t.updateLogPrefixLocked() + t.updateInfoLocked() // Reap the original leader. If it has a tracer, detach it instead of // waiting for it to acknowledge the original leader's death. oldLeader.exitParentNotified = true diff --git a/pkg/sentry/kernel/task_exit.go b/pkg/sentry/kernel/task_exit.go index 535f03e50..435761e5a 100644 --- a/pkg/sentry/kernel/task_exit.go +++ b/pkg/sentry/kernel/task_exit.go @@ -236,6 +236,7 @@ func (*runExit) execute(t *Task) taskRunState { type runExitMain struct{} func (*runExitMain) execute(t *Task) taskRunState { + t.traceExitEvent() lastExiter := t.exitThreadGroup() // If the task has a cleartid, and the thread group wasn't killed by a diff --git a/pkg/sentry/kernel/task_log.go b/pkg/sentry/kernel/task_log.go index a29e9b9eb..0fb3661de 100644 --- a/pkg/sentry/kernel/task_log.go +++ b/pkg/sentry/kernel/task_log.go @@ -16,6 +16,7 @@ package kernel import ( "fmt" + "runtime/trace" "sort" "gvisor.dev/gvisor/pkg/log" @@ -127,11 +128,88 @@ func (t *Task) debugDumpStack() { } } -// updateLogPrefix updates the task's cached log prefix to reflect its -// current thread ID. +// trace definitions. +// +// Note that all region names are prefixed by ':' in order to ensure that they +// are lexically ordered before all system calls, which use the naked system +// call name (e.g. "read") for maximum clarity. +const ( + traceCategory = "task" + runRegion = ":run" + blockRegion = ":block" + cpuidRegion = ":cpuid" + faultRegion = ":fault" +) + +// updateInfoLocked updates the task's cached log prefix and tracing +// information to reflect its current thread ID. // // Preconditions: The task's owning TaskSet.mu must be locked. -func (t *Task) updateLogPrefixLocked() { +func (t *Task) updateInfoLocked() { // Use the task's TID in the root PID namespace for logging. - t.logPrefix.Store(fmt.Sprintf("[% 4d] ", t.tg.pidns.owner.Root.tids[t])) + tid := t.tg.pidns.owner.Root.tids[t] + t.logPrefix.Store(fmt.Sprintf("[% 4d] ", tid)) + t.rebuildTraceContext(tid) +} + +// rebuildTraceContext rebuilds the trace context. +// +// Precondition: the passed tid must be the tid in the root namespace. +func (t *Task) rebuildTraceContext(tid ThreadID) { + // Re-initialize the trace context. + if t.traceTask != nil { + t.traceTask.End() + } + + // Note that we define the "task type" to be the dynamic TID. This does + // not align perfectly with the documentation for "tasks" in the + // tracing package. Tasks may be assumed to be bounded by analysis + // tools. However, if we just use a generic "task" type here, then the + // "user-defined tasks" page on the tracing dashboard becomes nearly + // unusable, as it loads all traces from all tasks. + // + // We can assume that the number of tasks in the system is not + // arbitrarily large (in general it won't be, especially for cases + // where we're collecting a brief profile), so using the TID is a + // reasonable compromise in this case. + t.traceContext, t.traceTask = trace.NewTask(t, fmt.Sprintf("tid:%d", tid)) +} + +// traceCloneEvent is called when a new task is spawned. +// +// ntid must be the new task's ThreadID in the root namespace. +func (t *Task) traceCloneEvent(ntid ThreadID) { + if !trace.IsEnabled() { + return + } + trace.Logf(t.traceContext, traceCategory, "spawn: %d", ntid) +} + +// traceExitEvent is called when a task exits. +func (t *Task) traceExitEvent() { + if !trace.IsEnabled() { + return + } + trace.Logf(t.traceContext, traceCategory, "exit status: 0x%x", t.exitStatus.Status()) +} + +// traceExecEvent is called when a task calls exec. +func (t *Task) traceExecEvent(tc *TaskContext) { + if !trace.IsEnabled() { + return + } + d := tc.MemoryManager.Executable() + if d == nil { + trace.Logf(t.traceContext, traceCategory, "exec: << unknown >>") + return + } + defer d.DecRef() + root := t.fsContext.RootDirectory() + if root == nil { + trace.Logf(t.traceContext, traceCategory, "exec: << no root directory >>") + return + } + defer root.DecRef() + n, _ := d.FullName(root) + trace.Logf(t.traceContext, traceCategory, "exec: %s", n) } diff --git a/pkg/sentry/kernel/task_run.go b/pkg/sentry/kernel/task_run.go index c92266c59..d97f8c189 100644 --- a/pkg/sentry/kernel/task_run.go +++ b/pkg/sentry/kernel/task_run.go @@ -17,6 +17,7 @@ package kernel import ( "bytes" "runtime" + "runtime/trace" "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" @@ -205,9 +206,11 @@ func (*runApp) execute(t *Task) taskRunState { t.tg.pidns.owner.mu.RUnlock() } + region := trace.StartRegion(t.traceContext, runRegion) t.accountTaskGoroutineEnter(TaskGoroutineRunningApp) info, at, err := t.p.Switch(t.MemoryManager().AddressSpace(), t.Arch(), t.rseqCPU) t.accountTaskGoroutineLeave(TaskGoroutineRunningApp) + region.End() if clearSinglestep { t.Arch().ClearSingleStep() @@ -225,6 +228,7 @@ func (*runApp) execute(t *Task) taskRunState { case platform.ErrContextSignalCPUID: // Is this a CPUID instruction? + region := trace.StartRegion(t.traceContext, cpuidRegion) expected := arch.CPUIDInstruction[:] found := make([]byte, len(expected)) _, err := t.CopyIn(usermem.Addr(t.Arch().IP()), &found) @@ -232,10 +236,12 @@ func (*runApp) execute(t *Task) taskRunState { // Skip the cpuid instruction. t.Arch().CPUIDEmulate(t) t.Arch().SetIP(t.Arch().IP() + uintptr(len(expected))) + region.End() // Resume execution. return (*runApp)(nil) } + region.End() // Not an actual CPUID, but required copy-in. // The instruction at the given RIP was not a CPUID, and we // fallthrough to the default signal deliver behavior below. @@ -251,8 +257,10 @@ func (*runApp) execute(t *Task) taskRunState { // an application-generated signal and we should continue execution // normally. if at.Any() { + region := trace.StartRegion(t.traceContext, faultRegion) addr := usermem.Addr(info.Addr()) err := t.MemoryManager().HandleUserFault(t, addr, at, usermem.Addr(t.Arch().Stack())) + region.End() if err == nil { // The fault was handled appropriately. // We can resume running the application. @@ -260,6 +268,12 @@ func (*runApp) execute(t *Task) taskRunState { } // Is this a vsyscall that we need emulate? + // + // Note that we don't track vsyscalls as part of a + // specific trace region. This is because regions don't + // stack, and the actual system call will count as a + // region. We should be able to easily identify + // vsyscalls by having a pair. if at.Execute { if sysno, ok := t.tc.st.LookupEmulate(addr); ok { return t.doVsyscall(addr, sysno) diff --git a/pkg/sentry/kernel/task_start.go b/pkg/sentry/kernel/task_start.go index ae6fc4025..3522a4ae5 100644 --- a/pkg/sentry/kernel/task_start.go +++ b/pkg/sentry/kernel/task_start.go @@ -154,10 +154,10 @@ func (ts *TaskSet) newTask(cfg *TaskConfig) (*Task, error) { // Below this point, newTask is expected not to fail (there is no rollback // of assignTIDsLocked or any of the following). - // Logging on t's behalf will panic if t.logPrefix hasn't been initialized. - // This is the earliest point at which we can do so (since t now has thread - // IDs). - t.updateLogPrefixLocked() + // Logging on t's behalf will panic if t.logPrefix hasn't been + // initialized. This is the earliest point at which we can do so + // (since t now has thread IDs). + t.updateInfoLocked() if cfg.InheritParent != nil { t.parent = cfg.InheritParent.parent diff --git a/pkg/sentry/kernel/task_syscall.go b/pkg/sentry/kernel/task_syscall.go index b543d536a..3180f5560 100644 --- a/pkg/sentry/kernel/task_syscall.go +++ b/pkg/sentry/kernel/task_syscall.go @@ -17,6 +17,7 @@ package kernel import ( "fmt" "os" + "runtime/trace" "syscall" "gvisor.dev/gvisor/pkg/abi/linux" @@ -160,6 +161,10 @@ func (t *Task) executeSyscall(sysno uintptr, args arch.SyscallArguments) (rval u ctrl = ctrlStopAndReinvokeSyscall } else { fn := s.Lookup(sysno) + var region *trace.Region // Only non-nil if tracing == true. + if trace.IsEnabled() { + region = trace.StartRegion(t.traceContext, s.LookupName(sysno)) + } if fn != nil { // Call our syscall implementation. rval, ctrl, err = fn(t, args) @@ -167,6 +172,9 @@ func (t *Task) executeSyscall(sysno uintptr, args arch.SyscallArguments) (rval u // Use the missing function if not found. rval, err = t.SyscallTable().Missing(t, sysno, args) } + if region != nil { + region.End() + } } if bits.IsOn32(fe, ExternalAfterEnable) && (s.ExternalFilterAfter == nil || s.ExternalFilterAfter(t, sysno, args)) { diff --git a/runsc/boot/controller.go b/runsc/boot/controller.go index f62be4c59..9c9e94864 100644 --- a/runsc/boot/controller.go +++ b/runsc/boot/controller.go @@ -152,7 +152,9 @@ func newController(fd int, l *Loader) (*controller, error) { srv.Register(&debug{}) srv.Register(&control.Logging{}) if l.conf.ProfileEnable { - srv.Register(&control.Profile{}) + srv.Register(&control.Profile{ + Kernel: l.k, + }) } return &controller{ diff --git a/runsc/cmd/debug.go b/runsc/cmd/debug.go index 7313e473f..38da7ee02 100644 --- a/runsc/cmd/debug.go +++ b/runsc/cmd/debug.go @@ -32,16 +32,16 @@ import ( // Debug implements subcommands.Command for the "debug" command. type Debug struct { - pid int - stacks bool - signal int - profileHeap string - profileCPU string - profileDelay int - trace string - strace string - logLevel string - logPackets string + pid int + stacks bool + signal int + profileHeap string + profileCPU string + trace string + strace string + logLevel string + logPackets string + duration time.Duration } // Name implements subcommands.Command. @@ -65,7 +65,7 @@ func (d *Debug) SetFlags(f *flag.FlagSet) { f.BoolVar(&d.stacks, "stacks", false, "if true, dumps all sandbox stacks to the log") f.StringVar(&d.profileHeap, "profile-heap", "", "writes heap profile to the given file.") f.StringVar(&d.profileCPU, "profile-cpu", "", "writes CPU profile to the given file.") - f.IntVar(&d.profileDelay, "profile-delay", 5, "amount of time to wait before stoping CPU profile") + f.DurationVar(&d.duration, "duration", time.Second, "amount of time to wait for CPU and trace profiles") f.StringVar(&d.trace, "trace", "", "writes an execution trace to the given file.") f.IntVar(&d.signal, "signal", -1, "sends signal to the sandbox") f.StringVar(&d.strace, "strace", "", `A comma separated list of syscalls to trace. "all" enables all traces, "off" disables all`) @@ -163,7 +163,7 @@ func (d *Debug) Execute(_ context.Context, f *flag.FlagSet, args ...interface{}) if err := c.Sandbox.StartCPUProfile(f); err != nil { return Errorf(err.Error()) } - log.Infof("CPU profile started for %d sec, writing to %q", d.profileDelay, d.profileCPU) + log.Infof("CPU profile started for %v, writing to %q", d.duration, d.profileCPU) } if d.trace != "" { delay = true @@ -181,8 +181,7 @@ func (d *Debug) Execute(_ context.Context, f *flag.FlagSet, args ...interface{}) if err := c.Sandbox.StartTrace(f); err != nil { return Errorf(err.Error()) } - log.Infof("Tracing started for %d sec, writing to %q", d.profileDelay, d.trace) - + log.Infof("Tracing started for %v, writing to %q", d.duration, d.trace) } if d.strace != "" || len(d.logLevel) != 0 || len(d.logPackets) != 0 { @@ -243,7 +242,7 @@ func (d *Debug) Execute(_ context.Context, f *flag.FlagSet, args ...interface{}) } if delay { - time.Sleep(time.Duration(d.profileDelay) * time.Second) + time.Sleep(d.duration) } return subcommands.ExitSuccess diff --git a/scripts/dev.sh b/scripts/dev.sh index c67003018..6238b4d0b 100755 --- a/scripts/dev.sh +++ b/scripts/dev.sh @@ -54,9 +54,10 @@ declare OUTPUT="$(build //runsc)" if [[ ${REFRESH} -eq 0 ]]; then install_runsc "${RUNTIME}" --net-raw install_runsc "${RUNTIME}-d" --net-raw --debug --strace --log-packets + install_runsc "${RUNTIME}-p" --net-raw --profile echo - echo "Runtimes ${RUNTIME} and ${RUNTIME}-d (debug enabled) setup." + echo "Runtimes ${RUNTIME}, ${RUNTIME}-d (debug enabled), and ${RUNTIME}-p installed." echo "Use --runtime="${RUNTIME}" with your Docker command." echo " docker run --rm --runtime="${RUNTIME}" hello-world" echo -- cgit v1.2.3 From b1d44be7ad893bd6bdfd164a54a7142f4462414b Mon Sep 17 00:00:00 2001 From: Mithun Iyer Date: Fri, 6 Dec 2019 17:15:52 -0800 Subject: Add TCP stats for connection close and keep-alive timeouts. Fix bugs in updates to TCP CurrentEstablished stat. Fixes #1277 PiperOrigin-RevId: 284292459 --- pkg/sentry/socket/netstack/netstack.go | 2 ++ pkg/tcpip/tcpip.go | 8 ++++++ pkg/tcpip/transport/tcp/connect.go | 5 ++-- pkg/tcpip/transport/tcp/snd.go | 1 - pkg/tcpip/transport/tcp/tcp_test.go | 46 ++++++++++++++++++++++++++++++++++ 5 files changed, 58 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index d92399efd..fe5a46aa3 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -151,6 +151,8 @@ var Metrics = tcpip.Stats{ PassiveConnectionOpenings: mustCreateMetric("/netstack/tcp/passive_connection_openings", "Number of connections opened successfully via Listen."), CurrentEstablished: mustCreateMetric("/netstack/tcp/current_established", "Number of connections in either ESTABLISHED or CLOSE-WAIT state now."), EstablishedResets: mustCreateMetric("/netstack/tcp/established_resets", "Number of times TCP connections have made a direct transition to the CLOSED state from either the ESTABLISHED state or the CLOSE-WAIT state"), + EstablishedClosed: mustCreateMetric("/netstack/tcp/established_closed", "number of times established TCP connections made a transition to CLOSED state."), + EstablishedTimedout: mustCreateMetric("/netstack/tcp/established_timedout", "Number of times an established connection was reset because of keep-alive time out."), ListenOverflowSynDrop: mustCreateMetric("/netstack/tcp/listen_overflow_syn_drop", "Number of times the listen queue overflowed and a SYN was dropped."), ListenOverflowAckDrop: mustCreateMetric("/netstack/tcp/listen_overflow_ack_drop", "Number of times the listen queue overflowed and the final ACK in the handshake was dropped."), ListenOverflowSynCookieSent: mustCreateMetric("/netstack/tcp/listen_overflow_syn_cookie_sent", "Number of times a SYN cookie was sent."), diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 5746043cc..d5bb5b6ed 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -924,6 +924,14 @@ type TCPStats struct { // ESTABLISHED state or the CLOSE-WAIT state. EstablishedResets *StatCounter + // EstablishedClosed is the number of times established TCP connections + // made a transition to CLOSED state. + EstablishedClosed *StatCounter + + // EstablishedTimedout is the number of times an established connection + // was reset because of keep-alive time out. + EstablishedTimedout *StatCounter + // ListenOverflowSynDrop is the number of times the listen queue overflowed // and a SYN was dropped. ListenOverflowSynDrop *StatCounter diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 2975a1c3c..3d059c302 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -924,6 +924,7 @@ func (e *endpoint) transitionToStateCloseLocked() { } e.cleanupLocked() e.state = StateClose + e.stack.Stats().TCP.EstablishedClosed.Increment() } // tryDeliverSegmentFromClosedEndpoint attempts to deliver the parsed @@ -1094,6 +1095,7 @@ func (e *endpoint) keepaliveTimerExpired() *tcpip.Error { if e.keepalive.unacked >= e.keepalive.count { e.keepalive.Unlock() + e.stack.Stats().TCP.EstablishedTimedout.Increment() return tcpip.ErrTimeout } @@ -1179,8 +1181,6 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { e.lastErrorMu.Unlock() e.mu.Lock() - e.stack.Stats().TCP.EstablishedResets.Increment() - e.stack.Stats().TCP.CurrentEstablished.Decrement() e.state = StateError e.HardError = err @@ -1389,7 +1389,6 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { // Mark endpoint as closed. e.mu.Lock() if e.state != StateError { - e.stack.Stats().TCP.EstablishedResets.Increment() e.stack.Stats().TCP.CurrentEstablished.Decrement() e.transitionToStateCloseLocked() } diff --git a/pkg/tcpip/transport/tcp/snd.go b/pkg/tcpip/transport/tcp/snd.go index d3f7c9125..8332a0179 100644 --- a/pkg/tcpip/transport/tcp/snd.go +++ b/pkg/tcpip/transport/tcp/snd.go @@ -674,7 +674,6 @@ func (s *sender) maybeSendSegment(seg *segment, limit int, end seqnum.Value) (se default: s.ep.state = StateFinWait1 } - s.ep.stack.Stats().TCP.CurrentEstablished.Decrement() s.ep.mu.Unlock() } else { // We're sending a non-FIN segment. diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index 52c2fa7e3..bc5cfcf0e 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -75,6 +75,20 @@ func TestGiveUpConnect(t *testing.T) { if err := ep.GetSockOpt(tcpip.ErrorOption{}); err != tcpip.ErrAborted { t.Fatalf("got ep.GetSockOpt(tcpip.ErrorOption{}) = %v, want = %v", err, tcpip.ErrAborted) } + + // Call Connect again to retreive the handshake failure status + // and stats updates. + if err := ep.Connect(tcpip.FullAddress{Addr: context.TestAddr, Port: context.TestPort}); err != tcpip.ErrAborted { + t.Fatalf("got ep.Connect(...) = %v, want = %v", err, tcpip.ErrAborted) + } + + if got := c.Stack().Stats().TCP.FailedConnectionAttempts.Value(); got != 1 { + t.Errorf("got stats.TCP.FailedConnectionAttempts.Value() = %v, want = 1", got) + } + + if got := c.Stack().Stats().TCP.CurrentEstablished.Value(); got != 0 { + t.Errorf("got stats.TCP.CurrentEstablished.Value() = %v, want = 0", got) + } } func TestConnectIncrementActiveConnection(t *testing.T) { @@ -548,6 +562,14 @@ func TestClosingWithEnqueuedSegments(t *testing.T) { t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) } + if got := c.Stack().Stats().TCP.EstablishedClosed.Value(); got != 1 { + t.Errorf("got c.Stack().Stats().TCP.EstablishedClosed = %v, want = 1", got) + } + + if got := c.Stack().Stats().TCP.CurrentEstablished.Value(); got != 0 { + t.Errorf("got stats.TCP.CurrentEstablished.Value() = %v, want = 0", got) + } + // Check if the endpoint was moved to CLOSED and netstack a reset in // response to the ACK packet that we sent after last-ACK. checker.IPv4(t, c.GetPacket(), @@ -2694,6 +2716,13 @@ loop: if tcp.EndpointState(c.EP.State()) != tcp.StateError { t.Fatalf("got EP state is not StateError") } + + if got := c.Stack().Stats().TCP.EstablishedResets.Value(); got != 1 { + t.Errorf("got stats.TCP.EstablishedResets.Value() = %v, want = 1", got) + } + if got := c.Stack().Stats().TCP.CurrentEstablished.Value(); got != 0 { + t.Errorf("got stats.TCP.CurrentEstablished.Value() = %v, want = 0", got) + } } func TestSendOnResetConnection(t *testing.T) { @@ -4363,9 +4392,17 @@ func TestKeepalive(t *testing.T) { ), ) + if got := c.Stack().Stats().TCP.EstablishedTimedout.Value(); got != 1 { + t.Errorf("got c.Stack().Stats().TCP.EstablishedTimedout.Value() = %v, want = 1", got) + } + if _, _, err := c.EP.Read(nil); err != tcpip.ErrTimeout { t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrTimeout) } + + if got := c.Stack().Stats().TCP.CurrentEstablished.Value(); got != 0 { + t.Errorf("got stats.TCP.CurrentEstablished.Value() = %v, want = 0", got) + } } func executeHandshake(t *testing.T, c *context.Context, srcPort uint16, synCookieInUse bool) (irs, iss seqnum.Value) { @@ -5992,6 +6029,8 @@ func TestTCPTimeWaitDuplicateFINExtendsTimeWait(t *testing.T) { t.Fatalf("c.stack.SetTransportProtocolOption(tcp, tcpip.TCPLingerTimeoutOption(%d) failed: %s", tcpTimeWaitTimeout, err) } + want := c.Stack().Stats().TCP.EstablishedClosed.Value() + 1 + wq := &waiter.Queue{} ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, wq) if err != nil { @@ -6120,6 +6159,13 @@ func TestTCPTimeWaitDuplicateFINExtendsTimeWait(t *testing.T) { checker.SeqNum(uint32(ackHeaders.AckNum)), checker.AckNum(uint32(ackHeaders.SeqNum)), checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck))) + + if got := c.Stack().Stats().TCP.EstablishedClosed.Value(); got != want { + t.Errorf("got c.Stack().Stats().TCP.EstablishedClosed = %v, want = %v", got, want) + } + if got := c.Stack().Stats().TCP.CurrentEstablished.Value(); got != 0 { + t.Errorf("got stats.TCP.CurrentEstablished.Value() = %v, want = 0", got) + } } func TestTCPCloseWithData(t *testing.T) { -- cgit v1.2.3 From 01eadf51ea54b8f478c49b755d712f11fff2b28c Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Fri, 6 Dec 2019 23:08:39 -0800 Subject: Bump up Go 1.13 as minimum requirement PiperOrigin-RevId: 284320186 --- pkg/sentry/sighandling/sighandling.go | 75 +++++------------------- pkg/sentry/sighandling/sighandling_unsafe.go | 26 -------- pkg/syncutil/BUILD | 2 - pkg/syncutil/downgradable_rwmutex_1_12_unsafe.go | 21 ------- pkg/syncutil/downgradable_rwmutex_1_13_unsafe.go | 16 ----- pkg/syncutil/downgradable_rwmutex_unsafe.go | 5 +- runsc/boot/loader.go | 50 +++++++--------- 7 files changed, 41 insertions(+), 154 deletions(-) delete mode 100644 pkg/syncutil/downgradable_rwmutex_1_12_unsafe.go delete mode 100644 pkg/syncutil/downgradable_rwmutex_1_13_unsafe.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/sighandling/sighandling.go b/pkg/sentry/sighandling/sighandling.go index 2f65db70b..ba1f9043d 100644 --- a/pkg/sentry/sighandling/sighandling.go +++ b/pkg/sentry/sighandling/sighandling.go @@ -16,7 +16,6 @@ package sighandling import ( - "fmt" "os" "os/signal" "reflect" @@ -31,37 +30,25 @@ const numSignals = 32 // handleSignals listens for incoming signals and calls the given handler // function. // -// It starts when the start channel is closed, stops when the stop channel -// is closed, and closes done once it will no longer deliver signals to k. -func handleSignals(sigchans []chan os.Signal, handler func(linux.Signal), start, stop, done chan struct{}) { +// It stops when the stop channel is closed. The done channel is closed once it +// will no longer deliver signals to k. +func handleSignals(sigchans []chan os.Signal, handler func(linux.Signal), stop, done chan struct{}) { // Build a select case. - sc := []reflect.SelectCase{{Dir: reflect.SelectRecv, Chan: reflect.ValueOf(start)}} + sc := []reflect.SelectCase{{Dir: reflect.SelectRecv, Chan: reflect.ValueOf(stop)}} for _, sigchan := range sigchans { sc = append(sc, reflect.SelectCase{Dir: reflect.SelectRecv, Chan: reflect.ValueOf(sigchan)}) } - started := false for { // Wait for a notification. index, _, ok := reflect.Select(sc) - // Was it the start / stop channel? + // Was it the stop channel? if index == 0 { if !ok { - if !started { - // start channel; start forwarding and - // swap this case for the stop channel - // to select stop requests. - started = true - sc[0] = reflect.SelectCase{Dir: reflect.SelectRecv, Chan: reflect.ValueOf(stop)} - } else { - // stop channel; stop forwarding and - // clear this case so it is never - // selected again. - started = false - close(done) - sc[0].Chan = reflect.Value{} - } + // Stop forwarding and notify that it's done. + close(done) + return } continue } @@ -73,44 +60,17 @@ func handleSignals(sigchans []chan os.Signal, handler func(linux.Signal), start, // Otherwise, it was a signal on channel N. Index 0 represents the stop // channel, so index N represents the channel for signal N. - signal := linux.Signal(index) - - if !started { - // Kernel cannot receive signals, either because it is - // not ready yet or is shutting down. - // - // Kill ourselves if this signal would have killed the - // process before PrepareForwarding was called. i.e., all - // _SigKill signals; see Go - // src/runtime/sigtab_linux_generic.go. - // - // Otherwise ignore the signal. - // - // TODO(b/114489875): Drop in Go 1.12, which uses tgkill - // in runtime.raise. - switch signal { - case linux.SIGHUP, linux.SIGINT, linux.SIGTERM: - dieFromSignal(signal) - panic(fmt.Sprintf("Failed to die from signal %d", signal)) - default: - continue - } - } - - // Pass the signal to the handler. - handler(signal) + handler(linux.Signal(index)) } } -// PrepareHandler ensures that synchronous signals are passed to the given -// handler function and returns a callback that starts signal delivery, which -// itself returns a callback that stops signal handling. +// StartSignalForwarding ensures that synchronous signals are passed to the +// given handler function and returns a callback that stops signal delivery. // // Note that this function permanently takes over signal handling. After the // stop callback, signals revert to the default Go runtime behavior, which // cannot be overridden with external calls to signal.Notify. -func PrepareHandler(handler func(linux.Signal)) func() func() { - start := make(chan struct{}) +func StartSignalForwarding(handler func(linux.Signal)) func() { stop := make(chan struct{}) done := make(chan struct{}) @@ -128,13 +88,10 @@ func PrepareHandler(handler func(linux.Signal)) func() func() { signal.Notify(sigchan, syscall.Signal(sig)) } // Start up our listener. - go handleSignals(sigchans, handler, start, stop, done) // S/R-SAFE: synchronized by Kernel.extMu. + go handleSignals(sigchans, handler, stop, done) // S/R-SAFE: synchronized by Kernel.extMu. - return func() func() { - close(start) - return func() { - close(stop) - <-done - } + return func() { + close(stop) + <-done } } diff --git a/pkg/sentry/sighandling/sighandling_unsafe.go b/pkg/sentry/sighandling/sighandling_unsafe.go index c303435d5..1ebe22d34 100644 --- a/pkg/sentry/sighandling/sighandling_unsafe.go +++ b/pkg/sentry/sighandling/sighandling_unsafe.go @@ -15,8 +15,6 @@ package sighandling import ( - "fmt" - "runtime" "syscall" "unsafe" @@ -48,27 +46,3 @@ func IgnoreChildStop() error { return nil } - -// dieFromSignal kills the current process with sig. -// -// Preconditions: The default action of sig is termination. -func dieFromSignal(sig linux.Signal) { - runtime.LockOSThread() - defer runtime.UnlockOSThread() - - sa := sigaction{handler: linux.SIG_DFL} - if _, _, e := syscall.RawSyscall6(syscall.SYS_RT_SIGACTION, uintptr(sig), uintptr(unsafe.Pointer(&sa)), 0, linux.SignalSetSize, 0, 0); e != 0 { - panic(fmt.Sprintf("rt_sigaction failed: %v", e)) - } - - set := linux.MakeSignalSet(sig) - if _, _, e := syscall.RawSyscall6(syscall.SYS_RT_SIGPROCMASK, linux.SIG_UNBLOCK, uintptr(unsafe.Pointer(&set)), 0, linux.SignalSetSize, 0, 0); e != 0 { - panic(fmt.Sprintf("rt_sigprocmask failed: %v", e)) - } - - if err := syscall.Tgkill(syscall.Getpid(), syscall.Gettid(), syscall.Signal(sig)); err != nil { - panic(fmt.Sprintf("tgkill failed: %v", err)) - } - - panic("failed to die") -} diff --git a/pkg/syncutil/BUILD b/pkg/syncutil/BUILD index b06a90bef..cb1f41628 100644 --- a/pkg/syncutil/BUILD +++ b/pkg/syncutil/BUILD @@ -31,8 +31,6 @@ go_template( go_library( name = "syncutil", srcs = [ - "downgradable_rwmutex_1_12_unsafe.go", - "downgradable_rwmutex_1_13_unsafe.go", "downgradable_rwmutex_unsafe.go", "memmove_unsafe.go", "norace_unsafe.go", diff --git a/pkg/syncutil/downgradable_rwmutex_1_12_unsafe.go b/pkg/syncutil/downgradable_rwmutex_1_12_unsafe.go deleted file mode 100644 index 7c6336e62..000000000 --- a/pkg/syncutil/downgradable_rwmutex_1_12_unsafe.go +++ /dev/null @@ -1,21 +0,0 @@ -// Copyright 2009 The Go Authors. All rights reserved. -// Copyright 2019 The gVisor Authors. -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -// +build go1.12 -// +build !go1.13 - -// TODO(b/133868570): Delete once Go 1.12 is no longer supported. - -package syncutil - -import _ "unsafe" - -//go:linkname runtimeSemrelease112 sync.runtime_Semrelease -func runtimeSemrelease112(s *uint32, handoff bool) - -func runtimeSemrelease(s *uint32, handoff bool, skipframes int) { - // 'skipframes' is only available starting from 1.13. - runtimeSemrelease112(s, handoff) -} diff --git a/pkg/syncutil/downgradable_rwmutex_1_13_unsafe.go b/pkg/syncutil/downgradable_rwmutex_1_13_unsafe.go deleted file mode 100644 index 3c3673119..000000000 --- a/pkg/syncutil/downgradable_rwmutex_1_13_unsafe.go +++ /dev/null @@ -1,16 +0,0 @@ -// Copyright 2009 The Go Authors. All rights reserved. -// Copyright 2019 The gVisor Authors. -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -// +build go1.13 -// +build !go1.15 - -// Check go:linkname function signatures when updating Go version. - -package syncutil - -import _ "unsafe" - -//go:linkname runtimeSemrelease sync.runtime_Semrelease -func runtimeSemrelease(s *uint32, handoff bool, skipframes int) diff --git a/pkg/syncutil/downgradable_rwmutex_unsafe.go b/pkg/syncutil/downgradable_rwmutex_unsafe.go index 07feca402..51e11555d 100644 --- a/pkg/syncutil/downgradable_rwmutex_unsafe.go +++ b/pkg/syncutil/downgradable_rwmutex_unsafe.go @@ -3,7 +3,7 @@ // Use of this source code is governed by a BSD-style // license that can be found in the LICENSE file. -// +build go1.12 +// +build go1.13 // +build !go1.15 // Check go:linkname function signatures when updating Go version. @@ -27,6 +27,9 @@ import ( //go:linkname runtimeSemacquire sync.runtime_Semacquire func runtimeSemacquire(s *uint32) +//go:linkname runtimeSemrelease sync.runtime_Semrelease +func runtimeSemrelease(s *uint32, handoff bool, skipframes int) + // DowngradableRWMutex is identical to sync.RWMutex, but adds the DowngradeLock // method. type DowngradableRWMutex struct { diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index df6052c88..bc1d0c1bb 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -93,10 +93,6 @@ type Loader struct { // spec is the base configuration for the root container. spec *specs.Spec - // startSignalForwarding enables forwarding of signals to the sandboxed - // container. It should be called after the init process is loaded. - startSignalForwarding func() func() - // stopSignalForwarding disables forwarding of signals to the sandboxed // container. It should be called when a sandbox is destroyed. stopSignalForwarding func() @@ -336,29 +332,6 @@ func New(args Args) (*Loader, error) { return nil, fmt.Errorf("ignore child stop signals failed: %v", err) } - // Handle signals by forwarding them to the root container process - // (except for panic signal, which should cause a panic). - l.startSignalForwarding = sighandling.PrepareHandler(func(sig linux.Signal) { - // Panic signal should cause a panic. - if args.Conf.PanicSignal != -1 && sig == linux.Signal(args.Conf.PanicSignal) { - panic("Signal-induced panic") - } - - // Otherwise forward to root container. - deliveryMode := DeliverToProcess - if args.Console { - // Since we are running with a console, we should - // forward the signal to the foreground process group - // so that job control signals like ^C can be handled - // properly. - deliveryMode = DeliverToForegroundProcessGroup - } - log.Infof("Received external signal %d, mode: %v", sig, deliveryMode) - if err := l.signal(args.ID, 0, int32(sig), deliveryMode); err != nil { - log.Warningf("error sending signal %v to container %q: %v", sig, args.ID, err) - } - }) - // Create the control server using the provided FD. // // This must be done *after* we have initialized the kernel since the @@ -566,8 +539,27 @@ func (l *Loader) run() error { ep.tty.InitForegroundProcessGroup(ep.tg.ProcessGroup()) } - // Start signal forwarding only after an init process is created. - l.stopSignalForwarding = l.startSignalForwarding() + // Handle signals by forwarding them to the root container process + // (except for panic signal, which should cause a panic). + l.stopSignalForwarding = sighandling.StartSignalForwarding(func(sig linux.Signal) { + // Panic signal should cause a panic. + if l.conf.PanicSignal != -1 && sig == linux.Signal(l.conf.PanicSignal) { + panic("Signal-induced panic") + } + + // Otherwise forward to root container. + deliveryMode := DeliverToProcess + if l.console { + // Since we are running with a console, we should forward the signal to + // the foreground process group so that job control signals like ^C can + // be handled properly. + deliveryMode = DeliverToForegroundProcessGroup + } + log.Infof("Received external signal %d, mode: %v", sig, deliveryMode) + if err := l.signal(l.sandboxID, 0, int32(sig), deliveryMode); err != nil { + log.Warningf("error sending signal %v to container %q: %v", sig, l.sandboxID, err) + } + }) log.Infof("Process should have started...") l.watchdog.Start() -- cgit v1.2.3 From 898dcc2f839a975a9171271824af32176c2e5c27 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Mon, 9 Dec 2019 12:03:31 -0800 Subject: Redirect TODOs to gvisor.dev PiperOrigin-RevId: 284606233 --- pkg/sentry/fs/gofer/session.go | 6 +++--- pkg/sentry/kernel/semaphore/semaphore.go | 6 +++--- pkg/sentry/syscalls/linux/linux64_amd64.go | 2 +- pkg/sentry/syscalls/linux/linux64_arm64.go | 2 +- 4 files changed, 8 insertions(+), 8 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/gofer/session.go b/pkg/sentry/fs/gofer/session.go index 0da608548..4e358a46a 100644 --- a/pkg/sentry/fs/gofer/session.go +++ b/pkg/sentry/fs/gofer/session.go @@ -143,9 +143,9 @@ type session struct { // socket files. This allows unix domain sockets to be used with paths that // belong to a gofer. // - // TODO(b/77154739): there are few possible races with someone stat'ing the - // file and another deleting it concurrently, where the file will not be - // reported as socket file. + // TODO(gvisor.dev/issue/1200): there are few possible races with someone + // stat'ing the file and another deleting it concurrently, where the file + // will not be reported as socket file. endpoints *endpointMaps `state:"wait"` } diff --git a/pkg/sentry/kernel/semaphore/semaphore.go b/pkg/sentry/kernel/semaphore/semaphore.go index 93fe68a3e..de9617e9d 100644 --- a/pkg/sentry/kernel/semaphore/semaphore.go +++ b/pkg/sentry/kernel/semaphore/semaphore.go @@ -302,7 +302,7 @@ func (s *Set) SetVal(ctx context.Context, num int32, val int16, creds *auth.Cred return syserror.ERANGE } - // TODO(b/29354920): Clear undo entries in all processes + // TODO(gvisor.dev/issue/137): Clear undo entries in all processes. sem.value = val sem.pid = pid s.changeTime = ktime.NowFromContext(ctx) @@ -336,7 +336,7 @@ func (s *Set) SetValAll(ctx context.Context, vals []uint16, creds *auth.Credenti for i, val := range vals { sem := &s.sems[i] - // TODO(b/29354920): Clear undo entries in all processes + // TODO(gvisor.dev/issue/137): Clear undo entries in all processes. sem.value = int16(val) sem.pid = pid sem.wakeWaiters() @@ -481,7 +481,7 @@ func (s *Set) executeOps(ctx context.Context, ops []linux.Sembuf, pid int32) (ch } // All operations succeeded, apply them. - // TODO(b/29354920): handle undo operations. + // TODO(gvisor.dev/issue/137): handle undo operations. for i, v := range tmpVals { s.sems[i].value = v s.sems[i].wakeWaiters() diff --git a/pkg/sentry/syscalls/linux/linux64_amd64.go b/pkg/sentry/syscalls/linux/linux64_amd64.go index 81e4f93a6..5642d69ea 100644 --- a/pkg/sentry/syscalls/linux/linux64_amd64.go +++ b/pkg/sentry/syscalls/linux/linux64_amd64.go @@ -260,7 +260,7 @@ var AMD64 = &kernel.SyscallTable{ 217: syscalls.Supported("getdents64", Getdents64), 218: syscalls.Supported("set_tid_address", SetTidAddress), 219: syscalls.Supported("restart_syscall", RestartSyscall), - 220: syscalls.ErrorWithEvent("semtimedop", syserror.ENOSYS, "", []string{"gvisor.dev/issue/137"}), // TODO(b/29354920) + 220: syscalls.ErrorWithEvent("semtimedop", syserror.ENOSYS, "", []string{"gvisor.dev/issue/137"}), 221: syscalls.PartiallySupported("fadvise64", Fadvise64, "Not all options are supported.", nil), 222: syscalls.Supported("timer_create", TimerCreate), 223: syscalls.Supported("timer_settime", TimerSettime), diff --git a/pkg/sentry/syscalls/linux/linux64_arm64.go b/pkg/sentry/syscalls/linux/linux64_arm64.go index f1dd4b0c0..f897bfff8 100644 --- a/pkg/sentry/syscalls/linux/linux64_arm64.go +++ b/pkg/sentry/syscalls/linux/linux64_arm64.go @@ -224,7 +224,7 @@ var ARM64 = &kernel.SyscallTable{ 189: syscalls.ErrorWithEvent("msgsnd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) 190: syscalls.Supported("semget", Semget), 191: syscalls.PartiallySupported("semctl", Semctl, "Options IPC_INFO, SEM_INFO, IPC_STAT, SEM_STAT, SEM_STAT_ANY, GETNCNT, GETZCNT not supported.", nil), - 192: syscalls.ErrorWithEvent("semtimedop", syserror.ENOSYS, "", []string{"gvisor.dev/issue/137"}), // TODO(b/29354920) + 192: syscalls.ErrorWithEvent("semtimedop", syserror.ENOSYS, "", []string{"gvisor.dev/issue/137"}), 193: syscalls.PartiallySupported("semop", Semop, "Option SEM_UNDO not supported.", nil), 194: syscalls.PartiallySupported("shmget", Shmget, "Option SHM_HUGETLB is not supported.", nil), 195: syscalls.PartiallySupported("shmctl", Shmctl, "Options SHM_LOCK, SHM_UNLOCK are not supported.", nil), -- cgit v1.2.3 From 30f7316dc4a5fe0346c6e2e8e6854bd48a316a82 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 10 Dec 2019 09:59:36 -0800 Subject: Make comments clearer for control message handling. PiperOrigin-RevId: 284791600 --- pkg/sentry/socket/control/control.go | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/control/control.go b/pkg/sentry/socket/control/control.go index 782a3cb92..fa3188d51 100644 --- a/pkg/sentry/socket/control/control.go +++ b/pkg/sentry/socket/control/control.go @@ -195,15 +195,15 @@ func putCmsg(buf []byte, flags int, msgType uint32, align uint, data []int32) ([ // the available space, we must align down. // // align must be >= 4 and each data int32 is 4 bytes. The length of the - // header is already aligned, so if we align to the with of the data there + // header is already aligned, so if we align to the width of the data there // are two cases: // 1. The aligned length is less than the length of the header. The // unaligned length was also less than the length of the header, so we // can't write anything. // 2. The aligned length is greater than or equal to the length of the - // header. We can write the header plus zero or more datas. We can't write - // a partial int32, so the length of the message will be - // min(aligned length, header + datas). + // header. We can write the header plus zero or more bytes of data. We can't + // write a partial int32, so the length of the message will be + // min(aligned length, header + data). if space < linux.SizeOfControlMessageHeader { flags |= linux.MSG_CTRUNC return buf, flags @@ -240,12 +240,12 @@ func putCmsgStruct(buf []byte, msgLevel, msgType uint32, align uint, data interf buf = binary.Marshal(buf, usermem.ByteOrder, data) - // Check if we went over. + // If the control message data brought us over capacity, omit it. if cap(buf) != cap(ob) { return hdrBuf } - // Fix up length. + // Update control message length to include data. putUint64(ob, uint64(len(buf)-len(ob))) return alignSlice(buf, align) -- cgit v1.2.3 From c15be3f8cfde692e94dbb936fc4111bc1560502c Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Tue, 10 Dec 2019 10:25:30 -0800 Subject: Add all upstream syscalls to tables Package strace is missing some syscalls we actually implement (e.g., getrandom). We also see newer syscalls sometimes (e.g., membarrier) that would be handy to have formatted. Let's go ahead and add all syscalls in the latest upstream release (v5.4), even though we only intend to implement v4.4. None of them are implemented, just included as placeholders. PiperOrigin-RevId: 284797577 --- pkg/sentry/strace/linux64.go | 28 ++++++++++++++++++++++++++++ pkg/sentry/syscalls/linux/linux64_amd64.go | 22 +++++++++++++++++++++- pkg/sentry/syscalls/linux/linux64_arm64.go | 19 +++++++++++++++++++ 3 files changed, 68 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/strace/linux64.go b/pkg/sentry/strace/linux64.go index 5d57b75af..f2763b3f1 100644 --- a/pkg/sentry/strace/linux64.go +++ b/pkg/sentry/strace/linux64.go @@ -335,5 +335,33 @@ var linuxAMD64 = SyscallMap{ 315: makeSyscallInfo("sched_getattr", Hex, Hex, Hex), 316: makeSyscallInfo("renameat2", FD, Path, Hex, Path, Hex), 317: makeSyscallInfo("seccomp", Hex, Hex, Hex), + 318: makeSyscallInfo("getrandom", Hex, Hex, Hex), + 319: makeSyscallInfo("memfd_create", Path, Hex), // Not quite a path, but close. + 320: makeSyscallInfo("kexec_file_load", FD, FD, Hex, Hex, Hex), + 321: makeSyscallInfo("bpf", Hex, Hex, Hex), + 322: makeSyscallInfo("execveat", FD, Path, ExecveStringVector, ExecveStringVector, Hex), + 323: makeSyscallInfo("userfaultfd", Hex), + 324: makeSyscallInfo("membarrier", Hex, Hex), + 325: makeSyscallInfo("mlock2", Hex, Hex, Hex), + 326: makeSyscallInfo("copy_file_range", FD, Hex, FD, Hex, Hex, Hex), + 327: makeSyscallInfo("preadv2", FD, ReadIOVec, Hex, Hex, Hex), + 328: makeSyscallInfo("pwritev2", FD, WriteIOVec, Hex, Hex, Hex), + 329: makeSyscallInfo("pkey_mprotect", Hex, Hex, Hex, Hex), + 330: makeSyscallInfo("pkey_alloc", Hex, Hex), + 331: makeSyscallInfo("pkey_free", Hex), 332: makeSyscallInfo("statx", FD, Path, Hex, Hex, Hex), + 333: makeSyscallInfo("io_pgetevents", Hex, Hex, Hex, Hex, Timespec, SigSet), + 334: makeSyscallInfo("rseq", Hex, Hex, Hex, Hex), + 424: makeSyscallInfo("pidfd_send_signal", FD, Signal, Hex, Hex), + 425: makeSyscallInfo("io_uring_setup", Hex, Hex), + 426: makeSyscallInfo("io_uring_enter", FD, Hex, Hex, Hex, SigSet, Hex), + 427: makeSyscallInfo("io_uring_register", FD, Hex, Hex, Hex), + 428: makeSyscallInfo("open_tree", FD, Path, Hex), + 429: makeSyscallInfo("move_mount", FD, Path, FD, Path, Hex), + 430: makeSyscallInfo("fsopen", Path, Hex), // Not quite a path, but close. + 431: makeSyscallInfo("fsconfig", FD, Hex, Hex, Hex, Hex), + 432: makeSyscallInfo("fsmount", FD, Hex, Hex), + 433: makeSyscallInfo("fspick", FD, Path, Hex), + 434: makeSyscallInfo("pidfd_open", Hex, Hex), + 435: makeSyscallInfo("clone3", Hex, Hex), } diff --git a/pkg/sentry/syscalls/linux/linux64_amd64.go b/pkg/sentry/syscalls/linux/linux64_amd64.go index 5642d69ea..797542d28 100644 --- a/pkg/sentry/syscalls/linux/linux64_amd64.go +++ b/pkg/sentry/syscalls/linux/linux64_amd64.go @@ -367,11 +367,31 @@ var AMD64 = &kernel.SyscallTable{ 324: syscalls.ErrorWithEvent("membarrier", syserror.ENOSYS, "", []string{"gvisor.dev/issue/267"}), // TODO(gvisor.dev/issue/267) 325: syscalls.PartiallySupported("mlock2", Mlock2, "Stub implementation. The sandbox lacks appropriate permissions.", nil), - // Syscalls after 325 are "backports" from versions of Linux after 4.4. + // Syscalls implemented after 325 are "backports" from versions + // of Linux after 4.4. 326: syscalls.ErrorWithEvent("copy_file_range", syserror.ENOSYS, "", nil), 327: syscalls.Supported("preadv2", Preadv2), 328: syscalls.PartiallySupported("pwritev2", Pwritev2, "Flag RWF_HIPRI is not supported.", nil), + 329: syscalls.ErrorWithEvent("pkey_mprotect", syserror.ENOSYS, "", nil), + 330: syscalls.ErrorWithEvent("pkey_alloc", syserror.ENOSYS, "", nil), + 331: syscalls.ErrorWithEvent("pkey_free", syserror.ENOSYS, "", nil), 332: syscalls.Supported("statx", Statx), + 333: syscalls.ErrorWithEvent("io_pgetevents", syserror.ENOSYS, "", nil), + 334: syscalls.ErrorWithEvent("rseq", syserror.ENOSYS, "", nil), + + // Linux skips ahead to syscall 424 to sync numbers between arches. + 424: syscalls.ErrorWithEvent("pidfd_send_signal", syserror.ENOSYS, "", nil), + 425: syscalls.ErrorWithEvent("io_uring_setup", syserror.ENOSYS, "", nil), + 426: syscalls.ErrorWithEvent("io_uring_enter", syserror.ENOSYS, "", nil), + 427: syscalls.ErrorWithEvent("io_uring_register", syserror.ENOSYS, "", nil), + 428: syscalls.ErrorWithEvent("open_tree", syserror.ENOSYS, "", nil), + 429: syscalls.ErrorWithEvent("move_mount", syserror.ENOSYS, "", nil), + 430: syscalls.ErrorWithEvent("fsopen", syserror.ENOSYS, "", nil), + 431: syscalls.ErrorWithEvent("fsconfig", syserror.ENOSYS, "", nil), + 432: syscalls.ErrorWithEvent("fsmount", syserror.ENOSYS, "", nil), + 433: syscalls.ErrorWithEvent("fspick", syserror.ENOSYS, "", nil), + 434: syscalls.ErrorWithEvent("pidfd_open", syserror.ENOSYS, "", nil), + 435: syscalls.ErrorWithEvent("clone3", syserror.ENOSYS, "", nil), }, Emulate: map[usermem.Addr]uintptr{ diff --git a/pkg/sentry/syscalls/linux/linux64_arm64.go b/pkg/sentry/syscalls/linux/linux64_arm64.go index f897bfff8..2bc7faff5 100644 --- a/pkg/sentry/syscalls/linux/linux64_arm64.go +++ b/pkg/sentry/syscalls/linux/linux64_arm64.go @@ -302,7 +302,26 @@ var ARM64 = &kernel.SyscallTable{ 285: syscalls.ErrorWithEvent("copy_file_range", syserror.ENOSYS, "", nil), 286: syscalls.Supported("preadv2", Preadv2), 287: syscalls.PartiallySupported("pwritev2", Pwritev2, "Flag RWF_HIPRI is not supported.", nil), + 288: syscalls.ErrorWithEvent("pkey_mprotect", syserror.ENOSYS, "", nil), + 289: syscalls.ErrorWithEvent("pkey_alloc", syserror.ENOSYS, "", nil), + 290: syscalls.ErrorWithEvent("pkey_free", syserror.ENOSYS, "", nil), 291: syscalls.Supported("statx", Statx), + 292: syscalls.ErrorWithEvent("io_pgetevents", syserror.ENOSYS, "", nil), + 293: syscalls.ErrorWithEvent("rseq", syserror.ENOSYS, "", nil), + + // Linux skips ahead to syscall 424 to sync numbers between arches. + 424: syscalls.ErrorWithEvent("pidfd_send_signal", syserror.ENOSYS, "", nil), + 425: syscalls.ErrorWithEvent("io_uring_setup", syserror.ENOSYS, "", nil), + 426: syscalls.ErrorWithEvent("io_uring_enter", syserror.ENOSYS, "", nil), + 427: syscalls.ErrorWithEvent("io_uring_register", syserror.ENOSYS, "", nil), + 428: syscalls.ErrorWithEvent("open_tree", syserror.ENOSYS, "", nil), + 429: syscalls.ErrorWithEvent("move_mount", syserror.ENOSYS, "", nil), + 430: syscalls.ErrorWithEvent("fsopen", syserror.ENOSYS, "", nil), + 431: syscalls.ErrorWithEvent("fsconfig", syserror.ENOSYS, "", nil), + 432: syscalls.ErrorWithEvent("fsmount", syserror.ENOSYS, "", nil), + 433: syscalls.ErrorWithEvent("fspick", syserror.ENOSYS, "", nil), + 434: syscalls.ErrorWithEvent("pidfd_open", syserror.ENOSYS, "", nil), + 435: syscalls.ErrorWithEvent("clone3", syserror.ENOSYS, "", nil), }, Emulate: map[usermem.Addr]uintptr{}, -- cgit v1.2.3 From f6e87be82f189d7d2176dc2ca4a2d261481a032a Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 10 Dec 2019 10:56:51 -0800 Subject: Let socket.ControlMessages Release() the underlying transport.ControlMessages. PiperOrigin-RevId: 284804370 --- pkg/sentry/socket/socket.go | 5 +++++ pkg/sentry/syscalls/linux/sys_socket.go | 8 ++++---- 2 files changed, 9 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/socket.go b/pkg/sentry/socket/socket.go index 8c250c325..2389a9cdb 100644 --- a/pkg/sentry/socket/socket.go +++ b/pkg/sentry/socket/socket.go @@ -43,6 +43,11 @@ type ControlMessages struct { IP tcpip.ControlMessages } +// Release releases Unix domain socket credentials and rights. +func (c *ControlMessages) Release() { + c.Unix.Release() +} + // Socket is the interface containing socket syscalls used by the syscall layer // to redirect them to the appropriate implementation. type Socket interface { diff --git a/pkg/sentry/syscalls/linux/sys_socket.go b/pkg/sentry/syscalls/linux/sys_socket.go index d8acae063..e3faa890b 100644 --- a/pkg/sentry/syscalls/linux/sys_socket.go +++ b/pkg/sentry/syscalls/linux/sys_socket.go @@ -764,7 +764,7 @@ func recvSingleMsg(t *kernel.Task, s socket.Socket, msgPtr usermem.Addr, flags i } if !cms.Unix.Empty() { mflags |= linux.MSG_CTRUNC - cms.Unix.Release() + cms.Release() } if int(msg.Flags) != mflags { @@ -784,7 +784,7 @@ func recvSingleMsg(t *kernel.Task, s socket.Socket, msgPtr usermem.Addr, flags i if e != nil { return 0, syserror.ConvertIntr(e.ToError(), kernel.ERESTARTSYS) } - defer cms.Unix.Release() + defer cms.Release() controlData := make([]byte, 0, msg.ControlLen) @@ -885,7 +885,7 @@ func recvFrom(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flag } n, _, sender, senderLen, cm, e := s.RecvMsg(t, dst, int(flags), haveDeadline, deadline, nameLenPtr != 0, 0) - cm.Unix.Release() + cm.Release() if e != nil { return 0, syserror.ConvertIntr(e.ToError(), kernel.ERESTARTSYS) } @@ -1071,7 +1071,7 @@ func sendSingleMsg(t *kernel.Task, s socket.Socket, file *fs.File, msgPtr userme n, e := s.SendMsg(t, src, to, int(flags), haveDeadline, deadline, controlMessages) err = handleIOError(t, n != 0, e.ToError(), kernel.ERESTARTSYS, "sendmsg", file) if err != nil { - controlMessages.Unix.Release() + controlMessages.Release() } return uintptr(n), err } -- cgit v1.2.3 From 39386d78bb9636e52d6a0487d5fa7bff6beab64e Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 10 Dec 2019 13:04:32 -0800 Subject: Format fd_set parameters in select(2)/pselect(2) for strace. I1202 14:55:06.835076 7991 x:0] [ 1] select_test E select(0xa, 0x7fc6ce924c28 [0 1], null, null, 0x7fc6ce924c08 {sec=0 usec=0}) I1202 14:55:06.835102 7991 x:0] [ 1] select_test X select(0xa, 0x7fc6ce924c28 [0 1], null, null, 0x7fc6ce924c08 {sec=0 usec=0}) PiperOrigin-RevId: 284831805 --- pkg/sentry/strace/BUILD | 1 + pkg/sentry/strace/linux64.go | 4 +- pkg/sentry/strace/select.go | 53 ++++++++++++++++++++++++++ pkg/sentry/strace/strace.go | 2 + pkg/sentry/strace/syscalls.go | 4 ++ pkg/sentry/syscalls/linux/sys_poll.go | 71 +++++++++++++++++------------------ 6 files changed, 96 insertions(+), 39 deletions(-) create mode 100644 pkg/sentry/strace/select.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/strace/BUILD b/pkg/sentry/strace/BUILD index 72ebf766d..d46421199 100644 --- a/pkg/sentry/strace/BUILD +++ b/pkg/sentry/strace/BUILD @@ -14,6 +14,7 @@ go_library( "open.go", "poll.go", "ptrace.go", + "select.go", "signal.go", "socket.go", "strace.go", diff --git a/pkg/sentry/strace/linux64.go b/pkg/sentry/strace/linux64.go index f2763b3f1..e603f858f 100644 --- a/pkg/sentry/strace/linux64.go +++ b/pkg/sentry/strace/linux64.go @@ -40,7 +40,7 @@ var linuxAMD64 = SyscallMap{ 20: makeSyscallInfo("writev", FD, WriteIOVec, Hex), 21: makeSyscallInfo("access", Path, Oct), 22: makeSyscallInfo("pipe", PipeFDs), - 23: makeSyscallInfo("select", Hex, Hex, Hex, Hex, Timeval), + 23: makeSyscallInfo("select", Hex, SelectFDSet, SelectFDSet, SelectFDSet, Timeval), 24: makeSyscallInfo("sched_yield"), 25: makeSyscallInfo("mremap", Hex, Hex, Hex, Hex, Hex), 26: makeSyscallInfo("msync", Hex, Hex, Hex), @@ -287,7 +287,7 @@ var linuxAMD64 = SyscallMap{ 267: makeSyscallInfo("readlinkat", FD, Path, ReadBuffer, Hex), 268: makeSyscallInfo("fchmodat", FD, Path, Mode), 269: makeSyscallInfo("faccessat", FD, Path, Oct, Hex), - 270: makeSyscallInfo("pselect6", Hex, Hex, Hex, Hex, Hex, Hex), + 270: makeSyscallInfo("pselect6", Hex, SelectFDSet, SelectFDSet, SelectFDSet, Timespec, SigSet), 271: makeSyscallInfo("ppoll", PollFDs, Hex, Timespec, SigSet, Hex), 272: makeSyscallInfo("unshare", CloneFlags), 273: makeSyscallInfo("set_robust_list", Hex, Hex), diff --git a/pkg/sentry/strace/select.go b/pkg/sentry/strace/select.go new file mode 100644 index 000000000..92c18083d --- /dev/null +++ b/pkg/sentry/strace/select.go @@ -0,0 +1,53 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package strace + +import ( + "fmt" + + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" + "gvisor.dev/gvisor/pkg/sentry/usermem" +) + +func fdsFromSet(t *kernel.Task, set []byte) []int { + var fds []int + // Append n if the n-th bit is 1. + for i, v := range set { + for j := 0; j < 8; j++ { + if (v>>uint(j))&1 == 1 { + fds = append(fds, i*8+j) + } + } + } + return fds +} + +func fdSet(t *kernel.Task, nfds int, addr usermem.Addr) string { + if addr == 0 { + return "null" + } + + // Calculate the size of the fd set (one bit per fd). + nBytes := (nfds + 7) / 8 + nBitsInLastPartialByte := uint(nfds % 8) + + set, err := linux.CopyInFDSet(t, addr, nBytes, nBitsInLastPartialByte) + if err != nil { + return fmt.Sprintf("%#x (error decoding fdset: %s)", addr, err) + } + + return fmt.Sprintf("%#x %v", addr, fdsFromSet(t, set)) +} diff --git a/pkg/sentry/strace/strace.go b/pkg/sentry/strace/strace.go index 311389547..629c1f308 100644 --- a/pkg/sentry/strace/strace.go +++ b/pkg/sentry/strace/strace.go @@ -439,6 +439,8 @@ func (i *SyscallInfo) pre(t *kernel.Task, args arch.SyscallArguments, maximumBlo output = append(output, capData(t, args[arg-1].Pointer(), args[arg].Pointer())) case PollFDs: output = append(output, pollFDs(t, args[arg].Pointer(), uint(args[arg+1].Uint()), false)) + case SelectFDSet: + output = append(output, fdSet(t, int(args[0].Int()), args[arg].Pointer())) case Oct: output = append(output, "0o"+strconv.FormatUint(args[arg].Uint64(), 8)) case Hex: diff --git a/pkg/sentry/strace/syscalls.go b/pkg/sentry/strace/syscalls.go index 3c389d375..e5d486c4e 100644 --- a/pkg/sentry/strace/syscalls.go +++ b/pkg/sentry/strace/syscalls.go @@ -206,6 +206,10 @@ const ( // PollFDs is an array of struct pollfd. The number of entries in the // array is in the next argument. PollFDs + + // SelectFDSet is an fd_set argument in select(2)/pselect(2). The number of + // fds represented must be the first argument. + SelectFDSet ) // defaultFormat is the syscall argument format to use if the actual format is diff --git a/pkg/sentry/syscalls/linux/sys_poll.go b/pkg/sentry/syscalls/linux/sys_poll.go index 7a13beac2..631dffec6 100644 --- a/pkg/sentry/syscalls/linux/sys_poll.go +++ b/pkg/sentry/syscalls/linux/sys_poll.go @@ -197,53 +197,51 @@ func doPoll(t *kernel.Task, addr usermem.Addr, nfds uint, timeout time.Duration) return remainingTimeout, n, err } +// CopyInFDSet copies an fd set from select(2)/pselect(2). +func CopyInFDSet(t *kernel.Task, addr usermem.Addr, nBytes int, nBitsInLastPartialByte uint) ([]byte, error) { + set := make([]byte, nBytes) + + if addr != 0 { + if _, err := t.CopyIn(addr, &set); err != nil { + return nil, err + } + // If we only use part of the last byte, mask out the extraneous bits. + // + // N.B. This only works on little-endian architectures. + if nBitsInLastPartialByte != 0 { + set[nBytes-1] &^= byte(0xff) << nBitsInLastPartialByte + } + } + return set, nil +} + func doSelect(t *kernel.Task, nfds int, readFDs, writeFDs, exceptFDs usermem.Addr, timeout time.Duration) (uintptr, error) { if nfds < 0 || nfds > fileCap { return 0, syserror.EINVAL } - // Capture all the provided input vectors. - // - // N.B. This only works on little-endian architectures. - byteCount := (nfds + 7) / 8 - - bitsInLastPartialByte := uint(nfds % 8) - r := make([]byte, byteCount) - w := make([]byte, byteCount) - e := make([]byte, byteCount) + // Calculate the size of the fd sets (one bit per fd). + nBytes := (nfds + 7) / 8 + nBitsInLastPartialByte := uint(nfds % 8) - if readFDs != 0 { - if _, err := t.CopyIn(readFDs, &r); err != nil { - return 0, err - } - // Mask out bits above nfds. - if bitsInLastPartialByte != 0 { - r[byteCount-1] &^= byte(0xff) << bitsInLastPartialByte - } + // Capture all the provided input vectors. + r, err := CopyInFDSet(t, readFDs, nBytes, nBitsInLastPartialByte) + if err != nil { + return 0, err } - - if writeFDs != 0 { - if _, err := t.CopyIn(writeFDs, &w); err != nil { - return 0, err - } - if bitsInLastPartialByte != 0 { - w[byteCount-1] &^= byte(0xff) << bitsInLastPartialByte - } + w, err := CopyInFDSet(t, writeFDs, nBytes, nBitsInLastPartialByte) + if err != nil { + return 0, err } - - if exceptFDs != 0 { - if _, err := t.CopyIn(exceptFDs, &e); err != nil { - return 0, err - } - if bitsInLastPartialByte != 0 { - e[byteCount-1] &^= byte(0xff) << bitsInLastPartialByte - } + e, err := CopyInFDSet(t, exceptFDs, nBytes, nBitsInLastPartialByte) + if err != nil { + return 0, err } // Count how many FDs are actually being requested so that we can build // a PollFD array. fdCount := 0 - for i := 0; i < byteCount; i++ { + for i := 0; i < nBytes; i++ { v := r[i] | w[i] | e[i] for v != 0 { v &= (v - 1) @@ -254,7 +252,7 @@ func doSelect(t *kernel.Task, nfds int, readFDs, writeFDs, exceptFDs usermem.Add // Build the PollFD array. pfd := make([]linux.PollFD, 0, fdCount) var fd int32 - for i := 0; i < byteCount; i++ { + for i := 0; i < nBytes; i++ { rV, wV, eV := r[i], w[i], e[i] v := rV | wV | eV m := byte(1) @@ -295,8 +293,7 @@ func doSelect(t *kernel.Task, nfds int, readFDs, writeFDs, exceptFDs usermem.Add } // Do the syscall, then count the number of bits set. - _, _, err := pollBlock(t, pfd, timeout) - if err != nil { + if _, _, err = pollBlock(t, pfd, timeout); err != nil { return 0, syserror.ConvertIntr(err, syserror.EINTR) } -- cgit v1.2.3 From 46651a7d26559bdc69d460bdeb4de5968212d615 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Tue, 10 Dec 2019 18:16:47 -0800 Subject: Add most VFS methods for syscalls. PiperOrigin-RevId: 284892289 --- pkg/abi/linux/file.go | 10 +- pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go | 6 +- pkg/sentry/fsimpl/ext/ext_test.go | 29 +- pkg/sentry/fsimpl/memfs/benchmark_test.go | 2 +- pkg/sentry/fsimpl/memfs/pipe_test.go | 6 +- pkg/sentry/vfs/BUILD | 1 - pkg/sentry/vfs/file_description.go | 93 ++++++ pkg/sentry/vfs/file_description_impl_util_test.go | 10 +- pkg/sentry/vfs/filesystem.go | 22 ++ pkg/sentry/vfs/mount.go | 69 +++- pkg/sentry/vfs/options.go | 12 + pkg/sentry/vfs/syscalls.go | 237 -------------- pkg/sentry/vfs/vfs.go | 378 ++++++++++++++++++++++ 13 files changed, 606 insertions(+), 269 deletions(-) delete mode 100644 pkg/sentry/vfs/syscalls.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/file.go b/pkg/abi/linux/file.go index c9ee098f4..0f014d27f 100644 --- a/pkg/abi/linux/file.go +++ b/pkg/abi/linux/file.go @@ -144,9 +144,13 @@ const ( ModeCharacterDevice = S_IFCHR ModeNamedPipe = S_IFIFO - ModeSetUID = 04000 - ModeSetGID = 02000 - ModeSticky = 01000 + S_ISUID = 04000 + S_ISGID = 02000 + S_ISVTX = 01000 + + ModeSetUID = S_ISUID + ModeSetGID = S_ISGID + ModeSticky = S_ISVTX ModeUserAll = 0700 ModeUserRead = 0400 diff --git a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go index 94cd74095..177ce2cb9 100644 --- a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go +++ b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go @@ -81,7 +81,11 @@ func mount(b *testing.B, imagePath string, vfsfs *vfs.VirtualFilesystem, pop *vf ctx := contexttest.Context(b) creds := auth.CredentialsFromContext(ctx) - if err := vfsfs.NewMount(ctx, creds, imagePath, pop, "extfs", &vfs.GetFilesystemOptions{InternalData: int(f.Fd())}); err != nil { + if err := vfsfs.MountAt(ctx, creds, imagePath, pop, "extfs", &vfs.MountOptions{ + GetFilesystemOptions: vfs.GetFilesystemOptions{ + InternalData: int(f.Fd()), + }, + }); err != nil { b.Fatalf("failed to mount tmpfs submount: %v", err) } return func() { diff --git a/pkg/sentry/fsimpl/ext/ext_test.go b/pkg/sentry/fsimpl/ext/ext_test.go index 307e4d68c..e9f756732 100644 --- a/pkg/sentry/fsimpl/ext/ext_test.go +++ b/pkg/sentry/fsimpl/ext/ext_test.go @@ -147,55 +147,54 @@ func TestSeek(t *testing.T) { t.Fatalf("vfsfs.OpenAt failed: %v", err) } - if n, err := fd.Impl().Seek(ctx, 0, linux.SEEK_SET); n != 0 || err != nil { + if n, err := fd.Seek(ctx, 0, linux.SEEK_SET); n != 0 || err != nil { t.Errorf("expected seek position 0, got %d and error %v", n, err) } - stat, err := fd.Impl().Stat(ctx, vfs.StatOptions{}) + stat, err := fd.Stat(ctx, vfs.StatOptions{}) if err != nil { t.Errorf("fd.stat failed for file %s in image %s: %v", test.path, test.image, err) } // We should be able to seek beyond the end of file. size := int64(stat.Size) - if n, err := fd.Impl().Seek(ctx, size, linux.SEEK_SET); n != size || err != nil { + if n, err := fd.Seek(ctx, size, linux.SEEK_SET); n != size || err != nil { t.Errorf("expected seek position %d, got %d and error %v", size, n, err) } // EINVAL should be returned if the resulting offset is negative. - if _, err := fd.Impl().Seek(ctx, -1, linux.SEEK_SET); err != syserror.EINVAL { + if _, err := fd.Seek(ctx, -1, linux.SEEK_SET); err != syserror.EINVAL { t.Errorf("expected error EINVAL but got %v", err) } - if n, err := fd.Impl().Seek(ctx, 3, linux.SEEK_CUR); n != size+3 || err != nil { + if n, err := fd.Seek(ctx, 3, linux.SEEK_CUR); n != size+3 || err != nil { t.Errorf("expected seek position %d, got %d and error %v", size+3, n, err) } // Make sure negative offsets work with SEEK_CUR. - if n, err := fd.Impl().Seek(ctx, -2, linux.SEEK_CUR); n != size+1 || err != nil { + if n, err := fd.Seek(ctx, -2, linux.SEEK_CUR); n != size+1 || err != nil { t.Errorf("expected seek position %d, got %d and error %v", size+1, n, err) } // EINVAL should be returned if the resulting offset is negative. - if _, err := fd.Impl().Seek(ctx, -(size + 2), linux.SEEK_CUR); err != syserror.EINVAL { + if _, err := fd.Seek(ctx, -(size + 2), linux.SEEK_CUR); err != syserror.EINVAL { t.Errorf("expected error EINVAL but got %v", err) } // Make sure SEEK_END works with regular files. - switch fd.Impl().(type) { - case *regularFileFD: + if _, ok := fd.Impl().(*regularFileFD); ok { // Seek back to 0. - if n, err := fd.Impl().Seek(ctx, -size, linux.SEEK_END); n != 0 || err != nil { + if n, err := fd.Seek(ctx, -size, linux.SEEK_END); n != 0 || err != nil { t.Errorf("expected seek position %d, got %d and error %v", 0, n, err) } // Seek forward beyond EOF. - if n, err := fd.Impl().Seek(ctx, 1, linux.SEEK_END); n != size+1 || err != nil { + if n, err := fd.Seek(ctx, 1, linux.SEEK_END); n != size+1 || err != nil { t.Errorf("expected seek position %d, got %d and error %v", size+1, n, err) } // EINVAL should be returned if the resulting offset is negative. - if _, err := fd.Impl().Seek(ctx, -(size + 1), linux.SEEK_END); err != syserror.EINVAL { + if _, err := fd.Seek(ctx, -(size + 1), linux.SEEK_END); err != syserror.EINVAL { t.Errorf("expected error EINVAL but got %v", err) } } @@ -456,7 +455,7 @@ func TestRead(t *testing.T) { want := make([]byte, 1) for { n, err := f.Read(want) - fd.Impl().Read(ctx, usermem.BytesIOSequence(got), vfs.ReadOptions{}) + fd.Read(ctx, usermem.BytesIOSequence(got), vfs.ReadOptions{}) if diff := cmp.Diff(got, want); diff != "" { t.Errorf("file data mismatch (-want +got):\n%s", diff) @@ -464,7 +463,7 @@ func TestRead(t *testing.T) { // Make sure there is no more file data left after getting EOF. if n == 0 || err == io.EOF { - if n, _ := fd.Impl().Read(ctx, usermem.BytesIOSequence(got), vfs.ReadOptions{}); n != 0 { + if n, _ := fd.Read(ctx, usermem.BytesIOSequence(got), vfs.ReadOptions{}); n != 0 { t.Errorf("extra unexpected file data in file %s in image %s", test.absPath, test.image) } @@ -574,7 +573,7 @@ func TestIterDirents(t *testing.T) { } cb := &iterDirentsCb{} - if err = fd.Impl().IterDirents(ctx, cb); err != nil { + if err = fd.IterDirents(ctx, cb); err != nil { t.Fatalf("dir fd.IterDirents() failed: %v", err) } diff --git a/pkg/sentry/fsimpl/memfs/benchmark_test.go b/pkg/sentry/fsimpl/memfs/benchmark_test.go index ea6417ce7..4a7a94a52 100644 --- a/pkg/sentry/fsimpl/memfs/benchmark_test.go +++ b/pkg/sentry/fsimpl/memfs/benchmark_test.go @@ -394,7 +394,7 @@ func BenchmarkVFS2MemfsMountStat(b *testing.B) { } defer mountPoint.DecRef() // Create and mount the submount. - if err := vfsObj.NewMount(ctx, creds, "", &pop, "memfs", &vfs.GetFilesystemOptions{}); err != nil { + if err := vfsObj.MountAt(ctx, creds, "", &pop, "memfs", &vfs.MountOptions{}); err != nil { b.Fatalf("failed to mount tmpfs submount: %v", err) } filePathBuilder.WriteString(mountPointName) diff --git a/pkg/sentry/fsimpl/memfs/pipe_test.go b/pkg/sentry/fsimpl/memfs/pipe_test.go index a3a870571..5bf527c80 100644 --- a/pkg/sentry/fsimpl/memfs/pipe_test.go +++ b/pkg/sentry/fsimpl/memfs/pipe_test.go @@ -194,7 +194,7 @@ func setup(t *testing.T) (context.Context, *auth.Credentials, *vfs.VirtualFilesy func checkEmpty(ctx context.Context, t *testing.T, fd *vfs.FileDescription) { readData := make([]byte, 1) dst := usermem.BytesIOSequence(readData) - bytesRead, err := fd.Impl().Read(ctx, dst, vfs.ReadOptions{}) + bytesRead, err := fd.Read(ctx, dst, vfs.ReadOptions{}) if err != syserror.ErrWouldBlock { t.Fatalf("expected ErrWouldBlock reading from empty pipe %q, but got: %v", fileName, err) } @@ -207,7 +207,7 @@ func checkEmpty(ctx context.Context, t *testing.T, fd *vfs.FileDescription) { func checkWrite(ctx context.Context, t *testing.T, fd *vfs.FileDescription, msg string) { writeData := []byte(msg) src := usermem.BytesIOSequence(writeData) - bytesWritten, err := fd.Impl().Write(ctx, src, vfs.WriteOptions{}) + bytesWritten, err := fd.Write(ctx, src, vfs.WriteOptions{}) if err != nil { t.Fatalf("error writing to pipe %q: %v", fileName, err) } @@ -220,7 +220,7 @@ func checkWrite(ctx context.Context, t *testing.T, fd *vfs.FileDescription, msg func checkRead(ctx context.Context, t *testing.T, fd *vfs.FileDescription, msg string) { readData := make([]byte, len(msg)) dst := usermem.BytesIOSequence(readData) - bytesRead, err := fd.Impl().Read(ctx, dst, vfs.ReadOptions{}) + bytesRead, err := fd.Read(ctx, dst, vfs.ReadOptions{}) if err != nil { t.Fatalf("error reading from pipe %q: %v", fileName, err) } diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index 74a325309..59237c3b9 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -19,7 +19,6 @@ go_library( "options.go", "permissions.go", "resolving_path.go", - "syscalls.go", "testutil.go", "vfs.go", ], diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 34007eb57..4473dfce8 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -241,3 +241,96 @@ type IterDirentsCallback interface { // called. Handle(dirent Dirent) bool } + +// OnClose is called when a file descriptor representing the FileDescription is +// closed. Returning a non-nil error should not prevent the file descriptor +// from being closed. +func (fd *FileDescription) OnClose(ctx context.Context) error { + return fd.impl.OnClose(ctx) +} + +// StatusFlags returns file description status flags, as for fcntl(F_GETFL). +func (fd *FileDescription) StatusFlags(ctx context.Context) (uint32, error) { + flags, err := fd.impl.StatusFlags(ctx) + flags |= linux.O_LARGEFILE + return flags, err +} + +// SetStatusFlags sets file description status flags, as for fcntl(F_SETFL). +func (fd *FileDescription) SetStatusFlags(ctx context.Context, flags uint32) error { + return fd.impl.SetStatusFlags(ctx, flags) +} + +// Stat returns metadata for the file represented by fd. +func (fd *FileDescription) Stat(ctx context.Context, opts StatOptions) (linux.Statx, error) { + return fd.impl.Stat(ctx, opts) +} + +// SetStat updates metadata for the file represented by fd. +func (fd *FileDescription) SetStat(ctx context.Context, opts SetStatOptions) error { + return fd.impl.SetStat(ctx, opts) +} + +// StatFS returns metadata for the filesystem containing the file represented +// by fd. +func (fd *FileDescription) StatFS(ctx context.Context) (linux.Statfs, error) { + return fd.impl.StatFS(ctx) +} + +// PRead reads from the file represented by fd into dst, starting at the given +// offset, and returns the number of bytes read. PRead is permitted to return +// partial reads with a nil error. +func (fd *FileDescription) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts ReadOptions) (int64, error) { + return fd.impl.PRead(ctx, dst, offset, opts) +} + +// Read is similar to PRead, but does not specify an offset. +func (fd *FileDescription) Read(ctx context.Context, dst usermem.IOSequence, opts ReadOptions) (int64, error) { + return fd.impl.Read(ctx, dst, opts) +} + +// PWrite writes src to the file represented by fd, starting at the given +// offset, and returns the number of bytes written. PWrite is permitted to +// return partial writes with a nil error. +func (fd *FileDescription) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts WriteOptions) (int64, error) { + return fd.impl.PWrite(ctx, src, offset, opts) +} + +// Write is similar to PWrite, but does not specify an offset. +func (fd *FileDescription) Write(ctx context.Context, src usermem.IOSequence, opts WriteOptions) (int64, error) { + return fd.impl.Write(ctx, src, opts) +} + +// IterDirents invokes cb on each entry in the directory represented by fd. If +// IterDirents has been called since the last call to Seek, it continues +// iteration from the end of the last call. +func (fd *FileDescription) IterDirents(ctx context.Context, cb IterDirentsCallback) error { + return fd.impl.IterDirents(ctx, cb) +} + +// Seek changes fd's offset (assuming one exists) and returns its new value. +func (fd *FileDescription) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + return fd.impl.Seek(ctx, offset, whence) +} + +// Sync has the semantics of fsync(2). +func (fd *FileDescription) Sync(ctx context.Context) error { + return fd.impl.Sync(ctx) +} + +// ConfigureMMap mutates opts to implement mmap(2) for the file represented by +// fd. +func (fd *FileDescription) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { + return fd.impl.ConfigureMMap(ctx, opts) +} + +// Ioctl implements the ioctl(2) syscall. +func (fd *FileDescription) Ioctl(ctx context.Context, uio usermem.IO, args arch.SyscallArguments) (uintptr, error) { + return fd.impl.Ioctl(ctx, uio, args) +} + +// SyncFS instructs the filesystem containing fd to execute the semantics of +// syncfs(2). +func (fd *FileDescription) SyncFS(ctx context.Context) error { + return fd.vd.mount.fs.impl.Sync(ctx) +} diff --git a/pkg/sentry/vfs/file_description_impl_util_test.go b/pkg/sentry/vfs/file_description_impl_util_test.go index a5561dcbe..ac7799296 100644 --- a/pkg/sentry/vfs/file_description_impl_util_test.go +++ b/pkg/sentry/vfs/file_description_impl_util_test.go @@ -103,7 +103,7 @@ func TestGenCountFD(t *testing.T) { // The first read causes Generate to be called to fill the FD's buffer. buf := make([]byte, 2) ioseq := usermem.BytesIOSequence(buf) - n, err := fd.Impl().Read(ctx, ioseq, ReadOptions{}) + n, err := fd.Read(ctx, ioseq, ReadOptions{}) if n != 1 || (err != nil && err != io.EOF) { t.Fatalf("first Read: got (%d, %v), wanted (1, nil or EOF)", n, err) } @@ -112,17 +112,17 @@ func TestGenCountFD(t *testing.T) { } // A second read without seeking is still at EOF. - n, err = fd.Impl().Read(ctx, ioseq, ReadOptions{}) + n, err = fd.Read(ctx, ioseq, ReadOptions{}) if n != 0 || err != io.EOF { t.Fatalf("second Read: got (%d, %v), wanted (0, EOF)", n, err) } // Seeking to the beginning of the file causes it to be regenerated. - n, err = fd.Impl().Seek(ctx, 0, linux.SEEK_SET) + n, err = fd.Seek(ctx, 0, linux.SEEK_SET) if n != 0 || err != nil { t.Fatalf("Seek: got (%d, %v), wanted (0, nil)", n, err) } - n, err = fd.Impl().Read(ctx, ioseq, ReadOptions{}) + n, err = fd.Read(ctx, ioseq, ReadOptions{}) if n != 1 || (err != nil && err != io.EOF) { t.Fatalf("Read after Seek: got (%d, %v), wanted (1, nil or EOF)", n, err) } @@ -131,7 +131,7 @@ func TestGenCountFD(t *testing.T) { } // PRead at the beginning of the file also causes it to be regenerated. - n, err = fd.Impl().PRead(ctx, ioseq, 0, ReadOptions{}) + n, err = fd.PRead(ctx, ioseq, 0, ReadOptions{}) if n != 1 || (err != nil && err != io.EOF) { t.Fatalf("PRead: got (%d, %v), wanted (1, nil or EOF)", n, err) } diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index 76ff8cf51..dfbd2372a 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -47,6 +47,9 @@ func (fs *Filesystem) Init(vfsObj *VirtualFilesystem, impl FilesystemImpl) { fs.refs = 1 fs.vfs = vfsObj fs.impl = impl + vfsObj.filesystemsMu.Lock() + vfsObj.filesystems[fs] = struct{}{} + vfsObj.filesystemsMu.Unlock() } // VirtualFilesystem returns the containing VirtualFilesystem. @@ -66,9 +69,28 @@ func (fs *Filesystem) IncRef() { } } +// TryIncRef increments fs' reference count and returns true. If fs' reference +// count is zero, TryIncRef does nothing and returns false. +// +// TryIncRef does not require that a reference is held on fs. +func (fs *Filesystem) TryIncRef() bool { + for { + refs := atomic.LoadInt64(&fs.refs) + if refs <= 0 { + return false + } + if atomic.CompareAndSwapInt64(&fs.refs, refs, refs+1) { + return true + } + } +} + // DecRef decrements fs' reference count. func (fs *Filesystem) DecRef() { if refs := atomic.AddInt64(&fs.refs, -1); refs == 0 { + fs.vfs.filesystemsMu.Lock() + delete(fs.vfs.filesystems, fs) + fs.vfs.filesystemsMu.Unlock() fs.impl.Release() } else if refs < 0 { panic("Filesystem.decRef() called without holding a reference") diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index 1c3b2e987..ec23ab0dd 100644 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go @@ -18,6 +18,7 @@ import ( "math" "sync/atomic" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/syserror" @@ -133,13 +134,13 @@ func (vfs *VirtualFilesystem) NewMountNamespace(ctx context.Context, creds *auth return mntns, nil } -// NewMount creates and mounts a Filesystem configured by the given arguments. -func (vfs *VirtualFilesystem) NewMount(ctx context.Context, creds *auth.Credentials, source string, target *PathOperation, fsTypeName string, opts *GetFilesystemOptions) error { +// MountAt creates and mounts a Filesystem configured by the given arguments. +func (vfs *VirtualFilesystem) MountAt(ctx context.Context, creds *auth.Credentials, source string, target *PathOperation, fsTypeName string, opts *MountOptions) error { fsType := vfs.getFilesystemType(fsTypeName) if fsType == nil { return syserror.ENODEV } - fs, root, err := fsType.GetFilesystem(ctx, vfs, creds, source, *opts) + fs, root, err := fsType.GetFilesystem(ctx, vfs, creds, source, opts.GetFilesystemOptions) if err != nil { return err } @@ -207,6 +208,68 @@ func (vfs *VirtualFilesystem) NewMount(ctx context.Context, creds *auth.Credenti return nil } +// UmountAt removes the Mount at the given path. +func (vfs *VirtualFilesystem) UmountAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *UmountOptions) error { + if opts.Flags&^(linux.MNT_FORCE|linux.MNT_DETACH) != 0 { + return syserror.EINVAL + } + + // MNT_FORCE is currently unimplemented except for the permission check. + if opts.Flags&linux.MNT_FORCE != 0 && creds.HasCapabilityIn(linux.CAP_SYS_ADMIN, creds.UserNamespace.Root()) { + return syserror.EPERM + } + + vd, err := vfs.GetDentryAt(ctx, creds, pop, &GetDentryOptions{}) + if err != nil { + return err + } + defer vd.DecRef() + if vd.dentry != vd.mount.root { + return syserror.EINVAL + } + vfs.mountMu.Lock() + if mntns := MountNamespaceFromContext(ctx); mntns != nil && mntns != vd.mount.ns { + vfs.mountMu.Unlock() + return syserror.EINVAL + } + + // TODO(jamieliu): Linux special-cases umount of the caller's root, which + // we don't implement yet (we'll just fail it since the caller holds a + // reference on it). + + vfs.mounts.seq.BeginWrite() + if opts.Flags&linux.MNT_DETACH == 0 { + if len(vd.mount.children) != 0 { + vfs.mounts.seq.EndWrite() + vfs.mountMu.Unlock() + return syserror.EBUSY + } + // We are holding a reference on vd.mount. + expectedRefs := int64(1) + if !vd.mount.umounted { + expectedRefs = 2 + } + if atomic.LoadInt64(&vd.mount.refs)&^math.MinInt64 != expectedRefs { // mask out MSB + vfs.mounts.seq.EndWrite() + vfs.mountMu.Unlock() + return syserror.EBUSY + } + } + vdsToDecRef, mountsToDecRef := vfs.umountRecursiveLocked(vd.mount, &umountRecursiveOptions{ + eager: opts.Flags&linux.MNT_DETACH == 0, + disconnectHierarchy: true, + }, nil, nil) + vfs.mounts.seq.EndWrite() + vfs.mountMu.Unlock() + for _, vd := range vdsToDecRef { + vd.DecRef() + } + for _, mnt := range mountsToDecRef { + mnt.DecRef() + } + return nil +} + type umountRecursiveOptions struct { // If eager is true, ensure that future calls to Mount.tryIncMountedRef() // on umounted mounts fail. diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go index 3aa73d911..3ecbc8fc1 100644 --- a/pkg/sentry/vfs/options.go +++ b/pkg/sentry/vfs/options.go @@ -46,6 +46,12 @@ type MknodOptions struct { DevMinor uint32 } +// MountOptions contains options to VirtualFilesystem.MountAt(). +type MountOptions struct { + // GetFilesystemOptions contains options to FilesystemType.GetFilesystem(). + GetFilesystemOptions GetFilesystemOptions +} + // OpenOptions contains options to VirtualFilesystem.OpenAt() and // FilesystemImpl.OpenAt(). type OpenOptions struct { @@ -114,6 +120,12 @@ type StatOptions struct { Sync uint32 } +// UmountOptions contains options to VirtualFilesystem.UmountAt(). +type UmountOptions struct { + // Flags contains flags as specified for umount2(2). + Flags uint32 +} + // WriteOptions contains options to FileDescription.PWrite(), // FileDescriptionImpl.PWrite(), FileDescription.Write(), and // FileDescriptionImpl.Write(). diff --git a/pkg/sentry/vfs/syscalls.go b/pkg/sentry/vfs/syscalls.go deleted file mode 100644 index 436151afa..000000000 --- a/pkg/sentry/vfs/syscalls.go +++ /dev/null @@ -1,237 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package vfs - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/syserror" -) - -// PathOperation specifies the path operated on by a VFS method. -// -// PathOperation is passed to VFS methods by pointer to reduce memory copying: -// it's somewhat large and should never escape. (Options structs are passed by -// pointer to VFS and FileDescription methods for the same reason.) -type PathOperation struct { - // Root is the VFS root. References on Root are borrowed from the provider - // of the PathOperation. - // - // Invariants: Root.Ok(). - Root VirtualDentry - - // Start is the starting point for the path traversal. References on Start - // are borrowed from the provider of the PathOperation (i.e. the caller of - // the VFS method to which the PathOperation was passed). - // - // Invariants: Start.Ok(). If Pathname.Absolute, then Start == Root. - Start VirtualDentry - - // Path is the pathname traversed by this operation. - Pathname string - - // If FollowFinalSymlink is true, and the Dentry traversed by the final - // path component represents a symbolic link, the symbolic link should be - // followed. - FollowFinalSymlink bool -} - -// GetDentryAt returns a VirtualDentry representing the given path, at which a -// file must exist. A reference is taken on the returned VirtualDentry. -func (vfs *VirtualFilesystem) GetDentryAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *GetDentryOptions) (VirtualDentry, error) { - rp, err := vfs.getResolvingPath(creds, pop) - if err != nil { - return VirtualDentry{}, err - } - for { - d, err := rp.mount.fs.impl.GetDentryAt(ctx, rp, *opts) - if err == nil { - vd := VirtualDentry{ - mount: rp.mount, - dentry: d, - } - rp.mount.IncRef() - vfs.putResolvingPath(rp) - return vd, nil - } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) - return VirtualDentry{}, err - } - } -} - -// MkdirAt creates a directory at the given path. -func (vfs *VirtualFilesystem) MkdirAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *MkdirOptions) error { - // "Under Linux, apart from the permission bits, the S_ISVTX mode bit is - // also honored." - mkdir(2) - opts.Mode &= 01777 - rp, err := vfs.getResolvingPath(creds, pop) - if err != nil { - return err - } - for { - err := rp.mount.fs.impl.MkdirAt(ctx, rp, *opts) - if err == nil { - vfs.putResolvingPath(rp) - return nil - } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) - return err - } - } -} - -// MknodAt creates a file of the given mode at the given path. It returns an -// error from the syserror package. -func (vfs *VirtualFilesystem) MknodAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *MknodOptions) error { - rp, err := vfs.getResolvingPath(creds, pop) - if err != nil { - return nil - } - for { - if err = rp.mount.fs.impl.MknodAt(ctx, rp, *opts); err == nil { - vfs.putResolvingPath(rp) - return nil - } - // Handle mount traversals. - if !rp.handleError(err) { - vfs.putResolvingPath(rp) - return err - } - } -} - -// OpenAt returns a FileDescription providing access to the file at the given -// path. A reference is taken on the returned FileDescription. -func (vfs *VirtualFilesystem) OpenAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *OpenOptions) (*FileDescription, error) { - // Remove: - // - // - O_LARGEFILE, which we always report in FileDescription status flags - // since only 64-bit architectures are supported at this time. - // - // - O_CLOEXEC, which affects file descriptors and therefore must be - // handled outside of VFS. - // - // - Unknown flags. - opts.Flags &= linux.O_ACCMODE | linux.O_CREAT | linux.O_EXCL | linux.O_NOCTTY | linux.O_TRUNC | linux.O_APPEND | linux.O_NONBLOCK | linux.O_DSYNC | linux.O_ASYNC | linux.O_DIRECT | linux.O_DIRECTORY | linux.O_NOFOLLOW | linux.O_NOATIME | linux.O_SYNC | linux.O_PATH | linux.O_TMPFILE - // Linux's __O_SYNC (which we call linux.O_SYNC) implies O_DSYNC. - if opts.Flags&linux.O_SYNC != 0 { - opts.Flags |= linux.O_DSYNC - } - // Linux's __O_TMPFILE (which we call linux.O_TMPFILE) must be specified - // with O_DIRECTORY and a writable access mode (to ensure that it fails on - // filesystem implementations that do not support it). - if opts.Flags&linux.O_TMPFILE != 0 { - if opts.Flags&linux.O_DIRECTORY == 0 { - return nil, syserror.EINVAL - } - if opts.Flags&linux.O_CREAT != 0 { - return nil, syserror.EINVAL - } - if opts.Flags&linux.O_ACCMODE == linux.O_RDONLY { - return nil, syserror.EINVAL - } - } - // O_PATH causes most other flags to be ignored. - if opts.Flags&linux.O_PATH != 0 { - opts.Flags &= linux.O_DIRECTORY | linux.O_NOFOLLOW | linux.O_PATH - } - // "On Linux, the following bits are also honored in mode: [S_ISUID, - // S_ISGID, S_ISVTX]" - open(2) - opts.Mode &= 07777 - - if opts.Flags&linux.O_NOFOLLOW != 0 { - pop.FollowFinalSymlink = false - } - rp, err := vfs.getResolvingPath(creds, pop) - if err != nil { - return nil, err - } - if opts.Flags&linux.O_DIRECTORY != 0 { - rp.mustBeDir = true - rp.mustBeDirOrig = true - } - for { - fd, err := rp.mount.fs.impl.OpenAt(ctx, rp, *opts) - if err == nil { - vfs.putResolvingPath(rp) - return fd, nil - } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) - return nil, err - } - } -} - -// StatAt returns metadata for the file at the given path. -func (vfs *VirtualFilesystem) StatAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *StatOptions) (linux.Statx, error) { - rp, err := vfs.getResolvingPath(creds, pop) - if err != nil { - return linux.Statx{}, err - } - for { - stat, err := rp.mount.fs.impl.StatAt(ctx, rp, *opts) - if err == nil { - vfs.putResolvingPath(rp) - return stat, nil - } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) - return linux.Statx{}, err - } - } -} - -// StatusFlags returns file description status flags. -func (fd *FileDescription) StatusFlags(ctx context.Context) (uint32, error) { - flags, err := fd.impl.StatusFlags(ctx) - flags |= linux.O_LARGEFILE - return flags, err -} - -// SetStatusFlags sets file description status flags. -func (fd *FileDescription) SetStatusFlags(ctx context.Context, flags uint32) error { - return fd.impl.SetStatusFlags(ctx, flags) -} - -// TODO: -// -// - VFS.SyncAllFilesystems() for sync(2) -// -// - Something for syncfs(2) -// -// - VFS.LinkAt() -// -// - VFS.ReadlinkAt() -// -// - VFS.RenameAt() -// -// - VFS.RmdirAt() -// -// - VFS.SetStatAt() -// -// - VFS.StatFSAt() -// -// - VFS.SymlinkAt() -// -// - VFS.UmountAt() -// -// - VFS.UnlinkAt() -// -// - FileDescription.(almost everything) diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index f0cd3ffe5..7262b0d0a 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -20,6 +20,7 @@ // VirtualFilesystem.mountMu // Dentry.mu // Locks acquired by FilesystemImpls between Prepare{Delete,Rename}Dentry and Commit{Delete,Rename*}Dentry +// VirtualFilesystem.filesystemsMu // VirtualFilesystem.fsTypesMu // // Locking Dentry.mu in multiple Dentries requires holding @@ -28,6 +29,11 @@ package vfs import ( "sync" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/syserror" ) // A VirtualFilesystem (VFS for short) combines Filesystems in trees of Mounts. @@ -67,6 +73,11 @@ type VirtualFilesystem struct { // mountpoints is analogous to Linux's mountpoint_hashtable. mountpoints map[*Dentry]map[*Mount]struct{} + // filesystems contains all Filesystems. filesystems is protected by + // filesystemsMu. + filesystemsMu sync.Mutex + filesystems map[*Filesystem]struct{} + // fsTypes contains all FilesystemTypes that are usable in the // VirtualFilesystem. fsTypes is protected by fsTypesMu. fsTypesMu sync.RWMutex @@ -77,12 +88,379 @@ type VirtualFilesystem struct { func New() *VirtualFilesystem { vfs := &VirtualFilesystem{ mountpoints: make(map[*Dentry]map[*Mount]struct{}), + filesystems: make(map[*Filesystem]struct{}), fsTypes: make(map[string]FilesystemType), } vfs.mounts.Init() return vfs } +// PathOperation specifies the path operated on by a VFS method. +// +// PathOperation is passed to VFS methods by pointer to reduce memory copying: +// it's somewhat large and should never escape. (Options structs are passed by +// pointer to VFS and FileDescription methods for the same reason.) +type PathOperation struct { + // Root is the VFS root. References on Root are borrowed from the provider + // of the PathOperation. + // + // Invariants: Root.Ok(). + Root VirtualDentry + + // Start is the starting point for the path traversal. References on Start + // are borrowed from the provider of the PathOperation (i.e. the caller of + // the VFS method to which the PathOperation was passed). + // + // Invariants: Start.Ok(). If Pathname.Absolute, then Start == Root. + Start VirtualDentry + + // Path is the pathname traversed by this operation. + Pathname string + + // If FollowFinalSymlink is true, and the Dentry traversed by the final + // path component represents a symbolic link, the symbolic link should be + // followed. + FollowFinalSymlink bool +} + +// GetDentryAt returns a VirtualDentry representing the given path, at which a +// file must exist. A reference is taken on the returned VirtualDentry. +func (vfs *VirtualFilesystem) GetDentryAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *GetDentryOptions) (VirtualDentry, error) { + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return VirtualDentry{}, err + } + for { + d, err := rp.mount.fs.impl.GetDentryAt(ctx, rp, *opts) + if err == nil { + vd := VirtualDentry{ + mount: rp.mount, + dentry: d, + } + rp.mount.IncRef() + vfs.putResolvingPath(rp) + return vd, nil + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return VirtualDentry{}, err + } + } +} + +// LinkAt creates a hard link at newpop representing the existing file at +// oldpop. +func (vfs *VirtualFilesystem) LinkAt(ctx context.Context, creds *auth.Credentials, oldpop, newpop *PathOperation) error { + oldVD, err := vfs.GetDentryAt(ctx, creds, oldpop, &GetDentryOptions{}) + if err != nil { + return err + } + rp, err := vfs.getResolvingPath(creds, newpop) + if err != nil { + oldVD.DecRef() + return err + } + for { + err := rp.mount.fs.impl.LinkAt(ctx, rp, oldVD) + if err == nil { + oldVD.DecRef() + vfs.putResolvingPath(rp) + return nil + } + if !rp.handleError(err) { + oldVD.DecRef() + vfs.putResolvingPath(rp) + return err + } + } +} + +// MkdirAt creates a directory at the given path. +func (vfs *VirtualFilesystem) MkdirAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *MkdirOptions) error { + // "Under Linux, apart from the permission bits, the S_ISVTX mode bit is + // also honored." - mkdir(2) + opts.Mode &= 0777 | linux.S_ISVTX + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return err + } + for { + err := rp.mount.fs.impl.MkdirAt(ctx, rp, *opts) + if err == nil { + vfs.putResolvingPath(rp) + return nil + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return err + } + } +} + +// MknodAt creates a file of the given mode at the given path. It returns an +// error from the syserror package. +func (vfs *VirtualFilesystem) MknodAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *MknodOptions) error { + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return nil + } + for { + if err = rp.mount.fs.impl.MknodAt(ctx, rp, *opts); err == nil { + vfs.putResolvingPath(rp) + return nil + } + // Handle mount traversals. + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return err + } + } +} + +// OpenAt returns a FileDescription providing access to the file at the given +// path. A reference is taken on the returned FileDescription. +func (vfs *VirtualFilesystem) OpenAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *OpenOptions) (*FileDescription, error) { + // Remove: + // + // - O_LARGEFILE, which we always report in FileDescription status flags + // since only 64-bit architectures are supported at this time. + // + // - O_CLOEXEC, which affects file descriptors and therefore must be + // handled outside of VFS. + // + // - Unknown flags. + opts.Flags &= linux.O_ACCMODE | linux.O_CREAT | linux.O_EXCL | linux.O_NOCTTY | linux.O_TRUNC | linux.O_APPEND | linux.O_NONBLOCK | linux.O_DSYNC | linux.O_ASYNC | linux.O_DIRECT | linux.O_DIRECTORY | linux.O_NOFOLLOW | linux.O_NOATIME | linux.O_SYNC | linux.O_PATH | linux.O_TMPFILE + // Linux's __O_SYNC (which we call linux.O_SYNC) implies O_DSYNC. + if opts.Flags&linux.O_SYNC != 0 { + opts.Flags |= linux.O_DSYNC + } + // Linux's __O_TMPFILE (which we call linux.O_TMPFILE) must be specified + // with O_DIRECTORY and a writable access mode (to ensure that it fails on + // filesystem implementations that do not support it). + if opts.Flags&linux.O_TMPFILE != 0 { + if opts.Flags&linux.O_DIRECTORY == 0 { + return nil, syserror.EINVAL + } + if opts.Flags&linux.O_CREAT != 0 { + return nil, syserror.EINVAL + } + if opts.Flags&linux.O_ACCMODE == linux.O_RDONLY { + return nil, syserror.EINVAL + } + } + // O_PATH causes most other flags to be ignored. + if opts.Flags&linux.O_PATH != 0 { + opts.Flags &= linux.O_DIRECTORY | linux.O_NOFOLLOW | linux.O_PATH + } + // "On Linux, the following bits are also honored in mode: [S_ISUID, + // S_ISGID, S_ISVTX]" - open(2) + opts.Mode &= 0777 | linux.S_ISUID | linux.S_ISGID | linux.S_ISVTX + + if opts.Flags&linux.O_NOFOLLOW != 0 { + pop.FollowFinalSymlink = false + } + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return nil, err + } + if opts.Flags&linux.O_DIRECTORY != 0 { + rp.mustBeDir = true + rp.mustBeDirOrig = true + } + for { + fd, err := rp.mount.fs.impl.OpenAt(ctx, rp, *opts) + if err == nil { + vfs.putResolvingPath(rp) + return fd, nil + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return nil, err + } + } +} + +// ReadlinkAt returns the target of the symbolic link at the given path. +func (vfs *VirtualFilesystem) ReadlinkAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation) (string, error) { + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return "", err + } + for { + target, err := rp.mount.fs.impl.ReadlinkAt(ctx, rp) + if err == nil { + vfs.putResolvingPath(rp) + return target, nil + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return "", err + } + } +} + +// RenameAt renames the file at oldpop to newpop. +func (vfs *VirtualFilesystem) RenameAt(ctx context.Context, creds *auth.Credentials, oldpop, newpop *PathOperation, opts *RenameOptions) error { + oldVD, err := vfs.GetDentryAt(ctx, creds, oldpop, &GetDentryOptions{}) + if err != nil { + return err + } + rp, err := vfs.getResolvingPath(creds, newpop) + if err != nil { + oldVD.DecRef() + return err + } + for { + err := rp.mount.fs.impl.RenameAt(ctx, rp, oldVD, *opts) + if err == nil { + oldVD.DecRef() + vfs.putResolvingPath(rp) + return nil + } + if !rp.handleError(err) { + oldVD.DecRef() + vfs.putResolvingPath(rp) + return err + } + } +} + +// RmdirAt removes the directory at the given path. +func (vfs *VirtualFilesystem) RmdirAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation) error { + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return err + } + for { + err := rp.mount.fs.impl.RmdirAt(ctx, rp) + if err == nil { + vfs.putResolvingPath(rp) + return nil + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return err + } + } +} + +// SetStatAt changes metadata for the file at the given path. +func (vfs *VirtualFilesystem) SetStatAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *SetStatOptions) error { + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return err + } + for { + err := rp.mount.fs.impl.SetStatAt(ctx, rp, *opts) + if err == nil { + vfs.putResolvingPath(rp) + return nil + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return err + } + } +} + +// StatAt returns metadata for the file at the given path. +func (vfs *VirtualFilesystem) StatAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *StatOptions) (linux.Statx, error) { + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return linux.Statx{}, err + } + for { + stat, err := rp.mount.fs.impl.StatAt(ctx, rp, *opts) + if err == nil { + vfs.putResolvingPath(rp) + return stat, nil + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return linux.Statx{}, err + } + } +} + +// StatFSAt returns metadata for the filesystem containing the file at the +// given path. +func (vfs *VirtualFilesystem) StatFSAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation) (linux.Statfs, error) { + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return linux.Statfs{}, err + } + for { + statfs, err := rp.mount.fs.impl.StatFSAt(ctx, rp) + if err == nil { + vfs.putResolvingPath(rp) + return statfs, nil + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return linux.Statfs{}, err + } + } +} + +// SymlinkAt creates a symbolic link at the given path with the given target. +func (vfs *VirtualFilesystem) SymlinkAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, target string) error { + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return err + } + for { + err := rp.mount.fs.impl.SymlinkAt(ctx, rp, target) + if err == nil { + vfs.putResolvingPath(rp) + return nil + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return err + } + } +} + +// UnlinkAt deletes the non-directory file at the given path. +func (vfs *VirtualFilesystem) UnlinkAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation) error { + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return err + } + for { + err := rp.mount.fs.impl.UnlinkAt(ctx, rp) + if err == nil { + vfs.putResolvingPath(rp) + return nil + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return err + } + } +} + +// SyncAllFilesystems has the semantics of Linux's sync(2). +func (vfs *VirtualFilesystem) SyncAllFilesystems(ctx context.Context) error { + fss := make(map[*Filesystem]struct{}) + vfs.filesystemsMu.Lock() + for fs := range vfs.filesystems { + if !fs.TryIncRef() { + continue + } + fss[fs] = struct{}{} + } + vfs.filesystemsMu.Unlock() + var retErr error + for fs := range fss { + if err := fs.impl.Sync(ctx); err != nil && retErr == nil { + retErr = err + } + fs.DecRef() + } + return retErr +} + // A VirtualDentry represents a node in a VFS tree, by combining a Dentry // (which represents a node in a Filesystem's tree) and a Mount (which // represents the Filesystem's position in a VFS mount tree). -- cgit v1.2.3 From 2e3b9b0a68d8f191d061008feda6e4a4ce202a78 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 10 Dec 2019 19:14:05 -0800 Subject: Deduplicate and simplify control message processing for recvmsg and sendmsg. Also, improve performance by calculating how much space is needed before making an allocation for sendmsg in hostinet. PiperOrigin-RevId: 284898581 --- pkg/sentry/socket/control/control.go | 51 ++++++++++++++++++++++----------- pkg/sentry/socket/hostinet/socket.go | 18 +++++++++--- pkg/sentry/syscalls/linux/sys_socket.go | 18 +----------- 3 files changed, 50 insertions(+), 37 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/control/control.go b/pkg/sentry/socket/control/control.go index fa3188d51..af1a4e95f 100644 --- a/pkg/sentry/socket/control/control.go +++ b/pkg/sentry/socket/control/control.go @@ -348,43 +348,62 @@ func PackTClass(t *kernel.Task, tClass int32, buf []byte) []byte { ) } -func addSpaceForCmsg(cmsgDataLen int, buf []byte) []byte { - newBuf := make([]byte, 0, len(buf)+linux.SizeOfControlMessageHeader+cmsgDataLen) - return append(newBuf, buf...) -} - -// PackControlMessages converts the given ControlMessages struct into a buffer. +// PackControlMessages packs control messages into the given buffer. +// // We skip control messages specific to Unix domain sockets. -func PackControlMessages(t *kernel.Task, cmsgs socket.ControlMessages) []byte { - var buf []byte - // The use of t.Arch().Width() is analogous to Linux's use of sizeof(long) in - // CMSG_ALIGN. - width := t.Arch().Width() - +// +// Note that some control messages may be truncated if they do not fit under +// the capacity of buf. +func PackControlMessages(t *kernel.Task, cmsgs socket.ControlMessages, buf []byte) []byte { if cmsgs.IP.HasTimestamp { - buf = addSpaceForCmsg(int(width), buf) buf = PackTimestamp(t, cmsgs.IP.Timestamp, buf) } if cmsgs.IP.HasInq { // In Linux, TCP_CM_INQ is added after SO_TIMESTAMP. - buf = addSpaceForCmsg(AlignUp(linux.SizeOfControlMessageInq, width), buf) buf = PackInq(t, cmsgs.IP.Inq, buf) } if cmsgs.IP.HasTOS { - buf = addSpaceForCmsg(AlignUp(linux.SizeOfControlMessageTOS, width), buf) buf = PackTOS(t, cmsgs.IP.TOS, buf) } if cmsgs.IP.HasTClass { - buf = addSpaceForCmsg(AlignUp(linux.SizeOfControlMessageTClass, width), buf) buf = PackTClass(t, cmsgs.IP.TClass, buf) } return buf } +// cmsgSpace is equivalent to CMSG_SPACE in Linux. +func cmsgSpace(t *kernel.Task, dataLen int) int { + return linux.SizeOfControlMessageHeader + AlignUp(dataLen, t.Arch().Width()) +} + +// CmsgsSpace returns the number of bytes needed to fit the control messages +// represented in cmsgs. +func CmsgsSpace(t *kernel.Task, cmsgs socket.ControlMessages) int { + space := 0 + + if cmsgs.IP.HasTimestamp { + space += cmsgSpace(t, linux.SizeOfTimeval) + } + + if cmsgs.IP.HasInq { + space += cmsgSpace(t, linux.SizeOfControlMessageInq) + } + + if cmsgs.IP.HasTOS { + space += cmsgSpace(t, linux.SizeOfControlMessageTOS) + } + + if cmsgs.IP.HasTClass { + space += cmsgSpace(t, linux.SizeOfControlMessageTClass) + } + + return space +} + // Parse parses a raw socket control message into portable objects. func Parse(t *kernel.Task, socketOrEndpoint interface{}, buf []byte) (socket.ControlMessages, error) { var ( diff --git a/pkg/sentry/socket/hostinet/socket.go b/pkg/sentry/socket/hostinet/socket.go index a8c152b54..c957b0f1d 100644 --- a/pkg/sentry/socket/hostinet/socket.go +++ b/pkg/sentry/socket/hostinet/socket.go @@ -45,7 +45,7 @@ const ( sizeofSockaddr = syscall.SizeofSockaddrInet6 // sizeof(sockaddr_in6) > sizeof(sockaddr_in) // maxControlLen is the maximum size of a control message buffer used in a - // recvmsg syscall. + // recvmsg or sendmsg syscall. maxControlLen = 1024 ) @@ -412,9 +412,12 @@ func (s *socketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags msg.Namelen = uint32(len(senderAddrBuf)) } if controlLen > 0 { - controlBuf = make([]byte, maxControlLen) + if controlLen > maxControlLen { + controlLen = maxControlLen + } + controlBuf = make([]byte, controlLen) msg.Control = &controlBuf[0] - msg.Controllen = maxControlLen + msg.Controllen = controlLen } n, err := recvmsg(s.fd, &msg, sysflags) if err != nil { @@ -489,7 +492,14 @@ func (s *socketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to [] return 0, syserr.ErrInvalidArgument } - controlBuf := control.PackControlMessages(t, controlMessages) + space := uint64(control.CmsgsSpace(t, controlMessages)) + if space > maxControlLen { + space = maxControlLen + } + controlBuf := make([]byte, 0, space) + // PackControlMessages will append up to space bytes to controlBuf. + controlBuf = control.PackControlMessages(t, controlMessages, controlBuf) + sendmsgFromBlocks := safemem.WriterFunc(func(srcs safemem.BlockSeq) (uint64, error) { // Refuse to do anything if any part of src.Addrs was unusable. if uint64(src.NumBytes()) != srcs.NumBytes() { diff --git a/pkg/sentry/syscalls/linux/sys_socket.go b/pkg/sentry/syscalls/linux/sys_socket.go index e3faa890b..4b5aafcc0 100644 --- a/pkg/sentry/syscalls/linux/sys_socket.go +++ b/pkg/sentry/syscalls/linux/sys_socket.go @@ -787,29 +787,13 @@ func recvSingleMsg(t *kernel.Task, s socket.Socket, msgPtr usermem.Addr, flags i defer cms.Release() controlData := make([]byte, 0, msg.ControlLen) + controlData = control.PackControlMessages(t, cms, controlData) if cr, ok := s.(transport.Credentialer); ok && cr.Passcred() { creds, _ := cms.Unix.Credentials.(control.SCMCredentials) controlData, mflags = control.PackCredentials(t, creds, controlData, mflags) } - if cms.IP.HasTimestamp { - controlData = control.PackTimestamp(t, cms.IP.Timestamp, controlData) - } - - if cms.IP.HasInq { - // In Linux, TCP_CM_INQ is added after SO_TIMESTAMP. - controlData = control.PackInq(t, cms.IP.Inq, controlData) - } - - if cms.IP.HasTOS { - controlData = control.PackTOS(t, cms.IP.TOS, controlData) - } - - if cms.IP.HasTClass { - controlData = control.PackTClass(t, cms.IP.TClass, controlData) - } - if cms.Unix.Rights != nil { controlData, mflags = control.PackRights(t, cms.Unix.Rights.(control.SCMRights), flags&linux.MSG_CMSG_CLOEXEC != 0, controlData, mflags) } -- cgit v1.2.3 From f8c5ad061bb529e0314bde17f4f1b4ddc82c0120 Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Wed, 11 Dec 2019 10:30:43 -0800 Subject: runsc/debug: add an option to list all processes runsc debug --ps list all processes with all threads. This option is added to the debug command but not to the ps command, because it is going to be used for debug purposes and we want to add any useful information without thinking about backward compatibility. This will help to investigate syzkaller issues. PiperOrigin-RevId: 285013668 --- pkg/sentry/control/proc.go | 34 +++++---- runsc/cmd/debug.go | 13 ++++ runsc/container/console_test.go | 11 +-- runsc/container/container_test.go | 120 +++++++++++++++++++------------- runsc/container/multi_container_test.go | 45 ++++++------ 5 files changed, 132 insertions(+), 91 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/control/proc.go b/pkg/sentry/control/proc.go index a6f90b2bb..ced51c66c 100644 --- a/pkg/sentry/control/proc.go +++ b/pkg/sentry/control/proc.go @@ -272,7 +272,8 @@ type Process struct { UID auth.KUID `json:"uid"` PID kernel.ThreadID `json:"pid"` // Parent PID - PPID kernel.ThreadID `json:"ppid"` + PPID kernel.ThreadID `json:"ppid"` + Threads []kernel.ThreadID `json:"threads"` // Processor utilization C int32 `json:"c"` // TTY name of the process. Will be of the form "pts/N" if there is a @@ -310,7 +311,7 @@ func ProcessListToTable(pl []*Process) string { // ProcessListToJSON will return the JSON representation of ps. func ProcessListToJSON(pl []*Process) (string, error) { - b, err := json.Marshal(pl) + b, err := json.MarshalIndent(pl, "", " ") if err != nil { return "", fmt.Errorf("couldn't marshal process list %v: %v", pl, err) } @@ -337,7 +338,9 @@ func Processes(k *kernel.Kernel, containerID string, out *[]*Process) error { ts := k.TaskSet() now := k.RealtimeClock().Now() for _, tg := range ts.Root.ThreadGroups() { - pid := tg.PIDNamespace().IDOfThreadGroup(tg) + pidns := tg.PIDNamespace() + pid := pidns.IDOfThreadGroup(tg) + // If tg has already been reaped ignore it. if pid == 0 { continue @@ -348,19 +351,20 @@ func Processes(k *kernel.Kernel, containerID string, out *[]*Process) error { ppid := kernel.ThreadID(0) if p := tg.Leader().Parent(); p != nil { - ppid = p.PIDNamespace().IDOfThreadGroup(p.ThreadGroup()) - } - p := Process{ - UID: tg.Leader().Credentials().EffectiveKUID, - PID: pid, - PPID: ppid, - STime: formatStartTime(now, tg.Leader().StartTime()), - C: percentCPU(tg.CPUStats(), tg.Leader().StartTime(), now), - Time: tg.CPUStats().SysTime.String(), - Cmd: tg.Leader().Name(), - TTY: ttyName(tg.TTY()), + ppid = pidns.IDOfThreadGroup(p.ThreadGroup()) } - *out = append(*out, &p) + threads := tg.MemberIDs(pidns) + *out = append(*out, &Process{ + UID: tg.Leader().Credentials().EffectiveKUID, + PID: pid, + PPID: ppid, + Threads: threads, + STime: formatStartTime(now, tg.Leader().StartTime()), + C: percentCPU(tg.CPUStats(), tg.Leader().StartTime(), now), + Time: tg.CPUStats().SysTime.String(), + Cmd: tg.Leader().Name(), + TTY: ttyName(tg.TTY()), + }) } sort.Slice(*out, func(i, j int) bool { return (*out)[i].PID < (*out)[j].PID }) return nil diff --git a/runsc/cmd/debug.go b/runsc/cmd/debug.go index 38da7ee02..f37415810 100644 --- a/runsc/cmd/debug.go +++ b/runsc/cmd/debug.go @@ -42,6 +42,7 @@ type Debug struct { logLevel string logPackets string duration time.Duration + ps bool } // Name implements subcommands.Command. @@ -71,6 +72,7 @@ func (d *Debug) SetFlags(f *flag.FlagSet) { f.StringVar(&d.strace, "strace", "", `A comma separated list of syscalls to trace. "all" enables all traces, "off" disables all`) f.StringVar(&d.logLevel, "log-level", "", "The log level to set: warning (0), info (1), or debug (2).") f.StringVar(&d.logPackets, "log-packets", "", "A boolean value to enable or disable packet logging: true or false.") + f.BoolVar(&d.ps, "ps", false, "lists processes") } // Execute implements subcommands.Command.Execute. @@ -240,6 +242,17 @@ func (d *Debug) Execute(_ context.Context, f *flag.FlagSet, args ...interface{}) } log.Infof("Logging options changed") } + if d.ps { + pList, err := c.Processes() + if err != nil { + Fatalf("getting processes for container: %v", err) + } + o, err := control.ProcessListToJSON(pList) + if err != nil { + Fatalf("generating JSON: %v", err) + } + log.Infof(o) + } if delay { time.Sleep(d.duration) diff --git a/runsc/container/console_test.go b/runsc/container/console_test.go index 7d67c3a75..5ed131a7f 100644 --- a/runsc/container/console_test.go +++ b/runsc/container/console_test.go @@ -28,6 +28,7 @@ import ( "github.com/kr/pty" "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/sentry/control" + "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/unet" "gvisor.dev/gvisor/pkg/urpc" "gvisor.dev/gvisor/runsc/testutil" @@ -219,9 +220,9 @@ func TestJobControlSignalExec(t *testing.T) { // Make sure all the processes are running. expectedPL := []*control.Process{ // Root container process. - {PID: 1, Cmd: "sleep"}, + {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, // Bash from exec process. - {PID: 2, Cmd: "bash"}, + {PID: 2, Cmd: "bash", Threads: []kernel.ThreadID{2}}, } if err := waitForProcessList(c, expectedPL); err != nil { t.Error(err) @@ -231,7 +232,7 @@ func TestJobControlSignalExec(t *testing.T) { ptyMaster.Write([]byte("sleep 100\n")) // Wait for it to start. Sleep's PPID is bash's PID. - expectedPL = append(expectedPL, &control.Process{PID: 3, PPID: 2, Cmd: "sleep"}) + expectedPL = append(expectedPL, &control.Process{PID: 3, PPID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{3}}) if err := waitForProcessList(c, expectedPL); err != nil { t.Error(err) } @@ -361,7 +362,7 @@ func TestJobControlSignalRootContainer(t *testing.T) { // Wait for bash to start. expectedPL := []*control.Process{ - {PID: 1, Cmd: "bash"}, + {PID: 1, Cmd: "bash", Threads: []kernel.ThreadID{1}}, } if err := waitForProcessList(c, expectedPL); err != nil { t.Fatal(err) @@ -371,7 +372,7 @@ func TestJobControlSignalRootContainer(t *testing.T) { ptyMaster.Write([]byte("sleep 100\n")) // Wait for sleep to start. - expectedPL = append(expectedPL, &control.Process{PID: 2, PPID: 1, Cmd: "sleep"}) + expectedPL = append(expectedPL, &control.Process{PID: 2, PPID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{2}}) if err := waitForProcessList(c, expectedPL); err != nil { t.Fatal(err) } diff --git a/runsc/container/container_test.go b/runsc/container/container_test.go index 1d06f2780..2ced028f6 100644 --- a/runsc/container/container_test.go +++ b/runsc/container/container_test.go @@ -37,6 +37,7 @@ import ( "gvisor.dev/gvisor/pkg/bits" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/control" + "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/runsc/boot" "gvisor.dev/gvisor/runsc/boot/platforms" @@ -52,13 +53,14 @@ func waitForProcessList(cont *Container, want []*control.Process) error { err = fmt.Errorf("error getting process data from container: %v", err) return &backoff.PermanentError{Err: err} } - if !procListsEqual(got, want) { - return fmt.Errorf("container got process list: %s, want: %s", procListToString(got), procListToString(want)) + if r, err := procListsEqual(got, want); !r { + return fmt.Errorf("container got process list: %s, want: %s: error: %v", + procListToString(got), procListToString(want), err) } return nil } // Gives plenty of time as tests can run slow under --race. - return testutil.Poll(cb, 30*time.Second) + return testutil.Poll(cb, 10*time.Second) } func waitForProcessCount(cont *Container, want int) error { @@ -91,9 +93,9 @@ func blockUntilWaitable(pid int) error { // procListsEqual is used to check whether 2 Process lists are equal for all // implemented fields. -func procListsEqual(got, want []*control.Process) bool { +func procListsEqual(got, want []*control.Process) (bool, error) { if len(got) != len(want) { - return false + return false, nil } for i := range got { pd1 := got[i] @@ -106,11 +108,19 @@ func procListsEqual(got, want []*control.Process) bool { // where we use this method. Tests that care about the TTY // field should check for it themselves. pd1.TTY = "" - if *pd1 != *pd2 { - return false + pd1Json, err := control.ProcessListToJSON([]*control.Process{pd1}) + if err != nil { + return false, err + } + pd2Json, err := control.ProcessListToJSON([]*control.Process{pd2}) + if err != nil { + return false, err + } + if pd1Json != pd2Json { + return false, nil } } - return true + return true, nil } // getAndCheckProcLists is similar to waitForProcessList, but does not wait and retry the @@ -120,7 +130,11 @@ func getAndCheckProcLists(cont *Container, want []*control.Process) error { if err != nil { return fmt.Errorf("error getting process data from container: %v", err) } - if procListsEqual(got, want) { + equal, err := procListsEqual(got, want) + if err != nil { + return err + } + if equal { return nil } return fmt.Errorf("container got process list: %s, want: %s", procListToString(got), procListToString(want)) @@ -292,11 +306,12 @@ func TestLifecycle(t *testing.T) { // expectedPL lists the expected process state of the container. expectedPL := []*control.Process{ { - UID: 0, - PID: 1, - PPID: 0, - C: 0, - Cmd: "sleep", + UID: 0, + PID: 1, + PPID: 0, + C: 0, + Cmd: "sleep", + Threads: []kernel.ThreadID{1}, }, } // Create the container. @@ -594,18 +609,20 @@ func TestExec(t *testing.T) { // expectedPL lists the expected process state of the container. expectedPL := []*control.Process{ { - UID: 0, - PID: 1, - PPID: 0, - C: 0, - Cmd: "sleep", + UID: 0, + PID: 1, + PPID: 0, + C: 0, + Cmd: "sleep", + Threads: []kernel.ThreadID{1}, }, { - UID: uid, - PID: 2, - PPID: 0, - C: 0, - Cmd: "sleep", + UID: uid, + PID: 2, + PPID: 0, + C: 0, + Cmd: "sleep", + Threads: []kernel.ThreadID{2}, }, } @@ -1066,18 +1083,20 @@ func TestPauseResume(t *testing.T) { // expectedPL lists the expected process state of the container. expectedPL := []*control.Process{ { - UID: 0, - PID: 1, - PPID: 0, - C: 0, - Cmd: "sleep", + UID: 0, + PID: 1, + PPID: 0, + C: 0, + Cmd: "sleep", + Threads: []kernel.ThreadID{1}, }, { - UID: uid, - PID: 2, - PPID: 0, - C: 0, - Cmd: "bash", + UID: uid, + PID: 2, + PPID: 0, + C: 0, + Cmd: "bash", + Threads: []kernel.ThreadID{2}, }, } @@ -1130,11 +1149,12 @@ func TestPauseResume(t *testing.T) { expectedPL2 := []*control.Process{ { - UID: 0, - PID: 1, - PPID: 0, - C: 0, - Cmd: "sleep", + UID: 0, + PID: 1, + PPID: 0, + C: 0, + Cmd: "sleep", + Threads: []kernel.ThreadID{1}, }, } @@ -1245,18 +1265,20 @@ func TestCapabilities(t *testing.T) { // expectedPL lists the expected process state of the container. expectedPL := []*control.Process{ { - UID: 0, - PID: 1, - PPID: 0, - C: 0, - Cmd: "sleep", + UID: 0, + PID: 1, + PPID: 0, + C: 0, + Cmd: "sleep", + Threads: []kernel.ThreadID{1}, }, { - UID: uid, - PID: 2, - PPID: 0, - C: 0, - Cmd: "exe", + UID: uid, + PID: 2, + PPID: 0, + C: 0, + Cmd: "exe", + Threads: []kernel.ThreadID{2}, }, } if err := waitForProcessList(cont, expectedPL[:1]); err != nil { diff --git a/runsc/container/multi_container_test.go b/runsc/container/multi_container_test.go index de2fd3cf2..4ad09ceab 100644 --- a/runsc/container/multi_container_test.go +++ b/runsc/container/multi_container_test.go @@ -156,13 +156,13 @@ func TestMultiContainerSanity(t *testing.T) { // Check via ps that multiple processes are running. expectedPL := []*control.Process{ - {PID: 1, Cmd: "sleep"}, + {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, } if err := waitForProcessList(containers[0], expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) } expectedPL = []*control.Process{ - {PID: 2, Cmd: "sleep"}, + {PID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{2}}, } if err := waitForProcessList(containers[1], expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) @@ -202,13 +202,13 @@ func TestMultiPIDNS(t *testing.T) { // Check via ps that multiple processes are running. expectedPL := []*control.Process{ - {PID: 1, Cmd: "sleep"}, + {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, } if err := waitForProcessList(containers[0], expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) } expectedPL = []*control.Process{ - {PID: 1, Cmd: "sleep"}, + {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, } if err := waitForProcessList(containers[1], expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) @@ -264,7 +264,7 @@ func TestMultiPIDNSPath(t *testing.T) { // Check via ps that multiple processes are running. expectedPL := []*control.Process{ - {PID: 1, Cmd: "sleep"}, + {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, } if err := waitForProcessList(containers[0], expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) @@ -274,7 +274,7 @@ func TestMultiPIDNSPath(t *testing.T) { } expectedPL = []*control.Process{ - {PID: 2, Cmd: "sleep"}, + {PID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{2}}, } if err := waitForProcessList(containers[1], expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) @@ -306,7 +306,7 @@ func TestMultiContainerWait(t *testing.T) { // Check via ps that multiple processes are running. expectedPL := []*control.Process{ - {PID: 2, Cmd: "sleep"}, + {PID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{2}}, } if err := waitForProcessList(containers[1], expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) @@ -351,7 +351,7 @@ func TestMultiContainerWait(t *testing.T) { // After Wait returns, ensure that the root container is running and // the child has finished. expectedPL = []*control.Process{ - {PID: 1, Cmd: "sleep"}, + {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, } if err := waitForProcessList(containers[0], expectedPL); err != nil { t.Errorf("failed to wait for %q to start: %v", strings.Join(containers[0].Spec.Process.Args, " "), err) @@ -383,7 +383,7 @@ func TestExecWait(t *testing.T) { // Check via ps that process is running. expectedPL := []*control.Process{ - {PID: 2, Cmd: "sleep"}, + {PID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{2}}, } if err := waitForProcessList(containers[1], expectedPL); err != nil { t.Fatalf("failed to wait for sleep to start: %v", err) @@ -418,7 +418,7 @@ func TestExecWait(t *testing.T) { // Wait for the exec'd process to exit. expectedPL = []*control.Process{ - {PID: 1, Cmd: "sleep"}, + {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, } if err := waitForProcessList(containers[0], expectedPL); err != nil { t.Fatalf("failed to wait for second container to stop: %v", err) @@ -505,7 +505,7 @@ func TestMultiContainerSignal(t *testing.T) { // Check via ps that container 1 process is running. expectedPL := []*control.Process{ - {PID: 2, Cmd: "sleep"}, + {PID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{2}}, } if err := waitForProcessList(containers[1], expectedPL); err != nil { @@ -519,7 +519,7 @@ func TestMultiContainerSignal(t *testing.T) { // Make sure process 1 is still running. expectedPL = []*control.Process{ - {PID: 1, Cmd: "sleep"}, + {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, } if err := waitForProcessList(containers[0], expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) @@ -633,9 +633,10 @@ func TestMultiContainerDestroy(t *testing.T) { if err != nil { t.Fatalf("error getting process data from sandbox: %v", err) } - expectedPL := []*control.Process{{PID: 1, Cmd: "sleep"}} - if !procListsEqual(pss, expectedPL) { - t.Errorf("container got process list: %s, want: %s", procListToString(pss), procListToString(expectedPL)) + expectedPL := []*control.Process{{PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}} + if r, err := procListsEqual(pss, expectedPL); !r { + t.Errorf("container got process list: %s, want: %s: error: %v", + procListToString(pss), procListToString(expectedPL), err) } // Check that cont.Destroy is safe to call multiple times. @@ -669,7 +670,7 @@ func TestMultiContainerProcesses(t *testing.T) { // Check root's container process list doesn't include other containers. expectedPL0 := []*control.Process{ - {PID: 1, Cmd: "sleep"}, + {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, } if err := waitForProcessList(containers[0], expectedPL0); err != nil { t.Errorf("failed to wait for process to start: %v", err) @@ -677,8 +678,8 @@ func TestMultiContainerProcesses(t *testing.T) { // Same for the other container. expectedPL1 := []*control.Process{ - {PID: 2, Cmd: "sh"}, - {PID: 3, PPID: 2, Cmd: "sleep"}, + {PID: 2, Cmd: "sh", Threads: []kernel.ThreadID{2}}, + {PID: 3, PPID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{3}}, } if err := waitForProcessList(containers[1], expectedPL1); err != nil { t.Errorf("failed to wait for process to start: %v", err) @@ -692,7 +693,7 @@ func TestMultiContainerProcesses(t *testing.T) { if _, err := containers[1].Execute(args); err != nil { t.Fatalf("error exec'ing: %v", err) } - expectedPL1 = append(expectedPL1, &control.Process{PID: 4, Cmd: "sleep"}) + expectedPL1 = append(expectedPL1, &control.Process{PID: 4, Cmd: "sleep", Threads: []kernel.ThreadID{4}}) if err := waitForProcessList(containers[1], expectedPL1); err != nil { t.Errorf("failed to wait for process to start: %v", err) } @@ -1513,7 +1514,7 @@ func TestMultiContainerGoferKilled(t *testing.T) { // Ensure container is running c := containers[2] expectedPL := []*control.Process{ - {PID: 3, Cmd: "sleep"}, + {PID: 3, Cmd: "sleep", Threads: []kernel.ThreadID{3}}, } if err := waitForProcessList(c, expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) @@ -1541,7 +1542,7 @@ func TestMultiContainerGoferKilled(t *testing.T) { continue // container[2] has been killed. } pl := []*control.Process{ - {PID: kernel.ThreadID(i + 1), Cmd: "sleep"}, + {PID: kernel.ThreadID(i + 1), Cmd: "sleep", Threads: []kernel.ThreadID{kernel.ThreadID(i + 1)}}, } if err := waitForProcessList(c, pl); err != nil { t.Errorf("Container %q was affected by another container: %v", c.ID, err) @@ -1561,7 +1562,7 @@ func TestMultiContainerGoferKilled(t *testing.T) { // Wait until sandbox stops. waitForProcessList will loop until sandbox exits // and RPC errors out. impossiblePL := []*control.Process{ - {PID: 100, Cmd: "non-existent-process"}, + {PID: 100, Cmd: "non-existent-process", Threads: []kernel.ThreadID{100}}, } if err := waitForProcessList(c, impossiblePL); err == nil { t.Fatalf("Sandbox was not killed after gofer death") -- cgit v1.2.3 From 481dbfa5ab24ec2c0752b9e748d3617285603c4e Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Wed, 11 Dec 2019 13:40:57 -0800 Subject: Add vfs.Pathname{WithDeleted,ForGetcwd}. The former is needed for vfs.FileDescription to implement memmap.MappingIdentity, and the latter is needed to implement getcwd(2). PiperOrigin-RevId: 285051855 --- pkg/sentry/fsimpl/ext/BUILD | 1 + pkg/sentry/fsimpl/ext/filesystem.go | 8 ++ pkg/sentry/fsimpl/memfs/BUILD | 1 + pkg/sentry/fsimpl/memfs/filesystem.go | 8 ++ pkg/sentry/vfs/BUILD | 1 + pkg/sentry/vfs/filesystem.go | 54 +++++++++++- pkg/sentry/vfs/filesystem_impl_util.go | 26 ++++++ pkg/sentry/vfs/pathname.go | 153 +++++++++++++++++++++++++++++++++ pkg/sentry/vfs/testutil.go | 9 ++ 9 files changed, 260 insertions(+), 1 deletion(-) create mode 100644 pkg/sentry/vfs/pathname.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/ext/BUILD b/pkg/sentry/fsimpl/ext/BUILD index 7ccff8b0d..880b7bcd3 100644 --- a/pkg/sentry/fsimpl/ext/BUILD +++ b/pkg/sentry/fsimpl/ext/BUILD @@ -38,6 +38,7 @@ go_library( "//pkg/abi/linux", "//pkg/binary", "//pkg/fd", + "//pkg/fspath", "//pkg/log", "//pkg/sentry/arch", "//pkg/sentry/context", diff --git a/pkg/sentry/fsimpl/ext/filesystem.go b/pkg/sentry/fsimpl/ext/filesystem.go index 2d15e8aaf..e7aa3b41b 100644 --- a/pkg/sentry/fsimpl/ext/filesystem.go +++ b/pkg/sentry/fsimpl/ext/filesystem.go @@ -20,6 +20,7 @@ import ( "sync" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" "gvisor.dev/gvisor/pkg/sentry/vfs" @@ -441,3 +442,10 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error return syserror.EROFS } + +// PrependPath implements vfs.FilesystemImpl.PrependPath. +func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { + fs.mu.RLock() + defer fs.mu.RUnlock() + return vfs.GenericPrependPath(vfsroot, vd, b) +} diff --git a/pkg/sentry/fsimpl/memfs/BUILD b/pkg/sentry/fsimpl/memfs/BUILD index bc5c0b591..0cc751eb8 100644 --- a/pkg/sentry/fsimpl/memfs/BUILD +++ b/pkg/sentry/fsimpl/memfs/BUILD @@ -32,6 +32,7 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/amutex", + "//pkg/fspath", "//pkg/sentry/arch", "//pkg/sentry/context", "//pkg/sentry/kernel/auth", diff --git a/pkg/sentry/fsimpl/memfs/filesystem.go b/pkg/sentry/fsimpl/memfs/filesystem.go index 08a9cb8ef..1f2a5122a 100644 --- a/pkg/sentry/fsimpl/memfs/filesystem.go +++ b/pkg/sentry/fsimpl/memfs/filesystem.go @@ -19,6 +19,7 @@ import ( "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" @@ -582,3 +583,10 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error inode.decLinksLocked() return nil } + +// PrependPath implements vfs.FilesystemImpl.PrependPath. +func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { + fs.mu.RLock() + defer fs.mu.RUnlock() + return vfs.GenericPrependPath(vfsroot, vd, b) +} diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index 59237c3b9..e3e554b88 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -17,6 +17,7 @@ go_library( "mount.go", "mount_unsafe.go", "options.go", + "pathname.go", "permissions.go", "resolving_path.go", "testutil.go", diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index dfbd2372a..8011eba3f 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -18,6 +18,7 @@ import ( "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/context" ) @@ -185,5 +186,56 @@ type FilesystemImpl interface { // UnlinkAt removes the non-directory file at rp. UnlinkAt(ctx context.Context, rp *ResolvingPath) error - // TODO: d_path(); extended attributes; inotify_add_watch(); bind() + // PrependPath prepends a path from vd to vd.Mount().Root() to b. + // + // If vfsroot.Ok(), it is the contextual VFS root; if it is encountered + // before vd.Mount().Root(), PrependPath should stop prepending path + // components and return a PrependPathAtVFSRootError. + // + // If traversal of vd.Dentry()'s ancestors encounters an independent + // ("root") Dentry that is not vd.Mount().Root() (i.e. vd.Dentry() is not a + // descendant of vd.Mount().Root()), PrependPath should stop prepending + // path components and return a PrependPathAtNonMountRootError. + // + // Filesystems for which Dentries do not have meaningful paths may prepend + // an arbitrary descriptive string to b and then return a + // PrependPathSyntheticError. + // + // Most implementations can acquire the appropriate locks to ensure that + // Dentry.Name() and Dentry.Parent() are fixed for vd.Dentry() and all of + // its ancestors, then call GenericPrependPath. + // + // Preconditions: vd.Mount().Filesystem().Impl() == this FilesystemImpl. + PrependPath(ctx context.Context, vfsroot, vd VirtualDentry, b *fspath.Builder) error + + // TODO: extended attributes; inotify_add_watch(); bind() +} + +// PrependPathAtVFSRootError is returned by implementations of +// FilesystemImpl.PrependPath() when they encounter the contextual VFS root. +type PrependPathAtVFSRootError struct{} + +// Error implements error.Error. +func (PrependPathAtVFSRootError) Error() string { + return "vfs.FilesystemImpl.PrependPath() reached VFS root" +} + +// PrependPathAtNonMountRootError is returned by implementations of +// FilesystemImpl.PrependPath() when they encounter an independent ancestor +// Dentry that is not the Mount root. +type PrependPathAtNonMountRootError struct{} + +// Error implements error.Error. +func (PrependPathAtNonMountRootError) Error() string { + return "vfs.FilesystemImpl.PrependPath() reached root other than Mount root" +} + +// PrependPathSyntheticError is returned by implementations of +// FilesystemImpl.PrependPath() for which prepended names do not represent real +// paths. +type PrependPathSyntheticError struct{} + +// Error implements error.Error. +func (PrependPathSyntheticError) Error() string { + return "vfs.FilesystemImpl.PrependPath() prepended synthetic name" } diff --git a/pkg/sentry/vfs/filesystem_impl_util.go b/pkg/sentry/vfs/filesystem_impl_util.go index 465e610e0..7315a588e 100644 --- a/pkg/sentry/vfs/filesystem_impl_util.go +++ b/pkg/sentry/vfs/filesystem_impl_util.go @@ -16,6 +16,8 @@ package vfs import ( "strings" + + "gvisor.dev/gvisor/pkg/fspath" ) // GenericParseMountOptions parses a comma-separated list of options of the @@ -41,3 +43,27 @@ func GenericParseMountOptions(str string) map[string]string { } return m } + +// GenericPrependPath may be used by implementations of +// FilesystemImpl.PrependPath() for which a single statically-determined lock +// or set of locks is sufficient to ensure its preconditions (as opposed to +// e.g. per-Dentry locks). +// +// Preconditions: Dentry.Name() and Dentry.Parent() must be held constant for +// vd.Dentry() and all of its ancestors. +func GenericPrependPath(vfsroot, vd VirtualDentry, b *fspath.Builder) error { + mnt, d := vd.mount, vd.dentry + for { + if mnt == vfsroot.mount && d == vfsroot.dentry { + return PrependPathAtVFSRootError{} + } + if d == mnt.root { + return nil + } + if d.parent == nil { + return PrependPathAtNonMountRootError{} + } + b.PrependComponent(d.name) + d = d.parent + } +} diff --git a/pkg/sentry/vfs/pathname.go b/pkg/sentry/vfs/pathname.go new file mode 100644 index 000000000..8e155654f --- /dev/null +++ b/pkg/sentry/vfs/pathname.go @@ -0,0 +1,153 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs + +import ( + "sync" + + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/syserror" +) + +var fspathBuilderPool = sync.Pool{ + New: func() interface{} { + return &fspath.Builder{} + }, +} + +func getFSPathBuilder() *fspath.Builder { + return fspathBuilderPool.Get().(*fspath.Builder) +} + +func putFSPathBuilder(b *fspath.Builder) { + // No methods can be called on b after b.String(), so reset it to its zero + // value (as returned by fspathBuilderPool.New) instead. + *b = fspath.Builder{} + fspathBuilderPool.Put(b) +} + +// PathnameWithDeleted returns an absolute pathname to vd, consistent with +// Linux's d_path(). In particular, if vd.Dentry() has been disowned, +// PathnameWithDeleted appends " (deleted)" to the returned pathname. +func (vfs *VirtualFilesystem) PathnameWithDeleted(ctx context.Context, vfsroot, vd VirtualDentry) (string, error) { + b := getFSPathBuilder() + defer putFSPathBuilder(b) + haveRef := false + defer func() { + if haveRef { + vd.DecRef() + } + }() + + origD := vd.dentry +loop: + for { + err := vd.mount.fs.impl.PrependPath(ctx, vfsroot, vd, b) + switch err.(type) { + case nil: + if vd.mount == vfsroot.mount && vd.mount.root == vfsroot.dentry { + // GenericPrependPath() will have returned + // PrependPathAtVFSRootError in this case since it checks + // against vfsroot before mnt.root, but other implementations + // of FilesystemImpl.PrependPath() may return nil instead. + break loop + } + nextVD := vfs.getMountpointAt(vd.mount, vfsroot) + if !nextVD.Ok() { + break loop + } + if haveRef { + vd.DecRef() + } + vd = nextVD + haveRef = true + // continue loop + case PrependPathSyntheticError: + // Skip prepending "/" and appending " (deleted)". + return b.String(), nil + case PrependPathAtVFSRootError, PrependPathAtNonMountRootError: + break loop + default: + return "", err + } + } + b.PrependByte('/') + if origD.IsDisowned() { + b.AppendString(" (deleted)") + } + return b.String(), nil +} + +// PathnameForGetcwd returns an absolute pathname to vd, consistent with +// Linux's sys_getcwd(). +func (vfs *VirtualFilesystem) PathnameForGetcwd(ctx context.Context, vfsroot, vd VirtualDentry) (string, error) { + if vd.dentry.IsDisowned() { + return "", syserror.ENOENT + } + + b := getFSPathBuilder() + defer putFSPathBuilder(b) + haveRef := false + defer func() { + if haveRef { + vd.DecRef() + } + }() + unreachable := false +loop: + for { + err := vd.mount.fs.impl.PrependPath(ctx, vfsroot, vd, b) + switch err.(type) { + case nil: + if vd.mount == vfsroot.mount && vd.mount.root == vfsroot.dentry { + break loop + } + nextVD := vfs.getMountpointAt(vd.mount, vfsroot) + if !nextVD.Ok() { + unreachable = true + break loop + } + if haveRef { + vd.DecRef() + } + vd = nextVD + haveRef = true + case PrependPathAtVFSRootError: + break loop + case PrependPathAtNonMountRootError, PrependPathSyntheticError: + unreachable = true + break loop + default: + return "", err + } + } + b.PrependByte('/') + if unreachable { + b.PrependString("(unreachable)") + } + return b.String(), nil +} + +// As of this writing, we do not have equivalents to: +// +// - d_absolute_path(), which returns EINVAL if (effectively) any call to +// FilesystemImpl.PrependPath() would return PrependPathAtNonMountRootError. +// +// - dentry_path(), which does not walk up mounts (and only returns the path +// relative to Filesystem root), but also appends "//deleted" for disowned +// Dentries. +// +// These should be added as necessary. diff --git a/pkg/sentry/vfs/testutil.go b/pkg/sentry/vfs/testutil.go index 593144cb7..7a1d9e383 100644 --- a/pkg/sentry/vfs/testutil.go +++ b/pkg/sentry/vfs/testutil.go @@ -15,7 +15,10 @@ package vfs import ( + "fmt" + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/syserror" @@ -114,6 +117,12 @@ func (fs *FDTestFilesystem) UnlinkAt(ctx context.Context, rp *ResolvingPath) err return syserror.EPERM } +// PrependPath implements FilesystemImpl.PrependPath. +func (fs *FDTestFilesystem) PrependPath(ctx context.Context, vfsroot, vd VirtualDentry, b *fspath.Builder) error { + b.PrependComponent(fmt.Sprintf("vfs.fdTestDentry:%p", vd.dentry.impl.(*fdTestDentry))) + return PrependPathSyntheticError{} +} + type fdTestDentry struct { vfsd Dentry } -- cgit v1.2.3 From 6fc9f0aefd89ce42ef2c38ea7853f9ba7c4bee04 Mon Sep 17 00:00:00 2001 From: Bhasker Hariharan Date: Wed, 11 Dec 2019 17:51:37 -0800 Subject: Add support for TCP_USER_TIMEOUT option. The implementation follows the linux behavior where specifying a TCP_USER_TIMEOUT will cause the resend timer to honor the user specified timeout rather than the default rto based timeout. Further it alters when connections are timedout due to keepalive failures. It does not alter the behavior of when keepalives are sent. This is as per the linux behavior. PiperOrigin-RevId: 285099795 --- pkg/sentry/socket/netstack/netstack.go | 23 ++++ pkg/tcpip/tcpip.go | 5 + pkg/tcpip/transport/tcp/BUILD | 1 + pkg/tcpip/transport/tcp/accept.go | 15 +++ pkg/tcpip/transport/tcp/connect.go | 19 ++- pkg/tcpip/transport/tcp/endpoint.go | 19 +++ pkg/tcpip/transport/tcp/protocol.go | 21 ++- pkg/tcpip/transport/tcp/rcv.go | 19 ++- pkg/tcpip/transport/tcp/rcv_state.go | 29 ++++ pkg/tcpip/transport/tcp/snd.go | 48 ++++++- pkg/tcpip/transport/tcp/snd_state.go | 10 ++ pkg/tcpip/transport/tcp/tcp_test.go | 194 ++++++++++++++++++++++++--- test/syscalls/linux/socket_inet_loopback.cc | 56 +++++++- test/syscalls/linux/socket_ip_tcp_generic.cc | 63 +++++++++ test/syscalls/linux/tcp_socket.cc | 25 ++++ 15 files changed, 509 insertions(+), 38 deletions(-) create mode 100644 pkg/tcpip/transport/tcp/rcv_state.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index fe5a46aa3..8a6522eac 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -1127,6 +1127,18 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa return int32(time.Duration(v) / time.Second), nil + case linux.TCP_USER_TIMEOUT: + if outLen < sizeOfInt32 { + return nil, syserr.ErrInvalidArgument + } + + var v tcpip.TCPUserTimeoutOption + if err := ep.GetSockOpt(&v); err != nil { + return nil, syserr.TranslateNetstackError(err) + } + + return int32(time.Duration(v) / time.Millisecond), nil + case linux.TCP_INFO: var v tcpip.TCPInfoOption if err := ep.GetSockOpt(&v); err != nil { @@ -1563,6 +1575,17 @@ func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) * } return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.KeepaliveIntervalOption(time.Second * time.Duration(v)))) + case linux.TCP_USER_TIMEOUT: + if len(optVal) < sizeOfInt32 { + return syserr.ErrInvalidArgument + } + + v := int32(usermem.ByteOrder.Uint32(optVal)) + if v < 0 { + return syserr.ErrInvalidArgument + } + return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.TCPUserTimeoutOption(time.Millisecond * time.Duration(v)))) + case linux.TCP_CONGESTION: v := tcpip.CongestionControlOption(optVal) if err := ep.SetSockOpt(v); err != nil { diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index d5bb5b6ed..f62fd729f 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -576,6 +576,11 @@ type KeepaliveIntervalOption time.Duration // closed. type KeepaliveCountOption int +// TCPUserTimeoutOption is used by SetSockOpt/GetSockOpt to specify a user +// specified timeout for a given TCP connection. +// See: RFC5482 for details. +type TCPUserTimeoutOption time.Duration + // CongestionControlOption is used by SetSockOpt/GetSockOpt to set/get // the current congestion control algorithm. type CongestionControlOption string diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD index 455a1c098..3b353d56c 100644 --- a/pkg/tcpip/transport/tcp/BUILD +++ b/pkg/tcpip/transport/tcp/BUILD @@ -28,6 +28,7 @@ go_library( "forwarder.go", "protocol.go", "rcv.go", + "rcv_state.go", "reno.go", "sack.go", "sack_scoreboard.go", diff --git a/pkg/tcpip/transport/tcp/accept.go b/pkg/tcpip/transport/tcp/accept.go index 74df3edfb..5422ae80c 100644 --- a/pkg/tcpip/transport/tcp/accept.go +++ b/pkg/tcpip/transport/tcp/accept.go @@ -242,6 +242,13 @@ func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, i n.initGSO() + // Now inherit any socket options that should be inherited from the + // listening endpoint. + // In case of Forwarder listenEP will be nil and hence this check. + if l.listenEP != nil { + l.listenEP.propagateInheritableOptions(n) + } + // Register new endpoint so that packets are routed to it. if err := n.stack.RegisterTransportEndpoint(n.boundNICID, n.effectiveNetProtos, ProtocolNumber, n.ID, n, n.reusePort, n.boundBindToDevice); err != nil { n.Close() @@ -350,6 +357,14 @@ func (e *endpoint) deliverAccepted(n *endpoint) { } } +// propagateInheritableOptions propagates any options set on the listening +// endpoint to the newly created endpoint. +func (e *endpoint) propagateInheritableOptions(n *endpoint) { + e.mu.Lock() + n.userTimeout = e.userTimeout + e.mu.Unlock() +} + // handleSynSegment is called in its own goroutine once the listening endpoint // receives a SYN segment. It is responsible for completing the handshake and // queueing the new endpoint for acceptance. diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 3d059c302..4c34fc9d2 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -862,7 +862,7 @@ func (e *endpoint) resetConnectionLocked(err *tcpip.Error) { } e.state = StateError e.HardError = err - if err != tcpip.ErrConnectionReset { + if err != tcpip.ErrConnectionReset && err != tcpip.ErrTimeout { // The exact sequence number to be used for the RST is the same as the // one used by Linux. We need to handle the case of window being shrunk // which can cause sndNxt to be outside the acceptable window on the @@ -1087,12 +1087,24 @@ func (e *endpoint) handleSegments() *tcpip.Error { // keepalive packets periodically when the connection is idle. If we don't hear // from the other side after a number of tries, we terminate the connection. func (e *endpoint) keepaliveTimerExpired() *tcpip.Error { + e.mu.RLock() + userTimeout := e.userTimeout + e.mu.RUnlock() + e.keepalive.Lock() if !e.keepalive.enabled || !e.keepalive.timer.checkExpiration() { e.keepalive.Unlock() return nil } + // If a userTimeout is set then abort the connection if it is + // exceeded. + if userTimeout != 0 && time.Since(e.rcv.lastRcvdAckTime) >= userTimeout && e.keepalive.unacked > 0 { + e.keepalive.Unlock() + e.stack.Stats().TCP.EstablishedTimedout.Increment() + return tcpip.ErrTimeout + } + if e.keepalive.unacked >= e.keepalive.count { e.keepalive.Unlock() e.stack.Stats().TCP.EstablishedTimedout.Increment() @@ -1112,7 +1124,6 @@ func (e *endpoint) keepaliveTimerExpired() *tcpip.Error { // whether it is enabled for this endpoint. func (e *endpoint) resetKeepaliveTimer(receivedData bool) { e.keepalive.Lock() - defer e.keepalive.Unlock() if receivedData { e.keepalive.unacked = 0 } @@ -1120,6 +1131,7 @@ func (e *endpoint) resetKeepaliveTimer(receivedData bool) { // data to send. if !e.keepalive.enabled || e.snd == nil || e.snd.sndUna != e.snd.sndNxt { e.keepalive.timer.disable() + e.keepalive.Unlock() return } if e.keepalive.unacked > 0 { @@ -1127,6 +1139,7 @@ func (e *endpoint) resetKeepaliveTimer(receivedData bool) { } else { e.keepalive.timer.enable(e.keepalive.idle) } + e.keepalive.Unlock() } // disableKeepaliveTimer stops the keepalive timer. @@ -1239,6 +1252,7 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { w: &e.snd.resendWaker, f: func() *tcpip.Error { if !e.snd.retransmitTimerExpired() { + e.stack.Stats().TCP.EstablishedTimedout.Increment() return tcpip.ErrTimeout } return nil @@ -1405,6 +1419,7 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { if s == nil { break } + e.tryDeliverSegmentFromClosedEndpoint(s) } diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 4861ab513..dd8b47cbe 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -341,6 +341,7 @@ type endpoint struct { // TCP should never broadcast but Linux nevertheless supports enabling/ // disabling SO_BROADCAST, albeit as a NOOP. broadcast bool + // Values used to reserve a port or register a transport endpoint // (which ever happens first). boundBindToDevice tcpip.NICID @@ -474,6 +475,12 @@ type endpoint struct { // without hearing a response, the connection is closed. keepalive keepalive + // userTimeout if non-zero specifies a user specified timeout for + // a connection w/ pending data to send. A connection that has pending + // unacked data will be forcibily aborted if the timeout is reached + // without any data being acked. + userTimeout time.Duration + // pendingAccepted is a synchronization primitive used to track number // of connections that are queued up to be delivered to the accepted // channel. We use this to ensure that all goroutines blocked on writing @@ -1333,6 +1340,12 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.notifyProtocolGoroutine(notifyKeepaliveChanged) return nil + case tcpip.TCPUserTimeoutOption: + e.mu.Lock() + e.userTimeout = time.Duration(v) + e.mu.Unlock() + return nil + case tcpip.BroadcastOption: e.mu.Lock() e.broadcast = v != 0 @@ -1591,6 +1604,12 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { e.keepalive.Unlock() return nil + case *tcpip.TCPUserTimeoutOption: + e.mu.Lock() + *o = tcpip.TCPUserTimeoutOption(e.userTimeout) + e.mu.Unlock() + return nil + case *tcpip.OutOfBandInlineOption: // We don't currently support disabling this option. *o = 1 diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go index 89b965c23..bc718064c 100644 --- a/pkg/tcpip/transport/tcp/protocol.go +++ b/pkg/tcpip/transport/tcp/protocol.go @@ -162,13 +162,26 @@ func (*protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Transpo func replyWithReset(s *segment) { // Get the seqnum from the packet if the ack flag is set. seq := seqnum.Value(0) + ack := seqnum.Value(0) + flags := byte(header.TCPFlagRst) + // As per RFC 793 page 35 (Reset Generation) + // 1. If the connection does not exist (CLOSED) then a reset is sent + // in response to any incoming segment except another reset. In + // particular, SYNs addressed to a non-existent connection are rejected + // by this means. + + // If the incoming segment has an ACK field, the reset takes its + // sequence number from the ACK field of the segment, otherwise the + // reset has sequence number zero and the ACK field is set to the sum + // of the sequence number and segment length of the incoming segment. + // The connection remains in the CLOSED state. if s.flagIsSet(header.TCPFlagAck) { seq = s.ackNumber + } else { + flags |= header.TCPFlagAck + ack = s.sequenceNumber.Add(s.logicalLen()) } - - ack := s.sequenceNumber.Add(s.logicalLen()) - - sendTCP(&s.route, s.id, buffer.VectorisedView{}, s.route.DefaultTTL(), stack.DefaultTOS, header.TCPFlagRst|header.TCPFlagAck, seq, ack, 0 /* rcvWnd */, nil /* options */, nil /* gso */) + sendTCP(&s.route, s.id, buffer.VectorisedView{}, s.route.DefaultTTL(), stack.DefaultTOS, flags, seq, ack, 0 /* rcvWnd */, nil /* options */, nil /* gso */) } // SetOption implements TransportProtocol.SetOption. diff --git a/pkg/tcpip/transport/tcp/rcv.go b/pkg/tcpip/transport/tcp/rcv.go index 5ee499c36..0a5534959 100644 --- a/pkg/tcpip/transport/tcp/rcv.go +++ b/pkg/tcpip/transport/tcp/rcv.go @@ -50,16 +50,20 @@ type receiver struct { pendingRcvdSegments segmentHeap pendingBufUsed seqnum.Size pendingBufSize seqnum.Size + + // Time when the last ack was received. + lastRcvdAckTime time.Time `state:".(unixTime)"` } func newReceiver(ep *endpoint, irs seqnum.Value, rcvWnd seqnum.Size, rcvWndScale uint8, pendingBufSize seqnum.Size) *receiver { return &receiver{ - ep: ep, - rcvNxt: irs + 1, - rcvAcc: irs.Add(rcvWnd + 1), - rcvWnd: rcvWnd, - rcvWndScale: rcvWndScale, - pendingBufSize: pendingBufSize, + ep: ep, + rcvNxt: irs + 1, + rcvAcc: irs.Add(rcvWnd + 1), + rcvWnd: rcvWnd, + rcvWndScale: rcvWndScale, + pendingBufSize: pendingBufSize, + lastRcvdAckTime: time.Now(), } } @@ -360,6 +364,9 @@ func (r *receiver) handleRcvdSegment(s *segment) (drop bool, err *tcpip.Error) { return true, nil } + // Store the time of the last ack. + r.lastRcvdAckTime = time.Now() + // Defer segment processing if it can't be consumed now. if !r.consumeSegment(s, segSeq, segLen) { if segLen > 0 || s.flagIsSet(header.TCPFlagFin) { diff --git a/pkg/tcpip/transport/tcp/rcv_state.go b/pkg/tcpip/transport/tcp/rcv_state.go new file mode 100644 index 000000000..2bf21a2e7 --- /dev/null +++ b/pkg/tcpip/transport/tcp/rcv_state.go @@ -0,0 +1,29 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tcp + +import ( + "time" +) + +// saveLastRcvdAckTime is invoked by stateify. +func (r *receiver) saveLastRcvdAckTime() unixTime { + return unixTime{r.lastRcvdAckTime.Unix(), r.lastRcvdAckTime.UnixNano()} +} + +// loadLastRcvdAckTime is invoked by stateify. +func (r *receiver) loadLastRcvdAckTime(unix unixTime) { + r.lastRcvdAckTime = time.Unix(unix.second, unix.nano) +} diff --git a/pkg/tcpip/transport/tcp/snd.go b/pkg/tcpip/transport/tcp/snd.go index 8332a0179..8a947dc66 100644 --- a/pkg/tcpip/transport/tcp/snd.go +++ b/pkg/tcpip/transport/tcp/snd.go @@ -28,8 +28,11 @@ import ( ) const ( - // minRTO is the minimum allowed value for the retransmit timeout. - minRTO = 200 * time.Millisecond + // MinRTO is the minimum allowed value for the retransmit timeout. + MinRTO = 200 * time.Millisecond + + // MaxRTO is the maximum allowed value for the retransmit timeout. + MaxRTO = 120 * time.Second // InitialCwnd is the initial congestion window. InitialCwnd = 10 @@ -134,6 +137,10 @@ type sender struct { // rttMeasureTime is the time when the rttMeasureSeqNum was sent. rttMeasureTime time.Time `state:".(unixTime)"` + // firstRetransmittedSegXmitTime is the original transmit time of + // the first segment that was retransmitted due to RTO expiration. + firstRetransmittedSegXmitTime time.Time `state:".(unixTime)"` + closed bool writeNext *segment writeList segmentList @@ -392,8 +399,8 @@ func (s *sender) updateRTO(rtt time.Duration) { s.rto = s.rtt.srtt + 4*s.rtt.rttvar s.rtt.Unlock() - if s.rto < minRTO { - s.rto = minRTO + if s.rto < MinRTO { + s.rto = MinRTO } } @@ -438,8 +445,30 @@ func (s *sender) retransmitTimerExpired() bool { s.ep.stack.Stats().TCP.Timeouts.Increment() s.ep.stats.SendErrors.Timeouts.Increment() - // Give up if we've waited more than a minute since the last resend. - if s.rto >= 60*time.Second { + // Give up if we've waited more than a minute since the last resend or + // if a user time out is set and we have exceeded the user specified + // timeout since the first retransmission. + s.ep.mu.RLock() + uto := s.ep.userTimeout + s.ep.mu.RUnlock() + + if s.firstRetransmittedSegXmitTime.IsZero() { + // We store the original xmitTime of the segment that we are + // about to retransmit as the retransmission time. This is + // required as by the time the retransmitTimer has expired the + // segment has already been sent and unacked for the RTO at the + // time the segment was sent. + s.firstRetransmittedSegXmitTime = s.writeList.Front().xmitTime + } + + elapsed := time.Since(s.firstRetransmittedSegXmitTime) + remaining := MaxRTO + if uto != 0 { + // Cap to the user specified timeout if one is specified. + remaining = uto - elapsed + } + + if remaining <= 0 || s.rto >= MaxRTO { return false } @@ -447,6 +476,11 @@ func (s *sender) retransmitTimerExpired() bool { // below. s.rto *= 2 + // Cap RTO to remaining time. + if s.rto > remaining { + s.rto = remaining + } + // See: https://tools.ietf.org/html/rfc6582#section-3.2 Step 4. // // Retransmit timeouts: @@ -1168,6 +1202,8 @@ func (s *sender) handleRcvdSegment(seg *segment) { // RFC 6298 Rule 5.3 if s.sndUna == s.sndNxt { s.outstanding = 0 + // Reset firstRetransmittedSegXmitTime to the zero value. + s.firstRetransmittedSegXmitTime = time.Time{} s.resendTimer.disable() } } diff --git a/pkg/tcpip/transport/tcp/snd_state.go b/pkg/tcpip/transport/tcp/snd_state.go index 12eff8afc..8b20c3455 100644 --- a/pkg/tcpip/transport/tcp/snd_state.go +++ b/pkg/tcpip/transport/tcp/snd_state.go @@ -48,3 +48,13 @@ func (s *sender) loadRttMeasureTime(unix unixTime) { func (s *sender) afterLoad() { s.resendTimer.init(&s.resendWaker) } + +// saveFirstRetransmittedSegXmitTime is invoked by stateify. +func (s *sender) saveFirstRetransmittedSegXmitTime() unixTime { + return unixTime{s.firstRetransmittedSegXmitTime.Unix(), s.firstRetransmittedSegXmitTime.UnixNano()} +} + +// loadFirstRetransmittedSegXmitTime is invoked by stateify. +func (s *sender) loadFirstRetransmittedSegXmitTime(unix unixTime) { + s.firstRetransmittedSegXmitTime = time.Unix(unix.second, unix.nano) +} diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index bc5cfcf0e..2a83f7bcc 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -323,8 +323,8 @@ func TestTCPResetSentForACKWhenNotUsingSynCookies(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.SeqNum(uint32(c.IRS+1)), - checker.AckNum(uint32(iss)+1), - checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck))) + checker.AckNum(0), + checker.TCPFlags(header.TCPFlagRst))) } func TestTCPResetsReceivedIncrement(t *testing.T) { @@ -460,18 +460,17 @@ func TestConnectResetAfterClose(t *testing.T) { checker.TCP( checker.DstPort(context.TestPort), checker.SeqNum(uint32(c.IRS)+2), - checker.AckNum(790), - checker.TCPFlags(header.TCPFlagAck|header.TCPFlagRst), + checker.AckNum(0), + checker.TCPFlags(header.TCPFlagRst), ), ) break } } -// TestClosingWithEnqueuedSegments tests handling of -// still enqueued segments when the endpoint transitions -// to StateClose. The in-flight segments would be re-enqueued -// to a any listening endpoint. +// TestClosingWithEnqueuedSegments tests handling of still enqueued segments +// when the endpoint transitions to StateClose. The in-flight segments would be +// re-enqueued to a any listening endpoint. func TestClosingWithEnqueuedSegments(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() @@ -576,8 +575,8 @@ func TestClosingWithEnqueuedSegments(t *testing.T) { checker.TCP( checker.DstPort(context.TestPort), checker.SeqNum(uint32(c.IRS)+2), - checker.AckNum(793), - checker.TCPFlags(header.TCPFlagAck|header.TCPFlagRst), + checker.AckNum(0), + checker.TCPFlags(header.TCPFlagRst), ), ) } @@ -914,7 +913,7 @@ func TestSendRstOnListenerRxAckV4(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck), + checker.TCPFlags(header.TCPFlagRst), checker.SeqNum(200))) } @@ -942,7 +941,7 @@ func TestSendRstOnListenerRxAckV6(t *testing.T) { checker.IPv6(t, c.GetV6Packet(), checker.TCP( checker.DstPort(context.TestPort), - checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck), + checker.TCPFlags(header.TCPFlagRst), checker.SeqNum(200))) } @@ -4291,8 +4290,9 @@ func TestKeepalive(t *testing.T) { c.CreateConnected(789, 30000, -1 /* epRcvBuf */) + const keepAliveInterval = 10 * time.Millisecond c.EP.SetSockOpt(tcpip.KeepaliveIdleOption(10 * time.Millisecond)) - c.EP.SetSockOpt(tcpip.KeepaliveIntervalOption(10 * time.Millisecond)) + c.EP.SetSockOpt(tcpip.KeepaliveIntervalOption(keepAliveInterval)) c.EP.SetSockOpt(tcpip.KeepaliveCountOption(5)) c.EP.SetSockOpt(tcpip.KeepaliveEnabledOption(1)) @@ -4382,13 +4382,29 @@ func TestKeepalive(t *testing.T) { ) } + // Sleep for a litte over the KeepAlive interval to make sure + // the timer has time to fire after the last ACK and close the + // close the socket. + time.Sleep(keepAliveInterval + 5*time.Millisecond) + // The connection should be terminated after 5 unacked keepalives. + // Send an ACK to trigger a RST from the stack as the endpoint should + // be dead. + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: c.Port, + Flags: header.TCPFlagAck, + SeqNum: 790, + AckNum: seqnum.Value(next), + RcvWnd: 30000, + }) + checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), checker.SeqNum(uint32(next)), - checker.AckNum(uint32(790)), - checker.TCPFlags(header.TCPFlagAck|header.TCPFlagRst), + checker.AckNum(uint32(0)), + checker.TCPFlags(header.TCPFlagRst), ), ) @@ -6157,8 +6173,8 @@ func TestTCPTimeWaitDuplicateFINExtendsTimeWait(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.SeqNum(uint32(ackHeaders.AckNum)), - checker.AckNum(uint32(ackHeaders.SeqNum)), - checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck))) + checker.AckNum(0), + checker.TCPFlags(header.TCPFlagRst))) if got := c.Stack().Stats().TCP.EstablishedClosed.Value(); got != want { t.Errorf("got c.Stack().Stats().TCP.EstablishedClosed = %v, want = %v", got, want) @@ -6336,7 +6352,147 @@ func TestTCPCloseWithData(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.SeqNum(uint32(ackHeaders.AckNum)), - checker.AckNum(uint32(ackHeaders.SeqNum)), - checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck))) + checker.AckNum(0), + checker.TCPFlags(header.TCPFlagRst))) +} + +func TestTCPUserTimeout(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) + + origEstablishedTimedout := c.Stack().Stats().TCP.EstablishedTimedout.Value() + userTimeout := 50 * time.Millisecond + c.EP.SetSockOpt(tcpip.TCPUserTimeoutOption(userTimeout)) + + // Send some data and wait before ACKing it. + view := buffer.NewView(3) + if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { + t.Fatalf("Write failed: %v", err) + } + + next := uint32(c.IRS) + 1 + checker.IPv4(t, c.GetPacket(), + checker.PayloadLen(len(view)+header.TCPMinimumSize), + checker.TCP( + checker.DstPort(context.TestPort), + checker.SeqNum(next), + checker.AckNum(790), + checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), + ), + ) + + // Wait for a little over the minimum retransmit timeout of 200ms for + // the retransmitTimer to fire and close the connection. + time.Sleep(tcp.MinRTO + 10*time.Millisecond) + + // No packet should be received as the connection should be silently + // closed due to timeout. + c.CheckNoPacket("unexpected packet received after userTimeout has expired") + + next += uint32(len(view)) + + // The connection should be terminated after userTimeout has expired. + // Send an ACK to trigger a RST from the stack as the endpoint should + // be dead. + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: c.Port, + Flags: header.TCPFlagAck, + SeqNum: 790, + AckNum: seqnum.Value(next), + RcvWnd: 30000, + }) + + checker.IPv4(t, c.GetPacket(), + checker.TCP( + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(next)), + checker.AckNum(uint32(0)), + checker.TCPFlags(header.TCPFlagRst), + ), + ) + + if _, _, err := c.EP.Read(nil); err != tcpip.ErrTimeout { + t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrTimeout) + } + + if got, want := c.Stack().Stats().TCP.EstablishedTimedout.Value(), origEstablishedTimedout+1; got != want { + t.Errorf("got c.Stack().Stats().TCP.EstablishedTimedout = %v, want = %v", got, want) + } +} + +func TestKeepaliveWithUserTimeout(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) + + origEstablishedTimedout := c.Stack().Stats().TCP.EstablishedTimedout.Value() + + const keepAliveInterval = 10 * time.Millisecond + c.EP.SetSockOpt(tcpip.KeepaliveIdleOption(10 * time.Millisecond)) + c.EP.SetSockOpt(tcpip.KeepaliveIntervalOption(keepAliveInterval)) + c.EP.SetSockOpt(tcpip.KeepaliveCountOption(10)) + c.EP.SetSockOpt(tcpip.KeepaliveEnabledOption(1)) + + // Set userTimeout to be the duration for 3 keepalive probes. + userTimeout := 30 * time.Millisecond + c.EP.SetSockOpt(tcpip.TCPUserTimeoutOption(userTimeout)) + + // Check that the connection is still alive. + if _, _, err := c.EP.Read(nil); err != tcpip.ErrWouldBlock { + t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrWouldBlock) + } + + // Now receive 2 keepalives, but don't ACK them. The connection should + // be reset when the 3rd one should be sent due to userTimeout being + // 30ms and each keepalive probe should be sent 10ms apart as set above after + // the connection has been idle for 10ms. + for i := 0; i < 2; i++ { + b := c.GetPacket() + checker.IPv4(t, b, + checker.TCP( + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS)), + checker.AckNum(uint32(790)), + checker.TCPFlags(header.TCPFlagAck), + ), + ) + } + + // Sleep for a litte over the KeepAlive interval to make sure + // the timer has time to fire after the last ACK and close the + // close the socket. + time.Sleep(keepAliveInterval + 5*time.Millisecond) + + // The connection should be terminated after 30ms. + // Send an ACK to trigger a RST from the stack as the endpoint should + // be dead. + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: c.Port, + Flags: header.TCPFlagAck, + SeqNum: 790, + AckNum: seqnum.Value(c.IRS + 1), + RcvWnd: 30000, + }) + + checker.IPv4(t, c.GetPacket(), + checker.TCP( + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+1)), + checker.AckNum(uint32(0)), + checker.TCPFlags(header.TCPFlagRst), + ), + ) + + if _, _, err := c.EP.Read(nil); err != tcpip.ErrTimeout { + t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrTimeout) + } + if got, want := c.Stack().Stats().TCP.EstablishedTimedout.Value(), origEstablishedTimedout+1; got != want { + t.Errorf("got c.Stack().Stats().TCP.EstablishedTimedout = %v, want = %v", got, want) + } } diff --git a/test/syscalls/linux/socket_inet_loopback.cc b/test/syscalls/linux/socket_inet_loopback.cc index fa4358ae4..761c3a9fe 100644 --- a/test/syscalls/linux/socket_inet_loopback.cc +++ b/test/syscalls/linux/socket_inet_loopback.cc @@ -206,7 +206,7 @@ TEST_P(SocketInetLoopbackTest, TCPListenClose) { } // TODO(b/138400178): Fix cooperative S/R failure when ds.reset() is invoked // before function end. - // ds.reset() + // ds.reset(); } TEST_P(SocketInetLoopbackTest, TCPbacklog) { @@ -603,6 +603,60 @@ TEST_P(SocketInetLoopbackTest, TCPTimeWaitTest_NoRandomSave) { SyscallSucceeds()); } +TEST_P(SocketInetLoopbackTest, AcceptedInheritsTCPUserTimeout) { + auto const& param = GetParam(); + TestAddress const& listener = param.listener; + TestAddress const& connector = param.connector; + + // Create the listening socket. + const FileDescriptor listen_fd = ASSERT_NO_ERRNO_AND_VALUE( + Socket(listener.family(), SOCK_STREAM, IPPROTO_TCP)); + sockaddr_storage listen_addr = listener.addr; + ASSERT_THAT(bind(listen_fd.get(), reinterpret_cast(&listen_addr), + listener.addr_len), + SyscallSucceeds()); + ASSERT_THAT(listen(listen_fd.get(), SOMAXCONN), SyscallSucceeds()); + + // Get the port bound by the listening socket. + socklen_t addrlen = listener.addr_len; + ASSERT_THAT(getsockname(listen_fd.get(), + reinterpret_cast(&listen_addr), &addrlen), + SyscallSucceeds()); + + const uint16_t port = + ASSERT_NO_ERRNO_AND_VALUE(AddrPort(listener.family(), listen_addr)); + + // Set the userTimeout on the listening socket. + constexpr int kUserTimeout = 10; + ASSERT_THAT(setsockopt(listen_fd.get(), IPPROTO_TCP, TCP_USER_TIMEOUT, + &kUserTimeout, sizeof(kUserTimeout)), + SyscallSucceeds()); + + // Connect to the listening socket. + FileDescriptor conn_fd = ASSERT_NO_ERRNO_AND_VALUE( + Socket(connector.family(), SOCK_STREAM, IPPROTO_TCP)); + + sockaddr_storage conn_addr = connector.addr; + ASSERT_NO_ERRNO(SetAddrPort(connector.family(), &conn_addr, port)); + ASSERT_THAT(RetryEINTR(connect)(conn_fd.get(), + reinterpret_cast(&conn_addr), + connector.addr_len), + SyscallSucceeds()); + + // Accept the connection. + auto accepted = + ASSERT_NO_ERRNO_AND_VALUE(Accept(listen_fd.get(), nullptr, nullptr)); + // Verify that the accepted socket inherited the user timeout set on + // listening socket. + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT( + getsockopt(accepted.get(), IPPROTO_TCP, TCP_USER_TIMEOUT, &get, &get_len), + SyscallSucceeds()); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kUserTimeout); +} + INSTANTIATE_TEST_SUITE_P( All, SocketInetLoopbackTest, ::testing::Values( diff --git a/test/syscalls/linux/socket_ip_tcp_generic.cc b/test/syscalls/linux/socket_ip_tcp_generic.cc index c74273436..57ce8e169 100644 --- a/test/syscalls/linux/socket_ip_tcp_generic.cc +++ b/test/syscalls/linux/socket_ip_tcp_generic.cc @@ -812,5 +812,68 @@ TEST_P(TCPSocketPairTest, TestTCPCloseWithData) { ASSERT_THAT(close(sockets->release_first_fd()), SyscallSucceeds()); } +TEST_P(TCPSocketPairTest, TCPUserTimeoutDefault) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT(getsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_USER_TIMEOUT, + &get, &get_len), + SyscallSucceeds()); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, 0); // 0 ms (disabled). +} + +TEST_P(TCPSocketPairTest, SetTCPUserTimeoutZero) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + constexpr int kZero = 0; + ASSERT_THAT(setsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_USER_TIMEOUT, + &kZero, sizeof(kZero)), + SyscallSucceeds()); + + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT(getsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_USER_TIMEOUT, + &get, &get_len), + SyscallSucceeds()); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, 0); // 0 ms (disabled). +} + +TEST_P(TCPSocketPairTest, SetTCPUserTimeoutBelowZero) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + constexpr int kNeg = -10; + EXPECT_THAT(setsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_USER_TIMEOUT, + &kNeg, sizeof(kNeg)), + SyscallFailsWithErrno(EINVAL)); + + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT(getsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_USER_TIMEOUT, + &get, &get_len), + SyscallSucceeds()); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, 0); // 0 ms (disabled). +} + +TEST_P(TCPSocketPairTest, SetTCPUserTimeoutAboveZero) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + constexpr int kAbove = 10; + ASSERT_THAT(setsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_USER_TIMEOUT, + &kAbove, sizeof(kAbove)), + SyscallSucceeds()); + + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT(getsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_USER_TIMEOUT, + &get, &get_len), + SyscallSucceeds()); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kAbove); +} + } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/tcp_socket.cc b/test/syscalls/linux/tcp_socket.cc index 99863b0ed..c503f3568 100644 --- a/test/syscalls/linux/tcp_socket.cc +++ b/test/syscalls/linux/tcp_socket.cc @@ -1175,6 +1175,31 @@ TEST_P(SimpleTcpSocketTest, SetMaxSegFailsForInvalidMSSValues) { } } +TEST_P(SimpleTcpSocketTest, SetTCPUserTimeout) { + FileDescriptor s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + + { + constexpr int kTCPUserTimeout = -1; + EXPECT_THAT(setsockopt(s.get(), IPPROTO_TCP, TCP_USER_TIMEOUT, + &kTCPUserTimeout, sizeof(kTCPUserTimeout)), + SyscallFailsWithErrno(EINVAL)); + } + + // kTCPUserTimeout is in milliseconds. + constexpr int kTCPUserTimeout = 100; + ASSERT_THAT(setsockopt(s.get(), IPPROTO_TCP, TCP_USER_TIMEOUT, + &kTCPUserTimeout, sizeof(kTCPUserTimeout)), + SyscallSucceedsWithValue(0)); + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT( + getsockopt(s.get(), IPPROTO_TCP, TCP_USER_TIMEOUT, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kTCPUserTimeout); +} + INSTANTIATE_TEST_SUITE_P(AllInetTests, SimpleTcpSocketTest, ::testing::Values(AF_INET, AF_INET6)); -- cgit v1.2.3 From 378d6c1f3697b8b939e6632e980562bfc8fb2781 Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Thu, 12 Dec 2019 11:07:25 -0800 Subject: unix: allow to bind unix sockets only to AF_UNIX addresses Reported-by: syzbot+2c0bcfd87fb4e8b7b009@syzkaller.appspotmail.com PiperOrigin-RevId: 285228312 --- pkg/sentry/socket/netstack/netstack.go | 2 +- pkg/sentry/socket/unix/unix.go | 3 +++ test/syscalls/linux/socket_unix.cc | 15 +++++++++++++++ 3 files changed, 19 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 8a6522eac..140851c17 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -326,7 +326,7 @@ func AddressAndFamily(sfamily int, addr []byte, strict bool) (tcpip.FullAddress, } family := usermem.ByteOrder.Uint16(addr) - if family != uint16(sfamily) && (!strict && family != linux.AF_UNSPEC) { + if family != uint16(sfamily) && (strict || family != linux.AF_UNSPEC) { return tcpip.FullAddress{}, family, syserr.ErrAddressFamilyNotSupported } diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go index 1aaae8487..885758054 100644 --- a/pkg/sentry/socket/unix/unix.go +++ b/pkg/sentry/socket/unix/unix.go @@ -118,6 +118,9 @@ func (s *SocketOperations) Endpoint() transport.Endpoint { func extractPath(sockaddr []byte) (string, *syserr.Error) { addr, _, err := netstack.AddressAndFamily(linux.AF_UNIX, sockaddr, true /* strict */) if err != nil { + if err == syserr.ErrAddressFamilyNotSupported { + err = syserr.ErrInvalidArgument + } return "", err } diff --git a/test/syscalls/linux/socket_unix.cc b/test/syscalls/linux/socket_unix.cc index 8a28202a8..4cf1f76f1 100644 --- a/test/syscalls/linux/socket_unix.cc +++ b/test/syscalls/linux/socket_unix.cc @@ -65,6 +65,21 @@ TEST_P(UnixSocketPairTest, BindToBadName) { SyscallFailsWithErrno(ENOENT)); } +TEST_P(UnixSocketPairTest, BindToBadFamily) { + auto pair = + ASSERT_NO_ERRNO_AND_VALUE(UnixDomainSocketPair(SOCK_SEQPACKET).Create()); + + constexpr char kBadName[] = "/some/path/that/does/not/exist"; + sockaddr_un sockaddr; + sockaddr.sun_family = AF_INET; + memcpy(sockaddr.sun_path, kBadName, sizeof(kBadName)); + + EXPECT_THAT( + bind(pair->first_fd(), reinterpret_cast(&sockaddr), + sizeof(sockaddr)), + SyscallFailsWithErrno(EINVAL)); +} + TEST_P(UnixSocketPairTest, RecvmmsgTimeoutAfterRecv) { auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); char sent_data[10]; -- cgit v1.2.3 From 007707a0726602bc99fc0dccaf2994ad967b9d96 Mon Sep 17 00:00:00 2001 From: Rahat Mahmood Date: Thu, 12 Dec 2019 11:19:18 -0800 Subject: Implement kernfs. PiperOrigin-RevId: 285231002 --- pkg/abi/linux/file.go | 23 + pkg/abi/linux/fs.go | 7 + pkg/sentry/fsimpl/kernfs/BUILD | 60 +++ pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go | 131 +++++ pkg/sentry/fsimpl/kernfs/fd_impl_util.go | 207 ++++++++ pkg/sentry/fsimpl/kernfs/filesystem.go | 691 +++++++++++++++++++++++++ pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 492 ++++++++++++++++++ pkg/sentry/fsimpl/kernfs/kernfs.go | 405 +++++++++++++++ pkg/sentry/fsimpl/kernfs/kernfs_test.go | 423 +++++++++++++++ 9 files changed, 2439 insertions(+) create mode 100644 pkg/sentry/fsimpl/kernfs/BUILD create mode 100644 pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go create mode 100644 pkg/sentry/fsimpl/kernfs/fd_impl_util.go create mode 100644 pkg/sentry/fsimpl/kernfs/filesystem.go create mode 100644 pkg/sentry/fsimpl/kernfs/inode_impl_util.go create mode 100644 pkg/sentry/fsimpl/kernfs/kernfs.go create mode 100644 pkg/sentry/fsimpl/kernfs/kernfs_test.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/file.go b/pkg/abi/linux/file.go index 0f014d27f..16791d03e 100644 --- a/pkg/abi/linux/file.go +++ b/pkg/abi/linux/file.go @@ -286,6 +286,29 @@ func (m FileMode) String() string { return strings.Join(s, "|") } +// DirentType maps file types to dirent types appropriate for (struct +// dirent)::d_type. +func (m FileMode) DirentType() uint8 { + switch m.FileType() { + case ModeSocket: + return DT_SOCK + case ModeSymlink: + return DT_LNK + case ModeRegular: + return DT_REG + case ModeBlockDevice: + return DT_BLK + case ModeDirectory: + return DT_DIR + case ModeCharacterDevice: + return DT_CHR + case ModeNamedPipe: + return DT_FIFO + default: + return DT_UNKNOWN + } +} + var modeExtraBits = abi.FlagSet{ { Flag: ModeSetUID, diff --git a/pkg/abi/linux/fs.go b/pkg/abi/linux/fs.go index b416e3472..2c652baa2 100644 --- a/pkg/abi/linux/fs.go +++ b/pkg/abi/linux/fs.go @@ -92,3 +92,10 @@ const ( SYNC_FILE_RANGE_WRITE = 2 SYNC_FILE_RANGE_WAIT_AFTER = 4 ) + +// Flag argument to renameat2(2), from include/uapi/linux/fs.h. +const ( + RENAME_NOREPLACE = (1 << 0) // Don't overwrite target. + RENAME_EXCHANGE = (1 << 1) // Exchange src and dst. + RENAME_WHITEOUT = (1 << 2) // Whiteout src. +) diff --git a/pkg/sentry/fsimpl/kernfs/BUILD b/pkg/sentry/fsimpl/kernfs/BUILD new file mode 100644 index 000000000..52596c090 --- /dev/null +++ b/pkg/sentry/fsimpl/kernfs/BUILD @@ -0,0 +1,60 @@ +load("//tools/go_stateify:defs.bzl", "go_library") +load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools/go_generics:defs.bzl", "go_template_instance") + +package(licenses = ["notice"]) + +go_template_instance( + name = "slot_list", + out = "slot_list.go", + package = "kernfs", + prefix = "slot", + template = "//pkg/ilist:generic_list", + types = { + "Element": "*slot", + "Linker": "*slot", + }, +) + +go_library( + name = "kernfs", + srcs = [ + "dynamic_bytes_file.go", + "fd_impl_util.go", + "filesystem.go", + "inode_impl_util.go", + "kernfs.go", + "slot_list.go", + ], + importpath = "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs", + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/abi/linux", + "//pkg/fspath", + "//pkg/log", + "//pkg/refs", + "//pkg/sentry/context", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/memmap", + "//pkg/sentry/usermem", + "//pkg/sentry/vfs", + "//pkg/syserror", + ], +) + +go_test( + name = "kernfs_test", + size = "small", + srcs = ["kernfs_test.go"], + deps = [ + ":kernfs", + "//pkg/abi/linux", + "//pkg/sentry/context", + "//pkg/sentry/context/contexttest", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/usermem", + "//pkg/sentry/vfs", + "//pkg/syserror", + "@com_github_google_go-cmp//cmp:go_default_library", + ], +) diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go new file mode 100644 index 000000000..30c06baf0 --- /dev/null +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -0,0 +1,131 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package kernfs + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// DynamicBytesFile implements kernfs.Inode and represents a read-only +// file whose contents are backed by a vfs.DynamicBytesSource. +// +// Must be initialized with Init before first use. +type DynamicBytesFile struct { + InodeAttrs + InodeNoopRefCount + InodeNotDirectory + InodeNotSymlink + + data vfs.DynamicBytesSource +} + +// Init intializes a dynamic bytes file. +func (f *DynamicBytesFile) Init(creds *auth.Credentials, ino uint64, data vfs.DynamicBytesSource) { + f.InodeAttrs.Init(creds, ino, linux.ModeRegular|0444) + f.data = data +} + +// Open implements Inode.Open. +func (f *DynamicBytesFile) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { + fd := &DynamicBytesFD{} + fd.Init(rp.Mount(), vfsd, f.data, flags) + return &fd.vfsfd, nil +} + +// SetStat implements Inode.SetStat. +func (f *DynamicBytesFile) SetStat(*vfs.Filesystem, vfs.SetStatOptions) error { + // DynamicBytesFiles are immutable. + return syserror.EPERM +} + +// DynamicBytesFD implements vfs.FileDescriptionImpl for an FD backed by a +// DynamicBytesFile. +// +// Must be initialized with Init before first use. +type DynamicBytesFD struct { + vfs.FileDescriptionDefaultImpl + vfs.DynamicBytesFileDescriptionImpl + + vfsfd vfs.FileDescription + inode Inode + flags uint32 +} + +// Init initializes a DynamicBytesFD. +func (fd *DynamicBytesFD) Init(m *vfs.Mount, d *vfs.Dentry, data vfs.DynamicBytesSource, flags uint32) { + m.IncRef() // DecRef in vfs.FileDescription.vd.DecRef on final ref. + d.IncRef() // DecRef in vfs.FileDescription.vd.DecRef on final ref. + fd.flags = flags + fd.inode = d.Impl().(*Dentry).inode + fd.SetDataSource(data) + fd.vfsfd.Init(fd, m, d) +} + +// Seek implements vfs.FileDescriptionImpl.Seek. +func (fd *DynamicBytesFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + return fd.DynamicBytesFileDescriptionImpl.Seek(ctx, offset, whence) +} + +// Read implmenets vfs.FileDescriptionImpl.Read. +func (fd *DynamicBytesFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + return fd.DynamicBytesFileDescriptionImpl.Read(ctx, dst, opts) +} + +// PRead implmenets vfs.FileDescriptionImpl.PRead. +func (fd *DynamicBytesFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + return fd.DynamicBytesFileDescriptionImpl.PRead(ctx, dst, offset, opts) +} + +// Write implements vfs.FileDescriptionImpl.Write. +func (fd *DynamicBytesFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + return fd.FileDescriptionDefaultImpl.Write(ctx, src, opts) +} + +// PWrite implements vfs.FileDescriptionImpl.PWrite. +func (fd *DynamicBytesFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + return fd.FileDescriptionDefaultImpl.PWrite(ctx, src, offset, opts) +} + +// Release implements vfs.FileDescriptionImpl.Release. +func (fd *DynamicBytesFD) Release() {} + +// Stat implements vfs.FileDescriptionImpl.Stat. +func (fd *DynamicBytesFD) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { + fs := fd.vfsfd.VirtualDentry().Mount().Filesystem() + return fd.inode.Stat(fs), nil +} + +// SetStat implements vfs.FileDescriptionImpl.SetStat. +func (fd *DynamicBytesFD) SetStat(context.Context, vfs.SetStatOptions) error { + // DynamicBytesFiles are immutable. + return syserror.EPERM +} + +// StatusFlags implements vfs.FileDescriptionImpl.StatusFlags. +func (fd *DynamicBytesFD) StatusFlags(ctx context.Context) (uint32, error) { + return fd.flags, nil +} + +// SetStatusFlags implements vfs.FileDescriptionImpl.SetStatusFlags. +func (fd *DynamicBytesFD) SetStatusFlags(ctx context.Context, flags uint32) error { + // None of the flags settable by fcntl(F_SETFL) are supported, so this is a + // no-op. + return nil +} diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go new file mode 100644 index 000000000..d6c18937a --- /dev/null +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -0,0 +1,207 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package kernfs + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/memmap" + "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// GenericDirectoryFD implements vfs.FileDescriptionImpl for a generic directory +// inode that uses OrderChildren to track child nodes. GenericDirectoryFD is not +// compatible with dynamic directories. +// +// Note that GenericDirectoryFD holds a lock over OrderedChildren while calling +// IterDirents callback. The IterDirents callback therefore cannot hash or +// unhash children, or recursively call IterDirents on the same underlying +// inode. +// +// Must be initialize with Init before first use. +type GenericDirectoryFD struct { + vfs.FileDescriptionDefaultImpl + vfs.DirectoryFileDescriptionDefaultImpl + + vfsfd vfs.FileDescription + children *OrderedChildren + flags uint32 + off int64 +} + +// Init initializes a GenericDirectoryFD. +func (fd *GenericDirectoryFD) Init(m *vfs.Mount, d *vfs.Dentry, children *OrderedChildren, flags uint32) { + m.IncRef() // DecRef in vfs.FileDescription.vd.DecRef on final ref. + d.IncRef() // DecRef in vfs.FileDescription.vd.DecRef on final ref. + fd.children = children + fd.flags = flags + fd.vfsfd.Init(fd, m, d) +} + +// VFSFileDescription returns a pointer to the vfs.FileDescription representing +// this object. +func (fd *GenericDirectoryFD) VFSFileDescription() *vfs.FileDescription { + return &fd.vfsfd +} + +// ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. +func (fd *GenericDirectoryFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { + return fd.FileDescriptionDefaultImpl.ConfigureMMap(ctx, opts) +} + +// Read implmenets vfs.FileDescriptionImpl.Read. +func (fd *GenericDirectoryFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + return fd.DirectoryFileDescriptionDefaultImpl.Read(ctx, dst, opts) +} + +// PRead implmenets vfs.FileDescriptionImpl.PRead. +func (fd *GenericDirectoryFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + return fd.DirectoryFileDescriptionDefaultImpl.PRead(ctx, dst, offset, opts) +} + +// Write implements vfs.FileDescriptionImpl.Write. +func (fd *GenericDirectoryFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + return fd.DirectoryFileDescriptionDefaultImpl.Write(ctx, src, opts) +} + +// PWrite implements vfs.FileDescriptionImpl.PWrite. +func (fd *GenericDirectoryFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + return fd.DirectoryFileDescriptionDefaultImpl.PWrite(ctx, src, offset, opts) +} + +// Release implements vfs.FileDecriptionImpl.Release. +func (fd *GenericDirectoryFD) Release() {} + +func (fd *GenericDirectoryFD) filesystem() *vfs.Filesystem { + return fd.vfsfd.VirtualDentry().Mount().Filesystem() +} + +func (fd *GenericDirectoryFD) inode() Inode { + return fd.vfsfd.VirtualDentry().Dentry().Impl().(*Dentry).inode +} + +// IterDirents implements vfs.FileDecriptionImpl.IterDirents. IterDirents holds +// o.mu when calling cb. +func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback) error { + vfsFS := fd.filesystem() + fs := vfsFS.Impl().(*Filesystem) + vfsd := fd.vfsfd.VirtualDentry().Dentry() + + fs.mu.Lock() + defer fs.mu.Unlock() + + // Handle ".". + if fd.off == 0 { + stat := fd.inode().Stat(vfsFS) + dirent := vfs.Dirent{ + Name: ".", + Type: linux.DT_DIR, + Ino: stat.Ino, + NextOff: 1, + } + if !cb.Handle(dirent) { + return nil + } + fd.off++ + } + + // Handle "..". + if fd.off == 1 { + parentInode := vfsd.ParentOrSelf().Impl().(*Dentry).inode + stat := parentInode.Stat(vfsFS) + dirent := vfs.Dirent{ + Name: "..", + Type: linux.FileMode(stat.Mode).DirentType(), + Ino: stat.Ino, + NextOff: 2, + } + if !cb.Handle(dirent) { + return nil + } + fd.off++ + } + + // Handle static children. + fd.children.mu.RLock() + defer fd.children.mu.RUnlock() + // fd.off accounts for "." and "..", but fd.children do not track + // these. + childIdx := fd.off - 2 + for it := fd.children.nthLocked(childIdx); it != nil; it = it.Next() { + inode := it.Dentry.Impl().(*Dentry).inode + stat := inode.Stat(vfsFS) + dirent := vfs.Dirent{ + Name: it.Name, + Type: linux.FileMode(stat.Mode).DirentType(), + Ino: stat.Ino, + NextOff: fd.off + 1, + } + if !cb.Handle(dirent) { + return nil + } + fd.off++ + } + + return nil +} + +// Seek implements vfs.FileDecriptionImpl.Seek. +func (fd *GenericDirectoryFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + fs := fd.filesystem().Impl().(*Filesystem) + fs.mu.Lock() + defer fs.mu.Unlock() + + switch whence { + case linux.SEEK_SET: + // Use offset as given. + case linux.SEEK_CUR: + offset += fd.off + default: + return 0, syserror.EINVAL + } + if offset < 0 { + return 0, syserror.EINVAL + } + fd.off = offset + return offset, nil +} + +// StatusFlags implements vfs.FileDescriptionImpl.StatusFlags. +func (fd *GenericDirectoryFD) StatusFlags(ctx context.Context) (uint32, error) { + return fd.flags, nil +} + +// SetStatusFlags implements vfs.FileDescriptionImpl.SetStatusFlags. +func (fd *GenericDirectoryFD) SetStatusFlags(ctx context.Context, flags uint32) error { + // None of the flags settable by fcntl(F_SETFL) are supported, so this is a + // no-op. + return nil +} + +// Stat implements vfs.FileDescriptionImpl.Stat. +func (fd *GenericDirectoryFD) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { + fs := fd.filesystem() + inode := fd.inode() + return inode.Stat(fs), nil +} + +// SetStat implements vfs.FileDescriptionImpl.SetStat. +func (fd *GenericDirectoryFD) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { + fs := fd.filesystem() + inode := fd.vfsfd.VirtualDentry().Dentry().Impl().(*Dentry).inode + return inode.SetStat(fs, opts) +} diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go new file mode 100644 index 000000000..db486b6c1 --- /dev/null +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -0,0 +1,691 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// This file implements vfs.FilesystemImpl for kernfs. + +package kernfs + +import ( + "fmt" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// stepExistingLocked resolves rp.Component() in parent directory vfsd. +// +// stepExistingLocked is loosely analogous to fs/namei.c:walk_component(). +// +// Preconditions: Filesystem.mu must be locked for at least reading. !rp.Done(). +// +// Postcondition: Caller must call fs.processDeferredDecRefs*. +func (fs *Filesystem) stepExistingLocked(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry) (*vfs.Dentry, error) { + d := vfsd.Impl().(*Dentry) + if !d.isDir() { + return nil, syserror.ENOTDIR + } + // Directory searchable? + if err := d.inode.CheckPermissions(rp.Credentials(), vfs.MayExec); err != nil { + return nil, err + } +afterSymlink: + d.dirMu.Lock() + nextVFSD, err := rp.ResolveComponent(vfsd) + d.dirMu.Unlock() + if err != nil { + return nil, err + } + if nextVFSD != nil { + // Cached dentry exists, revalidate. + next := nextVFSD.Impl().(*Dentry) + if !next.inode.Valid(ctx) { + d.dirMu.Lock() + rp.VirtualFilesystem().ForceDeleteDentry(nextVFSD) + d.dirMu.Unlock() + fs.deferDecRef(nextVFSD) // Reference from Lookup. + nextVFSD = nil + } + } + if nextVFSD == nil { + // Dentry isn't cached; it either doesn't exist or failed + // revalidation. Attempt to resolve it via Lookup. + name := rp.Component() + var err error + nextVFSD, err = d.inode.Lookup(ctx, name) + // Reference on nextVFSD dropped by a corresponding Valid. + if err != nil { + return nil, err + } + d.InsertChild(name, nextVFSD) + } + next := nextVFSD.Impl().(*Dentry) + + // Resolve any symlink at current path component. + if rp.ShouldFollowSymlink() && d.isSymlink() { + // TODO: VFS2 needs something extra for /proc/[pid]/fd/ "magic symlinks". + target, err := next.inode.Readlink(ctx) + if err != nil { + return nil, err + } + if err := rp.HandleSymlink(target); err != nil { + return nil, err + } + goto afterSymlink + + } + rp.Advance() + return nextVFSD, nil +} + +// walkExistingLocked resolves rp to an existing file. +// +// walkExistingLocked is loosely analogous to Linux's +// fs/namei.c:path_lookupat(). +// +// Preconditions: Filesystem.mu must be locked for at least reading. +// +// Postconditions: Caller must call fs.processDeferredDecRefs*. +func (fs *Filesystem) walkExistingLocked(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, Inode, error) { + vfsd := rp.Start() + for !rp.Done() { + var err error + vfsd, err = fs.stepExistingLocked(ctx, rp, vfsd) + if err != nil { + return nil, nil, err + } + } + d := vfsd.Impl().(*Dentry) + if rp.MustBeDir() && !d.isDir() { + return nil, nil, syserror.ENOTDIR + } + return vfsd, d.inode, nil +} + +// walkParentDirLocked resolves all but the last path component of rp to an +// existing directory. It does not check that the returned directory is +// searchable by the provider of rp. +// +// walkParentDirLocked is loosely analogous to Linux's +// fs/namei.c:path_parentat(). +// +// Preconditions: Filesystem.mu must be locked for at least reading. !rp.Done(). +// +// Postconditions: Caller must call fs.processDeferredDecRefs*. +func (fs *Filesystem) walkParentDirLocked(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, Inode, error) { + vfsd := rp.Start() + for !rp.Final() { + var err error + vfsd, err = fs.stepExistingLocked(ctx, rp, vfsd) + if err != nil { + return nil, nil, err + } + } + d := vfsd.Impl().(*Dentry) + if !d.isDir() { + return nil, nil, syserror.ENOTDIR + } + return vfsd, d.inode, nil +} + +// checkCreateLocked checks that a file named rp.Component() may be created in +// directory parentVFSD, then returns rp.Component(). +// +// Preconditions: Filesystem.mu must be locked for at least reading. parentInode +// == parentVFSD.Impl().(*Dentry).Inode. isDir(parentInode) == true. +func checkCreateLocked(rp *vfs.ResolvingPath, parentVFSD *vfs.Dentry, parentInode Inode) (string, error) { + if err := parentInode.CheckPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { + return "", err + } + pc := rp.Component() + if pc == "." || pc == ".." { + return "", syserror.EEXIST + } + childVFSD, err := rp.ResolveChild(parentVFSD, pc) + if err != nil { + return "", err + } + if childVFSD != nil { + return "", syserror.EEXIST + } + if parentVFSD.IsDisowned() { + return "", syserror.ENOENT + } + return pc, nil +} + +// checkDeleteLocked checks that the file represented by vfsd may be deleted. +// +// Preconditions: Filesystem.mu must be locked for at least reading. +func checkDeleteLocked(rp *vfs.ResolvingPath, vfsd *vfs.Dentry) error { + parentVFSD := vfsd.Parent() + if parentVFSD == nil { + return syserror.EBUSY + } + if parentVFSD.IsDisowned() { + return syserror.ENOENT + } + if err := parentVFSD.Impl().(*Dentry).inode.CheckPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { + return err + } + return nil +} + +// checkRenameLocked checks that a rename operation may be performed on the +// target dentry across the given set of parent directories. The target dentry +// may be nil. +// +// Precondition: isDir(dstInode) == true. +func checkRenameLocked(creds *auth.Credentials, src, dstDir *vfs.Dentry, dstInode Inode) error { + srcDir := src.Parent() + if srcDir == nil { + return syserror.EBUSY + } + if srcDir.IsDisowned() { + return syserror.ENOENT + } + if dstDir.IsDisowned() { + return syserror.ENOENT + } + // Check for creation permissions on dst dir. + if err := dstInode.CheckPermissions(creds, vfs.MayWrite|vfs.MayExec); err != nil { + return err + } + + return nil +} + +// Release implements vfs.FilesystemImpl.Release. +func (fs *Filesystem) Release() { +} + +// Sync implements vfs.FilesystemImpl.Sync. +func (fs *Filesystem) Sync(ctx context.Context) error { + // All filesystem state is in-memory. + return nil +} + +// GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. +func (fs *Filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { + fs.mu.RLock() + defer fs.processDeferredDecRefs() + defer fs.mu.RUnlock() + vfsd, inode, err := fs.walkExistingLocked(ctx, rp) + if err != nil { + return nil, err + } + + if opts.CheckSearchable { + d := vfsd.Impl().(*Dentry) + if !d.isDir() { + return nil, syserror.ENOTDIR + } + if err := inode.CheckPermissions(rp.Credentials(), vfs.MayExec); err != nil { + return nil, err + } + } + vfsd.IncRef() // Ownership transferred to caller. + return vfsd, nil +} + +// LinkAt implements vfs.FilesystemImpl.LinkAt. +func (fs *Filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry) error { + if rp.Done() { + return syserror.EEXIST + } + fs.mu.Lock() + defer fs.mu.Unlock() + parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) + fs.processDeferredDecRefsLocked() + if err != nil { + return err + } + pc, err := checkCreateLocked(rp, parentVFSD, parentInode) + if err != nil { + return err + } + if rp.Mount() != vd.Mount() { + return syserror.EXDEV + } + if err := rp.Mount().CheckBeginWrite(); err != nil { + return err + } + defer rp.Mount().EndWrite() + + d := vd.Dentry().Impl().(*Dentry) + if d.isDir() { + return syserror.EPERM + } + + child, err := parentInode.NewLink(ctx, pc, d.inode) + if err != nil { + return err + } + parentVFSD.Impl().(*Dentry).InsertChild(pc, child) + return nil +} + +// MkdirAt implements vfs.FilesystemImpl.MkdirAt. +func (fs *Filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error { + if rp.Done() { + return syserror.EEXIST + } + fs.mu.Lock() + defer fs.mu.Unlock() + parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) + fs.processDeferredDecRefsLocked() + if err != nil { + return err + } + pc, err := checkCreateLocked(rp, parentVFSD, parentInode) + if err != nil { + return err + } + if err := rp.Mount().CheckBeginWrite(); err != nil { + return err + } + defer rp.Mount().EndWrite() + child, err := parentInode.NewDir(ctx, pc, opts) + if err != nil { + return err + } + parentVFSD.Impl().(*Dentry).InsertChild(pc, child) + return nil +} + +// MknodAt implements vfs.FilesystemImpl.MknodAt. +func (fs *Filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MknodOptions) error { + if rp.Done() { + return syserror.EEXIST + } + fs.mu.Lock() + defer fs.mu.Unlock() + parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) + fs.processDeferredDecRefsLocked() + if err != nil { + return err + } + pc, err := checkCreateLocked(rp, parentVFSD, parentInode) + if err != nil { + return err + } + if err := rp.Mount().CheckBeginWrite(); err != nil { + return err + } + defer rp.Mount().EndWrite() + new, err := parentInode.NewNode(ctx, pc, opts) + if err != nil { + return err + } + parentVFSD.Impl().(*Dentry).InsertChild(pc, new) + return nil +} + +// OpenAt implements vfs.FilesystemImpl.OpenAt. +func (fs *Filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + // Filter out flags that are not supported by kernfs. O_DIRECTORY and + // O_NOFOLLOW have no effect here (they're handled by VFS by setting + // appropriate bits in rp), but are returned by + // FileDescriptionImpl.StatusFlags(). + opts.Flags &= linux.O_ACCMODE | linux.O_CREAT | linux.O_EXCL | linux.O_TRUNC | linux.O_DIRECTORY | linux.O_NOFOLLOW + ats := vfs.AccessTypesForOpenFlags(opts.Flags) + + // Do not create new file. + if opts.Flags&linux.O_CREAT == 0 { + fs.mu.RLock() + defer fs.processDeferredDecRefs() + defer fs.mu.RUnlock() + vfsd, inode, err := fs.walkExistingLocked(ctx, rp) + if err != nil { + return nil, err + } + if err := inode.CheckPermissions(rp.Credentials(), ats); err != nil { + return nil, err + } + return inode.Open(rp, vfsd, opts.Flags) + } + + // May create new file. + mustCreate := opts.Flags&linux.O_EXCL != 0 + vfsd := rp.Start() + inode := vfsd.Impl().(*Dentry).inode + fs.mu.Lock() + defer fs.mu.Unlock() + if rp.Done() { + if rp.MustBeDir() { + return nil, syserror.EISDIR + } + if mustCreate { + return nil, syserror.EEXIST + } + if err := inode.CheckPermissions(rp.Credentials(), ats); err != nil { + return nil, err + } + return inode.Open(rp, vfsd, opts.Flags) + } +afterTrailingSymlink: + parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) + fs.processDeferredDecRefsLocked() + if err != nil { + return nil, err + } + // Check for search permission in the parent directory. + if err := parentInode.CheckPermissions(rp.Credentials(), vfs.MayExec); err != nil { + return nil, err + } + // Reject attempts to open directories with O_CREAT. + if rp.MustBeDir() { + return nil, syserror.EISDIR + } + pc := rp.Component() + if pc == "." || pc == ".." { + return nil, syserror.EISDIR + } + // Determine whether or not we need to create a file. + childVFSD, err := rp.ResolveChild(parentVFSD, pc) + if err != nil { + return nil, err + } + if childVFSD == nil { + // Already checked for searchability above; now check for writability. + if err := parentInode.CheckPermissions(rp.Credentials(), vfs.MayWrite); err != nil { + return nil, err + } + if err := rp.Mount().CheckBeginWrite(); err != nil { + return nil, err + } + defer rp.Mount().EndWrite() + // Create and open the child. + child, err := parentInode.NewFile(ctx, pc, opts) + if err != nil { + return nil, err + } + parentVFSD.Impl().(*Dentry).InsertChild(pc, child) + return child.Impl().(*Dentry).inode.Open(rp, child, opts.Flags) + } + // Open existing file or follow symlink. + if mustCreate { + return nil, syserror.EEXIST + } + childDentry := childVFSD.Impl().(*Dentry) + childInode := childDentry.inode + if rp.ShouldFollowSymlink() { + if childDentry.isSymlink() { + target, err := childInode.Readlink(ctx) + if err != nil { + return nil, err + } + if err := rp.HandleSymlink(target); err != nil { + return nil, err + } + // rp.Final() may no longer be true since we now need to resolve the + // symlink target. + goto afterTrailingSymlink + } + } + if err := childInode.CheckPermissions(rp.Credentials(), ats); err != nil { + return nil, err + } + return childInode.Open(rp, childVFSD, opts.Flags) +} + +// ReadlinkAt implements vfs.FilesystemImpl.ReadlinkAt. +func (fs *Filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { + fs.mu.RLock() + d, inode, err := fs.walkExistingLocked(ctx, rp) + fs.mu.RUnlock() + fs.processDeferredDecRefs() + if err != nil { + return "", err + } + if !d.Impl().(*Dentry).isSymlink() { + return "", syserror.EINVAL + } + return inode.Readlink(ctx) +} + +// RenameAt implements vfs.FilesystemImpl.RenameAt. +func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry, opts vfs.RenameOptions) error { + noReplace := opts.Flags&linux.RENAME_NOREPLACE != 0 + exchange := opts.Flags&linux.RENAME_EXCHANGE != 0 + whiteout := opts.Flags&linux.RENAME_WHITEOUT != 0 + if exchange && (noReplace || whiteout) { + // Can't specify RENAME_NOREPLACE or RENAME_WHITEOUT with RENAME_EXCHANGE. + return syserror.EINVAL + } + if exchange || whiteout { + // Exchange and Whiteout flags are not supported on kernfs. + return syserror.EINVAL + } + + fs.mu.Lock() + defer fs.mu.Lock() + + mnt := rp.Mount() + if mnt != vd.Mount() { + return syserror.EXDEV + } + + if err := mnt.CheckBeginWrite(); err != nil { + return err + } + defer mnt.EndWrite() + + dstDirVFSD, dstDirInode, err := fs.walkParentDirLocked(ctx, rp) + fs.processDeferredDecRefsLocked() + if err != nil { + return err + } + + srcVFSD := vd.Dentry() + srcDirVFSD := srcVFSD.Parent() + + // Can we remove the src dentry? + if err := checkDeleteLocked(rp, srcVFSD); err != nil { + return err + } + + // Can we create the dst dentry? + var dstVFSD *vfs.Dentry + pc, err := checkCreateLocked(rp, dstDirVFSD, dstDirInode) + switch err { + case nil: + // Ok, continue with rename as replacement. + case syserror.EEXIST: + if noReplace { + // Won't overwrite existing node since RENAME_NOREPLACE was requested. + return syserror.EEXIST + } + dstVFSD, err = rp.ResolveChild(dstDirVFSD, pc) + if err != nil { + panic(fmt.Sprintf("Child %q for parent Dentry %+v disappeared inside atomic section?", pc, dstDirVFSD)) + } + default: + return err + } + + mntns := vfs.MountNamespaceFromContext(ctx) + virtfs := rp.VirtualFilesystem() + + srcDirDentry := srcDirVFSD.Impl().(*Dentry) + dstDirDentry := dstDirVFSD.Impl().(*Dentry) + + // We can't deadlock here due to lock ordering because we're protected from + // concurrent renames by fs.mu held for writing. + srcDirDentry.dirMu.Lock() + defer srcDirDentry.dirMu.Unlock() + dstDirDentry.dirMu.Lock() + defer dstDirDentry.dirMu.Unlock() + + if err := virtfs.PrepareRenameDentry(mntns, srcVFSD, dstVFSD); err != nil { + return err + } + srcDirInode := srcDirDentry.inode + replaced, err := srcDirInode.Rename(ctx, srcVFSD.Name(), pc, srcVFSD, dstDirVFSD) + if err != nil { + virtfs.AbortRenameDentry(srcVFSD, dstVFSD) + return err + } + virtfs.CommitRenameReplaceDentry(srcVFSD, dstDirVFSD, pc, replaced) + return nil +} + +// RmdirAt implements vfs.FilesystemImpl.RmdirAt. +func (fs *Filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error { + fs.mu.Lock() + defer fs.mu.Unlock() + vfsd, inode, err := fs.walkExistingLocked(ctx, rp) + fs.processDeferredDecRefsLocked() + if err != nil { + return err + } + if err := rp.Mount().CheckBeginWrite(); err != nil { + return err + } + defer rp.Mount().EndWrite() + if err := checkDeleteLocked(rp, vfsd); err != nil { + return err + } + if !vfsd.Impl().(*Dentry).isDir() { + return syserror.ENOTDIR + } + if inode.HasChildren() { + return syserror.ENOTEMPTY + } + virtfs := rp.VirtualFilesystem() + parentDentry := vfsd.Parent().Impl().(*Dentry) + parentDentry.dirMu.Lock() + defer parentDentry.dirMu.Unlock() + if err := virtfs.PrepareDeleteDentry(vfs.MountNamespaceFromContext(ctx), vfsd); err != nil { + return err + } + if err := parentDentry.inode.RmDir(ctx, rp.Component(), vfsd); err != nil { + virtfs.AbortDeleteDentry(vfsd) + return err + } + virtfs.CommitDeleteDentry(vfsd) + return nil +} + +// SetStatAt implements vfs.FilesystemImpl.SetStatAt. +func (fs *Filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { + fs.mu.RLock() + _, inode, err := fs.walkExistingLocked(ctx, rp) + fs.mu.RUnlock() + fs.processDeferredDecRefs() + if err != nil { + return err + } + if opts.Stat.Mask == 0 { + return nil + } + return inode.SetStat(fs.VFSFilesystem(), opts) +} + +// StatAt implements vfs.FilesystemImpl.StatAt. +func (fs *Filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { + fs.mu.RLock() + _, inode, err := fs.walkExistingLocked(ctx, rp) + fs.mu.RUnlock() + fs.processDeferredDecRefs() + if err != nil { + return linux.Statx{}, err + } + return inode.Stat(fs.VFSFilesystem()), nil +} + +// StatFSAt implements vfs.FilesystemImpl.StatFSAt. +func (fs *Filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { + fs.mu.RLock() + _, _, err := fs.walkExistingLocked(ctx, rp) + fs.mu.RUnlock() + fs.processDeferredDecRefs() + if err != nil { + return linux.Statfs{}, err + } + // TODO: actually implement statfs + return linux.Statfs{}, syserror.ENOSYS +} + +// SymlinkAt implements vfs.FilesystemImpl.SymlinkAt. +func (fs *Filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, target string) error { + if rp.Done() { + return syserror.EEXIST + } + fs.mu.Lock() + defer fs.mu.Unlock() + parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) + fs.processDeferredDecRefsLocked() + if err != nil { + return err + } + pc, err := checkCreateLocked(rp, parentVFSD, parentInode) + if err != nil { + return err + } + if err := rp.Mount().CheckBeginWrite(); err != nil { + return err + } + defer rp.Mount().EndWrite() + child, err := parentInode.NewSymlink(ctx, pc, target) + if err != nil { + return err + } + parentVFSD.Impl().(*Dentry).InsertChild(pc, child) + return nil +} + +// UnlinkAt implements vfs.FilesystemImpl.UnlinkAt. +func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error { + fs.mu.Lock() + defer fs.mu.Unlock() + vfsd, _, err := fs.walkExistingLocked(ctx, rp) + fs.processDeferredDecRefsLocked() + if err != nil { + return err + } + if err := rp.Mount().CheckBeginWrite(); err != nil { + return err + } + defer rp.Mount().EndWrite() + if err := checkDeleteLocked(rp, vfsd); err != nil { + return err + } + if vfsd.Impl().(*Dentry).isDir() { + return syserror.EISDIR + } + virtfs := rp.VirtualFilesystem() + parentDentry := vfsd.Parent().Impl().(*Dentry) + parentDentry.dirMu.Lock() + defer parentDentry.dirMu.Unlock() + if err := virtfs.PrepareDeleteDentry(vfs.MountNamespaceFromContext(ctx), vfsd); err != nil { + return err + } + if err := parentDentry.inode.Unlink(ctx, rp.Component(), vfsd); err != nil { + virtfs.AbortDeleteDentry(vfsd) + return err + } + virtfs.CommitDeleteDentry(vfsd) + return nil +} + +// PrependPath implements vfs.FilesystemImpl.PrependPath. +func (fs *Filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { + fs.mu.RLock() + defer fs.mu.RUnlock() + return vfs.GenericPrependPath(vfsroot, vd, b) +} diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go new file mode 100644 index 000000000..7b45b702a --- /dev/null +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -0,0 +1,492 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package kernfs + +import ( + "fmt" + "sync" + "sync/atomic" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/refs" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// InodeNoopRefCount partially implements the Inode interface, specifically the +// inodeRefs sub interface. InodeNoopRefCount implements a simple reference +// count for inodes, performing no extra actions when references are obtained or +// released. This is suitable for simple file inodes that don't reference any +// resources. +type InodeNoopRefCount struct { +} + +// IncRef implements Inode.IncRef. +func (n *InodeNoopRefCount) IncRef() { +} + +// DecRef implements Inode.DecRef. +func (n *InodeNoopRefCount) DecRef() { +} + +// TryIncRef implements Inode.TryIncRef. +func (n *InodeNoopRefCount) TryIncRef() bool { + return true +} + +// Destroy implements Inode.Destroy. +func (n *InodeNoopRefCount) Destroy() { +} + +// InodeDirectoryNoNewChildren partially implements the Inode interface. +// InodeDirectoryNoNewChildren represents a directory inode which does not +// support creation of new children. +type InodeDirectoryNoNewChildren struct{} + +// NewFile implements Inode.NewFile. +func (*InodeDirectoryNoNewChildren) NewFile(context.Context, string, vfs.OpenOptions) (*vfs.Dentry, error) { + return nil, syserror.EPERM +} + +// NewDir implements Inode.NewDir. +func (*InodeDirectoryNoNewChildren) NewDir(context.Context, string, vfs.MkdirOptions) (*vfs.Dentry, error) { + return nil, syserror.EPERM +} + +// NewLink implements Inode.NewLink. +func (*InodeDirectoryNoNewChildren) NewLink(context.Context, string, Inode) (*vfs.Dentry, error) { + return nil, syserror.EPERM +} + +// NewSymlink implements Inode.NewSymlink. +func (*InodeDirectoryNoNewChildren) NewSymlink(context.Context, string, string) (*vfs.Dentry, error) { + return nil, syserror.EPERM +} + +// NewNode implements Inode.NewNode. +func (*InodeDirectoryNoNewChildren) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, error) { + return nil, syserror.EPERM +} + +// InodeNotDirectory partially implements the Inode interface, specifically the +// inodeDirectory and inodeDynamicDirectory sub interfaces. Inodes that do not +// represent directories can embed this to provide no-op implementations for +// directory-related functions. +type InodeNotDirectory struct { +} + +// HasChildren implements Inode.HasChildren. +func (*InodeNotDirectory) HasChildren() bool { + return false +} + +// NewFile implements Inode.NewFile. +func (*InodeNotDirectory) NewFile(context.Context, string, vfs.OpenOptions) (*vfs.Dentry, error) { + panic("NewFile called on non-directory inode") +} + +// NewDir implements Inode.NewDir. +func (*InodeNotDirectory) NewDir(context.Context, string, vfs.MkdirOptions) (*vfs.Dentry, error) { + panic("NewDir called on non-directory inode") +} + +// NewLink implements Inode.NewLinkink. +func (*InodeNotDirectory) NewLink(context.Context, string, Inode) (*vfs.Dentry, error) { + panic("NewLink called on non-directory inode") +} + +// NewSymlink implements Inode.NewSymlink. +func (*InodeNotDirectory) NewSymlink(context.Context, string, string) (*vfs.Dentry, error) { + panic("NewSymlink called on non-directory inode") +} + +// NewNode implements Inode.NewNode. +func (*InodeNotDirectory) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, error) { + panic("NewNode called on non-directory inode") +} + +// Unlink implements Inode.Unlink. +func (*InodeNotDirectory) Unlink(context.Context, string, *vfs.Dentry) error { + panic("Unlink called on non-directory inode") +} + +// RmDir implements Inode.RmDir. +func (*InodeNotDirectory) RmDir(context.Context, string, *vfs.Dentry) error { + panic("RmDir called on non-directory inode") +} + +// Rename implements Inode.Rename. +func (*InodeNotDirectory) Rename(context.Context, string, string, *vfs.Dentry, *vfs.Dentry) (*vfs.Dentry, error) { + panic("Rename called on non-directory inode") +} + +// Lookup implements Inode.Lookup. +func (*InodeNotDirectory) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { + panic("Lookup called on non-directory inode") +} + +// Valid implements Inode.Valid. +func (*InodeNotDirectory) Valid(context.Context) bool { + return true +} + +// InodeNoDynamicLookup partially implements the Inode interface, specifically +// the inodeDynamicLookup sub interface. Directory inodes that do not support +// dymanic entries (i.e. entries that are not "hashed" into the +// vfs.Dentry.children) can embed this to provide no-op implementations for +// functions related to dynamic entries. +type InodeNoDynamicLookup struct{} + +// Lookup implements Inode.Lookup. +func (*InodeNoDynamicLookup) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { + return nil, syserror.ENOENT +} + +// Valid implements Inode.Valid. +func (*InodeNoDynamicLookup) Valid(ctx context.Context) bool { + return true +} + +// InodeNotSymlink partially implements the Inode interface, specifically the +// inodeSymlink sub interface. All inodes that are not symlinks may embed this +// to return the appropriate errors from symlink-related functions. +type InodeNotSymlink struct{} + +// Readlink implements Inode.Readlink. +func (*InodeNotSymlink) Readlink(context.Context) (string, error) { + return "", syserror.EINVAL +} + +// InodeAttrs partially implements the Inode interface, specifically the +// inodeMetadata sub interface. InodeAttrs provides functionality related to +// inode attributes. +// +// Must be initialized by Init prior to first use. +type InodeAttrs struct { + ino uint64 + mode uint32 + uid uint32 + gid uint32 + nlink uint32 +} + +// Init initializes this InodeAttrs. +func (a *InodeAttrs) Init(creds *auth.Credentials, ino uint64, mode linux.FileMode) { + if mode.FileType() == 0 { + panic(fmt.Sprintf("No file type specified in 'mode' for InodeAttrs.Init(): mode=0%o", mode)) + } + + nlink := uint32(1) + if mode.FileType() == linux.ModeDirectory { + nlink = 2 + } + atomic.StoreUint64(&a.ino, ino) + atomic.StoreUint32(&a.mode, uint32(mode)) + atomic.StoreUint32(&a.uid, uint32(creds.EffectiveKUID)) + atomic.StoreUint32(&a.gid, uint32(creds.EffectiveKGID)) + atomic.StoreUint32(&a.nlink, nlink) +} + +// Mode implements Inode.Mode. +func (a *InodeAttrs) Mode() linux.FileMode { + return linux.FileMode(atomic.LoadUint32(&a.mode)) +} + +// Stat partially implements Inode.Stat. Note that this function doesn't provide +// all the stat fields, and the embedder should consider extending the result +// with filesystem-specific fields. +func (a *InodeAttrs) Stat(*vfs.Filesystem) linux.Statx { + var stat linux.Statx + stat.Mask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_UID | linux.STATX_GID | linux.STATX_INO | linux.STATX_NLINK + stat.Ino = atomic.LoadUint64(&a.ino) + stat.Mode = uint16(a.Mode()) + stat.UID = atomic.LoadUint32(&a.uid) + stat.GID = atomic.LoadUint32(&a.gid) + stat.Nlink = atomic.LoadUint32(&a.nlink) + + // TODO: Implement other stat fields like timestamps. + + return stat +} + +// SetStat implements Inode.SetStat. +func (a *InodeAttrs) SetStat(_ *vfs.Filesystem, opts vfs.SetStatOptions) error { + stat := opts.Stat + if stat.Mask&linux.STATX_MODE != 0 { + for { + old := atomic.LoadUint32(&a.mode) + new := old | uint32(stat.Mode & ^uint16(linux.S_IFMT)) + if swapped := atomic.CompareAndSwapUint32(&a.mode, old, new); swapped { + break + } + } + } + + if stat.Mask&linux.STATX_UID != 0 { + atomic.StoreUint32(&a.uid, stat.UID) + } + if stat.Mask&linux.STATX_GID != 0 { + atomic.StoreUint32(&a.gid, stat.GID) + } + + // Note that not all fields are modifiable. For example, the file type and + // inode numbers are immutable after node creation. + + // TODO: Implement other stat fields like timestamps. + + return nil +} + +// CheckPermissions implements Inode.CheckPermissions. +func (a *InodeAttrs) CheckPermissions(creds *auth.Credentials, ats vfs.AccessTypes) error { + mode := a.Mode() + return vfs.GenericCheckPermissions( + creds, + ats, + mode.FileType() == linux.ModeDirectory, + uint16(mode), + auth.KUID(atomic.LoadUint32(&a.uid)), + auth.KGID(atomic.LoadUint32(&a.gid)), + ) +} + +// IncLinks implements Inode.IncLinks. +func (a *InodeAttrs) IncLinks(n uint32) { + if atomic.AddUint32(&a.nlink, n) <= n { + panic("InodeLink.IncLinks called with no existing links") + } +} + +// DecLinks implements Inode.DecLinks. +func (a *InodeAttrs) DecLinks() { + if nlink := atomic.AddUint32(&a.nlink, ^uint32(0)); nlink == ^uint32(0) { + // Negative overflow + panic("Inode.DecLinks called at 0 links") + } +} + +type slot struct { + Name string + Dentry *vfs.Dentry + slotEntry +} + +// OrderedChildrenOptions contains initialization options for OrderedChildren. +type OrderedChildrenOptions struct { + // Writable indicates whether vfs.FilesystemImpl methods implemented by + // OrderedChildren may modify the tracked children. This applies to + // operations related to rename, unlink and rmdir. If an OrderedChildren is + // not writable, these operations all fail with EPERM. + Writable bool +} + +// OrderedChildren partially implements the Inode interface. OrderedChildren can +// be embedded in directory inodes to keep track of the children in the +// directory, and can then be used to implement a generic directory FD -- see +// GenericDirectoryFD. OrderedChildren is not compatible with dynamic +// directories. +// +// Must be initialize with Init before first use. +type OrderedChildren struct { + refs.AtomicRefCount + + // Can children be modified by user syscalls? It set to false, interface + // methods that would modify the children return EPERM. Immutable. + writable bool + + mu sync.RWMutex + order slotList + set map[string]*slot +} + +// Init initializes an OrderedChildren. +func (o *OrderedChildren) Init(opts OrderedChildrenOptions) { + o.writable = opts.Writable + o.set = make(map[string]*slot) +} + +// DecRef implements Inode.DecRef. +func (o *OrderedChildren) DecRef() { + o.AtomicRefCount.DecRefWithDestructor(o.Destroy) +} + +// Destroy cleans up resources referenced by this OrderedChildren. +func (o *OrderedChildren) Destroy() { + o.mu.Lock() + defer o.mu.Unlock() + o.order.Reset() + o.set = nil +} + +// Populate inserts children into this OrderedChildren, and d's dentry +// cache. Populate returns the number of directories inserted, which the caller +// may use to update the link count for the parent directory. +// +// Precondition: d.Impl() must be a kernfs Dentry. d must represent a directory +// inode. children must not contain any conflicting entries already in o. +func (o *OrderedChildren) Populate(d *Dentry, children map[string]*Dentry) uint32 { + var links uint32 + for name, child := range children { + if child.isDir() { + links++ + } + if err := o.Insert(name, child.VFSDentry()); err != nil { + panic(fmt.Sprintf("Collision when attempting to insert child %q (%+v) into %+v", name, child, d)) + } + d.InsertChild(name, child.VFSDentry()) + } + return links +} + +// HasChildren implements Inode.HasChildren. +func (o *OrderedChildren) HasChildren() bool { + o.mu.RLock() + defer o.mu.RUnlock() + return len(o.set) > 0 +} + +// Insert inserts child into o. This ignores the writability of o, as this is +// not part of the vfs.FilesystemImpl interface, and is a lower-level operation. +func (o *OrderedChildren) Insert(name string, child *vfs.Dentry) error { + o.mu.Lock() + defer o.mu.Unlock() + if _, ok := o.set[name]; ok { + return syserror.EEXIST + } + s := &slot{ + Name: name, + Dentry: child, + } + o.order.PushBack(s) + o.set[name] = s + return nil +} + +// Precondition: caller must hold o.mu for writing. +func (o *OrderedChildren) removeLocked(name string) { + if s, ok := o.set[name]; ok { + delete(o.set, name) + o.order.Remove(s) + } +} + +// Precondition: caller must hold o.mu for writing. +func (o *OrderedChildren) replaceChildLocked(name string, new *vfs.Dentry) *vfs.Dentry { + if s, ok := o.set[name]; ok { + // Existing slot with given name, simply replace the dentry. + var old *vfs.Dentry + old, s.Dentry = s.Dentry, new + return old + } + + // No existing slot with given name, create and hash new slot. + s := &slot{ + Name: name, + Dentry: new, + } + o.order.PushBack(s) + o.set[name] = s + return nil +} + +// Precondition: caller must hold o.mu for reading or writing. +func (o *OrderedChildren) checkExistingLocked(name string, child *vfs.Dentry) error { + s, ok := o.set[name] + if !ok { + return syserror.ENOENT + } + if s.Dentry != child { + panic(fmt.Sprintf("Dentry hashed into inode doesn't match what vfs thinks! OrderedChild: %+v, vfs: %+v", s.Dentry, child)) + } + return nil +} + +// Unlink implements Inode.Unlink. +func (o *OrderedChildren) Unlink(ctx context.Context, name string, child *vfs.Dentry) error { + if !o.writable { + return syserror.EPERM + } + o.mu.Lock() + defer o.mu.Unlock() + if err := o.checkExistingLocked(name, child); err != nil { + return err + } + o.removeLocked(name) + return nil +} + +// Rmdir implements Inode.Rmdir. +func (o *OrderedChildren) RmDir(ctx context.Context, name string, child *vfs.Dentry) error { + // We're not responsible for checking that child is a directory, that it's + // empty, or updating any link counts; so this is the same as unlink. + return o.Unlink(ctx, name, child) +} + +type renameAcrossDifferentImplementationsError struct{} + +func (renameAcrossDifferentImplementationsError) Error() string { + return "rename across inodes with different implementations" +} + +// Rename implements Inode.Rename. +// +// Precondition: Rename may only be called across two directory inodes with +// identical implementations of Rename. Practically, this means filesystems that +// implement Rename by embedding OrderedChildren for any directory +// implementation must use OrderedChildren for all directory implementations +// that will support Rename. +// +// Postcondition: reference on any replaced dentry transferred to caller. +func (o *OrderedChildren) Rename(ctx context.Context, oldname, newname string, child, dstDir *vfs.Dentry) (*vfs.Dentry, error) { + dst, ok := dstDir.Impl().(*Dentry).inode.(interface{}).(*OrderedChildren) + if !ok { + return nil, renameAcrossDifferentImplementationsError{} + } + if !o.writable || !dst.writable { + return nil, syserror.EPERM + } + // Note: There's a potential deadlock below if concurrent calls to Rename + // refer to the same src and dst directories in reverse. We avoid any + // ordering issues because the caller is required to serialize concurrent + // calls to Rename in accordance with the interface declaration. + o.mu.Lock() + defer o.mu.Unlock() + if dst != o { + dst.mu.Lock() + defer dst.mu.Unlock() + } + if err := o.checkExistingLocked(oldname, child); err != nil { + return nil, err + } + replaced := dst.replaceChildLocked(newname, child) + return replaced, nil +} + +// nthLocked returns an iterator to the nth child tracked by this object. The +// iterator is valid until the caller releases o.mu. Returns nil if the +// requested index falls out of bounds. +// +// Preconditon: Caller must hold o.mu for reading. +func (o *OrderedChildren) nthLocked(i int64) *slot { + for it := o.order.Front(); it != nil && i >= 0; it = it.Next() { + if i == 0 { + return it + } + i-- + } + return nil +} diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go new file mode 100644 index 000000000..bb01c3d01 --- /dev/null +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -0,0 +1,405 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package kernfs provides the tools to implement inode-based filesystems. +// Kernfs has two main features: +// +// 1. The Inode interface, which maps VFS2's path-based filesystem operations to +// specific filesystem nodes. Kernfs uses the Inode interface to provide a +// blanket implementation for the vfs.FilesystemImpl. Kernfs also serves as +// the synchronization mechanism for all filesystem operations by holding a +// filesystem-wide lock across all operations. +// +// 2. Various utility types which provide generic implementations for various +// parts of the Inode and vfs.FileDescription interfaces. Client filesystems +// based on kernfs can embed the appropriate set of these to avoid having to +// reimplement common filesystem operations. See inode_impl_util.go and +// fd_impl_util.go. +// +// Reference Model: +// +// Kernfs dentries represents named pointers to inodes. Dentries and inode have +// independent lifetimes and reference counts. A child dentry unconditionally +// holds a reference on its parent directory's dentry. A dentry also holds a +// reference on the inode it points to. Multiple dentries can point to the same +// inode (for example, in the case of hardlinks). File descriptors hold a +// reference to the dentry they're opened on. +// +// Dentries are guaranteed to exist while holding Filesystem.mu for +// reading. Dropping dentries require holding Filesystem.mu for writing. To +// queue dentries for destruction from a read critical section, see +// Filesystem.deferDecRef. +// +// Lock ordering: +// +// kernfs.Filesystem.mu +// kernfs.Dentry.dirMu +// vfs.VirtualFilesystem.mountMu +// vfs.Dentry.mu +// kernfs.Filesystem.droppedDentriesMu +// (inode implementation locks, if any) +package kernfs + +import ( + "fmt" + "sync" + "sync/atomic" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/refs" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" +) + +// FilesystemType implements vfs.FilesystemType. +type FilesystemType struct{} + +// Filesystem mostly implements vfs.FilesystemImpl for a generic in-memory +// filesystem. Concrete implementations are expected to embed this in their own +// Filesystem type. +type Filesystem struct { + vfsfs vfs.Filesystem + + droppedDentriesMu sync.Mutex + + // droppedDentries is a list of dentries waiting to be DecRef()ed. This is + // used to defer dentry destruction until mu can be acquired for + // writing. Protected by droppedDentriesMu. + droppedDentries []*vfs.Dentry + + // mu synchronizes the lifetime of Dentries on this filesystem. Holding it + // for reading guarantees continued existence of any resolved dentries, but + // the dentry tree may be modified. + // + // Kernfs dentries can only be DecRef()ed while holding mu for writing. For + // example: + // + // fs.mu.Lock() + // defer fs.mu.Unlock() + // ... + // dentry1.DecRef() + // defer dentry2.DecRef() // Ok, will run before Unlock. + // + // If discarding dentries in a read context, use Filesystem.deferDecRef. For + // example: + // + // fs.mu.RLock() + // fs.mu.processDeferredDecRefs() + // defer fs.mu.RUnlock() + // ... + // fs.deferDecRef(dentry) + mu sync.RWMutex + + // nextInoMinusOne is used to to allocate inode numbers on this + // filesystem. Must be accessed by atomic operations. + nextInoMinusOne uint64 +} + +// deferDecRef defers dropping a dentry ref until the next call to +// processDeferredDecRefs{,Locked}. See comment on Filesystem.mu. +// +// Precondition: d must not already be pending destruction. +func (fs *Filesystem) deferDecRef(d *vfs.Dentry) { + fs.droppedDentriesMu.Lock() + fs.droppedDentries = append(fs.droppedDentries, d) + fs.droppedDentriesMu.Unlock() +} + +// processDeferredDecRefs calls vfs.Dentry.DecRef on all dentries in the +// droppedDentries list. See comment on Filesystem.mu. +func (fs *Filesystem) processDeferredDecRefs() { + fs.mu.Lock() + fs.processDeferredDecRefsLocked() + fs.mu.Unlock() +} + +// Precondition: fs.mu must be held for writing. +func (fs *Filesystem) processDeferredDecRefsLocked() { + fs.droppedDentriesMu.Lock() + for _, d := range fs.droppedDentries { + d.DecRef() + } + fs.droppedDentries = fs.droppedDentries[:0] // Keep slice memory for reuse. + fs.droppedDentriesMu.Unlock() +} + +// Init initializes a kernfs filesystem. This should be called from during +// vfs.FilesystemType.NewFilesystem for the concrete filesystem embedding +// kernfs. +func (fs *Filesystem) Init(vfsObj *vfs.VirtualFilesystem) { + fs.vfsfs.Init(vfsObj, fs) +} + +// VFSFilesystem returns the generic vfs filesystem object. +func (fs *Filesystem) VFSFilesystem() *vfs.Filesystem { + return &fs.vfsfs +} + +// NextIno allocates a new inode number on this filesystem. +func (fs *Filesystem) NextIno() uint64 { + return atomic.AddUint64(&fs.nextInoMinusOne, 1) +} + +// These consts are used in the Dentry.flags field. +const ( + // Dentry points to a directory inode. + dflagsIsDir = 1 << iota + + // Dentry points to a symlink inode. + dflagsIsSymlink +) + +// Dentry implements vfs.DentryImpl. +// +// A kernfs dentry is similar to a dentry in a traditional filesystem: it's a +// named reference to an inode. A dentry generally lives as long as it's part of +// a mounted filesystem tree. Kernfs doesn't cache dentries once all references +// to them are removed. Dentries hold a single reference to the inode they point +// to, and child dentries hold a reference on their parent. +// +// Must be initialized by Init prior to first use. +type Dentry struct { + refs.AtomicRefCount + + vfsd vfs.Dentry + inode Inode + + refs uint64 + + // flags caches useful information about the dentry from the inode. See the + // dflags* consts above. Must be accessed by atomic ops. + flags uint32 + + // dirMu protects vfsd.children for directory dentries. + dirMu sync.Mutex +} + +// Init initializes this dentry. +// +// Precondition: Caller must hold a reference on inode. +// +// Postcondition: Caller's reference on inode is transferred to the dentry. +func (d *Dentry) Init(inode Inode) { + d.vfsd.Init(d) + d.inode = inode + ftype := inode.Mode().FileType() + if ftype == linux.ModeDirectory { + d.flags |= dflagsIsDir + } + if ftype == linux.ModeSymlink { + d.flags |= dflagsIsSymlink + } +} + +// VFSDentry returns the generic vfs dentry for this kernfs dentry. +func (d *Dentry) VFSDentry() *vfs.Dentry { + return &d.vfsd +} + +// isDir checks whether the dentry points to a directory inode. +func (d *Dentry) isDir() bool { + return atomic.LoadUint32(&d.flags)&dflagsIsDir != 0 +} + +// isSymlink checks whether the dentry points to a symlink inode. +func (d *Dentry) isSymlink() bool { + return atomic.LoadUint32(&d.flags)&dflagsIsSymlink != 0 +} + +// DecRef implements vfs.DentryImpl.DecRef. +func (d *Dentry) DecRef() { + d.AtomicRefCount.DecRefWithDestructor(d.destroy) +} + +// Precondition: Dentry must be removed from VFS' dentry cache. +func (d *Dentry) destroy() { + d.inode.DecRef() // IncRef from Init. + d.inode = nil + if parent := d.vfsd.Parent(); parent != nil { + parent.DecRef() // IncRef from Dentry.InsertChild. + } +} + +// InsertChild inserts child into the vfs dentry cache with the given name under +// this dentry. This does not update the directory inode, so calling this on +// it's own isn't sufficient to insert a child into a directory. InsertChild +// updates the link count on d if required. +// +// Precondition: d must represent a directory inode. +func (d *Dentry) InsertChild(name string, child *vfs.Dentry) { + if !d.isDir() { + panic(fmt.Sprintf("InsertChild called on non-directory Dentry: %+v.", d)) + } + vfsDentry := d.VFSDentry() + vfsDentry.IncRef() // DecRef in child's Dentry.destroy. + d.dirMu.Lock() + vfsDentry.InsertChild(child, name) + d.dirMu.Unlock() +} + +// The Inode interface maps filesystem-level operations that operate on paths to +// equivalent operations on specific filesystem nodes. +// +// The interface methods are groups into logical categories as sub interfaces +// below. Generally, an implementation for each sub interface can be provided by +// embedding an appropriate type from inode_impl_utils.go. The sub interfaces +// are purely organizational. Methods declared directly in the main interface +// have no generic implementations, and should be explicitly provided by the +// client filesystem. +// +// Generally, implementations are not responsible for tasks that are common to +// all filesystems. These include: +// +// - Checking that dentries passed to methods are of the appropriate file type. +// - Checking permissions. +// - Updating link and reference counts. +// +// Specific responsibilities of implementations are documented below. +type Inode interface { + // Methods related to reference counting. A generic implementation is + // provided by InodeNoopRefCount. These methods are generally called by the + // equivalent Dentry methods. + inodeRefs + + // Methods related to node metadata. A generic implementation is provided by + // InodeAttrs. + inodeMetadata + + // Method for inodes that represent symlink. InodeNotSymlink provides a + // blanket implementation for all non-symlink inodes. + inodeSymlink + + // Method for inodes that represent directories. InodeNotDirectory provides + // a blanket implementation for all non-directory inodes. + inodeDirectory + + // Method for inodes that represent dynamic directories and their + // children. InodeNoDynamicLookup provides a blanket implementation for all + // non-dynamic-directory inodes. + inodeDynamicLookup + + // Open creates a file description for the filesystem object represented by + // this inode. The returned file description should hold a reference on the + // inode for its lifetime. + // + // Precondition: !rp.Done(). vfsd.Impl() must be a kernfs Dentry. + Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) +} + +type inodeRefs interface { + IncRef() + DecRef() + TryIncRef() bool + // Destroy is called when the inode reaches zero references. Destroy release + // all resources (references) on objects referenced by the inode, including + // any child dentries. + Destroy() +} + +type inodeMetadata interface { + // CheckPermissions checks that creds may access this inode for the + // requested access type, per the the rules of + // fs/namei.c:generic_permission(). + CheckPermissions(creds *auth.Credentials, atx vfs.AccessTypes) error + + // Mode returns the (struct stat)::st_mode value for this inode. This is + // separated from Stat for performance. + Mode() linux.FileMode + + // Stat returns the metadata for this inode. This corresponds to + // vfs.FilesystemImpl.StatAt. + Stat(fs *vfs.Filesystem) linux.Statx + + // SetStat updates the metadata for this inode. This corresponds to + // vfs.FilesystemImpl.SetStatAt. + SetStat(fs *vfs.Filesystem, opts vfs.SetStatOptions) error +} + +// Precondition: All methods in this interface may only be called on directory +// inodes. +type inodeDirectory interface { + // The New{File,Dir,Node,Symlink} methods below should return a new inode + // hashed into this inode. + // + // These inode constructors are inode-level operations rather than + // filesystem-level operations to allow client filesystems to mix different + // implementations based on the new node's location in the + // filesystem. + + // HasChildren returns true if the directory inode has any children. + HasChildren() bool + + // NewFile creates a new regular file inode. + NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (*vfs.Dentry, error) + + // NewDir creates a new directory inode. + NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (*vfs.Dentry, error) + + // NewLink creates a new hardlink to a specified inode in this + // directory. Implementations should create a new kernfs Dentry pointing to + // target, and update target's link count. + NewLink(ctx context.Context, name string, target Inode) (*vfs.Dentry, error) + + // NewSymlink creates a new symbolic link inode. + NewSymlink(ctx context.Context, name, target string) (*vfs.Dentry, error) + + // NewNode creates a new filesystem node for a mknod syscall. + NewNode(ctx context.Context, name string, opts vfs.MknodOptions) (*vfs.Dentry, error) + + // Unlink removes a child dentry from this directory inode. + Unlink(ctx context.Context, name string, child *vfs.Dentry) error + + // RmDir removes an empty child directory from this directory + // inode. Implementations must update the parent directory's link count, + // if required. Implementations are not responsible for checking that child + // is a directory, checking for an empty directory. + RmDir(ctx context.Context, name string, child *vfs.Dentry) error + + // Rename is called on the source directory containing an inode being + // renamed. child should point to the resolved child in the source + // directory. If Rename replaces a dentry in the destination directory, it + // should return the replaced dentry or nil otherwise. + // + // Precondition: Caller must serialize concurrent calls to Rename. + Rename(ctx context.Context, oldname, newname string, child, dstDir *vfs.Dentry) (replaced *vfs.Dentry, err error) +} + +type inodeDynamicLookup interface { + // Lookup should return an appropriate dentry if name should resolve to a + // child of this dynamic directory inode. This gives the directory an + // opportunity on every lookup to resolve additional entries that aren't + // hashed into the directory. This is only called when the inode is a + // directory. If the inode is not a directory, or if the directory only + // contains a static set of children, the implementer can unconditionally + // return an appropriate error (ENOTDIR and ENOENT respectively). + // + // The child returned by Lookup will be hashed into the VFS dentry tree. Its + // lifetime can be controlled by the filesystem implementation with an + // appropriate implementation of Valid. + // + // Lookup returns the child with an extra reference and the caller owns this + // reference. + Lookup(ctx context.Context, name string) (*vfs.Dentry, error) + + // Valid should return true if this inode is still valid, or needs to + // be resolved again by a call to Lookup. + Valid(ctx context.Context) bool +} + +type inodeSymlink interface { + // Readlink resolves the target of a symbolic link. If an inode is not a + // symlink, the implementation should return EINVAL. + Readlink(ctx context.Context) (string, error) +} diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go new file mode 100644 index 000000000..f78bb7b04 --- /dev/null +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -0,0 +1,423 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package kernfs_test + +import ( + "bytes" + "fmt" + "io" + "runtime" + "sync" + "testing" + + "github.com/google/go-cmp/cmp" + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +const defaultMode linux.FileMode = 01777 +const staticFileContent = "This is sample content for a static test file." + +// RootDentryFn is a generator function for creating the root dentry of a test +// filesystem. See newTestSystem. +type RootDentryFn func(*auth.Credentials, *filesystem) *kernfs.Dentry + +// TestSystem represents the context for a single test. +type TestSystem struct { + t *testing.T + ctx context.Context + creds *auth.Credentials + vfs *vfs.VirtualFilesystem + mns *vfs.MountNamespace + root vfs.VirtualDentry +} + +// newTestSystem sets up a minimal environment for running a test, including an +// instance of a test filesystem. Tests can control the contents of the +// filesystem by providing an appropriate rootFn, which should return a +// pre-populated root dentry. +func newTestSystem(t *testing.T, rootFn RootDentryFn) *TestSystem { + ctx := contexttest.Context(t) + creds := auth.CredentialsFromContext(ctx) + v := vfs.New() + v.MustRegisterFilesystemType("testfs", &fsType{rootFn: rootFn}) + mns, err := v.NewMountNamespace(ctx, creds, "", "testfs", &vfs.GetFilesystemOptions{}) + if err != nil { + t.Fatalf("Failed to create testfs root mount: %v", err) + } + + s := &TestSystem{ + t: t, + ctx: ctx, + creds: creds, + vfs: v, + mns: mns, + root: mns.Root(), + } + runtime.SetFinalizer(s, func(s *TestSystem) { s.root.DecRef() }) + return s +} + +// PathOpAtRoot constructs a vfs.PathOperation for a path from the +// root of the test filesystem. +// +// Precondition: path should be relative path. +func (s *TestSystem) PathOpAtRoot(path string) vfs.PathOperation { + return vfs.PathOperation{ + Root: s.root, + Start: s.root, + Pathname: path, + } +} + +// GetDentryOrDie attempts to resolve a dentry referred to by the +// provided path operation. If unsuccessful, the test fails. +func (s *TestSystem) GetDentryOrDie(pop vfs.PathOperation) vfs.VirtualDentry { + vd, err := s.vfs.GetDentryAt(s.ctx, s.creds, &pop, &vfs.GetDentryOptions{}) + if err != nil { + s.t.Fatalf("GetDentryAt(pop:%+v) failed: %v", pop, err) + } + return vd +} + +func (s *TestSystem) ReadToEnd(fd *vfs.FileDescription) (string, error) { + buf := make([]byte, usermem.PageSize) + bufIOSeq := usermem.BytesIOSequence(buf) + opts := vfs.ReadOptions{} + + var content bytes.Buffer + for { + n, err := fd.Impl().Read(s.ctx, bufIOSeq, opts) + if n == 0 || err != nil { + if err == io.EOF { + err = nil + } + return content.String(), err + } + content.Write(buf[:n]) + } +} + +type fsType struct { + rootFn RootDentryFn +} + +type filesystem struct { + kernfs.Filesystem +} + +type file struct { + kernfs.DynamicBytesFile + content string +} + +func (fs *filesystem) newFile(creds *auth.Credentials, content string) *kernfs.Dentry { + f := &file{} + f.content = content + f.DynamicBytesFile.Init(creds, fs.NextIno(), f) + + d := &kernfs.Dentry{} + d.Init(f) + return d +} + +func (f *file) Generate(ctx context.Context, buf *bytes.Buffer) error { + fmt.Fprintf(buf, "%s", f.content) + return nil +} + +type attrs struct { + kernfs.InodeAttrs +} + +func (a *attrs) SetStat(fs *vfs.Filesystem, opt vfs.SetStatOptions) error { + return syserror.EPERM +} + +type readonlyDir struct { + attrs + kernfs.InodeNotSymlink + kernfs.InodeNoDynamicLookup + kernfs.InodeDirectoryNoNewChildren + + kernfs.OrderedChildren + dentry kernfs.Dentry +} + +func (fs *filesystem) newReadonlyDir(creds *auth.Credentials, mode linux.FileMode, contents map[string]*kernfs.Dentry) *kernfs.Dentry { + dir := &readonlyDir{} + dir.attrs.Init(creds, fs.NextIno(), linux.ModeDirectory|mode) + dir.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) + dir.dentry.Init(dir) + + dir.IncLinks(dir.OrderedChildren.Populate(&dir.dentry, contents)) + + return &dir.dentry +} + +func (d *readonlyDir) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { + fd := &kernfs.GenericDirectoryFD{} + fd.Init(rp.Mount(), vfsd, &d.OrderedChildren, flags) + return fd.VFSFileDescription(), nil +} + +type dir struct { + attrs + kernfs.InodeNotSymlink + kernfs.InodeNoDynamicLookup + + fs *filesystem + dentry kernfs.Dentry + kernfs.OrderedChildren +} + +func (fs *filesystem) newDir(creds *auth.Credentials, mode linux.FileMode, contents map[string]*kernfs.Dentry) *kernfs.Dentry { + dir := &dir{} + dir.fs = fs + dir.attrs.Init(creds, fs.NextIno(), linux.ModeDirectory|mode) + dir.OrderedChildren.Init(kernfs.OrderedChildrenOptions{Writable: true}) + dir.dentry.Init(dir) + + dir.IncLinks(dir.OrderedChildren.Populate(&dir.dentry, contents)) + + return &dir.dentry +} + +func (d *dir) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { + fd := &kernfs.GenericDirectoryFD{} + fd.Init(rp.Mount(), vfsd, &d.OrderedChildren, flags) + return fd.VFSFileDescription(), nil +} + +func (d *dir) NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (*vfs.Dentry, error) { + creds := auth.CredentialsFromContext(ctx) + dir := d.fs.newDir(creds, opts.Mode, nil) + dirVFSD := dir.VFSDentry() + if err := d.OrderedChildren.Insert(name, dirVFSD); err != nil { + dir.DecRef() + return nil, err + } + d.IncLinks(1) + return dirVFSD, nil +} + +func (d *dir) NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (*vfs.Dentry, error) { + creds := auth.CredentialsFromContext(ctx) + f := d.fs.newFile(creds, "") + fVFSD := f.VFSDentry() + if err := d.OrderedChildren.Insert(name, fVFSD); err != nil { + f.DecRef() + return nil, err + } + return fVFSD, nil +} + +func (*dir) NewLink(context.Context, string, kernfs.Inode) (*vfs.Dentry, error) { + return nil, syserror.EPERM +} + +func (*dir) NewSymlink(context.Context, string, string) (*vfs.Dentry, error) { + return nil, syserror.EPERM +} + +func (*dir) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, error) { + return nil, syserror.EPERM +} + +func (fst *fsType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opt vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { + fs := &filesystem{} + fs.Init(vfsObj) + root := fst.rootFn(creds, fs) + return fs.VFSFilesystem(), root.VFSDentry(), nil +} + +// -------------------- Remainder of the file are test cases -------------------- + +func TestBasic(t *testing.T) { + sys := newTestSystem(t, func(creds *auth.Credentials, fs *filesystem) *kernfs.Dentry { + return fs.newReadonlyDir(creds, 0755, map[string]*kernfs.Dentry{ + "file1": fs.newFile(creds, staticFileContent), + }) + }) + sys.GetDentryOrDie(sys.PathOpAtRoot("file1")).DecRef() +} + +func TestMkdirGetDentry(t *testing.T) { + sys := newTestSystem(t, func(creds *auth.Credentials, fs *filesystem) *kernfs.Dentry { + return fs.newReadonlyDir(creds, 0755, map[string]*kernfs.Dentry{ + "dir1": fs.newDir(creds, 0755, nil), + }) + }) + + pop := sys.PathOpAtRoot("dir1/a new directory") + if err := sys.vfs.MkdirAt(sys.ctx, sys.creds, &pop, &vfs.MkdirOptions{Mode: 0755}); err != nil { + t.Fatalf("MkdirAt for PathOperation %+v failed: %v", pop, err) + } + sys.GetDentryOrDie(pop).DecRef() +} + +func TestReadStaticFile(t *testing.T) { + sys := newTestSystem(t, func(creds *auth.Credentials, fs *filesystem) *kernfs.Dentry { + return fs.newReadonlyDir(creds, 0755, map[string]*kernfs.Dentry{ + "file1": fs.newFile(creds, staticFileContent), + }) + }) + + pop := sys.PathOpAtRoot("file1") + fd, err := sys.vfs.OpenAt(sys.ctx, sys.creds, &pop, &vfs.OpenOptions{}) + if err != nil { + sys.t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) + } + defer fd.DecRef() + + content, err := sys.ReadToEnd(fd) + if err != nil { + sys.t.Fatalf("Read failed: %v", err) + } + if diff := cmp.Diff(staticFileContent, content); diff != "" { + sys.t.Fatalf("Read returned unexpected data:\n--- want\n+++ got\n%v", diff) + } +} + +func TestCreateNewFileInStaticDir(t *testing.T) { + sys := newTestSystem(t, func(creds *auth.Credentials, fs *filesystem) *kernfs.Dentry { + return fs.newReadonlyDir(creds, 0755, map[string]*kernfs.Dentry{ + "dir1": fs.newDir(creds, 0755, nil), + }) + }) + + pop := sys.PathOpAtRoot("dir1/newfile") + opts := &vfs.OpenOptions{Flags: linux.O_CREAT | linux.O_EXCL, Mode: defaultMode} + fd, err := sys.vfs.OpenAt(sys.ctx, sys.creds, &pop, opts) + if err != nil { + sys.t.Fatalf("OpenAt(pop:%+v, opts:%+v) failed: %v", pop, opts, err) + } + + // Close the file. The file should persist. + fd.DecRef() + + fd, err = sys.vfs.OpenAt(sys.ctx, sys.creds, &pop, &vfs.OpenOptions{}) + if err != nil { + sys.t.Fatalf("OpenAt(pop:%+v) = %+v failed: %v", pop, fd, err) + } + fd.DecRef() +} + +// direntCollector provides an implementation for vfs.IterDirentsCallback for +// testing. It simply iterates to the end of a given directory FD and collects +// all dirents emitted by the callback. +type direntCollector struct { + mu sync.Mutex + dirents map[string]vfs.Dirent +} + +// Handle implements vfs.IterDirentsCallback.Handle. +func (d *direntCollector) Handle(dirent vfs.Dirent) bool { + d.mu.Lock() + if d.dirents == nil { + d.dirents = make(map[string]vfs.Dirent) + } + d.dirents[dirent.Name] = dirent + d.mu.Unlock() + return true +} + +// count returns the number of dirents currently in the collector. +func (d *direntCollector) count() int { + d.mu.Lock() + defer d.mu.Unlock() + return len(d.dirents) +} + +// contains checks whether the collector has a dirent with the given name and +// type. +func (d *direntCollector) contains(name string, typ uint8) error { + d.mu.Lock() + defer d.mu.Unlock() + dirent, ok := d.dirents[name] + if !ok { + return fmt.Errorf("No dirent named %q found", name) + } + if dirent.Type != typ { + return fmt.Errorf("Dirent named %q found, but was expecting type %d, got: %+v", name, typ, dirent) + } + return nil +} + +func TestDirFDReadWrite(t *testing.T) { + sys := newTestSystem(t, func(creds *auth.Credentials, fs *filesystem) *kernfs.Dentry { + return fs.newReadonlyDir(creds, 0755, nil) + }) + + pop := sys.PathOpAtRoot("/") + fd, err := sys.vfs.OpenAt(sys.ctx, sys.creds, &pop, &vfs.OpenOptions{}) + if err != nil { + sys.t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) + } + defer fd.DecRef() + + // Read/Write should fail for directory FDs. + if _, err := fd.Read(sys.ctx, usermem.BytesIOSequence([]byte{}), vfs.ReadOptions{}); err != syserror.EISDIR { + sys.t.Fatalf("Read for directory FD failed with unexpected error: %v", err) + } + if _, err := fd.Write(sys.ctx, usermem.BytesIOSequence([]byte{}), vfs.WriteOptions{}); err != syserror.EISDIR { + sys.t.Fatalf("Wrire for directory FD failed with unexpected error: %v", err) + } +} + +func TestDirFDIterDirents(t *testing.T) { + sys := newTestSystem(t, func(creds *auth.Credentials, fs *filesystem) *kernfs.Dentry { + return fs.newReadonlyDir(creds, 0755, map[string]*kernfs.Dentry{ + // Fill root with nodes backed by various inode implementations. + "dir1": fs.newReadonlyDir(creds, 0755, nil), + "dir2": fs.newDir(creds, 0755, map[string]*kernfs.Dentry{ + "dir3": fs.newDir(creds, 0755, nil), + }), + "file1": fs.newFile(creds, staticFileContent), + }) + }) + + pop := sys.PathOpAtRoot("/") + fd, err := sys.vfs.OpenAt(sys.ctx, sys.creds, &pop, &vfs.OpenOptions{}) + if err != nil { + sys.t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) + } + defer fd.DecRef() + + collector := &direntCollector{} + if err := fd.IterDirents(sys.ctx, collector); err != nil { + sys.t.Fatalf("IterDirent failed: %v", err) + } + + // Root directory should contain ".", ".." and 3 children: + if collector.count() != 5 { + sys.t.Fatalf("IterDirent returned too many dirents") + } + for _, dirName := range []string{".", "..", "dir1", "dir2"} { + if err := collector.contains(dirName, linux.DT_DIR); err != nil { + sys.t.Fatalf("IterDirent had unexpected results: %v", err) + } + } + if err := collector.contains("file1", linux.DT_REG); err != nil { + sys.t.Fatalf("IterDirent had unexpected results: %v", err) + } + +} -- cgit v1.2.3 From 93d429d5b1e3801fb4c29568bcd40d6854c9fe94 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Thu, 12 Dec 2019 13:17:47 -0800 Subject: Implement memmap.MappingIdentity for vfs.FileDescription. PiperOrigin-RevId: 285255855 --- pkg/sentry/memmap/BUILD | 1 - pkg/sentry/memmap/memmap.go | 8 +++-- pkg/sentry/vfs/context.go | 13 ++++++++ pkg/sentry/vfs/file_description.go | 44 ++++++++++++++++++++++++++++ pkg/sentry/vfs/file_description_impl_util.go | 9 ++++++ 5 files changed, 71 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/memmap/BUILD b/pkg/sentry/memmap/BUILD index 3ef84245b..112794e9c 100644 --- a/pkg/sentry/memmap/BUILD +++ b/pkg/sentry/memmap/BUILD @@ -41,7 +41,6 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/log", - "//pkg/refs", "//pkg/sentry/context", "//pkg/sentry/platform", "//pkg/sentry/usermem", diff --git a/pkg/sentry/memmap/memmap.go b/pkg/sentry/memmap/memmap.go index 03b99aaea..16a722a13 100644 --- a/pkg/sentry/memmap/memmap.go +++ b/pkg/sentry/memmap/memmap.go @@ -18,7 +18,6 @@ package memmap import ( "fmt" - "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/usermem" @@ -235,8 +234,11 @@ type InvalidateOpts struct { // coincidental; fs.File implements MappingIdentity, and some // fs.InodeOperations implement Mappable.) type MappingIdentity interface { - // MappingIdentity is reference-counted. - refs.RefCounter + // IncRef increments the MappingIdentity's reference count. + IncRef() + + // DecRef decrements the MappingIdentity's reference count. + DecRef() // MappedName returns the application-visible name shown in // /proc/[pid]/maps. diff --git a/pkg/sentry/vfs/context.go b/pkg/sentry/vfs/context.go index 32cf9151b..705194ebc 100644 --- a/pkg/sentry/vfs/context.go +++ b/pkg/sentry/vfs/context.go @@ -24,6 +24,9 @@ type contextID int const ( // CtxMountNamespace is a Context.Value key for a MountNamespace. CtxMountNamespace contextID = iota + + // CtxRoot is a Context.Value key for a VFS root. + CtxRoot ) // MountNamespaceFromContext returns the MountNamespace used by ctx. It does @@ -35,3 +38,13 @@ func MountNamespaceFromContext(ctx context.Context) *MountNamespace { } return nil } + +// RootFromContext returns the VFS root used by ctx. It takes a reference on +// the returned VirtualDentry. If ctx does not have a specific VFS root, +// RootFromContext returns a zero-value VirtualDentry. +func RootFromContext(ctx context.Context) VirtualDentry { + if v := ctx.Value(CtxRoot); v != nil { + return v.(VirtualDentry) + } + return VirtualDentry{} +} diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 4473dfce8..6575afd16 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -334,3 +334,47 @@ func (fd *FileDescription) Ioctl(ctx context.Context, uio usermem.IO, args arch. func (fd *FileDescription) SyncFS(ctx context.Context) error { return fd.vd.mount.fs.impl.Sync(ctx) } + +// MappedName implements memmap.MappingIdentity.MappedName. +func (fd *FileDescription) MappedName(ctx context.Context) string { + vfsroot := RootFromContext(ctx) + s, _ := fd.vd.mount.vfs.PathnameWithDeleted(ctx, vfsroot, fd.vd) + if vfsroot.Ok() { + vfsroot.DecRef() + } + return s +} + +// DeviceID implements memmap.MappingIdentity.DeviceID. +func (fd *FileDescription) DeviceID() uint64 { + stat, err := fd.impl.Stat(context.Background(), StatOptions{ + // There is no STATX_DEV; we assume that Stat will return it if it's + // available regardless of mask. + Mask: 0, + // fs/proc/task_mmu.c:show_map_vma() just reads inode::i_sb->s_dev + // directly. + Sync: linux.AT_STATX_DONT_SYNC, + }) + if err != nil { + return 0 + } + return uint64(linux.MakeDeviceID(uint16(stat.DevMajor), stat.DevMinor)) +} + +// InodeID implements memmap.MappingIdentity.InodeID. +func (fd *FileDescription) InodeID() uint64 { + stat, err := fd.impl.Stat(context.Background(), StatOptions{ + Mask: linux.STATX_INO, + // fs/proc/task_mmu.c:show_map_vma() just reads inode::i_ino directly. + Sync: linux.AT_STATX_DONT_SYNC, + }) + if err != nil || stat.Mask&linux.STATX_INO == 0 { + return 0 + } + return stat.Ino +} + +// Msync implements memmap.MappingIdentity.Msync. +func (fd *FileDescription) Msync(ctx context.Context, mr memmap.MappableRange) error { + return fd.impl.Sync(ctx) +} diff --git a/pkg/sentry/vfs/file_description_impl_util.go b/pkg/sentry/vfs/file_description_impl_util.go index 4fbad7840..aae023254 100644 --- a/pkg/sentry/vfs/file_description_impl_util.go +++ b/pkg/sentry/vfs/file_description_impl_util.go @@ -252,3 +252,12 @@ func (fd *DynamicBytesFileDescriptionImpl) Seek(ctx context.Context, offset int6 fd.off = offset return offset, nil } + +// GenericConfigureMMap may be used by most implementations of +// FileDescriptionImpl.ConfigureMMap. +func GenericConfigureMMap(fd *FileDescription, m memmap.Mappable, opts *memmap.MMapOpts) error { + opts.Mappable = m + opts.MappingIdentity = fd + fd.IncRef() + return nil +} -- cgit v1.2.3 From 6b424530397e5100b08628efe8f6c62178daa70b Mon Sep 17 00:00:00 2001 From: lubinszARM <34124929+lubinszARM@users.noreply.github.com> Date: Fri, 13 Dec 2019 17:09:55 -0800 Subject: enable kvm to support arm64 There are 4 jobs were finished in this package: 1, Virtual machine initialization. 2, Bluepill implementation. 3, Move ring0.Vectors() into the address with 11-bits alignment. 4, Basic support for "SwitchToUser". Signed-off-by: Bin Lu COPYBARA_INTEGRATE_REVIEW=https://github.com/google/gvisor/pull/919 from lubinszARM:pr_kvm eedea52db451bf62722759009a9f14c54a69c55f PiperOrigin-RevId: 285501256 --- pkg/sentry/platform/kvm/BUILD | 10 +- pkg/sentry/platform/kvm/bluepill.go | 24 +- pkg/sentry/platform/kvm/bluepill_amd64.go | 20 +- pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go | 7 - pkg/sentry/platform/kvm/bluepill_arm64.go | 79 +++++ pkg/sentry/platform/kvm/bluepill_arm64.s | 87 ++++++ pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go | 28 ++ pkg/sentry/platform/kvm/bluepill_unsafe.go | 9 + pkg/sentry/platform/kvm/filters.go | 33 --- pkg/sentry/platform/kvm/filters_amd64.go | 33 +++ pkg/sentry/platform/kvm/filters_arm64.go | 32 ++ pkg/sentry/platform/kvm/kvm.go | 5 +- pkg/sentry/platform/kvm/kvm_amd64.go | 9 + pkg/sentry/platform/kvm/kvm_arm64.go | 83 ++++++ pkg/sentry/platform/kvm/kvm_arm64_unsafe.go | 39 +++ pkg/sentry/platform/kvm/kvm_const.go | 4 +- pkg/sentry/platform/kvm/kvm_const_arm64.go | 132 +++++++++ pkg/sentry/platform/kvm/machine_amd64_unsafe.go | 40 +++ pkg/sentry/platform/kvm/machine_arm64.go | 122 ++++++++ pkg/sentry/platform/kvm/machine_arm64_unsafe.go | 362 +++++++++++++++++++++++ pkg/sentry/platform/kvm/machine_unsafe.go | 40 --- 21 files changed, 1092 insertions(+), 106 deletions(-) create mode 100644 pkg/sentry/platform/kvm/bluepill_arm64.go create mode 100644 pkg/sentry/platform/kvm/bluepill_arm64.s create mode 100644 pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go delete mode 100644 pkg/sentry/platform/kvm/filters.go create mode 100644 pkg/sentry/platform/kvm/filters_amd64.go create mode 100644 pkg/sentry/platform/kvm/filters_arm64.go create mode 100644 pkg/sentry/platform/kvm/kvm_arm64.go create mode 100644 pkg/sentry/platform/kvm/kvm_arm64_unsafe.go create mode 100644 pkg/sentry/platform/kvm/kvm_const_arm64.go create mode 100644 pkg/sentry/platform/kvm/machine_arm64_unsafe.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/kvm/BUILD b/pkg/sentry/platform/kvm/BUILD index 6803d488c..5c52d4007 100644 --- a/pkg/sentry/platform/kvm/BUILD +++ b/pkg/sentry/platform/kvm/BUILD @@ -12,18 +12,26 @@ go_library( "bluepill_amd64.go", "bluepill_amd64.s", "bluepill_amd64_unsafe.go", + "bluepill_arm64.go", + "bluepill_arm64.s", + "bluepill_arm64_unsafe.go", "bluepill_fault.go", "bluepill_unsafe.go", "context.go", - "filters.go", + "filters_amd64.go", + "filters_arm64.go", "kvm.go", "kvm_amd64.go", "kvm_amd64_unsafe.go", + "kvm_arm64.go", + "kvm_arm64_unsafe.go", "kvm_const.go", + "kvm_const_arm64.go", "machine.go", "machine_amd64.go", "machine_amd64_unsafe.go", "machine_arm64.go", + "machine_arm64_unsafe.go", "machine_unsafe.go", "physical_map.go", "virtual_map.go", diff --git a/pkg/sentry/platform/kvm/bluepill.go b/pkg/sentry/platform/kvm/bluepill.go index 043de51b3..30dbb74d6 100644 --- a/pkg/sentry/platform/kvm/bluepill.go +++ b/pkg/sentry/platform/kvm/bluepill.go @@ -20,6 +20,7 @@ import ( "syscall" "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/platform/ring0" "gvisor.dev/gvisor/pkg/sentry/platform/safecopy" ) @@ -36,6 +37,18 @@ func sighandler() func dieTrampoline() var ( + // bounceSignal is the signal used for bouncing KVM. + // + // We use SIGCHLD because it is not masked by the runtime, and + // it will be ignored properly by other parts of the kernel. + bounceSignal = syscall.SIGCHLD + + // bounceSignalMask has only bounceSignal set. + bounceSignalMask = uint64(1 << (uint64(bounceSignal) - 1)) + + // bounce is the interrupt vector used to return to the kernel. + bounce = uint32(ring0.VirtualizationException) + // savedHandler is a pointer to the previous handler. // // This is called by bluepillHandler. @@ -45,6 +58,13 @@ var ( dieTrampolineAddr uintptr ) +// redpill invokes a syscall with -1. +// +//go:nosplit +func redpill() { + syscall.RawSyscall(^uintptr(0), 0, 0, 0) +} + // dieHandler is called by dieTrampoline. // //go:nosplit @@ -73,8 +93,8 @@ func (c *vCPU) die(context *arch.SignalContext64, msg string) { func init() { // Install the handler. - if err := safecopy.ReplaceSignalHandler(syscall.SIGSEGV, reflect.ValueOf(sighandler).Pointer(), &savedHandler); err != nil { - panic(fmt.Sprintf("Unable to set handler for signal %d: %v", syscall.SIGSEGV, err)) + if err := safecopy.ReplaceSignalHandler(bluepillSignal, reflect.ValueOf(sighandler).Pointer(), &savedHandler); err != nil { + panic(fmt.Sprintf("Unable to set handler for signal %d: %v", bluepillSignal, err)) } // Extract the address for the trampoline. diff --git a/pkg/sentry/platform/kvm/bluepill_amd64.go b/pkg/sentry/platform/kvm/bluepill_amd64.go index 421c88220..133c2203d 100644 --- a/pkg/sentry/platform/kvm/bluepill_amd64.go +++ b/pkg/sentry/platform/kvm/bluepill_amd64.go @@ -24,26 +24,10 @@ import ( ) var ( - // bounceSignal is the signal used for bouncing KVM. - // - // We use SIGCHLD because it is not masked by the runtime, and - // it will be ignored properly by other parts of the kernel. - bounceSignal = syscall.SIGCHLD - - // bounceSignalMask has only bounceSignal set. - bounceSignalMask = uint64(1 << (uint64(bounceSignal) - 1)) - - // bounce is the interrupt vector used to return to the kernel. - bounce = uint32(ring0.VirtualizationException) + // The action for bluepillSignal is changed by sigaction(). + bluepillSignal = syscall.SIGSEGV ) -// redpill on amd64 invokes a syscall with -1. -// -//go:nosplit -func redpill() { - syscall.RawSyscall(^uintptr(0), 0, 0, 0) -} - // bluepillArchEnter is called during bluepillEnter. // //go:nosplit diff --git a/pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go b/pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go index 9d8af143e..a63a6a071 100644 --- a/pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go @@ -23,13 +23,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/platform/ring0" ) -// bluepillArchContext returns the arch-specific context. -// -//go:nosplit -func bluepillArchContext(context unsafe.Pointer) *arch.SignalContext64 { - return &((*arch.UContext64)(context).MContext) -} - // dieArchSetup initializes the state for dieTrampoline. // // The amd64 dieTrampoline requires the vCPU to be set in BX, and the last RIP diff --git a/pkg/sentry/platform/kvm/bluepill_arm64.go b/pkg/sentry/platform/kvm/bluepill_arm64.go new file mode 100644 index 000000000..552341721 --- /dev/null +++ b/pkg/sentry/platform/kvm/bluepill_arm64.go @@ -0,0 +1,79 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package kvm + +import ( + "syscall" + + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/platform/ring0" +) + +var ( + // The action for bluepillSignal is changed by sigaction(). + bluepillSignal = syscall.SIGILL +) + +// bluepillArchEnter is called during bluepillEnter. +// +//go:nosplit +func bluepillArchEnter(context *arch.SignalContext64) (c *vCPU) { + c = vCPUPtr(uintptr(context.Regs[8])) + regs := c.CPU.Registers() + regs.Regs = context.Regs + regs.Sp = context.Sp + regs.Pc = context.Pc + regs.Pstate = context.Pstate + regs.Pstate &^= uint64(ring0.KernelFlagsClear) + regs.Pstate |= ring0.KernelFlagsSet + return +} + +// bluepillArchExit is called during bluepillEnter. +// +//go:nosplit +func bluepillArchExit(c *vCPU, context *arch.SignalContext64) { + regs := c.CPU.Registers() + context.Regs = regs.Regs + context.Sp = regs.Sp + context.Pc = regs.Pc + context.Pstate = regs.Pstate + context.Pstate &^= uint64(ring0.UserFlagsClear) + context.Pstate |= ring0.UserFlagsSet +} + +// KernelSyscall handles kernel syscalls. +// +//go:nosplit +func (c *vCPU) KernelSyscall() { + regs := c.Registers() + if regs.Regs[8] != ^uint64(0) { + regs.Pc -= 4 // Rewind. + } + ring0.Halt() +} + +// KernelException handles kernel exceptions. +// +//go:nosplit +func (c *vCPU) KernelException(vector ring0.Vector) { + regs := c.Registers() + if vector == ring0.Vector(bounce) { + regs.Pc = 0 + } + ring0.Halt() +} diff --git a/pkg/sentry/platform/kvm/bluepill_arm64.s b/pkg/sentry/platform/kvm/bluepill_arm64.s new file mode 100644 index 000000000..c61700892 --- /dev/null +++ b/pkg/sentry/platform/kvm/bluepill_arm64.s @@ -0,0 +1,87 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "textflag.h" + +// VCPU_CPU is the location of the CPU in the vCPU struct. +// +// This is guaranteed to be zero. +#define VCPU_CPU 0x0 + +// CPU_SELF is the self reference in ring0's percpu. +// +// This is guaranteed to be zero. +#define CPU_SELF 0x0 + +// Context offsets. +// +// Only limited use of the context is done in the assembly stub below, most is +// done in the Go handlers. +#define SIGINFO_SIGNO 0x0 +#define CONTEXT_PC 0x1B8 +#define CONTEXT_R0 0xB8 + +// See bluepill.go. +TEXT ·bluepill(SB),NOSPLIT,$0 +begin: + MOVD vcpu+0(FP), R8 + MOVD $VCPU_CPU(R8), R9 + ORR $0xffff000000000000, R9, R9 + // Trigger sigill. + // In ring0.Start(), the value of R8 will be stored into tpidr_el1. + // When the context was loaded into vcpu successfully, + // we will check if the value of R10 and R9 are the same. + WORD $0xd538d08a // MRS TPIDR_EL1, R10 +check_vcpu: + CMP R10, R9 + BEQ right_vCPU +wrong_vcpu: + CALL ·redpill(SB) + B begin +right_vCPU: + RET + +// sighandler: see bluepill.go for documentation. +// +// The arguments are the following: +// +// R0 - The signal number. +// R1 - Pointer to siginfo_t structure. +// R2 - Pointer to ucontext structure. +// +TEXT ·sighandler(SB),NOSPLIT,$0 + // si_signo should be sigill. + MOVD SIGINFO_SIGNO(R1), R7 + CMPW $4, R7 + BNE fallback + + MOVD CONTEXT_PC(R2), R7 + CMPW $0, R7 + BEQ fallback + + MOVD R2, 8(RSP) + BL ·bluepillHandler(SB) // Call the handler. + + RET + +fallback: + // Jump to the previous signal handler. + MOVD ·savedHandler(SB), R7 + B (R7) + +// dieTrampoline: see bluepill.go, bluepill_arm64_unsafe.go for documentation. +TEXT ·dieTrampoline(SB),NOSPLIT,$0 + // TODO(gvisor.dev/issue/1249): dieTrampoline supporting for Arm64. + MOVD R9, 8(RSP) + BL ·dieHandler(SB) diff --git a/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go b/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go new file mode 100644 index 000000000..e5fac0d6a --- /dev/null +++ b/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go @@ -0,0 +1,28 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package kvm + +import ( + "unsafe" + + "gvisor.dev/gvisor/pkg/sentry/arch" +) + +//go:nosplit +func dieArchSetup(c *vCPU, context *arch.SignalContext64, guestRegs *userRegs) { + // TODO(gvisor.dev/issue/1249): dieTrampoline supporting for Arm64. +} diff --git a/pkg/sentry/platform/kvm/bluepill_unsafe.go b/pkg/sentry/platform/kvm/bluepill_unsafe.go index ca011ef78..9add7c944 100644 --- a/pkg/sentry/platform/kvm/bluepill_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_unsafe.go @@ -23,6 +23,8 @@ import ( "sync/atomic" "syscall" "unsafe" + + "gvisor.dev/gvisor/pkg/sentry/arch" ) //go:linkname throw runtime.throw @@ -49,6 +51,13 @@ func uintptrValue(addr *byte) uintptr { return (uintptr)(unsafe.Pointer(addr)) } +// bluepillArchContext returns the UContext64. +// +//go:nosplit +func bluepillArchContext(context unsafe.Pointer) *arch.SignalContext64 { + return &((*arch.UContext64)(context).MContext) +} + // bluepillHandler is called from the signal stub. // // The world may be stopped while this is executing, and it executes on the diff --git a/pkg/sentry/platform/kvm/filters.go b/pkg/sentry/platform/kvm/filters.go deleted file mode 100644 index 7d949f1dd..000000000 --- a/pkg/sentry/platform/kvm/filters.go +++ /dev/null @@ -1,33 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package kvm - -import ( - "syscall" - - "gvisor.dev/gvisor/pkg/seccomp" -) - -// SyscallFilters returns syscalls made exclusively by the KVM platform. -func (*KVM) SyscallFilters() seccomp.SyscallRules { - return seccomp.SyscallRules{ - syscall.SYS_ARCH_PRCTL: {}, - syscall.SYS_IOCTL: {}, - syscall.SYS_MMAP: {}, - syscall.SYS_RT_SIGSUSPEND: {}, - syscall.SYS_RT_SIGTIMEDWAIT: {}, - 0xffffffffffffffff: {}, // KVM uses syscall -1 to transition to host. - } -} diff --git a/pkg/sentry/platform/kvm/filters_amd64.go b/pkg/sentry/platform/kvm/filters_amd64.go new file mode 100644 index 000000000..7d949f1dd --- /dev/null +++ b/pkg/sentry/platform/kvm/filters_amd64.go @@ -0,0 +1,33 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package kvm + +import ( + "syscall" + + "gvisor.dev/gvisor/pkg/seccomp" +) + +// SyscallFilters returns syscalls made exclusively by the KVM platform. +func (*KVM) SyscallFilters() seccomp.SyscallRules { + return seccomp.SyscallRules{ + syscall.SYS_ARCH_PRCTL: {}, + syscall.SYS_IOCTL: {}, + syscall.SYS_MMAP: {}, + syscall.SYS_RT_SIGSUSPEND: {}, + syscall.SYS_RT_SIGTIMEDWAIT: {}, + 0xffffffffffffffff: {}, // KVM uses syscall -1 to transition to host. + } +} diff --git a/pkg/sentry/platform/kvm/filters_arm64.go b/pkg/sentry/platform/kvm/filters_arm64.go new file mode 100644 index 000000000..9245d07c2 --- /dev/null +++ b/pkg/sentry/platform/kvm/filters_arm64.go @@ -0,0 +1,32 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package kvm + +import ( + "syscall" + + "gvisor.dev/gvisor/pkg/seccomp" +) + +// SyscallFilters returns syscalls made exclusively by the KVM platform. +func (*KVM) SyscallFilters() seccomp.SyscallRules { + return seccomp.SyscallRules{ + syscall.SYS_IOCTL: {}, + syscall.SYS_MMAP: {}, + syscall.SYS_RT_SIGSUSPEND: {}, + syscall.SYS_RT_SIGTIMEDWAIT: {}, + 0xffffffffffffffff: {}, // KVM uses syscall -1 to transition to host. + } +} diff --git a/pkg/sentry/platform/kvm/kvm.go b/pkg/sentry/platform/kvm/kvm.go index ee4cd2f4d..f2c2c059e 100644 --- a/pkg/sentry/platform/kvm/kvm.go +++ b/pkg/sentry/platform/kvm/kvm.go @@ -21,7 +21,6 @@ import ( "sync" "syscall" - "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/ring0" "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" @@ -56,9 +55,7 @@ func New(deviceFile *os.File) (*KVM, error) { // Ensure global initialization is done. globalOnce.Do(func() { - physicalInit() - globalErr = updateSystemValues(int(fd)) - ring0.Init(cpuid.HostFeatureSet()) + globalErr = updateGlobalOnce(int(fd)) }) if globalErr != nil { return nil, globalErr diff --git a/pkg/sentry/platform/kvm/kvm_amd64.go b/pkg/sentry/platform/kvm/kvm_amd64.go index 5d8ef4761..c5a6f9c7d 100644 --- a/pkg/sentry/platform/kvm/kvm_amd64.go +++ b/pkg/sentry/platform/kvm/kvm_amd64.go @@ -17,6 +17,7 @@ package kvm import ( + "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/sentry/platform/ring0" ) @@ -211,3 +212,11 @@ type cpuidEntries struct { _ uint32 entries [_KVM_NR_CPUID_ENTRIES]cpuidEntry } + +// updateGlobalOnce does global initialization. It has to be called only once. +func updateGlobalOnce(fd int) error { + physicalInit() + err := updateSystemValues(int(fd)) + ring0.Init(cpuid.HostFeatureSet()) + return err +} diff --git a/pkg/sentry/platform/kvm/kvm_arm64.go b/pkg/sentry/platform/kvm/kvm_arm64.go new file mode 100644 index 000000000..2319c86d3 --- /dev/null +++ b/pkg/sentry/platform/kvm/kvm_arm64.go @@ -0,0 +1,83 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package kvm + +import ( + "syscall" +) + +// userMemoryRegion is a region of physical memory. +// +// This mirrors kvm_memory_region. +type userMemoryRegion struct { + slot uint32 + flags uint32 + guestPhysAddr uint64 + memorySize uint64 + userspaceAddr uint64 +} + +type kvmOneReg struct { + id uint64 + addr uint64 +} + +const KVM_NR_SPSR = 5 + +type userFpsimdState struct { + vregs [64]uint64 + fpsr uint32 + fpcr uint32 + reserved [2]uint32 +} + +type userRegs struct { + Regs syscall.PtraceRegs + sp_el1 uint64 + elr_el1 uint64 + spsr [KVM_NR_SPSR]uint64 + fpRegs userFpsimdState +} + +// runData is the run structure. This may be mapped for synchronous register +// access (although that doesn't appear to be supported by my kernel at least). +// +// This mirrors kvm_run. +type runData struct { + requestInterruptWindow uint8 + _ [7]uint8 + + exitReason uint32 + readyForInterruptInjection uint8 + ifFlag uint8 + _ [2]uint8 + + cr8 uint64 + apicBase uint64 + + // This is the union data for exits. Interpretation depends entirely on + // the exitReason above (see vCPU code for more information). + data [32]uint64 +} + +// updateGlobalOnce does global initialization. It has to be called only once. +func updateGlobalOnce(fd int) error { + physicalInit() + err := updateSystemValues(int(fd)) + updateVectorTable() + return err +} diff --git a/pkg/sentry/platform/kvm/kvm_arm64_unsafe.go b/pkg/sentry/platform/kvm/kvm_arm64_unsafe.go new file mode 100644 index 000000000..6531bae1d --- /dev/null +++ b/pkg/sentry/platform/kvm/kvm_arm64_unsafe.go @@ -0,0 +1,39 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package kvm + +import ( + "fmt" + "syscall" +) + +var ( + runDataSize int +) + +func updateSystemValues(fd int) error { + // Extract the mmap size. + sz, _, errno := syscall.RawSyscall(syscall.SYS_IOCTL, uintptr(fd), _KVM_GET_VCPU_MMAP_SIZE, 0) + if errno != 0 { + return fmt.Errorf("getting VCPU mmap size: %v", errno) + } + // Save the data. + runDataSize = int(sz) + + // Success. + return nil +} diff --git a/pkg/sentry/platform/kvm/kvm_const.go b/pkg/sentry/platform/kvm/kvm_const.go index 766131d60..1d5c77ff4 100644 --- a/pkg/sentry/platform/kvm/kvm_const.go +++ b/pkg/sentry/platform/kvm/kvm_const.go @@ -49,11 +49,13 @@ const ( _KVM_EXIT_SHUTDOWN = 0x8 _KVM_EXIT_FAIL_ENTRY = 0x9 _KVM_EXIT_INTERNAL_ERROR = 0x11 + _KVM_EXIT_SYSTEM_EVENT = 0x18 ) // KVM capability options. const ( - _KVM_CAP_MAX_VCPUS = 0x42 + _KVM_CAP_MAX_VCPUS = 0x42 + _KVM_CAP_ARM_VM_IPA_SIZE = 0xa5 ) // KVM limits. diff --git a/pkg/sentry/platform/kvm/kvm_const_arm64.go b/pkg/sentry/platform/kvm/kvm_const_arm64.go new file mode 100644 index 000000000..5a74c6e36 --- /dev/null +++ b/pkg/sentry/platform/kvm/kvm_const_arm64.go @@ -0,0 +1,132 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package kvm + +// KVM ioctls for Arm64. +const ( + _KVM_GET_ONE_REG = 0x4010aeab + _KVM_SET_ONE_REG = 0x4010aeac + + _KVM_ARM_PREFERRED_TARGET = 0x8020aeaf + _KVM_ARM_VCPU_INIT = 0x4020aeae + _KVM_ARM64_REGS_PSTATE = 0x6030000000100042 + _KVM_ARM64_REGS_SP_EL1 = 0x6030000000100044 + _KVM_ARM64_REGS_R0 = 0x6030000000100000 + _KVM_ARM64_REGS_R1 = 0x6030000000100002 + _KVM_ARM64_REGS_R2 = 0x6030000000100004 + _KVM_ARM64_REGS_R3 = 0x6030000000100006 + _KVM_ARM64_REGS_R8 = 0x6030000000100010 + _KVM_ARM64_REGS_R18 = 0x6030000000100024 + _KVM_ARM64_REGS_PC = 0x6030000000100040 + _KVM_ARM64_REGS_MAIR_EL1 = 0x603000000013c510 + _KVM_ARM64_REGS_TCR_EL1 = 0x603000000013c102 + _KVM_ARM64_REGS_TTBR0_EL1 = 0x603000000013c100 + _KVM_ARM64_REGS_TTBR1_EL1 = 0x603000000013c101 + _KVM_ARM64_REGS_SCTLR_EL1 = 0x603000000013c080 + _KVM_ARM64_REGS_CPACR_EL1 = 0x603000000013c082 + _KVM_ARM64_REGS_VBAR_EL1 = 0x603000000013c600 +) + +// Arm64: Architectural Feature Access Control Register EL1. +const ( + _FPEN_NOTRAP = 0x3 + _FPEN_SHIFT = 0x20 +) + +// Arm64: System Control Register EL1. +const ( + _SCTLR_M = 1 << 0 + _SCTLR_C = 1 << 2 + _SCTLR_I = 1 << 12 +) + +// Arm64: Translation Control Register EL1. +const ( + _TCR_IPS_40BITS = 2 << 32 // PA=40 + _TCR_IPS_48BITS = 5 << 32 // PA=48 + + _TCR_T0SZ_OFFSET = 0 + _TCR_T1SZ_OFFSET = 16 + _TCR_IRGN0_SHIFT = 8 + _TCR_IRGN1_SHIFT = 24 + _TCR_ORGN0_SHIFT = 10 + _TCR_ORGN1_SHIFT = 26 + _TCR_SH0_SHIFT = 12 + _TCR_SH1_SHIFT = 28 + _TCR_TG0_SHIFT = 14 + _TCR_TG1_SHIFT = 30 + + _TCR_T0SZ_VA48 = 64 - 48 // VA=48 + _TCR_T1SZ_VA48 = 64 - 48 // VA=48 + + _TCR_ASID16 = 1 << 36 + _TCR_TBI0 = 1 << 37 + + _TCR_TXSZ_VA48 = (_TCR_T0SZ_VA48 << _TCR_T0SZ_OFFSET) | (_TCR_T1SZ_VA48 << _TCR_T1SZ_OFFSET) + + _TCR_TG0_4K = 0 << _TCR_TG0_SHIFT // 4K + _TCR_TG0_64K = 1 << _TCR_TG0_SHIFT // 64K + + _TCR_TG1_4K = 2 << _TCR_TG1_SHIFT + + _TCR_TG_FLAGS = _TCR_TG0_4K | _TCR_TG1_4K + + _TCR_IRGN0_WBWA = 1 << _TCR_IRGN0_SHIFT + _TCR_IRGN1_WBWA = 1 << _TCR_IRGN1_SHIFT + _TCR_IRGN_WBWA = _TCR_IRGN0_WBWA | _TCR_IRGN1_WBWA + + _TCR_ORGN0_WBWA = 1 << _TCR_ORGN0_SHIFT + _TCR_ORGN1_WBWA = 1 << _TCR_ORGN1_SHIFT + + _TCR_ORGN_WBWA = _TCR_ORGN0_WBWA | _TCR_ORGN1_WBWA + + _TCR_SHARED = (3 << _TCR_SH0_SHIFT) | (3 << _TCR_SH1_SHIFT) + + _TCR_CACHE_FLAGS = _TCR_IRGN_WBWA | _TCR_ORGN_WBWA +) + +// Arm64: Memory Attribute Indirection Register EL1. +const ( + _MT_DEVICE_nGnRnE = 0 + _MT_DEVICE_nGnRE = 1 + _MT_DEVICE_GRE = 2 + _MT_NORMAL_NC = 3 + _MT_NORMAL = 4 + _MT_NORMAL_WT = 5 + _MT_EL1_INIT = (0 << _MT_DEVICE_nGnRnE) | (0x4 << _MT_DEVICE_nGnRE * 8) | (0xc << _MT_DEVICE_GRE * 8) | (0x44 << _MT_NORMAL_NC * 8) | (0xff << _MT_NORMAL * 8) | (0xbb << _MT_NORMAL_WT * 8) +) + +const ( + _KVM_ARM_VCPU_POWER_OFF = 0 // CPU is started in OFF state + _KVM_ARM_VCPU_PSCI_0_2 = 2 // CPU uses PSCI v0.2 +) + +// Arm64: Exception Syndrome Register EL1. +const ( + _ESR_ELx_FSC = 0x3F + + _ESR_SEGV_MAPERR_L0 = 0x4 + _ESR_SEGV_MAPERR_L1 = 0x5 + _ESR_SEGV_MAPERR_L2 = 0x6 + _ESR_SEGV_MAPERR_L3 = 0x7 + + _ESR_SEGV_ACCERR_L1 = 0x9 + _ESR_SEGV_ACCERR_L2 = 0xa + _ESR_SEGV_ACCERR_L3 = 0xb + + _ESR_SEGV_PEMERR_L1 = 0xd + _ESR_SEGV_PEMERR_L2 = 0xe + _ESR_SEGV_PEMERR_L3 = 0xf +) diff --git a/pkg/sentry/platform/kvm/machine_amd64_unsafe.go b/pkg/sentry/platform/kvm/machine_amd64_unsafe.go index 61227cafb..7156c245f 100644 --- a/pkg/sentry/platform/kvm/machine_amd64_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_amd64_unsafe.go @@ -135,3 +135,43 @@ func (c *vCPU) setSignalMask() error { } return nil } + +// setUserRegisters sets user registers in the vCPU. +func (c *vCPU) setUserRegisters(uregs *userRegs) error { + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(c.fd), + _KVM_SET_REGS, + uintptr(unsafe.Pointer(uregs))); errno != 0 { + return fmt.Errorf("error setting user registers: %v", errno) + } + return nil +} + +// getUserRegisters reloads user registers in the vCPU. +// +// This is safe to call from a nosplit context. +// +//go:nosplit +func (c *vCPU) getUserRegisters(uregs *userRegs) syscall.Errno { + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(c.fd), + _KVM_GET_REGS, + uintptr(unsafe.Pointer(uregs))); errno != 0 { + return errno + } + return 0 +} + +// setSystemRegisters sets system registers. +func (c *vCPU) setSystemRegisters(sregs *systemRegs) error { + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(c.fd), + _KVM_SET_SREGS, + uintptr(unsafe.Pointer(sregs))); errno != 0 { + return fmt.Errorf("error setting system registers: %v", errno) + } + return nil +} diff --git a/pkg/sentry/platform/kvm/machine_arm64.go b/pkg/sentry/platform/kvm/machine_arm64.go index b7e2cfb9d..7ae47f291 100644 --- a/pkg/sentry/platform/kvm/machine_arm64.go +++ b/pkg/sentry/platform/kvm/machine_arm64.go @@ -12,8 +12,38 @@ // See the License for the specific language governing permissions and // limitations under the License. +// +build arm64 + package kvm +import ( + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/platform" + "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" + "gvisor.dev/gvisor/pkg/sentry/usermem" +) + +type vCPUArchState struct { + // PCIDs is the set of PCIDs for this vCPU. + // + // This starts above fixedKernelPCID. + PCIDs *pagetables.PCIDs +} + +const ( + // fixedKernelPCID is a fixed kernel PCID used for the kernel page + // tables. We must start allocating user PCIDs above this in order to + // avoid any conflict (see below). + fixedKernelPCID = 1 + + // poolPCIDs is the number of PCIDs to record in the database. As this + // grows, assignment can take longer, since it is a simple linear scan. + // Beyond a relatively small number, there are likely few perform + // benefits, since the TLB has likely long since lost any translations + // from more than a few PCIDs past. + poolPCIDs = 8 +) + // Get all read-only physicalRegions. func rdonlyRegionsForSetMem() (phyRegions []physicalRegion) { var rdonlyRegions []region @@ -59,3 +89,95 @@ func availableRegionsForSetMem() (phyRegions []physicalRegion) { return phyRegions } + +// dropPageTables drops cached page table entries. +func (m *machine) dropPageTables(pt *pagetables.PageTables) { + m.mu.Lock() + defer m.mu.Unlock() + + // Clear from all PCIDs. + for _, c := range m.vCPUs { + c.PCIDs.Drop(pt) + } +} + +// nonCanonical generates a canonical address return. +// +//go:nosplit +func nonCanonical(addr uint64, signal int32, info *arch.SignalInfo) (usermem.AccessType, error) { + *info = arch.SignalInfo{ + Signo: signal, + Code: arch.SignalInfoKernel, + } + info.SetAddr(addr) // Include address. + return usermem.NoAccess, platform.ErrContextSignal +} + +// fault generates an appropriate fault return. +// +//go:nosplit +func (c *vCPU) fault(signal int32, info *arch.SignalInfo) (usermem.AccessType, error) { + faultAddr := c.GetFaultAddr() + code, user := c.ErrorCode() + + // Reset the pointed SignalInfo. + *info = arch.SignalInfo{Signo: signal} + info.SetAddr(uint64(faultAddr)) + + read := true + write := false + execute := true + + ret := code & _ESR_ELx_FSC + switch ret { + case _ESR_SEGV_MAPERR_L0, _ESR_SEGV_MAPERR_L1, _ESR_SEGV_MAPERR_L2, _ESR_SEGV_MAPERR_L3: + info.Code = 1 //SEGV_MAPERR + read = false + write = true + execute = false + case _ESR_SEGV_ACCERR_L1, _ESR_SEGV_ACCERR_L2, _ESR_SEGV_ACCERR_L3, _ESR_SEGV_PEMERR_L1, _ESR_SEGV_PEMERR_L2, _ESR_SEGV_PEMERR_L3: + info.Code = 2 // SEGV_ACCERR. + read = true + write = false + execute = false + default: + info.Code = 2 + } + + if !user { + read = true + write = false + execute = true + + } + accessType := usermem.AccessType{ + Read: read, + Write: write, + Execute: execute, + } + + return accessType, platform.ErrContextSignal +} + +// retryInGuest runs the given function in guest mode. +// +// If the function does not complete in guest mode (due to execution of a +// system call due to a GC stall, for example), then it will be retried. The +// given function must be idempotent as a result of the retry mechanism. +func (m *machine) retryInGuest(fn func()) { + c := m.Get() + defer m.Put(c) + for { + c.ClearErrorCode() // See below. + bluepill(c) // Force guest mode. + fn() // Execute the given function. + _, user := c.ErrorCode() + if user { + // If user is set, then we haven't bailed back to host + // mode via a kernel exception or system call. We + // consider the full function to have executed in guest + // mode and we can return. + break + } + } +} diff --git a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go new file mode 100644 index 000000000..3f2f97a6b --- /dev/null +++ b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go @@ -0,0 +1,362 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package kvm + +import ( + "fmt" + "reflect" + "sync/atomic" + "syscall" + "unsafe" + + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/platform" + "gvisor.dev/gvisor/pkg/sentry/platform/ring0" + "gvisor.dev/gvisor/pkg/sentry/usermem" +) + +// setMemoryRegion initializes a region. +// +// This may be called from bluepillHandler, and therefore returns an errno +// directly (instead of wrapping in an error) to avoid allocations. +// +//go:nosplit +func (m *machine) setMemoryRegion(slot int, physical, length, virtual uintptr) syscall.Errno { + userRegion := userMemoryRegion{ + slot: uint32(slot), + flags: 0, + guestPhysAddr: uint64(physical), + memorySize: uint64(length), + userspaceAddr: uint64(virtual), + } + + // Set the region. + _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(m.fd), + _KVM_SET_USER_MEMORY_REGION, + uintptr(unsafe.Pointer(&userRegion))) + return errno +} + +type kvmVcpuInit struct { + target uint32 + features [7]uint32 +} + +var vcpuInit kvmVcpuInit + +// initArchState initializes architecture-specific state. +func (m *machine) initArchState() error { + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(m.fd), + _KVM_ARM_PREFERRED_TARGET, + uintptr(unsafe.Pointer(&vcpuInit))); errno != 0 { + panic(fmt.Sprintf("error setting KVM_ARM_PREFERRED_TARGET failed: %v", errno)) + } + return nil +} + +func getPageWithReflect(p uintptr) []byte { + return (*(*[0xFFFFFF]byte)(unsafe.Pointer(p & ^uintptr(syscall.Getpagesize()-1))))[:syscall.Getpagesize()] +} + +// Work around: move ring0.Vectors() into a specific address with 11-bits alignment. +// +// According to the design documentation of Arm64, +// the start address of exception vector table should be 11-bits aligned. +// Please see the code in linux kernel as reference: arch/arm64/kernel/entry.S +// But, we can't align a function's start address to a specific address by using golang. +// We have raised this question in golang community: +// https://groups.google.com/forum/m/#!topic/golang-dev/RPj90l5x86I +// This function will be removed when golang supports this feature. +// +// There are 2 jobs were implemented in this function: +// 1, move the start address of exception vector table into the specific address. +// 2, modify the offset of each instruction. +func updateVectorTable() { + fromLocation := reflect.ValueOf(ring0.Vectors).Pointer() + offset := fromLocation & (1<<11 - 1) + if offset != 0 { + offset = 1<<11 - offset + } + + toLocation := fromLocation + offset + page := getPageWithReflect(toLocation) + if err := syscall.Mprotect(page, syscall.PROT_READ|syscall.PROT_WRITE|syscall.PROT_EXEC); err != nil { + panic(err) + } + + page = getPageWithReflect(toLocation + 4096) + if err := syscall.Mprotect(page, syscall.PROT_READ|syscall.PROT_WRITE|syscall.PROT_EXEC); err != nil { + panic(err) + } + + // Move exception-vector-table into the specific address. + var entry *uint32 + var entryFrom *uint32 + for i := 1; i <= 0x800; i++ { + entry = (*uint32)(unsafe.Pointer(toLocation + 0x800 - uintptr(i))) + entryFrom = (*uint32)(unsafe.Pointer(fromLocation + 0x800 - uintptr(i))) + *entry = *entryFrom + } + + // The offset from the address of each unconditionally branch is changed. + // We should modify the offset of each instruction. + nums := []uint32{0x0, 0x80, 0x100, 0x180, 0x200, 0x280, 0x300, 0x380, 0x400, 0x480, 0x500, 0x580, 0x600, 0x680, 0x700, 0x780} + for _, num := range nums { + entry = (*uint32)(unsafe.Pointer(toLocation + uintptr(num))) + *entry = *entry - (uint32)(offset/4) + } + + page = getPageWithReflect(toLocation) + if err := syscall.Mprotect(page, syscall.PROT_READ|syscall.PROT_EXEC); err != nil { + panic(err) + } + + page = getPageWithReflect(toLocation + 4096) + if err := syscall.Mprotect(page, syscall.PROT_READ|syscall.PROT_EXEC); err != nil { + panic(err) + } +} + +// initArchState initializes architecture-specific state. +func (c *vCPU) initArchState() error { + var ( + reg kvmOneReg + data uint64 + regGet kvmOneReg + dataGet uint64 + ) + + reg.addr = uint64(reflect.ValueOf(&data).Pointer()) + regGet.addr = uint64(reflect.ValueOf(&dataGet).Pointer()) + + vcpuInit.features[0] |= (1 << _KVM_ARM_VCPU_PSCI_0_2) + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(c.fd), + _KVM_ARM_VCPU_INIT, + uintptr(unsafe.Pointer(&vcpuInit))); errno != 0 { + panic(fmt.Sprintf("error setting KVM_ARM_VCPU_INIT failed: %v", errno)) + } + + // cpacr_el1 + reg.id = _KVM_ARM64_REGS_CPACR_EL1 + data = (_FPEN_NOTRAP << _FPEN_SHIFT) + if err := c.setOneRegister(®); err != nil { + return err + } + + // sctlr_el1 + regGet.id = _KVM_ARM64_REGS_SCTLR_EL1 + if err := c.getOneRegister(®Get); err != nil { + return err + } + + dataGet |= (_SCTLR_M | _SCTLR_C | _SCTLR_I) + data = dataGet + reg.id = _KVM_ARM64_REGS_SCTLR_EL1 + if err := c.setOneRegister(®); err != nil { + return err + } + + // tcr_el1 + data = _TCR_TXSZ_VA48 | _TCR_CACHE_FLAGS | _TCR_SHARED | _TCR_TG_FLAGS | _TCR_ASID16 | _TCR_IPS_40BITS + reg.id = _KVM_ARM64_REGS_TCR_EL1 + if err := c.setOneRegister(®); err != nil { + return err + } + + // mair_el1 + data = _MT_EL1_INIT + reg.id = _KVM_ARM64_REGS_MAIR_EL1 + if err := c.setOneRegister(®); err != nil { + return err + } + + // ttbr0_el1 + data = c.machine.kernel.PageTables.TTBR0_EL1(false, 0) + + reg.id = _KVM_ARM64_REGS_TTBR0_EL1 + if err := c.setOneRegister(®); err != nil { + return err + } + + c.SetTtbr0Kvm(uintptr(data)) + + // ttbr1_el1 + data = c.machine.kernel.PageTables.TTBR1_EL1(false, 0) + + reg.id = _KVM_ARM64_REGS_TTBR1_EL1 + if err := c.setOneRegister(®); err != nil { + return err + } + + // sp_el1 + data = c.CPU.StackTop() + reg.id = _KVM_ARM64_REGS_SP_EL1 + if err := c.setOneRegister(®); err != nil { + return err + } + + // pc + reg.id = _KVM_ARM64_REGS_PC + data = uint64(reflect.ValueOf(ring0.Start).Pointer()) + if err := c.setOneRegister(®); err != nil { + return err + } + + // r8 + reg.id = _KVM_ARM64_REGS_R8 + data = uint64(reflect.ValueOf(&c.CPU).Pointer()) + if err := c.setOneRegister(®); err != nil { + return err + } + + // vbar_el1 + reg.id = _KVM_ARM64_REGS_VBAR_EL1 + + fromLocation := reflect.ValueOf(ring0.Vectors).Pointer() + offset := fromLocation & (1<<11 - 1) + if offset != 0 { + offset = 1<<11 - offset + } + + toLocation := fromLocation + offset + data = uint64(ring0.KernelStartAddress | toLocation) + if err := c.setOneRegister(®); err != nil { + return err + } + + data = ring0.PsrDefaultSet | ring0.KernelFlagsSet + reg.id = _KVM_ARM64_REGS_PSTATE + if err := c.setOneRegister(®); err != nil { + return err + } + + return nil +} + +//go:nosplit +func (c *vCPU) loadSegments(tid uint64) { + // TODO(gvisor.dev/issue/1238): TLS is not supported. + // Get TLS from tpidr_el0. + atomic.StoreUint64(&c.tid, tid) +} + +func (c *vCPU) setOneRegister(reg *kvmOneReg) error { + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(c.fd), + _KVM_SET_ONE_REG, + uintptr(unsafe.Pointer(reg))); errno != 0 { + return fmt.Errorf("error setting one register: %v", errno) + } + return nil +} + +func (c *vCPU) getOneRegister(reg *kvmOneReg) error { + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(c.fd), + _KVM_GET_ONE_REG, + uintptr(unsafe.Pointer(reg))); errno != 0 { + return fmt.Errorf("error setting one register: %v", errno) + } + return nil +} + +// setCPUID sets the CPUID to be used by the guest. +func (c *vCPU) setCPUID() error { + return nil +} + +// setSystemTime sets the TSC for the vCPU. +func (c *vCPU) setSystemTime() error { + return nil +} + +// setSignalMask sets the vCPU signal mask. +// +// This must be called prior to running the vCPU. +func (c *vCPU) setSignalMask() error { + // The layout of this structure implies that it will not necessarily be + // the same layout chosen by the Go compiler. It gets fudged here. + var data struct { + length uint32 + mask1 uint32 + mask2 uint32 + _ uint32 + } + data.length = 8 // Fixed sigset size. + data.mask1 = ^uint32(bounceSignalMask & 0xffffffff) + data.mask2 = ^uint32(bounceSignalMask >> 32) + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(c.fd), + _KVM_SET_SIGNAL_MASK, + uintptr(unsafe.Pointer(&data))); errno != 0 { + return fmt.Errorf("error setting signal mask: %v", errno) + } + + return nil +} + +// SwitchToUser unpacks architectural-details. +func (c *vCPU) SwitchToUser(switchOpts ring0.SwitchOpts, info *arch.SignalInfo) (usermem.AccessType, error) { + // Check for canonical addresses. + if regs := switchOpts.Registers; !ring0.IsCanonical(regs.Pc) { + return nonCanonical(regs.Pc, int32(syscall.SIGSEGV), info) + } else if !ring0.IsCanonical(regs.Sp) { + return nonCanonical(regs.Sp, int32(syscall.SIGBUS), info) + } + + var vector ring0.Vector + ttbr0App := switchOpts.PageTables.TTBR0_EL1(false, 0) + c.SetTtbr0App(uintptr(ttbr0App)) + + // TODO(gvisor.dev/issue/1238): full context-switch supporting for Arm64. + // The Arm64 user-mode execution state consists of: + // x0-x30 + // PC, SP, PSTATE + // V0-V31: 32 128-bit registers for floating point, and simd + // FPSR + // TPIDR_EL0, used for TLS + appRegs := switchOpts.Registers + c.SetAppAddr(ring0.KernelStartAddress | uintptr(unsafe.Pointer(appRegs))) + + entersyscall() + bluepill(c) + vector = c.CPU.SwitchToUser(switchOpts) + exitsyscall() + + switch vector { + case ring0.Syscall: + // Fast path: system call executed. + return usermem.NoAccess, nil + + case ring0.PageFault: + return c.fault(int32(syscall.SIGSEGV), info) + case 0xaa: + return usermem.NoAccess, nil + default: + return usermem.NoAccess, platform.ErrContextSignal + } + +} diff --git a/pkg/sentry/platform/kvm/machine_unsafe.go b/pkg/sentry/platform/kvm/machine_unsafe.go index ed9433311..f04be2ab5 100644 --- a/pkg/sentry/platform/kvm/machine_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_unsafe.go @@ -87,46 +87,6 @@ func unmapRunData(r *runData) error { return nil } -// setUserRegisters sets user registers in the vCPU. -func (c *vCPU) setUserRegisters(uregs *userRegs) error { - if _, _, errno := syscall.RawSyscall( - syscall.SYS_IOCTL, - uintptr(c.fd), - _KVM_SET_REGS, - uintptr(unsafe.Pointer(uregs))); errno != 0 { - return fmt.Errorf("error setting user registers: %v", errno) - } - return nil -} - -// getUserRegisters reloads user registers in the vCPU. -// -// This is safe to call from a nosplit context. -// -//go:nosplit -func (c *vCPU) getUserRegisters(uregs *userRegs) syscall.Errno { - if _, _, errno := syscall.RawSyscall( - syscall.SYS_IOCTL, - uintptr(c.fd), - _KVM_GET_REGS, - uintptr(unsafe.Pointer(uregs))); errno != 0 { - return errno - } - return 0 -} - -// setSystemRegisters sets system registers. -func (c *vCPU) setSystemRegisters(sregs *systemRegs) error { - if _, _, errno := syscall.RawSyscall( - syscall.SYS_IOCTL, - uintptr(c.fd), - _KVM_SET_SREGS, - uintptr(unsafe.Pointer(sregs))); errno != 0 { - return fmt.Errorf("error setting system registers: %v", errno) - } - return nil -} - // atomicAddressSpace is an atomic address space pointer. type atomicAddressSpace struct { pointer unsafe.Pointer -- cgit v1.2.3 From 8a46e83111c70c7ee81368dd4e9df53559213fa0 Mon Sep 17 00:00:00 2001 From: Yong He Date: Sun, 15 Dec 2019 11:20:12 +0800 Subject: Fix UDS bind cause fd leak in gofer After the finalizer optimize in 76039f895995c3fe0deef5958f843868685ecc38 commit, clientFile needs to closed before finalizer release it. The clientFile is not closed if it is created via gofer.(*inodeOperations).Bind, this will cause fd leak which is hold by gofer process. Fixes #1396 Signed-off-by: Yong He Signed-off-by: Jianfeng Tan --- pkg/sentry/fs/gofer/path.go | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/gofer/path.go b/pkg/sentry/fs/gofer/path.go index 8c17603f8..c09f3b71c 100644 --- a/pkg/sentry/fs/gofer/path.go +++ b/pkg/sentry/fs/gofer/path.go @@ -234,6 +234,8 @@ func (i *inodeOperations) Bind(ctx context.Context, dir *fs.Inode, name string, if err != nil { return nil, err } + // We're not going to use newFile after return. + defer newFile.close(ctx) // Stabilize the endpoint map while creation is in progress. unlock := i.session().endpoints.lock() @@ -254,7 +256,6 @@ func (i *inodeOperations) Bind(ctx context.Context, dir *fs.Inode, name string, // Get the attributes of the file to create inode key. qid, mask, attr, err := getattr(ctx, newFile) if err != nil { - newFile.close(ctx) return nil, err } @@ -270,7 +271,6 @@ func (i *inodeOperations) Bind(ctx context.Context, dir *fs.Inode, name string, // cloned and re-opened multiple times after creation. _, unopened, err := i.fileState.file.walk(ctx, []string{name}) if err != nil { - newFile.close(ctx) return nil, err } -- cgit v1.2.3 From bd5c7bf58dd656dc16b920b48a346ef08f1efba8 Mon Sep 17 00:00:00 2001 From: Yong He Date: Mon, 16 Dec 2019 14:04:03 +0800 Subject: Fix deadlock in overlay bind Copy up parent when binding UDS on overlayfs is supported in commit 02ab1f187cd24c67b754b004229421d189cee264. But the using of copyUp in overlayBind will cause sentry stuck, reason is dead lock in renameMu. 1 [Process A] Invoke a Unix socket bind operation renameMu is hold in fs.(*Dirent).genericCreate by process A 2 [Process B] Invoke a read syscall on /proc/task/mounts waitng on Lock of renameMu in fs.(*MountNamespace).FindMount 3 [Process A] Continue Unix socket bind operation wating on RLock of renameMu in fs.copyUp Root cause is recursive reading lock of reanmeMu in bind call trace, if there are writing lock between the two reading lock, then deadlock occured. Fixes #1397 --- pkg/sentry/fs/inode_overlay.go | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/inode_overlay.go b/pkg/sentry/fs/inode_overlay.go index a09147080..8d4303cc4 100644 --- a/pkg/sentry/fs/inode_overlay.go +++ b/pkg/sentry/fs/inode_overlay.go @@ -436,7 +436,7 @@ func overlayRename(ctx context.Context, o *overlayEntry, oldParent *Dirent, rena } func overlayBind(ctx context.Context, o *overlayEntry, parent *Dirent, name string, data transport.BoundEndpoint, perm FilePermissions) (*Dirent, error) { - if err := copyUp(ctx, parent); err != nil { + if err := copyUpLockedForRename(ctx, parent); err != nil { return nil, err } -- cgit v1.2.3 From e6f4124afd951c3b089f9c75c499c14f4d90a590 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Mon, 16 Dec 2019 13:18:36 -0800 Subject: Implement checks for get/setxattr at the syscall layer. Add checks for input arguments, file type, permissions, etc. that match the Linux implementation. A call to get/setxattr that passes all the checks will still currently return EOPNOTSUPP. Actual support will be added in following commits. Only allow user.* extended attributes for the time being. PiperOrigin-RevId: 285835159 --- pkg/abi/linux/BUILD | 1 + pkg/abi/linux/xattr.go | 27 +++++ pkg/sentry/fs/inode.go | 8 ++ pkg/sentry/fs/inode_overlay.go | 5 + pkg/sentry/syscalls/linux/BUILD | 1 + pkg/sentry/syscalls/linux/linux64_amd64.go | 4 +- pkg/sentry/syscalls/linux/linux64_arm64.go | 4 +- pkg/sentry/syscalls/linux/sys_xattr.go | 169 +++++++++++++++++++++++++++++ test/syscalls/linux/xattr.cc | 79 +++++++------- 9 files changed, 253 insertions(+), 45 deletions(-) create mode 100644 pkg/abi/linux/xattr.go create mode 100644 pkg/sentry/syscalls/linux/sys_xattr.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/BUILD b/pkg/abi/linux/BUILD index 51774c6b6..9553f164d 100644 --- a/pkg/abi/linux/BUILD +++ b/pkg/abi/linux/BUILD @@ -57,6 +57,7 @@ go_library( "uio.go", "utsname.go", "wait.go", + "xattr.go", ], importpath = "gvisor.dev/gvisor/pkg/abi/linux", visibility = ["//visibility:public"], diff --git a/pkg/abi/linux/xattr.go b/pkg/abi/linux/xattr.go new file mode 100644 index 000000000..a3b6406fa --- /dev/null +++ b/pkg/abi/linux/xattr.go @@ -0,0 +1,27 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package linux + +// Constants for extended attributes. +const ( + XATTR_NAME_MAX = 255 + XATTR_SIZE_MAX = 65536 + + XATTR_CREATE = 1 + XATTR_REPLACE = 2 + + XATTR_USER_PREFIX = "user." + XATTR_USER_PREFIX_LEN = len(XATTR_USER_PREFIX) +) diff --git a/pkg/sentry/fs/inode.go b/pkg/sentry/fs/inode.go index 2d43dff1d..91e2fde2f 100644 --- a/pkg/sentry/fs/inode.go +++ b/pkg/sentry/fs/inode.go @@ -270,6 +270,14 @@ func (i *Inode) Getxattr(name string) (string, error) { return i.InodeOperations.Getxattr(i, name) } +// Setxattr calls i.InodeOperations.Setxattr with i as the Inode. +func (i *Inode) Setxattr(name, value string) error { + if i.overlay != nil { + return overlaySetxattr(i.overlay, name, value) + } + return i.InodeOperations.Setxattr(i, name, value) +} + // Listxattr calls i.InodeOperations.Listxattr with i as the Inode. func (i *Inode) Listxattr() (map[string]struct{}, error) { if i.overlay != nil { diff --git a/pkg/sentry/fs/inode_overlay.go b/pkg/sentry/fs/inode_overlay.go index a09147080..63a991beb 100644 --- a/pkg/sentry/fs/inode_overlay.go +++ b/pkg/sentry/fs/inode_overlay.go @@ -552,6 +552,11 @@ func overlayGetxattr(o *overlayEntry, name string) (string, error) { return s, err } +// TODO(b/146028302): Support setxattr for overlayfs. +func overlaySetxattr(o *overlayEntry, name, value string) error { + return syserror.EOPNOTSUPP +} + func overlayListxattr(o *overlayEntry) (map[string]struct{}, error) { o.copyMu.RLock() defer o.copyMu.RUnlock() diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD index 4c0bf96e4..6766ba587 100644 --- a/pkg/sentry/syscalls/linux/BUILD +++ b/pkg/sentry/syscalls/linux/BUILD @@ -49,6 +49,7 @@ go_library( "sys_tls.go", "sys_utsname.go", "sys_write.go", + "sys_xattr.go", "timespec.go", ], importpath = "gvisor.dev/gvisor/pkg/sentry/syscalls/linux", diff --git a/pkg/sentry/syscalls/linux/linux64_amd64.go b/pkg/sentry/syscalls/linux/linux64_amd64.go index 797542d28..272ae9991 100644 --- a/pkg/sentry/syscalls/linux/linux64_amd64.go +++ b/pkg/sentry/syscalls/linux/linux64_amd64.go @@ -228,10 +228,10 @@ var AMD64 = &kernel.SyscallTable{ 185: syscalls.Error("security", syserror.ENOSYS, "Not implemented in Linux.", nil), 186: syscalls.Supported("gettid", Gettid), 187: syscalls.Supported("readahead", Readahead), - 188: syscalls.Error("setxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 188: syscalls.PartiallySupported("setxattr", Setxattr, "Only supported for tmpfs.", nil), 189: syscalls.Error("lsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 190: syscalls.Error("fsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 191: syscalls.ErrorWithEvent("getxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 191: syscalls.PartiallySupported("getxattr", Getxattr, "Only supported for tmpfs.", nil), 192: syscalls.ErrorWithEvent("lgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 193: syscalls.ErrorWithEvent("fgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 194: syscalls.ErrorWithEvent("listxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), diff --git a/pkg/sentry/syscalls/linux/linux64_arm64.go b/pkg/sentry/syscalls/linux/linux64_arm64.go index 2bc7faff5..3b584eed9 100644 --- a/pkg/sentry/syscalls/linux/linux64_arm64.go +++ b/pkg/sentry/syscalls/linux/linux64_arm64.go @@ -41,10 +41,10 @@ var ARM64 = &kernel.SyscallTable{ 2: syscalls.PartiallySupported("io_submit", IoSubmit, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), 3: syscalls.PartiallySupported("io_cancel", IoCancel, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), 4: syscalls.PartiallySupported("io_getevents", IoGetevents, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 5: syscalls.Error("setxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 5: syscalls.PartiallySupported("setxattr", Setxattr, "Only supported for tmpfs.", nil), 6: syscalls.Error("lsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 7: syscalls.Error("fsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 8: syscalls.ErrorWithEvent("getxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 8: syscalls.PartiallySupported("getxattr", Getxattr, "Only supported for tmpfs.", nil), 9: syscalls.ErrorWithEvent("lgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 10: syscalls.ErrorWithEvent("fgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 11: syscalls.ErrorWithEvent("listxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), diff --git a/pkg/sentry/syscalls/linux/sys_xattr.go b/pkg/sentry/syscalls/linux/sys_xattr.go new file mode 100644 index 000000000..97d9a65ea --- /dev/null +++ b/pkg/sentry/syscalls/linux/sys_xattr.go @@ -0,0 +1,169 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package linux + +import ( + "strings" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/syserror" +) + +// Getxattr implements linux syscall getxattr(2). +func Getxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + pathAddr := args[0].Pointer() + nameAddr := args[1].Pointer() + valueAddr := args[2].Pointer() + size := args[3].SizeT() + + path, dirPath, err := copyInPath(t, pathAddr, false /* allowEmpty */) + if err != nil { + return 0, nil, err + } + + valueLen := 0 + err = fileOpOn(t, linux.AT_FDCWD, path, true /* resolve */, func(root *fs.Dirent, d *fs.Dirent, _ uint) error { + value, err := getxattr(t, d, dirPath, nameAddr) + if err != nil { + return err + } + + valueLen = len(value) + if size == 0 { + return nil + } + if size > linux.XATTR_SIZE_MAX { + size = linux.XATTR_SIZE_MAX + } + if valueLen > int(size) { + return syserror.ERANGE + } + + _, err = t.CopyOutBytes(valueAddr, []byte(value)) + return err + }) + if err != nil { + return 0, nil, err + } + return uintptr(valueLen), nil, nil +} + +// getxattr implements getxattr from the given *fs.Dirent. +func getxattr(t *kernel.Task, d *fs.Dirent, dirPath bool, nameAddr usermem.Addr) (string, error) { + if dirPath && !fs.IsDir(d.Inode.StableAttr) { + return "", syserror.ENOTDIR + } + + if err := checkXattrPermissions(t, d.Inode, fs.PermMask{Read: true}); err != nil { + return "", err + } + + name, err := copyInXattrName(t, nameAddr) + if err != nil { + return "", err + } + + if !strings.HasPrefix(name, linux.XATTR_USER_PREFIX) { + return "", syserror.EOPNOTSUPP + } + + return d.Inode.Getxattr(name) +} + +// Setxattr implements linux syscall setxattr(2). +func Setxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + pathAddr := args[0].Pointer() + nameAddr := args[1].Pointer() + valueAddr := args[2].Pointer() + size := args[3].SizeT() + flags := args[4].Uint() + + path, dirPath, err := copyInPath(t, pathAddr, false /* allowEmpty */) + if err != nil { + return 0, nil, err + } + + if flags&^(linux.XATTR_CREATE|linux.XATTR_REPLACE) != 0 { + return 0, nil, syserror.EINVAL + } + + return 0, nil, fileOpOn(t, linux.AT_FDCWD, path, true /* resolve */, func(root *fs.Dirent, d *fs.Dirent, _ uint) error { + return setxattr(t, d, dirPath, nameAddr, valueAddr, size, flags) + }) +} + +// setxattr implements setxattr from the given *fs.Dirent. +func setxattr(t *kernel.Task, d *fs.Dirent, dirPath bool, nameAddr, valueAddr usermem.Addr, size uint, flags uint32) error { + if dirPath && !fs.IsDir(d.Inode.StableAttr) { + return syserror.ENOTDIR + } + + if err := checkXattrPermissions(t, d.Inode, fs.PermMask{Write: true}); err != nil { + return err + } + + name, err := copyInXattrName(t, nameAddr) + if err != nil { + return err + } + + if size > linux.XATTR_SIZE_MAX { + return syserror.E2BIG + } + buf := make([]byte, size) + if _, err = t.CopyInBytes(valueAddr, buf); err != nil { + return err + } + value := string(buf) + + if !strings.HasPrefix(name, linux.XATTR_USER_PREFIX) { + return syserror.EOPNOTSUPP + } + + return d.Inode.Setxattr(name, value) +} + +func copyInXattrName(t *kernel.Task, nameAddr usermem.Addr) (string, error) { + name, err := t.CopyInString(nameAddr, linux.XATTR_NAME_MAX+1) + if err != nil { + if err == syserror.ENAMETOOLONG { + return "", syserror.ERANGE + } + return "", err + } + if len(name) == 0 { + return "", syserror.ERANGE + } + return name, nil +} + +func checkXattrPermissions(t *kernel.Task, i *fs.Inode, perms fs.PermMask) error { + // Restrict xattrs to regular files and directories. + // + // In Linux, this restriction technically only applies to xattrs in the + // "user.*" namespace, but we don't allow any other xattr prefixes anyway. + if !fs.IsRegular(i.StableAttr) && !fs.IsDir(i.StableAttr) { + if perms.Write { + return syserror.EPERM + } + return syserror.ENODATA + } + + return i.CheckPermission(t, perms) +} diff --git a/test/syscalls/linux/xattr.cc b/test/syscalls/linux/xattr.cc index 3e07b634b..75740238c 100644 --- a/test/syscalls/linux/xattr.cc +++ b/test/syscalls/linux/xattr.cc @@ -28,6 +28,7 @@ #include "test/util/capability_util.h" #include "test/util/posix_error.h" #include "test/util/temp_path.h" +#include "test/util/test_util.h" namespace gvisor { namespace testing { @@ -37,9 +38,6 @@ namespace { class XattrTest : public FileTest {}; TEST_F(XattrTest, XattrNullName) { - // TODO(b/127675828): Support setxattr and getxattr. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); EXPECT_THAT(setxattr(path, nullptr, nullptr, 0, /*flags=*/0), @@ -49,9 +47,6 @@ TEST_F(XattrTest, XattrNullName) { } TEST_F(XattrTest, XattrEmptyName) { - // TODO(b/127675828): Support setxattr and getxattr. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); EXPECT_THAT(setxattr(path, "", nullptr, 0, /*flags=*/0), @@ -60,16 +55,17 @@ TEST_F(XattrTest, XattrEmptyName) { } TEST_F(XattrTest, XattrLargeName) { - // TODO(b/127675828): Support setxattr and getxattr. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); std::string name = "user."; name += std::string(XATTR_NAME_MAX - name.length(), 'a'); - EXPECT_THAT(setxattr(path, name.c_str(), nullptr, 0, /*flags=*/0), - SyscallSucceeds()); - EXPECT_THAT(getxattr(path, name.c_str(), nullptr, 0), - SyscallSucceedsWithValue(0)); + + // TODO(b/127675828): Support setxattr and getxattr. + if (!IsRunningOnGvisor()) { + EXPECT_THAT(setxattr(path, name.c_str(), nullptr, 0, /*flags=*/0), + SyscallSucceeds()); + EXPECT_THAT(getxattr(path, name.c_str(), nullptr, 0), + SyscallSucceedsWithValue(0)); + } name += "a"; EXPECT_THAT(setxattr(path, name.c_str(), nullptr, 0, /*flags=*/0), @@ -79,9 +75,6 @@ TEST_F(XattrTest, XattrLargeName) { } TEST_F(XattrTest, XattrInvalidPrefix) { - // TODO(b/127675828): Support setxattr and getxattr. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); std::string name(XATTR_NAME_MAX, 'a'); EXPECT_THAT(setxattr(path, name.c_str(), nullptr, 0, /*flags=*/0), @@ -91,9 +84,6 @@ TEST_F(XattrTest, XattrInvalidPrefix) { } TEST_F(XattrTest, XattrReadOnly) { - // TODO(b/127675828): Support setxattr and getxattr. - SKIP_IF(IsRunningOnGvisor()); - // Drop capabilities that allow us to override file and directory permissions. ASSERT_NO_ERRNO(SetCapability(CAP_DAC_OVERRIDE, false)); ASSERT_NO_ERRNO(SetCapability(CAP_DAC_READ_SEARCH, false)); @@ -102,22 +92,28 @@ TEST_F(XattrTest, XattrReadOnly) { char name[] = "user.abc"; char val = 'a'; size_t size = sizeof(val); - EXPECT_THAT(setxattr(path, name, &val, size, /*flags=*/0), SyscallSucceeds()); + + // TODO(b/127675828): Support setxattr and getxattr. + if (!IsRunningOnGvisor()) { + EXPECT_THAT(setxattr(path, name, &val, size, /*flags=*/0), + SyscallSucceeds()); + } ASSERT_NO_ERRNO(testing::Chmod(test_file_name_, S_IRUSR)); EXPECT_THAT(setxattr(path, name, &val, size, /*flags=*/0), SyscallFailsWithErrno(EACCES)); - char buf = '-'; - EXPECT_THAT(getxattr(path, name, &buf, size), SyscallSucceedsWithValue(size)); - EXPECT_EQ(buf, val); + // TODO(b/127675828): Support setxattr and getxattr. + if (!IsRunningOnGvisor()) { + char buf = '-'; + EXPECT_THAT(getxattr(path, name, &buf, size), + SyscallSucceedsWithValue(size)); + EXPECT_EQ(buf, val); + } } TEST_F(XattrTest, XattrWriteOnly) { - // TODO(b/127675828): Support setxattr and getxattr. - SKIP_IF(IsRunningOnGvisor()); - // Drop capabilities that allow us to override file and directory permissions. ASSERT_NO_ERRNO(SetCapability(CAP_DAC_OVERRIDE, false)); ASSERT_NO_ERRNO(SetCapability(CAP_DAC_READ_SEARCH, false)); @@ -128,7 +124,12 @@ TEST_F(XattrTest, XattrWriteOnly) { char name[] = "user.abc"; char val = 'a'; size_t size = sizeof(val); - EXPECT_THAT(setxattr(path, name, &val, size, /*flags=*/0), SyscallSucceeds()); + + // TODO(b/127675828): Support setxattr and getxattr. + if (!IsRunningOnGvisor()) { + EXPECT_THAT(setxattr(path, name, &val, size, /*flags=*/0), + SyscallSucceeds()); + } EXPECT_THAT(getxattr(path, name, nullptr, 0), SyscallFailsWithErrno(EACCES)); } @@ -172,9 +173,6 @@ TEST_F(XattrTest, XattrOnSymlink) { } TEST_F(XattrTest, XattrOnInvalidFileTypes) { - // TODO(b/127675828): Support setxattr and getxattr. - SKIP_IF(IsRunningOnGvisor()); - char name[] = "user.abc"; char char_device[] = "/dev/zero"; @@ -226,9 +224,6 @@ TEST_F(XattrTest, SetxattrZeroSize) { } TEST_F(XattrTest, SetxattrSizeTooLarge) { - // TODO(b/127675828): Support setxattr and getxattr. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); char name[] = "user.abc"; @@ -240,19 +235,24 @@ TEST_F(XattrTest, SetxattrSizeTooLarge) { EXPECT_THAT(setxattr(path, name, val.data(), size, /*flags=*/0), SyscallFailsWithErrno(E2BIG)); - EXPECT_THAT(getxattr(path, name, nullptr, 0), SyscallFailsWithErrno(ENODATA)); + // TODO(b/127675828): Support setxattr and getxattr. + if (!IsRunningOnGvisor()) { + EXPECT_THAT(getxattr(path, name, nullptr, 0), + SyscallFailsWithErrno(ENODATA)); + } } TEST_F(XattrTest, SetxattrNullValueAndNonzeroSize) { - // TODO(b/127675828): Support setxattr and getxattr. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); char name[] = "user.abc"; EXPECT_THAT(setxattr(path, name, nullptr, 1, /*flags=*/0), SyscallFailsWithErrno(EFAULT)); - EXPECT_THAT(getxattr(path, name, nullptr, 0), SyscallFailsWithErrno(ENODATA)); + // TODO(b/127675828): Support setxattr and getxattr. + if (!IsRunningOnGvisor()) { + EXPECT_THAT(getxattr(path, name, nullptr, 0), + SyscallFailsWithErrno(ENODATA)); + } } TEST_F(XattrTest, SetxattrNullValueAndZeroSize) { @@ -350,9 +350,6 @@ TEST_F(XattrTest, SetxattrReplaceFlag) { } TEST_F(XattrTest, SetxattrInvalidFlags) { - // TODO(b/127675828): Support setxattr and getxattr. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); int invalid_flags = 0xff; EXPECT_THAT(setxattr(path, nullptr, nullptr, 0, invalid_flags), -- cgit v1.2.3 From 3193b2fff8149fe43a3a59c266359e7f443a1563 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Mon, 16 Dec 2019 14:40:01 -0800 Subject: Drop unnecessary cast. Bitshift operators with signed int is supported in Go 1.13. PiperOrigin-RevId: 285853622 --- go.mod | 32 ++++++++++++++++---------------- pkg/sentry/strace/select.go | 4 ++-- pkg/sentry/syscalls/linux/sys_poll.go | 4 ++-- 3 files changed, 20 insertions(+), 20 deletions(-) (limited to 'pkg/sentry') diff --git a/go.mod b/go.mod index 821273d22..304b8bf13 100644 --- a/go.mod +++ b/go.mod @@ -1,21 +1,21 @@ module gvisor.dev/gvisor -go 1.12 +go 1.13 require ( - github.com/cenkalti/backoff v0.0.0-20190506075156-2146c9339422 - github.com/gofrs/flock v0.6.1-0.20180915234121-886344bea079 - github.com/golang/mock v1.3.1 - github.com/golang/protobuf v1.3.1 - github.com/google/btree v1.0.0 - github.com/google/go-cmp v0.2.0 - github.com/google/subcommands v0.0.0-20190508160503-636abe8753b8 - github.com/google/uuid v0.0.0-20171129191014-dec09d789f3d - github.com/kr/pty v1.1.1 - github.com/opencontainers/runtime-spec v0.1.2-0.20171211145439-b2d941ef6a78 - github.com/syndtr/gocapability v0.0.0-20180916011248-d98352740cb2 - github.com/vishvananda/netlink v1.0.1-0.20190318003149-adb577d4a45e - github.com/vishvananda/netns v0.0.0-20171111001504-be1fbeda1936 - golang.org/x/net v0.0.0-20190311183353-d8887717615a - golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a + github.com/cenkalti/backoff v0.0.0-20190506075156-2146c9339422 + github.com/gofrs/flock v0.6.1-0.20180915234121-886344bea079 + github.com/golang/mock v1.3.1 + github.com/golang/protobuf v1.3.1 + github.com/google/btree v1.0.0 + github.com/google/go-cmp v0.2.0 + github.com/google/subcommands v0.0.0-20190508160503-636abe8753b8 + github.com/google/uuid v0.0.0-20171129191014-dec09d789f3d + github.com/kr/pty v1.1.1 + github.com/opencontainers/runtime-spec v0.1.2-0.20171211145439-b2d941ef6a78 + github.com/syndtr/gocapability v0.0.0-20180916011248-d98352740cb2 + github.com/vishvananda/netlink v1.0.1-0.20190318003149-adb577d4a45e + github.com/vishvananda/netns v0.0.0-20171111001504-be1fbeda1936 + golang.org/x/net v0.0.0-20190311183353-d8887717615a + golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a ) diff --git a/pkg/sentry/strace/select.go b/pkg/sentry/strace/select.go index 92c18083d..dea309fda 100644 --- a/pkg/sentry/strace/select.go +++ b/pkg/sentry/strace/select.go @@ -27,7 +27,7 @@ func fdsFromSet(t *kernel.Task, set []byte) []int { // Append n if the n-th bit is 1. for i, v := range set { for j := 0; j < 8; j++ { - if (v>>uint(j))&1 == 1 { + if (v>>j)&1 == 1 { fds = append(fds, i*8+j) } } @@ -42,7 +42,7 @@ func fdSet(t *kernel.Task, nfds int, addr usermem.Addr) string { // Calculate the size of the fd set (one bit per fd). nBytes := (nfds + 7) / 8 - nBitsInLastPartialByte := uint(nfds % 8) + nBitsInLastPartialByte := nfds % 8 set, err := linux.CopyInFDSet(t, addr, nBytes, nBitsInLastPartialByte) if err != nil { diff --git a/pkg/sentry/syscalls/linux/sys_poll.go b/pkg/sentry/syscalls/linux/sys_poll.go index 631dffec6..2b2df989a 100644 --- a/pkg/sentry/syscalls/linux/sys_poll.go +++ b/pkg/sentry/syscalls/linux/sys_poll.go @@ -198,7 +198,7 @@ func doPoll(t *kernel.Task, addr usermem.Addr, nfds uint, timeout time.Duration) } // CopyInFDSet copies an fd set from select(2)/pselect(2). -func CopyInFDSet(t *kernel.Task, addr usermem.Addr, nBytes int, nBitsInLastPartialByte uint) ([]byte, error) { +func CopyInFDSet(t *kernel.Task, addr usermem.Addr, nBytes, nBitsInLastPartialByte int) ([]byte, error) { set := make([]byte, nBytes) if addr != 0 { @@ -222,7 +222,7 @@ func doSelect(t *kernel.Task, nfds int, readFDs, writeFDs, exceptFDs usermem.Add // Calculate the size of the fd sets (one bit per fd). nBytes := (nfds + 7) / 8 - nBitsInLastPartialByte := uint(nfds % 8) + nBitsInLastPartialByte := nfds % 8 // Capture all the provided input vectors. r, err := CopyInFDSet(t, readFDs, nBytes, nBitsInLastPartialByte) -- cgit v1.2.3 From 91f1ac731933ac1fe0f9ef30f4c9d06fa4031021 Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Tue, 17 Dec 2019 13:57:39 -0800 Subject: Mark enableCpuidFault nosplit This is called after fork, so it must be nosplit. Updates #1408 PiperOrigin-RevId: 286053054 --- pkg/sentry/platform/ptrace/subprocess_amd64.go | 10 ++++++++-- pkg/sentry/platform/ptrace/subprocess_arm64.go | 2 ++ 2 files changed, 10 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/ptrace/subprocess_amd64.go b/pkg/sentry/platform/ptrace/subprocess_amd64.go index a55cff507..606dc2b1d 100644 --- a/pkg/sentry/platform/ptrace/subprocess_amd64.go +++ b/pkg/sentry/platform/ptrace/subprocess_amd64.go @@ -167,8 +167,14 @@ func patchSignalInfo(regs *syscall.PtraceRegs, signalInfo *arch.SignalInfo) { } } -// enableCpuidFault enable cpuid-faulting; this may fail on older kernels or hardware, -// so we just disregard the result. Host CPUID will be enabled. +// enableCpuidFault enables cpuid-faulting. +// +// This may fail on older kernels or hardware, so we just disregard the result. +// Host CPUID will be enabled. +// +// This is safe to call in an afterFork context. +// +//go:nosplit func enableCpuidFault() { syscall.RawSyscall6(syscall.SYS_ARCH_PRCTL, linux.ARCH_SET_CPUID, 0, 0, 0, 0, 0) } diff --git a/pkg/sentry/platform/ptrace/subprocess_arm64.go b/pkg/sentry/platform/ptrace/subprocess_arm64.go index aed34e7ee..62a686ee7 100644 --- a/pkg/sentry/platform/ptrace/subprocess_arm64.go +++ b/pkg/sentry/platform/ptrace/subprocess_arm64.go @@ -151,6 +151,8 @@ func patchSignalInfo(regs *syscall.PtraceRegs, signalInfo *arch.SignalInfo) { } // Noop on arm64. +// +//go:nosplit func enableCpuidFault() { } -- cgit v1.2.3 From cb533f18cbb93e3f236ba191d1693e93716313b5 Mon Sep 17 00:00:00 2001 From: Haibo Xu Date: Mon, 18 Nov 2019 09:34:02 +0000 Subject: Enable pkg/sentry/strace support on arm64. Signed-off-by: Haibo Xu Change-Id: I006a1845b6aab2c2fdb9d80fffc1868a6a132ecd --- pkg/sentry/strace/BUILD | 3 +- pkg/sentry/strace/linux64.go | 367 ----------------------------------- pkg/sentry/strace/linux64_amd64.go | 382 +++++++++++++++++++++++++++++++++++++ pkg/sentry/strace/linux64_arm64.go | 323 +++++++++++++++++++++++++++++++ pkg/sentry/strace/syscalls.go | 9 +- 5 files changed, 708 insertions(+), 376 deletions(-) delete mode 100644 pkg/sentry/strace/linux64.go create mode 100644 pkg/sentry/strace/linux64_amd64.go create mode 100644 pkg/sentry/strace/linux64_arm64.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/strace/BUILD b/pkg/sentry/strace/BUILD index d46421199..aa1ac720c 100644 --- a/pkg/sentry/strace/BUILD +++ b/pkg/sentry/strace/BUILD @@ -10,7 +10,8 @@ go_library( "capability.go", "clone.go", "futex.go", - "linux64.go", + "linux64_amd64.go", + "linux64_arm64.go", "open.go", "poll.go", "ptrace.go", diff --git a/pkg/sentry/strace/linux64.go b/pkg/sentry/strace/linux64.go deleted file mode 100644 index e603f858f..000000000 --- a/pkg/sentry/strace/linux64.go +++ /dev/null @@ -1,367 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package strace - -// linuxAMD64 provides a mapping of the Linux amd64 syscalls and their argument -// types for display / formatting. -var linuxAMD64 = SyscallMap{ - 0: makeSyscallInfo("read", FD, ReadBuffer, Hex), - 1: makeSyscallInfo("write", FD, WriteBuffer, Hex), - 2: makeSyscallInfo("open", Path, OpenFlags, Mode), - 3: makeSyscallInfo("close", FD), - 4: makeSyscallInfo("stat", Path, Stat), - 5: makeSyscallInfo("fstat", FD, Stat), - 6: makeSyscallInfo("lstat", Path, Stat), - 7: makeSyscallInfo("poll", PollFDs, Hex, Hex), - 8: makeSyscallInfo("lseek", Hex, Hex, Hex), - 9: makeSyscallInfo("mmap", Hex, Hex, Hex, Hex, FD, Hex), - 10: makeSyscallInfo("mprotect", Hex, Hex, Hex), - 11: makeSyscallInfo("munmap", Hex, Hex), - 12: makeSyscallInfo("brk", Hex), - 13: makeSyscallInfo("rt_sigaction", Signal, SigAction, PostSigAction), - 14: makeSyscallInfo("rt_sigprocmask", SignalMaskAction, SigSet, PostSigSet, Hex), - 15: makeSyscallInfo("rt_sigreturn"), - 16: makeSyscallInfo("ioctl", FD, Hex, Hex), - 17: makeSyscallInfo("pread64", FD, ReadBuffer, Hex, Hex), - 18: makeSyscallInfo("pwrite64", FD, WriteBuffer, Hex, Hex), - 19: makeSyscallInfo("readv", FD, ReadIOVec, Hex), - 20: makeSyscallInfo("writev", FD, WriteIOVec, Hex), - 21: makeSyscallInfo("access", Path, Oct), - 22: makeSyscallInfo("pipe", PipeFDs), - 23: makeSyscallInfo("select", Hex, SelectFDSet, SelectFDSet, SelectFDSet, Timeval), - 24: makeSyscallInfo("sched_yield"), - 25: makeSyscallInfo("mremap", Hex, Hex, Hex, Hex, Hex), - 26: makeSyscallInfo("msync", Hex, Hex, Hex), - 27: makeSyscallInfo("mincore", Hex, Hex, Hex), - 28: makeSyscallInfo("madvise", Hex, Hex, Hex), - 29: makeSyscallInfo("shmget", Hex, Hex, Hex), - 30: makeSyscallInfo("shmat", Hex, Hex, Hex), - 31: makeSyscallInfo("shmctl", Hex, Hex, Hex), - 32: makeSyscallInfo("dup", FD), - 33: makeSyscallInfo("dup2", FD, FD), - 34: makeSyscallInfo("pause"), - 35: makeSyscallInfo("nanosleep", Timespec, PostTimespec), - 36: makeSyscallInfo("getitimer", ItimerType, PostItimerVal), - 37: makeSyscallInfo("alarm", Hex), - 38: makeSyscallInfo("setitimer", ItimerType, ItimerVal, PostItimerVal), - 39: makeSyscallInfo("getpid"), - 40: makeSyscallInfo("sendfile", FD, FD, Hex, Hex), - 41: makeSyscallInfo("socket", SockFamily, SockType, SockProtocol), - 42: makeSyscallInfo("connect", FD, SockAddr, Hex), - 43: makeSyscallInfo("accept", FD, PostSockAddr, SockLen), - 44: makeSyscallInfo("sendto", FD, Hex, Hex, Hex, SockAddr, Hex), - 45: makeSyscallInfo("recvfrom", FD, Hex, Hex, Hex, PostSockAddr, SockLen), - 46: makeSyscallInfo("sendmsg", FD, SendMsgHdr, Hex), - 47: makeSyscallInfo("recvmsg", FD, RecvMsgHdr, Hex), - 48: makeSyscallInfo("shutdown", FD, Hex), - 49: makeSyscallInfo("bind", FD, SockAddr, Hex), - 50: makeSyscallInfo("listen", FD, Hex), - 51: makeSyscallInfo("getsockname", FD, PostSockAddr, SockLen), - 52: makeSyscallInfo("getpeername", FD, PostSockAddr, SockLen), - 53: makeSyscallInfo("socketpair", SockFamily, SockType, SockProtocol, Hex), - 54: makeSyscallInfo("setsockopt", FD, Hex, Hex, Hex, Hex), - 55: makeSyscallInfo("getsockopt", FD, Hex, Hex, Hex, Hex), - 56: makeSyscallInfo("clone", CloneFlags, Hex, Hex, Hex, Hex), - 57: makeSyscallInfo("fork"), - 58: makeSyscallInfo("vfork"), - 59: makeSyscallInfo("execve", Path, ExecveStringVector, ExecveStringVector), - 60: makeSyscallInfo("exit", Hex), - 61: makeSyscallInfo("wait4", Hex, Hex, Hex, Rusage), - 62: makeSyscallInfo("kill", Hex, Signal), - 63: makeSyscallInfo("uname", Uname), - 64: makeSyscallInfo("semget", Hex, Hex, Hex), - 65: makeSyscallInfo("semop", Hex, Hex, Hex), - 66: makeSyscallInfo("semctl", Hex, Hex, Hex, Hex), - 67: makeSyscallInfo("shmdt", Hex), - 68: makeSyscallInfo("msgget", Hex, Hex), - 69: makeSyscallInfo("msgsnd", Hex, Hex, Hex, Hex), - 70: makeSyscallInfo("msgrcv", Hex, Hex, Hex, Hex, Hex), - 71: makeSyscallInfo("msgctl", Hex, Hex, Hex), - 72: makeSyscallInfo("fcntl", FD, Hex, Hex), - 73: makeSyscallInfo("flock", FD, Hex), - 74: makeSyscallInfo("fsync", FD), - 75: makeSyscallInfo("fdatasync", FD), - 76: makeSyscallInfo("truncate", Path, Hex), - 77: makeSyscallInfo("ftruncate", FD, Hex), - 78: makeSyscallInfo("getdents", FD, Hex, Hex), - 79: makeSyscallInfo("getcwd", PostPath, Hex), - 80: makeSyscallInfo("chdir", Path), - 81: makeSyscallInfo("fchdir", FD), - 82: makeSyscallInfo("rename", Path, Path), - 83: makeSyscallInfo("mkdir", Path, Oct), - 84: makeSyscallInfo("rmdir", Path), - 85: makeSyscallInfo("creat", Path, Oct), - 86: makeSyscallInfo("link", Path, Path), - 87: makeSyscallInfo("unlink", Path), - 88: makeSyscallInfo("symlink", Path, Path), - 89: makeSyscallInfo("readlink", Path, ReadBuffer, Hex), - 90: makeSyscallInfo("chmod", Path, Mode), - 91: makeSyscallInfo("fchmod", FD, Mode), - 92: makeSyscallInfo("chown", Path, Hex, Hex), - 93: makeSyscallInfo("fchown", FD, Hex, Hex), - 94: makeSyscallInfo("lchown", Path, Hex, Hex), - 95: makeSyscallInfo("umask", Hex), - 96: makeSyscallInfo("gettimeofday", Timeval, Hex), - 97: makeSyscallInfo("getrlimit", Hex, Hex), - 98: makeSyscallInfo("getrusage", Hex, Rusage), - 99: makeSyscallInfo("sysinfo", Hex), - 100: makeSyscallInfo("times", Hex), - 101: makeSyscallInfo("ptrace", PtraceRequest, Hex, Hex, Hex), - 102: makeSyscallInfo("getuid"), - 103: makeSyscallInfo("syslog", Hex, Hex, Hex), - 104: makeSyscallInfo("getgid"), - 105: makeSyscallInfo("setuid", Hex), - 106: makeSyscallInfo("setgid", Hex), - 107: makeSyscallInfo("geteuid"), - 108: makeSyscallInfo("getegid"), - 109: makeSyscallInfo("setpgid", Hex, Hex), - 110: makeSyscallInfo("getppid"), - 111: makeSyscallInfo("getpgrp"), - 112: makeSyscallInfo("setsid"), - 113: makeSyscallInfo("setreuid", Hex, Hex), - 114: makeSyscallInfo("setregid", Hex, Hex), - 115: makeSyscallInfo("getgroups", Hex, Hex), - 116: makeSyscallInfo("setgroups", Hex, Hex), - 117: makeSyscallInfo("setresuid", Hex, Hex, Hex), - 118: makeSyscallInfo("getresuid", Hex, Hex, Hex), - 119: makeSyscallInfo("setresgid", Hex, Hex, Hex), - 120: makeSyscallInfo("getresgid", Hex, Hex, Hex), - 121: makeSyscallInfo("getpgid", Hex), - 122: makeSyscallInfo("setfsuid", Hex), - 123: makeSyscallInfo("setfsgid", Hex), - 124: makeSyscallInfo("getsid", Hex), - 125: makeSyscallInfo("capget", CapHeader, PostCapData), - 126: makeSyscallInfo("capset", CapHeader, CapData), - 127: makeSyscallInfo("rt_sigpending", Hex), - 128: makeSyscallInfo("rt_sigtimedwait", SigSet, Hex, Timespec, Hex), - 129: makeSyscallInfo("rt_sigqueueinfo", Hex, Signal, Hex), - 130: makeSyscallInfo("rt_sigsuspend", Hex), - 131: makeSyscallInfo("sigaltstack", Hex, Hex), - 132: makeSyscallInfo("utime", Path, Utimbuf), - 133: makeSyscallInfo("mknod", Path, Mode, Hex), - 134: makeSyscallInfo("uselib", Hex), - 135: makeSyscallInfo("personality", Hex), - 136: makeSyscallInfo("ustat", Hex, Hex), - 137: makeSyscallInfo("statfs", Path, Hex), - 138: makeSyscallInfo("fstatfs", FD, Hex), - 139: makeSyscallInfo("sysfs", Hex, Hex, Hex), - 140: makeSyscallInfo("getpriority", Hex, Hex), - 141: makeSyscallInfo("setpriority", Hex, Hex, Hex), - 142: makeSyscallInfo("sched_setparam", Hex, Hex), - 143: makeSyscallInfo("sched_getparam", Hex, Hex), - 144: makeSyscallInfo("sched_setscheduler", Hex, Hex, Hex), - 145: makeSyscallInfo("sched_getscheduler", Hex), - 146: makeSyscallInfo("sched_get_priority_max", Hex), - 147: makeSyscallInfo("sched_get_priority_min", Hex), - 148: makeSyscallInfo("sched_rr_get_interval", Hex, Hex), - 149: makeSyscallInfo("mlock", Hex, Hex), - 150: makeSyscallInfo("munlock", Hex, Hex), - 151: makeSyscallInfo("mlockall", Hex), - 152: makeSyscallInfo("munlockall"), - 153: makeSyscallInfo("vhangup"), - 154: makeSyscallInfo("modify_ldt", Hex, Hex, Hex), - 155: makeSyscallInfo("pivot_root", Path, Path), - 156: makeSyscallInfo("_sysctl", Hex), - 157: makeSyscallInfo("prctl", Hex, Hex, Hex, Hex, Hex), - 158: makeSyscallInfo("arch_prctl", Hex, Hex), - 159: makeSyscallInfo("adjtimex", Hex), - 160: makeSyscallInfo("setrlimit", Hex, Hex), - 161: makeSyscallInfo("chroot", Path), - 162: makeSyscallInfo("sync"), - 163: makeSyscallInfo("acct", Hex), - 164: makeSyscallInfo("settimeofday", Timeval, Hex), - 165: makeSyscallInfo("mount", Path, Path, Path, Hex, Path), - 166: makeSyscallInfo("umount2", Path, Hex), - 167: makeSyscallInfo("swapon", Hex, Hex), - 168: makeSyscallInfo("swapoff", Hex), - 169: makeSyscallInfo("reboot", Hex, Hex, Hex, Hex), - 170: makeSyscallInfo("sethostname", Hex, Hex), - 171: makeSyscallInfo("setdomainname", Hex, Hex), - 172: makeSyscallInfo("iopl", Hex), - 173: makeSyscallInfo("ioperm", Hex, Hex, Hex), - 174: makeSyscallInfo("create_module", Path, Hex), - 175: makeSyscallInfo("init_module", Hex, Hex, Hex), - 176: makeSyscallInfo("delete_module", Hex, Hex), - 177: makeSyscallInfo("get_kernel_syms", Hex), - // 178: query_module (only present in Linux < 2.6) - 179: makeSyscallInfo("quotactl", Hex, Hex, Hex, Hex), - 180: makeSyscallInfo("nfsservctl", Hex, Hex, Hex), - // 181: getpmsg (not implemented in the Linux kernel) - // 182: putpmsg (not implemented in the Linux kernel) - // 183: afs_syscall (not implemented in the Linux kernel) - // 184: tuxcall (not implemented in the Linux kernel) - // 185: security (not implemented in the Linux kernel) - 186: makeSyscallInfo("gettid"), - 187: makeSyscallInfo("readahead", Hex, Hex, Hex), - 188: makeSyscallInfo("setxattr", Path, Path, Hex, Hex, Hex), - 189: makeSyscallInfo("lsetxattr", Path, Path, Hex, Hex, Hex), - 190: makeSyscallInfo("fsetxattr", FD, Path, Hex, Hex, Hex), - 191: makeSyscallInfo("getxattr", Path, Path, Hex, Hex), - 192: makeSyscallInfo("lgetxattr", Path, Path, Hex, Hex), - 193: makeSyscallInfo("fgetxattr", FD, Path, Hex, Hex), - 194: makeSyscallInfo("listxattr", Path, Path, Hex), - 195: makeSyscallInfo("llistxattr", Path, Path, Hex), - 196: makeSyscallInfo("flistxattr", FD, Path, Hex), - 197: makeSyscallInfo("removexattr", Path, Path), - 198: makeSyscallInfo("lremovexattr", Path, Path), - 199: makeSyscallInfo("fremovexattr", FD, Path), - 200: makeSyscallInfo("tkill", Hex, Signal), - 201: makeSyscallInfo("time", Hex), - 202: makeSyscallInfo("futex", Hex, FutexOp, Hex, Timespec, Hex, Hex), - 203: makeSyscallInfo("sched_setaffinity", Hex, Hex, Hex), - 204: makeSyscallInfo("sched_getaffinity", Hex, Hex, Hex), - 205: makeSyscallInfo("set_thread_area", Hex), - 206: makeSyscallInfo("io_setup", Hex, Hex), - 207: makeSyscallInfo("io_destroy", Hex), - 208: makeSyscallInfo("io_getevents", Hex, Hex, Hex, Hex, Timespec), - 209: makeSyscallInfo("io_submit", Hex, Hex, Hex), - 210: makeSyscallInfo("io_cancel", Hex, Hex, Hex), - 211: makeSyscallInfo("get_thread_area", Hex), - 212: makeSyscallInfo("lookup_dcookie", Hex, Hex, Hex), - 213: makeSyscallInfo("epoll_create", Hex), - // 214: epoll_ctl_old (not implemented in the Linux kernel) - // 215: epoll_wait_old (not implemented in the Linux kernel) - 216: makeSyscallInfo("remap_file_pages", Hex, Hex, Hex, Hex, Hex), - 217: makeSyscallInfo("getdents64", FD, Hex, Hex), - 218: makeSyscallInfo("set_tid_address", Hex), - 219: makeSyscallInfo("restart_syscall"), - 220: makeSyscallInfo("semtimedop", Hex, Hex, Hex, Hex), - 221: makeSyscallInfo("fadvise64", FD, Hex, Hex, Hex), - 222: makeSyscallInfo("timer_create", Hex, Hex, Hex), - 223: makeSyscallInfo("timer_settime", Hex, Hex, ItimerSpec, PostItimerSpec), - 224: makeSyscallInfo("timer_gettime", Hex, PostItimerSpec), - 225: makeSyscallInfo("timer_getoverrun", Hex), - 226: makeSyscallInfo("timer_delete", Hex), - 227: makeSyscallInfo("clock_settime", Hex, Timespec), - 228: makeSyscallInfo("clock_gettime", Hex, PostTimespec), - 229: makeSyscallInfo("clock_getres", Hex, PostTimespec), - 230: makeSyscallInfo("clock_nanosleep", Hex, Hex, Timespec, PostTimespec), - 231: makeSyscallInfo("exit_group", Hex), - 232: makeSyscallInfo("epoll_wait", Hex, Hex, Hex, Hex), - 233: makeSyscallInfo("epoll_ctl", Hex, Hex, FD, Hex), - 234: makeSyscallInfo("tgkill", Hex, Hex, Signal), - 235: makeSyscallInfo("utimes", Path, Timeval), - // 236: vserver (not implemented in the Linux kernel) - 237: makeSyscallInfo("mbind", Hex, Hex, Hex, Hex, Hex, Hex), - 238: makeSyscallInfo("set_mempolicy", Hex, Hex, Hex), - 239: makeSyscallInfo("get_mempolicy", Hex, Hex, Hex, Hex, Hex), - 240: makeSyscallInfo("mq_open", Hex, Hex, Hex, Hex), - 241: makeSyscallInfo("mq_unlink", Hex), - 242: makeSyscallInfo("mq_timedsend", Hex, Hex, Hex, Hex, Hex), - 243: makeSyscallInfo("mq_timedreceive", Hex, Hex, Hex, Hex, Hex), - 244: makeSyscallInfo("mq_notify", Hex, Hex), - 245: makeSyscallInfo("mq_getsetattr", Hex, Hex, Hex), - 246: makeSyscallInfo("kexec_load", Hex, Hex, Hex, Hex), - 247: makeSyscallInfo("waitid", Hex, Hex, Hex, Hex, Rusage), - 248: makeSyscallInfo("add_key", Hex, Hex, Hex, Hex, Hex), - 249: makeSyscallInfo("request_key", Hex, Hex, Hex, Hex), - 250: makeSyscallInfo("keyctl", Hex, Hex, Hex, Hex, Hex), - 251: makeSyscallInfo("ioprio_set", Hex, Hex, Hex), - 252: makeSyscallInfo("ioprio_get", Hex, Hex), - 253: makeSyscallInfo("inotify_init"), - 254: makeSyscallInfo("inotify_add_watch", Hex, Path, Hex), - 255: makeSyscallInfo("inotify_rm_watch", Hex, Hex), - 256: makeSyscallInfo("migrate_pages", Hex, Hex, Hex, Hex), - 257: makeSyscallInfo("openat", FD, Path, OpenFlags, Mode), - 258: makeSyscallInfo("mkdirat", FD, Path, Hex), - 259: makeSyscallInfo("mknodat", FD, Path, Mode, Hex), - 260: makeSyscallInfo("fchownat", FD, Path, Hex, Hex, Hex), - 261: makeSyscallInfo("futimesat", FD, Path, Hex), - 262: makeSyscallInfo("newfstatat", FD, Path, Stat, Hex), - 263: makeSyscallInfo("unlinkat", FD, Path, Hex), - 264: makeSyscallInfo("renameat", FD, Path, Hex, Path), - 265: makeSyscallInfo("linkat", FD, Path, Hex, Path, Hex), - 266: makeSyscallInfo("symlinkat", Path, Hex, Path), - 267: makeSyscallInfo("readlinkat", FD, Path, ReadBuffer, Hex), - 268: makeSyscallInfo("fchmodat", FD, Path, Mode), - 269: makeSyscallInfo("faccessat", FD, Path, Oct, Hex), - 270: makeSyscallInfo("pselect6", Hex, SelectFDSet, SelectFDSet, SelectFDSet, Timespec, SigSet), - 271: makeSyscallInfo("ppoll", PollFDs, Hex, Timespec, SigSet, Hex), - 272: makeSyscallInfo("unshare", CloneFlags), - 273: makeSyscallInfo("set_robust_list", Hex, Hex), - 274: makeSyscallInfo("get_robust_list", Hex, Hex, Hex), - 275: makeSyscallInfo("splice", FD, Hex, FD, Hex, Hex, Hex), - 276: makeSyscallInfo("tee", FD, FD, Hex, Hex), - 277: makeSyscallInfo("sync_file_range", FD, Hex, Hex, Hex), - 278: makeSyscallInfo("vmsplice", FD, Hex, Hex, Hex), - 279: makeSyscallInfo("move_pages", Hex, Hex, Hex, Hex, Hex, Hex), - 280: makeSyscallInfo("utimensat", FD, Path, UTimeTimespec, Hex), - 281: makeSyscallInfo("epoll_pwait", Hex, Hex, Hex, Hex, SigSet, Hex), - 282: makeSyscallInfo("signalfd", Hex, Hex, Hex), - 283: makeSyscallInfo("timerfd_create", Hex, Hex), - 284: makeSyscallInfo("eventfd", Hex), - 285: makeSyscallInfo("fallocate", FD, Hex, Hex, Hex), - 286: makeSyscallInfo("timerfd_settime", FD, Hex, ItimerSpec, PostItimerSpec), - 287: makeSyscallInfo("timerfd_gettime", FD, PostItimerSpec), - 288: makeSyscallInfo("accept4", FD, PostSockAddr, SockLen, SockFlags), - 289: makeSyscallInfo("signalfd4", Hex, Hex, Hex, Hex), - 290: makeSyscallInfo("eventfd2", Hex, Hex), - 291: makeSyscallInfo("epoll_create1", Hex), - 292: makeSyscallInfo("dup3", FD, FD, Hex), - 293: makeSyscallInfo("pipe2", PipeFDs, Hex), - 294: makeSyscallInfo("inotify_init1", Hex), - 295: makeSyscallInfo("preadv", FD, ReadIOVec, Hex, Hex), - 296: makeSyscallInfo("pwritev", FD, WriteIOVec, Hex, Hex), - 297: makeSyscallInfo("rt_tgsigqueueinfo", Hex, Hex, Signal, Hex), - 298: makeSyscallInfo("perf_event_open", Hex, Hex, Hex, Hex, Hex), - 299: makeSyscallInfo("recvmmsg", FD, Hex, Hex, Hex, Hex), - 300: makeSyscallInfo("fanotify_init", Hex, Hex), - 301: makeSyscallInfo("fanotify_mark", Hex, Hex, Hex, Hex, Hex), - 302: makeSyscallInfo("prlimit64", Hex, Hex, Hex, Hex), - 303: makeSyscallInfo("name_to_handle_at", FD, Hex, Hex, Hex, Hex), - 304: makeSyscallInfo("open_by_handle_at", FD, Hex, Hex), - 305: makeSyscallInfo("clock_adjtime", Hex, Hex), - 306: makeSyscallInfo("syncfs", FD), - 307: makeSyscallInfo("sendmmsg", FD, Hex, Hex, Hex), - 308: makeSyscallInfo("setns", FD, Hex), - 309: makeSyscallInfo("getcpu", Hex, Hex, Hex), - 310: makeSyscallInfo("process_vm_readv", Hex, ReadIOVec, Hex, IOVec, Hex, Hex), - 311: makeSyscallInfo("process_vm_writev", Hex, IOVec, Hex, WriteIOVec, Hex, Hex), - 312: makeSyscallInfo("kcmp", Hex, Hex, Hex, Hex, Hex), - 313: makeSyscallInfo("finit_module", Hex, Hex, Hex), - 314: makeSyscallInfo("sched_setattr", Hex, Hex, Hex), - 315: makeSyscallInfo("sched_getattr", Hex, Hex, Hex), - 316: makeSyscallInfo("renameat2", FD, Path, Hex, Path, Hex), - 317: makeSyscallInfo("seccomp", Hex, Hex, Hex), - 318: makeSyscallInfo("getrandom", Hex, Hex, Hex), - 319: makeSyscallInfo("memfd_create", Path, Hex), // Not quite a path, but close. - 320: makeSyscallInfo("kexec_file_load", FD, FD, Hex, Hex, Hex), - 321: makeSyscallInfo("bpf", Hex, Hex, Hex), - 322: makeSyscallInfo("execveat", FD, Path, ExecveStringVector, ExecveStringVector, Hex), - 323: makeSyscallInfo("userfaultfd", Hex), - 324: makeSyscallInfo("membarrier", Hex, Hex), - 325: makeSyscallInfo("mlock2", Hex, Hex, Hex), - 326: makeSyscallInfo("copy_file_range", FD, Hex, FD, Hex, Hex, Hex), - 327: makeSyscallInfo("preadv2", FD, ReadIOVec, Hex, Hex, Hex), - 328: makeSyscallInfo("pwritev2", FD, WriteIOVec, Hex, Hex, Hex), - 329: makeSyscallInfo("pkey_mprotect", Hex, Hex, Hex, Hex), - 330: makeSyscallInfo("pkey_alloc", Hex, Hex), - 331: makeSyscallInfo("pkey_free", Hex), - 332: makeSyscallInfo("statx", FD, Path, Hex, Hex, Hex), - 333: makeSyscallInfo("io_pgetevents", Hex, Hex, Hex, Hex, Timespec, SigSet), - 334: makeSyscallInfo("rseq", Hex, Hex, Hex, Hex), - 424: makeSyscallInfo("pidfd_send_signal", FD, Signal, Hex, Hex), - 425: makeSyscallInfo("io_uring_setup", Hex, Hex), - 426: makeSyscallInfo("io_uring_enter", FD, Hex, Hex, Hex, SigSet, Hex), - 427: makeSyscallInfo("io_uring_register", FD, Hex, Hex, Hex), - 428: makeSyscallInfo("open_tree", FD, Path, Hex), - 429: makeSyscallInfo("move_mount", FD, Path, FD, Path, Hex), - 430: makeSyscallInfo("fsopen", Path, Hex), // Not quite a path, but close. - 431: makeSyscallInfo("fsconfig", FD, Hex, Hex, Hex, Hex), - 432: makeSyscallInfo("fsmount", FD, Hex, Hex), - 433: makeSyscallInfo("fspick", FD, Path, Hex), - 434: makeSyscallInfo("pidfd_open", Hex, Hex), - 435: makeSyscallInfo("clone3", Hex, Hex), -} diff --git a/pkg/sentry/strace/linux64_amd64.go b/pkg/sentry/strace/linux64_amd64.go new file mode 100644 index 000000000..9fa2f0e16 --- /dev/null +++ b/pkg/sentry/strace/linux64_amd64.go @@ -0,0 +1,382 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build amd64 + +package strace + +import ( + "gvisor.dev/gvisor/pkg/abi" + "gvisor.dev/gvisor/pkg/sentry/arch" +) + +// linuxAMD64 provides a mapping of the Linux amd64 syscalls and their argument +// types for display / formatting. +var linuxAMD64 = SyscallMap{ + 0: makeSyscallInfo("read", FD, ReadBuffer, Hex), + 1: makeSyscallInfo("write", FD, WriteBuffer, Hex), + 2: makeSyscallInfo("open", Path, OpenFlags, Mode), + 3: makeSyscallInfo("close", FD), + 4: makeSyscallInfo("stat", Path, Stat), + 5: makeSyscallInfo("fstat", FD, Stat), + 6: makeSyscallInfo("lstat", Path, Stat), + 7: makeSyscallInfo("poll", PollFDs, Hex, Hex), + 8: makeSyscallInfo("lseek", Hex, Hex, Hex), + 9: makeSyscallInfo("mmap", Hex, Hex, Hex, Hex, FD, Hex), + 10: makeSyscallInfo("mprotect", Hex, Hex, Hex), + 11: makeSyscallInfo("munmap", Hex, Hex), + 12: makeSyscallInfo("brk", Hex), + 13: makeSyscallInfo("rt_sigaction", Signal, SigAction, PostSigAction), + 14: makeSyscallInfo("rt_sigprocmask", SignalMaskAction, SigSet, PostSigSet, Hex), + 15: makeSyscallInfo("rt_sigreturn"), + 16: makeSyscallInfo("ioctl", FD, Hex, Hex), + 17: makeSyscallInfo("pread64", FD, ReadBuffer, Hex, Hex), + 18: makeSyscallInfo("pwrite64", FD, WriteBuffer, Hex, Hex), + 19: makeSyscallInfo("readv", FD, ReadIOVec, Hex), + 20: makeSyscallInfo("writev", FD, WriteIOVec, Hex), + 21: makeSyscallInfo("access", Path, Oct), + 22: makeSyscallInfo("pipe", PipeFDs), + 23: makeSyscallInfo("select", Hex, SelectFDSet, SelectFDSet, SelectFDSet, Timeval), + 24: makeSyscallInfo("sched_yield"), + 25: makeSyscallInfo("mremap", Hex, Hex, Hex, Hex, Hex), + 26: makeSyscallInfo("msync", Hex, Hex, Hex), + 27: makeSyscallInfo("mincore", Hex, Hex, Hex), + 28: makeSyscallInfo("madvise", Hex, Hex, Hex), + 29: makeSyscallInfo("shmget", Hex, Hex, Hex), + 30: makeSyscallInfo("shmat", Hex, Hex, Hex), + 31: makeSyscallInfo("shmctl", Hex, Hex, Hex), + 32: makeSyscallInfo("dup", FD), + 33: makeSyscallInfo("dup2", FD, FD), + 34: makeSyscallInfo("pause"), + 35: makeSyscallInfo("nanosleep", Timespec, PostTimespec), + 36: makeSyscallInfo("getitimer", ItimerType, PostItimerVal), + 37: makeSyscallInfo("alarm", Hex), + 38: makeSyscallInfo("setitimer", ItimerType, ItimerVal, PostItimerVal), + 39: makeSyscallInfo("getpid"), + 40: makeSyscallInfo("sendfile", FD, FD, Hex, Hex), + 41: makeSyscallInfo("socket", SockFamily, SockType, SockProtocol), + 42: makeSyscallInfo("connect", FD, SockAddr, Hex), + 43: makeSyscallInfo("accept", FD, PostSockAddr, SockLen), + 44: makeSyscallInfo("sendto", FD, Hex, Hex, Hex, SockAddr, Hex), + 45: makeSyscallInfo("recvfrom", FD, Hex, Hex, Hex, PostSockAddr, SockLen), + 46: makeSyscallInfo("sendmsg", FD, SendMsgHdr, Hex), + 47: makeSyscallInfo("recvmsg", FD, RecvMsgHdr, Hex), + 48: makeSyscallInfo("shutdown", FD, Hex), + 49: makeSyscallInfo("bind", FD, SockAddr, Hex), + 50: makeSyscallInfo("listen", FD, Hex), + 51: makeSyscallInfo("getsockname", FD, PostSockAddr, SockLen), + 52: makeSyscallInfo("getpeername", FD, PostSockAddr, SockLen), + 53: makeSyscallInfo("socketpair", SockFamily, SockType, SockProtocol, Hex), + 54: makeSyscallInfo("setsockopt", FD, Hex, Hex, Hex, Hex), + 55: makeSyscallInfo("getsockopt", FD, Hex, Hex, Hex, Hex), + 56: makeSyscallInfo("clone", CloneFlags, Hex, Hex, Hex, Hex), + 57: makeSyscallInfo("fork"), + 58: makeSyscallInfo("vfork"), + 59: makeSyscallInfo("execve", Path, ExecveStringVector, ExecveStringVector), + 60: makeSyscallInfo("exit", Hex), + 61: makeSyscallInfo("wait4", Hex, Hex, Hex, Rusage), + 62: makeSyscallInfo("kill", Hex, Signal), + 63: makeSyscallInfo("uname", Uname), + 64: makeSyscallInfo("semget", Hex, Hex, Hex), + 65: makeSyscallInfo("semop", Hex, Hex, Hex), + 66: makeSyscallInfo("semctl", Hex, Hex, Hex, Hex), + 67: makeSyscallInfo("shmdt", Hex), + 68: makeSyscallInfo("msgget", Hex, Hex), + 69: makeSyscallInfo("msgsnd", Hex, Hex, Hex, Hex), + 70: makeSyscallInfo("msgrcv", Hex, Hex, Hex, Hex, Hex), + 71: makeSyscallInfo("msgctl", Hex, Hex, Hex), + 72: makeSyscallInfo("fcntl", FD, Hex, Hex), + 73: makeSyscallInfo("flock", FD, Hex), + 74: makeSyscallInfo("fsync", FD), + 75: makeSyscallInfo("fdatasync", FD), + 76: makeSyscallInfo("truncate", Path, Hex), + 77: makeSyscallInfo("ftruncate", FD, Hex), + 78: makeSyscallInfo("getdents", FD, Hex, Hex), + 79: makeSyscallInfo("getcwd", PostPath, Hex), + 80: makeSyscallInfo("chdir", Path), + 81: makeSyscallInfo("fchdir", FD), + 82: makeSyscallInfo("rename", Path, Path), + 83: makeSyscallInfo("mkdir", Path, Oct), + 84: makeSyscallInfo("rmdir", Path), + 85: makeSyscallInfo("creat", Path, Oct), + 86: makeSyscallInfo("link", Path, Path), + 87: makeSyscallInfo("unlink", Path), + 88: makeSyscallInfo("symlink", Path, Path), + 89: makeSyscallInfo("readlink", Path, ReadBuffer, Hex), + 90: makeSyscallInfo("chmod", Path, Mode), + 91: makeSyscallInfo("fchmod", FD, Mode), + 92: makeSyscallInfo("chown", Path, Hex, Hex), + 93: makeSyscallInfo("fchown", FD, Hex, Hex), + 94: makeSyscallInfo("lchown", Path, Hex, Hex), + 95: makeSyscallInfo("umask", Hex), + 96: makeSyscallInfo("gettimeofday", Timeval, Hex), + 97: makeSyscallInfo("getrlimit", Hex, Hex), + 98: makeSyscallInfo("getrusage", Hex, Rusage), + 99: makeSyscallInfo("sysinfo", Hex), + 100: makeSyscallInfo("times", Hex), + 101: makeSyscallInfo("ptrace", PtraceRequest, Hex, Hex, Hex), + 102: makeSyscallInfo("getuid"), + 103: makeSyscallInfo("syslog", Hex, Hex, Hex), + 104: makeSyscallInfo("getgid"), + 105: makeSyscallInfo("setuid", Hex), + 106: makeSyscallInfo("setgid", Hex), + 107: makeSyscallInfo("geteuid"), + 108: makeSyscallInfo("getegid"), + 109: makeSyscallInfo("setpgid", Hex, Hex), + 110: makeSyscallInfo("getppid"), + 111: makeSyscallInfo("getpgrp"), + 112: makeSyscallInfo("setsid"), + 113: makeSyscallInfo("setreuid", Hex, Hex), + 114: makeSyscallInfo("setregid", Hex, Hex), + 115: makeSyscallInfo("getgroups", Hex, Hex), + 116: makeSyscallInfo("setgroups", Hex, Hex), + 117: makeSyscallInfo("setresuid", Hex, Hex, Hex), + 118: makeSyscallInfo("getresuid", Hex, Hex, Hex), + 119: makeSyscallInfo("setresgid", Hex, Hex, Hex), + 120: makeSyscallInfo("getresgid", Hex, Hex, Hex), + 121: makeSyscallInfo("getpgid", Hex), + 122: makeSyscallInfo("setfsuid", Hex), + 123: makeSyscallInfo("setfsgid", Hex), + 124: makeSyscallInfo("getsid", Hex), + 125: makeSyscallInfo("capget", CapHeader, PostCapData), + 126: makeSyscallInfo("capset", CapHeader, CapData), + 127: makeSyscallInfo("rt_sigpending", Hex), + 128: makeSyscallInfo("rt_sigtimedwait", SigSet, Hex, Timespec, Hex), + 129: makeSyscallInfo("rt_sigqueueinfo", Hex, Signal, Hex), + 130: makeSyscallInfo("rt_sigsuspend", Hex), + 131: makeSyscallInfo("sigaltstack", Hex, Hex), + 132: makeSyscallInfo("utime", Path, Utimbuf), + 133: makeSyscallInfo("mknod", Path, Mode, Hex), + 134: makeSyscallInfo("uselib", Hex), + 135: makeSyscallInfo("personality", Hex), + 136: makeSyscallInfo("ustat", Hex, Hex), + 137: makeSyscallInfo("statfs", Path, Hex), + 138: makeSyscallInfo("fstatfs", FD, Hex), + 139: makeSyscallInfo("sysfs", Hex, Hex, Hex), + 140: makeSyscallInfo("getpriority", Hex, Hex), + 141: makeSyscallInfo("setpriority", Hex, Hex, Hex), + 142: makeSyscallInfo("sched_setparam", Hex, Hex), + 143: makeSyscallInfo("sched_getparam", Hex, Hex), + 144: makeSyscallInfo("sched_setscheduler", Hex, Hex, Hex), + 145: makeSyscallInfo("sched_getscheduler", Hex), + 146: makeSyscallInfo("sched_get_priority_max", Hex), + 147: makeSyscallInfo("sched_get_priority_min", Hex), + 148: makeSyscallInfo("sched_rr_get_interval", Hex, Hex), + 149: makeSyscallInfo("mlock", Hex, Hex), + 150: makeSyscallInfo("munlock", Hex, Hex), + 151: makeSyscallInfo("mlockall", Hex), + 152: makeSyscallInfo("munlockall"), + 153: makeSyscallInfo("vhangup"), + 154: makeSyscallInfo("modify_ldt", Hex, Hex, Hex), + 155: makeSyscallInfo("pivot_root", Path, Path), + 156: makeSyscallInfo("_sysctl", Hex), + 157: makeSyscallInfo("prctl", Hex, Hex, Hex, Hex, Hex), + 158: makeSyscallInfo("arch_prctl", Hex, Hex), + 159: makeSyscallInfo("adjtimex", Hex), + 160: makeSyscallInfo("setrlimit", Hex, Hex), + 161: makeSyscallInfo("chroot", Path), + 162: makeSyscallInfo("sync"), + 163: makeSyscallInfo("acct", Hex), + 164: makeSyscallInfo("settimeofday", Timeval, Hex), + 165: makeSyscallInfo("mount", Path, Path, Path, Hex, Path), + 166: makeSyscallInfo("umount2", Path, Hex), + 167: makeSyscallInfo("swapon", Hex, Hex), + 168: makeSyscallInfo("swapoff", Hex), + 169: makeSyscallInfo("reboot", Hex, Hex, Hex, Hex), + 170: makeSyscallInfo("sethostname", Hex, Hex), + 171: makeSyscallInfo("setdomainname", Hex, Hex), + 172: makeSyscallInfo("iopl", Hex), + 173: makeSyscallInfo("ioperm", Hex, Hex, Hex), + 174: makeSyscallInfo("create_module", Path, Hex), + 175: makeSyscallInfo("init_module", Hex, Hex, Hex), + 176: makeSyscallInfo("delete_module", Hex, Hex), + 177: makeSyscallInfo("get_kernel_syms", Hex), + // 178: query_module (only present in Linux < 2.6) + 179: makeSyscallInfo("quotactl", Hex, Hex, Hex, Hex), + 180: makeSyscallInfo("nfsservctl", Hex, Hex, Hex), + // 181: getpmsg (not implemented in the Linux kernel) + // 182: putpmsg (not implemented in the Linux kernel) + // 183: afs_syscall (not implemented in the Linux kernel) + // 184: tuxcall (not implemented in the Linux kernel) + // 185: security (not implemented in the Linux kernel) + 186: makeSyscallInfo("gettid"), + 187: makeSyscallInfo("readahead", Hex, Hex, Hex), + 188: makeSyscallInfo("setxattr", Path, Path, Hex, Hex, Hex), + 189: makeSyscallInfo("lsetxattr", Path, Path, Hex, Hex, Hex), + 190: makeSyscallInfo("fsetxattr", FD, Path, Hex, Hex, Hex), + 191: makeSyscallInfo("getxattr", Path, Path, Hex, Hex), + 192: makeSyscallInfo("lgetxattr", Path, Path, Hex, Hex), + 193: makeSyscallInfo("fgetxattr", FD, Path, Hex, Hex), + 194: makeSyscallInfo("listxattr", Path, Path, Hex), + 195: makeSyscallInfo("llistxattr", Path, Path, Hex), + 196: makeSyscallInfo("flistxattr", FD, Path, Hex), + 197: makeSyscallInfo("removexattr", Path, Path), + 198: makeSyscallInfo("lremovexattr", Path, Path), + 199: makeSyscallInfo("fremovexattr", FD, Path), + 200: makeSyscallInfo("tkill", Hex, Signal), + 201: makeSyscallInfo("time", Hex), + 202: makeSyscallInfo("futex", Hex, FutexOp, Hex, Timespec, Hex, Hex), + 203: makeSyscallInfo("sched_setaffinity", Hex, Hex, Hex), + 204: makeSyscallInfo("sched_getaffinity", Hex, Hex, Hex), + 205: makeSyscallInfo("set_thread_area", Hex), + 206: makeSyscallInfo("io_setup", Hex, Hex), + 207: makeSyscallInfo("io_destroy", Hex), + 208: makeSyscallInfo("io_getevents", Hex, Hex, Hex, Hex, Timespec), + 209: makeSyscallInfo("io_submit", Hex, Hex, Hex), + 210: makeSyscallInfo("io_cancel", Hex, Hex, Hex), + 211: makeSyscallInfo("get_thread_area", Hex), + 212: makeSyscallInfo("lookup_dcookie", Hex, Hex, Hex), + 213: makeSyscallInfo("epoll_create", Hex), + // 214: epoll_ctl_old (not implemented in the Linux kernel) + // 215: epoll_wait_old (not implemented in the Linux kernel) + 216: makeSyscallInfo("remap_file_pages", Hex, Hex, Hex, Hex, Hex), + 217: makeSyscallInfo("getdents64", FD, Hex, Hex), + 218: makeSyscallInfo("set_tid_address", Hex), + 219: makeSyscallInfo("restart_syscall"), + 220: makeSyscallInfo("semtimedop", Hex, Hex, Hex, Hex), + 221: makeSyscallInfo("fadvise64", FD, Hex, Hex, Hex), + 222: makeSyscallInfo("timer_create", Hex, Hex, Hex), + 223: makeSyscallInfo("timer_settime", Hex, Hex, ItimerSpec, PostItimerSpec), + 224: makeSyscallInfo("timer_gettime", Hex, PostItimerSpec), + 225: makeSyscallInfo("timer_getoverrun", Hex), + 226: makeSyscallInfo("timer_delete", Hex), + 227: makeSyscallInfo("clock_settime", Hex, Timespec), + 228: makeSyscallInfo("clock_gettime", Hex, PostTimespec), + 229: makeSyscallInfo("clock_getres", Hex, PostTimespec), + 230: makeSyscallInfo("clock_nanosleep", Hex, Hex, Timespec, PostTimespec), + 231: makeSyscallInfo("exit_group", Hex), + 232: makeSyscallInfo("epoll_wait", Hex, Hex, Hex, Hex), + 233: makeSyscallInfo("epoll_ctl", Hex, Hex, FD, Hex), + 234: makeSyscallInfo("tgkill", Hex, Hex, Signal), + 235: makeSyscallInfo("utimes", Path, Timeval), + // 236: vserver (not implemented in the Linux kernel) + 237: makeSyscallInfo("mbind", Hex, Hex, Hex, Hex, Hex, Hex), + 238: makeSyscallInfo("set_mempolicy", Hex, Hex, Hex), + 239: makeSyscallInfo("get_mempolicy", Hex, Hex, Hex, Hex, Hex), + 240: makeSyscallInfo("mq_open", Hex, Hex, Hex, Hex), + 241: makeSyscallInfo("mq_unlink", Hex), + 242: makeSyscallInfo("mq_timedsend", Hex, Hex, Hex, Hex, Hex), + 243: makeSyscallInfo("mq_timedreceive", Hex, Hex, Hex, Hex, Hex), + 244: makeSyscallInfo("mq_notify", Hex, Hex), + 245: makeSyscallInfo("mq_getsetattr", Hex, Hex, Hex), + 246: makeSyscallInfo("kexec_load", Hex, Hex, Hex, Hex), + 247: makeSyscallInfo("waitid", Hex, Hex, Hex, Hex, Rusage), + 248: makeSyscallInfo("add_key", Hex, Hex, Hex, Hex, Hex), + 249: makeSyscallInfo("request_key", Hex, Hex, Hex, Hex), + 250: makeSyscallInfo("keyctl", Hex, Hex, Hex, Hex, Hex), + 251: makeSyscallInfo("ioprio_set", Hex, Hex, Hex), + 252: makeSyscallInfo("ioprio_get", Hex, Hex), + 253: makeSyscallInfo("inotify_init"), + 254: makeSyscallInfo("inotify_add_watch", Hex, Path, Hex), + 255: makeSyscallInfo("inotify_rm_watch", Hex, Hex), + 256: makeSyscallInfo("migrate_pages", Hex, Hex, Hex, Hex), + 257: makeSyscallInfo("openat", FD, Path, OpenFlags, Mode), + 258: makeSyscallInfo("mkdirat", FD, Path, Hex), + 259: makeSyscallInfo("mknodat", FD, Path, Mode, Hex), + 260: makeSyscallInfo("fchownat", FD, Path, Hex, Hex, Hex), + 261: makeSyscallInfo("futimesat", FD, Path, Hex), + 262: makeSyscallInfo("newfstatat", FD, Path, Stat, Hex), + 263: makeSyscallInfo("unlinkat", FD, Path, Hex), + 264: makeSyscallInfo("renameat", FD, Path, Hex, Path), + 265: makeSyscallInfo("linkat", FD, Path, Hex, Path, Hex), + 266: makeSyscallInfo("symlinkat", Path, Hex, Path), + 267: makeSyscallInfo("readlinkat", FD, Path, ReadBuffer, Hex), + 268: makeSyscallInfo("fchmodat", FD, Path, Mode), + 269: makeSyscallInfo("faccessat", FD, Path, Oct, Hex), + 270: makeSyscallInfo("pselect6", Hex, SelectFDSet, SelectFDSet, SelectFDSet, Timespec, SigSet), + 271: makeSyscallInfo("ppoll", PollFDs, Hex, Timespec, SigSet, Hex), + 272: makeSyscallInfo("unshare", CloneFlags), + 273: makeSyscallInfo("set_robust_list", Hex, Hex), + 274: makeSyscallInfo("get_robust_list", Hex, Hex, Hex), + 275: makeSyscallInfo("splice", FD, Hex, FD, Hex, Hex, Hex), + 276: makeSyscallInfo("tee", FD, FD, Hex, Hex), + 277: makeSyscallInfo("sync_file_range", FD, Hex, Hex, Hex), + 278: makeSyscallInfo("vmsplice", FD, Hex, Hex, Hex), + 279: makeSyscallInfo("move_pages", Hex, Hex, Hex, Hex, Hex, Hex), + 280: makeSyscallInfo("utimensat", FD, Path, UTimeTimespec, Hex), + 281: makeSyscallInfo("epoll_pwait", Hex, Hex, Hex, Hex, SigSet, Hex), + 282: makeSyscallInfo("signalfd", Hex, Hex, Hex), + 283: makeSyscallInfo("timerfd_create", Hex, Hex), + 284: makeSyscallInfo("eventfd", Hex), + 285: makeSyscallInfo("fallocate", FD, Hex, Hex, Hex), + 286: makeSyscallInfo("timerfd_settime", FD, Hex, ItimerSpec, PostItimerSpec), + 287: makeSyscallInfo("timerfd_gettime", FD, PostItimerSpec), + 288: makeSyscallInfo("accept4", FD, PostSockAddr, SockLen, SockFlags), + 289: makeSyscallInfo("signalfd4", Hex, Hex, Hex, Hex), + 290: makeSyscallInfo("eventfd2", Hex, Hex), + 291: makeSyscallInfo("epoll_create1", Hex), + 292: makeSyscallInfo("dup3", FD, FD, Hex), + 293: makeSyscallInfo("pipe2", PipeFDs, Hex), + 294: makeSyscallInfo("inotify_init1", Hex), + 295: makeSyscallInfo("preadv", FD, ReadIOVec, Hex, Hex), + 296: makeSyscallInfo("pwritev", FD, WriteIOVec, Hex, Hex), + 297: makeSyscallInfo("rt_tgsigqueueinfo", Hex, Hex, Signal, Hex), + 298: makeSyscallInfo("perf_event_open", Hex, Hex, Hex, Hex, Hex), + 299: makeSyscallInfo("recvmmsg", FD, Hex, Hex, Hex, Hex), + 300: makeSyscallInfo("fanotify_init", Hex, Hex), + 301: makeSyscallInfo("fanotify_mark", Hex, Hex, Hex, Hex, Hex), + 302: makeSyscallInfo("prlimit64", Hex, Hex, Hex, Hex), + 303: makeSyscallInfo("name_to_handle_at", FD, Hex, Hex, Hex, Hex), + 304: makeSyscallInfo("open_by_handle_at", FD, Hex, Hex), + 305: makeSyscallInfo("clock_adjtime", Hex, Hex), + 306: makeSyscallInfo("syncfs", FD), + 307: makeSyscallInfo("sendmmsg", FD, Hex, Hex, Hex), + 308: makeSyscallInfo("setns", FD, Hex), + 309: makeSyscallInfo("getcpu", Hex, Hex, Hex), + 310: makeSyscallInfo("process_vm_readv", Hex, ReadIOVec, Hex, IOVec, Hex, Hex), + 311: makeSyscallInfo("process_vm_writev", Hex, IOVec, Hex, WriteIOVec, Hex, Hex), + 312: makeSyscallInfo("kcmp", Hex, Hex, Hex, Hex, Hex), + 313: makeSyscallInfo("finit_module", Hex, Hex, Hex), + 314: makeSyscallInfo("sched_setattr", Hex, Hex, Hex), + 315: makeSyscallInfo("sched_getattr", Hex, Hex, Hex), + 316: makeSyscallInfo("renameat2", FD, Path, Hex, Path, Hex), + 317: makeSyscallInfo("seccomp", Hex, Hex, Hex), + 318: makeSyscallInfo("getrandom", Hex, Hex, Hex), + 319: makeSyscallInfo("memfd_create", Path, Hex), // Not quite a path, but close. + 320: makeSyscallInfo("kexec_file_load", FD, FD, Hex, Hex, Hex), + 321: makeSyscallInfo("bpf", Hex, Hex, Hex), + 322: makeSyscallInfo("execveat", FD, Path, ExecveStringVector, ExecveStringVector, Hex), + 323: makeSyscallInfo("userfaultfd", Hex), + 324: makeSyscallInfo("membarrier", Hex, Hex), + 325: makeSyscallInfo("mlock2", Hex, Hex, Hex), + 326: makeSyscallInfo("copy_file_range", FD, Hex, FD, Hex, Hex, Hex), + 327: makeSyscallInfo("preadv2", FD, ReadIOVec, Hex, Hex, Hex), + 328: makeSyscallInfo("pwritev2", FD, WriteIOVec, Hex, Hex, Hex), + 329: makeSyscallInfo("pkey_mprotect", Hex, Hex, Hex, Hex), + 330: makeSyscallInfo("pkey_alloc", Hex, Hex), + 331: makeSyscallInfo("pkey_free", Hex), + 332: makeSyscallInfo("statx", FD, Path, Hex, Hex, Hex), + 333: makeSyscallInfo("io_pgetevents", Hex, Hex, Hex, Hex, Timespec, SigSet), + 334: makeSyscallInfo("rseq", Hex, Hex, Hex, Hex), + 424: makeSyscallInfo("pidfd_send_signal", FD, Signal, Hex, Hex), + 425: makeSyscallInfo("io_uring_setup", Hex, Hex), + 426: makeSyscallInfo("io_uring_enter", FD, Hex, Hex, Hex, SigSet, Hex), + 427: makeSyscallInfo("io_uring_register", FD, Hex, Hex, Hex), + 428: makeSyscallInfo("open_tree", FD, Path, Hex), + 429: makeSyscallInfo("move_mount", FD, Path, FD, Path, Hex), + 430: makeSyscallInfo("fsopen", Path, Hex), // Not quite a path, but close. + 431: makeSyscallInfo("fsconfig", FD, Hex, Hex, Hex, Hex), + 432: makeSyscallInfo("fsmount", FD, Hex, Hex), + 433: makeSyscallInfo("fspick", FD, Path, Hex), + 434: makeSyscallInfo("pidfd_open", Hex, Hex), + 435: makeSyscallInfo("clone3", Hex, Hex), +} + +func init() { + syscallTables = append(syscallTables, + syscallTable{ + os: abi.Linux, + arch: arch.AMD64, + syscalls: linuxAMD64}) +} diff --git a/pkg/sentry/strace/linux64_arm64.go b/pkg/sentry/strace/linux64_arm64.go new file mode 100644 index 000000000..c3ac5248d --- /dev/null +++ b/pkg/sentry/strace/linux64_arm64.go @@ -0,0 +1,323 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package strace + +import ( + "gvisor.dev/gvisor/pkg/abi" + "gvisor.dev/gvisor/pkg/sentry/arch" +) + +// linuxARM64 provides a mapping of the Linux arm64 syscalls and their argument +// types for display / formatting. +var linuxARM64 = SyscallMap{ + 0: makeSyscallInfo("io_setup", Hex, Hex), + 1: makeSyscallInfo("io_destroy", Hex), + 2: makeSyscallInfo("io_submit", Hex, Hex, Hex), + 3: makeSyscallInfo("io_cancel", Hex, Hex, Hex), + 4: makeSyscallInfo("io_getevents", Hex, Hex, Hex, Hex, Timespec), + 5: makeSyscallInfo("setxattr", Path, Path, Hex, Hex, Hex), + 6: makeSyscallInfo("lsetxattr", Path, Path, Hex, Hex, Hex), + 7: makeSyscallInfo("fsetxattr", FD, Path, Hex, Hex, Hex), + 8: makeSyscallInfo("getxattr", Path, Path, Hex, Hex), + 9: makeSyscallInfo("lgetxattr", Path, Path, Hex, Hex), + 10: makeSyscallInfo("fgetxattr", FD, Path, Hex, Hex), + 11: makeSyscallInfo("listxattr", Path, Path, Hex), + 12: makeSyscallInfo("llistxattr", Path, Path, Hex), + 13: makeSyscallInfo("flistxattr", FD, Path, Hex), + 14: makeSyscallInfo("removexattr", Path, Path), + 15: makeSyscallInfo("lremovexattr", Path, Path), + 16: makeSyscallInfo("fremovexattr", FD, Path), + 17: makeSyscallInfo("getcwd", PostPath, Hex), + 18: makeSyscallInfo("lookup_dcookie", Hex, Hex, Hex), + 19: makeSyscallInfo("eventfd2", Hex, Hex), + 20: makeSyscallInfo("epoll_create1", Hex), + 21: makeSyscallInfo("epoll_ctl", Hex, Hex, FD, Hex), + 22: makeSyscallInfo("epoll_pwait", Hex, Hex, Hex, Hex, SigSet, Hex), + 23: makeSyscallInfo("dup", FD), + 24: makeSyscallInfo("dup3", FD, FD, Hex), + 25: makeSyscallInfo("fcntl", FD, Hex, Hex), + 26: makeSyscallInfo("inotify_init1", Hex), + 27: makeSyscallInfo("inotify_add_watch", Hex, Path, Hex), + 28: makeSyscallInfo("inotify_rm_watch", Hex, Hex), + 29: makeSyscallInfo("ioctl", FD, Hex, Hex), + 30: makeSyscallInfo("ioprio_set", Hex, Hex, Hex), + 31: makeSyscallInfo("ioprio_get", Hex, Hex), + 32: makeSyscallInfo("flock", FD, Hex), + 33: makeSyscallInfo("mknodat", FD, Path, Mode, Hex), + 34: makeSyscallInfo("mkdirat", FD, Path, Hex), + 35: makeSyscallInfo("unlinkat", FD, Path, Hex), + 36: makeSyscallInfo("symlinkat", Path, Hex, Path), + 37: makeSyscallInfo("linkat", FD, Path, Hex, Path, Hex), + 38: makeSyscallInfo("renameat", FD, Path, Hex, Path), + 39: makeSyscallInfo("umount2", Path, Hex), + 40: makeSyscallInfo("mount", Path, Path, Path, Hex, Path), + 41: makeSyscallInfo("pivot_root", Path, Path), + 42: makeSyscallInfo("nfsservctl", Hex, Hex, Hex), + 43: makeSyscallInfo("statfs", Path, Hex), + 44: makeSyscallInfo("fstatfs", FD, Hex), + 45: makeSyscallInfo("truncate", Path, Hex), + 46: makeSyscallInfo("ftruncate", FD, Hex), + 47: makeSyscallInfo("fallocate", FD, Hex, Hex, Hex), + 48: makeSyscallInfo("faccessat", FD, Path, Oct, Hex), + 49: makeSyscallInfo("chdir", Path), + 50: makeSyscallInfo("fchdir", FD), + 51: makeSyscallInfo("chroot", Path), + 52: makeSyscallInfo("fchmod", FD, Mode), + 53: makeSyscallInfo("fchmodat", FD, Path, Mode), + 54: makeSyscallInfo("fchownat", FD, Path, Hex, Hex, Hex), + 55: makeSyscallInfo("fchown", FD, Hex, Hex), + 56: makeSyscallInfo("openat", FD, Path, OpenFlags, Mode), + 57: makeSyscallInfo("close", FD), + 58: makeSyscallInfo("vhangup"), + 59: makeSyscallInfo("pipe2", PipeFDs, Hex), + 60: makeSyscallInfo("quotactl", Hex, Hex, Hex, Hex), + 61: makeSyscallInfo("getdents64", FD, Hex, Hex), + 62: makeSyscallInfo("lseek", Hex, Hex, Hex), + 63: makeSyscallInfo("read", FD, ReadBuffer, Hex), + 64: makeSyscallInfo("write", FD, WriteBuffer, Hex), + 65: makeSyscallInfo("readv", FD, ReadIOVec, Hex), + 66: makeSyscallInfo("writev", FD, WriteIOVec, Hex), + 67: makeSyscallInfo("pread64", FD, ReadBuffer, Hex, Hex), + 68: makeSyscallInfo("pwrite64", FD, WriteBuffer, Hex, Hex), + 69: makeSyscallInfo("preadv", FD, ReadIOVec, Hex, Hex), + 70: makeSyscallInfo("pwritev", FD, WriteIOVec, Hex, Hex), + 71: makeSyscallInfo("sendfile", FD, FD, Hex, Hex), + 72: makeSyscallInfo("pselect6", Hex, Hex, Hex, Hex, Hex, Hex), + 73: makeSyscallInfo("ppoll", PollFDs, Hex, Timespec, SigSet, Hex), + 74: makeSyscallInfo("signalfd4", Hex, Hex, Hex, Hex), + 75: makeSyscallInfo("vmsplice", FD, Hex, Hex, Hex), + 76: makeSyscallInfo("splice", FD, Hex, FD, Hex, Hex, Hex), + 77: makeSyscallInfo("tee", FD, FD, Hex, Hex), + 78: makeSyscallInfo("readlinkat", FD, Path, ReadBuffer, Hex), + 79: makeSyscallInfo("fstatat", FD, Path, Stat, Hex), + 80: makeSyscallInfo("fstat", FD, Stat), + 81: makeSyscallInfo("sync"), + 82: makeSyscallInfo("fsync", FD), + 83: makeSyscallInfo("fdatasync", FD), + 84: makeSyscallInfo("sync_file_range", FD, Hex, Hex, Hex), + 85: makeSyscallInfo("timerfd_create", Hex, Hex), + 86: makeSyscallInfo("timerfd_settime", FD, Hex, ItimerSpec, PostItimerSpec), + 87: makeSyscallInfo("timerfd_gettime", FD, PostItimerSpec), + 88: makeSyscallInfo("utimensat", FD, Path, UTimeTimespec, Hex), + 89: makeSyscallInfo("acct", Hex), + 90: makeSyscallInfo("capget", CapHeader, PostCapData), + 91: makeSyscallInfo("capset", CapHeader, CapData), + 92: makeSyscallInfo("personality", Hex), + 93: makeSyscallInfo("exit", Hex), + 94: makeSyscallInfo("exit_group", Hex), + 95: makeSyscallInfo("waitid", Hex, Hex, Hex, Hex, Rusage), + 96: makeSyscallInfo("set_tid_address", Hex), + 97: makeSyscallInfo("unshare", CloneFlags), + 98: makeSyscallInfo("futex", Hex, FutexOp, Hex, Timespec, Hex, Hex), + 99: makeSyscallInfo("set_robust_list", Hex, Hex), + 100: makeSyscallInfo("get_robust_list", Hex, Hex, Hex), + 101: makeSyscallInfo("nanosleep", Timespec, PostTimespec), + 102: makeSyscallInfo("getitimer", ItimerType, PostItimerVal), + 103: makeSyscallInfo("setitimer", ItimerType, ItimerVal, PostItimerVal), + 104: makeSyscallInfo("kexec_load", Hex, Hex, Hex, Hex), + 105: makeSyscallInfo("init_module", Hex, Hex, Hex), + 106: makeSyscallInfo("delete_module", Hex, Hex), + 107: makeSyscallInfo("timer_create", Hex, Hex, Hex), + 108: makeSyscallInfo("timer_gettime", Hex, PostItimerSpec), + 109: makeSyscallInfo("timer_getoverrun", Hex), + 110: makeSyscallInfo("timer_settime", Hex, Hex, ItimerSpec, PostItimerSpec), + 111: makeSyscallInfo("timer_delete", Hex), + 112: makeSyscallInfo("clock_settime", Hex, Timespec), + 113: makeSyscallInfo("clock_gettime", Hex, PostTimespec), + 114: makeSyscallInfo("clock_getres", Hex, PostTimespec), + 115: makeSyscallInfo("clock_nanosleep", Hex, Hex, Timespec, PostTimespec), + 116: makeSyscallInfo("syslog", Hex, Hex, Hex), + 117: makeSyscallInfo("ptrace", PtraceRequest, Hex, Hex, Hex), + 118: makeSyscallInfo("sched_setparam", Hex, Hex), + 119: makeSyscallInfo("sched_setscheduler", Hex, Hex, Hex), + 120: makeSyscallInfo("sched_getscheduler", Hex), + 121: makeSyscallInfo("sched_getparam", Hex, Hex), + 122: makeSyscallInfo("sched_setaffinity", Hex, Hex, Hex), + 123: makeSyscallInfo("sched_getaffinity", Hex, Hex, Hex), + 124: makeSyscallInfo("sched_yield"), + 125: makeSyscallInfo("sched_get_priority_max", Hex), + 126: makeSyscallInfo("sched_get_priority_min", Hex), + 127: makeSyscallInfo("sched_rr_get_interval", Hex, Hex), + 128: makeSyscallInfo("restart_syscall"), + 129: makeSyscallInfo("kill", Hex, Signal), + 130: makeSyscallInfo("tkill", Hex, Signal), + 131: makeSyscallInfo("tgkill", Hex, Hex, Signal), + 132: makeSyscallInfo("sigaltstack", Hex, Hex), + 133: makeSyscallInfo("rt_sigsuspend", Hex), + 134: makeSyscallInfo("rt_sigaction", Signal, SigAction, PostSigAction), + 135: makeSyscallInfo("rt_sigprocmask", SignalMaskAction, SigSet, PostSigSet, Hex), + 136: makeSyscallInfo("rt_sigpending", Hex), + 137: makeSyscallInfo("rt_sigtimedwait", SigSet, Hex, Timespec, Hex), + 138: makeSyscallInfo("rt_sigqueueinfo", Hex, Signal, Hex), + 139: makeSyscallInfo("rt_sigreturn"), + 140: makeSyscallInfo("setpriority", Hex, Hex, Hex), + 141: makeSyscallInfo("getpriority", Hex, Hex), + 142: makeSyscallInfo("reboot", Hex, Hex, Hex, Hex), + 143: makeSyscallInfo("setregid", Hex, Hex), + 144: makeSyscallInfo("setgid", Hex), + 145: makeSyscallInfo("setreuid", Hex, Hex), + 146: makeSyscallInfo("setuid", Hex), + 147: makeSyscallInfo("setresuid", Hex, Hex, Hex), + 148: makeSyscallInfo("getresuid", Hex, Hex, Hex), + 149: makeSyscallInfo("setresgid", Hex, Hex, Hex), + 150: makeSyscallInfo("getresgid", Hex, Hex, Hex), + 151: makeSyscallInfo("setfsuid", Hex), + 152: makeSyscallInfo("setfsgid", Hex), + 153: makeSyscallInfo("times", Hex), + 154: makeSyscallInfo("setpgid", Hex, Hex), + 155: makeSyscallInfo("getpgid", Hex), + 156: makeSyscallInfo("getsid", Hex), + 157: makeSyscallInfo("setsid"), + 158: makeSyscallInfo("getgroups", Hex, Hex), + 159: makeSyscallInfo("setgroups", Hex, Hex), + 160: makeSyscallInfo("uname", Uname), + 161: makeSyscallInfo("sethostname", Hex, Hex), + 162: makeSyscallInfo("setdomainname", Hex, Hex), + 163: makeSyscallInfo("getrlimit", Hex, Hex), + 164: makeSyscallInfo("setrlimit", Hex, Hex), + 165: makeSyscallInfo("getrusage", Hex, Rusage), + 166: makeSyscallInfo("umask", Hex), + 167: makeSyscallInfo("prctl", Hex, Hex, Hex, Hex, Hex), + 168: makeSyscallInfo("getcpu", Hex, Hex, Hex), + 169: makeSyscallInfo("gettimeofday", Timeval, Hex), + 170: makeSyscallInfo("settimeofday", Timeval, Hex), + 171: makeSyscallInfo("adjtimex", Hex), + 172: makeSyscallInfo("getpid"), + 173: makeSyscallInfo("getppid"), + 174: makeSyscallInfo("getuid"), + 175: makeSyscallInfo("geteuid"), + 176: makeSyscallInfo("getgid"), + 177: makeSyscallInfo("getegid"), + 178: makeSyscallInfo("gettid"), + 179: makeSyscallInfo("sysinfo", Hex), + 180: makeSyscallInfo("mq_open", Hex, Hex, Hex, Hex), + 181: makeSyscallInfo("mq_unlink", Hex), + 182: makeSyscallInfo("mq_timedsend", Hex, Hex, Hex, Hex, Hex), + 183: makeSyscallInfo("mq_timedreceive", Hex, Hex, Hex, Hex, Hex), + 184: makeSyscallInfo("mq_notify", Hex, Hex), + 185: makeSyscallInfo("mq_getsetattr", Hex, Hex, Hex), + 186: makeSyscallInfo("msgget", Hex, Hex), + 187: makeSyscallInfo("msgctl", Hex, Hex, Hex), + 188: makeSyscallInfo("msgrcv", Hex, Hex, Hex, Hex, Hex), + 189: makeSyscallInfo("msgsnd", Hex, Hex, Hex, Hex), + 190: makeSyscallInfo("semget", Hex, Hex, Hex), + 191: makeSyscallInfo("semctl", Hex, Hex, Hex, Hex), + 192: makeSyscallInfo("semtimedop", Hex, Hex, Hex, Hex), + 193: makeSyscallInfo("semop", Hex, Hex, Hex), + 194: makeSyscallInfo("shmget", Hex, Hex, Hex), + 195: makeSyscallInfo("shmctl", Hex, Hex, Hex), + 196: makeSyscallInfo("shmat", Hex, Hex, Hex), + 197: makeSyscallInfo("shmdt", Hex), + 198: makeSyscallInfo("socket", SockFamily, SockType, SockProtocol), + 199: makeSyscallInfo("socketpair", SockFamily, SockType, SockProtocol, Hex), + 200: makeSyscallInfo("bind", FD, SockAddr, Hex), + 201: makeSyscallInfo("listen", FD, Hex), + 202: makeSyscallInfo("accept", FD, PostSockAddr, SockLen), + 203: makeSyscallInfo("connect", FD, SockAddr, Hex), + 204: makeSyscallInfo("getsockname", FD, PostSockAddr, SockLen), + 205: makeSyscallInfo("getpeername", FD, PostSockAddr, SockLen), + 206: makeSyscallInfo("sendto", FD, Hex, Hex, Hex, SockAddr, Hex), + 207: makeSyscallInfo("recvfrom", FD, Hex, Hex, Hex, PostSockAddr, SockLen), + 208: makeSyscallInfo("setsockopt", FD, Hex, Hex, Hex, Hex), + 209: makeSyscallInfo("getsockopt", FD, Hex, Hex, Hex, Hex), + 210: makeSyscallInfo("shutdown", FD, Hex), + 211: makeSyscallInfo("sendmsg", FD, SendMsgHdr, Hex), + 212: makeSyscallInfo("recvmsg", FD, RecvMsgHdr, Hex), + 213: makeSyscallInfo("readahead", Hex, Hex, Hex), + 214: makeSyscallInfo("brk", Hex), + 215: makeSyscallInfo("munmap", Hex, Hex), + 216: makeSyscallInfo("mremap", Hex, Hex, Hex, Hex, Hex), + 217: makeSyscallInfo("add_key", Hex, Hex, Hex, Hex, Hex), + 218: makeSyscallInfo("request_key", Hex, Hex, Hex, Hex), + 219: makeSyscallInfo("keyctl", Hex, Hex, Hex, Hex, Hex), + 220: makeSyscallInfo("clone", CloneFlags, Hex, Hex, Hex, Hex), + 221: makeSyscallInfo("execve", Path, ExecveStringVector, ExecveStringVector), + 222: makeSyscallInfo("mmap", Hex, Hex, Hex, Hex, FD, Hex), + 223: makeSyscallInfo("fadvise64", FD, Hex, Hex, Hex), + 224: makeSyscallInfo("swapon", Hex, Hex), + 225: makeSyscallInfo("swapoff", Hex), + 226: makeSyscallInfo("mprotect", Hex, Hex, Hex), + 227: makeSyscallInfo("msync", Hex, Hex, Hex), + 228: makeSyscallInfo("mlock", Hex, Hex), + 229: makeSyscallInfo("munlock", Hex, Hex), + 230: makeSyscallInfo("mlockall", Hex), + 231: makeSyscallInfo("munlockall"), + 232: makeSyscallInfo("mincore", Hex, Hex, Hex), + 233: makeSyscallInfo("madvise", Hex, Hex, Hex), + 234: makeSyscallInfo("remap_file_pages", Hex, Hex, Hex, Hex, Hex), + 235: makeSyscallInfo("mbind", Hex, Hex, Hex, Hex, Hex, Hex), + 236: makeSyscallInfo("get_mempolicy", Hex, Hex, Hex, Hex, Hex), + 237: makeSyscallInfo("set_mempolicy", Hex, Hex, Hex), + 238: makeSyscallInfo("migrate_pages", Hex, Hex, Hex, Hex), + 239: makeSyscallInfo("move_pages", Hex, Hex, Hex, Hex, Hex, Hex), + 240: makeSyscallInfo("rt_tgsigqueueinfo", Hex, Hex, Signal, Hex), + 241: makeSyscallInfo("perf_event_open", Hex, Hex, Hex, Hex, Hex), + 242: makeSyscallInfo("accept4", FD, PostSockAddr, SockLen, SockFlags), + 243: makeSyscallInfo("recvmmsg", FD, Hex, Hex, Hex, Hex), + + 260: makeSyscallInfo("wait4", Hex, Hex, Hex, Rusage), + 261: makeSyscallInfo("prlimit64", Hex, Hex, Hex, Hex), + 262: makeSyscallInfo("fanotify_init", Hex, Hex), + 263: makeSyscallInfo("fanotify_mark", Hex, Hex, Hex, Hex, Hex), + 264: makeSyscallInfo("name_to_handle_at", FD, Hex, Hex, Hex, Hex), + 265: makeSyscallInfo("open_by_handle_at", FD, Hex, Hex), + 266: makeSyscallInfo("clock_adjtime", Hex, Hex), + 267: makeSyscallInfo("syncfs", FD), + 268: makeSyscallInfo("setns", FD, Hex), + 269: makeSyscallInfo("sendmmsg", FD, Hex, Hex, Hex), + 270: makeSyscallInfo("process_vm_readv", Hex, ReadIOVec, Hex, IOVec, Hex, Hex), + 271: makeSyscallInfo("process_vm_writev", Hex, IOVec, Hex, WriteIOVec, Hex, Hex), + 272: makeSyscallInfo("kcmp", Hex, Hex, Hex, Hex, Hex), + 273: makeSyscallInfo("finit_module", Hex, Hex, Hex), + 274: makeSyscallInfo("sched_setattr", Hex, Hex, Hex), + 275: makeSyscallInfo("sched_getattr", Hex, Hex, Hex), + 276: makeSyscallInfo("renameat2", FD, Path, Hex, Path, Hex), + 277: makeSyscallInfo("seccomp", Hex, Hex, Hex), + 278: makeSyscallInfo("getrandom", Hex, Hex, Hex), + 279: makeSyscallInfo("memfd_create", Path, Hex), + 280: makeSyscallInfo("bpf", Hex, Hex, Hex), + 281: makeSyscallInfo("execveat", FD, Path, Hex, Hex, Hex), + 282: makeSyscallInfo("userfaultfd", Hex), + 283: makeSyscallInfo("membarrier", Hex), + 284: makeSyscallInfo("mlock2", Hex, Hex, Hex), + 285: makeSyscallInfo("copy_file_range", FD, Hex, FD, Hex, Hex, Hex), + 286: makeSyscallInfo("preadv2", FD, ReadIOVec, Hex, Hex, Hex), + 287: makeSyscallInfo("pwritev2", FD, WriteIOVec, Hex, Hex, Hex), + 291: makeSyscallInfo("statx", FD, Path, Hex, Hex, Hex), + 292: makeSyscallInfo("io_pgetevents", Hex, Hex, Hex, Hex, Timespec, SigSet), + 293: makeSyscallInfo("rseq", Hex, Hex, Hex, Hex), + 424: makeSyscallInfo("pidfd_send_signal", FD, Signal, Hex, Hex), + 425: makeSyscallInfo("io_uring_setup", Hex, Hex), + 426: makeSyscallInfo("io_uring_enter", FD, Hex, Hex, Hex, SigSet, Hex), + 427: makeSyscallInfo("io_uring_register", FD, Hex, Hex, Hex), + 428: makeSyscallInfo("open_tree", FD, Path, Hex), + 429: makeSyscallInfo("move_mount", FD, Path, FD, Path, Hex), + 430: makeSyscallInfo("fsopen", Path, Hex), // Not quite a path, but close. + 431: makeSyscallInfo("fsconfig", FD, Hex, Hex, Hex, Hex), + 432: makeSyscallInfo("fsmount", FD, Hex, Hex), + 433: makeSyscallInfo("fspick", FD, Path, Hex), + 434: makeSyscallInfo("pidfd_open", Hex, Hex), + 435: makeSyscallInfo("clone3", Hex, Hex), +} + +func init() { + syscallTables = append(syscallTables, + syscallTable{ + os: abi.Linux, + arch: arch.ARM64, + syscalls: linuxARM64}) +} diff --git a/pkg/sentry/strace/syscalls.go b/pkg/sentry/strace/syscalls.go index e5d486c4e..24e29a2ba 100644 --- a/pkg/sentry/strace/syscalls.go +++ b/pkg/sentry/strace/syscalls.go @@ -250,14 +250,7 @@ type syscallTable struct { syscalls SyscallMap } -// syscallTables contains all syscall tables. -var syscallTables = []syscallTable{ - { - os: abi.Linux, - arch: arch.AMD64, - syscalls: linuxAMD64, - }, -} +var syscallTables []syscallTable // Lookup returns the SyscallMap for the OS/Arch combination. The returned map // must not be changed. -- cgit v1.2.3 From eb868be7434600c41fe9aae8313166607869ecb3 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Wed, 4 Dec 2019 16:39:47 +0800 Subject: supporting lazy-fpsimd in guest on Arm64 Several jobs were finished in this patch: 1, provide functions to get/set fpcr/fpsr/vregs 2, support lazy-fpsimd-context-switch in el1 Signed-off-by: Bin Lu --- pkg/sentry/platform/ring0/BUILD | 1 + pkg/sentry/platform/ring0/defs_arm64.go | 3 + pkg/sentry/platform/ring0/entry_arm64.s | 26 +++++++ pkg/sentry/platform/ring0/lib_arm64.go | 26 +++++-- pkg/sentry/platform/ring0/lib_arm64.s | 118 +++++++++++++++++++++++++++++ pkg/sentry/platform/ring0/offsets_arm64.go | 1 + 6 files changed, 169 insertions(+), 6 deletions(-) create mode 100644 pkg/sentry/platform/ring0/lib_arm64.s (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/ring0/BUILD b/pkg/sentry/platform/ring0/BUILD index f1af18265..87f4552b5 100644 --- a/pkg/sentry/platform/ring0/BUILD +++ b/pkg/sentry/platform/ring0/BUILD @@ -71,6 +71,7 @@ go_library( "lib_amd64.go", "lib_amd64.s", "lib_arm64.go", + "lib_arm64.s", "ring0.go", ], importpath = "gvisor.dev/gvisor/pkg/sentry/platform/ring0", diff --git a/pkg/sentry/platform/ring0/defs_arm64.go b/pkg/sentry/platform/ring0/defs_arm64.go index fbfbd9bab..dc0eeec01 100644 --- a/pkg/sentry/platform/ring0/defs_arm64.go +++ b/pkg/sentry/platform/ring0/defs_arm64.go @@ -73,6 +73,9 @@ type CPUArchState struct { // application context pointer. appAddr uintptr + + // lazyVFP is the value of cpacr_el1. + lazyVFP uintptr } // ErrorCode returns the last error code. diff --git a/pkg/sentry/platform/ring0/entry_arm64.s b/pkg/sentry/platform/ring0/entry_arm64.s index 29c475882..add2c3e08 100644 --- a/pkg/sentry/platform/ring0/entry_arm64.s +++ b/pkg/sentry/platform/ring0/entry_arm64.s @@ -31,6 +31,11 @@ #define RSV_REG R18_PLATFORM #define RSV_REG_APP R9 +#define FPEN_NOTRAP 0x3 +#define FPEN_SHIFT 20 + +#define FPEN_ENABLE (FPEN_NOTRAP << FPEN_SHIFT) + #define REGISTERS_SAVE(reg, offset) \ MOVD R0, offset+PTRACE_R0(reg); \ MOVD R1, offset+PTRACE_R1(reg); \ @@ -279,6 +284,16 @@ #define IRQ_DISABLE \ MSR $2, DAIFClr; +#define VFP_ENABLE \ + MOVD $FPEN_ENABLE, R0; \ + WORD $0xd5181040; \ //MSR R0, CPACR_EL1 + ISB $15; + +#define VFP_DISABLE \ + MOVD $0x0, R0; \ + WORD $0xd5181040; \ //MSR R0, CPACR_EL1 + ISB $15; + #define KERNEL_ENTRY_FROM_EL0 \ SUB $16, RSP, RSP; \ // step1, save r18, r9 into kernel temporary stack. STP (RSV_REG, RSV_REG_APP), 16*0(RSP); \ @@ -318,6 +333,11 @@ TEXT ·Halt(SB),NOSPLIT,$0 BNE mmio_exit MOVD $0, CPU_REGISTERS+PTRACE_R9(RSV_REG) mmio_exit: + // Disable fpsimd. + WORD $0xd5381041 // MRS CPACR_EL1, R1 + MOVD R1, CPU_LAZY_VFP(RSV_REG) + VFP_DISABLE + // MMIO_EXIT. MOVD $0, R9 MOVD R0, 0xffff000000001000(R9) @@ -382,6 +402,8 @@ TEXT ·El1_sync(SB),NOSPLIT,$0 BEQ el1_svc CMP $ESR_ELx_EC_BREAKPT_CUR, R24 BGE el1_dbg + CMP $ESR_ELx_EC_FP_ASIMD, R24 + BEQ el1_fpsimd_acc B el1_invalid el1_da: @@ -402,6 +424,10 @@ el1_svc: el1_dbg: B ·Shutdown(SB) +el1_fpsimd_acc: + VFP_ENABLE + B ·kernelExitToEl1(SB) // Resume. + el1_invalid: B ·Shutdown(SB) diff --git a/pkg/sentry/platform/ring0/lib_arm64.go b/pkg/sentry/platform/ring0/lib_arm64.go index 900ee6380..3f3ab5cfb 100644 --- a/pkg/sentry/platform/ring0/lib_arm64.go +++ b/pkg/sentry/platform/ring0/lib_arm64.go @@ -16,10 +16,24 @@ package ring0 -// LoadFloatingPoint loads floating point state by the most efficient mechanism -// available (set by Init). -var LoadFloatingPoint func(*byte) +// CPACREL1 returns the value of the CPACR_EL1 register. +func CPACREL1() (value uintptr) -// SaveFloatingPoint saves floating point state by the most efficient mechanism -// available (set by Init). -var SaveFloatingPoint func(*byte) +// GetFPCR returns the value of FPCR register. +func GetFPCR() (value uintptr) + +// SetFPCR writes the FPCR value. +func SetFPCR(value uintptr) + +// GetFPSR returns the value of FPSR register. +func GetFPSR() (value uintptr) + +// SetFPSR writes the FPSR value. +func SetFPSR(value uintptr) + +// SaveVRegs saves V0-V31 registers. +// V0-V31: 32 128-bit registers for floating point and simd. +func SaveVRegs(*byte) + +// LoadVRegs loads V0-V31 registers. +func LoadVRegs(*byte) diff --git a/pkg/sentry/platform/ring0/lib_arm64.s b/pkg/sentry/platform/ring0/lib_arm64.s new file mode 100644 index 000000000..7c96e8e9c --- /dev/null +++ b/pkg/sentry/platform/ring0/lib_arm64.s @@ -0,0 +1,118 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +TEXT ·CPACREL1(SB),NOSPLIT,$0-8 + WORD $0xd5381041 // MRS CPACR_EL1, R1 + MOVD R1, ret+0(FP) + RET + +TEXT ·GetFPCR(SB),NOSPLIT,$0-8 + WORD $0xd53b4201 // MRS NZCV, R1 + MOVD R1, ret+0(FP) + RET + +TEXT ·GetFPSR(SB),NOSPLIT,$0-8 + WORD $0xd53b4421 // MRS FPSR, R1 + MOVD R1, ret+0(FP) + RET + +TEXT ·SetFPCR(SB),NOSPLIT,$0-8 + MOVD addr+0(FP), R1 + WORD $0xd51b4201 // MSR R1, NZCV + RET + +TEXT ·SetFPSR(SB),NOSPLIT,$0-8 + MOVD addr+0(FP), R1 + WORD $0xd51b4421 // MSR R1, FPSR + RET + +TEXT ·SaveVRegs(SB),NOSPLIT,$0-8 + MOVD addr+0(FP), R0 + + // Skip aarch64_ctx, fpsr, fpcr. + FMOVD F0, 16*1(R0) + FMOVD F1, 16*2(R0) + FMOVD F2, 16*3(R0) + FMOVD F3, 16*4(R0) + FMOVD F4, 16*5(R0) + FMOVD F5, 16*6(R0) + FMOVD F6, 16*7(R0) + FMOVD F7, 16*8(R0) + FMOVD F8, 16*9(R0) + FMOVD F9, 16*10(R0) + FMOVD F10, 16*11(R0) + FMOVD F11, 16*12(R0) + FMOVD F12, 16*13(R0) + FMOVD F13, 16*14(R0) + FMOVD F14, 16*15(R0) + FMOVD F15, 16*16(R0) + FMOVD F16, 16*17(R0) + FMOVD F17, 16*18(R0) + FMOVD F18, 16*19(R0) + FMOVD F19, 16*20(R0) + FMOVD F20, 16*21(R0) + FMOVD F21, 16*22(R0) + FMOVD F22, 16*23(R0) + FMOVD F23, 16*24(R0) + FMOVD F24, 16*25(R0) + FMOVD F25, 16*26(R0) + FMOVD F26, 16*27(R0) + FMOVD F27, 16*28(R0) + FMOVD F28, 16*29(R0) + FMOVD F29, 16*30(R0) + FMOVD F30, 16*31(R0) + FMOVD F31, 16*32(R0) + ISB $15 + + RET + +TEXT ·LoadVRegs(SB),NOSPLIT,$0-8 + MOVD addr+0(FP), R0 + + // Skip aarch64_ctx, fpsr, fpcr. + FMOVD 16*1(R0), F0 + FMOVD 16*2(R0), F1 + FMOVD 16*3(R0), F2 + FMOVD 16*4(R0), F3 + FMOVD 16*5(R0), F4 + FMOVD 16*6(R0), F5 + FMOVD 16*7(R0), F6 + FMOVD 16*8(R0), F7 + FMOVD 16*9(R0), F8 + FMOVD 16*10(R0), F9 + FMOVD 16*11(R0), F10 + FMOVD 16*12(R0), F11 + FMOVD 16*13(R0), F12 + FMOVD 16*14(R0), F13 + FMOVD 16*15(R0), F14 + FMOVD 16*16(R0), F15 + FMOVD 16*17(R0), F16 + FMOVD 16*18(R0), F17 + FMOVD 16*19(R0), F18 + FMOVD 16*20(R0), F19 + FMOVD 16*21(R0), F20 + FMOVD 16*22(R0), F21 + FMOVD 16*23(R0), F22 + FMOVD 16*24(R0), F23 + FMOVD 16*25(R0), F24 + FMOVD 16*26(R0), F25 + FMOVD 16*27(R0), F26 + FMOVD 16*28(R0), F27 + FMOVD 16*29(R0), F28 + FMOVD 16*30(R0), F29 + FMOVD 16*31(R0), F30 + FMOVD 16*32(R0), F31 + ISB $15 + + RET diff --git a/pkg/sentry/platform/ring0/offsets_arm64.go b/pkg/sentry/platform/ring0/offsets_arm64.go index d7aa1c7cc..cd2a65f97 100644 --- a/pkg/sentry/platform/ring0/offsets_arm64.go +++ b/pkg/sentry/platform/ring0/offsets_arm64.go @@ -39,6 +39,7 @@ func Emit(w io.Writer) { fmt.Fprintf(w, "#define CPU_TTBR0_APP 0x%02x\n", reflect.ValueOf(&c.ttbr0App).Pointer()-reflect.ValueOf(c).Pointer()) fmt.Fprintf(w, "#define CPU_VECTOR_CODE 0x%02x\n", reflect.ValueOf(&c.vecCode).Pointer()-reflect.ValueOf(c).Pointer()) fmt.Fprintf(w, "#define CPU_APP_ADDR 0x%02x\n", reflect.ValueOf(&c.appAddr).Pointer()-reflect.ValueOf(c).Pointer()) + fmt.Fprintf(w, "#define CPU_LAZY_VFP 0x%02x\n", reflect.ValueOf(&c.lazyVFP).Pointer()-reflect.ValueOf(c).Pointer()) fmt.Fprintf(w, "\n// Bits.\n") fmt.Fprintf(w, "#define _KERNEL_FLAGS 0x%02x\n", KernelFlagsSet) -- cgit v1.2.3 From 334a513f11f0ecc260abcce549b1f1a74edc2c51 Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Wed, 18 Dec 2019 12:59:32 -0800 Subject: Add Mems_allowed to /proc/PID/status PiperOrigin-RevId: 286248378 --- pkg/sentry/fs/proc/task.go | 4 ++++ pkg/sentry/fsimpl/proc/task.go | 4 ++++ 2 files changed, 8 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go index 0e46c5fb7..9bf4b4527 100644 --- a/pkg/sentry/fs/proc/task.go +++ b/pkg/sentry/fs/proc/task.go @@ -604,6 +604,10 @@ func (s *statusData) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ( fmt.Fprintf(&buf, "CapEff:\t%016x\n", creds.EffectiveCaps) fmt.Fprintf(&buf, "CapBnd:\t%016x\n", creds.BoundingCaps) fmt.Fprintf(&buf, "Seccomp:\t%d\n", s.t.SeccompMode()) + // We unconditionally report a single NUMA node. See + // pkg/sentry/syscalls/linux/sys_mempolicy.go. + fmt.Fprintf(&buf, "Mems_allowed:\t1\n") + fmt.Fprintf(&buf, "Mems_allowed_list:\t0\n") return []seqfile.SeqData{{Buf: buf.Bytes(), Handle: (*statusData)(nil)}}, 0 } diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index c46e05c3a..0d87be52b 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -229,6 +229,10 @@ func (s *statusData) Generate(ctx context.Context, buf *bytes.Buffer) error { fmt.Fprintf(buf, "CapEff:\t%016x\n", creds.EffectiveCaps) fmt.Fprintf(buf, "CapBnd:\t%016x\n", creds.BoundingCaps) fmt.Fprintf(buf, "Seccomp:\t%d\n", s.t.SeccompMode()) + // We unconditionally report a single NUMA node. See + // pkg/sentry/syscalls/linux/sys_mempolicy.go. + fmt.Fprintf(buf, "Mems_allowed:\t1\n") + fmt.Fprintf(buf, "Mems_allowed_list:\t0\n") return nil } -- cgit v1.2.3 From 744401297a8c93ce5992ba99aa84f3dcdc19ae9e Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Wed, 18 Dec 2019 15:47:24 -0800 Subject: Add VFS2 plumbing for extended attributes. PiperOrigin-RevId: 286281274 --- pkg/sentry/fsimpl/ext/filesystem.go | 36 ++++++++++++ pkg/sentry/fsimpl/kernfs/filesystem.go | 52 +++++++++++++++++ pkg/sentry/fsimpl/memfs/filesystem.go | 48 +++++++++++++++ pkg/sentry/vfs/file_description.go | 49 +++++++++++++++- pkg/sentry/vfs/file_description_impl_util.go | 25 ++++++++ pkg/sentry/vfs/filesystem.go | 16 ++++- pkg/sentry/vfs/options.go | 14 +++++ pkg/sentry/vfs/testutil.go | 20 +++++++ pkg/sentry/vfs/vfs.go | 87 ++++++++++++++++++++++++++++ 9 files changed, 345 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/ext/filesystem.go b/pkg/sentry/fsimpl/ext/filesystem.go index e7aa3b41b..d7e87979a 100644 --- a/pkg/sentry/fsimpl/ext/filesystem.go +++ b/pkg/sentry/fsimpl/ext/filesystem.go @@ -443,6 +443,42 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error return syserror.EROFS } +// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. +func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) { + _, _, err := fs.walk(rp, false) + if err != nil { + return nil, err + } + return nil, syserror.ENOTSUP +} + +// GetxattrAt implements vfs.FilesystemImpl.GetxattrAt. +func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) (string, error) { + _, _, err := fs.walk(rp, false) + if err != nil { + return "", err + } + return "", syserror.ENOTSUP +} + +// SetxattrAt implements vfs.FilesystemImpl.SetxattrAt. +func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { + _, _, err := fs.walk(rp, false) + if err != nil { + return err + } + return syserror.ENOTSUP +} + +// RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. +func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { + _, _, err := fs.walk(rp, false) + if err != nil { + return err + } + return syserror.ENOTSUP +} + // PrependPath implements vfs.FilesystemImpl.PrependPath. func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { fs.mu.RLock() diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index db486b6c1..3cbbe4b20 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -683,6 +683,58 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error return nil } +// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. +func (fs *Filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) { + fs.mu.RLock() + _, _, err := fs.walkExistingLocked(ctx, rp) + fs.mu.RUnlock() + fs.processDeferredDecRefs() + if err != nil { + return nil, err + } + // kernfs currently does not support extended attributes. + return nil, syserror.ENOTSUP +} + +// GetxattrAt implements vfs.FilesystemImpl.GetxattrAt. +func (fs *Filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) (string, error) { + fs.mu.RLock() + _, _, err := fs.walkExistingLocked(ctx, rp) + fs.mu.RUnlock() + fs.processDeferredDecRefs() + if err != nil { + return "", err + } + // kernfs currently does not support extended attributes. + return "", syserror.ENOTSUP +} + +// SetxattrAt implements vfs.FilesystemImpl.SetxattrAt. +func (fs *Filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { + fs.mu.RLock() + _, _, err := fs.walkExistingLocked(ctx, rp) + fs.mu.RUnlock() + fs.processDeferredDecRefs() + if err != nil { + return err + } + // kernfs currently does not support extended attributes. + return syserror.ENOTSUP +} + +// RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. +func (fs *Filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { + fs.mu.RLock() + _, _, err := fs.walkExistingLocked(ctx, rp) + fs.mu.RUnlock() + fs.processDeferredDecRefs() + if err != nil { + return err + } + // kernfs currently does not support extended attributes. + return syserror.ENOTSUP +} + // PrependPath implements vfs.FilesystemImpl.PrependPath. func (fs *Filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { fs.mu.RLock() diff --git a/pkg/sentry/fsimpl/memfs/filesystem.go b/pkg/sentry/fsimpl/memfs/filesystem.go index 1f2a5122a..22f1e811f 100644 --- a/pkg/sentry/fsimpl/memfs/filesystem.go +++ b/pkg/sentry/fsimpl/memfs/filesystem.go @@ -584,6 +584,54 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error return nil } +// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. +func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) { + fs.mu.RLock() + defer fs.mu.RUnlock() + _, _, err := walkExistingLocked(rp) + if err != nil { + return nil, err + } + // TODO(b/127675828): support extended attributes + return nil, syserror.ENOTSUP +} + +// GetxattrAt implements vfs.FilesystemImpl.GetxattrAt. +func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) (string, error) { + fs.mu.RLock() + defer fs.mu.RUnlock() + _, _, err := walkExistingLocked(rp) + if err != nil { + return "", err + } + // TODO(b/127675828): support extended attributes + return "", syserror.ENOTSUP +} + +// SetxattrAt implements vfs.FilesystemImpl.SetxattrAt. +func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { + fs.mu.RLock() + defer fs.mu.RUnlock() + _, _, err := walkExistingLocked(rp) + if err != nil { + return err + } + // TODO(b/127675828): support extended attributes + return syserror.ENOTSUP +} + +// RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. +func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { + fs.mu.RLock() + defer fs.mu.RUnlock() + _, _, err := walkExistingLocked(rp) + if err != nil { + return err + } + // TODO(b/127675828): support extended attributes + return syserror.ENOTSUP +} + // PrependPath implements vfs.FilesystemImpl.PrependPath. func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { fs.mu.RLock() diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 6575afd16..c5a9adca3 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -22,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/waiter" ) @@ -212,7 +213,21 @@ type FileDescriptionImpl interface { // Ioctl implements the ioctl(2) syscall. Ioctl(ctx context.Context, uio usermem.IO, args arch.SyscallArguments) (uintptr, error) - // TODO: extended attributes; file locking + // Listxattr returns all extended attribute names for the file. + Listxattr(ctx context.Context) ([]string, error) + + // Getxattr returns the value associated with the given extended attribute + // for the file. + Getxattr(ctx context.Context, name string) (string, error) + + // Setxattr changes the value associated with the given extended attribute + // for the file. + Setxattr(ctx context.Context, opts SetxattrOptions) error + + // Removexattr removes the given extended attribute from the file. + Removexattr(ctx context.Context, name string) error + + // TODO: file locking } // Dirent holds the information contained in struct linux_dirent64. @@ -329,6 +344,38 @@ func (fd *FileDescription) Ioctl(ctx context.Context, uio usermem.IO, args arch. return fd.impl.Ioctl(ctx, uio, args) } +// Listxattr returns all extended attribute names for the file represented by +// fd. +func (fd *FileDescription) Listxattr(ctx context.Context) ([]string, error) { + names, err := fd.impl.Listxattr(ctx) + if err == syserror.ENOTSUP { + // Linux doesn't actually return ENOTSUP in this case; instead, + // fs/xattr.c:vfs_listxattr() falls back to allowing the security + // subsystem to return security extended attributes, which by default + // don't exist. + return nil, nil + } + return names, err +} + +// Getxattr returns the value associated with the given extended attribute for +// the file represented by fd. +func (fd *FileDescription) Getxattr(ctx context.Context, name string) (string, error) { + return fd.impl.Getxattr(ctx, name) +} + +// Setxattr changes the value associated with the given extended attribute for +// the file represented by fd. +func (fd *FileDescription) Setxattr(ctx context.Context, opts SetxattrOptions) error { + return fd.impl.Setxattr(ctx, opts) +} + +// Removexattr removes the given extended attribute from the file represented +// by fd. +func (fd *FileDescription) Removexattr(ctx context.Context, name string) error { + return fd.impl.Removexattr(ctx, name) +} + // SyncFS instructs the filesystem containing fd to execute the semantics of // syncfs(2). func (fd *FileDescription) SyncFS(ctx context.Context) error { diff --git a/pkg/sentry/vfs/file_description_impl_util.go b/pkg/sentry/vfs/file_description_impl_util.go index aae023254..3df49991c 100644 --- a/pkg/sentry/vfs/file_description_impl_util.go +++ b/pkg/sentry/vfs/file_description_impl_util.go @@ -127,6 +127,31 @@ func (FileDescriptionDefaultImpl) Ioctl(ctx context.Context, uio usermem.IO, arg return 0, syserror.ENOTTY } +// Listxattr implements FileDescriptionImpl.Listxattr analogously to +// inode_operations::listxattr == NULL in Linux. +func (FileDescriptionDefaultImpl) Listxattr(ctx context.Context) ([]string, error) { + // This isn't exactly accurate; see FileDescription.Listxattr. + return nil, syserror.ENOTSUP +} + +// Getxattr implements FileDescriptionImpl.Getxattr analogously to +// inode::i_opflags & IOP_XATTR == 0 in Linux. +func (FileDescriptionDefaultImpl) Getxattr(ctx context.Context, name string) (string, error) { + return "", syserror.ENOTSUP +} + +// Setxattr implements FileDescriptionImpl.Setxattr analogously to +// inode::i_opflags & IOP_XATTR == 0 in Linux. +func (FileDescriptionDefaultImpl) Setxattr(ctx context.Context, opts SetxattrOptions) error { + return syserror.ENOTSUP +} + +// Removexattr implements FileDescriptionImpl.Removexattr analogously to +// inode::i_opflags & IOP_XATTR == 0 in Linux. +func (FileDescriptionDefaultImpl) Removexattr(ctx context.Context, name string) error { + return syserror.ENOTSUP +} + // DirectoryFileDescriptionDefaultImpl may be embedded by implementations of // FileDescriptionImpl that always represent directories to obtain // implementations of non-directory I/O methods that return EISDIR. diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index 8011eba3f..b766614e7 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -186,6 +186,20 @@ type FilesystemImpl interface { // UnlinkAt removes the non-directory file at rp. UnlinkAt(ctx context.Context, rp *ResolvingPath) error + // ListxattrAt returns all extended attribute names for the file at rp. + ListxattrAt(ctx context.Context, rp *ResolvingPath) ([]string, error) + + // GetxattrAt returns the value associated with the given extended + // attribute for the file at rp. + GetxattrAt(ctx context.Context, rp *ResolvingPath, name string) (string, error) + + // SetxattrAt changes the value associated with the given extended + // attribute for the file at rp. + SetxattrAt(ctx context.Context, rp *ResolvingPath, opts SetxattrOptions) error + + // RemovexattrAt removes the given extended attribute from the file at rp. + RemovexattrAt(ctx context.Context, rp *ResolvingPath, name string) error + // PrependPath prepends a path from vd to vd.Mount().Root() to b. // // If vfsroot.Ok(), it is the contextual VFS root; if it is encountered @@ -208,7 +222,7 @@ type FilesystemImpl interface { // Preconditions: vd.Mount().Filesystem().Impl() == this FilesystemImpl. PrependPath(ctx context.Context, vfsroot, vd VirtualDentry, b *fspath.Builder) error - // TODO: extended attributes; inotify_add_watch(); bind() + // TODO: inotify_add_watch(); bind() } // PrependPathAtVFSRootError is returned by implementations of diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go index 3ecbc8fc1..97ee4a446 100644 --- a/pkg/sentry/vfs/options.go +++ b/pkg/sentry/vfs/options.go @@ -101,6 +101,20 @@ type SetStatOptions struct { Stat linux.Statx } +// SetxattrOptions contains options to VirtualFilesystem.SetxattrAt(), +// FilesystemImpl.SetxattrAt(), FileDescription.Setxattr(), and +// FileDescriptionImpl.Setxattr(). +type SetxattrOptions struct { + // Name is the name of the extended attribute being mutated. + Name string + + // Value is the extended attribute's new value. + Value string + + // Flags contains flags as specified for setxattr/lsetxattr/fsetxattr(2). + Flags uint32 +} + // StatOptions contains options to VirtualFilesystem.StatAt(), // FilesystemImpl.StatAt(), FileDescription.Stat(), and // FileDescriptionImpl.Stat(). diff --git a/pkg/sentry/vfs/testutil.go b/pkg/sentry/vfs/testutil.go index 7a1d9e383..d94117bce 100644 --- a/pkg/sentry/vfs/testutil.go +++ b/pkg/sentry/vfs/testutil.go @@ -117,6 +117,26 @@ func (fs *FDTestFilesystem) UnlinkAt(ctx context.Context, rp *ResolvingPath) err return syserror.EPERM } +// ListxattrAt implements FilesystemImpl.ListxattrAt. +func (fs *FDTestFilesystem) ListxattrAt(ctx context.Context, rp *ResolvingPath) ([]string, error) { + return nil, syserror.EPERM +} + +// GetxattrAt implements FilesystemImpl.GetxattrAt. +func (fs *FDTestFilesystem) GetxattrAt(ctx context.Context, rp *ResolvingPath, name string) (string, error) { + return "", syserror.EPERM +} + +// SetxattrAt implements FilesystemImpl.SetxattrAt. +func (fs *FDTestFilesystem) SetxattrAt(ctx context.Context, rp *ResolvingPath, opts SetxattrOptions) error { + return syserror.EPERM +} + +// RemovexattrAt implements FilesystemImpl.RemovexattrAt. +func (fs *FDTestFilesystem) RemovexattrAt(ctx context.Context, rp *ResolvingPath, name string) error { + return syserror.EPERM +} + // PrependPath implements FilesystemImpl.PrependPath. func (fs *FDTestFilesystem) PrependPath(ctx context.Context, vfsroot, vd VirtualDentry, b *fspath.Builder) error { b.PrependComponent(fmt.Sprintf("vfs.fdTestDentry:%p", vd.dentry.impl.(*fdTestDentry))) diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 7262b0d0a..e60898d7c 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -440,6 +440,93 @@ func (vfs *VirtualFilesystem) UnlinkAt(ctx context.Context, creds *auth.Credenti } } +// ListxattrAt returns all extended attribute names for the file at the given +// path. +func (vfs *VirtualFilesystem) ListxattrAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation) ([]string, error) { + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return nil, err + } + for { + names, err := rp.mount.fs.impl.ListxattrAt(ctx, rp) + if err == nil { + vfs.putResolvingPath(rp) + return names, nil + } + if err == syserror.ENOTSUP { + // Linux doesn't actually return ENOTSUP in this case; instead, + // fs/xattr.c:vfs_listxattr() falls back to allowing the security + // subsystem to return security extended attributes, which by + // default don't exist. + vfs.putResolvingPath(rp) + return nil, nil + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return nil, err + } + } +} + +// GetxattrAt returns the value associated with the given extended attribute +// for the file at the given path. +func (vfs *VirtualFilesystem) GetxattrAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, name string) (string, error) { + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return "", err + } + for { + val, err := rp.mount.fs.impl.GetxattrAt(ctx, rp, name) + if err == nil { + vfs.putResolvingPath(rp) + return val, nil + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return "", err + } + } +} + +// SetxattrAt changes the value associated with the given extended attribute +// for the file at the given path. +func (vfs *VirtualFilesystem) SetxattrAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *SetxattrOptions) error { + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return err + } + for { + err := rp.mount.fs.impl.SetxattrAt(ctx, rp, *opts) + if err == nil { + vfs.putResolvingPath(rp) + return nil + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return err + } + } +} + +// RemovexattrAt removes the given extended attribute from the file at rp. +func (vfs *VirtualFilesystem) RemovexattrAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, name string) error { + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return err + } + for { + err := rp.mount.fs.impl.RemovexattrAt(ctx, rp, name) + if err == nil { + vfs.putResolvingPath(rp) + return nil + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return err + } + } +} + // SyncAllFilesystems has the semantics of Linux's sync(2). func (vfs *VirtualFilesystem) SyncAllFilesystems(ctx context.Context) error { fss := make(map[*Filesystem]struct{}) -- cgit v1.2.3 From bb00438f36ebd19968246a838c2ddd61b9e14b79 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Thu, 19 Dec 2019 14:29:57 -0800 Subject: Make masterInodeOperations.Truncate take a pointer receiver. Otherwise a copy happens, which triggers a data race when reading masterInodeOperations.SimpleFileOperations.uattr, which must be accessed with a lock held. PiperOrigin-RevId: 286464473 --- pkg/sentry/fs/tty/master.go | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/tty/master.go b/pkg/sentry/fs/tty/master.go index 934828c12..6b07f6bf2 100644 --- a/pkg/sentry/fs/tty/master.go +++ b/pkg/sentry/fs/tty/master.go @@ -77,7 +77,7 @@ func (mi *masterInodeOperations) Release(ctx context.Context) { } // Truncate implements fs.InodeOperations.Truncate. -func (masterInodeOperations) Truncate(context.Context, *fs.Inode, int64) error { +func (*masterInodeOperations) Truncate(context.Context, *fs.Inode, int64) error { return nil } -- cgit v1.2.3 From 29955a4797e8264f75886a989dbc81b2b5443f4c Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Thu, 19 Dec 2019 17:25:18 -0800 Subject: futex: wake one waiter if futex_wake is called with a non-positive value This change is needed to be compatible with the Linux kernel. There is no glibc wrapper for the futex system call, so it is easy to make a mistake and call syscall(__NR_futex, FUTEX_WAKE, addr) without the fourth argument. This works on Linux, because it wakes one waiter even if val is nonpositive. PiperOrigin-RevId: 286494396 --- pkg/sentry/syscalls/linux/sys_futex.go | 10 ++++++++++ test/syscalls/linux/futex.cc | 21 +++++++++++++++++++++ 2 files changed, 31 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/sys_futex.go b/pkg/sentry/syscalls/linux/sys_futex.go index b9bd25464..bde17a767 100644 --- a/pkg/sentry/syscalls/linux/sys_futex.go +++ b/pkg/sentry/syscalls/linux/sys_futex.go @@ -226,6 +226,11 @@ func Futex(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if mask == 0 { return 0, nil, syserror.EINVAL } + if val <= 0 { + // The Linux kernel wakes one waiter even if val is + // non-positive. + val = 1 + } n, err := t.Futex().Wake(t, addr, private, mask, val) return uintptr(n), nil, err @@ -242,6 +247,11 @@ func Futex(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall case linux.FUTEX_WAKE_OP: op := uint32(val3) + if val <= 0 { + // The Linux kernel wakes one waiter even if val is + // non-positive. + val = 1 + } n, err := t.Futex().WakeOp(t, addr, naddr, private, val, nreq, op) return uintptr(n), nil, err diff --git a/test/syscalls/linux/futex.cc b/test/syscalls/linux/futex.cc index d3e3f998c..40c80a6e1 100644 --- a/test/syscalls/linux/futex.cc +++ b/test/syscalls/linux/futex.cc @@ -239,6 +239,27 @@ TEST_P(PrivateAndSharedFutexTest, Wake1_NoRandomSave) { EXPECT_THAT(futex_wake(IsPrivate(), &a, 1), SyscallSucceedsWithValue(1)); } +TEST_P(PrivateAndSharedFutexTest, Wake0_NoRandomSave) { + constexpr int kInitialValue = 1; + std::atomic a = ATOMIC_VAR_INIT(kInitialValue); + + // Prevent save/restore from interrupting futex_wait, which will cause it to + // return EAGAIN instead of the expected result if futex_wait is restarted + // after we change the value of a below. + DisableSave ds; + ScopedThread thread([&] { + EXPECT_THAT(futex_wait(IsPrivate(), &a, kInitialValue), + SyscallSucceedsWithValue(0)); + }); + absl::SleepFor(kWaiterStartupDelay); + + // Change a so that if futex_wake happens before futex_wait, the latter + // returns EAGAIN instead of hanging the test. + a.fetch_add(1); + // The Linux kernel wakes one waiter even if val is 0 or negative. + EXPECT_THAT(futex_wake(IsPrivate(), &a, 0), SyscallSucceedsWithValue(1)); +} + TEST_P(PrivateAndSharedFutexTest, WakeAll_NoRandomSave) { constexpr int kInitialValue = 1; std::atomic a = ATOMIC_VAR_INIT(kInitialValue); -- cgit v1.2.3 From 822d847ccaa1e6016b818bee289b5e33335f9fee Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 20 Dec 2019 08:43:15 -0800 Subject: Check for valid nfds before copying in an fd set. Otherwise, CopyInFDSet will try to allocate a negative-length slice. PiperOrigin-RevId: 286584907 --- pkg/sentry/strace/select.go | 3 +++ 1 file changed, 3 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/strace/select.go b/pkg/sentry/strace/select.go index dea309fda..c77d418e6 100644 --- a/pkg/sentry/strace/select.go +++ b/pkg/sentry/strace/select.go @@ -36,6 +36,9 @@ func fdsFromSet(t *kernel.Task, set []byte) []int { } func fdSet(t *kernel.Task, nfds int, addr usermem.Addr) string { + if nfds < 0 { + return fmt.Sprintf("%#x (negative nfds)", addr) + } if addr == 0 { return "null" } -- cgit v1.2.3 From 3eb489ed6c67b069bc135ab92cb031ce80b40d8f Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Fri, 20 Dec 2019 11:52:24 -0800 Subject: Move VFS2 file description status flags to vfs.FileDescription. PiperOrigin-RevId: 286616668 --- pkg/sentry/fsimpl/ext/file_description.go | 19 --- pkg/sentry/fsimpl/ext/inode.go | 9 +- pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go | 16 +-- pkg/sentry/fsimpl/kernfs/fd_impl_util.go | 16 +-- pkg/sentry/fsimpl/memfs/filesystem.go | 11 +- pkg/sentry/fsimpl/memfs/memfs.go | 14 --- pkg/sentry/fsimpl/memfs/named_pipe.go | 2 +- pkg/sentry/vfs/file_description.go | 141 +++++++++++++++------- pkg/sentry/vfs/file_description_impl_util_test.go | 2 +- 9 files changed, 107 insertions(+), 123 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/ext/file_description.go b/pkg/sentry/fsimpl/ext/file_description.go index 5eca2b83f..841274daf 100644 --- a/pkg/sentry/fsimpl/ext/file_description.go +++ b/pkg/sentry/fsimpl/ext/file_description.go @@ -26,13 +26,6 @@ import ( type fileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl - - // flags is the same as vfs.OpenOptions.Flags which are passed to - // vfs.FilesystemImpl.OpenAt. - // TODO(b/134676337): syscalls like read(2), write(2), fchmod(2), fchown(2), - // fgetxattr(2), ioctl(2), mmap(2) should fail with EBADF if O_PATH is set. - // Only close(2), fstat(2), fstatfs(2) should work. - flags uint32 } func (fd *fileDescription) filesystem() *filesystem { @@ -43,18 +36,6 @@ func (fd *fileDescription) inode() *inode { return fd.vfsfd.Dentry().Impl().(*dentry).inode } -// StatusFlags implements vfs.FileDescriptionImpl.StatusFlags. -func (fd *fileDescription) StatusFlags(ctx context.Context) (uint32, error) { - return fd.flags, nil -} - -// SetStatusFlags implements vfs.FileDescriptionImpl.SetStatusFlags. -func (fd *fileDescription) SetStatusFlags(ctx context.Context, flags uint32) error { - // None of the flags settable by fcntl(F_SETFL) are supported, so this is a - // no-op. - return nil -} - // Stat implements vfs.FileDescriptionImpl.Stat. func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { var stat linux.Statx diff --git a/pkg/sentry/fsimpl/ext/inode.go b/pkg/sentry/fsimpl/ext/inode.go index 24249525c..b2cc826c7 100644 --- a/pkg/sentry/fsimpl/ext/inode.go +++ b/pkg/sentry/fsimpl/ext/inode.go @@ -157,10 +157,9 @@ func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*v switch in.impl.(type) { case *regularFile: var fd regularFileFD - fd.flags = flags mnt.IncRef() vfsd.IncRef() - fd.vfsfd.Init(&fd, mnt, vfsd) + fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{}) return &fd.vfsfd, nil case *directory: // Can't open directories writably. This check is not necessary for a read @@ -169,10 +168,9 @@ func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*v return nil, syserror.EISDIR } var fd directoryFD - fd.flags = flags mnt.IncRef() vfsd.IncRef() - fd.vfsfd.Init(&fd, mnt, vfsd) + fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{}) return &fd.vfsfd, nil case *symlink: if flags&linux.O_PATH == 0 { @@ -180,10 +178,9 @@ func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*v return nil, syserror.ELOOP } var fd symlinkFD - fd.flags = flags mnt.IncRef() vfsd.IncRef() - fd.vfsfd.Init(&fd, mnt, vfsd) + fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{}) return &fd.vfsfd, nil default: panic(fmt.Sprintf("unknown inode type: %T", in.impl)) diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go index 30c06baf0..51102ce48 100644 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -65,17 +65,15 @@ type DynamicBytesFD struct { vfsfd vfs.FileDescription inode Inode - flags uint32 } // Init initializes a DynamicBytesFD. func (fd *DynamicBytesFD) Init(m *vfs.Mount, d *vfs.Dentry, data vfs.DynamicBytesSource, flags uint32) { m.IncRef() // DecRef in vfs.FileDescription.vd.DecRef on final ref. d.IncRef() // DecRef in vfs.FileDescription.vd.DecRef on final ref. - fd.flags = flags fd.inode = d.Impl().(*Dentry).inode fd.SetDataSource(data) - fd.vfsfd.Init(fd, m, d) + fd.vfsfd.Init(fd, flags, m, d, &vfs.FileDescriptionOptions{}) } // Seek implements vfs.FileDescriptionImpl.Seek. @@ -117,15 +115,3 @@ func (fd *DynamicBytesFD) SetStat(context.Context, vfs.SetStatOptions) error { // DynamicBytesFiles are immutable. return syserror.EPERM } - -// StatusFlags implements vfs.FileDescriptionImpl.StatusFlags. -func (fd *DynamicBytesFD) StatusFlags(ctx context.Context) (uint32, error) { - return fd.flags, nil -} - -// SetStatusFlags implements vfs.FileDescriptionImpl.SetStatusFlags. -func (fd *DynamicBytesFD) SetStatusFlags(ctx context.Context, flags uint32) error { - // None of the flags settable by fcntl(F_SETFL) are supported, so this is a - // no-op. - return nil -} diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index d6c18937a..bd402330f 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -39,7 +39,6 @@ type GenericDirectoryFD struct { vfsfd vfs.FileDescription children *OrderedChildren - flags uint32 off int64 } @@ -48,8 +47,7 @@ func (fd *GenericDirectoryFD) Init(m *vfs.Mount, d *vfs.Dentry, children *Ordere m.IncRef() // DecRef in vfs.FileDescription.vd.DecRef on final ref. d.IncRef() // DecRef in vfs.FileDescription.vd.DecRef on final ref. fd.children = children - fd.flags = flags - fd.vfsfd.Init(fd, m, d) + fd.vfsfd.Init(fd, flags, m, d, &vfs.FileDescriptionOptions{}) } // VFSFileDescription returns a pointer to the vfs.FileDescription representing @@ -180,18 +178,6 @@ func (fd *GenericDirectoryFD) Seek(ctx context.Context, offset int64, whence int return offset, nil } -// StatusFlags implements vfs.FileDescriptionImpl.StatusFlags. -func (fd *GenericDirectoryFD) StatusFlags(ctx context.Context) (uint32, error) { - return fd.flags, nil -} - -// SetStatusFlags implements vfs.FileDescriptionImpl.SetStatusFlags. -func (fd *GenericDirectoryFD) SetStatusFlags(ctx context.Context, flags uint32) error { - // None of the flags settable by fcntl(F_SETFL) are supported, so this is a - // no-op. - return nil -} - // Stat implements vfs.FileDescriptionImpl.Stat. func (fd *GenericDirectoryFD) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { fs := fd.filesystem() diff --git a/pkg/sentry/fsimpl/memfs/filesystem.go b/pkg/sentry/fsimpl/memfs/filesystem.go index 22f1e811f..af4389459 100644 --- a/pkg/sentry/fsimpl/memfs/filesystem.go +++ b/pkg/sentry/fsimpl/memfs/filesystem.go @@ -282,9 +282,8 @@ func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.OpenOptions) (*vfs.FileDescription, error) { // Filter out flags that are not supported by memfs. O_DIRECTORY and // O_NOFOLLOW have no effect here (they're handled by VFS by setting - // appropriate bits in rp), but are returned by - // FileDescriptionImpl.StatusFlags(). O_NONBLOCK is supported only by - // pipes. + // appropriate bits in rp), but are visible in FD status flags. O_NONBLOCK + // is supported only by pipes. opts.Flags &= linux.O_ACCMODE | linux.O_CREAT | linux.O_EXCL | linux.O_TRUNC | linux.O_DIRECTORY | linux.O_NOFOLLOW | linux.O_NONBLOCK if opts.Flags&linux.O_CREAT == 0 { @@ -384,7 +383,6 @@ func (i *inode) open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentr switch impl := i.impl.(type) { case *regularFile: var fd regularFileFD - fd.flags = flags fd.readable = vfs.MayReadFileWithOpenFlags(flags) fd.writable = vfs.MayWriteFileWithOpenFlags(flags) if fd.writable { @@ -395,7 +393,7 @@ func (i *inode) open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentr } mnt.IncRef() vfsd.IncRef() - fd.vfsfd.Init(&fd, mnt, vfsd) + fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{}) if flags&linux.O_TRUNC != 0 { impl.mu.Lock() impl.data = impl.data[:0] @@ -411,8 +409,7 @@ func (i *inode) open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentr var fd directoryFD mnt.IncRef() vfsd.IncRef() - fd.vfsfd.Init(&fd, mnt, vfsd) - fd.flags = flags + fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{}) return &fd.vfsfd, nil case *symlink: // Can't open symlinks without O_PATH (which is unimplemented). diff --git a/pkg/sentry/fsimpl/memfs/memfs.go b/pkg/sentry/fsimpl/memfs/memfs.go index 4cb2a4e0f..9d509f6e4 100644 --- a/pkg/sentry/fsimpl/memfs/memfs.go +++ b/pkg/sentry/fsimpl/memfs/memfs.go @@ -261,8 +261,6 @@ func (i *inode) direntType() uint8 { type fileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl - - flags uint32 // status flags; immutable } func (fd *fileDescription) filesystem() *filesystem { @@ -273,18 +271,6 @@ func (fd *fileDescription) inode() *inode { return fd.vfsfd.Dentry().Impl().(*dentry).inode } -// StatusFlags implements vfs.FileDescriptionImpl.StatusFlags. -func (fd *fileDescription) StatusFlags(ctx context.Context) (uint32, error) { - return fd.flags, nil -} - -// SetStatusFlags implements vfs.FileDescriptionImpl.SetStatusFlags. -func (fd *fileDescription) SetStatusFlags(ctx context.Context, flags uint32) error { - // None of the flags settable by fcntl(F_SETFL) are supported, so this is a - // no-op. - return nil -} - // Stat implements vfs.FileDescriptionImpl.Stat. func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { var stat linux.Statx diff --git a/pkg/sentry/fsimpl/memfs/named_pipe.go b/pkg/sentry/fsimpl/memfs/named_pipe.go index 91cb4b1fc..d5060850e 100644 --- a/pkg/sentry/fsimpl/memfs/named_pipe.go +++ b/pkg/sentry/fsimpl/memfs/named_pipe.go @@ -57,6 +57,6 @@ func newNamedPipeFD(ctx context.Context, np *namedPipe, rp *vfs.ResolvingPath, v mnt := rp.Mount() mnt.IncRef() vfsd.IncRef() - fd.vfsfd.Init(&fd, mnt, vfsd) + fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{}) return &fd.vfsfd, nil } diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index c5a9adca3..df03886c3 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -20,6 +20,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" @@ -39,49 +40,43 @@ type FileDescription struct { // operations. refs int64 + // statusFlags contains status flags, "initialized by open(2) and possibly + // modified by fcntl()" - fcntl(2). statusFlags is accessed using atomic + // memory operations. + statusFlags uint32 + // vd is the filesystem location at which this FileDescription was opened. // A reference is held on vd. vd is immutable. vd VirtualDentry + opts FileDescriptionOptions + // impl is the FileDescriptionImpl associated with this Filesystem. impl is // immutable. This should be the last field in FileDescription. impl FileDescriptionImpl } +// FileDescriptionOptions contains options to FileDescription.Init(). +type FileDescriptionOptions struct { + // If AllowDirectIO is true, allow O_DIRECT to be set on the file. This is + // usually only the case if O_DIRECT would actually have an effect. + AllowDirectIO bool +} + // Init must be called before first use of fd. It takes ownership of references -// on mnt and d held by the caller. -func (fd *FileDescription) Init(impl FileDescriptionImpl, mnt *Mount, d *Dentry) { +// on mnt and d held by the caller. statusFlags is the initial file description +// status flags, which is usually the full set of flags passed to open(2). +func (fd *FileDescription) Init(impl FileDescriptionImpl, statusFlags uint32, mnt *Mount, d *Dentry, opts *FileDescriptionOptions) { fd.refs = 1 + fd.statusFlags = statusFlags | linux.O_LARGEFILE fd.vd = VirtualDentry{ mount: mnt, dentry: d, } + fd.opts = *opts fd.impl = impl } -// Impl returns the FileDescriptionImpl associated with fd. -func (fd *FileDescription) Impl() FileDescriptionImpl { - return fd.impl -} - -// Mount returns the mount on which fd was opened. It does not take a reference -// on the returned Mount. -func (fd *FileDescription) Mount() *Mount { - return fd.vd.mount -} - -// Dentry returns the dentry at which fd was opened. It does not take a -// reference on the returned Dentry. -func (fd *FileDescription) Dentry() *Dentry { - return fd.vd.dentry -} - -// VirtualDentry returns the location at which fd was opened. It does not take -// a reference on the returned VirtualDentry. -func (fd *FileDescription) VirtualDentry() VirtualDentry { - return fd.vd -} - // IncRef increments fd's reference count. func (fd *FileDescription) IncRef() { atomic.AddInt64(&fd.refs, 1) @@ -113,6 +108,82 @@ func (fd *FileDescription) DecRef() { } } +// Mount returns the mount on which fd was opened. It does not take a reference +// on the returned Mount. +func (fd *FileDescription) Mount() *Mount { + return fd.vd.mount +} + +// Dentry returns the dentry at which fd was opened. It does not take a +// reference on the returned Dentry. +func (fd *FileDescription) Dentry() *Dentry { + return fd.vd.dentry +} + +// VirtualDentry returns the location at which fd was opened. It does not take +// a reference on the returned VirtualDentry. +func (fd *FileDescription) VirtualDentry() VirtualDentry { + return fd.vd +} + +// StatusFlags returns file description status flags, as for fcntl(F_GETFL). +func (fd *FileDescription) StatusFlags() uint32 { + return atomic.LoadUint32(&fd.statusFlags) +} + +// SetStatusFlags sets file description status flags, as for fcntl(F_SETFL). +func (fd *FileDescription) SetStatusFlags(ctx context.Context, creds *auth.Credentials, flags uint32) error { + // Compare Linux's fs/fcntl.c:setfl(). + oldFlags := fd.StatusFlags() + // Linux documents this check as "O_APPEND cannot be cleared if the file is + // marked as append-only and the file is open for write", which would make + // sense. However, the check as actually implemented seems to be "O_APPEND + // cannot be changed if the file is marked as append-only". + if (flags^oldFlags)&linux.O_APPEND != 0 { + stat, err := fd.impl.Stat(ctx, StatOptions{ + // There is no mask bit for stx_attributes. + Mask: 0, + // Linux just reads inode::i_flags directly. + Sync: linux.AT_STATX_DONT_SYNC, + }) + if err != nil { + return err + } + if (stat.AttributesMask&linux.STATX_ATTR_APPEND != 0) && (stat.Attributes&linux.STATX_ATTR_APPEND != 0) { + return syserror.EPERM + } + } + if (flags&linux.O_NOATIME != 0) && (oldFlags&linux.O_NOATIME == 0) { + stat, err := fd.impl.Stat(ctx, StatOptions{ + Mask: linux.STATX_UID, + // Linux's inode_owner_or_capable() just reads inode::i_uid + // directly. + Sync: linux.AT_STATX_DONT_SYNC, + }) + if err != nil { + return err + } + if stat.Mask&linux.STATX_UID == 0 { + return syserror.EPERM + } + if !CanActAsOwner(creds, auth.KUID(stat.UID)) { + return syserror.EPERM + } + } + if flags&linux.O_DIRECT != 0 && !fd.opts.AllowDirectIO { + return syserror.EINVAL + } + // TODO(jamieliu): FileDescriptionImpl.SetOAsync()? + const settableFlags = linux.O_APPEND | linux.O_ASYNC | linux.O_DIRECT | linux.O_NOATIME | linux.O_NONBLOCK + atomic.StoreUint32(&fd.statusFlags, (oldFlags&^settableFlags)|(flags&settableFlags)) + return nil +} + +// Impl returns the FileDescriptionImpl associated with fd. +func (fd *FileDescription) Impl() FileDescriptionImpl { + return fd.impl +} + // FileDescriptionImpl contains implementation details for an FileDescription. // Implementations of FileDescriptionImpl should contain their associated // FileDescription by value as their first field. @@ -132,14 +203,6 @@ type FileDescriptionImpl interface { // prevent the file descriptor from being closed. OnClose(ctx context.Context) error - // StatusFlags returns file description status flags, as for - // fcntl(F_GETFL). - StatusFlags(ctx context.Context) (uint32, error) - - // SetStatusFlags sets file description status flags, as for - // fcntl(F_SETFL). - SetStatusFlags(ctx context.Context, flags uint32) error - // Stat returns metadata for the file represented by the FileDescription. Stat(ctx context.Context, opts StatOptions) (linux.Statx, error) @@ -264,18 +327,6 @@ func (fd *FileDescription) OnClose(ctx context.Context) error { return fd.impl.OnClose(ctx) } -// StatusFlags returns file description status flags, as for fcntl(F_GETFL). -func (fd *FileDescription) StatusFlags(ctx context.Context) (uint32, error) { - flags, err := fd.impl.StatusFlags(ctx) - flags |= linux.O_LARGEFILE - return flags, err -} - -// SetStatusFlags sets file description status flags, as for fcntl(F_SETFL). -func (fd *FileDescription) SetStatusFlags(ctx context.Context, flags uint32) error { - return fd.impl.SetStatusFlags(ctx, flags) -} - // Stat returns metadata for the file represented by fd. func (fd *FileDescription) Stat(ctx context.Context, opts StatOptions) (linux.Statx, error) { return fd.impl.Stat(ctx, opts) diff --git a/pkg/sentry/vfs/file_description_impl_util_test.go b/pkg/sentry/vfs/file_description_impl_util_test.go index ac7799296..678be07fe 100644 --- a/pkg/sentry/vfs/file_description_impl_util_test.go +++ b/pkg/sentry/vfs/file_description_impl_util_test.go @@ -48,7 +48,7 @@ type genCountFD struct { func newGenCountFD(mnt *Mount, vfsd *Dentry) *FileDescription { var fd genCountFD - fd.vfsfd.Init(&fd, mnt, vfsd) + fd.vfsfd.Init(&fd, 0 /* statusFlags */, mnt, vfsd, &FileDescriptionOptions{}) fd.DynamicBytesFileDescriptionImpl.SetDataSource(&fd) return &fd.vfsfd } -- cgit v1.2.3 From 21a14e9532365fc5eb51a5796ab66cf7f007ede3 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Fri, 20 Dec 2019 16:43:34 -0800 Subject: Add vfs.Dentry.Children(). PiperOrigin-RevId: 286660774 --- pkg/sentry/vfs/dentry.go | 18 +++++++++++++++--- 1 file changed, 15 insertions(+), 3 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/vfs/dentry.go b/pkg/sentry/vfs/dentry.go index 40f4c1d09..6209eb053 100644 --- a/pkg/sentry/vfs/dentry.go +++ b/pkg/sentry/vfs/dentry.go @@ -85,12 +85,12 @@ type Dentry struct { // mounts is accessed using atomic memory operations. mounts uint32 - // mu synchronizes disowning and mounting over this Dentry. - mu sync.Mutex - // children are child Dentries. children map[string]*Dentry + // mu synchronizes disowning and mounting over this Dentry. + mu sync.Mutex + // impl is the DentryImpl associated with this Dentry. impl is immutable. // This should be the last field in Dentry. impl DentryImpl @@ -199,6 +199,18 @@ func (d *Dentry) HasChildren() bool { return len(d.children) != 0 } +// Children returns a map containing all of d's children. +func (d *Dentry) Children() map[string]*Dentry { + if !d.HasChildren() { + return nil + } + m := make(map[string]*Dentry) + for name, child := range d.children { + m[name] = child + } + return m +} + // InsertChild makes child a child of d with the given name. // // InsertChild is a mutator of d and child. -- cgit v1.2.3 From 818eb22b11d6e0c056af6d4605e8cd246e622231 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Fri, 20 Dec 2019 17:40:18 -0800 Subject: Add vfs.ResolvingPath.HandleJump(). PiperOrigin-RevId: 286666533 --- pkg/sentry/vfs/resolving_path.go | 40 ++++++++++++++++++++++++++++++---------- 1 file changed, 30 insertions(+), 10 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/vfs/resolving_path.go b/pkg/sentry/vfs/resolving_path.go index 621f5a6f8..d580fd39e 100644 --- a/pkg/sentry/vfs/resolving_path.go +++ b/pkg/sentry/vfs/resolving_path.go @@ -85,11 +85,11 @@ func init() { // so error "constants" are really mutable vars, necessitating somewhat // expensive interface object comparisons. -type resolveMountRootError struct{} +type resolveMountRootOrJumpError struct{} // Error implements error.Error. -func (resolveMountRootError) Error() string { - return "resolving mount root" +func (resolveMountRootOrJumpError) Error() string { + return "resolving mount root or jump" } type resolveMountPointError struct{} @@ -274,7 +274,7 @@ func (rp *ResolvingPath) ResolveParent(d *Dentry) (*Dentry, error) { // ... of non-root mount. rp.nextMount = vd.mount rp.nextStart = vd.dentry - return nil, resolveMountRootError{} + return nil, resolveMountRootOrJumpError{} } // ... of root mount. parent = d @@ -385,11 +385,32 @@ func (rp *ResolvingPath) relpathPrepend(path fspath.Path) { } } +// HandleJump is called when the current path component is a "magic" link to +// the given VirtualDentry, like /proc/[pid]/fd/[fd]. If the calling Filesystem +// method should continue path traversal, HandleMagicSymlink updates the path +// component stream to reflect the magic link target and returns nil. Otherwise +// it returns a non-nil error. +// +// Preconditions: !rp.Done(). +func (rp *ResolvingPath) HandleJump(target VirtualDentry) error { + if rp.symlinks >= linux.MaxSymlinkTraversals { + return syserror.ELOOP + } + rp.symlinks++ + // Consume the path component that represented the magic link. + rp.Advance() + // Unconditionally return a resolveMountRootOrJumpError, even if the Mount + // isn't changing, to force restarting at the new Dentry. + target.IncRef() + rp.nextMount = target.mount + rp.nextStart = target.dentry + return resolveMountRootOrJumpError{} +} + func (rp *ResolvingPath) handleError(err error) bool { switch err.(type) { - case resolveMountRootError: - // Switch to the new Mount. We hold references on the Mount and Dentry - // (from VFS.getMountpointAt()). + case resolveMountRootOrJumpError: + // Switch to the new Mount. We hold references on the Mount and Dentry. rp.decRefStartAndMount() rp.mount = rp.nextMount rp.start = rp.nextStart @@ -407,9 +428,8 @@ func (rp *ResolvingPath) handleError(err error) bool { return true case resolveMountPointError: - // Switch to the new Mount. We hold a reference on the Mount (from - // VFS.getMountAt()), but borrow the reference on the mount root from - // the Mount. + // Switch to the new Mount. We hold a reference on the Mount, but + // borrow the reference on the mount root from the Mount. rp.decRefStartAndMount() rp.mount = rp.nextMount rp.start = rp.nextMount.root -- cgit v1.2.3 From b6e31aadaa74b235c46bfa5e08b5ac66a3b4f8c5 Mon Sep 17 00:00:00 2001 From: Haibo Xu Date: Mon, 23 Dec 2019 03:01:07 +0000 Subject: Replace syscall.PTRACE_SYSEMU with unix.PTRACE_SYSEMU Linux PTRACE_SYSEMU support on arm64 was merged to mainline from V5.3, and the corresponding support in go also enabled recently. Since the "syscall" package is locked down from go 1.4, so the ptrace PTRACE_SYSEMU definition can't be added to package "syscall" on arm64. According to the golang community, updates required by new systems or versions should use the corresponding package in the golang.org/x/sys repository instead(https://golang.org/pkg/syscall/). Signed-off-by: Haibo Xu Change-Id: I2f917bb2be62f990c3e158e2bb99e094ea03f751 --- pkg/sentry/platform/ptrace/subprocess.go | 5 +++-- pkg/sentry/platform/ptrace/subprocess_linux.go | 2 +- 2 files changed, 4 insertions(+), 3 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/ptrace/subprocess.go b/pkg/sentry/platform/ptrace/subprocess.go index ddb1f41e3..821f6848d 100644 --- a/pkg/sentry/platform/ptrace/subprocess.go +++ b/pkg/sentry/platform/ptrace/subprocess.go @@ -21,6 +21,7 @@ import ( "sync" "syscall" + "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/procid" "gvisor.dev/gvisor/pkg/sentry/arch" @@ -541,14 +542,14 @@ func (s *subprocess) switchToApp(c *context, ac arch.Context) bool { if isSingleStepping(regs) { if _, _, errno := syscall.RawSyscall6( syscall.SYS_PTRACE, - syscall.PTRACE_SYSEMU_SINGLESTEP, + unix.PTRACE_SYSEMU_SINGLESTEP, uintptr(t.tid), 0, 0, 0, 0); errno != 0 { panic(fmt.Sprintf("ptrace sysemu failed: %v", errno)) } } else { if _, _, errno := syscall.RawSyscall6( syscall.SYS_PTRACE, - syscall.PTRACE_SYSEMU, + unix.PTRACE_SYSEMU, uintptr(t.tid), 0, 0, 0, 0); errno != 0 { panic(fmt.Sprintf("ptrace sysemu failed: %v", errno)) } diff --git a/pkg/sentry/platform/ptrace/subprocess_linux.go b/pkg/sentry/platform/ptrace/subprocess_linux.go index cf13ea5e4..74968dfdf 100644 --- a/pkg/sentry/platform/ptrace/subprocess_linux.go +++ b/pkg/sentry/platform/ptrace/subprocess_linux.go @@ -54,7 +54,7 @@ func probeSeccomp() bool { for { // Attempt an emulation. - if _, _, errno := syscall.RawSyscall6(syscall.SYS_PTRACE, syscall.PTRACE_SYSEMU, uintptr(t.tid), 0, 0, 0, 0); errno != 0 { + if _, _, errno := syscall.RawSyscall6(syscall.SYS_PTRACE, unix.PTRACE_SYSEMU, uintptr(t.tid), 0, 0, 0, 0); errno != 0 { panic(fmt.Sprintf("ptrace syscall-enter failed: %v", errno)) } -- cgit v1.2.3 From f45df7505b0e7baf48a37f7c625f05051d144738 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Mon, 23 Dec 2019 13:17:29 -0800 Subject: Clean up vfs.FilesystemImpl methods that operate on parent directories. - Make FilesystemImpl methods that operate on parent directories require !rp.Done() (i.e. there is at least one path component to resolve) as precondition and postcondition (in cases where they do not finish path resolution due to mount boundary / absolute symlink), and require that they do not need to follow the last path component (the file being created / deleted) as a symlink. Check for these in VFS. - Add FilesystemImpl.GetParentDentryAt(), which is required to obtain the old parent directory for VFS.RenameAt(). (Passing the Dentry to be renamed instead has the wrong semantics if the file named by the old path is a mount point since the Dentry will be on the wrong Mount.) - Update memfs to implement these methods correctly (?), including RenameAt. - Change fspath.Parse() to allow empty paths (to simplify implementation of AT_EMPTY_PATH). - Change vfs.PathOperation to take a fspath.Path instead of a raw pathname; non-test callers will need to fspath.Parse() pathnames themselves anyway in order to detect absolute paths and select PathOperation.Start accordingly. PiperOrigin-RevId: 286934941 --- pkg/fspath/BUILD | 2 - pkg/fspath/fspath.go | 24 +- pkg/fspath/fspath_test.go | 25 +- pkg/sentry/fsimpl/ext/BUILD | 1 + pkg/sentry/fsimpl/ext/benchmark/BUILD | 1 + pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go | 11 +- pkg/sentry/fsimpl/ext/ext_test.go | 9 +- pkg/sentry/fsimpl/ext/filesystem.go | 12 +- pkg/sentry/fsimpl/kernfs/BUILD | 1 + pkg/sentry/fsimpl/kernfs/filesystem.go | 138 +++-- pkg/sentry/fsimpl/kernfs/kernfs_test.go | 7 +- pkg/sentry/fsimpl/memfs/BUILD | 2 + pkg/sentry/fsimpl/memfs/benchmark_test.go | 27 +- pkg/sentry/fsimpl/memfs/filesystem.go | 667 ++++++++++++---------- pkg/sentry/fsimpl/memfs/memfs.go | 29 +- pkg/sentry/fsimpl/memfs/pipe_test.go | 18 +- pkg/sentry/vfs/dentry.go | 29 +- pkg/sentry/vfs/file_description.go | 19 + pkg/sentry/vfs/filesystem.go | 251 +++++++- pkg/sentry/vfs/options.go | 3 + pkg/sentry/vfs/resolving_path.go | 46 +- pkg/sentry/vfs/testutil.go | 7 +- pkg/sentry/vfs/vfs.go | 259 ++++++--- pkg/syserror/syserror.go | 1 + 24 files changed, 1051 insertions(+), 538 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/fspath/BUILD b/pkg/fspath/BUILD index 0c5f50397..ca540363c 100644 --- a/pkg/fspath/BUILD +++ b/pkg/fspath/BUILD @@ -14,7 +14,6 @@ go_library( "fspath.go", ], importpath = "gvisor.dev/gvisor/pkg/fspath", - deps = ["//pkg/syserror"], ) go_test( @@ -25,5 +24,4 @@ go_test( "fspath_test.go", ], embed = [":fspath"], - deps = ["//pkg/syserror"], ) diff --git a/pkg/fspath/fspath.go b/pkg/fspath/fspath.go index f68752560..9fb3fee24 100644 --- a/pkg/fspath/fspath.go +++ b/pkg/fspath/fspath.go @@ -18,19 +18,17 @@ package fspath import ( "strings" - - "gvisor.dev/gvisor/pkg/syserror" ) const pathSep = '/' -// Parse parses a pathname as described by path_resolution(7). -func Parse(pathname string) (Path, error) { +// Parse parses a pathname as described by path_resolution(7), except that +// empty pathnames will be parsed successfully to a Path for which +// Path.Absolute == Path.Dir == Path.HasComponents() == false. (This is +// necessary to support AT_EMPTY_PATH.) +func Parse(pathname string) Path { if len(pathname) == 0 { - // "... POSIX decrees that an empty pathname must not be resolved - // successfully. Linux returns ENOENT in this case." - - // path_resolution(7) - return Path{}, syserror.ENOENT + return Path{} } // Skip leading path separators. i := 0 @@ -41,7 +39,7 @@ func Parse(pathname string) (Path, error) { return Path{ Absolute: true, Dir: true, - }, nil + } } } // Skip trailing path separators. This is required by Iterator.Next. This @@ -64,7 +62,7 @@ func Parse(pathname string) (Path, error) { }, Absolute: i != 0, Dir: j != len(pathname)-1, - }, nil + } } // Path contains the information contained in a pathname string. @@ -111,6 +109,12 @@ func (p Path) String() string { return b.String() } +// HasComponents returns true if p contains a non-zero number of path +// components. +func (p Path) HasComponents() bool { + return p.Begin.Ok() +} + // An Iterator represents either a path component in a Path or a terminal // iterator indicating that the end of the path has been reached. // diff --git a/pkg/fspath/fspath_test.go b/pkg/fspath/fspath_test.go index 215b35622..d5e9a549a 100644 --- a/pkg/fspath/fspath_test.go +++ b/pkg/fspath/fspath_test.go @@ -18,15 +18,10 @@ import ( "reflect" "strings" "testing" - - "gvisor.dev/gvisor/pkg/syserror" ) func TestParseIteratorPartialPathnames(t *testing.T) { - path, err := Parse("/foo//bar///baz////") - if err != nil { - t.Fatalf("Parse failed: %v", err) - } + path := Parse("/foo//bar///baz////") // Parse strips leading slashes, and records their presence as // Path.Absolute. if !path.Absolute { @@ -70,6 +65,12 @@ func TestParse(t *testing.T) { dir bool } tests := []testCase{ + { + pathname: "", + relpath: []string{}, + abs: false, + dir: false, + }, { pathname: "/", relpath: []string{}, @@ -113,10 +114,7 @@ func TestParse(t *testing.T) { for _, test := range tests { t.Run(test.pathname, func(t *testing.T) { - p, err := Parse(test.pathname) - if err != nil { - t.Fatalf("failed to parse pathname %q: %v", test.pathname, err) - } + p := Parse(test.pathname) t.Logf("pathname %q => path %q", test.pathname, p) if p.Absolute != test.abs { t.Errorf("path absoluteness: got %v, wanted %v", p.Absolute, test.abs) @@ -134,10 +132,3 @@ func TestParse(t *testing.T) { }) } } - -func TestParseEmptyPathname(t *testing.T) { - p, err := Parse("") - if err != syserror.ENOENT { - t.Errorf("parsing empty pathname: got (%v, %v), wanted (, ENOENT)", p, err) - } -} diff --git a/pkg/sentry/fsimpl/ext/BUILD b/pkg/sentry/fsimpl/ext/BUILD index 880b7bcd3..bc90330bc 100644 --- a/pkg/sentry/fsimpl/ext/BUILD +++ b/pkg/sentry/fsimpl/ext/BUILD @@ -74,6 +74,7 @@ go_test( deps = [ "//pkg/abi/linux", "//pkg/binary", + "//pkg/fspath", "//pkg/sentry/context", "//pkg/sentry/context/contexttest", "//pkg/sentry/fsimpl/ext/disklayout", diff --git a/pkg/sentry/fsimpl/ext/benchmark/BUILD b/pkg/sentry/fsimpl/ext/benchmark/BUILD index bfc46dfa6..4fc8296ef 100644 --- a/pkg/sentry/fsimpl/ext/benchmark/BUILD +++ b/pkg/sentry/fsimpl/ext/benchmark/BUILD @@ -7,6 +7,7 @@ go_test( size = "small", srcs = ["benchmark_test.go"], deps = [ + "//pkg/fspath", "//pkg/sentry/context", "//pkg/sentry/context/contexttest", "//pkg/sentry/fsimpl/ext", diff --git a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go index 177ce2cb9..2f46d2d13 100644 --- a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go +++ b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go @@ -24,6 +24,7 @@ import ( "strings" "testing" + "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/context/contexttest" "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext" @@ -121,7 +122,7 @@ func BenchmarkVFS2Ext4fsStat(b *testing.B) { stat, err := vfsfs.StatAt(ctx, creds, &vfs.PathOperation{ Root: *root, Start: *root, - Pathname: filePath, + Path: fspath.Parse(filePath), FollowFinalSymlink: true, }, &vfs.StatOptions{}) if err != nil { @@ -150,9 +151,9 @@ func BenchmarkVFS2ExtfsMountStat(b *testing.B) { creds := auth.CredentialsFromContext(ctx) mountPointName := "/1/" pop := vfs.PathOperation{ - Root: *root, - Start: *root, - Pathname: mountPointName, + Root: *root, + Start: *root, + Path: fspath.Parse(mountPointName), } // Save the mount point for later use. @@ -181,7 +182,7 @@ func BenchmarkVFS2ExtfsMountStat(b *testing.B) { stat, err := vfsfs.StatAt(ctx, creds, &vfs.PathOperation{ Root: *root, Start: *root, - Pathname: filePath, + Path: fspath.Parse(filePath), FollowFinalSymlink: true, }, &vfs.StatOptions{}) if err != nil { diff --git a/pkg/sentry/fsimpl/ext/ext_test.go b/pkg/sentry/fsimpl/ext/ext_test.go index e9f756732..5d6c999bd 100644 --- a/pkg/sentry/fsimpl/ext/ext_test.go +++ b/pkg/sentry/fsimpl/ext/ext_test.go @@ -25,6 +25,7 @@ import ( "github.com/google/go-cmp/cmp" "github.com/google/go-cmp/cmp/cmpopts" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/context/contexttest" "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" @@ -140,7 +141,7 @@ func TestSeek(t *testing.T) { fd, err := vfsfs.OpenAt( ctx, auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: *root, Start: *root, Pathname: test.path}, + &vfs.PathOperation{Root: *root, Start: *root, Path: fspath.Parse(test.path)}, &vfs.OpenOptions{}, ) if err != nil { @@ -359,7 +360,7 @@ func TestStatAt(t *testing.T) { got, err := vfsfs.StatAt(ctx, auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: *root, Start: *root, Pathname: test.path}, + &vfs.PathOperation{Root: *root, Start: *root, Path: fspath.Parse(test.path)}, &vfs.StatOptions{}, ) if err != nil { @@ -429,7 +430,7 @@ func TestRead(t *testing.T) { fd, err := vfsfs.OpenAt( ctx, auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: *root, Start: *root, Pathname: test.absPath}, + &vfs.PathOperation{Root: *root, Start: *root, Path: fspath.Parse(test.absPath)}, &vfs.OpenOptions{}, ) if err != nil { @@ -565,7 +566,7 @@ func TestIterDirents(t *testing.T) { fd, err := vfsfs.OpenAt( ctx, auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: *root, Start: *root, Pathname: test.path}, + &vfs.PathOperation{Root: *root, Start: *root, Path: fspath.Parse(test.path)}, &vfs.OpenOptions{}, ) if err != nil { diff --git a/pkg/sentry/fsimpl/ext/filesystem.go b/pkg/sentry/fsimpl/ext/filesystem.go index d7e87979a..616fc002a 100644 --- a/pkg/sentry/fsimpl/ext/filesystem.go +++ b/pkg/sentry/fsimpl/ext/filesystem.go @@ -275,6 +275,16 @@ func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, op return vfsd, nil } +// GetParentDentryAt implements vfs.FilesystemImpl.GetParentDentryAt. +func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { + vfsd, inode, err := fs.walk(rp, true) + if err != nil { + return nil, err + } + inode.incRef() + return vfsd, nil +} + // OpenAt implements vfs.FilesystemImpl.OpenAt. func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.OpenOptions) (*vfs.FileDescription, error) { vfsd, inode, err := fs.walk(rp, false) @@ -378,7 +388,7 @@ func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v } // RenameAt implements vfs.FilesystemImpl.RenameAt. -func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry, opts vfs.RenameOptions) error { +func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldParentVD vfs.VirtualDentry, oldName string, opts vfs.RenameOptions) error { if rp.Done() { return syserror.ENOENT } diff --git a/pkg/sentry/fsimpl/kernfs/BUILD b/pkg/sentry/fsimpl/kernfs/BUILD index 52596c090..59f7f39e2 100644 --- a/pkg/sentry/fsimpl/kernfs/BUILD +++ b/pkg/sentry/fsimpl/kernfs/BUILD @@ -49,6 +49,7 @@ go_test( deps = [ ":kernfs", "//pkg/abi/linux", + "//pkg/fspath", "//pkg/sentry/context", "//pkg/sentry/context/contexttest", "//pkg/sentry/kernel/auth", diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 3cbbe4b20..a6f9fced5 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -44,39 +44,37 @@ func (fs *Filesystem) stepExistingLocked(ctx context.Context, rp *vfs.ResolvingP return nil, err } afterSymlink: + name := rp.Component() + // Revalidation must be skipped if name is "." or ".."; d or its parent + // respectively can't be expected to transition from invalidated back to + // valid, so detecting invalidation and retrying would loop forever. This + // is consistent with Linux: fs/namei.c:walk_component() => lookup_fast() + // calls d_revalidate(), but walk_component() => handle_dots() does not. + if name == "." { + rp.Advance() + return vfsd, nil + } + if name == ".." { + nextVFSD, err := rp.ResolveParent(vfsd) + if err != nil { + return nil, err + } + rp.Advance() + return nextVFSD, nil + } d.dirMu.Lock() - nextVFSD, err := rp.ResolveComponent(vfsd) - d.dirMu.Unlock() + nextVFSD, err := rp.ResolveChild(vfsd, name) if err != nil { + d.dirMu.Unlock() return nil, err } - if nextVFSD != nil { - // Cached dentry exists, revalidate. - next := nextVFSD.Impl().(*Dentry) - if !next.inode.Valid(ctx) { - d.dirMu.Lock() - rp.VirtualFilesystem().ForceDeleteDentry(nextVFSD) - d.dirMu.Unlock() - fs.deferDecRef(nextVFSD) // Reference from Lookup. - nextVFSD = nil - } - } - if nextVFSD == nil { - // Dentry isn't cached; it either doesn't exist or failed - // revalidation. Attempt to resolve it via Lookup. - name := rp.Component() - var err error - nextVFSD, err = d.inode.Lookup(ctx, name) - // Reference on nextVFSD dropped by a corresponding Valid. - if err != nil { - return nil, err - } - d.InsertChild(name, nextVFSD) + next, err := fs.revalidateChildLocked(ctx, rp.VirtualFilesystem(), d, name, nextVFSD) + d.dirMu.Unlock() + if err != nil { + return nil, err } - next := nextVFSD.Impl().(*Dentry) - // Resolve any symlink at current path component. - if rp.ShouldFollowSymlink() && d.isSymlink() { + if rp.ShouldFollowSymlink() && next.isSymlink() { // TODO: VFS2 needs something extra for /proc/[pid]/fd/ "magic symlinks". target, err := next.inode.Readlink(ctx) if err != nil { @@ -89,7 +87,44 @@ afterSymlink: } rp.Advance() - return nextVFSD, nil + return &next.vfsd, nil +} + +// revalidateChildLocked must be called after a call to parent.vfsd.Child(name) +// or vfs.ResolvingPath.ResolveChild(name) returns childVFSD (which may be +// nil) to verify that the returned child (or lack thereof) is correct. +// +// Preconditions: Filesystem.mu must be locked for at least reading. +// parent.dirMu must be locked. parent.isDir(). name is not "." or "..". +// +// Postconditions: Caller must call fs.processDeferredDecRefs*. +func (fs *Filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.VirtualFilesystem, parent *Dentry, name string, childVFSD *vfs.Dentry) (*Dentry, error) { + if childVFSD != nil { + // Cached dentry exists, revalidate. + child := childVFSD.Impl().(*Dentry) + if !child.inode.Valid(ctx) { + vfsObj.ForceDeleteDentry(childVFSD) + fs.deferDecRef(childVFSD) // Reference from Lookup. + childVFSD = nil + } + } + if childVFSD == nil { + // Dentry isn't cached; it either doesn't exist or failed + // revalidation. Attempt to resolve it via Lookup. + // + // FIXME(b/144498111): Inode.Lookup() should return *(kernfs.)Dentry, + // not *vfs.Dentry, since (kernfs.)Filesystem assumes that all dentries + // in the filesystem are (kernfs.)Dentry and performs vfs.DentryImpl + // casts accordingly. + var err error + childVFSD, err = parent.inode.Lookup(ctx, name) + if err != nil { + return nil, err + } + // Reference on childVFSD dropped by a corresponding Valid. + parent.InsertChild(name, childVFSD) + } + return childVFSD.Impl().(*Dentry), nil } // walkExistingLocked resolves rp to an existing file. @@ -242,6 +277,19 @@ func (fs *Filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, op return vfsd, nil } +// GetParentDentryAt implements vfs.FilesystemImpl.GetParentDentryAt. +func (fs *Filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { + fs.mu.RLock() + defer fs.processDeferredDecRefs() + defer fs.mu.RUnlock() + vfsd, _, err := fs.walkParentDirLocked(ctx, rp) + if err != nil { + return nil, err + } + vfsd.IncRef() // Ownership transferred to caller. + return vfsd, nil +} + // LinkAt implements vfs.FilesystemImpl.LinkAt. func (fs *Filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry) error { if rp.Done() { @@ -459,40 +507,42 @@ func (fs *Filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (st } // RenameAt implements vfs.FilesystemImpl.RenameAt. -func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry, opts vfs.RenameOptions) error { - noReplace := opts.Flags&linux.RENAME_NOREPLACE != 0 - exchange := opts.Flags&linux.RENAME_EXCHANGE != 0 - whiteout := opts.Flags&linux.RENAME_WHITEOUT != 0 - if exchange && (noReplace || whiteout) { - // Can't specify RENAME_NOREPLACE or RENAME_WHITEOUT with RENAME_EXCHANGE. - return syserror.EINVAL - } - if exchange || whiteout { - // Exchange and Whiteout flags are not supported on kernfs. +func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldParentVD vfs.VirtualDentry, oldName string, opts vfs.RenameOptions) error { + // Only RENAME_NOREPLACE is supported. + if opts.Flags&^linux.RENAME_NOREPLACE != 0 { return syserror.EINVAL } + noReplace := opts.Flags&linux.RENAME_NOREPLACE != 0 fs.mu.Lock() defer fs.mu.Lock() + // Resolve the destination directory first to verify that it's on this + // Mount. + dstDirVFSD, dstDirInode, err := fs.walkParentDirLocked(ctx, rp) + fs.processDeferredDecRefsLocked() + if err != nil { + return err + } mnt := rp.Mount() - if mnt != vd.Mount() { + if mnt != oldParentVD.Mount() { return syserror.EXDEV } - if err := mnt.CheckBeginWrite(); err != nil { return err } defer mnt.EndWrite() - dstDirVFSD, dstDirInode, err := fs.walkParentDirLocked(ctx, rp) + srcDirVFSD := oldParentVD.Dentry() + srcDir := srcDirVFSD.Impl().(*Dentry) + srcDir.dirMu.Lock() + src, err := fs.revalidateChildLocked(ctx, rp.VirtualFilesystem(), srcDir, oldName, srcDirVFSD.Child(oldName)) + srcDir.dirMu.Unlock() fs.processDeferredDecRefsLocked() if err != nil { return err } - - srcVFSD := vd.Dentry() - srcDirVFSD := srcVFSD.Parent() + srcVFSD := &src.vfsd // Can we remove the src dentry? if err := checkDeleteLocked(rp, srcVFSD); err != nil { diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index f78bb7b04..73b6e43b5 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -24,6 +24,7 @@ import ( "github.com/google/go-cmp/cmp" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/context/contexttest" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" @@ -82,9 +83,9 @@ func newTestSystem(t *testing.T, rootFn RootDentryFn) *TestSystem { // Precondition: path should be relative path. func (s *TestSystem) PathOpAtRoot(path string) vfs.PathOperation { return vfs.PathOperation{ - Root: s.root, - Start: s.root, - Pathname: path, + Root: s.root, + Start: s.root, + Path: fspath.Parse(path), } } diff --git a/pkg/sentry/fsimpl/memfs/BUILD b/pkg/sentry/fsimpl/memfs/BUILD index 0cc751eb8..5689bed3b 100644 --- a/pkg/sentry/fsimpl/memfs/BUILD +++ b/pkg/sentry/fsimpl/memfs/BUILD @@ -50,6 +50,7 @@ go_test( deps = [ ":memfs", "//pkg/abi/linux", + "//pkg/fspath", "//pkg/refs", "//pkg/sentry/context", "//pkg/sentry/context/contexttest", @@ -68,6 +69,7 @@ go_test( embed = [":memfs"], deps = [ "//pkg/abi/linux", + "//pkg/fspath", "//pkg/sentry/context", "//pkg/sentry/context/contexttest", "//pkg/sentry/kernel/auth", diff --git a/pkg/sentry/fsimpl/memfs/benchmark_test.go b/pkg/sentry/fsimpl/memfs/benchmark_test.go index 4a7a94a52..6e987af88 100644 --- a/pkg/sentry/fsimpl/memfs/benchmark_test.go +++ b/pkg/sentry/fsimpl/memfs/benchmark_test.go @@ -21,6 +21,7 @@ import ( "testing" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/context/contexttest" @@ -193,9 +194,9 @@ func BenchmarkVFS2MemfsStat(b *testing.B) { for i := depth; i > 0; i-- { name := fmt.Sprintf("%d", i) pop := vfs.PathOperation{ - Root: root, - Start: vd, - Pathname: name, + Root: root, + Start: vd, + Path: fspath.Parse(name), } if err := vfsObj.MkdirAt(ctx, creds, &pop, &vfs.MkdirOptions{ Mode: 0755, @@ -216,7 +217,7 @@ func BenchmarkVFS2MemfsStat(b *testing.B) { fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ Root: root, Start: vd, - Pathname: filename, + Path: fspath.Parse(filename), FollowFinalSymlink: true, }, &vfs.OpenOptions{ Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL, @@ -237,7 +238,7 @@ func BenchmarkVFS2MemfsStat(b *testing.B) { stat, err := vfsObj.StatAt(ctx, creds, &vfs.PathOperation{ Root: root, Start: root, - Pathname: filePath, + Path: fspath.Parse(filePath), FollowFinalSymlink: true, }, &vfs.StatOptions{}) if err != nil { @@ -378,9 +379,9 @@ func BenchmarkVFS2MemfsMountStat(b *testing.B) { root := mntns.Root() defer root.DecRef() pop := vfs.PathOperation{ - Root: root, - Start: root, - Pathname: mountPointName, + Root: root, + Start: root, + Path: fspath.Parse(mountPointName), } if err := vfsObj.MkdirAt(ctx, creds, &pop, &vfs.MkdirOptions{ Mode: 0755, @@ -408,9 +409,9 @@ func BenchmarkVFS2MemfsMountStat(b *testing.B) { for i := depth; i > 0; i-- { name := fmt.Sprintf("%d", i) pop := vfs.PathOperation{ - Root: root, - Start: vd, - Pathname: name, + Root: root, + Start: vd, + Path: fspath.Parse(name), } if err := vfsObj.MkdirAt(ctx, creds, &pop, &vfs.MkdirOptions{ Mode: 0755, @@ -438,7 +439,7 @@ func BenchmarkVFS2MemfsMountStat(b *testing.B) { fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ Root: root, Start: vd, - Pathname: filename, + Path: fspath.Parse(filename), FollowFinalSymlink: true, }, &vfs.OpenOptions{ Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL, @@ -458,7 +459,7 @@ func BenchmarkVFS2MemfsMountStat(b *testing.B) { stat, err := vfsObj.StatAt(ctx, creds, &vfs.PathOperation{ Root: root, Start: root, - Pathname: filePath, + Path: fspath.Parse(filePath), FollowFinalSymlink: true, }, &vfs.StatOptions{}) if err != nil { diff --git a/pkg/sentry/fsimpl/memfs/filesystem.go b/pkg/sentry/fsimpl/memfs/filesystem.go index af4389459..4a83f310c 100644 --- a/pkg/sentry/fsimpl/memfs/filesystem.go +++ b/pkg/sentry/fsimpl/memfs/filesystem.go @@ -25,323 +25,283 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) -// stepLocked resolves rp.Component() in parent directory vfsd. +// Sync implements vfs.FilesystemImpl.Sync. +func (fs *filesystem) Sync(ctx context.Context) error { + // All filesystem state is in-memory. + return nil +} + +// stepLocked resolves rp.Component() to an existing file, starting from the +// given directory. // // stepLocked is loosely analogous to fs/namei.c:walk_component(). // -// Preconditions: filesystem.mu must be locked. !rp.Done(). inode == -// vfsd.Impl().(*dentry).inode. -func stepLocked(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, inode *inode) (*vfs.Dentry, *inode, error) { - if !inode.isDir() { - return nil, nil, syserror.ENOTDIR +// Preconditions: filesystem.mu must be locked. !rp.Done(). +func stepLocked(rp *vfs.ResolvingPath, d *dentry) (*dentry, error) { + if !d.inode.isDir() { + return nil, syserror.ENOTDIR } - if err := inode.checkPermissions(rp.Credentials(), vfs.MayExec, true); err != nil { - return nil, nil, err + if err := d.inode.checkPermissions(rp.Credentials(), vfs.MayExec, true); err != nil { + return nil, err } afterSymlink: - nextVFSD, err := rp.ResolveComponent(vfsd) + nextVFSD, err := rp.ResolveComponent(&d.vfsd) if err != nil { - return nil, nil, err + return nil, err } if nextVFSD == nil { // Since the Dentry tree is the sole source of truth for memfs, if it's // not in the Dentry tree, it doesn't exist. - return nil, nil, syserror.ENOENT + return nil, syserror.ENOENT } - nextInode := nextVFSD.Impl().(*dentry).inode - if symlink, ok := nextInode.impl.(*symlink); ok && rp.ShouldFollowSymlink() { + next := nextVFSD.Impl().(*dentry) + if symlink, ok := next.inode.impl.(*symlink); ok && rp.ShouldFollowSymlink() { // TODO: symlink traversals update access time if err := rp.HandleSymlink(symlink.target); err != nil { - return nil, nil, err + return nil, err } goto afterSymlink // don't check the current directory again } rp.Advance() - return nextVFSD, nextInode, nil + return next, nil } -// walkExistingLocked resolves rp to an existing file. +// walkParentDirLocked resolves all but the last path component of rp to an +// existing directory, starting from the given directory (which is usually +// rp.Start().Impl().(*dentry)). It does not check that the returned directory +// is searchable by the provider of rp. // -// walkExistingLocked is loosely analogous to Linux's -// fs/namei.c:path_lookupat(). +// walkParentDirLocked is loosely analogous to Linux's +// fs/namei.c:path_parentat(). // -// Preconditions: filesystem.mu must be locked. -func walkExistingLocked(rp *vfs.ResolvingPath) (*vfs.Dentry, *inode, error) { - vfsd := rp.Start() - inode := vfsd.Impl().(*dentry).inode - for !rp.Done() { - var err error - vfsd, inode, err = stepLocked(rp, vfsd, inode) +// Preconditions: filesystem.mu must be locked. !rp.Done(). +func walkParentDirLocked(rp *vfs.ResolvingPath, d *dentry) (*dentry, error) { + for !rp.Final() { + next, err := stepLocked(rp, d) if err != nil { - return nil, nil, err + return nil, err } + d = next } - if rp.MustBeDir() && !inode.isDir() { - return nil, nil, syserror.ENOTDIR + if !d.inode.isDir() { + return nil, syserror.ENOTDIR } - return vfsd, inode, nil + return d, nil } -// walkParentDirLocked resolves all but the last path component of rp to an -// existing directory. It does not check that the returned directory is -// searchable by the provider of rp. +// resolveLocked resolves rp to an existing file. // -// walkParentDirLocked is loosely analogous to Linux's -// fs/namei.c:path_parentat(). +// resolveLocked is loosely analogous to Linux's fs/namei.c:path_lookupat(). // -// Preconditions: filesystem.mu must be locked. !rp.Done(). -func walkParentDirLocked(rp *vfs.ResolvingPath) (*vfs.Dentry, *inode, error) { - vfsd := rp.Start() - inode := vfsd.Impl().(*dentry).inode - for !rp.Final() { - var err error - vfsd, inode, err = stepLocked(rp, vfsd, inode) +// Preconditions: filesystem.mu must be locked. +func resolveLocked(rp *vfs.ResolvingPath) (*dentry, error) { + d := rp.Start().Impl().(*dentry) + for !rp.Done() { + next, err := stepLocked(rp, d) if err != nil { - return nil, nil, err + return nil, err } + d = next } - if !inode.isDir() { - return nil, nil, syserror.ENOTDIR + if rp.MustBeDir() && !d.inode.isDir() { + return nil, syserror.ENOTDIR } - return vfsd, inode, nil + return d, nil } -// checkCreateLocked checks that a file named rp.Component() may be created in -// directory parentVFSD, then returns rp.Component(). +// doCreateAt checks that creating a file at rp is permitted, then invokes +// create to do so. // -// Preconditions: filesystem.mu must be locked. parentInode == -// parentVFSD.Impl().(*dentry).inode. parentInode.isDir() == true. -func checkCreateLocked(rp *vfs.ResolvingPath, parentVFSD *vfs.Dentry, parentInode *inode) (string, error) { - if err := parentInode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true); err != nil { - return "", err - } - pc := rp.Component() - if pc == "." || pc == ".." { - return "", syserror.EEXIST - } - childVFSD, err := rp.ResolveChild(parentVFSD, pc) +// doCreateAt is loosely analogous to a conjunction of Linux's +// fs/namei.c:filename_create() and done_path_create(). +// +// Preconditions: !rp.Done(). For the final path component in rp, +// !rp.ShouldFollowSymlink(). +func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(parent *dentry, name string) error) error { + fs.mu.Lock() + defer fs.mu.Unlock() + parent, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) if err != nil { - return "", err + return err } - if childVFSD != nil { - return "", syserror.EEXIST + if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true /* isDir */); err != nil { + return err } - if parentVFSD.IsDisowned() { - return "", syserror.ENOENT + name := rp.Component() + if name == "." || name == ".." { + return syserror.EEXIST } - return pc, nil -} - -// checkDeleteLocked checks that the file represented by vfsd may be deleted. -func checkDeleteLocked(vfsd *vfs.Dentry) error { - parentVFSD := vfsd.Parent() - if parentVFSD == nil { - return syserror.EBUSY + // Call parent.vfsd.Child() instead of stepLocked() or rp.ResolveChild(), + // because if the child exists we want to return EEXIST immediately instead + // of attempting symlink/mount traversal. + if parent.vfsd.Child(name) != nil { + return syserror.EEXIST } - if parentVFSD.IsDisowned() { + if !dir && rp.MustBeDir() { return syserror.ENOENT } - return nil + // In memfs, the only way to cause a dentry to be disowned is by removing + // it from the filesystem, so this check is equivalent to checking if + // parent has been removed. + if parent.vfsd.IsDisowned() { + return syserror.ENOENT + } + mnt := rp.Mount() + if err := mnt.CheckBeginWrite(); err != nil { + return err + } + defer mnt.EndWrite() + return create(parent, name) } // GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { fs.mu.RLock() defer fs.mu.RUnlock() - vfsd, inode, err := walkExistingLocked(rp) + d, err := resolveLocked(rp) if err != nil { return nil, err } if opts.CheckSearchable { - if !inode.isDir() { + if !d.inode.isDir() { return nil, syserror.ENOTDIR } - if err := inode.checkPermissions(rp.Credentials(), vfs.MayExec, true); err != nil { + if err := d.inode.checkPermissions(rp.Credentials(), vfs.MayExec, true /* isDir */); err != nil { return nil, err } } - inode.incRef() - return vfsd, nil + d.IncRef() + return &d.vfsd, nil } -// LinkAt implements vfs.FilesystemImpl.LinkAt. -func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry) error { - if rp.Done() { - return syserror.EEXIST - } - fs.mu.Lock() - defer fs.mu.Unlock() - parentVFSD, parentInode, err := walkParentDirLocked(rp) - if err != nil { - return err - } - pc, err := checkCreateLocked(rp, parentVFSD, parentInode) +// GetParentDentryAt implements vfs.FilesystemImpl.GetParentDentryAt. +func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { + fs.mu.RLock() + defer fs.mu.RUnlock() + d, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) if err != nil { - return err - } - if rp.Mount() != vd.Mount() { - return syserror.EXDEV - } - if err := rp.Mount().CheckBeginWrite(); err != nil { - return err - } - defer rp.Mount().EndWrite() - d := vd.Dentry().Impl().(*dentry) - if d.inode.isDir() { - return syserror.EPERM + return nil, err } - d.inode.incLinksLocked() - child := fs.newDentry(d.inode) - parentVFSD.InsertChild(&child.vfsd, pc) - parentInode.impl.(*directory).childList.PushBack(child) - return nil + d.IncRef() + return &d.vfsd, nil +} + +// LinkAt implements vfs.FilesystemImpl.LinkAt. +func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry) error { + return fs.doCreateAt(rp, false /* dir */, func(parent *dentry, name string) error { + if rp.Mount() != vd.Mount() { + return syserror.EXDEV + } + d := vd.Dentry().Impl().(*dentry) + if d.inode.isDir() { + return syserror.EPERM + } + if d.inode.nlink == 0 { + return syserror.ENOENT + } + if d.inode.nlink == maxLinks { + return syserror.EMLINK + } + d.inode.incLinksLocked() + child := fs.newDentry(d.inode) + parent.vfsd.InsertChild(&child.vfsd, name) + parent.inode.impl.(*directory).childList.PushBack(child) + return nil + }) } // MkdirAt implements vfs.FilesystemImpl.MkdirAt. func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error { - if rp.Done() { - return syserror.EEXIST - } - fs.mu.Lock() - defer fs.mu.Unlock() - parentVFSD, parentInode, err := walkParentDirLocked(rp) - if err != nil { - return err - } - pc, err := checkCreateLocked(rp, parentVFSD, parentInode) - if err != nil { - return err - } - if err := rp.Mount().CheckBeginWrite(); err != nil { - return err - } - defer rp.Mount().EndWrite() - child := fs.newDentry(fs.newDirectory(rp.Credentials(), opts.Mode)) - parentVFSD.InsertChild(&child.vfsd, pc) - parentInode.impl.(*directory).childList.PushBack(child) - parentInode.incLinksLocked() // from child's ".." - return nil + return fs.doCreateAt(rp, true /* dir */, func(parent *dentry, name string) error { + if parent.inode.nlink == maxLinks { + return syserror.EMLINK + } + parent.inode.incLinksLocked() // from child's ".." + child := fs.newDentry(fs.newDirectory(rp.Credentials(), opts.Mode)) + parent.vfsd.InsertChild(&child.vfsd, name) + parent.inode.impl.(*directory).childList.PushBack(child) + return nil + }) } // MknodAt implements vfs.FilesystemImpl.MknodAt. func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MknodOptions) error { - if rp.Done() { - return syserror.EEXIST - } - fs.mu.Lock() - defer fs.mu.Unlock() - parentVFSD, parentInode, err := walkParentDirLocked(rp) - if err != nil { - return err - } - pc, err := checkCreateLocked(rp, parentVFSD, parentInode) - if err != nil { - return err - } - if err := rp.Mount().CheckBeginWrite(); err != nil { - return err - } - defer rp.Mount().EndWrite() - - switch opts.Mode.FileType() { - case 0: - // "Zero file type is equivalent to type S_IFREG." - mknod(2) - fallthrough - case linux.ModeRegular: - // TODO(b/138862511): Implement. - return syserror.EINVAL - - case linux.ModeNamedPipe: - child := fs.newDentry(fs.newNamedPipe(rp.Credentials(), opts.Mode)) - parentVFSD.InsertChild(&child.vfsd, pc) - parentInode.impl.(*directory).childList.PushBack(child) - return nil - - case linux.ModeSocket: - // TODO(b/138862511): Implement. - return syserror.EINVAL - - case linux.ModeCharacterDevice: - fallthrough - case linux.ModeBlockDevice: - // TODO(b/72101894): We don't support creating block or character - // devices at the moment. - // - // When we start supporting block and character devices, we'll - // need to check for CAP_MKNOD here. - return syserror.EPERM - - default: - // "EINVAL - mode requested creation of something other than a - // regular file, device special file, FIFO or socket." - mknod(2) - return syserror.EINVAL - } + return fs.doCreateAt(rp, false /* dir */, func(parent *dentry, name string) error { + switch opts.Mode.FileType() { + case 0, linux.S_IFREG: + child := fs.newDentry(fs.newRegularFile(rp.Credentials(), opts.Mode)) + parent.vfsd.InsertChild(&child.vfsd, name) + parent.inode.impl.(*directory).childList.PushBack(child) + return nil + case linux.S_IFIFO: + child := fs.newDentry(fs.newNamedPipe(rp.Credentials(), opts.Mode)) + parent.vfsd.InsertChild(&child.vfsd, name) + parent.inode.impl.(*directory).childList.PushBack(child) + return nil + case linux.S_IFBLK, linux.S_IFCHR, linux.S_IFSOCK: + // Not yet supported. + return syserror.EPERM + default: + return syserror.EINVAL + } + }) } // OpenAt implements vfs.FilesystemImpl.OpenAt. func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - // Filter out flags that are not supported by memfs. O_DIRECTORY and - // O_NOFOLLOW have no effect here (they're handled by VFS by setting - // appropriate bits in rp), but are visible in FD status flags. O_NONBLOCK - // is supported only by pipes. - opts.Flags &= linux.O_ACCMODE | linux.O_CREAT | linux.O_EXCL | linux.O_TRUNC | linux.O_DIRECTORY | linux.O_NOFOLLOW | linux.O_NONBLOCK + if opts.Flags&linux.O_TMPFILE != 0 { + // Not yet supported. + return nil, syserror.EOPNOTSUPP + } + // Handle O_CREAT and !O_CREAT separately, since in the latter case we + // don't need fs.mu for writing. if opts.Flags&linux.O_CREAT == 0 { fs.mu.RLock() defer fs.mu.RUnlock() - vfsd, inode, err := walkExistingLocked(rp) + d, err := resolveLocked(rp) if err != nil { return nil, err } - return inode.open(ctx, rp, vfsd, opts.Flags, false) + return d.open(ctx, rp, opts.Flags, false /* afterCreate */) } mustCreate := opts.Flags&linux.O_EXCL != 0 - vfsd := rp.Start() - inode := vfsd.Impl().(*dentry).inode + start := rp.Start().Impl().(*dentry) fs.mu.Lock() defer fs.mu.Unlock() if rp.Done() { + // Reject attempts to open directories with O_CREAT. if rp.MustBeDir() { return nil, syserror.EISDIR } if mustCreate { return nil, syserror.EEXIST } - return inode.open(ctx, rp, vfsd, opts.Flags, false) + return start.open(ctx, rp, opts.Flags, false /* afterCreate */) } afterTrailingSymlink: - // Walk to the parent directory of the last path component. - for !rp.Final() { - var err error - vfsd, inode, err = stepLocked(rp, vfsd, inode) - if err != nil { - return nil, err - } - } - if !inode.isDir() { - return nil, syserror.ENOTDIR + parent, err := walkParentDirLocked(rp, start) + if err != nil { + return nil, err } // Check for search permission in the parent directory. - if err := inode.checkPermissions(rp.Credentials(), vfs.MayExec, true); err != nil { + if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayExec, true); err != nil { return nil, err } // Reject attempts to open directories with O_CREAT. if rp.MustBeDir() { return nil, syserror.EISDIR } - pc := rp.Component() - if pc == "." || pc == ".." { + name := rp.Component() + if name == "." || name == ".." { return nil, syserror.EISDIR } // Determine whether or not we need to create a file. - childVFSD, err := rp.ResolveChild(vfsd, pc) - if err != nil { - return nil, err - } - if childVFSD == nil { + child, err := stepLocked(rp, parent) + if err == syserror.ENOENT { // Already checked for searchability above; now check for writability. - if err := inode.checkPermissions(rp.Credentials(), vfs.MayWrite, true); err != nil { + if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite, true); err != nil { return nil, err } if err := rp.Mount().CheckBeginWrite(); err != nil { @@ -349,38 +309,35 @@ afterTrailingSymlink: } defer rp.Mount().EndWrite() // Create and open the child. - childInode := fs.newRegularFile(rp.Credentials(), opts.Mode) - child := fs.newDentry(childInode) - vfsd.InsertChild(&child.vfsd, pc) - inode.impl.(*directory).childList.PushBack(child) - return childInode.open(ctx, rp, &child.vfsd, opts.Flags, true) + child := fs.newDentry(fs.newRegularFile(rp.Credentials(), opts.Mode)) + parent.vfsd.InsertChild(&child.vfsd, name) + parent.inode.impl.(*directory).childList.PushBack(child) + return child.open(ctx, rp, opts.Flags, true) } - // Open existing file or follow symlink. - if mustCreate { - return nil, syserror.EEXIST + if err != nil { + return nil, err } - childInode := childVFSD.Impl().(*dentry).inode - if symlink, ok := childInode.impl.(*symlink); ok && rp.ShouldFollowSymlink() { - // TODO: symlink traversals update access time - if err := rp.HandleSymlink(symlink.target); err != nil { - return nil, err - } - // rp.Final() may no longer be true since we now need to resolve the - // symlink target. + // Do we need to resolve a trailing symlink? + if !rp.Done() { + start = parent goto afterTrailingSymlink } - return childInode.open(ctx, rp, childVFSD, opts.Flags, false) + // Open existing file. + if mustCreate { + return nil, syserror.EEXIST + } + return child.open(ctx, rp, opts.Flags, false) } -func (i *inode) open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32, afterCreate bool) (*vfs.FileDescription, error) { +func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, flags uint32, afterCreate bool) (*vfs.FileDescription, error) { ats := vfs.AccessTypesForOpenFlags(flags) if !afterCreate { - if err := i.checkPermissions(rp.Credentials(), ats, i.isDir()); err != nil { + if err := d.inode.checkPermissions(rp.Credentials(), ats, d.inode.isDir()); err != nil { return nil, err } } mnt := rp.Mount() - switch impl := i.impl.(type) { + switch impl := d.inode.impl.(type) { case *regularFile: var fd regularFileFD fd.readable = vfs.MayReadFileWithOpenFlags(flags) @@ -392,8 +349,8 @@ func (i *inode) open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentr // mnt.EndWrite() is called by regularFileFD.Release(). } mnt.IncRef() - vfsd.IncRef() - fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{}) + d.IncRef() + fd.vfsfd.Init(&fd, flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}) if flags&linux.O_TRUNC != 0 { impl.mu.Lock() impl.data = impl.data[:0] @@ -408,28 +365,28 @@ func (i *inode) open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentr } var fd directoryFD mnt.IncRef() - vfsd.IncRef() - fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{}) + d.IncRef() + fd.vfsfd.Init(&fd, flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}) return &fd.vfsfd, nil case *symlink: // Can't open symlinks without O_PATH (which is unimplemented). return nil, syserror.ELOOP case *namedPipe: - return newNamedPipeFD(ctx, impl, rp, vfsd, flags) + return newNamedPipeFD(ctx, impl, rp, &d.vfsd, flags) default: - panic(fmt.Sprintf("unknown inode type: %T", i.impl)) + panic(fmt.Sprintf("unknown inode type: %T", d.inode.impl)) } } // ReadlinkAt implements vfs.FilesystemImpl.ReadlinkAt. func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { fs.mu.RLock() - _, inode, err := walkExistingLocked(rp) - fs.mu.RUnlock() + defer fs.mu.RUnlock() + d, err := resolveLocked(rp) if err != nil { return "", err } - symlink, ok := inode.impl.(*symlink) + symlink, ok := d.inode.impl.(*symlink) if !ok { return "", syserror.EINVAL } @@ -437,63 +394,172 @@ func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (st } // RenameAt implements vfs.FilesystemImpl.RenameAt. -func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry, opts vfs.RenameOptions) error { - if rp.Done() { - return syserror.ENOENT +func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldParentVD vfs.VirtualDentry, oldName string, opts vfs.RenameOptions) error { + if opts.Flags != 0 { + // TODO(b/145974740): Support renameat2 flags. + return syserror.EINVAL } + + // Resolve newParent first to verify that it's on this Mount. fs.mu.Lock() defer fs.mu.Unlock() - parentVFSD, parentInode, err := walkParentDirLocked(rp) + newParent, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) if err != nil { return err } - _, err = checkCreateLocked(rp, parentVFSD, parentInode) - if err != nil { + newName := rp.Component() + if newName == "." || newName == ".." { + return syserror.EBUSY + } + mnt := rp.Mount() + if mnt != oldParentVD.Mount() { + return syserror.EXDEV + } + if err := mnt.CheckBeginWrite(); err != nil { return err } - if err := rp.Mount().CheckBeginWrite(); err != nil { + defer mnt.EndWrite() + + oldParent := oldParentVD.Dentry().Impl().(*dentry) + if err := oldParent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true /* isDir */); err != nil { return err } - defer rp.Mount().EndWrite() - // TODO: actually implement RenameAt - return syserror.EPERM + // Call vfs.Dentry.Child() instead of stepLocked() or rp.ResolveChild(), + // because if the existing child is a symlink or mount point then we want + // to rename over it rather than follow it. + renamedVFSD := oldParent.vfsd.Child(oldName) + if renamedVFSD == nil { + return syserror.ENOENT + } + renamed := renamedVFSD.Impl().(*dentry) + if renamed.inode.isDir() { + if renamed == newParent || renamedVFSD.IsAncestorOf(&newParent.vfsd) { + return syserror.EINVAL + } + if oldParent != newParent { + // Writability is needed to change renamed's "..". + if err := renamed.inode.checkPermissions(rp.Credentials(), vfs.MayWrite, true /* isDir */); err != nil { + return err + } + } + } else { + if opts.MustBeDir || rp.MustBeDir() { + return syserror.ENOTDIR + } + } + + if err := newParent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true /* isDir */); err != nil { + return err + } + replacedVFSD := newParent.vfsd.Child(newName) + var replaced *dentry + if replacedVFSD != nil { + replaced = replacedVFSD.Impl().(*dentry) + if replaced.inode.isDir() { + if !renamed.inode.isDir() { + return syserror.EISDIR + } + if replaced.vfsd.HasChildren() { + return syserror.ENOTEMPTY + } + } else { + if rp.MustBeDir() { + return syserror.ENOTDIR + } + if renamed.inode.isDir() { + return syserror.ENOTDIR + } + } + } else { + if renamed.inode.isDir() && newParent.inode.nlink == maxLinks { + return syserror.EMLINK + } + } + if newParent.vfsd.IsDisowned() { + return syserror.ENOENT + } + + // Linux places this check before some of those above; we do it here for + // simplicity, under the assumption that applications are not intentionally + // doing noop renames expecting them to succeed where non-noop renames + // would fail. + if renamedVFSD == replacedVFSD { + return nil + } + vfsObj := rp.VirtualFilesystem() + oldParentDir := oldParent.inode.impl.(*directory) + newParentDir := newParent.inode.impl.(*directory) + if err := vfsObj.PrepareRenameDentry(vfs.MountNamespaceFromContext(ctx), renamedVFSD, replacedVFSD); err != nil { + return err + } + if replaced != nil { + newParentDir.childList.Remove(replaced) + if replaced.inode.isDir() { + newParent.inode.decLinksLocked() // from replaced's ".." + } + replaced.inode.decLinksLocked() + } + oldParentDir.childList.Remove(renamed) + newParentDir.childList.PushBack(renamed) + if renamed.inode.isDir() { + oldParent.inode.decLinksLocked() + newParent.inode.incLinksLocked() + } + // TODO: update timestamps and parent directory sizes + vfsObj.CommitRenameReplaceDentry(renamedVFSD, &newParent.vfsd, newName, replacedVFSD) + return nil } // RmdirAt implements vfs.FilesystemImpl.RmdirAt. func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error { fs.mu.Lock() defer fs.mu.Unlock() - vfsd, inode, err := walkExistingLocked(rp) + parent, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) if err != nil { return err } - if err := rp.Mount().CheckBeginWrite(); err != nil { + if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true /* isDir */); err != nil { return err } - defer rp.Mount().EndWrite() - if err := checkDeleteLocked(vfsd); err != nil { - return err + name := rp.Component() + if name == "." { + return syserror.EINVAL } - if !inode.isDir() { + if name == ".." { + return syserror.ENOTEMPTY + } + childVFSD := parent.vfsd.Child(name) + if childVFSD == nil { + return syserror.ENOENT + } + child := childVFSD.Impl().(*dentry) + if !child.inode.isDir() { return syserror.ENOTDIR } - if vfsd.HasChildren() { + if childVFSD.HasChildren() { return syserror.ENOTEMPTY } - if err := rp.VirtualFilesystem().DeleteDentry(vfs.MountNamespaceFromContext(ctx), vfsd); err != nil { + mnt := rp.Mount() + if err := mnt.CheckBeginWrite(); err != nil { + return err + } + defer mnt.EndWrite() + vfsObj := rp.VirtualFilesystem() + if err := vfsObj.PrepareDeleteDentry(vfs.MountNamespaceFromContext(ctx), childVFSD); err != nil { return err } - // Remove from parent directory's childList. - vfsd.Parent().Impl().(*dentry).inode.impl.(*directory).childList.Remove(vfsd.Impl().(*dentry)) - inode.decRef() + parent.inode.impl.(*directory).childList.Remove(child) + parent.inode.decLinksLocked() // from child's ".." + child.inode.decLinksLocked() + vfsObj.CommitDeleteDentry(childVFSD) return nil } // SetStatAt implements vfs.FilesystemImpl.SetStatAt. func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { fs.mu.RLock() - _, _, err := walkExistingLocked(rp) - fs.mu.RUnlock() + defer fs.mu.RUnlock() + _, err := resolveLocked(rp) if err != nil { return err } @@ -507,21 +573,21 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts // StatAt implements vfs.FilesystemImpl.StatAt. func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { fs.mu.RLock() - _, inode, err := walkExistingLocked(rp) - fs.mu.RUnlock() + defer fs.mu.RUnlock() + d, err := resolveLocked(rp) if err != nil { return linux.Statx{}, err } var stat linux.Statx - inode.statTo(&stat) + d.inode.statTo(&stat) return stat, nil } // StatFSAt implements vfs.FilesystemImpl.StatFSAt. func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { fs.mu.RLock() - _, _, err := walkExistingLocked(rp) - fs.mu.RUnlock() + defer fs.mu.RUnlock() + _, err := resolveLocked(rp) if err != nil { return linux.Statfs{}, err } @@ -531,53 +597,52 @@ func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linu // SymlinkAt implements vfs.FilesystemImpl.SymlinkAt. func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, target string) error { - if rp.Done() { - return syserror.EEXIST - } - fs.mu.Lock() - defer fs.mu.Unlock() - parentVFSD, parentInode, err := walkParentDirLocked(rp) - if err != nil { - return err - } - pc, err := checkCreateLocked(rp, parentVFSD, parentInode) - if err != nil { - return err - } - if err := rp.Mount().CheckBeginWrite(); err != nil { - return err - } - defer rp.Mount().EndWrite() - child := fs.newDentry(fs.newSymlink(rp.Credentials(), target)) - parentVFSD.InsertChild(&child.vfsd, pc) - parentInode.impl.(*directory).childList.PushBack(child) - return nil + return fs.doCreateAt(rp, false /* dir */, func(parent *dentry, name string) error { + child := fs.newDentry(fs.newSymlink(rp.Credentials(), target)) + parent.vfsd.InsertChild(&child.vfsd, name) + parent.inode.impl.(*directory).childList.PushBack(child) + return nil + }) } // UnlinkAt implements vfs.FilesystemImpl.UnlinkAt. func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error { fs.mu.Lock() defer fs.mu.Unlock() - vfsd, inode, err := walkExistingLocked(rp) + parent, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) if err != nil { return err } - if err := rp.Mount().CheckBeginWrite(); err != nil { + if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true /* isDir */); err != nil { return err } - defer rp.Mount().EndWrite() - if err := checkDeleteLocked(vfsd); err != nil { - return err + name := rp.Component() + if name == "." || name == ".." { + return syserror.EISDIR } - if inode.isDir() { + childVFSD := parent.vfsd.Child(name) + if childVFSD == nil { + return syserror.ENOENT + } + child := childVFSD.Impl().(*dentry) + if child.inode.isDir() { return syserror.EISDIR } - if err := rp.VirtualFilesystem().DeleteDentry(vfs.MountNamespaceFromContext(ctx), vfsd); err != nil { + if !rp.MustBeDir() { + return syserror.ENOTDIR + } + mnt := rp.Mount() + if err := mnt.CheckBeginWrite(); err != nil { + return err + } + defer mnt.EndWrite() + vfsObj := rp.VirtualFilesystem() + if err := vfsObj.PrepareDeleteDentry(vfs.MountNamespaceFromContext(ctx), childVFSD); err != nil { return err } - // Remove from parent directory's childList. - vfsd.Parent().Impl().(*dentry).inode.impl.(*directory).childList.Remove(vfsd.Impl().(*dentry)) - inode.decLinksLocked() + parent.inode.impl.(*directory).childList.Remove(child) + child.inode.decLinksLocked() + vfsObj.CommitDeleteDentry(childVFSD) return nil } @@ -585,7 +650,7 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) { fs.mu.RLock() defer fs.mu.RUnlock() - _, _, err := walkExistingLocked(rp) + _, err := resolveLocked(rp) if err != nil { return nil, err } @@ -597,7 +662,7 @@ func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([ func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) (string, error) { fs.mu.RLock() defer fs.mu.RUnlock() - _, _, err := walkExistingLocked(rp) + _, err := resolveLocked(rp) if err != nil { return "", err } @@ -609,7 +674,7 @@ func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, nam func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { fs.mu.RLock() defer fs.mu.RUnlock() - _, _, err := walkExistingLocked(rp) + _, err := resolveLocked(rp) if err != nil { return err } @@ -621,7 +686,7 @@ func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { fs.mu.RLock() defer fs.mu.RUnlock() - _, _, err := walkExistingLocked(rp) + _, err := resolveLocked(rp) if err != nil { return err } diff --git a/pkg/sentry/fsimpl/memfs/memfs.go b/pkg/sentry/fsimpl/memfs/memfs.go index 9d509f6e4..8d0167c93 100644 --- a/pkg/sentry/fsimpl/memfs/memfs.go +++ b/pkg/sentry/fsimpl/memfs/memfs.go @@ -29,6 +29,7 @@ package memfs import ( "fmt" + "math" "sync" "sync/atomic" @@ -64,12 +65,6 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt func (fs *filesystem) Release() { } -// Sync implements vfs.FilesystemImpl.Sync. -func (fs *filesystem) Sync(ctx context.Context) error { - // All filesystem state is in-memory. - return nil -} - // dentry implements vfs.DentryImpl. type dentry struct { vfsd vfs.Dentry @@ -137,6 +132,8 @@ type inode struct { impl interface{} // immutable } +const maxLinks = math.MaxUint32 + func (i *inode) init(impl interface{}, fs *filesystem, creds *auth.Credentials, mode linux.FileMode) { i.refs = 1 i.mode = uint32(mode) @@ -147,20 +144,28 @@ func (i *inode) init(impl interface{}, fs *filesystem, creds *auth.Credentials, i.impl = impl } -// Preconditions: filesystem.mu must be locked for writing. +// incLinksLocked increments i's link count. +// +// Preconditions: filesystem.mu must be locked for writing. i.nlink != 0. +// i.nlink < maxLinks. func (i *inode) incLinksLocked() { - if atomic.AddUint32(&i.nlink, 1) <= 1 { + if i.nlink == 0 { panic("memfs.inode.incLinksLocked() called with no existing links") } + if i.nlink == maxLinks { + panic("memfs.inode.incLinksLocked() called with maximum link count") + } + atomic.AddUint32(&i.nlink, 1) } -// Preconditions: filesystem.mu must be locked for writing. +// decLinksLocked decrements i's link count. +// +// Preconditions: filesystem.mu must be locked for writing. i.nlink != 0. func (i *inode) decLinksLocked() { - if nlink := atomic.AddUint32(&i.nlink, ^uint32(0)); nlink == 0 { - i.decRef() - } else if nlink == ^uint32(0) { // negative overflow + if i.nlink == 0 { panic("memfs.inode.decLinksLocked() called with no existing links") } + atomic.AddUint32(&i.nlink, ^uint32(0)) } func (i *inode) incRef() { diff --git a/pkg/sentry/fsimpl/memfs/pipe_test.go b/pkg/sentry/fsimpl/memfs/pipe_test.go index 5bf527c80..be917aeee 100644 --- a/pkg/sentry/fsimpl/memfs/pipe_test.go +++ b/pkg/sentry/fsimpl/memfs/pipe_test.go @@ -19,6 +19,7 @@ import ( "testing" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/context/contexttest" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" @@ -38,7 +39,7 @@ func TestSeparateFDs(t *testing.T) { pop := vfs.PathOperation{ Root: root, Start: root, - Pathname: fileName, + Path: fspath.Parse(fileName), FollowFinalSymlink: true, } rfdchan := make(chan *vfs.FileDescription) @@ -76,7 +77,7 @@ func TestNonblockingRead(t *testing.T) { pop := vfs.PathOperation{ Root: root, Start: root, - Pathname: fileName, + Path: fspath.Parse(fileName), FollowFinalSymlink: true, } openOpts := vfs.OpenOptions{Flags: linux.O_RDONLY | linux.O_NONBLOCK} @@ -108,7 +109,7 @@ func TestNonblockingWriteError(t *testing.T) { pop := vfs.PathOperation{ Root: root, Start: root, - Pathname: fileName, + Path: fspath.Parse(fileName), FollowFinalSymlink: true, } openOpts := vfs.OpenOptions{Flags: linux.O_WRONLY | linux.O_NONBLOCK} @@ -126,7 +127,7 @@ func TestSingleFD(t *testing.T) { pop := vfs.PathOperation{ Root: root, Start: root, - Pathname: fileName, + Path: fspath.Parse(fileName), FollowFinalSymlink: true, } openOpts := vfs.OpenOptions{Flags: linux.O_RDWR} @@ -160,10 +161,9 @@ func setup(t *testing.T) (context.Context, *auth.Credentials, *vfs.VirtualFilesy // Create the pipe. root := mntns.Root() pop := vfs.PathOperation{ - Root: root, - Start: root, - Pathname: fileName, - FollowFinalSymlink: true, + Root: root, + Start: root, + Path: fspath.Parse(fileName), } mknodOpts := vfs.MknodOptions{Mode: linux.ModeNamedPipe | 0644} if err := vfsObj.MknodAt(ctx, creds, &pop, &mknodOpts); err != nil { @@ -174,7 +174,7 @@ func setup(t *testing.T) (context.Context, *auth.Credentials, *vfs.VirtualFilesy stat, err := vfsObj.StatAt(ctx, creds, &vfs.PathOperation{ Root: root, Start: root, - Pathname: fileName, + Path: fspath.Parse(fileName), FollowFinalSymlink: true, }, &vfs.StatOptions{}) if err != nil { diff --git a/pkg/sentry/vfs/dentry.go b/pkg/sentry/vfs/dentry.go index 6209eb053..1bc9c4a38 100644 --- a/pkg/sentry/vfs/dentry.go +++ b/pkg/sentry/vfs/dentry.go @@ -234,6 +234,18 @@ func (d *Dentry) InsertChild(child *Dentry, name string) { child.name = name } +// IsAncestorOf returns true if d is an ancestor of d2; that is, d is either +// d2's parent or an ancestor of d2's parent. +func (d *Dentry) IsAncestorOf(d2 *Dentry) bool { + for d2.parent != nil { + if d2.parent == d { + return true + } + d2 = d2.parent + } + return false +} + // PrepareDeleteDentry must be called before attempting to delete the file // represented by d. If PrepareDeleteDentry succeeds, the caller must call // AbortDeleteDentry or CommitDeleteDentry depending on the deletion's outcome. @@ -283,21 +295,6 @@ func (vfs *VirtualFilesystem) CommitDeleteDentry(d *Dentry) { } } -// DeleteDentry combines PrepareDeleteDentry and CommitDeleteDentry, as -// appropriate for in-memory filesystems that don't need to ensure that some -// external state change succeeds before committing the deletion. -// -// DeleteDentry is a mutator of d and d.Parent(). -// -// Preconditions: d is a child Dentry. -func (vfs *VirtualFilesystem) DeleteDentry(mntns *MountNamespace, d *Dentry) error { - if err := vfs.PrepareDeleteDentry(mntns, d); err != nil { - return err - } - vfs.CommitDeleteDentry(d) - return nil -} - // ForceDeleteDentry causes d to become disowned. It should only be used in // cases where VFS has no ability to stop the deletion (e.g. d represents the // local state of a file on a remote filesystem on which the file has already @@ -326,7 +323,7 @@ func (vfs *VirtualFilesystem) ForceDeleteDentry(d *Dentry) { // CommitRenameExchangeDentry depending on the rename's outcome. // // Preconditions: from is a child Dentry. If to is not nil, it must be a child -// Dentry from the same Filesystem. +// Dentry from the same Filesystem. from != to. func (vfs *VirtualFilesystem) PrepareRenameDentry(mntns *MountNamespace, from, to *Dentry) error { if checkInvariants { if from.parent == nil { diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index df03886c3..0b053201a 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -192,6 +192,8 @@ func (fd *FileDescription) Impl() FileDescriptionImpl { // be interpreted as IDs in the root UserNamespace (i.e. as auth.KUID and // auth.KGID respectively). // +// All methods may return errors not specified. +// // FileDescriptionImpl is analogous to Linux's struct file_operations. type FileDescriptionImpl interface { // Release is called when the associated FileDescription reaches zero @@ -220,6 +222,10 @@ type FileDescriptionImpl interface { // PRead reads from the file into dst, starting at the given offset, and // returns the number of bytes read. PRead is permitted to return partial // reads with a nil error. + // + // Errors: + // + // - If opts.Flags specifies unsupported options, PRead returns EOPNOTSUPP. PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts ReadOptions) (int64, error) // Read is similar to PRead, but does not specify an offset. @@ -229,6 +235,10 @@ type FileDescriptionImpl interface { // the number of bytes read; note that POSIX 2.9.7 "Thread Interactions // with Regular File Operations" requires that all operations that may // mutate the FileDescription offset are serialized. + // + // Errors: + // + // - If opts.Flags specifies unsupported options, Read returns EOPNOTSUPP. Read(ctx context.Context, dst usermem.IOSequence, opts ReadOptions) (int64, error) // PWrite writes src to the file, starting at the given offset, and returns @@ -238,6 +248,11 @@ type FileDescriptionImpl interface { // As in Linux (but not POSIX), if O_APPEND is in effect for the // FileDescription, PWrite should ignore the offset and append data to the // end of the file. + // + // Errors: + // + // - If opts.Flags specifies unsupported options, PWrite returns + // EOPNOTSUPP. PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts WriteOptions) (int64, error) // Write is similar to PWrite, but does not specify an offset, which is @@ -247,6 +262,10 @@ type FileDescriptionImpl interface { // PWrite that uses a FileDescription offset, to make it possible for // remote filesystems to implement O_APPEND correctly (i.e. atomically with // respect to writers outside the scope of VFS). + // + // Errors: + // + // - If opts.Flags specifies unsupported options, Write returns EOPNOTSUPP. Write(ctx context.Context, src usermem.IOSequence, opts WriteOptions) (int64, error) // IterDirents invokes cb on each entry in the directory represented by the diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index b766614e7..89bd58864 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -108,6 +108,24 @@ func (fs *Filesystem) DecRef() { // (responsible for actually implementing the operation) isn't known until path // resolution is complete. // +// Unless otherwise specified, FilesystemImpl methods are responsible for +// performing permission checks. In many cases, vfs package functions in +// permissions.go may be used to help perform these checks. +// +// When multiple specified error conditions apply to a given method call, the +// implementation may return any applicable errno unless otherwise specified, +// but returning the earliest error specified is preferable to maximize +// compatibility with Linux. +// +// All methods may return errors not specified, notably including: +// +// - ENOENT if a required path component does not exist. +// +// - ENOTDIR if an intermediate path component is not a directory. +// +// - Errors from vfs-package functions (ResolvingPath.Resolve*(), +// Mount.CheckBeginWrite(), permission-checking functions, etc.) +// // For all methods that take or return linux.Statx, Statx.Uid and Statx.Gid // should be interpreted as IDs in the root UserNamespace (i.e. as auth.KUID // and auth.KGID respectively). @@ -130,46 +148,223 @@ type FilesystemImpl interface { // GetDentryAt does not correspond directly to a Linux syscall; it is used // in the implementation of: // - // - Syscalls that need to resolve two paths: rename(), renameat(), - // renameat2(), link(), linkat(). + // - Syscalls that need to resolve two paths: link(), linkat(). // // - Syscalls that need to refer to a filesystem position outside the // context of a file description: chdir(), fchdir(), chroot(), mount(), // umount(). GetDentryAt(ctx context.Context, rp *ResolvingPath, opts GetDentryOptions) (*Dentry, error) + // GetParentDentryAt returns a Dentry representing the directory at the + // second-to-last path component in rp. (Note that, despite the name, this + // is not necessarily the parent directory of the file at rp, since the + // last path component in rp may be "." or "..".) A reference is taken on + // the returned Dentry. + // + // GetParentDentryAt does not correspond directly to a Linux syscall; it is + // used in the implementation of the rename() family of syscalls, which + // must resolve the parent directories of two paths. + // + // Preconditions: !rp.Done(). + // + // Postconditions: If GetParentDentryAt returns a nil error, then + // rp.Final(). If GetParentDentryAt returns an error returned by + // ResolvingPath.Resolve*(), then !rp.Done(). + GetParentDentryAt(ctx context.Context, rp *ResolvingPath) (*Dentry, error) + // LinkAt creates a hard link at rp representing the same file as vd. It // does not take ownership of references on vd. // - // The implementation is responsible for checking that vd.Mount() == - // rp.Mount(), and that vd does not represent a directory. + // Errors: + // + // - If the last path component in rp is "." or "..", LinkAt returns + // EEXIST. + // + // - If a file already exists at rp, LinkAt returns EEXIST. + // + // - If rp.MustBeDir(), LinkAt returns ENOENT. + // + // - If the directory in which the link would be created has been removed + // by RmdirAt or RenameAt, LinkAt returns ENOENT. + // + // - If rp.Mount != vd.Mount(), LinkAt returns EXDEV. + // + // - If vd represents a directory, LinkAt returns EPERM. + // + // - If vd represents a file for which all existing links have been + // removed, or a file created by open(O_TMPFILE|O_EXCL), LinkAt returns + // ENOENT. Equivalently, if vd represents a file with a link count of 0 not + // created by open(O_TMPFILE) without O_EXCL, LinkAt returns ENOENT. + // + // Preconditions: !rp.Done(). For the final path component in rp, + // !rp.ShouldFollowSymlink(). + // + // Postconditions: If LinkAt returns an error returned by + // ResolvingPath.Resolve*(), then !rp.Done(). LinkAt(ctx context.Context, rp *ResolvingPath, vd VirtualDentry) error // MkdirAt creates a directory at rp. + // + // Errors: + // + // - If the last path component in rp is "." or "..", MkdirAt returns + // EEXIST. + // + // - If a file already exists at rp, MkdirAt returns EEXIST. + // + // - If the directory in which the new directory would be created has been + // removed by RmdirAt or RenameAt, MkdirAt returns ENOENT. + // + // Preconditions: !rp.Done(). For the final path component in rp, + // !rp.ShouldFollowSymlink(). + // + // Postconditions: If MkdirAt returns an error returned by + // ResolvingPath.Resolve*(), then !rp.Done(). MkdirAt(ctx context.Context, rp *ResolvingPath, opts MkdirOptions) error // MknodAt creates a regular file, device special file, or named pipe at // rp. + // + // Errors: + // + // - If the last path component in rp is "." or "..", MknodAt returns + // EEXIST. + // + // - If a file already exists at rp, MknodAt returns EEXIST. + // + // - If rp.MustBeDir(), MknodAt returns ENOENT. + // + // - If the directory in which the file would be created has been removed + // by RmdirAt or RenameAt, MknodAt returns ENOENT. + // + // Preconditions: !rp.Done(). For the final path component in rp, + // !rp.ShouldFollowSymlink(). + // + // Postconditions: If MknodAt returns an error returned by + // ResolvingPath.Resolve*(), then !rp.Done(). MknodAt(ctx context.Context, rp *ResolvingPath, opts MknodOptions) error // OpenAt returns an FileDescription providing access to the file at rp. A // reference is taken on the returned FileDescription. + // + // Errors: + // + // - If opts.Flags specifies O_TMPFILE and this feature is unsupported by + // the implementation, OpenAt returns EOPNOTSUPP. (All other unsupported + // features are silently ignored, consistently with Linux's open*(2).) OpenAt(ctx context.Context, rp *ResolvingPath, opts OpenOptions) (*FileDescription, error) // ReadlinkAt returns the target of the symbolic link at rp. + // + // Errors: + // + // - If the file at rp is not a symbolic link, ReadlinkAt returns EINVAL. ReadlinkAt(ctx context.Context, rp *ResolvingPath) (string, error) - // RenameAt renames the Dentry represented by vd to rp. It does not take - // ownership of references on vd. + // RenameAt renames the file named oldName in directory oldParentVD to rp. + // It does not take ownership of references on oldParentVD. + // + // Errors [1]: + // + // - If opts.Flags specifies unsupported options, RenameAt returns EINVAL. + // + // - If the last path component in rp is "." or "..", and opts.Flags + // contains RENAME_NOREPLACE, RenameAt returns EEXIST. + // + // - If the last path component in rp is "." or "..", and opts.Flags does + // not contain RENAME_NOREPLACE, RenameAt returns EBUSY. + // + // - If rp.Mount != oldParentVD.Mount(), RenameAt returns EXDEV. // - // The implementation is responsible for checking that vd.Mount() == - // rp.Mount(). - RenameAt(ctx context.Context, rp *ResolvingPath, vd VirtualDentry, opts RenameOptions) error + // - If the renamed file is not a directory, and opts.MustBeDir is true, + // RenameAt returns ENOTDIR. + // + // - If renaming would replace an existing file and opts.Flags contains + // RENAME_NOREPLACE, RenameAt returns EEXIST. + // + // - If there is no existing file at rp and opts.Flags contains + // RENAME_EXCHANGE, RenameAt returns ENOENT. + // + // - If there is an existing non-directory file at rp, and rp.MustBeDir() + // is true, RenameAt returns ENOTDIR. + // + // - If the renamed file is not a directory, opts.Flags does not contain + // RENAME_EXCHANGE, and rp.MustBeDir() is true, RenameAt returns ENOTDIR. + // (This check is not subsumed by the check for directory replacement below + // since it applies even if there is no file to replace.) + // + // - If the renamed file is a directory, and the new parent directory of + // the renamed file is either the renamed directory or a descendant + // subdirectory of the renamed directory, RenameAt returns EINVAL. + // + // - If renaming would exchange the renamed file with an ancestor directory + // of the renamed file, RenameAt returns EINVAL. + // + // - If renaming would replace an ancestor directory of the renamed file, + // RenameAt returns ENOTEMPTY. (This check would be subsumed by the + // non-empty directory check below; however, this check takes place before + // the self-rename check.) + // + // - If the renamed file would replace or exchange with itself (i.e. the + // source and destination paths resolve to the same file), RenameAt returns + // nil, skipping the checks described below. + // + // - If the source or destination directory is not writable by the provider + // of rp.Credentials(), RenameAt returns EACCES. + // + // - If the renamed file is a directory, and renaming would replace a + // non-directory file, RenameAt returns ENOTDIR. + // + // - If the renamed file is not a directory, and renaming would replace a + // directory, RenameAt returns EISDIR. + // + // - If the new parent directory of the renamed file has been removed by + // RmdirAt or a preceding call to RenameAt, RenameAt returns ENOENT. + // + // - If the renamed file is a directory, it is not writable by the + // provider of rp.Credentials(), and the source and destination parent + // directories are different, RenameAt returns EACCES. (This is nominally + // required to change the ".." entry in the renamed directory.) + // + // - If renaming would replace a non-empty directory, RenameAt returns + // ENOTEMPTY. + // + // Preconditions: !rp.Done(). For the final path component in rp, + // !rp.ShouldFollowSymlink(). oldName is not "." or "..". + // + // Postconditions: If RenameAt returns an error returned by + // ResolvingPath.Resolve*(), then !rp.Done(). + // + // [1] "The worst of all namespace operations - renaming directory. + // "Perverted" doesn't even start to describe it. Somebody in UCB had a + // heck of a trip..." - fs/namei.c:vfs_rename() + RenameAt(ctx context.Context, rp *ResolvingPath, oldParentVD VirtualDentry, oldName string, opts RenameOptions) error // RmdirAt removes the directory at rp. + // + // Errors: + // + // - If the last path component in rp is ".", RmdirAt returns EINVAL. + // + // - If the last path component in rp is "..", RmdirAt returns ENOTEMPTY. + // + // - If no file exists at rp, RmdirAt returns ENOENT. + // + // - If the file at rp exists but is not a directory, RmdirAt returns + // ENOTDIR. + // + // Preconditions: !rp.Done(). For the final path component in rp, + // !rp.ShouldFollowSymlink(). + // + // Postconditions: If RmdirAt returns an error returned by + // ResolvingPath.Resolve*(), then !rp.Done(). RmdirAt(ctx context.Context, rp *ResolvingPath) error // SetStatAt updates metadata for the file at the given path. + // + // Errors: + // + // - If opts specifies unsupported options, SetStatAt returns EINVAL. SetStatAt(ctx context.Context, rp *ResolvingPath, opts SetStatOptions) error // StatAt returns metadata for the file at rp. @@ -181,9 +376,45 @@ type FilesystemImpl interface { StatFSAt(ctx context.Context, rp *ResolvingPath) (linux.Statfs, error) // SymlinkAt creates a symbolic link at rp referring to the given target. + // + // Errors: + // + // - If the last path component in rp is "." or "..", SymlinkAt returns + // EEXIST. + // + // - If a file already exists at rp, SymlinkAt returns EEXIST. + // + // - If rp.MustBeDir(), SymlinkAt returns ENOENT. + // + // - If the directory in which the symbolic link would be created has been + // removed by RmdirAt or RenameAt, SymlinkAt returns ENOENT. + // + // Preconditions: !rp.Done(). For the final path component in rp, + // !rp.ShouldFollowSymlink(). + // + // Postconditions: If SymlinkAt returns an error returned by + // ResolvingPath.Resolve*(), then !rp.Done(). SymlinkAt(ctx context.Context, rp *ResolvingPath, target string) error - // UnlinkAt removes the non-directory file at rp. + // UnlinkAt removes the file at rp. + // + // Errors: + // + // - If the last path component in rp is "." or "..", UnlinkAt returns + // EISDIR. + // + // - If no file exists at rp, UnlinkAt returns ENOENT. + // + // - If rp.MustBeDir(), and the file at rp exists and is not a directory, + // UnlinkAt returns ENOTDIR. + // + // - If the file at rp exists but is a directory, UnlinkAt returns EISDIR. + // + // Preconditions: !rp.Done(). For the final path component in rp, + // !rp.ShouldFollowSymlink(). + // + // Postconditions: If UnlinkAt returns an error returned by + // ResolvingPath.Resolve*(), then !rp.Done(). UnlinkAt(ctx context.Context, rp *ResolvingPath) error // ListxattrAt returns all extended attribute names for the file at rp. diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go index 97ee4a446..87d2b0d1c 100644 --- a/pkg/sentry/vfs/options.go +++ b/pkg/sentry/vfs/options.go @@ -83,6 +83,9 @@ type ReadOptions struct { type RenameOptions struct { // Flags contains flags as specified for renameat2(2). Flags uint32 + + // If MustBeDir is true, the renamed file must be a directory. + MustBeDir bool } // SetStatOptions contains options to VirtualFilesystem.SetStatAt(), diff --git a/pkg/sentry/vfs/resolving_path.go b/pkg/sentry/vfs/resolving_path.go index d580fd39e..f0641d314 100644 --- a/pkg/sentry/vfs/resolving_path.go +++ b/pkg/sentry/vfs/resolving_path.go @@ -112,30 +112,26 @@ var resolvingPathPool = sync.Pool{ }, } -func (vfs *VirtualFilesystem) getResolvingPath(creds *auth.Credentials, pop *PathOperation) (*ResolvingPath, error) { - path, err := fspath.Parse(pop.Pathname) - if err != nil { - return nil, err - } +func (vfs *VirtualFilesystem) getResolvingPath(creds *auth.Credentials, pop *PathOperation) *ResolvingPath { rp := resolvingPathPool.Get().(*ResolvingPath) rp.vfs = vfs rp.root = pop.Root rp.mount = pop.Start.mount rp.start = pop.Start.dentry - rp.pit = path.Begin + rp.pit = pop.Path.Begin rp.flags = 0 if pop.FollowFinalSymlink { rp.flags |= rpflagsFollowFinalSymlink } - rp.mustBeDir = path.Dir - rp.mustBeDirOrig = path.Dir + rp.mustBeDir = pop.Path.Dir + rp.mustBeDirOrig = pop.Path.Dir rp.symlinks = 0 rp.curPart = 0 rp.numOrigParts = 1 rp.creds = creds - rp.parts[0] = path.Begin - rp.origParts[0] = path.Begin - return rp, nil + rp.parts[0] = pop.Path.Begin + rp.origParts[0] = pop.Path.Begin + return rp } func (vfs *VirtualFilesystem) putResolvingPath(rp *ResolvingPath) { @@ -345,29 +341,34 @@ func (rp *ResolvingPath) ShouldFollowSymlink() bool { // symlink target and returns nil. Otherwise it returns a non-nil error. // // Preconditions: !rp.Done(). +// +// Postconditions: If HandleSymlink returns a nil error, then !rp.Done(). func (rp *ResolvingPath) HandleSymlink(target string) error { if rp.symlinks >= linux.MaxSymlinkTraversals { return syserror.ELOOP } - targetPath, err := fspath.Parse(target) - if err != nil { - return err + if len(target) == 0 { + return syserror.ENOENT } rp.symlinks++ + targetPath := fspath.Parse(target) if targetPath.Absolute { rp.absSymlinkTarget = targetPath return resolveAbsSymlinkError{} } - if !targetPath.Begin.Ok() { - panic(fmt.Sprintf("symbolic link has non-empty target %q that is both relative and has no path components?", target)) - } // Consume the path component that represented the symlink. rp.Advance() // Prepend the symlink target to the relative path. + if checkInvariants { + if !targetPath.HasComponents() { + panic(fmt.Sprintf("non-empty pathname %q parsed to relative path with no components", target)) + } + } rp.relpathPrepend(targetPath) return nil } +// Preconditions: path.HasComponents(). func (rp *ResolvingPath) relpathPrepend(path fspath.Path) { if rp.pit.Ok() { rp.parts[rp.curPart] = rp.pit @@ -467,6 +468,17 @@ func (rp *ResolvingPath) handleError(err error) bool { } } +// canHandleError returns true if err is an error returned by rp.Resolve*() +// that rp.handleError() may attempt to handle. +func (rp *ResolvingPath) canHandleError(err error) bool { + switch err.(type) { + case resolveMountRootOrJumpError, resolveMountPointError, resolveAbsSymlinkError: + return true + default: + return false + } +} + // MustBeDir returns true if the file traversed by rp must be a directory. func (rp *ResolvingPath) MustBeDir() bool { return rp.mustBeDir diff --git a/pkg/sentry/vfs/testutil.go b/pkg/sentry/vfs/testutil.go index d94117bce..ee5c8b9e2 100644 --- a/pkg/sentry/vfs/testutil.go +++ b/pkg/sentry/vfs/testutil.go @@ -57,6 +57,11 @@ func (fs *FDTestFilesystem) GetDentryAt(ctx context.Context, rp *ResolvingPath, return nil, syserror.EPERM } +// GetParentDentryAt implements FilesystemImpl.GetParentDentryAt. +func (fs *FDTestFilesystem) GetParentDentryAt(ctx context.Context, rp *ResolvingPath) (*Dentry, error) { + return nil, syserror.EPERM +} + // LinkAt implements FilesystemImpl.LinkAt. func (fs *FDTestFilesystem) LinkAt(ctx context.Context, rp *ResolvingPath, vd VirtualDentry) error { return syserror.EPERM @@ -83,7 +88,7 @@ func (fs *FDTestFilesystem) ReadlinkAt(ctx context.Context, rp *ResolvingPath) ( } // RenameAt implements FilesystemImpl.RenameAt. -func (fs *FDTestFilesystem) RenameAt(ctx context.Context, rp *ResolvingPath, vd VirtualDentry, opts RenameOptions) error { +func (fs *FDTestFilesystem) RenameAt(ctx context.Context, rp *ResolvingPath, oldParentVD VirtualDentry, oldName string, opts RenameOptions) error { return syserror.EPERM } diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index e60898d7c..3e4df8558 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -28,9 +28,11 @@ package vfs import ( + "fmt" "sync" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/syserror" @@ -111,11 +113,11 @@ type PathOperation struct { // are borrowed from the provider of the PathOperation (i.e. the caller of // the VFS method to which the PathOperation was passed). // - // Invariants: Start.Ok(). If Pathname.Absolute, then Start == Root. + // Invariants: Start.Ok(). If Path.Absolute, then Start == Root. Start VirtualDentry // Path is the pathname traversed by this operation. - Pathname string + Path fspath.Path // If FollowFinalSymlink is true, and the Dentry traversed by the final // path component represents a symbolic link, the symbolic link should be @@ -126,10 +128,7 @@ type PathOperation struct { // GetDentryAt returns a VirtualDentry representing the given path, at which a // file must exist. A reference is taken on the returned VirtualDentry. func (vfs *VirtualFilesystem) GetDentryAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *GetDentryOptions) (VirtualDentry, error) { - rp, err := vfs.getResolvingPath(creds, pop) - if err != nil { - return VirtualDentry{}, err - } + rp := vfs.getResolvingPath(creds, pop) for { d, err := rp.mount.fs.impl.GetDentryAt(ctx, rp, *opts) if err == nil { @@ -148,6 +147,33 @@ func (vfs *VirtualFilesystem) GetDentryAt(ctx context.Context, creds *auth.Crede } } +// Preconditions: pop.Path.Begin.Ok(). +func (vfs *VirtualFilesystem) getParentDirAndName(ctx context.Context, creds *auth.Credentials, pop *PathOperation) (VirtualDentry, string, error) { + rp := vfs.getResolvingPath(creds, pop) + for { + parent, err := rp.mount.fs.impl.GetParentDentryAt(ctx, rp) + if err == nil { + parentVD := VirtualDentry{ + mount: rp.mount, + dentry: parent, + } + rp.mount.IncRef() + name := rp.Component() + vfs.putResolvingPath(rp) + return parentVD, name, nil + } + if checkInvariants { + if rp.canHandleError(err) && rp.Done() { + panic(fmt.Sprintf("%T.GetParentDentryAt() consumed all path components and returned %T", rp.mount.fs.impl, err)) + } + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return VirtualDentry{}, "", err + } + } +} + // LinkAt creates a hard link at newpop representing the existing file at // oldpop. func (vfs *VirtualFilesystem) LinkAt(ctx context.Context, creds *auth.Credentials, oldpop, newpop *PathOperation) error { @@ -155,21 +181,36 @@ func (vfs *VirtualFilesystem) LinkAt(ctx context.Context, creds *auth.Credential if err != nil { return err } - rp, err := vfs.getResolvingPath(creds, newpop) - if err != nil { + + if !newpop.Path.Begin.Ok() { oldVD.DecRef() - return err + if newpop.Path.Absolute { + return syserror.EEXIST + } + return syserror.ENOENT } + if newpop.FollowFinalSymlink { + oldVD.DecRef() + ctx.Warningf("VirtualFilesystem.LinkAt: file creation paths can't follow final symlink") + return syserror.EINVAL + } + + rp := vfs.getResolvingPath(creds, newpop) for { err := rp.mount.fs.impl.LinkAt(ctx, rp, oldVD) if err == nil { - oldVD.DecRef() vfs.putResolvingPath(rp) + oldVD.DecRef() return nil } + if checkInvariants { + if rp.canHandleError(err) && rp.Done() { + panic(fmt.Sprintf("%T.LinkAt() consumed all path components and returned %T", rp.mount.fs.impl, err)) + } + } if !rp.handleError(err) { - oldVD.DecRef() vfs.putResolvingPath(rp) + oldVD.DecRef() return err } } @@ -177,19 +218,32 @@ func (vfs *VirtualFilesystem) LinkAt(ctx context.Context, creds *auth.Credential // MkdirAt creates a directory at the given path. func (vfs *VirtualFilesystem) MkdirAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *MkdirOptions) error { + if !pop.Path.Begin.Ok() { + if pop.Path.Absolute { + return syserror.EEXIST + } + return syserror.ENOENT + } + if pop.FollowFinalSymlink { + ctx.Warningf("VirtualFilesystem.MkdirAt: file creation paths can't follow final symlink") + return syserror.EINVAL + } // "Under Linux, apart from the permission bits, the S_ISVTX mode bit is // also honored." - mkdir(2) opts.Mode &= 0777 | linux.S_ISVTX - rp, err := vfs.getResolvingPath(creds, pop) - if err != nil { - return err - } + + rp := vfs.getResolvingPath(creds, pop) for { err := rp.mount.fs.impl.MkdirAt(ctx, rp, *opts) if err == nil { vfs.putResolvingPath(rp) return nil } + if checkInvariants { + if rp.canHandleError(err) && rp.Done() { + panic(fmt.Sprintf("%T.MkdirAt() consumed all path components and returned %T", rp.mount.fs.impl, err)) + } + } if !rp.handleError(err) { vfs.putResolvingPath(rp) return err @@ -200,16 +254,29 @@ func (vfs *VirtualFilesystem) MkdirAt(ctx context.Context, creds *auth.Credentia // MknodAt creates a file of the given mode at the given path. It returns an // error from the syserror package. func (vfs *VirtualFilesystem) MknodAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *MknodOptions) error { - rp, err := vfs.getResolvingPath(creds, pop) - if err != nil { - return nil + if !pop.Path.Begin.Ok() { + if pop.Path.Absolute { + return syserror.EEXIST + } + return syserror.ENOENT + } + if pop.FollowFinalSymlink { + ctx.Warningf("VirtualFilesystem.MknodAt: file creation paths can't follow final symlink") + return syserror.EINVAL } + + rp := vfs.getResolvingPath(creds, pop) for { - if err = rp.mount.fs.impl.MknodAt(ctx, rp, *opts); err == nil { + err := rp.mount.fs.impl.MknodAt(ctx, rp, *opts) + if err != nil { vfs.putResolvingPath(rp) return nil } - // Handle mount traversals. + if checkInvariants { + if rp.canHandleError(err) && rp.Done() { + panic(fmt.Sprintf("%T.MknodAt() consumed all path components and returned %T", rp.mount.fs.impl, err)) + } + } if !rp.handleError(err) { vfs.putResolvingPath(rp) return err @@ -259,10 +326,7 @@ func (vfs *VirtualFilesystem) OpenAt(ctx context.Context, creds *auth.Credential if opts.Flags&linux.O_NOFOLLOW != 0 { pop.FollowFinalSymlink = false } - rp, err := vfs.getResolvingPath(creds, pop) - if err != nil { - return nil, err - } + rp := vfs.getResolvingPath(creds, pop) if opts.Flags&linux.O_DIRECTORY != 0 { rp.mustBeDir = true rp.mustBeDirOrig = true @@ -282,10 +346,7 @@ func (vfs *VirtualFilesystem) OpenAt(ctx context.Context, creds *auth.Credential // ReadlinkAt returns the target of the symbolic link at the given path. func (vfs *VirtualFilesystem) ReadlinkAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation) (string, error) { - rp, err := vfs.getResolvingPath(creds, pop) - if err != nil { - return "", err - } + rp := vfs.getResolvingPath(creds, pop) for { target, err := rp.mount.fs.impl.ReadlinkAt(ctx, rp) if err == nil { @@ -301,25 +362,59 @@ func (vfs *VirtualFilesystem) ReadlinkAt(ctx context.Context, creds *auth.Creden // RenameAt renames the file at oldpop to newpop. func (vfs *VirtualFilesystem) RenameAt(ctx context.Context, creds *auth.Credentials, oldpop, newpop *PathOperation, opts *RenameOptions) error { - oldVD, err := vfs.GetDentryAt(ctx, creds, oldpop, &GetDentryOptions{}) - if err != nil { - return err + if !oldpop.Path.Begin.Ok() { + if oldpop.Path.Absolute { + return syserror.EBUSY + } + return syserror.ENOENT } - rp, err := vfs.getResolvingPath(creds, newpop) + if oldpop.FollowFinalSymlink { + ctx.Warningf("VirtualFilesystem.RenameAt: source path can't follow final symlink") + return syserror.EINVAL + } + + oldParentVD, oldName, err := vfs.getParentDirAndName(ctx, creds, oldpop) if err != nil { - oldVD.DecRef() return err } + if oldName == "." || oldName == ".." { + oldParentVD.DecRef() + return syserror.EBUSY + } + + if !newpop.Path.Begin.Ok() { + oldParentVD.DecRef() + if newpop.Path.Absolute { + return syserror.EBUSY + } + return syserror.ENOENT + } + if newpop.FollowFinalSymlink { + oldParentVD.DecRef() + ctx.Warningf("VirtualFilesystem.RenameAt: destination path can't follow final symlink") + return syserror.EINVAL + } + + rp := vfs.getResolvingPath(creds, newpop) + renameOpts := *opts + if oldpop.Path.Dir { + renameOpts.MustBeDir = true + } for { - err := rp.mount.fs.impl.RenameAt(ctx, rp, oldVD, *opts) + err := rp.mount.fs.impl.RenameAt(ctx, rp, oldParentVD, oldName, renameOpts) if err == nil { - oldVD.DecRef() vfs.putResolvingPath(rp) + oldParentVD.DecRef() return nil } + if checkInvariants { + if rp.canHandleError(err) && rp.Done() { + panic(fmt.Sprintf("%T.RenameAt() consumed all path components and returned %T", rp.mount.fs.impl, err)) + } + } if !rp.handleError(err) { - oldVD.DecRef() vfs.putResolvingPath(rp) + oldParentVD.DecRef() return err } } @@ -327,16 +422,29 @@ func (vfs *VirtualFilesystem) RenameAt(ctx context.Context, creds *auth.Credenti // RmdirAt removes the directory at the given path. func (vfs *VirtualFilesystem) RmdirAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation) error { - rp, err := vfs.getResolvingPath(creds, pop) - if err != nil { - return err + if !pop.Path.Begin.Ok() { + if pop.Path.Absolute { + return syserror.EBUSY + } + return syserror.ENOENT } + if pop.FollowFinalSymlink { + ctx.Warningf("VirtualFilesystem.RmdirAt: file deletion paths can't follow final symlink") + return syserror.EINVAL + } + + rp := vfs.getResolvingPath(creds, pop) for { err := rp.mount.fs.impl.RmdirAt(ctx, rp) if err == nil { vfs.putResolvingPath(rp) return nil } + if checkInvariants { + if rp.canHandleError(err) && rp.Done() { + panic(fmt.Sprintf("%T.RmdirAt() consumed all path components and returned %T", rp.mount.fs.impl, err)) + } + } if !rp.handleError(err) { vfs.putResolvingPath(rp) return err @@ -346,10 +454,7 @@ func (vfs *VirtualFilesystem) RmdirAt(ctx context.Context, creds *auth.Credentia // SetStatAt changes metadata for the file at the given path. func (vfs *VirtualFilesystem) SetStatAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *SetStatOptions) error { - rp, err := vfs.getResolvingPath(creds, pop) - if err != nil { - return err - } + rp := vfs.getResolvingPath(creds, pop) for { err := rp.mount.fs.impl.SetStatAt(ctx, rp, *opts) if err == nil { @@ -365,10 +470,7 @@ func (vfs *VirtualFilesystem) SetStatAt(ctx context.Context, creds *auth.Credent // StatAt returns metadata for the file at the given path. func (vfs *VirtualFilesystem) StatAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *StatOptions) (linux.Statx, error) { - rp, err := vfs.getResolvingPath(creds, pop) - if err != nil { - return linux.Statx{}, err - } + rp := vfs.getResolvingPath(creds, pop) for { stat, err := rp.mount.fs.impl.StatAt(ctx, rp, *opts) if err == nil { @@ -385,10 +487,7 @@ func (vfs *VirtualFilesystem) StatAt(ctx context.Context, creds *auth.Credential // StatFSAt returns metadata for the filesystem containing the file at the // given path. func (vfs *VirtualFilesystem) StatFSAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation) (linux.Statfs, error) { - rp, err := vfs.getResolvingPath(creds, pop) - if err != nil { - return linux.Statfs{}, err - } + rp := vfs.getResolvingPath(creds, pop) for { statfs, err := rp.mount.fs.impl.StatFSAt(ctx, rp) if err == nil { @@ -404,16 +503,29 @@ func (vfs *VirtualFilesystem) StatFSAt(ctx context.Context, creds *auth.Credenti // SymlinkAt creates a symbolic link at the given path with the given target. func (vfs *VirtualFilesystem) SymlinkAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, target string) error { - rp, err := vfs.getResolvingPath(creds, pop) - if err != nil { - return err + if !pop.Path.Begin.Ok() { + if pop.Path.Absolute { + return syserror.EEXIST + } + return syserror.ENOENT + } + if pop.FollowFinalSymlink { + ctx.Warningf("VirtualFilesystem.SymlinkAt: file creation paths can't follow final symlink") + return syserror.EINVAL } + + rp := vfs.getResolvingPath(creds, pop) for { err := rp.mount.fs.impl.SymlinkAt(ctx, rp, target) if err == nil { vfs.putResolvingPath(rp) return nil } + if checkInvariants { + if rp.canHandleError(err) && rp.Done() { + panic(fmt.Sprintf("%T.SymlinkAt() consumed all path components and returned %T", rp.mount.fs.impl, err)) + } + } if !rp.handleError(err) { vfs.putResolvingPath(rp) return err @@ -423,16 +535,29 @@ func (vfs *VirtualFilesystem) SymlinkAt(ctx context.Context, creds *auth.Credent // UnlinkAt deletes the non-directory file at the given path. func (vfs *VirtualFilesystem) UnlinkAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation) error { - rp, err := vfs.getResolvingPath(creds, pop) - if err != nil { - return err + if !pop.Path.Begin.Ok() { + if pop.Path.Absolute { + return syserror.EBUSY + } + return syserror.ENOENT + } + if pop.FollowFinalSymlink { + ctx.Warningf("VirtualFilesystem.UnlinkAt: file deletion paths can't follow final symlink") + return syserror.EINVAL } + + rp := vfs.getResolvingPath(creds, pop) for { err := rp.mount.fs.impl.UnlinkAt(ctx, rp) if err == nil { vfs.putResolvingPath(rp) return nil } + if checkInvariants { + if rp.canHandleError(err) && rp.Done() { + panic(fmt.Sprintf("%T.UnlinkAt() consumed all path components and returned %T", rp.mount.fs.impl, err)) + } + } if !rp.handleError(err) { vfs.putResolvingPath(rp) return err @@ -443,10 +568,7 @@ func (vfs *VirtualFilesystem) UnlinkAt(ctx context.Context, creds *auth.Credenti // ListxattrAt returns all extended attribute names for the file at the given // path. func (vfs *VirtualFilesystem) ListxattrAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation) ([]string, error) { - rp, err := vfs.getResolvingPath(creds, pop) - if err != nil { - return nil, err - } + rp := vfs.getResolvingPath(creds, pop) for { names, err := rp.mount.fs.impl.ListxattrAt(ctx, rp) if err == nil { @@ -471,10 +593,7 @@ func (vfs *VirtualFilesystem) ListxattrAt(ctx context.Context, creds *auth.Crede // GetxattrAt returns the value associated with the given extended attribute // for the file at the given path. func (vfs *VirtualFilesystem) GetxattrAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, name string) (string, error) { - rp, err := vfs.getResolvingPath(creds, pop) - if err != nil { - return "", err - } + rp := vfs.getResolvingPath(creds, pop) for { val, err := rp.mount.fs.impl.GetxattrAt(ctx, rp, name) if err == nil { @@ -491,10 +610,7 @@ func (vfs *VirtualFilesystem) GetxattrAt(ctx context.Context, creds *auth.Creden // SetxattrAt changes the value associated with the given extended attribute // for the file at the given path. func (vfs *VirtualFilesystem) SetxattrAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *SetxattrOptions) error { - rp, err := vfs.getResolvingPath(creds, pop) - if err != nil { - return err - } + rp := vfs.getResolvingPath(creds, pop) for { err := rp.mount.fs.impl.SetxattrAt(ctx, rp, *opts) if err == nil { @@ -510,10 +626,7 @@ func (vfs *VirtualFilesystem) SetxattrAt(ctx context.Context, creds *auth.Creden // RemovexattrAt removes the given extended attribute from the file at rp. func (vfs *VirtualFilesystem) RemovexattrAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, name string) error { - rp, err := vfs.getResolvingPath(creds, pop) - if err != nil { - return err - } + rp := vfs.getResolvingPath(creds, pop) for { err := rp.mount.fs.impl.RemovexattrAt(ctx, rp, name) if err == nil { diff --git a/pkg/syserror/syserror.go b/pkg/syserror/syserror.go index 1987e89cc..2269f6237 100644 --- a/pkg/syserror/syserror.go +++ b/pkg/syserror/syserror.go @@ -45,6 +45,7 @@ var ( ELIBBAD = error(syscall.ELIBBAD) ELOOP = error(syscall.ELOOP) EMFILE = error(syscall.EMFILE) + EMLINK = error(syscall.EMLINK) EMSGSIZE = error(syscall.EMSGSIZE) ENAMETOOLONG = error(syscall.ENAMETOOLONG) ENOATTR = ENODATA -- cgit v1.2.3 From 574e988f2bc6060078a17f37a377441703c52a22 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Mon, 23 Dec 2019 17:31:20 -0800 Subject: Fix deadlock in kernfs.Filesystem.revalidateChildLocked It was calling Dentry.InsertChild with the dentry's mutex already locked. Updates #1035 PiperOrigin-RevId: 286962742 --- pkg/sentry/fsimpl/kernfs/filesystem.go | 2 +- pkg/sentry/fsimpl/kernfs/kernfs.go | 12 ++++++++++-- 2 files changed, 11 insertions(+), 3 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index a6f9fced5..79759e0fc 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -122,7 +122,7 @@ func (fs *Filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.Vir return nil, err } // Reference on childVFSD dropped by a corresponding Valid. - parent.InsertChild(name, childVFSD) + parent.insertChildLocked(name, childVFSD) } return childVFSD.Impl().(*Dentry), nil } diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index bb01c3d01..ac802218d 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -239,14 +239,22 @@ func (d *Dentry) destroy() { // // Precondition: d must represent a directory inode. func (d *Dentry) InsertChild(name string, child *vfs.Dentry) { + d.dirMu.Lock() + d.insertChildLocked(name, child) + d.dirMu.Unlock() +} + +// insertChildLocked is equivalent to InsertChild, with additional +// preconditions. +// +// Precondition: d.dirMu must be locked. +func (d *Dentry) insertChildLocked(name string, child *vfs.Dentry) { if !d.isDir() { panic(fmt.Sprintf("InsertChild called on non-directory Dentry: %+v.", d)) } vfsDentry := d.VFSDentry() vfsDentry.IncRef() // DecRef in child's Dentry.destroy. - d.dirMu.Lock() vfsDentry.InsertChild(child, name) - d.dirMu.Unlock() } // The Inode interface maps filesystem-level operations that operate on paths to -- cgit v1.2.3 From 7b83d21856e569742397ab8b0146910eeff1462f Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Tue, 24 Dec 2019 10:50:05 +0800 Subject: slight changes to ring0&pagetables for Arm64 There are 2 jobs have been finished in this patch: 1, a comment was added to explain the purpose of the extra NOPs in Vectors(). 2, some merge errors were fixed. Signed-off-by: Bin Lu --- pkg/sentry/platform/ring0/entry_arm64.s | 4 ++++ pkg/sentry/platform/ring0/lib_arm64.s | 7 +++++-- pkg/sentry/platform/ring0/pagetables/BUILD | 5 ++++- 3 files changed, 13 insertions(+), 3 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/ring0/entry_arm64.s b/pkg/sentry/platform/ring0/entry_arm64.s index add2c3e08..813ef9822 100644 --- a/pkg/sentry/platform/ring0/entry_arm64.s +++ b/pkg/sentry/platform/ring0/entry_arm64.s @@ -554,6 +554,10 @@ TEXT ·Vectors(SB),NOSPLIT,$0 B ·El0_error_invalid(SB) nop31Instructions() + // The exception-vector-table is required to be 11-bits aligned. + // Please see Linux source code as reference: arch/arm64/kernel/entry.s. + // For gvisor, I defined it as 4K in length, filled the 2nd 2K part with NOPs. + // So that, I can safely move the 1st 2K part into the address with 11-bits alignment. WORD $0xd503201f //nop nop31Instructions() WORD $0xd503201f diff --git a/pkg/sentry/platform/ring0/lib_arm64.s b/pkg/sentry/platform/ring0/lib_arm64.s index 1c9171004..0e6a6235b 100644 --- a/pkg/sentry/platform/ring0/lib_arm64.s +++ b/pkg/sentry/platform/ring0/lib_arm64.s @@ -12,12 +12,15 @@ // See the License for the specific language governing permissions and // limitations under the License. +#include "funcdata.h" +#include "textflag.h" + TEXT ·CPACREL1(SB),NOSPLIT,$0-8 WORD $0xd5381041 // MRS CPACR_EL1, R1 MOVD R1, ret+0(FP) RET -TEXT ·FPCR(SB),NOSPLIT,$0-8 +TEXT ·GetFPCR(SB),NOSPLIT,$0-8 WORD $0xd53b4201 // MRS NZCV, R1 MOVD R1, ret+0(FP) RET @@ -27,7 +30,7 @@ TEXT ·GetFPSR(SB),NOSPLIT,$0-8 MOVD R1, ret+0(FP) RET -TEXT ·FPCR(SB),NOSPLIT,$0-8 +TEXT ·SetFPCR(SB),NOSPLIT,$0-8 MOVD addr+0(FP), R1 WORD $0xd51b4201 // MSR R1, NZCV RET diff --git a/pkg/sentry/platform/ring0/pagetables/BUILD b/pkg/sentry/platform/ring0/pagetables/BUILD index e2e15ba5c..09ecc3b09 100644 --- a/pkg/sentry/platform/ring0/pagetables/BUILD +++ b/pkg/sentry/platform/ring0/pagetables/BUILD @@ -11,7 +11,10 @@ config_setting( go_template( name = "generic_walker", - srcs = ["walker_amd64.go"], + srcs = select({ + ":aarch64": ["walker_arm64.go",], + "//conditions:default": ["walker_amd64.go",], + }), opt_types = [ "Visitor", ], -- cgit v1.2.3 From e013c48c78c9a7daf245b7de9563e3a0bd8a1e97 Mon Sep 17 00:00:00 2001 From: Ryan Heacock Date: Tue, 24 Dec 2019 08:48:14 -0800 Subject: Enable IP_RECVTOS socket option for datagram sockets Added the ability to get/set the IP_RECVTOS socket option on UDP endpoints. If enabled, TOS from the incoming Network Header passed as ancillary data in the ControlMessages. Test: * Added unit test to udp_test.go that tests getting/setting as well as verifying that we receive expected TOS from incoming packet. * Added a syscall test PiperOrigin-RevId: 287029703 --- pkg/sentry/socket/control/control.go | 2 +- pkg/sentry/socket/netstack/netstack.go | 42 ++++++++++++++++- pkg/tcpip/checker/checker.go | 16 +++++++ pkg/tcpip/stack/nic.go | 2 +- pkg/tcpip/stack/stack.go | 2 +- pkg/tcpip/tcpip.go | 6 ++- pkg/tcpip/transport/raw/endpoint.go | 2 +- pkg/tcpip/transport/udp/endpoint.go | 31 ++++++++++++- pkg/tcpip/transport/udp/udp_test.go | 69 ++++++++++++++++++++++++---- test/syscalls/linux/socket_ip_udp_generic.cc | 40 ++++++++++++++++ test/syscalls/linux/udp_socket_test_cases.cc | 8 ++-- 11 files changed, 201 insertions(+), 19 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/control/control.go b/pkg/sentry/socket/control/control.go index af1a4e95f..b649dd021 100644 --- a/pkg/sentry/socket/control/control.go +++ b/pkg/sentry/socket/control/control.go @@ -327,7 +327,7 @@ func PackInq(t *kernel.Task, inq int32, buf []byte) []byte { } // PackTOS packs an IP_TOS socket control message. -func PackTOS(t *kernel.Task, tos int8, buf []byte) []byte { +func PackTOS(t *kernel.Task, tos uint8, buf []byte) []byte { return putCmsgStruct( buf, linux.SOL_IP, diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 140851c17..d2f263402 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -1323,6 +1323,21 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in } return int32(v), nil + case linux.IP_RECVTOS: + if outLen < sizeOfInt32 { + return nil, syserr.ErrInvalidArgument + } + + var v tcpip.ReceiveTOSOption + if err := ep.GetSockOpt(&v); err != nil { + return nil, syserr.TranslateNetstackError(err) + } + + if v { + return int32(1), nil + } + return int32(0), nil + default: emitUnimplementedEventIP(t, name) } @@ -1808,6 +1823,16 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s } return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.IPv4TOSOption(v))) + case linux.IP_RECVTOS: + v, err := parseIntOrChar(optVal) + if err != nil { + return err + } + + return syserr.TranslateNetstackError(ep.SetSockOpt( + tcpip.ReceiveTOSOption(v != 0), + )) + case linux.IP_ADD_SOURCE_MEMBERSHIP, linux.IP_BIND_ADDRESS_NO_PORT, linux.IP_BLOCK_SOURCE, @@ -1828,7 +1853,6 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s linux.IP_RECVFRAGSIZE, linux.IP_RECVOPTS, linux.IP_RECVORIGDSTADDR, - linux.IP_RECVTOS, linux.IP_RECVTTL, linux.IP_RETOPTS, linux.IP_TRANSPARENT, @@ -2139,6 +2163,21 @@ func (s *SocketOperations) fillCmsgInq(cmsg *socket.ControlMessages) { cmsg.IP.Inq = int32(len(s.readView) + rcvBufUsed) } +func (s *SocketOperations) fillCmsgTOS(cmsg *socket.ControlMessages) { + if s.skType != linux.SOCK_DGRAM { + return + } + var receiveTOS tcpip.ReceiveTOSOption + if err := s.Endpoint.GetSockOpt(&receiveTOS); err != nil { + return + } + if !receiveTOS { + return + } + cmsg.IP.HasTOS = s.readCM.HasTOS + cmsg.IP.TOS = s.readCM.TOS +} + // nonBlockingRead issues a non-blocking read. // // TODO(b/78348848): Support timestamps for stream sockets. @@ -2244,6 +2283,7 @@ func (s *SocketOperations) nonBlockingRead(ctx context.Context, dst usermem.IOSe cmsg := s.controlMessages() s.fillCmsgInq(&cmsg) + s.fillCmsgTOS(&cmsg) return n, flags, addr, addrLen, cmsg, syserr.FromError(err) } diff --git a/pkg/tcpip/checker/checker.go b/pkg/tcpip/checker/checker.go index 2f15bf1f1..542abc99d 100644 --- a/pkg/tcpip/checker/checker.go +++ b/pkg/tcpip/checker/checker.go @@ -33,6 +33,9 @@ type NetworkChecker func(*testing.T, []header.Network) // TransportChecker is a function to check a property of a transport packet. type TransportChecker func(*testing.T, header.Transport) +// ControlMessagesChecker is a function to check a property of ancillary data. +type ControlMessagesChecker func(*testing.T, tcpip.ControlMessages) + // IPv4 checks the validity and properties of the given IPv4 packet. It is // expected to be used in conjunction with other network checkers for specific // properties. For example, to check the source and destination address, one @@ -158,6 +161,19 @@ func FragmentFlags(flags uint8) NetworkChecker { } } +// ReceiveTOS creates a checker that checks the TOS field in ControlMessages. +func ReceiveTOS(want uint8) ControlMessagesChecker { + return func(t *testing.T, cm tcpip.ControlMessages) { + t.Helper() + if !cm.HasTOS { + t.Fatalf("got cm.HasTOS = %t, want cm.TOS = %d", cm.HasTOS, want) + } + if got := cm.TOS; got != want { + t.Fatalf("got cm.TOS = %d, want %d", got, want) + } + } +} + // TOS creates a checker that checks the TOS field. func TOS(tos uint8, label uint32) NetworkChecker { return func(t *testing.T, h []header.Network) { diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index ddd014658..a4556674b 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -575,7 +575,7 @@ func (n *NIC) RemoveAddressRange(subnet tcpip.Subnet) { n.mu.Unlock() } -// Subnets returns the Subnets associated with this NIC. +// AddressRanges returns the Subnets associated with this NIC. func (n *NIC) AddressRanges() []tcpip.Subnet { n.mu.RLock() defer n.mu.RUnlock() diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index 7a9600679..251336224 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -829,7 +829,7 @@ func (s *Stack) CheckNIC(id tcpip.NICID) bool { return false } -// NICSubnets returns a map of NICIDs to their associated subnets. +// NICAddressRanges returns a map of NICIDs to their associated subnets. func (s *Stack) NICAddressRanges() map[tcpip.NICID][]tcpip.Subnet { s.mu.RLock() defer s.mu.RUnlock() diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index f62fd729f..5c7b2af88 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -322,7 +322,7 @@ type ControlMessages struct { HasTOS bool // TOS is the IPv4 type of service of the associated packet. - TOS int8 + TOS uint8 // HasTClass indicates whether Tclass is valid/set. HasTClass bool @@ -666,6 +666,10 @@ type IPv4TOSOption uint8 // for all subsequent outgoing IPv6 packets from the endpoint. type IPv6TrafficClassOption uint8 +// ReceiveTOSOption is used by SetSockOpt/GetSockOpt to specify if the TOS +// ancillary message is passed with incoming packets. +type ReceiveTOSOption bool + // Route is a row in the routing table. It specifies through which NIC (and // gateway) sets of packets should be routed. A row is considered viable if the // masked target address matches the destination address in the row. diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index 5aafe2615..6d23ab5a1 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -510,7 +510,7 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { } // SetSockOptInt implements tcpip.Endpoint.SetSockOptInt. -func (ep *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error { +func (e *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error { return tcpip.ErrUnknownProtocolOption } diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 1ac4705af..269470ed4 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -32,6 +32,7 @@ type udpPacket struct { senderAddress tcpip.FullAddress data buffer.VectorisedView `state:".(buffer.VectorisedView)"` timestamp int64 + tos uint8 } // EndpointState represents the state of a UDP endpoint. @@ -114,6 +115,10 @@ type endpoint struct { // applied while sending packets. Defaults to 0 as on Linux. sendTOS uint8 + // receiveTOS determines if the incoming IPv4 TOS header field is passed + // as ancillary data to ControlMessages on Read. + receiveTOS bool + // shutdownFlags represent the current shutdown state of the endpoint. shutdownFlags tcpip.ShutdownFlags @@ -244,7 +249,12 @@ func (e *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMess *addr = p.senderAddress } - return p.data.ToView(), tcpip.ControlMessages{HasTimestamp: true, Timestamp: p.timestamp}, nil + return p.data.ToView(), tcpip.ControlMessages{ + HasTimestamp: true, + Timestamp: p.timestamp, + HasTOS: e.receiveTOS, + TOS: p.tos, + }, nil } // prepareForWrite prepares the endpoint for sending data. In particular, it @@ -656,6 +666,12 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.sendTOS = uint8(v) e.mu.Unlock() return nil + + case tcpip.ReceiveTOSOption: + e.mu.Lock() + e.receiveTOS = bool(v) + e.mu.Unlock() + return nil } return nil } @@ -792,6 +808,12 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { e.mu.RUnlock() return nil + case *tcpip.ReceiveTOSOption: + e.mu.RLock() + *o = tcpip.ReceiveTOSOption(e.receiveTOS) + e.mu.RUnlock() + return nil + default: return tcpip.ErrUnknownProtocolOption } @@ -1238,6 +1260,13 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk e.rcvList.PushBack(packet) e.rcvBufSize += pkt.Data.Size() + // Save any useful information from the NetworkHeader to the packet. + switch r.NetProto { + case header.IPv4ProtocolNumber: + // This packet has already been validated before being passed up the stack. + packet.tos, _ = header.IPv4(pkt.NetworkHeader).TOS() + } + packet.timestamp = e.stack.NowNanoseconds() e.rcvMu.Unlock() diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index 7051a7a9c..43b8b35ba 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -56,6 +56,7 @@ const ( multicastAddr = "\xe8\x2b\xd3\xea" multicastV6Addr = "\xff\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00" broadcastAddr = header.IPv4Broadcast + testTOS = 0x80 // defaultMTU is the MTU, in bytes, used throughout the tests, except // where another value is explicitly used. It is chosen to match the MTU @@ -453,6 +454,7 @@ func (c *testContext) injectV4Packet(payload []byte, h *header4Tuple, valid bool ip := header.IPv4(buf) ip.Encode(&header.IPv4Fields{ IHL: header.IPv4MinimumSize, + TOS: testTOS, TotalLength: uint16(len(buf)), TTL: 65, Protocol: uint8(udp.ProtocolNumber), @@ -556,8 +558,8 @@ func TestBindToDeviceOption(t *testing.T) { // testReadInternal sends a packet of the given test flow into the stack by // injecting it into the link endpoint. It then attempts to read it from the // UDP endpoint and depending on if this was expected to succeed verifies its -// correctness. -func testReadInternal(c *testContext, flow testFlow, packetShouldBeDropped, expectReadError bool) { +// correctness including any additional checker functions provided. +func testReadInternal(c *testContext, flow testFlow, packetShouldBeDropped, expectReadError bool, checkers ...checker.ControlMessagesChecker) { c.t.Helper() payload := newPayload() @@ -572,12 +574,12 @@ func testReadInternal(c *testContext, flow testFlow, packetShouldBeDropped, expe epstats := c.ep.Stats().(*tcpip.TransportEndpointStats).Clone() var addr tcpip.FullAddress - v, _, err := c.ep.Read(&addr) + v, cm, err := c.ep.Read(&addr) if err == tcpip.ErrWouldBlock { // Wait for data to become available. select { case <-ch: - v, _, err = c.ep.Read(&addr) + v, cm, err = c.ep.Read(&addr) case <-time.After(300 * time.Millisecond): if packetShouldBeDropped { @@ -610,15 +612,21 @@ func testReadInternal(c *testContext, flow testFlow, packetShouldBeDropped, expe if !bytes.Equal(payload, v) { c.t.Fatalf("bad payload: got %x, want %x", v, payload) } + + // Run any checkers against the ControlMessages. + for _, f := range checkers { + f(c.t, cm) + } + c.checkEndpointReadStats(1, epstats, err) } // testRead sends a packet of the given test flow into the stack by injecting it // into the link endpoint. It then reads it from the UDP endpoint and verifies -// its correctness. -func testRead(c *testContext, flow testFlow) { +// its correctness including any additional checker functions provided. +func testRead(c *testContext, flow testFlow, checkers ...checker.ControlMessagesChecker) { c.t.Helper() - testReadInternal(c, flow, false /* packetShouldBeDropped */, false /* expectReadError */) + testReadInternal(c, flow, false /* packetShouldBeDropped */, false /* expectReadError */, checkers...) } // testFailingRead sends a packet of the given test flow into the stack by @@ -1286,7 +1294,7 @@ func TestTOSV4(t *testing.T) { c.createEndpointForFlow(flow) - const tos = 0xC0 + const tos = testTOS var v tcpip.IPv4TOSOption if err := c.ep.GetSockOpt(&v); err != nil { c.t.Errorf("GetSockopt failed: %s", err) @@ -1321,7 +1329,7 @@ func TestTOSV6(t *testing.T) { c.createEndpointForFlow(flow) - const tos = 0xC0 + const tos = testTOS var v tcpip.IPv6TrafficClassOption if err := c.ep.GetSockOpt(&v); err != nil { c.t.Errorf("GetSockopt failed: %s", err) @@ -1348,6 +1356,49 @@ func TestTOSV6(t *testing.T) { } } +func TestReceiveTOSV4(t *testing.T) { + for _, flow := range []testFlow{unicastV4, broadcast} { + t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpointForFlow(flow) + + // Verify that setting and reading the option works. + const recvTos = true + var v tcpip.ReceiveTOSOption + if err := c.ep.GetSockOpt(&v); err != nil { + c.t.Errorf("GetSockopt failed: %s", err) + } + // Test for expected default value. + if v != false { + c.t.Errorf("got GetSockOpt(...) = %t, want = %t", v, false) + } + + if err := c.ep.SetSockOpt(tcpip.ReceiveTOSOption(recvTos)); err != nil { + c.t.Errorf("SetSockOpt(%#v) failed: %s", tcpip.ReceiveTOSOption(recvTos), err) + } + + if err := c.ep.GetSockOpt(&v); err != nil { + c.t.Errorf("GetSockopt failed: %s", err) + } + + if want := tcpip.ReceiveTOSOption(recvTos); v != want { + c.t.Errorf("got GetSockOpt(...) = %t, want = %t", v, want) + } + + // Bind to wildcard. + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + c.t.Fatalf("Bind failed: %s", err) + } + + // Verify that the correct received TOS is actually handed through as + // ancillary data to the ControlMessages struct. + testRead(c, flow, checker.ReceiveTOS(testTOS)) + }) + } +} + func TestMulticastInterfaceOption(t *testing.T) { for _, flow := range []testFlow{multicastV4, multicastV4in6, multicastV6, multicastV6Only} { t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { diff --git a/test/syscalls/linux/socket_ip_udp_generic.cc b/test/syscalls/linux/socket_ip_udp_generic.cc index 66eb68857..53290bed7 100644 --- a/test/syscalls/linux/socket_ip_udp_generic.cc +++ b/test/syscalls/linux/socket_ip_udp_generic.cc @@ -209,6 +209,46 @@ TEST_P(UDPSocketPairTest, SetMulticastLoopChar) { EXPECT_EQ(get, kSockOptOn); } +// Ensure that Receiving TOS is off by default. +TEST_P(UDPSocketPairTest, RecvTosDefault) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT( + getsockopt(sockets->first_fd(), IPPROTO_IP, IP_RECVTOS, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kSockOptOff); +} + +// Test that setting and getting IP_RECVTOS works as expected. +TEST_P(UDPSocketPairTest, SetRecvTos) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + ASSERT_THAT(setsockopt(sockets->first_fd(), IPPROTO_IP, IP_RECVTOS, + &kSockOptOff, sizeof(kSockOptOff)), + SyscallSucceeds()); + + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT( + getsockopt(sockets->first_fd(), IPPROTO_IP, IP_RECVTOS, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kSockOptOff); + + ASSERT_THAT(setsockopt(sockets->first_fd(), IPPROTO_IP, IP_RECVTOS, + &kSockOptOn, sizeof(kSockOptOn)), + SyscallSucceeds()); + + ASSERT_THAT( + getsockopt(sockets->first_fd(), IPPROTO_IP, IP_RECVTOS, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kSockOptOn); +} + TEST_P(UDPSocketPairTest, ReuseAddrDefault) { auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); diff --git a/test/syscalls/linux/udp_socket_test_cases.cc b/test/syscalls/linux/udp_socket_test_cases.cc index dc35c2f50..68e0a8109 100644 --- a/test/syscalls/linux/udp_socket_test_cases.cc +++ b/test/syscalls/linux/udp_socket_test_cases.cc @@ -1349,8 +1349,9 @@ TEST_P(UdpSocketTest, TimestampIoctlPersistence) { // outgoing packets, and that a receiving socket with IP_RECVTOS or // IPV6_RECVTCLASS will create the corresponding control message. TEST_P(UdpSocketTest, SetAndReceiveTOS) { - // TODO(b/68320120): IP_RECVTOS/IPV6_RECVTCLASS not supported for netstack. - SKIP_IF(IsRunningOnGvisor() && !IsRunningWithHostinet()); + // TODO(b/68320120): IPV6_RECVTCLASS not supported for netstack. + SKIP_IF((GetParam() != AddressFamily::kIpv4) && IsRunningOnGvisor() && + !IsRunningWithHostinet()); ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); ASSERT_THAT(connect(t_, addr_[0], addrlen_), SyscallSucceeds()); @@ -1421,7 +1422,8 @@ TEST_P(UdpSocketTest, SetAndReceiveTOS) { // TOS byte on outgoing packets, and that a receiving socket with IP_RECVTOS or // IPV6_RECVTCLASS will create the corresponding control message. TEST_P(UdpSocketTest, SendAndReceiveTOS) { - // TODO(b/68320120): IP_RECVTOS/IPV6_RECVTCLASS not supported for netstack. + // TODO(b/68320120): IPV6_RECVTCLASS not supported for netstack. + // TODO(b/146661005): Setting TOS via cmsg not supported for netstack. SKIP_IF(IsRunningOnGvisor() && !IsRunningWithHostinet()); ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); ASSERT_THAT(connect(t_, addr_[0], addrlen_), SyscallSucceeds()); -- cgit v1.2.3 From 87e4d03fdf576348ac7023c599e0fc66ad4cccbd Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Thu, 26 Dec 2019 13:04:14 -0800 Subject: Automated rollback of changelist 287029703 PiperOrigin-RevId: 287217899 --- pkg/sentry/socket/control/control.go | 2 +- pkg/sentry/socket/netstack/netstack.go | 42 +---------------- pkg/tcpip/checker/checker.go | 16 ------- pkg/tcpip/stack/nic.go | 2 +- pkg/tcpip/stack/stack.go | 2 +- pkg/tcpip/tcpip.go | 6 +-- pkg/tcpip/transport/raw/endpoint.go | 2 +- pkg/tcpip/transport/udp/endpoint.go | 31 +------------ pkg/tcpip/transport/udp/udp_test.go | 69 ++++------------------------ test/syscalls/linux/socket_ip_udp_generic.cc | 40 ---------------- test/syscalls/linux/udp_socket_test_cases.cc | 8 ++-- 11 files changed, 19 insertions(+), 201 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/control/control.go b/pkg/sentry/socket/control/control.go index b649dd021..af1a4e95f 100644 --- a/pkg/sentry/socket/control/control.go +++ b/pkg/sentry/socket/control/control.go @@ -327,7 +327,7 @@ func PackInq(t *kernel.Task, inq int32, buf []byte) []byte { } // PackTOS packs an IP_TOS socket control message. -func PackTOS(t *kernel.Task, tos uint8, buf []byte) []byte { +func PackTOS(t *kernel.Task, tos int8, buf []byte) []byte { return putCmsgStruct( buf, linux.SOL_IP, diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index d2f263402..140851c17 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -1323,21 +1323,6 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in } return int32(v), nil - case linux.IP_RECVTOS: - if outLen < sizeOfInt32 { - return nil, syserr.ErrInvalidArgument - } - - var v tcpip.ReceiveTOSOption - if err := ep.GetSockOpt(&v); err != nil { - return nil, syserr.TranslateNetstackError(err) - } - - if v { - return int32(1), nil - } - return int32(0), nil - default: emitUnimplementedEventIP(t, name) } @@ -1823,16 +1808,6 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s } return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.IPv4TOSOption(v))) - case linux.IP_RECVTOS: - v, err := parseIntOrChar(optVal) - if err != nil { - return err - } - - return syserr.TranslateNetstackError(ep.SetSockOpt( - tcpip.ReceiveTOSOption(v != 0), - )) - case linux.IP_ADD_SOURCE_MEMBERSHIP, linux.IP_BIND_ADDRESS_NO_PORT, linux.IP_BLOCK_SOURCE, @@ -1853,6 +1828,7 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s linux.IP_RECVFRAGSIZE, linux.IP_RECVOPTS, linux.IP_RECVORIGDSTADDR, + linux.IP_RECVTOS, linux.IP_RECVTTL, linux.IP_RETOPTS, linux.IP_TRANSPARENT, @@ -2163,21 +2139,6 @@ func (s *SocketOperations) fillCmsgInq(cmsg *socket.ControlMessages) { cmsg.IP.Inq = int32(len(s.readView) + rcvBufUsed) } -func (s *SocketOperations) fillCmsgTOS(cmsg *socket.ControlMessages) { - if s.skType != linux.SOCK_DGRAM { - return - } - var receiveTOS tcpip.ReceiveTOSOption - if err := s.Endpoint.GetSockOpt(&receiveTOS); err != nil { - return - } - if !receiveTOS { - return - } - cmsg.IP.HasTOS = s.readCM.HasTOS - cmsg.IP.TOS = s.readCM.TOS -} - // nonBlockingRead issues a non-blocking read. // // TODO(b/78348848): Support timestamps for stream sockets. @@ -2283,7 +2244,6 @@ func (s *SocketOperations) nonBlockingRead(ctx context.Context, dst usermem.IOSe cmsg := s.controlMessages() s.fillCmsgInq(&cmsg) - s.fillCmsgTOS(&cmsg) return n, flags, addr, addrLen, cmsg, syserr.FromError(err) } diff --git a/pkg/tcpip/checker/checker.go b/pkg/tcpip/checker/checker.go index 542abc99d..2f15bf1f1 100644 --- a/pkg/tcpip/checker/checker.go +++ b/pkg/tcpip/checker/checker.go @@ -33,9 +33,6 @@ type NetworkChecker func(*testing.T, []header.Network) // TransportChecker is a function to check a property of a transport packet. type TransportChecker func(*testing.T, header.Transport) -// ControlMessagesChecker is a function to check a property of ancillary data. -type ControlMessagesChecker func(*testing.T, tcpip.ControlMessages) - // IPv4 checks the validity and properties of the given IPv4 packet. It is // expected to be used in conjunction with other network checkers for specific // properties. For example, to check the source and destination address, one @@ -161,19 +158,6 @@ func FragmentFlags(flags uint8) NetworkChecker { } } -// ReceiveTOS creates a checker that checks the TOS field in ControlMessages. -func ReceiveTOS(want uint8) ControlMessagesChecker { - return func(t *testing.T, cm tcpip.ControlMessages) { - t.Helper() - if !cm.HasTOS { - t.Fatalf("got cm.HasTOS = %t, want cm.TOS = %d", cm.HasTOS, want) - } - if got := cm.TOS; got != want { - t.Fatalf("got cm.TOS = %d, want %d", got, want) - } - } -} - // TOS creates a checker that checks the TOS field. func TOS(tos uint8, label uint32) NetworkChecker { return func(t *testing.T, h []header.Network) { diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index a4556674b..ddd014658 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -575,7 +575,7 @@ func (n *NIC) RemoveAddressRange(subnet tcpip.Subnet) { n.mu.Unlock() } -// AddressRanges returns the Subnets associated with this NIC. +// Subnets returns the Subnets associated with this NIC. func (n *NIC) AddressRanges() []tcpip.Subnet { n.mu.RLock() defer n.mu.RUnlock() diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index 251336224..7a9600679 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -829,7 +829,7 @@ func (s *Stack) CheckNIC(id tcpip.NICID) bool { return false } -// NICAddressRanges returns a map of NICIDs to their associated subnets. +// NICSubnets returns a map of NICIDs to their associated subnets. func (s *Stack) NICAddressRanges() map[tcpip.NICID][]tcpip.Subnet { s.mu.RLock() defer s.mu.RUnlock() diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 5c7b2af88..f62fd729f 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -322,7 +322,7 @@ type ControlMessages struct { HasTOS bool // TOS is the IPv4 type of service of the associated packet. - TOS uint8 + TOS int8 // HasTClass indicates whether Tclass is valid/set. HasTClass bool @@ -666,10 +666,6 @@ type IPv4TOSOption uint8 // for all subsequent outgoing IPv6 packets from the endpoint. type IPv6TrafficClassOption uint8 -// ReceiveTOSOption is used by SetSockOpt/GetSockOpt to specify if the TOS -// ancillary message is passed with incoming packets. -type ReceiveTOSOption bool - // Route is a row in the routing table. It specifies through which NIC (and // gateway) sets of packets should be routed. A row is considered viable if the // masked target address matches the destination address in the row. diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index 6d23ab5a1..5aafe2615 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -510,7 +510,7 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { } // SetSockOptInt implements tcpip.Endpoint.SetSockOptInt. -func (e *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error { +func (ep *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error { return tcpip.ErrUnknownProtocolOption } diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 269470ed4..1ac4705af 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -32,7 +32,6 @@ type udpPacket struct { senderAddress tcpip.FullAddress data buffer.VectorisedView `state:".(buffer.VectorisedView)"` timestamp int64 - tos uint8 } // EndpointState represents the state of a UDP endpoint. @@ -115,10 +114,6 @@ type endpoint struct { // applied while sending packets. Defaults to 0 as on Linux. sendTOS uint8 - // receiveTOS determines if the incoming IPv4 TOS header field is passed - // as ancillary data to ControlMessages on Read. - receiveTOS bool - // shutdownFlags represent the current shutdown state of the endpoint. shutdownFlags tcpip.ShutdownFlags @@ -249,12 +244,7 @@ func (e *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMess *addr = p.senderAddress } - return p.data.ToView(), tcpip.ControlMessages{ - HasTimestamp: true, - Timestamp: p.timestamp, - HasTOS: e.receiveTOS, - TOS: p.tos, - }, nil + return p.data.ToView(), tcpip.ControlMessages{HasTimestamp: true, Timestamp: p.timestamp}, nil } // prepareForWrite prepares the endpoint for sending data. In particular, it @@ -666,12 +656,6 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.sendTOS = uint8(v) e.mu.Unlock() return nil - - case tcpip.ReceiveTOSOption: - e.mu.Lock() - e.receiveTOS = bool(v) - e.mu.Unlock() - return nil } return nil } @@ -808,12 +792,6 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { e.mu.RUnlock() return nil - case *tcpip.ReceiveTOSOption: - e.mu.RLock() - *o = tcpip.ReceiveTOSOption(e.receiveTOS) - e.mu.RUnlock() - return nil - default: return tcpip.ErrUnknownProtocolOption } @@ -1260,13 +1238,6 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk e.rcvList.PushBack(packet) e.rcvBufSize += pkt.Data.Size() - // Save any useful information from the NetworkHeader to the packet. - switch r.NetProto { - case header.IPv4ProtocolNumber: - // This packet has already been validated before being passed up the stack. - packet.tos, _ = header.IPv4(pkt.NetworkHeader).TOS() - } - packet.timestamp = e.stack.NowNanoseconds() e.rcvMu.Unlock() diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index 43b8b35ba..7051a7a9c 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -56,7 +56,6 @@ const ( multicastAddr = "\xe8\x2b\xd3\xea" multicastV6Addr = "\xff\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00" broadcastAddr = header.IPv4Broadcast - testTOS = 0x80 // defaultMTU is the MTU, in bytes, used throughout the tests, except // where another value is explicitly used. It is chosen to match the MTU @@ -454,7 +453,6 @@ func (c *testContext) injectV4Packet(payload []byte, h *header4Tuple, valid bool ip := header.IPv4(buf) ip.Encode(&header.IPv4Fields{ IHL: header.IPv4MinimumSize, - TOS: testTOS, TotalLength: uint16(len(buf)), TTL: 65, Protocol: uint8(udp.ProtocolNumber), @@ -558,8 +556,8 @@ func TestBindToDeviceOption(t *testing.T) { // testReadInternal sends a packet of the given test flow into the stack by // injecting it into the link endpoint. It then attempts to read it from the // UDP endpoint and depending on if this was expected to succeed verifies its -// correctness including any additional checker functions provided. -func testReadInternal(c *testContext, flow testFlow, packetShouldBeDropped, expectReadError bool, checkers ...checker.ControlMessagesChecker) { +// correctness. +func testReadInternal(c *testContext, flow testFlow, packetShouldBeDropped, expectReadError bool) { c.t.Helper() payload := newPayload() @@ -574,12 +572,12 @@ func testReadInternal(c *testContext, flow testFlow, packetShouldBeDropped, expe epstats := c.ep.Stats().(*tcpip.TransportEndpointStats).Clone() var addr tcpip.FullAddress - v, cm, err := c.ep.Read(&addr) + v, _, err := c.ep.Read(&addr) if err == tcpip.ErrWouldBlock { // Wait for data to become available. select { case <-ch: - v, cm, err = c.ep.Read(&addr) + v, _, err = c.ep.Read(&addr) case <-time.After(300 * time.Millisecond): if packetShouldBeDropped { @@ -612,21 +610,15 @@ func testReadInternal(c *testContext, flow testFlow, packetShouldBeDropped, expe if !bytes.Equal(payload, v) { c.t.Fatalf("bad payload: got %x, want %x", v, payload) } - - // Run any checkers against the ControlMessages. - for _, f := range checkers { - f(c.t, cm) - } - c.checkEndpointReadStats(1, epstats, err) } // testRead sends a packet of the given test flow into the stack by injecting it // into the link endpoint. It then reads it from the UDP endpoint and verifies -// its correctness including any additional checker functions provided. -func testRead(c *testContext, flow testFlow, checkers ...checker.ControlMessagesChecker) { +// its correctness. +func testRead(c *testContext, flow testFlow) { c.t.Helper() - testReadInternal(c, flow, false /* packetShouldBeDropped */, false /* expectReadError */, checkers...) + testReadInternal(c, flow, false /* packetShouldBeDropped */, false /* expectReadError */) } // testFailingRead sends a packet of the given test flow into the stack by @@ -1294,7 +1286,7 @@ func TestTOSV4(t *testing.T) { c.createEndpointForFlow(flow) - const tos = testTOS + const tos = 0xC0 var v tcpip.IPv4TOSOption if err := c.ep.GetSockOpt(&v); err != nil { c.t.Errorf("GetSockopt failed: %s", err) @@ -1329,7 +1321,7 @@ func TestTOSV6(t *testing.T) { c.createEndpointForFlow(flow) - const tos = testTOS + const tos = 0xC0 var v tcpip.IPv6TrafficClassOption if err := c.ep.GetSockOpt(&v); err != nil { c.t.Errorf("GetSockopt failed: %s", err) @@ -1356,49 +1348,6 @@ func TestTOSV6(t *testing.T) { } } -func TestReceiveTOSV4(t *testing.T) { - for _, flow := range []testFlow{unicastV4, broadcast} { - t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpointForFlow(flow) - - // Verify that setting and reading the option works. - const recvTos = true - var v tcpip.ReceiveTOSOption - if err := c.ep.GetSockOpt(&v); err != nil { - c.t.Errorf("GetSockopt failed: %s", err) - } - // Test for expected default value. - if v != false { - c.t.Errorf("got GetSockOpt(...) = %t, want = %t", v, false) - } - - if err := c.ep.SetSockOpt(tcpip.ReceiveTOSOption(recvTos)); err != nil { - c.t.Errorf("SetSockOpt(%#v) failed: %s", tcpip.ReceiveTOSOption(recvTos), err) - } - - if err := c.ep.GetSockOpt(&v); err != nil { - c.t.Errorf("GetSockopt failed: %s", err) - } - - if want := tcpip.ReceiveTOSOption(recvTos); v != want { - c.t.Errorf("got GetSockOpt(...) = %t, want = %t", v, want) - } - - // Bind to wildcard. - if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %s", err) - } - - // Verify that the correct received TOS is actually handed through as - // ancillary data to the ControlMessages struct. - testRead(c, flow, checker.ReceiveTOS(testTOS)) - }) - } -} - func TestMulticastInterfaceOption(t *testing.T) { for _, flow := range []testFlow{multicastV4, multicastV4in6, multicastV6, multicastV6Only} { t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { diff --git a/test/syscalls/linux/socket_ip_udp_generic.cc b/test/syscalls/linux/socket_ip_udp_generic.cc index 53290bed7..66eb68857 100644 --- a/test/syscalls/linux/socket_ip_udp_generic.cc +++ b/test/syscalls/linux/socket_ip_udp_generic.cc @@ -209,46 +209,6 @@ TEST_P(UDPSocketPairTest, SetMulticastLoopChar) { EXPECT_EQ(get, kSockOptOn); } -// Ensure that Receiving TOS is off by default. -TEST_P(UDPSocketPairTest, RecvTosDefault) { - auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); - - int get = -1; - socklen_t get_len = sizeof(get); - ASSERT_THAT( - getsockopt(sockets->first_fd(), IPPROTO_IP, IP_RECVTOS, &get, &get_len), - SyscallSucceedsWithValue(0)); - EXPECT_EQ(get_len, sizeof(get)); - EXPECT_EQ(get, kSockOptOff); -} - -// Test that setting and getting IP_RECVTOS works as expected. -TEST_P(UDPSocketPairTest, SetRecvTos) { - auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); - - ASSERT_THAT(setsockopt(sockets->first_fd(), IPPROTO_IP, IP_RECVTOS, - &kSockOptOff, sizeof(kSockOptOff)), - SyscallSucceeds()); - - int get = -1; - socklen_t get_len = sizeof(get); - ASSERT_THAT( - getsockopt(sockets->first_fd(), IPPROTO_IP, IP_RECVTOS, &get, &get_len), - SyscallSucceedsWithValue(0)); - EXPECT_EQ(get_len, sizeof(get)); - EXPECT_EQ(get, kSockOptOff); - - ASSERT_THAT(setsockopt(sockets->first_fd(), IPPROTO_IP, IP_RECVTOS, - &kSockOptOn, sizeof(kSockOptOn)), - SyscallSucceeds()); - - ASSERT_THAT( - getsockopt(sockets->first_fd(), IPPROTO_IP, IP_RECVTOS, &get, &get_len), - SyscallSucceedsWithValue(0)); - EXPECT_EQ(get_len, sizeof(get)); - EXPECT_EQ(get, kSockOptOn); -} - TEST_P(UDPSocketPairTest, ReuseAddrDefault) { auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); diff --git a/test/syscalls/linux/udp_socket_test_cases.cc b/test/syscalls/linux/udp_socket_test_cases.cc index 68e0a8109..dc35c2f50 100644 --- a/test/syscalls/linux/udp_socket_test_cases.cc +++ b/test/syscalls/linux/udp_socket_test_cases.cc @@ -1349,9 +1349,8 @@ TEST_P(UdpSocketTest, TimestampIoctlPersistence) { // outgoing packets, and that a receiving socket with IP_RECVTOS or // IPV6_RECVTCLASS will create the corresponding control message. TEST_P(UdpSocketTest, SetAndReceiveTOS) { - // TODO(b/68320120): IPV6_RECVTCLASS not supported for netstack. - SKIP_IF((GetParam() != AddressFamily::kIpv4) && IsRunningOnGvisor() && - !IsRunningWithHostinet()); + // TODO(b/68320120): IP_RECVTOS/IPV6_RECVTCLASS not supported for netstack. + SKIP_IF(IsRunningOnGvisor() && !IsRunningWithHostinet()); ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); ASSERT_THAT(connect(t_, addr_[0], addrlen_), SyscallSucceeds()); @@ -1422,8 +1421,7 @@ TEST_P(UdpSocketTest, SetAndReceiveTOS) { // TOS byte on outgoing packets, and that a receiving socket with IP_RECVTOS or // IPV6_RECVTCLASS will create the corresponding control message. TEST_P(UdpSocketTest, SendAndReceiveTOS) { - // TODO(b/68320120): IPV6_RECVTCLASS not supported for netstack. - // TODO(b/146661005): Setting TOS via cmsg not supported for netstack. + // TODO(b/68320120): IP_RECVTOS/IPV6_RECVTCLASS not supported for netstack. SKIP_IF(IsRunningOnGvisor() && !IsRunningWithHostinet()); ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); ASSERT_THAT(connect(t_, addr_[0], addrlen_), SyscallSucceeds()); -- cgit v1.2.3 From 3c125eb21946e1f6bf8f22f4169baafb7f07bf60 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Thu, 26 Dec 2019 14:42:19 -0800 Subject: Initial procfs implementation in VFSv2 Updates #1195 PiperOrigin-RevId: 287227722 --- pkg/sentry/fsimpl/kernfs/BUILD | 1 + pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go | 20 +- pkg/sentry/fsimpl/kernfs/fd_impl_util.go | 5 +- pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 20 ++ pkg/sentry/fsimpl/kernfs/kernfs.go | 9 + pkg/sentry/fsimpl/kernfs/kernfs_test.go | 2 +- pkg/sentry/fsimpl/kernfs/symlink.go | 45 +++ pkg/sentry/fsimpl/proc/BUILD | 33 +- pkg/sentry/fsimpl/proc/boot_test.go | 149 +++++++++ pkg/sentry/fsimpl/proc/filesystem.go | 69 +++++ pkg/sentry/fsimpl/proc/filesystems.go | 25 -- pkg/sentry/fsimpl/proc/loadavg.go | 8 +- pkg/sentry/fsimpl/proc/meminfo.go | 6 +- pkg/sentry/fsimpl/proc/proc.go | 16 - pkg/sentry/fsimpl/proc/stat.go | 6 +- pkg/sentry/fsimpl/proc/task.go | 341 ++++++++------------ pkg/sentry/fsimpl/proc/task_files.go | 272 ++++++++++++++++ pkg/sentry/fsimpl/proc/tasks.go | 162 ++++++++++ pkg/sentry/fsimpl/proc/tasks_files.go | 92 ++++++ pkg/sentry/fsimpl/proc/tasks_test.go | 412 +++++++++++++++++++++++++ pkg/sentry/fsimpl/proc/version.go | 6 +- pkg/sentry/kernel/kernel.go | 7 +- pkg/sentry/kernel/task_clone.go | 2 +- pkg/sentry/kernel/thread_group.go | 8 +- pkg/sentry/vfs/file_description_impl_util.go | 11 + 25 files changed, 1454 insertions(+), 273 deletions(-) create mode 100644 pkg/sentry/fsimpl/kernfs/symlink.go create mode 100644 pkg/sentry/fsimpl/proc/boot_test.go create mode 100644 pkg/sentry/fsimpl/proc/filesystem.go delete mode 100644 pkg/sentry/fsimpl/proc/filesystems.go delete mode 100644 pkg/sentry/fsimpl/proc/proc.go create mode 100644 pkg/sentry/fsimpl/proc/task_files.go create mode 100644 pkg/sentry/fsimpl/proc/tasks.go create mode 100644 pkg/sentry/fsimpl/proc/tasks_files.go create mode 100644 pkg/sentry/fsimpl/proc/tasks_test.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/kernfs/BUILD b/pkg/sentry/fsimpl/kernfs/BUILD index 59f7f39e2..39c03ee9d 100644 --- a/pkg/sentry/fsimpl/kernfs/BUILD +++ b/pkg/sentry/fsimpl/kernfs/BUILD @@ -25,6 +25,7 @@ go_library( "inode_impl_util.go", "kernfs.go", "slot_list.go", + "symlink.go", ], importpath = "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs", visibility = ["//pkg/sentry:internal"], diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go index 51102ce48..c5fe65722 100644 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -15,6 +15,8 @@ package kernfs import ( + "fmt" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" @@ -26,7 +28,10 @@ import ( // DynamicBytesFile implements kernfs.Inode and represents a read-only // file whose contents are backed by a vfs.DynamicBytesSource. // -// Must be initialized with Init before first use. +// Must be instantiated with NewDynamicBytesFile or initialized with Init +// before first use. +// +// +stateify savable type DynamicBytesFile struct { InodeAttrs InodeNoopRefCount @@ -36,9 +41,14 @@ type DynamicBytesFile struct { data vfs.DynamicBytesSource } -// Init intializes a dynamic bytes file. -func (f *DynamicBytesFile) Init(creds *auth.Credentials, ino uint64, data vfs.DynamicBytesSource) { - f.InodeAttrs.Init(creds, ino, linux.ModeRegular|0444) +var _ Inode = (*DynamicBytesFile)(nil) + +// Init initializes a dynamic bytes file. +func (f *DynamicBytesFile) Init(creds *auth.Credentials, ino uint64, data vfs.DynamicBytesSource, perm linux.FileMode) { + if perm&^linux.PermissionsMask != 0 { + panic(fmt.Sprintf("Only permission mask must be set: %x", perm&linux.PermissionsMask)) + } + f.InodeAttrs.Init(creds, ino, linux.ModeRegular|perm) f.data = data } @@ -59,6 +69,8 @@ func (f *DynamicBytesFile) SetStat(*vfs.Filesystem, vfs.SetStatOptions) error { // DynamicBytesFile. // // Must be initialized with Init before first use. +// +// +stateify savable type DynamicBytesFD struct { vfs.FileDescriptionDefaultImpl vfs.DynamicBytesFileDescriptionImpl diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index bd402330f..77975583b 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -154,7 +154,10 @@ func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirent fd.off++ } - return nil + var err error + relOffset := fd.off - int64(len(fd.children.set)) - 2 + fd.off, err = fd.inode().IterDirents(ctx, cb, fd.off, relOffset) + return err } // Seek implements vfs.FileDecriptionImpl.Seek. diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 7b45b702a..752e0f659 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -139,6 +139,11 @@ func (*InodeNotDirectory) Lookup(ctx context.Context, name string) (*vfs.Dentry, panic("Lookup called on non-directory inode") } +// IterDirents implements Inode.IterDirents. +func (*InodeNotDirectory) IterDirents(ctx context.Context, callback vfs.IterDirentsCallback, offset, relOffset int64) (newOffset int64, err error) { + panic("IterDirents called on non-directory inode") +} + // Valid implements Inode.Valid. func (*InodeNotDirectory) Valid(context.Context) bool { return true @@ -156,6 +161,11 @@ func (*InodeNoDynamicLookup) Lookup(ctx context.Context, name string) (*vfs.Dent return nil, syserror.ENOENT } +// IterDirents implements Inode.IterDirents. +func (*InodeNoDynamicLookup) IterDirents(ctx context.Context, callback vfs.IterDirentsCallback, offset, relOffset int64) (int64, error) { + return offset, nil +} + // Valid implements Inode.Valid. func (*InodeNoDynamicLookup) Valid(ctx context.Context) bool { return true @@ -490,3 +500,13 @@ func (o *OrderedChildren) nthLocked(i int64) *slot { } return nil } + +// InodeSymlink partially implements Inode interface for symlinks. +type InodeSymlink struct { + InodeNotDirectory +} + +// Open implements Inode.Open. +func (InodeSymlink) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { + return nil, syserror.ELOOP +} diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index ac802218d..d69b299ae 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -404,6 +404,15 @@ type inodeDynamicLookup interface { // Valid should return true if this inode is still valid, or needs to // be resolved again by a call to Lookup. Valid(ctx context.Context) bool + + // IterDirents is used to iterate over dynamically created entries. It invokes + // cb on each entry in the directory represented by the FileDescription. + // 'offset' is the offset for the entire IterDirents call, which may include + // results from the caller. 'relOffset' is the offset inside the entries + // returned by this IterDirents invocation. In other words, + // 'offset+relOffset+1' is the value that should be set in vfs.Dirent.NextOff, + // while 'relOffset' is the place where iteration should start from. + IterDirents(ctx context.Context, callback vfs.IterDirentsCallback, offset, relOffset int64) (newOffset int64, err error) } type inodeSymlink interface { diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index 73b6e43b5..3db12caa0 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -133,7 +133,7 @@ type file struct { func (fs *filesystem) newFile(creds *auth.Credentials, content string) *kernfs.Dentry { f := &file{} f.content = content - f.DynamicBytesFile.Init(creds, fs.NextIno(), f) + f.DynamicBytesFile.Init(creds, fs.NextIno(), f, 0777) d := &kernfs.Dentry{} d.Init(f) diff --git a/pkg/sentry/fsimpl/kernfs/symlink.go b/pkg/sentry/fsimpl/kernfs/symlink.go new file mode 100644 index 000000000..068063f4e --- /dev/null +++ b/pkg/sentry/fsimpl/kernfs/symlink.go @@ -0,0 +1,45 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package kernfs + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" +) + +type staticSymlink struct { + InodeAttrs + InodeNoopRefCount + InodeSymlink + + target string +} + +var _ Inode = (*staticSymlink)(nil) + +// NewStaticSymlink creates a new symlink file pointing to 'target'. +func NewStaticSymlink(creds *auth.Credentials, ino uint64, perm linux.FileMode, target string) *Dentry { + inode := &staticSymlink{target: target} + inode.Init(creds, ino, linux.ModeSymlink|perm) + + d := &Dentry{} + d.Init(inode) + return d +} + +func (s *staticSymlink) Readlink(_ context.Context) (string, error) { + return s.target, nil +} diff --git a/pkg/sentry/fsimpl/proc/BUILD b/pkg/sentry/fsimpl/proc/BUILD index ade6ac946..1f44b3217 100644 --- a/pkg/sentry/fsimpl/proc/BUILD +++ b/pkg/sentry/fsimpl/proc/BUILD @@ -6,15 +6,17 @@ package(licenses = ["notice"]) go_library( name = "proc", srcs = [ - "filesystems.go", + "filesystem.go", "loadavg.go", "meminfo.go", "mounts.go", "net.go", - "proc.go", "stat.go", "sys.go", "task.go", + "task_files.go", + "tasks.go", + "tasks_files.go", "version.go", ], importpath = "gvisor.dev/gvisor/pkg/sentry/fsimpl/proc", @@ -24,8 +26,10 @@ go_library( "//pkg/log", "//pkg/sentry/context", "//pkg/sentry/fs", + "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/inet", "//pkg/sentry/kernel", + "//pkg/sentry/kernel/auth", "//pkg/sentry/limits", "//pkg/sentry/mm", "//pkg/sentry/socket", @@ -34,17 +38,40 @@ go_library( "//pkg/sentry/usage", "//pkg/sentry/usermem", "//pkg/sentry/vfs", + "//pkg/syserror", ], ) go_test( name = "proc_test", size = "small", - srcs = ["net_test.go"], + srcs = [ + "boot_test.go", + "net_test.go", + "tasks_test.go", + ], embed = [":proc"], deps = [ "//pkg/abi/linux", + "//pkg/cpuid", + "//pkg/fspath", + "//pkg/memutil", + "//pkg/sentry/context", "//pkg/sentry/context/contexttest", + "//pkg/sentry/fs", "//pkg/sentry/inet", + "//pkg/sentry/kernel", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/kernel/sched", + "//pkg/sentry/limits", + "//pkg/sentry/loader", + "//pkg/sentry/pgalloc", + "//pkg/sentry/platform", + "//pkg/sentry/platform/kvm", + "//pkg/sentry/platform/ptrace", + "//pkg/sentry/time", + "//pkg/sentry/usermem", + "//pkg/sentry/vfs", + "//pkg/syserror", ], ) diff --git a/pkg/sentry/fsimpl/proc/boot_test.go b/pkg/sentry/fsimpl/proc/boot_test.go new file mode 100644 index 000000000..84a93ee56 --- /dev/null +++ b/pkg/sentry/fsimpl/proc/boot_test.go @@ -0,0 +1,149 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package proc + +import ( + "flag" + "fmt" + "os" + "runtime" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/cpuid" + "gvisor.dev/gvisor/pkg/memutil" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/kernel/sched" + "gvisor.dev/gvisor/pkg/sentry/limits" + "gvisor.dev/gvisor/pkg/sentry/loader" + "gvisor.dev/gvisor/pkg/sentry/pgalloc" + "gvisor.dev/gvisor/pkg/sentry/platform" + "gvisor.dev/gvisor/pkg/sentry/time" + + // Platforms are plugable. + _ "gvisor.dev/gvisor/pkg/sentry/platform/kvm" + _ "gvisor.dev/gvisor/pkg/sentry/platform/ptrace" +) + +var ( + platformFlag = flag.String("platform", "ptrace", "specify which platform to use") +) + +// boot initializes a new bare bones kernel for test. +func boot() (*kernel.Kernel, error) { + platformCtr, err := platform.Lookup(*platformFlag) + if err != nil { + return nil, fmt.Errorf("platform not found: %v", err) + } + deviceFile, err := platformCtr.OpenDevice() + if err != nil { + return nil, fmt.Errorf("creating platform: %v", err) + } + plat, err := platformCtr.New(deviceFile) + if err != nil { + return nil, fmt.Errorf("creating platform: %v", err) + } + + k := &kernel.Kernel{ + Platform: plat, + } + + mf, err := createMemoryFile() + if err != nil { + return nil, err + } + k.SetMemoryFile(mf) + + // Pass k as the platform since it is savable, unlike the actual platform. + vdso, err := loader.PrepareVDSO(nil, k) + if err != nil { + return nil, fmt.Errorf("creating vdso: %v", err) + } + + // Create timekeeper. + tk, err := kernel.NewTimekeeper(k, vdso.ParamPage.FileRange()) + if err != nil { + return nil, fmt.Errorf("creating timekeeper: %v", err) + } + tk.SetClocks(time.NewCalibratedClocks()) + + creds := auth.NewRootCredentials(auth.NewRootUserNamespace()) + + // Initiate the Kernel object, which is required by the Context passed + // to createVFS in order to mount (among other things) procfs. + if err = k.Init(kernel.InitKernelArgs{ + ApplicationCores: uint(runtime.GOMAXPROCS(-1)), + FeatureSet: cpuid.HostFeatureSet(), + Timekeeper: tk, + RootUserNamespace: creds.UserNamespace, + Vdso: vdso, + RootUTSNamespace: kernel.NewUTSNamespace("hostname", "domain", creds.UserNamespace), + RootIPCNamespace: kernel.NewIPCNamespace(creds.UserNamespace), + RootAbstractSocketNamespace: kernel.NewAbstractSocketNamespace(), + PIDNamespace: kernel.NewRootPIDNamespace(creds.UserNamespace), + }); err != nil { + return nil, fmt.Errorf("initializing kernel: %v", err) + } + + ctx := k.SupervisorContext() + + // Create mount namespace without root as it's the minimum required to create + // the global thread group. + mntns, err := fs.NewMountNamespace(ctx, nil) + if err != nil { + return nil, err + } + ls, err := limits.NewLinuxLimitSet() + if err != nil { + return nil, err + } + tg := k.NewThreadGroup(mntns, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, ls) + k.TestOnly_SetGlobalInit(tg) + + return k, nil +} + +// createTask creates a new bare bones task for tests. +func createTask(ctx context.Context, name string, tc *kernel.ThreadGroup) (*kernel.Task, error) { + k := kernel.KernelFromContext(ctx) + config := &kernel.TaskConfig{ + Kernel: k, + ThreadGroup: tc, + TaskContext: &kernel.TaskContext{Name: name}, + Credentials: auth.CredentialsFromContext(ctx), + AllowedCPUMask: sched.NewFullCPUSet(k.ApplicationCores()), + UTSNamespace: kernel.UTSNamespaceFromContext(ctx), + IPCNamespace: kernel.IPCNamespaceFromContext(ctx), + AbstractSocketNamespace: kernel.NewAbstractSocketNamespace(), + } + return k.TaskSet().NewTask(config) +} + +func createMemoryFile() (*pgalloc.MemoryFile, error) { + const memfileName = "test-memory" + memfd, err := memutil.CreateMemFD(memfileName, 0) + if err != nil { + return nil, fmt.Errorf("error creating memfd: %v", err) + } + memfile := os.NewFile(uintptr(memfd), memfileName) + mf, err := pgalloc.NewMemoryFile(memfile, pgalloc.MemoryFileOpts{}) + if err != nil { + memfile.Close() + return nil, fmt.Errorf("error creating pgalloc.MemoryFile: %v", err) + } + return mf, nil +} diff --git a/pkg/sentry/fsimpl/proc/filesystem.go b/pkg/sentry/fsimpl/proc/filesystem.go new file mode 100644 index 000000000..d09182c77 --- /dev/null +++ b/pkg/sentry/fsimpl/proc/filesystem.go @@ -0,0 +1,69 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package proc implements a partial in-memory file system for procfs. +package proc + +import ( + "fmt" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" +) + +// procFSType is the factory class for procfs. +// +// +stateify savable +type procFSType struct{} + +var _ vfs.FilesystemType = (*procFSType)(nil) + +// GetFilesystem implements vfs.FilesystemType. +func (ft *procFSType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { + k := kernel.KernelFromContext(ctx) + if k == nil { + return nil, nil, fmt.Errorf("procfs requires a kernel") + } + pidns := kernel.PIDNamespaceFromContext(ctx) + if pidns == nil { + return nil, nil, fmt.Errorf("procfs requires a PID namespace") + } + + procfs := &kernfs.Filesystem{} + procfs.VFSFilesystem().Init(vfsObj, procfs) + + _, dentry := newTasksInode(procfs, k, pidns) + return procfs.VFSFilesystem(), dentry.VFSDentry(), nil +} + +// dynamicInode is an overfitted interface for common Inodes with +// dynamicByteSource types used in procfs. +type dynamicInode interface { + kernfs.Inode + vfs.DynamicBytesSource + + Init(creds *auth.Credentials, ino uint64, data vfs.DynamicBytesSource, perm linux.FileMode) +} + +func newDentry(creds *auth.Credentials, ino uint64, perm linux.FileMode, inode dynamicInode) *kernfs.Dentry { + inode.Init(creds, ino, inode, perm) + + d := &kernfs.Dentry{} + d.Init(inode) + return d +} diff --git a/pkg/sentry/fsimpl/proc/filesystems.go b/pkg/sentry/fsimpl/proc/filesystems.go deleted file mode 100644 index 0e016bca5..000000000 --- a/pkg/sentry/fsimpl/proc/filesystems.go +++ /dev/null @@ -1,25 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -// filesystemsData implements vfs.DynamicBytesSource for /proc/filesystems. -// -// +stateify savable -type filesystemsData struct{} - -// TODO(gvisor.dev/issue/1195): Implement vfs.DynamicBytesSource.Generate for -// filesystemsData. We would need to retrive filesystem names from -// vfs.VirtualFilesystem. Also needs vfs replacement for -// fs.Filesystem.AllowUserList() and fs.FilesystemRequiresDev. diff --git a/pkg/sentry/fsimpl/proc/loadavg.go b/pkg/sentry/fsimpl/proc/loadavg.go index 9135afef1..5351d86e8 100644 --- a/pkg/sentry/fsimpl/proc/loadavg.go +++ b/pkg/sentry/fsimpl/proc/loadavg.go @@ -19,15 +19,17 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" ) // loadavgData backs /proc/loadavg. // // +stateify savable -type loadavgData struct{} +type loadavgData struct { + kernfs.DynamicBytesFile +} -var _ vfs.DynamicBytesSource = (*loadavgData)(nil) +var _ dynamicInode = (*loadavgData)(nil) // Generate implements vfs.DynamicBytesSource.Generate. func (d *loadavgData) Generate(ctx context.Context, buf *bytes.Buffer) error { diff --git a/pkg/sentry/fsimpl/proc/meminfo.go b/pkg/sentry/fsimpl/proc/meminfo.go index 9a827cd66..cbdd4f3fc 100644 --- a/pkg/sentry/fsimpl/proc/meminfo.go +++ b/pkg/sentry/fsimpl/proc/meminfo.go @@ -19,21 +19,23 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/sentry/usermem" - "gvisor.dev/gvisor/pkg/sentry/vfs" ) // meminfoData implements vfs.DynamicBytesSource for /proc/meminfo. // // +stateify savable type meminfoData struct { + kernfs.DynamicBytesFile + // k is the owning Kernel. k *kernel.Kernel } -var _ vfs.DynamicBytesSource = (*meminfoData)(nil) +var _ dynamicInode = (*meminfoData)(nil) // Generate implements vfs.DynamicBytesSource.Generate. func (d *meminfoData) Generate(ctx context.Context, buf *bytes.Buffer) error { diff --git a/pkg/sentry/fsimpl/proc/proc.go b/pkg/sentry/fsimpl/proc/proc.go deleted file mode 100644 index 31dec36de..000000000 --- a/pkg/sentry/fsimpl/proc/proc.go +++ /dev/null @@ -1,16 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package proc implements a partial in-memory file system for procfs. -package proc diff --git a/pkg/sentry/fsimpl/proc/stat.go b/pkg/sentry/fsimpl/proc/stat.go index 720db3828..50894a534 100644 --- a/pkg/sentry/fsimpl/proc/stat.go +++ b/pkg/sentry/fsimpl/proc/stat.go @@ -20,8 +20,8 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/vfs" ) // cpuStats contains the breakdown of CPU time for /proc/stat. @@ -66,11 +66,13 @@ func (c cpuStats) String() string { // // +stateify savable type statData struct { + kernfs.DynamicBytesFile + // k is the owning Kernel. k *kernel.Kernel } -var _ vfs.DynamicBytesSource = (*statData)(nil) +var _ dynamicInode = (*statData)(nil) // Generate implements vfs.DynamicBytesSource.Generate. func (s *statData) Generate(ctx context.Context, buf *bytes.Buffer) error { diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index 0d87be52b..11a64c777 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -15,251 +15,176 @@ package proc import ( - "bytes" - "fmt" - "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/limits" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/mm" - "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" ) -// mapsCommon is embedded by mapsData and smapsData. -type mapsCommon struct { - t *kernel.Task -} - -// mm gets the kernel task's MemoryManager. No additional reference is taken on -// mm here. This is safe because MemoryManager.destroy is required to leave the -// MemoryManager in a state where it's still usable as a DynamicBytesSource. -func (md *mapsCommon) mm() *mm.MemoryManager { - var tmm *mm.MemoryManager - md.t.WithMuLocked(func(t *kernel.Task) { - if mm := t.MemoryManager(); mm != nil { - tmm = mm - } - }) - return tmm -} - -// mapsData implements vfs.DynamicBytesSource for /proc/[pid]/maps. +// taskInode represents the inode for /proc/PID/ directory. // // +stateify savable -type mapsData struct { - mapsCommon +type taskInode struct { + kernfs.InodeNotSymlink + kernfs.InodeDirectoryNoNewChildren + kernfs.InodeNoDynamicLookup + kernfs.InodeAttrs + kernfs.OrderedChildren + + task *kernel.Task } -var _ vfs.DynamicBytesSource = (*mapsData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (md *mapsData) Generate(ctx context.Context, buf *bytes.Buffer) error { - if mm := md.mm(); mm != nil { - mm.ReadMapsDataInto(ctx, buf) +var _ kernfs.Inode = (*taskInode)(nil) + +func newTaskInode(inoGen InoGenerator, task *kernel.Task, pidns *kernel.PIDNamespace, isThreadGroup bool) *kernfs.Dentry { + contents := map[string]*kernfs.Dentry{ + //"auxv": newAuxvec(t, msrc), + //"cmdline": newExecArgInode(t, msrc, cmdlineExecArg), + //"comm": newComm(t, msrc), + //"environ": newExecArgInode(t, msrc, environExecArg), + //"exe": newExe(t, msrc), + //"fd": newFdDir(t, msrc), + //"fdinfo": newFdInfoDir(t, msrc), + //"gid_map": newGIDMap(t, msrc), + "io": newTaskOwnedFile(task, inoGen.NextIno(), defaultPermission, newIO(task, isThreadGroup)), + "maps": newTaskOwnedFile(task, inoGen.NextIno(), defaultPermission, &mapsData{task: task}), + //"mountinfo": seqfile.NewSeqFileInode(t, &mountInfoFile{t: t}, msrc), + //"mounts": seqfile.NewSeqFileInode(t, &mountsFile{t: t}, msrc), + //"ns": newNamespaceDir(t, msrc), + "smaps": newTaskOwnedFile(task, inoGen.NextIno(), defaultPermission, &smapsData{task: task}), + "stat": newTaskOwnedFile(task, inoGen.NextIno(), defaultPermission, &taskStatData{t: task, pidns: pidns, tgstats: isThreadGroup}), + "statm": newTaskOwnedFile(task, inoGen.NextIno(), defaultPermission, &statmData{t: task}), + "status": newTaskOwnedFile(task, inoGen.NextIno(), defaultPermission, &statusData{t: task, pidns: pidns}), + //"uid_map": newUIDMap(t, msrc), } - return nil -} - -// smapsData implements vfs.DynamicBytesSource for /proc/[pid]/smaps. -// -// +stateify savable -type smapsData struct { - mapsCommon -} - -var _ vfs.DynamicBytesSource = (*smapsData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (sd *smapsData) Generate(ctx context.Context, buf *bytes.Buffer) error { - if mm := sd.mm(); mm != nil { - mm.ReadSmapsDataInto(ctx, buf) + if isThreadGroup { + //contents["task"] = p.newSubtasks(t, msrc) } - return nil -} - -// +stateify savable -type taskStatData struct { - t *kernel.Task + //if len(p.cgroupControllers) > 0 { + // contents["cgroup"] = newCGroupInode(t, msrc, p.cgroupControllers) + //} - // If tgstats is true, accumulate fault stats (not implemented) and CPU - // time across all tasks in t's thread group. - tgstats bool + taskInode := &taskInode{task: task} + // Note: credentials are overridden by taskOwnedInode. + taskInode.InodeAttrs.Init(task.Credentials(), inoGen.NextIno(), linux.ModeDirectory|0555) - // pidns is the PID namespace associated with the proc filesystem that - // includes the file using this statData. - pidns *kernel.PIDNamespace -} - -var _ vfs.DynamicBytesSource = (*taskStatData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (s *taskStatData) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "%d ", s.pidns.IDOfTask(s.t)) - fmt.Fprintf(buf, "(%s) ", s.t.Name()) - fmt.Fprintf(buf, "%c ", s.t.StateStatus()[0]) - ppid := kernel.ThreadID(0) - if parent := s.t.Parent(); parent != nil { - ppid = s.pidns.IDOfThreadGroup(parent.ThreadGroup()) - } - fmt.Fprintf(buf, "%d ", ppid) - fmt.Fprintf(buf, "%d ", s.pidns.IDOfProcessGroup(s.t.ThreadGroup().ProcessGroup())) - fmt.Fprintf(buf, "%d ", s.pidns.IDOfSession(s.t.ThreadGroup().Session())) - fmt.Fprintf(buf, "0 0 " /* tty_nr tpgid */) - fmt.Fprintf(buf, "0 " /* flags */) - fmt.Fprintf(buf, "0 0 0 0 " /* minflt cminflt majflt cmajflt */) - var cputime usage.CPUStats - if s.tgstats { - cputime = s.t.ThreadGroup().CPUStats() - } else { - cputime = s.t.CPUStats() - } - fmt.Fprintf(buf, "%d %d ", linux.ClockTFromDuration(cputime.UserTime), linux.ClockTFromDuration(cputime.SysTime)) - cputime = s.t.ThreadGroup().JoinedChildCPUStats() - fmt.Fprintf(buf, "%d %d ", linux.ClockTFromDuration(cputime.UserTime), linux.ClockTFromDuration(cputime.SysTime)) - fmt.Fprintf(buf, "%d %d ", s.t.Priority(), s.t.Niceness()) - fmt.Fprintf(buf, "%d ", s.t.ThreadGroup().Count()) + inode := &taskOwnedInode{Inode: taskInode, owner: task} + dentry := &kernfs.Dentry{} + dentry.Init(inode) - // itrealvalue. Since kernel 2.6.17, this field is no longer - // maintained, and is hard coded as 0. - fmt.Fprintf(buf, "0 ") + taskInode.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) + links := taskInode.OrderedChildren.Populate(dentry, contents) + taskInode.IncLinks(links) - // Start time is relative to boot time, expressed in clock ticks. - fmt.Fprintf(buf, "%d ", linux.ClockTFromDuration(s.t.StartTime().Sub(s.t.Kernel().Timekeeper().BootTime()))) + return dentry +} - var vss, rss uint64 - s.t.WithMuLocked(func(t *kernel.Task) { - if mm := t.MemoryManager(); mm != nil { - vss = mm.VirtualMemorySize() - rss = mm.ResidentSetSize() - } - }) - fmt.Fprintf(buf, "%d %d ", vss, rss/usermem.PageSize) +// Valid implements kernfs.inodeDynamicLookup. This inode remains valid as long +// as the task is still running. When it's dead, another tasks with the same +// PID could replace it. +func (i *taskInode) Valid(ctx context.Context) bool { + return i.task.ExitState() != kernel.TaskExitDead +} - // rsslim. - fmt.Fprintf(buf, "%d ", s.t.ThreadGroup().Limits().Get(limits.Rss).Cur) +// Open implements kernfs.Inode. +func (i *taskInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { + fd := &kernfs.GenericDirectoryFD{} + fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, flags) + return fd.VFSFileDescription(), nil +} - fmt.Fprintf(buf, "0 0 0 0 0 " /* startcode endcode startstack kstkesp kstkeip */) - fmt.Fprintf(buf, "0 0 0 0 0 " /* signal blocked sigignore sigcatch wchan */) - fmt.Fprintf(buf, "0 0 " /* nswap cnswap */) - terminationSignal := linux.Signal(0) - if s.t == s.t.ThreadGroup().Leader() { - terminationSignal = s.t.ThreadGroup().TerminationSignal() +// SetStat implements kernfs.Inode. +func (i *taskInode) SetStat(_ *vfs.Filesystem, opts vfs.SetStatOptions) error { + stat := opts.Stat + if stat.Mask&linux.STATX_MODE != 0 { + return syserror.EPERM } - fmt.Fprintf(buf, "%d ", terminationSignal) - fmt.Fprintf(buf, "0 0 0 " /* processor rt_priority policy */) - fmt.Fprintf(buf, "0 0 0 " /* delayacct_blkio_ticks guest_time cguest_time */) - fmt.Fprintf(buf, "0 0 0 0 0 0 0 " /* start_data end_data start_brk arg_start arg_end env_start env_end */) - fmt.Fprintf(buf, "0\n" /* exit_code */) - return nil } -// statmData implements vfs.DynamicBytesSource for /proc/[pid]/statm. -// -// +stateify savable -type statmData struct { - t *kernel.Task +// taskOwnedInode implements kernfs.Inode and overrides inode owner with task +// effective user and group. +type taskOwnedInode struct { + kernfs.Inode + + // owner is the task that owns this inode. + owner *kernel.Task } -var _ vfs.DynamicBytesSource = (*statmData)(nil) +var _ kernfs.Inode = (*taskOwnedInode)(nil) -// Generate implements vfs.DynamicBytesSource.Generate. -func (s *statmData) Generate(ctx context.Context, buf *bytes.Buffer) error { - var vss, rss uint64 - s.t.WithMuLocked(func(t *kernel.Task) { - if mm := t.MemoryManager(); mm != nil { - vss = mm.VirtualMemorySize() - rss = mm.ResidentSetSize() - } - }) +func newTaskOwnedFile(task *kernel.Task, ino uint64, perm linux.FileMode, inode dynamicInode) *kernfs.Dentry { + // Note: credentials are overridden by taskOwnedInode. + inode.Init(task.Credentials(), ino, inode, perm) - fmt.Fprintf(buf, "%d %d 0 0 0 0 0\n", vss/usermem.PageSize, rss/usermem.PageSize) - return nil + taskInode := &taskOwnedInode{Inode: inode, owner: task} + d := &kernfs.Dentry{} + d.Init(taskInode) + return d } -// statusData implements vfs.DynamicBytesSource for /proc/[pid]/status. -// -// +stateify savable -type statusData struct { - t *kernel.Task - pidns *kernel.PIDNamespace +// Stat implements kernfs.Inode. +func (i *taskOwnedInode) Stat(fs *vfs.Filesystem) linux.Statx { + stat := i.Inode.Stat(fs) + uid, gid := i.getOwner(linux.FileMode(stat.Mode)) + stat.UID = uint32(uid) + stat.GID = uint32(gid) + return stat } -var _ vfs.DynamicBytesSource = (*statusData)(nil) +// CheckPermissions implements kernfs.Inode. +func (i *taskOwnedInode) CheckPermissions(creds *auth.Credentials, ats vfs.AccessTypes) error { + mode := i.Mode() + uid, gid := i.getOwner(mode) + return vfs.GenericCheckPermissions( + creds, + ats, + mode.FileType() == linux.ModeDirectory, + uint16(mode), + uid, + gid, + ) +} -// Generate implements vfs.DynamicBytesSource.Generate. -func (s *statusData) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "Name:\t%s\n", s.t.Name()) - fmt.Fprintf(buf, "State:\t%s\n", s.t.StateStatus()) - fmt.Fprintf(buf, "Tgid:\t%d\n", s.pidns.IDOfThreadGroup(s.t.ThreadGroup())) - fmt.Fprintf(buf, "Pid:\t%d\n", s.pidns.IDOfTask(s.t)) - ppid := kernel.ThreadID(0) - if parent := s.t.Parent(); parent != nil { - ppid = s.pidns.IDOfThreadGroup(parent.ThreadGroup()) +func (i *taskOwnedInode) getOwner(mode linux.FileMode) (auth.KUID, auth.KGID) { + // By default, set the task owner as the file owner. + creds := i.owner.Credentials() + uid := creds.EffectiveKUID + gid := creds.EffectiveKGID + + // Linux doesn't apply dumpability adjustments to world readable/executable + // directories so that applications can stat /proc/PID to determine the + // effective UID of a process. See fs/proc/base.c:task_dump_owner. + if mode.FileType() == linux.ModeDirectory && mode.Permissions() == 0555 { + return uid, gid } - fmt.Fprintf(buf, "PPid:\t%d\n", ppid) - tpid := kernel.ThreadID(0) - if tracer := s.t.Tracer(); tracer != nil { - tpid = s.pidns.IDOfTask(tracer) + + // If the task is not dumpable, then root (in the namespace preferred) + // owns the file. + m := getMM(i.owner) + if m == nil { + return auth.RootKUID, auth.RootKGID } - fmt.Fprintf(buf, "TracerPid:\t%d\n", tpid) - var fds int - var vss, rss, data uint64 - s.t.WithMuLocked(func(t *kernel.Task) { - if fdTable := t.FDTable(); fdTable != nil { - fds = fdTable.Size() + if m.Dumpability() != mm.UserDumpable { + uid = auth.RootKUID + if kuid := creds.UserNamespace.MapToKUID(auth.RootUID); kuid.Ok() { + uid = kuid } - if mm := t.MemoryManager(); mm != nil { - vss = mm.VirtualMemorySize() - rss = mm.ResidentSetSize() - data = mm.VirtualDataSize() + gid = auth.RootKGID + if kgid := creds.UserNamespace.MapToKGID(auth.RootGID); kgid.Ok() { + gid = kgid } - }) - fmt.Fprintf(buf, "FDSize:\t%d\n", fds) - fmt.Fprintf(buf, "VmSize:\t%d kB\n", vss>>10) - fmt.Fprintf(buf, "VmRSS:\t%d kB\n", rss>>10) - fmt.Fprintf(buf, "VmData:\t%d kB\n", data>>10) - fmt.Fprintf(buf, "Threads:\t%d\n", s.t.ThreadGroup().Count()) - creds := s.t.Credentials() - fmt.Fprintf(buf, "CapInh:\t%016x\n", creds.InheritableCaps) - fmt.Fprintf(buf, "CapPrm:\t%016x\n", creds.PermittedCaps) - fmt.Fprintf(buf, "CapEff:\t%016x\n", creds.EffectiveCaps) - fmt.Fprintf(buf, "CapBnd:\t%016x\n", creds.BoundingCaps) - fmt.Fprintf(buf, "Seccomp:\t%d\n", s.t.SeccompMode()) - // We unconditionally report a single NUMA node. See - // pkg/sentry/syscalls/linux/sys_mempolicy.go. - fmt.Fprintf(buf, "Mems_allowed:\t1\n") - fmt.Fprintf(buf, "Mems_allowed_list:\t0\n") - return nil -} - -// ioUsage is the /proc//io and /proc//task//io data provider. -type ioUsage interface { - // IOUsage returns the io usage data. - IOUsage() *usage.IO -} - -// +stateify savable -type ioData struct { - ioUsage + } + return uid, gid } -var _ vfs.DynamicBytesSource = (*ioData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (i *ioData) Generate(ctx context.Context, buf *bytes.Buffer) error { - io := usage.IO{} - io.Accumulate(i.IOUsage()) - - fmt.Fprintf(buf, "char: %d\n", io.CharsRead) - fmt.Fprintf(buf, "wchar: %d\n", io.CharsWritten) - fmt.Fprintf(buf, "syscr: %d\n", io.ReadSyscalls) - fmt.Fprintf(buf, "syscw: %d\n", io.WriteSyscalls) - fmt.Fprintf(buf, "read_bytes: %d\n", io.BytesRead) - fmt.Fprintf(buf, "write_bytes: %d\n", io.BytesWritten) - fmt.Fprintf(buf, "cancelled_write_bytes: %d\n", io.BytesWriteCancelled) - return nil +func newIO(t *kernel.Task, isThreadGroup bool) *ioData { + if isThreadGroup { + return &ioData{ioUsage: t.ThreadGroup()} + } + return &ioData{ioUsage: t} } diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go new file mode 100644 index 000000000..93f0e1aa8 --- /dev/null +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -0,0 +1,272 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package proc + +import ( + "bytes" + "fmt" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/limits" + "gvisor.dev/gvisor/pkg/sentry/mm" + "gvisor.dev/gvisor/pkg/sentry/usage" + "gvisor.dev/gvisor/pkg/sentry/usermem" +) + +// mm gets the kernel task's MemoryManager. No additional reference is taken on +// mm here. This is safe because MemoryManager.destroy is required to leave the +// MemoryManager in a state where it's still usable as a DynamicBytesSource. +func getMM(task *kernel.Task) *mm.MemoryManager { + var tmm *mm.MemoryManager + task.WithMuLocked(func(t *kernel.Task) { + if mm := t.MemoryManager(); mm != nil { + tmm = mm + } + }) + return tmm +} + +// mapsData implements vfs.DynamicBytesSource for /proc/[pid]/maps. +// +// +stateify savable +type mapsData struct { + kernfs.DynamicBytesFile + + task *kernel.Task +} + +var _ dynamicInode = (*mapsData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (d *mapsData) Generate(ctx context.Context, buf *bytes.Buffer) error { + if mm := getMM(d.task); mm != nil { + mm.ReadMapsDataInto(ctx, buf) + } + return nil +} + +// smapsData implements vfs.DynamicBytesSource for /proc/[pid]/smaps. +// +// +stateify savable +type smapsData struct { + kernfs.DynamicBytesFile + + task *kernel.Task +} + +var _ dynamicInode = (*smapsData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (d *smapsData) Generate(ctx context.Context, buf *bytes.Buffer) error { + if mm := getMM(d.task); mm != nil { + mm.ReadSmapsDataInto(ctx, buf) + } + return nil +} + +// +stateify savable +type taskStatData struct { + kernfs.DynamicBytesFile + + t *kernel.Task + + // If tgstats is true, accumulate fault stats (not implemented) and CPU + // time across all tasks in t's thread group. + tgstats bool + + // pidns is the PID namespace associated with the proc filesystem that + // includes the file using this statData. + pidns *kernel.PIDNamespace +} + +var _ dynamicInode = (*taskStatData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (s *taskStatData) Generate(ctx context.Context, buf *bytes.Buffer) error { + fmt.Fprintf(buf, "%d ", s.pidns.IDOfTask(s.t)) + fmt.Fprintf(buf, "(%s) ", s.t.Name()) + fmt.Fprintf(buf, "%c ", s.t.StateStatus()[0]) + ppid := kernel.ThreadID(0) + if parent := s.t.Parent(); parent != nil { + ppid = s.pidns.IDOfThreadGroup(parent.ThreadGroup()) + } + fmt.Fprintf(buf, "%d ", ppid) + fmt.Fprintf(buf, "%d ", s.pidns.IDOfProcessGroup(s.t.ThreadGroup().ProcessGroup())) + fmt.Fprintf(buf, "%d ", s.pidns.IDOfSession(s.t.ThreadGroup().Session())) + fmt.Fprintf(buf, "0 0 " /* tty_nr tpgid */) + fmt.Fprintf(buf, "0 " /* flags */) + fmt.Fprintf(buf, "0 0 0 0 " /* minflt cminflt majflt cmajflt */) + var cputime usage.CPUStats + if s.tgstats { + cputime = s.t.ThreadGroup().CPUStats() + } else { + cputime = s.t.CPUStats() + } + fmt.Fprintf(buf, "%d %d ", linux.ClockTFromDuration(cputime.UserTime), linux.ClockTFromDuration(cputime.SysTime)) + cputime = s.t.ThreadGroup().JoinedChildCPUStats() + fmt.Fprintf(buf, "%d %d ", linux.ClockTFromDuration(cputime.UserTime), linux.ClockTFromDuration(cputime.SysTime)) + fmt.Fprintf(buf, "%d %d ", s.t.Priority(), s.t.Niceness()) + fmt.Fprintf(buf, "%d ", s.t.ThreadGroup().Count()) + + // itrealvalue. Since kernel 2.6.17, this field is no longer + // maintained, and is hard coded as 0. + fmt.Fprintf(buf, "0 ") + + // Start time is relative to boot time, expressed in clock ticks. + fmt.Fprintf(buf, "%d ", linux.ClockTFromDuration(s.t.StartTime().Sub(s.t.Kernel().Timekeeper().BootTime()))) + + var vss, rss uint64 + s.t.WithMuLocked(func(t *kernel.Task) { + if mm := t.MemoryManager(); mm != nil { + vss = mm.VirtualMemorySize() + rss = mm.ResidentSetSize() + } + }) + fmt.Fprintf(buf, "%d %d ", vss, rss/usermem.PageSize) + + // rsslim. + fmt.Fprintf(buf, "%d ", s.t.ThreadGroup().Limits().Get(limits.Rss).Cur) + + fmt.Fprintf(buf, "0 0 0 0 0 " /* startcode endcode startstack kstkesp kstkeip */) + fmt.Fprintf(buf, "0 0 0 0 0 " /* signal blocked sigignore sigcatch wchan */) + fmt.Fprintf(buf, "0 0 " /* nswap cnswap */) + terminationSignal := linux.Signal(0) + if s.t == s.t.ThreadGroup().Leader() { + terminationSignal = s.t.ThreadGroup().TerminationSignal() + } + fmt.Fprintf(buf, "%d ", terminationSignal) + fmt.Fprintf(buf, "0 0 0 " /* processor rt_priority policy */) + fmt.Fprintf(buf, "0 0 0 " /* delayacct_blkio_ticks guest_time cguest_time */) + fmt.Fprintf(buf, "0 0 0 0 0 0 0 " /* start_data end_data start_brk arg_start arg_end env_start env_end */) + fmt.Fprintf(buf, "0\n" /* exit_code */) + + return nil +} + +// statmData implements vfs.DynamicBytesSource for /proc/[pid]/statm. +// +// +stateify savable +type statmData struct { + kernfs.DynamicBytesFile + + t *kernel.Task +} + +var _ dynamicInode = (*statmData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (s *statmData) Generate(ctx context.Context, buf *bytes.Buffer) error { + var vss, rss uint64 + s.t.WithMuLocked(func(t *kernel.Task) { + if mm := t.MemoryManager(); mm != nil { + vss = mm.VirtualMemorySize() + rss = mm.ResidentSetSize() + } + }) + + fmt.Fprintf(buf, "%d %d 0 0 0 0 0\n", vss/usermem.PageSize, rss/usermem.PageSize) + return nil +} + +// statusData implements vfs.DynamicBytesSource for /proc/[pid]/status. +// +// +stateify savable +type statusData struct { + kernfs.DynamicBytesFile + + t *kernel.Task + pidns *kernel.PIDNamespace +} + +var _ dynamicInode = (*statusData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (s *statusData) Generate(ctx context.Context, buf *bytes.Buffer) error { + fmt.Fprintf(buf, "Name:\t%s\n", s.t.Name()) + fmt.Fprintf(buf, "State:\t%s\n", s.t.StateStatus()) + fmt.Fprintf(buf, "Tgid:\t%d\n", s.pidns.IDOfThreadGroup(s.t.ThreadGroup())) + fmt.Fprintf(buf, "Pid:\t%d\n", s.pidns.IDOfTask(s.t)) + ppid := kernel.ThreadID(0) + if parent := s.t.Parent(); parent != nil { + ppid = s.pidns.IDOfThreadGroup(parent.ThreadGroup()) + } + fmt.Fprintf(buf, "PPid:\t%d\n", ppid) + tpid := kernel.ThreadID(0) + if tracer := s.t.Tracer(); tracer != nil { + tpid = s.pidns.IDOfTask(tracer) + } + fmt.Fprintf(buf, "TracerPid:\t%d\n", tpid) + var fds int + var vss, rss, data uint64 + s.t.WithMuLocked(func(t *kernel.Task) { + if fdTable := t.FDTable(); fdTable != nil { + fds = fdTable.Size() + } + if mm := t.MemoryManager(); mm != nil { + vss = mm.VirtualMemorySize() + rss = mm.ResidentSetSize() + data = mm.VirtualDataSize() + } + }) + fmt.Fprintf(buf, "FDSize:\t%d\n", fds) + fmt.Fprintf(buf, "VmSize:\t%d kB\n", vss>>10) + fmt.Fprintf(buf, "VmRSS:\t%d kB\n", rss>>10) + fmt.Fprintf(buf, "VmData:\t%d kB\n", data>>10) + fmt.Fprintf(buf, "Threads:\t%d\n", s.t.ThreadGroup().Count()) + creds := s.t.Credentials() + fmt.Fprintf(buf, "CapInh:\t%016x\n", creds.InheritableCaps) + fmt.Fprintf(buf, "CapPrm:\t%016x\n", creds.PermittedCaps) + fmt.Fprintf(buf, "CapEff:\t%016x\n", creds.EffectiveCaps) + fmt.Fprintf(buf, "CapBnd:\t%016x\n", creds.BoundingCaps) + fmt.Fprintf(buf, "Seccomp:\t%d\n", s.t.SeccompMode()) + // We unconditionally report a single NUMA node. See + // pkg/sentry/syscalls/linux/sys_mempolicy.go. + fmt.Fprintf(buf, "Mems_allowed:\t1\n") + fmt.Fprintf(buf, "Mems_allowed_list:\t0\n") + return nil +} + +// ioUsage is the /proc//io and /proc//task//io data provider. +type ioUsage interface { + // IOUsage returns the io usage data. + IOUsage() *usage.IO +} + +// +stateify savable +type ioData struct { + kernfs.DynamicBytesFile + + ioUsage +} + +var _ dynamicInode = (*ioData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (i *ioData) Generate(ctx context.Context, buf *bytes.Buffer) error { + io := usage.IO{} + io.Accumulate(i.IOUsage()) + + fmt.Fprintf(buf, "char: %d\n", io.CharsRead) + fmt.Fprintf(buf, "wchar: %d\n", io.CharsWritten) + fmt.Fprintf(buf, "syscr: %d\n", io.ReadSyscalls) + fmt.Fprintf(buf, "syscw: %d\n", io.WriteSyscalls) + fmt.Fprintf(buf, "read_bytes: %d\n", io.BytesRead) + fmt.Fprintf(buf, "write_bytes: %d\n", io.BytesWritten) + fmt.Fprintf(buf, "cancelled_write_bytes: %d\n", io.BytesWriteCancelled) + return nil +} diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go new file mode 100644 index 000000000..50b2a832f --- /dev/null +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -0,0 +1,162 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package proc + +import ( + "sort" + "strconv" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +const defaultPermission = 0444 + +// InoGenerator generates unique inode numbers for a given filesystem. +type InoGenerator interface { + NextIno() uint64 +} + +// tasksInode represents the inode for /proc/ directory. +// +// +stateify savable +type tasksInode struct { + kernfs.InodeNotSymlink + kernfs.InodeDirectoryNoNewChildren + kernfs.InodeAttrs + kernfs.OrderedChildren + + inoGen InoGenerator + pidns *kernel.PIDNamespace +} + +var _ kernfs.Inode = (*tasksInode)(nil) + +func newTasksInode(inoGen InoGenerator, k *kernel.Kernel, pidns *kernel.PIDNamespace) (*tasksInode, *kernfs.Dentry) { + root := auth.NewRootCredentials(pidns.UserNamespace()) + contents := map[string]*kernfs.Dentry{ + //"cpuinfo": newCPUInfo(ctx, msrc), + //"filesystems": seqfile.NewSeqFileInode(ctx, &filesystemsData{}, msrc), + "loadavg": newDentry(root, inoGen.NextIno(), defaultPermission, &loadavgData{}), + "meminfo": newDentry(root, inoGen.NextIno(), defaultPermission, &meminfoData{k: k}), + "mounts": kernfs.NewStaticSymlink(root, inoGen.NextIno(), defaultPermission, "self/mounts"), + "self": newSelfSymlink(root, inoGen.NextIno(), defaultPermission, pidns), + "stat": newDentry(root, inoGen.NextIno(), defaultPermission, &statData{k: k}), + "thread-self": newThreadSelfSymlink(root, inoGen.NextIno(), defaultPermission, pidns), + //"uptime": newUptime(ctx, msrc), + //"version": newVersionData(root, inoGen.NextIno(), k), + "version": newDentry(root, inoGen.NextIno(), defaultPermission, &versionData{k: k}), + } + + inode := &tasksInode{ + pidns: pidns, + inoGen: inoGen, + } + inode.InodeAttrs.Init(root, inoGen.NextIno(), linux.ModeDirectory|0555) + + dentry := &kernfs.Dentry{} + dentry.Init(inode) + + inode.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) + links := inode.OrderedChildren.Populate(dentry, contents) + inode.IncLinks(links) + + return inode, dentry +} + +// Lookup implements kernfs.inodeDynamicLookup. +func (i *tasksInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { + // Try to lookup a corresponding task. + tid, err := strconv.ParseUint(name, 10, 64) + if err != nil { + return nil, syserror.ENOENT + } + + task := i.pidns.TaskWithID(kernel.ThreadID(tid)) + if task == nil { + return nil, syserror.ENOENT + } + + taskDentry := newTaskInode(i.inoGen, task, i.pidns, true) + return taskDentry.VFSDentry(), nil +} + +// Valid implements kernfs.inodeDynamicLookup. +func (i *tasksInode) Valid(ctx context.Context) bool { + return true +} + +// IterDirents implements kernfs.inodeDynamicLookup. +// +// TODO(gvisor.dev/issue/1195): Use tgid N offset = TGID_OFFSET + N. +func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, offset, relOffset int64) (int64, error) { + var tids []int + + // Collect all tasks. Per linux we only include it in directory listings if + // it's the leader. But for whatever crazy reason, you can still walk to the + // given node. + for _, tg := range i.pidns.ThreadGroups() { + if leader := tg.Leader(); leader != nil { + tids = append(tids, int(i.pidns.IDOfThreadGroup(tg))) + } + } + + if len(tids) == 0 { + return offset, nil + } + if relOffset >= int64(len(tids)) { + return offset, nil + } + + sort.Ints(tids) + for _, tid := range tids[relOffset:] { + dirent := vfs.Dirent{ + Name: strconv.FormatUint(uint64(tid), 10), + Type: linux.DT_DIR, + Ino: i.inoGen.NextIno(), + NextOff: offset + 1, + } + if !cb.Handle(dirent) { + return offset, nil + } + offset++ + } + return offset, nil +} + +// Open implements kernfs.Inode. +func (i *tasksInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { + fd := &kernfs.GenericDirectoryFD{} + fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, flags) + return fd.VFSFileDescription(), nil +} + +func (i *tasksInode) Stat(vsfs *vfs.Filesystem) linux.Statx { + stat := i.InodeAttrs.Stat(vsfs) + + // Add dynamic children to link count. + for _, tg := range i.pidns.ThreadGroups() { + if leader := tg.Leader(); leader != nil { + stat.Nlink++ + } + } + + return stat +} diff --git a/pkg/sentry/fsimpl/proc/tasks_files.go b/pkg/sentry/fsimpl/proc/tasks_files.go new file mode 100644 index 000000000..91f30a798 --- /dev/null +++ b/pkg/sentry/fsimpl/proc/tasks_files.go @@ -0,0 +1,92 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package proc + +import ( + "fmt" + "strconv" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/syserror" +) + +type selfSymlink struct { + kernfs.InodeAttrs + kernfs.InodeNoopRefCount + kernfs.InodeSymlink + + pidns *kernel.PIDNamespace +} + +var _ kernfs.Inode = (*selfSymlink)(nil) + +func newSelfSymlink(creds *auth.Credentials, ino uint64, perm linux.FileMode, pidns *kernel.PIDNamespace) *kernfs.Dentry { + inode := &selfSymlink{pidns: pidns} + inode.Init(creds, ino, linux.ModeSymlink|perm) + + d := &kernfs.Dentry{} + d.Init(inode) + return d +} + +func (s *selfSymlink) Readlink(ctx context.Context) (string, error) { + t := kernel.TaskFromContext(ctx) + if t == nil { + // Who is reading this link? + return "", syserror.EINVAL + } + tgid := s.pidns.IDOfThreadGroup(t.ThreadGroup()) + if tgid == 0 { + return "", syserror.ENOENT + } + return strconv.FormatUint(uint64(tgid), 10), nil +} + +type threadSelfSymlink struct { + kernfs.InodeAttrs + kernfs.InodeNoopRefCount + kernfs.InodeSymlink + + pidns *kernel.PIDNamespace +} + +var _ kernfs.Inode = (*threadSelfSymlink)(nil) + +func newThreadSelfSymlink(creds *auth.Credentials, ino uint64, perm linux.FileMode, pidns *kernel.PIDNamespace) *kernfs.Dentry { + inode := &threadSelfSymlink{pidns: pidns} + inode.Init(creds, ino, linux.ModeSymlink|perm) + + d := &kernfs.Dentry{} + d.Init(inode) + return d +} + +func (s *threadSelfSymlink) Readlink(ctx context.Context) (string, error) { + t := kernel.TaskFromContext(ctx) + if t == nil { + // Who is reading this link? + return "", syserror.EINVAL + } + tgid := s.pidns.IDOfThreadGroup(t.ThreadGroup()) + tid := s.pidns.IDOfTask(t) + if tid == 0 || tgid == 0 { + return "", syserror.ENOENT + } + return fmt.Sprintf("%d/task/%d", tgid, tid), nil +} diff --git a/pkg/sentry/fsimpl/proc/tasks_test.go b/pkg/sentry/fsimpl/proc/tasks_test.go new file mode 100644 index 000000000..48201d75a --- /dev/null +++ b/pkg/sentry/fsimpl/proc/tasks_test.go @@ -0,0 +1,412 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package proc + +import ( + "fmt" + "path" + "strconv" + "testing" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +type testIterDirentsCallback struct { + dirents []vfs.Dirent +} + +func (t *testIterDirentsCallback) Handle(d vfs.Dirent) bool { + t.dirents = append(t.dirents, d) + return true +} + +func checkDots(dirs []vfs.Dirent) ([]vfs.Dirent, error) { + if got := len(dirs); got < 2 { + return dirs, fmt.Errorf("wrong number of dirents, want at least: 2, got: %d: %v", got, dirs) + } + for i, want := range []string{".", ".."} { + if got := dirs[i].Name; got != want { + return dirs, fmt.Errorf("wrong name, want: %s, got: %s", want, got) + } + if got := dirs[i].Type; got != linux.DT_DIR { + return dirs, fmt.Errorf("wrong type, want: %d, got: %d", linux.DT_DIR, got) + } + } + return dirs[2:], nil +} + +func checkTasksStaticFiles(gots []vfs.Dirent) ([]vfs.Dirent, error) { + wants := map[string]vfs.Dirent{ + "loadavg": vfs.Dirent{Type: linux.DT_REG}, + "meminfo": vfs.Dirent{Type: linux.DT_REG}, + "mounts": vfs.Dirent{Type: linux.DT_LNK}, + "self": vfs.Dirent{Type: linux.DT_LNK}, + "stat": vfs.Dirent{Type: linux.DT_REG}, + "thread-self": vfs.Dirent{Type: linux.DT_LNK}, + "version": vfs.Dirent{Type: linux.DT_REG}, + } + return checkFiles(gots, wants) +} + +func checkTaskStaticFiles(gots []vfs.Dirent) ([]vfs.Dirent, error) { + wants := map[string]vfs.Dirent{ + "io": vfs.Dirent{Type: linux.DT_REG}, + "maps": vfs.Dirent{Type: linux.DT_REG}, + "smaps": vfs.Dirent{Type: linux.DT_REG}, + "stat": vfs.Dirent{Type: linux.DT_REG}, + "statm": vfs.Dirent{Type: linux.DT_REG}, + "status": vfs.Dirent{Type: linux.DT_REG}, + } + return checkFiles(gots, wants) +} + +func checkFiles(gots []vfs.Dirent, wants map[string]vfs.Dirent) ([]vfs.Dirent, error) { + // Go over all files, when there is a match, the file is removed from both + // 'gots' and 'wants'. wants is expected to reach 0, as all files must + // be present. Remaining files in 'gots', is returned to caller to decide + // whether this is valid or not. + for i := 0; i < len(gots); i++ { + got := gots[i] + want, ok := wants[got.Name] + if !ok { + continue + } + if want.Type != got.Type { + return gots, fmt.Errorf("wrong file type, want: %v, got: %v: %+v", want.Type, got.Type, got) + } + + delete(wants, got.Name) + gots = append(gots[0:i], gots[i+1:]...) + i-- + } + if len(wants) != 0 { + return gots, fmt.Errorf("not all files were found, missing: %+v", wants) + } + return gots, nil +} + +func setup() (context.Context, *vfs.VirtualFilesystem, vfs.VirtualDentry, error) { + k, err := boot() + if err != nil { + return nil, nil, vfs.VirtualDentry{}, fmt.Errorf("creating kernel: %v", err) + } + + ctx := k.SupervisorContext() + creds := auth.CredentialsFromContext(ctx) + + vfsObj := vfs.New() + vfsObj.MustRegisterFilesystemType("procfs", &procFSType{}) + mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "procfs", &vfs.GetFilesystemOptions{}) + if err != nil { + return nil, nil, vfs.VirtualDentry{}, fmt.Errorf("NewMountNamespace(): %v", err) + } + return ctx, vfsObj, mntns.Root(), nil +} + +func TestTasksEmpty(t *testing.T) { + ctx, vfsObj, root, err := setup() + if err != nil { + t.Fatalf("Setup failed: %v", err) + } + defer root.DecRef() + + fd, err := vfsObj.OpenAt( + ctx, + auth.CredentialsFromContext(ctx), + &vfs.PathOperation{Root: root, Start: root, Path: fspath.Parse("/")}, + &vfs.OpenOptions{}, + ) + if err != nil { + t.Fatalf("vfsfs.OpenAt failed: %v", err) + } + + cb := testIterDirentsCallback{} + if err := fd.Impl().IterDirents(ctx, &cb); err != nil { + t.Fatalf("IterDirents(): %v", err) + } + cb.dirents, err = checkDots(cb.dirents) + if err != nil { + t.Error(err.Error()) + } + cb.dirents, err = checkTasksStaticFiles(cb.dirents) + if err != nil { + t.Error(err.Error()) + } + if len(cb.dirents) != 0 { + t.Error("found more files than expected: %+v", cb.dirents) + } +} + +func TestTasks(t *testing.T) { + ctx, vfsObj, root, err := setup() + if err != nil { + t.Fatalf("Setup failed: %v", err) + } + defer root.DecRef() + + k := kernel.KernelFromContext(ctx) + var tasks []*kernel.Task + for i := 0; i < 5; i++ { + tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) + task, err := createTask(ctx, fmt.Sprintf("name-%d", i), tc) + if err != nil { + t.Fatalf("CreateTask(): %v", err) + } + tasks = append(tasks, task) + } + + fd, err := vfsObj.OpenAt( + ctx, + auth.CredentialsFromContext(ctx), + &vfs.PathOperation{Root: root, Start: root, Path: fspath.Parse("/")}, + &vfs.OpenOptions{}, + ) + if err != nil { + t.Fatalf("vfsfs.OpenAt(/) failed: %v", err) + } + + cb := testIterDirentsCallback{} + if err := fd.Impl().IterDirents(ctx, &cb); err != nil { + t.Fatalf("IterDirents(): %v", err) + } + cb.dirents, err = checkDots(cb.dirents) + if err != nil { + t.Error(err.Error()) + } + cb.dirents, err = checkTasksStaticFiles(cb.dirents) + if err != nil { + t.Error(err.Error()) + } + lastPid := 0 + for _, d := range cb.dirents { + pid, err := strconv.Atoi(d.Name) + if err != nil { + t.Fatalf("Invalid process directory %q", d.Name) + } + if lastPid > pid { + t.Errorf("pids not in order: %v", cb.dirents) + } + found := false + for _, t := range tasks { + if k.TaskSet().Root.IDOfTask(t) == kernel.ThreadID(pid) { + found = true + } + } + if !found { + t.Errorf("Additional task ID %d listed: %v", pid, tasks) + } + } + + // Test lookup. + for _, path := range []string{"/1", "/2"} { + fd, err := vfsObj.OpenAt( + ctx, + auth.CredentialsFromContext(ctx), + &vfs.PathOperation{Root: root, Start: root, Path: fspath.Parse(path)}, + &vfs.OpenOptions{}, + ) + if err != nil { + t.Fatalf("vfsfs.OpenAt(%q) failed: %v", path, err) + } + buf := make([]byte, 1) + bufIOSeq := usermem.BytesIOSequence(buf) + if _, err := fd.Read(ctx, bufIOSeq, vfs.ReadOptions{}); err != syserror.EISDIR { + t.Errorf("wrong error reading directory: %v", err) + } + } + + if _, err := vfsObj.OpenAt( + ctx, + auth.CredentialsFromContext(ctx), + &vfs.PathOperation{Root: root, Start: root, Path: fspath.Parse("/9999")}, + &vfs.OpenOptions{}, + ); err != syserror.ENOENT { + t.Fatalf("wrong error from vfsfs.OpenAt(/9999): %v", err) + } +} + +func TestTask(t *testing.T) { + ctx, vfsObj, root, err := setup() + if err != nil { + t.Fatalf("Setup failed: %v", err) + } + defer root.DecRef() + + k := kernel.KernelFromContext(ctx) + tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) + _, err = createTask(ctx, "name", tc) + if err != nil { + t.Fatalf("CreateTask(): %v", err) + } + + fd, err := vfsObj.OpenAt( + ctx, + auth.CredentialsFromContext(ctx), + &vfs.PathOperation{Root: root, Start: root, Path: fspath.Parse("/1")}, + &vfs.OpenOptions{}, + ) + if err != nil { + t.Fatalf("vfsfs.OpenAt(/1) failed: %v", err) + } + + cb := testIterDirentsCallback{} + if err := fd.Impl().IterDirents(ctx, &cb); err != nil { + t.Fatalf("IterDirents(): %v", err) + } + cb.dirents, err = checkDots(cb.dirents) + if err != nil { + t.Error(err.Error()) + } + cb.dirents, err = checkTaskStaticFiles(cb.dirents) + if err != nil { + t.Error(err.Error()) + } + if len(cb.dirents) != 0 { + t.Errorf("found more files than expected: %+v", cb.dirents) + } +} + +func TestProcSelf(t *testing.T) { + ctx, vfsObj, root, err := setup() + if err != nil { + t.Fatalf("Setup failed: %v", err) + } + defer root.DecRef() + + k := kernel.KernelFromContext(ctx) + tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) + task, err := createTask(ctx, "name", tc) + if err != nil { + t.Fatalf("CreateTask(): %v", err) + } + + fd, err := vfsObj.OpenAt( + task, + auth.CredentialsFromContext(ctx), + &vfs.PathOperation{Root: root, Start: root, Path: fspath.Parse("/self/"), FollowFinalSymlink: true}, + &vfs.OpenOptions{}, + ) + if err != nil { + t.Fatalf("vfsfs.OpenAt(/self/) failed: %v", err) + } + + cb := testIterDirentsCallback{} + if err := fd.Impl().IterDirents(ctx, &cb); err != nil { + t.Fatalf("IterDirents(): %v", err) + } + cb.dirents, err = checkDots(cb.dirents) + if err != nil { + t.Error(err.Error()) + } + cb.dirents, err = checkTaskStaticFiles(cb.dirents) + if err != nil { + t.Error(err.Error()) + } + if len(cb.dirents) != 0 { + t.Errorf("found more files than expected: %+v", cb.dirents) + } +} + +func iterateDir(ctx context.Context, t *testing.T, vfsObj *vfs.VirtualFilesystem, root vfs.VirtualDentry, fd *vfs.FileDescription) { + t.Logf("Iterating: /proc%s", fd.MappedName(ctx)) + + cb := testIterDirentsCallback{} + if err := fd.Impl().IterDirents(ctx, &cb); err != nil { + t.Fatalf("IterDirents(): %v", err) + } + var err error + cb.dirents, err = checkDots(cb.dirents) + if err != nil { + t.Error(err.Error()) + } + for _, d := range cb.dirents { + childPath := path.Join(fd.MappedName(ctx), d.Name) + if d.Type == linux.DT_LNK { + link, err := vfsObj.ReadlinkAt( + ctx, + auth.CredentialsFromContext(ctx), + &vfs.PathOperation{Root: root, Start: root, Path: fspath.Parse(childPath)}, + ) + if err != nil { + t.Errorf("vfsfs.ReadlinkAt(%v) failed: %v", childPath, err) + } else { + t.Logf("Skipping symlink: /proc%s => %s", childPath, link) + } + continue + } + + t.Logf("Opening: /proc%s", childPath) + child, err := vfsObj.OpenAt( + ctx, + auth.CredentialsFromContext(ctx), + &vfs.PathOperation{Root: root, Start: root, Path: fspath.Parse(childPath)}, + &vfs.OpenOptions{}, + ) + if err != nil { + t.Errorf("vfsfs.OpenAt(%v) failed: %v", childPath, err) + continue + } + stat, err := child.Stat(ctx, vfs.StatOptions{}) + if err != nil { + t.Errorf("Stat(%v) failed: %v", childPath, err) + } + if got := linux.FileMode(stat.Mode).DirentType(); got != d.Type { + t.Errorf("wrong file mode, stat: %v, dirent: %v", got, d.Type) + } + if d.Type == linux.DT_DIR { + // Found another dir, let's do it again! + iterateDir(ctx, t, vfsObj, root, child) + } + } +} + +// TestTree iterates all directories and stats every file. +func TestTree(t *testing.T) { + uberCtx, vfsObj, root, err := setup() + if err != nil { + t.Fatalf("Setup failed: %v", err) + } + defer root.DecRef() + + k := kernel.KernelFromContext(uberCtx) + var tasks []*kernel.Task + for i := 0; i < 5; i++ { + tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) + task, err := createTask(uberCtx, fmt.Sprintf("name-%d", i), tc) + if err != nil { + t.Fatalf("CreateTask(): %v", err) + } + tasks = append(tasks, task) + } + + ctx := tasks[0] + fd, err := vfsObj.OpenAt( + ctx, + auth.CredentialsFromContext(uberCtx), + &vfs.PathOperation{Root: root, Start: root, Path: fspath.Parse("/")}, + &vfs.OpenOptions{}, + ) + if err != nil { + t.Fatalf("vfsfs.OpenAt(/) failed: %v", err) + } + iterateDir(ctx, t, vfsObj, root, fd) +} diff --git a/pkg/sentry/fsimpl/proc/version.go b/pkg/sentry/fsimpl/proc/version.go index e1643d4e0..367f2396b 100644 --- a/pkg/sentry/fsimpl/proc/version.go +++ b/pkg/sentry/fsimpl/proc/version.go @@ -19,19 +19,21 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/vfs" ) // versionData implements vfs.DynamicBytesSource for /proc/version. // // +stateify savable type versionData struct { + kernfs.DynamicBytesFile + // k is the owning Kernel. k *kernel.Kernel } -var _ vfs.DynamicBytesSource = (*versionData)(nil) +var _ dynamicInode = (*versionData)(nil) // Generate implements vfs.DynamicBytesSource.Generate. func (v *versionData) Generate(ctx context.Context, buf *bytes.Buffer) error { diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index bd3fb4c03..8653d2f63 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -762,7 +762,7 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, mounts.IncRef() } - tg := k.newThreadGroup(mounts, args.PIDNamespace, NewSignalHandlers(), linux.SIGCHLD, args.Limits, k.monotonicClock) + tg := k.NewThreadGroup(mounts, args.PIDNamespace, NewSignalHandlers(), linux.SIGCHLD, args.Limits) ctx := args.NewContext(k) // Get the root directory from the MountNamespace. @@ -1191,6 +1191,11 @@ func (k *Kernel) GlobalInit() *ThreadGroup { return k.globalInit } +// TestOnly_SetGlobalInit sets the thread group with ID 1 in the root PID namespace. +func (k *Kernel) TestOnly_SetGlobalInit(tg *ThreadGroup) { + k.globalInit = tg +} + // ApplicationCores returns the number of CPUs visible to sandboxed // applications. func (k *Kernel) ApplicationCores() uint { diff --git a/pkg/sentry/kernel/task_clone.go b/pkg/sentry/kernel/task_clone.go index 3eadfedb4..5f3589493 100644 --- a/pkg/sentry/kernel/task_clone.go +++ b/pkg/sentry/kernel/task_clone.go @@ -243,7 +243,7 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { if opts.NewSignalHandlers { sh = sh.Fork() } - tg = t.k.newThreadGroup(tg.mounts, pidns, sh, opts.TerminationSignal, tg.limits.GetCopy(), t.k.monotonicClock) + tg = t.k.NewThreadGroup(tg.mounts, pidns, sh, opts.TerminationSignal, tg.limits.GetCopy()) } cfg := &TaskConfig{ diff --git a/pkg/sentry/kernel/thread_group.go b/pkg/sentry/kernel/thread_group.go index 72568d296..0cded73f6 100644 --- a/pkg/sentry/kernel/thread_group.go +++ b/pkg/sentry/kernel/thread_group.go @@ -256,20 +256,20 @@ type ThreadGroup struct { tty *TTY } -// newThreadGroup returns a new, empty thread group in PID namespace ns. The +// NewThreadGroup returns a new, empty thread group in PID namespace ns. The // thread group leader will send its parent terminationSignal when it exits. // The new thread group isn't visible to the system until a task has been // created inside of it by a successful call to TaskSet.NewTask. -func (k *Kernel) newThreadGroup(mounts *fs.MountNamespace, ns *PIDNamespace, sh *SignalHandlers, terminationSignal linux.Signal, limits *limits.LimitSet, monotonicClock *timekeeperClock) *ThreadGroup { +func (k *Kernel) NewThreadGroup(mntns *fs.MountNamespace, pidns *PIDNamespace, sh *SignalHandlers, terminationSignal linux.Signal, limits *limits.LimitSet) *ThreadGroup { tg := &ThreadGroup{ threadGroupNode: threadGroupNode{ - pidns: ns, + pidns: pidns, }, signalHandlers: sh, terminationSignal: terminationSignal, ioUsage: &usage.IO{}, limits: limits, - mounts: mounts, + mounts: mntns, } tg.itimerRealTimer = ktime.NewTimer(k.monotonicClock, &itimerRealListener{tg: tg}) tg.timers = make(map[linux.TimerID]*IntervalTimer) diff --git a/pkg/sentry/vfs/file_description_impl_util.go b/pkg/sentry/vfs/file_description_impl_util.go index 3df49991c..de782e577 100644 --- a/pkg/sentry/vfs/file_description_impl_util.go +++ b/pkg/sentry/vfs/file_description_impl_util.go @@ -199,6 +199,17 @@ type DynamicBytesSource interface { Generate(ctx context.Context, buf *bytes.Buffer) error } +// StaticData implements DynamicBytesSource over a static string. +type StaticData struct { + Data string +} + +// Generate implements DynamicBytesSource. +func (s *StaticData) Generate(ctx context.Context, buf *bytes.Buffer) error { + buf.WriteString(s.Data) + return nil +} + // SetDataSource must be called exactly once on fd before first use. func (fd *DynamicBytesFileDescriptionImpl) SetDataSource(data DynamicBytesSource) { fd.data = data -- cgit v1.2.3 From 796f53c0befc21570b185811e26b74e71950dfc3 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Fri, 27 Dec 2019 00:12:14 -0800 Subject: Add VFS2 support for /proc/filesystems. Updates #1195 PiperOrigin-RevId: 287269106 --- pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go | 4 +- pkg/sentry/fsimpl/ext/ext_test.go | 4 +- pkg/sentry/fsimpl/kernfs/kernfs_test.go | 4 +- pkg/sentry/fsimpl/memfs/benchmark_test.go | 8 +++- pkg/sentry/fsimpl/memfs/pipe_test.go | 4 +- pkg/sentry/fsimpl/proc/tasks_test.go | 30 +++++++------ pkg/sentry/vfs/file_description_impl_util_test.go | 2 +- pkg/sentry/vfs/filesystem_type.go | 55 ++++++++++++++++++++--- pkg/sentry/vfs/mount.go | 15 ++++--- pkg/sentry/vfs/options.go | 4 ++ pkg/sentry/vfs/vfs.go | 12 ++--- 11 files changed, 103 insertions(+), 39 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go index 2f46d2d13..a56b03711 100644 --- a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go +++ b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go @@ -50,7 +50,9 @@ func setUp(b *testing.B, imagePath string) (context.Context, *vfs.VirtualFilesys // Create VFS. vfsObj := vfs.New() - vfsObj.MustRegisterFilesystemType("extfs", ext.FilesystemType{}) + vfsObj.MustRegisterFilesystemType("extfs", ext.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + }) mntns, err := vfsObj.NewMountNamespace(ctx, creds, imagePath, "extfs", &vfs.GetFilesystemOptions{InternalData: int(f.Fd())}) if err != nil { f.Close() diff --git a/pkg/sentry/fsimpl/ext/ext_test.go b/pkg/sentry/fsimpl/ext/ext_test.go index 5d6c999bd..6c14a1e2d 100644 --- a/pkg/sentry/fsimpl/ext/ext_test.go +++ b/pkg/sentry/fsimpl/ext/ext_test.go @@ -66,7 +66,9 @@ func setUp(t *testing.T, imagePath string) (context.Context, *vfs.VirtualFilesys // Create VFS. vfsObj := vfs.New() - vfsObj.MustRegisterFilesystemType("extfs", FilesystemType{}) + vfsObj.MustRegisterFilesystemType("extfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + }) mntns, err := vfsObj.NewMountNamespace(ctx, creds, localImagePath, "extfs", &vfs.GetFilesystemOptions{InternalData: int(f.Fd())}) if err != nil { f.Close() diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index 3db12caa0..4b6b95f5f 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -59,7 +59,9 @@ func newTestSystem(t *testing.T, rootFn RootDentryFn) *TestSystem { ctx := contexttest.Context(t) creds := auth.CredentialsFromContext(ctx) v := vfs.New() - v.MustRegisterFilesystemType("testfs", &fsType{rootFn: rootFn}) + v.MustRegisterFilesystemType("testfs", &fsType{rootFn: rootFn}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + }) mns, err := v.NewMountNamespace(ctx, creds, "", "testfs", &vfs.GetFilesystemOptions{}) if err != nil { t.Fatalf("Failed to create testfs root mount: %v", err) diff --git a/pkg/sentry/fsimpl/memfs/benchmark_test.go b/pkg/sentry/fsimpl/memfs/benchmark_test.go index 6e987af88..a27876a4e 100644 --- a/pkg/sentry/fsimpl/memfs/benchmark_test.go +++ b/pkg/sentry/fsimpl/memfs/benchmark_test.go @@ -176,7 +176,9 @@ func BenchmarkVFS2MemfsStat(b *testing.B) { // Create VFS. vfsObj := vfs.New() - vfsObj.MustRegisterFilesystemType("memfs", memfs.FilesystemType{}) + vfsObj.MustRegisterFilesystemType("memfs", memfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + }) mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "memfs", &vfs.GetFilesystemOptions{}) if err != nil { b.Fatalf("failed to create tmpfs root mount: %v", err) @@ -365,7 +367,9 @@ func BenchmarkVFS2MemfsMountStat(b *testing.B) { // Create VFS. vfsObj := vfs.New() - vfsObj.MustRegisterFilesystemType("memfs", memfs.FilesystemType{}) + vfsObj.MustRegisterFilesystemType("memfs", memfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + }) mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "memfs", &vfs.GetFilesystemOptions{}) if err != nil { b.Fatalf("failed to create tmpfs root mount: %v", err) diff --git a/pkg/sentry/fsimpl/memfs/pipe_test.go b/pkg/sentry/fsimpl/memfs/pipe_test.go index be917aeee..807c1af7a 100644 --- a/pkg/sentry/fsimpl/memfs/pipe_test.go +++ b/pkg/sentry/fsimpl/memfs/pipe_test.go @@ -152,7 +152,9 @@ func setup(t *testing.T) (context.Context, *auth.Credentials, *vfs.VirtualFilesy // Create VFS. vfsObj := vfs.New() - vfsObj.MustRegisterFilesystemType("memfs", FilesystemType{}) + vfsObj.MustRegisterFilesystemType("memfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + }) mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "memfs", &vfs.GetFilesystemOptions{}) if err != nil { t.Fatalf("failed to create tmpfs root mount: %v", err) diff --git a/pkg/sentry/fsimpl/proc/tasks_test.go b/pkg/sentry/fsimpl/proc/tasks_test.go index 48201d75a..2560fcef9 100644 --- a/pkg/sentry/fsimpl/proc/tasks_test.go +++ b/pkg/sentry/fsimpl/proc/tasks_test.go @@ -56,25 +56,25 @@ func checkDots(dirs []vfs.Dirent) ([]vfs.Dirent, error) { func checkTasksStaticFiles(gots []vfs.Dirent) ([]vfs.Dirent, error) { wants := map[string]vfs.Dirent{ - "loadavg": vfs.Dirent{Type: linux.DT_REG}, - "meminfo": vfs.Dirent{Type: linux.DT_REG}, - "mounts": vfs.Dirent{Type: linux.DT_LNK}, - "self": vfs.Dirent{Type: linux.DT_LNK}, - "stat": vfs.Dirent{Type: linux.DT_REG}, - "thread-self": vfs.Dirent{Type: linux.DT_LNK}, - "version": vfs.Dirent{Type: linux.DT_REG}, + "loadavg": {Type: linux.DT_REG}, + "meminfo": {Type: linux.DT_REG}, + "mounts": {Type: linux.DT_LNK}, + "self": {Type: linux.DT_LNK}, + "stat": {Type: linux.DT_REG}, + "thread-self": {Type: linux.DT_LNK}, + "version": {Type: linux.DT_REG}, } return checkFiles(gots, wants) } func checkTaskStaticFiles(gots []vfs.Dirent) ([]vfs.Dirent, error) { wants := map[string]vfs.Dirent{ - "io": vfs.Dirent{Type: linux.DT_REG}, - "maps": vfs.Dirent{Type: linux.DT_REG}, - "smaps": vfs.Dirent{Type: linux.DT_REG}, - "stat": vfs.Dirent{Type: linux.DT_REG}, - "statm": vfs.Dirent{Type: linux.DT_REG}, - "status": vfs.Dirent{Type: linux.DT_REG}, + "io": {Type: linux.DT_REG}, + "maps": {Type: linux.DT_REG}, + "smaps": {Type: linux.DT_REG}, + "stat": {Type: linux.DT_REG}, + "statm": {Type: linux.DT_REG}, + "status": {Type: linux.DT_REG}, } return checkFiles(gots, wants) } @@ -114,7 +114,9 @@ func setup() (context.Context, *vfs.VirtualFilesystem, vfs.VirtualDentry, error) creds := auth.CredentialsFromContext(ctx) vfsObj := vfs.New() - vfsObj.MustRegisterFilesystemType("procfs", &procFSType{}) + vfsObj.MustRegisterFilesystemType("procfs", &procFSType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + }) mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "procfs", &vfs.GetFilesystemOptions{}) if err != nil { return nil, nil, vfs.VirtualDentry{}, fmt.Errorf("NewMountNamespace(): %v", err) diff --git a/pkg/sentry/vfs/file_description_impl_util_test.go b/pkg/sentry/vfs/file_description_impl_util_test.go index 678be07fe..9ed58512f 100644 --- a/pkg/sentry/vfs/file_description_impl_util_test.go +++ b/pkg/sentry/vfs/file_description_impl_util_test.go @@ -89,7 +89,7 @@ func TestGenCountFD(t *testing.T) { creds := auth.CredentialsFromContext(ctx) vfsObj := New() // vfs.New() - vfsObj.MustRegisterFilesystemType("testfs", FDTestFilesystemType{}) + vfsObj.MustRegisterFilesystemType("testfs", FDTestFilesystemType{}, &RegisterFilesystemTypeOptions{}) mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "testfs", &GetFilesystemOptions{}) if err != nil { t.Fatalf("failed to create testfs root mount: %v", err) diff --git a/pkg/sentry/vfs/filesystem_type.go b/pkg/sentry/vfs/filesystem_type.go index c335e206d..023301780 100644 --- a/pkg/sentry/vfs/filesystem_type.go +++ b/pkg/sentry/vfs/filesystem_type.go @@ -15,6 +15,7 @@ package vfs import ( + "bytes" "fmt" "gvisor.dev/gvisor/pkg/sentry/context" @@ -43,28 +44,70 @@ type GetFilesystemOptions struct { InternalData interface{} } +type registeredFilesystemType struct { + fsType FilesystemType + opts RegisterFilesystemTypeOptions +} + +// RegisterFilesystemTypeOptions contains options to +// VirtualFilesystem.RegisterFilesystem(). +type RegisterFilesystemTypeOptions struct { + // If AllowUserMount is true, allow calls to VirtualFilesystem.MountAt() + // for which MountOptions.InternalMount == false to use this filesystem + // type. + AllowUserMount bool + + // If AllowUserList is true, make this filesystem type visible in + // /proc/filesystems. + AllowUserList bool + + // If RequiresDevice is true, indicate that mounting this filesystem + // requires a block device as the mount source in /proc/filesystems. + RequiresDevice bool +} + // RegisterFilesystemType registers the given FilesystemType in vfs with the // given name. -func (vfs *VirtualFilesystem) RegisterFilesystemType(name string, fsType FilesystemType) error { +func (vfs *VirtualFilesystem) RegisterFilesystemType(name string, fsType FilesystemType, opts *RegisterFilesystemTypeOptions) error { vfs.fsTypesMu.Lock() defer vfs.fsTypesMu.Unlock() if existing, ok := vfs.fsTypes[name]; ok { - return fmt.Errorf("name %q is already registered to filesystem type %T", name, existing) + return fmt.Errorf("name %q is already registered to filesystem type %T", name, existing.fsType) + } + vfs.fsTypes[name] = ®isteredFilesystemType{ + fsType: fsType, + opts: *opts, } - vfs.fsTypes[name] = fsType return nil } // MustRegisterFilesystemType is equivalent to RegisterFilesystemType but // panics on failure. -func (vfs *VirtualFilesystem) MustRegisterFilesystemType(name string, fsType FilesystemType) { - if err := vfs.RegisterFilesystemType(name, fsType); err != nil { +func (vfs *VirtualFilesystem) MustRegisterFilesystemType(name string, fsType FilesystemType, opts *RegisterFilesystemTypeOptions) { + if err := vfs.RegisterFilesystemType(name, fsType, opts); err != nil { panic(fmt.Sprintf("failed to register filesystem type %T: %v", fsType, err)) } } -func (vfs *VirtualFilesystem) getFilesystemType(name string) FilesystemType { +func (vfs *VirtualFilesystem) getFilesystemType(name string) *registeredFilesystemType { vfs.fsTypesMu.RLock() defer vfs.fsTypesMu.RUnlock() return vfs.fsTypes[name] } + +// GenerateProcFilesystems emits the contents of /proc/filesystems for vfs to +// buf. +func (vfs *VirtualFilesystem) GenerateProcFilesystems(buf *bytes.Buffer) { + vfs.fsTypesMu.RLock() + defer vfs.fsTypesMu.RUnlock() + for name, rft := range vfs.fsTypes { + if !rft.opts.AllowUserList { + continue + } + var nodev string + if !rft.opts.RequiresDevice { + nodev = "nodev" + } + fmt.Fprintf(buf, "%s\t%s\n", nodev, name) + } +} diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index ec23ab0dd..00177b371 100644 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go @@ -112,11 +112,11 @@ type MountNamespace struct { // configured by the given arguments. A reference is taken on the returned // MountNamespace. func (vfs *VirtualFilesystem) NewMountNamespace(ctx context.Context, creds *auth.Credentials, source, fsTypeName string, opts *GetFilesystemOptions) (*MountNamespace, error) { - fsType := vfs.getFilesystemType(fsTypeName) - if fsType == nil { + rft := vfs.getFilesystemType(fsTypeName) + if rft == nil { return nil, syserror.ENODEV } - fs, root, err := fsType.GetFilesystem(ctx, vfs, creds, source, *opts) + fs, root, err := rft.fsType.GetFilesystem(ctx, vfs, creds, source, *opts) if err != nil { return nil, err } @@ -136,11 +136,14 @@ func (vfs *VirtualFilesystem) NewMountNamespace(ctx context.Context, creds *auth // MountAt creates and mounts a Filesystem configured by the given arguments. func (vfs *VirtualFilesystem) MountAt(ctx context.Context, creds *auth.Credentials, source string, target *PathOperation, fsTypeName string, opts *MountOptions) error { - fsType := vfs.getFilesystemType(fsTypeName) - if fsType == nil { + rft := vfs.getFilesystemType(fsTypeName) + if rft == nil { return syserror.ENODEV } - fs, root, err := fsType.GetFilesystem(ctx, vfs, creds, source, opts.GetFilesystemOptions) + if !opts.InternalMount && !rft.opts.AllowUserMount { + return syserror.ENODEV + } + fs, root, err := rft.fsType.GetFilesystem(ctx, vfs, creds, source, opts.GetFilesystemOptions) if err != nil { return err } diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go index 87d2b0d1c..b7774bf28 100644 --- a/pkg/sentry/vfs/options.go +++ b/pkg/sentry/vfs/options.go @@ -50,6 +50,10 @@ type MknodOptions struct { type MountOptions struct { // GetFilesystemOptions contains options to FilesystemType.GetFilesystem(). GetFilesystemOptions GetFilesystemOptions + + // If InternalMount is true, allow the use of filesystem types for which + // RegisterFilesystemTypeOptions.AllowUserMount == false. + InternalMount bool } // OpenOptions contains options to VirtualFilesystem.OpenAt() and diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 3e4df8558..a3bdb5805 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -75,23 +75,23 @@ type VirtualFilesystem struct { // mountpoints is analogous to Linux's mountpoint_hashtable. mountpoints map[*Dentry]map[*Mount]struct{} + // fsTypes contains all registered FilesystemTypes. fsTypes is protected by + // fsTypesMu. + fsTypesMu sync.RWMutex + fsTypes map[string]*registeredFilesystemType + // filesystems contains all Filesystems. filesystems is protected by // filesystemsMu. filesystemsMu sync.Mutex filesystems map[*Filesystem]struct{} - - // fsTypes contains all FilesystemTypes that are usable in the - // VirtualFilesystem. fsTypes is protected by fsTypesMu. - fsTypesMu sync.RWMutex - fsTypes map[string]FilesystemType } // New returns a new VirtualFilesystem with no mounts or FilesystemTypes. func New() *VirtualFilesystem { vfs := &VirtualFilesystem{ mountpoints: make(map[*Dentry]map[*Mount]struct{}), + fsTypes: make(map[string]*registeredFilesystemType), filesystems: make(map[*Filesystem]struct{}), - fsTypes: make(map[string]FilesystemType), } vfs.mounts.Init() return vfs -- cgit v1.2.3 From 1f384ac42b9ee8b52000dc2bff79d975853519ed Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Mon, 30 Dec 2019 11:35:06 -0800 Subject: Add VFS2 support for device special files. - Add FileDescriptionOptions.UseDentryMetadata, which reduces the amount of boilerplate needed for device FDs and the like between filesystems. - Switch back to having FileDescription.Init() take references on the Mount and Dentry; otherwise managing refcounts around failed calls to OpenDeviceSpecialFile() / Device.Open() is tricky. PiperOrigin-RevId: 287575574 --- pkg/sentry/fsimpl/ext/inode.go | 6 -- pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go | 2 - pkg/sentry/fsimpl/kernfs/fd_impl_util.go | 2 - pkg/sentry/fsimpl/memfs/filesystem.go | 4 - pkg/sentry/fsimpl/memfs/named_pipe.go | 2 - pkg/sentry/vfs/BUILD | 1 + pkg/sentry/vfs/device.go | 100 ++++++++++++++++++++++++ pkg/sentry/vfs/file_description.go | 101 +++++++++++++++++++++++-- pkg/sentry/vfs/file_description_impl_util.go | 15 ++++ pkg/sentry/vfs/filesystem.go | 21 +++++ pkg/sentry/vfs/vfs.go | 6 ++ 11 files changed, 236 insertions(+), 24 deletions(-) create mode 100644 pkg/sentry/vfs/device.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/ext/inode.go b/pkg/sentry/fsimpl/ext/inode.go index b2cc826c7..8608805bf 100644 --- a/pkg/sentry/fsimpl/ext/inode.go +++ b/pkg/sentry/fsimpl/ext/inode.go @@ -157,8 +157,6 @@ func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*v switch in.impl.(type) { case *regularFile: var fd regularFileFD - mnt.IncRef() - vfsd.IncRef() fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{}) return &fd.vfsfd, nil case *directory: @@ -168,8 +166,6 @@ func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*v return nil, syserror.EISDIR } var fd directoryFD - mnt.IncRef() - vfsd.IncRef() fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{}) return &fd.vfsfd, nil case *symlink: @@ -178,8 +174,6 @@ func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*v return nil, syserror.ELOOP } var fd symlinkFD - mnt.IncRef() - vfsd.IncRef() fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{}) return &fd.vfsfd, nil default: diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go index c5fe65722..606ca692d 100644 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -81,8 +81,6 @@ type DynamicBytesFD struct { // Init initializes a DynamicBytesFD. func (fd *DynamicBytesFD) Init(m *vfs.Mount, d *vfs.Dentry, data vfs.DynamicBytesSource, flags uint32) { - m.IncRef() // DecRef in vfs.FileDescription.vd.DecRef on final ref. - d.IncRef() // DecRef in vfs.FileDescription.vd.DecRef on final ref. fd.inode = d.Impl().(*Dentry).inode fd.SetDataSource(data) fd.vfsfd.Init(fd, flags, m, d, &vfs.FileDescriptionOptions{}) diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index 77975583b..bcf069b5f 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -44,8 +44,6 @@ type GenericDirectoryFD struct { // Init initializes a GenericDirectoryFD. func (fd *GenericDirectoryFD) Init(m *vfs.Mount, d *vfs.Dentry, children *OrderedChildren, flags uint32) { - m.IncRef() // DecRef in vfs.FileDescription.vd.DecRef on final ref. - d.IncRef() // DecRef in vfs.FileDescription.vd.DecRef on final ref. fd.children = children fd.vfsfd.Init(fd, flags, m, d, &vfs.FileDescriptionOptions{}) } diff --git a/pkg/sentry/fsimpl/memfs/filesystem.go b/pkg/sentry/fsimpl/memfs/filesystem.go index 4a83f310c..b063e09a3 100644 --- a/pkg/sentry/fsimpl/memfs/filesystem.go +++ b/pkg/sentry/fsimpl/memfs/filesystem.go @@ -348,8 +348,6 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, flags uint32, } // mnt.EndWrite() is called by regularFileFD.Release(). } - mnt.IncRef() - d.IncRef() fd.vfsfd.Init(&fd, flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}) if flags&linux.O_TRUNC != 0 { impl.mu.Lock() @@ -364,8 +362,6 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, flags uint32, return nil, syserror.EISDIR } var fd directoryFD - mnt.IncRef() - d.IncRef() fd.vfsfd.Init(&fd, flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}) return &fd.vfsfd, nil case *symlink: diff --git a/pkg/sentry/fsimpl/memfs/named_pipe.go b/pkg/sentry/fsimpl/memfs/named_pipe.go index d5060850e..b5a204438 100644 --- a/pkg/sentry/fsimpl/memfs/named_pipe.go +++ b/pkg/sentry/fsimpl/memfs/named_pipe.go @@ -55,8 +55,6 @@ func newNamedPipeFD(ctx context.Context, np *namedPipe, rp *vfs.ResolvingPath, v return nil, err } mnt := rp.Mount() - mnt.IncRef() - vfsd.IncRef() fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{}) return &fd.vfsfd, nil } diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index e3e554b88..4c6aa04a1 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -9,6 +9,7 @@ go_library( "context.go", "debug.go", "dentry.go", + "device.go", "file_description.go", "file_description_impl_util.go", "filesystem.go", diff --git a/pkg/sentry/vfs/device.go b/pkg/sentry/vfs/device.go new file mode 100644 index 000000000..cb672e36f --- /dev/null +++ b/pkg/sentry/vfs/device.go @@ -0,0 +1,100 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs + +import ( + "fmt" + + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/syserror" +) + +// DeviceKind indicates whether a device is a block or character device. +type DeviceKind uint32 + +const ( + // BlockDevice indicates a block device. + BlockDevice DeviceKind = iota + + // CharDevice indicates a character device. + CharDevice +) + +// String implements fmt.Stringer.String. +func (kind DeviceKind) String() string { + switch kind { + case BlockDevice: + return "block" + case CharDevice: + return "character" + default: + return fmt.Sprintf("invalid device kind %d", kind) + } +} + +type devTuple struct { + kind DeviceKind + major uint32 + minor uint32 +} + +// A Device backs device special files. +type Device interface { + // Open returns a FileDescription representing this device. + Open(ctx context.Context, mnt *Mount, d *Dentry, opts OpenOptions) (*FileDescription, error) +} + +type registeredDevice struct { + dev Device + opts RegisterDeviceOptions +} + +// RegisterDeviceOptions contains options to +// VirtualFilesystem.RegisterDevice(). +type RegisterDeviceOptions struct { + // GroupName is the name shown for this device registration in + // /proc/devices. If GroupName is empty, this registration will not be + // shown in /proc/devices. + GroupName string +} + +// RegisterDevice registers the given Device in vfs with the given major and +// minor device numbers. +func (vfs *VirtualFilesystem) RegisterDevice(kind DeviceKind, major, minor uint32, dev Device, opts *RegisterDeviceOptions) error { + tup := devTuple{kind, major, minor} + vfs.devicesMu.Lock() + defer vfs.devicesMu.Unlock() + if existing, ok := vfs.devices[tup]; ok { + return fmt.Errorf("%s device number (%d, %d) is already registered to device type %T", kind, major, minor, existing.dev) + } + vfs.devices[tup] = ®isteredDevice{ + dev: dev, + opts: *opts, + } + return nil +} + +// OpenDeviceSpecialFile returns a FileDescription representing the given +// device. +func (vfs *VirtualFilesystem) OpenDeviceSpecialFile(ctx context.Context, mnt *Mount, d *Dentry, kind DeviceKind, major, minor uint32, opts *OpenOptions) (*FileDescription, error) { + tup := devTuple{kind, major, minor} + vfs.devicesMu.RLock() + defer vfs.devicesMu.RUnlock() + rd, ok := vfs.devices[tup] + if !ok { + return nil, syserror.ENXIO + } + return rd.dev.Open(ctx, mnt, d, *opts) +} diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 0b053201a..6afe280bc 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -61,11 +61,25 @@ type FileDescriptionOptions struct { // If AllowDirectIO is true, allow O_DIRECT to be set on the file. This is // usually only the case if O_DIRECT would actually have an effect. AllowDirectIO bool + + // If UseDentryMetadata is true, calls to FileDescription methods that + // interact with file and filesystem metadata (Stat, SetStat, StatFS, + // Listxattr, Getxattr, Setxattr, Removexattr) are implemented by calling + // the corresponding FilesystemImpl methods instead of the corresponding + // FileDescriptionImpl methods. + // + // UseDentryMetadata is intended for file descriptions that are implemented + // outside of individual filesystems, such as pipes, sockets, and device + // special files. FileDescriptions for which UseDentryMetadata is true may + // embed DentryMetadataFileDescriptionImpl to obtain appropriate + // implementations of FileDescriptionImpl methods that should not be + // called. + UseDentryMetadata bool } -// Init must be called before first use of fd. It takes ownership of references -// on mnt and d held by the caller. statusFlags is the initial file description -// status flags, which is usually the full set of flags passed to open(2). +// Init must be called before first use of fd. It takes references on mnt and +// d. statusFlags is the initial file description status flags, which is +// usually the full set of flags passed to open(2). func (fd *FileDescription) Init(impl FileDescriptionImpl, statusFlags uint32, mnt *Mount, d *Dentry, opts *FileDescriptionOptions) { fd.refs = 1 fd.statusFlags = statusFlags | linux.O_LARGEFILE @@ -73,6 +87,7 @@ func (fd *FileDescription) Init(impl FileDescriptionImpl, statusFlags uint32, mn mount: mnt, dentry: d, } + fd.vd.IncRef() fd.opts = *opts fd.impl = impl } @@ -140,7 +155,7 @@ func (fd *FileDescription) SetStatusFlags(ctx context.Context, creds *auth.Crede // sense. However, the check as actually implemented seems to be "O_APPEND // cannot be changed if the file is marked as append-only". if (flags^oldFlags)&linux.O_APPEND != 0 { - stat, err := fd.impl.Stat(ctx, StatOptions{ + stat, err := fd.Stat(ctx, StatOptions{ // There is no mask bit for stx_attributes. Mask: 0, // Linux just reads inode::i_flags directly. @@ -154,7 +169,7 @@ func (fd *FileDescription) SetStatusFlags(ctx context.Context, creds *auth.Crede } } if (flags&linux.O_NOATIME != 0) && (oldFlags&linux.O_NOATIME == 0) { - stat, err := fd.impl.Stat(ctx, StatOptions{ + stat, err := fd.Stat(ctx, StatOptions{ Mask: linux.STATX_UID, // Linux's inode_owner_or_capable() just reads inode::i_uid // directly. @@ -348,17 +363,47 @@ func (fd *FileDescription) OnClose(ctx context.Context) error { // Stat returns metadata for the file represented by fd. func (fd *FileDescription) Stat(ctx context.Context, opts StatOptions) (linux.Statx, error) { + if fd.opts.UseDentryMetadata { + vfsObj := fd.vd.mount.vfs + rp := vfsObj.getResolvingPath(auth.CredentialsFromContext(ctx), &PathOperation{ + Root: fd.vd, + Start: fd.vd, + }) + stat, err := fd.vd.mount.fs.impl.StatAt(ctx, rp, opts) + vfsObj.putResolvingPath(rp) + return stat, err + } return fd.impl.Stat(ctx, opts) } // SetStat updates metadata for the file represented by fd. func (fd *FileDescription) SetStat(ctx context.Context, opts SetStatOptions) error { + if fd.opts.UseDentryMetadata { + vfsObj := fd.vd.mount.vfs + rp := vfsObj.getResolvingPath(auth.CredentialsFromContext(ctx), &PathOperation{ + Root: fd.vd, + Start: fd.vd, + }) + err := fd.vd.mount.fs.impl.SetStatAt(ctx, rp, opts) + vfsObj.putResolvingPath(rp) + return err + } return fd.impl.SetStat(ctx, opts) } // StatFS returns metadata for the filesystem containing the file represented // by fd. func (fd *FileDescription) StatFS(ctx context.Context) (linux.Statfs, error) { + if fd.opts.UseDentryMetadata { + vfsObj := fd.vd.mount.vfs + rp := vfsObj.getResolvingPath(auth.CredentialsFromContext(ctx), &PathOperation{ + Root: fd.vd, + Start: fd.vd, + }) + statfs, err := fd.vd.mount.fs.impl.StatFSAt(ctx, rp) + vfsObj.putResolvingPath(rp) + return statfs, err + } return fd.impl.StatFS(ctx) } @@ -417,6 +462,16 @@ func (fd *FileDescription) Ioctl(ctx context.Context, uio usermem.IO, args arch. // Listxattr returns all extended attribute names for the file represented by // fd. func (fd *FileDescription) Listxattr(ctx context.Context) ([]string, error) { + if fd.opts.UseDentryMetadata { + vfsObj := fd.vd.mount.vfs + rp := vfsObj.getResolvingPath(auth.CredentialsFromContext(ctx), &PathOperation{ + Root: fd.vd, + Start: fd.vd, + }) + names, err := fd.vd.mount.fs.impl.ListxattrAt(ctx, rp) + vfsObj.putResolvingPath(rp) + return names, err + } names, err := fd.impl.Listxattr(ctx) if err == syserror.ENOTSUP { // Linux doesn't actually return ENOTSUP in this case; instead, @@ -431,18 +486,48 @@ func (fd *FileDescription) Listxattr(ctx context.Context) ([]string, error) { // Getxattr returns the value associated with the given extended attribute for // the file represented by fd. func (fd *FileDescription) Getxattr(ctx context.Context, name string) (string, error) { + if fd.opts.UseDentryMetadata { + vfsObj := fd.vd.mount.vfs + rp := vfsObj.getResolvingPath(auth.CredentialsFromContext(ctx), &PathOperation{ + Root: fd.vd, + Start: fd.vd, + }) + val, err := fd.vd.mount.fs.impl.GetxattrAt(ctx, rp, name) + vfsObj.putResolvingPath(rp) + return val, err + } return fd.impl.Getxattr(ctx, name) } // Setxattr changes the value associated with the given extended attribute for // the file represented by fd. func (fd *FileDescription) Setxattr(ctx context.Context, opts SetxattrOptions) error { + if fd.opts.UseDentryMetadata { + vfsObj := fd.vd.mount.vfs + rp := vfsObj.getResolvingPath(auth.CredentialsFromContext(ctx), &PathOperation{ + Root: fd.vd, + Start: fd.vd, + }) + err := fd.vd.mount.fs.impl.SetxattrAt(ctx, rp, opts) + vfsObj.putResolvingPath(rp) + return err + } return fd.impl.Setxattr(ctx, opts) } // Removexattr removes the given extended attribute from the file represented // by fd. func (fd *FileDescription) Removexattr(ctx context.Context, name string) error { + if fd.opts.UseDentryMetadata { + vfsObj := fd.vd.mount.vfs + rp := vfsObj.getResolvingPath(auth.CredentialsFromContext(ctx), &PathOperation{ + Root: fd.vd, + Start: fd.vd, + }) + err := fd.vd.mount.fs.impl.RemovexattrAt(ctx, rp, name) + vfsObj.putResolvingPath(rp) + return err + } return fd.impl.Removexattr(ctx, name) } @@ -464,7 +549,7 @@ func (fd *FileDescription) MappedName(ctx context.Context) string { // DeviceID implements memmap.MappingIdentity.DeviceID. func (fd *FileDescription) DeviceID() uint64 { - stat, err := fd.impl.Stat(context.Background(), StatOptions{ + stat, err := fd.Stat(context.Background(), StatOptions{ // There is no STATX_DEV; we assume that Stat will return it if it's // available regardless of mask. Mask: 0, @@ -480,7 +565,7 @@ func (fd *FileDescription) DeviceID() uint64 { // InodeID implements memmap.MappingIdentity.InodeID. func (fd *FileDescription) InodeID() uint64 { - stat, err := fd.impl.Stat(context.Background(), StatOptions{ + stat, err := fd.Stat(context.Background(), StatOptions{ Mask: linux.STATX_INO, // fs/proc/task_mmu.c:show_map_vma() just reads inode::i_ino directly. Sync: linux.AT_STATX_DONT_SYNC, @@ -493,5 +578,5 @@ func (fd *FileDescription) InodeID() uint64 { // Msync implements memmap.MappingIdentity.Msync. func (fd *FileDescription) Msync(ctx context.Context, mr memmap.MappableRange) error { - return fd.impl.Sync(ctx) + return fd.Sync(ctx) } diff --git a/pkg/sentry/vfs/file_description_impl_util.go b/pkg/sentry/vfs/file_description_impl_util.go index de782e577..66eb57bc2 100644 --- a/pkg/sentry/vfs/file_description_impl_util.go +++ b/pkg/sentry/vfs/file_description_impl_util.go @@ -177,6 +177,21 @@ func (DirectoryFileDescriptionDefaultImpl) Write(ctx context.Context, src userme return 0, syserror.EISDIR } +// DentryMetadataFileDescriptionImpl may be embedded by implementations of +// FileDescriptionImpl for which FileDescriptionOptions.UseDentryMetadata is +// true to obtain implementations of Stat and SetStat that panic. +type DentryMetadataFileDescriptionImpl struct{} + +// Stat implements FileDescriptionImpl.Stat. +func (DentryMetadataFileDescriptionImpl) Stat(ctx context.Context, opts StatOptions) (linux.Statx, error) { + panic("illegal call to DentryMetadataFileDescriptionImpl.Stat") +} + +// SetStat implements FileDescriptionImpl.SetStat. +func (DentryMetadataFileDescriptionImpl) SetStat(ctx context.Context, opts SetStatOptions) error { + panic("illegal call to DentryMetadataFileDescriptionImpl.SetStat") +} + // DynamicBytesFileDescriptionImpl may be embedded by implementations of // FileDescriptionImpl that represent read-only regular files whose contents // are backed by a bytes.Buffer that is regenerated when necessary, consistent diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index 89bd58864..ea78f555b 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -418,17 +418,38 @@ type FilesystemImpl interface { UnlinkAt(ctx context.Context, rp *ResolvingPath) error // ListxattrAt returns all extended attribute names for the file at rp. + // + // Errors: + // + // - If extended attributes are not supported by the filesystem, + // ListxattrAt returns nil. (See FileDescription.Listxattr for an + // explanation.) ListxattrAt(ctx context.Context, rp *ResolvingPath) ([]string, error) // GetxattrAt returns the value associated with the given extended // attribute for the file at rp. + // + // Errors: + // + // - If extended attributes are not supported by the filesystem, GetxattrAt + // returns ENOTSUP. GetxattrAt(ctx context.Context, rp *ResolvingPath, name string) (string, error) // SetxattrAt changes the value associated with the given extended // attribute for the file at rp. + // + // Errors: + // + // - If extended attributes are not supported by the filesystem, SetxattrAt + // returns ENOTSUP. SetxattrAt(ctx context.Context, rp *ResolvingPath, opts SetxattrOptions) error // RemovexattrAt removes the given extended attribute from the file at rp. + // + // Errors: + // + // - If extended attributes are not supported by the filesystem, + // RemovexattrAt returns ENOTSUP. RemovexattrAt(ctx context.Context, rp *ResolvingPath, name string) error // PrependPath prepends a path from vd to vd.Mount().Root() to b. diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index a3bdb5805..ea2db7031 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -75,6 +75,11 @@ type VirtualFilesystem struct { // mountpoints is analogous to Linux's mountpoint_hashtable. mountpoints map[*Dentry]map[*Mount]struct{} + // devices contains all registered Devices. devices is protected by + // devicesMu. + devicesMu sync.RWMutex + devices map[devTuple]*registeredDevice + // fsTypes contains all registered FilesystemTypes. fsTypes is protected by // fsTypesMu. fsTypesMu sync.RWMutex @@ -90,6 +95,7 @@ type VirtualFilesystem struct { func New() *VirtualFilesystem { vfs := &VirtualFilesystem{ mountpoints: make(map[*Dentry]map[*Mount]struct{}), + devices: make(map[devTuple]*registeredDevice), fsTypes: make(map[string]*registeredFilesystemType), filesystems: make(map[*Filesystem]struct{}), } -- cgit v1.2.3 From bf53d325ddcd533d202efcab40047535078a02f3 Mon Sep 17 00:00:00 2001 From: Zach Koopmans Date: Fri, 3 Jan 2020 17:46:04 -0800 Subject: Remove FIXME comments to close old bug. PiperOrigin-RevId: 288075400 --- pkg/sentry/mm/procfs.go | 12 ------------ 1 file changed, 12 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/mm/procfs.go b/pkg/sentry/mm/procfs.go index 8c2246bb4..79610acb7 100644 --- a/pkg/sentry/mm/procfs.go +++ b/pkg/sentry/mm/procfs.go @@ -66,8 +66,6 @@ func (mm *MemoryManager) ReadMapsDataInto(ctx context.Context, buf *bytes.Buffer var start usermem.Addr for vseg := mm.vmas.LowerBoundSegment(start); vseg.Ok(); vseg = vseg.NextSegment() { - // FIXME(b/30793614): If we use a usermem.Addr for the handle, we get - // "panic: autosave error: type usermem.Addr is not registered". mm.appendVMAMapsEntryLocked(ctx, vseg, buf) } @@ -81,7 +79,6 @@ func (mm *MemoryManager) ReadMapsDataInto(ctx context.Context, buf *bytes.Buffer // // Artifically adjust the seqfile handle so we only output vsyscall entry once. if start != vsyscallEnd { - // FIXME(b/30793614): Can't get a pointer to constant vsyscallEnd. buf.WriteString(vsyscallMapsEntry) } } @@ -97,8 +94,6 @@ func (mm *MemoryManager) ReadMapsSeqFileData(ctx context.Context, handle seqfile start = *handle.(*usermem.Addr) } for vseg := mm.vmas.LowerBoundSegment(start); vseg.Ok(); vseg = vseg.NextSegment() { - // FIXME(b/30793614): If we use a usermem.Addr for the handle, we get - // "panic: autosave error: type usermem.Addr is not registered". vmaAddr := vseg.End() data = append(data, seqfile.SeqData{ Buf: mm.vmaMapsEntryLocked(ctx, vseg), @@ -116,7 +111,6 @@ func (mm *MemoryManager) ReadMapsSeqFileData(ctx context.Context, handle seqfile // // Artifically adjust the seqfile handle so we only output vsyscall entry once. if start != vsyscallEnd { - // FIXME(b/30793614): Can't get a pointer to constant vsyscallEnd. vmaAddr := vsyscallEnd data = append(data, seqfile.SeqData{ Buf: []byte(vsyscallMapsEntry), @@ -187,15 +181,12 @@ func (mm *MemoryManager) ReadSmapsDataInto(ctx context.Context, buf *bytes.Buffe var start usermem.Addr for vseg := mm.vmas.LowerBoundSegment(start); vseg.Ok(); vseg = vseg.NextSegment() { - // FIXME(b/30793614): If we use a usermem.Addr for the handle, we get - // "panic: autosave error: type usermem.Addr is not registered". mm.vmaSmapsEntryIntoLocked(ctx, vseg, buf) } // We always emulate vsyscall, so advertise it here. See // ReadMapsSeqFileData for additional commentary. if start != vsyscallEnd { - // FIXME(b/30793614): Can't get a pointer to constant vsyscallEnd. buf.WriteString(vsyscallSmapsEntry) } } @@ -211,8 +202,6 @@ func (mm *MemoryManager) ReadSmapsSeqFileData(ctx context.Context, handle seqfil start = *handle.(*usermem.Addr) } for vseg := mm.vmas.LowerBoundSegment(start); vseg.Ok(); vseg = vseg.NextSegment() { - // FIXME(b/30793614): If we use a usermem.Addr for the handle, we get - // "panic: autosave error: type usermem.Addr is not registered". vmaAddr := vseg.End() data = append(data, seqfile.SeqData{ Buf: mm.vmaSmapsEntryLocked(ctx, vseg), @@ -223,7 +212,6 @@ func (mm *MemoryManager) ReadSmapsSeqFileData(ctx context.Context, handle seqfil // We always emulate vsyscall, so advertise it here. See // ReadMapsSeqFileData for additional commentary. if start != vsyscallEnd { - // FIXME(b/30793614): Can't get a pointer to constant vsyscallEnd. vmaAddr := vsyscallEnd data = append(data, seqfile.SeqData{ Buf: []byte(vsyscallSmapsEntry), -- cgit v1.2.3 From 6410387ff9b4f0dbe88325ea0e30776f5f3efd5d Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Mon, 6 Jan 2020 09:27:35 -0800 Subject: Cleanup Shm reference handling Currently, shm.Registry.FindByID will return Shm instances without taking an additional reference on them, making it possible for them to disappear. More explicitly handle references. All callers hold a reference for the duration that they hold the instance. Registry.shms may transitively hold Shms with no references, so it must TryIncRef to determine if they are still valid. PiperOrigin-RevId: 288314529 --- pkg/sentry/kernel/shm/shm.go | 85 +++++++++++++++++++++++++----------- pkg/sentry/syscalls/linux/sys_shm.go | 7 ++- 2 files changed, 66 insertions(+), 26 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/shm/shm.go b/pkg/sentry/kernel/shm/shm.go index 5bd610f68..19034a21e 100644 --- a/pkg/sentry/kernel/shm/shm.go +++ b/pkg/sentry/kernel/shm/shm.go @@ -71,9 +71,20 @@ type Registry struct { mu sync.Mutex `state:"nosave"` // shms maps segment ids to segments. + // + // shms holds all referenced segments, which are removed on the last + // DecRef. Thus, it cannot itself hold a reference on the Shm. + // + // Since removal only occurs after the last (unlocked) DecRef, there + // exists a short window during which a Shm still exists in Shm, but is + // unreferenced. Users must use TryIncRef to determine if the Shm is + // still valid. shms map[ID]*Shm // keysToShms maps segment keys to segments. + // + // Shms in keysToShms are guaranteed to be referenced, as they are + // removed by disassociateKey before the last DecRef. keysToShms map[Key]*Shm // Sum of the sizes of all existing segments rounded up to page size, in @@ -95,10 +106,18 @@ func NewRegistry(userNS *auth.UserNamespace) *Registry { } // FindByID looks up a segment given an ID. +// +// FindByID returns a reference on Shm. func (r *Registry) FindByID(id ID) *Shm { r.mu.Lock() defer r.mu.Unlock() - return r.shms[id] + s := r.shms[id] + // Take a reference on s. If TryIncRef fails, s has reached the last + // DecRef, but hasn't quite been removed from r.shms yet. + if s != nil && s.TryIncRef() { + return s + } + return nil } // dissociateKey removes the association between a segment and its key, @@ -119,6 +138,8 @@ func (r *Registry) dissociateKey(s *Shm) { // FindOrCreate looks up or creates a segment in the registry. It's functionally // analogous to open(2). +// +// FindOrCreate returns a reference on Shm. func (r *Registry) FindOrCreate(ctx context.Context, pid int32, key Key, size uint64, mode linux.FileMode, private, create, exclusive bool) (*Shm, error) { if (create || private) && (size < linux.SHMMIN || size > linux.SHMMAX) { // "A new segment was to be created and size is less than SHMMIN or @@ -166,6 +187,7 @@ func (r *Registry) FindOrCreate(ctx context.Context, pid int32, key Key, size ui return nil, syserror.EEXIST } + shm.IncRef() return shm, nil } @@ -193,7 +215,14 @@ func (r *Registry) FindOrCreate(ctx context.Context, pid int32, key Key, size ui // Need to create a new segment. creator := fs.FileOwnerFromContext(ctx) perms := fs.FilePermsFromMode(mode) - return r.newShm(ctx, pid, key, creator, perms, size) + s, err := r.newShm(ctx, pid, key, creator, perms, size) + if err != nil { + return nil, err + } + // The initial reference is held by s itself. Take another to return to + // the caller. + s.IncRef() + return s, nil } // newShm creates a new segment in the registry. @@ -296,22 +325,26 @@ func (r *Registry) remove(s *Shm) { // Shm represents a single shared memory segment. // -// Shm segment are backed directly by an allocation from platform -// memory. Segments are always mapped as a whole, greatly simplifying how -// mappings are tracked. However note that mremap and munmap calls may cause the -// vma for a segment to become fragmented; which requires special care when -// unmapping a segment. See mm/shm.go. +// Shm segment are backed directly by an allocation from platform memory. +// Segments are always mapped as a whole, greatly simplifying how mappings are +// tracked. However note that mremap and munmap calls may cause the vma for a +// segment to become fragmented; which requires special care when unmapping a +// segment. See mm/shm.go. // // Segments persist until they are explicitly marked for destruction via -// shmctl(SHM_RMID). +// MarkDestroyed(). // // Shm implements memmap.Mappable and memmap.MappingIdentity. // // +stateify savable type Shm struct { - // AtomicRefCount tracks the number of references to this segment from - // maps. A segment always holds a reference to itself, until it's marked for + // AtomicRefCount tracks the number of references to this segment. + // + // A segment holds a reference to itself until it is marked for // destruction. + // + // In addition to direct users, the MemoryManager will hold references + // via MappingIdentity. refs.AtomicRefCount mfp pgalloc.MemoryFileProvider @@ -484,9 +517,8 @@ type AttachOpts struct { // ConfigureAttach creates an mmap configuration for the segment with the // requested attach options. // -// ConfigureAttach returns with a ref on s on success. The caller should drop -// this once the map is installed. This reference prevents s from being -// destroyed before the returned configuration is used. +// Postconditions: The returned MMapOpts are valid only as long as a reference +// continues to be held on s. func (s *Shm) ConfigureAttach(ctx context.Context, addr usermem.Addr, opts AttachOpts) (memmap.MMapOpts, error) { s.mu.Lock() defer s.mu.Unlock() @@ -504,7 +536,6 @@ func (s *Shm) ConfigureAttach(ctx context.Context, addr usermem.Addr, opts Attac // in the user namespace that governs its IPC namespace." - man shmat(2) return memmap.MMapOpts{}, syserror.EACCES } - s.IncRef() return memmap.MMapOpts{ Length: s.size, Offset: 0, @@ -549,10 +580,15 @@ func (s *Shm) IPCStat(ctx context.Context) (*linux.ShmidDS, error) { } creds := auth.CredentialsFromContext(ctx) - nattach := uint64(s.ReadRefs()) - // Don't report the self-reference we keep prior to being marked for - // destruction. However, also don't report a count of -1 for segments marked - // as destroyed, with no mappings. + // Use the reference count as a rudimentary count of the number of + // attaches. We exclude: + // + // 1. The reference the caller holds. + // 2. The self-reference held by s prior to destruction. + // + // Note that this may still overcount by including transient references + // used in concurrent calls. + nattach := uint64(s.ReadRefs()) - 1 if !s.pendingDestruction { nattach-- } @@ -620,18 +656,17 @@ func (s *Shm) MarkDestroyed() { s.registry.dissociateKey(s) s.mu.Lock() - // Only drop the segment's self-reference once, when destruction is - // requested. Otherwise, repeated calls to shmctl(IPC_RMID) would force a - // segment to be destroyed prematurely, potentially with active maps to the - // segment's address range. Remaining references are dropped when the - // segment is detached or unmaped. + defer s.mu.Unlock() if !s.pendingDestruction { s.pendingDestruction = true - s.mu.Unlock() // Must release s.mu before calling s.DecRef. + // Drop the self-reference so destruction occurs when all + // external references are gone. + // + // N.B. This cannot be the final DecRef, as the caller also + // holds a reference. s.DecRef() return } - s.mu.Unlock() } // checkOwnership verifies whether a segment may be accessed by ctx as an diff --git a/pkg/sentry/syscalls/linux/sys_shm.go b/pkg/sentry/syscalls/linux/sys_shm.go index d57ffb3a1..4a8bc24a2 100644 --- a/pkg/sentry/syscalls/linux/sys_shm.go +++ b/pkg/sentry/syscalls/linux/sys_shm.go @@ -39,10 +39,13 @@ func Shmget(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, err } + defer segment.DecRef() return uintptr(segment.ID), nil, nil } // findSegment retrives a shm segment by the given id. +// +// findSegment returns a reference on Shm. func findSegment(t *kernel.Task, id shm.ID) (*shm.Shm, error) { r := t.IPCNamespace().ShmRegistry() segment := r.FindByID(id) @@ -63,6 +66,7 @@ func Shmat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if err != nil { return 0, nil, syserror.EINVAL } + defer segment.DecRef() opts, err := segment.ConfigureAttach(t, addr, shm.AttachOpts{ Execute: flag&linux.SHM_EXEC == linux.SHM_EXEC, @@ -72,7 +76,6 @@ func Shmat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if err != nil { return 0, nil, err } - defer segment.DecRef() addr, err = t.MemoryManager().MMap(t, opts) return uintptr(addr), nil, err } @@ -105,6 +108,7 @@ func Shmctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, syserror.EINVAL } + defer segment.DecRef() stat, err := segment.IPCStat(t) if err == nil { @@ -128,6 +132,7 @@ func Shmctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, syserror.EINVAL } + defer segment.DecRef() switch cmd { case linux.IPC_SET: -- cgit v1.2.3 From 354a15a234c1270bcb9b902503f61835b2ccd2d0 Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Mon, 6 Jan 2020 11:41:13 -0800 Subject: Implement rseq(2) PiperOrigin-RevId: 288342928 --- pkg/abi/linux/BUILD | 1 + pkg/abi/linux/rseq.go | 130 ++++++++++ pkg/sentry/arch/arch.go | 6 +- pkg/sentry/arch/arch_amd64.go | 4 +- pkg/sentry/kernel/rseq.go | 383 +++++++++++++++++++++++++---- pkg/sentry/kernel/task.go | 43 +++- pkg/sentry/kernel/task_clone.go | 7 + pkg/sentry/kernel/task_exec.go | 6 +- pkg/sentry/kernel/task_run.go | 16 +- pkg/sentry/kernel/task_start.go | 10 + pkg/sentry/kernel/thread_group.go | 18 +- pkg/sentry/syscalls/linux/BUILD | 1 + pkg/sentry/syscalls/linux/linux64_amd64.go | 2 +- pkg/sentry/syscalls/linux/linux64_arm64.go | 2 +- pkg/sentry/syscalls/linux/sys_rseq.go | 48 ++++ 15 files changed, 598 insertions(+), 79 deletions(-) create mode 100644 pkg/abi/linux/rseq.go create mode 100644 pkg/sentry/syscalls/linux/sys_rseq.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/BUILD b/pkg/abi/linux/BUILD index 9553f164d..716ff22d2 100644 --- a/pkg/abi/linux/BUILD +++ b/pkg/abi/linux/BUILD @@ -41,6 +41,7 @@ go_library( "poll.go", "prctl.go", "ptrace.go", + "rseq.go", "rusage.go", "sched.go", "seccomp.go", diff --git a/pkg/abi/linux/rseq.go b/pkg/abi/linux/rseq.go new file mode 100644 index 000000000..76253ba30 --- /dev/null +++ b/pkg/abi/linux/rseq.go @@ -0,0 +1,130 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package linux + +// Flags passed to rseq(2). +// +// Defined in include/uapi/linux/rseq.h. +const ( + // RSEQ_FLAG_UNREGISTER unregisters the current thread. + RSEQ_FLAG_UNREGISTER = 1 << 0 +) + +// Critical section flags used in RSeqCriticalSection.Flags and RSeq.Flags. +// +// Defined in include/uapi/linux/rseq.h. +const ( + // RSEQ_CS_FLAG_NO_RESTART_ON_PREEMPT inhibits restart on preemption. + RSEQ_CS_FLAG_NO_RESTART_ON_PREEMPT = 1 << 0 + + // RSEQ_CS_FLAG_NO_RESTART_ON_SIGNAL inhibits restart on signal + // delivery. + RSEQ_CS_FLAG_NO_RESTART_ON_SIGNAL = 1 << 1 + + // RSEQ_CS_FLAG_NO_RESTART_ON_MIGRATE inhibits restart on CPU + // migration. + RSEQ_CS_FLAG_NO_RESTART_ON_MIGRATE = 1 << 2 +) + +// RSeqCriticalSection describes a restartable sequences critical section. It +// is equivalent to struct rseq_cs, defined in include/uapi/linux/rseq.h. +// +// In userspace, this structure is always aligned to 32 bytes. +// +// +marshal +type RSeqCriticalSection struct { + // Version is the version of this structure. Version 0 is defined here. + Version uint32 + + // Flags are the critical section flags, defined above. + Flags uint32 + + // Start is the start address of the critical section. + Start uint64 + + // PostCommitOffset is the offset from Start of the first instruction + // outside of the critical section. + PostCommitOffset uint64 + + // Abort is the abort address. It must be outside the critical section, + // and the 4 bytes prior must match the abort signature. + Abort uint64 +} + +const ( + // SizeOfRSeqCriticalSection is the size of RSeqCriticalSection. + SizeOfRSeqCriticalSection = 32 + + // SizeOfRSeqSignature is the size of the signature immediately + // preceding RSeqCriticalSection.Abort. + SizeOfRSeqSignature = 4 +) + +// Special values for RSeq.CPUID, defined in include/uapi/linux/rseq.h. +const ( + // RSEQ_CPU_ID_UNINITIALIZED indicates that this thread has not + // performed rseq initialization. + RSEQ_CPU_ID_UNINITIALIZED = ^uint32(0) // -1 + + // RSEQ_CPU_ID_REGISTRATION_FAILED indicates that rseq initialization + // failed. + RSEQ_CPU_ID_REGISTRATION_FAILED = ^uint32(1) // -2 +) + +// RSeq is the thread-local restartable sequences config/status. It +// is equivalent to struct rseq, defined in include/uapi/linux/rseq.h. +// +// In userspace, this structure is always aligned to 32 bytes. +type RSeq struct { + // CPUIDStart contains the current CPU ID if rseq is initialized. + // + // This field should only be read by the thread which registered this + // structure, and must be read atomically. + CPUIDStart uint32 + + // CPUID contains the current CPU ID or one of the CPU ID special + // values defined above. + // + // This field should only be read by the thread which registered this + // structure, and must be read atomically. + CPUID uint32 + + // RSeqCriticalSection is a pointer to the current RSeqCriticalSection + // block, or NULL. It is reset to NULL by the kernel on restart or + // non-restarting preempt/signal. + // + // This field should only be written by the thread which registered + // this structure, and must be written atomically. + RSeqCriticalSection uint64 + + // Flags are the critical section flags that apply to all critical + // sections on this thread, defined above. + Flags uint32 +} + +const ( + // SizeOfRSeq is the size of RSeq. + // + // Note that RSeq is naively 24 bytes. However, it has 32-byte + // alignment, which in C increases sizeof to 32. That is the size that + // the Linux kernel uses. + SizeOfRSeq = 32 + + // AlignOfRSeq is the standard alignment of RSeq. + AlignOfRSeq = 32 + + // OffsetOfRSeqCriticalSection is the offset of RSeqCriticalSection in RSeq. + OffsetOfRSeqCriticalSection = 8 +) diff --git a/pkg/sentry/arch/arch.go b/pkg/sentry/arch/arch.go index 498ca4669..81ec98a77 100644 --- a/pkg/sentry/arch/arch.go +++ b/pkg/sentry/arch/arch.go @@ -125,9 +125,9 @@ type Context interface { // SetTLS sets the current TLS pointer. Returns false if value is invalid. SetTLS(value uintptr) bool - // SetRSEQInterruptedIP sets the register that contains the old IP when a - // restartable sequence is interrupted. - SetRSEQInterruptedIP(value uintptr) + // SetOldRSeqInterruptedIP sets the register that contains the old IP + // when an "old rseq" restartable sequence is interrupted. + SetOldRSeqInterruptedIP(value uintptr) // StateData returns a pointer to underlying architecture state. StateData() *State diff --git a/pkg/sentry/arch/arch_amd64.go b/pkg/sentry/arch/arch_amd64.go index 67daa6c24..2aa08b1a9 100644 --- a/pkg/sentry/arch/arch_amd64.go +++ b/pkg/sentry/arch/arch_amd64.go @@ -174,8 +174,8 @@ func (c *context64) SetTLS(value uintptr) bool { return true } -// SetRSEQInterruptedIP implements Context.SetRSEQInterruptedIP. -func (c *context64) SetRSEQInterruptedIP(value uintptr) { +// SetOldRSeqInterruptedIP implements Context.SetOldRSeqInterruptedIP. +func (c *context64) SetOldRSeqInterruptedIP(value uintptr) { c.Regs.R10 = uint64(value) } diff --git a/pkg/sentry/kernel/rseq.go b/pkg/sentry/kernel/rseq.go index 24ea002ba..b14429854 100644 --- a/pkg/sentry/kernel/rseq.go +++ b/pkg/sentry/kernel/rseq.go @@ -15,17 +15,29 @@ package kernel import ( + "fmt" + + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/hostcpu" "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" ) -// Restartable sequences, as described in https://lwn.net/Articles/650333/. +// Restartable sequences. +// +// We support two different APIs for restartable sequences. +// +// 1. The upstream interface added in v4.18. +// 2. The interface described in https://lwn.net/Articles/650333/. +// +// Throughout this file and other parts of the kernel, the latter is referred +// to as "old rseq". This interface was never merged upstream, but is supported +// for a limited set of applications that use it regardless. -// RSEQCriticalRegion describes a restartable sequence critical region. +// OldRSeqCriticalRegion describes an old rseq critical region. // // +stateify savable -type RSEQCriticalRegion struct { +type OldRSeqCriticalRegion struct { // When a task in this thread group has its CPU preempted (as defined by // platform.ErrContextCPUPreempted) or has a signal delivered to an // application handler while its instruction pointer is in CriticalSection, @@ -35,86 +47,359 @@ type RSEQCriticalRegion struct { Restart usermem.Addr } -// RSEQAvailable returns true if t supports restartable sequences. -func (t *Task) RSEQAvailable() bool { +// RSeqAvailable returns true if t supports (old and new) restartable sequences. +func (t *Task) RSeqAvailable() bool { return t.k.useHostCores && t.k.Platform.DetectsCPUPreemption() } -// RSEQCriticalRegion returns a copy of t's thread group's current restartable -// sequence. -func (t *Task) RSEQCriticalRegion() RSEQCriticalRegion { - return *t.tg.rscr.Load().(*RSEQCriticalRegion) +// SetRSeq registers addr as this thread's rseq structure. +// +// Preconditions: The caller must be running on the task goroutine. +func (t *Task) SetRSeq(addr usermem.Addr, length, signature uint32) error { + if t.rseqAddr != 0 { + if t.rseqAddr != addr { + return syserror.EINVAL + } + if t.rseqSignature != signature { + return syserror.EINVAL + } + return syserror.EBUSY + } + + // rseq must be aligned and correctly sized. + if addr&(linux.AlignOfRSeq-1) != 0 { + return syserror.EINVAL + } + if length != linux.SizeOfRSeq { + return syserror.EINVAL + } + if _, ok := t.MemoryManager().CheckIORange(addr, linux.SizeOfRSeq); !ok { + return syserror.EFAULT + } + + t.rseqAddr = addr + t.rseqSignature = signature + + // Initialize the CPUID. + // + // Linux implicitly does this on return from userspace, where failure + // would cause SIGSEGV. + if err := t.rseqUpdateCPU(); err != nil { + t.rseqAddr = 0 + t.rseqSignature = 0 + + t.Debugf("Failed to copy CPU to %#x for rseq: %v", t.rseqAddr, err) + t.forceSignal(linux.SIGSEGV, false /* unconditional */) + t.SendSignal(SignalInfoPriv(linux.SIGSEGV)) + return syserror.EFAULT + } + + return nil } -// SetRSEQCriticalRegion replaces t's thread group's restartable sequence. +// ClearRSeq unregisters addr as this thread's rseq structure. // -// Preconditions: t.RSEQAvailable() == true. -func (t *Task) SetRSEQCriticalRegion(rscr RSEQCriticalRegion) error { +// Preconditions: The caller must be running on the task goroutine. +func (t *Task) ClearRSeq(addr usermem.Addr, length, signature uint32) error { + if t.rseqAddr == 0 { + return syserror.EINVAL + } + if t.rseqAddr != addr { + return syserror.EINVAL + } + if length != linux.SizeOfRSeq { + return syserror.EINVAL + } + if t.rseqSignature != signature { + return syserror.EPERM + } + + if err := t.rseqClearCPU(); err != nil { + return err + } + + t.rseqAddr = 0 + t.rseqSignature = 0 + + if t.oldRSeqCPUAddr == 0 { + // rseqCPU no longer needed. + t.rseqCPU = -1 + } + + return nil +} + +// OldRSeqCriticalRegion returns a copy of t's thread group's current +// old restartable sequence. +func (t *Task) OldRSeqCriticalRegion() OldRSeqCriticalRegion { + return *t.tg.oldRSeqCritical.Load().(*OldRSeqCriticalRegion) +} + +// SetOldRSeqCriticalRegion replaces t's thread group's old restartable +// sequence. +// +// Preconditions: t.RSeqAvailable() == true. +func (t *Task) SetOldRSeqCriticalRegion(r OldRSeqCriticalRegion) error { // These checks are somewhat more lenient than in Linux, which (bizarrely) - // requires rscr.CriticalSection to be non-empty and rscr.Restart to be - // outside of rscr.CriticalSection, even if rscr.CriticalSection.Start == 0 + // requires r.CriticalSection to be non-empty and r.Restart to be + // outside of r.CriticalSection, even if r.CriticalSection.Start == 0 // (which disables the critical region). - if rscr.CriticalSection.Start == 0 { - rscr.CriticalSection.End = 0 - rscr.Restart = 0 - t.tg.rscr.Store(&rscr) + if r.CriticalSection.Start == 0 { + r.CriticalSection.End = 0 + r.Restart = 0 + t.tg.oldRSeqCritical.Store(&r) return nil } - if rscr.CriticalSection.Start >= rscr.CriticalSection.End { + if r.CriticalSection.Start >= r.CriticalSection.End { return syserror.EINVAL } - if rscr.CriticalSection.Contains(rscr.Restart) { + if r.CriticalSection.Contains(r.Restart) { return syserror.EINVAL } - // TODO(jamieliu): check that rscr.CriticalSection and rscr.Restart are in - // the application address range, for consistency with Linux - t.tg.rscr.Store(&rscr) + // TODO(jamieliu): check that r.CriticalSection and r.Restart are in + // the application address range, for consistency with Linux. + t.tg.oldRSeqCritical.Store(&r) return nil } -// RSEQCPUAddr returns the address that RSEQ will keep updated with t's CPU -// number. +// OldRSeqCPUAddr returns the address that old rseq will keep updated with t's +// CPU number. // // Preconditions: The caller must be running on the task goroutine. -func (t *Task) RSEQCPUAddr() usermem.Addr { - return t.rseqCPUAddr +func (t *Task) OldRSeqCPUAddr() usermem.Addr { + return t.oldRSeqCPUAddr } -// SetRSEQCPUAddr replaces the address that RSEQ will keep updated with t's CPU -// number. +// SetOldRSeqCPUAddr replaces the address that old rseq will keep updated with +// t's CPU number. // -// Preconditions: t.RSEQAvailable() == true. The caller must be running on the +// Preconditions: t.RSeqAvailable() == true. The caller must be running on the // task goroutine. t's AddressSpace must be active. -func (t *Task) SetRSEQCPUAddr(addr usermem.Addr) error { - t.rseqCPUAddr = addr - if addr != 0 { - t.rseqCPU = int32(hostcpu.GetCPU()) - if err := t.rseqCopyOutCPU(); err != nil { - t.rseqCPUAddr = 0 - t.rseqCPU = -1 - return syserror.EINVAL // yes, EINVAL, not err or EFAULT - } - } else { - t.rseqCPU = -1 +func (t *Task) SetOldRSeqCPUAddr(addr usermem.Addr) error { + t.oldRSeqCPUAddr = addr + + // Check that addr is writable. + // + // N.B. rseqUpdateCPU may fail on a bad t.rseqAddr as well. That's + // unfortunate, but unlikely in a correct program. + if err := t.rseqUpdateCPU(); err != nil { + t.oldRSeqCPUAddr = 0 + return syserror.EINVAL // yes, EINVAL, not err or EFAULT } return nil } // Preconditions: The caller must be running on the task goroutine. t's // AddressSpace must be active. -func (t *Task) rseqCopyOutCPU() error { +func (t *Task) rseqUpdateCPU() error { + if t.rseqAddr == 0 && t.oldRSeqCPUAddr == 0 { + t.rseqCPU = -1 + return nil + } + + t.rseqCPU = int32(hostcpu.GetCPU()) + + // Update both CPUs, even if one fails. + rerr := t.rseqCopyOutCPU() + oerr := t.oldRSeqCopyOutCPU() + + if rerr != nil { + return rerr + } + return oerr +} + +// Preconditions: The caller must be running on the task goroutine. t's +// AddressSpace must be active. +func (t *Task) oldRSeqCopyOutCPU() error { + if t.oldRSeqCPUAddr == 0 { + return nil + } + buf := t.CopyScratchBuffer(4) usermem.ByteOrder.PutUint32(buf, uint32(t.rseqCPU)) - _, err := t.CopyOutBytes(t.rseqCPUAddr, buf) + _, err := t.CopyOutBytes(t.oldRSeqCPUAddr, buf) + return err +} + +// Preconditions: The caller must be running on the task goroutine. t's +// AddressSpace must be active. +func (t *Task) rseqCopyOutCPU() error { + if t.rseqAddr == 0 { + return nil + } + + buf := t.CopyScratchBuffer(8) + // CPUIDStart and CPUID are the first two fields in linux.RSeq. + usermem.ByteOrder.PutUint32(buf, uint32(t.rseqCPU)) // CPUIDStart + usermem.ByteOrder.PutUint32(buf[4:], uint32(t.rseqCPU)) // CPUID + // N.B. This write is not atomic, but since this occurs on the task + // goroutine then as long as userspace uses a single-instruction read + // it can't see an invalid value. + _, err := t.CopyOutBytes(t.rseqAddr, buf) + return err +} + +// Preconditions: The caller must be running on the task goroutine. t's +// AddressSpace must be active. +func (t *Task) rseqClearCPU() error { + buf := t.CopyScratchBuffer(8) + // CPUIDStart and CPUID are the first two fields in linux.RSeq. + usermem.ByteOrder.PutUint32(buf, 0) // CPUIDStart + usermem.ByteOrder.PutUint32(buf[4:], linux.RSEQ_CPU_ID_UNINITIALIZED) // CPUID + // N.B. This write is not atomic, but since this occurs on the task + // goroutine then as long as userspace uses a single-instruction read + // it can't see an invalid value. + _, err := t.CopyOutBytes(t.rseqAddr, buf) return err } +// rseqAddrInterrupt checks if IP is in a critical section, and aborts if so. +// +// This is a bit complex since both the RSeq and RSeqCriticalSection structs +// are stored in userspace. So we must: +// +// 1. Copy in the address of RSeqCriticalSection from RSeq. +// 2. Copy in RSeqCriticalSection itself. +// 3. Validate critical section struct version, address range, abort address. +// 4. Validate the abort signature (4 bytes preceding abort IP match expected +// signature). +// 5. Clear address of RSeqCriticalSection from RSeq. +// 6. Finally, conditionally abort. +// +// See kernel/rseq.c:rseq_ip_fixup for reference. +// +// Preconditions: The caller must be running on the task goroutine. t's +// AddressSpace must be active. +func (t *Task) rseqAddrInterrupt() { + if t.rseqAddr == 0 { + return + } + + critAddrAddr, ok := t.rseqAddr.AddLength(linux.OffsetOfRSeqCriticalSection) + if !ok { + // SetRSeq should validate this. + panic(fmt.Sprintf("t.rseqAddr (%#x) not large enough", t.rseqAddr)) + } + + if t.Arch().Width() != 8 { + // We only handle 64-bit for now. + t.Debugf("Only 64-bit rseq supported.") + t.forceSignal(linux.SIGSEGV, false /* unconditional */) + t.SendSignal(SignalInfoPriv(linux.SIGSEGV)) + return + } + + buf := t.CopyScratchBuffer(8) + if _, err := t.CopyInBytes(critAddrAddr, buf); err != nil { + t.Debugf("Failed to copy critical section address from %#x for rseq: %v", critAddrAddr, err) + t.forceSignal(linux.SIGSEGV, false /* unconditional */) + t.SendSignal(SignalInfoPriv(linux.SIGSEGV)) + return + } + + critAddr := usermem.Addr(usermem.ByteOrder.Uint64(buf)) + if critAddr == 0 { + return + } + + buf = t.CopyScratchBuffer(linux.SizeOfRSeqCriticalSection) + if _, err := t.CopyInBytes(critAddr, buf); err != nil { + t.Debugf("Failed to copy critical section from %#x for rseq: %v", critAddr, err) + t.forceSignal(linux.SIGSEGV, false /* unconditional */) + t.SendSignal(SignalInfoPriv(linux.SIGSEGV)) + return + } + + // Manually marshal RSeqCriticalSection as this is in the hot path when + // rseq is enabled. It must be as fast as possible. + // + // TODO(b/130243041): Replace with go_marshal. + cs := linux.RSeqCriticalSection{ + Version: usermem.ByteOrder.Uint32(buf[0:4]), + Flags: usermem.ByteOrder.Uint32(buf[4:8]), + Start: usermem.ByteOrder.Uint64(buf[8:16]), + PostCommitOffset: usermem.ByteOrder.Uint64(buf[16:24]), + Abort: usermem.ByteOrder.Uint64(buf[24:32]), + } + + if cs.Version != 0 { + t.Debugf("Unknown version in %+v", cs) + t.forceSignal(linux.SIGSEGV, false /* unconditional */) + t.SendSignal(SignalInfoPriv(linux.SIGSEGV)) + return + } + + start := usermem.Addr(cs.Start) + critRange, ok := start.ToRange(cs.PostCommitOffset) + if !ok { + t.Debugf("Invalid start and offset in %+v", cs) + t.forceSignal(linux.SIGSEGV, false /* unconditional */) + t.SendSignal(SignalInfoPriv(linux.SIGSEGV)) + return + } + + abort := usermem.Addr(cs.Abort) + if critRange.Contains(abort) { + t.Debugf("Abort in critical section in %+v", cs) + t.forceSignal(linux.SIGSEGV, false /* unconditional */) + t.SendSignal(SignalInfoPriv(linux.SIGSEGV)) + return + } + + // Verify signature. + sigAddr := abort - linux.SizeOfRSeqSignature + + buf = t.CopyScratchBuffer(linux.SizeOfRSeqSignature) + if _, err := t.CopyInBytes(sigAddr, buf); err != nil { + t.Debugf("Failed to copy critical section signature from %#x for rseq: %v", sigAddr, err) + t.forceSignal(linux.SIGSEGV, false /* unconditional */) + t.SendSignal(SignalInfoPriv(linux.SIGSEGV)) + return + } + + sig := usermem.ByteOrder.Uint32(buf) + if sig != t.rseqSignature { + t.Debugf("Mismatched rseq signature %d != %d", sig, t.rseqSignature) + t.forceSignal(linux.SIGSEGV, false /* unconditional */) + t.SendSignal(SignalInfoPriv(linux.SIGSEGV)) + return + } + + // Clear the critical section address. + // + // NOTE(b/143949567): We don't support any rseq flags, so we always + // restart if we are in the critical section, and thus *always* clear + // critAddrAddr. + if _, err := t.MemoryManager().ZeroOut(t, critAddrAddr, int64(t.Arch().Width()), usermem.IOOpts{ + AddressSpaceActive: true, + }); err != nil { + t.Debugf("Failed to clear critical section address from %#x for rseq: %v", critAddrAddr, err) + t.forceSignal(linux.SIGSEGV, false /* unconditional */) + t.SendSignal(SignalInfoPriv(linux.SIGSEGV)) + return + } + + // Finally we can actually decide whether or not to restart. + if !critRange.Contains(usermem.Addr(t.Arch().IP())) { + return + } + + t.Arch().SetIP(uintptr(cs.Abort)) +} + // Preconditions: The caller must be running on the task goroutine. -func (t *Task) rseqInterrupt() { - rscr := t.tg.rscr.Load().(*RSEQCriticalRegion) - if ip := t.Arch().IP(); rscr.CriticalSection.Contains(usermem.Addr(ip)) { - t.Debugf("Interrupted RSEQ critical section at %#x; restarting at %#x", ip, rscr.Restart) - t.Arch().SetIP(uintptr(rscr.Restart)) - t.Arch().SetRSEQInterruptedIP(ip) +func (t *Task) oldRSeqInterrupt() { + r := t.tg.oldRSeqCritical.Load().(*OldRSeqCriticalRegion) + if ip := t.Arch().IP(); r.CriticalSection.Contains(usermem.Addr(ip)) { + t.Debugf("Interrupted rseq critical section at %#x; restarting at %#x", ip, r.Restart) + t.Arch().SetIP(uintptr(r.Restart)) + t.Arch().SetOldRSeqInterruptedIP(ip) } } + +// Preconditions: The caller must be running on the task goroutine. +func (t *Task) rseqInterrupt() { + t.rseqAddrInterrupt() + t.oldRSeqInterrupt() +} diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index ab0c6c4aa..d25a7903b 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -489,18 +489,43 @@ type Task struct { // netns is protected by mu. netns is owned by the task goroutine. netns bool - // If rseqPreempted is true, before the next call to p.Switch(), interrupt - // RSEQ critical regions as defined by tg.rseq and write the task - // goroutine's CPU number to rseqCPUAddr. rseqCPU is the last CPU number - // written to rseqCPUAddr. + // If rseqPreempted is true, before the next call to p.Switch(), + // interrupt rseq critical regions as defined by rseqAddr and + // tg.oldRSeqCritical and write the task goroutine's CPU number to + // rseqAddr/oldRSeqCPUAddr. // - // If rseqCPUAddr is 0, rseqCPU is -1. + // We support two ABIs for restartable sequences: // - // rseqCPUAddr, rseqCPU, and rseqPreempted are exclusive to the task - // goroutine. + // 1. The upstream interface added in v4.18, + // 2. An "old" interface never merged upstream. In the implementation, + // this is referred to as "old rseq". + // + // rseqPreempted is exclusive to the task goroutine. rseqPreempted bool `state:"nosave"` - rseqCPUAddr usermem.Addr - rseqCPU int32 + + // rseqCPU is the last CPU number written to rseqAddr/oldRSeqCPUAddr. + // + // If rseq is unused, rseqCPU is -1 for convenient use in + // platform.Context.Switch. + // + // rseqCPU is exclusive to the task goroutine. + rseqCPU int32 + + // oldRSeqCPUAddr is a pointer to the userspace old rseq CPU variable. + // + // oldRSeqCPUAddr is exclusive to the task goroutine. + oldRSeqCPUAddr usermem.Addr + + // rseqAddr is a pointer to the userspace linux.RSeq structure. + // + // rseqAddr is exclusive to the task goroutine. + rseqAddr usermem.Addr + + // rseqSignature is the signature that the rseq abort IP must be signed + // with. + // + // rseqSignature is exclusive to the task goroutine. + rseqSignature uint32 // copyScratchBuffer is a buffer available to CopyIn/CopyOut // implementations that require an intermediate buffer to copy data diff --git a/pkg/sentry/kernel/task_clone.go b/pkg/sentry/kernel/task_clone.go index 5f3589493..247bd4aba 100644 --- a/pkg/sentry/kernel/task_clone.go +++ b/pkg/sentry/kernel/task_clone.go @@ -236,7 +236,10 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { } else if opts.NewPIDNamespace { pidns = pidns.NewChild(userns) } + tg := t.tg + rseqAddr := usermem.Addr(0) + rseqSignature := uint32(0) if opts.NewThreadGroup { tg.mounts.IncRef() sh := t.tg.signalHandlers @@ -244,6 +247,8 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { sh = sh.Fork() } tg = t.k.NewThreadGroup(tg.mounts, pidns, sh, opts.TerminationSignal, tg.limits.GetCopy()) + rseqAddr = t.rseqAddr + rseqSignature = t.rseqSignature } cfg := &TaskConfig{ @@ -260,6 +265,8 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { UTSNamespace: utsns, IPCNamespace: ipcns, AbstractSocketNamespace: t.abstractSockets, + RSeqAddr: rseqAddr, + RSeqSignature: rseqSignature, ContainerID: t.ContainerID(), } if opts.NewThreadGroup { diff --git a/pkg/sentry/kernel/task_exec.go b/pkg/sentry/kernel/task_exec.go index 90a6190f1..fa6528386 100644 --- a/pkg/sentry/kernel/task_exec.go +++ b/pkg/sentry/kernel/task_exec.go @@ -190,9 +190,11 @@ func (r *runSyscallAfterExecStop) execute(t *Task) taskRunState { t.updateRSSLocked() // Restartable sequence state is discarded. t.rseqPreempted = false - t.rseqCPUAddr = 0 t.rseqCPU = -1 - t.tg.rscr.Store(&RSEQCriticalRegion{}) + t.rseqAddr = 0 + t.rseqSignature = 0 + t.oldRSeqCPUAddr = 0 + t.tg.oldRSeqCritical.Store(&OldRSeqCriticalRegion{}) t.tg.pidns.owner.mu.Unlock() // Remove FDs with the CloseOnExec flag set. diff --git a/pkg/sentry/kernel/task_run.go b/pkg/sentry/kernel/task_run.go index d97f8c189..6357273d3 100644 --- a/pkg/sentry/kernel/task_run.go +++ b/pkg/sentry/kernel/task_run.go @@ -169,12 +169,22 @@ func (*runApp) execute(t *Task) taskRunState { // Apply restartable sequences. if t.rseqPreempted { t.rseqPreempted = false - if t.rseqCPUAddr != 0 { + if t.rseqAddr != 0 || t.oldRSeqCPUAddr != 0 { + // Linux writes the CPU on every preemption. We only do + // so if it changed. Thus we may delay delivery of + // SIGSEGV if rseqAddr/oldRSeqCPUAddr is invalid. cpu := int32(hostcpu.GetCPU()) if t.rseqCPU != cpu { t.rseqCPU = cpu if err := t.rseqCopyOutCPU(); err != nil { - t.Warningf("Failed to copy CPU to %#x for RSEQ: %v", t.rseqCPUAddr, err) + t.Debugf("Failed to copy CPU to %#x for rseq: %v", t.rseqAddr, err) + t.forceSignal(linux.SIGSEGV, false) + t.SendSignal(SignalInfoPriv(linux.SIGSEGV)) + // Re-enter the task run loop for signal delivery. + return (*runApp)(nil) + } + if err := t.oldRSeqCopyOutCPU(); err != nil { + t.Debugf("Failed to copy CPU to %#x for old rseq: %v", t.oldRSeqCPUAddr, err) t.forceSignal(linux.SIGSEGV, false) t.SendSignal(SignalInfoPriv(linux.SIGSEGV)) // Re-enter the task run loop for signal delivery. @@ -320,7 +330,7 @@ func (*runApp) execute(t *Task) taskRunState { return (*runApp)(nil) case platform.ErrContextCPUPreempted: - // Ensure that RSEQ critical sections are interrupted and per-thread + // Ensure that rseq critical sections are interrupted and per-thread // CPU values are updated before the next platform.Context.Switch(). t.rseqPreempted = true return (*runApp)(nil) diff --git a/pkg/sentry/kernel/task_start.go b/pkg/sentry/kernel/task_start.go index 3522a4ae5..58af16ee2 100644 --- a/pkg/sentry/kernel/task_start.go +++ b/pkg/sentry/kernel/task_start.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/futex" "gvisor.dev/gvisor/pkg/sentry/kernel/sched" "gvisor.dev/gvisor/pkg/sentry/usage" + "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" ) @@ -79,6 +80,13 @@ type TaskConfig struct { // AbstractSocketNamespace is the AbstractSocketNamespace of the new task. AbstractSocketNamespace *AbstractSocketNamespace + // RSeqAddr is a pointer to the the userspace linux.RSeq structure. + RSeqAddr usermem.Addr + + // RSeqSignature is the signature that the rseq abort IP must be signed + // with. + RSeqSignature uint32 + // ContainerID is the container the new task belongs to. ContainerID string } @@ -126,6 +134,8 @@ func (ts *TaskSet) newTask(cfg *TaskConfig) (*Task, error) { ipcns: cfg.IPCNamespace, abstractSockets: cfg.AbstractSocketNamespace, rseqCPU: -1, + rseqAddr: cfg.RSeqAddr, + rseqSignature: cfg.RSeqSignature, futexWaiter: futex.NewWaiter(), containerID: cfg.ContainerID, } diff --git a/pkg/sentry/kernel/thread_group.go b/pkg/sentry/kernel/thread_group.go index 0cded73f6..c0197a563 100644 --- a/pkg/sentry/kernel/thread_group.go +++ b/pkg/sentry/kernel/thread_group.go @@ -238,8 +238,8 @@ type ThreadGroup struct { // execed is protected by the TaskSet mutex. execed bool - // rscr is the thread group's RSEQ critical region. - rscr atomic.Value `state:".(*RSEQCriticalRegion)"` + // oldRSeqCritical is the thread group's old rseq critical region. + oldRSeqCritical atomic.Value `state:".(*OldRSeqCriticalRegion)"` // mounts is the thread group's mount namespace. This does not really // correspond to a "mount namespace" in Linux, but is more like a @@ -273,18 +273,18 @@ func (k *Kernel) NewThreadGroup(mntns *fs.MountNamespace, pidns *PIDNamespace, s } tg.itimerRealTimer = ktime.NewTimer(k.monotonicClock, &itimerRealListener{tg: tg}) tg.timers = make(map[linux.TimerID]*IntervalTimer) - tg.rscr.Store(&RSEQCriticalRegion{}) + tg.oldRSeqCritical.Store(&OldRSeqCriticalRegion{}) return tg } -// saveRscr is invoked by stateify. -func (tg *ThreadGroup) saveRscr() *RSEQCriticalRegion { - return tg.rscr.Load().(*RSEQCriticalRegion) +// saveOldRSeqCritical is invoked by stateify. +func (tg *ThreadGroup) saveOldRSeqCritical() *OldRSeqCriticalRegion { + return tg.oldRSeqCritical.Load().(*OldRSeqCriticalRegion) } -// loadRscr is invoked by stateify. -func (tg *ThreadGroup) loadRscr(rscr *RSEQCriticalRegion) { - tg.rscr.Store(rscr) +// loadOldRSeqCritical is invoked by stateify. +func (tg *ThreadGroup) loadOldRSeqCritical(r *OldRSeqCriticalRegion) { + tg.oldRSeqCritical.Store(r) } // SignalHandlers returns the signal handlers used by tg. diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD index 6766ba587..a76975cee 100644 --- a/pkg/sentry/syscalls/linux/BUILD +++ b/pkg/sentry/syscalls/linux/BUILD @@ -30,6 +30,7 @@ go_library( "sys_random.go", "sys_read.go", "sys_rlimit.go", + "sys_rseq.go", "sys_rusage.go", "sys_sched.go", "sys_seccomp.go", diff --git a/pkg/sentry/syscalls/linux/linux64_amd64.go b/pkg/sentry/syscalls/linux/linux64_amd64.go index 272ae9991..479c5f6ff 100644 --- a/pkg/sentry/syscalls/linux/linux64_amd64.go +++ b/pkg/sentry/syscalls/linux/linux64_amd64.go @@ -377,7 +377,7 @@ var AMD64 = &kernel.SyscallTable{ 331: syscalls.ErrorWithEvent("pkey_free", syserror.ENOSYS, "", nil), 332: syscalls.Supported("statx", Statx), 333: syscalls.ErrorWithEvent("io_pgetevents", syserror.ENOSYS, "", nil), - 334: syscalls.ErrorWithEvent("rseq", syserror.ENOSYS, "", nil), + 334: syscalls.PartiallySupported("rseq", RSeq, "Not supported on all platforms.", nil), // Linux skips ahead to syscall 424 to sync numbers between arches. 424: syscalls.ErrorWithEvent("pidfd_send_signal", syserror.ENOSYS, "", nil), diff --git a/pkg/sentry/syscalls/linux/linux64_arm64.go b/pkg/sentry/syscalls/linux/linux64_arm64.go index 3b584eed9..d3f61f5e8 100644 --- a/pkg/sentry/syscalls/linux/linux64_arm64.go +++ b/pkg/sentry/syscalls/linux/linux64_arm64.go @@ -307,7 +307,7 @@ var ARM64 = &kernel.SyscallTable{ 290: syscalls.ErrorWithEvent("pkey_free", syserror.ENOSYS, "", nil), 291: syscalls.Supported("statx", Statx), 292: syscalls.ErrorWithEvent("io_pgetevents", syserror.ENOSYS, "", nil), - 293: syscalls.ErrorWithEvent("rseq", syserror.ENOSYS, "", nil), + 293: syscalls.PartiallySupported("rseq", RSeq, "Not supported on all platforms.", nil), // Linux skips ahead to syscall 424 to sync numbers between arches. 424: syscalls.ErrorWithEvent("pidfd_send_signal", syserror.ENOSYS, "", nil), diff --git a/pkg/sentry/syscalls/linux/sys_rseq.go b/pkg/sentry/syscalls/linux/sys_rseq.go new file mode 100644 index 000000000..90db10ea6 --- /dev/null +++ b/pkg/sentry/syscalls/linux/sys_rseq.go @@ -0,0 +1,48 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package linux + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/syserror" +) + +// RSeq implements syscall rseq(2). +func RSeq(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + addr := args[0].Pointer() + length := args[1].Uint() + flags := args[2].Int() + signature := args[3].Uint() + + if !t.RSeqAvailable() { + // Event for applications that want rseq on a configuration + // that doesn't support them. + t.Kernel().EmitUnimplementedEvent(t) + return 0, nil, syserror.ENOSYS + } + + switch flags { + case 0: + // Register. + return 0, nil, t.SetRSeq(addr, length, signature) + case linux.RSEQ_FLAG_UNREGISTER: + return 0, nil, t.ClearRSeq(addr, length, signature) + default: + // Unknown flag. + return 0, nil, syserror.EINVAL + } +} -- cgit v1.2.3 From 51f3ab85e024fcd74c49d273ce5202a207577d31 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Mon, 6 Jan 2020 12:51:35 -0800 Subject: Convert memfs into proto-tmpfs. - Renamed memfs to tmpfs. - Copied fileRangeSet bits from fs/fsutil/ to fsimpl/tmpfs/ - Changed tmpfs to be backed by filemem instead of byte slice. - regularFileReadWriter uses a sync.Pool, similar to gofer client. PiperOrigin-RevId: 288356380 --- pkg/sentry/fs/fsutil/BUILD | 2 +- pkg/sentry/fs/fsutil/file_range_set.go | 14 +- pkg/sentry/fsimpl/memfs/BUILD | 80 --- pkg/sentry/fsimpl/memfs/benchmark_test.go | 487 ------------------- pkg/sentry/fsimpl/memfs/directory.go | 187 ------- pkg/sentry/fsimpl/memfs/filesystem.go | 698 --------------------------- pkg/sentry/fsimpl/memfs/memfs.go | 293 ----------- pkg/sentry/fsimpl/memfs/named_pipe.go | 60 --- pkg/sentry/fsimpl/memfs/pipe_test.go | 235 --------- pkg/sentry/fsimpl/memfs/regular_file.go | 154 ------ pkg/sentry/fsimpl/memfs/symlink.go | 36 -- pkg/sentry/fsimpl/tmpfs/BUILD | 92 ++++ pkg/sentry/fsimpl/tmpfs/benchmark_test.go | 487 +++++++++++++++++++ pkg/sentry/fsimpl/tmpfs/directory.go | 187 +++++++ pkg/sentry/fsimpl/tmpfs/filesystem.go | 698 +++++++++++++++++++++++++++ pkg/sentry/fsimpl/tmpfs/named_pipe.go | 60 +++ pkg/sentry/fsimpl/tmpfs/pipe_test.go | 235 +++++++++ pkg/sentry/fsimpl/tmpfs/regular_file.go | 357 ++++++++++++++ pkg/sentry/fsimpl/tmpfs/regular_file_test.go | 224 +++++++++ pkg/sentry/fsimpl/tmpfs/symlink.go | 36 ++ pkg/sentry/fsimpl/tmpfs/tmpfs.go | 299 ++++++++++++ 21 files changed, 2683 insertions(+), 2238 deletions(-) delete mode 100644 pkg/sentry/fsimpl/memfs/BUILD delete mode 100644 pkg/sentry/fsimpl/memfs/benchmark_test.go delete mode 100644 pkg/sentry/fsimpl/memfs/directory.go delete mode 100644 pkg/sentry/fsimpl/memfs/filesystem.go delete mode 100644 pkg/sentry/fsimpl/memfs/memfs.go delete mode 100644 pkg/sentry/fsimpl/memfs/named_pipe.go delete mode 100644 pkg/sentry/fsimpl/memfs/pipe_test.go delete mode 100644 pkg/sentry/fsimpl/memfs/regular_file.go delete mode 100644 pkg/sentry/fsimpl/memfs/symlink.go create mode 100644 pkg/sentry/fsimpl/tmpfs/BUILD create mode 100644 pkg/sentry/fsimpl/tmpfs/benchmark_test.go create mode 100644 pkg/sentry/fsimpl/tmpfs/directory.go create mode 100644 pkg/sentry/fsimpl/tmpfs/filesystem.go create mode 100644 pkg/sentry/fsimpl/tmpfs/named_pipe.go create mode 100644 pkg/sentry/fsimpl/tmpfs/pipe_test.go create mode 100644 pkg/sentry/fsimpl/tmpfs/regular_file.go create mode 100644 pkg/sentry/fsimpl/tmpfs/regular_file_test.go create mode 100644 pkg/sentry/fsimpl/tmpfs/symlink.go create mode 100644 pkg/sentry/fsimpl/tmpfs/tmpfs.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/fsutil/BUILD b/pkg/sentry/fs/fsutil/BUILD index b2e8d9c77..9ca695a95 100644 --- a/pkg/sentry/fs/fsutil/BUILD +++ b/pkg/sentry/fs/fsutil/BUILD @@ -53,7 +53,7 @@ go_template_instance( "Key": "uint64", "Range": "memmap.MappableRange", "Value": "uint64", - "Functions": "fileRangeSetFunctions", + "Functions": "FileRangeSetFunctions", }, ) diff --git a/pkg/sentry/fs/fsutil/file_range_set.go b/pkg/sentry/fs/fsutil/file_range_set.go index 0a5466b0a..f52d712e3 100644 --- a/pkg/sentry/fs/fsutil/file_range_set.go +++ b/pkg/sentry/fs/fsutil/file_range_set.go @@ -34,25 +34,25 @@ import ( // // type FileRangeSet -// fileRangeSetFunctions implements segment.Functions for FileRangeSet. -type fileRangeSetFunctions struct{} +// FileRangeSetFunctions implements segment.Functions for FileRangeSet. +type FileRangeSetFunctions struct{} // MinKey implements segment.Functions.MinKey. -func (fileRangeSetFunctions) MinKey() uint64 { +func (FileRangeSetFunctions) MinKey() uint64 { return 0 } // MaxKey implements segment.Functions.MaxKey. -func (fileRangeSetFunctions) MaxKey() uint64 { +func (FileRangeSetFunctions) MaxKey() uint64 { return math.MaxUint64 } // ClearValue implements segment.Functions.ClearValue. -func (fileRangeSetFunctions) ClearValue(_ *uint64) { +func (FileRangeSetFunctions) ClearValue(_ *uint64) { } // Merge implements segment.Functions.Merge. -func (fileRangeSetFunctions) Merge(mr1 memmap.MappableRange, frstart1 uint64, _ memmap.MappableRange, frstart2 uint64) (uint64, bool) { +func (FileRangeSetFunctions) Merge(mr1 memmap.MappableRange, frstart1 uint64, _ memmap.MappableRange, frstart2 uint64) (uint64, bool) { if frstart1+mr1.Length() != frstart2 { return 0, false } @@ -60,7 +60,7 @@ func (fileRangeSetFunctions) Merge(mr1 memmap.MappableRange, frstart1 uint64, _ } // Split implements segment.Functions.Split. -func (fileRangeSetFunctions) Split(mr memmap.MappableRange, frstart uint64, split uint64) (uint64, uint64) { +func (FileRangeSetFunctions) Split(mr memmap.MappableRange, frstart uint64, split uint64) (uint64, uint64) { return frstart, frstart + (split - mr.Start) } diff --git a/pkg/sentry/fsimpl/memfs/BUILD b/pkg/sentry/fsimpl/memfs/BUILD deleted file mode 100644 index 5689bed3b..000000000 --- a/pkg/sentry/fsimpl/memfs/BUILD +++ /dev/null @@ -1,80 +0,0 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") - -package(licenses = ["notice"]) - -load("//tools/go_generics:defs.bzl", "go_template_instance") - -go_template_instance( - name = "dentry_list", - out = "dentry_list.go", - package = "memfs", - prefix = "dentry", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*dentry", - "Linker": "*dentry", - }, -) - -go_library( - name = "memfs", - srcs = [ - "dentry_list.go", - "directory.go", - "filesystem.go", - "memfs.go", - "named_pipe.go", - "regular_file.go", - "symlink.go", - ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fsimpl/memfs", - deps = [ - "//pkg/abi/linux", - "//pkg/amutex", - "//pkg/fspath", - "//pkg/sentry/arch", - "//pkg/sentry/context", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/pipe", - "//pkg/sentry/usermem", - "//pkg/sentry/vfs", - "//pkg/syserror", - ], -) - -go_test( - name = "benchmark_test", - size = "small", - srcs = ["benchmark_test.go"], - deps = [ - ":memfs", - "//pkg/abi/linux", - "//pkg/fspath", - "//pkg/refs", - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", - "//pkg/sentry/fs", - "//pkg/sentry/fs/tmpfs", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/vfs", - "//pkg/syserror", - ], -) - -go_test( - name = "memfs_test", - size = "small", - srcs = ["pipe_test.go"], - embed = [":memfs"], - deps = [ - "//pkg/abi/linux", - "//pkg/fspath", - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/usermem", - "//pkg/sentry/vfs", - "//pkg/syserror", - ], -) diff --git a/pkg/sentry/fsimpl/memfs/benchmark_test.go b/pkg/sentry/fsimpl/memfs/benchmark_test.go deleted file mode 100644 index a27876a4e..000000000 --- a/pkg/sentry/fsimpl/memfs/benchmark_test.go +++ /dev/null @@ -1,487 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package benchmark_test - -import ( - "fmt" - "runtime" - "strings" - "testing" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" - _ "gvisor.dev/gvisor/pkg/sentry/fs/tmpfs" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/memfs" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -// Differences from stat_benchmark: -// -// - Syscall interception, CopyInPath, copyOutStat, and overlayfs overheads are -// not included. -// -// - *MountStat benchmarks use a tmpfs root mount and a tmpfs submount at /tmp. -// Non-MountStat benchmarks use a tmpfs root mount and no submounts. -// stat_benchmark uses a varying root mount, a tmpfs submount at /tmp, and a -// subdirectory /tmp/ (assuming TEST_TMPDIR == "/tmp"). Thus -// stat_benchmark at depth 1 does a comparable amount of work to *MountStat -// benchmarks at depth 2, and non-MountStat benchmarks at depth 3. -var depths = []int{1, 2, 3, 8, 64, 100} - -const ( - mountPointName = "tmp" - filename = "gvisor_test_temp_0_1557494568" -) - -// This is copied from syscalls/linux/sys_file.go, with the dependency on -// kernel.Task stripped out. -func fileOpOn(ctx context.Context, mntns *fs.MountNamespace, root, wd *fs.Dirent, dirFD int32, path string, resolve bool, fn func(root *fs.Dirent, d *fs.Dirent) error) error { - var ( - d *fs.Dirent // The file. - rel *fs.Dirent // The relative directory for search (if required.) - err error - ) - - // Extract the working directory (maybe). - if len(path) > 0 && path[0] == '/' { - // Absolute path; rel can be nil. - } else if dirFD == linux.AT_FDCWD { - // Need to reference the working directory. - rel = wd - } else { - // Need to extract the given FD. - return syserror.EBADF - } - - // Lookup the node. - remainingTraversals := uint(linux.MaxSymlinkTraversals) - if resolve { - d, err = mntns.FindInode(ctx, root, rel, path, &remainingTraversals) - } else { - d, err = mntns.FindLink(ctx, root, rel, path, &remainingTraversals) - } - if err != nil { - return err - } - - err = fn(root, d) - d.DecRef() - return err -} - -func BenchmarkVFS1TmpfsStat(b *testing.B) { - for _, depth := range depths { - b.Run(fmt.Sprintf("%d", depth), func(b *testing.B) { - ctx := contexttest.Context(b) - - // Create VFS. - tmpfsFS, ok := fs.FindFilesystem("tmpfs") - if !ok { - b.Fatalf("failed to find tmpfs filesystem type") - } - rootInode, err := tmpfsFS.Mount(ctx, "tmpfs", fs.MountSourceFlags{}, "", nil) - if err != nil { - b.Fatalf("failed to create tmpfs root mount: %v", err) - } - mntns, err := fs.NewMountNamespace(ctx, rootInode) - if err != nil { - b.Fatalf("failed to create mount namespace: %v", err) - } - defer mntns.DecRef() - - var filePathBuilder strings.Builder - filePathBuilder.WriteByte('/') - - // Create nested directories with given depth. - root := mntns.Root() - defer root.DecRef() - d := root - d.IncRef() - defer d.DecRef() - for i := depth; i > 0; i-- { - name := fmt.Sprintf("%d", i) - if err := d.Inode.CreateDirectory(ctx, d, name, fs.FilePermsFromMode(0755)); err != nil { - b.Fatalf("failed to create directory %q: %v", name, err) - } - next, err := d.Walk(ctx, root, name) - if err != nil { - b.Fatalf("failed to walk to directory %q: %v", name, err) - } - d.DecRef() - d = next - filePathBuilder.WriteString(name) - filePathBuilder.WriteByte('/') - } - - // Create the file that will be stat'd. - file, err := d.Inode.Create(ctx, d, filename, fs.FileFlags{Read: true, Write: true}, fs.FilePermsFromMode(0644)) - if err != nil { - b.Fatalf("failed to create file %q: %v", filename, err) - } - file.DecRef() - filePathBuilder.WriteString(filename) - filePath := filePathBuilder.String() - - dirPath := false - runtime.GC() - b.ResetTimer() - for i := 0; i < b.N; i++ { - err := fileOpOn(ctx, mntns, root, root, linux.AT_FDCWD, filePath, true /* resolve */, func(root *fs.Dirent, d *fs.Dirent) error { - if dirPath && !fs.IsDir(d.Inode.StableAttr) { - return syserror.ENOTDIR - } - uattr, err := d.Inode.UnstableAttr(ctx) - if err != nil { - return err - } - // Sanity check. - if uattr.Perms.User.Execute { - b.Fatalf("got wrong permissions (%0o)", uattr.Perms.LinuxMode()) - } - return nil - }) - if err != nil { - b.Fatalf("stat(%q) failed: %v", filePath, err) - } - } - // Don't include deferred cleanup in benchmark time. - b.StopTimer() - }) - } -} - -func BenchmarkVFS2MemfsStat(b *testing.B) { - for _, depth := range depths { - b.Run(fmt.Sprintf("%d", depth), func(b *testing.B) { - ctx := contexttest.Context(b) - creds := auth.CredentialsFromContext(ctx) - - // Create VFS. - vfsObj := vfs.New() - vfsObj.MustRegisterFilesystemType("memfs", memfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "memfs", &vfs.GetFilesystemOptions{}) - if err != nil { - b.Fatalf("failed to create tmpfs root mount: %v", err) - } - defer mntns.DecRef(vfsObj) - - var filePathBuilder strings.Builder - filePathBuilder.WriteByte('/') - - // Create nested directories with given depth. - root := mntns.Root() - defer root.DecRef() - vd := root - vd.IncRef() - for i := depth; i > 0; i-- { - name := fmt.Sprintf("%d", i) - pop := vfs.PathOperation{ - Root: root, - Start: vd, - Path: fspath.Parse(name), - } - if err := vfsObj.MkdirAt(ctx, creds, &pop, &vfs.MkdirOptions{ - Mode: 0755, - }); err != nil { - b.Fatalf("failed to create directory %q: %v", name, err) - } - nextVD, err := vfsObj.GetDentryAt(ctx, creds, &pop, &vfs.GetDentryOptions{}) - if err != nil { - b.Fatalf("failed to walk to directory %q: %v", name, err) - } - vd.DecRef() - vd = nextVD - filePathBuilder.WriteString(name) - filePathBuilder.WriteByte('/') - } - - // Create the file that will be stat'd. - fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: vd, - Path: fspath.Parse(filename), - FollowFinalSymlink: true, - }, &vfs.OpenOptions{ - Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL, - Mode: 0644, - }) - vd.DecRef() - vd = vfs.VirtualDentry{} - if err != nil { - b.Fatalf("failed to create file %q: %v", filename, err) - } - defer fd.DecRef() - filePathBuilder.WriteString(filename) - filePath := filePathBuilder.String() - - runtime.GC() - b.ResetTimer() - for i := 0; i < b.N; i++ { - stat, err := vfsObj.StatAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(filePath), - FollowFinalSymlink: true, - }, &vfs.StatOptions{}) - if err != nil { - b.Fatalf("stat(%q) failed: %v", filePath, err) - } - // Sanity check. - if stat.Mode&^linux.S_IFMT != 0644 { - b.Fatalf("got wrong permissions (%0o)", stat.Mode) - } - } - // Don't include deferred cleanup in benchmark time. - b.StopTimer() - }) - } -} - -func BenchmarkVFS1TmpfsMountStat(b *testing.B) { - for _, depth := range depths { - b.Run(fmt.Sprintf("%d", depth), func(b *testing.B) { - ctx := contexttest.Context(b) - - // Create VFS. - tmpfsFS, ok := fs.FindFilesystem("tmpfs") - if !ok { - b.Fatalf("failed to find tmpfs filesystem type") - } - rootInode, err := tmpfsFS.Mount(ctx, "tmpfs", fs.MountSourceFlags{}, "", nil) - if err != nil { - b.Fatalf("failed to create tmpfs root mount: %v", err) - } - mntns, err := fs.NewMountNamespace(ctx, rootInode) - if err != nil { - b.Fatalf("failed to create mount namespace: %v", err) - } - defer mntns.DecRef() - - var filePathBuilder strings.Builder - filePathBuilder.WriteByte('/') - - // Create and mount the submount. - root := mntns.Root() - defer root.DecRef() - if err := root.Inode.CreateDirectory(ctx, root, mountPointName, fs.FilePermsFromMode(0755)); err != nil { - b.Fatalf("failed to create mount point: %v", err) - } - mountPoint, err := root.Walk(ctx, root, mountPointName) - if err != nil { - b.Fatalf("failed to walk to mount point: %v", err) - } - defer mountPoint.DecRef() - submountInode, err := tmpfsFS.Mount(ctx, "tmpfs", fs.MountSourceFlags{}, "", nil) - if err != nil { - b.Fatalf("failed to create tmpfs submount: %v", err) - } - if err := mntns.Mount(ctx, mountPoint, submountInode); err != nil { - b.Fatalf("failed to mount tmpfs submount: %v", err) - } - filePathBuilder.WriteString(mountPointName) - filePathBuilder.WriteByte('/') - - // Create nested directories with given depth. - d, err := root.Walk(ctx, root, mountPointName) - if err != nil { - b.Fatalf("failed to walk to mount root: %v", err) - } - defer d.DecRef() - for i := depth; i > 0; i-- { - name := fmt.Sprintf("%d", i) - if err := d.Inode.CreateDirectory(ctx, d, name, fs.FilePermsFromMode(0755)); err != nil { - b.Fatalf("failed to create directory %q: %v", name, err) - } - next, err := d.Walk(ctx, root, name) - if err != nil { - b.Fatalf("failed to walk to directory %q: %v", name, err) - } - d.DecRef() - d = next - filePathBuilder.WriteString(name) - filePathBuilder.WriteByte('/') - } - - // Create the file that will be stat'd. - file, err := d.Inode.Create(ctx, d, filename, fs.FileFlags{Read: true, Write: true}, fs.FilePermsFromMode(0644)) - if err != nil { - b.Fatalf("failed to create file %q: %v", filename, err) - } - file.DecRef() - filePathBuilder.WriteString(filename) - filePath := filePathBuilder.String() - - dirPath := false - runtime.GC() - b.ResetTimer() - for i := 0; i < b.N; i++ { - err := fileOpOn(ctx, mntns, root, root, linux.AT_FDCWD, filePath, true /* resolve */, func(root *fs.Dirent, d *fs.Dirent) error { - if dirPath && !fs.IsDir(d.Inode.StableAttr) { - return syserror.ENOTDIR - } - uattr, err := d.Inode.UnstableAttr(ctx) - if err != nil { - return err - } - // Sanity check. - if uattr.Perms.User.Execute { - b.Fatalf("got wrong permissions (%0o)", uattr.Perms.LinuxMode()) - } - return nil - }) - if err != nil { - b.Fatalf("stat(%q) failed: %v", filePath, err) - } - } - // Don't include deferred cleanup in benchmark time. - b.StopTimer() - }) - } -} - -func BenchmarkVFS2MemfsMountStat(b *testing.B) { - for _, depth := range depths { - b.Run(fmt.Sprintf("%d", depth), func(b *testing.B) { - ctx := contexttest.Context(b) - creds := auth.CredentialsFromContext(ctx) - - // Create VFS. - vfsObj := vfs.New() - vfsObj.MustRegisterFilesystemType("memfs", memfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "memfs", &vfs.GetFilesystemOptions{}) - if err != nil { - b.Fatalf("failed to create tmpfs root mount: %v", err) - } - defer mntns.DecRef(vfsObj) - - var filePathBuilder strings.Builder - filePathBuilder.WriteByte('/') - - // Create the mount point. - root := mntns.Root() - defer root.DecRef() - pop := vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(mountPointName), - } - if err := vfsObj.MkdirAt(ctx, creds, &pop, &vfs.MkdirOptions{ - Mode: 0755, - }); err != nil { - b.Fatalf("failed to create mount point: %v", err) - } - // Save the mount point for later use. - mountPoint, err := vfsObj.GetDentryAt(ctx, creds, &pop, &vfs.GetDentryOptions{}) - if err != nil { - b.Fatalf("failed to walk to mount point: %v", err) - } - defer mountPoint.DecRef() - // Create and mount the submount. - if err := vfsObj.MountAt(ctx, creds, "", &pop, "memfs", &vfs.MountOptions{}); err != nil { - b.Fatalf("failed to mount tmpfs submount: %v", err) - } - filePathBuilder.WriteString(mountPointName) - filePathBuilder.WriteByte('/') - - // Create nested directories with given depth. - vd, err := vfsObj.GetDentryAt(ctx, creds, &pop, &vfs.GetDentryOptions{}) - if err != nil { - b.Fatalf("failed to walk to mount root: %v", err) - } - for i := depth; i > 0; i-- { - name := fmt.Sprintf("%d", i) - pop := vfs.PathOperation{ - Root: root, - Start: vd, - Path: fspath.Parse(name), - } - if err := vfsObj.MkdirAt(ctx, creds, &pop, &vfs.MkdirOptions{ - Mode: 0755, - }); err != nil { - b.Fatalf("failed to create directory %q: %v", name, err) - } - nextVD, err := vfsObj.GetDentryAt(ctx, creds, &pop, &vfs.GetDentryOptions{}) - if err != nil { - b.Fatalf("failed to walk to directory %q: %v", name, err) - } - vd.DecRef() - vd = nextVD - filePathBuilder.WriteString(name) - filePathBuilder.WriteByte('/') - } - - // Verify that we didn't create any directories under the mount - // point (i.e. they were all created on the submount). - firstDirName := fmt.Sprintf("%d", depth) - if child := mountPoint.Dentry().Child(firstDirName); child != nil { - b.Fatalf("created directory %q under root mount, not submount", firstDirName) - } - - // Create the file that will be stat'd. - fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: vd, - Path: fspath.Parse(filename), - FollowFinalSymlink: true, - }, &vfs.OpenOptions{ - Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL, - Mode: 0644, - }) - vd.DecRef() - if err != nil { - b.Fatalf("failed to create file %q: %v", filename, err) - } - fd.DecRef() - filePathBuilder.WriteString(filename) - filePath := filePathBuilder.String() - - runtime.GC() - b.ResetTimer() - for i := 0; i < b.N; i++ { - stat, err := vfsObj.StatAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(filePath), - FollowFinalSymlink: true, - }, &vfs.StatOptions{}) - if err != nil { - b.Fatalf("stat(%q) failed: %v", filePath, err) - } - // Sanity check. - if stat.Mode&^linux.S_IFMT != 0644 { - b.Fatalf("got wrong permissions (%0o)", stat.Mode) - } - } - // Don't include deferred cleanup in benchmark time. - b.StopTimer() - }) - } -} - -func init() { - // Turn off reference leak checking for a fair comparison between vfs1 and - // vfs2. - refs.SetLeakMode(refs.NoLeakChecking) -} diff --git a/pkg/sentry/fsimpl/memfs/directory.go b/pkg/sentry/fsimpl/memfs/directory.go deleted file mode 100644 index 0bd82e480..000000000 --- a/pkg/sentry/fsimpl/memfs/directory.go +++ /dev/null @@ -1,187 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package memfs - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -type directory struct { - inode inode - - // childList is a list containing (1) child Dentries and (2) fake Dentries - // (with inode == nil) that represent the iteration position of - // directoryFDs. childList is used to support directoryFD.IterDirents() - // efficiently. childList is protected by filesystem.mu. - childList dentryList -} - -func (fs *filesystem) newDirectory(creds *auth.Credentials, mode linux.FileMode) *inode { - dir := &directory{} - dir.inode.init(dir, fs, creds, mode) - dir.inode.nlink = 2 // from "." and parent directory or ".." for root - return &dir.inode -} - -func (i *inode) isDir() bool { - _, ok := i.impl.(*directory) - return ok -} - -type directoryFD struct { - fileDescription - vfs.DirectoryFileDescriptionDefaultImpl - - // Protected by filesystem.mu. - iter *dentry - off int64 -} - -// Release implements vfs.FileDescriptionImpl.Release. -func (fd *directoryFD) Release() { - if fd.iter != nil { - fs := fd.filesystem() - dir := fd.inode().impl.(*directory) - fs.mu.Lock() - dir.childList.Remove(fd.iter) - fs.mu.Unlock() - fd.iter = nil - } -} - -// IterDirents implements vfs.FileDescriptionImpl.IterDirents. -func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback) error { - fs := fd.filesystem() - vfsd := fd.vfsfd.VirtualDentry().Dentry() - - fs.mu.Lock() - defer fs.mu.Unlock() - - if fd.off == 0 { - if !cb.Handle(vfs.Dirent{ - Name: ".", - Type: linux.DT_DIR, - Ino: vfsd.Impl().(*dentry).inode.ino, - NextOff: 1, - }) { - return nil - } - fd.off++ - } - if fd.off == 1 { - parentInode := vfsd.ParentOrSelf().Impl().(*dentry).inode - if !cb.Handle(vfs.Dirent{ - Name: "..", - Type: parentInode.direntType(), - Ino: parentInode.ino, - NextOff: 2, - }) { - return nil - } - fd.off++ - } - - dir := vfsd.Impl().(*dentry).inode.impl.(*directory) - var child *dentry - if fd.iter == nil { - // Start iteration at the beginning of dir. - child = dir.childList.Front() - fd.iter = &dentry{} - } else { - // Continue iteration from where we left off. - child = fd.iter.Next() - dir.childList.Remove(fd.iter) - } - for child != nil { - // Skip other directoryFD iterators. - if child.inode != nil { - if !cb.Handle(vfs.Dirent{ - Name: child.vfsd.Name(), - Type: child.inode.direntType(), - Ino: child.inode.ino, - NextOff: fd.off + 1, - }) { - dir.childList.InsertBefore(child, fd.iter) - return nil - } - fd.off++ - } - child = child.Next() - } - dir.childList.PushBack(fd.iter) - return nil -} - -// Seek implements vfs.FileDescriptionImpl.Seek. -func (fd *directoryFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - fs := fd.filesystem() - fs.mu.Lock() - defer fs.mu.Unlock() - - switch whence { - case linux.SEEK_SET: - // Use offset as given. - case linux.SEEK_CUR: - offset += fd.off - default: - return 0, syserror.EINVAL - } - if offset < 0 { - return 0, syserror.EINVAL - } - - // If the offset isn't changing (e.g. due to lseek(0, SEEK_CUR)), don't - // seek even if doing so might reposition the iterator due to concurrent - // mutation of the directory. Compare fs/libfs.c:dcache_dir_lseek(). - if fd.off == offset { - return offset, nil - } - - fd.off = offset - // Compensate for "." and "..". - remChildren := int64(0) - if offset >= 2 { - remChildren = offset - 2 - } - - dir := fd.inode().impl.(*directory) - - // Ensure that fd.iter exists and is not linked into dir.childList. - if fd.iter == nil { - fd.iter = &dentry{} - } else { - dir.childList.Remove(fd.iter) - } - // Insert fd.iter before the remChildren'th child, or at the end of the - // list if remChildren >= number of children. - child := dir.childList.Front() - for child != nil { - // Skip other directoryFD iterators. - if child.inode != nil { - if remChildren == 0 { - dir.childList.InsertBefore(child, fd.iter) - return offset, nil - } - remChildren-- - } - child = child.Next() - } - dir.childList.PushBack(fd.iter) - return offset, nil -} diff --git a/pkg/sentry/fsimpl/memfs/filesystem.go b/pkg/sentry/fsimpl/memfs/filesystem.go deleted file mode 100644 index b063e09a3..000000000 --- a/pkg/sentry/fsimpl/memfs/filesystem.go +++ /dev/null @@ -1,698 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package memfs - -import ( - "fmt" - "sync/atomic" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -// Sync implements vfs.FilesystemImpl.Sync. -func (fs *filesystem) Sync(ctx context.Context) error { - // All filesystem state is in-memory. - return nil -} - -// stepLocked resolves rp.Component() to an existing file, starting from the -// given directory. -// -// stepLocked is loosely analogous to fs/namei.c:walk_component(). -// -// Preconditions: filesystem.mu must be locked. !rp.Done(). -func stepLocked(rp *vfs.ResolvingPath, d *dentry) (*dentry, error) { - if !d.inode.isDir() { - return nil, syserror.ENOTDIR - } - if err := d.inode.checkPermissions(rp.Credentials(), vfs.MayExec, true); err != nil { - return nil, err - } -afterSymlink: - nextVFSD, err := rp.ResolveComponent(&d.vfsd) - if err != nil { - return nil, err - } - if nextVFSD == nil { - // Since the Dentry tree is the sole source of truth for memfs, if it's - // not in the Dentry tree, it doesn't exist. - return nil, syserror.ENOENT - } - next := nextVFSD.Impl().(*dentry) - if symlink, ok := next.inode.impl.(*symlink); ok && rp.ShouldFollowSymlink() { - // TODO: symlink traversals update access time - if err := rp.HandleSymlink(symlink.target); err != nil { - return nil, err - } - goto afterSymlink // don't check the current directory again - } - rp.Advance() - return next, nil -} - -// walkParentDirLocked resolves all but the last path component of rp to an -// existing directory, starting from the given directory (which is usually -// rp.Start().Impl().(*dentry)). It does not check that the returned directory -// is searchable by the provider of rp. -// -// walkParentDirLocked is loosely analogous to Linux's -// fs/namei.c:path_parentat(). -// -// Preconditions: filesystem.mu must be locked. !rp.Done(). -func walkParentDirLocked(rp *vfs.ResolvingPath, d *dentry) (*dentry, error) { - for !rp.Final() { - next, err := stepLocked(rp, d) - if err != nil { - return nil, err - } - d = next - } - if !d.inode.isDir() { - return nil, syserror.ENOTDIR - } - return d, nil -} - -// resolveLocked resolves rp to an existing file. -// -// resolveLocked is loosely analogous to Linux's fs/namei.c:path_lookupat(). -// -// Preconditions: filesystem.mu must be locked. -func resolveLocked(rp *vfs.ResolvingPath) (*dentry, error) { - d := rp.Start().Impl().(*dentry) - for !rp.Done() { - next, err := stepLocked(rp, d) - if err != nil { - return nil, err - } - d = next - } - if rp.MustBeDir() && !d.inode.isDir() { - return nil, syserror.ENOTDIR - } - return d, nil -} - -// doCreateAt checks that creating a file at rp is permitted, then invokes -// create to do so. -// -// doCreateAt is loosely analogous to a conjunction of Linux's -// fs/namei.c:filename_create() and done_path_create(). -// -// Preconditions: !rp.Done(). For the final path component in rp, -// !rp.ShouldFollowSymlink(). -func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(parent *dentry, name string) error) error { - fs.mu.Lock() - defer fs.mu.Unlock() - parent, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) - if err != nil { - return err - } - if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true /* isDir */); err != nil { - return err - } - name := rp.Component() - if name == "." || name == ".." { - return syserror.EEXIST - } - // Call parent.vfsd.Child() instead of stepLocked() or rp.ResolveChild(), - // because if the child exists we want to return EEXIST immediately instead - // of attempting symlink/mount traversal. - if parent.vfsd.Child(name) != nil { - return syserror.EEXIST - } - if !dir && rp.MustBeDir() { - return syserror.ENOENT - } - // In memfs, the only way to cause a dentry to be disowned is by removing - // it from the filesystem, so this check is equivalent to checking if - // parent has been removed. - if parent.vfsd.IsDisowned() { - return syserror.ENOENT - } - mnt := rp.Mount() - if err := mnt.CheckBeginWrite(); err != nil { - return err - } - defer mnt.EndWrite() - return create(parent, name) -} - -// GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. -func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { - fs.mu.RLock() - defer fs.mu.RUnlock() - d, err := resolveLocked(rp) - if err != nil { - return nil, err - } - if opts.CheckSearchable { - if !d.inode.isDir() { - return nil, syserror.ENOTDIR - } - if err := d.inode.checkPermissions(rp.Credentials(), vfs.MayExec, true /* isDir */); err != nil { - return nil, err - } - } - d.IncRef() - return &d.vfsd, nil -} - -// GetParentDentryAt implements vfs.FilesystemImpl.GetParentDentryAt. -func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { - fs.mu.RLock() - defer fs.mu.RUnlock() - d, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) - if err != nil { - return nil, err - } - d.IncRef() - return &d.vfsd, nil -} - -// LinkAt implements vfs.FilesystemImpl.LinkAt. -func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry) error { - return fs.doCreateAt(rp, false /* dir */, func(parent *dentry, name string) error { - if rp.Mount() != vd.Mount() { - return syserror.EXDEV - } - d := vd.Dentry().Impl().(*dentry) - if d.inode.isDir() { - return syserror.EPERM - } - if d.inode.nlink == 0 { - return syserror.ENOENT - } - if d.inode.nlink == maxLinks { - return syserror.EMLINK - } - d.inode.incLinksLocked() - child := fs.newDentry(d.inode) - parent.vfsd.InsertChild(&child.vfsd, name) - parent.inode.impl.(*directory).childList.PushBack(child) - return nil - }) -} - -// MkdirAt implements vfs.FilesystemImpl.MkdirAt. -func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error { - return fs.doCreateAt(rp, true /* dir */, func(parent *dentry, name string) error { - if parent.inode.nlink == maxLinks { - return syserror.EMLINK - } - parent.inode.incLinksLocked() // from child's ".." - child := fs.newDentry(fs.newDirectory(rp.Credentials(), opts.Mode)) - parent.vfsd.InsertChild(&child.vfsd, name) - parent.inode.impl.(*directory).childList.PushBack(child) - return nil - }) -} - -// MknodAt implements vfs.FilesystemImpl.MknodAt. -func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MknodOptions) error { - return fs.doCreateAt(rp, false /* dir */, func(parent *dentry, name string) error { - switch opts.Mode.FileType() { - case 0, linux.S_IFREG: - child := fs.newDentry(fs.newRegularFile(rp.Credentials(), opts.Mode)) - parent.vfsd.InsertChild(&child.vfsd, name) - parent.inode.impl.(*directory).childList.PushBack(child) - return nil - case linux.S_IFIFO: - child := fs.newDentry(fs.newNamedPipe(rp.Credentials(), opts.Mode)) - parent.vfsd.InsertChild(&child.vfsd, name) - parent.inode.impl.(*directory).childList.PushBack(child) - return nil - case linux.S_IFBLK, linux.S_IFCHR, linux.S_IFSOCK: - // Not yet supported. - return syserror.EPERM - default: - return syserror.EINVAL - } - }) -} - -// OpenAt implements vfs.FilesystemImpl.OpenAt. -func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - if opts.Flags&linux.O_TMPFILE != 0 { - // Not yet supported. - return nil, syserror.EOPNOTSUPP - } - - // Handle O_CREAT and !O_CREAT separately, since in the latter case we - // don't need fs.mu for writing. - if opts.Flags&linux.O_CREAT == 0 { - fs.mu.RLock() - defer fs.mu.RUnlock() - d, err := resolveLocked(rp) - if err != nil { - return nil, err - } - return d.open(ctx, rp, opts.Flags, false /* afterCreate */) - } - - mustCreate := opts.Flags&linux.O_EXCL != 0 - start := rp.Start().Impl().(*dentry) - fs.mu.Lock() - defer fs.mu.Unlock() - if rp.Done() { - // Reject attempts to open directories with O_CREAT. - if rp.MustBeDir() { - return nil, syserror.EISDIR - } - if mustCreate { - return nil, syserror.EEXIST - } - return start.open(ctx, rp, opts.Flags, false /* afterCreate */) - } -afterTrailingSymlink: - parent, err := walkParentDirLocked(rp, start) - if err != nil { - return nil, err - } - // Check for search permission in the parent directory. - if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayExec, true); err != nil { - return nil, err - } - // Reject attempts to open directories with O_CREAT. - if rp.MustBeDir() { - return nil, syserror.EISDIR - } - name := rp.Component() - if name == "." || name == ".." { - return nil, syserror.EISDIR - } - // Determine whether or not we need to create a file. - child, err := stepLocked(rp, parent) - if err == syserror.ENOENT { - // Already checked for searchability above; now check for writability. - if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite, true); err != nil { - return nil, err - } - if err := rp.Mount().CheckBeginWrite(); err != nil { - return nil, err - } - defer rp.Mount().EndWrite() - // Create and open the child. - child := fs.newDentry(fs.newRegularFile(rp.Credentials(), opts.Mode)) - parent.vfsd.InsertChild(&child.vfsd, name) - parent.inode.impl.(*directory).childList.PushBack(child) - return child.open(ctx, rp, opts.Flags, true) - } - if err != nil { - return nil, err - } - // Do we need to resolve a trailing symlink? - if !rp.Done() { - start = parent - goto afterTrailingSymlink - } - // Open existing file. - if mustCreate { - return nil, syserror.EEXIST - } - return child.open(ctx, rp, opts.Flags, false) -} - -func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, flags uint32, afterCreate bool) (*vfs.FileDescription, error) { - ats := vfs.AccessTypesForOpenFlags(flags) - if !afterCreate { - if err := d.inode.checkPermissions(rp.Credentials(), ats, d.inode.isDir()); err != nil { - return nil, err - } - } - mnt := rp.Mount() - switch impl := d.inode.impl.(type) { - case *regularFile: - var fd regularFileFD - fd.readable = vfs.MayReadFileWithOpenFlags(flags) - fd.writable = vfs.MayWriteFileWithOpenFlags(flags) - if fd.writable { - if err := mnt.CheckBeginWrite(); err != nil { - return nil, err - } - // mnt.EndWrite() is called by regularFileFD.Release(). - } - fd.vfsfd.Init(&fd, flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}) - if flags&linux.O_TRUNC != 0 { - impl.mu.Lock() - impl.data = impl.data[:0] - atomic.StoreInt64(&impl.dataLen, 0) - impl.mu.Unlock() - } - return &fd.vfsfd, nil - case *directory: - // Can't open directories writably. - if ats&vfs.MayWrite != 0 { - return nil, syserror.EISDIR - } - var fd directoryFD - fd.vfsfd.Init(&fd, flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}) - return &fd.vfsfd, nil - case *symlink: - // Can't open symlinks without O_PATH (which is unimplemented). - return nil, syserror.ELOOP - case *namedPipe: - return newNamedPipeFD(ctx, impl, rp, &d.vfsd, flags) - default: - panic(fmt.Sprintf("unknown inode type: %T", d.inode.impl)) - } -} - -// ReadlinkAt implements vfs.FilesystemImpl.ReadlinkAt. -func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { - fs.mu.RLock() - defer fs.mu.RUnlock() - d, err := resolveLocked(rp) - if err != nil { - return "", err - } - symlink, ok := d.inode.impl.(*symlink) - if !ok { - return "", syserror.EINVAL - } - return symlink.target, nil -} - -// RenameAt implements vfs.FilesystemImpl.RenameAt. -func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldParentVD vfs.VirtualDentry, oldName string, opts vfs.RenameOptions) error { - if opts.Flags != 0 { - // TODO(b/145974740): Support renameat2 flags. - return syserror.EINVAL - } - - // Resolve newParent first to verify that it's on this Mount. - fs.mu.Lock() - defer fs.mu.Unlock() - newParent, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) - if err != nil { - return err - } - newName := rp.Component() - if newName == "." || newName == ".." { - return syserror.EBUSY - } - mnt := rp.Mount() - if mnt != oldParentVD.Mount() { - return syserror.EXDEV - } - if err := mnt.CheckBeginWrite(); err != nil { - return err - } - defer mnt.EndWrite() - - oldParent := oldParentVD.Dentry().Impl().(*dentry) - if err := oldParent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true /* isDir */); err != nil { - return err - } - // Call vfs.Dentry.Child() instead of stepLocked() or rp.ResolveChild(), - // because if the existing child is a symlink or mount point then we want - // to rename over it rather than follow it. - renamedVFSD := oldParent.vfsd.Child(oldName) - if renamedVFSD == nil { - return syserror.ENOENT - } - renamed := renamedVFSD.Impl().(*dentry) - if renamed.inode.isDir() { - if renamed == newParent || renamedVFSD.IsAncestorOf(&newParent.vfsd) { - return syserror.EINVAL - } - if oldParent != newParent { - // Writability is needed to change renamed's "..". - if err := renamed.inode.checkPermissions(rp.Credentials(), vfs.MayWrite, true /* isDir */); err != nil { - return err - } - } - } else { - if opts.MustBeDir || rp.MustBeDir() { - return syserror.ENOTDIR - } - } - - if err := newParent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true /* isDir */); err != nil { - return err - } - replacedVFSD := newParent.vfsd.Child(newName) - var replaced *dentry - if replacedVFSD != nil { - replaced = replacedVFSD.Impl().(*dentry) - if replaced.inode.isDir() { - if !renamed.inode.isDir() { - return syserror.EISDIR - } - if replaced.vfsd.HasChildren() { - return syserror.ENOTEMPTY - } - } else { - if rp.MustBeDir() { - return syserror.ENOTDIR - } - if renamed.inode.isDir() { - return syserror.ENOTDIR - } - } - } else { - if renamed.inode.isDir() && newParent.inode.nlink == maxLinks { - return syserror.EMLINK - } - } - if newParent.vfsd.IsDisowned() { - return syserror.ENOENT - } - - // Linux places this check before some of those above; we do it here for - // simplicity, under the assumption that applications are not intentionally - // doing noop renames expecting them to succeed where non-noop renames - // would fail. - if renamedVFSD == replacedVFSD { - return nil - } - vfsObj := rp.VirtualFilesystem() - oldParentDir := oldParent.inode.impl.(*directory) - newParentDir := newParent.inode.impl.(*directory) - if err := vfsObj.PrepareRenameDentry(vfs.MountNamespaceFromContext(ctx), renamedVFSD, replacedVFSD); err != nil { - return err - } - if replaced != nil { - newParentDir.childList.Remove(replaced) - if replaced.inode.isDir() { - newParent.inode.decLinksLocked() // from replaced's ".." - } - replaced.inode.decLinksLocked() - } - oldParentDir.childList.Remove(renamed) - newParentDir.childList.PushBack(renamed) - if renamed.inode.isDir() { - oldParent.inode.decLinksLocked() - newParent.inode.incLinksLocked() - } - // TODO: update timestamps and parent directory sizes - vfsObj.CommitRenameReplaceDentry(renamedVFSD, &newParent.vfsd, newName, replacedVFSD) - return nil -} - -// RmdirAt implements vfs.FilesystemImpl.RmdirAt. -func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error { - fs.mu.Lock() - defer fs.mu.Unlock() - parent, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) - if err != nil { - return err - } - if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true /* isDir */); err != nil { - return err - } - name := rp.Component() - if name == "." { - return syserror.EINVAL - } - if name == ".." { - return syserror.ENOTEMPTY - } - childVFSD := parent.vfsd.Child(name) - if childVFSD == nil { - return syserror.ENOENT - } - child := childVFSD.Impl().(*dentry) - if !child.inode.isDir() { - return syserror.ENOTDIR - } - if childVFSD.HasChildren() { - return syserror.ENOTEMPTY - } - mnt := rp.Mount() - if err := mnt.CheckBeginWrite(); err != nil { - return err - } - defer mnt.EndWrite() - vfsObj := rp.VirtualFilesystem() - if err := vfsObj.PrepareDeleteDentry(vfs.MountNamespaceFromContext(ctx), childVFSD); err != nil { - return err - } - parent.inode.impl.(*directory).childList.Remove(child) - parent.inode.decLinksLocked() // from child's ".." - child.inode.decLinksLocked() - vfsObj.CommitDeleteDentry(childVFSD) - return nil -} - -// SetStatAt implements vfs.FilesystemImpl.SetStatAt. -func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { - fs.mu.RLock() - defer fs.mu.RUnlock() - _, err := resolveLocked(rp) - if err != nil { - return err - } - if opts.Stat.Mask == 0 { - return nil - } - // TODO: implement inode.setStat - return syserror.EPERM -} - -// StatAt implements vfs.FilesystemImpl.StatAt. -func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { - fs.mu.RLock() - defer fs.mu.RUnlock() - d, err := resolveLocked(rp) - if err != nil { - return linux.Statx{}, err - } - var stat linux.Statx - d.inode.statTo(&stat) - return stat, nil -} - -// StatFSAt implements vfs.FilesystemImpl.StatFSAt. -func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { - fs.mu.RLock() - defer fs.mu.RUnlock() - _, err := resolveLocked(rp) - if err != nil { - return linux.Statfs{}, err - } - // TODO: actually implement statfs - return linux.Statfs{}, syserror.ENOSYS -} - -// SymlinkAt implements vfs.FilesystemImpl.SymlinkAt. -func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, target string) error { - return fs.doCreateAt(rp, false /* dir */, func(parent *dentry, name string) error { - child := fs.newDentry(fs.newSymlink(rp.Credentials(), target)) - parent.vfsd.InsertChild(&child.vfsd, name) - parent.inode.impl.(*directory).childList.PushBack(child) - return nil - }) -} - -// UnlinkAt implements vfs.FilesystemImpl.UnlinkAt. -func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error { - fs.mu.Lock() - defer fs.mu.Unlock() - parent, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) - if err != nil { - return err - } - if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true /* isDir */); err != nil { - return err - } - name := rp.Component() - if name == "." || name == ".." { - return syserror.EISDIR - } - childVFSD := parent.vfsd.Child(name) - if childVFSD == nil { - return syserror.ENOENT - } - child := childVFSD.Impl().(*dentry) - if child.inode.isDir() { - return syserror.EISDIR - } - if !rp.MustBeDir() { - return syserror.ENOTDIR - } - mnt := rp.Mount() - if err := mnt.CheckBeginWrite(); err != nil { - return err - } - defer mnt.EndWrite() - vfsObj := rp.VirtualFilesystem() - if err := vfsObj.PrepareDeleteDentry(vfs.MountNamespaceFromContext(ctx), childVFSD); err != nil { - return err - } - parent.inode.impl.(*directory).childList.Remove(child) - child.inode.decLinksLocked() - vfsObj.CommitDeleteDentry(childVFSD) - return nil -} - -// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. -func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) { - fs.mu.RLock() - defer fs.mu.RUnlock() - _, err := resolveLocked(rp) - if err != nil { - return nil, err - } - // TODO(b/127675828): support extended attributes - return nil, syserror.ENOTSUP -} - -// GetxattrAt implements vfs.FilesystemImpl.GetxattrAt. -func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) (string, error) { - fs.mu.RLock() - defer fs.mu.RUnlock() - _, err := resolveLocked(rp) - if err != nil { - return "", err - } - // TODO(b/127675828): support extended attributes - return "", syserror.ENOTSUP -} - -// SetxattrAt implements vfs.FilesystemImpl.SetxattrAt. -func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { - fs.mu.RLock() - defer fs.mu.RUnlock() - _, err := resolveLocked(rp) - if err != nil { - return err - } - // TODO(b/127675828): support extended attributes - return syserror.ENOTSUP -} - -// RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. -func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { - fs.mu.RLock() - defer fs.mu.RUnlock() - _, err := resolveLocked(rp) - if err != nil { - return err - } - // TODO(b/127675828): support extended attributes - return syserror.ENOTSUP -} - -// PrependPath implements vfs.FilesystemImpl.PrependPath. -func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { - fs.mu.RLock() - defer fs.mu.RUnlock() - return vfs.GenericPrependPath(vfsroot, vd, b) -} diff --git a/pkg/sentry/fsimpl/memfs/memfs.go b/pkg/sentry/fsimpl/memfs/memfs.go deleted file mode 100644 index 8d0167c93..000000000 --- a/pkg/sentry/fsimpl/memfs/memfs.go +++ /dev/null @@ -1,293 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package memfs provides a filesystem implementation that behaves like tmpfs: -// the Dentry tree is the sole source of truth for the state of the filesystem. -// -// memfs is intended primarily to demonstrate filesystem implementation -// patterns. Real uses cases for an in-memory filesystem should use tmpfs -// instead. -// -// Lock order: -// -// filesystem.mu -// regularFileFD.offMu -// regularFile.mu -// inode.mu -package memfs - -import ( - "fmt" - "math" - "sync" - "sync/atomic" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -// FilesystemType implements vfs.FilesystemType. -type FilesystemType struct{} - -// filesystem implements vfs.FilesystemImpl. -type filesystem struct { - vfsfs vfs.Filesystem - - // mu serializes changes to the Dentry tree. - mu sync.RWMutex - - nextInoMinusOne uint64 // accessed using atomic memory operations -} - -// GetFilesystem implements vfs.FilesystemType.GetFilesystem. -func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { - var fs filesystem - fs.vfsfs.Init(vfsObj, &fs) - root := fs.newDentry(fs.newDirectory(creds, 01777)) - return &fs.vfsfs, &root.vfsd, nil -} - -// Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { -} - -// dentry implements vfs.DentryImpl. -type dentry struct { - vfsd vfs.Dentry - - // inode is the inode represented by this dentry. Multiple Dentries may - // share a single non-directory inode (with hard links). inode is - // immutable. - inode *inode - - // memfs doesn't count references on dentries; because the dentry tree is - // the sole source of truth, it is by definition always consistent with the - // state of the filesystem. However, it does count references on inodes, - // because inode resources are released when all references are dropped. - // (memfs doesn't really have resources to release, but we implement - // reference counting because tmpfs regular files will.) - - // dentryEntry (ugh) links dentries into their parent directory.childList. - dentryEntry -} - -func (fs *filesystem) newDentry(inode *inode) *dentry { - d := &dentry{ - inode: inode, - } - d.vfsd.Init(d) - return d -} - -// IncRef implements vfs.DentryImpl.IncRef. -func (d *dentry) IncRef() { - d.inode.incRef() -} - -// TryIncRef implements vfs.DentryImpl.TryIncRef. -func (d *dentry) TryIncRef() bool { - return d.inode.tryIncRef() -} - -// DecRef implements vfs.DentryImpl.DecRef. -func (d *dentry) DecRef() { - d.inode.decRef() -} - -// inode represents a filesystem object. -type inode struct { - // refs is a reference count. refs is accessed using atomic memory - // operations. - // - // A reference is held on all inodes that are reachable in the filesystem - // tree. For non-directories (which may have multiple hard links), this - // means that a reference is dropped when nlink reaches 0. For directories, - // nlink never reaches 0 due to the "." entry; instead, - // filesystem.RmdirAt() drops the reference. - refs int64 - - // Inode metadata; protected by mu and accessed using atomic memory - // operations unless otherwise specified. - mu sync.RWMutex - mode uint32 // excluding file type bits, which are based on impl - nlink uint32 // protected by filesystem.mu instead of inode.mu - uid uint32 // auth.KUID, but stored as raw uint32 for sync/atomic - gid uint32 // auth.KGID, but ... - ino uint64 // immutable - - impl interface{} // immutable -} - -const maxLinks = math.MaxUint32 - -func (i *inode) init(impl interface{}, fs *filesystem, creds *auth.Credentials, mode linux.FileMode) { - i.refs = 1 - i.mode = uint32(mode) - i.uid = uint32(creds.EffectiveKUID) - i.gid = uint32(creds.EffectiveKGID) - i.ino = atomic.AddUint64(&fs.nextInoMinusOne, 1) - // i.nlink initialized by caller - i.impl = impl -} - -// incLinksLocked increments i's link count. -// -// Preconditions: filesystem.mu must be locked for writing. i.nlink != 0. -// i.nlink < maxLinks. -func (i *inode) incLinksLocked() { - if i.nlink == 0 { - panic("memfs.inode.incLinksLocked() called with no existing links") - } - if i.nlink == maxLinks { - panic("memfs.inode.incLinksLocked() called with maximum link count") - } - atomic.AddUint32(&i.nlink, 1) -} - -// decLinksLocked decrements i's link count. -// -// Preconditions: filesystem.mu must be locked for writing. i.nlink != 0. -func (i *inode) decLinksLocked() { - if i.nlink == 0 { - panic("memfs.inode.decLinksLocked() called with no existing links") - } - atomic.AddUint32(&i.nlink, ^uint32(0)) -} - -func (i *inode) incRef() { - if atomic.AddInt64(&i.refs, 1) <= 1 { - panic("memfs.inode.incRef() called without holding a reference") - } -} - -func (i *inode) tryIncRef() bool { - for { - refs := atomic.LoadInt64(&i.refs) - if refs == 0 { - return false - } - if atomic.CompareAndSwapInt64(&i.refs, refs, refs+1) { - return true - } - } -} - -func (i *inode) decRef() { - if refs := atomic.AddInt64(&i.refs, -1); refs == 0 { - // This is unnecessary; it's mostly to simulate what tmpfs would do. - if regfile, ok := i.impl.(*regularFile); ok { - regfile.mu.Lock() - regfile.data = nil - atomic.StoreInt64(®file.dataLen, 0) - regfile.mu.Unlock() - } - } else if refs < 0 { - panic("memfs.inode.decRef() called without holding a reference") - } -} - -func (i *inode) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes, isDir bool) error { - return vfs.GenericCheckPermissions(creds, ats, isDir, uint16(atomic.LoadUint32(&i.mode)), auth.KUID(atomic.LoadUint32(&i.uid)), auth.KGID(atomic.LoadUint32(&i.gid))) -} - -// Go won't inline this function, and returning linux.Statx (which is quite -// big) means spending a lot of time in runtime.duffcopy(), so instead it's an -// output parameter. -func (i *inode) statTo(stat *linux.Statx) { - stat.Mask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_NLINK | linux.STATX_UID | linux.STATX_GID | linux.STATX_INO - stat.Blksize = 1 // usermem.PageSize in tmpfs - stat.Nlink = atomic.LoadUint32(&i.nlink) - stat.UID = atomic.LoadUint32(&i.uid) - stat.GID = atomic.LoadUint32(&i.gid) - stat.Mode = uint16(atomic.LoadUint32(&i.mode)) - stat.Ino = i.ino - // TODO: device number - switch impl := i.impl.(type) { - case *regularFile: - stat.Mode |= linux.S_IFREG - stat.Mask |= linux.STATX_SIZE | linux.STATX_BLOCKS - stat.Size = uint64(atomic.LoadInt64(&impl.dataLen)) - // In tmpfs, this will be FileRangeSet.Span() / 512 (but also cached in - // a uint64 accessed using atomic memory operations to avoid taking - // locks). - stat.Blocks = allocatedBlocksForSize(stat.Size) - case *directory: - stat.Mode |= linux.S_IFDIR - case *symlink: - stat.Mode |= linux.S_IFLNK - stat.Mask |= linux.STATX_SIZE | linux.STATX_BLOCKS - stat.Size = uint64(len(impl.target)) - stat.Blocks = allocatedBlocksForSize(stat.Size) - case *namedPipe: - stat.Mode |= linux.S_IFIFO - default: - panic(fmt.Sprintf("unknown inode type: %T", i.impl)) - } -} - -// allocatedBlocksForSize returns the number of 512B blocks needed to -// accommodate the given size in bytes, as appropriate for struct -// stat::st_blocks and struct statx::stx_blocks. (Note that this 512B block -// size is independent of the "preferred block size for I/O", struct -// stat::st_blksize and struct statx::stx_blksize.) -func allocatedBlocksForSize(size uint64) uint64 { - return (size + 511) / 512 -} - -func (i *inode) direntType() uint8 { - switch i.impl.(type) { - case *regularFile: - return linux.DT_REG - case *directory: - return linux.DT_DIR - case *symlink: - return linux.DT_LNK - default: - panic(fmt.Sprintf("unknown inode type: %T", i.impl)) - } -} - -// fileDescription is embedded by memfs implementations of -// vfs.FileDescriptionImpl. -type fileDescription struct { - vfsfd vfs.FileDescription - vfs.FileDescriptionDefaultImpl -} - -func (fd *fileDescription) filesystem() *filesystem { - return fd.vfsfd.Mount().Filesystem().Impl().(*filesystem) -} - -func (fd *fileDescription) inode() *inode { - return fd.vfsfd.Dentry().Impl().(*dentry).inode -} - -// Stat implements vfs.FileDescriptionImpl.Stat. -func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { - var stat linux.Statx - fd.inode().statTo(&stat) - return stat, nil -} - -// SetStat implements vfs.FileDescriptionImpl.SetStat. -func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { - if opts.Stat.Mask == 0 { - return nil - } - // TODO: implement inode.setStat - return syserror.EPERM -} diff --git a/pkg/sentry/fsimpl/memfs/named_pipe.go b/pkg/sentry/fsimpl/memfs/named_pipe.go deleted file mode 100644 index b5a204438..000000000 --- a/pkg/sentry/fsimpl/memfs/named_pipe.go +++ /dev/null @@ -1,60 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package memfs - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" - "gvisor.dev/gvisor/pkg/sentry/usermem" - "gvisor.dev/gvisor/pkg/sentry/vfs" -) - -type namedPipe struct { - inode inode - - pipe *pipe.VFSPipe -} - -// Preconditions: -// * fs.mu must be locked. -// * rp.Mount().CheckBeginWrite() has been called successfully. -func (fs *filesystem) newNamedPipe(creds *auth.Credentials, mode linux.FileMode) *inode { - file := &namedPipe{pipe: pipe.NewVFSPipe(pipe.DefaultPipeSize, usermem.PageSize)} - file.inode.init(file, fs, creds, mode) - file.inode.nlink = 1 // Only the parent has a link. - return &file.inode -} - -// namedPipeFD implements vfs.FileDescriptionImpl. Methods are implemented -// entirely via struct embedding. -type namedPipeFD struct { - fileDescription - - *pipe.VFSPipeFD -} - -func newNamedPipeFD(ctx context.Context, np *namedPipe, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { - var err error - var fd namedPipeFD - fd.VFSPipeFD, err = np.pipe.NewVFSPipeFD(ctx, rp, vfsd, &fd.vfsfd, flags) - if err != nil { - return nil, err - } - mnt := rp.Mount() - fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{}) - return &fd.vfsfd, nil -} diff --git a/pkg/sentry/fsimpl/memfs/pipe_test.go b/pkg/sentry/fsimpl/memfs/pipe_test.go deleted file mode 100644 index 807c1af7a..000000000 --- a/pkg/sentry/fsimpl/memfs/pipe_test.go +++ /dev/null @@ -1,235 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package memfs - -import ( - "bytes" - "testing" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/usermem" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -const fileName = "mypipe" - -func TestSeparateFDs(t *testing.T) { - ctx, creds, vfsObj, root := setup(t) - defer root.DecRef() - - // Open the read side. This is done in a concurrently because opening - // One end the pipe blocks until the other end is opened. - pop := vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(fileName), - FollowFinalSymlink: true, - } - rfdchan := make(chan *vfs.FileDescription) - go func() { - openOpts := vfs.OpenOptions{Flags: linux.O_RDONLY} - rfd, _ := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) - rfdchan <- rfd - }() - - // Open the write side. - openOpts := vfs.OpenOptions{Flags: linux.O_WRONLY} - wfd, err := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) - if err != nil { - t.Fatalf("failed to open pipe for writing %q: %v", fileName, err) - } - defer wfd.DecRef() - - rfd, ok := <-rfdchan - if !ok { - t.Fatalf("failed to open pipe for reading %q", fileName) - } - defer rfd.DecRef() - - const msg = "vamos azul" - checkEmpty(ctx, t, rfd) - checkWrite(ctx, t, wfd, msg) - checkRead(ctx, t, rfd, msg) -} - -func TestNonblockingRead(t *testing.T) { - ctx, creds, vfsObj, root := setup(t) - defer root.DecRef() - - // Open the read side as nonblocking. - pop := vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(fileName), - FollowFinalSymlink: true, - } - openOpts := vfs.OpenOptions{Flags: linux.O_RDONLY | linux.O_NONBLOCK} - rfd, err := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) - if err != nil { - t.Fatalf("failed to open pipe for reading %q: %v", fileName, err) - } - defer rfd.DecRef() - - // Open the write side. - openOpts = vfs.OpenOptions{Flags: linux.O_WRONLY} - wfd, err := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) - if err != nil { - t.Fatalf("failed to open pipe for writing %q: %v", fileName, err) - } - defer wfd.DecRef() - - const msg = "geh blau" - checkEmpty(ctx, t, rfd) - checkWrite(ctx, t, wfd, msg) - checkRead(ctx, t, rfd, msg) -} - -func TestNonblockingWriteError(t *testing.T) { - ctx, creds, vfsObj, root := setup(t) - defer root.DecRef() - - // Open the write side as nonblocking, which should return ENXIO. - pop := vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(fileName), - FollowFinalSymlink: true, - } - openOpts := vfs.OpenOptions{Flags: linux.O_WRONLY | linux.O_NONBLOCK} - _, err := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) - if err != syserror.ENXIO { - t.Fatalf("expected ENXIO, but got error: %v", err) - } -} - -func TestSingleFD(t *testing.T) { - ctx, creds, vfsObj, root := setup(t) - defer root.DecRef() - - // Open the pipe as readable and writable. - pop := vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(fileName), - FollowFinalSymlink: true, - } - openOpts := vfs.OpenOptions{Flags: linux.O_RDWR} - fd, err := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) - if err != nil { - t.Fatalf("failed to open pipe for writing %q: %v", fileName, err) - } - defer fd.DecRef() - - const msg = "forza blu" - checkEmpty(ctx, t, fd) - checkWrite(ctx, t, fd, msg) - checkRead(ctx, t, fd, msg) -} - -// setup creates a VFS with a pipe in the root directory at path fileName. The -// returned VirtualDentry must be DecRef()'d be the caller. It calls t.Fatal -// upon failure. -func setup(t *testing.T) (context.Context, *auth.Credentials, *vfs.VirtualFilesystem, vfs.VirtualDentry) { - ctx := contexttest.Context(t) - creds := auth.CredentialsFromContext(ctx) - - // Create VFS. - vfsObj := vfs.New() - vfsObj.MustRegisterFilesystemType("memfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "memfs", &vfs.GetFilesystemOptions{}) - if err != nil { - t.Fatalf("failed to create tmpfs root mount: %v", err) - } - - // Create the pipe. - root := mntns.Root() - pop := vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(fileName), - } - mknodOpts := vfs.MknodOptions{Mode: linux.ModeNamedPipe | 0644} - if err := vfsObj.MknodAt(ctx, creds, &pop, &mknodOpts); err != nil { - t.Fatalf("failed to create file %q: %v", fileName, err) - } - - // Sanity check: the file pipe exists and has the correct mode. - stat, err := vfsObj.StatAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(fileName), - FollowFinalSymlink: true, - }, &vfs.StatOptions{}) - if err != nil { - t.Fatalf("stat(%q) failed: %v", fileName, err) - } - if stat.Mode&^linux.S_IFMT != 0644 { - t.Errorf("got wrong permissions (%0o)", stat.Mode) - } - if stat.Mode&linux.S_IFMT != linux.ModeNamedPipe { - t.Errorf("got wrong file type (%0o)", stat.Mode) - } - - return ctx, creds, vfsObj, root -} - -// checkEmpty calls t.Fatal if the pipe in fd is not empty. -func checkEmpty(ctx context.Context, t *testing.T, fd *vfs.FileDescription) { - readData := make([]byte, 1) - dst := usermem.BytesIOSequence(readData) - bytesRead, err := fd.Read(ctx, dst, vfs.ReadOptions{}) - if err != syserror.ErrWouldBlock { - t.Fatalf("expected ErrWouldBlock reading from empty pipe %q, but got: %v", fileName, err) - } - if bytesRead != 0 { - t.Fatalf("expected to read 0 bytes, but got %d", bytesRead) - } -} - -// checkWrite calls t.Fatal if it fails to write all of msg to fd. -func checkWrite(ctx context.Context, t *testing.T, fd *vfs.FileDescription, msg string) { - writeData := []byte(msg) - src := usermem.BytesIOSequence(writeData) - bytesWritten, err := fd.Write(ctx, src, vfs.WriteOptions{}) - if err != nil { - t.Fatalf("error writing to pipe %q: %v", fileName, err) - } - if bytesWritten != int64(len(writeData)) { - t.Fatalf("expected to write %d bytes, but wrote %d", len(writeData), bytesWritten) - } -} - -// checkRead calls t.Fatal if it fails to read msg from fd. -func checkRead(ctx context.Context, t *testing.T, fd *vfs.FileDescription, msg string) { - readData := make([]byte, len(msg)) - dst := usermem.BytesIOSequence(readData) - bytesRead, err := fd.Read(ctx, dst, vfs.ReadOptions{}) - if err != nil { - t.Fatalf("error reading from pipe %q: %v", fileName, err) - } - if bytesRead != int64(len(msg)) { - t.Fatalf("expected to read %d bytes, but got %d", len(msg), bytesRead) - } - if !bytes.Equal(readData, []byte(msg)) { - t.Fatalf("expected to read %q from pipe, but got %q", msg, string(readData)) - } -} diff --git a/pkg/sentry/fsimpl/memfs/regular_file.go b/pkg/sentry/fsimpl/memfs/regular_file.go deleted file mode 100644 index b7f4853b3..000000000 --- a/pkg/sentry/fsimpl/memfs/regular_file.go +++ /dev/null @@ -1,154 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package memfs - -import ( - "io" - "sync" - "sync/atomic" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/usermem" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -type regularFile struct { - inode inode - - mu sync.RWMutex - data []byte - // dataLen is len(data), but accessed using atomic memory operations to - // avoid locking in inode.stat(). - dataLen int64 -} - -func (fs *filesystem) newRegularFile(creds *auth.Credentials, mode linux.FileMode) *inode { - file := ®ularFile{} - file.inode.init(file, fs, creds, mode) - file.inode.nlink = 1 // from parent directory - return &file.inode -} - -type regularFileFD struct { - fileDescription - - // These are immutable. - readable bool - writable bool - - // off is the file offset. off is accessed using atomic memory operations. - // offMu serializes operations that may mutate off. - off int64 - offMu sync.Mutex -} - -// Release implements vfs.FileDescriptionImpl.Release. -func (fd *regularFileFD) Release() { - if fd.writable { - fd.vfsfd.VirtualDentry().Mount().EndWrite() - } -} - -// PRead implements vfs.FileDescriptionImpl.PRead. -func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - if !fd.readable { - return 0, syserror.EINVAL - } - f := fd.inode().impl.(*regularFile) - f.mu.RLock() - if offset >= int64(len(f.data)) { - f.mu.RUnlock() - return 0, io.EOF - } - n, err := dst.CopyOut(ctx, f.data[offset:]) - f.mu.RUnlock() - return int64(n), err -} - -// Read implements vfs.FileDescriptionImpl.Read. -func (fd *regularFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - fd.offMu.Lock() - n, err := fd.PRead(ctx, dst, fd.off, opts) - fd.off += n - fd.offMu.Unlock() - return n, err -} - -// PWrite implements vfs.FileDescriptionImpl.PWrite. -func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - if !fd.writable { - return 0, syserror.EINVAL - } - if offset < 0 { - return 0, syserror.EINVAL - } - srclen := src.NumBytes() - if srclen == 0 { - return 0, nil - } - f := fd.inode().impl.(*regularFile) - f.mu.Lock() - end := offset + srclen - if end < offset { - // Overflow. - f.mu.Unlock() - return 0, syserror.EFBIG - } - if end > f.dataLen { - f.data = append(f.data, make([]byte, end-f.dataLen)...) - atomic.StoreInt64(&f.dataLen, end) - } - n, err := src.CopyIn(ctx, f.data[offset:end]) - f.mu.Unlock() - return int64(n), err -} - -// Write implements vfs.FileDescriptionImpl.Write. -func (fd *regularFileFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { - fd.offMu.Lock() - n, err := fd.PWrite(ctx, src, fd.off, opts) - fd.off += n - fd.offMu.Unlock() - return n, err -} - -// Seek implements vfs.FileDescriptionImpl.Seek. -func (fd *regularFileFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - fd.offMu.Lock() - defer fd.offMu.Unlock() - switch whence { - case linux.SEEK_SET: - // use offset as specified - case linux.SEEK_CUR: - offset += fd.off - case linux.SEEK_END: - offset += atomic.LoadInt64(&fd.inode().impl.(*regularFile).dataLen) - default: - return 0, syserror.EINVAL - } - if offset < 0 { - return 0, syserror.EINVAL - } - fd.off = offset - return offset, nil -} - -// Sync implements vfs.FileDescriptionImpl.Sync. -func (fd *regularFileFD) Sync(ctx context.Context) error { - return nil -} diff --git a/pkg/sentry/fsimpl/memfs/symlink.go b/pkg/sentry/fsimpl/memfs/symlink.go deleted file mode 100644 index b2ac2cbeb..000000000 --- a/pkg/sentry/fsimpl/memfs/symlink.go +++ /dev/null @@ -1,36 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package memfs - -import ( - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" -) - -type symlink struct { - inode inode - target string // immutable -} - -func (fs *filesystem) newSymlink(creds *auth.Credentials, target string) *inode { - link := &symlink{ - target: target, - } - link.inode.init(link, fs, creds, 0777) - link.inode.nlink = 1 // from parent directory - return &link.inode -} - -// O_PATH is unimplemented, so there's no way to get a FileDescription -// representing a symlink yet. diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD new file mode 100644 index 000000000..a5b285987 --- /dev/null +++ b/pkg/sentry/fsimpl/tmpfs/BUILD @@ -0,0 +1,92 @@ +load("//tools/go_stateify:defs.bzl", "go_library") +load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools/go_generics:defs.bzl", "go_template_instance") + +package(licenses = ["notice"]) + +go_template_instance( + name = "dentry_list", + out = "dentry_list.go", + package = "tmpfs", + prefix = "dentry", + template = "//pkg/ilist:generic_list", + types = { + "Element": "*dentry", + "Linker": "*dentry", + }, +) + +go_library( + name = "tmpfs", + srcs = [ + "dentry_list.go", + "directory.go", + "filesystem.go", + "named_pipe.go", + "regular_file.go", + "symlink.go", + "tmpfs.go", + ], + importpath = "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs", + deps = [ + "//pkg/abi/linux", + "//pkg/amutex", + "//pkg/fspath", + "//pkg/log", + "//pkg/sentry/arch", + "//pkg/sentry/context", + "//pkg/sentry/fs", + "//pkg/sentry/fs/fsutil", + "//pkg/sentry/kernel", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/kernel/pipe", + "//pkg/sentry/memmap", + "//pkg/sentry/pgalloc", + "//pkg/sentry/platform", + "//pkg/sentry/safemem", + "//pkg/sentry/usage", + "//pkg/sentry/usermem", + "//pkg/sentry/vfs", + "//pkg/syserror", + ], +) + +go_test( + name = "benchmark_test", + size = "small", + srcs = ["benchmark_test.go"], + deps = [ + ":tmpfs", + "//pkg/abi/linux", + "//pkg/fspath", + "//pkg/refs", + "//pkg/sentry/context", + "//pkg/sentry/context/contexttest", + "//pkg/sentry/fs", + "//pkg/sentry/fs/tmpfs", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/vfs", + "//pkg/syserror", + ], +) + +go_test( + name = "tmpfs_test", + size = "small", + srcs = [ + "pipe_test.go", + "regular_file_test.go", + ], + embed = [":tmpfs"], + deps = [ + "//pkg/abi/linux", + "//pkg/fspath", + "//pkg/sentry/context", + "//pkg/sentry/context/contexttest", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/kernel/contexttest", + "//pkg/sentry/usermem", + "//pkg/sentry/vfs", + "//pkg/syserror", + ], +) diff --git a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go new file mode 100644 index 000000000..d88c83499 --- /dev/null +++ b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go @@ -0,0 +1,487 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package benchmark_test + +import ( + "fmt" + "runtime" + "strings" + "testing" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/refs" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/sentry/fs" + _ "gvisor.dev/gvisor/pkg/sentry/fs/tmpfs" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// Differences from stat_benchmark: +// +// - Syscall interception, CopyInPath, copyOutStat, and overlayfs overheads are +// not included. +// +// - *MountStat benchmarks use a tmpfs root mount and a tmpfs submount at /tmp. +// Non-MountStat benchmarks use a tmpfs root mount and no submounts. +// stat_benchmark uses a varying root mount, a tmpfs submount at /tmp, and a +// subdirectory /tmp/ (assuming TEST_TMPDIR == "/tmp"). Thus +// stat_benchmark at depth 1 does a comparable amount of work to *MountStat +// benchmarks at depth 2, and non-MountStat benchmarks at depth 3. +var depths = []int{1, 2, 3, 8, 64, 100} + +const ( + mountPointName = "tmp" + filename = "gvisor_test_temp_0_1557494568" +) + +// This is copied from syscalls/linux/sys_file.go, with the dependency on +// kernel.Task stripped out. +func fileOpOn(ctx context.Context, mntns *fs.MountNamespace, root, wd *fs.Dirent, dirFD int32, path string, resolve bool, fn func(root *fs.Dirent, d *fs.Dirent) error) error { + var ( + d *fs.Dirent // The file. + rel *fs.Dirent // The relative directory for search (if required.) + err error + ) + + // Extract the working directory (maybe). + if len(path) > 0 && path[0] == '/' { + // Absolute path; rel can be nil. + } else if dirFD == linux.AT_FDCWD { + // Need to reference the working directory. + rel = wd + } else { + // Need to extract the given FD. + return syserror.EBADF + } + + // Lookup the node. + remainingTraversals := uint(linux.MaxSymlinkTraversals) + if resolve { + d, err = mntns.FindInode(ctx, root, rel, path, &remainingTraversals) + } else { + d, err = mntns.FindLink(ctx, root, rel, path, &remainingTraversals) + } + if err != nil { + return err + } + + err = fn(root, d) + d.DecRef() + return err +} + +func BenchmarkVFS1TmpfsStat(b *testing.B) { + for _, depth := range depths { + b.Run(fmt.Sprintf("%d", depth), func(b *testing.B) { + ctx := contexttest.Context(b) + + // Create VFS. + tmpfsFS, ok := fs.FindFilesystem("tmpfs") + if !ok { + b.Fatalf("failed to find tmpfs filesystem type") + } + rootInode, err := tmpfsFS.Mount(ctx, "tmpfs", fs.MountSourceFlags{}, "", nil) + if err != nil { + b.Fatalf("failed to create tmpfs root mount: %v", err) + } + mntns, err := fs.NewMountNamespace(ctx, rootInode) + if err != nil { + b.Fatalf("failed to create mount namespace: %v", err) + } + defer mntns.DecRef() + + var filePathBuilder strings.Builder + filePathBuilder.WriteByte('/') + + // Create nested directories with given depth. + root := mntns.Root() + defer root.DecRef() + d := root + d.IncRef() + defer d.DecRef() + for i := depth; i > 0; i-- { + name := fmt.Sprintf("%d", i) + if err := d.Inode.CreateDirectory(ctx, d, name, fs.FilePermsFromMode(0755)); err != nil { + b.Fatalf("failed to create directory %q: %v", name, err) + } + next, err := d.Walk(ctx, root, name) + if err != nil { + b.Fatalf("failed to walk to directory %q: %v", name, err) + } + d.DecRef() + d = next + filePathBuilder.WriteString(name) + filePathBuilder.WriteByte('/') + } + + // Create the file that will be stat'd. + file, err := d.Inode.Create(ctx, d, filename, fs.FileFlags{Read: true, Write: true}, fs.FilePermsFromMode(0644)) + if err != nil { + b.Fatalf("failed to create file %q: %v", filename, err) + } + file.DecRef() + filePathBuilder.WriteString(filename) + filePath := filePathBuilder.String() + + dirPath := false + runtime.GC() + b.ResetTimer() + for i := 0; i < b.N; i++ { + err := fileOpOn(ctx, mntns, root, root, linux.AT_FDCWD, filePath, true /* resolve */, func(root *fs.Dirent, d *fs.Dirent) error { + if dirPath && !fs.IsDir(d.Inode.StableAttr) { + return syserror.ENOTDIR + } + uattr, err := d.Inode.UnstableAttr(ctx) + if err != nil { + return err + } + // Sanity check. + if uattr.Perms.User.Execute { + b.Fatalf("got wrong permissions (%0o)", uattr.Perms.LinuxMode()) + } + return nil + }) + if err != nil { + b.Fatalf("stat(%q) failed: %v", filePath, err) + } + } + // Don't include deferred cleanup in benchmark time. + b.StopTimer() + }) + } +} + +func BenchmarkVFS2MemfsStat(b *testing.B) { + for _, depth := range depths { + b.Run(fmt.Sprintf("%d", depth), func(b *testing.B) { + ctx := contexttest.Context(b) + creds := auth.CredentialsFromContext(ctx) + + // Create VFS. + vfsObj := vfs.New() + vfsObj.MustRegisterFilesystemType("tmpfs", tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + }) + mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{}) + if err != nil { + b.Fatalf("failed to create tmpfs root mount: %v", err) + } + defer mntns.DecRef(vfsObj) + + var filePathBuilder strings.Builder + filePathBuilder.WriteByte('/') + + // Create nested directories with given depth. + root := mntns.Root() + defer root.DecRef() + vd := root + vd.IncRef() + for i := depth; i > 0; i-- { + name := fmt.Sprintf("%d", i) + pop := vfs.PathOperation{ + Root: root, + Start: vd, + Path: fspath.Parse(name), + } + if err := vfsObj.MkdirAt(ctx, creds, &pop, &vfs.MkdirOptions{ + Mode: 0755, + }); err != nil { + b.Fatalf("failed to create directory %q: %v", name, err) + } + nextVD, err := vfsObj.GetDentryAt(ctx, creds, &pop, &vfs.GetDentryOptions{}) + if err != nil { + b.Fatalf("failed to walk to directory %q: %v", name, err) + } + vd.DecRef() + vd = nextVD + filePathBuilder.WriteString(name) + filePathBuilder.WriteByte('/') + } + + // Create the file that will be stat'd. + fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ + Root: root, + Start: vd, + Path: fspath.Parse(filename), + FollowFinalSymlink: true, + }, &vfs.OpenOptions{ + Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL, + Mode: 0644, + }) + vd.DecRef() + vd = vfs.VirtualDentry{} + if err != nil { + b.Fatalf("failed to create file %q: %v", filename, err) + } + defer fd.DecRef() + filePathBuilder.WriteString(filename) + filePath := filePathBuilder.String() + + runtime.GC() + b.ResetTimer() + for i := 0; i < b.N; i++ { + stat, err := vfsObj.StatAt(ctx, creds, &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(filePath), + FollowFinalSymlink: true, + }, &vfs.StatOptions{}) + if err != nil { + b.Fatalf("stat(%q) failed: %v", filePath, err) + } + // Sanity check. + if stat.Mode&^linux.S_IFMT != 0644 { + b.Fatalf("got wrong permissions (%0o)", stat.Mode) + } + } + // Don't include deferred cleanup in benchmark time. + b.StopTimer() + }) + } +} + +func BenchmarkVFS1TmpfsMountStat(b *testing.B) { + for _, depth := range depths { + b.Run(fmt.Sprintf("%d", depth), func(b *testing.B) { + ctx := contexttest.Context(b) + + // Create VFS. + tmpfsFS, ok := fs.FindFilesystem("tmpfs") + if !ok { + b.Fatalf("failed to find tmpfs filesystem type") + } + rootInode, err := tmpfsFS.Mount(ctx, "tmpfs", fs.MountSourceFlags{}, "", nil) + if err != nil { + b.Fatalf("failed to create tmpfs root mount: %v", err) + } + mntns, err := fs.NewMountNamespace(ctx, rootInode) + if err != nil { + b.Fatalf("failed to create mount namespace: %v", err) + } + defer mntns.DecRef() + + var filePathBuilder strings.Builder + filePathBuilder.WriteByte('/') + + // Create and mount the submount. + root := mntns.Root() + defer root.DecRef() + if err := root.Inode.CreateDirectory(ctx, root, mountPointName, fs.FilePermsFromMode(0755)); err != nil { + b.Fatalf("failed to create mount point: %v", err) + } + mountPoint, err := root.Walk(ctx, root, mountPointName) + if err != nil { + b.Fatalf("failed to walk to mount point: %v", err) + } + defer mountPoint.DecRef() + submountInode, err := tmpfsFS.Mount(ctx, "tmpfs", fs.MountSourceFlags{}, "", nil) + if err != nil { + b.Fatalf("failed to create tmpfs submount: %v", err) + } + if err := mntns.Mount(ctx, mountPoint, submountInode); err != nil { + b.Fatalf("failed to mount tmpfs submount: %v", err) + } + filePathBuilder.WriteString(mountPointName) + filePathBuilder.WriteByte('/') + + // Create nested directories with given depth. + d, err := root.Walk(ctx, root, mountPointName) + if err != nil { + b.Fatalf("failed to walk to mount root: %v", err) + } + defer d.DecRef() + for i := depth; i > 0; i-- { + name := fmt.Sprintf("%d", i) + if err := d.Inode.CreateDirectory(ctx, d, name, fs.FilePermsFromMode(0755)); err != nil { + b.Fatalf("failed to create directory %q: %v", name, err) + } + next, err := d.Walk(ctx, root, name) + if err != nil { + b.Fatalf("failed to walk to directory %q: %v", name, err) + } + d.DecRef() + d = next + filePathBuilder.WriteString(name) + filePathBuilder.WriteByte('/') + } + + // Create the file that will be stat'd. + file, err := d.Inode.Create(ctx, d, filename, fs.FileFlags{Read: true, Write: true}, fs.FilePermsFromMode(0644)) + if err != nil { + b.Fatalf("failed to create file %q: %v", filename, err) + } + file.DecRef() + filePathBuilder.WriteString(filename) + filePath := filePathBuilder.String() + + dirPath := false + runtime.GC() + b.ResetTimer() + for i := 0; i < b.N; i++ { + err := fileOpOn(ctx, mntns, root, root, linux.AT_FDCWD, filePath, true /* resolve */, func(root *fs.Dirent, d *fs.Dirent) error { + if dirPath && !fs.IsDir(d.Inode.StableAttr) { + return syserror.ENOTDIR + } + uattr, err := d.Inode.UnstableAttr(ctx) + if err != nil { + return err + } + // Sanity check. + if uattr.Perms.User.Execute { + b.Fatalf("got wrong permissions (%0o)", uattr.Perms.LinuxMode()) + } + return nil + }) + if err != nil { + b.Fatalf("stat(%q) failed: %v", filePath, err) + } + } + // Don't include deferred cleanup in benchmark time. + b.StopTimer() + }) + } +} + +func BenchmarkVFS2MemfsMountStat(b *testing.B) { + for _, depth := range depths { + b.Run(fmt.Sprintf("%d", depth), func(b *testing.B) { + ctx := contexttest.Context(b) + creds := auth.CredentialsFromContext(ctx) + + // Create VFS. + vfsObj := vfs.New() + vfsObj.MustRegisterFilesystemType("tmpfs", tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + }) + mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{}) + if err != nil { + b.Fatalf("failed to create tmpfs root mount: %v", err) + } + defer mntns.DecRef(vfsObj) + + var filePathBuilder strings.Builder + filePathBuilder.WriteByte('/') + + // Create the mount point. + root := mntns.Root() + defer root.DecRef() + pop := vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(mountPointName), + } + if err := vfsObj.MkdirAt(ctx, creds, &pop, &vfs.MkdirOptions{ + Mode: 0755, + }); err != nil { + b.Fatalf("failed to create mount point: %v", err) + } + // Save the mount point for later use. + mountPoint, err := vfsObj.GetDentryAt(ctx, creds, &pop, &vfs.GetDentryOptions{}) + if err != nil { + b.Fatalf("failed to walk to mount point: %v", err) + } + defer mountPoint.DecRef() + // Create and mount the submount. + if err := vfsObj.MountAt(ctx, creds, "", &pop, "tmpfs", &vfs.MountOptions{}); err != nil { + b.Fatalf("failed to mount tmpfs submount: %v", err) + } + filePathBuilder.WriteString(mountPointName) + filePathBuilder.WriteByte('/') + + // Create nested directories with given depth. + vd, err := vfsObj.GetDentryAt(ctx, creds, &pop, &vfs.GetDentryOptions{}) + if err != nil { + b.Fatalf("failed to walk to mount root: %v", err) + } + for i := depth; i > 0; i-- { + name := fmt.Sprintf("%d", i) + pop := vfs.PathOperation{ + Root: root, + Start: vd, + Path: fspath.Parse(name), + } + if err := vfsObj.MkdirAt(ctx, creds, &pop, &vfs.MkdirOptions{ + Mode: 0755, + }); err != nil { + b.Fatalf("failed to create directory %q: %v", name, err) + } + nextVD, err := vfsObj.GetDentryAt(ctx, creds, &pop, &vfs.GetDentryOptions{}) + if err != nil { + b.Fatalf("failed to walk to directory %q: %v", name, err) + } + vd.DecRef() + vd = nextVD + filePathBuilder.WriteString(name) + filePathBuilder.WriteByte('/') + } + + // Verify that we didn't create any directories under the mount + // point (i.e. they were all created on the submount). + firstDirName := fmt.Sprintf("%d", depth) + if child := mountPoint.Dentry().Child(firstDirName); child != nil { + b.Fatalf("created directory %q under root mount, not submount", firstDirName) + } + + // Create the file that will be stat'd. + fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ + Root: root, + Start: vd, + Path: fspath.Parse(filename), + FollowFinalSymlink: true, + }, &vfs.OpenOptions{ + Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL, + Mode: 0644, + }) + vd.DecRef() + if err != nil { + b.Fatalf("failed to create file %q: %v", filename, err) + } + fd.DecRef() + filePathBuilder.WriteString(filename) + filePath := filePathBuilder.String() + + runtime.GC() + b.ResetTimer() + for i := 0; i < b.N; i++ { + stat, err := vfsObj.StatAt(ctx, creds, &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(filePath), + FollowFinalSymlink: true, + }, &vfs.StatOptions{}) + if err != nil { + b.Fatalf("stat(%q) failed: %v", filePath, err) + } + // Sanity check. + if stat.Mode&^linux.S_IFMT != 0644 { + b.Fatalf("got wrong permissions (%0o)", stat.Mode) + } + } + // Don't include deferred cleanup in benchmark time. + b.StopTimer() + }) + } +} + +func init() { + // Turn off reference leak checking for a fair comparison between vfs1 and + // vfs2. + refs.SetLeakMode(refs.NoLeakChecking) +} diff --git a/pkg/sentry/fsimpl/tmpfs/directory.go b/pkg/sentry/fsimpl/tmpfs/directory.go new file mode 100644 index 000000000..887ca2619 --- /dev/null +++ b/pkg/sentry/fsimpl/tmpfs/directory.go @@ -0,0 +1,187 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tmpfs + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +type directory struct { + inode inode + + // childList is a list containing (1) child Dentries and (2) fake Dentries + // (with inode == nil) that represent the iteration position of + // directoryFDs. childList is used to support directoryFD.IterDirents() + // efficiently. childList is protected by filesystem.mu. + childList dentryList +} + +func (fs *filesystem) newDirectory(creds *auth.Credentials, mode linux.FileMode) *inode { + dir := &directory{} + dir.inode.init(dir, fs, creds, mode) + dir.inode.nlink = 2 // from "." and parent directory or ".." for root + return &dir.inode +} + +func (i *inode) isDir() bool { + _, ok := i.impl.(*directory) + return ok +} + +type directoryFD struct { + fileDescription + vfs.DirectoryFileDescriptionDefaultImpl + + // Protected by filesystem.mu. + iter *dentry + off int64 +} + +// Release implements vfs.FileDescriptionImpl.Release. +func (fd *directoryFD) Release() { + if fd.iter != nil { + fs := fd.filesystem() + dir := fd.inode().impl.(*directory) + fs.mu.Lock() + dir.childList.Remove(fd.iter) + fs.mu.Unlock() + fd.iter = nil + } +} + +// IterDirents implements vfs.FileDescriptionImpl.IterDirents. +func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback) error { + fs := fd.filesystem() + vfsd := fd.vfsfd.VirtualDentry().Dentry() + + fs.mu.Lock() + defer fs.mu.Unlock() + + if fd.off == 0 { + if !cb.Handle(vfs.Dirent{ + Name: ".", + Type: linux.DT_DIR, + Ino: vfsd.Impl().(*dentry).inode.ino, + NextOff: 1, + }) { + return nil + } + fd.off++ + } + if fd.off == 1 { + parentInode := vfsd.ParentOrSelf().Impl().(*dentry).inode + if !cb.Handle(vfs.Dirent{ + Name: "..", + Type: parentInode.direntType(), + Ino: parentInode.ino, + NextOff: 2, + }) { + return nil + } + fd.off++ + } + + dir := vfsd.Impl().(*dentry).inode.impl.(*directory) + var child *dentry + if fd.iter == nil { + // Start iteration at the beginning of dir. + child = dir.childList.Front() + fd.iter = &dentry{} + } else { + // Continue iteration from where we left off. + child = fd.iter.Next() + dir.childList.Remove(fd.iter) + } + for child != nil { + // Skip other directoryFD iterators. + if child.inode != nil { + if !cb.Handle(vfs.Dirent{ + Name: child.vfsd.Name(), + Type: child.inode.direntType(), + Ino: child.inode.ino, + NextOff: fd.off + 1, + }) { + dir.childList.InsertBefore(child, fd.iter) + return nil + } + fd.off++ + } + child = child.Next() + } + dir.childList.PushBack(fd.iter) + return nil +} + +// Seek implements vfs.FileDescriptionImpl.Seek. +func (fd *directoryFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + fs := fd.filesystem() + fs.mu.Lock() + defer fs.mu.Unlock() + + switch whence { + case linux.SEEK_SET: + // Use offset as given. + case linux.SEEK_CUR: + offset += fd.off + default: + return 0, syserror.EINVAL + } + if offset < 0 { + return 0, syserror.EINVAL + } + + // If the offset isn't changing (e.g. due to lseek(0, SEEK_CUR)), don't + // seek even if doing so might reposition the iterator due to concurrent + // mutation of the directory. Compare fs/libfs.c:dcache_dir_lseek(). + if fd.off == offset { + return offset, nil + } + + fd.off = offset + // Compensate for "." and "..". + remChildren := int64(0) + if offset >= 2 { + remChildren = offset - 2 + } + + dir := fd.inode().impl.(*directory) + + // Ensure that fd.iter exists and is not linked into dir.childList. + if fd.iter == nil { + fd.iter = &dentry{} + } else { + dir.childList.Remove(fd.iter) + } + // Insert fd.iter before the remChildren'th child, or at the end of the + // list if remChildren >= number of children. + child := dir.childList.Front() + for child != nil { + // Skip other directoryFD iterators. + if child.inode != nil { + if remChildren == 0 { + dir.childList.InsertBefore(child, fd.iter) + return offset, nil + } + remChildren-- + } + child = child.Next() + } + dir.childList.PushBack(fd.iter) + return offset, nil +} diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go new file mode 100644 index 000000000..26979729e --- /dev/null +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -0,0 +1,698 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tmpfs + +import ( + "fmt" + "sync/atomic" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// Sync implements vfs.FilesystemImpl.Sync. +func (fs *filesystem) Sync(ctx context.Context) error { + // All filesystem state is in-memory. + return nil +} + +// stepLocked resolves rp.Component() to an existing file, starting from the +// given directory. +// +// stepLocked is loosely analogous to fs/namei.c:walk_component(). +// +// Preconditions: filesystem.mu must be locked. !rp.Done(). +func stepLocked(rp *vfs.ResolvingPath, d *dentry) (*dentry, error) { + if !d.inode.isDir() { + return nil, syserror.ENOTDIR + } + if err := d.inode.checkPermissions(rp.Credentials(), vfs.MayExec, true); err != nil { + return nil, err + } +afterSymlink: + nextVFSD, err := rp.ResolveComponent(&d.vfsd) + if err != nil { + return nil, err + } + if nextVFSD == nil { + // Since the Dentry tree is the sole source of truth for tmpfs, if it's + // not in the Dentry tree, it doesn't exist. + return nil, syserror.ENOENT + } + next := nextVFSD.Impl().(*dentry) + if symlink, ok := next.inode.impl.(*symlink); ok && rp.ShouldFollowSymlink() { + // TODO: symlink traversals update access time + if err := rp.HandleSymlink(symlink.target); err != nil { + return nil, err + } + goto afterSymlink // don't check the current directory again + } + rp.Advance() + return next, nil +} + +// walkParentDirLocked resolves all but the last path component of rp to an +// existing directory, starting from the given directory (which is usually +// rp.Start().Impl().(*dentry)). It does not check that the returned directory +// is searchable by the provider of rp. +// +// walkParentDirLocked is loosely analogous to Linux's +// fs/namei.c:path_parentat(). +// +// Preconditions: filesystem.mu must be locked. !rp.Done(). +func walkParentDirLocked(rp *vfs.ResolvingPath, d *dentry) (*dentry, error) { + for !rp.Final() { + next, err := stepLocked(rp, d) + if err != nil { + return nil, err + } + d = next + } + if !d.inode.isDir() { + return nil, syserror.ENOTDIR + } + return d, nil +} + +// resolveLocked resolves rp to an existing file. +// +// resolveLocked is loosely analogous to Linux's fs/namei.c:path_lookupat(). +// +// Preconditions: filesystem.mu must be locked. +func resolveLocked(rp *vfs.ResolvingPath) (*dentry, error) { + d := rp.Start().Impl().(*dentry) + for !rp.Done() { + next, err := stepLocked(rp, d) + if err != nil { + return nil, err + } + d = next + } + if rp.MustBeDir() && !d.inode.isDir() { + return nil, syserror.ENOTDIR + } + return d, nil +} + +// doCreateAt checks that creating a file at rp is permitted, then invokes +// create to do so. +// +// doCreateAt is loosely analogous to a conjunction of Linux's +// fs/namei.c:filename_create() and done_path_create(). +// +// Preconditions: !rp.Done(). For the final path component in rp, +// !rp.ShouldFollowSymlink(). +func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(parent *dentry, name string) error) error { + fs.mu.Lock() + defer fs.mu.Unlock() + parent, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) + if err != nil { + return err + } + if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true /* isDir */); err != nil { + return err + } + name := rp.Component() + if name == "." || name == ".." { + return syserror.EEXIST + } + // Call parent.vfsd.Child() instead of stepLocked() or rp.ResolveChild(), + // because if the child exists we want to return EEXIST immediately instead + // of attempting symlink/mount traversal. + if parent.vfsd.Child(name) != nil { + return syserror.EEXIST + } + if !dir && rp.MustBeDir() { + return syserror.ENOENT + } + // In memfs, the only way to cause a dentry to be disowned is by removing + // it from the filesystem, so this check is equivalent to checking if + // parent has been removed. + if parent.vfsd.IsDisowned() { + return syserror.ENOENT + } + mnt := rp.Mount() + if err := mnt.CheckBeginWrite(); err != nil { + return err + } + defer mnt.EndWrite() + return create(parent, name) +} + +// GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. +func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { + fs.mu.RLock() + defer fs.mu.RUnlock() + d, err := resolveLocked(rp) + if err != nil { + return nil, err + } + if opts.CheckSearchable { + if !d.inode.isDir() { + return nil, syserror.ENOTDIR + } + if err := d.inode.checkPermissions(rp.Credentials(), vfs.MayExec, true /* isDir */); err != nil { + return nil, err + } + } + d.IncRef() + return &d.vfsd, nil +} + +// GetParentDentryAt implements vfs.FilesystemImpl.GetParentDentryAt. +func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { + fs.mu.RLock() + defer fs.mu.RUnlock() + d, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) + if err != nil { + return nil, err + } + d.IncRef() + return &d.vfsd, nil +} + +// LinkAt implements vfs.FilesystemImpl.LinkAt. +func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry) error { + return fs.doCreateAt(rp, false /* dir */, func(parent *dentry, name string) error { + if rp.Mount() != vd.Mount() { + return syserror.EXDEV + } + d := vd.Dentry().Impl().(*dentry) + if d.inode.isDir() { + return syserror.EPERM + } + if d.inode.nlink == 0 { + return syserror.ENOENT + } + if d.inode.nlink == maxLinks { + return syserror.EMLINK + } + d.inode.incLinksLocked() + child := fs.newDentry(d.inode) + parent.vfsd.InsertChild(&child.vfsd, name) + parent.inode.impl.(*directory).childList.PushBack(child) + return nil + }) +} + +// MkdirAt implements vfs.FilesystemImpl.MkdirAt. +func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error { + return fs.doCreateAt(rp, true /* dir */, func(parent *dentry, name string) error { + if parent.inode.nlink == maxLinks { + return syserror.EMLINK + } + parent.inode.incLinksLocked() // from child's ".." + child := fs.newDentry(fs.newDirectory(rp.Credentials(), opts.Mode)) + parent.vfsd.InsertChild(&child.vfsd, name) + parent.inode.impl.(*directory).childList.PushBack(child) + return nil + }) +} + +// MknodAt implements vfs.FilesystemImpl.MknodAt. +func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MknodOptions) error { + return fs.doCreateAt(rp, false /* dir */, func(parent *dentry, name string) error { + switch opts.Mode.FileType() { + case 0, linux.S_IFREG: + child := fs.newDentry(fs.newRegularFile(rp.Credentials(), opts.Mode)) + parent.vfsd.InsertChild(&child.vfsd, name) + parent.inode.impl.(*directory).childList.PushBack(child) + return nil + case linux.S_IFIFO: + child := fs.newDentry(fs.newNamedPipe(rp.Credentials(), opts.Mode)) + parent.vfsd.InsertChild(&child.vfsd, name) + parent.inode.impl.(*directory).childList.PushBack(child) + return nil + case linux.S_IFBLK, linux.S_IFCHR, linux.S_IFSOCK: + // Not yet supported. + return syserror.EPERM + default: + return syserror.EINVAL + } + }) +} + +// OpenAt implements vfs.FilesystemImpl.OpenAt. +func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + if opts.Flags&linux.O_TMPFILE != 0 { + // Not yet supported. + return nil, syserror.EOPNOTSUPP + } + + // Handle O_CREAT and !O_CREAT separately, since in the latter case we + // don't need fs.mu for writing. + if opts.Flags&linux.O_CREAT == 0 { + fs.mu.RLock() + defer fs.mu.RUnlock() + d, err := resolveLocked(rp) + if err != nil { + return nil, err + } + return d.open(ctx, rp, opts.Flags, false /* afterCreate */) + } + + mustCreate := opts.Flags&linux.O_EXCL != 0 + start := rp.Start().Impl().(*dentry) + fs.mu.Lock() + defer fs.mu.Unlock() + if rp.Done() { + // Reject attempts to open directories with O_CREAT. + if rp.MustBeDir() { + return nil, syserror.EISDIR + } + if mustCreate { + return nil, syserror.EEXIST + } + return start.open(ctx, rp, opts.Flags, false /* afterCreate */) + } +afterTrailingSymlink: + parent, err := walkParentDirLocked(rp, start) + if err != nil { + return nil, err + } + // Check for search permission in the parent directory. + if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayExec, true); err != nil { + return nil, err + } + // Reject attempts to open directories with O_CREAT. + if rp.MustBeDir() { + return nil, syserror.EISDIR + } + name := rp.Component() + if name == "." || name == ".." { + return nil, syserror.EISDIR + } + // Determine whether or not we need to create a file. + child, err := stepLocked(rp, parent) + if err == syserror.ENOENT { + // Already checked for searchability above; now check for writability. + if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite, true); err != nil { + return nil, err + } + if err := rp.Mount().CheckBeginWrite(); err != nil { + return nil, err + } + defer rp.Mount().EndWrite() + // Create and open the child. + child := fs.newDentry(fs.newRegularFile(rp.Credentials(), opts.Mode)) + parent.vfsd.InsertChild(&child.vfsd, name) + parent.inode.impl.(*directory).childList.PushBack(child) + return child.open(ctx, rp, opts.Flags, true) + } + if err != nil { + return nil, err + } + // Do we need to resolve a trailing symlink? + if !rp.Done() { + start = parent + goto afterTrailingSymlink + } + // Open existing file. + if mustCreate { + return nil, syserror.EEXIST + } + return child.open(ctx, rp, opts.Flags, false) +} + +func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, flags uint32, afterCreate bool) (*vfs.FileDescription, error) { + ats := vfs.AccessTypesForOpenFlags(flags) + if !afterCreate { + if err := d.inode.checkPermissions(rp.Credentials(), ats, d.inode.isDir()); err != nil { + return nil, err + } + } + mnt := rp.Mount() + switch impl := d.inode.impl.(type) { + case *regularFile: + var fd regularFileFD + fd.readable = vfs.MayReadFileWithOpenFlags(flags) + fd.writable = vfs.MayWriteFileWithOpenFlags(flags) + if fd.writable { + if err := mnt.CheckBeginWrite(); err != nil { + return nil, err + } + // mnt.EndWrite() is called by regularFileFD.Release(). + } + fd.vfsfd.Init(&fd, flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}) + if flags&linux.O_TRUNC != 0 { + impl.mu.Lock() + impl.data.Truncate(0, impl.memFile) + atomic.StoreUint64(&impl.size, 0) + impl.mu.Unlock() + } + return &fd.vfsfd, nil + case *directory: + // Can't open directories writably. + if ats&vfs.MayWrite != 0 { + return nil, syserror.EISDIR + } + var fd directoryFD + fd.vfsfd.Init(&fd, flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}) + return &fd.vfsfd, nil + case *symlink: + // Can't open symlinks without O_PATH (which is unimplemented). + return nil, syserror.ELOOP + case *namedPipe: + return newNamedPipeFD(ctx, impl, rp, &d.vfsd, flags) + default: + panic(fmt.Sprintf("unknown inode type: %T", d.inode.impl)) + } +} + +// ReadlinkAt implements vfs.FilesystemImpl.ReadlinkAt. +func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { + fs.mu.RLock() + defer fs.mu.RUnlock() + d, err := resolveLocked(rp) + if err != nil { + return "", err + } + symlink, ok := d.inode.impl.(*symlink) + if !ok { + return "", syserror.EINVAL + } + return symlink.target, nil +} + +// RenameAt implements vfs.FilesystemImpl.RenameAt. +func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldParentVD vfs.VirtualDentry, oldName string, opts vfs.RenameOptions) error { + if opts.Flags != 0 { + // TODO(b/145974740): Support renameat2 flags. + return syserror.EINVAL + } + + // Resolve newParent first to verify that it's on this Mount. + fs.mu.Lock() + defer fs.mu.Unlock() + newParent, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) + if err != nil { + return err + } + newName := rp.Component() + if newName == "." || newName == ".." { + return syserror.EBUSY + } + mnt := rp.Mount() + if mnt != oldParentVD.Mount() { + return syserror.EXDEV + } + if err := mnt.CheckBeginWrite(); err != nil { + return err + } + defer mnt.EndWrite() + + oldParent := oldParentVD.Dentry().Impl().(*dentry) + if err := oldParent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true /* isDir */); err != nil { + return err + } + // Call vfs.Dentry.Child() instead of stepLocked() or rp.ResolveChild(), + // because if the existing child is a symlink or mount point then we want + // to rename over it rather than follow it. + renamedVFSD := oldParent.vfsd.Child(oldName) + if renamedVFSD == nil { + return syserror.ENOENT + } + renamed := renamedVFSD.Impl().(*dentry) + if renamed.inode.isDir() { + if renamed == newParent || renamedVFSD.IsAncestorOf(&newParent.vfsd) { + return syserror.EINVAL + } + if oldParent != newParent { + // Writability is needed to change renamed's "..". + if err := renamed.inode.checkPermissions(rp.Credentials(), vfs.MayWrite, true /* isDir */); err != nil { + return err + } + } + } else { + if opts.MustBeDir || rp.MustBeDir() { + return syserror.ENOTDIR + } + } + + if err := newParent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true /* isDir */); err != nil { + return err + } + replacedVFSD := newParent.vfsd.Child(newName) + var replaced *dentry + if replacedVFSD != nil { + replaced = replacedVFSD.Impl().(*dentry) + if replaced.inode.isDir() { + if !renamed.inode.isDir() { + return syserror.EISDIR + } + if replaced.vfsd.HasChildren() { + return syserror.ENOTEMPTY + } + } else { + if rp.MustBeDir() { + return syserror.ENOTDIR + } + if renamed.inode.isDir() { + return syserror.ENOTDIR + } + } + } else { + if renamed.inode.isDir() && newParent.inode.nlink == maxLinks { + return syserror.EMLINK + } + } + if newParent.vfsd.IsDisowned() { + return syserror.ENOENT + } + + // Linux places this check before some of those above; we do it here for + // simplicity, under the assumption that applications are not intentionally + // doing noop renames expecting them to succeed where non-noop renames + // would fail. + if renamedVFSD == replacedVFSD { + return nil + } + vfsObj := rp.VirtualFilesystem() + oldParentDir := oldParent.inode.impl.(*directory) + newParentDir := newParent.inode.impl.(*directory) + if err := vfsObj.PrepareRenameDentry(vfs.MountNamespaceFromContext(ctx), renamedVFSD, replacedVFSD); err != nil { + return err + } + if replaced != nil { + newParentDir.childList.Remove(replaced) + if replaced.inode.isDir() { + newParent.inode.decLinksLocked() // from replaced's ".." + } + replaced.inode.decLinksLocked() + } + oldParentDir.childList.Remove(renamed) + newParentDir.childList.PushBack(renamed) + if renamed.inode.isDir() { + oldParent.inode.decLinksLocked() + newParent.inode.incLinksLocked() + } + // TODO: update timestamps and parent directory sizes + vfsObj.CommitRenameReplaceDentry(renamedVFSD, &newParent.vfsd, newName, replacedVFSD) + return nil +} + +// RmdirAt implements vfs.FilesystemImpl.RmdirAt. +func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error { + fs.mu.Lock() + defer fs.mu.Unlock() + parent, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) + if err != nil { + return err + } + if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true /* isDir */); err != nil { + return err + } + name := rp.Component() + if name == "." { + return syserror.EINVAL + } + if name == ".." { + return syserror.ENOTEMPTY + } + childVFSD := parent.vfsd.Child(name) + if childVFSD == nil { + return syserror.ENOENT + } + child := childVFSD.Impl().(*dentry) + if !child.inode.isDir() { + return syserror.ENOTDIR + } + if childVFSD.HasChildren() { + return syserror.ENOTEMPTY + } + mnt := rp.Mount() + if err := mnt.CheckBeginWrite(); err != nil { + return err + } + defer mnt.EndWrite() + vfsObj := rp.VirtualFilesystem() + if err := vfsObj.PrepareDeleteDentry(vfs.MountNamespaceFromContext(ctx), childVFSD); err != nil { + return err + } + parent.inode.impl.(*directory).childList.Remove(child) + parent.inode.decLinksLocked() // from child's ".." + child.inode.decLinksLocked() + vfsObj.CommitDeleteDentry(childVFSD) + return nil +} + +// SetStatAt implements vfs.FilesystemImpl.SetStatAt. +func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { + fs.mu.RLock() + defer fs.mu.RUnlock() + _, err := resolveLocked(rp) + if err != nil { + return err + } + if opts.Stat.Mask == 0 { + return nil + } + // TODO: implement inode.setStat + return syserror.EPERM +} + +// StatAt implements vfs.FilesystemImpl.StatAt. +func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { + fs.mu.RLock() + defer fs.mu.RUnlock() + d, err := resolveLocked(rp) + if err != nil { + return linux.Statx{}, err + } + var stat linux.Statx + d.inode.statTo(&stat) + return stat, nil +} + +// StatFSAt implements vfs.FilesystemImpl.StatFSAt. +func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { + fs.mu.RLock() + defer fs.mu.RUnlock() + _, err := resolveLocked(rp) + if err != nil { + return linux.Statfs{}, err + } + // TODO: actually implement statfs + return linux.Statfs{}, syserror.ENOSYS +} + +// SymlinkAt implements vfs.FilesystemImpl.SymlinkAt. +func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, target string) error { + return fs.doCreateAt(rp, false /* dir */, func(parent *dentry, name string) error { + child := fs.newDentry(fs.newSymlink(rp.Credentials(), target)) + parent.vfsd.InsertChild(&child.vfsd, name) + parent.inode.impl.(*directory).childList.PushBack(child) + return nil + }) +} + +// UnlinkAt implements vfs.FilesystemImpl.UnlinkAt. +func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error { + fs.mu.Lock() + defer fs.mu.Unlock() + parent, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) + if err != nil { + return err + } + if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true /* isDir */); err != nil { + return err + } + name := rp.Component() + if name == "." || name == ".." { + return syserror.EISDIR + } + childVFSD := parent.vfsd.Child(name) + if childVFSD == nil { + return syserror.ENOENT + } + child := childVFSD.Impl().(*dentry) + if child.inode.isDir() { + return syserror.EISDIR + } + if !rp.MustBeDir() { + return syserror.ENOTDIR + } + mnt := rp.Mount() + if err := mnt.CheckBeginWrite(); err != nil { + return err + } + defer mnt.EndWrite() + vfsObj := rp.VirtualFilesystem() + if err := vfsObj.PrepareDeleteDentry(vfs.MountNamespaceFromContext(ctx), childVFSD); err != nil { + return err + } + parent.inode.impl.(*directory).childList.Remove(child) + child.inode.decLinksLocked() + vfsObj.CommitDeleteDentry(childVFSD) + return nil +} + +// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. +func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) { + fs.mu.RLock() + defer fs.mu.RUnlock() + _, err := resolveLocked(rp) + if err != nil { + return nil, err + } + // TODO(b/127675828): support extended attributes + return nil, syserror.ENOTSUP +} + +// GetxattrAt implements vfs.FilesystemImpl.GetxattrAt. +func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) (string, error) { + fs.mu.RLock() + defer fs.mu.RUnlock() + _, err := resolveLocked(rp) + if err != nil { + return "", err + } + // TODO(b/127675828): support extended attributes + return "", syserror.ENOTSUP +} + +// SetxattrAt implements vfs.FilesystemImpl.SetxattrAt. +func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { + fs.mu.RLock() + defer fs.mu.RUnlock() + _, err := resolveLocked(rp) + if err != nil { + return err + } + // TODO(b/127675828): support extended attributes + return syserror.ENOTSUP +} + +// RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. +func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { + fs.mu.RLock() + defer fs.mu.RUnlock() + _, err := resolveLocked(rp) + if err != nil { + return err + } + // TODO(b/127675828): support extended attributes + return syserror.ENOTSUP +} + +// PrependPath implements vfs.FilesystemImpl.PrependPath. +func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { + fs.mu.RLock() + defer fs.mu.RUnlock() + return vfs.GenericPrependPath(vfsroot, vd, b) +} diff --git a/pkg/sentry/fsimpl/tmpfs/named_pipe.go b/pkg/sentry/fsimpl/tmpfs/named_pipe.go new file mode 100644 index 000000000..40bde54de --- /dev/null +++ b/pkg/sentry/fsimpl/tmpfs/named_pipe.go @@ -0,0 +1,60 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tmpfs + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" + "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sentry/vfs" +) + +type namedPipe struct { + inode inode + + pipe *pipe.VFSPipe +} + +// Preconditions: +// * fs.mu must be locked. +// * rp.Mount().CheckBeginWrite() has been called successfully. +func (fs *filesystem) newNamedPipe(creds *auth.Credentials, mode linux.FileMode) *inode { + file := &namedPipe{pipe: pipe.NewVFSPipe(pipe.DefaultPipeSize, usermem.PageSize)} + file.inode.init(file, fs, creds, mode) + file.inode.nlink = 1 // Only the parent has a link. + return &file.inode +} + +// namedPipeFD implements vfs.FileDescriptionImpl. Methods are implemented +// entirely via struct embedding. +type namedPipeFD struct { + fileDescription + + *pipe.VFSPipeFD +} + +func newNamedPipeFD(ctx context.Context, np *namedPipe, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { + var err error + var fd namedPipeFD + fd.VFSPipeFD, err = np.pipe.NewVFSPipeFD(ctx, rp, vfsd, &fd.vfsfd, flags) + if err != nil { + return nil, err + } + mnt := rp.Mount() + fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{}) + return &fd.vfsfd, nil +} diff --git a/pkg/sentry/fsimpl/tmpfs/pipe_test.go b/pkg/sentry/fsimpl/tmpfs/pipe_test.go new file mode 100644 index 000000000..70b42a6ec --- /dev/null +++ b/pkg/sentry/fsimpl/tmpfs/pipe_test.go @@ -0,0 +1,235 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tmpfs + +import ( + "bytes" + "testing" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +const fileName = "mypipe" + +func TestSeparateFDs(t *testing.T) { + ctx, creds, vfsObj, root := setup(t) + defer root.DecRef() + + // Open the read side. This is done in a concurrently because opening + // One end the pipe blocks until the other end is opened. + pop := vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(fileName), + FollowFinalSymlink: true, + } + rfdchan := make(chan *vfs.FileDescription) + go func() { + openOpts := vfs.OpenOptions{Flags: linux.O_RDONLY} + rfd, _ := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) + rfdchan <- rfd + }() + + // Open the write side. + openOpts := vfs.OpenOptions{Flags: linux.O_WRONLY} + wfd, err := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) + if err != nil { + t.Fatalf("failed to open pipe for writing %q: %v", fileName, err) + } + defer wfd.DecRef() + + rfd, ok := <-rfdchan + if !ok { + t.Fatalf("failed to open pipe for reading %q", fileName) + } + defer rfd.DecRef() + + const msg = "vamos azul" + checkEmpty(ctx, t, rfd) + checkWrite(ctx, t, wfd, msg) + checkRead(ctx, t, rfd, msg) +} + +func TestNonblockingRead(t *testing.T) { + ctx, creds, vfsObj, root := setup(t) + defer root.DecRef() + + // Open the read side as nonblocking. + pop := vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(fileName), + FollowFinalSymlink: true, + } + openOpts := vfs.OpenOptions{Flags: linux.O_RDONLY | linux.O_NONBLOCK} + rfd, err := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) + if err != nil { + t.Fatalf("failed to open pipe for reading %q: %v", fileName, err) + } + defer rfd.DecRef() + + // Open the write side. + openOpts = vfs.OpenOptions{Flags: linux.O_WRONLY} + wfd, err := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) + if err != nil { + t.Fatalf("failed to open pipe for writing %q: %v", fileName, err) + } + defer wfd.DecRef() + + const msg = "geh blau" + checkEmpty(ctx, t, rfd) + checkWrite(ctx, t, wfd, msg) + checkRead(ctx, t, rfd, msg) +} + +func TestNonblockingWriteError(t *testing.T) { + ctx, creds, vfsObj, root := setup(t) + defer root.DecRef() + + // Open the write side as nonblocking, which should return ENXIO. + pop := vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(fileName), + FollowFinalSymlink: true, + } + openOpts := vfs.OpenOptions{Flags: linux.O_WRONLY | linux.O_NONBLOCK} + _, err := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) + if err != syserror.ENXIO { + t.Fatalf("expected ENXIO, but got error: %v", err) + } +} + +func TestSingleFD(t *testing.T) { + ctx, creds, vfsObj, root := setup(t) + defer root.DecRef() + + // Open the pipe as readable and writable. + pop := vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(fileName), + FollowFinalSymlink: true, + } + openOpts := vfs.OpenOptions{Flags: linux.O_RDWR} + fd, err := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) + if err != nil { + t.Fatalf("failed to open pipe for writing %q: %v", fileName, err) + } + defer fd.DecRef() + + const msg = "forza blu" + checkEmpty(ctx, t, fd) + checkWrite(ctx, t, fd, msg) + checkRead(ctx, t, fd, msg) +} + +// setup creates a VFS with a pipe in the root directory at path fileName. The +// returned VirtualDentry must be DecRef()'d be the caller. It calls t.Fatal +// upon failure. +func setup(t *testing.T) (context.Context, *auth.Credentials, *vfs.VirtualFilesystem, vfs.VirtualDentry) { + ctx := contexttest.Context(t) + creds := auth.CredentialsFromContext(ctx) + + // Create VFS. + vfsObj := vfs.New() + vfsObj.MustRegisterFilesystemType("tmpfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + }) + mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{}) + if err != nil { + t.Fatalf("failed to create tmpfs root mount: %v", err) + } + + // Create the pipe. + root := mntns.Root() + pop := vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(fileName), + } + mknodOpts := vfs.MknodOptions{Mode: linux.ModeNamedPipe | 0644} + if err := vfsObj.MknodAt(ctx, creds, &pop, &mknodOpts); err != nil { + t.Fatalf("failed to create file %q: %v", fileName, err) + } + + // Sanity check: the file pipe exists and has the correct mode. + stat, err := vfsObj.StatAt(ctx, creds, &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(fileName), + FollowFinalSymlink: true, + }, &vfs.StatOptions{}) + if err != nil { + t.Fatalf("stat(%q) failed: %v", fileName, err) + } + if stat.Mode&^linux.S_IFMT != 0644 { + t.Errorf("got wrong permissions (%0o)", stat.Mode) + } + if stat.Mode&linux.S_IFMT != linux.ModeNamedPipe { + t.Errorf("got wrong file type (%0o)", stat.Mode) + } + + return ctx, creds, vfsObj, root +} + +// checkEmpty calls t.Fatal if the pipe in fd is not empty. +func checkEmpty(ctx context.Context, t *testing.T, fd *vfs.FileDescription) { + readData := make([]byte, 1) + dst := usermem.BytesIOSequence(readData) + bytesRead, err := fd.Read(ctx, dst, vfs.ReadOptions{}) + if err != syserror.ErrWouldBlock { + t.Fatalf("expected ErrWouldBlock reading from empty pipe %q, but got: %v", fileName, err) + } + if bytesRead != 0 { + t.Fatalf("expected to read 0 bytes, but got %d", bytesRead) + } +} + +// checkWrite calls t.Fatal if it fails to write all of msg to fd. +func checkWrite(ctx context.Context, t *testing.T, fd *vfs.FileDescription, msg string) { + writeData := []byte(msg) + src := usermem.BytesIOSequence(writeData) + bytesWritten, err := fd.Write(ctx, src, vfs.WriteOptions{}) + if err != nil { + t.Fatalf("error writing to pipe %q: %v", fileName, err) + } + if bytesWritten != int64(len(writeData)) { + t.Fatalf("expected to write %d bytes, but wrote %d", len(writeData), bytesWritten) + } +} + +// checkRead calls t.Fatal if it fails to read msg from fd. +func checkRead(ctx context.Context, t *testing.T, fd *vfs.FileDescription, msg string) { + readData := make([]byte, len(msg)) + dst := usermem.BytesIOSequence(readData) + bytesRead, err := fd.Read(ctx, dst, vfs.ReadOptions{}) + if err != nil { + t.Fatalf("error reading from pipe %q: %v", fileName, err) + } + if bytesRead != int64(len(msg)) { + t.Fatalf("expected to read %d bytes, but got %d", len(msg), bytesRead) + } + if !bytes.Equal(readData, []byte(msg)) { + t.Fatalf("expected to read %q from pipe, but got %q", msg, string(readData)) + } +} diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go new file mode 100644 index 000000000..f51e247a7 --- /dev/null +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -0,0 +1,357 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tmpfs + +import ( + "io" + "math" + "sync" + "sync/atomic" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/memmap" + "gvisor.dev/gvisor/pkg/sentry/pgalloc" + "gvisor.dev/gvisor/pkg/sentry/safemem" + "gvisor.dev/gvisor/pkg/sentry/usage" + "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +type regularFile struct { + inode inode + + // memFile is a platform.File used to allocate pages to this regularFile. + memFile *pgalloc.MemoryFile + + // mu protects the fields below. + mu sync.RWMutex + + // data maps offsets into the file to offsets into memFile that store + // the file's data. + data fsutil.FileRangeSet + + // size is the size of data, but accessed using atomic memory + // operations to avoid locking in inode.stat(). + size uint64 + + // seals represents file seals on this inode. + seals uint32 +} + +func (fs *filesystem) newRegularFile(creds *auth.Credentials, mode linux.FileMode) *inode { + file := ®ularFile{ + memFile: fs.memFile, + } + file.inode.init(file, fs, creds, mode) + file.inode.nlink = 1 // from parent directory + return &file.inode +} + +type regularFileFD struct { + fileDescription + + // These are immutable. + readable bool + writable bool + + // off is the file offset. off is accessed using atomic memory operations. + // offMu serializes operations that may mutate off. + off int64 + offMu sync.Mutex +} + +// Release implements vfs.FileDescriptionImpl.Release. +func (fd *regularFileFD) Release() { + if fd.writable { + fd.vfsfd.VirtualDentry().Mount().EndWrite() + } +} + +// PRead implements vfs.FileDescriptionImpl.PRead. +func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + if !fd.readable { + return 0, syserror.EINVAL + } + if offset < 0 { + return 0, syserror.EINVAL + } + if dst.NumBytes() == 0 { + return 0, nil + } + f := fd.inode().impl.(*regularFile) + rw := getRegularFileReadWriter(f, offset) + n, err := dst.CopyOutFrom(ctx, rw) + putRegularFileReadWriter(rw) + return int64(n), err +} + +// Read implements vfs.FileDescriptionImpl.Read. +func (fd *regularFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + fd.offMu.Lock() + n, err := fd.PRead(ctx, dst, fd.off, opts) + fd.off += n + fd.offMu.Unlock() + return n, err +} + +// PWrite implements vfs.FileDescriptionImpl.PWrite. +func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + if !fd.writable { + return 0, syserror.EINVAL + } + if offset < 0 { + return 0, syserror.EINVAL + } + srclen := src.NumBytes() + if srclen == 0 { + return 0, nil + } + f := fd.inode().impl.(*regularFile) + end := offset + srclen + if end < offset { + // Overflow. + return 0, syserror.EFBIG + } + rw := getRegularFileReadWriter(f, offset) + n, err := src.CopyInTo(ctx, rw) + putRegularFileReadWriter(rw) + return n, err +} + +// Write implements vfs.FileDescriptionImpl.Write. +func (fd *regularFileFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + fd.offMu.Lock() + n, err := fd.PWrite(ctx, src, fd.off, opts) + fd.off += n + fd.offMu.Unlock() + return n, err +} + +// Seek implements vfs.FileDescriptionImpl.Seek. +func (fd *regularFileFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + fd.offMu.Lock() + defer fd.offMu.Unlock() + switch whence { + case linux.SEEK_SET: + // use offset as specified + case linux.SEEK_CUR: + offset += fd.off + case linux.SEEK_END: + offset += int64(atomic.LoadUint64(&fd.inode().impl.(*regularFile).size)) + default: + return 0, syserror.EINVAL + } + if offset < 0 { + return 0, syserror.EINVAL + } + fd.off = offset + return offset, nil +} + +// Sync implements vfs.FileDescriptionImpl.Sync. +func (fd *regularFileFD) Sync(ctx context.Context) error { + return nil +} + +// regularFileReadWriter implements safemem.Reader and Safemem.Writer. +type regularFileReadWriter struct { + file *regularFile + + // Offset into the file to read/write at. Note that this may be + // different from the FD offset if PRead/PWrite is used. + off uint64 +} + +var regularFileReadWriterPool = sync.Pool{ + New: func() interface{} { + return ®ularFileReadWriter{} + }, +} + +func getRegularFileReadWriter(file *regularFile, offset int64) *regularFileReadWriter { + rw := regularFileReadWriterPool.Get().(*regularFileReadWriter) + rw.file = file + rw.off = uint64(offset) + return rw +} + +func putRegularFileReadWriter(rw *regularFileReadWriter) { + rw.file = nil + regularFileReadWriterPool.Put(rw) +} + +// ReadToBlocks implements safemem.Reader.ReadToBlocks. +func (rw *regularFileReadWriter) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) { + rw.file.mu.RLock() + + // Compute the range to read (limited by file size and overflow-checked). + if rw.off >= rw.file.size { + rw.file.mu.RUnlock() + return 0, io.EOF + } + end := rw.file.size + if rend := rw.off + dsts.NumBytes(); rend > rw.off && rend < end { + end = rend + } + + var done uint64 + seg, gap := rw.file.data.Find(uint64(rw.off)) + for rw.off < end { + mr := memmap.MappableRange{uint64(rw.off), uint64(end)} + switch { + case seg.Ok(): + // Get internal mappings. + ims, err := rw.file.memFile.MapInternal(seg.FileRangeOf(seg.Range().Intersect(mr)), usermem.Read) + if err != nil { + rw.file.mu.RUnlock() + return done, err + } + + // Copy from internal mappings. + n, err := safemem.CopySeq(dsts, ims) + done += n + rw.off += uint64(n) + dsts = dsts.DropFirst64(n) + if err != nil { + rw.file.mu.RUnlock() + return done, err + } + + // Continue. + seg, gap = seg.NextNonEmpty() + + case gap.Ok(): + // Tmpfs holes are zero-filled. + gapmr := gap.Range().Intersect(mr) + dst := dsts.TakeFirst64(gapmr.Length()) + n, err := safemem.ZeroSeq(dst) + done += n + rw.off += uint64(n) + dsts = dsts.DropFirst64(n) + if err != nil { + rw.file.mu.RUnlock() + return done, err + } + + // Continue. + seg, gap = gap.NextSegment(), fsutil.FileRangeGapIterator{} + } + } + rw.file.mu.RUnlock() + return done, nil +} + +// WriteFromBlocks implements safemem.Writer.WriteFromBlocks. +func (rw *regularFileReadWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) { + rw.file.mu.Lock() + + // Compute the range to write (overflow-checked). + end := rw.off + srcs.NumBytes() + if end <= rw.off { + end = math.MaxInt64 + } + + // Check if seals prevent either file growth or all writes. + switch { + case rw.file.seals&linux.F_SEAL_WRITE != 0: // Write sealed + rw.file.mu.Unlock() + return 0, syserror.EPERM + case end > rw.file.size && rw.file.seals&linux.F_SEAL_GROW != 0: // Grow sealed + // When growth is sealed, Linux effectively allows writes which would + // normally grow the file to partially succeed up to the current EOF, + // rounded down to the page boundary before the EOF. + // + // This happens because writes (and thus the growth check) for tmpfs + // files proceed page-by-page on Linux, and the final write to the page + // containing EOF fails, resulting in a partial write up to the start of + // that page. + // + // To emulate this behaviour, artifically truncate the write to the + // start of the page containing the current EOF. + // + // See Linux, mm/filemap.c:generic_perform_write() and + // mm/shmem.c:shmem_write_begin(). + if pgstart := uint64(usermem.Addr(rw.file.size).RoundDown()); end > pgstart { + end = pgstart + } + if end <= rw.off { + // Truncation would result in no data being written. + rw.file.mu.Unlock() + return 0, syserror.EPERM + } + } + + // Page-aligned mr for when we need to allocate memory. RoundUp can't + // overflow since end is an int64. + pgstartaddr := usermem.Addr(rw.off).RoundDown() + pgendaddr, _ := usermem.Addr(end).RoundUp() + pgMR := memmap.MappableRange{uint64(pgstartaddr), uint64(pgendaddr)} + + var ( + done uint64 + retErr error + ) + seg, gap := rw.file.data.Find(uint64(rw.off)) + for rw.off < end { + mr := memmap.MappableRange{uint64(rw.off), uint64(end)} + switch { + case seg.Ok(): + // Get internal mappings. + ims, err := rw.file.memFile.MapInternal(seg.FileRangeOf(seg.Range().Intersect(mr)), usermem.Write) + if err != nil { + retErr = err + goto exitLoop + } + + // Copy to internal mappings. + n, err := safemem.CopySeq(ims, srcs) + done += n + rw.off += uint64(n) + srcs = srcs.DropFirst64(n) + if err != nil { + retErr = err + goto exitLoop + } + + // Continue. + seg, gap = seg.NextNonEmpty() + + case gap.Ok(): + // Allocate memory for the write. + gapMR := gap.Range().Intersect(pgMR) + fr, err := rw.file.memFile.Allocate(gapMR.Length(), usage.Tmpfs) + if err != nil { + retErr = err + goto exitLoop + } + + // Write to that memory as usual. + seg, gap = rw.file.data.Insert(gap, gapMR, fr.Start), fsutil.FileRangeGapIterator{} + } + } +exitLoop: + // If the write ends beyond the file's previous size, it causes the + // file to grow. + if rw.off > rw.file.size { + atomic.StoreUint64(&rw.file.size, rw.off) + } + + rw.file.mu.Unlock() + return done, retErr +} diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go new file mode 100644 index 000000000..3731c5b6f --- /dev/null +++ b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go @@ -0,0 +1,224 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tmpfs + +import ( + "bytes" + "fmt" + "io" + "testing" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" + "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sentry/vfs" +) + +// newFileFD creates a new file in a new tmpfs mount, and returns the FD. If +// the returned err is not nil, then cleanup should be called when the FD is no +// longer needed. +func newFileFD(ctx context.Context, filename string) (*vfs.FileDescription, func(), error) { + creds := auth.CredentialsFromContext(ctx) + + vfsObj := vfs.New() + vfsObj.MustRegisterFilesystemType("tmpfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + }) + mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{}) + if err != nil { + return nil, nil, fmt.Errorf("failed to create tmpfs root mount: %v", err) + } + root := mntns.Root() + + // Create the file that will be write/read. + fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(filename), + FollowFinalSymlink: true, + }, &vfs.OpenOptions{ + Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL, + Mode: 0644, + }) + if err != nil { + root.DecRef() + mntns.DecRef(vfsObj) + return nil, nil, fmt.Errorf("failed to create file %q: %v", filename, err) + } + + return fd, func() { + root.DecRef() + mntns.DecRef(vfsObj) + }, nil +} + +// Test that we can write some data to a file and read it back.` +func TestSimpleWriteRead(t *testing.T) { + ctx := contexttest.Context(t) + fd, cleanup, err := newFileFD(ctx, "simpleReadWrite") + if err != nil { + t.Fatal(err) + } + defer cleanup() + + // Write. + data := []byte("foobarbaz") + n, err := fd.Write(ctx, usermem.BytesIOSequence(data), vfs.WriteOptions{}) + if err != nil { + t.Fatalf("fd.Write failed: %v", err) + } + if n != int64(len(data)) { + t.Errorf("fd.Write got short write length %d, want %d", n, len(data)) + } + if got, want := fd.Impl().(*regularFileFD).off, int64(len(data)); got != want { + t.Errorf("fd.Write left offset at %d, want %d", got, want) + } + + // Seek back to beginning. + if _, err := fd.Seek(ctx, 0, linux.SEEK_SET); err != nil { + t.Fatalf("fd.Seek failed: %v", err) + } + if got, want := fd.Impl().(*regularFileFD).off, int64(0); got != want { + t.Errorf("fd.Seek(0) left offset at %d, want %d", got, want) + } + + // Read. + buf := make([]byte, len(data)) + n, err = fd.Read(ctx, usermem.BytesIOSequence(buf), vfs.ReadOptions{}) + if err != nil && err != io.EOF { + t.Fatalf("fd.Read failed: %v", err) + } + if n != int64(len(data)) { + t.Errorf("fd.Read got short read length %d, want %d", n, len(data)) + } + if got, want := string(buf), string(data); got != want { + t.Errorf("Read got %q want %s", got, want) + } + if got, want := fd.Impl().(*regularFileFD).off, int64(len(data)); got != want { + t.Errorf("fd.Write left offset at %d, want %d", got, want) + } +} + +func TestPWrite(t *testing.T) { + ctx := contexttest.Context(t) + fd, cleanup, err := newFileFD(ctx, "PRead") + if err != nil { + t.Fatal(err) + } + defer cleanup() + + // Fill file with 1k 'a's. + data := bytes.Repeat([]byte{'a'}, 1000) + n, err := fd.Write(ctx, usermem.BytesIOSequence(data), vfs.WriteOptions{}) + if err != nil { + t.Fatalf("fd.Write failed: %v", err) + } + if n != int64(len(data)) { + t.Errorf("fd.Write got short write length %d, want %d", n, len(data)) + } + + // Write "gVisor is awesome" at various offsets. + buf := []byte("gVisor is awesome") + offsets := []int{0, 1, 2, 10, 20, 50, 100, len(data) - 100, len(data) - 1, len(data), len(data) + 1} + for _, offset := range offsets { + name := fmt.Sprintf("PWrite offset=%d", offset) + t.Run(name, func(t *testing.T) { + n, err := fd.PWrite(ctx, usermem.BytesIOSequence(buf), int64(offset), vfs.WriteOptions{}) + if err != nil { + t.Errorf("fd.PWrite got err %v want nil", err) + } + if n != int64(len(buf)) { + t.Errorf("fd.PWrite got %d bytes want %d", n, len(buf)) + } + + // Update data to reflect expected file contents. + if len(data) < offset+len(buf) { + data = append(data, make([]byte, (offset+len(buf))-len(data))...) + } + copy(data[offset:], buf) + + // Read the whole file and compare with data. + readBuf := make([]byte, len(data)) + n, err = fd.PRead(ctx, usermem.BytesIOSequence(readBuf), 0, vfs.ReadOptions{}) + if err != nil { + t.Fatalf("fd.PRead failed: %v", err) + } + if n != int64(len(data)) { + t.Errorf("fd.PRead got short read length %d, want %d", n, len(data)) + } + if got, want := string(readBuf), string(data); got != want { + t.Errorf("PRead got %q want %s", got, want) + } + + }) + } +} + +func TestPRead(t *testing.T) { + ctx := contexttest.Context(t) + fd, cleanup, err := newFileFD(ctx, "PRead") + if err != nil { + t.Fatal(err) + } + defer cleanup() + + // Write 100 sequences of 'gVisor is awesome'. + data := bytes.Repeat([]byte("gVisor is awsome"), 100) + n, err := fd.Write(ctx, usermem.BytesIOSequence(data), vfs.WriteOptions{}) + if err != nil { + t.Fatalf("fd.Write failed: %v", err) + } + if n != int64(len(data)) { + t.Errorf("fd.Write got short write length %d, want %d", n, len(data)) + } + + // Read various sizes from various offsets. + sizes := []int{0, 1, 2, 10, 20, 50, 100, 1000} + offsets := []int{0, 1, 2, 10, 20, 50, 100, 1000, len(data) - 100, len(data) - 1, len(data), len(data) + 1} + + for _, size := range sizes { + for _, offset := range offsets { + name := fmt.Sprintf("PRead offset=%d size=%d", offset, size) + t.Run(name, func(t *testing.T) { + var ( + wantRead []byte + wantErr error + ) + if offset < len(data) { + wantRead = data[offset:] + } else if size > 0 { + wantErr = io.EOF + } + if offset+size < len(data) { + wantRead = wantRead[:size] + } + buf := make([]byte, size) + n, err := fd.PRead(ctx, usermem.BytesIOSequence(buf), int64(offset), vfs.ReadOptions{}) + if err != wantErr { + t.Errorf("fd.PRead got err %v want %v", err, wantErr) + } + if n != int64(len(wantRead)) { + t.Errorf("fd.PRead got %d bytes want %d", n, len(wantRead)) + } + if got := string(buf[:n]); got != string(wantRead) { + t.Errorf("fd.PRead got %q want %q", got, string(wantRead)) + } + }) + } + } +} diff --git a/pkg/sentry/fsimpl/tmpfs/symlink.go b/pkg/sentry/fsimpl/tmpfs/symlink.go new file mode 100644 index 000000000..5246aca84 --- /dev/null +++ b/pkg/sentry/fsimpl/tmpfs/symlink.go @@ -0,0 +1,36 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tmpfs + +import ( + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" +) + +type symlink struct { + inode inode + target string // immutable +} + +func (fs *filesystem) newSymlink(creds *auth.Credentials, target string) *inode { + link := &symlink{ + target: target, + } + link.inode.init(link, fs, creds, 0777) + link.inode.nlink = 1 // from parent directory + return &link.inode +} + +// O_PATH is unimplemented, so there's no way to get a FileDescription +// representing a symlink yet. diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go new file mode 100644 index 000000000..7be6faa5b --- /dev/null +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -0,0 +1,299 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package tmpfs provides a filesystem implementation that behaves like tmpfs: +// the Dentry tree is the sole source of truth for the state of the filesystem. +// +// Lock order: +// +// filesystem.mu +// regularFileFD.offMu +// regularFile.mu +// inode.mu +package tmpfs + +import ( + "fmt" + "math" + "sync" + "sync/atomic" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/pgalloc" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// FilesystemType implements vfs.FilesystemType. +type FilesystemType struct{} + +// filesystem implements vfs.FilesystemImpl. +type filesystem struct { + vfsfs vfs.Filesystem + + // memFile is used to allocate pages to for regular files. + memFile *pgalloc.MemoryFile + + // mu serializes changes to the Dentry tree. + mu sync.RWMutex + + nextInoMinusOne uint64 // accessed using atomic memory operations +} + +// GetFilesystem implements vfs.FilesystemType.GetFilesystem. +func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { + memFileProvider := pgalloc.MemoryFileProviderFromContext(ctx) + if memFileProvider == nil { + panic("MemoryFileProviderFromContext returned nil") + } + fs := filesystem{ + memFile: memFileProvider.MemoryFile(), + } + fs.vfsfs.Init(vfsObj, &fs) + root := fs.newDentry(fs.newDirectory(creds, 01777)) + return &fs.vfsfs, &root.vfsd, nil +} + +// Release implements vfs.FilesystemImpl.Release. +func (fs *filesystem) Release() { +} + +// dentry implements vfs.DentryImpl. +type dentry struct { + vfsd vfs.Dentry + + // inode is the inode represented by this dentry. Multiple Dentries may + // share a single non-directory inode (with hard links). inode is + // immutable. + inode *inode + + // tmpfs doesn't count references on dentries; because the dentry tree is + // the sole source of truth, it is by definition always consistent with the + // state of the filesystem. However, it does count references on inodes, + // because inode resources are released when all references are dropped. + // (tmpfs doesn't really have resources to release, but we implement + // reference counting because tmpfs regular files will.) + + // dentryEntry (ugh) links dentries into their parent directory.childList. + dentryEntry +} + +func (fs *filesystem) newDentry(inode *inode) *dentry { + d := &dentry{ + inode: inode, + } + d.vfsd.Init(d) + return d +} + +// IncRef implements vfs.DentryImpl.IncRef. +func (d *dentry) IncRef() { + d.inode.incRef() +} + +// TryIncRef implements vfs.DentryImpl.TryIncRef. +func (d *dentry) TryIncRef() bool { + return d.inode.tryIncRef() +} + +// DecRef implements vfs.DentryImpl.DecRef. +func (d *dentry) DecRef() { + d.inode.decRef() +} + +// inode represents a filesystem object. +type inode struct { + // refs is a reference count. refs is accessed using atomic memory + // operations. + // + // A reference is held on all inodes that are reachable in the filesystem + // tree. For non-directories (which may have multiple hard links), this + // means that a reference is dropped when nlink reaches 0. For directories, + // nlink never reaches 0 due to the "." entry; instead, + // filesystem.RmdirAt() drops the reference. + refs int64 + + // Inode metadata; protected by mu and accessed using atomic memory + // operations unless otherwise specified. + mu sync.RWMutex + mode uint32 // excluding file type bits, which are based on impl + nlink uint32 // protected by filesystem.mu instead of inode.mu + uid uint32 // auth.KUID, but stored as raw uint32 for sync/atomic + gid uint32 // auth.KGID, but ... + ino uint64 // immutable + + impl interface{} // immutable +} + +const maxLinks = math.MaxUint32 + +func (i *inode) init(impl interface{}, fs *filesystem, creds *auth.Credentials, mode linux.FileMode) { + i.refs = 1 + i.mode = uint32(mode) + i.uid = uint32(creds.EffectiveKUID) + i.gid = uint32(creds.EffectiveKGID) + i.ino = atomic.AddUint64(&fs.nextInoMinusOne, 1) + // i.nlink initialized by caller + i.impl = impl +} + +// incLinksLocked increments i's link count. +// +// Preconditions: filesystem.mu must be locked for writing. i.nlink != 0. +// i.nlink < maxLinks. +func (i *inode) incLinksLocked() { + if i.nlink == 0 { + panic("tmpfs.inode.incLinksLocked() called with no existing links") + } + if i.nlink == maxLinks { + panic("memfs.inode.incLinksLocked() called with maximum link count") + } + atomic.AddUint32(&i.nlink, 1) +} + +// decLinksLocked decrements i's link count. +// +// Preconditions: filesystem.mu must be locked for writing. i.nlink != 0. +func (i *inode) decLinksLocked() { + if i.nlink == 0 { + panic("tmpfs.inode.decLinksLocked() called with no existing links") + } + atomic.AddUint32(&i.nlink, ^uint32(0)) +} + +func (i *inode) incRef() { + if atomic.AddInt64(&i.refs, 1) <= 1 { + panic("tmpfs.inode.incRef() called without holding a reference") + } +} + +func (i *inode) tryIncRef() bool { + for { + refs := atomic.LoadInt64(&i.refs) + if refs == 0 { + return false + } + if atomic.CompareAndSwapInt64(&i.refs, refs, refs+1) { + return true + } + } +} + +func (i *inode) decRef() { + if refs := atomic.AddInt64(&i.refs, -1); refs == 0 { + // This is unnecessary; it's mostly to simulate what tmpfs would do. + if regFile, ok := i.impl.(*regularFile); ok { + regFile.mu.Lock() + regFile.data.DropAll(regFile.memFile) + atomic.StoreUint64(®File.size, 0) + regFile.mu.Unlock() + } + } else if refs < 0 { + panic("tmpfs.inode.decRef() called without holding a reference") + } +} + +func (i *inode) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes, isDir bool) error { + return vfs.GenericCheckPermissions(creds, ats, isDir, uint16(atomic.LoadUint32(&i.mode)), auth.KUID(atomic.LoadUint32(&i.uid)), auth.KGID(atomic.LoadUint32(&i.gid))) +} + +// Go won't inline this function, and returning linux.Statx (which is quite +// big) means spending a lot of time in runtime.duffcopy(), so instead it's an +// output parameter. +func (i *inode) statTo(stat *linux.Statx) { + stat.Mask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_NLINK | linux.STATX_UID | linux.STATX_GID | linux.STATX_INO + stat.Blksize = 1 // usermem.PageSize in tmpfs + stat.Nlink = atomic.LoadUint32(&i.nlink) + stat.UID = atomic.LoadUint32(&i.uid) + stat.GID = atomic.LoadUint32(&i.gid) + stat.Mode = uint16(atomic.LoadUint32(&i.mode)) + stat.Ino = i.ino + // TODO: device number + switch impl := i.impl.(type) { + case *regularFile: + stat.Mode |= linux.S_IFREG + stat.Mask |= linux.STATX_SIZE | linux.STATX_BLOCKS + stat.Size = uint64(atomic.LoadUint64(&impl.size)) + // In tmpfs, this will be FileRangeSet.Span() / 512 (but also cached in + // a uint64 accessed using atomic memory operations to avoid taking + // locks). + stat.Blocks = allocatedBlocksForSize(stat.Size) + case *directory: + stat.Mode |= linux.S_IFDIR + case *symlink: + stat.Mode |= linux.S_IFLNK + stat.Mask |= linux.STATX_SIZE | linux.STATX_BLOCKS + stat.Size = uint64(len(impl.target)) + stat.Blocks = allocatedBlocksForSize(stat.Size) + case *namedPipe: + stat.Mode |= linux.S_IFIFO + default: + panic(fmt.Sprintf("unknown inode type: %T", i.impl)) + } +} + +// allocatedBlocksForSize returns the number of 512B blocks needed to +// accommodate the given size in bytes, as appropriate for struct +// stat::st_blocks and struct statx::stx_blocks. (Note that this 512B block +// size is independent of the "preferred block size for I/O", struct +// stat::st_blksize and struct statx::stx_blksize.) +func allocatedBlocksForSize(size uint64) uint64 { + return (size + 511) / 512 +} + +func (i *inode) direntType() uint8 { + switch i.impl.(type) { + case *regularFile: + return linux.DT_REG + case *directory: + return linux.DT_DIR + case *symlink: + return linux.DT_LNK + default: + panic(fmt.Sprintf("unknown inode type: %T", i.impl)) + } +} + +// fileDescription is embedded by tmpfs implementations of +// vfs.FileDescriptionImpl. +type fileDescription struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl +} + +func (fd *fileDescription) filesystem() *filesystem { + return fd.vfsfd.Mount().Filesystem().Impl().(*filesystem) +} + +func (fd *fileDescription) inode() *inode { + return fd.vfsfd.Dentry().Impl().(*dentry).inode +} + +// Stat implements vfs.FileDescriptionImpl.Stat. +func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { + var stat linux.Statx + fd.inode().statTo(&stat) + return stat, nil +} + +// SetStat implements vfs.FileDescriptionImpl.SetStat. +func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { + if opts.Stat.Mask == 0 { + return nil + } + // TODO: implement inode.setStat + return syserror.EPERM +} -- cgit v1.2.3 From 17c18241cdeb66e75738c3892730f1a434a4bd60 Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Mon, 6 Jan 2020 15:53:21 -0800 Subject: platform/syscall: use syscall + int3 to execute a system call in a stub process Right now, we need to call ptrace(PTRACE_SYSCALL) and wait() twice to execute one system call in a stub process. With these changes, we will need to call ptrace + wait only once. In addition, this allows to workaround the kernel bug when a stub process doesn't stop on syscall-exit-stop and starts executing the next system call. Reported-by: syzbot+37143cafa8dc3b5008ee@syzkaller.appspotmail.com PiperOrigin-RevId: 288393029 --- pkg/sentry/platform/ptrace/stub_amd64.s | 29 ++++++++++++++----------- pkg/sentry/platform/ptrace/stub_arm64.s | 30 +++++++++++++++----------- pkg/sentry/platform/ptrace/subprocess.go | 20 +++++------------ pkg/sentry/platform/ptrace/subprocess_amd64.go | 4 +++- pkg/sentry/platform/ptrace/subprocess_arm64.go | 2 ++ 5 files changed, 45 insertions(+), 40 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/ptrace/stub_amd64.s b/pkg/sentry/platform/ptrace/stub_amd64.s index 64c718d21..16f9c523e 100644 --- a/pkg/sentry/platform/ptrace/stub_amd64.s +++ b/pkg/sentry/platform/ptrace/stub_amd64.s @@ -64,6 +64,8 @@ begin: CMPQ AX, $0 JL error + MOVQ $0, BX + // SIGSTOP to wait for attach. // // The SYSCALL instruction will be used for future syscall injection by @@ -73,23 +75,26 @@ begin: MOVQ $SIGSTOP, SI SYSCALL - // The tracer may "detach" and/or allow code execution here in three cases: - // - // 1. New (traced) stub threads are explicitly detached by the - // goroutine in newSubprocess. However, they are detached while in - // group-stop, so they do not execute code here. - // - // 2. If a tracer thread exits, it implicitly detaches from the stub, - // potentially allowing code execution here. However, the Go runtime - // never exits individual threads, so this case never occurs. - // - // 3. subprocess.createStub clones a new stub process that is untraced, + // The sentry sets BX to 1 when creating stub process. + CMPQ BX, $1 + JE clone + + // Notify the Sentry that syscall exited. +done: + INT $3 + // Be paranoid. + JMP done +clone: + // subprocess.createStub clones a new stub process that is untraced, // thus executing this code. We setup the PDEATHSIG before SIGSTOPing // ourselves for attach by the tracer. // // R15 has been updated with the expected PPID. - JMP begin + CMPQ AX, $0 + JE begin + // The clone syscall returns a non-zero value. + JMP done error: // Exit with -errno. MOVQ AX, DI diff --git a/pkg/sentry/platform/ptrace/stub_arm64.s b/pkg/sentry/platform/ptrace/stub_arm64.s index 2c5e4d5cb..6162df02a 100644 --- a/pkg/sentry/platform/ptrace/stub_arm64.s +++ b/pkg/sentry/platform/ptrace/stub_arm64.s @@ -59,6 +59,8 @@ begin: CMP $0x0, R0 BLT error + MOVD $0, R9 + // SIGSTOP to wait for attach. // // The SYSCALL instruction will be used for future syscall injection by @@ -66,22 +68,26 @@ begin: MOVD $SYS_KILL, R8 MOVD $SIGSTOP, R1 SVC - // The tracer may "detach" and/or allow code execution here in three cases: - // - // 1. New (traced) stub threads are explicitly detached by the - // goroutine in newSubprocess. However, they are detached while in - // group-stop, so they do not execute code here. - // - // 2. If a tracer thread exits, it implicitly detaches from the stub, - // potentially allowing code execution here. However, the Go runtime - // never exits individual threads, so this case never occurs. - // - // 3. subprocess.createStub clones a new stub process that is untraced, + + // The sentry sets R9 to 1 when creating stub process. + CMP $1, R9 + BEQ clone + +done: + // Notify the Sentry that syscall exited. + BRK $3 + B done // Be paranoid. +clone: + // subprocess.createStub clones a new stub process that is untraced, // thus executing this code. We setup the PDEATHSIG before SIGSTOPing // ourselves for attach by the tracer. // // R7 has been updated with the expected PPID. - B begin + CMP $0, R0 + BEQ begin + + // The clone system call returned a non-zero value. + B done error: // Exit with -errno. diff --git a/pkg/sentry/platform/ptrace/subprocess.go b/pkg/sentry/platform/ptrace/subprocess.go index 821f6848d..20244fd95 100644 --- a/pkg/sentry/platform/ptrace/subprocess.go +++ b/pkg/sentry/platform/ptrace/subprocess.go @@ -430,13 +430,15 @@ func (t *thread) syscall(regs *syscall.PtraceRegs) (uintptr, error) { } for { - // Execute the syscall instruction. - if _, _, errno := syscall.RawSyscall6(syscall.SYS_PTRACE, syscall.PTRACE_SYSCALL, uintptr(t.tid), 0, 0, 0, 0); errno != 0 { + // Execute the syscall instruction. The task has to stop on the + // trap instruction which is right after the syscall + // instruction. + if _, _, errno := syscall.RawSyscall6(syscall.SYS_PTRACE, syscall.PTRACE_CONT, uintptr(t.tid), 0, 0, 0, 0); errno != 0 { panic(fmt.Sprintf("ptrace syscall-enter failed: %v", errno)) } sig := t.wait(stopped) - if sig == (syscallEvent | syscall.SIGTRAP) { + if sig == syscall.SIGTRAP { // Reached syscall-enter-stop. break } else { @@ -448,18 +450,6 @@ func (t *thread) syscall(regs *syscall.PtraceRegs) (uintptr, error) { } } - // Complete the actual system call. - if _, _, errno := syscall.RawSyscall6(syscall.SYS_PTRACE, syscall.PTRACE_SYSCALL, uintptr(t.tid), 0, 0, 0, 0); errno != 0 { - panic(fmt.Sprintf("ptrace syscall-enter failed: %v", errno)) - } - - // Wait for syscall-exit-stop. "[Signal-delivery-stop] never happens - // between syscall-enter-stop and syscall-exit-stop; it happens *after* - // syscall-exit-stop.)" - ptrace(2), "Syscall-stops" - if sig := t.wait(stopped); sig != (syscallEvent | syscall.SIGTRAP) { - t.dumpAndPanic(fmt.Sprintf("wait failed: expected SIGTRAP, got %v [%d]", sig, sig)) - } - // Grab registers. if err := t.getRegs(regs); err != nil { panic(fmt.Sprintf("ptrace get regs failed: %v", err)) diff --git a/pkg/sentry/platform/ptrace/subprocess_amd64.go b/pkg/sentry/platform/ptrace/subprocess_amd64.go index 606dc2b1d..e99798c56 100644 --- a/pkg/sentry/platform/ptrace/subprocess_amd64.go +++ b/pkg/sentry/platform/ptrace/subprocess_amd64.go @@ -141,9 +141,11 @@ func (t *thread) adjustInitRegsRip() { t.initRegs.Rip -= initRegsRipAdjustment } -// Pass the expected PPID to the child via R15 when creating stub process +// Pass the expected PPID to the child via R15 when creating stub process. func initChildProcessPPID(initregs *syscall.PtraceRegs, ppid int32) { initregs.R15 = uint64(ppid) + // Rbx has to be set to 1 when creating stub process. + initregs.Rbx = 1 } // patchSignalInfo patches the signal info to account for hitting the seccomp diff --git a/pkg/sentry/platform/ptrace/subprocess_arm64.go b/pkg/sentry/platform/ptrace/subprocess_arm64.go index 62a686ee7..7b975137f 100644 --- a/pkg/sentry/platform/ptrace/subprocess_arm64.go +++ b/pkg/sentry/platform/ptrace/subprocess_arm64.go @@ -127,6 +127,8 @@ func (t *thread) adjustInitRegsRip() { // Pass the expected PPID to the child via X7 when creating stub process func initChildProcessPPID(initregs *syscall.PtraceRegs, ppid int32) { initregs.Regs[7] = uint64(ppid) + // R9 has to be set to 1 when creating stub process. + initregs.Regs[9] = 1 } // patchSignalInfo patches the signal info to account for hitting the seccomp -- cgit v1.2.3 From a53ac7307abfeb7172e67f48d0a7aaa4b5c3f31e Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Tue, 7 Jan 2020 23:52:59 -0800 Subject: fs/splice: don't report a partialResult error if there is no data loss PiperOrigin-RevId: 288642552 --- pkg/sentry/fs/file.go | 7 +++++++ pkg/sentry/fs/splice.go | 5 +++++ test/syscalls/linux/inotify.cc | 28 ++++++++++++++++++++++++++++ 3 files changed, 40 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/file.go b/pkg/sentry/fs/file.go index c0a6e884b..a2f966cb6 100644 --- a/pkg/sentry/fs/file.go +++ b/pkg/sentry/fs/file.go @@ -555,6 +555,10 @@ type lockedWriter struct { // // This applies only to Write, not WriteAt. Offset int64 + + // Err contains the first error encountered while copying. This is + // useful to determine whether Writer or Reader failed during io.Copy. + Err error } // Write implements io.Writer.Write. @@ -590,5 +594,8 @@ func (w *lockedWriter) WriteAt(buf []byte, offset int64) (int, error) { break } } + if w.Err == nil { + w.Err = err + } return written, err } diff --git a/pkg/sentry/fs/splice.go b/pkg/sentry/fs/splice.go index 311798811..389c330a0 100644 --- a/pkg/sentry/fs/splice.go +++ b/pkg/sentry/fs/splice.go @@ -167,6 +167,11 @@ func Splice(ctx context.Context, dst *File, src *File, opts SpliceOpts) (int64, if !srcPipe && !opts.SrcOffset { atomic.StoreInt64(&src.offset, src.offset+n) } + + // Don't report any errors if we have some progress without data loss. + if w.Err == nil { + err = nil + } } // Drop locks. diff --git a/test/syscalls/linux/inotify.cc b/test/syscalls/linux/inotify.cc index 7384c27dc..59ec9940a 100644 --- a/test/syscalls/linux/inotify.cc +++ b/test/syscalls/linux/inotify.cc @@ -1591,6 +1591,34 @@ TEST(Inotify, EpollNoDeadlock) { } } +TEST(Inotify, SpliceEvent) { + int pipes[2]; + ASSERT_THAT(pipe2(pipes, O_NONBLOCK), SyscallSucceeds()); + + const TempPath root = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + const TempPath file1 = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileWith( + root.path(), "some content", TempPath::kDefaultFileMode)); + + const FileDescriptor file1_fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(file1.path(), O_RDONLY)); + const int watcher = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(fd.get(), file1.path(), IN_ALL_EVENTS)); + + char buf; + EXPECT_THAT(read(file1_fd.get(), &buf, 1), SyscallSucceeds()); + + EXPECT_THAT(splice(fd.get(), nullptr, pipes[1], nullptr, + sizeof(struct inotify_event) + 1, SPLICE_F_NONBLOCK), + SyscallSucceedsWithValue(sizeof(struct inotify_event))); + + const FileDescriptor read_fd(pipes[0]); + const std::vector events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(read_fd.get())); + ASSERT_THAT(events, Are({Event(IN_ACCESS, watcher)})); +} + } // namespace } // namespace testing } // namespace gvisor -- cgit v1.2.3 From 8cc1c35bbdc5c9bd6b3965311497885ce72317a8 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Thu, 12 Dec 2019 15:48:24 -0800 Subject: Write simple ACCEPT rules to the filter table. This gets us closer to passing the iptables tests and opens up iptables so it can be worked on by multiple people. A few restrictions are enforced for security (i.e. we don't want to let users write a bunch of iptables rules and then just not enforce them): - Only the filter table is writable. - Only ACCEPT rules with no matching criteria can be added. --- go.mod | 31 +-- go.sum | 10 + pkg/abi/linux/netfilter.go | 82 +++--- pkg/sentry/socket/netfilter/BUILD | 1 + pkg/sentry/socket/netfilter/netfilter.go | 411 ++++++++++++++++++++++++------- pkg/sentry/socket/netstack/netstack.go | 23 +- pkg/tcpip/iptables/iptables.go | 114 ++++++--- pkg/tcpip/iptables/targets.go | 8 + pkg/tcpip/iptables/types.go | 55 ++--- test/iptables/filter_input.go | 1 + test/iptables/iptables_test.go | 23 +- 11 files changed, 550 insertions(+), 209 deletions(-) (limited to 'pkg/sentry') diff --git a/go.mod b/go.mod index 304b8bf13..4802359f8 100644 --- a/go.mod +++ b/go.mod @@ -3,19 +3,20 @@ module gvisor.dev/gvisor go 1.13 require ( - github.com/cenkalti/backoff v0.0.0-20190506075156-2146c9339422 - github.com/gofrs/flock v0.6.1-0.20180915234121-886344bea079 - github.com/golang/mock v1.3.1 - github.com/golang/protobuf v1.3.1 - github.com/google/btree v1.0.0 - github.com/google/go-cmp v0.2.0 - github.com/google/subcommands v0.0.0-20190508160503-636abe8753b8 - github.com/google/uuid v0.0.0-20171129191014-dec09d789f3d - github.com/kr/pty v1.1.1 - github.com/opencontainers/runtime-spec v0.1.2-0.20171211145439-b2d941ef6a78 - github.com/syndtr/gocapability v0.0.0-20180916011248-d98352740cb2 - github.com/vishvananda/netlink v1.0.1-0.20190318003149-adb577d4a45e - github.com/vishvananda/netns v0.0.0-20171111001504-be1fbeda1936 - golang.org/x/net v0.0.0-20190311183353-d8887717615a - golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a + github.com/cenkalti/backoff v0.0.0-20190506075156-2146c9339422 + github.com/gofrs/flock v0.6.1-0.20180915234121-886344bea079 + github.com/golang/mock v1.3.1 + github.com/golang/protobuf v1.3.1 + github.com/google/btree v1.0.0 + github.com/google/go-cmp v0.2.0 + github.com/google/subcommands v0.0.0-20190508160503-636abe8753b8 + github.com/google/uuid v0.0.0-20171129191014-dec09d789f3d + github.com/kr/pty v1.1.1 + github.com/opencontainers/runtime-spec v0.1.2-0.20171211145439-b2d941ef6a78 + github.com/syndtr/gocapability v0.0.0-20180916011248-d98352740cb2 + github.com/vishvananda/netlink v1.0.1-0.20190318003149-adb577d4a45e + github.com/vishvananda/netns v0.0.0-20171111001504-be1fbeda1936 + golang.org/x/net v0.0.0-20190311183353-d8887717615a + golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a + golang.org/x/time v0.0.0-20191024005414-555d28b269f0 ) diff --git a/go.sum b/go.sum index 7a0bc175a..cf092956e 100644 --- a/go.sum +++ b/go.sum @@ -1,19 +1,29 @@ +github.com/cenkalti/backoff v0.0.0-20190506075156-2146c9339422 h1:+FKjzBIdfBHYDvxCv+djmDJdes/AoDtg8gpcxowBlF8= github.com/cenkalti/backoff v0.0.0-20190506075156-2146c9339422/go.mod h1:b6Nc7NRH5C4aCISLry0tLnTjcuTEvoiqcWDdsU0sOGM= github.com/gofrs/flock v0.6.1-0.20180915234121-886344bea079/go.mod h1:F1TvTiK9OcQqauNUHlbJvyl9Qa1QvF/gOUDKA14jxHU= github.com/golang/mock v1.3.1/go.mod h1:sBzyDLLjw3U8JLTeZvSv8jJB+tU5PVekmnlKIyFUx0Y= +github.com/golang/protobuf v1.3.1 h1:YF8+flBXS5eO826T4nzqPrxfhQThhXl0YzfuUPu4SBg= github.com/golang/protobuf v1.3.1/go.mod h1:6lQm79b+lXiMfvg/cZm0SGofjICqVBUtrP5yJMmIC1U= +github.com/google/btree v1.0.0 h1:0udJVsspx3VBr5FwtLhQQtuAsVc79tTq0ocGIPAU6qo= github.com/google/btree v1.0.0/go.mod h1:lNA+9X1NB3Zf8V7Ke586lFgjr2dZNuvo3lPJSGZ5JPQ= +github.com/google/go-cmp v0.2.0 h1:+dTQ8DZQJz0Mb/HjFlkptS1FeQ4cWSnN941F8aEG4SQ= github.com/google/go-cmp v0.2.0/go.mod h1:oXzfMopK8JAjlY9xF4vHSVASa0yLyX7SntLO5aqRK0M= github.com/google/subcommands v0.0.0-20190508160503-636abe8753b8/go.mod h1:ZjhPrFU+Olkh9WazFPsl27BQ4UPiG37m3yTrtFlrHVk= github.com/google/uuid v0.0.0-20171129191014-dec09d789f3d/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo= +github.com/kr/pty v1.1.1 h1:VkoXIwSboBpnk99O/KFauAEILuNHv5DVFKZMBN/gUgw= github.com/kr/pty v1.1.1/go.mod h1:pFQYn66WHrOpPYNljwOMqo10TkYh1fy3cYio2l3bCsQ= +github.com/opencontainers/runtime-spec v0.1.2-0.20171211145439-b2d941ef6a78 h1:d9F+LNYwMyi3BDN4GzZdaSiq4otb8duVEWyZjeUtOQI= github.com/opencontainers/runtime-spec v0.1.2-0.20171211145439-b2d941ef6a78/go.mod h1:jwyrGlmzljRJv/Fgzds9SsS/C5hL+LL3ko9hs6T5lQ0= +github.com/syndtr/gocapability v0.0.0-20180916011248-d98352740cb2 h1:b6uOv7YOFK0TYG7HtkIgExQo+2RdLuwRft63jn2HWj8= github.com/syndtr/gocapability v0.0.0-20180916011248-d98352740cb2/go.mod h1:hkRG7XYTFWNJGYcbNJQlaLq0fg1yr4J4t/NcTQtrfww= github.com/vishvananda/netlink v1.0.1-0.20190318003149-adb577d4a45e/go.mod h1:+SR5DhBJrl6ZM7CoCKvpw5BKroDKQ+PJqOg65H/2ktk= github.com/vishvananda/netns v0.0.0-20171111001504-be1fbeda1936/go.mod h1:ZjcWmFBXmLKZu9Nxj3WKYEafiSqer2rnvPr0en9UNpI= golang.org/x/crypto v0.0.0-20190308221718-c2843e01d9a2/go.mod h1:djNgcEr1/C05ACkg1iLfiJU5Ep61QUkGW8qpdssI0+w= golang.org/x/net v0.0.0-20190311183353-d8887717615a/go.mod h1:t9HGtf8HONx5eT2rtn7q6eTqICYqUVnKs3thJo3Qplg= golang.org/x/sync v0.0.0-20190423024810-112230192c58/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM= +golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a h1:1BGLXjeY4akVXGgbC9HugT3Jv3hCI0z56oJR5vAMgBU= golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY= golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ= +golang.org/x/time v0.0.0-20191024005414-555d28b269f0 h1:/5xXl8Y5W96D+TtHSlonuFqGHIWVuyCkGJLwGh9JJFs= +golang.org/x/time v0.0.0-20191024005414-555d28b269f0/go.mod h1:tRJNPiyCQ0inRvYxbN9jk5I+vvW/OXSQhTDSoE431IQ= golang.org/x/tools v0.0.0-20190425150028-36563e24a262/go.mod h1:RgjU9mgBXZiqYHBnxXauZ1Gv1EHHAz9KjViQ78xBX0Q= diff --git a/pkg/abi/linux/netfilter.go b/pkg/abi/linux/netfilter.go index 269ba5567..0bcb232de 100644 --- a/pkg/abi/linux/netfilter.go +++ b/pkg/abi/linux/netfilter.go @@ -42,6 +42,13 @@ const ( NF_RETURN = -NF_REPEAT - 1 ) +var VerdictStrings = map[int32]string{ + -NF_DROP - 1: "DROP", + -NF_ACCEPT - 1: "ACCEPT", + -NF_QUEUE - 1: "QUEUE", + NF_RETURN: "RETURN", +} + // Socket options. These correspond to values in // include/uapi/linux/netfilter_ipv4/ip_tables.h. const ( @@ -179,7 +186,7 @@ const SizeOfXTCounters = 16 // the user data. type XTEntryMatch struct { MatchSize uint16 - Name [XT_EXTENSION_MAXNAMELEN]byte + Name ExtensionName Revision uint8 // Data is omitted here because it would cause XTEntryMatch to be an // extra byte larger (see http://www.catb.org/esr/structure-packing/). @@ -199,7 +206,7 @@ const SizeOfXTEntryMatch = 32 // the user data. type XTEntryTarget struct { TargetSize uint16 - Name [XT_EXTENSION_MAXNAMELEN]byte + Name ExtensionName Revision uint8 // Data is omitted here because it would cause XTEntryTarget to be an // extra byte larger (see http://www.catb.org/esr/structure-packing/). @@ -226,9 +233,9 @@ const SizeOfXTStandardTarget = 40 // ErrorName. It corresponds to struct xt_error_target in // include/uapi/linux/netfilter/x_tables.h. type XTErrorTarget struct { - Target XTEntryTarget - ErrorName [XT_FUNCTION_MAXNAMELEN]byte - _ [2]byte + Target XTEntryTarget + Name ErrorName + _ [2]byte } // SizeOfXTErrorTarget is the size of an XTErrorTarget. @@ -237,7 +244,7 @@ const SizeOfXTErrorTarget = 64 // IPTGetinfo is the argument for the IPT_SO_GET_INFO sockopt. It corresponds // to struct ipt_getinfo in include/uapi/linux/netfilter_ipv4/ip_tables.h. type IPTGetinfo struct { - Name [XT_TABLE_MAXNAMELEN]byte + Name TableName ValidHooks uint32 HookEntry [NF_INET_NUMHOOKS]uint32 Underflow [NF_INET_NUMHOOKS]uint32 @@ -248,16 +255,11 @@ type IPTGetinfo struct { // SizeOfIPTGetinfo is the size of an IPTGetinfo. const SizeOfIPTGetinfo = 84 -// TableName returns the table name. -func (info *IPTGetinfo) TableName() string { - return tableName(info.Name[:]) -} - // IPTGetEntries is the argument for the IPT_SO_GET_ENTRIES sockopt. It // corresponds to struct ipt_get_entries in // include/uapi/linux/netfilter_ipv4/ip_tables.h. type IPTGetEntries struct { - Name [XT_TABLE_MAXNAMELEN]byte + Name TableName Size uint32 _ [4]byte // Entrytable is omitted here because it would cause IPTGetEntries to @@ -266,34 +268,22 @@ type IPTGetEntries struct { // Entrytable [0]IPTEntry } -// TableName returns the entries' table name. -func (entries *IPTGetEntries) TableName() string { - return tableName(entries.Name[:]) -} - // SizeOfIPTGetEntries is the size of an IPTGetEntries. const SizeOfIPTGetEntries = 40 -// KernelIPTGetEntries is identical to IPTEntry, but includes the Elems field. -// This struct marshaled via the binary package to write an KernelIPTGetEntries -// to userspace. +// KernelIPTGetEntries is identical to IPTGetEntries, but includes the +// Entrytable field. This struct marshaled via the binary package to write an +// KernelIPTGetEntries to userspace. type KernelIPTGetEntries struct { - Name [XT_TABLE_MAXNAMELEN]byte - Size uint32 - _ [4]byte + IPTGetEntries Entrytable []KernelIPTEntry } -// TableName returns the entries' table name. -func (entries *KernelIPTGetEntries) TableName() string { - return tableName(entries.Name[:]) -} - // IPTReplace is the argument for the IPT_SO_SET_REPLACE sockopt. It // corresponds to struct ipt_replace in // include/uapi/linux/netfilter_ipv4/ip_tables.h. type IPTReplace struct { - Name [XT_TABLE_MAXNAMELEN]byte + Name TableName ValidHooks uint32 NumEntries uint32 Size uint32 @@ -306,14 +296,40 @@ type IPTReplace struct { // Entries [0]IPTEntry } +type KernelIPTReplace struct { + IPTReplace + Entries [0]IPTEntry +} + // SizeOfIPTReplace is the size of an IPTReplace. const SizeOfIPTReplace = 96 -func tableName(name []byte) string { - for i, c := range name { +type ExtensionName [XT_EXTENSION_MAXNAMELEN]byte + +// String implements fmt.Stringer. +func (en ExtensionName) String() string { + return name(en[:]) +} + +type TableName [XT_TABLE_MAXNAMELEN]byte + +// String implements fmt.Stringer. +func (tn TableName) String() string { + return name(tn[:]) +} + +type ErrorName [XT_FUNCTION_MAXNAMELEN]byte + +// String implements fmt.Stringer. +func (fn ErrorName) String() string { + return name(fn[:]) +} + +func name(cstring []byte) string { + for i, c := range cstring { if c == 0 { - return string(name[:i]) + return string(cstring[:i]) } } - return string(name) + return string(cstring) } diff --git a/pkg/sentry/socket/netfilter/BUILD b/pkg/sentry/socket/netfilter/BUILD index 5eb06bbf4..b70047d81 100644 --- a/pkg/sentry/socket/netfilter/BUILD +++ b/pkg/sentry/socket/netfilter/BUILD @@ -14,6 +14,7 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/binary", + "//pkg/log", "//pkg/sentry/kernel", "//pkg/sentry/usermem", "//pkg/syserr", diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 9f87c32f1..8c7f3c7fc 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserr" @@ -35,6 +36,7 @@ const errorTargetName = "ERROR" // metadata is opaque to netstack. It holds data that we need to translate // between Linux's and netstack's iptables representations. +// TODO(gvisor.dev/issue/170): This might be removable. type metadata struct { HookEntry [linux.NF_INET_NUMHOOKS]uint32 Underflow [linux.NF_INET_NUMHOOKS]uint32 @@ -51,7 +53,7 @@ func GetInfo(t *kernel.Task, ep tcpip.Endpoint, outPtr usermem.Addr) (linux.IPTG } // Find the appropriate table. - table, err := findTable(ep, info.TableName()) + table, err := findTable(ep, info.Name.String()) if err != nil { return linux.IPTGetinfo{}, err } @@ -82,18 +84,19 @@ func GetEntries(t *kernel.Task, ep tcpip.Endpoint, outPtr usermem.Addr, outLen i } // Find the appropriate table. - table, err := findTable(ep, userEntries.TableName()) + table, err := findTable(ep, userEntries.Name.String()) if err != nil { return linux.KernelIPTGetEntries{}, err } // Convert netstack's iptables rules to something that the iptables // tool can understand. - entries, _, err := convertNetstackToBinary(userEntries.TableName(), table) + entries, _, err := convertNetstackToBinary(userEntries.Name.String(), table) if err != nil { return linux.KernelIPTGetEntries{}, err } if binary.Size(entries) > uintptr(outLen) { + log.Infof("Insufficient GetEntries output size: %d", uintptr(outLen)) return linux.KernelIPTGetEntries{}, syserr.ErrInvalidArgument } @@ -142,103 +145,63 @@ func convertNetstackToBinary(name string, table iptables.Table) (linux.KernelIPT // The table name has to fit in the struct. if linux.XT_TABLE_MAXNAMELEN < len(name) { + log.Infof("Table name too long.") return linux.KernelIPTGetEntries{}, metadata{}, syserr.ErrInvalidArgument } copy(entries.Name[:], name) - // Deal with the built in chains first (INPUT, OUTPUT, etc.). Each of - // these chains ends with an unconditional policy entry. - for hook := iptables.Prerouting; hook < iptables.NumHooks; hook++ { - chain, ok := table.BuiltinChains[hook] - if !ok { - // This table doesn't support this hook. - continue - } - - // Sanity check. - if len(chain.Rules) < 1 { - return linux.KernelIPTGetEntries{}, metadata{}, syserr.ErrInvalidArgument - } - - for ruleIdx, rule := range chain.Rules { - // If this is the first rule of a builtin chain, set - // the metadata hook entry point. - if ruleIdx == 0 { + for ruleIdx, rule := range table.Rules { + // Is this a chain entry point? + for hook, hookRuleIdx := range table.BuiltinChains { + if hookRuleIdx == ruleIdx { meta.HookEntry[hook] = entries.Size } - - // Each rule corresponds to an entry. - entry := linux.KernelIPTEntry{ - IPTEntry: linux.IPTEntry{ - NextOffset: linux.SizeOfIPTEntry, - TargetOffset: linux.SizeOfIPTEntry, - }, + } + // Is this a chain underflow point? The underflow rule is the last rule + // in the chain, and is an unconditional rule (i.e. it matches any + // packet). This is enforced when saving iptables. + for underflow, underflowRuleIdx := range table.Underflows { + if underflowRuleIdx == ruleIdx { + meta.Underflow[underflow] = entries.Size } + } - for _, matcher := range rule.Matchers { - // Serialize the matcher and add it to the - // entry. - serialized := marshalMatcher(matcher) - entry.Elems = append(entry.Elems, serialized...) - entry.NextOffset += uint16(len(serialized)) - entry.TargetOffset += uint16(len(serialized)) - } + // Each rule corresponds to an entry. + entry := linux.KernelIPTEntry{ + IPTEntry: linux.IPTEntry{ + NextOffset: linux.SizeOfIPTEntry, + TargetOffset: linux.SizeOfIPTEntry, + }, + } - // Serialize and append the target. - serialized := marshalTarget(rule.Target) + for _, matcher := range rule.Matchers { + // Serialize the matcher and add it to the + // entry. + serialized := marshalMatcher(matcher) entry.Elems = append(entry.Elems, serialized...) entry.NextOffset += uint16(len(serialized)) - - // The underflow rule is the last rule in the chain, - // and is an unconditional rule (i.e. it matches any - // packet). This is enforced when saving iptables. - if ruleIdx == len(chain.Rules)-1 { - meta.Underflow[hook] = entries.Size - } - - entries.Size += uint32(entry.NextOffset) - entries.Entrytable = append(entries.Entrytable, entry) - meta.NumEntries++ + entry.TargetOffset += uint16(len(serialized)) } - } + // Serialize and append the target. + serialized := marshalTarget(rule.Target) + entry.Elems = append(entry.Elems, serialized...) + entry.NextOffset += uint16(len(serialized)) - // TODO(gvisor.dev/issue/170): Deal with the user chains here. Each of - // these starts with an error node holding the chain's name and ends - // with an unconditional return. - - // Lastly, each table ends with an unconditional error target rule as - // its final entry. - errorEntry := linux.KernelIPTEntry{ - IPTEntry: linux.IPTEntry{ - NextOffset: linux.SizeOfIPTEntry, - TargetOffset: linux.SizeOfIPTEntry, - }, + entries.Size += uint32(entry.NextOffset) + entries.Entrytable = append(entries.Entrytable, entry) + meta.NumEntries++ } - var errorTarget linux.XTErrorTarget - errorTarget.Target.TargetSize = linux.SizeOfXTErrorTarget - copy(errorTarget.ErrorName[:], errorTargetName) - copy(errorTarget.Target.Name[:], errorTargetName) - - // Serialize and add it to the list of entries. - errorTargetBuf := make([]byte, 0, linux.SizeOfXTErrorTarget) - serializedErrorTarget := binary.Marshal(errorTargetBuf, usermem.ByteOrder, errorTarget) - errorEntry.Elems = append(errorEntry.Elems, serializedErrorTarget...) - errorEntry.NextOffset += uint16(len(serializedErrorTarget)) - - entries.Size += uint32(errorEntry.NextOffset) - entries.Entrytable = append(entries.Entrytable, errorEntry) - meta.NumEntries++ - meta.Size = entries.Size + meta.Size = entries.Size return entries, meta, nil } func marshalMatcher(matcher iptables.Matcher) []byte { switch matcher.(type) { default: - // TODO(gvisor.dev/issue/170): We don't support any matchers yet, so - // any call to marshalMatcher will panic. + // TODO(gvisor.dev/issue/170): We don't support any matchers + // yet, so any call to marshalMatcher will panic. panic(fmt.Errorf("unknown matcher of type %T", matcher)) } } @@ -246,28 +209,46 @@ func marshalMatcher(matcher iptables.Matcher) []byte { func marshalTarget(target iptables.Target) []byte { switch target.(type) { case iptables.UnconditionalAcceptTarget: - return marshalUnconditionalAcceptTarget() + return marshalStandardTarget(iptables.Accept) + case iptables.UnconditionalDropTarget: + return marshalStandardTarget(iptables.Drop) + case iptables.PanicTarget: + return marshalPanicTarget() default: panic(fmt.Errorf("unknown target of type %T", target)) } } -func marshalUnconditionalAcceptTarget() []byte { +func marshalStandardTarget(verdict iptables.Verdict) []byte { // The target's name will be the empty string. target := linux.XTStandardTarget{ Target: linux.XTEntryTarget{ TargetSize: linux.SizeOfXTStandardTarget, }, - Verdict: translateStandardVerdict(iptables.Accept), + Verdict: translateFromStandardVerdict(verdict), } ret := make([]byte, 0, linux.SizeOfXTStandardTarget) return binary.Marshal(ret, usermem.ByteOrder, target) } -// translateStandardVerdict translates verdicts the same way as the iptables +func marshalPanicTarget() []byte { + // This is an error target named error + target := linux.XTErrorTarget{ + Target: linux.XTEntryTarget{ + TargetSize: linux.SizeOfXTErrorTarget, + }, + } + copy(target.Name[:], errorTargetName) + copy(target.Target.Name[:], errorTargetName) + + ret := make([]byte, 0, linux.SizeOfXTErrorTarget) + return binary.Marshal(ret, usermem.ByteOrder, target) +} + +// translateFromStandardVerdict translates verdicts the same way as the iptables // tool. -func translateStandardVerdict(verdict iptables.Verdict) int32 { +func translateFromStandardVerdict(verdict iptables.Verdict) int32 { switch verdict { case iptables.Accept: return -linux.NF_ACCEPT - 1 @@ -280,7 +261,269 @@ func translateStandardVerdict(verdict iptables.Verdict) int32 { case iptables.Jump: // TODO(gvisor.dev/issue/170): Support Jump. panic("Jump isn't supported yet") + } + panic(fmt.Sprintf("unknown standard verdict: %d", verdict)) +} + +// translateToStandardVerdict translates from the value in a +// linux.XTStandardTarget to an iptables.Verdict. +func translateToStandardVerdict(val int32) (iptables.Verdict, *syserr.Error) { + // TODO(gvisor.dev/issue/170): Support other verdicts. + switch val { + case -linux.NF_ACCEPT - 1: + return iptables.Accept, nil + case -linux.NF_DROP - 1: + return iptables.Drop, nil + case -linux.NF_QUEUE - 1: + log.Infof("Unsupported iptables verdict QUEUE.") + case linux.NF_RETURN: + log.Infof("Unsupported iptables verdict RETURN.") + } + log.Infof("Unknown iptables verdict %d.", val) + return iptables.Invalid, syserr.ErrInvalidArgument +} + +// SetEntries sets iptables rules for a single table. See +// net/ipv4/netfilter/ip_tables.c:translate_table for reference. +func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { + printReplace(optVal) + + // Get the basic rules data (struct ipt_replace). + if len(optVal) < linux.SizeOfIPTReplace { + return syserr.ErrInvalidArgument + } + var replace linux.IPTReplace + replaceBuf := optVal[:linux.SizeOfIPTReplace] + optVal = optVal[linux.SizeOfIPTReplace:] + binary.Unmarshal(replaceBuf, usermem.ByteOrder, &replace) + + // TODO(gvisor.dev/issue/170): Support other tables. + var table iptables.Table + switch replace.Name.String() { + case iptables.TablenameFilter: + table = iptables.EmptyFilterTable() default: - panic(fmt.Sprintf("unknown standard verdict: %d", verdict)) + log.Infof(fmt.Sprintf("We don't yet support writing to the %q table (gvisor.dev/issue/170)", replace.Name.String())) + return syserr.ErrInvalidArgument + } + + // Convert input into a list of rules and their offsets. + var offset uint32 + var offsets []uint32 + for entryIdx := uint32(0); entryIdx < replace.NumEntries; entryIdx++ { + // Get the struct ipt_entry. + if len(optVal) < linux.SizeOfIPTEntry { + return syserr.ErrInvalidArgument + } + var entry linux.IPTEntry + buf := optVal[:linux.SizeOfIPTEntry] + optVal = optVal[linux.SizeOfIPTEntry:] + binary.Unmarshal(buf, usermem.ByteOrder, &entry) + if entry.TargetOffset != linux.SizeOfIPTEntry { + // TODO(gvisor.dev/issue/170): Support matchers. + return syserr.ErrInvalidArgument + } + + // TODO(gvisor.dev/issue/170): We should support IPTIP + // filtering. We reject any nonzero IPTIP values for now. + emptyIPTIP := linux.IPTIP{} + if entry.IP != emptyIPTIP { + return syserr.ErrInvalidArgument + } + + // Get the target of the rule. + target, consumed, err := parseTarget(optVal) + if err != nil { + return err + } + optVal = optVal[consumed:] + + table.Rules = append(table.Rules, iptables.Rule{Target: target}) + offsets = append(offsets, offset) + offset += linux.SizeOfIPTEntry + consumed + } + + // Go through the list of supported hooks for this table and, for each + // one, set the rule it corresponds to. + for hook, _ := range replace.HookEntry { + if table.ValidHooks()&uint32(hook) != 0 { + hk := hookFromLinux(hook) + for ruleIdx, offset := range offsets { + if offset == replace.HookEntry[hook] { + table.BuiltinChains[hk] = ruleIdx + } + if offset == replace.Underflow[hook] { + table.Underflows[hk] = ruleIdx + } + } + if ruleIdx := table.BuiltinChains[hk]; ruleIdx == iptables.HookUnset { + log.Infof("Hook %v is unset.", hk) + return syserr.ErrInvalidArgument + } + if ruleIdx := table.Underflows[hk]; ruleIdx == iptables.HookUnset { + log.Infof("Underflow %v is unset.", hk) + return syserr.ErrInvalidArgument + } + } + } + + ipt := stack.IPTables() + table.SetMetadata(metadata{ + HookEntry: replace.HookEntry, + Underflow: replace.Underflow, + NumEntries: replace.NumEntries, + Size: replace.Size, + }) + ipt.Tables[replace.Name.String()] = table + // TODO: Do we need to worry about locking? We could write rules while + // packets traverse tables. + stack.SetIPTables(ipt) + + return nil +} + +// parseTarget parses a target from the start of optVal and returns the target +// along with the number of bytes it occupies in optVal. +func parseTarget(optVal []byte) (iptables.Target, uint32, *syserr.Error) { + if len(optVal) < linux.SizeOfXTEntryTarget { + return nil, 0, syserr.ErrInvalidArgument + } + var target linux.XTEntryTarget + buf := optVal[:linux.SizeOfXTEntryTarget] + binary.Unmarshal(buf, usermem.ByteOrder, &target) + switch target.Name.String() { + case "": + // Standard target. + if len(optVal) < linux.SizeOfXTStandardTarget { + return nil, 0, syserr.ErrInvalidArgument + } + var target linux.XTStandardTarget + buf = optVal[:linux.SizeOfXTStandardTarget] + binary.Unmarshal(buf, usermem.ByteOrder, &target) + + verdict, err := translateToStandardVerdict(target.Verdict) + if err != nil { + return nil, 0, err + } + switch verdict { + case iptables.Accept: + return iptables.UnconditionalAcceptTarget{}, linux.SizeOfXTStandardTarget, nil + case iptables.Drop: + // TODO(gvisor.dev/issue/170): Return an + // iptables.UnconditionalDropTarget to support DROP. + log.Infof("netfilter DROP is not supported yet.") + return nil, 0, syserr.ErrInvalidArgument + default: + panic(fmt.Sprintf("Unknown verdict: %v", verdict)) + } + + case errorTargetName: + // Error target. + if len(optVal) < linux.SizeOfXTErrorTarget { + return nil, 0, syserr.ErrInvalidArgument + } + var target linux.XTErrorTarget + buf = optVal[:linux.SizeOfXTErrorTarget] + binary.Unmarshal(buf, usermem.ByteOrder, &target) + + // Error targets are used in 2 cases: + // * An actual error case. These rules have an error + // named errorTargetName. The last entry of the table + // is usually an error case to catch any packets that + // somehow fall through every rule. + // * To mark the start of a user defined chain. These + // rules have an error with the name of the chain. + switch target.Name.String() { + case errorTargetName: + return iptables.PanicTarget{}, linux.SizeOfXTErrorTarget, nil + default: + log.Infof("Unknown error target %q doesn't exist or isn't supported yet.", target.Name.String()) + return nil, 0, syserr.ErrInvalidArgument + } + } + + // Unknown target. + log.Infof("Unknown target %q doesn't exist or isn't supported yet.", target.Name.String()) + return nil, 0, syserr.ErrInvalidArgument +} + +func chainNameFromHook(hook int) string { + switch hook { + case linux.NF_INET_PRE_ROUTING: + return iptables.ChainNamePrerouting + case linux.NF_INET_LOCAL_IN: + return iptables.ChainNameInput + case linux.NF_INET_FORWARD: + return iptables.ChainNameForward + case linux.NF_INET_LOCAL_OUT: + return iptables.ChainNameOutput + case linux.NF_INET_POST_ROUTING: + return iptables.ChainNamePostrouting + } + panic(fmt.Sprintf("Unknown hook %d does not correspond to a builtin chain")) +} + +func hookFromLinux(hook int) iptables.Hook { + switch hook { + case linux.NF_INET_PRE_ROUTING: + return iptables.Prerouting + case linux.NF_INET_LOCAL_IN: + return iptables.Input + case linux.NF_INET_FORWARD: + return iptables.Forward + case linux.NF_INET_LOCAL_OUT: + return iptables.Output + case linux.NF_INET_POST_ROUTING: + return iptables.Postrouting + } + panic(fmt.Sprintf("Unknown hook %d does not correspond to a builtin chain")) +} + +// printReplace prints information about the struct ipt_replace in optVal. It +// is only for debugging. +func printReplace(optVal []byte) { + // Basic replace info. + var replace linux.IPTReplace + replaceBuf := optVal[:linux.SizeOfIPTReplace] + optVal = optVal[linux.SizeOfIPTReplace:] + binary.Unmarshal(replaceBuf, usermem.ByteOrder, &replace) + log.Infof("kevin: Replacing table %q: %+v", replace.Name.String(), replace) + + // Read in the list of entries at the end of replace. + var totalOffset uint16 + for entryIdx := uint32(0); entryIdx < replace.NumEntries; entryIdx++ { + var entry linux.IPTEntry + entryBuf := optVal[:linux.SizeOfIPTEntry] + binary.Unmarshal(entryBuf, usermem.ByteOrder, &entry) + log.Infof("kevin: Entry %d (total offset %d): %+v", entryIdx, totalOffset, entry) + + totalOffset += entry.NextOffset + if entry.TargetOffset == linux.SizeOfIPTEntry { + log.Infof("kevin: Entry has no matches.") + } else { + log.Infof("kevin: Entry has matches.") + } + + var target linux.XTEntryTarget + targetBuf := optVal[entry.TargetOffset : entry.TargetOffset+linux.SizeOfXTEntryTarget] + binary.Unmarshal(targetBuf, usermem.ByteOrder, &target) + log.Infof("kevin: Target named %q: %+v", target.Name.String(), target) + + switch target.Name.String() { + case "": + var standardTarget linux.XTStandardTarget + stBuf := optVal[entry.TargetOffset : entry.TargetOffset+linux.SizeOfXTStandardTarget] + binary.Unmarshal(stBuf, usermem.ByteOrder, &standardTarget) + log.Infof("kevin: Standard target with verdict %q (%d).", linux.VerdictStrings[standardTarget.Verdict], standardTarget.Verdict) + case errorTargetName: + var errorTarget linux.XTErrorTarget + etBuf := optVal[entry.TargetOffset : entry.TargetOffset+linux.SizeOfXTErrorTarget] + binary.Unmarshal(etBuf, usermem.ByteOrder, &errorTarget) + log.Infof("kevin: Error target with name %q.", errorTarget.Name.String()) + default: + log.Infof("kevin: Unknown target type.") + } + + optVal = optVal[entry.NextOffset:] } } diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 140851c17..f7caa45b4 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -326,7 +326,7 @@ func AddressAndFamily(sfamily int, addr []byte, strict bool) (tcpip.FullAddress, } family := usermem.ByteOrder.Uint16(addr) - if family != uint16(sfamily) && (strict || family != linux.AF_UNSPEC) { + if family != uint16(sfamily) && (!strict && family != linux.AF_UNSPEC) { return tcpip.FullAddress{}, family, syserr.ErrAddressFamilyNotSupported } @@ -1356,6 +1356,27 @@ func (s *SocketOperations) SetSockOpt(t *kernel.Task, level int, name int, optVa return nil } + if s.skType == linux.SOCK_RAW && level == linux.IPPROTO_IP { + if name == linux.IPT_SO_SET_REPLACE { + if len(optVal) < linux.SizeOfIPTReplace { + return syserr.ErrInvalidArgument + } + + stack := inet.StackFromContext(t) + if stack == nil { + return syserr.ErrNoDevice + } + // Stack must be a netstack stack. + if err := netfilter.SetEntries(stack.(*Stack).Stack, optVal); err != nil { + return err + } + return nil + } else if name == linux.IPT_SO_SET_ADD_COUNTERS { + // TODO(gvisor.dev/issue/170): Counter support. + return nil + } + } + return SetSockOpt(t, s, s.Endpoint, level, name, optVal) } diff --git a/pkg/tcpip/iptables/iptables.go b/pkg/tcpip/iptables/iptables.go index 68c68d4aa..9e7005374 100644 --- a/pkg/tcpip/iptables/iptables.go +++ b/pkg/tcpip/iptables/iptables.go @@ -17,65 +17,107 @@ package iptables const ( - tablenameNat = "nat" - tablenameMangle = "mangle" + TablenameNat = "nat" + TablenameMangle = "mangle" + TablenameFilter = "filter" ) +// TODO: Make this an iota? Faster! Do it. // Chain names as defined by net/ipv4/netfilter/ip_tables.c. const ( - chainNamePrerouting = "PREROUTING" - chainNameInput = "INPUT" - chainNameForward = "FORWARD" - chainNameOutput = "OUTPUT" - chainNamePostrouting = "POSTROUTING" + ChainNamePrerouting = "PREROUTING" + ChainNameInput = "INPUT" + ChainNameForward = "FORWARD" + ChainNameOutput = "OUTPUT" + ChainNamePostrouting = "POSTROUTING" ) +const HookUnset = -1 + // DefaultTables returns a default set of tables. Each chain is set to accept // all packets. func DefaultTables() IPTables { return IPTables{ Tables: map[string]Table{ - tablenameNat: Table{ - BuiltinChains: map[Hook]Chain{ - Prerouting: unconditionalAcceptChain(chainNamePrerouting), - Input: unconditionalAcceptChain(chainNameInput), - Output: unconditionalAcceptChain(chainNameOutput), - Postrouting: unconditionalAcceptChain(chainNamePostrouting), + TablenameNat: Table{ + Rules: []Rule{ + Rule{Target: UnconditionalAcceptTarget{}}, + Rule{Target: UnconditionalAcceptTarget{}}, + Rule{Target: UnconditionalAcceptTarget{}}, + Rule{Target: UnconditionalAcceptTarget{}}, + Rule{Target: PanicTarget{}}, + }, + BuiltinChains: map[Hook]int{ + Prerouting: 0, + Input: 1, + Output: 2, + Postrouting: 3, }, - DefaultTargets: map[Hook]Target{ - Prerouting: UnconditionalAcceptTarget{}, - Input: UnconditionalAcceptTarget{}, - Output: UnconditionalAcceptTarget{}, - Postrouting: UnconditionalAcceptTarget{}, + Underflows: map[Hook]int{ + Prerouting: 0, + Input: 1, + Output: 2, + Postrouting: 3, }, - UserChains: map[string]Chain{}, + UserChains: map[string]int{}, }, - tablenameMangle: Table{ - BuiltinChains: map[Hook]Chain{ - Prerouting: unconditionalAcceptChain(chainNamePrerouting), - Output: unconditionalAcceptChain(chainNameOutput), + TablenameMangle: Table{ + Rules: []Rule{ + Rule{Target: UnconditionalAcceptTarget{}}, + Rule{Target: UnconditionalAcceptTarget{}}, + Rule{Target: PanicTarget{}}, + }, + BuiltinChains: map[Hook]int{ + Prerouting: 0, + Output: 1, }, - DefaultTargets: map[Hook]Target{ - Prerouting: UnconditionalAcceptTarget{}, - Output: UnconditionalAcceptTarget{}, + Underflows: map[Hook]int{ + Prerouting: 0, + Output: 1, }, - UserChains: map[string]Chain{}, + UserChains: map[string]int{}, + }, + TablenameFilter: Table{ + Rules: []Rule{ + Rule{Target: UnconditionalAcceptTarget{}}, + Rule{Target: UnconditionalAcceptTarget{}}, + Rule{Target: UnconditionalAcceptTarget{}}, + Rule{Target: PanicTarget{}}, + }, + BuiltinChains: map[Hook]int{ + Input: 0, + Forward: 1, + Output: 2, + }, + Underflows: map[Hook]int{ + Input: 0, + Forward: 1, + Output: 2, + }, + UserChains: map[string]int{}, }, }, Priorities: map[Hook][]string{ - Prerouting: []string{tablenameMangle, tablenameNat}, - Output: []string{tablenameMangle, tablenameNat}, + Input: []string{TablenameNat, TablenameFilter}, + Prerouting: []string{TablenameMangle, TablenameNat}, + Output: []string{TablenameMangle, TablenameNat, TablenameFilter}, }, } } -func unconditionalAcceptChain(name string) Chain { - return Chain{ - Name: name, - Rules: []Rule{ - Rule{ - Target: UnconditionalAcceptTarget{}, - }, +func EmptyFilterTable() Table { + return Table{ + Rules: []Rule{}, + BuiltinChains: map[Hook]int{ + Input: HookUnset, + Forward: HookUnset, + Output: HookUnset, + }, + Underflows: map[Hook]int{ + Input: HookUnset, + Forward: HookUnset, + Output: HookUnset, }, + UserChains: map[string]int{}, } } diff --git a/pkg/tcpip/iptables/targets.go b/pkg/tcpip/iptables/targets.go index 19a7f77e3..03c9f19ff 100644 --- a/pkg/tcpip/iptables/targets.go +++ b/pkg/tcpip/iptables/targets.go @@ -33,3 +33,11 @@ type UnconditionalDropTarget struct{} func (UnconditionalDropTarget) Action(packet buffer.VectorisedView) (Verdict, string) { return Drop, "" } + +// PanicTarget just panics. +type PanicTarget struct{} + +// Actions implements Target.Action. +func (PanicTarget) Action(packet buffer.VectorisedView) (Verdict, string) { + panic("PanicTarget triggered.") +} diff --git a/pkg/tcpip/iptables/types.go b/pkg/tcpip/iptables/types.go index 42a79ef9f..76364ff1f 100644 --- a/pkg/tcpip/iptables/types.go +++ b/pkg/tcpip/iptables/types.go @@ -61,9 +61,12 @@ const ( type Verdict int const ( + // Invalid indicates an unkonwn or erroneous verdict. + Invalid Verdict = iota + // Accept indicates the packet should continue traversing netstack as // normal. - Accept Verdict = iota + Accept // Drop inicates the packet should be dropped, stopping traversing // netstack. @@ -109,24 +112,18 @@ type IPTables struct { // * nat // * mangle type Table struct { - // BuiltinChains holds the un-deletable chains built into netstack. If - // a hook isn't present in the map, this table doesn't utilize that - // hook. - BuiltinChains map[Hook]Chain + // A table is just a list of rules with some entrypoints. + Rules []Rule + + BuiltinChains map[Hook]int + + Underflows map[Hook]int - // DefaultTargets holds a target for each hook that will be executed if - // chain traversal doesn't yield a verdict. - DefaultTargets map[Hook]Target + // DefaultTargets map[Hook]int // UserChains holds user-defined chains for the keyed by name. Users // can give their chains arbitrary names. - UserChains map[string]Chain - - // Chains maps names to chains for both builtin and user-defined chains. - // Its entries point to Chains already either in BuiltinChains or - // UserChains, and its purpose is to make looking up tables by name - // fast. - Chains map[string]*Chain + UserChains map[string]int // Metadata holds information about the Table that is useful to users // of IPTables, but not to the netstack IPTables code itself. @@ -152,20 +149,20 @@ func (table *Table) SetMetadata(metadata interface{}) { table.metadata = metadata } -// A Chain defines a list of rules for packet processing. When a packet -// traverses a chain, it is checked against each rule until either a rule -// returns a verdict or the chain ends. -// -// By convention, builtin chains end with a rule that matches everything and -// returns either Accept or Drop. User-defined chains end with Return. These -// aren't strictly necessary here, but the iptables tool writes tables this way. -type Chain struct { - // Name is the chain name. - Name string - - // Rules is the list of rules to traverse. - Rules []Rule -} +//// A Chain defines a list of rules for packet processing. When a packet +//// traverses a chain, it is checked against each rule until either a rule +//// returns a verdict or the chain ends. +//// +//// By convention, builtin chains end with a rule that matches everything and +//// returns either Accept or Drop. User-defined chains end with Return. These +//// aren't strictly necessary here, but the iptables tool writes tables this way. +//type Chain struct { +// // Name is the chain name. +// Name string + +// // Rules is the list of rules to traverse. +// Rules []Rule +//} // A Rule is a packet processing rule. It consists of two pieces. First it // contains zero or more matchers, each of which is a specification of which diff --git a/test/iptables/filter_input.go b/test/iptables/filter_input.go index 923f44e68..0cb668635 100644 --- a/test/iptables/filter_input.go +++ b/test/iptables/filter_input.go @@ -44,6 +44,7 @@ func (FilterInputDropUDP) Name() string { // ContainerAction implements TestCase.ContainerAction. func (FilterInputDropUDP) ContainerAction(ip net.IP) error { if err := filterTable("-A", "INPUT", "-p", "udp", "-j", "DROP"); err != nil { + // if err := filterTable("-A", "INPUT", "-j", "ACCEPT"); err != nil { return err } diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go index bfbf1bb87..e761e0f2f 100644 --- a/test/iptables/iptables_test.go +++ b/test/iptables/iptables_test.go @@ -23,6 +23,7 @@ import ( "time" "flag" + "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/runsc/dockerutil" "gvisor.dev/gvisor/runsc/testutil" @@ -166,14 +167,14 @@ func TestFilterInputDropUDP(t *testing.T) { } } -func TestFilterInputDropUDPPort(t *testing.T) { - if err := singleTest(FilterInputDropUDPPort{}); err != nil { - t.Fatal(err) - } -} - -func TestFilterInputDropDifferentUDPPort(t *testing.T) { - if err := singleTest(FilterInputDropDifferentUDPPort{}); err != nil { - t.Fatal(err) - } -} +// func TestFilterInputDropUDPPort(t *testing.T) { +// if err := singleTest(FilterInputDropUDPPort{}); err != nil { +// t.Fatal(err) +// } +// } + +// func TestFilterInputDropDifferentUDPPort(t *testing.T) { +// if err := singleTest(FilterInputDropDifferentUDPPort{}); err != nil { +// t.Fatal(err) +// } +// } -- cgit v1.2.3 From db376e13924be59182ed4df95762328febf26298 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Wed, 8 Jan 2020 10:30:50 -0800 Subject: Make /proc/[pid] offset start at TGID_OFFSET Updates #1195 PiperOrigin-RevId: 288725745 --- pkg/sentry/fsimpl/proc/tasks.go | 102 ++++++++++++++++++------ pkg/sentry/fsimpl/proc/tasks_test.go | 147 ++++++++++++++++++++++++++++++++++- 2 files changed, 223 insertions(+), 26 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index 50b2a832f..d8f92d52f 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -27,7 +27,11 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) -const defaultPermission = 0444 +const ( + defaultPermission = 0444 + selfName = "self" + threadSelfName = "thread-self" +) // InoGenerator generates unique inode numbers for a given filesystem. type InoGenerator interface { @@ -45,6 +49,11 @@ type tasksInode struct { inoGen InoGenerator pidns *kernel.PIDNamespace + + // '/proc/self' and '/proc/thread-self' have custom directory offsets in + // Linux. So handle them outside of OrderedChildren. + selfSymlink *vfs.Dentry + threadSelfSymlink *vfs.Dentry } var _ kernfs.Inode = (*tasksInode)(nil) @@ -54,20 +63,20 @@ func newTasksInode(inoGen InoGenerator, k *kernel.Kernel, pidns *kernel.PIDNames contents := map[string]*kernfs.Dentry{ //"cpuinfo": newCPUInfo(ctx, msrc), //"filesystems": seqfile.NewSeqFileInode(ctx, &filesystemsData{}, msrc), - "loadavg": newDentry(root, inoGen.NextIno(), defaultPermission, &loadavgData{}), - "meminfo": newDentry(root, inoGen.NextIno(), defaultPermission, &meminfoData{k: k}), - "mounts": kernfs.NewStaticSymlink(root, inoGen.NextIno(), defaultPermission, "self/mounts"), - "self": newSelfSymlink(root, inoGen.NextIno(), defaultPermission, pidns), - "stat": newDentry(root, inoGen.NextIno(), defaultPermission, &statData{k: k}), - "thread-self": newThreadSelfSymlink(root, inoGen.NextIno(), defaultPermission, pidns), + "loadavg": newDentry(root, inoGen.NextIno(), defaultPermission, &loadavgData{}), + "meminfo": newDentry(root, inoGen.NextIno(), defaultPermission, &meminfoData{k: k}), + "mounts": kernfs.NewStaticSymlink(root, inoGen.NextIno(), defaultPermission, "self/mounts"), + "stat": newDentry(root, inoGen.NextIno(), defaultPermission, &statData{k: k}), //"uptime": newUptime(ctx, msrc), //"version": newVersionData(root, inoGen.NextIno(), k), "version": newDentry(root, inoGen.NextIno(), defaultPermission, &versionData{k: k}), } inode := &tasksInode{ - pidns: pidns, - inoGen: inoGen, + pidns: pidns, + inoGen: inoGen, + selfSymlink: newSelfSymlink(root, inoGen.NextIno(), 0444, pidns).VFSDentry(), + threadSelfSymlink: newThreadSelfSymlink(root, inoGen.NextIno(), 0444, pidns).VFSDentry(), } inode.InodeAttrs.Init(root, inoGen.NextIno(), linux.ModeDirectory|0555) @@ -86,6 +95,13 @@ func (i *tasksInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, erro // Try to lookup a corresponding task. tid, err := strconv.ParseUint(name, 10, 64) if err != nil { + // If it failed to parse, check if it's one of the special handled files. + switch name { + case selfName: + return i.selfSymlink, nil + case threadSelfName: + return i.threadSelfSymlink, nil + } return nil, syserror.ENOENT } @@ -104,41 +120,81 @@ func (i *tasksInode) Valid(ctx context.Context) bool { } // IterDirents implements kernfs.inodeDynamicLookup. -// -// TODO(gvisor.dev/issue/1195): Use tgid N offset = TGID_OFFSET + N. -func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, offset, relOffset int64) (int64, error) { - var tids []int +func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, offset, _ int64) (int64, error) { + // fs/proc/internal.h: #define FIRST_PROCESS_ENTRY 256 + const FIRST_PROCESS_ENTRY = 256 + + // Use maxTaskID to shortcut searches that will result in 0 entries. + const maxTaskID = kernel.TasksLimit + 1 + if offset >= maxTaskID { + return offset, nil + } + + // According to Linux (fs/proc/base.c:proc_pid_readdir()), process directories + // start at offset FIRST_PROCESS_ENTRY with '/proc/self', followed by + // '/proc/thread-self' and then '/proc/[pid]'. + if offset < FIRST_PROCESS_ENTRY { + offset = FIRST_PROCESS_ENTRY + } + + if offset == FIRST_PROCESS_ENTRY { + dirent := vfs.Dirent{ + Name: selfName, + Type: linux.DT_LNK, + Ino: i.inoGen.NextIno(), + NextOff: offset + 1, + } + if !cb.Handle(dirent) { + return offset, nil + } + offset++ + } + if offset == FIRST_PROCESS_ENTRY+1 { + dirent := vfs.Dirent{ + Name: threadSelfName, + Type: linux.DT_LNK, + Ino: i.inoGen.NextIno(), + NextOff: offset + 1, + } + if !cb.Handle(dirent) { + return offset, nil + } + offset++ + } - // Collect all tasks. Per linux we only include it in directory listings if - // it's the leader. But for whatever crazy reason, you can still walk to the - // given node. + // Collect all tasks that TGIDs are greater than the offset specified. Per + // Linux we only include in directory listings if it's the leader. But for + // whatever crazy reason, you can still walk to the given node. + var tids []int + startTid := offset - FIRST_PROCESS_ENTRY - 2 for _, tg := range i.pidns.ThreadGroups() { + tid := i.pidns.IDOfThreadGroup(tg) + if int64(tid) < startTid { + continue + } if leader := tg.Leader(); leader != nil { - tids = append(tids, int(i.pidns.IDOfThreadGroup(tg))) + tids = append(tids, int(tid)) } } if len(tids) == 0 { return offset, nil } - if relOffset >= int64(len(tids)) { - return offset, nil - } sort.Ints(tids) - for _, tid := range tids[relOffset:] { + for _, tid := range tids { dirent := vfs.Dirent{ Name: strconv.FormatUint(uint64(tid), 10), Type: linux.DT_DIR, Ino: i.inoGen.NextIno(), - NextOff: offset + 1, + NextOff: FIRST_PROCESS_ENTRY + 2 + int64(tid) + 1, } if !cb.Handle(dirent) { return offset, nil } offset++ } - return offset, nil + return maxTaskID, nil } // Open implements kernfs.Inode. diff --git a/pkg/sentry/fsimpl/proc/tasks_test.go b/pkg/sentry/fsimpl/proc/tasks_test.go index 2560fcef9..ca8c87ec2 100644 --- a/pkg/sentry/fsimpl/proc/tasks_test.go +++ b/pkg/sentry/fsimpl/proc/tasks_test.go @@ -16,6 +16,7 @@ package proc import ( "fmt" + "math" "path" "strconv" "testing" @@ -30,6 +31,18 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) +var ( + // Next offset 256 by convention. Adds 1 for the next offset. + selfLink = vfs.Dirent{Type: linux.DT_LNK, NextOff: 256 + 0 + 1} + threadSelfLink = vfs.Dirent{Type: linux.DT_LNK, NextOff: 256 + 1 + 1} + + // /proc/[pid] next offset starts at 256+2 (files above), then adds the + // PID, and adds 1 for the next offset. + proc1 = vfs.Dirent{Type: linux.DT_DIR, NextOff: 258 + 1 + 1} + proc2 = vfs.Dirent{Type: linux.DT_DIR, NextOff: 258 + 2 + 1} + proc3 = vfs.Dirent{Type: linux.DT_DIR, NextOff: 258 + 3 + 1} +) + type testIterDirentsCallback struct { dirents []vfs.Dirent } @@ -59,9 +72,9 @@ func checkTasksStaticFiles(gots []vfs.Dirent) ([]vfs.Dirent, error) { "loadavg": {Type: linux.DT_REG}, "meminfo": {Type: linux.DT_REG}, "mounts": {Type: linux.DT_LNK}, - "self": {Type: linux.DT_LNK}, + "self": selfLink, "stat": {Type: linux.DT_REG}, - "thread-self": {Type: linux.DT_LNK}, + "thread-self": threadSelfLink, "version": {Type: linux.DT_REG}, } return checkFiles(gots, wants) @@ -93,6 +106,9 @@ func checkFiles(gots []vfs.Dirent, wants map[string]vfs.Dirent) ([]vfs.Dirent, e if want.Type != got.Type { return gots, fmt.Errorf("wrong file type, want: %v, got: %v: %+v", want.Type, got.Type, got) } + if want.NextOff != 0 && want.NextOff != got.NextOff { + return gots, fmt.Errorf("wrong dirent offset, want: %v, got: %v: %+v", want.NextOff, got.NextOff, got) + } delete(wants, got.Name) gots = append(gots[0:i], gots[i+1:]...) @@ -154,7 +170,7 @@ func TestTasksEmpty(t *testing.T) { t.Error(err.Error()) } if len(cb.dirents) != 0 { - t.Error("found more files than expected: %+v", cb.dirents) + t.Errorf("found more files than expected: %+v", cb.dirents) } } @@ -216,6 +232,11 @@ func TestTasks(t *testing.T) { if !found { t.Errorf("Additional task ID %d listed: %v", pid, tasks) } + // Next offset starts at 256+2 ('self' and 'thread-self'), then adds the + // PID, and adds 1 for the next offset. + if want := int64(256 + 2 + pid + 1); d.NextOff != want { + t.Errorf("Wrong dirent offset want: %d got: %d: %+v", want, d.NextOff, d) + } } // Test lookup. @@ -246,6 +267,126 @@ func TestTasks(t *testing.T) { } } +func TestTasksOffset(t *testing.T) { + ctx, vfsObj, root, err := setup() + if err != nil { + t.Fatalf("Setup failed: %v", err) + } + defer root.DecRef() + + k := kernel.KernelFromContext(ctx) + for i := 0; i < 3; i++ { + tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) + if _, err := createTask(ctx, fmt.Sprintf("name-%d", i), tc); err != nil { + t.Fatalf("CreateTask(): %v", err) + } + } + + for _, tc := range []struct { + name string + offset int64 + wants map[string]vfs.Dirent + }{ + { + name: "small offset", + offset: 100, + wants: map[string]vfs.Dirent{ + "self": selfLink, + "thread-self": threadSelfLink, + "1": proc1, + "2": proc2, + "3": proc3, + }, + }, + { + name: "offset at start", + offset: 256, + wants: map[string]vfs.Dirent{ + "self": selfLink, + "thread-self": threadSelfLink, + "1": proc1, + "2": proc2, + "3": proc3, + }, + }, + { + name: "skip /proc/self", + offset: 257, + wants: map[string]vfs.Dirent{ + "thread-self": threadSelfLink, + "1": proc1, + "2": proc2, + "3": proc3, + }, + }, + { + name: "skip symlinks", + offset: 258, + wants: map[string]vfs.Dirent{ + "1": proc1, + "2": proc2, + "3": proc3, + }, + }, + { + name: "skip first process", + offset: 260, + wants: map[string]vfs.Dirent{ + "2": proc2, + "3": proc3, + }, + }, + { + name: "last process", + offset: 261, + wants: map[string]vfs.Dirent{ + "3": proc3, + }, + }, + { + name: "after last", + offset: 262, + wants: nil, + }, + { + name: "TaskLimit+1", + offset: kernel.TasksLimit + 1, + wants: nil, + }, + { + name: "max", + offset: math.MaxInt64, + wants: nil, + }, + } { + t.Run(tc.name, func(t *testing.T) { + fd, err := vfsObj.OpenAt( + ctx, + auth.CredentialsFromContext(ctx), + &vfs.PathOperation{Root: root, Start: root, Path: fspath.Parse("/")}, + &vfs.OpenOptions{}, + ) + if err != nil { + t.Fatalf("vfsfs.OpenAt(/) failed: %v", err) + } + if _, err := fd.Impl().Seek(ctx, tc.offset, linux.SEEK_SET); err != nil { + t.Fatalf("Seek(%d, SEEK_SET): %v", tc.offset, err) + } + + cb := testIterDirentsCallback{} + if err := fd.Impl().IterDirents(ctx, &cb); err != nil { + t.Fatalf("IterDirents(): %v", err) + } + if cb.dirents, err = checkFiles(cb.dirents, tc.wants); err != nil { + t.Error(err.Error()) + } + if len(cb.dirents) != 0 { + t.Errorf("found more files than expected: %+v", cb.dirents) + } + }) + } +} + func TestTask(t *testing.T) { ctx, vfsObj, root, err := setup() if err != nil { -- cgit v1.2.3 From 1e1921e2acdb7357972257219fdffb9edf17bf55 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Wed, 8 Jan 2020 11:15:46 -0800 Subject: Minor fixes to comments and logging --- pkg/sentry/socket/netfilter/netfilter.go | 10 +++++++--- pkg/sentry/socket/netstack/netstack.go | 8 +++++--- pkg/sentry/syscalls/linux/sys_socket.go | 2 +- pkg/tcpip/iptables/targets.go | 2 +- pkg/tcpip/iptables/types.go | 28 ++++++---------------------- test/iptables/filter_input.go | 3 +-- test/iptables/iptables_test.go | 22 +++++++++++----------- 7 files changed, 32 insertions(+), 43 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 8c7f3c7fc..b7867a576 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -157,9 +157,7 @@ func convertNetstackToBinary(name string, table iptables.Table) (linux.KernelIPT meta.HookEntry[hook] = entries.Size } } - // Is this a chain underflow point? The underflow rule is the last rule - // in the chain, and is an unconditional rule (i.e. it matches any - // packet). This is enforced when saving iptables. + // Is this a chain underflow point? for underflow, underflowRuleIdx := range table.Underflows { if underflowRuleIdx == ruleIdx { meta.Underflow[underflow] = entries.Size @@ -290,6 +288,7 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { // Get the basic rules data (struct ipt_replace). if len(optVal) < linux.SizeOfIPTReplace { + log.Infof("netfilter.SetEntries: optVal has insufficient size for replace %d", len(optVal)) return syserr.ErrInvalidArgument } var replace linux.IPTReplace @@ -313,6 +312,7 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { for entryIdx := uint32(0); entryIdx < replace.NumEntries; entryIdx++ { // Get the struct ipt_entry. if len(optVal) < linux.SizeOfIPTEntry { + log.Infof("netfilter: optVal has insufficient size for entry %d", len(optVal)) return syserr.ErrInvalidArgument } var entry linux.IPTEntry @@ -328,6 +328,7 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { // filtering. We reject any nonzero IPTIP values for now. emptyIPTIP := linux.IPTIP{} if entry.IP != emptyIPTIP { + log.Infof("netfilter: non-empty struct iptip found") return syserr.ErrInvalidArgument } @@ -386,6 +387,7 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { // along with the number of bytes it occupies in optVal. func parseTarget(optVal []byte) (iptables.Target, uint32, *syserr.Error) { if len(optVal) < linux.SizeOfXTEntryTarget { + log.Infof("netfilter: optVal has insufficient size for entry target %d", len(optVal)) return nil, 0, syserr.ErrInvalidArgument } var target linux.XTEntryTarget @@ -395,6 +397,7 @@ func parseTarget(optVal []byte) (iptables.Target, uint32, *syserr.Error) { case "": // Standard target. if len(optVal) < linux.SizeOfXTStandardTarget { + log.Infof("netfilter.SetEntries: optVal has insufficient size for standard target %d", len(optVal)) return nil, 0, syserr.ErrInvalidArgument } var target linux.XTStandardTarget @@ -420,6 +423,7 @@ func parseTarget(optVal []byte) (iptables.Target, uint32, *syserr.Error) { case errorTargetName: // Error target. if len(optVal) < linux.SizeOfXTErrorTarget { + log.Infof("netfilter.SetEntries: optVal has insufficient size for error target %d", len(optVal)) return nil, 0, syserr.ErrInvalidArgument } var target linux.XTErrorTarget diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index f7caa45b4..8c07eef4b 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -326,7 +326,7 @@ func AddressAndFamily(sfamily int, addr []byte, strict bool) (tcpip.FullAddress, } family := usermem.ByteOrder.Uint16(addr) - if family != uint16(sfamily) && (!strict && family != linux.AF_UNSPEC) { + if family != uint16(sfamily) && (strict || family != linux.AF_UNSPEC) { return tcpip.FullAddress{}, family, syserr.ErrAddressFamilyNotSupported } @@ -1357,7 +1357,8 @@ func (s *SocketOperations) SetSockOpt(t *kernel.Task, level int, name int, optVa } if s.skType == linux.SOCK_RAW && level == linux.IPPROTO_IP { - if name == linux.IPT_SO_SET_REPLACE { + switch name { + case linux.IPT_SO_SET_REPLACE: if len(optVal) < linux.SizeOfIPTReplace { return syserr.ErrInvalidArgument } @@ -1371,7 +1372,8 @@ func (s *SocketOperations) SetSockOpt(t *kernel.Task, level int, name int, optVa return err } return nil - } else if name == linux.IPT_SO_SET_ADD_COUNTERS { + + case linux.IPT_SO_SET_ADD_COUNTERS: // TODO(gvisor.dev/issue/170): Counter support. return nil } diff --git a/pkg/sentry/syscalls/linux/sys_socket.go b/pkg/sentry/syscalls/linux/sys_socket.go index 4b5aafcc0..cda517a81 100644 --- a/pkg/sentry/syscalls/linux/sys_socket.go +++ b/pkg/sentry/syscalls/linux/sys_socket.go @@ -41,7 +41,7 @@ const maxListenBacklog = 1024 const maxAddrLen = 200 // maxOptLen is the maximum sockopt parameter length we're willing to accept. -const maxOptLen = 1024 +const maxOptLen = 1024 * 8 // maxControlLen is the maximum length of the msghdr.msg_control buffer we're // willing to accept. Note that this limit is smaller than Linux, which allows diff --git a/pkg/tcpip/iptables/targets.go b/pkg/tcpip/iptables/targets.go index 03c9f19ff..2c3598e3d 100644 --- a/pkg/tcpip/iptables/targets.go +++ b/pkg/tcpip/iptables/targets.go @@ -34,7 +34,7 @@ func (UnconditionalDropTarget) Action(packet buffer.VectorisedView) (Verdict, st return Drop, "" } -// PanicTarget just panics. +// PanicTarget just panics. It represents a target that should be unreachable. type PanicTarget struct{} // Actions implements Target.Action. diff --git a/pkg/tcpip/iptables/types.go b/pkg/tcpip/iptables/types.go index 76364ff1f..fe0394a31 100644 --- a/pkg/tcpip/iptables/types.go +++ b/pkg/tcpip/iptables/types.go @@ -107,20 +107,19 @@ type IPTables struct { Priorities map[Hook][]string } -// A Table defines a set of chains and hooks into the network stack. The -// currently supported tables are: -// * nat -// * mangle +// A Table defines a set of chains and hooks into the network stack. It is +// really just a list of rules with some metadata for entrypoints and such. type Table struct { - // A table is just a list of rules with some entrypoints. + // Rules holds the rules that make up the table. Rules []Rule + // BuiltinChains maps builtin chains to their entrypoints. BuiltinChains map[Hook]int + // Underflows maps builtin chains to their underflow point (i.e. the + // rule to execute if the chain returns without a verdict). Underflows map[Hook]int - // DefaultTargets map[Hook]int - // UserChains holds user-defined chains for the keyed by name. Users // can give their chains arbitrary names. UserChains map[string]int @@ -149,21 +148,6 @@ func (table *Table) SetMetadata(metadata interface{}) { table.metadata = metadata } -//// A Chain defines a list of rules for packet processing. When a packet -//// traverses a chain, it is checked against each rule until either a rule -//// returns a verdict or the chain ends. -//// -//// By convention, builtin chains end with a rule that matches everything and -//// returns either Accept or Drop. User-defined chains end with Return. These -//// aren't strictly necessary here, but the iptables tool writes tables this way. -//type Chain struct { -// // Name is the chain name. -// Name string - -// // Rules is the list of rules to traverse. -// Rules []Rule -//} - // A Rule is a packet processing rule. It consists of two pieces. First it // contains zero or more matchers, each of which is a specification of which // packets this rule applies to. If there are no matchers in the rule, it diff --git a/test/iptables/filter_input.go b/test/iptables/filter_input.go index 0cb668635..34a85db97 100644 --- a/test/iptables/filter_input.go +++ b/test/iptables/filter_input.go @@ -43,8 +43,7 @@ func (FilterInputDropUDP) Name() string { // ContainerAction implements TestCase.ContainerAction. func (FilterInputDropUDP) ContainerAction(ip net.IP) error { - if err := filterTable("-A", "INPUT", "-p", "udp", "-j", "DROP"); err != nil { - // if err := filterTable("-A", "INPUT", "-j", "ACCEPT"); err != nil { + if err := filterTable("-A", "INPUT", "-j", "ACCEPT"); err != nil { return err } diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go index e761e0f2f..2465a4e65 100644 --- a/test/iptables/iptables_test.go +++ b/test/iptables/iptables_test.go @@ -167,14 +167,14 @@ func TestFilterInputDropUDP(t *testing.T) { } } -// func TestFilterInputDropUDPPort(t *testing.T) { -// if err := singleTest(FilterInputDropUDPPort{}); err != nil { -// t.Fatal(err) -// } -// } - -// func TestFilterInputDropDifferentUDPPort(t *testing.T) { -// if err := singleTest(FilterInputDropDifferentUDPPort{}); err != nil { -// t.Fatal(err) -// } -// } +func TestFilterInputDropUDPPort(t *testing.T) { + if err := singleTest(FilterInputDropUDPPort{}); err != nil { + t.Fatal(err) + } +} + +func TestFilterInputDropDifferentUDPPort(t *testing.T) { + if err := singleTest(FilterInputDropDifferentUDPPort{}); err != nil { + t.Fatal(err) + } +} -- cgit v1.2.3 From 446a250996d9c946d9a5279f7fd081cc1be0bd11 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Wed, 8 Jan 2020 11:20:48 -0800 Subject: Comment cleanup. --- pkg/abi/linux/netfilter.go | 2 ++ pkg/sentry/socket/netfilter/netfilter.go | 2 -- pkg/tcpip/iptables/types.go | 6 +++--- 3 files changed, 5 insertions(+), 5 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/netfilter.go b/pkg/abi/linux/netfilter.go index 0bcb232de..35d66d622 100644 --- a/pkg/abi/linux/netfilter.go +++ b/pkg/abi/linux/netfilter.go @@ -42,6 +42,8 @@ const ( NF_RETURN = -NF_REPEAT - 1 ) +// VerdictStrings maps int verdicts to the strings they represent. It is used +// for debugging. var VerdictStrings = map[int32]string{ -NF_DROP - 1: "DROP", -NF_ACCEPT - 1: "ACCEPT", diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index b7867a576..347342f98 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -376,8 +376,6 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { Size: replace.Size, }) ipt.Tables[replace.Name.String()] = table - // TODO: Do we need to worry about locking? We could write rules while - // packets traverse tables. stack.SetIPTables(ipt) return nil diff --git a/pkg/tcpip/iptables/types.go b/pkg/tcpip/iptables/types.go index fe0394a31..540f8c0b4 100644 --- a/pkg/tcpip/iptables/types.go +++ b/pkg/tcpip/iptables/types.go @@ -113,11 +113,11 @@ type Table struct { // Rules holds the rules that make up the table. Rules []Rule - // BuiltinChains maps builtin chains to their entrypoints. + // BuiltinChains maps builtin chains to their entrypoint rule in Rules. BuiltinChains map[Hook]int - // Underflows maps builtin chains to their underflow point (i.e. the - // rule to execute if the chain returns without a verdict). + // Underflows maps builtin chains to their underflow rule in Rules + // (i.e. the rule to execute if the chain returns without a verdict). Underflows map[Hook]int // UserChains holds user-defined chains for the keyed by name. Users -- cgit v1.2.3 From 7cebd77806d164a3baec52eaeb05662e8c404967 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Wed, 8 Jan 2020 12:43:46 -0800 Subject: First commit -- re-adding DROP --- pkg/sentry/socket/netfilter/netfilter.go | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 347342f98..e4c493141 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -410,10 +410,7 @@ func parseTarget(optVal []byte) (iptables.Target, uint32, *syserr.Error) { case iptables.Accept: return iptables.UnconditionalAcceptTarget{}, linux.SizeOfXTStandardTarget, nil case iptables.Drop: - // TODO(gvisor.dev/issue/170): Return an - // iptables.UnconditionalDropTarget to support DROP. - log.Infof("netfilter DROP is not supported yet.") - return nil, 0, syserr.ErrInvalidArgument + return iptables.UnconditionalDropTarget{}, linux.SizeOfXTStandardTarget, nil default: panic(fmt.Sprintf("Unknown verdict: %v", verdict)) } -- cgit v1.2.3 From a271bccfc61390be64ca0175b8fc7d20e66d05b6 Mon Sep 17 00:00:00 2001 From: Tamir Duberstein Date: Wed, 8 Jan 2020 14:16:38 -0800 Subject: Rename tcpip.SockOpt{,Int} PiperOrigin-RevId: 288772878 --- pkg/sentry/socket/netstack/netstack.go | 4 ++-- pkg/sentry/socket/unix/transport/unix.go | 8 ++++---- pkg/tcpip/stack/transport_test.go | 4 ++-- pkg/tcpip/tcpip.go | 10 +++++----- pkg/tcpip/transport/icmp/endpoint.go | 4 ++-- pkg/tcpip/transport/packet/endpoint.go | 4 ++-- pkg/tcpip/transport/raw/endpoint.go | 4 ++-- pkg/tcpip/transport/tcp/endpoint.go | 4 ++-- pkg/tcpip/transport/udp/endpoint.go | 4 ++-- 9 files changed, 23 insertions(+), 23 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 140851c17..5f91a0d1a 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -224,7 +224,7 @@ type commonEndpoint interface { // SetSockOptInt implements tcpip.Endpoint.SetSockOptInt and // transport.Endpoint.SetSockOptInt. - SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error + SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error // GetSockOpt implements tcpip.Endpoint.GetSockOpt and // transport.Endpoint.GetSockOpt. @@ -232,7 +232,7 @@ type commonEndpoint interface { // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt and // transport.Endpoint.GetSockOpt. - GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) + GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) } // SocketOperations encapsulates all the state needed to represent a network stack diff --git a/pkg/sentry/socket/unix/transport/unix.go b/pkg/sentry/socket/unix/transport/unix.go index 529a7a7a9..fcba49435 100644 --- a/pkg/sentry/socket/unix/transport/unix.go +++ b/pkg/sentry/socket/unix/transport/unix.go @@ -177,7 +177,7 @@ type Endpoint interface { // SetSockOptInt sets a socket option for simple cases when a value has // the int type. - SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error + SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error // GetSockOpt gets a socket option. opt should be a pointer to one of the // tcpip.*Option types. @@ -185,7 +185,7 @@ type Endpoint interface { // GetSockOptInt gets a socket option for simple cases when a return // value has the int type. - GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) + GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) // State returns the current state of the socket, as represented by Linux in // procfs. @@ -851,11 +851,11 @@ func (e *baseEndpoint) SetSockOpt(opt interface{}) *tcpip.Error { return nil } -func (e *baseEndpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error { +func (e *baseEndpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { return nil } -func (e *baseEndpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) { +func (e *baseEndpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { switch opt { case tcpip.ReceiveQueueSizeOption: v := 0 diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go index 748ce4ea5..095346f0b 100644 --- a/pkg/tcpip/stack/transport_test.go +++ b/pkg/tcpip/stack/transport_test.go @@ -103,12 +103,12 @@ func (*fakeTransportEndpoint) SetSockOpt(interface{}) *tcpip.Error { } // SetSockOptInt sets a socket option. Currently not supported. -func (*fakeTransportEndpoint) SetSockOptInt(tcpip.SockOpt, int) *tcpip.Error { +func (*fakeTransportEndpoint) SetSockOptInt(tcpip.SockOptInt, int) *tcpip.Error { return tcpip.ErrInvalidEndpointState } // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. -func (*fakeTransportEndpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) { +func (*fakeTransportEndpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { return -1, tcpip.ErrUnknownProtocolOption } diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index f62fd729f..b172d71b0 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -425,7 +425,7 @@ type Endpoint interface { // SetSockOptInt sets a socket option, for simple cases where a value // has the int type. - SetSockOptInt(opt SockOpt, v int) *Error + SetSockOptInt(opt SockOptInt, v int) *Error // GetSockOpt gets a socket option. opt should be a pointer to one of the // *Option types. @@ -433,7 +433,7 @@ type Endpoint interface { // GetSockOptInt gets a socket option for simple cases where a return // value has the int type. - GetSockOptInt(SockOpt) (int, *Error) + GetSockOptInt(SockOptInt) (int, *Error) // State returns a socket's lifecycle state. The returned value is // protocol-specific and is primarily used for diagnostics. @@ -488,13 +488,13 @@ type WriteOptions struct { Atomic bool } -// SockOpt represents socket options which values have the int type. -type SockOpt int +// SockOptInt represents socket options which values have the int type. +type SockOptInt int const ( // ReceiveQueueSizeOption is used in GetSockOptInt to specify that the // number of unread bytes in the input buffer should be returned. - ReceiveQueueSizeOption SockOpt = iota + ReceiveQueueSizeOption SockOptInt = iota // SendBufferSizeOption is used by SetSockOptInt/GetSockOptInt to // specify the send buffer size option. diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index 9c40931b5..5816ce49a 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -351,12 +351,12 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { } // SetSockOptInt sets a socket option. Currently not supported. -func (e *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error { +func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { return nil } // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. -func (e *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) { +func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { switch opt { case tcpip.ReceiveQueueSizeOption: v := 0 diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go index 0010b5e5f..6360ce880 100644 --- a/pkg/tcpip/transport/packet/endpoint.go +++ b/pkg/tcpip/transport/packet/endpoint.go @@ -251,12 +251,12 @@ func (ep *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { } // SetSockOptInt implements tcpip.Endpoint.SetSockOptInt. -func (ep *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error { +func (ep *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { return tcpip.ErrUnknownProtocolOption } // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. -func (ep *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) { +func (ep *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { return 0, tcpip.ErrNotSupported } diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index 5aafe2615..0fd9c456a 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -510,12 +510,12 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { } // SetSockOptInt implements tcpip.Endpoint.SetSockOptInt. -func (ep *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error { +func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { return tcpip.ErrUnknownProtocolOption } // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. -func (e *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) { +func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { switch opt { case tcpip.ReceiveQueueSizeOption: v := 0 diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index fe629aa40..f79154b95 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -1146,7 +1146,7 @@ func (e *endpoint) zeroReceiveWindow(scale uint8) bool { } // SetSockOptInt sets a socket option. -func (e *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error { +func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { switch opt { case tcpip.ReceiveBufferSizeOption: // Make sure the receive buffer size is within the min and max @@ -1447,7 +1447,7 @@ func (e *endpoint) readyReceiveSize() (int, *tcpip.Error) { } // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. -func (e *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) { +func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { switch opt { case tcpip.ReceiveQueueSizeOption: return e.readyReceiveSize() diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 1ac4705af..dae373ea7 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -457,7 +457,7 @@ func (e *endpoint) Peek([][]byte) (int64, tcpip.ControlMessages, *tcpip.Error) { } // SetSockOptInt implements tcpip.Endpoint.SetSockOptInt. -func (e *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error { +func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { return nil } @@ -661,7 +661,7 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { } // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. -func (e *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) { +func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { switch opt { case tcpip.ReceiveQueueSizeOption: v := 0 -- cgit v1.2.3 From b2a881784c8e525c1fea71c6f23663413d107f05 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Wed, 8 Jan 2020 14:48:47 -0800 Subject: Built dead-simple traversal, but now getting depedency cycle error :'( --- pkg/sentry/socket/netfilter/netfilter.go | 4 +++ pkg/tcpip/iptables/BUILD | 4 ++- pkg/tcpip/iptables/iptables.go | 59 ++++++++++++++++++++++++++++++++ pkg/tcpip/network/ipv4/ipv4.go | 6 ++++ test/iptables/filter_input.go | 2 +- 5 files changed, 73 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index e4c493141..57785220e 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -368,6 +368,10 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { } } + // TODO(gvisor.dev/issue/170): Check the following conditions: + // - There are no loops. + // - There are no chains without an unconditional final rule. + ipt := stack.IPTables() table.SetMetadata(metadata{ HookEntry: replace.HookEntry, diff --git a/pkg/tcpip/iptables/BUILD b/pkg/tcpip/iptables/BUILD index cc5f531e2..6ed7c6da0 100644 --- a/pkg/tcpip/iptables/BUILD +++ b/pkg/tcpip/iptables/BUILD @@ -11,5 +11,7 @@ go_library( ], importpath = "gvisor.dev/gvisor/pkg/tcpip/iptables", visibility = ["//visibility:public"], - deps = ["//pkg/tcpip/buffer"], + deps = [ + "//pkg/tcpip", + ], ) diff --git a/pkg/tcpip/iptables/iptables.go b/pkg/tcpip/iptables/iptables.go index 9e7005374..025a4679d 100644 --- a/pkg/tcpip/iptables/iptables.go +++ b/pkg/tcpip/iptables/iptables.go @@ -16,6 +16,8 @@ // tool. package iptables +import "github.com/google/netstack/tcpip" + const ( TablenameNat = "nat" TablenameMangle = "mangle" @@ -121,3 +123,60 @@ func EmptyFilterTable() Table { UserChains: map[string]int{}, } } + +// Check runs pkt through the rules for hook. It returns true when the packet +// should continue traversing the network stack and false when it should be +// dropped. +func (it *IPTables) Check(hook Hook, pkt tcpip.PacketBuffer) bool { + // TODO(gvisor.dev/issue/170): A lot of this is uncomplicated because + // we're missing features. Jumps, the call stack, etc. aren't checked + // for yet because we're yet to support them. + log.Infof("kevin: iptables.IPTables: checking hook %v", hook) + + // Go through each table containing the hook. + for _, tablename := range it.Priorities[hook] { + verdict := it.checkTable(tablename) + switch verdict { + // TODO: We either got a final verdict or simply continue on. + } + } +} + +func (it *IPTables) checkTable(hook Hook, pkt tcpip.PacketBuffer, tablename string) bool { + log.Infof("kevin: iptables.IPTables: checking table %q", tablename) + table := it.Tables[tablename] + ruleIdx := table.BuiltinChains[hook] + + // Start from ruleIdx and go down until a rule gives us a verdict. + for ruleIdx := table.BuiltinChains[hook]; ruleIdx < len(table.Rules); ruleIdx++ { + verdict := checkRule(hook, pkt, table, ruleIdx) + switch verdict { + case Accept, Drop: + return verdict + case Continue: + continue + case Stolen, Queue, Repeat, None, Jump, Return: + } + } + + panic("Traversed past the entire list of iptables rules.") +} + +func (it *IPTables) checkRule(hook Hook, pkt tcpip.PacketBuffer, table Table, ruleIdx int) Verdict { + rule := table.Rules[ruleIdx] + // Go through each rule matcher. If they all match, run + // the rule target. + for _, matcher := range rule.Matchers { + matches, hotdrop := matcher.Match(hook, pkt, "") + if hotdrop { + return Drop + } + if !matches { + return Continue + } + } + + // All the matchers matched, so run the target. + verdict, _ := rule.Target.Action(pkt) + return verdict +} diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index e645cf62c..bbb5aafee 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -350,6 +350,12 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt tcpip.PacketBuffer) { } pkt.NetworkHeader = headerView[:h.HeaderLength()] + // iptables filtering. + if ok := iptables.Check(iptables.Input, pkt); !ok { + // iptables is telling us to drop the packet. + return + } + hlen := int(h.HeaderLength()) tlen := int(h.TotalLength()) pkt.Data.TrimFront(hlen) diff --git a/test/iptables/filter_input.go b/test/iptables/filter_input.go index 1723a4d3e..7c4d469fa 100644 --- a/test/iptables/filter_input.go +++ b/test/iptables/filter_input.go @@ -138,7 +138,7 @@ func (FilterInputDropAll) ContainerAction(ip net.IP) error { return err } - // Listen for All packets on dropPort. + // Listen for all packets on dropPort. if err := listenUDP(dropPort, sendloopDuration); err == nil { return fmt.Errorf("packets should have been dropped, but got a packet") } else if netErr, ok := err.(net.Error); !ok || !netErr.Timeout() { -- cgit v1.2.3 From d530df2f95c3f75488ecc56b8fd205c3ee0966f8 Mon Sep 17 00:00:00 2001 From: Tamir Duberstein Date: Wed, 8 Jan 2020 15:39:22 -0800 Subject: Introduce tcpip.SockOptBool ...and port V6OnlyOption to it. PiperOrigin-RevId: 288789451 --- pkg/sentry/socket/netstack/netstack.go | 21 ++++-- pkg/sentry/socket/unix/transport/unix.go | 16 +++++ pkg/tcpip/stack/ndp_test.go | 15 ++--- pkg/tcpip/stack/transport_demuxer_test.go | 48 +++++++------- pkg/tcpip/stack/transport_test.go | 10 +++ pkg/tcpip/tcpip.go | 21 ++++-- pkg/tcpip/transport/icmp/endpoint.go | 10 +++ pkg/tcpip/transport/packet/endpoint.go | 22 +++++-- pkg/tcpip/transport/raw/endpoint.go | 40 +++++++----- pkg/tcpip/transport/tcp/dual_stack_test.go | 8 +-- pkg/tcpip/transport/tcp/endpoint.go | 75 ++++++++++++---------- pkg/tcpip/transport/tcp/tcp_test.go | 8 +-- pkg/tcpip/transport/tcp/testing/context/context.go | 6 +- pkg/tcpip/transport/udp/endpoint.go | 60 +++++++++-------- pkg/tcpip/transport/udp/udp_test.go | 2 +- 15 files changed, 224 insertions(+), 138 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 5f91a0d1a..9e0d69046 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -222,6 +222,10 @@ type commonEndpoint interface { // transport.Endpoint.SetSockOpt. SetSockOpt(interface{}) *tcpip.Error + // SetSockOptBool implements tcpip.Endpoint.SetSockOptBool and + // transport.Endpoint.SetSockOptBool. + SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error + // SetSockOptInt implements tcpip.Endpoint.SetSockOptInt and // transport.Endpoint.SetSockOptInt. SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error @@ -230,6 +234,10 @@ type commonEndpoint interface { // transport.Endpoint.GetSockOpt. GetSockOpt(interface{}) *tcpip.Error + // GetSockOptBool implements tcpip.Endpoint.GetSockOptBool and + // transport.Endpoint.GetSockOpt. + GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) + // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt and // transport.Endpoint.GetSockOpt. GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) @@ -1213,12 +1221,15 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf return nil, syserr.ErrInvalidArgument } - var v tcpip.V6OnlyOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptBool(tcpip.V6OnlyOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(v), nil + var o uint32 + if v { + o = 1 + } + return int32(o), nil case linux.IPV6_PATHMTU: t.Kernel().EmitUnimplementedEvent(t) @@ -1621,7 +1632,7 @@ func setSockOptIPv6(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.V6OnlyOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.V6OnlyOption, v != 0)) case linux.IPV6_ADD_MEMBERSHIP, linux.IPV6_DROP_MEMBERSHIP, diff --git a/pkg/sentry/socket/unix/transport/unix.go b/pkg/sentry/socket/unix/transport/unix.go index fcba49435..37c7ac3c1 100644 --- a/pkg/sentry/socket/unix/transport/unix.go +++ b/pkg/sentry/socket/unix/transport/unix.go @@ -175,6 +175,10 @@ type Endpoint interface { // types. SetSockOpt(opt interface{}) *tcpip.Error + // SetSockOptBool sets a socket option for simple cases when a value has + // the int type. + SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error + // SetSockOptInt sets a socket option for simple cases when a value has // the int type. SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error @@ -183,6 +187,10 @@ type Endpoint interface { // tcpip.*Option types. GetSockOpt(opt interface{}) *tcpip.Error + // GetSockOptBool gets a socket option for simple cases when a return + // value has the int type. + GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) + // GetSockOptInt gets a socket option for simple cases when a return // value has the int type. GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) @@ -851,10 +859,18 @@ func (e *baseEndpoint) SetSockOpt(opt interface{}) *tcpip.Error { return nil } +func (e *baseEndpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { + return nil +} + func (e *baseEndpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { return nil } +func (e *baseEndpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { + return false, tcpip.ErrUnknownProtocolOption +} + func (e *baseEndpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { switch opt { case tcpip.ReceiveQueueSizeOption: diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go index 070d80c8d..e51462a55 100644 --- a/pkg/tcpip/stack/ndp_test.go +++ b/pkg/tcpip/stack/ndp_test.go @@ -1701,9 +1701,8 @@ func addrForNewConnection(t *testing.T, s *stack.Stack) tcpip.Address { t.Fatalf("s.NewEndpoint(%d, %d, _): %s", header.UDPProtocolNumber, header.IPv6ProtocolNumber, err) } defer ep.Close() - v := tcpip.V6OnlyOption(1) - if err := ep.SetSockOpt(v); err != nil { - t.Fatalf("SetSockOpt(%+v): %s", v, err) + if err := ep.SetSockOptBool(tcpip.V6OnlyOption, true); err != nil { + t.Fatalf("SetSockOpt(tcpip.V6OnlyOption, true): %s", err) } if err := ep.Connect(dstAddr); err != nil { t.Fatalf("ep.Connect(%+v): %s", dstAddr, err) @@ -1728,9 +1727,8 @@ func addrForNewConnectionWithAddr(t *testing.T, s *stack.Stack, addr tcpip.FullA t.Fatalf("s.NewEndpoint(%d, %d, _): %s", header.UDPProtocolNumber, header.IPv6ProtocolNumber, err) } defer ep.Close() - v := tcpip.V6OnlyOption(1) - if err := ep.SetSockOpt(v); err != nil { - t.Fatalf("SetSockOpt(%+v): %s", v, err) + if err := ep.SetSockOptBool(tcpip.V6OnlyOption, true); err != nil { + t.Fatalf("SetSockOpt(tcpip.V6OnlyOption, true): %s", err) } if err := ep.Bind(addr); err != nil { t.Fatalf("ep.Bind(%+v): %s", addr, err) @@ -2066,9 +2064,8 @@ func TestAutoGenAddrTimerDeprecation(t *testing.T) { t.Fatalf("s.NewEndpoint(%d, %d, _): %s", header.UDPProtocolNumber, header.IPv6ProtocolNumber, err) } defer ep.Close() - v := tcpip.V6OnlyOption(1) - if err := ep.SetSockOpt(v); err != nil { - t.Fatalf("SetSockOpt(%+v): %s", v, err) + if err := ep.SetSockOptBool(tcpip.V6OnlyOption, true); err != nil { + t.Fatalf("SetSockOpt(tcpip.V6OnlyOption, true): %s", err) } if err := ep.Connect(dstAddr); err != tcpip.ErrNoRoute { diff --git a/pkg/tcpip/stack/transport_demuxer_test.go b/pkg/tcpip/stack/transport_demuxer_test.go index 33dbc0536..df5ced887 100644 --- a/pkg/tcpip/stack/transport_demuxer_test.go +++ b/pkg/tcpip/stack/transport_demuxer_test.go @@ -61,11 +61,7 @@ func (c *testContext) createV6Endpoint(v6only bool) { c.t.Fatalf("NewEndpoint failed: %v", err) } - var v tcpip.V6OnlyOption - if v6only { - v = 1 - } - if err := c.ep.SetSockOpt(v); err != nil { + if err := c.ep.SetSockOptBool(tcpip.V6OnlyOption, v6only); err != nil { c.t.Fatalf("SetSockOpt failed: %v", err) } } @@ -201,54 +197,54 @@ func TestDistribution(t *testing.T) { "BindPortReuse", // 5 endpoints that all have reuse set. []endpointSockopts{ - endpointSockopts{1, ""}, - endpointSockopts{1, ""}, - endpointSockopts{1, ""}, - endpointSockopts{1, ""}, - endpointSockopts{1, ""}, + {1, ""}, + {1, ""}, + {1, ""}, + {1, ""}, + {1, ""}, }, map[string][]float64{ // Injected packets on dev0 get distributed evenly. - "dev0": []float64{0.2, 0.2, 0.2, 0.2, 0.2}, + "dev0": {0.2, 0.2, 0.2, 0.2, 0.2}, }, }, { "BindToDevice", // 3 endpoints with various bindings. []endpointSockopts{ - endpointSockopts{0, "dev0"}, - endpointSockopts{0, "dev1"}, - endpointSockopts{0, "dev2"}, + {0, "dev0"}, + {0, "dev1"}, + {0, "dev2"}, }, map[string][]float64{ // Injected packets on dev0 go only to the endpoint bound to dev0. - "dev0": []float64{1, 0, 0}, + "dev0": {1, 0, 0}, // Injected packets on dev1 go only to the endpoint bound to dev1. - "dev1": []float64{0, 1, 0}, + "dev1": {0, 1, 0}, // Injected packets on dev2 go only to the endpoint bound to dev2. - "dev2": []float64{0, 0, 1}, + "dev2": {0, 0, 1}, }, }, { "ReuseAndBindToDevice", // 6 endpoints with various bindings. []endpointSockopts{ - endpointSockopts{1, "dev0"}, - endpointSockopts{1, "dev0"}, - endpointSockopts{1, "dev1"}, - endpointSockopts{1, "dev1"}, - endpointSockopts{1, "dev1"}, - endpointSockopts{1, ""}, + {1, "dev0"}, + {1, "dev0"}, + {1, "dev1"}, + {1, "dev1"}, + {1, "dev1"}, + {1, ""}, }, map[string][]float64{ // Injected packets on dev0 get distributed among endpoints bound to // dev0. - "dev0": []float64{0.5, 0.5, 0, 0, 0, 0}, + "dev0": {0.5, 0.5, 0, 0, 0, 0}, // Injected packets on dev1 get distributed among endpoints bound to // dev1 or unbound. - "dev1": []float64{0, 0, 1. / 3, 1. / 3, 1. / 3, 0}, + "dev1": {0, 0, 1. / 3, 1. / 3, 1. / 3, 0}, // Injected packets on dev999 go only to the unbound. - "dev999": []float64{0, 0, 0, 0, 0, 1}, + "dev999": {0, 0, 0, 0, 0, 1}, }, }, } { diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go index 095346f0b..f50604a8a 100644 --- a/pkg/tcpip/stack/transport_test.go +++ b/pkg/tcpip/stack/transport_test.go @@ -102,11 +102,21 @@ func (*fakeTransportEndpoint) SetSockOpt(interface{}) *tcpip.Error { return tcpip.ErrInvalidEndpointState } +// SetSockOptBool sets a socket option. Currently not supported. +func (*fakeTransportEndpoint) SetSockOptBool(tcpip.SockOptBool, bool) *tcpip.Error { + return tcpip.ErrInvalidEndpointState +} + // SetSockOptInt sets a socket option. Currently not supported. func (*fakeTransportEndpoint) SetSockOptInt(tcpip.SockOptInt, int) *tcpip.Error { return tcpip.ErrInvalidEndpointState } +// GetSockOptBool implements tcpip.Endpoint.GetSockOptBool. +func (*fakeTransportEndpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { + return false, tcpip.ErrUnknownProtocolOption +} + // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. func (*fakeTransportEndpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { return -1, tcpip.ErrUnknownProtocolOption diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index b172d71b0..1eca76c30 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -423,6 +423,10 @@ type Endpoint interface { // SetSockOpt sets a socket option. opt should be one of the *Option types. SetSockOpt(opt interface{}) *Error + // SetSockOptBool sets a socket option, for simple cases where a value + // has the bool type. + SetSockOptBool(opt SockOptBool, v bool) *Error + // SetSockOptInt sets a socket option, for simple cases where a value // has the int type. SetSockOptInt(opt SockOptInt, v int) *Error @@ -431,6 +435,10 @@ type Endpoint interface { // *Option types. GetSockOpt(opt interface{}) *Error + // GetSockOptBool gets a socket option for simple cases where a return + // value has the bool type. + GetSockOptBool(SockOptBool) (bool, *Error) + // GetSockOptInt gets a socket option for simple cases where a return // value has the int type. GetSockOptInt(SockOptInt) (int, *Error) @@ -488,6 +496,15 @@ type WriteOptions struct { Atomic bool } +// SockOptBool represents socket options which values have the bool type. +type SockOptBool int + +const ( + // V6OnlyOption is used by {G,S}etSockOptBool to specify whether an IPv6 + // socket is to be restricted to sending and receiving IPv6 packets only. + V6OnlyOption SockOptBool = iota +) + // SockOptInt represents socket options which values have the int type. type SockOptInt int @@ -521,10 +538,6 @@ const ( // the endpoint should be cleared and returned. type ErrorOption struct{} -// V6OnlyOption is used by SetSockOpt/GetSockOpt to specify whether an IPv6 -// socket is to be restricted to sending and receiving IPv6 packets only. -type V6OnlyOption int - // CorkOption is used by SetSockOpt/GetSockOpt to specify if data should be // held until segments are full by the TCP transport protocol. type CorkOption int diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index 5816ce49a..c7ce74cdd 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -350,11 +350,21 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { return nil } +// SetSockOptBool sets a socket option. Currently not supported. +func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { + return nil +} + // SetSockOptInt sets a socket option. Currently not supported. func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { return nil } +// GetSockOptBool implements tcpip.Endpoint.GetSockOptBool. +func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { + return false, tcpip.ErrUnknownProtocolOption +} + // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { switch opt { diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go index 6360ce880..07ffa8aba 100644 --- a/pkg/tcpip/transport/packet/endpoint.go +++ b/pkg/tcpip/transport/packet/endpoint.go @@ -247,17 +247,17 @@ func (ep *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { // used with SetSockOpt, and this function always returns // tcpip.ErrNotSupported. func (ep *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { - return tcpip.ErrNotSupported + return tcpip.ErrUnknownProtocolOption } -// SetSockOptInt implements tcpip.Endpoint.SetSockOptInt. -func (ep *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { +// SetSockOptBool implements tcpip.Endpoint.SetSockOptBool. +func (ep *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { return tcpip.ErrUnknownProtocolOption } -// GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. -func (ep *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { - return 0, tcpip.ErrNotSupported +// SetSockOptInt implements tcpip.Endpoint.SetSockOptInt. +func (ep *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { + return tcpip.ErrUnknownProtocolOption } // GetSockOpt implements tcpip.Endpoint.GetSockOpt. @@ -265,6 +265,16 @@ func (ep *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { return tcpip.ErrNotSupported } +// GetSockOptBool implements tcpip.Endpoint.GetSockOptBool. +func (ep *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { + return false, tcpip.ErrNotSupported +} + +// GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. +func (ep *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { + return 0, tcpip.ErrNotSupported +} + // HandlePacket implements stack.PacketEndpoint.HandlePacket. func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { ep.rcvMu.Lock() diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index 0fd9c456a..85f7eb76b 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -509,11 +509,36 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { return tcpip.ErrUnknownProtocolOption } +// SetSockOptBool implements tcpip.Endpoint.SetSockOptBool. +func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { + return tcpip.ErrUnknownProtocolOption +} + // SetSockOptInt implements tcpip.Endpoint.SetSockOptInt. func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { return tcpip.ErrUnknownProtocolOption } +// GetSockOpt implements tcpip.Endpoint.GetSockOpt. +func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { + switch o := opt.(type) { + case tcpip.ErrorOption: + return nil + + case *tcpip.KeepaliveEnabledOption: + *o = 0 + return nil + + default: + return tcpip.ErrUnknownProtocolOption + } +} + +// GetSockOptBool implements tcpip.Endpoint.GetSockOptBool. +func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { + return false, tcpip.ErrUnknownProtocolOption +} + // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { switch opt { @@ -544,21 +569,6 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { return -1, tcpip.ErrUnknownProtocolOption } -// GetSockOpt implements tcpip.Endpoint.GetSockOpt. -func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { - switch o := opt.(type) { - case tcpip.ErrorOption: - return nil - - case *tcpip.KeepaliveEnabledOption: - *o = 0 - return nil - - default: - return tcpip.ErrUnknownProtocolOption - } -} - // HandlePacket implements stack.RawTransportEndpoint.HandlePacket. func (e *endpoint) HandlePacket(route *stack.Route, pkt tcpip.PacketBuffer) { e.rcvMu.Lock() diff --git a/pkg/tcpip/transport/tcp/dual_stack_test.go b/pkg/tcpip/transport/tcp/dual_stack_test.go index dfaa4a559..4f361b226 100644 --- a/pkg/tcpip/transport/tcp/dual_stack_test.go +++ b/pkg/tcpip/transport/tcp/dual_stack_test.go @@ -391,9 +391,8 @@ func testV4Accept(t *testing.T, c *context.Context) { // Make sure we get the same error when calling the original ep and the // new one. This validates that v4-mapped endpoints are still able to // query the V6Only flag, whereas pure v4 endpoints are not. - var v tcpip.V6OnlyOption - expected := c.EP.GetSockOpt(&v) - if err := nep.GetSockOpt(&v); err != expected { + _, expected := c.EP.GetSockOptBool(tcpip.V6OnlyOption) + if _, err := nep.GetSockOptBool(tcpip.V6OnlyOption); err != expected { t.Fatalf("GetSockOpt returned unexpected value: got %v, want %v", err, expected) } @@ -531,8 +530,7 @@ func TestV6AcceptOnV6(t *testing.T) { // Make sure we can still query the v6 only status of the new endpoint, // that is, that it is in fact a v6 socket. - var v tcpip.V6OnlyOption - if err := nep.GetSockOpt(&v); err != nil { + if _, err := nep.GetSockOptBool(tcpip.V6OnlyOption); err != nil { t.Fatalf("GetSockOpt failed failed: %v", err) } diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index f79154b95..2ac1b6877 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -1145,6 +1145,29 @@ func (e *endpoint) zeroReceiveWindow(scale uint8) bool { return ((e.rcvBufSize - e.rcvBufUsed) >> scale) == 0 } +// SetSockOptBool sets a socket option. +func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { + switch opt { + case tcpip.V6OnlyOption: + // We only recognize this option on v6 endpoints. + if e.NetProto != header.IPv6ProtocolNumber { + return tcpip.ErrInvalidEndpointState + } + + e.mu.Lock() + defer e.mu.Unlock() + + // We only allow this to be set when we're in the initial state. + if e.state != StateInitial { + return tcpip.ErrInvalidEndpointState + } + + e.v6only = v + } + + return nil +} + // SetSockOptInt sets a socket option. func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { switch opt { @@ -1289,23 +1312,6 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.notifyProtocolGoroutine(notifyMSSChanged) return nil - case tcpip.V6OnlyOption: - // We only recognize this option on v6 endpoints. - if e.NetProto != header.IPv6ProtocolNumber { - return tcpip.ErrInvalidEndpointState - } - - e.mu.Lock() - defer e.mu.Unlock() - - // We only allow this to be set when we're in the initial state. - if e.state != StateInitial { - return tcpip.ErrInvalidEndpointState - } - - e.v6only = v != 0 - return nil - case tcpip.TTLOption: e.mu.Lock() e.ttl = uint8(v) @@ -1446,6 +1452,25 @@ func (e *endpoint) readyReceiveSize() (int, *tcpip.Error) { return e.rcvBufUsed, nil } +// GetSockOptBool implements tcpip.Endpoint.GetSockOptBool. +func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { + switch opt { + case tcpip.V6OnlyOption: + // We only recognize this option on v6 endpoints. + if e.NetProto != header.IPv6ProtocolNumber { + return false, tcpip.ErrUnknownProtocolOption + } + + e.mu.Lock() + v := e.v6only + e.mu.Unlock() + + return v, nil + } + + return false, tcpip.ErrUnknownProtocolOption +} + // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { switch opt { @@ -1540,22 +1565,6 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { } return nil - case *tcpip.V6OnlyOption: - // We only recognize this option on v6 endpoints. - if e.NetProto != header.IPv6ProtocolNumber { - return tcpip.ErrUnknownProtocolOption - } - - e.mu.Lock() - v := e.v6only - e.mu.Unlock() - - *o = 0 - if v { - *o = 1 - } - return nil - case *tcpip.TTLOption: e.mu.Lock() *o = tcpip.TTLOption(e.ttl) diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index 9d7b0910d..15745ebd4 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -4028,12 +4028,12 @@ func TestConnectAvoidsBoundPorts(t *testing.T) { switch network { case "ipv4": case "ipv6": - if err := ep.SetSockOpt(tcpip.V6OnlyOption(1)); err != nil { - t.Fatalf("SetSockOpt(V6OnlyOption(1)) failed: %v", err) + if err := ep.SetSockOptBool(tcpip.V6OnlyOption, true); err != nil { + t.Fatalf("SetSockOpt(V6OnlyOption(true)) failed: %v", err) } case "dual": - if err := ep.SetSockOpt(tcpip.V6OnlyOption(0)); err != nil { - t.Fatalf("SetSockOpt(V6OnlyOption(0)) failed: %v", err) + if err := ep.SetSockOptBool(tcpip.V6OnlyOption, false); err != nil { + t.Fatalf("SetSockOpt(V6OnlyOption(false)) failed: %v", err) } default: t.Fatalf("unknown network: '%s'", network) diff --git a/pkg/tcpip/transport/tcp/testing/context/context.go b/pkg/tcpip/transport/tcp/testing/context/context.go index 50c81aa65..822907998 100644 --- a/pkg/tcpip/transport/tcp/testing/context/context.go +++ b/pkg/tcpip/transport/tcp/testing/context/context.go @@ -475,11 +475,7 @@ func (c *Context) CreateV6Endpoint(v6only bool) { c.t.Fatalf("NewEndpoint failed: %v", err) } - var v tcpip.V6OnlyOption - if v6only { - v = 1 - } - if err := c.EP.SetSockOpt(v); err != nil { + if err := c.EP.SetSockOptBool(tcpip.V6OnlyOption, v6only); err != nil { c.t.Fatalf("SetSockOpt failed failed: %v", err) } } diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index dae373ea7..1a5ee6317 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -456,14 +456,9 @@ func (e *endpoint) Peek([][]byte) (int64, tcpip.ControlMessages, *tcpip.Error) { return 0, tcpip.ControlMessages{}, nil } -// SetSockOptInt implements tcpip.Endpoint.SetSockOptInt. -func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { - return nil -} - -// SetSockOpt implements tcpip.Endpoint.SetSockOpt. -func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { - switch v := opt.(type) { +// SetSockOptBool implements tcpip.Endpoint.SetSockOptBool. +func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { + switch opt { case tcpip.V6OnlyOption: // We only recognize this option on v6 endpoints. if e.NetProto != header.IPv6ProtocolNumber { @@ -478,8 +473,20 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { return tcpip.ErrInvalidEndpointState } - e.v6only = v != 0 + e.v6only = v + } + + return nil +} +// SetSockOptInt implements tcpip.Endpoint.SetSockOptInt. +func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { + return nil +} + +// SetSockOpt implements tcpip.Endpoint.SetSockOpt. +func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { + switch v := opt.(type) { case tcpip.TTLOption: e.mu.Lock() e.ttl = uint8(v) @@ -660,6 +667,25 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { return nil } +// GetSockOptBool implements tcpip.Endpoint.GetSockOptBool. +func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { + switch opt { + case tcpip.V6OnlyOption: + // We only recognize this option on v6 endpoints. + if e.NetProto != header.IPv6ProtocolNumber { + return false, tcpip.ErrUnknownProtocolOption + } + + e.mu.Lock() + v := e.v6only + e.mu.Unlock() + + return v, nil + } + + return false, tcpip.ErrUnknownProtocolOption +} + // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { switch opt { @@ -695,22 +721,6 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { case tcpip.ErrorOption: return nil - case *tcpip.V6OnlyOption: - // We only recognize this option on v6 endpoints. - if e.NetProto != header.IPv6ProtocolNumber { - return tcpip.ErrUnknownProtocolOption - } - - e.mu.Lock() - v := e.v6only - e.mu.Unlock() - - *o = 0 - if v { - *o = 1 - } - return nil - case *tcpip.TTLOption: e.mu.Lock() *o = tcpip.TTLOption(e.ttl) diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index 65382b7f1..149fff999 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -335,7 +335,7 @@ func (c *testContext) createEndpointForFlow(flow testFlow) { c.createEndpoint(flow.sockProto()) if flow.isV6Only() { - if err := c.ep.SetSockOpt(tcpip.V6OnlyOption(1)); err != nil { + if err := c.ep.SetSockOptBool(tcpip.V6OnlyOption, true); err != nil { c.t.Fatalf("SetSockOpt failed: %v", err) } } else if flow.isBroadcast() { -- cgit v1.2.3 From 0999ae8b34d83a4b2ea8342d0459c8131c35d6e1 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Wed, 8 Jan 2020 15:57:25 -0800 Subject: Getting a panic when running tests. For some reason the filter table is ending up with the wrong chains and is indexing -1 into rules. --- pkg/sentry/socket/netfilter/netfilter.go | 17 ++++++---------- pkg/sentry/socket/netstack/netstack.go | 12 +++++++++-- pkg/tcpip/BUILD | 1 - pkg/tcpip/iptables/BUILD | 1 + pkg/tcpip/iptables/iptables.go | 35 +++++++++++++++++++++++++------- pkg/tcpip/iptables/targets.go | 8 ++++---- pkg/tcpip/iptables/types.go | 8 +++----- pkg/tcpip/network/arp/arp.go | 2 +- pkg/tcpip/network/ipv4/BUILD | 1 + pkg/tcpip/network/ipv4/ipv4.go | 8 ++++++-- pkg/tcpip/network/ipv6/ipv6.go | 2 +- pkg/tcpip/stack/nic.go | 2 +- pkg/tcpip/stack/registration.go | 2 +- pkg/tcpip/tcpip.go | 4 ---- 14 files changed, 63 insertions(+), 40 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 57785220e..3a857ef6d 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -25,7 +25,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserr" - "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/iptables" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -45,7 +44,7 @@ type metadata struct { } // GetInfo returns information about iptables. -func GetInfo(t *kernel.Task, ep tcpip.Endpoint, outPtr usermem.Addr) (linux.IPTGetinfo, *syserr.Error) { +func GetInfo(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr) (linux.IPTGetinfo, *syserr.Error) { // Read in the struct and table name. var info linux.IPTGetinfo if _, err := t.CopyIn(outPtr, &info); err != nil { @@ -53,7 +52,7 @@ func GetInfo(t *kernel.Task, ep tcpip.Endpoint, outPtr usermem.Addr) (linux.IPTG } // Find the appropriate table. - table, err := findTable(ep, info.Name.String()) + table, err := findTable(stack, info.Name.String()) if err != nil { return linux.IPTGetinfo{}, err } @@ -76,7 +75,7 @@ func GetInfo(t *kernel.Task, ep tcpip.Endpoint, outPtr usermem.Addr) (linux.IPTG } // GetEntries returns netstack's iptables rules encoded for the iptables tool. -func GetEntries(t *kernel.Task, ep tcpip.Endpoint, outPtr usermem.Addr, outLen int) (linux.KernelIPTGetEntries, *syserr.Error) { +func GetEntries(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr, outLen int) (linux.KernelIPTGetEntries, *syserr.Error) { // Read in the struct and table name. var userEntries linux.IPTGetEntries if _, err := t.CopyIn(outPtr, &userEntries); err != nil { @@ -84,7 +83,7 @@ func GetEntries(t *kernel.Task, ep tcpip.Endpoint, outPtr usermem.Addr, outLen i } // Find the appropriate table. - table, err := findTable(ep, userEntries.Name.String()) + table, err := findTable(stack, userEntries.Name.String()) if err != nil { return linux.KernelIPTGetEntries{}, err } @@ -103,12 +102,8 @@ func GetEntries(t *kernel.Task, ep tcpip.Endpoint, outPtr usermem.Addr, outLen i return entries, nil } -func findTable(ep tcpip.Endpoint, tableName string) (iptables.Table, *syserr.Error) { - ipt, err := ep.IPTables() - if err != nil { - return iptables.Table{}, syserr.FromError(err) - } - table, ok := ipt.Tables[tableName] +func findTable(stack *stack.Stack, tableName string) (iptables.Table, *syserr.Error) { + table, ok := stack.IPTables().Tables[tableName] if !ok { return iptables.Table{}, syserr.ErrInvalidArgument } diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 8c07eef4b..86a8104df 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -826,7 +826,11 @@ func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr us return nil, syserr.ErrInvalidArgument } - info, err := netfilter.GetInfo(t, s.Endpoint, outPtr) + stack := inet.StackFromContext(t) + if stack == nil { + return nil, syserr.ErrNoDevice + } + info, err := netfilter.GetInfo(t, stack.(*Stack).Stack, outPtr) if err != nil { return nil, err } @@ -837,7 +841,11 @@ func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr us return nil, syserr.ErrInvalidArgument } - entries, err := netfilter.GetEntries(t, s.Endpoint, outPtr, outLen) + stack := inet.StackFromContext(t) + if stack == nil { + return nil, syserr.ErrNoDevice + } + entries, err := netfilter.GetEntries(t, stack.(*Stack).Stack, outPtr, outLen) if err != nil { return nil, err } diff --git a/pkg/tcpip/BUILD b/pkg/tcpip/BUILD index 65d4d0cd8..36bc3a63b 100644 --- a/pkg/tcpip/BUILD +++ b/pkg/tcpip/BUILD @@ -15,7 +15,6 @@ go_library( visibility = ["//visibility:public"], deps = [ "//pkg/tcpip/buffer", - "//pkg/tcpip/iptables", "//pkg/waiter", ], ) diff --git a/pkg/tcpip/iptables/BUILD b/pkg/tcpip/iptables/BUILD index 6ed7c6da0..2893c80cd 100644 --- a/pkg/tcpip/iptables/BUILD +++ b/pkg/tcpip/iptables/BUILD @@ -12,6 +12,7 @@ go_library( importpath = "gvisor.dev/gvisor/pkg/tcpip/iptables", visibility = ["//visibility:public"], deps = [ + "//pkg/log", "//pkg/tcpip", ], ) diff --git a/pkg/tcpip/iptables/iptables.go b/pkg/tcpip/iptables/iptables.go index 025a4679d..aff8a680b 100644 --- a/pkg/tcpip/iptables/iptables.go +++ b/pkg/tcpip/iptables/iptables.go @@ -16,7 +16,12 @@ // tool. package iptables -import "github.com/google/netstack/tcpip" +import ( + "fmt" + + "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/tcpip" +) const ( TablenameNat = "nat" @@ -135,31 +140,47 @@ func (it *IPTables) Check(hook Hook, pkt tcpip.PacketBuffer) bool { // Go through each table containing the hook. for _, tablename := range it.Priorities[hook] { - verdict := it.checkTable(tablename) + verdict := it.checkTable(hook, pkt, tablename) switch verdict { - // TODO: We either got a final verdict or simply continue on. + // If the table returns Accept, move on to the next table. + case Accept: + continue + // The Drop verdict is final. + case Drop: + log.Infof("kevin: Packet dropped") + return false + case Stolen, Queue, Repeat, None, Jump, Return, Continue: + panic(fmt.Sprintf("Unimplemented verdict %v.", verdict)) } } + + // Every table returned Accept. + log.Infof("kevin: Packet accepted") + return true } -func (it *IPTables) checkTable(hook Hook, pkt tcpip.PacketBuffer, tablename string) bool { +func (it *IPTables) checkTable(hook Hook, pkt tcpip.PacketBuffer, tablename string) Verdict { log.Infof("kevin: iptables.IPTables: checking table %q", tablename) table := it.Tables[tablename] - ruleIdx := table.BuiltinChains[hook] + log.Infof("kevin: iptables.IPTables: table %+v", table) // Start from ruleIdx and go down until a rule gives us a verdict. for ruleIdx := table.BuiltinChains[hook]; ruleIdx < len(table.Rules); ruleIdx++ { - verdict := checkRule(hook, pkt, table, ruleIdx) + verdict := it.checkRule(hook, pkt, table, ruleIdx) switch verdict { + // For either of these cases, this table is done with the + // packet. case Accept, Drop: return verdict + // Continue traversing the rules of the table. case Continue: continue case Stolen, Queue, Repeat, None, Jump, Return: + panic(fmt.Sprintf("Unimplemented verdict %v.", verdict)) } } - panic("Traversed past the entire list of iptables rules.") + panic(fmt.Sprintf("Traversed past the entire list of iptables rules in table %q.", tablename)) } func (it *IPTables) checkRule(hook Hook, pkt tcpip.PacketBuffer, table Table, ruleIdx int) Verdict { diff --git a/pkg/tcpip/iptables/targets.go b/pkg/tcpip/iptables/targets.go index 2c3598e3d..cb3ac1aff 100644 --- a/pkg/tcpip/iptables/targets.go +++ b/pkg/tcpip/iptables/targets.go @@ -16,13 +16,13 @@ package iptables -import "gvisor.dev/gvisor/pkg/tcpip/buffer" +import "gvisor.dev/gvisor/pkg/tcpip" // UnconditionalAcceptTarget accepts all packets. type UnconditionalAcceptTarget struct{} // Action implements Target.Action. -func (UnconditionalAcceptTarget) Action(packet buffer.VectorisedView) (Verdict, string) { +func (UnconditionalAcceptTarget) Action(packet tcpip.PacketBuffer) (Verdict, string) { return Accept, "" } @@ -30,7 +30,7 @@ func (UnconditionalAcceptTarget) Action(packet buffer.VectorisedView) (Verdict, type UnconditionalDropTarget struct{} // Action implements Target.Action. -func (UnconditionalDropTarget) Action(packet buffer.VectorisedView) (Verdict, string) { +func (UnconditionalDropTarget) Action(packet tcpip.PacketBuffer) (Verdict, string) { return Drop, "" } @@ -38,6 +38,6 @@ func (UnconditionalDropTarget) Action(packet buffer.VectorisedView) (Verdict, st type PanicTarget struct{} // Actions implements Target.Action. -func (PanicTarget) Action(packet buffer.VectorisedView) (Verdict, string) { +func (PanicTarget) Action(packet tcpip.PacketBuffer) (Verdict, string) { panic("PanicTarget triggered.") } diff --git a/pkg/tcpip/iptables/types.go b/pkg/tcpip/iptables/types.go index 540f8c0b4..9f6906100 100644 --- a/pkg/tcpip/iptables/types.go +++ b/pkg/tcpip/iptables/types.go @@ -14,9 +14,7 @@ package iptables -import ( - "gvisor.dev/gvisor/pkg/tcpip/buffer" -) +import "gvisor.dev/gvisor/pkg/tcpip" // A Hook specifies one of the hooks built into the network stack. // @@ -165,7 +163,7 @@ type Matcher interface { // Match returns whether the packet matches and whether the packet // should be "hotdropped", i.e. dropped immediately. This is usually // used for suspicious packets. - Match(hook Hook, packet buffer.VectorisedView, interfaceName string) (matches bool, hotdrop bool) + Match(hook Hook, packet tcpip.PacketBuffer, interfaceName string) (matches bool, hotdrop bool) } // A Target is the interface for taking an action for a packet. @@ -173,5 +171,5 @@ type Target interface { // Action takes an action on the packet and returns a verdict on how // traversal should (or should not) continue. If the return value is // Jump, it also returns the name of the chain to jump to. - Action(packet buffer.VectorisedView) (Verdict, string) + Action(packet tcpip.PacketBuffer) (Verdict, string) } diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index da8482509..d88119f68 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -137,7 +137,7 @@ func (*protocol) ParseAddresses(v buffer.View) (src, dst tcpip.Address) { return tcpip.Address(h.ProtocolAddressSender()), ProtocolAddress } -func (p *protocol) NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, sender stack.LinkEndpoint) (stack.NetworkEndpoint, *tcpip.Error) { +func (p *protocol) NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, sender stack.LinkEndpoint, st *stack.Stack) (stack.NetworkEndpoint, *tcpip.Error) { if addrWithPrefix.Address != ProtocolAddress { return nil, tcpip.ErrBadLocalAddress } diff --git a/pkg/tcpip/network/ipv4/BUILD b/pkg/tcpip/network/ipv4/BUILD index aeddfcdd4..4e2aae9a3 100644 --- a/pkg/tcpip/network/ipv4/BUILD +++ b/pkg/tcpip/network/ipv4/BUILD @@ -15,6 +15,7 @@ go_library( "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/header", + "//pkg/tcpip/iptables", "//pkg/tcpip/network/fragmentation", "//pkg/tcpip/network/hash", "//pkg/tcpip/stack", diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index bbb5aafee..f856081e6 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -26,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/iptables" "gvisor.dev/gvisor/pkg/tcpip/network/fragmentation" "gvisor.dev/gvisor/pkg/tcpip/network/hash" "gvisor.dev/gvisor/pkg/tcpip/stack" @@ -54,10 +55,11 @@ type endpoint struct { dispatcher stack.TransportDispatcher fragmentation *fragmentation.Fragmentation protocol *protocol + stack *stack.Stack } // NewEndpoint creates a new ipv4 endpoint. -func (p *protocol) NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, linkEP stack.LinkEndpoint) (stack.NetworkEndpoint, *tcpip.Error) { +func (p *protocol) NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, linkEP stack.LinkEndpoint, st *stack.Stack) (stack.NetworkEndpoint, *tcpip.Error) { e := &endpoint{ nicID: nicID, id: stack.NetworkEndpointID{LocalAddress: addrWithPrefix.Address}, @@ -66,6 +68,7 @@ func (p *protocol) NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWi dispatcher: dispatcher, fragmentation: fragmentation.NewFragmentation(fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, fragmentation.DefaultReassembleTimeout), protocol: p, + stack: st, } return e, nil @@ -351,7 +354,8 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt tcpip.PacketBuffer) { pkt.NetworkHeader = headerView[:h.HeaderLength()] // iptables filtering. - if ok := iptables.Check(iptables.Input, pkt); !ok { + ipt := e.stack.IPTables() + if ok := ipt.Check(iptables.Input, pkt); !ok { // iptables is telling us to drop the packet. return } diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index e13f1fabf..4c940e9e5 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -221,7 +221,7 @@ func (*protocol) ParseAddresses(v buffer.View) (src, dst tcpip.Address) { } // NewEndpoint creates a new ipv6 endpoint. -func (p *protocol) NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, linkEP stack.LinkEndpoint) (stack.NetworkEndpoint, *tcpip.Error) { +func (p *protocol) NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, linkEP stack.LinkEndpoint, st *stack.Stack) (stack.NetworkEndpoint, *tcpip.Error) { return &endpoint{ nicID: nicID, id: stack.NetworkEndpointID{LocalAddress: addrWithPrefix.Address}, diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 4144d5d0f..f2d338bd1 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -467,7 +467,7 @@ func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb Primar } // Create the new network endpoint. - ep, err := netProto.NewEndpoint(n.id, protocolAddress.AddressWithPrefix, n.stack, n, n.linkEP) + ep, err := netProto.NewEndpoint(n.id, protocolAddress.AddressWithPrefix, n.stack, n, n.linkEP, n.stack) if err != nil { return nil, err } diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go index 61fd46d66..754323e82 100644 --- a/pkg/tcpip/stack/registration.go +++ b/pkg/tcpip/stack/registration.go @@ -282,7 +282,7 @@ type NetworkProtocol interface { ParseAddresses(v buffer.View) (src, dst tcpip.Address) // NewEndpoint creates a new endpoint of this protocol. - NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache LinkAddressCache, dispatcher TransportDispatcher, sender LinkEndpoint) (NetworkEndpoint, *tcpip.Error) + NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache LinkAddressCache, dispatcher TransportDispatcher, sender LinkEndpoint, st *Stack) (NetworkEndpoint, *tcpip.Error) // SetOption allows enabling/disabling protocol specific features. // SetOption returns an error if the option is not supported or the diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index f62fd729f..d02950c7a 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -40,7 +40,6 @@ import ( "time" "gvisor.dev/gvisor/pkg/tcpip/buffer" - "gvisor.dev/gvisor/pkg/tcpip/iptables" "gvisor.dev/gvisor/pkg/waiter" ) @@ -446,9 +445,6 @@ type Endpoint interface { // NOTE: This method is a no-op for sockets other than TCP. ModerateRecvBuf(copied int) - // IPTables returns the iptables for this endpoint's stack. - IPTables() (iptables.IPTables, error) - // Info returns a copy to the transport endpoint info. Info() EndpointInfo -- cgit v1.2.3 From b3ae8a62cfdf13821d35467d4150ed983ac556f1 Mon Sep 17 00:00:00 2001 From: Ting-Yu Wang Date: Wed, 8 Jan 2020 16:29:12 -0800 Subject: Fix slice bounds out of range panic in parsing socket control message. Panic found by syzakller. PiperOrigin-RevId: 288799046 --- pkg/sentry/socket/control/control.go | 6 ++++++ test/syscalls/linux/socket_ip_unbound.cc | 33 ++++++++++++++++++++++++++++++++ 2 files changed, 39 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/control/control.go b/pkg/sentry/socket/control/control.go index af1a4e95f..4301b697c 100644 --- a/pkg/sentry/socket/control/control.go +++ b/pkg/sentry/socket/control/control.go @@ -471,6 +471,9 @@ func Parse(t *kernel.Task, socketOrEndpoint interface{}, buf []byte) (socket.Con case linux.SOL_IP: switch h.Type { case linux.IP_TOS: + if length < linux.SizeOfControlMessageTOS { + return socket.ControlMessages{}, syserror.EINVAL + } cmsgs.IP.HasTOS = true binary.Unmarshal(buf[i:i+linux.SizeOfControlMessageTOS], usermem.ByteOrder, &cmsgs.IP.TOS) i += AlignUp(length, width) @@ -481,6 +484,9 @@ func Parse(t *kernel.Task, socketOrEndpoint interface{}, buf []byte) (socket.Con case linux.SOL_IPV6: switch h.Type { case linux.IPV6_TCLASS: + if length < linux.SizeOfControlMessageTClass { + return socket.ControlMessages{}, syserror.EINVAL + } cmsgs.IP.HasTClass = true binary.Unmarshal(buf[i:i+linux.SizeOfControlMessageTClass], usermem.ByteOrder, &cmsgs.IP.TClass) i += AlignUp(length, width) diff --git a/test/syscalls/linux/socket_ip_unbound.cc b/test/syscalls/linux/socket_ip_unbound.cc index b6754111f..ca597e267 100644 --- a/test/syscalls/linux/socket_ip_unbound.cc +++ b/test/syscalls/linux/socket_ip_unbound.cc @@ -129,6 +129,7 @@ TEST_P(IPUnboundSocketTest, InvalidNegativeTtl) { struct TOSOption { int level; int option; + int cmsg_level; }; constexpr int INET_ECN_MASK = 3; @@ -139,10 +140,12 @@ static TOSOption GetTOSOption(int domain) { case AF_INET: opt.level = IPPROTO_IP; opt.option = IP_TOS; + opt.cmsg_level = SOL_IP; break; case AF_INET6: opt.level = IPPROTO_IPV6; opt.option = IPV6_TCLASS; + opt.cmsg_level = SOL_IPV6; break; } return opt; @@ -386,6 +389,36 @@ TEST_P(IPUnboundSocketTest, NullTOS) { SyscallFailsWithErrno(EFAULT)); } +TEST_P(IPUnboundSocketTest, InsufficientBufferTOS) { + SKIP_IF(GetParam().protocol == IPPROTO_TCP); + + auto socket = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + TOSOption t = GetTOSOption(GetParam().domain); + + in_addr addr4; + in6_addr addr6; + ASSERT_THAT(inet_pton(AF_INET, "127.0.0.1", &addr4), ::testing::Eq(1)); + ASSERT_THAT(inet_pton(AF_INET6, "fe80::", &addr6), ::testing::Eq(1)); + + cmsghdr cmsg = {}; + cmsg.cmsg_len = sizeof(cmsg); + cmsg.cmsg_level = t.cmsg_level; + cmsg.cmsg_type = t.option; + + msghdr msg = {}; + msg.msg_control = &cmsg; + msg.msg_controllen = sizeof(cmsg); + if (GetParam().domain == AF_INET) { + msg.msg_name = &addr4; + msg.msg_namelen = sizeof(addr4); + } else { + msg.msg_name = &addr6; + msg.msg_namelen = sizeof(addr6); + } + + EXPECT_THAT(sendmsg(socket->get(), &msg, 0), SyscallFailsWithErrno(EINVAL)); +} + INSTANTIATE_TEST_SUITE_P( IPUnboundSockets, IPUnboundSocketTest, ::testing::ValuesIn(VecCat(VecCat( -- cgit v1.2.3 From f26a576984052a235b63ec79081a8c4a8c8ffc00 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Wed, 8 Jan 2020 16:35:01 -0800 Subject: Addressed GH comments --- pkg/abi/linux/netfilter.go | 15 ++++-- pkg/sentry/socket/netfilter/netfilter.go | 87 +++++++++++++------------------- pkg/sentry/socket/netstack/netstack.go | 5 +- 3 files changed, 47 insertions(+), 60 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/netfilter.go b/pkg/abi/linux/netfilter.go index 35d66d622..c4f4ea0b1 100644 --- a/pkg/abi/linux/netfilter.go +++ b/pkg/abi/linux/netfilter.go @@ -298,6 +298,7 @@ type IPTReplace struct { // Entries [0]IPTEntry } +// KernelIPTEntry is identical to IPTReplace, but includes the Entries field. type KernelIPTReplace struct { IPTReplace Entries [0]IPTEntry @@ -306,28 +307,32 @@ type KernelIPTReplace struct { // SizeOfIPTReplace is the size of an IPTReplace. const SizeOfIPTReplace = 96 +// ExtensionName holds the name of a netfilter extension. type ExtensionName [XT_EXTENSION_MAXNAMELEN]byte // String implements fmt.Stringer. func (en ExtensionName) String() string { - return name(en[:]) + return goString(en[:]) } +// ExtensionName holds the name of a netfilter table. type TableName [XT_TABLE_MAXNAMELEN]byte // String implements fmt.Stringer. func (tn TableName) String() string { - return name(tn[:]) + return goString(tn[:]) } +// ExtensionName holds the name of a netfilter error. These can also hold +// user-defined chains. type ErrorName [XT_FUNCTION_MAXNAMELEN]byte // String implements fmt.Stringer. -func (fn ErrorName) String() string { - return name(fn[:]) +func (en ErrorName) String() string { + return goString(en[:]) } -func name(cstring []byte) string { +func goString(cstring []byte) string { for i, c := range cstring { if c == 0 { return string(cstring[:i]) diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 347342f98..799865b03 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -53,7 +53,7 @@ func GetInfo(t *kernel.Task, ep tcpip.Endpoint, outPtr usermem.Addr) (linux.IPTG } // Find the appropriate table. - table, err := findTable(ep, info.Name.String()) + table, err := findTable(ep, info.Name) if err != nil { return linux.IPTGetinfo{}, err } @@ -84,7 +84,7 @@ func GetEntries(t *kernel.Task, ep tcpip.Endpoint, outPtr usermem.Addr, outLen i } // Find the appropriate table. - table, err := findTable(ep, userEntries.Name.String()) + table, err := findTable(ep, userEntries.Name) if err != nil { return linux.KernelIPTGetEntries{}, err } @@ -96,19 +96,19 @@ func GetEntries(t *kernel.Task, ep tcpip.Endpoint, outPtr usermem.Addr, outLen i return linux.KernelIPTGetEntries{}, err } if binary.Size(entries) > uintptr(outLen) { - log.Infof("Insufficient GetEntries output size: %d", uintptr(outLen)) + log.Warningf("Insufficient GetEntries output size: %d", uintptr(outLen)) return linux.KernelIPTGetEntries{}, syserr.ErrInvalidArgument } return entries, nil } -func findTable(ep tcpip.Endpoint, tableName string) (iptables.Table, *syserr.Error) { +func findTable(ep tcpip.Endpoint, tablename linux.TableName) (iptables.Table, *syserr.Error) { ipt, err := ep.IPTables() if err != nil { return iptables.Table{}, syserr.FromError(err) } - table, ok := ipt.Tables[tableName] + table, ok := ipt.Tables[tablename.String()] if !ok { return iptables.Table{}, syserr.ErrInvalidArgument } @@ -138,17 +138,17 @@ func FillDefaultIPTables(stack *stack.Stack) { // format expected by the iptables tool. Linux stores each table as a binary // blob that can only be traversed by parsing a bit, reading some offsets, // jumping to those offsets, parsing again, etc. -func convertNetstackToBinary(name string, table iptables.Table) (linux.KernelIPTGetEntries, metadata, *syserr.Error) { +func convertNetstackToBinary(tablename string, table iptables.Table) (linux.KernelIPTGetEntries, metadata, *syserr.Error) { // Return values. var entries linux.KernelIPTGetEntries var meta metadata // The table name has to fit in the struct. - if linux.XT_TABLE_MAXNAMELEN < len(name) { - log.Infof("Table name too long.") + if linux.XT_TABLE_MAXNAMELEN < len(tablename) { + log.Warningf("Table name %q too long.", tablename) return linux.KernelIPTGetEntries{}, metadata{}, syserr.ErrInvalidArgument } - copy(entries.Name[:], name) + copy(entries.Name[:], tablename) for ruleIdx, rule := range table.Rules { // Is this a chain entry point? @@ -273,11 +273,12 @@ func translateToStandardVerdict(val int32) (iptables.Verdict, *syserr.Error) { case -linux.NF_DROP - 1: return iptables.Drop, nil case -linux.NF_QUEUE - 1: - log.Infof("Unsupported iptables verdict QUEUE.") + log.Warningf("Unsupported iptables verdict QUEUE.") case linux.NF_RETURN: - log.Infof("Unsupported iptables verdict RETURN.") + log.Warningf("Unsupported iptables verdict RETURN.") + default: + log.Warningf("Unknown iptables verdict %d.", val) } - log.Infof("Unknown iptables verdict %d.", val) return iptables.Invalid, syserr.ErrInvalidArgument } @@ -288,7 +289,7 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { // Get the basic rules data (struct ipt_replace). if len(optVal) < linux.SizeOfIPTReplace { - log.Infof("netfilter.SetEntries: optVal has insufficient size for replace %d", len(optVal)) + log.Warningf("netfilter.SetEntries: optVal has insufficient size for replace %d", len(optVal)) return syserr.ErrInvalidArgument } var replace linux.IPTReplace @@ -302,7 +303,7 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { case iptables.TablenameFilter: table = iptables.EmptyFilterTable() default: - log.Infof(fmt.Sprintf("We don't yet support writing to the %q table (gvisor.dev/issue/170)", replace.Name.String())) + log.Warningf("We don't yet support writing to the %q table (gvisor.dev/issue/170)", replace.Name.String()) return syserr.ErrInvalidArgument } @@ -312,7 +313,7 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { for entryIdx := uint32(0); entryIdx < replace.NumEntries; entryIdx++ { // Get the struct ipt_entry. if len(optVal) < linux.SizeOfIPTEntry { - log.Infof("netfilter: optVal has insufficient size for entry %d", len(optVal)) + log.Warningf("netfilter: optVal has insufficient size for entry %d", len(optVal)) return syserr.ErrInvalidArgument } var entry linux.IPTEntry @@ -328,7 +329,7 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { // filtering. We reject any nonzero IPTIP values for now. emptyIPTIP := linux.IPTIP{} if entry.IP != emptyIPTIP { - log.Infof("netfilter: non-empty struct iptip found") + log.Warningf("netfilter: non-empty struct iptip found") return syserr.ErrInvalidArgument } @@ -358,11 +359,11 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { } } if ruleIdx := table.BuiltinChains[hk]; ruleIdx == iptables.HookUnset { - log.Infof("Hook %v is unset.", hk) + log.Warningf("Hook %v is unset.", hk) return syserr.ErrInvalidArgument } if ruleIdx := table.Underflows[hk]; ruleIdx == iptables.HookUnset { - log.Infof("Underflow %v is unset.", hk) + log.Warningf("Underflow %v is unset.", hk) return syserr.ErrInvalidArgument } } @@ -385,7 +386,7 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { // along with the number of bytes it occupies in optVal. func parseTarget(optVal []byte) (iptables.Target, uint32, *syserr.Error) { if len(optVal) < linux.SizeOfXTEntryTarget { - log.Infof("netfilter: optVal has insufficient size for entry target %d", len(optVal)) + log.Warningf("netfilter: optVal has insufficient size for entry target %d", len(optVal)) return nil, 0, syserr.ErrInvalidArgument } var target linux.XTEntryTarget @@ -395,14 +396,14 @@ func parseTarget(optVal []byte) (iptables.Target, uint32, *syserr.Error) { case "": // Standard target. if len(optVal) < linux.SizeOfXTStandardTarget { - log.Infof("netfilter.SetEntries: optVal has insufficient size for standard target %d", len(optVal)) + log.Warningf("netfilter.SetEntries: optVal has insufficient size for standard target %d", len(optVal)) return nil, 0, syserr.ErrInvalidArgument } - var target linux.XTStandardTarget + var standardTarget linux.XTStandardTarget buf = optVal[:linux.SizeOfXTStandardTarget] - binary.Unmarshal(buf, usermem.ByteOrder, &target) + binary.Unmarshal(buf, usermem.ByteOrder, &standardTarget) - verdict, err := translateToStandardVerdict(target.Verdict) + verdict, err := translateToStandardVerdict(standardTarget.Verdict) if err != nil { return nil, 0, err } @@ -424,9 +425,9 @@ func parseTarget(optVal []byte) (iptables.Target, uint32, *syserr.Error) { log.Infof("netfilter.SetEntries: optVal has insufficient size for error target %d", len(optVal)) return nil, 0, syserr.ErrInvalidArgument } - var target linux.XTErrorTarget + var errorTarget linux.XTErrorTarget buf = optVal[:linux.SizeOfXTErrorTarget] - binary.Unmarshal(buf, usermem.ByteOrder, &target) + binary.Unmarshal(buf, usermem.ByteOrder, &errorTarget) // Error targets are used in 2 cases: // * An actual error case. These rules have an error @@ -435,11 +436,11 @@ func parseTarget(optVal []byte) (iptables.Target, uint32, *syserr.Error) { // somehow fall through every rule. // * To mark the start of a user defined chain. These // rules have an error with the name of the chain. - switch target.Name.String() { + switch errorTarget.Name.String() { case errorTargetName: return iptables.PanicTarget{}, linux.SizeOfXTErrorTarget, nil default: - log.Infof("Unknown error target %q doesn't exist or isn't supported yet.", target.Name.String()) + log.Infof("Unknown error target %q doesn't exist or isn't supported yet.", errorTarget.Name.String()) return nil, 0, syserr.ErrInvalidArgument } } @@ -449,22 +450,6 @@ func parseTarget(optVal []byte) (iptables.Target, uint32, *syserr.Error) { return nil, 0, syserr.ErrInvalidArgument } -func chainNameFromHook(hook int) string { - switch hook { - case linux.NF_INET_PRE_ROUTING: - return iptables.ChainNamePrerouting - case linux.NF_INET_LOCAL_IN: - return iptables.ChainNameInput - case linux.NF_INET_FORWARD: - return iptables.ChainNameForward - case linux.NF_INET_LOCAL_OUT: - return iptables.ChainNameOutput - case linux.NF_INET_POST_ROUTING: - return iptables.ChainNamePostrouting - } - panic(fmt.Sprintf("Unknown hook %d does not correspond to a builtin chain")) -} - func hookFromLinux(hook int) iptables.Hook { switch hook { case linux.NF_INET_PRE_ROUTING: @@ -489,7 +474,7 @@ func printReplace(optVal []byte) { replaceBuf := optVal[:linux.SizeOfIPTReplace] optVal = optVal[linux.SizeOfIPTReplace:] binary.Unmarshal(replaceBuf, usermem.ByteOrder, &replace) - log.Infof("kevin: Replacing table %q: %+v", replace.Name.String(), replace) + log.Infof("Replacing table %q: %+v", replace.Name.String(), replace) // Read in the list of entries at the end of replace. var totalOffset uint16 @@ -497,33 +482,33 @@ func printReplace(optVal []byte) { var entry linux.IPTEntry entryBuf := optVal[:linux.SizeOfIPTEntry] binary.Unmarshal(entryBuf, usermem.ByteOrder, &entry) - log.Infof("kevin: Entry %d (total offset %d): %+v", entryIdx, totalOffset, entry) + log.Infof("Entry %d (total offset %d): %+v", entryIdx, totalOffset, entry) totalOffset += entry.NextOffset if entry.TargetOffset == linux.SizeOfIPTEntry { - log.Infof("kevin: Entry has no matches.") + log.Infof("Entry has no matches.") } else { - log.Infof("kevin: Entry has matches.") + log.Infof("Entry has matches.") } var target linux.XTEntryTarget targetBuf := optVal[entry.TargetOffset : entry.TargetOffset+linux.SizeOfXTEntryTarget] binary.Unmarshal(targetBuf, usermem.ByteOrder, &target) - log.Infof("kevin: Target named %q: %+v", target.Name.String(), target) + log.Infof("Target named %q: %+v", target.Name.String(), target) switch target.Name.String() { case "": var standardTarget linux.XTStandardTarget stBuf := optVal[entry.TargetOffset : entry.TargetOffset+linux.SizeOfXTStandardTarget] binary.Unmarshal(stBuf, usermem.ByteOrder, &standardTarget) - log.Infof("kevin: Standard target with verdict %q (%d).", linux.VerdictStrings[standardTarget.Verdict], standardTarget.Verdict) + log.Infof("Standard target with verdict %q (%d).", linux.VerdictStrings[standardTarget.Verdict], standardTarget.Verdict) case errorTargetName: var errorTarget linux.XTErrorTarget etBuf := optVal[entry.TargetOffset : entry.TargetOffset+linux.SizeOfXTErrorTarget] binary.Unmarshal(etBuf, usermem.ByteOrder, &errorTarget) - log.Infof("kevin: Error target with name %q.", errorTarget.Name.String()) + log.Infof("Error target with name %q.", errorTarget.Name.String()) default: - log.Infof("kevin: Unknown target type.") + log.Infof("Unknown target type.") } optVal = optVal[entry.NextOffset:] diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 8c07eef4b..cd3dd1a53 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -1368,10 +1368,7 @@ func (s *SocketOperations) SetSockOpt(t *kernel.Task, level int, name int, optVa return syserr.ErrNoDevice } // Stack must be a netstack stack. - if err := netfilter.SetEntries(stack.(*Stack).Stack, optVal); err != nil { - return err - } - return nil + return netfilter.SetEntries(stack.(*Stack).Stack, optVal) case linux.IPT_SO_SET_ADD_COUNTERS: // TODO(gvisor.dev/issue/170): Counter support. -- cgit v1.2.3 From 565b64148314018e1234196182b55c4f01772e77 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Wed, 8 Jan 2020 16:32:50 -0800 Subject: Define sizes for extent headers and entries separately to improve clarity. PiperOrigin-RevId: 288799694 --- pkg/sentry/fsimpl/ext/disklayout/extent.go | 10 +++++++--- pkg/sentry/fsimpl/ext/disklayout/extent_test.go | 6 +++--- pkg/sentry/fsimpl/ext/extent_file.go | 8 ++++---- 3 files changed, 14 insertions(+), 10 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/ext/disklayout/extent.go b/pkg/sentry/fsimpl/ext/disklayout/extent.go index 567523d32..4110649ab 100644 --- a/pkg/sentry/fsimpl/ext/disklayout/extent.go +++ b/pkg/sentry/fsimpl/ext/disklayout/extent.go @@ -29,8 +29,12 @@ package disklayout // byte (i * sb.BlockSize()) to ((i+1) * sb.BlockSize()). const ( - // ExtentStructsSize is the size of all the three extent on-disk structs. - ExtentStructsSize = 12 + // ExtentHeaderSize is the size of the header of an extent tree node. + ExtentHeaderSize = 12 + + // ExtentEntrySize is the size of an entry in an extent tree node. + // This size is the same for both leaf and internal nodes. + ExtentEntrySize = 12 // ExtentMagic is the magic number which must be present in the header. ExtentMagic = 0xf30a @@ -57,7 +61,7 @@ type ExtentNode struct { Entries []ExtentEntryPair } -// ExtentEntry reprsents an extent tree node entry. The entry can either be +// ExtentEntry represents an extent tree node entry. The entry can either be // an ExtentIdx or Extent itself. This exists to simplify navigation logic. type ExtentEntry interface { // FileBlock returns the first file block number covered by this entry. diff --git a/pkg/sentry/fsimpl/ext/disklayout/extent_test.go b/pkg/sentry/fsimpl/ext/disklayout/extent_test.go index b0fad9b71..8762b90db 100644 --- a/pkg/sentry/fsimpl/ext/disklayout/extent_test.go +++ b/pkg/sentry/fsimpl/ext/disklayout/extent_test.go @@ -21,7 +21,7 @@ import ( // TestExtentSize tests that the extent structs are of the correct // size. func TestExtentSize(t *testing.T) { - assertSize(t, ExtentHeader{}, ExtentStructsSize) - assertSize(t, ExtentIdx{}, ExtentStructsSize) - assertSize(t, Extent{}, ExtentStructsSize) + assertSize(t, ExtentHeader{}, ExtentHeaderSize) + assertSize(t, ExtentIdx{}, ExtentEntrySize) + assertSize(t, Extent{}, ExtentEntrySize) } diff --git a/pkg/sentry/fsimpl/ext/extent_file.go b/pkg/sentry/fsimpl/ext/extent_file.go index 3d3ebaca6..11dcc0346 100644 --- a/pkg/sentry/fsimpl/ext/extent_file.go +++ b/pkg/sentry/fsimpl/ext/extent_file.go @@ -57,7 +57,7 @@ func newExtentFile(regFile regularFile) (*extentFile, error) { func (f *extentFile) buildExtTree() error { rootNodeData := f.regFile.inode.diskInode.Data() - binary.Unmarshal(rootNodeData[:disklayout.ExtentStructsSize], binary.LittleEndian, &f.root.Header) + binary.Unmarshal(rootNodeData[:disklayout.ExtentHeaderSize], binary.LittleEndian, &f.root.Header) // Root node can not have more than 4 entries: 60 bytes = 1 header + 4 entries. if f.root.Header.NumEntries > 4 { @@ -67,7 +67,7 @@ func (f *extentFile) buildExtTree() error { } f.root.Entries = make([]disklayout.ExtentEntryPair, f.root.Header.NumEntries) - for i, off := uint16(0), disklayout.ExtentStructsSize; i < f.root.Header.NumEntries; i, off = i+1, off+disklayout.ExtentStructsSize { + for i, off := uint16(0), disklayout.ExtentEntrySize; i < f.root.Header.NumEntries; i, off = i+1, off+disklayout.ExtentEntrySize { var curEntry disklayout.ExtentEntry if f.root.Header.Height == 0 { // Leaf node. @@ -76,7 +76,7 @@ func (f *extentFile) buildExtTree() error { // Internal node. curEntry = &disklayout.ExtentIdx{} } - binary.Unmarshal(rootNodeData[off:off+disklayout.ExtentStructsSize], binary.LittleEndian, curEntry) + binary.Unmarshal(rootNodeData[off:off+disklayout.ExtentEntrySize], binary.LittleEndian, curEntry) f.root.Entries[i].Entry = curEntry } @@ -105,7 +105,7 @@ func (f *extentFile) buildExtTreeFromDisk(entry disklayout.ExtentEntry) (*diskla } entries := make([]disklayout.ExtentEntryPair, header.NumEntries) - for i, off := uint16(0), off+disklayout.ExtentStructsSize; i < header.NumEntries; i, off = i+1, off+disklayout.ExtentStructsSize { + for i, off := uint16(0), off+disklayout.ExtentEntrySize; i < header.NumEntries; i, off = i+1, off+disklayout.ExtentEntrySize { var curEntry disklayout.ExtentEntry if header.Height == 0 { // Leaf node. -- cgit v1.2.3 From fbb2c008e26a7e9d860f6cbf796ea7c375858502 Mon Sep 17 00:00:00 2001 From: Ian Lewis Date: Wed, 8 Jan 2020 16:35:43 -0800 Subject: Return correct length with MSG_TRUNC for unix sockets. This change calls a new Truncate method on the EndpointReader in RecvMsg for both netlink and unix sockets. This allows readers such as sockets to peek at the length of data without actually reading it to a buffer. Fixes #993 #1240 PiperOrigin-RevId: 288800167 --- pkg/sentry/socket/netlink/BUILD | 1 - pkg/sentry/socket/netlink/socket.go | 29 +++--- pkg/sentry/socket/unix/io.go | 13 +++ pkg/sentry/socket/unix/unix.go | 23 ++++- test/syscalls/linux/BUILD | 1 - test/syscalls/linux/socket_non_stream.cc | 113 +++++++++++++++++++++- test/syscalls/linux/socket_non_stream_blocking.cc | 37 +++++++ test/syscalls/linux/socket_stream.cc | 55 ++++++++++- 8 files changed, 250 insertions(+), 22 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netlink/BUILD b/pkg/sentry/socket/netlink/BUILD index 79589e3c8..136821963 100644 --- a/pkg/sentry/socket/netlink/BUILD +++ b/pkg/sentry/socket/netlink/BUILD @@ -22,7 +22,6 @@ go_library( "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/time", - "//pkg/sentry/safemem", "//pkg/sentry/socket", "//pkg/sentry/socket/netlink/port", "//pkg/sentry/socket/unix", diff --git a/pkg/sentry/socket/netlink/socket.go b/pkg/sentry/socket/netlink/socket.go index 4a1b87a9a..d2e3644a6 100644 --- a/pkg/sentry/socket/netlink/socket.go +++ b/pkg/sentry/socket/netlink/socket.go @@ -29,7 +29,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/netlink/port" "gvisor.dev/gvisor/pkg/sentry/socket/unix" @@ -500,29 +499,29 @@ func (s *Socket) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, have trunc := flags&linux.MSG_TRUNC != 0 r := unix.EndpointReader{ + Ctx: t, Endpoint: s.ep, Peek: flags&linux.MSG_PEEK != 0, } + doRead := func() (int64, error) { + return dst.CopyOutFrom(t, &r) + } + // If MSG_TRUNC is set with a zero byte destination then we still need // to read the message and discard it, or in the case where MSG_PEEK is // set, leave it be. In both cases the full message length must be - // returned. However, the memory manager for the destination will not read - // the endpoint if the destination is zero length. - // - // In order for the endpoint to be read when the destination size is zero, - // we must cause a read of the endpoint by using a separate fake zero - // length block sequence and calling the EndpointReader directly. + // returned. if trunc && dst.Addrs.NumBytes() == 0 { - // Perform a read to a zero byte block sequence. We can ignore the - // original destination since it was zero bytes. The length returned by - // ReadToBlocks is ignored and we return the full message length to comply - // with MSG_TRUNC. - _, err := r.ReadToBlocks(safemem.BlockSeqOf(safemem.BlockFromSafeSlice(make([]byte, 0)))) - return int(r.MsgSize), linux.MSG_TRUNC, from, fromLen, socket.ControlMessages{}, syserr.FromError(err) + doRead = func() (int64, error) { + err := r.Truncate() + // Always return zero for bytes read since the destination size is + // zero. + return 0, err + } } - if n, err := dst.CopyOutFrom(t, &r); err != syserror.ErrWouldBlock || flags&linux.MSG_DONTWAIT != 0 { + if n, err := doRead(); err != syserror.ErrWouldBlock || flags&linux.MSG_DONTWAIT != 0 { var mflags int if n < int64(r.MsgSize) { mflags |= linux.MSG_TRUNC @@ -540,7 +539,7 @@ func (s *Socket) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, have defer s.EventUnregister(&e) for { - if n, err := dst.CopyOutFrom(t, &r); err != syserror.ErrWouldBlock { + if n, err := doRead(); err != syserror.ErrWouldBlock { var mflags int if n < int64(r.MsgSize) { mflags |= linux.MSG_TRUNC diff --git a/pkg/sentry/socket/unix/io.go b/pkg/sentry/socket/unix/io.go index 2ec1a662d..2447f24ef 100644 --- a/pkg/sentry/socket/unix/io.go +++ b/pkg/sentry/socket/unix/io.go @@ -83,6 +83,19 @@ type EndpointReader struct { ControlTrunc bool } +// Truncate calls RecvMsg on the endpoint without writing to a destination. +func (r *EndpointReader) Truncate() error { + // Ignore bytes read since it will always be zero. + _, ms, c, ct, err := r.Endpoint.RecvMsg(r.Ctx, [][]byte{}, r.Creds, r.NumRights, r.Peek, r.From) + r.Control = c + r.ControlTrunc = ct + r.MsgSize = ms + if err != nil { + return err.ToError() + } + return nil +} + // ReadToBlocks implements safemem.Reader.ReadToBlocks. func (r *EndpointReader) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) { return safemem.FromVecReaderFunc{func(bufs [][]byte) (int64, error) { diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go index 885758054..91effe89a 100644 --- a/pkg/sentry/socket/unix/unix.go +++ b/pkg/sentry/socket/unix/unix.go @@ -544,8 +544,27 @@ func (s *SocketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags if senderRequested { r.From = &tcpip.FullAddress{} } + + doRead := func() (int64, error) { + return dst.CopyOutFrom(t, &r) + } + + // If MSG_TRUNC is set with a zero byte destination then we still need + // to read the message and discard it, or in the case where MSG_PEEK is + // set, leave it be. In both cases the full message length must be + // returned. + if trunc && dst.Addrs.NumBytes() == 0 { + doRead = func() (int64, error) { + err := r.Truncate() + // Always return zero for bytes read since the destination size is + // zero. + return 0, err + } + + } + var total int64 - if n, err := dst.CopyOutFrom(t, &r); err != syserror.ErrWouldBlock || dontWait { + if n, err := doRead(); err != syserror.ErrWouldBlock || dontWait { var from linux.SockAddr var fromLen uint32 if r.From != nil && len([]byte(r.From.Addr)) != 0 { @@ -580,7 +599,7 @@ func (s *SocketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags defer s.EventUnregister(&e) for { - if n, err := dst.CopyOutFrom(t, &r); err != syserror.ErrWouldBlock { + if n, err := doRead(); err != syserror.ErrWouldBlock { var from linux.SockAddr var fromLen uint32 if r.From != nil { diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 064ce8429..ce8abe217 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -2888,7 +2888,6 @@ cc_library( ":unix_domain_socket_test_util", "//test/util:test_util", "//test/util:thread_util", - "//test/util:timer_util", "@com_google_absl//absl/time", "@com_google_googletest//:gtest", ], diff --git a/test/syscalls/linux/socket_non_stream.cc b/test/syscalls/linux/socket_non_stream.cc index d91c5ed39..c61817f14 100644 --- a/test/syscalls/linux/socket_non_stream.cc +++ b/test/syscalls/linux/socket_non_stream.cc @@ -113,7 +113,7 @@ TEST_P(NonStreamSocketPairTest, RecvmsgMsghdrFlagMsgTrunc) { EXPECT_EQ(0, memcmp(received_data, sent_data, sizeof(received_data))); // Check that msghdr flags were updated. - EXPECT_EQ(msg.msg_flags, MSG_TRUNC); + EXPECT_EQ(msg.msg_flags & MSG_TRUNC, MSG_TRUNC); } // Stream sockets allow data sent with multiple sends to be peeked at in a @@ -193,7 +193,7 @@ TEST_P(NonStreamSocketPairTest, MsgTruncTruncationRecvmsgMsghdrFlagMsgTrunc) { EXPECT_EQ(0, memcmp(received_data, sent_data, sizeof(received_data))); // Check that msghdr flags were updated. - EXPECT_EQ(msg.msg_flags, MSG_TRUNC); + EXPECT_EQ(msg.msg_flags & MSG_TRUNC, MSG_TRUNC); } TEST_P(NonStreamSocketPairTest, MsgTruncSameSize) { @@ -224,5 +224,114 @@ TEST_P(NonStreamSocketPairTest, MsgTruncNotFull) { EXPECT_EQ(0, memcmp(sent_data, received_data, sizeof(sent_data))); } +// This test tests reading from a socket with MSG_TRUNC and a zero length +// receive buffer. The user should be able to get the message length. +TEST_P(NonStreamSocketPairTest, RecvmsgMsgTruncZeroLen) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + char sent_data[10]; + RandomizeBuffer(sent_data, sizeof(sent_data)); + ASSERT_THAT( + RetryEINTR(send)(sockets->first_fd(), sent_data, sizeof(sent_data), 0), + SyscallSucceedsWithValue(sizeof(sent_data))); + + // The receive buffer is of zero length. + char received_data[0] = {}; + + struct iovec iov; + iov.iov_base = received_data; + iov.iov_len = sizeof(received_data); + struct msghdr msg = {}; + msg.msg_flags = -1; + msg.msg_iov = &iov; + msg.msg_iovlen = 1; + + // The syscall succeeds returning the full size of the message on the socket. + ASSERT_THAT(RetryEINTR(recvmsg)(sockets->second_fd(), &msg, MSG_TRUNC), + SyscallSucceedsWithValue(sizeof(sent_data))); + + // Check that MSG_TRUNC is set on msghdr flags. + EXPECT_EQ(msg.msg_flags & MSG_TRUNC, MSG_TRUNC); +} + +// This test tests reading from a socket with MSG_TRUNC | MSG_PEEK and a zero +// length receive buffer. The user should be able to get the message length +// without reading data off the socket. +TEST_P(NonStreamSocketPairTest, RecvmsgMsgTruncMsgPeekZeroLen) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + char sent_data[10]; + RandomizeBuffer(sent_data, sizeof(sent_data)); + ASSERT_THAT( + RetryEINTR(send)(sockets->first_fd(), sent_data, sizeof(sent_data), 0), + SyscallSucceedsWithValue(sizeof(sent_data))); + + // The receive buffer is of zero length. + char peek_data[0] = {}; + + struct iovec peek_iov; + peek_iov.iov_base = peek_data; + peek_iov.iov_len = sizeof(peek_data); + struct msghdr peek_msg = {}; + peek_msg.msg_flags = -1; + peek_msg.msg_iov = &peek_iov; + peek_msg.msg_iovlen = 1; + + // The syscall succeeds returning the full size of the message on the socket. + ASSERT_THAT(RetryEINTR(recvmsg)(sockets->second_fd(), &peek_msg, + MSG_TRUNC | MSG_PEEK), + SyscallSucceedsWithValue(sizeof(sent_data))); + + // Check that MSG_TRUNC is set on msghdr flags because the receive buffer is + // smaller than the message size. + EXPECT_EQ(peek_msg.msg_flags & MSG_TRUNC, MSG_TRUNC); + + char received_data[sizeof(sent_data)] = {}; + + struct iovec received_iov; + received_iov.iov_base = received_data; + received_iov.iov_len = sizeof(received_data); + struct msghdr received_msg = {}; + received_msg.msg_flags = -1; + received_msg.msg_iov = &received_iov; + received_msg.msg_iovlen = 1; + + // Next we can read the actual data. + ASSERT_THAT( + RetryEINTR(recvmsg)(sockets->second_fd(), &received_msg, MSG_TRUNC), + SyscallSucceedsWithValue(sizeof(sent_data))); + + EXPECT_EQ(0, memcmp(sent_data, received_data, sizeof(sent_data))); + + // Check that MSG_TRUNC is not set on msghdr flags because we read the whole + // message. + EXPECT_EQ(received_msg.msg_flags & MSG_TRUNC, 0); +} + +// This test tests reading from a socket with MSG_TRUNC | MSG_PEEK and a zero +// length receive buffer and MSG_DONTWAIT. The user should be able to get an +// EAGAIN or EWOULDBLOCK error response. +TEST_P(NonStreamSocketPairTest, RecvmsgTruncPeekDontwaitZeroLen) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + // NOTE: We don't send any data on the socket. + + // The receive buffer is of zero length. + char peek_data[0] = {}; + + struct iovec peek_iov; + peek_iov.iov_base = peek_data; + peek_iov.iov_len = sizeof(peek_data); + struct msghdr peek_msg = {}; + peek_msg.msg_flags = -1; + peek_msg.msg_iov = &peek_iov; + peek_msg.msg_iovlen = 1; + + // recvmsg fails with EAGAIN because no data is available on the socket. + ASSERT_THAT(RetryEINTR(recvmsg)(sockets->second_fd(), &peek_msg, + MSG_TRUNC | MSG_PEEK | MSG_DONTWAIT), + SyscallFailsWithErrno(EAGAIN)); +} + } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_non_stream_blocking.cc b/test/syscalls/linux/socket_non_stream_blocking.cc index 62d87c1af..b052f6e61 100644 --- a/test/syscalls/linux/socket_non_stream_blocking.cc +++ b/test/syscalls/linux/socket_non_stream_blocking.cc @@ -25,6 +25,7 @@ #include "test/syscalls/linux/socket_test_util.h" #include "test/syscalls/linux/unix_domain_socket_test_util.h" #include "test/util/test_util.h" +#include "test/util/thread_util.h" namespace gvisor { namespace testing { @@ -44,5 +45,41 @@ TEST_P(BlockingNonStreamSocketPairTest, RecvLessThanBufferWaitAll) { SyscallSucceedsWithValue(sizeof(sent_data))); } +// This test tests reading from a socket with MSG_TRUNC | MSG_PEEK and a zero +// length receive buffer and MSG_DONTWAIT. The recvmsg call should block on +// reading the data. +TEST_P(BlockingNonStreamSocketPairTest, + RecvmsgTruncPeekDontwaitZeroLenBlocking) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + // NOTE: We don't initially send any data on the socket. + const int data_size = 10; + char sent_data[data_size]; + RandomizeBuffer(sent_data, data_size); + + // The receive buffer is of zero length. + char peek_data[0] = {}; + + struct iovec peek_iov; + peek_iov.iov_base = peek_data; + peek_iov.iov_len = sizeof(peek_data); + struct msghdr peek_msg = {}; + peek_msg.msg_flags = -1; + peek_msg.msg_iov = &peek_iov; + peek_msg.msg_iovlen = 1; + + ScopedThread t([&]() { + // The syscall succeeds returning the full size of the message on the + // socket. This should block until there is data on the socket. + ASSERT_THAT(RetryEINTR(recvmsg)(sockets->second_fd(), &peek_msg, + MSG_TRUNC | MSG_PEEK), + SyscallSucceedsWithValue(data_size)); + }); + + absl::SleepFor(absl::Seconds(1)); + ASSERT_THAT(RetryEINTR(send)(sockets->first_fd(), sent_data, data_size, 0), + SyscallSucceedsWithValue(data_size)); +} + } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_stream.cc b/test/syscalls/linux/socket_stream.cc index 346443f96..6522b2e01 100644 --- a/test/syscalls/linux/socket_stream.cc +++ b/test/syscalls/linux/socket_stream.cc @@ -104,7 +104,60 @@ TEST_P(StreamSocketPairTest, RecvmsgMsghdrFlagsNoMsgTrunc) { EXPECT_EQ(0, memcmp(received_data, sent_data, sizeof(received_data))); // Check that msghdr flags were cleared (MSG_TRUNC was not set). - EXPECT_EQ(msg.msg_flags, 0); + ASSERT_EQ(msg.msg_flags & MSG_TRUNC, 0); +} + +TEST_P(StreamSocketPairTest, RecvmsgTruncZeroLen) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + char sent_data[10]; + RandomizeBuffer(sent_data, sizeof(sent_data)); + ASSERT_THAT( + RetryEINTR(send)(sockets->first_fd(), sent_data, sizeof(sent_data), 0), + SyscallSucceedsWithValue(sizeof(sent_data))); + + char received_data[0] = {}; + + struct iovec iov; + iov.iov_base = received_data; + iov.iov_len = sizeof(received_data); + struct msghdr msg = {}; + msg.msg_flags = -1; + msg.msg_iov = &iov; + msg.msg_iovlen = 1; + + ASSERT_THAT(RetryEINTR(recvmsg)(sockets->second_fd(), &msg, MSG_TRUNC), + SyscallSucceedsWithValue(0)); + + // Check that msghdr flags were cleared (MSG_TRUNC was not set). + ASSERT_EQ(msg.msg_flags & MSG_TRUNC, 0); +} + +TEST_P(StreamSocketPairTest, RecvmsgTruncPeekZeroLen) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + char sent_data[10]; + RandomizeBuffer(sent_data, sizeof(sent_data)); + ASSERT_THAT( + RetryEINTR(send)(sockets->first_fd(), sent_data, sizeof(sent_data), 0), + SyscallSucceedsWithValue(sizeof(sent_data))); + + char received_data[0] = {}; + + struct iovec iov; + iov.iov_base = received_data; + iov.iov_len = sizeof(received_data); + struct msghdr msg = {}; + msg.msg_flags = -1; + msg.msg_iov = &iov; + msg.msg_iovlen = 1; + + ASSERT_THAT( + RetryEINTR(recvmsg)(sockets->second_fd(), &msg, MSG_TRUNC | MSG_PEEK), + SyscallSucceedsWithValue(0)); + + // Check that msghdr flags were cleared (MSG_TRUNC was not set). + ASSERT_EQ(msg.msg_flags & MSG_TRUNC, 0); } TEST_P(StreamSocketPairTest, MsgTrunc) { -- cgit v1.2.3 From ae060a63d9ad1bfb65b84a2ccbaf2893c5a50b76 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Wed, 8 Jan 2020 17:30:08 -0800 Subject: More GH comments. --- pkg/abi/linux/netfilter.go | 6 +++--- pkg/sentry/socket/netfilter/netfilter.go | 8 ++++---- pkg/tcpip/iptables/BUILD | 5 ++++- pkg/tcpip/iptables/iptables.go | 6 +++--- pkg/tcpip/iptables/targets.go | 16 +++++++++++----- 5 files changed, 25 insertions(+), 16 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/netfilter.go b/pkg/abi/linux/netfilter.go index c4f4ea0b1..33fcc6c95 100644 --- a/pkg/abi/linux/netfilter.go +++ b/pkg/abi/linux/netfilter.go @@ -298,7 +298,7 @@ type IPTReplace struct { // Entries [0]IPTEntry } -// KernelIPTEntry is identical to IPTReplace, but includes the Entries field. +// KernelIPTReplace is identical to IPTReplace, but includes the Entries field. type KernelIPTReplace struct { IPTReplace Entries [0]IPTEntry @@ -315,7 +315,7 @@ func (en ExtensionName) String() string { return goString(en[:]) } -// ExtensionName holds the name of a netfilter table. +// TableName holds the name of a netfilter table. type TableName [XT_TABLE_MAXNAMELEN]byte // String implements fmt.Stringer. @@ -323,7 +323,7 @@ func (tn TableName) String() string { return goString(tn[:]) } -// ExtensionName holds the name of a netfilter error. These can also hold +// ErrorName holds the name of a netfilter error. These can also hold // user-defined chains. type ErrorName [XT_FUNCTION_MAXNAMELEN]byte diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 799865b03..60bb30a9f 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -210,8 +210,8 @@ func marshalTarget(target iptables.Target) []byte { return marshalStandardTarget(iptables.Accept) case iptables.UnconditionalDropTarget: return marshalStandardTarget(iptables.Drop) - case iptables.PanicTarget: - return marshalPanicTarget() + case iptables.ErrorTarget: + return marshalErrorTarget() default: panic(fmt.Errorf("unknown target of type %T", target)) } @@ -230,7 +230,7 @@ func marshalStandardTarget(verdict iptables.Verdict) []byte { return binary.Marshal(ret, usermem.ByteOrder, target) } -func marshalPanicTarget() []byte { +func marshalErrorTarget() []byte { // This is an error target named error target := linux.XTErrorTarget{ Target: linux.XTEntryTarget{ @@ -438,7 +438,7 @@ func parseTarget(optVal []byte) (iptables.Target, uint32, *syserr.Error) { // rules have an error with the name of the chain. switch errorTarget.Name.String() { case errorTargetName: - return iptables.PanicTarget{}, linux.SizeOfXTErrorTarget, nil + return iptables.ErrorTarget{}, linux.SizeOfXTErrorTarget, nil default: log.Infof("Unknown error target %q doesn't exist or isn't supported yet.", errorTarget.Name.String()) return nil, 0, syserr.ErrInvalidArgument diff --git a/pkg/tcpip/iptables/BUILD b/pkg/tcpip/iptables/BUILD index cc5f531e2..64769c333 100644 --- a/pkg/tcpip/iptables/BUILD +++ b/pkg/tcpip/iptables/BUILD @@ -11,5 +11,8 @@ go_library( ], importpath = "gvisor.dev/gvisor/pkg/tcpip/iptables", visibility = ["//visibility:public"], - deps = ["//pkg/tcpip/buffer"], + deps = [ + "//pkg/log", + "//pkg/tcpip/buffer", + ], ) diff --git a/pkg/tcpip/iptables/iptables.go b/pkg/tcpip/iptables/iptables.go index 9e7005374..db0450a21 100644 --- a/pkg/tcpip/iptables/iptables.go +++ b/pkg/tcpip/iptables/iptables.go @@ -45,7 +45,7 @@ func DefaultTables() IPTables { Rule{Target: UnconditionalAcceptTarget{}}, Rule{Target: UnconditionalAcceptTarget{}}, Rule{Target: UnconditionalAcceptTarget{}}, - Rule{Target: PanicTarget{}}, + Rule{Target: ErrorTarget{}}, }, BuiltinChains: map[Hook]int{ Prerouting: 0, @@ -65,7 +65,7 @@ func DefaultTables() IPTables { Rules: []Rule{ Rule{Target: UnconditionalAcceptTarget{}}, Rule{Target: UnconditionalAcceptTarget{}}, - Rule{Target: PanicTarget{}}, + Rule{Target: ErrorTarget{}}, }, BuiltinChains: map[Hook]int{ Prerouting: 0, @@ -82,7 +82,7 @@ func DefaultTables() IPTables { Rule{Target: UnconditionalAcceptTarget{}}, Rule{Target: UnconditionalAcceptTarget{}}, Rule{Target: UnconditionalAcceptTarget{}}, - Rule{Target: PanicTarget{}}, + Rule{Target: ErrorTarget{}}, }, BuiltinChains: map[Hook]int{ Input: 0, diff --git a/pkg/tcpip/iptables/targets.go b/pkg/tcpip/iptables/targets.go index 2c3598e3d..d65ed8df5 100644 --- a/pkg/tcpip/iptables/targets.go +++ b/pkg/tcpip/iptables/targets.go @@ -16,7 +16,10 @@ package iptables -import "gvisor.dev/gvisor/pkg/tcpip/buffer" +import ( + "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/tcpip/buffer" +) // UnconditionalAcceptTarget accepts all packets. type UnconditionalAcceptTarget struct{} @@ -34,10 +37,13 @@ func (UnconditionalDropTarget) Action(packet buffer.VectorisedView) (Verdict, st return Drop, "" } -// PanicTarget just panics. It represents a target that should be unreachable. -type PanicTarget struct{} +// ErrorTarget logs an error and drops the packet. It represents a target that +// should be unreachable. +type ErrorTarget struct{} // Actions implements Target.Action. -func (PanicTarget) Action(packet buffer.VectorisedView) (Verdict, string) { - panic("PanicTarget triggered.") +func (ErrorTarget) Action(packet buffer.VectorisedView) (Verdict, string) { + log.Warningf("ErrorTarget triggered.") + return Drop, "" + } -- cgit v1.2.3 From 781a68eeb65b4db09ae6591a4273e27a2bf60999 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Wed, 8 Jan 2020 21:39:13 -0800 Subject: It works! It drops some packets. --- pkg/sentry/socket/netfilter/netfilter.go | 13 +++++-------- 1 file changed, 5 insertions(+), 8 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 2fc7aeea3..014dfa625 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -52,7 +52,7 @@ func GetInfo(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr) (linux.IPT } // Find the appropriate table. - table, err := findTable(ep, info.Name) + table, err := findTable(stack, info.Name) if err != nil { return linux.IPTGetinfo{}, err } @@ -83,7 +83,7 @@ func GetEntries(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr, outLen } // Find the appropriate table. - table, err := findTable(ep, userEntries.Name) + table, err := findTable(stack, userEntries.Name) if err != nil { return linux.KernelIPTGetEntries{}, err } @@ -102,11 +102,8 @@ func GetEntries(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr, outLen return entries, nil } -func findTable(ep tcpip.Endpoint, tablename linux.TableName) (iptables.Table, *syserr.Error) { - ipt, err := ep.IPTables() - if err != nil { - return iptables.Table{}, syserr.FromError(err) - } +func findTable(stack *stack.Stack, tablename linux.TableName) (iptables.Table, *syserr.Error) { + ipt := stack.IPTables() table, ok := ipt.Tables[tablename.String()] if !ok { return iptables.Table{}, syserr.ErrInvalidArgument @@ -347,7 +344,7 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { // Go through the list of supported hooks for this table and, for each // one, set the rule it corresponds to. for hook, _ := range replace.HookEntry { - if table.ValidHooks()&uint32(hook) != 0 { + if table.ValidHooks()&(1< Date: Thu, 9 Jan 2020 09:01:17 +0000 Subject: Avoid panic when c.PCIDs is nil When PCID is disabled, there would throw a panic when dropPageTables() access to c.PCID without check. Signed-off-by: Lai Jiangshan --- pkg/sentry/platform/kvm/machine_amd64.go | 4 +++- pkg/sentry/platform/kvm/machine_arm64.go | 4 +++- 2 files changed, 6 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/kvm/machine_amd64.go b/pkg/sentry/platform/kvm/machine_amd64.go index b99fe425e..873e39dc7 100644 --- a/pkg/sentry/platform/kvm/machine_amd64.go +++ b/pkg/sentry/platform/kvm/machine_amd64.go @@ -90,7 +90,9 @@ func (m *machine) dropPageTables(pt *pagetables.PageTables) { // Clear from all PCIDs. for _, c := range m.vCPUs { - c.PCIDs.Drop(pt) + if c.PCIDs != nil { + c.PCIDs.Drop(pt) + } } } diff --git a/pkg/sentry/platform/kvm/machine_arm64.go b/pkg/sentry/platform/kvm/machine_arm64.go index 7ae47f291..3b1f20219 100644 --- a/pkg/sentry/platform/kvm/machine_arm64.go +++ b/pkg/sentry/platform/kvm/machine_arm64.go @@ -97,7 +97,9 @@ func (m *machine) dropPageTables(pt *pagetables.PageTables) { // Clear from all PCIDs. for _, c := range m.vCPUs { - c.PCIDs.Drop(pt) + if c.PCIDs != nil { + c.PCIDs.Drop(pt) + } } } -- cgit v1.2.3 From 8643933d6e58492cbe9d5c78124873ab40f65feb Mon Sep 17 00:00:00 2001 From: Eyal Soha Date: Thu, 9 Jan 2020 13:06:24 -0800 Subject: Change BindToDeviceOption to store NICID This makes it possible to call the sockopt from go even when the NIC has no name. PiperOrigin-RevId: 288955236 --- pkg/sentry/socket/netstack/netstack.go | 29 ++++++++-- pkg/tcpip/stack/stack.go | 8 +++ pkg/tcpip/stack/transport_demuxer_test.go | 89 +++++++++++++++---------------- pkg/tcpip/tcpip.go | 2 +- pkg/tcpip/transport/tcp/endpoint.go | 27 ++++------ pkg/tcpip/transport/tcp/tcp_test.go | 42 +++++++-------- pkg/tcpip/transport/udp/endpoint.go | 27 ++++------ pkg/tcpip/transport/udp/udp_test.go | 31 +++++------ 8 files changed, 127 insertions(+), 128 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 9e0d69046..764f11a6b 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -985,13 +985,23 @@ func getSockOptSocket(t *kernel.Task, s socket.Socket, ep commonEndpoint, family if err := ep.GetSockOpt(&v); err != nil { return nil, syserr.TranslateNetstackError(err) } - if len(v) == 0 { + if v == 0 { return []byte{}, nil } if outLen < linux.IFNAMSIZ { return nil, syserr.ErrInvalidArgument } - return append([]byte(v), 0), nil + s := t.NetworkContext() + if s == nil { + return nil, syserr.ErrNoDevice + } + nic, ok := s.Interfaces()[int32(v)] + if !ok { + // The NICID no longer indicates a valid interface, probably because that + // interface was removed. + return nil, syserr.ErrUnknownDevice + } + return append([]byte(nic.Name), 0), nil case linux.SO_BROADCAST: if outLen < sizeOfInt32 { @@ -1438,7 +1448,20 @@ func setSockOptSocket(t *kernel.Task, s socket.Socket, ep commonEndpoint, name i if n == -1 { n = len(optVal) } - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.BindToDeviceOption(optVal[:n]))) + name := string(optVal[:n]) + if name == "" { + return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.BindToDeviceOption(0))) + } + s := t.NetworkContext() + if s == nil { + return syserr.ErrNoDevice + } + for nicID, nic := range s.Interfaces() { + if nic.Name == name { + return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.BindToDeviceOption(nicID))) + } + } + return syserr.ErrUnknownDevice case linux.SO_BROADCAST: if len(optVal) < sizeOfInt32 { diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index e2a2edb2c..41bf9fd9b 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -901,6 +901,14 @@ type NICInfo struct { Context NICContext } +// HasNIC returns true if the NICID is defined in the stack. +func (s *Stack) HasNIC(id tcpip.NICID) bool { + s.mu.RLock() + _, ok := s.nics[id] + s.mu.RUnlock() + return ok +} + // NICInfo returns a map of NICIDs to their associated information. func (s *Stack) NICInfo() map[tcpip.NICID]NICInfo { s.mu.RLock() diff --git a/pkg/tcpip/stack/transport_demuxer_test.go b/pkg/tcpip/stack/transport_demuxer_test.go index df5ced887..5e9237de9 100644 --- a/pkg/tcpip/stack/transport_demuxer_test.go +++ b/pkg/tcpip/stack/transport_demuxer_test.go @@ -41,7 +41,7 @@ const ( type testContext struct { t *testing.T - linkEPs map[string]*channel.Endpoint + linkEps map[tcpip.NICID]*channel.Endpoint s *stack.Stack ep tcpip.Endpoint @@ -66,27 +66,24 @@ func (c *testContext) createV6Endpoint(v6only bool) { } } -// newDualTestContextMultiNic creates the testing context and also linkEpNames -// named NICs. -func newDualTestContextMultiNic(t *testing.T, mtu uint32, linkEpNames []string) *testContext { +// newDualTestContextMultiNIC creates the testing context and also linkEpIDs NICs. +func newDualTestContextMultiNIC(t *testing.T, mtu uint32, linkEpIDs []tcpip.NICID) *testContext { s := stack.New(stack.Options{ NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}}) - linkEPs := make(map[string]*channel.Endpoint) - for i, linkEpName := range linkEpNames { - channelEP := channel.New(256, mtu, "") - nicID := tcpip.NICID(i + 1) - opts := stack.NICOptions{Name: linkEpName} - if err := s.CreateNICWithOptions(nicID, channelEP, opts); err != nil { - t.Fatalf("CreateNICWithOptions(_, _, %+v) failed: %v", opts, err) + linkEps := make(map[tcpip.NICID]*channel.Endpoint) + for _, linkEpID := range linkEpIDs { + channelEp := channel.New(256, mtu, "") + if err := s.CreateNIC(linkEpID, channelEp); err != nil { + t.Fatalf("CreateNIC failed: %v", err) } - linkEPs[linkEpName] = channelEP + linkEps[linkEpID] = channelEp - if err := s.AddAddress(nicID, ipv4.ProtocolNumber, stackAddr); err != nil { + if err := s.AddAddress(linkEpID, ipv4.ProtocolNumber, stackAddr); err != nil { t.Fatalf("AddAddress IPv4 failed: %v", err) } - if err := s.AddAddress(nicID, ipv6.ProtocolNumber, stackV6Addr); err != nil { + if err := s.AddAddress(linkEpID, ipv6.ProtocolNumber, stackV6Addr); err != nil { t.Fatalf("AddAddress IPv6 failed: %v", err) } } @@ -105,7 +102,7 @@ func newDualTestContextMultiNic(t *testing.T, mtu uint32, linkEpNames []string) return &testContext{ t: t, s: s, - linkEPs: linkEPs, + linkEps: linkEps, } } @@ -122,7 +119,7 @@ func newPayload() []byte { return b } -func (c *testContext) sendV6Packet(payload []byte, h *headers, linkEpName string) { +func (c *testContext) sendV6Packet(payload []byte, h *headers, linkEpID tcpip.NICID) { // Allocate a buffer for data and headers. buf := buffer.NewView(header.UDPMinimumSize + header.IPv6MinimumSize + len(payload)) copy(buf[len(buf)-len(payload):], payload) @@ -153,7 +150,7 @@ func (c *testContext) sendV6Packet(payload []byte, h *headers, linkEpName string u.SetChecksum(^u.CalculateChecksum(xsum)) // Inject packet. - c.linkEPs[linkEpName].InjectInbound(ipv6.ProtocolNumber, tcpip.PacketBuffer{ + c.linkEps[linkEpID].InjectInbound(ipv6.ProtocolNumber, tcpip.PacketBuffer{ Data: buf.ToVectorisedView(), }) } @@ -183,7 +180,7 @@ func TestTransportDemuxerRegister(t *testing.T) { func TestDistribution(t *testing.T) { type endpointSockopts struct { reuse int - bindToDevice string + bindToDevice tcpip.NICID } for _, test := range []struct { name string @@ -191,71 +188,71 @@ func TestDistribution(t *testing.T) { endpoints []endpointSockopts // wantedDistribution is the wanted ratio of packets received on each // endpoint for each NIC on which packets are injected. - wantedDistributions map[string][]float64 + wantedDistributions map[tcpip.NICID][]float64 }{ { "BindPortReuse", // 5 endpoints that all have reuse set. []endpointSockopts{ - {1, ""}, - {1, ""}, - {1, ""}, - {1, ""}, - {1, ""}, + {1, 0}, + {1, 0}, + {1, 0}, + {1, 0}, + {1, 0}, }, - map[string][]float64{ + map[tcpip.NICID][]float64{ // Injected packets on dev0 get distributed evenly. - "dev0": {0.2, 0.2, 0.2, 0.2, 0.2}, + 1: {0.2, 0.2, 0.2, 0.2, 0.2}, }, }, { "BindToDevice", // 3 endpoints with various bindings. []endpointSockopts{ - {0, "dev0"}, - {0, "dev1"}, - {0, "dev2"}, + {0, 1}, + {0, 2}, + {0, 3}, }, - map[string][]float64{ + map[tcpip.NICID][]float64{ // Injected packets on dev0 go only to the endpoint bound to dev0. - "dev0": {1, 0, 0}, + 1: {1, 0, 0}, // Injected packets on dev1 go only to the endpoint bound to dev1. - "dev1": {0, 1, 0}, + 2: {0, 1, 0}, // Injected packets on dev2 go only to the endpoint bound to dev2. - "dev2": {0, 0, 1}, + 3: {0, 0, 1}, }, }, { "ReuseAndBindToDevice", // 6 endpoints with various bindings. []endpointSockopts{ - {1, "dev0"}, - {1, "dev0"}, - {1, "dev1"}, - {1, "dev1"}, - {1, "dev1"}, - {1, ""}, + {1, 1}, + {1, 1}, + {1, 2}, + {1, 2}, + {1, 2}, + {1, 0}, }, - map[string][]float64{ + map[tcpip.NICID][]float64{ // Injected packets on dev0 get distributed among endpoints bound to // dev0. - "dev0": {0.5, 0.5, 0, 0, 0, 0}, + 1: {0.5, 0.5, 0, 0, 0, 0}, // Injected packets on dev1 get distributed among endpoints bound to // dev1 or unbound. - "dev1": {0, 0, 1. / 3, 1. / 3, 1. / 3, 0}, + 2: {0, 0, 1. / 3, 1. / 3, 1. / 3, 0}, // Injected packets on dev999 go only to the unbound. - "dev999": {0, 0, 0, 0, 0, 1}, + 1000: {0, 0, 0, 0, 0, 1}, }, }, } { t.Run(test.name, func(t *testing.T) { for device, wantedDistribution := range test.wantedDistributions { - t.Run(device, func(t *testing.T) { - var devices []string + t.Run(string(device), func(t *testing.T) { + var devices []tcpip.NICID for d := range test.wantedDistributions { devices = append(devices, d) } - c := newDualTestContextMultiNic(t, defaultMTU, devices) + c := newDualTestContextMultiNIC(t, defaultMTU, devices) defer c.cleanup() c.createV6Endpoint(false) diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 1eca76c30..72b5ce179 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -552,7 +552,7 @@ type ReusePortOption int // BindToDeviceOption is used by SetSockOpt/GetSockOpt to specify that sockets // should bind only on a specific NIC. -type BindToDeviceOption string +type BindToDeviceOption NICID // QuickAckOption is stubbed out in SetSockOpt/GetSockOpt. type QuickAckOption int diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 2ac1b6877..920b24975 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -1279,19 +1279,14 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { return nil case tcpip.BindToDeviceOption: - e.mu.Lock() - defer e.mu.Unlock() - if v == "" { - e.bindToDevice = 0 - return nil - } - for nicID, nic := range e.stack.NICInfo() { - if nic.Name == string(v) { - e.bindToDevice = nicID - return nil - } + id := tcpip.NICID(v) + if id != 0 && !e.stack.HasNIC(id) { + return tcpip.ErrUnknownDevice } - return tcpip.ErrUnknownDevice + e.mu.Lock() + e.bindToDevice = id + e.mu.Unlock() + return nil case tcpip.QuickAckOption: if v == 0 { @@ -1550,12 +1545,8 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { case *tcpip.BindToDeviceOption: e.mu.RLock() - defer e.mu.RUnlock() - if nic, ok := e.stack.NICInfo()[e.bindToDevice]; ok { - *o = tcpip.BindToDeviceOption(nic.Name) - return nil - } - *o = "" + *o = tcpip.BindToDeviceOption(e.bindToDevice) + e.mu.RUnlock() return nil case *tcpip.QuickAckOption: diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index 15745ebd4..1aa0733d0 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -1083,12 +1083,12 @@ func TestTrafficClassV6(t *testing.T) { func TestConnectBindToDevice(t *testing.T) { for _, test := range []struct { name string - device string + device tcpip.NICID want tcp.EndpointState }{ - {"RightDevice", "nic1", tcp.StateEstablished}, - {"WrongDevice", "nic2", tcp.StateSynSent}, - {"AnyDevice", "", tcp.StateEstablished}, + {"RightDevice", 1, tcp.StateEstablished}, + {"WrongDevice", 2, tcp.StateSynSent}, + {"AnyDevice", 0, tcp.StateEstablished}, } { t.Run(test.name, func(t *testing.T) { c := context.New(t, defaultMTU) @@ -3794,47 +3794,41 @@ func TestBindToDeviceOption(t *testing.T) { } defer ep.Close() - opts := stack.NICOptions{Name: "my_device"} - if err := s.CreateNICWithOptions(321, loopback.New(), opts); err != nil { - t.Errorf("CreateNICWithOptions(_, _, %+v) failed: %v", opts, err) - } - - // Make an nameless NIC. - if err := s.CreateNIC(54321, loopback.New()); err != nil { + if err := s.CreateNIC(321, loopback.New()); err != nil { t.Errorf("CreateNIC failed: %v", err) } - // strPtr is used instead of taking the address of string literals, which is + // nicIDPtr is used instead of taking the address of NICID literals, which is // a compiler error. - strPtr := func(s string) *string { + nicIDPtr := func(s tcpip.NICID) *tcpip.NICID { return &s } testActions := []struct { name string - setBindToDevice *string + setBindToDevice *tcpip.NICID setBindToDeviceError *tcpip.Error getBindToDevice tcpip.BindToDeviceOption }{ - {"GetDefaultValue", nil, nil, ""}, - {"BindToNonExistent", strPtr("non_existent_device"), tcpip.ErrUnknownDevice, ""}, - {"BindToExistent", strPtr("my_device"), nil, "my_device"}, - {"UnbindToDevice", strPtr(""), nil, ""}, + {"GetDefaultValue", nil, nil, 0}, + {"BindToNonExistent", nicIDPtr(999), tcpip.ErrUnknownDevice, 0}, + {"BindToExistent", nicIDPtr(321), nil, 321}, + {"UnbindToDevice", nicIDPtr(0), nil, 0}, } for _, testAction := range testActions { t.Run(testAction.name, func(t *testing.T) { if testAction.setBindToDevice != nil { bindToDevice := tcpip.BindToDeviceOption(*testAction.setBindToDevice) - if got, want := ep.SetSockOpt(bindToDevice), testAction.setBindToDeviceError; got != want { - t.Errorf("SetSockOpt(%v) got %v, want %v", bindToDevice, got, want) + if gotErr, wantErr := ep.SetSockOpt(bindToDevice), testAction.setBindToDeviceError; gotErr != wantErr { + t.Errorf("SetSockOpt(%v) got %v, want %v", bindToDevice, gotErr, wantErr) } } - bindToDevice := tcpip.BindToDeviceOption("to be modified by GetSockOpt") - if ep.GetSockOpt(&bindToDevice) != nil { - t.Errorf("GetSockOpt got %v, want %v", ep.GetSockOpt(&bindToDevice), nil) + bindToDevice := tcpip.BindToDeviceOption(88888) + if err := ep.GetSockOpt(&bindToDevice); err != nil { + t.Errorf("GetSockOpt got %v, want %v", err, nil) } if got, want := bindToDevice, testAction.getBindToDevice; got != want { - t.Errorf("bindToDevice got %q, want %q", got, want) + t.Errorf("bindToDevice got %d, want %d", got, want) } }) } diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 1a5ee6317..864dc8733 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -631,19 +631,14 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.mu.Unlock() case tcpip.BindToDeviceOption: - e.mu.Lock() - defer e.mu.Unlock() - if v == "" { - e.bindToDevice = 0 - return nil - } - for nicID, nic := range e.stack.NICInfo() { - if nic.Name == string(v) { - e.bindToDevice = nicID - return nil - } + id := tcpip.NICID(v) + if id != 0 && !e.stack.HasNIC(id) { + return tcpip.ErrUnknownDevice } - return tcpip.ErrUnknownDevice + e.mu.Lock() + e.bindToDevice = id + e.mu.Unlock() + return nil case tcpip.BroadcastOption: e.mu.Lock() @@ -767,12 +762,8 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { case *tcpip.BindToDeviceOption: e.mu.RLock() - defer e.mu.RUnlock() - if nic, ok := e.stack.NICInfo()[e.bindToDevice]; ok { - *o = tcpip.BindToDeviceOption(nic.Name) - return nil - } - *o = tcpip.BindToDeviceOption("") + *o = tcpip.BindToDeviceOption(e.bindToDevice) + e.mu.RUnlock() return nil case *tcpip.KeepaliveEnabledOption: diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index 149fff999..0a82bc4fa 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -513,42 +513,37 @@ func TestBindToDeviceOption(t *testing.T) { t.Errorf("CreateNICWithOptions(_, _, %+v) failed: %v", opts, err) } - // Make an nameless NIC. - if err := s.CreateNIC(54321, loopback.New()); err != nil { - t.Errorf("CreateNIC failed: %v", err) - } - - // strPtr is used instead of taking the address of string literals, which is + // nicIDPtr is used instead of taking the address of NICID literals, which is // a compiler error. - strPtr := func(s string) *string { + nicIDPtr := func(s tcpip.NICID) *tcpip.NICID { return &s } testActions := []struct { name string - setBindToDevice *string + setBindToDevice *tcpip.NICID setBindToDeviceError *tcpip.Error getBindToDevice tcpip.BindToDeviceOption }{ - {"GetDefaultValue", nil, nil, ""}, - {"BindToNonExistent", strPtr("non_existent_device"), tcpip.ErrUnknownDevice, ""}, - {"BindToExistent", strPtr("my_device"), nil, "my_device"}, - {"UnbindToDevice", strPtr(""), nil, ""}, + {"GetDefaultValue", nil, nil, 0}, + {"BindToNonExistent", nicIDPtr(999), tcpip.ErrUnknownDevice, 0}, + {"BindToExistent", nicIDPtr(321), nil, 321}, + {"UnbindToDevice", nicIDPtr(0), nil, 0}, } for _, testAction := range testActions { t.Run(testAction.name, func(t *testing.T) { if testAction.setBindToDevice != nil { bindToDevice := tcpip.BindToDeviceOption(*testAction.setBindToDevice) - if got, want := ep.SetSockOpt(bindToDevice), testAction.setBindToDeviceError; got != want { - t.Errorf("SetSockOpt(%v) got %v, want %v", bindToDevice, got, want) + if gotErr, wantErr := ep.SetSockOpt(bindToDevice), testAction.setBindToDeviceError; gotErr != wantErr { + t.Errorf("SetSockOpt(%v) got %v, want %v", bindToDevice, gotErr, wantErr) } } - bindToDevice := tcpip.BindToDeviceOption("to be modified by GetSockOpt") - if ep.GetSockOpt(&bindToDevice) != nil { - t.Errorf("GetSockOpt got %v, want %v", ep.GetSockOpt(&bindToDevice), nil) + bindToDevice := tcpip.BindToDeviceOption(88888) + if err := ep.GetSockOpt(&bindToDevice); err != nil { + t.Errorf("GetSockOpt got %v, want %v", err, nil) } if got, want := bindToDevice, testAction.getBindToDevice; got != want { - t.Errorf("bindToDevice got %q, want %q", got, want) + t.Errorf("bindToDevice got %d, want %d", got, want) } }) } -- cgit v1.2.3 From 89d11b4d96b0c40e373f14ba72d570c9b894f976 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Thu, 9 Jan 2020 13:41:52 -0800 Subject: Added a test that we don't pass yet --- pkg/sentry/socket/netfilter/netfilter.go | 37 +++++++++++++++++++++++++---- pkg/tcpip/iptables/iptables.go | 7 ++++++ pkg/tcpip/iptables/types.go | 4 ++-- test/iptables/BUILD | 4 ++++ test/iptables/filter_input.go | 30 ++++++++++++++++++++++++ test/iptables/iptables_test.go | 16 +++++++++---- test/iptables/iptables_util.go | 40 ++++++++++++++++++++++++++++++++ test/iptables/runner/BUILD | 1 + 8 files changed, 127 insertions(+), 12 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 014dfa625..f30461936 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -323,10 +323,9 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { // TODO(gvisor.dev/issue/170): We should support IPTIP // filtering. We reject any nonzero IPTIP values for now. - emptyIPTIP := linux.IPTIP{} - if entry.IP != emptyIPTIP { - log.Warningf("netfilter: non-empty struct iptip found") - return syserr.ErrInvalidArgument + filter, err := filterFromIPTIP(entry.IP) + if err != nil { + return err } // Get the target of the rule. @@ -336,7 +335,10 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { } optVal = optVal[consumed:] - table.Rules = append(table.Rules, iptables.Rule{Target: target}) + table.Rules = append(table.Rules, iptables.Rule{ + Filter: filter, + Target: target, + }) offsets = append(offsets, offset) offset += linux.SizeOfIPTEntry + consumed } @@ -447,6 +449,31 @@ func parseTarget(optVal []byte) (iptables.Target, uint32, *syserr.Error) { return nil, 0, syserr.ErrInvalidArgument } +func filterFromIPTIP(iptip linux.IPTIP) (iptables.IPHeaderFilter, *syserr.Error) { + if containsUnsupportedFields(iptip) { + log.Warningf("netfilter: unsupported fields in struct iptip: %+v") + return iptables.IPHeaderFilter{}, syserr.ErrInvalidArgument + } + return iptables.IPHeaderFilter{ + Protocol: iptip.Protocol, + }, nil +} + +func containsUnsupportedFields(iptip linux.IPTIP) bool { + // Currently we check that everything except protocol is zeroed. + var emptyInetAddr = linux.InetAddr{} + var emptyInterface = [linux.IFNAMSIZ]byte{} + return iptip.Dst != emptyInetAddr || + iptip.SrcMask != emptyInetAddr || + iptip.DstMask != emptyInetAddr || + iptip.InputInterface != emptyInterface || + iptip.OutputInterface != emptyInterface || + iptip.InputInterfaceMask != emptyInterface || + iptip.OutputInterfaceMask != emptyInterface || + iptip.Flags != 0 || + iptip.InverseFlags != 0 +} + func hookFromLinux(hook int) iptables.Hook { switch hook { case linux.NF_INET_PRE_ROUTING: diff --git a/pkg/tcpip/iptables/iptables.go b/pkg/tcpip/iptables/iptables.go index 91abbbea8..b8d70ec1e 100644 --- a/pkg/tcpip/iptables/iptables.go +++ b/pkg/tcpip/iptables/iptables.go @@ -185,6 +185,13 @@ func (it *IPTables) checkTable(hook Hook, pkt tcpip.PacketBuffer, tablename stri func (it *IPTables) checkRule(hook Hook, pkt tcpip.PacketBuffer, table Table, ruleIdx int) Verdict { rule := table.Rules[ruleIdx] + + // First check whether the packet matches the IP header filter. + // TODO(gvisor.dev/issue/170): Support other fields of the filter. + // if rule.Filter.Protocol != pkt.Protocol { + // return Continue + // } + // Go through each rule matcher. If they all match, run // the rule target. for _, matcher := range rule.Matchers { diff --git a/pkg/tcpip/iptables/types.go b/pkg/tcpip/iptables/types.go index 4b2a9c294..4bedd9bc8 100644 --- a/pkg/tcpip/iptables/types.go +++ b/pkg/tcpip/iptables/types.go @@ -151,8 +151,8 @@ func (table *Table) SetMetadata(metadata interface{}) { // packets this rule applies to. If there are no matchers in the rule, it // applies to any packet. type Rule struct { - // IPHeaderFilters holds basic IP filtering fields common to every rule. - IPHeaderFilter IPHeaderFilter + // IPHeaderFilter holds basic IP filtering fields common to every rule. + Filter IPHeaderFilter // Matchers is the list of matchers for this rule. Matchers []Matcher diff --git a/test/iptables/BUILD b/test/iptables/BUILD index fa833c3b2..6a9d05828 100644 --- a/test/iptables/BUILD +++ b/test/iptables/BUILD @@ -4,6 +4,7 @@ package(licenses = ["notice"]) go_library( name = "iptables", + testonly = 1, srcs = [ "filter_input.go", "iptables.go", @@ -11,6 +12,9 @@ go_library( ], importpath = "gvisor.dev/gvisor/test/iptables", visibility = ["//test/iptables:__subpackages__"], + deps = [ + "//runsc/testutil", + ], ) go_test( diff --git a/test/iptables/filter_input.go b/test/iptables/filter_input.go index 7c4d469fa..a3f0052b5 100644 --- a/test/iptables/filter_input.go +++ b/test/iptables/filter_input.go @@ -28,6 +28,7 @@ const ( ) func init() { + RegisterTestCase(FilterInputDropOnlyUDP{}) RegisterTestCase(FilterInputDropUDP{}) RegisterTestCase(FilterInputDropUDPPort{}) RegisterTestCase(FilterInputDropDifferentUDPPort{}) @@ -65,6 +66,35 @@ func (FilterInputDropUDP) LocalAction(ip net.IP) error { return sendUDPLoop(ip, dropPort, sendloopDuration) } +// FilterInputDropOnlyUDP tests that "-p udp -j DROP" only affects UDP traffic. +type FilterInputDropOnlyUDP struct{} + +// Name implements TestCase.Name. +func (FilterInputDropOnlyUDP) Name() string { + return "FilterInputDropOnlyUDP" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterInputDropOnlyUDP) ContainerAction(ip net.IP) error { + if err := filterTable("-A", "INPUT", "-p", "udp", "-j", "DROP"); err != nil { + return err + } + + // Listen for a TCP connection, which should be allowed. + if err := listenTCP(acceptPort, sendloopDuration); err != nil { + return fmt.Errorf("failed to establish a connection %v", err) + } + + return nil +} + +// LocalAction implements TestCase.LocalAction. +func (FilterInputDropOnlyUDP) LocalAction(ip net.IP) error { + // Try to establish a TCP connection with the container, which should + // succeed. + return connectLoopTCP(ip, acceptPort, sendloopDuration) +} + // FilterInputDropUDPPort tests that we can drop UDP traffic by port. type FilterInputDropUDPPort struct{} diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go index d040e971a..beaaf519c 100644 --- a/test/iptables/iptables_test.go +++ b/test/iptables/iptables_test.go @@ -160,11 +160,11 @@ func logContainer(output string, err error) { log.Infof(msg) } -func TestFilterInputDropUDP(t *testing.T) { - if err := singleTest(FilterInputDropUDP{}); err != nil { - t.Fatal(err) - } -} +// func TestFilterInputDropUDP(t *testing.T) { +// if err := singleTest(FilterInputDropUDP{}); err != nil { +// t.Fatal(err) +// } +// } // func TestFilterInputDropUDPPort(t *testing.T) { // if err := singleTest(FilterInputDropUDPPort{}); err != nil { @@ -183,3 +183,9 @@ func TestFilterInputDropUDP(t *testing.T) { // t.Fatal(err) // } // } + +func TestFilterInputDropOnlyUDP(t *testing.T) { + if err := singleTest(FilterInputDropOnlyUDP{}); err != nil { + t.Fatal(err) + } +} diff --git a/test/iptables/iptables_util.go b/test/iptables/iptables_util.go index 3a4d11f1a..3dcaafb79 100644 --- a/test/iptables/iptables_util.go +++ b/test/iptables/iptables_util.go @@ -19,6 +19,8 @@ import ( "net" "os/exec" "time" + + "gvisor.dev/gvisor/runsc/testutil" ) const iptablesBinary = "iptables" @@ -80,3 +82,41 @@ func sendUDPLoop(ip net.IP, port int, duration time.Duration) error { return nil } + +func listenTCP(port int, timeout time.Duration) error { + localAddr := net.TCPAddr{Port: acceptPort} + listener, err := net.ListenTCP("tcp4", &localAddr) + if err != nil { + return err + } + defer listener.Close() + listener.SetDeadline(time.Now().Add(timeout)) + conn, err := listener.AcceptTCP() + if err != nil { + return fmt.Errorf("failed to establish a connection %v", err) + } + defer conn.Close() + + return nil +} + +func connectLoopTCP(ip net.IP, port int, timeout time.Duration) error { + contAddr := net.TCPAddr{ + IP: ip, + Port: port, + } + // The container may not be listening when we first connect, so retry + // upon error. + cb := func() error { + conn, err := net.DialTCP("tcp4", nil, &contAddr) + if conn != nil { + conn.Close() + } + return err + } + if err := testutil.Poll(cb, timeout); err != nil { + return fmt.Errorf("timed out waiting to send IP, most recent error: %v", err) + } + + return nil +} diff --git a/test/iptables/runner/BUILD b/test/iptables/runner/BUILD index c6c42d870..a5b6f082c 100644 --- a/test/iptables/runner/BUILD +++ b/test/iptables/runner/BUILD @@ -10,6 +10,7 @@ container_image( go_image( name = "runner", + testonly = 1, srcs = ["main.go"], base = ":iptables-base", deps = ["//test/iptables"], -- cgit v1.2.3 From 27500d529f7fb87eef8812278fd1bbca67bcba72 Mon Sep 17 00:00:00 2001 From: Ian Gudger Date: Thu, 9 Jan 2020 22:00:42 -0800 Subject: New sync package. * Rename syncutil to sync. * Add aliases to sync types. * Replace existing usage of standard library sync package. This will make it easier to swap out synchronization primitives. For example, this will allow us to use primitives from github.com/sasha-s/go-deadlock to check for lock ordering violations. Updates #1472 PiperOrigin-RevId: 289033387 --- pkg/amutex/BUILD | 1 + pkg/amutex/amutex_test.go | 3 +- pkg/atomicbitops/BUILD | 1 + pkg/atomicbitops/atomic_bitops_test.go | 3 +- pkg/compressio/BUILD | 5 +- pkg/compressio/compressio.go | 2 +- pkg/control/server/BUILD | 1 + pkg/control/server/server.go | 2 +- pkg/eventchannel/BUILD | 2 + pkg/eventchannel/event.go | 2 +- pkg/eventchannel/event_test.go | 2 +- pkg/fdchannel/BUILD | 1 + pkg/fdchannel/fdchannel_test.go | 3 +- pkg/fdnotifier/BUILD | 1 + pkg/fdnotifier/fdnotifier.go | 2 +- pkg/flipcall/BUILD | 3 +- pkg/flipcall/flipcall_example_test.go | 3 +- pkg/flipcall/flipcall_test.go | 3 +- pkg/flipcall/flipcall_unsafe.go | 10 +- pkg/gate/BUILD | 1 + pkg/gate/gate_test.go | 2 +- pkg/linewriter/BUILD | 1 + pkg/linewriter/linewriter.go | 3 +- pkg/log/BUILD | 5 +- pkg/log/log.go | 2 +- pkg/metric/BUILD | 1 + pkg/metric/metric.go | 2 +- pkg/p9/BUILD | 1 + pkg/p9/client.go | 2 +- pkg/p9/p9test/BUILD | 2 + pkg/p9/p9test/client_test.go | 2 +- pkg/p9/p9test/p9test.go | 2 +- pkg/p9/path_tree.go | 3 +- pkg/p9/pool.go | 2 +- pkg/p9/server.go | 2 +- pkg/p9/transport.go | 2 +- pkg/procid/BUILD | 2 + pkg/procid/procid_test.go | 3 +- pkg/rand/BUILD | 5 +- pkg/rand/rand_linux.go | 2 +- pkg/refs/BUILD | 2 + pkg/refs/refcounter.go | 2 +- pkg/refs/refcounter_test.go | 3 +- pkg/sentry/arch/BUILD | 1 + pkg/sentry/arch/arch_x86.go | 2 +- pkg/sentry/control/BUILD | 1 + pkg/sentry/control/pprof.go | 2 +- pkg/sentry/device/BUILD | 5 +- pkg/sentry/device/device.go | 2 +- pkg/sentry/fs/BUILD | 3 +- pkg/sentry/fs/copy_up.go | 2 +- pkg/sentry/fs/copy_up_test.go | 2 +- pkg/sentry/fs/dirent.go | 2 +- pkg/sentry/fs/dirent_cache.go | 3 +- pkg/sentry/fs/dirent_cache_limiter.go | 3 +- pkg/sentry/fs/fdpipe/BUILD | 1 + pkg/sentry/fs/fdpipe/pipe.go | 2 +- pkg/sentry/fs/fdpipe/pipe_state.go | 2 +- pkg/sentry/fs/file.go | 2 +- pkg/sentry/fs/file_overlay.go | 2 +- pkg/sentry/fs/filesystems.go | 2 +- pkg/sentry/fs/fs.go | 3 +- pkg/sentry/fs/fsutil/BUILD | 1 + pkg/sentry/fs/fsutil/host_file_mapper.go | 2 +- pkg/sentry/fs/fsutil/host_mappable.go | 2 +- pkg/sentry/fs/fsutil/inode.go | 3 +- pkg/sentry/fs/fsutil/inode_cached.go | 2 +- pkg/sentry/fs/gofer/BUILD | 1 + pkg/sentry/fs/gofer/inode.go | 2 +- pkg/sentry/fs/gofer/session.go | 2 +- pkg/sentry/fs/host/BUILD | 1 + pkg/sentry/fs/host/inode.go | 2 +- pkg/sentry/fs/host/socket.go | 2 +- pkg/sentry/fs/host/tty.go | 3 +- pkg/sentry/fs/inode.go | 3 +- pkg/sentry/fs/inode_inotify.go | 3 +- pkg/sentry/fs/inotify.go | 2 +- pkg/sentry/fs/inotify_watch.go | 2 +- pkg/sentry/fs/lock/BUILD | 1 + pkg/sentry/fs/lock/lock.go | 2 +- pkg/sentry/fs/mounts.go | 2 +- pkg/sentry/fs/overlay.go | 5 +- pkg/sentry/fs/proc/BUILD | 1 + pkg/sentry/fs/proc/seqfile/BUILD | 1 + pkg/sentry/fs/proc/seqfile/seqfile.go | 2 +- pkg/sentry/fs/proc/sys_net.go | 2 +- pkg/sentry/fs/ramfs/BUILD | 1 + pkg/sentry/fs/ramfs/dir.go | 2 +- pkg/sentry/fs/restore.go | 2 +- pkg/sentry/fs/tmpfs/BUILD | 1 + pkg/sentry/fs/tmpfs/inode_file.go | 2 +- pkg/sentry/fs/tty/BUILD | 1 + pkg/sentry/fs/tty/dir.go | 2 +- pkg/sentry/fs/tty/line_discipline.go | 2 +- pkg/sentry/fs/tty/queue.go | 3 +- pkg/sentry/fsimpl/ext/BUILD | 1 + pkg/sentry/fsimpl/ext/directory.go | 3 +- pkg/sentry/fsimpl/ext/filesystem.go | 2 +- pkg/sentry/fsimpl/ext/regular_file.go | 2 +- pkg/sentry/fsimpl/kernfs/BUILD | 2 + pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 2 +- pkg/sentry/fsimpl/kernfs/kernfs.go | 2 +- pkg/sentry/fsimpl/kernfs/kernfs_test.go | 2 +- pkg/sentry/fsimpl/tmpfs/BUILD | 1 + pkg/sentry/fsimpl/tmpfs/regular_file.go | 2 +- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 2 +- pkg/sentry/kernel/BUILD | 5 +- pkg/sentry/kernel/abstract_socket_namespace.go | 2 +- pkg/sentry/kernel/auth/BUILD | 3 +- pkg/sentry/kernel/auth/user_namespace.go | 2 +- pkg/sentry/kernel/epoll/BUILD | 1 + pkg/sentry/kernel/epoll/epoll.go | 2 +- pkg/sentry/kernel/eventfd/BUILD | 1 + pkg/sentry/kernel/eventfd/eventfd.go | 2 +- pkg/sentry/kernel/fasync/BUILD | 1 + pkg/sentry/kernel/fasync/fasync.go | 3 +- pkg/sentry/kernel/fd_table.go | 2 +- pkg/sentry/kernel/fd_table_test.go | 2 +- pkg/sentry/kernel/fs_context.go | 2 +- pkg/sentry/kernel/futex/BUILD | 8 +- pkg/sentry/kernel/futex/futex.go | 3 +- pkg/sentry/kernel/futex/futex_test.go | 2 +- pkg/sentry/kernel/kernel.go | 2 +- pkg/sentry/kernel/memevent/BUILD | 1 + pkg/sentry/kernel/memevent/memory_events.go | 2 +- pkg/sentry/kernel/pipe/BUILD | 1 + pkg/sentry/kernel/pipe/buffer.go | 2 +- pkg/sentry/kernel/pipe/node.go | 3 +- pkg/sentry/kernel/pipe/pipe.go | 2 +- pkg/sentry/kernel/pipe/pipe_util.go | 2 +- pkg/sentry/kernel/pipe/vfs.go | 3 +- pkg/sentry/kernel/semaphore/BUILD | 1 + pkg/sentry/kernel/semaphore/semaphore.go | 2 +- pkg/sentry/kernel/shm/BUILD | 1 + pkg/sentry/kernel/shm/shm.go | 2 +- pkg/sentry/kernel/signal_handlers.go | 3 +- pkg/sentry/kernel/signalfd/BUILD | 1 + pkg/sentry/kernel/signalfd/signalfd.go | 3 +- pkg/sentry/kernel/syscalls.go | 2 +- pkg/sentry/kernel/syslog.go | 3 +- pkg/sentry/kernel/task.go | 5 +- pkg/sentry/kernel/thread_group.go | 2 +- pkg/sentry/kernel/threads.go | 2 +- pkg/sentry/kernel/time/BUILD | 1 + pkg/sentry/kernel/time/time.go | 2 +- pkg/sentry/kernel/timekeeper.go | 2 +- pkg/sentry/kernel/tty.go | 2 +- pkg/sentry/kernel/uts_namespace.go | 3 +- pkg/sentry/limits/BUILD | 1 + pkg/sentry/limits/limits.go | 3 +- pkg/sentry/mm/BUILD | 2 +- pkg/sentry/mm/aio_context.go | 3 +- pkg/sentry/mm/mm.go | 8 +- pkg/sentry/pgalloc/BUILD | 1 + pkg/sentry/pgalloc/pgalloc.go | 2 +- pkg/sentry/platform/interrupt/BUILD | 1 + pkg/sentry/platform/interrupt/interrupt.go | 3 +- pkg/sentry/platform/kvm/BUILD | 1 + pkg/sentry/platform/kvm/address_space.go | 2 +- pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go | 2 - pkg/sentry/platform/kvm/kvm.go | 2 +- pkg/sentry/platform/kvm/machine.go | 2 +- pkg/sentry/platform/ptrace/BUILD | 1 + pkg/sentry/platform/ptrace/ptrace.go | 2 +- pkg/sentry/platform/ptrace/subprocess.go | 2 +- .../platform/ptrace/subprocess_linux_unsafe.go | 2 +- pkg/sentry/platform/ring0/defs.go | 2 +- pkg/sentry/platform/ring0/defs_amd64.go | 1 + pkg/sentry/platform/ring0/defs_arm64.go | 1 + pkg/sentry/platform/ring0/pagetables/BUILD | 5 +- pkg/sentry/platform/ring0/pagetables/pcids_x86.go | 2 +- pkg/sentry/socket/netlink/BUILD | 1 + pkg/sentry/socket/netlink/port/BUILD | 1 + pkg/sentry/socket/netlink/port/port.go | 3 +- pkg/sentry/socket/netlink/socket.go | 2 +- pkg/sentry/socket/netstack/BUILD | 1 + pkg/sentry/socket/netstack/netstack.go | 2 +- pkg/sentry/socket/rpcinet/conn/BUILD | 1 + pkg/sentry/socket/rpcinet/conn/conn.go | 2 +- pkg/sentry/socket/rpcinet/notifier/BUILD | 1 + pkg/sentry/socket/rpcinet/notifier/notifier.go | 2 +- pkg/sentry/socket/unix/transport/BUILD | 1 + pkg/sentry/socket/unix/transport/connectioned.go | 3 +- pkg/sentry/socket/unix/transport/queue.go | 3 +- pkg/sentry/socket/unix/transport/unix.go | 2 +- pkg/sentry/syscalls/linux/BUILD | 1 + pkg/sentry/syscalls/linux/error.go | 2 +- pkg/sentry/time/BUILD | 4 +- pkg/sentry/time/calibrated_clock.go | 2 +- pkg/sentry/usage/BUILD | 1 + pkg/sentry/usage/memory.go | 2 +- pkg/sentry/vfs/BUILD | 3 +- pkg/sentry/vfs/dentry.go | 2 +- pkg/sentry/vfs/file_description_impl_util.go | 2 +- pkg/sentry/vfs/mount_test.go | 3 +- pkg/sentry/vfs/mount_unsafe.go | 4 +- pkg/sentry/vfs/pathname.go | 3 +- pkg/sentry/vfs/resolving_path.go | 2 +- pkg/sentry/vfs/vfs.go | 2 +- pkg/sentry/watchdog/BUILD | 1 + pkg/sentry/watchdog/watchdog.go | 2 +- pkg/sync/BUILD | 53 +++++++ pkg/sync/LICENSE | 27 ++++ pkg/sync/README.md | 5 + pkg/sync/aliases.go | 37 +++++ pkg/sync/atomicptr_unsafe.go | 47 +++++++ pkg/sync/atomicptrtest/BUILD | 29 ++++ pkg/sync/atomicptrtest/atomicptr_test.go | 31 +++++ pkg/sync/downgradable_rwmutex_test.go | 150 ++++++++++++++++++++ pkg/sync/downgradable_rwmutex_unsafe.go | 146 ++++++++++++++++++++ pkg/sync/memmove_unsafe.go | 28 ++++ pkg/sync/norace_unsafe.go | 35 +++++ pkg/sync/race_unsafe.go | 41 ++++++ pkg/sync/seqatomic_unsafe.go | 72 ++++++++++ pkg/sync/seqatomictest/BUILD | 33 +++++ pkg/sync/seqatomictest/seqatomic_test.go | 132 ++++++++++++++++++ pkg/sync/seqcount.go | 149 ++++++++++++++++++++ pkg/sync/seqcount_test.go | 153 +++++++++++++++++++++ pkg/sync/syncutil.go | 7 + pkg/syncutil/BUILD | 52 ------- pkg/syncutil/LICENSE | 27 ---- pkg/syncutil/README.md | 5 - pkg/syncutil/atomicptr_unsafe.go | 47 ------- pkg/syncutil/atomicptrtest/BUILD | 29 ---- pkg/syncutil/atomicptrtest/atomicptr_test.go | 31 ----- pkg/syncutil/downgradable_rwmutex_test.go | 150 -------------------- pkg/syncutil/downgradable_rwmutex_unsafe.go | 146 -------------------- pkg/syncutil/memmove_unsafe.go | 28 ---- pkg/syncutil/norace_unsafe.go | 35 ----- pkg/syncutil/race_unsafe.go | 41 ------ pkg/syncutil/seqatomic_unsafe.go | 72 ---------- pkg/syncutil/seqatomictest/BUILD | 35 ----- pkg/syncutil/seqatomictest/seqatomic_test.go | 132 ------------------ pkg/syncutil/seqcount.go | 149 -------------------- pkg/syncutil/seqcount_test.go | 153 --------------------- pkg/syncutil/syncutil.go | 7 - pkg/tcpip/BUILD | 1 + pkg/tcpip/adapters/gonet/BUILD | 1 + pkg/tcpip/adapters/gonet/gonet.go | 2 +- pkg/tcpip/link/fdbased/BUILD | 1 + pkg/tcpip/link/fdbased/endpoint.go | 2 +- pkg/tcpip/link/sharedmem/BUILD | 2 + pkg/tcpip/link/sharedmem/pipe/BUILD | 1 + pkg/tcpip/link/sharedmem/pipe/pipe_test.go | 3 +- pkg/tcpip/link/sharedmem/sharedmem.go | 2 +- pkg/tcpip/link/sharedmem/sharedmem_test.go | 2 +- pkg/tcpip/network/fragmentation/BUILD | 1 + pkg/tcpip/network/fragmentation/fragmentation.go | 2 +- pkg/tcpip/network/fragmentation/reassembler.go | 2 +- pkg/tcpip/ports/BUILD | 1 + pkg/tcpip/ports/ports.go | 2 +- pkg/tcpip/stack/BUILD | 2 + pkg/tcpip/stack/linkaddrcache.go | 2 +- pkg/tcpip/stack/linkaddrcache_test.go | 2 +- pkg/tcpip/stack/nic.go | 2 +- pkg/tcpip/stack/stack.go | 2 +- pkg/tcpip/stack/transport_demuxer.go | 2 +- pkg/tcpip/tcpip.go | 2 +- pkg/tcpip/transport/icmp/BUILD | 1 + pkg/tcpip/transport/icmp/endpoint.go | 3 +- pkg/tcpip/transport/packet/BUILD | 1 + pkg/tcpip/transport/packet/endpoint.go | 3 +- pkg/tcpip/transport/raw/BUILD | 1 + pkg/tcpip/transport/raw/endpoint.go | 3 +- pkg/tcpip/transport/tcp/BUILD | 1 + pkg/tcpip/transport/tcp/accept.go | 2 +- pkg/tcpip/transport/tcp/connect.go | 2 +- pkg/tcpip/transport/tcp/endpoint.go | 2 +- pkg/tcpip/transport/tcp/endpoint_state.go | 2 +- pkg/tcpip/transport/tcp/forwarder.go | 3 +- pkg/tcpip/transport/tcp/protocol.go | 2 +- pkg/tcpip/transport/tcp/segment_queue.go | 2 +- pkg/tcpip/transport/tcp/snd.go | 2 +- pkg/tcpip/transport/udp/BUILD | 1 + pkg/tcpip/transport/udp/endpoint.go | 3 +- pkg/tmutex/BUILD | 1 + pkg/tmutex/tmutex_test.go | 3 +- pkg/unet/BUILD | 1 + pkg/unet/unet_test.go | 3 +- pkg/urpc/BUILD | 1 + pkg/urpc/urpc.go | 2 +- pkg/waiter/BUILD | 1 + pkg/waiter/waiter.go | 2 +- runsc/boot/BUILD | 2 + runsc/boot/compat.go | 2 +- runsc/boot/limits.go | 2 +- runsc/boot/loader.go | 2 +- runsc/boot/loader_test.go | 2 +- runsc/cmd/BUILD | 1 + runsc/cmd/create.go | 1 + runsc/cmd/gofer.go | 2 +- runsc/cmd/start.go | 1 + runsc/container/BUILD | 2 + runsc/container/console_test.go | 2 +- runsc/container/container_test.go | 2 +- runsc/container/multi_container_test.go | 2 +- runsc/container/state_file.go | 2 +- runsc/fsgofer/BUILD | 1 + runsc/fsgofer/fsgofer.go | 2 +- runsc/sandbox/BUILD | 1 + runsc/sandbox/sandbox.go | 2 +- runsc/testutil/BUILD | 1 + runsc/testutil/testutil.go | 2 +- 303 files changed, 1507 insertions(+), 1368 deletions(-) create mode 100644 pkg/sync/BUILD create mode 100644 pkg/sync/LICENSE create mode 100644 pkg/sync/README.md create mode 100644 pkg/sync/aliases.go create mode 100644 pkg/sync/atomicptr_unsafe.go create mode 100644 pkg/sync/atomicptrtest/BUILD create mode 100644 pkg/sync/atomicptrtest/atomicptr_test.go create mode 100644 pkg/sync/downgradable_rwmutex_test.go create mode 100644 pkg/sync/downgradable_rwmutex_unsafe.go create mode 100644 pkg/sync/memmove_unsafe.go create mode 100644 pkg/sync/norace_unsafe.go create mode 100644 pkg/sync/race_unsafe.go create mode 100644 pkg/sync/seqatomic_unsafe.go create mode 100644 pkg/sync/seqatomictest/BUILD create mode 100644 pkg/sync/seqatomictest/seqatomic_test.go create mode 100644 pkg/sync/seqcount.go create mode 100644 pkg/sync/seqcount_test.go create mode 100644 pkg/sync/syncutil.go delete mode 100644 pkg/syncutil/BUILD delete mode 100644 pkg/syncutil/LICENSE delete mode 100644 pkg/syncutil/README.md delete mode 100644 pkg/syncutil/atomicptr_unsafe.go delete mode 100644 pkg/syncutil/atomicptrtest/BUILD delete mode 100644 pkg/syncutil/atomicptrtest/atomicptr_test.go delete mode 100644 pkg/syncutil/downgradable_rwmutex_test.go delete mode 100644 pkg/syncutil/downgradable_rwmutex_unsafe.go delete mode 100644 pkg/syncutil/memmove_unsafe.go delete mode 100644 pkg/syncutil/norace_unsafe.go delete mode 100644 pkg/syncutil/race_unsafe.go delete mode 100644 pkg/syncutil/seqatomic_unsafe.go delete mode 100644 pkg/syncutil/seqatomictest/BUILD delete mode 100644 pkg/syncutil/seqatomictest/seqatomic_test.go delete mode 100644 pkg/syncutil/seqcount.go delete mode 100644 pkg/syncutil/seqcount_test.go delete mode 100644 pkg/syncutil/syncutil.go (limited to 'pkg/sentry') diff --git a/pkg/amutex/BUILD b/pkg/amutex/BUILD index 6bc486b62..d99e37b40 100644 --- a/pkg/amutex/BUILD +++ b/pkg/amutex/BUILD @@ -15,4 +15,5 @@ go_test( size = "small", srcs = ["amutex_test.go"], embed = [":amutex"], + deps = ["//pkg/sync"], ) diff --git a/pkg/amutex/amutex_test.go b/pkg/amutex/amutex_test.go index 1d7f45641..8a3952f2a 100644 --- a/pkg/amutex/amutex_test.go +++ b/pkg/amutex/amutex_test.go @@ -15,9 +15,10 @@ package amutex import ( - "sync" "testing" "time" + + "gvisor.dev/gvisor/pkg/sync" ) type sleeper struct { diff --git a/pkg/atomicbitops/BUILD b/pkg/atomicbitops/BUILD index 36beaade9..6403c60c2 100644 --- a/pkg/atomicbitops/BUILD +++ b/pkg/atomicbitops/BUILD @@ -20,4 +20,5 @@ go_test( size = "small", srcs = ["atomic_bitops_test.go"], embed = [":atomicbitops"], + deps = ["//pkg/sync"], ) diff --git a/pkg/atomicbitops/atomic_bitops_test.go b/pkg/atomicbitops/atomic_bitops_test.go index 965e9be79..9466d3e23 100644 --- a/pkg/atomicbitops/atomic_bitops_test.go +++ b/pkg/atomicbitops/atomic_bitops_test.go @@ -16,8 +16,9 @@ package atomicbitops import ( "runtime" - "sync" "testing" + + "gvisor.dev/gvisor/pkg/sync" ) const iterations = 100 diff --git a/pkg/compressio/BUILD b/pkg/compressio/BUILD index a0b21d4bd..2bb581b18 100644 --- a/pkg/compressio/BUILD +++ b/pkg/compressio/BUILD @@ -8,7 +8,10 @@ go_library( srcs = ["compressio.go"], importpath = "gvisor.dev/gvisor/pkg/compressio", visibility = ["//:sandbox"], - deps = ["//pkg/binary"], + deps = [ + "//pkg/binary", + "//pkg/sync", + ], ) go_test( diff --git a/pkg/compressio/compressio.go b/pkg/compressio/compressio.go index 3b0bb086e..5f52cbe74 100644 --- a/pkg/compressio/compressio.go +++ b/pkg/compressio/compressio.go @@ -52,9 +52,9 @@ import ( "hash" "io" "runtime" - "sync" "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/sync" ) var bufPool = sync.Pool{ diff --git a/pkg/control/server/BUILD b/pkg/control/server/BUILD index 21adf3adf..adbd1e3f8 100644 --- a/pkg/control/server/BUILD +++ b/pkg/control/server/BUILD @@ -9,6 +9,7 @@ go_library( visibility = ["//:sandbox"], deps = [ "//pkg/log", + "//pkg/sync", "//pkg/unet", "//pkg/urpc", ], diff --git a/pkg/control/server/server.go b/pkg/control/server/server.go index a56152d10..41abe1f2d 100644 --- a/pkg/control/server/server.go +++ b/pkg/control/server/server.go @@ -22,9 +22,9 @@ package server import ( "os" - "sync" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/unet" "gvisor.dev/gvisor/pkg/urpc" ) diff --git a/pkg/eventchannel/BUILD b/pkg/eventchannel/BUILD index 0b4b7cc44..9d68682c7 100644 --- a/pkg/eventchannel/BUILD +++ b/pkg/eventchannel/BUILD @@ -15,6 +15,7 @@ go_library( deps = [ ":eventchannel_go_proto", "//pkg/log", + "//pkg/sync", "//pkg/unet", "@com_github_golang_protobuf//proto:go_default_library", "@com_github_golang_protobuf//ptypes:go_default_library_gen", @@ -40,6 +41,7 @@ go_test( srcs = ["event_test.go"], embed = [":eventchannel"], deps = [ + "//pkg/sync", "@com_github_golang_protobuf//proto:go_default_library", ], ) diff --git a/pkg/eventchannel/event.go b/pkg/eventchannel/event.go index d37ad0428..9a29c58bd 100644 --- a/pkg/eventchannel/event.go +++ b/pkg/eventchannel/event.go @@ -22,13 +22,13 @@ package eventchannel import ( "encoding/binary" "fmt" - "sync" "syscall" "github.com/golang/protobuf/proto" "github.com/golang/protobuf/ptypes" pb "gvisor.dev/gvisor/pkg/eventchannel/eventchannel_go_proto" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/unet" ) diff --git a/pkg/eventchannel/event_test.go b/pkg/eventchannel/event_test.go index 3649097d6..7f41b4a27 100644 --- a/pkg/eventchannel/event_test.go +++ b/pkg/eventchannel/event_test.go @@ -16,11 +16,11 @@ package eventchannel import ( "fmt" - "sync" "testing" "time" "github.com/golang/protobuf/proto" + "gvisor.dev/gvisor/pkg/sync" ) // testEmitter is an emitter that can be used in tests. It records all events diff --git a/pkg/fdchannel/BUILD b/pkg/fdchannel/BUILD index 56495cbd9..b0478c672 100644 --- a/pkg/fdchannel/BUILD +++ b/pkg/fdchannel/BUILD @@ -15,4 +15,5 @@ go_test( size = "small", srcs = ["fdchannel_test.go"], embed = [":fdchannel"], + deps = ["//pkg/sync"], ) diff --git a/pkg/fdchannel/fdchannel_test.go b/pkg/fdchannel/fdchannel_test.go index 5d01dc636..7a8a63a59 100644 --- a/pkg/fdchannel/fdchannel_test.go +++ b/pkg/fdchannel/fdchannel_test.go @@ -17,10 +17,11 @@ package fdchannel import ( "io/ioutil" "os" - "sync" "syscall" "testing" "time" + + "gvisor.dev/gvisor/pkg/sync" ) func TestSendRecvFD(t *testing.T) { diff --git a/pkg/fdnotifier/BUILD b/pkg/fdnotifier/BUILD index aca2d8a82..91a202a30 100644 --- a/pkg/fdnotifier/BUILD +++ b/pkg/fdnotifier/BUILD @@ -11,6 +11,7 @@ go_library( importpath = "gvisor.dev/gvisor/pkg/fdnotifier", visibility = ["//:sandbox"], deps = [ + "//pkg/sync", "//pkg/waiter", "@org_golang_x_sys//unix:go_default_library", ], diff --git a/pkg/fdnotifier/fdnotifier.go b/pkg/fdnotifier/fdnotifier.go index f4aae1953..a6b63c982 100644 --- a/pkg/fdnotifier/fdnotifier.go +++ b/pkg/fdnotifier/fdnotifier.go @@ -22,10 +22,10 @@ package fdnotifier import ( "fmt" - "sync" "syscall" "golang.org/x/sys/unix" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/flipcall/BUILD b/pkg/flipcall/BUILD index e590a71ba..85bd83af1 100644 --- a/pkg/flipcall/BUILD +++ b/pkg/flipcall/BUILD @@ -19,7 +19,7 @@ go_library( "//pkg/abi/linux", "//pkg/log", "//pkg/memutil", - "//pkg/syncutil", + "//pkg/sync", ], ) @@ -31,4 +31,5 @@ go_test( "flipcall_test.go", ], embed = [":flipcall"], + deps = ["//pkg/sync"], ) diff --git a/pkg/flipcall/flipcall_example_test.go b/pkg/flipcall/flipcall_example_test.go index 8d88b845d..2e28a149a 100644 --- a/pkg/flipcall/flipcall_example_test.go +++ b/pkg/flipcall/flipcall_example_test.go @@ -17,7 +17,8 @@ package flipcall import ( "bytes" "fmt" - "sync" + + "gvisor.dev/gvisor/pkg/sync" ) func Example() { diff --git a/pkg/flipcall/flipcall_test.go b/pkg/flipcall/flipcall_test.go index 168a487ec..33fd55a44 100644 --- a/pkg/flipcall/flipcall_test.go +++ b/pkg/flipcall/flipcall_test.go @@ -16,9 +16,10 @@ package flipcall import ( "runtime" - "sync" "testing" "time" + + "gvisor.dev/gvisor/pkg/sync" ) var testPacketWindowSize = pageSize diff --git a/pkg/flipcall/flipcall_unsafe.go b/pkg/flipcall/flipcall_unsafe.go index 27b8939fc..ac974b232 100644 --- a/pkg/flipcall/flipcall_unsafe.go +++ b/pkg/flipcall/flipcall_unsafe.go @@ -18,7 +18,7 @@ import ( "reflect" "unsafe" - "gvisor.dev/gvisor/pkg/syncutil" + "gvisor.dev/gvisor/pkg/sync" ) // Packets consist of a 16-byte header followed by an arbitrarily-sized @@ -75,13 +75,13 @@ func (ep *Endpoint) Data() []byte { var ioSync int64 func raceBecomeActive() { - if syncutil.RaceEnabled { - syncutil.RaceAcquire((unsafe.Pointer)(&ioSync)) + if sync.RaceEnabled { + sync.RaceAcquire((unsafe.Pointer)(&ioSync)) } } func raceBecomeInactive() { - if syncutil.RaceEnabled { - syncutil.RaceReleaseMerge((unsafe.Pointer)(&ioSync)) + if sync.RaceEnabled { + sync.RaceReleaseMerge((unsafe.Pointer)(&ioSync)) } } diff --git a/pkg/gate/BUILD b/pkg/gate/BUILD index 4b9321711..f22bd070d 100644 --- a/pkg/gate/BUILD +++ b/pkg/gate/BUILD @@ -19,5 +19,6 @@ go_test( ], deps = [ ":gate", + "//pkg/sync", ], ) diff --git a/pkg/gate/gate_test.go b/pkg/gate/gate_test.go index 5dbd8d712..850693df8 100644 --- a/pkg/gate/gate_test.go +++ b/pkg/gate/gate_test.go @@ -15,11 +15,11 @@ package gate_test import ( - "sync" "testing" "time" "gvisor.dev/gvisor/pkg/gate" + "gvisor.dev/gvisor/pkg/sync" ) func TestBasicEnter(t *testing.T) { diff --git a/pkg/linewriter/BUILD b/pkg/linewriter/BUILD index a5d980d14..bcde6d308 100644 --- a/pkg/linewriter/BUILD +++ b/pkg/linewriter/BUILD @@ -8,6 +8,7 @@ go_library( srcs = ["linewriter.go"], importpath = "gvisor.dev/gvisor/pkg/linewriter", visibility = ["//visibility:public"], + deps = ["//pkg/sync"], ) go_test( diff --git a/pkg/linewriter/linewriter.go b/pkg/linewriter/linewriter.go index cd6e4e2ce..a1b1285d4 100644 --- a/pkg/linewriter/linewriter.go +++ b/pkg/linewriter/linewriter.go @@ -17,7 +17,8 @@ package linewriter import ( "bytes" - "sync" + + "gvisor.dev/gvisor/pkg/sync" ) // Writer is an io.Writer which buffers input, flushing diff --git a/pkg/log/BUILD b/pkg/log/BUILD index fc5f5779b..0df0f2849 100644 --- a/pkg/log/BUILD +++ b/pkg/log/BUILD @@ -16,7 +16,10 @@ go_library( visibility = [ "//visibility:public", ], - deps = ["//pkg/linewriter"], + deps = [ + "//pkg/linewriter", + "//pkg/sync", + ], ) go_test( diff --git a/pkg/log/log.go b/pkg/log/log.go index 9387586e6..91a81b288 100644 --- a/pkg/log/log.go +++ b/pkg/log/log.go @@ -25,12 +25,12 @@ import ( stdlog "log" "os" "runtime" - "sync" "sync/atomic" "syscall" "time" "gvisor.dev/gvisor/pkg/linewriter" + "gvisor.dev/gvisor/pkg/sync" ) // Level is the log level. diff --git a/pkg/metric/BUILD b/pkg/metric/BUILD index dd6ca6d39..9145f3233 100644 --- a/pkg/metric/BUILD +++ b/pkg/metric/BUILD @@ -14,6 +14,7 @@ go_library( ":metric_go_proto", "//pkg/eventchannel", "//pkg/log", + "//pkg/sync", ], ) diff --git a/pkg/metric/metric.go b/pkg/metric/metric.go index eadde06e4..93d4f2b8c 100644 --- a/pkg/metric/metric.go +++ b/pkg/metric/metric.go @@ -18,12 +18,12 @@ package metric import ( "errors" "fmt" - "sync" "sync/atomic" "gvisor.dev/gvisor/pkg/eventchannel" "gvisor.dev/gvisor/pkg/log" pb "gvisor.dev/gvisor/pkg/metric/metric_go_proto" + "gvisor.dev/gvisor/pkg/sync" ) var ( diff --git a/pkg/p9/BUILD b/pkg/p9/BUILD index f32244c69..a3e05c96d 100644 --- a/pkg/p9/BUILD +++ b/pkg/p9/BUILD @@ -29,6 +29,7 @@ go_library( "//pkg/fdchannel", "//pkg/flipcall", "//pkg/log", + "//pkg/sync", "//pkg/unet", "@org_golang_x_sys//unix:go_default_library", ], diff --git a/pkg/p9/client.go b/pkg/p9/client.go index 221516c6c..4045e41fa 100644 --- a/pkg/p9/client.go +++ b/pkg/p9/client.go @@ -17,12 +17,12 @@ package p9 import ( "errors" "fmt" - "sync" "syscall" "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/flipcall" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/unet" ) diff --git a/pkg/p9/p9test/BUILD b/pkg/p9/p9test/BUILD index 28707c0ca..f4edd68b2 100644 --- a/pkg/p9/p9test/BUILD +++ b/pkg/p9/p9test/BUILD @@ -70,6 +70,7 @@ go_library( "//pkg/fd", "//pkg/log", "//pkg/p9", + "//pkg/sync", "//pkg/unet", "@com_github_golang_mock//gomock:go_default_library", ], @@ -83,6 +84,7 @@ go_test( deps = [ "//pkg/fd", "//pkg/p9", + "//pkg/sync", "@com_github_golang_mock//gomock:go_default_library", ], ) diff --git a/pkg/p9/p9test/client_test.go b/pkg/p9/p9test/client_test.go index 6e758148d..6e7bb3db2 100644 --- a/pkg/p9/p9test/client_test.go +++ b/pkg/p9/p9test/client_test.go @@ -22,7 +22,6 @@ import ( "os" "reflect" "strings" - "sync" "syscall" "testing" "time" @@ -30,6 +29,7 @@ import ( "github.com/golang/mock/gomock" "gvisor.dev/gvisor/pkg/fd" "gvisor.dev/gvisor/pkg/p9" + "gvisor.dev/gvisor/pkg/sync" ) func TestPanic(t *testing.T) { diff --git a/pkg/p9/p9test/p9test.go b/pkg/p9/p9test/p9test.go index 4d3271b37..dd8b01b6d 100644 --- a/pkg/p9/p9test/p9test.go +++ b/pkg/p9/p9test/p9test.go @@ -17,13 +17,13 @@ package p9test import ( "fmt" - "sync" "sync/atomic" "syscall" "testing" "github.com/golang/mock/gomock" "gvisor.dev/gvisor/pkg/p9" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/unet" ) diff --git a/pkg/p9/path_tree.go b/pkg/p9/path_tree.go index 865459411..72ef53313 100644 --- a/pkg/p9/path_tree.go +++ b/pkg/p9/path_tree.go @@ -16,7 +16,8 @@ package p9 import ( "fmt" - "sync" + + "gvisor.dev/gvisor/pkg/sync" ) // pathNode is a single node in a path traversal. diff --git a/pkg/p9/pool.go b/pkg/p9/pool.go index 52de889e1..2b14a5ce3 100644 --- a/pkg/p9/pool.go +++ b/pkg/p9/pool.go @@ -15,7 +15,7 @@ package p9 import ( - "sync" + "gvisor.dev/gvisor/pkg/sync" ) // pool is a simple allocator. diff --git a/pkg/p9/server.go b/pkg/p9/server.go index 40b8fa023..fdfa83648 100644 --- a/pkg/p9/server.go +++ b/pkg/p9/server.go @@ -17,7 +17,6 @@ package p9 import ( "io" "runtime/debug" - "sync" "sync/atomic" "syscall" @@ -25,6 +24,7 @@ import ( "gvisor.dev/gvisor/pkg/fdchannel" "gvisor.dev/gvisor/pkg/flipcall" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/unet" ) diff --git a/pkg/p9/transport.go b/pkg/p9/transport.go index 6e8b4bbcd..9c11e28ce 100644 --- a/pkg/p9/transport.go +++ b/pkg/p9/transport.go @@ -19,11 +19,11 @@ import ( "fmt" "io" "io/ioutil" - "sync" "syscall" "gvisor.dev/gvisor/pkg/fd" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/unet" ) diff --git a/pkg/procid/BUILD b/pkg/procid/BUILD index 078f084b2..b506813f0 100644 --- a/pkg/procid/BUILD +++ b/pkg/procid/BUILD @@ -21,6 +21,7 @@ go_test( "procid_test.go", ], embed = [":procid"], + deps = ["//pkg/sync"], ) go_test( @@ -31,4 +32,5 @@ go_test( "procid_test.go", ], embed = [":procid"], + deps = ["//pkg/sync"], ) diff --git a/pkg/procid/procid_test.go b/pkg/procid/procid_test.go index 88dd0b3ae..9ec08c3d6 100644 --- a/pkg/procid/procid_test.go +++ b/pkg/procid/procid_test.go @@ -17,9 +17,10 @@ package procid import ( "os" "runtime" - "sync" "syscall" "testing" + + "gvisor.dev/gvisor/pkg/sync" ) // runOnMain is used to send functions to run on the main (initial) thread. diff --git a/pkg/rand/BUILD b/pkg/rand/BUILD index f4f2001f3..9d5b4859b 100644 --- a/pkg/rand/BUILD +++ b/pkg/rand/BUILD @@ -10,5 +10,8 @@ go_library( ], importpath = "gvisor.dev/gvisor/pkg/rand", visibility = ["//:sandbox"], - deps = ["@org_golang_x_sys//unix:go_default_library"], + deps = [ + "//pkg/sync", + "@org_golang_x_sys//unix:go_default_library", + ], ) diff --git a/pkg/rand/rand_linux.go b/pkg/rand/rand_linux.go index 2b92db3e6..0bdad5fad 100644 --- a/pkg/rand/rand_linux.go +++ b/pkg/rand/rand_linux.go @@ -19,9 +19,9 @@ package rand import ( "crypto/rand" "io" - "sync" "golang.org/x/sys/unix" + "gvisor.dev/gvisor/pkg/sync" ) // reader implements an io.Reader that returns pseudorandom bytes. diff --git a/pkg/refs/BUILD b/pkg/refs/BUILD index 7ad59dfd7..974d9af9b 100644 --- a/pkg/refs/BUILD +++ b/pkg/refs/BUILD @@ -27,6 +27,7 @@ go_library( visibility = ["//:sandbox"], deps = [ "//pkg/log", + "//pkg/sync", ], ) @@ -35,4 +36,5 @@ go_test( size = "small", srcs = ["refcounter_test.go"], embed = [":refs"], + deps = ["//pkg/sync"], ) diff --git a/pkg/refs/refcounter.go b/pkg/refs/refcounter.go index ad69e0757..c45ba8200 100644 --- a/pkg/refs/refcounter.go +++ b/pkg/refs/refcounter.go @@ -21,10 +21,10 @@ import ( "fmt" "reflect" "runtime" - "sync" "sync/atomic" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/sync" ) // RefCounter is the interface to be implemented by objects that are reference diff --git a/pkg/refs/refcounter_test.go b/pkg/refs/refcounter_test.go index ffd3d3f07..1ab4a4440 100644 --- a/pkg/refs/refcounter_test.go +++ b/pkg/refs/refcounter_test.go @@ -16,8 +16,9 @@ package refs import ( "reflect" - "sync" "testing" + + "gvisor.dev/gvisor/pkg/sync" ) type testCounter struct { diff --git a/pkg/sentry/arch/BUILD b/pkg/sentry/arch/BUILD index 18c73cc24..ae3e364cd 100644 --- a/pkg/sentry/arch/BUILD +++ b/pkg/sentry/arch/BUILD @@ -32,6 +32,7 @@ go_library( "//pkg/sentry/context", "//pkg/sentry/limits", "//pkg/sentry/usermem", + "//pkg/sync", "//pkg/syserror", ], ) diff --git a/pkg/sentry/arch/arch_x86.go b/pkg/sentry/arch/arch_x86.go index 9294ac773..9f41e566f 100644 --- a/pkg/sentry/arch/arch_x86.go +++ b/pkg/sentry/arch/arch_x86.go @@ -19,7 +19,6 @@ package arch import ( "fmt" "io" - "sync" "syscall" "gvisor.dev/gvisor/pkg/binary" @@ -27,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/log" rpb "gvisor.dev/gvisor/pkg/sentry/arch/registers_go_proto" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/control/BUILD b/pkg/sentry/control/BUILD index 5522cecd0..2561a6109 100644 --- a/pkg/sentry/control/BUILD +++ b/pkg/sentry/control/BUILD @@ -30,6 +30,7 @@ go_library( "//pkg/sentry/strace", "//pkg/sentry/usage", "//pkg/sentry/watchdog", + "//pkg/sync", "//pkg/tcpip/link/sniffer", "//pkg/urpc", ], diff --git a/pkg/sentry/control/pprof.go b/pkg/sentry/control/pprof.go index e1f2fea60..151808911 100644 --- a/pkg/sentry/control/pprof.go +++ b/pkg/sentry/control/pprof.go @@ -19,10 +19,10 @@ import ( "runtime" "runtime/pprof" "runtime/trace" - "sync" "gvisor.dev/gvisor/pkg/fd" "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/urpc" ) diff --git a/pkg/sentry/device/BUILD b/pkg/sentry/device/BUILD index 1098ed777..97fa1512c 100644 --- a/pkg/sentry/device/BUILD +++ b/pkg/sentry/device/BUILD @@ -8,7 +8,10 @@ go_library( srcs = ["device.go"], importpath = "gvisor.dev/gvisor/pkg/sentry/device", visibility = ["//pkg/sentry:internal"], - deps = ["//pkg/abi/linux"], + deps = [ + "//pkg/abi/linux", + "//pkg/sync", + ], ) go_test( diff --git a/pkg/sentry/device/device.go b/pkg/sentry/device/device.go index 47945d1a7..69e71e322 100644 --- a/pkg/sentry/device/device.go +++ b/pkg/sentry/device/device.go @@ -19,10 +19,10 @@ package device import ( "bytes" "fmt" - "sync" "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sync" ) // Registry tracks all simple devices and related state on the system for diff --git a/pkg/sentry/fs/BUILD b/pkg/sentry/fs/BUILD index c035ffff7..7d5d72d5a 100644 --- a/pkg/sentry/fs/BUILD +++ b/pkg/sentry/fs/BUILD @@ -68,7 +68,7 @@ go_library( "//pkg/sentry/usage", "//pkg/sentry/usermem", "//pkg/state", - "//pkg/syncutil", + "//pkg/sync", "//pkg/syserror", "//pkg/waiter", ], @@ -115,6 +115,7 @@ go_test( "//pkg/sentry/fs/tmpfs", "//pkg/sentry/kernel/contexttest", "//pkg/sentry/usermem", + "//pkg/sync", "//pkg/syserror", ], ) diff --git a/pkg/sentry/fs/copy_up.go b/pkg/sentry/fs/copy_up.go index 9ac62c84d..734177e90 100644 --- a/pkg/sentry/fs/copy_up.go +++ b/pkg/sentry/fs/copy_up.go @@ -17,12 +17,12 @@ package fs import ( "fmt" "io" - "sync" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/fs/copy_up_test.go b/pkg/sentry/fs/copy_up_test.go index 1d80bf15a..738580c5f 100644 --- a/pkg/sentry/fs/copy_up_test.go +++ b/pkg/sentry/fs/copy_up_test.go @@ -19,13 +19,13 @@ import ( "crypto/rand" "fmt" "io" - "sync" "testing" "gvisor.dev/gvisor/pkg/sentry/fs" _ "gvisor.dev/gvisor/pkg/sentry/fs/tmpfs" "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" ) const ( diff --git a/pkg/sentry/fs/dirent.go b/pkg/sentry/fs/dirent.go index 3cb73bd78..31fc4d87b 100644 --- a/pkg/sentry/fs/dirent.go +++ b/pkg/sentry/fs/dirent.go @@ -18,7 +18,6 @@ import ( "fmt" "path" "sort" - "sync" "sync/atomic" "syscall" @@ -28,6 +27,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/uniqueid" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/fs/dirent_cache.go b/pkg/sentry/fs/dirent_cache.go index 60a15a275..25514ace4 100644 --- a/pkg/sentry/fs/dirent_cache.go +++ b/pkg/sentry/fs/dirent_cache.go @@ -16,7 +16,8 @@ package fs import ( "fmt" - "sync" + + "gvisor.dev/gvisor/pkg/sync" ) // DirentCache is an LRU cache of Dirents. The Dirent's refCount is diff --git a/pkg/sentry/fs/dirent_cache_limiter.go b/pkg/sentry/fs/dirent_cache_limiter.go index ebb80bd50..525ee25f9 100644 --- a/pkg/sentry/fs/dirent_cache_limiter.go +++ b/pkg/sentry/fs/dirent_cache_limiter.go @@ -16,7 +16,8 @@ package fs import ( "fmt" - "sync" + + "gvisor.dev/gvisor/pkg/sync" ) // DirentCacheLimiter acts as a global limit for all dirent caches in the diff --git a/pkg/sentry/fs/fdpipe/BUILD b/pkg/sentry/fs/fdpipe/BUILD index 277ee4c31..cc43de69d 100644 --- a/pkg/sentry/fs/fdpipe/BUILD +++ b/pkg/sentry/fs/fdpipe/BUILD @@ -23,6 +23,7 @@ go_library( "//pkg/sentry/fs/fsutil", "//pkg/sentry/safemem", "//pkg/sentry/usermem", + "//pkg/sync", "//pkg/syserror", "//pkg/waiter", ], diff --git a/pkg/sentry/fs/fdpipe/pipe.go b/pkg/sentry/fs/fdpipe/pipe.go index 669ffcb75..5b6cfeb0a 100644 --- a/pkg/sentry/fs/fdpipe/pipe.go +++ b/pkg/sentry/fs/fdpipe/pipe.go @@ -17,7 +17,6 @@ package fdpipe import ( "os" - "sync" "syscall" "gvisor.dev/gvisor/pkg/fd" @@ -29,6 +28,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/fdpipe/pipe_state.go b/pkg/sentry/fs/fdpipe/pipe_state.go index 29175fb3d..cee87f726 100644 --- a/pkg/sentry/fs/fdpipe/pipe_state.go +++ b/pkg/sentry/fs/fdpipe/pipe_state.go @@ -17,10 +17,10 @@ package fdpipe import ( "fmt" "io/ioutil" - "sync" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sync" ) // beforeSave is invoked by stateify. diff --git a/pkg/sentry/fs/file.go b/pkg/sentry/fs/file.go index a2f966cb6..7c4586296 100644 --- a/pkg/sentry/fs/file.go +++ b/pkg/sentry/fs/file.go @@ -16,7 +16,6 @@ package fs import ( "math" - "sync" "sync/atomic" "time" @@ -29,6 +28,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/uniqueid" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/file_overlay.go b/pkg/sentry/fs/file_overlay.go index 225e40186..8a633b1ba 100644 --- a/pkg/sentry/fs/file_overlay.go +++ b/pkg/sentry/fs/file_overlay.go @@ -16,13 +16,13 @@ package fs import ( "io" - "sync" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/filesystems.go b/pkg/sentry/fs/filesystems.go index b157fd228..c5b51620a 100644 --- a/pkg/sentry/fs/filesystems.go +++ b/pkg/sentry/fs/filesystems.go @@ -18,9 +18,9 @@ import ( "fmt" "sort" "strings" - "sync" "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sync" ) // FilesystemFlags matches include/linux/fs.h:file_system_type.fs_flags. diff --git a/pkg/sentry/fs/fs.go b/pkg/sentry/fs/fs.go index 8b2a5e6b2..26abf49e2 100644 --- a/pkg/sentry/fs/fs.go +++ b/pkg/sentry/fs/fs.go @@ -54,10 +54,9 @@ package fs import ( - "sync" - "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sync" ) var ( diff --git a/pkg/sentry/fs/fsutil/BUILD b/pkg/sentry/fs/fsutil/BUILD index 9ca695a95..945b6270d 100644 --- a/pkg/sentry/fs/fsutil/BUILD +++ b/pkg/sentry/fs/fsutil/BUILD @@ -93,6 +93,7 @@ go_library( "//pkg/sentry/usage", "//pkg/sentry/usermem", "//pkg/state", + "//pkg/sync", "//pkg/syserror", "//pkg/waiter", ], diff --git a/pkg/sentry/fs/fsutil/host_file_mapper.go b/pkg/sentry/fs/fsutil/host_file_mapper.go index b06a71cc2..837fc70b5 100644 --- a/pkg/sentry/fs/fsutil/host_file_mapper.go +++ b/pkg/sentry/fs/fsutil/host_file_mapper.go @@ -16,7 +16,6 @@ package fsutil import ( "fmt" - "sync" "syscall" "gvisor.dev/gvisor/pkg/log" @@ -24,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" ) // HostFileMapper caches mappings of an arbitrary host file descriptor. It is diff --git a/pkg/sentry/fs/fsutil/host_mappable.go b/pkg/sentry/fs/fsutil/host_mappable.go index 30475f340..a625f0e26 100644 --- a/pkg/sentry/fs/fsutil/host_mappable.go +++ b/pkg/sentry/fs/fsutil/host_mappable.go @@ -16,7 +16,6 @@ package fsutil import ( "math" - "sync" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" @@ -24,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" ) // HostMappable implements memmap.Mappable and platform.File over a diff --git a/pkg/sentry/fs/fsutil/inode.go b/pkg/sentry/fs/fsutil/inode.go index 4e100a402..adf5ec69c 100644 --- a/pkg/sentry/fs/fsutil/inode.go +++ b/pkg/sentry/fs/fsutil/inode.go @@ -15,13 +15,12 @@ package fsutil import ( - "sync" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/fsutil/inode_cached.go b/pkg/sentry/fs/fsutil/inode_cached.go index 798920d18..20a014402 100644 --- a/pkg/sentry/fs/fsutil/inode_cached.go +++ b/pkg/sentry/fs/fsutil/inode_cached.go @@ -17,7 +17,6 @@ package fsutil import ( "fmt" "io" - "sync" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/context" @@ -30,6 +29,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" ) // Lock order (compare the lock order model in mm/mm.go): diff --git a/pkg/sentry/fs/gofer/BUILD b/pkg/sentry/fs/gofer/BUILD index 4a005c605..fd870e8e1 100644 --- a/pkg/sentry/fs/gofer/BUILD +++ b/pkg/sentry/fs/gofer/BUILD @@ -44,6 +44,7 @@ go_library( "//pkg/sentry/safemem", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usermem", + "//pkg/sync", "//pkg/syserr", "//pkg/syserror", "//pkg/unet", diff --git a/pkg/sentry/fs/gofer/inode.go b/pkg/sentry/fs/gofer/inode.go index 91263ebdc..245fe2ef1 100644 --- a/pkg/sentry/fs/gofer/inode.go +++ b/pkg/sentry/fs/gofer/inode.go @@ -16,7 +16,6 @@ package gofer import ( "errors" - "sync" "syscall" "gvisor.dev/gvisor/pkg/abi/linux" @@ -31,6 +30,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs/host" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/safemem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/fs/gofer/session.go b/pkg/sentry/fs/gofer/session.go index 4e358a46a..edc796ce0 100644 --- a/pkg/sentry/fs/gofer/session.go +++ b/pkg/sentry/fs/gofer/session.go @@ -16,7 +16,6 @@ package gofer import ( "fmt" - "sync" "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/refs" @@ -25,6 +24,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/unet" ) diff --git a/pkg/sentry/fs/host/BUILD b/pkg/sentry/fs/host/BUILD index 23daeb528..2b581aa69 100644 --- a/pkg/sentry/fs/host/BUILD +++ b/pkg/sentry/fs/host/BUILD @@ -50,6 +50,7 @@ go_library( "//pkg/sentry/unimpl", "//pkg/sentry/uniqueid", "//pkg/sentry/usermem", + "//pkg/sync", "//pkg/syserr", "//pkg/syserror", "//pkg/tcpip", diff --git a/pkg/sentry/fs/host/inode.go b/pkg/sentry/fs/host/inode.go index a6e4a09e3..873a1c52d 100644 --- a/pkg/sentry/fs/host/inode.go +++ b/pkg/sentry/fs/host/inode.go @@ -15,7 +15,6 @@ package host import ( - "sync" "syscall" "gvisor.dev/gvisor/pkg/abi/linux" @@ -28,6 +27,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/host/socket.go b/pkg/sentry/fs/host/socket.go index 107336a3e..c076d5bdd 100644 --- a/pkg/sentry/fs/host/socket.go +++ b/pkg/sentry/fs/host/socket.go @@ -16,7 +16,6 @@ package host import ( "fmt" - "sync" "syscall" "gvisor.dev/gvisor/pkg/abi/linux" @@ -30,6 +29,7 @@ import ( unixsocket "gvisor.dev/gvisor/pkg/sentry/socket/unix" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/uniqueid" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip" diff --git a/pkg/sentry/fs/host/tty.go b/pkg/sentry/fs/host/tty.go index 90331e3b2..753ef8cd6 100644 --- a/pkg/sentry/fs/host/tty.go +++ b/pkg/sentry/fs/host/tty.go @@ -15,8 +15,6 @@ package host import ( - "sync" - "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/context" @@ -24,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/unimpl" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/fs/inode.go b/pkg/sentry/fs/inode.go index 91e2fde2f..468043df0 100644 --- a/pkg/sentry/fs/inode.go +++ b/pkg/sentry/fs/inode.go @@ -15,8 +15,6 @@ package fs import ( - "sync" - "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/metric" @@ -26,6 +24,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/fs/inode_inotify.go b/pkg/sentry/fs/inode_inotify.go index 0f2a66a79..efd3c962b 100644 --- a/pkg/sentry/fs/inode_inotify.go +++ b/pkg/sentry/fs/inode_inotify.go @@ -16,7 +16,8 @@ package fs import ( "fmt" - "sync" + + "gvisor.dev/gvisor/pkg/sync" ) // Watches is the collection of inotify watches on an inode. diff --git a/pkg/sentry/fs/inotify.go b/pkg/sentry/fs/inotify.go index ba3e0233d..cc7dd1c92 100644 --- a/pkg/sentry/fs/inotify.go +++ b/pkg/sentry/fs/inotify.go @@ -16,7 +16,6 @@ package fs import ( "io" - "sync" "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" @@ -25,6 +24,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/uniqueid" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/inotify_watch.go b/pkg/sentry/fs/inotify_watch.go index 0aa0a5e9b..900cba3ca 100644 --- a/pkg/sentry/fs/inotify_watch.go +++ b/pkg/sentry/fs/inotify_watch.go @@ -15,10 +15,10 @@ package fs import ( - "sync" "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sync" ) // Watch represent a particular inotify watch created by inotify_add_watch. diff --git a/pkg/sentry/fs/lock/BUILD b/pkg/sentry/fs/lock/BUILD index 8d62642e7..2c332a82a 100644 --- a/pkg/sentry/fs/lock/BUILD +++ b/pkg/sentry/fs/lock/BUILD @@ -44,6 +44,7 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/log", + "//pkg/sync", "//pkg/waiter", ], ) diff --git a/pkg/sentry/fs/lock/lock.go b/pkg/sentry/fs/lock/lock.go index 636484424..41b040818 100644 --- a/pkg/sentry/fs/lock/lock.go +++ b/pkg/sentry/fs/lock/lock.go @@ -52,9 +52,9 @@ package lock import ( "fmt" "math" - "sync" "syscall" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/mounts.go b/pkg/sentry/fs/mounts.go index ac0398bd9..db3dfd096 100644 --- a/pkg/sentry/fs/mounts.go +++ b/pkg/sentry/fs/mounts.go @@ -19,7 +19,6 @@ import ( "math" "path" "strings" - "sync" "syscall" "gvisor.dev/gvisor/pkg/abi/linux" @@ -27,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/fs/overlay.go b/pkg/sentry/fs/overlay.go index 25573e986..4cad55327 100644 --- a/pkg/sentry/fs/overlay.go +++ b/pkg/sentry/fs/overlay.go @@ -17,13 +17,12 @@ package fs import ( "fmt" "strings" - "sync" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/usermem" - "gvisor.dev/gvisor/pkg/syncutil" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) @@ -199,7 +198,7 @@ type overlayEntry struct { upper *Inode // dirCacheMu protects dirCache. - dirCacheMu syncutil.DowngradableRWMutex `state:"nosave"` + dirCacheMu sync.DowngradableRWMutex `state:"nosave"` // dirCache is cache of DentAttrs from upper and lower Inodes. dirCache *SortedDentryMap diff --git a/pkg/sentry/fs/proc/BUILD b/pkg/sentry/fs/proc/BUILD index 75cbb0622..94d46ab1b 100644 --- a/pkg/sentry/fs/proc/BUILD +++ b/pkg/sentry/fs/proc/BUILD @@ -51,6 +51,7 @@ go_library( "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usage", "//pkg/sentry/usermem", + "//pkg/sync", "//pkg/syserror", "//pkg/tcpip/header", "//pkg/waiter", diff --git a/pkg/sentry/fs/proc/seqfile/BUILD b/pkg/sentry/fs/proc/seqfile/BUILD index fe7067be1..38b246dff 100644 --- a/pkg/sentry/fs/proc/seqfile/BUILD +++ b/pkg/sentry/fs/proc/seqfile/BUILD @@ -16,6 +16,7 @@ go_library( "//pkg/sentry/fs/proc/device", "//pkg/sentry/kernel/time", "//pkg/sentry/usermem", + "//pkg/sync", "//pkg/syserror", "//pkg/waiter", ], diff --git a/pkg/sentry/fs/proc/seqfile/seqfile.go b/pkg/sentry/fs/proc/seqfile/seqfile.go index 5fe823000..f9af191d5 100644 --- a/pkg/sentry/fs/proc/seqfile/seqfile.go +++ b/pkg/sentry/fs/proc/seqfile/seqfile.go @@ -17,7 +17,6 @@ package seqfile import ( "io" - "sync" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/context" @@ -26,6 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs/proc/device" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/proc/sys_net.go b/pkg/sentry/fs/proc/sys_net.go index bd93f83fa..a37e1fa06 100644 --- a/pkg/sentry/fs/proc/sys_net.go +++ b/pkg/sentry/fs/proc/sys_net.go @@ -17,7 +17,6 @@ package proc import ( "fmt" "io" - "sync" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/context" @@ -27,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/ramfs/BUILD b/pkg/sentry/fs/ramfs/BUILD index 012cb3e44..3fb7b0633 100644 --- a/pkg/sentry/fs/ramfs/BUILD +++ b/pkg/sentry/fs/ramfs/BUILD @@ -21,6 +21,7 @@ go_library( "//pkg/sentry/fs/fsutil", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usermem", + "//pkg/sync", "//pkg/syserror", "//pkg/waiter", ], diff --git a/pkg/sentry/fs/ramfs/dir.go b/pkg/sentry/fs/ramfs/dir.go index 78e082b8e..dcbb8eb2e 100644 --- a/pkg/sentry/fs/ramfs/dir.go +++ b/pkg/sentry/fs/ramfs/dir.go @@ -17,7 +17,6 @@ package ramfs import ( "fmt" - "sync" "syscall" "gvisor.dev/gvisor/pkg/abi/linux" @@ -25,6 +24,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/fs/restore.go b/pkg/sentry/fs/restore.go index f10168125..64c6a6ae9 100644 --- a/pkg/sentry/fs/restore.go +++ b/pkg/sentry/fs/restore.go @@ -15,7 +15,7 @@ package fs import ( - "sync" + "gvisor.dev/gvisor/pkg/sync" ) // RestoreEnvironment is the restore environment for file systems. It consists diff --git a/pkg/sentry/fs/tmpfs/BUILD b/pkg/sentry/fs/tmpfs/BUILD index 59ce400c2..3400b940c 100644 --- a/pkg/sentry/fs/tmpfs/BUILD +++ b/pkg/sentry/fs/tmpfs/BUILD @@ -31,6 +31,7 @@ go_library( "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usage", "//pkg/sentry/usermem", + "//pkg/sync", "//pkg/syserror", "//pkg/waiter", ], diff --git a/pkg/sentry/fs/tmpfs/inode_file.go b/pkg/sentry/fs/tmpfs/inode_file.go index f86dfaa36..f1c87fe41 100644 --- a/pkg/sentry/fs/tmpfs/inode_file.go +++ b/pkg/sentry/fs/tmpfs/inode_file.go @@ -17,7 +17,6 @@ package tmpfs import ( "fmt" "io" - "sync" "time" "gvisor.dev/gvisor/pkg/abi/linux" @@ -31,6 +30,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/fs/tty/BUILD b/pkg/sentry/fs/tty/BUILD index 95ad98cb0..f6f60d0cf 100644 --- a/pkg/sentry/fs/tty/BUILD +++ b/pkg/sentry/fs/tty/BUILD @@ -30,6 +30,7 @@ go_library( "//pkg/sentry/socket/unix/transport", "//pkg/sentry/unimpl", "//pkg/sentry/usermem", + "//pkg/sync", "//pkg/syserror", "//pkg/waiter", ], diff --git a/pkg/sentry/fs/tty/dir.go b/pkg/sentry/fs/tty/dir.go index 2f639c823..88aa66b24 100644 --- a/pkg/sentry/fs/tty/dir.go +++ b/pkg/sentry/fs/tty/dir.go @@ -19,7 +19,6 @@ import ( "fmt" "math" "strconv" - "sync" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/context" @@ -28,6 +27,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/tty/line_discipline.go b/pkg/sentry/fs/tty/line_discipline.go index 7cc0eb409..894964260 100644 --- a/pkg/sentry/fs/tty/line_discipline.go +++ b/pkg/sentry/fs/tty/line_discipline.go @@ -16,13 +16,13 @@ package tty import ( "bytes" - "sync" "unicode/utf8" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/tty/queue.go b/pkg/sentry/fs/tty/queue.go index 231e4e6eb..8b5d4699a 100644 --- a/pkg/sentry/fs/tty/queue.go +++ b/pkg/sentry/fs/tty/queue.go @@ -15,13 +15,12 @@ package tty import ( - "sync" - "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fsimpl/ext/BUILD b/pkg/sentry/fsimpl/ext/BUILD index bc90330bc..903874141 100644 --- a/pkg/sentry/fsimpl/ext/BUILD +++ b/pkg/sentry/fsimpl/ext/BUILD @@ -50,6 +50,7 @@ go_library( "//pkg/sentry/syscalls/linux", "//pkg/sentry/usermem", "//pkg/sentry/vfs", + "//pkg/sync", "//pkg/syserror", "//pkg/waiter", ], diff --git a/pkg/sentry/fsimpl/ext/directory.go b/pkg/sentry/fsimpl/ext/directory.go index 91802dc1e..8944171c8 100644 --- a/pkg/sentry/fsimpl/ext/directory.go +++ b/pkg/sentry/fsimpl/ext/directory.go @@ -15,8 +15,6 @@ package ext import ( - "sync" - "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/log" @@ -25,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/fsimpl/ext/filesystem.go b/pkg/sentry/fsimpl/ext/filesystem.go index 616fc002a..9afb1a84c 100644 --- a/pkg/sentry/fsimpl/ext/filesystem.go +++ b/pkg/sentry/fsimpl/ext/filesystem.go @@ -17,13 +17,13 @@ package ext import ( "errors" "io" - "sync" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/fsimpl/ext/regular_file.go b/pkg/sentry/fsimpl/ext/regular_file.go index aec33e00a..d11153c90 100644 --- a/pkg/sentry/fsimpl/ext/regular_file.go +++ b/pkg/sentry/fsimpl/ext/regular_file.go @@ -16,7 +16,6 @@ package ext import ( "io" - "sync" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/context" @@ -24,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/fsimpl/kernfs/BUILD b/pkg/sentry/fsimpl/kernfs/BUILD index 39c03ee9d..809178250 100644 --- a/pkg/sentry/fsimpl/kernfs/BUILD +++ b/pkg/sentry/fsimpl/kernfs/BUILD @@ -39,6 +39,7 @@ go_library( "//pkg/sentry/memmap", "//pkg/sentry/usermem", "//pkg/sentry/vfs", + "//pkg/sync", "//pkg/syserror", ], ) @@ -56,6 +57,7 @@ go_test( "//pkg/sentry/kernel/auth", "//pkg/sentry/usermem", "//pkg/sentry/vfs", + "//pkg/sync", "//pkg/syserror", "@com_github_google_go-cmp//cmp:go_default_library", ], diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 752e0f659..1d469a0db 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -16,7 +16,6 @@ package kernfs import ( "fmt" - "sync" "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" @@ -24,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index d69b299ae..bb12f39a2 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -53,7 +53,6 @@ package kernfs import ( "fmt" - "sync" "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" @@ -61,6 +60,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sync" ) // FilesystemType implements vfs.FilesystemType. diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index 4b6b95f5f..5c9d580e1 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -19,7 +19,6 @@ import ( "fmt" "io" "runtime" - "sync" "testing" "github.com/google/go-cmp/cmp" @@ -31,6 +30,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD index a5b285987..82f5c2f41 100644 --- a/pkg/sentry/fsimpl/tmpfs/BUILD +++ b/pkg/sentry/fsimpl/tmpfs/BUILD @@ -47,6 +47,7 @@ go_library( "//pkg/sentry/usage", "//pkg/sentry/usermem", "//pkg/sentry/vfs", + "//pkg/sync", "//pkg/syserror", ], ) diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index f51e247a7..f200e767d 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -17,7 +17,6 @@ package tmpfs import ( "io" "math" - "sync" "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" @@ -30,6 +29,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 7be6faa5b..701826f90 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -26,7 +26,6 @@ package tmpfs import ( "fmt" "math" - "sync" "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" @@ -34,6 +33,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD index 2706927ff..ac85ba0c8 100644 --- a/pkg/sentry/kernel/BUILD +++ b/pkg/sentry/kernel/BUILD @@ -35,7 +35,7 @@ go_template_instance( out = "seqatomic_taskgoroutineschedinfo_unsafe.go", package = "kernel", suffix = "TaskGoroutineSchedInfo", - template = "//pkg/syncutil:generic_seqatomic", + template = "//pkg/sync:generic_seqatomic", types = { "Value": "TaskGoroutineSchedInfo", }, @@ -209,7 +209,7 @@ go_library( "//pkg/sentry/usermem", "//pkg/state", "//pkg/state/statefile", - "//pkg/syncutil", + "//pkg/sync", "//pkg/syserr", "//pkg/syserror", "//pkg/tcpip", @@ -241,6 +241,7 @@ go_test( "//pkg/sentry/time", "//pkg/sentry/usage", "//pkg/sentry/usermem", + "//pkg/sync", "//pkg/syserror", ], ) diff --git a/pkg/sentry/kernel/abstract_socket_namespace.go b/pkg/sentry/kernel/abstract_socket_namespace.go index 244655b5c..920fe4329 100644 --- a/pkg/sentry/kernel/abstract_socket_namespace.go +++ b/pkg/sentry/kernel/abstract_socket_namespace.go @@ -15,11 +15,11 @@ package kernel import ( - "sync" "syscall" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" + "gvisor.dev/gvisor/pkg/sync" ) // +stateify savable diff --git a/pkg/sentry/kernel/auth/BUILD b/pkg/sentry/kernel/auth/BUILD index 04c244447..1aa72fa47 100644 --- a/pkg/sentry/kernel/auth/BUILD +++ b/pkg/sentry/kernel/auth/BUILD @@ -8,7 +8,7 @@ go_template_instance( out = "atomicptr_credentials_unsafe.go", package = "auth", suffix = "Credentials", - template = "//pkg/syncutil:generic_atomicptr", + template = "//pkg/sync:generic_atomicptr", types = { "Value": "Credentials", }, @@ -64,6 +64,7 @@ go_library( "//pkg/bits", "//pkg/log", "//pkg/sentry/context", + "//pkg/sync", "//pkg/syserror", ], ) diff --git a/pkg/sentry/kernel/auth/user_namespace.go b/pkg/sentry/kernel/auth/user_namespace.go index af28ccc65..9dd52c860 100644 --- a/pkg/sentry/kernel/auth/user_namespace.go +++ b/pkg/sentry/kernel/auth/user_namespace.go @@ -16,8 +16,8 @@ package auth import ( "math" - "sync" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/kernel/epoll/BUILD b/pkg/sentry/kernel/epoll/BUILD index 3361e8b7d..c47f6b6fc 100644 --- a/pkg/sentry/kernel/epoll/BUILD +++ b/pkg/sentry/kernel/epoll/BUILD @@ -32,6 +32,7 @@ go_library( "//pkg/sentry/fs/anon", "//pkg/sentry/fs/fsutil", "//pkg/sentry/usermem", + "//pkg/sync", "//pkg/waiter", ], ) diff --git a/pkg/sentry/kernel/epoll/epoll.go b/pkg/sentry/kernel/epoll/epoll.go index 9c0a4e1b4..430311cc0 100644 --- a/pkg/sentry/kernel/epoll/epoll.go +++ b/pkg/sentry/kernel/epoll/epoll.go @@ -18,7 +18,6 @@ package epoll import ( "fmt" - "sync" "syscall" "gvisor.dev/gvisor/pkg/refs" @@ -27,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs/anon" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/kernel/eventfd/BUILD b/pkg/sentry/kernel/eventfd/BUILD index e65b961e8..c831fbab2 100644 --- a/pkg/sentry/kernel/eventfd/BUILD +++ b/pkg/sentry/kernel/eventfd/BUILD @@ -16,6 +16,7 @@ go_library( "//pkg/sentry/fs/anon", "//pkg/sentry/fs/fsutil", "//pkg/sentry/usermem", + "//pkg/sync", "//pkg/syserror", "//pkg/waiter", ], diff --git a/pkg/sentry/kernel/eventfd/eventfd.go b/pkg/sentry/kernel/eventfd/eventfd.go index 12f0d429b..687690679 100644 --- a/pkg/sentry/kernel/eventfd/eventfd.go +++ b/pkg/sentry/kernel/eventfd/eventfd.go @@ -18,7 +18,6 @@ package eventfd import ( "math" - "sync" "syscall" "gvisor.dev/gvisor/pkg/abi/linux" @@ -28,6 +27,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs/anon" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/kernel/fasync/BUILD b/pkg/sentry/kernel/fasync/BUILD index 49d81b712..6b36bc63e 100644 --- a/pkg/sentry/kernel/fasync/BUILD +++ b/pkg/sentry/kernel/fasync/BUILD @@ -12,6 +12,7 @@ go_library( "//pkg/sentry/fs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", + "//pkg/sync", "//pkg/waiter", ], ) diff --git a/pkg/sentry/kernel/fasync/fasync.go b/pkg/sentry/kernel/fasync/fasync.go index 6b0bb0324..d32c3e90a 100644 --- a/pkg/sentry/kernel/fasync/fasync.go +++ b/pkg/sentry/kernel/fasync/fasync.go @@ -16,12 +16,11 @@ package fasync import ( - "sync" - "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/kernel/fd_table.go b/pkg/sentry/kernel/fd_table.go index 11f613a11..cd1501f85 100644 --- a/pkg/sentry/kernel/fd_table.go +++ b/pkg/sentry/kernel/fd_table.go @@ -18,7 +18,6 @@ import ( "bytes" "fmt" "math" - "sync" "sync/atomic" "syscall" @@ -28,6 +27,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/limits" + "gvisor.dev/gvisor/pkg/sync" ) // FDFlags define flags for an individual descriptor. diff --git a/pkg/sentry/kernel/fd_table_test.go b/pkg/sentry/kernel/fd_table_test.go index 2bcb6216a..eccb7d1e7 100644 --- a/pkg/sentry/kernel/fd_table_test.go +++ b/pkg/sentry/kernel/fd_table_test.go @@ -16,7 +16,6 @@ package kernel import ( "runtime" - "sync" "testing" "gvisor.dev/gvisor/pkg/sentry/context" @@ -24,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/filetest" "gvisor.dev/gvisor/pkg/sentry/limits" + "gvisor.dev/gvisor/pkg/sync" ) const ( diff --git a/pkg/sentry/kernel/fs_context.go b/pkg/sentry/kernel/fs_context.go index ded27d668..2448c1d99 100644 --- a/pkg/sentry/kernel/fs_context.go +++ b/pkg/sentry/kernel/fs_context.go @@ -16,10 +16,10 @@ package kernel import ( "fmt" - "sync" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sync" ) // FSContext contains filesystem context. diff --git a/pkg/sentry/kernel/futex/BUILD b/pkg/sentry/kernel/futex/BUILD index 75ec31761..50db443ce 100644 --- a/pkg/sentry/kernel/futex/BUILD +++ b/pkg/sentry/kernel/futex/BUILD @@ -9,7 +9,7 @@ go_template_instance( out = "atomicptr_bucket_unsafe.go", package = "futex", suffix = "Bucket", - template = "//pkg/syncutil:generic_atomicptr", + template = "//pkg/sync:generic_atomicptr", types = { "Value": "bucket", }, @@ -42,6 +42,7 @@ go_library( "//pkg/sentry/context", "//pkg/sentry/memmap", "//pkg/sentry/usermem", + "//pkg/sync", "//pkg/syserror", ], ) @@ -51,5 +52,8 @@ go_test( size = "small", srcs = ["futex_test.go"], embed = [":futex"], - deps = ["//pkg/sentry/usermem"], + deps = [ + "//pkg/sentry/usermem", + "//pkg/sync", + ], ) diff --git a/pkg/sentry/kernel/futex/futex.go b/pkg/sentry/kernel/futex/futex.go index 278cc8143..d1931c8f4 100644 --- a/pkg/sentry/kernel/futex/futex.go +++ b/pkg/sentry/kernel/futex/futex.go @@ -18,11 +18,10 @@ package futex import ( - "sync" - "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/kernel/futex/futex_test.go b/pkg/sentry/kernel/futex/futex_test.go index 65e5d1428..c23126ca5 100644 --- a/pkg/sentry/kernel/futex/futex_test.go +++ b/pkg/sentry/kernel/futex/futex_test.go @@ -17,13 +17,13 @@ package futex import ( "math" "runtime" - "sync" "sync/atomic" "syscall" "testing" "unsafe" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" ) // testData implements the Target interface, and allows us to diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 8653d2f63..c85e97fef 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -36,7 +36,6 @@ import ( "fmt" "io" "path/filepath" - "sync" "sync/atomic" "time" @@ -67,6 +66,7 @@ import ( uspb "gvisor.dev/gvisor/pkg/sentry/unimpl/unimplemented_syscall_go_proto" "gvisor.dev/gvisor/pkg/sentry/uniqueid" "gvisor.dev/gvisor/pkg/state" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" ) diff --git a/pkg/sentry/kernel/memevent/BUILD b/pkg/sentry/kernel/memevent/BUILD index d7a7d1169..7f36252a9 100644 --- a/pkg/sentry/kernel/memevent/BUILD +++ b/pkg/sentry/kernel/memevent/BUILD @@ -16,6 +16,7 @@ go_library( "//pkg/metric", "//pkg/sentry/kernel", "//pkg/sentry/usage", + "//pkg/sync", ], ) diff --git a/pkg/sentry/kernel/memevent/memory_events.go b/pkg/sentry/kernel/memevent/memory_events.go index b0d98e7f0..200565bb8 100644 --- a/pkg/sentry/kernel/memevent/memory_events.go +++ b/pkg/sentry/kernel/memevent/memory_events.go @@ -17,7 +17,6 @@ package memevent import ( - "sync" "time" "gvisor.dev/gvisor/pkg/eventchannel" @@ -26,6 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" pb "gvisor.dev/gvisor/pkg/sentry/kernel/memevent/memory_events_go_proto" "gvisor.dev/gvisor/pkg/sentry/usage" + "gvisor.dev/gvisor/pkg/sync" ) var totalTicks = metric.MustCreateNewUint64Metric("/memory_events/ticks", false /*sync*/, "Total number of memory event periods that have elapsed since startup.") diff --git a/pkg/sentry/kernel/pipe/BUILD b/pkg/sentry/kernel/pipe/BUILD index 9d34f6d4d..5eeaeff66 100644 --- a/pkg/sentry/kernel/pipe/BUILD +++ b/pkg/sentry/kernel/pipe/BUILD @@ -43,6 +43,7 @@ go_library( "//pkg/sentry/safemem", "//pkg/sentry/usermem", "//pkg/sentry/vfs", + "//pkg/sync", "//pkg/syserror", "//pkg/waiter", ], diff --git a/pkg/sentry/kernel/pipe/buffer.go b/pkg/sentry/kernel/pipe/buffer.go index 95bee2d37..1c0f34269 100644 --- a/pkg/sentry/kernel/pipe/buffer.go +++ b/pkg/sentry/kernel/pipe/buffer.go @@ -16,9 +16,9 @@ package pipe import ( "io" - "sync" "gvisor.dev/gvisor/pkg/sentry/safemem" + "gvisor.dev/gvisor/pkg/sync" ) // buffer encapsulates a queueable byte buffer. diff --git a/pkg/sentry/kernel/pipe/node.go b/pkg/sentry/kernel/pipe/node.go index 4a19ab7ce..716f589af 100644 --- a/pkg/sentry/kernel/pipe/node.go +++ b/pkg/sentry/kernel/pipe/node.go @@ -15,12 +15,11 @@ package pipe import ( - "sync" - "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/kernel/pipe/pipe.go b/pkg/sentry/kernel/pipe/pipe.go index 1a1b38f83..e4fd7d420 100644 --- a/pkg/sentry/kernel/pipe/pipe.go +++ b/pkg/sentry/kernel/pipe/pipe.go @@ -17,12 +17,12 @@ package pipe import ( "fmt" - "sync" "sync/atomic" "syscall" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/kernel/pipe/pipe_util.go b/pkg/sentry/kernel/pipe/pipe_util.go index ef9641e6a..8394eb78b 100644 --- a/pkg/sentry/kernel/pipe/pipe_util.go +++ b/pkg/sentry/kernel/pipe/pipe_util.go @@ -17,7 +17,6 @@ package pipe import ( "io" "math" - "sync" "syscall" "gvisor.dev/gvisor/pkg/abi/linux" @@ -25,6 +24,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/kernel/pipe/vfs.go b/pkg/sentry/kernel/pipe/vfs.go index 6416e0dd8..bf7461cbb 100644 --- a/pkg/sentry/kernel/pipe/vfs.go +++ b/pkg/sentry/kernel/pipe/vfs.go @@ -15,13 +15,12 @@ package pipe import ( - "sync" - "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/kernel/semaphore/BUILD b/pkg/sentry/kernel/semaphore/BUILD index f4c00cd86..13a961594 100644 --- a/pkg/sentry/kernel/semaphore/BUILD +++ b/pkg/sentry/kernel/semaphore/BUILD @@ -31,6 +31,7 @@ go_library( "//pkg/sentry/fs", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/time", + "//pkg/sync", "//pkg/syserror", ], ) diff --git a/pkg/sentry/kernel/semaphore/semaphore.go b/pkg/sentry/kernel/semaphore/semaphore.go index de9617e9d..18299814e 100644 --- a/pkg/sentry/kernel/semaphore/semaphore.go +++ b/pkg/sentry/kernel/semaphore/semaphore.go @@ -17,7 +17,6 @@ package semaphore import ( "fmt" - "sync" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/log" @@ -25,6 +24,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/kernel/shm/BUILD b/pkg/sentry/kernel/shm/BUILD index cd48945e6..7321b22ed 100644 --- a/pkg/sentry/kernel/shm/BUILD +++ b/pkg/sentry/kernel/shm/BUILD @@ -24,6 +24,7 @@ go_library( "//pkg/sentry/platform", "//pkg/sentry/usage", "//pkg/sentry/usermem", + "//pkg/sync", "//pkg/syserror", ], ) diff --git a/pkg/sentry/kernel/shm/shm.go b/pkg/sentry/kernel/shm/shm.go index 19034a21e..8ddef7eb8 100644 --- a/pkg/sentry/kernel/shm/shm.go +++ b/pkg/sentry/kernel/shm/shm.go @@ -35,7 +35,6 @@ package shm import ( "fmt" - "sync" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/log" @@ -49,6 +48,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/kernel/signal_handlers.go b/pkg/sentry/kernel/signal_handlers.go index a16f3d57f..768fda220 100644 --- a/pkg/sentry/kernel/signal_handlers.go +++ b/pkg/sentry/kernel/signal_handlers.go @@ -15,10 +15,9 @@ package kernel import ( - "sync" - "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sync" ) // SignalHandlers holds information about signal actions. diff --git a/pkg/sentry/kernel/signalfd/BUILD b/pkg/sentry/kernel/signalfd/BUILD index 9f7e19b4d..89e4d84b1 100644 --- a/pkg/sentry/kernel/signalfd/BUILD +++ b/pkg/sentry/kernel/signalfd/BUILD @@ -16,6 +16,7 @@ go_library( "//pkg/sentry/fs/fsutil", "//pkg/sentry/kernel", "//pkg/sentry/usermem", + "//pkg/sync", "//pkg/syserror", "//pkg/waiter", ], diff --git a/pkg/sentry/kernel/signalfd/signalfd.go b/pkg/sentry/kernel/signalfd/signalfd.go index 4b08d7d72..28be4a939 100644 --- a/pkg/sentry/kernel/signalfd/signalfd.go +++ b/pkg/sentry/kernel/signalfd/signalfd.go @@ -16,8 +16,6 @@ package signalfd import ( - "sync" - "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/sentry/context" @@ -26,6 +24,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/kernel/syscalls.go b/pkg/sentry/kernel/syscalls.go index 2fdee0282..d2d01add4 100644 --- a/pkg/sentry/kernel/syscalls.go +++ b/pkg/sentry/kernel/syscalls.go @@ -16,13 +16,13 @@ package kernel import ( "fmt" - "sync" "sync/atomic" "gvisor.dev/gvisor/pkg/abi" "gvisor.dev/gvisor/pkg/bits" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" ) // maxSyscallNum is the highest supported syscall number. diff --git a/pkg/sentry/kernel/syslog.go b/pkg/sentry/kernel/syslog.go index 8227ecf1d..4607cde2f 100644 --- a/pkg/sentry/kernel/syslog.go +++ b/pkg/sentry/kernel/syslog.go @@ -17,7 +17,8 @@ package kernel import ( "fmt" "math/rand" - "sync" + + "gvisor.dev/gvisor/pkg/sync" ) // syslog represents a sentry-global kernel log. diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index d25a7903b..978d66da8 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -17,7 +17,6 @@ package kernel import ( gocontext "context" "runtime/trace" - "sync" "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" @@ -37,7 +36,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/uniqueid" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/sentry/usermem" - "gvisor.dev/gvisor/pkg/syncutil" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/waiter" ) @@ -85,7 +84,7 @@ type Task struct { // // gosched is protected by goschedSeq. gosched is owned by the task // goroutine. - goschedSeq syncutil.SeqCount `state:"nosave"` + goschedSeq sync.SeqCount `state:"nosave"` gosched TaskGoroutineSchedInfo // yieldCount is the number of times the task goroutine has called diff --git a/pkg/sentry/kernel/thread_group.go b/pkg/sentry/kernel/thread_group.go index c0197a563..768e958d2 100644 --- a/pkg/sentry/kernel/thread_group.go +++ b/pkg/sentry/kernel/thread_group.go @@ -15,7 +15,6 @@ package kernel import ( - "sync" "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" @@ -25,6 +24,7 @@ import ( ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/usage" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/kernel/threads.go b/pkg/sentry/kernel/threads.go index 8267929a6..bf2dabb6e 100644 --- a/pkg/sentry/kernel/threads.go +++ b/pkg/sentry/kernel/threads.go @@ -16,9 +16,9 @@ package kernel import ( "fmt" - "sync" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/kernel/time/BUILD b/pkg/sentry/kernel/time/BUILD index 31847e1df..4e4de0512 100644 --- a/pkg/sentry/kernel/time/BUILD +++ b/pkg/sentry/kernel/time/BUILD @@ -13,6 +13,7 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/sentry/context", + "//pkg/sync", "//pkg/syserror", "//pkg/waiter", ], diff --git a/pkg/sentry/kernel/time/time.go b/pkg/sentry/kernel/time/time.go index 107394183..706de83ef 100644 --- a/pkg/sentry/kernel/time/time.go +++ b/pkg/sentry/kernel/time/time.go @@ -19,10 +19,10 @@ package time import ( "fmt" "math" - "sync" "time" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/kernel/timekeeper.go b/pkg/sentry/kernel/timekeeper.go index 76417342a..dc99301de 100644 --- a/pkg/sentry/kernel/timekeeper.go +++ b/pkg/sentry/kernel/timekeeper.go @@ -16,7 +16,6 @@ package kernel import ( "fmt" - "sync" "time" "gvisor.dev/gvisor/pkg/log" @@ -24,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/platform" sentrytime "gvisor.dev/gvisor/pkg/sentry/time" + "gvisor.dev/gvisor/pkg/sync" ) // Timekeeper manages all of the kernel clocks. diff --git a/pkg/sentry/kernel/tty.go b/pkg/sentry/kernel/tty.go index 048de26dc..464d2306a 100644 --- a/pkg/sentry/kernel/tty.go +++ b/pkg/sentry/kernel/tty.go @@ -14,7 +14,7 @@ package kernel -import "sync" +import "gvisor.dev/gvisor/pkg/sync" // TTY defines the relationship between a thread group and its controlling // terminal. diff --git a/pkg/sentry/kernel/uts_namespace.go b/pkg/sentry/kernel/uts_namespace.go index 0a563e715..8ccf04bd1 100644 --- a/pkg/sentry/kernel/uts_namespace.go +++ b/pkg/sentry/kernel/uts_namespace.go @@ -15,9 +15,8 @@ package kernel import ( - "sync" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sync" ) // UTSNamespace represents a UTS namespace, a holder of two system identifiers: diff --git a/pkg/sentry/limits/BUILD b/pkg/sentry/limits/BUILD index 156e67bf8..9fa841e8b 100644 --- a/pkg/sentry/limits/BUILD +++ b/pkg/sentry/limits/BUILD @@ -15,6 +15,7 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/sentry/context", + "//pkg/sync", ], ) diff --git a/pkg/sentry/limits/limits.go b/pkg/sentry/limits/limits.go index b6c22656b..31b9e9ff6 100644 --- a/pkg/sentry/limits/limits.go +++ b/pkg/sentry/limits/limits.go @@ -16,8 +16,9 @@ package limits import ( - "sync" "syscall" + + "gvisor.dev/gvisor/pkg/sync" ) // LimitType defines a type of resource limit. diff --git a/pkg/sentry/mm/BUILD b/pkg/sentry/mm/BUILD index 839931f67..83e248431 100644 --- a/pkg/sentry/mm/BUILD +++ b/pkg/sentry/mm/BUILD @@ -118,7 +118,7 @@ go_library( "//pkg/sentry/safemem", "//pkg/sentry/usage", "//pkg/sentry/usermem", - "//pkg/syncutil", + "//pkg/sync", "//pkg/syserror", "//pkg/tcpip/buffer", ], diff --git a/pkg/sentry/mm/aio_context.go b/pkg/sentry/mm/aio_context.go index 1b746d030..4b48866ad 100644 --- a/pkg/sentry/mm/aio_context.go +++ b/pkg/sentry/mm/aio_context.go @@ -15,8 +15,6 @@ package mm import ( - "sync" - "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/context" @@ -25,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/mm/mm.go b/pkg/sentry/mm/mm.go index 58a5c186d..fa86ebced 100644 --- a/pkg/sentry/mm/mm.go +++ b/pkg/sentry/mm/mm.go @@ -35,8 +35,6 @@ package mm import ( - "sync" - "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/memmap" @@ -44,7 +42,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/usermem" - "gvisor.dev/gvisor/pkg/syncutil" + "gvisor.dev/gvisor/pkg/sync" ) // MemoryManager implements a virtual address space. @@ -82,7 +80,7 @@ type MemoryManager struct { users int32 // mappingMu is analogous to Linux's struct mm_struct::mmap_sem. - mappingMu syncutil.DowngradableRWMutex `state:"nosave"` + mappingMu sync.DowngradableRWMutex `state:"nosave"` // vmas stores virtual memory areas. Since vmas are stored by value, // clients should usually use vmaIterator.ValuePtr() instead of @@ -125,7 +123,7 @@ type MemoryManager struct { // activeMu is loosely analogous to Linux's struct // mm_struct::page_table_lock. - activeMu syncutil.DowngradableRWMutex `state:"nosave"` + activeMu sync.DowngradableRWMutex `state:"nosave"` // pmas stores platform mapping areas used to implement vmas. Since pmas // are stored by value, clients should usually use pmaIterator.ValuePtr() diff --git a/pkg/sentry/pgalloc/BUILD b/pkg/sentry/pgalloc/BUILD index f404107af..a9a2642c5 100644 --- a/pkg/sentry/pgalloc/BUILD +++ b/pkg/sentry/pgalloc/BUILD @@ -73,6 +73,7 @@ go_library( "//pkg/sentry/usage", "//pkg/sentry/usermem", "//pkg/state", + "//pkg/sync", "//pkg/syserror", ], ) diff --git a/pkg/sentry/pgalloc/pgalloc.go b/pkg/sentry/pgalloc/pgalloc.go index f7f7298c4..c99e023d9 100644 --- a/pkg/sentry/pgalloc/pgalloc.go +++ b/pkg/sentry/pgalloc/pgalloc.go @@ -25,7 +25,6 @@ import ( "fmt" "math" "os" - "sync" "sync/atomic" "syscall" "time" @@ -37,6 +36,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/platform/interrupt/BUILD b/pkg/sentry/platform/interrupt/BUILD index b6d008dbe..85e882df9 100644 --- a/pkg/sentry/platform/interrupt/BUILD +++ b/pkg/sentry/platform/interrupt/BUILD @@ -10,6 +10,7 @@ go_library( ], importpath = "gvisor.dev/gvisor/pkg/sentry/platform/interrupt", visibility = ["//pkg/sentry:internal"], + deps = ["//pkg/sync"], ) go_test( diff --git a/pkg/sentry/platform/interrupt/interrupt.go b/pkg/sentry/platform/interrupt/interrupt.go index a4651f500..57be41647 100644 --- a/pkg/sentry/platform/interrupt/interrupt.go +++ b/pkg/sentry/platform/interrupt/interrupt.go @@ -17,7 +17,8 @@ package interrupt import ( "fmt" - "sync" + + "gvisor.dev/gvisor/pkg/sync" ) // Receiver receives interrupt notifications from a Forwarder. diff --git a/pkg/sentry/platform/kvm/BUILD b/pkg/sentry/platform/kvm/BUILD index f3afd98da..6a358d1d4 100644 --- a/pkg/sentry/platform/kvm/BUILD +++ b/pkg/sentry/platform/kvm/BUILD @@ -55,6 +55,7 @@ go_library( "//pkg/sentry/platform/safecopy", "//pkg/sentry/time", "//pkg/sentry/usermem", + "//pkg/sync", ], ) diff --git a/pkg/sentry/platform/kvm/address_space.go b/pkg/sentry/platform/kvm/address_space.go index ea8b9632e..a25f3c449 100644 --- a/pkg/sentry/platform/kvm/address_space.go +++ b/pkg/sentry/platform/kvm/address_space.go @@ -15,13 +15,13 @@ package kvm import ( - "sync" "sync/atomic" "gvisor.dev/gvisor/pkg/atomicbitops" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" ) // dirtySet tracks vCPUs for invalidation. diff --git a/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go b/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go index e5fac0d6a..2f02c03cf 100644 --- a/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go @@ -17,8 +17,6 @@ package kvm import ( - "unsafe" - "gvisor.dev/gvisor/pkg/sentry/arch" ) diff --git a/pkg/sentry/platform/kvm/kvm.go b/pkg/sentry/platform/kvm/kvm.go index f2c2c059e..a7850faed 100644 --- a/pkg/sentry/platform/kvm/kvm.go +++ b/pkg/sentry/platform/kvm/kvm.go @@ -18,13 +18,13 @@ package kvm import ( "fmt" "os" - "sync" "syscall" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/ring0" "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" ) // KVM represents a lightweight VM context. diff --git a/pkg/sentry/platform/kvm/machine.go b/pkg/sentry/platform/kvm/machine.go index 7d02ebf19..e6d912168 100644 --- a/pkg/sentry/platform/kvm/machine.go +++ b/pkg/sentry/platform/kvm/machine.go @@ -17,7 +17,6 @@ package kvm import ( "fmt" "runtime" - "sync" "sync/atomic" "syscall" @@ -27,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/platform/ring0" "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" ) // machine contains state associated with the VM as a whole. diff --git a/pkg/sentry/platform/ptrace/BUILD b/pkg/sentry/platform/ptrace/BUILD index 0df8cfa0f..cd13390c3 100644 --- a/pkg/sentry/platform/ptrace/BUILD +++ b/pkg/sentry/platform/ptrace/BUILD @@ -33,6 +33,7 @@ go_library( "//pkg/sentry/platform/interrupt", "//pkg/sentry/platform/safecopy", "//pkg/sentry/usermem", + "//pkg/sync", "@org_golang_x_sys//unix:go_default_library", ], ) diff --git a/pkg/sentry/platform/ptrace/ptrace.go b/pkg/sentry/platform/ptrace/ptrace.go index 7b120a15d..bb0e03880 100644 --- a/pkg/sentry/platform/ptrace/ptrace.go +++ b/pkg/sentry/platform/ptrace/ptrace.go @@ -46,13 +46,13 @@ package ptrace import ( "os" - "sync" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/interrupt" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" ) var ( diff --git a/pkg/sentry/platform/ptrace/subprocess.go b/pkg/sentry/platform/ptrace/subprocess.go index 20244fd95..15dc46a5b 100644 --- a/pkg/sentry/platform/ptrace/subprocess.go +++ b/pkg/sentry/platform/ptrace/subprocess.go @@ -18,7 +18,6 @@ import ( "fmt" "os" "runtime" - "sync" "syscall" "golang.org/x/sys/unix" @@ -27,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" ) // Linux kernel errnos which "should never be seen by user programs", but will diff --git a/pkg/sentry/platform/ptrace/subprocess_linux_unsafe.go b/pkg/sentry/platform/ptrace/subprocess_linux_unsafe.go index 2e6fbe488..245b20722 100644 --- a/pkg/sentry/platform/ptrace/subprocess_linux_unsafe.go +++ b/pkg/sentry/platform/ptrace/subprocess_linux_unsafe.go @@ -18,7 +18,6 @@ package ptrace import ( - "sync" "sync/atomic" "syscall" "unsafe" @@ -26,6 +25,7 @@ import ( "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/hostcpu" + "gvisor.dev/gvisor/pkg/sync" ) // maskPool contains reusable CPU masks for setting affinity. Unfortunately, diff --git a/pkg/sentry/platform/ring0/defs.go b/pkg/sentry/platform/ring0/defs.go index 3f094c2a7..86fd5ed58 100644 --- a/pkg/sentry/platform/ring0/defs.go +++ b/pkg/sentry/platform/ring0/defs.go @@ -17,7 +17,7 @@ package ring0 import ( "syscall" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" ) // Kernel is a global kernel object. diff --git a/pkg/sentry/platform/ring0/defs_amd64.go b/pkg/sentry/platform/ring0/defs_amd64.go index 10dbd381f..9dae0dccb 100644 --- a/pkg/sentry/platform/ring0/defs_amd64.go +++ b/pkg/sentry/platform/ring0/defs_amd64.go @@ -18,6 +18,7 @@ package ring0 import ( "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" + "gvisor.dev/gvisor/pkg/sentry/usermem" ) var ( diff --git a/pkg/sentry/platform/ring0/defs_arm64.go b/pkg/sentry/platform/ring0/defs_arm64.go index dc0eeec01..a850ce6cf 100644 --- a/pkg/sentry/platform/ring0/defs_arm64.go +++ b/pkg/sentry/platform/ring0/defs_arm64.go @@ -18,6 +18,7 @@ package ring0 import ( "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" + "gvisor.dev/gvisor/pkg/sentry/usermem" ) var ( diff --git a/pkg/sentry/platform/ring0/pagetables/BUILD b/pkg/sentry/platform/ring0/pagetables/BUILD index e2e15ba5c..387a7f6c3 100644 --- a/pkg/sentry/platform/ring0/pagetables/BUILD +++ b/pkg/sentry/platform/ring0/pagetables/BUILD @@ -96,7 +96,10 @@ go_library( "//pkg/sentry/platform/kvm:__subpackages__", "//pkg/sentry/platform/ring0:__subpackages__", ], - deps = ["//pkg/sentry/usermem"], + deps = [ + "//pkg/sentry/usermem", + "//pkg/sync", + ], ) go_test( diff --git a/pkg/sentry/platform/ring0/pagetables/pcids_x86.go b/pkg/sentry/platform/ring0/pagetables/pcids_x86.go index 0f029f25d..e199bae18 100644 --- a/pkg/sentry/platform/ring0/pagetables/pcids_x86.go +++ b/pkg/sentry/platform/ring0/pagetables/pcids_x86.go @@ -17,7 +17,7 @@ package pagetables import ( - "sync" + "gvisor.dev/gvisor/pkg/sync" ) // limitPCID is the number of valid PCIDs. diff --git a/pkg/sentry/socket/netlink/BUILD b/pkg/sentry/socket/netlink/BUILD index 136821963..103933144 100644 --- a/pkg/sentry/socket/netlink/BUILD +++ b/pkg/sentry/socket/netlink/BUILD @@ -27,6 +27,7 @@ go_library( "//pkg/sentry/socket/unix", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usermem", + "//pkg/sync", "//pkg/syserr", "//pkg/syserror", "//pkg/tcpip", diff --git a/pkg/sentry/socket/netlink/port/BUILD b/pkg/sentry/socket/netlink/port/BUILD index 463544c1a..2d9f4ba9b 100644 --- a/pkg/sentry/socket/netlink/port/BUILD +++ b/pkg/sentry/socket/netlink/port/BUILD @@ -8,6 +8,7 @@ go_library( srcs = ["port.go"], importpath = "gvisor.dev/gvisor/pkg/sentry/socket/netlink/port", visibility = ["//pkg/sentry:internal"], + deps = ["//pkg/sync"], ) go_test( diff --git a/pkg/sentry/socket/netlink/port/port.go b/pkg/sentry/socket/netlink/port/port.go index e9d3275b1..2cd3afc22 100644 --- a/pkg/sentry/socket/netlink/port/port.go +++ b/pkg/sentry/socket/netlink/port/port.go @@ -24,7 +24,8 @@ import ( "fmt" "math" "math/rand" - "sync" + + "gvisor.dev/gvisor/pkg/sync" ) // maxPorts is a sanity limit on the maximum number of ports to allocate per diff --git a/pkg/sentry/socket/netlink/socket.go b/pkg/sentry/socket/netlink/socket.go index d2e3644a6..cea56f4ed 100644 --- a/pkg/sentry/socket/netlink/socket.go +++ b/pkg/sentry/socket/netlink/socket.go @@ -17,7 +17,6 @@ package netlink import ( "math" - "sync" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" @@ -34,6 +33,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/socket/unix" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip" diff --git a/pkg/sentry/socket/netstack/BUILD b/pkg/sentry/socket/netstack/BUILD index e414d8055..f78784569 100644 --- a/pkg/sentry/socket/netstack/BUILD +++ b/pkg/sentry/socket/netstack/BUILD @@ -34,6 +34,7 @@ go_library( "//pkg/sentry/socket/netfilter", "//pkg/sentry/unimpl", "//pkg/sentry/usermem", + "//pkg/sync", "//pkg/syserr", "//pkg/syserror", "//pkg/tcpip", diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 764f11a6b..0affb8071 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -29,7 +29,6 @@ import ( "io" "math" "reflect" - "sync" "syscall" "time" @@ -49,6 +48,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/socket/netfilter" "gvisor.dev/gvisor/pkg/sentry/unimpl" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip" diff --git a/pkg/sentry/socket/rpcinet/conn/BUILD b/pkg/sentry/socket/rpcinet/conn/BUILD index 23eadcb1b..b2677c659 100644 --- a/pkg/sentry/socket/rpcinet/conn/BUILD +++ b/pkg/sentry/socket/rpcinet/conn/BUILD @@ -10,6 +10,7 @@ go_library( deps = [ "//pkg/binary", "//pkg/sentry/socket/rpcinet:syscall_rpc_go_proto", + "//pkg/sync", "//pkg/syserr", "//pkg/unet", "@com_github_golang_protobuf//proto:go_default_library", diff --git a/pkg/sentry/socket/rpcinet/conn/conn.go b/pkg/sentry/socket/rpcinet/conn/conn.go index 356adad99..02f39c767 100644 --- a/pkg/sentry/socket/rpcinet/conn/conn.go +++ b/pkg/sentry/socket/rpcinet/conn/conn.go @@ -17,12 +17,12 @@ package conn import ( "fmt" - "sync" "sync/atomic" "syscall" "github.com/golang/protobuf/proto" "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/unet" diff --git a/pkg/sentry/socket/rpcinet/notifier/BUILD b/pkg/sentry/socket/rpcinet/notifier/BUILD index a3585e10d..a5954f22b 100644 --- a/pkg/sentry/socket/rpcinet/notifier/BUILD +++ b/pkg/sentry/socket/rpcinet/notifier/BUILD @@ -10,6 +10,7 @@ go_library( deps = [ "//pkg/sentry/socket/rpcinet:syscall_rpc_go_proto", "//pkg/sentry/socket/rpcinet/conn", + "//pkg/sync", "//pkg/waiter", "@org_golang_x_sys//unix:go_default_library", ], diff --git a/pkg/sentry/socket/rpcinet/notifier/notifier.go b/pkg/sentry/socket/rpcinet/notifier/notifier.go index 7efe4301f..82b75d6dd 100644 --- a/pkg/sentry/socket/rpcinet/notifier/notifier.go +++ b/pkg/sentry/socket/rpcinet/notifier/notifier.go @@ -17,12 +17,12 @@ package notifier import ( "fmt" - "sync" "syscall" "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/conn" pb "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/syscall_rpc_go_proto" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/socket/unix/transport/BUILD b/pkg/sentry/socket/unix/transport/BUILD index 788ad70d2..d7ba95dff 100644 --- a/pkg/sentry/socket/unix/transport/BUILD +++ b/pkg/sentry/socket/unix/transport/BUILD @@ -32,6 +32,7 @@ go_library( "//pkg/ilist", "//pkg/refs", "//pkg/sentry/context", + "//pkg/sync", "//pkg/syserr", "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/sentry/socket/unix/transport/connectioned.go b/pkg/sentry/socket/unix/transport/connectioned.go index dea11e253..9e6fbc111 100644 --- a/pkg/sentry/socket/unix/transport/connectioned.go +++ b/pkg/sentry/socket/unix/transport/connectioned.go @@ -15,10 +15,9 @@ package transport import ( - "sync" - "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/waiter" diff --git a/pkg/sentry/socket/unix/transport/queue.go b/pkg/sentry/socket/unix/transport/queue.go index e27b1c714..5dcd3d95e 100644 --- a/pkg/sentry/socket/unix/transport/queue.go +++ b/pkg/sentry/socket/unix/transport/queue.go @@ -15,9 +15,8 @@ package transport import ( - "sync" - "gvisor.dev/gvisor/pkg/refs" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/socket/unix/transport/unix.go b/pkg/sentry/socket/unix/transport/unix.go index 37c7ac3c1..fcc0da332 100644 --- a/pkg/sentry/socket/unix/transport/unix.go +++ b/pkg/sentry/socket/unix/transport/unix.go @@ -16,11 +16,11 @@ package transport import ( - "sync" "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD index a76975cee..aa05e208a 100644 --- a/pkg/sentry/syscalls/linux/BUILD +++ b/pkg/sentry/syscalls/linux/BUILD @@ -91,6 +91,7 @@ go_library( "//pkg/sentry/syscalls", "//pkg/sentry/usage", "//pkg/sentry/usermem", + "//pkg/sync", "//pkg/syserr", "//pkg/syserror", "//pkg/waiter", diff --git a/pkg/sentry/syscalls/linux/error.go b/pkg/sentry/syscalls/linux/error.go index 1d9018c96..60469549d 100644 --- a/pkg/sentry/syscalls/linux/error.go +++ b/pkg/sentry/syscalls/linux/error.go @@ -16,13 +16,13 @@ package linux import ( "io" - "sync" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/metric" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/time/BUILD b/pkg/sentry/time/BUILD index 18e212dff..3cde3a0be 100644 --- a/pkg/sentry/time/BUILD +++ b/pkg/sentry/time/BUILD @@ -9,7 +9,7 @@ go_template_instance( out = "seqatomic_parameters_unsafe.go", package = "time", suffix = "Parameters", - template = "//pkg/syncutil:generic_seqatomic", + template = "//pkg/sync:generic_seqatomic", types = { "Value": "Parameters", }, @@ -36,7 +36,7 @@ go_library( deps = [ "//pkg/log", "//pkg/metric", - "//pkg/syncutil", + "//pkg/sync", "//pkg/syserror", ], ) diff --git a/pkg/sentry/time/calibrated_clock.go b/pkg/sentry/time/calibrated_clock.go index 318503277..f9a93115d 100644 --- a/pkg/sentry/time/calibrated_clock.go +++ b/pkg/sentry/time/calibrated_clock.go @@ -17,11 +17,11 @@ package time import ( - "sync" "time" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/metric" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/usage/BUILD b/pkg/sentry/usage/BUILD index c32fe3241..5518ac3d0 100644 --- a/pkg/sentry/usage/BUILD +++ b/pkg/sentry/usage/BUILD @@ -18,5 +18,6 @@ go_library( deps = [ "//pkg/bits", "//pkg/memutil", + "//pkg/sync", ], ) diff --git a/pkg/sentry/usage/memory.go b/pkg/sentry/usage/memory.go index d6ef644d8..538c645eb 100644 --- a/pkg/sentry/usage/memory.go +++ b/pkg/sentry/usage/memory.go @@ -17,12 +17,12 @@ package usage import ( "fmt" "os" - "sync" "sync/atomic" "syscall" "gvisor.dev/gvisor/pkg/bits" "gvisor.dev/gvisor/pkg/memutil" + "gvisor.dev/gvisor/pkg/sync" ) // MemoryKind represents a type of memory used by the application. diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index 4c6aa04a1..35c7be259 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -34,7 +34,7 @@ go_library( "//pkg/sentry/kernel/auth", "//pkg/sentry/memmap", "//pkg/sentry/usermem", - "//pkg/syncutil", + "//pkg/sync", "//pkg/syserror", "//pkg/waiter", ], @@ -54,6 +54,7 @@ go_test( "//pkg/sentry/context/contexttest", "//pkg/sentry/kernel/auth", "//pkg/sentry/usermem", + "//pkg/sync", "//pkg/syserror", ], ) diff --git a/pkg/sentry/vfs/dentry.go b/pkg/sentry/vfs/dentry.go index 1bc9c4a38..486a76475 100644 --- a/pkg/sentry/vfs/dentry.go +++ b/pkg/sentry/vfs/dentry.go @@ -16,9 +16,9 @@ package vfs import ( "fmt" - "sync" "sync/atomic" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/vfs/file_description_impl_util.go b/pkg/sentry/vfs/file_description_impl_util.go index 66eb57bc2..c00b3c84b 100644 --- a/pkg/sentry/vfs/file_description_impl_util.go +++ b/pkg/sentry/vfs/file_description_impl_util.go @@ -17,13 +17,13 @@ package vfs import ( "bytes" "io" - "sync" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/vfs/mount_test.go b/pkg/sentry/vfs/mount_test.go index adff0b94b..3b933468d 100644 --- a/pkg/sentry/vfs/mount_test.go +++ b/pkg/sentry/vfs/mount_test.go @@ -17,8 +17,9 @@ package vfs import ( "fmt" "runtime" - "sync" "testing" + + "gvisor.dev/gvisor/pkg/sync" ) func TestMountTableLookupEmpty(t *testing.T) { diff --git a/pkg/sentry/vfs/mount_unsafe.go b/pkg/sentry/vfs/mount_unsafe.go index ab13fa461..bd90d36c4 100644 --- a/pkg/sentry/vfs/mount_unsafe.go +++ b/pkg/sentry/vfs/mount_unsafe.go @@ -26,7 +26,7 @@ import ( "sync/atomic" "unsafe" - "gvisor.dev/gvisor/pkg/syncutil" + "gvisor.dev/gvisor/pkg/sync" ) // mountKey represents the location at which a Mount is mounted. It is @@ -75,7 +75,7 @@ type mountTable struct { // intrinsics and inline assembly, limiting the performance of this // approach.) - seq syncutil.SeqCount + seq sync.SeqCount seed uint32 // for hashing keys // size holds both length (number of elements) and capacity (number of diff --git a/pkg/sentry/vfs/pathname.go b/pkg/sentry/vfs/pathname.go index 8e155654f..cf80df90e 100644 --- a/pkg/sentry/vfs/pathname.go +++ b/pkg/sentry/vfs/pathname.go @@ -15,10 +15,9 @@ package vfs import ( - "sync" - "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/vfs/resolving_path.go b/pkg/sentry/vfs/resolving_path.go index f0641d314..8a0b382f6 100644 --- a/pkg/sentry/vfs/resolving_path.go +++ b/pkg/sentry/vfs/resolving_path.go @@ -16,11 +16,11 @@ package vfs import ( "fmt" - "sync" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index ea2db7031..1f21b0b31 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -29,12 +29,12 @@ package vfs import ( "fmt" - "sync" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/watchdog/BUILD b/pkg/sentry/watchdog/BUILD index 4d8435265..28f21f13d 100644 --- a/pkg/sentry/watchdog/BUILD +++ b/pkg/sentry/watchdog/BUILD @@ -13,5 +13,6 @@ go_library( "//pkg/metric", "//pkg/sentry/kernel", "//pkg/sentry/kernel/time", + "//pkg/sync", ], ) diff --git a/pkg/sentry/watchdog/watchdog.go b/pkg/sentry/watchdog/watchdog.go index 5e4611333..bfb2fac26 100644 --- a/pkg/sentry/watchdog/watchdog.go +++ b/pkg/sentry/watchdog/watchdog.go @@ -32,7 +32,6 @@ package watchdog import ( "bytes" "fmt" - "sync" "time" "gvisor.dev/gvisor/pkg/abi/linux" @@ -40,6 +39,7 @@ import ( "gvisor.dev/gvisor/pkg/metric" "gvisor.dev/gvisor/pkg/sentry/kernel" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" + "gvisor.dev/gvisor/pkg/sync" ) // Opts configures the watchdog. diff --git a/pkg/sync/BUILD b/pkg/sync/BUILD new file mode 100644 index 000000000..e8cd16b8f --- /dev/null +++ b/pkg/sync/BUILD @@ -0,0 +1,53 @@ +load("//tools/go_stateify:defs.bzl", "go_library") +load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools/go_generics:defs.bzl", "go_template") + +package( + default_visibility = ["//:sandbox"], + licenses = ["notice"], +) + +exports_files(["LICENSE"]) + +go_template( + name = "generic_atomicptr", + srcs = ["atomicptr_unsafe.go"], + types = [ + "Value", + ], +) + +go_template( + name = "generic_seqatomic", + srcs = ["seqatomic_unsafe.go"], + types = [ + "Value", + ], + deps = [ + ":sync", + ], +) + +go_library( + name = "sync", + srcs = [ + "aliases.go", + "downgradable_rwmutex_unsafe.go", + "memmove_unsafe.go", + "norace_unsafe.go", + "race_unsafe.go", + "seqcount.go", + "syncutil.go", + ], + importpath = "gvisor.dev/gvisor/pkg/sync", +) + +go_test( + name = "sync_test", + size = "small", + srcs = [ + "downgradable_rwmutex_test.go", + "seqcount_test.go", + ], + embed = [":sync"], +) diff --git a/pkg/sync/LICENSE b/pkg/sync/LICENSE new file mode 100644 index 000000000..6a66aea5e --- /dev/null +++ b/pkg/sync/LICENSE @@ -0,0 +1,27 @@ +Copyright (c) 2009 The Go Authors. All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are +met: + + * Redistributions of source code must retain the above copyright +notice, this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above +copyright notice, this list of conditions and the following disclaimer +in the documentation and/or other materials provided with the +distribution. + * Neither the name of Google Inc. nor the names of its +contributors may be used to endorse or promote products derived from +this software without specific prior written permission. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/pkg/sync/README.md b/pkg/sync/README.md new file mode 100644 index 000000000..2183c4e20 --- /dev/null +++ b/pkg/sync/README.md @@ -0,0 +1,5 @@ +# Syncutil + +This package provides additional synchronization primitives not provided by the +Go stdlib 'sync' package. It is partially derived from the upstream 'sync' +package from go1.10. diff --git a/pkg/sync/aliases.go b/pkg/sync/aliases.go new file mode 100644 index 000000000..20c7ca041 --- /dev/null +++ b/pkg/sync/aliases.go @@ -0,0 +1,37 @@ +// Copyright 2020 The gVisor Authors. +// +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package sync + +import ( + "sync" +) + +// Aliases of standard library types. +type ( + // Mutex is an alias of sync.Mutex. + Mutex = sync.Mutex + + // RWMutex is an alias of sync.RWMutex. + RWMutex = sync.RWMutex + + // Cond is an alias of sync.Cond. + Cond = sync.Cond + + // Locker is an alias of sync.Locker. + Locker = sync.Locker + + // Once is an alias of sync.Once. + Once = sync.Once + + // Pool is an alias of sync.Pool. + Pool = sync.Pool + + // WaitGroup is an alias of sync.WaitGroup. + WaitGroup = sync.WaitGroup + + // Map is an alias of sync.Map. + Map = sync.Map +) diff --git a/pkg/sync/atomicptr_unsafe.go b/pkg/sync/atomicptr_unsafe.go new file mode 100644 index 000000000..525c4beed --- /dev/null +++ b/pkg/sync/atomicptr_unsafe.go @@ -0,0 +1,47 @@ +// Copyright 2019 The gVisor Authors. +// +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// Package template doesn't exist. This file must be instantiated using the +// go_template_instance rule in tools/go_generics/defs.bzl. +package template + +import ( + "sync/atomic" + "unsafe" +) + +// Value is a required type parameter. +type Value struct{} + +// An AtomicPtr is a pointer to a value of type Value that can be atomically +// loaded and stored. The zero value of an AtomicPtr represents nil. +// +// Note that copying AtomicPtr by value performs a non-atomic read of the +// stored pointer, which is unsafe if Store() can be called concurrently; in +// this case, do `dst.Store(src.Load())` instead. +// +// +stateify savable +type AtomicPtr struct { + ptr unsafe.Pointer `state:".(*Value)"` +} + +func (p *AtomicPtr) savePtr() *Value { + return p.Load() +} + +func (p *AtomicPtr) loadPtr(v *Value) { + p.Store(v) +} + +// Load returns the value set by the most recent Store. It returns nil if there +// has been no previous call to Store. +func (p *AtomicPtr) Load() *Value { + return (*Value)(atomic.LoadPointer(&p.ptr)) +} + +// Store sets the value returned by Load to x. +func (p *AtomicPtr) Store(x *Value) { + atomic.StorePointer(&p.ptr, (unsafe.Pointer)(x)) +} diff --git a/pkg/sync/atomicptrtest/BUILD b/pkg/sync/atomicptrtest/BUILD new file mode 100644 index 000000000..418eda29c --- /dev/null +++ b/pkg/sync/atomicptrtest/BUILD @@ -0,0 +1,29 @@ +load("//tools/go_stateify:defs.bzl", "go_library") +load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools/go_generics:defs.bzl", "go_template_instance") + +package(licenses = ["notice"]) + +go_template_instance( + name = "atomicptr_int", + out = "atomicptr_int_unsafe.go", + package = "atomicptr", + suffix = "Int", + template = "//pkg/sync:generic_atomicptr", + types = { + "Value": "int", + }, +) + +go_library( + name = "atomicptr", + srcs = ["atomicptr_int_unsafe.go"], + importpath = "gvisor.dev/gvisor/pkg/sync/atomicptr", +) + +go_test( + name = "atomicptr_test", + size = "small", + srcs = ["atomicptr_test.go"], + embed = [":atomicptr"], +) diff --git a/pkg/sync/atomicptrtest/atomicptr_test.go b/pkg/sync/atomicptrtest/atomicptr_test.go new file mode 100644 index 000000000..8fdc5112e --- /dev/null +++ b/pkg/sync/atomicptrtest/atomicptr_test.go @@ -0,0 +1,31 @@ +// Copyright 2019 The gVisor Authors. +// +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package atomicptr + +import ( + "testing" +) + +func newInt(val int) *int { + return &val +} + +func TestAtomicPtr(t *testing.T) { + var p AtomicPtrInt + if got := p.Load(); got != nil { + t.Errorf("initial value is %p (%v), wanted nil", got, got) + } + want := newInt(42) + p.Store(want) + if got := p.Load(); got != want { + t.Errorf("wrong value: got %p (%v), wanted %p (%v)", got, got, want, want) + } + want = newInt(100) + p.Store(want) + if got := p.Load(); got != want { + t.Errorf("wrong value: got %p (%v), wanted %p (%v)", got, got, want, want) + } +} diff --git a/pkg/sync/downgradable_rwmutex_test.go b/pkg/sync/downgradable_rwmutex_test.go new file mode 100644 index 000000000..f04496bc5 --- /dev/null +++ b/pkg/sync/downgradable_rwmutex_test.go @@ -0,0 +1,150 @@ +// Copyright 2009 The Go Authors. All rights reserved. +// Copyright 2019 The gVisor Authors. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// GOMAXPROCS=10 go test + +// Copy/pasted from the standard library's sync/rwmutex_test.go, except for the +// addition of downgradingWriter and the renaming of num_iterations to +// numIterations to shut up Golint. + +package sync + +import ( + "fmt" + "runtime" + "sync/atomic" + "testing" +) + +func parallelReader(m *DowngradableRWMutex, clocked, cunlock, cdone chan bool) { + m.RLock() + clocked <- true + <-cunlock + m.RUnlock() + cdone <- true +} + +func doTestParallelReaders(numReaders, gomaxprocs int) { + runtime.GOMAXPROCS(gomaxprocs) + var m DowngradableRWMutex + clocked := make(chan bool) + cunlock := make(chan bool) + cdone := make(chan bool) + for i := 0; i < numReaders; i++ { + go parallelReader(&m, clocked, cunlock, cdone) + } + // Wait for all parallel RLock()s to succeed. + for i := 0; i < numReaders; i++ { + <-clocked + } + for i := 0; i < numReaders; i++ { + cunlock <- true + } + // Wait for the goroutines to finish. + for i := 0; i < numReaders; i++ { + <-cdone + } +} + +func TestParallelReaders(t *testing.T) { + defer runtime.GOMAXPROCS(runtime.GOMAXPROCS(-1)) + doTestParallelReaders(1, 4) + doTestParallelReaders(3, 4) + doTestParallelReaders(4, 2) +} + +func reader(rwm *DowngradableRWMutex, numIterations int, activity *int32, cdone chan bool) { + for i := 0; i < numIterations; i++ { + rwm.RLock() + n := atomic.AddInt32(activity, 1) + if n < 1 || n >= 10000 { + panic(fmt.Sprintf("wlock(%d)\n", n)) + } + for i := 0; i < 100; i++ { + } + atomic.AddInt32(activity, -1) + rwm.RUnlock() + } + cdone <- true +} + +func writer(rwm *DowngradableRWMutex, numIterations int, activity *int32, cdone chan bool) { + for i := 0; i < numIterations; i++ { + rwm.Lock() + n := atomic.AddInt32(activity, 10000) + if n != 10000 { + panic(fmt.Sprintf("wlock(%d)\n", n)) + } + for i := 0; i < 100; i++ { + } + atomic.AddInt32(activity, -10000) + rwm.Unlock() + } + cdone <- true +} + +func downgradingWriter(rwm *DowngradableRWMutex, numIterations int, activity *int32, cdone chan bool) { + for i := 0; i < numIterations; i++ { + rwm.Lock() + n := atomic.AddInt32(activity, 10000) + if n != 10000 { + panic(fmt.Sprintf("wlock(%d)\n", n)) + } + for i := 0; i < 100; i++ { + } + atomic.AddInt32(activity, -10000) + rwm.DowngradeLock() + n = atomic.AddInt32(activity, 1) + if n < 1 || n >= 10000 { + panic(fmt.Sprintf("wlock(%d)\n", n)) + } + for i := 0; i < 100; i++ { + } + n = atomic.AddInt32(activity, -1) + rwm.RUnlock() + } + cdone <- true +} + +func HammerDowngradableRWMutex(gomaxprocs, numReaders, numIterations int) { + runtime.GOMAXPROCS(gomaxprocs) + // Number of active readers + 10000 * number of active writers. + var activity int32 + var rwm DowngradableRWMutex + cdone := make(chan bool) + go writer(&rwm, numIterations, &activity, cdone) + go downgradingWriter(&rwm, numIterations, &activity, cdone) + var i int + for i = 0; i < numReaders/2; i++ { + go reader(&rwm, numIterations, &activity, cdone) + } + go writer(&rwm, numIterations, &activity, cdone) + go downgradingWriter(&rwm, numIterations, &activity, cdone) + for ; i < numReaders; i++ { + go reader(&rwm, numIterations, &activity, cdone) + } + // Wait for the 4 writers and all readers to finish. + for i := 0; i < 4+numReaders; i++ { + <-cdone + } +} + +func TestDowngradableRWMutex(t *testing.T) { + defer runtime.GOMAXPROCS(runtime.GOMAXPROCS(-1)) + n := 1000 + if testing.Short() { + n = 5 + } + HammerDowngradableRWMutex(1, 1, n) + HammerDowngradableRWMutex(1, 3, n) + HammerDowngradableRWMutex(1, 10, n) + HammerDowngradableRWMutex(4, 1, n) + HammerDowngradableRWMutex(4, 3, n) + HammerDowngradableRWMutex(4, 10, n) + HammerDowngradableRWMutex(10, 1, n) + HammerDowngradableRWMutex(10, 3, n) + HammerDowngradableRWMutex(10, 10, n) + HammerDowngradableRWMutex(10, 5, n) +} diff --git a/pkg/sync/downgradable_rwmutex_unsafe.go b/pkg/sync/downgradable_rwmutex_unsafe.go new file mode 100644 index 000000000..9bb55cd3a --- /dev/null +++ b/pkg/sync/downgradable_rwmutex_unsafe.go @@ -0,0 +1,146 @@ +// Copyright 2009 The Go Authors. All rights reserved. +// Copyright 2019 The gVisor Authors. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// +build go1.13 +// +build !go1.15 + +// Check go:linkname function signatures when updating Go version. + +// This is mostly copied from the standard library's sync/rwmutex.go. +// +// Happens-before relationships indicated to the race detector: +// - Unlock -> Lock (via writerSem) +// - Unlock -> RLock (via readerSem) +// - RUnlock -> Lock (via writerSem) +// - DowngradeLock -> RLock (via readerSem) + +package sync + +import ( + "sync" + "sync/atomic" + "unsafe" +) + +//go:linkname runtimeSemacquire sync.runtime_Semacquire +func runtimeSemacquire(s *uint32) + +//go:linkname runtimeSemrelease sync.runtime_Semrelease +func runtimeSemrelease(s *uint32, handoff bool, skipframes int) + +// DowngradableRWMutex is identical to sync.RWMutex, but adds the DowngradeLock +// method. +type DowngradableRWMutex struct { + w sync.Mutex // held if there are pending writers + writerSem uint32 // semaphore for writers to wait for completing readers + readerSem uint32 // semaphore for readers to wait for completing writers + readerCount int32 // number of pending readers + readerWait int32 // number of departing readers +} + +const rwmutexMaxReaders = 1 << 30 + +// RLock locks rw for reading. +func (rw *DowngradableRWMutex) RLock() { + if RaceEnabled { + RaceDisable() + } + if atomic.AddInt32(&rw.readerCount, 1) < 0 { + // A writer is pending, wait for it. + runtimeSemacquire(&rw.readerSem) + } + if RaceEnabled { + RaceEnable() + RaceAcquire(unsafe.Pointer(&rw.readerSem)) + } +} + +// RUnlock undoes a single RLock call. +func (rw *DowngradableRWMutex) RUnlock() { + if RaceEnabled { + RaceReleaseMerge(unsafe.Pointer(&rw.writerSem)) + RaceDisable() + } + if r := atomic.AddInt32(&rw.readerCount, -1); r < 0 { + if r+1 == 0 || r+1 == -rwmutexMaxReaders { + panic("RUnlock of unlocked DowngradableRWMutex") + } + // A writer is pending. + if atomic.AddInt32(&rw.readerWait, -1) == 0 { + // The last reader unblocks the writer. + runtimeSemrelease(&rw.writerSem, false, 0) + } + } + if RaceEnabled { + RaceEnable() + } +} + +// Lock locks rw for writing. +func (rw *DowngradableRWMutex) Lock() { + if RaceEnabled { + RaceDisable() + } + // First, resolve competition with other writers. + rw.w.Lock() + // Announce to readers there is a pending writer. + r := atomic.AddInt32(&rw.readerCount, -rwmutexMaxReaders) + rwmutexMaxReaders + // Wait for active readers. + if r != 0 && atomic.AddInt32(&rw.readerWait, r) != 0 { + runtimeSemacquire(&rw.writerSem) + } + if RaceEnabled { + RaceEnable() + RaceAcquire(unsafe.Pointer(&rw.writerSem)) + } +} + +// Unlock unlocks rw for writing. +func (rw *DowngradableRWMutex) Unlock() { + if RaceEnabled { + RaceRelease(unsafe.Pointer(&rw.writerSem)) + RaceRelease(unsafe.Pointer(&rw.readerSem)) + RaceDisable() + } + // Announce to readers there is no active writer. + r := atomic.AddInt32(&rw.readerCount, rwmutexMaxReaders) + if r >= rwmutexMaxReaders { + panic("Unlock of unlocked DowngradableRWMutex") + } + // Unblock blocked readers, if any. + for i := 0; i < int(r); i++ { + runtimeSemrelease(&rw.readerSem, false, 0) + } + // Allow other writers to proceed. + rw.w.Unlock() + if RaceEnabled { + RaceEnable() + } +} + +// DowngradeLock atomically unlocks rw for writing and locks it for reading. +func (rw *DowngradableRWMutex) DowngradeLock() { + if RaceEnabled { + RaceRelease(unsafe.Pointer(&rw.readerSem)) + RaceDisable() + } + // Announce to readers there is no active writer and one additional reader. + r := atomic.AddInt32(&rw.readerCount, rwmutexMaxReaders+1) + if r >= rwmutexMaxReaders+1 { + panic("DowngradeLock of unlocked DowngradableRWMutex") + } + // Unblock blocked readers, if any. Note that this loop starts as 1 since r + // includes this goroutine. + for i := 1; i < int(r); i++ { + runtimeSemrelease(&rw.readerSem, false, 0) + } + // Allow other writers to proceed to rw.w.Lock(). Note that they will still + // block on rw.writerSem since at least this reader exists, such that + // DowngradeLock() is atomic with the previous write lock. + rw.w.Unlock() + if RaceEnabled { + RaceEnable() + } +} diff --git a/pkg/sync/memmove_unsafe.go b/pkg/sync/memmove_unsafe.go new file mode 100644 index 000000000..ad4a3a37e --- /dev/null +++ b/pkg/sync/memmove_unsafe.go @@ -0,0 +1,28 @@ +// Copyright 2019 The gVisor Authors. +// +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// +build go1.12 +// +build !go1.15 + +// Check go:linkname function signatures when updating Go version. + +package sync + +import ( + "unsafe" +) + +//go:linkname memmove runtime.memmove +//go:noescape +func memmove(to, from unsafe.Pointer, n uintptr) + +// Memmove is exported for SeqAtomicLoad/SeqAtomicTryLoad, which can't +// define it because go_generics can't update the go:linkname annotation. +// Furthermore, go:linkname silently doesn't work if the local name is exported +// (this is of course undocumented), which is why this indirection is +// necessary. +func Memmove(to, from unsafe.Pointer, n uintptr) { + memmove(to, from, n) +} diff --git a/pkg/sync/norace_unsafe.go b/pkg/sync/norace_unsafe.go new file mode 100644 index 000000000..006055dd6 --- /dev/null +++ b/pkg/sync/norace_unsafe.go @@ -0,0 +1,35 @@ +// Copyright 2019 The gVisor Authors. +// +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// +build !race + +package sync + +import ( + "unsafe" +) + +// RaceEnabled is true if the Go data race detector is enabled. +const RaceEnabled = false + +// RaceDisable has the same semantics as runtime.RaceDisable. +func RaceDisable() { +} + +// RaceEnable has the same semantics as runtime.RaceEnable. +func RaceEnable() { +} + +// RaceAcquire has the same semantics as runtime.RaceAcquire. +func RaceAcquire(addr unsafe.Pointer) { +} + +// RaceRelease has the same semantics as runtime.RaceRelease. +func RaceRelease(addr unsafe.Pointer) { +} + +// RaceReleaseMerge has the same semantics as runtime.RaceReleaseMerge. +func RaceReleaseMerge(addr unsafe.Pointer) { +} diff --git a/pkg/sync/race_unsafe.go b/pkg/sync/race_unsafe.go new file mode 100644 index 000000000..31d8fa9a6 --- /dev/null +++ b/pkg/sync/race_unsafe.go @@ -0,0 +1,41 @@ +// Copyright 2019 The gVisor Authors. +// +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// +build race + +package sync + +import ( + "runtime" + "unsafe" +) + +// RaceEnabled is true if the Go data race detector is enabled. +const RaceEnabled = true + +// RaceDisable has the same semantics as runtime.RaceDisable. +func RaceDisable() { + runtime.RaceDisable() +} + +// RaceEnable has the same semantics as runtime.RaceEnable. +func RaceEnable() { + runtime.RaceEnable() +} + +// RaceAcquire has the same semantics as runtime.RaceAcquire. +func RaceAcquire(addr unsafe.Pointer) { + runtime.RaceAcquire(addr) +} + +// RaceRelease has the same semantics as runtime.RaceRelease. +func RaceRelease(addr unsafe.Pointer) { + runtime.RaceRelease(addr) +} + +// RaceReleaseMerge has the same semantics as runtime.RaceReleaseMerge. +func RaceReleaseMerge(addr unsafe.Pointer) { + runtime.RaceReleaseMerge(addr) +} diff --git a/pkg/sync/seqatomic_unsafe.go b/pkg/sync/seqatomic_unsafe.go new file mode 100644 index 000000000..eda6fb131 --- /dev/null +++ b/pkg/sync/seqatomic_unsafe.go @@ -0,0 +1,72 @@ +// Copyright 2019 The gVisor Authors. +// +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// Package template doesn't exist. This file must be instantiated using the +// go_template_instance rule in tools/go_generics/defs.bzl. +package template + +import ( + "fmt" + "reflect" + "strings" + "unsafe" + + "gvisor.dev/gvisor/pkg/sync" +) + +// Value is a required type parameter. +// +// Value must not contain any pointers, including interface objects, function +// objects, slices, maps, channels, unsafe.Pointer, and arrays or structs +// containing any of the above. An init() function will panic if this property +// does not hold. +type Value struct{} + +// SeqAtomicLoad returns a copy of *ptr, ensuring that the read does not race +// with any writer critical sections in sc. +func SeqAtomicLoad(sc *sync.SeqCount, ptr *Value) Value { + // This function doesn't use SeqAtomicTryLoad because doing so is + // measurably, significantly (~20%) slower; Go is awful at inlining. + var val Value + for { + epoch := sc.BeginRead() + if sync.RaceEnabled { + // runtime.RaceDisable() doesn't actually stop the race detector, + // so it can't help us here. Instead, call runtime.memmove + // directly, which is not instrumented by the race detector. + sync.Memmove(unsafe.Pointer(&val), unsafe.Pointer(ptr), unsafe.Sizeof(val)) + } else { + // This is ~40% faster for short reads than going through memmove. + val = *ptr + } + if sc.ReadOk(epoch) { + break + } + } + return val +} + +// SeqAtomicTryLoad returns a copy of *ptr while in a reader critical section +// in sc initiated by a call to sc.BeginRead() that returned epoch. If the read +// would race with a writer critical section, SeqAtomicTryLoad returns +// (unspecified, false). +func SeqAtomicTryLoad(sc *sync.SeqCount, epoch sync.SeqCountEpoch, ptr *Value) (Value, bool) { + var val Value + if sync.RaceEnabled { + sync.Memmove(unsafe.Pointer(&val), unsafe.Pointer(ptr), unsafe.Sizeof(val)) + } else { + val = *ptr + } + return val, sc.ReadOk(epoch) +} + +func init() { + var val Value + typ := reflect.TypeOf(val) + name := typ.Name() + if ptrs := sync.PointersInType(typ, name); len(ptrs) != 0 { + panic(fmt.Sprintf("SeqAtomicLoad<%s> is invalid since values %s of type %s contain pointers:\n%s", typ, name, typ, strings.Join(ptrs, "\n"))) + } +} diff --git a/pkg/sync/seqatomictest/BUILD b/pkg/sync/seqatomictest/BUILD new file mode 100644 index 000000000..eba21518d --- /dev/null +++ b/pkg/sync/seqatomictest/BUILD @@ -0,0 +1,33 @@ +load("//tools/go_stateify:defs.bzl", "go_library") +load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools/go_generics:defs.bzl", "go_template_instance") + +package(licenses = ["notice"]) + +go_template_instance( + name = "seqatomic_int", + out = "seqatomic_int_unsafe.go", + package = "seqatomic", + suffix = "Int", + template = "//pkg/sync:generic_seqatomic", + types = { + "Value": "int", + }, +) + +go_library( + name = "seqatomic", + srcs = ["seqatomic_int_unsafe.go"], + importpath = "gvisor.dev/gvisor/pkg/sync/seqatomic", + deps = [ + "//pkg/sync", + ], +) + +go_test( + name = "seqatomic_test", + size = "small", + srcs = ["seqatomic_test.go"], + embed = [":seqatomic"], + deps = ["//pkg/sync"], +) diff --git a/pkg/sync/seqatomictest/seqatomic_test.go b/pkg/sync/seqatomictest/seqatomic_test.go new file mode 100644 index 000000000..2c4568b07 --- /dev/null +++ b/pkg/sync/seqatomictest/seqatomic_test.go @@ -0,0 +1,132 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package seqatomic + +import ( + "sync/atomic" + "testing" + "time" + + "gvisor.dev/gvisor/pkg/sync" +) + +func TestSeqAtomicLoadUncontended(t *testing.T) { + var seq sync.SeqCount + const want = 1 + data := want + if got := SeqAtomicLoadInt(&seq, &data); got != want { + t.Errorf("SeqAtomicLoadInt: got %v, wanted %v", got, want) + } +} + +func TestSeqAtomicLoadAfterWrite(t *testing.T) { + var seq sync.SeqCount + var data int + const want = 1 + seq.BeginWrite() + data = want + seq.EndWrite() + if got := SeqAtomicLoadInt(&seq, &data); got != want { + t.Errorf("SeqAtomicLoadInt: got %v, wanted %v", got, want) + } +} + +func TestSeqAtomicLoadDuringWrite(t *testing.T) { + var seq sync.SeqCount + var data int + const want = 1 + seq.BeginWrite() + go func() { + time.Sleep(time.Second) + data = want + seq.EndWrite() + }() + if got := SeqAtomicLoadInt(&seq, &data); got != want { + t.Errorf("SeqAtomicLoadInt: got %v, wanted %v", got, want) + } +} + +func TestSeqAtomicTryLoadUncontended(t *testing.T) { + var seq sync.SeqCount + const want = 1 + data := want + epoch := seq.BeginRead() + if got, ok := SeqAtomicTryLoadInt(&seq, epoch, &data); !ok || got != want { + t.Errorf("SeqAtomicTryLoadInt: got (%v, %v), wanted (%v, true)", got, ok, want) + } +} + +func TestSeqAtomicTryLoadDuringWrite(t *testing.T) { + var seq sync.SeqCount + var data int + epoch := seq.BeginRead() + seq.BeginWrite() + if got, ok := SeqAtomicTryLoadInt(&seq, epoch, &data); ok { + t.Errorf("SeqAtomicTryLoadInt: got (%v, true), wanted (_, false)", got) + } + seq.EndWrite() +} + +func TestSeqAtomicTryLoadAfterWrite(t *testing.T) { + var seq sync.SeqCount + var data int + epoch := seq.BeginRead() + seq.BeginWrite() + seq.EndWrite() + if got, ok := SeqAtomicTryLoadInt(&seq, epoch, &data); ok { + t.Errorf("SeqAtomicTryLoadInt: got (%v, true), wanted (_, false)", got) + } +} + +func BenchmarkSeqAtomicLoadIntUncontended(b *testing.B) { + var seq sync.SeqCount + const want = 42 + data := want + b.RunParallel(func(pb *testing.PB) { + for pb.Next() { + if got := SeqAtomicLoadInt(&seq, &data); got != want { + b.Fatalf("SeqAtomicLoadInt: got %v, wanted %v", got, want) + } + } + }) +} + +func BenchmarkSeqAtomicTryLoadIntUncontended(b *testing.B) { + var seq sync.SeqCount + const want = 42 + data := want + b.RunParallel(func(pb *testing.PB) { + epoch := seq.BeginRead() + for pb.Next() { + if got, ok := SeqAtomicTryLoadInt(&seq, epoch, &data); !ok || got != want { + b.Fatalf("SeqAtomicTryLoadInt: got (%v, %v), wanted (%v, true)", got, ok, want) + } + } + }) +} + +// For comparison: +func BenchmarkAtomicValueLoadIntUncontended(b *testing.B) { + var a atomic.Value + const want = 42 + a.Store(int(want)) + b.RunParallel(func(pb *testing.PB) { + for pb.Next() { + if got := a.Load().(int); got != want { + b.Fatalf("atomic.Value.Load: got %v, wanted %v", got, want) + } + } + }) +} diff --git a/pkg/sync/seqcount.go b/pkg/sync/seqcount.go new file mode 100644 index 000000000..a1e895352 --- /dev/null +++ b/pkg/sync/seqcount.go @@ -0,0 +1,149 @@ +// Copyright 2019 The gVisor Authors. +// +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package sync + +import ( + "fmt" + "reflect" + "runtime" + "sync/atomic" +) + +// SeqCount is a synchronization primitive for optimistic reader/writer +// synchronization in cases where readers can work with stale data and +// therefore do not need to block writers. +// +// Compared to sync/atomic.Value: +// +// - Mutation of SeqCount-protected data does not require memory allocation, +// whereas atomic.Value generally does. This is a significant advantage when +// writes are common. +// +// - Atomic reads of SeqCount-protected data require copying. This is a +// disadvantage when atomic reads are common. +// +// - SeqCount may be more flexible: correct use of SeqCount.ReadOk allows other +// operations to be made atomic with reads of SeqCount-protected data. +// +// - SeqCount may be less flexible: as of this writing, SeqCount-protected data +// cannot include pointers. +// +// - SeqCount is more cumbersome to use; atomic reads of SeqCount-protected +// data require instantiating function templates using go_generics (see +// seqatomic.go). +type SeqCount struct { + // epoch is incremented by BeginWrite and EndWrite, such that epoch is odd + // if a writer critical section is active, and a read from data protected + // by this SeqCount is atomic iff epoch is the same even value before and + // after the read. + epoch uint32 +} + +// SeqCountEpoch tracks writer critical sections in a SeqCount. +type SeqCountEpoch struct { + val uint32 +} + +// We assume that: +// +// - All functions in sync/atomic that perform a memory read are at least a +// read fence: memory reads before calls to such functions cannot be reordered +// after the call, and memory reads after calls to such functions cannot be +// reordered before the call, even if those reads do not use sync/atomic. +// +// - All functions in sync/atomic that perform a memory write are at least a +// write fence: memory writes before calls to such functions cannot be +// reordered after the call, and memory writes after calls to such functions +// cannot be reordered before the call, even if those writes do not use +// sync/atomic. +// +// As of this writing, the Go memory model completely fails to describe +// sync/atomic, but these properties are implied by +// https://groups.google.com/forum/#!topic/golang-nuts/7EnEhM3U7B8. + +// BeginRead indicates the beginning of a reader critical section. Reader +// critical sections DO NOT BLOCK writer critical sections, so operations in a +// reader critical section MAY RACE with writer critical sections. Races are +// detected by ReadOk at the end of the reader critical section. Thus, the +// low-level structure of readers is generally: +// +// for { +// epoch := seq.BeginRead() +// // do something idempotent with seq-protected data +// if seq.ReadOk(epoch) { +// break +// } +// } +// +// However, since reader critical sections may race with writer critical +// sections, the Go race detector will (accurately) flag data races in readers +// using this pattern. Most users of SeqCount will need to use the +// SeqAtomicLoad function template in seqatomic.go. +func (s *SeqCount) BeginRead() SeqCountEpoch { + epoch := atomic.LoadUint32(&s.epoch) + for epoch&1 != 0 { + runtime.Gosched() + epoch = atomic.LoadUint32(&s.epoch) + } + return SeqCountEpoch{epoch} +} + +// ReadOk returns true if the reader critical section initiated by a previous +// call to BeginRead() that returned epoch did not race with any writer critical +// sections. +// +// ReadOk may be called any number of times during a reader critical section. +// Reader critical sections do not need to be explicitly terminated; the last +// call to ReadOk is implicitly the end of the reader critical section. +func (s *SeqCount) ReadOk(epoch SeqCountEpoch) bool { + return atomic.LoadUint32(&s.epoch) == epoch.val +} + +// BeginWrite indicates the beginning of a writer critical section. +// +// SeqCount does not support concurrent writer critical sections; clients with +// concurrent writers must synchronize them using e.g. sync.Mutex. +func (s *SeqCount) BeginWrite() { + if epoch := atomic.AddUint32(&s.epoch, 1); epoch&1 == 0 { + panic("SeqCount.BeginWrite during writer critical section") + } +} + +// EndWrite ends the effect of a preceding BeginWrite. +func (s *SeqCount) EndWrite() { + if epoch := atomic.AddUint32(&s.epoch, 1); epoch&1 != 0 { + panic("SeqCount.EndWrite outside writer critical section") + } +} + +// PointersInType returns a list of pointers reachable from values named +// valName of the given type. +// +// PointersInType is not exhaustive, but it is guaranteed that if typ contains +// at least one pointer, then PointersInTypeOf returns a non-empty list. +func PointersInType(typ reflect.Type, valName string) []string { + switch kind := typ.Kind(); kind { + case reflect.Bool, reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64, reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr, reflect.Float32, reflect.Float64, reflect.Complex64, reflect.Complex128: + return nil + + case reflect.Chan, reflect.Func, reflect.Interface, reflect.Map, reflect.Ptr, reflect.Slice, reflect.String, reflect.UnsafePointer: + return []string{valName} + + case reflect.Array: + return PointersInType(typ.Elem(), valName+"[]") + + case reflect.Struct: + var ptrs []string + for i, n := 0, typ.NumField(); i < n; i++ { + field := typ.Field(i) + ptrs = append(ptrs, PointersInType(field.Type, fmt.Sprintf("%s.%s", valName, field.Name))...) + } + return ptrs + + default: + return []string{fmt.Sprintf("%s (of type %s with unknown kind %s)", valName, typ, kind)} + } +} diff --git a/pkg/sync/seqcount_test.go b/pkg/sync/seqcount_test.go new file mode 100644 index 000000000..6eb7b4b59 --- /dev/null +++ b/pkg/sync/seqcount_test.go @@ -0,0 +1,153 @@ +// Copyright 2019 The gVisor Authors. +// +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package sync + +import ( + "reflect" + "testing" + "time" +) + +func TestSeqCountWriteUncontended(t *testing.T) { + var seq SeqCount + seq.BeginWrite() + seq.EndWrite() +} + +func TestSeqCountReadUncontended(t *testing.T) { + var seq SeqCount + epoch := seq.BeginRead() + if !seq.ReadOk(epoch) { + t.Errorf("ReadOk: got false, wanted true") + } +} + +func TestSeqCountBeginReadAfterWrite(t *testing.T) { + var seq SeqCount + var data int32 + const want = 1 + seq.BeginWrite() + data = want + seq.EndWrite() + epoch := seq.BeginRead() + if data != want { + t.Errorf("Reader: got %v, wanted %v", data, want) + } + if !seq.ReadOk(epoch) { + t.Errorf("ReadOk: got false, wanted true") + } +} + +func TestSeqCountBeginReadDuringWrite(t *testing.T) { + var seq SeqCount + var data int + const want = 1 + seq.BeginWrite() + go func() { + time.Sleep(time.Second) + data = want + seq.EndWrite() + }() + epoch := seq.BeginRead() + if data != want { + t.Errorf("Reader: got %v, wanted %v", data, want) + } + if !seq.ReadOk(epoch) { + t.Errorf("ReadOk: got false, wanted true") + } +} + +func TestSeqCountReadOkAfterWrite(t *testing.T) { + var seq SeqCount + epoch := seq.BeginRead() + seq.BeginWrite() + seq.EndWrite() + if seq.ReadOk(epoch) { + t.Errorf("ReadOk: got true, wanted false") + } +} + +func TestSeqCountReadOkDuringWrite(t *testing.T) { + var seq SeqCount + epoch := seq.BeginRead() + seq.BeginWrite() + if seq.ReadOk(epoch) { + t.Errorf("ReadOk: got true, wanted false") + } + seq.EndWrite() +} + +func BenchmarkSeqCountWriteUncontended(b *testing.B) { + var seq SeqCount + for i := 0; i < b.N; i++ { + seq.BeginWrite() + seq.EndWrite() + } +} + +func BenchmarkSeqCountReadUncontended(b *testing.B) { + var seq SeqCount + b.RunParallel(func(pb *testing.PB) { + for pb.Next() { + epoch := seq.BeginRead() + if !seq.ReadOk(epoch) { + b.Fatalf("ReadOk: got false, wanted true") + } + } + }) +} + +func TestPointersInType(t *testing.T) { + for _, test := range []struct { + name string // used for both test and value name + val interface{} + ptrs []string + }{ + { + name: "EmptyStruct", + val: struct{}{}, + }, + { + name: "Int", + val: int(0), + }, + { + name: "MixedStruct", + val: struct { + b bool + I int + ExportedPtr *struct{} + unexportedPtr *struct{} + arr [2]int + ptrArr [2]*int + nestedStruct struct { + nestedNonptr int + nestedPtr *int + } + structArr [1]struct { + nonptr int + ptr *int + } + }{}, + ptrs: []string{ + "MixedStruct.ExportedPtr", + "MixedStruct.unexportedPtr", + "MixedStruct.ptrArr[]", + "MixedStruct.nestedStruct.nestedPtr", + "MixedStruct.structArr[].ptr", + }, + }, + } { + t.Run(test.name, func(t *testing.T) { + typ := reflect.TypeOf(test.val) + ptrs := PointersInType(typ, test.name) + t.Logf("Found pointers: %v", ptrs) + if (len(ptrs) != 0 || len(test.ptrs) != 0) && !reflect.DeepEqual(ptrs, test.ptrs) { + t.Errorf("Got %v, wanted %v", ptrs, test.ptrs) + } + }) + } +} diff --git a/pkg/sync/syncutil.go b/pkg/sync/syncutil.go new file mode 100644 index 000000000..b16cf5333 --- /dev/null +++ b/pkg/sync/syncutil.go @@ -0,0 +1,7 @@ +// Copyright 2019 The gVisor Authors. +// +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// Package sync provides synchronization primitives. +package sync diff --git a/pkg/syncutil/BUILD b/pkg/syncutil/BUILD deleted file mode 100644 index cb1f41628..000000000 --- a/pkg/syncutil/BUILD +++ /dev/null @@ -1,52 +0,0 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_generics:defs.bzl", "go_template") - -package( - default_visibility = ["//:sandbox"], - licenses = ["notice"], -) - -exports_files(["LICENSE"]) - -go_template( - name = "generic_atomicptr", - srcs = ["atomicptr_unsafe.go"], - types = [ - "Value", - ], -) - -go_template( - name = "generic_seqatomic", - srcs = ["seqatomic_unsafe.go"], - types = [ - "Value", - ], - deps = [ - ":sync", - ], -) - -go_library( - name = "syncutil", - srcs = [ - "downgradable_rwmutex_unsafe.go", - "memmove_unsafe.go", - "norace_unsafe.go", - "race_unsafe.go", - "seqcount.go", - "syncutil.go", - ], - importpath = "gvisor.dev/gvisor/pkg/syncutil", -) - -go_test( - name = "syncutil_test", - size = "small", - srcs = [ - "downgradable_rwmutex_test.go", - "seqcount_test.go", - ], - embed = [":syncutil"], -) diff --git a/pkg/syncutil/LICENSE b/pkg/syncutil/LICENSE deleted file mode 100644 index 6a66aea5e..000000000 --- a/pkg/syncutil/LICENSE +++ /dev/null @@ -1,27 +0,0 @@ -Copyright (c) 2009 The Go Authors. All rights reserved. - -Redistribution and use in source and binary forms, with or without -modification, are permitted provided that the following conditions are -met: - - * Redistributions of source code must retain the above copyright -notice, this list of conditions and the following disclaimer. - * Redistributions in binary form must reproduce the above -copyright notice, this list of conditions and the following disclaimer -in the documentation and/or other materials provided with the -distribution. - * Neither the name of Google Inc. nor the names of its -contributors may be used to endorse or promote products derived from -this software without specific prior written permission. - -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/pkg/syncutil/README.md b/pkg/syncutil/README.md deleted file mode 100644 index 2183c4e20..000000000 --- a/pkg/syncutil/README.md +++ /dev/null @@ -1,5 +0,0 @@ -# Syncutil - -This package provides additional synchronization primitives not provided by the -Go stdlib 'sync' package. It is partially derived from the upstream 'sync' -package from go1.10. diff --git a/pkg/syncutil/atomicptr_unsafe.go b/pkg/syncutil/atomicptr_unsafe.go deleted file mode 100644 index 525c4beed..000000000 --- a/pkg/syncutil/atomicptr_unsafe.go +++ /dev/null @@ -1,47 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -// Package template doesn't exist. This file must be instantiated using the -// go_template_instance rule in tools/go_generics/defs.bzl. -package template - -import ( - "sync/atomic" - "unsafe" -) - -// Value is a required type parameter. -type Value struct{} - -// An AtomicPtr is a pointer to a value of type Value that can be atomically -// loaded and stored. The zero value of an AtomicPtr represents nil. -// -// Note that copying AtomicPtr by value performs a non-atomic read of the -// stored pointer, which is unsafe if Store() can be called concurrently; in -// this case, do `dst.Store(src.Load())` instead. -// -// +stateify savable -type AtomicPtr struct { - ptr unsafe.Pointer `state:".(*Value)"` -} - -func (p *AtomicPtr) savePtr() *Value { - return p.Load() -} - -func (p *AtomicPtr) loadPtr(v *Value) { - p.Store(v) -} - -// Load returns the value set by the most recent Store. It returns nil if there -// has been no previous call to Store. -func (p *AtomicPtr) Load() *Value { - return (*Value)(atomic.LoadPointer(&p.ptr)) -} - -// Store sets the value returned by Load to x. -func (p *AtomicPtr) Store(x *Value) { - atomic.StorePointer(&p.ptr, (unsafe.Pointer)(x)) -} diff --git a/pkg/syncutil/atomicptrtest/BUILD b/pkg/syncutil/atomicptrtest/BUILD deleted file mode 100644 index 63f411a90..000000000 --- a/pkg/syncutil/atomicptrtest/BUILD +++ /dev/null @@ -1,29 +0,0 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -package(licenses = ["notice"]) - -go_template_instance( - name = "atomicptr_int", - out = "atomicptr_int_unsafe.go", - package = "atomicptr", - suffix = "Int", - template = "//pkg/syncutil:generic_atomicptr", - types = { - "Value": "int", - }, -) - -go_library( - name = "atomicptr", - srcs = ["atomicptr_int_unsafe.go"], - importpath = "gvisor.dev/gvisor/pkg/syncutil/atomicptr", -) - -go_test( - name = "atomicptr_test", - size = "small", - srcs = ["atomicptr_test.go"], - embed = [":atomicptr"], -) diff --git a/pkg/syncutil/atomicptrtest/atomicptr_test.go b/pkg/syncutil/atomicptrtest/atomicptr_test.go deleted file mode 100644 index 8fdc5112e..000000000 --- a/pkg/syncutil/atomicptrtest/atomicptr_test.go +++ /dev/null @@ -1,31 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -package atomicptr - -import ( - "testing" -) - -func newInt(val int) *int { - return &val -} - -func TestAtomicPtr(t *testing.T) { - var p AtomicPtrInt - if got := p.Load(); got != nil { - t.Errorf("initial value is %p (%v), wanted nil", got, got) - } - want := newInt(42) - p.Store(want) - if got := p.Load(); got != want { - t.Errorf("wrong value: got %p (%v), wanted %p (%v)", got, got, want, want) - } - want = newInt(100) - p.Store(want) - if got := p.Load(); got != want { - t.Errorf("wrong value: got %p (%v), wanted %p (%v)", got, got, want, want) - } -} diff --git a/pkg/syncutil/downgradable_rwmutex_test.go b/pkg/syncutil/downgradable_rwmutex_test.go deleted file mode 100644 index ffaf7ecc7..000000000 --- a/pkg/syncutil/downgradable_rwmutex_test.go +++ /dev/null @@ -1,150 +0,0 @@ -// Copyright 2009 The Go Authors. All rights reserved. -// Copyright 2019 The gVisor Authors. -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -// GOMAXPROCS=10 go test - -// Copy/pasted from the standard library's sync/rwmutex_test.go, except for the -// addition of downgradingWriter and the renaming of num_iterations to -// numIterations to shut up Golint. - -package syncutil - -import ( - "fmt" - "runtime" - "sync/atomic" - "testing" -) - -func parallelReader(m *DowngradableRWMutex, clocked, cunlock, cdone chan bool) { - m.RLock() - clocked <- true - <-cunlock - m.RUnlock() - cdone <- true -} - -func doTestParallelReaders(numReaders, gomaxprocs int) { - runtime.GOMAXPROCS(gomaxprocs) - var m DowngradableRWMutex - clocked := make(chan bool) - cunlock := make(chan bool) - cdone := make(chan bool) - for i := 0; i < numReaders; i++ { - go parallelReader(&m, clocked, cunlock, cdone) - } - // Wait for all parallel RLock()s to succeed. - for i := 0; i < numReaders; i++ { - <-clocked - } - for i := 0; i < numReaders; i++ { - cunlock <- true - } - // Wait for the goroutines to finish. - for i := 0; i < numReaders; i++ { - <-cdone - } -} - -func TestParallelReaders(t *testing.T) { - defer runtime.GOMAXPROCS(runtime.GOMAXPROCS(-1)) - doTestParallelReaders(1, 4) - doTestParallelReaders(3, 4) - doTestParallelReaders(4, 2) -} - -func reader(rwm *DowngradableRWMutex, numIterations int, activity *int32, cdone chan bool) { - for i := 0; i < numIterations; i++ { - rwm.RLock() - n := atomic.AddInt32(activity, 1) - if n < 1 || n >= 10000 { - panic(fmt.Sprintf("wlock(%d)\n", n)) - } - for i := 0; i < 100; i++ { - } - atomic.AddInt32(activity, -1) - rwm.RUnlock() - } - cdone <- true -} - -func writer(rwm *DowngradableRWMutex, numIterations int, activity *int32, cdone chan bool) { - for i := 0; i < numIterations; i++ { - rwm.Lock() - n := atomic.AddInt32(activity, 10000) - if n != 10000 { - panic(fmt.Sprintf("wlock(%d)\n", n)) - } - for i := 0; i < 100; i++ { - } - atomic.AddInt32(activity, -10000) - rwm.Unlock() - } - cdone <- true -} - -func downgradingWriter(rwm *DowngradableRWMutex, numIterations int, activity *int32, cdone chan bool) { - for i := 0; i < numIterations; i++ { - rwm.Lock() - n := atomic.AddInt32(activity, 10000) - if n != 10000 { - panic(fmt.Sprintf("wlock(%d)\n", n)) - } - for i := 0; i < 100; i++ { - } - atomic.AddInt32(activity, -10000) - rwm.DowngradeLock() - n = atomic.AddInt32(activity, 1) - if n < 1 || n >= 10000 { - panic(fmt.Sprintf("wlock(%d)\n", n)) - } - for i := 0; i < 100; i++ { - } - n = atomic.AddInt32(activity, -1) - rwm.RUnlock() - } - cdone <- true -} - -func HammerDowngradableRWMutex(gomaxprocs, numReaders, numIterations int) { - runtime.GOMAXPROCS(gomaxprocs) - // Number of active readers + 10000 * number of active writers. - var activity int32 - var rwm DowngradableRWMutex - cdone := make(chan bool) - go writer(&rwm, numIterations, &activity, cdone) - go downgradingWriter(&rwm, numIterations, &activity, cdone) - var i int - for i = 0; i < numReaders/2; i++ { - go reader(&rwm, numIterations, &activity, cdone) - } - go writer(&rwm, numIterations, &activity, cdone) - go downgradingWriter(&rwm, numIterations, &activity, cdone) - for ; i < numReaders; i++ { - go reader(&rwm, numIterations, &activity, cdone) - } - // Wait for the 4 writers and all readers to finish. - for i := 0; i < 4+numReaders; i++ { - <-cdone - } -} - -func TestDowngradableRWMutex(t *testing.T) { - defer runtime.GOMAXPROCS(runtime.GOMAXPROCS(-1)) - n := 1000 - if testing.Short() { - n = 5 - } - HammerDowngradableRWMutex(1, 1, n) - HammerDowngradableRWMutex(1, 3, n) - HammerDowngradableRWMutex(1, 10, n) - HammerDowngradableRWMutex(4, 1, n) - HammerDowngradableRWMutex(4, 3, n) - HammerDowngradableRWMutex(4, 10, n) - HammerDowngradableRWMutex(10, 1, n) - HammerDowngradableRWMutex(10, 3, n) - HammerDowngradableRWMutex(10, 10, n) - HammerDowngradableRWMutex(10, 5, n) -} diff --git a/pkg/syncutil/downgradable_rwmutex_unsafe.go b/pkg/syncutil/downgradable_rwmutex_unsafe.go deleted file mode 100644 index 51e11555d..000000000 --- a/pkg/syncutil/downgradable_rwmutex_unsafe.go +++ /dev/null @@ -1,146 +0,0 @@ -// Copyright 2009 The Go Authors. All rights reserved. -// Copyright 2019 The gVisor Authors. -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -// +build go1.13 -// +build !go1.15 - -// Check go:linkname function signatures when updating Go version. - -// This is mostly copied from the standard library's sync/rwmutex.go. -// -// Happens-before relationships indicated to the race detector: -// - Unlock -> Lock (via writerSem) -// - Unlock -> RLock (via readerSem) -// - RUnlock -> Lock (via writerSem) -// - DowngradeLock -> RLock (via readerSem) - -package syncutil - -import ( - "sync" - "sync/atomic" - "unsafe" -) - -//go:linkname runtimeSemacquire sync.runtime_Semacquire -func runtimeSemacquire(s *uint32) - -//go:linkname runtimeSemrelease sync.runtime_Semrelease -func runtimeSemrelease(s *uint32, handoff bool, skipframes int) - -// DowngradableRWMutex is identical to sync.RWMutex, but adds the DowngradeLock -// method. -type DowngradableRWMutex struct { - w sync.Mutex // held if there are pending writers - writerSem uint32 // semaphore for writers to wait for completing readers - readerSem uint32 // semaphore for readers to wait for completing writers - readerCount int32 // number of pending readers - readerWait int32 // number of departing readers -} - -const rwmutexMaxReaders = 1 << 30 - -// RLock locks rw for reading. -func (rw *DowngradableRWMutex) RLock() { - if RaceEnabled { - RaceDisable() - } - if atomic.AddInt32(&rw.readerCount, 1) < 0 { - // A writer is pending, wait for it. - runtimeSemacquire(&rw.readerSem) - } - if RaceEnabled { - RaceEnable() - RaceAcquire(unsafe.Pointer(&rw.readerSem)) - } -} - -// RUnlock undoes a single RLock call. -func (rw *DowngradableRWMutex) RUnlock() { - if RaceEnabled { - RaceReleaseMerge(unsafe.Pointer(&rw.writerSem)) - RaceDisable() - } - if r := atomic.AddInt32(&rw.readerCount, -1); r < 0 { - if r+1 == 0 || r+1 == -rwmutexMaxReaders { - panic("RUnlock of unlocked DowngradableRWMutex") - } - // A writer is pending. - if atomic.AddInt32(&rw.readerWait, -1) == 0 { - // The last reader unblocks the writer. - runtimeSemrelease(&rw.writerSem, false, 0) - } - } - if RaceEnabled { - RaceEnable() - } -} - -// Lock locks rw for writing. -func (rw *DowngradableRWMutex) Lock() { - if RaceEnabled { - RaceDisable() - } - // First, resolve competition with other writers. - rw.w.Lock() - // Announce to readers there is a pending writer. - r := atomic.AddInt32(&rw.readerCount, -rwmutexMaxReaders) + rwmutexMaxReaders - // Wait for active readers. - if r != 0 && atomic.AddInt32(&rw.readerWait, r) != 0 { - runtimeSemacquire(&rw.writerSem) - } - if RaceEnabled { - RaceEnable() - RaceAcquire(unsafe.Pointer(&rw.writerSem)) - } -} - -// Unlock unlocks rw for writing. -func (rw *DowngradableRWMutex) Unlock() { - if RaceEnabled { - RaceRelease(unsafe.Pointer(&rw.writerSem)) - RaceRelease(unsafe.Pointer(&rw.readerSem)) - RaceDisable() - } - // Announce to readers there is no active writer. - r := atomic.AddInt32(&rw.readerCount, rwmutexMaxReaders) - if r >= rwmutexMaxReaders { - panic("Unlock of unlocked DowngradableRWMutex") - } - // Unblock blocked readers, if any. - for i := 0; i < int(r); i++ { - runtimeSemrelease(&rw.readerSem, false, 0) - } - // Allow other writers to proceed. - rw.w.Unlock() - if RaceEnabled { - RaceEnable() - } -} - -// DowngradeLock atomically unlocks rw for writing and locks it for reading. -func (rw *DowngradableRWMutex) DowngradeLock() { - if RaceEnabled { - RaceRelease(unsafe.Pointer(&rw.readerSem)) - RaceDisable() - } - // Announce to readers there is no active writer and one additional reader. - r := atomic.AddInt32(&rw.readerCount, rwmutexMaxReaders+1) - if r >= rwmutexMaxReaders+1 { - panic("DowngradeLock of unlocked DowngradableRWMutex") - } - // Unblock blocked readers, if any. Note that this loop starts as 1 since r - // includes this goroutine. - for i := 1; i < int(r); i++ { - runtimeSemrelease(&rw.readerSem, false, 0) - } - // Allow other writers to proceed to rw.w.Lock(). Note that they will still - // block on rw.writerSem since at least this reader exists, such that - // DowngradeLock() is atomic with the previous write lock. - rw.w.Unlock() - if RaceEnabled { - RaceEnable() - } -} diff --git a/pkg/syncutil/memmove_unsafe.go b/pkg/syncutil/memmove_unsafe.go deleted file mode 100644 index 348675baa..000000000 --- a/pkg/syncutil/memmove_unsafe.go +++ /dev/null @@ -1,28 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -// +build go1.12 -// +build !go1.15 - -// Check go:linkname function signatures when updating Go version. - -package syncutil - -import ( - "unsafe" -) - -//go:linkname memmove runtime.memmove -//go:noescape -func memmove(to, from unsafe.Pointer, n uintptr) - -// Memmove is exported for SeqAtomicLoad/SeqAtomicTryLoad, which can't -// define it because go_generics can't update the go:linkname annotation. -// Furthermore, go:linkname silently doesn't work if the local name is exported -// (this is of course undocumented), which is why this indirection is -// necessary. -func Memmove(to, from unsafe.Pointer, n uintptr) { - memmove(to, from, n) -} diff --git a/pkg/syncutil/norace_unsafe.go b/pkg/syncutil/norace_unsafe.go deleted file mode 100644 index 0a0a9deda..000000000 --- a/pkg/syncutil/norace_unsafe.go +++ /dev/null @@ -1,35 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -// +build !race - -package syncutil - -import ( - "unsafe" -) - -// RaceEnabled is true if the Go data race detector is enabled. -const RaceEnabled = false - -// RaceDisable has the same semantics as runtime.RaceDisable. -func RaceDisable() { -} - -// RaceEnable has the same semantics as runtime.RaceEnable. -func RaceEnable() { -} - -// RaceAcquire has the same semantics as runtime.RaceAcquire. -func RaceAcquire(addr unsafe.Pointer) { -} - -// RaceRelease has the same semantics as runtime.RaceRelease. -func RaceRelease(addr unsafe.Pointer) { -} - -// RaceReleaseMerge has the same semantics as runtime.RaceReleaseMerge. -func RaceReleaseMerge(addr unsafe.Pointer) { -} diff --git a/pkg/syncutil/race_unsafe.go b/pkg/syncutil/race_unsafe.go deleted file mode 100644 index 206067ec1..000000000 --- a/pkg/syncutil/race_unsafe.go +++ /dev/null @@ -1,41 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -// +build race - -package syncutil - -import ( - "runtime" - "unsafe" -) - -// RaceEnabled is true if the Go data race detector is enabled. -const RaceEnabled = true - -// RaceDisable has the same semantics as runtime.RaceDisable. -func RaceDisable() { - runtime.RaceDisable() -} - -// RaceEnable has the same semantics as runtime.RaceEnable. -func RaceEnable() { - runtime.RaceEnable() -} - -// RaceAcquire has the same semantics as runtime.RaceAcquire. -func RaceAcquire(addr unsafe.Pointer) { - runtime.RaceAcquire(addr) -} - -// RaceRelease has the same semantics as runtime.RaceRelease. -func RaceRelease(addr unsafe.Pointer) { - runtime.RaceRelease(addr) -} - -// RaceReleaseMerge has the same semantics as runtime.RaceReleaseMerge. -func RaceReleaseMerge(addr unsafe.Pointer) { - runtime.RaceReleaseMerge(addr) -} diff --git a/pkg/syncutil/seqatomic_unsafe.go b/pkg/syncutil/seqatomic_unsafe.go deleted file mode 100644 index cb6d2eb22..000000000 --- a/pkg/syncutil/seqatomic_unsafe.go +++ /dev/null @@ -1,72 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -// Package template doesn't exist. This file must be instantiated using the -// go_template_instance rule in tools/go_generics/defs.bzl. -package template - -import ( - "fmt" - "reflect" - "strings" - "unsafe" - - "gvisor.dev/gvisor/pkg/syncutil" -) - -// Value is a required type parameter. -// -// Value must not contain any pointers, including interface objects, function -// objects, slices, maps, channels, unsafe.Pointer, and arrays or structs -// containing any of the above. An init() function will panic if this property -// does not hold. -type Value struct{} - -// SeqAtomicLoad returns a copy of *ptr, ensuring that the read does not race -// with any writer critical sections in sc. -func SeqAtomicLoad(sc *syncutil.SeqCount, ptr *Value) Value { - // This function doesn't use SeqAtomicTryLoad because doing so is - // measurably, significantly (~20%) slower; Go is awful at inlining. - var val Value - for { - epoch := sc.BeginRead() - if syncutil.RaceEnabled { - // runtime.RaceDisable() doesn't actually stop the race detector, - // so it can't help us here. Instead, call runtime.memmove - // directly, which is not instrumented by the race detector. - syncutil.Memmove(unsafe.Pointer(&val), unsafe.Pointer(ptr), unsafe.Sizeof(val)) - } else { - // This is ~40% faster for short reads than going through memmove. - val = *ptr - } - if sc.ReadOk(epoch) { - break - } - } - return val -} - -// SeqAtomicTryLoad returns a copy of *ptr while in a reader critical section -// in sc initiated by a call to sc.BeginRead() that returned epoch. If the read -// would race with a writer critical section, SeqAtomicTryLoad returns -// (unspecified, false). -func SeqAtomicTryLoad(sc *syncutil.SeqCount, epoch syncutil.SeqCountEpoch, ptr *Value) (Value, bool) { - var val Value - if syncutil.RaceEnabled { - syncutil.Memmove(unsafe.Pointer(&val), unsafe.Pointer(ptr), unsafe.Sizeof(val)) - } else { - val = *ptr - } - return val, sc.ReadOk(epoch) -} - -func init() { - var val Value - typ := reflect.TypeOf(val) - name := typ.Name() - if ptrs := syncutil.PointersInType(typ, name); len(ptrs) != 0 { - panic(fmt.Sprintf("SeqAtomicLoad<%s> is invalid since values %s of type %s contain pointers:\n%s", typ, name, typ, strings.Join(ptrs, "\n"))) - } -} diff --git a/pkg/syncutil/seqatomictest/BUILD b/pkg/syncutil/seqatomictest/BUILD deleted file mode 100644 index ba18f3238..000000000 --- a/pkg/syncutil/seqatomictest/BUILD +++ /dev/null @@ -1,35 +0,0 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -package(licenses = ["notice"]) - -go_template_instance( - name = "seqatomic_int", - out = "seqatomic_int_unsafe.go", - package = "seqatomic", - suffix = "Int", - template = "//pkg/syncutil:generic_seqatomic", - types = { - "Value": "int", - }, -) - -go_library( - name = "seqatomic", - srcs = ["seqatomic_int_unsafe.go"], - importpath = "gvisor.dev/gvisor/pkg/syncutil/seqatomic", - deps = [ - "//pkg/syncutil", - ], -) - -go_test( - name = "seqatomic_test", - size = "small", - srcs = ["seqatomic_test.go"], - embed = [":seqatomic"], - deps = [ - "//pkg/syncutil", - ], -) diff --git a/pkg/syncutil/seqatomictest/seqatomic_test.go b/pkg/syncutil/seqatomictest/seqatomic_test.go deleted file mode 100644 index b0db44999..000000000 --- a/pkg/syncutil/seqatomictest/seqatomic_test.go +++ /dev/null @@ -1,132 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package seqatomic - -import ( - "sync/atomic" - "testing" - "time" - - "gvisor.dev/gvisor/pkg/syncutil" -) - -func TestSeqAtomicLoadUncontended(t *testing.T) { - var seq syncutil.SeqCount - const want = 1 - data := want - if got := SeqAtomicLoadInt(&seq, &data); got != want { - t.Errorf("SeqAtomicLoadInt: got %v, wanted %v", got, want) - } -} - -func TestSeqAtomicLoadAfterWrite(t *testing.T) { - var seq syncutil.SeqCount - var data int - const want = 1 - seq.BeginWrite() - data = want - seq.EndWrite() - if got := SeqAtomicLoadInt(&seq, &data); got != want { - t.Errorf("SeqAtomicLoadInt: got %v, wanted %v", got, want) - } -} - -func TestSeqAtomicLoadDuringWrite(t *testing.T) { - var seq syncutil.SeqCount - var data int - const want = 1 - seq.BeginWrite() - go func() { - time.Sleep(time.Second) - data = want - seq.EndWrite() - }() - if got := SeqAtomicLoadInt(&seq, &data); got != want { - t.Errorf("SeqAtomicLoadInt: got %v, wanted %v", got, want) - } -} - -func TestSeqAtomicTryLoadUncontended(t *testing.T) { - var seq syncutil.SeqCount - const want = 1 - data := want - epoch := seq.BeginRead() - if got, ok := SeqAtomicTryLoadInt(&seq, epoch, &data); !ok || got != want { - t.Errorf("SeqAtomicTryLoadInt: got (%v, %v), wanted (%v, true)", got, ok, want) - } -} - -func TestSeqAtomicTryLoadDuringWrite(t *testing.T) { - var seq syncutil.SeqCount - var data int - epoch := seq.BeginRead() - seq.BeginWrite() - if got, ok := SeqAtomicTryLoadInt(&seq, epoch, &data); ok { - t.Errorf("SeqAtomicTryLoadInt: got (%v, true), wanted (_, false)", got) - } - seq.EndWrite() -} - -func TestSeqAtomicTryLoadAfterWrite(t *testing.T) { - var seq syncutil.SeqCount - var data int - epoch := seq.BeginRead() - seq.BeginWrite() - seq.EndWrite() - if got, ok := SeqAtomicTryLoadInt(&seq, epoch, &data); ok { - t.Errorf("SeqAtomicTryLoadInt: got (%v, true), wanted (_, false)", got) - } -} - -func BenchmarkSeqAtomicLoadIntUncontended(b *testing.B) { - var seq syncutil.SeqCount - const want = 42 - data := want - b.RunParallel(func(pb *testing.PB) { - for pb.Next() { - if got := SeqAtomicLoadInt(&seq, &data); got != want { - b.Fatalf("SeqAtomicLoadInt: got %v, wanted %v", got, want) - } - } - }) -} - -func BenchmarkSeqAtomicTryLoadIntUncontended(b *testing.B) { - var seq syncutil.SeqCount - const want = 42 - data := want - b.RunParallel(func(pb *testing.PB) { - epoch := seq.BeginRead() - for pb.Next() { - if got, ok := SeqAtomicTryLoadInt(&seq, epoch, &data); !ok || got != want { - b.Fatalf("SeqAtomicTryLoadInt: got (%v, %v), wanted (%v, true)", got, ok, want) - } - } - }) -} - -// For comparison: -func BenchmarkAtomicValueLoadIntUncontended(b *testing.B) { - var a atomic.Value - const want = 42 - a.Store(int(want)) - b.RunParallel(func(pb *testing.PB) { - for pb.Next() { - if got := a.Load().(int); got != want { - b.Fatalf("atomic.Value.Load: got %v, wanted %v", got, want) - } - } - }) -} diff --git a/pkg/syncutil/seqcount.go b/pkg/syncutil/seqcount.go deleted file mode 100644 index 11d8dbfaa..000000000 --- a/pkg/syncutil/seqcount.go +++ /dev/null @@ -1,149 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -package syncutil - -import ( - "fmt" - "reflect" - "runtime" - "sync/atomic" -) - -// SeqCount is a synchronization primitive for optimistic reader/writer -// synchronization in cases where readers can work with stale data and -// therefore do not need to block writers. -// -// Compared to sync/atomic.Value: -// -// - Mutation of SeqCount-protected data does not require memory allocation, -// whereas atomic.Value generally does. This is a significant advantage when -// writes are common. -// -// - Atomic reads of SeqCount-protected data require copying. This is a -// disadvantage when atomic reads are common. -// -// - SeqCount may be more flexible: correct use of SeqCount.ReadOk allows other -// operations to be made atomic with reads of SeqCount-protected data. -// -// - SeqCount may be less flexible: as of this writing, SeqCount-protected data -// cannot include pointers. -// -// - SeqCount is more cumbersome to use; atomic reads of SeqCount-protected -// data require instantiating function templates using go_generics (see -// seqatomic.go). -type SeqCount struct { - // epoch is incremented by BeginWrite and EndWrite, such that epoch is odd - // if a writer critical section is active, and a read from data protected - // by this SeqCount is atomic iff epoch is the same even value before and - // after the read. - epoch uint32 -} - -// SeqCountEpoch tracks writer critical sections in a SeqCount. -type SeqCountEpoch struct { - val uint32 -} - -// We assume that: -// -// - All functions in sync/atomic that perform a memory read are at least a -// read fence: memory reads before calls to such functions cannot be reordered -// after the call, and memory reads after calls to such functions cannot be -// reordered before the call, even if those reads do not use sync/atomic. -// -// - All functions in sync/atomic that perform a memory write are at least a -// write fence: memory writes before calls to such functions cannot be -// reordered after the call, and memory writes after calls to such functions -// cannot be reordered before the call, even if those writes do not use -// sync/atomic. -// -// As of this writing, the Go memory model completely fails to describe -// sync/atomic, but these properties are implied by -// https://groups.google.com/forum/#!topic/golang-nuts/7EnEhM3U7B8. - -// BeginRead indicates the beginning of a reader critical section. Reader -// critical sections DO NOT BLOCK writer critical sections, so operations in a -// reader critical section MAY RACE with writer critical sections. Races are -// detected by ReadOk at the end of the reader critical section. Thus, the -// low-level structure of readers is generally: -// -// for { -// epoch := seq.BeginRead() -// // do something idempotent with seq-protected data -// if seq.ReadOk(epoch) { -// break -// } -// } -// -// However, since reader critical sections may race with writer critical -// sections, the Go race detector will (accurately) flag data races in readers -// using this pattern. Most users of SeqCount will need to use the -// SeqAtomicLoad function template in seqatomic.go. -func (s *SeqCount) BeginRead() SeqCountEpoch { - epoch := atomic.LoadUint32(&s.epoch) - for epoch&1 != 0 { - runtime.Gosched() - epoch = atomic.LoadUint32(&s.epoch) - } - return SeqCountEpoch{epoch} -} - -// ReadOk returns true if the reader critical section initiated by a previous -// call to BeginRead() that returned epoch did not race with any writer critical -// sections. -// -// ReadOk may be called any number of times during a reader critical section. -// Reader critical sections do not need to be explicitly terminated; the last -// call to ReadOk is implicitly the end of the reader critical section. -func (s *SeqCount) ReadOk(epoch SeqCountEpoch) bool { - return atomic.LoadUint32(&s.epoch) == epoch.val -} - -// BeginWrite indicates the beginning of a writer critical section. -// -// SeqCount does not support concurrent writer critical sections; clients with -// concurrent writers must synchronize them using e.g. sync.Mutex. -func (s *SeqCount) BeginWrite() { - if epoch := atomic.AddUint32(&s.epoch, 1); epoch&1 == 0 { - panic("SeqCount.BeginWrite during writer critical section") - } -} - -// EndWrite ends the effect of a preceding BeginWrite. -func (s *SeqCount) EndWrite() { - if epoch := atomic.AddUint32(&s.epoch, 1); epoch&1 != 0 { - panic("SeqCount.EndWrite outside writer critical section") - } -} - -// PointersInType returns a list of pointers reachable from values named -// valName of the given type. -// -// PointersInType is not exhaustive, but it is guaranteed that if typ contains -// at least one pointer, then PointersInTypeOf returns a non-empty list. -func PointersInType(typ reflect.Type, valName string) []string { - switch kind := typ.Kind(); kind { - case reflect.Bool, reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64, reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr, reflect.Float32, reflect.Float64, reflect.Complex64, reflect.Complex128: - return nil - - case reflect.Chan, reflect.Func, reflect.Interface, reflect.Map, reflect.Ptr, reflect.Slice, reflect.String, reflect.UnsafePointer: - return []string{valName} - - case reflect.Array: - return PointersInType(typ.Elem(), valName+"[]") - - case reflect.Struct: - var ptrs []string - for i, n := 0, typ.NumField(); i < n; i++ { - field := typ.Field(i) - ptrs = append(ptrs, PointersInType(field.Type, fmt.Sprintf("%s.%s", valName, field.Name))...) - } - return ptrs - - default: - return []string{fmt.Sprintf("%s (of type %s with unknown kind %s)", valName, typ, kind)} - } -} diff --git a/pkg/syncutil/seqcount_test.go b/pkg/syncutil/seqcount_test.go deleted file mode 100644 index 14d6aedea..000000000 --- a/pkg/syncutil/seqcount_test.go +++ /dev/null @@ -1,153 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -package syncutil - -import ( - "reflect" - "testing" - "time" -) - -func TestSeqCountWriteUncontended(t *testing.T) { - var seq SeqCount - seq.BeginWrite() - seq.EndWrite() -} - -func TestSeqCountReadUncontended(t *testing.T) { - var seq SeqCount - epoch := seq.BeginRead() - if !seq.ReadOk(epoch) { - t.Errorf("ReadOk: got false, wanted true") - } -} - -func TestSeqCountBeginReadAfterWrite(t *testing.T) { - var seq SeqCount - var data int32 - const want = 1 - seq.BeginWrite() - data = want - seq.EndWrite() - epoch := seq.BeginRead() - if data != want { - t.Errorf("Reader: got %v, wanted %v", data, want) - } - if !seq.ReadOk(epoch) { - t.Errorf("ReadOk: got false, wanted true") - } -} - -func TestSeqCountBeginReadDuringWrite(t *testing.T) { - var seq SeqCount - var data int - const want = 1 - seq.BeginWrite() - go func() { - time.Sleep(time.Second) - data = want - seq.EndWrite() - }() - epoch := seq.BeginRead() - if data != want { - t.Errorf("Reader: got %v, wanted %v", data, want) - } - if !seq.ReadOk(epoch) { - t.Errorf("ReadOk: got false, wanted true") - } -} - -func TestSeqCountReadOkAfterWrite(t *testing.T) { - var seq SeqCount - epoch := seq.BeginRead() - seq.BeginWrite() - seq.EndWrite() - if seq.ReadOk(epoch) { - t.Errorf("ReadOk: got true, wanted false") - } -} - -func TestSeqCountReadOkDuringWrite(t *testing.T) { - var seq SeqCount - epoch := seq.BeginRead() - seq.BeginWrite() - if seq.ReadOk(epoch) { - t.Errorf("ReadOk: got true, wanted false") - } - seq.EndWrite() -} - -func BenchmarkSeqCountWriteUncontended(b *testing.B) { - var seq SeqCount - for i := 0; i < b.N; i++ { - seq.BeginWrite() - seq.EndWrite() - } -} - -func BenchmarkSeqCountReadUncontended(b *testing.B) { - var seq SeqCount - b.RunParallel(func(pb *testing.PB) { - for pb.Next() { - epoch := seq.BeginRead() - if !seq.ReadOk(epoch) { - b.Fatalf("ReadOk: got false, wanted true") - } - } - }) -} - -func TestPointersInType(t *testing.T) { - for _, test := range []struct { - name string // used for both test and value name - val interface{} - ptrs []string - }{ - { - name: "EmptyStruct", - val: struct{}{}, - }, - { - name: "Int", - val: int(0), - }, - { - name: "MixedStruct", - val: struct { - b bool - I int - ExportedPtr *struct{} - unexportedPtr *struct{} - arr [2]int - ptrArr [2]*int - nestedStruct struct { - nestedNonptr int - nestedPtr *int - } - structArr [1]struct { - nonptr int - ptr *int - } - }{}, - ptrs: []string{ - "MixedStruct.ExportedPtr", - "MixedStruct.unexportedPtr", - "MixedStruct.ptrArr[]", - "MixedStruct.nestedStruct.nestedPtr", - "MixedStruct.structArr[].ptr", - }, - }, - } { - t.Run(test.name, func(t *testing.T) { - typ := reflect.TypeOf(test.val) - ptrs := PointersInType(typ, test.name) - t.Logf("Found pointers: %v", ptrs) - if (len(ptrs) != 0 || len(test.ptrs) != 0) && !reflect.DeepEqual(ptrs, test.ptrs) { - t.Errorf("Got %v, wanted %v", ptrs, test.ptrs) - } - }) - } -} diff --git a/pkg/syncutil/syncutil.go b/pkg/syncutil/syncutil.go deleted file mode 100644 index 66e750d06..000000000 --- a/pkg/syncutil/syncutil.go +++ /dev/null @@ -1,7 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -// Package syncutil provides synchronization primitives. -package syncutil diff --git a/pkg/tcpip/BUILD b/pkg/tcpip/BUILD index e07ebd153..db06d02c6 100644 --- a/pkg/tcpip/BUILD +++ b/pkg/tcpip/BUILD @@ -15,6 +15,7 @@ go_library( importpath = "gvisor.dev/gvisor/pkg/tcpip", visibility = ["//visibility:public"], deps = [ + "//pkg/sync", "//pkg/tcpip/buffer", "//pkg/tcpip/iptables", "//pkg/waiter", diff --git a/pkg/tcpip/adapters/gonet/BUILD b/pkg/tcpip/adapters/gonet/BUILD index 78df5a0b1..3df7d18d3 100644 --- a/pkg/tcpip/adapters/gonet/BUILD +++ b/pkg/tcpip/adapters/gonet/BUILD @@ -9,6 +9,7 @@ go_library( importpath = "gvisor.dev/gvisor/pkg/tcpip/adapters/gonet", visibility = ["//visibility:public"], deps = [ + "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/stack", diff --git a/pkg/tcpip/adapters/gonet/gonet.go b/pkg/tcpip/adapters/gonet/gonet.go index cd6ce930a..a2f44b496 100644 --- a/pkg/tcpip/adapters/gonet/gonet.go +++ b/pkg/tcpip/adapters/gonet/gonet.go @@ -20,9 +20,9 @@ import ( "errors" "io" "net" - "sync" "time" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/stack" diff --git a/pkg/tcpip/link/fdbased/BUILD b/pkg/tcpip/link/fdbased/BUILD index 897c94821..66cc53ed4 100644 --- a/pkg/tcpip/link/fdbased/BUILD +++ b/pkg/tcpip/link/fdbased/BUILD @@ -16,6 +16,7 @@ go_library( importpath = "gvisor.dev/gvisor/pkg/tcpip/link/fdbased", visibility = ["//visibility:public"], deps = [ + "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/header", diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go index fa8a703d9..b7f60178e 100644 --- a/pkg/tcpip/link/fdbased/endpoint.go +++ b/pkg/tcpip/link/fdbased/endpoint.go @@ -41,10 +41,10 @@ package fdbased import ( "fmt" - "sync" "syscall" "golang.org/x/sys/unix" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" diff --git a/pkg/tcpip/link/sharedmem/BUILD b/pkg/tcpip/link/sharedmem/BUILD index a4f9cdd69..09165dd4c 100644 --- a/pkg/tcpip/link/sharedmem/BUILD +++ b/pkg/tcpip/link/sharedmem/BUILD @@ -15,6 +15,7 @@ go_library( visibility = ["//visibility:public"], deps = [ "//pkg/log", + "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/header", @@ -31,6 +32,7 @@ go_test( ], embed = [":sharedmem"], deps = [ + "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/header", diff --git a/pkg/tcpip/link/sharedmem/pipe/BUILD b/pkg/tcpip/link/sharedmem/pipe/BUILD index 6b5bc542c..a0d4ad0be 100644 --- a/pkg/tcpip/link/sharedmem/pipe/BUILD +++ b/pkg/tcpip/link/sharedmem/pipe/BUILD @@ -21,4 +21,5 @@ go_test( "pipe_test.go", ], embed = [":pipe"], + deps = ["//pkg/sync"], ) diff --git a/pkg/tcpip/link/sharedmem/pipe/pipe_test.go b/pkg/tcpip/link/sharedmem/pipe/pipe_test.go index 59ef69a8b..dc239a0d0 100644 --- a/pkg/tcpip/link/sharedmem/pipe/pipe_test.go +++ b/pkg/tcpip/link/sharedmem/pipe/pipe_test.go @@ -18,8 +18,9 @@ import ( "math/rand" "reflect" "runtime" - "sync" "testing" + + "gvisor.dev/gvisor/pkg/sync" ) func TestSimpleReadWrite(t *testing.T) { diff --git a/pkg/tcpip/link/sharedmem/sharedmem.go b/pkg/tcpip/link/sharedmem/sharedmem.go index 080f9d667..655e537c4 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem.go +++ b/pkg/tcpip/link/sharedmem/sharedmem.go @@ -23,11 +23,11 @@ package sharedmem import ( - "sync" "sync/atomic" "syscall" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" diff --git a/pkg/tcpip/link/sharedmem/sharedmem_test.go b/pkg/tcpip/link/sharedmem/sharedmem_test.go index 89603c48f..5c729a439 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem_test.go +++ b/pkg/tcpip/link/sharedmem/sharedmem_test.go @@ -22,11 +22,11 @@ import ( "math/rand" "os" "strings" - "sync" "syscall" "testing" "time" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" diff --git a/pkg/tcpip/network/fragmentation/BUILD b/pkg/tcpip/network/fragmentation/BUILD index acf1e022c..ed16076fd 100644 --- a/pkg/tcpip/network/fragmentation/BUILD +++ b/pkg/tcpip/network/fragmentation/BUILD @@ -28,6 +28,7 @@ go_library( visibility = ["//visibility:public"], deps = [ "//pkg/log", + "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", ], diff --git a/pkg/tcpip/network/fragmentation/fragmentation.go b/pkg/tcpip/network/fragmentation/fragmentation.go index 6da5238ec..92f2aa13a 100644 --- a/pkg/tcpip/network/fragmentation/fragmentation.go +++ b/pkg/tcpip/network/fragmentation/fragmentation.go @@ -19,9 +19,9 @@ package fragmentation import ( "fmt" "log" - "sync" "time" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip/buffer" ) diff --git a/pkg/tcpip/network/fragmentation/reassembler.go b/pkg/tcpip/network/fragmentation/reassembler.go index 9e002e396..0a83d81f2 100644 --- a/pkg/tcpip/network/fragmentation/reassembler.go +++ b/pkg/tcpip/network/fragmentation/reassembler.go @@ -18,9 +18,9 @@ import ( "container/heap" "fmt" "math" - "sync" "time" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip/buffer" ) diff --git a/pkg/tcpip/ports/BUILD b/pkg/tcpip/ports/BUILD index e156b01f6..a6ef3bdcc 100644 --- a/pkg/tcpip/ports/BUILD +++ b/pkg/tcpip/ports/BUILD @@ -9,6 +9,7 @@ go_library( importpath = "gvisor.dev/gvisor/pkg/tcpip/ports", visibility = ["//visibility:public"], deps = [ + "//pkg/sync", "//pkg/tcpip", ], ) diff --git a/pkg/tcpip/ports/ports.go b/pkg/tcpip/ports/ports.go index 6c5e19e8f..b937cb84b 100644 --- a/pkg/tcpip/ports/ports.go +++ b/pkg/tcpip/ports/ports.go @@ -18,9 +18,9 @@ package ports import ( "math" "math/rand" - "sync" "sync/atomic" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" ) diff --git a/pkg/tcpip/stack/BUILD b/pkg/tcpip/stack/BUILD index 826fca4de..6a8654105 100644 --- a/pkg/tcpip/stack/BUILD +++ b/pkg/tcpip/stack/BUILD @@ -36,6 +36,7 @@ go_library( "//pkg/ilist", "//pkg/rand", "//pkg/sleep", + "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/hash/jenkins", @@ -80,6 +81,7 @@ go_test( embed = [":stack"], deps = [ "//pkg/sleep", + "//pkg/sync", "//pkg/tcpip", ], ) diff --git a/pkg/tcpip/stack/linkaddrcache.go b/pkg/tcpip/stack/linkaddrcache.go index 267df60d1..403557fd7 100644 --- a/pkg/tcpip/stack/linkaddrcache.go +++ b/pkg/tcpip/stack/linkaddrcache.go @@ -16,10 +16,10 @@ package stack import ( "fmt" - "sync" "time" "gvisor.dev/gvisor/pkg/sleep" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" ) diff --git a/pkg/tcpip/stack/linkaddrcache_test.go b/pkg/tcpip/stack/linkaddrcache_test.go index 9946b8fe8..1baa498d0 100644 --- a/pkg/tcpip/stack/linkaddrcache_test.go +++ b/pkg/tcpip/stack/linkaddrcache_test.go @@ -16,12 +16,12 @@ package stack import ( "fmt" - "sync" "sync/atomic" "testing" "time" "gvisor.dev/gvisor/pkg/sleep" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" ) diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 3810c6602..fe557ccbd 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -16,9 +16,9 @@ package stack import ( "strings" - "sync" "sync/atomic" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index 41bf9fd9b..a47ceba54 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -21,13 +21,13 @@ package stack import ( "encoding/binary" - "sync" "sync/atomic" "time" "golang.org/x/time/rate" "gvisor.dev/gvisor/pkg/rand" "gvisor.dev/gvisor/pkg/sleep" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" diff --git a/pkg/tcpip/stack/transport_demuxer.go b/pkg/tcpip/stack/transport_demuxer.go index 67c21be42..f384a91de 100644 --- a/pkg/tcpip/stack/transport_demuxer.go +++ b/pkg/tcpip/stack/transport_demuxer.go @@ -18,8 +18,8 @@ import ( "fmt" "math/rand" "sort" - "sync" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/hash/jenkins" "gvisor.dev/gvisor/pkg/tcpip/header" diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 72b5ce179..4a090ac86 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -35,10 +35,10 @@ import ( "reflect" "strconv" "strings" - "sync" "sync/atomic" "time" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/iptables" "gvisor.dev/gvisor/pkg/waiter" diff --git a/pkg/tcpip/transport/icmp/BUILD b/pkg/tcpip/transport/icmp/BUILD index d8c5b5058..3aa23d529 100644 --- a/pkg/tcpip/transport/icmp/BUILD +++ b/pkg/tcpip/transport/icmp/BUILD @@ -28,6 +28,7 @@ go_library( visibility = ["//visibility:public"], deps = [ "//pkg/sleep", + "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/header", diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index c7ce74cdd..330786f4c 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -15,8 +15,7 @@ package icmp import ( - "sync" - + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" diff --git a/pkg/tcpip/transport/packet/BUILD b/pkg/tcpip/transport/packet/BUILD index 44b58ff6b..4858d150c 100644 --- a/pkg/tcpip/transport/packet/BUILD +++ b/pkg/tcpip/transport/packet/BUILD @@ -28,6 +28,7 @@ go_library( deps = [ "//pkg/log", "//pkg/sleep", + "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/header", diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go index 07ffa8aba..fc5bc69fa 100644 --- a/pkg/tcpip/transport/packet/endpoint.go +++ b/pkg/tcpip/transport/packet/endpoint.go @@ -25,8 +25,7 @@ package packet import ( - "sync" - + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" diff --git a/pkg/tcpip/transport/raw/BUILD b/pkg/tcpip/transport/raw/BUILD index 00991ac8e..2f2131ff7 100644 --- a/pkg/tcpip/transport/raw/BUILD +++ b/pkg/tcpip/transport/raw/BUILD @@ -29,6 +29,7 @@ go_library( deps = [ "//pkg/log", "//pkg/sleep", + "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/header", diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index 85f7eb76b..ee9c4c58b 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -26,8 +26,7 @@ package raw import ( - "sync" - + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD index 3b353d56c..353bd06f4 100644 --- a/pkg/tcpip/transport/tcp/BUILD +++ b/pkg/tcpip/transport/tcp/BUILD @@ -48,6 +48,7 @@ go_library( "//pkg/log", "//pkg/rand", "//pkg/sleep", + "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/hash/jenkins", diff --git a/pkg/tcpip/transport/tcp/accept.go b/pkg/tcpip/transport/tcp/accept.go index 5422ae80c..1ea996936 100644 --- a/pkg/tcpip/transport/tcp/accept.go +++ b/pkg/tcpip/transport/tcp/accept.go @@ -19,11 +19,11 @@ import ( "encoding/binary" "hash" "io" - "sync" "time" "gvisor.dev/gvisor/pkg/rand" "gvisor.dev/gvisor/pkg/sleep" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index cdd69f360..613ec1775 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -16,11 +16,11 @@ package tcp import ( "encoding/binary" - "sync" "time" "gvisor.dev/gvisor/pkg/rand" "gvisor.dev/gvisor/pkg/sleep" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/hash/jenkins" diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 830bc1e3e..cca511fb9 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -19,12 +19,12 @@ import ( "fmt" "math" "strings" - "sync" "sync/atomic" "time" "gvisor.dev/gvisor/pkg/rand" "gvisor.dev/gvisor/pkg/sleep" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/hash/jenkins" diff --git a/pkg/tcpip/transport/tcp/endpoint_state.go b/pkg/tcpip/transport/tcp/endpoint_state.go index 7aa4c3f0e..4b8d867bc 100644 --- a/pkg/tcpip/transport/tcp/endpoint_state.go +++ b/pkg/tcpip/transport/tcp/endpoint_state.go @@ -16,9 +16,9 @@ package tcp import ( "fmt" - "sync" "time" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" diff --git a/pkg/tcpip/transport/tcp/forwarder.go b/pkg/tcpip/transport/tcp/forwarder.go index 4983bca81..7eb613be5 100644 --- a/pkg/tcpip/transport/tcp/forwarder.go +++ b/pkg/tcpip/transport/tcp/forwarder.go @@ -15,8 +15,7 @@ package tcp import ( - "sync" - + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/seqnum" diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go index bc718064c..9a8f64aa6 100644 --- a/pkg/tcpip/transport/tcp/protocol.go +++ b/pkg/tcpip/transport/tcp/protocol.go @@ -22,9 +22,9 @@ package tcp import ( "strings" - "sync" "time" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" diff --git a/pkg/tcpip/transport/tcp/segment_queue.go b/pkg/tcpip/transport/tcp/segment_queue.go index e0759225e..bd20a7ee9 100644 --- a/pkg/tcpip/transport/tcp/segment_queue.go +++ b/pkg/tcpip/transport/tcp/segment_queue.go @@ -15,7 +15,7 @@ package tcp import ( - "sync" + "gvisor.dev/gvisor/pkg/sync" ) // segmentQueue is a bounded, thread-safe queue of TCP segments. diff --git a/pkg/tcpip/transport/tcp/snd.go b/pkg/tcpip/transport/tcp/snd.go index 8a947dc66..79f2d274b 100644 --- a/pkg/tcpip/transport/tcp/snd.go +++ b/pkg/tcpip/transport/tcp/snd.go @@ -16,11 +16,11 @@ package tcp import ( "math" - "sync" "sync/atomic" "time" "gvisor.dev/gvisor/pkg/sleep" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" diff --git a/pkg/tcpip/transport/udp/BUILD b/pkg/tcpip/transport/udp/BUILD index 97e4d5825..57ff123e3 100644 --- a/pkg/tcpip/transport/udp/BUILD +++ b/pkg/tcpip/transport/udp/BUILD @@ -30,6 +30,7 @@ go_library( visibility = ["//visibility:public"], deps = [ "//pkg/sleep", + "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/header", diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 864dc8733..a4ff29a7d 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -15,8 +15,7 @@ package udp import ( - "sync" - + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" diff --git a/pkg/tmutex/BUILD b/pkg/tmutex/BUILD index 6afdb29b7..07778e4f7 100644 --- a/pkg/tmutex/BUILD +++ b/pkg/tmutex/BUILD @@ -15,4 +15,5 @@ go_test( size = "medium", srcs = ["tmutex_test.go"], embed = [":tmutex"], + deps = ["//pkg/sync"], ) diff --git a/pkg/tmutex/tmutex_test.go b/pkg/tmutex/tmutex_test.go index ce34c7962..05540696a 100644 --- a/pkg/tmutex/tmutex_test.go +++ b/pkg/tmutex/tmutex_test.go @@ -17,10 +17,11 @@ package tmutex import ( "fmt" "runtime" - "sync" "sync/atomic" "testing" "time" + + "gvisor.dev/gvisor/pkg/sync" ) func TestBasicLock(t *testing.T) { diff --git a/pkg/unet/BUILD b/pkg/unet/BUILD index 8f6f180e5..d1885ae66 100644 --- a/pkg/unet/BUILD +++ b/pkg/unet/BUILD @@ -24,4 +24,5 @@ go_test( "unet_test.go", ], embed = [":unet"], + deps = ["//pkg/sync"], ) diff --git a/pkg/unet/unet_test.go b/pkg/unet/unet_test.go index a3cc6f5d3..5c4b9e8e9 100644 --- a/pkg/unet/unet_test.go +++ b/pkg/unet/unet_test.go @@ -19,10 +19,11 @@ import ( "os" "path/filepath" "reflect" - "sync" "syscall" "testing" "time" + + "gvisor.dev/gvisor/pkg/sync" ) func randomFilename() (string, error) { diff --git a/pkg/urpc/BUILD b/pkg/urpc/BUILD index b6bbb0ea2..b8fdc3125 100644 --- a/pkg/urpc/BUILD +++ b/pkg/urpc/BUILD @@ -11,6 +11,7 @@ go_library( deps = [ "//pkg/fd", "//pkg/log", + "//pkg/sync", "//pkg/unet", ], ) diff --git a/pkg/urpc/urpc.go b/pkg/urpc/urpc.go index df59ffab1..13b2ea314 100644 --- a/pkg/urpc/urpc.go +++ b/pkg/urpc/urpc.go @@ -27,10 +27,10 @@ import ( "os" "reflect" "runtime" - "sync" "gvisor.dev/gvisor/pkg/fd" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/unet" ) diff --git a/pkg/waiter/BUILD b/pkg/waiter/BUILD index 0427bc41f..1c6890e52 100644 --- a/pkg/waiter/BUILD +++ b/pkg/waiter/BUILD @@ -24,6 +24,7 @@ go_library( ], importpath = "gvisor.dev/gvisor/pkg/waiter", visibility = ["//visibility:public"], + deps = ["//pkg/sync"], ) go_test( diff --git a/pkg/waiter/waiter.go b/pkg/waiter/waiter.go index 8a65ed164..f708e95fa 100644 --- a/pkg/waiter/waiter.go +++ b/pkg/waiter/waiter.go @@ -58,7 +58,7 @@ package waiter import ( - "sync" + "gvisor.dev/gvisor/pkg/sync" ) // EventMask represents io events as used in the poll() syscall. diff --git a/runsc/boot/BUILD b/runsc/boot/BUILD index 6226b63f8..3e20f8f2f 100644 --- a/runsc/boot/BUILD +++ b/runsc/boot/BUILD @@ -74,6 +74,7 @@ go_library( "//pkg/sentry/usage", "//pkg/sentry/usermem", "//pkg/sentry/watchdog", + "//pkg/sync", "//pkg/syserror", "//pkg/tcpip", "//pkg/tcpip/link/fdbased", @@ -114,6 +115,7 @@ go_test( "//pkg/sentry/context/contexttest", "//pkg/sentry/fs", "//pkg/sentry/kernel/auth", + "//pkg/sync", "//pkg/unet", "//runsc/fsgofer", "@com_github_opencontainers_runtime-spec//specs-go:go_default_library", diff --git a/runsc/boot/compat.go b/runsc/boot/compat.go index 352e710d2..9c23b9553 100644 --- a/runsc/boot/compat.go +++ b/runsc/boot/compat.go @@ -17,7 +17,6 @@ package boot import ( "fmt" "os" - "sync" "syscall" "github.com/golang/protobuf/proto" @@ -27,6 +26,7 @@ import ( ucspb "gvisor.dev/gvisor/pkg/sentry/kernel/uncaught_signal_go_proto" "gvisor.dev/gvisor/pkg/sentry/strace" spb "gvisor.dev/gvisor/pkg/sentry/unimpl/unimplemented_syscall_go_proto" + "gvisor.dev/gvisor/pkg/sync" ) func initCompatLogs(fd int) error { diff --git a/runsc/boot/limits.go b/runsc/boot/limits.go index d1c0bb9b5..ce62236e5 100644 --- a/runsc/boot/limits.go +++ b/runsc/boot/limits.go @@ -16,12 +16,12 @@ package boot import ( "fmt" - "sync" "syscall" specs "github.com/opencontainers/runtime-spec/specs-go" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/limits" + "gvisor.dev/gvisor/pkg/sync" ) // Mapping from linux resource names to limits.LimitType. diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index bc1d0c1bb..fad72f4ab 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -20,7 +20,6 @@ import ( mrand "math/rand" "os" "runtime" - "sync" "sync/atomic" "syscall" gtime "time" @@ -46,6 +45,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/time" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/sentry/watchdog" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/link/sniffer" "gvisor.dev/gvisor/pkg/tcpip/network/arp" diff --git a/runsc/boot/loader_test.go b/runsc/boot/loader_test.go index 147ff7703..bec0dc292 100644 --- a/runsc/boot/loader_test.go +++ b/runsc/boot/loader_test.go @@ -19,7 +19,6 @@ import ( "math/rand" "os" "reflect" - "sync" "syscall" "testing" "time" @@ -30,6 +29,7 @@ import ( "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/sentry/context/contexttest" "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/unet" "gvisor.dev/gvisor/runsc/fsgofer" ) diff --git a/runsc/cmd/BUILD b/runsc/cmd/BUILD index 250845ad7..b94bc4fa0 100644 --- a/runsc/cmd/BUILD +++ b/runsc/cmd/BUILD @@ -44,6 +44,7 @@ go_library( "//pkg/sentry/control", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", + "//pkg/sync", "//pkg/unet", "//pkg/urpc", "//runsc/boot", diff --git a/runsc/cmd/create.go b/runsc/cmd/create.go index a4e3071b3..1815c93b9 100644 --- a/runsc/cmd/create.go +++ b/runsc/cmd/create.go @@ -16,6 +16,7 @@ package cmd import ( "context" + "flag" "github.com/google/subcommands" "gvisor.dev/gvisor/runsc/boot" diff --git a/runsc/cmd/gofer.go b/runsc/cmd/gofer.go index 4831210c0..7df7995f0 100644 --- a/runsc/cmd/gofer.go +++ b/runsc/cmd/gofer.go @@ -21,7 +21,6 @@ import ( "os" "path/filepath" "strings" - "sync" "syscall" "flag" @@ -30,6 +29,7 @@ import ( "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/p9" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/unet" "gvisor.dev/gvisor/runsc/boot" "gvisor.dev/gvisor/runsc/fsgofer" diff --git a/runsc/cmd/start.go b/runsc/cmd/start.go index de2115dff..5e9bc53ab 100644 --- a/runsc/cmd/start.go +++ b/runsc/cmd/start.go @@ -16,6 +16,7 @@ package cmd import ( "context" + "flag" "github.com/google/subcommands" "gvisor.dev/gvisor/runsc/boot" diff --git a/runsc/container/BUILD b/runsc/container/BUILD index 2bd12120d..6dea179e4 100644 --- a/runsc/container/BUILD +++ b/runsc/container/BUILD @@ -18,6 +18,7 @@ go_library( deps = [ "//pkg/log", "//pkg/sentry/control", + "//pkg/sync", "//runsc/boot", "//runsc/cgroup", "//runsc/sandbox", @@ -53,6 +54,7 @@ go_test( "//pkg/sentry/control", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", + "//pkg/sync", "//pkg/unet", "//pkg/urpc", "//runsc/boot", diff --git a/runsc/container/console_test.go b/runsc/container/console_test.go index 5ed131a7f..060b63bf3 100644 --- a/runsc/container/console_test.go +++ b/runsc/container/console_test.go @@ -20,7 +20,6 @@ import ( "io" "os" "path/filepath" - "sync" "syscall" "testing" "time" @@ -29,6 +28,7 @@ import ( "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/sentry/control" "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/unet" "gvisor.dev/gvisor/pkg/urpc" "gvisor.dev/gvisor/runsc/testutil" diff --git a/runsc/container/container_test.go b/runsc/container/container_test.go index c10f85992..b54d8f712 100644 --- a/runsc/container/container_test.go +++ b/runsc/container/container_test.go @@ -26,7 +26,6 @@ import ( "reflect" "strconv" "strings" - "sync" "syscall" "testing" "time" @@ -39,6 +38,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/control" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/runsc/boot" "gvisor.dev/gvisor/runsc/boot/platforms" "gvisor.dev/gvisor/runsc/specutils" diff --git a/runsc/container/multi_container_test.go b/runsc/container/multi_container_test.go index 4ad09ceab..2da93ec5b 100644 --- a/runsc/container/multi_container_test.go +++ b/runsc/container/multi_container_test.go @@ -22,7 +22,6 @@ import ( "path" "path/filepath" "strings" - "sync" "syscall" "testing" "time" @@ -30,6 +29,7 @@ import ( specs "github.com/opencontainers/runtime-spec/specs-go" "gvisor.dev/gvisor/pkg/sentry/control" "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/runsc/boot" "gvisor.dev/gvisor/runsc/specutils" "gvisor.dev/gvisor/runsc/testutil" diff --git a/runsc/container/state_file.go b/runsc/container/state_file.go index d95151ea5..17a251530 100644 --- a/runsc/container/state_file.go +++ b/runsc/container/state_file.go @@ -20,10 +20,10 @@ import ( "io/ioutil" "os" "path/filepath" - "sync" "github.com/gofrs/flock" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/sync" ) const stateFileExtension = ".state" diff --git a/runsc/fsgofer/BUILD b/runsc/fsgofer/BUILD index afcb41801..a9582d92b 100644 --- a/runsc/fsgofer/BUILD +++ b/runsc/fsgofer/BUILD @@ -19,6 +19,7 @@ go_library( "//pkg/fd", "//pkg/log", "//pkg/p9", + "//pkg/sync", "//pkg/syserr", "//runsc/specutils", "@org_golang_x_sys//unix:go_default_library", diff --git a/runsc/fsgofer/fsgofer.go b/runsc/fsgofer/fsgofer.go index b59e1a70e..93606d051 100644 --- a/runsc/fsgofer/fsgofer.go +++ b/runsc/fsgofer/fsgofer.go @@ -29,7 +29,6 @@ import ( "path/filepath" "runtime" "strconv" - "sync" "syscall" "golang.org/x/sys/unix" @@ -37,6 +36,7 @@ import ( "gvisor.dev/gvisor/pkg/fd" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/p9" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/runsc/specutils" ) diff --git a/runsc/sandbox/BUILD b/runsc/sandbox/BUILD index 8001949d5..ddbc37456 100644 --- a/runsc/sandbox/BUILD +++ b/runsc/sandbox/BUILD @@ -19,6 +19,7 @@ go_library( "//pkg/log", "//pkg/sentry/control", "//pkg/sentry/platform", + "//pkg/sync", "//pkg/tcpip/header", "//pkg/tcpip/stack", "//pkg/urpc", diff --git a/runsc/sandbox/sandbox.go b/runsc/sandbox/sandbox.go index ce1452b87..ec72bdbfd 100644 --- a/runsc/sandbox/sandbox.go +++ b/runsc/sandbox/sandbox.go @@ -22,7 +22,6 @@ import ( "os" "os/exec" "strconv" - "sync" "syscall" "time" @@ -34,6 +33,7 @@ import ( "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/control" "gvisor.dev/gvisor/pkg/sentry/platform" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/urpc" "gvisor.dev/gvisor/runsc/boot" "gvisor.dev/gvisor/runsc/boot/platforms" diff --git a/runsc/testutil/BUILD b/runsc/testutil/BUILD index c96ca2eb6..3c3027cb5 100644 --- a/runsc/testutil/BUILD +++ b/runsc/testutil/BUILD @@ -10,6 +10,7 @@ go_library( visibility = ["//:sandbox"], deps = [ "//pkg/log", + "//pkg/sync", "//runsc/boot", "//runsc/specutils", "@com_github_cenkalti_backoff//:go_default_library", diff --git a/runsc/testutil/testutil.go b/runsc/testutil/testutil.go index 9632776d2..fb22eae39 100644 --- a/runsc/testutil/testutil.go +++ b/runsc/testutil/testutil.go @@ -34,7 +34,6 @@ import ( "path/filepath" "strconv" "strings" - "sync" "sync/atomic" "syscall" "time" @@ -42,6 +41,7 @@ import ( "github.com/cenkalti/backoff" specs "github.com/opencontainers/runtime-spec/specs-go" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/runsc/boot" "gvisor.dev/gvisor/runsc/specutils" ) -- cgit v1.2.3 From d793677cd424fef10ac0b080871d181db0bcdec0 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Fri, 10 Jan 2020 18:07:15 -0800 Subject: I think INPUT works with protocol --- pkg/sentry/socket/netfilter/netfilter.go | 3 ++- pkg/tcpip/iptables/BUILD | 1 + pkg/tcpip/iptables/iptables.go | 4 +++- pkg/tcpip/iptables/types.go | 2 +- pkg/tcpip/network/ipv4/ipv4.go | 3 ++- pkg/tcpip/packet_buffer.go | 25 ++++++++++++++++++++++++- 6 files changed, 33 insertions(+), 5 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index f30461936..175466f19 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -25,6 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserr" + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/iptables" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -455,7 +456,7 @@ func filterFromIPTIP(iptip linux.IPTIP) (iptables.IPHeaderFilter, *syserr.Error) return iptables.IPHeaderFilter{}, syserr.ErrInvalidArgument } return iptables.IPHeaderFilter{ - Protocol: iptip.Protocol, + Protocol: tcpip.TransportProtocolNumber(iptip.Protocol), }, nil } diff --git a/pkg/tcpip/iptables/BUILD b/pkg/tcpip/iptables/BUILD index 2893c80cd..297eaccaf 100644 --- a/pkg/tcpip/iptables/BUILD +++ b/pkg/tcpip/iptables/BUILD @@ -14,5 +14,6 @@ go_library( deps = [ "//pkg/log", "//pkg/tcpip", + "//pkg/tcpip/header", ], ) diff --git a/pkg/tcpip/iptables/iptables.go b/pkg/tcpip/iptables/iptables.go index 4e1700fdb..3cff879a2 100644 --- a/pkg/tcpip/iptables/iptables.go +++ b/pkg/tcpip/iptables/iptables.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/header" ) const ( @@ -183,12 +184,13 @@ func (it *IPTables) checkTable(hook Hook, pkt tcpip.PacketBuffer, tablename stri panic(fmt.Sprintf("Traversed past the entire list of iptables rules in table %q.", tablename)) } +// Precondition: pk.NetworkHeader is set. func (it *IPTables) checkRule(hook Hook, pkt tcpip.PacketBuffer, table Table, ruleIdx int) Verdict { rule := table.Rules[ruleIdx] // First check whether the packet matches the IP header filter. // TODO(gvisor.dev/issue/170): Support other fields of the filter. - if rule.Filter.Protocol != pkt.Protocol { + if rule.Filter.Protocol != header.IPv4(pkt.NetworkHeader).TransportProtocol() { return Continue } diff --git a/pkg/tcpip/iptables/types.go b/pkg/tcpip/iptables/types.go index 4bedd9bc8..4f2a4d65e 100644 --- a/pkg/tcpip/iptables/types.go +++ b/pkg/tcpip/iptables/types.go @@ -173,7 +173,7 @@ type IPHeaderFilter struct { InputInterface string OutputInterfaceMask string InputInterfaceMask string - Protocol uint16 + Protocol tcpip.TransportProtocolNumber Flags uint8 InverseFlags uint8 } diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index f856081e6..5388d2549 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -353,7 +353,8 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt tcpip.PacketBuffer) { } pkt.NetworkHeader = headerView[:h.HeaderLength()] - // iptables filtering. + // iptables filtering. All packets that reach here are intended for + // this machine and will not be forwarded. ipt := e.stack.IPTables() if ok := ipt.Check(iptables.Input, pkt); !ok { // iptables is telling us to drop the packet. diff --git a/pkg/tcpip/packet_buffer.go b/pkg/tcpip/packet_buffer.go index ab24372e7..7a036b93c 100644 --- a/pkg/tcpip/packet_buffer.go +++ b/pkg/tcpip/packet_buffer.go @@ -13,7 +13,9 @@ package tcpip -import "gvisor.dev/gvisor/pkg/tcpip/buffer" +import ( + "gvisor.dev/gvisor/pkg/tcpip/buffer" +) // A PacketBuffer contains all the data of a network packet. // @@ -65,3 +67,24 @@ func (pk PacketBuffer) Clone() PacketBuffer { pk.Data = pk.Data.Clone(nil) return pk } + +//// TransportProtocol returns the transport protocol of pk. +//// +//// Precondition: pk.NetworkHeader is set. +//func (pk PacketBuffer) TransportProtocolIPv4() uint16 { +// if pk.NetworkHeader == nil { +// panic("This should only be called when pk.NetworkHeader is set.") +// } +// return header.IPv4(pk.NetworkHeader).TransportProtocol() +//} + +// func (pk Packet) findNetHeader() header.IPv4 { +// // Inbound: +// // Data holds everything, but may have had some headers shaved off. +// // Figure out whether it's set or still somewhere in data and return +// // appropriately. + +// // Outbound: +// // NetworkHeader will be set if we've added one. Otherwise there's no +// // header. +// } -- cgit v1.2.3 From 36641a21953b72d64d4378d4974ef467e901a5fe Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Mon, 13 Jan 2020 14:14:49 -0800 Subject: Only allow INPUT modifications. --- pkg/sentry/socket/netfilter/netfilter.go | 13 +++++++++++++ 1 file changed, 13 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 37f726295..507a77483 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -365,9 +365,22 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { } } + // TODO(gvisor.dev/issue/170): Support other chains. + // Since we only support modifying the INPUT chain right now, make sure + // all other chains point to ACCEPT rules. + for hook, ruleIdx := range table.BuiltinChains { + if hook != iptables.Input { + if _, ok := table.Rules[ruleIdx].Target.(iptables.UnconditionalAcceptTarget); !ok { + log.Warningf("Hook %d is unsupported.", hook) + return syserr.ErrInvalidArgument + } + } + } + // TODO(gvisor.dev/issue/170): Check the following conditions: // - There are no loops. // - There are no chains without an unconditional final rule. + // - There are no chains without an unconditional underflow rule. ipt := stack.IPTables() table.SetMetadata(metadata{ -- cgit v1.2.3 From debd213da61cf35d7c91346820e93fc87bfa5896 Mon Sep 17 00:00:00 2001 From: Tamir Duberstein Date: Mon, 13 Jan 2020 14:45:31 -0800 Subject: Allow dual stack sockets to operate on AF_INET Fixes #1490 Fixes #1495 PiperOrigin-RevId: 289523250 --- pkg/sentry/socket/netstack/netstack.go | 65 +++++++++--- pkg/sentry/socket/unix/unix.go | 5 +- pkg/sentry/strace/socket.go | 2 +- pkg/tcpip/stack/stack.go | 43 ++++++++ pkg/tcpip/transport/icmp/endpoint.go | 22 ++-- pkg/tcpip/transport/tcp/endpoint.go | 23 +--- pkg/tcpip/transport/udp/endpoint.go | 41 ++------ scripts/common.sh | 2 +- test/syscalls/linux/BUILD | 1 + test/syscalls/linux/socket_inet_loopback.cc | 156 ++++++++++++++++++++++++++++ 10 files changed, 278 insertions(+), 82 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 099319327..c020c11cb 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -324,22 +324,15 @@ func bytesToIPAddress(addr []byte) tcpip.Address { // converts it to the FullAddress format. It supports AF_UNIX, AF_INET, // AF_INET6, and AF_PACKET addresses. // -// strict indicates whether addresses with the AF_UNSPEC family are accepted of not. -// // AddressAndFamily returns an address and its family. -func AddressAndFamily(sfamily int, addr []byte, strict bool) (tcpip.FullAddress, uint16, *syserr.Error) { +func AddressAndFamily(addr []byte) (tcpip.FullAddress, uint16, *syserr.Error) { // Make sure we have at least 2 bytes for the address family. if len(addr) < 2 { return tcpip.FullAddress{}, 0, syserr.ErrInvalidArgument } - family := usermem.ByteOrder.Uint16(addr) - if family != uint16(sfamily) && (strict || family != linux.AF_UNSPEC) { - return tcpip.FullAddress{}, family, syserr.ErrAddressFamilyNotSupported - } - // Get the rest of the fields based on the address family. - switch family { + switch family := usermem.ByteOrder.Uint16(addr); family { case linux.AF_UNIX: path := addr[2:] if len(path) > linux.UnixPathMax { @@ -638,10 +631,40 @@ func (s *SocketOperations) Readiness(mask waiter.EventMask) waiter.EventMask { return r } +func (s *SocketOperations) checkFamily(family uint16, exact bool) *syserr.Error { + if family == uint16(s.family) { + return nil + } + if !exact && family == linux.AF_INET && s.family == linux.AF_INET6 { + v, err := s.Endpoint.GetSockOptBool(tcpip.V6OnlyOption) + if err != nil { + return syserr.TranslateNetstackError(err) + } + if !v { + return nil + } + } + return syserr.ErrInvalidArgument +} + +// mapFamily maps the AF_INET ANY address to the IPv4-mapped IPv6 ANY if the +// receiver's family is AF_INET6. +// +// This is a hack to work around the fact that both IPv4 and IPv6 ANY are +// represented by the empty string. +// +// TODO(gvisor.dev/issues/1556): remove this function. +func (s *SocketOperations) mapFamily(addr tcpip.FullAddress, family uint16) tcpip.FullAddress { + if len(addr.Addr) == 0 && s.family == linux.AF_INET6 && family == linux.AF_INET { + addr.Addr = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\xff\xff\x00\x00\x00\x00" + } + return addr +} + // Connect implements the linux syscall connect(2) for sockets backed by // tpcip.Endpoint. func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking bool) *syserr.Error { - addr, family, err := AddressAndFamily(s.family, sockaddr, false /* strict */) + addr, family, err := AddressAndFamily(sockaddr) if err != nil { return err } @@ -653,6 +676,12 @@ func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking boo } return syserr.TranslateNetstackError(err) } + + if err := s.checkFamily(family, false /* exact */); err != nil { + return err + } + addr = s.mapFamily(addr, family) + // Always return right away in the non-blocking case. if !blocking { return syserr.TranslateNetstackError(s.Endpoint.Connect(addr)) @@ -681,10 +710,14 @@ func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking boo // Bind implements the linux syscall bind(2) for sockets backed by // tcpip.Endpoint. func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { - addr, _, err := AddressAndFamily(s.family, sockaddr, true /* strict */) + addr, family, err := AddressAndFamily(sockaddr) if err != nil { return err } + if err := s.checkFamily(family, true /* exact */); err != nil { + return err + } + addr = s.mapFamily(addr, family) // Issue the bind request to the endpoint. return syserr.TranslateNetstackError(s.Endpoint.Bind(addr)) @@ -2080,8 +2113,8 @@ func ConvertAddress(family int, addr tcpip.FullAddress) (linux.SockAddr, uint32) case linux.AF_INET6: var out linux.SockAddrInet6 - if len(addr.Addr) == 4 { - // Copy address is v4-mapped format. + if len(addr.Addr) == header.IPv4AddressSize { + // Copy address in v4-mapped format. copy(out.Addr[12:], addr.Addr) out.Addr[10] = 0xff out.Addr[11] = 0xff @@ -2395,10 +2428,14 @@ func (s *SocketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to [] var addr *tcpip.FullAddress if len(to) > 0 { - addrBuf, _, err := AddressAndFamily(s.family, to, true /* strict */) + addrBuf, family, err := AddressAndFamily(to) if err != nil { return 0, err } + if err := s.checkFamily(family, false /* exact */); err != nil { + return 0, err + } + addrBuf = s.mapFamily(addrBuf, family) addr = &addrBuf } diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go index 91effe89a..7f49ba864 100644 --- a/pkg/sentry/socket/unix/unix.go +++ b/pkg/sentry/socket/unix/unix.go @@ -116,13 +116,16 @@ func (s *SocketOperations) Endpoint() transport.Endpoint { // extractPath extracts and validates the address. func extractPath(sockaddr []byte) (string, *syserr.Error) { - addr, _, err := netstack.AddressAndFamily(linux.AF_UNIX, sockaddr, true /* strict */) + addr, family, err := netstack.AddressAndFamily(sockaddr) if err != nil { if err == syserr.ErrAddressFamilyNotSupported { err = syserr.ErrInvalidArgument } return "", err } + if family != linux.AF_UNIX { + return "", syserr.ErrInvalidArgument + } // The address is trimmed by GetAddress. p := string(addr.Addr) diff --git a/pkg/sentry/strace/socket.go b/pkg/sentry/strace/socket.go index 51f2efb39..b6d7177f4 100644 --- a/pkg/sentry/strace/socket.go +++ b/pkg/sentry/strace/socket.go @@ -341,7 +341,7 @@ func sockAddr(t *kernel.Task, addr usermem.Addr, length uint32) string { switch family { case linux.AF_INET, linux.AF_INET6, linux.AF_UNIX: - fa, _, err := netstack.AddressAndFamily(int(family), b, true /* strict */) + fa, _, err := netstack.AddressAndFamily(b) if err != nil { return fmt.Sprintf("%#x {Family: %s, error extracting address: %v}", addr, familyStr, err) } diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index a47ceba54..113b457fb 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -547,6 +547,49 @@ type TransportEndpointInfo struct { RegisterNICID tcpip.NICID } +// AddrNetProto unwraps the specified address if it is a V4-mapped V6 address +// and returns the network protocol number to be used to communicate with the +// specified address. It returns an error if the passed address is incompatible +// with the receiver. +func (e *TransportEndpointInfo) AddrNetProto(addr tcpip.FullAddress, v6only bool) (tcpip.FullAddress, tcpip.NetworkProtocolNumber, *tcpip.Error) { + netProto := e.NetProto + switch len(addr.Addr) { + case header.IPv4AddressSize: + netProto = header.IPv4ProtocolNumber + case header.IPv6AddressSize: + if header.IsV4MappedAddress(addr.Addr) { + netProto = header.IPv4ProtocolNumber + addr.Addr = addr.Addr[header.IPv6AddressSize-header.IPv4AddressSize:] + if addr.Addr == header.IPv4Any { + addr.Addr = "" + } + } + } + + switch len(e.ID.LocalAddress) { + case header.IPv4AddressSize: + if len(addr.Addr) == header.IPv6AddressSize { + return tcpip.FullAddress{}, 0, tcpip.ErrInvalidEndpointState + } + case header.IPv6AddressSize: + if len(addr.Addr) == header.IPv4AddressSize { + return tcpip.FullAddress{}, 0, tcpip.ErrNetworkUnreachable + } + } + + switch { + case netProto == e.NetProto: + case netProto == header.IPv4ProtocolNumber && e.NetProto == header.IPv6ProtocolNumber: + if v6only { + return tcpip.FullAddress{}, 0, tcpip.ErrNoRoute + } + default: + return tcpip.FullAddress{}, 0, tcpip.ErrInvalidEndpointState + } + + return addr, netProto, nil +} + // IsEndpointInfo is an empty method to implement the tcpip.EndpointInfo // marker interface. func (*TransportEndpointInfo) IsEndpointInfo() {} diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index 330786f4c..42afb3f5b 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -288,7 +288,7 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c toCopy := *to to = &toCopy - netProto, err := e.checkV4Mapped(to, true) + netProto, err := e.checkV4Mapped(to) if err != nil { return 0, nil, err } @@ -475,18 +475,12 @@ func send6(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Err }) } -func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress, allowMismatch bool) (tcpip.NetworkProtocolNumber, *tcpip.Error) { - netProto := e.NetProto - if header.IsV4MappedAddress(addr.Addr) { - return 0, tcpip.ErrNoRoute - } - - // Fail if we're bound to an address length different from the one we're - // checking. - if l := len(e.ID.LocalAddress); !allowMismatch && l != 0 && l != len(addr.Addr) { - return 0, tcpip.ErrInvalidEndpointState +func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress) (tcpip.NetworkProtocolNumber, *tcpip.Error) { + unwrapped, netProto, err := e.TransportEndpointInfo.AddrNetProto(*addr, false /* v6only */) + if err != nil { + return 0, err } - + *addr = unwrapped return netProto, nil } @@ -518,7 +512,7 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { return tcpip.ErrInvalidEndpointState } - netProto, err := e.checkV4Mapped(&addr, false) + netProto, err := e.checkV4Mapped(&addr) if err != nil { return err } @@ -631,7 +625,7 @@ func (e *endpoint) bindLocked(addr tcpip.FullAddress) *tcpip.Error { return tcpip.ErrInvalidEndpointState } - netProto, err := e.checkV4Mapped(&addr, false) + netProto, err := e.checkV4Mapped(&addr) if err != nil { return err } diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index cca511fb9..cc8b533c8 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -1691,26 +1691,11 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { } func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress) (tcpip.NetworkProtocolNumber, *tcpip.Error) { - netProto := e.NetProto - if header.IsV4MappedAddress(addr.Addr) { - // Fail if using a v4 mapped address on a v6only endpoint. - if e.v6only { - return 0, tcpip.ErrNoRoute - } - - netProto = header.IPv4ProtocolNumber - addr.Addr = addr.Addr[header.IPv6AddressSize-header.IPv4AddressSize:] - if addr.Addr == header.IPv4Any { - addr.Addr = "" - } - } - - // Fail if we're bound to an address length different from the one we're - // checking. - if l := len(e.ID.LocalAddress); l != 0 && len(addr.Addr) != 0 && l != len(addr.Addr) { - return 0, tcpip.ErrInvalidEndpointState + unwrapped, netProto, err := e.TransportEndpointInfo.AddrNetProto(*addr, e.v6only) + if err != nil { + return 0, err } - + *addr = unwrapped return netProto, nil } diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index a4ff29a7d..13446f5d9 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -402,7 +402,7 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c return 0, nil, tcpip.ErrBroadcastDisabled } - netProto, err := e.checkV4Mapped(to, false) + netProto, err := e.checkV4Mapped(to) if err != nil { return 0, nil, err } @@ -501,7 +501,7 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { defer e.mu.Unlock() fa := tcpip.FullAddress{Addr: v.InterfaceAddr} - netProto, err := e.checkV4Mapped(&fa, false) + netProto, err := e.checkV4Mapped(&fa) if err != nil { return err } @@ -839,35 +839,12 @@ func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort u return nil } -func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress, allowMismatch bool) (tcpip.NetworkProtocolNumber, *tcpip.Error) { - netProto := e.NetProto - if len(addr.Addr) == 0 { - return netProto, nil - } - if header.IsV4MappedAddress(addr.Addr) { - // Fail if using a v4 mapped address on a v6only endpoint. - if e.v6only { - return 0, tcpip.ErrNoRoute - } - - netProto = header.IPv4ProtocolNumber - addr.Addr = addr.Addr[header.IPv6AddressSize-header.IPv4AddressSize:] - if addr.Addr == header.IPv4Any { - addr.Addr = "" - } - - // Fail if we are bound to an IPv6 address. - if !allowMismatch && len(e.ID.LocalAddress) == 16 { - return 0, tcpip.ErrNetworkUnreachable - } - } - - // Fail if we're bound to an address length different from the one we're - // checking. - if l := len(e.ID.LocalAddress); l != 0 && l != len(addr.Addr) { - return 0, tcpip.ErrInvalidEndpointState +func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress) (tcpip.NetworkProtocolNumber, *tcpip.Error) { + unwrapped, netProto, err := e.TransportEndpointInfo.AddrNetProto(*addr, e.v6only) + if err != nil { + return 0, err } - + *addr = unwrapped return netProto, nil } @@ -916,7 +893,7 @@ func (e *endpoint) Disconnect() *tcpip.Error { // Connect connects the endpoint to its peer. Specifying a NIC is optional. func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { - netProto, err := e.checkV4Mapped(&addr, false) + netProto, err := e.checkV4Mapped(&addr) if err != nil { return err } @@ -1074,7 +1051,7 @@ func (e *endpoint) bindLocked(addr tcpip.FullAddress) *tcpip.Error { return tcpip.ErrInvalidEndpointState } - netProto, err := e.checkV4Mapped(&addr, true) + netProto, err := e.checkV4Mapped(&addr) if err != nil { return err } diff --git a/scripts/common.sh b/scripts/common.sh index 6dabad141..fdb1aa142 100755 --- a/scripts/common.sh +++ b/scripts/common.sh @@ -73,7 +73,7 @@ function install_runsc() { sudo "${RUNSC_BIN}" install --experimental=true --runtime="${runtime}" -- --debug-log "${RUNSC_LOGS}" "$@" # Clear old logs files that may exist. - sudo rm -f "${RUNSC_LOGS_DIR}"/* + sudo rm -f "${RUNSC_LOGS_DIR}"/'*' # Restart docker to pick up the new runtime configuration. sudo systemctl restart docker diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index ce8abe217..4c7ec3f06 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -2693,6 +2693,7 @@ cc_binary( srcs = ["socket_inet_loopback.cc"], linkstatic = 1, deps = [ + ":ip_socket_test_util", ":socket_test_util", "//test/util:file_descriptor", "//test/util:posix_error", diff --git a/test/syscalls/linux/socket_inet_loopback.cc b/test/syscalls/linux/socket_inet_loopback.cc index 138024d9e..5d114d460 100644 --- a/test/syscalls/linux/socket_inet_loopback.cc +++ b/test/syscalls/linux/socket_inet_loopback.cc @@ -32,6 +32,7 @@ #include "absl/strings/str_cat.h" #include "absl/time/clock.h" #include "absl/time/time.h" +#include "test/syscalls/linux/ip_socket_test_util.h" #include "test/syscalls/linux/socket_test_util.h" #include "test/util/file_descriptor.h" #include "test/util/posix_error.h" @@ -102,6 +103,161 @@ TEST(BadSocketPairArgs, ValidateErrForBadCallsToSocketPair) { SyscallFailsWithErrno(EAFNOSUPPORT)); } +enum class Operation { + Bind, + Connect, + SendTo, +}; + +std::string OperationToString(Operation operation) { + switch (operation) { + case Operation::Bind: + return "Bind"; + case Operation::Connect: + return "Connect"; + case Operation::SendTo: + return "SendTo"; + } +} + +using OperationSequence = std::vector; + +using DualStackSocketTest = + ::testing::TestWithParam>; + +TEST_P(DualStackSocketTest, AddressOperations) { + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET6, SOCK_DGRAM, 0)); + + const TestAddress& addr = std::get<0>(GetParam()); + const OperationSequence& operations = std::get<1>(GetParam()); + + auto addr_in = reinterpret_cast(&addr.addr); + + // sockets may only be bound once. Both `connect` and `sendto` cause a socket + // to be bound. + bool bound = false; + for (const Operation& operation : operations) { + bool sockname = false; + bool peername = false; + switch (operation) { + case Operation::Bind: { + ASSERT_NO_ERRNO(SetAddrPort( + addr.family(), const_cast(&addr.addr), 0)); + + int bind_ret = bind(fd.get(), addr_in, addr.addr_len); + + // Dual stack sockets may only be bound to AF_INET6. + if (!bound && addr.family() == AF_INET6) { + EXPECT_THAT(bind_ret, SyscallSucceeds()); + bound = true; + + sockname = true; + } else { + EXPECT_THAT(bind_ret, SyscallFailsWithErrno(EINVAL)); + } + break; + } + case Operation::Connect: { + ASSERT_NO_ERRNO(SetAddrPort( + addr.family(), const_cast(&addr.addr), 1337)); + + EXPECT_THAT(connect(fd.get(), addr_in, addr.addr_len), + SyscallSucceeds()) + << GetAddrStr(addr_in); + bound = true; + + sockname = true; + peername = true; + + break; + } + case Operation::SendTo: { + const char payload[] = "hello"; + ASSERT_NO_ERRNO(SetAddrPort( + addr.family(), const_cast(&addr.addr), 1337)); + + ssize_t sendto_ret = sendto(fd.get(), &payload, sizeof(payload), 0, + addr_in, addr.addr_len); + + EXPECT_THAT(sendto_ret, SyscallSucceedsWithValue(sizeof(payload))); + sockname = !bound; + bound = true; + break; + } + } + + if (sockname) { + sockaddr_storage sock_addr; + socklen_t addrlen = sizeof(sock_addr); + ASSERT_THAT(getsockname(fd.get(), reinterpret_cast(&sock_addr), + &addrlen), + SyscallSucceeds()); + ASSERT_EQ(addrlen, sizeof(struct sockaddr_in6)); + + auto sock_addr_in6 = reinterpret_cast(&sock_addr); + + if (operation == Operation::SendTo) { + EXPECT_EQ(sock_addr_in6->sin6_family, AF_INET6); + EXPECT_TRUE(IN6_IS_ADDR_UNSPECIFIED(sock_addr_in6->sin6_addr.s6_addr32)) + << OperationToString(operation) << " getsocknam=" + << GetAddrStr(reinterpret_cast(&sock_addr)); + + EXPECT_NE(sock_addr_in6->sin6_port, 0); + } else if (IN6_IS_ADDR_V4MAPPED( + reinterpret_cast(addr_in) + ->sin6_addr.s6_addr32)) { + EXPECT_TRUE(IN6_IS_ADDR_V4MAPPED(sock_addr_in6->sin6_addr.s6_addr32)) + << OperationToString(operation) << " getsocknam=" + << GetAddrStr(reinterpret_cast(&sock_addr)); + } + } + + if (peername) { + sockaddr_storage peer_addr; + socklen_t addrlen = sizeof(peer_addr); + ASSERT_THAT(getpeername(fd.get(), reinterpret_cast(&peer_addr), + &addrlen), + SyscallSucceeds()); + ASSERT_EQ(addrlen, sizeof(struct sockaddr_in6)); + + if (addr.family() == AF_INET || + IN6_IS_ADDR_V4MAPPED(reinterpret_cast(addr_in) + ->sin6_addr.s6_addr32)) { + EXPECT_TRUE(IN6_IS_ADDR_V4MAPPED( + reinterpret_cast(&peer_addr) + ->sin6_addr.s6_addr32)) + << OperationToString(operation) << " getpeername=" + << GetAddrStr(reinterpret_cast(&peer_addr)); + } + } + } +} + +// TODO(gvisor.dev/issues/1556): uncomment V4MappedAny. +INSTANTIATE_TEST_SUITE_P( + All, DualStackSocketTest, + ::testing::Combine( + ::testing::Values(V4Any(), V4Loopback(), /*V4MappedAny(),*/ + V4MappedLoopback(), V6Any(), V6Loopback()), + ::testing::ValuesIn( + {{Operation::Bind, Operation::Connect, Operation::SendTo}, + {Operation::Bind, Operation::SendTo, Operation::Connect}, + {Operation::Connect, Operation::Bind, Operation::SendTo}, + {Operation::Connect, Operation::SendTo, Operation::Bind}, + {Operation::SendTo, Operation::Bind, Operation::Connect}, + {Operation::SendTo, Operation::Connect, Operation::Bind}})), + [](::testing::TestParamInfo< + std::tuple> const& info) { + const TestAddress& addr = std::get<0>(info.param); + const OperationSequence& operations = std::get<1>(info.param); + std::string s = addr.description; + for (const Operation& operation : operations) { + absl::StrAppend(&s, OperationToString(operation)); + } + return s; + }); + void tcpSimpleConnectTest(TestAddress const& listener, TestAddress const& connector, bool unbound) { // Create the listening socket. -- cgit v1.2.3 From 50625cee59aaff834c7968771ab385ad0e7b0e1f Mon Sep 17 00:00:00 2001 From: Tamir Duberstein Date: Tue, 14 Jan 2020 13:31:52 -0800 Subject: Implement {g,s}etsockopt(IP_RECVTOS) for UDP sockets PiperOrigin-RevId: 289718534 --- pkg/sentry/socket/control/control.go | 2 +- pkg/sentry/socket/netstack/netstack.go | 36 +++++++++++++-- pkg/tcpip/checker/checker.go | 16 +++++++ pkg/tcpip/stack/nic.go | 2 +- pkg/tcpip/stack/stack.go | 2 +- pkg/tcpip/tcpip.go | 8 +++- pkg/tcpip/transport/udp/endpoint.go | 40 +++++++++++++++-- pkg/tcpip/transport/udp/udp_test.go | 67 ++++++++++++++++++++++++---- test/syscalls/linux/socket_ip_udp_generic.cc | 40 +++++++++++++++++ test/syscalls/linux/udp_socket_test_cases.cc | 8 ++-- 10 files changed, 197 insertions(+), 24 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/control/control.go b/pkg/sentry/socket/control/control.go index 4301b697c..1684dfc24 100644 --- a/pkg/sentry/socket/control/control.go +++ b/pkg/sentry/socket/control/control.go @@ -327,7 +327,7 @@ func PackInq(t *kernel.Task, inq int32, buf []byte) []byte { } // PackTOS packs an IP_TOS socket control message. -func PackTOS(t *kernel.Task, tos int8, buf []byte) []byte { +func PackTOS(t *kernel.Task, tos uint8, buf []byte) []byte { return putCmsgStruct( buf, linux.SOL_IP, diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index c020c11cb..d2f7e987d 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -1268,11 +1268,11 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf if err != nil { return nil, syserr.TranslateNetstackError(err) } - var o uint32 + var o int32 if v { o = 1 } - return int32(o), nil + return o, nil case linux.IPV6_PATHMTU: t.Kernel().EmitUnimplementedEvent(t) @@ -1377,6 +1377,21 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in } return int32(v), nil + case linux.IP_RECVTOS: + if outLen < sizeOfInt32 { + return nil, syserr.ErrInvalidArgument + } + + v, err := ep.GetSockOptBool(tcpip.ReceiveTOSOption) + if err != nil { + return nil, syserr.TranslateNetstackError(err) + } + var o int32 + if v { + o = 1 + } + return o, nil + default: emitUnimplementedEventIP(t, name) } @@ -1895,6 +1910,13 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s } return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.IPv4TOSOption(v))) + case linux.IP_RECVTOS: + v, err := parseIntOrChar(optVal) + if err != nil { + return err + } + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.ReceiveTOSOption, v != 0)) + case linux.IP_ADD_SOURCE_MEMBERSHIP, linux.IP_BIND_ADDRESS_NO_PORT, linux.IP_BLOCK_SOURCE, @@ -1915,7 +1937,6 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s linux.IP_RECVFRAGSIZE, linux.IP_RECVOPTS, linux.IP_RECVORIGDSTADDR, - linux.IP_RECVTOS, linux.IP_RECVTTL, linux.IP_RETOPTS, linux.IP_TRANSPARENT, @@ -2335,7 +2356,14 @@ func (s *SocketOperations) nonBlockingRead(ctx context.Context, dst usermem.IOSe } func (s *SocketOperations) controlMessages() socket.ControlMessages { - return socket.ControlMessages{IP: tcpip.ControlMessages{HasTimestamp: s.readCM.HasTimestamp && s.sockOptTimestamp, Timestamp: s.readCM.Timestamp}} + return socket.ControlMessages{ + IP: tcpip.ControlMessages{ + HasTimestamp: s.readCM.HasTimestamp && s.sockOptTimestamp, + Timestamp: s.readCM.Timestamp, + HasTOS: s.readCM.HasTOS, + TOS: s.readCM.TOS, + }, + } } // updateTimestamp sets the timestamp for SIOCGSTAMP. It should be called after diff --git a/pkg/tcpip/checker/checker.go b/pkg/tcpip/checker/checker.go index 2f15bf1f1..542abc99d 100644 --- a/pkg/tcpip/checker/checker.go +++ b/pkg/tcpip/checker/checker.go @@ -33,6 +33,9 @@ type NetworkChecker func(*testing.T, []header.Network) // TransportChecker is a function to check a property of a transport packet. type TransportChecker func(*testing.T, header.Transport) +// ControlMessagesChecker is a function to check a property of ancillary data. +type ControlMessagesChecker func(*testing.T, tcpip.ControlMessages) + // IPv4 checks the validity and properties of the given IPv4 packet. It is // expected to be used in conjunction with other network checkers for specific // properties. For example, to check the source and destination address, one @@ -158,6 +161,19 @@ func FragmentFlags(flags uint8) NetworkChecker { } } +// ReceiveTOS creates a checker that checks the TOS field in ControlMessages. +func ReceiveTOS(want uint8) ControlMessagesChecker { + return func(t *testing.T, cm tcpip.ControlMessages) { + t.Helper() + if !cm.HasTOS { + t.Fatalf("got cm.HasTOS = %t, want cm.TOS = %d", cm.HasTOS, want) + } + if got := cm.TOS; got != want { + t.Fatalf("got cm.TOS = %d, want %d", got, want) + } + } +} + // TOS creates a checker that checks the TOS field. func TOS(tos uint8, label uint32) NetworkChecker { return func(t *testing.T, h []header.Network) { diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index abf73fe33..071221d5a 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -763,7 +763,7 @@ func (n *NIC) RemoveAddressRange(subnet tcpip.Subnet) { n.mu.Unlock() } -// Subnets returns the Subnets associated with this NIC. +// AddressRanges returns the Subnets associated with this NIC. func (n *NIC) AddressRanges() []tcpip.Subnet { n.mu.RLock() defer n.mu.RUnlock() diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index f8d89248e..386eb6eec 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -912,7 +912,7 @@ func (s *Stack) CheckNIC(id tcpip.NICID) bool { return false } -// NICSubnets returns a map of NICIDs to their associated subnets. +// NICAddressRanges returns a map of NICIDs to their associated subnets. func (s *Stack) NICAddressRanges() map[tcpip.NICID][]tcpip.Subnet { s.mu.RLock() defer s.mu.RUnlock() diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 4a090ac86..b7813cbc0 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -322,7 +322,7 @@ type ControlMessages struct { HasTOS bool // TOS is the IPv4 type of service of the associated packet. - TOS int8 + TOS uint8 // HasTClass indicates whether Tclass is valid/set. HasTClass bool @@ -500,9 +500,13 @@ type WriteOptions struct { type SockOptBool int const ( + // ReceiveTOSOption is used by SetSockOpt/GetSockOpt to specify if the TOS + // ancillary message is passed with incoming packets. + ReceiveTOSOption SockOptBool = iota + // V6OnlyOption is used by {G,S}etSockOptBool to specify whether an IPv6 // socket is to be restricted to sending and receiving IPv6 packets only. - V6OnlyOption SockOptBool = iota + V6OnlyOption ) // SockOptInt represents socket options which values have the int type. diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 13446f5d9..c9cbed8f4 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -31,6 +31,7 @@ type udpPacket struct { senderAddress tcpip.FullAddress data buffer.VectorisedView `state:".(buffer.VectorisedView)"` timestamp int64 + tos uint8 } // EndpointState represents the state of a UDP endpoint. @@ -113,6 +114,10 @@ type endpoint struct { // applied while sending packets. Defaults to 0 as on Linux. sendTOS uint8 + // receiveTOS determines if the incoming IPv4 TOS header field is passed + // as ancillary data to ControlMessages on Read. + receiveTOS bool + // shutdownFlags represent the current shutdown state of the endpoint. shutdownFlags tcpip.ShutdownFlags @@ -243,7 +248,18 @@ func (e *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMess *addr = p.senderAddress } - return p.data.ToView(), tcpip.ControlMessages{HasTimestamp: true, Timestamp: p.timestamp}, nil + cm := tcpip.ControlMessages{ + HasTimestamp: true, + Timestamp: p.timestamp, + } + e.mu.RLock() + receiveTOS := e.receiveTOS + e.mu.RUnlock() + if receiveTOS { + cm.HasTOS = true + cm.TOS = p.tos + } + return p.data.ToView(), cm, nil } // prepareForWrite prepares the endpoint for sending data. In particular, it @@ -458,6 +474,12 @@ func (e *endpoint) Peek([][]byte) (int64, tcpip.ControlMessages, *tcpip.Error) { // SetSockOptBool implements tcpip.Endpoint.SetSockOptBool. func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { switch opt { + case tcpip.ReceiveTOSOption: + e.mu.Lock() + e.receiveTOS = v + e.mu.Unlock() + return nil + case tcpip.V6OnlyOption: // We only recognize this option on v6 endpoints. if e.NetProto != header.IPv6ProtocolNumber { @@ -664,15 +686,21 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { // GetSockOptBool implements tcpip.Endpoint.GetSockOptBool. func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { switch opt { + case tcpip.ReceiveTOSOption: + e.mu.RLock() + v := e.receiveTOS + e.mu.RUnlock() + return v, nil + case tcpip.V6OnlyOption: // We only recognize this option on v6 endpoints. if e.NetProto != header.IPv6ProtocolNumber { return false, tcpip.ErrUnknownProtocolOption } - e.mu.Lock() + e.mu.RLock() v := e.v6only - e.mu.Unlock() + e.mu.RUnlock() return v, nil } @@ -1215,6 +1243,12 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk e.rcvList.PushBack(packet) e.rcvBufSize += pkt.Data.Size() + // Save any useful information from the network header to the packet. + switch r.NetProto { + case header.IPv4ProtocolNumber: + packet.tos, _ = header.IPv4(pkt.NetworkHeader).TOS() + } + packet.timestamp = e.stack.NowNanoseconds() e.rcvMu.Unlock() diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index 0a82bc4fa..ee9d10555 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -56,6 +56,7 @@ const ( multicastAddr = "\xe8\x2b\xd3\xea" multicastV6Addr = "\xff\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00" broadcastAddr = header.IPv4Broadcast + testTOS = 0x80 // defaultMTU is the MTU, in bytes, used throughout the tests, except // where another value is explicitly used. It is chosen to match the MTU @@ -453,6 +454,7 @@ func (c *testContext) injectV4Packet(payload []byte, h *header4Tuple, valid bool ip := header.IPv4(buf) ip.Encode(&header.IPv4Fields{ IHL: header.IPv4MinimumSize, + TOS: testTOS, TotalLength: uint16(len(buf)), TTL: 65, Protocol: uint8(udp.ProtocolNumber), @@ -552,8 +554,8 @@ func TestBindToDeviceOption(t *testing.T) { // testReadInternal sends a packet of the given test flow into the stack by // injecting it into the link endpoint. It then attempts to read it from the // UDP endpoint and depending on if this was expected to succeed verifies its -// correctness. -func testReadInternal(c *testContext, flow testFlow, packetShouldBeDropped, expectReadError bool) { +// correctness including any additional checker functions provided. +func testReadInternal(c *testContext, flow testFlow, packetShouldBeDropped, expectReadError bool, checkers ...checker.ControlMessagesChecker) { c.t.Helper() payload := newPayload() @@ -568,12 +570,12 @@ func testReadInternal(c *testContext, flow testFlow, packetShouldBeDropped, expe epstats := c.ep.Stats().(*tcpip.TransportEndpointStats).Clone() var addr tcpip.FullAddress - v, _, err := c.ep.Read(&addr) + v, cm, err := c.ep.Read(&addr) if err == tcpip.ErrWouldBlock { // Wait for data to become available. select { case <-ch: - v, _, err = c.ep.Read(&addr) + v, cm, err = c.ep.Read(&addr) case <-time.After(300 * time.Millisecond): if packetShouldBeDropped { @@ -606,15 +608,21 @@ func testReadInternal(c *testContext, flow testFlow, packetShouldBeDropped, expe if !bytes.Equal(payload, v) { c.t.Fatalf("bad payload: got %x, want %x", v, payload) } + + // Run any checkers against the ControlMessages. + for _, f := range checkers { + f(c.t, cm) + } + c.checkEndpointReadStats(1, epstats, err) } // testRead sends a packet of the given test flow into the stack by injecting it // into the link endpoint. It then reads it from the UDP endpoint and verifies -// its correctness. -func testRead(c *testContext, flow testFlow) { +// its correctness including any additional checker functions provided. +func testRead(c *testContext, flow testFlow, checkers ...checker.ControlMessagesChecker) { c.t.Helper() - testReadInternal(c, flow, false /* packetShouldBeDropped */, false /* expectReadError */) + testReadInternal(c, flow, false /* packetShouldBeDropped */, false /* expectReadError */, checkers...) } // testFailingRead sends a packet of the given test flow into the stack by @@ -1282,7 +1290,7 @@ func TestTOSV4(t *testing.T) { c.createEndpointForFlow(flow) - const tos = 0xC0 + const tos = testTOS var v tcpip.IPv4TOSOption if err := c.ep.GetSockOpt(&v); err != nil { c.t.Errorf("GetSockopt failed: %s", err) @@ -1317,7 +1325,7 @@ func TestTOSV6(t *testing.T) { c.createEndpointForFlow(flow) - const tos = 0xC0 + const tos = testTOS var v tcpip.IPv6TrafficClassOption if err := c.ep.GetSockOpt(&v); err != nil { c.t.Errorf("GetSockopt failed: %s", err) @@ -1344,6 +1352,47 @@ func TestTOSV6(t *testing.T) { } } +func TestReceiveTOSV4(t *testing.T) { + for _, flow := range []testFlow{unicastV4, broadcast} { + t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpointForFlow(flow) + + // Verify that setting and reading the option works. + v, err := c.ep.GetSockOptBool(tcpip.ReceiveTOSOption) + if err != nil { + c.t.Fatal("GetSockOptBool(tcpip.ReceiveTOSOption) failed:", err) + } + // Test for expected default value. + if v != false { + c.t.Errorf("got GetSockOptBool(tcpip.ReceiveTOSOption) = %t, want = %t", v, false) + } + + want := true + if err := c.ep.SetSockOptBool(tcpip.ReceiveTOSOption, want); err != nil { + c.t.Fatalf("SetSockOptBool(tcpip.ReceiveTOSOption, %t) failed: %s", want, err) + } + + got, err := c.ep.GetSockOptBool(tcpip.ReceiveTOSOption) + if err != nil { + c.t.Fatal("GetSockOptBool(tcpip.ReceiveTOSOption) failed:", err) + } + if got != want { + c.t.Fatalf("got GetSockOptBool(tcpip.ReceiveTOSOption) = %t, want = %t", got, want) + } + + // Verify that the correct received TOS is handed through as + // ancillary data to the ControlMessages struct. + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + c.t.Fatal("Bind failed:", err) + } + testRead(c, flow, checker.ReceiveTOS(testTOS)) + }) + } +} + func TestMulticastInterfaceOption(t *testing.T) { for _, flow := range []testFlow{multicastV4, multicastV4in6, multicastV6, multicastV6Only} { t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { diff --git a/test/syscalls/linux/socket_ip_udp_generic.cc b/test/syscalls/linux/socket_ip_udp_generic.cc index 66eb68857..53290bed7 100644 --- a/test/syscalls/linux/socket_ip_udp_generic.cc +++ b/test/syscalls/linux/socket_ip_udp_generic.cc @@ -209,6 +209,46 @@ TEST_P(UDPSocketPairTest, SetMulticastLoopChar) { EXPECT_EQ(get, kSockOptOn); } +// Ensure that Receiving TOS is off by default. +TEST_P(UDPSocketPairTest, RecvTosDefault) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT( + getsockopt(sockets->first_fd(), IPPROTO_IP, IP_RECVTOS, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kSockOptOff); +} + +// Test that setting and getting IP_RECVTOS works as expected. +TEST_P(UDPSocketPairTest, SetRecvTos) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + ASSERT_THAT(setsockopt(sockets->first_fd(), IPPROTO_IP, IP_RECVTOS, + &kSockOptOff, sizeof(kSockOptOff)), + SyscallSucceeds()); + + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT( + getsockopt(sockets->first_fd(), IPPROTO_IP, IP_RECVTOS, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kSockOptOff); + + ASSERT_THAT(setsockopt(sockets->first_fd(), IPPROTO_IP, IP_RECVTOS, + &kSockOptOn, sizeof(kSockOptOn)), + SyscallSucceeds()); + + ASSERT_THAT( + getsockopt(sockets->first_fd(), IPPROTO_IP, IP_RECVTOS, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kSockOptOn); +} + TEST_P(UDPSocketPairTest, ReuseAddrDefault) { auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); diff --git a/test/syscalls/linux/udp_socket_test_cases.cc b/test/syscalls/linux/udp_socket_test_cases.cc index dc35c2f50..68e0a8109 100644 --- a/test/syscalls/linux/udp_socket_test_cases.cc +++ b/test/syscalls/linux/udp_socket_test_cases.cc @@ -1349,8 +1349,9 @@ TEST_P(UdpSocketTest, TimestampIoctlPersistence) { // outgoing packets, and that a receiving socket with IP_RECVTOS or // IPV6_RECVTCLASS will create the corresponding control message. TEST_P(UdpSocketTest, SetAndReceiveTOS) { - // TODO(b/68320120): IP_RECVTOS/IPV6_RECVTCLASS not supported for netstack. - SKIP_IF(IsRunningOnGvisor() && !IsRunningWithHostinet()); + // TODO(b/68320120): IPV6_RECVTCLASS not supported for netstack. + SKIP_IF((GetParam() != AddressFamily::kIpv4) && IsRunningOnGvisor() && + !IsRunningWithHostinet()); ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); ASSERT_THAT(connect(t_, addr_[0], addrlen_), SyscallSucceeds()); @@ -1421,7 +1422,8 @@ TEST_P(UdpSocketTest, SetAndReceiveTOS) { // TOS byte on outgoing packets, and that a receiving socket with IP_RECVTOS or // IPV6_RECVTCLASS will create the corresponding control message. TEST_P(UdpSocketTest, SendAndReceiveTOS) { - // TODO(b/68320120): IP_RECVTOS/IPV6_RECVTCLASS not supported for netstack. + // TODO(b/68320120): IPV6_RECVTCLASS not supported for netstack. + // TODO(b/146661005): Setting TOS via cmsg not supported for netstack. SKIP_IF(IsRunningOnGvisor() && !IsRunningWithHostinet()); ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); ASSERT_THAT(connect(t_, addr_[0], addrlen_), SyscallSucceeds()); -- cgit v1.2.3 From 376a777c55680f8139313d87bd460248fd251941 Mon Sep 17 00:00:00 2001 From: "chris.zn" Date: Tue, 14 Jan 2020 18:59:33 +0800 Subject: Fix "unlock of unlocked mutex" crash when getting tty This patch holds taskset.mu when getting tty. If we don't do this, it may cause a "unlock of unlocked mutex" problem, since signalHandlers may be replaced by CopyForExec() in runSyscallAfterExecStop after the signalHandlers.mu has been holded in TTY(). The problem is easy to reproduce with keeping to do "runsc ps". The crash log is : fatal error: sync: unlock of unlocked mutex goroutine 5801304 [running]: runtime.throw(0xfd019c, 0x1e) GOROOT/src/runtime/panic.go:774 +0x72 fp=0xc001ba47b0 sp=0xc001ba4780 pc=0x431702 sync.throw(0xfd019c, 0x1e) GOROOT/src/runtime/panic.go:760 +0x35 fp=0xc001ba47d0 sp=0xc001ba47b0 pc=0x431685 sync.(*Mutex).unlockSlow(0xc00cf94a30, 0xc0ffffffff) GOROOT/src/sync/mutex.go:196 +0xd6 fp=0xc001ba47f8 sp=0xc001ba47d0 pc=0x4707d6 sync.(*Mutex).Unlock(0xc00cf94a30) GOROOT/src/sync/mutex.go:190 +0x48 fp=0xc001ba4818 sp=0xc001ba47f8 pc=0x4706e8 gvisor.dev/gvisor/pkg/sentry/kernel.(*ThreadGroup).TTY(0xc011a9e800, 0x0) pkg/sentry/kernel/tty.go:38 +0x88 fp=0xc001ba4868 sp=0xc001ba4818 pc=0x835fa8 gvisor.dev/gvisor/pkg/sentry/control.Processes(0xc00025ae00, 0xc013e397c0, 0x40, 0xc0137b9800, 0x1, 0x7f292e9a4cc0) pkg/sentry/control/proc.go:366 +0x355 fp=0xc001ba49a0 sp=0xc001ba4868 pc=0x9ac4a5 gvisor.dev/gvisor/runsc/boot.(*containerManager).Processes(0xc0003b62c0, 0xc0051423d0, 0xc0137b9800, 0x0, 0x0) runsc/boot/controller.go:228 +0xdf fp=0xc001ba49e8 sp=0xc001ba49a0 pc=0xaf06cf Signed-off-by: chris.zn --- pkg/sentry/kernel/tty.go | 2 ++ 1 file changed, 2 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/tty.go b/pkg/sentry/kernel/tty.go index 464d2306a..d0e0810e8 100644 --- a/pkg/sentry/kernel/tty.go +++ b/pkg/sentry/kernel/tty.go @@ -33,6 +33,8 @@ type TTY struct { // TTY returns the thread group's controlling terminal. If nil, there is no // controlling terminal. func (tg *ThreadGroup) TTY() *TTY { + tg.pidns.owner.mu.RLock() + defer tg.pidns.owner.mu.RUnlock() tg.signalHandlers.mu.Lock() defer tg.signalHandlers.mu.Unlock() return tg.tty -- cgit v1.2.3 From ff78a721700f8b7d3c8dae14fc14c04f3a82b970 Mon Sep 17 00:00:00 2001 From: lubinszARM <34124929+lubinszARM@users.noreply.github.com> Date: Tue, 14 Jan 2020 22:22:45 -0800 Subject: enable pkg/sentry/arch to support arm64 basically Signed-off-by: Bin Lu Change-Id: I9cce23db4e5caec82ce42b4970fdb7f7e8c08f1d COPYBARA_INTEGRATE_REVIEW=https://github.com/google/gvisor/pull/773 from lubinszARM:pr_arch_basic 3fe2fd8e6286766bbe489ef971dce204f924feba PiperOrigin-RevId: 289795569 --- pkg/sentry/arch/BUILD | 6 + pkg/sentry/arch/arch_aarch64.go | 293 ++++++++++++++++++++++++++++++++++ pkg/sentry/arch/arch_arm64.go | 266 ++++++++++++++++++++++++++++++ pkg/sentry/arch/arch_state_aarch64.go | 38 +++++ pkg/sentry/arch/arch_state_x86.go | 2 + pkg/sentry/arch/registers.proto | 37 +++++ pkg/sentry/arch/signal.go | 250 +++++++++++++++++++++++++++++ pkg/sentry/arch/signal_amd64.go | 230 -------------------------- pkg/sentry/arch/signal_arm64.go | 126 +++++++++++++++ pkg/sentry/arch/signal_stack.go | 2 +- pkg/sentry/arch/syscalls_arm64.go | 62 +++++++ 11 files changed, 1081 insertions(+), 231 deletions(-) create mode 100644 pkg/sentry/arch/arch_aarch64.go create mode 100644 pkg/sentry/arch/arch_arm64.go create mode 100644 pkg/sentry/arch/arch_state_aarch64.go create mode 100644 pkg/sentry/arch/signal.go create mode 100644 pkg/sentry/arch/signal_arm64.go create mode 100644 pkg/sentry/arch/syscalls_arm64.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/arch/BUILD b/pkg/sentry/arch/BUILD index ae3e364cd..65f22af2b 100644 --- a/pkg/sentry/arch/BUILD +++ b/pkg/sentry/arch/BUILD @@ -9,17 +9,23 @@ go_library( srcs = [ "aligned.go", "arch.go", + "arch_aarch64.go", "arch_amd64.go", "arch_amd64.s", + "arch_arm64.go", + "arch_state_aarch64.go", "arch_state_x86.go", "arch_x86.go", "auxv.go", + "signal.go", "signal_act.go", "signal_amd64.go", + "signal_arm64.go", "signal_info.go", "signal_stack.go", "stack.go", "syscalls_amd64.go", + "syscalls_arm64.go", ], importpath = "gvisor.dev/gvisor/pkg/sentry/arch", visibility = ["//:sandbox"], diff --git a/pkg/sentry/arch/arch_aarch64.go b/pkg/sentry/arch/arch_aarch64.go new file mode 100644 index 000000000..ea4dedbdf --- /dev/null +++ b/pkg/sentry/arch/arch_aarch64.go @@ -0,0 +1,293 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package arch + +import ( + "fmt" + "io" + "syscall" + + "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/cpuid" + "gvisor.dev/gvisor/pkg/log" + rpb "gvisor.dev/gvisor/pkg/sentry/arch/registers_go_proto" + "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/syserror" +) + +const ( + // SyscallWidth is the width of insturctions. + SyscallWidth = 4 +) + +// aarch64FPState is aarch64 floating point state. +type aarch64FPState []byte + +// initAarch64FPState (defined in asm files) sets up initial state. +func initAarch64FPState(data *FloatingPointData) { + // TODO(gvisor.dev/issue/1238): floating-point is not supported. +} + +func newAarch64FPStateSlice() []byte { + return alignedBytes(4096, 32)[:4096] +} + +// newAarch64FPState returns an initialized floating point state. +// +// The returned state is large enough to store all floating point state +// supported by host, even if the app won't use much of it due to a restricted +// FeatureSet. Since they may still be able to see state not advertised by +// CPUID we must ensure it does not contain any sentry state. +func newAarch64FPState() aarch64FPState { + f := aarch64FPState(newAarch64FPStateSlice()) + initAarch64FPState(f.FloatingPointData()) + return f +} + +// fork creates and returns an identical copy of the aarch64 floating point state. +func (f aarch64FPState) fork() aarch64FPState { + n := aarch64FPState(newAarch64FPStateSlice()) + copy(n, f) + return n +} + +// FloatingPointData returns the raw data pointer. +func (f aarch64FPState) FloatingPointData() *FloatingPointData { + return (*FloatingPointData)(&f[0]) +} + +// NewFloatingPointData returns a new floating point data blob. +// +// This is primarily for use in tests. +func NewFloatingPointData() *FloatingPointData { + return (*FloatingPointData)(&(newAarch64FPState()[0])) +} + +// State contains the common architecture bits for aarch64 (the build tag of this +// file ensures it's only built on aarch64). +type State struct { + // The system registers. + Regs syscall.PtraceRegs `state:".(syscallPtraceRegs)"` + + // Our floating point state. + aarch64FPState `state:"wait"` + + // FeatureSet is a pointer to the currently active feature set. + FeatureSet *cpuid.FeatureSet +} + +// Proto returns a protobuf representation of the system registers in State. +func (s State) Proto() *rpb.Registers { + regs := &rpb.ARM64Registers{ + R0: s.Regs.Regs[0], + R1: s.Regs.Regs[1], + R2: s.Regs.Regs[2], + R3: s.Regs.Regs[3], + R4: s.Regs.Regs[4], + R5: s.Regs.Regs[5], + R6: s.Regs.Regs[6], + R7: s.Regs.Regs[7], + R8: s.Regs.Regs[8], + R9: s.Regs.Regs[9], + R10: s.Regs.Regs[10], + R11: s.Regs.Regs[11], + R12: s.Regs.Regs[12], + R13: s.Regs.Regs[13], + R14: s.Regs.Regs[14], + R15: s.Regs.Regs[15], + R16: s.Regs.Regs[16], + R17: s.Regs.Regs[17], + R18: s.Regs.Regs[18], + R19: s.Regs.Regs[19], + R20: s.Regs.Regs[20], + R21: s.Regs.Regs[21], + R22: s.Regs.Regs[22], + R23: s.Regs.Regs[23], + R24: s.Regs.Regs[24], + R25: s.Regs.Regs[25], + R26: s.Regs.Regs[26], + R27: s.Regs.Regs[27], + R28: s.Regs.Regs[28], + R29: s.Regs.Regs[29], + R30: s.Regs.Regs[30], + Sp: s.Regs.Sp, + Pc: s.Regs.Pc, + Pstate: s.Regs.Pstate, + } + return &rpb.Registers{Arch: &rpb.Registers_Arm64{Arm64: regs}} +} + +// Fork creates and returns an identical copy of the state. +func (s *State) Fork() State { + // TODO(gvisor.dev/issue/1238): floating-point is not supported. + return State{ + Regs: s.Regs, + FeatureSet: s.FeatureSet, + } +} + +// StateData implements Context.StateData. +func (s *State) StateData() *State { + return s +} + +// CPUIDEmulate emulates a cpuid instruction. +func (s *State) CPUIDEmulate(l log.Logger) { + // TODO(gvisor.dev/issue/1255): cpuid is not supported. +} + +// SingleStep implements Context.SingleStep. +func (s *State) SingleStep() bool { + return false +} + +// SetSingleStep enables single stepping. +func (s *State) SetSingleStep() { + // Set the trap flag. + // TODO(gvisor.dev/issue/1239): ptrace single-step is not supported. +} + +// ClearSingleStep enables single stepping. +func (s *State) ClearSingleStep() { + // Clear the trap flag. + // TODO(gvisor.dev/issue/1239): ptrace single-step is not supported. +} + +// RegisterMap returns a map of all registers. +func (s *State) RegisterMap() (map[string]uintptr, error) { + return map[string]uintptr{ + "R0": uintptr(s.Regs.Regs[0]), + "R1": uintptr(s.Regs.Regs[1]), + "R2": uintptr(s.Regs.Regs[2]), + "R3": uintptr(s.Regs.Regs[3]), + "R4": uintptr(s.Regs.Regs[4]), + "R5": uintptr(s.Regs.Regs[5]), + "R6": uintptr(s.Regs.Regs[6]), + "R7": uintptr(s.Regs.Regs[7]), + "R8": uintptr(s.Regs.Regs[8]), + "R9": uintptr(s.Regs.Regs[9]), + "R10": uintptr(s.Regs.Regs[10]), + "R11": uintptr(s.Regs.Regs[11]), + "R12": uintptr(s.Regs.Regs[12]), + "R13": uintptr(s.Regs.Regs[13]), + "R14": uintptr(s.Regs.Regs[14]), + "R15": uintptr(s.Regs.Regs[15]), + "R16": uintptr(s.Regs.Regs[16]), + "R17": uintptr(s.Regs.Regs[17]), + "R18": uintptr(s.Regs.Regs[18]), + "R19": uintptr(s.Regs.Regs[19]), + "R20": uintptr(s.Regs.Regs[20]), + "R21": uintptr(s.Regs.Regs[21]), + "R22": uintptr(s.Regs.Regs[22]), + "R23": uintptr(s.Regs.Regs[23]), + "R24": uintptr(s.Regs.Regs[24]), + "R25": uintptr(s.Regs.Regs[25]), + "R26": uintptr(s.Regs.Regs[26]), + "R27": uintptr(s.Regs.Regs[27]), + "R28": uintptr(s.Regs.Regs[28]), + "R29": uintptr(s.Regs.Regs[29]), + "R30": uintptr(s.Regs.Regs[30]), + "Sp": uintptr(s.Regs.Sp), + "Pc": uintptr(s.Regs.Pc), + "Pstate": uintptr(s.Regs.Pstate), + }, nil +} + +// PtraceGetRegs implements Context.PtraceGetRegs. +func (s *State) PtraceGetRegs(dst io.Writer) (int, error) { + return dst.Write(binary.Marshal(nil, usermem.ByteOrder, s.ptraceGetRegs())) +} + +func (s *State) ptraceGetRegs() syscall.PtraceRegs { + return s.Regs +} + +var ptraceRegsSize = int(binary.Size(syscall.PtraceRegs{})) + +// PtraceSetRegs implements Context.PtraceSetRegs. +func (s *State) PtraceSetRegs(src io.Reader) (int, error) { + var regs syscall.PtraceRegs + buf := make([]byte, ptraceRegsSize) + if _, err := io.ReadFull(src, buf); err != nil { + return 0, err + } + binary.Unmarshal(buf, usermem.ByteOrder, ®s) + s.Regs = regs + return ptraceRegsSize, nil +} + +// PtraceGetFPRegs implements Context.PtraceGetFPRegs. +func (s *State) PtraceGetFPRegs(dst io.Writer) (int, error) { + // TODO(gvisor.dev/issue/1238): floating-point is not supported. + return 0, nil +} + +// PtraceSetFPRegs implements Context.PtraceSetFPRegs. +func (s *State) PtraceSetFPRegs(src io.Reader) (int, error) { + // TODO(gvisor.dev/issue/1238): floating-point is not supported. + return 0, nil +} + +// Register sets defined in include/uapi/linux/elf.h. +const ( + _NT_PRSTATUS = 1 + _NT_PRFPREG = 2 +) + +// PtraceGetRegSet implements Context.PtraceGetRegSet. +func (s *State) PtraceGetRegSet(regset uintptr, dst io.Writer, maxlen int) (int, error) { + switch regset { + case _NT_PRSTATUS: + if maxlen < ptraceRegsSize { + return 0, syserror.EFAULT + } + return s.PtraceGetRegs(dst) + default: + return 0, syserror.EINVAL + } +} + +// PtraceSetRegSet implements Context.PtraceSetRegSet. +func (s *State) PtraceSetRegSet(regset uintptr, src io.Reader, maxlen int) (int, error) { + switch regset { + case _NT_PRSTATUS: + if maxlen < ptraceRegsSize { + return 0, syserror.EFAULT + } + return s.PtraceSetRegs(src) + default: + return 0, syserror.EINVAL + } +} + +// FullRestore indicates whether a full restore is required. +func (s *State) FullRestore() bool { + return false +} + +// New returns a new architecture context. +func New(arch Arch, fs *cpuid.FeatureSet) Context { + switch arch { + case ARM64: + return &context64{ + State{ + FeatureSet: fs, + }, + } + } + panic(fmt.Sprintf("unknown architecture %v", arch)) +} diff --git a/pkg/sentry/arch/arch_arm64.go b/pkg/sentry/arch/arch_arm64.go new file mode 100644 index 000000000..0d5b7d317 --- /dev/null +++ b/pkg/sentry/arch/arch_arm64.go @@ -0,0 +1,266 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package arch + +import ( + "fmt" + "math/rand" + "syscall" + + "gvisor.dev/gvisor/pkg/cpuid" + "gvisor.dev/gvisor/pkg/sentry/limits" + "gvisor.dev/gvisor/pkg/sentry/usermem" +) + +// Host specifies the host architecture. +const Host = ARM64 + +// These constants come directly from Linux. +const ( + // maxAddr64 is the maximum userspace address. It is TASK_SIZE in Linux + // for a 64-bit process. + maxAddr64 usermem.Addr = (1 << 48) + + // maxStackRand64 is the maximum randomization to apply to the stack. + // It is defined by arch/arm64/mm/mmap.c:(STACK_RND_MASK << PAGE_SHIFT) in Linux. + maxStackRand64 = 0x3ffff << 12 // 16 GB + + // maxMmapRand64 is the maximum randomization to apply to the mmap + // layout. It is defined by arch/arm64/mm/mmap.c:arch_mmap_rnd in Linux. + maxMmapRand64 = (1 << 33) * usermem.PageSize + + // minGap64 is the minimum gap to leave at the top of the address space + // for the stack. It is defined by arch/arm64/mm/mmap.c:MIN_GAP in Linux. + minGap64 = (128 << 20) + maxStackRand64 + + // preferredPIELoadAddr is the standard Linux position-independent + // executable base load address. It is ELF_ET_DYN_BASE in Linux. + // + // The Platform {Min,Max}UserAddress() may preclude loading at this + // address. See other preferredFoo comments below. + preferredPIELoadAddr usermem.Addr = maxAddr64 / 6 * 5 +) + +// These constants are selected as heuristics to help make the Platform's +// potentially limited address space conform as closely to Linux as possible. +const ( + preferredTopDownAllocMin usermem.Addr = 0x7e8000000000 + preferredAllocationGap = 128 << 30 // 128 GB + preferredTopDownBaseMin = preferredTopDownAllocMin + preferredAllocationGap + + // minMmapRand64 is the smallest we are willing to make the + // randomization to stay above preferredTopDownBaseMin. + minMmapRand64 = (1 << 18) * usermem.PageSize +) + +// context64 represents an ARM64 context. +type context64 struct { + State +} + +// Arch implements Context.Arch. +func (c *context64) Arch() Arch { + return ARM64 +} + +// Fork returns an exact copy of this context. +func (c *context64) Fork() Context { + return &context64{ + State: c.State.Fork(), + } +} + +// General purpose registers usage on Arm64: +// R0...R7: parameter/result registers. +// R8: indirect result location register. +// R9...R15: temporary rgisters. +// R16: the first intra-procedure-call scratch register. +// R17: the second intra-procedure-call scratch register. +// R18: the platform register. +// R19...R28: callee-saved registers. +// R29: the frame pointer. +// R30: the link register. + +// Return returns the current syscall return value. +func (c *context64) Return() uintptr { + return uintptr(c.Regs.Regs[0]) +} + +// SetReturn sets the syscall return value. +func (c *context64) SetReturn(value uintptr) { + c.Regs.Regs[0] = uint64(value) +} + +// IP returns the current instruction pointer. +func (c *context64) IP() uintptr { + return uintptr(c.Regs.Pc) +} + +// SetIP sets the current instruction pointer. +func (c *context64) SetIP(value uintptr) { + c.Regs.Pc = uint64(value) +} + +// Stack returns the current stack pointer. +func (c *context64) Stack() uintptr { + return uintptr(c.Regs.Sp) +} + +// SetStack sets the current stack pointer. +func (c *context64) SetStack(value uintptr) { + c.Regs.Sp = uint64(value) +} + +// TLS returns the current TLS pointer. +func (c *context64) TLS() uintptr { + // TODO(gvisor.dev/issue/1238): TLS is not supported. + // MRS_TPIDR_EL0 + return 0 +} + +// SetTLS sets the current TLS pointer. Returns false if value is invalid. +func (c *context64) SetTLS(value uintptr) bool { + // TODO(gvisor.dev/issue/1238): TLS is not supported. + // MSR_TPIDR_EL0 + return false +} + +// SetRSEQInterruptedIP implements Context.SetRSEQInterruptedIP. +func (c *context64) SetRSEQInterruptedIP(value uintptr) { + c.Regs.Regs[3] = uint64(value) +} + +// Native returns the native type for the given val. +func (c *context64) Native(val uintptr) interface{} { + v := uint64(val) + return &v +} + +// Value returns the generic val for the given native type. +func (c *context64) Value(val interface{}) uintptr { + return uintptr(*val.(*uint64)) +} + +// Width returns the byte width of this architecture. +func (c *context64) Width() uint { + return 8 +} + +// FeatureSet returns the FeatureSet in use. +func (c *context64) FeatureSet() *cpuid.FeatureSet { + return c.State.FeatureSet +} + +// mmapRand returns a random adjustment for randomizing an mmap layout. +func mmapRand(max uint64) usermem.Addr { + return usermem.Addr(rand.Int63n(int64(max))).RoundDown() +} + +// NewMmapLayout implements Context.NewMmapLayout consistently with Linux. +func (c *context64) NewMmapLayout(min, max usermem.Addr, r *limits.LimitSet) (MmapLayout, error) { + min, ok := min.RoundUp() + if !ok { + return MmapLayout{}, syscall.EINVAL + } + if max > maxAddr64 { + max = maxAddr64 + } + max = max.RoundDown() + + if min > max { + return MmapLayout{}, syscall.EINVAL + } + + stackSize := r.Get(limits.Stack) + + // MAX_GAP in Linux. + maxGap := (max / 6) * 5 + gap := usermem.Addr(stackSize.Cur) + if gap < minGap64 { + gap = minGap64 + } + if gap > maxGap { + gap = maxGap + } + defaultDir := MmapTopDown + if stackSize.Cur == limits.Infinity { + defaultDir = MmapBottomUp + } + + topDownMin := max - gap - maxMmapRand64 + maxRand := usermem.Addr(maxMmapRand64) + if topDownMin < preferredTopDownBaseMin { + // Try to keep TopDownBase above preferredTopDownBaseMin by + // shrinking maxRand. + maxAdjust := maxRand - minMmapRand64 + needAdjust := preferredTopDownBaseMin - topDownMin + if needAdjust <= maxAdjust { + maxRand -= needAdjust + } + } + + rnd := mmapRand(uint64(maxRand)) + l := MmapLayout{ + MinAddr: min, + MaxAddr: max, + // TASK_UNMAPPED_BASE in Linux. + BottomUpBase: (max/3 + rnd).RoundDown(), + TopDownBase: (max - gap - rnd).RoundDown(), + DefaultDirection: defaultDir, + // We may have reduced the maximum randomization to keep + // TopDownBase above preferredTopDownBaseMin while maintaining + // our stack gap. Stack allocations must use that max + // randomization to avoiding eating into the gap. + MaxStackRand: uint64(maxRand), + } + + // Final sanity check on the layout. + if !l.Valid() { + panic(fmt.Sprintf("Invalid MmapLayout: %+v", l)) + } + + return l, nil +} + +// PIELoadAddress implements Context.PIELoadAddress. +func (c *context64) PIELoadAddress(l MmapLayout) usermem.Addr { + base := preferredPIELoadAddr + max, ok := base.AddLength(maxMmapRand64) + if !ok { + panic(fmt.Sprintf("preferredPIELoadAddr %#x too large", base)) + } + + if max > l.MaxAddr { + // preferredPIELoadAddr won't fit; fall back to the standard + // Linux behavior of 2/3 of TopDownBase. TSAN won't like this. + // + // Don't bother trying to shrink the randomization for now. + base = l.TopDownBase / 3 * 2 + } + + return base + mmapRand(maxMmapRand64) +} + +// PtracePeekUser implements Context.PtracePeekUser. +func (c *context64) PtracePeekUser(addr uintptr) (interface{}, error) { + // TODO(gvisor.dev/issue/1239): Full ptrace supporting for Arm64. + return c.Native(0), nil +} + +// PtracePokeUser implements Context.PtracePokeUser. +func (c *context64) PtracePokeUser(addr, data uintptr) error { + // TODO(gvisor.dev/issue/1239): Full ptrace supporting for Arm64. + return nil +} diff --git a/pkg/sentry/arch/arch_state_aarch64.go b/pkg/sentry/arch/arch_state_aarch64.go new file mode 100644 index 000000000..0136a85ad --- /dev/null +++ b/pkg/sentry/arch/arch_state_aarch64.go @@ -0,0 +1,38 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package arch + +import ( + "syscall" +) + +type syscallPtraceRegs struct { + Regs [31]uint64 + Sp uint64 + Pc uint64 + Pstate uint64 +} + +// saveRegs is invoked by stateify. +func (s *State) saveRegs() syscallPtraceRegs { + return syscallPtraceRegs(s.Regs) +} + +// loadRegs is invoked by stateify. +func (s *State) loadRegs(r syscallPtraceRegs) { + s.Regs = syscall.PtraceRegs(r) +} diff --git a/pkg/sentry/arch/arch_state_x86.go b/pkg/sentry/arch/arch_state_x86.go index 9061fcc86..84f11b0d1 100644 --- a/pkg/sentry/arch/arch_state_x86.go +++ b/pkg/sentry/arch/arch_state_x86.go @@ -12,6 +12,8 @@ // See the License for the specific language governing permissions and // limitations under the License. +// +build amd64 i386 + package arch import ( diff --git a/pkg/sentry/arch/registers.proto b/pkg/sentry/arch/registers.proto index 9dc83e241..60c027aab 100644 --- a/pkg/sentry/arch/registers.proto +++ b/pkg/sentry/arch/registers.proto @@ -48,8 +48,45 @@ message AMD64Registers { uint64 gs_base = 27; } +message ARM64Registers { + uint64 r0 = 1; + uint64 r1 = 2; + uint64 r2 = 3; + uint64 r3 = 4; + uint64 r4 = 5; + uint64 r5 = 6; + uint64 r6 = 7; + uint64 r7 = 8; + uint64 r8 = 9; + uint64 r9 = 10; + uint64 r10 = 11; + uint64 r11 = 12; + uint64 r12 = 13; + uint64 r13 = 14; + uint64 r14 = 15; + uint64 r15 = 16; + uint64 r16 = 17; + uint64 r17 = 18; + uint64 r18 = 19; + uint64 r19 = 20; + uint64 r20 = 21; + uint64 r21 = 22; + uint64 r22 = 23; + uint64 r23 = 24; + uint64 r24 = 25; + uint64 r25 = 26; + uint64 r26 = 27; + uint64 r27 = 28; + uint64 r28 = 29; + uint64 r29 = 30; + uint64 r30 = 31; + uint64 sp = 32; + uint64 pc = 33; + uint64 pstate = 34; +} message Registers { oneof arch { AMD64Registers amd64 = 1; + ARM64Registers arm64 = 2; } } diff --git a/pkg/sentry/arch/signal.go b/pkg/sentry/arch/signal.go new file mode 100644 index 000000000..402e46025 --- /dev/null +++ b/pkg/sentry/arch/signal.go @@ -0,0 +1,250 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package arch + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/usermem" +) + +// SignalAct represents the action that should be taken when a signal is +// delivered, and is equivalent to struct sigaction. +// +// +stateify savable +type SignalAct struct { + Handler uint64 + Flags uint64 + Restorer uint64 // Only used on amd64. + Mask linux.SignalSet +} + +// SerializeFrom implements NativeSignalAct.SerializeFrom. +func (s *SignalAct) SerializeFrom(other *SignalAct) { + *s = *other +} + +// DeserializeTo implements NativeSignalAct.DeserializeTo. +func (s *SignalAct) DeserializeTo(other *SignalAct) { + *other = *s +} + +// SignalStack represents information about a user stack, and is equivalent to +// stack_t. +// +// +stateify savable +type SignalStack struct { + Addr uint64 + Flags uint32 + _ uint32 + Size uint64 +} + +// SerializeFrom implements NativeSignalStack.SerializeFrom. +func (s *SignalStack) SerializeFrom(other *SignalStack) { + *s = *other +} + +// DeserializeTo implements NativeSignalStack.DeserializeTo. +func (s *SignalStack) DeserializeTo(other *SignalStack) { + *other = *s +} + +// SignalInfo represents information about a signal being delivered, and is +// equivalent to struct siginfo in linux kernel(linux/include/uapi/asm-generic/siginfo.h). +// +// +stateify savable +type SignalInfo struct { + Signo int32 // Signal number + Errno int32 // Errno value + Code int32 // Signal code + _ uint32 + + // struct siginfo::_sifields is a union. In SignalInfo, fields in the union + // are accessed through methods. + // + // For reference, here is the definition of _sifields: (_sigfault._trapno, + // which does not exist on x86, omitted for clarity) + // + // union { + // int _pad[SI_PAD_SIZE]; + // + // /* kill() */ + // struct { + // __kernel_pid_t _pid; /* sender's pid */ + // __ARCH_SI_UID_T _uid; /* sender's uid */ + // } _kill; + // + // /* POSIX.1b timers */ + // struct { + // __kernel_timer_t _tid; /* timer id */ + // int _overrun; /* overrun count */ + // char _pad[sizeof( __ARCH_SI_UID_T) - sizeof(int)]; + // sigval_t _sigval; /* same as below */ + // int _sys_private; /* not to be passed to user */ + // } _timer; + // + // /* POSIX.1b signals */ + // struct { + // __kernel_pid_t _pid; /* sender's pid */ + // __ARCH_SI_UID_T _uid; /* sender's uid */ + // sigval_t _sigval; + // } _rt; + // + // /* SIGCHLD */ + // struct { + // __kernel_pid_t _pid; /* which child */ + // __ARCH_SI_UID_T _uid; /* sender's uid */ + // int _status; /* exit code */ + // __ARCH_SI_CLOCK_T _utime; + // __ARCH_SI_CLOCK_T _stime; + // } _sigchld; + // + // /* SIGILL, SIGFPE, SIGSEGV, SIGBUS */ + // struct { + // void *_addr; /* faulting insn/memory ref. */ + // short _addr_lsb; /* LSB of the reported address */ + // } _sigfault; + // + // /* SIGPOLL */ + // struct { + // __ARCH_SI_BAND_T _band; /* POLL_IN, POLL_OUT, POLL_MSG */ + // int _fd; + // } _sigpoll; + // + // /* SIGSYS */ + // struct { + // void *_call_addr; /* calling user insn */ + // int _syscall; /* triggering system call number */ + // unsigned int _arch; /* AUDIT_ARCH_* of syscall */ + // } _sigsys; + // } _sifields; + // + // _sifields is padded so that the size of siginfo is SI_MAX_SIZE = 128 + // bytes. + Fields [128 - 16]byte +} + +// FixSignalCodeForUser fixes up si_code. +// +// The si_code we get from Linux may contain the kernel-specific code in the +// top 16 bits if it's positive (e.g., from ptrace). Linux's +// copy_siginfo_to_user does +// err |= __put_user((short)from->si_code, &to->si_code); +// to mask out those bits and we need to do the same. +func (s *SignalInfo) FixSignalCodeForUser() { + if s.Code > 0 { + s.Code &= 0x0000ffff + } +} + +// Pid returns the si_pid field. +func (s *SignalInfo) Pid() int32 { + return int32(usermem.ByteOrder.Uint32(s.Fields[0:4])) +} + +// SetPid mutates the si_pid field. +func (s *SignalInfo) SetPid(val int32) { + usermem.ByteOrder.PutUint32(s.Fields[0:4], uint32(val)) +} + +// Uid returns the si_uid field. +func (s *SignalInfo) Uid() int32 { + return int32(usermem.ByteOrder.Uint32(s.Fields[4:8])) +} + +// SetUid mutates the si_uid field. +func (s *SignalInfo) SetUid(val int32) { + usermem.ByteOrder.PutUint32(s.Fields[4:8], uint32(val)) +} + +// Sigval returns the sigval field, which is aliased to both si_int and si_ptr. +func (s *SignalInfo) Sigval() uint64 { + return usermem.ByteOrder.Uint64(s.Fields[8:16]) +} + +// SetSigval mutates the sigval field. +func (s *SignalInfo) SetSigval(val uint64) { + usermem.ByteOrder.PutUint64(s.Fields[8:16], val) +} + +// TimerID returns the si_timerid field. +func (s *SignalInfo) TimerID() linux.TimerID { + return linux.TimerID(usermem.ByteOrder.Uint32(s.Fields[0:4])) +} + +// SetTimerID sets the si_timerid field. +func (s *SignalInfo) SetTimerID(val linux.TimerID) { + usermem.ByteOrder.PutUint32(s.Fields[0:4], uint32(val)) +} + +// Overrun returns the si_overrun field. +func (s *SignalInfo) Overrun() int32 { + return int32(usermem.ByteOrder.Uint32(s.Fields[4:8])) +} + +// SetOverrun sets the si_overrun field. +func (s *SignalInfo) SetOverrun(val int32) { + usermem.ByteOrder.PutUint32(s.Fields[4:8], uint32(val)) +} + +// Addr returns the si_addr field. +func (s *SignalInfo) Addr() uint64 { + return usermem.ByteOrder.Uint64(s.Fields[0:8]) +} + +// SetAddr sets the si_addr field. +func (s *SignalInfo) SetAddr(val uint64) { + usermem.ByteOrder.PutUint64(s.Fields[0:8], val) +} + +// Status returns the si_status field. +func (s *SignalInfo) Status() int32 { + return int32(usermem.ByteOrder.Uint32(s.Fields[8:12])) +} + +// SetStatus mutates the si_status field. +func (s *SignalInfo) SetStatus(val int32) { + usermem.ByteOrder.PutUint32(s.Fields[8:12], uint32(val)) +} + +// CallAddr returns the si_call_addr field. +func (s *SignalInfo) CallAddr() uint64 { + return usermem.ByteOrder.Uint64(s.Fields[0:8]) +} + +// SetCallAddr mutates the si_call_addr field. +func (s *SignalInfo) SetCallAddr(val uint64) { + usermem.ByteOrder.PutUint64(s.Fields[0:8], val) +} + +// Syscall returns the si_syscall field. +func (s *SignalInfo) Syscall() int32 { + return int32(usermem.ByteOrder.Uint32(s.Fields[8:12])) +} + +// SetSyscall mutates the si_syscall field. +func (s *SignalInfo) SetSyscall(val int32) { + usermem.ByteOrder.PutUint32(s.Fields[8:12], uint32(val)) +} + +// Arch returns the si_arch field. +func (s *SignalInfo) Arch() uint32 { + return usermem.ByteOrder.Uint32(s.Fields[12:16]) +} + +// SetArch mutates the si_arch field. +func (s *SignalInfo) SetArch(val uint32) { + usermem.ByteOrder.PutUint32(s.Fields[12:16], val) +} diff --git a/pkg/sentry/arch/signal_amd64.go b/pkg/sentry/arch/signal_amd64.go index febd6f9b9..1e4f9c3c2 100644 --- a/pkg/sentry/arch/signal_amd64.go +++ b/pkg/sentry/arch/signal_amd64.go @@ -26,236 +26,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/usermem" ) -// SignalAct represents the action that should be taken when a signal is -// delivered, and is equivalent to struct sigaction on 64-bit x86. -// -// +stateify savable -type SignalAct struct { - Handler uint64 - Flags uint64 - Restorer uint64 - Mask linux.SignalSet -} - -// SerializeFrom implements NativeSignalAct.SerializeFrom. -func (s *SignalAct) SerializeFrom(other *SignalAct) { - *s = *other -} - -// DeserializeTo implements NativeSignalAct.DeserializeTo. -func (s *SignalAct) DeserializeTo(other *SignalAct) { - *other = *s -} - -// SignalStack represents information about a user stack, and is equivalent to -// stack_t on 64-bit x86. -// -// +stateify savable -type SignalStack struct { - Addr uint64 - Flags uint32 - _ uint32 - Size uint64 -} - -// SerializeFrom implements NativeSignalStack.SerializeFrom. -func (s *SignalStack) SerializeFrom(other *SignalStack) { - *s = *other -} - -// DeserializeTo implements NativeSignalStack.DeserializeTo. -func (s *SignalStack) DeserializeTo(other *SignalStack) { - *other = *s -} - -// SignalInfo represents information about a signal being delivered, and is -// equivalent to struct siginfo on 64-bit x86. -// -// +stateify savable -type SignalInfo struct { - Signo int32 // Signal number - Errno int32 // Errno value - Code int32 // Signal code - _ uint32 - - // struct siginfo::_sifields is a union. In SignalInfo, fields in the union - // are accessed through methods. - // - // For reference, here is the definition of _sifields: (_sigfault._trapno, - // which does not exist on x86, omitted for clarity) - // - // union { - // int _pad[SI_PAD_SIZE]; - // - // /* kill() */ - // struct { - // __kernel_pid_t _pid; /* sender's pid */ - // __ARCH_SI_UID_T _uid; /* sender's uid */ - // } _kill; - // - // /* POSIX.1b timers */ - // struct { - // __kernel_timer_t _tid; /* timer id */ - // int _overrun; /* overrun count */ - // char _pad[sizeof( __ARCH_SI_UID_T) - sizeof(int)]; - // sigval_t _sigval; /* same as below */ - // int _sys_private; /* not to be passed to user */ - // } _timer; - // - // /* POSIX.1b signals */ - // struct { - // __kernel_pid_t _pid; /* sender's pid */ - // __ARCH_SI_UID_T _uid; /* sender's uid */ - // sigval_t _sigval; - // } _rt; - // - // /* SIGCHLD */ - // struct { - // __kernel_pid_t _pid; /* which child */ - // __ARCH_SI_UID_T _uid; /* sender's uid */ - // int _status; /* exit code */ - // __ARCH_SI_CLOCK_T _utime; - // __ARCH_SI_CLOCK_T _stime; - // } _sigchld; - // - // /* SIGILL, SIGFPE, SIGSEGV, SIGBUS */ - // struct { - // void *_addr; /* faulting insn/memory ref. */ - // short _addr_lsb; /* LSB of the reported address */ - // } _sigfault; - // - // /* SIGPOLL */ - // struct { - // __ARCH_SI_BAND_T _band; /* POLL_IN, POLL_OUT, POLL_MSG */ - // int _fd; - // } _sigpoll; - // - // /* SIGSYS */ - // struct { - // void *_call_addr; /* calling user insn */ - // int _syscall; /* triggering system call number */ - // unsigned int _arch; /* AUDIT_ARCH_* of syscall */ - // } _sigsys; - // } _sifields; - // - // _sifields is padded so that the size of siginfo is SI_MAX_SIZE = 128 - // bytes. - Fields [128 - 16]byte -} - -// FixSignalCodeForUser fixes up si_code. -// -// The si_code we get from Linux may contain the kernel-specific code in the -// top 16 bits if it's positive (e.g., from ptrace). Linux's -// copy_siginfo_to_user does -// err |= __put_user((short)from->si_code, &to->si_code); -// to mask out those bits and we need to do the same. -func (s *SignalInfo) FixSignalCodeForUser() { - if s.Code > 0 { - s.Code &= 0x0000ffff - } -} - -// Pid returns the si_pid field. -func (s *SignalInfo) Pid() int32 { - return int32(usermem.ByteOrder.Uint32(s.Fields[0:4])) -} - -// SetPid mutates the si_pid field. -func (s *SignalInfo) SetPid(val int32) { - usermem.ByteOrder.PutUint32(s.Fields[0:4], uint32(val)) -} - -// Uid returns the si_uid field. -func (s *SignalInfo) Uid() int32 { - return int32(usermem.ByteOrder.Uint32(s.Fields[4:8])) -} - -// SetUid mutates the si_uid field. -func (s *SignalInfo) SetUid(val int32) { - usermem.ByteOrder.PutUint32(s.Fields[4:8], uint32(val)) -} - -// Sigval returns the sigval field, which is aliased to both si_int and si_ptr. -func (s *SignalInfo) Sigval() uint64 { - return usermem.ByteOrder.Uint64(s.Fields[8:16]) -} - -// SetSigval mutates the sigval field. -func (s *SignalInfo) SetSigval(val uint64) { - usermem.ByteOrder.PutUint64(s.Fields[8:16], val) -} - -// TimerID returns the si_timerid field. -func (s *SignalInfo) TimerID() linux.TimerID { - return linux.TimerID(usermem.ByteOrder.Uint32(s.Fields[0:4])) -} - -// SetTimerID sets the si_timerid field. -func (s *SignalInfo) SetTimerID(val linux.TimerID) { - usermem.ByteOrder.PutUint32(s.Fields[0:4], uint32(val)) -} - -// Overrun returns the si_overrun field. -func (s *SignalInfo) Overrun() int32 { - return int32(usermem.ByteOrder.Uint32(s.Fields[4:8])) -} - -// SetOverrun sets the si_overrun field. -func (s *SignalInfo) SetOverrun(val int32) { - usermem.ByteOrder.PutUint32(s.Fields[4:8], uint32(val)) -} - -// Addr returns the si_addr field. -func (s *SignalInfo) Addr() uint64 { - return usermem.ByteOrder.Uint64(s.Fields[0:8]) -} - -// SetAddr sets the si_addr field. -func (s *SignalInfo) SetAddr(val uint64) { - usermem.ByteOrder.PutUint64(s.Fields[0:8], val) -} - -// Status returns the si_status field. -func (s *SignalInfo) Status() int32 { - return int32(usermem.ByteOrder.Uint32(s.Fields[8:12])) -} - -// SetStatus mutates the si_status field. -func (s *SignalInfo) SetStatus(val int32) { - usermem.ByteOrder.PutUint32(s.Fields[8:12], uint32(val)) -} - -// CallAddr returns the si_call_addr field. -func (s *SignalInfo) CallAddr() uint64 { - return usermem.ByteOrder.Uint64(s.Fields[0:8]) -} - -// SetCallAddr mutates the si_call_addr field. -func (s *SignalInfo) SetCallAddr(val uint64) { - usermem.ByteOrder.PutUint64(s.Fields[0:8], val) -} - -// Syscall returns the si_syscall field. -func (s *SignalInfo) Syscall() int32 { - return int32(usermem.ByteOrder.Uint32(s.Fields[8:12])) -} - -// SetSyscall mutates the si_syscall field. -func (s *SignalInfo) SetSyscall(val int32) { - usermem.ByteOrder.PutUint32(s.Fields[8:12], uint32(val)) -} - -// Arch returns the si_arch field. -func (s *SignalInfo) Arch() uint32 { - return usermem.ByteOrder.Uint32(s.Fields[12:16]) -} - -// SetArch mutates the si_arch field. -func (s *SignalInfo) SetArch(val uint32) { - usermem.ByteOrder.PutUint32(s.Fields[12:16], val) -} - // SignalContext64 is equivalent to struct sigcontext, the type passed as the // second argument to signal handlers set by signal(2). type SignalContext64 struct { diff --git a/pkg/sentry/arch/signal_arm64.go b/pkg/sentry/arch/signal_arm64.go new file mode 100644 index 000000000..7d0e98935 --- /dev/null +++ b/pkg/sentry/arch/signal_arm64.go @@ -0,0 +1,126 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package arch + +import ( + "encoding/binary" + "syscall" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/usermem" +) + +// SignalContext64 is equivalent to struct sigcontext, the type passed as the +// second argument to signal handlers set by signal(2). +type SignalContext64 struct { + FaultAddr uint64 + Regs [31]uint64 + Sp uint64 + Pc uint64 + Pstate uint64 + _pad [8]byte // __attribute__((__aligned__(16))) + Reserved [4096]uint8 +} + +// UContext64 is equivalent to ucontext on arm64(arch/arm64/include/uapi/asm/ucontext.h). +type UContext64 struct { + Flags uint64 + Link *UContext64 + Stack SignalStack + Sigset linux.SignalSet + // glibc uses a 1024-bit sigset_t + _pad [(1024 - 64) / 8]byte + // sigcontext must be aligned to 16-byte + _pad2 [8]byte + // last for future expansion + MContext SignalContext64 +} + +// NewSignalAct implements Context.NewSignalAct. +func (c *context64) NewSignalAct() NativeSignalAct { + return &SignalAct{} +} + +// NewSignalStack implements Context.NewSignalStack. +func (c *context64) NewSignalStack() NativeSignalStack { + return &SignalStack{} +} + +// SignalSetup implements Context.SignalSetup. +func (c *context64) SignalSetup(st *Stack, act *SignalAct, info *SignalInfo, alt *SignalStack, sigset linux.SignalSet) error { + sp := st.Bottom + + if !(alt.IsEnabled() && sp == alt.Top()) { + sp -= 128 + } + + // Construct the UContext64 now since we need its size. + uc := &UContext64{ + Flags: 0, + Stack: *alt, + MContext: SignalContext64{ + Regs: c.Regs.Regs, + Sp: c.Regs.Sp, + Pc: c.Regs.Pc, + Pstate: c.Regs.Pstate, + }, + Sigset: sigset, + } + + ucSize := binary.Size(uc) + if ucSize < 0 { + panic("can't get size of UContext64") + } + // st.Arch.Width() is for the restorer address. sizeof(siginfo) == 128. + frameSize := int(st.Arch.Width()) + ucSize + 128 + frameBottom := (sp-usermem.Addr(frameSize)) & ^usermem.Addr(15) - 8 + sp = frameBottom + usermem.Addr(frameSize) + st.Bottom = sp + + // Prior to proceeding, figure out if the frame will exhaust the range + // for the signal stack. This is not allowed, and should immediately + // force signal delivery (reverting to the default handler). + if act.IsOnStack() && alt.IsEnabled() && !alt.Contains(frameBottom) { + return syscall.EFAULT + } + + // Adjust the code. + info.FixSignalCodeForUser() + + // Set up the stack frame. + infoAddr, err := st.Push(info) + if err != nil { + return err + } + ucAddr, err := st.Push(uc) + if err != nil { + return err + } + + // Set up registers. + c.Regs.Sp = uint64(st.Bottom) + c.Regs.Pc = act.Handler + c.Regs.Regs[0] = uint64(info.Signo) + c.Regs.Regs[1] = uint64(infoAddr) + c.Regs.Regs[2] = uint64(ucAddr) + + return nil +} + +// SignalRestore implements Context.SignalRestore. +// Only used on intel. +func (c *context64) SignalRestore(st *Stack, rt bool) (linux.SignalSet, SignalStack, error) { + return 0, SignalStack{}, nil +} diff --git a/pkg/sentry/arch/signal_stack.go b/pkg/sentry/arch/signal_stack.go index 5a3228113..d324da705 100644 --- a/pkg/sentry/arch/signal_stack.go +++ b/pkg/sentry/arch/signal_stack.go @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -// +build i386 amd64 +// +build i386 amd64 arm64 package arch diff --git a/pkg/sentry/arch/syscalls_arm64.go b/pkg/sentry/arch/syscalls_arm64.go new file mode 100644 index 000000000..00d5ef461 --- /dev/null +++ b/pkg/sentry/arch/syscalls_arm64.go @@ -0,0 +1,62 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package arch + +const restartSyscallNr = uintptr(128) + +// SyscallNo returns the syscall number according to the 64-bit convention. +func (c *context64) SyscallNo() uintptr { + return uintptr(c.Regs.Regs[8]) +} + +// SyscallArgs provides syscall arguments according to the 64-bit convention. +// +// Due to the way addresses are mapped for the sentry this binary *must* be +// built in 64-bit mode. So we can just assume the syscall numbers that come +// back match the expected host system call numbers. +// General purpose registers usage on Arm64: +// R0...R7: parameter/result registers. +// R8: indirect result location register. +// R9...R15: temporary registers. +// R16: the first intra-procedure-call scratch register. +// R17: the second intra-procedure-call scratch register. +// R18: the platform register. +// R19...R28: callee-saved registers. +// R29: the frame pointer. +// R30: the link register. +func (c *context64) SyscallArgs() SyscallArguments { + return SyscallArguments{ + SyscallArgument{Value: uintptr(c.Regs.Regs[0])}, + SyscallArgument{Value: uintptr(c.Regs.Regs[1])}, + SyscallArgument{Value: uintptr(c.Regs.Regs[2])}, + SyscallArgument{Value: uintptr(c.Regs.Regs[3])}, + SyscallArgument{Value: uintptr(c.Regs.Regs[4])}, + SyscallArgument{Value: uintptr(c.Regs.Regs[5])}, + } +} + +// RestartSyscall implements Context.RestartSyscall. +func (c *context64) RestartSyscall() { + c.Regs.Pc -= SyscallWidth + c.Regs.Regs[8] = uint64(restartSyscallNr) +} + +// RestartSyscallWithRestartBlock implements Context.RestartSyscallWithRestartBlock. +func (c *context64) RestartSyscallWithRestartBlock() { + c.Regs.Pc -= SyscallWidth + c.Regs.Regs[8] = uint64(restartSyscallNr) +} -- cgit v1.2.3 From d6fb1ec6c7c76040dd20e915b32f9ed795ae7077 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Wed, 15 Jan 2020 16:31:24 -0800 Subject: Add timestamps to VFS2 tmpfs, and implement some of SetStat. PiperOrigin-RevId: 289962040 --- pkg/abi/linux/time.go | 13 ++ pkg/sentry/fsimpl/tmpfs/BUILD | 2 + pkg/sentry/fsimpl/tmpfs/filesystem.go | 16 +- pkg/sentry/fsimpl/tmpfs/regular_file_test.go | 129 ++++++++++++--- pkg/sentry/fsimpl/tmpfs/stat_test.go | 232 +++++++++++++++++++++++++++ pkg/sentry/fsimpl/tmpfs/tmpfs.go | 72 +++++++-- 6 files changed, 425 insertions(+), 39 deletions(-) create mode 100644 pkg/sentry/fsimpl/tmpfs/stat_test.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/time.go b/pkg/abi/linux/time.go index 546668bca..5c5a58cd4 100644 --- a/pkg/abi/linux/time.go +++ b/pkg/abi/linux/time.go @@ -234,6 +234,19 @@ type StatxTimestamp struct { _ int32 } +// ToNsec returns the nanosecond representation. +func (sxts StatxTimestamp) ToNsec() int64 { + return int64(sxts.Sec)*1e9 + int64(sxts.Nsec) +} + +// ToNsecCapped returns the safe nanosecond representation. +func (sxts StatxTimestamp) ToNsecCapped() int64 { + if sxts.Sec > maxSecInDuration { + return math.MaxInt64 + } + return sxts.ToNsec() +} + // NsecToStatxTimestamp translates nanoseconds to StatxTimestamp. func NsecToStatxTimestamp(nsec int64) (ts StatxTimestamp) { return StatxTimestamp{ diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD index 82f5c2f41..7601c7c04 100644 --- a/pkg/sentry/fsimpl/tmpfs/BUILD +++ b/pkg/sentry/fsimpl/tmpfs/BUILD @@ -40,6 +40,7 @@ go_library( "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/pipe", + "//pkg/sentry/kernel/time", "//pkg/sentry/memmap", "//pkg/sentry/pgalloc", "//pkg/sentry/platform", @@ -77,6 +78,7 @@ go_test( srcs = [ "pipe_test.go", "regular_file_test.go", + "stat_test.go", ], embed = [":tmpfs"], deps = [ diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 26979729e..4cd7e9aea 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -56,7 +56,8 @@ afterSymlink: } next := nextVFSD.Impl().(*dentry) if symlink, ok := next.inode.impl.(*symlink); ok && rp.ShouldFollowSymlink() { - // TODO: symlink traversals update access time + // TODO(gvisor.dev/issues/1197): Symlink traversals updates + // access time. if err := rp.HandleSymlink(symlink.target); err != nil { return nil, err } @@ -501,7 +502,8 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa oldParent.inode.decLinksLocked() newParent.inode.incLinksLocked() } - // TODO: update timestamps and parent directory sizes + // TODO(gvisor.dev/issues/1197): Update timestamps and parent directory + // sizes. vfsObj.CommitRenameReplaceDentry(renamedVFSD, &newParent.vfsd, newName, replacedVFSD) return nil } @@ -555,15 +557,11 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { fs.mu.RLock() defer fs.mu.RUnlock() - _, err := resolveLocked(rp) + d, err := resolveLocked(rp) if err != nil { return err } - if opts.Stat.Mask == 0 { - return nil - } - // TODO: implement inode.setStat - return syserror.EPERM + return d.inode.setStat(opts.Stat) } // StatAt implements vfs.FilesystemImpl.StatAt. @@ -587,7 +585,7 @@ func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linu if err != nil { return linux.Statfs{}, err } - // TODO: actually implement statfs + // TODO(gvisor.dev/issues/1197): Actually implement statfs. return linux.Statfs{}, syserror.ENOSYS } diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go index 3731c5b6f..7b0a962f0 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go @@ -18,6 +18,7 @@ import ( "bytes" "fmt" "io" + "sync/atomic" "testing" "gvisor.dev/gvisor/pkg/abi/linux" @@ -29,10 +30,12 @@ import ( "gvisor.dev/gvisor/pkg/sentry/vfs" ) -// newFileFD creates a new file in a new tmpfs mount, and returns the FD. If -// the returned err is not nil, then cleanup should be called when the FD is no -// longer needed. -func newFileFD(ctx context.Context, filename string) (*vfs.FileDescription, func(), error) { +// nextFileID is used to generate unique file names. +var nextFileID int64 + +// newTmpfsRoot creates a new tmpfs mount, and returns the root. If the error +// is not nil, then cleanup should be called when the root is no longer needed. +func newTmpfsRoot(ctx context.Context) (*vfs.VirtualFilesystem, vfs.VirtualDentry, func(), error) { creds := auth.CredentialsFromContext(ctx) vfsObj := vfs.New() @@ -41,36 +44,124 @@ func newFileFD(ctx context.Context, filename string) (*vfs.FileDescription, func }) mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{}) if err != nil { - return nil, nil, fmt.Errorf("failed to create tmpfs root mount: %v", err) + return nil, vfs.VirtualDentry{}, nil, fmt.Errorf("failed to create tmpfs root mount: %v", err) } root := mntns.Root() + return vfsObj, root, func() { + root.DecRef() + mntns.DecRef(vfsObj) + }, nil +} + +// newFileFD creates a new file in a new tmpfs mount, and returns the FD. If +// the returned err is not nil, then cleanup should be called when the FD is no +// longer needed. +func newFileFD(ctx context.Context, mode linux.FileMode) (*vfs.FileDescription, func(), error) { + creds := auth.CredentialsFromContext(ctx) + vfsObj, root, cleanup, err := newTmpfsRoot(ctx) + if err != nil { + return nil, nil, err + } + + filename := fmt.Sprintf("tmpfs-test-file-%d", atomic.AddInt64(&nextFileID, 1)) // Create the file that will be write/read. fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(filename), - FollowFinalSymlink: true, + Root: root, + Start: root, + Path: fspath.Parse(filename), }, &vfs.OpenOptions{ Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL, - Mode: 0644, + Mode: linux.ModeRegular | mode, }) if err != nil { - root.DecRef() - mntns.DecRef(vfsObj) + cleanup() return nil, nil, fmt.Errorf("failed to create file %q: %v", filename, err) } - return fd, func() { - root.DecRef() - mntns.DecRef(vfsObj) - }, nil + return fd, cleanup, nil +} + +// newDirFD is like newFileFD, but for directories. +func newDirFD(ctx context.Context, mode linux.FileMode) (*vfs.FileDescription, func(), error) { + creds := auth.CredentialsFromContext(ctx) + vfsObj, root, cleanup, err := newTmpfsRoot(ctx) + if err != nil { + return nil, nil, err + } + + dirname := fmt.Sprintf("tmpfs-test-dir-%d", atomic.AddInt64(&nextFileID, 1)) + + // Create the dir. + if err := vfsObj.MkdirAt(ctx, creds, &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(dirname), + }, &vfs.MkdirOptions{ + Mode: linux.ModeDirectory | mode, + }); err != nil { + cleanup() + return nil, nil, fmt.Errorf("failed to create directory %q: %v", dirname, err) + } + + // Open the dir and return it. + fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(dirname), + }, &vfs.OpenOptions{ + Flags: linux.O_RDONLY | linux.O_DIRECTORY, + }) + if err != nil { + cleanup() + return nil, nil, fmt.Errorf("failed to open directory %q: %v", dirname, err) + } + + return fd, cleanup, nil +} + +// newPipeFD is like newFileFD, but for pipes. +func newPipeFD(ctx context.Context, mode linux.FileMode) (*vfs.FileDescription, func(), error) { + creds := auth.CredentialsFromContext(ctx) + vfsObj, root, cleanup, err := newTmpfsRoot(ctx) + if err != nil { + return nil, nil, err + } + + pipename := fmt.Sprintf("tmpfs-test-pipe-%d", atomic.AddInt64(&nextFileID, 1)) + + // Create the pipe. + if err := vfsObj.MknodAt(ctx, creds, &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(pipename), + }, &vfs.MknodOptions{ + Mode: linux.ModeNamedPipe | mode, + }); err != nil { + cleanup() + return nil, nil, fmt.Errorf("failed to create pipe %q: %v", pipename, err) + } + + // Open the pipe and return it. + fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(pipename), + }, &vfs.OpenOptions{ + Flags: linux.O_RDWR, + }) + if err != nil { + cleanup() + return nil, nil, fmt.Errorf("failed to open pipe %q: %v", pipename, err) + } + + return fd, cleanup, nil } // Test that we can write some data to a file and read it back.` func TestSimpleWriteRead(t *testing.T) { ctx := contexttest.Context(t) - fd, cleanup, err := newFileFD(ctx, "simpleReadWrite") + fd, cleanup, err := newFileFD(ctx, 0644) if err != nil { t.Fatal(err) } @@ -116,7 +207,7 @@ func TestSimpleWriteRead(t *testing.T) { func TestPWrite(t *testing.T) { ctx := contexttest.Context(t) - fd, cleanup, err := newFileFD(ctx, "PRead") + fd, cleanup, err := newFileFD(ctx, 0644) if err != nil { t.Fatal(err) } @@ -171,7 +262,7 @@ func TestPWrite(t *testing.T) { func TestPRead(t *testing.T) { ctx := contexttest.Context(t) - fd, cleanup, err := newFileFD(ctx, "PRead") + fd, cleanup, err := newFileFD(ctx, 0644) if err != nil { t.Fatal(err) } diff --git a/pkg/sentry/fsimpl/tmpfs/stat_test.go b/pkg/sentry/fsimpl/tmpfs/stat_test.go new file mode 100644 index 000000000..ebe035dee --- /dev/null +++ b/pkg/sentry/fsimpl/tmpfs/stat_test.go @@ -0,0 +1,232 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tmpfs + +import ( + "fmt" + "testing" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" + "gvisor.dev/gvisor/pkg/sentry/vfs" +) + +func TestStatAfterCreate(t *testing.T) { + ctx := contexttest.Context(t) + mode := linux.FileMode(0644) + + // Run with different file types. + // TODO(gvisor.dev/issues/1197): Also test symlinks and sockets. + for _, typ := range []string{"file", "dir", "pipe"} { + t.Run(fmt.Sprintf("type=%q", typ), func(t *testing.T) { + var ( + fd *vfs.FileDescription + cleanup func() + err error + ) + switch typ { + case "file": + fd, cleanup, err = newFileFD(ctx, mode) + case "dir": + fd, cleanup, err = newDirFD(ctx, mode) + case "pipe": + fd, cleanup, err = newPipeFD(ctx, mode) + default: + panic(fmt.Sprintf("unknown typ %q", typ)) + } + if err != nil { + t.Fatal(err) + } + defer cleanup() + + got, err := fd.Stat(ctx, vfs.StatOptions{}) + if err != nil { + t.Fatalf("Stat failed: %v", err) + } + + // Atime, Ctime, Mtime should all be current time (non-zero). + atime, ctime, mtime := got.Atime.ToNsec(), got.Ctime.ToNsec(), got.Mtime.ToNsec() + if atime != ctime || ctime != mtime { + t.Errorf("got atime=%d ctime=%d mtime=%d, wanted equal values", atime, ctime, mtime) + } + if atime == 0 { + t.Errorf("got atime=%d, want non-zero", atime) + } + + // Btime should be 0, as it is not set by tmpfs. + if btime := got.Btime.ToNsec(); btime != 0 { + t.Errorf("got btime %d, want 0", got.Btime.ToNsec()) + } + + // Size should be 0. + if got.Size != 0 { + t.Errorf("got size %d, want 0", got.Size) + } + + // Nlink should be 1 for files, 2 for dirs. + wantNlink := uint32(1) + if typ == "dir" { + wantNlink = 2 + } + if got.Nlink != wantNlink { + t.Errorf("got nlink %d, want %d", got.Nlink, wantNlink) + } + + // UID and GID are set from context creds. + creds := auth.CredentialsFromContext(ctx) + if got.UID != uint32(creds.EffectiveKUID) { + t.Errorf("got uid %d, want %d", got.UID, uint32(creds.EffectiveKUID)) + } + if got.GID != uint32(creds.EffectiveKGID) { + t.Errorf("got gid %d, want %d", got.GID, uint32(creds.EffectiveKGID)) + } + + // Mode. + wantMode := uint16(mode) + switch typ { + case "file": + wantMode |= linux.S_IFREG + case "dir": + wantMode |= linux.S_IFDIR + case "pipe": + wantMode |= linux.S_IFIFO + default: + panic(fmt.Sprintf("unknown typ %q", typ)) + } + + if got.Mode != wantMode { + t.Errorf("got mode %x, want %x", got.Mode, wantMode) + } + + // Ino. + if got.Ino == 0 { + t.Errorf("got ino %d, want not 0", got.Ino) + } + }) + } +} + +func TestSetStatAtime(t *testing.T) { + ctx := contexttest.Context(t) + fd, cleanup, err := newFileFD(ctx, 0644) + if err != nil { + t.Fatal(err) + } + defer cleanup() + + allStatOptions := vfs.StatOptions{Mask: linux.STATX_ALL} + + // Get initial stat. + initialStat, err := fd.Stat(ctx, allStatOptions) + if err != nil { + t.Fatalf("Stat failed: %v", err) + } + + // Set atime, but without the mask. + if err := fd.SetStat(ctx, vfs.SetStatOptions{Stat: linux.Statx{ + Mask: 0, + Atime: linux.NsecToStatxTimestamp(100), + }}); err != nil { + t.Errorf("SetStat atime without mask failed: %v") + } + // Atime should be unchanged. + if gotStat, err := fd.Stat(ctx, allStatOptions); err != nil { + t.Errorf("Stat got error: %v", err) + } else if gotStat.Atime != initialStat.Atime { + t.Errorf("Stat got atime %d, want %d", gotStat.Atime, initialStat.Atime) + } + + // Set atime, this time included in the mask. + setStat := linux.Statx{ + Mask: linux.STATX_ATIME, + Atime: linux.NsecToStatxTimestamp(100), + } + if err := fd.SetStat(ctx, vfs.SetStatOptions{Stat: setStat}); err != nil { + t.Errorf("SetStat atime with mask failed: %v") + } + if gotStat, err := fd.Stat(ctx, allStatOptions); err != nil { + t.Errorf("Stat got error: %v", err) + } else if gotStat.Atime != setStat.Atime { + t.Errorf("Stat got atime %d, want %d", gotStat.Atime, setStat.Atime) + } +} + +func TestSetStat(t *testing.T) { + ctx := contexttest.Context(t) + mode := linux.FileMode(0644) + + // Run with different file types. + // TODO(gvisor.dev/issues/1197): Also test symlinks and sockets. + for _, typ := range []string{"file", "dir", "pipe"} { + t.Run(fmt.Sprintf("type=%q", typ), func(t *testing.T) { + var ( + fd *vfs.FileDescription + cleanup func() + err error + ) + switch typ { + case "file": + fd, cleanup, err = newFileFD(ctx, mode) + case "dir": + fd, cleanup, err = newDirFD(ctx, mode) + case "pipe": + fd, cleanup, err = newPipeFD(ctx, mode) + default: + panic(fmt.Sprintf("unknown typ %q", typ)) + } + if err != nil { + t.Fatal(err) + } + defer cleanup() + + allStatOptions := vfs.StatOptions{Mask: linux.STATX_ALL} + + // Get initial stat. + initialStat, err := fd.Stat(ctx, allStatOptions) + if err != nil { + t.Fatalf("Stat failed: %v", err) + } + + // Set atime, but without the mask. + if err := fd.SetStat(ctx, vfs.SetStatOptions{Stat: linux.Statx{ + Mask: 0, + Atime: linux.NsecToStatxTimestamp(100), + }}); err != nil { + t.Errorf("SetStat atime without mask failed: %v") + } + // Atime should be unchanged. + if gotStat, err := fd.Stat(ctx, allStatOptions); err != nil { + t.Errorf("Stat got error: %v", err) + } else if gotStat.Atime != initialStat.Atime { + t.Errorf("Stat got atime %d, want %d", gotStat.Atime, initialStat.Atime) + } + + // Set atime, this time included in the mask. + setStat := linux.Statx{ + Mask: linux.STATX_ATIME, + Atime: linux.NsecToStatxTimestamp(100), + } + if err := fd.SetStat(ctx, vfs.SetStatOptions{Stat: setStat}); err != nil { + t.Errorf("SetStat atime with mask failed: %v") + } + if gotStat, err := fd.Stat(ctx, allStatOptions); err != nil { + t.Errorf("Stat got error: %v", err) + } else if gotStat.Atime != setStat.Atime { + t.Errorf("Stat got atime %d, want %d", gotStat.Atime, setStat.Atime) + } + }) + } +} diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 701826f90..d6960ee47 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -31,10 +31,10 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/syserror" ) // FilesystemType implements vfs.FilesystemType. @@ -47,6 +47,9 @@ type filesystem struct { // memFile is used to allocate pages to for regular files. memFile *pgalloc.MemoryFile + // clock is a realtime clock used to set timestamps in file operations. + clock time.Clock + // mu serializes changes to the Dentry tree. mu sync.RWMutex @@ -59,8 +62,10 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt if memFileProvider == nil { panic("MemoryFileProviderFromContext returned nil") } + clock := time.RealtimeClockFromContext(ctx) fs := filesystem{ memFile: memFileProvider.MemoryFile(), + clock: clock, } fs.vfsfs.Init(vfsObj, &fs) root := fs.newDentry(fs.newDirectory(creds, 01777)) @@ -126,26 +131,36 @@ type inode struct { // filesystem.RmdirAt() drops the reference. refs int64 - // Inode metadata; protected by mu and accessed using atomic memory - // operations unless otherwise specified. - mu sync.RWMutex + // Inode metadata. Writing multiple fields atomically requires holding + // mu, othewise atomic operations can be used. + mu sync.Mutex mode uint32 // excluding file type bits, which are based on impl nlink uint32 // protected by filesystem.mu instead of inode.mu uid uint32 // auth.KUID, but stored as raw uint32 for sync/atomic gid uint32 // auth.KGID, but ... ino uint64 // immutable + // Linux's tmpfs has no concept of btime. + atime int64 // nanoseconds + ctime int64 // nanoseconds + mtime int64 // nanoseconds + impl interface{} // immutable } const maxLinks = math.MaxUint32 func (i *inode) init(impl interface{}, fs *filesystem, creds *auth.Credentials, mode linux.FileMode) { + now := fs.clock.Now().Nanoseconds() i.refs = 1 i.mode = uint32(mode) i.uid = uint32(creds.EffectiveKUID) i.gid = uint32(creds.EffectiveKGID) i.ino = atomic.AddUint64(&fs.nextInoMinusOne, 1) + // Tmpfs creation sets atime, ctime, and mtime to current time. + i.atime = now + i.ctime = now + i.mtime = now // i.nlink initialized by caller i.impl = impl } @@ -213,15 +228,24 @@ func (i *inode) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes, i // Go won't inline this function, and returning linux.Statx (which is quite // big) means spending a lot of time in runtime.duffcopy(), so instead it's an // output parameter. +// +// Note that Linux does not guarantee to return consistent data (in the case of +// a concurrent modification), so we do not require holding inode.mu. func (i *inode) statTo(stat *linux.Statx) { - stat.Mask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_NLINK | linux.STATX_UID | linux.STATX_GID | linux.STATX_INO + stat.Mask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_NLINK | + linux.STATX_UID | linux.STATX_GID | linux.STATX_INO | linux.STATX_ATIME | + linux.STATX_BTIME | linux.STATX_CTIME | linux.STATX_MTIME stat.Blksize = 1 // usermem.PageSize in tmpfs stat.Nlink = atomic.LoadUint32(&i.nlink) stat.UID = atomic.LoadUint32(&i.uid) stat.GID = atomic.LoadUint32(&i.gid) stat.Mode = uint16(atomic.LoadUint32(&i.mode)) stat.Ino = i.ino - // TODO: device number + // Linux's tmpfs has no concept of btime, so zero-value is returned. + stat.Atime = linux.NsecToStatxTimestamp(i.atime) + stat.Ctime = linux.NsecToStatxTimestamp(i.ctime) + stat.Mtime = linux.NsecToStatxTimestamp(i.mtime) + // TODO(gvisor.dev/issues/1197): Device number. switch impl := i.impl.(type) { case *regularFile: stat.Mode |= linux.S_IFREG @@ -245,6 +269,36 @@ func (i *inode) statTo(stat *linux.Statx) { } } +func (i *inode) setStat(stat linux.Statx) error { + // TODO(gvisor.dev/issues/1197): Handle stat.Size by growing/shrinking + // the file. + if stat.Mask == 0 { + return nil + } + i.mu.Lock() + mask := stat.Mask + if mask&linux.STATX_MODE != 0 { + atomic.StoreUint32(&i.mode, uint32(stat.Mode)) + } + if mask&linux.STATX_UID != 0 { + atomic.StoreUint32(&i.uid, stat.UID) + } + if mask&linux.STATX_GID != 0 { + atomic.StoreUint32(&i.gid, stat.GID) + } + if mask&linux.STATX_ATIME != 0 { + atomic.StoreInt64(&i.atime, stat.Atime.ToNsecCapped()) + } + if mask&linux.STATX_CTIME != 0 { + atomic.StoreInt64(&i.ctime, stat.Ctime.ToNsecCapped()) + } + if mask&linux.STATX_MTIME != 0 { + atomic.StoreInt64(&i.mtime, stat.Mtime.ToNsecCapped()) + } + i.mu.Unlock() + return nil +} + // allocatedBlocksForSize returns the number of 512B blocks needed to // accommodate the given size in bytes, as appropriate for struct // stat::st_blocks and struct statx::stx_blocks. (Note that this 512B block @@ -291,9 +345,5 @@ func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linu // SetStat implements vfs.FileDescriptionImpl.SetStat. func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { - if opts.Stat.Mask == 0 { - return nil - } - // TODO: implement inode.setStat - return syserror.EPERM + return fd.inode().setStat(opts.Stat) } -- cgit v1.2.3 From 420d335fc9495ec18a20f710869770d0708d9a49 Mon Sep 17 00:00:00 2001 From: Haibo Xu Date: Thu, 16 Jan 2020 10:26:23 -0800 Subject: Enable clone syscall support on arm64. sys_clone has many flavors in Linux, and amd64 chose a different one from x86(different arguments order). Ref kernel/fork.c for more info. Signed-off-by: Haibo Xu Change-Id: I6c8cbc685f4a6e786b171715ab68292fc95cbf48 COPYBARA_INTEGRATE_REVIEW=https://github.com/google/gvisor/pull/1545 from xiaobo55x:clone 156bd2dfbc63ef5291627b0578ddea77997393b2 PiperOrigin-RevId: 290093953 --- pkg/sentry/syscalls/linux/BUILD | 2 ++ pkg/sentry/syscalls/linux/sys_clone_amd64.go | 35 ++++++++++++++++++++++++++++ pkg/sentry/syscalls/linux/sys_clone_arm64.go | 35 ++++++++++++++++++++++++++++ pkg/sentry/syscalls/linux/sys_thread.go | 13 ----------- 4 files changed, 72 insertions(+), 13 deletions(-) create mode 100644 pkg/sentry/syscalls/linux/sys_clone_amd64.go create mode 100644 pkg/sentry/syscalls/linux/sys_clone_arm64.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD index aa05e208a..430d796ba 100644 --- a/pkg/sentry/syscalls/linux/BUILD +++ b/pkg/sentry/syscalls/linux/BUILD @@ -13,6 +13,8 @@ go_library( "sigset.go", "sys_aio.go", "sys_capability.go", + "sys_clone_amd64.go", + "sys_clone_arm64.go", "sys_epoll.go", "sys_eventfd.go", "sys_file.go", diff --git a/pkg/sentry/syscalls/linux/sys_clone_amd64.go b/pkg/sentry/syscalls/linux/sys_clone_amd64.go new file mode 100644 index 000000000..dd43cf18d --- /dev/null +++ b/pkg/sentry/syscalls/linux/sys_clone_amd64.go @@ -0,0 +1,35 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build amd64 + +package linux + +import ( + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" +) + +// Clone implements linux syscall clone(2). +// sys_clone has so many flavors. We implement the default one in linux 3.11 +// x86_64: +// sys_clone(clone_flags, newsp, parent_tidptr, child_tidptr, tls_val) +func Clone(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + flags := int(args[0].Int()) + stack := args[1].Pointer() + parentTID := args[2].Pointer() + childTID := args[3].Pointer() + tls := args[4].Pointer() + return clone(t, flags, stack, parentTID, childTID, tls) +} diff --git a/pkg/sentry/syscalls/linux/sys_clone_arm64.go b/pkg/sentry/syscalls/linux/sys_clone_arm64.go new file mode 100644 index 000000000..cf68a8949 --- /dev/null +++ b/pkg/sentry/syscalls/linux/sys_clone_arm64.go @@ -0,0 +1,35 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package linux + +import ( + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" +) + +// Clone implements linux syscall clone(2). +// sys_clone has so many flavors, and we implement the default one in linux 3.11 +// arm64(kernel/fork.c with CONFIG_CLONE_BACKWARDS defined in the config file): +// sys_clone(clone_flags, newsp, parent_tidptr, tls_val, child_tidptr) +func Clone(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + flags := int(args[0].Int()) + stack := args[1].Pointer() + parentTID := args[2].Pointer() + tls := args[3].Pointer() + childTID := args[4].Pointer() + return clone(t, flags, stack, parentTID, childTID, tls) +} diff --git a/pkg/sentry/syscalls/linux/sys_thread.go b/pkg/sentry/syscalls/linux/sys_thread.go index 4115116ff..b47c3b5c4 100644 --- a/pkg/sentry/syscalls/linux/sys_thread.go +++ b/pkg/sentry/syscalls/linux/sys_thread.go @@ -220,19 +220,6 @@ func clone(t *kernel.Task, flags int, stack usermem.Addr, parentTID usermem.Addr return uintptr(ntid), ctrl, err } -// Clone implements linux syscall clone(2). -// sys_clone has so many flavors. We implement the default one in linux 3.11 -// x86_64: -// sys_clone(clone_flags, newsp, parent_tidptr, child_tidptr, tls_val) -func Clone(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - flags := int(args[0].Int()) - stack := args[1].Pointer() - parentTID := args[2].Pointer() - childTID := args[3].Pointer() - tls := args[4].Pointer() - return clone(t, flags, stack, parentTID, childTID, tls) -} - // Fork implements Linux syscall fork(2). func Fork(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { // "A call to fork() is equivalent to a call to clone(2) specifying flags -- cgit v1.2.3 From 7b7c31820b83abcfe43f7170eff1f7953f3f27e2 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Thu, 16 Jan 2020 12:28:44 -0800 Subject: Add remaining /proc/* and /proc/sys/* files Except for one under /proc/sys/net/ipv4/tcp_sack. /proc/pid/* is still incomplete. Updates #1195 PiperOrigin-RevId: 290120438 --- pkg/cpuid/cpuid.go | 44 ++-- pkg/sentry/fs/proc/cpuinfo.go | 8 +- pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 40 ++++ pkg/sentry/fsimpl/proc/BUILD | 11 +- pkg/sentry/fsimpl/proc/filesystem.go | 11 + pkg/sentry/fsimpl/proc/loadavg.go | 42 ---- pkg/sentry/fsimpl/proc/meminfo.go | 79 ------- pkg/sentry/fsimpl/proc/net.go | 338 ---------------------------- pkg/sentry/fsimpl/proc/net_test.go | 78 ------- pkg/sentry/fsimpl/proc/stat.go | 129 ----------- pkg/sentry/fsimpl/proc/sys.go | 51 ----- pkg/sentry/fsimpl/proc/task.go | 12 +- pkg/sentry/fsimpl/proc/tasks.go | 41 +++- pkg/sentry/fsimpl/proc/tasks_files.go | 245 ++++++++++++++++++++ pkg/sentry/fsimpl/proc/tasks_net.go | 337 +++++++++++++++++++++++++++ pkg/sentry/fsimpl/proc/tasks_sys.go | 143 ++++++++++++ pkg/sentry/fsimpl/proc/tasks_sys_test.go | 78 +++++++ pkg/sentry/fsimpl/proc/tasks_test.go | 3 + pkg/sentry/fsimpl/proc/version.go | 70 ------ 19 files changed, 922 insertions(+), 838 deletions(-) delete mode 100644 pkg/sentry/fsimpl/proc/loadavg.go delete mode 100644 pkg/sentry/fsimpl/proc/meminfo.go delete mode 100644 pkg/sentry/fsimpl/proc/net.go delete mode 100644 pkg/sentry/fsimpl/proc/net_test.go delete mode 100644 pkg/sentry/fsimpl/proc/stat.go delete mode 100644 pkg/sentry/fsimpl/proc/sys.go create mode 100644 pkg/sentry/fsimpl/proc/tasks_net.go create mode 100644 pkg/sentry/fsimpl/proc/tasks_sys.go create mode 100644 pkg/sentry/fsimpl/proc/tasks_sys_test.go delete mode 100644 pkg/sentry/fsimpl/proc/version.go (limited to 'pkg/sentry') diff --git a/pkg/cpuid/cpuid.go b/pkg/cpuid/cpuid.go index d37047368..cf50ee53f 100644 --- a/pkg/cpuid/cpuid.go +++ b/pkg/cpuid/cpuid.go @@ -657,30 +657,28 @@ func (fs *FeatureSet) FlagsString(cpuinfoOnly bool) string { return strings.Join(s, " ") } -// CPUInfo is to generate a section of one cpu in /proc/cpuinfo. This is a -// minimal /proc/cpuinfo, it is missing some fields like "microcode" that are +// WriteCPUInfoTo is to generate a section of one cpu in /proc/cpuinfo. This is +// a minimal /proc/cpuinfo, it is missing some fields like "microcode" that are // not always printed in Linux. The bogomips field is simply made up. -func (fs FeatureSet) CPUInfo(cpu uint) string { - var b bytes.Buffer - fmt.Fprintf(&b, "processor\t: %d\n", cpu) - fmt.Fprintf(&b, "vendor_id\t: %s\n", fs.VendorID) - fmt.Fprintf(&b, "cpu family\t: %d\n", ((fs.ExtendedFamily<<4)&0xff)|fs.Family) - fmt.Fprintf(&b, "model\t\t: %d\n", ((fs.ExtendedModel<<4)&0xff)|fs.Model) - fmt.Fprintf(&b, "model name\t: %s\n", "unknown") // Unknown for now. - fmt.Fprintf(&b, "stepping\t: %s\n", "unknown") // Unknown for now. - fmt.Fprintf(&b, "cpu MHz\t\t: %.3f\n", cpuFreqMHz) - fmt.Fprintln(&b, "fpu\t\t: yes") - fmt.Fprintln(&b, "fpu_exception\t: yes") - fmt.Fprintf(&b, "cpuid level\t: %d\n", uint32(xSaveInfo)) // Same as ax in vendorID. - fmt.Fprintln(&b, "wp\t\t: yes") - fmt.Fprintf(&b, "flags\t\t: %s\n", fs.FlagsString(true)) - fmt.Fprintf(&b, "bogomips\t: %.02f\n", cpuFreqMHz) // It's bogus anyway. - fmt.Fprintf(&b, "clflush size\t: %d\n", fs.CacheLine) - fmt.Fprintf(&b, "cache_alignment\t: %d\n", fs.CacheLine) - fmt.Fprintf(&b, "address sizes\t: %d bits physical, %d bits virtual\n", 46, 48) - fmt.Fprintln(&b, "power management:") // This is always here, but can be blank. - fmt.Fprintln(&b, "") // The /proc/cpuinfo file ends with an extra newline. - return b.String() +func (fs FeatureSet) WriteCPUInfoTo(cpu uint, b *bytes.Buffer) { + fmt.Fprintf(b, "processor\t: %d\n", cpu) + fmt.Fprintf(b, "vendor_id\t: %s\n", fs.VendorID) + fmt.Fprintf(b, "cpu family\t: %d\n", ((fs.ExtendedFamily<<4)&0xff)|fs.Family) + fmt.Fprintf(b, "model\t\t: %d\n", ((fs.ExtendedModel<<4)&0xff)|fs.Model) + fmt.Fprintf(b, "model name\t: %s\n", "unknown") // Unknown for now. + fmt.Fprintf(b, "stepping\t: %s\n", "unknown") // Unknown for now. + fmt.Fprintf(b, "cpu MHz\t\t: %.3f\n", cpuFreqMHz) + fmt.Fprintln(b, "fpu\t\t: yes") + fmt.Fprintln(b, "fpu_exception\t: yes") + fmt.Fprintf(b, "cpuid level\t: %d\n", uint32(xSaveInfo)) // Same as ax in vendorID. + fmt.Fprintln(b, "wp\t\t: yes") + fmt.Fprintf(b, "flags\t\t: %s\n", fs.FlagsString(true)) + fmt.Fprintf(b, "bogomips\t: %.02f\n", cpuFreqMHz) // It's bogus anyway. + fmt.Fprintf(b, "clflush size\t: %d\n", fs.CacheLine) + fmt.Fprintf(b, "cache_alignment\t: %d\n", fs.CacheLine) + fmt.Fprintf(b, "address sizes\t: %d bits physical, %d bits virtual\n", 46, 48) + fmt.Fprintln(b, "power management:") // This is always here, but can be blank. + fmt.Fprintln(b, "") // The /proc/cpuinfo file ends with an extra newline. } const ( diff --git a/pkg/sentry/fs/proc/cpuinfo.go b/pkg/sentry/fs/proc/cpuinfo.go index 3edf36780..6330337eb 100644 --- a/pkg/sentry/fs/proc/cpuinfo.go +++ b/pkg/sentry/fs/proc/cpuinfo.go @@ -15,6 +15,8 @@ package proc import ( + "bytes" + "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -27,9 +29,9 @@ func newCPUInfo(ctx context.Context, msrc *fs.MountSource) *fs.Inode { // Kernel is always initialized with a FeatureSet. panic("cpuinfo read with nil FeatureSet") } - contents := make([]byte, 0, 1024) + var buf bytes.Buffer for i, max := uint(0), k.ApplicationCores(); i < max; i++ { - contents = append(contents, []byte(features.CPUInfo(i))...) + features.WriteCPUInfoTo(i, &buf) } - return newStaticProcInode(ctx, msrc, contents) + return newStaticProcInode(ctx, msrc, buf.Bytes()) } diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 1d469a0db..6aff3d39a 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -510,3 +510,43 @@ type InodeSymlink struct { func (InodeSymlink) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { return nil, syserror.ELOOP } + +// StaticDirectory is a standard implementation of a directory with static +// contents. +// +// +stateify savable +type StaticDirectory struct { + InodeNotSymlink + InodeDirectoryNoNewChildren + InodeAttrs + InodeNoDynamicLookup + OrderedChildren +} + +var _ Inode = (*StaticDirectory)(nil) + +// NewStaticDir creates a new static directory and returns its dentry. +func NewStaticDir(creds *auth.Credentials, ino uint64, perm linux.FileMode, children map[string]*Dentry) *Dentry { + if perm&^linux.PermissionsMask != 0 { + panic(fmt.Sprintf("Only permission mask must be set: %x", perm&linux.PermissionsMask)) + } + + inode := &StaticDirectory{} + inode.InodeAttrs.Init(creds, ino, linux.ModeDirectory|perm) + + dentry := &Dentry{} + dentry.Init(inode) + + inode.OrderedChildren.Init(OrderedChildrenOptions{}) + links := inode.OrderedChildren.Populate(dentry, children) + inode.IncLinks(links) + + return dentry +} + +// Open implements kernfs.Inode. +func (s *StaticDirectory) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { + fd := &GenericDirectoryFD{} + fd.Init(rp.Mount(), vfsd, &s.OrderedChildren, flags) + return fd.VFSFileDescription(), nil +} diff --git a/pkg/sentry/fsimpl/proc/BUILD b/pkg/sentry/fsimpl/proc/BUILD index 1f44b3217..6cd18cec8 100644 --- a/pkg/sentry/fsimpl/proc/BUILD +++ b/pkg/sentry/fsimpl/proc/BUILD @@ -7,17 +7,13 @@ go_library( name = "proc", srcs = [ "filesystem.go", - "loadavg.go", - "meminfo.go", "mounts.go", - "net.go", - "stat.go", - "sys.go", "task.go", "task_files.go", "tasks.go", "tasks_files.go", - "version.go", + "tasks_net.go", + "tasks_sys.go", ], importpath = "gvisor.dev/gvisor/pkg/sentry/fsimpl/proc", deps = [ @@ -30,6 +26,7 @@ go_library( "//pkg/sentry/inet", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", + "//pkg/sentry/kernel/time", "//pkg/sentry/limits", "//pkg/sentry/mm", "//pkg/sentry/socket", @@ -47,7 +44,7 @@ go_test( size = "small", srcs = [ "boot_test.go", - "net_test.go", + "tasks_sys_test.go", "tasks_test.go", ], embed = [":proc"], diff --git a/pkg/sentry/fsimpl/proc/filesystem.go b/pkg/sentry/fsimpl/proc/filesystem.go index d09182c77..e9cb7895f 100644 --- a/pkg/sentry/fsimpl/proc/filesystem.go +++ b/pkg/sentry/fsimpl/proc/filesystem.go @@ -67,3 +67,14 @@ func newDentry(creds *auth.Credentials, ino uint64, perm linux.FileMode, inode d d.Init(inode) return d } + +type staticFile struct { + kernfs.DynamicBytesFile + vfs.StaticData +} + +var _ dynamicInode = (*staticFile)(nil) + +func newStaticFile(data string) *staticFile { + return &staticFile{StaticData: vfs.StaticData{Data: data}} +} diff --git a/pkg/sentry/fsimpl/proc/loadavg.go b/pkg/sentry/fsimpl/proc/loadavg.go deleted file mode 100644 index 5351d86e8..000000000 --- a/pkg/sentry/fsimpl/proc/loadavg.go +++ /dev/null @@ -1,42 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "bytes" - "fmt" - - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" -) - -// loadavgData backs /proc/loadavg. -// -// +stateify savable -type loadavgData struct { - kernfs.DynamicBytesFile -} - -var _ dynamicInode = (*loadavgData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *loadavgData) Generate(ctx context.Context, buf *bytes.Buffer) error { - // TODO(b/62345059): Include real data in fields. - // Column 1-3: CPU and IO utilization of the last 1, 5, and 10 minute periods. - // Column 4-5: currently running processes and the total number of processes. - // Column 6: the last process ID used. - fmt.Fprintf(buf, "%.2f %.2f %.2f %d/%d %d\n", 0.00, 0.00, 0.00, 0, 0, 0) - return nil -} diff --git a/pkg/sentry/fsimpl/proc/meminfo.go b/pkg/sentry/fsimpl/proc/meminfo.go deleted file mode 100644 index cbdd4f3fc..000000000 --- a/pkg/sentry/fsimpl/proc/meminfo.go +++ /dev/null @@ -1,79 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "bytes" - "fmt" - - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" -) - -// meminfoData implements vfs.DynamicBytesSource for /proc/meminfo. -// -// +stateify savable -type meminfoData struct { - kernfs.DynamicBytesFile - - // k is the owning Kernel. - k *kernel.Kernel -} - -var _ dynamicInode = (*meminfoData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *meminfoData) Generate(ctx context.Context, buf *bytes.Buffer) error { - mf := d.k.MemoryFile() - mf.UpdateUsage() - snapshot, totalUsage := usage.MemoryAccounting.Copy() - totalSize := usage.TotalMemory(mf.TotalSize(), totalUsage) - anon := snapshot.Anonymous + snapshot.Tmpfs - file := snapshot.PageCache + snapshot.Mapped - // We don't actually have active/inactive LRUs, so just make up numbers. - activeFile := (file / 2) &^ (usermem.PageSize - 1) - inactiveFile := file - activeFile - - fmt.Fprintf(buf, "MemTotal: %8d kB\n", totalSize/1024) - memFree := (totalSize - totalUsage) / 1024 - // We use MemFree as MemAvailable because we don't swap. - // TODO(rahat): When reclaim is implemented the value of MemAvailable - // should change. - fmt.Fprintf(buf, "MemFree: %8d kB\n", memFree) - fmt.Fprintf(buf, "MemAvailable: %8d kB\n", memFree) - fmt.Fprintf(buf, "Buffers: 0 kB\n") // memory usage by block devices - fmt.Fprintf(buf, "Cached: %8d kB\n", (file+snapshot.Tmpfs)/1024) - // Emulate a system with no swap, which disables inactivation of anon pages. - fmt.Fprintf(buf, "SwapCache: 0 kB\n") - fmt.Fprintf(buf, "Active: %8d kB\n", (anon+activeFile)/1024) - fmt.Fprintf(buf, "Inactive: %8d kB\n", inactiveFile/1024) - fmt.Fprintf(buf, "Active(anon): %8d kB\n", anon/1024) - fmt.Fprintf(buf, "Inactive(anon): 0 kB\n") - fmt.Fprintf(buf, "Active(file): %8d kB\n", activeFile/1024) - fmt.Fprintf(buf, "Inactive(file): %8d kB\n", inactiveFile/1024) - fmt.Fprintf(buf, "Unevictable: 0 kB\n") // TODO(b/31823263) - fmt.Fprintf(buf, "Mlocked: 0 kB\n") // TODO(b/31823263) - fmt.Fprintf(buf, "SwapTotal: 0 kB\n") - fmt.Fprintf(buf, "SwapFree: 0 kB\n") - fmt.Fprintf(buf, "Dirty: 0 kB\n") - fmt.Fprintf(buf, "Writeback: 0 kB\n") - fmt.Fprintf(buf, "AnonPages: %8d kB\n", anon/1024) - fmt.Fprintf(buf, "Mapped: %8d kB\n", file/1024) // doesn't count mapped tmpfs, which we don't know - fmt.Fprintf(buf, "Shmem: %8d kB\n", snapshot.Tmpfs/1024) - return nil -} diff --git a/pkg/sentry/fsimpl/proc/net.go b/pkg/sentry/fsimpl/proc/net.go deleted file mode 100644 index fd46eebf8..000000000 --- a/pkg/sentry/fsimpl/proc/net.go +++ /dev/null @@ -1,338 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "bytes" - "fmt" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/inet" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/socket" - "gvisor.dev/gvisor/pkg/sentry/socket/unix" - "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" - "gvisor.dev/gvisor/pkg/sentry/vfs" -) - -// ifinet6 implements vfs.DynamicBytesSource for /proc/net/if_inet6. -// -// +stateify savable -type ifinet6 struct { - s inet.Stack -} - -var _ vfs.DynamicBytesSource = (*ifinet6)(nil) - -func (n *ifinet6) contents() []string { - var lines []string - nics := n.s.Interfaces() - for id, naddrs := range n.s.InterfaceAddrs() { - nic, ok := nics[id] - if !ok { - // NIC was added after NICNames was called. We'll just - // ignore it. - continue - } - - for _, a := range naddrs { - // IPv6 only. - if a.Family != linux.AF_INET6 { - continue - } - - // Fields: - // IPv6 address displayed in 32 hexadecimal chars without colons - // Netlink device number (interface index) in hexadecimal (use nic id) - // Prefix length in hexadecimal - // Scope value (use 0) - // Interface flags - // Device name - lines = append(lines, fmt.Sprintf("%032x %02x %02x %02x %02x %8s\n", a.Addr, id, a.PrefixLen, 0, a.Flags, nic.Name)) - } - } - return lines -} - -// Generate implements vfs.DynamicBytesSource.Generate. -func (n *ifinet6) Generate(ctx context.Context, buf *bytes.Buffer) error { - for _, l := range n.contents() { - buf.WriteString(l) - } - return nil -} - -// netDev implements vfs.DynamicBytesSource for /proc/net/dev. -// -// +stateify savable -type netDev struct { - s inet.Stack -} - -var _ vfs.DynamicBytesSource = (*netDev)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (n *netDev) Generate(ctx context.Context, buf *bytes.Buffer) error { - interfaces := n.s.Interfaces() - buf.WriteString("Inter-| Receive | Transmit\n") - buf.WriteString(" face |bytes packets errs drop fifo frame compressed multicast|bytes packets errs drop fifo colls carrier compressed\n") - - for _, i := range interfaces { - // Implements the same format as - // net/core/net-procfs.c:dev_seq_printf_stats. - var stats inet.StatDev - if err := n.s.Statistics(&stats, i.Name); err != nil { - log.Warningf("Failed to retrieve interface statistics for %v: %v", i.Name, err) - continue - } - fmt.Fprintf( - buf, - "%6s: %7d %7d %4d %4d %4d %5d %10d %9d %8d %7d %4d %4d %4d %5d %7d %10d\n", - i.Name, - // Received - stats[0], // bytes - stats[1], // packets - stats[2], // errors - stats[3], // dropped - stats[4], // fifo - stats[5], // frame - stats[6], // compressed - stats[7], // multicast - // Transmitted - stats[8], // bytes - stats[9], // packets - stats[10], // errors - stats[11], // dropped - stats[12], // fifo - stats[13], // frame - stats[14], // compressed - stats[15], // multicast - ) - } - - return nil -} - -// netUnix implements vfs.DynamicBytesSource for /proc/net/unix. -// -// +stateify savable -type netUnix struct { - k *kernel.Kernel -} - -var _ vfs.DynamicBytesSource = (*netUnix)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (n *netUnix) Generate(ctx context.Context, buf *bytes.Buffer) error { - buf.WriteString("Num RefCount Protocol Flags Type St Inode Path\n") - for _, se := range n.k.ListSockets() { - s := se.Sock.Get() - if s == nil { - log.Debugf("Couldn't resolve weakref %v in socket table, racing with destruction?", se.Sock) - continue - } - sfile := s.(*fs.File) - if family, _, _ := sfile.FileOperations.(socket.Socket).Type(); family != linux.AF_UNIX { - s.DecRef() - // Not a unix socket. - continue - } - sops := sfile.FileOperations.(*unix.SocketOperations) - - addr, err := sops.Endpoint().GetLocalAddress() - if err != nil { - log.Warningf("Failed to retrieve socket name from %+v: %v", sfile, err) - addr.Addr = "" - } - - sockFlags := 0 - if ce, ok := sops.Endpoint().(transport.ConnectingEndpoint); ok { - if ce.Listening() { - // For unix domain sockets, linux reports a single flag - // value if the socket is listening, of __SO_ACCEPTCON. - sockFlags = linux.SO_ACCEPTCON - } - } - - // In the socket entry below, the value for the 'Num' field requires - // some consideration. Linux prints the address to the struct - // unix_sock representing a socket in the kernel, but may redact the - // value for unprivileged users depending on the kptr_restrict - // sysctl. - // - // One use for this field is to allow a privileged user to - // introspect into the kernel memory to determine information about - // a socket not available through procfs, such as the socket's peer. - // - // In gvisor, returning a pointer to our internal structures would - // be pointless, as it wouldn't match the memory layout for struct - // unix_sock, making introspection difficult. We could populate a - // struct unix_sock with the appropriate data, but even that - // requires consideration for which kernel version to emulate, as - // the definition of this struct changes over time. - // - // For now, we always redact this pointer. - fmt.Fprintf(buf, "%#016p: %08X %08X %08X %04X %02X %5d", - (*unix.SocketOperations)(nil), // Num, pointer to kernel socket struct. - sfile.ReadRefs()-1, // RefCount, don't count our own ref. - 0, // Protocol, always 0 for UDS. - sockFlags, // Flags. - sops.Endpoint().Type(), // Type. - sops.State(), // State. - sfile.InodeID(), // Inode. - ) - - // Path - if len(addr.Addr) != 0 { - if addr.Addr[0] == 0 { - // Abstract path. - fmt.Fprintf(buf, " @%s", string(addr.Addr[1:])) - } else { - fmt.Fprintf(buf, " %s", string(addr.Addr)) - } - } - fmt.Fprintf(buf, "\n") - - s.DecRef() - } - return nil -} - -// netTCP implements vfs.DynamicBytesSource for /proc/net/tcp. -// -// +stateify savable -type netTCP struct { - k *kernel.Kernel -} - -var _ vfs.DynamicBytesSource = (*netTCP)(nil) - -func (n *netTCP) Generate(ctx context.Context, buf *bytes.Buffer) error { - t := kernel.TaskFromContext(ctx) - buf.WriteString(" sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode \n") - for _, se := range n.k.ListSockets() { - s := se.Sock.Get() - if s == nil { - log.Debugf("Couldn't resolve weakref %+v in socket table, racing with destruction?", se.Sock) - continue - } - sfile := s.(*fs.File) - sops, ok := sfile.FileOperations.(socket.Socket) - if !ok { - panic(fmt.Sprintf("Found non-socket file in socket table: %+v", sfile)) - } - if family, stype, _ := sops.Type(); !(family == linux.AF_INET && stype == linux.SOCK_STREAM) { - s.DecRef() - // Not tcp4 sockets. - continue - } - - // Linux's documentation for the fields below can be found at - // https://www.kernel.org/doc/Documentation/networking/proc_net_tcp.txt. - // For Linux's implementation, see net/ipv4/tcp_ipv4.c:get_tcp4_sock(). - // Note that the header doesn't contain labels for all the fields. - - // Field: sl; entry number. - fmt.Fprintf(buf, "%4d: ", se.ID) - - portBuf := make([]byte, 2) - - // Field: local_adddress. - var localAddr linux.SockAddrInet - if local, _, err := sops.GetSockName(t); err == nil { - localAddr = *local.(*linux.SockAddrInet) - } - binary.LittleEndian.PutUint16(portBuf, localAddr.Port) - fmt.Fprintf(buf, "%08X:%04X ", - binary.LittleEndian.Uint32(localAddr.Addr[:]), - portBuf) - - // Field: rem_address. - var remoteAddr linux.SockAddrInet - if remote, _, err := sops.GetPeerName(t); err == nil { - remoteAddr = *remote.(*linux.SockAddrInet) - } - binary.LittleEndian.PutUint16(portBuf, remoteAddr.Port) - fmt.Fprintf(buf, "%08X:%04X ", - binary.LittleEndian.Uint32(remoteAddr.Addr[:]), - portBuf) - - // Field: state; socket state. - fmt.Fprintf(buf, "%02X ", sops.State()) - - // Field: tx_queue, rx_queue; number of packets in the transmit and - // receive queue. Unimplemented. - fmt.Fprintf(buf, "%08X:%08X ", 0, 0) - - // Field: tr, tm->when; timer active state and number of jiffies - // until timer expires. Unimplemented. - fmt.Fprintf(buf, "%02X:%08X ", 0, 0) - - // Field: retrnsmt; number of unrecovered RTO timeouts. - // Unimplemented. - fmt.Fprintf(buf, "%08X ", 0) - - // Field: uid. - uattr, err := sfile.Dirent.Inode.UnstableAttr(ctx) - if err != nil { - log.Warningf("Failed to retrieve unstable attr for socket file: %v", err) - fmt.Fprintf(buf, "%5d ", 0) - } else { - fmt.Fprintf(buf, "%5d ", uint32(uattr.Owner.UID.In(t.UserNamespace()).OrOverflow())) - } - - // Field: timeout; number of unanswered 0-window probes. - // Unimplemented. - fmt.Fprintf(buf, "%8d ", 0) - - // Field: inode. - fmt.Fprintf(buf, "%8d ", sfile.InodeID()) - - // Field: refcount. Don't count the ref we obtain while deferencing - // the weakref to this socket. - fmt.Fprintf(buf, "%d ", sfile.ReadRefs()-1) - - // Field: Socket struct address. Redacted due to the same reason as - // the 'Num' field in /proc/net/unix, see netUnix.ReadSeqFileData. - fmt.Fprintf(buf, "%#016p ", (*socket.Socket)(nil)) - - // Field: retransmit timeout. Unimplemented. - fmt.Fprintf(buf, "%d ", 0) - - // Field: predicted tick of soft clock (delayed ACK control data). - // Unimplemented. - fmt.Fprintf(buf, "%d ", 0) - - // Field: (ack.quick<<1)|ack.pingpong, Unimplemented. - fmt.Fprintf(buf, "%d ", 0) - - // Field: sending congestion window, Unimplemented. - fmt.Fprintf(buf, "%d ", 0) - - // Field: Slow start size threshold, -1 if threshold >= 0xFFFF. - // Unimplemented, report as large threshold. - fmt.Fprintf(buf, "%d", -1) - - fmt.Fprintf(buf, "\n") - - s.DecRef() - } - - return nil -} diff --git a/pkg/sentry/fsimpl/proc/net_test.go b/pkg/sentry/fsimpl/proc/net_test.go deleted file mode 100644 index 20a77a8ca..000000000 --- a/pkg/sentry/fsimpl/proc/net_test.go +++ /dev/null @@ -1,78 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "bytes" - "reflect" - "testing" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" - "gvisor.dev/gvisor/pkg/sentry/inet" -) - -func newIPv6TestStack() *inet.TestStack { - s := inet.NewTestStack() - s.SupportsIPv6Flag = true - return s -} - -func TestIfinet6NoAddresses(t *testing.T) { - n := &ifinet6{s: newIPv6TestStack()} - var buf bytes.Buffer - n.Generate(contexttest.Context(t), &buf) - if buf.Len() > 0 { - t.Errorf("n.Generate() generated = %v, want = %v", buf.Bytes(), []byte{}) - } -} - -func TestIfinet6(t *testing.T) { - s := newIPv6TestStack() - s.InterfacesMap[1] = inet.Interface{Name: "eth0"} - s.InterfaceAddrsMap[1] = []inet.InterfaceAddr{ - { - Family: linux.AF_INET6, - PrefixLen: 128, - Addr: []byte("\x00\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x0f"), - }, - } - s.InterfacesMap[2] = inet.Interface{Name: "eth1"} - s.InterfaceAddrsMap[2] = []inet.InterfaceAddr{ - { - Family: linux.AF_INET6, - PrefixLen: 128, - Addr: []byte("\x10\x11\x12\x13\x14\x15\x16\x17\x18\x19\x1a\x1b\x1c\x1d\x1e\x1f"), - }, - } - want := map[string]struct{}{ - "000102030405060708090a0b0c0d0e0f 01 80 00 00 eth0\n": {}, - "101112131415161718191a1b1c1d1e1f 02 80 00 00 eth1\n": {}, - } - - n := &ifinet6{s: s} - contents := n.contents() - if len(contents) != len(want) { - t.Errorf("Got len(n.contents()) = %d, want = %d", len(contents), len(want)) - } - got := map[string]struct{}{} - for _, l := range contents { - got[l] = struct{}{} - } - - if !reflect.DeepEqual(got, want) { - t.Errorf("Got n.contents() = %v, want = %v", got, want) - } -} diff --git a/pkg/sentry/fsimpl/proc/stat.go b/pkg/sentry/fsimpl/proc/stat.go deleted file mode 100644 index 50894a534..000000000 --- a/pkg/sentry/fsimpl/proc/stat.go +++ /dev/null @@ -1,129 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "bytes" - "fmt" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" -) - -// cpuStats contains the breakdown of CPU time for /proc/stat. -type cpuStats struct { - // user is time spent in userspace tasks with non-positive niceness. - user uint64 - - // nice is time spent in userspace tasks with positive niceness. - nice uint64 - - // system is time spent in non-interrupt kernel context. - system uint64 - - // idle is time spent idle. - idle uint64 - - // ioWait is time spent waiting for IO. - ioWait uint64 - - // irq is time spent in interrupt context. - irq uint64 - - // softirq is time spent in software interrupt context. - softirq uint64 - - // steal is involuntary wait time. - steal uint64 - - // guest is time spent in guests with non-positive niceness. - guest uint64 - - // guestNice is time spent in guests with positive niceness. - guestNice uint64 -} - -// String implements fmt.Stringer. -func (c cpuStats) String() string { - return fmt.Sprintf("%d %d %d %d %d %d %d %d %d %d", c.user, c.nice, c.system, c.idle, c.ioWait, c.irq, c.softirq, c.steal, c.guest, c.guestNice) -} - -// statData implements vfs.DynamicBytesSource for /proc/stat. -// -// +stateify savable -type statData struct { - kernfs.DynamicBytesFile - - // k is the owning Kernel. - k *kernel.Kernel -} - -var _ dynamicInode = (*statData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (s *statData) Generate(ctx context.Context, buf *bytes.Buffer) error { - // TODO(b/37226836): We currently export only zero CPU stats. We could - // at least provide some aggregate stats. - var cpu cpuStats - fmt.Fprintf(buf, "cpu %s\n", cpu) - - for c, max := uint(0), s.k.ApplicationCores(); c < max; c++ { - fmt.Fprintf(buf, "cpu%d %s\n", c, cpu) - } - - // The total number of interrupts is dependent on the CPUs and PCI - // devices on the system. See arch_probe_nr_irqs. - // - // Since we don't report real interrupt stats, just choose an arbitrary - // value from a representative VM. - const numInterrupts = 256 - - // The Kernel doesn't handle real interrupts, so report all zeroes. - // TODO(b/37226836): We could count page faults as #PF. - fmt.Fprintf(buf, "intr 0") // total - for i := 0; i < numInterrupts; i++ { - fmt.Fprintf(buf, " 0") - } - fmt.Fprintf(buf, "\n") - - // Total number of context switches. - // TODO(b/37226836): Count this. - fmt.Fprintf(buf, "ctxt 0\n") - - // CLOCK_REALTIME timestamp from boot, in seconds. - fmt.Fprintf(buf, "btime %d\n", s.k.Timekeeper().BootTime().Seconds()) - - // Total number of clones. - // TODO(b/37226836): Count this. - fmt.Fprintf(buf, "processes 0\n") - - // Number of runnable tasks. - // TODO(b/37226836): Count this. - fmt.Fprintf(buf, "procs_running 0\n") - - // Number of tasks waiting on IO. - // TODO(b/37226836): Count this. - fmt.Fprintf(buf, "procs_blocked 0\n") - - // Number of each softirq handled. - fmt.Fprintf(buf, "softirq 0") // total - for i := 0; i < linux.NumSoftIRQ; i++ { - fmt.Fprintf(buf, " 0") - } - fmt.Fprintf(buf, "\n") - return nil -} diff --git a/pkg/sentry/fsimpl/proc/sys.go b/pkg/sentry/fsimpl/proc/sys.go deleted file mode 100644 index b88256e12..000000000 --- a/pkg/sentry/fsimpl/proc/sys.go +++ /dev/null @@ -1,51 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "bytes" - "fmt" - - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/vfs" -) - -// mmapMinAddrData implements vfs.DynamicBytesSource for -// /proc/sys/vm/mmap_min_addr. -// -// +stateify savable -type mmapMinAddrData struct { - k *kernel.Kernel -} - -var _ vfs.DynamicBytesSource = (*mmapMinAddrData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *mmapMinAddrData) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "%d\n", d.k.Platform.MinUserAddress()) - return nil -} - -// +stateify savable -type overcommitMemory struct{} - -var _ vfs.DynamicBytesSource = (*overcommitMemory)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *overcommitMemory) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "0\n") - return nil -} diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index 11a64c777..5a384817f 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -50,15 +50,15 @@ func newTaskInode(inoGen InoGenerator, task *kernel.Task, pidns *kernel.PIDNames //"fd": newFdDir(t, msrc), //"fdinfo": newFdInfoDir(t, msrc), //"gid_map": newGIDMap(t, msrc), - "io": newTaskOwnedFile(task, inoGen.NextIno(), defaultPermission, newIO(task, isThreadGroup)), - "maps": newTaskOwnedFile(task, inoGen.NextIno(), defaultPermission, &mapsData{task: task}), + "io": newTaskOwnedFile(task, inoGen.NextIno(), 0400, newIO(task, isThreadGroup)), + "maps": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &mapsData{task: task}), //"mountinfo": seqfile.NewSeqFileInode(t, &mountInfoFile{t: t}, msrc), //"mounts": seqfile.NewSeqFileInode(t, &mountsFile{t: t}, msrc), //"ns": newNamespaceDir(t, msrc), - "smaps": newTaskOwnedFile(task, inoGen.NextIno(), defaultPermission, &smapsData{task: task}), - "stat": newTaskOwnedFile(task, inoGen.NextIno(), defaultPermission, &taskStatData{t: task, pidns: pidns, tgstats: isThreadGroup}), - "statm": newTaskOwnedFile(task, inoGen.NextIno(), defaultPermission, &statmData{t: task}), - "status": newTaskOwnedFile(task, inoGen.NextIno(), defaultPermission, &statusData{t: task, pidns: pidns}), + "smaps": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &smapsData{task: task}), + "stat": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &taskStatData{t: task, pidns: pidns, tgstats: isThreadGroup}), + "statm": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &statmData{t: task}), + "status": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &statusData{t: task, pidns: pidns}), //"uid_map": newUIDMap(t, msrc), } if isThreadGroup { diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index d8f92d52f..72315d25c 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -15,6 +15,7 @@ package proc import ( + "bytes" "sort" "strconv" @@ -28,9 +29,8 @@ import ( ) const ( - defaultPermission = 0444 - selfName = "self" - threadSelfName = "thread-self" + selfName = "self" + threadSelfName = "thread-self" ) // InoGenerator generates unique inode numbers for a given filesystem. @@ -61,15 +61,15 @@ var _ kernfs.Inode = (*tasksInode)(nil) func newTasksInode(inoGen InoGenerator, k *kernel.Kernel, pidns *kernel.PIDNamespace) (*tasksInode, *kernfs.Dentry) { root := auth.NewRootCredentials(pidns.UserNamespace()) contents := map[string]*kernfs.Dentry{ - //"cpuinfo": newCPUInfo(ctx, msrc), - //"filesystems": seqfile.NewSeqFileInode(ctx, &filesystemsData{}, msrc), - "loadavg": newDentry(root, inoGen.NextIno(), defaultPermission, &loadavgData{}), - "meminfo": newDentry(root, inoGen.NextIno(), defaultPermission, &meminfoData{k: k}), - "mounts": kernfs.NewStaticSymlink(root, inoGen.NextIno(), defaultPermission, "self/mounts"), - "stat": newDentry(root, inoGen.NextIno(), defaultPermission, &statData{k: k}), - //"uptime": newUptime(ctx, msrc), - //"version": newVersionData(root, inoGen.NextIno(), k), - "version": newDentry(root, inoGen.NextIno(), defaultPermission, &versionData{k: k}), + "cpuinfo": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(cpuInfoData(k))), + //"filesystems": newDentry(root, inoGen.NextIno(), 0444, &filesystemsData{}), + "loadavg": newDentry(root, inoGen.NextIno(), 0444, &loadavgData{}), + "sys": newSysDir(root, inoGen), + "meminfo": newDentry(root, inoGen.NextIno(), 0444, &meminfoData{}), + "mounts": kernfs.NewStaticSymlink(root, inoGen.NextIno(), 0777, "self/mounts"), + "stat": newDentry(root, inoGen.NextIno(), 0444, &statData{}), + "uptime": newDentry(root, inoGen.NextIno(), 0444, &uptimeData{}), + "version": newDentry(root, inoGen.NextIno(), 0444, &versionData{}), } inode := &tasksInode{ @@ -216,3 +216,20 @@ func (i *tasksInode) Stat(vsfs *vfs.Filesystem) linux.Statx { return stat } + +func cpuInfoData(k *kernel.Kernel) string { + features := k.FeatureSet() + if features == nil { + // Kernel is always initialized with a FeatureSet. + panic("cpuinfo read with nil FeatureSet") + } + var buf bytes.Buffer + for i, max := uint(0), k.ApplicationCores(); i < max; i++ { + features.WriteCPUInfoTo(i, &buf) + } + return buf.String() +} + +func shmData(v uint64) dynamicInode { + return newStaticFile(strconv.FormatUint(v, 10)) +} diff --git a/pkg/sentry/fsimpl/proc/tasks_files.go b/pkg/sentry/fsimpl/proc/tasks_files.go index 91f30a798..ad3760e39 100644 --- a/pkg/sentry/fsimpl/proc/tasks_files.go +++ b/pkg/sentry/fsimpl/proc/tasks_files.go @@ -15,6 +15,7 @@ package proc import ( + "bytes" "fmt" "strconv" @@ -23,6 +24,9 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/kernel/time" + "gvisor.dev/gvisor/pkg/sentry/usage" + "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" ) @@ -90,3 +94,244 @@ func (s *threadSelfSymlink) Readlink(ctx context.Context) (string, error) { } return fmt.Sprintf("%d/task/%d", tgid, tid), nil } + +// cpuStats contains the breakdown of CPU time for /proc/stat. +type cpuStats struct { + // user is time spent in userspace tasks with non-positive niceness. + user uint64 + + // nice is time spent in userspace tasks with positive niceness. + nice uint64 + + // system is time spent in non-interrupt kernel context. + system uint64 + + // idle is time spent idle. + idle uint64 + + // ioWait is time spent waiting for IO. + ioWait uint64 + + // irq is time spent in interrupt context. + irq uint64 + + // softirq is time spent in software interrupt context. + softirq uint64 + + // steal is involuntary wait time. + steal uint64 + + // guest is time spent in guests with non-positive niceness. + guest uint64 + + // guestNice is time spent in guests with positive niceness. + guestNice uint64 +} + +// String implements fmt.Stringer. +func (c cpuStats) String() string { + return fmt.Sprintf("%d %d %d %d %d %d %d %d %d %d", c.user, c.nice, c.system, c.idle, c.ioWait, c.irq, c.softirq, c.steal, c.guest, c.guestNice) +} + +// statData implements vfs.DynamicBytesSource for /proc/stat. +// +// +stateify savable +type statData struct { + kernfs.DynamicBytesFile + + // k is the owning Kernel. + k *kernel.Kernel +} + +var _ dynamicInode = (*statData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (s *statData) Generate(ctx context.Context, buf *bytes.Buffer) error { + // TODO(b/37226836): We currently export only zero CPU stats. We could + // at least provide some aggregate stats. + var cpu cpuStats + fmt.Fprintf(buf, "cpu %s\n", cpu) + + for c, max := uint(0), s.k.ApplicationCores(); c < max; c++ { + fmt.Fprintf(buf, "cpu%d %s\n", c, cpu) + } + + // The total number of interrupts is dependent on the CPUs and PCI + // devices on the system. See arch_probe_nr_irqs. + // + // Since we don't report real interrupt stats, just choose an arbitrary + // value from a representative VM. + const numInterrupts = 256 + + // The Kernel doesn't handle real interrupts, so report all zeroes. + // TODO(b/37226836): We could count page faults as #PF. + fmt.Fprintf(buf, "intr 0") // total + for i := 0; i < numInterrupts; i++ { + fmt.Fprintf(buf, " 0") + } + fmt.Fprintf(buf, "\n") + + // Total number of context switches. + // TODO(b/37226836): Count this. + fmt.Fprintf(buf, "ctxt 0\n") + + // CLOCK_REALTIME timestamp from boot, in seconds. + fmt.Fprintf(buf, "btime %d\n", s.k.Timekeeper().BootTime().Seconds()) + + // Total number of clones. + // TODO(b/37226836): Count this. + fmt.Fprintf(buf, "processes 0\n") + + // Number of runnable tasks. + // TODO(b/37226836): Count this. + fmt.Fprintf(buf, "procs_running 0\n") + + // Number of tasks waiting on IO. + // TODO(b/37226836): Count this. + fmt.Fprintf(buf, "procs_blocked 0\n") + + // Number of each softirq handled. + fmt.Fprintf(buf, "softirq 0") // total + for i := 0; i < linux.NumSoftIRQ; i++ { + fmt.Fprintf(buf, " 0") + } + fmt.Fprintf(buf, "\n") + return nil +} + +// loadavgData backs /proc/loadavg. +// +// +stateify savable +type loadavgData struct { + kernfs.DynamicBytesFile +} + +var _ dynamicInode = (*loadavgData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (d *loadavgData) Generate(ctx context.Context, buf *bytes.Buffer) error { + // TODO(b/62345059): Include real data in fields. + // Column 1-3: CPU and IO utilization of the last 1, 5, and 10 minute periods. + // Column 4-5: currently running processes and the total number of processes. + // Column 6: the last process ID used. + fmt.Fprintf(buf, "%.2f %.2f %.2f %d/%d %d\n", 0.00, 0.00, 0.00, 0, 0, 0) + return nil +} + +// meminfoData implements vfs.DynamicBytesSource for /proc/meminfo. +// +// +stateify savable +type meminfoData struct { + kernfs.DynamicBytesFile + + // k is the owning Kernel. + k *kernel.Kernel +} + +var _ dynamicInode = (*meminfoData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (d *meminfoData) Generate(ctx context.Context, buf *bytes.Buffer) error { + mf := d.k.MemoryFile() + mf.UpdateUsage() + snapshot, totalUsage := usage.MemoryAccounting.Copy() + totalSize := usage.TotalMemory(mf.TotalSize(), totalUsage) + anon := snapshot.Anonymous + snapshot.Tmpfs + file := snapshot.PageCache + snapshot.Mapped + // We don't actually have active/inactive LRUs, so just make up numbers. + activeFile := (file / 2) &^ (usermem.PageSize - 1) + inactiveFile := file - activeFile + + fmt.Fprintf(buf, "MemTotal: %8d kB\n", totalSize/1024) + memFree := (totalSize - totalUsage) / 1024 + // We use MemFree as MemAvailable because we don't swap. + // TODO(rahat): When reclaim is implemented the value of MemAvailable + // should change. + fmt.Fprintf(buf, "MemFree: %8d kB\n", memFree) + fmt.Fprintf(buf, "MemAvailable: %8d kB\n", memFree) + fmt.Fprintf(buf, "Buffers: 0 kB\n") // memory usage by block devices + fmt.Fprintf(buf, "Cached: %8d kB\n", (file+snapshot.Tmpfs)/1024) + // Emulate a system with no swap, which disables inactivation of anon pages. + fmt.Fprintf(buf, "SwapCache: 0 kB\n") + fmt.Fprintf(buf, "Active: %8d kB\n", (anon+activeFile)/1024) + fmt.Fprintf(buf, "Inactive: %8d kB\n", inactiveFile/1024) + fmt.Fprintf(buf, "Active(anon): %8d kB\n", anon/1024) + fmt.Fprintf(buf, "Inactive(anon): 0 kB\n") + fmt.Fprintf(buf, "Active(file): %8d kB\n", activeFile/1024) + fmt.Fprintf(buf, "Inactive(file): %8d kB\n", inactiveFile/1024) + fmt.Fprintf(buf, "Unevictable: 0 kB\n") // TODO(b/31823263) + fmt.Fprintf(buf, "Mlocked: 0 kB\n") // TODO(b/31823263) + fmt.Fprintf(buf, "SwapTotal: 0 kB\n") + fmt.Fprintf(buf, "SwapFree: 0 kB\n") + fmt.Fprintf(buf, "Dirty: 0 kB\n") + fmt.Fprintf(buf, "Writeback: 0 kB\n") + fmt.Fprintf(buf, "AnonPages: %8d kB\n", anon/1024) + fmt.Fprintf(buf, "Mapped: %8d kB\n", file/1024) // doesn't count mapped tmpfs, which we don't know + fmt.Fprintf(buf, "Shmem: %8d kB\n", snapshot.Tmpfs/1024) + return nil +} + +// uptimeData implements vfs.DynamicBytesSource for /proc/uptime. +// +// +stateify savable +type uptimeData struct { + kernfs.DynamicBytesFile +} + +var _ dynamicInode = (*uptimeData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (*uptimeData) Generate(ctx context.Context, buf *bytes.Buffer) error { + k := kernel.KernelFromContext(ctx) + now := time.NowFromContext(ctx) + + // Pretend that we've spent zero time sleeping (second number). + fmt.Fprintf(buf, "%.2f 0.00\n", now.Sub(k.Timekeeper().BootTime()).Seconds()) + return nil +} + +// versionData implements vfs.DynamicBytesSource for /proc/version. +// +// +stateify savable +type versionData struct { + kernfs.DynamicBytesFile + + // k is the owning Kernel. + k *kernel.Kernel +} + +var _ dynamicInode = (*versionData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (v *versionData) Generate(ctx context.Context, buf *bytes.Buffer) error { + init := v.k.GlobalInit() + if init == nil { + // Attempted to read before the init Task is created. This can + // only occur during startup, which should never need to read + // this file. + panic("Attempted to read version before initial Task is available") + } + + // /proc/version takes the form: + // + // "SYSNAME version RELEASE (COMPILE_USER@COMPILE_HOST) + // (COMPILER_VERSION) VERSION" + // + // where: + // - SYSNAME, RELEASE, and VERSION are the same as returned by + // sys_utsname + // - COMPILE_USER is the user that build the kernel + // - COMPILE_HOST is the hostname of the machine on which the kernel + // was built + // - COMPILER_VERSION is the version reported by the building compiler + // + // Since we don't really want to expose build information to + // applications, those fields are omitted. + // + // FIXME(mpratt): Using Version from the init task SyscallTable + // disregards the different version a task may have (e.g., in a uts + // namespace). + ver := init.Leader().SyscallTable().Version + fmt.Fprintf(buf, "%s version %s %s\n", ver.Sysname, ver.Release, ver.Version) + return nil +} diff --git a/pkg/sentry/fsimpl/proc/tasks_net.go b/pkg/sentry/fsimpl/proc/tasks_net.go new file mode 100644 index 000000000..06dc43c26 --- /dev/null +++ b/pkg/sentry/fsimpl/proc/tasks_net.go @@ -0,0 +1,337 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package proc + +import ( + "bytes" + "fmt" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/inet" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/socket" + "gvisor.dev/gvisor/pkg/sentry/socket/unix" + "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" + "gvisor.dev/gvisor/pkg/sentry/vfs" +) + +// ifinet6 implements vfs.DynamicBytesSource for /proc/net/if_inet6. +// +// +stateify savable +type ifinet6 struct { + s inet.Stack +} + +var _ vfs.DynamicBytesSource = (*ifinet6)(nil) + +func (n *ifinet6) contents() []string { + var lines []string + nics := n.s.Interfaces() + for id, naddrs := range n.s.InterfaceAddrs() { + nic, ok := nics[id] + if !ok { + // NIC was added after NICNames was called. We'll just ignore it. + continue + } + + for _, a := range naddrs { + // IPv6 only. + if a.Family != linux.AF_INET6 { + continue + } + + // Fields: + // IPv6 address displayed in 32 hexadecimal chars without colons + // Netlink device number (interface index) in hexadecimal (use nic id) + // Prefix length in hexadecimal + // Scope value (use 0) + // Interface flags + // Device name + lines = append(lines, fmt.Sprintf("%032x %02x %02x %02x %02x %8s\n", a.Addr, id, a.PrefixLen, 0, a.Flags, nic.Name)) + } + } + return lines +} + +// Generate implements vfs.DynamicBytesSource.Generate. +func (n *ifinet6) Generate(ctx context.Context, buf *bytes.Buffer) error { + for _, l := range n.contents() { + buf.WriteString(l) + } + return nil +} + +// netDev implements vfs.DynamicBytesSource for /proc/net/dev. +// +// +stateify savable +type netDev struct { + s inet.Stack +} + +var _ vfs.DynamicBytesSource = (*netDev)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (n *netDev) Generate(ctx context.Context, buf *bytes.Buffer) error { + interfaces := n.s.Interfaces() + buf.WriteString("Inter-| Receive | Transmit\n") + buf.WriteString(" face |bytes packets errs drop fifo frame compressed multicast|bytes packets errs drop fifo colls carrier compressed\n") + + for _, i := range interfaces { + // Implements the same format as + // net/core/net-procfs.c:dev_seq_printf_stats. + var stats inet.StatDev + if err := n.s.Statistics(&stats, i.Name); err != nil { + log.Warningf("Failed to retrieve interface statistics for %v: %v", i.Name, err) + continue + } + fmt.Fprintf( + buf, + "%6s: %7d %7d %4d %4d %4d %5d %10d %9d %8d %7d %4d %4d %4d %5d %7d %10d\n", + i.Name, + // Received + stats[0], // bytes + stats[1], // packets + stats[2], // errors + stats[3], // dropped + stats[4], // fifo + stats[5], // frame + stats[6], // compressed + stats[7], // multicast + // Transmitted + stats[8], // bytes + stats[9], // packets + stats[10], // errors + stats[11], // dropped + stats[12], // fifo + stats[13], // frame + stats[14], // compressed + stats[15], // multicast + ) + } + + return nil +} + +// netUnix implements vfs.DynamicBytesSource for /proc/net/unix. +// +// +stateify savable +type netUnix struct { + k *kernel.Kernel +} + +var _ vfs.DynamicBytesSource = (*netUnix)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (n *netUnix) Generate(ctx context.Context, buf *bytes.Buffer) error { + buf.WriteString("Num RefCount Protocol Flags Type St Inode Path\n") + for _, se := range n.k.ListSockets() { + s := se.Sock.Get() + if s == nil { + log.Debugf("Couldn't resolve weakref %v in socket table, racing with destruction?", se.Sock) + continue + } + sfile := s.(*fs.File) + if family, _, _ := sfile.FileOperations.(socket.Socket).Type(); family != linux.AF_UNIX { + s.DecRef() + // Not a unix socket. + continue + } + sops := sfile.FileOperations.(*unix.SocketOperations) + + addr, err := sops.Endpoint().GetLocalAddress() + if err != nil { + log.Warningf("Failed to retrieve socket name from %+v: %v", sfile, err) + addr.Addr = "" + } + + sockFlags := 0 + if ce, ok := sops.Endpoint().(transport.ConnectingEndpoint); ok { + if ce.Listening() { + // For unix domain sockets, linux reports a single flag + // value if the socket is listening, of __SO_ACCEPTCON. + sockFlags = linux.SO_ACCEPTCON + } + } + + // In the socket entry below, the value for the 'Num' field requires + // some consideration. Linux prints the address to the struct + // unix_sock representing a socket in the kernel, but may redact the + // value for unprivileged users depending on the kptr_restrict + // sysctl. + // + // One use for this field is to allow a privileged user to + // introspect into the kernel memory to determine information about + // a socket not available through procfs, such as the socket's peer. + // + // In gvisor, returning a pointer to our internal structures would + // be pointless, as it wouldn't match the memory layout for struct + // unix_sock, making introspection difficult. We could populate a + // struct unix_sock with the appropriate data, but even that + // requires consideration for which kernel version to emulate, as + // the definition of this struct changes over time. + // + // For now, we always redact this pointer. + fmt.Fprintf(buf, "%#016p: %08X %08X %08X %04X %02X %5d", + (*unix.SocketOperations)(nil), // Num, pointer to kernel socket struct. + sfile.ReadRefs()-1, // RefCount, don't count our own ref. + 0, // Protocol, always 0 for UDS. + sockFlags, // Flags. + sops.Endpoint().Type(), // Type. + sops.State(), // State. + sfile.InodeID(), // Inode. + ) + + // Path + if len(addr.Addr) != 0 { + if addr.Addr[0] == 0 { + // Abstract path. + fmt.Fprintf(buf, " @%s", string(addr.Addr[1:])) + } else { + fmt.Fprintf(buf, " %s", string(addr.Addr)) + } + } + fmt.Fprintf(buf, "\n") + + s.DecRef() + } + return nil +} + +// netTCP implements vfs.DynamicBytesSource for /proc/net/tcp. +// +// +stateify savable +type netTCP struct { + k *kernel.Kernel +} + +var _ vfs.DynamicBytesSource = (*netTCP)(nil) + +func (n *netTCP) Generate(ctx context.Context, buf *bytes.Buffer) error { + t := kernel.TaskFromContext(ctx) + buf.WriteString(" sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode \n") + for _, se := range n.k.ListSockets() { + s := se.Sock.Get() + if s == nil { + log.Debugf("Couldn't resolve weakref %+v in socket table, racing with destruction?", se.Sock) + continue + } + sfile := s.(*fs.File) + sops, ok := sfile.FileOperations.(socket.Socket) + if !ok { + panic(fmt.Sprintf("Found non-socket file in socket table: %+v", sfile)) + } + if family, stype, _ := sops.Type(); !(family == linux.AF_INET && stype == linux.SOCK_STREAM) { + s.DecRef() + // Not tcp4 sockets. + continue + } + + // Linux's documentation for the fields below can be found at + // https://www.kernel.org/doc/Documentation/networking/proc_net_tcp.txt. + // For Linux's implementation, see net/ipv4/tcp_ipv4.c:get_tcp4_sock(). + // Note that the header doesn't contain labels for all the fields. + + // Field: sl; entry number. + fmt.Fprintf(buf, "%4d: ", se.ID) + + portBuf := make([]byte, 2) + + // Field: local_adddress. + var localAddr linux.SockAddrInet + if local, _, err := sops.GetSockName(t); err == nil { + localAddr = *local.(*linux.SockAddrInet) + } + binary.LittleEndian.PutUint16(portBuf, localAddr.Port) + fmt.Fprintf(buf, "%08X:%04X ", + binary.LittleEndian.Uint32(localAddr.Addr[:]), + portBuf) + + // Field: rem_address. + var remoteAddr linux.SockAddrInet + if remote, _, err := sops.GetPeerName(t); err == nil { + remoteAddr = *remote.(*linux.SockAddrInet) + } + binary.LittleEndian.PutUint16(portBuf, remoteAddr.Port) + fmt.Fprintf(buf, "%08X:%04X ", + binary.LittleEndian.Uint32(remoteAddr.Addr[:]), + portBuf) + + // Field: state; socket state. + fmt.Fprintf(buf, "%02X ", sops.State()) + + // Field: tx_queue, rx_queue; number of packets in the transmit and + // receive queue. Unimplemented. + fmt.Fprintf(buf, "%08X:%08X ", 0, 0) + + // Field: tr, tm->when; timer active state and number of jiffies + // until timer expires. Unimplemented. + fmt.Fprintf(buf, "%02X:%08X ", 0, 0) + + // Field: retrnsmt; number of unrecovered RTO timeouts. + // Unimplemented. + fmt.Fprintf(buf, "%08X ", 0) + + // Field: uid. + uattr, err := sfile.Dirent.Inode.UnstableAttr(ctx) + if err != nil { + log.Warningf("Failed to retrieve unstable attr for socket file: %v", err) + fmt.Fprintf(buf, "%5d ", 0) + } else { + fmt.Fprintf(buf, "%5d ", uint32(uattr.Owner.UID.In(t.UserNamespace()).OrOverflow())) + } + + // Field: timeout; number of unanswered 0-window probes. + // Unimplemented. + fmt.Fprintf(buf, "%8d ", 0) + + // Field: inode. + fmt.Fprintf(buf, "%8d ", sfile.InodeID()) + + // Field: refcount. Don't count the ref we obtain while deferencing + // the weakref to this socket. + fmt.Fprintf(buf, "%d ", sfile.ReadRefs()-1) + + // Field: Socket struct address. Redacted due to the same reason as + // the 'Num' field in /proc/net/unix, see netUnix.ReadSeqFileData. + fmt.Fprintf(buf, "%#016p ", (*socket.Socket)(nil)) + + // Field: retransmit timeout. Unimplemented. + fmt.Fprintf(buf, "%d ", 0) + + // Field: predicted tick of soft clock (delayed ACK control data). + // Unimplemented. + fmt.Fprintf(buf, "%d ", 0) + + // Field: (ack.quick<<1)|ack.pingpong, Unimplemented. + fmt.Fprintf(buf, "%d ", 0) + + // Field: sending congestion window, Unimplemented. + fmt.Fprintf(buf, "%d ", 0) + + // Field: Slow start size threshold, -1 if threshold >= 0xFFFF. + // Unimplemented, report as large threshold. + fmt.Fprintf(buf, "%d", -1) + + fmt.Fprintf(buf, "\n") + + s.DecRef() + } + + return nil +} diff --git a/pkg/sentry/fsimpl/proc/tasks_sys.go b/pkg/sentry/fsimpl/proc/tasks_sys.go new file mode 100644 index 000000000..aabf2bf0c --- /dev/null +++ b/pkg/sentry/fsimpl/proc/tasks_sys.go @@ -0,0 +1,143 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package proc + +import ( + "bytes" + "fmt" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" +) + +// newSysDir returns the dentry corresponding to /proc/sys directory. +func newSysDir(root *auth.Credentials, inoGen InoGenerator) *kernfs.Dentry { + return kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ + "kernel": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ + "hostname": newDentry(root, inoGen.NextIno(), 0444, &hostnameData{}), + "shmall": newDentry(root, inoGen.NextIno(), 0444, shmData(linux.SHMALL)), + "shmmax": newDentry(root, inoGen.NextIno(), 0444, shmData(linux.SHMMAX)), + "shmmni": newDentry(root, inoGen.NextIno(), 0444, shmData(linux.SHMMNI)), + }), + "vm": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ + "mmap_min_addr": newDentry(root, inoGen.NextIno(), 0444, &mmapMinAddrData{}), + "overcommit_memory": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0\n")), + }), + "net": newSysNetDir(root, inoGen), + }) +} + +// newSysNetDir returns the dentry corresponding to /proc/sys/net directory. +func newSysNetDir(root *auth.Credentials, inoGen InoGenerator) *kernfs.Dentry { + return kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ + "net": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ + "ipv4": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ + // Add tcp_sack. + // TODO(gvisor.dev/issue/1195): tcp_sack allows write(2) + // "tcp_sack": newTCPSackInode(ctx, msrc, s), + + // The following files are simple stubs until they are implemented in + // netstack, most of these files are configuration related. We use the + // value closest to the actual netstack behavior or any empty file, all + // of these files will have mode 0444 (read-only for all users). + "ip_local_port_range": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("16000 65535")), + "ip_local_reserved_ports": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("")), + "ipfrag_time": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("30")), + "ip_nonlocal_bind": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), + "ip_no_pmtu_disc": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), + + // tcp_allowed_congestion_control tell the user what they are able to + // do as an unprivledged process so we leave it empty. + "tcp_allowed_congestion_control": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("")), + "tcp_available_congestion_control": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("reno")), + "tcp_congestion_control": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("reno")), + + // Many of the following stub files are features netstack doesn't + // support. The unsupported features return "0" to indicate they are + // disabled. + "tcp_base_mss": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1280")), + "tcp_dsack": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), + "tcp_early_retrans": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), + "tcp_fack": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), + "tcp_fastopen": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), + "tcp_fastopen_key": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("")), + "tcp_invalid_ratelimit": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), + "tcp_keepalive_intvl": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), + "tcp_keepalive_probes": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), + "tcp_keepalive_time": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("7200")), + "tcp_mtu_probing": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), + "tcp_no_metrics_save": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), + "tcp_probe_interval": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), + "tcp_probe_threshold": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), + "tcp_retries1": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("3")), + "tcp_retries2": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("15")), + "tcp_rfc1337": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), + "tcp_slow_start_after_idle": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), + "tcp_synack_retries": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("5")), + "tcp_syn_retries": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("3")), + "tcp_timestamps": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), + }), + "core": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ + "default_qdisc": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("pfifo_fast")), + "message_burst": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("10")), + "message_cost": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("5")), + "optmem_max": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), + "rmem_default": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), + "rmem_max": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), + "somaxconn": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("128")), + "wmem_default": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), + "wmem_max": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), + }), + }), + }) +} + +// mmapMinAddrData implements vfs.DynamicBytesSource for +// /proc/sys/vm/mmap_min_addr. +// +// +stateify savable +type mmapMinAddrData struct { + kernfs.DynamicBytesFile + + k *kernel.Kernel +} + +var _ dynamicInode = (*mmapMinAddrData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (d *mmapMinAddrData) Generate(ctx context.Context, buf *bytes.Buffer) error { + fmt.Fprintf(buf, "%d\n", d.k.Platform.MinUserAddress()) + return nil +} + +// hostnameData implements vfs.DynamicBytesSource for /proc/sys/kernel/hostname. +// +// +stateify savable +type hostnameData struct { + kernfs.DynamicBytesFile +} + +var _ dynamicInode = (*hostnameData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (*hostnameData) Generate(ctx context.Context, buf *bytes.Buffer) error { + utsns := kernel.UTSNamespaceFromContext(ctx) + buf.WriteString(utsns.HostName()) + buf.WriteString("\n") + return nil +} diff --git a/pkg/sentry/fsimpl/proc/tasks_sys_test.go b/pkg/sentry/fsimpl/proc/tasks_sys_test.go new file mode 100644 index 000000000..20a77a8ca --- /dev/null +++ b/pkg/sentry/fsimpl/proc/tasks_sys_test.go @@ -0,0 +1,78 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package proc + +import ( + "bytes" + "reflect" + "testing" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/sentry/inet" +) + +func newIPv6TestStack() *inet.TestStack { + s := inet.NewTestStack() + s.SupportsIPv6Flag = true + return s +} + +func TestIfinet6NoAddresses(t *testing.T) { + n := &ifinet6{s: newIPv6TestStack()} + var buf bytes.Buffer + n.Generate(contexttest.Context(t), &buf) + if buf.Len() > 0 { + t.Errorf("n.Generate() generated = %v, want = %v", buf.Bytes(), []byte{}) + } +} + +func TestIfinet6(t *testing.T) { + s := newIPv6TestStack() + s.InterfacesMap[1] = inet.Interface{Name: "eth0"} + s.InterfaceAddrsMap[1] = []inet.InterfaceAddr{ + { + Family: linux.AF_INET6, + PrefixLen: 128, + Addr: []byte("\x00\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x0f"), + }, + } + s.InterfacesMap[2] = inet.Interface{Name: "eth1"} + s.InterfaceAddrsMap[2] = []inet.InterfaceAddr{ + { + Family: linux.AF_INET6, + PrefixLen: 128, + Addr: []byte("\x10\x11\x12\x13\x14\x15\x16\x17\x18\x19\x1a\x1b\x1c\x1d\x1e\x1f"), + }, + } + want := map[string]struct{}{ + "000102030405060708090a0b0c0d0e0f 01 80 00 00 eth0\n": {}, + "101112131415161718191a1b1c1d1e1f 02 80 00 00 eth1\n": {}, + } + + n := &ifinet6{s: s} + contents := n.contents() + if len(contents) != len(want) { + t.Errorf("Got len(n.contents()) = %d, want = %d", len(contents), len(want)) + } + got := map[string]struct{}{} + for _, l := range contents { + got[l] = struct{}{} + } + + if !reflect.DeepEqual(got, want) { + t.Errorf("Got n.contents() = %v, want = %v", got, want) + } +} diff --git a/pkg/sentry/fsimpl/proc/tasks_test.go b/pkg/sentry/fsimpl/proc/tasks_test.go index ca8c87ec2..76eafe593 100644 --- a/pkg/sentry/fsimpl/proc/tasks_test.go +++ b/pkg/sentry/fsimpl/proc/tasks_test.go @@ -69,12 +69,15 @@ func checkDots(dirs []vfs.Dirent) ([]vfs.Dirent, error) { func checkTasksStaticFiles(gots []vfs.Dirent) ([]vfs.Dirent, error) { wants := map[string]vfs.Dirent{ + "cpuinfo": {Type: linux.DT_REG}, "loadavg": {Type: linux.DT_REG}, "meminfo": {Type: linux.DT_REG}, "mounts": {Type: linux.DT_LNK}, "self": selfLink, "stat": {Type: linux.DT_REG}, + "sys": {Type: linux.DT_DIR}, "thread-self": threadSelfLink, + "uptime": {Type: linux.DT_REG}, "version": {Type: linux.DT_REG}, } return checkFiles(gots, wants) diff --git a/pkg/sentry/fsimpl/proc/version.go b/pkg/sentry/fsimpl/proc/version.go deleted file mode 100644 index 367f2396b..000000000 --- a/pkg/sentry/fsimpl/proc/version.go +++ /dev/null @@ -1,70 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "bytes" - "fmt" - - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" -) - -// versionData implements vfs.DynamicBytesSource for /proc/version. -// -// +stateify savable -type versionData struct { - kernfs.DynamicBytesFile - - // k is the owning Kernel. - k *kernel.Kernel -} - -var _ dynamicInode = (*versionData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (v *versionData) Generate(ctx context.Context, buf *bytes.Buffer) error { - init := v.k.GlobalInit() - if init == nil { - // Attempted to read before the init Task is created. This can - // only occur during startup, which should never need to read - // this file. - panic("Attempted to read version before initial Task is available") - } - - // /proc/version takes the form: - // - // "SYSNAME version RELEASE (COMPILE_USER@COMPILE_HOST) - // (COMPILER_VERSION) VERSION" - // - // where: - // - SYSNAME, RELEASE, and VERSION are the same as returned by - // sys_utsname - // - COMPILE_USER is the user that build the kernel - // - COMPILE_HOST is the hostname of the machine on which the kernel - // was built - // - COMPILER_VERSION is the version reported by the building compiler - // - // Since we don't really want to expose build information to - // applications, those fields are omitted. - // - // FIXME(mpratt): Using Version from the init task SyscallTable - // disregards the different version a task may have (e.g., in a uts - // namespace). - ver := init.Leader().SyscallTable().Version - fmt.Fprintf(buf, "%s version %s %s\n", ver.Sysname, ver.Release, ver.Version) - return nil -} -- cgit v1.2.3 From 07f258497932e53f4651b80a086117ffda843fe3 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Thu, 16 Jan 2020 12:33:07 -0800 Subject: Plumb getting/setting xattrs through InodeOperations and 9p gofer interfaces. There was a very bare get/setxattr in the InodeOperations interface. Add context.Context to both, size to getxattr, and flags to setxattr. Note that extended attributes are passed around as strings in this implementation, so size is automatically encoded into the value. Size is added in getxattr so that implementations can return ERANGE if a value is larger than can fit in the user-allocated buffer. This prevents us from unnecessarily passing around an arbitrarily large xattr when the user buffer is actually too small. Don't use the existing xattrwalk and xattrcreate messages and define our own, mainly for the sake of simplicity. Extended attributes will be implemented in future commits. PiperOrigin-RevId: 290121300 --- pkg/p9/client_file.go | 23 +++++ pkg/p9/file.go | 16 ++++ pkg/p9/handlers.go | 29 +++++++ pkg/p9/messages.go | 129 +++++++++++++++++++++++++++++ pkg/p9/messages_test.go | 15 ++++ pkg/p9/p9.go | 4 + pkg/sentry/fs/copy_up.go | 9 +- pkg/sentry/fs/file_overlay.go | 2 +- pkg/sentry/fs/fsutil/inode.go | 41 +++++---- pkg/sentry/fs/gofer/context_file.go | 14 ++++ pkg/sentry/fs/gofer/inode.go | 18 +++- pkg/sentry/fs/inode.go | 24 +++--- pkg/sentry/fs/inode_operations.go | 25 ++++-- pkg/sentry/fs/inode_overlay.go | 30 +++---- pkg/sentry/fs/inode_overlay_test.go | 4 +- pkg/sentry/fs/tmpfs/tmpfs.go | 18 ++-- pkg/sentry/syscalls/linux/linux64_amd64.go | 4 +- pkg/sentry/syscalls/linux/linux64_arm64.go | 4 +- pkg/sentry/syscalls/linux/sys_xattr.go | 48 ++++++----- runsc/fsgofer/fsgofer.go | 10 +++ 20 files changed, 374 insertions(+), 93 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/p9/client_file.go b/pkg/p9/client_file.go index de9357389..04b584383 100644 --- a/pkg/p9/client_file.go +++ b/pkg/p9/client_file.go @@ -165,6 +165,29 @@ func (c *clientFile) SetAttr(valid SetAttrMask, attr SetAttr) error { return c.client.sendRecv(&Tsetattr{FID: c.fid, Valid: valid, SetAttr: attr}, &Rsetattr{}) } +// GetXattr implements File.GetXattr. +func (c *clientFile) GetXattr(name string, size uint64) (string, error) { + if atomic.LoadUint32(&c.closed) != 0 { + return "", syscall.EBADF + } + + rgetxattr := Rgetxattr{} + if err := c.client.sendRecv(&Tgetxattr{FID: c.fid, Name: name, Size: size}, &rgetxattr); err != nil { + return "", err + } + + return rgetxattr.Value, nil +} + +// SetXattr implements File.SetXattr. +func (c *clientFile) SetXattr(name, value string, flags uint32) error { + if atomic.LoadUint32(&c.closed) != 0 { + return syscall.EBADF + } + + return c.client.sendRecv(&Tsetxattr{FID: c.fid, Name: name, Value: value, Flags: flags}, &Rsetxattr{}) +} + // Allocate implements File.Allocate. func (c *clientFile) Allocate(mode AllocateMode, offset, length uint64) error { if atomic.LoadUint32(&c.closed) != 0 { diff --git a/pkg/p9/file.go b/pkg/p9/file.go index 96d1f2a8e..4607cfcdf 100644 --- a/pkg/p9/file.go +++ b/pkg/p9/file.go @@ -89,6 +89,22 @@ type File interface { // On the server, SetAttr has a write concurrency guarantee. SetAttr(valid SetAttrMask, attr SetAttr) error + // GetXattr returns extended attributes of this node. + // + // Size indicates the size of the buffer that has been allocated to hold the + // attribute value. If the value is larger than size, implementations may + // return ERANGE to indicate that the buffer is too small, but they are also + // free to ignore the hint entirely (i.e. the value returned may be larger + // than size). All size checking is done independently at the syscall layer. + // + // TODO(b/127675828): Determine concurrency guarantees once implemented. + GetXattr(name string, size uint64) (string, error) + + // SetXattr sets extended attributes on this node. + // + // TODO(b/127675828): Determine concurrency guarantees once implemented. + SetXattr(name, value string, flags uint32) error + // Allocate allows the caller to directly manipulate the allocated disk space // for the file. See fallocate(2) for more details. Allocate(mode AllocateMode, offset, length uint64) error diff --git a/pkg/p9/handlers.go b/pkg/p9/handlers.go index b9582c07f..7d6653a07 100644 --- a/pkg/p9/handlers.go +++ b/pkg/p9/handlers.go @@ -912,6 +912,35 @@ func (t *Txattrcreate) handle(cs *connState) message { return newErr(syscall.ENOSYS) } +// handle implements handler.handle. +func (t *Tgetxattr) handle(cs *connState) message { + ref, ok := cs.LookupFID(t.FID) + if !ok { + return newErr(syscall.EBADF) + } + defer ref.DecRef() + + val, err := ref.file.GetXattr(t.Name, t.Size) + if err != nil { + return newErr(err) + } + return &Rgetxattr{Value: val} +} + +// handle implements handler.handle. +func (t *Tsetxattr) handle(cs *connState) message { + ref, ok := cs.LookupFID(t.FID) + if !ok { + return newErr(syscall.EBADF) + } + defer ref.DecRef() + + if err := ref.file.SetXattr(t.Name, t.Value, t.Flags); err != nil { + return newErr(err) + } + return &Rsetxattr{} +} + // handle implements handler.handle. func (t *Treaddir) handle(cs *connState) message { ref, ok := cs.LookupFID(t.Directory) diff --git a/pkg/p9/messages.go b/pkg/p9/messages.go index ffdd7e8c6..ceb723d86 100644 --- a/pkg/p9/messages.go +++ b/pkg/p9/messages.go @@ -1611,6 +1611,131 @@ func (r *Rxattrcreate) String() string { return fmt.Sprintf("Rxattrcreate{}") } +// Tgetxattr is a getxattr request. +type Tgetxattr struct { + // FID refers to the file for which to get xattrs. + FID FID + + // Name is the xattr to get. + Name string + + // Size is the buffer size for the xattr to get. + Size uint64 +} + +// Decode implements encoder.Decode. +func (t *Tgetxattr) Decode(b *buffer) { + t.FID = b.ReadFID() + t.Name = b.ReadString() + t.Size = b.Read64() +} + +// Encode implements encoder.Encode. +func (t *Tgetxattr) Encode(b *buffer) { + b.WriteFID(t.FID) + b.WriteString(t.Name) + b.Write64(t.Size) +} + +// Type implements message.Type. +func (*Tgetxattr) Type() MsgType { + return MsgTgetxattr +} + +// String implements fmt.Stringer. +func (t *Tgetxattr) String() string { + return fmt.Sprintf("Tgetxattr{FID: %d, Name: %s, Size: %d}", t.FID, t.Name, t.Size) +} + +// Rgetxattr is a getxattr response. +type Rgetxattr struct { + // Value is the extended attribute value. + Value string +} + +// Decode implements encoder.Decode. +func (r *Rgetxattr) Decode(b *buffer) { + r.Value = b.ReadString() +} + +// Encode implements encoder.Encode. +func (r *Rgetxattr) Encode(b *buffer) { + b.WriteString(r.Value) +} + +// Type implements message.Type. +func (*Rgetxattr) Type() MsgType { + return MsgRgetxattr +} + +// String implements fmt.Stringer. +func (r *Rgetxattr) String() string { + return fmt.Sprintf("Rgetxattr{Value: %s}", r.Value) +} + +// Tsetxattr sets extended attributes. +type Tsetxattr struct { + // FID refers to the file on which to set xattrs. + FID FID + + // Name is the attribute name. + Name string + + // Value is the attribute value. + Value string + + // Linux setxattr(2) flags. + Flags uint32 +} + +// Decode implements encoder.Decode. +func (t *Tsetxattr) Decode(b *buffer) { + t.FID = b.ReadFID() + t.Name = b.ReadString() + t.Value = b.ReadString() + t.Flags = b.Read32() +} + +// Encode implements encoder.Encode. +func (t *Tsetxattr) Encode(b *buffer) { + b.WriteFID(t.FID) + b.WriteString(t.Name) + b.WriteString(t.Value) + b.Write32(t.Flags) +} + +// Type implements message.Type. +func (*Tsetxattr) Type() MsgType { + return MsgTsetxattr +} + +// String implements fmt.Stringer. +func (t *Tsetxattr) String() string { + return fmt.Sprintf("Tsetxattr{FID: %d, Name: %s, Value: %s, Flags: %d}", t.FID, t.Name, t.Value, t.Flags) +} + +// Rsetxattr is a setxattr response. +type Rsetxattr struct { +} + +// Decode implements encoder.Decode. +func (r *Rsetxattr) Decode(b *buffer) { +} + +// Encode implements encoder.Encode. +func (r *Rsetxattr) Encode(b *buffer) { +} + +// Type implements message.Type. +func (*Rsetxattr) Type() MsgType { + return MsgRsetxattr +} + +// String implements fmt.Stringer. +func (r *Rsetxattr) String() string { + return fmt.Sprintf("Rsetxattr{}") +} + // Treaddir is a readdir request. type Treaddir struct { // Directory is the directory FID to read. @@ -2363,6 +2488,10 @@ func init() { msgRegistry.register(MsgRxattrwalk, func() message { return &Rxattrwalk{} }) msgRegistry.register(MsgTxattrcreate, func() message { return &Txattrcreate{} }) msgRegistry.register(MsgRxattrcreate, func() message { return &Rxattrcreate{} }) + msgRegistry.register(MsgTgetxattr, func() message { return &Tgetxattr{} }) + msgRegistry.register(MsgRgetxattr, func() message { return &Rgetxattr{} }) + msgRegistry.register(MsgTsetxattr, func() message { return &Tsetxattr{} }) + msgRegistry.register(MsgRsetxattr, func() message { return &Rsetxattr{} }) msgRegistry.register(MsgTreaddir, func() message { return &Treaddir{} }) msgRegistry.register(MsgRreaddir, func() message { return &Rreaddir{} }) msgRegistry.register(MsgTfsync, func() message { return &Tfsync{} }) diff --git a/pkg/p9/messages_test.go b/pkg/p9/messages_test.go index 6ba6a1654..825c939da 100644 --- a/pkg/p9/messages_test.go +++ b/pkg/p9/messages_test.go @@ -194,6 +194,21 @@ func TestEncodeDecode(t *testing.T) { Flags: 3, }, &Rxattrcreate{}, + &Tgetxattr{ + FID: 1, + Name: "abc", + Size: 2, + }, + &Rgetxattr{ + Value: "xyz", + }, + &Tsetxattr{ + FID: 1, + Name: "abc", + Value: "xyz", + Flags: 2, + }, + &Rsetxattr{}, &Treaddir{ Directory: 1, Offset: 2, diff --git a/pkg/p9/p9.go b/pkg/p9/p9.go index d3090535a..5ab00d625 100644 --- a/pkg/p9/p9.go +++ b/pkg/p9/p9.go @@ -339,6 +339,10 @@ const ( MsgRxattrwalk = 31 MsgTxattrcreate = 32 MsgRxattrcreate = 33 + MsgTgetxattr = 34 + MsgRgetxattr = 35 + MsgTsetxattr = 36 + MsgRsetxattr = 37 MsgTreaddir = 40 MsgRreaddir = 41 MsgTfsync = 50 diff --git a/pkg/sentry/fs/copy_up.go b/pkg/sentry/fs/copy_up.go index 734177e90..e03e3e417 100644 --- a/pkg/sentry/fs/copy_up.go +++ b/pkg/sentry/fs/copy_up.go @@ -18,6 +18,7 @@ import ( "fmt" "io" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/memmap" @@ -395,12 +396,12 @@ func copyContentsLocked(ctx context.Context, upper *Inode, lower *Inode, size in // Size and permissions are set on upper when the file content is copied // and when the file is created respectively. func copyAttributesLocked(ctx context.Context, upper *Inode, lower *Inode) error { - // Extract attributes fro the lower filesystem. + // Extract attributes from the lower filesystem. lowerAttr, err := lower.UnstableAttr(ctx) if err != nil { return err } - lowerXattr, err := lower.Listxattr() + lowerXattr, err := lower.ListXattr(ctx) if err != nil && err != syserror.EOPNOTSUPP { return err } @@ -421,11 +422,11 @@ func copyAttributesLocked(ctx context.Context, upper *Inode, lower *Inode) error if isXattrOverlay(name) { continue } - value, err := lower.Getxattr(name) + value, err := lower.GetXattr(ctx, name, linux.XATTR_SIZE_MAX) if err != nil { return err } - if err := upper.InodeOperations.Setxattr(upper, name, value); err != nil { + if err := upper.InodeOperations.SetXattr(ctx, upper, name, value, 0 /* flags */); err != nil { return err } } diff --git a/pkg/sentry/fs/file_overlay.go b/pkg/sentry/fs/file_overlay.go index 8a633b1ba..8991207b4 100644 --- a/pkg/sentry/fs/file_overlay.go +++ b/pkg/sentry/fs/file_overlay.go @@ -475,7 +475,7 @@ func readdirEntries(ctx context.Context, o *overlayEntry) (*SortedDentryMap, err // Skip this name if it is a negative entry in the // upper or there exists a whiteout for it. if o.upper != nil { - if overlayHasWhiteout(o.upper, name) { + if overlayHasWhiteout(ctx, o.upper, name) { continue } } diff --git a/pkg/sentry/fs/fsutil/inode.go b/pkg/sentry/fs/fsutil/inode.go index adf5ec69c..df7b74855 100644 --- a/pkg/sentry/fs/fsutil/inode.go +++ b/pkg/sentry/fs/fsutil/inode.go @@ -15,6 +15,7 @@ package fsutil import ( + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" @@ -202,7 +203,7 @@ func (i *InodeSimpleAttributes) NotifyModificationAndStatusChange(ctx context.Co } // InodeSimpleExtendedAttributes implements -// fs.InodeOperations.{Get,Set,List}xattr. +// fs.InodeOperations.{Get,Set,List}Xattr. // // +stateify savable type InodeSimpleExtendedAttributes struct { @@ -211,8 +212,8 @@ type InodeSimpleExtendedAttributes struct { xattrs map[string]string } -// Getxattr implements fs.InodeOperations.Getxattr. -func (i *InodeSimpleExtendedAttributes) Getxattr(_ *fs.Inode, name string) (string, error) { +// GetXattr implements fs.InodeOperations.GetXattr. +func (i *InodeSimpleExtendedAttributes) GetXattr(_ context.Context, _ *fs.Inode, name string, _ uint64) (string, error) { i.mu.RLock() value, ok := i.xattrs[name] i.mu.RUnlock() @@ -222,19 +223,31 @@ func (i *InodeSimpleExtendedAttributes) Getxattr(_ *fs.Inode, name string) (stri return value, nil } -// Setxattr implements fs.InodeOperations.Setxattr. -func (i *InodeSimpleExtendedAttributes) Setxattr(_ *fs.Inode, name, value string) error { +// SetXattr implements fs.InodeOperations.SetXattr. +func (i *InodeSimpleExtendedAttributes) SetXattr(_ context.Context, _ *fs.Inode, name, value string, flags uint32) error { i.mu.Lock() + defer i.mu.Unlock() if i.xattrs == nil { + if flags&linux.XATTR_REPLACE != 0 { + return syserror.ENODATA + } i.xattrs = make(map[string]string) } + + _, ok := i.xattrs[name] + if ok && flags&linux.XATTR_CREATE != 0 { + return syserror.EEXIST + } + if !ok && flags&linux.XATTR_REPLACE != 0 { + return syserror.ENODATA + } + i.xattrs[name] = value - i.mu.Unlock() return nil } -// Listxattr implements fs.InodeOperations.Listxattr. -func (i *InodeSimpleExtendedAttributes) Listxattr(_ *fs.Inode) (map[string]struct{}, error) { +// ListXattr implements fs.InodeOperations.ListXattr. +func (i *InodeSimpleExtendedAttributes) ListXattr(context.Context, *fs.Inode) (map[string]struct{}, error) { i.mu.RLock() names := make(map[string]struct{}, len(i.xattrs)) for name := range i.xattrs { @@ -436,18 +449,18 @@ func (InodeNotSymlink) Getlink(context.Context, *fs.Inode) (*fs.Dirent, error) { // extended attributes. type InodeNoExtendedAttributes struct{} -// Getxattr implements fs.InodeOperations.Getxattr. -func (InodeNoExtendedAttributes) Getxattr(*fs.Inode, string) (string, error) { +// GetXattr implements fs.InodeOperations.GetXattr. +func (InodeNoExtendedAttributes) GetXattr(context.Context, *fs.Inode, string, uint64) (string, error) { return "", syserror.EOPNOTSUPP } -// Setxattr implements fs.InodeOperations.Setxattr. -func (InodeNoExtendedAttributes) Setxattr(*fs.Inode, string, string) error { +// SetXattr implements fs.InodeOperations.SetXattr. +func (InodeNoExtendedAttributes) SetXattr(context.Context, *fs.Inode, string, string, uint32) error { return syserror.EOPNOTSUPP } -// Listxattr implements fs.InodeOperations.Listxattr. -func (InodeNoExtendedAttributes) Listxattr(*fs.Inode) (map[string]struct{}, error) { +// ListXattr implements fs.InodeOperations.ListXattr. +func (InodeNoExtendedAttributes) ListXattr(context.Context, *fs.Inode) (map[string]struct{}, error) { return nil, syserror.EOPNOTSUPP } diff --git a/pkg/sentry/fs/gofer/context_file.go b/pkg/sentry/fs/gofer/context_file.go index 44b72582a..2125dafef 100644 --- a/pkg/sentry/fs/gofer/context_file.go +++ b/pkg/sentry/fs/gofer/context_file.go @@ -59,6 +59,20 @@ func (c *contextFile) setAttr(ctx context.Context, valid p9.SetAttrMask, attr p9 return err } +func (c *contextFile) getXattr(ctx context.Context, name string, size uint64) (string, error) { + ctx.UninterruptibleSleepStart(false) + val, err := c.file.GetXattr(name, size) + ctx.UninterruptibleSleepFinish(false) + return val, err +} + +func (c *contextFile) setXattr(ctx context.Context, name, value string, flags uint32) error { + ctx.UninterruptibleSleepStart(false) + err := c.file.SetXattr(name, value, flags) + ctx.UninterruptibleSleepFinish(false) + return err +} + func (c *contextFile) allocate(ctx context.Context, mode p9.AllocateMode, offset, length uint64) error { ctx.UninterruptibleSleepStart(false) err := c.file.Allocate(mode, offset, length) diff --git a/pkg/sentry/fs/gofer/inode.go b/pkg/sentry/fs/gofer/inode.go index 245fe2ef1..98d1a8a48 100644 --- a/pkg/sentry/fs/gofer/inode.go +++ b/pkg/sentry/fs/gofer/inode.go @@ -38,8 +38,7 @@ import ( // // +stateify savable type inodeOperations struct { - fsutil.InodeNotVirtual `state:"nosave"` - fsutil.InodeNoExtendedAttributes `state:"nosave"` + fsutil.InodeNotVirtual `state:"nosave"` // fileState implements fs.CachedFileObject. It exists // to break a circular load dependency between inodeOperations @@ -604,6 +603,21 @@ func (i *inodeOperations) Truncate(ctx context.Context, inode *fs.Inode, length return i.fileState.file.setAttr(ctx, p9.SetAttrMask{Size: true}, p9.SetAttr{Size: uint64(length)}) } +// GetXattr implements fs.InodeOperations.GetXattr. +func (i *inodeOperations) GetXattr(ctx context.Context, inode *fs.Inode, name string, size uint64) (string, error) { + return i.fileState.file.getXattr(ctx, name, size) +} + +// SetXattr implements fs.InodeOperations.SetXattr. +func (i *inodeOperations) SetXattr(ctx context.Context, inode *fs.Inode, name string, value string, flags uint32) error { + return i.fileState.file.setXattr(ctx, name, value, flags) +} + +// ListXattr implements fs.InodeOperations.ListXattr. +func (i *inodeOperations) ListXattr(context.Context, *fs.Inode) (map[string]struct{}, error) { + return nil, syscall.EOPNOTSUPP +} + // Allocate implements fs.InodeOperations.Allocate. func (i *inodeOperations) Allocate(ctx context.Context, inode *fs.Inode, offset, length int64) error { // This can only be called for files anyway. diff --git a/pkg/sentry/fs/inode.go b/pkg/sentry/fs/inode.go index 468043df0..ee9d301ef 100644 --- a/pkg/sentry/fs/inode.go +++ b/pkg/sentry/fs/inode.go @@ -261,28 +261,28 @@ func (i *Inode) UnstableAttr(ctx context.Context) (UnstableAttr, error) { return i.InodeOperations.UnstableAttr(ctx, i) } -// Getxattr calls i.InodeOperations.Getxattr with i as the Inode. -func (i *Inode) Getxattr(name string) (string, error) { +// GetXattr calls i.InodeOperations.GetXattr with i as the Inode. +func (i *Inode) GetXattr(ctx context.Context, name string, size uint64) (string, error) { if i.overlay != nil { - return overlayGetxattr(i.overlay, name) + return overlayGetXattr(ctx, i.overlay, name, size) } - return i.InodeOperations.Getxattr(i, name) + return i.InodeOperations.GetXattr(ctx, i, name, size) } -// Setxattr calls i.InodeOperations.Setxattr with i as the Inode. -func (i *Inode) Setxattr(name, value string) error { +// SetXattr calls i.InodeOperations.SetXattr with i as the Inode. +func (i *Inode) SetXattr(ctx context.Context, name, value string, flags uint32) error { if i.overlay != nil { - return overlaySetxattr(i.overlay, name, value) + return overlaySetxattr(ctx, i.overlay, name, value, flags) } - return i.InodeOperations.Setxattr(i, name, value) + return i.InodeOperations.SetXattr(ctx, i, name, value, flags) } -// Listxattr calls i.InodeOperations.Listxattr with i as the Inode. -func (i *Inode) Listxattr() (map[string]struct{}, error) { +// ListXattr calls i.InodeOperations.ListXattr with i as the Inode. +func (i *Inode) ListXattr(ctx context.Context) (map[string]struct{}, error) { if i.overlay != nil { - return overlayListxattr(i.overlay) + return overlayListXattr(ctx, i.overlay) } - return i.InodeOperations.Listxattr(i) + return i.InodeOperations.ListXattr(ctx, i) } // CheckPermission will check if the caller may access this file in the diff --git a/pkg/sentry/fs/inode_operations.go b/pkg/sentry/fs/inode_operations.go index 5cde9d215..13261cb81 100644 --- a/pkg/sentry/fs/inode_operations.go +++ b/pkg/sentry/fs/inode_operations.go @@ -170,20 +170,27 @@ type InodeOperations interface { // file system events. UnstableAttr(ctx context.Context, inode *Inode) (UnstableAttr, error) - // Getxattr retrieves the value of extended attribute name. Inodes that - // do not support extended attributes return EOPNOTSUPP. Inodes that - // support extended attributes but don't have a value at name return + // GetXattr retrieves the value of extended attribute specified by name. + // Inodes that do not support extended attributes return EOPNOTSUPP. Inodes + // that support extended attributes but don't have a value at name return // ENODATA. - Getxattr(inode *Inode, name string) (string, error) + // + // If this is called through the getxattr(2) syscall, size indicates the + // size of the buffer that the application has allocated to hold the + // attribute value. If the value is larger than size, implementations may + // return ERANGE to indicate that the buffer is too small, but they are also + // free to ignore the hint entirely (i.e. the value returned may be larger + // than size). All size checking is done independently at the syscall layer. + GetXattr(ctx context.Context, inode *Inode, name string, size uint64) (string, error) - // Setxattr sets the value of extended attribute name. Inodes that - // do not support extended attributes return EOPNOTSUPP. - Setxattr(inode *Inode, name, value string) error + // SetXattr sets the value of extended attribute specified by name. Inodes + // that do not support extended attributes return EOPNOTSUPP. + SetXattr(ctx context.Context, inode *Inode, name, value string, flags uint32) error - // Listxattr returns the set of all extended attributes names that + // ListXattr returns the set of all extended attributes names that // have values. Inodes that do not support extended attributes return // EOPNOTSUPP. - Listxattr(inode *Inode) (map[string]struct{}, error) + ListXattr(ctx context.Context, inode *Inode) (map[string]struct{}, error) // Check determines whether an Inode can be accessed with the // requested permission mask using the context (which gives access diff --git a/pkg/sentry/fs/inode_overlay.go b/pkg/sentry/fs/inode_overlay.go index 13d11e001..b90da20d0 100644 --- a/pkg/sentry/fs/inode_overlay.go +++ b/pkg/sentry/fs/inode_overlay.go @@ -25,13 +25,13 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) -func overlayHasWhiteout(parent *Inode, name string) bool { - s, err := parent.Getxattr(XattrOverlayWhiteout(name)) +func overlayHasWhiteout(ctx context.Context, parent *Inode, name string) bool { + s, err := parent.GetXattr(ctx, XattrOverlayWhiteout(name), 1) return err == nil && s == "y" } -func overlayCreateWhiteout(parent *Inode, name string) error { - return parent.InodeOperations.Setxattr(parent, XattrOverlayWhiteout(name), "y") +func overlayCreateWhiteout(ctx context.Context, parent *Inode, name string) error { + return parent.InodeOperations.SetXattr(ctx, parent, XattrOverlayWhiteout(name), "y", 0 /* flags */) } func overlayWriteOut(ctx context.Context, o *overlayEntry) error { @@ -89,7 +89,7 @@ func overlayLookup(ctx context.Context, parent *overlayEntry, inode *Inode, name } // Are we done? - if overlayHasWhiteout(parent.upper, name) { + if overlayHasWhiteout(ctx, parent.upper, name) { if upperInode == nil { parent.copyMu.RUnlock() if negativeUpperChild { @@ -345,7 +345,7 @@ func overlayRemove(ctx context.Context, o *overlayEntry, parent *Dirent, child * } } if child.Inode.overlay.lowerExists { - if err := overlayCreateWhiteout(o.upper, child.name); err != nil { + if err := overlayCreateWhiteout(ctx, o.upper, child.name); err != nil { return err } } @@ -426,7 +426,7 @@ func overlayRename(ctx context.Context, o *overlayEntry, oldParent *Dirent, rena return err } if renamed.Inode.overlay.lowerExists { - if err := overlayCreateWhiteout(oldParent.Inode.overlay.upper, oldName); err != nil { + if err := overlayCreateWhiteout(ctx, oldParent.Inode.overlay.upper, oldName); err != nil { return err } } @@ -528,7 +528,7 @@ func overlayUnstableAttr(ctx context.Context, o *overlayEntry) (UnstableAttr, er return attr, err } -func overlayGetxattr(o *overlayEntry, name string) (string, error) { +func overlayGetXattr(ctx context.Context, o *overlayEntry, name string, size uint64) (string, error) { // Hot path. This is how the overlay checks for whiteout files. // Avoid defers. var ( @@ -544,31 +544,31 @@ func overlayGetxattr(o *overlayEntry, name string) (string, error) { o.copyMu.RLock() if o.upper != nil { - s, err = o.upper.Getxattr(name) + s, err = o.upper.GetXattr(ctx, name, size) } else { - s, err = o.lower.Getxattr(name) + s, err = o.lower.GetXattr(ctx, name, size) } o.copyMu.RUnlock() return s, err } // TODO(b/146028302): Support setxattr for overlayfs. -func overlaySetxattr(o *overlayEntry, name, value string) error { +func overlaySetxattr(ctx context.Context, o *overlayEntry, name, value string, flags uint32) error { return syserror.EOPNOTSUPP } -func overlayListxattr(o *overlayEntry) (map[string]struct{}, error) { +func overlayListXattr(ctx context.Context, o *overlayEntry) (map[string]struct{}, error) { o.copyMu.RLock() defer o.copyMu.RUnlock() var names map[string]struct{} var err error if o.upper != nil { - names, err = o.upper.Listxattr() + names, err = o.upper.ListXattr(ctx) } else { - names, err = o.lower.Listxattr() + names, err = o.lower.ListXattr(ctx) } for name := range names { - // Same as overlayGetxattr, we shouldn't forward along + // Same as overlayGetXattr, we shouldn't forward along // overlay attributes. if strings.HasPrefix(XattrOverlayPrefix, name) { delete(names, name) diff --git a/pkg/sentry/fs/inode_overlay_test.go b/pkg/sentry/fs/inode_overlay_test.go index 8935aad65..493d98c36 100644 --- a/pkg/sentry/fs/inode_overlay_test.go +++ b/pkg/sentry/fs/inode_overlay_test.go @@ -382,8 +382,8 @@ type dir struct { ReaddirCalled bool } -// Getxattr implements InodeOperations.Getxattr. -func (d *dir) Getxattr(inode *fs.Inode, name string) (string, error) { +// GetXattr implements InodeOperations.GetXattr. +func (d *dir) GetXattr(_ context.Context, _ *fs.Inode, name string, _ uint64) (string, error) { for _, n := range d.negative { if name == fs.XattrOverlayWhiteout(n) { return "y", nil diff --git a/pkg/sentry/fs/tmpfs/tmpfs.go b/pkg/sentry/fs/tmpfs/tmpfs.go index 69089c8a8..0f718e236 100644 --- a/pkg/sentry/fs/tmpfs/tmpfs.go +++ b/pkg/sentry/fs/tmpfs/tmpfs.go @@ -148,19 +148,19 @@ func (d *Dir) CreateFifo(ctx context.Context, dir *fs.Inode, name string, perms return d.ramfsDir.CreateFifo(ctx, dir, name, perms) } -// Getxattr implements fs.InodeOperations.Getxattr. -func (d *Dir) Getxattr(i *fs.Inode, name string) (string, error) { - return d.ramfsDir.Getxattr(i, name) +// GetXattr implements fs.InodeOperations.GetXattr. +func (d *Dir) GetXattr(ctx context.Context, i *fs.Inode, name string, size uint64) (string, error) { + return d.ramfsDir.GetXattr(ctx, i, name, size) } -// Setxattr implements fs.InodeOperations.Setxattr. -func (d *Dir) Setxattr(i *fs.Inode, name, value string) error { - return d.ramfsDir.Setxattr(i, name, value) +// SetXattr implements fs.InodeOperations.SetXattr. +func (d *Dir) SetXattr(ctx context.Context, i *fs.Inode, name, value string, flags uint32) error { + return d.ramfsDir.SetXattr(ctx, i, name, value, flags) } -// Listxattr implements fs.InodeOperations.Listxattr. -func (d *Dir) Listxattr(i *fs.Inode) (map[string]struct{}, error) { - return d.ramfsDir.Listxattr(i) +// ListXattr implements fs.InodeOperations.ListXattr. +func (d *Dir) ListXattr(ctx context.Context, i *fs.Inode) (map[string]struct{}, error) { + return d.ramfsDir.ListXattr(ctx, i) } // Lookup implements fs.InodeOperations.Lookup. diff --git a/pkg/sentry/syscalls/linux/linux64_amd64.go b/pkg/sentry/syscalls/linux/linux64_amd64.go index 479c5f6ff..6b2920900 100644 --- a/pkg/sentry/syscalls/linux/linux64_amd64.go +++ b/pkg/sentry/syscalls/linux/linux64_amd64.go @@ -228,10 +228,10 @@ var AMD64 = &kernel.SyscallTable{ 185: syscalls.Error("security", syserror.ENOSYS, "Not implemented in Linux.", nil), 186: syscalls.Supported("gettid", Gettid), 187: syscalls.Supported("readahead", Readahead), - 188: syscalls.PartiallySupported("setxattr", Setxattr, "Only supported for tmpfs.", nil), + 188: syscalls.PartiallySupported("setxattr", SetXattr, "Only supported for tmpfs.", nil), 189: syscalls.Error("lsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 190: syscalls.Error("fsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 191: syscalls.PartiallySupported("getxattr", Getxattr, "Only supported for tmpfs.", nil), + 191: syscalls.PartiallySupported("getxattr", GetXattr, "Only supported for tmpfs.", nil), 192: syscalls.ErrorWithEvent("lgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 193: syscalls.ErrorWithEvent("fgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 194: syscalls.ErrorWithEvent("listxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), diff --git a/pkg/sentry/syscalls/linux/linux64_arm64.go b/pkg/sentry/syscalls/linux/linux64_arm64.go index d3f61f5e8..8c1b20911 100644 --- a/pkg/sentry/syscalls/linux/linux64_arm64.go +++ b/pkg/sentry/syscalls/linux/linux64_arm64.go @@ -41,10 +41,10 @@ var ARM64 = &kernel.SyscallTable{ 2: syscalls.PartiallySupported("io_submit", IoSubmit, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), 3: syscalls.PartiallySupported("io_cancel", IoCancel, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), 4: syscalls.PartiallySupported("io_getevents", IoGetevents, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 5: syscalls.PartiallySupported("setxattr", Setxattr, "Only supported for tmpfs.", nil), + 5: syscalls.PartiallySupported("setxattr", SetXattr, "Only supported for tmpfs.", nil), 6: syscalls.Error("lsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 7: syscalls.Error("fsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 8: syscalls.PartiallySupported("getxattr", Getxattr, "Only supported for tmpfs.", nil), + 8: syscalls.PartiallySupported("getxattr", GetXattr, "Only supported for tmpfs.", nil), 9: syscalls.ErrorWithEvent("lgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 10: syscalls.ErrorWithEvent("fgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 11: syscalls.ErrorWithEvent("listxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), diff --git a/pkg/sentry/syscalls/linux/sys_xattr.go b/pkg/sentry/syscalls/linux/sys_xattr.go index 97d9a65ea..816352218 100644 --- a/pkg/sentry/syscalls/linux/sys_xattr.go +++ b/pkg/sentry/syscalls/linux/sys_xattr.go @@ -25,12 +25,12 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) -// Getxattr implements linux syscall getxattr(2). -func Getxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { +// GetXattr implements linux syscall getxattr(2). +func GetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { pathAddr := args[0].Pointer() nameAddr := args[1].Pointer() valueAddr := args[2].Pointer() - size := args[3].SizeT() + size := uint64(args[3].SizeT()) path, dirPath, err := copyInPath(t, pathAddr, false /* allowEmpty */) if err != nil { @@ -39,22 +39,28 @@ func Getxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc valueLen := 0 err = fileOpOn(t, linux.AT_FDCWD, path, true /* resolve */, func(root *fs.Dirent, d *fs.Dirent, _ uint) error { - value, err := getxattr(t, d, dirPath, nameAddr) + // If getxattr(2) is called with size 0, the size of the value will be + // returned successfully even if it is nonzero. In that case, we need to + // retrieve the entire attribute value so we can return the correct size. + requestedSize := size + if size == 0 || size > linux.XATTR_SIZE_MAX { + requestedSize = linux.XATTR_SIZE_MAX + } + + value, err := getXattr(t, d, dirPath, nameAddr, uint64(requestedSize)) if err != nil { return err } valueLen = len(value) - if size == 0 { - return nil - } - if size > linux.XATTR_SIZE_MAX { - size = linux.XATTR_SIZE_MAX - } - if valueLen > int(size) { + if uint64(valueLen) > requestedSize { return syserror.ERANGE } + // Skip copying out the attribute value if size is 0. + if size == 0 { + return nil + } _, err = t.CopyOutBytes(valueAddr, []byte(value)) return err }) @@ -64,8 +70,8 @@ func Getxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc return uintptr(valueLen), nil, nil } -// getxattr implements getxattr from the given *fs.Dirent. -func getxattr(t *kernel.Task, d *fs.Dirent, dirPath bool, nameAddr usermem.Addr) (string, error) { +// getXattr implements getxattr(2) from the given *fs.Dirent. +func getXattr(t *kernel.Task, d *fs.Dirent, dirPath bool, nameAddr usermem.Addr, size uint64) (string, error) { if dirPath && !fs.IsDir(d.Inode.StableAttr) { return "", syserror.ENOTDIR } @@ -83,15 +89,15 @@ func getxattr(t *kernel.Task, d *fs.Dirent, dirPath bool, nameAddr usermem.Addr) return "", syserror.EOPNOTSUPP } - return d.Inode.Getxattr(name) + return d.Inode.GetXattr(t, name, size) } -// Setxattr implements linux syscall setxattr(2). -func Setxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { +// SetXattr implements linux syscall setxattr(2). +func SetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { pathAddr := args[0].Pointer() nameAddr := args[1].Pointer() valueAddr := args[2].Pointer() - size := args[3].SizeT() + size := uint64(args[3].SizeT()) flags := args[4].Uint() path, dirPath, err := copyInPath(t, pathAddr, false /* allowEmpty */) @@ -104,12 +110,12 @@ func Setxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc } return 0, nil, fileOpOn(t, linux.AT_FDCWD, path, true /* resolve */, func(root *fs.Dirent, d *fs.Dirent, _ uint) error { - return setxattr(t, d, dirPath, nameAddr, valueAddr, size, flags) + return setXattr(t, d, dirPath, nameAddr, valueAddr, uint64(size), flags) }) } -// setxattr implements setxattr from the given *fs.Dirent. -func setxattr(t *kernel.Task, d *fs.Dirent, dirPath bool, nameAddr, valueAddr usermem.Addr, size uint, flags uint32) error { +// setXattr implements setxattr(2) from the given *fs.Dirent. +func setXattr(t *kernel.Task, d *fs.Dirent, dirPath bool, nameAddr, valueAddr usermem.Addr, size uint64, flags uint32) error { if dirPath && !fs.IsDir(d.Inode.StableAttr) { return syserror.ENOTDIR } @@ -136,7 +142,7 @@ func setxattr(t *kernel.Task, d *fs.Dirent, dirPath bool, nameAddr, valueAddr us return syserror.EOPNOTSUPP } - return d.Inode.Setxattr(name, value) + return d.Inode.SetXattr(t, name, value, flags) } func copyInXattrName(t *kernel.Task, nameAddr usermem.Addr) (string, error) { diff --git a/runsc/fsgofer/fsgofer.go b/runsc/fsgofer/fsgofer.go index 93606d051..4d84ad999 100644 --- a/runsc/fsgofer/fsgofer.go +++ b/runsc/fsgofer/fsgofer.go @@ -767,6 +767,16 @@ func (l *localFile) SetAttr(valid p9.SetAttrMask, attr p9.SetAttr) error { return err } +// TODO(b/127675828): support getxattr. +func (l *localFile) GetXattr(name string, size uint64) (string, error) { + return "", syscall.EOPNOTSUPP +} + +// TODO(b/127675828): support setxattr. +func (l *localFile) SetXattr(name, value string, flags uint32) error { + return syscall.EOPNOTSUPP +} + // Allocate implements p9.File. func (l *localFile) Allocate(mode p9.AllocateMode, offset, length uint64) error { if !l.isOpen() { -- cgit v1.2.3 From 3dd3275da7b665cf2ca297e4bf566fcc77025af8 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Thu, 16 Jan 2020 13:13:22 -0800 Subject: Add more files to /proc/[pid]/* Files not implemented require VFSv2 plumbing into the kernel. Also, cgroup is not implemented yet. Updates #1195 PiperOrigin-RevId: 290129176 --- pkg/sentry/fsimpl/kernfs/filesystem.go | 63 ++---- pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 16 +- pkg/sentry/fsimpl/kernfs/kernfs.go | 2 +- pkg/sentry/fsimpl/kernfs/symlink.go | 21 +- pkg/sentry/fsimpl/proc/BUILD | 3 +- pkg/sentry/fsimpl/proc/mounts.go | 33 --- pkg/sentry/fsimpl/proc/subtasks.go | 126 +++++++++++ pkg/sentry/fsimpl/proc/task.go | 69 ++++-- pkg/sentry/fsimpl/proc/task_files.go | 315 +++++++++++++++++++++++++--- pkg/sentry/fsimpl/proc/tasks.go | 2 +- pkg/sentry/fsimpl/proc/tasks_test.go | 20 +- pkg/sentry/vfs/permissions.go | 24 ++- 12 files changed, 549 insertions(+), 145 deletions(-) delete mode 100644 pkg/sentry/fsimpl/proc/mounts.go create mode 100644 pkg/sentry/fsimpl/proc/subtasks.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 79759e0fc..a4600ad47 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -22,7 +22,6 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" ) @@ -40,7 +39,7 @@ func (fs *Filesystem) stepExistingLocked(ctx context.Context, rp *vfs.ResolvingP return nil, syserror.ENOTDIR } // Directory searchable? - if err := d.inode.CheckPermissions(rp.Credentials(), vfs.MayExec); err != nil { + if err := d.inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayExec); err != nil { return nil, err } afterSymlink: @@ -182,8 +181,8 @@ func (fs *Filesystem) walkParentDirLocked(ctx context.Context, rp *vfs.Resolving // // Preconditions: Filesystem.mu must be locked for at least reading. parentInode // == parentVFSD.Impl().(*Dentry).Inode. isDir(parentInode) == true. -func checkCreateLocked(rp *vfs.ResolvingPath, parentVFSD *vfs.Dentry, parentInode Inode) (string, error) { - if err := parentInode.CheckPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { +func checkCreateLocked(ctx context.Context, rp *vfs.ResolvingPath, parentVFSD *vfs.Dentry, parentInode Inode) (string, error) { + if err := parentInode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { return "", err } pc := rp.Component() @@ -206,7 +205,7 @@ func checkCreateLocked(rp *vfs.ResolvingPath, parentVFSD *vfs.Dentry, parentInod // checkDeleteLocked checks that the file represented by vfsd may be deleted. // // Preconditions: Filesystem.mu must be locked for at least reading. -func checkDeleteLocked(rp *vfs.ResolvingPath, vfsd *vfs.Dentry) error { +func checkDeleteLocked(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry) error { parentVFSD := vfsd.Parent() if parentVFSD == nil { return syserror.EBUSY @@ -214,36 +213,12 @@ func checkDeleteLocked(rp *vfs.ResolvingPath, vfsd *vfs.Dentry) error { if parentVFSD.IsDisowned() { return syserror.ENOENT } - if err := parentVFSD.Impl().(*Dentry).inode.CheckPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { + if err := parentVFSD.Impl().(*Dentry).inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { return err } return nil } -// checkRenameLocked checks that a rename operation may be performed on the -// target dentry across the given set of parent directories. The target dentry -// may be nil. -// -// Precondition: isDir(dstInode) == true. -func checkRenameLocked(creds *auth.Credentials, src, dstDir *vfs.Dentry, dstInode Inode) error { - srcDir := src.Parent() - if srcDir == nil { - return syserror.EBUSY - } - if srcDir.IsDisowned() { - return syserror.ENOENT - } - if dstDir.IsDisowned() { - return syserror.ENOENT - } - // Check for creation permissions on dst dir. - if err := dstInode.CheckPermissions(creds, vfs.MayWrite|vfs.MayExec); err != nil { - return err - } - - return nil -} - // Release implements vfs.FilesystemImpl.Release. func (fs *Filesystem) Release() { } @@ -269,7 +244,7 @@ func (fs *Filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, op if !d.isDir() { return nil, syserror.ENOTDIR } - if err := inode.CheckPermissions(rp.Credentials(), vfs.MayExec); err != nil { + if err := inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayExec); err != nil { return nil, err } } @@ -302,7 +277,7 @@ func (fs *Filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. if err != nil { return err } - pc, err := checkCreateLocked(rp, parentVFSD, parentInode) + pc, err := checkCreateLocked(ctx, rp, parentVFSD, parentInode) if err != nil { return err } @@ -339,7 +314,7 @@ func (fs *Filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts v if err != nil { return err } - pc, err := checkCreateLocked(rp, parentVFSD, parentInode) + pc, err := checkCreateLocked(ctx, rp, parentVFSD, parentInode) if err != nil { return err } @@ -367,7 +342,7 @@ func (fs *Filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v if err != nil { return err } - pc, err := checkCreateLocked(rp, parentVFSD, parentInode) + pc, err := checkCreateLocked(ctx, rp, parentVFSD, parentInode) if err != nil { return err } @@ -401,7 +376,7 @@ func (fs *Filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf if err != nil { return nil, err } - if err := inode.CheckPermissions(rp.Credentials(), ats); err != nil { + if err := inode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { return nil, err } return inode.Open(rp, vfsd, opts.Flags) @@ -420,7 +395,7 @@ func (fs *Filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf if mustCreate { return nil, syserror.EEXIST } - if err := inode.CheckPermissions(rp.Credentials(), ats); err != nil { + if err := inode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { return nil, err } return inode.Open(rp, vfsd, opts.Flags) @@ -432,7 +407,7 @@ afterTrailingSymlink: return nil, err } // Check for search permission in the parent directory. - if err := parentInode.CheckPermissions(rp.Credentials(), vfs.MayExec); err != nil { + if err := parentInode.CheckPermissions(ctx, rp.Credentials(), vfs.MayExec); err != nil { return nil, err } // Reject attempts to open directories with O_CREAT. @@ -450,7 +425,7 @@ afterTrailingSymlink: } if childVFSD == nil { // Already checked for searchability above; now check for writability. - if err := parentInode.CheckPermissions(rp.Credentials(), vfs.MayWrite); err != nil { + if err := parentInode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite); err != nil { return nil, err } if err := rp.Mount().CheckBeginWrite(); err != nil { @@ -485,7 +460,7 @@ afterTrailingSymlink: goto afterTrailingSymlink } } - if err := childInode.CheckPermissions(rp.Credentials(), ats); err != nil { + if err := childInode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { return nil, err } return childInode.Open(rp, childVFSD, opts.Flags) @@ -545,13 +520,13 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa srcVFSD := &src.vfsd // Can we remove the src dentry? - if err := checkDeleteLocked(rp, srcVFSD); err != nil { + if err := checkDeleteLocked(ctx, rp, srcVFSD); err != nil { return err } // Can we create the dst dentry? var dstVFSD *vfs.Dentry - pc, err := checkCreateLocked(rp, dstDirVFSD, dstDirInode) + pc, err := checkCreateLocked(ctx, rp, dstDirVFSD, dstDirInode) switch err { case nil: // Ok, continue with rename as replacement. @@ -607,7 +582,7 @@ func (fs *Filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error return err } defer rp.Mount().EndWrite() - if err := checkDeleteLocked(rp, vfsd); err != nil { + if err := checkDeleteLocked(ctx, rp, vfsd); err != nil { return err } if !vfsd.Impl().(*Dentry).isDir() { @@ -683,7 +658,7 @@ func (fs *Filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, targ if err != nil { return err } - pc, err := checkCreateLocked(rp, parentVFSD, parentInode) + pc, err := checkCreateLocked(ctx, rp, parentVFSD, parentInode) if err != nil { return err } @@ -712,7 +687,7 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error return err } defer rp.Mount().EndWrite() - if err := checkDeleteLocked(rp, vfsd); err != nil { + if err := checkDeleteLocked(ctx, rp, vfsd); err != nil { return err } if vfsd.Impl().(*Dentry).isDir() { diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 6aff3d39a..1700fffd9 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -262,7 +262,7 @@ func (a *InodeAttrs) SetStat(_ *vfs.Filesystem, opts vfs.SetStatOptions) error { } // CheckPermissions implements Inode.CheckPermissions. -func (a *InodeAttrs) CheckPermissions(creds *auth.Credentials, ats vfs.AccessTypes) error { +func (a *InodeAttrs) CheckPermissions(_ context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error { mode := a.Mode() return vfs.GenericCheckPermissions( creds, @@ -527,12 +527,8 @@ var _ Inode = (*StaticDirectory)(nil) // NewStaticDir creates a new static directory and returns its dentry. func NewStaticDir(creds *auth.Credentials, ino uint64, perm linux.FileMode, children map[string]*Dentry) *Dentry { - if perm&^linux.PermissionsMask != 0 { - panic(fmt.Sprintf("Only permission mask must be set: %x", perm&linux.PermissionsMask)) - } - inode := &StaticDirectory{} - inode.InodeAttrs.Init(creds, ino, linux.ModeDirectory|perm) + inode.Init(creds, ino, perm) dentry := &Dentry{} dentry.Init(inode) @@ -544,6 +540,14 @@ func NewStaticDir(creds *auth.Credentials, ino uint64, perm linux.FileMode, chil return dentry } +// Init initializes StaticDirectory. +func (s *StaticDirectory) Init(creds *auth.Credentials, ino uint64, perm linux.FileMode) { + if perm&^linux.PermissionsMask != 0 { + panic(fmt.Sprintf("Only permission mask must be set: %x", perm&linux.PermissionsMask)) + } + s.InodeAttrs.Init(creds, ino, linux.ModeDirectory|perm) +} + // Open implements kernfs.Inode. func (s *StaticDirectory) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { fd := &GenericDirectoryFD{} diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index bb12f39a2..85bcdcc57 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -320,7 +320,7 @@ type inodeMetadata interface { // CheckPermissions checks that creds may access this inode for the // requested access type, per the the rules of // fs/namei.c:generic_permission(). - CheckPermissions(creds *auth.Credentials, atx vfs.AccessTypes) error + CheckPermissions(ctx context.Context, creds *auth.Credentials, atx vfs.AccessTypes) error // Mode returns the (struct stat)::st_mode value for this inode. This is // separated from Stat for performance. diff --git a/pkg/sentry/fsimpl/kernfs/symlink.go b/pkg/sentry/fsimpl/kernfs/symlink.go index 068063f4e..f19f12854 100644 --- a/pkg/sentry/fsimpl/kernfs/symlink.go +++ b/pkg/sentry/fsimpl/kernfs/symlink.go @@ -20,7 +20,9 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ) -type staticSymlink struct { +// StaticSymlink provides an Inode implementation for symlinks that point to +// a immutable target. +type StaticSymlink struct { InodeAttrs InodeNoopRefCount InodeSymlink @@ -28,18 +30,25 @@ type staticSymlink struct { target string } -var _ Inode = (*staticSymlink)(nil) +var _ Inode = (*StaticSymlink)(nil) // NewStaticSymlink creates a new symlink file pointing to 'target'. -func NewStaticSymlink(creds *auth.Credentials, ino uint64, perm linux.FileMode, target string) *Dentry { - inode := &staticSymlink{target: target} - inode.Init(creds, ino, linux.ModeSymlink|perm) +func NewStaticSymlink(creds *auth.Credentials, ino uint64, target string) *Dentry { + inode := &StaticSymlink{} + inode.Init(creds, ino, target) d := &Dentry{} d.Init(inode) return d } -func (s *staticSymlink) Readlink(_ context.Context) (string, error) { +// Init initializes the instance. +func (s *StaticSymlink) Init(creds *auth.Credentials, ino uint64, target string) { + s.target = target + s.InodeAttrs.Init(creds, ino, linux.ModeSymlink|0777) +} + +// Readlink implements Inode. +func (s *StaticSymlink) Readlink(_ context.Context) (string, error) { return s.target, nil } diff --git a/pkg/sentry/fsimpl/proc/BUILD b/pkg/sentry/fsimpl/proc/BUILD index 6cd18cec8..e92564b5d 100644 --- a/pkg/sentry/fsimpl/proc/BUILD +++ b/pkg/sentry/fsimpl/proc/BUILD @@ -7,7 +7,7 @@ go_library( name = "proc", srcs = [ "filesystem.go", - "mounts.go", + "subtasks.go", "task.go", "task_files.go", "tasks.go", @@ -29,6 +29,7 @@ go_library( "//pkg/sentry/kernel/time", "//pkg/sentry/limits", "//pkg/sentry/mm", + "//pkg/sentry/safemem", "//pkg/sentry/socket", "//pkg/sentry/socket/unix", "//pkg/sentry/socket/unix/transport", diff --git a/pkg/sentry/fsimpl/proc/mounts.go b/pkg/sentry/fsimpl/proc/mounts.go deleted file mode 100644 index 8683cf677..000000000 --- a/pkg/sentry/fsimpl/proc/mounts.go +++ /dev/null @@ -1,33 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import "gvisor.dev/gvisor/pkg/sentry/kernel" - -// TODO(gvisor.dev/issue/1195): Implement mountInfoFile and mountsFile. - -// mountInfoFile implements vfs.DynamicBytesSource for /proc/[pid]/mountinfo. -// -// +stateify savable -type mountInfoFile struct { - t *kernel.Task -} - -// mountsFile implements vfs.DynamicBytesSource for /proc/[pid]/mounts. -// -// +stateify savable -type mountsFile struct { - t *kernel.Task -} diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go new file mode 100644 index 000000000..8892c5a11 --- /dev/null +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -0,0 +1,126 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package proc + +import ( + "sort" + "strconv" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// subtasksInode represents the inode for /proc/[pid]/task/ directory. +// +// +stateify savable +type subtasksInode struct { + kernfs.InodeNotSymlink + kernfs.InodeDirectoryNoNewChildren + kernfs.InodeAttrs + kernfs.OrderedChildren + + task *kernel.Task + pidns *kernel.PIDNamespace + inoGen InoGenerator +} + +var _ kernfs.Inode = (*subtasksInode)(nil) + +func newSubtasks(task *kernel.Task, pidns *kernel.PIDNamespace, inoGen InoGenerator) *kernfs.Dentry { + subInode := &subtasksInode{ + task: task, + pidns: pidns, + inoGen: inoGen, + } + // Note: credentials are overridden by taskOwnedInode. + subInode.InodeAttrs.Init(task.Credentials(), inoGen.NextIno(), linux.ModeDirectory|0555) + subInode.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) + + inode := &taskOwnedInode{Inode: subInode, owner: task} + dentry := &kernfs.Dentry{} + dentry.Init(inode) + + return dentry +} + +// Valid implements kernfs.inodeDynamicLookup. +func (i *subtasksInode) Valid(ctx context.Context) bool { + return true +} + +// Lookup implements kernfs.inodeDynamicLookup. +func (i *subtasksInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { + tid, err := strconv.ParseUint(name, 10, 32) + if err != nil { + return nil, syserror.ENOENT + } + + subTask := i.pidns.TaskWithID(kernel.ThreadID(tid)) + if subTask == nil { + return nil, syserror.ENOENT + } + if subTask.ThreadGroup() != i.task.ThreadGroup() { + return nil, syserror.ENOENT + } + + subTaskDentry := newTaskInode(i.inoGen, subTask, i.pidns, false) + return subTaskDentry.VFSDentry(), nil +} + +// IterDirents implements kernfs.inodeDynamicLookup. +func (i *subtasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, offset, relOffset int64) (int64, error) { + tasks := i.task.ThreadGroup().MemberIDs(i.pidns) + if len(tasks) == 0 { + return offset, syserror.ENOENT + } + + tids := make([]int, 0, len(tasks)) + for _, tid := range tasks { + tids = append(tids, int(tid)) + } + + sort.Ints(tids) + for _, tid := range tids[relOffset:] { + dirent := vfs.Dirent{ + Name: strconv.FormatUint(uint64(tid), 10), + Type: linux.DT_DIR, + Ino: i.inoGen.NextIno(), + NextOff: offset + 1, + } + if !cb.Handle(dirent) { + return offset, nil + } + offset++ + } + return offset, nil +} + +// Open implements kernfs.Inode. +func (i *subtasksInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { + fd := &kernfs.GenericDirectoryFD{} + fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, flags) + return fd.VFSFileDescription(), nil +} + +// Stat implements kernfs.Inode. +func (i *subtasksInode) Stat(vsfs *vfs.Filesystem) linux.Statx { + stat := i.InodeAttrs.Stat(vsfs) + stat.Nlink += uint32(i.task.ThreadGroup().Count()) + return stat +} diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index 5a384817f..621c17cfe 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -15,6 +15,8 @@ package proc import ( + "fmt" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" @@ -42,27 +44,31 @@ var _ kernfs.Inode = (*taskInode)(nil) func newTaskInode(inoGen InoGenerator, task *kernel.Task, pidns *kernel.PIDNamespace, isThreadGroup bool) *kernfs.Dentry { contents := map[string]*kernfs.Dentry{ - //"auxv": newAuxvec(t, msrc), - //"cmdline": newExecArgInode(t, msrc, cmdlineExecArg), - //"comm": newComm(t, msrc), - //"environ": newExecArgInode(t, msrc, environExecArg), + "auxv": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &auxvData{task: task}), + "cmdline": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &cmdlineData{task: task, arg: cmdlineDataArg}), + "comm": newComm(task, inoGen.NextIno(), 0444), + "environ": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &cmdlineData{task: task, arg: environDataArg}), //"exe": newExe(t, msrc), //"fd": newFdDir(t, msrc), //"fdinfo": newFdInfoDir(t, msrc), - //"gid_map": newGIDMap(t, msrc), - "io": newTaskOwnedFile(task, inoGen.NextIno(), 0400, newIO(task, isThreadGroup)), - "maps": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &mapsData{task: task}), + "gid_map": newTaskOwnedFile(task, inoGen.NextIno(), 0644, &idMapData{task: task, gids: true}), + "io": newTaskOwnedFile(task, inoGen.NextIno(), 0400, newIO(task, isThreadGroup)), + "maps": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &mapsData{task: task}), //"mountinfo": seqfile.NewSeqFileInode(t, &mountInfoFile{t: t}, msrc), //"mounts": seqfile.NewSeqFileInode(t, &mountsFile{t: t}, msrc), - //"ns": newNamespaceDir(t, msrc), - "smaps": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &smapsData{task: task}), - "stat": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &taskStatData{t: task, pidns: pidns, tgstats: isThreadGroup}), - "statm": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &statmData{t: task}), - "status": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &statusData{t: task, pidns: pidns}), - //"uid_map": newUIDMap(t, msrc), + "ns": newTaskOwnedDir(task, inoGen.NextIno(), 0511, map[string]*kernfs.Dentry{ + "net": newNamespaceSymlink(task, inoGen.NextIno(), "net"), + "pid": newNamespaceSymlink(task, inoGen.NextIno(), "pid"), + "user": newNamespaceSymlink(task, inoGen.NextIno(), "user"), + }), + "smaps": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &smapsData{task: task}), + "stat": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &taskStatData{task: task, pidns: pidns, tgstats: isThreadGroup}), + "statm": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &statmData{task: task}), + "status": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &statusData{task: task, pidns: pidns}), + "uid_map": newTaskOwnedFile(task, inoGen.NextIno(), 0644, &idMapData{task: task, gids: false}), } if isThreadGroup { - //contents["task"] = p.newSubtasks(t, msrc) + contents["task"] = newSubtasks(task, pidns, inoGen) } //if len(p.cgroupControllers) > 0 { // contents["cgroup"] = newCGroupInode(t, msrc, p.cgroupControllers) @@ -127,6 +133,23 @@ func newTaskOwnedFile(task *kernel.Task, ino uint64, perm linux.FileMode, inode return d } +func newTaskOwnedDir(task *kernel.Task, ino uint64, perm linux.FileMode, children map[string]*kernfs.Dentry) *kernfs.Dentry { + dir := &kernfs.StaticDirectory{} + + // Note: credentials are overridden by taskOwnedInode. + dir.Init(task.Credentials(), ino, perm) + + inode := &taskOwnedInode{Inode: dir, owner: task} + d := &kernfs.Dentry{} + d.Init(inode) + + dir.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) + links := dir.OrderedChildren.Populate(d, children) + dir.IncLinks(links) + + return d +} + // Stat implements kernfs.Inode. func (i *taskOwnedInode) Stat(fs *vfs.Filesystem) linux.Statx { stat := i.Inode.Stat(fs) @@ -137,7 +160,7 @@ func (i *taskOwnedInode) Stat(fs *vfs.Filesystem) linux.Statx { } // CheckPermissions implements kernfs.Inode. -func (i *taskOwnedInode) CheckPermissions(creds *auth.Credentials, ats vfs.AccessTypes) error { +func (i *taskOwnedInode) CheckPermissions(_ context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error { mode := i.Mode() uid, gid := i.getOwner(mode) return vfs.GenericCheckPermissions( @@ -188,3 +211,19 @@ func newIO(t *kernel.Task, isThreadGroup bool) *ioData { } return &ioData{ioUsage: t} } + +func newNamespaceSymlink(task *kernel.Task, ino uint64, ns string) *kernfs.Dentry { + // Namespace symlinks should contain the namespace name and the inode number + // for the namespace instance, so for example user:[123456]. We currently fake + // the inode number by sticking the symlink inode in its place. + target := fmt.Sprintf("%s:[%d]", ns, ino) + + inode := &kernfs.StaticSymlink{} + // Note: credentials are overridden by taskOwnedInode. + inode.Init(task.Credentials(), ino, target) + + taskInode := &taskOwnedInode{Inode: inode, owner: task} + d := &kernfs.Dentry{} + d.Init(taskInode) + return d +} diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index 93f0e1aa8..7bc352ae9 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -17,15 +17,20 @@ package proc import ( "bytes" "fmt" + "io" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/mm" + "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" ) // mm gets the kernel task's MemoryManager. No additional reference is taken on @@ -41,6 +46,256 @@ func getMM(task *kernel.Task) *mm.MemoryManager { return tmm } +// getMMIncRef returns t's MemoryManager. If getMMIncRef succeeds, the +// MemoryManager's users count is incremented, and must be decremented by the +// caller when it is no longer in use. +func getMMIncRef(task *kernel.Task) (*mm.MemoryManager, error) { + if task.ExitState() == kernel.TaskExitDead { + return nil, syserror.ESRCH + } + var m *mm.MemoryManager + task.WithMuLocked(func(t *kernel.Task) { + m = t.MemoryManager() + }) + if m == nil || !m.IncUsers() { + return nil, io.EOF + } + return m, nil +} + +type bufferWriter struct { + buf *bytes.Buffer +} + +// WriteFromBlocks writes up to srcs.NumBytes() bytes from srcs and returns +// the number of bytes written. It may return a partial write without an +// error (i.e. (n, nil) where 0 < n < srcs.NumBytes()). It should not +// return a full write with an error (i.e. srcs.NumBytes(), err) where err +// != nil). +func (w *bufferWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) { + written := srcs.NumBytes() + for !srcs.IsEmpty() { + w.buf.Write(srcs.Head().ToSlice()) + srcs = srcs.Tail() + } + return written, nil +} + +// auxvData implements vfs.DynamicBytesSource for /proc/[pid]/auxv. +// +// +stateify savable +type auxvData struct { + kernfs.DynamicBytesFile + + task *kernel.Task +} + +var _ dynamicInode = (*auxvData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (d *auxvData) Generate(ctx context.Context, buf *bytes.Buffer) error { + m, err := getMMIncRef(d.task) + if err != nil { + return err + } + defer m.DecUsers(ctx) + + // Space for buffer with AT_NULL (0) terminator at the end. + auxv := m.Auxv() + buf.Grow((len(auxv) + 1) * 16) + for _, e := range auxv { + var tmp [8]byte + usermem.ByteOrder.PutUint64(tmp[:], e.Key) + buf.Write(tmp[:]) + + usermem.ByteOrder.PutUint64(tmp[:], uint64(e.Value)) + buf.Write(tmp[:]) + } + return nil +} + +// execArgType enumerates the types of exec arguments that are exposed through +// proc. +type execArgType int + +const ( + cmdlineDataArg execArgType = iota + environDataArg +) + +// cmdlineData implements vfs.DynamicBytesSource for /proc/[pid]/cmdline. +// +// +stateify savable +type cmdlineData struct { + kernfs.DynamicBytesFile + + task *kernel.Task + + // arg is the type of exec argument this file contains. + arg execArgType +} + +var _ dynamicInode = (*cmdlineData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (d *cmdlineData) Generate(ctx context.Context, buf *bytes.Buffer) error { + m, err := getMMIncRef(d.task) + if err != nil { + return err + } + defer m.DecUsers(ctx) + + // Figure out the bounds of the exec arg we are trying to read. + var ar usermem.AddrRange + switch d.arg { + case cmdlineDataArg: + ar = usermem.AddrRange{ + Start: m.ArgvStart(), + End: m.ArgvEnd(), + } + case environDataArg: + ar = usermem.AddrRange{ + Start: m.EnvvStart(), + End: m.EnvvEnd(), + } + default: + panic(fmt.Sprintf("unknown exec arg type %v", d.arg)) + } + if ar.Start == 0 || ar.End == 0 { + // Don't attempt to read before the start/end are set up. + return io.EOF + } + + // N.B. Technically this should be usermem.IOOpts.IgnorePermissions = true + // until Linux 4.9 (272ddc8b3735 "proc: don't use FOLL_FORCE for reading + // cmdline and environment"). + writer := &bufferWriter{buf: buf} + if n, err := m.CopyInTo(ctx, usermem.AddrRangeSeqOf(ar), writer, usermem.IOOpts{}); n == 0 || err != nil { + // Nothing to copy or something went wrong. + return err + } + + // On Linux, if the NULL byte at the end of the argument vector has been + // overwritten, it continues reading the environment vector as part of + // the argument vector. + if d.arg == cmdlineDataArg && buf.Bytes()[buf.Len()-1] != 0 { + if end := bytes.IndexByte(buf.Bytes(), 0); end != -1 { + // If we found a NULL character somewhere else in argv, truncate the + // return up to the NULL terminator (including it). + buf.Truncate(end) + return nil + } + + // There is no NULL terminator in the string, return into envp. + arEnvv := usermem.AddrRange{ + Start: m.EnvvStart(), + End: m.EnvvEnd(), + } + + // Upstream limits the returned amount to one page of slop. + // https://elixir.bootlin.com/linux/v4.20/source/fs/proc/base.c#L208 + // we'll return one page total between argv and envp because of the + // above page restrictions. + if buf.Len() >= usermem.PageSize { + // Returned at least one page already, nothing else to add. + return nil + } + remaining := usermem.PageSize - buf.Len() + if int(arEnvv.Length()) > remaining { + end, ok := arEnvv.Start.AddLength(uint64(remaining)) + if !ok { + return syserror.EFAULT + } + arEnvv.End = end + } + if _, err := m.CopyInTo(ctx, usermem.AddrRangeSeqOf(arEnvv), writer, usermem.IOOpts{}); err != nil { + return err + } + + // Linux will return envp up to and including the first NULL character, + // so find it. + if end := bytes.IndexByte(buf.Bytes()[ar.Length():], 0); end != -1 { + buf.Truncate(end) + } + } + + return nil +} + +// +stateify savable +type commInode struct { + kernfs.DynamicBytesFile + + task *kernel.Task +} + +func newComm(task *kernel.Task, ino uint64, perm linux.FileMode) *kernfs.Dentry { + inode := &commInode{task: task} + inode.DynamicBytesFile.Init(task.Credentials(), ino, &commData{task: task}, perm) + + d := &kernfs.Dentry{} + d.Init(inode) + return d +} + +func (i *commInode) CheckPermissions(ctx context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error { + // This file can always be read or written by members of the same thread + // group. See fs/proc/base.c:proc_tid_comm_permission. + // + // N.B. This check is currently a no-op as we don't yet support writing and + // this file is world-readable anyways. + t := kernel.TaskFromContext(ctx) + if t != nil && t.ThreadGroup() == i.task.ThreadGroup() && !ats.MayExec() { + return nil + } + + return i.DynamicBytesFile.CheckPermissions(ctx, creds, ats) +} + +// commData implements vfs.DynamicBytesSource for /proc/[pid]/comm. +// +// +stateify savable +type commData struct { + kernfs.DynamicBytesFile + + task *kernel.Task +} + +var _ dynamicInode = (*commData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (d *commData) Generate(ctx context.Context, buf *bytes.Buffer) error { + buf.WriteString(d.task.Name()) + buf.WriteString("\n") + return nil +} + +// idMapData implements vfs.DynamicBytesSource for /proc/[pid]/{gid_map|uid_map}. +// +// +stateify savable +type idMapData struct { + kernfs.DynamicBytesFile + + task *kernel.Task + gids bool +} + +var _ dynamicInode = (*idMapData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (d *idMapData) Generate(ctx context.Context, buf *bytes.Buffer) error { + var entries []auth.IDMapEntry + if d.gids { + entries = d.task.UserNamespace().GIDMap() + } else { + entries = d.task.UserNamespace().UIDMap() + } + for _, e := range entries { + fmt.Fprintf(buf, "%10d %10d %10d\n", e.FirstID, e.FirstParentID, e.Length) + } + return nil +} + // mapsData implements vfs.DynamicBytesSource for /proc/[pid]/maps. // // +stateify savable @@ -83,7 +338,7 @@ func (d *smapsData) Generate(ctx context.Context, buf *bytes.Buffer) error { type taskStatData struct { kernfs.DynamicBytesFile - t *kernel.Task + task *kernel.Task // If tgstats is true, accumulate fault stats (not implemented) and CPU // time across all tasks in t's thread group. @@ -98,40 +353,40 @@ var _ dynamicInode = (*taskStatData)(nil) // Generate implements vfs.DynamicBytesSource.Generate. func (s *taskStatData) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "%d ", s.pidns.IDOfTask(s.t)) - fmt.Fprintf(buf, "(%s) ", s.t.Name()) - fmt.Fprintf(buf, "%c ", s.t.StateStatus()[0]) + fmt.Fprintf(buf, "%d ", s.pidns.IDOfTask(s.task)) + fmt.Fprintf(buf, "(%s) ", s.task.Name()) + fmt.Fprintf(buf, "%c ", s.task.StateStatus()[0]) ppid := kernel.ThreadID(0) - if parent := s.t.Parent(); parent != nil { + if parent := s.task.Parent(); parent != nil { ppid = s.pidns.IDOfThreadGroup(parent.ThreadGroup()) } fmt.Fprintf(buf, "%d ", ppid) - fmt.Fprintf(buf, "%d ", s.pidns.IDOfProcessGroup(s.t.ThreadGroup().ProcessGroup())) - fmt.Fprintf(buf, "%d ", s.pidns.IDOfSession(s.t.ThreadGroup().Session())) + fmt.Fprintf(buf, "%d ", s.pidns.IDOfProcessGroup(s.task.ThreadGroup().ProcessGroup())) + fmt.Fprintf(buf, "%d ", s.pidns.IDOfSession(s.task.ThreadGroup().Session())) fmt.Fprintf(buf, "0 0 " /* tty_nr tpgid */) fmt.Fprintf(buf, "0 " /* flags */) fmt.Fprintf(buf, "0 0 0 0 " /* minflt cminflt majflt cmajflt */) var cputime usage.CPUStats if s.tgstats { - cputime = s.t.ThreadGroup().CPUStats() + cputime = s.task.ThreadGroup().CPUStats() } else { - cputime = s.t.CPUStats() + cputime = s.task.CPUStats() } fmt.Fprintf(buf, "%d %d ", linux.ClockTFromDuration(cputime.UserTime), linux.ClockTFromDuration(cputime.SysTime)) - cputime = s.t.ThreadGroup().JoinedChildCPUStats() + cputime = s.task.ThreadGroup().JoinedChildCPUStats() fmt.Fprintf(buf, "%d %d ", linux.ClockTFromDuration(cputime.UserTime), linux.ClockTFromDuration(cputime.SysTime)) - fmt.Fprintf(buf, "%d %d ", s.t.Priority(), s.t.Niceness()) - fmt.Fprintf(buf, "%d ", s.t.ThreadGroup().Count()) + fmt.Fprintf(buf, "%d %d ", s.task.Priority(), s.task.Niceness()) + fmt.Fprintf(buf, "%d ", s.task.ThreadGroup().Count()) // itrealvalue. Since kernel 2.6.17, this field is no longer // maintained, and is hard coded as 0. fmt.Fprintf(buf, "0 ") // Start time is relative to boot time, expressed in clock ticks. - fmt.Fprintf(buf, "%d ", linux.ClockTFromDuration(s.t.StartTime().Sub(s.t.Kernel().Timekeeper().BootTime()))) + fmt.Fprintf(buf, "%d ", linux.ClockTFromDuration(s.task.StartTime().Sub(s.task.Kernel().Timekeeper().BootTime()))) var vss, rss uint64 - s.t.WithMuLocked(func(t *kernel.Task) { + s.task.WithMuLocked(func(t *kernel.Task) { if mm := t.MemoryManager(); mm != nil { vss = mm.VirtualMemorySize() rss = mm.ResidentSetSize() @@ -140,14 +395,14 @@ func (s *taskStatData) Generate(ctx context.Context, buf *bytes.Buffer) error { fmt.Fprintf(buf, "%d %d ", vss, rss/usermem.PageSize) // rsslim. - fmt.Fprintf(buf, "%d ", s.t.ThreadGroup().Limits().Get(limits.Rss).Cur) + fmt.Fprintf(buf, "%d ", s.task.ThreadGroup().Limits().Get(limits.Rss).Cur) fmt.Fprintf(buf, "0 0 0 0 0 " /* startcode endcode startstack kstkesp kstkeip */) fmt.Fprintf(buf, "0 0 0 0 0 " /* signal blocked sigignore sigcatch wchan */) fmt.Fprintf(buf, "0 0 " /* nswap cnswap */) terminationSignal := linux.Signal(0) - if s.t == s.t.ThreadGroup().Leader() { - terminationSignal = s.t.ThreadGroup().TerminationSignal() + if s.task == s.task.ThreadGroup().Leader() { + terminationSignal = s.task.ThreadGroup().TerminationSignal() } fmt.Fprintf(buf, "%d ", terminationSignal) fmt.Fprintf(buf, "0 0 0 " /* processor rt_priority policy */) @@ -164,7 +419,7 @@ func (s *taskStatData) Generate(ctx context.Context, buf *bytes.Buffer) error { type statmData struct { kernfs.DynamicBytesFile - t *kernel.Task + task *kernel.Task } var _ dynamicInode = (*statmData)(nil) @@ -172,7 +427,7 @@ var _ dynamicInode = (*statmData)(nil) // Generate implements vfs.DynamicBytesSource.Generate. func (s *statmData) Generate(ctx context.Context, buf *bytes.Buffer) error { var vss, rss uint64 - s.t.WithMuLocked(func(t *kernel.Task) { + s.task.WithMuLocked(func(t *kernel.Task) { if mm := t.MemoryManager(); mm != nil { vss = mm.VirtualMemorySize() rss = mm.ResidentSetSize() @@ -189,7 +444,7 @@ func (s *statmData) Generate(ctx context.Context, buf *bytes.Buffer) error { type statusData struct { kernfs.DynamicBytesFile - t *kernel.Task + task *kernel.Task pidns *kernel.PIDNamespace } @@ -197,23 +452,23 @@ var _ dynamicInode = (*statusData)(nil) // Generate implements vfs.DynamicBytesSource.Generate. func (s *statusData) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "Name:\t%s\n", s.t.Name()) - fmt.Fprintf(buf, "State:\t%s\n", s.t.StateStatus()) - fmt.Fprintf(buf, "Tgid:\t%d\n", s.pidns.IDOfThreadGroup(s.t.ThreadGroup())) - fmt.Fprintf(buf, "Pid:\t%d\n", s.pidns.IDOfTask(s.t)) + fmt.Fprintf(buf, "Name:\t%s\n", s.task.Name()) + fmt.Fprintf(buf, "State:\t%s\n", s.task.StateStatus()) + fmt.Fprintf(buf, "Tgid:\t%d\n", s.pidns.IDOfThreadGroup(s.task.ThreadGroup())) + fmt.Fprintf(buf, "Pid:\t%d\n", s.pidns.IDOfTask(s.task)) ppid := kernel.ThreadID(0) - if parent := s.t.Parent(); parent != nil { + if parent := s.task.Parent(); parent != nil { ppid = s.pidns.IDOfThreadGroup(parent.ThreadGroup()) } fmt.Fprintf(buf, "PPid:\t%d\n", ppid) tpid := kernel.ThreadID(0) - if tracer := s.t.Tracer(); tracer != nil { + if tracer := s.task.Tracer(); tracer != nil { tpid = s.pidns.IDOfTask(tracer) } fmt.Fprintf(buf, "TracerPid:\t%d\n", tpid) var fds int var vss, rss, data uint64 - s.t.WithMuLocked(func(t *kernel.Task) { + s.task.WithMuLocked(func(t *kernel.Task) { if fdTable := t.FDTable(); fdTable != nil { fds = fdTable.Size() } @@ -227,13 +482,13 @@ func (s *statusData) Generate(ctx context.Context, buf *bytes.Buffer) error { fmt.Fprintf(buf, "VmSize:\t%d kB\n", vss>>10) fmt.Fprintf(buf, "VmRSS:\t%d kB\n", rss>>10) fmt.Fprintf(buf, "VmData:\t%d kB\n", data>>10) - fmt.Fprintf(buf, "Threads:\t%d\n", s.t.ThreadGroup().Count()) - creds := s.t.Credentials() + fmt.Fprintf(buf, "Threads:\t%d\n", s.task.ThreadGroup().Count()) + creds := s.task.Credentials() fmt.Fprintf(buf, "CapInh:\t%016x\n", creds.InheritableCaps) fmt.Fprintf(buf, "CapPrm:\t%016x\n", creds.PermittedCaps) fmt.Fprintf(buf, "CapEff:\t%016x\n", creds.EffectiveCaps) fmt.Fprintf(buf, "CapBnd:\t%016x\n", creds.BoundingCaps) - fmt.Fprintf(buf, "Seccomp:\t%d\n", s.t.SeccompMode()) + fmt.Fprintf(buf, "Seccomp:\t%d\n", s.task.SeccompMode()) // We unconditionally report a single NUMA node. See // pkg/sentry/syscalls/linux/sys_mempolicy.go. fmt.Fprintf(buf, "Mems_allowed:\t1\n") diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index 72315d25c..a97b1753a 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -66,7 +66,7 @@ func newTasksInode(inoGen InoGenerator, k *kernel.Kernel, pidns *kernel.PIDNames "loadavg": newDentry(root, inoGen.NextIno(), 0444, &loadavgData{}), "sys": newSysDir(root, inoGen), "meminfo": newDentry(root, inoGen.NextIno(), 0444, &meminfoData{}), - "mounts": kernfs.NewStaticSymlink(root, inoGen.NextIno(), 0777, "self/mounts"), + "mounts": kernfs.NewStaticSymlink(root, inoGen.NextIno(), "self/mounts"), "stat": newDentry(root, inoGen.NextIno(), 0444, &statData{}), "uptime": newDentry(root, inoGen.NextIno(), 0444, &uptimeData{}), "version": newDentry(root, inoGen.NextIno(), 0444, &versionData{}), diff --git a/pkg/sentry/fsimpl/proc/tasks_test.go b/pkg/sentry/fsimpl/proc/tasks_test.go index 76eafe593..6b58c16b9 100644 --- a/pkg/sentry/fsimpl/proc/tasks_test.go +++ b/pkg/sentry/fsimpl/proc/tasks_test.go @@ -85,12 +85,20 @@ func checkTasksStaticFiles(gots []vfs.Dirent) ([]vfs.Dirent, error) { func checkTaskStaticFiles(gots []vfs.Dirent) ([]vfs.Dirent, error) { wants := map[string]vfs.Dirent{ - "io": {Type: linux.DT_REG}, - "maps": {Type: linux.DT_REG}, - "smaps": {Type: linux.DT_REG}, - "stat": {Type: linux.DT_REG}, - "statm": {Type: linux.DT_REG}, - "status": {Type: linux.DT_REG}, + "auxv": {Type: linux.DT_REG}, + "cmdline": {Type: linux.DT_REG}, + "comm": {Type: linux.DT_REG}, + "environ": {Type: linux.DT_REG}, + "gid_map": {Type: linux.DT_REG}, + "io": {Type: linux.DT_REG}, + "maps": {Type: linux.DT_REG}, + "ns": {Type: linux.DT_DIR}, + "smaps": {Type: linux.DT_REG}, + "stat": {Type: linux.DT_REG}, + "statm": {Type: linux.DT_REG}, + "status": {Type: linux.DT_REG}, + "task": {Type: linux.DT_DIR}, + "uid_map": {Type: linux.DT_REG}, } return checkFiles(gots, wants) } diff --git a/pkg/sentry/vfs/permissions.go b/pkg/sentry/vfs/permissions.go index f1edb0680..d279d05ca 100644 --- a/pkg/sentry/vfs/permissions.go +++ b/pkg/sentry/vfs/permissions.go @@ -30,6 +30,26 @@ const ( MayExec = 1 ) +// OnlyRead returns true if access _only_ allows read. +func (a AccessTypes) OnlyRead() bool { + return a == MayRead +} + +// MayRead returns true if access allows read. +func (a AccessTypes) MayRead() bool { + return a&MayRead != 0 +} + +// MayWrite returns true if access allows write. +func (a AccessTypes) MayWrite() bool { + return a&MayWrite != 0 +} + +// MayExec returns true if access allows exec. +func (a AccessTypes) MayExec() bool { + return a&MayExec != 0 +} + // GenericCheckPermissions checks that creds has the given access rights on a // file with the given permissions, UID, and GID, subject to the rules of // fs/namei.c:generic_permission(). isDir is true if the file is a directory. @@ -53,7 +73,7 @@ func GenericCheckPermissions(creds *auth.Credentials, ats AccessTypes, isDir boo } // CAP_DAC_READ_SEARCH allows the caller to read and search arbitrary // directories, and read arbitrary non-directory files. - if (isDir && (ats&MayWrite == 0)) || ats == MayRead { + if (isDir && !ats.MayWrite()) || ats.OnlyRead() { if creds.HasCapability(linux.CAP_DAC_READ_SEARCH) { return nil } @@ -61,7 +81,7 @@ func GenericCheckPermissions(creds *auth.Credentials, ats AccessTypes, isDir boo // CAP_DAC_OVERRIDE allows arbitrary access to directories, read/write // access to non-directory files, and execute access to non-directory files // for which at least one execute bit is set. - if isDir || (ats&MayExec == 0) || (mode&0111 != 0) { + if isDir || !ats.MayExec() || (mode&0111 != 0) { if creds.HasCapability(linux.CAP_DAC_OVERRIDE) { return nil } -- cgit v1.2.3 From 70d7c52bd7583393d39177a7935cca57372d67f1 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Thu, 16 Jan 2020 13:58:25 -0800 Subject: Implement tmpfs.SetStat with a size argument. This is similar to 'Truncate' in vfs1. Updates https://github.com/google/gvisor/issues/1197 PiperOrigin-RevId: 290139140 --- pkg/sentry/fsimpl/tmpfs/regular_file.go | 35 ++++++++ pkg/sentry/fsimpl/tmpfs/regular_file_test.go | 121 +++++++++++++++++++++++++++ pkg/sentry/fsimpl/tmpfs/tmpfs.go | 54 ++++++++++-- 3 files changed, 205 insertions(+), 5 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index f200e767d..5fa70cc6d 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -63,6 +63,41 @@ func (fs *filesystem) newRegularFile(creds *auth.Credentials, mode linux.FileMod return &file.inode } +// truncate grows or shrinks the file to the given size. It returns true if the +// file size was updated. +func (rf *regularFile) truncate(size uint64) (bool, error) { + rf.mu.Lock() + defer rf.mu.Unlock() + + if size == rf.size { + // Nothing to do. + return false, nil + } + + if size > rf.size { + // Growing the file. + if rf.seals&linux.F_SEAL_GROW != 0 { + // Seal does not allow growth. + return false, syserror.EPERM + } + rf.size = size + return true, nil + } + + // Shrinking the file + if rf.seals&linux.F_SEAL_SHRINK != 0 { + // Seal does not allow shrink. + return false, syserror.EPERM + } + + // TODO(gvisor.dev/issues/1197): Invalidate mappings once we have + // mappings. + + rf.data.Truncate(size, rf.memFile) + rf.size = size + return true, nil +} + type regularFileFD struct { fileDescription diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go index 7b0a962f0..034a29fdb 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go @@ -313,3 +313,124 @@ func TestPRead(t *testing.T) { } } } + +func TestTruncate(t *testing.T) { + ctx := contexttest.Context(t) + fd, cleanup, err := newFileFD(ctx, 0644) + if err != nil { + t.Fatal(err) + } + defer cleanup() + + // Fill the file with some data. + data := bytes.Repeat([]byte("gVisor is awsome"), 100) + written, err := fd.Write(ctx, usermem.BytesIOSequence(data), vfs.WriteOptions{}) + if err != nil { + t.Fatalf("fd.Write failed: %v", err) + } + + // Size should be same as written. + sizeStatOpts := vfs.StatOptions{Mask: linux.STATX_SIZE} + stat, err := fd.Stat(ctx, sizeStatOpts) + if err != nil { + t.Fatalf("fd.Stat failed: %v", err) + } + if got, want := int64(stat.Size), written; got != want { + t.Errorf("fd.Stat got size %d, want %d", got, want) + } + + // Truncate down. + newSize := uint64(10) + if err := fd.SetStat(ctx, vfs.SetStatOptions{ + Stat: linux.Statx{ + Mask: linux.STATX_SIZE, + Size: newSize, + }, + }); err != nil { + t.Errorf("fd.Truncate failed: %v", err) + } + // Size should be updated. + statAfterTruncateDown, err := fd.Stat(ctx, sizeStatOpts) + if err != nil { + t.Fatalf("fd.Stat failed: %v", err) + } + if got, want := statAfterTruncateDown.Size, newSize; got != want { + t.Errorf("fd.Stat got size %d, want %d", got, want) + } + // We should only read newSize worth of data. + buf := make([]byte, 1000) + if n, err := fd.PRead(ctx, usermem.BytesIOSequence(buf), 0, vfs.ReadOptions{}); err != nil && err != io.EOF { + t.Fatalf("fd.PRead failed: %v", err) + } else if uint64(n) != newSize { + t.Errorf("fd.PRead got size %d, want %d", n, newSize) + } + // Mtime and Ctime should be bumped. + if got := statAfterTruncateDown.Mtime.ToNsec(); got <= stat.Mtime.ToNsec() { + t.Errorf("fd.Stat got Mtime %v, want > %v", got, stat.Mtime) + } + if got := statAfterTruncateDown.Ctime.ToNsec(); got <= stat.Ctime.ToNsec() { + t.Errorf("fd.Stat got Ctime %v, want > %v", got, stat.Ctime) + } + + // Truncate up. + newSize = 100 + if err := fd.SetStat(ctx, vfs.SetStatOptions{ + Stat: linux.Statx{ + Mask: linux.STATX_SIZE, + Size: newSize, + }, + }); err != nil { + t.Errorf("fd.Truncate failed: %v", err) + } + // Size should be updated. + statAfterTruncateUp, err := fd.Stat(ctx, sizeStatOpts) + if err != nil { + t.Fatalf("fd.Stat failed: %v", err) + } + if got, want := statAfterTruncateUp.Size, newSize; got != want { + t.Errorf("fd.Stat got size %d, want %d", got, want) + } + // We should read newSize worth of data. + buf = make([]byte, 1000) + if n, err := fd.PRead(ctx, usermem.BytesIOSequence(buf), 0, vfs.ReadOptions{}); err != nil && err != io.EOF { + t.Fatalf("fd.PRead failed: %v", err) + } else if uint64(n) != newSize { + t.Errorf("fd.PRead got size %d, want %d", n, newSize) + } + // Bytes should be null after 10, since we previously truncated to 10. + for i := uint64(10); i < newSize; i++ { + if buf[i] != 0 { + t.Errorf("fd.PRead got byte %d=%x, want 0", i, buf[i]) + break + } + } + // Mtime and Ctime should be bumped. + if got := statAfterTruncateUp.Mtime.ToNsec(); got <= statAfterTruncateDown.Mtime.ToNsec() { + t.Errorf("fd.Stat got Mtime %v, want > %v", got, statAfterTruncateDown.Mtime) + } + if got := statAfterTruncateUp.Ctime.ToNsec(); got <= statAfterTruncateDown.Ctime.ToNsec() { + t.Errorf("fd.Stat got Ctime %v, want > %v", got, stat.Ctime) + } + + // Truncate to the current size. + newSize = statAfterTruncateUp.Size + if err := fd.SetStat(ctx, vfs.SetStatOptions{ + Stat: linux.Statx{ + Mask: linux.STATX_SIZE, + Size: newSize, + }, + }); err != nil { + t.Errorf("fd.Truncate failed: %v", err) + } + statAfterTruncateNoop, err := fd.Stat(ctx, sizeStatOpts) + if err != nil { + t.Fatalf("fd.Stat failed: %v", err) + } + // Mtime and Ctime should not be bumped, since operation is a noop. + if got := statAfterTruncateNoop.Mtime.ToNsec(); got != statAfterTruncateUp.Mtime.ToNsec() { + t.Errorf("fd.Stat got Mtime %v, want %v", got, statAfterTruncateUp.Mtime) + } + if got := statAfterTruncateNoop.Ctime.ToNsec(); got != statAfterTruncateUp.Ctime.ToNsec() { + t.Errorf("fd.Stat got Ctime %v, want %v", got, statAfterTruncateUp.Ctime) + } +} diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index d6960ee47..1d4889c89 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -35,6 +35,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/syserror" ) // FilesystemType implements vfs.FilesystemType. @@ -121,6 +122,9 @@ func (d *dentry) DecRef() { // inode represents a filesystem object. type inode struct { + // clock is a realtime clock used to set timestamps in file operations. + clock time.Clock + // refs is a reference count. refs is accessed using atomic memory // operations. // @@ -151,13 +155,14 @@ type inode struct { const maxLinks = math.MaxUint32 func (i *inode) init(impl interface{}, fs *filesystem, creds *auth.Credentials, mode linux.FileMode) { - now := fs.clock.Now().Nanoseconds() + i.clock = fs.clock i.refs = 1 i.mode = uint32(mode) i.uid = uint32(creds.EffectiveKUID) i.gid = uint32(creds.EffectiveKGID) i.ino = atomic.AddUint64(&fs.nextInoMinusOne, 1) // Tmpfs creation sets atime, ctime, and mtime to current time. + now := i.clock.Now().Nanoseconds() i.atime = now i.ctime = now i.mtime = now @@ -270,30 +275,69 @@ func (i *inode) statTo(stat *linux.Statx) { } func (i *inode) setStat(stat linux.Statx) error { - // TODO(gvisor.dev/issues/1197): Handle stat.Size by growing/shrinking - // the file. if stat.Mask == 0 { return nil } i.mu.Lock() + var ( + needsMtimeBump bool + needsCtimeBump bool + ) mask := stat.Mask if mask&linux.STATX_MODE != 0 { atomic.StoreUint32(&i.mode, uint32(stat.Mode)) + needsCtimeBump = true } if mask&linux.STATX_UID != 0 { atomic.StoreUint32(&i.uid, stat.UID) + needsCtimeBump = true } if mask&linux.STATX_GID != 0 { atomic.StoreUint32(&i.gid, stat.GID) + needsCtimeBump = true + } + if mask&linux.STATX_SIZE != 0 { + switch impl := i.impl.(type) { + case *regularFile: + updated, err := impl.truncate(stat.Size) + if err != nil { + return err + } + if updated { + needsMtimeBump = true + needsCtimeBump = true + } + case *directory: + return syserror.EISDIR + case *symlink: + return syserror.EINVAL + case *namedPipe: + // Nothing. + default: + panic(fmt.Sprintf("unknown inode type: %T", i.impl)) + } } if mask&linux.STATX_ATIME != 0 { atomic.StoreInt64(&i.atime, stat.Atime.ToNsecCapped()) + needsCtimeBump = true + } + if mask&linux.STATX_MTIME != 0 { + atomic.StoreInt64(&i.mtime, stat.Mtime.ToNsecCapped()) + needsCtimeBump = true + // Ignore the mtime bump, since we just set it ourselves. + needsMtimeBump = false } if mask&linux.STATX_CTIME != 0 { atomic.StoreInt64(&i.ctime, stat.Ctime.ToNsecCapped()) + // Ignore the ctime bump, since we just set it ourselves. + needsCtimeBump = false } - if mask&linux.STATX_MTIME != 0 { - atomic.StoreInt64(&i.mtime, stat.Mtime.ToNsecCapped()) + now := i.clock.Now().Nanoseconds() + if needsMtimeBump { + atomic.StoreInt64(&i.mtime, now) + } + if needsCtimeBump { + atomic.StoreInt64(&i.ctime, now) } i.mu.Unlock() return nil -- cgit v1.2.3 From ab48112e41427579ecf585f6280be1e2d58acf06 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Thu, 16 Jan 2020 13:58:50 -0800 Subject: Add IfChange/ThenChange reminders in fs/proc There is a lot of code duplication for VFSv2 and this serves as remind to keep the copies in sync. Updates #1195 PiperOrigin-RevId: 290139234 --- pkg/sentry/fs/proc/cgroup.go | 4 ++++ pkg/sentry/fs/proc/cpuinfo.go | 4 ++++ pkg/sentry/fs/proc/exec_args.go | 4 ++++ pkg/sentry/fs/proc/fds.go | 4 ++++ pkg/sentry/fs/proc/filesystems.go | 4 ++++ pkg/sentry/fs/proc/fs.go | 4 ++++ pkg/sentry/fs/proc/inode.go | 4 ++++ pkg/sentry/fs/proc/loadavg.go | 4 ++++ pkg/sentry/fs/proc/meminfo.go | 4 ++++ pkg/sentry/fs/proc/mounts.go | 4 ++++ pkg/sentry/fs/proc/net.go | 4 ++++ pkg/sentry/fs/proc/proc.go | 4 ++++ pkg/sentry/fs/proc/stat.go | 4 ++++ pkg/sentry/fs/proc/sys.go | 4 ++++ pkg/sentry/fs/proc/sys_net.go | 4 ++++ pkg/sentry/fs/proc/task.go | 4 ++++ pkg/sentry/fs/proc/uid_gid_map.go | 4 ++++ pkg/sentry/fs/proc/uptime.go | 4 ++++ pkg/sentry/fs/proc/version.go | 4 ++++ 19 files changed, 76 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/proc/cgroup.go b/pkg/sentry/fs/proc/cgroup.go index 05e31c55d..c4abe319d 100644 --- a/pkg/sentry/fs/proc/cgroup.go +++ b/pkg/sentry/fs/proc/cgroup.go @@ -21,6 +21,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" ) +// LINT.IfChange + func newCGroupInode(ctx context.Context, msrc *fs.MountSource, cgroupControllers map[string]string) *fs.Inode { // From man 7 cgroups: "For each cgroup hierarchy of which the process // is a member, there is one entry containing three colon-separated @@ -39,3 +41,5 @@ func newCGroupInode(ctx context.Context, msrc *fs.MountSource, cgroupControllers return newStaticProcInode(ctx, msrc, []byte(data)) } + +// LINT.ThenChange(../../fsimpl/proc/tasks_files.go) diff --git a/pkg/sentry/fs/proc/cpuinfo.go b/pkg/sentry/fs/proc/cpuinfo.go index 6330337eb..df0c4e3a7 100644 --- a/pkg/sentry/fs/proc/cpuinfo.go +++ b/pkg/sentry/fs/proc/cpuinfo.go @@ -22,6 +22,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" ) +// LINT.IfChange + func newCPUInfo(ctx context.Context, msrc *fs.MountSource) *fs.Inode { k := kernel.KernelFromContext(ctx) features := k.FeatureSet() @@ -35,3 +37,5 @@ func newCPUInfo(ctx context.Context, msrc *fs.MountSource) *fs.Inode { } return newStaticProcInode(ctx, msrc, buf.Bytes()) } + +// LINT.ThenChange(../../fsimpl/proc/tasks_files.go) diff --git a/pkg/sentry/fs/proc/exec_args.go b/pkg/sentry/fs/proc/exec_args.go index 1d3a2d426..9aaeb780b 100644 --- a/pkg/sentry/fs/proc/exec_args.go +++ b/pkg/sentry/fs/proc/exec_args.go @@ -29,6 +29,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +// LINT.IfChange + // execArgType enumerates the types of exec arguments that are exposed through // proc. type execArgType int @@ -201,3 +203,5 @@ func (f *execArgFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequen } return int64(n), err } + +// LINT.ThenChange(../../fsimpl/proc/task.go) diff --git a/pkg/sentry/fs/proc/fds.go b/pkg/sentry/fs/proc/fds.go index bee421d76..2fa3cfa7d 100644 --- a/pkg/sentry/fs/proc/fds.go +++ b/pkg/sentry/fs/proc/fds.go @@ -28,6 +28,8 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) +// LINT.IfChange + // walkDescriptors finds the descriptor (file-flag pair) for the fd identified // by p, and calls the toInodeOperations callback with that descriptor. This is a helper // method for implementing fs.InodeOperations.Lookup. @@ -277,3 +279,5 @@ func (fdid *fdInfoDir) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs. } return fs.NewFile(ctx, dirent, flags, fops), nil } + +// LINT.ThenChange(../../fsimpl/proc/task_files.go) diff --git a/pkg/sentry/fs/proc/filesystems.go b/pkg/sentry/fs/proc/filesystems.go index e9250c51c..7b3b974ab 100644 --- a/pkg/sentry/fs/proc/filesystems.go +++ b/pkg/sentry/fs/proc/filesystems.go @@ -23,6 +23,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs/proc/seqfile" ) +// LINT.IfChange + // filesystemsData backs /proc/filesystems. // // +stateify savable @@ -59,3 +61,5 @@ func (*filesystemsData) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle // Return the SeqData and advance the generation counter. return []seqfile.SeqData{{Buf: buf.Bytes(), Handle: (*filesystemsData)(nil)}}, 1 } + +// LINT.ThenChange(../../fsimpl/proc/filesystem.go) diff --git a/pkg/sentry/fs/proc/fs.go b/pkg/sentry/fs/proc/fs.go index f14833805..761d24462 100644 --- a/pkg/sentry/fs/proc/fs.go +++ b/pkg/sentry/fs/proc/fs.go @@ -21,6 +21,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" ) +// LINT.IfChange + // filesystem is a procfs. // // +stateify savable @@ -79,3 +81,5 @@ func (f *filesystem) Mount(ctx context.Context, device string, flags fs.MountSou // never want them cached. return New(ctx, fs.NewNonCachingMountSource(ctx, f, flags), cgroups) } + +// LINT.ThenChange(../../fsimpl/proc/filesystem.go) diff --git a/pkg/sentry/fs/proc/inode.go b/pkg/sentry/fs/proc/inode.go index 0c04f81fa..723f6b661 100644 --- a/pkg/sentry/fs/proc/inode.go +++ b/pkg/sentry/fs/proc/inode.go @@ -26,6 +26,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/usermem" ) +// LINT.IfChange + // taskOwnedInodeOps wraps an fs.InodeOperations and overrides the UnstableAttr // method to return either the task or root as the owner, depending on the // task's dumpability. @@ -131,3 +133,5 @@ func newProcInode(ctx context.Context, iops fs.InodeOperations, msrc *fs.MountSo } return fs.NewInode(ctx, iops, msrc, sattr) } + +// LINT.ThenChange(../../fsimpl/proc/tasks.go) diff --git a/pkg/sentry/fs/proc/loadavg.go b/pkg/sentry/fs/proc/loadavg.go index 8602b7426..d7d2afcb7 100644 --- a/pkg/sentry/fs/proc/loadavg.go +++ b/pkg/sentry/fs/proc/loadavg.go @@ -22,6 +22,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs/proc/seqfile" ) +// LINT.IfChange + // loadavgData backs /proc/loadavg. // // +stateify savable @@ -53,3 +55,5 @@ func (d *loadavgData) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) }, }, 0 } + +// LINT.ThenChange(../../fsimpl/proc/tasks_files.go) diff --git a/pkg/sentry/fs/proc/meminfo.go b/pkg/sentry/fs/proc/meminfo.go index 495f3e3ba..313c6a32b 100644 --- a/pkg/sentry/fs/proc/meminfo.go +++ b/pkg/sentry/fs/proc/meminfo.go @@ -25,6 +25,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/usermem" ) +// LINT.IfChange + // meminfoData backs /proc/meminfo. // // +stateify savable @@ -83,3 +85,5 @@ func (d *meminfoData) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) fmt.Fprintf(&buf, "Shmem: %8d kB\n", snapshot.Tmpfs/1024) return []seqfile.SeqData{{Buf: buf.Bytes(), Handle: (*meminfoData)(nil)}}, 0 } + +// LINT.ThenChange(../../fsimpl/proc/tasks_files.go) diff --git a/pkg/sentry/fs/proc/mounts.go b/pkg/sentry/fs/proc/mounts.go index e33c4a460..5aedae799 100644 --- a/pkg/sentry/fs/proc/mounts.go +++ b/pkg/sentry/fs/proc/mounts.go @@ -25,6 +25,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" ) +// LINT.IfChange + // forEachMountSource runs f for the process root mount and each mount that is a // descendant of the root. func forEachMount(t *kernel.Task, fn func(string, *fs.Mount)) { @@ -195,3 +197,5 @@ func (mf *mountsFile) ReadSeqFileData(ctx context.Context, handle seqfile.SeqHan return []seqfile.SeqData{{Buf: buf.Bytes(), Handle: (*mountsFile)(nil)}}, 0 } + +// LINT.ThenChange(../../fsimpl/proc/tasks_files.go) diff --git a/pkg/sentry/fs/proc/net.go b/pkg/sentry/fs/proc/net.go index 402919924..3f17e98ea 100644 --- a/pkg/sentry/fs/proc/net.go +++ b/pkg/sentry/fs/proc/net.go @@ -38,6 +38,8 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/header" ) +// LINT.IfChange + // newNet creates a new proc net entry. func (p *proc) newNetDir(ctx context.Context, k *kernel.Kernel, msrc *fs.MountSource) *fs.Inode { var contents map[string]*fs.Inode @@ -831,3 +833,5 @@ func (n *netUDP) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]se } return data, 0 } + +// LINT.ThenChange(../../fsimpl/proc/tasks_net.go) diff --git a/pkg/sentry/fs/proc/proc.go b/pkg/sentry/fs/proc/proc.go index 56e92721e..da9341e4e 100644 --- a/pkg/sentry/fs/proc/proc.go +++ b/pkg/sentry/fs/proc/proc.go @@ -31,6 +31,8 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) +// LINT.IfChange + // proc is a root proc node. // // +stateify savable @@ -249,3 +251,5 @@ func (rpf *rootProcFile) Readdir(ctx context.Context, file *fs.File, ser fs.Dent } return offset, nil } + +// LINT.ThenChange(../../fsimpl/proc/tasks.go) diff --git a/pkg/sentry/fs/proc/stat.go b/pkg/sentry/fs/proc/stat.go index b641effbb..bc5b2bc7b 100644 --- a/pkg/sentry/fs/proc/stat.go +++ b/pkg/sentry/fs/proc/stat.go @@ -24,6 +24,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" ) +// LINT.IfChange + // statData backs /proc/stat. // // +stateify savable @@ -140,3 +142,5 @@ func (s *statData) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([] }, }, 0 } + +// LINT.ThenChange(../../fsimpl/proc/task_files.go) diff --git a/pkg/sentry/fs/proc/sys.go b/pkg/sentry/fs/proc/sys.go index cd37776c8..1062bd852 100644 --- a/pkg/sentry/fs/proc/sys.go +++ b/pkg/sentry/fs/proc/sys.go @@ -31,6 +31,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +// LINT.IfChange + // mmapMinAddrData backs /proc/sys/vm/mmap_min_addr. // // +stateify savable @@ -160,3 +162,5 @@ func (hf *hostnameFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequ } var _ fs.FileOperations = (*hostnameFile)(nil) + +// LINT.ThenChange(../../fsimpl/proc/tasks_sys.go) diff --git a/pkg/sentry/fs/proc/sys_net.go b/pkg/sentry/fs/proc/sys_net.go index a37e1fa06..b9e8ef35f 100644 --- a/pkg/sentry/fs/proc/sys_net.go +++ b/pkg/sentry/fs/proc/sys_net.go @@ -30,6 +30,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +// LINT.IfChange + type tcpMemDir int const ( @@ -364,3 +366,5 @@ func (p *proc) newSysNetDir(ctx context.Context, msrc *fs.MountSource) *fs.Inode d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) return newProcInode(ctx, d, msrc, fs.SpecialDirectory, nil) } + +// LINT.ThenChange(../../fsimpl/proc/tasks_sys.go) diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go index 9bf4b4527..7358d6ef9 100644 --- a/pkg/sentry/fs/proc/task.go +++ b/pkg/sentry/fs/proc/task.go @@ -37,6 +37,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +// LINT.IfChange + // getTaskMM returns t's MemoryManager. If getTaskMM succeeds, the MemoryManager's // users count is incremented, and must be decremented by the caller when it is // no longer in use. @@ -800,3 +802,5 @@ func (f *auxvecFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequenc n, err := dst.CopyOut(ctx, buf[offset:]) return int64(n), err } + +// LINT.ThenChange(../../fsimpl/proc/task.go|../../fsimpl/proc/task_files.go) diff --git a/pkg/sentry/fs/proc/uid_gid_map.go b/pkg/sentry/fs/proc/uid_gid_map.go index eea37d15c..3eacc9265 100644 --- a/pkg/sentry/fs/proc/uid_gid_map.go +++ b/pkg/sentry/fs/proc/uid_gid_map.go @@ -30,6 +30,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +// LINT.IfChange + // idMapInodeOperations implements fs.InodeOperations for // /proc/[pid]/{uid,gid}_map. // @@ -177,3 +179,5 @@ func (imfo *idMapFileOperations) Write(ctx context.Context, file *fs.File, src u // count, even if fewer bytes were used. return int64(srclen), nil } + +// LINT.ThenChange(../../fsimpl/proc/task_files.go) diff --git a/pkg/sentry/fs/proc/uptime.go b/pkg/sentry/fs/proc/uptime.go index 4e903917a..adfe58adb 100644 --- a/pkg/sentry/fs/proc/uptime.go +++ b/pkg/sentry/fs/proc/uptime.go @@ -28,6 +28,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +// LINT.IfChange + // uptime is a file containing the system uptime. // // +stateify savable @@ -85,3 +87,5 @@ func (f *uptimeFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequenc n, err := dst.CopyOut(ctx, s[offset:]) return int64(n), err } + +// LINT.ThenChange(../../fsimpl/proc/tasks_files.go) diff --git a/pkg/sentry/fs/proc/version.go b/pkg/sentry/fs/proc/version.go index a6d2c3cd3..27fd5b1cb 100644 --- a/pkg/sentry/fs/proc/version.go +++ b/pkg/sentry/fs/proc/version.go @@ -22,6 +22,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" ) +// LINT.IfChange + // versionData backs /proc/version. // // +stateify savable @@ -76,3 +78,5 @@ func (v *versionData) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) }, }, 0 } + +// LINT.ThenChange(../../fsimpl/proc/task_files.go) -- cgit v1.2.3 From 7a45ae7e67438697296fc12345202e3c76304096 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Thu, 16 Jan 2020 18:13:27 -0800 Subject: Implement setxattr for overlays. PiperOrigin-RevId: 290186303 --- pkg/sentry/fs/inode.go | 4 ++-- pkg/sentry/fs/inode_overlay.go | 13 ++++++++++--- pkg/sentry/syscalls/linux/sys_xattr.go | 2 +- 3 files changed, 13 insertions(+), 6 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/inode.go b/pkg/sentry/fs/inode.go index ee9d301ef..e4cf5a570 100644 --- a/pkg/sentry/fs/inode.go +++ b/pkg/sentry/fs/inode.go @@ -270,9 +270,9 @@ func (i *Inode) GetXattr(ctx context.Context, name string, size uint64) (string, } // SetXattr calls i.InodeOperations.SetXattr with i as the Inode. -func (i *Inode) SetXattr(ctx context.Context, name, value string, flags uint32) error { +func (i *Inode) SetXattr(ctx context.Context, d *Dirent, name, value string, flags uint32) error { if i.overlay != nil { - return overlaySetxattr(ctx, i.overlay, name, value, flags) + return overlaySetxattr(ctx, i.overlay, d, name, value, flags) } return i.InodeOperations.SetXattr(ctx, i, name, value, flags) } diff --git a/pkg/sentry/fs/inode_overlay.go b/pkg/sentry/fs/inode_overlay.go index b90da20d0..c477de837 100644 --- a/pkg/sentry/fs/inode_overlay.go +++ b/pkg/sentry/fs/inode_overlay.go @@ -552,9 +552,16 @@ func overlayGetXattr(ctx context.Context, o *overlayEntry, name string, size uin return s, err } -// TODO(b/146028302): Support setxattr for overlayfs. -func overlaySetxattr(ctx context.Context, o *overlayEntry, name, value string, flags uint32) error { - return syserror.EOPNOTSUPP +func overlaySetxattr(ctx context.Context, o *overlayEntry, d *Dirent, name, value string, flags uint32) error { + // Don't allow changes to overlay xattrs through a setxattr syscall. + if strings.HasPrefix(XattrOverlayPrefix, name) { + return syserror.EPERM + } + + if err := copyUp(ctx, d); err != nil { + return err + } + return o.upper.SetXattr(ctx, d, name, value, flags) } func overlayListXattr(ctx context.Context, o *overlayEntry) (map[string]struct{}, error) { diff --git a/pkg/sentry/syscalls/linux/sys_xattr.go b/pkg/sentry/syscalls/linux/sys_xattr.go index 816352218..23d20da6f 100644 --- a/pkg/sentry/syscalls/linux/sys_xattr.go +++ b/pkg/sentry/syscalls/linux/sys_xattr.go @@ -142,7 +142,7 @@ func setXattr(t *kernel.Task, d *fs.Dirent, dirPath bool, nameAddr, valueAddr us return syserror.EOPNOTSUPP } - return d.Inode.SetXattr(t, name, value, flags) + return d.Inode.SetXattr(t, d, name, value, flags) } func copyInXattrName(t *kernel.Task, nameAddr usermem.Addr) (string, error) { -- cgit v1.2.3 From 19b4653147c8ec1cd2cf6e2a8f9bfc7865a5f850 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Thu, 16 Jan 2020 20:19:40 -0800 Subject: Remove unused rpcinet. PiperOrigin-RevId: 290198756 --- pkg/sentry/fs/proc/BUILD | 2 - pkg/sentry/fs/proc/proc.go | 9 +- pkg/sentry/fs/proc/rpcinet_proc.go | 217 ------ pkg/sentry/fs/proc/sys.go | 9 +- pkg/sentry/socket/rpcinet/BUILD | 69 -- pkg/sentry/socket/rpcinet/conn/BUILD | 18 - pkg/sentry/socket/rpcinet/conn/conn.go | 187 ----- pkg/sentry/socket/rpcinet/device.go | 19 - pkg/sentry/socket/rpcinet/notifier/BUILD | 17 - pkg/sentry/socket/rpcinet/notifier/notifier.go | 231 ------- pkg/sentry/socket/rpcinet/rpcinet.go | 16 - pkg/sentry/socket/rpcinet/socket.go | 909 ------------------------- pkg/sentry/socket/rpcinet/stack.go | 177 ----- pkg/sentry/socket/rpcinet/stack_unsafe.go | 193 ------ pkg/sentry/socket/rpcinet/syscall_rpc.proto | 352 ---------- 15 files changed, 2 insertions(+), 2423 deletions(-) delete mode 100644 pkg/sentry/fs/proc/rpcinet_proc.go delete mode 100644 pkg/sentry/socket/rpcinet/BUILD delete mode 100644 pkg/sentry/socket/rpcinet/conn/BUILD delete mode 100644 pkg/sentry/socket/rpcinet/conn/conn.go delete mode 100644 pkg/sentry/socket/rpcinet/device.go delete mode 100644 pkg/sentry/socket/rpcinet/notifier/BUILD delete mode 100644 pkg/sentry/socket/rpcinet/notifier/notifier.go delete mode 100644 pkg/sentry/socket/rpcinet/rpcinet.go delete mode 100644 pkg/sentry/socket/rpcinet/socket.go delete mode 100644 pkg/sentry/socket/rpcinet/stack.go delete mode 100644 pkg/sentry/socket/rpcinet/stack_unsafe.go delete mode 100644 pkg/sentry/socket/rpcinet/syscall_rpc.proto (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/proc/BUILD b/pkg/sentry/fs/proc/BUILD index 94d46ab1b..cb37c6c6b 100644 --- a/pkg/sentry/fs/proc/BUILD +++ b/pkg/sentry/fs/proc/BUILD @@ -18,7 +18,6 @@ go_library( "mounts.go", "net.go", "proc.go", - "rpcinet_proc.go", "stat.go", "sys.go", "sys_net.go", @@ -46,7 +45,6 @@ go_library( "//pkg/sentry/limits", "//pkg/sentry/mm", "//pkg/sentry/socket", - "//pkg/sentry/socket/rpcinet", "//pkg/sentry/socket/unix", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usage", diff --git a/pkg/sentry/fs/proc/proc.go b/pkg/sentry/fs/proc/proc.go index da9341e4e..29867dc3a 100644 --- a/pkg/sentry/fs/proc/proc.go +++ b/pkg/sentry/fs/proc/proc.go @@ -27,7 +27,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs/proc/seqfile" "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet" "gvisor.dev/gvisor/pkg/syserror" ) @@ -87,15 +86,9 @@ func New(ctx context.Context, msrc *fs.MountSource, cgroupControllers map[string } // Add more contents that need proc to be initialized. + p.AddChild(ctx, "net", p.newNetDir(ctx, k, msrc)) p.AddChild(ctx, "sys", p.newSysDir(ctx, msrc)) - // If we're using rpcinet we will let it manage /proc/net. - if _, ok := p.k.NetworkStack().(*rpcinet.Stack); ok { - p.AddChild(ctx, "net", newRPCInetProcNet(ctx, msrc)) - } else { - p.AddChild(ctx, "net", p.newNetDir(ctx, k, msrc)) - } - return newProcInode(ctx, p, msrc, fs.SpecialDirectory, nil), nil } diff --git a/pkg/sentry/fs/proc/rpcinet_proc.go b/pkg/sentry/fs/proc/rpcinet_proc.go deleted file mode 100644 index 01ac97530..000000000 --- a/pkg/sentry/fs/proc/rpcinet_proc.go +++ /dev/null @@ -1,217 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "io" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet" - "gvisor.dev/gvisor/pkg/sentry/usermem" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/waiter" -) - -// rpcInetInode implements fs.InodeOperations. -type rpcInetInode struct { - fsutil.SimpleFileInode - - // filepath is the full path of this rpcInetInode. - filepath string - - k *kernel.Kernel -} - -func newRPCInetInode(ctx context.Context, msrc *fs.MountSource, filepath string, mode linux.FileMode) *fs.Inode { - f := &rpcInetInode{ - SimpleFileInode: *fsutil.NewSimpleFileInode(ctx, fs.RootOwner, fs.FilePermsFromMode(mode), linux.PROC_SUPER_MAGIC), - filepath: filepath, - k: kernel.KernelFromContext(ctx), - } - return newProcInode(ctx, f, msrc, fs.SpecialFile, nil) -} - -// GetFile implements fs.InodeOperations.GetFile. -func (i *rpcInetInode) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { - flags.Pread = true - flags.Pwrite = true - fops := &rpcInetFile{ - inode: i, - } - return fs.NewFile(ctx, dirent, flags, fops), nil -} - -// rpcInetFile implements fs.FileOperations as RPCs. -type rpcInetFile struct { - fsutil.FileGenericSeek `state:"nosave"` - fsutil.FileNoIoctl `state:"nosave"` - fsutil.FileNoMMap `state:"nosave"` - fsutil.FileNoSplice `state:"nosave"` - fsutil.FileNoopFlush `state:"nosave"` - fsutil.FileNoopFsync `state:"nosave"` - fsutil.FileNoopRelease `state:"nosave"` - fsutil.FileNotDirReaddir `state:"nosave"` - fsutil.FileUseInodeUnstableAttr `state:"nosave"` - waiter.AlwaysReady `state:"nosave"` - - inode *rpcInetInode -} - -// Read implements fs.FileOperations.Read. -// -// This method can panic if an rpcInetInode was created without an rpcinet -// stack. -func (f *rpcInetFile) Read(ctx context.Context, file *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { - if offset < 0 { - return 0, syserror.EINVAL - } - s, ok := f.inode.k.NetworkStack().(*rpcinet.Stack) - if !ok { - panic("Network stack is not a rpcinet.") - } - - contents, se := s.RPCReadFile(f.inode.filepath) - if se != nil || offset >= int64(len(contents)) { - return 0, io.EOF - } - - n, err := dst.CopyOut(ctx, contents[offset:]) - return int64(n), err -} - -// Write implements fs.FileOperations.Write. -// -// This method can panic if an rpcInetInode was created without an rpcInet -// stack. -func (f *rpcInetFile) Write(ctx context.Context, file *fs.File, src usermem.IOSequence, offset int64) (int64, error) { - s, ok := f.inode.k.NetworkStack().(*rpcinet.Stack) - if !ok { - panic("Network stack is not a rpcinet.") - } - - if src.NumBytes() == 0 { - return 0, nil - } - - b := make([]byte, src.NumBytes(), src.NumBytes()) - n, err := src.CopyIn(ctx, b) - if err != nil { - return int64(n), err - } - - written, se := s.RPCWriteFile(f.inode.filepath, b) - return int64(written), se.ToError() -} - -// newRPCInetProcNet will build an inode for /proc/net. -func newRPCInetProcNet(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - contents := map[string]*fs.Inode{ - "arp": newRPCInetInode(ctx, msrc, "/proc/net/arp", 0444), - "dev": newRPCInetInode(ctx, msrc, "/proc/net/dev", 0444), - "if_inet6": newRPCInetInode(ctx, msrc, "/proc/net/if_inet6", 0444), - "ipv6_route": newRPCInetInode(ctx, msrc, "/proc/net/ipv6_route", 0444), - "netlink": newRPCInetInode(ctx, msrc, "/proc/net/netlink", 0444), - "netstat": newRPCInetInode(ctx, msrc, "/proc/net/netstat", 0444), - "packet": newRPCInetInode(ctx, msrc, "/proc/net/packet", 0444), - "protocols": newRPCInetInode(ctx, msrc, "/proc/net/protocols", 0444), - "psched": newRPCInetInode(ctx, msrc, "/proc/net/psched", 0444), - "ptype": newRPCInetInode(ctx, msrc, "/proc/net/ptype", 0444), - "route": newRPCInetInode(ctx, msrc, "/proc/net/route", 0444), - "tcp": newRPCInetInode(ctx, msrc, "/proc/net/tcp", 0444), - "tcp6": newRPCInetInode(ctx, msrc, "/proc/net/tcp6", 0444), - "udp": newRPCInetInode(ctx, msrc, "/proc/net/udp", 0444), - "udp6": newRPCInetInode(ctx, msrc, "/proc/net/udp6", 0444), - } - - d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) - return newProcInode(ctx, d, msrc, fs.SpecialDirectory, nil) -} - -// newRPCInetProcSysNet will build an inode for /proc/sys/net. -func newRPCInetProcSysNet(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - contents := map[string]*fs.Inode{ - "ipv4": newRPCInetSysNetIPv4Dir(ctx, msrc), - "core": newRPCInetSysNetCore(ctx, msrc), - } - - d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) - return newProcInode(ctx, d, msrc, fs.SpecialDirectory, nil) -} - -// newRPCInetSysNetCore builds the /proc/sys/net/core directory. -func newRPCInetSysNetCore(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - contents := map[string]*fs.Inode{ - "default_qdisc": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/default_qdisc", 0444), - "message_burst": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/message_burst", 0444), - "message_cost": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/message_cost", 0444), - "optmem_max": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/optmem_max", 0444), - "rmem_default": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/rmem_default", 0444), - "rmem_max": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/rmem_max", 0444), - "somaxconn": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/somaxconn", 0444), - "wmem_default": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/wmem_default", 0444), - "wmem_max": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/wmem_max", 0444), - } - - d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) - return newProcInode(ctx, d, msrc, fs.SpecialDirectory, nil) -} - -// newRPCInetSysNetIPv4Dir builds the /proc/sys/net/ipv4 directory. -func newRPCInetSysNetIPv4Dir(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - contents := map[string]*fs.Inode{ - "ip_local_port_range": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/ip_local_port_range", 0444), - "ip_local_reserved_ports": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/ip_local_reserved_ports", 0444), - "ipfrag_time": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/ipfrag_time", 0444), - "ip_nonlocal_bind": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/ip_nonlocal_bind", 0444), - "ip_no_pmtu_disc": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/ip_no_pmtu_disc", 0444), - "tcp_allowed_congestion_control": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_allowed_congestion_control", 0444), - "tcp_available_congestion_control": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_available_congestion_control", 0444), - "tcp_base_mss": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_base_mss", 0444), - "tcp_congestion_control": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_congestion_control", 0644), - "tcp_dsack": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_dsack", 0644), - "tcp_early_retrans": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_early_retrans", 0644), - "tcp_fack": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_fack", 0644), - "tcp_fastopen": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_fastopen", 0644), - "tcp_fastopen_key": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_fastopen_key", 0444), - "tcp_fin_timeout": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_fin_timeout", 0644), - "tcp_invalid_ratelimit": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_invalid_ratelimit", 0444), - "tcp_keepalive_intvl": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_keepalive_intvl", 0644), - "tcp_keepalive_probes": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_keepalive_probes", 0644), - "tcp_keepalive_time": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_keepalive_time", 0644), - "tcp_mem": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_mem", 0444), - "tcp_mtu_probing": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_mtu_probing", 0644), - "tcp_no_metrics_save": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_no_metrics_save", 0444), - "tcp_probe_interval": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_probe_interval", 0444), - "tcp_probe_threshold": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_probe_threshold", 0444), - "tcp_retries1": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_retries1", 0644), - "tcp_retries2": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_retries2", 0644), - "tcp_rfc1337": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_rfc1337", 0444), - "tcp_rmem": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_rmem", 0444), - "tcp_sack": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_sack", 0644), - "tcp_slow_start_after_idle": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_slow_start_after_idle", 0644), - "tcp_synack_retries": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_synack_retries", 0644), - "tcp_syn_retries": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_syn_retries", 0644), - "tcp_timestamps": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_timestamps", 0644), - "tcp_wmem": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_wmem", 0444), - } - - d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) - return newProcInode(ctx, d, msrc, fs.SpecialDirectory, nil) -} diff --git a/pkg/sentry/fs/proc/sys.go b/pkg/sentry/fs/proc/sys.go index 1062bd852..2bdcf5f70 100644 --- a/pkg/sentry/fs/proc/sys.go +++ b/pkg/sentry/fs/proc/sys.go @@ -26,7 +26,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs/proc/seqfile" "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet" "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/waiter" ) @@ -106,16 +105,10 @@ func (p *proc) newVMDir(ctx context.Context, msrc *fs.MountSource) *fs.Inode { func (p *proc) newSysDir(ctx context.Context, msrc *fs.MountSource) *fs.Inode { children := map[string]*fs.Inode{ "kernel": p.newKernelDir(ctx, msrc), + "net": p.newSysNetDir(ctx, msrc), "vm": p.newVMDir(ctx, msrc), } - // If we're using rpcinet we will let it manage /proc/sys/net. - if _, ok := p.k.NetworkStack().(*rpcinet.Stack); ok { - children["net"] = newRPCInetProcSysNet(ctx, msrc) - } else { - children["net"] = p.newSysNetDir(ctx, msrc) - } - d := ramfs.NewDir(ctx, children, fs.RootOwner, fs.FilePermsFromMode(0555)) return newProcInode(ctx, d, msrc, fs.SpecialDirectory, nil) } diff --git a/pkg/sentry/socket/rpcinet/BUILD b/pkg/sentry/socket/rpcinet/BUILD deleted file mode 100644 index 4668b87d1..000000000 --- a/pkg/sentry/socket/rpcinet/BUILD +++ /dev/null @@ -1,69 +0,0 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//proto:def.bzl", "go_proto_library") -load("@rules_cc//cc:defs.bzl", "cc_proto_library") - -package(licenses = ["notice"]) - -go_library( - name = "rpcinet", - srcs = [ - "device.go", - "rpcinet.go", - "socket.go", - "stack.go", - "stack_unsafe.go", - ], - importpath = "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet", - visibility = ["//pkg/sentry:internal"], - deps = [ - ":syscall_rpc_go_proto", - "//pkg/abi/linux", - "//pkg/binary", - "//pkg/sentry/arch", - "//pkg/sentry/context", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/inet", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/time", - "//pkg/sentry/socket", - "//pkg/sentry/socket/hostinet", - "//pkg/sentry/socket/rpcinet/conn", - "//pkg/sentry/socket/rpcinet/notifier", - "//pkg/sentry/unimpl", - "//pkg/sentry/usermem", - "//pkg/syserr", - "//pkg/syserror", - "//pkg/tcpip", - "//pkg/tcpip/buffer", - "//pkg/tcpip/stack", - "//pkg/unet", - "//pkg/waiter", - ], -) - -proto_library( - name = "syscall_rpc_proto", - srcs = ["syscall_rpc.proto"], - visibility = [ - "//visibility:public", - ], -) - -cc_proto_library( - name = "syscall_rpc_cc_proto", - visibility = [ - "//visibility:public", - ], - deps = [":syscall_rpc_proto"], -) - -go_proto_library( - name = "syscall_rpc_go_proto", - importpath = "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/syscall_rpc_go_proto", - proto = ":syscall_rpc_proto", - visibility = [ - "//visibility:public", - ], -) diff --git a/pkg/sentry/socket/rpcinet/conn/BUILD b/pkg/sentry/socket/rpcinet/conn/BUILD deleted file mode 100644 index b2677c659..000000000 --- a/pkg/sentry/socket/rpcinet/conn/BUILD +++ /dev/null @@ -1,18 +0,0 @@ -load("//tools/go_stateify:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "conn", - srcs = ["conn.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/conn", - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/binary", - "//pkg/sentry/socket/rpcinet:syscall_rpc_go_proto", - "//pkg/sync", - "//pkg/syserr", - "//pkg/unet", - "@com_github_golang_protobuf//proto:go_default_library", - ], -) diff --git a/pkg/sentry/socket/rpcinet/conn/conn.go b/pkg/sentry/socket/rpcinet/conn/conn.go deleted file mode 100644 index 02f39c767..000000000 --- a/pkg/sentry/socket/rpcinet/conn/conn.go +++ /dev/null @@ -1,187 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package conn is an RPC connection to a syscall RPC server. -package conn - -import ( - "fmt" - "sync/atomic" - "syscall" - - "github.com/golang/protobuf/proto" - "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/syserr" - "gvisor.dev/gvisor/pkg/unet" - - pb "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/syscall_rpc_go_proto" -) - -type request struct { - response []byte - ready chan struct{} - ignoreResult bool -} - -// RPCConnection represents a single RPC connection to a syscall gofer. -type RPCConnection struct { - // reqID is the ID of the last request and must be accessed atomically. - reqID uint64 - - sendMu sync.Mutex - socket *unet.Socket - - reqMu sync.Mutex - requests map[uint64]request -} - -// NewRPCConnection initializes a RPC connection to a socket gofer. -func NewRPCConnection(s *unet.Socket) *RPCConnection { - conn := &RPCConnection{socket: s, requests: map[uint64]request{}} - go func() { // S/R-FIXME(b/77962828) - var nums [16]byte - for { - for n := 0; n < len(nums); { - nn, err := conn.socket.Read(nums[n:]) - if err != nil { - panic(fmt.Sprint("error reading length from socket rpc gofer: ", err)) - } - n += nn - } - - b := make([]byte, binary.LittleEndian.Uint64(nums[:8])) - id := binary.LittleEndian.Uint64(nums[8:]) - - for n := 0; n < len(b); { - nn, err := conn.socket.Read(b[n:]) - if err != nil { - panic(fmt.Sprint("error reading request from socket rpc gofer: ", err)) - } - n += nn - } - - conn.reqMu.Lock() - r := conn.requests[id] - if r.ignoreResult { - delete(conn.requests, id) - } else { - r.response = b - conn.requests[id] = r - } - conn.reqMu.Unlock() - close(r.ready) - } - }() - return conn -} - -// NewRequest makes a request to the RPC gofer and returns the request ID and a -// channel which will be closed once the request completes. -func (c *RPCConnection) NewRequest(req pb.SyscallRequest, ignoreResult bool) (uint64, chan struct{}) { - b, err := proto.Marshal(&req) - if err != nil { - panic(fmt.Sprint("invalid proto: ", err)) - } - - id := atomic.AddUint64(&c.reqID, 1) - ch := make(chan struct{}) - - c.reqMu.Lock() - c.requests[id] = request{ready: ch, ignoreResult: ignoreResult} - c.reqMu.Unlock() - - c.sendMu.Lock() - defer c.sendMu.Unlock() - - var nums [16]byte - binary.LittleEndian.PutUint64(nums[:8], uint64(len(b))) - binary.LittleEndian.PutUint64(nums[8:], id) - for n := 0; n < len(nums); { - nn, err := c.socket.Write(nums[n:]) - if err != nil { - panic(fmt.Sprint("error writing length and ID to socket gofer: ", err)) - } - n += nn - } - - for n := 0; n < len(b); { - nn, err := c.socket.Write(b[n:]) - if err != nil { - panic(fmt.Sprint("error writing request to socket gofer: ", err)) - } - n += nn - } - - return id, ch -} - -// RPCReadFile will execute the ReadFile helper RPC method which avoids the -// common pattern of open(2), read(2), close(2) by doing all three operations -// as a single RPC. It will read the entire file or return EFBIG if the file -// was too large. -func (c *RPCConnection) RPCReadFile(path string) ([]byte, *syserr.Error) { - req := &pb.SyscallRequest_ReadFile{&pb.ReadFileRequest{ - Path: path, - }} - - id, ch := c.NewRequest(pb.SyscallRequest{Args: req}, false /* ignoreResult */) - <-ch - - res := c.Request(id).Result.(*pb.SyscallResponse_ReadFile).ReadFile.Result - if e, ok := res.(*pb.ReadFileResponse_ErrorNumber); ok { - return nil, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - - return res.(*pb.ReadFileResponse_Data).Data, nil -} - -// RPCWriteFile will execute the WriteFile helper RPC method which avoids the -// common pattern of open(2), write(2), write(2), close(2) by doing all -// operations as a single RPC. -func (c *RPCConnection) RPCWriteFile(path string, data []byte) (int64, *syserr.Error) { - req := &pb.SyscallRequest_WriteFile{&pb.WriteFileRequest{ - Path: path, - Content: data, - }} - - id, ch := c.NewRequest(pb.SyscallRequest{Args: req}, false /* ignoreResult */) - <-ch - - res := c.Request(id).Result.(*pb.SyscallResponse_WriteFile).WriteFile - if e := res.ErrorNumber; e != 0 { - return int64(res.Written), syserr.FromHost(syscall.Errno(e)) - } - - return int64(res.Written), nil -} - -// Request retrieves the request corresponding to the given request ID. -// -// The channel returned by NewRequest must have been closed before Request can -// be called. This will happen automatically, do not manually close the -// channel. -func (c *RPCConnection) Request(id uint64) pb.SyscallResponse { - c.reqMu.Lock() - r := c.requests[id] - delete(c.requests, id) - c.reqMu.Unlock() - - var resp pb.SyscallResponse - if err := proto.Unmarshal(r.response, &resp); err != nil { - panic(fmt.Sprint("invalid proto: ", err)) - } - - return resp -} diff --git a/pkg/sentry/socket/rpcinet/device.go b/pkg/sentry/socket/rpcinet/device.go deleted file mode 100644 index 8cfd5f6e5..000000000 --- a/pkg/sentry/socket/rpcinet/device.go +++ /dev/null @@ -1,19 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package rpcinet - -import "gvisor.dev/gvisor/pkg/sentry/device" - -var socketDevice = device.NewAnonDevice() diff --git a/pkg/sentry/socket/rpcinet/notifier/BUILD b/pkg/sentry/socket/rpcinet/notifier/BUILD deleted file mode 100644 index a5954f22b..000000000 --- a/pkg/sentry/socket/rpcinet/notifier/BUILD +++ /dev/null @@ -1,17 +0,0 @@ -load("//tools/go_stateify:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "notifier", - srcs = ["notifier.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/notifier", - visibility = ["//:sandbox"], - deps = [ - "//pkg/sentry/socket/rpcinet:syscall_rpc_go_proto", - "//pkg/sentry/socket/rpcinet/conn", - "//pkg/sync", - "//pkg/waiter", - "@org_golang_x_sys//unix:go_default_library", - ], -) diff --git a/pkg/sentry/socket/rpcinet/notifier/notifier.go b/pkg/sentry/socket/rpcinet/notifier/notifier.go deleted file mode 100644 index 82b75d6dd..000000000 --- a/pkg/sentry/socket/rpcinet/notifier/notifier.go +++ /dev/null @@ -1,231 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package notifier implements an FD notifier implementation over RPC. -package notifier - -import ( - "fmt" - "syscall" - - "golang.org/x/sys/unix" - "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/conn" - pb "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/syscall_rpc_go_proto" - "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/waiter" -) - -type fdInfo struct { - queue *waiter.Queue - waiting bool -} - -// Notifier holds all the state necessary to issue notifications when IO events -// occur in the observed FDs. -type Notifier struct { - // rpcConn is the connection that is used for sending RPCs. - rpcConn *conn.RPCConnection - - // epFD is the epoll file descriptor used to register for io - // notifications. - epFD uint32 - - // mu protects fdMap. - mu sync.Mutex - - // fdMap maps file descriptors to their notification queues and waiting - // status. - fdMap map[uint32]*fdInfo -} - -// NewRPCNotifier creates a new notifier object. -func NewRPCNotifier(cn *conn.RPCConnection) (*Notifier, error) { - id, c := cn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_EpollCreate1{&pb.EpollCreate1Request{}}}, false /* ignoreResult */) - <-c - - res := cn.Request(id).Result.(*pb.SyscallResponse_EpollCreate1).EpollCreate1.Result - if e, ok := res.(*pb.EpollCreate1Response_ErrorNumber); ok { - return nil, syscall.Errno(e.ErrorNumber) - } - - w := &Notifier{ - rpcConn: cn, - epFD: res.(*pb.EpollCreate1Response_Fd).Fd, - fdMap: make(map[uint32]*fdInfo), - } - - go w.waitAndNotify() // S/R-FIXME(b/77962828) - - return w, nil -} - -// waitFD waits on mask for fd. The fdMap mutex must be hold. -func (n *Notifier) waitFD(fd uint32, fi *fdInfo, mask waiter.EventMask) error { - if !fi.waiting && mask == 0 { - return nil - } - - e := pb.EpollEvent{ - Events: mask.ToLinux() | unix.EPOLLET, - Fd: fd, - } - - switch { - case !fi.waiting && mask != 0: - id, c := n.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_EpollCtl{&pb.EpollCtlRequest{Epfd: n.epFD, Op: syscall.EPOLL_CTL_ADD, Fd: fd, Event: &e}}}, false /* ignoreResult */) - <-c - - e := n.rpcConn.Request(id).Result.(*pb.SyscallResponse_EpollCtl).EpollCtl.ErrorNumber - if e != 0 { - return syscall.Errno(e) - } - - fi.waiting = true - case fi.waiting && mask == 0: - id, c := n.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_EpollCtl{&pb.EpollCtlRequest{Epfd: n.epFD, Op: syscall.EPOLL_CTL_DEL, Fd: fd}}}, false /* ignoreResult */) - <-c - n.rpcConn.Request(id) - - fi.waiting = false - case fi.waiting && mask != 0: - id, c := n.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_EpollCtl{&pb.EpollCtlRequest{Epfd: n.epFD, Op: syscall.EPOLL_CTL_MOD, Fd: fd, Event: &e}}}, false /* ignoreResult */) - <-c - - e := n.rpcConn.Request(id).Result.(*pb.SyscallResponse_EpollCtl).EpollCtl.ErrorNumber - if e != 0 { - return syscall.Errno(e) - } - } - - return nil -} - -// addFD adds an FD to the list of FDs observed by n. -func (n *Notifier) addFD(fd uint32, queue *waiter.Queue) { - n.mu.Lock() - defer n.mu.Unlock() - - // Panic if we're already notifying on this FD. - if _, ok := n.fdMap[fd]; ok { - panic(fmt.Sprintf("File descriptor %d added twice", fd)) - } - - // We have nothing to wait for at the moment. Just add it to the map. - n.fdMap[fd] = &fdInfo{queue: queue} -} - -// updateFD updates the set of events the FD needs to be notified on. -func (n *Notifier) updateFD(fd uint32) error { - n.mu.Lock() - defer n.mu.Unlock() - - if fi, ok := n.fdMap[fd]; ok { - return n.waitFD(fd, fi, fi.queue.Events()) - } - - return nil -} - -// RemoveFD removes an FD from the list of FDs observed by n. -func (n *Notifier) removeFD(fd uint32) { - n.mu.Lock() - defer n.mu.Unlock() - - // Remove from map, then from epoll object. - n.waitFD(fd, n.fdMap[fd], 0) - delete(n.fdMap, fd) -} - -// hasFD returns true if the FD is in the list of observed FDs. -func (n *Notifier) hasFD(fd uint32) bool { - n.mu.Lock() - defer n.mu.Unlock() - - _, ok := n.fdMap[fd] - return ok -} - -// waitAndNotify loops waiting for io event notifications from the epoll -// object. Once notifications arrive, they are dispatched to the -// registered queue. -func (n *Notifier) waitAndNotify() error { - for { - id, c := n.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_EpollWait{&pb.EpollWaitRequest{Fd: n.epFD, NumEvents: 100, Msec: -1}}}, false /* ignoreResult */) - <-c - - res := n.rpcConn.Request(id).Result.(*pb.SyscallResponse_EpollWait).EpollWait.Result - if e, ok := res.(*pb.EpollWaitResponse_ErrorNumber); ok { - err := syscall.Errno(e.ErrorNumber) - // NOTE(magi): I don't think epoll_wait can return EAGAIN but I'm being - // conseratively careful here since exiting the notification thread - // would be really bad. - if err == syscall.EINTR || err == syscall.EAGAIN { - continue - } - return err - } - - n.mu.Lock() - for _, e := range res.(*pb.EpollWaitResponse_Events).Events.Events { - if fi, ok := n.fdMap[e.Fd]; ok { - fi.queue.Notify(waiter.EventMaskFromLinux(e.Events)) - } - } - n.mu.Unlock() - } -} - -// AddFD adds an FD to the list of observed FDs. -func (n *Notifier) AddFD(fd uint32, queue *waiter.Queue) error { - n.addFD(fd, queue) - return nil -} - -// UpdateFD updates the set of events the FD needs to be notified on. -func (n *Notifier) UpdateFD(fd uint32) error { - return n.updateFD(fd) -} - -// RemoveFD removes an FD from the list of observed FDs. -func (n *Notifier) RemoveFD(fd uint32) { - n.removeFD(fd) -} - -// HasFD returns true if the FD is in the list of observed FDs. -// -// This should only be used by tests to assert that FDs are correctly -// registered. -func (n *Notifier) HasFD(fd uint32) bool { - return n.hasFD(fd) -} - -// NonBlockingPoll polls the given fd in non-blocking fashion. It is used just -// to query the FD's current state; this method will block on the RPC response -// although the syscall is non-blocking. -func (n *Notifier) NonBlockingPoll(fd uint32, mask waiter.EventMask) waiter.EventMask { - for { - id, c := n.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_Poll{&pb.PollRequest{Fd: fd, Events: mask.ToLinux()}}}, false /* ignoreResult */) - <-c - - res := n.rpcConn.Request(id).Result.(*pb.SyscallResponse_Poll).Poll.Result - if e, ok := res.(*pb.PollResponse_ErrorNumber); ok { - if syscall.Errno(e.ErrorNumber) == syscall.EINTR { - continue - } - return mask - } - - return waiter.EventMaskFromLinux(res.(*pb.PollResponse_Events).Events) - } -} diff --git a/pkg/sentry/socket/rpcinet/rpcinet.go b/pkg/sentry/socket/rpcinet/rpcinet.go deleted file mode 100644 index 5d4fd4dac..000000000 --- a/pkg/sentry/socket/rpcinet/rpcinet.go +++ /dev/null @@ -1,16 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package rpcinet implements sockets using an RPC for each syscall. -package rpcinet diff --git a/pkg/sentry/socket/rpcinet/socket.go b/pkg/sentry/socket/rpcinet/socket.go deleted file mode 100644 index ddb76d9d4..000000000 --- a/pkg/sentry/socket/rpcinet/socket.go +++ /dev/null @@ -1,909 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package rpcinet - -import ( - "sync/atomic" - "syscall" - "time" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/kernel" - ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/socket" - "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/conn" - "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/notifier" - pb "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/syscall_rpc_go_proto" - "gvisor.dev/gvisor/pkg/sentry/unimpl" - "gvisor.dev/gvisor/pkg/sentry/usermem" - "gvisor.dev/gvisor/pkg/syserr" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/buffer" - "gvisor.dev/gvisor/pkg/waiter" -) - -// socketOperations implements fs.FileOperations and socket.Socket for a socket -// implemented using a host socket. -type socketOperations struct { - fsutil.FilePipeSeek `state:"nosave"` - fsutil.FileNotDirReaddir `state:"nosave"` - fsutil.FileNoFsync `state:"nosave"` - fsutil.FileNoMMap `state:"nosave"` - fsutil.FileNoSplice `state:"nosave"` - fsutil.FileNoopFlush `state:"nosave"` - fsutil.FileUseInodeUnstableAttr `state:"nosave"` - socket.SendReceiveTimeout - - family int // Read-only. - stype linux.SockType // Read-only. - protocol int // Read-only. - - fd uint32 // must be O_NONBLOCK - wq *waiter.Queue - rpcConn *conn.RPCConnection - notifier *notifier.Notifier - - // shState is the state of the connection with respect to shutdown. Because - // we're mixing non-blocking semantics on the other side we have to adapt for - // some strange differences between blocking and non-blocking sockets. - shState int32 -} - -// Verify that we actually implement socket.Socket. -var _ = socket.Socket(&socketOperations{}) - -// New creates a new RPC socket. -func newSocketFile(ctx context.Context, stack *Stack, family int, skType linux.SockType, protocol int) (*fs.File, *syserr.Error) { - id, c := stack.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_Socket{&pb.SocketRequest{Family: int64(family), Type: int64(skType | syscall.SOCK_NONBLOCK), Protocol: int64(protocol)}}}, false /* ignoreResult */) - <-c - - res := stack.rpcConn.Request(id).Result.(*pb.SyscallResponse_Socket).Socket.Result - if e, ok := res.(*pb.SocketResponse_ErrorNumber); ok { - return nil, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - fd := res.(*pb.SocketResponse_Fd).Fd - - var wq waiter.Queue - stack.notifier.AddFD(fd, &wq) - - dirent := socket.NewDirent(ctx, socketDevice) - defer dirent.DecRef() - return fs.NewFile(ctx, dirent, fs.FileFlags{Read: true, Write: true}, &socketOperations{ - family: family, - stype: skType, - protocol: protocol, - wq: &wq, - fd: fd, - rpcConn: stack.rpcConn, - notifier: stack.notifier, - }), nil -} - -func isBlockingErrno(err error) bool { - return err == syscall.EAGAIN || err == syscall.EWOULDBLOCK -} - -func translateIOSyscallError(err error) error { - if isBlockingErrno(err) { - return syserror.ErrWouldBlock - } - return err -} - -// setShutdownFlags will set the shutdown flag so we can handle blocking reads -// after a read shutdown. -func (s *socketOperations) setShutdownFlags(how int) { - var f tcpip.ShutdownFlags - switch how { - case linux.SHUT_RD: - f = tcpip.ShutdownRead - case linux.SHUT_WR: - f = tcpip.ShutdownWrite - case linux.SHUT_RDWR: - f = tcpip.ShutdownWrite | tcpip.ShutdownRead - } - - // Atomically update the flags. - for { - old := atomic.LoadInt32(&s.shState) - if atomic.CompareAndSwapInt32(&s.shState, old, old|int32(f)) { - break - } - } -} - -func (s *socketOperations) resetShutdownFlags() { - atomic.StoreInt32(&s.shState, 0) -} - -func (s *socketOperations) isShutRdSet() bool { - return atomic.LoadInt32(&s.shState)&int32(tcpip.ShutdownRead) != 0 -} - -func (s *socketOperations) isShutWrSet() bool { - return atomic.LoadInt32(&s.shState)&int32(tcpip.ShutdownWrite) != 0 -} - -// Release implements fs.FileOperations.Release. -func (s *socketOperations) Release() { - s.notifier.RemoveFD(s.fd) - - // We always need to close the FD. - _, _ = s.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_Close{&pb.CloseRequest{Fd: s.fd}}}, true /* ignoreResult */) -} - -// Readiness implements waiter.Waitable.Readiness. -func (s *socketOperations) Readiness(mask waiter.EventMask) waiter.EventMask { - return s.notifier.NonBlockingPoll(s.fd, mask) -} - -// EventRegister implements waiter.Waitable.EventRegister. -func (s *socketOperations) EventRegister(e *waiter.Entry, mask waiter.EventMask) { - s.wq.EventRegister(e, mask) - s.notifier.UpdateFD(s.fd) -} - -// EventUnregister implements waiter.Waitable.EventUnregister. -func (s *socketOperations) EventUnregister(e *waiter.Entry) { - s.wq.EventUnregister(e) - s.notifier.UpdateFD(s.fd) -} - -func rpcRead(t *kernel.Task, req *pb.SyscallRequest_Read) (*pb.ReadResponse_Data, *syserr.Error) { - s := t.NetworkContext().(*Stack) - id, c := s.rpcConn.NewRequest(pb.SyscallRequest{Args: req}, false /* ignoreResult */) - <-c - - res := s.rpcConn.Request(id).Result.(*pb.SyscallResponse_Read).Read.Result - if e, ok := res.(*pb.ReadResponse_ErrorNumber); ok { - return nil, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - - return res.(*pb.ReadResponse_Data), nil -} - -// Read implements fs.FileOperations.Read. -func (s *socketOperations) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, _ int64) (int64, error) { - req := &pb.SyscallRequest_Read{&pb.ReadRequest{ - Fd: s.fd, - Length: uint32(dst.NumBytes()), - }} - - res, se := rpcRead(ctx.(*kernel.Task), req) - if se == nil { - n, e := dst.CopyOut(ctx, res.Data) - return int64(n), e - } - - return 0, se.ToError() -} - -func rpcWrite(t *kernel.Task, req *pb.SyscallRequest_Write) (uint32, *syserr.Error) { - s := t.NetworkContext().(*Stack) - id, c := s.rpcConn.NewRequest(pb.SyscallRequest{Args: req}, false /* ignoreResult */) - <-c - - res := s.rpcConn.Request(id).Result.(*pb.SyscallResponse_Write).Write.Result - if e, ok := res.(*pb.WriteResponse_ErrorNumber); ok { - return 0, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - - return res.(*pb.WriteResponse_Length).Length, nil -} - -// Write implements fs.FileOperations.Write. -func (s *socketOperations) Write(ctx context.Context, _ *fs.File, src usermem.IOSequence, _ int64) (int64, error) { - t := ctx.(*kernel.Task) - v := buffer.NewView(int(src.NumBytes())) - - // Copy all the data into the buffer. - if _, err := src.CopyIn(t, v); err != nil { - return 0, err - } - - n, err := rpcWrite(t, &pb.SyscallRequest_Write{&pb.WriteRequest{Fd: s.fd, Data: v}}) - if n > 0 && n < uint32(src.NumBytes()) { - // The FileOperations.Write interface expects us to return ErrWouldBlock in - // the event of a partial write. - return int64(n), syserror.ErrWouldBlock - } - return int64(n), err.ToError() -} - -func rpcConnect(t *kernel.Task, fd uint32, sockaddr []byte) *syserr.Error { - s := t.NetworkContext().(*Stack) - id, c := s.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_Connect{&pb.ConnectRequest{Fd: uint32(fd), Address: sockaddr}}}, false /* ignoreResult */) - <-c - - if e := s.rpcConn.Request(id).Result.(*pb.SyscallResponse_Connect).Connect.ErrorNumber; e != 0 { - return syserr.FromHost(syscall.Errno(e)) - } - return nil -} - -// Connect implements socket.Socket.Connect. -func (s *socketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking bool) *syserr.Error { - if !blocking { - e := rpcConnect(t, s.fd, sockaddr) - if e == nil { - // Reset the shutdown state on new connects. - s.resetShutdownFlags() - } - return e - } - - // Register for notification when the endpoint becomes writable, then - // initiate the connection. - e, ch := waiter.NewChannelEntry(nil) - s.EventRegister(&e, waiter.EventOut|waiter.EventIn|waiter.EventHUp) - defer s.EventUnregister(&e) - for { - if err := rpcConnect(t, s.fd, sockaddr); err == nil || err != syserr.ErrInProgress && err != syserr.ErrAlreadyInProgress { - if err == nil { - // Reset the shutdown state on new connects. - s.resetShutdownFlags() - } - return err - } - - // It's pending, so we have to wait for a notification, and fetch the - // result once the wait completes. - if err := t.Block(ch); err != nil { - return syserr.FromError(err) - } - } -} - -func rpcAccept(t *kernel.Task, fd uint32, peer bool) (*pb.AcceptResponse_ResultPayload, *syserr.Error) { - stack := t.NetworkContext().(*Stack) - id, c := stack.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_Accept{&pb.AcceptRequest{Fd: fd, Peer: peer, Flags: syscall.SOCK_NONBLOCK}}}, false /* ignoreResult */) - <-c - - res := stack.rpcConn.Request(id).Result.(*pb.SyscallResponse_Accept).Accept.Result - if e, ok := res.(*pb.AcceptResponse_ErrorNumber); ok { - return nil, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - return res.(*pb.AcceptResponse_Payload).Payload, nil -} - -// Accept implements socket.Socket.Accept. -func (s *socketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (int32, linux.SockAddr, uint32, *syserr.Error) { - payload, se := rpcAccept(t, s.fd, peerRequested) - - // Check if we need to block. - if blocking && se == syserr.ErrTryAgain { - // Register for notifications. - e, ch := waiter.NewChannelEntry(nil) - // FIXME(b/119878986): This waiter.EventHUp is a partial - // measure, need to figure out how to translate linux events to - // internal events. - s.EventRegister(&e, waiter.EventIn|waiter.EventHUp) - defer s.EventUnregister(&e) - - // Try to accept the connection again; if it fails, then wait until we - // get a notification. - for { - if payload, se = rpcAccept(t, s.fd, peerRequested); se != syserr.ErrTryAgain { - break - } - - if err := t.Block(ch); err != nil { - return 0, nil, 0, syserr.FromError(err) - } - } - } - - // Handle any error from accept. - if se != nil { - return 0, nil, 0, se - } - - var wq waiter.Queue - s.notifier.AddFD(payload.Fd, &wq) - - dirent := socket.NewDirent(t, socketDevice) - defer dirent.DecRef() - fileFlags := fs.FileFlags{ - Read: true, - Write: true, - NonSeekable: true, - NonBlocking: flags&linux.SOCK_NONBLOCK != 0, - } - file := fs.NewFile(t, dirent, fileFlags, &socketOperations{ - family: s.family, - stype: s.stype, - protocol: s.protocol, - wq: &wq, - fd: payload.Fd, - rpcConn: s.rpcConn, - notifier: s.notifier, - }) - defer file.DecRef() - - fd, err := t.NewFDFrom(0, file, kernel.FDFlags{ - CloseOnExec: flags&linux.SOCK_CLOEXEC != 0, - }) - if err != nil { - return 0, nil, 0, syserr.FromError(err) - } - t.Kernel().RecordSocket(file) - - if peerRequested { - return fd, socket.UnmarshalSockAddr(s.family, payload.Address.Address), payload.Address.Length, nil - } - - return fd, nil, 0, nil -} - -// Bind implements socket.Socket.Bind. -func (s *socketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { - stack := t.NetworkContext().(*Stack) - id, c := stack.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_Bind{&pb.BindRequest{Fd: s.fd, Address: sockaddr}}}, false /* ignoreResult */) - <-c - - if e := stack.rpcConn.Request(id).Result.(*pb.SyscallResponse_Bind).Bind.ErrorNumber; e != 0 { - return syserr.FromHost(syscall.Errno(e)) - } - return nil -} - -// Listen implements socket.Socket.Listen. -func (s *socketOperations) Listen(t *kernel.Task, backlog int) *syserr.Error { - stack := t.NetworkContext().(*Stack) - id, c := stack.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_Listen{&pb.ListenRequest{Fd: s.fd, Backlog: int64(backlog)}}}, false /* ignoreResult */) - <-c - - if e := stack.rpcConn.Request(id).Result.(*pb.SyscallResponse_Listen).Listen.ErrorNumber; e != 0 { - return syserr.FromHost(syscall.Errno(e)) - } - return nil -} - -// Shutdown implements socket.Socket.Shutdown. -func (s *socketOperations) Shutdown(t *kernel.Task, how int) *syserr.Error { - // We save the shutdown state because of strange differences on linux - // related to recvs on blocking vs. non-blocking sockets after a SHUT_RD. - // We need to emulate that behavior on the blocking side. - // TODO(b/120096741): There is a possible race that can exist with loopback, - // where data could possibly be lost. - s.setShutdownFlags(how) - - stack := t.NetworkContext().(*Stack) - id, c := stack.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_Shutdown{&pb.ShutdownRequest{Fd: s.fd, How: int64(how)}}}, false /* ignoreResult */) - <-c - - if e := stack.rpcConn.Request(id).Result.(*pb.SyscallResponse_Shutdown).Shutdown.ErrorNumber; e != 0 { - return syserr.FromHost(syscall.Errno(e)) - } - - return nil -} - -// GetSockOpt implements socket.Socket.GetSockOpt. -func (s *socketOperations) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) { - // SO_RCVTIMEO and SO_SNDTIMEO are special because blocking is performed - // within the sentry. - if level == linux.SOL_SOCKET && name == linux.SO_RCVTIMEO { - if outLen < linux.SizeOfTimeval { - return nil, syserr.ErrInvalidArgument - } - - return linux.NsecToTimeval(s.RecvTimeout()), nil - } - if level == linux.SOL_SOCKET && name == linux.SO_SNDTIMEO { - if outLen < linux.SizeOfTimeval { - return nil, syserr.ErrInvalidArgument - } - - return linux.NsecToTimeval(s.SendTimeout()), nil - } - - stack := t.NetworkContext().(*Stack) - id, c := stack.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_GetSockOpt{&pb.GetSockOptRequest{Fd: s.fd, Level: int64(level), Name: int64(name), Length: uint32(outLen)}}}, false /* ignoreResult */) - <-c - - res := stack.rpcConn.Request(id).Result.(*pb.SyscallResponse_GetSockOpt).GetSockOpt.Result - if e, ok := res.(*pb.GetSockOptResponse_ErrorNumber); ok { - return nil, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - - return res.(*pb.GetSockOptResponse_Opt).Opt, nil -} - -// SetSockOpt implements socket.Socket.SetSockOpt. -func (s *socketOperations) SetSockOpt(t *kernel.Task, level int, name int, opt []byte) *syserr.Error { - // Because blocking actually happens within the sentry we need to inspect - // this socket option to determine if it's a SO_RCVTIMEO or SO_SNDTIMEO, - // and if so, we will save it and use it as the deadline for recv(2) - // or send(2) related syscalls. - if level == linux.SOL_SOCKET && name == linux.SO_RCVTIMEO { - if len(opt) < linux.SizeOfTimeval { - return syserr.ErrInvalidArgument - } - - var v linux.Timeval - binary.Unmarshal(opt[:linux.SizeOfTimeval], usermem.ByteOrder, &v) - if v.Usec < 0 || v.Usec >= int64(time.Second/time.Microsecond) { - return syserr.ErrDomain - } - s.SetRecvTimeout(v.ToNsecCapped()) - return nil - } - if level == linux.SOL_SOCKET && name == linux.SO_SNDTIMEO { - if len(opt) < linux.SizeOfTimeval { - return syserr.ErrInvalidArgument - } - - var v linux.Timeval - binary.Unmarshal(opt[:linux.SizeOfTimeval], usermem.ByteOrder, &v) - if v.Usec < 0 || v.Usec >= int64(time.Second/time.Microsecond) { - return syserr.ErrDomain - } - s.SetSendTimeout(v.ToNsecCapped()) - return nil - } - - stack := t.NetworkContext().(*Stack) - id, c := stack.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_SetSockOpt{&pb.SetSockOptRequest{Fd: s.fd, Level: int64(level), Name: int64(name), Opt: opt}}}, false /* ignoreResult */) - <-c - - if e := stack.rpcConn.Request(id).Result.(*pb.SyscallResponse_SetSockOpt).SetSockOpt.ErrorNumber; e != 0 { - return syserr.FromHost(syscall.Errno(e)) - } - return nil -} - -// GetPeerName implements socket.Socket.GetPeerName. -func (s *socketOperations) GetPeerName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) { - stack := t.NetworkContext().(*Stack) - id, c := stack.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_GetPeerName{&pb.GetPeerNameRequest{Fd: s.fd}}}, false /* ignoreResult */) - <-c - - res := stack.rpcConn.Request(id).Result.(*pb.SyscallResponse_GetPeerName).GetPeerName.Result - if e, ok := res.(*pb.GetPeerNameResponse_ErrorNumber); ok { - return nil, 0, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - - addr := res.(*pb.GetPeerNameResponse_Address).Address - return socket.UnmarshalSockAddr(s.family, addr.Address), addr.Length, nil -} - -// GetSockName implements socket.Socket.GetSockName. -func (s *socketOperations) GetSockName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) { - stack := t.NetworkContext().(*Stack) - id, c := stack.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_GetSockName{&pb.GetSockNameRequest{Fd: s.fd}}}, false /* ignoreResult */) - <-c - - res := stack.rpcConn.Request(id).Result.(*pb.SyscallResponse_GetSockName).GetSockName.Result - if e, ok := res.(*pb.GetSockNameResponse_ErrorNumber); ok { - return nil, 0, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - - addr := res.(*pb.GetSockNameResponse_Address).Address - return socket.UnmarshalSockAddr(s.family, addr.Address), addr.Length, nil -} - -func rpcIoctl(t *kernel.Task, fd, cmd uint32, arg []byte) ([]byte, error) { - stack := t.NetworkContext().(*Stack) - - id, c := stack.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_Ioctl{&pb.IOCtlRequest{Fd: fd, Cmd: cmd, Arg: arg}}}, false /* ignoreResult */) - <-c - - res := stack.rpcConn.Request(id).Result.(*pb.SyscallResponse_Ioctl).Ioctl.Result - if e, ok := res.(*pb.IOCtlResponse_ErrorNumber); ok { - return nil, syscall.Errno(e.ErrorNumber) - } - - return res.(*pb.IOCtlResponse_Value).Value, nil -} - -// ifconfIoctlFromStack populates a struct ifconf for the SIOCGIFCONF ioctl. -func ifconfIoctlFromStack(ctx context.Context, io usermem.IO, ifc *linux.IFConf) error { - // If Ptr is NULL, return the necessary buffer size via Len. - // Otherwise, write up to Len bytes starting at Ptr containing ifreq - // structs. - t := ctx.(*kernel.Task) - s := t.NetworkContext().(*Stack) - if s == nil { - return syserr.ErrNoDevice.ToError() - } - - if ifc.Ptr == 0 { - ifc.Len = int32(len(s.Interfaces())) * int32(linux.SizeOfIFReq) - return nil - } - - max := ifc.Len - ifc.Len = 0 - for key, ifaceAddrs := range s.InterfaceAddrs() { - iface := s.Interfaces()[key] - for _, ifaceAddr := range ifaceAddrs { - // Don't write past the end of the buffer. - if ifc.Len+int32(linux.SizeOfIFReq) > max { - break - } - if ifaceAddr.Family != linux.AF_INET { - continue - } - - // Populate ifr.ifr_addr. - ifr := linux.IFReq{} - ifr.SetName(iface.Name) - usermem.ByteOrder.PutUint16(ifr.Data[0:2], uint16(ifaceAddr.Family)) - usermem.ByteOrder.PutUint16(ifr.Data[2:4], 0) - copy(ifr.Data[4:8], ifaceAddr.Addr[:4]) - - // Copy the ifr to userspace. - dst := uintptr(ifc.Ptr) + uintptr(ifc.Len) - ifc.Len += int32(linux.SizeOfIFReq) - if _, err := usermem.CopyObjectOut(ctx, io, usermem.Addr(dst), ifr, usermem.IOOpts{ - AddressSpaceActive: true, - }); err != nil { - return err - } - } - } - return nil -} - -// Ioctl implements fs.FileOperations.Ioctl. -func (s *socketOperations) Ioctl(ctx context.Context, _ *fs.File, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { - t := ctx.(*kernel.Task) - - cmd := uint32(args[1].Int()) - arg := args[2].Pointer() - - var buf []byte - switch cmd { - // The following ioctls take 4 byte argument parameters. - case syscall.TIOCINQ, - syscall.TIOCOUTQ: - buf = make([]byte, 4) - // The following ioctls have args which are sizeof(struct ifreq). - case syscall.SIOCGIFADDR, - syscall.SIOCGIFBRDADDR, - syscall.SIOCGIFDSTADDR, - syscall.SIOCGIFFLAGS, - syscall.SIOCGIFHWADDR, - syscall.SIOCGIFINDEX, - syscall.SIOCGIFMAP, - syscall.SIOCGIFMETRIC, - syscall.SIOCGIFMTU, - syscall.SIOCGIFNAME, - syscall.SIOCGIFNETMASK, - syscall.SIOCGIFTXQLEN: - buf = make([]byte, linux.SizeOfIFReq) - case syscall.SIOCGIFCONF: - // SIOCGIFCONF has slightly different behavior than the others, in that it - // will need to populate the array of ifreqs. - var ifc linux.IFConf - if _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &ifc, usermem.IOOpts{ - AddressSpaceActive: true, - }); err != nil { - return 0, err - } - - if err := ifconfIoctlFromStack(ctx, io, &ifc); err != nil { - return 0, err - } - _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), ifc, usermem.IOOpts{ - AddressSpaceActive: true, - }) - - return 0, err - - case linux.SIOCGIFMEM, linux.SIOCGIFPFLAGS, linux.SIOCGMIIPHY, linux.SIOCGMIIREG: - unimpl.EmitUnimplementedEvent(ctx) - - default: - return 0, syserror.ENOTTY - } - - _, err := io.CopyIn(ctx, arg, buf, usermem.IOOpts{ - AddressSpaceActive: true, - }) - - if err != nil { - return 0, err - } - - v, err := rpcIoctl(t, s.fd, cmd, buf) - if err != nil { - return 0, err - } - - if len(v) != len(buf) { - return 0, syserror.EINVAL - } - - _, err = io.CopyOut(ctx, arg, v, usermem.IOOpts{ - AddressSpaceActive: true, - }) - return 0, err -} - -func rpcRecvMsg(t *kernel.Task, req *pb.SyscallRequest_Recvmsg) (*pb.RecvmsgResponse_ResultPayload, *syserr.Error) { - s := t.NetworkContext().(*Stack) - id, c := s.rpcConn.NewRequest(pb.SyscallRequest{Args: req}, false /* ignoreResult */) - <-c - - res := s.rpcConn.Request(id).Result.(*pb.SyscallResponse_Recvmsg).Recvmsg.Result - if e, ok := res.(*pb.RecvmsgResponse_ErrorNumber); ok { - return nil, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - - return res.(*pb.RecvmsgResponse_Payload).Payload, nil -} - -// Because we only support SO_TIMESTAMP we will search control messages for -// that value and set it if so, all other control messages will be ignored. -func (s *socketOperations) extractControlMessages(payload *pb.RecvmsgResponse_ResultPayload) socket.ControlMessages { - c := socket.ControlMessages{} - if len(payload.GetCmsgData()) > 0 { - // Parse the control messages looking for SO_TIMESTAMP. - msgs, e := syscall.ParseSocketControlMessage(payload.GetCmsgData()) - if e != nil { - return socket.ControlMessages{} - } - for _, m := range msgs { - if m.Header.Level != linux.SOL_SOCKET || m.Header.Type != linux.SO_TIMESTAMP { - continue - } - - // Let's parse the time stamp and set it. - if len(m.Data) < linux.SizeOfTimeval { - // Give up on locating the SO_TIMESTAMP option. - return socket.ControlMessages{} - } - - var v linux.Timeval - binary.Unmarshal(m.Data[:linux.SizeOfTimeval], usermem.ByteOrder, &v) - c.IP.HasTimestamp = true - c.IP.Timestamp = v.ToNsecCapped() - break - } - } - return c -} - -// RecvMsg implements socket.Socket.RecvMsg. -func (s *socketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, haveDeadline bool, deadline ktime.Time, senderRequested bool, controlDataLen uint64) (int, int, linux.SockAddr, uint32, socket.ControlMessages, *syserr.Error) { - req := &pb.SyscallRequest_Recvmsg{&pb.RecvmsgRequest{ - Fd: s.fd, - Length: uint32(dst.NumBytes()), - Sender: senderRequested, - Trunc: flags&linux.MSG_TRUNC != 0, - Peek: flags&linux.MSG_PEEK != 0, - CmsgLength: uint32(controlDataLen), - }} - - res, err := rpcRecvMsg(t, req) - if err == nil { - var e error - var n int - if len(res.Data) > 0 { - n, e = dst.CopyOut(t, res.Data) - if e == nil && n != len(res.Data) { - panic("CopyOut failed to copy full buffer") - } - } - c := s.extractControlMessages(res) - return int(res.Length), 0, socket.UnmarshalSockAddr(s.family, res.Address.GetAddress()), res.Address.GetLength(), c, syserr.FromError(e) - } - if err != syserr.ErrWouldBlock && err != syserr.ErrTryAgain || flags&linux.MSG_DONTWAIT != 0 { - return 0, 0, nil, 0, socket.ControlMessages{}, err - } - - // We'll have to block. Register for notifications and keep trying to - // send all the data. - e, ch := waiter.NewChannelEntry(nil) - s.EventRegister(&e, waiter.EventIn) - defer s.EventUnregister(&e) - - for { - res, err := rpcRecvMsg(t, req) - if err == nil { - var e error - var n int - if len(res.Data) > 0 { - n, e = dst.CopyOut(t, res.Data) - if e == nil && n != len(res.Data) { - panic("CopyOut failed to copy full buffer") - } - } - c := s.extractControlMessages(res) - return int(res.Length), 0, socket.UnmarshalSockAddr(s.family, res.Address.GetAddress()), res.Address.GetLength(), c, syserr.FromError(e) - } - if err != syserr.ErrWouldBlock && err != syserr.ErrTryAgain { - return 0, 0, nil, 0, socket.ControlMessages{}, err - } - - if s.isShutRdSet() { - // Blocking would have caused us to block indefinitely so we return 0, - // this is the same behavior as Linux. - return 0, 0, nil, 0, socket.ControlMessages{}, nil - } - - if err := t.BlockWithDeadline(ch, haveDeadline, deadline); err != nil { - if err == syserror.ETIMEDOUT { - return 0, 0, nil, 0, socket.ControlMessages{}, syserr.ErrTryAgain - } - return 0, 0, nil, 0, socket.ControlMessages{}, syserr.FromError(err) - } - } -} - -func rpcSendMsg(t *kernel.Task, req *pb.SyscallRequest_Sendmsg) (uint32, *syserr.Error) { - s := t.NetworkContext().(*Stack) - id, c := s.rpcConn.NewRequest(pb.SyscallRequest{Args: req}, false /* ignoreResult */) - <-c - - res := s.rpcConn.Request(id).Result.(*pb.SyscallResponse_Sendmsg).Sendmsg.Result - if e, ok := res.(*pb.SendmsgResponse_ErrorNumber); ok { - return 0, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - - return res.(*pb.SendmsgResponse_Length).Length, nil -} - -// SendMsg implements socket.Socket.SendMsg. -func (s *socketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to []byte, flags int, haveDeadline bool, deadline ktime.Time, controlMessages socket.ControlMessages) (int, *syserr.Error) { - // Whitelist flags. - if flags&^(syscall.MSG_DONTWAIT|syscall.MSG_EOR|syscall.MSG_FASTOPEN|syscall.MSG_MORE|syscall.MSG_NOSIGNAL) != 0 { - return 0, syserr.ErrInvalidArgument - } - - // Reject Unix control messages. - if !controlMessages.Unix.Empty() { - return 0, syserr.ErrInvalidArgument - } - - v := buffer.NewView(int(src.NumBytes())) - - // Copy all the data into the buffer. - if _, err := src.CopyIn(t, v); err != nil { - return 0, syserr.FromError(err) - } - - // TODO(bgeffon): this needs to change to map directly to a SendMsg syscall - // in the RPC. - totalWritten := 0 - n, err := rpcSendMsg(t, &pb.SyscallRequest_Sendmsg{&pb.SendmsgRequest{ - Fd: uint32(s.fd), - Data: v, - Address: to, - More: flags&linux.MSG_MORE != 0, - EndOfRecord: flags&linux.MSG_EOR != 0, - }}) - - if err != syserr.ErrWouldBlock && err != syserr.ErrTryAgain || flags&linux.MSG_DONTWAIT != 0 { - return int(n), err - } - - if n > 0 { - totalWritten += int(n) - v.TrimFront(int(n)) - } - - // We'll have to block. Register for notification and keep trying to - // send all the data. - e, ch := waiter.NewChannelEntry(nil) - s.EventRegister(&e, waiter.EventOut) - defer s.EventUnregister(&e) - - for { - n, err := rpcSendMsg(t, &pb.SyscallRequest_Sendmsg{&pb.SendmsgRequest{ - Fd: uint32(s.fd), - Data: v, - Address: to, - More: flags&linux.MSG_MORE != 0, - EndOfRecord: flags&linux.MSG_EOR != 0, - }}) - - if n > 0 { - totalWritten += int(n) - v.TrimFront(int(n)) - - if err == nil && totalWritten < int(src.NumBytes()) { - continue - } - } - - if err != syserr.ErrWouldBlock && err != syserr.ErrTryAgain { - // We eat the error in this situation. - return int(totalWritten), nil - } - - if err := t.BlockWithDeadline(ch, haveDeadline, deadline); err != nil { - if err == syserror.ETIMEDOUT { - return int(totalWritten), syserr.ErrTryAgain - } - return int(totalWritten), syserr.FromError(err) - } - } -} - -// State implements socket.Socket.State. -func (s *socketOperations) State() uint32 { - // TODO(b/127845868): Define a new rpc to query the socket state. - return 0 -} - -// Type implements socket.Socket.Type. -func (s *socketOperations) Type() (family int, skType linux.SockType, protocol int) { - return s.family, s.stype, s.protocol -} - -type socketProvider struct { - family int -} - -// Socket implements socket.Provider.Socket. -func (p *socketProvider) Socket(t *kernel.Task, stypeflags linux.SockType, protocol int) (*fs.File, *syserr.Error) { - // Check that we are using the RPC network stack. - stack := t.NetworkContext() - if stack == nil { - return nil, nil - } - - s, ok := stack.(*Stack) - if !ok { - return nil, nil - } - - // Only accept TCP and UDP. - // - // Try to restrict the flags we will accept to minimize backwards - // incompatibility with netstack. - stype := stypeflags & linux.SOCK_TYPE_MASK - switch stype { - case syscall.SOCK_STREAM: - switch protocol { - case 0, syscall.IPPROTO_TCP: - // ok - default: - return nil, nil - } - case syscall.SOCK_DGRAM: - switch protocol { - case 0, syscall.IPPROTO_UDP: - // ok - default: - return nil, nil - } - default: - return nil, nil - } - - return newSocketFile(t, s, p.family, stype, protocol) -} - -// Pair implements socket.Provider.Pair. -func (p *socketProvider) Pair(t *kernel.Task, stype linux.SockType, protocol int) (*fs.File, *fs.File, *syserr.Error) { - // Not supported by AF_INET/AF_INET6. - return nil, nil, nil -} - -func init() { - for _, family := range []int{syscall.AF_INET, syscall.AF_INET6} { - socket.RegisterProvider(family, &socketProvider{family}) - } -} diff --git a/pkg/sentry/socket/rpcinet/stack.go b/pkg/sentry/socket/rpcinet/stack.go deleted file mode 100644 index f7878a760..000000000 --- a/pkg/sentry/socket/rpcinet/stack.go +++ /dev/null @@ -1,177 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package rpcinet - -import ( - "fmt" - "syscall" - - "gvisor.dev/gvisor/pkg/sentry/inet" - "gvisor.dev/gvisor/pkg/sentry/socket/hostinet" - "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/conn" - "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/notifier" - "gvisor.dev/gvisor/pkg/syserr" - "gvisor.dev/gvisor/pkg/tcpip/stack" - "gvisor.dev/gvisor/pkg/unet" -) - -// Stack implements inet.Stack for RPC backed sockets. -type Stack struct { - interfaces map[int32]inet.Interface - interfaceAddrs map[int32][]inet.InterfaceAddr - routes []inet.Route - rpcConn *conn.RPCConnection - notifier *notifier.Notifier -} - -// NewStack returns a Stack containing the current state of the host network -// stack. -func NewStack(fd int32) (*Stack, error) { - sock, err := unet.NewSocket(int(fd)) - if err != nil { - return nil, err - } - - stack := &Stack{ - interfaces: make(map[int32]inet.Interface), - interfaceAddrs: make(map[int32][]inet.InterfaceAddr), - rpcConn: conn.NewRPCConnection(sock), - } - - var e error - stack.notifier, e = notifier.NewRPCNotifier(stack.rpcConn) - if e != nil { - return nil, e - } - - links, err := stack.DoNetlinkRouteRequest(syscall.RTM_GETLINK) - if err != nil { - return nil, fmt.Errorf("RTM_GETLINK failed: %v", err) - } - - addrs, err := stack.DoNetlinkRouteRequest(syscall.RTM_GETADDR) - if err != nil { - return nil, fmt.Errorf("RTM_GETADDR failed: %v", err) - } - - e = hostinet.ExtractHostInterfaces(links, addrs, stack.interfaces, stack.interfaceAddrs) - if e != nil { - return nil, e - } - - routes, err := stack.DoNetlinkRouteRequest(syscall.RTM_GETROUTE) - if err != nil { - return nil, fmt.Errorf("RTM_GETROUTE failed: %v", err) - } - - stack.routes, e = hostinet.ExtractHostRoutes(routes) - if e != nil { - return nil, e - } - - return stack, nil -} - -// RPCReadFile will execute the ReadFile helper RPC method which avoids the -// common pattern of open(2), read(2), close(2) by doing all three operations -// as a single RPC. It will read the entire file or return EFBIG if the file -// was too large. -func (s *Stack) RPCReadFile(path string) ([]byte, *syserr.Error) { - return s.rpcConn.RPCReadFile(path) -} - -// RPCWriteFile will execute the WriteFile helper RPC method which avoids the -// common pattern of open(2), write(2), write(2), close(2) by doing all -// operations as a single RPC. -func (s *Stack) RPCWriteFile(path string, data []byte) (int64, *syserr.Error) { - return s.rpcConn.RPCWriteFile(path, data) -} - -// Interfaces implements inet.Stack.Interfaces. -func (s *Stack) Interfaces() map[int32]inet.Interface { - interfaces := make(map[int32]inet.Interface) - for k, v := range s.interfaces { - interfaces[k] = v - } - return interfaces -} - -// InterfaceAddrs implements inet.Stack.InterfaceAddrs. -func (s *Stack) InterfaceAddrs() map[int32][]inet.InterfaceAddr { - addrs := make(map[int32][]inet.InterfaceAddr) - for k, v := range s.interfaceAddrs { - addrs[k] = append([]inet.InterfaceAddr(nil), v...) - } - return addrs -} - -// SupportsIPv6 implements inet.Stack.SupportsIPv6. -func (s *Stack) SupportsIPv6() bool { - panic("rpcinet handles procfs directly this method should not be called") -} - -// TCPReceiveBufferSize implements inet.Stack.TCPReceiveBufferSize. -func (s *Stack) TCPReceiveBufferSize() (inet.TCPBufferSize, error) { - panic("rpcinet handles procfs directly this method should not be called") -} - -// SetTCPReceiveBufferSize implements inet.Stack.SetTCPReceiveBufferSize. -func (s *Stack) SetTCPReceiveBufferSize(size inet.TCPBufferSize) error { - panic("rpcinet handles procfs directly this method should not be called") - -} - -// TCPSendBufferSize implements inet.Stack.TCPSendBufferSize. -func (s *Stack) TCPSendBufferSize() (inet.TCPBufferSize, error) { - panic("rpcinet handles procfs directly this method should not be called") - -} - -// SetTCPSendBufferSize implements inet.Stack.SetTCPSendBufferSize. -func (s *Stack) SetTCPSendBufferSize(size inet.TCPBufferSize) error { - panic("rpcinet handles procfs directly this method should not be called") -} - -// TCPSACKEnabled implements inet.Stack.TCPSACKEnabled. -func (s *Stack) TCPSACKEnabled() (bool, error) { - panic("rpcinet handles procfs directly this method should not be called") -} - -// SetTCPSACKEnabled implements inet.Stack.SetTCPSACKEnabled. -func (s *Stack) SetTCPSACKEnabled(enabled bool) error { - panic("rpcinet handles procfs directly this method should not be called") -} - -// Statistics implements inet.Stack.Statistics. -func (s *Stack) Statistics(stat interface{}, arg string) error { - return syserr.ErrEndpointOperation.ToError() -} - -// RouteTable implements inet.Stack.RouteTable. -func (s *Stack) RouteTable() []inet.Route { - return append([]inet.Route(nil), s.routes...) -} - -// Resume implements inet.Stack.Resume. -func (s *Stack) Resume() {} - -// RegisteredEndpoints implements inet.Stack.RegisteredEndpoints. -func (s *Stack) RegisteredEndpoints() []stack.TransportEndpoint { return nil } - -// CleanupEndpoints implements inet.Stack.CleanupEndpoints. -func (s *Stack) CleanupEndpoints() []stack.TransportEndpoint { return nil } - -// RestoreCleanupEndpoints implements inet.Stack.RestoreCleanupEndpoints. -func (s *Stack) RestoreCleanupEndpoints([]stack.TransportEndpoint) {} diff --git a/pkg/sentry/socket/rpcinet/stack_unsafe.go b/pkg/sentry/socket/rpcinet/stack_unsafe.go deleted file mode 100644 index a94bdad83..000000000 --- a/pkg/sentry/socket/rpcinet/stack_unsafe.go +++ /dev/null @@ -1,193 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package rpcinet - -import ( - "syscall" - "unsafe" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/binary" - pb "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/syscall_rpc_go_proto" - "gvisor.dev/gvisor/pkg/sentry/usermem" - "gvisor.dev/gvisor/pkg/syserr" -) - -// NewNetlinkRouteRequest builds a netlink message for getting the RIB, -// the routing information base. -func newNetlinkRouteRequest(proto, seq, family int) []byte { - rr := &syscall.NetlinkRouteRequest{} - rr.Header.Len = uint32(syscall.NLMSG_HDRLEN + syscall.SizeofRtGenmsg) - rr.Header.Type = uint16(proto) - rr.Header.Flags = syscall.NLM_F_DUMP | syscall.NLM_F_REQUEST - rr.Header.Seq = uint32(seq) - rr.Data.Family = uint8(family) - return netlinkRRtoWireFormat(rr) -} - -func netlinkRRtoWireFormat(rr *syscall.NetlinkRouteRequest) []byte { - b := make([]byte, rr.Header.Len) - *(*uint32)(unsafe.Pointer(&b[0:4][0])) = rr.Header.Len - *(*uint16)(unsafe.Pointer(&b[4:6][0])) = rr.Header.Type - *(*uint16)(unsafe.Pointer(&b[6:8][0])) = rr.Header.Flags - *(*uint32)(unsafe.Pointer(&b[8:12][0])) = rr.Header.Seq - *(*uint32)(unsafe.Pointer(&b[12:16][0])) = rr.Header.Pid - b[16] = byte(rr.Data.Family) - return b -} - -func (s *Stack) getNetlinkFd() (uint32, *syserr.Error) { - id, c := s.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_Socket{&pb.SocketRequest{Family: int64(syscall.AF_NETLINK), Type: int64(syscall.SOCK_RAW | syscall.SOCK_NONBLOCK), Protocol: int64(syscall.NETLINK_ROUTE)}}}, false /* ignoreResult */) - <-c - - res := s.rpcConn.Request(id).Result.(*pb.SyscallResponse_Socket).Socket.Result - if e, ok := res.(*pb.SocketResponse_ErrorNumber); ok { - return 0, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - return res.(*pb.SocketResponse_Fd).Fd, nil -} - -func (s *Stack) bindNetlinkFd(fd uint32, sockaddr []byte) *syserr.Error { - id, c := s.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_Bind{&pb.BindRequest{Fd: fd, Address: sockaddr}}}, false /* ignoreResult */) - <-c - - if e := s.rpcConn.Request(id).Result.(*pb.SyscallResponse_Bind).Bind.ErrorNumber; e != 0 { - return syserr.FromHost(syscall.Errno(e)) - } - return nil -} - -func (s *Stack) closeNetlinkFd(fd uint32) { - _, _ = s.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_Close{&pb.CloseRequest{Fd: fd}}}, true /* ignoreResult */) -} - -func (s *Stack) rpcSendMsg(req *pb.SyscallRequest_Sendmsg) (uint32, *syserr.Error) { - id, c := s.rpcConn.NewRequest(pb.SyscallRequest{Args: req}, false /* ignoreResult */) - <-c - - res := s.rpcConn.Request(id).Result.(*pb.SyscallResponse_Sendmsg).Sendmsg.Result - if e, ok := res.(*pb.SendmsgResponse_ErrorNumber); ok { - return 0, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - - return res.(*pb.SendmsgResponse_Length).Length, nil -} - -func (s *Stack) sendMsg(fd uint32, buf []byte, to []byte, flags int) (int, *syserr.Error) { - // Whitelist flags. - if flags&^(syscall.MSG_DONTWAIT|syscall.MSG_EOR|syscall.MSG_FASTOPEN|syscall.MSG_MORE|syscall.MSG_NOSIGNAL) != 0 { - return 0, syserr.ErrInvalidArgument - } - - req := &pb.SyscallRequest_Sendmsg{&pb.SendmsgRequest{ - Fd: fd, - Data: buf, - Address: to, - More: flags&linux.MSG_MORE != 0, - EndOfRecord: flags&linux.MSG_EOR != 0, - }} - - n, err := s.rpcSendMsg(req) - return int(n), err -} - -func (s *Stack) rpcRecvMsg(req *pb.SyscallRequest_Recvmsg) (*pb.RecvmsgResponse_ResultPayload, *syserr.Error) { - id, c := s.rpcConn.NewRequest(pb.SyscallRequest{Args: req}, false /* ignoreResult */) - <-c - - res := s.rpcConn.Request(id).Result.(*pb.SyscallResponse_Recvmsg).Recvmsg.Result - if e, ok := res.(*pb.RecvmsgResponse_ErrorNumber); ok { - return nil, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - - return res.(*pb.RecvmsgResponse_Payload).Payload, nil -} - -func (s *Stack) recvMsg(fd, l, flags uint32) ([]byte, *syserr.Error) { - req := &pb.SyscallRequest_Recvmsg{&pb.RecvmsgRequest{ - Fd: fd, - Length: l, - Sender: false, - Trunc: flags&linux.MSG_TRUNC != 0, - Peek: flags&linux.MSG_PEEK != 0, - }} - - res, err := s.rpcRecvMsg(req) - if err != nil { - return nil, err - } - return res.Data, nil -} - -func (s *Stack) netlinkRequest(proto, family int) ([]byte, error) { - fd, err := s.getNetlinkFd() - if err != nil { - return nil, err.ToError() - } - defer s.closeNetlinkFd(fd) - - lsa := syscall.SockaddrNetlink{Family: syscall.AF_NETLINK} - b := binary.Marshal(nil, usermem.ByteOrder, &lsa) - if err := s.bindNetlinkFd(fd, b); err != nil { - return nil, err.ToError() - } - - wb := newNetlinkRouteRequest(proto, 1, family) - _, err = s.sendMsg(fd, wb, b, 0) - if err != nil { - return nil, err.ToError() - } - - var tab []byte -done: - for { - rb, err := s.recvMsg(fd, uint32(syscall.Getpagesize()), 0) - nr := len(rb) - if err != nil { - return nil, err.ToError() - } - - if nr < syscall.NLMSG_HDRLEN { - return nil, syserr.ErrInvalidArgument.ToError() - } - - tab = append(tab, rb...) - msgs, e := syscall.ParseNetlinkMessage(rb) - if e != nil { - return nil, e - } - - for _, m := range msgs { - if m.Header.Type == syscall.NLMSG_DONE { - break done - } - if m.Header.Type == syscall.NLMSG_ERROR { - return nil, syserr.ErrInvalidArgument.ToError() - } - } - } - - return tab, nil -} - -// DoNetlinkRouteRequest returns routing information base, also known as RIB, -// which consists of network facility information, states and parameters. -func (s *Stack) DoNetlinkRouteRequest(req int) ([]syscall.NetlinkMessage, error) { - data, err := s.netlinkRequest(req, syscall.AF_UNSPEC) - if err != nil { - return nil, err - } - return syscall.ParseNetlinkMessage(data) -} diff --git a/pkg/sentry/socket/rpcinet/syscall_rpc.proto b/pkg/sentry/socket/rpcinet/syscall_rpc.proto deleted file mode 100644 index b677e9eb3..000000000 --- a/pkg/sentry/socket/rpcinet/syscall_rpc.proto +++ /dev/null @@ -1,352 +0,0 @@ -syntax = "proto3"; - -// package syscall_rpc is a set of networking related system calls that can be -// forwarded to a socket gofer. -// -package syscall_rpc; - -message SendmsgRequest { - uint32 fd = 1; - bytes data = 2 [ctype = CORD]; - bytes address = 3; - bool more = 4; - bool end_of_record = 5; -} - -message SendmsgResponse { - oneof result { - uint32 error_number = 1; - uint32 length = 2; - } -} - -message IOCtlRequest { - uint32 fd = 1; - uint32 cmd = 2; - bytes arg = 3; -} - -message IOCtlResponse { - oneof result { - uint32 error_number = 1; - bytes value = 2; - } -} - -message RecvmsgRequest { - uint32 fd = 1; - uint32 length = 2; - bool sender = 3; - bool peek = 4; - bool trunc = 5; - uint32 cmsg_length = 6; -} - -message OpenRequest { - bytes path = 1; - uint32 flags = 2; - uint32 mode = 3; -} - -message OpenResponse { - oneof result { - uint32 error_number = 1; - uint32 fd = 2; - } -} - -message ReadRequest { - uint32 fd = 1; - uint32 length = 2; -} - -message ReadResponse { - oneof result { - uint32 error_number = 1; - bytes data = 2 [ctype = CORD]; - } -} - -message ReadFileRequest { - string path = 1; -} - -message ReadFileResponse { - oneof result { - uint32 error_number = 1; - bytes data = 2 [ctype = CORD]; - } -} - -message WriteRequest { - uint32 fd = 1; - bytes data = 2 [ctype = CORD]; -} - -message WriteResponse { - oneof result { - uint32 error_number = 1; - uint32 length = 2; - } -} - -message WriteFileRequest { - string path = 1; - bytes content = 2; -} - -message WriteFileResponse { - uint32 error_number = 1; - uint32 written = 2; -} - -message AddressResponse { - bytes address = 1; - uint32 length = 2; -} - -message RecvmsgResponse { - message ResultPayload { - bytes data = 1 [ctype = CORD]; - AddressResponse address = 2; - uint32 length = 3; - bytes cmsg_data = 4; - } - oneof result { - uint32 error_number = 1; - ResultPayload payload = 2; - } -} - -message BindRequest { - uint32 fd = 1; - bytes address = 2; -} - -message BindResponse { - uint32 error_number = 1; -} - -message AcceptRequest { - uint32 fd = 1; - bool peer = 2; - int64 flags = 3; -} - -message AcceptResponse { - message ResultPayload { - uint32 fd = 1; - AddressResponse address = 2; - } - oneof result { - uint32 error_number = 1; - ResultPayload payload = 2; - } -} - -message ConnectRequest { - uint32 fd = 1; - bytes address = 2; -} - -message ConnectResponse { - uint32 error_number = 1; -} - -message ListenRequest { - uint32 fd = 1; - int64 backlog = 2; -} - -message ListenResponse { - uint32 error_number = 1; -} - -message ShutdownRequest { - uint32 fd = 1; - int64 how = 2; -} - -message ShutdownResponse { - uint32 error_number = 1; -} - -message CloseRequest { - uint32 fd = 1; -} - -message CloseResponse { - uint32 error_number = 1; -} - -message GetSockOptRequest { - uint32 fd = 1; - int64 level = 2; - int64 name = 3; - uint32 length = 4; -} - -message GetSockOptResponse { - oneof result { - uint32 error_number = 1; - bytes opt = 2; - } -} - -message SetSockOptRequest { - uint32 fd = 1; - int64 level = 2; - int64 name = 3; - bytes opt = 4; -} - -message SetSockOptResponse { - uint32 error_number = 1; -} - -message GetSockNameRequest { - uint32 fd = 1; -} - -message GetSockNameResponse { - oneof result { - uint32 error_number = 1; - AddressResponse address = 2; - } -} - -message GetPeerNameRequest { - uint32 fd = 1; -} - -message GetPeerNameResponse { - oneof result { - uint32 error_number = 1; - AddressResponse address = 2; - } -} - -message SocketRequest { - int64 family = 1; - int64 type = 2; - int64 protocol = 3; -} - -message SocketResponse { - oneof result { - uint32 error_number = 1; - uint32 fd = 2; - } -} - -message EpollWaitRequest { - uint32 fd = 1; - uint32 num_events = 2; - sint64 msec = 3; -} - -message EpollEvent { - uint32 fd = 1; - uint32 events = 2; -} - -message EpollEvents { - repeated EpollEvent events = 1; -} - -message EpollWaitResponse { - oneof result { - uint32 error_number = 1; - EpollEvents events = 2; - } -} - -message EpollCtlRequest { - uint32 epfd = 1; - int64 op = 2; - uint32 fd = 3; - EpollEvent event = 4; -} - -message EpollCtlResponse { - uint32 error_number = 1; -} - -message EpollCreate1Request { - int64 flag = 1; -} - -message EpollCreate1Response { - oneof result { - uint32 error_number = 1; - uint32 fd = 2; - } -} - -message PollRequest { - uint32 fd = 1; - uint32 events = 2; -} - -message PollResponse { - oneof result { - uint32 error_number = 1; - uint32 events = 2; - } -} - -message SyscallRequest { - oneof args { - SocketRequest socket = 1; - SendmsgRequest sendmsg = 2; - RecvmsgRequest recvmsg = 3; - BindRequest bind = 4; - AcceptRequest accept = 5; - ConnectRequest connect = 6; - ListenRequest listen = 7; - ShutdownRequest shutdown = 8; - CloseRequest close = 9; - GetSockOptRequest get_sock_opt = 10; - SetSockOptRequest set_sock_opt = 11; - GetSockNameRequest get_sock_name = 12; - GetPeerNameRequest get_peer_name = 13; - EpollWaitRequest epoll_wait = 14; - EpollCtlRequest epoll_ctl = 15; - EpollCreate1Request epoll_create1 = 16; - PollRequest poll = 17; - ReadRequest read = 18; - WriteRequest write = 19; - OpenRequest open = 20; - IOCtlRequest ioctl = 21; - WriteFileRequest write_file = 22; - ReadFileRequest read_file = 23; - } -} - -message SyscallResponse { - oneof result { - SocketResponse socket = 1; - SendmsgResponse sendmsg = 2; - RecvmsgResponse recvmsg = 3; - BindResponse bind = 4; - AcceptResponse accept = 5; - ConnectResponse connect = 6; - ListenResponse listen = 7; - ShutdownResponse shutdown = 8; - CloseResponse close = 9; - GetSockOptResponse get_sock_opt = 10; - SetSockOptResponse set_sock_opt = 11; - GetSockNameResponse get_sock_name = 12; - GetPeerNameResponse get_peer_name = 13; - EpollWaitResponse epoll_wait = 14; - EpollCtlResponse epoll_ctl = 15; - EpollCreate1Response epoll_create1 = 16; - PollResponse poll = 17; - ReadResponse read = 18; - WriteResponse write = 19; - OpenResponse open = 20; - IOCtlResponse ioctl = 21; - WriteFileResponse write_file = 22; - ReadFileResponse read_file = 23; - } -} -- cgit v1.2.3 From 345df7cab48ac79bccf2620900cd972b3026296d Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 17 Jan 2020 08:10:14 -0800 Subject: Add explanation for implementation of BSD full file locks. PiperOrigin-RevId: 290272560 --- pkg/sentry/fs/lock/lock.go | 3 +++ 1 file changed, 3 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/lock/lock.go b/pkg/sentry/fs/lock/lock.go index 41b040818..926538d90 100644 --- a/pkg/sentry/fs/lock/lock.go +++ b/pkg/sentry/fs/lock/lock.go @@ -78,6 +78,9 @@ const ( ) // LockEOF is the maximal possible end of a regional file lock. +// +// A BSD-style full file lock can be represented as a regional file lock from +// offset 0 to LockEOF. const LockEOF = math.MaxUint64 // Lock is a regional file lock. It consists of either a single writer -- cgit v1.2.3 From acf2d6dcc34501d2573f9c3f2b6da80308f3267e Mon Sep 17 00:00:00 2001 From: Haibo Xu Date: Fri, 17 Jan 2020 08:22:21 -0800 Subject: Enable stat syscall support on arm64. x86 and arm64 use a different stat struct in Linux kernel, so the stat() syscall implementation has to handle the file stat data separately. Signed-off-by: Haibo Xu Change-Id: If3986e915a667362257a54e7fbbcc1fe18951015 COPYBARA_INTEGRATE_REVIEW=https://github.com/google/gvisor/pull/1493 from xiaobo55x:stat f15a216d9297eb9a96d2c483d396a9919145d7fa PiperOrigin-RevId: 290274287 --- pkg/sentry/syscalls/linux/BUILD | 2 + pkg/sentry/syscalls/linux/linux64_arm64.go | 8 +++ pkg/sentry/syscalls/linux/sys_stat.go | 51 ------------------- pkg/sentry/syscalls/linux/sys_stat_amd64.go | 75 ++++++++++++++++++++++++++++ pkg/sentry/syscalls/linux/sys_stat_arm64.go | 77 +++++++++++++++++++++++++++++ 5 files changed, 162 insertions(+), 51 deletions(-) create mode 100644 pkg/sentry/syscalls/linux/sys_stat_amd64.go create mode 100644 pkg/sentry/syscalls/linux/sys_stat_arm64.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD index 430d796ba..917f74e07 100644 --- a/pkg/sentry/syscalls/linux/BUILD +++ b/pkg/sentry/syscalls/linux/BUILD @@ -42,6 +42,8 @@ go_library( "sys_socket.go", "sys_splice.go", "sys_stat.go", + "sys_stat_amd64.go", + "sys_stat_arm64.go", "sys_sync.go", "sys_sysinfo.go", "sys_syslog.go", diff --git a/pkg/sentry/syscalls/linux/linux64_arm64.go b/pkg/sentry/syscalls/linux/linux64_arm64.go index 8c1b20911..c9629f6f3 100644 --- a/pkg/sentry/syscalls/linux/linux64_arm64.go +++ b/pkg/sentry/syscalls/linux/linux64_arm64.go @@ -61,6 +61,7 @@ var ARM64 = &kernel.SyscallTable{ 22: syscalls.Supported("epoll_pwait", EpollPwait), 23: syscalls.Supported("dup", Dup), 24: syscalls.Supported("dup3", Dup3), + 25: syscalls.PartiallySupported("fcntl", Fcntl, "Not all options are supported.", nil), 26: syscalls.Supported("inotify_init1", InotifyInit1), 27: syscalls.PartiallySupported("inotify_add_watch", InotifyAddWatch, "inotify events are only available inside the sandbox.", nil), 28: syscalls.PartiallySupported("inotify_rm_watch", InotifyRmWatch, "inotify events are only available inside the sandbox.", nil), @@ -78,7 +79,9 @@ var ARM64 = &kernel.SyscallTable{ 40: syscalls.PartiallySupported("mount", Mount, "Not all options or file systems are supported.", nil), 41: syscalls.Error("pivot_root", syserror.EPERM, "", nil), 42: syscalls.Error("nfsservctl", syserror.ENOSYS, "Removed after Linux 3.1.", nil), + 43: syscalls.PartiallySupported("statfs", Statfs, "Depends on the backing file system implementation.", nil), 44: syscalls.PartiallySupported("fstatfs", Fstatfs, "Depends on the backing file system implementation.", nil), + 45: syscalls.Supported("truncate", Truncate), 46: syscalls.Supported("ftruncate", Ftruncate), 47: syscalls.PartiallySupported("fallocate", Fallocate, "Not all options are supported.", nil), 48: syscalls.Supported("faccessat", Faccessat), @@ -112,6 +115,7 @@ var ARM64 = &kernel.SyscallTable{ 76: syscalls.PartiallySupported("splice", Splice, "Stub implementation.", []string{"gvisor.dev/issue/138"}), // TODO(b/29354098) 77: syscalls.Supported("tee", Tee), 78: syscalls.Supported("readlinkat", Readlinkat), + 79: syscalls.Supported("fstatat", Fstatat), 80: syscalls.Supported("fstat", Fstat), 81: syscalls.PartiallySupported("sync", Sync, "Full data flush is not guaranteed at this time.", nil), 82: syscalls.PartiallySupported("fsync", Fsync, "Full data flush is not guaranteed at this time.", nil), @@ -254,6 +258,8 @@ var ARM64 = &kernel.SyscallTable{ 219: syscalls.Error("keyctl", syserror.EACCES, "Not available to user.", nil), 220: syscalls.PartiallySupported("clone", Clone, "Mount namespace (CLONE_NEWNS) not supported. Options CLONE_PARENT, CLONE_SYSVSEM not supported.", nil), 221: syscalls.Supported("execve", Execve), + 222: syscalls.PartiallySupported("mmap", Mmap, "Generally supported with exceptions. Options MAP_FIXED_NOREPLACE, MAP_SHARED_VALIDATE, MAP_SYNC MAP_GROWSDOWN, MAP_HUGETLB are not supported.", nil), + 223: syscalls.PartiallySupported("fadvise64", Fadvise64, "Not all options are supported.", nil), 224: syscalls.CapError("swapon", linux.CAP_SYS_ADMIN, "", nil), 225: syscalls.CapError("swapoff", linux.CAP_SYS_ADMIN, "", nil), 226: syscalls.Supported("mprotect", Mprotect), @@ -299,6 +305,8 @@ var ARM64 = &kernel.SyscallTable{ 282: syscalls.ErrorWithEvent("userfaultfd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/266"}), // TODO(b/118906345) 283: syscalls.ErrorWithEvent("membarrier", syserror.ENOSYS, "", []string{"gvisor.dev/issue/267"}), // TODO(gvisor.dev/issue/267) 284: syscalls.PartiallySupported("mlock2", Mlock2, "Stub implementation. The sandbox lacks appropriate permissions.", nil), + + // Syscalls after 284 are "backports" from versions of Linux after 4.4. 285: syscalls.ErrorWithEvent("copy_file_range", syserror.ENOSYS, "", nil), 286: syscalls.Supported("preadv2", Preadv2), 287: syscalls.PartiallySupported("pwritev2", Pwritev2, "Flag RWF_HIPRI is not supported.", nil), diff --git a/pkg/sentry/syscalls/linux/sys_stat.go b/pkg/sentry/syscalls/linux/sys_stat.go index 5556bc276..69b17b799 100644 --- a/pkg/sentry/syscalls/linux/sys_stat.go +++ b/pkg/sentry/syscalls/linux/sys_stat.go @@ -16,7 +16,6 @@ package linux import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -125,56 +124,6 @@ func fstat(t *kernel.Task, f *fs.File, statAddr usermem.Addr) error { return copyOutStat(t, statAddr, f.Dirent.Inode.StableAttr, uattr) } -// copyOutStat copies the attributes (sattr, uattr) to the struct stat at -// address dst in t's address space. It encodes the stat struct to bytes -// manually, as stat() is a very common syscall for many applications, and -// t.CopyObjectOut has noticeable performance impact due to its many slice -// allocations and use of reflection. -func copyOutStat(t *kernel.Task, dst usermem.Addr, sattr fs.StableAttr, uattr fs.UnstableAttr) error { - b := t.CopyScratchBuffer(int(linux.SizeOfStat))[:0] - - // Dev (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(sattr.DeviceID)) - // Ino (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(sattr.InodeID)) - // Nlink (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uattr.Links) - // Mode (uint32) - b = binary.AppendUint32(b, usermem.ByteOrder, sattr.Type.LinuxType()|uint32(uattr.Perms.LinuxMode())) - // UID (uint32) - b = binary.AppendUint32(b, usermem.ByteOrder, uint32(uattr.Owner.UID.In(t.UserNamespace()).OrOverflow())) - // GID (uint32) - b = binary.AppendUint32(b, usermem.ByteOrder, uint32(uattr.Owner.GID.In(t.UserNamespace()).OrOverflow())) - // Padding (uint32) - b = binary.AppendUint32(b, usermem.ByteOrder, 0) - // Rdev (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(linux.MakeDeviceID(sattr.DeviceFileMajor, sattr.DeviceFileMinor))) - // Size (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(uattr.Size)) - // Blksize (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(sattr.BlockSize)) - // Blocks (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(uattr.Usage/512)) - - // ATime - atime := uattr.AccessTime.Timespec() - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(atime.Sec)) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(atime.Nsec)) - - // MTime - mtime := uattr.ModificationTime.Timespec() - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(mtime.Sec)) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(mtime.Nsec)) - - // CTime - ctime := uattr.StatusChangeTime.Timespec() - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(ctime.Sec)) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(ctime.Nsec)) - - _, err := t.CopyOutBytes(dst, b) - return err -} - // Statx implements linux syscall statx(2). func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { fd := args[0].Int() diff --git a/pkg/sentry/syscalls/linux/sys_stat_amd64.go b/pkg/sentry/syscalls/linux/sys_stat_amd64.go new file mode 100644 index 000000000..58afb4a9a --- /dev/null +++ b/pkg/sentry/syscalls/linux/sys_stat_amd64.go @@ -0,0 +1,75 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +//+build amd64 + +package linux + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/usermem" +) + +// copyOutStat copies the attributes (sattr, uattr) to the struct stat at +// address dst in t's address space. It encodes the stat struct to bytes +// manually, as stat() is a very common syscall for many applications, and +// t.CopyObjectOut has noticeable performance impact due to its many slice +// allocations and use of reflection. +func copyOutStat(t *kernel.Task, dst usermem.Addr, sattr fs.StableAttr, uattr fs.UnstableAttr) error { + b := t.CopyScratchBuffer(int(linux.SizeOfStat))[:0] + + // Dev (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(sattr.DeviceID)) + // Ino (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(sattr.InodeID)) + // Nlink (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, uattr.Links) + // Mode (uint32) + b = binary.AppendUint32(b, usermem.ByteOrder, sattr.Type.LinuxType()|uint32(uattr.Perms.LinuxMode())) + // UID (uint32) + b = binary.AppendUint32(b, usermem.ByteOrder, uint32(uattr.Owner.UID.In(t.UserNamespace()).OrOverflow())) + // GID (uint32) + b = binary.AppendUint32(b, usermem.ByteOrder, uint32(uattr.Owner.GID.In(t.UserNamespace()).OrOverflow())) + // Padding (uint32) + b = binary.AppendUint32(b, usermem.ByteOrder, 0) + // Rdev (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(linux.MakeDeviceID(sattr.DeviceFileMajor, sattr.DeviceFileMinor))) + // Size (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(uattr.Size)) + // Blksize (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(sattr.BlockSize)) + // Blocks (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(uattr.Usage/512)) + + // ATime + atime := uattr.AccessTime.Timespec() + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(atime.Sec)) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(atime.Nsec)) + + // MTime + mtime := uattr.ModificationTime.Timespec() + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(mtime.Sec)) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(mtime.Nsec)) + + // CTime + ctime := uattr.StatusChangeTime.Timespec() + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(ctime.Sec)) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(ctime.Nsec)) + + _, err := t.CopyOutBytes(dst, b) + return err +} diff --git a/pkg/sentry/syscalls/linux/sys_stat_arm64.go b/pkg/sentry/syscalls/linux/sys_stat_arm64.go new file mode 100644 index 000000000..3e1251e0b --- /dev/null +++ b/pkg/sentry/syscalls/linux/sys_stat_arm64.go @@ -0,0 +1,77 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +//+build arm64 + +package linux + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/usermem" +) + +// copyOutStat copies the attributes (sattr, uattr) to the struct stat at +// address dst in t's address space. It encodes the stat struct to bytes +// manually, as stat() is a very common syscall for many applications, and +// t.CopyObjectOut has noticeable performance impact due to its many slice +// allocations and use of reflection. +func copyOutStat(t *kernel.Task, dst usermem.Addr, sattr fs.StableAttr, uattr fs.UnstableAttr) error { + b := t.CopyScratchBuffer(int(linux.SizeOfStat))[:0] + + // Dev (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(sattr.DeviceID)) + // Ino (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(sattr.InodeID)) + // Mode (uint32) + b = binary.AppendUint32(b, usermem.ByteOrder, sattr.Type.LinuxType()|uint32(uattr.Perms.LinuxMode())) + // Nlink (uint32) + b = binary.AppendUint32(b, usermem.ByteOrder, uint32(uattr.Links)) + // UID (uint32) + b = binary.AppendUint32(b, usermem.ByteOrder, uint32(uattr.Owner.UID.In(t.UserNamespace()).OrOverflow())) + // GID (uint32) + b = binary.AppendUint32(b, usermem.ByteOrder, uint32(uattr.Owner.GID.In(t.UserNamespace()).OrOverflow())) + // Rdev (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(linux.MakeDeviceID(sattr.DeviceFileMajor, sattr.DeviceFileMinor))) + // Padding (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, 0) + // Size (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(uattr.Size)) + // Blksize (uint32) + b = binary.AppendUint32(b, usermem.ByteOrder, uint32(sattr.BlockSize)) + // Padding (uint32) + b = binary.AppendUint32(b, usermem.ByteOrder, 0) + // Blocks (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(uattr.Usage/512)) + + // ATime + atime := uattr.AccessTime.Timespec() + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(atime.Sec)) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(atime.Nsec)) + + // MTime + mtime := uattr.ModificationTime.Timespec() + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(mtime.Sec)) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(mtime.Nsec)) + + // CTime + ctime := uattr.StatusChangeTime.Timespec() + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(ctime.Sec)) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(ctime.Nsec)) + + _, err := t.CopyOutBytes(dst, b) + return err +} -- cgit v1.2.3 From ff9960985848a48863c01f91acd5b34d3e83a9c5 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Fri, 17 Jan 2020 09:33:14 -0800 Subject: Add /proc/net/* files Updates #1195 PiperOrigin-RevId: 290285420 --- pkg/sentry/fsimpl/proc/BUILD | 2 +- pkg/sentry/fsimpl/proc/tasks.go | 1 + pkg/sentry/fsimpl/proc/tasks_net.go | 541 ++++++++++++++++++++++++++++--- pkg/sentry/fsimpl/proc/tasks_sys_test.go | 4 +- pkg/sentry/fsimpl/proc/tasks_test.go | 1 + 5 files changed, 499 insertions(+), 50 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/proc/BUILD b/pkg/sentry/fsimpl/proc/BUILD index e92564b5d..f69aa19c4 100644 --- a/pkg/sentry/fsimpl/proc/BUILD +++ b/pkg/sentry/fsimpl/proc/BUILD @@ -18,7 +18,6 @@ go_library( importpath = "gvisor.dev/gvisor/pkg/sentry/fsimpl/proc", deps = [ "//pkg/abi/linux", - "//pkg/binary", "//pkg/log", "//pkg/sentry/context", "//pkg/sentry/fs", @@ -37,6 +36,7 @@ go_library( "//pkg/sentry/usermem", "//pkg/sentry/vfs", "//pkg/syserror", + "//pkg/tcpip/header", ], ) diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index a97b1753a..5646c602a 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -67,6 +67,7 @@ func newTasksInode(inoGen InoGenerator, k *kernel.Kernel, pidns *kernel.PIDNames "sys": newSysDir(root, inoGen), "meminfo": newDentry(root, inoGen.NextIno(), 0444, &meminfoData{}), "mounts": kernfs.NewStaticSymlink(root, inoGen.NextIno(), "self/mounts"), + "net": newNetDir(root, inoGen, k), "stat": newDentry(root, inoGen.NextIno(), 0444, &statData{}), "uptime": newDentry(root, inoGen.NextIno(), 0444, &uptimeData{}), "version": newDentry(root, inoGen.NextIno(), 0444, &versionData{}), diff --git a/pkg/sentry/fsimpl/proc/tasks_net.go b/pkg/sentry/fsimpl/proc/tasks_net.go index 06dc43c26..3dbf3ba41 100644 --- a/pkg/sentry/fsimpl/proc/tasks_net.go +++ b/pkg/sentry/fsimpl/proc/tasks_net.go @@ -17,33 +17,88 @@ package proc import ( "bytes" "fmt" + "io" + "reflect" + "time" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/unix" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" - "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/tcpip/header" ) +func newNetDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) *kernfs.Dentry { + var contents map[string]*kernfs.Dentry + if stack := k.NetworkStack(); stack != nil { + const ( + arp = "IP address HW type Flags HW address Mask Device" + netlink = "sk Eth Pid Groups Rmem Wmem Dump Locks Drops Inode" + packet = "sk RefCnt Type Proto Iface R Rmem User Inode" + protocols = "protocol size sockets memory press maxhdr slab module cl co di ac io in de sh ss gs se re sp bi br ha uh gp em" + ptype = "Type Device Function" + upd6 = " sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode" + ) + psched := fmt.Sprintf("%08x %08x %08x %08x\n", uint64(time.Microsecond/time.Nanosecond), 64, 1000000, uint64(time.Second/time.Nanosecond)) + + contents = map[string]*kernfs.Dentry{ + "dev": newDentry(root, inoGen.NextIno(), 0444, &netDevData{stack: stack}), + "snmp": newDentry(root, inoGen.NextIno(), 0444, &netSnmpData{stack: stack}), + + // The following files are simple stubs until they are implemented in + // netstack, if the file contains a header the stub is just the header + // otherwise it is an empty file. + "arp": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(arp)), + "netlink": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(netlink)), + "netstat": newDentry(root, inoGen.NextIno(), 0444, &netStatData{}), + "packet": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(packet)), + "protocols": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(protocols)), + + // Linux sets psched values to: nsec per usec, psched tick in ns, 1000000, + // high res timer ticks per sec (ClockGetres returns 1ns resolution). + "psched": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(psched)), + "ptype": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(ptype)), + "route": newDentry(root, inoGen.NextIno(), 0444, &netRouteData{stack: stack}), + "tcp": newDentry(root, inoGen.NextIno(), 0444, &netTCPData{kernel: k}), + "udp": newDentry(root, inoGen.NextIno(), 0444, &netUDPData{kernel: k}), + "unix": newDentry(root, inoGen.NextIno(), 0444, &netUnixData{kernel: k}), + } + + if stack.SupportsIPv6() { + contents["if_inet6"] = newDentry(root, inoGen.NextIno(), 0444, &ifinet6{stack: stack}) + contents["ipv6_route"] = newDentry(root, inoGen.NextIno(), 0444, newStaticFile("")) + contents["tcp6"] = newDentry(root, inoGen.NextIno(), 0444, &netTCP6Data{kernel: k}) + contents["udp6"] = newDentry(root, inoGen.NextIno(), 0444, newStaticFile(upd6)) + } + } + + return kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, contents) +} + // ifinet6 implements vfs.DynamicBytesSource for /proc/net/if_inet6. // // +stateify savable type ifinet6 struct { - s inet.Stack + kernfs.DynamicBytesFile + + stack inet.Stack } -var _ vfs.DynamicBytesSource = (*ifinet6)(nil) +var _ dynamicInode = (*ifinet6)(nil) func (n *ifinet6) contents() []string { var lines []string - nics := n.s.Interfaces() - for id, naddrs := range n.s.InterfaceAddrs() { + nics := n.stack.Interfaces() + for id, naddrs := range n.stack.InterfaceAddrs() { nic, ok := nics[id] if !ok { // NIC was added after NICNames was called. We'll just ignore it. @@ -77,18 +132,20 @@ func (n *ifinet6) Generate(ctx context.Context, buf *bytes.Buffer) error { return nil } -// netDev implements vfs.DynamicBytesSource for /proc/net/dev. +// netDevData implements vfs.DynamicBytesSource for /proc/net/dev. // // +stateify savable -type netDev struct { - s inet.Stack +type netDevData struct { + kernfs.DynamicBytesFile + + stack inet.Stack } -var _ vfs.DynamicBytesSource = (*netDev)(nil) +var _ dynamicInode = (*netDevData)(nil) // Generate implements vfs.DynamicBytesSource.Generate. -func (n *netDev) Generate(ctx context.Context, buf *bytes.Buffer) error { - interfaces := n.s.Interfaces() +func (n *netDevData) Generate(ctx context.Context, buf *bytes.Buffer) error { + interfaces := n.stack.Interfaces() buf.WriteString("Inter-| Receive | Transmit\n") buf.WriteString(" face |bytes packets errs drop fifo frame compressed multicast|bytes packets errs drop fifo colls carrier compressed\n") @@ -96,7 +153,7 @@ func (n *netDev) Generate(ctx context.Context, buf *bytes.Buffer) error { // Implements the same format as // net/core/net-procfs.c:dev_seq_printf_stats. var stats inet.StatDev - if err := n.s.Statistics(&stats, i.Name); err != nil { + if err := n.stack.Statistics(&stats, i.Name); err != nil { log.Warningf("Failed to retrieve interface statistics for %v: %v", i.Name, err) continue } @@ -128,19 +185,21 @@ func (n *netDev) Generate(ctx context.Context, buf *bytes.Buffer) error { return nil } -// netUnix implements vfs.DynamicBytesSource for /proc/net/unix. +// netUnixData implements vfs.DynamicBytesSource for /proc/net/unix. // // +stateify savable -type netUnix struct { - k *kernel.Kernel +type netUnixData struct { + kernfs.DynamicBytesFile + + kernel *kernel.Kernel } -var _ vfs.DynamicBytesSource = (*netUnix)(nil) +var _ dynamicInode = (*netUnixData)(nil) // Generate implements vfs.DynamicBytesSource.Generate. -func (n *netUnix) Generate(ctx context.Context, buf *bytes.Buffer) error { +func (n *netUnixData) Generate(ctx context.Context, buf *bytes.Buffer) error { buf.WriteString("Num RefCount Protocol Flags Type St Inode Path\n") - for _, se := range n.k.ListSockets() { + for _, se := range n.kernel.ListSockets() { s := se.Sock.Get() if s == nil { log.Debugf("Couldn't resolve weakref %v in socket table, racing with destruction?", se.Sock) @@ -213,22 +272,72 @@ func (n *netUnix) Generate(ctx context.Context, buf *bytes.Buffer) error { return nil } -// netTCP implements vfs.DynamicBytesSource for /proc/net/tcp. -// -// +stateify savable -type netTCP struct { - k *kernel.Kernel +func networkToHost16(n uint16) uint16 { + // n is in network byte order, so is big-endian. The most-significant byte + // should be stored in the lower address. + // + // We manually inline binary.BigEndian.Uint16() because Go does not support + // non-primitive consts, so binary.BigEndian is a (mutable) var, so calls to + // binary.BigEndian.Uint16() require a read of binary.BigEndian and an + // interface method call, defeating inlining. + buf := [2]byte{byte(n >> 8 & 0xff), byte(n & 0xff)} + return usermem.ByteOrder.Uint16(buf[:]) } -var _ vfs.DynamicBytesSource = (*netTCP)(nil) +func writeInetAddr(w io.Writer, family int, i linux.SockAddr) { + switch family { + case linux.AF_INET: + var a linux.SockAddrInet + if i != nil { + a = *i.(*linux.SockAddrInet) + } + + // linux.SockAddrInet.Port is stored in the network byte order and is + // printed like a number in host byte order. Note that all numbers in host + // byte order are printed with the most-significant byte first when + // formatted with %X. See get_tcp4_sock() and udp4_format_sock() in Linux. + port := networkToHost16(a.Port) + + // linux.SockAddrInet.Addr is stored as a byte slice in big-endian order + // (i.e. most-significant byte in index 0). Linux represents this as a + // __be32 which is a typedef for an unsigned int, and is printed with + // %X. This means that for a little-endian machine, Linux prints the + // least-significant byte of the address first. To emulate this, we first + // invert the byte order for the address using usermem.ByteOrder.Uint32, + // which makes it have the equivalent encoding to a __be32 on a little + // endian machine. Note that this operation is a no-op on a big endian + // machine. Then similar to Linux, we format it with %X, which will print + // the most-significant byte of the __be32 address first, which is now + // actually the least-significant byte of the original address in + // linux.SockAddrInet.Addr on little endian machines, due to the conversion. + addr := usermem.ByteOrder.Uint32(a.Addr[:]) + + fmt.Fprintf(w, "%08X:%04X ", addr, port) + case linux.AF_INET6: + var a linux.SockAddrInet6 + if i != nil { + a = *i.(*linux.SockAddrInet6) + } -func (n *netTCP) Generate(ctx context.Context, buf *bytes.Buffer) error { + port := networkToHost16(a.Port) + addr0 := usermem.ByteOrder.Uint32(a.Addr[0:4]) + addr1 := usermem.ByteOrder.Uint32(a.Addr[4:8]) + addr2 := usermem.ByteOrder.Uint32(a.Addr[8:12]) + addr3 := usermem.ByteOrder.Uint32(a.Addr[12:16]) + fmt.Fprintf(w, "%08X%08X%08X%08X:%04X ", addr0, addr1, addr2, addr3, port) + } +} + +func commonGenerateTCP(ctx context.Context, buf *bytes.Buffer, k *kernel.Kernel, family int) error { + // t may be nil here if our caller is not part of a task goroutine. This can + // happen for example if we're here for "sentryctl cat". When t is nil, + // degrade gracefully and retrieve what we can. t := kernel.TaskFromContext(ctx) - buf.WriteString(" sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode \n") - for _, se := range n.k.ListSockets() { + + for _, se := range k.ListSockets() { s := se.Sock.Get() if s == nil { - log.Debugf("Couldn't resolve weakref %+v in socket table, racing with destruction?", se.Sock) + log.Debugf("Couldn't resolve weakref with ID %v in socket table, racing with destruction?", se.ID) continue } sfile := s.(*fs.File) @@ -236,7 +345,7 @@ func (n *netTCP) Generate(ctx context.Context, buf *bytes.Buffer) error { if !ok { panic(fmt.Sprintf("Found non-socket file in socket table: %+v", sfile)) } - if family, stype, _ := sops.Type(); !(family == linux.AF_INET && stype == linux.SOCK_STREAM) { + if fa, stype, _ := sops.Type(); !(family == fa && stype == linux.SOCK_STREAM) { s.DecRef() // Not tcp4 sockets. continue @@ -250,27 +359,23 @@ func (n *netTCP) Generate(ctx context.Context, buf *bytes.Buffer) error { // Field: sl; entry number. fmt.Fprintf(buf, "%4d: ", se.ID) - portBuf := make([]byte, 2) - // Field: local_adddress. - var localAddr linux.SockAddrInet - if local, _, err := sops.GetSockName(t); err == nil { - localAddr = *local.(*linux.SockAddrInet) + var localAddr linux.SockAddr + if t != nil { + if local, _, err := sops.GetSockName(t); err == nil { + localAddr = local + } } - binary.LittleEndian.PutUint16(portBuf, localAddr.Port) - fmt.Fprintf(buf, "%08X:%04X ", - binary.LittleEndian.Uint32(localAddr.Addr[:]), - portBuf) + writeInetAddr(buf, family, localAddr) // Field: rem_address. - var remoteAddr linux.SockAddrInet - if remote, _, err := sops.GetPeerName(t); err == nil { - remoteAddr = *remote.(*linux.SockAddrInet) + var remoteAddr linux.SockAddr + if t != nil { + if remote, _, err := sops.GetPeerName(t); err == nil { + remoteAddr = remote + } } - binary.LittleEndian.PutUint16(portBuf, remoteAddr.Port) - fmt.Fprintf(buf, "%08X:%04X ", - binary.LittleEndian.Uint32(remoteAddr.Addr[:]), - portBuf) + writeInetAddr(buf, family, remoteAddr) // Field: state; socket state. fmt.Fprintf(buf, "%02X ", sops.State()) @@ -293,7 +398,8 @@ func (n *netTCP) Generate(ctx context.Context, buf *bytes.Buffer) error { log.Warningf("Failed to retrieve unstable attr for socket file: %v", err) fmt.Fprintf(buf, "%5d ", 0) } else { - fmt.Fprintf(buf, "%5d ", uint32(uattr.Owner.UID.In(t.UserNamespace()).OrOverflow())) + creds := auth.CredentialsFromContext(ctx) + fmt.Fprintf(buf, "%5d ", uint32(uattr.Owner.UID.In(creds.UserNamespace).OrOverflow())) } // Field: timeout; number of unanswered 0-window probes. @@ -335,3 +441,344 @@ func (n *netTCP) Generate(ctx context.Context, buf *bytes.Buffer) error { return nil } + +// netTCPData implements vfs.DynamicBytesSource for /proc/net/tcp. +// +// +stateify savable +type netTCPData struct { + kernfs.DynamicBytesFile + + kernel *kernel.Kernel +} + +var _ dynamicInode = (*netTCPData)(nil) + +func (d *netTCPData) Generate(ctx context.Context, buf *bytes.Buffer) error { + buf.WriteString(" sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode \n") + return commonGenerateTCP(ctx, buf, d.kernel, linux.AF_INET) +} + +// netTCP6Data implements vfs.DynamicBytesSource for /proc/net/tcp6. +// +// +stateify savable +type netTCP6Data struct { + kernfs.DynamicBytesFile + + kernel *kernel.Kernel +} + +var _ dynamicInode = (*netTCP6Data)(nil) + +func (d *netTCP6Data) Generate(ctx context.Context, buf *bytes.Buffer) error { + buf.WriteString(" sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode\n") + return commonGenerateTCP(ctx, buf, d.kernel, linux.AF_INET6) +} + +// netUDPData implements vfs.DynamicBytesSource for /proc/net/udp. +// +// +stateify savable +type netUDPData struct { + kernfs.DynamicBytesFile + + kernel *kernel.Kernel +} + +var _ dynamicInode = (*netUDPData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (d *netUDPData) Generate(ctx context.Context, buf *bytes.Buffer) error { + // t may be nil here if our caller is not part of a task goroutine. This can + // happen for example if we're here for "sentryctl cat". When t is nil, + // degrade gracefully and retrieve what we can. + t := kernel.TaskFromContext(ctx) + + for _, se := range d.kernel.ListSockets() { + s := se.Sock.Get() + if s == nil { + log.Debugf("Couldn't resolve weakref with ID %v in socket table, racing with destruction?", se.ID) + continue + } + sfile := s.(*fs.File) + sops, ok := sfile.FileOperations.(socket.Socket) + if !ok { + panic(fmt.Sprintf("Found non-socket file in socket table: %+v", sfile)) + } + if family, stype, _ := sops.Type(); family != linux.AF_INET || stype != linux.SOCK_DGRAM { + s.DecRef() + // Not udp4 socket. + continue + } + + // For Linux's implementation, see net/ipv4/udp.c:udp4_format_sock(). + + // Field: sl; entry number. + fmt.Fprintf(buf, "%5d: ", se.ID) + + // Field: local_adddress. + var localAddr linux.SockAddrInet + if t != nil { + if local, _, err := sops.GetSockName(t); err == nil { + localAddr = *local.(*linux.SockAddrInet) + } + } + writeInetAddr(buf, linux.AF_INET, &localAddr) + + // Field: rem_address. + var remoteAddr linux.SockAddrInet + if t != nil { + if remote, _, err := sops.GetPeerName(t); err == nil { + remoteAddr = *remote.(*linux.SockAddrInet) + } + } + writeInetAddr(buf, linux.AF_INET, &remoteAddr) + + // Field: state; socket state. + fmt.Fprintf(buf, "%02X ", sops.State()) + + // Field: tx_queue, rx_queue; number of packets in the transmit and + // receive queue. Unimplemented. + fmt.Fprintf(buf, "%08X:%08X ", 0, 0) + + // Field: tr, tm->when. Always 0 for UDP. + fmt.Fprintf(buf, "%02X:%08X ", 0, 0) + + // Field: retrnsmt. Always 0 for UDP. + fmt.Fprintf(buf, "%08X ", 0) + + // Field: uid. + uattr, err := sfile.Dirent.Inode.UnstableAttr(ctx) + if err != nil { + log.Warningf("Failed to retrieve unstable attr for socket file: %v", err) + fmt.Fprintf(buf, "%5d ", 0) + } else { + creds := auth.CredentialsFromContext(ctx) + fmt.Fprintf(buf, "%5d ", uint32(uattr.Owner.UID.In(creds.UserNamespace).OrOverflow())) + } + + // Field: timeout. Always 0 for UDP. + fmt.Fprintf(buf, "%8d ", 0) + + // Field: inode. + fmt.Fprintf(buf, "%8d ", sfile.InodeID()) + + // Field: ref; reference count on the socket inode. Don't count the ref + // we obtain while deferencing the weakref to this socket. + fmt.Fprintf(buf, "%d ", sfile.ReadRefs()-1) + + // Field: Socket struct address. Redacted due to the same reason as + // the 'Num' field in /proc/net/unix, see netUnix.ReadSeqFileData. + fmt.Fprintf(buf, "%#016p ", (*socket.Socket)(nil)) + + // Field: drops; number of dropped packets. Unimplemented. + fmt.Fprintf(buf, "%d", 0) + + fmt.Fprintf(buf, "\n") + + s.DecRef() + } + return nil +} + +// netSnmpData implements vfs.DynamicBytesSource for /proc/net/snmp. +// +// +stateify savable +type netSnmpData struct { + kernfs.DynamicBytesFile + + stack inet.Stack +} + +var _ dynamicInode = (*netSnmpData)(nil) + +type snmpLine struct { + prefix string + header string +} + +var snmp = []snmpLine{ + { + prefix: "Ip", + header: "Forwarding DefaultTTL InReceives InHdrErrors InAddrErrors ForwDatagrams InUnknownProtos InDiscards InDelivers OutRequests OutDiscards OutNoRoutes ReasmTimeout ReasmReqds ReasmOKs ReasmFails FragOKs FragFails FragCreates", + }, + { + prefix: "Icmp", + header: "InMsgs InErrors InCsumErrors InDestUnreachs InTimeExcds InParmProbs InSrcQuenchs InRedirects InEchos InEchoReps InTimestamps InTimestampReps InAddrMasks InAddrMaskReps OutMsgs OutErrors OutDestUnreachs OutTimeExcds OutParmProbs OutSrcQuenchs OutRedirects OutEchos OutEchoReps OutTimestamps OutTimestampReps OutAddrMasks OutAddrMaskReps", + }, + { + prefix: "IcmpMsg", + }, + { + prefix: "Tcp", + header: "RtoAlgorithm RtoMin RtoMax MaxConn ActiveOpens PassiveOpens AttemptFails EstabResets CurrEstab InSegs OutSegs RetransSegs InErrs OutRsts InCsumErrors", + }, + { + prefix: "Udp", + header: "InDatagrams NoPorts InErrors OutDatagrams RcvbufErrors SndbufErrors InCsumErrors IgnoredMulti", + }, + { + prefix: "UdpLite", + header: "InDatagrams NoPorts InErrors OutDatagrams RcvbufErrors SndbufErrors InCsumErrors IgnoredMulti", + }, +} + +func toSlice(a interface{}) []uint64 { + v := reflect.Indirect(reflect.ValueOf(a)) + return v.Slice(0, v.Len()).Interface().([]uint64) +} + +func sprintSlice(s []uint64) string { + if len(s) == 0 { + return "" + } + r := fmt.Sprint(s) + return r[1 : len(r)-1] // Remove "[]" introduced by fmt of slice. +} + +// Generate implements vfs.DynamicBytesSource. +func (d *netSnmpData) Generate(ctx context.Context, buf *bytes.Buffer) error { + types := []interface{}{ + &inet.StatSNMPIP{}, + &inet.StatSNMPICMP{}, + nil, // TODO(gvisor.dev/issue/628): Support IcmpMsg stats. + &inet.StatSNMPTCP{}, + &inet.StatSNMPUDP{}, + &inet.StatSNMPUDPLite{}, + } + for i, stat := range types { + line := snmp[i] + if stat == nil { + fmt.Fprintf(buf, "%s:\n", line.prefix) + fmt.Fprintf(buf, "%s:\n", line.prefix) + continue + } + if err := d.stack.Statistics(stat, line.prefix); err != nil { + if err == syserror.EOPNOTSUPP { + log.Infof("Failed to retrieve %s of /proc/net/snmp: %v", line.prefix, err) + } else { + log.Warningf("Failed to retrieve %s of /proc/net/snmp: %v", line.prefix, err) + } + } + + fmt.Fprintf(buf, "%s: %s\n", line.prefix, line.header) + + if line.prefix == "Tcp" { + tcp := stat.(*inet.StatSNMPTCP) + // "Tcp" needs special processing because MaxConn is signed. RFC 2012. + fmt.Sprintf("%s: %s %d %s\n", line.prefix, sprintSlice(tcp[:3]), int64(tcp[3]), sprintSlice(tcp[4:])) + } else { + fmt.Sprintf("%s: %s\n", line.prefix, sprintSlice(toSlice(stat))) + } + } + return nil +} + +// netRouteData implements vfs.DynamicBytesSource for /proc/net/route. +// +// +stateify savable +type netRouteData struct { + kernfs.DynamicBytesFile + + stack inet.Stack +} + +var _ dynamicInode = (*netRouteData)(nil) + +// Generate implements vfs.DynamicBytesSource. +// See Linux's net/ipv4/fib_trie.c:fib_route_seq_show. +func (d *netRouteData) Generate(ctx context.Context, buf *bytes.Buffer) error { + fmt.Fprintf(buf, "%-127s\n", "Iface\tDestination\tGateway\tFlags\tRefCnt\tUse\tMetric\tMask\tMTU\tWindow\tIRTT") + + interfaces := d.stack.Interfaces() + for _, rt := range d.stack.RouteTable() { + // /proc/net/route only includes ipv4 routes. + if rt.Family != linux.AF_INET { + continue + } + + // /proc/net/route does not include broadcast or multicast routes. + if rt.Type == linux.RTN_BROADCAST || rt.Type == linux.RTN_MULTICAST { + continue + } + + iface, ok := interfaces[rt.OutputInterface] + if !ok || iface.Name == "lo" { + continue + } + + var ( + gw uint32 + prefix uint32 + flags = linux.RTF_UP + ) + if len(rt.GatewayAddr) == header.IPv4AddressSize { + flags |= linux.RTF_GATEWAY + gw = usermem.ByteOrder.Uint32(rt.GatewayAddr) + } + if len(rt.DstAddr) == header.IPv4AddressSize { + prefix = usermem.ByteOrder.Uint32(rt.DstAddr) + } + l := fmt.Sprintf( + "%s\t%08X\t%08X\t%04X\t%d\t%d\t%d\t%08X\t%d\t%d\t%d", + iface.Name, + prefix, + gw, + flags, + 0, // RefCnt. + 0, // Use. + 0, // Metric. + (uint32(1)< 0 { @@ -62,7 +62,7 @@ func TestIfinet6(t *testing.T) { "101112131415161718191a1b1c1d1e1f 02 80 00 00 eth1\n": {}, } - n := &ifinet6{s: s} + n := &ifinet6{stack: s} contents := n.contents() if len(contents) != len(want) { t.Errorf("Got len(n.contents()) = %d, want = %d", len(contents), len(want)) diff --git a/pkg/sentry/fsimpl/proc/tasks_test.go b/pkg/sentry/fsimpl/proc/tasks_test.go index 6b58c16b9..8eddf95e0 100644 --- a/pkg/sentry/fsimpl/proc/tasks_test.go +++ b/pkg/sentry/fsimpl/proc/tasks_test.go @@ -73,6 +73,7 @@ func checkTasksStaticFiles(gots []vfs.Dirent) ([]vfs.Dirent, error) { "loadavg": {Type: linux.DT_REG}, "meminfo": {Type: linux.DT_REG}, "mounts": {Type: linux.DT_LNK}, + "net": {Type: linux.DT_DIR}, "self": selfLink, "stat": {Type: linux.DT_REG}, "sys": {Type: linux.DT_DIR}, -- cgit v1.2.3 From 8e8d0f96f651ce161dfe6003d738dbda28f7cb0e Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Fri, 17 Jan 2020 10:39:24 -0800 Subject: Add /proc/[pid]/cgroups file Updates #1195 PiperOrigin-RevId: 290298266 --- pkg/sentry/fsimpl/proc/filesystem.go | 13 ++++++++++++- pkg/sentry/fsimpl/proc/subtasks.go | 18 ++++++++++-------- pkg/sentry/fsimpl/proc/task.go | 30 +++++++++++++++++++++++++----- pkg/sentry/fsimpl/proc/tasks.go | 10 ++++++++-- pkg/sentry/fsimpl/proc/tasks_test.go | 11 ++++++++++- 5 files changed, 65 insertions(+), 17 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/proc/filesystem.go b/pkg/sentry/fsimpl/proc/filesystem.go index e9cb7895f..f49819187 100644 --- a/pkg/sentry/fsimpl/proc/filesystem.go +++ b/pkg/sentry/fsimpl/proc/filesystem.go @@ -47,7 +47,12 @@ func (ft *procFSType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFile procfs := &kernfs.Filesystem{} procfs.VFSFilesystem().Init(vfsObj, procfs) - _, dentry := newTasksInode(procfs, k, pidns) + var data *InternalData + if opts.InternalData != nil { + data = opts.InternalData.(*InternalData) + } + + _, dentry := newTasksInode(procfs, k, pidns, data.Cgroups) return procfs.VFSFilesystem(), dentry.VFSDentry(), nil } @@ -78,3 +83,9 @@ var _ dynamicInode = (*staticFile)(nil) func newStaticFile(data string) *staticFile { return &staticFile{StaticData: vfs.StaticData{Data: data}} } + +// InternalData contains internal data passed in to the procfs mount via +// vfs.GetFilesystemOptions.InternalData. +type InternalData struct { + Cgroups map[string]string +} diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go index 8892c5a11..91eded415 100644 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -35,18 +35,20 @@ type subtasksInode struct { kernfs.InodeAttrs kernfs.OrderedChildren - task *kernel.Task - pidns *kernel.PIDNamespace - inoGen InoGenerator + task *kernel.Task + pidns *kernel.PIDNamespace + inoGen InoGenerator + cgroupControllers map[string]string } var _ kernfs.Inode = (*subtasksInode)(nil) -func newSubtasks(task *kernel.Task, pidns *kernel.PIDNamespace, inoGen InoGenerator) *kernfs.Dentry { +func newSubtasks(task *kernel.Task, pidns *kernel.PIDNamespace, inoGen InoGenerator, cgroupControllers map[string]string) *kernfs.Dentry { subInode := &subtasksInode{ - task: task, - pidns: pidns, - inoGen: inoGen, + task: task, + pidns: pidns, + inoGen: inoGen, + cgroupControllers: cgroupControllers, } // Note: credentials are overridden by taskOwnedInode. subInode.InodeAttrs.Init(task.Credentials(), inoGen.NextIno(), linux.ModeDirectory|0555) @@ -79,7 +81,7 @@ func (i *subtasksInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, e return nil, syserror.ENOENT } - subTaskDentry := newTaskInode(i.inoGen, subTask, i.pidns, false) + subTaskDentry := newTaskInode(i.inoGen, subTask, i.pidns, false, i.cgroupControllers) return subTaskDentry.VFSDentry(), nil } diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index 621c17cfe..a0580f20d 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -15,6 +15,7 @@ package proc import ( + "bytes" "fmt" "gvisor.dev/gvisor/pkg/abi/linux" @@ -42,7 +43,7 @@ type taskInode struct { var _ kernfs.Inode = (*taskInode)(nil) -func newTaskInode(inoGen InoGenerator, task *kernel.Task, pidns *kernel.PIDNamespace, isThreadGroup bool) *kernfs.Dentry { +func newTaskInode(inoGen InoGenerator, task *kernel.Task, pidns *kernel.PIDNamespace, isThreadGroup bool, cgroupControllers map[string]string) *kernfs.Dentry { contents := map[string]*kernfs.Dentry{ "auxv": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &auxvData{task: task}), "cmdline": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &cmdlineData{task: task, arg: cmdlineDataArg}), @@ -68,11 +69,11 @@ func newTaskInode(inoGen InoGenerator, task *kernel.Task, pidns *kernel.PIDNames "uid_map": newTaskOwnedFile(task, inoGen.NextIno(), 0644, &idMapData{task: task, gids: false}), } if isThreadGroup { - contents["task"] = newSubtasks(task, pidns, inoGen) + contents["task"] = newSubtasks(task, pidns, inoGen, cgroupControllers) + } + if len(cgroupControllers) > 0 { + contents["cgroup"] = newTaskOwnedFile(task, inoGen.NextIno(), 0444, newCgroupData(cgroupControllers)) } - //if len(p.cgroupControllers) > 0 { - // contents["cgroup"] = newCGroupInode(t, msrc, p.cgroupControllers) - //} taskInode := &taskInode{task: task} // Note: credentials are overridden by taskOwnedInode. @@ -227,3 +228,22 @@ func newNamespaceSymlink(task *kernel.Task, ino uint64, ns string) *kernfs.Dentr d.Init(taskInode) return d } + +// newCgroupData creates inode that shows cgroup information. +// From man 7 cgroups: "For each cgroup hierarchy of which the process is a +// member, there is one entry containing three colon-separated fields: +// hierarchy-ID:controller-list:cgroup-path" +func newCgroupData(controllers map[string]string) dynamicInode { + buf := bytes.Buffer{} + + // The hierarchy ids must be positive integers (for cgroup v1), but the + // exact number does not matter, so long as they are unique. We can + // just use a counter, but since linux sorts this file in descending + // order, we must count down to preserve this behavior. + i := len(controllers) + for name, dir := range controllers { + fmt.Fprintf(&buf, "%d:%s:%s\n", i, name, dir) + i-- + } + return newStaticFile(buf.String()) +} diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index 5646c602a..51f634716 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -54,11 +54,16 @@ type tasksInode struct { // Linux. So handle them outside of OrderedChildren. selfSymlink *vfs.Dentry threadSelfSymlink *vfs.Dentry + + // cgroupControllers is a map of controller name to directory in the + // cgroup hierarchy. These controllers are immutable and will be listed + // in /proc/pid/cgroup if not nil. + cgroupControllers map[string]string } var _ kernfs.Inode = (*tasksInode)(nil) -func newTasksInode(inoGen InoGenerator, k *kernel.Kernel, pidns *kernel.PIDNamespace) (*tasksInode, *kernfs.Dentry) { +func newTasksInode(inoGen InoGenerator, k *kernel.Kernel, pidns *kernel.PIDNamespace, cgroupControllers map[string]string) (*tasksInode, *kernfs.Dentry) { root := auth.NewRootCredentials(pidns.UserNamespace()) contents := map[string]*kernfs.Dentry{ "cpuinfo": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(cpuInfoData(k))), @@ -78,6 +83,7 @@ func newTasksInode(inoGen InoGenerator, k *kernel.Kernel, pidns *kernel.PIDNames inoGen: inoGen, selfSymlink: newSelfSymlink(root, inoGen.NextIno(), 0444, pidns).VFSDentry(), threadSelfSymlink: newThreadSelfSymlink(root, inoGen.NextIno(), 0444, pidns).VFSDentry(), + cgroupControllers: cgroupControllers, } inode.InodeAttrs.Init(root, inoGen.NextIno(), linux.ModeDirectory|0555) @@ -111,7 +117,7 @@ func (i *tasksInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, erro return nil, syserror.ENOENT } - taskDentry := newTaskInode(i.inoGen, task, i.pidns, true) + taskDentry := newTaskInode(i.inoGen, task, i.pidns, true, i.cgroupControllers) return taskDentry.VFSDentry(), nil } diff --git a/pkg/sentry/fsimpl/proc/tasks_test.go b/pkg/sentry/fsimpl/proc/tasks_test.go index 8eddf95e0..002d2f73b 100644 --- a/pkg/sentry/fsimpl/proc/tasks_test.go +++ b/pkg/sentry/fsimpl/proc/tasks_test.go @@ -87,6 +87,7 @@ func checkTasksStaticFiles(gots []vfs.Dirent) ([]vfs.Dirent, error) { func checkTaskStaticFiles(gots []vfs.Dirent) ([]vfs.Dirent, error) { wants := map[string]vfs.Dirent{ "auxv": {Type: linux.DT_REG}, + "cgroup": {Type: linux.DT_REG}, "cmdline": {Type: linux.DT_REG}, "comm": {Type: linux.DT_REG}, "environ": {Type: linux.DT_REG}, @@ -145,7 +146,15 @@ func setup() (context.Context, *vfs.VirtualFilesystem, vfs.VirtualDentry, error) vfsObj.MustRegisterFilesystemType("procfs", &procFSType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, }) - mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "procfs", &vfs.GetFilesystemOptions{}) + fsOpts := vfs.GetFilesystemOptions{ + InternalData: &InternalData{ + Cgroups: map[string]string{ + "cpuset": "/foo/cpuset", + "memory": "/foo/memory", + }, + }, + } + mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "procfs", &fsOpts) if err != nil { return nil, nil, vfs.VirtualDentry{}, fmt.Errorf("NewMountNamespace(): %v", err) } -- cgit v1.2.3 From 80d0f9304484897e4307c9701ddbfaacb925715d Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Fri, 17 Jan 2020 11:20:29 -0800 Subject: Fix data race in tty.queue.readableSize. We were setting queue.readable without holding the lock. PiperOrigin-RevId: 290306922 --- pkg/sentry/fs/tty/line_discipline.go | 4 +++- pkg/sentry/fs/tty/queue.go | 11 ++--------- 2 files changed, 5 insertions(+), 10 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/tty/line_discipline.go b/pkg/sentry/fs/tty/line_discipline.go index 894964260..9fe02657e 100644 --- a/pkg/sentry/fs/tty/line_discipline.go +++ b/pkg/sentry/fs/tty/line_discipline.go @@ -140,8 +140,10 @@ func (l *lineDiscipline) setTermios(ctx context.Context, io usermem.IO, args arc // buffer to its read buffer. Anything already in the read buffer is // now readable. if oldCanonEnabled && !l.termios.LEnabled(linux.ICANON) { - l.inQueue.pushWaitBuf(l) + l.inQueue.mu.Lock() + l.inQueue.pushWaitBufLocked(l) l.inQueue.readable = true + l.inQueue.mu.Unlock() l.slaveWaiter.Notify(waiter.EventIn) } diff --git a/pkg/sentry/fs/tty/queue.go b/pkg/sentry/fs/tty/queue.go index 8b5d4699a..21ccc6f32 100644 --- a/pkg/sentry/fs/tty/queue.go +++ b/pkg/sentry/fs/tty/queue.go @@ -197,16 +197,9 @@ func (q *queue) writeBytes(b []byte, l *lineDiscipline) { q.pushWaitBufLocked(l) } -// pushWaitBuf fills the queue's read buffer with data from the wait buffer. +// pushWaitBufLocked fills the queue's read buffer with data from the wait +// buffer. // -// Preconditions: -// * l.termiosMu must be held for reading. -func (q *queue) pushWaitBuf(l *lineDiscipline) int { - q.mu.Lock() - defer q.mu.Unlock() - return q.pushWaitBufLocked(l) -} - // Preconditions: // * l.termiosMu must be held for reading. // * q.mu must be locked. -- cgit v1.2.3 From f1a5178c589dbd9a1fe4f1b9fb943fbe64791b58 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Fri, 17 Jan 2020 14:20:00 -0800 Subject: Fix data race in MountNamespace.resolve. We must hold fs.renameMu to access Dirent.parent. PiperOrigin-RevId: 290340804 --- pkg/sentry/fs/mounts.go | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/mounts.go b/pkg/sentry/fs/mounts.go index db3dfd096..a9627a9d1 100644 --- a/pkg/sentry/fs/mounts.go +++ b/pkg/sentry/fs/mounts.go @@ -609,8 +609,11 @@ func (mns *MountNamespace) resolve(ctx context.Context, root, node *Dirent, rema } // Find the node; we resolve relative to the current symlink's parent. + renameMu.RLock() + parent := node.parent + renameMu.RUnlock() *remainingTraversals-- - d, err := mns.FindInode(ctx, root, node.parent, targetPath, remainingTraversals) + d, err := mns.FindInode(ctx, root, parent, targetPath, remainingTraversals) if err != nil { return nil, err } -- cgit v1.2.3 From 47d85257d3d015f0b9f7739c81af0ee9f510aaf5 Mon Sep 17 00:00:00 2001 From: Eyal Soha Date: Fri, 17 Jan 2020 18:24:39 -0800 Subject: Filter out received packets with a local source IP address. CERT Advisory CA-96.21 III. Solution advises that devices drop packets which could not have correctly arrived on the wire, such as receiving a packet where the source IP address is owned by the device that sent it. Fixes #1507 PiperOrigin-RevId: 290378240 --- pkg/sentry/socket/netstack/netstack.go | 15 +++++----- pkg/sentry/socket/netstack/stack.go | 38 ++++++++++++------------- pkg/tcpip/stack/nic.go | 14 +++++++-- pkg/tcpip/tcpip.go | 10 +++++-- pkg/tcpip/transport/udp/udp_test.go | 52 ++++++++++++++++++++++++++++++++-- 5 files changed, 95 insertions(+), 34 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index d2f7e987d..fec575357 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -138,13 +138,14 @@ var Metrics = tcpip.Stats{ }, }, IP: tcpip.IPStats{ - PacketsReceived: mustCreateMetric("/netstack/ip/packets_received", "Total number of IP packets received from the link layer in nic.DeliverNetworkPacket."), - InvalidAddressesReceived: mustCreateMetric("/netstack/ip/invalid_addresses_received", "Total number of IP packets received with an unknown or invalid destination address."), - PacketsDelivered: mustCreateMetric("/netstack/ip/packets_delivered", "Total number of incoming IP packets that are successfully delivered to the transport layer via HandlePacket."), - PacketsSent: mustCreateMetric("/netstack/ip/packets_sent", "Total number of IP packets sent via WritePacket."), - OutgoingPacketErrors: mustCreateMetric("/netstack/ip/outgoing_packet_errors", "Total number of IP packets which failed to write to a link-layer endpoint."), - MalformedPacketsReceived: mustCreateMetric("/netstack/ip/malformed_packets_received", "Total number of IP packets which failed IP header validation checks."), - MalformedFragmentsReceived: mustCreateMetric("/netstack/ip/malformed_fragments_received", "Total number of IP fragments which failed IP fragment validation checks."), + PacketsReceived: mustCreateMetric("/netstack/ip/packets_received", "Total number of IP packets received from the link layer in nic.DeliverNetworkPacket."), + InvalidDestinationAddressesReceived: mustCreateMetric("/netstack/ip/invalid_addresses_received", "Total number of IP packets received with an unknown or invalid destination address."), + InvalidSourceAddressesReceived: mustCreateMetric("/netstack/ip/invalid_source_addresses_received", "Total number of IP packets received with an unknown or invalid source address."), + PacketsDelivered: mustCreateMetric("/netstack/ip/packets_delivered", "Total number of incoming IP packets that are successfully delivered to the transport layer via HandlePacket."), + PacketsSent: mustCreateMetric("/netstack/ip/packets_sent", "Total number of IP packets sent via WritePacket."), + OutgoingPacketErrors: mustCreateMetric("/netstack/ip/outgoing_packet_errors", "Total number of IP packets which failed to write to a link-layer endpoint."), + MalformedPacketsReceived: mustCreateMetric("/netstack/ip/malformed_packets_received", "Total number of IP packets which failed IP header validation checks."), + MalformedFragmentsReceived: mustCreateMetric("/netstack/ip/malformed_fragments_received", "Total number of IP fragments which failed IP fragment validation checks."), }, TCP: tcpip.TCPStats{ ActiveConnectionOpenings: mustCreateMetric("/netstack/tcp/active_connection_openings", "Number of connections opened successfully via Connect."), diff --git a/pkg/sentry/socket/netstack/stack.go b/pkg/sentry/socket/netstack/stack.go index a0db2d4fd..31ea66eca 100644 --- a/pkg/sentry/socket/netstack/stack.go +++ b/pkg/sentry/socket/netstack/stack.go @@ -148,25 +148,25 @@ func (s *Stack) Statistics(stat interface{}, arg string) error { case *inet.StatSNMPIP: ip := Metrics.IP *stats = inet.StatSNMPIP{ - 0, // TODO(gvisor.dev/issue/969): Support Ip/Forwarding. - 0, // TODO(gvisor.dev/issue/969): Support Ip/DefaultTTL. - ip.PacketsReceived.Value(), // InReceives. - 0, // TODO(gvisor.dev/issue/969): Support Ip/InHdrErrors. - ip.InvalidAddressesReceived.Value(), // InAddrErrors. - 0, // TODO(gvisor.dev/issue/969): Support Ip/ForwDatagrams. - 0, // TODO(gvisor.dev/issue/969): Support Ip/InUnknownProtos. - 0, // TODO(gvisor.dev/issue/969): Support Ip/InDiscards. - ip.PacketsDelivered.Value(), // InDelivers. - ip.PacketsSent.Value(), // OutRequests. - ip.OutgoingPacketErrors.Value(), // OutDiscards. - 0, // TODO(gvisor.dev/issue/969): Support Ip/OutNoRoutes. - 0, // TODO(gvisor.dev/issue/969): Support Ip/ReasmTimeout. - 0, // TODO(gvisor.dev/issue/969): Support Ip/ReasmReqds. - 0, // TODO(gvisor.dev/issue/969): Support Ip/ReasmOKs. - 0, // TODO(gvisor.dev/issue/969): Support Ip/ReasmFails. - 0, // TODO(gvisor.dev/issue/969): Support Ip/FragOKs. - 0, // TODO(gvisor.dev/issue/969): Support Ip/FragFails. - 0, // TODO(gvisor.dev/issue/969): Support Ip/FragCreates. + 0, // TODO(gvisor.dev/issue/969): Support Ip/Forwarding. + 0, // TODO(gvisor.dev/issue/969): Support Ip/DefaultTTL. + ip.PacketsReceived.Value(), // InReceives. + 0, // TODO(gvisor.dev/issue/969): Support Ip/InHdrErrors. + ip.InvalidDestinationAddressesReceived.Value(), // InAddrErrors. + 0, // TODO(gvisor.dev/issue/969): Support Ip/ForwDatagrams. + 0, // TODO(gvisor.dev/issue/969): Support Ip/InUnknownProtos. + 0, // TODO(gvisor.dev/issue/969): Support Ip/InDiscards. + ip.PacketsDelivered.Value(), // InDelivers. + ip.PacketsSent.Value(), // OutRequests. + ip.OutgoingPacketErrors.Value(), // OutDiscards. + 0, // TODO(gvisor.dev/issue/969): Support Ip/OutNoRoutes. + 0, // TODO(gvisor.dev/issue/969): Support Ip/ReasmTimeout. + 0, // TODO(gvisor.dev/issue/969): Support Ip/ReasmReqds. + 0, // TODO(gvisor.dev/issue/969): Support Ip/ReasmOKs. + 0, // TODO(gvisor.dev/issue/969): Support Ip/ReasmFails. + 0, // TODO(gvisor.dev/issue/969): Support Ip/FragOKs. + 0, // TODO(gvisor.dev/issue/969): Support Ip/FragFails. + 0, // TODO(gvisor.dev/issue/969): Support Ip/FragCreates. } case *inet.StatSNMPICMP: in := Metrics.ICMP.V4PacketsReceived.ICMPv4PacketStats diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 53abf29e5..4afe7b744 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -984,7 +984,7 @@ func handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, // DeliverNetworkPacket finds the appropriate network protocol endpoint and // hands the packet over for further processing. This function is called when -// the NIC receives a packet from the physical interface. +// the NIC receives a packet from the link endpoint. // Note that the ownership of the slice backing vv is retained by the caller. // This rule applies only to the slice itself, not to the items of the slice; // the ownership of the items is not retained by the caller. @@ -1029,6 +1029,14 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link src, dst := netProto.ParseAddresses(pkt.Data.First()) + if n.stack.handleLocal && !n.isLoopback() && n.getRef(protocol, src) != nil { + // The source address is one of our own, so we never should have gotten a + // packet like this unless handleLocal is false. Loopback also calls this + // function even though the packets didn't come from the physical interface + // so don't drop those. + n.stack.stats.IP.InvalidSourceAddressesReceived.Increment() + return + } if ref := n.getRef(protocol, dst); ref != nil { handlePacket(protocol, dst, src, linkEP.LinkAddress(), remote, ref, pkt) return @@ -1041,7 +1049,7 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link if n.stack.Forwarding() { r, err := n.stack.FindRoute(0, "", dst, protocol, false /* multicastLoop */) if err != nil { - n.stack.stats.IP.InvalidAddressesReceived.Increment() + n.stack.stats.IP.InvalidDestinationAddressesReceived.Increment() return } defer r.Release() @@ -1079,7 +1087,7 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link // If a packet socket handled the packet, don't treat it as invalid. if len(packetEPs) == 0 { - n.stack.stats.IP.InvalidAddressesReceived.Increment() + n.stack.stats.IP.InvalidDestinationAddressesReceived.Increment() } } diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index b7813cbc0..6243762e3 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -903,9 +903,13 @@ type IPStats struct { // link layer in nic.DeliverNetworkPacket. PacketsReceived *StatCounter - // InvalidAddressesReceived is the total number of IP packets received - // with an unknown or invalid destination address. - InvalidAddressesReceived *StatCounter + // InvalidDestinationAddressesReceived is the total number of IP packets + // received with an unknown or invalid destination address. + InvalidDestinationAddressesReceived *StatCounter + + // InvalidSourceAddressesReceived is the total number of IP packets received + // with a source address that should never have been received on the wire. + InvalidSourceAddressesReceived *StatCounter // PacketsDelivered is the total number of incoming IP packets that // are successfully delivered to the transport layer via HandlePacket. diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index ee9d10555..51bb61167 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -274,11 +274,16 @@ type testContext struct { func newDualTestContext(t *testing.T, mtu uint32) *testContext { t.Helper() - - s := stack.New(stack.Options{ + return newDualTestContextWithOptions(t, mtu, stack.Options{ NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, }) +} + +func newDualTestContextWithOptions(t *testing.T, mtu uint32, options stack.Options) *testContext { + t.Helper() + + s := stack.New(options) ep := channel.New(256, mtu, "") wep := stack.LinkEndpoint(ep) @@ -763,6 +768,49 @@ func TestV6ReadOnV6(t *testing.T) { testRead(c, unicastV6) } +// TestV4ReadSelfSource checks that packets coming from a local IP address are +// correctly dropped when handleLocal is true and not otherwise. +func TestV4ReadSelfSource(t *testing.T) { + for _, tt := range []struct { + name string + handleLocal bool + wantErr *tcpip.Error + wantInvalidSource uint64 + }{ + {"HandleLocal", false, nil, 0}, + {"NoHandleLocal", true, tcpip.ErrWouldBlock, 1}, + } { + t.Run(tt.name, func(t *testing.T) { + c := newDualTestContextWithOptions(t, defaultMTU, stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + HandleLocal: tt.handleLocal, + }) + defer c.cleanup() + + c.createEndpointForFlow(unicastV4) + + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + t.Fatalf("Bind failed: %s", err) + } + + payload := newPayload() + h := unicastV4.header4Tuple(incoming) + h.srcAddr = h.dstAddr + + c.injectV4Packet(payload, &h, true /* valid */) + + if got := c.s.Stats().IP.InvalidSourceAddressesReceived.Value(); got != tt.wantInvalidSource { + t.Errorf("c.s.Stats().IP.InvalidSourceAddressesReceived got %d, want %d", got, tt.wantInvalidSource) + } + + if _, _, err := c.ep.Read(nil); err != tt.wantErr { + t.Errorf("c.ep.Read() got error %v, want %v", err, tt.wantErr) + } + }) + } +} + func TestV4ReadOnV4(t *testing.T) { c := newDualTestContext(t, defaultMTU) defer c.cleanup() -- cgit v1.2.3 From 10401599e104d90644a220c1cce3e4c2f224f0b3 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Sat, 18 Jan 2020 09:32:39 -0800 Subject: Include the cgroup name in the superblock options in /proc/self/mountinfo. Java 11 parses /proc/self/mountinfo for cgroup information. Java 11.0.4 uses the mount path to determine what cgroups existed, but Java 11.0.5 reads the cgroup names from the superblock options. This CL adds the cgroup name to the superblock options if the filesystem type is "cgroup". Since gVisor doesn't actually support cgroups yet, we just infer the cgroup name from the path. PiperOrigin-RevId: 290434323 --- pkg/sentry/fs/proc/mounts.go | 28 +++++++++++++++++++++++++--- 1 file changed, 25 insertions(+), 3 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/proc/mounts.go b/pkg/sentry/fs/proc/mounts.go index 5aedae799..d4efc86e0 100644 --- a/pkg/sentry/fs/proc/mounts.go +++ b/pkg/sentry/fs/proc/mounts.go @@ -18,6 +18,7 @@ import ( "bytes" "fmt" "sort" + "strings" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" @@ -146,14 +147,35 @@ func (mif *mountInfoFile) ReadSeqFileData(ctx context.Context, handle seqfile.Se // (10) Mount source: filesystem-specific information or "none". fmt.Fprintf(&buf, "none ") - // (11) Superblock options. Only "ro/rw" is supported for now, - // and is the same as the filesystem option. - fmt.Fprintf(&buf, "%s\n", opts) + // (11) Superblock options, and final newline. + fmt.Fprintf(&buf, "%s\n", superBlockOpts(mountPath, mroot.Inode.MountSource)) }) return []seqfile.SeqData{{Buf: buf.Bytes(), Handle: (*mountInfoFile)(nil)}}, 0 } +func superBlockOpts(mountPath string, msrc *fs.MountSource) string { + // gVisor doesn't (yet) have a concept of super block options, so we + // use the ro/rw bit from the mount flag. + opts := "rw" + if msrc.Flags.ReadOnly { + opts = "ro" + } + + // NOTE(b/147673608): If the mount is a cgroup, we also need to include + // the cgroup name in the options. For now we just read that from the + // path. + // TODO(gvisor.dev/issues/190): Once gVisor has full cgroup support, we + // should get this value from the cgroup itself, and not rely on the + // path. + if msrc.FilesystemType == "cgroup" { + splitPath := strings.Split(mountPath, "/") + cgroupType := splitPath[len(splitPath)-1] + opts += "," + cgroupType + } + return opts +} + // mountsFile is used to implement /proc/[pid]/mounts. // // +stateify savable -- cgit v1.2.3 From c0e39a8271198f10407009ec1994e5d9efac796c Mon Sep 17 00:00:00 2001 From: Haibo Xu Date: Fri, 20 Dec 2019 06:00:30 +0000 Subject: Enable uname syscall support on arm64. Signed-off-by: Haibo Xu Change-Id: I206f38416a64d7c6a8531d8eb305c6ea239616b8 --- pkg/sentry/syscalls/linux/sys_utsname.go | 12 ++++++++++-- 1 file changed, 10 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/sys_utsname.go b/pkg/sentry/syscalls/linux/sys_utsname.go index 748e8dd8d..a393e28c1 100644 --- a/pkg/sentry/syscalls/linux/sys_utsname.go +++ b/pkg/sentry/syscalls/linux/sys_utsname.go @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -// +build amd64 +// +build amd64 arm64 package linux @@ -35,7 +35,15 @@ func Uname(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall copy(u.Nodename[:], uts.HostName()) copy(u.Release[:], version.Release) copy(u.Version[:], version.Version) - copy(u.Machine[:], "x86_64") // build tag above. + // build tag above. + switch t.SyscallTable().Arch { + case arch.AMD64: + copy(u.Machine[:], "x86_64") + case arch.ARM64: + copy(u.Machine[:], "aarch64") + default: + copy(u.Machine[:], "unknown") + } copy(u.Domainname[:], uts.DomainName()) // Copy out the result. -- cgit v1.2.3 From 2ba6198851dc1e293295d7cadf8c0ae456b68beb Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 21 Jan 2020 12:41:50 -0800 Subject: Add syscalls for lgetxattr, fgetxattr, lsetxattr, and fsetxattr. Note that these simply will use the same logic as getxattr and setxattr, which is not yet implemented for most filesystems. PiperOrigin-RevId: 290800960 --- pkg/sentry/syscalls/linux/linux64_amd64.go | 8 +- pkg/sentry/syscalls/linux/linux64_arm64.go | 8 +- pkg/sentry/syscalls/linux/sys_xattr.go | 136 +++++++++++++++++++++-------- test/syscalls/linux/xattr.cc | 41 +++++++++ 4 files changed, 150 insertions(+), 43 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/linux64_amd64.go b/pkg/sentry/syscalls/linux/linux64_amd64.go index 6b2920900..c76771a54 100644 --- a/pkg/sentry/syscalls/linux/linux64_amd64.go +++ b/pkg/sentry/syscalls/linux/linux64_amd64.go @@ -229,11 +229,11 @@ var AMD64 = &kernel.SyscallTable{ 186: syscalls.Supported("gettid", Gettid), 187: syscalls.Supported("readahead", Readahead), 188: syscalls.PartiallySupported("setxattr", SetXattr, "Only supported for tmpfs.", nil), - 189: syscalls.Error("lsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 190: syscalls.Error("fsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 189: syscalls.PartiallySupported("lsetxattr", LSetXattr, "Only supported for tmpfs.", nil), + 190: syscalls.PartiallySupported("fsetxattr", FSetXattr, "Only supported for tmpfs.", nil), 191: syscalls.PartiallySupported("getxattr", GetXattr, "Only supported for tmpfs.", nil), - 192: syscalls.ErrorWithEvent("lgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 193: syscalls.ErrorWithEvent("fgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 192: syscalls.PartiallySupported("lgetxattr", LGetXattr, "Only supported for tmpfs.", nil), + 193: syscalls.PartiallySupported("fgetxattr", FGetXattr, "Only supported for tmpfs.", nil), 194: syscalls.ErrorWithEvent("listxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 195: syscalls.ErrorWithEvent("llistxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 196: syscalls.ErrorWithEvent("flistxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), diff --git a/pkg/sentry/syscalls/linux/linux64_arm64.go b/pkg/sentry/syscalls/linux/linux64_arm64.go index c9629f6f3..d3587fda6 100644 --- a/pkg/sentry/syscalls/linux/linux64_arm64.go +++ b/pkg/sentry/syscalls/linux/linux64_arm64.go @@ -42,11 +42,11 @@ var ARM64 = &kernel.SyscallTable{ 3: syscalls.PartiallySupported("io_cancel", IoCancel, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), 4: syscalls.PartiallySupported("io_getevents", IoGetevents, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), 5: syscalls.PartiallySupported("setxattr", SetXattr, "Only supported for tmpfs.", nil), - 6: syscalls.Error("lsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 7: syscalls.Error("fsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 6: syscalls.PartiallySupported("lsetxattr", LSetXattr, "Only supported for tmpfs.", nil), + 7: syscalls.PartiallySupported("fsetxattr", FSetXattr, "Only supported for tmpfs.", nil), 8: syscalls.PartiallySupported("getxattr", GetXattr, "Only supported for tmpfs.", nil), - 9: syscalls.ErrorWithEvent("lgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 10: syscalls.ErrorWithEvent("fgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 9: syscalls.PartiallySupported("lgetxattr", LGetXattr, "Only supported for tmpfs.", nil), + 10: syscalls.PartiallySupported("fgetxattr", FGetXattr, "Only supported for tmpfs.", nil), 11: syscalls.ErrorWithEvent("listxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 12: syscalls.ErrorWithEvent("llistxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 13: syscalls.ErrorWithEvent("flistxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), diff --git a/pkg/sentry/syscalls/linux/sys_xattr.go b/pkg/sentry/syscalls/linux/sys_xattr.go index 23d20da6f..e35c077d6 100644 --- a/pkg/sentry/syscalls/linux/sys_xattr.go +++ b/pkg/sentry/syscalls/linux/sys_xattr.go @@ -27,6 +27,40 @@ import ( // GetXattr implements linux syscall getxattr(2). func GetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + return getXattrFromPath(t, args, true) +} + +// LGetXattr implements linux syscall lgetxattr(2). +func LGetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + return getXattrFromPath(t, args, false) +} + +// FGetXattr implements linux syscall fgetxattr(2). +func FGetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + nameAddr := args[1].Pointer() + valueAddr := args[2].Pointer() + size := uint64(args[3].SizeT()) + + // TODO(b/113957122): Return EBADF if the fd was opened with O_PATH. + f := t.GetFile(fd) + if f == nil { + return 0, nil, syserror.EBADF + } + defer f.DecRef() + + n, value, err := getXattr(t, f.Dirent, nameAddr, size) + if err != nil { + return 0, nil, err + } + + if _, err := t.CopyOutBytes(valueAddr, []byte(value)); err != nil { + return 0, nil, err + } + return uintptr(n), nil, nil +} + +func getXattrFromPath(t *kernel.Task, args arch.SyscallArguments, resolveSymlink bool) (uintptr, *kernel.SyscallControl, error) { pathAddr := args[0].Pointer() nameAddr := args[1].Pointer() valueAddr := args[2].Pointer() @@ -38,29 +72,17 @@ func GetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc } valueLen := 0 - err = fileOpOn(t, linux.AT_FDCWD, path, true /* resolve */, func(root *fs.Dirent, d *fs.Dirent, _ uint) error { - // If getxattr(2) is called with size 0, the size of the value will be - // returned successfully even if it is nonzero. In that case, we need to - // retrieve the entire attribute value so we can return the correct size. - requestedSize := size - if size == 0 || size > linux.XATTR_SIZE_MAX { - requestedSize = linux.XATTR_SIZE_MAX + err = fileOpOn(t, linux.AT_FDCWD, path, resolveSymlink, func(root *fs.Dirent, d *fs.Dirent, _ uint) error { + if dirPath && !fs.IsDir(d.Inode.StableAttr) { + return syserror.ENOTDIR } - value, err := getXattr(t, d, dirPath, nameAddr, uint64(requestedSize)) + n, value, err := getXattr(t, d, nameAddr, size) + valueLen = n if err != nil { return err } - valueLen = len(value) - if uint64(valueLen) > requestedSize { - return syserror.ERANGE - } - - // Skip copying out the attribute value if size is 0. - if size == 0 { - return nil - } _, err = t.CopyOutBytes(valueAddr, []byte(value)) return err }) @@ -71,29 +93,73 @@ func GetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc } // getXattr implements getxattr(2) from the given *fs.Dirent. -func getXattr(t *kernel.Task, d *fs.Dirent, dirPath bool, nameAddr usermem.Addr, size uint64) (string, error) { - if dirPath && !fs.IsDir(d.Inode.StableAttr) { - return "", syserror.ENOTDIR - } - +func getXattr(t *kernel.Task, d *fs.Dirent, nameAddr usermem.Addr, size uint64) (int, string, error) { if err := checkXattrPermissions(t, d.Inode, fs.PermMask{Read: true}); err != nil { - return "", err + return 0, "", err } name, err := copyInXattrName(t, nameAddr) if err != nil { - return "", err + return 0, "", err } if !strings.HasPrefix(name, linux.XATTR_USER_PREFIX) { - return "", syserror.EOPNOTSUPP + return 0, "", syserror.EOPNOTSUPP } - return d.Inode.GetXattr(t, name, size) + // If getxattr(2) is called with size 0, the size of the value will be + // returned successfully even if it is nonzero. In that case, we need to + // retrieve the entire attribute value so we can return the correct size. + requestedSize := size + if size == 0 || size > linux.XATTR_SIZE_MAX { + requestedSize = linux.XATTR_SIZE_MAX + } + + value, err := d.Inode.GetXattr(t, name, requestedSize) + if err != nil { + return 0, "", err + } + n := len(value) + if uint64(n) > requestedSize { + return 0, "", syserror.ERANGE + } + + // Don't copy out the attribute value if size is 0. + if size == 0 { + return n, "", nil + } + return n, value, nil } // SetXattr implements linux syscall setxattr(2). func SetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + return setXattrFromPath(t, args, true) +} + +// LSetXattr implements linux syscall lsetxattr(2). +func LSetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + return setXattrFromPath(t, args, false) +} + +// FSetXattr implements linux syscall fsetxattr(2). +func FSetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + nameAddr := args[1].Pointer() + valueAddr := args[2].Pointer() + size := uint64(args[3].SizeT()) + flags := args[4].Uint() + + // TODO(b/113957122): Return EBADF if the fd was opened with O_PATH. + f := t.GetFile(fd) + if f == nil { + return 0, nil, syserror.EBADF + } + defer f.DecRef() + + return 0, nil, setXattr(t, f.Dirent, nameAddr, valueAddr, uint64(size), flags) +} + +func setXattrFromPath(t *kernel.Task, args arch.SyscallArguments, resolveSymlink bool) (uintptr, *kernel.SyscallControl, error) { pathAddr := args[0].Pointer() nameAddr := args[1].Pointer() valueAddr := args[2].Pointer() @@ -105,19 +171,19 @@ func SetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc return 0, nil, err } - if flags&^(linux.XATTR_CREATE|linux.XATTR_REPLACE) != 0 { - return 0, nil, syserror.EINVAL - } + return 0, nil, fileOpOn(t, linux.AT_FDCWD, path, resolveSymlink, func(root *fs.Dirent, d *fs.Dirent, _ uint) error { + if dirPath && !fs.IsDir(d.Inode.StableAttr) { + return syserror.ENOTDIR + } - return 0, nil, fileOpOn(t, linux.AT_FDCWD, path, true /* resolve */, func(root *fs.Dirent, d *fs.Dirent, _ uint) error { - return setXattr(t, d, dirPath, nameAddr, valueAddr, uint64(size), flags) + return setXattr(t, d, nameAddr, valueAddr, uint64(size), flags) }) } // setXattr implements setxattr(2) from the given *fs.Dirent. -func setXattr(t *kernel.Task, d *fs.Dirent, dirPath bool, nameAddr, valueAddr usermem.Addr, size uint64, flags uint32) error { - if dirPath && !fs.IsDir(d.Inode.StableAttr) { - return syserror.ENOTDIR +func setXattr(t *kernel.Task, d *fs.Dirent, nameAddr, valueAddr usermem.Addr, size uint64, flags uint32) error { + if flags&^(linux.XATTR_CREATE|linux.XATTR_REPLACE) != 0 { + return syserror.EINVAL } if err := checkXattrPermissions(t, d.Inode, fs.PermMask{Write: true}); err != nil { @@ -133,7 +199,7 @@ func setXattr(t *kernel.Task, d *fs.Dirent, dirPath bool, nameAddr, valueAddr us return syserror.E2BIG } buf := make([]byte, size) - if _, err = t.CopyInBytes(valueAddr, buf); err != nil { + if _, err := t.CopyInBytes(valueAddr, buf); err != nil { return err } value := string(buf) diff --git a/test/syscalls/linux/xattr.cc b/test/syscalls/linux/xattr.cc index b3bc3463e..e77c355d7 100644 --- a/test/syscalls/linux/xattr.cc +++ b/test/syscalls/linux/xattr.cc @@ -26,6 +26,7 @@ #include "gtest/gtest.h" #include "test/syscalls/linux/file_base.h" #include "test/util/capability_util.h" +#include "test/util/file_descriptor.h" #include "test/util/posix_error.h" #include "test/util/temp_path.h" #include "test/util/test_util.h" @@ -414,6 +415,46 @@ TEST_F(XattrTest, GetxattrNonexistentName) { EXPECT_THAT(getxattr(path, name, nullptr, 0), SyscallFailsWithErrno(ENODATA)); } +TEST_F(XattrTest, LGetSetxattrOnSymlink) { + TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + TempPath link = ASSERT_NO_ERRNO_AND_VALUE( + TempPath::CreateSymlinkTo(dir.path(), test_file_name_)); + + EXPECT_THAT(lsetxattr(link.path().c_str(), nullptr, nullptr, 0, 0), + SyscallFailsWithErrno(EPERM)); + EXPECT_THAT(lgetxattr(link.path().c_str(), nullptr, nullptr, 0), + SyscallFailsWithErrno(ENODATA)); +} + +TEST_F(XattrTest, LGetSetxattrOnNonsymlink) { + const char* path = test_file_name_.c_str(); + const char name[] = "user.test"; + int val = 1234; + size_t size = sizeof(val); + EXPECT_THAT(lsetxattr(path, name, &val, size, /*flags=*/0), + SyscallSucceeds()); + + int buf = 0; + EXPECT_THAT(lgetxattr(path, name, &buf, size), + SyscallSucceedsWithValue(size)); + EXPECT_EQ(buf, val); +} + +TEST_F(XattrTest, FGetSetxattr) { + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(test_file_name_.c_str(), 0)); + const char name[] = "user.test"; + int val = 1234; + size_t size = sizeof(val); + EXPECT_THAT(fsetxattr(fd.get(), name, &val, size, /*flags=*/0), + SyscallSucceeds()); + + int buf = 0; + EXPECT_THAT(fgetxattr(fd.get(), name, &buf, size), + SyscallSucceedsWithValue(size)); + EXPECT_EQ(buf, val); +} + } // namespace } // namespace testing -- cgit v1.2.3 From d46c397a1cd38f1e2aa5c864c1bb8594fb87bb63 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Tue, 21 Jan 2020 13:26:50 -0800 Subject: Add line break to /proc/net files Some files were missing the last line break. PiperOrigin-RevId: 290808898 --- pkg/sentry/fs/proc/net.go | 14 +++++++------- pkg/sentry/fsimpl/proc/tasks_net.go | 14 +++++++------- 2 files changed, 14 insertions(+), 14 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/proc/net.go b/pkg/sentry/fs/proc/net.go index 3f17e98ea..bad445f3f 100644 --- a/pkg/sentry/fs/proc/net.go +++ b/pkg/sentry/fs/proc/net.go @@ -52,17 +52,17 @@ func (p *proc) newNetDir(ctx context.Context, k *kernel.Kernel, msrc *fs.MountSo // implemented in netstack, if the file contains a // header the stub is just the header otherwise it is // an empty file. - "arp": newStaticProcInode(ctx, msrc, []byte("IP address HW type Flags HW address Mask Device")), + "arp": newStaticProcInode(ctx, msrc, []byte("IP address HW type Flags HW address Mask Device\n")), - "netlink": newStaticProcInode(ctx, msrc, []byte("sk Eth Pid Groups Rmem Wmem Dump Locks Drops Inode")), - "netstat": newStaticProcInode(ctx, msrc, []byte("TcpExt: SyncookiesSent SyncookiesRecv SyncookiesFailed EmbryonicRsts PruneCalled RcvPruned OfoPruned OutOfWindowIcmps LockDroppedIcmps ArpFilter TW TWRecycled TWKilled PAWSPassive PAWSActive PAWSEstab DelayedACKs DelayedACKLocked DelayedACKLost ListenOverflows ListenDrops TCPPrequeued TCPDirectCopyFromBacklog TCPDirectCopyFromPrequeue TCPPrequeueDropped TCPHPHits TCPHPHitsToUser TCPPureAcks TCPHPAcks TCPRenoRecovery TCPSackRecovery TCPSACKReneging TCPFACKReorder TCPSACKReorder TCPRenoReorder TCPTSReorder TCPFullUndo TCPPartialUndo TCPDSACKUndo TCPLossUndo TCPLostRetransmit TCPRenoFailures TCPSackFailures TCPLossFailures TCPFastRetrans TCPForwardRetrans TCPSlowStartRetrans TCPTimeouts TCPLossProbes TCPLossProbeRecovery TCPRenoRecoveryFail TCPSackRecoveryFail TCPSchedulerFailed TCPRcvCollapsed TCPDSACKOldSent TCPDSACKOfoSent TCPDSACKRecv TCPDSACKOfoRecv TCPAbortOnData TCPAbortOnClose TCPAbortOnMemory TCPAbortOnTimeout TCPAbortOnLinger TCPAbortFailed TCPMemoryPressures TCPSACKDiscard TCPDSACKIgnoredOld TCPDSACKIgnoredNoUndo TCPSpuriousRTOs TCPMD5NotFound TCPMD5Unexpected TCPMD5Failure TCPSackShifted TCPSackMerged TCPSackShiftFallback TCPBacklogDrop TCPMinTTLDrop TCPDeferAcceptDrop IPReversePathFilter TCPTimeWaitOverflow TCPReqQFullDoCookies TCPReqQFullDrop TCPRetransFail TCPRcvCoalesce TCPOFOQueue TCPOFODrop TCPOFOMerge TCPChallengeACK TCPSYNChallenge TCPFastOpenActive TCPFastOpenActiveFail TCPFastOpenPassive TCPFastOpenPassiveFail TCPFastOpenListenOverflow TCPFastOpenCookieReqd TCPSpuriousRtxHostQueues BusyPollRxPackets TCPAutoCorking TCPFromZeroWindowAdv TCPToZeroWindowAdv TCPWantZeroWindowAdv TCPSynRetrans TCPOrigDataSent TCPHystartTrainDetect TCPHystartTrainCwnd TCPHystartDelayDetect TCPHystartDelayCwnd TCPACKSkippedSynRecv TCPACKSkippedPAWS TCPACKSkippedSeq TCPACKSkippedFinWait2 TCPACKSkippedTimeWait TCPACKSkippedChallenge TCPWinProbe TCPKeepAlive TCPMTUPFail TCPMTUPSuccess")), - "packet": newStaticProcInode(ctx, msrc, []byte("sk RefCnt Type Proto Iface R Rmem User Inode")), - "protocols": newStaticProcInode(ctx, msrc, []byte("protocol size sockets memory press maxhdr slab module cl co di ac io in de sh ss gs se re sp bi br ha uh gp em")), + "netlink": newStaticProcInode(ctx, msrc, []byte("sk Eth Pid Groups Rmem Wmem Dump Locks Drops Inode\n")), + "netstat": newStaticProcInode(ctx, msrc, []byte("TcpExt: SyncookiesSent SyncookiesRecv SyncookiesFailed EmbryonicRsts PruneCalled RcvPruned OfoPruned OutOfWindowIcmps LockDroppedIcmps ArpFilter TW TWRecycled TWKilled PAWSPassive PAWSActive PAWSEstab DelayedACKs DelayedACKLocked DelayedACKLost ListenOverflows ListenDrops TCPPrequeued TCPDirectCopyFromBacklog TCPDirectCopyFromPrequeue TCPPrequeueDropped TCPHPHits TCPHPHitsToUser TCPPureAcks TCPHPAcks TCPRenoRecovery TCPSackRecovery TCPSACKReneging TCPFACKReorder TCPSACKReorder TCPRenoReorder TCPTSReorder TCPFullUndo TCPPartialUndo TCPDSACKUndo TCPLossUndo TCPLostRetransmit TCPRenoFailures TCPSackFailures TCPLossFailures TCPFastRetrans TCPForwardRetrans TCPSlowStartRetrans TCPTimeouts TCPLossProbes TCPLossProbeRecovery TCPRenoRecoveryFail TCPSackRecoveryFail TCPSchedulerFailed TCPRcvCollapsed TCPDSACKOldSent TCPDSACKOfoSent TCPDSACKRecv TCPDSACKOfoRecv TCPAbortOnData TCPAbortOnClose TCPAbortOnMemory TCPAbortOnTimeout TCPAbortOnLinger TCPAbortFailed TCPMemoryPressures TCPSACKDiscard TCPDSACKIgnoredOld TCPDSACKIgnoredNoUndo TCPSpuriousRTOs TCPMD5NotFound TCPMD5Unexpected TCPMD5Failure TCPSackShifted TCPSackMerged TCPSackShiftFallback TCPBacklogDrop TCPMinTTLDrop TCPDeferAcceptDrop IPReversePathFilter TCPTimeWaitOverflow TCPReqQFullDoCookies TCPReqQFullDrop TCPRetransFail TCPRcvCoalesce TCPOFOQueue TCPOFODrop TCPOFOMerge TCPChallengeACK TCPSYNChallenge TCPFastOpenActive TCPFastOpenActiveFail TCPFastOpenPassive TCPFastOpenPassiveFail TCPFastOpenListenOverflow TCPFastOpenCookieReqd TCPSpuriousRtxHostQueues BusyPollRxPackets TCPAutoCorking TCPFromZeroWindowAdv TCPToZeroWindowAdv TCPWantZeroWindowAdv TCPSynRetrans TCPOrigDataSent TCPHystartTrainDetect TCPHystartTrainCwnd TCPHystartDelayDetect TCPHystartDelayCwnd TCPACKSkippedSynRecv TCPACKSkippedPAWS TCPACKSkippedSeq TCPACKSkippedFinWait2 TCPACKSkippedTimeWait TCPACKSkippedChallenge TCPWinProbe TCPKeepAlive TCPMTUPFail TCPMTUPSuccess\n")), + "packet": newStaticProcInode(ctx, msrc, []byte("sk RefCnt Type Proto Iface R Rmem User Inode\n")), + "protocols": newStaticProcInode(ctx, msrc, []byte("protocol size sockets memory press maxhdr slab module cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n")), // Linux sets psched values to: nsec per usec, psched // tick in ns, 1000000, high res timer ticks per sec // (ClockGetres returns 1ns resolution). "psched": newStaticProcInode(ctx, msrc, []byte(fmt.Sprintf("%08x %08x %08x %08x\n", uint64(time.Microsecond/time.Nanosecond), 64, 1000000, uint64(time.Second/time.Nanosecond)))), - "ptype": newStaticProcInode(ctx, msrc, []byte("Type Device Function")), + "ptype": newStaticProcInode(ctx, msrc, []byte("Type Device Function\n")), "route": seqfile.NewSeqFileInode(ctx, &netRoute{s: s}, msrc), "tcp": seqfile.NewSeqFileInode(ctx, &netTCP{k: k}, msrc), "udp": seqfile.NewSeqFileInode(ctx, &netUDP{k: k}, msrc), @@ -73,7 +73,7 @@ func (p *proc) newNetDir(ctx context.Context, k *kernel.Kernel, msrc *fs.MountSo contents["if_inet6"] = seqfile.NewSeqFileInode(ctx, &ifinet6{s: s}, msrc) contents["ipv6_route"] = newStaticProcInode(ctx, msrc, []byte("")) contents["tcp6"] = seqfile.NewSeqFileInode(ctx, &netTCP6{k: k}, msrc) - contents["udp6"] = newStaticProcInode(ctx, msrc, []byte(" sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode")) + contents["udp6"] = newStaticProcInode(ctx, msrc, []byte(" sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode\n")) } } d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) diff --git a/pkg/sentry/fsimpl/proc/tasks_net.go b/pkg/sentry/fsimpl/proc/tasks_net.go index 3dbf3ba41..4aaf23e97 100644 --- a/pkg/sentry/fsimpl/proc/tasks_net.go +++ b/pkg/sentry/fsimpl/proc/tasks_net.go @@ -41,12 +41,12 @@ func newNetDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) *k var contents map[string]*kernfs.Dentry if stack := k.NetworkStack(); stack != nil { const ( - arp = "IP address HW type Flags HW address Mask Device" - netlink = "sk Eth Pid Groups Rmem Wmem Dump Locks Drops Inode" - packet = "sk RefCnt Type Proto Iface R Rmem User Inode" - protocols = "protocol size sockets memory press maxhdr slab module cl co di ac io in de sh ss gs se re sp bi br ha uh gp em" - ptype = "Type Device Function" - upd6 = " sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode" + arp = "IP address HW type Flags HW address Mask Device\n" + netlink = "sk Eth Pid Groups Rmem Wmem Dump Locks Drops Inode\n" + packet = "sk RefCnt Type Proto Iface R Rmem User Inode\n" + protocols = "protocol size sockets memory press maxhdr slab module cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n" + ptype = "Type Device Function\n" + upd6 = " sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode\n" ) psched := fmt.Sprintf("%08x %08x %08x %08x\n", uint64(time.Microsecond/time.Nanosecond), 64, 1000000, uint64(time.Second/time.Nanosecond)) @@ -779,6 +779,6 @@ func (d *netStatData) Generate(ctx context.Context, buf *bytes.Buffer) error { "TCPHystartTrainCwnd TCPHystartDelayDetect TCPHystartDelayCwnd " + "TCPACKSkippedSynRecv TCPACKSkippedPAWS TCPACKSkippedSeq " + "TCPACKSkippedFinWait2 TCPACKSkippedTimeWait TCPACKSkippedChallenge " + - "TCPWinProbe TCPKeepAlive TCPMTUPFail TCPMTUPSuccess") + "TCPWinProbe TCPKeepAlive TCPMTUPFail TCPMTUPSuccess\n") return nil } -- cgit v1.2.3 From 47bc7550c0b8fcde7b3452bf536082e955882026 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Tue, 21 Jan 2020 13:37:25 -0800 Subject: Fixing stuff --- pkg/sentry/socket/netfilter/netfilter.go | 3 ++ pkg/tcpip/iptables/types.go | 21 ++++---------- pkg/tcpip/packet_buffer.go | 25 +---------------- test/iptables/iptables_test.go | 47 ++++++++++++++++---------------- 4 files changed, 34 insertions(+), 62 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 09a3276c7..4ef8123ac 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -164,6 +164,9 @@ func convertNetstackToBinary(tablename string, table iptables.Table) (linux.Kern // Each rule corresponds to an entry. entry := linux.KernelIPTEntry{ IPTEntry: linux.IPTEntry{ + IP: linux.IPTIP{ + Protocol: uint16(rule.Filter.Protocol), + }, NextOffset: linux.SizeOfIPTEntry, TargetOffset: linux.SizeOfIPTEntry, }, diff --git a/pkg/tcpip/iptables/types.go b/pkg/tcpip/iptables/types.go index 4f2a4d65e..a0bfc8b41 100644 --- a/pkg/tcpip/iptables/types.go +++ b/pkg/tcpip/iptables/types.go @@ -14,7 +14,9 @@ package iptables -import "gvisor.dev/gvisor/pkg/tcpip" +import ( + "gvisor.dev/gvisor/pkg/tcpip" +) // A Hook specifies one of the hooks built into the network stack. // @@ -161,21 +163,10 @@ type Rule struct { Target Target } -// TODO: This is gross. -// TODO: Save this in SetEntries. -// TODO: Utilize this when traversing tables. +// IPHeaderFilter holds basic IP filtering data common to every rule. type IPHeaderFilter struct { - Source [4]byte - Destination [4]byte - SourceMask [4]byte - DestinationMask [4]byte - OutputInterface string - InputInterface string - OutputInterfaceMask string - InputInterfaceMask string - Protocol tcpip.TransportProtocolNumber - Flags uint8 - InverseFlags uint8 + // Protocol matches the transport protocol. + Protocol tcpip.TransportProtocolNumber } // A Matcher is the interface for matching packets. diff --git a/pkg/tcpip/packet_buffer.go b/pkg/tcpip/packet_buffer.go index 7a036b93c..ab24372e7 100644 --- a/pkg/tcpip/packet_buffer.go +++ b/pkg/tcpip/packet_buffer.go @@ -13,9 +13,7 @@ package tcpip -import ( - "gvisor.dev/gvisor/pkg/tcpip/buffer" -) +import "gvisor.dev/gvisor/pkg/tcpip/buffer" // A PacketBuffer contains all the data of a network packet. // @@ -67,24 +65,3 @@ func (pk PacketBuffer) Clone() PacketBuffer { pk.Data = pk.Data.Clone(nil) return pk } - -//// TransportProtocol returns the transport protocol of pk. -//// -//// Precondition: pk.NetworkHeader is set. -//func (pk PacketBuffer) TransportProtocolIPv4() uint16 { -// if pk.NetworkHeader == nil { -// panic("This should only be called when pk.NetworkHeader is set.") -// } -// return header.IPv4(pk.NetworkHeader).TransportProtocol() -//} - -// func (pk Packet) findNetHeader() header.IPv4 { -// // Inbound: -// // Data holds everything, but may have had some headers shaved off. -// // Figure out whether it's set or still somewhere in data and return -// // appropriately. - -// // Outbound: -// // NetworkHeader will be set if we've added one. Otherwise there's no -// // header. -// } diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go index 63e691af6..150b44e42 100644 --- a/test/iptables/iptables_test.go +++ b/test/iptables/iptables_test.go @@ -23,6 +23,7 @@ import ( "time" "flag" + "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/runsc/dockerutil" "gvisor.dev/gvisor/runsc/testutil" @@ -160,29 +161,29 @@ func logContainer(output string, err error) { log.Infof(msg) } -// func TestFilterInputDropUDP(t *testing.T) { -// if err := singleTest(FilterInputDropUDP{}); err != nil { -// t.Fatal(err) -// } -// } - -// func TestFilterInputDropUDPPort(t *testing.T) { -// if err := singleTest(FilterInputDropUDPPort{}); err != nil { -// t.Fatal(err) -// } -// } - -// func TestFilterInputDropDifferentUDPPort(t *testing.T) { -// if err := singleTest(FilterInputDropDifferentUDPPort{}); err != nil { -// t.Fatal(err) -// } -// } - -// func TestFilterInputDropAll(t *testing.T) { -// if err := singleTest(FilterInputDropAll{}); err != nil { -// t.Fatal(err) -// } -// } +func TestFilterInputDropUDP(t *testing.T) { + if err := singleTest(FilterInputDropUDP{}); err != nil { + t.Fatal(err) + } +} + +func TestFilterInputDropUDPPort(t *testing.T) { + if err := singleTest(FilterInputDropUDPPort{}); err != nil { + t.Fatal(err) + } +} + +func TestFilterInputDropDifferentUDPPort(t *testing.T) { + if err := singleTest(FilterInputDropDifferentUDPPort{}); err != nil { + t.Fatal(err) + } +} + +func TestFilterInputDropAll(t *testing.T) { + if err := singleTest(FilterInputDropAll{}); err != nil { + t.Fatal(err) + } +} func TestFilterInputDropOnlyUDP(t *testing.T) { if err := singleTest(FilterInputDropOnlyUDP{}); err != nil { -- cgit v1.2.3 From 9f736ac6a7747917f690596ac9b072c108b5670c Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Tue, 21 Jan 2020 13:39:48 -0800 Subject: More little fixes. --- pkg/sentry/socket/netfilter/netfilter.go | 4 ++-- test/iptables/iptables_test.go | 3 +-- 2 files changed, 3 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 4ef8123ac..e1f2bacce 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -325,8 +325,8 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { return syserr.ErrInvalidArgument } - // TODO(gvisor.dev/issue/170): We should support IPTIP - // filtering. We reject any nonzero IPTIP values for now. + // TODO(gvisor.dev/issue/170): We should support more IPTIP + // filtering fields. filter, err := filterFromIPTIP(entry.IP) if err != nil { return err diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go index 150b44e42..679a29bef 100644 --- a/test/iptables/iptables_test.go +++ b/test/iptables/iptables_test.go @@ -15,6 +15,7 @@ package iptables import ( + "flag" "fmt" "net" "os" @@ -22,8 +23,6 @@ import ( "testing" "time" - "flag" - "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/runsc/dockerutil" "gvisor.dev/gvisor/runsc/testutil" -- cgit v1.2.3 From 9143fcd7fd38243dd40f927dafaeb75f6ef8ef49 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Tue, 21 Jan 2020 14:47:17 -0800 Subject: Add UDP matchers. --- pkg/abi/linux/netfilter.go | 92 +++++++++++++++++ pkg/sentry/socket/netfilter/netfilter.go | 164 ++++++++++++++++++++++++++----- pkg/tcpip/iptables/BUILD | 2 + pkg/tcpip/iptables/iptables.go | 3 + pkg/tcpip/iptables/tcp_matcher.go | 122 +++++++++++++++++++++++ pkg/tcpip/iptables/types.go | 17 ++++ pkg/tcpip/iptables/udp_matcher.go | 127 ++++++++++++++++++++++++ pkg/tcpip/network/ipv4/ipv4.go | 10 +- test/iptables/filter_input.go | 46 +++++++++ 9 files changed, 555 insertions(+), 28 deletions(-) create mode 100644 pkg/tcpip/iptables/tcp_matcher.go create mode 100644 pkg/tcpip/iptables/udp_matcher.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/netfilter.go b/pkg/abi/linux/netfilter.go index 33fcc6c95..fb4588272 100644 --- a/pkg/abi/linux/netfilter.go +++ b/pkg/abi/linux/netfilter.go @@ -340,3 +340,95 @@ func goString(cstring []byte) string { } return string(cstring) } + +// XTTCP holds data for matching TCP packets. It corresponds to struct xt_tcp +// in include/uapi/linux/netfilter/xt_tcpudp.h. +type XTTCP struct { + // SourcePortStart specifies the inclusive start of the range of source + // ports to which the matcher applies. + SourcePortStart uint16 + + // SourcePortEnd specifies the inclusive end of the range of source ports + // to which the matcher applies. + SourcePortEnd uint16 + + // DestinationPortStart specifies the start of the destination port + // range to which the matcher applies. + DestinationPortStart uint16 + + // DestinationPortEnd specifies the start of the destination port + // range to which the matcher applies. + DestinationPortEnd uint16 + + // Option specifies that a particular TCP option must be set. + Option uint8 + + // FlagMask masks the FlagCompare byte when comparing to the TCP flag + // fields. + FlagMask uint8 + + // FlagCompare is binary and-ed with the TCP flag fields. + FlagCompare uint8 + + // InverseFlags flips the meaning of certain fields. See the + // TX_TCP_INV_* flags. + InverseFlags uint8 +} + +// SizeOfXTTCP is the size of an XTTCP. +const SizeOfXTTCP = 12 + +// Flags in XTTCP.InverseFlags. Corresponding constants are in +// include/uapi/linux/netfilter/xt_tcpudp.h. +const ( + // Invert the meaning of SourcePortStart/End. + XT_TCP_INV_SRCPT = 0x01 + // Invert the meaning of DestinationPortStart/End. + XT_TCP_INV_DSTPT = 0x02 + // Invert the meaning of FlagCompare. + XT_TCP_INV_FLAGS = 0x04 + // Invert the meaning of Option. + XT_TCP_INV_OPTION = 0x08 + // Enable all flags. + XT_TCP_INV_MASK = 0x0F +) + +// XTUDP holds data for matching UDP packets. It corresponds to struct xt_udp +// in include/uapi/linux/netfilter/xt_tcpudp.h. +type XTUDP struct { + // SourcePortStart specifies the inclusive start of the range of source + // ports to which the matcher applies. + SourcePortStart uint16 + + // SourcePortEnd specifies the inclusive end of the range of source ports + // to which the matcher applies. + SourcePortEnd uint16 + + // DestinationPortStart specifies the start of the destination port + // range to which the matcher applies. + DestinationPortStart uint16 + + // DestinationPortEnd specifies the start of the destination port + // range to which the matcher applies. + DestinationPortEnd uint16 + + // InverseFlags flips the meaning of certain fields. See the + // TX_UDP_INV_* flags. + InverseFlags uint8 + + _ uint8 +} + +// SizeOfXTUDP is the size of an XTUDP. +const SizeOfXTUDP = 10 + +// Flags in XTUDP.InverseFlags. Corresponding constants are in +// include/uapi/linux/netfilter/xt_tcpudp.h. +const ( + // Invert the meaning of SourcePortStart/End. + XT_UDP_INV_SRCPT = 0x01 + // Invert the meaning of DestinationPortStart/End. + XT_UDP_INV_DSTPT = 0x02 + // Enable all flags. + XT_UDP_INV_MASK = 0x03 +) diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index e1f2bacce..45296b339 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -131,6 +131,7 @@ func FillDefaultIPTables(stack *stack.Stack) { stack.SetIPTables(ipt) } +// TODO: Return proto. // convertNetstackToBinary converts the iptables as stored in netstack to the // format expected by the iptables tool. Linux stores each table as a binary // blob that can only be traversed by parsing a bit, reading some offsets, @@ -318,10 +319,12 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { } var entry linux.IPTEntry buf := optVal[:linux.SizeOfIPTEntry] - optVal = optVal[linux.SizeOfIPTEntry:] binary.Unmarshal(buf, usermem.ByteOrder, &entry) - if entry.TargetOffset != linux.SizeOfIPTEntry { - // TODO(gvisor.dev/issue/170): Support matchers. + initialOptValLen := len(optVal) + optVal = optVal[linux.SizeOfIPTEntry:] + + if entry.TargetOffset < linux.SizeOfIPTEntry { + log.Warningf("netfilter: entry has too-small target offset %d", entry.TargetOffset) return syserr.ErrInvalidArgument } @@ -332,19 +335,41 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { return err } + // TODO: Matchers (and maybe targets) can specify that they only work for certiain protocols, hooks, tables. + // Get matchers. + matchersSize := entry.TargetOffset - linux.SizeOfIPTEntry + if len(optVal) < int(matchersSize) { + log.Warningf("netfilter: entry doesn't have enough room for its matchers (only %d bytes remain)", len(optVal)) + } + matchers, err := parseMatchers(filter, optVal[:matchersSize]) + if err != nil { + log.Warningf("netfilter: failed to parse matchers: %v", err) + return err + } + optVal = optVal[matchersSize:] + // Get the target of the rule. - target, consumed, err := parseTarget(optVal) + targetSize := entry.NextOffset - entry.TargetOffset + if len(optVal) < int(targetSize) { + log.Warningf("netfilter: entry doesn't have enough room for its target (only %d bytes remain)", len(optVal)) + } + target, err := parseTarget(optVal[:targetSize]) if err != nil { return err } - optVal = optVal[consumed:] + optVal = optVal[targetSize:] table.Rules = append(table.Rules, iptables.Rule{ - Filter: filter, - Target: target, + Filter: filter, + Target: target, + Matchers: matchers, }) offsets = append(offsets, offset) - offset += linux.SizeOfIPTEntry + consumed + offset += uint32(entry.NextOffset) + + if initialOptValLen-len(optVal) != int(entry.NextOffset) { + log.Warningf("netfilter: entry NextOffset is %d, but entry took up %d bytes", entry.NextOffset, initialOptValLen-len(optVal)) + } } // Go through the list of supported hooks for this table and, for each @@ -401,12 +426,105 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { return nil } -// parseTarget parses a target from the start of optVal and returns the target -// along with the number of bytes it occupies in optVal. -func parseTarget(optVal []byte) (iptables.Target, uint32, *syserr.Error) { +// parseMatchers parses 0 or more matchers from optVal. optVal should contain +// only the matchers. +func parseMatchers(filter iptables.IPHeaderFilter, optVal []byte) ([]iptables.Matcher, *syserr.Error) { + var matchers []iptables.Matcher + for len(optVal) > 0 { + log.Infof("parseMatchers: optVal has len %d", len(optVal)) + // Get the XTEntryMatch. + if len(optVal) < linux.SizeOfXTEntryMatch { + log.Warningf("netfilter: optVal has insufficient size for entry match: %d", len(optVal)) + return nil, syserr.ErrInvalidArgument + } + var match linux.XTEntryMatch + buf := optVal[:linux.SizeOfXTEntryMatch] + binary.Unmarshal(buf, usermem.ByteOrder, &match) + log.Infof("parseMatchers: parsed entry match %q: %+v", match.Name.String(), match) + + // Check some invariants. + if match.MatchSize < linux.SizeOfXTEntryMatch { + log.Warningf("netfilter: match size is too small, must be at least %d", linux.SizeOfXTEntryMatch) + return nil, syserr.ErrInvalidArgument + } + if len(optVal) < int(match.MatchSize) { + log.Warningf("netfilter: optVal has insufficient size for match: %d", len(optVal)) + return nil, syserr.ErrInvalidArgument + } + + buf = optVal[linux.SizeOfXTEntryMatch:match.MatchSize] + var matcher iptables.Matcher + var err error + switch match.Name.String() { + case "tcp": + if len(buf) < linux.SizeOfXTTCP { + log.Warningf("netfilter: optVal has insufficient size for TCP match: %d", len(optVal)) + return nil, syserr.ErrInvalidArgument + } + var matchData linux.XTTCP + // For alignment reasons, the match's total size may exceed what's + // strictly necessary to hold matchData. + binary.Unmarshal(buf[:linux.SizeOfXTUDP], usermem.ByteOrder, &matchData) + log.Infof("parseMatchers: parsed XTTCP: %+v", matchData) + matcher, err = iptables.NewTCPMatcher(filter, iptables.TCPMatcherData{ + SourcePortStart: matchData.SourcePortStart, + SourcePortEnd: matchData.SourcePortEnd, + DestinationPortStart: matchData.DestinationPortStart, + DestinationPortEnd: matchData.DestinationPortEnd, + Option: matchData.Option, + FlagMask: matchData.FlagMask, + FlagCompare: matchData.FlagCompare, + InverseFlags: matchData.InverseFlags, + }) + if err != nil { + log.Warningf("netfilter: failed to create TCP matcher: %v", err) + return nil, syserr.ErrInvalidArgument + } + + case "udp": + if len(buf) < linux.SizeOfXTUDP { + log.Warningf("netfilter: optVal has insufficient size for UDP match: %d", len(optVal)) + return nil, syserr.ErrInvalidArgument + } + var matchData linux.XTUDP + // For alignment reasons, the match's total size may exceed what's + // strictly necessary to hold matchData. + binary.Unmarshal(buf[:linux.SizeOfXTUDP], usermem.ByteOrder, &matchData) + log.Infof("parseMatchers: parsed XTUDP: %+v", matchData) + matcher, err = iptables.NewUDPMatcher(filter, iptables.UDPMatcherData{ + SourcePortStart: matchData.SourcePortStart, + SourcePortEnd: matchData.SourcePortEnd, + DestinationPortStart: matchData.DestinationPortStart, + DestinationPortEnd: matchData.DestinationPortEnd, + InverseFlags: matchData.InverseFlags, + }) + if err != nil { + log.Warningf("netfilter: failed to create UDP matcher: %v", err) + return nil, syserr.ErrInvalidArgument + } + + default: + log.Warningf("netfilter: unsupported matcher with name %q", match.Name.String()) + return nil, syserr.ErrInvalidArgument + } + + matchers = append(matchers, matcher) + + // TODO: Support revision. + // TODO: Support proto -- matchers usually specify which proto(s) they work with. + optVal = optVal[match.MatchSize:] + } + + // TODO: Check that optVal is exhausted. + return matchers, nil +} + +// parseTarget parses a target from optVal. optVal should contain only the +// target. +func parseTarget(optVal []byte) (iptables.Target, *syserr.Error) { if len(optVal) < linux.SizeOfXTEntryTarget { log.Warningf("netfilter: optVal has insufficient size for entry target %d", len(optVal)) - return nil, 0, syserr.ErrInvalidArgument + return nil, syserr.ErrInvalidArgument } var target linux.XTEntryTarget buf := optVal[:linux.SizeOfXTEntryTarget] @@ -414,9 +532,9 @@ func parseTarget(optVal []byte) (iptables.Target, uint32, *syserr.Error) { switch target.Name.String() { case "": // Standard target. - if len(optVal) < linux.SizeOfXTStandardTarget { - log.Warningf("netfilter.SetEntries: optVal has insufficient size for standard target %d", len(optVal)) - return nil, 0, syserr.ErrInvalidArgument + if len(optVal) != linux.SizeOfXTStandardTarget { + log.Warningf("netfilter.SetEntries: optVal has wrong size for standard target %d", len(optVal)) + return nil, syserr.ErrInvalidArgument } var standardTarget linux.XTStandardTarget buf = optVal[:linux.SizeOfXTStandardTarget] @@ -424,22 +542,22 @@ func parseTarget(optVal []byte) (iptables.Target, uint32, *syserr.Error) { verdict, err := translateToStandardVerdict(standardTarget.Verdict) if err != nil { - return nil, 0, err + return nil, err } switch verdict { case iptables.Accept: - return iptables.UnconditionalAcceptTarget{}, linux.SizeOfXTStandardTarget, nil + return iptables.UnconditionalAcceptTarget{}, nil case iptables.Drop: - return iptables.UnconditionalDropTarget{}, linux.SizeOfXTStandardTarget, nil + return iptables.UnconditionalDropTarget{}, nil default: panic(fmt.Sprintf("Unknown verdict: %v", verdict)) } case errorTargetName: // Error target. - if len(optVal) < linux.SizeOfXTErrorTarget { + if len(optVal) != linux.SizeOfXTErrorTarget { log.Infof("netfilter.SetEntries: optVal has insufficient size for error target %d", len(optVal)) - return nil, 0, syserr.ErrInvalidArgument + return nil, syserr.ErrInvalidArgument } var errorTarget linux.XTErrorTarget buf = optVal[:linux.SizeOfXTErrorTarget] @@ -454,16 +572,16 @@ func parseTarget(optVal []byte) (iptables.Target, uint32, *syserr.Error) { // rules have an error with the name of the chain. switch errorTarget.Name.String() { case errorTargetName: - return iptables.ErrorTarget{}, linux.SizeOfXTErrorTarget, nil + return iptables.ErrorTarget{}, nil default: log.Infof("Unknown error target %q doesn't exist or isn't supported yet.", errorTarget.Name.String()) - return nil, 0, syserr.ErrInvalidArgument + return nil, syserr.ErrInvalidArgument } } // Unknown target. log.Infof("Unknown target %q doesn't exist or isn't supported yet.", target.Name.String()) - return nil, 0, syserr.ErrInvalidArgument + return nil, syserr.ErrInvalidArgument } func filterFromIPTIP(iptip linux.IPTIP) (iptables.IPHeaderFilter, *syserr.Error) { diff --git a/pkg/tcpip/iptables/BUILD b/pkg/tcpip/iptables/BUILD index 297eaccaf..ff4e3c932 100644 --- a/pkg/tcpip/iptables/BUILD +++ b/pkg/tcpip/iptables/BUILD @@ -7,7 +7,9 @@ go_library( srcs = [ "iptables.go", "targets.go", + "tcp_matcher.go", "types.go", + "udp_matcher.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/iptables", visibility = ["//visibility:public"], diff --git a/pkg/tcpip/iptables/iptables.go b/pkg/tcpip/iptables/iptables.go index fc06b5b87..accedba1e 100644 --- a/pkg/tcpip/iptables/iptables.go +++ b/pkg/tcpip/iptables/iptables.go @@ -138,6 +138,8 @@ func EmptyFilterTable() Table { // Check runs pkt through the rules for hook. It returns true when the packet // should continue traversing the network stack and false when it should be // dropped. +// +// Precondition: pkt.NetworkHeader is set. func (it *IPTables) Check(hook Hook, pkt tcpip.PacketBuffer) bool { // TODO(gvisor.dev/issue/170): A lot of this is uncomplicated because // we're missing features. Jumps, the call stack, etc. aren't checked @@ -163,6 +165,7 @@ func (it *IPTables) Check(hook Hook, pkt tcpip.PacketBuffer) bool { return true } +// Precondition: pkt.NetworkHeader is set. func (it *IPTables) checkTable(hook Hook, pkt tcpip.PacketBuffer, tablename string) Verdict { // Start from ruleIdx and walk the list of rules until a rule gives us // a verdict. diff --git a/pkg/tcpip/iptables/tcp_matcher.go b/pkg/tcpip/iptables/tcp_matcher.go new file mode 100644 index 000000000..6acbd6eb9 --- /dev/null +++ b/pkg/tcpip/iptables/tcp_matcher.go @@ -0,0 +1,122 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package iptables + +import ( + "fmt" + + "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/header" +) + +type TCPMatcher struct { + data TCPMatcherData + + // tablename string + // unsigned int matchsize; + // unsigned int usersize; + // #ifdef CONFIG_COMPAT + // unsigned int compatsize; + // #endif + // unsigned int hooks; + // unsigned short proto; + // unsigned short family; +} + +// TODO: Delete? +// MatchCheckEntryParams + +type TCPMatcherData struct { + // Filter IPHeaderFilter + + SourcePortStart uint16 + SourcePortEnd uint16 + DestinationPortStart uint16 + DestinationPortEnd uint16 + Option uint8 + FlagMask uint8 + FlagCompare uint8 + InverseFlags uint8 +} + +func NewTCPMatcher(filter IPHeaderFilter, data TCPMatcherData) (Matcher, error) { + // TODO: We currently only support source port and destination port. + log.Infof("Adding rule with TCPMatcherData: %+v", data) + + if data.Option != 0 || + data.FlagMask != 0 || + data.FlagCompare != 0 || + data.InverseFlags != 0 { + return nil, fmt.Errorf("unsupported TCP matcher flags set") + } + + if filter.Protocol != header.TCPProtocolNumber { + log.Warningf("TCP matching is only valid for protocol %d.", header.TCPProtocolNumber) + } + + return &TCPMatcher{data: data}, nil +} + +// TODO: Check xt_tcpudp.c. Need to check for same things (e.g. fragments). +func (tm *TCPMatcher) Match(hook Hook, pkt tcpip.PacketBuffer, interfaceName string) (bool, bool) { + netHeader := header.IPv4(pkt.NetworkHeader) + + // TODO: Do we check proto here or elsewhere? I think elsewhere (check + // codesearch). + if netHeader.TransportProtocol() != header.TCPProtocolNumber { + return false, false + } + + // We dont't match fragments. + if frag := netHeader.FragmentOffset(); frag != 0 { + if frag == 1 { + log.Warningf("Dropping TCP packet: malicious packet with fragment with fragment offest of 1.") + return false, true + } + return false, false + } + + // Now we need the transport header. However, this may not have been set + // yet. + // TODO + var tcpHeader header.TCP + if pkt.TransportHeader != nil { + tcpHeader = header.TCP(pkt.TransportHeader) + } else { + // The TCP header hasn't been parsed yet. We have to do it here. + if len(pkt.Data.First()) < header.TCPMinimumSize { + // There's no valid TCP header here, so we hotdrop the + // packet. + // TODO: Stats. + log.Warningf("Dropping TCP packet: size to small.") + return false, true + } + tcpHeader = header.TCP(pkt.Data.First()) + } + + // Check whether the source and destination ports are within the + // matching range. + sourcePort := tcpHeader.SourcePort() + destinationPort := tcpHeader.DestinationPort() + if sourcePort < tm.data.SourcePortStart || tm.data.SourcePortEnd < sourcePort { + return false, false + } + if destinationPort < tm.data.DestinationPortStart || tm.data.DestinationPortEnd < destinationPort { + return false, false + } + + return true, false +} diff --git a/pkg/tcpip/iptables/types.go b/pkg/tcpip/iptables/types.go index a0bfc8b41..54e66f09a 100644 --- a/pkg/tcpip/iptables/types.go +++ b/pkg/tcpip/iptables/types.go @@ -169,12 +169,29 @@ type IPHeaderFilter struct { Protocol tcpip.TransportProtocolNumber } +// TODO: Should these be able to marshal/unmarshal themselves? +// TODO: Something has to map the name to the matcher. // A Matcher is the interface for matching packets. type Matcher interface { // Match returns whether the packet matches and whether the packet // should be "hotdropped", i.e. dropped immediately. This is usually // used for suspicious packets. + // + // Precondition: packet.NetworkHeader is set. Match(hook Hook, packet tcpip.PacketBuffer, interfaceName string) (matches bool, hotdrop bool) + + // TODO: Make this typesafe by having each Matcher have their own, typed CheckEntry? + // CheckEntry(params MatchCheckEntryParams) bool +} + +// TODO: Unused? +type MatchCheckEntryParams struct { + Table string // TODO: Tables should be an enum... + Filter IPHeaderFilter + Info interface{} // TODO: Type unsafe. + // HookMask uint8 + // Family uint8 + // NFTCompat bool } // A Target is the interface for taking an action for a packet. diff --git a/pkg/tcpip/iptables/udp_matcher.go b/pkg/tcpip/iptables/udp_matcher.go new file mode 100644 index 000000000..ce4368a3d --- /dev/null +++ b/pkg/tcpip/iptables/udp_matcher.go @@ -0,0 +1,127 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package iptables + +import ( + "fmt" + "runtime/debug" + + "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/header" +) + +type UDPMatcher struct { + data UDPMatcherData + + // tablename string + // unsigned int matchsize; + // unsigned int usersize; + // #ifdef CONFIG_COMPAT + // unsigned int compatsize; + // #endif + // unsigned int hooks; + // unsigned short proto; + // unsigned short family; +} + +// TODO: Delete? +// MatchCheckEntryParams + +type UDPMatcherData struct { + // Filter IPHeaderFilter + + SourcePortStart uint16 + SourcePortEnd uint16 + DestinationPortStart uint16 + DestinationPortEnd uint16 + InverseFlags uint8 +} + +func NewUDPMatcher(filter IPHeaderFilter, data UDPMatcherData) (Matcher, error) { + // TODO: We currently only support source port and destination port. + log.Infof("Adding rule with UDPMatcherData: %+v", data) + + if data.InverseFlags != 0 { + return nil, fmt.Errorf("unsupported UDP matcher flags set") + } + + if filter.Protocol != header.UDPProtocolNumber { + log.Warningf("UDP matching is only valid for protocol %d.", header.UDPProtocolNumber) + } + + return &UDPMatcher{data: data}, nil +} + +// TODO: Check xt_tcpudp.c. Need to check for same things (e.g. fragments). +func (tm *UDPMatcher) Match(hook Hook, pkt tcpip.PacketBuffer, interfaceName string) (bool, bool) { + log.Infof("UDPMatcher called from: %s", string(debug.Stack())) + netHeader := header.IPv4(pkt.NetworkHeader) + + // TODO: Do we check proto here or elsewhere? I think elsewhere (check + // codesearch). + if netHeader.TransportProtocol() != header.UDPProtocolNumber { + log.Infof("UDPMatcher: wrong protocol number") + return false, false + } + + // We dont't match fragments. + if frag := netHeader.FragmentOffset(); frag != 0 { + log.Infof("UDPMatcher: it's a fragment") + if frag == 1 { + return false, true + } + log.Warningf("Dropping UDP packet: malicious fragmented packet.") + return false, false + } + + // Now we need the transport header. However, this may not have been set + // yet. + // TODO + var udpHeader header.UDP + if pkt.TransportHeader != nil { + log.Infof("UDPMatcher: transport header is not nil") + udpHeader = header.UDP(pkt.TransportHeader) + } else { + log.Infof("UDPMatcher: transport header is nil") + log.Infof("UDPMatcher: is network header nil: %t", pkt.NetworkHeader == nil) + // The UDP header hasn't been parsed yet. We have to do it here. + if len(pkt.Data.First()) < header.UDPMinimumSize { + // There's no valid UDP header here, so we hotdrop the + // packet. + // TODO: Stats. + log.Warningf("Dropping UDP packet: size to small.") + return false, true + } + udpHeader = header.UDP(pkt.Data.First()) + } + + // Check whether the source and destination ports are within the + // matching range. + sourcePort := udpHeader.SourcePort() + destinationPort := udpHeader.DestinationPort() + log.Infof("UDPMatcher: sport and dport are %d and %d. sports and dport start and end are (%d, %d) and (%d, %d)", + udpHeader.SourcePort(), udpHeader.DestinationPort(), + tm.data.SourcePortStart, tm.data.SourcePortEnd, + tm.data.DestinationPortStart, tm.data.DestinationPortEnd) + if sourcePort < tm.data.SourcePortStart || tm.data.SourcePortEnd < sourcePort { + return false, false + } + if destinationPort < tm.data.DestinationPortStart || tm.data.DestinationPortEnd < destinationPort { + return false, false + } + + return true, false +} diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index 85512f9b2..6597e6781 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -353,6 +353,11 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt tcpip.PacketBuffer) { } pkt.NetworkHeader = headerView[:h.HeaderLength()] + hlen := int(h.HeaderLength()) + tlen := int(h.TotalLength()) + pkt.Data.TrimFront(hlen) + pkt.Data.CapLength(tlen - hlen) + // iptables filtering. All packets that reach here are intended for // this machine and will not be forwarded. ipt := e.stack.IPTables() @@ -361,11 +366,6 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt tcpip.PacketBuffer) { return } - hlen := int(h.HeaderLength()) - tlen := int(h.TotalLength()) - pkt.Data.TrimFront(hlen) - pkt.Data.CapLength(tlen - hlen) - more := (h.Flags() & header.IPv4FlagMoreFragments) != 0 if more || h.FragmentOffset() != 0 { if pkt.Data.Size() == 0 { diff --git a/test/iptables/filter_input.go b/test/iptables/filter_input.go index fd02ff2ff..bc963d40e 100644 --- a/test/iptables/filter_input.go +++ b/test/iptables/filter_input.go @@ -15,6 +15,7 @@ package iptables import ( + "errors" "fmt" "net" "time" @@ -248,3 +249,48 @@ func (FilterInputDropAll) ContainerAction(ip net.IP) error { func (FilterInputDropAll) LocalAction(ip net.IP) error { return sendUDPLoop(ip, dropPort, sendloopDuration) } + +// FilterInputMultiUDPRules verifies that multiple UDP rules are applied +// correctly. This has the added benefit of testing whether we're serializing +// rules correctly -- if we do it incorrectly, the iptables tool will +// misunderstand and save the wrong tables. +type FilterInputMultiUDPRules struct{} + +func (FilterInputMultiUDPRules) Name() string { + return "FilterInputMultiUDPRules" +} + +func (FilterInputMultiUDPRules) ContainerAction(ip net.IP) error { + if err := filterTable("-A", "INPUT", "-p", "udp", "-m", "udp", "--destination-port", fmt.Sprintf("%d", dropPort), "-j", "DROP"); err != nil { + return err + } + // if err := filterTable("-A", "INPUT", "-p", "udp", "-m", "udp", "--destination-port", fmt.Sprintf("%d", acceptPort), "-j", "ACCEPT"); err != nil { + // return err + // } + return filterTable("-L") +} + +func (FilterInputMultiUDPRules) LocalAction(ip net.IP) error { + // No-op. + return nil +} + +// FilterInputRequireProtocolUDP checks that "-m udp" requires "-p udp" to be +// specified. +type FilterInputRequireProtocolUDP struct{} + +func (FilterInputRequireProtocolUDP) Name() string { + return "FilterInputRequireProtocolUDP" +} + +func (FilterInputRequireProtocolUDP) ContainerAction(ip net.IP) error { + if err := filterTable("-A", "INPUT", "-m", "udp", "--destination-port", fmt.Sprintf("%d", dropPort), "-j", "DROP"); err == nil { + return errors.New("expected iptables to fail with out \"-p udp\", but succeeded") + } + return nil +} + +func (FilterInputRequireProtocolUDP) LocalAction(ip net.IP) error { + // No-op. + return nil +} -- cgit v1.2.3 From 2661101ad470548cb15dce0afc694296668d780a Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Tue, 21 Jan 2020 14:51:28 -0800 Subject: Removed TCP work (saved in ipt-tcp-match). --- pkg/abi/linux/netfilter.go | 52 ------------- pkg/sentry/socket/netfilter/netfilter.go | 26 ------- pkg/tcpip/iptables/tcp_matcher.go | 122 ------------------------------- 3 files changed, 200 deletions(-) delete mode 100644 pkg/tcpip/iptables/tcp_matcher.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/netfilter.go b/pkg/abi/linux/netfilter.go index fb4588272..f0e544f9c 100644 --- a/pkg/abi/linux/netfilter.go +++ b/pkg/abi/linux/netfilter.go @@ -341,58 +341,6 @@ func goString(cstring []byte) string { return string(cstring) } -// XTTCP holds data for matching TCP packets. It corresponds to struct xt_tcp -// in include/uapi/linux/netfilter/xt_tcpudp.h. -type XTTCP struct { - // SourcePortStart specifies the inclusive start of the range of source - // ports to which the matcher applies. - SourcePortStart uint16 - - // SourcePortEnd specifies the inclusive end of the range of source ports - // to which the matcher applies. - SourcePortEnd uint16 - - // DestinationPortStart specifies the start of the destination port - // range to which the matcher applies. - DestinationPortStart uint16 - - // DestinationPortEnd specifies the start of the destination port - // range to which the matcher applies. - DestinationPortEnd uint16 - - // Option specifies that a particular TCP option must be set. - Option uint8 - - // FlagMask masks the FlagCompare byte when comparing to the TCP flag - // fields. - FlagMask uint8 - - // FlagCompare is binary and-ed with the TCP flag fields. - FlagCompare uint8 - - // InverseFlags flips the meaning of certain fields. See the - // TX_TCP_INV_* flags. - InverseFlags uint8 -} - -// SizeOfXTTCP is the size of an XTTCP. -const SizeOfXTTCP = 12 - -// Flags in XTTCP.InverseFlags. Corresponding constants are in -// include/uapi/linux/netfilter/xt_tcpudp.h. -const ( - // Invert the meaning of SourcePortStart/End. - XT_TCP_INV_SRCPT = 0x01 - // Invert the meaning of DestinationPortStart/End. - XT_TCP_INV_DSTPT = 0x02 - // Invert the meaning of FlagCompare. - XT_TCP_INV_FLAGS = 0x04 - // Invert the meaning of Option. - XT_TCP_INV_OPTION = 0x08 - // Enable all flags. - XT_TCP_INV_MASK = 0x0F -) - // XTUDP holds data for matching UDP packets. It corresponds to struct xt_udp // in include/uapi/linux/netfilter/xt_tcpudp.h. type XTUDP struct { diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 45296b339..f8ed1acbc 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -131,7 +131,6 @@ func FillDefaultIPTables(stack *stack.Stack) { stack.SetIPTables(ipt) } -// TODO: Return proto. // convertNetstackToBinary converts the iptables as stored in netstack to the // format expected by the iptables tool. Linux stores each table as a binary // blob that can only be traversed by parsing a bit, reading some offsets, @@ -456,31 +455,6 @@ func parseMatchers(filter iptables.IPHeaderFilter, optVal []byte) ([]iptables.Ma var matcher iptables.Matcher var err error switch match.Name.String() { - case "tcp": - if len(buf) < linux.SizeOfXTTCP { - log.Warningf("netfilter: optVal has insufficient size for TCP match: %d", len(optVal)) - return nil, syserr.ErrInvalidArgument - } - var matchData linux.XTTCP - // For alignment reasons, the match's total size may exceed what's - // strictly necessary to hold matchData. - binary.Unmarshal(buf[:linux.SizeOfXTUDP], usermem.ByteOrder, &matchData) - log.Infof("parseMatchers: parsed XTTCP: %+v", matchData) - matcher, err = iptables.NewTCPMatcher(filter, iptables.TCPMatcherData{ - SourcePortStart: matchData.SourcePortStart, - SourcePortEnd: matchData.SourcePortEnd, - DestinationPortStart: matchData.DestinationPortStart, - DestinationPortEnd: matchData.DestinationPortEnd, - Option: matchData.Option, - FlagMask: matchData.FlagMask, - FlagCompare: matchData.FlagCompare, - InverseFlags: matchData.InverseFlags, - }) - if err != nil { - log.Warningf("netfilter: failed to create TCP matcher: %v", err) - return nil, syserr.ErrInvalidArgument - } - case "udp": if len(buf) < linux.SizeOfXTUDP { log.Warningf("netfilter: optVal has insufficient size for UDP match: %d", len(optVal)) diff --git a/pkg/tcpip/iptables/tcp_matcher.go b/pkg/tcpip/iptables/tcp_matcher.go deleted file mode 100644 index 6acbd6eb9..000000000 --- a/pkg/tcpip/iptables/tcp_matcher.go +++ /dev/null @@ -1,122 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package iptables - -import ( - "fmt" - - "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/header" -) - -type TCPMatcher struct { - data TCPMatcherData - - // tablename string - // unsigned int matchsize; - // unsigned int usersize; - // #ifdef CONFIG_COMPAT - // unsigned int compatsize; - // #endif - // unsigned int hooks; - // unsigned short proto; - // unsigned short family; -} - -// TODO: Delete? -// MatchCheckEntryParams - -type TCPMatcherData struct { - // Filter IPHeaderFilter - - SourcePortStart uint16 - SourcePortEnd uint16 - DestinationPortStart uint16 - DestinationPortEnd uint16 - Option uint8 - FlagMask uint8 - FlagCompare uint8 - InverseFlags uint8 -} - -func NewTCPMatcher(filter IPHeaderFilter, data TCPMatcherData) (Matcher, error) { - // TODO: We currently only support source port and destination port. - log.Infof("Adding rule with TCPMatcherData: %+v", data) - - if data.Option != 0 || - data.FlagMask != 0 || - data.FlagCompare != 0 || - data.InverseFlags != 0 { - return nil, fmt.Errorf("unsupported TCP matcher flags set") - } - - if filter.Protocol != header.TCPProtocolNumber { - log.Warningf("TCP matching is only valid for protocol %d.", header.TCPProtocolNumber) - } - - return &TCPMatcher{data: data}, nil -} - -// TODO: Check xt_tcpudp.c. Need to check for same things (e.g. fragments). -func (tm *TCPMatcher) Match(hook Hook, pkt tcpip.PacketBuffer, interfaceName string) (bool, bool) { - netHeader := header.IPv4(pkt.NetworkHeader) - - // TODO: Do we check proto here or elsewhere? I think elsewhere (check - // codesearch). - if netHeader.TransportProtocol() != header.TCPProtocolNumber { - return false, false - } - - // We dont't match fragments. - if frag := netHeader.FragmentOffset(); frag != 0 { - if frag == 1 { - log.Warningf("Dropping TCP packet: malicious packet with fragment with fragment offest of 1.") - return false, true - } - return false, false - } - - // Now we need the transport header. However, this may not have been set - // yet. - // TODO - var tcpHeader header.TCP - if pkt.TransportHeader != nil { - tcpHeader = header.TCP(pkt.TransportHeader) - } else { - // The TCP header hasn't been parsed yet. We have to do it here. - if len(pkt.Data.First()) < header.TCPMinimumSize { - // There's no valid TCP header here, so we hotdrop the - // packet. - // TODO: Stats. - log.Warningf("Dropping TCP packet: size to small.") - return false, true - } - tcpHeader = header.TCP(pkt.Data.First()) - } - - // Check whether the source and destination ports are within the - // matching range. - sourcePort := tcpHeader.SourcePort() - destinationPort := tcpHeader.DestinationPort() - if sourcePort < tm.data.SourcePortStart || tm.data.SourcePortEnd < sourcePort { - return false, false - } - if destinationPort < tm.data.DestinationPortStart || tm.data.DestinationPortEnd < destinationPort { - return false, false - } - - return true, false -} -- cgit v1.2.3 From ad1968ed5665c7541d6920edbd7c7492b7db3046 Mon Sep 17 00:00:00 2001 From: Rahat Mahmood Date: Tue, 21 Jan 2020 14:25:14 -0800 Subject: Implement sysfs. PiperOrigin-RevId: 290822487 --- pkg/abi/linux/file.go | 13 +++ pkg/sentry/fsimpl/kernfs/BUILD | 3 +- pkg/sentry/fsimpl/kernfs/kernfs_test.go | 175 ++++++------------------------- pkg/sentry/fsimpl/proc/BUILD | 13 +-- pkg/sentry/fsimpl/proc/boot_test.go | 149 -------------------------- pkg/sentry/fsimpl/proc/tasks_test.go | 13 +-- pkg/sentry/fsimpl/sys/BUILD | 35 +++++++ pkg/sentry/fsimpl/sys/sys.go | 124 ++++++++++++++++++++++ pkg/sentry/fsimpl/sys/sys_test.go | 90 ++++++++++++++++ pkg/sentry/fsimpl/testutil/BUILD | 36 +++++++ pkg/sentry/fsimpl/testutil/kernel.go | 149 ++++++++++++++++++++++++++ pkg/sentry/fsimpl/testutil/testutil.go | 178 ++++++++++++++++++++++++++++++++ 12 files changed, 663 insertions(+), 315 deletions(-) delete mode 100644 pkg/sentry/fsimpl/proc/boot_test.go create mode 100644 pkg/sentry/fsimpl/sys/BUILD create mode 100644 pkg/sentry/fsimpl/sys/sys.go create mode 100644 pkg/sentry/fsimpl/sys/sys_test.go create mode 100644 pkg/sentry/fsimpl/testutil/BUILD create mode 100644 pkg/sentry/fsimpl/testutil/kernel.go create mode 100644 pkg/sentry/fsimpl/testutil/testutil.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/file.go b/pkg/abi/linux/file.go index 16791d03e..6fbdd668d 100644 --- a/pkg/abi/linux/file.go +++ b/pkg/abi/linux/file.go @@ -180,6 +180,19 @@ const ( DT_WHT = 14 ) +// DirentType are the friendly strings for linux_dirent64.d_type. +var DirentType = abi.ValueSet{ + DT_UNKNOWN: "DT_UNKNOWN", + DT_FIFO: "DT_FIFO", + DT_CHR: "DT_CHR", + DT_DIR: "DT_DIR", + DT_BLK: "DT_BLK", + DT_REG: "DT_REG", + DT_LNK: "DT_LNK", + DT_SOCK: "DT_SOCK", + DT_WHT: "DT_WHT", +} + // Values for preadv2/pwritev2. const ( // Note: gVisor does not implement the RWF_HIPRI feature, but the flag is diff --git a/pkg/sentry/fsimpl/kernfs/BUILD b/pkg/sentry/fsimpl/kernfs/BUILD index 809178250..66d409785 100644 --- a/pkg/sentry/fsimpl/kernfs/BUILD +++ b/pkg/sentry/fsimpl/kernfs/BUILD @@ -51,13 +51,12 @@ go_test( deps = [ ":kernfs", "//pkg/abi/linux", - "//pkg/fspath", "//pkg/sentry/context", "//pkg/sentry/context/contexttest", + "//pkg/sentry/fsimpl/testutil", "//pkg/sentry/kernel/auth", "//pkg/sentry/usermem", "//pkg/sentry/vfs", - "//pkg/sync", "//pkg/syserror", "@com_github_google_go-cmp//cmp:go_default_library", ], diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index 5c9d580e1..a5fdfbde5 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -17,20 +17,17 @@ package kernfs_test import ( "bytes" "fmt" - "io" - "runtime" "testing" "github.com/google/go-cmp/cmp" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/context/contexttest" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/testutil" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) @@ -41,21 +38,11 @@ const staticFileContent = "This is sample content for a static test file." // filesystem. See newTestSystem. type RootDentryFn func(*auth.Credentials, *filesystem) *kernfs.Dentry -// TestSystem represents the context for a single test. -type TestSystem struct { - t *testing.T - ctx context.Context - creds *auth.Credentials - vfs *vfs.VirtualFilesystem - mns *vfs.MountNamespace - root vfs.VirtualDentry -} - // newTestSystem sets up a minimal environment for running a test, including an // instance of a test filesystem. Tests can control the contents of the // filesystem by providing an appropriate rootFn, which should return a // pre-populated root dentry. -func newTestSystem(t *testing.T, rootFn RootDentryFn) *TestSystem { +func newTestSystem(t *testing.T, rootFn RootDentryFn) *testutil.System { ctx := contexttest.Context(t) creds := auth.CredentialsFromContext(ctx) v := vfs.New() @@ -66,57 +53,7 @@ func newTestSystem(t *testing.T, rootFn RootDentryFn) *TestSystem { if err != nil { t.Fatalf("Failed to create testfs root mount: %v", err) } - - s := &TestSystem{ - t: t, - ctx: ctx, - creds: creds, - vfs: v, - mns: mns, - root: mns.Root(), - } - runtime.SetFinalizer(s, func(s *TestSystem) { s.root.DecRef() }) - return s -} - -// PathOpAtRoot constructs a vfs.PathOperation for a path from the -// root of the test filesystem. -// -// Precondition: path should be relative path. -func (s *TestSystem) PathOpAtRoot(path string) vfs.PathOperation { - return vfs.PathOperation{ - Root: s.root, - Start: s.root, - Path: fspath.Parse(path), - } -} - -// GetDentryOrDie attempts to resolve a dentry referred to by the -// provided path operation. If unsuccessful, the test fails. -func (s *TestSystem) GetDentryOrDie(pop vfs.PathOperation) vfs.VirtualDentry { - vd, err := s.vfs.GetDentryAt(s.ctx, s.creds, &pop, &vfs.GetDentryOptions{}) - if err != nil { - s.t.Fatalf("GetDentryAt(pop:%+v) failed: %v", pop, err) - } - return vd -} - -func (s *TestSystem) ReadToEnd(fd *vfs.FileDescription) (string, error) { - buf := make([]byte, usermem.PageSize) - bufIOSeq := usermem.BytesIOSequence(buf) - opts := vfs.ReadOptions{} - - var content bytes.Buffer - for { - n, err := fd.Impl().Read(s.ctx, bufIOSeq, opts) - if n == 0 || err != nil { - if err == io.EOF { - err = nil - } - return content.String(), err - } - content.Write(buf[:n]) - } + return testutil.NewSystem(ctx, t, v, mns) } type fsType struct { @@ -260,6 +197,7 @@ func TestBasic(t *testing.T) { "file1": fs.newFile(creds, staticFileContent), }) }) + defer sys.Destroy() sys.GetDentryOrDie(sys.PathOpAtRoot("file1")).DecRef() } @@ -269,9 +207,10 @@ func TestMkdirGetDentry(t *testing.T) { "dir1": fs.newDir(creds, 0755, nil), }) }) + defer sys.Destroy() pop := sys.PathOpAtRoot("dir1/a new directory") - if err := sys.vfs.MkdirAt(sys.ctx, sys.creds, &pop, &vfs.MkdirOptions{Mode: 0755}); err != nil { + if err := sys.VFS.MkdirAt(sys.Ctx, sys.Creds, &pop, &vfs.MkdirOptions{Mode: 0755}); err != nil { t.Fatalf("MkdirAt for PathOperation %+v failed: %v", pop, err) } sys.GetDentryOrDie(pop).DecRef() @@ -283,20 +222,21 @@ func TestReadStaticFile(t *testing.T) { "file1": fs.newFile(creds, staticFileContent), }) }) + defer sys.Destroy() pop := sys.PathOpAtRoot("file1") - fd, err := sys.vfs.OpenAt(sys.ctx, sys.creds, &pop, &vfs.OpenOptions{}) + fd, err := sys.VFS.OpenAt(sys.Ctx, sys.Creds, &pop, &vfs.OpenOptions{}) if err != nil { - sys.t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) + t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) } defer fd.DecRef() content, err := sys.ReadToEnd(fd) if err != nil { - sys.t.Fatalf("Read failed: %v", err) + t.Fatalf("Read failed: %v", err) } if diff := cmp.Diff(staticFileContent, content); diff != "" { - sys.t.Fatalf("Read returned unexpected data:\n--- want\n+++ got\n%v", diff) + t.Fatalf("Read returned unexpected data:\n--- want\n+++ got\n%v", diff) } } @@ -306,83 +246,44 @@ func TestCreateNewFileInStaticDir(t *testing.T) { "dir1": fs.newDir(creds, 0755, nil), }) }) + defer sys.Destroy() pop := sys.PathOpAtRoot("dir1/newfile") opts := &vfs.OpenOptions{Flags: linux.O_CREAT | linux.O_EXCL, Mode: defaultMode} - fd, err := sys.vfs.OpenAt(sys.ctx, sys.creds, &pop, opts) + fd, err := sys.VFS.OpenAt(sys.Ctx, sys.Creds, &pop, opts) if err != nil { - sys.t.Fatalf("OpenAt(pop:%+v, opts:%+v) failed: %v", pop, opts, err) + t.Fatalf("OpenAt(pop:%+v, opts:%+v) failed: %v", pop, opts, err) } // Close the file. The file should persist. fd.DecRef() - fd, err = sys.vfs.OpenAt(sys.ctx, sys.creds, &pop, &vfs.OpenOptions{}) + fd, err = sys.VFS.OpenAt(sys.Ctx, sys.Creds, &pop, &vfs.OpenOptions{}) if err != nil { - sys.t.Fatalf("OpenAt(pop:%+v) = %+v failed: %v", pop, fd, err) + t.Fatalf("OpenAt(pop:%+v) = %+v failed: %v", pop, fd, err) } fd.DecRef() } -// direntCollector provides an implementation for vfs.IterDirentsCallback for -// testing. It simply iterates to the end of a given directory FD and collects -// all dirents emitted by the callback. -type direntCollector struct { - mu sync.Mutex - dirents map[string]vfs.Dirent -} - -// Handle implements vfs.IterDirentsCallback.Handle. -func (d *direntCollector) Handle(dirent vfs.Dirent) bool { - d.mu.Lock() - if d.dirents == nil { - d.dirents = make(map[string]vfs.Dirent) - } - d.dirents[dirent.Name] = dirent - d.mu.Unlock() - return true -} - -// count returns the number of dirents currently in the collector. -func (d *direntCollector) count() int { - d.mu.Lock() - defer d.mu.Unlock() - return len(d.dirents) -} - -// contains checks whether the collector has a dirent with the given name and -// type. -func (d *direntCollector) contains(name string, typ uint8) error { - d.mu.Lock() - defer d.mu.Unlock() - dirent, ok := d.dirents[name] - if !ok { - return fmt.Errorf("No dirent named %q found", name) - } - if dirent.Type != typ { - return fmt.Errorf("Dirent named %q found, but was expecting type %d, got: %+v", name, typ, dirent) - } - return nil -} - func TestDirFDReadWrite(t *testing.T) { sys := newTestSystem(t, func(creds *auth.Credentials, fs *filesystem) *kernfs.Dentry { return fs.newReadonlyDir(creds, 0755, nil) }) + defer sys.Destroy() pop := sys.PathOpAtRoot("/") - fd, err := sys.vfs.OpenAt(sys.ctx, sys.creds, &pop, &vfs.OpenOptions{}) + fd, err := sys.VFS.OpenAt(sys.Ctx, sys.Creds, &pop, &vfs.OpenOptions{}) if err != nil { - sys.t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) + t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) } defer fd.DecRef() // Read/Write should fail for directory FDs. - if _, err := fd.Read(sys.ctx, usermem.BytesIOSequence([]byte{}), vfs.ReadOptions{}); err != syserror.EISDIR { - sys.t.Fatalf("Read for directory FD failed with unexpected error: %v", err) + if _, err := fd.Read(sys.Ctx, usermem.BytesIOSequence([]byte{}), vfs.ReadOptions{}); err != syserror.EISDIR { + t.Fatalf("Read for directory FD failed with unexpected error: %v", err) } - if _, err := fd.Write(sys.ctx, usermem.BytesIOSequence([]byte{}), vfs.WriteOptions{}); err != syserror.EISDIR { - sys.t.Fatalf("Wrire for directory FD failed with unexpected error: %v", err) + if _, err := fd.Write(sys.Ctx, usermem.BytesIOSequence([]byte{}), vfs.WriteOptions{}); err != syserror.EISDIR { + t.Fatalf("Write for directory FD failed with unexpected error: %v", err) } } @@ -397,30 +298,12 @@ func TestDirFDIterDirents(t *testing.T) { "file1": fs.newFile(creds, staticFileContent), }) }) + defer sys.Destroy() pop := sys.PathOpAtRoot("/") - fd, err := sys.vfs.OpenAt(sys.ctx, sys.creds, &pop, &vfs.OpenOptions{}) - if err != nil { - sys.t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) - } - defer fd.DecRef() - - collector := &direntCollector{} - if err := fd.IterDirents(sys.ctx, collector); err != nil { - sys.t.Fatalf("IterDirent failed: %v", err) - } - - // Root directory should contain ".", ".." and 3 children: - if collector.count() != 5 { - sys.t.Fatalf("IterDirent returned too many dirents") - } - for _, dirName := range []string{".", "..", "dir1", "dir2"} { - if err := collector.contains(dirName, linux.DT_DIR); err != nil { - sys.t.Fatalf("IterDirent had unexpected results: %v", err) - } - } - if err := collector.contains("file1", linux.DT_REG); err != nil { - sys.t.Fatalf("IterDirent had unexpected results: %v", err) - } - + sys.AssertDirectoryContains(&pop, map[string]testutil.DirentType{ + "dir1": linux.DT_DIR, + "dir2": linux.DT_DIR, + "file1": linux.DT_REG, + }) } diff --git a/pkg/sentry/fsimpl/proc/BUILD b/pkg/sentry/fsimpl/proc/BUILD index f69aa19c4..c5b79fb38 100644 --- a/pkg/sentry/fsimpl/proc/BUILD +++ b/pkg/sentry/fsimpl/proc/BUILD @@ -44,30 +44,19 @@ go_test( name = "proc_test", size = "small", srcs = [ - "boot_test.go", "tasks_sys_test.go", "tasks_test.go", ], embed = [":proc"], deps = [ "//pkg/abi/linux", - "//pkg/cpuid", "//pkg/fspath", - "//pkg/memutil", "//pkg/sentry/context", "//pkg/sentry/context/contexttest", - "//pkg/sentry/fs", + "//pkg/sentry/fsimpl/testutil", "//pkg/sentry/inet", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/sched", - "//pkg/sentry/limits", - "//pkg/sentry/loader", - "//pkg/sentry/pgalloc", - "//pkg/sentry/platform", - "//pkg/sentry/platform/kvm", - "//pkg/sentry/platform/ptrace", - "//pkg/sentry/time", "//pkg/sentry/usermem", "//pkg/sentry/vfs", "//pkg/syserror", diff --git a/pkg/sentry/fsimpl/proc/boot_test.go b/pkg/sentry/fsimpl/proc/boot_test.go deleted file mode 100644 index 84a93ee56..000000000 --- a/pkg/sentry/fsimpl/proc/boot_test.go +++ /dev/null @@ -1,149 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "flag" - "fmt" - "os" - "runtime" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/cpuid" - "gvisor.dev/gvisor/pkg/memutil" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/kernel/sched" - "gvisor.dev/gvisor/pkg/sentry/limits" - "gvisor.dev/gvisor/pkg/sentry/loader" - "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/time" - - // Platforms are plugable. - _ "gvisor.dev/gvisor/pkg/sentry/platform/kvm" - _ "gvisor.dev/gvisor/pkg/sentry/platform/ptrace" -) - -var ( - platformFlag = flag.String("platform", "ptrace", "specify which platform to use") -) - -// boot initializes a new bare bones kernel for test. -func boot() (*kernel.Kernel, error) { - platformCtr, err := platform.Lookup(*platformFlag) - if err != nil { - return nil, fmt.Errorf("platform not found: %v", err) - } - deviceFile, err := platformCtr.OpenDevice() - if err != nil { - return nil, fmt.Errorf("creating platform: %v", err) - } - plat, err := platformCtr.New(deviceFile) - if err != nil { - return nil, fmt.Errorf("creating platform: %v", err) - } - - k := &kernel.Kernel{ - Platform: plat, - } - - mf, err := createMemoryFile() - if err != nil { - return nil, err - } - k.SetMemoryFile(mf) - - // Pass k as the platform since it is savable, unlike the actual platform. - vdso, err := loader.PrepareVDSO(nil, k) - if err != nil { - return nil, fmt.Errorf("creating vdso: %v", err) - } - - // Create timekeeper. - tk, err := kernel.NewTimekeeper(k, vdso.ParamPage.FileRange()) - if err != nil { - return nil, fmt.Errorf("creating timekeeper: %v", err) - } - tk.SetClocks(time.NewCalibratedClocks()) - - creds := auth.NewRootCredentials(auth.NewRootUserNamespace()) - - // Initiate the Kernel object, which is required by the Context passed - // to createVFS in order to mount (among other things) procfs. - if err = k.Init(kernel.InitKernelArgs{ - ApplicationCores: uint(runtime.GOMAXPROCS(-1)), - FeatureSet: cpuid.HostFeatureSet(), - Timekeeper: tk, - RootUserNamespace: creds.UserNamespace, - Vdso: vdso, - RootUTSNamespace: kernel.NewUTSNamespace("hostname", "domain", creds.UserNamespace), - RootIPCNamespace: kernel.NewIPCNamespace(creds.UserNamespace), - RootAbstractSocketNamespace: kernel.NewAbstractSocketNamespace(), - PIDNamespace: kernel.NewRootPIDNamespace(creds.UserNamespace), - }); err != nil { - return nil, fmt.Errorf("initializing kernel: %v", err) - } - - ctx := k.SupervisorContext() - - // Create mount namespace without root as it's the minimum required to create - // the global thread group. - mntns, err := fs.NewMountNamespace(ctx, nil) - if err != nil { - return nil, err - } - ls, err := limits.NewLinuxLimitSet() - if err != nil { - return nil, err - } - tg := k.NewThreadGroup(mntns, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, ls) - k.TestOnly_SetGlobalInit(tg) - - return k, nil -} - -// createTask creates a new bare bones task for tests. -func createTask(ctx context.Context, name string, tc *kernel.ThreadGroup) (*kernel.Task, error) { - k := kernel.KernelFromContext(ctx) - config := &kernel.TaskConfig{ - Kernel: k, - ThreadGroup: tc, - TaskContext: &kernel.TaskContext{Name: name}, - Credentials: auth.CredentialsFromContext(ctx), - AllowedCPUMask: sched.NewFullCPUSet(k.ApplicationCores()), - UTSNamespace: kernel.UTSNamespaceFromContext(ctx), - IPCNamespace: kernel.IPCNamespaceFromContext(ctx), - AbstractSocketNamespace: kernel.NewAbstractSocketNamespace(), - } - return k.TaskSet().NewTask(config) -} - -func createMemoryFile() (*pgalloc.MemoryFile, error) { - const memfileName = "test-memory" - memfd, err := memutil.CreateMemFD(memfileName, 0) - if err != nil { - return nil, fmt.Errorf("error creating memfd: %v", err) - } - memfile := os.NewFile(uintptr(memfd), memfileName) - mf, err := pgalloc.NewMemoryFile(memfile, pgalloc.MemoryFileOpts{}) - if err != nil { - memfile.Close() - return nil, fmt.Errorf("error creating pgalloc.MemoryFile: %v", err) - } - return mf, nil -} diff --git a/pkg/sentry/fsimpl/proc/tasks_test.go b/pkg/sentry/fsimpl/proc/tasks_test.go index 002d2f73b..41977d816 100644 --- a/pkg/sentry/fsimpl/proc/tasks_test.go +++ b/pkg/sentry/fsimpl/proc/tasks_test.go @@ -24,6 +24,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/testutil" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/usermem" @@ -134,7 +135,7 @@ func checkFiles(gots []vfs.Dirent, wants map[string]vfs.Dirent) ([]vfs.Dirent, e } func setup() (context.Context, *vfs.VirtualFilesystem, vfs.VirtualDentry, error) { - k, err := boot() + k, err := testutil.Boot() if err != nil { return nil, nil, vfs.VirtualDentry{}, fmt.Errorf("creating kernel: %v", err) } @@ -206,7 +207,7 @@ func TestTasks(t *testing.T) { var tasks []*kernel.Task for i := 0; i < 5; i++ { tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - task, err := createTask(ctx, fmt.Sprintf("name-%d", i), tc) + task, err := testutil.CreateTask(ctx, fmt.Sprintf("name-%d", i), tc) if err != nil { t.Fatalf("CreateTask(): %v", err) } @@ -298,7 +299,7 @@ func TestTasksOffset(t *testing.T) { k := kernel.KernelFromContext(ctx) for i := 0; i < 3; i++ { tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - if _, err := createTask(ctx, fmt.Sprintf("name-%d", i), tc); err != nil { + if _, err := testutil.CreateTask(ctx, fmt.Sprintf("name-%d", i), tc); err != nil { t.Fatalf("CreateTask(): %v", err) } } @@ -417,7 +418,7 @@ func TestTask(t *testing.T) { k := kernel.KernelFromContext(ctx) tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - _, err = createTask(ctx, "name", tc) + _, err = testutil.CreateTask(ctx, "name", tc) if err != nil { t.Fatalf("CreateTask(): %v", err) } @@ -458,7 +459,7 @@ func TestProcSelf(t *testing.T) { k := kernel.KernelFromContext(ctx) tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - task, err := createTask(ctx, "name", tc) + task, err := testutil.CreateTask(ctx, "name", tc) if err != nil { t.Fatalf("CreateTask(): %v", err) } @@ -555,7 +556,7 @@ func TestTree(t *testing.T) { var tasks []*kernel.Task for i := 0; i < 5; i++ { tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - task, err := createTask(uberCtx, fmt.Sprintf("name-%d", i), tc) + task, err := testutil.CreateTask(uberCtx, fmt.Sprintf("name-%d", i), tc) if err != nil { t.Fatalf("CreateTask(): %v", err) } diff --git a/pkg/sentry/fsimpl/sys/BUILD b/pkg/sentry/fsimpl/sys/BUILD new file mode 100644 index 000000000..ee3c842bd --- /dev/null +++ b/pkg/sentry/fsimpl/sys/BUILD @@ -0,0 +1,35 @@ +load("//tools/go_stateify:defs.bzl", "go_library") +load("@io_bazel_rules_go//go:def.bzl", "go_test") + +package(licenses = ["notice"]) + +go_library( + name = "sys", + srcs = [ + "sys.go", + ], + importpath = "gvisor.dev/gvisor/pkg/sentry/fsimpl/sys", + deps = [ + "//pkg/abi/linux", + "//pkg/sentry/context", + "//pkg/sentry/fsimpl/kernfs", + "//pkg/sentry/kernel", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/vfs", + "//pkg/syserror", + ], +) + +go_test( + name = "sys_test", + srcs = ["sys_test.go"], + deps = [ + ":sys", + "//pkg/abi/linux", + "//pkg/sentry/fsimpl/testutil", + "//pkg/sentry/kernel", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/vfs", + "@com_github_google_go-cmp//cmp:go_default_library", + ], +) diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go new file mode 100644 index 000000000..1305ad01d --- /dev/null +++ b/pkg/sentry/fsimpl/sys/sys.go @@ -0,0 +1,124 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package sys implements sysfs. +package sys + +import ( + "bytes" + "fmt" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// FilesystemType implements vfs.FilesystemType. +type FilesystemType struct{} + +// filesystem implements vfs.FilesystemImpl. +type filesystem struct { + kernfs.Filesystem +} + +// GetFilesystem implements vfs.FilesystemType.GetFilesystem. +func (FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { + fs := &filesystem{} + fs.Filesystem.Init(vfsObj) + k := kernel.KernelFromContext(ctx) + maxCPUCores := k.ApplicationCores() + defaultSysDirMode := linux.FileMode(0755) + + root := fs.newDir(creds, defaultSysDirMode, map[string]*kernfs.Dentry{ + "block": fs.newDir(creds, defaultSysDirMode, nil), + "bus": fs.newDir(creds, defaultSysDirMode, nil), + "class": fs.newDir(creds, defaultSysDirMode, map[string]*kernfs.Dentry{ + "power_supply": fs.newDir(creds, defaultSysDirMode, nil), + }), + "dev": fs.newDir(creds, defaultSysDirMode, nil), + "devices": fs.newDir(creds, defaultSysDirMode, map[string]*kernfs.Dentry{ + "system": fs.newDir(creds, defaultSysDirMode, map[string]*kernfs.Dentry{ + "cpu": fs.newDir(creds, defaultSysDirMode, map[string]*kernfs.Dentry{ + "online": fs.newCPUFile(creds, maxCPUCores, linux.FileMode(0444)), + "possible": fs.newCPUFile(creds, maxCPUCores, linux.FileMode(0444)), + "present": fs.newCPUFile(creds, maxCPUCores, linux.FileMode(0444)), + }), + }), + }), + "firmware": fs.newDir(creds, defaultSysDirMode, nil), + "fs": fs.newDir(creds, defaultSysDirMode, nil), + "kernel": fs.newDir(creds, defaultSysDirMode, nil), + "module": fs.newDir(creds, defaultSysDirMode, nil), + "power": fs.newDir(creds, defaultSysDirMode, nil), + }) + return fs.VFSFilesystem(), root.VFSDentry(), nil +} + +// dir implements kernfs.Inode. +type dir struct { + kernfs.InodeAttrs + kernfs.InodeNoDynamicLookup + kernfs.InodeNotSymlink + kernfs.InodeDirectoryNoNewChildren + + kernfs.OrderedChildren + dentry kernfs.Dentry +} + +func (fs *filesystem) newDir(creds *auth.Credentials, mode linux.FileMode, contents map[string]*kernfs.Dentry) *kernfs.Dentry { + d := &dir{} + d.InodeAttrs.Init(creds, fs.NextIno(), linux.ModeDirectory|0755) + d.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) + d.dentry.Init(d) + + d.IncLinks(d.OrderedChildren.Populate(&d.dentry, contents)) + + return &d.dentry +} + +// SetStat implements kernfs.Inode.SetStat. +func (d *dir) SetStat(fs *vfs.Filesystem, opts vfs.SetStatOptions) error { + return syserror.EPERM +} + +// Open implements kernfs.Inode.Open. +func (d *dir) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { + fd := &kernfs.GenericDirectoryFD{} + fd.Init(rp.Mount(), vfsd, &d.OrderedChildren, flags) + return fd.VFSFileDescription(), nil +} + +// cpuFile implements kernfs.Inode. +type cpuFile struct { + kernfs.DynamicBytesFile + maxCores uint +} + +// Generate implements vfs.DynamicBytesSource.Generate. +func (c *cpuFile) Generate(ctx context.Context, buf *bytes.Buffer) error { + fmt.Fprintf(buf, "0-%d", c.maxCores-1) + return nil +} + +func (fs *filesystem) newCPUFile(creds *auth.Credentials, maxCores uint, mode linux.FileMode) *kernfs.Dentry { + c := &cpuFile{maxCores: maxCores} + c.DynamicBytesFile.Init(creds, fs.NextIno(), c, mode) + d := &kernfs.Dentry{} + d.Init(c) + return d +} diff --git a/pkg/sentry/fsimpl/sys/sys_test.go b/pkg/sentry/fsimpl/sys/sys_test.go new file mode 100644 index 000000000..60a1634a9 --- /dev/null +++ b/pkg/sentry/fsimpl/sys/sys_test.go @@ -0,0 +1,90 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package sys_test + +import ( + "fmt" + "testing" + + "github.com/google/go-cmp/cmp" + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/sys" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/testutil" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" +) + +func newTestSystem(t *testing.T) *testutil.System { + k, err := testutil.Boot() + if err != nil { + t.Fatalf("Failed to create test kernel: %v", err) + } + ctx := k.SupervisorContext() + creds := auth.CredentialsFromContext(ctx) + v := vfs.New() + v.MustRegisterFilesystemType("sysfs", sys.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + }) + + mns, err := v.NewMountNamespace(ctx, creds, "", "sysfs", &vfs.GetFilesystemOptions{}) + if err != nil { + t.Fatalf("Failed to create new mount namespace: %v", err) + } + return testutil.NewSystem(ctx, t, v, mns) +} + +func TestReadCPUFile(t *testing.T) { + s := newTestSystem(t) + defer s.Destroy() + k := kernel.KernelFromContext(s.Ctx) + maxCPUCores := k.ApplicationCores() + + expected := fmt.Sprintf("0-%d", maxCPUCores-1) + + for _, fname := range []string{"online", "possible", "present"} { + pop := s.PathOpAtRoot(fmt.Sprintf("devices/system/cpu/%s", fname)) + fd, err := s.VFS.OpenAt(s.Ctx, s.Creds, &pop, &vfs.OpenOptions{}) + if err != nil { + t.Fatalf("OpenAt(pop:%+v) = %+v failed: %v", pop, fd, err) + } + defer fd.DecRef() + content, err := s.ReadToEnd(fd) + if err != nil { + t.Fatalf("Read failed: %v", err) + } + if diff := cmp.Diff(expected, content); diff != "" { + t.Fatalf("Read returned unexpected data:\n--- want\n+++ got\n%v", diff) + } + } +} + +func TestSysRootContainsExpectedEntries(t *testing.T) { + s := newTestSystem(t) + defer s.Destroy() + pop := s.PathOpAtRoot("/") + s.AssertDirectoryContains(&pop, map[string]testutil.DirentType{ + "block": linux.DT_DIR, + "bus": linux.DT_DIR, + "class": linux.DT_DIR, + "dev": linux.DT_DIR, + "devices": linux.DT_DIR, + "firmware": linux.DT_DIR, + "fs": linux.DT_DIR, + "kernel": linux.DT_DIR, + "module": linux.DT_DIR, + "power": linux.DT_DIR, + }) +} diff --git a/pkg/sentry/fsimpl/testutil/BUILD b/pkg/sentry/fsimpl/testutil/BUILD new file mode 100644 index 000000000..4e70d84a7 --- /dev/null +++ b/pkg/sentry/fsimpl/testutil/BUILD @@ -0,0 +1,36 @@ +load("//tools/go_stateify:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_library( + name = "testutil", + testonly = 1, + srcs = [ + "kernel.go", + "testutil.go", + ], + importpath = "gvisor.dev/gvisor/pkg/sentry/fsimpl/testutil", + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/abi/linux", + "//pkg/cpuid", + "//pkg/fspath", + "//pkg/memutil", + "//pkg/sentry/context", + "//pkg/sentry/fs", + "//pkg/sentry/kernel", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/kernel/sched", + "//pkg/sentry/limits", + "//pkg/sentry/loader", + "//pkg/sentry/pgalloc", + "//pkg/sentry/platform", + "//pkg/sentry/platform/kvm", + "//pkg/sentry/platform/ptrace", + "//pkg/sentry/time", + "//pkg/sentry/usermem", + "//pkg/sentry/vfs", + "//pkg/sync", + "@com_github_google_go-cmp//cmp:go_default_library", + ], +) diff --git a/pkg/sentry/fsimpl/testutil/kernel.go b/pkg/sentry/fsimpl/testutil/kernel.go new file mode 100644 index 000000000..295da2d52 --- /dev/null +++ b/pkg/sentry/fsimpl/testutil/kernel.go @@ -0,0 +1,149 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package testutil + +import ( + "flag" + "fmt" + "os" + "runtime" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/cpuid" + "gvisor.dev/gvisor/pkg/memutil" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/kernel/sched" + "gvisor.dev/gvisor/pkg/sentry/limits" + "gvisor.dev/gvisor/pkg/sentry/loader" + "gvisor.dev/gvisor/pkg/sentry/pgalloc" + "gvisor.dev/gvisor/pkg/sentry/platform" + "gvisor.dev/gvisor/pkg/sentry/time" + + // Platforms are plugable. + _ "gvisor.dev/gvisor/pkg/sentry/platform/kvm" + _ "gvisor.dev/gvisor/pkg/sentry/platform/ptrace" +) + +var ( + platformFlag = flag.String("platform", "ptrace", "specify which platform to use") +) + +// Boot initializes a new bare bones kernel for test. +func Boot() (*kernel.Kernel, error) { + platformCtr, err := platform.Lookup(*platformFlag) + if err != nil { + return nil, fmt.Errorf("platform not found: %v", err) + } + deviceFile, err := platformCtr.OpenDevice() + if err != nil { + return nil, fmt.Errorf("creating platform: %v", err) + } + plat, err := platformCtr.New(deviceFile) + if err != nil { + return nil, fmt.Errorf("creating platform: %v", err) + } + + k := &kernel.Kernel{ + Platform: plat, + } + + mf, err := createMemoryFile() + if err != nil { + return nil, err + } + k.SetMemoryFile(mf) + + // Pass k as the platform since it is savable, unlike the actual platform. + vdso, err := loader.PrepareVDSO(nil, k) + if err != nil { + return nil, fmt.Errorf("creating vdso: %v", err) + } + + // Create timekeeper. + tk, err := kernel.NewTimekeeper(k, vdso.ParamPage.FileRange()) + if err != nil { + return nil, fmt.Errorf("creating timekeeper: %v", err) + } + tk.SetClocks(time.NewCalibratedClocks()) + + creds := auth.NewRootCredentials(auth.NewRootUserNamespace()) + + // Initiate the Kernel object, which is required by the Context passed + // to createVFS in order to mount (among other things) procfs. + if err = k.Init(kernel.InitKernelArgs{ + ApplicationCores: uint(runtime.GOMAXPROCS(-1)), + FeatureSet: cpuid.HostFeatureSet(), + Timekeeper: tk, + RootUserNamespace: creds.UserNamespace, + Vdso: vdso, + RootUTSNamespace: kernel.NewUTSNamespace("hostname", "domain", creds.UserNamespace), + RootIPCNamespace: kernel.NewIPCNamespace(creds.UserNamespace), + RootAbstractSocketNamespace: kernel.NewAbstractSocketNamespace(), + PIDNamespace: kernel.NewRootPIDNamespace(creds.UserNamespace), + }); err != nil { + return nil, fmt.Errorf("initializing kernel: %v", err) + } + + ctx := k.SupervisorContext() + + // Create mount namespace without root as it's the minimum required to create + // the global thread group. + mntns, err := fs.NewMountNamespace(ctx, nil) + if err != nil { + return nil, err + } + ls, err := limits.NewLinuxLimitSet() + if err != nil { + return nil, err + } + tg := k.NewThreadGroup(mntns, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, ls) + k.TestOnly_SetGlobalInit(tg) + + return k, nil +} + +// CreateTask creates a new bare bones task for tests. +func CreateTask(ctx context.Context, name string, tc *kernel.ThreadGroup) (*kernel.Task, error) { + k := kernel.KernelFromContext(ctx) + config := &kernel.TaskConfig{ + Kernel: k, + ThreadGroup: tc, + TaskContext: &kernel.TaskContext{Name: name}, + Credentials: auth.CredentialsFromContext(ctx), + AllowedCPUMask: sched.NewFullCPUSet(k.ApplicationCores()), + UTSNamespace: kernel.UTSNamespaceFromContext(ctx), + IPCNamespace: kernel.IPCNamespaceFromContext(ctx), + AbstractSocketNamespace: kernel.NewAbstractSocketNamespace(), + } + return k.TaskSet().NewTask(config) +} + +func createMemoryFile() (*pgalloc.MemoryFile, error) { + const memfileName = "test-memory" + memfd, err := memutil.CreateMemFD(memfileName, 0) + if err != nil { + return nil, fmt.Errorf("error creating memfd: %v", err) + } + memfile := os.NewFile(uintptr(memfd), memfileName) + mf, err := pgalloc.NewMemoryFile(memfile, pgalloc.MemoryFileOpts{}) + if err != nil { + memfile.Close() + return nil, fmt.Errorf("error creating pgalloc.MemoryFile: %v", err) + } + return mf, nil +} diff --git a/pkg/sentry/fsimpl/testutil/testutil.go b/pkg/sentry/fsimpl/testutil/testutil.go new file mode 100644 index 000000000..eada31d94 --- /dev/null +++ b/pkg/sentry/fsimpl/testutil/testutil.go @@ -0,0 +1,178 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package testutil provides common test utilities for kernfs-based +// filesystems. +package testutil + +import ( + "fmt" + "io" + "strings" + "testing" + + "github.com/google/go-cmp/cmp" + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sync" +) + +// System represents the context for a single test. +// +// Test systems must be explicitly destroyed with System.Destroy. +type System struct { + t *testing.T + Ctx context.Context + Creds *auth.Credentials + VFS *vfs.VirtualFilesystem + mns *vfs.MountNamespace + root vfs.VirtualDentry +} + +// NewSystem constructs a System. +// +// Precondition: Caller must hold a reference on mns, whose ownership +// is transferred to the new System. +func NewSystem(ctx context.Context, t *testing.T, v *vfs.VirtualFilesystem, mns *vfs.MountNamespace) *System { + s := &System{ + t: t, + Ctx: ctx, + Creds: auth.CredentialsFromContext(ctx), + VFS: v, + mns: mns, + root: mns.Root(), + } + return s +} + +// Destroy release resources associated with a test system. +func (s *System) Destroy() { + s.root.DecRef() + s.mns.DecRef(s.VFS) // Reference on mns passed to NewSystem. +} + +// ReadToEnd reads the contents of fd until EOF to a string. +func (s *System) ReadToEnd(fd *vfs.FileDescription) (string, error) { + buf := make([]byte, usermem.PageSize) + bufIOSeq := usermem.BytesIOSequence(buf) + opts := vfs.ReadOptions{} + + var content strings.Builder + for { + n, err := fd.Read(s.Ctx, bufIOSeq, opts) + if n == 0 || err != nil { + if err == io.EOF { + err = nil + } + return content.String(), err + } + content.Write(buf[:n]) + } +} + +// PathOpAtRoot constructs a PathOperation with the given path from +// the root of the filesystem. +func (s *System) PathOpAtRoot(path string) vfs.PathOperation { + return vfs.PathOperation{ + Root: s.root, + Start: s.root, + Path: fspath.Parse(path), + } +} + +// GetDentryOrDie attempts to resolve a dentry referred to by the +// provided path operation. If unsuccessful, the test fails. +func (s *System) GetDentryOrDie(pop vfs.PathOperation) vfs.VirtualDentry { + vd, err := s.VFS.GetDentryAt(s.Ctx, s.Creds, &pop, &vfs.GetDentryOptions{}) + if err != nil { + s.t.Fatalf("GetDentryAt(pop:%+v) failed: %v", pop, err) + } + return vd +} + +// DirentType is an alias for values for linux_dirent64.d_type. +type DirentType = uint8 + +// AssertDirectoryContains verifies that a directory at pop contains the entries +// specified. AssertDirectoryContains implicitly checks for "." and "..", these +// need not be included in entries. +func (s *System) AssertDirectoryContains(pop *vfs.PathOperation, entries map[string]DirentType) { + // Also implicitly check for "." and "..". + entries["."] = linux.DT_DIR + entries[".."] = linux.DT_DIR + + fd, err := s.VFS.OpenAt(s.Ctx, s.Creds, pop, &vfs.OpenOptions{Flags: linux.O_RDONLY}) + if err != nil { + s.t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) + } + defer fd.DecRef() + + collector := &DirentCollector{} + if err := fd.IterDirents(s.Ctx, collector); err != nil { + s.t.Fatalf("IterDirent failed: %v", err) + } + + collectedEntries := make(map[string]DirentType) + for _, dirent := range collector.dirents { + collectedEntries[dirent.Name] = dirent.Type + } + if diff := cmp.Diff(entries, collectedEntries); diff != "" { + s.t.Fatalf("IterDirent had unexpected results:\n--- want\n+++ got\n%v", diff) + } +} + +// DirentCollector provides an implementation for vfs.IterDirentsCallback for +// testing. It simply iterates to the end of a given directory FD and collects +// all dirents emitted by the callback. +type DirentCollector struct { + mu sync.Mutex + dirents map[string]vfs.Dirent +} + +// Handle implements vfs.IterDirentsCallback.Handle. +func (d *DirentCollector) Handle(dirent vfs.Dirent) bool { + d.mu.Lock() + if d.dirents == nil { + d.dirents = make(map[string]vfs.Dirent) + } + d.dirents[dirent.Name] = dirent + d.mu.Unlock() + return true +} + +// Count returns the number of dirents currently in the collector. +func (d *DirentCollector) Count() int { + d.mu.Lock() + defer d.mu.Unlock() + return len(d.dirents) +} + +// Contains checks whether the collector has a dirent with the given name and +// type. +func (d *DirentCollector) Contains(name string, typ uint8) error { + d.mu.Lock() + defer d.mu.Unlock() + dirent, ok := d.dirents[name] + if !ok { + return fmt.Errorf("No dirent named %q found", name) + } + if dirent.Type != typ { + return fmt.Errorf("Dirent named %q found, but was expecting type %s, got: %+v", name, linux.DirentType.Parse(uint64(typ)), dirent) + } + return nil +} -- cgit v1.2.3 From 7e6fbc6afe797752efe066a8aa86f9eca973f3a4 Mon Sep 17 00:00:00 2001 From: Mithun Iyer Date: Tue, 21 Jan 2020 14:47:04 -0800 Subject: Add a new TCP stat for current open connections. Such a stat accounts for all connections that are currently established and not yet transitioned to close state. Also fix bug in double increment of CurrentEstablished stat. Fixes #1579 PiperOrigin-RevId: 290827365 --- pkg/sentry/socket/netstack/netstack.go | 3 +- pkg/tcpip/tcpip.go | 6 ++- pkg/tcpip/transport/tcp/accept.go | 1 - pkg/tcpip/transport/tcp/connect.go | 1 + pkg/tcpip/transport/tcp/endpoint.go | 1 + pkg/tcpip/transport/tcp/tcp_test.go | 83 ++++++++++++++++++++++++++++++++++ 6 files changed, 92 insertions(+), 3 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 2662fbc0f..318acbeff 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -150,7 +150,8 @@ var Metrics = tcpip.Stats{ TCP: tcpip.TCPStats{ ActiveConnectionOpenings: mustCreateMetric("/netstack/tcp/active_connection_openings", "Number of connections opened successfully via Connect."), PassiveConnectionOpenings: mustCreateMetric("/netstack/tcp/passive_connection_openings", "Number of connections opened successfully via Listen."), - CurrentEstablished: mustCreateMetric("/netstack/tcp/current_established", "Number of connections in either ESTABLISHED or CLOSE-WAIT state now."), + CurrentEstablished: mustCreateMetric("/netstack/tcp/current_established", "Number of connections in ESTABLISHED state now."), + CurrentConnected: mustCreateMetric("/netstack/tcp/current_open", "Number of connections that are in connected state."), EstablishedResets: mustCreateMetric("/netstack/tcp/established_resets", "Number of times TCP connections have made a direct transition to the CLOSED state from either the ESTABLISHED state or the CLOSE-WAIT state"), EstablishedClosed: mustCreateMetric("/netstack/tcp/established_closed", "number of times established TCP connections made a transition to CLOSED state."), EstablishedTimedout: mustCreateMetric("/netstack/tcp/established_timedout", "Number of times an established connection was reset because of keep-alive time out."), diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 3fc823a36..59c9b3fb0 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -938,9 +938,13 @@ type TCPStats struct { PassiveConnectionOpenings *StatCounter // CurrentEstablished is the number of TCP connections for which the - // current state is either ESTABLISHED or CLOSE-WAIT. + // current state is ESTABLISHED. CurrentEstablished *StatCounter + // CurrentConnected is the number of TCP connections that + // are in connected state. + CurrentConnected *StatCounter + // EstablishedResets is the number of times TCP connections have made // a direct transition to the CLOSED state from either the // ESTABLISHED state or the CLOSE-WAIT state. diff --git a/pkg/tcpip/transport/tcp/accept.go b/pkg/tcpip/transport/tcp/accept.go index 1a2e3efa9..d469758eb 100644 --- a/pkg/tcpip/transport/tcp/accept.go +++ b/pkg/tcpip/transport/tcp/accept.go @@ -562,7 +562,6 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { // Switch state to connected. // We do not use transitionToStateEstablishedLocked here as there is // no handshake state available when doing a SYN cookie based accept. - n.stack.Stats().TCP.CurrentEstablished.Increment() n.isConnectNotified = true n.setEndpointState(StateEstablished) diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index a2f384384..4e3c5419c 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -934,6 +934,7 @@ func (e *endpoint) transitionToStateCloseLocked() { // Mark the endpoint as fully closed for reads/writes. e.cleanupLocked() e.setEndpointState(StateClose) + e.stack.Stats().TCP.CurrentConnected.Decrement() e.stack.Stats().TCP.EstablishedClosed.Increment() } diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 4797f11d1..13718ff55 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -594,6 +594,7 @@ func (e *endpoint) setEndpointState(state EndpointState) { switch state { case StateEstablished: e.stack.Stats().TCP.CurrentEstablished.Increment() + e.stack.Stats().TCP.CurrentConnected.Increment() case StateError: fallthrough case StateClose: diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index a9dfbe857..df2fb1071 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -470,6 +470,89 @@ func TestConnectResetAfterClose(t *testing.T) { } } +// TestCurrentConnectedIncrement tests increment of the current +// established and connected counters. +func TestCurrentConnectedIncrement(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + // Set TCPTimeWaitTimeout to 1 seconds so that sockets are marked closed + // after 1 second in TIME_WAIT state. + tcpTimeWaitTimeout := 1 * time.Second + if err := c.Stack().SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.TCPTimeWaitTimeoutOption(tcpTimeWaitTimeout)); err != nil { + t.Fatalf("c.stack.SetTransportProtocolOption(tcp, tcpip.TCPTimeWaitTimeout(%d) failed: %s", tcpTimeWaitTimeout, err) + } + + c.CreateConnected(789, 30000, -1 /* epRcvBuf */) + ep := c.EP + c.EP = nil + + if got := c.Stack().Stats().TCP.CurrentEstablished.Value(); got != 1 { + t.Errorf("got stats.TCP.CurrentEstablished.Value() = %v, want = 1", got) + } + gotConnected := c.Stack().Stats().TCP.CurrentConnected.Value() + if gotConnected != 1 { + t.Errorf("got stats.TCP.CurrentConnected.Value() = %v, want = 1", gotConnected) + } + + ep.Close() + + checker.IPv4(t, c.GetPacket(), + checker.TCP( + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS)+1), + checker.AckNum(790), + checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), + ), + ) + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: c.Port, + Flags: header.TCPFlagAck, + SeqNum: 790, + AckNum: c.IRS.Add(2), + RcvWnd: 30000, + }) + + if got := c.Stack().Stats().TCP.CurrentEstablished.Value(); got != 0 { + t.Errorf("got stats.TCP.CurrentEstablished.Value() = %v, want = 0", got) + } + if got := c.Stack().Stats().TCP.CurrentConnected.Value(); got != gotConnected { + t.Errorf("got stats.TCP.CurrentConnected.Value() = %v, want = %v", got, gotConnected) + } + + // Ack and send FIN as well. + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: c.Port, + Flags: header.TCPFlagAck | header.TCPFlagFin, + SeqNum: 790, + AckNum: c.IRS.Add(2), + RcvWnd: 30000, + }) + + // Check that the stack acks the FIN. + checker.IPv4(t, c.GetPacket(), + checker.PayloadLen(header.TCPMinimumSize), + checker.TCP( + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS)+2), + checker.AckNum(791), + checker.TCPFlags(header.TCPFlagAck), + ), + ) + + // Wait for the TIME-WAIT state to transition to CLOSED. + time.Sleep(1 * time.Second) + + if got := c.Stack().Stats().TCP.CurrentEstablished.Value(); got != 0 { + t.Errorf("got stats.TCP.CurrentEstablished.Value() = %v, want = 0", got) + } + if got := c.Stack().Stats().TCP.CurrentConnected.Value(); got != 0 { + t.Errorf("got stats.TCP.CurrentConnected.Value() = %v, want = 0", got) + } +} + // TestClosingWithEnqueuedSegments tests handling of still enqueued segments // when the endpoint transitions to StateClose. The in-flight segments would be // re-enqueued to a any listening endpoint. -- cgit v1.2.3 From 538053538dfb378aa8bc512d484ea305177e617b Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Tue, 21 Jan 2020 16:51:17 -0800 Subject: Adding serialization. --- pkg/sentry/socket/netfilter/netfilter.go | 29 ++++++++++++++++++++++++++++- pkg/tcpip/iptables/udp_matcher.go | 14 +++++++------- 2 files changed, 35 insertions(+), 8 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index f8ed1acbc..3caabca9a 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -196,7 +196,9 @@ func convertNetstackToBinary(tablename string, table iptables.Table) (linux.Kern } func marshalMatcher(matcher iptables.Matcher) []byte { - switch matcher.(type) { + switch m := matcher.(type) { + case *iptables.UDPMatcher: + return marshalUDPMatcher(m) default: // TODO(gvisor.dev/issue/170): We don't support any matchers // yet, so any call to marshalMatcher will panic. @@ -204,6 +206,31 @@ func marshalMatcher(matcher iptables.Matcher) []byte { } } +func marshalUDPMatcher(matcher *iptables.UDPMatcher) []byte { + type udpMatch struct { + linux.XTEntryMatch + linux.XTUDP + } + linuxMatcher := udpMatch{ + XTEntryMatch: linux.XTEntryMatch{ + MatchSize: linux.SizeOfXTEntryMatch + linux.SizeOfXTUDP, + // Name: "udp", + }, + XTUDP: linux.XTUDP{ + SourcePortStart: matcher.Data.SourcePortStart, + SourcePortEnd: matcher.Data.SourcePortEnd, + DestinationPortStart: matcher.Data.DestinationPortStart, + DestinationPortEnd: matcher.Data.DestinationPortEnd, + InverseFlags: matcher.Data.InverseFlags, + }, + } + copy(linuxMatcher.Name[:], "udp") + + var buf [linux.SizeOfXTEntryMatch + linux.SizeOfXTUDP]byte + binary.Marshal(buf[:], usermem.ByteOrder, linuxMatcher) + return buf[:] +} + func marshalTarget(target iptables.Target) []byte { switch target.(type) { case iptables.UnconditionalAcceptTarget: diff --git a/pkg/tcpip/iptables/udp_matcher.go b/pkg/tcpip/iptables/udp_matcher.go index ce4368a3d..fca457199 100644 --- a/pkg/tcpip/iptables/udp_matcher.go +++ b/pkg/tcpip/iptables/udp_matcher.go @@ -24,7 +24,7 @@ import ( ) type UDPMatcher struct { - data UDPMatcherData + Data UDPMatcherData // tablename string // unsigned int matchsize; @@ -62,11 +62,11 @@ func NewUDPMatcher(filter IPHeaderFilter, data UDPMatcherData) (Matcher, error) log.Warningf("UDP matching is only valid for protocol %d.", header.UDPProtocolNumber) } - return &UDPMatcher{data: data}, nil + return &UDPMatcher{Data: data}, nil } // TODO: Check xt_tcpudp.c. Need to check for same things (e.g. fragments). -func (tm *UDPMatcher) Match(hook Hook, pkt tcpip.PacketBuffer, interfaceName string) (bool, bool) { +func (um *UDPMatcher) Match(hook Hook, pkt tcpip.PacketBuffer, interfaceName string) (bool, bool) { log.Infof("UDPMatcher called from: %s", string(debug.Stack())) netHeader := header.IPv4(pkt.NetworkHeader) @@ -114,12 +114,12 @@ func (tm *UDPMatcher) Match(hook Hook, pkt tcpip.PacketBuffer, interfaceName str destinationPort := udpHeader.DestinationPort() log.Infof("UDPMatcher: sport and dport are %d and %d. sports and dport start and end are (%d, %d) and (%d, %d)", udpHeader.SourcePort(), udpHeader.DestinationPort(), - tm.data.SourcePortStart, tm.data.SourcePortEnd, - tm.data.DestinationPortStart, tm.data.DestinationPortEnd) - if sourcePort < tm.data.SourcePortStart || tm.data.SourcePortEnd < sourcePort { + um.Data.SourcePortStart, um.Data.SourcePortEnd, + um.Data.DestinationPortStart, um.Data.DestinationPortEnd) + if sourcePort < um.Data.SourcePortStart || um.Data.SourcePortEnd < sourcePort { return false, false } - if destinationPort < tm.data.DestinationPortStart || tm.data.DestinationPortEnd < destinationPort { + if destinationPort < um.Data.DestinationPortStart || um.Data.DestinationPortEnd < destinationPort { return false, false } -- cgit v1.2.3 From 6a59e7f510a7b12f8b3bd768dfe569033ef07d30 Mon Sep 17 00:00:00 2001 From: Ian Gudger Date: Tue, 21 Jan 2020 19:23:26 -0800 Subject: Rename DowngradableRWMutex to RWmutex. Also renames TMutex to Mutex. These custom mutexes aren't any worse than the standard library versions (same code), so having both seems redundant. PiperOrigin-RevId: 290873587 --- pkg/sentry/fs/overlay.go | 2 +- pkg/sentry/mm/mm.go | 4 +-- pkg/sync/aliases.go | 6 ---- pkg/sync/downgradable_rwmutex_test.go | 50 ++++++++++++++++----------------- pkg/sync/downgradable_rwmutex_unsafe.go | 26 ++++++++--------- pkg/sync/tmutex_test.go | 8 +++--- pkg/sync/tmutex_unsafe.go | 8 +++--- 7 files changed, 49 insertions(+), 55 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/overlay.go b/pkg/sentry/fs/overlay.go index 4cad55327..f7702f8f4 100644 --- a/pkg/sentry/fs/overlay.go +++ b/pkg/sentry/fs/overlay.go @@ -198,7 +198,7 @@ type overlayEntry struct { upper *Inode // dirCacheMu protects dirCache. - dirCacheMu sync.DowngradableRWMutex `state:"nosave"` + dirCacheMu sync.RWMutex `state:"nosave"` // dirCache is cache of DentAttrs from upper and lower Inodes. dirCache *SortedDentryMap diff --git a/pkg/sentry/mm/mm.go b/pkg/sentry/mm/mm.go index fa86ebced..78cc9e6e4 100644 --- a/pkg/sentry/mm/mm.go +++ b/pkg/sentry/mm/mm.go @@ -80,7 +80,7 @@ type MemoryManager struct { users int32 // mappingMu is analogous to Linux's struct mm_struct::mmap_sem. - mappingMu sync.DowngradableRWMutex `state:"nosave"` + mappingMu sync.RWMutex `state:"nosave"` // vmas stores virtual memory areas. Since vmas are stored by value, // clients should usually use vmaIterator.ValuePtr() instead of @@ -123,7 +123,7 @@ type MemoryManager struct { // activeMu is loosely analogous to Linux's struct // mm_struct::page_table_lock. - activeMu sync.DowngradableRWMutex `state:"nosave"` + activeMu sync.RWMutex `state:"nosave"` // pmas stores platform mapping areas used to implement vmas. Since pmas // are stored by value, clients should usually use pmaIterator.ValuePtr() diff --git a/pkg/sync/aliases.go b/pkg/sync/aliases.go index 20c7ca041..d2d7132fa 100644 --- a/pkg/sync/aliases.go +++ b/pkg/sync/aliases.go @@ -11,12 +11,6 @@ import ( // Aliases of standard library types. type ( - // Mutex is an alias of sync.Mutex. - Mutex = sync.Mutex - - // RWMutex is an alias of sync.RWMutex. - RWMutex = sync.RWMutex - // Cond is an alias of sync.Cond. Cond = sync.Cond diff --git a/pkg/sync/downgradable_rwmutex_test.go b/pkg/sync/downgradable_rwmutex_test.go index b5cb28ec0..ce667e825 100644 --- a/pkg/sync/downgradable_rwmutex_test.go +++ b/pkg/sync/downgradable_rwmutex_test.go @@ -18,7 +18,7 @@ import ( "testing" ) -func parallelReader(m *DowngradableRWMutex, clocked, cunlock, cdone chan bool) { +func parallelReader(m *RWMutex, clocked, cunlock, cdone chan bool) { m.RLock() clocked <- true <-cunlock @@ -28,7 +28,7 @@ func parallelReader(m *DowngradableRWMutex, clocked, cunlock, cdone chan bool) { func doTestParallelReaders(numReaders, gomaxprocs int) { runtime.GOMAXPROCS(gomaxprocs) - var m DowngradableRWMutex + var m RWMutex clocked := make(chan bool) cunlock := make(chan bool) cdone := make(chan bool) @@ -55,7 +55,7 @@ func TestParallelReaders(t *testing.T) { doTestParallelReaders(4, 2) } -func reader(rwm *DowngradableRWMutex, numIterations int, activity *int32, cdone chan bool) { +func reader(rwm *RWMutex, numIterations int, activity *int32, cdone chan bool) { for i := 0; i < numIterations; i++ { rwm.RLock() n := atomic.AddInt32(activity, 1) @@ -70,7 +70,7 @@ func reader(rwm *DowngradableRWMutex, numIterations int, activity *int32, cdone cdone <- true } -func writer(rwm *DowngradableRWMutex, numIterations int, activity *int32, cdone chan bool) { +func writer(rwm *RWMutex, numIterations int, activity *int32, cdone chan bool) { for i := 0; i < numIterations; i++ { rwm.Lock() n := atomic.AddInt32(activity, 10000) @@ -85,7 +85,7 @@ func writer(rwm *DowngradableRWMutex, numIterations int, activity *int32, cdone cdone <- true } -func downgradingWriter(rwm *DowngradableRWMutex, numIterations int, activity *int32, cdone chan bool) { +func downgradingWriter(rwm *RWMutex, numIterations int, activity *int32, cdone chan bool) { for i := 0; i < numIterations; i++ { rwm.Lock() n := atomic.AddInt32(activity, 10000) @@ -112,7 +112,7 @@ func HammerDowngradableRWMutex(gomaxprocs, numReaders, numIterations int) { runtime.GOMAXPROCS(gomaxprocs) // Number of active readers + 10000 * number of active writers. var activity int32 - var rwm DowngradableRWMutex + var rwm RWMutex cdone := make(chan bool) go writer(&rwm, numIterations, &activity, cdone) go downgradingWriter(&rwm, numIterations, &activity, cdone) @@ -150,56 +150,56 @@ func TestDowngradableRWMutex(t *testing.T) { } func TestRWDoubleTryLock(t *testing.T) { - var m DowngradableRWMutex - if !m.TryLock() { + var rwm RWMutex + if !rwm.TryLock() { t.Fatal("failed to aquire lock") } - if m.TryLock() { + if rwm.TryLock() { t.Fatal("unexpectedly succeeded in aquiring locked mutex") } } func TestRWTryLockAfterLock(t *testing.T) { - var m DowngradableRWMutex - m.Lock() - if m.TryLock() { + var rwm RWMutex + rwm.Lock() + if rwm.TryLock() { t.Fatal("unexpectedly succeeded in aquiring locked mutex") } } func TestRWTryLockUnlock(t *testing.T) { - var m DowngradableRWMutex - if !m.TryLock() { + var rwm RWMutex + if !rwm.TryLock() { t.Fatal("failed to aquire lock") } - m.Unlock() - if !m.TryLock() { + rwm.Unlock() + if !rwm.TryLock() { t.Fatal("failed to aquire lock after unlock") } } func TestTryRLockAfterLock(t *testing.T) { - var m DowngradableRWMutex - m.Lock() - if m.TryRLock() { + var rwm RWMutex + rwm.Lock() + if rwm.TryRLock() { t.Fatal("unexpectedly succeeded in aquiring locked mutex") } } func TestTryLockAfterRLock(t *testing.T) { - var m DowngradableRWMutex - m.RLock() - if m.TryLock() { + var rwm RWMutex + rwm.RLock() + if rwm.TryLock() { t.Fatal("unexpectedly succeeded in aquiring locked mutex") } } func TestDoubleTryRLock(t *testing.T) { - var m DowngradableRWMutex - if !m.TryRLock() { + var rwm RWMutex + if !rwm.TryRLock() { t.Fatal("failed to aquire lock") } - if !m.TryRLock() { + if !rwm.TryRLock() { t.Fatal("failed to read aquire read locked lock") } } diff --git a/pkg/sync/downgradable_rwmutex_unsafe.go b/pkg/sync/downgradable_rwmutex_unsafe.go index 0d321f5e3..ea6cdc447 100644 --- a/pkg/sync/downgradable_rwmutex_unsafe.go +++ b/pkg/sync/downgradable_rwmutex_unsafe.go @@ -29,10 +29,10 @@ func runtimeSemacquire(s *uint32) //go:linkname runtimeSemrelease sync.runtime_Semrelease func runtimeSemrelease(s *uint32, handoff bool, skipframes int) -// DowngradableRWMutex is identical to sync.RWMutex, but adds the DowngradeLock, +// RWMutex is identical to sync.RWMutex, but adds the DowngradeLock, // TryLock and TryRLock methods. -type DowngradableRWMutex struct { - w TMutex // held if there are pending writers +type RWMutex struct { + w Mutex // held if there are pending writers writerSem uint32 // semaphore for writers to wait for completing readers readerSem uint32 // semaphore for readers to wait for completing writers readerCount int32 // number of pending readers @@ -43,7 +43,7 @@ const rwmutexMaxReaders = 1 << 30 // TryRLock locks rw for reading. It returns true if it succeeds and false // otherwise. It does not block. -func (rw *DowngradableRWMutex) TryRLock() bool { +func (rw *RWMutex) TryRLock() bool { if RaceEnabled { RaceDisable() } @@ -67,7 +67,7 @@ func (rw *DowngradableRWMutex) TryRLock() bool { } // RLock locks rw for reading. -func (rw *DowngradableRWMutex) RLock() { +func (rw *RWMutex) RLock() { if RaceEnabled { RaceDisable() } @@ -82,14 +82,14 @@ func (rw *DowngradableRWMutex) RLock() { } // RUnlock undoes a single RLock call. -func (rw *DowngradableRWMutex) RUnlock() { +func (rw *RWMutex) RUnlock() { if RaceEnabled { RaceReleaseMerge(unsafe.Pointer(&rw.writerSem)) RaceDisable() } if r := atomic.AddInt32(&rw.readerCount, -1); r < 0 { if r+1 == 0 || r+1 == -rwmutexMaxReaders { - panic("RUnlock of unlocked DowngradableRWMutex") + panic("RUnlock of unlocked RWMutex") } // A writer is pending. if atomic.AddInt32(&rw.readerWait, -1) == 0 { @@ -104,7 +104,7 @@ func (rw *DowngradableRWMutex) RUnlock() { // TryLock locks rw for writing. It returns true if it succeeds and false // otherwise. It does not block. -func (rw *DowngradableRWMutex) TryLock() bool { +func (rw *RWMutex) TryLock() bool { if RaceEnabled { RaceDisable() } @@ -131,7 +131,7 @@ func (rw *DowngradableRWMutex) TryLock() bool { } // Lock locks rw for writing. -func (rw *DowngradableRWMutex) Lock() { +func (rw *RWMutex) Lock() { if RaceEnabled { RaceDisable() } @@ -150,7 +150,7 @@ func (rw *DowngradableRWMutex) Lock() { } // Unlock unlocks rw for writing. -func (rw *DowngradableRWMutex) Unlock() { +func (rw *RWMutex) Unlock() { if RaceEnabled { RaceRelease(unsafe.Pointer(&rw.writerSem)) RaceRelease(unsafe.Pointer(&rw.readerSem)) @@ -159,7 +159,7 @@ func (rw *DowngradableRWMutex) Unlock() { // Announce to readers there is no active writer. r := atomic.AddInt32(&rw.readerCount, rwmutexMaxReaders) if r >= rwmutexMaxReaders { - panic("Unlock of unlocked DowngradableRWMutex") + panic("Unlock of unlocked RWMutex") } // Unblock blocked readers, if any. for i := 0; i < int(r); i++ { @@ -173,7 +173,7 @@ func (rw *DowngradableRWMutex) Unlock() { } // DowngradeLock atomically unlocks rw for writing and locks it for reading. -func (rw *DowngradableRWMutex) DowngradeLock() { +func (rw *RWMutex) DowngradeLock() { if RaceEnabled { RaceRelease(unsafe.Pointer(&rw.readerSem)) RaceDisable() @@ -181,7 +181,7 @@ func (rw *DowngradableRWMutex) DowngradeLock() { // Announce to readers there is no active writer and one additional reader. r := atomic.AddInt32(&rw.readerCount, rwmutexMaxReaders+1) if r >= rwmutexMaxReaders+1 { - panic("DowngradeLock of unlocked DowngradableRWMutex") + panic("DowngradeLock of unlocked RWMutex") } // Unblock blocked readers, if any. Note that this loop starts as 1 since r // includes this goroutine. diff --git a/pkg/sync/tmutex_test.go b/pkg/sync/tmutex_test.go index c640bae23..0838248b4 100644 --- a/pkg/sync/tmutex_test.go +++ b/pkg/sync/tmutex_test.go @@ -30,7 +30,7 @@ func TestStructSize(t *testing.T) { // // The correctness of this package relies on these remaining in sync. func TestFieldValues(t *testing.T) { - var m TMutex + var m Mutex m.Lock() if got := *m.state(); got != mutexLocked { t.Errorf("got locked sync.Mutex.state = %d, want = %d", got, mutexLocked) @@ -42,7 +42,7 @@ func TestFieldValues(t *testing.T) { } func TestDoubleTryLock(t *testing.T) { - var m TMutex + var m Mutex if !m.TryLock() { t.Fatal("failed to aquire lock") } @@ -52,7 +52,7 @@ func TestDoubleTryLock(t *testing.T) { } func TestTryLockAfterLock(t *testing.T) { - var m TMutex + var m Mutex m.Lock() if m.TryLock() { t.Fatal("unexpectedly succeeded in aquiring locked mutex") @@ -60,7 +60,7 @@ func TestTryLockAfterLock(t *testing.T) { } func TestTryLockUnlock(t *testing.T) { - var m TMutex + var m Mutex if !m.TryLock() { t.Fatal("failed to aquire lock") } diff --git a/pkg/sync/tmutex_unsafe.go b/pkg/sync/tmutex_unsafe.go index 3c32f8371..3dd15578b 100644 --- a/pkg/sync/tmutex_unsafe.go +++ b/pkg/sync/tmutex_unsafe.go @@ -17,8 +17,8 @@ import ( "unsafe" ) -// TMutex is a try lock. -type TMutex struct { +// Mutex is a try lock. +type Mutex struct { sync.Mutex } @@ -27,7 +27,7 @@ type syncMutex struct { sema uint32 } -func (m *TMutex) state() *int32 { +func (m *Mutex) state() *int32 { return &(*syncMutex)(unsafe.Pointer(&m.Mutex)).state } @@ -38,7 +38,7 @@ const ( // TryLock tries to aquire the mutex. It returns true if it succeeds and false // otherwise. TryLock does not block. -func (m *TMutex) TryLock() bool { +func (m *Mutex) TryLock() bool { if atomic.CompareAndSwapInt32(m.state(), mutexUnlocked, mutexLocked) { if RaceEnabled { RaceAcquire(unsafe.Pointer(&m.Mutex)) -- cgit v1.2.3 From 5ab1213a6c405071546c783d6d93b4e9af52842e Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Wed, 22 Jan 2020 12:27:16 -0800 Subject: Move VFS2 handling of FD readability/writability to vfs.FileDescription. PiperOrigin-RevId: 291006713 --- pkg/sentry/fsimpl/ext/inode.go | 8 +++- pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go | 11 +++-- pkg/sentry/fsimpl/kernfs/fd_impl_util.go | 11 ++++- pkg/sentry/fsimpl/kernfs/kernfs_test.go | 18 +++++-- pkg/sentry/fsimpl/tmpfs/filesystem.go | 15 ++---- pkg/sentry/fsimpl/tmpfs/named_pipe.go | 5 +- pkg/sentry/fsimpl/tmpfs/regular_file.go | 14 +----- pkg/sentry/kernel/pipe/vfs.go | 12 ++--- pkg/sentry/vfs/file_description.go | 66 ++++++++++++++++++++++++-- pkg/sentry/vfs/permissions.go | 5 +- 10 files changed, 111 insertions(+), 54 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/ext/inode.go b/pkg/sentry/fsimpl/ext/inode.go index 8608805bf..191b39970 100644 --- a/pkg/sentry/fsimpl/ext/inode.go +++ b/pkg/sentry/fsimpl/ext/inode.go @@ -157,7 +157,9 @@ func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*v switch in.impl.(type) { case *regularFile: var fd regularFileFD - fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{}) + if err := fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{}); err != nil { + return nil, err + } return &fd.vfsfd, nil case *directory: // Can't open directories writably. This check is not necessary for a read @@ -166,7 +168,9 @@ func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*v return nil, syserror.EISDIR } var fd directoryFD - fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{}) + if err := fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{}); err != nil { + return nil, err + } return &fd.vfsfd, nil case *symlink: if flags&linux.O_PATH == 0 { diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go index 606ca692d..75624e0b1 100644 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -55,7 +55,9 @@ func (f *DynamicBytesFile) Init(creds *auth.Credentials, ino uint64, data vfs.Dy // Open implements Inode.Open. func (f *DynamicBytesFile) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { fd := &DynamicBytesFD{} - fd.Init(rp.Mount(), vfsd, f.data, flags) + if err := fd.Init(rp.Mount(), vfsd, f.data, flags); err != nil { + return nil, err + } return &fd.vfsfd, nil } @@ -80,10 +82,13 @@ type DynamicBytesFD struct { } // Init initializes a DynamicBytesFD. -func (fd *DynamicBytesFD) Init(m *vfs.Mount, d *vfs.Dentry, data vfs.DynamicBytesSource, flags uint32) { +func (fd *DynamicBytesFD) Init(m *vfs.Mount, d *vfs.Dentry, data vfs.DynamicBytesSource, flags uint32) error { + if err := fd.vfsfd.Init(fd, flags, m, d, &vfs.FileDescriptionOptions{}); err != nil { + return err + } fd.inode = d.Impl().(*Dentry).inode fd.SetDataSource(data) - fd.vfsfd.Init(fd, flags, m, d, &vfs.FileDescriptionOptions{}) + return nil } // Seek implements vfs.FileDescriptionImpl.Seek. diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index bcf069b5f..5fa1fa67b 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -43,9 +43,16 @@ type GenericDirectoryFD struct { } // Init initializes a GenericDirectoryFD. -func (fd *GenericDirectoryFD) Init(m *vfs.Mount, d *vfs.Dentry, children *OrderedChildren, flags uint32) { +func (fd *GenericDirectoryFD) Init(m *vfs.Mount, d *vfs.Dentry, children *OrderedChildren, flags uint32) error { + if vfs.AccessTypesForOpenFlags(flags)&vfs.MayWrite != 0 { + // Can't open directories for writing. + return syserror.EISDIR + } + if err := fd.vfsfd.Init(fd, flags, m, d, &vfs.FileDescriptionOptions{}); err != nil { + return err + } fd.children = children - fd.vfsfd.Init(fd, flags, m, d, &vfs.FileDescriptionOptions{}) + return nil } // VFSFileDescription returns a pointer to the vfs.FileDescription representing diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index a5fdfbde5..aa3fe76ee 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -115,7 +115,9 @@ func (fs *filesystem) newReadonlyDir(creds *auth.Credentials, mode linux.FileMod func (d *readonlyDir) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &d.OrderedChildren, flags) + if err := fd.Init(rp.Mount(), vfsd, &d.OrderedChildren, flags); err != nil { + return nil, err + } return fd.VFSFileDescription(), nil } @@ -225,7 +227,9 @@ func TestReadStaticFile(t *testing.T) { defer sys.Destroy() pop := sys.PathOpAtRoot("file1") - fd, err := sys.VFS.OpenAt(sys.Ctx, sys.Creds, &pop, &vfs.OpenOptions{}) + fd, err := sys.VFS.OpenAt(sys.Ctx, sys.Creds, &pop, &vfs.OpenOptions{ + Flags: linux.O_RDONLY, + }) if err != nil { t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) } @@ -258,7 +262,9 @@ func TestCreateNewFileInStaticDir(t *testing.T) { // Close the file. The file should persist. fd.DecRef() - fd, err = sys.VFS.OpenAt(sys.Ctx, sys.Creds, &pop, &vfs.OpenOptions{}) + fd, err = sys.VFS.OpenAt(sys.Ctx, sys.Creds, &pop, &vfs.OpenOptions{ + Flags: linux.O_RDONLY, + }) if err != nil { t.Fatalf("OpenAt(pop:%+v) = %+v failed: %v", pop, fd, err) } @@ -272,7 +278,9 @@ func TestDirFDReadWrite(t *testing.T) { defer sys.Destroy() pop := sys.PathOpAtRoot("/") - fd, err := sys.VFS.OpenAt(sys.Ctx, sys.Creds, &pop, &vfs.OpenOptions{}) + fd, err := sys.VFS.OpenAt(sys.Ctx, sys.Creds, &pop, &vfs.OpenOptions{ + Flags: linux.O_RDONLY, + }) if err != nil { t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) } @@ -282,7 +290,7 @@ func TestDirFDReadWrite(t *testing.T) { if _, err := fd.Read(sys.Ctx, usermem.BytesIOSequence([]byte{}), vfs.ReadOptions{}); err != syserror.EISDIR { t.Fatalf("Read for directory FD failed with unexpected error: %v", err) } - if _, err := fd.Write(sys.Ctx, usermem.BytesIOSequence([]byte{}), vfs.WriteOptions{}); err != syserror.EISDIR { + if _, err := fd.Write(sys.Ctx, usermem.BytesIOSequence([]byte{}), vfs.WriteOptions{}); err != syserror.EBADF { t.Fatalf("Write for directory FD failed with unexpected error: %v", err) } } diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 4cd7e9aea..a9f66a42a 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -337,19 +337,12 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, flags uint32, return nil, err } } - mnt := rp.Mount() switch impl := d.inode.impl.(type) { case *regularFile: var fd regularFileFD - fd.readable = vfs.MayReadFileWithOpenFlags(flags) - fd.writable = vfs.MayWriteFileWithOpenFlags(flags) - if fd.writable { - if err := mnt.CheckBeginWrite(); err != nil { - return nil, err - } - // mnt.EndWrite() is called by regularFileFD.Release(). + if err := fd.vfsfd.Init(&fd, flags, rp.Mount(), &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { + return nil, err } - fd.vfsfd.Init(&fd, flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}) if flags&linux.O_TRUNC != 0 { impl.mu.Lock() impl.data.Truncate(0, impl.memFile) @@ -363,7 +356,9 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, flags uint32, return nil, syserror.EISDIR } var fd directoryFD - fd.vfsfd.Init(&fd, flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}) + if err := fd.vfsfd.Init(&fd, flags, rp.Mount(), &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { + return nil, err + } return &fd.vfsfd, nil case *symlink: // Can't open symlinks without O_PATH (which is unimplemented). diff --git a/pkg/sentry/fsimpl/tmpfs/named_pipe.go b/pkg/sentry/fsimpl/tmpfs/named_pipe.go index 40bde54de..482aabd52 100644 --- a/pkg/sentry/fsimpl/tmpfs/named_pipe.go +++ b/pkg/sentry/fsimpl/tmpfs/named_pipe.go @@ -50,11 +50,10 @@ type namedPipeFD struct { func newNamedPipeFD(ctx context.Context, np *namedPipe, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { var err error var fd namedPipeFD - fd.VFSPipeFD, err = np.pipe.NewVFSPipeFD(ctx, rp, vfsd, &fd.vfsfd, flags) + fd.VFSPipeFD, err = np.pipe.NewVFSPipeFD(ctx, vfsd, &fd.vfsfd, flags) if err != nil { return nil, err } - mnt := rp.Mount() - fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{}) + fd.vfsfd.Init(&fd, flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}) return &fd.vfsfd, nil } diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index 5fa70cc6d..7c633c1b0 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -101,10 +101,6 @@ func (rf *regularFile) truncate(size uint64) (bool, error) { type regularFileFD struct { fileDescription - // These are immutable. - readable bool - writable bool - // off is the file offset. off is accessed using atomic memory operations. // offMu serializes operations that may mutate off. off int64 @@ -113,16 +109,11 @@ type regularFileFD struct { // Release implements vfs.FileDescriptionImpl.Release. func (fd *regularFileFD) Release() { - if fd.writable { - fd.vfsfd.VirtualDentry().Mount().EndWrite() - } + // noop } // PRead implements vfs.FileDescriptionImpl.PRead. func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - if !fd.readable { - return 0, syserror.EINVAL - } if offset < 0 { return 0, syserror.EINVAL } @@ -147,9 +138,6 @@ func (fd *regularFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts // PWrite implements vfs.FileDescriptionImpl.PWrite. func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - if !fd.writable { - return 0, syserror.EINVAL - } if offset < 0 { return 0, syserror.EINVAL } diff --git a/pkg/sentry/kernel/pipe/vfs.go b/pkg/sentry/kernel/pipe/vfs.go index bf7461cbb..6f83e3cee 100644 --- a/pkg/sentry/kernel/pipe/vfs.go +++ b/pkg/sentry/kernel/pipe/vfs.go @@ -66,7 +66,7 @@ func NewVFSPipe(sizeBytes, atomicIOBytes int64) *VFSPipe { // for read and write will succeed both in blocking and nonblocking mode. POSIX // leaves this behavior undefined. This can be used to open a FIFO for writing // while there are no readers available." - fifo(7) -func (vp *VFSPipe) NewVFSPipeFD(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, vfsfd *vfs.FileDescription, flags uint32) (*VFSPipeFD, error) { +func (vp *VFSPipe) NewVFSPipeFD(ctx context.Context, vfsd *vfs.Dentry, vfsfd *vfs.FileDescription, flags uint32) (*VFSPipeFD, error) { vp.mu.Lock() defer vp.mu.Unlock() @@ -76,7 +76,7 @@ func (vp *VFSPipe) NewVFSPipeFD(ctx context.Context, rp *vfs.ResolvingPath, vfsd return nil, syserror.EINVAL } - vfd, err := vp.open(rp, vfsd, vfsfd, flags) + vfd, err := vp.open(vfsd, vfsfd, flags) if err != nil { return nil, err } @@ -118,19 +118,13 @@ func (vp *VFSPipe) NewVFSPipeFD(ctx context.Context, rp *vfs.ResolvingPath, vfsd } // Preconditions: vp.mu must be held. -func (vp *VFSPipe) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, vfsfd *vfs.FileDescription, flags uint32) (*VFSPipeFD, error) { +func (vp *VFSPipe) open(vfsd *vfs.Dentry, vfsfd *vfs.FileDescription, flags uint32) (*VFSPipeFD, error) { var fd VFSPipeFD fd.flags = flags fd.readable = vfs.MayReadFileWithOpenFlags(flags) fd.writable = vfs.MayWriteFileWithOpenFlags(flags) fd.vfsfd = vfsfd fd.pipe = &vp.pipe - if fd.writable { - // The corresponding Mount.EndWrite() is in VFSPipe.Release(). - if err := rp.Mount().CheckBeginWrite(); err != nil { - return nil, err - } - } switch { case fd.readable && fd.writable: diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 6afe280bc..51c95c2d9 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -49,8 +49,23 @@ type FileDescription struct { // A reference is held on vd. vd is immutable. vd VirtualDentry + // opts contains options passed to FileDescription.Init(). opts is + // immutable. opts FileDescriptionOptions + // readable is MayReadFileWithOpenFlags(statusFlags). readable is + // immutable. + // + // readable is analogous to Linux's FMODE_READ. + readable bool + + // writable is MayWriteFileWithOpenFlags(statusFlags). If writable is true, + // the FileDescription holds a write count on vd.mount. writable is + // immutable. + // + // writable is analogous to Linux's FMODE_WRITE. + writable bool + // impl is the FileDescriptionImpl associated with this Filesystem. impl is // immutable. This should be the last field in FileDescription. impl FileDescriptionImpl @@ -77,10 +92,17 @@ type FileDescriptionOptions struct { UseDentryMetadata bool } -// Init must be called before first use of fd. It takes references on mnt and -// d. statusFlags is the initial file description status flags, which is -// usually the full set of flags passed to open(2). -func (fd *FileDescription) Init(impl FileDescriptionImpl, statusFlags uint32, mnt *Mount, d *Dentry, opts *FileDescriptionOptions) { +// Init must be called before first use of fd. If it succeeds, it takes +// references on mnt and d. statusFlags is the initial file description status +// flags, which is usually the full set of flags passed to open(2). +func (fd *FileDescription) Init(impl FileDescriptionImpl, statusFlags uint32, mnt *Mount, d *Dentry, opts *FileDescriptionOptions) error { + writable := MayWriteFileWithOpenFlags(statusFlags) + if writable { + if err := mnt.CheckBeginWrite(); err != nil { + return err + } + } + fd.refs = 1 fd.statusFlags = statusFlags | linux.O_LARGEFILE fd.vd = VirtualDentry{ @@ -89,7 +111,10 @@ func (fd *FileDescription) Init(impl FileDescriptionImpl, statusFlags uint32, mn } fd.vd.IncRef() fd.opts = *opts + fd.readable = MayReadFileWithOpenFlags(statusFlags) + fd.writable = writable fd.impl = impl + return nil } // IncRef increments fd's reference count. @@ -117,6 +142,9 @@ func (fd *FileDescription) TryIncRef() bool { func (fd *FileDescription) DecRef() { if refs := atomic.AddInt64(&fd.refs, -1); refs == 0 { fd.impl.Release() + if fd.writable { + fd.vd.mount.EndWrite() + } fd.vd.DecRef() } else if refs < 0 { panic("FileDescription.DecRef() called without holding a reference") @@ -194,6 +222,16 @@ func (fd *FileDescription) SetStatusFlags(ctx context.Context, creds *auth.Crede return nil } +// IsReadable returns true if fd was opened for reading. +func (fd *FileDescription) IsReadable() bool { + return fd.readable +} + +// IsWritable returns true if fd was opened for writing. +func (fd *FileDescription) IsWritable() bool { + return fd.writable +} + // Impl returns the FileDescriptionImpl associated with fd. func (fd *FileDescription) Impl() FileDescriptionImpl { return fd.impl @@ -241,6 +279,8 @@ type FileDescriptionImpl interface { // Errors: // // - If opts.Flags specifies unsupported options, PRead returns EOPNOTSUPP. + // + // Preconditions: The FileDescription was opened for reading. PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts ReadOptions) (int64, error) // Read is similar to PRead, but does not specify an offset. @@ -254,6 +294,8 @@ type FileDescriptionImpl interface { // Errors: // // - If opts.Flags specifies unsupported options, Read returns EOPNOTSUPP. + // + // Preconditions: The FileDescription was opened for reading. Read(ctx context.Context, dst usermem.IOSequence, opts ReadOptions) (int64, error) // PWrite writes src to the file, starting at the given offset, and returns @@ -268,6 +310,8 @@ type FileDescriptionImpl interface { // // - If opts.Flags specifies unsupported options, PWrite returns // EOPNOTSUPP. + // + // Preconditions: The FileDescription was opened for writing. PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts WriteOptions) (int64, error) // Write is similar to PWrite, but does not specify an offset, which is @@ -281,6 +325,8 @@ type FileDescriptionImpl interface { // Errors: // // - If opts.Flags specifies unsupported options, Write returns EOPNOTSUPP. + // + // Preconditions: The FileDescription was opened for writing. Write(ctx context.Context, src usermem.IOSequence, opts WriteOptions) (int64, error) // IterDirents invokes cb on each entry in the directory represented by the @@ -411,11 +457,17 @@ func (fd *FileDescription) StatFS(ctx context.Context) (linux.Statfs, error) { // offset, and returns the number of bytes read. PRead is permitted to return // partial reads with a nil error. func (fd *FileDescription) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts ReadOptions) (int64, error) { + if !fd.readable { + return 0, syserror.EBADF + } return fd.impl.PRead(ctx, dst, offset, opts) } // Read is similar to PRead, but does not specify an offset. func (fd *FileDescription) Read(ctx context.Context, dst usermem.IOSequence, opts ReadOptions) (int64, error) { + if !fd.readable { + return 0, syserror.EBADF + } return fd.impl.Read(ctx, dst, opts) } @@ -423,11 +475,17 @@ func (fd *FileDescription) Read(ctx context.Context, dst usermem.IOSequence, opt // offset, and returns the number of bytes written. PWrite is permitted to // return partial writes with a nil error. func (fd *FileDescription) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts WriteOptions) (int64, error) { + if !fd.writable { + return 0, syserror.EBADF + } return fd.impl.PWrite(ctx, src, offset, opts) } // Write is similar to PWrite, but does not specify an offset. func (fd *FileDescription) Write(ctx context.Context, src usermem.IOSequence, opts WriteOptions) (int64, error) { + if !fd.writable { + return 0, syserror.EBADF + } return fd.impl.Write(ctx, src, opts) } diff --git a/pkg/sentry/vfs/permissions.go b/pkg/sentry/vfs/permissions.go index d279d05ca..f664581f4 100644 --- a/pkg/sentry/vfs/permissions.go +++ b/pkg/sentry/vfs/permissions.go @@ -94,14 +94,13 @@ func GenericCheckPermissions(creds *auth.Credentials, ats AccessTypes, isDir boo // the set of accesses permitted for the opened file: // // - O_TRUNC causes MayWrite to be set in the returned AccessTypes (since it -// mutates the file), but does not permit the opened to write to the file +// mutates the file), but does not permit writing to the open file description // thereafter. // // - "Linux reserves the special, nonstandard access mode 3 (binary 11) in // flags to mean: check for read and write permission on the file and return a // file descriptor that can't be used for reading or writing." - open(2). Thus -// AccessTypesForOpenFlags returns MayRead|MayWrite in this case, but -// filesystems are responsible for ensuring that access is denied. +// AccessTypesForOpenFlags returns MayRead|MayWrite in this case. // // Use May{Read,Write}FileWithOpenFlags() for these checks instead. func AccessTypesForOpenFlags(flags uint32) AccessTypes { -- cgit v1.2.3 From b7853f688b4bcd3465c0c3087fcbd8d53bdf26ae Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Wed, 22 Jan 2020 14:46:15 -0800 Subject: Error marshalling the matcher. The iptables binary is looking for libxt_.so when it should be looking for libxt_udp.so, so it's having an issue reading the data in xt_match_entry. I think it may be an alignment issue. Trying to fix this is leading to me fighting with the metadata struct, so I'm gonna go kill that. --- pkg/abi/linux/netfilter.go | 5 +++++ pkg/sentry/socket/netfilter/netfilter.go | 35 ++++++++++++++++++++------------ pkg/tcpip/iptables/udp_matcher.go | 2 +- 3 files changed, 28 insertions(+), 14 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/netfilter.go b/pkg/abi/linux/netfilter.go index f0e544f9c..effed7976 100644 --- a/pkg/abi/linux/netfilter.go +++ b/pkg/abi/linux/netfilter.go @@ -198,6 +198,11 @@ type XTEntryMatch struct { // SizeOfXTEntryMatch is the size of an XTEntryMatch. const SizeOfXTEntryMatch = 32 +type KernelXTEntryMatch struct { + XTEntryMatch + Data []byte +} + // XTEntryTarget holds a target for a rule. For example, it can specify that // packets matching the rule should DROP, ACCEPT, or use an extension target. // iptables-extension(8) has a list of possible targets. diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 3caabca9a..b49fe5b3e 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -207,26 +207,34 @@ func marshalMatcher(matcher iptables.Matcher) []byte { } func marshalUDPMatcher(matcher *iptables.UDPMatcher) []byte { - type udpMatch struct { - linux.XTEntryMatch - linux.XTUDP - } - linuxMatcher := udpMatch{ + linuxMatcher := linux.KernelXTEntryMatch{ XTEntryMatch: linux.XTEntryMatch{ MatchSize: linux.SizeOfXTEntryMatch + linux.SizeOfXTUDP, // Name: "udp", }, - XTUDP: linux.XTUDP{ - SourcePortStart: matcher.Data.SourcePortStart, - SourcePortEnd: matcher.Data.SourcePortEnd, - DestinationPortStart: matcher.Data.DestinationPortStart, - DestinationPortEnd: matcher.Data.DestinationPortEnd, - InverseFlags: matcher.Data.InverseFlags, - }, + Data: make([]byte, linux.SizeOfXTUDP+22), } + // copy(linuxMatcher.Name[:], "udp") copy(linuxMatcher.Name[:], "udp") - var buf [linux.SizeOfXTEntryMatch + linux.SizeOfXTUDP]byte + // TODO: Must be aligned. + xtudp := linux.XTUDP{ + SourcePortStart: matcher.Data.SourcePortStart, + SourcePortEnd: matcher.Data.SourcePortEnd, + DestinationPortStart: matcher.Data.DestinationPortStart, + DestinationPortEnd: matcher.Data.DestinationPortEnd, + InverseFlags: matcher.Data.InverseFlags, + } + binary.Marshal(linuxMatcher.Data[:linux.SizeOfXTUDP], usermem.ByteOrder, xtudp) + + if binary.Size(linuxMatcher)%64 != 0 { + panic(fmt.Sprintf("size is actually: %d", binary.Size(linuxMatcher))) + } + + var buf [linux.SizeOfXTEntryMatch + linux.SizeOfXTUDP + 22]byte + if len(buf)%64 != 0 { + panic(fmt.Sprintf("len is actually: %d", len(buf))) + } binary.Marshal(buf[:], usermem.ByteOrder, linuxMatcher) return buf[:] } @@ -245,6 +253,7 @@ func marshalTarget(target iptables.Target) []byte { } func marshalStandardTarget(verdict iptables.Verdict) []byte { + // TODO: Must be aligned. // The target's name will be the empty string. target := linux.XTStandardTarget{ Target: linux.XTEntryTarget{ diff --git a/pkg/tcpip/iptables/udp_matcher.go b/pkg/tcpip/iptables/udp_matcher.go index fca457199..65ae7f9e0 100644 --- a/pkg/tcpip/iptables/udp_matcher.go +++ b/pkg/tcpip/iptables/udp_matcher.go @@ -59,7 +59,7 @@ func NewUDPMatcher(filter IPHeaderFilter, data UDPMatcherData) (Matcher, error) } if filter.Protocol != header.UDPProtocolNumber { - log.Warningf("UDP matching is only valid for protocol %d.", header.UDPProtocolNumber) + return nil, fmt.Errorf("UDP matching is only valid for protocol %d.", header.UDPProtocolNumber) } return &UDPMatcher{Data: data}, nil -- cgit v1.2.3 From 896bd654b6622d20cbaf8e82b4554a5375addf81 Mon Sep 17 00:00:00 2001 From: Rahat Mahmood Date: Wed, 22 Jan 2020 15:14:43 -0800 Subject: De-duplicate common test functionality for VFS2 filesystems. PiperOrigin-RevId: 291041576 --- pkg/sentry/fsimpl/kernfs/kernfs_test.go | 12 +- pkg/sentry/fsimpl/proc/tasks_test.go | 411 +++++++++++--------------------- pkg/sentry/fsimpl/sys/sys_test.go | 4 +- pkg/sentry/fsimpl/testutil/testutil.go | 149 ++++++++++-- 4 files changed, 278 insertions(+), 298 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index aa3fe76ee..fade59491 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -212,7 +212,7 @@ func TestMkdirGetDentry(t *testing.T) { defer sys.Destroy() pop := sys.PathOpAtRoot("dir1/a new directory") - if err := sys.VFS.MkdirAt(sys.Ctx, sys.Creds, &pop, &vfs.MkdirOptions{Mode: 0755}); err != nil { + if err := sys.VFS.MkdirAt(sys.Ctx, sys.Creds, pop, &vfs.MkdirOptions{Mode: 0755}); err != nil { t.Fatalf("MkdirAt for PathOperation %+v failed: %v", pop, err) } sys.GetDentryOrDie(pop).DecRef() @@ -227,7 +227,7 @@ func TestReadStaticFile(t *testing.T) { defer sys.Destroy() pop := sys.PathOpAtRoot("file1") - fd, err := sys.VFS.OpenAt(sys.Ctx, sys.Creds, &pop, &vfs.OpenOptions{ + fd, err := sys.VFS.OpenAt(sys.Ctx, sys.Creds, pop, &vfs.OpenOptions{ Flags: linux.O_RDONLY, }) if err != nil { @@ -254,7 +254,7 @@ func TestCreateNewFileInStaticDir(t *testing.T) { pop := sys.PathOpAtRoot("dir1/newfile") opts := &vfs.OpenOptions{Flags: linux.O_CREAT | linux.O_EXCL, Mode: defaultMode} - fd, err := sys.VFS.OpenAt(sys.Ctx, sys.Creds, &pop, opts) + fd, err := sys.VFS.OpenAt(sys.Ctx, sys.Creds, pop, opts) if err != nil { t.Fatalf("OpenAt(pop:%+v, opts:%+v) failed: %v", pop, opts, err) } @@ -262,7 +262,7 @@ func TestCreateNewFileInStaticDir(t *testing.T) { // Close the file. The file should persist. fd.DecRef() - fd, err = sys.VFS.OpenAt(sys.Ctx, sys.Creds, &pop, &vfs.OpenOptions{ + fd, err = sys.VFS.OpenAt(sys.Ctx, sys.Creds, pop, &vfs.OpenOptions{ Flags: linux.O_RDONLY, }) if err != nil { @@ -278,7 +278,7 @@ func TestDirFDReadWrite(t *testing.T) { defer sys.Destroy() pop := sys.PathOpAtRoot("/") - fd, err := sys.VFS.OpenAt(sys.Ctx, sys.Creds, &pop, &vfs.OpenOptions{ + fd, err := sys.VFS.OpenAt(sys.Ctx, sys.Creds, pop, &vfs.OpenOptions{ Flags: linux.O_RDONLY, }) if err != nil { @@ -309,7 +309,7 @@ func TestDirFDIterDirents(t *testing.T) { defer sys.Destroy() pop := sys.PathOpAtRoot("/") - sys.AssertDirectoryContains(&pop, map[string]testutil.DirentType{ + sys.AssertAllDirentTypes(sys.ListDirents(pop), map[string]testutil.DirentType{ "dir1": linux.DT_DIR, "dir2": linux.DT_DIR, "file1": linux.DT_REG, diff --git a/pkg/sentry/fsimpl/proc/tasks_test.go b/pkg/sentry/fsimpl/proc/tasks_test.go index 41977d816..2c1635f33 100644 --- a/pkg/sentry/fsimpl/proc/tasks_test.go +++ b/pkg/sentry/fsimpl/proc/tasks_test.go @@ -44,100 +44,47 @@ var ( proc3 = vfs.Dirent{Type: linux.DT_DIR, NextOff: 258 + 3 + 1} ) -type testIterDirentsCallback struct { - dirents []vfs.Dirent -} - -func (t *testIterDirentsCallback) Handle(d vfs.Dirent) bool { - t.dirents = append(t.dirents, d) - return true -} - -func checkDots(dirs []vfs.Dirent) ([]vfs.Dirent, error) { - if got := len(dirs); got < 2 { - return dirs, fmt.Errorf("wrong number of dirents, want at least: 2, got: %d: %v", got, dirs) - } - for i, want := range []string{".", ".."} { - if got := dirs[i].Name; got != want { - return dirs, fmt.Errorf("wrong name, want: %s, got: %s", want, got) - } - if got := dirs[i].Type; got != linux.DT_DIR { - return dirs, fmt.Errorf("wrong type, want: %d, got: %d", linux.DT_DIR, got) - } - } - return dirs[2:], nil -} - -func checkTasksStaticFiles(gots []vfs.Dirent) ([]vfs.Dirent, error) { - wants := map[string]vfs.Dirent{ - "cpuinfo": {Type: linux.DT_REG}, - "loadavg": {Type: linux.DT_REG}, - "meminfo": {Type: linux.DT_REG}, - "mounts": {Type: linux.DT_LNK}, - "net": {Type: linux.DT_DIR}, - "self": selfLink, - "stat": {Type: linux.DT_REG}, - "sys": {Type: linux.DT_DIR}, - "thread-self": threadSelfLink, - "uptime": {Type: linux.DT_REG}, - "version": {Type: linux.DT_REG}, - } - return checkFiles(gots, wants) -} - -func checkTaskStaticFiles(gots []vfs.Dirent) ([]vfs.Dirent, error) { - wants := map[string]vfs.Dirent{ - "auxv": {Type: linux.DT_REG}, - "cgroup": {Type: linux.DT_REG}, - "cmdline": {Type: linux.DT_REG}, - "comm": {Type: linux.DT_REG}, - "environ": {Type: linux.DT_REG}, - "gid_map": {Type: linux.DT_REG}, - "io": {Type: linux.DT_REG}, - "maps": {Type: linux.DT_REG}, - "ns": {Type: linux.DT_DIR}, - "smaps": {Type: linux.DT_REG}, - "stat": {Type: linux.DT_REG}, - "statm": {Type: linux.DT_REG}, - "status": {Type: linux.DT_REG}, - "task": {Type: linux.DT_DIR}, - "uid_map": {Type: linux.DT_REG}, - } - return checkFiles(gots, wants) -} - -func checkFiles(gots []vfs.Dirent, wants map[string]vfs.Dirent) ([]vfs.Dirent, error) { - // Go over all files, when there is a match, the file is removed from both - // 'gots' and 'wants'. wants is expected to reach 0, as all files must - // be present. Remaining files in 'gots', is returned to caller to decide - // whether this is valid or not. - for i := 0; i < len(gots); i++ { - got := gots[i] - want, ok := wants[got.Name] - if !ok { - continue - } - if want.Type != got.Type { - return gots, fmt.Errorf("wrong file type, want: %v, got: %v: %+v", want.Type, got.Type, got) - } - if want.NextOff != 0 && want.NextOff != got.NextOff { - return gots, fmt.Errorf("wrong dirent offset, want: %v, got: %v: %+v", want.NextOff, got.NextOff, got) - } - - delete(wants, got.Name) - gots = append(gots[0:i], gots[i+1:]...) - i-- - } - if len(wants) != 0 { - return gots, fmt.Errorf("not all files were found, missing: %+v", wants) +var ( + tasksStaticFiles = map[string]testutil.DirentType{ + "cpuinfo": linux.DT_REG, + "loadavg": linux.DT_REG, + "meminfo": linux.DT_REG, + "mounts": linux.DT_LNK, + "net": linux.DT_DIR, + "self": linux.DT_LNK, + "stat": linux.DT_REG, + "sys": linux.DT_DIR, + "thread-self": linux.DT_LNK, + "uptime": linux.DT_REG, + "version": linux.DT_REG, + } + tasksStaticFilesNextOffs = map[string]int64{ + "self": selfLink.NextOff, + "thread-self": threadSelfLink.NextOff, + } + taskStaticFiles = map[string]testutil.DirentType{ + "auxv": linux.DT_REG, + "cgroup": linux.DT_REG, + "cmdline": linux.DT_REG, + "comm": linux.DT_REG, + "environ": linux.DT_REG, + "gid_map": linux.DT_REG, + "io": linux.DT_REG, + "maps": linux.DT_REG, + "ns": linux.DT_DIR, + "smaps": linux.DT_REG, + "stat": linux.DT_REG, + "statm": linux.DT_REG, + "status": linux.DT_REG, + "task": linux.DT_DIR, + "uid_map": linux.DT_REG, } - return gots, nil -} +) -func setup() (context.Context, *vfs.VirtualFilesystem, vfs.VirtualDentry, error) { +func setup(t *testing.T) *testutil.System { k, err := testutil.Boot() if err != nil { - return nil, nil, vfs.VirtualDentry{}, fmt.Errorf("creating kernel: %v", err) + t.Fatalf("Error creating kernel: %v", err) } ctx := k.SupervisorContext() @@ -157,93 +104,60 @@ func setup() (context.Context, *vfs.VirtualFilesystem, vfs.VirtualDentry, error) } mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "procfs", &fsOpts) if err != nil { - return nil, nil, vfs.VirtualDentry{}, fmt.Errorf("NewMountNamespace(): %v", err) + t.Fatalf("NewMountNamespace(): %v", err) } - return ctx, vfsObj, mntns.Root(), nil + return testutil.NewSystem(ctx, t, vfsObj, mntns) } func TestTasksEmpty(t *testing.T) { - ctx, vfsObj, root, err := setup() - if err != nil { - t.Fatalf("Setup failed: %v", err) - } - defer root.DecRef() - - fd, err := vfsObj.OpenAt( - ctx, - auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: root, Start: root, Path: fspath.Parse("/")}, - &vfs.OpenOptions{}, - ) - if err != nil { - t.Fatalf("vfsfs.OpenAt failed: %v", err) - } + s := setup(t) + defer s.Destroy() - cb := testIterDirentsCallback{} - if err := fd.Impl().IterDirents(ctx, &cb); err != nil { - t.Fatalf("IterDirents(): %v", err) - } - cb.dirents, err = checkDots(cb.dirents) - if err != nil { - t.Error(err.Error()) - } - cb.dirents, err = checkTasksStaticFiles(cb.dirents) - if err != nil { - t.Error(err.Error()) - } - if len(cb.dirents) != 0 { - t.Errorf("found more files than expected: %+v", cb.dirents) - } + collector := s.ListDirents(s.PathOpAtRoot("/")) + s.AssertAllDirentTypes(collector, tasksStaticFiles) + s.AssertDirentOffsets(collector, tasksStaticFilesNextOffs) } func TestTasks(t *testing.T) { - ctx, vfsObj, root, err := setup() - if err != nil { - t.Fatalf("Setup failed: %v", err) + s := setup(t) + defer s.Destroy() + + expectedDirents := make(map[string]testutil.DirentType) + for n, d := range tasksStaticFiles { + expectedDirents[n] = d } - defer root.DecRef() - k := kernel.KernelFromContext(ctx) + k := kernel.KernelFromContext(s.Ctx) var tasks []*kernel.Task for i := 0; i < 5; i++ { tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - task, err := testutil.CreateTask(ctx, fmt.Sprintf("name-%d", i), tc) + task, err := testutil.CreateTask(s.Ctx, fmt.Sprintf("name-%d", i), tc) if err != nil { t.Fatalf("CreateTask(): %v", err) } tasks = append(tasks, task) + expectedDirents[fmt.Sprintf("%d", i+1)] = linux.DT_DIR } - fd, err := vfsObj.OpenAt( - ctx, - auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: root, Start: root, Path: fspath.Parse("/")}, - &vfs.OpenOptions{}, - ) - if err != nil { - t.Fatalf("vfsfs.OpenAt(/) failed: %v", err) - } + collector := s.ListDirents(s.PathOpAtRoot("/")) + s.AssertAllDirentTypes(collector, expectedDirents) + s.AssertDirentOffsets(collector, tasksStaticFilesNextOffs) - cb := testIterDirentsCallback{} - if err := fd.Impl().IterDirents(ctx, &cb); err != nil { - t.Fatalf("IterDirents(): %v", err) - } - cb.dirents, err = checkDots(cb.dirents) - if err != nil { - t.Error(err.Error()) - } - cb.dirents, err = checkTasksStaticFiles(cb.dirents) - if err != nil { - t.Error(err.Error()) - } lastPid := 0 - for _, d := range cb.dirents { + dirents := collector.OrderedDirents() + doneSkippingNonTaskDirs := false + for _, d := range dirents { pid, err := strconv.Atoi(d.Name) if err != nil { + if !doneSkippingNonTaskDirs { + // We haven't gotten to the task dirs yet. + continue + } t.Fatalf("Invalid process directory %q", d.Name) } + doneSkippingNonTaskDirs = true if lastPid > pid { - t.Errorf("pids not in order: %v", cb.dirents) + t.Errorf("pids not in order: %v", dirents) } found := false for _, t := range tasks { @@ -260,13 +174,16 @@ func TestTasks(t *testing.T) { t.Errorf("Wrong dirent offset want: %d got: %d: %+v", want, d.NextOff, d) } } + if !doneSkippingNonTaskDirs { + t.Fatalf("Never found any process directories.") + } // Test lookup. for _, path := range []string{"/1", "/2"} { - fd, err := vfsObj.OpenAt( - ctx, - auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: root, Start: root, Path: fspath.Parse(path)}, + fd, err := s.VFS.OpenAt( + s.Ctx, + s.Creds, + s.PathOpAtRoot(path), &vfs.OpenOptions{}, ) if err != nil { @@ -274,15 +191,15 @@ func TestTasks(t *testing.T) { } buf := make([]byte, 1) bufIOSeq := usermem.BytesIOSequence(buf) - if _, err := fd.Read(ctx, bufIOSeq, vfs.ReadOptions{}); err != syserror.EISDIR { + if _, err := fd.Read(s.Ctx, bufIOSeq, vfs.ReadOptions{}); err != syserror.EISDIR { t.Errorf("wrong error reading directory: %v", err) } } - if _, err := vfsObj.OpenAt( - ctx, - auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: root, Start: root, Path: fspath.Parse("/9999")}, + if _, err := s.VFS.OpenAt( + s.Ctx, + s.Creds, + s.PathOpAtRoot("/9999"), &vfs.OpenOptions{}, ); err != syserror.ENOENT { t.Fatalf("wrong error from vfsfs.OpenAt(/9999): %v", err) @@ -290,16 +207,13 @@ func TestTasks(t *testing.T) { } func TestTasksOffset(t *testing.T) { - ctx, vfsObj, root, err := setup() - if err != nil { - t.Fatalf("Setup failed: %v", err) - } - defer root.DecRef() + s := setup(t) + defer s.Destroy() - k := kernel.KernelFromContext(ctx) + k := kernel.KernelFromContext(s.Ctx) for i := 0; i < 3; i++ { tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - if _, err := testutil.CreateTask(ctx, fmt.Sprintf("name-%d", i), tc); err != nil { + if _, err := testutil.CreateTask(s.Ctx, fmt.Sprintf("name-%d", i), tc); err != nil { t.Fatalf("CreateTask(): %v", err) } } @@ -382,134 +296,100 @@ func TestTasksOffset(t *testing.T) { }, } { t.Run(tc.name, func(t *testing.T) { - fd, err := vfsObj.OpenAt( - ctx, - auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: root, Start: root, Path: fspath.Parse("/")}, + s := s.WithSubtest(t) + fd, err := s.VFS.OpenAt( + s.Ctx, + s.Creds, + s.PathOpAtRoot("/"), &vfs.OpenOptions{}, ) if err != nil { t.Fatalf("vfsfs.OpenAt(/) failed: %v", err) } - if _, err := fd.Impl().Seek(ctx, tc.offset, linux.SEEK_SET); err != nil { + if _, err := fd.Seek(s.Ctx, tc.offset, linux.SEEK_SET); err != nil { t.Fatalf("Seek(%d, SEEK_SET): %v", tc.offset, err) } - cb := testIterDirentsCallback{} - if err := fd.Impl().IterDirents(ctx, &cb); err != nil { - t.Fatalf("IterDirents(): %v", err) + var collector testutil.DirentCollector + if err := fd.IterDirents(s.Ctx, &collector); err != nil { + t.Fatalf("IterDirent(): %v", err) } - if cb.dirents, err = checkFiles(cb.dirents, tc.wants); err != nil { - t.Error(err.Error()) - } - if len(cb.dirents) != 0 { - t.Errorf("found more files than expected: %+v", cb.dirents) + + expectedTypes := make(map[string]testutil.DirentType) + expectedOffsets := make(map[string]int64) + for name, want := range tc.wants { + expectedTypes[name] = want.Type + if want.NextOff != 0 { + expectedOffsets[name] = want.NextOff + } } + + collector.SkipDotsChecks(true) // We seek()ed past the dots. + s.AssertAllDirentTypes(&collector, expectedTypes) + s.AssertDirentOffsets(&collector, expectedOffsets) }) } } func TestTask(t *testing.T) { - ctx, vfsObj, root, err := setup() - if err != nil { - t.Fatalf("Setup failed: %v", err) - } - defer root.DecRef() + s := setup(t) + defer s.Destroy() - k := kernel.KernelFromContext(ctx) + k := kernel.KernelFromContext(s.Ctx) tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - _, err = testutil.CreateTask(ctx, "name", tc) + _, err := testutil.CreateTask(s.Ctx, "name", tc) if err != nil { t.Fatalf("CreateTask(): %v", err) } - fd, err := vfsObj.OpenAt( - ctx, - auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: root, Start: root, Path: fspath.Parse("/1")}, - &vfs.OpenOptions{}, - ) - if err != nil { - t.Fatalf("vfsfs.OpenAt(/1) failed: %v", err) - } - - cb := testIterDirentsCallback{} - if err := fd.Impl().IterDirents(ctx, &cb); err != nil { - t.Fatalf("IterDirents(): %v", err) - } - cb.dirents, err = checkDots(cb.dirents) - if err != nil { - t.Error(err.Error()) - } - cb.dirents, err = checkTaskStaticFiles(cb.dirents) - if err != nil { - t.Error(err.Error()) - } - if len(cb.dirents) != 0 { - t.Errorf("found more files than expected: %+v", cb.dirents) - } + collector := s.ListDirents(s.PathOpAtRoot("/1")) + s.AssertAllDirentTypes(collector, taskStaticFiles) } func TestProcSelf(t *testing.T) { - ctx, vfsObj, root, err := setup() - if err != nil { - t.Fatalf("Setup failed: %v", err) - } - defer root.DecRef() + s := setup(t) + defer s.Destroy() - k := kernel.KernelFromContext(ctx) + k := kernel.KernelFromContext(s.Ctx) tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - task, err := testutil.CreateTask(ctx, "name", tc) + task, err := testutil.CreateTask(s.Ctx, "name", tc) if err != nil { t.Fatalf("CreateTask(): %v", err) } - fd, err := vfsObj.OpenAt( - task, - auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: root, Start: root, Path: fspath.Parse("/self/"), FollowFinalSymlink: true}, - &vfs.OpenOptions{}, - ) - if err != nil { - t.Fatalf("vfsfs.OpenAt(/self/) failed: %v", err) - } - - cb := testIterDirentsCallback{} - if err := fd.Impl().IterDirents(ctx, &cb); err != nil { - t.Fatalf("IterDirents(): %v", err) - } - cb.dirents, err = checkDots(cb.dirents) - if err != nil { - t.Error(err.Error()) - } - cb.dirents, err = checkTaskStaticFiles(cb.dirents) - if err != nil { - t.Error(err.Error()) - } - if len(cb.dirents) != 0 { - t.Errorf("found more files than expected: %+v", cb.dirents) - } + collector := s.WithTemporaryContext(task).ListDirents(&vfs.PathOperation{ + Root: s.Root, + Start: s.Root, + Path: fspath.Parse("/self/"), + FollowFinalSymlink: true, + }) + s.AssertAllDirentTypes(collector, taskStaticFiles) } -func iterateDir(ctx context.Context, t *testing.T, vfsObj *vfs.VirtualFilesystem, root vfs.VirtualDentry, fd *vfs.FileDescription) { +func iterateDir(ctx context.Context, t *testing.T, s *testutil.System, fd *vfs.FileDescription) { t.Logf("Iterating: /proc%s", fd.MappedName(ctx)) - cb := testIterDirentsCallback{} - if err := fd.Impl().IterDirents(ctx, &cb); err != nil { + var collector testutil.DirentCollector + if err := fd.IterDirents(ctx, &collector); err != nil { t.Fatalf("IterDirents(): %v", err) } - var err error - cb.dirents, err = checkDots(cb.dirents) - if err != nil { + if err := collector.Contains(".", linux.DT_DIR); err != nil { t.Error(err.Error()) } - for _, d := range cb.dirents { + if err := collector.Contains("..", linux.DT_DIR); err != nil { + t.Error(err.Error()) + } + + for _, d := range collector.Dirents() { + if d.Name == "." || d.Name == ".." { + continue + } childPath := path.Join(fd.MappedName(ctx), d.Name) if d.Type == linux.DT_LNK { - link, err := vfsObj.ReadlinkAt( + link, err := s.VFS.ReadlinkAt( ctx, auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: root, Start: root, Path: fspath.Parse(childPath)}, + &vfs.PathOperation{Root: s.Root, Start: s.Root, Path: fspath.Parse(childPath)}, ) if err != nil { t.Errorf("vfsfs.ReadlinkAt(%v) failed: %v", childPath, err) @@ -520,10 +400,10 @@ func iterateDir(ctx context.Context, t *testing.T, vfsObj *vfs.VirtualFilesystem } t.Logf("Opening: /proc%s", childPath) - child, err := vfsObj.OpenAt( + child, err := s.VFS.OpenAt( ctx, auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: root, Start: root, Path: fspath.Parse(childPath)}, + &vfs.PathOperation{Root: s.Root, Start: s.Root, Path: fspath.Parse(childPath)}, &vfs.OpenOptions{}, ) if err != nil { @@ -539,24 +419,21 @@ func iterateDir(ctx context.Context, t *testing.T, vfsObj *vfs.VirtualFilesystem } if d.Type == linux.DT_DIR { // Found another dir, let's do it again! - iterateDir(ctx, t, vfsObj, root, child) + iterateDir(ctx, t, s, child) } } } // TestTree iterates all directories and stats every file. func TestTree(t *testing.T) { - uberCtx, vfsObj, root, err := setup() - if err != nil { - t.Fatalf("Setup failed: %v", err) - } - defer root.DecRef() + s := setup(t) + defer s.Destroy() - k := kernel.KernelFromContext(uberCtx) + k := kernel.KernelFromContext(s.Ctx) var tasks []*kernel.Task for i := 0; i < 5; i++ { tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - task, err := testutil.CreateTask(uberCtx, fmt.Sprintf("name-%d", i), tc) + task, err := testutil.CreateTask(s.Ctx, fmt.Sprintf("name-%d", i), tc) if err != nil { t.Fatalf("CreateTask(): %v", err) } @@ -564,14 +441,14 @@ func TestTree(t *testing.T) { } ctx := tasks[0] - fd, err := vfsObj.OpenAt( + fd, err := s.VFS.OpenAt( ctx, - auth.CredentialsFromContext(uberCtx), - &vfs.PathOperation{Root: root, Start: root, Path: fspath.Parse("/")}, + auth.CredentialsFromContext(s.Ctx), + &vfs.PathOperation{Root: s.Root, Start: s.Root, Path: fspath.Parse("/")}, &vfs.OpenOptions{}, ) if err != nil { t.Fatalf("vfsfs.OpenAt(/) failed: %v", err) } - iterateDir(ctx, t, vfsObj, root, fd) + iterateDir(ctx, t, s, fd) } diff --git a/pkg/sentry/fsimpl/sys/sys_test.go b/pkg/sentry/fsimpl/sys/sys_test.go index 60a1634a9..8b1cf0bd0 100644 --- a/pkg/sentry/fsimpl/sys/sys_test.go +++ b/pkg/sentry/fsimpl/sys/sys_test.go @@ -56,7 +56,7 @@ func TestReadCPUFile(t *testing.T) { for _, fname := range []string{"online", "possible", "present"} { pop := s.PathOpAtRoot(fmt.Sprintf("devices/system/cpu/%s", fname)) - fd, err := s.VFS.OpenAt(s.Ctx, s.Creds, &pop, &vfs.OpenOptions{}) + fd, err := s.VFS.OpenAt(s.Ctx, s.Creds, pop, &vfs.OpenOptions{}) if err != nil { t.Fatalf("OpenAt(pop:%+v) = %+v failed: %v", pop, fd, err) } @@ -75,7 +75,7 @@ func TestSysRootContainsExpectedEntries(t *testing.T) { s := newTestSystem(t) defer s.Destroy() pop := s.PathOpAtRoot("/") - s.AssertDirectoryContains(&pop, map[string]testutil.DirentType{ + s.AssertAllDirentTypes(s.ListDirents(pop), map[string]testutil.DirentType{ "block": linux.DT_DIR, "bus": linux.DT_DIR, "class": linux.DT_DIR, diff --git a/pkg/sentry/fsimpl/testutil/testutil.go b/pkg/sentry/fsimpl/testutil/testutil.go index eada31d94..2a723a89f 100644 --- a/pkg/sentry/fsimpl/testutil/testutil.go +++ b/pkg/sentry/fsimpl/testutil/testutil.go @@ -40,8 +40,8 @@ type System struct { Ctx context.Context Creds *auth.Credentials VFS *vfs.VirtualFilesystem + Root vfs.VirtualDentry mns *vfs.MountNamespace - root vfs.VirtualDentry } // NewSystem constructs a System. @@ -55,14 +55,49 @@ func NewSystem(ctx context.Context, t *testing.T, v *vfs.VirtualFilesystem, mns Creds: auth.CredentialsFromContext(ctx), VFS: v, mns: mns, - root: mns.Root(), + Root: mns.Root(), } return s } +// WithSubtest creates a temporary test system with a new test harness, +// referencing all other resources from the original system. This is useful when +// a system is reused for multiple subtests, and the T needs to change for each +// case. Note that this is safe when test cases run in parallel, as all +// resources referenced by the system are immutable, or handle interior +// mutations in a thread-safe manner. +// +// The returned system must not outlive the original and should not be destroyed +// via System.Destroy. +func (s *System) WithSubtest(t *testing.T) *System { + return &System{ + t: t, + Ctx: s.Ctx, + Creds: s.Creds, + VFS: s.VFS, + mns: s.mns, + Root: s.Root, + } +} + +// WithTemporaryContext constructs a temporary test system with a new context +// ctx. The temporary system borrows all resources and references from the +// original system. The returned temporary system must not outlive the original +// system, and should not be destroyed via System.Destroy. +func (s *System) WithTemporaryContext(ctx context.Context) *System { + return &System{ + t: s.t, + Ctx: ctx, + Creds: s.Creds, + VFS: s.VFS, + mns: s.mns, + Root: s.Root, + } +} + // Destroy release resources associated with a test system. func (s *System) Destroy() { - s.root.DecRef() + s.Root.DecRef() s.mns.DecRef(s.VFS) // Reference on mns passed to NewSystem. } @@ -87,18 +122,18 @@ func (s *System) ReadToEnd(fd *vfs.FileDescription) (string, error) { // PathOpAtRoot constructs a PathOperation with the given path from // the root of the filesystem. -func (s *System) PathOpAtRoot(path string) vfs.PathOperation { - return vfs.PathOperation{ - Root: s.root, - Start: s.root, +func (s *System) PathOpAtRoot(path string) *vfs.PathOperation { + return &vfs.PathOperation{ + Root: s.Root, + Start: s.Root, Path: fspath.Parse(path), } } // GetDentryOrDie attempts to resolve a dentry referred to by the // provided path operation. If unsuccessful, the test fails. -func (s *System) GetDentryOrDie(pop vfs.PathOperation) vfs.VirtualDentry { - vd, err := s.VFS.GetDentryAt(s.Ctx, s.Creds, &pop, &vfs.GetDentryOptions{}) +func (s *System) GetDentryOrDie(pop *vfs.PathOperation) vfs.VirtualDentry { + vd, err := s.VFS.GetDentryAt(s.Ctx, s.Creds, pop, &vfs.GetDentryOptions{}) if err != nil { s.t.Fatalf("GetDentryAt(pop:%+v) failed: %v", pop, err) } @@ -108,14 +143,8 @@ func (s *System) GetDentryOrDie(pop vfs.PathOperation) vfs.VirtualDentry { // DirentType is an alias for values for linux_dirent64.d_type. type DirentType = uint8 -// AssertDirectoryContains verifies that a directory at pop contains the entries -// specified. AssertDirectoryContains implicitly checks for "." and "..", these -// need not be included in entries. -func (s *System) AssertDirectoryContains(pop *vfs.PathOperation, entries map[string]DirentType) { - // Also implicitly check for "." and "..". - entries["."] = linux.DT_DIR - entries[".."] = linux.DT_DIR - +// ListDirents lists the Dirents for a directory at pop. +func (s *System) ListDirents(pop *vfs.PathOperation) *DirentCollector { fd, err := s.VFS.OpenAt(s.Ctx, s.Creds, pop, &vfs.OpenOptions{Flags: linux.O_RDONLY}) if err != nil { s.t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) @@ -126,12 +155,52 @@ func (s *System) AssertDirectoryContains(pop *vfs.PathOperation, entries map[str if err := fd.IterDirents(s.Ctx, collector); err != nil { s.t.Fatalf("IterDirent failed: %v", err) } + return collector +} + +// AssertAllDirentTypes verifies that the set of dirents in collector contains +// exactly the specified set of expected entries. AssertAllDirentTypes respects +// collector.skipDots, and implicitly checks for "." and ".." accordingly. +func (s *System) AssertAllDirentTypes(collector *DirentCollector, expected map[string]DirentType) { + // Also implicitly check for "." and "..", if enabled. + if !collector.skipDots { + expected["."] = linux.DT_DIR + expected[".."] = linux.DT_DIR + } - collectedEntries := make(map[string]DirentType) + dentryTypes := make(map[string]DirentType) + collector.mu.Lock() for _, dirent := range collector.dirents { - collectedEntries[dirent.Name] = dirent.Type + dentryTypes[dirent.Name] = dirent.Type } - if diff := cmp.Diff(entries, collectedEntries); diff != "" { + collector.mu.Unlock() + if diff := cmp.Diff(expected, dentryTypes); diff != "" { + s.t.Fatalf("IterDirent had unexpected results:\n--- want\n+++ got\n%v", diff) + } +} + +// AssertDirentOffsets verifies that collector contains at least the entries +// specified in expected, with the given NextOff field. Entries specified in +// expected but missing from collector result in failure. Extra entries in +// collector are ignored. AssertDirentOffsets respects collector.skipDots, and +// implicitly checks for "." and ".." accordingly. +func (s *System) AssertDirentOffsets(collector *DirentCollector, expected map[string]int64) { + // Also implicitly check for "." and "..", if enabled. + if !collector.skipDots { + expected["."] = 1 + expected[".."] = 2 + } + + dentryNextOffs := make(map[string]int64) + collector.mu.Lock() + for _, dirent := range collector.dirents { + // Ignore extra entries in dentries that are not in expected. + if _, ok := expected[dirent.Name]; ok { + dentryNextOffs[dirent.Name] = dirent.NextOff + } + } + collector.mu.Unlock() + if diff := cmp.Diff(expected, dentryNextOffs); diff != "" { s.t.Fatalf("IterDirent had unexpected results:\n--- want\n+++ got\n%v", diff) } } @@ -141,16 +210,29 @@ func (s *System) AssertDirectoryContains(pop *vfs.PathOperation, entries map[str // all dirents emitted by the callback. type DirentCollector struct { mu sync.Mutex - dirents map[string]vfs.Dirent + order []*vfs.Dirent + dirents map[string]*vfs.Dirent + // When the collector is used in various Assert* functions, should "." and + // ".." be implicitly checked? + skipDots bool +} + +// SkipDotsChecks enables or disables the implicit checks on "." and ".." when +// the collector is used in various Assert* functions. Note that "." and ".." +// are still collected if passed to d.Handle, so the caller should only disable +// the checks when they aren't expected. +func (d *DirentCollector) SkipDotsChecks(value bool) { + d.skipDots = value } // Handle implements vfs.IterDirentsCallback.Handle. func (d *DirentCollector) Handle(dirent vfs.Dirent) bool { d.mu.Lock() if d.dirents == nil { - d.dirents = make(map[string]vfs.Dirent) + d.dirents = make(map[string]*vfs.Dirent) } - d.dirents[dirent.Name] = dirent + d.order = append(d.order, &dirent) + d.dirents[dirent.Name] = &dirent d.mu.Unlock() return true } @@ -176,3 +258,24 @@ func (d *DirentCollector) Contains(name string, typ uint8) error { } return nil } + +// Dirents returns all dirents discovered by this collector. +func (d *DirentCollector) Dirents() map[string]*vfs.Dirent { + d.mu.Lock() + dirents := make(map[string]*vfs.Dirent) + for n, d := range d.dirents { + dirents[n] = d + } + d.mu.Unlock() + return dirents +} + +// OrderedDirents returns an ordered list of dirents as discovered by this +// collector. +func (d *DirentCollector) OrderedDirents() []*vfs.Dirent { + d.mu.Lock() + dirents := make([]*vfs.Dirent, len(d.order)) + copy(dirents, d.order) + d.mu.Unlock() + return dirents +} -- cgit v1.2.3 From 7a79715504e92be9fc9aebc12fbd65aa46049054 Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Thu, 23 Jan 2020 11:47:30 -0800 Subject: Check for EINTR from KVM_CREATE_VM The kernel may return EINTR from: kvm_create_vm kvm_init_mmu_notifier mmu_notifier_register do_mmu_notifier_register mm_take_all_locks Go 1.14's preemptive scheduling signals make hitting this much more likely. PiperOrigin-RevId: 291212669 --- pkg/sentry/platform/kvm/kvm.go | 16 +++++++++++++--- 1 file changed, 13 insertions(+), 3 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/kvm/kvm.go b/pkg/sentry/platform/kvm/kvm.go index a7850faed..d337c5c7c 100644 --- a/pkg/sentry/platform/kvm/kvm.go +++ b/pkg/sentry/platform/kvm/kvm.go @@ -62,9 +62,19 @@ func New(deviceFile *os.File) (*KVM, error) { } // Create a new VM fd. - vm, _, errno := syscall.RawSyscall(syscall.SYS_IOCTL, fd, _KVM_CREATE_VM, 0) - if errno != 0 { - return nil, fmt.Errorf("creating VM: %v", errno) + var ( + vm uintptr + errno syscall.Errno + ) + for { + vm, _, errno = syscall.Syscall(syscall.SYS_IOCTL, fd, _KVM_CREATE_VM, 0) + if errno == syscall.EINTR { + continue + } + if errno != 0 { + return nil, fmt.Errorf("creating VM: %v", errno) + } + break } // We are done with the device file. deviceFile.Close() -- cgit v1.2.3 From 3db317390b5cc491d680fc4a5fc7b8372890b4da Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Thu, 23 Jan 2020 16:17:50 -0800 Subject: Remove epoll entry from map when dropping it. This pattern (delete from map when dropping) is also used in epoll.RemoveEntry, and seems like generally a good idea. PiperOrigin-RevId: 291268208 --- pkg/sentry/kernel/epoll/epoll.go | 1 + 1 file changed, 1 insertion(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/epoll/epoll.go b/pkg/sentry/kernel/epoll/epoll.go index 430311cc0..e84742993 100644 --- a/pkg/sentry/kernel/epoll/epoll.go +++ b/pkg/sentry/kernel/epoll/epoll.go @@ -174,6 +174,7 @@ func (e *EventPoll) Release() { entry.id.File.EventUnregister(&entry.waiter) entry.file.Drop() } + e.files = nil } // Read implements fs.FileOperations.Read. -- cgit v1.2.3 From 390bb9c241c2b05c311579562d95cc39d899157b Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Fri, 24 Jan 2020 11:58:13 -0800 Subject: Ignore external SIGURG Go 1.14+ sends SIGURG to Ms to attempt asynchronous preemption of a G. Since it can't guarantee that a SIGURG is only related to preemption, it continues to forward them to signal.Notify (see runtime.sighandler). We should ignore these signals, as applications shouldn't receive them. Note that this means that truly external SIGURG can no longer be sent to the application (as with SIGCHLD). PiperOrigin-RevId: 291415357 --- pkg/sentry/kernel/signal.go | 3 +++ 1 file changed, 3 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/signal.go b/pkg/sentry/kernel/signal.go index 02eede93d..e8cce37d0 100644 --- a/pkg/sentry/kernel/signal.go +++ b/pkg/sentry/kernel/signal.go @@ -38,6 +38,9 @@ const SignalPanic = linux.SIGUSR2 // Preconditions: Kernel must have an init process. func (k *Kernel) sendExternalSignal(info *arch.SignalInfo, context string) { switch linux.Signal(info.Signo) { + case linux.SIGURG: + // Sent by the Go 1.14+ runtime for asynchronous goroutine preemption. + case platform.SignalInterrupt: // Assume that a call to platform.Context.Interrupt() misfired. -- cgit v1.2.3 From d135b5abf6eafa92d2745dc98d48ef39d2f90e75 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Fri, 24 Jan 2020 12:53:29 -0800 Subject: Add anonymous device number allocation to VFS2. Note that in VFS2, filesystem device numbers are per-vfs.FilesystemImpl rather than global, avoiding the need for a "registry" type to handle save/restore. (This is more consistent with Linux anyway: compare e.g. mm/shmem.c:shmem_mount() => fs/super.c:mount_nodev() => (indirectly) set_anon_super().) PiperOrigin-RevId: 291425193 --- pkg/sentry/vfs/device.go | 29 +++++++++++++++++++++++++++++ pkg/sentry/vfs/vfs.go | 18 ++++++++++++++---- 2 files changed, 43 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/vfs/device.go b/pkg/sentry/vfs/device.go index cb672e36f..9f9d6e783 100644 --- a/pkg/sentry/vfs/device.go +++ b/pkg/sentry/vfs/device.go @@ -98,3 +98,32 @@ func (vfs *VirtualFilesystem) OpenDeviceSpecialFile(ctx context.Context, mnt *Mo } return rd.dev.Open(ctx, mnt, d, *opts) } + +// GetAnonBlockDevMinor allocates and returns an unused minor device number for +// an "anonymous" block device with major number 0. +func (vfs *VirtualFilesystem) GetAnonBlockDevMinor() (uint32, error) { + vfs.anonBlockDevMinorMu.Lock() + defer vfs.anonBlockDevMinorMu.Unlock() + minor := vfs.anonBlockDevMinorNext + const maxDevMinor = (1 << 20) - 1 + for minor < maxDevMinor { + if _, ok := vfs.anonBlockDevMinor[minor]; !ok { + vfs.anonBlockDevMinor[minor] = struct{}{} + vfs.anonBlockDevMinorNext = minor + 1 + return minor, nil + } + minor++ + } + return 0, syserror.EMFILE +} + +// PutAnonBlockDevMinor deallocates a minor device number returned by a +// previous call to GetAnonBlockDevMinor. +func (vfs *VirtualFilesystem) PutAnonBlockDevMinor(minor uint32) { + vfs.anonBlockDevMinorMu.Lock() + defer vfs.anonBlockDevMinorMu.Unlock() + delete(vfs.anonBlockDevMinor, minor) + if minor < vfs.anonBlockDevMinorNext { + vfs.anonBlockDevMinorNext = minor + } +} diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 1f21b0b31..1f6f56293 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -80,6 +80,14 @@ type VirtualFilesystem struct { devicesMu sync.RWMutex devices map[devTuple]*registeredDevice + // anonBlockDevMinor contains all allocated anonymous block device minor + // numbers. anonBlockDevMinorNext is a lower bound for the smallest + // unallocated anonymous block device number. anonBlockDevMinorNext and + // anonBlockDevMinor are protected by anonBlockDevMinorMu. + anonBlockDevMinorMu sync.Mutex + anonBlockDevMinorNext uint32 + anonBlockDevMinor map[uint32]struct{} + // fsTypes contains all registered FilesystemTypes. fsTypes is protected by // fsTypesMu. fsTypesMu sync.RWMutex @@ -94,10 +102,12 @@ type VirtualFilesystem struct { // New returns a new VirtualFilesystem with no mounts or FilesystemTypes. func New() *VirtualFilesystem { vfs := &VirtualFilesystem{ - mountpoints: make(map[*Dentry]map[*Mount]struct{}), - devices: make(map[devTuple]*registeredDevice), - fsTypes: make(map[string]*registeredFilesystemType), - filesystems: make(map[*Filesystem]struct{}), + mountpoints: make(map[*Dentry]map[*Mount]struct{}), + devices: make(map[devTuple]*registeredDevice), + anonBlockDevMinorNext: 1, + anonBlockDevMinor: make(map[uint32]struct{}), + fsTypes: make(map[string]*registeredFilesystemType), + filesystems: make(map[*Filesystem]struct{}), } vfs.mounts.Init() return vfs -- cgit v1.2.3 From 18a7e1309decb9bc09879e337adbc00f81d420c5 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Fri, 24 Jan 2020 17:06:30 -0800 Subject: Add support for device special files to VFS2 tmpfs. PiperOrigin-RevId: 291471892 --- pkg/sentry/fsimpl/tmpfs/BUILD | 1 + pkg/sentry/fsimpl/tmpfs/device_file.go | 39 ++++++++++++++++++++++++++++++ pkg/sentry/fsimpl/tmpfs/filesystem.go | 43 +++++++++++++++++++--------------- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 30 +++++++++++++++++++----- 4 files changed, 88 insertions(+), 25 deletions(-) create mode 100644 pkg/sentry/fsimpl/tmpfs/device_file.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD index 7601c7c04..691476b4f 100644 --- a/pkg/sentry/fsimpl/tmpfs/BUILD +++ b/pkg/sentry/fsimpl/tmpfs/BUILD @@ -20,6 +20,7 @@ go_library( name = "tmpfs", srcs = [ "dentry_list.go", + "device_file.go", "directory.go", "filesystem.go", "named_pipe.go", diff --git a/pkg/sentry/fsimpl/tmpfs/device_file.go b/pkg/sentry/fsimpl/tmpfs/device_file.go new file mode 100644 index 000000000..84b181b90 --- /dev/null +++ b/pkg/sentry/fsimpl/tmpfs/device_file.go @@ -0,0 +1,39 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tmpfs + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" +) + +type deviceFile struct { + inode inode + kind vfs.DeviceKind + major uint32 + minor uint32 +} + +func (fs *filesystem) newDeviceFile(creds *auth.Credentials, mode linux.FileMode, kind vfs.DeviceKind, major, minor uint32) *inode { + file := &deviceFile{ + kind: kind, + major: major, + minor: minor, + } + file.inode.init(file, fs, creds, mode) + file.inode.nlink = 1 // from parent directory + return &file.inode +} diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index a9f66a42a..d726f03c5 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -228,23 +228,26 @@ func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts v // MknodAt implements vfs.FilesystemImpl.MknodAt. func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MknodOptions) error { return fs.doCreateAt(rp, false /* dir */, func(parent *dentry, name string) error { + var childInode *inode switch opts.Mode.FileType() { case 0, linux.S_IFREG: - child := fs.newDentry(fs.newRegularFile(rp.Credentials(), opts.Mode)) - parent.vfsd.InsertChild(&child.vfsd, name) - parent.inode.impl.(*directory).childList.PushBack(child) - return nil + childInode = fs.newRegularFile(rp.Credentials(), opts.Mode) case linux.S_IFIFO: - child := fs.newDentry(fs.newNamedPipe(rp.Credentials(), opts.Mode)) - parent.vfsd.InsertChild(&child.vfsd, name) - parent.inode.impl.(*directory).childList.PushBack(child) - return nil - case linux.S_IFBLK, linux.S_IFCHR, linux.S_IFSOCK: + childInode = fs.newNamedPipe(rp.Credentials(), opts.Mode) + case linux.S_IFBLK: + childInode = fs.newDeviceFile(rp.Credentials(), opts.Mode, vfs.BlockDevice, opts.DevMajor, opts.DevMinor) + case linux.S_IFCHR: + childInode = fs.newDeviceFile(rp.Credentials(), opts.Mode, vfs.CharDevice, opts.DevMajor, opts.DevMinor) + case linux.S_IFSOCK: // Not yet supported. return syserror.EPERM default: return syserror.EINVAL } + child := fs.newDentry(childInode) + parent.vfsd.InsertChild(&child.vfsd, name) + parent.inode.impl.(*directory).childList.PushBack(child) + return nil }) } @@ -264,7 +267,7 @@ func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf if err != nil { return nil, err } - return d.open(ctx, rp, opts.Flags, false /* afterCreate */) + return d.open(ctx, rp, &opts, false /* afterCreate */) } mustCreate := opts.Flags&linux.O_EXCL != 0 @@ -279,7 +282,7 @@ func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf if mustCreate { return nil, syserror.EEXIST } - return start.open(ctx, rp, opts.Flags, false /* afterCreate */) + return start.open(ctx, rp, &opts, false /* afterCreate */) } afterTrailingSymlink: parent, err := walkParentDirLocked(rp, start) @@ -313,7 +316,7 @@ afterTrailingSymlink: child := fs.newDentry(fs.newRegularFile(rp.Credentials(), opts.Mode)) parent.vfsd.InsertChild(&child.vfsd, name) parent.inode.impl.(*directory).childList.PushBack(child) - return child.open(ctx, rp, opts.Flags, true) + return child.open(ctx, rp, &opts, true) } if err != nil { return nil, err @@ -327,11 +330,11 @@ afterTrailingSymlink: if mustCreate { return nil, syserror.EEXIST } - return child.open(ctx, rp, opts.Flags, false) + return child.open(ctx, rp, &opts, false) } -func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, flags uint32, afterCreate bool) (*vfs.FileDescription, error) { - ats := vfs.AccessTypesForOpenFlags(flags) +func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.OpenOptions, afterCreate bool) (*vfs.FileDescription, error) { + ats := vfs.AccessTypesForOpenFlags(opts.Flags) if !afterCreate { if err := d.inode.checkPermissions(rp.Credentials(), ats, d.inode.isDir()); err != nil { return nil, err @@ -340,10 +343,10 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, flags uint32, switch impl := d.inode.impl.(type) { case *regularFile: var fd regularFileFD - if err := fd.vfsfd.Init(&fd, flags, rp.Mount(), &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { + if err := fd.vfsfd.Init(&fd, opts.Flags, rp.Mount(), &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { return nil, err } - if flags&linux.O_TRUNC != 0 { + if opts.Flags&linux.O_TRUNC != 0 { impl.mu.Lock() impl.data.Truncate(0, impl.memFile) atomic.StoreUint64(&impl.size, 0) @@ -356,7 +359,7 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, flags uint32, return nil, syserror.EISDIR } var fd directoryFD - if err := fd.vfsfd.Init(&fd, flags, rp.Mount(), &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { + if err := fd.vfsfd.Init(&fd, opts.Flags, rp.Mount(), &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { return nil, err } return &fd.vfsfd, nil @@ -364,7 +367,9 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, flags uint32, // Can't open symlinks without O_PATH (which is unimplemented). return nil, syserror.ELOOP case *namedPipe: - return newNamedPipeFD(ctx, impl, rp, &d.vfsd, flags) + return newNamedPipeFD(ctx, impl, rp, &d.vfsd, opts.Flags) + case *deviceFile: + return rp.VirtualFilesystem().OpenDeviceSpecialFile(ctx, rp.Mount(), &d.vfsd, impl.kind, impl.major, impl.minor, opts) default: panic(fmt.Sprintf("unknown inode type: %T", d.inode.impl)) } diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 1d4889c89..515f033f2 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -149,6 +149,10 @@ type inode struct { ctime int64 // nanoseconds mtime int64 // nanoseconds + // Only meaningful for device special files. + rdevMajor uint32 + rdevMinor uint32 + impl interface{} // immutable } @@ -269,6 +273,15 @@ func (i *inode) statTo(stat *linux.Statx) { stat.Blocks = allocatedBlocksForSize(stat.Size) case *namedPipe: stat.Mode |= linux.S_IFIFO + case *deviceFile: + switch impl.kind { + case vfs.BlockDevice: + stat.Mode |= linux.S_IFBLK + case vfs.CharDevice: + stat.Mode |= linux.S_IFCHR + } + stat.RdevMajor = impl.major + stat.RdevMinor = impl.minor default: panic(fmt.Sprintf("unknown inode type: %T", i.impl)) } @@ -309,12 +322,8 @@ func (i *inode) setStat(stat linux.Statx) error { } case *directory: return syserror.EISDIR - case *symlink: - return syserror.EINVAL - case *namedPipe: - // Nothing. default: - panic(fmt.Sprintf("unknown inode type: %T", i.impl)) + return syserror.EINVAL } } if mask&linux.STATX_ATIME != 0 { @@ -353,13 +362,22 @@ func allocatedBlocksForSize(size uint64) uint64 { } func (i *inode) direntType() uint8 { - switch i.impl.(type) { + switch impl := i.impl.(type) { case *regularFile: return linux.DT_REG case *directory: return linux.DT_DIR case *symlink: return linux.DT_LNK + case *deviceFile: + switch impl.kind { + case vfs.BlockDevice: + return linux.DT_BLK + case vfs.CharDevice: + return linux.DT_CHR + default: + panic(fmt.Sprintf("unknown vfs.DeviceKind: %v", impl.kind)) + } default: panic(fmt.Sprintf("unknown inode type: %T", i.impl)) } -- cgit v1.2.3 From 2946fe81627afa223853769ed736e2a56e0144b7 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Fri, 24 Jan 2020 17:12:03 -0800 Subject: We can now actually write out the udp matcher. --- pkg/sentry/socket/netfilter/netfilter.go | 78 ++++++++++++++++++++++++-------- 1 file changed, 58 insertions(+), 20 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 3ca22932d..6c88a50a6 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -36,7 +36,7 @@ const errorTargetName = "ERROR" // metadata is opaque to netstack. It holds data that we need to translate // between Linux's and netstack's iptables representations. -// TODO(gvisor.dev/issue/170): This might be removable. +// TODO(gvisor.dev/issue/170): Use metadata to check correctness. type metadata struct { HookEntry [linux.NF_INET_NUMHOOKS]uint32 Underflow [linux.NF_INET_NUMHOOKS]uint32 @@ -44,6 +44,14 @@ type metadata struct { Size uint32 } +const enableDebugLog = true + +func nflog(format string, args ...interface{}) { + if enableDebugLog { + log.Infof("netfilter: "+format, args...) + } +} + // GetInfo returns information about iptables. func GetInfo(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr) (linux.IPTGetinfo, *syserr.Error) { // Read in the struct and table name. @@ -72,6 +80,8 @@ func GetInfo(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr) (linux.IPT info.NumEntries = metadata.NumEntries info.Size = metadata.Size + nflog("GetInfo returning info: %+v", info) + return info, nil } @@ -80,21 +90,26 @@ func GetEntries(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr, outLen // Read in the struct and table name. var userEntries linux.IPTGetEntries if _, err := t.CopyIn(outPtr, &userEntries); err != nil { + log.Warningf("netfilter: couldn't copy in entries %q", userEntries.Name) return linux.KernelIPTGetEntries{}, syserr.FromError(err) } // Find the appropriate table. table, err := findTable(stack, userEntries.Name) if err != nil { + log.Warningf("netfilter: couldn't find table %q", userEntries.Name) return linux.KernelIPTGetEntries{}, err } // Convert netstack's iptables rules to something that the iptables // tool can understand. - entries, _, err := convertNetstackToBinary(userEntries.Name.String(), table) + entries, meta, err := convertNetstackToBinary(userEntries.Name.String(), table) if err != nil { return linux.KernelIPTGetEntries{}, err } + if meta != table.Metadata().(metadata) { + panic(fmt.Sprintf("Table %q metadata changed between writing and reading. Was saved as %+v, but is now %+v", userEntries.Name.String(), table.Metadata().(metadata), meta)) + } if binary.Size(entries) > uintptr(outLen) { log.Warningf("Insufficient GetEntries output size: %d", uintptr(outLen)) return linux.KernelIPTGetEntries{}, syserr.ErrInvalidArgument @@ -148,15 +163,19 @@ func convertNetstackToBinary(tablename string, table iptables.Table) (linux.Kern copy(entries.Name[:], tablename) for ruleIdx, rule := range table.Rules { + nflog("Current offset: %d", entries.Size) + // Is this a chain entry point? for hook, hookRuleIdx := range table.BuiltinChains { if hookRuleIdx == ruleIdx { + nflog("Found hook %d at offset %d", hook, entries.Size) meta.HookEntry[hook] = entries.Size } } // Is this a chain underflow point? for underflow, underflowRuleIdx := range table.Underflows { if underflowRuleIdx == ruleIdx { + nflog("Found underflow %d at offset %d", underflow, entries.Size) meta.Underflow[underflow] = entries.Size } } @@ -176,6 +195,10 @@ func convertNetstackToBinary(tablename string, table iptables.Table) (linux.Kern // Serialize the matcher and add it to the // entry. serialized := marshalMatcher(matcher) + nflog("matcher serialized as: %v", serialized) + if len(serialized)%8 != 0 { + panic(fmt.Sprintf("matcher %T is not 64-bit aligned", matcher)) + } entry.Elems = append(entry.Elems, serialized...) entry.NextOffset += uint16(len(serialized)) entry.TargetOffset += uint16(len(serialized)) @@ -183,18 +206,25 @@ func convertNetstackToBinary(tablename string, table iptables.Table) (linux.Kern // Serialize and append the target. serialized := marshalTarget(rule.Target) + if len(serialized)%8 != 0 { + panic(fmt.Sprintf("target %T is not 64-bit aligned", rule.Target)) + } entry.Elems = append(entry.Elems, serialized...) entry.NextOffset += uint16(len(serialized)) + nflog("Adding entry: %+v", entry) + entries.Size += uint32(entry.NextOffset) entries.Entrytable = append(entries.Entrytable, entry) meta.NumEntries++ } + nflog("Finished with an marshalled size of %d", meta.Size) meta.Size = entries.Size return entries, meta, nil } +// TODO: SOMEHOW THIS IS NOT GETTING APPENDED! func marshalMatcher(matcher iptables.Matcher) []byte { switch m := matcher.(type) { case *iptables.UDPMatcher: @@ -207,17 +237,17 @@ func marshalMatcher(matcher iptables.Matcher) []byte { } func marshalUDPMatcher(matcher *iptables.UDPMatcher) []byte { + nflog("Marshalling UDP matcher: %+v", matcher) + linuxMatcher := linux.KernelXTEntryMatch{ XTEntryMatch: linux.XTEntryMatch{ - MatchSize: linux.SizeOfXTEntryMatch + linux.SizeOfXTUDP, + MatchSize: linux.SizeOfXTEntryMatch + linux.SizeOfXTUDP + 6, // Name: "udp", }, - Data: make([]byte, linux.SizeOfXTUDP+22), + Data: make([]byte, 0, linux.SizeOfXTUDP), } - // copy(linuxMatcher.Name[:], "udp") copy(linuxMatcher.Name[:], "udp") - // TODO: Must be aligned. xtudp := linux.XTUDP{ SourcePortStart: matcher.Data.SourcePortStart, SourcePortEnd: matcher.Data.SourcePortEnd, @@ -225,17 +255,17 @@ func marshalUDPMatcher(matcher *iptables.UDPMatcher) []byte { DestinationPortEnd: matcher.Data.DestinationPortEnd, InverseFlags: matcher.Data.InverseFlags, } - binary.Marshal(linuxMatcher.Data[:linux.SizeOfXTUDP], usermem.ByteOrder, xtudp) - - if binary.Size(linuxMatcher)%64 != 0 { - panic(fmt.Sprintf("size is actually: %d", binary.Size(linuxMatcher))) - } - - var buf [linux.SizeOfXTEntryMatch + linux.SizeOfXTUDP + 22]byte - if len(buf)%64 != 0 { - panic(fmt.Sprintf("len is actually: %d", len(buf))) - } - binary.Marshal(buf[:], usermem.ByteOrder, linuxMatcher) + nflog("marshalUDPMatcher: xtudp: %+v", xtudp) + linuxMatcher.Data = binary.Marshal(linuxMatcher.Data, usermem.ByteOrder, xtudp) + nflog("marshalUDPMatcher: linuxMatcher: %+v", linuxMatcher) + + // We have to pad this struct size to a multiple of 8 bytes, so we make + // this a little longer than it needs to be. + buf := make([]byte, 0, linux.SizeOfXTEntryMatch+linux.SizeOfXTUDP+6) + buf = binary.Marshal(buf, usermem.ByteOrder, linuxMatcher) + buf = append(buf, []byte{0, 0, 0, 0, 0, 0}...) + nflog("Marshalled into matcher of size %d", len(buf)) + nflog("marshalUDPMatcher: buf is: %v", buf) return buf[:] } @@ -253,6 +283,8 @@ func marshalTarget(target iptables.Target) []byte { } func marshalStandardTarget(verdict iptables.Verdict) []byte { + nflog("Marshalling standard target with size %d", linux.SizeOfXTStandardTarget) + // TODO: Must be aligned. // The target's name will be the empty string. target := linux.XTStandardTarget{ @@ -321,7 +353,7 @@ func translateToStandardVerdict(val int32) (iptables.Verdict, *syserr.Error) { // SetEntries sets iptables rules for a single table. See // net/ipv4/netfilter/ip_tables.c:translate_table for reference. func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { - printReplace(optVal) + // printReplace(optVal) // Get the basic rules data (struct ipt_replace). if len(optVal) < linux.SizeOfIPTReplace { @@ -343,10 +375,14 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { return syserr.ErrInvalidArgument } + nflog("Setting entries in table %q", replace.Name.String()) + // Convert input into a list of rules and their offsets. var offset uint32 var offsets []uint32 for entryIdx := uint32(0); entryIdx < replace.NumEntries; entryIdx++ { + nflog("Processing entry at offset %d", offset) + // Get the struct ipt_entry. if len(optVal) < linux.SizeOfIPTEntry { log.Warningf("netfilter: optVal has insufficient size for entry %d", len(optVal)) @@ -464,9 +500,10 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { // parseMatchers parses 0 or more matchers from optVal. optVal should contain // only the matchers. func parseMatchers(filter iptables.IPHeaderFilter, optVal []byte) ([]iptables.Matcher, *syserr.Error) { + nflog("Parsing matchers of size %d", len(optVal)) var matchers []iptables.Matcher for len(optVal) > 0 { - log.Infof("parseMatchers: optVal has len %d", len(optVal)) + nflog("parseMatchers: optVal has len %d", len(optVal)) // Get the XTEntryMatch. if len(optVal) < linux.SizeOfXTEntryMatch { log.Warningf("netfilter: optVal has insufficient size for entry match: %d", len(optVal)) @@ -475,7 +512,7 @@ func parseMatchers(filter iptables.IPHeaderFilter, optVal []byte) ([]iptables.Ma var match linux.XTEntryMatch buf := optVal[:linux.SizeOfXTEntryMatch] binary.Unmarshal(buf, usermem.ByteOrder, &match) - log.Infof("parseMatchers: parsed entry match %q: %+v", match.Name.String(), match) + nflog("parseMatchers: parsed entry match %q: %+v", match.Name.String(), match) // Check some invariants. if match.MatchSize < linux.SizeOfXTEntryMatch { @@ -532,6 +569,7 @@ func parseMatchers(filter iptables.IPHeaderFilter, optVal []byte) ([]iptables.Ma // parseTarget parses a target from optVal. optVal should contain only the // target. func parseTarget(optVal []byte) (iptables.Target, *syserr.Error) { + nflog("Parsing target of size %d", len(optVal)) if len(optVal) < linux.SizeOfXTEntryTarget { log.Warningf("netfilter: optVal has insufficient size for entry target %d", len(optVal)) return nil, syserr.ErrInvalidArgument -- cgit v1.2.3 From d29e59af9fbd420e34378bcbf7ae543134070217 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Mon, 27 Jan 2020 10:04:07 -0800 Subject: Standardize on tools directory. PiperOrigin-RevId: 291745021 --- .bazelrc | 8 +- BUILD | 49 ++++++- benchmarks/defs.bzl | 18 --- benchmarks/harness/BUILD | 74 +++++----- benchmarks/harness/machine_producers/BUILD | 4 +- benchmarks/runner/BUILD | 24 ++-- benchmarks/tcp/BUILD | 3 +- benchmarks/workloads/ab/BUILD | 19 ++- benchmarks/workloads/absl/BUILD | 19 ++- benchmarks/workloads/curl/BUILD | 2 +- benchmarks/workloads/ffmpeg/BUILD | 2 +- benchmarks/workloads/fio/BUILD | 19 ++- benchmarks/workloads/httpd/BUILD | 2 +- benchmarks/workloads/iperf/BUILD | 19 ++- benchmarks/workloads/netcat/BUILD | 2 +- benchmarks/workloads/nginx/BUILD | 2 +- benchmarks/workloads/node/BUILD | 2 +- benchmarks/workloads/node_template/BUILD | 2 +- benchmarks/workloads/redis/BUILD | 2 +- benchmarks/workloads/redisbenchmark/BUILD | 19 ++- benchmarks/workloads/ruby/BUILD | 2 +- benchmarks/workloads/ruby_template/BUILD | 2 +- benchmarks/workloads/sleep/BUILD | 2 +- benchmarks/workloads/sysbench/BUILD | 19 ++- benchmarks/workloads/syscall/BUILD | 19 ++- benchmarks/workloads/tensorflow/BUILD | 2 +- benchmarks/workloads/true/BUILD | 2 +- pkg/abi/BUILD | 3 +- pkg/abi/linux/BUILD | 6 +- pkg/amutex/BUILD | 6 +- pkg/atomicbitops/BUILD | 6 +- pkg/binary/BUILD | 6 +- pkg/bits/BUILD | 6 +- pkg/bpf/BUILD | 6 +- pkg/compressio/BUILD | 6 +- pkg/control/client/BUILD | 3 +- pkg/control/server/BUILD | 3 +- pkg/cpuid/BUILD | 8 +- pkg/eventchannel/BUILD | 16 +-- pkg/fd/BUILD | 6 +- pkg/fdchannel/BUILD | 8 +- pkg/fdnotifier/BUILD | 3 +- pkg/flipcall/BUILD | 8 +- pkg/fspath/BUILD | 13 +- pkg/gate/BUILD | 4 +- pkg/goid/BUILD | 6 +- pkg/ilist/BUILD | 6 +- pkg/linewriter/BUILD | 6 +- pkg/log/BUILD | 6 +- pkg/memutil/BUILD | 3 +- pkg/metric/BUILD | 23 +-- pkg/p9/BUILD | 6 +- pkg/p9/p9test/BUILD | 6 +- pkg/procid/BUILD | 8 +- pkg/rand/BUILD | 3 +- pkg/refs/BUILD | 6 +- pkg/seccomp/BUILD | 6 +- pkg/secio/BUILD | 6 +- pkg/segment/test/BUILD | 6 +- pkg/sentry/BUILD | 2 + pkg/sentry/arch/BUILD | 20 +-- pkg/sentry/context/BUILD | 3 +- pkg/sentry/context/contexttest/BUILD | 3 +- pkg/sentry/control/BUILD | 8 +- pkg/sentry/device/BUILD | 6 +- pkg/sentry/fs/BUILD | 6 +- pkg/sentry/fs/anon/BUILD | 3 +- pkg/sentry/fs/dev/BUILD | 3 +- pkg/sentry/fs/fdpipe/BUILD | 6 +- pkg/sentry/fs/filetest/BUILD | 3 +- pkg/sentry/fs/fsutil/BUILD | 6 +- pkg/sentry/fs/gofer/BUILD | 6 +- pkg/sentry/fs/host/BUILD | 6 +- pkg/sentry/fs/lock/BUILD | 6 +- pkg/sentry/fs/proc/BUILD | 6 +- pkg/sentry/fs/proc/device/BUILD | 3 +- pkg/sentry/fs/proc/seqfile/BUILD | 6 +- pkg/sentry/fs/ramfs/BUILD | 6 +- pkg/sentry/fs/sys/BUILD | 3 +- pkg/sentry/fs/timerfd/BUILD | 3 +- pkg/sentry/fs/tmpfs/BUILD | 6 +- pkg/sentry/fs/tty/BUILD | 6 +- pkg/sentry/fsimpl/ext/BUILD | 6 +- pkg/sentry/fsimpl/ext/benchmark/BUILD | 2 +- pkg/sentry/fsimpl/ext/disklayout/BUILD | 6 +- pkg/sentry/fsimpl/kernfs/BUILD | 6 +- pkg/sentry/fsimpl/proc/BUILD | 8 +- pkg/sentry/fsimpl/sys/BUILD | 6 +- pkg/sentry/fsimpl/testutil/BUILD | 5 +- pkg/sentry/fsimpl/tmpfs/BUILD | 8 +- pkg/sentry/hostcpu/BUILD | 6 +- pkg/sentry/hostmm/BUILD | 3 +- pkg/sentry/inet/BUILD | 3 +- pkg/sentry/kernel/BUILD | 24 +--- pkg/sentry/kernel/auth/BUILD | 3 +- pkg/sentry/kernel/contexttest/BUILD | 3 +- pkg/sentry/kernel/epoll/BUILD | 6 +- pkg/sentry/kernel/eventfd/BUILD | 6 +- pkg/sentry/kernel/fasync/BUILD | 3 +- pkg/sentry/kernel/futex/BUILD | 6 +- pkg/sentry/kernel/memevent/BUILD | 20 +-- pkg/sentry/kernel/pipe/BUILD | 6 +- pkg/sentry/kernel/sched/BUILD | 6 +- pkg/sentry/kernel/semaphore/BUILD | 6 +- pkg/sentry/kernel/shm/BUILD | 3 +- pkg/sentry/kernel/signalfd/BUILD | 5 +- pkg/sentry/kernel/time/BUILD | 3 +- pkg/sentry/limits/BUILD | 6 +- pkg/sentry/loader/BUILD | 4 +- pkg/sentry/memmap/BUILD | 6 +- pkg/sentry/mm/BUILD | 6 +- pkg/sentry/pgalloc/BUILD | 6 +- pkg/sentry/platform/BUILD | 3 +- pkg/sentry/platform/interrupt/BUILD | 6 +- pkg/sentry/platform/kvm/BUILD | 6 +- pkg/sentry/platform/kvm/testutil/BUILD | 3 +- pkg/sentry/platform/ptrace/BUILD | 3 +- pkg/sentry/platform/ring0/BUILD | 3 +- pkg/sentry/platform/ring0/gen_offsets/BUILD | 2 +- pkg/sentry/platform/ring0/pagetables/BUILD | 16 +-- pkg/sentry/platform/safecopy/BUILD | 6 +- pkg/sentry/safemem/BUILD | 6 +- pkg/sentry/sighandling/BUILD | 3 +- pkg/sentry/socket/BUILD | 3 +- pkg/sentry/socket/control/BUILD | 3 +- pkg/sentry/socket/hostinet/BUILD | 3 +- pkg/sentry/socket/netfilter/BUILD | 3 +- pkg/sentry/socket/netlink/BUILD | 3 +- pkg/sentry/socket/netlink/port/BUILD | 6 +- pkg/sentry/socket/netlink/route/BUILD | 3 +- pkg/sentry/socket/netlink/uevent/BUILD | 3 +- pkg/sentry/socket/netstack/BUILD | 3 +- pkg/sentry/socket/unix/BUILD | 3 +- pkg/sentry/socket/unix/transport/BUILD | 3 +- pkg/sentry/state/BUILD | 3 +- pkg/sentry/strace/BUILD | 20 +-- pkg/sentry/syscalls/BUILD | 3 +- pkg/sentry/syscalls/linux/BUILD | 3 +- pkg/sentry/time/BUILD | 6 +- pkg/sentry/unimpl/BUILD | 21 +-- pkg/sentry/uniqueid/BUILD | 3 +- pkg/sentry/usage/BUILD | 5 +- pkg/sentry/usermem/BUILD | 7 +- pkg/sentry/vfs/BUILD | 8 +- pkg/sentry/watchdog/BUILD | 3 +- pkg/sleep/BUILD | 6 +- pkg/state/BUILD | 17 +-- pkg/state/statefile/BUILD | 6 +- pkg/sync/BUILD | 6 +- pkg/sync/atomicptrtest/BUILD | 6 +- pkg/sync/seqatomictest/BUILD | 6 +- pkg/syserr/BUILD | 3 +- pkg/syserror/BUILD | 4 +- pkg/tcpip/BUILD | 6 +- pkg/tcpip/adapters/gonet/BUILD | 6 +- pkg/tcpip/buffer/BUILD | 6 +- pkg/tcpip/checker/BUILD | 3 +- pkg/tcpip/hash/jenkins/BUILD | 6 +- pkg/tcpip/header/BUILD | 6 +- pkg/tcpip/iptables/BUILD | 3 +- pkg/tcpip/link/channel/BUILD | 3 +- pkg/tcpip/link/fdbased/BUILD | 6 +- pkg/tcpip/link/loopback/BUILD | 3 +- pkg/tcpip/link/muxed/BUILD | 6 +- pkg/tcpip/link/rawfile/BUILD | 3 +- pkg/tcpip/link/sharedmem/BUILD | 6 +- pkg/tcpip/link/sharedmem/pipe/BUILD | 6 +- pkg/tcpip/link/sharedmem/queue/BUILD | 6 +- pkg/tcpip/link/sniffer/BUILD | 3 +- pkg/tcpip/link/tun/BUILD | 3 +- pkg/tcpip/link/waitable/BUILD | 6 +- pkg/tcpip/network/BUILD | 2 +- pkg/tcpip/network/arp/BUILD | 4 +- pkg/tcpip/network/fragmentation/BUILD | 6 +- pkg/tcpip/network/hash/BUILD | 3 +- pkg/tcpip/network/ipv4/BUILD | 4 +- pkg/tcpip/network/ipv6/BUILD | 6 +- pkg/tcpip/ports/BUILD | 6 +- pkg/tcpip/sample/tun_tcp_connect/BUILD | 2 +- pkg/tcpip/sample/tun_tcp_echo/BUILD | 2 +- pkg/tcpip/seqnum/BUILD | 3 +- pkg/tcpip/stack/BUILD | 6 +- pkg/tcpip/transport/icmp/BUILD | 3 +- pkg/tcpip/transport/packet/BUILD | 3 +- pkg/tcpip/transport/raw/BUILD | 3 +- pkg/tcpip/transport/tcp/BUILD | 4 +- pkg/tcpip/transport/tcp/testing/context/BUILD | 3 +- pkg/tcpip/transport/tcpconntrack/BUILD | 4 +- pkg/tcpip/transport/udp/BUILD | 4 +- pkg/tmutex/BUILD | 6 +- pkg/unet/BUILD | 6 +- pkg/urpc/BUILD | 6 +- pkg/waiter/BUILD | 6 +- runsc/BUILD | 27 ++-- runsc/boot/BUILD | 5 +- runsc/boot/filter/BUILD | 3 +- runsc/boot/platforms/BUILD | 3 +- runsc/cgroup/BUILD | 5 +- runsc/cmd/BUILD | 5 +- runsc/console/BUILD | 3 +- runsc/container/BUILD | 5 +- runsc/container/test_app/BUILD | 4 +- runsc/criutil/BUILD | 3 +- runsc/dockerutil/BUILD | 3 +- runsc/fsgofer/BUILD | 9 +- runsc/fsgofer/filter/BUILD | 3 +- runsc/sandbox/BUILD | 3 +- runsc/specutils/BUILD | 5 +- runsc/testutil/BUILD | 3 +- runsc/version_test.sh | 2 +- scripts/common.sh | 6 +- scripts/common_bazel.sh | 99 ------------- scripts/common_build.sh | 99 +++++++++++++ test/BUILD | 45 +----- test/e2e/BUILD | 5 +- test/image/BUILD | 5 +- test/iptables/BUILD | 5 +- test/iptables/runner/BUILD | 12 +- test/root/BUILD | 5 +- test/root/testdata/BUILD | 3 +- test/runtimes/BUILD | 4 +- test/runtimes/build_defs.bzl | 5 +- test/runtimes/images/proctor/BUILD | 4 +- test/syscalls/BUILD | 2 +- test/syscalls/build_defs.bzl | 6 +- test/syscalls/gtest/BUILD | 7 +- test/syscalls/linux/BUILD | 23 ++- test/syscalls/linux/arch_prctl.cc | 2 + test/syscalls/linux/rseq/BUILD | 5 +- .../linux/udp_socket_errqueue_test_case.cc | 4 + test/uds/BUILD | 3 +- test/util/BUILD | 27 ++-- test/util/save_util_linux.cc | 4 + test/util/save_util_other.cc | 4 + test/util/test_util_runfiles.cc | 4 + tools/BUILD | 3 + tools/build/BUILD | 10 ++ tools/build/defs.bzl | 91 ++++++++++++ tools/checkunsafe/BUILD | 3 +- tools/defs.bzl | 154 +++++++++++++++++++++ tools/go_generics/BUILD | 2 +- tools/go_generics/globals/BUILD | 4 +- tools/go_generics/go_merge/BUILD | 2 +- tools/go_generics/rules_tests/BUILD | 2 +- tools/go_marshal/BUILD | 4 +- tools/go_marshal/README.md | 52 +------ tools/go_marshal/analysis/BUILD | 5 +- tools/go_marshal/defs.bzl | 112 ++------------- tools/go_marshal/gomarshal/BUILD | 6 +- tools/go_marshal/gomarshal/generator.go | 20 ++- tools/go_marshal/gomarshal/generator_tests.go | 6 +- tools/go_marshal/main.go | 11 +- tools/go_marshal/marshal/BUILD | 5 +- tools/go_marshal/test/BUILD | 7 +- tools/go_marshal/test/external/BUILD | 6 +- tools/go_stateify/BUILD | 2 +- tools/go_stateify/defs.bzl | 79 +---------- tools/images/BUILD | 2 +- tools/images/defs.bzl | 6 +- tools/issue_reviver/BUILD | 2 +- tools/issue_reviver/github/BUILD | 3 +- tools/issue_reviver/reviver/BUILD | 5 +- tools/workspace_status.sh | 2 +- vdso/BUILD | 33 ++--- 264 files changed, 1012 insertions(+), 1380 deletions(-) delete mode 100644 benchmarks/defs.bzl delete mode 100755 scripts/common_bazel.sh create mode 100755 scripts/common_build.sh create mode 100644 tools/BUILD create mode 100644 tools/build/BUILD create mode 100644 tools/build/defs.bzl create mode 100644 tools/defs.bzl (limited to 'pkg/sentry') diff --git a/.bazelrc b/.bazelrc index 9c35c5e7b..ef214bcfa 100644 --- a/.bazelrc +++ b/.bazelrc @@ -30,10 +30,10 @@ build:remote --auth_scope="https://www.googleapis.com/auth/cloud-source-tools" # Add a custom platform and toolchain that builds in a privileged docker # container, which is required by our syscall tests. -build:remote --host_platform=//test:rbe_ubuntu1604 -build:remote --extra_toolchains=//test:cc-toolchain-clang-x86_64-default -build:remote --extra_execution_platforms=//test:rbe_ubuntu1604 -build:remote --platforms=//test:rbe_ubuntu1604 +build:remote --host_platform=//:rbe_ubuntu1604 +build:remote --extra_toolchains=//:cc-toolchain-clang-x86_64-default +build:remote --extra_execution_platforms=//:rbe_ubuntu1604 +build:remote --platforms=//:rbe_ubuntu1604 build:remote --crosstool_top=@rbe_default//cc:toolchain build:remote --jobs=50 build:remote --remote_timeout=3600 diff --git a/BUILD b/BUILD index 76286174f..5fd929378 100644 --- a/BUILD +++ b/BUILD @@ -1,8 +1,8 @@ -package(licenses = ["notice"]) # Apache 2.0 - load("@io_bazel_rules_go//go:def.bzl", "go_path", "nogo") load("@bazel_gazelle//:def.bzl", "gazelle") +package(licenses = ["notice"]) + # The sandbox filegroup is used for sandbox-internal dependencies. package_group( name = "sandbox", @@ -49,9 +49,52 @@ gazelle(name = "gazelle") # live in the tools subdirectory (unless they are standard). nogo( name = "nogo", - config = "tools/nogo.js", + config = "//tools:nogo.js", visibility = ["//visibility:public"], deps = [ "//tools/checkunsafe", ], ) + +# We need to define a bazel platform and toolchain to specify dockerPrivileged +# and dockerRunAsRoot options, they are required to run tests on the RBE +# cluster in Kokoro. +alias( + name = "rbe_ubuntu1604", + actual = ":rbe_ubuntu1604_r346485", +) + +platform( + name = "rbe_ubuntu1604_r346485", + constraint_values = [ + "@bazel_tools//platforms:x86_64", + "@bazel_tools//platforms:linux", + "@bazel_tools//tools/cpp:clang", + "@bazel_toolchains//constraints:xenial", + "@bazel_toolchains//constraints/sanitizers:support_msan", + ], + remote_execution_properties = """ + properties: { + name: "container-image" + value:"docker://gcr.io/cloud-marketplace/google/rbe-ubuntu16-04@sha256:93f7e127196b9b653d39830c50f8b05d49ef6fd8739a9b5b8ab16e1df5399e50" + } + properties: { + name: "dockerAddCapabilities" + value: "SYS_ADMIN" + } + properties: { + name: "dockerPrivileged" + value: "true" + } + """, +) + +toolchain( + name = "cc-toolchain-clang-x86_64-default", + exec_compatible_with = [ + ], + target_compatible_with = [ + ], + toolchain = "@bazel_toolchains//configs/ubuntu16_04_clang/10.0.0/bazel_2.0.0/cc:cc-compiler-k8", + toolchain_type = "@bazel_tools//tools/cpp:toolchain_type", +) diff --git a/benchmarks/defs.bzl b/benchmarks/defs.bzl deleted file mode 100644 index 79e6cdbc8..000000000 --- a/benchmarks/defs.bzl +++ /dev/null @@ -1,18 +0,0 @@ -"""Provides python helper functions.""" - -load("@pydeps//:requirements.bzl", _requirement = "requirement") - -def filter_deps(deps = None): - if deps == None: - deps = [] - return [dep for dep in deps if dep] - -def py_library(deps = None, **kwargs): - return native.py_library(deps = filter_deps(deps), **kwargs) - -def py_test(deps = None, **kwargs): - return native.py_test(deps = filter_deps(deps), **kwargs) - -def requirement(name, direct = True): - """ requirement returns the required dependency. """ - return _requirement(name) diff --git a/benchmarks/harness/BUILD b/benchmarks/harness/BUILD index 081a74243..52d4e42f8 100644 --- a/benchmarks/harness/BUILD +++ b/benchmarks/harness/BUILD @@ -1,4 +1,4 @@ -load("//benchmarks:defs.bzl", "py_library", "requirement") +load("//tools:defs.bzl", "py_library", "py_requirement") package( default_visibility = ["//benchmarks:__subpackages__"], @@ -25,16 +25,16 @@ py_library( srcs = ["container.py"], deps = [ "//benchmarks/workloads", - requirement("asn1crypto", False), - requirement("chardet", False), - requirement("certifi", False), - requirement("docker", True), - requirement("docker-pycreds", False), - requirement("idna", False), - requirement("ptyprocess", False), - requirement("requests", False), - requirement("urllib3", False), - requirement("websocket-client", False), + py_requirement("asn1crypto", False), + py_requirement("chardet", False), + py_requirement("certifi", False), + py_requirement("docker", True), + py_requirement("docker-pycreds", False), + py_requirement("idna", False), + py_requirement("ptyprocess", False), + py_requirement("requests", False), + py_requirement("urllib3", False), + py_requirement("websocket-client", False), ], ) @@ -47,17 +47,17 @@ py_library( "//benchmarks/harness:ssh_connection", "//benchmarks/harness:tunnel_dispatcher", "//benchmarks/harness/machine_mocks", - requirement("asn1crypto", False), - requirement("chardet", False), - requirement("certifi", False), - requirement("docker", True), - requirement("docker-pycreds", False), - requirement("idna", False), - requirement("ptyprocess", False), - requirement("requests", False), - requirement("six", False), - requirement("urllib3", False), - requirement("websocket-client", False), + py_requirement("asn1crypto", False), + py_requirement("chardet", False), + py_requirement("certifi", False), + py_requirement("docker", True), + py_requirement("docker-pycreds", False), + py_requirement("idna", False), + py_requirement("ptyprocess", False), + py_requirement("requests", False), + py_requirement("six", False), + py_requirement("urllib3", False), + py_requirement("websocket-client", False), ], ) @@ -66,10 +66,10 @@ py_library( srcs = ["ssh_connection.py"], deps = [ "//benchmarks/harness", - requirement("bcrypt", False), - requirement("cffi", True), - requirement("paramiko", True), - requirement("cryptography", False), + py_requirement("bcrypt", False), + py_requirement("cffi", True), + py_requirement("paramiko", True), + py_requirement("cryptography", False), ], ) @@ -77,16 +77,16 @@ py_library( name = "tunnel_dispatcher", srcs = ["tunnel_dispatcher.py"], deps = [ - requirement("asn1crypto", False), - requirement("chardet", False), - requirement("certifi", False), - requirement("docker", True), - requirement("docker-pycreds", False), - requirement("idna", False), - requirement("pexpect", True), - requirement("ptyprocess", False), - requirement("requests", False), - requirement("urllib3", False), - requirement("websocket-client", False), + py_requirement("asn1crypto", False), + py_requirement("chardet", False), + py_requirement("certifi", False), + py_requirement("docker", True), + py_requirement("docker-pycreds", False), + py_requirement("idna", False), + py_requirement("pexpect", True), + py_requirement("ptyprocess", False), + py_requirement("requests", False), + py_requirement("urllib3", False), + py_requirement("websocket-client", False), ], ) diff --git a/benchmarks/harness/machine_producers/BUILD b/benchmarks/harness/machine_producers/BUILD index c4e943882..48ea0ef39 100644 --- a/benchmarks/harness/machine_producers/BUILD +++ b/benchmarks/harness/machine_producers/BUILD @@ -1,4 +1,4 @@ -load("//benchmarks:defs.bzl", "py_library", "requirement") +load("//tools:defs.bzl", "py_library", "py_requirement") package( default_visibility = ["//benchmarks:__subpackages__"], @@ -31,7 +31,7 @@ py_library( deps = [ "//benchmarks/harness:machine", "//benchmarks/harness/machine_producers:machine_producer", - requirement("PyYAML", False), + py_requirement("PyYAML", False), ], ) diff --git a/benchmarks/runner/BUILD b/benchmarks/runner/BUILD index e1b2ea550..fae0ca800 100644 --- a/benchmarks/runner/BUILD +++ b/benchmarks/runner/BUILD @@ -1,4 +1,4 @@ -load("//benchmarks:defs.bzl", "py_library", "py_test", "requirement") +load("//tools:defs.bzl", "py_library", "py_requirement", "py_test") package(licenses = ["notice"]) @@ -28,7 +28,7 @@ py_library( "//benchmarks/suites:startup", "//benchmarks/suites:sysbench", "//benchmarks/suites:syscall", - requirement("click", True), + py_requirement("click", True), ], ) @@ -36,7 +36,7 @@ py_library( name = "commands", srcs = ["commands.py"], deps = [ - requirement("click", True), + py_requirement("click", True), ], ) @@ -50,14 +50,14 @@ py_test( ], deps = [ ":runner", - requirement("click", True), - requirement("attrs", False), - requirement("atomicwrites", False), - requirement("more-itertools", False), - requirement("pathlib2", False), - requirement("pluggy", False), - requirement("py", False), - requirement("pytest", True), - requirement("six", False), + py_requirement("click", True), + py_requirement("attrs", False), + py_requirement("atomicwrites", False), + py_requirement("more-itertools", False), + py_requirement("pathlib2", False), + py_requirement("pluggy", False), + py_requirement("py", False), + py_requirement("pytest", True), + py_requirement("six", False), ], ) diff --git a/benchmarks/tcp/BUILD b/benchmarks/tcp/BUILD index 735d7127f..d5e401acc 100644 --- a/benchmarks/tcp/BUILD +++ b/benchmarks/tcp/BUILD @@ -1,5 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_binary") -load("@rules_cc//cc:defs.bzl", "cc_binary") +load("//tools:defs.bzl", "cc_binary", "go_binary") package(licenses = ["notice"]) diff --git a/benchmarks/workloads/ab/BUILD b/benchmarks/workloads/ab/BUILD index 4fc0ab735..4dd91ceb3 100644 --- a/benchmarks/workloads/ab/BUILD +++ b/benchmarks/workloads/ab/BUILD @@ -1,5 +1,4 @@ -load("//benchmarks:defs.bzl", "py_library", "py_test", "requirement") -load("@rules_pkg//:pkg.bzl", "pkg_tar") +load("//tools:defs.bzl", "pkg_tar", "py_library", "py_requirement", "py_test") package( default_visibility = ["//benchmarks:__subpackages__"], @@ -17,14 +16,14 @@ py_test( python_version = "PY3", deps = [ ":ab", - requirement("attrs", False), - requirement("atomicwrites", False), - requirement("more-itertools", False), - requirement("pathlib2", False), - requirement("pluggy", False), - requirement("py", False), - requirement("pytest", True), - requirement("six", False), + py_requirement("attrs", False), + py_requirement("atomicwrites", False), + py_requirement("more-itertools", False), + py_requirement("pathlib2", False), + py_requirement("pluggy", False), + py_requirement("py", False), + py_requirement("pytest", True), + py_requirement("six", False), ], ) diff --git a/benchmarks/workloads/absl/BUILD b/benchmarks/workloads/absl/BUILD index 61e010096..55dae3baa 100644 --- a/benchmarks/workloads/absl/BUILD +++ b/benchmarks/workloads/absl/BUILD @@ -1,5 +1,4 @@ -load("//benchmarks:defs.bzl", "py_library", "py_test", "requirement") -load("@rules_pkg//:pkg.bzl", "pkg_tar") +load("//tools:defs.bzl", "pkg_tar", "py_library", "py_requirement", "py_test") package( default_visibility = ["//benchmarks:__subpackages__"], @@ -17,14 +16,14 @@ py_test( python_version = "PY3", deps = [ ":absl", - requirement("attrs", False), - requirement("atomicwrites", False), - requirement("more-itertools", False), - requirement("pathlib2", False), - requirement("pluggy", False), - requirement("py", False), - requirement("pytest", True), - requirement("six", False), + py_requirement("attrs", False), + py_requirement("atomicwrites", False), + py_requirement("more-itertools", False), + py_requirement("pathlib2", False), + py_requirement("pluggy", False), + py_requirement("py", False), + py_requirement("pytest", True), + py_requirement("six", False), ], ) diff --git a/benchmarks/workloads/curl/BUILD b/benchmarks/workloads/curl/BUILD index eb0fb6165..a70873065 100644 --- a/benchmarks/workloads/curl/BUILD +++ b/benchmarks/workloads/curl/BUILD @@ -1,4 +1,4 @@ -load("@rules_pkg//:pkg.bzl", "pkg_tar") +load("//tools:defs.bzl", "pkg_tar") package( default_visibility = ["//benchmarks:__subpackages__"], diff --git a/benchmarks/workloads/ffmpeg/BUILD b/benchmarks/workloads/ffmpeg/BUILD index be472dfb2..7c41ba631 100644 --- a/benchmarks/workloads/ffmpeg/BUILD +++ b/benchmarks/workloads/ffmpeg/BUILD @@ -1,4 +1,4 @@ -load("@rules_pkg//:pkg.bzl", "pkg_tar") +load("//tools:defs.bzl", "pkg_tar") package( default_visibility = ["//benchmarks:__subpackages__"], diff --git a/benchmarks/workloads/fio/BUILD b/benchmarks/workloads/fio/BUILD index de257adad..7b78e8e75 100644 --- a/benchmarks/workloads/fio/BUILD +++ b/benchmarks/workloads/fio/BUILD @@ -1,5 +1,4 @@ -load("//benchmarks:defs.bzl", "py_library", "py_test", "requirement") -load("@rules_pkg//:pkg.bzl", "pkg_tar") +load("//tools:defs.bzl", "pkg_tar", "py_library", "py_requirement", "py_test") package( default_visibility = ["//benchmarks:__subpackages__"], @@ -17,14 +16,14 @@ py_test( python_version = "PY3", deps = [ ":fio", - requirement("attrs", False), - requirement("atomicwrites", False), - requirement("more-itertools", False), - requirement("pathlib2", False), - requirement("pluggy", False), - requirement("py", False), - requirement("pytest", True), - requirement("six", False), + py_requirement("attrs", False), + py_requirement("atomicwrites", False), + py_requirement("more-itertools", False), + py_requirement("pathlib2", False), + py_requirement("pluggy", False), + py_requirement("py", False), + py_requirement("pytest", True), + py_requirement("six", False), ], ) diff --git a/benchmarks/workloads/httpd/BUILD b/benchmarks/workloads/httpd/BUILD index eb0fb6165..a70873065 100644 --- a/benchmarks/workloads/httpd/BUILD +++ b/benchmarks/workloads/httpd/BUILD @@ -1,4 +1,4 @@ -load("@rules_pkg//:pkg.bzl", "pkg_tar") +load("//tools:defs.bzl", "pkg_tar") package( default_visibility = ["//benchmarks:__subpackages__"], diff --git a/benchmarks/workloads/iperf/BUILD b/benchmarks/workloads/iperf/BUILD index 8832a996c..570f40148 100644 --- a/benchmarks/workloads/iperf/BUILD +++ b/benchmarks/workloads/iperf/BUILD @@ -1,5 +1,4 @@ -load("//benchmarks:defs.bzl", "py_library", "py_test", "requirement") -load("@rules_pkg//:pkg.bzl", "pkg_tar") +load("//tools:defs.bzl", "pkg_tar", "py_library", "py_requirement", "py_test") package( default_visibility = ["//benchmarks:__subpackages__"], @@ -17,14 +16,14 @@ py_test( python_version = "PY3", deps = [ ":iperf", - requirement("attrs", False), - requirement("atomicwrites", False), - requirement("more-itertools", False), - requirement("pathlib2", False), - requirement("pluggy", False), - requirement("py", False), - requirement("pytest", True), - requirement("six", False), + py_requirement("attrs", False), + py_requirement("atomicwrites", False), + py_requirement("more-itertools", False), + py_requirement("pathlib2", False), + py_requirement("pluggy", False), + py_requirement("py", False), + py_requirement("pytest", True), + py_requirement("six", False), ], ) diff --git a/benchmarks/workloads/netcat/BUILD b/benchmarks/workloads/netcat/BUILD index eb0fb6165..a70873065 100644 --- a/benchmarks/workloads/netcat/BUILD +++ b/benchmarks/workloads/netcat/BUILD @@ -1,4 +1,4 @@ -load("@rules_pkg//:pkg.bzl", "pkg_tar") +load("//tools:defs.bzl", "pkg_tar") package( default_visibility = ["//benchmarks:__subpackages__"], diff --git a/benchmarks/workloads/nginx/BUILD b/benchmarks/workloads/nginx/BUILD index eb0fb6165..a70873065 100644 --- a/benchmarks/workloads/nginx/BUILD +++ b/benchmarks/workloads/nginx/BUILD @@ -1,4 +1,4 @@ -load("@rules_pkg//:pkg.bzl", "pkg_tar") +load("//tools:defs.bzl", "pkg_tar") package( default_visibility = ["//benchmarks:__subpackages__"], diff --git a/benchmarks/workloads/node/BUILD b/benchmarks/workloads/node/BUILD index 71cd9f519..bfcf78cf9 100644 --- a/benchmarks/workloads/node/BUILD +++ b/benchmarks/workloads/node/BUILD @@ -1,4 +1,4 @@ -load("@rules_pkg//:pkg.bzl", "pkg_tar") +load("//tools:defs.bzl", "pkg_tar") package( default_visibility = ["//benchmarks:__subpackages__"], diff --git a/benchmarks/workloads/node_template/BUILD b/benchmarks/workloads/node_template/BUILD index ca996f068..e142f082a 100644 --- a/benchmarks/workloads/node_template/BUILD +++ b/benchmarks/workloads/node_template/BUILD @@ -1,4 +1,4 @@ -load("@rules_pkg//:pkg.bzl", "pkg_tar") +load("//tools:defs.bzl", "pkg_tar") package( default_visibility = ["//benchmarks:__subpackages__"], diff --git a/benchmarks/workloads/redis/BUILD b/benchmarks/workloads/redis/BUILD index eb0fb6165..a70873065 100644 --- a/benchmarks/workloads/redis/BUILD +++ b/benchmarks/workloads/redis/BUILD @@ -1,4 +1,4 @@ -load("@rules_pkg//:pkg.bzl", "pkg_tar") +load("//tools:defs.bzl", "pkg_tar") package( default_visibility = ["//benchmarks:__subpackages__"], diff --git a/benchmarks/workloads/redisbenchmark/BUILD b/benchmarks/workloads/redisbenchmark/BUILD index f5994a815..f472a4443 100644 --- a/benchmarks/workloads/redisbenchmark/BUILD +++ b/benchmarks/workloads/redisbenchmark/BUILD @@ -1,5 +1,4 @@ -load("//benchmarks:defs.bzl", "py_library", "py_test", "requirement") -load("@rules_pkg//:pkg.bzl", "pkg_tar") +load("//tools:defs.bzl", "pkg_tar", "py_library", "py_requirement", "py_test") package( default_visibility = ["//benchmarks:__subpackages__"], @@ -17,14 +16,14 @@ py_test( python_version = "PY3", deps = [ ":redisbenchmark", - requirement("attrs", False), - requirement("atomicwrites", False), - requirement("more-itertools", False), - requirement("pathlib2", False), - requirement("pluggy", False), - requirement("py", False), - requirement("pytest", True), - requirement("six", False), + py_requirement("attrs", False), + py_requirement("atomicwrites", False), + py_requirement("more-itertools", False), + py_requirement("pathlib2", False), + py_requirement("pluggy", False), + py_requirement("py", False), + py_requirement("pytest", True), + py_requirement("six", False), ], ) diff --git a/benchmarks/workloads/ruby/BUILD b/benchmarks/workloads/ruby/BUILD index e37d77804..a3be4fe92 100644 --- a/benchmarks/workloads/ruby/BUILD +++ b/benchmarks/workloads/ruby/BUILD @@ -1,4 +1,4 @@ -load("@rules_pkg//:pkg.bzl", "pkg_tar") +load("//tools:defs.bzl", "pkg_tar") package( default_visibility = ["//benchmarks:__subpackages__"], diff --git a/benchmarks/workloads/ruby_template/BUILD b/benchmarks/workloads/ruby_template/BUILD index 27f7c0c46..59443b14a 100644 --- a/benchmarks/workloads/ruby_template/BUILD +++ b/benchmarks/workloads/ruby_template/BUILD @@ -1,4 +1,4 @@ -load("@rules_pkg//:pkg.bzl", "pkg_tar") +load("//tools:defs.bzl", "pkg_tar") package( default_visibility = ["//benchmarks:__subpackages__"], diff --git a/benchmarks/workloads/sleep/BUILD b/benchmarks/workloads/sleep/BUILD index eb0fb6165..a70873065 100644 --- a/benchmarks/workloads/sleep/BUILD +++ b/benchmarks/workloads/sleep/BUILD @@ -1,4 +1,4 @@ -load("@rules_pkg//:pkg.bzl", "pkg_tar") +load("//tools:defs.bzl", "pkg_tar") package( default_visibility = ["//benchmarks:__subpackages__"], diff --git a/benchmarks/workloads/sysbench/BUILD b/benchmarks/workloads/sysbench/BUILD index fd2f8f03d..3834af7ed 100644 --- a/benchmarks/workloads/sysbench/BUILD +++ b/benchmarks/workloads/sysbench/BUILD @@ -1,5 +1,4 @@ -load("//benchmarks:defs.bzl", "py_library", "py_test", "requirement") -load("@rules_pkg//:pkg.bzl", "pkg_tar") +load("//tools:defs.bzl", "pkg_tar", "py_library", "py_requirement", "py_test") package( default_visibility = ["//benchmarks:__subpackages__"], @@ -17,14 +16,14 @@ py_test( python_version = "PY3", deps = [ ":sysbench", - requirement("attrs", False), - requirement("atomicwrites", False), - requirement("more-itertools", False), - requirement("pathlib2", False), - requirement("pluggy", False), - requirement("py", False), - requirement("pytest", True), - requirement("six", False), + py_requirement("attrs", False), + py_requirement("atomicwrites", False), + py_requirement("more-itertools", False), + py_requirement("pathlib2", False), + py_requirement("pluggy", False), + py_requirement("py", False), + py_requirement("pytest", True), + py_requirement("six", False), ], ) diff --git a/benchmarks/workloads/syscall/BUILD b/benchmarks/workloads/syscall/BUILD index 5100cbb21..dba4bb1e7 100644 --- a/benchmarks/workloads/syscall/BUILD +++ b/benchmarks/workloads/syscall/BUILD @@ -1,5 +1,4 @@ -load("//benchmarks:defs.bzl", "py_library", "py_test", "requirement") -load("@rules_pkg//:pkg.bzl", "pkg_tar") +load("//tools:defs.bzl", "pkg_tar", "py_library", "py_requirement", "py_test") package( default_visibility = ["//benchmarks:__subpackages__"], @@ -17,14 +16,14 @@ py_test( python_version = "PY3", deps = [ ":syscall", - requirement("attrs", False), - requirement("atomicwrites", False), - requirement("more-itertools", False), - requirement("pathlib2", False), - requirement("pluggy", False), - requirement("py", False), - requirement("pytest", True), - requirement("six", False), + py_requirement("attrs", False), + py_requirement("atomicwrites", False), + py_requirement("more-itertools", False), + py_requirement("pathlib2", False), + py_requirement("pluggy", False), + py_requirement("py", False), + py_requirement("pytest", True), + py_requirement("six", False), ], ) diff --git a/benchmarks/workloads/tensorflow/BUILD b/benchmarks/workloads/tensorflow/BUILD index 026c3b316..a7b7742f4 100644 --- a/benchmarks/workloads/tensorflow/BUILD +++ b/benchmarks/workloads/tensorflow/BUILD @@ -1,4 +1,4 @@ -load("@rules_pkg//:pkg.bzl", "pkg_tar") +load("//tools:defs.bzl", "pkg_tar") package( default_visibility = ["//benchmarks:__subpackages__"], diff --git a/benchmarks/workloads/true/BUILD b/benchmarks/workloads/true/BUILD index 221c4b9a7..eba23d325 100644 --- a/benchmarks/workloads/true/BUILD +++ b/benchmarks/workloads/true/BUILD @@ -1,4 +1,4 @@ -load("@rules_pkg//:pkg.bzl", "pkg_tar") +load("//tools:defs.bzl", "pkg_tar") package( default_visibility = ["//benchmarks:__subpackages__"], diff --git a/pkg/abi/BUILD b/pkg/abi/BUILD index f5c08ea06..839f822eb 100644 --- a/pkg/abi/BUILD +++ b/pkg/abi/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -9,6 +9,5 @@ go_library( "abi_linux.go", "flag.go", ], - importpath = "gvisor.dev/gvisor/pkg/abi", visibility = ["//:sandbox"], ) diff --git a/pkg/abi/linux/BUILD b/pkg/abi/linux/BUILD index 716ff22d2..1f3c0c687 100644 --- a/pkg/abi/linux/BUILD +++ b/pkg/abi/linux/BUILD @@ -1,5 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") # Package linux contains the constants and types needed to interface with a # Linux kernel. It should be used instead of syscall or golang.org/x/sys/unix @@ -60,7 +59,6 @@ go_library( "wait.go", "xattr.go", ], - importpath = "gvisor.dev/gvisor/pkg/abi/linux", visibility = ["//visibility:public"], deps = [ "//pkg/abi", @@ -73,7 +71,7 @@ go_test( name = "linux_test", size = "small", srcs = ["netfilter_test.go"], - embed = [":linux"], + library = ":linux", deps = [ "//pkg/binary", ], diff --git a/pkg/amutex/BUILD b/pkg/amutex/BUILD index d99e37b40..9612f072e 100644 --- a/pkg/amutex/BUILD +++ b/pkg/amutex/BUILD @@ -1,12 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) go_library( name = "amutex", srcs = ["amutex.go"], - importpath = "gvisor.dev/gvisor/pkg/amutex", visibility = ["//:sandbox"], ) @@ -14,6 +12,6 @@ go_test( name = "amutex_test", size = "small", srcs = ["amutex_test.go"], - embed = [":amutex"], + library = ":amutex", deps = ["//pkg/sync"], ) diff --git a/pkg/atomicbitops/BUILD b/pkg/atomicbitops/BUILD index 6403c60c2..3948074ba 100644 --- a/pkg/atomicbitops/BUILD +++ b/pkg/atomicbitops/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -11,7 +10,6 @@ go_library( "atomic_bitops_arm64.s", "atomic_bitops_common.go", ], - importpath = "gvisor.dev/gvisor/pkg/atomicbitops", visibility = ["//:sandbox"], ) @@ -19,6 +17,6 @@ go_test( name = "atomicbitops_test", size = "small", srcs = ["atomic_bitops_test.go"], - embed = [":atomicbitops"], + library = ":atomicbitops", deps = ["//pkg/sync"], ) diff --git a/pkg/binary/BUILD b/pkg/binary/BUILD index 543fb54bf..7ca2fda90 100644 --- a/pkg/binary/BUILD +++ b/pkg/binary/BUILD @@ -1,12 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) go_library( name = "binary", srcs = ["binary.go"], - importpath = "gvisor.dev/gvisor/pkg/binary", visibility = ["//:sandbox"], ) @@ -14,5 +12,5 @@ go_test( name = "binary_test", size = "small", srcs = ["binary_test.go"], - embed = [":binary"], + library = ":binary", ) diff --git a/pkg/bits/BUILD b/pkg/bits/BUILD index 93b88a29a..63f4670d7 100644 --- a/pkg/bits/BUILD +++ b/pkg/bits/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template", "go_template_instance") package(licenses = ["notice"]) @@ -15,7 +14,6 @@ go_library( "uint64_arch_arm64_asm.s", "uint64_arch_generic.go", ], - importpath = "gvisor.dev/gvisor/pkg/bits", visibility = ["//:sandbox"], ) @@ -53,5 +51,5 @@ go_test( name = "bits_test", size = "small", srcs = ["uint64_test.go"], - embed = [":bits"], + library = ":bits", ) diff --git a/pkg/bpf/BUILD b/pkg/bpf/BUILD index fba5643e8..2a6977f85 100644 --- a/pkg/bpf/BUILD +++ b/pkg/bpf/BUILD @@ -1,5 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -12,7 +11,6 @@ go_library( "interpreter.go", "program_builder.go", ], - importpath = "gvisor.dev/gvisor/pkg/bpf", visibility = ["//visibility:public"], deps = ["//pkg/abi/linux"], ) @@ -25,7 +23,7 @@ go_test( "interpreter_test.go", "program_builder_test.go", ], - embed = [":bpf"], + library = ":bpf", deps = [ "//pkg/abi/linux", "//pkg/binary", diff --git a/pkg/compressio/BUILD b/pkg/compressio/BUILD index 2bb581b18..1f75319a7 100644 --- a/pkg/compressio/BUILD +++ b/pkg/compressio/BUILD @@ -1,12 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) go_library( name = "compressio", srcs = ["compressio.go"], - importpath = "gvisor.dev/gvisor/pkg/compressio", visibility = ["//:sandbox"], deps = [ "//pkg/binary", @@ -18,5 +16,5 @@ go_test( name = "compressio_test", size = "medium", srcs = ["compressio_test.go"], - embed = [":compressio"], + library = ":compressio", ) diff --git a/pkg/control/client/BUILD b/pkg/control/client/BUILD index 066d7b1a1..1b9e10ee7 100644 --- a/pkg/control/client/BUILD +++ b/pkg/control/client/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -7,7 +7,6 @@ go_library( srcs = [ "client.go", ], - importpath = "gvisor.dev/gvisor/pkg/control/client", visibility = ["//:sandbox"], deps = [ "//pkg/unet", diff --git a/pkg/control/server/BUILD b/pkg/control/server/BUILD index adbd1e3f8..002d2ef44 100644 --- a/pkg/control/server/BUILD +++ b/pkg/control/server/BUILD @@ -1,11 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) go_library( name = "server", srcs = ["server.go"], - importpath = "gvisor.dev/gvisor/pkg/control/server", visibility = ["//:sandbox"], deps = [ "//pkg/log", diff --git a/pkg/cpuid/BUILD b/pkg/cpuid/BUILD index ed111fd2a..43a432190 100644 --- a/pkg/cpuid/BUILD +++ b/pkg/cpuid/BUILD @@ -1,5 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -9,7 +8,6 @@ go_library( "cpu_amd64.s", "cpuid.go", ], - importpath = "gvisor.dev/gvisor/pkg/cpuid", visibility = ["//:sandbox"], deps = ["//pkg/log"], ) @@ -18,7 +16,7 @@ go_test( name = "cpuid_test", size = "small", srcs = ["cpuid_test.go"], - embed = [":cpuid"], + library = ":cpuid", ) go_test( @@ -27,6 +25,6 @@ go_test( srcs = [ "cpuid_parse_test.go", ], - embed = [":cpuid"], + library = ":cpuid", tags = ["manual"], ) diff --git a/pkg/eventchannel/BUILD b/pkg/eventchannel/BUILD index 9d68682c7..bee28b68d 100644 --- a/pkg/eventchannel/BUILD +++ b/pkg/eventchannel/BUILD @@ -1,6 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//proto:def.bzl", "go_proto_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test", "proto_library") package(licenses = ["notice"]) @@ -10,7 +8,6 @@ go_library( "event.go", "rate.go", ], - importpath = "gvisor.dev/gvisor/pkg/eventchannel", visibility = ["//:sandbox"], deps = [ ":eventchannel_go_proto", @@ -24,22 +21,15 @@ go_library( ) proto_library( - name = "eventchannel_proto", + name = "eventchannel", srcs = ["event.proto"], visibility = ["//:sandbox"], ) -go_proto_library( - name = "eventchannel_go_proto", - importpath = "gvisor.dev/gvisor/pkg/eventchannel/eventchannel_go_proto", - proto = ":eventchannel_proto", - visibility = ["//:sandbox"], -) - go_test( name = "eventchannel_test", srcs = ["event_test.go"], - embed = [":eventchannel"], + library = ":eventchannel", deps = [ "//pkg/sync", "@com_github_golang_protobuf//proto:go_default_library", diff --git a/pkg/fd/BUILD b/pkg/fd/BUILD index afa8f7659..872361546 100644 --- a/pkg/fd/BUILD +++ b/pkg/fd/BUILD @@ -1,12 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) go_library( name = "fd", srcs = ["fd.go"], - importpath = "gvisor.dev/gvisor/pkg/fd", visibility = ["//visibility:public"], ) @@ -14,5 +12,5 @@ go_test( name = "fd_test", size = "small", srcs = ["fd_test.go"], - embed = [":fd"], + library = ":fd", ) diff --git a/pkg/fdchannel/BUILD b/pkg/fdchannel/BUILD index b0478c672..d9104ef02 100644 --- a/pkg/fdchannel/BUILD +++ b/pkg/fdchannel/BUILD @@ -1,12 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") -package(licenses = ["notice"]) +licenses(["notice"]) go_library( name = "fdchannel", srcs = ["fdchannel_unsafe.go"], - importpath = "gvisor.dev/gvisor/pkg/fdchannel", visibility = ["//visibility:public"], ) @@ -14,6 +12,6 @@ go_test( name = "fdchannel_test", size = "small", srcs = ["fdchannel_test.go"], - embed = [":fdchannel"], + library = ":fdchannel", deps = ["//pkg/sync"], ) diff --git a/pkg/fdnotifier/BUILD b/pkg/fdnotifier/BUILD index 91a202a30..235dcc490 100644 --- a/pkg/fdnotifier/BUILD +++ b/pkg/fdnotifier/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -8,7 +8,6 @@ go_library( "fdnotifier.go", "poll_unsafe.go", ], - importpath = "gvisor.dev/gvisor/pkg/fdnotifier", visibility = ["//:sandbox"], deps = [ "//pkg/sync", diff --git a/pkg/flipcall/BUILD b/pkg/flipcall/BUILD index 85bd83af1..9c5ad500b 100644 --- a/pkg/flipcall/BUILD +++ b/pkg/flipcall/BUILD @@ -1,7 +1,6 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") -package(licenses = ["notice"]) +licenses(["notice"]) go_library( name = "flipcall", @@ -13,7 +12,6 @@ go_library( "io.go", "packet_window_allocator.go", ], - importpath = "gvisor.dev/gvisor/pkg/flipcall", visibility = ["//visibility:public"], deps = [ "//pkg/abi/linux", @@ -30,6 +28,6 @@ go_test( "flipcall_example_test.go", "flipcall_test.go", ], - embed = [":flipcall"], + library = ":flipcall", deps = ["//pkg/sync"], ) diff --git a/pkg/fspath/BUILD b/pkg/fspath/BUILD index ca540363c..ee84471b2 100644 --- a/pkg/fspath/BUILD +++ b/pkg/fspath/BUILD @@ -1,10 +1,8 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") -package( - default_visibility = ["//visibility:public"], - licenses = ["notice"], -) +package(default_visibility = ["//visibility:public"]) + +licenses(["notice"]) go_library( name = "fspath", @@ -13,7 +11,6 @@ go_library( "builder_unsafe.go", "fspath.go", ], - importpath = "gvisor.dev/gvisor/pkg/fspath", ) go_test( @@ -23,5 +20,5 @@ go_test( "builder_test.go", "fspath_test.go", ], - embed = [":fspath"], + library = ":fspath", ) diff --git a/pkg/gate/BUILD b/pkg/gate/BUILD index f22bd070d..dd3141143 100644 --- a/pkg/gate/BUILD +++ b/pkg/gate/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -8,7 +7,6 @@ go_library( srcs = [ "gate.go", ], - importpath = "gvisor.dev/gvisor/pkg/gate", visibility = ["//visibility:public"], ) diff --git a/pkg/goid/BUILD b/pkg/goid/BUILD index 5d31e5366..ea8d2422c 100644 --- a/pkg/goid/BUILD +++ b/pkg/goid/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -11,7 +10,6 @@ go_library( "goid_race.go", "goid_unsafe.go", ], - importpath = "gvisor.dev/gvisor/pkg/goid", visibility = ["//visibility:public"], ) @@ -22,5 +20,5 @@ go_test( "empty_test.go", "goid_test.go", ], - embed = [":goid"], + library = ":goid", ) diff --git a/pkg/ilist/BUILD b/pkg/ilist/BUILD index 34d2673ef..3f6eb07df 100644 --- a/pkg/ilist/BUILD +++ b/pkg/ilist/BUILD @@ -1,6 +1,5 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -9,7 +8,6 @@ go_library( srcs = [ "interface_list.go", ], - importpath = "gvisor.dev/gvisor/pkg/ilist", visibility = ["//visibility:public"], ) @@ -41,7 +39,7 @@ go_test( "list_test.go", "test_list.go", ], - embed = [":ilist"], + library = ":ilist", ) go_template( diff --git a/pkg/linewriter/BUILD b/pkg/linewriter/BUILD index bcde6d308..41bf104d0 100644 --- a/pkg/linewriter/BUILD +++ b/pkg/linewriter/BUILD @@ -1,12 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) go_library( name = "linewriter", srcs = ["linewriter.go"], - importpath = "gvisor.dev/gvisor/pkg/linewriter", visibility = ["//visibility:public"], deps = ["//pkg/sync"], ) @@ -14,5 +12,5 @@ go_library( go_test( name = "linewriter_test", srcs = ["linewriter_test.go"], - embed = [":linewriter"], + library = ":linewriter", ) diff --git a/pkg/log/BUILD b/pkg/log/BUILD index 0df0f2849..935d06963 100644 --- a/pkg/log/BUILD +++ b/pkg/log/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -12,7 +11,6 @@ go_library( "json_k8s.go", "log.go", ], - importpath = "gvisor.dev/gvisor/pkg/log", visibility = [ "//visibility:public", ], @@ -29,5 +27,5 @@ go_test( "json_test.go", "log_test.go", ], - embed = [":log"], + library = ":log", ) diff --git a/pkg/memutil/BUILD b/pkg/memutil/BUILD index 7b50e2b28..9d07d98b4 100644 --- a/pkg/memutil/BUILD +++ b/pkg/memutil/BUILD @@ -1,11 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) go_library( name = "memutil", srcs = ["memutil_unsafe.go"], - importpath = "gvisor.dev/gvisor/pkg/memutil", visibility = ["//visibility:public"], deps = ["@org_golang_x_sys//unix:go_default_library"], ) diff --git a/pkg/metric/BUILD b/pkg/metric/BUILD index 9145f3233..58305009d 100644 --- a/pkg/metric/BUILD +++ b/pkg/metric/BUILD @@ -1,14 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//proto:def.bzl", "go_proto_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("@rules_cc//cc:defs.bzl", "cc_proto_library") +load("//tools:defs.bzl", "go_library", "go_test", "proto_library") package(licenses = ["notice"]) go_library( name = "metric", srcs = ["metric.go"], - importpath = "gvisor.dev/gvisor/pkg/metric", visibility = ["//:sandbox"], deps = [ ":metric_go_proto", @@ -19,28 +15,15 @@ go_library( ) proto_library( - name = "metric_proto", + name = "metric", srcs = ["metric.proto"], visibility = ["//:sandbox"], ) -cc_proto_library( - name = "metric_cc_proto", - visibility = ["//:sandbox"], - deps = [":metric_proto"], -) - -go_proto_library( - name = "metric_go_proto", - importpath = "gvisor.dev/gvisor/pkg/metric/metric_go_proto", - proto = ":metric_proto", - visibility = ["//:sandbox"], -) - go_test( name = "metric_test", srcs = ["metric_test.go"], - embed = [":metric"], + library = ":metric", deps = [ ":metric_go_proto", "//pkg/eventchannel", diff --git a/pkg/p9/BUILD b/pkg/p9/BUILD index a3e05c96d..4ccc1de86 100644 --- a/pkg/p9/BUILD +++ b/pkg/p9/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package( default_visibility = ["//visibility:public"], @@ -23,7 +22,6 @@ go_library( "transport_flipcall.go", "version.go", ], - importpath = "gvisor.dev/gvisor/pkg/p9", deps = [ "//pkg/fd", "//pkg/fdchannel", @@ -47,7 +45,7 @@ go_test( "transport_test.go", "version_test.go", ], - embed = [":p9"], + library = ":p9", deps = [ "//pkg/fd", "//pkg/unet", diff --git a/pkg/p9/p9test/BUILD b/pkg/p9/p9test/BUILD index f4edd68b2..7ca67cb19 100644 --- a/pkg/p9/p9test/BUILD +++ b/pkg/p9/p9test/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_binary", "go_test") +load("//tools:defs.bzl", "go_binary", "go_library", "go_test") package(licenses = ["notice"]) @@ -64,7 +63,6 @@ go_library( "mocks.go", "p9test.go", ], - importpath = "gvisor.dev/gvisor/pkg/p9/p9test", visibility = ["//:sandbox"], deps = [ "//pkg/fd", @@ -80,7 +78,7 @@ go_test( name = "client_test", size = "medium", srcs = ["client_test.go"], - embed = [":p9test"], + library = ":p9test", deps = [ "//pkg/fd", "//pkg/p9", diff --git a/pkg/procid/BUILD b/pkg/procid/BUILD index b506813f0..aa3e3ac0b 100644 --- a/pkg/procid/BUILD +++ b/pkg/procid/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -10,7 +9,6 @@ go_library( "procid_amd64.s", "procid_arm64.s", ], - importpath = "gvisor.dev/gvisor/pkg/procid", visibility = ["//visibility:public"], ) @@ -20,7 +18,7 @@ go_test( srcs = [ "procid_test.go", ], - embed = [":procid"], + library = ":procid", deps = ["//pkg/sync"], ) @@ -31,6 +29,6 @@ go_test( "procid_net_test.go", "procid_test.go", ], - embed = [":procid"], + library = ":procid", deps = ["//pkg/sync"], ) diff --git a/pkg/rand/BUILD b/pkg/rand/BUILD index 9d5b4859b..80b8ceb02 100644 --- a/pkg/rand/BUILD +++ b/pkg/rand/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -8,7 +8,6 @@ go_library( "rand.go", "rand_linux.go", ], - importpath = "gvisor.dev/gvisor/pkg/rand", visibility = ["//:sandbox"], deps = [ "//pkg/sync", diff --git a/pkg/refs/BUILD b/pkg/refs/BUILD index 974d9af9b..74affc887 100644 --- a/pkg/refs/BUILD +++ b/pkg/refs/BUILD @@ -1,6 +1,5 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -23,7 +22,6 @@ go_library( "refcounter_state.go", "weak_ref_list.go", ], - importpath = "gvisor.dev/gvisor/pkg/refs", visibility = ["//:sandbox"], deps = [ "//pkg/log", @@ -35,6 +33,6 @@ go_test( name = "refs_test", size = "small", srcs = ["refcounter_test.go"], - embed = [":refs"], + library = ":refs", deps = ["//pkg/sync"], ) diff --git a/pkg/seccomp/BUILD b/pkg/seccomp/BUILD index af94e944d..742c8b79b 100644 --- a/pkg/seccomp/BUILD +++ b/pkg/seccomp/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_binary", "go_embed_data", "go_test") +load("//tools:defs.bzl", "go_binary", "go_embed_data", "go_library", "go_test") package(licenses = ["notice"]) @@ -27,7 +26,6 @@ go_library( "seccomp_rules.go", "seccomp_unsafe.go", ], - importpath = "gvisor.dev/gvisor/pkg/seccomp", visibility = ["//visibility:public"], deps = [ "//pkg/abi/linux", @@ -43,7 +41,7 @@ go_test( "seccomp_test.go", ":victim_data", ], - embed = [":seccomp"], + library = ":seccomp", deps = [ "//pkg/abi/linux", "//pkg/binary", diff --git a/pkg/secio/BUILD b/pkg/secio/BUILD index 22abdc69f..60f63c7a6 100644 --- a/pkg/secio/BUILD +++ b/pkg/secio/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -9,7 +8,6 @@ go_library( "full_reader.go", "secio.go", ], - importpath = "gvisor.dev/gvisor/pkg/secio", visibility = ["//pkg/sentry:internal"], ) @@ -17,5 +15,5 @@ go_test( name = "secio_test", size = "small", srcs = ["secio_test.go"], - embed = [":secio"], + library = ":secio", ) diff --git a/pkg/segment/test/BUILD b/pkg/segment/test/BUILD index a27c35e21..f2d8462d8 100644 --- a/pkg/segment/test/BUILD +++ b/pkg/segment/test/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") package( @@ -38,7 +37,6 @@ go_library( "int_set.go", "set_functions.go", ], - importpath = "gvisor.dev/gvisor/pkg/segment/segment", deps = [ "//pkg/state", ], @@ -48,5 +46,5 @@ go_test( name = "segment_test", size = "small", srcs = ["segment_test.go"], - embed = [":segment"], + library = ":segment", ) diff --git a/pkg/sentry/BUILD b/pkg/sentry/BUILD index 2d6379c86..e8b794179 100644 --- a/pkg/sentry/BUILD +++ b/pkg/sentry/BUILD @@ -6,6 +6,8 @@ package(licenses = ["notice"]) package_group( name = "internal", packages = [ + "//cloud/gvisor/gopkg/sentry/...", + "//cloud/gvisor/sentry/...", "//pkg/sentry/...", "//runsc/...", # Code generated by go_marshal relies on go_marshal libraries. diff --git a/pkg/sentry/arch/BUILD b/pkg/sentry/arch/BUILD index 65f22af2b..51ca09b24 100644 --- a/pkg/sentry/arch/BUILD +++ b/pkg/sentry/arch/BUILD @@ -1,6 +1,4 @@ -load("@io_bazel_rules_go//proto:def.bzl", "go_proto_library") -load("@rules_cc//cc:defs.bzl", "cc_proto_library") -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "proto_library") package(licenses = ["notice"]) @@ -27,7 +25,6 @@ go_library( "syscalls_amd64.go", "syscalls_arm64.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/arch", visibility = ["//:sandbox"], deps = [ ":registers_go_proto", @@ -44,20 +41,7 @@ go_library( ) proto_library( - name = "registers_proto", + name = "registers", srcs = ["registers.proto"], visibility = ["//visibility:public"], ) - -cc_proto_library( - name = "registers_cc_proto", - visibility = ["//visibility:public"], - deps = [":registers_proto"], -) - -go_proto_library( - name = "registers_go_proto", - importpath = "gvisor.dev/gvisor/pkg/sentry/arch/registers_go_proto", - proto = ":registers_proto", - visibility = ["//visibility:public"], -) diff --git a/pkg/sentry/context/BUILD b/pkg/sentry/context/BUILD index 8dc1a77b1..e13a9ce20 100644 --- a/pkg/sentry/context/BUILD +++ b/pkg/sentry/context/BUILD @@ -1,11 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) go_library( name = "context", srcs = ["context.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/context", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/amutex", diff --git a/pkg/sentry/context/contexttest/BUILD b/pkg/sentry/context/contexttest/BUILD index 581e7aa96..f91a6d4ed 100644 --- a/pkg/sentry/context/contexttest/BUILD +++ b/pkg/sentry/context/contexttest/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -6,7 +6,6 @@ go_library( name = "contexttest", testonly = 1, srcs = ["contexttest.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/context/contexttest", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/memutil", diff --git a/pkg/sentry/control/BUILD b/pkg/sentry/control/BUILD index 2561a6109..e69496477 100644 --- a/pkg/sentry/control/BUILD +++ b/pkg/sentry/control/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -12,9 +11,8 @@ go_library( "proc.go", "state.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/control", visibility = [ - "//pkg/sentry:internal", + "//:sandbox", ], deps = [ "//pkg/abi/linux", @@ -40,7 +38,7 @@ go_test( name = "control_test", size = "small", srcs = ["proc_test.go"], - embed = [":control"], + library = ":control", deps = [ "//pkg/log", "//pkg/sentry/kernel/time", diff --git a/pkg/sentry/device/BUILD b/pkg/sentry/device/BUILD index 97fa1512c..e403cbd8b 100644 --- a/pkg/sentry/device/BUILD +++ b/pkg/sentry/device/BUILD @@ -1,12 +1,10 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) go_library( name = "device", srcs = ["device.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/device", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", @@ -18,5 +16,5 @@ go_test( name = "device_test", size = "small", srcs = ["device_test.go"], - embed = [":device"], + library = ":device", ) diff --git a/pkg/sentry/fs/BUILD b/pkg/sentry/fs/BUILD index 7d5d72d5a..605d61dbe 100644 --- a/pkg/sentry/fs/BUILD +++ b/pkg/sentry/fs/BUILD @@ -1,6 +1,5 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -44,7 +43,6 @@ go_library( "splice.go", "sync.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", @@ -129,7 +127,7 @@ go_test( "mount_test.go", "path_test.go", ], - embed = [":fs"], + library = ":fs", deps = [ "//pkg/sentry/context", "//pkg/sentry/context/contexttest", diff --git a/pkg/sentry/fs/anon/BUILD b/pkg/sentry/fs/anon/BUILD index ae1c9cf76..c14e5405e 100644 --- a/pkg/sentry/fs/anon/BUILD +++ b/pkg/sentry/fs/anon/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -8,7 +8,6 @@ go_library( "anon.go", "device.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/anon", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", diff --git a/pkg/sentry/fs/dev/BUILD b/pkg/sentry/fs/dev/BUILD index a0d9e8496..0c7247bd7 100644 --- a/pkg/sentry/fs/dev/BUILD +++ b/pkg/sentry/fs/dev/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -13,7 +13,6 @@ go_library( "random.go", "tty.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/dev", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", diff --git a/pkg/sentry/fs/fdpipe/BUILD b/pkg/sentry/fs/fdpipe/BUILD index cc43de69d..25ef96299 100644 --- a/pkg/sentry/fs/fdpipe/BUILD +++ b/pkg/sentry/fs/fdpipe/BUILD @@ -1,5 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -10,7 +9,6 @@ go_library( "pipe_opener.go", "pipe_state.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/fdpipe", imports = ["gvisor.dev/gvisor/pkg/sentry/fs"], visibility = ["//pkg/sentry:internal"], deps = [ @@ -36,7 +34,7 @@ go_test( "pipe_opener_test.go", "pipe_test.go", ], - embed = [":fdpipe"], + library = ":fdpipe", deps = [ "//pkg/fd", "//pkg/fdnotifier", diff --git a/pkg/sentry/fs/filetest/BUILD b/pkg/sentry/fs/filetest/BUILD index 358dc2be3..9a7608cae 100644 --- a/pkg/sentry/fs/filetest/BUILD +++ b/pkg/sentry/fs/filetest/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -6,7 +6,6 @@ go_library( name = "filetest", testonly = 1, srcs = ["filetest.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/filetest", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/sentry/context", diff --git a/pkg/sentry/fs/fsutil/BUILD b/pkg/sentry/fs/fsutil/BUILD index 945b6270d..9142f5bdf 100644 --- a/pkg/sentry/fs/fsutil/BUILD +++ b/pkg/sentry/fs/fsutil/BUILD @@ -1,6 +1,5 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -75,7 +74,6 @@ go_library( "inode.go", "inode_cached.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/fsutil", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", @@ -106,7 +104,7 @@ go_test( "dirty_set_test.go", "inode_cached_test.go", ], - embed = [":fsutil"], + library = ":fsutil", deps = [ "//pkg/sentry/context", "//pkg/sentry/context/contexttest", diff --git a/pkg/sentry/fs/gofer/BUILD b/pkg/sentry/fs/gofer/BUILD index fd870e8e1..cf48e7c03 100644 --- a/pkg/sentry/fs/gofer/BUILD +++ b/pkg/sentry/fs/gofer/BUILD @@ -1,5 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -22,7 +21,6 @@ go_library( "socket.go", "util.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/gofer", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", @@ -56,7 +54,7 @@ go_test( name = "gofer_test", size = "small", srcs = ["gofer_test.go"], - embed = [":gofer"], + library = ":gofer", deps = [ "//pkg/p9", "//pkg/p9/p9test", diff --git a/pkg/sentry/fs/host/BUILD b/pkg/sentry/fs/host/BUILD index 2b581aa69..f586f47c1 100644 --- a/pkg/sentry/fs/host/BUILD +++ b/pkg/sentry/fs/host/BUILD @@ -1,5 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -25,7 +24,6 @@ go_library( "util_arm64_unsafe.go", "util_unsafe.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/host", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", @@ -69,7 +67,7 @@ go_test( "socket_test.go", "wait_test.go", ], - embed = [":host"], + library = ":host", deps = [ "//pkg/fd", "//pkg/fdnotifier", diff --git a/pkg/sentry/fs/lock/BUILD b/pkg/sentry/fs/lock/BUILD index 2c332a82a..ae3331737 100644 --- a/pkg/sentry/fs/lock/BUILD +++ b/pkg/sentry/fs/lock/BUILD @@ -1,6 +1,5 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -40,7 +39,6 @@ go_library( "lock_set.go", "lock_set_functions.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/lock", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/log", @@ -56,5 +54,5 @@ go_test( "lock_range_test.go", "lock_test.go", ], - embed = [":lock"], + library = ":lock", ) diff --git a/pkg/sentry/fs/proc/BUILD b/pkg/sentry/fs/proc/BUILD index cb37c6c6b..b06bead41 100644 --- a/pkg/sentry/fs/proc/BUILD +++ b/pkg/sentry/fs/proc/BUILD @@ -1,5 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -27,7 +26,6 @@ go_library( "uptime.go", "version.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/proc", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", @@ -63,7 +61,7 @@ go_test( "net_test.go", "sys_net_test.go", ], - embed = [":proc"], + library = ":proc", deps = [ "//pkg/abi/linux", "//pkg/sentry/context", diff --git a/pkg/sentry/fs/proc/device/BUILD b/pkg/sentry/fs/proc/device/BUILD index 0394451d4..52c9aa93d 100644 --- a/pkg/sentry/fs/proc/device/BUILD +++ b/pkg/sentry/fs/proc/device/BUILD @@ -1,11 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) go_library( name = "device", srcs = ["device.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/proc/device", visibility = ["//pkg/sentry:internal"], deps = ["//pkg/sentry/device"], ) diff --git a/pkg/sentry/fs/proc/seqfile/BUILD b/pkg/sentry/fs/proc/seqfile/BUILD index 38b246dff..310d8dd52 100644 --- a/pkg/sentry/fs/proc/seqfile/BUILD +++ b/pkg/sentry/fs/proc/seqfile/BUILD @@ -1,12 +1,10 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) go_library( name = "seqfile", srcs = ["seqfile.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/proc/seqfile", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", @@ -26,7 +24,7 @@ go_test( name = "seqfile_test", size = "small", srcs = ["seqfile_test.go"], - embed = [":seqfile"], + library = ":seqfile", deps = [ "//pkg/sentry/context", "//pkg/sentry/context/contexttest", diff --git a/pkg/sentry/fs/ramfs/BUILD b/pkg/sentry/fs/ramfs/BUILD index 3fb7b0633..39c4b84f8 100644 --- a/pkg/sentry/fs/ramfs/BUILD +++ b/pkg/sentry/fs/ramfs/BUILD @@ -1,5 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -11,7 +10,6 @@ go_library( "symlink.go", "tree.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/ramfs", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", @@ -31,7 +29,7 @@ go_test( name = "ramfs_test", size = "small", srcs = ["tree_test.go"], - embed = [":ramfs"], + library = ":ramfs", deps = [ "//pkg/sentry/context/contexttest", "//pkg/sentry/fs", diff --git a/pkg/sentry/fs/sys/BUILD b/pkg/sentry/fs/sys/BUILD index 25f0f124e..cc6b3bfbf 100644 --- a/pkg/sentry/fs/sys/BUILD +++ b/pkg/sentry/fs/sys/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -10,7 +10,6 @@ go_library( "fs.go", "sys.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/sys", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", diff --git a/pkg/sentry/fs/timerfd/BUILD b/pkg/sentry/fs/timerfd/BUILD index a215c1b95..092668e8d 100644 --- a/pkg/sentry/fs/timerfd/BUILD +++ b/pkg/sentry/fs/timerfd/BUILD @@ -1,11 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) go_library( name = "timerfd", srcs = ["timerfd.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/timerfd", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/sentry/context", diff --git a/pkg/sentry/fs/tmpfs/BUILD b/pkg/sentry/fs/tmpfs/BUILD index 3400b940c..04776555f 100644 --- a/pkg/sentry/fs/tmpfs/BUILD +++ b/pkg/sentry/fs/tmpfs/BUILD @@ -1,5 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -12,7 +11,6 @@ go_library( "inode_file.go", "tmpfs.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/tmpfs", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", @@ -41,7 +39,7 @@ go_test( name = "tmpfs_test", size = "small", srcs = ["file_test.go"], - embed = [":tmpfs"], + library = ":tmpfs", deps = [ "//pkg/sentry/context", "//pkg/sentry/fs", diff --git a/pkg/sentry/fs/tty/BUILD b/pkg/sentry/fs/tty/BUILD index f6f60d0cf..29f804c6c 100644 --- a/pkg/sentry/fs/tty/BUILD +++ b/pkg/sentry/fs/tty/BUILD @@ -1,5 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -14,7 +13,6 @@ go_library( "slave.go", "terminal.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/tty", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", @@ -40,7 +38,7 @@ go_test( name = "tty_test", size = "small", srcs = ["tty_test.go"], - embed = [":tty"], + library = ":tty", deps = [ "//pkg/abi/linux", "//pkg/sentry/context/contexttest", diff --git a/pkg/sentry/fsimpl/ext/BUILD b/pkg/sentry/fsimpl/ext/BUILD index 903874141..a718920d5 100644 --- a/pkg/sentry/fsimpl/ext/BUILD +++ b/pkg/sentry/fsimpl/ext/BUILD @@ -1,5 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") package(licenses = ["notice"]) @@ -32,7 +31,6 @@ go_library( "symlink.go", "utils.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", @@ -71,7 +69,7 @@ go_test( "//pkg/sentry/fsimpl/ext:assets/tiny.ext3", "//pkg/sentry/fsimpl/ext:assets/tiny.ext4", ], - embed = [":ext"], + library = ":ext", deps = [ "//pkg/abi/linux", "//pkg/binary", diff --git a/pkg/sentry/fsimpl/ext/benchmark/BUILD b/pkg/sentry/fsimpl/ext/benchmark/BUILD index 4fc8296ef..12f3990c1 100644 --- a/pkg/sentry/fsimpl/ext/benchmark/BUILD +++ b/pkg/sentry/fsimpl/ext/benchmark/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_test") package(licenses = ["notice"]) diff --git a/pkg/sentry/fsimpl/ext/disklayout/BUILD b/pkg/sentry/fsimpl/ext/disklayout/BUILD index fcfaf5c3e..9bd9c76c0 100644 --- a/pkg/sentry/fsimpl/ext/disklayout/BUILD +++ b/pkg/sentry/fsimpl/ext/disklayout/BUILD @@ -1,5 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -23,7 +22,6 @@ go_library( "superblock_old.go", "test_utils.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", @@ -44,6 +42,6 @@ go_test( "inode_test.go", "superblock_test.go", ], - embed = [":disklayout"], + library = ":disklayout", deps = ["//pkg/sentry/kernel/time"], ) diff --git a/pkg/sentry/fsimpl/kernfs/BUILD b/pkg/sentry/fsimpl/kernfs/BUILD index 66d409785..7bf83ccba 100644 --- a/pkg/sentry/fsimpl/kernfs/BUILD +++ b/pkg/sentry/fsimpl/kernfs/BUILD @@ -1,8 +1,7 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") -package(licenses = ["notice"]) +licenses(["notice"]) go_template_instance( name = "slot_list", @@ -27,7 +26,6 @@ go_library( "slot_list.go", "symlink.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", diff --git a/pkg/sentry/fsimpl/proc/BUILD b/pkg/sentry/fsimpl/proc/BUILD index c5b79fb38..3768f55b2 100644 --- a/pkg/sentry/fsimpl/proc/BUILD +++ b/pkg/sentry/fsimpl/proc/BUILD @@ -1,7 +1,6 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") -package(licenses = ["notice"]) +licenses(["notice"]) go_library( name = "proc", @@ -15,7 +14,6 @@ go_library( "tasks_net.go", "tasks_sys.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fsimpl/proc", deps = [ "//pkg/abi/linux", "//pkg/log", @@ -47,7 +45,7 @@ go_test( "tasks_sys_test.go", "tasks_test.go", ], - embed = [":proc"], + library = ":proc", deps = [ "//pkg/abi/linux", "//pkg/fspath", diff --git a/pkg/sentry/fsimpl/sys/BUILD b/pkg/sentry/fsimpl/sys/BUILD index ee3c842bd..beda141f1 100644 --- a/pkg/sentry/fsimpl/sys/BUILD +++ b/pkg/sentry/fsimpl/sys/BUILD @@ -1,14 +1,12 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") -package(licenses = ["notice"]) +licenses(["notice"]) go_library( name = "sys", srcs = [ "sys.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fsimpl/sys", deps = [ "//pkg/abi/linux", "//pkg/sentry/context", diff --git a/pkg/sentry/fsimpl/testutil/BUILD b/pkg/sentry/fsimpl/testutil/BUILD index 4e70d84a7..12053a5b6 100644 --- a/pkg/sentry/fsimpl/testutil/BUILD +++ b/pkg/sentry/fsimpl/testutil/BUILD @@ -1,6 +1,6 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") -package(licenses = ["notice"]) +licenses(["notice"]) go_library( name = "testutil", @@ -9,7 +9,6 @@ go_library( "kernel.go", "testutil.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fsimpl/testutil", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD index 691476b4f..857e98bc5 100644 --- a/pkg/sentry/fsimpl/tmpfs/BUILD +++ b/pkg/sentry/fsimpl/tmpfs/BUILD @@ -1,8 +1,7 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") -package(licenses = ["notice"]) +licenses(["notice"]) go_template_instance( name = "dentry_list", @@ -28,7 +27,6 @@ go_library( "symlink.go", "tmpfs.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs", deps = [ "//pkg/abi/linux", "//pkg/amutex", @@ -81,7 +79,7 @@ go_test( "regular_file_test.go", "stat_test.go", ], - embed = [":tmpfs"], + library = ":tmpfs", deps = [ "//pkg/abi/linux", "//pkg/fspath", diff --git a/pkg/sentry/hostcpu/BUILD b/pkg/sentry/hostcpu/BUILD index 359468ccc..e6933aa70 100644 --- a/pkg/sentry/hostcpu/BUILD +++ b/pkg/sentry/hostcpu/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -10,7 +9,6 @@ go_library( "getcpu_arm64.s", "hostcpu.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/hostcpu", visibility = ["//:sandbox"], ) @@ -18,5 +16,5 @@ go_test( name = "hostcpu_test", size = "small", srcs = ["hostcpu_test.go"], - embed = [":hostcpu"], + library = ":hostcpu", ) diff --git a/pkg/sentry/hostmm/BUILD b/pkg/sentry/hostmm/BUILD index 67831d5a1..a145a5ca3 100644 --- a/pkg/sentry/hostmm/BUILD +++ b/pkg/sentry/hostmm/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -8,7 +8,6 @@ go_library( "cgroup.go", "hostmm.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/hostmm", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/fd", diff --git a/pkg/sentry/inet/BUILD b/pkg/sentry/inet/BUILD index 8d60ad4ad..aa621b724 100644 --- a/pkg/sentry/inet/BUILD +++ b/pkg/sentry/inet/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package( default_visibility = ["//:sandbox"], @@ -12,7 +12,6 @@ go_library( "inet.go", "test_stack.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/inet", deps = [ "//pkg/sentry/context", "//pkg/tcpip/stack", diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD index ac85ba0c8..cebaccd92 100644 --- a/pkg/sentry/kernel/BUILD +++ b/pkg/sentry/kernel/BUILD @@ -1,8 +1,5 @@ -load("@io_bazel_rules_go//proto:def.bzl", "go_proto_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("@rules_cc//cc:defs.bzl", "cc_proto_library") +load("//tools:defs.bzl", "go_library", "go_test", "proto_library") load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -78,26 +75,12 @@ go_template_instance( ) proto_library( - name = "uncaught_signal_proto", + name = "uncaught_signal", srcs = ["uncaught_signal.proto"], visibility = ["//visibility:public"], deps = ["//pkg/sentry/arch:registers_proto"], ) -cc_proto_library( - name = "uncaught_signal_cc_proto", - visibility = ["//visibility:public"], - deps = [":uncaught_signal_proto"], -) - -go_proto_library( - name = "uncaught_signal_go_proto", - importpath = "gvisor.dev/gvisor/pkg/sentry/kernel/uncaught_signal_go_proto", - proto = ":uncaught_signal_proto", - visibility = ["//visibility:public"], - deps = ["//pkg/sentry/arch:registers_go_proto"], -) - go_library( name = "kernel", srcs = [ @@ -156,7 +139,6 @@ go_library( "vdso.go", "version.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/kernel", imports = [ "gvisor.dev/gvisor/pkg/bpf", "gvisor.dev/gvisor/pkg/sentry/device", @@ -227,7 +209,7 @@ go_test( "task_test.go", "timekeeper_test.go", ], - embed = [":kernel"], + library = ":kernel", deps = [ "//pkg/abi", "//pkg/sentry/arch", diff --git a/pkg/sentry/kernel/auth/BUILD b/pkg/sentry/kernel/auth/BUILD index 1aa72fa47..64537c9be 100644 --- a/pkg/sentry/kernel/auth/BUILD +++ b/pkg/sentry/kernel/auth/BUILD @@ -1,5 +1,5 @@ +load("//tools:defs.bzl", "go_library") load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -57,7 +57,6 @@ go_library( "id_map_set.go", "user_namespace.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/kernel/auth", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", diff --git a/pkg/sentry/kernel/contexttest/BUILD b/pkg/sentry/kernel/contexttest/BUILD index 3a88a585c..daff608d7 100644 --- a/pkg/sentry/kernel/contexttest/BUILD +++ b/pkg/sentry/kernel/contexttest/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -6,7 +6,6 @@ go_library( name = "contexttest", testonly = 1, srcs = ["contexttest.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/sentry/context", diff --git a/pkg/sentry/kernel/epoll/BUILD b/pkg/sentry/kernel/epoll/BUILD index c47f6b6fc..19e16ab3a 100644 --- a/pkg/sentry/kernel/epoll/BUILD +++ b/pkg/sentry/kernel/epoll/BUILD @@ -1,6 +1,5 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -23,7 +22,6 @@ go_library( "epoll_list.go", "epoll_state.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/kernel/epoll", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/refs", @@ -43,7 +41,7 @@ go_test( srcs = [ "epoll_test.go", ], - embed = [":epoll"], + library = ":epoll", deps = [ "//pkg/sentry/context/contexttest", "//pkg/sentry/fs/filetest", diff --git a/pkg/sentry/kernel/eventfd/BUILD b/pkg/sentry/kernel/eventfd/BUILD index c831fbab2..ee2d74864 100644 --- a/pkg/sentry/kernel/eventfd/BUILD +++ b/pkg/sentry/kernel/eventfd/BUILD @@ -1,12 +1,10 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) go_library( name = "eventfd", srcs = ["eventfd.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/kernel/eventfd", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", @@ -26,7 +24,7 @@ go_test( name = "eventfd_test", size = "small", srcs = ["eventfd_test.go"], - embed = [":eventfd"], + library = ":eventfd", deps = [ "//pkg/sentry/context/contexttest", "//pkg/sentry/usermem", diff --git a/pkg/sentry/kernel/fasync/BUILD b/pkg/sentry/kernel/fasync/BUILD index 6b36bc63e..b9126e946 100644 --- a/pkg/sentry/kernel/fasync/BUILD +++ b/pkg/sentry/kernel/fasync/BUILD @@ -1,11 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) go_library( name = "fasync", srcs = ["fasync.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/kernel/fasync", visibility = ["//:sandbox"], deps = [ "//pkg/abi/linux", diff --git a/pkg/sentry/kernel/futex/BUILD b/pkg/sentry/kernel/futex/BUILD index 50db443ce..f413d8ae2 100644 --- a/pkg/sentry/kernel/futex/BUILD +++ b/pkg/sentry/kernel/futex/BUILD @@ -1,6 +1,5 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -34,7 +33,6 @@ go_library( "futex.go", "waiter_list.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/kernel/futex", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", @@ -51,7 +49,7 @@ go_test( name = "futex_test", size = "small", srcs = ["futex_test.go"], - embed = [":futex"], + library = ":futex", deps = [ "//pkg/sentry/usermem", "//pkg/sync", diff --git a/pkg/sentry/kernel/memevent/BUILD b/pkg/sentry/kernel/memevent/BUILD index 7f36252a9..4486848d2 100644 --- a/pkg/sentry/kernel/memevent/BUILD +++ b/pkg/sentry/kernel/memevent/BUILD @@ -1,13 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//proto:def.bzl", "go_proto_library") -load("@rules_cc//cc:defs.bzl", "cc_proto_library") +load("//tools:defs.bzl", "go_library", "proto_library") package(licenses = ["notice"]) go_library( name = "memevent", srcs = ["memory_events.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/kernel/memevent", visibility = ["//:sandbox"], deps = [ ":memory_events_go_proto", @@ -21,20 +18,7 @@ go_library( ) proto_library( - name = "memory_events_proto", + name = "memory_events", srcs = ["memory_events.proto"], visibility = ["//visibility:public"], ) - -cc_proto_library( - name = "memory_events_cc_proto", - visibility = ["//visibility:public"], - deps = [":memory_events_proto"], -) - -go_proto_library( - name = "memory_events_go_proto", - importpath = "gvisor.dev/gvisor/pkg/sentry/kernel/memevent/memory_events_go_proto", - proto = ":memory_events_proto", - visibility = ["//visibility:public"], -) diff --git a/pkg/sentry/kernel/pipe/BUILD b/pkg/sentry/kernel/pipe/BUILD index 5eeaeff66..2c7b6206f 100644 --- a/pkg/sentry/kernel/pipe/BUILD +++ b/pkg/sentry/kernel/pipe/BUILD @@ -1,6 +1,5 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -30,7 +29,6 @@ go_library( "vfs.go", "writer.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/kernel/pipe", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", @@ -57,7 +55,7 @@ go_test( "node_test.go", "pipe_test.go", ], - embed = [":pipe"], + library = ":pipe", deps = [ "//pkg/sentry/context", "//pkg/sentry/context/contexttest", diff --git a/pkg/sentry/kernel/sched/BUILD b/pkg/sentry/kernel/sched/BUILD index 98ea7a0d8..1b82e087b 100644 --- a/pkg/sentry/kernel/sched/BUILD +++ b/pkg/sentry/kernel/sched/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -9,7 +8,6 @@ go_library( "cpuset.go", "sched.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/kernel/sched", visibility = ["//pkg/sentry:internal"], ) @@ -17,5 +15,5 @@ go_test( name = "sched_test", size = "small", srcs = ["cpuset_test.go"], - embed = [":sched"], + library = ":sched", ) diff --git a/pkg/sentry/kernel/semaphore/BUILD b/pkg/sentry/kernel/semaphore/BUILD index 13a961594..76e19b551 100644 --- a/pkg/sentry/kernel/semaphore/BUILD +++ b/pkg/sentry/kernel/semaphore/BUILD @@ -1,6 +1,5 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -22,7 +21,6 @@ go_library( "semaphore.go", "waiter_list.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/kernel/semaphore", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", @@ -40,7 +38,7 @@ go_test( name = "semaphore_test", size = "small", srcs = ["semaphore_test.go"], - embed = [":semaphore"], + library = ":semaphore", deps = [ "//pkg/abi/linux", "//pkg/sentry/context", diff --git a/pkg/sentry/kernel/shm/BUILD b/pkg/sentry/kernel/shm/BUILD index 7321b22ed..5547c5abf 100644 --- a/pkg/sentry/kernel/shm/BUILD +++ b/pkg/sentry/kernel/shm/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -8,7 +8,6 @@ go_library( "device.go", "shm.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/kernel/shm", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", diff --git a/pkg/sentry/kernel/signalfd/BUILD b/pkg/sentry/kernel/signalfd/BUILD index 89e4d84b1..5d44773d4 100644 --- a/pkg/sentry/kernel/signalfd/BUILD +++ b/pkg/sentry/kernel/signalfd/BUILD @@ -1,11 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") -package(licenses = ["notice"]) +licenses(["notice"]) go_library( name = "signalfd", srcs = ["signalfd.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/kernel/signalfd", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", diff --git a/pkg/sentry/kernel/time/BUILD b/pkg/sentry/kernel/time/BUILD index 4e4de0512..d49594d9f 100644 --- a/pkg/sentry/kernel/time/BUILD +++ b/pkg/sentry/kernel/time/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -8,7 +8,6 @@ go_library( "context.go", "time.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/kernel/time", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", diff --git a/pkg/sentry/limits/BUILD b/pkg/sentry/limits/BUILD index 9fa841e8b..67869757f 100644 --- a/pkg/sentry/limits/BUILD +++ b/pkg/sentry/limits/BUILD @@ -1,5 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -10,7 +9,6 @@ go_library( "limits.go", "linux.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/limits", visibility = ["//:sandbox"], deps = [ "//pkg/abi/linux", @@ -25,5 +23,5 @@ go_test( srcs = [ "limits_test.go", ], - embed = [":limits"], + library = ":limits", ) diff --git a/pkg/sentry/loader/BUILD b/pkg/sentry/loader/BUILD index 2890393bd..d4ad2bd6c 100644 --- a/pkg/sentry/loader/BUILD +++ b/pkg/sentry/loader/BUILD @@ -1,5 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_embed_data") -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_embed_data", "go_library") package(licenses = ["notice"]) @@ -20,7 +19,6 @@ go_library( "vdso_state.go", ":vdso_bin", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/loader", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi", diff --git a/pkg/sentry/memmap/BUILD b/pkg/sentry/memmap/BUILD index 112794e9c..f9a65f086 100644 --- a/pkg/sentry/memmap/BUILD +++ b/pkg/sentry/memmap/BUILD @@ -1,6 +1,5 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -37,7 +36,6 @@ go_library( "mapping_set_impl.go", "memmap.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/memmap", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/log", @@ -52,6 +50,6 @@ go_test( name = "memmap_test", size = "small", srcs = ["mapping_set_test.go"], - embed = [":memmap"], + library = ":memmap", deps = ["//pkg/sentry/usermem"], ) diff --git a/pkg/sentry/mm/BUILD b/pkg/sentry/mm/BUILD index 83e248431..bd6399fa2 100644 --- a/pkg/sentry/mm/BUILD +++ b/pkg/sentry/mm/BUILD @@ -1,6 +1,5 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -96,7 +95,6 @@ go_library( "vma.go", "vma_set.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/mm", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", @@ -128,7 +126,7 @@ go_test( name = "mm_test", size = "small", srcs = ["mm_test.go"], - embed = [":mm"], + library = ":mm", deps = [ "//pkg/sentry/arch", "//pkg/sentry/context", diff --git a/pkg/sentry/pgalloc/BUILD b/pkg/sentry/pgalloc/BUILD index a9a2642c5..02385a3ce 100644 --- a/pkg/sentry/pgalloc/BUILD +++ b/pkg/sentry/pgalloc/BUILD @@ -1,6 +1,5 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -60,7 +59,6 @@ go_library( "save_restore.go", "usage_set.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/pgalloc", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/log", @@ -82,6 +80,6 @@ go_test( name = "pgalloc_test", size = "small", srcs = ["pgalloc_test.go"], - embed = [":pgalloc"], + library = ":pgalloc", deps = ["//pkg/sentry/usermem"], ) diff --git a/pkg/sentry/platform/BUILD b/pkg/sentry/platform/BUILD index 157bffa81..006450b2d 100644 --- a/pkg/sentry/platform/BUILD +++ b/pkg/sentry/platform/BUILD @@ -1,5 +1,5 @@ +load("//tools:defs.bzl", "go_library") load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -22,7 +22,6 @@ go_library( "mmap_min_addr.go", "platform.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/platform", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", diff --git a/pkg/sentry/platform/interrupt/BUILD b/pkg/sentry/platform/interrupt/BUILD index 85e882df9..83b385f14 100644 --- a/pkg/sentry/platform/interrupt/BUILD +++ b/pkg/sentry/platform/interrupt/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -8,7 +7,6 @@ go_library( srcs = [ "interrupt.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/platform/interrupt", visibility = ["//pkg/sentry:internal"], deps = ["//pkg/sync"], ) @@ -17,5 +15,5 @@ go_test( name = "interrupt_test", size = "small", srcs = ["interrupt_test.go"], - embed = [":interrupt"], + library = ":interrupt", ) diff --git a/pkg/sentry/platform/kvm/BUILD b/pkg/sentry/platform/kvm/BUILD index 6a358d1d4..a4532a766 100644 --- a/pkg/sentry/platform/kvm/BUILD +++ b/pkg/sentry/platform/kvm/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -38,7 +37,6 @@ go_library( "physical_map_arm64.go", "virtual_map.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/platform/kvm", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", @@ -65,7 +63,7 @@ go_test( "kvm_test.go", "virtual_map_test.go", ], - embed = [":kvm"], + library = ":kvm", tags = [ "manual", "nogotsan", diff --git a/pkg/sentry/platform/kvm/testutil/BUILD b/pkg/sentry/platform/kvm/testutil/BUILD index b0e45f159..f7605df8a 100644 --- a/pkg/sentry/platform/kvm/testutil/BUILD +++ b/pkg/sentry/platform/kvm/testutil/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -12,6 +12,5 @@ go_library( "testutil_arm64.go", "testutil_arm64.s", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/platform/kvm/testutil", visibility = ["//pkg/sentry/platform/kvm:__pkg__"], ) diff --git a/pkg/sentry/platform/ptrace/BUILD b/pkg/sentry/platform/ptrace/BUILD index cd13390c3..3bcc5e040 100644 --- a/pkg/sentry/platform/ptrace/BUILD +++ b/pkg/sentry/platform/ptrace/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -20,7 +20,6 @@ go_library( "subprocess_linux_unsafe.go", "subprocess_unsafe.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/platform/ptrace", visibility = ["//:sandbox"], deps = [ "//pkg/abi/linux", diff --git a/pkg/sentry/platform/ring0/BUILD b/pkg/sentry/platform/ring0/BUILD index 87f4552b5..6dee8fcc5 100644 --- a/pkg/sentry/platform/ring0/BUILD +++ b/pkg/sentry/platform/ring0/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") load("//tools/go_generics:defs.bzl", "go_template", "go_template_instance") package(licenses = ["notice"]) @@ -74,7 +74,6 @@ go_library( "lib_arm64.s", "ring0.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/platform/ring0", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/cpuid", diff --git a/pkg/sentry/platform/ring0/gen_offsets/BUILD b/pkg/sentry/platform/ring0/gen_offsets/BUILD index 42076fb04..147311ed3 100644 --- a/pkg/sentry/platform/ring0/gen_offsets/BUILD +++ b/pkg/sentry/platform/ring0/gen_offsets/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_binary") +load("//tools:defs.bzl", "go_binary") load("//tools/go_generics:defs.bzl", "go_template_instance") package(licenses = ["notice"]) diff --git a/pkg/sentry/platform/ring0/pagetables/BUILD b/pkg/sentry/platform/ring0/pagetables/BUILD index 387a7f6c3..8b5cdd6c1 100644 --- a/pkg/sentry/platform/ring0/pagetables/BUILD +++ b/pkg/sentry/platform/ring0/pagetables/BUILD @@ -1,17 +1,14 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test", "select_arch") load("//tools/go_generics:defs.bzl", "go_template", "go_template_instance") package(licenses = ["notice"]) -config_setting( - name = "aarch64", - constraint_values = ["@bazel_tools//platforms:aarch64"], -) - go_template( name = "generic_walker", - srcs = ["walker_amd64.go"], + srcs = select_arch( + amd64 = ["walker_amd64.go"], + arm64 = ["walker_amd64.go"], + ), opt_types = [ "Visitor", ], @@ -91,7 +88,6 @@ go_library( "walker_map.go", "walker_unmap.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables", visibility = [ "//pkg/sentry/platform/kvm:__subpackages__", "//pkg/sentry/platform/ring0:__subpackages__", @@ -111,6 +107,6 @@ go_test( "pagetables_test.go", "walker_check.go", ], - embed = [":pagetables"], + library = ":pagetables", deps = ["//pkg/sentry/usermem"], ) diff --git a/pkg/sentry/platform/safecopy/BUILD b/pkg/sentry/platform/safecopy/BUILD index 6769cd0a5..b8747585b 100644 --- a/pkg/sentry/platform/safecopy/BUILD +++ b/pkg/sentry/platform/safecopy/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -17,7 +16,6 @@ go_library( "sighandler_amd64.s", "sighandler_arm64.s", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/platform/safecopy", visibility = ["//pkg/sentry:internal"], deps = ["//pkg/syserror"], ) @@ -27,5 +25,5 @@ go_test( srcs = [ "safecopy_test.go", ], - embed = [":safecopy"], + library = ":safecopy", ) diff --git a/pkg/sentry/safemem/BUILD b/pkg/sentry/safemem/BUILD index 884020f7b..3ab76da97 100644 --- a/pkg/sentry/safemem/BUILD +++ b/pkg/sentry/safemem/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -11,7 +10,6 @@ go_library( "safemem.go", "seq_unsafe.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/safemem", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/sentry/platform/safecopy", @@ -25,5 +23,5 @@ go_test( "io_test.go", "seq_test.go", ], - embed = [":safemem"], + library = ":safemem", ) diff --git a/pkg/sentry/sighandling/BUILD b/pkg/sentry/sighandling/BUILD index f561670c7..6c38a3f44 100644 --- a/pkg/sentry/sighandling/BUILD +++ b/pkg/sentry/sighandling/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -8,7 +8,6 @@ go_library( "sighandling.go", "sighandling_unsafe.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/sighandling", visibility = ["//pkg/sentry:internal"], deps = ["//pkg/abi/linux"], ) diff --git a/pkg/sentry/socket/BUILD b/pkg/sentry/socket/BUILD index 26176b10d..8e2b97afb 100644 --- a/pkg/sentry/socket/BUILD +++ b/pkg/sentry/socket/BUILD @@ -1,11 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) go_library( name = "socket", srcs = ["socket.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/socket", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", diff --git a/pkg/sentry/socket/control/BUILD b/pkg/sentry/socket/control/BUILD index 357517ed4..3850f6345 100644 --- a/pkg/sentry/socket/control/BUILD +++ b/pkg/sentry/socket/control/BUILD @@ -1,11 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) go_library( name = "control", srcs = ["control.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/socket/control", imports = [ "gvisor.dev/gvisor/pkg/sentry/fs", ], diff --git a/pkg/sentry/socket/hostinet/BUILD b/pkg/sentry/socket/hostinet/BUILD index 4c44c7c0f..42bf7be6a 100644 --- a/pkg/sentry/socket/hostinet/BUILD +++ b/pkg/sentry/socket/hostinet/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -12,7 +12,6 @@ go_library( "socket_unsafe.go", "stack.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/socket/hostinet", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", diff --git a/pkg/sentry/socket/netfilter/BUILD b/pkg/sentry/socket/netfilter/BUILD index b70047d81..ed34a8308 100644 --- a/pkg/sentry/socket/netfilter/BUILD +++ b/pkg/sentry/socket/netfilter/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -7,7 +7,6 @@ go_library( srcs = [ "netfilter.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/socket/netfilter", # This target depends on netstack and should only be used by epsocket, # which is allowed to depend on netstack. visibility = ["//pkg/sentry:internal"], diff --git a/pkg/sentry/socket/netlink/BUILD b/pkg/sentry/socket/netlink/BUILD index 103933144..baaac13c6 100644 --- a/pkg/sentry/socket/netlink/BUILD +++ b/pkg/sentry/socket/netlink/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -9,7 +9,6 @@ go_library( "provider.go", "socket.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/socket/netlink", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", diff --git a/pkg/sentry/socket/netlink/port/BUILD b/pkg/sentry/socket/netlink/port/BUILD index 2d9f4ba9b..3a22923d8 100644 --- a/pkg/sentry/socket/netlink/port/BUILD +++ b/pkg/sentry/socket/netlink/port/BUILD @@ -1,12 +1,10 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) go_library( name = "port", srcs = ["port.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/socket/netlink/port", visibility = ["//pkg/sentry:internal"], deps = ["//pkg/sync"], ) @@ -14,5 +12,5 @@ go_library( go_test( name = "port_test", srcs = ["port_test.go"], - embed = [":port"], + library = ":port", ) diff --git a/pkg/sentry/socket/netlink/route/BUILD b/pkg/sentry/socket/netlink/route/BUILD index 1d4912753..2137c7aeb 100644 --- a/pkg/sentry/socket/netlink/route/BUILD +++ b/pkg/sentry/socket/netlink/route/BUILD @@ -1,11 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) go_library( name = "route", srcs = ["protocol.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/socket/netlink/route", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", diff --git a/pkg/sentry/socket/netlink/uevent/BUILD b/pkg/sentry/socket/netlink/uevent/BUILD index 0777f3baf..73fbdf1eb 100644 --- a/pkg/sentry/socket/netlink/uevent/BUILD +++ b/pkg/sentry/socket/netlink/uevent/BUILD @@ -1,11 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) go_library( name = "uevent", srcs = ["protocol.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/socket/netlink/uevent", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", diff --git a/pkg/sentry/socket/netstack/BUILD b/pkg/sentry/socket/netstack/BUILD index f78784569..e3d1f90cb 100644 --- a/pkg/sentry/socket/netstack/BUILD +++ b/pkg/sentry/socket/netstack/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -11,7 +11,6 @@ go_library( "save_restore.go", "stack.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/socket/netstack", visibility = [ "//pkg/sentry:internal", ], diff --git a/pkg/sentry/socket/unix/BUILD b/pkg/sentry/socket/unix/BUILD index 5b6a154f6..bade18686 100644 --- a/pkg/sentry/socket/unix/BUILD +++ b/pkg/sentry/socket/unix/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -9,7 +9,6 @@ go_library( "io.go", "unix.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/socket/unix", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", diff --git a/pkg/sentry/socket/unix/transport/BUILD b/pkg/sentry/socket/unix/transport/BUILD index d7ba95dff..4bdfc9208 100644 --- a/pkg/sentry/socket/unix/transport/BUILD +++ b/pkg/sentry/socket/unix/transport/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") load("//tools/go_generics:defs.bzl", "go_template_instance") package(licenses = ["notice"]) @@ -25,7 +25,6 @@ go_library( "transport_message_list.go", "unix.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport", visibility = ["//:sandbox"], deps = [ "//pkg/abi/linux", diff --git a/pkg/sentry/state/BUILD b/pkg/sentry/state/BUILD index 88765f4d6..0ea4aab8b 100644 --- a/pkg/sentry/state/BUILD +++ b/pkg/sentry/state/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -9,7 +9,6 @@ go_library( "state_metadata.go", "state_unsafe.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/state", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", diff --git a/pkg/sentry/strace/BUILD b/pkg/sentry/strace/BUILD index aa1ac720c..ff6fafa63 100644 --- a/pkg/sentry/strace/BUILD +++ b/pkg/sentry/strace/BUILD @@ -1,6 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//proto:def.bzl", "go_proto_library") -load("@rules_cc//cc:defs.bzl", "cc_proto_library") +load("//tools:defs.bzl", "go_library", "proto_library") package(licenses = ["notice"]) @@ -21,7 +19,6 @@ go_library( "strace.go", "syscalls.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/strace", visibility = ["//:sandbox"], deps = [ ":strace_go_proto", @@ -42,20 +39,7 @@ go_library( ) proto_library( - name = "strace_proto", + name = "strace", srcs = ["strace.proto"], visibility = ["//visibility:public"], ) - -cc_proto_library( - name = "strace_cc_proto", - visibility = ["//visibility:public"], - deps = [":strace_proto"], -) - -go_proto_library( - name = "strace_go_proto", - importpath = "gvisor.dev/gvisor/pkg/sentry/strace/strace_go_proto", - proto = ":strace_proto", - visibility = ["//visibility:public"], -) diff --git a/pkg/sentry/syscalls/BUILD b/pkg/sentry/syscalls/BUILD index 79d972202..b8d1bd415 100644 --- a/pkg/sentry/syscalls/BUILD +++ b/pkg/sentry/syscalls/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -8,7 +8,6 @@ go_library( "epoll.go", "syscalls.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/syscalls", visibility = ["//:sandbox"], deps = [ "//pkg/abi/linux", diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD index 917f74e07..7d74e0f70 100644 --- a/pkg/sentry/syscalls/linux/BUILD +++ b/pkg/sentry/syscalls/linux/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -57,7 +57,6 @@ go_library( "sys_xattr.go", "timespec.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/syscalls/linux", visibility = ["//:sandbox"], deps = [ "//pkg/abi", diff --git a/pkg/sentry/time/BUILD b/pkg/sentry/time/BUILD index 3cde3a0be..04f81a35b 100644 --- a/pkg/sentry/time/BUILD +++ b/pkg/sentry/time/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") package(licenses = ["notice"]) @@ -31,7 +30,6 @@ go_library( "tsc_amd64.s", "tsc_arm64.s", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/time", visibility = ["//:sandbox"], deps = [ "//pkg/log", @@ -48,5 +46,5 @@ go_test( "parameters_test.go", "sampler_test.go", ], - embed = [":time"], + library = ":time", ) diff --git a/pkg/sentry/unimpl/BUILD b/pkg/sentry/unimpl/BUILD index fc7614fff..370fa6ec5 100644 --- a/pkg/sentry/unimpl/BUILD +++ b/pkg/sentry/unimpl/BUILD @@ -1,34 +1,17 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//proto:def.bzl", "go_proto_library") -load("@rules_cc//cc:defs.bzl", "cc_proto_library") +load("//tools:defs.bzl", "go_library", "proto_library") package(licenses = ["notice"]) proto_library( - name = "unimplemented_syscall_proto", + name = "unimplemented_syscall", srcs = ["unimplemented_syscall.proto"], visibility = ["//visibility:public"], deps = ["//pkg/sentry/arch:registers_proto"], ) -cc_proto_library( - name = "unimplemented_syscall_cc_proto", - visibility = ["//visibility:public"], - deps = [":unimplemented_syscall_proto"], -) - -go_proto_library( - name = "unimplemented_syscall_go_proto", - importpath = "gvisor.dev/gvisor/pkg/sentry/unimpl/unimplemented_syscall_go_proto", - proto = ":unimplemented_syscall_proto", - visibility = ["//visibility:public"], - deps = ["//pkg/sentry/arch:registers_go_proto"], -) - go_library( name = "unimpl", srcs = ["events.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/unimpl", visibility = ["//:sandbox"], deps = [ "//pkg/log", diff --git a/pkg/sentry/uniqueid/BUILD b/pkg/sentry/uniqueid/BUILD index 86a87edd4..e9c18f170 100644 --- a/pkg/sentry/uniqueid/BUILD +++ b/pkg/sentry/uniqueid/BUILD @@ -1,11 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) go_library( name = "uniqueid", srcs = ["context.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/uniqueid", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/sentry/context", diff --git a/pkg/sentry/usage/BUILD b/pkg/sentry/usage/BUILD index 5518ac3d0..099315613 100644 --- a/pkg/sentry/usage/BUILD +++ b/pkg/sentry/usage/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -11,9 +11,8 @@ go_library( "memory_unsafe.go", "usage.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/usage", visibility = [ - "//pkg/sentry:internal", + "//:sandbox", ], deps = [ "//pkg/bits", diff --git a/pkg/sentry/usermem/BUILD b/pkg/sentry/usermem/BUILD index 684f59a6b..c8322e29e 100644 --- a/pkg/sentry/usermem/BUILD +++ b/pkg/sentry/usermem/BUILD @@ -1,6 +1,5 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -29,7 +28,6 @@ go_library( "usermem_unsafe.go", "usermem_x86.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/usermem", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/atomicbitops", @@ -38,7 +36,6 @@ go_library( "//pkg/sentry/context", "//pkg/sentry/safemem", "//pkg/syserror", - "//pkg/tcpip/buffer", ], ) @@ -49,7 +46,7 @@ go_test( "addr_range_seq_test.go", "usermem_test.go", ], - embed = [":usermem"], + library = ":usermem", deps = [ "//pkg/sentry/context", "//pkg/sentry/safemem", diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index 35c7be259..51acdc4e9 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -1,7 +1,6 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") -package(licenses = ["notice"]) +licenses(["notice"]) go_library( name = "vfs", @@ -24,7 +23,6 @@ go_library( "testutil.go", "vfs.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/vfs", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", @@ -47,7 +45,7 @@ go_test( "file_description_impl_util_test.go", "mount_test.go", ], - embed = [":vfs"], + library = ":vfs", deps = [ "//pkg/abi/linux", "//pkg/sentry/context", diff --git a/pkg/sentry/watchdog/BUILD b/pkg/sentry/watchdog/BUILD index 28f21f13d..1c5a1c9b6 100644 --- a/pkg/sentry/watchdog/BUILD +++ b/pkg/sentry/watchdog/BUILD @@ -1,11 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) go_library( name = "watchdog", srcs = ["watchdog.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/watchdog", visibility = ["//:sandbox"], deps = [ "//pkg/abi/linux", diff --git a/pkg/sleep/BUILD b/pkg/sleep/BUILD index a23c86fb1..e131455f7 100644 --- a/pkg/sleep/BUILD +++ b/pkg/sleep/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -12,7 +11,6 @@ go_library( "commit_noasm.go", "sleep_unsafe.go", ], - importpath = "gvisor.dev/gvisor/pkg/sleep", visibility = ["//:sandbox"], ) @@ -22,5 +20,5 @@ go_test( srcs = [ "sleep_test.go", ], - embed = [":sleep"], + library = ":sleep", ) diff --git a/pkg/state/BUILD b/pkg/state/BUILD index be93750bf..921af9d63 100644 --- a/pkg/state/BUILD +++ b/pkg/state/BUILD @@ -1,6 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//proto:def.bzl", "go_proto_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test", "proto_library") load("//tools/go_generics:defs.bzl", "go_template_instance") package(licenses = ["notice"]) @@ -49,7 +47,7 @@ go_library( "state.go", "stats.go", ], - importpath = "gvisor.dev/gvisor/pkg/state", + stateify = False, visibility = ["//:sandbox"], deps = [ ":object_go_proto", @@ -58,21 +56,14 @@ go_library( ) proto_library( - name = "object_proto", + name = "object", srcs = ["object.proto"], visibility = ["//:sandbox"], ) -go_proto_library( - name = "object_go_proto", - importpath = "gvisor.dev/gvisor/pkg/state/object_go_proto", - proto = ":object_proto", - visibility = ["//:sandbox"], -) - go_test( name = "state_test", timeout = "long", srcs = ["state_test.go"], - embed = [":state"], + library = ":state", ) diff --git a/pkg/state/statefile/BUILD b/pkg/state/statefile/BUILD index 8a865d229..e7581c09b 100644 --- a/pkg/state/statefile/BUILD +++ b/pkg/state/statefile/BUILD @@ -1,12 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) go_library( name = "statefile", srcs = ["statefile.go"], - importpath = "gvisor.dev/gvisor/pkg/state/statefile", visibility = ["//:sandbox"], deps = [ "//pkg/binary", @@ -18,6 +16,6 @@ go_test( name = "statefile_test", size = "small", srcs = ["statefile_test.go"], - embed = [":statefile"], + library = ":statefile", deps = ["//pkg/compressio"], ) diff --git a/pkg/sync/BUILD b/pkg/sync/BUILD index 97c4b3b1e..5340cf0d6 100644 --- a/pkg/sync/BUILD +++ b/pkg/sync/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template") package( @@ -40,7 +39,6 @@ go_library( "syncutil.go", "tmutex_unsafe.go", ], - importpath = "gvisor.dev/gvisor/pkg/sync", ) go_test( @@ -51,5 +49,5 @@ go_test( "seqcount_test.go", "tmutex_test.go", ], - embed = [":sync"], + library = ":sync", ) diff --git a/pkg/sync/atomicptrtest/BUILD b/pkg/sync/atomicptrtest/BUILD index 418eda29c..e97553254 100644 --- a/pkg/sync/atomicptrtest/BUILD +++ b/pkg/sync/atomicptrtest/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") package(licenses = ["notice"]) @@ -18,12 +17,11 @@ go_template_instance( go_library( name = "atomicptr", srcs = ["atomicptr_int_unsafe.go"], - importpath = "gvisor.dev/gvisor/pkg/sync/atomicptr", ) go_test( name = "atomicptr_test", size = "small", srcs = ["atomicptr_test.go"], - embed = [":atomicptr"], + library = ":atomicptr", ) diff --git a/pkg/sync/seqatomictest/BUILD b/pkg/sync/seqatomictest/BUILD index eba21518d..5c38c783e 100644 --- a/pkg/sync/seqatomictest/BUILD +++ b/pkg/sync/seqatomictest/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") package(licenses = ["notice"]) @@ -18,7 +17,6 @@ go_template_instance( go_library( name = "seqatomic", srcs = ["seqatomic_int_unsafe.go"], - importpath = "gvisor.dev/gvisor/pkg/sync/seqatomic", deps = [ "//pkg/sync", ], @@ -28,6 +26,6 @@ go_test( name = "seqatomic_test", size = "small", srcs = ["seqatomic_test.go"], - embed = [":seqatomic"], + library = ":seqatomic", deps = ["//pkg/sync"], ) diff --git a/pkg/syserr/BUILD b/pkg/syserr/BUILD index 5665ad4ee..7d760344a 100644 --- a/pkg/syserr/BUILD +++ b/pkg/syserr/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -9,7 +9,6 @@ go_library( "netstack.go", "syserr.go", ], - importpath = "gvisor.dev/gvisor/pkg/syserr", visibility = ["//visibility:public"], deps = [ "//pkg/abi/linux", diff --git a/pkg/syserror/BUILD b/pkg/syserror/BUILD index bd3f9fd28..b13c15d9b 100644 --- a/pkg/syserror/BUILD +++ b/pkg/syserror/BUILD @@ -1,12 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) go_library( name = "syserror", srcs = ["syserror.go"], - importpath = "gvisor.dev/gvisor/pkg/syserror", visibility = ["//visibility:public"], ) diff --git a/pkg/tcpip/BUILD b/pkg/tcpip/BUILD index 23e4b09e7..26f7ba86b 100644 --- a/pkg/tcpip/BUILD +++ b/pkg/tcpip/BUILD @@ -1,5 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -12,7 +11,6 @@ go_library( "time_unsafe.go", "timer.go", ], - importpath = "gvisor.dev/gvisor/pkg/tcpip", visibility = ["//visibility:public"], deps = [ "//pkg/sync", @@ -25,7 +23,7 @@ go_test( name = "tcpip_test", size = "small", srcs = ["tcpip_test.go"], - embed = [":tcpip"], + library = ":tcpip", ) go_test( diff --git a/pkg/tcpip/adapters/gonet/BUILD b/pkg/tcpip/adapters/gonet/BUILD index 3df7d18d3..a984f1712 100644 --- a/pkg/tcpip/adapters/gonet/BUILD +++ b/pkg/tcpip/adapters/gonet/BUILD @@ -1,12 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) go_library( name = "gonet", srcs = ["gonet.go"], - importpath = "gvisor.dev/gvisor/pkg/tcpip/adapters/gonet", visibility = ["//visibility:public"], deps = [ "//pkg/sync", @@ -23,7 +21,7 @@ go_test( name = "gonet_test", size = "small", srcs = ["gonet_test.go"], - embed = [":gonet"], + library = ":gonet", deps = [ "//pkg/tcpip", "//pkg/tcpip/header", diff --git a/pkg/tcpip/buffer/BUILD b/pkg/tcpip/buffer/BUILD index d6c31bfa2..563bc78ea 100644 --- a/pkg/tcpip/buffer/BUILD +++ b/pkg/tcpip/buffer/BUILD @@ -1,5 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -9,7 +8,6 @@ go_library( "prependable.go", "view.go", ], - importpath = "gvisor.dev/gvisor/pkg/tcpip/buffer", visibility = ["//visibility:public"], ) @@ -17,5 +15,5 @@ go_test( name = "buffer_test", size = "small", srcs = ["view_test.go"], - embed = [":buffer"], + library = ":buffer", ) diff --git a/pkg/tcpip/checker/BUILD b/pkg/tcpip/checker/BUILD index b6fa6fc37..ed434807f 100644 --- a/pkg/tcpip/checker/BUILD +++ b/pkg/tcpip/checker/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -6,7 +6,6 @@ go_library( name = "checker", testonly = 1, srcs = ["checker.go"], - importpath = "gvisor.dev/gvisor/pkg/tcpip/checker", visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", diff --git a/pkg/tcpip/hash/jenkins/BUILD b/pkg/tcpip/hash/jenkins/BUILD index e648efa71..ff2719291 100644 --- a/pkg/tcpip/hash/jenkins/BUILD +++ b/pkg/tcpip/hash/jenkins/BUILD @@ -1,12 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) go_library( name = "jenkins", srcs = ["jenkins.go"], - importpath = "gvisor.dev/gvisor/pkg/tcpip/hash/jenkins", visibility = ["//visibility:public"], ) @@ -16,5 +14,5 @@ go_test( srcs = [ "jenkins_test.go", ], - embed = [":jenkins"], + library = ":jenkins", ) diff --git a/pkg/tcpip/header/BUILD b/pkg/tcpip/header/BUILD index cd747d100..9da0d71f8 100644 --- a/pkg/tcpip/header/BUILD +++ b/pkg/tcpip/header/BUILD @@ -1,5 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -24,7 +23,6 @@ go_library( "tcp.go", "udp.go", ], - importpath = "gvisor.dev/gvisor/pkg/tcpip/header", visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", @@ -59,7 +57,7 @@ go_test( "eth_test.go", "ndp_test.go", ], - embed = [":header"], + library = ":header", deps = [ "//pkg/tcpip", "@com_github_google_go-cmp//cmp:go_default_library", diff --git a/pkg/tcpip/iptables/BUILD b/pkg/tcpip/iptables/BUILD index 297eaccaf..d1b73cfdf 100644 --- a/pkg/tcpip/iptables/BUILD +++ b/pkg/tcpip/iptables/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -9,7 +9,6 @@ go_library( "targets.go", "types.go", ], - importpath = "gvisor.dev/gvisor/pkg/tcpip/iptables", visibility = ["//visibility:public"], deps = [ "//pkg/log", diff --git a/pkg/tcpip/link/channel/BUILD b/pkg/tcpip/link/channel/BUILD index 7dbc05754..3974c464e 100644 --- a/pkg/tcpip/link/channel/BUILD +++ b/pkg/tcpip/link/channel/BUILD @@ -1,11 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) go_library( name = "channel", srcs = ["channel.go"], - importpath = "gvisor.dev/gvisor/pkg/tcpip/link/channel", visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", diff --git a/pkg/tcpip/link/fdbased/BUILD b/pkg/tcpip/link/fdbased/BUILD index 66cc53ed4..abe725548 100644 --- a/pkg/tcpip/link/fdbased/BUILD +++ b/pkg/tcpip/link/fdbased/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -13,7 +12,6 @@ go_library( "mmap_unsafe.go", "packet_dispatchers.go", ], - importpath = "gvisor.dev/gvisor/pkg/tcpip/link/fdbased", visibility = ["//visibility:public"], deps = [ "//pkg/sync", @@ -30,7 +28,7 @@ go_test( name = "fdbased_test", size = "small", srcs = ["endpoint_test.go"], - embed = [":fdbased"], + library = ":fdbased", deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/link/loopback/BUILD b/pkg/tcpip/link/loopback/BUILD index f35fcdff4..6bf3805b7 100644 --- a/pkg/tcpip/link/loopback/BUILD +++ b/pkg/tcpip/link/loopback/BUILD @@ -1,11 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) go_library( name = "loopback", srcs = ["loopback.go"], - importpath = "gvisor.dev/gvisor/pkg/tcpip/link/loopback", visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", diff --git a/pkg/tcpip/link/muxed/BUILD b/pkg/tcpip/link/muxed/BUILD index 1ac7948b6..82b441b79 100644 --- a/pkg/tcpip/link/muxed/BUILD +++ b/pkg/tcpip/link/muxed/BUILD @@ -1,12 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) go_library( name = "muxed", srcs = ["injectable.go"], - importpath = "gvisor.dev/gvisor/pkg/tcpip/link/muxed", visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", @@ -19,7 +17,7 @@ go_test( name = "muxed_test", size = "small", srcs = ["injectable_test.go"], - embed = [":muxed"], + library = ":muxed", deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/link/rawfile/BUILD b/pkg/tcpip/link/rawfile/BUILD index d8211e93d..14b527bc2 100644 --- a/pkg/tcpip/link/rawfile/BUILD +++ b/pkg/tcpip/link/rawfile/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -12,7 +12,6 @@ go_library( "errors.go", "rawfile_unsafe.go", ], - importpath = "gvisor.dev/gvisor/pkg/tcpip/link/rawfile", visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", diff --git a/pkg/tcpip/link/sharedmem/BUILD b/pkg/tcpip/link/sharedmem/BUILD index 09165dd4c..13243ebbb 100644 --- a/pkg/tcpip/link/sharedmem/BUILD +++ b/pkg/tcpip/link/sharedmem/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -11,7 +10,6 @@ go_library( "sharedmem_unsafe.go", "tx.go", ], - importpath = "gvisor.dev/gvisor/pkg/tcpip/link/sharedmem", visibility = ["//visibility:public"], deps = [ "//pkg/log", @@ -30,7 +28,7 @@ go_test( srcs = [ "sharedmem_test.go", ], - embed = [":sharedmem"], + library = ":sharedmem", deps = [ "//pkg/sync", "//pkg/tcpip", diff --git a/pkg/tcpip/link/sharedmem/pipe/BUILD b/pkg/tcpip/link/sharedmem/pipe/BUILD index a0d4ad0be..87020ec08 100644 --- a/pkg/tcpip/link/sharedmem/pipe/BUILD +++ b/pkg/tcpip/link/sharedmem/pipe/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -11,7 +10,6 @@ go_library( "rx.go", "tx.go", ], - importpath = "gvisor.dev/gvisor/pkg/tcpip/link/sharedmem/pipe", visibility = ["//visibility:public"], ) @@ -20,6 +18,6 @@ go_test( srcs = [ "pipe_test.go", ], - embed = [":pipe"], + library = ":pipe", deps = ["//pkg/sync"], ) diff --git a/pkg/tcpip/link/sharedmem/queue/BUILD b/pkg/tcpip/link/sharedmem/queue/BUILD index 8c9234d54..3ba06af73 100644 --- a/pkg/tcpip/link/sharedmem/queue/BUILD +++ b/pkg/tcpip/link/sharedmem/queue/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -9,7 +8,6 @@ go_library( "rx.go", "tx.go", ], - importpath = "gvisor.dev/gvisor/pkg/tcpip/link/sharedmem/queue", visibility = ["//visibility:public"], deps = [ "//pkg/log", @@ -22,7 +20,7 @@ go_test( srcs = [ "queue_test.go", ], - embed = [":queue"], + library = ":queue", deps = [ "//pkg/tcpip/link/sharedmem/pipe", ], diff --git a/pkg/tcpip/link/sniffer/BUILD b/pkg/tcpip/link/sniffer/BUILD index d6ae0368a..230a8d53a 100644 --- a/pkg/tcpip/link/sniffer/BUILD +++ b/pkg/tcpip/link/sniffer/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -8,7 +8,6 @@ go_library( "pcap.go", "sniffer.go", ], - importpath = "gvisor.dev/gvisor/pkg/tcpip/link/sniffer", visibility = ["//visibility:public"], deps = [ "//pkg/log", diff --git a/pkg/tcpip/link/tun/BUILD b/pkg/tcpip/link/tun/BUILD index a71a493fc..e5096ea38 100644 --- a/pkg/tcpip/link/tun/BUILD +++ b/pkg/tcpip/link/tun/BUILD @@ -1,10 +1,9 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) go_library( name = "tun", srcs = ["tun_unsafe.go"], - importpath = "gvisor.dev/gvisor/pkg/tcpip/link/tun", visibility = ["//visibility:public"], ) diff --git a/pkg/tcpip/link/waitable/BUILD b/pkg/tcpip/link/waitable/BUILD index 134837943..0956d2c65 100644 --- a/pkg/tcpip/link/waitable/BUILD +++ b/pkg/tcpip/link/waitable/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -8,7 +7,6 @@ go_library( srcs = [ "waitable.go", ], - importpath = "gvisor.dev/gvisor/pkg/tcpip/link/waitable", visibility = ["//visibility:public"], deps = [ "//pkg/gate", @@ -23,7 +21,7 @@ go_test( srcs = [ "waitable_test.go", ], - embed = [":waitable"], + library = ":waitable", deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/network/BUILD b/pkg/tcpip/network/BUILD index 9d16ff8c9..6a4839fb8 100644 --- a/pkg/tcpip/network/BUILD +++ b/pkg/tcpip/network/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_test") package(licenses = ["notice"]) diff --git a/pkg/tcpip/network/arp/BUILD b/pkg/tcpip/network/arp/BUILD index e7617229b..eddf7b725 100644 --- a/pkg/tcpip/network/arp/BUILD +++ b/pkg/tcpip/network/arp/BUILD @@ -1,12 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) go_library( name = "arp", srcs = ["arp.go"], - importpath = "gvisor.dev/gvisor/pkg/tcpip/network/arp", visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", diff --git a/pkg/tcpip/network/fragmentation/BUILD b/pkg/tcpip/network/fragmentation/BUILD index ed16076fd..d1c728ccf 100644 --- a/pkg/tcpip/network/fragmentation/BUILD +++ b/pkg/tcpip/network/fragmentation/BUILD @@ -1,6 +1,5 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -24,7 +23,6 @@ go_library( "reassembler.go", "reassembler_list.go", ], - importpath = "gvisor.dev/gvisor/pkg/tcpip/network/fragmentation", visibility = ["//visibility:public"], deps = [ "//pkg/log", @@ -42,6 +40,6 @@ go_test( "fragmentation_test.go", "reassembler_test.go", ], - embed = [":fragmentation"], + library = ":fragmentation", deps = ["//pkg/tcpip/buffer"], ) diff --git a/pkg/tcpip/network/hash/BUILD b/pkg/tcpip/network/hash/BUILD index e6db5c0b0..872165866 100644 --- a/pkg/tcpip/network/hash/BUILD +++ b/pkg/tcpip/network/hash/BUILD @@ -1,11 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) go_library( name = "hash", srcs = ["hash.go"], - importpath = "gvisor.dev/gvisor/pkg/tcpip/network/hash", visibility = ["//visibility:public"], deps = [ "//pkg/rand", diff --git a/pkg/tcpip/network/ipv4/BUILD b/pkg/tcpip/network/ipv4/BUILD index 4e2aae9a3..0fef2b1f1 100644 --- a/pkg/tcpip/network/ipv4/BUILD +++ b/pkg/tcpip/network/ipv4/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -9,7 +8,6 @@ go_library( "icmp.go", "ipv4.go", ], - importpath = "gvisor.dev/gvisor/pkg/tcpip/network/ipv4", visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", diff --git a/pkg/tcpip/network/ipv6/BUILD b/pkg/tcpip/network/ipv6/BUILD index e4e273460..fb11874c6 100644 --- a/pkg/tcpip/network/ipv6/BUILD +++ b/pkg/tcpip/network/ipv6/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -9,7 +8,6 @@ go_library( "icmp.go", "ipv6.go", ], - importpath = "gvisor.dev/gvisor/pkg/tcpip/network/ipv6", visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", @@ -27,7 +25,7 @@ go_test( "ipv6_test.go", "ndp_test.go", ], - embed = [":ipv6"], + library = ":ipv6", deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/ports/BUILD b/pkg/tcpip/ports/BUILD index a6ef3bdcc..2bad05a2e 100644 --- a/pkg/tcpip/ports/BUILD +++ b/pkg/tcpip/ports/BUILD @@ -1,12 +1,10 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) go_library( name = "ports", srcs = ["ports.go"], - importpath = "gvisor.dev/gvisor/pkg/tcpip/ports", visibility = ["//visibility:public"], deps = [ "//pkg/sync", @@ -17,7 +15,7 @@ go_library( go_test( name = "ports_test", srcs = ["ports_test.go"], - embed = [":ports"], + library = ":ports", deps = [ "//pkg/tcpip", ], diff --git a/pkg/tcpip/sample/tun_tcp_connect/BUILD b/pkg/tcpip/sample/tun_tcp_connect/BUILD index d7496fde6..cf0a5fefe 100644 --- a/pkg/tcpip/sample/tun_tcp_connect/BUILD +++ b/pkg/tcpip/sample/tun_tcp_connect/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_binary") +load("//tools:defs.bzl", "go_binary") package(licenses = ["notice"]) diff --git a/pkg/tcpip/sample/tun_tcp_echo/BUILD b/pkg/tcpip/sample/tun_tcp_echo/BUILD index 875561566..43264b76d 100644 --- a/pkg/tcpip/sample/tun_tcp_echo/BUILD +++ b/pkg/tcpip/sample/tun_tcp_echo/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_binary") +load("//tools:defs.bzl", "go_binary") package(licenses = ["notice"]) diff --git a/pkg/tcpip/seqnum/BUILD b/pkg/tcpip/seqnum/BUILD index b31ddba2f..45f503845 100644 --- a/pkg/tcpip/seqnum/BUILD +++ b/pkg/tcpip/seqnum/BUILD @@ -1,10 +1,9 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) go_library( name = "seqnum", srcs = ["seqnum.go"], - importpath = "gvisor.dev/gvisor/pkg/tcpip/seqnum", visibility = ["//visibility:public"], ) diff --git a/pkg/tcpip/stack/BUILD b/pkg/tcpip/stack/BUILD index 783351a69..f5b750046 100644 --- a/pkg/tcpip/stack/BUILD +++ b/pkg/tcpip/stack/BUILD @@ -1,6 +1,5 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -30,7 +29,6 @@ go_library( "stack_global_state.go", "transport_demuxer.go", ], - importpath = "gvisor.dev/gvisor/pkg/tcpip/stack", visibility = ["//visibility:public"], deps = [ "//pkg/ilist", @@ -81,7 +79,7 @@ go_test( name = "stack_test", size = "small", srcs = ["linkaddrcache_test.go"], - embed = [":stack"], + library = ":stack", deps = [ "//pkg/sleep", "//pkg/sync", diff --git a/pkg/tcpip/transport/icmp/BUILD b/pkg/tcpip/transport/icmp/BUILD index 3aa23d529..ac18ec5b1 100644 --- a/pkg/tcpip/transport/icmp/BUILD +++ b/pkg/tcpip/transport/icmp/BUILD @@ -1,5 +1,5 @@ +load("//tools:defs.bzl", "go_library") load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -23,7 +23,6 @@ go_library( "icmp_packet_list.go", "protocol.go", ], - importpath = "gvisor.dev/gvisor/pkg/tcpip/transport/icmp", imports = ["gvisor.dev/gvisor/pkg/tcpip/buffer"], visibility = ["//visibility:public"], deps = [ diff --git a/pkg/tcpip/transport/packet/BUILD b/pkg/tcpip/transport/packet/BUILD index 4858d150c..d22de6b26 100644 --- a/pkg/tcpip/transport/packet/BUILD +++ b/pkg/tcpip/transport/packet/BUILD @@ -1,5 +1,5 @@ +load("//tools:defs.bzl", "go_library") load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -22,7 +22,6 @@ go_library( "endpoint_state.go", "packet_list.go", ], - importpath = "gvisor.dev/gvisor/pkg/tcpip/transport/packet", imports = ["gvisor.dev/gvisor/pkg/tcpip/buffer"], visibility = ["//visibility:public"], deps = [ diff --git a/pkg/tcpip/transport/raw/BUILD b/pkg/tcpip/transport/raw/BUILD index 2f2131ff7..c9baf4600 100644 --- a/pkg/tcpip/transport/raw/BUILD +++ b/pkg/tcpip/transport/raw/BUILD @@ -1,5 +1,5 @@ +load("//tools:defs.bzl", "go_library") load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -23,7 +23,6 @@ go_library( "protocol.go", "raw_packet_list.go", ], - importpath = "gvisor.dev/gvisor/pkg/tcpip/transport/raw", imports = ["gvisor.dev/gvisor/pkg/tcpip/buffer"], visibility = ["//visibility:public"], deps = [ diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD index 0e3ab05ad..4acd9fb9a 100644 --- a/pkg/tcpip/transport/tcp/BUILD +++ b/pkg/tcpip/transport/tcp/BUILD @@ -1,6 +1,5 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -55,7 +54,6 @@ go_library( "tcp_segment_list.go", "timer.go", ], - importpath = "gvisor.dev/gvisor/pkg/tcpip/transport/tcp", imports = ["gvisor.dev/gvisor/pkg/tcpip/buffer"], visibility = ["//visibility:public"], deps = [ diff --git a/pkg/tcpip/transport/tcp/testing/context/BUILD b/pkg/tcpip/transport/tcp/testing/context/BUILD index b33ec2087..ce6a2c31d 100644 --- a/pkg/tcpip/transport/tcp/testing/context/BUILD +++ b/pkg/tcpip/transport/tcp/testing/context/BUILD @@ -1,4 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -6,7 +6,6 @@ go_library( name = "context", testonly = 1, srcs = ["context.go"], - importpath = "gvisor.dev/gvisor/pkg/tcpip/transport/tcp/testing/context", visibility = [ "//visibility:public", ], diff --git a/pkg/tcpip/transport/tcpconntrack/BUILD b/pkg/tcpip/transport/tcpconntrack/BUILD index 43fcc27f0..3ad6994a7 100644 --- a/pkg/tcpip/transport/tcpconntrack/BUILD +++ b/pkg/tcpip/transport/tcpconntrack/BUILD @@ -1,12 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) go_library( name = "tcpconntrack", srcs = ["tcp_conntrack.go"], - importpath = "gvisor.dev/gvisor/pkg/tcpip/transport/tcpconntrack", visibility = ["//visibility:public"], deps = [ "//pkg/tcpip/header", diff --git a/pkg/tcpip/transport/udp/BUILD b/pkg/tcpip/transport/udp/BUILD index 57ff123e3..adc908e24 100644 --- a/pkg/tcpip/transport/udp/BUILD +++ b/pkg/tcpip/transport/udp/BUILD @@ -1,6 +1,5 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -25,7 +24,6 @@ go_library( "protocol.go", "udp_packet_list.go", ], - importpath = "gvisor.dev/gvisor/pkg/tcpip/transport/udp", imports = ["gvisor.dev/gvisor/pkg/tcpip/buffer"], visibility = ["//visibility:public"], deps = [ diff --git a/pkg/tmutex/BUILD b/pkg/tmutex/BUILD index 07778e4f7..2dcba84ae 100644 --- a/pkg/tmutex/BUILD +++ b/pkg/tmutex/BUILD @@ -1,12 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) go_library( name = "tmutex", srcs = ["tmutex.go"], - importpath = "gvisor.dev/gvisor/pkg/tmutex", visibility = ["//:sandbox"], ) @@ -14,6 +12,6 @@ go_test( name = "tmutex_test", size = "medium", srcs = ["tmutex_test.go"], - embed = [":tmutex"], + library = ":tmutex", deps = ["//pkg/sync"], ) diff --git a/pkg/unet/BUILD b/pkg/unet/BUILD index d1885ae66..a86501fa2 100644 --- a/pkg/unet/BUILD +++ b/pkg/unet/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -9,7 +8,6 @@ go_library( "unet.go", "unet_unsafe.go", ], - importpath = "gvisor.dev/gvisor/pkg/unet", visibility = ["//visibility:public"], deps = [ "//pkg/gate", @@ -23,6 +21,6 @@ go_test( srcs = [ "unet_test.go", ], - embed = [":unet"], + library = ":unet", deps = ["//pkg/sync"], ) diff --git a/pkg/urpc/BUILD b/pkg/urpc/BUILD index b8fdc3125..850c34ed0 100644 --- a/pkg/urpc/BUILD +++ b/pkg/urpc/BUILD @@ -1,12 +1,10 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) go_library( name = "urpc", srcs = ["urpc.go"], - importpath = "gvisor.dev/gvisor/pkg/urpc", visibility = ["//:sandbox"], deps = [ "//pkg/fd", @@ -20,6 +18,6 @@ go_test( name = "urpc_test", size = "small", srcs = ["urpc_test.go"], - embed = [":urpc"], + library = ":urpc", deps = ["//pkg/unet"], ) diff --git a/pkg/waiter/BUILD b/pkg/waiter/BUILD index 1c6890e52..852480a09 100644 --- a/pkg/waiter/BUILD +++ b/pkg/waiter/BUILD @@ -1,6 +1,5 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -22,7 +21,6 @@ go_library( "waiter.go", "waiter_list.go", ], - importpath = "gvisor.dev/gvisor/pkg/waiter", visibility = ["//visibility:public"], deps = ["//pkg/sync"], ) @@ -33,5 +31,5 @@ go_test( srcs = [ "waiter_test.go", ], - embed = [":waiter"], + library = ":waiter", ) diff --git a/runsc/BUILD b/runsc/BUILD index e5587421d..b35b41d81 100644 --- a/runsc/BUILD +++ b/runsc/BUILD @@ -1,7 +1,6 @@ -package(licenses = ["notice"]) # Apache 2.0 +load("//tools:defs.bzl", "go_binary", "pkg_deb", "pkg_tar") -load("@io_bazel_rules_go//go:def.bzl", "go_binary") -load("@rules_pkg//:pkg.bzl", "pkg_deb", "pkg_tar") +package(licenses = ["notice"]) go_binary( name = "runsc", @@ -9,7 +8,7 @@ go_binary( "main.go", "version.go", ], - pure = "on", + pure = True, visibility = [ "//visibility:public", ], @@ -26,10 +25,12 @@ go_binary( ) # The runsc-race target is a race-compatible BUILD target. This must be built -# via "bazel build --features=race //runsc:runsc-race", since the race feature -# must apply to all dependencies due a bug in gazelle file selection. The pure -# attribute must be off because the race detector requires linking with non-Go -# components, although we still require a static binary. +# via: bazel build --features=race //runsc:runsc-race +# +# This is neccessary because the race feature must apply to all dependencies +# due a bug in gazelle file selection. The pure attribute must be off because +# the race detector requires linking with non-Go components, although we still +# require a static binary. # # Note that in the future this might be convertible to a compatible target by # using the pure and static attributes within a select function, but select is @@ -42,7 +43,7 @@ go_binary( "main.go", "version.go", ], - static = "on", + static = True, visibility = [ "//visibility:public", ], @@ -82,7 +83,12 @@ genrule( # because they are assumes to be hermetic). srcs = [":runsc"], outs = ["version.txt"], - cmd = "$(location :runsc) -version | grep 'runsc version' | sed 's/^[^0-9]*//' > $@", + # Note that the little dance here is necessary because files in the $(SRCS) + # attribute are not executable by default, and we can't touch in place. + cmd = "cp $(location :runsc) $(@D)/runsc && \ + chmod a+x $(@D)/runsc && \ + $(@D)/runsc -version | grep version | sed 's/^[^0-9]*//' > $@ && \ + rm -f $(@D)/runsc", stamp = 1, ) @@ -109,5 +115,6 @@ sh_test( name = "version_test", size = "small", srcs = ["version_test.sh"], + args = ["$(location :runsc)"], data = [":runsc"], ) diff --git a/runsc/boot/BUILD b/runsc/boot/BUILD index 3e20f8f2f..f3ebc0231 100644 --- a/runsc/boot/BUILD +++ b/runsc/boot/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -23,7 +23,6 @@ go_library( "strace.go", "user.go", ], - importpath = "gvisor.dev/gvisor/runsc/boot", visibility = [ "//runsc:__subpackages__", "//test:__subpackages__", @@ -107,7 +106,7 @@ go_test( "loader_test.go", "user_test.go", ], - embed = [":boot"], + library = ":boot", deps = [ "//pkg/control/server", "//pkg/log", diff --git a/runsc/boot/filter/BUILD b/runsc/boot/filter/BUILD index 3a9dcfc04..ce30f6c53 100644 --- a/runsc/boot/filter/BUILD +++ b/runsc/boot/filter/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -13,7 +13,6 @@ go_library( "extra_filters_race.go", "filter.go", ], - importpath = "gvisor.dev/gvisor/runsc/boot/filter", visibility = [ "//runsc/boot:__subpackages__", ], diff --git a/runsc/boot/platforms/BUILD b/runsc/boot/platforms/BUILD index 03391cdca..77774f43c 100644 --- a/runsc/boot/platforms/BUILD +++ b/runsc/boot/platforms/BUILD @@ -1,11 +1,10 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) go_library( name = "platforms", srcs = ["platforms.go"], - importpath = "gvisor.dev/gvisor/runsc/boot/platforms", visibility = [ "//runsc:__subpackages__", ], diff --git a/runsc/cgroup/BUILD b/runsc/cgroup/BUILD index d6165f9e5..d4c7bdfbb 100644 --- a/runsc/cgroup/BUILD +++ b/runsc/cgroup/BUILD @@ -1,11 +1,10 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) go_library( name = "cgroup", srcs = ["cgroup.go"], - importpath = "gvisor.dev/gvisor/runsc/cgroup", visibility = ["//:sandbox"], deps = [ "//pkg/log", @@ -19,6 +18,6 @@ go_test( name = "cgroup_test", size = "small", srcs = ["cgroup_test.go"], - embed = [":cgroup"], + library = ":cgroup", tags = ["local"], ) diff --git a/runsc/cmd/BUILD b/runsc/cmd/BUILD index b94bc4fa0..09aa46434 100644 --- a/runsc/cmd/BUILD +++ b/runsc/cmd/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -34,7 +34,6 @@ go_library( "syscalls.go", "wait.go", ], - importpath = "gvisor.dev/gvisor/runsc/cmd", visibility = [ "//runsc:__subpackages__", ], @@ -73,7 +72,7 @@ go_test( data = [ "//runsc", ], - embed = [":cmd"], + library = ":cmd", deps = [ "//pkg/abi/linux", "//pkg/log", diff --git a/runsc/console/BUILD b/runsc/console/BUILD index e623c1a0f..06924bccd 100644 --- a/runsc/console/BUILD +++ b/runsc/console/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -7,7 +7,6 @@ go_library( srcs = [ "console.go", ], - importpath = "gvisor.dev/gvisor/runsc/console", visibility = [ "//runsc:__subpackages__", ], diff --git a/runsc/container/BUILD b/runsc/container/BUILD index 6dea179e4..e21431e4c 100644 --- a/runsc/container/BUILD +++ b/runsc/container/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -10,7 +10,6 @@ go_library( "state_file.go", "status.go", ], - importpath = "gvisor.dev/gvisor/runsc/container", visibility = [ "//runsc:__subpackages__", "//test:__subpackages__", @@ -42,7 +41,7 @@ go_test( "//runsc", "//runsc/container/test_app", ], - embed = [":container"], + library = ":container", shard_count = 5, tags = [ "requires-kvm", diff --git a/runsc/container/test_app/BUILD b/runsc/container/test_app/BUILD index bfd338bb6..e200bafd9 100644 --- a/runsc/container/test_app/BUILD +++ b/runsc/container/test_app/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_binary") +load("//tools:defs.bzl", "go_binary") package(licenses = ["notice"]) @@ -9,7 +9,7 @@ go_binary( "fds.go", "test_app.go", ], - pure = "on", + pure = True, visibility = ["//runsc/container:__pkg__"], deps = [ "//pkg/unet", diff --git a/runsc/criutil/BUILD b/runsc/criutil/BUILD index 558133a0e..8a571a000 100644 --- a/runsc/criutil/BUILD +++ b/runsc/criutil/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -6,7 +6,6 @@ go_library( name = "criutil", testonly = 1, srcs = ["criutil.go"], - importpath = "gvisor.dev/gvisor/runsc/criutil", visibility = ["//:sandbox"], deps = ["//runsc/testutil"], ) diff --git a/runsc/dockerutil/BUILD b/runsc/dockerutil/BUILD index 0e0423504..8621af901 100644 --- a/runsc/dockerutil/BUILD +++ b/runsc/dockerutil/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -6,7 +6,6 @@ go_library( name = "dockerutil", testonly = 1, srcs = ["dockerutil.go"], - importpath = "gvisor.dev/gvisor/runsc/dockerutil", visibility = ["//:sandbox"], deps = [ "//runsc/testutil", diff --git a/runsc/fsgofer/BUILD b/runsc/fsgofer/BUILD index a9582d92b..64a406ae2 100644 --- a/runsc/fsgofer/BUILD +++ b/runsc/fsgofer/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -10,10 +10,7 @@ go_library( "fsgofer_arm64_unsafe.go", "fsgofer_unsafe.go", ], - importpath = "gvisor.dev/gvisor/runsc/fsgofer", - visibility = [ - "//runsc:__subpackages__", - ], + visibility = ["//runsc:__subpackages__"], deps = [ "//pkg/abi/linux", "//pkg/fd", @@ -30,7 +27,7 @@ go_test( name = "fsgofer_test", size = "small", srcs = ["fsgofer_test.go"], - embed = [":fsgofer"], + library = ":fsgofer", deps = [ "//pkg/log", "//pkg/p9", diff --git a/runsc/fsgofer/filter/BUILD b/runsc/fsgofer/filter/BUILD index bac73f89d..82b48ef32 100644 --- a/runsc/fsgofer/filter/BUILD +++ b/runsc/fsgofer/filter/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -13,7 +13,6 @@ go_library( "extra_filters_race.go", "filter.go", ], - importpath = "gvisor.dev/gvisor/runsc/fsgofer/filter", visibility = [ "//runsc:__subpackages__", ], diff --git a/runsc/sandbox/BUILD b/runsc/sandbox/BUILD index ddbc37456..c95d50294 100644 --- a/runsc/sandbox/BUILD +++ b/runsc/sandbox/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -9,7 +9,6 @@ go_library( "network_unsafe.go", "sandbox.go", ], - importpath = "gvisor.dev/gvisor/runsc/sandbox", visibility = [ "//runsc:__subpackages__", ], diff --git a/runsc/specutils/BUILD b/runsc/specutils/BUILD index 205638803..4ccd77f63 100644 --- a/runsc/specutils/BUILD +++ b/runsc/specutils/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -10,7 +10,6 @@ go_library( "namespace.go", "specutils.go", ], - importpath = "gvisor.dev/gvisor/runsc/specutils", visibility = ["//:sandbox"], deps = [ "//pkg/abi/linux", @@ -28,6 +27,6 @@ go_test( name = "specutils_test", size = "small", srcs = ["specutils_test.go"], - embed = [":specutils"], + library = ":specutils", deps = ["@com_github_opencontainers_runtime-spec//specs-go:go_default_library"], ) diff --git a/runsc/testutil/BUILD b/runsc/testutil/BUILD index 3c3027cb5..f845120b0 100644 --- a/runsc/testutil/BUILD +++ b/runsc/testutil/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -6,7 +6,6 @@ go_library( name = "testutil", testonly = 1, srcs = ["testutil.go"], - importpath = "gvisor.dev/gvisor/runsc/testutil", visibility = ["//:sandbox"], deps = [ "//pkg/log", diff --git a/runsc/version_test.sh b/runsc/version_test.sh index cc0ca3f05..747350654 100755 --- a/runsc/version_test.sh +++ b/runsc/version_test.sh @@ -16,7 +16,7 @@ set -euf -x -o pipefail -readonly runsc="${TEST_SRCDIR}/__main__/runsc/linux_amd64_pure_stripped/runsc" +readonly runsc="$1" readonly version=$($runsc --version) # Version should should not match VERSION, which is the default and which will diff --git a/scripts/common.sh b/scripts/common.sh index fdb1aa142..cd91b9f8e 100755 --- a/scripts/common.sh +++ b/scripts/common.sh @@ -16,11 +16,7 @@ set -xeou pipefail -if [[ -f $(dirname $0)/common_google.sh ]]; then - source $(dirname $0)/common_google.sh -else - source $(dirname $0)/common_bazel.sh -fi +source $(dirname $0)/common_build.sh # Ensure it attempts to collect logs in all cases. trap collect_logs EXIT diff --git a/scripts/common_bazel.sh b/scripts/common_bazel.sh deleted file mode 100755 index a473a88a4..000000000 --- a/scripts/common_bazel.sh +++ /dev/null @@ -1,99 +0,0 @@ -#!/bin/bash - -# Copyright 2019 The gVisor Authors. -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -# Install the latest version of Bazel and log the version. -(which use_bazel.sh && use_bazel.sh latest) || which bazel -bazel version - -# Switch into the workspace; only necessary if run with kokoro. -if [[ -v KOKORO_GIT_COMMIT ]] && [[ -d git/repo ]]; then - cd git/repo -elif [[ -v KOKORO_GIT_COMMIT ]] && [[ -d github/repo ]]; then - cd github/repo -fi - -# Set the standard bazel flags. -declare -r BAZEL_FLAGS=( - "--show_timestamps" - "--test_output=errors" - "--keep_going" - "--verbose_failures=true" -) -if [[ -v KOKORO_BAZEL_AUTH_CREDENTIAL ]]; then - declare -r BAZEL_RBE_AUTH_FLAGS=( - "--auth_credentials=${KOKORO_BAZEL_AUTH_CREDENTIAL}" - ) - declare -r BAZEL_RBE_FLAGS=("--config=remote") -fi - -# Wrap bazel. -function build() { - bazel build "${BAZEL_RBE_FLAGS[@]}" "${BAZEL_RBE_AUTH_FLAGS[@]}" "${BAZEL_FLAGS[@]}" "$@" 2>&1 | - tee /dev/fd/2 | grep -E '^ bazel-bin/' | awk '{ print $1; }' -} - -function test() { - bazel test "${BAZEL_RBE_FLAGS[@]}" "${BAZEL_RBE_AUTH_FLAGS[@]}" "${BAZEL_FLAGS[@]}" "$@" -} - -function run() { - local binary=$1 - shift - bazel run "${binary}" -- "$@" -} - -function run_as_root() { - local binary=$1 - shift - bazel run --run_under="sudo" "${binary}" -- "$@" -} - -function collect_logs() { - # Zip out everything into a convenient form. - if [[ -v KOKORO_ARTIFACTS_DIR ]] && [[ -e bazel-testlogs ]]; then - # Merge results files of all shards for each test suite. - for d in `find -L "bazel-testlogs" -name 'shard_*_of_*' | xargs dirname | sort | uniq`; do - junitparser merge `find $d -name test.xml` $d/test.xml - cat $d/shard_*_of_*/test.log > $d/test.log - ls -l $d/shard_*_of_*/test.outputs/outputs.zip && zip -r -1 $d/outputs.zip $d/shard_*_of_*/test.outputs/outputs.zip - done - find -L "bazel-testlogs" -name 'shard_*_of_*' | xargs rm -rf - # Move test logs to Kokoro directory. tar is used to conveniently perform - # renames while moving files. - find -L "bazel-testlogs" -name "test.xml" -o -name "test.log" -o -name "outputs.zip" | - tar --create --files-from - --transform 's/test\./sponge_log./' | - tar --extract --directory ${KOKORO_ARTIFACTS_DIR} - - # Collect sentry logs, if any. - if [[ -v RUNSC_LOGS_DIR ]] && [[ -d "${RUNSC_LOGS_DIR}" ]]; then - # Check if the directory is empty or not (only the first line it needed). - local -r logs=$(ls "${RUNSC_LOGS_DIR}" | head -n1) - if [[ "${logs}" ]]; then - local -r archive=runsc_logs_"${RUNTIME}".tar.gz - if [[ -v KOKORO_BUILD_ARTIFACTS_SUBDIR ]]; then - echo "runsc logs will be uploaded to:" - echo " gsutil cp gs://gvisor/logs/${KOKORO_BUILD_ARTIFACTS_SUBDIR}/${archive} /tmp" - echo " https://storage.cloud.google.com/gvisor/logs/${KOKORO_BUILD_ARTIFACTS_SUBDIR}/${archive}" - fi - tar --create --gzip --file="${KOKORO_ARTIFACTS_DIR}/${archive}" -C "${RUNSC_LOGS_DIR}" . - fi - fi - fi -} - -function find_branch_name() { - git branch --show-current || git rev-parse HEAD || bazel info workspace | xargs basename -} diff --git a/scripts/common_build.sh b/scripts/common_build.sh new file mode 100755 index 000000000..a473a88a4 --- /dev/null +++ b/scripts/common_build.sh @@ -0,0 +1,99 @@ +#!/bin/bash + +# Copyright 2019 The gVisor Authors. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +# Install the latest version of Bazel and log the version. +(which use_bazel.sh && use_bazel.sh latest) || which bazel +bazel version + +# Switch into the workspace; only necessary if run with kokoro. +if [[ -v KOKORO_GIT_COMMIT ]] && [[ -d git/repo ]]; then + cd git/repo +elif [[ -v KOKORO_GIT_COMMIT ]] && [[ -d github/repo ]]; then + cd github/repo +fi + +# Set the standard bazel flags. +declare -r BAZEL_FLAGS=( + "--show_timestamps" + "--test_output=errors" + "--keep_going" + "--verbose_failures=true" +) +if [[ -v KOKORO_BAZEL_AUTH_CREDENTIAL ]]; then + declare -r BAZEL_RBE_AUTH_FLAGS=( + "--auth_credentials=${KOKORO_BAZEL_AUTH_CREDENTIAL}" + ) + declare -r BAZEL_RBE_FLAGS=("--config=remote") +fi + +# Wrap bazel. +function build() { + bazel build "${BAZEL_RBE_FLAGS[@]}" "${BAZEL_RBE_AUTH_FLAGS[@]}" "${BAZEL_FLAGS[@]}" "$@" 2>&1 | + tee /dev/fd/2 | grep -E '^ bazel-bin/' | awk '{ print $1; }' +} + +function test() { + bazel test "${BAZEL_RBE_FLAGS[@]}" "${BAZEL_RBE_AUTH_FLAGS[@]}" "${BAZEL_FLAGS[@]}" "$@" +} + +function run() { + local binary=$1 + shift + bazel run "${binary}" -- "$@" +} + +function run_as_root() { + local binary=$1 + shift + bazel run --run_under="sudo" "${binary}" -- "$@" +} + +function collect_logs() { + # Zip out everything into a convenient form. + if [[ -v KOKORO_ARTIFACTS_DIR ]] && [[ -e bazel-testlogs ]]; then + # Merge results files of all shards for each test suite. + for d in `find -L "bazel-testlogs" -name 'shard_*_of_*' | xargs dirname | sort | uniq`; do + junitparser merge `find $d -name test.xml` $d/test.xml + cat $d/shard_*_of_*/test.log > $d/test.log + ls -l $d/shard_*_of_*/test.outputs/outputs.zip && zip -r -1 $d/outputs.zip $d/shard_*_of_*/test.outputs/outputs.zip + done + find -L "bazel-testlogs" -name 'shard_*_of_*' | xargs rm -rf + # Move test logs to Kokoro directory. tar is used to conveniently perform + # renames while moving files. + find -L "bazel-testlogs" -name "test.xml" -o -name "test.log" -o -name "outputs.zip" | + tar --create --files-from - --transform 's/test\./sponge_log./' | + tar --extract --directory ${KOKORO_ARTIFACTS_DIR} + + # Collect sentry logs, if any. + if [[ -v RUNSC_LOGS_DIR ]] && [[ -d "${RUNSC_LOGS_DIR}" ]]; then + # Check if the directory is empty or not (only the first line it needed). + local -r logs=$(ls "${RUNSC_LOGS_DIR}" | head -n1) + if [[ "${logs}" ]]; then + local -r archive=runsc_logs_"${RUNTIME}".tar.gz + if [[ -v KOKORO_BUILD_ARTIFACTS_SUBDIR ]]; then + echo "runsc logs will be uploaded to:" + echo " gsutil cp gs://gvisor/logs/${KOKORO_BUILD_ARTIFACTS_SUBDIR}/${archive} /tmp" + echo " https://storage.cloud.google.com/gvisor/logs/${KOKORO_BUILD_ARTIFACTS_SUBDIR}/${archive}" + fi + tar --create --gzip --file="${KOKORO_ARTIFACTS_DIR}/${archive}" -C "${RUNSC_LOGS_DIR}" . + fi + fi + fi +} + +function find_branch_name() { + git branch --show-current || git rev-parse HEAD || bazel info workspace | xargs basename +} diff --git a/test/BUILD b/test/BUILD index bf834d994..34b950644 100644 --- a/test/BUILD +++ b/test/BUILD @@ -1,44 +1 @@ -package(licenses = ["notice"]) # Apache 2.0 - -# We need to define a bazel platform and toolchain to specify dockerPrivileged -# and dockerRunAsRoot options, they are required to run tests on the RBE -# cluster in Kokoro. -alias( - name = "rbe_ubuntu1604", - actual = ":rbe_ubuntu1604_r346485", -) - -platform( - name = "rbe_ubuntu1604_r346485", - constraint_values = [ - "@bazel_tools//platforms:x86_64", - "@bazel_tools//platforms:linux", - "@bazel_tools//tools/cpp:clang", - "@bazel_toolchains//constraints:xenial", - "@bazel_toolchains//constraints/sanitizers:support_msan", - ], - remote_execution_properties = """ - properties: { - name: "container-image" - value:"docker://gcr.io/cloud-marketplace/google/rbe-ubuntu16-04@sha256:93f7e127196b9b653d39830c50f8b05d49ef6fd8739a9b5b8ab16e1df5399e50" - } - properties: { - name: "dockerAddCapabilities" - value: "SYS_ADMIN" - } - properties: { - name: "dockerPrivileged" - value: "true" - } - """, -) - -toolchain( - name = "cc-toolchain-clang-x86_64-default", - exec_compatible_with = [ - ], - target_compatible_with = [ - ], - toolchain = "@bazel_toolchains//configs/ubuntu16_04_clang/10.0.0/bazel_2.0.0/cc:cc-compiler-k8", - toolchain_type = "@bazel_tools//tools/cpp:toolchain_type", -) +package(licenses = ["notice"]) diff --git a/test/e2e/BUILD b/test/e2e/BUILD index 4fe03a220..76e04f878 100644 --- a/test/e2e/BUILD +++ b/test/e2e/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -10,7 +10,7 @@ go_test( "integration_test.go", "regression_test.go", ], - embed = [":integration"], + library = ":integration", tags = [ # Requires docker and runsc to be configured before the test runs. "manual", @@ -29,5 +29,4 @@ go_test( go_library( name = "integration", srcs = ["integration.go"], - importpath = "gvisor.dev/gvisor/test/integration", ) diff --git a/test/image/BUILD b/test/image/BUILD index 09b0a0ad5..7392ac54e 100644 --- a/test/image/BUILD +++ b/test/image/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -14,7 +14,7 @@ go_test( "ruby.rb", "ruby.sh", ], - embed = [":image"], + library = ":image", tags = [ # Requires docker and runsc to be configured before the test runs. "manual", @@ -30,5 +30,4 @@ go_test( go_library( name = "image", srcs = ["image.go"], - importpath = "gvisor.dev/gvisor/test/image", ) diff --git a/test/iptables/BUILD b/test/iptables/BUILD index 22f470092..6bb3b82b5 100644 --- a/test/iptables/BUILD +++ b/test/iptables/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -12,7 +12,6 @@ go_library( "iptables_util.go", "nat.go", ], - importpath = "gvisor.dev/gvisor/test/iptables", visibility = ["//test/iptables:__subpackages__"], deps = [ "//runsc/testutil", @@ -24,7 +23,7 @@ go_test( srcs = [ "iptables_test.go", ], - embed = [":iptables"], + library = ":iptables", tags = [ "local", "manual", diff --git a/test/iptables/runner/BUILD b/test/iptables/runner/BUILD index a5b6f082c..b9199387a 100644 --- a/test/iptables/runner/BUILD +++ b/test/iptables/runner/BUILD @@ -1,15 +1,21 @@ -load("@io_bazel_rules_docker//go:image.bzl", "go_image") -load("@io_bazel_rules_docker//container:container.bzl", "container_image") +load("//tools:defs.bzl", "container_image", "go_binary", "go_image") package(licenses = ["notice"]) +go_binary( + name = "runner", + testonly = 1, + srcs = ["main.go"], + deps = ["//test/iptables"], +) + container_image( name = "iptables-base", base = "@iptables-test//image", ) go_image( - name = "runner", + name = "runner-image", testonly = 1, srcs = ["main.go"], base = ":iptables-base", diff --git a/test/root/BUILD b/test/root/BUILD index d5dd9bca2..23ce2a70f 100644 --- a/test/root/BUILD +++ b/test/root/BUILD @@ -1,11 +1,10 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) go_library( name = "root", srcs = ["root.go"], - importpath = "gvisor.dev/gvisor/test/root", ) go_test( @@ -21,7 +20,7 @@ go_test( data = [ "//runsc", ], - embed = [":root"], + library = ":root", tags = [ # Requires docker and runsc to be configured before the test runs. # Also test only runs as root. diff --git a/test/root/testdata/BUILD b/test/root/testdata/BUILD index 125633680..bca5f9cab 100644 --- a/test/root/testdata/BUILD +++ b/test/root/testdata/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -12,7 +12,6 @@ go_library( "sandbox.go", "simple.go", ], - importpath = "gvisor.dev/gvisor/test/root/testdata", visibility = [ "//visibility:public", ], diff --git a/test/runtimes/BUILD b/test/runtimes/BUILD index 367295206..2c472bf8d 100644 --- a/test/runtimes/BUILD +++ b/test/runtimes/BUILD @@ -1,6 +1,6 @@ # These packages are used to run language runtime tests inside gVisor sandboxes. -load("@io_bazel_rules_go//go:def.bzl", "go_binary", "go_test") +load("//tools:defs.bzl", "go_binary", "go_test") load("//test/runtimes:build_defs.bzl", "runtime_test") package(licenses = ["notice"]) @@ -49,5 +49,5 @@ go_test( name = "blacklist_test", size = "small", srcs = ["blacklist_test.go"], - embed = [":runner"], + library = ":runner", ) diff --git a/test/runtimes/build_defs.bzl b/test/runtimes/build_defs.bzl index 6f84ca852..92e275a76 100644 --- a/test/runtimes/build_defs.bzl +++ b/test/runtimes/build_defs.bzl @@ -1,6 +1,6 @@ """Defines a rule for runtime test targets.""" -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_test", "loopback") def runtime_test( name, @@ -34,6 +34,7 @@ def runtime_test( ] data = [ ":runner", + loopback, ] if blacklist_file: args += ["--blacklist_file", "test/runtimes/" + blacklist_file] @@ -61,7 +62,7 @@ def blacklist_test(name, blacklist_file): """Test that a blacklist parses correctly.""" go_test( name = name + "_blacklist_test", - embed = [":runner"], + library = ":runner", srcs = ["blacklist_test.go"], args = ["--blacklist_file", "test/runtimes/" + blacklist_file], data = [blacklist_file], diff --git a/test/runtimes/images/proctor/BUILD b/test/runtimes/images/proctor/BUILD index 09dc6c42f..85e004c45 100644 --- a/test/runtimes/images/proctor/BUILD +++ b/test/runtimes/images/proctor/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_binary", "go_test") +load("//tools:defs.bzl", "go_binary", "go_test") package(licenses = ["notice"]) @@ -19,7 +19,7 @@ go_test( name = "proctor_test", size = "small", srcs = ["proctor_test.go"], - embed = [":proctor"], + library = ":proctor", deps = [ "//runsc/testutil", ], diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 90d52e73b..40e974314 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_binary") +load("//tools:defs.bzl", "go_binary") load("//test/syscalls:build_defs.bzl", "syscall_test") package(licenses = ["notice"]) diff --git a/test/syscalls/build_defs.bzl b/test/syscalls/build_defs.bzl index aaf77c65b..1df761dd0 100644 --- a/test/syscalls/build_defs.bzl +++ b/test/syscalls/build_defs.bzl @@ -1,5 +1,7 @@ """Defines a rule for syscall test targets.""" +load("//tools:defs.bzl", "loopback") + # syscall_test is a macro that will create targets to run the given test target # on the host (native) and runsc. def syscall_test( @@ -135,6 +137,7 @@ def _syscall_test( name = name, data = [ ":syscall_test_runner", + loopback, test, ], args = args, @@ -148,6 +151,3 @@ def sh_test(**kwargs): native.sh_test( **kwargs ) - -def select_for_linux(for_linux, for_others = []): - return for_linux diff --git a/test/syscalls/gtest/BUILD b/test/syscalls/gtest/BUILD index 9293f25cb..de4b2727c 100644 --- a/test/syscalls/gtest/BUILD +++ b/test/syscalls/gtest/BUILD @@ -1,12 +1,9 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) go_library( name = "gtest", srcs = ["gtest.go"], - importpath = "gvisor.dev/gvisor/test/syscalls/gtest", - visibility = [ - "//test:__subpackages__", - ], + visibility = ["//:sandbox"], ) diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 4c7ec3f06..c2ef50c1d 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -1,5 +1,4 @@ -load("@rules_cc//cc:defs.bzl", "cc_binary", "cc_library") -load("//test/syscalls:build_defs.bzl", "select_for_linux") +load("//tools:defs.bzl", "cc_binary", "cc_library", "default_net_util", "select_system") package( default_visibility = ["//:sandbox"], @@ -126,13 +125,11 @@ cc_library( testonly = 1, srcs = [ "socket_test_util.cc", - ] + select_for_linux( - [ - "socket_test_util_impl.cc", - ], - ), + "socket_test_util_impl.cc", + ], hdrs = ["socket_test_util.h"], - deps = [ + defines = select_system(), + deps = default_net_util() + [ "@com_google_googletest//:gtest", "@com_google_absl//absl/memory", "@com_google_absl//absl/strings", @@ -143,8 +140,7 @@ cc_library( "//test/util:temp_path", "//test/util:test_util", "//test/util:thread_util", - ] + select_for_linux([ - ]), + ], ) cc_library( @@ -1443,6 +1439,7 @@ cc_binary( srcs = ["arch_prctl.cc"], linkstatic = 1, deps = [ + "//test/util:file_descriptor", "//test/util:test_main", "//test/util:test_util", "@com_google_googletest//:gtest", @@ -3383,11 +3380,11 @@ cc_library( name = "udp_socket_test_cases", testonly = 1, srcs = [ - "udp_socket_test_cases.cc", - ] + select_for_linux([ "udp_socket_errqueue_test_case.cc", - ]), + "udp_socket_test_cases.cc", + ], hdrs = ["udp_socket_test_cases.h"], + defines = select_system(), deps = [ ":socket_test_util", ":unix_domain_socket_test_util", diff --git a/test/syscalls/linux/arch_prctl.cc b/test/syscalls/linux/arch_prctl.cc index 81bf5a775..3a901faf5 100644 --- a/test/syscalls/linux/arch_prctl.cc +++ b/test/syscalls/linux/arch_prctl.cc @@ -14,8 +14,10 @@ #include #include +#include #include "gtest/gtest.h" +#include "test/util/file_descriptor.h" #include "test/util/test_util.h" // glibc does not provide a prototype for arch_prctl() so declare it here. diff --git a/test/syscalls/linux/rseq/BUILD b/test/syscalls/linux/rseq/BUILD index 5cfe4e56f..ed488dbc2 100644 --- a/test/syscalls/linux/rseq/BUILD +++ b/test/syscalls/linux/rseq/BUILD @@ -1,8 +1,7 @@ # This package contains a standalone rseq test binary. This binary must not # depend on libc, which might use rseq itself. -load("@bazel_tools//tools/cpp:cc_flags_supplier.bzl", "cc_flags_supplier") -load("@rules_cc//cc:defs.bzl", "cc_library") +load("//tools:defs.bzl", "cc_flags_supplier", "cc_library", "cc_toolchain") package(licenses = ["notice"]) @@ -37,8 +36,8 @@ genrule( "$(location start.S)", ]), toolchains = [ + cc_toolchain, ":no_pie_cc_flags", - "@bazel_tools//tools/cpp:current_cc_toolchain", ], visibility = ["//:sandbox"], ) diff --git a/test/syscalls/linux/udp_socket_errqueue_test_case.cc b/test/syscalls/linux/udp_socket_errqueue_test_case.cc index 147978f46..9a24e1df0 100644 --- a/test/syscalls/linux/udp_socket_errqueue_test_case.cc +++ b/test/syscalls/linux/udp_socket_errqueue_test_case.cc @@ -12,6 +12,8 @@ // See the License for the specific language governing permissions and // limitations under the License. +#ifndef __fuchsia__ + #include "test/syscalls/linux/udp_socket_test_cases.h" #include @@ -52,3 +54,5 @@ TEST_P(UdpSocketTest, ErrorQueue) { } // namespace testing } // namespace gvisor + +#endif // __fuchsia__ diff --git a/test/uds/BUILD b/test/uds/BUILD index a3843e699..51e2c7ce8 100644 --- a/test/uds/BUILD +++ b/test/uds/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package( default_visibility = ["//:sandbox"], @@ -9,7 +9,6 @@ go_library( name = "uds", testonly = 1, srcs = ["uds.go"], - importpath = "gvisor.dev/gvisor/test/uds", deps = [ "//pkg/log", "//pkg/unet", diff --git a/test/util/BUILD b/test/util/BUILD index cbc728159..3c732be62 100644 --- a/test/util/BUILD +++ b/test/util/BUILD @@ -1,5 +1,4 @@ -load("@rules_cc//cc:defs.bzl", "cc_library", "cc_test") -load("//test/syscalls:build_defs.bzl", "select_for_linux") +load("//tools:defs.bzl", "cc_library", "cc_test", "select_system") package( default_visibility = ["//:sandbox"], @@ -142,12 +141,13 @@ cc_library( cc_library( name = "save_util", testonly = 1, - srcs = ["save_util.cc"] + - select_for_linux( - ["save_util_linux.cc"], - ["save_util_other.cc"], - ), + srcs = [ + "save_util.cc", + "save_util_linux.cc", + "save_util_other.cc", + ], hdrs = ["save_util.h"], + defines = select_system(), ) cc_library( @@ -234,13 +234,16 @@ cc_library( testonly = 1, srcs = [ "test_util.cc", - ] + select_for_linux( - [ - "test_util_impl.cc", - "test_util_runfiles.cc", + "test_util_impl.cc", + "test_util_runfiles.cc", + ], + hdrs = ["test_util.h"], + defines = select_system( + fuchsia = [ + "__opensource__", + "__fuchsia__", ], ), - hdrs = ["test_util.h"], deps = [ ":fs_util", ":logging", diff --git a/test/util/save_util_linux.cc b/test/util/save_util_linux.cc index cd56118c0..d0aea8e6a 100644 --- a/test/util/save_util_linux.cc +++ b/test/util/save_util_linux.cc @@ -12,6 +12,8 @@ // See the License for the specific language governing permissions and // limitations under the License. +#ifdef __linux__ + #include #include #include @@ -43,3 +45,5 @@ void MaybeSave() { } // namespace testing } // namespace gvisor + +#endif diff --git a/test/util/save_util_other.cc b/test/util/save_util_other.cc index 1aca663b7..931af2c29 100644 --- a/test/util/save_util_other.cc +++ b/test/util/save_util_other.cc @@ -12,6 +12,8 @@ // See the License for the specific language governing permissions and // limitations under the License. +#ifndef __linux__ + namespace gvisor { namespace testing { @@ -21,3 +23,5 @@ void MaybeSave() { } // namespace testing } // namespace gvisor + +#endif diff --git a/test/util/test_util_runfiles.cc b/test/util/test_util_runfiles.cc index 7210094eb..694d21692 100644 --- a/test/util/test_util_runfiles.cc +++ b/test/util/test_util_runfiles.cc @@ -12,6 +12,8 @@ // See the License for the specific language governing permissions and // limitations under the License. +#ifndef __fuchsia__ + #include #include @@ -44,3 +46,5 @@ std::string RunfilePath(std::string path) { } // namespace testing } // namespace gvisor + +#endif // __fuchsia__ diff --git a/tools/BUILD b/tools/BUILD new file mode 100644 index 000000000..e73a9c885 --- /dev/null +++ b/tools/BUILD @@ -0,0 +1,3 @@ +package(licenses = ["notice"]) + +exports_files(["nogo.js"]) diff --git a/tools/build/BUILD b/tools/build/BUILD new file mode 100644 index 000000000..0c0ce3f4d --- /dev/null +++ b/tools/build/BUILD @@ -0,0 +1,10 @@ +package(licenses = ["notice"]) + +# In bazel, no special support is required for loopback networking. This is +# just a dummy data target that does not change the test environment. +genrule( + name = "loopback", + outs = ["loopback.txt"], + cmd = "touch $@", + visibility = ["//visibility:public"], +) diff --git a/tools/build/defs.bzl b/tools/build/defs.bzl new file mode 100644 index 000000000..d0556abd1 --- /dev/null +++ b/tools/build/defs.bzl @@ -0,0 +1,91 @@ +"""Bazel implementations of standard rules.""" + +load("@bazel_tools//tools/cpp:cc_flags_supplier.bzl", _cc_flags_supplier = "cc_flags_supplier") +load("@io_bazel_rules_go//go:def.bzl", _go_binary = "go_binary", _go_embed_data = "go_embed_data", _go_library = "go_library", _go_test = "go_test", _go_tool_library = "go_tool_library") +load("@io_bazel_rules_go//proto:def.bzl", _go_proto_library = "go_proto_library") +load("@rules_cc//cc:defs.bzl", _cc_binary = "cc_binary", _cc_library = "cc_library", _cc_proto_library = "cc_proto_library", _cc_test = "cc_test") +load("@rules_pkg//:pkg.bzl", _pkg_deb = "pkg_deb", _pkg_tar = "pkg_tar") +load("@io_bazel_rules_docker//go:image.bzl", _go_image = "go_image") +load("@io_bazel_rules_docker//container:container.bzl", _container_image = "container_image") +load("@pydeps//:requirements.bzl", _py_requirement = "requirement") + +container_image = _container_image +cc_binary = _cc_binary +cc_library = _cc_library +cc_flags_supplier = _cc_flags_supplier +cc_proto_library = _cc_proto_library +cc_test = _cc_test +cc_toolchain = "@bazel_tools//tools/cpp:current_cc_toolchain" +go_image = _go_image +go_embed_data = _go_embed_data +loopback = "//tools/build:loopback" +proto_library = native.proto_library +pkg_deb = _pkg_deb +pkg_tar = _pkg_tar +py_library = native.py_library +py_binary = native.py_binary +py_test = native.py_test + +def go_binary(name, static = False, pure = False, **kwargs): + if static: + kwargs["static"] = "on" + if pure: + kwargs["pure"] = "on" + _go_binary( + name = name, + **kwargs + ) + +def go_library(name, **kwargs): + _go_library( + name = name, + importpath = "gvisor.dev/gvisor/" + native.package_name(), + **kwargs + ) + +def go_tool_library(name, **kwargs): + _go_tool_library( + name = name, + importpath = "gvisor.dev/gvisor/" + native.package_name(), + **kwargs + ) + +def go_proto_library(name, proto, **kwargs): + deps = kwargs.pop("deps", []) + _go_proto_library( + name = name, + importpath = "gvisor.dev/gvisor/" + native.package_name() + "/" + name, + proto = proto, + deps = [dep.replace("_proto", "_go_proto") for dep in deps], + **kwargs + ) + +def go_test(name, **kwargs): + library = kwargs.pop("library", None) + if library: + kwargs["embed"] = [library] + _go_test( + name = name, + **kwargs + ) + +def py_requirement(name, direct = False): + return _py_requirement(name) + +def select_arch(amd64 = "amd64", arm64 = "arm64", default = None, **kwargs): + values = { + "@bazel_tools//src/conditions:linux_x86_64": amd64, + "@bazel_tools//src/conditions:linux_aarch64": arm64, + } + if default: + values["//conditions:default"] = default + return select(values, **kwargs) + +def select_system(linux = ["__linux__"], **kwargs): + return linux # Only Linux supported. + +def default_installer(): + return None + +def default_net_util(): + return [] # Nothing needed. diff --git a/tools/checkunsafe/BUILD b/tools/checkunsafe/BUILD index d85c56131..92ba8ab06 100644 --- a/tools/checkunsafe/BUILD +++ b/tools/checkunsafe/BUILD @@ -1,11 +1,10 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_tool_library") +load("//tools:defs.bzl", "go_tool_library") package(licenses = ["notice"]) go_tool_library( name = "checkunsafe", srcs = ["check_unsafe.go"], - importpath = "checkunsafe", visibility = ["//visibility:public"], deps = [ "@org_golang_x_tools//go/analysis:go_tool_library", diff --git a/tools/defs.bzl b/tools/defs.bzl new file mode 100644 index 000000000..819f12b0d --- /dev/null +++ b/tools/defs.bzl @@ -0,0 +1,154 @@ +"""Wrappers for common build rules. + +These wrappers apply common BUILD configurations (e.g., proto_library +automagically creating cc_ and go_ proto targets) and act as a single point of +change for Google-internal and bazel-compatible rules. +""" + +load("//tools/go_stateify:defs.bzl", "go_stateify") +load("//tools/go_marshal:defs.bzl", "go_marshal", "marshal_deps", "marshal_test_deps") +load("//tools/build:defs.bzl", _cc_binary = "cc_binary", _cc_flags_supplier = "cc_flags_supplier", _cc_library = "cc_library", _cc_proto_library = "cc_proto_library", _cc_test = "cc_test", _cc_toolchain = "cc_toolchain", _container_image = "container_image", _default_installer = "default_installer", _default_net_util = "default_net_util", _go_binary = "go_binary", _go_embed_data = "go_embed_data", _go_image = "go_image", _go_library = "go_library", _go_proto_library = "go_proto_library", _go_test = "go_test", _go_tool_library = "go_tool_library", _loopback = "loopback", _pkg_deb = "pkg_deb", _pkg_tar = "pkg_tar", _proto_library = "proto_library", _py_binary = "py_binary", _py_library = "py_library", _py_requirement = "py_requirement", _py_test = "py_test", _select_arch = "select_arch", _select_system = "select_system") + +# Delegate directly. +cc_binary = _cc_binary +cc_library = _cc_library +cc_test = _cc_test +cc_toolchain = _cc_toolchain +cc_flags_supplier = _cc_flags_supplier +container_image = _container_image +go_embed_data = _go_embed_data +go_image = _go_image +go_test = _go_test +go_tool_library = _go_tool_library +pkg_deb = _pkg_deb +pkg_tar = _pkg_tar +py_library = _py_library +py_binary = _py_binary +py_test = _py_test +py_requirement = _py_requirement +select_arch = _select_arch +select_system = _select_system +loopback = _loopback +default_installer = _default_installer +default_net_util = _default_net_util + +def go_binary(name, **kwargs): + """Wraps the standard go_binary. + + Args: + name: the rule name. + **kwargs: standard go_binary arguments. + """ + _go_binary( + name = name, + **kwargs + ) + +def go_library(name, srcs, deps = [], imports = [], stateify = True, marshal = False, **kwargs): + """Wraps the standard go_library and does stateification and marshalling. + + The recommended way is to use this rule with mostly identical configuration as the native + go_library rule. + + These definitions provide additional flags (stateify, marshal) that can be used + with the generators to automatically supplement the library code. + + load("//tools:defs.bzl", "go_library") + + go_library( + name = "foo", + srcs = ["foo.go"], + ) + + Args: + name: the rule name. + srcs: the library sources. + deps: the library dependencies. + imports: imports required for stateify. + stateify: whether statify is enabled (default: true). + marshal: whether marshal is enabled (default: false). + **kwargs: standard go_library arguments. + """ + if stateify: + # Only do stateification for non-state packages without manual autogen. + go_stateify( + name = name + "_state_autogen", + srcs = [src for src in srcs if src.endswith(".go")], + imports = imports, + package = name, + arch = select_arch(), + out = name + "_state_autogen.go", + ) + all_srcs = srcs + [name + "_state_autogen.go"] + if "//pkg/state" not in deps: + all_deps = deps + ["//pkg/state"] + else: + all_deps = deps + else: + all_deps = deps + all_srcs = srcs + if marshal: + go_marshal( + name = name + "_abi_autogen", + srcs = [src for src in srcs if src.endswith(".go")], + debug = False, + imports = imports, + package = name, + ) + extra_deps = [ + dep + for dep in marshal_deps + if not dep in all_deps + ] + all_deps = all_deps + extra_deps + all_srcs = srcs + [name + "_abi_autogen_unsafe.go"] + + _go_library( + name = name, + srcs = all_srcs, + deps = all_deps, + **kwargs + ) + + if marshal: + # Ignore importpath for go_test. + kwargs.pop("importpath", None) + + _go_test( + name = name + "_abi_autogen_test", + srcs = [name + "_abi_autogen_test.go"], + library = ":" + name, + deps = marshal_test_deps, + **kwargs + ) + +def proto_library(name, srcs, **kwargs): + """Wraps the standard proto_library. + + Given a proto_library named "foo", this produces three different targets: + - foo_proto: proto_library rule. + - foo_go_proto: go_proto_library rule. + - foo_cc_proto: cc_proto_library rule. + + Args: + srcs: the proto sources. + **kwargs: standard proto_library arguments. + """ + deps = kwargs.pop("deps", []) + _proto_library( + name = name + "_proto", + srcs = srcs, + deps = deps, + **kwargs + ) + _go_proto_library( + name = name + "_go_proto", + proto = ":" + name + "_proto", + deps = deps, + **kwargs + ) + _cc_proto_library( + name = name + "_cc_proto", + deps = [":" + name + "_proto"], + **kwargs + ) diff --git a/tools/go_generics/BUILD b/tools/go_generics/BUILD index 39318b877..069df3856 100644 --- a/tools/go_generics/BUILD +++ b/tools/go_generics/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_binary") +load("//tools:defs.bzl", "go_binary") package(licenses = ["notice"]) diff --git a/tools/go_generics/globals/BUILD b/tools/go_generics/globals/BUILD index 74853c7d2..38caa3ce7 100644 --- a/tools/go_generics/globals/BUILD +++ b/tools/go_generics/globals/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -8,6 +8,6 @@ go_library( "globals_visitor.go", "scope.go", ], - importpath = "gvisor.dev/gvisor/tools/go_generics/globals", + stateify = False, visibility = ["//tools/go_generics:__pkg__"], ) diff --git a/tools/go_generics/go_merge/BUILD b/tools/go_generics/go_merge/BUILD index 02b09120e..b7d35e272 100644 --- a/tools/go_generics/go_merge/BUILD +++ b/tools/go_generics/go_merge/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_binary") +load("//tools:defs.bzl", "go_binary") package(licenses = ["notice"]) diff --git a/tools/go_generics/rules_tests/BUILD b/tools/go_generics/rules_tests/BUILD index 9d26a88b7..8a329dfc6 100644 --- a/tools/go_generics/rules_tests/BUILD +++ b/tools/go_generics/rules_tests/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_test") load("//tools/go_generics:defs.bzl", "go_template", "go_template_instance") package(licenses = ["notice"]) diff --git a/tools/go_marshal/BUILD b/tools/go_marshal/BUILD index c862b277c..80d9c0504 100644 --- a/tools/go_marshal/BUILD +++ b/tools/go_marshal/BUILD @@ -1,6 +1,6 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_binary") +load("//tools:defs.bzl", "go_binary") -package(licenses = ["notice"]) +licenses(["notice"]) go_binary( name = "go_marshal", diff --git a/tools/go_marshal/README.md b/tools/go_marshal/README.md index 481575bd3..4886efddf 100644 --- a/tools/go_marshal/README.md +++ b/tools/go_marshal/README.md @@ -20,19 +20,7 @@ comment `// +marshal`. # Usage -See `defs.bzl`: two new rules are provided, `go_marshal` and `go_library`. - -The recommended way to generate a go library with marshalling is to use the -`go_library` with mostly identical configuration as the native go_library rule. - -``` -load("/gvisor/tools/go_marshal:defs.bzl", "go_library") - -go_library( - name = "foo", - srcs = ["foo.go"], -) -``` +See `defs.bzl`: a new rule is provided, `go_marshal`. Under the hood, the `go_marshal` rule is used to generate a file that will appear in a Go target; the output file should appear explicitly in a srcs list. @@ -54,11 +42,7 @@ go_library( "foo.go", "foo_abi.go", ], - deps = [ - "/gvisor/pkg/abi", - "/gvisor/pkg/sentry/safemem/safemem", - "/gvisor/pkg/sentry/usermem/usermem", - ], + ... ) ``` @@ -69,22 +53,6 @@ These tests use reflection to verify properties of the ABI struct, and should be considered part of the generated interfaces (but are too expensive to execute at runtime). Ensure these tests run at some point. -``` -$ cat BUILD -load("/gvisor/tools/go_marshal:defs.bzl", "go_library") - -go_library( - name = "foo", - srcs = ["foo.go"], -) -$ blaze build :foo -$ blaze query ... -:foo_abi_autogen -:foo_abi_autogen_test -$ blaze test :foo_abi_autogen_test - -``` - # Restrictions Not all valid go type definitions can be used with `go_marshal`. `go_marshal` is @@ -131,22 +99,6 @@ for embedded structs that are not aligned. Because of this, it's generally best to avoid using `marshal:"unaligned"` and insert explicit padding fields instead. -## Debugging go_marshal - -To enable debugging output from the go marshal tool, pass the `-debug` flag to -the tool. When using the build rules from above, add a `debug = True` field to -the build rule like this: - -``` -load("/gvisor/tools/go_marshal:defs.bzl", "go_library") - -go_library( - name = "foo", - srcs = ["foo.go"], - debug = True, -) -``` - ## Modifying the `go_marshal` Tool The following are some guidelines for modifying the `go_marshal` tool: diff --git a/tools/go_marshal/analysis/BUILD b/tools/go_marshal/analysis/BUILD index c859ced77..c2a4d45c4 100644 --- a/tools/go_marshal/analysis/BUILD +++ b/tools/go_marshal/analysis/BUILD @@ -1,12 +1,11 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library") +load("//tools:defs.bzl", "go_library") -package(licenses = ["notice"]) +licenses(["notice"]) go_library( name = "analysis", testonly = 1, srcs = ["analysis_unsafe.go"], - importpath = "gvisor.dev/gvisor/tools/go_marshal/analysis", visibility = [ "//:sandbox", ], diff --git a/tools/go_marshal/defs.bzl b/tools/go_marshal/defs.bzl index c32eb559f..2918ceffe 100644 --- a/tools/go_marshal/defs.bzl +++ b/tools/go_marshal/defs.bzl @@ -1,57 +1,14 @@ -"""Marshal is a tool for generating marshalling interfaces for Go types. - -The recommended way is to use the go_library rule defined below with mostly -identical configuration as the native go_library rule. - -load("//tools/go_marshal:defs.bzl", "go_library") - -go_library( - name = "foo", - srcs = ["foo.go"], -) - -Under the hood, the go_marshal rule is used to generate a file that will -appear in a Go target; the output file should appear explicitly in a srcs list. -For example (the above is still the preferred way): - -load("//tools/go_marshal:defs.bzl", "go_marshal") - -go_marshal( - name = "foo_abi", - srcs = ["foo.go"], - out = "foo_abi.go", - package = "foo", -) - -go_library( - name = "foo", - srcs = [ - "foo.go", - "foo_abi.go", - ], - deps = [ - "//tools/go_marshal:marshal", - "//pkg/sentry/platform/safecopy", - "//pkg/sentry/usermem", - ], -) -""" - -load("@io_bazel_rules_go//go:def.bzl", _go_library = "go_library", _go_test = "go_test") +"""Marshal is a tool for generating marshalling interfaces for Go types.""" def _go_marshal_impl(ctx): """Execute the go_marshal tool.""" output = ctx.outputs.lib output_test = ctx.outputs.test - (build_dir, _, _) = ctx.build_file_path.rpartition("/BUILD") - - decl = "/".join(["gvisor.dev/gvisor", build_dir]) # Run the marshal command. args = ["-output=%s" % output.path] args += ["-pkg=%s" % ctx.attr.package] args += ["-output_test=%s" % output_test.path] - args += ["-declarationPkg=%s" % decl] if ctx.attr.debug: args += ["-debug"] @@ -83,7 +40,6 @@ go_marshal = rule( implementation = _go_marshal_impl, attrs = { "srcs": attr.label_list(mandatory = True, allow_files = True), - "libname": attr.string(mandatory = True), "imports": attr.string_list(mandatory = False), "package": attr.string(mandatory = True), "debug": attr.bool(doc = "enable debugging output from the go_marshal tool"), @@ -95,58 +51,14 @@ go_marshal = rule( }, ) -def go_library(name, srcs, deps = [], imports = [], debug = False, **kwargs): - """wraps the standard go_library and does mashalling interface generation. - - Args: - name: Same as native go_library. - srcs: Same as native go_library. - deps: Same as native go_library. - imports: Extra import paths to pass to the go_marshal tool. - debug: Enables debugging output from the go_marshal tool. - **kwargs: Remaining args to pass to the native go_library rule unmodified. - """ - go_marshal( - name = name + "_abi_autogen", - libname = name, - srcs = [src for src in srcs if src.endswith(".go")], - debug = debug, - imports = imports, - package = name, - ) - - extra_deps = [ - "//tools/go_marshal/marshal", - "//pkg/sentry/platform/safecopy", - "//pkg/sentry/usermem", - ] - - all_srcs = srcs + [name + "_abi_autogen_unsafe.go"] - all_deps = deps + [] # + extra_deps - - for extra in extra_deps: - if extra not in deps: - all_deps.append(extra) - - _go_library( - name = name, - srcs = all_srcs, - deps = all_deps, - **kwargs - ) - - # Don't pass importpath arg to go_test. - kwargs.pop("importpath", "") - - _go_test( - name = name + "_abi_autogen_test", - srcs = [name + "_abi_autogen_test.go"], - # Generated test has a fixed set of dependencies since we generate these - # tests. They should only depend on the library generated above, and the - # Marshallable interface. - deps = [ - ":" + name, - "//tools/go_marshal/analysis", - ], - **kwargs - ) +# marshal_deps are the dependencies requied by generated code. +marshal_deps = [ + "//tools/go_marshal/marshal", + "//pkg/sentry/platform/safecopy", + "//pkg/sentry/usermem", +] + +# marshal_test_deps are required by test targets. +marshal_test_deps = [ + "//tools/go_marshal/analysis", +] diff --git a/tools/go_marshal/gomarshal/BUILD b/tools/go_marshal/gomarshal/BUILD index a0eae6492..c92b59dd6 100644 --- a/tools/go_marshal/gomarshal/BUILD +++ b/tools/go_marshal/gomarshal/BUILD @@ -1,6 +1,6 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library") +load("//tools:defs.bzl", "go_library") -package(licenses = ["notice"]) +licenses(["notice"]) go_library( name = "gomarshal", @@ -10,7 +10,7 @@ go_library( "generator_tests.go", "util.go", ], - importpath = "gvisor.dev/gvisor/tools/go_marshal/gomarshal", + stateify = False, visibility = [ "//:sandbox", ], diff --git a/tools/go_marshal/gomarshal/generator.go b/tools/go_marshal/gomarshal/generator.go index 641ccd938..8392f3f6d 100644 --- a/tools/go_marshal/gomarshal/generator.go +++ b/tools/go_marshal/gomarshal/generator.go @@ -62,15 +62,12 @@ type Generator struct { outputTest *os.File // Package name for the generated file. pkg string - // Go import path for package we're processing. This package should directly - // declare the type we're generating code for. - declaration string // Set of extra packages to import in the generated file. imports *importTable } // NewGenerator creates a new code Generator. -func NewGenerator(srcs []string, out, outTest, pkg, declaration string, imports []string) (*Generator, error) { +func NewGenerator(srcs []string, out, outTest, pkg string, imports []string) (*Generator, error) { f, err := os.OpenFile(out, os.O_CREATE|os.O_WRONLY|os.O_TRUNC, 0644) if err != nil { return nil, fmt.Errorf("Couldn't open output file %q: %v", out, err) @@ -80,12 +77,11 @@ func NewGenerator(srcs []string, out, outTest, pkg, declaration string, imports return nil, fmt.Errorf("Couldn't open test output file %q: %v", out, err) } g := Generator{ - inputs: srcs, - output: f, - outputTest: fTest, - pkg: pkg, - declaration: declaration, - imports: newImportTable(), + inputs: srcs, + output: f, + outputTest: fTest, + pkg: pkg, + imports: newImportTable(), } for _, i := range imports { // All imports on the extra imports list are unconditionally marked as @@ -264,7 +260,7 @@ func (g *Generator) generateOne(t *ast.TypeSpec, fset *token.FileSet) *interface // generateOneTestSuite generates a test suite for the automatically generated // implementations type t. func (g *Generator) generateOneTestSuite(t *ast.TypeSpec) *testGenerator { - i := newTestGenerator(t, g.declaration) + i := newTestGenerator(t) i.emitTests() return i } @@ -359,7 +355,7 @@ func (g *Generator) Run() error { // source file. func (g *Generator) writeTests(ts []*testGenerator) error { var b sourceBuffer - b.emit("package %s_test\n\n", g.pkg) + b.emit("package %s\n\n", g.pkg) if err := b.write(g.outputTest); err != nil { return err } diff --git a/tools/go_marshal/gomarshal/generator_tests.go b/tools/go_marshal/gomarshal/generator_tests.go index df25cb5b2..bcda17c3b 100644 --- a/tools/go_marshal/gomarshal/generator_tests.go +++ b/tools/go_marshal/gomarshal/generator_tests.go @@ -46,7 +46,7 @@ type testGenerator struct { decl *importStmt } -func newTestGenerator(t *ast.TypeSpec, declaration string) *testGenerator { +func newTestGenerator(t *ast.TypeSpec) *testGenerator { if _, ok := t.Type.(*ast.StructType); !ok { panic(fmt.Sprintf("Attempting to generate code for a not struct type %v", t)) } @@ -59,14 +59,12 @@ func newTestGenerator(t *ast.TypeSpec, declaration string) *testGenerator { for _, i := range standardImports { g.imports.add(i).markUsed() } - g.decl = g.imports.add(declaration) - g.decl.markUsed() return g } func (g *testGenerator) typeName() string { - return fmt.Sprintf("%s.%s", g.decl.name, g.t.Name.Name) + return g.t.Name.Name } func (g *testGenerator) forEachField(fn func(f *ast.Field)) { diff --git a/tools/go_marshal/main.go b/tools/go_marshal/main.go index 3d12eb93c..e1a97b311 100644 --- a/tools/go_marshal/main.go +++ b/tools/go_marshal/main.go @@ -31,11 +31,10 @@ import ( ) var ( - pkg = flag.String("pkg", "", "output package") - output = flag.String("output", "", "output file") - outputTest = flag.String("output_test", "", "output file for tests") - imports = flag.String("imports", "", "comma-separated list of extra packages to import in generated code") - declarationPkg = flag.String("declarationPkg", "", "import path of target declaring the types we're generating on") + pkg = flag.String("pkg", "", "output package") + output = flag.String("output", "", "output file") + outputTest = flag.String("output_test", "", "output file for tests") + imports = flag.String("imports", "", "comma-separated list of extra packages to import in generated code") ) func main() { @@ -62,7 +61,7 @@ func main() { // as an import. extraImports = strings.Split(*imports, ",") } - g, err := gomarshal.NewGenerator(flag.Args(), *output, *outputTest, *pkg, *declarationPkg, extraImports) + g, err := gomarshal.NewGenerator(flag.Args(), *output, *outputTest, *pkg, extraImports) if err != nil { panic(err) } diff --git a/tools/go_marshal/marshal/BUILD b/tools/go_marshal/marshal/BUILD index 47dda97a1..ad508c72f 100644 --- a/tools/go_marshal/marshal/BUILD +++ b/tools/go_marshal/marshal/BUILD @@ -1,13 +1,12 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library") +load("//tools:defs.bzl", "go_library") -package(licenses = ["notice"]) +licenses(["notice"]) go_library( name = "marshal", srcs = [ "marshal.go", ], - importpath = "gvisor.dev/gvisor/tools/go_marshal/marshal", visibility = [ "//:sandbox", ], diff --git a/tools/go_marshal/test/BUILD b/tools/go_marshal/test/BUILD index d412e1ccf..38ba49fed 100644 --- a/tools/go_marshal/test/BUILD +++ b/tools/go_marshal/test/BUILD @@ -1,7 +1,6 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") -load("//tools/go_marshal:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") -package(licenses = ["notice"]) +licenses(["notice"]) package_group( name = "gomarshal_test", @@ -25,6 +24,6 @@ go_library( name = "test", testonly = 1, srcs = ["test.go"], - importpath = "gvisor.dev/gvisor/tools/go_marshal/test", + marshal = True, deps = ["//tools/go_marshal/test/external"], ) diff --git a/tools/go_marshal/test/external/BUILD b/tools/go_marshal/test/external/BUILD index 9bb89e1da..0cf6da603 100644 --- a/tools/go_marshal/test/external/BUILD +++ b/tools/go_marshal/test/external/BUILD @@ -1,11 +1,11 @@ -load("//tools/go_marshal:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library") -package(licenses = ["notice"]) +licenses(["notice"]) go_library( name = "external", testonly = 1, srcs = ["external.go"], - importpath = "gvisor.dev/gvisor/tools/go_marshal/test/external", + marshal = True, visibility = ["//tools/go_marshal/test:gomarshal_test"], ) diff --git a/tools/go_stateify/BUILD b/tools/go_stateify/BUILD index bb53f8ae9..a133d6f8b 100644 --- a/tools/go_stateify/BUILD +++ b/tools/go_stateify/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_binary") +load("//tools:defs.bzl", "go_binary") package(licenses = ["notice"]) diff --git a/tools/go_stateify/defs.bzl b/tools/go_stateify/defs.bzl index 33267c074..0f261d89f 100644 --- a/tools/go_stateify/defs.bzl +++ b/tools/go_stateify/defs.bzl @@ -1,41 +1,4 @@ -"""Stateify is a tool for generating state wrappers for Go types. - -The recommended way is to use the go_library rule defined below with mostly -identical configuration as the native go_library rule. - -load("//tools/go_stateify:defs.bzl", "go_library") - -go_library( - name = "foo", - srcs = ["foo.go"], -) - -Under the hood, the go_stateify rule is used to generate a file that will -appear in a Go target; the output file should appear explicitly in a srcs list. -For example (the above is still the preferred way): - -load("//tools/go_stateify:defs.bzl", "go_stateify") - -go_stateify( - name = "foo_state", - srcs = ["foo.go"], - out = "foo_state.go", - package = "foo", -) - -go_library( - name = "foo", - srcs = [ - "foo.go", - "foo_state.go", - ], - deps = [ - "//pkg/state", - ], -) -""" - -load("@io_bazel_rules_go//go:def.bzl", _go_library = "go_library") +"""Stateify is a tool for generating state wrappers for Go types.""" def _go_stateify_impl(ctx): """Implementation for the stateify tool.""" @@ -103,43 +66,3 @@ files and must be added to the srcs of the relevant go_library. "_statepkg": attr.string(default = "gvisor.dev/gvisor/pkg/state"), }, ) - -def go_library(name, srcs, deps = [], imports = [], **kwargs): - """Standard go_library wrapped which generates state source files. - - Args: - name: the name of the go_library rule. - srcs: sources of the go_library. Each will be processed for stateify - annotations. - deps: dependencies for the go_library. - imports: an optional list of extra non-aliased, Go-style absolute import - paths required for stateified types. - **kwargs: passed to go_library. - """ - if "encode_unsafe.go" not in srcs and (name + "_state_autogen.go") not in srcs: - # Only do stateification for non-state packages without manual autogen. - go_stateify( - name = name + "_state_autogen", - srcs = [src for src in srcs if src.endswith(".go")], - imports = imports, - package = name, - arch = select({ - "@bazel_tools//src/conditions:linux_aarch64": "arm64", - "//conditions:default": "amd64", - }), - out = name + "_state_autogen.go", - ) - all_srcs = srcs + [name + "_state_autogen.go"] - if "//pkg/state" not in deps: - all_deps = deps + ["//pkg/state"] - else: - all_deps = deps - else: - all_deps = deps - all_srcs = srcs - _go_library( - name = name, - srcs = all_srcs, - deps = all_deps, - **kwargs - ) diff --git a/tools/images/BUILD b/tools/images/BUILD index 2b77c2737..f1699b184 100644 --- a/tools/images/BUILD +++ b/tools/images/BUILD @@ -1,4 +1,4 @@ -load("@rules_cc//cc:defs.bzl", "cc_binary") +load("//tools:defs.bzl", "cc_binary") load("//tools/images:defs.bzl", "vm_image", "vm_test") package( diff --git a/tools/images/defs.bzl b/tools/images/defs.bzl index d8e422a5d..32235813a 100644 --- a/tools/images/defs.bzl +++ b/tools/images/defs.bzl @@ -28,6 +28,8 @@ The vm_test rule can be used to execute a command remotely. For example, ) """ +load("//tools:defs.bzl", "default_installer") + def _vm_image_impl(ctx): script_paths = [] for script in ctx.files.scripts: @@ -165,8 +167,8 @@ def vm_test( targets = kwargs.pop("targets", []) if installer: targets = [installer] + targets - targets = [ - ] + targets + if default_installer(): + targets = [default_installer()] + targets _vm_test( tags = [ "local", diff --git a/tools/issue_reviver/BUILD b/tools/issue_reviver/BUILD index ee7ea11fd..4ef1a3124 100644 --- a/tools/issue_reviver/BUILD +++ b/tools/issue_reviver/BUILD @@ -1,4 +1,4 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_binary") +load("//tools:defs.bzl", "go_binary") package(licenses = ["notice"]) diff --git a/tools/issue_reviver/github/BUILD b/tools/issue_reviver/github/BUILD index 6da22ba1c..da4133472 100644 --- a/tools/issue_reviver/github/BUILD +++ b/tools/issue_reviver/github/BUILD @@ -1,11 +1,10 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) go_library( name = "github", srcs = ["github.go"], - importpath = "gvisor.dev/gvisor/tools/issue_reviver/github", visibility = [ "//tools/issue_reviver:__subpackages__", ], diff --git a/tools/issue_reviver/reviver/BUILD b/tools/issue_reviver/reviver/BUILD index 2c3675977..d262932bd 100644 --- a/tools/issue_reviver/reviver/BUILD +++ b/tools/issue_reviver/reviver/BUILD @@ -1,11 +1,10 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_library", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) go_library( name = "reviver", srcs = ["reviver.go"], - importpath = "gvisor.dev/gvisor/tools/issue_reviver/reviver", visibility = [ "//tools/issue_reviver:__subpackages__", ], @@ -15,5 +14,5 @@ go_test( name = "reviver_test", size = "small", srcs = ["reviver_test.go"], - embed = [":reviver"], + library = ":reviver", ) diff --git a/tools/workspace_status.sh b/tools/workspace_status.sh index fb09ff331..a22c8c9f2 100755 --- a/tools/workspace_status.sh +++ b/tools/workspace_status.sh @@ -15,4 +15,4 @@ # limitations under the License. # The STABLE_ prefix will trigger a re-link if it changes. -echo STABLE_VERSION $(git describe --always --tags --abbrev=12 --dirty) +echo STABLE_VERSION $(git describe --always --tags --abbrev=12 --dirty || echo 0.0.0) diff --git a/vdso/BUILD b/vdso/BUILD index 2b6744c26..d37d4266d 100644 --- a/vdso/BUILD +++ b/vdso/BUILD @@ -3,20 +3,10 @@ # normal system VDSO (time, gettimeofday, clock_gettimeofday) but which uses # timekeeping parameters managed by the sandbox kernel. -load("@bazel_tools//tools/cpp:cc_flags_supplier.bzl", "cc_flags_supplier") +load("//tools:defs.bzl", "cc_flags_supplier", "cc_toolchain", "select_arch") package(licenses = ["notice"]) -config_setting( - name = "x86_64", - constraint_values = ["@bazel_tools//platforms:x86_64"], -) - -config_setting( - name = "aarch64", - constraint_values = ["@bazel_tools//platforms:aarch64"], -) - genrule( name = "vdso", srcs = [ @@ -39,14 +29,15 @@ genrule( "-O2 " + "-std=c++11 " + "-fPIC " + + "-fno-sanitize=all " + # Some toolchains enable stack protector by default. Disable it, the # VDSO has no hooks to handle failures. "-fno-stack-protector " + "-fuse-ld=gold " + - select({ - ":x86_64": "-m64 ", - "//conditions:default": "", - }) + + select_arch( + amd64 = "-m64 ", + arm64 = "", + ) + "-shared " + "-nostdlib " + "-Wl,-soname=linux-vdso.so.1 " + @@ -55,12 +46,10 @@ genrule( "-Wl,-Bsymbolic " + "-Wl,-z,max-page-size=4096 " + "-Wl,-z,common-page-size=4096 " + - select( - { - ":x86_64": "-Wl,-T$(location vdso_amd64.lds) ", - ":aarch64": "-Wl,-T$(location vdso_arm64.lds) ", - }, - no_match_error = "Unsupported architecture", + select_arch( + amd64 = "-Wl,-T$(location vdso_amd64.lds) ", + arm64 = "-Wl,-T$(location vdso_arm64.lds) ", + no_match_error = "unsupported architecture", ) + "-o $(location vdso.so) " + "$(location vdso.cc) " + @@ -73,7 +62,7 @@ genrule( ], features = ["-pie"], toolchains = [ - "@bazel_tools//tools/cpp:current_cc_toolchain", + cc_toolchain, ":no_pie_cc_flags", ], visibility = ["//:sandbox"], -- cgit v1.2.3 From 29316e66adfc49c158425554761e34c12338f1d9 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Mon, 27 Jan 2020 12:27:04 -0800 Subject: Cleanup for GH review. --- pkg/abi/linux/netfilter.go | 14 +-- pkg/sentry/socket/netfilter/netfilter.go | 144 ++++++++++++------------------- pkg/tcpip/iptables/types.go | 15 ---- pkg/tcpip/iptables/udp_matcher.go | 62 ++++++------- test/iptables/filter_input.go | 6 +- 5 files changed, 88 insertions(+), 153 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/netfilter.go b/pkg/abi/linux/netfilter.go index effed7976..8e40bcc62 100644 --- a/pkg/abi/linux/netfilter.go +++ b/pkg/abi/linux/netfilter.go @@ -198,6 +198,8 @@ type XTEntryMatch struct { // SizeOfXTEntryMatch is the size of an XTEntryMatch. const SizeOfXTEntryMatch = 32 +// KernelXTEntryMatch is identical to XTEntryMatch, but contains +// variable-length Data field. type KernelXTEntryMatch struct { XTEntryMatch Data []byte @@ -349,19 +351,19 @@ func goString(cstring []byte) string { // XTUDP holds data for matching UDP packets. It corresponds to struct xt_udp // in include/uapi/linux/netfilter/xt_tcpudp.h. type XTUDP struct { - // SourcePortStart specifies the inclusive start of the range of source - // ports to which the matcher applies. + // SourcePortStart is the inclusive start of the range of source ports + // to which the matcher applies. SourcePortStart uint16 - // SourcePortEnd specifies the inclusive end of the range of source ports - // to which the matcher applies. + // SourcePortEnd is the inclusive end of the range of source ports to + // which the matcher applies. SourcePortEnd uint16 - // DestinationPortStart specifies the start of the destination port + // DestinationPortStart is the inclusive start of the destination port // range to which the matcher applies. DestinationPortStart uint16 - // DestinationPortEnd specifies the start of the destination port + // DestinationPortEnd is the inclusive end of the destination port // range to which the matcher applies. DestinationPortEnd uint16 diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 6c88a50a6..b8848f08a 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -34,9 +34,16 @@ import ( // shouldn't be reached - an error has occurred if we fall through to one. const errorTargetName = "ERROR" -// metadata is opaque to netstack. It holds data that we need to translate -// between Linux's and netstack's iptables representations. -// TODO(gvisor.dev/issue/170): Use metadata to check correctness. +const ( + matcherNameUDP = "udp" +) + +// Metadata is used to verify that we are correctly serializing and +// deserializing iptables into structs consumable by the iptables tool. We save +// a metadata struct when the tables are written, and when they are read out we +// verify that certain fields are the same. +// +// metadata is opaque to netstack. type metadata struct { HookEntry [linux.NF_INET_NUMHOOKS]uint32 Underflow [linux.NF_INET_NUMHOOKS]uint32 @@ -44,10 +51,12 @@ type metadata struct { Size uint32 } -const enableDebugLog = true +const enableDebug = false +// nflog logs messages related to the writing and reading of iptables, but only +// when enableDebug is true. func nflog(format string, args ...interface{}) { - if enableDebugLog { + if enableDebug { log.Infof("netfilter: "+format, args...) } } @@ -80,7 +89,7 @@ func GetInfo(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr) (linux.IPT info.NumEntries = metadata.NumEntries info.Size = metadata.Size - nflog("GetInfo returning info: %+v", info) + nflog("returning info: %+v", info) return info, nil } @@ -163,19 +172,19 @@ func convertNetstackToBinary(tablename string, table iptables.Table) (linux.Kern copy(entries.Name[:], tablename) for ruleIdx, rule := range table.Rules { - nflog("Current offset: %d", entries.Size) + nflog("convert to binary: current offset: %d", entries.Size) // Is this a chain entry point? for hook, hookRuleIdx := range table.BuiltinChains { if hookRuleIdx == ruleIdx { - nflog("Found hook %d at offset %d", hook, entries.Size) + nflog("convert to binary: found hook %d at offset %d", hook, entries.Size) meta.HookEntry[hook] = entries.Size } } // Is this a chain underflow point? for underflow, underflowRuleIdx := range table.Underflows { if underflowRuleIdx == ruleIdx { - nflog("Found underflow %d at offset %d", underflow, entries.Size) + nflog("convert to binary: found underflow %d at offset %d", underflow, entries.Size) meta.Underflow[underflow] = entries.Size } } @@ -195,7 +204,7 @@ func convertNetstackToBinary(tablename string, table iptables.Table) (linux.Kern // Serialize the matcher and add it to the // entry. serialized := marshalMatcher(matcher) - nflog("matcher serialized as: %v", serialized) + nflog("convert to binary: matcher serialized as: %v", serialized) if len(serialized)%8 != 0 { panic(fmt.Sprintf("matcher %T is not 64-bit aligned", matcher)) } @@ -212,14 +221,14 @@ func convertNetstackToBinary(tablename string, table iptables.Table) (linux.Kern entry.Elems = append(entry.Elems, serialized...) entry.NextOffset += uint16(len(serialized)) - nflog("Adding entry: %+v", entry) + nflog("convert to binary: adding entry: %+v", entry) entries.Size += uint32(entry.NextOffset) entries.Entrytable = append(entries.Entrytable, entry) meta.NumEntries++ } - nflog("Finished with an marshalled size of %d", meta.Size) + nflog("convert to binary: finished with an marshalled size of %d", meta.Size) meta.Size = entries.Size return entries, meta, nil } @@ -237,16 +246,18 @@ func marshalMatcher(matcher iptables.Matcher) []byte { } func marshalUDPMatcher(matcher *iptables.UDPMatcher) []byte { - nflog("Marshalling UDP matcher: %+v", matcher) + nflog("convert to binary: marshalling UDP matcher: %+v", matcher) + + // We have to pad this struct size to a multiple of 8 bytes. + const size = linux.SizeOfXTEntryMatch + linux.SizeOfXTUDP + 6 linuxMatcher := linux.KernelXTEntryMatch{ XTEntryMatch: linux.XTEntryMatch{ - MatchSize: linux.SizeOfXTEntryMatch + linux.SizeOfXTUDP + 6, - // Name: "udp", + MatchSize: size, }, Data: make([]byte, 0, linux.SizeOfXTUDP), } - copy(linuxMatcher.Name[:], "udp") + copy(linuxMatcher.Name[:], matcherNameUDP) xtudp := linux.XTUDP{ SourcePortStart: matcher.Data.SourcePortStart, @@ -255,17 +266,12 @@ func marshalUDPMatcher(matcher *iptables.UDPMatcher) []byte { DestinationPortEnd: matcher.Data.DestinationPortEnd, InverseFlags: matcher.Data.InverseFlags, } - nflog("marshalUDPMatcher: xtudp: %+v", xtudp) linuxMatcher.Data = binary.Marshal(linuxMatcher.Data, usermem.ByteOrder, xtudp) - nflog("marshalUDPMatcher: linuxMatcher: %+v", linuxMatcher) - // We have to pad this struct size to a multiple of 8 bytes, so we make - // this a little longer than it needs to be. - buf := make([]byte, 0, linux.SizeOfXTEntryMatch+linux.SizeOfXTUDP+6) + buf := make([]byte, 0, size) buf = binary.Marshal(buf, usermem.ByteOrder, linuxMatcher) buf = append(buf, []byte{0, 0, 0, 0, 0, 0}...) - nflog("Marshalled into matcher of size %d", len(buf)) - nflog("marshalUDPMatcher: buf is: %v", buf) + nflog("convert to binary: marshalled UDP matcher into %v", buf) return buf[:] } @@ -283,9 +289,8 @@ func marshalTarget(target iptables.Target) []byte { } func marshalStandardTarget(verdict iptables.Verdict) []byte { - nflog("Marshalling standard target with size %d", linux.SizeOfXTStandardTarget) + nflog("convert to binary: marshalling standard target with size %d", linux.SizeOfXTStandardTarget) - // TODO: Must be aligned. // The target's name will be the empty string. target := linux.XTStandardTarget{ Target: linux.XTEntryTarget{ @@ -353,8 +358,6 @@ func translateToStandardVerdict(val int32) (iptables.Verdict, *syserr.Error) { // SetEntries sets iptables rules for a single table. See // net/ipv4/netfilter/ip_tables.c:translate_table for reference. func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { - // printReplace(optVal) - // Get the basic rules data (struct ipt_replace). if len(optVal) < linux.SizeOfIPTReplace { log.Warningf("netfilter.SetEntries: optVal has insufficient size for replace %d", len(optVal)) @@ -375,13 +378,13 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { return syserr.ErrInvalidArgument } - nflog("Setting entries in table %q", replace.Name.String()) + nflog("set entries: setting entries in table %q", replace.Name.String()) // Convert input into a list of rules and their offsets. var offset uint32 var offsets []uint32 for entryIdx := uint32(0); entryIdx < replace.NumEntries; entryIdx++ { - nflog("Processing entry at offset %d", offset) + nflog("set entries: processing entry at offset %d", offset) // Get the struct ipt_entry. if len(optVal) < linux.SizeOfIPTEntry { @@ -406,11 +409,13 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { return err } - // TODO: Matchers (and maybe targets) can specify that they only work for certiain protocols, hooks, tables. + // TODO(gvisor.dev/issue/170): Matchers and targets can specify + // that they only work for certiain protocols, hooks, tables. // Get matchers. matchersSize := entry.TargetOffset - linux.SizeOfIPTEntry if len(optVal) < int(matchersSize) { log.Warningf("netfilter: entry doesn't have enough room for its matchers (only %d bytes remain)", len(optVal)) + return syserr.ErrInvalidArgument } matchers, err := parseMatchers(filter, optVal[:matchersSize]) if err != nil { @@ -423,6 +428,7 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { targetSize := entry.NextOffset - entry.TargetOffset if len(optVal) < int(targetSize) { log.Warningf("netfilter: entry doesn't have enough room for its target (only %d bytes remain)", len(optVal)) + return syserr.ErrInvalidArgument } target, err := parseTarget(optVal[:targetSize]) if err != nil { @@ -500,10 +506,11 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { // parseMatchers parses 0 or more matchers from optVal. optVal should contain // only the matchers. func parseMatchers(filter iptables.IPHeaderFilter, optVal []byte) ([]iptables.Matcher, *syserr.Error) { - nflog("Parsing matchers of size %d", len(optVal)) + nflog("set entries: parsing matchers of size %d", len(optVal)) var matchers []iptables.Matcher for len(optVal) > 0 { - nflog("parseMatchers: optVal has len %d", len(optVal)) + nflog("set entries: optVal has len %d", len(optVal)) + // Get the XTEntryMatch. if len(optVal) < linux.SizeOfXTEntryMatch { log.Warningf("netfilter: optVal has insufficient size for entry match: %d", len(optVal)) @@ -512,7 +519,7 @@ func parseMatchers(filter iptables.IPHeaderFilter, optVal []byte) ([]iptables.Ma var match linux.XTEntryMatch buf := optVal[:linux.SizeOfXTEntryMatch] binary.Unmarshal(buf, usermem.ByteOrder, &match) - nflog("parseMatchers: parsed entry match %q: %+v", match.Name.String(), match) + nflog("set entries: parsed entry match %q: %+v", match.Name.String(), match) // Check some invariants. if match.MatchSize < linux.SizeOfXTEntryMatch { @@ -528,17 +535,17 @@ func parseMatchers(filter iptables.IPHeaderFilter, optVal []byte) ([]iptables.Ma var matcher iptables.Matcher var err error switch match.Name.String() { - case "udp": + case matcherNameUDP: if len(buf) < linux.SizeOfXTUDP { log.Warningf("netfilter: optVal has insufficient size for UDP match: %d", len(optVal)) return nil, syserr.ErrInvalidArgument } + // For alignment reasons, the match's total size may + // exceed what's strictly necessary to hold matchData. var matchData linux.XTUDP - // For alignment reasons, the match's total size may exceed what's - // strictly necessary to hold matchData. binary.Unmarshal(buf[:linux.SizeOfXTUDP], usermem.ByteOrder, &matchData) log.Infof("parseMatchers: parsed XTUDP: %+v", matchData) - matcher, err = iptables.NewUDPMatcher(filter, iptables.UDPMatcherData{ + matcher, err = iptables.NewUDPMatcher(filter, iptables.UDPMatcherParams{ SourcePortStart: matchData.SourcePortStart, SourcePortEnd: matchData.SourcePortEnd, DestinationPortStart: matchData.DestinationPortStart, @@ -557,19 +564,22 @@ func parseMatchers(filter iptables.IPHeaderFilter, optVal []byte) ([]iptables.Ma matchers = append(matchers, matcher) - // TODO: Support revision. - // TODO: Support proto -- matchers usually specify which proto(s) they work with. + // TODO(gvisor.dev/issue/170): Check the revision field. optVal = optVal[match.MatchSize:] } - // TODO: Check that optVal is exhausted. + if len(optVal) != 0 { + log.Warningf("netfilter: optVal should be exhausted after parsing matchers") + return nil, syserr.ErrInvalidArgument + } + return matchers, nil } // parseTarget parses a target from optVal. optVal should contain only the // target. func parseTarget(optVal []byte) (iptables.Target, *syserr.Error) { - nflog("Parsing target of size %d", len(optVal)) + nflog("set entries: parsing target of size %d", len(optVal)) if len(optVal) < linux.SizeOfXTEntryTarget { log.Warningf("netfilter: optVal has insufficient size for entry target %d", len(optVal)) return nil, syserr.ErrInvalidArgument @@ -598,7 +608,8 @@ func parseTarget(optVal []byte) (iptables.Target, *syserr.Error) { case iptables.Drop: return iptables.UnconditionalDropTarget{}, nil default: - panic(fmt.Sprintf("Unknown verdict: %v", verdict)) + log.Warningf("Unknown verdict: %v", verdict) + return nil, syserr.ErrInvalidArgument } case errorTargetName: @@ -673,52 +684,3 @@ func hookFromLinux(hook int) iptables.Hook { } panic(fmt.Sprintf("Unknown hook %d does not correspond to a builtin chain", hook)) } - -// printReplace prints information about the struct ipt_replace in optVal. It -// is only for debugging. -func printReplace(optVal []byte) { - // Basic replace info. - var replace linux.IPTReplace - replaceBuf := optVal[:linux.SizeOfIPTReplace] - optVal = optVal[linux.SizeOfIPTReplace:] - binary.Unmarshal(replaceBuf, usermem.ByteOrder, &replace) - log.Infof("Replacing table %q: %+v", replace.Name.String(), replace) - - // Read in the list of entries at the end of replace. - var totalOffset uint16 - for entryIdx := uint32(0); entryIdx < replace.NumEntries; entryIdx++ { - var entry linux.IPTEntry - entryBuf := optVal[:linux.SizeOfIPTEntry] - binary.Unmarshal(entryBuf, usermem.ByteOrder, &entry) - log.Infof("Entry %d (total offset %d): %+v", entryIdx, totalOffset, entry) - - totalOffset += entry.NextOffset - if entry.TargetOffset == linux.SizeOfIPTEntry { - log.Infof("Entry has no matches.") - } else { - log.Infof("Entry has matches.") - } - - var target linux.XTEntryTarget - targetBuf := optVal[entry.TargetOffset : entry.TargetOffset+linux.SizeOfXTEntryTarget] - binary.Unmarshal(targetBuf, usermem.ByteOrder, &target) - log.Infof("Target named %q: %+v", target.Name.String(), target) - - switch target.Name.String() { - case "": - var standardTarget linux.XTStandardTarget - stBuf := optVal[entry.TargetOffset : entry.TargetOffset+linux.SizeOfXTStandardTarget] - binary.Unmarshal(stBuf, usermem.ByteOrder, &standardTarget) - log.Infof("Standard target with verdict %q (%d).", linux.VerdictStrings[standardTarget.Verdict], standardTarget.Verdict) - case errorTargetName: - var errorTarget linux.XTErrorTarget - etBuf := optVal[entry.TargetOffset : entry.TargetOffset+linux.SizeOfXTErrorTarget] - binary.Unmarshal(etBuf, usermem.ByteOrder, &errorTarget) - log.Infof("Error target with name %q.", errorTarget.Name.String()) - default: - log.Infof("Unknown target type.") - } - - optVal = optVal[entry.NextOffset:] - } -} diff --git a/pkg/tcpip/iptables/types.go b/pkg/tcpip/iptables/types.go index d47447d40..ba5ed75b4 100644 --- a/pkg/tcpip/iptables/types.go +++ b/pkg/tcpip/iptables/types.go @@ -169,8 +169,6 @@ type IPHeaderFilter struct { Protocol tcpip.TransportProtocolNumber } -// TODO: Should these be able to marshal/unmarshal themselves? -// TODO: Something has to map the name to the matcher. // A Matcher is the interface for matching packets. type Matcher interface { // Match returns whether the packet matches and whether the packet @@ -179,19 +177,6 @@ type Matcher interface { // // Precondition: packet.NetworkHeader is set. Match(hook Hook, packet tcpip.PacketBuffer, interfaceName string) (matches bool, hotdrop bool) - - // TODO: Make this typesafe by having each Matcher have their own, typed CheckEntry? - // CheckEntry(params MatchCheckEntryParams) bool -} - -// TODO: Unused? -type MatchCheckEntryParams struct { - Table string // TODO: Tables should be an enum... - Filter IPHeaderFilter - Info interface{} // TODO: Type unsafe. - // HookMask uint8 - // Family uint8 - // NFTCompat bool } // A Target is the interface for taking an action for a packet. diff --git a/pkg/tcpip/iptables/udp_matcher.go b/pkg/tcpip/iptables/udp_matcher.go index 65ae7f9e0..f59ca2027 100644 --- a/pkg/tcpip/iptables/udp_matcher.go +++ b/pkg/tcpip/iptables/udp_matcher.go @@ -16,33 +16,28 @@ package iptables import ( "fmt" - "runtime/debug" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/header" ) +// TODO(gvisor.dev/issue/170): The following per-matcher params should be +// supported: +// - Table name +// - Match size +// - User size +// - Hooks +// - Proto +// - Family + +// UDPMatcher matches UDP packets and their headers. It implements Matcher. type UDPMatcher struct { - Data UDPMatcherData - - // tablename string - // unsigned int matchsize; - // unsigned int usersize; - // #ifdef CONFIG_COMPAT - // unsigned int compatsize; - // #endif - // unsigned int hooks; - // unsigned short proto; - // unsigned short family; + Data UDPMatcherParams } -// TODO: Delete? -// MatchCheckEntryParams - -type UDPMatcherData struct { - // Filter IPHeaderFilter - +// UDPMatcherParams are the parameters used to create a UDPMatcher. +type UDPMatcherParams struct { SourcePortStart uint16 SourcePortEnd uint16 DestinationPortStart uint16 @@ -50,12 +45,12 @@ type UDPMatcherData struct { InverseFlags uint8 } -func NewUDPMatcher(filter IPHeaderFilter, data UDPMatcherData) (Matcher, error) { - // TODO: We currently only support source port and destination port. - log.Infof("Adding rule with UDPMatcherData: %+v", data) +// NewUDPMatcher returns a new instance of UDPMatcher. +func NewUDPMatcher(filter IPHeaderFilter, data UDPMatcherParams) (Matcher, error) { + log.Infof("Adding rule with UDPMatcherParams: %+v", data) if data.InverseFlags != 0 { - return nil, fmt.Errorf("unsupported UDP matcher flags set") + return nil, fmt.Errorf("unsupported UDP matcher inverse flags set") } if filter.Protocol != header.UDPProtocolNumber { @@ -65,21 +60,18 @@ func NewUDPMatcher(filter IPHeaderFilter, data UDPMatcherData) (Matcher, error) return &UDPMatcher{Data: data}, nil } -// TODO: Check xt_tcpudp.c. Need to check for same things (e.g. fragments). +// Match implements Matcher.Match. func (um *UDPMatcher) Match(hook Hook, pkt tcpip.PacketBuffer, interfaceName string) (bool, bool) { - log.Infof("UDPMatcher called from: %s", string(debug.Stack())) netHeader := header.IPv4(pkt.NetworkHeader) - // TODO: Do we check proto here or elsewhere? I think elsewhere (check - // codesearch). + // TODO(gvisor.dev/issue/170): Proto checks should ultimately be moved + // into the iptables.Check codepath as matchers are added. if netHeader.TransportProtocol() != header.UDPProtocolNumber { - log.Infof("UDPMatcher: wrong protocol number") return false, false } // We dont't match fragments. if frag := netHeader.FragmentOffset(); frag != 0 { - log.Infof("UDPMatcher: it's a fragment") if frag == 1 { return false, true } @@ -89,20 +81,18 @@ func (um *UDPMatcher) Match(hook Hook, pkt tcpip.PacketBuffer, interfaceName str // Now we need the transport header. However, this may not have been set // yet. - // TODO + // TODO(gvisor.dev/issue/170): Parsing the transport header should + // ultimately be moved into the iptables.Check codepath as matchers are + // added. var udpHeader header.UDP if pkt.TransportHeader != nil { - log.Infof("UDPMatcher: transport header is not nil") udpHeader = header.UDP(pkt.TransportHeader) } else { - log.Infof("UDPMatcher: transport header is nil") - log.Infof("UDPMatcher: is network header nil: %t", pkt.NetworkHeader == nil) // The UDP header hasn't been parsed yet. We have to do it here. if len(pkt.Data.First()) < header.UDPMinimumSize { // There's no valid UDP header here, so we hotdrop the // packet. - // TODO: Stats. - log.Warningf("Dropping UDP packet: size to small.") + log.Warningf("Dropping UDP packet: size too small.") return false, true } udpHeader = header.UDP(pkt.Data.First()) @@ -112,10 +102,6 @@ func (um *UDPMatcher) Match(hook Hook, pkt tcpip.PacketBuffer, interfaceName str // matching range. sourcePort := udpHeader.SourcePort() destinationPort := udpHeader.DestinationPort() - log.Infof("UDPMatcher: sport and dport are %d and %d. sports and dport start and end are (%d, %d) and (%d, %d)", - udpHeader.SourcePort(), udpHeader.DestinationPort(), - um.Data.SourcePortStart, um.Data.SourcePortEnd, - um.Data.DestinationPortStart, um.Data.DestinationPortEnd) if sourcePort < um.Data.SourcePortStart || um.Data.SourcePortEnd < sourcePort { return false, false } diff --git a/test/iptables/filter_input.go b/test/iptables/filter_input.go index bc963d40e..e9f0978eb 100644 --- a/test/iptables/filter_input.go +++ b/test/iptables/filter_input.go @@ -264,9 +264,9 @@ func (FilterInputMultiUDPRules) ContainerAction(ip net.IP) error { if err := filterTable("-A", "INPUT", "-p", "udp", "-m", "udp", "--destination-port", fmt.Sprintf("%d", dropPort), "-j", "DROP"); err != nil { return err } - // if err := filterTable("-A", "INPUT", "-p", "udp", "-m", "udp", "--destination-port", fmt.Sprintf("%d", acceptPort), "-j", "ACCEPT"); err != nil { - // return err - // } + if err := filterTable("-A", "INPUT", "-p", "udp", "-m", "udp", "--destination-port", fmt.Sprintf("%d", acceptPort), "-j", "ACCEPT"); err != nil { + return err + } return filterTable("-L") } -- cgit v1.2.3 From 13c1f38dfa215ab3e3cc70642721f55ab226d5b7 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Mon, 27 Jan 2020 12:19:20 -0800 Subject: Update bug number for supporting extended attribute namespaces. PiperOrigin-RevId: 291774815 --- pkg/sentry/syscalls/linux/sys_xattr.go | 1 + test/syscalls/linux/xattr.cc | 2 +- 2 files changed, 2 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/sys_xattr.go b/pkg/sentry/syscalls/linux/sys_xattr.go index e35c077d6..77deb8980 100644 --- a/pkg/sentry/syscalls/linux/sys_xattr.go +++ b/pkg/sentry/syscalls/linux/sys_xattr.go @@ -103,6 +103,7 @@ func getXattr(t *kernel.Task, d *fs.Dirent, nameAddr usermem.Addr, size uint64) return 0, "", err } + // TODO(b/148380782): Support xattrs in namespaces other than "user". if !strings.HasPrefix(name, linux.XATTR_USER_PREFIX) { return 0, "", syserror.EOPNOTSUPP } diff --git a/test/syscalls/linux/xattr.cc b/test/syscalls/linux/xattr.cc index e77c355d7..ab21d68c6 100644 --- a/test/syscalls/linux/xattr.cc +++ b/test/syscalls/linux/xattr.cc @@ -131,7 +131,7 @@ TEST_F(XattrTest, XattrWriteOnly_NoRandomSave) { } TEST_F(XattrTest, XattrTrustedWithNonadmin) { - // TODO(b/127675828): Support setxattr and getxattr with "trusted" prefix. + // TODO(b/148380782): Support setxattr and getxattr with "trusted" prefix. SKIP_IF(IsRunningOnGvisor()); SKIP_IF(ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SYS_ADMIN))); -- cgit v1.2.3 From 90ec5961667a1c4a21702e64adb383403af8ad25 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Mon, 27 Jan 2020 13:22:50 -0800 Subject: Fix licenses. The preferred Copyright holder is "The gVisor Authors". PiperOrigin-RevId: 291786657 --- pkg/sentry/kernel/fd_table.go | 2 +- pkg/sentry/kernel/fd_table_test.go | 2 +- pkg/sentry/kernel/fd_table_unsafe.go | 2 +- pkg/sentry/platform/ring0/entry_arm64.go | 2 +- pkg/sentry/platform/ring0/kernel_arm64.go | 2 +- pkg/sentry/platform/ring0/lib_arm64.go | 2 +- pkg/sentry/platform/ring0/offsets_arm64.go | 2 +- pkg/tcpip/iptables/iptables.go | 2 +- pkg/tcpip/iptables/types.go | 2 +- runsc/cmd/help.go | 2 +- tools/go_marshal/main.go | 2 +- 11 files changed, 11 insertions(+), 11 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/fd_table.go b/pkg/sentry/kernel/fd_table.go index cd1501f85..0ad4135b3 100644 --- a/pkg/sentry/kernel/fd_table.go +++ b/pkg/sentry/kernel/fd_table.go @@ -1,4 +1,4 @@ -// Copyright 2018 Google LLC +// Copyright 2018 The gVisor Authors. // // Licensed under the Apache License, Version 2.0 (the "License"); // you may not use this file except in compliance with the License. diff --git a/pkg/sentry/kernel/fd_table_test.go b/pkg/sentry/kernel/fd_table_test.go index eccb7d1e7..86164df49 100644 --- a/pkg/sentry/kernel/fd_table_test.go +++ b/pkg/sentry/kernel/fd_table_test.go @@ -1,4 +1,4 @@ -// Copyright 2018 Google LLC +// Copyright 2018 The gVisor Authors. // // Licensed under the Apache License, Version 2.0 (the "License"); // you may not use this file except in compliance with the License. diff --git a/pkg/sentry/kernel/fd_table_unsafe.go b/pkg/sentry/kernel/fd_table_unsafe.go index e009df974..e9fdb0917 100644 --- a/pkg/sentry/kernel/fd_table_unsafe.go +++ b/pkg/sentry/kernel/fd_table_unsafe.go @@ -1,4 +1,4 @@ -// Copyright 2018 Google LLC +// Copyright 2018 The gVisor Authors. // // Licensed under the Apache License, Version 2.0 (the "License"); // you may not use this file except in compliance with the License. diff --git a/pkg/sentry/platform/ring0/entry_arm64.go b/pkg/sentry/platform/ring0/entry_arm64.go index 0dfa42c36..62a93f3d6 100644 --- a/pkg/sentry/platform/ring0/entry_arm64.go +++ b/pkg/sentry/platform/ring0/entry_arm64.go @@ -1,4 +1,4 @@ -// Copyright 2019 Google Inc. +// Copyright 2019 The gVisor Authors. // // Licensed under the Apache License, Version 2.0 (the "License"); // you may not use this file except in compliance with the License. diff --git a/pkg/sentry/platform/ring0/kernel_arm64.go b/pkg/sentry/platform/ring0/kernel_arm64.go index ed82a131e..c3d341998 100644 --- a/pkg/sentry/platform/ring0/kernel_arm64.go +++ b/pkg/sentry/platform/ring0/kernel_arm64.go @@ -1,4 +1,4 @@ -// Copyright 2019 Google Inc. +// Copyright 2019 The gVisor Authors. // // Licensed under the Apache License, Version 2.0 (the "License"); // you may not use this file except in compliance with the License. diff --git a/pkg/sentry/platform/ring0/lib_arm64.go b/pkg/sentry/platform/ring0/lib_arm64.go index 8bcfe1032..af075aae4 100644 --- a/pkg/sentry/platform/ring0/lib_arm64.go +++ b/pkg/sentry/platform/ring0/lib_arm64.go @@ -1,4 +1,4 @@ -// Copyright 2019 Google Inc. +// Copyright 2019 The gVisor Authors. // // Licensed under the Apache License, Version 2.0 (the "License"); // you may not use this file except in compliance with the License. diff --git a/pkg/sentry/platform/ring0/offsets_arm64.go b/pkg/sentry/platform/ring0/offsets_arm64.go index cd2a65f97..8c960c749 100644 --- a/pkg/sentry/platform/ring0/offsets_arm64.go +++ b/pkg/sentry/platform/ring0/offsets_arm64.go @@ -1,4 +1,4 @@ -// Copyright 2019 Google Inc. +// Copyright 2019 The gVisor Authors. // // Licensed under the Apache License, Version 2.0 (the "License"); // you may not use this file except in compliance with the License. diff --git a/pkg/tcpip/iptables/iptables.go b/pkg/tcpip/iptables/iptables.go index fc06b5b87..4bfb3149e 100644 --- a/pkg/tcpip/iptables/iptables.go +++ b/pkg/tcpip/iptables/iptables.go @@ -1,4 +1,4 @@ -// Copyright 2019 The gVisor authors. +// Copyright 2019 The gVisor Authors. // // Licensed under the Apache License, Version 2.0 (the "License"); // you may not use this file except in compliance with the License. diff --git a/pkg/tcpip/iptables/types.go b/pkg/tcpip/iptables/types.go index a8b972f1b..50893cc55 100644 --- a/pkg/tcpip/iptables/types.go +++ b/pkg/tcpip/iptables/types.go @@ -1,4 +1,4 @@ -// Copyright 2019 The gVisor authors. +// Copyright 2019 The gVisor Authors. // // Licensed under the Apache License, Version 2.0 (the "License"); // you may not use this file except in compliance with the License. diff --git a/runsc/cmd/help.go b/runsc/cmd/help.go index ff4f901cb..930e8454f 100644 --- a/runsc/cmd/help.go +++ b/runsc/cmd/help.go @@ -1,4 +1,4 @@ -// Copyright 2018 Google LLC +// Copyright 2018 The gVisor Authors. // // Licensed under the Apache License, Version 2.0 (the "License"); // you may not use this file except in compliance with the License. diff --git a/tools/go_marshal/main.go b/tools/go_marshal/main.go index e1a97b311..f74be5c29 100644 --- a/tools/go_marshal/main.go +++ b/tools/go_marshal/main.go @@ -1,4 +1,4 @@ -// Copyright 2019 Google LLC +// Copyright 2019 The gVisor Authors. // // Licensed under the Apache License, Version 2.0 (the "License"); // you may not use this file except in compliance with the License. -- cgit v1.2.3 From 0e2f1b7abd219f39d67cc2cecd00c441a13eeb29 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Mon, 27 Jan 2020 15:17:58 -0800 Subject: Update package locations. Because the abi will depend on the core types for marshalling (usermem, context, safemem, safecopy), these need to be flattened from the sentry directory. These packages contain no sentry-specific details. PiperOrigin-RevId: 291811289 --- pkg/abi/abi.go | 4 + pkg/context/BUILD | 13 + pkg/context/context.go | 141 +++++ pkg/safecopy/BUILD | 29 + pkg/safecopy/LICENSE | 27 + pkg/safecopy/atomic_amd64.s | 136 +++++ pkg/safecopy/atomic_arm64.s | 126 +++++ pkg/safecopy/memclr_amd64.s | 147 +++++ pkg/safecopy/memclr_arm64.s | 74 +++ pkg/safecopy/memcpy_amd64.s | 250 +++++++++ pkg/safecopy/memcpy_arm64.s | 78 +++ pkg/safecopy/safecopy.go | 144 +++++ pkg/safecopy/safecopy_test.go | 617 +++++++++++++++++++++ pkg/safecopy/safecopy_unsafe.go | 335 +++++++++++ pkg/safecopy/sighandler_amd64.s | 133 +++++ pkg/safecopy/sighandler_arm64.s | 143 +++++ pkg/safemem/BUILD | 27 + pkg/safemem/block_unsafe.go | 279 ++++++++++ pkg/safemem/io.go | 392 +++++++++++++ pkg/safemem/io_test.go | 199 +++++++ pkg/safemem/safemem.go | 16 + pkg/safemem/seq_test.go | 196 +++++++ pkg/safemem/seq_unsafe.go | 299 ++++++++++ pkg/sentry/arch/BUILD | 4 +- pkg/sentry/arch/arch.go | 2 +- pkg/sentry/arch/arch_aarch64.go | 2 +- pkg/sentry/arch/arch_amd64.go | 2 +- pkg/sentry/arch/arch_arm64.go | 2 +- pkg/sentry/arch/arch_state_x86.go | 2 +- pkg/sentry/arch/arch_x86.go | 2 +- pkg/sentry/arch/auxv.go | 2 +- pkg/sentry/arch/signal.go | 2 +- pkg/sentry/arch/signal_amd64.go | 2 +- pkg/sentry/arch/signal_arm64.go | 2 +- pkg/sentry/arch/signal_stack.go | 2 +- pkg/sentry/arch/stack.go | 4 +- pkg/sentry/context/BUILD | 13 - pkg/sentry/context/context.go | 141 ----- pkg/sentry/context/contexttest/BUILD | 21 - pkg/sentry/context/contexttest/contexttest.go | 188 ------- pkg/sentry/contexttest/BUILD | 21 + pkg/sentry/contexttest/contexttest.go | 188 +++++++ pkg/sentry/fs/BUILD | 12 +- pkg/sentry/fs/anon/BUILD | 4 +- pkg/sentry/fs/anon/anon.go | 4 +- pkg/sentry/fs/attr.go | 2 +- pkg/sentry/fs/context.go | 2 +- pkg/sentry/fs/copy_up.go | 4 +- pkg/sentry/fs/copy_up_test.go | 2 +- pkg/sentry/fs/dev/BUILD | 6 +- pkg/sentry/fs/dev/dev.go | 4 +- pkg/sentry/fs/dev/fs.go | 2 +- pkg/sentry/fs/dev/full.go | 4 +- pkg/sentry/fs/dev/null.go | 2 +- pkg/sentry/fs/dev/random.go | 6 +- pkg/sentry/fs/dev/tty.go | 2 +- pkg/sentry/fs/dirent.go | 2 +- pkg/sentry/fs/dirent_refs_test.go | 4 +- pkg/sentry/fs/fdpipe/BUILD | 12 +- pkg/sentry/fs/fdpipe/pipe.go | 6 +- pkg/sentry/fs/fdpipe/pipe_opener.go | 2 +- pkg/sentry/fs/fdpipe/pipe_opener_test.go | 6 +- pkg/sentry/fs/fdpipe/pipe_state.go | 2 +- pkg/sentry/fs/fdpipe/pipe_test.go | 4 +- pkg/sentry/fs/file.go | 4 +- pkg/sentry/fs/file_operations.go | 4 +- pkg/sentry/fs/file_overlay.go | 4 +- pkg/sentry/fs/file_overlay_test.go | 2 +- pkg/sentry/fs/filesystems.go | 2 +- pkg/sentry/fs/filetest/BUILD | 6 +- pkg/sentry/fs/filetest/filetest.go | 6 +- pkg/sentry/fs/fs.go | 2 +- pkg/sentry/fs/fsutil/BUILD | 14 +- pkg/sentry/fs/fsutil/dirty_set.go | 6 +- pkg/sentry/fs/fsutil/dirty_set_test.go | 2 +- pkg/sentry/fs/fsutil/file.go | 4 +- pkg/sentry/fs/fsutil/file_range_set.go | 6 +- pkg/sentry/fs/fsutil/host_file_mapper.go | 4 +- pkg/sentry/fs/fsutil/host_file_mapper_unsafe.go | 2 +- pkg/sentry/fs/fsutil/host_mappable.go | 6 +- pkg/sentry/fs/fsutil/inode.go | 2 +- pkg/sentry/fs/fsutil/inode_cached.go | 6 +- pkg/sentry/fs/fsutil/inode_cached_test.go | 8 +- pkg/sentry/fs/gofer/BUILD | 10 +- pkg/sentry/fs/gofer/attr.go | 4 +- pkg/sentry/fs/gofer/cache_policy.go | 2 +- pkg/sentry/fs/gofer/context_file.go | 2 +- pkg/sentry/fs/gofer/file.go | 4 +- pkg/sentry/fs/gofer/file_state.go | 2 +- pkg/sentry/fs/gofer/fs.go | 2 +- pkg/sentry/fs/gofer/gofer_test.go | 4 +- pkg/sentry/fs/gofer/handles.go | 4 +- pkg/sentry/fs/gofer/inode.go | 4 +- pkg/sentry/fs/gofer/inode_state.go | 2 +- pkg/sentry/fs/gofer/path.go | 2 +- pkg/sentry/fs/gofer/session.go | 2 +- pkg/sentry/fs/gofer/session_state.go | 2 +- pkg/sentry/fs/gofer/socket.go | 2 +- pkg/sentry/fs/gofer/util.go | 2 +- pkg/sentry/fs/host/BUILD | 12 +- pkg/sentry/fs/host/control.go | 2 +- pkg/sentry/fs/host/file.go | 6 +- pkg/sentry/fs/host/fs.go | 2 +- pkg/sentry/fs/host/fs_test.go | 4 +- pkg/sentry/fs/host/inode.go | 4 +- pkg/sentry/fs/host/inode_state.go | 2 +- pkg/sentry/fs/host/inode_test.go | 2 +- pkg/sentry/fs/host/socket.go | 2 +- pkg/sentry/fs/host/socket_test.go | 4 +- pkg/sentry/fs/host/tty.go | 4 +- pkg/sentry/fs/host/wait_test.go | 2 +- pkg/sentry/fs/inode.go | 2 +- pkg/sentry/fs/inode_operations.go | 2 +- pkg/sentry/fs/inode_overlay.go | 2 +- pkg/sentry/fs/inode_overlay_test.go | 2 +- pkg/sentry/fs/inotify.go | 4 +- pkg/sentry/fs/inotify_event.go | 4 +- pkg/sentry/fs/mock.go | 2 +- pkg/sentry/fs/mount.go | 2 +- pkg/sentry/fs/mount_overlay.go | 2 +- pkg/sentry/fs/mount_test.go | 2 +- pkg/sentry/fs/mounts.go | 2 +- pkg/sentry/fs/mounts_test.go | 2 +- pkg/sentry/fs/offset.go | 2 +- pkg/sentry/fs/overlay.go | 4 +- pkg/sentry/fs/proc/BUILD | 8 +- pkg/sentry/fs/proc/cgroup.go | 2 +- pkg/sentry/fs/proc/cpuinfo.go | 2 +- pkg/sentry/fs/proc/exec_args.go | 4 +- pkg/sentry/fs/proc/fds.go | 2 +- pkg/sentry/fs/proc/filesystems.go | 2 +- pkg/sentry/fs/proc/fs.go | 2 +- pkg/sentry/fs/proc/inode.go | 4 +- pkg/sentry/fs/proc/loadavg.go | 2 +- pkg/sentry/fs/proc/meminfo.go | 4 +- pkg/sentry/fs/proc/mounts.go | 2 +- pkg/sentry/fs/proc/net.go | 4 +- pkg/sentry/fs/proc/proc.go | 2 +- pkg/sentry/fs/proc/seqfile/BUILD | 10 +- pkg/sentry/fs/proc/seqfile/seqfile.go | 4 +- pkg/sentry/fs/proc/seqfile/seqfile_test.go | 6 +- pkg/sentry/fs/proc/stat.go | 2 +- pkg/sentry/fs/proc/sys.go | 4 +- pkg/sentry/fs/proc/sys_net.go | 4 +- pkg/sentry/fs/proc/sys_net_test.go | 4 +- pkg/sentry/fs/proc/task.go | 4 +- pkg/sentry/fs/proc/uid_gid_map.go | 4 +- pkg/sentry/fs/proc/uptime.go | 4 +- pkg/sentry/fs/proc/version.go | 2 +- pkg/sentry/fs/ramfs/BUILD | 6 +- pkg/sentry/fs/ramfs/dir.go | 2 +- pkg/sentry/fs/ramfs/socket.go | 2 +- pkg/sentry/fs/ramfs/symlink.go | 2 +- pkg/sentry/fs/ramfs/tree.go | 4 +- pkg/sentry/fs/ramfs/tree_test.go | 2 +- pkg/sentry/fs/splice.go | 2 +- pkg/sentry/fs/sys/BUILD | 4 +- pkg/sentry/fs/sys/devices.go | 2 +- pkg/sentry/fs/sys/fs.go | 2 +- pkg/sentry/fs/sys/sys.go | 4 +- pkg/sentry/fs/timerfd/BUILD | 4 +- pkg/sentry/fs/timerfd/timerfd.go | 4 +- pkg/sentry/fs/tmpfs/BUILD | 10 +- pkg/sentry/fs/tmpfs/file_regular.go | 4 +- pkg/sentry/fs/tmpfs/file_test.go | 4 +- pkg/sentry/fs/tmpfs/fs.go | 2 +- pkg/sentry/fs/tmpfs/inode_file.go | 6 +- pkg/sentry/fs/tmpfs/tmpfs.go | 4 +- pkg/sentry/fs/tty/BUILD | 10 +- pkg/sentry/fs/tty/dir.go | 4 +- pkg/sentry/fs/tty/fs.go | 2 +- pkg/sentry/fs/tty/line_discipline.go | 4 +- pkg/sentry/fs/tty/master.go | 4 +- pkg/sentry/fs/tty/queue.go | 6 +- pkg/sentry/fs/tty/slave.go | 4 +- pkg/sentry/fs/tty/terminal.go | 4 +- pkg/sentry/fs/tty/tty_test.go | 4 +- pkg/sentry/fsimpl/ext/BUILD | 12 +- pkg/sentry/fsimpl/ext/benchmark/BUILD | 4 +- pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go | 4 +- pkg/sentry/fsimpl/ext/directory.go | 2 +- pkg/sentry/fsimpl/ext/ext.go | 2 +- pkg/sentry/fsimpl/ext/ext_test.go | 6 +- pkg/sentry/fsimpl/ext/file_description.go | 2 +- pkg/sentry/fsimpl/ext/filesystem.go | 2 +- pkg/sentry/fsimpl/ext/regular_file.go | 6 +- pkg/sentry/fsimpl/ext/symlink.go | 4 +- pkg/sentry/fsimpl/kernfs/BUILD | 10 +- pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go | 4 +- pkg/sentry/fsimpl/kernfs/fd_impl_util.go | 4 +- pkg/sentry/fsimpl/kernfs/filesystem.go | 2 +- pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 2 +- pkg/sentry/fsimpl/kernfs/kernfs.go | 2 +- pkg/sentry/fsimpl/kernfs/kernfs_test.go | 6 +- pkg/sentry/fsimpl/kernfs/symlink.go | 2 +- pkg/sentry/fsimpl/proc/BUILD | 12 +- pkg/sentry/fsimpl/proc/filesystem.go | 2 +- pkg/sentry/fsimpl/proc/subtasks.go | 2 +- pkg/sentry/fsimpl/proc/task.go | 2 +- pkg/sentry/fsimpl/proc/task_files.go | 6 +- pkg/sentry/fsimpl/proc/tasks.go | 2 +- pkg/sentry/fsimpl/proc/tasks_files.go | 4 +- pkg/sentry/fsimpl/proc/tasks_net.go | 4 +- pkg/sentry/fsimpl/proc/tasks_sys.go | 2 +- pkg/sentry/fsimpl/proc/tasks_sys_test.go | 2 +- pkg/sentry/fsimpl/proc/tasks_test.go | 4 +- pkg/sentry/fsimpl/sys/BUILD | 2 +- pkg/sentry/fsimpl/sys/sys.go | 2 +- pkg/sentry/fsimpl/testutil/BUILD | 4 +- pkg/sentry/fsimpl/testutil/kernel.go | 2 +- pkg/sentry/fsimpl/testutil/testutil.go | 4 +- pkg/sentry/fsimpl/tmpfs/BUILD | 16 +- pkg/sentry/fsimpl/tmpfs/benchmark_test.go | 4 +- pkg/sentry/fsimpl/tmpfs/directory.go | 2 +- pkg/sentry/fsimpl/tmpfs/filesystem.go | 2 +- pkg/sentry/fsimpl/tmpfs/named_pipe.go | 4 +- pkg/sentry/fsimpl/tmpfs/pipe_test.go | 6 +- pkg/sentry/fsimpl/tmpfs/regular_file.go | 6 +- pkg/sentry/fsimpl/tmpfs/regular_file_test.go | 4 +- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 2 +- pkg/sentry/hostmm/BUILD | 2 +- pkg/sentry/hostmm/hostmm.go | 2 +- pkg/sentry/inet/BUILD | 2 +- pkg/sentry/inet/context.go | 2 +- pkg/sentry/kernel/BUILD | 12 +- pkg/sentry/kernel/auth/BUILD | 2 +- pkg/sentry/kernel/auth/context.go | 2 +- pkg/sentry/kernel/auth/id_map.go | 2 +- pkg/sentry/kernel/context.go | 2 +- pkg/sentry/kernel/contexttest/BUILD | 4 +- pkg/sentry/kernel/contexttest/contexttest.go | 4 +- pkg/sentry/kernel/epoll/BUILD | 6 +- pkg/sentry/kernel/epoll/epoll.go | 4 +- pkg/sentry/kernel/epoll/epoll_test.go | 2 +- pkg/sentry/kernel/eventfd/BUILD | 8 +- pkg/sentry/kernel/eventfd/eventfd.go | 4 +- pkg/sentry/kernel/eventfd/eventfd_test.go | 4 +- pkg/sentry/kernel/fd_table.go | 2 +- pkg/sentry/kernel/fd_table_test.go | 4 +- pkg/sentry/kernel/futex/BUILD | 6 +- pkg/sentry/kernel/futex/futex.go | 2 +- pkg/sentry/kernel/futex/futex_test.go | 2 +- pkg/sentry/kernel/kernel.go | 2 +- pkg/sentry/kernel/pipe/BUILD | 12 +- pkg/sentry/kernel/pipe/buffer.go | 2 +- pkg/sentry/kernel/pipe/buffer_test.go | 2 +- pkg/sentry/kernel/pipe/node.go | 2 +- pkg/sentry/kernel/pipe/node_test.go | 6 +- pkg/sentry/kernel/pipe/pipe.go | 2 +- pkg/sentry/kernel/pipe/pipe_test.go | 4 +- pkg/sentry/kernel/pipe/pipe_util.go | 4 +- pkg/sentry/kernel/pipe/reader_writer.go | 4 +- pkg/sentry/kernel/pipe/vfs.go | 4 +- pkg/sentry/kernel/ptrace.go | 2 +- pkg/sentry/kernel/ptrace_amd64.go | 2 +- pkg/sentry/kernel/ptrace_arm64.go | 2 +- pkg/sentry/kernel/rseq.go | 2 +- pkg/sentry/kernel/seccomp.go | 2 +- pkg/sentry/kernel/semaphore/BUILD | 6 +- pkg/sentry/kernel/semaphore/semaphore.go | 2 +- pkg/sentry/kernel/semaphore/semaphore_test.go | 4 +- pkg/sentry/kernel/shm/BUILD | 4 +- pkg/sentry/kernel/shm/shm.go | 4 +- pkg/sentry/kernel/signalfd/BUILD | 4 +- pkg/sentry/kernel/signalfd/signalfd.go | 4 +- pkg/sentry/kernel/syscalls.go | 2 +- pkg/sentry/kernel/task.go | 4 +- pkg/sentry/kernel/task_clone.go | 2 +- pkg/sentry/kernel/task_context.go | 4 +- pkg/sentry/kernel/task_futex.go | 2 +- pkg/sentry/kernel/task_log.go | 2 +- pkg/sentry/kernel/task_run.go | 2 +- pkg/sentry/kernel/task_signals.go | 2 +- pkg/sentry/kernel/task_start.go | 2 +- pkg/sentry/kernel/task_syscall.go | 2 +- pkg/sentry/kernel/task_usermem.go | 2 +- pkg/sentry/kernel/time/BUILD | 2 +- pkg/sentry/kernel/time/context.go | 2 +- pkg/sentry/kernel/timekeeper_test.go | 4 +- pkg/sentry/kernel/vdso.go | 4 +- pkg/sentry/limits/BUILD | 2 +- pkg/sentry/limits/context.go | 2 +- pkg/sentry/loader/BUILD | 6 +- pkg/sentry/loader/elf.go | 4 +- pkg/sentry/loader/interpreter.go | 4 +- pkg/sentry/loader/loader.go | 4 +- pkg/sentry/loader/vdso.go | 6 +- pkg/sentry/memmap/BUILD | 6 +- pkg/sentry/memmap/mapping_set.go | 2 +- pkg/sentry/memmap/mapping_set_test.go | 2 +- pkg/sentry/memmap/memmap.go | 4 +- pkg/sentry/mm/BUILD | 18 +- pkg/sentry/mm/address_space.go | 2 +- pkg/sentry/mm/aio_context.go | 4 +- pkg/sentry/mm/debug.go | 2 +- pkg/sentry/mm/io.go | 6 +- pkg/sentry/mm/lifecycle.go | 4 +- pkg/sentry/mm/metadata.go | 2 +- pkg/sentry/mm/mm.go | 4 +- pkg/sentry/mm/mm_test.go | 6 +- pkg/sentry/mm/pma.go | 8 +- pkg/sentry/mm/procfs.go | 4 +- pkg/sentry/mm/save_restore.go | 2 +- pkg/sentry/mm/shm.go | 4 +- pkg/sentry/mm/special_mappable.go | 4 +- pkg/sentry/mm/syscalls.go | 4 +- pkg/sentry/mm/vma.go | 4 +- pkg/sentry/pgalloc/BUILD | 8 +- pkg/sentry/pgalloc/context.go | 2 +- pkg/sentry/pgalloc/pgalloc.go | 6 +- pkg/sentry/pgalloc/pgalloc_test.go | 2 +- pkg/sentry/pgalloc/save_restore.go | 2 +- pkg/sentry/platform/BUILD | 8 +- pkg/sentry/platform/context.go | 2 +- pkg/sentry/platform/kvm/BUILD | 6 +- pkg/sentry/platform/kvm/address_space.go | 2 +- pkg/sentry/platform/kvm/bluepill.go | 2 +- pkg/sentry/platform/kvm/bluepill_fault.go | 2 +- pkg/sentry/platform/kvm/context.go | 2 +- pkg/sentry/platform/kvm/kvm.go | 2 +- pkg/sentry/platform/kvm/kvm_test.go | 2 +- pkg/sentry/platform/kvm/machine.go | 2 +- pkg/sentry/platform/kvm/machine_amd64.go | 2 +- pkg/sentry/platform/kvm/machine_arm64.go | 2 +- pkg/sentry/platform/kvm/machine_arm64_unsafe.go | 2 +- pkg/sentry/platform/kvm/physical_map.go | 2 +- pkg/sentry/platform/kvm/virtual_map.go | 2 +- pkg/sentry/platform/kvm/virtual_map_test.go | 2 +- pkg/sentry/platform/mmap_min_addr.go | 2 +- pkg/sentry/platform/platform.go | 4 +- pkg/sentry/platform/ptrace/BUILD | 4 +- pkg/sentry/platform/ptrace/ptrace.go | 2 +- pkg/sentry/platform/ptrace/ptrace_unsafe.go | 2 +- pkg/sentry/platform/ptrace/stub_unsafe.go | 4 +- pkg/sentry/platform/ptrace/subprocess.go | 2 +- pkg/sentry/platform/ring0/BUILD | 2 +- pkg/sentry/platform/ring0/defs_amd64.go | 2 +- pkg/sentry/platform/ring0/defs_arm64.go | 2 +- pkg/sentry/platform/ring0/gen_offsets/BUILD | 2 +- pkg/sentry/platform/ring0/pagetables/BUILD | 4 +- .../platform/ring0/pagetables/allocator_unsafe.go | 2 +- pkg/sentry/platform/ring0/pagetables/pagetables.go | 2 +- .../ring0/pagetables/pagetables_aarch64.go | 2 +- .../ring0/pagetables/pagetables_amd64_test.go | 2 +- .../ring0/pagetables/pagetables_arm64_test.go | 2 +- .../platform/ring0/pagetables/pagetables_test.go | 2 +- .../platform/ring0/pagetables/pagetables_x86.go | 2 +- pkg/sentry/platform/safecopy/BUILD | 29 - pkg/sentry/platform/safecopy/LICENSE | 27 - pkg/sentry/platform/safecopy/atomic_amd64.s | 136 ----- pkg/sentry/platform/safecopy/atomic_arm64.s | 126 ----- pkg/sentry/platform/safecopy/memclr_amd64.s | 147 ----- pkg/sentry/platform/safecopy/memclr_arm64.s | 74 --- pkg/sentry/platform/safecopy/memcpy_amd64.s | 250 --------- pkg/sentry/platform/safecopy/memcpy_arm64.s | 78 --- pkg/sentry/platform/safecopy/safecopy.go | 144 ----- pkg/sentry/platform/safecopy/safecopy_test.go | 617 --------------------- pkg/sentry/platform/safecopy/safecopy_unsafe.go | 335 ----------- pkg/sentry/platform/safecopy/sighandler_amd64.s | 133 ----- pkg/sentry/platform/safecopy/sighandler_arm64.s | 143 ----- pkg/sentry/safemem/BUILD | 27 - pkg/sentry/safemem/block_unsafe.go | 279 ---------- pkg/sentry/safemem/io.go | 392 ------------- pkg/sentry/safemem/io_test.go | 199 ------- pkg/sentry/safemem/safemem.go | 16 - pkg/sentry/safemem/seq_test.go | 196 ------- pkg/sentry/safemem/seq_unsafe.go | 299 ---------- pkg/sentry/socket/BUILD | 4 +- pkg/sentry/socket/control/BUILD | 4 +- pkg/sentry/socket/control/control.go | 4 +- pkg/sentry/socket/hostinet/BUILD | 6 +- pkg/sentry/socket/hostinet/socket.go | 6 +- pkg/sentry/socket/hostinet/socket_unsafe.go | 4 +- pkg/sentry/socket/hostinet/stack.go | 4 +- pkg/sentry/socket/netfilter/BUILD | 2 +- pkg/sentry/socket/netfilter/netfilter.go | 2 +- pkg/sentry/socket/netlink/BUILD | 4 +- pkg/sentry/socket/netlink/message.go | 2 +- pkg/sentry/socket/netlink/provider.go | 2 +- pkg/sentry/socket/netlink/route/BUILD | 2 +- pkg/sentry/socket/netlink/route/protocol.go | 2 +- pkg/sentry/socket/netlink/socket.go | 4 +- pkg/sentry/socket/netlink/uevent/BUILD | 2 +- pkg/sentry/socket/netlink/uevent/protocol.go | 2 +- pkg/sentry/socket/netstack/BUILD | 6 +- pkg/sentry/socket/netstack/netstack.go | 6 +- pkg/sentry/socket/netstack/provider.go | 2 +- pkg/sentry/socket/socket.go | 4 +- pkg/sentry/socket/unix/BUILD | 6 +- pkg/sentry/socket/unix/io.go | 4 +- pkg/sentry/socket/unix/transport/BUILD | 2 +- pkg/sentry/socket/unix/transport/connectioned.go | 2 +- pkg/sentry/socket/unix/transport/connectionless.go | 2 +- pkg/sentry/socket/unix/transport/unix.go | 2 +- pkg/sentry/socket/unix/unix.go | 4 +- pkg/sentry/strace/BUILD | 2 +- pkg/sentry/strace/poll.go | 2 +- pkg/sentry/strace/select.go | 2 +- pkg/sentry/strace/signal.go | 2 +- pkg/sentry/strace/socket.go | 2 +- pkg/sentry/strace/strace.go | 2 +- pkg/sentry/syscalls/linux/BUILD | 6 +- pkg/sentry/syscalls/linux/linux64_amd64.go | 2 +- pkg/sentry/syscalls/linux/linux64_arm64.go | 2 +- pkg/sentry/syscalls/linux/sigset.go | 2 +- pkg/sentry/syscalls/linux/sys_aio.go | 2 +- pkg/sentry/syscalls/linux/sys_epoll.go | 2 +- pkg/sentry/syscalls/linux/sys_file.go | 4 +- pkg/sentry/syscalls/linux/sys_futex.go | 2 +- pkg/sentry/syscalls/linux/sys_getdents.go | 2 +- pkg/sentry/syscalls/linux/sys_mempolicy.go | 2 +- pkg/sentry/syscalls/linux/sys_mmap.go | 2 +- pkg/sentry/syscalls/linux/sys_mount.go | 2 +- pkg/sentry/syscalls/linux/sys_pipe.go | 2 +- pkg/sentry/syscalls/linux/sys_poll.go | 2 +- pkg/sentry/syscalls/linux/sys_random.go | 4 +- pkg/sentry/syscalls/linux/sys_read.go | 2 +- pkg/sentry/syscalls/linux/sys_rlimit.go | 2 +- pkg/sentry/syscalls/linux/sys_seccomp.go | 2 +- pkg/sentry/syscalls/linux/sys_sem.go | 2 +- pkg/sentry/syscalls/linux/sys_signal.go | 2 +- pkg/sentry/syscalls/linux/sys_socket.go | 2 +- pkg/sentry/syscalls/linux/sys_stat.go | 2 +- pkg/sentry/syscalls/linux/sys_stat_amd64.go | 2 +- pkg/sentry/syscalls/linux/sys_stat_arm64.go | 2 +- pkg/sentry/syscalls/linux/sys_thread.go | 2 +- pkg/sentry/syscalls/linux/sys_time.go | 2 +- pkg/sentry/syscalls/linux/sys_timer.go | 2 +- pkg/sentry/syscalls/linux/sys_write.go | 2 +- pkg/sentry/syscalls/linux/sys_xattr.go | 2 +- pkg/sentry/syscalls/linux/timespec.go | 2 +- pkg/sentry/unimpl/BUILD | 2 +- pkg/sentry/unimpl/events.go | 2 +- pkg/sentry/uniqueid/BUILD | 2 +- pkg/sentry/uniqueid/context.go | 2 +- pkg/sentry/usermem/BUILD | 55 -- pkg/sentry/usermem/README.md | 31 -- pkg/sentry/usermem/access_type.go | 128 ----- pkg/sentry/usermem/addr.go | 108 ---- pkg/sentry/usermem/addr_range_seq_test.go | 197 ------- pkg/sentry/usermem/addr_range_seq_unsafe.go | 277 --------- pkg/sentry/usermem/bytes_io.go | 141 ----- pkg/sentry/usermem/bytes_io_unsafe.go | 47 -- pkg/sentry/usermem/usermem.go | 597 -------------------- pkg/sentry/usermem/usermem_arm64.go | 53 -- pkg/sentry/usermem/usermem_test.go | 424 -------------- pkg/sentry/usermem/usermem_unsafe.go | 27 - pkg/sentry/usermem/usermem_x86.go | 38 -- pkg/sentry/vfs/BUILD | 10 +- pkg/sentry/vfs/context.go | 2 +- pkg/sentry/vfs/device.go | 2 +- pkg/sentry/vfs/file_description.go | 4 +- pkg/sentry/vfs/file_description_impl_util.go | 4 +- pkg/sentry/vfs/file_description_impl_util_test.go | 6 +- pkg/sentry/vfs/filesystem.go | 2 +- pkg/sentry/vfs/filesystem_type.go | 2 +- pkg/sentry/vfs/mount.go | 2 +- pkg/sentry/vfs/pathname.go | 2 +- pkg/sentry/vfs/testutil.go | 2 +- pkg/sentry/vfs/vfs.go | 2 +- pkg/usermem/BUILD | 55 ++ pkg/usermem/README.md | 31 ++ pkg/usermem/access_type.go | 128 +++++ pkg/usermem/addr.go | 108 ++++ pkg/usermem/addr_range_seq_test.go | 197 +++++++ pkg/usermem/addr_range_seq_unsafe.go | 277 +++++++++ pkg/usermem/bytes_io.go | 141 +++++ pkg/usermem/bytes_io_unsafe.go | 47 ++ pkg/usermem/usermem.go | 597 ++++++++++++++++++++ pkg/usermem/usermem_arm64.go | 53 ++ pkg/usermem/usermem_test.go | 424 ++++++++++++++ pkg/usermem/usermem_unsafe.go | 27 + pkg/usermem/usermem_x86.go | 38 ++ runsc/boot/BUILD | 6 +- runsc/boot/fds.go | 2 +- runsc/boot/fs.go | 2 +- runsc/boot/loader_test.go | 2 +- runsc/boot/user.go | 4 +- runsc/boot/user_test.go | 2 +- tools/go_marshal/defs.bzl | 4 +- tools/go_marshal/gomarshal/generator.go | 4 +- tools/go_marshal/test/BUILD | 2 +- tools/go_marshal/test/benchmark_test.go | 2 +- 483 files changed, 6839 insertions(+), 6835 deletions(-) create mode 100644 pkg/context/BUILD create mode 100644 pkg/context/context.go create mode 100644 pkg/safecopy/BUILD create mode 100644 pkg/safecopy/LICENSE create mode 100644 pkg/safecopy/atomic_amd64.s create mode 100644 pkg/safecopy/atomic_arm64.s create mode 100644 pkg/safecopy/memclr_amd64.s create mode 100644 pkg/safecopy/memclr_arm64.s create mode 100644 pkg/safecopy/memcpy_amd64.s create mode 100644 pkg/safecopy/memcpy_arm64.s create mode 100644 pkg/safecopy/safecopy.go create mode 100644 pkg/safecopy/safecopy_test.go create mode 100644 pkg/safecopy/safecopy_unsafe.go create mode 100644 pkg/safecopy/sighandler_amd64.s create mode 100644 pkg/safecopy/sighandler_arm64.s create mode 100644 pkg/safemem/BUILD create mode 100644 pkg/safemem/block_unsafe.go create mode 100644 pkg/safemem/io.go create mode 100644 pkg/safemem/io_test.go create mode 100644 pkg/safemem/safemem.go create mode 100644 pkg/safemem/seq_test.go create mode 100644 pkg/safemem/seq_unsafe.go delete mode 100644 pkg/sentry/context/BUILD delete mode 100644 pkg/sentry/context/context.go delete mode 100644 pkg/sentry/context/contexttest/BUILD delete mode 100644 pkg/sentry/context/contexttest/contexttest.go create mode 100644 pkg/sentry/contexttest/BUILD create mode 100644 pkg/sentry/contexttest/contexttest.go delete mode 100644 pkg/sentry/platform/safecopy/BUILD delete mode 100644 pkg/sentry/platform/safecopy/LICENSE delete mode 100644 pkg/sentry/platform/safecopy/atomic_amd64.s delete mode 100644 pkg/sentry/platform/safecopy/atomic_arm64.s delete mode 100644 pkg/sentry/platform/safecopy/memclr_amd64.s delete mode 100644 pkg/sentry/platform/safecopy/memclr_arm64.s delete mode 100644 pkg/sentry/platform/safecopy/memcpy_amd64.s delete mode 100644 pkg/sentry/platform/safecopy/memcpy_arm64.s delete mode 100644 pkg/sentry/platform/safecopy/safecopy.go delete mode 100644 pkg/sentry/platform/safecopy/safecopy_test.go delete mode 100644 pkg/sentry/platform/safecopy/safecopy_unsafe.go delete mode 100644 pkg/sentry/platform/safecopy/sighandler_amd64.s delete mode 100644 pkg/sentry/platform/safecopy/sighandler_arm64.s delete mode 100644 pkg/sentry/safemem/BUILD delete mode 100644 pkg/sentry/safemem/block_unsafe.go delete mode 100644 pkg/sentry/safemem/io.go delete mode 100644 pkg/sentry/safemem/io_test.go delete mode 100644 pkg/sentry/safemem/safemem.go delete mode 100644 pkg/sentry/safemem/seq_test.go delete mode 100644 pkg/sentry/safemem/seq_unsafe.go delete mode 100644 pkg/sentry/usermem/BUILD delete mode 100644 pkg/sentry/usermem/README.md delete mode 100644 pkg/sentry/usermem/access_type.go delete mode 100644 pkg/sentry/usermem/addr.go delete mode 100644 pkg/sentry/usermem/addr_range_seq_test.go delete mode 100644 pkg/sentry/usermem/addr_range_seq_unsafe.go delete mode 100644 pkg/sentry/usermem/bytes_io.go delete mode 100644 pkg/sentry/usermem/bytes_io_unsafe.go delete mode 100644 pkg/sentry/usermem/usermem.go delete mode 100644 pkg/sentry/usermem/usermem_arm64.go delete mode 100644 pkg/sentry/usermem/usermem_test.go delete mode 100644 pkg/sentry/usermem/usermem_unsafe.go delete mode 100644 pkg/sentry/usermem/usermem_x86.go create mode 100644 pkg/usermem/BUILD create mode 100644 pkg/usermem/README.md create mode 100644 pkg/usermem/access_type.go create mode 100644 pkg/usermem/addr.go create mode 100644 pkg/usermem/addr_range_seq_test.go create mode 100644 pkg/usermem/addr_range_seq_unsafe.go create mode 100644 pkg/usermem/bytes_io.go create mode 100644 pkg/usermem/bytes_io_unsafe.go create mode 100644 pkg/usermem/usermem.go create mode 100644 pkg/usermem/usermem_arm64.go create mode 100644 pkg/usermem/usermem_test.go create mode 100644 pkg/usermem/usermem_unsafe.go create mode 100644 pkg/usermem/usermem_x86.go (limited to 'pkg/sentry') diff --git a/pkg/abi/abi.go b/pkg/abi/abi.go index d56c481c9..e6be93c3a 100644 --- a/pkg/abi/abi.go +++ b/pkg/abi/abi.go @@ -39,3 +39,7 @@ func (o OS) String() string { return fmt.Sprintf("OS(%d)", o) } } + +// ABI is an interface that defines OS-specific interactions. +type ABI interface { +} diff --git a/pkg/context/BUILD b/pkg/context/BUILD new file mode 100644 index 000000000..239f31149 --- /dev/null +++ b/pkg/context/BUILD @@ -0,0 +1,13 @@ +load("//tools:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_library( + name = "context", + srcs = ["context.go"], + visibility = ["//:sandbox"], + deps = [ + "//pkg/amutex", + "//pkg/log", + ], +) diff --git a/pkg/context/context.go b/pkg/context/context.go new file mode 100644 index 000000000..23e009ef3 --- /dev/null +++ b/pkg/context/context.go @@ -0,0 +1,141 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package context defines an internal context type. +// +// The given Context conforms to the standard Go context, but mandates +// additional methods that are specific to the kernel internals. Note however, +// that the Context described by this package carries additional constraints +// regarding concurrent access and retaining beyond the scope of a call. +// +// See the Context type for complete details. +package context + +import ( + "context" + "time" + + "gvisor.dev/gvisor/pkg/amutex" + "gvisor.dev/gvisor/pkg/log" +) + +type contextID int + +// Globally accessible values from a context. These keys are defined in the +// context package to resolve dependency cycles by not requiring the caller to +// import packages usually required to get these information. +const ( + // CtxThreadGroupID is the current thread group ID when a context represents + // a task context. The value is represented as an int32. + CtxThreadGroupID contextID = iota +) + +// ThreadGroupIDFromContext returns the current thread group ID when ctx +// represents a task context. +func ThreadGroupIDFromContext(ctx Context) (tgid int32, ok bool) { + if tgid := ctx.Value(CtxThreadGroupID); tgid != nil { + return tgid.(int32), true + } + return 0, false +} + +// A Context represents a thread of execution (hereafter "goroutine" to reflect +// Go idiosyncrasy). It carries state associated with the goroutine across API +// boundaries. +// +// While Context exists for essentially the same reasons as Go's standard +// context.Context, the standard type represents the state of an operation +// rather than that of a goroutine. This is a critical distinction: +// +// - Unlike context.Context, which "may be passed to functions running in +// different goroutines", it is *not safe* to use the same Context in multiple +// concurrent goroutines. +// +// - It is *not safe* to retain a Context passed to a function beyond the scope +// of that function call. +// +// In both cases, values extracted from the Context should be used instead. +type Context interface { + log.Logger + amutex.Sleeper + context.Context + + // UninterruptibleSleepStart indicates the beginning of an uninterruptible + // sleep state (equivalent to Linux's TASK_UNINTERRUPTIBLE). If deactivate + // is true and the Context represents a Task, the Task's AddressSpace is + // deactivated. + UninterruptibleSleepStart(deactivate bool) + + // UninterruptibleSleepFinish indicates the end of an uninterruptible sleep + // state that was begun by a previous call to UninterruptibleSleepStart. If + // activate is true and the Context represents a Task, the Task's + // AddressSpace is activated. Normally activate is the same value as the + // deactivate parameter passed to UninterruptibleSleepStart. + UninterruptibleSleepFinish(activate bool) +} + +// NoopSleeper is a noop implementation of amutex.Sleeper and UninterruptibleSleep +// methods for anonymous embedding in other types that do not implement sleeps. +type NoopSleeper struct { + amutex.NoopSleeper +} + +// UninterruptibleSleepStart does nothing. +func (NoopSleeper) UninterruptibleSleepStart(bool) {} + +// UninterruptibleSleepFinish does nothing. +func (NoopSleeper) UninterruptibleSleepFinish(bool) {} + +// Deadline returns zero values, meaning no deadline. +func (NoopSleeper) Deadline() (time.Time, bool) { + return time.Time{}, false +} + +// Done returns nil. +func (NoopSleeper) Done() <-chan struct{} { + return nil +} + +// Err returns nil. +func (NoopSleeper) Err() error { + return nil +} + +// logContext implements basic logging. +type logContext struct { + log.Logger + NoopSleeper +} + +// Value implements Context.Value. +func (logContext) Value(key interface{}) interface{} { + return nil +} + +// bgContext is the context returned by context.Background. +var bgContext = &logContext{Logger: log.Log()} + +// Background returns an empty context using the default logger. +// +// Users should be wary of using a Background context. Please tag any use with +// FIXME(b/38173783) and a note to remove this use. +// +// Generally, one should use the Task as their context when available, or avoid +// having to use a context in places where a Task is unavailable. +// +// Using a Background context for tests is fine, as long as no values are +// needed from the context in the tested code paths. +func Background() Context { + return bgContext +} diff --git a/pkg/safecopy/BUILD b/pkg/safecopy/BUILD new file mode 100644 index 000000000..426ef30c9 --- /dev/null +++ b/pkg/safecopy/BUILD @@ -0,0 +1,29 @@ +load("//tools:defs.bzl", "go_library", "go_test") + +package(licenses = ["notice"]) + +go_library( + name = "safecopy", + srcs = [ + "atomic_amd64.s", + "atomic_arm64.s", + "memclr_amd64.s", + "memclr_arm64.s", + "memcpy_amd64.s", + "memcpy_arm64.s", + "safecopy.go", + "safecopy_unsafe.go", + "sighandler_amd64.s", + "sighandler_arm64.s", + ], + visibility = ["//:sandbox"], + deps = ["//pkg/syserror"], +) + +go_test( + name = "safecopy_test", + srcs = [ + "safecopy_test.go", + ], + library = ":safecopy", +) diff --git a/pkg/safecopy/LICENSE b/pkg/safecopy/LICENSE new file mode 100644 index 000000000..6a66aea5e --- /dev/null +++ b/pkg/safecopy/LICENSE @@ -0,0 +1,27 @@ +Copyright (c) 2009 The Go Authors. All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are +met: + + * Redistributions of source code must retain the above copyright +notice, this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above +copyright notice, this list of conditions and the following disclaimer +in the documentation and/or other materials provided with the +distribution. + * Neither the name of Google Inc. nor the names of its +contributors may be used to endorse or promote products derived from +this software without specific prior written permission. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/pkg/safecopy/atomic_amd64.s b/pkg/safecopy/atomic_amd64.s new file mode 100644 index 000000000..a0cd78f33 --- /dev/null +++ b/pkg/safecopy/atomic_amd64.s @@ -0,0 +1,136 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "textflag.h" + +// handleSwapUint32Fault returns the value stored in DI. Control is transferred +// to it when swapUint32 below receives SIGSEGV or SIGBUS, with the signal +// number stored in DI. +// +// It must have the same frame configuration as swapUint32 so that it can undo +// any potential call frame set up by the assembler. +TEXT handleSwapUint32Fault(SB), NOSPLIT, $0-24 + MOVL DI, sig+20(FP) + RET + +// swapUint32 atomically stores new into *addr and returns (the previous *addr +// value, 0). If a SIGSEGV or SIGBUS signal is received during the swap, the +// value of old is unspecified, and sig is the number of the signal that was +// received. +// +// Preconditions: addr must be aligned to a 4-byte boundary. +// +//func swapUint32(ptr unsafe.Pointer, new uint32) (old uint32, sig int32) +TEXT ·swapUint32(SB), NOSPLIT, $0-24 + // Store 0 as the returned signal number. If we run to completion, + // this is the value the caller will see; if a signal is received, + // handleSwapUint32Fault will store a different value in this address. + MOVL $0, sig+20(FP) + + MOVQ addr+0(FP), DI + MOVL new+8(FP), AX + XCHGL AX, 0(DI) + MOVL AX, old+16(FP) + RET + +// handleSwapUint64Fault returns the value stored in DI. Control is transferred +// to it when swapUint64 below receives SIGSEGV or SIGBUS, with the signal +// number stored in DI. +// +// It must have the same frame configuration as swapUint64 so that it can undo +// any potential call frame set up by the assembler. +TEXT handleSwapUint64Fault(SB), NOSPLIT, $0-28 + MOVL DI, sig+24(FP) + RET + +// swapUint64 atomically stores new into *addr and returns (the previous *addr +// value, 0). If a SIGSEGV or SIGBUS signal is received during the swap, the +// value of old is unspecified, and sig is the number of the signal that was +// received. +// +// Preconditions: addr must be aligned to a 8-byte boundary. +// +//func swapUint64(ptr unsafe.Pointer, new uint64) (old uint64, sig int32) +TEXT ·swapUint64(SB), NOSPLIT, $0-28 + // Store 0 as the returned signal number. If we run to completion, + // this is the value the caller will see; if a signal is received, + // handleSwapUint64Fault will store a different value in this address. + MOVL $0, sig+24(FP) + + MOVQ addr+0(FP), DI + MOVQ new+8(FP), AX + XCHGQ AX, 0(DI) + MOVQ AX, old+16(FP) + RET + +// handleCompareAndSwapUint32Fault returns the value stored in DI. Control is +// transferred to it when swapUint64 below receives SIGSEGV or SIGBUS, with the +// signal number stored in DI. +// +// It must have the same frame configuration as compareAndSwapUint32 so that it +// can undo any potential call frame set up by the assembler. +TEXT handleCompareAndSwapUint32Fault(SB), NOSPLIT, $0-24 + MOVL DI, sig+20(FP) + RET + +// compareAndSwapUint32 is like sync/atomic.CompareAndSwapUint32, but returns +// (the value previously stored at addr, 0). If a SIGSEGV or SIGBUS signal is +// received during the operation, the value of prev is unspecified, and sig is +// the number of the signal that was received. +// +// Preconditions: addr must be aligned to a 4-byte boundary. +// +//func compareAndSwapUint32(ptr unsafe.Pointer, old, new uint32) (prev uint32, sig int32) +TEXT ·compareAndSwapUint32(SB), NOSPLIT, $0-24 + // Store 0 as the returned signal number. If we run to completion, this is + // the value the caller will see; if a signal is received, + // handleCompareAndSwapUint32Fault will store a different value in this + // address. + MOVL $0, sig+20(FP) + + MOVQ addr+0(FP), DI + MOVL old+8(FP), AX + MOVL new+12(FP), DX + LOCK + CMPXCHGL DX, 0(DI) + MOVL AX, prev+16(FP) + RET + +// handleLoadUint32Fault returns the value stored in DI. Control is transferred +// to it when LoadUint32 below receives SIGSEGV or SIGBUS, with the signal +// number stored in DI. +// +// It must have the same frame configuration as loadUint32 so that it can undo +// any potential call frame set up by the assembler. +TEXT handleLoadUint32Fault(SB), NOSPLIT, $0-16 + MOVL DI, sig+12(FP) + RET + +// loadUint32 atomically loads *addr and returns it. If a SIGSEGV or SIGBUS +// signal is received, the value returned is unspecified, and sig is the number +// of the signal that was received. +// +// Preconditions: addr must be aligned to a 4-byte boundary. +// +//func loadUint32(ptr unsafe.Pointer) (val uint32, sig int32) +TEXT ·loadUint32(SB), NOSPLIT, $0-16 + // Store 0 as the returned signal number. If we run to completion, + // this is the value the caller will see; if a signal is received, + // handleLoadUint32Fault will store a different value in this address. + MOVL $0, sig+12(FP) + + MOVQ addr+0(FP), AX + MOVL (AX), BX + MOVL BX, val+8(FP) + RET diff --git a/pkg/safecopy/atomic_arm64.s b/pkg/safecopy/atomic_arm64.s new file mode 100644 index 000000000..d58ed71f7 --- /dev/null +++ b/pkg/safecopy/atomic_arm64.s @@ -0,0 +1,126 @@ +// Copyright 2014 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +#include "textflag.h" + +// handleSwapUint32Fault returns the value stored in R1. Control is transferred +// to it when swapUint32 below receives SIGSEGV or SIGBUS, with the signal +// number stored in R1. +// +// It must have the same frame configuration as swapUint32 so that it can undo +// any potential call frame set up by the assembler. +TEXT handleSwapUint32Fault(SB), NOSPLIT, $0-24 + MOVW R1, sig+20(FP) + RET + +// See the corresponding doc in safecopy_unsafe.go +// +// The code is derived from Go source runtime/internal/atomic.Xchg. +// +//func swapUint32(ptr unsafe.Pointer, new uint32) (old uint32, sig int32) +TEXT ·swapUint32(SB), NOSPLIT, $0-24 + // Store 0 as the returned signal number. If we run to completion, + // this is the value the caller will see; if a signal is received, + // handleSwapUint32Fault will store a different value in this address. + MOVW $0, sig+20(FP) +again: + MOVD addr+0(FP), R0 + MOVW new+8(FP), R1 + LDAXRW (R0), R2 + STLXRW R1, (R0), R3 + CBNZ R3, again + MOVW R2, old+16(FP) + RET + +// handleSwapUint64Fault returns the value stored in R1. Control is transferred +// to it when swapUint64 below receives SIGSEGV or SIGBUS, with the signal +// number stored in R1. +// +// It must have the same frame configuration as swapUint64 so that it can undo +// any potential call frame set up by the assembler. +TEXT handleSwapUint64Fault(SB), NOSPLIT, $0-28 + MOVW R1, sig+24(FP) + RET + +// See the corresponding doc in safecopy_unsafe.go +// +// The code is derived from Go source runtime/internal/atomic.Xchg64. +// +//func swapUint64(ptr unsafe.Pointer, new uint64) (old uint64, sig int32) +TEXT ·swapUint64(SB), NOSPLIT, $0-28 + // Store 0 as the returned signal number. If we run to completion, + // this is the value the caller will see; if a signal is received, + // handleSwapUint64Fault will store a different value in this address. + MOVW $0, sig+24(FP) +again: + MOVD addr+0(FP), R0 + MOVD new+8(FP), R1 + LDAXR (R0), R2 + STLXR R1, (R0), R3 + CBNZ R3, again + MOVD R2, old+16(FP) + RET + +// handleCompareAndSwapUint32Fault returns the value stored in R1. Control is +// transferred to it when compareAndSwapUint32 below receives SIGSEGV or SIGBUS, +// with the signal number stored in R1. +// +// It must have the same frame configuration as compareAndSwapUint32 so that it +// can undo any potential call frame set up by the assembler. +TEXT handleCompareAndSwapUint32Fault(SB), NOSPLIT, $0-24 + MOVW R1, sig+20(FP) + RET + +// See the corresponding doc in safecopy_unsafe.go +// +// The code is derived from Go source runtime/internal/atomic.Cas. +// +//func compareAndSwapUint32(ptr unsafe.Pointer, old, new uint32) (prev uint32, sig int32) +TEXT ·compareAndSwapUint32(SB), NOSPLIT, $0-24 + // Store 0 as the returned signal number. If we run to completion, this is + // the value the caller will see; if a signal is received, + // handleCompareAndSwapUint32Fault will store a different value in this + // address. + MOVW $0, sig+20(FP) + + MOVD addr+0(FP), R0 + MOVW old+8(FP), R1 + MOVW new+12(FP), R2 +again: + LDAXRW (R0), R3 + CMPW R1, R3 + BNE done + STLXRW R2, (R0), R4 + CBNZ R4, again +done: + MOVW R3, prev+16(FP) + RET + +// handleLoadUint32Fault returns the value stored in DI. Control is transferred +// to it when LoadUint32 below receives SIGSEGV or SIGBUS, with the signal +// number stored in DI. +// +// It must have the same frame configuration as loadUint32 so that it can undo +// any potential call frame set up by the assembler. +TEXT handleLoadUint32Fault(SB), NOSPLIT, $0-16 + MOVW R1, sig+12(FP) + RET + +// loadUint32 atomically loads *addr and returns it. If a SIGSEGV or SIGBUS +// signal is received, the value returned is unspecified, and sig is the number +// of the signal that was received. +// +// Preconditions: addr must be aligned to a 4-byte boundary. +// +//func loadUint32(ptr unsafe.Pointer) (val uint32, sig int32) +TEXT ·loadUint32(SB), NOSPLIT, $0-16 + // Store 0 as the returned signal number. If we run to completion, + // this is the value the caller will see; if a signal is received, + // handleLoadUint32Fault will store a different value in this address. + MOVW $0, sig+12(FP) + + MOVD addr+0(FP), R0 + LDARW (R0), R1 + MOVW R1, val+8(FP) + RET diff --git a/pkg/safecopy/memclr_amd64.s b/pkg/safecopy/memclr_amd64.s new file mode 100644 index 000000000..64cf32f05 --- /dev/null +++ b/pkg/safecopy/memclr_amd64.s @@ -0,0 +1,147 @@ +// Copyright 2014 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +#include "textflag.h" + +// handleMemclrFault returns (the value stored in AX, the value stored in DI). +// Control is transferred to it when memclr below receives SIGSEGV or SIGBUS, +// with the faulting address stored in AX and the signal number stored in DI. +// +// It must have the same frame configuration as memclr so that it can undo any +// potential call frame set up by the assembler. +TEXT handleMemclrFault(SB), NOSPLIT, $0-28 + MOVQ AX, addr+16(FP) + MOVL DI, sig+24(FP) + RET + +// memclr sets the n bytes following ptr to zeroes. If a SIGSEGV or SIGBUS +// signal is received during the write, it returns the address that caused the +// fault and the number of the signal that was received. Otherwise, it returns +// an unspecified address and a signal number of 0. +// +// Data is written in order, such that if a fault happens at address p, it is +// safe to assume that all data before p-maxRegisterSize has already been +// successfully written. +// +// The code is derived from runtime.memclrNoHeapPointers. +// +// func memclr(ptr unsafe.Pointer, n uintptr) (fault unsafe.Pointer, sig int32) +TEXT ·memclr(SB), NOSPLIT, $0-28 + // Store 0 as the returned signal number. If we run to completion, + // this is the value the caller will see; if a signal is received, + // handleMemclrFault will store a different value in this address. + MOVL $0, sig+24(FP) + + MOVQ ptr+0(FP), DI + MOVQ n+8(FP), BX + XORQ AX, AX + + // MOVOU seems always faster than REP STOSQ. +tail: + TESTQ BX, BX + JEQ _0 + CMPQ BX, $2 + JBE _1or2 + CMPQ BX, $4 + JBE _3or4 + CMPQ BX, $8 + JB _5through7 + JE _8 + CMPQ BX, $16 + JBE _9through16 + PXOR X0, X0 + CMPQ BX, $32 + JBE _17through32 + CMPQ BX, $64 + JBE _33through64 + CMPQ BX, $128 + JBE _65through128 + CMPQ BX, $256 + JBE _129through256 + // TODO: use branch table and BSR to make this just a single dispatch + // TODO: for really big clears, use MOVNTDQ, even without AVX2. + +loop: + MOVOU X0, 0(DI) + MOVOU X0, 16(DI) + MOVOU X0, 32(DI) + MOVOU X0, 48(DI) + MOVOU X0, 64(DI) + MOVOU X0, 80(DI) + MOVOU X0, 96(DI) + MOVOU X0, 112(DI) + MOVOU X0, 128(DI) + MOVOU X0, 144(DI) + MOVOU X0, 160(DI) + MOVOU X0, 176(DI) + MOVOU X0, 192(DI) + MOVOU X0, 208(DI) + MOVOU X0, 224(DI) + MOVOU X0, 240(DI) + SUBQ $256, BX + ADDQ $256, DI + CMPQ BX, $256 + JAE loop + JMP tail + +_1or2: + MOVB AX, (DI) + MOVB AX, -1(DI)(BX*1) + RET +_0: + RET +_3or4: + MOVW AX, (DI) + MOVW AX, -2(DI)(BX*1) + RET +_5through7: + MOVL AX, (DI) + MOVL AX, -4(DI)(BX*1) + RET +_8: + // We need a separate case for 8 to make sure we clear pointers atomically. + MOVQ AX, (DI) + RET +_9through16: + MOVQ AX, (DI) + MOVQ AX, -8(DI)(BX*1) + RET +_17through32: + MOVOU X0, (DI) + MOVOU X0, -16(DI)(BX*1) + RET +_33through64: + MOVOU X0, (DI) + MOVOU X0, 16(DI) + MOVOU X0, -32(DI)(BX*1) + MOVOU X0, -16(DI)(BX*1) + RET +_65through128: + MOVOU X0, (DI) + MOVOU X0, 16(DI) + MOVOU X0, 32(DI) + MOVOU X0, 48(DI) + MOVOU X0, -64(DI)(BX*1) + MOVOU X0, -48(DI)(BX*1) + MOVOU X0, -32(DI)(BX*1) + MOVOU X0, -16(DI)(BX*1) + RET +_129through256: + MOVOU X0, (DI) + MOVOU X0, 16(DI) + MOVOU X0, 32(DI) + MOVOU X0, 48(DI) + MOVOU X0, 64(DI) + MOVOU X0, 80(DI) + MOVOU X0, 96(DI) + MOVOU X0, 112(DI) + MOVOU X0, -128(DI)(BX*1) + MOVOU X0, -112(DI)(BX*1) + MOVOU X0, -96(DI)(BX*1) + MOVOU X0, -80(DI)(BX*1) + MOVOU X0, -64(DI)(BX*1) + MOVOU X0, -48(DI)(BX*1) + MOVOU X0, -32(DI)(BX*1) + MOVOU X0, -16(DI)(BX*1) + RET diff --git a/pkg/safecopy/memclr_arm64.s b/pkg/safecopy/memclr_arm64.s new file mode 100644 index 000000000..7361b9067 --- /dev/null +++ b/pkg/safecopy/memclr_arm64.s @@ -0,0 +1,74 @@ +// Copyright 2014 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +#include "textflag.h" + +// handleMemclrFault returns (the value stored in R0, the value stored in R1). +// Control is transferred to it when memclr below receives SIGSEGV or SIGBUS, +// with the faulting address stored in R0 and the signal number stored in R1. +// +// It must have the same frame configuration as memclr so that it can undo any +// potential call frame set up by the assembler. +TEXT handleMemclrFault(SB), NOSPLIT, $0-28 + MOVD R0, addr+16(FP) + MOVW R1, sig+24(FP) + RET + +// See the corresponding doc in safecopy_unsafe.go +// +// The code is derived from runtime.memclrNoHeapPointers. +// +// func memclr(ptr unsafe.Pointer, n uintptr) (fault unsafe.Pointer, sig int32) +TEXT ·memclr(SB), NOSPLIT, $0-28 + // Store 0 as the returned signal number. If we run to completion, + // this is the value the caller will see; if a signal is received, + // handleMemclrFault will store a different value in this address. + MOVW $0, sig+24(FP) + MOVD ptr+0(FP), R0 + MOVD n+8(FP), R1 + + // If size is less than 16 bytes, use tail_zero to zero what remains + CMP $16, R1 + BLT tail_zero + // Get buffer offset into 16 byte aligned address for better performance + ANDS $15, R0, ZR + BNE unaligned_to_16 +aligned_to_16: + LSR $4, R1, R2 +zero_by_16: + STP.P (ZR, ZR), 16(R0) // Store pair with post index. + SUBS $1, R2, R2 + BNE zero_by_16 + ANDS $15, R1, R1 + BEQ end + + // Zero buffer with size=R1 < 16 +tail_zero: + TBZ $3, R1, tail_zero_4 + MOVD.P ZR, 8(R0) +tail_zero_4: + TBZ $2, R1, tail_zero_2 + MOVW.P ZR, 4(R0) +tail_zero_2: + TBZ $1, R1, tail_zero_1 + MOVH.P ZR, 2(R0) +tail_zero_1: + TBZ $0, R1, end + MOVB ZR, (R0) +end: + RET + +unaligned_to_16: + MOVD R0, R2 +head_loop: + MOVBU.P ZR, 1(R0) + ANDS $15, R0, ZR + BNE head_loop + // Adjust length for what remains + SUB R2, R0, R3 + SUB R3, R1 + // If size is less than 16 bytes, use tail_zero to zero what remains + CMP $16, R1 + BLT tail_zero + B aligned_to_16 diff --git a/pkg/safecopy/memcpy_amd64.s b/pkg/safecopy/memcpy_amd64.s new file mode 100644 index 000000000..129691d68 --- /dev/null +++ b/pkg/safecopy/memcpy_amd64.s @@ -0,0 +1,250 @@ +// Copyright © 1994-1999 Lucent Technologies Inc. All rights reserved. +// Revisions Copyright © 2000-2007 Vita Nuova Holdings Limited (www.vitanuova.com). All rights reserved. +// Portions Copyright 2009 The Go Authors. All rights reserved. +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +// THE SOFTWARE. + +#include "textflag.h" + +// handleMemcpyFault returns (the value stored in AX, the value stored in DI). +// Control is transferred to it when memcpy below receives SIGSEGV or SIGBUS, +// with the faulting address stored in AX and the signal number stored in DI. +// +// It must have the same frame configuration as memcpy so that it can undo any +// potential call frame set up by the assembler. +TEXT handleMemcpyFault(SB), NOSPLIT, $0-36 + MOVQ AX, addr+24(FP) + MOVL DI, sig+32(FP) + RET + +// memcpy copies data from src to dst. If a SIGSEGV or SIGBUS signal is received +// during the copy, it returns the address that caused the fault and the number +// of the signal that was received. Otherwise, it returns an unspecified address +// and a signal number of 0. +// +// Data is copied in order, such that if a fault happens at address p, it is +// safe to assume that all data before p-maxRegisterSize has already been +// successfully copied. +// +// The code is derived from the forward copying part of runtime.memmove. +// +// func memcpy(dst, src unsafe.Pointer, n uintptr) (fault unsafe.Pointer, sig int32) +TEXT ·memcpy(SB), NOSPLIT, $0-36 + // Store 0 as the returned signal number. If we run to completion, + // this is the value the caller will see; if a signal is received, + // handleMemcpyFault will store a different value in this address. + MOVL $0, sig+32(FP) + + MOVQ to+0(FP), DI + MOVQ from+8(FP), SI + MOVQ n+16(FP), BX + + // REP instructions have a high startup cost, so we handle small sizes + // with some straightline code. The REP MOVSQ instruction is really fast + // for large sizes. The cutover is approximately 2K. +tail: + // move_129through256 or smaller work whether or not the source and the + // destination memory regions overlap because they load all data into + // registers before writing it back. move_256through2048 on the other + // hand can be used only when the memory regions don't overlap or the copy + // direction is forward. + TESTQ BX, BX + JEQ move_0 + CMPQ BX, $2 + JBE move_1or2 + CMPQ BX, $4 + JBE move_3or4 + CMPQ BX, $8 + JB move_5through7 + JE move_8 + CMPQ BX, $16 + JBE move_9through16 + CMPQ BX, $32 + JBE move_17through32 + CMPQ BX, $64 + JBE move_33through64 + CMPQ BX, $128 + JBE move_65through128 + CMPQ BX, $256 + JBE move_129through256 + // TODO: use branch table and BSR to make this just a single dispatch + +/* + * forward copy loop + */ + CMPQ BX, $2048 + JLS move_256through2048 + + // Check alignment + MOVL SI, AX + ORL DI, AX + TESTL $7, AX + JEQ fwdBy8 + + // Do 1 byte at a time + MOVQ BX, CX + REP; MOVSB + RET + +fwdBy8: + // Do 8 bytes at a time + MOVQ BX, CX + SHRQ $3, CX + ANDQ $7, BX + REP; MOVSQ + JMP tail + +move_1or2: + MOVB (SI), AX + MOVB AX, (DI) + MOVB -1(SI)(BX*1), CX + MOVB CX, -1(DI)(BX*1) + RET +move_0: + RET +move_3or4: + MOVW (SI), AX + MOVW AX, (DI) + MOVW -2(SI)(BX*1), CX + MOVW CX, -2(DI)(BX*1) + RET +move_5through7: + MOVL (SI), AX + MOVL AX, (DI) + MOVL -4(SI)(BX*1), CX + MOVL CX, -4(DI)(BX*1) + RET +move_8: + // We need a separate case for 8 to make sure we write pointers atomically. + MOVQ (SI), AX + MOVQ AX, (DI) + RET +move_9through16: + MOVQ (SI), AX + MOVQ AX, (DI) + MOVQ -8(SI)(BX*1), CX + MOVQ CX, -8(DI)(BX*1) + RET +move_17through32: + MOVOU (SI), X0 + MOVOU X0, (DI) + MOVOU -16(SI)(BX*1), X1 + MOVOU X1, -16(DI)(BX*1) + RET +move_33through64: + MOVOU (SI), X0 + MOVOU X0, (DI) + MOVOU 16(SI), X1 + MOVOU X1, 16(DI) + MOVOU -32(SI)(BX*1), X2 + MOVOU X2, -32(DI)(BX*1) + MOVOU -16(SI)(BX*1), X3 + MOVOU X3, -16(DI)(BX*1) + RET +move_65through128: + MOVOU (SI), X0 + MOVOU X0, (DI) + MOVOU 16(SI), X1 + MOVOU X1, 16(DI) + MOVOU 32(SI), X2 + MOVOU X2, 32(DI) + MOVOU 48(SI), X3 + MOVOU X3, 48(DI) + MOVOU -64(SI)(BX*1), X4 + MOVOU X4, -64(DI)(BX*1) + MOVOU -48(SI)(BX*1), X5 + MOVOU X5, -48(DI)(BX*1) + MOVOU -32(SI)(BX*1), X6 + MOVOU X6, -32(DI)(BX*1) + MOVOU -16(SI)(BX*1), X7 + MOVOU X7, -16(DI)(BX*1) + RET +move_129through256: + MOVOU (SI), X0 + MOVOU X0, (DI) + MOVOU 16(SI), X1 + MOVOU X1, 16(DI) + MOVOU 32(SI), X2 + MOVOU X2, 32(DI) + MOVOU 48(SI), X3 + MOVOU X3, 48(DI) + MOVOU 64(SI), X4 + MOVOU X4, 64(DI) + MOVOU 80(SI), X5 + MOVOU X5, 80(DI) + MOVOU 96(SI), X6 + MOVOU X6, 96(DI) + MOVOU 112(SI), X7 + MOVOU X7, 112(DI) + MOVOU -128(SI)(BX*1), X8 + MOVOU X8, -128(DI)(BX*1) + MOVOU -112(SI)(BX*1), X9 + MOVOU X9, -112(DI)(BX*1) + MOVOU -96(SI)(BX*1), X10 + MOVOU X10, -96(DI)(BX*1) + MOVOU -80(SI)(BX*1), X11 + MOVOU X11, -80(DI)(BX*1) + MOVOU -64(SI)(BX*1), X12 + MOVOU X12, -64(DI)(BX*1) + MOVOU -48(SI)(BX*1), X13 + MOVOU X13, -48(DI)(BX*1) + MOVOU -32(SI)(BX*1), X14 + MOVOU X14, -32(DI)(BX*1) + MOVOU -16(SI)(BX*1), X15 + MOVOU X15, -16(DI)(BX*1) + RET +move_256through2048: + SUBQ $256, BX + MOVOU (SI), X0 + MOVOU X0, (DI) + MOVOU 16(SI), X1 + MOVOU X1, 16(DI) + MOVOU 32(SI), X2 + MOVOU X2, 32(DI) + MOVOU 48(SI), X3 + MOVOU X3, 48(DI) + MOVOU 64(SI), X4 + MOVOU X4, 64(DI) + MOVOU 80(SI), X5 + MOVOU X5, 80(DI) + MOVOU 96(SI), X6 + MOVOU X6, 96(DI) + MOVOU 112(SI), X7 + MOVOU X7, 112(DI) + MOVOU 128(SI), X8 + MOVOU X8, 128(DI) + MOVOU 144(SI), X9 + MOVOU X9, 144(DI) + MOVOU 160(SI), X10 + MOVOU X10, 160(DI) + MOVOU 176(SI), X11 + MOVOU X11, 176(DI) + MOVOU 192(SI), X12 + MOVOU X12, 192(DI) + MOVOU 208(SI), X13 + MOVOU X13, 208(DI) + MOVOU 224(SI), X14 + MOVOU X14, 224(DI) + MOVOU 240(SI), X15 + MOVOU X15, 240(DI) + CMPQ BX, $256 + LEAQ 256(SI), SI + LEAQ 256(DI), DI + JGE move_256through2048 + JMP tail diff --git a/pkg/safecopy/memcpy_arm64.s b/pkg/safecopy/memcpy_arm64.s new file mode 100644 index 000000000..e7e541565 --- /dev/null +++ b/pkg/safecopy/memcpy_arm64.s @@ -0,0 +1,78 @@ +// Copyright 2014 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +#include "textflag.h" + +// handleMemcpyFault returns (the value stored in R0, the value stored in R1). +// Control is transferred to it when memcpy below receives SIGSEGV or SIGBUS, +// with the faulting address stored in R0 and the signal number stored in R1. +// +// It must have the same frame configuration as memcpy so that it can undo any +// potential call frame set up by the assembler. +TEXT handleMemcpyFault(SB), NOSPLIT, $0-36 + MOVD R0, addr+24(FP) + MOVW R1, sig+32(FP) + RET + +// memcpy copies data from src to dst. If a SIGSEGV or SIGBUS signal is received +// during the copy, it returns the address that caused the fault and the number +// of the signal that was received. Otherwise, it returns an unspecified address +// and a signal number of 0. +// +// Data is copied in order, such that if a fault happens at address p, it is +// safe to assume that all data before p-maxRegisterSize has already been +// successfully copied. +// +// The code is derived from the Go source runtime.memmove. +// +// func memcpy(dst, src unsafe.Pointer, n uintptr) (fault unsafe.Pointer, sig int32) +TEXT ·memcpy(SB), NOSPLIT, $-8-36 + // Store 0 as the returned signal number. If we run to completion, + // this is the value the caller will see; if a signal is received, + // handleMemcpyFault will store a different value in this address. + MOVW $0, sig+32(FP) + + MOVD to+0(FP), R3 + MOVD from+8(FP), R4 + MOVD n+16(FP), R5 + CMP $0, R5 + BNE check + RET + +check: + AND $~7, R5, R7 // R7 is N&~7. + SUB R7, R5, R6 // R6 is N&7. + + // Copying forward proceeds by copying R7/8 words then copying R6 bytes. + // R3 and R4 are advanced as we copy. + + // (There may be implementations of armv8 where copying by bytes until + // at least one of source or dest is word aligned is a worthwhile + // optimization, but the on the one tested so far (xgene) it did not + // make a significance difference.) + + CMP $0, R7 // Do we need to do any word-by-word copying? + BEQ noforwardlarge + ADD R3, R7, R9 // R9 points just past where we copy by word. + +forwardlargeloop: + MOVD.P 8(R4), R8 // R8 is just a scratch register. + MOVD.P R8, 8(R3) + CMP R3, R9 + BNE forwardlargeloop + +noforwardlarge: + CMP $0, R6 // Do we need to do any byte-by-byte copying? + BNE forwardtail + RET + +forwardtail: + ADD R3, R6, R9 // R9 points just past the destination memory. + +forwardtailloop: + MOVBU.P 1(R4), R8 + MOVBU.P R8, 1(R3) + CMP R3, R9 + BNE forwardtailloop + RET diff --git a/pkg/safecopy/safecopy.go b/pkg/safecopy/safecopy.go new file mode 100644 index 000000000..2fb7e5809 --- /dev/null +++ b/pkg/safecopy/safecopy.go @@ -0,0 +1,144 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package safecopy provides an efficient implementation of functions to access +// memory that may result in SIGSEGV or SIGBUS being sent to the accessor. +package safecopy + +import ( + "fmt" + "reflect" + "runtime" + "syscall" + + "gvisor.dev/gvisor/pkg/syserror" +) + +// SegvError is returned when a safecopy function receives SIGSEGV. +type SegvError struct { + // Addr is the address at which the SIGSEGV occurred. + Addr uintptr +} + +// Error implements error.Error. +func (e SegvError) Error() string { + return fmt.Sprintf("SIGSEGV at %#x", e.Addr) +} + +// BusError is returned when a safecopy function receives SIGBUS. +type BusError struct { + // Addr is the address at which the SIGBUS occurred. + Addr uintptr +} + +// Error implements error.Error. +func (e BusError) Error() string { + return fmt.Sprintf("SIGBUS at %#x", e.Addr) +} + +// AlignmentError is returned when a safecopy function is passed an address +// that does not meet alignment requirements. +type AlignmentError struct { + // Addr is the invalid address. + Addr uintptr + + // Alignment is the required alignment. + Alignment uintptr +} + +// Error implements error.Error. +func (e AlignmentError) Error() string { + return fmt.Sprintf("address %#x is not aligned to a %d-byte boundary", e.Addr, e.Alignment) +} + +var ( + // The begin and end addresses below are for the functions that are + // checked by the signal handler. + memcpyBegin uintptr + memcpyEnd uintptr + memclrBegin uintptr + memclrEnd uintptr + swapUint32Begin uintptr + swapUint32End uintptr + swapUint64Begin uintptr + swapUint64End uintptr + compareAndSwapUint32Begin uintptr + compareAndSwapUint32End uintptr + loadUint32Begin uintptr + loadUint32End uintptr + + // savedSigSegVHandler is a pointer to the SIGSEGV handler that was + // configured before we replaced it with our own. We still call into it + // when we get a SIGSEGV that is not interesting to us. + savedSigSegVHandler uintptr + + // same a above, but for SIGBUS signals. + savedSigBusHandler uintptr +) + +// signalHandler is our replacement signal handler for SIGSEGV and SIGBUS +// signals. +func signalHandler() + +// FindEndAddress returns the end address (one byte beyond the last) of the +// function that contains the specified address (begin). +func FindEndAddress(begin uintptr) uintptr { + f := runtime.FuncForPC(begin) + if f != nil { + for p := begin; ; p++ { + g := runtime.FuncForPC(p) + if f != g { + return p + } + } + } + return begin +} + +// initializeAddresses initializes the addresses used by the signal handler. +func initializeAddresses() { + // The following functions are written in assembly language, so they won't + // be inlined by the existing compiler/linker. Tests will fail if this + // assumption is violated. + memcpyBegin = reflect.ValueOf(memcpy).Pointer() + memcpyEnd = FindEndAddress(memcpyBegin) + memclrBegin = reflect.ValueOf(memclr).Pointer() + memclrEnd = FindEndAddress(memclrBegin) + swapUint32Begin = reflect.ValueOf(swapUint32).Pointer() + swapUint32End = FindEndAddress(swapUint32Begin) + swapUint64Begin = reflect.ValueOf(swapUint64).Pointer() + swapUint64End = FindEndAddress(swapUint64Begin) + compareAndSwapUint32Begin = reflect.ValueOf(compareAndSwapUint32).Pointer() + compareAndSwapUint32End = FindEndAddress(compareAndSwapUint32Begin) + loadUint32Begin = reflect.ValueOf(loadUint32).Pointer() + loadUint32End = FindEndAddress(loadUint32Begin) +} + +func init() { + initializeAddresses() + if err := ReplaceSignalHandler(syscall.SIGSEGV, reflect.ValueOf(signalHandler).Pointer(), &savedSigSegVHandler); err != nil { + panic(fmt.Sprintf("Unable to set handler for SIGSEGV: %v", err)) + } + if err := ReplaceSignalHandler(syscall.SIGBUS, reflect.ValueOf(signalHandler).Pointer(), &savedSigBusHandler); err != nil { + panic(fmt.Sprintf("Unable to set handler for SIGBUS: %v", err)) + } + syserror.AddErrorUnwrapper(func(e error) (syscall.Errno, bool) { + switch e.(type) { + case SegvError, BusError, AlignmentError: + return syscall.EFAULT, true + default: + return 0, false + } + }) +} diff --git a/pkg/safecopy/safecopy_test.go b/pkg/safecopy/safecopy_test.go new file mode 100644 index 000000000..5818f7f9b --- /dev/null +++ b/pkg/safecopy/safecopy_test.go @@ -0,0 +1,617 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package safecopy + +import ( + "bytes" + "fmt" + "io/ioutil" + "math/rand" + "os" + "runtime/debug" + "syscall" + "testing" + "unsafe" +) + +// Size of a page in bytes. Cloned from usermem.PageSize to avoid a circular +// dependency. +const pageSize = 4096 + +func initRandom(b []byte) { + for i := range b { + b[i] = byte(rand.Intn(256)) + } +} + +func randBuf(size int) []byte { + b := make([]byte, size) + initRandom(b) + return b +} + +func TestCopyInSuccess(t *testing.T) { + // Test that CopyIn does not return an error when all pages are accessible. + const bufLen = 8192 + a := randBuf(bufLen) + b := make([]byte, bufLen) + + n, err := CopyIn(b, unsafe.Pointer(&a[0])) + if n != bufLen { + t.Errorf("Unexpected copy length, got %v, want %v", n, bufLen) + } + if err != nil { + t.Errorf("Unexpected error: %v", err) + } + if !bytes.Equal(a, b) { + t.Errorf("Buffers are not equal when they should be: %v %v", a, b) + } +} + +func TestCopyOutSuccess(t *testing.T) { + // Test that CopyOut does not return an error when all pages are + // accessible. + const bufLen = 8192 + a := randBuf(bufLen) + b := make([]byte, bufLen) + + n, err := CopyOut(unsafe.Pointer(&b[0]), a) + if n != bufLen { + t.Errorf("Unexpected copy length, got %v, want %v", n, bufLen) + } + if err != nil { + t.Errorf("Unexpected error: %v", err) + } + if !bytes.Equal(a, b) { + t.Errorf("Buffers are not equal when they should be: %v %v", a, b) + } +} + +func TestCopySuccess(t *testing.T) { + // Test that Copy does not return an error when all pages are accessible. + const bufLen = 8192 + a := randBuf(bufLen) + b := make([]byte, bufLen) + + n, err := Copy(unsafe.Pointer(&b[0]), unsafe.Pointer(&a[0]), bufLen) + if n != bufLen { + t.Errorf("Unexpected copy length, got %v, want %v", n, bufLen) + } + if err != nil { + t.Errorf("Unexpected error: %v", err) + } + if !bytes.Equal(a, b) { + t.Errorf("Buffers are not equal when they should be: %v %v", a, b) + } +} + +func TestZeroOutSuccess(t *testing.T) { + // Test that ZeroOut does not return an error when all pages are + // accessible. + const bufLen = 8192 + a := make([]byte, bufLen) + b := randBuf(bufLen) + + n, err := ZeroOut(unsafe.Pointer(&b[0]), bufLen) + if n != bufLen { + t.Errorf("Unexpected copy length, got %v, want %v", n, bufLen) + } + if err != nil { + t.Errorf("Unexpected error: %v", err) + } + if !bytes.Equal(a, b) { + t.Errorf("Buffers are not equal when they should be: %v %v", a, b) + } +} + +func TestSwapUint32Success(t *testing.T) { + // Test that SwapUint32 does not return an error when the page is + // accessible. + before := uint32(rand.Int31()) + after := uint32(rand.Int31()) + val := before + + old, err := SwapUint32(unsafe.Pointer(&val), after) + if err != nil { + t.Errorf("Unexpected error: %v", err) + } + if old != before { + t.Errorf("Unexpected old value: got %v, want %v", old, before) + } + if val != after { + t.Errorf("Unexpected new value: got %v, want %v", val, after) + } +} + +func TestSwapUint32AlignmentError(t *testing.T) { + // Test that SwapUint32 returns an AlignmentError when passed an unaligned + // address. + data := new(struct{ val uint64 }) + addr := uintptr(unsafe.Pointer(&data.val)) + 1 + want := AlignmentError{Addr: addr, Alignment: 4} + if _, err := SwapUint32(unsafe.Pointer(addr), 1); err != want { + t.Errorf("Unexpected error: got %v, want %v", err, want) + } +} + +func TestSwapUint64Success(t *testing.T) { + // Test that SwapUint64 does not return an error when the page is + // accessible. + before := uint64(rand.Int63()) + after := uint64(rand.Int63()) + // "The first word in ... an allocated struct or slice can be relied upon + // to be 64-bit aligned." - sync/atomic docs + data := new(struct{ val uint64 }) + data.val = before + + old, err := SwapUint64(unsafe.Pointer(&data.val), after) + if err != nil { + t.Errorf("Unexpected error: %v", err) + } + if old != before { + t.Errorf("Unexpected old value: got %v, want %v", old, before) + } + if data.val != after { + t.Errorf("Unexpected new value: got %v, want %v", data.val, after) + } +} + +func TestSwapUint64AlignmentError(t *testing.T) { + // Test that SwapUint64 returns an AlignmentError when passed an unaligned + // address. + data := new(struct{ val1, val2 uint64 }) + addr := uintptr(unsafe.Pointer(&data.val1)) + 1 + want := AlignmentError{Addr: addr, Alignment: 8} + if _, err := SwapUint64(unsafe.Pointer(addr), 1); err != want { + t.Errorf("Unexpected error: got %v, want %v", err, want) + } +} + +func TestCompareAndSwapUint32Success(t *testing.T) { + // Test that CompareAndSwapUint32 does not return an error when the page is + // accessible. + before := uint32(rand.Int31()) + after := uint32(rand.Int31()) + val := before + + old, err := CompareAndSwapUint32(unsafe.Pointer(&val), before, after) + if err != nil { + t.Errorf("Unexpected error: %v", err) + } + if old != before { + t.Errorf("Unexpected old value: got %v, want %v", old, before) + } + if val != after { + t.Errorf("Unexpected new value: got %v, want %v", val, after) + } +} + +func TestCompareAndSwapUint32AlignmentError(t *testing.T) { + // Test that CompareAndSwapUint32 returns an AlignmentError when passed an + // unaligned address. + data := new(struct{ val uint64 }) + addr := uintptr(unsafe.Pointer(&data.val)) + 1 + want := AlignmentError{Addr: addr, Alignment: 4} + if _, err := CompareAndSwapUint32(unsafe.Pointer(addr), 0, 1); err != want { + t.Errorf("Unexpected error: got %v, want %v", err, want) + } +} + +// withSegvErrorTestMapping calls fn with a two-page mapping. The first page +// contains random data, and the second page generates SIGSEGV when accessed. +func withSegvErrorTestMapping(t *testing.T, fn func(m []byte)) { + mapping, err := syscall.Mmap(-1, 0, 2*pageSize, syscall.PROT_READ|syscall.PROT_WRITE, syscall.MAP_ANONYMOUS|syscall.MAP_PRIVATE) + if err != nil { + t.Fatalf("Mmap failed: %v", err) + } + defer syscall.Munmap(mapping) + if err := syscall.Mprotect(mapping[pageSize:], syscall.PROT_NONE); err != nil { + t.Fatalf("Mprotect failed: %v", err) + } + initRandom(mapping[:pageSize]) + + fn(mapping) +} + +// withBusErrorTestMapping calls fn with a two-page mapping. The first page +// contains random data, and the second page generates SIGBUS when accessed. +func withBusErrorTestMapping(t *testing.T, fn func(m []byte)) { + f, err := ioutil.TempFile("", "sigbus_test") + if err != nil { + t.Fatalf("TempFile failed: %v", err) + } + defer f.Close() + if err := f.Truncate(pageSize); err != nil { + t.Fatalf("Truncate failed: %v", err) + } + mapping, err := syscall.Mmap(int(f.Fd()), 0, 2*pageSize, syscall.PROT_READ|syscall.PROT_WRITE, syscall.MAP_SHARED) + if err != nil { + t.Fatalf("Mmap failed: %v", err) + } + defer syscall.Munmap(mapping) + initRandom(mapping[:pageSize]) + + fn(mapping) +} + +func TestCopyInSegvError(t *testing.T) { + // Test that CopyIn returns a SegvError when reaching a page that signals + // SIGSEGV. + for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { + t.Run(fmt.Sprintf("starting copy %d bytes before SIGSEGV", bytesBeforeFault), func(t *testing.T) { + withSegvErrorTestMapping(t, func(mapping []byte) { + secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize + src := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) + dst := randBuf(pageSize) + n, err := CopyIn(dst, src) + if n != bytesBeforeFault { + t.Errorf("Unexpected copy length: got %v, want %v", n, bytesBeforeFault) + } + if want := (SegvError{secondPage}); err != want { + t.Errorf("Unexpected error: got %v, want %v", err, want) + } + if got, want := dst[:bytesBeforeFault], mapping[pageSize-bytesBeforeFault:pageSize]; !bytes.Equal(got, want) { + t.Errorf("Buffers are not equal when they should be: %v %v", got, want) + } + }) + }) + } +} + +func TestCopyInBusError(t *testing.T) { + // Test that CopyIn returns a BusError when reaching a page that signals + // SIGBUS. + for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { + t.Run(fmt.Sprintf("starting copy %d bytes before SIGBUS", bytesBeforeFault), func(t *testing.T) { + withBusErrorTestMapping(t, func(mapping []byte) { + secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize + src := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) + dst := randBuf(pageSize) + n, err := CopyIn(dst, src) + if n != bytesBeforeFault { + t.Errorf("Unexpected copy length: got %v, want %v", n, bytesBeforeFault) + } + if want := (BusError{secondPage}); err != want { + t.Errorf("Unexpected error: got %v, want %v", err, want) + } + if got, want := dst[:bytesBeforeFault], mapping[pageSize-bytesBeforeFault:pageSize]; !bytes.Equal(got, want) { + t.Errorf("Buffers are not equal when they should be: %v %v", got, want) + } + }) + }) + } +} + +func TestCopyOutSegvError(t *testing.T) { + // Test that CopyOut returns a SegvError when reaching a page that signals + // SIGSEGV. + for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { + t.Run(fmt.Sprintf("starting copy %d bytes before SIGSEGV", bytesBeforeFault), func(t *testing.T) { + withSegvErrorTestMapping(t, func(mapping []byte) { + secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize + dst := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) + src := randBuf(pageSize) + n, err := CopyOut(dst, src) + if n != bytesBeforeFault { + t.Errorf("Unexpected copy length: got %v, want %v", n, bytesBeforeFault) + } + if want := (SegvError{secondPage}); err != want { + t.Errorf("Unexpected error: got %v, want %v", err, want) + } + if got, want := mapping[pageSize-bytesBeforeFault:pageSize], src[:bytesBeforeFault]; !bytes.Equal(got, want) { + t.Errorf("Buffers are not equal when they should be: %v %v", got, want) + } + }) + }) + } +} + +func TestCopyOutBusError(t *testing.T) { + // Test that CopyOut returns a BusError when reaching a page that signals + // SIGBUS. + for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { + t.Run(fmt.Sprintf("starting copy %d bytes before SIGSEGV", bytesBeforeFault), func(t *testing.T) { + withBusErrorTestMapping(t, func(mapping []byte) { + secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize + dst := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) + src := randBuf(pageSize) + n, err := CopyOut(dst, src) + if n != bytesBeforeFault { + t.Errorf("Unexpected copy length: got %v, want %v", n, bytesBeforeFault) + } + if want := (BusError{secondPage}); err != want { + t.Errorf("Unexpected error: got %v, want %v", err, want) + } + if got, want := mapping[pageSize-bytesBeforeFault:pageSize], src[:bytesBeforeFault]; !bytes.Equal(got, want) { + t.Errorf("Buffers are not equal when they should be: %v %v", got, want) + } + }) + }) + } +} + +func TestCopySourceSegvError(t *testing.T) { + // Test that Copy returns a SegvError when copying from a page that signals + // SIGSEGV. + for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { + t.Run(fmt.Sprintf("starting copy %d bytes before SIGSEGV", bytesBeforeFault), func(t *testing.T) { + withSegvErrorTestMapping(t, func(mapping []byte) { + secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize + src := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) + dst := randBuf(pageSize) + n, err := Copy(unsafe.Pointer(&dst[0]), src, pageSize) + if n != uintptr(bytesBeforeFault) { + t.Errorf("Unexpected copy length: got %v, want %v", n, bytesBeforeFault) + } + if want := (SegvError{secondPage}); err != want { + t.Errorf("Unexpected error: got %v, want %v", err, want) + } + if got, want := dst[:bytesBeforeFault], mapping[pageSize-bytesBeforeFault:pageSize]; !bytes.Equal(got, want) { + t.Errorf("Buffers are not equal when they should be: %v %v", got, want) + } + }) + }) + } +} + +func TestCopySourceBusError(t *testing.T) { + // Test that Copy returns a BusError when copying from a page that signals + // SIGBUS. + for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { + t.Run(fmt.Sprintf("starting copy %d bytes before SIGBUS", bytesBeforeFault), func(t *testing.T) { + withBusErrorTestMapping(t, func(mapping []byte) { + secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize + src := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) + dst := randBuf(pageSize) + n, err := Copy(unsafe.Pointer(&dst[0]), src, pageSize) + if n != uintptr(bytesBeforeFault) { + t.Errorf("Unexpected copy length: got %v, want %v", n, bytesBeforeFault) + } + if want := (BusError{secondPage}); err != want { + t.Errorf("Unexpected error: got %v, want %v", err, want) + } + if got, want := dst[:bytesBeforeFault], mapping[pageSize-bytesBeforeFault:pageSize]; !bytes.Equal(got, want) { + t.Errorf("Buffers are not equal when they should be: %v %v", got, want) + } + }) + }) + } +} + +func TestCopyDestinationSegvError(t *testing.T) { + // Test that Copy returns a SegvError when copying to a page that signals + // SIGSEGV. + for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { + t.Run(fmt.Sprintf("starting copy %d bytes before SIGSEGV", bytesBeforeFault), func(t *testing.T) { + withSegvErrorTestMapping(t, func(mapping []byte) { + secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize + dst := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) + src := randBuf(pageSize) + n, err := Copy(dst, unsafe.Pointer(&src[0]), pageSize) + if n != uintptr(bytesBeforeFault) { + t.Errorf("Unexpected copy length: got %v, want %v", n, bytesBeforeFault) + } + if want := (SegvError{secondPage}); err != want { + t.Errorf("Unexpected error: got %v, want %v", err, want) + } + if got, want := mapping[pageSize-bytesBeforeFault:pageSize], src[:bytesBeforeFault]; !bytes.Equal(got, want) { + t.Errorf("Buffers are not equal when they should be: %v %v", got, want) + } + }) + }) + } +} + +func TestCopyDestinationBusError(t *testing.T) { + // Test that Copy returns a BusError when copying to a page that signals + // SIGBUS. + for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { + t.Run(fmt.Sprintf("starting copy %d bytes before SIGBUS", bytesBeforeFault), func(t *testing.T) { + withBusErrorTestMapping(t, func(mapping []byte) { + secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize + dst := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) + src := randBuf(pageSize) + n, err := Copy(dst, unsafe.Pointer(&src[0]), pageSize) + if n != uintptr(bytesBeforeFault) { + t.Errorf("Unexpected copy length: got %v, want %v", n, bytesBeforeFault) + } + if want := (BusError{secondPage}); err != want { + t.Errorf("Unexpected error: got %v, want %v", err, want) + } + if got, want := mapping[pageSize-bytesBeforeFault:pageSize], src[:bytesBeforeFault]; !bytes.Equal(got, want) { + t.Errorf("Buffers are not equal when they should be: %v %v", got, want) + } + }) + }) + } +} + +func TestZeroOutSegvError(t *testing.T) { + // Test that ZeroOut returns a SegvError when reaching a page that signals + // SIGSEGV. + for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { + t.Run(fmt.Sprintf("starting write %d bytes before SIGSEGV", bytesBeforeFault), func(t *testing.T) { + withSegvErrorTestMapping(t, func(mapping []byte) { + secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize + dst := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) + n, err := ZeroOut(dst, pageSize) + if n != uintptr(bytesBeforeFault) { + t.Errorf("Unexpected write length: got %v, want %v", n, bytesBeforeFault) + } + if want := (SegvError{secondPage}); err != want { + t.Errorf("Unexpected error: got %v, want %v", err, want) + } + if got, want := mapping[pageSize-bytesBeforeFault:pageSize], make([]byte, bytesBeforeFault); !bytes.Equal(got, want) { + t.Errorf("Non-zero bytes in written part of mapping: %v", got) + } + }) + }) + } +} + +func TestZeroOutBusError(t *testing.T) { + // Test that ZeroOut returns a BusError when reaching a page that signals + // SIGBUS. + for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { + t.Run(fmt.Sprintf("starting write %d bytes before SIGBUS", bytesBeforeFault), func(t *testing.T) { + withBusErrorTestMapping(t, func(mapping []byte) { + secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize + dst := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) + n, err := ZeroOut(dst, pageSize) + if n != uintptr(bytesBeforeFault) { + t.Errorf("Unexpected write length: got %v, want %v", n, bytesBeforeFault) + } + if want := (BusError{secondPage}); err != want { + t.Errorf("Unexpected error: got %v, want %v", err, want) + } + if got, want := mapping[pageSize-bytesBeforeFault:pageSize], make([]byte, bytesBeforeFault); !bytes.Equal(got, want) { + t.Errorf("Non-zero bytes in written part of mapping: %v", got) + } + }) + }) + } +} + +func TestSwapUint32SegvError(t *testing.T) { + // Test that SwapUint32 returns a SegvError when reaching a page that + // signals SIGSEGV. + withSegvErrorTestMapping(t, func(mapping []byte) { + secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize + _, err := SwapUint32(unsafe.Pointer(secondPage), 1) + if want := (SegvError{secondPage}); err != want { + t.Errorf("Unexpected error: got %v, want %v", err, want) + } + }) +} + +func TestSwapUint32BusError(t *testing.T) { + // Test that SwapUint32 returns a BusError when reaching a page that + // signals SIGBUS. + withBusErrorTestMapping(t, func(mapping []byte) { + secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize + _, err := SwapUint32(unsafe.Pointer(secondPage), 1) + if want := (BusError{secondPage}); err != want { + t.Errorf("Unexpected error: got %v, want %v", err, want) + } + }) +} + +func TestSwapUint64SegvError(t *testing.T) { + // Test that SwapUint64 returns a SegvError when reaching a page that + // signals SIGSEGV. + withSegvErrorTestMapping(t, func(mapping []byte) { + secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize + _, err := SwapUint64(unsafe.Pointer(secondPage), 1) + if want := (SegvError{secondPage}); err != want { + t.Errorf("Unexpected error: got %v, want %v", err, want) + } + }) +} + +func TestSwapUint64BusError(t *testing.T) { + // Test that SwapUint64 returns a BusError when reaching a page that + // signals SIGBUS. + withBusErrorTestMapping(t, func(mapping []byte) { + secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize + _, err := SwapUint64(unsafe.Pointer(secondPage), 1) + if want := (BusError{secondPage}); err != want { + t.Errorf("Unexpected error: got %v, want %v", err, want) + } + }) +} + +func TestCompareAndSwapUint32SegvError(t *testing.T) { + // Test that CompareAndSwapUint32 returns a SegvError when reaching a page + // that signals SIGSEGV. + withSegvErrorTestMapping(t, func(mapping []byte) { + secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize + _, err := CompareAndSwapUint32(unsafe.Pointer(secondPage), 0, 1) + if want := (SegvError{secondPage}); err != want { + t.Errorf("Unexpected error: got %v, want %v", err, want) + } + }) +} + +func TestCompareAndSwapUint32BusError(t *testing.T) { + // Test that CompareAndSwapUint32 returns a BusError when reaching a page + // that signals SIGBUS. + withBusErrorTestMapping(t, func(mapping []byte) { + secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize + _, err := CompareAndSwapUint32(unsafe.Pointer(secondPage), 0, 1) + if want := (BusError{secondPage}); err != want { + t.Errorf("Unexpected error: got %v, want %v", err, want) + } + }) +} + +func testCopy(dst, src []byte) (panicked bool) { + defer func() { + if r := recover(); r != nil { + panicked = true + } + }() + debug.SetPanicOnFault(true) + copy(dst, src) + return +} + +func TestSegVOnMemmove(t *testing.T) { + // Test that SIGSEGVs received by runtime.memmove when *not* doing + // CopyIn or CopyOut work gets propagated to the runtime. + const bufLen = pageSize + a, err := syscall.Mmap(-1, 0, bufLen, syscall.PROT_NONE, syscall.MAP_ANON|syscall.MAP_PRIVATE) + if err != nil { + t.Fatalf("Mmap failed: %v", err) + + } + defer syscall.Munmap(a) + b := randBuf(bufLen) + + if !testCopy(b, a) { + t.Fatalf("testCopy didn't panic when it should have") + } + + if !testCopy(a, b) { + t.Fatalf("testCopy didn't panic when it should have") + } +} + +func TestSigbusOnMemmove(t *testing.T) { + // Test that SIGBUS received by runtime.memmove when *not* doing + // CopyIn or CopyOut work gets propagated to the runtime. + const bufLen = pageSize + f, err := ioutil.TempFile("", "sigbus_test") + if err != nil { + t.Fatalf("TempFile failed: %v", err) + } + os.Remove(f.Name()) + defer f.Close() + + a, err := syscall.Mmap(int(f.Fd()), 0, bufLen, syscall.PROT_READ|syscall.PROT_WRITE, syscall.MAP_SHARED) + if err != nil { + t.Fatalf("Mmap failed: %v", err) + + } + defer syscall.Munmap(a) + b := randBuf(bufLen) + + if !testCopy(b, a) { + t.Fatalf("testCopy didn't panic when it should have") + } + + if !testCopy(a, b) { + t.Fatalf("testCopy didn't panic when it should have") + } +} diff --git a/pkg/safecopy/safecopy_unsafe.go b/pkg/safecopy/safecopy_unsafe.go new file mode 100644 index 000000000..eef028e68 --- /dev/null +++ b/pkg/safecopy/safecopy_unsafe.go @@ -0,0 +1,335 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package safecopy + +import ( + "fmt" + "syscall" + "unsafe" +) + +// maxRegisterSize is the maximum register size used in memcpy and memclr. It +// is used to decide by how much to rewind the copy (for memcpy) or zeroing +// (for memclr) before proceeding. +const maxRegisterSize = 16 + +// memcpy copies data from src to dst. If a SIGSEGV or SIGBUS signal is received +// during the copy, it returns the address that caused the fault and the number +// of the signal that was received. Otherwise, it returns an unspecified address +// and a signal number of 0. +// +// Data is copied in order, such that if a fault happens at address p, it is +// safe to assume that all data before p-maxRegisterSize has already been +// successfully copied. +// +//go:noescape +func memcpy(dst, src unsafe.Pointer, n uintptr) (fault unsafe.Pointer, sig int32) + +// memclr sets the n bytes following ptr to zeroes. If a SIGSEGV or SIGBUS +// signal is received during the write, it returns the address that caused the +// fault and the number of the signal that was received. Otherwise, it returns +// an unspecified address and a signal number of 0. +// +// Data is written in order, such that if a fault happens at address p, it is +// safe to assume that all data before p-maxRegisterSize has already been +// successfully written. +// +//go:noescape +func memclr(ptr unsafe.Pointer, n uintptr) (fault unsafe.Pointer, sig int32) + +// swapUint32 atomically stores new into *ptr and returns (the previous *ptr +// value, 0). If a SIGSEGV or SIGBUS signal is received during the swap, the +// value of old is unspecified, and sig is the number of the signal that was +// received. +// +// Preconditions: ptr must be aligned to a 4-byte boundary. +// +//go:noescape +func swapUint32(ptr unsafe.Pointer, new uint32) (old uint32, sig int32) + +// swapUint64 atomically stores new into *ptr and returns (the previous *ptr +// value, 0). If a SIGSEGV or SIGBUS signal is received during the swap, the +// value of old is unspecified, and sig is the number of the signal that was +// received. +// +// Preconditions: ptr must be aligned to a 8-byte boundary. +// +//go:noescape +func swapUint64(ptr unsafe.Pointer, new uint64) (old uint64, sig int32) + +// compareAndSwapUint32 is like sync/atomic.CompareAndSwapUint32, but returns +// (the value previously stored at ptr, 0). If a SIGSEGV or SIGBUS signal is +// received during the operation, the value of prev is unspecified, and sig is +// the number of the signal that was received. +// +// Preconditions: ptr must be aligned to a 4-byte boundary. +// +//go:noescape +func compareAndSwapUint32(ptr unsafe.Pointer, old, new uint32) (prev uint32, sig int32) + +// LoadUint32 is like sync/atomic.LoadUint32, but operates with user memory. It +// may fail with SIGSEGV or SIGBUS if it is received while reading from ptr. +// +// Preconditions: ptr must be aligned to a 4-byte boundary. +// +//go:noescape +func loadUint32(ptr unsafe.Pointer) (val uint32, sig int32) + +// CopyIn copies len(dst) bytes from src to dst. It returns the number of bytes +// copied and an error if SIGSEGV or SIGBUS is received while reading from src. +func CopyIn(dst []byte, src unsafe.Pointer) (int, error) { + toCopy := uintptr(len(dst)) + if len(dst) == 0 { + return 0, nil + } + + fault, sig := memcpy(unsafe.Pointer(&dst[0]), src, toCopy) + if sig == 0 { + return len(dst), nil + } + + faultN, srcN := uintptr(fault), uintptr(src) + if faultN < srcN || faultN >= srcN+toCopy { + panic(fmt.Sprintf("CopyIn raised signal %d at %#x, which is outside source [%#x, %#x)", sig, faultN, srcN, srcN+toCopy)) + } + + // memcpy might have ended the copy up to maxRegisterSize bytes before + // fault, if an instruction caused a memory access that straddled two + // pages, and the second one faulted. Try to copy up to the fault. + var done int + if faultN-srcN > maxRegisterSize { + done = int(faultN - srcN - maxRegisterSize) + } + n, err := CopyIn(dst[done:int(faultN-srcN)], unsafe.Pointer(srcN+uintptr(done))) + done += n + if err != nil { + return done, err + } + return done, errorFromFaultSignal(fault, sig) +} + +// CopyOut copies len(src) bytes from src to dst. If returns the number of +// bytes done and an error if SIGSEGV or SIGBUS is received while writing to +// dst. +func CopyOut(dst unsafe.Pointer, src []byte) (int, error) { + toCopy := uintptr(len(src)) + if toCopy == 0 { + return 0, nil + } + + fault, sig := memcpy(dst, unsafe.Pointer(&src[0]), toCopy) + if sig == 0 { + return len(src), nil + } + + faultN, dstN := uintptr(fault), uintptr(dst) + if faultN < dstN || faultN >= dstN+toCopy { + panic(fmt.Sprintf("CopyOut raised signal %d at %#x, which is outside destination [%#x, %#x)", sig, faultN, dstN, dstN+toCopy)) + } + + // memcpy might have ended the copy up to maxRegisterSize bytes before + // fault, if an instruction caused a memory access that straddled two + // pages, and the second one faulted. Try to copy up to the fault. + var done int + if faultN-dstN > maxRegisterSize { + done = int(faultN - dstN - maxRegisterSize) + } + n, err := CopyOut(unsafe.Pointer(dstN+uintptr(done)), src[done:int(faultN-dstN)]) + done += n + if err != nil { + return done, err + } + return done, errorFromFaultSignal(fault, sig) +} + +// Copy copies toCopy bytes from src to dst. It returns the number of bytes +// copied and an error if SIGSEGV or SIGBUS is received while reading from src +// or writing to dst. +// +// Data is copied in order; if [src, src+toCopy) and [dst, dst+toCopy) overlap, +// the resulting contents of dst are unspecified. +func Copy(dst, src unsafe.Pointer, toCopy uintptr) (uintptr, error) { + if toCopy == 0 { + return 0, nil + } + + fault, sig := memcpy(dst, src, toCopy) + if sig == 0 { + return toCopy, nil + } + + // Did the fault occur while reading from src or writing to dst? + faultN, srcN, dstN := uintptr(fault), uintptr(src), uintptr(dst) + faultAfterSrc := ^uintptr(0) + if faultN >= srcN { + faultAfterSrc = faultN - srcN + } + faultAfterDst := ^uintptr(0) + if faultN >= dstN { + faultAfterDst = faultN - dstN + } + if faultAfterSrc >= toCopy && faultAfterDst >= toCopy { + panic(fmt.Sprintf("Copy raised signal %d at %#x, which is outside source [%#x, %#x) and destination [%#x, %#x)", sig, faultN, srcN, srcN+toCopy, dstN, dstN+toCopy)) + } + faultedAfter := faultAfterSrc + if faultedAfter > faultAfterDst { + faultedAfter = faultAfterDst + } + + // memcpy might have ended the copy up to maxRegisterSize bytes before + // fault, if an instruction caused a memory access that straddled two + // pages, and the second one faulted. Try to copy up to the fault. + var done uintptr + if faultedAfter > maxRegisterSize { + done = faultedAfter - maxRegisterSize + } + n, err := Copy(unsafe.Pointer(dstN+done), unsafe.Pointer(srcN+done), faultedAfter-done) + done += n + if err != nil { + return done, err + } + return done, errorFromFaultSignal(fault, sig) +} + +// ZeroOut writes toZero zero bytes to dst. It returns the number of bytes +// written and an error if SIGSEGV or SIGBUS is received while writing to dst. +func ZeroOut(dst unsafe.Pointer, toZero uintptr) (uintptr, error) { + if toZero == 0 { + return 0, nil + } + + fault, sig := memclr(dst, toZero) + if sig == 0 { + return toZero, nil + } + + faultN, dstN := uintptr(fault), uintptr(dst) + if faultN < dstN || faultN >= dstN+toZero { + panic(fmt.Sprintf("ZeroOut raised signal %d at %#x, which is outside destination [%#x, %#x)", sig, faultN, dstN, dstN+toZero)) + } + + // memclr might have ended the write up to maxRegisterSize bytes before + // fault, if an instruction caused a memory access that straddled two + // pages, and the second one faulted. Try to write up to the fault. + var done uintptr + if faultN-dstN > maxRegisterSize { + done = faultN - dstN - maxRegisterSize + } + n, err := ZeroOut(unsafe.Pointer(dstN+done), faultN-dstN-done) + done += n + if err != nil { + return done, err + } + return done, errorFromFaultSignal(fault, sig) +} + +// SwapUint32 is equivalent to sync/atomic.SwapUint32, except that it returns +// an error if SIGSEGV or SIGBUS is received while accessing ptr, or if ptr is +// not aligned to a 4-byte boundary. +func SwapUint32(ptr unsafe.Pointer, new uint32) (uint32, error) { + if addr := uintptr(ptr); addr&3 != 0 { + return 0, AlignmentError{addr, 4} + } + old, sig := swapUint32(ptr, new) + return old, errorFromFaultSignal(ptr, sig) +} + +// SwapUint64 is equivalent to sync/atomic.SwapUint64, except that it returns +// an error if SIGSEGV or SIGBUS is received while accessing ptr, or if ptr is +// not aligned to an 8-byte boundary. +func SwapUint64(ptr unsafe.Pointer, new uint64) (uint64, error) { + if addr := uintptr(ptr); addr&7 != 0 { + return 0, AlignmentError{addr, 8} + } + old, sig := swapUint64(ptr, new) + return old, errorFromFaultSignal(ptr, sig) +} + +// CompareAndSwapUint32 is equivalent to atomicbitops.CompareAndSwapUint32, +// except that it returns an error if SIGSEGV or SIGBUS is received while +// accessing ptr, or if ptr is not aligned to a 4-byte boundary. +func CompareAndSwapUint32(ptr unsafe.Pointer, old, new uint32) (uint32, error) { + if addr := uintptr(ptr); addr&3 != 0 { + return 0, AlignmentError{addr, 4} + } + prev, sig := compareAndSwapUint32(ptr, old, new) + return prev, errorFromFaultSignal(ptr, sig) +} + +// LoadUint32 is like sync/atomic.LoadUint32, but operates with user memory. It +// may fail with SIGSEGV or SIGBUS if it is received while reading from ptr. +// +// Preconditions: ptr must be aligned to a 4-byte boundary. +func LoadUint32(ptr unsafe.Pointer) (uint32, error) { + if addr := uintptr(ptr); addr&3 != 0 { + return 0, AlignmentError{addr, 4} + } + val, sig := loadUint32(ptr) + return val, errorFromFaultSignal(ptr, sig) +} + +func errorFromFaultSignal(addr unsafe.Pointer, sig int32) error { + switch sig { + case 0: + return nil + case int32(syscall.SIGSEGV): + return SegvError{uintptr(addr)} + case int32(syscall.SIGBUS): + return BusError{uintptr(addr)} + default: + panic(fmt.Sprintf("safecopy got unexpected signal %d at address %#x", sig, addr)) + } +} + +// ReplaceSignalHandler replaces the existing signal handler for the provided +// signal with the one that handles faults in safecopy-protected functions. +// +// It stores the value of the previously set handler in previous. +// +// This function will be called on initialization in order to install safecopy +// handlers for appropriate signals. These handlers will call the previous +// handler however, and if this is function is being used externally then the +// same courtesy is expected. +func ReplaceSignalHandler(sig syscall.Signal, handler uintptr, previous *uintptr) error { + var sa struct { + handler uintptr + flags uint64 + restorer uintptr + mask uint64 + } + const maskLen = 8 + + // Get the existing signal handler information, and save the current + // handler. Once we replace it, we will use this pointer to fall back to + // it when we receive other signals. + if _, _, e := syscall.RawSyscall6(syscall.SYS_RT_SIGACTION, uintptr(sig), 0, uintptr(unsafe.Pointer(&sa)), maskLen, 0, 0); e != 0 { + return e + } + + // Fail if there isn't a previous handler. + if sa.handler == 0 { + return fmt.Errorf("previous handler for signal %x isn't set", sig) + } + + *previous = sa.handler + + // Install our own handler. + sa.handler = handler + if _, _, e := syscall.RawSyscall6(syscall.SYS_RT_SIGACTION, uintptr(sig), uintptr(unsafe.Pointer(&sa)), 0, maskLen, 0, 0); e != 0 { + return e + } + + return nil +} diff --git a/pkg/safecopy/sighandler_amd64.s b/pkg/safecopy/sighandler_amd64.s new file mode 100644 index 000000000..475ae48e9 --- /dev/null +++ b/pkg/safecopy/sighandler_amd64.s @@ -0,0 +1,133 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "textflag.h" + +// The signals handled by sigHandler. +#define SIGBUS 7 +#define SIGSEGV 11 + +// Offsets to the registers in context->uc_mcontext.gregs[]. +#define REG_RDI 0x68 +#define REG_RAX 0x90 +#define REG_IP 0xa8 + +// Offset to the si_addr field of siginfo. +#define SI_CODE 0x08 +#define SI_ADDR 0x10 + +// signalHandler is the signal handler for SIGSEGV and SIGBUS signals. It must +// not be set up as a handler to any other signals. +// +// If the instruction causing the signal is within a safecopy-protected +// function, the signal is handled such that execution resumes in the +// appropriate fault handling stub with AX containing the faulting address and +// DI containing the signal number. Otherwise control is transferred to the +// previously configured signal handler (savedSigSegvHandler or +// savedSigBusHandler). +// +// This function cannot be written in go because it runs whenever a signal is +// received by the thread (preempting whatever was running), which includes when +// garbage collector has stopped or isn't expecting any interactions (like +// barriers). +// +// The arguments are the following: +// DI - The signal number. +// SI - Pointer to siginfo_t structure. +// DX - Pointer to ucontext structure. +TEXT ·signalHandler(SB),NOSPLIT,$0 + // Check if the signal is from the kernel. + MOVQ $0x0, CX + CMPL CX, SI_CODE(SI) + JGE original_handler + + // Check if RIP is within the area we care about. + MOVQ REG_IP(DX), CX + CMPQ CX, ·memcpyBegin(SB) + JB not_memcpy + CMPQ CX, ·memcpyEnd(SB) + JAE not_memcpy + + // Modify the context such that execution will resume in the fault + // handler. + LEAQ handleMemcpyFault(SB), CX + JMP handle_fault + +not_memcpy: + CMPQ CX, ·memclrBegin(SB) + JB not_memclr + CMPQ CX, ·memclrEnd(SB) + JAE not_memclr + + LEAQ handleMemclrFault(SB), CX + JMP handle_fault + +not_memclr: + CMPQ CX, ·swapUint32Begin(SB) + JB not_swapuint32 + CMPQ CX, ·swapUint32End(SB) + JAE not_swapuint32 + + LEAQ handleSwapUint32Fault(SB), CX + JMP handle_fault + +not_swapuint32: + CMPQ CX, ·swapUint64Begin(SB) + JB not_swapuint64 + CMPQ CX, ·swapUint64End(SB) + JAE not_swapuint64 + + LEAQ handleSwapUint64Fault(SB), CX + JMP handle_fault + +not_swapuint64: + CMPQ CX, ·compareAndSwapUint32Begin(SB) + JB not_casuint32 + CMPQ CX, ·compareAndSwapUint32End(SB) + JAE not_casuint32 + + LEAQ handleCompareAndSwapUint32Fault(SB), CX + JMP handle_fault + +not_casuint32: + CMPQ CX, ·loadUint32Begin(SB) + JB not_loaduint32 + CMPQ CX, ·loadUint32End(SB) + JAE not_loaduint32 + + LEAQ handleLoadUint32Fault(SB), CX + JMP handle_fault + +not_loaduint32: +original_handler: + // Jump to the previous signal handler, which is likely the golang one. + XORQ CX, CX + MOVQ ·savedSigBusHandler(SB), AX + CMPL DI, $SIGSEGV + CMOVQEQ ·savedSigSegVHandler(SB), AX + JMP AX + +handle_fault: + // Entered with the address of the fault handler in RCX; store it in + // RIP. + MOVQ CX, REG_IP(DX) + + // Store the faulting address in RAX. + MOVQ SI_ADDR(SI), CX + MOVQ CX, REG_RAX(DX) + + // Store the signal number in EDI. + MOVL DI, REG_RDI(DX) + + RET diff --git a/pkg/safecopy/sighandler_arm64.s b/pkg/safecopy/sighandler_arm64.s new file mode 100644 index 000000000..53e4ac2c1 --- /dev/null +++ b/pkg/safecopy/sighandler_arm64.s @@ -0,0 +1,143 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "textflag.h" + +// The signals handled by sigHandler. +#define SIGBUS 7 +#define SIGSEGV 11 + +// Offsets to the registers in context->uc_mcontext.gregs[]. +#define REG_R0 0xB8 +#define REG_R1 0xC0 +#define REG_PC 0x1B8 + +// Offset to the si_addr field of siginfo. +#define SI_CODE 0x08 +#define SI_ADDR 0x10 + +// signalHandler is the signal handler for SIGSEGV and SIGBUS signals. It must +// not be set up as a handler to any other signals. +// +// If the instruction causing the signal is within a safecopy-protected +// function, the signal is handled such that execution resumes in the +// appropriate fault handling stub with R0 containing the faulting address and +// R1 containing the signal number. Otherwise control is transferred to the +// previously configured signal handler (savedSigSegvHandler or +// savedSigBusHandler). +// +// This function cannot be written in go because it runs whenever a signal is +// received by the thread (preempting whatever was running), which includes when +// garbage collector has stopped or isn't expecting any interactions (like +// barriers). +// +// The arguments are the following: +// R0 - The signal number. +// R1 - Pointer to siginfo_t structure. +// R2 - Pointer to ucontext structure. +TEXT ·signalHandler(SB),NOSPLIT,$0 + // Check if the signal is from the kernel, si_code > 0 means a kernel signal. + MOVD SI_CODE(R1), R7 + CMPW $0x0, R7 + BLE original_handler + + // Check if PC is within the area we care about. + MOVD REG_PC(R2), R7 + MOVD ·memcpyBegin(SB), R8 + CMP R8, R7 + BLO not_memcpy + MOVD ·memcpyEnd(SB), R8 + CMP R8, R7 + BHS not_memcpy + + // Modify the context such that execution will resume in the fault handler. + MOVD $handleMemcpyFault(SB), R7 + B handle_fault + +not_memcpy: + MOVD ·memclrBegin(SB), R8 + CMP R8, R7 + BLO not_memclr + MOVD ·memclrEnd(SB), R8 + CMP R8, R7 + BHS not_memclr + + MOVD $handleMemclrFault(SB), R7 + B handle_fault + +not_memclr: + MOVD ·swapUint32Begin(SB), R8 + CMP R8, R7 + BLO not_swapuint32 + MOVD ·swapUint32End(SB), R8 + CMP R8, R7 + BHS not_swapuint32 + + MOVD $handleSwapUint32Fault(SB), R7 + B handle_fault + +not_swapuint32: + MOVD ·swapUint64Begin(SB), R8 + CMP R8, R7 + BLO not_swapuint64 + MOVD ·swapUint64End(SB), R8 + CMP R8, R7 + BHS not_swapuint64 + + MOVD $handleSwapUint64Fault(SB), R7 + B handle_fault + +not_swapuint64: + MOVD ·compareAndSwapUint32Begin(SB), R8 + CMP R8, R7 + BLO not_casuint32 + MOVD ·compareAndSwapUint32End(SB), R8 + CMP R8, R7 + BHS not_casuint32 + + MOVD $handleCompareAndSwapUint32Fault(SB), R7 + B handle_fault + +not_casuint32: + MOVD ·loadUint32Begin(SB), R8 + CMP R8, R7 + BLO not_loaduint32 + MOVD ·loadUint32End(SB), R8 + CMP R8, R7 + BHS not_loaduint32 + + MOVD $handleLoadUint32Fault(SB), R7 + B handle_fault + +not_loaduint32: +original_handler: + // Jump to the previous signal handler, which is likely the golang one. + MOVD ·savedSigBusHandler(SB), R7 + MOVD ·savedSigSegVHandler(SB), R8 + CMPW $SIGSEGV, R0 + CSEL EQ, R8, R7, R7 + B (R7) + +handle_fault: + // Entered with the address of the fault handler in R7; store it in PC. + MOVD R7, REG_PC(R2) + + // Store the faulting address in R0. + MOVD SI_ADDR(R1), R7 + MOVD R7, REG_R0(R2) + + // Store the signal number in R1. + MOVW R0, REG_R1(R2) + + RET diff --git a/pkg/safemem/BUILD b/pkg/safemem/BUILD new file mode 100644 index 000000000..ce30382ab --- /dev/null +++ b/pkg/safemem/BUILD @@ -0,0 +1,27 @@ +load("//tools:defs.bzl", "go_library", "go_test") + +package(licenses = ["notice"]) + +go_library( + name = "safemem", + srcs = [ + "block_unsafe.go", + "io.go", + "safemem.go", + "seq_unsafe.go", + ], + visibility = ["//:sandbox"], + deps = [ + "//pkg/safecopy", + ], +) + +go_test( + name = "safemem_test", + size = "small", + srcs = [ + "io_test.go", + "seq_test.go", + ], + library = ":safemem", +) diff --git a/pkg/safemem/block_unsafe.go b/pkg/safemem/block_unsafe.go new file mode 100644 index 000000000..e7fd30743 --- /dev/null +++ b/pkg/safemem/block_unsafe.go @@ -0,0 +1,279 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package safemem + +import ( + "fmt" + "reflect" + "unsafe" + + "gvisor.dev/gvisor/pkg/safecopy" +) + +// A Block is a range of contiguous bytes, similar to []byte but with the +// following differences: +// +// - The memory represented by a Block may require the use of safecopy to +// access. +// +// - Block does not carry a capacity and cannot be expanded. +// +// Blocks are immutable and may be copied by value. The zero value of Block +// represents an empty range, analogous to a nil []byte. +type Block struct { + // [start, start+length) is the represented memory. + // + // start is an unsafe.Pointer to ensure that Block prevents the represented + // memory from being garbage-collected. + start unsafe.Pointer + length int + + // needSafecopy is true if accessing the represented memory requires the + // use of safecopy. + needSafecopy bool +} + +// BlockFromSafeSlice returns a Block equivalent to slice, which is safe to +// access without safecopy. +func BlockFromSafeSlice(slice []byte) Block { + return blockFromSlice(slice, false) +} + +// BlockFromUnsafeSlice returns a Block equivalent to bs, which is not safe to +// access without safecopy. +func BlockFromUnsafeSlice(slice []byte) Block { + return blockFromSlice(slice, true) +} + +func blockFromSlice(slice []byte, needSafecopy bool) Block { + if len(slice) == 0 { + return Block{} + } + return Block{ + start: unsafe.Pointer(&slice[0]), + length: len(slice), + needSafecopy: needSafecopy, + } +} + +// BlockFromSafePointer returns a Block equivalent to [ptr, ptr+len), which is +// safe to access without safecopy. +// +// Preconditions: ptr+len does not overflow. +func BlockFromSafePointer(ptr unsafe.Pointer, len int) Block { + return blockFromPointer(ptr, len, false) +} + +// BlockFromUnsafePointer returns a Block equivalent to [ptr, ptr+len), which +// is not safe to access without safecopy. +// +// Preconditions: ptr+len does not overflow. +func BlockFromUnsafePointer(ptr unsafe.Pointer, len int) Block { + return blockFromPointer(ptr, len, true) +} + +func blockFromPointer(ptr unsafe.Pointer, len int, needSafecopy bool) Block { + if uptr := uintptr(ptr); uptr+uintptr(len) < uptr { + panic(fmt.Sprintf("ptr %#x + len %#x overflows", ptr, len)) + } + return Block{ + start: ptr, + length: len, + needSafecopy: needSafecopy, + } +} + +// DropFirst returns a Block equivalent to b, but with the first n bytes +// omitted. It is analogous to the [n:] operation on a slice, except that if n +// > b.Len(), DropFirst returns an empty Block instead of panicking. +// +// Preconditions: n >= 0. +func (b Block) DropFirst(n int) Block { + if n < 0 { + panic(fmt.Sprintf("invalid n: %d", n)) + } + return b.DropFirst64(uint64(n)) +} + +// DropFirst64 is equivalent to DropFirst but takes a uint64. +func (b Block) DropFirst64(n uint64) Block { + if n >= uint64(b.length) { + return Block{} + } + return Block{ + start: unsafe.Pointer(uintptr(b.start) + uintptr(n)), + length: b.length - int(n), + needSafecopy: b.needSafecopy, + } +} + +// TakeFirst returns a Block equivalent to the first n bytes of b. It is +// analogous to the [:n] operation on a slice, except that if n > b.Len(), +// TakeFirst returns a copy of b instead of panicking. +// +// Preconditions: n >= 0. +func (b Block) TakeFirst(n int) Block { + if n < 0 { + panic(fmt.Sprintf("invalid n: %d", n)) + } + return b.TakeFirst64(uint64(n)) +} + +// TakeFirst64 is equivalent to TakeFirst but takes a uint64. +func (b Block) TakeFirst64(n uint64) Block { + if n == 0 { + return Block{} + } + if n >= uint64(b.length) { + return b + } + return Block{ + start: b.start, + length: int(n), + needSafecopy: b.needSafecopy, + } +} + +// ToSlice returns a []byte equivalent to b. +func (b Block) ToSlice() []byte { + var bs []byte + hdr := (*reflect.SliceHeader)(unsafe.Pointer(&bs)) + hdr.Data = uintptr(b.start) + hdr.Len = b.length + hdr.Cap = b.length + return bs +} + +// Addr returns b's start address as a uintptr. It returns uintptr instead of +// unsafe.Pointer so that code using safemem cannot obtain unsafe.Pointers +// without importing the unsafe package explicitly. +// +// Note that a uintptr is not recognized as a pointer by the garbage collector, +// such that if there are no uses of b after a call to b.Addr() and the address +// is to Go-managed memory, the returned uintptr does not prevent garbage +// collection of the pointee. +func (b Block) Addr() uintptr { + return uintptr(b.start) +} + +// Len returns b's length in bytes. +func (b Block) Len() int { + return b.length +} + +// NeedSafecopy returns true if accessing b.ToSlice() requires the use of safecopy. +func (b Block) NeedSafecopy() bool { + return b.needSafecopy +} + +// String implements fmt.Stringer.String. +func (b Block) String() string { + if uintptr(b.start) == 0 && b.length == 0 { + return "" + } + var suffix string + if b.needSafecopy { + suffix = "*" + } + return fmt.Sprintf("[%#x-%#x)%s", uintptr(b.start), uintptr(b.start)+uintptr(b.length), suffix) +} + +// Copy copies src.Len() or dst.Len() bytes, whichever is less, from src +// to dst and returns the number of bytes copied. +// +// If src and dst overlap, the data stored in dst is unspecified. +func Copy(dst, src Block) (int, error) { + if !dst.needSafecopy && !src.needSafecopy { + return copy(dst.ToSlice(), src.ToSlice()), nil + } + + n := dst.length + if n > src.length { + n = src.length + } + if n == 0 { + return 0, nil + } + + switch { + case dst.needSafecopy && !src.needSafecopy: + return safecopy.CopyOut(dst.start, src.TakeFirst(n).ToSlice()) + case !dst.needSafecopy && src.needSafecopy: + return safecopy.CopyIn(dst.TakeFirst(n).ToSlice(), src.start) + case dst.needSafecopy && src.needSafecopy: + n64, err := safecopy.Copy(dst.start, src.start, uintptr(n)) + return int(n64), err + default: + panic("unreachable") + } +} + +// Zero sets all bytes in dst to 0 and returns the number of bytes zeroed. +func Zero(dst Block) (int, error) { + if !dst.needSafecopy { + bs := dst.ToSlice() + for i := range bs { + bs[i] = 0 + } + return len(bs), nil + } + + n64, err := safecopy.ZeroOut(dst.start, uintptr(dst.length)) + return int(n64), err +} + +// Safecopy atomics are no slower than non-safecopy atomics, so use the former +// even when !b.needSafecopy to get consistent alignment checking. + +// SwapUint32 invokes safecopy.SwapUint32 on the first 4 bytes of b. +// +// Preconditions: b.Len() >= 4. +func SwapUint32(b Block, new uint32) (uint32, error) { + if b.length < 4 { + panic(fmt.Sprintf("insufficient length: %d", b.length)) + } + return safecopy.SwapUint32(b.start, new) +} + +// SwapUint64 invokes safecopy.SwapUint64 on the first 8 bytes of b. +// +// Preconditions: b.Len() >= 8. +func SwapUint64(b Block, new uint64) (uint64, error) { + if b.length < 8 { + panic(fmt.Sprintf("insufficient length: %d", b.length)) + } + return safecopy.SwapUint64(b.start, new) +} + +// CompareAndSwapUint32 invokes safecopy.CompareAndSwapUint32 on the first 4 +// bytes of b. +// +// Preconditions: b.Len() >= 4. +func CompareAndSwapUint32(b Block, old, new uint32) (uint32, error) { + if b.length < 4 { + panic(fmt.Sprintf("insufficient length: %d", b.length)) + } + return safecopy.CompareAndSwapUint32(b.start, old, new) +} + +// LoadUint32 invokes safecopy.LoadUint32 on the first 4 bytes of b. +// +// Preconditions: b.Len() >= 4. +func LoadUint32(b Block) (uint32, error) { + if b.length < 4 { + panic(fmt.Sprintf("insufficient length: %d", b.length)) + } + return safecopy.LoadUint32(b.start) +} diff --git a/pkg/safemem/io.go b/pkg/safemem/io.go new file mode 100644 index 000000000..f039a5c34 --- /dev/null +++ b/pkg/safemem/io.go @@ -0,0 +1,392 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package safemem + +import ( + "errors" + "io" + "math" +) + +// ErrEndOfBlockSeq is returned by BlockSeqWriter when attempting to write +// beyond the end of the BlockSeq. +var ErrEndOfBlockSeq = errors.New("write beyond end of BlockSeq") + +// Reader represents a streaming byte source like io.Reader. +type Reader interface { + // ReadToBlocks reads up to dsts.NumBytes() bytes into dsts and returns the + // number of bytes read. It may return a partial read without an error + // (i.e. (n, nil) where 0 < n < dsts.NumBytes()). It should not return a + // full read with an error (i.e. (dsts.NumBytes(), err) where err != nil); + // note that this differs from io.Reader.Read (in particular, io.EOF should + // not be returned if ReadToBlocks successfully reads dsts.NumBytes() + // bytes.) + ReadToBlocks(dsts BlockSeq) (uint64, error) +} + +// Writer represents a streaming byte sink like io.Writer. +type Writer interface { + // WriteFromBlocks writes up to srcs.NumBytes() bytes from srcs and returns + // the number of bytes written. It may return a partial write without an + // error (i.e. (n, nil) where 0 < n < srcs.NumBytes()). It should not + // return a full write with an error (i.e. srcs.NumBytes(), err) where err + // != nil). + WriteFromBlocks(srcs BlockSeq) (uint64, error) +} + +// ReadFullToBlocks repeatedly invokes r.ReadToBlocks until dsts.NumBytes() +// bytes have been read or ReadToBlocks returns an error. +func ReadFullToBlocks(r Reader, dsts BlockSeq) (uint64, error) { + var done uint64 + for !dsts.IsEmpty() { + n, err := r.ReadToBlocks(dsts) + done += n + if err != nil { + return done, err + } + dsts = dsts.DropFirst64(n) + } + return done, nil +} + +// WriteFullFromBlocks repeatedly invokes w.WriteFromBlocks until +// srcs.NumBytes() bytes have been written or WriteFromBlocks returns an error. +func WriteFullFromBlocks(w Writer, srcs BlockSeq) (uint64, error) { + var done uint64 + for !srcs.IsEmpty() { + n, err := w.WriteFromBlocks(srcs) + done += n + if err != nil { + return done, err + } + srcs = srcs.DropFirst64(n) + } + return done, nil +} + +// BlockSeqReader implements Reader by reading from a BlockSeq. +type BlockSeqReader struct { + Blocks BlockSeq +} + +// ReadToBlocks implements Reader.ReadToBlocks. +func (r *BlockSeqReader) ReadToBlocks(dsts BlockSeq) (uint64, error) { + n, err := CopySeq(dsts, r.Blocks) + r.Blocks = r.Blocks.DropFirst64(n) + if err != nil { + return n, err + } + if n < dsts.NumBytes() { + return n, io.EOF + } + return n, nil +} + +// BlockSeqWriter implements Writer by writing to a BlockSeq. +type BlockSeqWriter struct { + Blocks BlockSeq +} + +// WriteFromBlocks implements Writer.WriteFromBlocks. +func (w *BlockSeqWriter) WriteFromBlocks(srcs BlockSeq) (uint64, error) { + n, err := CopySeq(w.Blocks, srcs) + w.Blocks = w.Blocks.DropFirst64(n) + if err != nil { + return n, err + } + if n < srcs.NumBytes() { + return n, ErrEndOfBlockSeq + } + return n, nil +} + +// ReaderFunc implements Reader for a function with the semantics of +// Reader.ReadToBlocks. +type ReaderFunc func(dsts BlockSeq) (uint64, error) + +// ReadToBlocks implements Reader.ReadToBlocks. +func (f ReaderFunc) ReadToBlocks(dsts BlockSeq) (uint64, error) { + return f(dsts) +} + +// WriterFunc implements Writer for a function with the semantics of +// Writer.WriteFromBlocks. +type WriterFunc func(srcs BlockSeq) (uint64, error) + +// WriteFromBlocks implements Writer.WriteFromBlocks. +func (f WriterFunc) WriteFromBlocks(srcs BlockSeq) (uint64, error) { + return f(srcs) +} + +// ToIOReader implements io.Reader for a (safemem.)Reader. +// +// ToIOReader will return a successful partial read iff Reader.ReadToBlocks does +// so. +type ToIOReader struct { + Reader Reader +} + +// Read implements io.Reader.Read. +func (r ToIOReader) Read(dst []byte) (int, error) { + n, err := r.Reader.ReadToBlocks(BlockSeqOf(BlockFromSafeSlice(dst))) + return int(n), err +} + +// ToIOWriter implements io.Writer for a (safemem.)Writer. +type ToIOWriter struct { + Writer Writer +} + +// Write implements io.Writer.Write. +func (w ToIOWriter) Write(src []byte) (int, error) { + // io.Writer does not permit partial writes. + n, err := WriteFullFromBlocks(w.Writer, BlockSeqOf(BlockFromSafeSlice(src))) + return int(n), err +} + +// FromIOReader implements Reader for an io.Reader by repeatedly invoking +// io.Reader.Read until it returns an error or partial read. This is not +// thread-safe. +// +// FromIOReader will return a successful partial read iff Reader.Read does so. +type FromIOReader struct { + Reader io.Reader +} + +// ReadToBlocks implements Reader.ReadToBlocks. +func (r FromIOReader) ReadToBlocks(dsts BlockSeq) (uint64, error) { + var buf []byte + var done uint64 + for !dsts.IsEmpty() { + dst := dsts.Head() + var n int + var err error + n, buf, err = r.readToBlock(dst, buf) + done += uint64(n) + if n != dst.Len() { + return done, err + } + dsts = dsts.Tail() + if err != nil { + if dsts.IsEmpty() && err == io.EOF { + return done, nil + } + return done, err + } + } + return done, nil +} + +func (r FromIOReader) readToBlock(dst Block, buf []byte) (int, []byte, error) { + // io.Reader isn't safecopy-aware, so we have to buffer Blocks that require + // safecopy. + if !dst.NeedSafecopy() { + n, err := r.Reader.Read(dst.ToSlice()) + return n, buf, err + } + if len(buf) < dst.Len() { + buf = make([]byte, dst.Len()) + } + rn, rerr := r.Reader.Read(buf[:dst.Len()]) + wbn, wberr := Copy(dst, BlockFromSafeSlice(buf[:rn])) + if wberr != nil { + return wbn, buf, wberr + } + return wbn, buf, rerr +} + +// FromIOReaderAt implements Reader for an io.ReaderAt. Does not repeatedly +// invoke io.ReaderAt.ReadAt because ReadAt is more strict than Read. A partial +// read indicates an error. This is not thread-safe. +type FromIOReaderAt struct { + ReaderAt io.ReaderAt + Offset int64 +} + +// ReadToBlocks implements Reader.ReadToBlocks. +func (r FromIOReaderAt) ReadToBlocks(dsts BlockSeq) (uint64, error) { + var buf []byte + var done uint64 + for !dsts.IsEmpty() { + dst := dsts.Head() + var n int + var err error + n, buf, err = r.readToBlock(dst, buf) + done += uint64(n) + if n != dst.Len() { + return done, err + } + dsts = dsts.Tail() + if err != nil { + if dsts.IsEmpty() && err == io.EOF { + return done, nil + } + return done, err + } + } + return done, nil +} + +func (r FromIOReaderAt) readToBlock(dst Block, buf []byte) (int, []byte, error) { + // io.Reader isn't safecopy-aware, so we have to buffer Blocks that require + // safecopy. + if !dst.NeedSafecopy() { + n, err := r.ReaderAt.ReadAt(dst.ToSlice(), r.Offset) + r.Offset += int64(n) + return n, buf, err + } + if len(buf) < dst.Len() { + buf = make([]byte, dst.Len()) + } + rn, rerr := r.ReaderAt.ReadAt(buf[:dst.Len()], r.Offset) + r.Offset += int64(rn) + wbn, wberr := Copy(dst, BlockFromSafeSlice(buf[:rn])) + if wberr != nil { + return wbn, buf, wberr + } + return wbn, buf, rerr +} + +// FromIOWriter implements Writer for an io.Writer by repeatedly invoking +// io.Writer.Write until it returns an error or partial write. +// +// FromIOWriter will tolerate implementations of io.Writer.Write that return +// partial writes with a nil error in contravention of io.Writer's +// requirements, since Writer is permitted to do so. FromIOWriter will return a +// successful partial write iff Writer.Write does so. +type FromIOWriter struct { + Writer io.Writer +} + +// WriteFromBlocks implements Writer.WriteFromBlocks. +func (w FromIOWriter) WriteFromBlocks(srcs BlockSeq) (uint64, error) { + var buf []byte + var done uint64 + for !srcs.IsEmpty() { + src := srcs.Head() + var n int + var err error + n, buf, err = w.writeFromBlock(src, buf) + done += uint64(n) + if n != src.Len() || err != nil { + return done, err + } + srcs = srcs.Tail() + } + return done, nil +} + +func (w FromIOWriter) writeFromBlock(src Block, buf []byte) (int, []byte, error) { + // io.Writer isn't safecopy-aware, so we have to buffer Blocks that require + // safecopy. + if !src.NeedSafecopy() { + n, err := w.Writer.Write(src.ToSlice()) + return n, buf, err + } + if len(buf) < src.Len() { + buf = make([]byte, src.Len()) + } + bufn, buferr := Copy(BlockFromSafeSlice(buf[:src.Len()]), src) + wn, werr := w.Writer.Write(buf[:bufn]) + if werr != nil { + return wn, buf, werr + } + return wn, buf, buferr +} + +// FromVecReaderFunc implements Reader for a function that reads data into a +// [][]byte and returns the number of bytes read as an int64. +type FromVecReaderFunc struct { + ReadVec func(dsts [][]byte) (int64, error) +} + +// ReadToBlocks implements Reader.ReadToBlocks. +// +// ReadToBlocks calls r.ReadVec at most once. +func (r FromVecReaderFunc) ReadToBlocks(dsts BlockSeq) (uint64, error) { + if dsts.IsEmpty() { + return 0, nil + } + // Ensure that we don't pass a [][]byte with a total length > MaxInt64. + dsts = dsts.TakeFirst64(uint64(math.MaxInt64)) + dstSlices := make([][]byte, 0, dsts.NumBlocks()) + // Buffer Blocks that require safecopy. + for tmp := dsts; !tmp.IsEmpty(); tmp = tmp.Tail() { + dst := tmp.Head() + if dst.NeedSafecopy() { + dstSlices = append(dstSlices, make([]byte, dst.Len())) + } else { + dstSlices = append(dstSlices, dst.ToSlice()) + } + } + rn, rerr := r.ReadVec(dstSlices) + dsts = dsts.TakeFirst64(uint64(rn)) + var done uint64 + var i int + for !dsts.IsEmpty() { + dst := dsts.Head() + if dst.NeedSafecopy() { + n, err := Copy(dst, BlockFromSafeSlice(dstSlices[i])) + done += uint64(n) + if err != nil { + return done, err + } + } else { + done += uint64(dst.Len()) + } + dsts = dsts.Tail() + i++ + } + return done, rerr +} + +// FromVecWriterFunc implements Writer for a function that writes data from a +// [][]byte and returns the number of bytes written. +type FromVecWriterFunc struct { + WriteVec func(srcs [][]byte) (int64, error) +} + +// WriteFromBlocks implements Writer.WriteFromBlocks. +// +// WriteFromBlocks calls w.WriteVec at most once. +func (w FromVecWriterFunc) WriteFromBlocks(srcs BlockSeq) (uint64, error) { + if srcs.IsEmpty() { + return 0, nil + } + // Ensure that we don't pass a [][]byte with a total length > MaxInt64. + srcs = srcs.TakeFirst64(uint64(math.MaxInt64)) + srcSlices := make([][]byte, 0, srcs.NumBlocks()) + // Buffer Blocks that require safecopy. + var buferr error + for tmp := srcs; !tmp.IsEmpty(); tmp = tmp.Tail() { + src := tmp.Head() + if src.NeedSafecopy() { + slice := make([]byte, src.Len()) + n, err := Copy(BlockFromSafeSlice(slice), src) + srcSlices = append(srcSlices, slice[:n]) + if err != nil { + buferr = err + break + } + } else { + srcSlices = append(srcSlices, src.ToSlice()) + } + } + n, err := w.WriteVec(srcSlices) + if err != nil { + return uint64(n), err + } + return uint64(n), buferr +} diff --git a/pkg/safemem/io_test.go b/pkg/safemem/io_test.go new file mode 100644 index 000000000..629741bee --- /dev/null +++ b/pkg/safemem/io_test.go @@ -0,0 +1,199 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package safemem + +import ( + "bytes" + "io" + "testing" +) + +func makeBlocks(slices ...[]byte) []Block { + blocks := make([]Block, 0, len(slices)) + for _, s := range slices { + blocks = append(blocks, BlockFromSafeSlice(s)) + } + return blocks +} + +func TestFromIOReaderFullRead(t *testing.T) { + r := FromIOReader{bytes.NewBufferString("foobar")} + dsts := makeBlocks(make([]byte, 3), make([]byte, 3)) + n, err := r.ReadToBlocks(BlockSeqFromSlice(dsts)) + if wantN := uint64(6); n != wantN || err != nil { + t.Errorf("ReadToBlocks: got (%v, %v), wanted (%v, nil)", n, err, wantN) + } + for i, want := range [][]byte{[]byte("foo"), []byte("bar")} { + if got := dsts[i].ToSlice(); !bytes.Equal(got, want) { + t.Errorf("dsts[%d]: got %q, wanted %q", i, got, want) + } + } +} + +type eofHidingReader struct { + Reader io.Reader +} + +func (r eofHidingReader) Read(dst []byte) (int, error) { + n, err := r.Reader.Read(dst) + if err == io.EOF { + return n, nil + } + return n, err +} + +func TestFromIOReaderPartialRead(t *testing.T) { + r := FromIOReader{eofHidingReader{bytes.NewBufferString("foob")}} + dsts := makeBlocks(make([]byte, 3), make([]byte, 3)) + n, err := r.ReadToBlocks(BlockSeqFromSlice(dsts)) + // FromIOReader should stop after the eofHidingReader returns (1, nil) + // for a 3-byte read. + if wantN := uint64(4); n != wantN || err != nil { + t.Errorf("ReadToBlocks: got (%v, %v), wanted (%v, nil)", n, err, wantN) + } + for i, want := range [][]byte{[]byte("foo"), []byte("b\x00\x00")} { + if got := dsts[i].ToSlice(); !bytes.Equal(got, want) { + t.Errorf("dsts[%d]: got %q, wanted %q", i, got, want) + } + } +} + +type singleByteReader struct { + Reader io.Reader +} + +func (r singleByteReader) Read(dst []byte) (int, error) { + if len(dst) == 0 { + return r.Reader.Read(dst) + } + return r.Reader.Read(dst[:1]) +} + +func TestSingleByteReader(t *testing.T) { + r := FromIOReader{singleByteReader{bytes.NewBufferString("foobar")}} + dsts := makeBlocks(make([]byte, 3), make([]byte, 3)) + n, err := r.ReadToBlocks(BlockSeqFromSlice(dsts)) + // FromIOReader should stop after the singleByteReader returns (1, nil) + // for a 3-byte read. + if wantN := uint64(1); n != wantN || err != nil { + t.Errorf("ReadToBlocks: got (%v, %v), wanted (%v, nil)", n, err, wantN) + } + for i, want := range [][]byte{[]byte("f\x00\x00"), []byte("\x00\x00\x00")} { + if got := dsts[i].ToSlice(); !bytes.Equal(got, want) { + t.Errorf("dsts[%d]: got %q, wanted %q", i, got, want) + } + } +} + +func TestReadFullToBlocks(t *testing.T) { + r := FromIOReader{singleByteReader{bytes.NewBufferString("foobar")}} + dsts := makeBlocks(make([]byte, 3), make([]byte, 3)) + n, err := ReadFullToBlocks(r, BlockSeqFromSlice(dsts)) + // ReadFullToBlocks should call into FromIOReader => singleByteReader + // repeatedly until dsts is exhausted. + if wantN := uint64(6); n != wantN || err != nil { + t.Errorf("ReadFullToBlocks: got (%v, %v), wanted (%v, nil)", n, err, wantN) + } + for i, want := range [][]byte{[]byte("foo"), []byte("bar")} { + if got := dsts[i].ToSlice(); !bytes.Equal(got, want) { + t.Errorf("dsts[%d]: got %q, wanted %q", i, got, want) + } + } +} + +func TestFromIOWriterFullWrite(t *testing.T) { + srcs := makeBlocks([]byte("foo"), []byte("bar")) + var dst bytes.Buffer + w := FromIOWriter{&dst} + n, err := w.WriteFromBlocks(BlockSeqFromSlice(srcs)) + if wantN := uint64(6); n != wantN || err != nil { + t.Errorf("WriteFromBlocks: got (%v, %v), wanted (%v, nil)", n, err, wantN) + } + if got, want := dst.Bytes(), []byte("foobar"); !bytes.Equal(got, want) { + t.Errorf("dst: got %q, wanted %q", got, want) + } +} + +type limitedWriter struct { + Writer io.Writer + Done int + Limit int +} + +func (w *limitedWriter) Write(src []byte) (int, error) { + count := len(src) + if count > (w.Limit - w.Done) { + count = w.Limit - w.Done + } + n, err := w.Writer.Write(src[:count]) + w.Done += n + return n, err +} + +func TestFromIOWriterPartialWrite(t *testing.T) { + srcs := makeBlocks([]byte("foo"), []byte("bar")) + var dst bytes.Buffer + w := FromIOWriter{&limitedWriter{&dst, 0, 4}} + n, err := w.WriteFromBlocks(BlockSeqFromSlice(srcs)) + // FromIOWriter should stop after the limitedWriter returns (1, nil) for a + // 3-byte write. + if wantN := uint64(4); n != wantN || err != nil { + t.Errorf("WriteFromBlocks: got (%v, %v), wanted (%v, nil)", n, err, wantN) + } + if got, want := dst.Bytes(), []byte("foob"); !bytes.Equal(got, want) { + t.Errorf("dst: got %q, wanted %q", got, want) + } +} + +type singleByteWriter struct { + Writer io.Writer +} + +func (w singleByteWriter) Write(src []byte) (int, error) { + if len(src) == 0 { + return w.Writer.Write(src) + } + return w.Writer.Write(src[:1]) +} + +func TestSingleByteWriter(t *testing.T) { + srcs := makeBlocks([]byte("foo"), []byte("bar")) + var dst bytes.Buffer + w := FromIOWriter{singleByteWriter{&dst}} + n, err := w.WriteFromBlocks(BlockSeqFromSlice(srcs)) + // FromIOWriter should stop after the singleByteWriter returns (1, nil) + // for a 3-byte write. + if wantN := uint64(1); n != wantN || err != nil { + t.Errorf("WriteFromBlocks: got (%v, %v), wanted (%v, nil)", n, err, wantN) + } + if got, want := dst.Bytes(), []byte("f"); !bytes.Equal(got, want) { + t.Errorf("dst: got %q, wanted %q", got, want) + } +} + +func TestWriteFullToBlocks(t *testing.T) { + srcs := makeBlocks([]byte("foo"), []byte("bar")) + var dst bytes.Buffer + w := FromIOWriter{singleByteWriter{&dst}} + n, err := WriteFullFromBlocks(w, BlockSeqFromSlice(srcs)) + // WriteFullToBlocks should call into FromIOWriter => singleByteWriter + // repeatedly until srcs is exhausted. + if wantN := uint64(6); n != wantN || err != nil { + t.Errorf("WriteFullFromBlocks: got (%v, %v), wanted (%v, nil)", n, err, wantN) + } + if got, want := dst.Bytes(), []byte("foobar"); !bytes.Equal(got, want) { + t.Errorf("dst: got %q, wanted %q", got, want) + } +} diff --git a/pkg/safemem/safemem.go b/pkg/safemem/safemem.go new file mode 100644 index 000000000..3e70d33a2 --- /dev/null +++ b/pkg/safemem/safemem.go @@ -0,0 +1,16 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package safemem provides the Block and BlockSeq types. +package safemem diff --git a/pkg/safemem/seq_test.go b/pkg/safemem/seq_test.go new file mode 100644 index 000000000..eba4bb535 --- /dev/null +++ b/pkg/safemem/seq_test.go @@ -0,0 +1,196 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package safemem + +import ( + "bytes" + "reflect" + "testing" +) + +type blockSeqTest struct { + desc string + + pieces []string + haveOffset bool + offset uint64 + haveLimit bool + limit uint64 + + want string +} + +func (t blockSeqTest) NonEmptyByteSlices() [][]byte { + // t is a value, so we can mutate it freely. + slices := make([][]byte, 0, len(t.pieces)) + for _, str := range t.pieces { + if t.haveOffset { + strOff := t.offset + if strOff > uint64(len(str)) { + strOff = uint64(len(str)) + } + str = str[strOff:] + t.offset -= strOff + } + if t.haveLimit { + strLim := t.limit + if strLim > uint64(len(str)) { + strLim = uint64(len(str)) + } + str = str[:strLim] + t.limit -= strLim + } + if len(str) != 0 { + slices = append(slices, []byte(str)) + } + } + return slices +} + +func (t blockSeqTest) BlockSeq() BlockSeq { + blocks := make([]Block, 0, len(t.pieces)) + for _, str := range t.pieces { + blocks = append(blocks, BlockFromSafeSlice([]byte(str))) + } + bs := BlockSeqFromSlice(blocks) + if t.haveOffset { + bs = bs.DropFirst64(t.offset) + } + if t.haveLimit { + bs = bs.TakeFirst64(t.limit) + } + return bs +} + +var blockSeqTests = []blockSeqTest{ + { + desc: "Empty sequence", + }, + { + desc: "Sequence of length 1", + pieces: []string{"foobar"}, + want: "foobar", + }, + { + desc: "Sequence of length 2", + pieces: []string{"foo", "bar"}, + want: "foobar", + }, + { + desc: "Empty Blocks", + pieces: []string{"", "foo", "", "", "bar", ""}, + want: "foobar", + }, + { + desc: "Sequence with non-zero offset", + pieces: []string{"foo", "bar"}, + haveOffset: true, + offset: 2, + want: "obar", + }, + { + desc: "Sequence with non-maximal limit", + pieces: []string{"foo", "bar"}, + haveLimit: true, + limit: 5, + want: "fooba", + }, + { + desc: "Sequence with offset and limit", + pieces: []string{"foo", "bar"}, + haveOffset: true, + offset: 2, + haveLimit: true, + limit: 3, + want: "oba", + }, +} + +func TestBlockSeqNumBytes(t *testing.T) { + for _, test := range blockSeqTests { + t.Run(test.desc, func(t *testing.T) { + if got, want := test.BlockSeq().NumBytes(), uint64(len(test.want)); got != want { + t.Errorf("NumBytes: got %d, wanted %d", got, want) + } + }) + } +} + +func TestBlockSeqIterBlocks(t *testing.T) { + // Tests BlockSeq iteration using Head/Tail. + for _, test := range blockSeqTests { + t.Run(test.desc, func(t *testing.T) { + srcs := test.BlockSeq() + // "Note that a non-nil empty slice and a nil slice ... are not + // deeply equal." - reflect + slices := make([][]byte, 0, 0) + for !srcs.IsEmpty() { + src := srcs.Head() + slices = append(slices, src.ToSlice()) + nextSrcs := srcs.Tail() + if got, want := nextSrcs.NumBytes(), srcs.NumBytes()-uint64(src.Len()); got != want { + t.Fatalf("%v.Tail(): got %v (%d bytes), wanted %d bytes", srcs, nextSrcs, got, want) + } + srcs = nextSrcs + } + if wantSlices := test.NonEmptyByteSlices(); !reflect.DeepEqual(slices, wantSlices) { + t.Errorf("Accumulated slices: got %v, wanted %v", slices, wantSlices) + } + }) + } +} + +func TestBlockSeqIterBytes(t *testing.T) { + // Tests BlockSeq iteration using Head/DropFirst. + for _, test := range blockSeqTests { + t.Run(test.desc, func(t *testing.T) { + srcs := test.BlockSeq() + var dst bytes.Buffer + for !srcs.IsEmpty() { + src := srcs.Head() + var b [1]byte + n, err := Copy(BlockFromSafeSlice(b[:]), src) + if n != 1 || err != nil { + t.Fatalf("Copy: got (%v, %v), wanted (1, nil)", n, err) + } + dst.WriteByte(b[0]) + nextSrcs := srcs.DropFirst(1) + if got, want := nextSrcs.NumBytes(), srcs.NumBytes()-1; got != want { + t.Fatalf("%v.DropFirst(1): got %v (%d bytes), wanted %d bytes", srcs, nextSrcs, got, want) + } + srcs = nextSrcs + } + if got := string(dst.Bytes()); got != test.want { + t.Errorf("Copied string: got %q, wanted %q", got, test.want) + } + }) + } +} + +func TestBlockSeqDropBeyondLimit(t *testing.T) { + blocks := []Block{BlockFromSafeSlice([]byte("123")), BlockFromSafeSlice([]byte("4"))} + bs := BlockSeqFromSlice(blocks) + if got, want := bs.NumBytes(), uint64(4); got != want { + t.Errorf("%v.NumBytes(): got %d, wanted %d", bs, got, want) + } + bs = bs.TakeFirst(1) + if got, want := bs.NumBytes(), uint64(1); got != want { + t.Errorf("%v.NumBytes(): got %d, wanted %d", bs, got, want) + } + bs = bs.DropFirst(2) + if got, want := bs.NumBytes(), uint64(0); got != want { + t.Errorf("%v.NumBytes(): got %d, wanted %d", bs, got, want) + } +} diff --git a/pkg/safemem/seq_unsafe.go b/pkg/safemem/seq_unsafe.go new file mode 100644 index 000000000..354a95dde --- /dev/null +++ b/pkg/safemem/seq_unsafe.go @@ -0,0 +1,299 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package safemem + +import ( + "bytes" + "fmt" + "reflect" + "unsafe" +) + +// A BlockSeq represents a sequence of Blocks, each of which has non-zero +// length. +// +// BlockSeqs are immutable and may be copied by value. The zero value of +// BlockSeq represents an empty sequence. +type BlockSeq struct { + // If length is 0, then the BlockSeq is empty. Invariants: data == 0; + // offset == 0; limit == 0. + // + // If length is -1, then the BlockSeq represents the single Block{data, + // limit, false}. Invariants: offset == 0; limit > 0; limit does not + // overflow the range of an int. + // + // If length is -2, then the BlockSeq represents the single Block{data, + // limit, true}. Invariants: offset == 0; limit > 0; limit does not + // overflow the range of an int. + // + // Otherwise, length >= 2, and the BlockSeq represents the `length` Blocks + // in the array of Blocks starting at address `data`, starting at `offset` + // bytes into the first Block and limited to the following `limit` bytes. + // Invariants: data != 0; offset < len(data[0]); limit > 0; offset+limit <= + // the combined length of all Blocks in the array; the first Block in the + // array has non-zero length. + // + // length is never 1; sequences consisting of a single Block are always + // stored inline (with length < 0). + data unsafe.Pointer + length int + offset int + limit uint64 +} + +// BlockSeqOf returns a BlockSeq representing the single Block b. +func BlockSeqOf(b Block) BlockSeq { + bs := BlockSeq{ + data: b.start, + length: -1, + limit: uint64(b.length), + } + if b.needSafecopy { + bs.length = -2 + } + return bs +} + +// BlockSeqFromSlice returns a BlockSeq representing all Blocks in slice. +// If slice contains Blocks with zero length, BlockSeq will skip them during +// iteration. +// +// Whether the returned BlockSeq shares memory with slice is unspecified; +// clients should avoid mutating slices passed to BlockSeqFromSlice. +// +// Preconditions: The combined length of all Blocks in slice <= math.MaxUint64. +func BlockSeqFromSlice(slice []Block) BlockSeq { + slice = skipEmpty(slice) + var limit uint64 + for _, b := range slice { + sum := limit + uint64(b.Len()) + if sum < limit { + panic("BlockSeq length overflows uint64") + } + limit = sum + } + return blockSeqFromSliceLimited(slice, limit) +} + +// Preconditions: The combined length of all Blocks in slice <= limit. If +// len(slice) != 0, the first Block in slice has non-zero length, and limit > +// 0. +func blockSeqFromSliceLimited(slice []Block, limit uint64) BlockSeq { + switch len(slice) { + case 0: + return BlockSeq{} + case 1: + return BlockSeqOf(slice[0].TakeFirst64(limit)) + default: + return BlockSeq{ + data: unsafe.Pointer(&slice[0]), + length: len(slice), + limit: limit, + } + } +} + +func skipEmpty(slice []Block) []Block { + for i, b := range slice { + if b.Len() != 0 { + return slice[i:] + } + } + return nil +} + +// IsEmpty returns true if bs contains no Blocks. +// +// Invariants: bs.IsEmpty() == (bs.NumBlocks() == 0) == (bs.NumBytes() == 0). +// (Of these, prefer to use bs.IsEmpty().) +func (bs BlockSeq) IsEmpty() bool { + return bs.length == 0 +} + +// NumBlocks returns the number of Blocks in bs. +func (bs BlockSeq) NumBlocks() int { + // In general, we have to count: if bs represents a windowed slice then the + // slice may contain Blocks with zero length, and bs.length may be larger + // than the actual number of Blocks due to bs.limit. + var n int + for !bs.IsEmpty() { + n++ + bs = bs.Tail() + } + return n +} + +// NumBytes returns the sum of Block.Len() for all Blocks in bs. +func (bs BlockSeq) NumBytes() uint64 { + return bs.limit +} + +// Head returns the first Block in bs. +// +// Preconditions: !bs.IsEmpty(). +func (bs BlockSeq) Head() Block { + if bs.length == 0 { + panic("empty BlockSeq") + } + if bs.length < 0 { + return bs.internalBlock() + } + return (*Block)(bs.data).DropFirst(bs.offset).TakeFirst64(bs.limit) +} + +// Preconditions: bs.length < 0. +func (bs BlockSeq) internalBlock() Block { + return Block{ + start: bs.data, + length: int(bs.limit), + needSafecopy: bs.length == -2, + } +} + +// Tail returns a BlockSeq consisting of all Blocks in bs after the first. +// +// Preconditions: !bs.IsEmpty(). +func (bs BlockSeq) Tail() BlockSeq { + if bs.length == 0 { + panic("empty BlockSeq") + } + if bs.length < 0 { + return BlockSeq{} + } + head := (*Block)(bs.data).DropFirst(bs.offset) + headLen := uint64(head.Len()) + if headLen >= bs.limit { + // The head Block exhausts the limit, so the tail is empty. + return BlockSeq{} + } + var extSlice []Block + extSliceHdr := (*reflect.SliceHeader)(unsafe.Pointer(&extSlice)) + extSliceHdr.Data = uintptr(bs.data) + extSliceHdr.Len = bs.length + extSliceHdr.Cap = bs.length + tailSlice := skipEmpty(extSlice[1:]) + tailLimit := bs.limit - headLen + return blockSeqFromSliceLimited(tailSlice, tailLimit) +} + +// DropFirst returns a BlockSeq equivalent to bs, but with the first n bytes +// omitted. If n > bs.NumBytes(), DropFirst returns an empty BlockSeq. +// +// Preconditions: n >= 0. +func (bs BlockSeq) DropFirst(n int) BlockSeq { + if n < 0 { + panic(fmt.Sprintf("invalid n: %d", n)) + } + return bs.DropFirst64(uint64(n)) +} + +// DropFirst64 is equivalent to DropFirst but takes an uint64. +func (bs BlockSeq) DropFirst64(n uint64) BlockSeq { + if n >= bs.limit { + return BlockSeq{} + } + for { + // Calling bs.Head() here is surprisingly expensive, so inline getting + // the head's length. + var headLen uint64 + if bs.length < 0 { + headLen = bs.limit + } else { + headLen = uint64((*Block)(bs.data).Len() - bs.offset) + } + if n < headLen { + // Dropping ends partway through the head Block. + if bs.length < 0 { + return BlockSeqOf(bs.internalBlock().DropFirst64(n)) + } + bs.offset += int(n) + bs.limit -= n + return bs + } + n -= headLen + bs = bs.Tail() + } +} + +// TakeFirst returns a BlockSeq equivalent to the first n bytes of bs. If n > +// bs.NumBytes(), TakeFirst returns a BlockSeq equivalent to bs. +// +// Preconditions: n >= 0. +func (bs BlockSeq) TakeFirst(n int) BlockSeq { + if n < 0 { + panic(fmt.Sprintf("invalid n: %d", n)) + } + return bs.TakeFirst64(uint64(n)) +} + +// TakeFirst64 is equivalent to TakeFirst but takes a uint64. +func (bs BlockSeq) TakeFirst64(n uint64) BlockSeq { + if n == 0 { + return BlockSeq{} + } + if bs.limit > n { + bs.limit = n + } + return bs +} + +// String implements fmt.Stringer.String. +func (bs BlockSeq) String() string { + var buf bytes.Buffer + buf.WriteByte('[') + var sep string + for !bs.IsEmpty() { + buf.WriteString(sep) + sep = " " + buf.WriteString(bs.Head().String()) + bs = bs.Tail() + } + buf.WriteByte(']') + return buf.String() +} + +// CopySeq copies srcs.NumBytes() or dsts.NumBytes() bytes, whichever is less, +// from srcs to dsts and returns the number of bytes copied. +// +// If srcs and dsts overlap, the data stored in dsts is unspecified. +func CopySeq(dsts, srcs BlockSeq) (uint64, error) { + var done uint64 + for !dsts.IsEmpty() && !srcs.IsEmpty() { + dst := dsts.Head() + src := srcs.Head() + n, err := Copy(dst, src) + done += uint64(n) + if err != nil { + return done, err + } + dsts = dsts.DropFirst(n) + srcs = srcs.DropFirst(n) + } + return done, nil +} + +// ZeroSeq sets all bytes in dsts to 0 and returns the number of bytes zeroed. +func ZeroSeq(dsts BlockSeq) (uint64, error) { + var done uint64 + for !dsts.IsEmpty() { + n, err := Zero(dsts.Head()) + done += uint64(n) + if err != nil { + return done, err + } + dsts = dsts.DropFirst(n) + } + return done, nil +} diff --git a/pkg/sentry/arch/BUILD b/pkg/sentry/arch/BUILD index 51ca09b24..34c0a867d 100644 --- a/pkg/sentry/arch/BUILD +++ b/pkg/sentry/arch/BUILD @@ -30,13 +30,13 @@ go_library( ":registers_go_proto", "//pkg/abi/linux", "//pkg/binary", + "//pkg/context", "//pkg/cpuid", "//pkg/log", - "//pkg/sentry/context", "//pkg/sentry/limits", - "//pkg/sentry/usermem", "//pkg/sync", "//pkg/syserror", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/arch/arch.go b/pkg/sentry/arch/arch.go index 81ec98a77..1d11cc472 100644 --- a/pkg/sentry/arch/arch.go +++ b/pkg/sentry/arch/arch.go @@ -24,7 +24,7 @@ import ( "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/limits" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // Arch describes an architecture. diff --git a/pkg/sentry/arch/arch_aarch64.go b/pkg/sentry/arch/arch_aarch64.go index ea4dedbdf..3b6987665 100644 --- a/pkg/sentry/arch/arch_aarch64.go +++ b/pkg/sentry/arch/arch_aarch64.go @@ -25,8 +25,8 @@ import ( "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/log" rpb "gvisor.dev/gvisor/pkg/sentry/arch/registers_go_proto" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) const ( diff --git a/pkg/sentry/arch/arch_amd64.go b/pkg/sentry/arch/arch_amd64.go index 2aa08b1a9..85d6acc0f 100644 --- a/pkg/sentry/arch/arch_amd64.go +++ b/pkg/sentry/arch/arch_amd64.go @@ -25,7 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/sentry/limits" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // Host specifies the host architecture. diff --git a/pkg/sentry/arch/arch_arm64.go b/pkg/sentry/arch/arch_arm64.go index 0d5b7d317..94f1a808f 100644 --- a/pkg/sentry/arch/arch_arm64.go +++ b/pkg/sentry/arch/arch_arm64.go @@ -21,7 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/sentry/limits" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // Host specifies the host architecture. diff --git a/pkg/sentry/arch/arch_state_x86.go b/pkg/sentry/arch/arch_state_x86.go index 84f11b0d1..d388ee9cf 100644 --- a/pkg/sentry/arch/arch_state_x86.go +++ b/pkg/sentry/arch/arch_state_x86.go @@ -21,7 +21,7 @@ import ( "syscall" "gvisor.dev/gvisor/pkg/cpuid" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // ErrFloatingPoint indicates a failed restore due to unusable floating point diff --git a/pkg/sentry/arch/arch_x86.go b/pkg/sentry/arch/arch_x86.go index 9f41e566f..a18093155 100644 --- a/pkg/sentry/arch/arch_x86.go +++ b/pkg/sentry/arch/arch_x86.go @@ -25,9 +25,9 @@ import ( "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/log" rpb "gvisor.dev/gvisor/pkg/sentry/arch/registers_go_proto" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // System-related constants for x86. diff --git a/pkg/sentry/arch/auxv.go b/pkg/sentry/arch/auxv.go index 4546b2ef9..2b4c8f3fc 100644 --- a/pkg/sentry/arch/auxv.go +++ b/pkg/sentry/arch/auxv.go @@ -15,7 +15,7 @@ package arch import ( - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // An AuxEntry represents an entry in an ELF auxiliary vector. diff --git a/pkg/sentry/arch/signal.go b/pkg/sentry/arch/signal.go index 402e46025..8b03d0187 100644 --- a/pkg/sentry/arch/signal.go +++ b/pkg/sentry/arch/signal.go @@ -16,7 +16,7 @@ package arch import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // SignalAct represents the action that should be taken when a signal is diff --git a/pkg/sentry/arch/signal_amd64.go b/pkg/sentry/arch/signal_amd64.go index 1e4f9c3c2..81b92bb43 100644 --- a/pkg/sentry/arch/signal_amd64.go +++ b/pkg/sentry/arch/signal_amd64.go @@ -23,7 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // SignalContext64 is equivalent to struct sigcontext, the type passed as the diff --git a/pkg/sentry/arch/signal_arm64.go b/pkg/sentry/arch/signal_arm64.go index 7d0e98935..4f4cc46a8 100644 --- a/pkg/sentry/arch/signal_arm64.go +++ b/pkg/sentry/arch/signal_arm64.go @@ -19,7 +19,7 @@ import ( "syscall" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // SignalContext64 is equivalent to struct sigcontext, the type passed as the diff --git a/pkg/sentry/arch/signal_stack.go b/pkg/sentry/arch/signal_stack.go index d324da705..1a6056171 100644 --- a/pkg/sentry/arch/signal_stack.go +++ b/pkg/sentry/arch/signal_stack.go @@ -17,7 +17,7 @@ package arch import ( - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) const ( diff --git a/pkg/sentry/arch/stack.go b/pkg/sentry/arch/stack.go index 7472c3c61..09bceabc9 100644 --- a/pkg/sentry/arch/stack.go +++ b/pkg/sentry/arch/stack.go @@ -18,8 +18,8 @@ import ( "encoding/binary" "fmt" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/usermem" ) // Stack is a simple wrapper around a usermem.IO and an address. diff --git a/pkg/sentry/context/BUILD b/pkg/sentry/context/BUILD deleted file mode 100644 index e13a9ce20..000000000 --- a/pkg/sentry/context/BUILD +++ /dev/null @@ -1,13 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "context", - srcs = ["context.go"], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/amutex", - "//pkg/log", - ], -) diff --git a/pkg/sentry/context/context.go b/pkg/sentry/context/context.go deleted file mode 100644 index 23e009ef3..000000000 --- a/pkg/sentry/context/context.go +++ /dev/null @@ -1,141 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package context defines an internal context type. -// -// The given Context conforms to the standard Go context, but mandates -// additional methods that are specific to the kernel internals. Note however, -// that the Context described by this package carries additional constraints -// regarding concurrent access and retaining beyond the scope of a call. -// -// See the Context type for complete details. -package context - -import ( - "context" - "time" - - "gvisor.dev/gvisor/pkg/amutex" - "gvisor.dev/gvisor/pkg/log" -) - -type contextID int - -// Globally accessible values from a context. These keys are defined in the -// context package to resolve dependency cycles by not requiring the caller to -// import packages usually required to get these information. -const ( - // CtxThreadGroupID is the current thread group ID when a context represents - // a task context. The value is represented as an int32. - CtxThreadGroupID contextID = iota -) - -// ThreadGroupIDFromContext returns the current thread group ID when ctx -// represents a task context. -func ThreadGroupIDFromContext(ctx Context) (tgid int32, ok bool) { - if tgid := ctx.Value(CtxThreadGroupID); tgid != nil { - return tgid.(int32), true - } - return 0, false -} - -// A Context represents a thread of execution (hereafter "goroutine" to reflect -// Go idiosyncrasy). It carries state associated with the goroutine across API -// boundaries. -// -// While Context exists for essentially the same reasons as Go's standard -// context.Context, the standard type represents the state of an operation -// rather than that of a goroutine. This is a critical distinction: -// -// - Unlike context.Context, which "may be passed to functions running in -// different goroutines", it is *not safe* to use the same Context in multiple -// concurrent goroutines. -// -// - It is *not safe* to retain a Context passed to a function beyond the scope -// of that function call. -// -// In both cases, values extracted from the Context should be used instead. -type Context interface { - log.Logger - amutex.Sleeper - context.Context - - // UninterruptibleSleepStart indicates the beginning of an uninterruptible - // sleep state (equivalent to Linux's TASK_UNINTERRUPTIBLE). If deactivate - // is true and the Context represents a Task, the Task's AddressSpace is - // deactivated. - UninterruptibleSleepStart(deactivate bool) - - // UninterruptibleSleepFinish indicates the end of an uninterruptible sleep - // state that was begun by a previous call to UninterruptibleSleepStart. If - // activate is true and the Context represents a Task, the Task's - // AddressSpace is activated. Normally activate is the same value as the - // deactivate parameter passed to UninterruptibleSleepStart. - UninterruptibleSleepFinish(activate bool) -} - -// NoopSleeper is a noop implementation of amutex.Sleeper and UninterruptibleSleep -// methods for anonymous embedding in other types that do not implement sleeps. -type NoopSleeper struct { - amutex.NoopSleeper -} - -// UninterruptibleSleepStart does nothing. -func (NoopSleeper) UninterruptibleSleepStart(bool) {} - -// UninterruptibleSleepFinish does nothing. -func (NoopSleeper) UninterruptibleSleepFinish(bool) {} - -// Deadline returns zero values, meaning no deadline. -func (NoopSleeper) Deadline() (time.Time, bool) { - return time.Time{}, false -} - -// Done returns nil. -func (NoopSleeper) Done() <-chan struct{} { - return nil -} - -// Err returns nil. -func (NoopSleeper) Err() error { - return nil -} - -// logContext implements basic logging. -type logContext struct { - log.Logger - NoopSleeper -} - -// Value implements Context.Value. -func (logContext) Value(key interface{}) interface{} { - return nil -} - -// bgContext is the context returned by context.Background. -var bgContext = &logContext{Logger: log.Log()} - -// Background returns an empty context using the default logger. -// -// Users should be wary of using a Background context. Please tag any use with -// FIXME(b/38173783) and a note to remove this use. -// -// Generally, one should use the Task as their context when available, or avoid -// having to use a context in places where a Task is unavailable. -// -// Using a Background context for tests is fine, as long as no values are -// needed from the context in the tested code paths. -func Background() Context { - return bgContext -} diff --git a/pkg/sentry/context/contexttest/BUILD b/pkg/sentry/context/contexttest/BUILD deleted file mode 100644 index f91a6d4ed..000000000 --- a/pkg/sentry/context/contexttest/BUILD +++ /dev/null @@ -1,21 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "contexttest", - testonly = 1, - srcs = ["contexttest.go"], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/memutil", - "//pkg/sentry/context", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/time", - "//pkg/sentry/limits", - "//pkg/sentry/pgalloc", - "//pkg/sentry/platform", - "//pkg/sentry/platform/ptrace", - "//pkg/sentry/uniqueid", - ], -) diff --git a/pkg/sentry/context/contexttest/contexttest.go b/pkg/sentry/context/contexttest/contexttest.go deleted file mode 100644 index 15cf086a9..000000000 --- a/pkg/sentry/context/contexttest/contexttest.go +++ /dev/null @@ -1,188 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package contexttest builds a test context.Context. -package contexttest - -import ( - "os" - "sync/atomic" - "testing" - "time" - - "gvisor.dev/gvisor/pkg/memutil" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/limits" - "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/platform/ptrace" - "gvisor.dev/gvisor/pkg/sentry/uniqueid" -) - -// Context returns a Context that may be used in tests. Uses ptrace as the -// platform.Platform. -// -// Note that some filesystems may require a minimal kernel for testing, which -// this test context does not provide. For such tests, see kernel/contexttest. -func Context(tb testing.TB) context.Context { - const memfileName = "contexttest-memory" - memfd, err := memutil.CreateMemFD(memfileName, 0) - if err != nil { - tb.Fatalf("error creating application memory file: %v", err) - } - memfile := os.NewFile(uintptr(memfd), memfileName) - mf, err := pgalloc.NewMemoryFile(memfile, pgalloc.MemoryFileOpts{}) - if err != nil { - memfile.Close() - tb.Fatalf("error creating pgalloc.MemoryFile: %v", err) - } - p, err := ptrace.New() - if err != nil { - tb.Fatal(err) - } - // Test usage of context.Background is fine. - return &TestContext{ - Context: context.Background(), - l: limits.NewLimitSet(), - mf: mf, - platform: p, - creds: auth.NewAnonymousCredentials(), - otherValues: make(map[interface{}]interface{}), - } -} - -// TestContext represents a context with minimal functionality suitable for -// running tests. -type TestContext struct { - context.Context - l *limits.LimitSet - mf *pgalloc.MemoryFile - platform platform.Platform - creds *auth.Credentials - otherValues map[interface{}]interface{} -} - -// globalUniqueID tracks incremental unique identifiers for tests. -var globalUniqueID uint64 - -// globalUniqueIDProvider implements unix.UniqueIDProvider. -type globalUniqueIDProvider struct{} - -// UniqueID implements unix.UniqueIDProvider.UniqueID. -func (*globalUniqueIDProvider) UniqueID() uint64 { - return atomic.AddUint64(&globalUniqueID, 1) -} - -// lastInotifyCookie is a monotonically increasing counter for generating unique -// inotify cookies. Must be accessed using atomic ops. -var lastInotifyCookie uint32 - -// hostClock implements ktime.Clock. -type hostClock struct { - ktime.WallRateClock - ktime.NoClockEvents -} - -// Now implements ktime.Clock.Now. -func (hostClock) Now() ktime.Time { - return ktime.FromNanoseconds(time.Now().UnixNano()) -} - -// RegisterValue registers additional values with this test context. Useful for -// providing values from external packages that contexttest can't depend on. -func (t *TestContext) RegisterValue(key, value interface{}) { - t.otherValues[key] = value -} - -// Value implements context.Context. -func (t *TestContext) Value(key interface{}) interface{} { - switch key { - case auth.CtxCredentials: - return t.creds - case limits.CtxLimits: - return t.l - case pgalloc.CtxMemoryFile: - return t.mf - case pgalloc.CtxMemoryFileProvider: - return t - case platform.CtxPlatform: - return t.platform - case uniqueid.CtxGlobalUniqueID: - return (*globalUniqueIDProvider).UniqueID(nil) - case uniqueid.CtxGlobalUniqueIDProvider: - return &globalUniqueIDProvider{} - case uniqueid.CtxInotifyCookie: - return atomic.AddUint32(&lastInotifyCookie, 1) - case ktime.CtxRealtimeClock: - return hostClock{} - default: - if val, ok := t.otherValues[key]; ok { - return val - } - return t.Context.Value(key) - } -} - -// MemoryFile implements pgalloc.MemoryFileProvider.MemoryFile. -func (t *TestContext) MemoryFile() *pgalloc.MemoryFile { - return t.mf -} - -// RootContext returns a Context that may be used in tests that need root -// credentials. Uses ptrace as the platform.Platform. -func RootContext(tb testing.TB) context.Context { - return WithCreds(Context(tb), auth.NewRootCredentials(auth.NewRootUserNamespace())) -} - -// WithCreds returns a copy of ctx carrying creds. -func WithCreds(ctx context.Context, creds *auth.Credentials) context.Context { - return &authContext{ctx, creds} -} - -type authContext struct { - context.Context - creds *auth.Credentials -} - -// Value implements context.Context. -func (ac *authContext) Value(key interface{}) interface{} { - switch key { - case auth.CtxCredentials: - return ac.creds - default: - return ac.Context.Value(key) - } -} - -// WithLimitSet returns a copy of ctx carrying l. -func WithLimitSet(ctx context.Context, l *limits.LimitSet) context.Context { - return limitContext{ctx, l} -} - -type limitContext struct { - context.Context - l *limits.LimitSet -} - -// Value implements context.Context. -func (lc limitContext) Value(key interface{}) interface{} { - switch key { - case limits.CtxLimits: - return lc.l - default: - return lc.Context.Value(key) - } -} diff --git a/pkg/sentry/contexttest/BUILD b/pkg/sentry/contexttest/BUILD new file mode 100644 index 000000000..6f4c86684 --- /dev/null +++ b/pkg/sentry/contexttest/BUILD @@ -0,0 +1,21 @@ +load("//tools:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_library( + name = "contexttest", + testonly = 1, + srcs = ["contexttest.go"], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/context", + "//pkg/memutil", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/kernel/time", + "//pkg/sentry/limits", + "//pkg/sentry/pgalloc", + "//pkg/sentry/platform", + "//pkg/sentry/platform/ptrace", + "//pkg/sentry/uniqueid", + ], +) diff --git a/pkg/sentry/contexttest/contexttest.go b/pkg/sentry/contexttest/contexttest.go new file mode 100644 index 000000000..031fc64ec --- /dev/null +++ b/pkg/sentry/contexttest/contexttest.go @@ -0,0 +1,188 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package contexttest builds a test context.Context. +package contexttest + +import ( + "os" + "sync/atomic" + "testing" + "time" + + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/memutil" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" + "gvisor.dev/gvisor/pkg/sentry/limits" + "gvisor.dev/gvisor/pkg/sentry/pgalloc" + "gvisor.dev/gvisor/pkg/sentry/platform" + "gvisor.dev/gvisor/pkg/sentry/platform/ptrace" + "gvisor.dev/gvisor/pkg/sentry/uniqueid" +) + +// Context returns a Context that may be used in tests. Uses ptrace as the +// platform.Platform. +// +// Note that some filesystems may require a minimal kernel for testing, which +// this test context does not provide. For such tests, see kernel/contexttest. +func Context(tb testing.TB) context.Context { + const memfileName = "contexttest-memory" + memfd, err := memutil.CreateMemFD(memfileName, 0) + if err != nil { + tb.Fatalf("error creating application memory file: %v", err) + } + memfile := os.NewFile(uintptr(memfd), memfileName) + mf, err := pgalloc.NewMemoryFile(memfile, pgalloc.MemoryFileOpts{}) + if err != nil { + memfile.Close() + tb.Fatalf("error creating pgalloc.MemoryFile: %v", err) + } + p, err := ptrace.New() + if err != nil { + tb.Fatal(err) + } + // Test usage of context.Background is fine. + return &TestContext{ + Context: context.Background(), + l: limits.NewLimitSet(), + mf: mf, + platform: p, + creds: auth.NewAnonymousCredentials(), + otherValues: make(map[interface{}]interface{}), + } +} + +// TestContext represents a context with minimal functionality suitable for +// running tests. +type TestContext struct { + context.Context + l *limits.LimitSet + mf *pgalloc.MemoryFile + platform platform.Platform + creds *auth.Credentials + otherValues map[interface{}]interface{} +} + +// globalUniqueID tracks incremental unique identifiers for tests. +var globalUniqueID uint64 + +// globalUniqueIDProvider implements unix.UniqueIDProvider. +type globalUniqueIDProvider struct{} + +// UniqueID implements unix.UniqueIDProvider.UniqueID. +func (*globalUniqueIDProvider) UniqueID() uint64 { + return atomic.AddUint64(&globalUniqueID, 1) +} + +// lastInotifyCookie is a monotonically increasing counter for generating unique +// inotify cookies. Must be accessed using atomic ops. +var lastInotifyCookie uint32 + +// hostClock implements ktime.Clock. +type hostClock struct { + ktime.WallRateClock + ktime.NoClockEvents +} + +// Now implements ktime.Clock.Now. +func (hostClock) Now() ktime.Time { + return ktime.FromNanoseconds(time.Now().UnixNano()) +} + +// RegisterValue registers additional values with this test context. Useful for +// providing values from external packages that contexttest can't depend on. +func (t *TestContext) RegisterValue(key, value interface{}) { + t.otherValues[key] = value +} + +// Value implements context.Context. +func (t *TestContext) Value(key interface{}) interface{} { + switch key { + case auth.CtxCredentials: + return t.creds + case limits.CtxLimits: + return t.l + case pgalloc.CtxMemoryFile: + return t.mf + case pgalloc.CtxMemoryFileProvider: + return t + case platform.CtxPlatform: + return t.platform + case uniqueid.CtxGlobalUniqueID: + return (*globalUniqueIDProvider).UniqueID(nil) + case uniqueid.CtxGlobalUniqueIDProvider: + return &globalUniqueIDProvider{} + case uniqueid.CtxInotifyCookie: + return atomic.AddUint32(&lastInotifyCookie, 1) + case ktime.CtxRealtimeClock: + return hostClock{} + default: + if val, ok := t.otherValues[key]; ok { + return val + } + return t.Context.Value(key) + } +} + +// MemoryFile implements pgalloc.MemoryFileProvider.MemoryFile. +func (t *TestContext) MemoryFile() *pgalloc.MemoryFile { + return t.mf +} + +// RootContext returns a Context that may be used in tests that need root +// credentials. Uses ptrace as the platform.Platform. +func RootContext(tb testing.TB) context.Context { + return WithCreds(Context(tb), auth.NewRootCredentials(auth.NewRootUserNamespace())) +} + +// WithCreds returns a copy of ctx carrying creds. +func WithCreds(ctx context.Context, creds *auth.Credentials) context.Context { + return &authContext{ctx, creds} +} + +type authContext struct { + context.Context + creds *auth.Credentials +} + +// Value implements context.Context. +func (ac *authContext) Value(key interface{}) interface{} { + switch key { + case auth.CtxCredentials: + return ac.creds + default: + return ac.Context.Value(key) + } +} + +// WithLimitSet returns a copy of ctx carrying l. +func WithLimitSet(ctx context.Context, l *limits.LimitSet) context.Context { + return limitContext{ctx, l} +} + +type limitContext struct { + context.Context + l *limits.LimitSet +} + +// Value implements context.Context. +func (lc limitContext) Value(key interface{}) interface{} { + switch key { + case limits.CtxLimits: + return lc.l + default: + return lc.Context.Value(key) + } +} diff --git a/pkg/sentry/fs/BUILD b/pkg/sentry/fs/BUILD index 605d61dbe..ea85ab33c 100644 --- a/pkg/sentry/fs/BUILD +++ b/pkg/sentry/fs/BUILD @@ -47,13 +47,13 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/amutex", + "//pkg/context", "//pkg/log", "//pkg/metric", "//pkg/p9", "//pkg/refs", "//pkg/secio", "//pkg/sentry/arch", - "//pkg/sentry/context", "//pkg/sentry/device", "//pkg/sentry/fs/lock", "//pkg/sentry/kernel/auth", @@ -64,10 +64,10 @@ go_library( "//pkg/sentry/socket/unix/transport", "//pkg/sentry/uniqueid", "//pkg/sentry/usage", - "//pkg/sentry/usermem", "//pkg/state", "//pkg/sync", "//pkg/syserror", + "//pkg/usermem", "//pkg/waiter", ], ) @@ -107,14 +107,14 @@ go_test( ], deps = [ ":fs", - "//pkg/sentry/context", + "//pkg/context", "//pkg/sentry/fs/fsutil", "//pkg/sentry/fs/ramfs", "//pkg/sentry/fs/tmpfs", "//pkg/sentry/kernel/contexttest", - "//pkg/sentry/usermem", "//pkg/sync", "//pkg/syserror", + "//pkg/usermem", ], ) @@ -129,7 +129,7 @@ go_test( ], library = ":fs", deps = [ - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", + "//pkg/context", + "//pkg/sentry/contexttest", ], ) diff --git a/pkg/sentry/fs/anon/BUILD b/pkg/sentry/fs/anon/BUILD index c14e5405e..aedcecfa1 100644 --- a/pkg/sentry/fs/anon/BUILD +++ b/pkg/sentry/fs/anon/BUILD @@ -11,10 +11,10 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", - "//pkg/sentry/context", + "//pkg/context", "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", - "//pkg/sentry/usermem", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/fs/anon/anon.go b/pkg/sentry/fs/anon/anon.go index 7323c7222..5c421f5fb 100644 --- a/pkg/sentry/fs/anon/anon.go +++ b/pkg/sentry/fs/anon/anon.go @@ -18,10 +18,10 @@ package anon import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // NewInode constructs an anonymous Inode that is not associated diff --git a/pkg/sentry/fs/attr.go b/pkg/sentry/fs/attr.go index 4f3d6410e..fa9e7d517 100644 --- a/pkg/sentry/fs/attr.go +++ b/pkg/sentry/fs/attr.go @@ -20,8 +20,8 @@ import ( "syscall" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/p9" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" ) diff --git a/pkg/sentry/fs/context.go b/pkg/sentry/fs/context.go index dd427de5d..0fbd60056 100644 --- a/pkg/sentry/fs/context.go +++ b/pkg/sentry/fs/context.go @@ -16,7 +16,7 @@ package fs import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ) diff --git a/pkg/sentry/fs/copy_up.go b/pkg/sentry/fs/copy_up.go index e03e3e417..f6c79e51b 100644 --- a/pkg/sentry/fs/copy_up.go +++ b/pkg/sentry/fs/copy_up.go @@ -19,12 +19,12 @@ import ( "io" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // copyUp copies a file in an overlay from a lower filesystem to an diff --git a/pkg/sentry/fs/copy_up_test.go b/pkg/sentry/fs/copy_up_test.go index 738580c5f..91792d9fe 100644 --- a/pkg/sentry/fs/copy_up_test.go +++ b/pkg/sentry/fs/copy_up_test.go @@ -24,8 +24,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" _ "gvisor.dev/gvisor/pkg/sentry/fs/tmpfs" "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/usermem" ) const ( diff --git a/pkg/sentry/fs/dev/BUILD b/pkg/sentry/fs/dev/BUILD index 0c7247bd7..4c4b7d5cc 100644 --- a/pkg/sentry/fs/dev/BUILD +++ b/pkg/sentry/fs/dev/BUILD @@ -16,8 +16,9 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/rand", - "//pkg/sentry/context", + "//pkg/safemem", "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", @@ -26,9 +27,8 @@ go_library( "//pkg/sentry/memmap", "//pkg/sentry/mm", "//pkg/sentry/pgalloc", - "//pkg/sentry/safemem", - "//pkg/sentry/usermem", "//pkg/syserror", + "//pkg/usermem", "//pkg/waiter", ], ) diff --git a/pkg/sentry/fs/dev/dev.go b/pkg/sentry/fs/dev/dev.go index f739c476c..35bd23991 100644 --- a/pkg/sentry/fs/dev/dev.go +++ b/pkg/sentry/fs/dev/dev.go @@ -18,11 +18,11 @@ package dev import ( "math" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" "gvisor.dev/gvisor/pkg/sentry/fs/tmpfs" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // Memory device numbers are from Linux's drivers/char/mem.c diff --git a/pkg/sentry/fs/dev/fs.go b/pkg/sentry/fs/dev/fs.go index 55f8af704..5e518fb63 100644 --- a/pkg/sentry/fs/dev/fs.go +++ b/pkg/sentry/fs/dev/fs.go @@ -15,7 +15,7 @@ package dev import ( - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" ) diff --git a/pkg/sentry/fs/dev/full.go b/pkg/sentry/fs/dev/full.go index 07e0ea010..deb9c6ad8 100644 --- a/pkg/sentry/fs/dev/full.go +++ b/pkg/sentry/fs/dev/full.go @@ -16,11 +16,11 @@ package dev import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/dev/null.go b/pkg/sentry/fs/dev/null.go index 4404b97ef..aec33d0d9 100644 --- a/pkg/sentry/fs/dev/null.go +++ b/pkg/sentry/fs/dev/null.go @@ -16,7 +16,7 @@ package dev import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/memmap" diff --git a/pkg/sentry/fs/dev/random.go b/pkg/sentry/fs/dev/random.go index 49cb92f6e..2a9bbeb18 100644 --- a/pkg/sentry/fs/dev/random.go +++ b/pkg/sentry/fs/dev/random.go @@ -16,12 +16,12 @@ package dev import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/rand" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/safemem" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/dev/tty.go b/pkg/sentry/fs/dev/tty.go index 87d80e292..760ca563d 100644 --- a/pkg/sentry/fs/dev/tty.go +++ b/pkg/sentry/fs/dev/tty.go @@ -16,7 +16,7 @@ package dev import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/waiter" diff --git a/pkg/sentry/fs/dirent.go b/pkg/sentry/fs/dirent.go index 31fc4d87b..acab0411a 100644 --- a/pkg/sentry/fs/dirent.go +++ b/pkg/sentry/fs/dirent.go @@ -22,8 +22,8 @@ import ( "syscall" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/uniqueid" diff --git a/pkg/sentry/fs/dirent_refs_test.go b/pkg/sentry/fs/dirent_refs_test.go index 47bc72a88..98d69c6f2 100644 --- a/pkg/sentry/fs/dirent_refs_test.go +++ b/pkg/sentry/fs/dirent_refs_test.go @@ -18,8 +18,8 @@ import ( "syscall" "testing" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/contexttest" ) func newMockDirInode(ctx context.Context, cache *DirentCache) *Inode { diff --git a/pkg/sentry/fs/fdpipe/BUILD b/pkg/sentry/fs/fdpipe/BUILD index 25ef96299..1d09e983c 100644 --- a/pkg/sentry/fs/fdpipe/BUILD +++ b/pkg/sentry/fs/fdpipe/BUILD @@ -12,17 +12,17 @@ go_library( imports = ["gvisor.dev/gvisor/pkg/sentry/fs"], visibility = ["//pkg/sentry:internal"], deps = [ + "//pkg/context", "//pkg/fd", "//pkg/fdnotifier", "//pkg/log", + "//pkg/safemem", "//pkg/secio", - "//pkg/sentry/context", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", - "//pkg/sentry/safemem", - "//pkg/sentry/usermem", "//pkg/sync", "//pkg/syserror", + "//pkg/usermem", "//pkg/waiter", ], ) @@ -36,13 +36,13 @@ go_test( ], library = ":fdpipe", deps = [ + "//pkg/context", "//pkg/fd", "//pkg/fdnotifier", - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", + "//pkg/sentry/contexttest", "//pkg/sentry/fs", - "//pkg/sentry/usermem", "//pkg/syserror", + "//pkg/usermem", "@com_github_google_uuid//:go_default_library", ], ) diff --git a/pkg/sentry/fs/fdpipe/pipe.go b/pkg/sentry/fs/fdpipe/pipe.go index 5b6cfeb0a..9fce177ad 100644 --- a/pkg/sentry/fs/fdpipe/pipe.go +++ b/pkg/sentry/fs/fdpipe/pipe.go @@ -19,17 +19,17 @@ import ( "os" "syscall" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fd" "gvisor.dev/gvisor/pkg/fdnotifier" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/secio" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/safemem" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/fdpipe/pipe_opener.go b/pkg/sentry/fs/fdpipe/pipe_opener.go index 64b558975..0c3595998 100644 --- a/pkg/sentry/fs/fdpipe/pipe_opener.go +++ b/pkg/sentry/fs/fdpipe/pipe_opener.go @@ -20,8 +20,8 @@ import ( "syscall" "time" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fd" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/fs/fdpipe/pipe_opener_test.go b/pkg/sentry/fs/fdpipe/pipe_opener_test.go index 577445148..e556da48a 100644 --- a/pkg/sentry/fs/fdpipe/pipe_opener_test.go +++ b/pkg/sentry/fs/fdpipe/pipe_opener_test.go @@ -26,12 +26,12 @@ import ( "github.com/google/uuid" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fd" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) type hostOpener struct { diff --git a/pkg/sentry/fs/fdpipe/pipe_state.go b/pkg/sentry/fs/fdpipe/pipe_state.go index cee87f726..af8230a7d 100644 --- a/pkg/sentry/fs/fdpipe/pipe_state.go +++ b/pkg/sentry/fs/fdpipe/pipe_state.go @@ -18,7 +18,7 @@ import ( "fmt" "io/ioutil" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sync" ) diff --git a/pkg/sentry/fs/fdpipe/pipe_test.go b/pkg/sentry/fs/fdpipe/pipe_test.go index 69abc1e71..5aff0cc95 100644 --- a/pkg/sentry/fs/fdpipe/pipe_test.go +++ b/pkg/sentry/fs/fdpipe/pipe_test.go @@ -23,10 +23,10 @@ import ( "gvisor.dev/gvisor/pkg/fd" "gvisor.dev/gvisor/pkg/fdnotifier" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) func singlePipeFD() (int, error) { diff --git a/pkg/sentry/fs/file.go b/pkg/sentry/fs/file.go index 7c4586296..ca3466f4f 100644 --- a/pkg/sentry/fs/file.go +++ b/pkg/sentry/fs/file.go @@ -20,16 +20,16 @@ import ( "time" "gvisor.dev/gvisor/pkg/amutex" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/metric" "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/uniqueid" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/file_operations.go b/pkg/sentry/fs/file_operations.go index b88303f17..beba0f771 100644 --- a/pkg/sentry/fs/file_operations.go +++ b/pkg/sentry/fs/file_operations.go @@ -17,10 +17,10 @@ package fs import ( "io" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/file_overlay.go b/pkg/sentry/fs/file_overlay.go index 8991207b4..dcc1df38f 100644 --- a/pkg/sentry/fs/file_overlay.go +++ b/pkg/sentry/fs/file_overlay.go @@ -17,13 +17,13 @@ package fs import ( "io" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/file_overlay_test.go b/pkg/sentry/fs/file_overlay_test.go index 2fb824d5c..02538bb4f 100644 --- a/pkg/sentry/fs/file_overlay_test.go +++ b/pkg/sentry/fs/file_overlay_test.go @@ -18,7 +18,7 @@ import ( "reflect" "testing" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" diff --git a/pkg/sentry/fs/filesystems.go b/pkg/sentry/fs/filesystems.go index c5b51620a..084da2a8d 100644 --- a/pkg/sentry/fs/filesystems.go +++ b/pkg/sentry/fs/filesystems.go @@ -19,7 +19,7 @@ import ( "sort" "strings" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sync" ) diff --git a/pkg/sentry/fs/filetest/BUILD b/pkg/sentry/fs/filetest/BUILD index 9a7608cae..a8000e010 100644 --- a/pkg/sentry/fs/filetest/BUILD +++ b/pkg/sentry/fs/filetest/BUILD @@ -8,12 +8,12 @@ go_library( srcs = ["filetest.go"], visibility = ["//pkg/sentry:internal"], deps = [ - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", + "//pkg/context", + "//pkg/sentry/contexttest", "//pkg/sentry/fs", "//pkg/sentry/fs/anon", "//pkg/sentry/fs/fsutil", - "//pkg/sentry/usermem", + "//pkg/usermem", "//pkg/waiter", ], ) diff --git a/pkg/sentry/fs/filetest/filetest.go b/pkg/sentry/fs/filetest/filetest.go index 22270a494..8049538f2 100644 --- a/pkg/sentry/fs/filetest/filetest.go +++ b/pkg/sentry/fs/filetest/filetest.go @@ -19,12 +19,12 @@ import ( "fmt" "testing" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/anon" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/fs.go b/pkg/sentry/fs/fs.go index 26abf49e2..bdba6efe5 100644 --- a/pkg/sentry/fs/fs.go +++ b/pkg/sentry/fs/fs.go @@ -54,8 +54,8 @@ package fs import ( + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sync" ) diff --git a/pkg/sentry/fs/fsutil/BUILD b/pkg/sentry/fs/fsutil/BUILD index 9142f5bdf..4ab2a384f 100644 --- a/pkg/sentry/fs/fsutil/BUILD +++ b/pkg/sentry/fs/fsutil/BUILD @@ -77,22 +77,22 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/log", + "//pkg/safemem", "//pkg/sentry/arch", - "//pkg/sentry/context", "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/kernel/time", "//pkg/sentry/memmap", "//pkg/sentry/pgalloc", "//pkg/sentry/platform", - "//pkg/sentry/safemem", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usage", - "//pkg/sentry/usermem", "//pkg/state", "//pkg/sync", "//pkg/syserror", + "//pkg/usermem", "//pkg/waiter", ], ) @@ -106,13 +106,13 @@ go_test( ], library = ":fsutil", deps = [ - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", + "//pkg/context", + "//pkg/safemem", + "//pkg/sentry/contexttest", "//pkg/sentry/fs", "//pkg/sentry/kernel/time", "//pkg/sentry/memmap", - "//pkg/sentry/safemem", - "//pkg/sentry/usermem", "//pkg/syserror", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/fs/fsutil/dirty_set.go b/pkg/sentry/fs/fsutil/dirty_set.go index 12132680b..c6cd45087 100644 --- a/pkg/sentry/fs/fsutil/dirty_set.go +++ b/pkg/sentry/fs/fsutil/dirty_set.go @@ -17,11 +17,11 @@ package fsutil import ( "math" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/safemem" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // DirtySet maps offsets into a memmap.Mappable to DirtyInfo. It is used to diff --git a/pkg/sentry/fs/fsutil/dirty_set_test.go b/pkg/sentry/fs/fsutil/dirty_set_test.go index 75575d994..e3579c23c 100644 --- a/pkg/sentry/fs/fsutil/dirty_set_test.go +++ b/pkg/sentry/fs/fsutil/dirty_set_test.go @@ -19,7 +19,7 @@ import ( "testing" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) func TestDirtySet(t *testing.T) { diff --git a/pkg/sentry/fs/fsutil/file.go b/pkg/sentry/fs/fsutil/file.go index fc5b3b1a1..08695391c 100644 --- a/pkg/sentry/fs/fsutil/file.go +++ b/pkg/sentry/fs/fsutil/file.go @@ -17,12 +17,12 @@ package fsutil import ( "io" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/fsutil/file_range_set.go b/pkg/sentry/fs/fsutil/file_range_set.go index f52d712e3..5643cdac9 100644 --- a/pkg/sentry/fs/fsutil/file_range_set.go +++ b/pkg/sentry/fs/fsutil/file_range_set.go @@ -19,13 +19,13 @@ import ( "io" "math" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // FileRangeSet maps offsets into a memmap.Mappable to offsets into a diff --git a/pkg/sentry/fs/fsutil/host_file_mapper.go b/pkg/sentry/fs/fsutil/host_file_mapper.go index 837fc70b5..67278aa86 100644 --- a/pkg/sentry/fs/fsutil/host_file_mapper.go +++ b/pkg/sentry/fs/fsutil/host_file_mapper.go @@ -19,11 +19,11 @@ import ( "syscall" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/safemem" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/usermem" ) // HostFileMapper caches mappings of an arbitrary host file descriptor. It is diff --git a/pkg/sentry/fs/fsutil/host_file_mapper_unsafe.go b/pkg/sentry/fs/fsutil/host_file_mapper_unsafe.go index ad11a0573..2d4778d64 100644 --- a/pkg/sentry/fs/fsutil/host_file_mapper_unsafe.go +++ b/pkg/sentry/fs/fsutil/host_file_mapper_unsafe.go @@ -17,7 +17,7 @@ package fsutil import ( "unsafe" - "gvisor.dev/gvisor/pkg/sentry/safemem" + "gvisor.dev/gvisor/pkg/safemem" ) func (*HostFileMapper) unsafeBlockFromChunkMapping(addr uintptr) safemem.Block { diff --git a/pkg/sentry/fs/fsutil/host_mappable.go b/pkg/sentry/fs/fsutil/host_mappable.go index a625f0e26..78fec553e 100644 --- a/pkg/sentry/fs/fsutil/host_mappable.go +++ b/pkg/sentry/fs/fsutil/host_mappable.go @@ -17,13 +17,13 @@ package fsutil import ( "math" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/safemem" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/usermem" ) // HostMappable implements memmap.Mappable and platform.File over a diff --git a/pkg/sentry/fs/fsutil/inode.go b/pkg/sentry/fs/fsutil/inode.go index df7b74855..252830572 100644 --- a/pkg/sentry/fs/fsutil/inode.go +++ b/pkg/sentry/fs/fsutil/inode.go @@ -16,7 +16,7 @@ package fsutil import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/memmap" diff --git a/pkg/sentry/fs/fsutil/inode_cached.go b/pkg/sentry/fs/fsutil/inode_cached.go index 20a014402..573b8586e 100644 --- a/pkg/sentry/fs/fsutil/inode_cached.go +++ b/pkg/sentry/fs/fsutil/inode_cached.go @@ -18,18 +18,18 @@ import ( "fmt" "io" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel/time" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/usermem" ) // Lock order (compare the lock order model in mm/mm.go): diff --git a/pkg/sentry/fs/fsutil/inode_cached_test.go b/pkg/sentry/fs/fsutil/inode_cached_test.go index 129f314c8..1547584c5 100644 --- a/pkg/sentry/fs/fsutil/inode_cached_test.go +++ b/pkg/sentry/fs/fsutil/inode_cached_test.go @@ -19,14 +19,14 @@ import ( "io" "testing" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/safemem" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/fs" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/safemem" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) type noopBackingFile struct{} diff --git a/pkg/sentry/fs/gofer/BUILD b/pkg/sentry/fs/gofer/BUILD index cf48e7c03..971d3718e 100644 --- a/pkg/sentry/fs/gofer/BUILD +++ b/pkg/sentry/fs/gofer/BUILD @@ -24,13 +24,14 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/fd", "//pkg/log", "//pkg/metric", "//pkg/p9", "//pkg/refs", + "//pkg/safemem", "//pkg/secio", - "//pkg/sentry/context", "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fdpipe", @@ -39,13 +40,12 @@ go_library( "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/time", "//pkg/sentry/memmap", - "//pkg/sentry/safemem", "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/usermem", "//pkg/sync", "//pkg/syserr", "//pkg/syserror", "//pkg/unet", + "//pkg/usermem", "//pkg/waiter", ], ) @@ -56,10 +56,10 @@ go_test( srcs = ["gofer_test.go"], library = ":gofer", deps = [ + "//pkg/context", "//pkg/p9", "//pkg/p9/p9test", - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", + "//pkg/sentry/contexttest", "//pkg/sentry/fs", ], ) diff --git a/pkg/sentry/fs/gofer/attr.go b/pkg/sentry/fs/gofer/attr.go index 4848e2374..71cccdc34 100644 --- a/pkg/sentry/fs/gofer/attr.go +++ b/pkg/sentry/fs/gofer/attr.go @@ -17,12 +17,12 @@ package gofer import ( "syscall" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/p9" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // getattr returns the 9p attributes of the p9.File. On success, Mode, Size, and RDev diff --git a/pkg/sentry/fs/gofer/cache_policy.go b/pkg/sentry/fs/gofer/cache_policy.go index cc11c6339..ebea03c42 100644 --- a/pkg/sentry/fs/gofer/cache_policy.go +++ b/pkg/sentry/fs/gofer/cache_policy.go @@ -17,7 +17,7 @@ package gofer import ( "fmt" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" ) diff --git a/pkg/sentry/fs/gofer/context_file.go b/pkg/sentry/fs/gofer/context_file.go index 2125dafef..3da818aed 100644 --- a/pkg/sentry/fs/gofer/context_file.go +++ b/pkg/sentry/fs/gofer/context_file.go @@ -15,9 +15,9 @@ package gofer import ( + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fd" "gvisor.dev/gvisor/pkg/p9" - "gvisor.dev/gvisor/pkg/sentry/context" ) // contextFile is a wrapper around p9.File that notifies the context that diff --git a/pkg/sentry/fs/gofer/file.go b/pkg/sentry/fs/gofer/file.go index 7960b9c7b..23296f246 100644 --- a/pkg/sentry/fs/gofer/file.go +++ b/pkg/sentry/fs/gofer/file.go @@ -19,16 +19,16 @@ import ( "syscall" "time" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/metric" "gvisor.dev/gvisor/pkg/p9" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/device" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/gofer/file_state.go b/pkg/sentry/fs/gofer/file_state.go index bb8312849..ff96b28ba 100644 --- a/pkg/sentry/fs/gofer/file_state.go +++ b/pkg/sentry/fs/gofer/file_state.go @@ -17,7 +17,7 @@ package gofer import ( "fmt" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" ) diff --git a/pkg/sentry/fs/gofer/fs.go b/pkg/sentry/fs/gofer/fs.go index cf96dd9fa..9d41fcbdb 100644 --- a/pkg/sentry/fs/gofer/fs.go +++ b/pkg/sentry/fs/gofer/fs.go @@ -20,8 +20,8 @@ import ( "fmt" "strconv" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/p9" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" ) diff --git a/pkg/sentry/fs/gofer/gofer_test.go b/pkg/sentry/fs/gofer/gofer_test.go index 7fc3c32ae..0c2f89ae8 100644 --- a/pkg/sentry/fs/gofer/gofer_test.go +++ b/pkg/sentry/fs/gofer/gofer_test.go @@ -20,10 +20,10 @@ import ( "testing" "time" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/p9/p9test" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/fs" ) diff --git a/pkg/sentry/fs/gofer/handles.go b/pkg/sentry/fs/gofer/handles.go index b86c49b39..9f7c3e89f 100644 --- a/pkg/sentry/fs/gofer/handles.go +++ b/pkg/sentry/fs/gofer/handles.go @@ -17,14 +17,14 @@ package gofer import ( "io" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fd" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/refs" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/secio" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/safemem" ) // handles are the open handles of a gofer file. They are reference counted to diff --git a/pkg/sentry/fs/gofer/inode.go b/pkg/sentry/fs/gofer/inode.go index 98d1a8a48..ac28174d2 100644 --- a/pkg/sentry/fs/gofer/inode.go +++ b/pkg/sentry/fs/gofer/inode.go @@ -19,17 +19,17 @@ import ( "syscall" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fd" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/p9" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/device" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fdpipe" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/fs/host" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/fs/gofer/inode_state.go b/pkg/sentry/fs/gofer/inode_state.go index 0b2eedb7c..238f7804c 100644 --- a/pkg/sentry/fs/gofer/inode_state.go +++ b/pkg/sentry/fs/gofer/inode_state.go @@ -20,8 +20,8 @@ import ( "path/filepath" "strings" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/p9" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/device" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel/time" diff --git a/pkg/sentry/fs/gofer/path.go b/pkg/sentry/fs/gofer/path.go index c09f3b71c..0c1be05ef 100644 --- a/pkg/sentry/fs/gofer/path.go +++ b/pkg/sentry/fs/gofer/path.go @@ -18,9 +18,9 @@ import ( "fmt" "syscall" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/p9" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/device" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" diff --git a/pkg/sentry/fs/gofer/session.go b/pkg/sentry/fs/gofer/session.go index edc796ce0..498c4645a 100644 --- a/pkg/sentry/fs/gofer/session.go +++ b/pkg/sentry/fs/gofer/session.go @@ -17,9 +17,9 @@ package gofer import ( "fmt" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/device" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" diff --git a/pkg/sentry/fs/gofer/session_state.go b/pkg/sentry/fs/gofer/session_state.go index d045e04ff..0285c5361 100644 --- a/pkg/sentry/fs/gofer/session_state.go +++ b/pkg/sentry/fs/gofer/session_state.go @@ -17,8 +17,8 @@ package gofer import ( "fmt" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/p9" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/unet" ) diff --git a/pkg/sentry/fs/gofer/socket.go b/pkg/sentry/fs/gofer/socket.go index a45a8f36c..376cfce2c 100644 --- a/pkg/sentry/fs/gofer/socket.go +++ b/pkg/sentry/fs/gofer/socket.go @@ -16,9 +16,9 @@ package gofer import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/p9" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/host" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" diff --git a/pkg/sentry/fs/gofer/util.go b/pkg/sentry/fs/gofer/util.go index 848e6812b..2d8d3a2ea 100644 --- a/pkg/sentry/fs/gofer/util.go +++ b/pkg/sentry/fs/gofer/util.go @@ -17,8 +17,8 @@ package gofer import ( "syscall" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/p9" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" ) diff --git a/pkg/sentry/fs/host/BUILD b/pkg/sentry/fs/host/BUILD index f586f47c1..21003ea45 100644 --- a/pkg/sentry/fs/host/BUILD +++ b/pkg/sentry/fs/host/BUILD @@ -27,13 +27,14 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/fd", "//pkg/fdnotifier", "//pkg/log", "//pkg/refs", + "//pkg/safemem", "//pkg/secio", "//pkg/sentry/arch", - "//pkg/sentry/context", "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", @@ -41,18 +42,17 @@ go_library( "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/time", "//pkg/sentry/memmap", - "//pkg/sentry/safemem", "//pkg/sentry/socket/control", "//pkg/sentry/socket/unix", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/unimpl", "//pkg/sentry/uniqueid", - "//pkg/sentry/usermem", "//pkg/sync", "//pkg/syserr", "//pkg/syserror", "//pkg/tcpip", "//pkg/unet", + "//pkg/usermem", "//pkg/waiter", ], ) @@ -69,17 +69,17 @@ go_test( ], library = ":host", deps = [ + "//pkg/context", "//pkg/fd", "//pkg/fdnotifier", - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", + "//pkg/sentry/contexttest", "//pkg/sentry/fs", "//pkg/sentry/kernel/time", "//pkg/sentry/socket", "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/usermem", "//pkg/syserr", "//pkg/tcpip", + "//pkg/usermem", "//pkg/waiter", ], ) diff --git a/pkg/sentry/fs/host/control.go b/pkg/sentry/fs/host/control.go index 5532ff5a0..1658979fc 100644 --- a/pkg/sentry/fs/host/control.go +++ b/pkg/sentry/fs/host/control.go @@ -17,7 +17,7 @@ package host import ( "syscall" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/socket/control" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" diff --git a/pkg/sentry/fs/host/file.go b/pkg/sentry/fs/host/file.go index f6c626f2c..e08f56d04 100644 --- a/pkg/sentry/fs/host/file.go +++ b/pkg/sentry/fs/host/file.go @@ -18,17 +18,17 @@ import ( "fmt" "syscall" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fd" "gvisor.dev/gvisor/pkg/fdnotifier" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/secio" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/safemem" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/host/fs.go b/pkg/sentry/fs/host/fs.go index 68d2697c0..d3e8e3a36 100644 --- a/pkg/sentry/fs/host/fs.go +++ b/pkg/sentry/fs/host/fs.go @@ -23,8 +23,8 @@ import ( "strconv" "strings" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" ) diff --git a/pkg/sentry/fs/host/fs_test.go b/pkg/sentry/fs/host/fs_test.go index c6852ee30..3111d2df9 100644 --- a/pkg/sentry/fs/host/fs_test.go +++ b/pkg/sentry/fs/host/fs_test.go @@ -23,8 +23,8 @@ import ( "sort" "testing" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/fs" ) diff --git a/pkg/sentry/fs/host/inode.go b/pkg/sentry/fs/host/inode.go index 873a1c52d..6fa39caab 100644 --- a/pkg/sentry/fs/host/inode.go +++ b/pkg/sentry/fs/host/inode.go @@ -18,14 +18,14 @@ import ( "syscall" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fd" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/secio" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/device" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" diff --git a/pkg/sentry/fs/host/inode_state.go b/pkg/sentry/fs/host/inode_state.go index b267ec305..299e0e0b0 100644 --- a/pkg/sentry/fs/host/inode_state.go +++ b/pkg/sentry/fs/host/inode_state.go @@ -18,7 +18,7 @@ import ( "fmt" "syscall" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/device" "gvisor.dev/gvisor/pkg/sentry/fs" ) diff --git a/pkg/sentry/fs/host/inode_test.go b/pkg/sentry/fs/host/inode_test.go index 2d959f10d..7221bc825 100644 --- a/pkg/sentry/fs/host/inode_test.go +++ b/pkg/sentry/fs/host/inode_test.go @@ -21,7 +21,7 @@ import ( "syscall" "testing" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/fs" ) diff --git a/pkg/sentry/fs/host/socket.go b/pkg/sentry/fs/host/socket.go index c076d5bdd..06fc2d80a 100644 --- a/pkg/sentry/fs/host/socket.go +++ b/pkg/sentry/fs/host/socket.go @@ -19,11 +19,11 @@ import ( "syscall" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fd" "gvisor.dev/gvisor/pkg/fdnotifier" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/socket/control" unixsocket "gvisor.dev/gvisor/pkg/sentry/socket/unix" diff --git a/pkg/sentry/fs/host/socket_test.go b/pkg/sentry/fs/host/socket_test.go index 68b38fd1c..eb4afe520 100644 --- a/pkg/sentry/fs/host/socket_test.go +++ b/pkg/sentry/fs/host/socket_test.go @@ -21,13 +21,13 @@ import ( "gvisor.dev/gvisor/pkg/fd" "gvisor.dev/gvisor/pkg/fdnotifier" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/sentry/contexttest" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/host/tty.go b/pkg/sentry/fs/host/tty.go index 753ef8cd6..3f218b4a7 100644 --- a/pkg/sentry/fs/host/tty.go +++ b/pkg/sentry/fs/host/tty.go @@ -16,14 +16,14 @@ package host import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/unimpl" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // TTYFileOperations implements fs.FileOperations for a host file descriptor diff --git a/pkg/sentry/fs/host/wait_test.go b/pkg/sentry/fs/host/wait_test.go index 88d24d693..d49c3a635 100644 --- a/pkg/sentry/fs/host/wait_test.go +++ b/pkg/sentry/fs/host/wait_test.go @@ -19,7 +19,7 @@ import ( "testing" "time" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/inode.go b/pkg/sentry/fs/inode.go index e4cf5a570..b66c091ab 100644 --- a/pkg/sentry/fs/inode.go +++ b/pkg/sentry/fs/inode.go @@ -16,10 +16,10 @@ package fs import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/metric" "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/memmap" diff --git a/pkg/sentry/fs/inode_operations.go b/pkg/sentry/fs/inode_operations.go index 13261cb81..70f2eae96 100644 --- a/pkg/sentry/fs/inode_operations.go +++ b/pkg/sentry/fs/inode_operations.go @@ -17,7 +17,7 @@ package fs import ( "errors" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" diff --git a/pkg/sentry/fs/inode_overlay.go b/pkg/sentry/fs/inode_overlay.go index c477de837..4729b4aac 100644 --- a/pkg/sentry/fs/inode_overlay.go +++ b/pkg/sentry/fs/inode_overlay.go @@ -19,8 +19,8 @@ import ( "strings" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/fs/inode_overlay_test.go b/pkg/sentry/fs/inode_overlay_test.go index 493d98c36..389c219d6 100644 --- a/pkg/sentry/fs/inode_overlay_test.go +++ b/pkg/sentry/fs/inode_overlay_test.go @@ -17,7 +17,7 @@ package fs_test import ( "testing" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" diff --git a/pkg/sentry/fs/inotify.go b/pkg/sentry/fs/inotify.go index cc7dd1c92..928c90aa0 100644 --- a/pkg/sentry/fs/inotify.go +++ b/pkg/sentry/fs/inotify.go @@ -19,13 +19,13 @@ import ( "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/uniqueid" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/inotify_event.go b/pkg/sentry/fs/inotify_event.go index 9f70a3e82..686e1b1cd 100644 --- a/pkg/sentry/fs/inotify_event.go +++ b/pkg/sentry/fs/inotify_event.go @@ -18,8 +18,8 @@ import ( "bytes" "fmt" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/usermem" ) // inotifyEventBaseSize is the base size of linux's struct inotify_event. This diff --git a/pkg/sentry/fs/mock.go b/pkg/sentry/fs/mock.go index 7a24c6f1b..1d6ea5736 100644 --- a/pkg/sentry/fs/mock.go +++ b/pkg/sentry/fs/mock.go @@ -15,7 +15,7 @@ package fs import ( - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/fs/mount.go b/pkg/sentry/fs/mount.go index 7a9692800..37bae6810 100644 --- a/pkg/sentry/fs/mount.go +++ b/pkg/sentry/fs/mount.go @@ -19,8 +19,8 @@ import ( "fmt" "sync/atomic" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/context" ) // DirentOperations provide file systems greater control over how long a Dirent diff --git a/pkg/sentry/fs/mount_overlay.go b/pkg/sentry/fs/mount_overlay.go index 299712cd7..78e35b1e6 100644 --- a/pkg/sentry/fs/mount_overlay.go +++ b/pkg/sentry/fs/mount_overlay.go @@ -15,7 +15,7 @@ package fs import ( - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" ) // overlayMountSourceOperations implements MountSourceOperations for an overlay diff --git a/pkg/sentry/fs/mount_test.go b/pkg/sentry/fs/mount_test.go index 0b84732aa..e672a438c 100644 --- a/pkg/sentry/fs/mount_test.go +++ b/pkg/sentry/fs/mount_test.go @@ -18,7 +18,7 @@ import ( "fmt" "testing" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/sentry/contexttest" ) // cacheReallyContains iterates through the dirent cache to determine whether diff --git a/pkg/sentry/fs/mounts.go b/pkg/sentry/fs/mounts.go index a9627a9d1..574a2cc91 100644 --- a/pkg/sentry/fs/mounts.go +++ b/pkg/sentry/fs/mounts.go @@ -22,9 +22,9 @@ import ( "syscall" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" diff --git a/pkg/sentry/fs/mounts_test.go b/pkg/sentry/fs/mounts_test.go index c4c771f2c..a69b41468 100644 --- a/pkg/sentry/fs/mounts_test.go +++ b/pkg/sentry/fs/mounts_test.go @@ -17,7 +17,7 @@ package fs_test import ( "testing" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" diff --git a/pkg/sentry/fs/offset.go b/pkg/sentry/fs/offset.go index f7d844ce7..53b5df175 100644 --- a/pkg/sentry/fs/offset.go +++ b/pkg/sentry/fs/offset.go @@ -17,7 +17,7 @@ package fs import ( "math" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // OffsetPageEnd returns the file offset rounded up to the nearest diff --git a/pkg/sentry/fs/overlay.go b/pkg/sentry/fs/overlay.go index f7702f8f4..a8ae7d81d 100644 --- a/pkg/sentry/fs/overlay.go +++ b/pkg/sentry/fs/overlay.go @@ -18,12 +18,12 @@ import ( "fmt" "strings" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // The virtual filesystem implements an overlay configuration. For a high-level diff --git a/pkg/sentry/fs/proc/BUILD b/pkg/sentry/fs/proc/BUILD index b06bead41..280093c5e 100644 --- a/pkg/sentry/fs/proc/BUILD +++ b/pkg/sentry/fs/proc/BUILD @@ -29,8 +29,8 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/log", - "//pkg/sentry/context", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", "//pkg/sentry/fs/proc/device", @@ -46,10 +46,10 @@ go_library( "//pkg/sentry/socket/unix", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usage", - "//pkg/sentry/usermem", "//pkg/sync", "//pkg/syserror", "//pkg/tcpip/header", + "//pkg/usermem", "//pkg/waiter", ], ) @@ -64,8 +64,8 @@ go_test( library = ":proc", deps = [ "//pkg/abi/linux", - "//pkg/sentry/context", + "//pkg/context", "//pkg/sentry/inet", - "//pkg/sentry/usermem", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/fs/proc/cgroup.go b/pkg/sentry/fs/proc/cgroup.go index c4abe319d..7c1d9e7e9 100644 --- a/pkg/sentry/fs/proc/cgroup.go +++ b/pkg/sentry/fs/proc/cgroup.go @@ -17,7 +17,7 @@ package proc import ( "fmt" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" ) diff --git a/pkg/sentry/fs/proc/cpuinfo.go b/pkg/sentry/fs/proc/cpuinfo.go index df0c4e3a7..c96533401 100644 --- a/pkg/sentry/fs/proc/cpuinfo.go +++ b/pkg/sentry/fs/proc/cpuinfo.go @@ -17,7 +17,7 @@ package proc import ( "bytes" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" ) diff --git a/pkg/sentry/fs/proc/exec_args.go b/pkg/sentry/fs/proc/exec_args.go index 9aaeb780b..8fe626e1c 100644 --- a/pkg/sentry/fs/proc/exec_args.go +++ b/pkg/sentry/fs/proc/exec_args.go @@ -20,12 +20,12 @@ import ( "io" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/proc/fds.go b/pkg/sentry/fs/proc/fds.go index 2fa3cfa7d..35972e23c 100644 --- a/pkg/sentry/fs/proc/fds.go +++ b/pkg/sentry/fs/proc/fds.go @@ -19,7 +19,7 @@ import ( "sort" "strconv" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/fs/proc/device" diff --git a/pkg/sentry/fs/proc/filesystems.go b/pkg/sentry/fs/proc/filesystems.go index 7b3b974ab..0a58ac34c 100644 --- a/pkg/sentry/fs/proc/filesystems.go +++ b/pkg/sentry/fs/proc/filesystems.go @@ -18,7 +18,7 @@ import ( "bytes" "fmt" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/proc/seqfile" ) diff --git a/pkg/sentry/fs/proc/fs.go b/pkg/sentry/fs/proc/fs.go index 761d24462..daf1ba781 100644 --- a/pkg/sentry/fs/proc/fs.go +++ b/pkg/sentry/fs/proc/fs.go @@ -17,7 +17,7 @@ package proc import ( "fmt" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" ) diff --git a/pkg/sentry/fs/proc/inode.go b/pkg/sentry/fs/proc/inode.go index 723f6b661..d2859a4c2 100644 --- a/pkg/sentry/fs/proc/inode.go +++ b/pkg/sentry/fs/proc/inode.go @@ -16,14 +16,14 @@ package proc import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/fs/proc/device" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/mm" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // LINT.IfChange diff --git a/pkg/sentry/fs/proc/loadavg.go b/pkg/sentry/fs/proc/loadavg.go index d7d2afcb7..139d49c34 100644 --- a/pkg/sentry/fs/proc/loadavg.go +++ b/pkg/sentry/fs/proc/loadavg.go @@ -18,7 +18,7 @@ import ( "bytes" "fmt" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs/proc/seqfile" ) diff --git a/pkg/sentry/fs/proc/meminfo.go b/pkg/sentry/fs/proc/meminfo.go index 313c6a32b..465b47da9 100644 --- a/pkg/sentry/fs/proc/meminfo.go +++ b/pkg/sentry/fs/proc/meminfo.go @@ -18,11 +18,11 @@ import ( "bytes" "fmt" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs/proc/seqfile" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // LINT.IfChange diff --git a/pkg/sentry/fs/proc/mounts.go b/pkg/sentry/fs/proc/mounts.go index d4efc86e0..c10888100 100644 --- a/pkg/sentry/fs/proc/mounts.go +++ b/pkg/sentry/fs/proc/mounts.go @@ -20,7 +20,7 @@ import ( "sort" "strings" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/proc/seqfile" "gvisor.dev/gvisor/pkg/sentry/kernel" diff --git a/pkg/sentry/fs/proc/net.go b/pkg/sentry/fs/proc/net.go index bad445f3f..6f2775344 100644 --- a/pkg/sentry/fs/proc/net.go +++ b/pkg/sentry/fs/proc/net.go @@ -22,8 +22,8 @@ import ( "time" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/proc/seqfile" "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" @@ -33,9 +33,9 @@ import ( "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/unix" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/usermem" ) // LINT.IfChange diff --git a/pkg/sentry/fs/proc/proc.go b/pkg/sentry/fs/proc/proc.go index 29867dc3a..c8abb5052 100644 --- a/pkg/sentry/fs/proc/proc.go +++ b/pkg/sentry/fs/proc/proc.go @@ -20,7 +20,7 @@ import ( "sort" "strconv" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/fs/proc/device" diff --git a/pkg/sentry/fs/proc/seqfile/BUILD b/pkg/sentry/fs/proc/seqfile/BUILD index 310d8dd52..21338d912 100644 --- a/pkg/sentry/fs/proc/seqfile/BUILD +++ b/pkg/sentry/fs/proc/seqfile/BUILD @@ -8,14 +8,14 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", - "//pkg/sentry/context", + "//pkg/context", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", "//pkg/sentry/fs/proc/device", "//pkg/sentry/kernel/time", - "//pkg/sentry/usermem", "//pkg/sync", "//pkg/syserror", + "//pkg/usermem", "//pkg/waiter", ], ) @@ -26,10 +26,10 @@ go_test( srcs = ["seqfile_test.go"], library = ":seqfile", deps = [ - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", + "//pkg/context", + "//pkg/sentry/contexttest", "//pkg/sentry/fs", "//pkg/sentry/fs/ramfs", - "//pkg/sentry/usermem", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/fs/proc/seqfile/seqfile.go b/pkg/sentry/fs/proc/seqfile/seqfile.go index f9af191d5..6121f0e95 100644 --- a/pkg/sentry/fs/proc/seqfile/seqfile.go +++ b/pkg/sentry/fs/proc/seqfile/seqfile.go @@ -19,14 +19,14 @@ import ( "io" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/fs/proc/device" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/proc/seqfile/seqfile_test.go b/pkg/sentry/fs/proc/seqfile/seqfile_test.go index ebfeee835..98e394569 100644 --- a/pkg/sentry/fs/proc/seqfile/seqfile_test.go +++ b/pkg/sentry/fs/proc/seqfile/seqfile_test.go @@ -20,11 +20,11 @@ import ( "io" "testing" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) type seqTest struct { diff --git a/pkg/sentry/fs/proc/stat.go b/pkg/sentry/fs/proc/stat.go index bc5b2bc7b..d4fbd76ac 100644 --- a/pkg/sentry/fs/proc/stat.go +++ b/pkg/sentry/fs/proc/stat.go @@ -19,7 +19,7 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs/proc/seqfile" "gvisor.dev/gvisor/pkg/sentry/kernel" ) diff --git a/pkg/sentry/fs/proc/sys.go b/pkg/sentry/fs/proc/sys.go index 2bdcf5f70..f8aad2dbd 100644 --- a/pkg/sentry/fs/proc/sys.go +++ b/pkg/sentry/fs/proc/sys.go @@ -20,13 +20,13 @@ import ( "strconv" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/fs/proc/seqfile" "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/proc/sys_net.go b/pkg/sentry/fs/proc/sys_net.go index b9e8ef35f..0772d4ae4 100644 --- a/pkg/sentry/fs/proc/sys_net.go +++ b/pkg/sentry/fs/proc/sys_net.go @@ -19,14 +19,14 @@ import ( "io" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/fs/proc/device" "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" "gvisor.dev/gvisor/pkg/sentry/inet" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/proc/sys_net_test.go b/pkg/sentry/fs/proc/sys_net_test.go index 6abae7a60..355e83d47 100644 --- a/pkg/sentry/fs/proc/sys_net_test.go +++ b/pkg/sentry/fs/proc/sys_net_test.go @@ -17,9 +17,9 @@ package proc import ( "testing" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/inet" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) func TestQuerySendBufferSize(t *testing.T) { diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go index 7358d6ef9..ca020e11e 100644 --- a/pkg/sentry/fs/proc/task.go +++ b/pkg/sentry/fs/proc/task.go @@ -22,7 +22,7 @@ import ( "strconv" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/fs/proc/device" @@ -32,8 +32,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/mm" "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/proc/uid_gid_map.go b/pkg/sentry/fs/proc/uid_gid_map.go index 3eacc9265..8d9517b95 100644 --- a/pkg/sentry/fs/proc/uid_gid_map.go +++ b/pkg/sentry/fs/proc/uid_gid_map.go @@ -20,13 +20,13 @@ import ( "io" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/proc/uptime.go b/pkg/sentry/fs/proc/uptime.go index adfe58adb..c0f6fb802 100644 --- a/pkg/sentry/fs/proc/uptime.go +++ b/pkg/sentry/fs/proc/uptime.go @@ -19,12 +19,12 @@ import ( "io" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/proc/version.go b/pkg/sentry/fs/proc/version.go index 27fd5b1cb..35e258ff6 100644 --- a/pkg/sentry/fs/proc/version.go +++ b/pkg/sentry/fs/proc/version.go @@ -17,7 +17,7 @@ package proc import ( "fmt" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs/proc/seqfile" "gvisor.dev/gvisor/pkg/sentry/kernel" ) diff --git a/pkg/sentry/fs/ramfs/BUILD b/pkg/sentry/fs/ramfs/BUILD index 39c4b84f8..8ca823fb3 100644 --- a/pkg/sentry/fs/ramfs/BUILD +++ b/pkg/sentry/fs/ramfs/BUILD @@ -13,14 +13,14 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", - "//pkg/sentry/context", + "//pkg/context", "//pkg/sentry/fs", "//pkg/sentry/fs/anon", "//pkg/sentry/fs/fsutil", "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/usermem", "//pkg/sync", "//pkg/syserror", + "//pkg/usermem", "//pkg/waiter", ], ) @@ -31,7 +31,7 @@ go_test( srcs = ["tree_test.go"], library = ":ramfs", deps = [ - "//pkg/sentry/context/contexttest", + "//pkg/sentry/contexttest", "//pkg/sentry/fs", ], ) diff --git a/pkg/sentry/fs/ramfs/dir.go b/pkg/sentry/fs/ramfs/dir.go index dcbb8eb2e..bfa304552 100644 --- a/pkg/sentry/fs/ramfs/dir.go +++ b/pkg/sentry/fs/ramfs/dir.go @@ -20,7 +20,7 @@ import ( "syscall" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" diff --git a/pkg/sentry/fs/ramfs/socket.go b/pkg/sentry/fs/ramfs/socket.go index a24fe2ea2..29ff004f2 100644 --- a/pkg/sentry/fs/ramfs/socket.go +++ b/pkg/sentry/fs/ramfs/socket.go @@ -16,7 +16,7 @@ package ramfs import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" diff --git a/pkg/sentry/fs/ramfs/symlink.go b/pkg/sentry/fs/ramfs/symlink.go index fcfaa29aa..d988349aa 100644 --- a/pkg/sentry/fs/ramfs/symlink.go +++ b/pkg/sentry/fs/ramfs/symlink.go @@ -16,7 +16,7 @@ package ramfs import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/waiter" diff --git a/pkg/sentry/fs/ramfs/tree.go b/pkg/sentry/fs/ramfs/tree.go index 702cc4a1e..dfc9d3453 100644 --- a/pkg/sentry/fs/ramfs/tree.go +++ b/pkg/sentry/fs/ramfs/tree.go @@ -19,10 +19,10 @@ import ( "path" "strings" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/anon" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // MakeDirectoryTree constructs a ramfs tree of all directories containing diff --git a/pkg/sentry/fs/ramfs/tree_test.go b/pkg/sentry/fs/ramfs/tree_test.go index 61a7e2900..a6ed8b2c5 100644 --- a/pkg/sentry/fs/ramfs/tree_test.go +++ b/pkg/sentry/fs/ramfs/tree_test.go @@ -17,7 +17,7 @@ package ramfs import ( "testing" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/fs" ) diff --git a/pkg/sentry/fs/splice.go b/pkg/sentry/fs/splice.go index 389c330a0..791d1526c 100644 --- a/pkg/sentry/fs/splice.go +++ b/pkg/sentry/fs/splice.go @@ -18,7 +18,7 @@ import ( "io" "sync/atomic" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/fs/sys/BUILD b/pkg/sentry/fs/sys/BUILD index cc6b3bfbf..f2e8b9932 100644 --- a/pkg/sentry/fs/sys/BUILD +++ b/pkg/sentry/fs/sys/BUILD @@ -13,12 +13,12 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", - "//pkg/sentry/context", + "//pkg/context", "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", "//pkg/sentry/fs/ramfs", "//pkg/sentry/kernel", - "//pkg/sentry/usermem", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/fs/sys/devices.go b/pkg/sentry/fs/sys/devices.go index 4f78ca8d2..b67065956 100644 --- a/pkg/sentry/fs/sys/devices.go +++ b/pkg/sentry/fs/sys/devices.go @@ -18,7 +18,7 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/kernel" diff --git a/pkg/sentry/fs/sys/fs.go b/pkg/sentry/fs/sys/fs.go index e60b63e75..fd03a4e38 100644 --- a/pkg/sentry/fs/sys/fs.go +++ b/pkg/sentry/fs/sys/fs.go @@ -15,7 +15,7 @@ package sys import ( - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" ) diff --git a/pkg/sentry/fs/sys/sys.go b/pkg/sentry/fs/sys/sys.go index b14bf3f55..0891645e4 100644 --- a/pkg/sentry/fs/sys/sys.go +++ b/pkg/sentry/fs/sys/sys.go @@ -16,10 +16,10 @@ package sys import ( - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) func newFile(ctx context.Context, node fs.InodeOperations, msrc *fs.MountSource) *fs.Inode { diff --git a/pkg/sentry/fs/timerfd/BUILD b/pkg/sentry/fs/timerfd/BUILD index 092668e8d..d16cdb4df 100644 --- a/pkg/sentry/fs/timerfd/BUILD +++ b/pkg/sentry/fs/timerfd/BUILD @@ -7,13 +7,13 @@ go_library( srcs = ["timerfd.go"], visibility = ["//pkg/sentry:internal"], deps = [ - "//pkg/sentry/context", + "//pkg/context", "//pkg/sentry/fs", "//pkg/sentry/fs/anon", "//pkg/sentry/fs/fsutil", "//pkg/sentry/kernel/time", - "//pkg/sentry/usermem", "//pkg/syserror", + "//pkg/usermem", "//pkg/waiter", ], ) diff --git a/pkg/sentry/fs/timerfd/timerfd.go b/pkg/sentry/fs/timerfd/timerfd.go index f8bf663bb..88c344089 100644 --- a/pkg/sentry/fs/timerfd/timerfd.go +++ b/pkg/sentry/fs/timerfd/timerfd.go @@ -19,13 +19,13 @@ package timerfd import ( "sync/atomic" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/anon" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/tmpfs/BUILD b/pkg/sentry/fs/tmpfs/BUILD index 04776555f..aa7199014 100644 --- a/pkg/sentry/fs/tmpfs/BUILD +++ b/pkg/sentry/fs/tmpfs/BUILD @@ -14,8 +14,9 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/metric", - "//pkg/sentry/context", + "//pkg/safemem", "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", @@ -25,12 +26,11 @@ go_library( "//pkg/sentry/kernel/pipe", "//pkg/sentry/kernel/time", "//pkg/sentry/memmap", - "//pkg/sentry/safemem", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usage", - "//pkg/sentry/usermem", "//pkg/sync", "//pkg/syserror", + "//pkg/usermem", "//pkg/waiter", ], ) @@ -41,10 +41,10 @@ go_test( srcs = ["file_test.go"], library = ":tmpfs", deps = [ - "//pkg/sentry/context", + "//pkg/context", "//pkg/sentry/fs", "//pkg/sentry/kernel/contexttest", "//pkg/sentry/usage", - "//pkg/sentry/usermem", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/fs/tmpfs/file_regular.go b/pkg/sentry/fs/tmpfs/file_regular.go index 9a6943fe4..614f8f8a1 100644 --- a/pkg/sentry/fs/tmpfs/file_regular.go +++ b/pkg/sentry/fs/tmpfs/file_regular.go @@ -15,11 +15,11 @@ package tmpfs import ( - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/tmpfs/file_test.go b/pkg/sentry/fs/tmpfs/file_test.go index 0075ef023..aaba35502 100644 --- a/pkg/sentry/fs/tmpfs/file_test.go +++ b/pkg/sentry/fs/tmpfs/file_test.go @@ -18,11 +18,11 @@ import ( "bytes" "testing" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) func newFileInode(ctx context.Context) *fs.Inode { diff --git a/pkg/sentry/fs/tmpfs/fs.go b/pkg/sentry/fs/tmpfs/fs.go index be98ad751..d5be56c3f 100644 --- a/pkg/sentry/fs/tmpfs/fs.go +++ b/pkg/sentry/fs/tmpfs/fs.go @@ -19,7 +19,7 @@ import ( "strconv" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ) diff --git a/pkg/sentry/fs/tmpfs/inode_file.go b/pkg/sentry/fs/tmpfs/inode_file.go index f1c87fe41..dabc10662 100644 --- a/pkg/sentry/fs/tmpfs/inode_file.go +++ b/pkg/sentry/fs/tmpfs/inode_file.go @@ -20,18 +20,18 @@ import ( "time" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/metric" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/kernel" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) var ( diff --git a/pkg/sentry/fs/tmpfs/tmpfs.go b/pkg/sentry/fs/tmpfs/tmpfs.go index 0f718e236..c00cef0a5 100644 --- a/pkg/sentry/fs/tmpfs/tmpfs.go +++ b/pkg/sentry/fs/tmpfs/tmpfs.go @@ -17,7 +17,7 @@ package tmpfs import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" @@ -25,8 +25,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) var fsInfo = fs.Info{ diff --git a/pkg/sentry/fs/tty/BUILD b/pkg/sentry/fs/tty/BUILD index 29f804c6c..5cb0e0417 100644 --- a/pkg/sentry/fs/tty/BUILD +++ b/pkg/sentry/fs/tty/BUILD @@ -16,20 +16,20 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/refs", + "//pkg/safemem", "//pkg/sentry/arch", - "//pkg/sentry/context", "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", - "//pkg/sentry/safemem", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/unimpl", - "//pkg/sentry/usermem", "//pkg/sync", "//pkg/syserror", + "//pkg/usermem", "//pkg/waiter", ], ) @@ -41,7 +41,7 @@ go_test( library = ":tty", deps = [ "//pkg/abi/linux", - "//pkg/sentry/context/contexttest", - "//pkg/sentry/usermem", + "//pkg/sentry/contexttest", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/fs/tty/dir.go b/pkg/sentry/fs/tty/dir.go index 88aa66b24..108654827 100644 --- a/pkg/sentry/fs/tty/dir.go +++ b/pkg/sentry/fs/tty/dir.go @@ -21,14 +21,14 @@ import ( "strconv" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/tty/fs.go b/pkg/sentry/fs/tty/fs.go index edee56c12..8fe05ebe5 100644 --- a/pkg/sentry/fs/tty/fs.go +++ b/pkg/sentry/fs/tty/fs.go @@ -15,7 +15,7 @@ package tty import ( - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/device" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/syserror" diff --git a/pkg/sentry/fs/tty/line_discipline.go b/pkg/sentry/fs/tty/line_discipline.go index 9fe02657e..12b1c6097 100644 --- a/pkg/sentry/fs/tty/line_discipline.go +++ b/pkg/sentry/fs/tty/line_discipline.go @@ -19,11 +19,11 @@ import ( "unicode/utf8" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/tty/master.go b/pkg/sentry/fs/tty/master.go index 6b07f6bf2..f62da49bd 100644 --- a/pkg/sentry/fs/tty/master.go +++ b/pkg/sentry/fs/tty/master.go @@ -16,13 +16,13 @@ package tty import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/unimpl" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/tty/queue.go b/pkg/sentry/fs/tty/queue.go index 21ccc6f32..1ca79c0b2 100644 --- a/pkg/sentry/fs/tty/queue.go +++ b/pkg/sentry/fs/tty/queue.go @@ -16,12 +16,12 @@ package tty import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/safemem" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/tty/slave.go b/pkg/sentry/fs/tty/slave.go index 2a51e6bab..db55cdc48 100644 --- a/pkg/sentry/fs/tty/slave.go +++ b/pkg/sentry/fs/tty/slave.go @@ -16,12 +16,12 @@ package tty import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/fs/tty/terminal.go b/pkg/sentry/fs/tty/terminal.go index 917f90cc0..5883f26db 100644 --- a/pkg/sentry/fs/tty/terminal.go +++ b/pkg/sentry/fs/tty/terminal.go @@ -16,11 +16,11 @@ package tty import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // Terminal is a pseudoterminal. diff --git a/pkg/sentry/fs/tty/tty_test.go b/pkg/sentry/fs/tty/tty_test.go index 59f07ff8e..2cbc05678 100644 --- a/pkg/sentry/fs/tty/tty_test.go +++ b/pkg/sentry/fs/tty/tty_test.go @@ -18,8 +18,8 @@ import ( "testing" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sentry/contexttest" + "gvisor.dev/gvisor/pkg/usermem" ) func TestSimpleMasterToSlave(t *testing.T) { diff --git a/pkg/sentry/fsimpl/ext/BUILD b/pkg/sentry/fsimpl/ext/BUILD index a718920d5..6f78f478f 100644 --- a/pkg/sentry/fsimpl/ext/BUILD +++ b/pkg/sentry/fsimpl/ext/BUILD @@ -35,21 +35,21 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/binary", + "//pkg/context", "//pkg/fd", "//pkg/fspath", "//pkg/log", + "//pkg/safemem", "//pkg/sentry/arch", - "//pkg/sentry/context", "//pkg/sentry/fs", "//pkg/sentry/fsimpl/ext/disklayout", "//pkg/sentry/kernel/auth", "//pkg/sentry/memmap", - "//pkg/sentry/safemem", "//pkg/sentry/syscalls/linux", - "//pkg/sentry/usermem", "//pkg/sentry/vfs", "//pkg/sync", "//pkg/syserror", + "//pkg/usermem", "//pkg/waiter", ], ) @@ -73,14 +73,14 @@ go_test( deps = [ "//pkg/abi/linux", "//pkg/binary", + "//pkg/context", "//pkg/fspath", - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", + "//pkg/sentry/contexttest", "//pkg/sentry/fsimpl/ext/disklayout", "//pkg/sentry/kernel/auth", - "//pkg/sentry/usermem", "//pkg/sentry/vfs", "//pkg/syserror", + "//pkg/usermem", "//runsc/testutil", "@com_github_google_go-cmp//cmp:go_default_library", "@com_github_google_go-cmp//cmp/cmpopts:go_default_library", diff --git a/pkg/sentry/fsimpl/ext/benchmark/BUILD b/pkg/sentry/fsimpl/ext/benchmark/BUILD index 12f3990c1..6c5a559fd 100644 --- a/pkg/sentry/fsimpl/ext/benchmark/BUILD +++ b/pkg/sentry/fsimpl/ext/benchmark/BUILD @@ -7,9 +7,9 @@ go_test( size = "small", srcs = ["benchmark_test.go"], deps = [ + "//pkg/context", "//pkg/fspath", - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", + "//pkg/sentry/contexttest", "//pkg/sentry/fsimpl/ext", "//pkg/sentry/kernel/auth", "//pkg/sentry/vfs", diff --git a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go index a56b03711..d1436b943 100644 --- a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go +++ b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go @@ -24,9 +24,9 @@ import ( "strings" "testing" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" diff --git a/pkg/sentry/fsimpl/ext/directory.go b/pkg/sentry/fsimpl/ext/directory.go index 8944171c8..ebb72b75e 100644 --- a/pkg/sentry/fsimpl/ext/directory.go +++ b/pkg/sentry/fsimpl/ext/directory.go @@ -17,8 +17,8 @@ package ext import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" "gvisor.dev/gvisor/pkg/sentry/memmap" diff --git a/pkg/sentry/fsimpl/ext/ext.go b/pkg/sentry/fsimpl/ext/ext.go index 4b7d17dc6..373d23b74 100644 --- a/pkg/sentry/fsimpl/ext/ext.go +++ b/pkg/sentry/fsimpl/ext/ext.go @@ -21,9 +21,9 @@ import ( "io" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fd" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" diff --git a/pkg/sentry/fsimpl/ext/ext_test.go b/pkg/sentry/fsimpl/ext/ext_test.go index 6c14a1e2d..05f992826 100644 --- a/pkg/sentry/fsimpl/ext/ext_test.go +++ b/pkg/sentry/fsimpl/ext/ext_test.go @@ -25,14 +25,14 @@ import ( "github.com/google/go-cmp/cmp" "github.com/google/go-cmp/cmp/cmpopts" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/runsc/testutil" ) diff --git a/pkg/sentry/fsimpl/ext/file_description.go b/pkg/sentry/fsimpl/ext/file_description.go index 841274daf..92f7da40d 100644 --- a/pkg/sentry/fsimpl/ext/file_description.go +++ b/pkg/sentry/fsimpl/ext/file_description.go @@ -16,7 +16,7 @@ package ext import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/fsimpl/ext/filesystem.go b/pkg/sentry/fsimpl/ext/filesystem.go index 9afb1a84c..07bf58953 100644 --- a/pkg/sentry/fsimpl/ext/filesystem.go +++ b/pkg/sentry/fsimpl/ext/filesystem.go @@ -19,8 +19,8 @@ import ( "io" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" diff --git a/pkg/sentry/fsimpl/ext/regular_file.go b/pkg/sentry/fsimpl/ext/regular_file.go index d11153c90..30135ddb0 100644 --- a/pkg/sentry/fsimpl/ext/regular_file.go +++ b/pkg/sentry/fsimpl/ext/regular_file.go @@ -18,13 +18,13 @@ import ( "io" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/safemem" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // regularFile represents a regular file's inode. This too follows the diff --git a/pkg/sentry/fsimpl/ext/symlink.go b/pkg/sentry/fsimpl/ext/symlink.go index bdf8705c1..1447a4dc1 100644 --- a/pkg/sentry/fsimpl/ext/symlink.go +++ b/pkg/sentry/fsimpl/ext/symlink.go @@ -15,11 +15,11 @@ package ext import ( - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // symlink represents a symlink inode. diff --git a/pkg/sentry/fsimpl/kernfs/BUILD b/pkg/sentry/fsimpl/kernfs/BUILD index 7bf83ccba..e73f1f857 100644 --- a/pkg/sentry/fsimpl/kernfs/BUILD +++ b/pkg/sentry/fsimpl/kernfs/BUILD @@ -29,16 +29,16 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/fspath", "//pkg/log", "//pkg/refs", - "//pkg/sentry/context", "//pkg/sentry/kernel/auth", "//pkg/sentry/memmap", - "//pkg/sentry/usermem", "//pkg/sentry/vfs", "//pkg/sync", "//pkg/syserror", + "//pkg/usermem", ], ) @@ -49,13 +49,13 @@ go_test( deps = [ ":kernfs", "//pkg/abi/linux", - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", + "//pkg/context", + "//pkg/sentry/contexttest", "//pkg/sentry/fsimpl/testutil", "//pkg/sentry/kernel/auth", - "//pkg/sentry/usermem", "//pkg/sentry/vfs", "//pkg/syserror", + "//pkg/usermem", "@com_github_google_go-cmp//cmp:go_default_library", ], ) diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go index 75624e0b1..373f801ff 100644 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -18,11 +18,11 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // DynamicBytesFile implements kernfs.Inode and represents a read-only diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index 5fa1fa67b..6104751c8 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -16,11 +16,11 @@ package kernfs import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // GenericDirectoryFD implements vfs.FileDescriptionImpl for a generic directory diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index a4600ad47..9d65d0179 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -20,8 +20,8 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 1700fffd9..adca2313f 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -19,8 +19,8 @@ import ( "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index 85bcdcc57..79ebea8a5 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -56,8 +56,8 @@ import ( "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index fade59491..ee65cf491 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -21,14 +21,14 @@ import ( "github.com/google/go-cmp/cmp" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/testutil" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) const defaultMode linux.FileMode = 01777 diff --git a/pkg/sentry/fsimpl/kernfs/symlink.go b/pkg/sentry/fsimpl/kernfs/symlink.go index f19f12854..0ee7eb9b7 100644 --- a/pkg/sentry/fsimpl/kernfs/symlink.go +++ b/pkg/sentry/fsimpl/kernfs/symlink.go @@ -16,7 +16,7 @@ package kernfs import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ) diff --git a/pkg/sentry/fsimpl/proc/BUILD b/pkg/sentry/fsimpl/proc/BUILD index 3768f55b2..12aac2e6a 100644 --- a/pkg/sentry/fsimpl/proc/BUILD +++ b/pkg/sentry/fsimpl/proc/BUILD @@ -16,8 +16,9 @@ go_library( ], deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/log", - "//pkg/sentry/context", + "//pkg/safemem", "//pkg/sentry/fs", "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/inet", @@ -26,15 +27,14 @@ go_library( "//pkg/sentry/kernel/time", "//pkg/sentry/limits", "//pkg/sentry/mm", - "//pkg/sentry/safemem", "//pkg/sentry/socket", "//pkg/sentry/socket/unix", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usage", - "//pkg/sentry/usermem", "//pkg/sentry/vfs", "//pkg/syserror", "//pkg/tcpip/header", + "//pkg/usermem", ], ) @@ -48,15 +48,15 @@ go_test( library = ":proc", deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/fspath", - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", + "//pkg/sentry/contexttest", "//pkg/sentry/fsimpl/testutil", "//pkg/sentry/inet", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", - "//pkg/sentry/usermem", "//pkg/sentry/vfs", "//pkg/syserror", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/fsimpl/proc/filesystem.go b/pkg/sentry/fsimpl/proc/filesystem.go index f49819187..11477b6a9 100644 --- a/pkg/sentry/fsimpl/proc/filesystem.go +++ b/pkg/sentry/fsimpl/proc/filesystem.go @@ -19,7 +19,7 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go index 91eded415..353e37195 100644 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -19,7 +19,7 @@ import ( "strconv" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/vfs" diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index a0580f20d..eb5bc62c0 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -19,7 +19,7 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index 7bc352ae9..efd3b3453 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -20,17 +20,17 @@ import ( "io" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/mm" - "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // mm gets the kernel task's MemoryManager. No additional reference is taken on diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index 51f634716..e0cb9c47b 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -20,7 +20,7 @@ import ( "strconv" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" diff --git a/pkg/sentry/fsimpl/proc/tasks_files.go b/pkg/sentry/fsimpl/proc/tasks_files.go index ad3760e39..434998910 100644 --- a/pkg/sentry/fsimpl/proc/tasks_files.go +++ b/pkg/sentry/fsimpl/proc/tasks_files.go @@ -20,14 +20,14 @@ import ( "strconv" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) type selfSymlink struct { diff --git a/pkg/sentry/fsimpl/proc/tasks_net.go b/pkg/sentry/fsimpl/proc/tasks_net.go index 4aaf23e97..608fec017 100644 --- a/pkg/sentry/fsimpl/proc/tasks_net.go +++ b/pkg/sentry/fsimpl/proc/tasks_net.go @@ -22,8 +22,8 @@ import ( "time" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/inet" @@ -32,9 +32,9 @@ import ( "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/unix" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/usermem" ) func newNetDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) *kernfs.Dentry { diff --git a/pkg/sentry/fsimpl/proc/tasks_sys.go b/pkg/sentry/fsimpl/proc/tasks_sys.go index aabf2bf0c..ad963870b 100644 --- a/pkg/sentry/fsimpl/proc/tasks_sys.go +++ b/pkg/sentry/fsimpl/proc/tasks_sys.go @@ -19,7 +19,7 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" diff --git a/pkg/sentry/fsimpl/proc/tasks_sys_test.go b/pkg/sentry/fsimpl/proc/tasks_sys_test.go index 0a1d3f34b..be54897bb 100644 --- a/pkg/sentry/fsimpl/proc/tasks_sys_test.go +++ b/pkg/sentry/fsimpl/proc/tasks_sys_test.go @@ -20,7 +20,7 @@ import ( "testing" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/inet" ) diff --git a/pkg/sentry/fsimpl/proc/tasks_test.go b/pkg/sentry/fsimpl/proc/tasks_test.go index 2c1635f33..6fc3524db 100644 --- a/pkg/sentry/fsimpl/proc/tasks_test.go +++ b/pkg/sentry/fsimpl/proc/tasks_test.go @@ -22,14 +22,14 @@ import ( "testing" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fsimpl/testutil" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) var ( diff --git a/pkg/sentry/fsimpl/sys/BUILD b/pkg/sentry/fsimpl/sys/BUILD index beda141f1..66c0d8bc8 100644 --- a/pkg/sentry/fsimpl/sys/BUILD +++ b/pkg/sentry/fsimpl/sys/BUILD @@ -9,7 +9,7 @@ go_library( ], deps = [ "//pkg/abi/linux", - "//pkg/sentry/context", + "//pkg/context", "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go index 1305ad01d..e35d52d17 100644 --- a/pkg/sentry/fsimpl/sys/sys.go +++ b/pkg/sentry/fsimpl/sys/sys.go @@ -20,7 +20,7 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" diff --git a/pkg/sentry/fsimpl/testutil/BUILD b/pkg/sentry/fsimpl/testutil/BUILD index 12053a5b6..efd5974c4 100644 --- a/pkg/sentry/fsimpl/testutil/BUILD +++ b/pkg/sentry/fsimpl/testutil/BUILD @@ -12,10 +12,10 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/cpuid", "//pkg/fspath", "//pkg/memutil", - "//pkg/sentry/context", "//pkg/sentry/fs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", @@ -27,9 +27,9 @@ go_library( "//pkg/sentry/platform/kvm", "//pkg/sentry/platform/ptrace", "//pkg/sentry/time", - "//pkg/sentry/usermem", "//pkg/sentry/vfs", "//pkg/sync", + "//pkg/usermem", "@com_github_google_go-cmp//cmp:go_default_library", ], ) diff --git a/pkg/sentry/fsimpl/testutil/kernel.go b/pkg/sentry/fsimpl/testutil/kernel.go index 295da2d52..89f8c4915 100644 --- a/pkg/sentry/fsimpl/testutil/kernel.go +++ b/pkg/sentry/fsimpl/testutil/kernel.go @@ -21,9 +21,9 @@ import ( "runtime" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/memutil" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" diff --git a/pkg/sentry/fsimpl/testutil/testutil.go b/pkg/sentry/fsimpl/testutil/testutil.go index 2a723a89f..1c98335c1 100644 --- a/pkg/sentry/fsimpl/testutil/testutil.go +++ b/pkg/sentry/fsimpl/testutil/testutil.go @@ -24,12 +24,12 @@ import ( "github.com/google/go-cmp/cmp" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/usermem" ) // System represents the context for a single test. diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD index 857e98bc5..fb436860c 100644 --- a/pkg/sentry/fsimpl/tmpfs/BUILD +++ b/pkg/sentry/fsimpl/tmpfs/BUILD @@ -30,10 +30,11 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/amutex", + "//pkg/context", "//pkg/fspath", "//pkg/log", + "//pkg/safemem", "//pkg/sentry/arch", - "//pkg/sentry/context", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", "//pkg/sentry/kernel", @@ -43,12 +44,11 @@ go_library( "//pkg/sentry/memmap", "//pkg/sentry/pgalloc", "//pkg/sentry/platform", - "//pkg/sentry/safemem", "//pkg/sentry/usage", - "//pkg/sentry/usermem", "//pkg/sentry/vfs", "//pkg/sync", "//pkg/syserror", + "//pkg/usermem", ], ) @@ -59,10 +59,10 @@ go_test( deps = [ ":tmpfs", "//pkg/abi/linux", + "//pkg/context", "//pkg/fspath", "//pkg/refs", - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", + "//pkg/sentry/contexttest", "//pkg/sentry/fs", "//pkg/sentry/fs/tmpfs", "//pkg/sentry/kernel/auth", @@ -82,13 +82,13 @@ go_test( library = ":tmpfs", deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/fspath", - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", + "//pkg/sentry/contexttest", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/contexttest", - "//pkg/sentry/usermem", "//pkg/sentry/vfs", "//pkg/syserror", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go index d88c83499..54241c8e8 100644 --- a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go +++ b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go @@ -21,10 +21,10 @@ import ( "testing" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/fs" _ "gvisor.dev/gvisor/pkg/sentry/fs/tmpfs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" diff --git a/pkg/sentry/fsimpl/tmpfs/directory.go b/pkg/sentry/fsimpl/tmpfs/directory.go index 887ca2619..dc0d27cf9 100644 --- a/pkg/sentry/fsimpl/tmpfs/directory.go +++ b/pkg/sentry/fsimpl/tmpfs/directory.go @@ -16,7 +16,7 @@ package tmpfs import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index d726f03c5..5ee9cf1e9 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -19,8 +19,8 @@ import ( "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/fsimpl/tmpfs/named_pipe.go b/pkg/sentry/fsimpl/tmpfs/named_pipe.go index 482aabd52..0c57fdca3 100644 --- a/pkg/sentry/fsimpl/tmpfs/named_pipe.go +++ b/pkg/sentry/fsimpl/tmpfs/named_pipe.go @@ -16,11 +16,11 @@ package tmpfs import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/usermem" ) type namedPipe struct { diff --git a/pkg/sentry/fsimpl/tmpfs/pipe_test.go b/pkg/sentry/fsimpl/tmpfs/pipe_test.go index 70b42a6ec..5ee7f2a72 100644 --- a/pkg/sentry/fsimpl/tmpfs/pipe_test.go +++ b/pkg/sentry/fsimpl/tmpfs/pipe_test.go @@ -19,13 +19,13 @@ import ( "testing" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) const fileName = "mypipe" diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index 7c633c1b0..e9e6faf67 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -20,17 +20,17 @@ import ( "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) type regularFile struct { diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go index 034a29fdb..32552e261 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go @@ -22,12 +22,12 @@ import ( "testing" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/usermem" ) // nextFileID is used to generate unique file names. diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 515f033f2..88dbd6e35 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -29,7 +29,7 @@ import ( "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/pgalloc" diff --git a/pkg/sentry/hostmm/BUILD b/pkg/sentry/hostmm/BUILD index a145a5ca3..61c78569d 100644 --- a/pkg/sentry/hostmm/BUILD +++ b/pkg/sentry/hostmm/BUILD @@ -12,6 +12,6 @@ go_library( deps = [ "//pkg/fd", "//pkg/log", - "//pkg/sentry/usermem", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/hostmm/hostmm.go b/pkg/sentry/hostmm/hostmm.go index 19335ca73..506c7864a 100644 --- a/pkg/sentry/hostmm/hostmm.go +++ b/pkg/sentry/hostmm/hostmm.go @@ -24,7 +24,7 @@ import ( "gvisor.dev/gvisor/pkg/fd" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // NotifyCurrentMemcgPressureCallback requests that f is called whenever the diff --git a/pkg/sentry/inet/BUILD b/pkg/sentry/inet/BUILD index aa621b724..334432abf 100644 --- a/pkg/sentry/inet/BUILD +++ b/pkg/sentry/inet/BUILD @@ -13,7 +13,7 @@ go_library( "test_stack.go", ], deps = [ - "//pkg/sentry/context", + "//pkg/context", "//pkg/tcpip/stack", ], ) diff --git a/pkg/sentry/inet/context.go b/pkg/sentry/inet/context.go index 4eda7dd1f..e8cc1bffd 100644 --- a/pkg/sentry/inet/context.go +++ b/pkg/sentry/inet/context.go @@ -15,7 +15,7 @@ package inet import ( - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" ) // contextID is the inet package's type for context.Context.Value keys. diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD index cebaccd92..0738946d9 100644 --- a/pkg/sentry/kernel/BUILD +++ b/pkg/sentry/kernel/BUILD @@ -153,14 +153,15 @@ go_library( "//pkg/binary", "//pkg/bits", "//pkg/bpf", + "//pkg/context", "//pkg/cpuid", "//pkg/eventchannel", "//pkg/log", "//pkg/metric", "//pkg/refs", + "//pkg/safemem", "//pkg/secio", "//pkg/sentry/arch", - "//pkg/sentry/context", "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/lock", @@ -180,7 +181,6 @@ go_library( "//pkg/sentry/mm", "//pkg/sentry/pgalloc", "//pkg/sentry/platform", - "//pkg/sentry/safemem", "//pkg/sentry/socket/netlink/port", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/time", @@ -188,7 +188,6 @@ go_library( "//pkg/sentry/unimpl:unimplemented_syscall_go_proto", "//pkg/sentry/uniqueid", "//pkg/sentry/usage", - "//pkg/sentry/usermem", "//pkg/state", "//pkg/state/statefile", "//pkg/sync", @@ -196,6 +195,7 @@ go_library( "//pkg/syserror", "//pkg/tcpip", "//pkg/tcpip/stack", + "//pkg/usermem", "//pkg/waiter", ], ) @@ -212,9 +212,9 @@ go_test( library = ":kernel", deps = [ "//pkg/abi", + "//pkg/context", "//pkg/sentry/arch", - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", + "//pkg/sentry/contexttest", "//pkg/sentry/fs", "//pkg/sentry/fs/filetest", "//pkg/sentry/kernel/sched", @@ -222,8 +222,8 @@ go_test( "//pkg/sentry/pgalloc", "//pkg/sentry/time", "//pkg/sentry/usage", - "//pkg/sentry/usermem", "//pkg/sync", "//pkg/syserror", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/kernel/auth/BUILD b/pkg/sentry/kernel/auth/BUILD index 64537c9be..2bc49483a 100644 --- a/pkg/sentry/kernel/auth/BUILD +++ b/pkg/sentry/kernel/auth/BUILD @@ -61,8 +61,8 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/bits", + "//pkg/context", "//pkg/log", - "//pkg/sentry/context", "//pkg/sync", "//pkg/syserror", ], diff --git a/pkg/sentry/kernel/auth/context.go b/pkg/sentry/kernel/auth/context.go index 5c0e7d6b6..ef5723127 100644 --- a/pkg/sentry/kernel/auth/context.go +++ b/pkg/sentry/kernel/auth/context.go @@ -15,7 +15,7 @@ package auth import ( - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" ) // contextID is the auth package's type for context.Context.Value keys. diff --git a/pkg/sentry/kernel/auth/id_map.go b/pkg/sentry/kernel/auth/id_map.go index 3d74bc610..28cbe159d 100644 --- a/pkg/sentry/kernel/auth/id_map.go +++ b/pkg/sentry/kernel/auth/id_map.go @@ -16,7 +16,7 @@ package auth import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/kernel/context.go b/pkg/sentry/kernel/context.go index 3c9dceaba..0c40bf315 100644 --- a/pkg/sentry/kernel/context.go +++ b/pkg/sentry/kernel/context.go @@ -17,8 +17,8 @@ package kernel import ( "time" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/context" ) // contextID is the kernel package's type for context.Context.Value keys. diff --git a/pkg/sentry/kernel/contexttest/BUILD b/pkg/sentry/kernel/contexttest/BUILD index daff608d7..9d26392c0 100644 --- a/pkg/sentry/kernel/contexttest/BUILD +++ b/pkg/sentry/kernel/contexttest/BUILD @@ -8,8 +8,8 @@ go_library( srcs = ["contexttest.go"], visibility = ["//pkg/sentry:internal"], deps = [ - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", + "//pkg/context", + "//pkg/sentry/contexttest", "//pkg/sentry/kernel", "//pkg/sentry/pgalloc", "//pkg/sentry/platform", diff --git a/pkg/sentry/kernel/contexttest/contexttest.go b/pkg/sentry/kernel/contexttest/contexttest.go index 82f9d8922..22c340e56 100644 --- a/pkg/sentry/kernel/contexttest/contexttest.go +++ b/pkg/sentry/kernel/contexttest/contexttest.go @@ -19,8 +19,8 @@ package contexttest import ( "testing" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/platform" diff --git a/pkg/sentry/kernel/epoll/BUILD b/pkg/sentry/kernel/epoll/BUILD index 19e16ab3a..dedf0fa15 100644 --- a/pkg/sentry/kernel/epoll/BUILD +++ b/pkg/sentry/kernel/epoll/BUILD @@ -24,13 +24,13 @@ go_library( ], visibility = ["//pkg/sentry:internal"], deps = [ + "//pkg/context", "//pkg/refs", - "//pkg/sentry/context", "//pkg/sentry/fs", "//pkg/sentry/fs/anon", "//pkg/sentry/fs/fsutil", - "//pkg/sentry/usermem", "//pkg/sync", + "//pkg/usermem", "//pkg/waiter", ], ) @@ -43,7 +43,7 @@ go_test( ], library = ":epoll", deps = [ - "//pkg/sentry/context/contexttest", + "//pkg/sentry/contexttest", "//pkg/sentry/fs/filetest", "//pkg/waiter", ], diff --git a/pkg/sentry/kernel/epoll/epoll.go b/pkg/sentry/kernel/epoll/epoll.go index e84742993..8bffb78fc 100644 --- a/pkg/sentry/kernel/epoll/epoll.go +++ b/pkg/sentry/kernel/epoll/epoll.go @@ -20,13 +20,13 @@ import ( "fmt" "syscall" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/anon" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/kernel/epoll/epoll_test.go b/pkg/sentry/kernel/epoll/epoll_test.go index 4a20d4c82..22630e9c5 100644 --- a/pkg/sentry/kernel/epoll/epoll_test.go +++ b/pkg/sentry/kernel/epoll/epoll_test.go @@ -17,7 +17,7 @@ package epoll import ( "testing" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/fs/filetest" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/kernel/eventfd/BUILD b/pkg/sentry/kernel/eventfd/BUILD index ee2d74864..9983a32e5 100644 --- a/pkg/sentry/kernel/eventfd/BUILD +++ b/pkg/sentry/kernel/eventfd/BUILD @@ -8,14 +8,14 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/fdnotifier", - "//pkg/sentry/context", "//pkg/sentry/fs", "//pkg/sentry/fs/anon", "//pkg/sentry/fs/fsutil", - "//pkg/sentry/usermem", "//pkg/sync", "//pkg/syserror", + "//pkg/usermem", "//pkg/waiter", ], ) @@ -26,8 +26,8 @@ go_test( srcs = ["eventfd_test.go"], library = ":eventfd", deps = [ - "//pkg/sentry/context/contexttest", - "//pkg/sentry/usermem", + "//pkg/sentry/contexttest", + "//pkg/usermem", "//pkg/waiter", ], ) diff --git a/pkg/sentry/kernel/eventfd/eventfd.go b/pkg/sentry/kernel/eventfd/eventfd.go index 687690679..87951adeb 100644 --- a/pkg/sentry/kernel/eventfd/eventfd.go +++ b/pkg/sentry/kernel/eventfd/eventfd.go @@ -21,14 +21,14 @@ import ( "syscall" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fdnotifier" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/anon" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/kernel/eventfd/eventfd_test.go b/pkg/sentry/kernel/eventfd/eventfd_test.go index 018c7f3ef..9b4892f74 100644 --- a/pkg/sentry/kernel/eventfd/eventfd_test.go +++ b/pkg/sentry/kernel/eventfd/eventfd_test.go @@ -17,8 +17,8 @@ package eventfd import ( "testing" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sentry/contexttest" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/kernel/fd_table.go b/pkg/sentry/kernel/fd_table.go index 0ad4135b3..9460bb235 100644 --- a/pkg/sentry/kernel/fd_table.go +++ b/pkg/sentry/kernel/fd_table.go @@ -22,8 +22,8 @@ import ( "syscall" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/limits" diff --git a/pkg/sentry/kernel/fd_table_test.go b/pkg/sentry/kernel/fd_table_test.go index 86164df49..261b815f2 100644 --- a/pkg/sentry/kernel/fd_table_test.go +++ b/pkg/sentry/kernel/fd_table_test.go @@ -18,8 +18,8 @@ import ( "runtime" "testing" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/filetest" "gvisor.dev/gvisor/pkg/sentry/limits" diff --git a/pkg/sentry/kernel/futex/BUILD b/pkg/sentry/kernel/futex/BUILD index f413d8ae2..c5021f2db 100644 --- a/pkg/sentry/kernel/futex/BUILD +++ b/pkg/sentry/kernel/futex/BUILD @@ -36,12 +36,12 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/log", - "//pkg/sentry/context", "//pkg/sentry/memmap", - "//pkg/sentry/usermem", "//pkg/sync", "//pkg/syserror", + "//pkg/usermem", ], ) @@ -51,7 +51,7 @@ go_test( srcs = ["futex_test.go"], library = ":futex", deps = [ - "//pkg/sentry/usermem", "//pkg/sync", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/kernel/futex/futex.go b/pkg/sentry/kernel/futex/futex.go index d1931c8f4..732e66da4 100644 --- a/pkg/sentry/kernel/futex/futex.go +++ b/pkg/sentry/kernel/futex/futex.go @@ -20,9 +20,9 @@ package futex import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // KeyKind indicates the type of a Key. diff --git a/pkg/sentry/kernel/futex/futex_test.go b/pkg/sentry/kernel/futex/futex_test.go index c23126ca5..7c5c7665b 100644 --- a/pkg/sentry/kernel/futex/futex_test.go +++ b/pkg/sentry/kernel/futex/futex_test.go @@ -22,8 +22,8 @@ import ( "testing" "unsafe" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/usermem" ) // testData implements the Target interface, and allows us to diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index c85e97fef..7b90fac5a 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -40,12 +40,12 @@ import ( "time" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/eventchannel" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/timerfd" "gvisor.dev/gvisor/pkg/sentry/hostcpu" diff --git a/pkg/sentry/kernel/pipe/BUILD b/pkg/sentry/kernel/pipe/BUILD index 2c7b6206f..4c049d5b4 100644 --- a/pkg/sentry/kernel/pipe/BUILD +++ b/pkg/sentry/kernel/pipe/BUILD @@ -33,16 +33,16 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/amutex", + "//pkg/context", + "//pkg/safemem", "//pkg/sentry/arch", - "//pkg/sentry/context", "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", - "//pkg/sentry/safemem", - "//pkg/sentry/usermem", "//pkg/sentry/vfs", "//pkg/sync", "//pkg/syserror", + "//pkg/usermem", "//pkg/waiter", ], ) @@ -57,11 +57,11 @@ go_test( ], library = ":pipe", deps = [ - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", + "//pkg/context", + "//pkg/sentry/contexttest", "//pkg/sentry/fs", - "//pkg/sentry/usermem", "//pkg/syserror", + "//pkg/usermem", "//pkg/waiter", ], ) diff --git a/pkg/sentry/kernel/pipe/buffer.go b/pkg/sentry/kernel/pipe/buffer.go index 1c0f34269..fe3be5dbd 100644 --- a/pkg/sentry/kernel/pipe/buffer.go +++ b/pkg/sentry/kernel/pipe/buffer.go @@ -17,7 +17,7 @@ package pipe import ( "io" - "gvisor.dev/gvisor/pkg/sentry/safemem" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sync" ) diff --git a/pkg/sentry/kernel/pipe/buffer_test.go b/pkg/sentry/kernel/pipe/buffer_test.go index ee1b90115..4d54b8b8f 100644 --- a/pkg/sentry/kernel/pipe/buffer_test.go +++ b/pkg/sentry/kernel/pipe/buffer_test.go @@ -18,7 +18,7 @@ import ( "testing" "unsafe" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) func TestBufferSize(t *testing.T) { diff --git a/pkg/sentry/kernel/pipe/node.go b/pkg/sentry/kernel/pipe/node.go index 716f589af..4b688c627 100644 --- a/pkg/sentry/kernel/pipe/node.go +++ b/pkg/sentry/kernel/pipe/node.go @@ -16,7 +16,7 @@ package pipe import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sync" diff --git a/pkg/sentry/kernel/pipe/node_test.go b/pkg/sentry/kernel/pipe/node_test.go index 16fa80abe..ab75a87ff 100644 --- a/pkg/sentry/kernel/pipe/node_test.go +++ b/pkg/sentry/kernel/pipe/node_test.go @@ -18,11 +18,11 @@ import ( "testing" "time" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) type sleeper struct { diff --git a/pkg/sentry/kernel/pipe/pipe.go b/pkg/sentry/kernel/pipe/pipe.go index e4fd7d420..08410283f 100644 --- a/pkg/sentry/kernel/pipe/pipe.go +++ b/pkg/sentry/kernel/pipe/pipe.go @@ -20,7 +20,7 @@ import ( "sync/atomic" "syscall" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" diff --git a/pkg/sentry/kernel/pipe/pipe_test.go b/pkg/sentry/kernel/pipe/pipe_test.go index e3a14b665..bda739dbe 100644 --- a/pkg/sentry/kernel/pipe/pipe_test.go +++ b/pkg/sentry/kernel/pipe/pipe_test.go @@ -18,9 +18,9 @@ import ( "bytes" "testing" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/kernel/pipe/pipe_util.go b/pkg/sentry/kernel/pipe/pipe_util.go index 8394eb78b..80158239e 100644 --- a/pkg/sentry/kernel/pipe/pipe_util.go +++ b/pkg/sentry/kernel/pipe/pipe_util.go @@ -21,10 +21,10 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/amutex" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/kernel/pipe/reader_writer.go b/pkg/sentry/kernel/pipe/reader_writer.go index b4d29fc77..b2b5691ee 100644 --- a/pkg/sentry/kernel/pipe/reader_writer.go +++ b/pkg/sentry/kernel/pipe/reader_writer.go @@ -17,11 +17,11 @@ package pipe import ( "io" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // ReaderWriter satisfies the FileOperations interface and services both diff --git a/pkg/sentry/kernel/pipe/vfs.go b/pkg/sentry/kernel/pipe/vfs.go index 6f83e3cee..a5675bd70 100644 --- a/pkg/sentry/kernel/pipe/vfs.go +++ b/pkg/sentry/kernel/pipe/vfs.go @@ -16,12 +16,12 @@ package pipe import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/kernel/ptrace.go b/pkg/sentry/kernel/ptrace.go index 3be171cdc..35ad97d5d 100644 --- a/pkg/sentry/kernel/ptrace.go +++ b/pkg/sentry/kernel/ptrace.go @@ -20,8 +20,8 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/mm" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // ptraceOptions are the subset of options controlling a task's ptrace behavior diff --git a/pkg/sentry/kernel/ptrace_amd64.go b/pkg/sentry/kernel/ptrace_amd64.go index 5514cf432..cef1276ec 100644 --- a/pkg/sentry/kernel/ptrace_amd64.go +++ b/pkg/sentry/kernel/ptrace_amd64.go @@ -18,8 +18,8 @@ package kernel import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // ptraceArch implements arch-specific ptrace commands. diff --git a/pkg/sentry/kernel/ptrace_arm64.go b/pkg/sentry/kernel/ptrace_arm64.go index 61e412911..d971b96b3 100644 --- a/pkg/sentry/kernel/ptrace_arm64.go +++ b/pkg/sentry/kernel/ptrace_arm64.go @@ -17,8 +17,8 @@ package kernel import ( - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // ptraceArch implements arch-specific ptrace commands. diff --git a/pkg/sentry/kernel/rseq.go b/pkg/sentry/kernel/rseq.go index b14429854..efebfd872 100644 --- a/pkg/sentry/kernel/rseq.go +++ b/pkg/sentry/kernel/rseq.go @@ -19,8 +19,8 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/hostcpu" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // Restartable sequences. diff --git a/pkg/sentry/kernel/seccomp.go b/pkg/sentry/kernel/seccomp.go index 2347dcf36..c38c5a40c 100644 --- a/pkg/sentry/kernel/seccomp.go +++ b/pkg/sentry/kernel/seccomp.go @@ -21,8 +21,8 @@ import ( "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/bpf" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) const maxSyscallFilterInstructions = 1 << 15 diff --git a/pkg/sentry/kernel/semaphore/BUILD b/pkg/sentry/kernel/semaphore/BUILD index 76e19b551..65e5427c1 100644 --- a/pkg/sentry/kernel/semaphore/BUILD +++ b/pkg/sentry/kernel/semaphore/BUILD @@ -24,8 +24,8 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/log", - "//pkg/sentry/context", "//pkg/sentry/fs", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/time", @@ -41,8 +41,8 @@ go_test( library = ":semaphore", deps = [ "//pkg/abi/linux", - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", + "//pkg/context", + "//pkg/sentry/contexttest", "//pkg/sentry/kernel/auth", "//pkg/syserror", ], diff --git a/pkg/sentry/kernel/semaphore/semaphore.go b/pkg/sentry/kernel/semaphore/semaphore.go index 18299814e..1000f3287 100644 --- a/pkg/sentry/kernel/semaphore/semaphore.go +++ b/pkg/sentry/kernel/semaphore/semaphore.go @@ -19,8 +19,8 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" diff --git a/pkg/sentry/kernel/semaphore/semaphore_test.go b/pkg/sentry/kernel/semaphore/semaphore_test.go index c235f6ca4..e47acefdf 100644 --- a/pkg/sentry/kernel/semaphore/semaphore_test.go +++ b/pkg/sentry/kernel/semaphore/semaphore_test.go @@ -18,8 +18,8 @@ import ( "testing" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/kernel/shm/BUILD b/pkg/sentry/kernel/shm/BUILD index 5547c5abf..bfd779837 100644 --- a/pkg/sentry/kernel/shm/BUILD +++ b/pkg/sentry/kernel/shm/BUILD @@ -11,9 +11,9 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/log", "//pkg/refs", - "//pkg/sentry/context", "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/kernel/auth", @@ -22,8 +22,8 @@ go_library( "//pkg/sentry/pgalloc", "//pkg/sentry/platform", "//pkg/sentry/usage", - "//pkg/sentry/usermem", "//pkg/sync", "//pkg/syserror", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/kernel/shm/shm.go b/pkg/sentry/kernel/shm/shm.go index 8ddef7eb8..208569057 100644 --- a/pkg/sentry/kernel/shm/shm.go +++ b/pkg/sentry/kernel/shm/shm.go @@ -37,9 +37,9 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" @@ -47,9 +47,9 @@ import ( "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // Key represents a shm segment key. Analogous to a file name. diff --git a/pkg/sentry/kernel/signalfd/BUILD b/pkg/sentry/kernel/signalfd/BUILD index 5d44773d4..3eb78e91b 100644 --- a/pkg/sentry/kernel/signalfd/BUILD +++ b/pkg/sentry/kernel/signalfd/BUILD @@ -9,14 +9,14 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/binary", - "//pkg/sentry/context", + "//pkg/context", "//pkg/sentry/fs", "//pkg/sentry/fs/anon", "//pkg/sentry/fs/fsutil", "//pkg/sentry/kernel", - "//pkg/sentry/usermem", "//pkg/sync", "//pkg/syserror", + "//pkg/usermem", "//pkg/waiter", ], ) diff --git a/pkg/sentry/kernel/signalfd/signalfd.go b/pkg/sentry/kernel/signalfd/signalfd.go index 28be4a939..8243bb93e 100644 --- a/pkg/sentry/kernel/signalfd/signalfd.go +++ b/pkg/sentry/kernel/signalfd/signalfd.go @@ -18,14 +18,14 @@ package signalfd import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/anon" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/kernel/syscalls.go b/pkg/sentry/kernel/syscalls.go index d2d01add4..93c4fe969 100644 --- a/pkg/sentry/kernel/syscalls.go +++ b/pkg/sentry/kernel/syscalls.go @@ -21,8 +21,8 @@ import ( "gvisor.dev/gvisor/pkg/abi" "gvisor.dev/gvisor/pkg/bits" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/usermem" ) // maxSyscallNum is the highest supported syscall number. diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index 978d66da8..95adf2778 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -21,8 +21,8 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/bpf" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" @@ -35,8 +35,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/unimpl" "gvisor.dev/gvisor/pkg/sentry/uniqueid" "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/kernel/task_clone.go b/pkg/sentry/kernel/task_clone.go index 247bd4aba..53d4d211b 100644 --- a/pkg/sentry/kernel/task_clone.go +++ b/pkg/sentry/kernel/task_clone.go @@ -17,8 +17,8 @@ package kernel import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/bpf" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // SharingOptions controls what resources are shared by a new task created by diff --git a/pkg/sentry/kernel/task_context.go b/pkg/sentry/kernel/task_context.go index bb5560acf..2d6e7733c 100644 --- a/pkg/sentry/kernel/task_context.go +++ b/pkg/sentry/kernel/task_context.go @@ -18,13 +18,13 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel/futex" "gvisor.dev/gvisor/pkg/sentry/loader" "gvisor.dev/gvisor/pkg/sentry/mm" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserr" + "gvisor.dev/gvisor/pkg/usermem" ) var errNoSyscalls = syserr.New("no syscall table found", linux.ENOEXEC) diff --git a/pkg/sentry/kernel/task_futex.go b/pkg/sentry/kernel/task_futex.go index c211b5b74..a53e77c9f 100644 --- a/pkg/sentry/kernel/task_futex.go +++ b/pkg/sentry/kernel/task_futex.go @@ -16,7 +16,7 @@ package kernel import ( "gvisor.dev/gvisor/pkg/sentry/kernel/futex" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // Futex returns t's futex manager. diff --git a/pkg/sentry/kernel/task_log.go b/pkg/sentry/kernel/task_log.go index 0fb3661de..41259210c 100644 --- a/pkg/sentry/kernel/task_log.go +++ b/pkg/sentry/kernel/task_log.go @@ -20,7 +20,7 @@ import ( "sort" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) const ( diff --git a/pkg/sentry/kernel/task_run.go b/pkg/sentry/kernel/task_run.go index 6357273d3..5568c91bc 100644 --- a/pkg/sentry/kernel/task_run.go +++ b/pkg/sentry/kernel/task_run.go @@ -26,7 +26,7 @@ import ( ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // A taskRunState is a reified state in the task state machine. See README.md diff --git a/pkg/sentry/kernel/task_signals.go b/pkg/sentry/kernel/task_signals.go index 39cd1340d..8802db142 100644 --- a/pkg/sentry/kernel/task_signals.go +++ b/pkg/sentry/kernel/task_signals.go @@ -26,8 +26,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ucspb "gvisor.dev/gvisor/pkg/sentry/kernel/uncaught_signal_go_proto" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/kernel/task_start.go b/pkg/sentry/kernel/task_start.go index 58af16ee2..de838beef 100644 --- a/pkg/sentry/kernel/task_start.go +++ b/pkg/sentry/kernel/task_start.go @@ -21,8 +21,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/futex" "gvisor.dev/gvisor/pkg/sentry/kernel/sched" "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // TaskConfig defines the configuration of a new Task (see below). diff --git a/pkg/sentry/kernel/task_syscall.go b/pkg/sentry/kernel/task_syscall.go index 3180f5560..d555d69a8 100644 --- a/pkg/sentry/kernel/task_syscall.go +++ b/pkg/sentry/kernel/task_syscall.go @@ -25,8 +25,8 @@ import ( "gvisor.dev/gvisor/pkg/metric" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // SyscallRestartErrno represents a ERESTART* errno defined in the Linux's kernel diff --git a/pkg/sentry/kernel/task_usermem.go b/pkg/sentry/kernel/task_usermem.go index 518bfe1bd..2bf3ce8a8 100644 --- a/pkg/sentry/kernel/task_usermem.go +++ b/pkg/sentry/kernel/task_usermem.go @@ -18,8 +18,8 @@ import ( "math" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // MAX_RW_COUNT is the maximum size in bytes of a single read or write. diff --git a/pkg/sentry/kernel/time/BUILD b/pkg/sentry/kernel/time/BUILD index d49594d9f..7ba7dc50c 100644 --- a/pkg/sentry/kernel/time/BUILD +++ b/pkg/sentry/kernel/time/BUILD @@ -11,7 +11,7 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", - "//pkg/sentry/context", + "//pkg/context", "//pkg/sync", "//pkg/syserror", "//pkg/waiter", diff --git a/pkg/sentry/kernel/time/context.go b/pkg/sentry/kernel/time/context.go index 8ef483dd3..00b729d88 100644 --- a/pkg/sentry/kernel/time/context.go +++ b/pkg/sentry/kernel/time/context.go @@ -15,7 +15,7 @@ package time import ( - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" ) // contextID is the time package's type for context.Context.Value keys. diff --git a/pkg/sentry/kernel/timekeeper_test.go b/pkg/sentry/kernel/timekeeper_test.go index 849c5b646..cf2f7ca72 100644 --- a/pkg/sentry/kernel/timekeeper_test.go +++ b/pkg/sentry/kernel/timekeeper_test.go @@ -17,12 +17,12 @@ package kernel import ( "testing" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/pgalloc" sentrytime "gvisor.dev/gvisor/pkg/sentry/time" "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // mockClocks is a sentrytime.Clocks that simply returns the times in the diff --git a/pkg/sentry/kernel/vdso.go b/pkg/sentry/kernel/vdso.go index fdd10c56c..f1b3c212c 100644 --- a/pkg/sentry/kernel/vdso.go +++ b/pkg/sentry/kernel/vdso.go @@ -18,10 +18,10 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/safemem" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // vdsoParams are the parameters exposed to the VDSO. diff --git a/pkg/sentry/limits/BUILD b/pkg/sentry/limits/BUILD index 67869757f..cf591c4c1 100644 --- a/pkg/sentry/limits/BUILD +++ b/pkg/sentry/limits/BUILD @@ -12,7 +12,7 @@ go_library( visibility = ["//:sandbox"], deps = [ "//pkg/abi/linux", - "//pkg/sentry/context", + "//pkg/context", "//pkg/sync", ], ) diff --git a/pkg/sentry/limits/context.go b/pkg/sentry/limits/context.go index 6972749ed..77e1fe217 100644 --- a/pkg/sentry/limits/context.go +++ b/pkg/sentry/limits/context.go @@ -15,7 +15,7 @@ package limits import ( - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" ) // contextID is the limit package's type for context.Context.Value keys. diff --git a/pkg/sentry/loader/BUILD b/pkg/sentry/loader/BUILD index d4ad2bd6c..23790378a 100644 --- a/pkg/sentry/loader/BUILD +++ b/pkg/sentry/loader/BUILD @@ -24,11 +24,12 @@ go_library( "//pkg/abi", "//pkg/abi/linux", "//pkg/binary", + "//pkg/context", "//pkg/cpuid", "//pkg/log", "//pkg/rand", + "//pkg/safemem", "//pkg/sentry/arch", - "//pkg/sentry/context", "//pkg/sentry/fs", "//pkg/sentry/fs/anon", "//pkg/sentry/fs/fsutil", @@ -37,12 +38,11 @@ go_library( "//pkg/sentry/memmap", "//pkg/sentry/mm", "//pkg/sentry/pgalloc", - "//pkg/sentry/safemem", "//pkg/sentry/uniqueid", "//pkg/sentry/usage", - "//pkg/sentry/usermem", "//pkg/syserr", "//pkg/syserror", + "//pkg/usermem", "//pkg/waiter", ], ) diff --git a/pkg/sentry/loader/elf.go b/pkg/sentry/loader/elf.go index 6299a3e2f..122ed05c2 100644 --- a/pkg/sentry/loader/elf.go +++ b/pkg/sentry/loader/elf.go @@ -23,16 +23,16 @@ import ( "gvisor.dev/gvisor/pkg/abi" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/mm" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) const ( diff --git a/pkg/sentry/loader/interpreter.go b/pkg/sentry/loader/interpreter.go index ccf909cac..098a45d36 100644 --- a/pkg/sentry/loader/interpreter.go +++ b/pkg/sentry/loader/interpreter.go @@ -18,10 +18,10 @@ import ( "bytes" "io" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) const ( diff --git a/pkg/sentry/loader/loader.go b/pkg/sentry/loader/loader.go index b03eeb005..9a613d6b7 100644 --- a/pkg/sentry/loader/loader.go +++ b/pkg/sentry/loader/loader.go @@ -24,16 +24,16 @@ import ( "gvisor.dev/gvisor/pkg/abi" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/rand" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/mm" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // LoadArgs holds specifications for an executable file to be loaded. diff --git a/pkg/sentry/loader/vdso.go b/pkg/sentry/loader/vdso.go index df8a81907..52f446ed7 100644 --- a/pkg/sentry/loader/vdso.go +++ b/pkg/sentry/loader/vdso.go @@ -20,20 +20,20 @@ import ( "io" "gvisor.dev/gvisor/pkg/abi" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/anon" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/mm" "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/uniqueid" "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/memmap/BUILD b/pkg/sentry/memmap/BUILD index f9a65f086..a98b66de1 100644 --- a/pkg/sentry/memmap/BUILD +++ b/pkg/sentry/memmap/BUILD @@ -38,11 +38,11 @@ go_library( ], visibility = ["//pkg/sentry:internal"], deps = [ + "//pkg/context", "//pkg/log", - "//pkg/sentry/context", "//pkg/sentry/platform", - "//pkg/sentry/usermem", "//pkg/syserror", + "//pkg/usermem", ], ) @@ -51,5 +51,5 @@ go_test( size = "small", srcs = ["mapping_set_test.go"], library = ":memmap", - deps = ["//pkg/sentry/usermem"], + deps = ["//pkg/usermem"], ) diff --git a/pkg/sentry/memmap/mapping_set.go b/pkg/sentry/memmap/mapping_set.go index 0a5b7ce45..d609c1ae0 100644 --- a/pkg/sentry/memmap/mapping_set.go +++ b/pkg/sentry/memmap/mapping_set.go @@ -18,7 +18,7 @@ import ( "fmt" "math" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // MappingSet maps offsets into a Mappable to mappings of those offsets. It is diff --git a/pkg/sentry/memmap/mapping_set_test.go b/pkg/sentry/memmap/mapping_set_test.go index f9b11a59c..d39efe38f 100644 --- a/pkg/sentry/memmap/mapping_set_test.go +++ b/pkg/sentry/memmap/mapping_set_test.go @@ -18,7 +18,7 @@ import ( "reflect" "testing" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) type testMappingSpace struct { diff --git a/pkg/sentry/memmap/memmap.go b/pkg/sentry/memmap/memmap.go index 16a722a13..c6db9fc8f 100644 --- a/pkg/sentry/memmap/memmap.go +++ b/pkg/sentry/memmap/memmap.go @@ -18,9 +18,9 @@ package memmap import ( "fmt" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // Mappable represents a memory-mappable object, a mutable mapping from uint64 diff --git a/pkg/sentry/mm/BUILD b/pkg/sentry/mm/BUILD index bd6399fa2..e5729ced5 100644 --- a/pkg/sentry/mm/BUILD +++ b/pkg/sentry/mm/BUILD @@ -27,7 +27,7 @@ go_template_instance( "minDegree": "8", }, imports = { - "usermem": "gvisor.dev/gvisor/pkg/sentry/usermem", + "usermem": "gvisor.dev/gvisor/pkg/usermem", }, package = "mm", prefix = "vma", @@ -47,7 +47,7 @@ go_template_instance( "minDegree": "8", }, imports = { - "usermem": "gvisor.dev/gvisor/pkg/sentry/usermem", + "usermem": "gvisor.dev/gvisor/pkg/usermem", }, package = "mm", prefix = "pma", @@ -99,10 +99,12 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/atomicbitops", + "//pkg/context", "//pkg/log", "//pkg/refs", + "//pkg/safecopy", + "//pkg/safemem", "//pkg/sentry/arch", - "//pkg/sentry/context", "//pkg/sentry/fs", "//pkg/sentry/fs/proc/seqfile", "//pkg/sentry/kernel/auth", @@ -112,13 +114,11 @@ go_library( "//pkg/sentry/memmap", "//pkg/sentry/pgalloc", "//pkg/sentry/platform", - "//pkg/sentry/platform/safecopy", - "//pkg/sentry/safemem", "//pkg/sentry/usage", - "//pkg/sentry/usermem", "//pkg/sync", "//pkg/syserror", "//pkg/tcpip/buffer", + "//pkg/usermem", ], ) @@ -128,14 +128,14 @@ go_test( srcs = ["mm_test.go"], library = ":mm", deps = [ + "//pkg/context", "//pkg/sentry/arch", - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", + "//pkg/sentry/contexttest", "//pkg/sentry/limits", "//pkg/sentry/memmap", "//pkg/sentry/pgalloc", "//pkg/sentry/platform", - "//pkg/sentry/usermem", "//pkg/syserror", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/mm/address_space.go b/pkg/sentry/mm/address_space.go index cfebcfd42..e58a63deb 100644 --- a/pkg/sentry/mm/address_space.go +++ b/pkg/sentry/mm/address_space.go @@ -20,7 +20,7 @@ import ( "gvisor.dev/gvisor/pkg/atomicbitops" "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // AddressSpace returns the platform.AddressSpace bound to mm. diff --git a/pkg/sentry/mm/aio_context.go b/pkg/sentry/mm/aio_context.go index 4b48866ad..cb29d94b0 100644 --- a/pkg/sentry/mm/aio_context.go +++ b/pkg/sentry/mm/aio_context.go @@ -16,15 +16,15 @@ package mm import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // aioManager creates and manages asynchronous I/O contexts. diff --git a/pkg/sentry/mm/debug.go b/pkg/sentry/mm/debug.go index df9adf708..c273c982e 100644 --- a/pkg/sentry/mm/debug.go +++ b/pkg/sentry/mm/debug.go @@ -18,7 +18,7 @@ import ( "bytes" "fmt" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" ) const ( diff --git a/pkg/sentry/mm/io.go b/pkg/sentry/mm/io.go index b03e7d020..fa776f9c6 100644 --- a/pkg/sentry/mm/io.go +++ b/pkg/sentry/mm/io.go @@ -15,11 +15,11 @@ package mm import ( - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/safemem" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // There are two supported ways to copy data to/from application virtual diff --git a/pkg/sentry/mm/lifecycle.go b/pkg/sentry/mm/lifecycle.go index 4e9ca1de6..47b8fbf43 100644 --- a/pkg/sentry/mm/lifecycle.go +++ b/pkg/sentry/mm/lifecycle.go @@ -19,13 +19,13 @@ import ( "sync/atomic" "gvisor.dev/gvisor/pkg/atomicbitops" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // NewMemoryManager returns a new MemoryManager with no mappings and 1 user. diff --git a/pkg/sentry/mm/metadata.go b/pkg/sentry/mm/metadata.go index d2a01d48a..f550acae0 100644 --- a/pkg/sentry/mm/metadata.go +++ b/pkg/sentry/mm/metadata.go @@ -17,7 +17,7 @@ package mm import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // Dumpability describes if and how core dumps should be created. diff --git a/pkg/sentry/mm/mm.go b/pkg/sentry/mm/mm.go index 78cc9e6e4..09e582dd3 100644 --- a/pkg/sentry/mm/mm.go +++ b/pkg/sentry/mm/mm.go @@ -35,14 +35,14 @@ package mm import ( + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/safemem" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/usermem" ) // MemoryManager implements a virtual address space. diff --git a/pkg/sentry/mm/mm_test.go b/pkg/sentry/mm/mm_test.go index 4d2bfaaed..edacca741 100644 --- a/pkg/sentry/mm/mm_test.go +++ b/pkg/sentry/mm/mm_test.go @@ -17,15 +17,15 @@ package mm import ( "testing" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) func testMemoryManager(ctx context.Context) *MemoryManager { diff --git a/pkg/sentry/mm/pma.go b/pkg/sentry/mm/pma.go index c976c6f45..62e4c20af 100644 --- a/pkg/sentry/mm/pma.go +++ b/pkg/sentry/mm/pma.go @@ -17,14 +17,14 @@ package mm import ( "fmt" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/safecopy" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/platform/safecopy" - "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // existingPMAsLocked checks that pmas exist for all addresses in ar, and diff --git a/pkg/sentry/mm/procfs.go b/pkg/sentry/mm/procfs.go index 79610acb7..1ab92f046 100644 --- a/pkg/sentry/mm/procfs.go +++ b/pkg/sentry/mm/procfs.go @@ -19,10 +19,10 @@ import ( "fmt" "strings" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs/proc/seqfile" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) const ( diff --git a/pkg/sentry/mm/save_restore.go b/pkg/sentry/mm/save_restore.go index 93259c5a3..f56215d9a 100644 --- a/pkg/sentry/mm/save_restore.go +++ b/pkg/sentry/mm/save_restore.go @@ -17,7 +17,7 @@ package mm import ( "fmt" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" ) // InvalidateUnsavable invokes memmap.Mappable.InvalidateUnsavable on all diff --git a/pkg/sentry/mm/shm.go b/pkg/sentry/mm/shm.go index b9f2d23e5..6432731d4 100644 --- a/pkg/sentry/mm/shm.go +++ b/pkg/sentry/mm/shm.go @@ -15,10 +15,10 @@ package mm import ( - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/kernel/shm" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // DetachShm unmaps a sysv shared memory segment. diff --git a/pkg/sentry/mm/special_mappable.go b/pkg/sentry/mm/special_mappable.go index ea2d7af74..9ad52082d 100644 --- a/pkg/sentry/mm/special_mappable.go +++ b/pkg/sentry/mm/special_mappable.go @@ -15,14 +15,14 @@ package mm import ( + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // SpecialMappable implements memmap.MappingIdentity and memmap.Mappable with diff --git a/pkg/sentry/mm/syscalls.go b/pkg/sentry/mm/syscalls.go index c2466c988..c5dfa5972 100644 --- a/pkg/sentry/mm/syscalls.go +++ b/pkg/sentry/mm/syscalls.go @@ -19,14 +19,14 @@ import ( mrand "math/rand" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/futex" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // HandleUserFault handles an application page fault. sp is the faulting diff --git a/pkg/sentry/mm/vma.go b/pkg/sentry/mm/vma.go index f2fd70799..9a14e69e6 100644 --- a/pkg/sentry/mm/vma.go +++ b/pkg/sentry/mm/vma.go @@ -18,13 +18,13 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // Preconditions: mm.mappingMu must be locked for writing. opts must be valid diff --git a/pkg/sentry/pgalloc/BUILD b/pkg/sentry/pgalloc/BUILD index 02385a3ce..1eeb9f317 100644 --- a/pkg/sentry/pgalloc/BUILD +++ b/pkg/sentry/pgalloc/BUILD @@ -61,18 +61,18 @@ go_library( ], visibility = ["//pkg/sentry:internal"], deps = [ + "//pkg/context", "//pkg/log", "//pkg/memutil", + "//pkg/safemem", "//pkg/sentry/arch", - "//pkg/sentry/context", "//pkg/sentry/hostmm", "//pkg/sentry/platform", - "//pkg/sentry/safemem", "//pkg/sentry/usage", - "//pkg/sentry/usermem", "//pkg/state", "//pkg/sync", "//pkg/syserror", + "//pkg/usermem", ], ) @@ -81,5 +81,5 @@ go_test( size = "small", srcs = ["pgalloc_test.go"], library = ":pgalloc", - deps = ["//pkg/sentry/usermem"], + deps = ["//pkg/usermem"], ) diff --git a/pkg/sentry/pgalloc/context.go b/pkg/sentry/pgalloc/context.go index 11ccf897b..d25215418 100644 --- a/pkg/sentry/pgalloc/context.go +++ b/pkg/sentry/pgalloc/context.go @@ -15,7 +15,7 @@ package pgalloc import ( - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" ) // contextID is this package's type for context.Context.Value keys. diff --git a/pkg/sentry/pgalloc/pgalloc.go b/pkg/sentry/pgalloc/pgalloc.go index c99e023d9..577e9306a 100644 --- a/pkg/sentry/pgalloc/pgalloc.go +++ b/pkg/sentry/pgalloc/pgalloc.go @@ -29,15 +29,15 @@ import ( "syscall" "time" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/hostmm" "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // MemoryFile is a platform.File whose pages may be allocated to arbitrary diff --git a/pkg/sentry/pgalloc/pgalloc_test.go b/pkg/sentry/pgalloc/pgalloc_test.go index 428e6a859..293f22c6b 100644 --- a/pkg/sentry/pgalloc/pgalloc_test.go +++ b/pkg/sentry/pgalloc/pgalloc_test.go @@ -17,7 +17,7 @@ package pgalloc import ( "testing" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) const ( diff --git a/pkg/sentry/pgalloc/save_restore.go b/pkg/sentry/pgalloc/save_restore.go index aafce1d00..f8385c146 100644 --- a/pkg/sentry/pgalloc/save_restore.go +++ b/pkg/sentry/pgalloc/save_restore.go @@ -25,8 +25,8 @@ import ( "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/state" + "gvisor.dev/gvisor/pkg/usermem" ) // SaveTo writes f's state to the given stream. diff --git a/pkg/sentry/platform/BUILD b/pkg/sentry/platform/BUILD index 006450b2d..453241eca 100644 --- a/pkg/sentry/platform/BUILD +++ b/pkg/sentry/platform/BUILD @@ -26,14 +26,14 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/atomicbitops", + "//pkg/context", "//pkg/log", + "//pkg/safecopy", + "//pkg/safemem", "//pkg/seccomp", "//pkg/sentry/arch", - "//pkg/sentry/context", - "//pkg/sentry/platform/safecopy", - "//pkg/sentry/safemem", "//pkg/sentry/usage", - "//pkg/sentry/usermem", "//pkg/syserror", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/platform/context.go b/pkg/sentry/platform/context.go index e29bc4485..6759cda65 100644 --- a/pkg/sentry/platform/context.go +++ b/pkg/sentry/platform/context.go @@ -15,7 +15,7 @@ package platform import ( - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" ) // contextID is the auth package's type for context.Context.Value keys. diff --git a/pkg/sentry/platform/kvm/BUILD b/pkg/sentry/platform/kvm/BUILD index a4532a766..159f7eafd 100644 --- a/pkg/sentry/platform/kvm/BUILD +++ b/pkg/sentry/platform/kvm/BUILD @@ -44,16 +44,16 @@ go_library( "//pkg/cpuid", "//pkg/log", "//pkg/procid", + "//pkg/safecopy", "//pkg/seccomp", "//pkg/sentry/arch", "//pkg/sentry/platform", "//pkg/sentry/platform/interrupt", "//pkg/sentry/platform/ring0", "//pkg/sentry/platform/ring0/pagetables", - "//pkg/sentry/platform/safecopy", "//pkg/sentry/time", - "//pkg/sentry/usermem", "//pkg/sync", + "//pkg/usermem", ], ) @@ -75,6 +75,6 @@ go_test( "//pkg/sentry/platform/kvm/testutil", "//pkg/sentry/platform/ring0", "//pkg/sentry/platform/ring0/pagetables", - "//pkg/sentry/usermem", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/platform/kvm/address_space.go b/pkg/sentry/platform/kvm/address_space.go index a25f3c449..be213bfe8 100644 --- a/pkg/sentry/platform/kvm/address_space.go +++ b/pkg/sentry/platform/kvm/address_space.go @@ -20,8 +20,8 @@ import ( "gvisor.dev/gvisor/pkg/atomicbitops" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/usermem" ) // dirtySet tracks vCPUs for invalidation. diff --git a/pkg/sentry/platform/kvm/bluepill.go b/pkg/sentry/platform/kvm/bluepill.go index 30dbb74d6..35cd55fef 100644 --- a/pkg/sentry/platform/kvm/bluepill.go +++ b/pkg/sentry/platform/kvm/bluepill.go @@ -19,9 +19,9 @@ import ( "reflect" "syscall" + "gvisor.dev/gvisor/pkg/safecopy" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/platform/ring0" - "gvisor.dev/gvisor/pkg/sentry/platform/safecopy" ) // bluepill enters guest mode. diff --git a/pkg/sentry/platform/kvm/bluepill_fault.go b/pkg/sentry/platform/kvm/bluepill_fault.go index f6459cda9..e34f46aeb 100644 --- a/pkg/sentry/platform/kvm/bluepill_fault.go +++ b/pkg/sentry/platform/kvm/bluepill_fault.go @@ -18,7 +18,7 @@ import ( "sync/atomic" "syscall" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) const ( diff --git a/pkg/sentry/platform/kvm/context.go b/pkg/sentry/platform/kvm/context.go index 99450d22d..c769ac7b4 100644 --- a/pkg/sentry/platform/kvm/context.go +++ b/pkg/sentry/platform/kvm/context.go @@ -19,7 +19,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/interrupt" "gvisor.dev/gvisor/pkg/sentry/platform/ring0" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // context is an implementation of the platform context. diff --git a/pkg/sentry/platform/kvm/kvm.go b/pkg/sentry/platform/kvm/kvm.go index d337c5c7c..972ba85c3 100644 --- a/pkg/sentry/platform/kvm/kvm.go +++ b/pkg/sentry/platform/kvm/kvm.go @@ -23,8 +23,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/ring0" "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/usermem" ) // KVM represents a lightweight VM context. diff --git a/pkg/sentry/platform/kvm/kvm_test.go b/pkg/sentry/platform/kvm/kvm_test.go index 30df725d4..c42752d50 100644 --- a/pkg/sentry/platform/kvm/kvm_test.go +++ b/pkg/sentry/platform/kvm/kvm_test.go @@ -27,7 +27,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/platform/kvm/testutil" "gvisor.dev/gvisor/pkg/sentry/platform/ring0" "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) var dummyFPState = (*byte)(arch.NewFloatingPointData()) diff --git a/pkg/sentry/platform/kvm/machine.go b/pkg/sentry/platform/kvm/machine.go index e6d912168..8076c7529 100644 --- a/pkg/sentry/platform/kvm/machine.go +++ b/pkg/sentry/platform/kvm/machine.go @@ -25,8 +25,8 @@ import ( "gvisor.dev/gvisor/pkg/procid" "gvisor.dev/gvisor/pkg/sentry/platform/ring0" "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/usermem" ) // machine contains state associated with the VM as a whole. diff --git a/pkg/sentry/platform/kvm/machine_amd64.go b/pkg/sentry/platform/kvm/machine_amd64.go index 873e39dc7..923ce3909 100644 --- a/pkg/sentry/platform/kvm/machine_amd64.go +++ b/pkg/sentry/platform/kvm/machine_amd64.go @@ -26,7 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/ring0" "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // initArchState initializes architecture-specific state. diff --git a/pkg/sentry/platform/kvm/machine_arm64.go b/pkg/sentry/platform/kvm/machine_arm64.go index 3b1f20219..09552837a 100644 --- a/pkg/sentry/platform/kvm/machine_arm64.go +++ b/pkg/sentry/platform/kvm/machine_arm64.go @@ -20,7 +20,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) type vCPUArchState struct { diff --git a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go index 3f2f97a6b..1c8384e6b 100644 --- a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go @@ -26,7 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/ring0" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // setMemoryRegion initializes a region. diff --git a/pkg/sentry/platform/kvm/physical_map.go b/pkg/sentry/platform/kvm/physical_map.go index 91de5dab1..f7fa2f98d 100644 --- a/pkg/sentry/platform/kvm/physical_map.go +++ b/pkg/sentry/platform/kvm/physical_map.go @@ -21,7 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/platform/ring0" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) type region struct { diff --git a/pkg/sentry/platform/kvm/virtual_map.go b/pkg/sentry/platform/kvm/virtual_map.go index 2d68855ef..c8897d34f 100644 --- a/pkg/sentry/platform/kvm/virtual_map.go +++ b/pkg/sentry/platform/kvm/virtual_map.go @@ -22,7 +22,7 @@ import ( "regexp" "strconv" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) type virtualRegion struct { diff --git a/pkg/sentry/platform/kvm/virtual_map_test.go b/pkg/sentry/platform/kvm/virtual_map_test.go index 6a2f145be..327e2be4f 100644 --- a/pkg/sentry/platform/kvm/virtual_map_test.go +++ b/pkg/sentry/platform/kvm/virtual_map_test.go @@ -18,7 +18,7 @@ import ( "syscall" "testing" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) type checker struct { diff --git a/pkg/sentry/platform/mmap_min_addr.go b/pkg/sentry/platform/mmap_min_addr.go index 999787462..091c2e365 100644 --- a/pkg/sentry/platform/mmap_min_addr.go +++ b/pkg/sentry/platform/mmap_min_addr.go @@ -20,7 +20,7 @@ import ( "strconv" "strings" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // systemMMapMinAddrSource is the source file. diff --git a/pkg/sentry/platform/platform.go b/pkg/sentry/platform/platform.go index ec22dbf87..2ca696382 100644 --- a/pkg/sentry/platform/platform.go +++ b/pkg/sentry/platform/platform.go @@ -22,10 +22,10 @@ import ( "os" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/seccomp" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/safemem" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // Platform provides abstractions for execution contexts (Context, diff --git a/pkg/sentry/platform/ptrace/BUILD b/pkg/sentry/platform/ptrace/BUILD index 3bcc5e040..95abd321e 100644 --- a/pkg/sentry/platform/ptrace/BUILD +++ b/pkg/sentry/platform/ptrace/BUILD @@ -25,14 +25,14 @@ go_library( "//pkg/abi/linux", "//pkg/log", "//pkg/procid", + "//pkg/safecopy", "//pkg/seccomp", "//pkg/sentry/arch", "//pkg/sentry/hostcpu", "//pkg/sentry/platform", "//pkg/sentry/platform/interrupt", - "//pkg/sentry/platform/safecopy", - "//pkg/sentry/usermem", "//pkg/sync", + "//pkg/usermem", "@org_golang_x_sys//unix:go_default_library", ], ) diff --git a/pkg/sentry/platform/ptrace/ptrace.go b/pkg/sentry/platform/ptrace/ptrace.go index bb0e03880..03adb624b 100644 --- a/pkg/sentry/platform/ptrace/ptrace.go +++ b/pkg/sentry/platform/ptrace/ptrace.go @@ -51,8 +51,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/interrupt" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/usermem" ) var ( diff --git a/pkg/sentry/platform/ptrace/ptrace_unsafe.go b/pkg/sentry/platform/ptrace/ptrace_unsafe.go index 72c7ec564..6c0ed7b3e 100644 --- a/pkg/sentry/platform/ptrace/ptrace_unsafe.go +++ b/pkg/sentry/platform/ptrace/ptrace_unsafe.go @@ -20,7 +20,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // getRegs gets the general purpose register set. diff --git a/pkg/sentry/platform/ptrace/stub_unsafe.go b/pkg/sentry/platform/ptrace/stub_unsafe.go index aa1b87237..341dde143 100644 --- a/pkg/sentry/platform/ptrace/stub_unsafe.go +++ b/pkg/sentry/platform/ptrace/stub_unsafe.go @@ -19,8 +19,8 @@ import ( "syscall" "unsafe" - "gvisor.dev/gvisor/pkg/sentry/platform/safecopy" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/safecopy" + "gvisor.dev/gvisor/pkg/usermem" ) // stub is defined in arch-specific assembly. diff --git a/pkg/sentry/platform/ptrace/subprocess.go b/pkg/sentry/platform/ptrace/subprocess.go index 15dc46a5b..31b7cec53 100644 --- a/pkg/sentry/platform/ptrace/subprocess.go +++ b/pkg/sentry/platform/ptrace/subprocess.go @@ -25,8 +25,8 @@ import ( "gvisor.dev/gvisor/pkg/procid" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/usermem" ) // Linux kernel errnos which "should never be seen by user programs", but will diff --git a/pkg/sentry/platform/ring0/BUILD b/pkg/sentry/platform/ring0/BUILD index 6dee8fcc5..934b6fbcd 100644 --- a/pkg/sentry/platform/ring0/BUILD +++ b/pkg/sentry/platform/ring0/BUILD @@ -78,6 +78,6 @@ go_library( deps = [ "//pkg/cpuid", "//pkg/sentry/platform/ring0/pagetables", - "//pkg/sentry/usermem", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/platform/ring0/defs_amd64.go b/pkg/sentry/platform/ring0/defs_amd64.go index 9dae0dccb..9c6c2cf5c 100644 --- a/pkg/sentry/platform/ring0/defs_amd64.go +++ b/pkg/sentry/platform/ring0/defs_amd64.go @@ -18,7 +18,7 @@ package ring0 import ( "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) var ( diff --git a/pkg/sentry/platform/ring0/defs_arm64.go b/pkg/sentry/platform/ring0/defs_arm64.go index a850ce6cf..1583dda12 100644 --- a/pkg/sentry/platform/ring0/defs_arm64.go +++ b/pkg/sentry/platform/ring0/defs_arm64.go @@ -18,7 +18,7 @@ package ring0 import ( "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) var ( diff --git a/pkg/sentry/platform/ring0/gen_offsets/BUILD b/pkg/sentry/platform/ring0/gen_offsets/BUILD index 147311ed3..4cae10459 100644 --- a/pkg/sentry/platform/ring0/gen_offsets/BUILD +++ b/pkg/sentry/platform/ring0/gen_offsets/BUILD @@ -28,6 +28,6 @@ go_binary( deps = [ "//pkg/cpuid", "//pkg/sentry/platform/ring0/pagetables", - "//pkg/sentry/usermem", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/platform/ring0/pagetables/BUILD b/pkg/sentry/platform/ring0/pagetables/BUILD index 8b5cdd6c1..971eed7fa 100644 --- a/pkg/sentry/platform/ring0/pagetables/BUILD +++ b/pkg/sentry/platform/ring0/pagetables/BUILD @@ -93,8 +93,8 @@ go_library( "//pkg/sentry/platform/ring0:__subpackages__", ], deps = [ - "//pkg/sentry/usermem", "//pkg/sync", + "//pkg/usermem", ], ) @@ -108,5 +108,5 @@ go_test( "walker_check.go", ], library = ":pagetables", - deps = ["//pkg/sentry/usermem"], + deps = ["//pkg/usermem"], ) diff --git a/pkg/sentry/platform/ring0/pagetables/allocator_unsafe.go b/pkg/sentry/platform/ring0/pagetables/allocator_unsafe.go index a90394a33..d08bfdeb3 100644 --- a/pkg/sentry/platform/ring0/pagetables/allocator_unsafe.go +++ b/pkg/sentry/platform/ring0/pagetables/allocator_unsafe.go @@ -17,7 +17,7 @@ package pagetables import ( "unsafe" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // newAlignedPTEs returns a set of aligned PTEs. diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables.go b/pkg/sentry/platform/ring0/pagetables/pagetables.go index 30c64a372..87e88e97d 100644 --- a/pkg/sentry/platform/ring0/pagetables/pagetables.go +++ b/pkg/sentry/platform/ring0/pagetables/pagetables.go @@ -21,7 +21,7 @@ package pagetables import ( - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // PageTables is a set of page tables. diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64.go b/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64.go index e78424766..78510ebed 100644 --- a/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64.go +++ b/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64.go @@ -19,7 +19,7 @@ package pagetables import ( "sync/atomic" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // archPageTables is architecture-specific data. diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_amd64_test.go b/pkg/sentry/platform/ring0/pagetables/pagetables_amd64_test.go index 35e917526..54e8e554f 100644 --- a/pkg/sentry/platform/ring0/pagetables/pagetables_amd64_test.go +++ b/pkg/sentry/platform/ring0/pagetables/pagetables_amd64_test.go @@ -19,7 +19,7 @@ package pagetables import ( "testing" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) func Test2MAnd4K(t *testing.T) { diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_arm64_test.go b/pkg/sentry/platform/ring0/pagetables/pagetables_arm64_test.go index 254116233..2f73d424f 100644 --- a/pkg/sentry/platform/ring0/pagetables/pagetables_arm64_test.go +++ b/pkg/sentry/platform/ring0/pagetables/pagetables_arm64_test.go @@ -19,7 +19,7 @@ package pagetables import ( "testing" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) func Test2MAnd4K(t *testing.T) { diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_test.go b/pkg/sentry/platform/ring0/pagetables/pagetables_test.go index 6e95ad2b9..5c88d087d 100644 --- a/pkg/sentry/platform/ring0/pagetables/pagetables_test.go +++ b/pkg/sentry/platform/ring0/pagetables/pagetables_test.go @@ -17,7 +17,7 @@ package pagetables import ( "testing" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) type mapping struct { diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_x86.go b/pkg/sentry/platform/ring0/pagetables/pagetables_x86.go index 3e2383c5e..dcf061df9 100644 --- a/pkg/sentry/platform/ring0/pagetables/pagetables_x86.go +++ b/pkg/sentry/platform/ring0/pagetables/pagetables_x86.go @@ -19,7 +19,7 @@ package pagetables import ( "sync/atomic" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // archPageTables is architecture-specific data. diff --git a/pkg/sentry/platform/safecopy/BUILD b/pkg/sentry/platform/safecopy/BUILD deleted file mode 100644 index b8747585b..000000000 --- a/pkg/sentry/platform/safecopy/BUILD +++ /dev/null @@ -1,29 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -package(licenses = ["notice"]) - -go_library( - name = "safecopy", - srcs = [ - "atomic_amd64.s", - "atomic_arm64.s", - "memclr_amd64.s", - "memclr_arm64.s", - "memcpy_amd64.s", - "memcpy_arm64.s", - "safecopy.go", - "safecopy_unsafe.go", - "sighandler_amd64.s", - "sighandler_arm64.s", - ], - visibility = ["//pkg/sentry:internal"], - deps = ["//pkg/syserror"], -) - -go_test( - name = "safecopy_test", - srcs = [ - "safecopy_test.go", - ], - library = ":safecopy", -) diff --git a/pkg/sentry/platform/safecopy/LICENSE b/pkg/sentry/platform/safecopy/LICENSE deleted file mode 100644 index 6a66aea5e..000000000 --- a/pkg/sentry/platform/safecopy/LICENSE +++ /dev/null @@ -1,27 +0,0 @@ -Copyright (c) 2009 The Go Authors. All rights reserved. - -Redistribution and use in source and binary forms, with or without -modification, are permitted provided that the following conditions are -met: - - * Redistributions of source code must retain the above copyright -notice, this list of conditions and the following disclaimer. - * Redistributions in binary form must reproduce the above -copyright notice, this list of conditions and the following disclaimer -in the documentation and/or other materials provided with the -distribution. - * Neither the name of Google Inc. nor the names of its -contributors may be used to endorse or promote products derived from -this software without specific prior written permission. - -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/pkg/sentry/platform/safecopy/atomic_amd64.s b/pkg/sentry/platform/safecopy/atomic_amd64.s deleted file mode 100644 index a0cd78f33..000000000 --- a/pkg/sentry/platform/safecopy/atomic_amd64.s +++ /dev/null @@ -1,136 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -#include "textflag.h" - -// handleSwapUint32Fault returns the value stored in DI. Control is transferred -// to it when swapUint32 below receives SIGSEGV or SIGBUS, with the signal -// number stored in DI. -// -// It must have the same frame configuration as swapUint32 so that it can undo -// any potential call frame set up by the assembler. -TEXT handleSwapUint32Fault(SB), NOSPLIT, $0-24 - MOVL DI, sig+20(FP) - RET - -// swapUint32 atomically stores new into *addr and returns (the previous *addr -// value, 0). If a SIGSEGV or SIGBUS signal is received during the swap, the -// value of old is unspecified, and sig is the number of the signal that was -// received. -// -// Preconditions: addr must be aligned to a 4-byte boundary. -// -//func swapUint32(ptr unsafe.Pointer, new uint32) (old uint32, sig int32) -TEXT ·swapUint32(SB), NOSPLIT, $0-24 - // Store 0 as the returned signal number. If we run to completion, - // this is the value the caller will see; if a signal is received, - // handleSwapUint32Fault will store a different value in this address. - MOVL $0, sig+20(FP) - - MOVQ addr+0(FP), DI - MOVL new+8(FP), AX - XCHGL AX, 0(DI) - MOVL AX, old+16(FP) - RET - -// handleSwapUint64Fault returns the value stored in DI. Control is transferred -// to it when swapUint64 below receives SIGSEGV or SIGBUS, with the signal -// number stored in DI. -// -// It must have the same frame configuration as swapUint64 so that it can undo -// any potential call frame set up by the assembler. -TEXT handleSwapUint64Fault(SB), NOSPLIT, $0-28 - MOVL DI, sig+24(FP) - RET - -// swapUint64 atomically stores new into *addr and returns (the previous *addr -// value, 0). If a SIGSEGV or SIGBUS signal is received during the swap, the -// value of old is unspecified, and sig is the number of the signal that was -// received. -// -// Preconditions: addr must be aligned to a 8-byte boundary. -// -//func swapUint64(ptr unsafe.Pointer, new uint64) (old uint64, sig int32) -TEXT ·swapUint64(SB), NOSPLIT, $0-28 - // Store 0 as the returned signal number. If we run to completion, - // this is the value the caller will see; if a signal is received, - // handleSwapUint64Fault will store a different value in this address. - MOVL $0, sig+24(FP) - - MOVQ addr+0(FP), DI - MOVQ new+8(FP), AX - XCHGQ AX, 0(DI) - MOVQ AX, old+16(FP) - RET - -// handleCompareAndSwapUint32Fault returns the value stored in DI. Control is -// transferred to it when swapUint64 below receives SIGSEGV or SIGBUS, with the -// signal number stored in DI. -// -// It must have the same frame configuration as compareAndSwapUint32 so that it -// can undo any potential call frame set up by the assembler. -TEXT handleCompareAndSwapUint32Fault(SB), NOSPLIT, $0-24 - MOVL DI, sig+20(FP) - RET - -// compareAndSwapUint32 is like sync/atomic.CompareAndSwapUint32, but returns -// (the value previously stored at addr, 0). If a SIGSEGV or SIGBUS signal is -// received during the operation, the value of prev is unspecified, and sig is -// the number of the signal that was received. -// -// Preconditions: addr must be aligned to a 4-byte boundary. -// -//func compareAndSwapUint32(ptr unsafe.Pointer, old, new uint32) (prev uint32, sig int32) -TEXT ·compareAndSwapUint32(SB), NOSPLIT, $0-24 - // Store 0 as the returned signal number. If we run to completion, this is - // the value the caller will see; if a signal is received, - // handleCompareAndSwapUint32Fault will store a different value in this - // address. - MOVL $0, sig+20(FP) - - MOVQ addr+0(FP), DI - MOVL old+8(FP), AX - MOVL new+12(FP), DX - LOCK - CMPXCHGL DX, 0(DI) - MOVL AX, prev+16(FP) - RET - -// handleLoadUint32Fault returns the value stored in DI. Control is transferred -// to it when LoadUint32 below receives SIGSEGV or SIGBUS, with the signal -// number stored in DI. -// -// It must have the same frame configuration as loadUint32 so that it can undo -// any potential call frame set up by the assembler. -TEXT handleLoadUint32Fault(SB), NOSPLIT, $0-16 - MOVL DI, sig+12(FP) - RET - -// loadUint32 atomically loads *addr and returns it. If a SIGSEGV or SIGBUS -// signal is received, the value returned is unspecified, and sig is the number -// of the signal that was received. -// -// Preconditions: addr must be aligned to a 4-byte boundary. -// -//func loadUint32(ptr unsafe.Pointer) (val uint32, sig int32) -TEXT ·loadUint32(SB), NOSPLIT, $0-16 - // Store 0 as the returned signal number. If we run to completion, - // this is the value the caller will see; if a signal is received, - // handleLoadUint32Fault will store a different value in this address. - MOVL $0, sig+12(FP) - - MOVQ addr+0(FP), AX - MOVL (AX), BX - MOVL BX, val+8(FP) - RET diff --git a/pkg/sentry/platform/safecopy/atomic_arm64.s b/pkg/sentry/platform/safecopy/atomic_arm64.s deleted file mode 100644 index d58ed71f7..000000000 --- a/pkg/sentry/platform/safecopy/atomic_arm64.s +++ /dev/null @@ -1,126 +0,0 @@ -// Copyright 2014 The Go Authors. All rights reserved. -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -#include "textflag.h" - -// handleSwapUint32Fault returns the value stored in R1. Control is transferred -// to it when swapUint32 below receives SIGSEGV or SIGBUS, with the signal -// number stored in R1. -// -// It must have the same frame configuration as swapUint32 so that it can undo -// any potential call frame set up by the assembler. -TEXT handleSwapUint32Fault(SB), NOSPLIT, $0-24 - MOVW R1, sig+20(FP) - RET - -// See the corresponding doc in safecopy_unsafe.go -// -// The code is derived from Go source runtime/internal/atomic.Xchg. -// -//func swapUint32(ptr unsafe.Pointer, new uint32) (old uint32, sig int32) -TEXT ·swapUint32(SB), NOSPLIT, $0-24 - // Store 0 as the returned signal number. If we run to completion, - // this is the value the caller will see; if a signal is received, - // handleSwapUint32Fault will store a different value in this address. - MOVW $0, sig+20(FP) -again: - MOVD addr+0(FP), R0 - MOVW new+8(FP), R1 - LDAXRW (R0), R2 - STLXRW R1, (R0), R3 - CBNZ R3, again - MOVW R2, old+16(FP) - RET - -// handleSwapUint64Fault returns the value stored in R1. Control is transferred -// to it when swapUint64 below receives SIGSEGV or SIGBUS, with the signal -// number stored in R1. -// -// It must have the same frame configuration as swapUint64 so that it can undo -// any potential call frame set up by the assembler. -TEXT handleSwapUint64Fault(SB), NOSPLIT, $0-28 - MOVW R1, sig+24(FP) - RET - -// See the corresponding doc in safecopy_unsafe.go -// -// The code is derived from Go source runtime/internal/atomic.Xchg64. -// -//func swapUint64(ptr unsafe.Pointer, new uint64) (old uint64, sig int32) -TEXT ·swapUint64(SB), NOSPLIT, $0-28 - // Store 0 as the returned signal number. If we run to completion, - // this is the value the caller will see; if a signal is received, - // handleSwapUint64Fault will store a different value in this address. - MOVW $0, sig+24(FP) -again: - MOVD addr+0(FP), R0 - MOVD new+8(FP), R1 - LDAXR (R0), R2 - STLXR R1, (R0), R3 - CBNZ R3, again - MOVD R2, old+16(FP) - RET - -// handleCompareAndSwapUint32Fault returns the value stored in R1. Control is -// transferred to it when compareAndSwapUint32 below receives SIGSEGV or SIGBUS, -// with the signal number stored in R1. -// -// It must have the same frame configuration as compareAndSwapUint32 so that it -// can undo any potential call frame set up by the assembler. -TEXT handleCompareAndSwapUint32Fault(SB), NOSPLIT, $0-24 - MOVW R1, sig+20(FP) - RET - -// See the corresponding doc in safecopy_unsafe.go -// -// The code is derived from Go source runtime/internal/atomic.Cas. -// -//func compareAndSwapUint32(ptr unsafe.Pointer, old, new uint32) (prev uint32, sig int32) -TEXT ·compareAndSwapUint32(SB), NOSPLIT, $0-24 - // Store 0 as the returned signal number. If we run to completion, this is - // the value the caller will see; if a signal is received, - // handleCompareAndSwapUint32Fault will store a different value in this - // address. - MOVW $0, sig+20(FP) - - MOVD addr+0(FP), R0 - MOVW old+8(FP), R1 - MOVW new+12(FP), R2 -again: - LDAXRW (R0), R3 - CMPW R1, R3 - BNE done - STLXRW R2, (R0), R4 - CBNZ R4, again -done: - MOVW R3, prev+16(FP) - RET - -// handleLoadUint32Fault returns the value stored in DI. Control is transferred -// to it when LoadUint32 below receives SIGSEGV or SIGBUS, with the signal -// number stored in DI. -// -// It must have the same frame configuration as loadUint32 so that it can undo -// any potential call frame set up by the assembler. -TEXT handleLoadUint32Fault(SB), NOSPLIT, $0-16 - MOVW R1, sig+12(FP) - RET - -// loadUint32 atomically loads *addr and returns it. If a SIGSEGV or SIGBUS -// signal is received, the value returned is unspecified, and sig is the number -// of the signal that was received. -// -// Preconditions: addr must be aligned to a 4-byte boundary. -// -//func loadUint32(ptr unsafe.Pointer) (val uint32, sig int32) -TEXT ·loadUint32(SB), NOSPLIT, $0-16 - // Store 0 as the returned signal number. If we run to completion, - // this is the value the caller will see; if a signal is received, - // handleLoadUint32Fault will store a different value in this address. - MOVW $0, sig+12(FP) - - MOVD addr+0(FP), R0 - LDARW (R0), R1 - MOVW R1, val+8(FP) - RET diff --git a/pkg/sentry/platform/safecopy/memclr_amd64.s b/pkg/sentry/platform/safecopy/memclr_amd64.s deleted file mode 100644 index 64cf32f05..000000000 --- a/pkg/sentry/platform/safecopy/memclr_amd64.s +++ /dev/null @@ -1,147 +0,0 @@ -// Copyright 2014 The Go Authors. All rights reserved. -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -#include "textflag.h" - -// handleMemclrFault returns (the value stored in AX, the value stored in DI). -// Control is transferred to it when memclr below receives SIGSEGV or SIGBUS, -// with the faulting address stored in AX and the signal number stored in DI. -// -// It must have the same frame configuration as memclr so that it can undo any -// potential call frame set up by the assembler. -TEXT handleMemclrFault(SB), NOSPLIT, $0-28 - MOVQ AX, addr+16(FP) - MOVL DI, sig+24(FP) - RET - -// memclr sets the n bytes following ptr to zeroes. If a SIGSEGV or SIGBUS -// signal is received during the write, it returns the address that caused the -// fault and the number of the signal that was received. Otherwise, it returns -// an unspecified address and a signal number of 0. -// -// Data is written in order, such that if a fault happens at address p, it is -// safe to assume that all data before p-maxRegisterSize has already been -// successfully written. -// -// The code is derived from runtime.memclrNoHeapPointers. -// -// func memclr(ptr unsafe.Pointer, n uintptr) (fault unsafe.Pointer, sig int32) -TEXT ·memclr(SB), NOSPLIT, $0-28 - // Store 0 as the returned signal number. If we run to completion, - // this is the value the caller will see; if a signal is received, - // handleMemclrFault will store a different value in this address. - MOVL $0, sig+24(FP) - - MOVQ ptr+0(FP), DI - MOVQ n+8(FP), BX - XORQ AX, AX - - // MOVOU seems always faster than REP STOSQ. -tail: - TESTQ BX, BX - JEQ _0 - CMPQ BX, $2 - JBE _1or2 - CMPQ BX, $4 - JBE _3or4 - CMPQ BX, $8 - JB _5through7 - JE _8 - CMPQ BX, $16 - JBE _9through16 - PXOR X0, X0 - CMPQ BX, $32 - JBE _17through32 - CMPQ BX, $64 - JBE _33through64 - CMPQ BX, $128 - JBE _65through128 - CMPQ BX, $256 - JBE _129through256 - // TODO: use branch table and BSR to make this just a single dispatch - // TODO: for really big clears, use MOVNTDQ, even without AVX2. - -loop: - MOVOU X0, 0(DI) - MOVOU X0, 16(DI) - MOVOU X0, 32(DI) - MOVOU X0, 48(DI) - MOVOU X0, 64(DI) - MOVOU X0, 80(DI) - MOVOU X0, 96(DI) - MOVOU X0, 112(DI) - MOVOU X0, 128(DI) - MOVOU X0, 144(DI) - MOVOU X0, 160(DI) - MOVOU X0, 176(DI) - MOVOU X0, 192(DI) - MOVOU X0, 208(DI) - MOVOU X0, 224(DI) - MOVOU X0, 240(DI) - SUBQ $256, BX - ADDQ $256, DI - CMPQ BX, $256 - JAE loop - JMP tail - -_1or2: - MOVB AX, (DI) - MOVB AX, -1(DI)(BX*1) - RET -_0: - RET -_3or4: - MOVW AX, (DI) - MOVW AX, -2(DI)(BX*1) - RET -_5through7: - MOVL AX, (DI) - MOVL AX, -4(DI)(BX*1) - RET -_8: - // We need a separate case for 8 to make sure we clear pointers atomically. - MOVQ AX, (DI) - RET -_9through16: - MOVQ AX, (DI) - MOVQ AX, -8(DI)(BX*1) - RET -_17through32: - MOVOU X0, (DI) - MOVOU X0, -16(DI)(BX*1) - RET -_33through64: - MOVOU X0, (DI) - MOVOU X0, 16(DI) - MOVOU X0, -32(DI)(BX*1) - MOVOU X0, -16(DI)(BX*1) - RET -_65through128: - MOVOU X0, (DI) - MOVOU X0, 16(DI) - MOVOU X0, 32(DI) - MOVOU X0, 48(DI) - MOVOU X0, -64(DI)(BX*1) - MOVOU X0, -48(DI)(BX*1) - MOVOU X0, -32(DI)(BX*1) - MOVOU X0, -16(DI)(BX*1) - RET -_129through256: - MOVOU X0, (DI) - MOVOU X0, 16(DI) - MOVOU X0, 32(DI) - MOVOU X0, 48(DI) - MOVOU X0, 64(DI) - MOVOU X0, 80(DI) - MOVOU X0, 96(DI) - MOVOU X0, 112(DI) - MOVOU X0, -128(DI)(BX*1) - MOVOU X0, -112(DI)(BX*1) - MOVOU X0, -96(DI)(BX*1) - MOVOU X0, -80(DI)(BX*1) - MOVOU X0, -64(DI)(BX*1) - MOVOU X0, -48(DI)(BX*1) - MOVOU X0, -32(DI)(BX*1) - MOVOU X0, -16(DI)(BX*1) - RET diff --git a/pkg/sentry/platform/safecopy/memclr_arm64.s b/pkg/sentry/platform/safecopy/memclr_arm64.s deleted file mode 100644 index 7361b9067..000000000 --- a/pkg/sentry/platform/safecopy/memclr_arm64.s +++ /dev/null @@ -1,74 +0,0 @@ -// Copyright 2014 The Go Authors. All rights reserved. -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -#include "textflag.h" - -// handleMemclrFault returns (the value stored in R0, the value stored in R1). -// Control is transferred to it when memclr below receives SIGSEGV or SIGBUS, -// with the faulting address stored in R0 and the signal number stored in R1. -// -// It must have the same frame configuration as memclr so that it can undo any -// potential call frame set up by the assembler. -TEXT handleMemclrFault(SB), NOSPLIT, $0-28 - MOVD R0, addr+16(FP) - MOVW R1, sig+24(FP) - RET - -// See the corresponding doc in safecopy_unsafe.go -// -// The code is derived from runtime.memclrNoHeapPointers. -// -// func memclr(ptr unsafe.Pointer, n uintptr) (fault unsafe.Pointer, sig int32) -TEXT ·memclr(SB), NOSPLIT, $0-28 - // Store 0 as the returned signal number. If we run to completion, - // this is the value the caller will see; if a signal is received, - // handleMemclrFault will store a different value in this address. - MOVW $0, sig+24(FP) - MOVD ptr+0(FP), R0 - MOVD n+8(FP), R1 - - // If size is less than 16 bytes, use tail_zero to zero what remains - CMP $16, R1 - BLT tail_zero - // Get buffer offset into 16 byte aligned address for better performance - ANDS $15, R0, ZR - BNE unaligned_to_16 -aligned_to_16: - LSR $4, R1, R2 -zero_by_16: - STP.P (ZR, ZR), 16(R0) // Store pair with post index. - SUBS $1, R2, R2 - BNE zero_by_16 - ANDS $15, R1, R1 - BEQ end - - // Zero buffer with size=R1 < 16 -tail_zero: - TBZ $3, R1, tail_zero_4 - MOVD.P ZR, 8(R0) -tail_zero_4: - TBZ $2, R1, tail_zero_2 - MOVW.P ZR, 4(R0) -tail_zero_2: - TBZ $1, R1, tail_zero_1 - MOVH.P ZR, 2(R0) -tail_zero_1: - TBZ $0, R1, end - MOVB ZR, (R0) -end: - RET - -unaligned_to_16: - MOVD R0, R2 -head_loop: - MOVBU.P ZR, 1(R0) - ANDS $15, R0, ZR - BNE head_loop - // Adjust length for what remains - SUB R2, R0, R3 - SUB R3, R1 - // If size is less than 16 bytes, use tail_zero to zero what remains - CMP $16, R1 - BLT tail_zero - B aligned_to_16 diff --git a/pkg/sentry/platform/safecopy/memcpy_amd64.s b/pkg/sentry/platform/safecopy/memcpy_amd64.s deleted file mode 100644 index 129691d68..000000000 --- a/pkg/sentry/platform/safecopy/memcpy_amd64.s +++ /dev/null @@ -1,250 +0,0 @@ -// Copyright © 1994-1999 Lucent Technologies Inc. All rights reserved. -// Revisions Copyright © 2000-2007 Vita Nuova Holdings Limited (www.vitanuova.com). All rights reserved. -// Portions Copyright 2009 The Go Authors. All rights reserved. -// -// Permission is hereby granted, free of charge, to any person obtaining a copy -// of this software and associated documentation files (the "Software"), to deal -// in the Software without restriction, including without limitation the rights -// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell -// copies of the Software, and to permit persons to whom the Software is -// furnished to do so, subject to the following conditions: -// -// The above copyright notice and this permission notice shall be included in -// all copies or substantial portions of the Software. -// -// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR -// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, -// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE -// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER -// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, -// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN -// THE SOFTWARE. - -#include "textflag.h" - -// handleMemcpyFault returns (the value stored in AX, the value stored in DI). -// Control is transferred to it when memcpy below receives SIGSEGV or SIGBUS, -// with the faulting address stored in AX and the signal number stored in DI. -// -// It must have the same frame configuration as memcpy so that it can undo any -// potential call frame set up by the assembler. -TEXT handleMemcpyFault(SB), NOSPLIT, $0-36 - MOVQ AX, addr+24(FP) - MOVL DI, sig+32(FP) - RET - -// memcpy copies data from src to dst. If a SIGSEGV or SIGBUS signal is received -// during the copy, it returns the address that caused the fault and the number -// of the signal that was received. Otherwise, it returns an unspecified address -// and a signal number of 0. -// -// Data is copied in order, such that if a fault happens at address p, it is -// safe to assume that all data before p-maxRegisterSize has already been -// successfully copied. -// -// The code is derived from the forward copying part of runtime.memmove. -// -// func memcpy(dst, src unsafe.Pointer, n uintptr) (fault unsafe.Pointer, sig int32) -TEXT ·memcpy(SB), NOSPLIT, $0-36 - // Store 0 as the returned signal number. If we run to completion, - // this is the value the caller will see; if a signal is received, - // handleMemcpyFault will store a different value in this address. - MOVL $0, sig+32(FP) - - MOVQ to+0(FP), DI - MOVQ from+8(FP), SI - MOVQ n+16(FP), BX - - // REP instructions have a high startup cost, so we handle small sizes - // with some straightline code. The REP MOVSQ instruction is really fast - // for large sizes. The cutover is approximately 2K. -tail: - // move_129through256 or smaller work whether or not the source and the - // destination memory regions overlap because they load all data into - // registers before writing it back. move_256through2048 on the other - // hand can be used only when the memory regions don't overlap or the copy - // direction is forward. - TESTQ BX, BX - JEQ move_0 - CMPQ BX, $2 - JBE move_1or2 - CMPQ BX, $4 - JBE move_3or4 - CMPQ BX, $8 - JB move_5through7 - JE move_8 - CMPQ BX, $16 - JBE move_9through16 - CMPQ BX, $32 - JBE move_17through32 - CMPQ BX, $64 - JBE move_33through64 - CMPQ BX, $128 - JBE move_65through128 - CMPQ BX, $256 - JBE move_129through256 - // TODO: use branch table and BSR to make this just a single dispatch - -/* - * forward copy loop - */ - CMPQ BX, $2048 - JLS move_256through2048 - - // Check alignment - MOVL SI, AX - ORL DI, AX - TESTL $7, AX - JEQ fwdBy8 - - // Do 1 byte at a time - MOVQ BX, CX - REP; MOVSB - RET - -fwdBy8: - // Do 8 bytes at a time - MOVQ BX, CX - SHRQ $3, CX - ANDQ $7, BX - REP; MOVSQ - JMP tail - -move_1or2: - MOVB (SI), AX - MOVB AX, (DI) - MOVB -1(SI)(BX*1), CX - MOVB CX, -1(DI)(BX*1) - RET -move_0: - RET -move_3or4: - MOVW (SI), AX - MOVW AX, (DI) - MOVW -2(SI)(BX*1), CX - MOVW CX, -2(DI)(BX*1) - RET -move_5through7: - MOVL (SI), AX - MOVL AX, (DI) - MOVL -4(SI)(BX*1), CX - MOVL CX, -4(DI)(BX*1) - RET -move_8: - // We need a separate case for 8 to make sure we write pointers atomically. - MOVQ (SI), AX - MOVQ AX, (DI) - RET -move_9through16: - MOVQ (SI), AX - MOVQ AX, (DI) - MOVQ -8(SI)(BX*1), CX - MOVQ CX, -8(DI)(BX*1) - RET -move_17through32: - MOVOU (SI), X0 - MOVOU X0, (DI) - MOVOU -16(SI)(BX*1), X1 - MOVOU X1, -16(DI)(BX*1) - RET -move_33through64: - MOVOU (SI), X0 - MOVOU X0, (DI) - MOVOU 16(SI), X1 - MOVOU X1, 16(DI) - MOVOU -32(SI)(BX*1), X2 - MOVOU X2, -32(DI)(BX*1) - MOVOU -16(SI)(BX*1), X3 - MOVOU X3, -16(DI)(BX*1) - RET -move_65through128: - MOVOU (SI), X0 - MOVOU X0, (DI) - MOVOU 16(SI), X1 - MOVOU X1, 16(DI) - MOVOU 32(SI), X2 - MOVOU X2, 32(DI) - MOVOU 48(SI), X3 - MOVOU X3, 48(DI) - MOVOU -64(SI)(BX*1), X4 - MOVOU X4, -64(DI)(BX*1) - MOVOU -48(SI)(BX*1), X5 - MOVOU X5, -48(DI)(BX*1) - MOVOU -32(SI)(BX*1), X6 - MOVOU X6, -32(DI)(BX*1) - MOVOU -16(SI)(BX*1), X7 - MOVOU X7, -16(DI)(BX*1) - RET -move_129through256: - MOVOU (SI), X0 - MOVOU X0, (DI) - MOVOU 16(SI), X1 - MOVOU X1, 16(DI) - MOVOU 32(SI), X2 - MOVOU X2, 32(DI) - MOVOU 48(SI), X3 - MOVOU X3, 48(DI) - MOVOU 64(SI), X4 - MOVOU X4, 64(DI) - MOVOU 80(SI), X5 - MOVOU X5, 80(DI) - MOVOU 96(SI), X6 - MOVOU X6, 96(DI) - MOVOU 112(SI), X7 - MOVOU X7, 112(DI) - MOVOU -128(SI)(BX*1), X8 - MOVOU X8, -128(DI)(BX*1) - MOVOU -112(SI)(BX*1), X9 - MOVOU X9, -112(DI)(BX*1) - MOVOU -96(SI)(BX*1), X10 - MOVOU X10, -96(DI)(BX*1) - MOVOU -80(SI)(BX*1), X11 - MOVOU X11, -80(DI)(BX*1) - MOVOU -64(SI)(BX*1), X12 - MOVOU X12, -64(DI)(BX*1) - MOVOU -48(SI)(BX*1), X13 - MOVOU X13, -48(DI)(BX*1) - MOVOU -32(SI)(BX*1), X14 - MOVOU X14, -32(DI)(BX*1) - MOVOU -16(SI)(BX*1), X15 - MOVOU X15, -16(DI)(BX*1) - RET -move_256through2048: - SUBQ $256, BX - MOVOU (SI), X0 - MOVOU X0, (DI) - MOVOU 16(SI), X1 - MOVOU X1, 16(DI) - MOVOU 32(SI), X2 - MOVOU X2, 32(DI) - MOVOU 48(SI), X3 - MOVOU X3, 48(DI) - MOVOU 64(SI), X4 - MOVOU X4, 64(DI) - MOVOU 80(SI), X5 - MOVOU X5, 80(DI) - MOVOU 96(SI), X6 - MOVOU X6, 96(DI) - MOVOU 112(SI), X7 - MOVOU X7, 112(DI) - MOVOU 128(SI), X8 - MOVOU X8, 128(DI) - MOVOU 144(SI), X9 - MOVOU X9, 144(DI) - MOVOU 160(SI), X10 - MOVOU X10, 160(DI) - MOVOU 176(SI), X11 - MOVOU X11, 176(DI) - MOVOU 192(SI), X12 - MOVOU X12, 192(DI) - MOVOU 208(SI), X13 - MOVOU X13, 208(DI) - MOVOU 224(SI), X14 - MOVOU X14, 224(DI) - MOVOU 240(SI), X15 - MOVOU X15, 240(DI) - CMPQ BX, $256 - LEAQ 256(SI), SI - LEAQ 256(DI), DI - JGE move_256through2048 - JMP tail diff --git a/pkg/sentry/platform/safecopy/memcpy_arm64.s b/pkg/sentry/platform/safecopy/memcpy_arm64.s deleted file mode 100644 index e7e541565..000000000 --- a/pkg/sentry/platform/safecopy/memcpy_arm64.s +++ /dev/null @@ -1,78 +0,0 @@ -// Copyright 2014 The Go Authors. All rights reserved. -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -#include "textflag.h" - -// handleMemcpyFault returns (the value stored in R0, the value stored in R1). -// Control is transferred to it when memcpy below receives SIGSEGV or SIGBUS, -// with the faulting address stored in R0 and the signal number stored in R1. -// -// It must have the same frame configuration as memcpy so that it can undo any -// potential call frame set up by the assembler. -TEXT handleMemcpyFault(SB), NOSPLIT, $0-36 - MOVD R0, addr+24(FP) - MOVW R1, sig+32(FP) - RET - -// memcpy copies data from src to dst. If a SIGSEGV or SIGBUS signal is received -// during the copy, it returns the address that caused the fault and the number -// of the signal that was received. Otherwise, it returns an unspecified address -// and a signal number of 0. -// -// Data is copied in order, such that if a fault happens at address p, it is -// safe to assume that all data before p-maxRegisterSize has already been -// successfully copied. -// -// The code is derived from the Go source runtime.memmove. -// -// func memcpy(dst, src unsafe.Pointer, n uintptr) (fault unsafe.Pointer, sig int32) -TEXT ·memcpy(SB), NOSPLIT, $-8-36 - // Store 0 as the returned signal number. If we run to completion, - // this is the value the caller will see; if a signal is received, - // handleMemcpyFault will store a different value in this address. - MOVW $0, sig+32(FP) - - MOVD to+0(FP), R3 - MOVD from+8(FP), R4 - MOVD n+16(FP), R5 - CMP $0, R5 - BNE check - RET - -check: - AND $~7, R5, R7 // R7 is N&~7. - SUB R7, R5, R6 // R6 is N&7. - - // Copying forward proceeds by copying R7/8 words then copying R6 bytes. - // R3 and R4 are advanced as we copy. - - // (There may be implementations of armv8 where copying by bytes until - // at least one of source or dest is word aligned is a worthwhile - // optimization, but the on the one tested so far (xgene) it did not - // make a significance difference.) - - CMP $0, R7 // Do we need to do any word-by-word copying? - BEQ noforwardlarge - ADD R3, R7, R9 // R9 points just past where we copy by word. - -forwardlargeloop: - MOVD.P 8(R4), R8 // R8 is just a scratch register. - MOVD.P R8, 8(R3) - CMP R3, R9 - BNE forwardlargeloop - -noforwardlarge: - CMP $0, R6 // Do we need to do any byte-by-byte copying? - BNE forwardtail - RET - -forwardtail: - ADD R3, R6, R9 // R9 points just past the destination memory. - -forwardtailloop: - MOVBU.P 1(R4), R8 - MOVBU.P R8, 1(R3) - CMP R3, R9 - BNE forwardtailloop - RET diff --git a/pkg/sentry/platform/safecopy/safecopy.go b/pkg/sentry/platform/safecopy/safecopy.go deleted file mode 100644 index 2fb7e5809..000000000 --- a/pkg/sentry/platform/safecopy/safecopy.go +++ /dev/null @@ -1,144 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package safecopy provides an efficient implementation of functions to access -// memory that may result in SIGSEGV or SIGBUS being sent to the accessor. -package safecopy - -import ( - "fmt" - "reflect" - "runtime" - "syscall" - - "gvisor.dev/gvisor/pkg/syserror" -) - -// SegvError is returned when a safecopy function receives SIGSEGV. -type SegvError struct { - // Addr is the address at which the SIGSEGV occurred. - Addr uintptr -} - -// Error implements error.Error. -func (e SegvError) Error() string { - return fmt.Sprintf("SIGSEGV at %#x", e.Addr) -} - -// BusError is returned when a safecopy function receives SIGBUS. -type BusError struct { - // Addr is the address at which the SIGBUS occurred. - Addr uintptr -} - -// Error implements error.Error. -func (e BusError) Error() string { - return fmt.Sprintf("SIGBUS at %#x", e.Addr) -} - -// AlignmentError is returned when a safecopy function is passed an address -// that does not meet alignment requirements. -type AlignmentError struct { - // Addr is the invalid address. - Addr uintptr - - // Alignment is the required alignment. - Alignment uintptr -} - -// Error implements error.Error. -func (e AlignmentError) Error() string { - return fmt.Sprintf("address %#x is not aligned to a %d-byte boundary", e.Addr, e.Alignment) -} - -var ( - // The begin and end addresses below are for the functions that are - // checked by the signal handler. - memcpyBegin uintptr - memcpyEnd uintptr - memclrBegin uintptr - memclrEnd uintptr - swapUint32Begin uintptr - swapUint32End uintptr - swapUint64Begin uintptr - swapUint64End uintptr - compareAndSwapUint32Begin uintptr - compareAndSwapUint32End uintptr - loadUint32Begin uintptr - loadUint32End uintptr - - // savedSigSegVHandler is a pointer to the SIGSEGV handler that was - // configured before we replaced it with our own. We still call into it - // when we get a SIGSEGV that is not interesting to us. - savedSigSegVHandler uintptr - - // same a above, but for SIGBUS signals. - savedSigBusHandler uintptr -) - -// signalHandler is our replacement signal handler for SIGSEGV and SIGBUS -// signals. -func signalHandler() - -// FindEndAddress returns the end address (one byte beyond the last) of the -// function that contains the specified address (begin). -func FindEndAddress(begin uintptr) uintptr { - f := runtime.FuncForPC(begin) - if f != nil { - for p := begin; ; p++ { - g := runtime.FuncForPC(p) - if f != g { - return p - } - } - } - return begin -} - -// initializeAddresses initializes the addresses used by the signal handler. -func initializeAddresses() { - // The following functions are written in assembly language, so they won't - // be inlined by the existing compiler/linker. Tests will fail if this - // assumption is violated. - memcpyBegin = reflect.ValueOf(memcpy).Pointer() - memcpyEnd = FindEndAddress(memcpyBegin) - memclrBegin = reflect.ValueOf(memclr).Pointer() - memclrEnd = FindEndAddress(memclrBegin) - swapUint32Begin = reflect.ValueOf(swapUint32).Pointer() - swapUint32End = FindEndAddress(swapUint32Begin) - swapUint64Begin = reflect.ValueOf(swapUint64).Pointer() - swapUint64End = FindEndAddress(swapUint64Begin) - compareAndSwapUint32Begin = reflect.ValueOf(compareAndSwapUint32).Pointer() - compareAndSwapUint32End = FindEndAddress(compareAndSwapUint32Begin) - loadUint32Begin = reflect.ValueOf(loadUint32).Pointer() - loadUint32End = FindEndAddress(loadUint32Begin) -} - -func init() { - initializeAddresses() - if err := ReplaceSignalHandler(syscall.SIGSEGV, reflect.ValueOf(signalHandler).Pointer(), &savedSigSegVHandler); err != nil { - panic(fmt.Sprintf("Unable to set handler for SIGSEGV: %v", err)) - } - if err := ReplaceSignalHandler(syscall.SIGBUS, reflect.ValueOf(signalHandler).Pointer(), &savedSigBusHandler); err != nil { - panic(fmt.Sprintf("Unable to set handler for SIGBUS: %v", err)) - } - syserror.AddErrorUnwrapper(func(e error) (syscall.Errno, bool) { - switch e.(type) { - case SegvError, BusError, AlignmentError: - return syscall.EFAULT, true - default: - return 0, false - } - }) -} diff --git a/pkg/sentry/platform/safecopy/safecopy_test.go b/pkg/sentry/platform/safecopy/safecopy_test.go deleted file mode 100644 index 5818f7f9b..000000000 --- a/pkg/sentry/platform/safecopy/safecopy_test.go +++ /dev/null @@ -1,617 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package safecopy - -import ( - "bytes" - "fmt" - "io/ioutil" - "math/rand" - "os" - "runtime/debug" - "syscall" - "testing" - "unsafe" -) - -// Size of a page in bytes. Cloned from usermem.PageSize to avoid a circular -// dependency. -const pageSize = 4096 - -func initRandom(b []byte) { - for i := range b { - b[i] = byte(rand.Intn(256)) - } -} - -func randBuf(size int) []byte { - b := make([]byte, size) - initRandom(b) - return b -} - -func TestCopyInSuccess(t *testing.T) { - // Test that CopyIn does not return an error when all pages are accessible. - const bufLen = 8192 - a := randBuf(bufLen) - b := make([]byte, bufLen) - - n, err := CopyIn(b, unsafe.Pointer(&a[0])) - if n != bufLen { - t.Errorf("Unexpected copy length, got %v, want %v", n, bufLen) - } - if err != nil { - t.Errorf("Unexpected error: %v", err) - } - if !bytes.Equal(a, b) { - t.Errorf("Buffers are not equal when they should be: %v %v", a, b) - } -} - -func TestCopyOutSuccess(t *testing.T) { - // Test that CopyOut does not return an error when all pages are - // accessible. - const bufLen = 8192 - a := randBuf(bufLen) - b := make([]byte, bufLen) - - n, err := CopyOut(unsafe.Pointer(&b[0]), a) - if n != bufLen { - t.Errorf("Unexpected copy length, got %v, want %v", n, bufLen) - } - if err != nil { - t.Errorf("Unexpected error: %v", err) - } - if !bytes.Equal(a, b) { - t.Errorf("Buffers are not equal when they should be: %v %v", a, b) - } -} - -func TestCopySuccess(t *testing.T) { - // Test that Copy does not return an error when all pages are accessible. - const bufLen = 8192 - a := randBuf(bufLen) - b := make([]byte, bufLen) - - n, err := Copy(unsafe.Pointer(&b[0]), unsafe.Pointer(&a[0]), bufLen) - if n != bufLen { - t.Errorf("Unexpected copy length, got %v, want %v", n, bufLen) - } - if err != nil { - t.Errorf("Unexpected error: %v", err) - } - if !bytes.Equal(a, b) { - t.Errorf("Buffers are not equal when they should be: %v %v", a, b) - } -} - -func TestZeroOutSuccess(t *testing.T) { - // Test that ZeroOut does not return an error when all pages are - // accessible. - const bufLen = 8192 - a := make([]byte, bufLen) - b := randBuf(bufLen) - - n, err := ZeroOut(unsafe.Pointer(&b[0]), bufLen) - if n != bufLen { - t.Errorf("Unexpected copy length, got %v, want %v", n, bufLen) - } - if err != nil { - t.Errorf("Unexpected error: %v", err) - } - if !bytes.Equal(a, b) { - t.Errorf("Buffers are not equal when they should be: %v %v", a, b) - } -} - -func TestSwapUint32Success(t *testing.T) { - // Test that SwapUint32 does not return an error when the page is - // accessible. - before := uint32(rand.Int31()) - after := uint32(rand.Int31()) - val := before - - old, err := SwapUint32(unsafe.Pointer(&val), after) - if err != nil { - t.Errorf("Unexpected error: %v", err) - } - if old != before { - t.Errorf("Unexpected old value: got %v, want %v", old, before) - } - if val != after { - t.Errorf("Unexpected new value: got %v, want %v", val, after) - } -} - -func TestSwapUint32AlignmentError(t *testing.T) { - // Test that SwapUint32 returns an AlignmentError when passed an unaligned - // address. - data := new(struct{ val uint64 }) - addr := uintptr(unsafe.Pointer(&data.val)) + 1 - want := AlignmentError{Addr: addr, Alignment: 4} - if _, err := SwapUint32(unsafe.Pointer(addr), 1); err != want { - t.Errorf("Unexpected error: got %v, want %v", err, want) - } -} - -func TestSwapUint64Success(t *testing.T) { - // Test that SwapUint64 does not return an error when the page is - // accessible. - before := uint64(rand.Int63()) - after := uint64(rand.Int63()) - // "The first word in ... an allocated struct or slice can be relied upon - // to be 64-bit aligned." - sync/atomic docs - data := new(struct{ val uint64 }) - data.val = before - - old, err := SwapUint64(unsafe.Pointer(&data.val), after) - if err != nil { - t.Errorf("Unexpected error: %v", err) - } - if old != before { - t.Errorf("Unexpected old value: got %v, want %v", old, before) - } - if data.val != after { - t.Errorf("Unexpected new value: got %v, want %v", data.val, after) - } -} - -func TestSwapUint64AlignmentError(t *testing.T) { - // Test that SwapUint64 returns an AlignmentError when passed an unaligned - // address. - data := new(struct{ val1, val2 uint64 }) - addr := uintptr(unsafe.Pointer(&data.val1)) + 1 - want := AlignmentError{Addr: addr, Alignment: 8} - if _, err := SwapUint64(unsafe.Pointer(addr), 1); err != want { - t.Errorf("Unexpected error: got %v, want %v", err, want) - } -} - -func TestCompareAndSwapUint32Success(t *testing.T) { - // Test that CompareAndSwapUint32 does not return an error when the page is - // accessible. - before := uint32(rand.Int31()) - after := uint32(rand.Int31()) - val := before - - old, err := CompareAndSwapUint32(unsafe.Pointer(&val), before, after) - if err != nil { - t.Errorf("Unexpected error: %v", err) - } - if old != before { - t.Errorf("Unexpected old value: got %v, want %v", old, before) - } - if val != after { - t.Errorf("Unexpected new value: got %v, want %v", val, after) - } -} - -func TestCompareAndSwapUint32AlignmentError(t *testing.T) { - // Test that CompareAndSwapUint32 returns an AlignmentError when passed an - // unaligned address. - data := new(struct{ val uint64 }) - addr := uintptr(unsafe.Pointer(&data.val)) + 1 - want := AlignmentError{Addr: addr, Alignment: 4} - if _, err := CompareAndSwapUint32(unsafe.Pointer(addr), 0, 1); err != want { - t.Errorf("Unexpected error: got %v, want %v", err, want) - } -} - -// withSegvErrorTestMapping calls fn with a two-page mapping. The first page -// contains random data, and the second page generates SIGSEGV when accessed. -func withSegvErrorTestMapping(t *testing.T, fn func(m []byte)) { - mapping, err := syscall.Mmap(-1, 0, 2*pageSize, syscall.PROT_READ|syscall.PROT_WRITE, syscall.MAP_ANONYMOUS|syscall.MAP_PRIVATE) - if err != nil { - t.Fatalf("Mmap failed: %v", err) - } - defer syscall.Munmap(mapping) - if err := syscall.Mprotect(mapping[pageSize:], syscall.PROT_NONE); err != nil { - t.Fatalf("Mprotect failed: %v", err) - } - initRandom(mapping[:pageSize]) - - fn(mapping) -} - -// withBusErrorTestMapping calls fn with a two-page mapping. The first page -// contains random data, and the second page generates SIGBUS when accessed. -func withBusErrorTestMapping(t *testing.T, fn func(m []byte)) { - f, err := ioutil.TempFile("", "sigbus_test") - if err != nil { - t.Fatalf("TempFile failed: %v", err) - } - defer f.Close() - if err := f.Truncate(pageSize); err != nil { - t.Fatalf("Truncate failed: %v", err) - } - mapping, err := syscall.Mmap(int(f.Fd()), 0, 2*pageSize, syscall.PROT_READ|syscall.PROT_WRITE, syscall.MAP_SHARED) - if err != nil { - t.Fatalf("Mmap failed: %v", err) - } - defer syscall.Munmap(mapping) - initRandom(mapping[:pageSize]) - - fn(mapping) -} - -func TestCopyInSegvError(t *testing.T) { - // Test that CopyIn returns a SegvError when reaching a page that signals - // SIGSEGV. - for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { - t.Run(fmt.Sprintf("starting copy %d bytes before SIGSEGV", bytesBeforeFault), func(t *testing.T) { - withSegvErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - src := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) - dst := randBuf(pageSize) - n, err := CopyIn(dst, src) - if n != bytesBeforeFault { - t.Errorf("Unexpected copy length: got %v, want %v", n, bytesBeforeFault) - } - if want := (SegvError{secondPage}); err != want { - t.Errorf("Unexpected error: got %v, want %v", err, want) - } - if got, want := dst[:bytesBeforeFault], mapping[pageSize-bytesBeforeFault:pageSize]; !bytes.Equal(got, want) { - t.Errorf("Buffers are not equal when they should be: %v %v", got, want) - } - }) - }) - } -} - -func TestCopyInBusError(t *testing.T) { - // Test that CopyIn returns a BusError when reaching a page that signals - // SIGBUS. - for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { - t.Run(fmt.Sprintf("starting copy %d bytes before SIGBUS", bytesBeforeFault), func(t *testing.T) { - withBusErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - src := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) - dst := randBuf(pageSize) - n, err := CopyIn(dst, src) - if n != bytesBeforeFault { - t.Errorf("Unexpected copy length: got %v, want %v", n, bytesBeforeFault) - } - if want := (BusError{secondPage}); err != want { - t.Errorf("Unexpected error: got %v, want %v", err, want) - } - if got, want := dst[:bytesBeforeFault], mapping[pageSize-bytesBeforeFault:pageSize]; !bytes.Equal(got, want) { - t.Errorf("Buffers are not equal when they should be: %v %v", got, want) - } - }) - }) - } -} - -func TestCopyOutSegvError(t *testing.T) { - // Test that CopyOut returns a SegvError when reaching a page that signals - // SIGSEGV. - for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { - t.Run(fmt.Sprintf("starting copy %d bytes before SIGSEGV", bytesBeforeFault), func(t *testing.T) { - withSegvErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - dst := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) - src := randBuf(pageSize) - n, err := CopyOut(dst, src) - if n != bytesBeforeFault { - t.Errorf("Unexpected copy length: got %v, want %v", n, bytesBeforeFault) - } - if want := (SegvError{secondPage}); err != want { - t.Errorf("Unexpected error: got %v, want %v", err, want) - } - if got, want := mapping[pageSize-bytesBeforeFault:pageSize], src[:bytesBeforeFault]; !bytes.Equal(got, want) { - t.Errorf("Buffers are not equal when they should be: %v %v", got, want) - } - }) - }) - } -} - -func TestCopyOutBusError(t *testing.T) { - // Test that CopyOut returns a BusError when reaching a page that signals - // SIGBUS. - for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { - t.Run(fmt.Sprintf("starting copy %d bytes before SIGSEGV", bytesBeforeFault), func(t *testing.T) { - withBusErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - dst := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) - src := randBuf(pageSize) - n, err := CopyOut(dst, src) - if n != bytesBeforeFault { - t.Errorf("Unexpected copy length: got %v, want %v", n, bytesBeforeFault) - } - if want := (BusError{secondPage}); err != want { - t.Errorf("Unexpected error: got %v, want %v", err, want) - } - if got, want := mapping[pageSize-bytesBeforeFault:pageSize], src[:bytesBeforeFault]; !bytes.Equal(got, want) { - t.Errorf("Buffers are not equal when they should be: %v %v", got, want) - } - }) - }) - } -} - -func TestCopySourceSegvError(t *testing.T) { - // Test that Copy returns a SegvError when copying from a page that signals - // SIGSEGV. - for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { - t.Run(fmt.Sprintf("starting copy %d bytes before SIGSEGV", bytesBeforeFault), func(t *testing.T) { - withSegvErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - src := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) - dst := randBuf(pageSize) - n, err := Copy(unsafe.Pointer(&dst[0]), src, pageSize) - if n != uintptr(bytesBeforeFault) { - t.Errorf("Unexpected copy length: got %v, want %v", n, bytesBeforeFault) - } - if want := (SegvError{secondPage}); err != want { - t.Errorf("Unexpected error: got %v, want %v", err, want) - } - if got, want := dst[:bytesBeforeFault], mapping[pageSize-bytesBeforeFault:pageSize]; !bytes.Equal(got, want) { - t.Errorf("Buffers are not equal when they should be: %v %v", got, want) - } - }) - }) - } -} - -func TestCopySourceBusError(t *testing.T) { - // Test that Copy returns a BusError when copying from a page that signals - // SIGBUS. - for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { - t.Run(fmt.Sprintf("starting copy %d bytes before SIGBUS", bytesBeforeFault), func(t *testing.T) { - withBusErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - src := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) - dst := randBuf(pageSize) - n, err := Copy(unsafe.Pointer(&dst[0]), src, pageSize) - if n != uintptr(bytesBeforeFault) { - t.Errorf("Unexpected copy length: got %v, want %v", n, bytesBeforeFault) - } - if want := (BusError{secondPage}); err != want { - t.Errorf("Unexpected error: got %v, want %v", err, want) - } - if got, want := dst[:bytesBeforeFault], mapping[pageSize-bytesBeforeFault:pageSize]; !bytes.Equal(got, want) { - t.Errorf("Buffers are not equal when they should be: %v %v", got, want) - } - }) - }) - } -} - -func TestCopyDestinationSegvError(t *testing.T) { - // Test that Copy returns a SegvError when copying to a page that signals - // SIGSEGV. - for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { - t.Run(fmt.Sprintf("starting copy %d bytes before SIGSEGV", bytesBeforeFault), func(t *testing.T) { - withSegvErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - dst := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) - src := randBuf(pageSize) - n, err := Copy(dst, unsafe.Pointer(&src[0]), pageSize) - if n != uintptr(bytesBeforeFault) { - t.Errorf("Unexpected copy length: got %v, want %v", n, bytesBeforeFault) - } - if want := (SegvError{secondPage}); err != want { - t.Errorf("Unexpected error: got %v, want %v", err, want) - } - if got, want := mapping[pageSize-bytesBeforeFault:pageSize], src[:bytesBeforeFault]; !bytes.Equal(got, want) { - t.Errorf("Buffers are not equal when they should be: %v %v", got, want) - } - }) - }) - } -} - -func TestCopyDestinationBusError(t *testing.T) { - // Test that Copy returns a BusError when copying to a page that signals - // SIGBUS. - for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { - t.Run(fmt.Sprintf("starting copy %d bytes before SIGBUS", bytesBeforeFault), func(t *testing.T) { - withBusErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - dst := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) - src := randBuf(pageSize) - n, err := Copy(dst, unsafe.Pointer(&src[0]), pageSize) - if n != uintptr(bytesBeforeFault) { - t.Errorf("Unexpected copy length: got %v, want %v", n, bytesBeforeFault) - } - if want := (BusError{secondPage}); err != want { - t.Errorf("Unexpected error: got %v, want %v", err, want) - } - if got, want := mapping[pageSize-bytesBeforeFault:pageSize], src[:bytesBeforeFault]; !bytes.Equal(got, want) { - t.Errorf("Buffers are not equal when they should be: %v %v", got, want) - } - }) - }) - } -} - -func TestZeroOutSegvError(t *testing.T) { - // Test that ZeroOut returns a SegvError when reaching a page that signals - // SIGSEGV. - for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { - t.Run(fmt.Sprintf("starting write %d bytes before SIGSEGV", bytesBeforeFault), func(t *testing.T) { - withSegvErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - dst := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) - n, err := ZeroOut(dst, pageSize) - if n != uintptr(bytesBeforeFault) { - t.Errorf("Unexpected write length: got %v, want %v", n, bytesBeforeFault) - } - if want := (SegvError{secondPage}); err != want { - t.Errorf("Unexpected error: got %v, want %v", err, want) - } - if got, want := mapping[pageSize-bytesBeforeFault:pageSize], make([]byte, bytesBeforeFault); !bytes.Equal(got, want) { - t.Errorf("Non-zero bytes in written part of mapping: %v", got) - } - }) - }) - } -} - -func TestZeroOutBusError(t *testing.T) { - // Test that ZeroOut returns a BusError when reaching a page that signals - // SIGBUS. - for bytesBeforeFault := 0; bytesBeforeFault <= 2*maxRegisterSize; bytesBeforeFault++ { - t.Run(fmt.Sprintf("starting write %d bytes before SIGBUS", bytesBeforeFault), func(t *testing.T) { - withBusErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - dst := unsafe.Pointer(secondPage - uintptr(bytesBeforeFault)) - n, err := ZeroOut(dst, pageSize) - if n != uintptr(bytesBeforeFault) { - t.Errorf("Unexpected write length: got %v, want %v", n, bytesBeforeFault) - } - if want := (BusError{secondPage}); err != want { - t.Errorf("Unexpected error: got %v, want %v", err, want) - } - if got, want := mapping[pageSize-bytesBeforeFault:pageSize], make([]byte, bytesBeforeFault); !bytes.Equal(got, want) { - t.Errorf("Non-zero bytes in written part of mapping: %v", got) - } - }) - }) - } -} - -func TestSwapUint32SegvError(t *testing.T) { - // Test that SwapUint32 returns a SegvError when reaching a page that - // signals SIGSEGV. - withSegvErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - _, err := SwapUint32(unsafe.Pointer(secondPage), 1) - if want := (SegvError{secondPage}); err != want { - t.Errorf("Unexpected error: got %v, want %v", err, want) - } - }) -} - -func TestSwapUint32BusError(t *testing.T) { - // Test that SwapUint32 returns a BusError when reaching a page that - // signals SIGBUS. - withBusErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - _, err := SwapUint32(unsafe.Pointer(secondPage), 1) - if want := (BusError{secondPage}); err != want { - t.Errorf("Unexpected error: got %v, want %v", err, want) - } - }) -} - -func TestSwapUint64SegvError(t *testing.T) { - // Test that SwapUint64 returns a SegvError when reaching a page that - // signals SIGSEGV. - withSegvErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - _, err := SwapUint64(unsafe.Pointer(secondPage), 1) - if want := (SegvError{secondPage}); err != want { - t.Errorf("Unexpected error: got %v, want %v", err, want) - } - }) -} - -func TestSwapUint64BusError(t *testing.T) { - // Test that SwapUint64 returns a BusError when reaching a page that - // signals SIGBUS. - withBusErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - _, err := SwapUint64(unsafe.Pointer(secondPage), 1) - if want := (BusError{secondPage}); err != want { - t.Errorf("Unexpected error: got %v, want %v", err, want) - } - }) -} - -func TestCompareAndSwapUint32SegvError(t *testing.T) { - // Test that CompareAndSwapUint32 returns a SegvError when reaching a page - // that signals SIGSEGV. - withSegvErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - _, err := CompareAndSwapUint32(unsafe.Pointer(secondPage), 0, 1) - if want := (SegvError{secondPage}); err != want { - t.Errorf("Unexpected error: got %v, want %v", err, want) - } - }) -} - -func TestCompareAndSwapUint32BusError(t *testing.T) { - // Test that CompareAndSwapUint32 returns a BusError when reaching a page - // that signals SIGBUS. - withBusErrorTestMapping(t, func(mapping []byte) { - secondPage := uintptr(unsafe.Pointer(&mapping[0])) + pageSize - _, err := CompareAndSwapUint32(unsafe.Pointer(secondPage), 0, 1) - if want := (BusError{secondPage}); err != want { - t.Errorf("Unexpected error: got %v, want %v", err, want) - } - }) -} - -func testCopy(dst, src []byte) (panicked bool) { - defer func() { - if r := recover(); r != nil { - panicked = true - } - }() - debug.SetPanicOnFault(true) - copy(dst, src) - return -} - -func TestSegVOnMemmove(t *testing.T) { - // Test that SIGSEGVs received by runtime.memmove when *not* doing - // CopyIn or CopyOut work gets propagated to the runtime. - const bufLen = pageSize - a, err := syscall.Mmap(-1, 0, bufLen, syscall.PROT_NONE, syscall.MAP_ANON|syscall.MAP_PRIVATE) - if err != nil { - t.Fatalf("Mmap failed: %v", err) - - } - defer syscall.Munmap(a) - b := randBuf(bufLen) - - if !testCopy(b, a) { - t.Fatalf("testCopy didn't panic when it should have") - } - - if !testCopy(a, b) { - t.Fatalf("testCopy didn't panic when it should have") - } -} - -func TestSigbusOnMemmove(t *testing.T) { - // Test that SIGBUS received by runtime.memmove when *not* doing - // CopyIn or CopyOut work gets propagated to the runtime. - const bufLen = pageSize - f, err := ioutil.TempFile("", "sigbus_test") - if err != nil { - t.Fatalf("TempFile failed: %v", err) - } - os.Remove(f.Name()) - defer f.Close() - - a, err := syscall.Mmap(int(f.Fd()), 0, bufLen, syscall.PROT_READ|syscall.PROT_WRITE, syscall.MAP_SHARED) - if err != nil { - t.Fatalf("Mmap failed: %v", err) - - } - defer syscall.Munmap(a) - b := randBuf(bufLen) - - if !testCopy(b, a) { - t.Fatalf("testCopy didn't panic when it should have") - } - - if !testCopy(a, b) { - t.Fatalf("testCopy didn't panic when it should have") - } -} diff --git a/pkg/sentry/platform/safecopy/safecopy_unsafe.go b/pkg/sentry/platform/safecopy/safecopy_unsafe.go deleted file mode 100644 index eef028e68..000000000 --- a/pkg/sentry/platform/safecopy/safecopy_unsafe.go +++ /dev/null @@ -1,335 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package safecopy - -import ( - "fmt" - "syscall" - "unsafe" -) - -// maxRegisterSize is the maximum register size used in memcpy and memclr. It -// is used to decide by how much to rewind the copy (for memcpy) or zeroing -// (for memclr) before proceeding. -const maxRegisterSize = 16 - -// memcpy copies data from src to dst. If a SIGSEGV or SIGBUS signal is received -// during the copy, it returns the address that caused the fault and the number -// of the signal that was received. Otherwise, it returns an unspecified address -// and a signal number of 0. -// -// Data is copied in order, such that if a fault happens at address p, it is -// safe to assume that all data before p-maxRegisterSize has already been -// successfully copied. -// -//go:noescape -func memcpy(dst, src unsafe.Pointer, n uintptr) (fault unsafe.Pointer, sig int32) - -// memclr sets the n bytes following ptr to zeroes. If a SIGSEGV or SIGBUS -// signal is received during the write, it returns the address that caused the -// fault and the number of the signal that was received. Otherwise, it returns -// an unspecified address and a signal number of 0. -// -// Data is written in order, such that if a fault happens at address p, it is -// safe to assume that all data before p-maxRegisterSize has already been -// successfully written. -// -//go:noescape -func memclr(ptr unsafe.Pointer, n uintptr) (fault unsafe.Pointer, sig int32) - -// swapUint32 atomically stores new into *ptr and returns (the previous *ptr -// value, 0). If a SIGSEGV or SIGBUS signal is received during the swap, the -// value of old is unspecified, and sig is the number of the signal that was -// received. -// -// Preconditions: ptr must be aligned to a 4-byte boundary. -// -//go:noescape -func swapUint32(ptr unsafe.Pointer, new uint32) (old uint32, sig int32) - -// swapUint64 atomically stores new into *ptr and returns (the previous *ptr -// value, 0). If a SIGSEGV or SIGBUS signal is received during the swap, the -// value of old is unspecified, and sig is the number of the signal that was -// received. -// -// Preconditions: ptr must be aligned to a 8-byte boundary. -// -//go:noescape -func swapUint64(ptr unsafe.Pointer, new uint64) (old uint64, sig int32) - -// compareAndSwapUint32 is like sync/atomic.CompareAndSwapUint32, but returns -// (the value previously stored at ptr, 0). If a SIGSEGV or SIGBUS signal is -// received during the operation, the value of prev is unspecified, and sig is -// the number of the signal that was received. -// -// Preconditions: ptr must be aligned to a 4-byte boundary. -// -//go:noescape -func compareAndSwapUint32(ptr unsafe.Pointer, old, new uint32) (prev uint32, sig int32) - -// LoadUint32 is like sync/atomic.LoadUint32, but operates with user memory. It -// may fail with SIGSEGV or SIGBUS if it is received while reading from ptr. -// -// Preconditions: ptr must be aligned to a 4-byte boundary. -// -//go:noescape -func loadUint32(ptr unsafe.Pointer) (val uint32, sig int32) - -// CopyIn copies len(dst) bytes from src to dst. It returns the number of bytes -// copied and an error if SIGSEGV or SIGBUS is received while reading from src. -func CopyIn(dst []byte, src unsafe.Pointer) (int, error) { - toCopy := uintptr(len(dst)) - if len(dst) == 0 { - return 0, nil - } - - fault, sig := memcpy(unsafe.Pointer(&dst[0]), src, toCopy) - if sig == 0 { - return len(dst), nil - } - - faultN, srcN := uintptr(fault), uintptr(src) - if faultN < srcN || faultN >= srcN+toCopy { - panic(fmt.Sprintf("CopyIn raised signal %d at %#x, which is outside source [%#x, %#x)", sig, faultN, srcN, srcN+toCopy)) - } - - // memcpy might have ended the copy up to maxRegisterSize bytes before - // fault, if an instruction caused a memory access that straddled two - // pages, and the second one faulted. Try to copy up to the fault. - var done int - if faultN-srcN > maxRegisterSize { - done = int(faultN - srcN - maxRegisterSize) - } - n, err := CopyIn(dst[done:int(faultN-srcN)], unsafe.Pointer(srcN+uintptr(done))) - done += n - if err != nil { - return done, err - } - return done, errorFromFaultSignal(fault, sig) -} - -// CopyOut copies len(src) bytes from src to dst. If returns the number of -// bytes done and an error if SIGSEGV or SIGBUS is received while writing to -// dst. -func CopyOut(dst unsafe.Pointer, src []byte) (int, error) { - toCopy := uintptr(len(src)) - if toCopy == 0 { - return 0, nil - } - - fault, sig := memcpy(dst, unsafe.Pointer(&src[0]), toCopy) - if sig == 0 { - return len(src), nil - } - - faultN, dstN := uintptr(fault), uintptr(dst) - if faultN < dstN || faultN >= dstN+toCopy { - panic(fmt.Sprintf("CopyOut raised signal %d at %#x, which is outside destination [%#x, %#x)", sig, faultN, dstN, dstN+toCopy)) - } - - // memcpy might have ended the copy up to maxRegisterSize bytes before - // fault, if an instruction caused a memory access that straddled two - // pages, and the second one faulted. Try to copy up to the fault. - var done int - if faultN-dstN > maxRegisterSize { - done = int(faultN - dstN - maxRegisterSize) - } - n, err := CopyOut(unsafe.Pointer(dstN+uintptr(done)), src[done:int(faultN-dstN)]) - done += n - if err != nil { - return done, err - } - return done, errorFromFaultSignal(fault, sig) -} - -// Copy copies toCopy bytes from src to dst. It returns the number of bytes -// copied and an error if SIGSEGV or SIGBUS is received while reading from src -// or writing to dst. -// -// Data is copied in order; if [src, src+toCopy) and [dst, dst+toCopy) overlap, -// the resulting contents of dst are unspecified. -func Copy(dst, src unsafe.Pointer, toCopy uintptr) (uintptr, error) { - if toCopy == 0 { - return 0, nil - } - - fault, sig := memcpy(dst, src, toCopy) - if sig == 0 { - return toCopy, nil - } - - // Did the fault occur while reading from src or writing to dst? - faultN, srcN, dstN := uintptr(fault), uintptr(src), uintptr(dst) - faultAfterSrc := ^uintptr(0) - if faultN >= srcN { - faultAfterSrc = faultN - srcN - } - faultAfterDst := ^uintptr(0) - if faultN >= dstN { - faultAfterDst = faultN - dstN - } - if faultAfterSrc >= toCopy && faultAfterDst >= toCopy { - panic(fmt.Sprintf("Copy raised signal %d at %#x, which is outside source [%#x, %#x) and destination [%#x, %#x)", sig, faultN, srcN, srcN+toCopy, dstN, dstN+toCopy)) - } - faultedAfter := faultAfterSrc - if faultedAfter > faultAfterDst { - faultedAfter = faultAfterDst - } - - // memcpy might have ended the copy up to maxRegisterSize bytes before - // fault, if an instruction caused a memory access that straddled two - // pages, and the second one faulted. Try to copy up to the fault. - var done uintptr - if faultedAfter > maxRegisterSize { - done = faultedAfter - maxRegisterSize - } - n, err := Copy(unsafe.Pointer(dstN+done), unsafe.Pointer(srcN+done), faultedAfter-done) - done += n - if err != nil { - return done, err - } - return done, errorFromFaultSignal(fault, sig) -} - -// ZeroOut writes toZero zero bytes to dst. It returns the number of bytes -// written and an error if SIGSEGV or SIGBUS is received while writing to dst. -func ZeroOut(dst unsafe.Pointer, toZero uintptr) (uintptr, error) { - if toZero == 0 { - return 0, nil - } - - fault, sig := memclr(dst, toZero) - if sig == 0 { - return toZero, nil - } - - faultN, dstN := uintptr(fault), uintptr(dst) - if faultN < dstN || faultN >= dstN+toZero { - panic(fmt.Sprintf("ZeroOut raised signal %d at %#x, which is outside destination [%#x, %#x)", sig, faultN, dstN, dstN+toZero)) - } - - // memclr might have ended the write up to maxRegisterSize bytes before - // fault, if an instruction caused a memory access that straddled two - // pages, and the second one faulted. Try to write up to the fault. - var done uintptr - if faultN-dstN > maxRegisterSize { - done = faultN - dstN - maxRegisterSize - } - n, err := ZeroOut(unsafe.Pointer(dstN+done), faultN-dstN-done) - done += n - if err != nil { - return done, err - } - return done, errorFromFaultSignal(fault, sig) -} - -// SwapUint32 is equivalent to sync/atomic.SwapUint32, except that it returns -// an error if SIGSEGV or SIGBUS is received while accessing ptr, or if ptr is -// not aligned to a 4-byte boundary. -func SwapUint32(ptr unsafe.Pointer, new uint32) (uint32, error) { - if addr := uintptr(ptr); addr&3 != 0 { - return 0, AlignmentError{addr, 4} - } - old, sig := swapUint32(ptr, new) - return old, errorFromFaultSignal(ptr, sig) -} - -// SwapUint64 is equivalent to sync/atomic.SwapUint64, except that it returns -// an error if SIGSEGV or SIGBUS is received while accessing ptr, or if ptr is -// not aligned to an 8-byte boundary. -func SwapUint64(ptr unsafe.Pointer, new uint64) (uint64, error) { - if addr := uintptr(ptr); addr&7 != 0 { - return 0, AlignmentError{addr, 8} - } - old, sig := swapUint64(ptr, new) - return old, errorFromFaultSignal(ptr, sig) -} - -// CompareAndSwapUint32 is equivalent to atomicbitops.CompareAndSwapUint32, -// except that it returns an error if SIGSEGV or SIGBUS is received while -// accessing ptr, or if ptr is not aligned to a 4-byte boundary. -func CompareAndSwapUint32(ptr unsafe.Pointer, old, new uint32) (uint32, error) { - if addr := uintptr(ptr); addr&3 != 0 { - return 0, AlignmentError{addr, 4} - } - prev, sig := compareAndSwapUint32(ptr, old, new) - return prev, errorFromFaultSignal(ptr, sig) -} - -// LoadUint32 is like sync/atomic.LoadUint32, but operates with user memory. It -// may fail with SIGSEGV or SIGBUS if it is received while reading from ptr. -// -// Preconditions: ptr must be aligned to a 4-byte boundary. -func LoadUint32(ptr unsafe.Pointer) (uint32, error) { - if addr := uintptr(ptr); addr&3 != 0 { - return 0, AlignmentError{addr, 4} - } - val, sig := loadUint32(ptr) - return val, errorFromFaultSignal(ptr, sig) -} - -func errorFromFaultSignal(addr unsafe.Pointer, sig int32) error { - switch sig { - case 0: - return nil - case int32(syscall.SIGSEGV): - return SegvError{uintptr(addr)} - case int32(syscall.SIGBUS): - return BusError{uintptr(addr)} - default: - panic(fmt.Sprintf("safecopy got unexpected signal %d at address %#x", sig, addr)) - } -} - -// ReplaceSignalHandler replaces the existing signal handler for the provided -// signal with the one that handles faults in safecopy-protected functions. -// -// It stores the value of the previously set handler in previous. -// -// This function will be called on initialization in order to install safecopy -// handlers for appropriate signals. These handlers will call the previous -// handler however, and if this is function is being used externally then the -// same courtesy is expected. -func ReplaceSignalHandler(sig syscall.Signal, handler uintptr, previous *uintptr) error { - var sa struct { - handler uintptr - flags uint64 - restorer uintptr - mask uint64 - } - const maskLen = 8 - - // Get the existing signal handler information, and save the current - // handler. Once we replace it, we will use this pointer to fall back to - // it when we receive other signals. - if _, _, e := syscall.RawSyscall6(syscall.SYS_RT_SIGACTION, uintptr(sig), 0, uintptr(unsafe.Pointer(&sa)), maskLen, 0, 0); e != 0 { - return e - } - - // Fail if there isn't a previous handler. - if sa.handler == 0 { - return fmt.Errorf("previous handler for signal %x isn't set", sig) - } - - *previous = sa.handler - - // Install our own handler. - sa.handler = handler - if _, _, e := syscall.RawSyscall6(syscall.SYS_RT_SIGACTION, uintptr(sig), uintptr(unsafe.Pointer(&sa)), 0, maskLen, 0, 0); e != 0 { - return e - } - - return nil -} diff --git a/pkg/sentry/platform/safecopy/sighandler_amd64.s b/pkg/sentry/platform/safecopy/sighandler_amd64.s deleted file mode 100644 index 475ae48e9..000000000 --- a/pkg/sentry/platform/safecopy/sighandler_amd64.s +++ /dev/null @@ -1,133 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -#include "textflag.h" - -// The signals handled by sigHandler. -#define SIGBUS 7 -#define SIGSEGV 11 - -// Offsets to the registers in context->uc_mcontext.gregs[]. -#define REG_RDI 0x68 -#define REG_RAX 0x90 -#define REG_IP 0xa8 - -// Offset to the si_addr field of siginfo. -#define SI_CODE 0x08 -#define SI_ADDR 0x10 - -// signalHandler is the signal handler for SIGSEGV and SIGBUS signals. It must -// not be set up as a handler to any other signals. -// -// If the instruction causing the signal is within a safecopy-protected -// function, the signal is handled such that execution resumes in the -// appropriate fault handling stub with AX containing the faulting address and -// DI containing the signal number. Otherwise control is transferred to the -// previously configured signal handler (savedSigSegvHandler or -// savedSigBusHandler). -// -// This function cannot be written in go because it runs whenever a signal is -// received by the thread (preempting whatever was running), which includes when -// garbage collector has stopped or isn't expecting any interactions (like -// barriers). -// -// The arguments are the following: -// DI - The signal number. -// SI - Pointer to siginfo_t structure. -// DX - Pointer to ucontext structure. -TEXT ·signalHandler(SB),NOSPLIT,$0 - // Check if the signal is from the kernel. - MOVQ $0x0, CX - CMPL CX, SI_CODE(SI) - JGE original_handler - - // Check if RIP is within the area we care about. - MOVQ REG_IP(DX), CX - CMPQ CX, ·memcpyBegin(SB) - JB not_memcpy - CMPQ CX, ·memcpyEnd(SB) - JAE not_memcpy - - // Modify the context such that execution will resume in the fault - // handler. - LEAQ handleMemcpyFault(SB), CX - JMP handle_fault - -not_memcpy: - CMPQ CX, ·memclrBegin(SB) - JB not_memclr - CMPQ CX, ·memclrEnd(SB) - JAE not_memclr - - LEAQ handleMemclrFault(SB), CX - JMP handle_fault - -not_memclr: - CMPQ CX, ·swapUint32Begin(SB) - JB not_swapuint32 - CMPQ CX, ·swapUint32End(SB) - JAE not_swapuint32 - - LEAQ handleSwapUint32Fault(SB), CX - JMP handle_fault - -not_swapuint32: - CMPQ CX, ·swapUint64Begin(SB) - JB not_swapuint64 - CMPQ CX, ·swapUint64End(SB) - JAE not_swapuint64 - - LEAQ handleSwapUint64Fault(SB), CX - JMP handle_fault - -not_swapuint64: - CMPQ CX, ·compareAndSwapUint32Begin(SB) - JB not_casuint32 - CMPQ CX, ·compareAndSwapUint32End(SB) - JAE not_casuint32 - - LEAQ handleCompareAndSwapUint32Fault(SB), CX - JMP handle_fault - -not_casuint32: - CMPQ CX, ·loadUint32Begin(SB) - JB not_loaduint32 - CMPQ CX, ·loadUint32End(SB) - JAE not_loaduint32 - - LEAQ handleLoadUint32Fault(SB), CX - JMP handle_fault - -not_loaduint32: -original_handler: - // Jump to the previous signal handler, which is likely the golang one. - XORQ CX, CX - MOVQ ·savedSigBusHandler(SB), AX - CMPL DI, $SIGSEGV - CMOVQEQ ·savedSigSegVHandler(SB), AX - JMP AX - -handle_fault: - // Entered with the address of the fault handler in RCX; store it in - // RIP. - MOVQ CX, REG_IP(DX) - - // Store the faulting address in RAX. - MOVQ SI_ADDR(SI), CX - MOVQ CX, REG_RAX(DX) - - // Store the signal number in EDI. - MOVL DI, REG_RDI(DX) - - RET diff --git a/pkg/sentry/platform/safecopy/sighandler_arm64.s b/pkg/sentry/platform/safecopy/sighandler_arm64.s deleted file mode 100644 index 53e4ac2c1..000000000 --- a/pkg/sentry/platform/safecopy/sighandler_arm64.s +++ /dev/null @@ -1,143 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -#include "textflag.h" - -// The signals handled by sigHandler. -#define SIGBUS 7 -#define SIGSEGV 11 - -// Offsets to the registers in context->uc_mcontext.gregs[]. -#define REG_R0 0xB8 -#define REG_R1 0xC0 -#define REG_PC 0x1B8 - -// Offset to the si_addr field of siginfo. -#define SI_CODE 0x08 -#define SI_ADDR 0x10 - -// signalHandler is the signal handler for SIGSEGV and SIGBUS signals. It must -// not be set up as a handler to any other signals. -// -// If the instruction causing the signal is within a safecopy-protected -// function, the signal is handled such that execution resumes in the -// appropriate fault handling stub with R0 containing the faulting address and -// R1 containing the signal number. Otherwise control is transferred to the -// previously configured signal handler (savedSigSegvHandler or -// savedSigBusHandler). -// -// This function cannot be written in go because it runs whenever a signal is -// received by the thread (preempting whatever was running), which includes when -// garbage collector has stopped or isn't expecting any interactions (like -// barriers). -// -// The arguments are the following: -// R0 - The signal number. -// R1 - Pointer to siginfo_t structure. -// R2 - Pointer to ucontext structure. -TEXT ·signalHandler(SB),NOSPLIT,$0 - // Check if the signal is from the kernel, si_code > 0 means a kernel signal. - MOVD SI_CODE(R1), R7 - CMPW $0x0, R7 - BLE original_handler - - // Check if PC is within the area we care about. - MOVD REG_PC(R2), R7 - MOVD ·memcpyBegin(SB), R8 - CMP R8, R7 - BLO not_memcpy - MOVD ·memcpyEnd(SB), R8 - CMP R8, R7 - BHS not_memcpy - - // Modify the context such that execution will resume in the fault handler. - MOVD $handleMemcpyFault(SB), R7 - B handle_fault - -not_memcpy: - MOVD ·memclrBegin(SB), R8 - CMP R8, R7 - BLO not_memclr - MOVD ·memclrEnd(SB), R8 - CMP R8, R7 - BHS not_memclr - - MOVD $handleMemclrFault(SB), R7 - B handle_fault - -not_memclr: - MOVD ·swapUint32Begin(SB), R8 - CMP R8, R7 - BLO not_swapuint32 - MOVD ·swapUint32End(SB), R8 - CMP R8, R7 - BHS not_swapuint32 - - MOVD $handleSwapUint32Fault(SB), R7 - B handle_fault - -not_swapuint32: - MOVD ·swapUint64Begin(SB), R8 - CMP R8, R7 - BLO not_swapuint64 - MOVD ·swapUint64End(SB), R8 - CMP R8, R7 - BHS not_swapuint64 - - MOVD $handleSwapUint64Fault(SB), R7 - B handle_fault - -not_swapuint64: - MOVD ·compareAndSwapUint32Begin(SB), R8 - CMP R8, R7 - BLO not_casuint32 - MOVD ·compareAndSwapUint32End(SB), R8 - CMP R8, R7 - BHS not_casuint32 - - MOVD $handleCompareAndSwapUint32Fault(SB), R7 - B handle_fault - -not_casuint32: - MOVD ·loadUint32Begin(SB), R8 - CMP R8, R7 - BLO not_loaduint32 - MOVD ·loadUint32End(SB), R8 - CMP R8, R7 - BHS not_loaduint32 - - MOVD $handleLoadUint32Fault(SB), R7 - B handle_fault - -not_loaduint32: -original_handler: - // Jump to the previous signal handler, which is likely the golang one. - MOVD ·savedSigBusHandler(SB), R7 - MOVD ·savedSigSegVHandler(SB), R8 - CMPW $SIGSEGV, R0 - CSEL EQ, R8, R7, R7 - B (R7) - -handle_fault: - // Entered with the address of the fault handler in R7; store it in PC. - MOVD R7, REG_PC(R2) - - // Store the faulting address in R0. - MOVD SI_ADDR(R1), R7 - MOVD R7, REG_R0(R2) - - // Store the signal number in R1. - MOVW R0, REG_R1(R2) - - RET diff --git a/pkg/sentry/safemem/BUILD b/pkg/sentry/safemem/BUILD deleted file mode 100644 index 3ab76da97..000000000 --- a/pkg/sentry/safemem/BUILD +++ /dev/null @@ -1,27 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -package(licenses = ["notice"]) - -go_library( - name = "safemem", - srcs = [ - "block_unsafe.go", - "io.go", - "safemem.go", - "seq_unsafe.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/sentry/platform/safecopy", - ], -) - -go_test( - name = "safemem_test", - size = "small", - srcs = [ - "io_test.go", - "seq_test.go", - ], - library = ":safemem", -) diff --git a/pkg/sentry/safemem/block_unsafe.go b/pkg/sentry/safemem/block_unsafe.go deleted file mode 100644 index 6f03c94bf..000000000 --- a/pkg/sentry/safemem/block_unsafe.go +++ /dev/null @@ -1,279 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package safemem - -import ( - "fmt" - "reflect" - "unsafe" - - "gvisor.dev/gvisor/pkg/sentry/platform/safecopy" -) - -// A Block is a range of contiguous bytes, similar to []byte but with the -// following differences: -// -// - The memory represented by a Block may require the use of safecopy to -// access. -// -// - Block does not carry a capacity and cannot be expanded. -// -// Blocks are immutable and may be copied by value. The zero value of Block -// represents an empty range, analogous to a nil []byte. -type Block struct { - // [start, start+length) is the represented memory. - // - // start is an unsafe.Pointer to ensure that Block prevents the represented - // memory from being garbage-collected. - start unsafe.Pointer - length int - - // needSafecopy is true if accessing the represented memory requires the - // use of safecopy. - needSafecopy bool -} - -// BlockFromSafeSlice returns a Block equivalent to slice, which is safe to -// access without safecopy. -func BlockFromSafeSlice(slice []byte) Block { - return blockFromSlice(slice, false) -} - -// BlockFromUnsafeSlice returns a Block equivalent to bs, which is not safe to -// access without safecopy. -func BlockFromUnsafeSlice(slice []byte) Block { - return blockFromSlice(slice, true) -} - -func blockFromSlice(slice []byte, needSafecopy bool) Block { - if len(slice) == 0 { - return Block{} - } - return Block{ - start: unsafe.Pointer(&slice[0]), - length: len(slice), - needSafecopy: needSafecopy, - } -} - -// BlockFromSafePointer returns a Block equivalent to [ptr, ptr+len), which is -// safe to access without safecopy. -// -// Preconditions: ptr+len does not overflow. -func BlockFromSafePointer(ptr unsafe.Pointer, len int) Block { - return blockFromPointer(ptr, len, false) -} - -// BlockFromUnsafePointer returns a Block equivalent to [ptr, ptr+len), which -// is not safe to access without safecopy. -// -// Preconditions: ptr+len does not overflow. -func BlockFromUnsafePointer(ptr unsafe.Pointer, len int) Block { - return blockFromPointer(ptr, len, true) -} - -func blockFromPointer(ptr unsafe.Pointer, len int, needSafecopy bool) Block { - if uptr := uintptr(ptr); uptr+uintptr(len) < uptr { - panic(fmt.Sprintf("ptr %#x + len %#x overflows", ptr, len)) - } - return Block{ - start: ptr, - length: len, - needSafecopy: needSafecopy, - } -} - -// DropFirst returns a Block equivalent to b, but with the first n bytes -// omitted. It is analogous to the [n:] operation on a slice, except that if n -// > b.Len(), DropFirst returns an empty Block instead of panicking. -// -// Preconditions: n >= 0. -func (b Block) DropFirst(n int) Block { - if n < 0 { - panic(fmt.Sprintf("invalid n: %d", n)) - } - return b.DropFirst64(uint64(n)) -} - -// DropFirst64 is equivalent to DropFirst but takes a uint64. -func (b Block) DropFirst64(n uint64) Block { - if n >= uint64(b.length) { - return Block{} - } - return Block{ - start: unsafe.Pointer(uintptr(b.start) + uintptr(n)), - length: b.length - int(n), - needSafecopy: b.needSafecopy, - } -} - -// TakeFirst returns a Block equivalent to the first n bytes of b. It is -// analogous to the [:n] operation on a slice, except that if n > b.Len(), -// TakeFirst returns a copy of b instead of panicking. -// -// Preconditions: n >= 0. -func (b Block) TakeFirst(n int) Block { - if n < 0 { - panic(fmt.Sprintf("invalid n: %d", n)) - } - return b.TakeFirst64(uint64(n)) -} - -// TakeFirst64 is equivalent to TakeFirst but takes a uint64. -func (b Block) TakeFirst64(n uint64) Block { - if n == 0 { - return Block{} - } - if n >= uint64(b.length) { - return b - } - return Block{ - start: b.start, - length: int(n), - needSafecopy: b.needSafecopy, - } -} - -// ToSlice returns a []byte equivalent to b. -func (b Block) ToSlice() []byte { - var bs []byte - hdr := (*reflect.SliceHeader)(unsafe.Pointer(&bs)) - hdr.Data = uintptr(b.start) - hdr.Len = b.length - hdr.Cap = b.length - return bs -} - -// Addr returns b's start address as a uintptr. It returns uintptr instead of -// unsafe.Pointer so that code using safemem cannot obtain unsafe.Pointers -// without importing the unsafe package explicitly. -// -// Note that a uintptr is not recognized as a pointer by the garbage collector, -// such that if there are no uses of b after a call to b.Addr() and the address -// is to Go-managed memory, the returned uintptr does not prevent garbage -// collection of the pointee. -func (b Block) Addr() uintptr { - return uintptr(b.start) -} - -// Len returns b's length in bytes. -func (b Block) Len() int { - return b.length -} - -// NeedSafecopy returns true if accessing b.ToSlice() requires the use of safecopy. -func (b Block) NeedSafecopy() bool { - return b.needSafecopy -} - -// String implements fmt.Stringer.String. -func (b Block) String() string { - if uintptr(b.start) == 0 && b.length == 0 { - return "" - } - var suffix string - if b.needSafecopy { - suffix = "*" - } - return fmt.Sprintf("[%#x-%#x)%s", uintptr(b.start), uintptr(b.start)+uintptr(b.length), suffix) -} - -// Copy copies src.Len() or dst.Len() bytes, whichever is less, from src -// to dst and returns the number of bytes copied. -// -// If src and dst overlap, the data stored in dst is unspecified. -func Copy(dst, src Block) (int, error) { - if !dst.needSafecopy && !src.needSafecopy { - return copy(dst.ToSlice(), src.ToSlice()), nil - } - - n := dst.length - if n > src.length { - n = src.length - } - if n == 0 { - return 0, nil - } - - switch { - case dst.needSafecopy && !src.needSafecopy: - return safecopy.CopyOut(dst.start, src.TakeFirst(n).ToSlice()) - case !dst.needSafecopy && src.needSafecopy: - return safecopy.CopyIn(dst.TakeFirst(n).ToSlice(), src.start) - case dst.needSafecopy && src.needSafecopy: - n64, err := safecopy.Copy(dst.start, src.start, uintptr(n)) - return int(n64), err - default: - panic("unreachable") - } -} - -// Zero sets all bytes in dst to 0 and returns the number of bytes zeroed. -func Zero(dst Block) (int, error) { - if !dst.needSafecopy { - bs := dst.ToSlice() - for i := range bs { - bs[i] = 0 - } - return len(bs), nil - } - - n64, err := safecopy.ZeroOut(dst.start, uintptr(dst.length)) - return int(n64), err -} - -// Safecopy atomics are no slower than non-safecopy atomics, so use the former -// even when !b.needSafecopy to get consistent alignment checking. - -// SwapUint32 invokes safecopy.SwapUint32 on the first 4 bytes of b. -// -// Preconditions: b.Len() >= 4. -func SwapUint32(b Block, new uint32) (uint32, error) { - if b.length < 4 { - panic(fmt.Sprintf("insufficient length: %d", b.length)) - } - return safecopy.SwapUint32(b.start, new) -} - -// SwapUint64 invokes safecopy.SwapUint64 on the first 8 bytes of b. -// -// Preconditions: b.Len() >= 8. -func SwapUint64(b Block, new uint64) (uint64, error) { - if b.length < 8 { - panic(fmt.Sprintf("insufficient length: %d", b.length)) - } - return safecopy.SwapUint64(b.start, new) -} - -// CompareAndSwapUint32 invokes safecopy.CompareAndSwapUint32 on the first 4 -// bytes of b. -// -// Preconditions: b.Len() >= 4. -func CompareAndSwapUint32(b Block, old, new uint32) (uint32, error) { - if b.length < 4 { - panic(fmt.Sprintf("insufficient length: %d", b.length)) - } - return safecopy.CompareAndSwapUint32(b.start, old, new) -} - -// LoadUint32 invokes safecopy.LoadUint32 on the first 4 bytes of b. -// -// Preconditions: b.Len() >= 4. -func LoadUint32(b Block) (uint32, error) { - if b.length < 4 { - panic(fmt.Sprintf("insufficient length: %d", b.length)) - } - return safecopy.LoadUint32(b.start) -} diff --git a/pkg/sentry/safemem/io.go b/pkg/sentry/safemem/io.go deleted file mode 100644 index f039a5c34..000000000 --- a/pkg/sentry/safemem/io.go +++ /dev/null @@ -1,392 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package safemem - -import ( - "errors" - "io" - "math" -) - -// ErrEndOfBlockSeq is returned by BlockSeqWriter when attempting to write -// beyond the end of the BlockSeq. -var ErrEndOfBlockSeq = errors.New("write beyond end of BlockSeq") - -// Reader represents a streaming byte source like io.Reader. -type Reader interface { - // ReadToBlocks reads up to dsts.NumBytes() bytes into dsts and returns the - // number of bytes read. It may return a partial read without an error - // (i.e. (n, nil) where 0 < n < dsts.NumBytes()). It should not return a - // full read with an error (i.e. (dsts.NumBytes(), err) where err != nil); - // note that this differs from io.Reader.Read (in particular, io.EOF should - // not be returned if ReadToBlocks successfully reads dsts.NumBytes() - // bytes.) - ReadToBlocks(dsts BlockSeq) (uint64, error) -} - -// Writer represents a streaming byte sink like io.Writer. -type Writer interface { - // WriteFromBlocks writes up to srcs.NumBytes() bytes from srcs and returns - // the number of bytes written. It may return a partial write without an - // error (i.e. (n, nil) where 0 < n < srcs.NumBytes()). It should not - // return a full write with an error (i.e. srcs.NumBytes(), err) where err - // != nil). - WriteFromBlocks(srcs BlockSeq) (uint64, error) -} - -// ReadFullToBlocks repeatedly invokes r.ReadToBlocks until dsts.NumBytes() -// bytes have been read or ReadToBlocks returns an error. -func ReadFullToBlocks(r Reader, dsts BlockSeq) (uint64, error) { - var done uint64 - for !dsts.IsEmpty() { - n, err := r.ReadToBlocks(dsts) - done += n - if err != nil { - return done, err - } - dsts = dsts.DropFirst64(n) - } - return done, nil -} - -// WriteFullFromBlocks repeatedly invokes w.WriteFromBlocks until -// srcs.NumBytes() bytes have been written or WriteFromBlocks returns an error. -func WriteFullFromBlocks(w Writer, srcs BlockSeq) (uint64, error) { - var done uint64 - for !srcs.IsEmpty() { - n, err := w.WriteFromBlocks(srcs) - done += n - if err != nil { - return done, err - } - srcs = srcs.DropFirst64(n) - } - return done, nil -} - -// BlockSeqReader implements Reader by reading from a BlockSeq. -type BlockSeqReader struct { - Blocks BlockSeq -} - -// ReadToBlocks implements Reader.ReadToBlocks. -func (r *BlockSeqReader) ReadToBlocks(dsts BlockSeq) (uint64, error) { - n, err := CopySeq(dsts, r.Blocks) - r.Blocks = r.Blocks.DropFirst64(n) - if err != nil { - return n, err - } - if n < dsts.NumBytes() { - return n, io.EOF - } - return n, nil -} - -// BlockSeqWriter implements Writer by writing to a BlockSeq. -type BlockSeqWriter struct { - Blocks BlockSeq -} - -// WriteFromBlocks implements Writer.WriteFromBlocks. -func (w *BlockSeqWriter) WriteFromBlocks(srcs BlockSeq) (uint64, error) { - n, err := CopySeq(w.Blocks, srcs) - w.Blocks = w.Blocks.DropFirst64(n) - if err != nil { - return n, err - } - if n < srcs.NumBytes() { - return n, ErrEndOfBlockSeq - } - return n, nil -} - -// ReaderFunc implements Reader for a function with the semantics of -// Reader.ReadToBlocks. -type ReaderFunc func(dsts BlockSeq) (uint64, error) - -// ReadToBlocks implements Reader.ReadToBlocks. -func (f ReaderFunc) ReadToBlocks(dsts BlockSeq) (uint64, error) { - return f(dsts) -} - -// WriterFunc implements Writer for a function with the semantics of -// Writer.WriteFromBlocks. -type WriterFunc func(srcs BlockSeq) (uint64, error) - -// WriteFromBlocks implements Writer.WriteFromBlocks. -func (f WriterFunc) WriteFromBlocks(srcs BlockSeq) (uint64, error) { - return f(srcs) -} - -// ToIOReader implements io.Reader for a (safemem.)Reader. -// -// ToIOReader will return a successful partial read iff Reader.ReadToBlocks does -// so. -type ToIOReader struct { - Reader Reader -} - -// Read implements io.Reader.Read. -func (r ToIOReader) Read(dst []byte) (int, error) { - n, err := r.Reader.ReadToBlocks(BlockSeqOf(BlockFromSafeSlice(dst))) - return int(n), err -} - -// ToIOWriter implements io.Writer for a (safemem.)Writer. -type ToIOWriter struct { - Writer Writer -} - -// Write implements io.Writer.Write. -func (w ToIOWriter) Write(src []byte) (int, error) { - // io.Writer does not permit partial writes. - n, err := WriteFullFromBlocks(w.Writer, BlockSeqOf(BlockFromSafeSlice(src))) - return int(n), err -} - -// FromIOReader implements Reader for an io.Reader by repeatedly invoking -// io.Reader.Read until it returns an error or partial read. This is not -// thread-safe. -// -// FromIOReader will return a successful partial read iff Reader.Read does so. -type FromIOReader struct { - Reader io.Reader -} - -// ReadToBlocks implements Reader.ReadToBlocks. -func (r FromIOReader) ReadToBlocks(dsts BlockSeq) (uint64, error) { - var buf []byte - var done uint64 - for !dsts.IsEmpty() { - dst := dsts.Head() - var n int - var err error - n, buf, err = r.readToBlock(dst, buf) - done += uint64(n) - if n != dst.Len() { - return done, err - } - dsts = dsts.Tail() - if err != nil { - if dsts.IsEmpty() && err == io.EOF { - return done, nil - } - return done, err - } - } - return done, nil -} - -func (r FromIOReader) readToBlock(dst Block, buf []byte) (int, []byte, error) { - // io.Reader isn't safecopy-aware, so we have to buffer Blocks that require - // safecopy. - if !dst.NeedSafecopy() { - n, err := r.Reader.Read(dst.ToSlice()) - return n, buf, err - } - if len(buf) < dst.Len() { - buf = make([]byte, dst.Len()) - } - rn, rerr := r.Reader.Read(buf[:dst.Len()]) - wbn, wberr := Copy(dst, BlockFromSafeSlice(buf[:rn])) - if wberr != nil { - return wbn, buf, wberr - } - return wbn, buf, rerr -} - -// FromIOReaderAt implements Reader for an io.ReaderAt. Does not repeatedly -// invoke io.ReaderAt.ReadAt because ReadAt is more strict than Read. A partial -// read indicates an error. This is not thread-safe. -type FromIOReaderAt struct { - ReaderAt io.ReaderAt - Offset int64 -} - -// ReadToBlocks implements Reader.ReadToBlocks. -func (r FromIOReaderAt) ReadToBlocks(dsts BlockSeq) (uint64, error) { - var buf []byte - var done uint64 - for !dsts.IsEmpty() { - dst := dsts.Head() - var n int - var err error - n, buf, err = r.readToBlock(dst, buf) - done += uint64(n) - if n != dst.Len() { - return done, err - } - dsts = dsts.Tail() - if err != nil { - if dsts.IsEmpty() && err == io.EOF { - return done, nil - } - return done, err - } - } - return done, nil -} - -func (r FromIOReaderAt) readToBlock(dst Block, buf []byte) (int, []byte, error) { - // io.Reader isn't safecopy-aware, so we have to buffer Blocks that require - // safecopy. - if !dst.NeedSafecopy() { - n, err := r.ReaderAt.ReadAt(dst.ToSlice(), r.Offset) - r.Offset += int64(n) - return n, buf, err - } - if len(buf) < dst.Len() { - buf = make([]byte, dst.Len()) - } - rn, rerr := r.ReaderAt.ReadAt(buf[:dst.Len()], r.Offset) - r.Offset += int64(rn) - wbn, wberr := Copy(dst, BlockFromSafeSlice(buf[:rn])) - if wberr != nil { - return wbn, buf, wberr - } - return wbn, buf, rerr -} - -// FromIOWriter implements Writer for an io.Writer by repeatedly invoking -// io.Writer.Write until it returns an error or partial write. -// -// FromIOWriter will tolerate implementations of io.Writer.Write that return -// partial writes with a nil error in contravention of io.Writer's -// requirements, since Writer is permitted to do so. FromIOWriter will return a -// successful partial write iff Writer.Write does so. -type FromIOWriter struct { - Writer io.Writer -} - -// WriteFromBlocks implements Writer.WriteFromBlocks. -func (w FromIOWriter) WriteFromBlocks(srcs BlockSeq) (uint64, error) { - var buf []byte - var done uint64 - for !srcs.IsEmpty() { - src := srcs.Head() - var n int - var err error - n, buf, err = w.writeFromBlock(src, buf) - done += uint64(n) - if n != src.Len() || err != nil { - return done, err - } - srcs = srcs.Tail() - } - return done, nil -} - -func (w FromIOWriter) writeFromBlock(src Block, buf []byte) (int, []byte, error) { - // io.Writer isn't safecopy-aware, so we have to buffer Blocks that require - // safecopy. - if !src.NeedSafecopy() { - n, err := w.Writer.Write(src.ToSlice()) - return n, buf, err - } - if len(buf) < src.Len() { - buf = make([]byte, src.Len()) - } - bufn, buferr := Copy(BlockFromSafeSlice(buf[:src.Len()]), src) - wn, werr := w.Writer.Write(buf[:bufn]) - if werr != nil { - return wn, buf, werr - } - return wn, buf, buferr -} - -// FromVecReaderFunc implements Reader for a function that reads data into a -// [][]byte and returns the number of bytes read as an int64. -type FromVecReaderFunc struct { - ReadVec func(dsts [][]byte) (int64, error) -} - -// ReadToBlocks implements Reader.ReadToBlocks. -// -// ReadToBlocks calls r.ReadVec at most once. -func (r FromVecReaderFunc) ReadToBlocks(dsts BlockSeq) (uint64, error) { - if dsts.IsEmpty() { - return 0, nil - } - // Ensure that we don't pass a [][]byte with a total length > MaxInt64. - dsts = dsts.TakeFirst64(uint64(math.MaxInt64)) - dstSlices := make([][]byte, 0, dsts.NumBlocks()) - // Buffer Blocks that require safecopy. - for tmp := dsts; !tmp.IsEmpty(); tmp = tmp.Tail() { - dst := tmp.Head() - if dst.NeedSafecopy() { - dstSlices = append(dstSlices, make([]byte, dst.Len())) - } else { - dstSlices = append(dstSlices, dst.ToSlice()) - } - } - rn, rerr := r.ReadVec(dstSlices) - dsts = dsts.TakeFirst64(uint64(rn)) - var done uint64 - var i int - for !dsts.IsEmpty() { - dst := dsts.Head() - if dst.NeedSafecopy() { - n, err := Copy(dst, BlockFromSafeSlice(dstSlices[i])) - done += uint64(n) - if err != nil { - return done, err - } - } else { - done += uint64(dst.Len()) - } - dsts = dsts.Tail() - i++ - } - return done, rerr -} - -// FromVecWriterFunc implements Writer for a function that writes data from a -// [][]byte and returns the number of bytes written. -type FromVecWriterFunc struct { - WriteVec func(srcs [][]byte) (int64, error) -} - -// WriteFromBlocks implements Writer.WriteFromBlocks. -// -// WriteFromBlocks calls w.WriteVec at most once. -func (w FromVecWriterFunc) WriteFromBlocks(srcs BlockSeq) (uint64, error) { - if srcs.IsEmpty() { - return 0, nil - } - // Ensure that we don't pass a [][]byte with a total length > MaxInt64. - srcs = srcs.TakeFirst64(uint64(math.MaxInt64)) - srcSlices := make([][]byte, 0, srcs.NumBlocks()) - // Buffer Blocks that require safecopy. - var buferr error - for tmp := srcs; !tmp.IsEmpty(); tmp = tmp.Tail() { - src := tmp.Head() - if src.NeedSafecopy() { - slice := make([]byte, src.Len()) - n, err := Copy(BlockFromSafeSlice(slice), src) - srcSlices = append(srcSlices, slice[:n]) - if err != nil { - buferr = err - break - } - } else { - srcSlices = append(srcSlices, src.ToSlice()) - } - } - n, err := w.WriteVec(srcSlices) - if err != nil { - return uint64(n), err - } - return uint64(n), buferr -} diff --git a/pkg/sentry/safemem/io_test.go b/pkg/sentry/safemem/io_test.go deleted file mode 100644 index 629741bee..000000000 --- a/pkg/sentry/safemem/io_test.go +++ /dev/null @@ -1,199 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package safemem - -import ( - "bytes" - "io" - "testing" -) - -func makeBlocks(slices ...[]byte) []Block { - blocks := make([]Block, 0, len(slices)) - for _, s := range slices { - blocks = append(blocks, BlockFromSafeSlice(s)) - } - return blocks -} - -func TestFromIOReaderFullRead(t *testing.T) { - r := FromIOReader{bytes.NewBufferString("foobar")} - dsts := makeBlocks(make([]byte, 3), make([]byte, 3)) - n, err := r.ReadToBlocks(BlockSeqFromSlice(dsts)) - if wantN := uint64(6); n != wantN || err != nil { - t.Errorf("ReadToBlocks: got (%v, %v), wanted (%v, nil)", n, err, wantN) - } - for i, want := range [][]byte{[]byte("foo"), []byte("bar")} { - if got := dsts[i].ToSlice(); !bytes.Equal(got, want) { - t.Errorf("dsts[%d]: got %q, wanted %q", i, got, want) - } - } -} - -type eofHidingReader struct { - Reader io.Reader -} - -func (r eofHidingReader) Read(dst []byte) (int, error) { - n, err := r.Reader.Read(dst) - if err == io.EOF { - return n, nil - } - return n, err -} - -func TestFromIOReaderPartialRead(t *testing.T) { - r := FromIOReader{eofHidingReader{bytes.NewBufferString("foob")}} - dsts := makeBlocks(make([]byte, 3), make([]byte, 3)) - n, err := r.ReadToBlocks(BlockSeqFromSlice(dsts)) - // FromIOReader should stop after the eofHidingReader returns (1, nil) - // for a 3-byte read. - if wantN := uint64(4); n != wantN || err != nil { - t.Errorf("ReadToBlocks: got (%v, %v), wanted (%v, nil)", n, err, wantN) - } - for i, want := range [][]byte{[]byte("foo"), []byte("b\x00\x00")} { - if got := dsts[i].ToSlice(); !bytes.Equal(got, want) { - t.Errorf("dsts[%d]: got %q, wanted %q", i, got, want) - } - } -} - -type singleByteReader struct { - Reader io.Reader -} - -func (r singleByteReader) Read(dst []byte) (int, error) { - if len(dst) == 0 { - return r.Reader.Read(dst) - } - return r.Reader.Read(dst[:1]) -} - -func TestSingleByteReader(t *testing.T) { - r := FromIOReader{singleByteReader{bytes.NewBufferString("foobar")}} - dsts := makeBlocks(make([]byte, 3), make([]byte, 3)) - n, err := r.ReadToBlocks(BlockSeqFromSlice(dsts)) - // FromIOReader should stop after the singleByteReader returns (1, nil) - // for a 3-byte read. - if wantN := uint64(1); n != wantN || err != nil { - t.Errorf("ReadToBlocks: got (%v, %v), wanted (%v, nil)", n, err, wantN) - } - for i, want := range [][]byte{[]byte("f\x00\x00"), []byte("\x00\x00\x00")} { - if got := dsts[i].ToSlice(); !bytes.Equal(got, want) { - t.Errorf("dsts[%d]: got %q, wanted %q", i, got, want) - } - } -} - -func TestReadFullToBlocks(t *testing.T) { - r := FromIOReader{singleByteReader{bytes.NewBufferString("foobar")}} - dsts := makeBlocks(make([]byte, 3), make([]byte, 3)) - n, err := ReadFullToBlocks(r, BlockSeqFromSlice(dsts)) - // ReadFullToBlocks should call into FromIOReader => singleByteReader - // repeatedly until dsts is exhausted. - if wantN := uint64(6); n != wantN || err != nil { - t.Errorf("ReadFullToBlocks: got (%v, %v), wanted (%v, nil)", n, err, wantN) - } - for i, want := range [][]byte{[]byte("foo"), []byte("bar")} { - if got := dsts[i].ToSlice(); !bytes.Equal(got, want) { - t.Errorf("dsts[%d]: got %q, wanted %q", i, got, want) - } - } -} - -func TestFromIOWriterFullWrite(t *testing.T) { - srcs := makeBlocks([]byte("foo"), []byte("bar")) - var dst bytes.Buffer - w := FromIOWriter{&dst} - n, err := w.WriteFromBlocks(BlockSeqFromSlice(srcs)) - if wantN := uint64(6); n != wantN || err != nil { - t.Errorf("WriteFromBlocks: got (%v, %v), wanted (%v, nil)", n, err, wantN) - } - if got, want := dst.Bytes(), []byte("foobar"); !bytes.Equal(got, want) { - t.Errorf("dst: got %q, wanted %q", got, want) - } -} - -type limitedWriter struct { - Writer io.Writer - Done int - Limit int -} - -func (w *limitedWriter) Write(src []byte) (int, error) { - count := len(src) - if count > (w.Limit - w.Done) { - count = w.Limit - w.Done - } - n, err := w.Writer.Write(src[:count]) - w.Done += n - return n, err -} - -func TestFromIOWriterPartialWrite(t *testing.T) { - srcs := makeBlocks([]byte("foo"), []byte("bar")) - var dst bytes.Buffer - w := FromIOWriter{&limitedWriter{&dst, 0, 4}} - n, err := w.WriteFromBlocks(BlockSeqFromSlice(srcs)) - // FromIOWriter should stop after the limitedWriter returns (1, nil) for a - // 3-byte write. - if wantN := uint64(4); n != wantN || err != nil { - t.Errorf("WriteFromBlocks: got (%v, %v), wanted (%v, nil)", n, err, wantN) - } - if got, want := dst.Bytes(), []byte("foob"); !bytes.Equal(got, want) { - t.Errorf("dst: got %q, wanted %q", got, want) - } -} - -type singleByteWriter struct { - Writer io.Writer -} - -func (w singleByteWriter) Write(src []byte) (int, error) { - if len(src) == 0 { - return w.Writer.Write(src) - } - return w.Writer.Write(src[:1]) -} - -func TestSingleByteWriter(t *testing.T) { - srcs := makeBlocks([]byte("foo"), []byte("bar")) - var dst bytes.Buffer - w := FromIOWriter{singleByteWriter{&dst}} - n, err := w.WriteFromBlocks(BlockSeqFromSlice(srcs)) - // FromIOWriter should stop after the singleByteWriter returns (1, nil) - // for a 3-byte write. - if wantN := uint64(1); n != wantN || err != nil { - t.Errorf("WriteFromBlocks: got (%v, %v), wanted (%v, nil)", n, err, wantN) - } - if got, want := dst.Bytes(), []byte("f"); !bytes.Equal(got, want) { - t.Errorf("dst: got %q, wanted %q", got, want) - } -} - -func TestWriteFullToBlocks(t *testing.T) { - srcs := makeBlocks([]byte("foo"), []byte("bar")) - var dst bytes.Buffer - w := FromIOWriter{singleByteWriter{&dst}} - n, err := WriteFullFromBlocks(w, BlockSeqFromSlice(srcs)) - // WriteFullToBlocks should call into FromIOWriter => singleByteWriter - // repeatedly until srcs is exhausted. - if wantN := uint64(6); n != wantN || err != nil { - t.Errorf("WriteFullFromBlocks: got (%v, %v), wanted (%v, nil)", n, err, wantN) - } - if got, want := dst.Bytes(), []byte("foobar"); !bytes.Equal(got, want) { - t.Errorf("dst: got %q, wanted %q", got, want) - } -} diff --git a/pkg/sentry/safemem/safemem.go b/pkg/sentry/safemem/safemem.go deleted file mode 100644 index 3e70d33a2..000000000 --- a/pkg/sentry/safemem/safemem.go +++ /dev/null @@ -1,16 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package safemem provides the Block and BlockSeq types. -package safemem diff --git a/pkg/sentry/safemem/seq_test.go b/pkg/sentry/safemem/seq_test.go deleted file mode 100644 index eba4bb535..000000000 --- a/pkg/sentry/safemem/seq_test.go +++ /dev/null @@ -1,196 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package safemem - -import ( - "bytes" - "reflect" - "testing" -) - -type blockSeqTest struct { - desc string - - pieces []string - haveOffset bool - offset uint64 - haveLimit bool - limit uint64 - - want string -} - -func (t blockSeqTest) NonEmptyByteSlices() [][]byte { - // t is a value, so we can mutate it freely. - slices := make([][]byte, 0, len(t.pieces)) - for _, str := range t.pieces { - if t.haveOffset { - strOff := t.offset - if strOff > uint64(len(str)) { - strOff = uint64(len(str)) - } - str = str[strOff:] - t.offset -= strOff - } - if t.haveLimit { - strLim := t.limit - if strLim > uint64(len(str)) { - strLim = uint64(len(str)) - } - str = str[:strLim] - t.limit -= strLim - } - if len(str) != 0 { - slices = append(slices, []byte(str)) - } - } - return slices -} - -func (t blockSeqTest) BlockSeq() BlockSeq { - blocks := make([]Block, 0, len(t.pieces)) - for _, str := range t.pieces { - blocks = append(blocks, BlockFromSafeSlice([]byte(str))) - } - bs := BlockSeqFromSlice(blocks) - if t.haveOffset { - bs = bs.DropFirst64(t.offset) - } - if t.haveLimit { - bs = bs.TakeFirst64(t.limit) - } - return bs -} - -var blockSeqTests = []blockSeqTest{ - { - desc: "Empty sequence", - }, - { - desc: "Sequence of length 1", - pieces: []string{"foobar"}, - want: "foobar", - }, - { - desc: "Sequence of length 2", - pieces: []string{"foo", "bar"}, - want: "foobar", - }, - { - desc: "Empty Blocks", - pieces: []string{"", "foo", "", "", "bar", ""}, - want: "foobar", - }, - { - desc: "Sequence with non-zero offset", - pieces: []string{"foo", "bar"}, - haveOffset: true, - offset: 2, - want: "obar", - }, - { - desc: "Sequence with non-maximal limit", - pieces: []string{"foo", "bar"}, - haveLimit: true, - limit: 5, - want: "fooba", - }, - { - desc: "Sequence with offset and limit", - pieces: []string{"foo", "bar"}, - haveOffset: true, - offset: 2, - haveLimit: true, - limit: 3, - want: "oba", - }, -} - -func TestBlockSeqNumBytes(t *testing.T) { - for _, test := range blockSeqTests { - t.Run(test.desc, func(t *testing.T) { - if got, want := test.BlockSeq().NumBytes(), uint64(len(test.want)); got != want { - t.Errorf("NumBytes: got %d, wanted %d", got, want) - } - }) - } -} - -func TestBlockSeqIterBlocks(t *testing.T) { - // Tests BlockSeq iteration using Head/Tail. - for _, test := range blockSeqTests { - t.Run(test.desc, func(t *testing.T) { - srcs := test.BlockSeq() - // "Note that a non-nil empty slice and a nil slice ... are not - // deeply equal." - reflect - slices := make([][]byte, 0, 0) - for !srcs.IsEmpty() { - src := srcs.Head() - slices = append(slices, src.ToSlice()) - nextSrcs := srcs.Tail() - if got, want := nextSrcs.NumBytes(), srcs.NumBytes()-uint64(src.Len()); got != want { - t.Fatalf("%v.Tail(): got %v (%d bytes), wanted %d bytes", srcs, nextSrcs, got, want) - } - srcs = nextSrcs - } - if wantSlices := test.NonEmptyByteSlices(); !reflect.DeepEqual(slices, wantSlices) { - t.Errorf("Accumulated slices: got %v, wanted %v", slices, wantSlices) - } - }) - } -} - -func TestBlockSeqIterBytes(t *testing.T) { - // Tests BlockSeq iteration using Head/DropFirst. - for _, test := range blockSeqTests { - t.Run(test.desc, func(t *testing.T) { - srcs := test.BlockSeq() - var dst bytes.Buffer - for !srcs.IsEmpty() { - src := srcs.Head() - var b [1]byte - n, err := Copy(BlockFromSafeSlice(b[:]), src) - if n != 1 || err != nil { - t.Fatalf("Copy: got (%v, %v), wanted (1, nil)", n, err) - } - dst.WriteByte(b[0]) - nextSrcs := srcs.DropFirst(1) - if got, want := nextSrcs.NumBytes(), srcs.NumBytes()-1; got != want { - t.Fatalf("%v.DropFirst(1): got %v (%d bytes), wanted %d bytes", srcs, nextSrcs, got, want) - } - srcs = nextSrcs - } - if got := string(dst.Bytes()); got != test.want { - t.Errorf("Copied string: got %q, wanted %q", got, test.want) - } - }) - } -} - -func TestBlockSeqDropBeyondLimit(t *testing.T) { - blocks := []Block{BlockFromSafeSlice([]byte("123")), BlockFromSafeSlice([]byte("4"))} - bs := BlockSeqFromSlice(blocks) - if got, want := bs.NumBytes(), uint64(4); got != want { - t.Errorf("%v.NumBytes(): got %d, wanted %d", bs, got, want) - } - bs = bs.TakeFirst(1) - if got, want := bs.NumBytes(), uint64(1); got != want { - t.Errorf("%v.NumBytes(): got %d, wanted %d", bs, got, want) - } - bs = bs.DropFirst(2) - if got, want := bs.NumBytes(), uint64(0); got != want { - t.Errorf("%v.NumBytes(): got %d, wanted %d", bs, got, want) - } -} diff --git a/pkg/sentry/safemem/seq_unsafe.go b/pkg/sentry/safemem/seq_unsafe.go deleted file mode 100644 index 354a95dde..000000000 --- a/pkg/sentry/safemem/seq_unsafe.go +++ /dev/null @@ -1,299 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package safemem - -import ( - "bytes" - "fmt" - "reflect" - "unsafe" -) - -// A BlockSeq represents a sequence of Blocks, each of which has non-zero -// length. -// -// BlockSeqs are immutable and may be copied by value. The zero value of -// BlockSeq represents an empty sequence. -type BlockSeq struct { - // If length is 0, then the BlockSeq is empty. Invariants: data == 0; - // offset == 0; limit == 0. - // - // If length is -1, then the BlockSeq represents the single Block{data, - // limit, false}. Invariants: offset == 0; limit > 0; limit does not - // overflow the range of an int. - // - // If length is -2, then the BlockSeq represents the single Block{data, - // limit, true}. Invariants: offset == 0; limit > 0; limit does not - // overflow the range of an int. - // - // Otherwise, length >= 2, and the BlockSeq represents the `length` Blocks - // in the array of Blocks starting at address `data`, starting at `offset` - // bytes into the first Block and limited to the following `limit` bytes. - // Invariants: data != 0; offset < len(data[0]); limit > 0; offset+limit <= - // the combined length of all Blocks in the array; the first Block in the - // array has non-zero length. - // - // length is never 1; sequences consisting of a single Block are always - // stored inline (with length < 0). - data unsafe.Pointer - length int - offset int - limit uint64 -} - -// BlockSeqOf returns a BlockSeq representing the single Block b. -func BlockSeqOf(b Block) BlockSeq { - bs := BlockSeq{ - data: b.start, - length: -1, - limit: uint64(b.length), - } - if b.needSafecopy { - bs.length = -2 - } - return bs -} - -// BlockSeqFromSlice returns a BlockSeq representing all Blocks in slice. -// If slice contains Blocks with zero length, BlockSeq will skip them during -// iteration. -// -// Whether the returned BlockSeq shares memory with slice is unspecified; -// clients should avoid mutating slices passed to BlockSeqFromSlice. -// -// Preconditions: The combined length of all Blocks in slice <= math.MaxUint64. -func BlockSeqFromSlice(slice []Block) BlockSeq { - slice = skipEmpty(slice) - var limit uint64 - for _, b := range slice { - sum := limit + uint64(b.Len()) - if sum < limit { - panic("BlockSeq length overflows uint64") - } - limit = sum - } - return blockSeqFromSliceLimited(slice, limit) -} - -// Preconditions: The combined length of all Blocks in slice <= limit. If -// len(slice) != 0, the first Block in slice has non-zero length, and limit > -// 0. -func blockSeqFromSliceLimited(slice []Block, limit uint64) BlockSeq { - switch len(slice) { - case 0: - return BlockSeq{} - case 1: - return BlockSeqOf(slice[0].TakeFirst64(limit)) - default: - return BlockSeq{ - data: unsafe.Pointer(&slice[0]), - length: len(slice), - limit: limit, - } - } -} - -func skipEmpty(slice []Block) []Block { - for i, b := range slice { - if b.Len() != 0 { - return slice[i:] - } - } - return nil -} - -// IsEmpty returns true if bs contains no Blocks. -// -// Invariants: bs.IsEmpty() == (bs.NumBlocks() == 0) == (bs.NumBytes() == 0). -// (Of these, prefer to use bs.IsEmpty().) -func (bs BlockSeq) IsEmpty() bool { - return bs.length == 0 -} - -// NumBlocks returns the number of Blocks in bs. -func (bs BlockSeq) NumBlocks() int { - // In general, we have to count: if bs represents a windowed slice then the - // slice may contain Blocks with zero length, and bs.length may be larger - // than the actual number of Blocks due to bs.limit. - var n int - for !bs.IsEmpty() { - n++ - bs = bs.Tail() - } - return n -} - -// NumBytes returns the sum of Block.Len() for all Blocks in bs. -func (bs BlockSeq) NumBytes() uint64 { - return bs.limit -} - -// Head returns the first Block in bs. -// -// Preconditions: !bs.IsEmpty(). -func (bs BlockSeq) Head() Block { - if bs.length == 0 { - panic("empty BlockSeq") - } - if bs.length < 0 { - return bs.internalBlock() - } - return (*Block)(bs.data).DropFirst(bs.offset).TakeFirst64(bs.limit) -} - -// Preconditions: bs.length < 0. -func (bs BlockSeq) internalBlock() Block { - return Block{ - start: bs.data, - length: int(bs.limit), - needSafecopy: bs.length == -2, - } -} - -// Tail returns a BlockSeq consisting of all Blocks in bs after the first. -// -// Preconditions: !bs.IsEmpty(). -func (bs BlockSeq) Tail() BlockSeq { - if bs.length == 0 { - panic("empty BlockSeq") - } - if bs.length < 0 { - return BlockSeq{} - } - head := (*Block)(bs.data).DropFirst(bs.offset) - headLen := uint64(head.Len()) - if headLen >= bs.limit { - // The head Block exhausts the limit, so the tail is empty. - return BlockSeq{} - } - var extSlice []Block - extSliceHdr := (*reflect.SliceHeader)(unsafe.Pointer(&extSlice)) - extSliceHdr.Data = uintptr(bs.data) - extSliceHdr.Len = bs.length - extSliceHdr.Cap = bs.length - tailSlice := skipEmpty(extSlice[1:]) - tailLimit := bs.limit - headLen - return blockSeqFromSliceLimited(tailSlice, tailLimit) -} - -// DropFirst returns a BlockSeq equivalent to bs, but with the first n bytes -// omitted. If n > bs.NumBytes(), DropFirst returns an empty BlockSeq. -// -// Preconditions: n >= 0. -func (bs BlockSeq) DropFirst(n int) BlockSeq { - if n < 0 { - panic(fmt.Sprintf("invalid n: %d", n)) - } - return bs.DropFirst64(uint64(n)) -} - -// DropFirst64 is equivalent to DropFirst but takes an uint64. -func (bs BlockSeq) DropFirst64(n uint64) BlockSeq { - if n >= bs.limit { - return BlockSeq{} - } - for { - // Calling bs.Head() here is surprisingly expensive, so inline getting - // the head's length. - var headLen uint64 - if bs.length < 0 { - headLen = bs.limit - } else { - headLen = uint64((*Block)(bs.data).Len() - bs.offset) - } - if n < headLen { - // Dropping ends partway through the head Block. - if bs.length < 0 { - return BlockSeqOf(bs.internalBlock().DropFirst64(n)) - } - bs.offset += int(n) - bs.limit -= n - return bs - } - n -= headLen - bs = bs.Tail() - } -} - -// TakeFirst returns a BlockSeq equivalent to the first n bytes of bs. If n > -// bs.NumBytes(), TakeFirst returns a BlockSeq equivalent to bs. -// -// Preconditions: n >= 0. -func (bs BlockSeq) TakeFirst(n int) BlockSeq { - if n < 0 { - panic(fmt.Sprintf("invalid n: %d", n)) - } - return bs.TakeFirst64(uint64(n)) -} - -// TakeFirst64 is equivalent to TakeFirst but takes a uint64. -func (bs BlockSeq) TakeFirst64(n uint64) BlockSeq { - if n == 0 { - return BlockSeq{} - } - if bs.limit > n { - bs.limit = n - } - return bs -} - -// String implements fmt.Stringer.String. -func (bs BlockSeq) String() string { - var buf bytes.Buffer - buf.WriteByte('[') - var sep string - for !bs.IsEmpty() { - buf.WriteString(sep) - sep = " " - buf.WriteString(bs.Head().String()) - bs = bs.Tail() - } - buf.WriteByte(']') - return buf.String() -} - -// CopySeq copies srcs.NumBytes() or dsts.NumBytes() bytes, whichever is less, -// from srcs to dsts and returns the number of bytes copied. -// -// If srcs and dsts overlap, the data stored in dsts is unspecified. -func CopySeq(dsts, srcs BlockSeq) (uint64, error) { - var done uint64 - for !dsts.IsEmpty() && !srcs.IsEmpty() { - dst := dsts.Head() - src := srcs.Head() - n, err := Copy(dst, src) - done += uint64(n) - if err != nil { - return done, err - } - dsts = dsts.DropFirst(n) - srcs = srcs.DropFirst(n) - } - return done, nil -} - -// ZeroSeq sets all bytes in dsts to 0 and returns the number of bytes zeroed. -func ZeroSeq(dsts BlockSeq) (uint64, error) { - var done uint64 - for !dsts.IsEmpty() { - n, err := Zero(dsts.Head()) - done += uint64(n) - if err != nil { - return done, err - } - dsts = dsts.DropFirst(n) - } - return done, nil -} diff --git a/pkg/sentry/socket/BUILD b/pkg/sentry/socket/BUILD index 8e2b97afb..611fa22c3 100644 --- a/pkg/sentry/socket/BUILD +++ b/pkg/sentry/socket/BUILD @@ -9,15 +9,15 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/binary", - "//pkg/sentry/context", + "//pkg/context", "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", "//pkg/sentry/kernel", "//pkg/sentry/kernel/time", "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/usermem", "//pkg/syserr", "//pkg/tcpip", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/socket/control/BUILD b/pkg/sentry/socket/control/BUILD index 3850f6345..79e16d6e8 100644 --- a/pkg/sentry/socket/control/BUILD +++ b/pkg/sentry/socket/control/BUILD @@ -12,13 +12,13 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/binary", - "//pkg/sentry/context", + "//pkg/context", "//pkg/sentry/fs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/socket", "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/usermem", "//pkg/syserror", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/socket/control/control.go b/pkg/sentry/socket/control/control.go index 1684dfc24..00265f15b 100644 --- a/pkg/sentry/socket/control/control.go +++ b/pkg/sentry/socket/control/control.go @@ -19,14 +19,14 @@ package control import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) const maxInt = int(^uint(0) >> 1) diff --git a/pkg/sentry/socket/hostinet/BUILD b/pkg/sentry/socket/hostinet/BUILD index 42bf7be6a..5a07d5d0e 100644 --- a/pkg/sentry/socket/hostinet/BUILD +++ b/pkg/sentry/socket/hostinet/BUILD @@ -16,23 +16,23 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/binary", + "//pkg/context", "//pkg/fdnotifier", "//pkg/log", + "//pkg/safemem", "//pkg/sentry/arch", - "//pkg/sentry/context", "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", "//pkg/sentry/inet", "//pkg/sentry/kernel", "//pkg/sentry/kernel/time", - "//pkg/sentry/safemem", "//pkg/sentry/socket", "//pkg/sentry/socket/control", - "//pkg/sentry/usermem", "//pkg/syserr", "//pkg/syserror", "//pkg/tcpip/stack", + "//pkg/usermem", "//pkg/waiter", "@org_golang_x_sys//unix:go_default_library", ], diff --git a/pkg/sentry/socket/hostinet/socket.go b/pkg/sentry/socket/hostinet/socket.go index c957b0f1d..bde4c7a1e 100644 --- a/pkg/sentry/socket/hostinet/socket.go +++ b/pkg/sentry/socket/hostinet/socket.go @@ -21,19 +21,19 @@ import ( "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fdnotifier" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/kernel" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/control" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/socket/hostinet/socket_unsafe.go b/pkg/sentry/socket/hostinet/socket_unsafe.go index e69ec38c2..cd67234d2 100644 --- a/pkg/sentry/socket/hostinet/socket_unsafe.go +++ b/pkg/sentry/socket/hostinet/socket_unsafe.go @@ -19,14 +19,14 @@ import ( "unsafe" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/socket" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) func firstBytePtr(bs []byte) unsafe.Pointer { diff --git a/pkg/sentry/socket/hostinet/stack.go b/pkg/sentry/socket/hostinet/stack.go index e67b46c9e..034eca676 100644 --- a/pkg/sentry/socket/hostinet/stack.go +++ b/pkg/sentry/socket/hostinet/stack.go @@ -25,13 +25,13 @@ import ( "syscall" "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/inet" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip/stack" + "gvisor.dev/gvisor/pkg/usermem" ) var defaultRecvBufSize = inet.TCPBufferSize{ diff --git a/pkg/sentry/socket/netfilter/BUILD b/pkg/sentry/socket/netfilter/BUILD index ed34a8308..fa2a2cb66 100644 --- a/pkg/sentry/socket/netfilter/BUILD +++ b/pkg/sentry/socket/netfilter/BUILD @@ -15,10 +15,10 @@ go_library( "//pkg/binary", "//pkg/log", "//pkg/sentry/kernel", - "//pkg/sentry/usermem", "//pkg/syserr", "//pkg/tcpip", "//pkg/tcpip/iptables", "//pkg/tcpip/stack", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index c65c36081..6ef740463 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -23,11 +23,11 @@ import ( "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/iptables" "gvisor.dev/gvisor/pkg/tcpip/stack" + "gvisor.dev/gvisor/pkg/usermem" ) // errorTargetName is used to mark targets as error targets. Error targets diff --git a/pkg/sentry/socket/netlink/BUILD b/pkg/sentry/socket/netlink/BUILD index baaac13c6..f8b8e467d 100644 --- a/pkg/sentry/socket/netlink/BUILD +++ b/pkg/sentry/socket/netlink/BUILD @@ -13,8 +13,8 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/binary", + "//pkg/context", "//pkg/sentry/arch", - "//pkg/sentry/context", "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", @@ -25,11 +25,11 @@ go_library( "//pkg/sentry/socket/netlink/port", "//pkg/sentry/socket/unix", "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/usermem", "//pkg/sync", "//pkg/syserr", "//pkg/syserror", "//pkg/tcpip", + "//pkg/usermem", "//pkg/waiter", ], ) diff --git a/pkg/sentry/socket/netlink/message.go b/pkg/sentry/socket/netlink/message.go index ce0a1afd0..b21e0ca4b 100644 --- a/pkg/sentry/socket/netlink/message.go +++ b/pkg/sentry/socket/netlink/message.go @@ -20,7 +20,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // alignUp rounds a length up to an alignment. diff --git a/pkg/sentry/socket/netlink/provider.go b/pkg/sentry/socket/netlink/provider.go index be005df24..07f860a49 100644 --- a/pkg/sentry/socket/netlink/provider.go +++ b/pkg/sentry/socket/netlink/provider.go @@ -18,7 +18,7 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/socket" diff --git a/pkg/sentry/socket/netlink/route/BUILD b/pkg/sentry/socket/netlink/route/BUILD index 2137c7aeb..0234aadde 100644 --- a/pkg/sentry/socket/netlink/route/BUILD +++ b/pkg/sentry/socket/netlink/route/BUILD @@ -8,7 +8,7 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", - "//pkg/sentry/context", + "//pkg/context", "//pkg/sentry/inet", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", diff --git a/pkg/sentry/socket/netlink/route/protocol.go b/pkg/sentry/socket/netlink/route/protocol.go index 6b4a0ecf4..80a15d6cb 100644 --- a/pkg/sentry/socket/netlink/route/protocol.go +++ b/pkg/sentry/socket/netlink/route/protocol.go @@ -19,7 +19,7 @@ import ( "bytes" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" diff --git a/pkg/sentry/socket/netlink/socket.go b/pkg/sentry/socket/netlink/socket.go index cea56f4ed..c4b95debb 100644 --- a/pkg/sentry/socket/netlink/socket.go +++ b/pkg/sentry/socket/netlink/socket.go @@ -20,8 +20,8 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/device" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" @@ -32,11 +32,11 @@ import ( "gvisor.dev/gvisor/pkg/sentry/socket/netlink/port" "gvisor.dev/gvisor/pkg/sentry/socket/unix" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/socket/netlink/uevent/BUILD b/pkg/sentry/socket/netlink/uevent/BUILD index 73fbdf1eb..b6434923c 100644 --- a/pkg/sentry/socket/netlink/uevent/BUILD +++ b/pkg/sentry/socket/netlink/uevent/BUILD @@ -8,7 +8,7 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", - "//pkg/sentry/context", + "//pkg/context", "//pkg/sentry/kernel", "//pkg/sentry/socket/netlink", "//pkg/syserr", diff --git a/pkg/sentry/socket/netlink/uevent/protocol.go b/pkg/sentry/socket/netlink/uevent/protocol.go index b5d7808d7..1ee4296bc 100644 --- a/pkg/sentry/socket/netlink/uevent/protocol.go +++ b/pkg/sentry/socket/netlink/uevent/protocol.go @@ -20,7 +20,7 @@ package uevent import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/socket/netlink" "gvisor.dev/gvisor/pkg/syserr" diff --git a/pkg/sentry/socket/netstack/BUILD b/pkg/sentry/socket/netstack/BUILD index e3d1f90cb..ab01cb4fa 100644 --- a/pkg/sentry/socket/netstack/BUILD +++ b/pkg/sentry/socket/netstack/BUILD @@ -17,10 +17,11 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/binary", + "//pkg/context", "//pkg/log", "//pkg/metric", + "//pkg/safemem", "//pkg/sentry/arch", - "//pkg/sentry/context", "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", @@ -28,11 +29,9 @@ go_library( "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/time", - "//pkg/sentry/safemem", "//pkg/sentry/socket", "//pkg/sentry/socket/netfilter", "//pkg/sentry/unimpl", - "//pkg/sentry/usermem", "//pkg/sync", "//pkg/syserr", "//pkg/syserror", @@ -45,6 +44,7 @@ go_library( "//pkg/tcpip/stack", "//pkg/tcpip/transport/tcp", "//pkg/tcpip/transport/udp", + "//pkg/usermem", "//pkg/waiter", ], ) diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 318acbeff..8619cc506 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -34,20 +34,19 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/metric" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/netfilter" "gvisor.dev/gvisor/pkg/sentry/unimpl" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" @@ -57,6 +56,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/tcpip/transport/tcp" "gvisor.dev/gvisor/pkg/tcpip/transport/udp" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/socket/netstack/provider.go b/pkg/sentry/socket/netstack/provider.go index 2d2c1ba2a..5afff2564 100644 --- a/pkg/sentry/socket/netstack/provider.go +++ b/pkg/sentry/socket/netstack/provider.go @@ -18,7 +18,7 @@ import ( "syscall" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" diff --git a/pkg/sentry/socket/socket.go b/pkg/sentry/socket/socket.go index 2389a9cdb..50d9744e6 100644 --- a/pkg/sentry/socket/socket.go +++ b/pkg/sentry/socket/socket.go @@ -24,16 +24,16 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/device" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/kernel" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/usermem" ) // ControlMessages represents the union of unix control messages and tcpip diff --git a/pkg/sentry/socket/unix/BUILD b/pkg/sentry/socket/unix/BUILD index bade18686..08743deba 100644 --- a/pkg/sentry/socket/unix/BUILD +++ b/pkg/sentry/socket/unix/BUILD @@ -12,23 +12,23 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/refs", + "//pkg/safemem", "//pkg/sentry/arch", - "//pkg/sentry/context", "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", "//pkg/sentry/kernel", "//pkg/sentry/kernel/time", - "//pkg/sentry/safemem", "//pkg/sentry/socket", "//pkg/sentry/socket/control", "//pkg/sentry/socket/netstack", "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/usermem", "//pkg/syserr", "//pkg/syserror", "//pkg/tcpip", + "//pkg/usermem", "//pkg/waiter", ], ) diff --git a/pkg/sentry/socket/unix/io.go b/pkg/sentry/socket/unix/io.go index 2447f24ef..129949990 100644 --- a/pkg/sentry/socket/unix/io.go +++ b/pkg/sentry/socket/unix/io.go @@ -15,8 +15,8 @@ package unix import ( - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/safemem" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/tcpip" ) diff --git a/pkg/sentry/socket/unix/transport/BUILD b/pkg/sentry/socket/unix/transport/BUILD index 4bdfc9208..74bcd6300 100644 --- a/pkg/sentry/socket/unix/transport/BUILD +++ b/pkg/sentry/socket/unix/transport/BUILD @@ -28,9 +28,9 @@ go_library( visibility = ["//:sandbox"], deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/ilist", "//pkg/refs", - "//pkg/sentry/context", "//pkg/sync", "//pkg/syserr", "//pkg/tcpip", diff --git a/pkg/sentry/socket/unix/transport/connectioned.go b/pkg/sentry/socket/unix/transport/connectioned.go index 9e6fbc111..ce5b94ee7 100644 --- a/pkg/sentry/socket/unix/transport/connectioned.go +++ b/pkg/sentry/socket/unix/transport/connectioned.go @@ -16,7 +16,7 @@ package transport import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/tcpip" diff --git a/pkg/sentry/socket/unix/transport/connectionless.go b/pkg/sentry/socket/unix/transport/connectionless.go index 0322dec0b..4b06d63ac 100644 --- a/pkg/sentry/socket/unix/transport/connectionless.go +++ b/pkg/sentry/socket/unix/transport/connectionless.go @@ -16,7 +16,7 @@ package transport import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/waiter" diff --git a/pkg/sentry/socket/unix/transport/unix.go b/pkg/sentry/socket/unix/transport/unix.go index fcc0da332..dcbafe0e5 100644 --- a/pkg/sentry/socket/unix/transport/unix.go +++ b/pkg/sentry/socket/unix/transport/unix.go @@ -19,7 +19,7 @@ import ( "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/tcpip" diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go index 7f49ba864..4d30aa714 100644 --- a/pkg/sentry/socket/unix/unix.go +++ b/pkg/sentry/socket/unix/unix.go @@ -22,9 +22,9 @@ import ( "syscall" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -33,10 +33,10 @@ import ( "gvisor.dev/gvisor/pkg/sentry/socket/control" "gvisor.dev/gvisor/pkg/sentry/socket/netstack" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/strace/BUILD b/pkg/sentry/strace/BUILD index ff6fafa63..762a946fe 100644 --- a/pkg/sentry/strace/BUILD +++ b/pkg/sentry/strace/BUILD @@ -34,7 +34,7 @@ go_library( "//pkg/sentry/socket/netlink", "//pkg/sentry/socket/netstack", "//pkg/sentry/syscalls/linux", - "//pkg/sentry/usermem", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/strace/poll.go b/pkg/sentry/strace/poll.go index 5187594a7..074e80f9b 100644 --- a/pkg/sentry/strace/poll.go +++ b/pkg/sentry/strace/poll.go @@ -22,7 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/kernel" slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // PollEventSet is the set of poll(2) event flags. diff --git a/pkg/sentry/strace/select.go b/pkg/sentry/strace/select.go index c77d418e6..3a4c32aa0 100644 --- a/pkg/sentry/strace/select.go +++ b/pkg/sentry/strace/select.go @@ -19,7 +19,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) func fdsFromSet(t *kernel.Task, set []byte) []int { diff --git a/pkg/sentry/strace/signal.go b/pkg/sentry/strace/signal.go index 5656d53eb..c41f36e3f 100644 --- a/pkg/sentry/strace/signal.go +++ b/pkg/sentry/strace/signal.go @@ -21,7 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/abi" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // signalNames contains the names of all named signals. diff --git a/pkg/sentry/strace/socket.go b/pkg/sentry/strace/socket.go index b6d7177f4..d2079c85f 100644 --- a/pkg/sentry/strace/socket.go +++ b/pkg/sentry/strace/socket.go @@ -26,7 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/socket/netlink" "gvisor.dev/gvisor/pkg/sentry/socket/netstack" slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // SocketFamily are the possible socket(2) families. diff --git a/pkg/sentry/strace/strace.go b/pkg/sentry/strace/strace.go index 629c1f308..3fc4a47fc 100644 --- a/pkg/sentry/strace/strace.go +++ b/pkg/sentry/strace/strace.go @@ -33,7 +33,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" pb "gvisor.dev/gvisor/pkg/sentry/strace/strace_go_proto" slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // DefaultLogMaximumSize is the default LogMaximumSize. diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD index 7d74e0f70..8d6c52850 100644 --- a/pkg/sentry/syscalls/linux/BUILD +++ b/pkg/sentry/syscalls/linux/BUILD @@ -63,11 +63,12 @@ go_library( "//pkg/abi/linux", "//pkg/binary", "//pkg/bpf", + "//pkg/context", "//pkg/log", "//pkg/metric", "//pkg/rand", + "//pkg/safemem", "//pkg/sentry/arch", - "//pkg/sentry/context", "//pkg/sentry/fs", "//pkg/sentry/fs/anon", "//pkg/sentry/fs/lock", @@ -87,16 +88,15 @@ go_library( "//pkg/sentry/loader", "//pkg/sentry/memmap", "//pkg/sentry/mm", - "//pkg/sentry/safemem", "//pkg/sentry/socket", "//pkg/sentry/socket/control", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/syscalls", "//pkg/sentry/usage", - "//pkg/sentry/usermem", "//pkg/sync", "//pkg/syserr", "//pkg/syserror", + "//pkg/usermem", "//pkg/waiter", ], ) diff --git a/pkg/sentry/syscalls/linux/linux64_amd64.go b/pkg/sentry/syscalls/linux/linux64_amd64.go index c76771a54..7435b50bf 100644 --- a/pkg/sentry/syscalls/linux/linux64_amd64.go +++ b/pkg/sentry/syscalls/linux/linux64_amd64.go @@ -20,8 +20,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/syscalls" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // AMD64 is a table of Linux amd64 syscall API with the corresponding syscall diff --git a/pkg/sentry/syscalls/linux/linux64_arm64.go b/pkg/sentry/syscalls/linux/linux64_arm64.go index d3587fda6..03a39fe65 100644 --- a/pkg/sentry/syscalls/linux/linux64_arm64.go +++ b/pkg/sentry/syscalls/linux/linux64_arm64.go @@ -20,8 +20,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/syscalls" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // ARM64 is a table of Linux arm64 syscall API with the corresponding syscall diff --git a/pkg/sentry/syscalls/linux/sigset.go b/pkg/sentry/syscalls/linux/sigset.go index 333013d8c..2ddb2b146 100644 --- a/pkg/sentry/syscalls/linux/sigset.go +++ b/pkg/sentry/syscalls/linux/sigset.go @@ -17,8 +17,8 @@ package linux import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // copyInSigSet copies in a sigset_t, checks its size, and ensures that KILL and diff --git a/pkg/sentry/syscalls/linux/sys_aio.go b/pkg/sentry/syscalls/linux/sys_aio.go index f56411bfe..b401978db 100644 --- a/pkg/sentry/syscalls/linux/sys_aio.go +++ b/pkg/sentry/syscalls/linux/sys_aio.go @@ -23,8 +23,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/eventfd" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/mm" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // I/O commands. diff --git a/pkg/sentry/syscalls/linux/sys_epoll.go b/pkg/sentry/syscalls/linux/sys_epoll.go index 65b4a227b..5f11b496c 100644 --- a/pkg/sentry/syscalls/linux/sys_epoll.go +++ b/pkg/sentry/syscalls/linux/sys_epoll.go @@ -20,8 +20,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/epoll" "gvisor.dev/gvisor/pkg/sentry/syscalls" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go index 9bc2445a5..c54735148 100644 --- a/pkg/sentry/syscalls/linux/sys_file.go +++ b/pkg/sentry/syscalls/linux/sys_file.go @@ -18,8 +18,8 @@ import ( "syscall" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/fs/tmpfs" @@ -28,8 +28,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/fasync" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/limits" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // fileOpAt performs an operation on the second last component in the path. diff --git a/pkg/sentry/syscalls/linux/sys_futex.go b/pkg/sentry/syscalls/linux/sys_futex.go index bde17a767..b68261f72 100644 --- a/pkg/sentry/syscalls/linux/sys_futex.go +++ b/pkg/sentry/syscalls/linux/sys_futex.go @@ -21,8 +21,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // futexWaitRestartBlock encapsulates the state required to restart futex(2) diff --git a/pkg/sentry/syscalls/linux/sys_getdents.go b/pkg/sentry/syscalls/linux/sys_getdents.go index 912cbe4ff..f66f4ffde 100644 --- a/pkg/sentry/syscalls/linux/sys_getdents.go +++ b/pkg/sentry/syscalls/linux/sys_getdents.go @@ -23,8 +23,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // Getdents implements linux syscall getdents(2) for 64bit systems. diff --git a/pkg/sentry/syscalls/linux/sys_mempolicy.go b/pkg/sentry/syscalls/linux/sys_mempolicy.go index f5a519d8a..ac934dc6f 100644 --- a/pkg/sentry/syscalls/linux/sys_mempolicy.go +++ b/pkg/sentry/syscalls/linux/sys_mempolicy.go @@ -20,8 +20,8 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // We unconditionally report a single NUMA node. This also means that our diff --git a/pkg/sentry/syscalls/linux/sys_mmap.go b/pkg/sentry/syscalls/linux/sys_mmap.go index 58a05b5bb..9959f6e61 100644 --- a/pkg/sentry/syscalls/linux/sys_mmap.go +++ b/pkg/sentry/syscalls/linux/sys_mmap.go @@ -22,8 +22,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/mm" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // Brk implements linux syscall brk(2). diff --git a/pkg/sentry/syscalls/linux/sys_mount.go b/pkg/sentry/syscalls/linux/sys_mount.go index 8c13e2d82..eb5ff48f5 100644 --- a/pkg/sentry/syscalls/linux/sys_mount.go +++ b/pkg/sentry/syscalls/linux/sys_mount.go @@ -19,8 +19,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // Mount implements Linux syscall mount(2). diff --git a/pkg/sentry/syscalls/linux/sys_pipe.go b/pkg/sentry/syscalls/linux/sys_pipe.go index 418d7fa5f..798344042 100644 --- a/pkg/sentry/syscalls/linux/sys_pipe.go +++ b/pkg/sentry/syscalls/linux/sys_pipe.go @@ -20,8 +20,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // pipe2 implements the actual system call with flags. diff --git a/pkg/sentry/syscalls/linux/sys_poll.go b/pkg/sentry/syscalls/linux/sys_poll.go index 2b2df989a..4f8762d7d 100644 --- a/pkg/sentry/syscalls/linux/sys_poll.go +++ b/pkg/sentry/syscalls/linux/sys_poll.go @@ -23,8 +23,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/limits" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/syscalls/linux/sys_random.go b/pkg/sentry/syscalls/linux/sys_random.go index bc4c588bf..c0aa0fd60 100644 --- a/pkg/sentry/syscalls/linux/sys_random.go +++ b/pkg/sentry/syscalls/linux/sys_random.go @@ -19,11 +19,11 @@ import ( "math" "gvisor.dev/gvisor/pkg/rand" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/safemem" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) const ( diff --git a/pkg/sentry/syscalls/linux/sys_read.go b/pkg/sentry/syscalls/linux/sys_read.go index cd31e0649..f9f594190 100644 --- a/pkg/sentry/syscalls/linux/sys_read.go +++ b/pkg/sentry/syscalls/linux/sys_read.go @@ -23,8 +23,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/socket" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/syscalls/linux/sys_rlimit.go b/pkg/sentry/syscalls/linux/sys_rlimit.go index 51e3f836b..e08c333d6 100644 --- a/pkg/sentry/syscalls/linux/sys_rlimit.go +++ b/pkg/sentry/syscalls/linux/sys_rlimit.go @@ -19,8 +19,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/limits" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // rlimit describes an implementation of 'struct rlimit', which may vary from diff --git a/pkg/sentry/syscalls/linux/sys_seccomp.go b/pkg/sentry/syscalls/linux/sys_seccomp.go index 18510ead8..5b7a66f4d 100644 --- a/pkg/sentry/syscalls/linux/sys_seccomp.go +++ b/pkg/sentry/syscalls/linux/sys_seccomp.go @@ -19,8 +19,8 @@ import ( "gvisor.dev/gvisor/pkg/bpf" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // userSockFprog is equivalent to Linux's struct sock_fprog on amd64. diff --git a/pkg/sentry/syscalls/linux/sys_sem.go b/pkg/sentry/syscalls/linux/sys_sem.go index cde3b54e7..5f54f2456 100644 --- a/pkg/sentry/syscalls/linux/sys_sem.go +++ b/pkg/sentry/syscalls/linux/sys_sem.go @@ -22,8 +22,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) const opsMax = 500 // SEMOPM diff --git a/pkg/sentry/syscalls/linux/sys_signal.go b/pkg/sentry/syscalls/linux/sys_signal.go index fb6efd5d8..209be2990 100644 --- a/pkg/sentry/syscalls/linux/sys_signal.go +++ b/pkg/sentry/syscalls/linux/sys_signal.go @@ -23,8 +23,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/signalfd" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // "For a process to have permission to send a signal it must diff --git a/pkg/sentry/syscalls/linux/sys_socket.go b/pkg/sentry/syscalls/linux/sys_socket.go index cda517a81..2919228d0 100644 --- a/pkg/sentry/syscalls/linux/sys_socket.go +++ b/pkg/sentry/syscalls/linux/sys_socket.go @@ -26,9 +26,9 @@ import ( "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/control" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // minListenBacklog is the minimum reasonable backlog for listening sockets. diff --git a/pkg/sentry/syscalls/linux/sys_stat.go b/pkg/sentry/syscalls/linux/sys_stat.go index 69b17b799..c841abccb 100644 --- a/pkg/sentry/syscalls/linux/sys_stat.go +++ b/pkg/sentry/syscalls/linux/sys_stat.go @@ -19,8 +19,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // Stat implements linux syscall stat(2). diff --git a/pkg/sentry/syscalls/linux/sys_stat_amd64.go b/pkg/sentry/syscalls/linux/sys_stat_amd64.go index 58afb4a9a..75a567bd4 100644 --- a/pkg/sentry/syscalls/linux/sys_stat_amd64.go +++ b/pkg/sentry/syscalls/linux/sys_stat_amd64.go @@ -21,7 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // copyOutStat copies the attributes (sattr, uattr) to the struct stat at diff --git a/pkg/sentry/syscalls/linux/sys_stat_arm64.go b/pkg/sentry/syscalls/linux/sys_stat_arm64.go index 3e1251e0b..80c98d05c 100644 --- a/pkg/sentry/syscalls/linux/sys_stat_arm64.go +++ b/pkg/sentry/syscalls/linux/sys_stat_arm64.go @@ -21,7 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // copyOutStat copies the attributes (sattr, uattr) to the struct stat at diff --git a/pkg/sentry/syscalls/linux/sys_thread.go b/pkg/sentry/syscalls/linux/sys_thread.go index b47c3b5c4..0c9e2255d 100644 --- a/pkg/sentry/syscalls/linux/sys_thread.go +++ b/pkg/sentry/syscalls/linux/sys_thread.go @@ -24,8 +24,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/sched" "gvisor.dev/gvisor/pkg/sentry/loader" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) const ( diff --git a/pkg/sentry/syscalls/linux/sys_time.go b/pkg/sentry/syscalls/linux/sys_time.go index b887fa9d7..2d2aa0819 100644 --- a/pkg/sentry/syscalls/linux/sys_time.go +++ b/pkg/sentry/syscalls/linux/sys_time.go @@ -22,8 +22,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // The most significant 29 bits hold either a pid or a file descriptor. diff --git a/pkg/sentry/syscalls/linux/sys_timer.go b/pkg/sentry/syscalls/linux/sys_timer.go index d4134207b..432351917 100644 --- a/pkg/sentry/syscalls/linux/sys_timer.go +++ b/pkg/sentry/syscalls/linux/sys_timer.go @@ -20,8 +20,8 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) const nsecPerSec = int64(time.Second) diff --git a/pkg/sentry/syscalls/linux/sys_write.go b/pkg/sentry/syscalls/linux/sys_write.go index ad4b67806..aba892939 100644 --- a/pkg/sentry/syscalls/linux/sys_write.go +++ b/pkg/sentry/syscalls/linux/sys_write.go @@ -23,8 +23,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/socket" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/syscalls/linux/sys_xattr.go b/pkg/sentry/syscalls/linux/sys_xattr.go index 77deb8980..efb95555c 100644 --- a/pkg/sentry/syscalls/linux/sys_xattr.go +++ b/pkg/sentry/syscalls/linux/sys_xattr.go @@ -21,8 +21,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // GetXattr implements linux syscall getxattr(2). diff --git a/pkg/sentry/syscalls/linux/timespec.go b/pkg/sentry/syscalls/linux/timespec.go index 4ff8f9234..ddc3ee26e 100644 --- a/pkg/sentry/syscalls/linux/timespec.go +++ b/pkg/sentry/syscalls/linux/timespec.go @@ -19,8 +19,8 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // copyTimespecIn copies a Timespec from the untrusted app range to the kernel. diff --git a/pkg/sentry/unimpl/BUILD b/pkg/sentry/unimpl/BUILD index 370fa6ec5..5d4aa3a63 100644 --- a/pkg/sentry/unimpl/BUILD +++ b/pkg/sentry/unimpl/BUILD @@ -14,7 +14,7 @@ go_library( srcs = ["events.go"], visibility = ["//:sandbox"], deps = [ + "//pkg/context", "//pkg/log", - "//pkg/sentry/context", ], ) diff --git a/pkg/sentry/unimpl/events.go b/pkg/sentry/unimpl/events.go index 79b5de9e4..73ed9372f 100644 --- a/pkg/sentry/unimpl/events.go +++ b/pkg/sentry/unimpl/events.go @@ -17,8 +17,8 @@ package unimpl import ( + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/context" ) // contextID is the events package's type for context.Context.Value keys. diff --git a/pkg/sentry/uniqueid/BUILD b/pkg/sentry/uniqueid/BUILD index e9c18f170..7467e6398 100644 --- a/pkg/sentry/uniqueid/BUILD +++ b/pkg/sentry/uniqueid/BUILD @@ -7,7 +7,7 @@ go_library( srcs = ["context.go"], visibility = ["//pkg/sentry:internal"], deps = [ - "//pkg/sentry/context", + "//pkg/context", "//pkg/sentry/socket/unix/transport", ], ) diff --git a/pkg/sentry/uniqueid/context.go b/pkg/sentry/uniqueid/context.go index 4e466d66d..1fb884a90 100644 --- a/pkg/sentry/uniqueid/context.go +++ b/pkg/sentry/uniqueid/context.go @@ -17,7 +17,7 @@ package uniqueid import ( - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" ) diff --git a/pkg/sentry/usermem/BUILD b/pkg/sentry/usermem/BUILD deleted file mode 100644 index c8322e29e..000000000 --- a/pkg/sentry/usermem/BUILD +++ /dev/null @@ -1,55 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -package(licenses = ["notice"]) - -go_template_instance( - name = "addr_range", - out = "addr_range.go", - package = "usermem", - prefix = "Addr", - template = "//pkg/segment:generic_range", - types = { - "T": "Addr", - }, -) - -go_library( - name = "usermem", - srcs = [ - "access_type.go", - "addr.go", - "addr_range.go", - "addr_range_seq_unsafe.go", - "bytes_io.go", - "bytes_io_unsafe.go", - "usermem.go", - "usermem_arm64.go", - "usermem_unsafe.go", - "usermem_x86.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/atomicbitops", - "//pkg/binary", - "//pkg/log", - "//pkg/sentry/context", - "//pkg/sentry/safemem", - "//pkg/syserror", - ], -) - -go_test( - name = "usermem_test", - size = "small", - srcs = [ - "addr_range_seq_test.go", - "usermem_test.go", - ], - library = ":usermem", - deps = [ - "//pkg/sentry/context", - "//pkg/sentry/safemem", - "//pkg/syserror", - ], -) diff --git a/pkg/sentry/usermem/README.md b/pkg/sentry/usermem/README.md deleted file mode 100644 index f6d2137eb..000000000 --- a/pkg/sentry/usermem/README.md +++ /dev/null @@ -1,31 +0,0 @@ -This package defines primitives for sentry access to application memory. - -Major types: - -- The `IO` interface represents a virtual address space and provides I/O - methods on that address space. `IO` is the lowest-level primitive. The - primary implementation of the `IO` interface is `mm.MemoryManager`. - -- `IOSequence` represents a collection of individually-contiguous address - ranges in a `IO` that is operated on sequentially, analogous to Linux's - `struct iov_iter`. - -Major usage patterns: - -- Access to a task's virtual memory, subject to the application's memory - protections and while running on that task's goroutine, from a context that - is at or above the level of the `kernel` package (e.g. most syscall - implementations in `syscalls/linux`); use the `kernel.Task.Copy*` wrappers - defined in `kernel/task_usermem.go`. - -- Access to a task's virtual memory, from a context that is at or above the - level of the `kernel` package, but where any of the above constraints does - not hold (e.g. `PTRACE_POKEDATA`, which ignores application memory - protections); obtain the task's `mm.MemoryManager` by calling - `kernel.Task.MemoryManager`, and call its `IO` methods directly. - -- Access to a task's virtual memory, from a context that is below the level of - the `kernel` package (e.g. filesystem I/O); clients must pass I/O arguments - from higher layers, usually in the form of an `IOSequence`. The - `kernel.Task.SingleIOSequence` and `kernel.Task.IovecsIOSequence` functions - in `kernel/task_usermem.go` are convenience functions for doing so. diff --git a/pkg/sentry/usermem/access_type.go b/pkg/sentry/usermem/access_type.go deleted file mode 100644 index 9c1742a59..000000000 --- a/pkg/sentry/usermem/access_type.go +++ /dev/null @@ -1,128 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package usermem - -import ( - "syscall" -) - -// AccessType specifies memory access types. This is used for -// setting mapping permissions, as well as communicating faults. -// -// +stateify savable -type AccessType struct { - // Read is read access. - Read bool - - // Write is write access. - Write bool - - // Execute is executable access. - Execute bool -} - -// String returns a pretty representation of access. This looks like the -// familiar r-x, rw-, etc. and can be relied on as such. -func (a AccessType) String() string { - bits := [3]byte{'-', '-', '-'} - if a.Read { - bits[0] = 'r' - } - if a.Write { - bits[1] = 'w' - } - if a.Execute { - bits[2] = 'x' - } - return string(bits[:]) -} - -// Any returns true iff at least one of Read, Write or Execute is true. -func (a AccessType) Any() bool { - return a.Read || a.Write || a.Execute -} - -// Prot returns the system prot (syscall.PROT_READ, etc.) for this access. -func (a AccessType) Prot() int { - var prot int - if a.Read { - prot |= syscall.PROT_READ - } - if a.Write { - prot |= syscall.PROT_WRITE - } - if a.Execute { - prot |= syscall.PROT_EXEC - } - return prot -} - -// SupersetOf returns true iff the access types in a are a superset of the -// access types in other. -func (a AccessType) SupersetOf(other AccessType) bool { - if !a.Read && other.Read { - return false - } - if !a.Write && other.Write { - return false - } - if !a.Execute && other.Execute { - return false - } - return true -} - -// Intersect returns the access types set in both a and other. -func (a AccessType) Intersect(other AccessType) AccessType { - return AccessType{ - Read: a.Read && other.Read, - Write: a.Write && other.Write, - Execute: a.Execute && other.Execute, - } -} - -// Union returns the access types set in either a or other. -func (a AccessType) Union(other AccessType) AccessType { - return AccessType{ - Read: a.Read || other.Read, - Write: a.Write || other.Write, - Execute: a.Execute || other.Execute, - } -} - -// Effective returns the set of effective access types allowed by a, even if -// some types are not explicitly allowed. -func (a AccessType) Effective() AccessType { - // In Linux, Write and Execute access generally imply Read access. See - // mm/mmap.c:protection_map. - // - // The notable exception is get_user_pages, which only checks against - // the original vma flags. That said, most user memory accesses do not - // use GUP. - if a.Write || a.Execute { - a.Read = true - } - return a -} - -// Convenient access types. -var ( - NoAccess = AccessType{} - Read = AccessType{Read: true} - Write = AccessType{Write: true} - Execute = AccessType{Execute: true} - ReadWrite = AccessType{Read: true, Write: true} - AnyAccess = AccessType{Read: true, Write: true, Execute: true} -) diff --git a/pkg/sentry/usermem/addr.go b/pkg/sentry/usermem/addr.go deleted file mode 100644 index e79210804..000000000 --- a/pkg/sentry/usermem/addr.go +++ /dev/null @@ -1,108 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package usermem - -import ( - "fmt" -) - -// Addr represents a generic virtual address. -// -// +stateify savable -type Addr uintptr - -// AddLength adds the given length to start and returns the result. ok is true -// iff adding the length did not overflow the range of Addr. -// -// Note: This function is usually used to get the end of an address range -// defined by its start address and length. Since the resulting end is -// exclusive, end == 0 is technically valid, and corresponds to a range that -// extends to the end of the address space, but ok will be false. This isn't -// expected to ever come up in practice. -func (v Addr) AddLength(length uint64) (end Addr, ok bool) { - end = v + Addr(length) - // The second half of the following check is needed in case uintptr is - // smaller than 64 bits. - ok = end >= v && length <= uint64(^Addr(0)) - return -} - -// RoundDown returns the address rounded down to the nearest page boundary. -func (v Addr) RoundDown() Addr { - return v & ^Addr(PageSize-1) -} - -// RoundUp returns the address rounded up to the nearest page boundary. ok is -// true iff rounding up did not wrap around. -func (v Addr) RoundUp() (addr Addr, ok bool) { - addr = Addr(v + PageSize - 1).RoundDown() - ok = addr >= v - return -} - -// MustRoundUp is equivalent to RoundUp, but panics if rounding up wraps -// around. -func (v Addr) MustRoundUp() Addr { - addr, ok := v.RoundUp() - if !ok { - panic(fmt.Sprintf("usermem.Addr(%d).RoundUp() wraps", v)) - } - return addr -} - -// HugeRoundDown returns the address rounded down to the nearest huge page -// boundary. -func (v Addr) HugeRoundDown() Addr { - return v & ^Addr(HugePageSize-1) -} - -// HugeRoundUp returns the address rounded up to the nearest huge page boundary. -// ok is true iff rounding up did not wrap around. -func (v Addr) HugeRoundUp() (addr Addr, ok bool) { - addr = Addr(v + HugePageSize - 1).HugeRoundDown() - ok = addr >= v - return -} - -// PageOffset returns the offset of v into the current page. -func (v Addr) PageOffset() uint64 { - return uint64(v & Addr(PageSize-1)) -} - -// IsPageAligned returns true if v.PageOffset() == 0. -func (v Addr) IsPageAligned() bool { - return v.PageOffset() == 0 -} - -// AddrRange is a range of Addrs. -// -// type AddrRange - -// ToRange returns [v, v+length). -func (v Addr) ToRange(length uint64) (AddrRange, bool) { - end, ok := v.AddLength(length) - return AddrRange{v, end}, ok -} - -// IsPageAligned returns true if ar.Start.IsPageAligned() and -// ar.End.IsPageAligned(). -func (ar AddrRange) IsPageAligned() bool { - return ar.Start.IsPageAligned() && ar.End.IsPageAligned() -} - -// String implements fmt.Stringer.String. -func (ar AddrRange) String() string { - return fmt.Sprintf("[%#x, %#x)", ar.Start, ar.End) -} diff --git a/pkg/sentry/usermem/addr_range_seq_test.go b/pkg/sentry/usermem/addr_range_seq_test.go deleted file mode 100644 index 82f735026..000000000 --- a/pkg/sentry/usermem/addr_range_seq_test.go +++ /dev/null @@ -1,197 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package usermem - -import ( - "testing" -) - -var addrRangeSeqTests = []struct { - desc string - ranges []AddrRange -}{ - { - desc: "Empty sequence", - }, - { - desc: "Single empty AddrRange", - ranges: []AddrRange{ - {0x10, 0x10}, - }, - }, - { - desc: "Single non-empty AddrRange of length 1", - ranges: []AddrRange{ - {0x10, 0x11}, - }, - }, - { - desc: "Single non-empty AddrRange of length 2", - ranges: []AddrRange{ - {0x10, 0x12}, - }, - }, - { - desc: "Multiple non-empty AddrRanges", - ranges: []AddrRange{ - {0x10, 0x11}, - {0x20, 0x22}, - }, - }, - { - desc: "Multiple AddrRanges including empty AddrRanges", - ranges: []AddrRange{ - {0x10, 0x10}, - {0x20, 0x20}, - {0x30, 0x33}, - {0x40, 0x44}, - {0x50, 0x50}, - {0x60, 0x60}, - {0x70, 0x77}, - {0x80, 0x88}, - {0x90, 0x90}, - {0xa0, 0xa0}, - }, - }, -} - -func testAddrRangeSeqEqualityWithTailIteration(t *testing.T, ars AddrRangeSeq, wantRanges []AddrRange) { - var wantLen int64 - for _, ar := range wantRanges { - wantLen += int64(ar.Length()) - } - - var i int - for !ars.IsEmpty() { - if gotLen := ars.NumBytes(); gotLen != wantLen { - t.Errorf("Iteration %d: %v.NumBytes(): got %d, wanted %d", i, ars, gotLen, wantLen) - } - if gotN, wantN := ars.NumRanges(), len(wantRanges)-i; gotN != wantN { - t.Errorf("Iteration %d: %v.NumRanges(): got %d, wanted %d", i, ars, gotN, wantN) - } - got := ars.Head() - if i >= len(wantRanges) { - t.Errorf("Iteration %d: %v.Head(): got %s, wanted ", i, ars, got) - } else if want := wantRanges[i]; got != want { - t.Errorf("Iteration %d: %v.Head(): got %s, wanted %s", i, ars, got, want) - } - ars = ars.Tail() - wantLen -= int64(got.Length()) - i++ - } - if gotLen := ars.NumBytes(); gotLen != 0 || wantLen != 0 { - t.Errorf("Iteration %d: %v.NumBytes(): got %d, wanted %d (which should be 0)", i, ars, gotLen, wantLen) - } - if gotN := ars.NumRanges(); gotN != 0 { - t.Errorf("Iteration %d: %v.NumRanges(): got %d, wanted 0", i, ars, gotN) - } -} - -func TestAddrRangeSeqTailIteration(t *testing.T) { - for _, test := range addrRangeSeqTests { - t.Run(test.desc, func(t *testing.T) { - testAddrRangeSeqEqualityWithTailIteration(t, AddrRangeSeqFromSlice(test.ranges), test.ranges) - }) - } -} - -func TestAddrRangeSeqDropFirstEmpty(t *testing.T) { - var ars AddrRangeSeq - if got, want := ars.DropFirst(1), ars; got != want { - t.Errorf("%v.DropFirst(1): got %v, wanted %v", ars, got, want) - } -} - -func TestAddrRangeSeqDropSingleByteIteration(t *testing.T) { - // Tests AddrRangeSeq iteration using Head/DropFirst, simulating - // I/O-per-AddrRange. - for _, test := range addrRangeSeqTests { - t.Run(test.desc, func(t *testing.T) { - // Figure out what AddrRanges we expect to see. - var wantLen int64 - var wantRanges []AddrRange - for _, ar := range test.ranges { - wantLen += int64(ar.Length()) - wantRanges = append(wantRanges, ar) - if ar.Length() == 0 { - // We "do" 0 bytes of I/O and then call DropFirst(0), - // advancing to the next AddrRange. - continue - } - // Otherwise we "do" 1 byte of I/O and then call DropFirst(1), - // advancing the AddrRange by 1 byte, or to the next AddrRange - // if this one is exhausted. - for ar.Start++; ar.Length() != 0; ar.Start++ { - wantRanges = append(wantRanges, ar) - } - } - t.Logf("Expected AddrRanges: %s (%d bytes)", wantRanges, wantLen) - - ars := AddrRangeSeqFromSlice(test.ranges) - var i int - for !ars.IsEmpty() { - if gotLen := ars.NumBytes(); gotLen != wantLen { - t.Errorf("Iteration %d: %v.NumBytes(): got %d, wanted %d", i, ars, gotLen, wantLen) - } - got := ars.Head() - if i >= len(wantRanges) { - t.Errorf("Iteration %d: %v.Head(): got %s, wanted ", i, ars, got) - } else if want := wantRanges[i]; got != want { - t.Errorf("Iteration %d: %v.Head(): got %s, wanted %s", i, ars, got, want) - } - if got.Length() == 0 { - ars = ars.DropFirst(0) - } else { - ars = ars.DropFirst(1) - wantLen-- - } - i++ - } - if gotLen := ars.NumBytes(); gotLen != 0 || wantLen != 0 { - t.Errorf("Iteration %d: %v.NumBytes(): got %d, wanted %d (which should be 0)", i, ars, gotLen, wantLen) - } - }) - } -} - -func TestAddrRangeSeqTakeFirstEmpty(t *testing.T) { - var ars AddrRangeSeq - if got, want := ars.TakeFirst(1), ars; got != want { - t.Errorf("%v.TakeFirst(1): got %v, wanted %v", ars, got, want) - } -} - -func TestAddrRangeSeqTakeFirst(t *testing.T) { - ranges := []AddrRange{ - {0x10, 0x11}, - {0x20, 0x22}, - {0x30, 0x30}, - {0x40, 0x44}, - {0x50, 0x55}, - {0x60, 0x60}, - {0x70, 0x77}, - } - ars := AddrRangeSeqFromSlice(ranges).TakeFirst(5) - want := []AddrRange{ - {0x10, 0x11}, // +1 byte (total 1 byte), not truncated - {0x20, 0x22}, // +2 bytes (total 3 bytes), not truncated - {0x30, 0x30}, // +0 bytes (total 3 bytes), no change - {0x40, 0x42}, // +2 bytes (total 5 bytes), partially truncated - {0x50, 0x50}, // +0 bytes (total 5 bytes), fully truncated - {0x60, 0x60}, // +0 bytes (total 5 bytes), "fully truncated" (no change) - {0x70, 0x70}, // +0 bytes (total 5 bytes), fully truncated - } - testAddrRangeSeqEqualityWithTailIteration(t, ars, want) -} diff --git a/pkg/sentry/usermem/addr_range_seq_unsafe.go b/pkg/sentry/usermem/addr_range_seq_unsafe.go deleted file mode 100644 index c09337c15..000000000 --- a/pkg/sentry/usermem/addr_range_seq_unsafe.go +++ /dev/null @@ -1,277 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package usermem - -import ( - "bytes" - "fmt" - "reflect" - "unsafe" -) - -// An AddrRangeSeq represents a sequence of AddrRanges. -// -// AddrRangeSeqs are immutable and may be copied by value. The zero value of -// AddrRangeSeq represents an empty sequence. -// -// An AddrRangeSeq may contain AddrRanges with a length of 0. This is necessary -// since zero-length AddrRanges are significant to MM bounds checks. -type AddrRangeSeq struct { - // If length is 0, then the AddrRangeSeq represents no AddrRanges. - // Invariants: data == 0; offset == 0; limit == 0. - // - // If length is 1, then the AddrRangeSeq represents the single - // AddrRange{offset, offset+limit}. Invariants: data == 0. - // - // Otherwise, length >= 2, and the AddrRangeSeq represents the `length` - // AddrRanges in the array of AddrRanges starting at address `data`, - // starting at `offset` bytes into the first AddrRange and limited to the - // following `limit` bytes. (AddrRanges after `limit` are still iterated, - // but are truncated to a length of 0.) Invariants: data != 0; offset <= - // data[0].Length(); limit > 0; offset+limit <= the combined length of all - // AddrRanges in the array. - data unsafe.Pointer - length int - offset Addr - limit Addr -} - -// AddrRangeSeqOf returns an AddrRangeSeq representing the single AddrRange ar. -func AddrRangeSeqOf(ar AddrRange) AddrRangeSeq { - return AddrRangeSeq{ - length: 1, - offset: ar.Start, - limit: ar.Length(), - } -} - -// AddrRangeSeqFromSlice returns an AddrRangeSeq representing all AddrRanges in -// slice. -// -// Whether the returned AddrRangeSeq shares memory with slice is unspecified; -// clients should avoid mutating slices passed to AddrRangeSeqFromSlice. -// -// Preconditions: The combined length of all AddrRanges in slice <= -// math.MaxInt64. -func AddrRangeSeqFromSlice(slice []AddrRange) AddrRangeSeq { - var limit int64 - for _, ar := range slice { - len64 := int64(ar.Length()) - if len64 < 0 { - panic(fmt.Sprintf("Length of AddrRange %v overflows int64", ar)) - } - sum := limit + len64 - if sum < limit { - panic(fmt.Sprintf("Total length of AddrRanges %v overflows int64", slice)) - } - limit = sum - } - return addrRangeSeqFromSliceLimited(slice, limit) -} - -// Preconditions: The combined length of all AddrRanges in slice <= limit. -// limit >= 0. If len(slice) != 0, then limit > 0. -func addrRangeSeqFromSliceLimited(slice []AddrRange, limit int64) AddrRangeSeq { - switch len(slice) { - case 0: - return AddrRangeSeq{} - case 1: - return AddrRangeSeq{ - length: 1, - offset: slice[0].Start, - limit: Addr(limit), - } - default: - return AddrRangeSeq{ - data: unsafe.Pointer(&slice[0]), - length: len(slice), - limit: Addr(limit), - } - } -} - -// IsEmpty returns true if ars.NumRanges() == 0. -// -// Note that since AddrRangeSeq may contain AddrRanges with a length of zero, -// an AddrRange representing 0 bytes (AddrRangeSeq.NumBytes() == 0) is not -// necessarily empty. -func (ars AddrRangeSeq) IsEmpty() bool { - return ars.length == 0 -} - -// NumRanges returns the number of AddrRanges in ars. -func (ars AddrRangeSeq) NumRanges() int { - return ars.length -} - -// NumBytes returns the number of bytes represented by ars. -func (ars AddrRangeSeq) NumBytes() int64 { - return int64(ars.limit) -} - -// Head returns the first AddrRange in ars. -// -// Preconditions: !ars.IsEmpty(). -func (ars AddrRangeSeq) Head() AddrRange { - if ars.length == 0 { - panic("empty AddrRangeSeq") - } - if ars.length == 1 { - return AddrRange{ars.offset, ars.offset + ars.limit} - } - ar := *(*AddrRange)(ars.data) - ar.Start += ars.offset - if ar.Length() > ars.limit { - ar.End = ar.Start + ars.limit - } - return ar -} - -// Tail returns an AddrRangeSeq consisting of all AddrRanges in ars after the -// first. -// -// Preconditions: !ars.IsEmpty(). -func (ars AddrRangeSeq) Tail() AddrRangeSeq { - if ars.length == 0 { - panic("empty AddrRangeSeq") - } - if ars.length == 1 { - return AddrRangeSeq{} - } - return ars.externalTail() -} - -// Preconditions: ars.length >= 2. -func (ars AddrRangeSeq) externalTail() AddrRangeSeq { - headLen := (*AddrRange)(ars.data).Length() - ars.offset - var tailLimit int64 - if ars.limit > headLen { - tailLimit = int64(ars.limit - headLen) - } - var extSlice []AddrRange - extSliceHdr := (*reflect.SliceHeader)(unsafe.Pointer(&extSlice)) - extSliceHdr.Data = uintptr(ars.data) - extSliceHdr.Len = ars.length - extSliceHdr.Cap = ars.length - return addrRangeSeqFromSliceLimited(extSlice[1:], tailLimit) -} - -// DropFirst returns an AddrRangeSeq equivalent to ars, but with the first n -// bytes omitted. If n > ars.NumBytes(), DropFirst returns an empty -// AddrRangeSeq. -// -// If !ars.IsEmpty() and ars.Head().Length() == 0, DropFirst will always omit -// at least ars.Head(), even if n == 0. This guarantees that the basic pattern -// of: -// -// for !ars.IsEmpty() { -// n, err = doIOWith(ars.Head()) -// if err != nil { -// return err -// } -// ars = ars.DropFirst(n) -// } -// -// works even in the presence of zero-length AddrRanges. -// -// Preconditions: n >= 0. -func (ars AddrRangeSeq) DropFirst(n int) AddrRangeSeq { - if n < 0 { - panic(fmt.Sprintf("invalid n: %d", n)) - } - return ars.DropFirst64(int64(n)) -} - -// DropFirst64 is equivalent to DropFirst but takes an int64. -func (ars AddrRangeSeq) DropFirst64(n int64) AddrRangeSeq { - if n < 0 { - panic(fmt.Sprintf("invalid n: %d", n)) - } - if Addr(n) > ars.limit { - return AddrRangeSeq{} - } - // Handle initial empty AddrRange. - switch ars.length { - case 0: - return AddrRangeSeq{} - case 1: - if ars.limit == 0 { - return AddrRangeSeq{} - } - default: - if rawHeadLen := (*AddrRange)(ars.data).Length(); ars.offset == rawHeadLen { - ars = ars.externalTail() - } - } - for n != 0 { - // Calling ars.Head() here is surprisingly expensive, so inline getting - // the head's length. - var headLen Addr - if ars.length == 1 { - headLen = ars.limit - } else { - headLen = (*AddrRange)(ars.data).Length() - ars.offset - } - if Addr(n) < headLen { - // Dropping ends partway through the head AddrRange. - ars.offset += Addr(n) - ars.limit -= Addr(n) - return ars - } - n -= int64(headLen) - ars = ars.Tail() - } - return ars -} - -// TakeFirst returns an AddrRangeSeq equivalent to ars, but iterating at most n -// bytes. TakeFirst never removes AddrRanges from ars; AddrRanges beyond the -// first n bytes are reduced to a length of zero, but will still be iterated. -// -// Preconditions: n >= 0. -func (ars AddrRangeSeq) TakeFirst(n int) AddrRangeSeq { - if n < 0 { - panic(fmt.Sprintf("invalid n: %d", n)) - } - return ars.TakeFirst64(int64(n)) -} - -// TakeFirst64 is equivalent to TakeFirst but takes an int64. -func (ars AddrRangeSeq) TakeFirst64(n int64) AddrRangeSeq { - if n < 0 { - panic(fmt.Sprintf("invalid n: %d", n)) - } - if ars.limit > Addr(n) { - ars.limit = Addr(n) - } - return ars -} - -// String implements fmt.Stringer.String. -func (ars AddrRangeSeq) String() string { - // This is deliberately chosen to be the same as fmt's automatic stringer - // for []AddrRange. - var buf bytes.Buffer - buf.WriteByte('[') - var sep string - for !ars.IsEmpty() { - buf.WriteString(sep) - sep = " " - buf.WriteString(ars.Head().String()) - ars = ars.Tail() - } - buf.WriteByte(']') - return buf.String() -} diff --git a/pkg/sentry/usermem/bytes_io.go b/pkg/sentry/usermem/bytes_io.go deleted file mode 100644 index 7898851b3..000000000 --- a/pkg/sentry/usermem/bytes_io.go +++ /dev/null @@ -1,141 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package usermem - -import ( - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/safemem" - "gvisor.dev/gvisor/pkg/syserror" -) - -const maxInt = int(^uint(0) >> 1) - -// BytesIO implements IO using a byte slice. Addresses are interpreted as -// offsets into the slice. Reads and writes beyond the end of the slice return -// EFAULT. -type BytesIO struct { - Bytes []byte -} - -// CopyOut implements IO.CopyOut. -func (b *BytesIO) CopyOut(ctx context.Context, addr Addr, src []byte, opts IOOpts) (int, error) { - rngN, rngErr := b.rangeCheck(addr, len(src)) - if rngN == 0 { - return 0, rngErr - } - return copy(b.Bytes[int(addr):], src[:rngN]), rngErr -} - -// CopyIn implements IO.CopyIn. -func (b *BytesIO) CopyIn(ctx context.Context, addr Addr, dst []byte, opts IOOpts) (int, error) { - rngN, rngErr := b.rangeCheck(addr, len(dst)) - if rngN == 0 { - return 0, rngErr - } - return copy(dst[:rngN], b.Bytes[int(addr):]), rngErr -} - -// ZeroOut implements IO.ZeroOut. -func (b *BytesIO) ZeroOut(ctx context.Context, addr Addr, toZero int64, opts IOOpts) (int64, error) { - if toZero > int64(maxInt) { - return 0, syserror.EINVAL - } - rngN, rngErr := b.rangeCheck(addr, int(toZero)) - if rngN == 0 { - return 0, rngErr - } - zeroSlice := b.Bytes[int(addr) : int(addr)+rngN] - for i := range zeroSlice { - zeroSlice[i] = 0 - } - return int64(rngN), rngErr -} - -// CopyOutFrom implements IO.CopyOutFrom. -func (b *BytesIO) CopyOutFrom(ctx context.Context, ars AddrRangeSeq, src safemem.Reader, opts IOOpts) (int64, error) { - dsts, rngErr := b.blocksFromAddrRanges(ars) - n, err := src.ReadToBlocks(dsts) - if err != nil { - return int64(n), err - } - return int64(n), rngErr -} - -// CopyInTo implements IO.CopyInTo. -func (b *BytesIO) CopyInTo(ctx context.Context, ars AddrRangeSeq, dst safemem.Writer, opts IOOpts) (int64, error) { - srcs, rngErr := b.blocksFromAddrRanges(ars) - n, err := dst.WriteFromBlocks(srcs) - if err != nil { - return int64(n), err - } - return int64(n), rngErr -} - -func (b *BytesIO) rangeCheck(addr Addr, length int) (int, error) { - if length == 0 { - return 0, nil - } - if length < 0 { - return 0, syserror.EINVAL - } - max := Addr(len(b.Bytes)) - if addr >= max { - return 0, syserror.EFAULT - } - end, ok := addr.AddLength(uint64(length)) - if !ok || end > max { - return int(max - addr), syserror.EFAULT - } - return length, nil -} - -func (b *BytesIO) blocksFromAddrRanges(ars AddrRangeSeq) (safemem.BlockSeq, error) { - switch ars.NumRanges() { - case 0: - return safemem.BlockSeq{}, nil - case 1: - block, err := b.blockFromAddrRange(ars.Head()) - return safemem.BlockSeqOf(block), err - default: - blocks := make([]safemem.Block, 0, ars.NumRanges()) - for !ars.IsEmpty() { - block, err := b.blockFromAddrRange(ars.Head()) - if block.Len() != 0 { - blocks = append(blocks, block) - } - if err != nil { - return safemem.BlockSeqFromSlice(blocks), err - } - ars = ars.Tail() - } - return safemem.BlockSeqFromSlice(blocks), nil - } -} - -func (b *BytesIO) blockFromAddrRange(ar AddrRange) (safemem.Block, error) { - n, err := b.rangeCheck(ar.Start, int(ar.Length())) - if n == 0 { - return safemem.Block{}, err - } - return safemem.BlockFromSafeSlice(b.Bytes[int(ar.Start) : int(ar.Start)+n]), err -} - -// BytesIOSequence returns an IOSequence representing the given byte slice. -func BytesIOSequence(buf []byte) IOSequence { - return IOSequence{ - IO: &BytesIO{buf}, - Addrs: AddrRangeSeqOf(AddrRange{0, Addr(len(buf))}), - } -} diff --git a/pkg/sentry/usermem/bytes_io_unsafe.go b/pkg/sentry/usermem/bytes_io_unsafe.go deleted file mode 100644 index fca5952f4..000000000 --- a/pkg/sentry/usermem/bytes_io_unsafe.go +++ /dev/null @@ -1,47 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package usermem - -import ( - "sync/atomic" - "unsafe" - - "gvisor.dev/gvisor/pkg/atomicbitops" - "gvisor.dev/gvisor/pkg/sentry/context" -) - -// SwapUint32 implements IO.SwapUint32. -func (b *BytesIO) SwapUint32(ctx context.Context, addr Addr, new uint32, opts IOOpts) (uint32, error) { - if _, rngErr := b.rangeCheck(addr, 4); rngErr != nil { - return 0, rngErr - } - return atomic.SwapUint32((*uint32)(unsafe.Pointer(&b.Bytes[int(addr)])), new), nil -} - -// CompareAndSwapUint32 implements IO.CompareAndSwapUint32. -func (b *BytesIO) CompareAndSwapUint32(ctx context.Context, addr Addr, old, new uint32, opts IOOpts) (uint32, error) { - if _, rngErr := b.rangeCheck(addr, 4); rngErr != nil { - return 0, rngErr - } - return atomicbitops.CompareAndSwapUint32((*uint32)(unsafe.Pointer(&b.Bytes[int(addr)])), old, new), nil -} - -// LoadUint32 implements IO.LoadUint32. -func (b *BytesIO) LoadUint32(ctx context.Context, addr Addr, opts IOOpts) (uint32, error) { - if _, err := b.rangeCheck(addr, 4); err != nil { - return 0, err - } - return atomic.LoadUint32((*uint32)(unsafe.Pointer(&b.Bytes[int(addr)]))), nil -} diff --git a/pkg/sentry/usermem/usermem.go b/pkg/sentry/usermem/usermem.go deleted file mode 100644 index 7b1f312b1..000000000 --- a/pkg/sentry/usermem/usermem.go +++ /dev/null @@ -1,597 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package usermem governs access to user memory. -package usermem - -import ( - "bytes" - "errors" - "io" - "strconv" - - "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/safemem" - "gvisor.dev/gvisor/pkg/syserror" -) - -// IO provides access to the contents of a virtual memory space. -// -// FIXME(b/38173783): Implementations of IO cannot expect ctx to contain any -// meaningful data. -type IO interface { - // CopyOut copies len(src) bytes from src to the memory mapped at addr. It - // returns the number of bytes copied. If the number of bytes copied is < - // len(src), it returns a non-nil error explaining why. - // - // Preconditions: The caller must not hold mm.MemoryManager.mappingMu or - // any following locks in the lock order. - // - // Postconditions: CopyOut does not retain src. - CopyOut(ctx context.Context, addr Addr, src []byte, opts IOOpts) (int, error) - - // CopyIn copies len(dst) bytes from the memory mapped at addr to dst. - // It returns the number of bytes copied. If the number of bytes copied is - // < len(dst), it returns a non-nil error explaining why. - // - // Preconditions: The caller must not hold mm.MemoryManager.mappingMu or - // any following locks in the lock order. - // - // Postconditions: CopyIn does not retain dst. - CopyIn(ctx context.Context, addr Addr, dst []byte, opts IOOpts) (int, error) - - // ZeroOut sets toZero bytes to 0, starting at addr. It returns the number - // of bytes zeroed. If the number of bytes zeroed is < toZero, it returns a - // non-nil error explaining why. - // - // Preconditions: The caller must not hold mm.MemoryManager.mappingMu or - // any following locks in the lock order. toZero >= 0. - ZeroOut(ctx context.Context, addr Addr, toZero int64, opts IOOpts) (int64, error) - - // CopyOutFrom copies ars.NumBytes() bytes from src to the memory mapped at - // ars. It returns the number of bytes copied, which may be less than the - // number of bytes read from src if copying fails. CopyOutFrom may return a - // partial copy without an error iff src.ReadToBlocks returns a partial - // read without an error. - // - // CopyOutFrom calls src.ReadToBlocks at most once. - // - // Preconditions: The caller must not hold mm.MemoryManager.mappingMu or - // any following locks in the lock order. src.ReadToBlocks must not block - // on mm.MemoryManager.activeMu or any preceding locks in the lock order. - CopyOutFrom(ctx context.Context, ars AddrRangeSeq, src safemem.Reader, opts IOOpts) (int64, error) - - // CopyInTo copies ars.NumBytes() bytes from the memory mapped at ars to - // dst. It returns the number of bytes copied. CopyInTo may return a - // partial copy without an error iff dst.WriteFromBlocks returns a partial - // write without an error. - // - // CopyInTo calls dst.WriteFromBlocks at most once. - // - // Preconditions: The caller must not hold mm.MemoryManager.mappingMu or - // any following locks in the lock order. dst.WriteFromBlocks must not - // block on mm.MemoryManager.activeMu or any preceding locks in the lock - // order. - CopyInTo(ctx context.Context, ars AddrRangeSeq, dst safemem.Writer, opts IOOpts) (int64, error) - - // TODO(jamieliu): The requirement that CopyOutFrom/CopyInTo call src/dst - // at most once, which is unnecessary in most cases, forces implementations - // to gather safemem.Blocks into a single slice to pass to src/dst. Add - // CopyOutFromIter/CopyInToIter, which relaxes this restriction, to avoid - // this allocation. - - // SwapUint32 atomically sets the uint32 value at addr to new and - // returns the previous value. - // - // Preconditions: The caller must not hold mm.MemoryManager.mappingMu or - // any following locks in the lock order. addr must be aligned to a 4-byte - // boundary. - SwapUint32(ctx context.Context, addr Addr, new uint32, opts IOOpts) (uint32, error) - - // CompareAndSwapUint32 atomically compares the uint32 value at addr to - // old; if they are equal, the value in memory is replaced by new. In - // either case, the previous value stored in memory is returned. - // - // Preconditions: The caller must not hold mm.MemoryManager.mappingMu or - // any following locks in the lock order. addr must be aligned to a 4-byte - // boundary. - CompareAndSwapUint32(ctx context.Context, addr Addr, old, new uint32, opts IOOpts) (uint32, error) - - // LoadUint32 atomically loads the uint32 value at addr and returns it. - // - // Preconditions: The caller must not hold mm.MemoryManager.mappingMu or - // any following locks in the lock order. addr must be aligned to a 4-byte - // boundary. - LoadUint32(ctx context.Context, addr Addr, opts IOOpts) (uint32, error) -} - -// IOOpts contains options applicable to all IO methods. -type IOOpts struct { - // If IgnorePermissions is true, application-defined memory protections set - // by mmap(2) or mprotect(2) will be ignored. (Memory protections required - // by the target of the mapping are never ignored.) - IgnorePermissions bool - - // If AddressSpaceActive is true, the IO implementation may assume that it - // has an active AddressSpace and can therefore use AddressSpace copying - // without performing activation. See mm/io.go for details. - AddressSpaceActive bool -} - -// IOReadWriter is an io.ReadWriter that reads from / writes to addresses -// starting at addr in IO. The preconditions that apply to IO.CopyIn and -// IO.CopyOut also apply to IOReadWriter.Read and IOReadWriter.Write -// respectively. -type IOReadWriter struct { - Ctx context.Context - IO IO - Addr Addr - Opts IOOpts -} - -// Read implements io.Reader.Read. -// -// Note that an address space does not have an "end of file", so Read can only -// return io.EOF if IO.CopyIn returns io.EOF. Attempts to read unmapped or -// unreadable memory, or beyond the end of the address space, should return -// EFAULT. -func (rw *IOReadWriter) Read(dst []byte) (int, error) { - n, err := rw.IO.CopyIn(rw.Ctx, rw.Addr, dst, rw.Opts) - end, ok := rw.Addr.AddLength(uint64(n)) - if ok { - rw.Addr = end - } else { - // Disallow wraparound. - rw.Addr = ^Addr(0) - if err != nil { - err = syserror.EFAULT - } - } - return n, err -} - -// Writer implements io.Writer.Write. -func (rw *IOReadWriter) Write(src []byte) (int, error) { - n, err := rw.IO.CopyOut(rw.Ctx, rw.Addr, src, rw.Opts) - end, ok := rw.Addr.AddLength(uint64(n)) - if ok { - rw.Addr = end - } else { - // Disallow wraparound. - rw.Addr = ^Addr(0) - if err != nil { - err = syserror.EFAULT - } - } - return n, err -} - -// CopyObjectOut copies a fixed-size value or slice of fixed-size values from -// src to the memory mapped at addr in uio. It returns the number of bytes -// copied. -// -// CopyObjectOut must use reflection to encode src; performance-sensitive -// clients should do encoding manually and use uio.CopyOut directly. -// -// Preconditions: As for IO.CopyOut. -func CopyObjectOut(ctx context.Context, uio IO, addr Addr, src interface{}, opts IOOpts) (int, error) { - w := &IOReadWriter{ - Ctx: ctx, - IO: uio, - Addr: addr, - Opts: opts, - } - // Allocate a byte slice the size of the object being marshaled. This - // adds an extra reflection call, but avoids needing to grow the slice - // during encoding, which can result in many heap-allocated slices. - b := make([]byte, 0, binary.Size(src)) - return w.Write(binary.Marshal(b, ByteOrder, src)) -} - -// CopyObjectIn copies a fixed-size value or slice of fixed-size values from -// the memory mapped at addr in uio to dst. It returns the number of bytes -// copied. -// -// CopyObjectIn must use reflection to decode dst; performance-sensitive -// clients should use uio.CopyIn directly and do decoding manually. -// -// Preconditions: As for IO.CopyIn. -func CopyObjectIn(ctx context.Context, uio IO, addr Addr, dst interface{}, opts IOOpts) (int, error) { - r := &IOReadWriter{ - Ctx: ctx, - IO: uio, - Addr: addr, - Opts: opts, - } - buf := make([]byte, binary.Size(dst)) - if _, err := io.ReadFull(r, buf); err != nil { - return 0, err - } - binary.Unmarshal(buf, ByteOrder, dst) - return int(r.Addr - addr), nil -} - -// CopyStringIn tuning parameters, defined outside that function for tests. -const ( - copyStringIncrement = 64 - copyStringMaxInitBufLen = 256 -) - -// CopyStringIn copies a NUL-terminated string of unknown length from the -// memory mapped at addr in uio and returns it as a string (not including the -// trailing NUL). If the length of the string, including the terminating NUL, -// would exceed maxlen, CopyStringIn returns the string truncated to maxlen and -// ENAMETOOLONG. -// -// Preconditions: As for IO.CopyFromUser. maxlen >= 0. -func CopyStringIn(ctx context.Context, uio IO, addr Addr, maxlen int, opts IOOpts) (string, error) { - initLen := maxlen - if initLen > copyStringMaxInitBufLen { - initLen = copyStringMaxInitBufLen - } - buf := make([]byte, initLen) - var done int - for done < maxlen { - // Read up to copyStringIncrement bytes at a time. - readlen := copyStringIncrement - if readlen > maxlen-done { - readlen = maxlen - done - } - end, ok := addr.AddLength(uint64(readlen)) - if !ok { - return stringFromImmutableBytes(buf[:done]), syserror.EFAULT - } - // Shorten the read to avoid crossing page boundaries, since faulting - // in a page unnecessarily is expensive. This also ensures that partial - // copies up to the end of application-mappable memory succeed. - if addr.RoundDown() != end.RoundDown() { - end = end.RoundDown() - readlen = int(end - addr) - } - // Ensure that our buffer is large enough to accommodate the read. - if done+readlen > len(buf) { - newBufLen := len(buf) * 2 - if newBufLen > maxlen { - newBufLen = maxlen - } - buf = append(buf, make([]byte, newBufLen-len(buf))...) - } - n, err := uio.CopyIn(ctx, addr, buf[done:done+readlen], opts) - // Look for the terminating zero byte, which may have occurred before - // hitting err. - if i := bytes.IndexByte(buf[done:done+n], byte(0)); i >= 0 { - return stringFromImmutableBytes(buf[:done+i]), nil - } - - done += n - if err != nil { - return stringFromImmutableBytes(buf[:done]), err - } - addr = end - } - return stringFromImmutableBytes(buf), syserror.ENAMETOOLONG -} - -// CopyOutVec copies bytes from src to the memory mapped at ars in uio. The -// maximum number of bytes copied is ars.NumBytes() or len(src), whichever is -// less. CopyOutVec returns the number of bytes copied; if this is less than -// the maximum, it returns a non-nil error explaining why. -// -// Preconditions: As for IO.CopyOut. -func CopyOutVec(ctx context.Context, uio IO, ars AddrRangeSeq, src []byte, opts IOOpts) (int, error) { - var done int - for !ars.IsEmpty() && done < len(src) { - ar := ars.Head() - cplen := len(src) - done - if Addr(cplen) >= ar.Length() { - cplen = int(ar.Length()) - } - n, err := uio.CopyOut(ctx, ar.Start, src[done:done+cplen], opts) - done += n - if err != nil { - return done, err - } - ars = ars.DropFirst(n) - } - return done, nil -} - -// CopyInVec copies bytes from the memory mapped at ars in uio to dst. The -// maximum number of bytes copied is ars.NumBytes() or len(dst), whichever is -// less. CopyInVec returns the number of bytes copied; if this is less than the -// maximum, it returns a non-nil error explaining why. -// -// Preconditions: As for IO.CopyIn. -func CopyInVec(ctx context.Context, uio IO, ars AddrRangeSeq, dst []byte, opts IOOpts) (int, error) { - var done int - for !ars.IsEmpty() && done < len(dst) { - ar := ars.Head() - cplen := len(dst) - done - if Addr(cplen) >= ar.Length() { - cplen = int(ar.Length()) - } - n, err := uio.CopyIn(ctx, ar.Start, dst[done:done+cplen], opts) - done += n - if err != nil { - return done, err - } - ars = ars.DropFirst(n) - } - return done, nil -} - -// ZeroOutVec writes zeroes to the memory mapped at ars in uio. The maximum -// number of bytes written is ars.NumBytes() or toZero, whichever is less. -// ZeroOutVec returns the number of bytes written; if this is less than the -// maximum, it returns a non-nil error explaining why. -// -// Preconditions: As for IO.ZeroOut. -func ZeroOutVec(ctx context.Context, uio IO, ars AddrRangeSeq, toZero int64, opts IOOpts) (int64, error) { - var done int64 - for !ars.IsEmpty() && done < toZero { - ar := ars.Head() - cplen := toZero - done - if Addr(cplen) >= ar.Length() { - cplen = int64(ar.Length()) - } - n, err := uio.ZeroOut(ctx, ar.Start, cplen, opts) - done += n - if err != nil { - return done, err - } - ars = ars.DropFirst64(n) - } - return done, nil -} - -func isASCIIWhitespace(b byte) bool { - // Compare Linux include/linux/ctype.h, lib/ctype.c. - // 9 => horizontal tab '\t' - // 10 => line feed '\n' - // 11 => vertical tab '\v' - // 12 => form feed '\c' - // 13 => carriage return '\r' - return b == ' ' || (b >= 9 && b <= 13) -} - -// CopyInt32StringsInVec copies up to len(dsts) whitespace-separated decimal -// strings from the memory mapped at ars in uio and converts them to int32 -// values in dsts. It returns the number of bytes read. -// -// CopyInt32StringsInVec shares the following properties with Linux's -// kernel/sysctl.c:proc_dointvec(write=1): -// -// - If any read value overflows the range of int32, or any invalid characters -// are encountered during the read, CopyInt32StringsInVec returns EINVAL. -// -// - If, upon reaching the end of ars, fewer than len(dsts) values have been -// read, CopyInt32StringsInVec returns no error if at least 1 value was read -// and EINVAL otherwise. -// -// - Trailing whitespace after the last successfully read value is counted in -// the number of bytes read. -// -// Unlike proc_dointvec(): -// -// - CopyInt32StringsInVec does not implicitly limit ars.NumBytes() to -// PageSize-1; callers that require this must do so explicitly. -// -// - CopyInt32StringsInVec returns EINVAL if ars.NumBytes() == 0. -// -// Preconditions: As for CopyInVec. -func CopyInt32StringsInVec(ctx context.Context, uio IO, ars AddrRangeSeq, dsts []int32, opts IOOpts) (int64, error) { - if len(dsts) == 0 { - return 0, nil - } - - buf := make([]byte, ars.NumBytes()) - n, cperr := CopyInVec(ctx, uio, ars, buf, opts) - buf = buf[:n] - - var i, j int - for ; j < len(dsts); j++ { - // Skip leading whitespace. - for i < len(buf) && isASCIIWhitespace(buf[i]) { - i++ - } - if i == len(buf) { - break - } - - // Find the end of the value to be parsed (next whitespace or end of string). - nextI := i + 1 - for nextI < len(buf) && !isASCIIWhitespace(buf[nextI]) { - nextI++ - } - - // Parse a single value. - val, err := strconv.ParseInt(string(buf[i:nextI]), 10, 32) - if err != nil { - return int64(i), syserror.EINVAL - } - dsts[j] = int32(val) - - i = nextI - } - - // Skip trailing whitespace. - for i < len(buf) && isASCIIWhitespace(buf[i]) { - i++ - } - - if cperr != nil { - return int64(i), cperr - } - if j == 0 { - return int64(i), syserror.EINVAL - } - return int64(i), nil -} - -// CopyInt32StringInVec is equivalent to CopyInt32StringsInVec, but copies at -// most one int32. -func CopyInt32StringInVec(ctx context.Context, uio IO, ars AddrRangeSeq, dst *int32, opts IOOpts) (int64, error) { - dsts := [1]int32{*dst} - n, err := CopyInt32StringsInVec(ctx, uio, ars, dsts[:], opts) - *dst = dsts[0] - return n, err -} - -// IOSequence holds arguments to IO methods. -type IOSequence struct { - IO IO - Addrs AddrRangeSeq - Opts IOOpts -} - -// NumBytes returns s.Addrs.NumBytes(). -// -// Note that NumBytes() may return 0 even if !s.Addrs.IsEmpty(), since -// s.Addrs may contain a non-zero number of zero-length AddrRanges. -// Many clients of -// IOSequence currently do something like: -// -// if ioseq.NumBytes() == 0 { -// return 0, nil -// } -// if f.availableBytes == 0 { -// return 0, syserror.ErrWouldBlock -// } -// return ioseq.CopyOutFrom(..., reader) -// -// In such cases, using s.Addrs.IsEmpty() will cause them to have the wrong -// behavior for zero-length I/O. However, using s.NumBytes() == 0 instead means -// that we will return success for zero-length I/O in cases where Linux would -// return EFAULT due to a failed access_ok() check, so in the long term we -// should move checks for ErrWouldBlock etc. into the body of -// reader.ReadToBlocks and use s.Addrs.IsEmpty() instead. -func (s IOSequence) NumBytes() int64 { - return s.Addrs.NumBytes() -} - -// DropFirst returns a copy of s with s.Addrs.DropFirst(n). -// -// Preconditions: As for AddrRangeSeq.DropFirst. -func (s IOSequence) DropFirst(n int) IOSequence { - return IOSequence{s.IO, s.Addrs.DropFirst(n), s.Opts} -} - -// DropFirst64 returns a copy of s with s.Addrs.DropFirst64(n). -// -// Preconditions: As for AddrRangeSeq.DropFirst64. -func (s IOSequence) DropFirst64(n int64) IOSequence { - return IOSequence{s.IO, s.Addrs.DropFirst64(n), s.Opts} -} - -// TakeFirst returns a copy of s with s.Addrs.TakeFirst(n). -// -// Preconditions: As for AddrRangeSeq.TakeFirst. -func (s IOSequence) TakeFirst(n int) IOSequence { - return IOSequence{s.IO, s.Addrs.TakeFirst(n), s.Opts} -} - -// TakeFirst64 returns a copy of s with s.Addrs.TakeFirst64(n). -// -// Preconditions: As for AddrRangeSeq.TakeFirst64. -func (s IOSequence) TakeFirst64(n int64) IOSequence { - return IOSequence{s.IO, s.Addrs.TakeFirst64(n), s.Opts} -} - -// CopyOut invokes CopyOutVec over s.Addrs. -// -// As with CopyOutVec, if s.NumBytes() < len(src), the copy will be truncated -// to s.NumBytes(), and a nil error will be returned. -// -// Preconditions: As for CopyOutVec. -func (s IOSequence) CopyOut(ctx context.Context, src []byte) (int, error) { - return CopyOutVec(ctx, s.IO, s.Addrs, src, s.Opts) -} - -// CopyIn invokes CopyInVec over s.Addrs. -// -// As with CopyInVec, if s.NumBytes() < len(dst), the copy will be truncated to -// s.NumBytes(), and a nil error will be returned. -// -// Preconditions: As for CopyInVec. -func (s IOSequence) CopyIn(ctx context.Context, dst []byte) (int, error) { - return CopyInVec(ctx, s.IO, s.Addrs, dst, s.Opts) -} - -// ZeroOut invokes ZeroOutVec over s.Addrs. -// -// As with ZeroOutVec, if s.NumBytes() < toZero, the write will be truncated -// to s.NumBytes(), and a nil error will be returned. -// -// Preconditions: As for ZeroOutVec. -func (s IOSequence) ZeroOut(ctx context.Context, toZero int64) (int64, error) { - return ZeroOutVec(ctx, s.IO, s.Addrs, toZero, s.Opts) -} - -// CopyOutFrom invokes s.CopyOutFrom over s.Addrs. -// -// Preconditions: As for IO.CopyOutFrom. -func (s IOSequence) CopyOutFrom(ctx context.Context, src safemem.Reader) (int64, error) { - return s.IO.CopyOutFrom(ctx, s.Addrs, src, s.Opts) -} - -// CopyInTo invokes s.CopyInTo over s.Addrs. -// -// Preconditions: As for IO.CopyInTo. -func (s IOSequence) CopyInTo(ctx context.Context, dst safemem.Writer) (int64, error) { - return s.IO.CopyInTo(ctx, s.Addrs, dst, s.Opts) -} - -// Reader returns an io.Reader that reads from s. Reads beyond the end of s -// return io.EOF. The preconditions that apply to s.CopyIn also apply to the -// returned io.Reader.Read. -func (s IOSequence) Reader(ctx context.Context) io.Reader { - return &ioSequenceReadWriter{ctx, s} -} - -// Writer returns an io.Writer that writes to s. Writes beyond the end of s -// return ErrEndOfIOSequence. The preconditions that apply to s.CopyOut also -// apply to the returned io.Writer.Write. -func (s IOSequence) Writer(ctx context.Context) io.Writer { - return &ioSequenceReadWriter{ctx, s} -} - -// ErrEndOfIOSequence is returned by IOSequence.Writer().Write() when -// attempting to write beyond the end of the IOSequence. -var ErrEndOfIOSequence = errors.New("write beyond end of IOSequence") - -type ioSequenceReadWriter struct { - ctx context.Context - s IOSequence -} - -// Read implements io.Reader.Read. -func (rw *ioSequenceReadWriter) Read(dst []byte) (int, error) { - n, err := rw.s.CopyIn(rw.ctx, dst) - rw.s = rw.s.DropFirst(n) - if err == nil && rw.s.NumBytes() == 0 { - err = io.EOF - } - return n, err -} - -// Write implements io.Writer.Write. -func (rw *ioSequenceReadWriter) Write(src []byte) (int, error) { - n, err := rw.s.CopyOut(rw.ctx, src) - rw.s = rw.s.DropFirst(n) - if err == nil && n < len(src) { - err = ErrEndOfIOSequence - } - return n, err -} diff --git a/pkg/sentry/usermem/usermem_arm64.go b/pkg/sentry/usermem/usermem_arm64.go deleted file mode 100644 index fdfc30a66..000000000 --- a/pkg/sentry/usermem/usermem_arm64.go +++ /dev/null @@ -1,53 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build arm64 - -package usermem - -import ( - "encoding/binary" - "syscall" -) - -const ( - // PageSize is the system page size. - // arm64 support 4K/16K/64K page size, - // which can be get by syscall.Getpagesize(). - // Currently, only 4K page size is supported. - PageSize = 1 << PageShift - - // HugePageSize is the system huge page size. - HugePageSize = 1 << HugePageShift - - // PageShift is the binary log of the system page size. - PageShift = 12 - - // HugePageShift is the binary log of the system huge page size. - // Should be calculated by "PageShift + (PageShift - 3)" - // when multiple page size support is ready. - HugePageShift = 21 -) - -var ( - // ByteOrder is the native byte order (little endian). - ByteOrder = binary.LittleEndian -) - -func init() { - // Make sure the page size is 4K on arm64 platform. - if size := syscall.Getpagesize(); size != PageSize { - panic("Only 4K page size is supported on arm64!") - } -} diff --git a/pkg/sentry/usermem/usermem_test.go b/pkg/sentry/usermem/usermem_test.go deleted file mode 100644 index 299f64754..000000000 --- a/pkg/sentry/usermem/usermem_test.go +++ /dev/null @@ -1,424 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package usermem - -import ( - "bytes" - "encoding/binary" - "fmt" - "reflect" - "strings" - "testing" - - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/safemem" - "gvisor.dev/gvisor/pkg/syserror" -) - -// newContext returns a context.Context that we can use in these tests (we -// can't use contexttest because it depends on usermem). -func newContext() context.Context { - return context.Background() -} - -func newBytesIOString(s string) *BytesIO { - return &BytesIO{[]byte(s)} -} - -func TestBytesIOCopyOutSuccess(t *testing.T) { - b := newBytesIOString("ABCDE") - n, err := b.CopyOut(newContext(), 1, []byte("foo"), IOOpts{}) - if wantN := 3; n != wantN || err != nil { - t.Errorf("CopyOut: got (%v, %v), wanted (%v, nil)", n, err, wantN) - } - if got, want := b.Bytes, []byte("AfooE"); !bytes.Equal(got, want) { - t.Errorf("Bytes: got %q, wanted %q", got, want) - } -} - -func TestBytesIOCopyOutFailure(t *testing.T) { - b := newBytesIOString("ABC") - n, err := b.CopyOut(newContext(), 1, []byte("foo"), IOOpts{}) - if wantN, wantErr := 2, syserror.EFAULT; n != wantN || err != wantErr { - t.Errorf("CopyOut: got (%v, %v), wanted (%v, %v)", n, err, wantN, wantErr) - } - if got, want := b.Bytes, []byte("Afo"); !bytes.Equal(got, want) { - t.Errorf("Bytes: got %q, wanted %q", got, want) - } -} - -func TestBytesIOCopyInSuccess(t *testing.T) { - b := newBytesIOString("AfooE") - var dst [3]byte - n, err := b.CopyIn(newContext(), 1, dst[:], IOOpts{}) - if wantN := 3; n != wantN || err != nil { - t.Errorf("CopyIn: got (%v, %v), wanted (%v, nil)", n, err, wantN) - } - if got, want := dst[:], []byte("foo"); !bytes.Equal(got, want) { - t.Errorf("dst: got %q, wanted %q", got, want) - } -} - -func TestBytesIOCopyInFailure(t *testing.T) { - b := newBytesIOString("Afo") - var dst [3]byte - n, err := b.CopyIn(newContext(), 1, dst[:], IOOpts{}) - if wantN, wantErr := 2, syserror.EFAULT; n != wantN || err != wantErr { - t.Errorf("CopyIn: got (%v, %v), wanted (%v, %v)", n, err, wantN, wantErr) - } - if got, want := dst[:], []byte("fo\x00"); !bytes.Equal(got, want) { - t.Errorf("dst: got %q, wanted %q", got, want) - } -} - -func TestBytesIOZeroOutSuccess(t *testing.T) { - b := newBytesIOString("ABCD") - n, err := b.ZeroOut(newContext(), 1, 2, IOOpts{}) - if wantN := int64(2); n != wantN || err != nil { - t.Errorf("ZeroOut: got (%v, %v), wanted (%v, nil)", n, err, wantN) - } - if got, want := b.Bytes, []byte("A\x00\x00D"); !bytes.Equal(got, want) { - t.Errorf("Bytes: got %q, wanted %q", got, want) - } -} - -func TestBytesIOZeroOutFailure(t *testing.T) { - b := newBytesIOString("ABC") - n, err := b.ZeroOut(newContext(), 1, 3, IOOpts{}) - if wantN, wantErr := int64(2), syserror.EFAULT; n != wantN || err != wantErr { - t.Errorf("ZeroOut: got (%v, %v), wanted (%v, %v)", n, err, wantN, wantErr) - } - if got, want := b.Bytes, []byte("A\x00\x00"); !bytes.Equal(got, want) { - t.Errorf("Bytes: got %q, wanted %q", got, want) - } -} - -func TestBytesIOCopyOutFromSuccess(t *testing.T) { - b := newBytesIOString("ABCDEFGH") - n, err := b.CopyOutFrom(newContext(), AddrRangeSeqFromSlice([]AddrRange{ - {Start: 4, End: 7}, - {Start: 1, End: 4}, - }), safemem.FromIOReader{bytes.NewBufferString("barfoo")}, IOOpts{}) - if wantN := int64(6); n != wantN || err != nil { - t.Errorf("CopyOutFrom: got (%v, %v), wanted (%v, nil)", n, err, wantN) - } - if got, want := b.Bytes, []byte("AfoobarH"); !bytes.Equal(got, want) { - t.Errorf("Bytes: got %q, wanted %q", got, want) - } -} - -func TestBytesIOCopyOutFromFailure(t *testing.T) { - b := newBytesIOString("ABCDE") - n, err := b.CopyOutFrom(newContext(), AddrRangeSeqFromSlice([]AddrRange{ - {Start: 1, End: 4}, - {Start: 4, End: 7}, - }), safemem.FromIOReader{bytes.NewBufferString("foobar")}, IOOpts{}) - if wantN, wantErr := int64(4), syserror.EFAULT; n != wantN || err != wantErr { - t.Errorf("CopyOutFrom: got (%v, %v), wanted (%v, %v)", n, err, wantN, wantErr) - } - if got, want := b.Bytes, []byte("Afoob"); !bytes.Equal(got, want) { - t.Errorf("Bytes: got %q, wanted %q", got, want) - } -} - -func TestBytesIOCopyInToSuccess(t *testing.T) { - b := newBytesIOString("AfoobarH") - var dst bytes.Buffer - n, err := b.CopyInTo(newContext(), AddrRangeSeqFromSlice([]AddrRange{ - {Start: 4, End: 7}, - {Start: 1, End: 4}, - }), safemem.FromIOWriter{&dst}, IOOpts{}) - if wantN := int64(6); n != wantN || err != nil { - t.Errorf("CopyInTo: got (%v, %v), wanted (%v, nil)", n, err, wantN) - } - if got, want := dst.Bytes(), []byte("barfoo"); !bytes.Equal(got, want) { - t.Errorf("dst.Bytes(): got %q, wanted %q", got, want) - } -} - -func TestBytesIOCopyInToFailure(t *testing.T) { - b := newBytesIOString("Afoob") - var dst bytes.Buffer - n, err := b.CopyInTo(newContext(), AddrRangeSeqFromSlice([]AddrRange{ - {Start: 1, End: 4}, - {Start: 4, End: 7}, - }), safemem.FromIOWriter{&dst}, IOOpts{}) - if wantN, wantErr := int64(4), syserror.EFAULT; n != wantN || err != wantErr { - t.Errorf("CopyOutFrom: got (%v, %v), wanted (%v, %v)", n, err, wantN, wantErr) - } - if got, want := dst.Bytes(), []byte("foob"); !bytes.Equal(got, want) { - t.Errorf("dst.Bytes(): got %q, wanted %q", got, want) - } -} - -type testStruct struct { - Int8 int8 - Uint8 uint8 - Int16 int16 - Uint16 uint16 - Int32 int32 - Uint32 uint32 - Int64 int64 - Uint64 uint64 -} - -func TestCopyObject(t *testing.T) { - wantObj := testStruct{1, 2, 3, 4, 5, 6, 7, 8} - wantN := binary.Size(wantObj) - b := &BytesIO{make([]byte, wantN)} - ctx := newContext() - if n, err := CopyObjectOut(ctx, b, 0, &wantObj, IOOpts{}); n != wantN || err != nil { - t.Fatalf("CopyObjectOut: got (%v, %v), wanted (%v, nil)", n, err, wantN) - } - var gotObj testStruct - if n, err := CopyObjectIn(ctx, b, 0, &gotObj, IOOpts{}); n != wantN || err != nil { - t.Errorf("CopyObjectIn: got (%v, %v), wanted (%v, nil)", n, err, wantN) - } - if gotObj != wantObj { - t.Errorf("CopyObject round trip: got %+v, wanted %+v", gotObj, wantObj) - } -} - -func TestCopyStringInShort(t *testing.T) { - // Tests for string length <= copyStringIncrement. - want := strings.Repeat("A", copyStringIncrement-2) - mem := want + "\x00" - if got, err := CopyStringIn(newContext(), newBytesIOString(mem), 0, 2*copyStringIncrement, IOOpts{}); got != want || err != nil { - t.Errorf("CopyStringIn: got (%q, %v), wanted (%q, nil)", got, err, want) - } -} - -func TestCopyStringInLong(t *testing.T) { - // Tests for copyStringIncrement < string length <= copyStringMaxInitBufLen - // (requiring multiple calls to IO.CopyIn()). - want := strings.Repeat("A", copyStringIncrement*3/4) + strings.Repeat("B", copyStringIncrement*3/4) - mem := want + "\x00" - if got, err := CopyStringIn(newContext(), newBytesIOString(mem), 0, 2*copyStringIncrement, IOOpts{}); got != want || err != nil { - t.Errorf("CopyStringIn: got (%q, %v), wanted (%q, nil)", got, err, want) - } -} - -func TestCopyStringInVeryLong(t *testing.T) { - // Tests for string length > copyStringMaxInitBufLen (requiring buffer - // reallocation). - want := strings.Repeat("A", copyStringMaxInitBufLen*3/4) + strings.Repeat("B", copyStringMaxInitBufLen*3/4) - mem := want + "\x00" - if got, err := CopyStringIn(newContext(), newBytesIOString(mem), 0, 2*copyStringMaxInitBufLen, IOOpts{}); got != want || err != nil { - t.Errorf("CopyStringIn: got (%q, %v), wanted (%q, nil)", got, err, want) - } -} - -func TestCopyStringInNoTerminatingZeroByte(t *testing.T) { - want := strings.Repeat("A", copyStringIncrement-1) - got, err := CopyStringIn(newContext(), newBytesIOString(want), 0, 2*copyStringIncrement, IOOpts{}) - if wantErr := syserror.EFAULT; got != want || err != wantErr { - t.Errorf("CopyStringIn: got (%q, %v), wanted (%q, %v)", got, err, want, wantErr) - } -} - -func TestCopyStringInTruncatedByMaxlen(t *testing.T) { - got, err := CopyStringIn(newContext(), newBytesIOString(strings.Repeat("A", 10)), 0, 5, IOOpts{}) - if want, wantErr := strings.Repeat("A", 5), syserror.ENAMETOOLONG; got != want || err != wantErr { - t.Errorf("CopyStringIn: got (%q, %v), wanted (%q, %v)", got, err, want, wantErr) - } -} - -func TestCopyInt32StringsInVec(t *testing.T) { - for _, test := range []struct { - str string - n int - initial []int32 - final []int32 - }{ - { - str: "100 200", - n: len("100 200"), - initial: []int32{1, 2}, - final: []int32{100, 200}, - }, - { - // Fewer values ok - str: "100", - n: len("100"), - initial: []int32{1, 2}, - final: []int32{100, 2}, - }, - { - // Extra values ok - str: "100 200 300", - n: len("100 200 "), - initial: []int32{1, 2}, - final: []int32{100, 200}, - }, - { - // Leading and trailing whitespace ok - str: " 100\t200\n", - n: len(" 100\t200\n"), - initial: []int32{1, 2}, - final: []int32{100, 200}, - }, - } { - t.Run(fmt.Sprintf("%q", test.str), func(t *testing.T) { - src := BytesIOSequence([]byte(test.str)) - dsts := append([]int32(nil), test.initial...) - if n, err := CopyInt32StringsInVec(newContext(), src.IO, src.Addrs, dsts, src.Opts); n != int64(test.n) || err != nil { - t.Errorf("CopyInt32StringsInVec: got (%d, %v), wanted (%d, nil)", n, err, test.n) - } - if !reflect.DeepEqual(dsts, test.final) { - t.Errorf("dsts: got %v, wanted %v", dsts, test.final) - } - }) - } -} - -func TestCopyInt32StringsInVecRequiresOneValidValue(t *testing.T) { - for _, s := range []string{"", "\n", "a123"} { - t.Run(fmt.Sprintf("%q", s), func(t *testing.T) { - src := BytesIOSequence([]byte(s)) - initial := []int32{1, 2} - dsts := append([]int32(nil), initial...) - if n, err := CopyInt32StringsInVec(newContext(), src.IO, src.Addrs, dsts, src.Opts); err != syserror.EINVAL { - t.Errorf("CopyInt32StringsInVec: got (%d, %v), wanted (_, %v)", n, err, syserror.EINVAL) - } - if !reflect.DeepEqual(dsts, initial) { - t.Errorf("dsts: got %v, wanted %v", dsts, initial) - } - }) - } -} - -func TestIOSequenceCopyOut(t *testing.T) { - buf := []byte("ABCD") - s := BytesIOSequence(buf) - - // CopyOut limited by len(src). - n, err := s.CopyOut(newContext(), []byte("fo")) - if wantN := 2; n != wantN || err != nil { - t.Errorf("CopyOut: got (%v, %v), wanted (%v, nil)", n, err, wantN) - } - if want := []byte("foCD"); !bytes.Equal(buf, want) { - t.Errorf("buf: got %q, wanted %q", buf, want) - } - s = s.DropFirst(2) - if got, want := s.NumBytes(), int64(2); got != want { - t.Errorf("NumBytes: got %v, wanted %v", got, want) - } - - // CopyOut limited by s.NumBytes(). - n, err = s.CopyOut(newContext(), []byte("obar")) - if wantN := 2; n != wantN || err != nil { - t.Errorf("CopyOut: got (%v, %v), wanted (%v, nil)", n, err, wantN) - } - if want := []byte("foob"); !bytes.Equal(buf, want) { - t.Errorf("buf: got %q, wanted %q", buf, want) - } - s = s.DropFirst(2) - if got, want := s.NumBytes(), int64(0); got != want { - t.Errorf("NumBytes: got %v, wanted %v", got, want) - } -} - -func TestIOSequenceCopyIn(t *testing.T) { - s := BytesIOSequence([]byte("foob")) - dst := []byte("ABCDEF") - - // CopyIn limited by len(dst). - n, err := s.CopyIn(newContext(), dst[:2]) - if wantN := 2; n != wantN || err != nil { - t.Errorf("CopyIn: got (%v, %v), wanted (%v, nil)", n, err, wantN) - } - if want := []byte("foCDEF"); !bytes.Equal(dst, want) { - t.Errorf("dst: got %q, wanted %q", dst, want) - } - s = s.DropFirst(2) - if got, want := s.NumBytes(), int64(2); got != want { - t.Errorf("NumBytes: got %v, wanted %v", got, want) - } - - // CopyIn limited by s.Remaining(). - n, err = s.CopyIn(newContext(), dst[2:]) - if wantN := 2; n != wantN || err != nil { - t.Errorf("CopyIn: got (%v, %v), wanted (%v, nil)", n, err, wantN) - } - if want := []byte("foobEF"); !bytes.Equal(dst, want) { - t.Errorf("dst: got %q, wanted %q", dst, want) - } - s = s.DropFirst(2) - if got, want := s.NumBytes(), int64(0); got != want { - t.Errorf("NumBytes: got %v, wanted %v", got, want) - } -} - -func TestIOSequenceZeroOut(t *testing.T) { - buf := []byte("ABCD") - s := BytesIOSequence(buf) - - // ZeroOut limited by toZero. - n, err := s.ZeroOut(newContext(), 2) - if wantN := int64(2); n != wantN || err != nil { - t.Errorf("ZeroOut: got (%v, %v), wanted (%v, nil)", n, err, wantN) - } - if want := []byte("\x00\x00CD"); !bytes.Equal(buf, want) { - t.Errorf("buf: got %q, wanted %q", buf, want) - } - s = s.DropFirst(2) - if got, want := s.NumBytes(), int64(2); got != want { - t.Errorf("NumBytes: got %v, wanted %v", got, want) - } - - // ZeroOut limited by s.NumBytes(). - n, err = s.ZeroOut(newContext(), 4) - if wantN := int64(2); n != wantN || err != nil { - t.Errorf("CopyOut: got (%v, %v), wanted (%v, nil)", n, err, wantN) - } - if want := []byte("\x00\x00\x00\x00"); !bytes.Equal(buf, want) { - t.Errorf("buf: got %q, wanted %q", buf, want) - } - s = s.DropFirst(2) - if got, want := s.NumBytes(), int64(0); got != want { - t.Errorf("NumBytes: got %v, wanted %v", got, want) - } -} - -func TestIOSequenceTakeFirst(t *testing.T) { - s := BytesIOSequence([]byte("foobar")) - if got, want := s.NumBytes(), int64(6); got != want { - t.Errorf("NumBytes: got %v, wanted %v", got, want) - } - - s = s.TakeFirst(3) - if got, want := s.NumBytes(), int64(3); got != want { - t.Errorf("NumBytes: got %v, wanted %v", got, want) - } - - // TakeFirst(n) where n > s.NumBytes() is a no-op. - s = s.TakeFirst(9) - if got, want := s.NumBytes(), int64(3); got != want { - t.Errorf("NumBytes: got %v, wanted %v", got, want) - } - - var dst [3]byte - n, err := s.CopyIn(newContext(), dst[:]) - if wantN := 3; n != wantN || err != nil { - t.Errorf("CopyIn: got (%v, %v), wanted (%v, nil)", n, err, wantN) - } - if got, want := dst[:], []byte("foo"); !bytes.Equal(got, want) { - t.Errorf("dst: got %q, wanted %q", got, want) - } - s = s.DropFirst(3) - if got, want := s.NumBytes(), int64(0); got != want { - t.Errorf("NumBytes: got %v, wanted %v", got, want) - } -} diff --git a/pkg/sentry/usermem/usermem_unsafe.go b/pkg/sentry/usermem/usermem_unsafe.go deleted file mode 100644 index 876783e78..000000000 --- a/pkg/sentry/usermem/usermem_unsafe.go +++ /dev/null @@ -1,27 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package usermem - -import ( - "unsafe" -) - -// stringFromImmutableBytes is equivalent to string(bs), except that it never -// copies even if escape analysis can't prove that bs does not escape. This is -// only valid if bs is never mutated after stringFromImmutableBytes returns. -func stringFromImmutableBytes(bs []byte) string { - // Compare strings.Builder.String(). - return *(*string)(unsafe.Pointer(&bs)) -} diff --git a/pkg/sentry/usermem/usermem_x86.go b/pkg/sentry/usermem/usermem_x86.go deleted file mode 100644 index 8059b72d2..000000000 --- a/pkg/sentry/usermem/usermem_x86.go +++ /dev/null @@ -1,38 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build amd64 i386 - -package usermem - -import "encoding/binary" - -const ( - // PageSize is the system page size. - PageSize = 1 << PageShift - - // HugePageSize is the system huge page size. - HugePageSize = 1 << HugePageShift - - // PageShift is the binary log of the system page size. - PageShift = 12 - - // HugePageShift is the binary log of the system huge page size. - HugePageShift = 21 -) - -var ( - // ByteOrder is the native byte order (little endian). - ByteOrder = binary.LittleEndian -) diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index 51acdc4e9..6b1009328 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -26,14 +26,14 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/fspath", "//pkg/sentry/arch", - "//pkg/sentry/context", "//pkg/sentry/kernel/auth", "//pkg/sentry/memmap", - "//pkg/sentry/usermem", "//pkg/sync", "//pkg/syserror", + "//pkg/usermem", "//pkg/waiter", ], ) @@ -48,11 +48,11 @@ go_test( library = ":vfs", deps = [ "//pkg/abi/linux", - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", + "//pkg/context", + "//pkg/sentry/contexttest", "//pkg/sentry/kernel/auth", - "//pkg/sentry/usermem", "//pkg/sync", "//pkg/syserror", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/vfs/context.go b/pkg/sentry/vfs/context.go index 705194ebc..d97362b9a 100644 --- a/pkg/sentry/vfs/context.go +++ b/pkg/sentry/vfs/context.go @@ -15,7 +15,7 @@ package vfs import ( - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" ) // contextID is this package's type for context.Context.Value keys. diff --git a/pkg/sentry/vfs/device.go b/pkg/sentry/vfs/device.go index 9f9d6e783..3af2aa58d 100644 --- a/pkg/sentry/vfs/device.go +++ b/pkg/sentry/vfs/device.go @@ -17,7 +17,7 @@ package vfs import ( "fmt" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 51c95c2d9..225024463 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -18,12 +18,12 @@ import ( "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/vfs/file_description_impl_util.go b/pkg/sentry/vfs/file_description_impl_util.go index c00b3c84b..fb9b87fdc 100644 --- a/pkg/sentry/vfs/file_description_impl_util.go +++ b/pkg/sentry/vfs/file_description_impl_util.go @@ -19,12 +19,12 @@ import ( "io" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/vfs/file_description_impl_util_test.go b/pkg/sentry/vfs/file_description_impl_util_test.go index 9ed58512f..1720d325d 100644 --- a/pkg/sentry/vfs/file_description_impl_util_test.go +++ b/pkg/sentry/vfs/file_description_impl_util_test.go @@ -22,11 +22,11 @@ import ( "testing" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // fileDescription is the common fd struct which a filesystem implementation diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index ea78f555b..a06a6caf3 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -18,8 +18,8 @@ import ( "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/context" ) // A Filesystem is a tree of nodes represented by Dentries, which forms part of diff --git a/pkg/sentry/vfs/filesystem_type.go b/pkg/sentry/vfs/filesystem_type.go index 023301780..c58b70728 100644 --- a/pkg/sentry/vfs/filesystem_type.go +++ b/pkg/sentry/vfs/filesystem_type.go @@ -18,7 +18,7 @@ import ( "bytes" "fmt" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ) diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index 00177b371..d39528051 100644 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go @@ -19,7 +19,7 @@ import ( "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/vfs/pathname.go b/pkg/sentry/vfs/pathname.go index cf80df90e..b318c681a 100644 --- a/pkg/sentry/vfs/pathname.go +++ b/pkg/sentry/vfs/pathname.go @@ -15,8 +15,8 @@ package vfs import ( + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/vfs/testutil.go b/pkg/sentry/vfs/testutil.go index ee5c8b9e2..392c7611e 100644 --- a/pkg/sentry/vfs/testutil.go +++ b/pkg/sentry/vfs/testutil.go @@ -18,8 +18,8 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/syserror" ) diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 1f6f56293..b2bf48853 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -31,8 +31,8 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" diff --git a/pkg/usermem/BUILD b/pkg/usermem/BUILD new file mode 100644 index 000000000..ff8b9e91a --- /dev/null +++ b/pkg/usermem/BUILD @@ -0,0 +1,55 @@ +load("//tools:defs.bzl", "go_library", "go_test") +load("//tools/go_generics:defs.bzl", "go_template_instance") + +package(licenses = ["notice"]) + +go_template_instance( + name = "addr_range", + out = "addr_range.go", + package = "usermem", + prefix = "Addr", + template = "//pkg/segment:generic_range", + types = { + "T": "Addr", + }, +) + +go_library( + name = "usermem", + srcs = [ + "access_type.go", + "addr.go", + "addr_range.go", + "addr_range_seq_unsafe.go", + "bytes_io.go", + "bytes_io_unsafe.go", + "usermem.go", + "usermem_arm64.go", + "usermem_unsafe.go", + "usermem_x86.go", + ], + visibility = ["//:sandbox"], + deps = [ + "//pkg/atomicbitops", + "//pkg/binary", + "//pkg/context", + "//pkg/log", + "//pkg/safemem", + "//pkg/syserror", + ], +) + +go_test( + name = "usermem_test", + size = "small", + srcs = [ + "addr_range_seq_test.go", + "usermem_test.go", + ], + library = ":usermem", + deps = [ + "//pkg/context", + "//pkg/safemem", + "//pkg/syserror", + ], +) diff --git a/pkg/usermem/README.md b/pkg/usermem/README.md new file mode 100644 index 000000000..f6d2137eb --- /dev/null +++ b/pkg/usermem/README.md @@ -0,0 +1,31 @@ +This package defines primitives for sentry access to application memory. + +Major types: + +- The `IO` interface represents a virtual address space and provides I/O + methods on that address space. `IO` is the lowest-level primitive. The + primary implementation of the `IO` interface is `mm.MemoryManager`. + +- `IOSequence` represents a collection of individually-contiguous address + ranges in a `IO` that is operated on sequentially, analogous to Linux's + `struct iov_iter`. + +Major usage patterns: + +- Access to a task's virtual memory, subject to the application's memory + protections and while running on that task's goroutine, from a context that + is at or above the level of the `kernel` package (e.g. most syscall + implementations in `syscalls/linux`); use the `kernel.Task.Copy*` wrappers + defined in `kernel/task_usermem.go`. + +- Access to a task's virtual memory, from a context that is at or above the + level of the `kernel` package, but where any of the above constraints does + not hold (e.g. `PTRACE_POKEDATA`, which ignores application memory + protections); obtain the task's `mm.MemoryManager` by calling + `kernel.Task.MemoryManager`, and call its `IO` methods directly. + +- Access to a task's virtual memory, from a context that is below the level of + the `kernel` package (e.g. filesystem I/O); clients must pass I/O arguments + from higher layers, usually in the form of an `IOSequence`. The + `kernel.Task.SingleIOSequence` and `kernel.Task.IovecsIOSequence` functions + in `kernel/task_usermem.go` are convenience functions for doing so. diff --git a/pkg/usermem/access_type.go b/pkg/usermem/access_type.go new file mode 100644 index 000000000..9c1742a59 --- /dev/null +++ b/pkg/usermem/access_type.go @@ -0,0 +1,128 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package usermem + +import ( + "syscall" +) + +// AccessType specifies memory access types. This is used for +// setting mapping permissions, as well as communicating faults. +// +// +stateify savable +type AccessType struct { + // Read is read access. + Read bool + + // Write is write access. + Write bool + + // Execute is executable access. + Execute bool +} + +// String returns a pretty representation of access. This looks like the +// familiar r-x, rw-, etc. and can be relied on as such. +func (a AccessType) String() string { + bits := [3]byte{'-', '-', '-'} + if a.Read { + bits[0] = 'r' + } + if a.Write { + bits[1] = 'w' + } + if a.Execute { + bits[2] = 'x' + } + return string(bits[:]) +} + +// Any returns true iff at least one of Read, Write or Execute is true. +func (a AccessType) Any() bool { + return a.Read || a.Write || a.Execute +} + +// Prot returns the system prot (syscall.PROT_READ, etc.) for this access. +func (a AccessType) Prot() int { + var prot int + if a.Read { + prot |= syscall.PROT_READ + } + if a.Write { + prot |= syscall.PROT_WRITE + } + if a.Execute { + prot |= syscall.PROT_EXEC + } + return prot +} + +// SupersetOf returns true iff the access types in a are a superset of the +// access types in other. +func (a AccessType) SupersetOf(other AccessType) bool { + if !a.Read && other.Read { + return false + } + if !a.Write && other.Write { + return false + } + if !a.Execute && other.Execute { + return false + } + return true +} + +// Intersect returns the access types set in both a and other. +func (a AccessType) Intersect(other AccessType) AccessType { + return AccessType{ + Read: a.Read && other.Read, + Write: a.Write && other.Write, + Execute: a.Execute && other.Execute, + } +} + +// Union returns the access types set in either a or other. +func (a AccessType) Union(other AccessType) AccessType { + return AccessType{ + Read: a.Read || other.Read, + Write: a.Write || other.Write, + Execute: a.Execute || other.Execute, + } +} + +// Effective returns the set of effective access types allowed by a, even if +// some types are not explicitly allowed. +func (a AccessType) Effective() AccessType { + // In Linux, Write and Execute access generally imply Read access. See + // mm/mmap.c:protection_map. + // + // The notable exception is get_user_pages, which only checks against + // the original vma flags. That said, most user memory accesses do not + // use GUP. + if a.Write || a.Execute { + a.Read = true + } + return a +} + +// Convenient access types. +var ( + NoAccess = AccessType{} + Read = AccessType{Read: true} + Write = AccessType{Write: true} + Execute = AccessType{Execute: true} + ReadWrite = AccessType{Read: true, Write: true} + AnyAccess = AccessType{Read: true, Write: true, Execute: true} +) diff --git a/pkg/usermem/addr.go b/pkg/usermem/addr.go new file mode 100644 index 000000000..e79210804 --- /dev/null +++ b/pkg/usermem/addr.go @@ -0,0 +1,108 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package usermem + +import ( + "fmt" +) + +// Addr represents a generic virtual address. +// +// +stateify savable +type Addr uintptr + +// AddLength adds the given length to start and returns the result. ok is true +// iff adding the length did not overflow the range of Addr. +// +// Note: This function is usually used to get the end of an address range +// defined by its start address and length. Since the resulting end is +// exclusive, end == 0 is technically valid, and corresponds to a range that +// extends to the end of the address space, but ok will be false. This isn't +// expected to ever come up in practice. +func (v Addr) AddLength(length uint64) (end Addr, ok bool) { + end = v + Addr(length) + // The second half of the following check is needed in case uintptr is + // smaller than 64 bits. + ok = end >= v && length <= uint64(^Addr(0)) + return +} + +// RoundDown returns the address rounded down to the nearest page boundary. +func (v Addr) RoundDown() Addr { + return v & ^Addr(PageSize-1) +} + +// RoundUp returns the address rounded up to the nearest page boundary. ok is +// true iff rounding up did not wrap around. +func (v Addr) RoundUp() (addr Addr, ok bool) { + addr = Addr(v + PageSize - 1).RoundDown() + ok = addr >= v + return +} + +// MustRoundUp is equivalent to RoundUp, but panics if rounding up wraps +// around. +func (v Addr) MustRoundUp() Addr { + addr, ok := v.RoundUp() + if !ok { + panic(fmt.Sprintf("usermem.Addr(%d).RoundUp() wraps", v)) + } + return addr +} + +// HugeRoundDown returns the address rounded down to the nearest huge page +// boundary. +func (v Addr) HugeRoundDown() Addr { + return v & ^Addr(HugePageSize-1) +} + +// HugeRoundUp returns the address rounded up to the nearest huge page boundary. +// ok is true iff rounding up did not wrap around. +func (v Addr) HugeRoundUp() (addr Addr, ok bool) { + addr = Addr(v + HugePageSize - 1).HugeRoundDown() + ok = addr >= v + return +} + +// PageOffset returns the offset of v into the current page. +func (v Addr) PageOffset() uint64 { + return uint64(v & Addr(PageSize-1)) +} + +// IsPageAligned returns true if v.PageOffset() == 0. +func (v Addr) IsPageAligned() bool { + return v.PageOffset() == 0 +} + +// AddrRange is a range of Addrs. +// +// type AddrRange + +// ToRange returns [v, v+length). +func (v Addr) ToRange(length uint64) (AddrRange, bool) { + end, ok := v.AddLength(length) + return AddrRange{v, end}, ok +} + +// IsPageAligned returns true if ar.Start.IsPageAligned() and +// ar.End.IsPageAligned(). +func (ar AddrRange) IsPageAligned() bool { + return ar.Start.IsPageAligned() && ar.End.IsPageAligned() +} + +// String implements fmt.Stringer.String. +func (ar AddrRange) String() string { + return fmt.Sprintf("[%#x, %#x)", ar.Start, ar.End) +} diff --git a/pkg/usermem/addr_range_seq_test.go b/pkg/usermem/addr_range_seq_test.go new file mode 100644 index 000000000..82f735026 --- /dev/null +++ b/pkg/usermem/addr_range_seq_test.go @@ -0,0 +1,197 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package usermem + +import ( + "testing" +) + +var addrRangeSeqTests = []struct { + desc string + ranges []AddrRange +}{ + { + desc: "Empty sequence", + }, + { + desc: "Single empty AddrRange", + ranges: []AddrRange{ + {0x10, 0x10}, + }, + }, + { + desc: "Single non-empty AddrRange of length 1", + ranges: []AddrRange{ + {0x10, 0x11}, + }, + }, + { + desc: "Single non-empty AddrRange of length 2", + ranges: []AddrRange{ + {0x10, 0x12}, + }, + }, + { + desc: "Multiple non-empty AddrRanges", + ranges: []AddrRange{ + {0x10, 0x11}, + {0x20, 0x22}, + }, + }, + { + desc: "Multiple AddrRanges including empty AddrRanges", + ranges: []AddrRange{ + {0x10, 0x10}, + {0x20, 0x20}, + {0x30, 0x33}, + {0x40, 0x44}, + {0x50, 0x50}, + {0x60, 0x60}, + {0x70, 0x77}, + {0x80, 0x88}, + {0x90, 0x90}, + {0xa0, 0xa0}, + }, + }, +} + +func testAddrRangeSeqEqualityWithTailIteration(t *testing.T, ars AddrRangeSeq, wantRanges []AddrRange) { + var wantLen int64 + for _, ar := range wantRanges { + wantLen += int64(ar.Length()) + } + + var i int + for !ars.IsEmpty() { + if gotLen := ars.NumBytes(); gotLen != wantLen { + t.Errorf("Iteration %d: %v.NumBytes(): got %d, wanted %d", i, ars, gotLen, wantLen) + } + if gotN, wantN := ars.NumRanges(), len(wantRanges)-i; gotN != wantN { + t.Errorf("Iteration %d: %v.NumRanges(): got %d, wanted %d", i, ars, gotN, wantN) + } + got := ars.Head() + if i >= len(wantRanges) { + t.Errorf("Iteration %d: %v.Head(): got %s, wanted ", i, ars, got) + } else if want := wantRanges[i]; got != want { + t.Errorf("Iteration %d: %v.Head(): got %s, wanted %s", i, ars, got, want) + } + ars = ars.Tail() + wantLen -= int64(got.Length()) + i++ + } + if gotLen := ars.NumBytes(); gotLen != 0 || wantLen != 0 { + t.Errorf("Iteration %d: %v.NumBytes(): got %d, wanted %d (which should be 0)", i, ars, gotLen, wantLen) + } + if gotN := ars.NumRanges(); gotN != 0 { + t.Errorf("Iteration %d: %v.NumRanges(): got %d, wanted 0", i, ars, gotN) + } +} + +func TestAddrRangeSeqTailIteration(t *testing.T) { + for _, test := range addrRangeSeqTests { + t.Run(test.desc, func(t *testing.T) { + testAddrRangeSeqEqualityWithTailIteration(t, AddrRangeSeqFromSlice(test.ranges), test.ranges) + }) + } +} + +func TestAddrRangeSeqDropFirstEmpty(t *testing.T) { + var ars AddrRangeSeq + if got, want := ars.DropFirst(1), ars; got != want { + t.Errorf("%v.DropFirst(1): got %v, wanted %v", ars, got, want) + } +} + +func TestAddrRangeSeqDropSingleByteIteration(t *testing.T) { + // Tests AddrRangeSeq iteration using Head/DropFirst, simulating + // I/O-per-AddrRange. + for _, test := range addrRangeSeqTests { + t.Run(test.desc, func(t *testing.T) { + // Figure out what AddrRanges we expect to see. + var wantLen int64 + var wantRanges []AddrRange + for _, ar := range test.ranges { + wantLen += int64(ar.Length()) + wantRanges = append(wantRanges, ar) + if ar.Length() == 0 { + // We "do" 0 bytes of I/O and then call DropFirst(0), + // advancing to the next AddrRange. + continue + } + // Otherwise we "do" 1 byte of I/O and then call DropFirst(1), + // advancing the AddrRange by 1 byte, or to the next AddrRange + // if this one is exhausted. + for ar.Start++; ar.Length() != 0; ar.Start++ { + wantRanges = append(wantRanges, ar) + } + } + t.Logf("Expected AddrRanges: %s (%d bytes)", wantRanges, wantLen) + + ars := AddrRangeSeqFromSlice(test.ranges) + var i int + for !ars.IsEmpty() { + if gotLen := ars.NumBytes(); gotLen != wantLen { + t.Errorf("Iteration %d: %v.NumBytes(): got %d, wanted %d", i, ars, gotLen, wantLen) + } + got := ars.Head() + if i >= len(wantRanges) { + t.Errorf("Iteration %d: %v.Head(): got %s, wanted ", i, ars, got) + } else if want := wantRanges[i]; got != want { + t.Errorf("Iteration %d: %v.Head(): got %s, wanted %s", i, ars, got, want) + } + if got.Length() == 0 { + ars = ars.DropFirst(0) + } else { + ars = ars.DropFirst(1) + wantLen-- + } + i++ + } + if gotLen := ars.NumBytes(); gotLen != 0 || wantLen != 0 { + t.Errorf("Iteration %d: %v.NumBytes(): got %d, wanted %d (which should be 0)", i, ars, gotLen, wantLen) + } + }) + } +} + +func TestAddrRangeSeqTakeFirstEmpty(t *testing.T) { + var ars AddrRangeSeq + if got, want := ars.TakeFirst(1), ars; got != want { + t.Errorf("%v.TakeFirst(1): got %v, wanted %v", ars, got, want) + } +} + +func TestAddrRangeSeqTakeFirst(t *testing.T) { + ranges := []AddrRange{ + {0x10, 0x11}, + {0x20, 0x22}, + {0x30, 0x30}, + {0x40, 0x44}, + {0x50, 0x55}, + {0x60, 0x60}, + {0x70, 0x77}, + } + ars := AddrRangeSeqFromSlice(ranges).TakeFirst(5) + want := []AddrRange{ + {0x10, 0x11}, // +1 byte (total 1 byte), not truncated + {0x20, 0x22}, // +2 bytes (total 3 bytes), not truncated + {0x30, 0x30}, // +0 bytes (total 3 bytes), no change + {0x40, 0x42}, // +2 bytes (total 5 bytes), partially truncated + {0x50, 0x50}, // +0 bytes (total 5 bytes), fully truncated + {0x60, 0x60}, // +0 bytes (total 5 bytes), "fully truncated" (no change) + {0x70, 0x70}, // +0 bytes (total 5 bytes), fully truncated + } + testAddrRangeSeqEqualityWithTailIteration(t, ars, want) +} diff --git a/pkg/usermem/addr_range_seq_unsafe.go b/pkg/usermem/addr_range_seq_unsafe.go new file mode 100644 index 000000000..c09337c15 --- /dev/null +++ b/pkg/usermem/addr_range_seq_unsafe.go @@ -0,0 +1,277 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package usermem + +import ( + "bytes" + "fmt" + "reflect" + "unsafe" +) + +// An AddrRangeSeq represents a sequence of AddrRanges. +// +// AddrRangeSeqs are immutable and may be copied by value. The zero value of +// AddrRangeSeq represents an empty sequence. +// +// An AddrRangeSeq may contain AddrRanges with a length of 0. This is necessary +// since zero-length AddrRanges are significant to MM bounds checks. +type AddrRangeSeq struct { + // If length is 0, then the AddrRangeSeq represents no AddrRanges. + // Invariants: data == 0; offset == 0; limit == 0. + // + // If length is 1, then the AddrRangeSeq represents the single + // AddrRange{offset, offset+limit}. Invariants: data == 0. + // + // Otherwise, length >= 2, and the AddrRangeSeq represents the `length` + // AddrRanges in the array of AddrRanges starting at address `data`, + // starting at `offset` bytes into the first AddrRange and limited to the + // following `limit` bytes. (AddrRanges after `limit` are still iterated, + // but are truncated to a length of 0.) Invariants: data != 0; offset <= + // data[0].Length(); limit > 0; offset+limit <= the combined length of all + // AddrRanges in the array. + data unsafe.Pointer + length int + offset Addr + limit Addr +} + +// AddrRangeSeqOf returns an AddrRangeSeq representing the single AddrRange ar. +func AddrRangeSeqOf(ar AddrRange) AddrRangeSeq { + return AddrRangeSeq{ + length: 1, + offset: ar.Start, + limit: ar.Length(), + } +} + +// AddrRangeSeqFromSlice returns an AddrRangeSeq representing all AddrRanges in +// slice. +// +// Whether the returned AddrRangeSeq shares memory with slice is unspecified; +// clients should avoid mutating slices passed to AddrRangeSeqFromSlice. +// +// Preconditions: The combined length of all AddrRanges in slice <= +// math.MaxInt64. +func AddrRangeSeqFromSlice(slice []AddrRange) AddrRangeSeq { + var limit int64 + for _, ar := range slice { + len64 := int64(ar.Length()) + if len64 < 0 { + panic(fmt.Sprintf("Length of AddrRange %v overflows int64", ar)) + } + sum := limit + len64 + if sum < limit { + panic(fmt.Sprintf("Total length of AddrRanges %v overflows int64", slice)) + } + limit = sum + } + return addrRangeSeqFromSliceLimited(slice, limit) +} + +// Preconditions: The combined length of all AddrRanges in slice <= limit. +// limit >= 0. If len(slice) != 0, then limit > 0. +func addrRangeSeqFromSliceLimited(slice []AddrRange, limit int64) AddrRangeSeq { + switch len(slice) { + case 0: + return AddrRangeSeq{} + case 1: + return AddrRangeSeq{ + length: 1, + offset: slice[0].Start, + limit: Addr(limit), + } + default: + return AddrRangeSeq{ + data: unsafe.Pointer(&slice[0]), + length: len(slice), + limit: Addr(limit), + } + } +} + +// IsEmpty returns true if ars.NumRanges() == 0. +// +// Note that since AddrRangeSeq may contain AddrRanges with a length of zero, +// an AddrRange representing 0 bytes (AddrRangeSeq.NumBytes() == 0) is not +// necessarily empty. +func (ars AddrRangeSeq) IsEmpty() bool { + return ars.length == 0 +} + +// NumRanges returns the number of AddrRanges in ars. +func (ars AddrRangeSeq) NumRanges() int { + return ars.length +} + +// NumBytes returns the number of bytes represented by ars. +func (ars AddrRangeSeq) NumBytes() int64 { + return int64(ars.limit) +} + +// Head returns the first AddrRange in ars. +// +// Preconditions: !ars.IsEmpty(). +func (ars AddrRangeSeq) Head() AddrRange { + if ars.length == 0 { + panic("empty AddrRangeSeq") + } + if ars.length == 1 { + return AddrRange{ars.offset, ars.offset + ars.limit} + } + ar := *(*AddrRange)(ars.data) + ar.Start += ars.offset + if ar.Length() > ars.limit { + ar.End = ar.Start + ars.limit + } + return ar +} + +// Tail returns an AddrRangeSeq consisting of all AddrRanges in ars after the +// first. +// +// Preconditions: !ars.IsEmpty(). +func (ars AddrRangeSeq) Tail() AddrRangeSeq { + if ars.length == 0 { + panic("empty AddrRangeSeq") + } + if ars.length == 1 { + return AddrRangeSeq{} + } + return ars.externalTail() +} + +// Preconditions: ars.length >= 2. +func (ars AddrRangeSeq) externalTail() AddrRangeSeq { + headLen := (*AddrRange)(ars.data).Length() - ars.offset + var tailLimit int64 + if ars.limit > headLen { + tailLimit = int64(ars.limit - headLen) + } + var extSlice []AddrRange + extSliceHdr := (*reflect.SliceHeader)(unsafe.Pointer(&extSlice)) + extSliceHdr.Data = uintptr(ars.data) + extSliceHdr.Len = ars.length + extSliceHdr.Cap = ars.length + return addrRangeSeqFromSliceLimited(extSlice[1:], tailLimit) +} + +// DropFirst returns an AddrRangeSeq equivalent to ars, but with the first n +// bytes omitted. If n > ars.NumBytes(), DropFirst returns an empty +// AddrRangeSeq. +// +// If !ars.IsEmpty() and ars.Head().Length() == 0, DropFirst will always omit +// at least ars.Head(), even if n == 0. This guarantees that the basic pattern +// of: +// +// for !ars.IsEmpty() { +// n, err = doIOWith(ars.Head()) +// if err != nil { +// return err +// } +// ars = ars.DropFirst(n) +// } +// +// works even in the presence of zero-length AddrRanges. +// +// Preconditions: n >= 0. +func (ars AddrRangeSeq) DropFirst(n int) AddrRangeSeq { + if n < 0 { + panic(fmt.Sprintf("invalid n: %d", n)) + } + return ars.DropFirst64(int64(n)) +} + +// DropFirst64 is equivalent to DropFirst but takes an int64. +func (ars AddrRangeSeq) DropFirst64(n int64) AddrRangeSeq { + if n < 0 { + panic(fmt.Sprintf("invalid n: %d", n)) + } + if Addr(n) > ars.limit { + return AddrRangeSeq{} + } + // Handle initial empty AddrRange. + switch ars.length { + case 0: + return AddrRangeSeq{} + case 1: + if ars.limit == 0 { + return AddrRangeSeq{} + } + default: + if rawHeadLen := (*AddrRange)(ars.data).Length(); ars.offset == rawHeadLen { + ars = ars.externalTail() + } + } + for n != 0 { + // Calling ars.Head() here is surprisingly expensive, so inline getting + // the head's length. + var headLen Addr + if ars.length == 1 { + headLen = ars.limit + } else { + headLen = (*AddrRange)(ars.data).Length() - ars.offset + } + if Addr(n) < headLen { + // Dropping ends partway through the head AddrRange. + ars.offset += Addr(n) + ars.limit -= Addr(n) + return ars + } + n -= int64(headLen) + ars = ars.Tail() + } + return ars +} + +// TakeFirst returns an AddrRangeSeq equivalent to ars, but iterating at most n +// bytes. TakeFirst never removes AddrRanges from ars; AddrRanges beyond the +// first n bytes are reduced to a length of zero, but will still be iterated. +// +// Preconditions: n >= 0. +func (ars AddrRangeSeq) TakeFirst(n int) AddrRangeSeq { + if n < 0 { + panic(fmt.Sprintf("invalid n: %d", n)) + } + return ars.TakeFirst64(int64(n)) +} + +// TakeFirst64 is equivalent to TakeFirst but takes an int64. +func (ars AddrRangeSeq) TakeFirst64(n int64) AddrRangeSeq { + if n < 0 { + panic(fmt.Sprintf("invalid n: %d", n)) + } + if ars.limit > Addr(n) { + ars.limit = Addr(n) + } + return ars +} + +// String implements fmt.Stringer.String. +func (ars AddrRangeSeq) String() string { + // This is deliberately chosen to be the same as fmt's automatic stringer + // for []AddrRange. + var buf bytes.Buffer + buf.WriteByte('[') + var sep string + for !ars.IsEmpty() { + buf.WriteString(sep) + sep = " " + buf.WriteString(ars.Head().String()) + ars = ars.Tail() + } + buf.WriteByte(']') + return buf.String() +} diff --git a/pkg/usermem/bytes_io.go b/pkg/usermem/bytes_io.go new file mode 100644 index 000000000..e177d30eb --- /dev/null +++ b/pkg/usermem/bytes_io.go @@ -0,0 +1,141 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package usermem + +import ( + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/safemem" + "gvisor.dev/gvisor/pkg/syserror" +) + +const maxInt = int(^uint(0) >> 1) + +// BytesIO implements IO using a byte slice. Addresses are interpreted as +// offsets into the slice. Reads and writes beyond the end of the slice return +// EFAULT. +type BytesIO struct { + Bytes []byte +} + +// CopyOut implements IO.CopyOut. +func (b *BytesIO) CopyOut(ctx context.Context, addr Addr, src []byte, opts IOOpts) (int, error) { + rngN, rngErr := b.rangeCheck(addr, len(src)) + if rngN == 0 { + return 0, rngErr + } + return copy(b.Bytes[int(addr):], src[:rngN]), rngErr +} + +// CopyIn implements IO.CopyIn. +func (b *BytesIO) CopyIn(ctx context.Context, addr Addr, dst []byte, opts IOOpts) (int, error) { + rngN, rngErr := b.rangeCheck(addr, len(dst)) + if rngN == 0 { + return 0, rngErr + } + return copy(dst[:rngN], b.Bytes[int(addr):]), rngErr +} + +// ZeroOut implements IO.ZeroOut. +func (b *BytesIO) ZeroOut(ctx context.Context, addr Addr, toZero int64, opts IOOpts) (int64, error) { + if toZero > int64(maxInt) { + return 0, syserror.EINVAL + } + rngN, rngErr := b.rangeCheck(addr, int(toZero)) + if rngN == 0 { + return 0, rngErr + } + zeroSlice := b.Bytes[int(addr) : int(addr)+rngN] + for i := range zeroSlice { + zeroSlice[i] = 0 + } + return int64(rngN), rngErr +} + +// CopyOutFrom implements IO.CopyOutFrom. +func (b *BytesIO) CopyOutFrom(ctx context.Context, ars AddrRangeSeq, src safemem.Reader, opts IOOpts) (int64, error) { + dsts, rngErr := b.blocksFromAddrRanges(ars) + n, err := src.ReadToBlocks(dsts) + if err != nil { + return int64(n), err + } + return int64(n), rngErr +} + +// CopyInTo implements IO.CopyInTo. +func (b *BytesIO) CopyInTo(ctx context.Context, ars AddrRangeSeq, dst safemem.Writer, opts IOOpts) (int64, error) { + srcs, rngErr := b.blocksFromAddrRanges(ars) + n, err := dst.WriteFromBlocks(srcs) + if err != nil { + return int64(n), err + } + return int64(n), rngErr +} + +func (b *BytesIO) rangeCheck(addr Addr, length int) (int, error) { + if length == 0 { + return 0, nil + } + if length < 0 { + return 0, syserror.EINVAL + } + max := Addr(len(b.Bytes)) + if addr >= max { + return 0, syserror.EFAULT + } + end, ok := addr.AddLength(uint64(length)) + if !ok || end > max { + return int(max - addr), syserror.EFAULT + } + return length, nil +} + +func (b *BytesIO) blocksFromAddrRanges(ars AddrRangeSeq) (safemem.BlockSeq, error) { + switch ars.NumRanges() { + case 0: + return safemem.BlockSeq{}, nil + case 1: + block, err := b.blockFromAddrRange(ars.Head()) + return safemem.BlockSeqOf(block), err + default: + blocks := make([]safemem.Block, 0, ars.NumRanges()) + for !ars.IsEmpty() { + block, err := b.blockFromAddrRange(ars.Head()) + if block.Len() != 0 { + blocks = append(blocks, block) + } + if err != nil { + return safemem.BlockSeqFromSlice(blocks), err + } + ars = ars.Tail() + } + return safemem.BlockSeqFromSlice(blocks), nil + } +} + +func (b *BytesIO) blockFromAddrRange(ar AddrRange) (safemem.Block, error) { + n, err := b.rangeCheck(ar.Start, int(ar.Length())) + if n == 0 { + return safemem.Block{}, err + } + return safemem.BlockFromSafeSlice(b.Bytes[int(ar.Start) : int(ar.Start)+n]), err +} + +// BytesIOSequence returns an IOSequence representing the given byte slice. +func BytesIOSequence(buf []byte) IOSequence { + return IOSequence{ + IO: &BytesIO{buf}, + Addrs: AddrRangeSeqOf(AddrRange{0, Addr(len(buf))}), + } +} diff --git a/pkg/usermem/bytes_io_unsafe.go b/pkg/usermem/bytes_io_unsafe.go new file mode 100644 index 000000000..20de5037d --- /dev/null +++ b/pkg/usermem/bytes_io_unsafe.go @@ -0,0 +1,47 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package usermem + +import ( + "sync/atomic" + "unsafe" + + "gvisor.dev/gvisor/pkg/atomicbitops" + "gvisor.dev/gvisor/pkg/context" +) + +// SwapUint32 implements IO.SwapUint32. +func (b *BytesIO) SwapUint32(ctx context.Context, addr Addr, new uint32, opts IOOpts) (uint32, error) { + if _, rngErr := b.rangeCheck(addr, 4); rngErr != nil { + return 0, rngErr + } + return atomic.SwapUint32((*uint32)(unsafe.Pointer(&b.Bytes[int(addr)])), new), nil +} + +// CompareAndSwapUint32 implements IO.CompareAndSwapUint32. +func (b *BytesIO) CompareAndSwapUint32(ctx context.Context, addr Addr, old, new uint32, opts IOOpts) (uint32, error) { + if _, rngErr := b.rangeCheck(addr, 4); rngErr != nil { + return 0, rngErr + } + return atomicbitops.CompareAndSwapUint32((*uint32)(unsafe.Pointer(&b.Bytes[int(addr)])), old, new), nil +} + +// LoadUint32 implements IO.LoadUint32. +func (b *BytesIO) LoadUint32(ctx context.Context, addr Addr, opts IOOpts) (uint32, error) { + if _, err := b.rangeCheck(addr, 4); err != nil { + return 0, err + } + return atomic.LoadUint32((*uint32)(unsafe.Pointer(&b.Bytes[int(addr)]))), nil +} diff --git a/pkg/usermem/usermem.go b/pkg/usermem/usermem.go new file mode 100644 index 000000000..71fd4e155 --- /dev/null +++ b/pkg/usermem/usermem.go @@ -0,0 +1,597 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package usermem governs access to user memory. +package usermem + +import ( + "bytes" + "errors" + "io" + "strconv" + + "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/safemem" + "gvisor.dev/gvisor/pkg/syserror" +) + +// IO provides access to the contents of a virtual memory space. +// +// FIXME(b/38173783): Implementations of IO cannot expect ctx to contain any +// meaningful data. +type IO interface { + // CopyOut copies len(src) bytes from src to the memory mapped at addr. It + // returns the number of bytes copied. If the number of bytes copied is < + // len(src), it returns a non-nil error explaining why. + // + // Preconditions: The caller must not hold mm.MemoryManager.mappingMu or + // any following locks in the lock order. + // + // Postconditions: CopyOut does not retain src. + CopyOut(ctx context.Context, addr Addr, src []byte, opts IOOpts) (int, error) + + // CopyIn copies len(dst) bytes from the memory mapped at addr to dst. + // It returns the number of bytes copied. If the number of bytes copied is + // < len(dst), it returns a non-nil error explaining why. + // + // Preconditions: The caller must not hold mm.MemoryManager.mappingMu or + // any following locks in the lock order. + // + // Postconditions: CopyIn does not retain dst. + CopyIn(ctx context.Context, addr Addr, dst []byte, opts IOOpts) (int, error) + + // ZeroOut sets toZero bytes to 0, starting at addr. It returns the number + // of bytes zeroed. If the number of bytes zeroed is < toZero, it returns a + // non-nil error explaining why. + // + // Preconditions: The caller must not hold mm.MemoryManager.mappingMu or + // any following locks in the lock order. toZero >= 0. + ZeroOut(ctx context.Context, addr Addr, toZero int64, opts IOOpts) (int64, error) + + // CopyOutFrom copies ars.NumBytes() bytes from src to the memory mapped at + // ars. It returns the number of bytes copied, which may be less than the + // number of bytes read from src if copying fails. CopyOutFrom may return a + // partial copy without an error iff src.ReadToBlocks returns a partial + // read without an error. + // + // CopyOutFrom calls src.ReadToBlocks at most once. + // + // Preconditions: The caller must not hold mm.MemoryManager.mappingMu or + // any following locks in the lock order. src.ReadToBlocks must not block + // on mm.MemoryManager.activeMu or any preceding locks in the lock order. + CopyOutFrom(ctx context.Context, ars AddrRangeSeq, src safemem.Reader, opts IOOpts) (int64, error) + + // CopyInTo copies ars.NumBytes() bytes from the memory mapped at ars to + // dst. It returns the number of bytes copied. CopyInTo may return a + // partial copy without an error iff dst.WriteFromBlocks returns a partial + // write without an error. + // + // CopyInTo calls dst.WriteFromBlocks at most once. + // + // Preconditions: The caller must not hold mm.MemoryManager.mappingMu or + // any following locks in the lock order. dst.WriteFromBlocks must not + // block on mm.MemoryManager.activeMu or any preceding locks in the lock + // order. + CopyInTo(ctx context.Context, ars AddrRangeSeq, dst safemem.Writer, opts IOOpts) (int64, error) + + // TODO(jamieliu): The requirement that CopyOutFrom/CopyInTo call src/dst + // at most once, which is unnecessary in most cases, forces implementations + // to gather safemem.Blocks into a single slice to pass to src/dst. Add + // CopyOutFromIter/CopyInToIter, which relaxes this restriction, to avoid + // this allocation. + + // SwapUint32 atomically sets the uint32 value at addr to new and + // returns the previous value. + // + // Preconditions: The caller must not hold mm.MemoryManager.mappingMu or + // any following locks in the lock order. addr must be aligned to a 4-byte + // boundary. + SwapUint32(ctx context.Context, addr Addr, new uint32, opts IOOpts) (uint32, error) + + // CompareAndSwapUint32 atomically compares the uint32 value at addr to + // old; if they are equal, the value in memory is replaced by new. In + // either case, the previous value stored in memory is returned. + // + // Preconditions: The caller must not hold mm.MemoryManager.mappingMu or + // any following locks in the lock order. addr must be aligned to a 4-byte + // boundary. + CompareAndSwapUint32(ctx context.Context, addr Addr, old, new uint32, opts IOOpts) (uint32, error) + + // LoadUint32 atomically loads the uint32 value at addr and returns it. + // + // Preconditions: The caller must not hold mm.MemoryManager.mappingMu or + // any following locks in the lock order. addr must be aligned to a 4-byte + // boundary. + LoadUint32(ctx context.Context, addr Addr, opts IOOpts) (uint32, error) +} + +// IOOpts contains options applicable to all IO methods. +type IOOpts struct { + // If IgnorePermissions is true, application-defined memory protections set + // by mmap(2) or mprotect(2) will be ignored. (Memory protections required + // by the target of the mapping are never ignored.) + IgnorePermissions bool + + // If AddressSpaceActive is true, the IO implementation may assume that it + // has an active AddressSpace and can therefore use AddressSpace copying + // without performing activation. See mm/io.go for details. + AddressSpaceActive bool +} + +// IOReadWriter is an io.ReadWriter that reads from / writes to addresses +// starting at addr in IO. The preconditions that apply to IO.CopyIn and +// IO.CopyOut also apply to IOReadWriter.Read and IOReadWriter.Write +// respectively. +type IOReadWriter struct { + Ctx context.Context + IO IO + Addr Addr + Opts IOOpts +} + +// Read implements io.Reader.Read. +// +// Note that an address space does not have an "end of file", so Read can only +// return io.EOF if IO.CopyIn returns io.EOF. Attempts to read unmapped or +// unreadable memory, or beyond the end of the address space, should return +// EFAULT. +func (rw *IOReadWriter) Read(dst []byte) (int, error) { + n, err := rw.IO.CopyIn(rw.Ctx, rw.Addr, dst, rw.Opts) + end, ok := rw.Addr.AddLength(uint64(n)) + if ok { + rw.Addr = end + } else { + // Disallow wraparound. + rw.Addr = ^Addr(0) + if err != nil { + err = syserror.EFAULT + } + } + return n, err +} + +// Writer implements io.Writer.Write. +func (rw *IOReadWriter) Write(src []byte) (int, error) { + n, err := rw.IO.CopyOut(rw.Ctx, rw.Addr, src, rw.Opts) + end, ok := rw.Addr.AddLength(uint64(n)) + if ok { + rw.Addr = end + } else { + // Disallow wraparound. + rw.Addr = ^Addr(0) + if err != nil { + err = syserror.EFAULT + } + } + return n, err +} + +// CopyObjectOut copies a fixed-size value or slice of fixed-size values from +// src to the memory mapped at addr in uio. It returns the number of bytes +// copied. +// +// CopyObjectOut must use reflection to encode src; performance-sensitive +// clients should do encoding manually and use uio.CopyOut directly. +// +// Preconditions: As for IO.CopyOut. +func CopyObjectOut(ctx context.Context, uio IO, addr Addr, src interface{}, opts IOOpts) (int, error) { + w := &IOReadWriter{ + Ctx: ctx, + IO: uio, + Addr: addr, + Opts: opts, + } + // Allocate a byte slice the size of the object being marshaled. This + // adds an extra reflection call, but avoids needing to grow the slice + // during encoding, which can result in many heap-allocated slices. + b := make([]byte, 0, binary.Size(src)) + return w.Write(binary.Marshal(b, ByteOrder, src)) +} + +// CopyObjectIn copies a fixed-size value or slice of fixed-size values from +// the memory mapped at addr in uio to dst. It returns the number of bytes +// copied. +// +// CopyObjectIn must use reflection to decode dst; performance-sensitive +// clients should use uio.CopyIn directly and do decoding manually. +// +// Preconditions: As for IO.CopyIn. +func CopyObjectIn(ctx context.Context, uio IO, addr Addr, dst interface{}, opts IOOpts) (int, error) { + r := &IOReadWriter{ + Ctx: ctx, + IO: uio, + Addr: addr, + Opts: opts, + } + buf := make([]byte, binary.Size(dst)) + if _, err := io.ReadFull(r, buf); err != nil { + return 0, err + } + binary.Unmarshal(buf, ByteOrder, dst) + return int(r.Addr - addr), nil +} + +// CopyStringIn tuning parameters, defined outside that function for tests. +const ( + copyStringIncrement = 64 + copyStringMaxInitBufLen = 256 +) + +// CopyStringIn copies a NUL-terminated string of unknown length from the +// memory mapped at addr in uio and returns it as a string (not including the +// trailing NUL). If the length of the string, including the terminating NUL, +// would exceed maxlen, CopyStringIn returns the string truncated to maxlen and +// ENAMETOOLONG. +// +// Preconditions: As for IO.CopyFromUser. maxlen >= 0. +func CopyStringIn(ctx context.Context, uio IO, addr Addr, maxlen int, opts IOOpts) (string, error) { + initLen := maxlen + if initLen > copyStringMaxInitBufLen { + initLen = copyStringMaxInitBufLen + } + buf := make([]byte, initLen) + var done int + for done < maxlen { + // Read up to copyStringIncrement bytes at a time. + readlen := copyStringIncrement + if readlen > maxlen-done { + readlen = maxlen - done + } + end, ok := addr.AddLength(uint64(readlen)) + if !ok { + return stringFromImmutableBytes(buf[:done]), syserror.EFAULT + } + // Shorten the read to avoid crossing page boundaries, since faulting + // in a page unnecessarily is expensive. This also ensures that partial + // copies up to the end of application-mappable memory succeed. + if addr.RoundDown() != end.RoundDown() { + end = end.RoundDown() + readlen = int(end - addr) + } + // Ensure that our buffer is large enough to accommodate the read. + if done+readlen > len(buf) { + newBufLen := len(buf) * 2 + if newBufLen > maxlen { + newBufLen = maxlen + } + buf = append(buf, make([]byte, newBufLen-len(buf))...) + } + n, err := uio.CopyIn(ctx, addr, buf[done:done+readlen], opts) + // Look for the terminating zero byte, which may have occurred before + // hitting err. + if i := bytes.IndexByte(buf[done:done+n], byte(0)); i >= 0 { + return stringFromImmutableBytes(buf[:done+i]), nil + } + + done += n + if err != nil { + return stringFromImmutableBytes(buf[:done]), err + } + addr = end + } + return stringFromImmutableBytes(buf), syserror.ENAMETOOLONG +} + +// CopyOutVec copies bytes from src to the memory mapped at ars in uio. The +// maximum number of bytes copied is ars.NumBytes() or len(src), whichever is +// less. CopyOutVec returns the number of bytes copied; if this is less than +// the maximum, it returns a non-nil error explaining why. +// +// Preconditions: As for IO.CopyOut. +func CopyOutVec(ctx context.Context, uio IO, ars AddrRangeSeq, src []byte, opts IOOpts) (int, error) { + var done int + for !ars.IsEmpty() && done < len(src) { + ar := ars.Head() + cplen := len(src) - done + if Addr(cplen) >= ar.Length() { + cplen = int(ar.Length()) + } + n, err := uio.CopyOut(ctx, ar.Start, src[done:done+cplen], opts) + done += n + if err != nil { + return done, err + } + ars = ars.DropFirst(n) + } + return done, nil +} + +// CopyInVec copies bytes from the memory mapped at ars in uio to dst. The +// maximum number of bytes copied is ars.NumBytes() or len(dst), whichever is +// less. CopyInVec returns the number of bytes copied; if this is less than the +// maximum, it returns a non-nil error explaining why. +// +// Preconditions: As for IO.CopyIn. +func CopyInVec(ctx context.Context, uio IO, ars AddrRangeSeq, dst []byte, opts IOOpts) (int, error) { + var done int + for !ars.IsEmpty() && done < len(dst) { + ar := ars.Head() + cplen := len(dst) - done + if Addr(cplen) >= ar.Length() { + cplen = int(ar.Length()) + } + n, err := uio.CopyIn(ctx, ar.Start, dst[done:done+cplen], opts) + done += n + if err != nil { + return done, err + } + ars = ars.DropFirst(n) + } + return done, nil +} + +// ZeroOutVec writes zeroes to the memory mapped at ars in uio. The maximum +// number of bytes written is ars.NumBytes() or toZero, whichever is less. +// ZeroOutVec returns the number of bytes written; if this is less than the +// maximum, it returns a non-nil error explaining why. +// +// Preconditions: As for IO.ZeroOut. +func ZeroOutVec(ctx context.Context, uio IO, ars AddrRangeSeq, toZero int64, opts IOOpts) (int64, error) { + var done int64 + for !ars.IsEmpty() && done < toZero { + ar := ars.Head() + cplen := toZero - done + if Addr(cplen) >= ar.Length() { + cplen = int64(ar.Length()) + } + n, err := uio.ZeroOut(ctx, ar.Start, cplen, opts) + done += n + if err != nil { + return done, err + } + ars = ars.DropFirst64(n) + } + return done, nil +} + +func isASCIIWhitespace(b byte) bool { + // Compare Linux include/linux/ctype.h, lib/ctype.c. + // 9 => horizontal tab '\t' + // 10 => line feed '\n' + // 11 => vertical tab '\v' + // 12 => form feed '\c' + // 13 => carriage return '\r' + return b == ' ' || (b >= 9 && b <= 13) +} + +// CopyInt32StringsInVec copies up to len(dsts) whitespace-separated decimal +// strings from the memory mapped at ars in uio and converts them to int32 +// values in dsts. It returns the number of bytes read. +// +// CopyInt32StringsInVec shares the following properties with Linux's +// kernel/sysctl.c:proc_dointvec(write=1): +// +// - If any read value overflows the range of int32, or any invalid characters +// are encountered during the read, CopyInt32StringsInVec returns EINVAL. +// +// - If, upon reaching the end of ars, fewer than len(dsts) values have been +// read, CopyInt32StringsInVec returns no error if at least 1 value was read +// and EINVAL otherwise. +// +// - Trailing whitespace after the last successfully read value is counted in +// the number of bytes read. +// +// Unlike proc_dointvec(): +// +// - CopyInt32StringsInVec does not implicitly limit ars.NumBytes() to +// PageSize-1; callers that require this must do so explicitly. +// +// - CopyInt32StringsInVec returns EINVAL if ars.NumBytes() == 0. +// +// Preconditions: As for CopyInVec. +func CopyInt32StringsInVec(ctx context.Context, uio IO, ars AddrRangeSeq, dsts []int32, opts IOOpts) (int64, error) { + if len(dsts) == 0 { + return 0, nil + } + + buf := make([]byte, ars.NumBytes()) + n, cperr := CopyInVec(ctx, uio, ars, buf, opts) + buf = buf[:n] + + var i, j int + for ; j < len(dsts); j++ { + // Skip leading whitespace. + for i < len(buf) && isASCIIWhitespace(buf[i]) { + i++ + } + if i == len(buf) { + break + } + + // Find the end of the value to be parsed (next whitespace or end of string). + nextI := i + 1 + for nextI < len(buf) && !isASCIIWhitespace(buf[nextI]) { + nextI++ + } + + // Parse a single value. + val, err := strconv.ParseInt(string(buf[i:nextI]), 10, 32) + if err != nil { + return int64(i), syserror.EINVAL + } + dsts[j] = int32(val) + + i = nextI + } + + // Skip trailing whitespace. + for i < len(buf) && isASCIIWhitespace(buf[i]) { + i++ + } + + if cperr != nil { + return int64(i), cperr + } + if j == 0 { + return int64(i), syserror.EINVAL + } + return int64(i), nil +} + +// CopyInt32StringInVec is equivalent to CopyInt32StringsInVec, but copies at +// most one int32. +func CopyInt32StringInVec(ctx context.Context, uio IO, ars AddrRangeSeq, dst *int32, opts IOOpts) (int64, error) { + dsts := [1]int32{*dst} + n, err := CopyInt32StringsInVec(ctx, uio, ars, dsts[:], opts) + *dst = dsts[0] + return n, err +} + +// IOSequence holds arguments to IO methods. +type IOSequence struct { + IO IO + Addrs AddrRangeSeq + Opts IOOpts +} + +// NumBytes returns s.Addrs.NumBytes(). +// +// Note that NumBytes() may return 0 even if !s.Addrs.IsEmpty(), since +// s.Addrs may contain a non-zero number of zero-length AddrRanges. +// Many clients of +// IOSequence currently do something like: +// +// if ioseq.NumBytes() == 0 { +// return 0, nil +// } +// if f.availableBytes == 0 { +// return 0, syserror.ErrWouldBlock +// } +// return ioseq.CopyOutFrom(..., reader) +// +// In such cases, using s.Addrs.IsEmpty() will cause them to have the wrong +// behavior for zero-length I/O. However, using s.NumBytes() == 0 instead means +// that we will return success for zero-length I/O in cases where Linux would +// return EFAULT due to a failed access_ok() check, so in the long term we +// should move checks for ErrWouldBlock etc. into the body of +// reader.ReadToBlocks and use s.Addrs.IsEmpty() instead. +func (s IOSequence) NumBytes() int64 { + return s.Addrs.NumBytes() +} + +// DropFirst returns a copy of s with s.Addrs.DropFirst(n). +// +// Preconditions: As for AddrRangeSeq.DropFirst. +func (s IOSequence) DropFirst(n int) IOSequence { + return IOSequence{s.IO, s.Addrs.DropFirst(n), s.Opts} +} + +// DropFirst64 returns a copy of s with s.Addrs.DropFirst64(n). +// +// Preconditions: As for AddrRangeSeq.DropFirst64. +func (s IOSequence) DropFirst64(n int64) IOSequence { + return IOSequence{s.IO, s.Addrs.DropFirst64(n), s.Opts} +} + +// TakeFirst returns a copy of s with s.Addrs.TakeFirst(n). +// +// Preconditions: As for AddrRangeSeq.TakeFirst. +func (s IOSequence) TakeFirst(n int) IOSequence { + return IOSequence{s.IO, s.Addrs.TakeFirst(n), s.Opts} +} + +// TakeFirst64 returns a copy of s with s.Addrs.TakeFirst64(n). +// +// Preconditions: As for AddrRangeSeq.TakeFirst64. +func (s IOSequence) TakeFirst64(n int64) IOSequence { + return IOSequence{s.IO, s.Addrs.TakeFirst64(n), s.Opts} +} + +// CopyOut invokes CopyOutVec over s.Addrs. +// +// As with CopyOutVec, if s.NumBytes() < len(src), the copy will be truncated +// to s.NumBytes(), and a nil error will be returned. +// +// Preconditions: As for CopyOutVec. +func (s IOSequence) CopyOut(ctx context.Context, src []byte) (int, error) { + return CopyOutVec(ctx, s.IO, s.Addrs, src, s.Opts) +} + +// CopyIn invokes CopyInVec over s.Addrs. +// +// As with CopyInVec, if s.NumBytes() < len(dst), the copy will be truncated to +// s.NumBytes(), and a nil error will be returned. +// +// Preconditions: As for CopyInVec. +func (s IOSequence) CopyIn(ctx context.Context, dst []byte) (int, error) { + return CopyInVec(ctx, s.IO, s.Addrs, dst, s.Opts) +} + +// ZeroOut invokes ZeroOutVec over s.Addrs. +// +// As with ZeroOutVec, if s.NumBytes() < toZero, the write will be truncated +// to s.NumBytes(), and a nil error will be returned. +// +// Preconditions: As for ZeroOutVec. +func (s IOSequence) ZeroOut(ctx context.Context, toZero int64) (int64, error) { + return ZeroOutVec(ctx, s.IO, s.Addrs, toZero, s.Opts) +} + +// CopyOutFrom invokes s.CopyOutFrom over s.Addrs. +// +// Preconditions: As for IO.CopyOutFrom. +func (s IOSequence) CopyOutFrom(ctx context.Context, src safemem.Reader) (int64, error) { + return s.IO.CopyOutFrom(ctx, s.Addrs, src, s.Opts) +} + +// CopyInTo invokes s.CopyInTo over s.Addrs. +// +// Preconditions: As for IO.CopyInTo. +func (s IOSequence) CopyInTo(ctx context.Context, dst safemem.Writer) (int64, error) { + return s.IO.CopyInTo(ctx, s.Addrs, dst, s.Opts) +} + +// Reader returns an io.Reader that reads from s. Reads beyond the end of s +// return io.EOF. The preconditions that apply to s.CopyIn also apply to the +// returned io.Reader.Read. +func (s IOSequence) Reader(ctx context.Context) io.Reader { + return &ioSequenceReadWriter{ctx, s} +} + +// Writer returns an io.Writer that writes to s. Writes beyond the end of s +// return ErrEndOfIOSequence. The preconditions that apply to s.CopyOut also +// apply to the returned io.Writer.Write. +func (s IOSequence) Writer(ctx context.Context) io.Writer { + return &ioSequenceReadWriter{ctx, s} +} + +// ErrEndOfIOSequence is returned by IOSequence.Writer().Write() when +// attempting to write beyond the end of the IOSequence. +var ErrEndOfIOSequence = errors.New("write beyond end of IOSequence") + +type ioSequenceReadWriter struct { + ctx context.Context + s IOSequence +} + +// Read implements io.Reader.Read. +func (rw *ioSequenceReadWriter) Read(dst []byte) (int, error) { + n, err := rw.s.CopyIn(rw.ctx, dst) + rw.s = rw.s.DropFirst(n) + if err == nil && rw.s.NumBytes() == 0 { + err = io.EOF + } + return n, err +} + +// Write implements io.Writer.Write. +func (rw *ioSequenceReadWriter) Write(src []byte) (int, error) { + n, err := rw.s.CopyOut(rw.ctx, src) + rw.s = rw.s.DropFirst(n) + if err == nil && n < len(src) { + err = ErrEndOfIOSequence + } + return n, err +} diff --git a/pkg/usermem/usermem_arm64.go b/pkg/usermem/usermem_arm64.go new file mode 100644 index 000000000..fdfc30a66 --- /dev/null +++ b/pkg/usermem/usermem_arm64.go @@ -0,0 +1,53 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package usermem + +import ( + "encoding/binary" + "syscall" +) + +const ( + // PageSize is the system page size. + // arm64 support 4K/16K/64K page size, + // which can be get by syscall.Getpagesize(). + // Currently, only 4K page size is supported. + PageSize = 1 << PageShift + + // HugePageSize is the system huge page size. + HugePageSize = 1 << HugePageShift + + // PageShift is the binary log of the system page size. + PageShift = 12 + + // HugePageShift is the binary log of the system huge page size. + // Should be calculated by "PageShift + (PageShift - 3)" + // when multiple page size support is ready. + HugePageShift = 21 +) + +var ( + // ByteOrder is the native byte order (little endian). + ByteOrder = binary.LittleEndian +) + +func init() { + // Make sure the page size is 4K on arm64 platform. + if size := syscall.Getpagesize(); size != PageSize { + panic("Only 4K page size is supported on arm64!") + } +} diff --git a/pkg/usermem/usermem_test.go b/pkg/usermem/usermem_test.go new file mode 100644 index 000000000..bf3c5df2b --- /dev/null +++ b/pkg/usermem/usermem_test.go @@ -0,0 +1,424 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package usermem + +import ( + "bytes" + "encoding/binary" + "fmt" + "reflect" + "strings" + "testing" + + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/safemem" + "gvisor.dev/gvisor/pkg/syserror" +) + +// newContext returns a context.Context that we can use in these tests (we +// can't use contexttest because it depends on usermem). +func newContext() context.Context { + return context.Background() +} + +func newBytesIOString(s string) *BytesIO { + return &BytesIO{[]byte(s)} +} + +func TestBytesIOCopyOutSuccess(t *testing.T) { + b := newBytesIOString("ABCDE") + n, err := b.CopyOut(newContext(), 1, []byte("foo"), IOOpts{}) + if wantN := 3; n != wantN || err != nil { + t.Errorf("CopyOut: got (%v, %v), wanted (%v, nil)", n, err, wantN) + } + if got, want := b.Bytes, []byte("AfooE"); !bytes.Equal(got, want) { + t.Errorf("Bytes: got %q, wanted %q", got, want) + } +} + +func TestBytesIOCopyOutFailure(t *testing.T) { + b := newBytesIOString("ABC") + n, err := b.CopyOut(newContext(), 1, []byte("foo"), IOOpts{}) + if wantN, wantErr := 2, syserror.EFAULT; n != wantN || err != wantErr { + t.Errorf("CopyOut: got (%v, %v), wanted (%v, %v)", n, err, wantN, wantErr) + } + if got, want := b.Bytes, []byte("Afo"); !bytes.Equal(got, want) { + t.Errorf("Bytes: got %q, wanted %q", got, want) + } +} + +func TestBytesIOCopyInSuccess(t *testing.T) { + b := newBytesIOString("AfooE") + var dst [3]byte + n, err := b.CopyIn(newContext(), 1, dst[:], IOOpts{}) + if wantN := 3; n != wantN || err != nil { + t.Errorf("CopyIn: got (%v, %v), wanted (%v, nil)", n, err, wantN) + } + if got, want := dst[:], []byte("foo"); !bytes.Equal(got, want) { + t.Errorf("dst: got %q, wanted %q", got, want) + } +} + +func TestBytesIOCopyInFailure(t *testing.T) { + b := newBytesIOString("Afo") + var dst [3]byte + n, err := b.CopyIn(newContext(), 1, dst[:], IOOpts{}) + if wantN, wantErr := 2, syserror.EFAULT; n != wantN || err != wantErr { + t.Errorf("CopyIn: got (%v, %v), wanted (%v, %v)", n, err, wantN, wantErr) + } + if got, want := dst[:], []byte("fo\x00"); !bytes.Equal(got, want) { + t.Errorf("dst: got %q, wanted %q", got, want) + } +} + +func TestBytesIOZeroOutSuccess(t *testing.T) { + b := newBytesIOString("ABCD") + n, err := b.ZeroOut(newContext(), 1, 2, IOOpts{}) + if wantN := int64(2); n != wantN || err != nil { + t.Errorf("ZeroOut: got (%v, %v), wanted (%v, nil)", n, err, wantN) + } + if got, want := b.Bytes, []byte("A\x00\x00D"); !bytes.Equal(got, want) { + t.Errorf("Bytes: got %q, wanted %q", got, want) + } +} + +func TestBytesIOZeroOutFailure(t *testing.T) { + b := newBytesIOString("ABC") + n, err := b.ZeroOut(newContext(), 1, 3, IOOpts{}) + if wantN, wantErr := int64(2), syserror.EFAULT; n != wantN || err != wantErr { + t.Errorf("ZeroOut: got (%v, %v), wanted (%v, %v)", n, err, wantN, wantErr) + } + if got, want := b.Bytes, []byte("A\x00\x00"); !bytes.Equal(got, want) { + t.Errorf("Bytes: got %q, wanted %q", got, want) + } +} + +func TestBytesIOCopyOutFromSuccess(t *testing.T) { + b := newBytesIOString("ABCDEFGH") + n, err := b.CopyOutFrom(newContext(), AddrRangeSeqFromSlice([]AddrRange{ + {Start: 4, End: 7}, + {Start: 1, End: 4}, + }), safemem.FromIOReader{bytes.NewBufferString("barfoo")}, IOOpts{}) + if wantN := int64(6); n != wantN || err != nil { + t.Errorf("CopyOutFrom: got (%v, %v), wanted (%v, nil)", n, err, wantN) + } + if got, want := b.Bytes, []byte("AfoobarH"); !bytes.Equal(got, want) { + t.Errorf("Bytes: got %q, wanted %q", got, want) + } +} + +func TestBytesIOCopyOutFromFailure(t *testing.T) { + b := newBytesIOString("ABCDE") + n, err := b.CopyOutFrom(newContext(), AddrRangeSeqFromSlice([]AddrRange{ + {Start: 1, End: 4}, + {Start: 4, End: 7}, + }), safemem.FromIOReader{bytes.NewBufferString("foobar")}, IOOpts{}) + if wantN, wantErr := int64(4), syserror.EFAULT; n != wantN || err != wantErr { + t.Errorf("CopyOutFrom: got (%v, %v), wanted (%v, %v)", n, err, wantN, wantErr) + } + if got, want := b.Bytes, []byte("Afoob"); !bytes.Equal(got, want) { + t.Errorf("Bytes: got %q, wanted %q", got, want) + } +} + +func TestBytesIOCopyInToSuccess(t *testing.T) { + b := newBytesIOString("AfoobarH") + var dst bytes.Buffer + n, err := b.CopyInTo(newContext(), AddrRangeSeqFromSlice([]AddrRange{ + {Start: 4, End: 7}, + {Start: 1, End: 4}, + }), safemem.FromIOWriter{&dst}, IOOpts{}) + if wantN := int64(6); n != wantN || err != nil { + t.Errorf("CopyInTo: got (%v, %v), wanted (%v, nil)", n, err, wantN) + } + if got, want := dst.Bytes(), []byte("barfoo"); !bytes.Equal(got, want) { + t.Errorf("dst.Bytes(): got %q, wanted %q", got, want) + } +} + +func TestBytesIOCopyInToFailure(t *testing.T) { + b := newBytesIOString("Afoob") + var dst bytes.Buffer + n, err := b.CopyInTo(newContext(), AddrRangeSeqFromSlice([]AddrRange{ + {Start: 1, End: 4}, + {Start: 4, End: 7}, + }), safemem.FromIOWriter{&dst}, IOOpts{}) + if wantN, wantErr := int64(4), syserror.EFAULT; n != wantN || err != wantErr { + t.Errorf("CopyOutFrom: got (%v, %v), wanted (%v, %v)", n, err, wantN, wantErr) + } + if got, want := dst.Bytes(), []byte("foob"); !bytes.Equal(got, want) { + t.Errorf("dst.Bytes(): got %q, wanted %q", got, want) + } +} + +type testStruct struct { + Int8 int8 + Uint8 uint8 + Int16 int16 + Uint16 uint16 + Int32 int32 + Uint32 uint32 + Int64 int64 + Uint64 uint64 +} + +func TestCopyObject(t *testing.T) { + wantObj := testStruct{1, 2, 3, 4, 5, 6, 7, 8} + wantN := binary.Size(wantObj) + b := &BytesIO{make([]byte, wantN)} + ctx := newContext() + if n, err := CopyObjectOut(ctx, b, 0, &wantObj, IOOpts{}); n != wantN || err != nil { + t.Fatalf("CopyObjectOut: got (%v, %v), wanted (%v, nil)", n, err, wantN) + } + var gotObj testStruct + if n, err := CopyObjectIn(ctx, b, 0, &gotObj, IOOpts{}); n != wantN || err != nil { + t.Errorf("CopyObjectIn: got (%v, %v), wanted (%v, nil)", n, err, wantN) + } + if gotObj != wantObj { + t.Errorf("CopyObject round trip: got %+v, wanted %+v", gotObj, wantObj) + } +} + +func TestCopyStringInShort(t *testing.T) { + // Tests for string length <= copyStringIncrement. + want := strings.Repeat("A", copyStringIncrement-2) + mem := want + "\x00" + if got, err := CopyStringIn(newContext(), newBytesIOString(mem), 0, 2*copyStringIncrement, IOOpts{}); got != want || err != nil { + t.Errorf("CopyStringIn: got (%q, %v), wanted (%q, nil)", got, err, want) + } +} + +func TestCopyStringInLong(t *testing.T) { + // Tests for copyStringIncrement < string length <= copyStringMaxInitBufLen + // (requiring multiple calls to IO.CopyIn()). + want := strings.Repeat("A", copyStringIncrement*3/4) + strings.Repeat("B", copyStringIncrement*3/4) + mem := want + "\x00" + if got, err := CopyStringIn(newContext(), newBytesIOString(mem), 0, 2*copyStringIncrement, IOOpts{}); got != want || err != nil { + t.Errorf("CopyStringIn: got (%q, %v), wanted (%q, nil)", got, err, want) + } +} + +func TestCopyStringInVeryLong(t *testing.T) { + // Tests for string length > copyStringMaxInitBufLen (requiring buffer + // reallocation). + want := strings.Repeat("A", copyStringMaxInitBufLen*3/4) + strings.Repeat("B", copyStringMaxInitBufLen*3/4) + mem := want + "\x00" + if got, err := CopyStringIn(newContext(), newBytesIOString(mem), 0, 2*copyStringMaxInitBufLen, IOOpts{}); got != want || err != nil { + t.Errorf("CopyStringIn: got (%q, %v), wanted (%q, nil)", got, err, want) + } +} + +func TestCopyStringInNoTerminatingZeroByte(t *testing.T) { + want := strings.Repeat("A", copyStringIncrement-1) + got, err := CopyStringIn(newContext(), newBytesIOString(want), 0, 2*copyStringIncrement, IOOpts{}) + if wantErr := syserror.EFAULT; got != want || err != wantErr { + t.Errorf("CopyStringIn: got (%q, %v), wanted (%q, %v)", got, err, want, wantErr) + } +} + +func TestCopyStringInTruncatedByMaxlen(t *testing.T) { + got, err := CopyStringIn(newContext(), newBytesIOString(strings.Repeat("A", 10)), 0, 5, IOOpts{}) + if want, wantErr := strings.Repeat("A", 5), syserror.ENAMETOOLONG; got != want || err != wantErr { + t.Errorf("CopyStringIn: got (%q, %v), wanted (%q, %v)", got, err, want, wantErr) + } +} + +func TestCopyInt32StringsInVec(t *testing.T) { + for _, test := range []struct { + str string + n int + initial []int32 + final []int32 + }{ + { + str: "100 200", + n: len("100 200"), + initial: []int32{1, 2}, + final: []int32{100, 200}, + }, + { + // Fewer values ok + str: "100", + n: len("100"), + initial: []int32{1, 2}, + final: []int32{100, 2}, + }, + { + // Extra values ok + str: "100 200 300", + n: len("100 200 "), + initial: []int32{1, 2}, + final: []int32{100, 200}, + }, + { + // Leading and trailing whitespace ok + str: " 100\t200\n", + n: len(" 100\t200\n"), + initial: []int32{1, 2}, + final: []int32{100, 200}, + }, + } { + t.Run(fmt.Sprintf("%q", test.str), func(t *testing.T) { + src := BytesIOSequence([]byte(test.str)) + dsts := append([]int32(nil), test.initial...) + if n, err := CopyInt32StringsInVec(newContext(), src.IO, src.Addrs, dsts, src.Opts); n != int64(test.n) || err != nil { + t.Errorf("CopyInt32StringsInVec: got (%d, %v), wanted (%d, nil)", n, err, test.n) + } + if !reflect.DeepEqual(dsts, test.final) { + t.Errorf("dsts: got %v, wanted %v", dsts, test.final) + } + }) + } +} + +func TestCopyInt32StringsInVecRequiresOneValidValue(t *testing.T) { + for _, s := range []string{"", "\n", "a123"} { + t.Run(fmt.Sprintf("%q", s), func(t *testing.T) { + src := BytesIOSequence([]byte(s)) + initial := []int32{1, 2} + dsts := append([]int32(nil), initial...) + if n, err := CopyInt32StringsInVec(newContext(), src.IO, src.Addrs, dsts, src.Opts); err != syserror.EINVAL { + t.Errorf("CopyInt32StringsInVec: got (%d, %v), wanted (_, %v)", n, err, syserror.EINVAL) + } + if !reflect.DeepEqual(dsts, initial) { + t.Errorf("dsts: got %v, wanted %v", dsts, initial) + } + }) + } +} + +func TestIOSequenceCopyOut(t *testing.T) { + buf := []byte("ABCD") + s := BytesIOSequence(buf) + + // CopyOut limited by len(src). + n, err := s.CopyOut(newContext(), []byte("fo")) + if wantN := 2; n != wantN || err != nil { + t.Errorf("CopyOut: got (%v, %v), wanted (%v, nil)", n, err, wantN) + } + if want := []byte("foCD"); !bytes.Equal(buf, want) { + t.Errorf("buf: got %q, wanted %q", buf, want) + } + s = s.DropFirst(2) + if got, want := s.NumBytes(), int64(2); got != want { + t.Errorf("NumBytes: got %v, wanted %v", got, want) + } + + // CopyOut limited by s.NumBytes(). + n, err = s.CopyOut(newContext(), []byte("obar")) + if wantN := 2; n != wantN || err != nil { + t.Errorf("CopyOut: got (%v, %v), wanted (%v, nil)", n, err, wantN) + } + if want := []byte("foob"); !bytes.Equal(buf, want) { + t.Errorf("buf: got %q, wanted %q", buf, want) + } + s = s.DropFirst(2) + if got, want := s.NumBytes(), int64(0); got != want { + t.Errorf("NumBytes: got %v, wanted %v", got, want) + } +} + +func TestIOSequenceCopyIn(t *testing.T) { + s := BytesIOSequence([]byte("foob")) + dst := []byte("ABCDEF") + + // CopyIn limited by len(dst). + n, err := s.CopyIn(newContext(), dst[:2]) + if wantN := 2; n != wantN || err != nil { + t.Errorf("CopyIn: got (%v, %v), wanted (%v, nil)", n, err, wantN) + } + if want := []byte("foCDEF"); !bytes.Equal(dst, want) { + t.Errorf("dst: got %q, wanted %q", dst, want) + } + s = s.DropFirst(2) + if got, want := s.NumBytes(), int64(2); got != want { + t.Errorf("NumBytes: got %v, wanted %v", got, want) + } + + // CopyIn limited by s.Remaining(). + n, err = s.CopyIn(newContext(), dst[2:]) + if wantN := 2; n != wantN || err != nil { + t.Errorf("CopyIn: got (%v, %v), wanted (%v, nil)", n, err, wantN) + } + if want := []byte("foobEF"); !bytes.Equal(dst, want) { + t.Errorf("dst: got %q, wanted %q", dst, want) + } + s = s.DropFirst(2) + if got, want := s.NumBytes(), int64(0); got != want { + t.Errorf("NumBytes: got %v, wanted %v", got, want) + } +} + +func TestIOSequenceZeroOut(t *testing.T) { + buf := []byte("ABCD") + s := BytesIOSequence(buf) + + // ZeroOut limited by toZero. + n, err := s.ZeroOut(newContext(), 2) + if wantN := int64(2); n != wantN || err != nil { + t.Errorf("ZeroOut: got (%v, %v), wanted (%v, nil)", n, err, wantN) + } + if want := []byte("\x00\x00CD"); !bytes.Equal(buf, want) { + t.Errorf("buf: got %q, wanted %q", buf, want) + } + s = s.DropFirst(2) + if got, want := s.NumBytes(), int64(2); got != want { + t.Errorf("NumBytes: got %v, wanted %v", got, want) + } + + // ZeroOut limited by s.NumBytes(). + n, err = s.ZeroOut(newContext(), 4) + if wantN := int64(2); n != wantN || err != nil { + t.Errorf("CopyOut: got (%v, %v), wanted (%v, nil)", n, err, wantN) + } + if want := []byte("\x00\x00\x00\x00"); !bytes.Equal(buf, want) { + t.Errorf("buf: got %q, wanted %q", buf, want) + } + s = s.DropFirst(2) + if got, want := s.NumBytes(), int64(0); got != want { + t.Errorf("NumBytes: got %v, wanted %v", got, want) + } +} + +func TestIOSequenceTakeFirst(t *testing.T) { + s := BytesIOSequence([]byte("foobar")) + if got, want := s.NumBytes(), int64(6); got != want { + t.Errorf("NumBytes: got %v, wanted %v", got, want) + } + + s = s.TakeFirst(3) + if got, want := s.NumBytes(), int64(3); got != want { + t.Errorf("NumBytes: got %v, wanted %v", got, want) + } + + // TakeFirst(n) where n > s.NumBytes() is a no-op. + s = s.TakeFirst(9) + if got, want := s.NumBytes(), int64(3); got != want { + t.Errorf("NumBytes: got %v, wanted %v", got, want) + } + + var dst [3]byte + n, err := s.CopyIn(newContext(), dst[:]) + if wantN := 3; n != wantN || err != nil { + t.Errorf("CopyIn: got (%v, %v), wanted (%v, nil)", n, err, wantN) + } + if got, want := dst[:], []byte("foo"); !bytes.Equal(got, want) { + t.Errorf("dst: got %q, wanted %q", got, want) + } + s = s.DropFirst(3) + if got, want := s.NumBytes(), int64(0); got != want { + t.Errorf("NumBytes: got %v, wanted %v", got, want) + } +} diff --git a/pkg/usermem/usermem_unsafe.go b/pkg/usermem/usermem_unsafe.go new file mode 100644 index 000000000..876783e78 --- /dev/null +++ b/pkg/usermem/usermem_unsafe.go @@ -0,0 +1,27 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package usermem + +import ( + "unsafe" +) + +// stringFromImmutableBytes is equivalent to string(bs), except that it never +// copies even if escape analysis can't prove that bs does not escape. This is +// only valid if bs is never mutated after stringFromImmutableBytes returns. +func stringFromImmutableBytes(bs []byte) string { + // Compare strings.Builder.String(). + return *(*string)(unsafe.Pointer(&bs)) +} diff --git a/pkg/usermem/usermem_x86.go b/pkg/usermem/usermem_x86.go new file mode 100644 index 000000000..8059b72d2 --- /dev/null +++ b/pkg/usermem/usermem_x86.go @@ -0,0 +1,38 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build amd64 i386 + +package usermem + +import "encoding/binary" + +const ( + // PageSize is the system page size. + PageSize = 1 << PageShift + + // HugePageSize is the system huge page size. + HugePageSize = 1 << HugePageShift + + // PageShift is the binary log of the system page size. + PageShift = 12 + + // HugePageShift is the binary log of the system huge page size. + HugePageShift = 21 +) + +var ( + // ByteOrder is the native byte order (little endian). + ByteOrder = binary.LittleEndian +) diff --git a/runsc/boot/BUILD b/runsc/boot/BUILD index f3ebc0231..a96c80261 100644 --- a/runsc/boot/BUILD +++ b/runsc/boot/BUILD @@ -30,6 +30,7 @@ go_library( deps = [ "//pkg/abi", "//pkg/abi/linux", + "//pkg/context", "//pkg/control/server", "//pkg/cpuid", "//pkg/eventchannel", @@ -39,7 +40,6 @@ go_library( "//pkg/refs", "//pkg/sentry/arch", "//pkg/sentry/arch:registers_go_proto", - "//pkg/sentry/context", "//pkg/sentry/control", "//pkg/sentry/fs", "//pkg/sentry/fs/dev", @@ -71,7 +71,6 @@ go_library( "//pkg/sentry/time", "//pkg/sentry/unimpl:unimplemented_syscall_go_proto", "//pkg/sentry/usage", - "//pkg/sentry/usermem", "//pkg/sentry/watchdog", "//pkg/sync", "//pkg/syserror", @@ -88,6 +87,7 @@ go_library( "//pkg/tcpip/transport/tcp", "//pkg/tcpip/transport/udp", "//pkg/urpc", + "//pkg/usermem", "//runsc/boot/filter", "//runsc/boot/platforms", "//runsc/specutils", @@ -111,7 +111,7 @@ go_test( "//pkg/control/server", "//pkg/log", "//pkg/p9", - "//pkg/sentry/context/contexttest", + "//pkg/sentry/contexttest", "//pkg/sentry/fs", "//pkg/sentry/kernel/auth", "//pkg/sync", diff --git a/runsc/boot/fds.go b/runsc/boot/fds.go index e5de1f3d7..417d2d5fb 100644 --- a/runsc/boot/fds.go +++ b/runsc/boot/fds.go @@ -17,7 +17,7 @@ package boot import ( "fmt" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/host" "gvisor.dev/gvisor/pkg/sentry/kernel" diff --git a/runsc/boot/fs.go b/runsc/boot/fs.go index 421ccd255..0f62842ea 100644 --- a/runsc/boot/fs.go +++ b/runsc/boot/fs.go @@ -32,8 +32,8 @@ import ( specs "github.com/opencontainers/runtime-spec/specs-go" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/gofer" "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" diff --git a/runsc/boot/loader_test.go b/runsc/boot/loader_test.go index bec0dc292..44aa63196 100644 --- a/runsc/boot/loader_test.go +++ b/runsc/boot/loader_test.go @@ -27,7 +27,7 @@ import ( "gvisor.dev/gvisor/pkg/control/server" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/p9" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/unet" diff --git a/runsc/boot/user.go b/runsc/boot/user.go index 56cc12ee0..f0aa52135 100644 --- a/runsc/boot/user.go +++ b/runsc/boot/user.go @@ -22,10 +22,10 @@ import ( "strings" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) type fileReader struct { diff --git a/runsc/boot/user_test.go b/runsc/boot/user_test.go index 9aee2ad07..fb4e13dfb 100644 --- a/runsc/boot/user_test.go +++ b/runsc/boot/user_test.go @@ -23,7 +23,7 @@ import ( "testing" specs "github.com/opencontainers/runtime-spec/specs-go" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ) diff --git a/tools/go_marshal/defs.bzl b/tools/go_marshal/defs.bzl index 2918ceffe..d79786a68 100644 --- a/tools/go_marshal/defs.bzl +++ b/tools/go_marshal/defs.bzl @@ -54,8 +54,8 @@ go_marshal = rule( # marshal_deps are the dependencies requied by generated code. marshal_deps = [ "//tools/go_marshal/marshal", - "//pkg/sentry/platform/safecopy", - "//pkg/sentry/usermem", + "//pkg/safecopy", + "//pkg/usermem", ] # marshal_test_deps are required by test targets. diff --git a/tools/go_marshal/gomarshal/generator.go b/tools/go_marshal/gomarshal/generator.go index 8392f3f6d..af90bdecb 100644 --- a/tools/go_marshal/gomarshal/generator.go +++ b/tools/go_marshal/gomarshal/generator.go @@ -27,8 +27,8 @@ import ( const ( marshalImport = "gvisor.dev/gvisor/tools/go_marshal/marshal" - usermemImport = "gvisor.dev/gvisor/pkg/sentry/usermem" - safecopyImport = "gvisor.dev/gvisor/pkg/sentry/platform/safecopy" + safecopyImport = "gvisor.dev/gvisor/pkg/safecopy" + usermemImport = "gvisor.dev/gvisor/pkg/usermem" ) // List of identifiers we use in generated code, that may conflict a diff --git a/tools/go_marshal/test/BUILD b/tools/go_marshal/test/BUILD index 38ba49fed..e345e3a8e 100644 --- a/tools/go_marshal/test/BUILD +++ b/tools/go_marshal/test/BUILD @@ -15,7 +15,7 @@ go_test( deps = [ ":test", "//pkg/binary", - "//pkg/sentry/usermem", + "//pkg/usermem", "//tools/go_marshal/analysis", ], ) diff --git a/tools/go_marshal/test/benchmark_test.go b/tools/go_marshal/test/benchmark_test.go index e70db06d8..e12403741 100644 --- a/tools/go_marshal/test/benchmark_test.go +++ b/tools/go_marshal/test/benchmark_test.go @@ -22,7 +22,7 @@ import ( "testing" "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/tools/go_marshal/analysis" test "gvisor.dev/gvisor/tools/go_marshal/test" ) -- cgit v1.2.3 From d6a2e01d3e57e0837c7e5cfda3b56c4dcfbb4627 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Mon, 27 Jan 2020 16:40:46 -0800 Subject: Address GH comments. --- pkg/sentry/socket/netfilter/netfilter.go | 25 ++++++++++++------------- 1 file changed, 12 insertions(+), 13 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index b8848f08a..a06562743 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -43,7 +43,7 @@ const ( // a metadata struct when the tables are written, and when they are read out we // verify that certain fields are the same. // -// metadata is opaque to netstack. +// metadata is used by this serialization/deserializing code, not netstack. type metadata struct { HookEntry [linux.NF_INET_NUMHOOKS]uint32 Underflow [linux.NF_INET_NUMHOOKS]uint32 @@ -51,14 +51,10 @@ type metadata struct { Size uint32 } -const enableDebug = false - // nflog logs messages related to the writing and reading of iptables, but only // when enableDebug is true. func nflog(format string, args ...interface{}) { - if enableDebug { - log.Infof("netfilter: "+format, args...) - } + log.Infof("netfilter: "+format, args...) } // GetInfo returns information about iptables. @@ -233,14 +229,12 @@ func convertNetstackToBinary(tablename string, table iptables.Table) (linux.Kern return entries, meta, nil } -// TODO: SOMEHOW THIS IS NOT GETTING APPENDED! func marshalMatcher(matcher iptables.Matcher) []byte { switch m := matcher.(type) { case *iptables.UDPMatcher: return marshalUDPMatcher(m) default: - // TODO(gvisor.dev/issue/170): We don't support any matchers - // yet, so any call to marshalMatcher will panic. + // TODO(gvisor.dev/issue/170): Support other matchers. panic(fmt.Errorf("unknown matcher of type %T", matcher)) } } @@ -249,11 +243,11 @@ func marshalUDPMatcher(matcher *iptables.UDPMatcher) []byte { nflog("convert to binary: marshalling UDP matcher: %+v", matcher) // We have to pad this struct size to a multiple of 8 bytes. - const size = linux.SizeOfXTEntryMatch + linux.SizeOfXTUDP + 6 + size := alignUp(linux.SizeOfXTEntryMatch+linux.SizeOfXTUDP, 8) linuxMatcher := linux.KernelXTEntryMatch{ XTEntryMatch: linux.XTEntryMatch{ - MatchSize: size, + MatchSize: uint16(size), }, Data: make([]byte, 0, linux.SizeOfXTUDP), } @@ -270,7 +264,7 @@ func marshalUDPMatcher(matcher *iptables.UDPMatcher) []byte { buf := make([]byte, 0, size) buf = binary.Marshal(buf, usermem.ByteOrder, linuxMatcher) - buf = append(buf, []byte{0, 0, 0, 0, 0, 0}...) + buf = append(buf, make([]byte, size-len(buf))...) nflog("convert to binary: marshalled UDP matcher into %v", buf) return buf[:] } @@ -410,7 +404,7 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { } // TODO(gvisor.dev/issue/170): Matchers and targets can specify - // that they only work for certiain protocols, hooks, tables. + // that they only work for certain protocols, hooks, tables. // Get matchers. matchersSize := entry.TargetOffset - linux.SizeOfIPTEntry if len(optVal) < int(matchersSize) { @@ -684,3 +678,8 @@ func hookFromLinux(hook int) iptables.Hook { } panic(fmt.Sprintf("Unknown hook %d does not correspond to a builtin chain", hook)) } + +// alignUp rounds a length up to an alignment. align must be a power of 2. +func alignUp(length int, align uint) int { + return (length + int(align) - 1) & ^(int(align) - 1) +} -- cgit v1.2.3 From 76483b8b1ec4ee1fb6b6efb6bdcfaf6dba7be4ce Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Tue, 28 Jan 2020 11:12:01 -0800 Subject: Check sigsetsize in rt_sigaction This isn't in the libc wrapper, but it is in the syscall itself. Discovered by @xiaobo55x in #1625. PiperOrigin-RevId: 291973931 --- pkg/sentry/strace/linux64_amd64.go | 2 +- pkg/sentry/strace/linux64_arm64.go | 2 +- pkg/sentry/syscalls/linux/sys_signal.go | 5 ++++ test/syscalls/linux/sigaction.cc | 53 +++++++++++++++++++-------------- 4 files changed, 38 insertions(+), 24 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/strace/linux64_amd64.go b/pkg/sentry/strace/linux64_amd64.go index 1e823b685..85ec66fd3 100644 --- a/pkg/sentry/strace/linux64_amd64.go +++ b/pkg/sentry/strace/linux64_amd64.go @@ -37,7 +37,7 @@ var linuxAMD64 = SyscallMap{ 10: makeSyscallInfo("mprotect", Hex, Hex, Hex), 11: makeSyscallInfo("munmap", Hex, Hex), 12: makeSyscallInfo("brk", Hex), - 13: makeSyscallInfo("rt_sigaction", Signal, SigAction, PostSigAction), + 13: makeSyscallInfo("rt_sigaction", Signal, SigAction, PostSigAction, Hex), 14: makeSyscallInfo("rt_sigprocmask", SignalMaskAction, SigSet, PostSigSet, Hex), 15: makeSyscallInfo("rt_sigreturn"), 16: makeSyscallInfo("ioctl", FD, Hex, Hex), diff --git a/pkg/sentry/strace/linux64_arm64.go b/pkg/sentry/strace/linux64_arm64.go index c3ac5248d..8bc38545f 100644 --- a/pkg/sentry/strace/linux64_arm64.go +++ b/pkg/sentry/strace/linux64_arm64.go @@ -158,7 +158,7 @@ var linuxARM64 = SyscallMap{ 131: makeSyscallInfo("tgkill", Hex, Hex, Signal), 132: makeSyscallInfo("sigaltstack", Hex, Hex), 133: makeSyscallInfo("rt_sigsuspend", Hex), - 134: makeSyscallInfo("rt_sigaction", Signal, SigAction, PostSigAction), + 134: makeSyscallInfo("rt_sigaction", Signal, SigAction, PostSigAction, Hex), 135: makeSyscallInfo("rt_sigprocmask", SignalMaskAction, SigSet, PostSigSet, Hex), 136: makeSyscallInfo("rt_sigpending", Hex), 137: makeSyscallInfo("rt_sigtimedwait", SigSet, Hex, Timespec, Hex), diff --git a/pkg/sentry/syscalls/linux/sys_signal.go b/pkg/sentry/syscalls/linux/sys_signal.go index 209be2990..7e1747a0c 100644 --- a/pkg/sentry/syscalls/linux/sys_signal.go +++ b/pkg/sentry/syscalls/linux/sys_signal.go @@ -245,6 +245,11 @@ func RtSigaction(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S sig := linux.Signal(args[0].Int()) newactarg := args[1].Pointer() oldactarg := args[2].Pointer() + sigsetsize := args[3].SizeT() + + if sigsetsize != linux.SignalSetSize { + return 0, nil, syserror.EINVAL + } var newactptr *arch.SignalAct if newactarg != 0 { diff --git a/test/syscalls/linux/sigaction.cc b/test/syscalls/linux/sigaction.cc index 9a53fd3e0..9d9dd57a8 100644 --- a/test/syscalls/linux/sigaction.cc +++ b/test/syscalls/linux/sigaction.cc @@ -13,6 +13,7 @@ // limitations under the License. #include +#include #include "gtest/gtest.h" #include "test/util/test_util.h" @@ -23,45 +24,53 @@ namespace testing { namespace { TEST(SigactionTest, GetLessThanOrEqualToZeroFails) { - struct sigaction act; - memset(&act, 0, sizeof(act)); - ASSERT_THAT(sigaction(-1, NULL, &act), SyscallFailsWithErrno(EINVAL)); - ASSERT_THAT(sigaction(0, NULL, &act), SyscallFailsWithErrno(EINVAL)); + struct sigaction act = {}; + ASSERT_THAT(sigaction(-1, nullptr, &act), SyscallFailsWithErrno(EINVAL)); + ASSERT_THAT(sigaction(0, nullptr, &act), SyscallFailsWithErrno(EINVAL)); } TEST(SigactionTest, SetLessThanOrEqualToZeroFails) { - struct sigaction act; - memset(&act, 0, sizeof(act)); - ASSERT_THAT(sigaction(0, &act, NULL), SyscallFailsWithErrno(EINVAL)); - ASSERT_THAT(sigaction(0, &act, NULL), SyscallFailsWithErrno(EINVAL)); + struct sigaction act = {}; + ASSERT_THAT(sigaction(0, &act, nullptr), SyscallFailsWithErrno(EINVAL)); + ASSERT_THAT(sigaction(0, &act, nullptr), SyscallFailsWithErrno(EINVAL)); } TEST(SigactionTest, GetGreaterThanMaxFails) { - struct sigaction act; - memset(&act, 0, sizeof(act)); - ASSERT_THAT(sigaction(SIGRTMAX + 1, NULL, &act), + struct sigaction act = {}; + ASSERT_THAT(sigaction(SIGRTMAX + 1, nullptr, &act), SyscallFailsWithErrno(EINVAL)); } TEST(SigactionTest, SetGreaterThanMaxFails) { - struct sigaction act; - memset(&act, 0, sizeof(act)); - ASSERT_THAT(sigaction(SIGRTMAX + 1, &act, NULL), + struct sigaction act = {}; + ASSERT_THAT(sigaction(SIGRTMAX + 1, &act, nullptr), SyscallFailsWithErrno(EINVAL)); } TEST(SigactionTest, SetSigkillFails) { - struct sigaction act; - memset(&act, 0, sizeof(act)); - ASSERT_THAT(sigaction(SIGKILL, NULL, &act), SyscallSucceeds()); - ASSERT_THAT(sigaction(SIGKILL, &act, NULL), SyscallFailsWithErrno(EINVAL)); + struct sigaction act = {}; + ASSERT_THAT(sigaction(SIGKILL, nullptr, &act), SyscallSucceeds()); + ASSERT_THAT(sigaction(SIGKILL, &act, nullptr), SyscallFailsWithErrno(EINVAL)); } TEST(SigactionTest, SetSigstopFails) { - struct sigaction act; - memset(&act, 0, sizeof(act)); - ASSERT_THAT(sigaction(SIGSTOP, NULL, &act), SyscallSucceeds()); - ASSERT_THAT(sigaction(SIGSTOP, &act, NULL), SyscallFailsWithErrno(EINVAL)); + struct sigaction act = {}; + ASSERT_THAT(sigaction(SIGSTOP, nullptr, &act), SyscallSucceeds()); + ASSERT_THAT(sigaction(SIGSTOP, &act, nullptr), SyscallFailsWithErrno(EINVAL)); +} + +TEST(SigactionTest, BadSigsetFails) { + constexpr size_t kWrongSigSetSize = 43; + + struct sigaction act = {}; + + // The syscall itself (rather than the libc wrapper) takes the sigset_t size. + ASSERT_THAT( + syscall(SYS_rt_sigaction, SIGTERM, nullptr, &act, kWrongSigSetSize), + SyscallFailsWithErrno(EINVAL)); + ASSERT_THAT( + syscall(SYS_rt_sigaction, SIGTERM, &act, nullptr, kWrongSigSetSize), + SyscallFailsWithErrno(EINVAL)); } } // namespace -- cgit v1.2.3 From 1119644080ae57c206b9b0d8d127cf48423af7f2 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Tue, 28 Jan 2020 12:06:58 -0800 Subject: Implement an anon_inode equivalent for VFS2. PiperOrigin-RevId: 291986033 --- pkg/sentry/vfs/BUILD | 3 +- pkg/sentry/vfs/anonfs.go | 259 ++++++++++++++++++++++ pkg/sentry/vfs/file_description_impl_util_test.go | 18 +- pkg/sentry/vfs/testutil.go | 173 --------------- pkg/sentry/vfs/vfs.go | 24 ++ 5 files changed, 289 insertions(+), 188 deletions(-) create mode 100644 pkg/sentry/vfs/anonfs.go delete mode 100644 pkg/sentry/vfs/testutil.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index 6b1009328..33516e6f7 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -5,6 +5,7 @@ licenses(["notice"]) go_library( name = "vfs", srcs = [ + "anonfs.go", "context.go", "debug.go", "dentry.go", @@ -20,7 +21,6 @@ go_library( "pathname.go", "permissions.go", "resolving_path.go", - "testutil.go", "vfs.go", ], visibility = ["//pkg/sentry:internal"], @@ -50,7 +50,6 @@ go_test( "//pkg/abi/linux", "//pkg/context", "//pkg/sentry/contexttest", - "//pkg/sentry/kernel/auth", "//pkg/sync", "//pkg/syserror", "//pkg/usermem", diff --git a/pkg/sentry/vfs/anonfs.go b/pkg/sentry/vfs/anonfs.go new file mode 100644 index 000000000..2db25be49 --- /dev/null +++ b/pkg/sentry/vfs/anonfs.go @@ -0,0 +1,259 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs + +import ( + "fmt" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +// NewAnonVirtualDentry returns a VirtualDentry with the given synthetic name, +// consistent with Linux's fs/anon_inodes.c:anon_inode_getfile(). References +// are taken on the returned VirtualDentry. +func (vfs *VirtualFilesystem) NewAnonVirtualDentry(name string) VirtualDentry { + d := anonDentry{ + name: name, + } + d.vfsd.Init(&d) + vfs.anonMount.IncRef() + // anonDentry no-ops refcounting. + return VirtualDentry{ + mount: vfs.anonMount, + dentry: &d.vfsd, + } +} + +const anonfsBlockSize = usermem.PageSize // via fs/libfs.c:pseudo_fs_fill_super() + +// anonFilesystem is the implementation of FilesystemImpl that backs +// VirtualDentries returned by VirtualFilesystem.NewAnonVirtualDentry(). +// +// Since all Dentries in anonFilesystem are non-directories, all FilesystemImpl +// methods that would require an anonDentry to be a directory return ENOTDIR. +type anonFilesystem struct { + vfsfs Filesystem + + devMinor uint32 +} + +type anonDentry struct { + vfsd Dentry + + name string +} + +// Release implements FilesystemImpl.Release. +func (fs *anonFilesystem) Release() { +} + +// Sync implements FilesystemImpl.Sync. +func (fs *anonFilesystem) Sync(ctx context.Context) error { + return nil +} + +// GetDentryAt implements FilesystemImpl.GetDentryAt. +func (fs *anonFilesystem) GetDentryAt(ctx context.Context, rp *ResolvingPath, opts GetDentryOptions) (*Dentry, error) { + if !rp.Done() { + return nil, syserror.ENOTDIR + } + if opts.CheckSearchable { + return nil, syserror.ENOTDIR + } + // anonDentry no-ops refcounting. + return rp.Start(), nil +} + +// GetParentDentryAt implements FilesystemImpl.GetParentDentryAt. +func (fs *anonFilesystem) GetParentDentryAt(ctx context.Context, rp *ResolvingPath) (*Dentry, error) { + if !rp.Final() { + return nil, syserror.ENOTDIR + } + // anonDentry no-ops refcounting. + return rp.Start(), nil +} + +// LinkAt implements FilesystemImpl.LinkAt. +func (fs *anonFilesystem) LinkAt(ctx context.Context, rp *ResolvingPath, vd VirtualDentry) error { + if !rp.Final() { + return syserror.ENOTDIR + } + return syserror.EPERM +} + +// MkdirAt implements FilesystemImpl.MkdirAt. +func (fs *anonFilesystem) MkdirAt(ctx context.Context, rp *ResolvingPath, opts MkdirOptions) error { + if !rp.Final() { + return syserror.ENOTDIR + } + return syserror.EPERM +} + +// MknodAt implements FilesystemImpl.MknodAt. +func (fs *anonFilesystem) MknodAt(ctx context.Context, rp *ResolvingPath, opts MknodOptions) error { + if !rp.Final() { + return syserror.ENOTDIR + } + return syserror.EPERM +} + +// OpenAt implements FilesystemImpl.OpenAt. +func (fs *anonFilesystem) OpenAt(ctx context.Context, rp *ResolvingPath, opts OpenOptions) (*FileDescription, error) { + if !rp.Done() { + return nil, syserror.ENOTDIR + } + return nil, syserror.ENODEV +} + +// ReadlinkAt implements FilesystemImpl.ReadlinkAt. +func (fs *anonFilesystem) ReadlinkAt(ctx context.Context, rp *ResolvingPath) (string, error) { + if !rp.Done() { + return "", syserror.ENOTDIR + } + return "", syserror.EINVAL +} + +// RenameAt implements FilesystemImpl.RenameAt. +func (fs *anonFilesystem) RenameAt(ctx context.Context, rp *ResolvingPath, oldParentVD VirtualDentry, oldName string, opts RenameOptions) error { + if !rp.Final() { + return syserror.ENOTDIR + } + return syserror.EPERM +} + +// RmdirAt implements FilesystemImpl.RmdirAt. +func (fs *anonFilesystem) RmdirAt(ctx context.Context, rp *ResolvingPath) error { + if !rp.Final() { + return syserror.ENOTDIR + } + return syserror.EPERM +} + +// SetStatAt implements FilesystemImpl.SetStatAt. +func (fs *anonFilesystem) SetStatAt(ctx context.Context, rp *ResolvingPath, opts SetStatOptions) error { + if !rp.Done() { + return syserror.ENOTDIR + } + // Linux actually permits anon_inode_inode's metadata to be set, which is + // visible to all users of anon_inode_inode. We just silently ignore + // metadata changes. + return nil +} + +// StatAt implements FilesystemImpl.StatAt. +func (fs *anonFilesystem) StatAt(ctx context.Context, rp *ResolvingPath, opts StatOptions) (linux.Statx, error) { + if !rp.Done() { + return linux.Statx{}, syserror.ENOTDIR + } + // See fs/anon_inodes.c:anon_inode_init() => fs/libfs.c:alloc_anon_inode(). + return linux.Statx{ + Mask: linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_NLINK | linux.STATX_UID | linux.STATX_GID | linux.STATX_INO | linux.STATX_SIZE | linux.STATX_BLOCKS, + Blksize: anonfsBlockSize, + Nlink: 1, + UID: uint32(auth.RootKUID), + GID: uint32(auth.RootKGID), + Mode: 0600, // no type is correct + Ino: 1, + Size: 0, + Blocks: 0, + DevMajor: 0, + DevMinor: fs.devMinor, + }, nil +} + +// StatFSAt implements FilesystemImpl.StatFSAt. +func (fs *anonFilesystem) StatFSAt(ctx context.Context, rp *ResolvingPath) (linux.Statfs, error) { + if !rp.Done() { + return linux.Statfs{}, syserror.ENOTDIR + } + return linux.Statfs{ + Type: linux.ANON_INODE_FS_MAGIC, + BlockSize: anonfsBlockSize, + }, nil +} + +// SymlinkAt implements FilesystemImpl.SymlinkAt. +func (fs *anonFilesystem) SymlinkAt(ctx context.Context, rp *ResolvingPath, target string) error { + if !rp.Final() { + return syserror.ENOTDIR + } + return syserror.EPERM +} + +// UnlinkAt implements FilesystemImpl.UnlinkAt. +func (fs *anonFilesystem) UnlinkAt(ctx context.Context, rp *ResolvingPath) error { + if !rp.Final() { + return syserror.ENOTDIR + } + return syserror.EPERM +} + +// ListxattrAt implements FilesystemImpl.ListxattrAt. +func (fs *anonFilesystem) ListxattrAt(ctx context.Context, rp *ResolvingPath) ([]string, error) { + if !rp.Done() { + return nil, syserror.ENOTDIR + } + return nil, nil +} + +// GetxattrAt implements FilesystemImpl.GetxattrAt. +func (fs *anonFilesystem) GetxattrAt(ctx context.Context, rp *ResolvingPath, name string) (string, error) { + if !rp.Done() { + return "", syserror.ENOTDIR + } + return "", syserror.ENOTSUP +} + +// SetxattrAt implements FilesystemImpl.SetxattrAt. +func (fs *anonFilesystem) SetxattrAt(ctx context.Context, rp *ResolvingPath, opts SetxattrOptions) error { + if !rp.Done() { + return syserror.ENOTDIR + } + return syserror.EPERM +} + +// RemovexattrAt implements FilesystemImpl.RemovexattrAt. +func (fs *anonFilesystem) RemovexattrAt(ctx context.Context, rp *ResolvingPath, name string) error { + if !rp.Done() { + return syserror.ENOTDIR + } + return syserror.EPERM +} + +// PrependPath implements FilesystemImpl.PrependPath. +func (fs *anonFilesystem) PrependPath(ctx context.Context, vfsroot, vd VirtualDentry, b *fspath.Builder) error { + b.PrependComponent(fmt.Sprintf("anon_inode:%s", vd.dentry.impl.(*anonDentry).name)) + return PrependPathSyntheticError{} +} + +// IncRef implements DentryImpl.IncRef. +func (d *anonDentry) IncRef() { + // no-op +} + +// TryIncRef implements DentryImpl.TryIncRef. +func (d *anonDentry) TryIncRef() bool { + return true +} + +// DecRef implements DentryImpl.DecRef. +func (d *anonDentry) DecRef() { + // no-op +} diff --git a/pkg/sentry/vfs/file_description_impl_util_test.go b/pkg/sentry/vfs/file_description_impl_util_test.go index 1720d325d..0f44e7c8c 100644 --- a/pkg/sentry/vfs/file_description_impl_util_test.go +++ b/pkg/sentry/vfs/file_description_impl_util_test.go @@ -24,7 +24,6 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" ) @@ -46,9 +45,11 @@ type genCountFD struct { count uint64 // accessed using atomic memory ops } -func newGenCountFD(mnt *Mount, vfsd *Dentry) *FileDescription { +func newGenCountFD(vfsObj *VirtualFilesystem) *FileDescription { + vd := vfsObj.NewAnonVirtualDentry("genCountFD") + defer vd.DecRef() var fd genCountFD - fd.vfsfd.Init(&fd, 0 /* statusFlags */, mnt, vfsd, &FileDescriptionOptions{}) + fd.vfsfd.Init(&fd, 0 /* statusFlags */, vd.Mount(), vd.Dentry(), &FileDescriptionOptions{}) fd.DynamicBytesFileDescriptionImpl.SetDataSource(&fd) return &fd.vfsfd } @@ -86,18 +87,9 @@ func (fd *genCountFD) Generate(ctx context.Context, buf *bytes.Buffer) error { func TestGenCountFD(t *testing.T) { ctx := contexttest.Context(t) - creds := auth.CredentialsFromContext(ctx) vfsObj := New() // vfs.New() - vfsObj.MustRegisterFilesystemType("testfs", FDTestFilesystemType{}, &RegisterFilesystemTypeOptions{}) - mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "testfs", &GetFilesystemOptions{}) - if err != nil { - t.Fatalf("failed to create testfs root mount: %v", err) - } - vd := mntns.Root() - defer vd.DecRef() - - fd := newGenCountFD(vd.Mount(), vd.Dentry()) + fd := newGenCountFD(vfsObj) defer fd.DecRef() // The first read causes Generate to be called to fill the FD's buffer. diff --git a/pkg/sentry/vfs/testutil.go b/pkg/sentry/vfs/testutil.go deleted file mode 100644 index 392c7611e..000000000 --- a/pkg/sentry/vfs/testutil.go +++ /dev/null @@ -1,173 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package vfs - -import ( - "fmt" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/syserror" -) - -// FDTestFilesystemType is a test-only FilesystemType that produces Filesystems -// for which all FilesystemImpl methods taking a path return EPERM. It is used -// to produce Mounts and Dentries for testing of FileDescriptionImpls that do -// not depend on their originating Filesystem. -type FDTestFilesystemType struct{} - -// FDTestFilesystem is a test-only FilesystemImpl produced by -// FDTestFilesystemType. -type FDTestFilesystem struct { - vfsfs Filesystem -} - -// GetFilesystem implements FilesystemType.GetFilesystem. -func (fstype FDTestFilesystemType) GetFilesystem(ctx context.Context, vfsObj *VirtualFilesystem, creds *auth.Credentials, source string, opts GetFilesystemOptions) (*Filesystem, *Dentry, error) { - var fs FDTestFilesystem - fs.vfsfs.Init(vfsObj, &fs) - return &fs.vfsfs, fs.NewDentry(), nil -} - -// Release implements FilesystemImpl.Release. -func (fs *FDTestFilesystem) Release() { -} - -// Sync implements FilesystemImpl.Sync. -func (fs *FDTestFilesystem) Sync(ctx context.Context) error { - return nil -} - -// GetDentryAt implements FilesystemImpl.GetDentryAt. -func (fs *FDTestFilesystem) GetDentryAt(ctx context.Context, rp *ResolvingPath, opts GetDentryOptions) (*Dentry, error) { - return nil, syserror.EPERM -} - -// GetParentDentryAt implements FilesystemImpl.GetParentDentryAt. -func (fs *FDTestFilesystem) GetParentDentryAt(ctx context.Context, rp *ResolvingPath) (*Dentry, error) { - return nil, syserror.EPERM -} - -// LinkAt implements FilesystemImpl.LinkAt. -func (fs *FDTestFilesystem) LinkAt(ctx context.Context, rp *ResolvingPath, vd VirtualDentry) error { - return syserror.EPERM -} - -// MkdirAt implements FilesystemImpl.MkdirAt. -func (fs *FDTestFilesystem) MkdirAt(ctx context.Context, rp *ResolvingPath, opts MkdirOptions) error { - return syserror.EPERM -} - -// MknodAt implements FilesystemImpl.MknodAt. -func (fs *FDTestFilesystem) MknodAt(ctx context.Context, rp *ResolvingPath, opts MknodOptions) error { - return syserror.EPERM -} - -// OpenAt implements FilesystemImpl.OpenAt. -func (fs *FDTestFilesystem) OpenAt(ctx context.Context, rp *ResolvingPath, opts OpenOptions) (*FileDescription, error) { - return nil, syserror.EPERM -} - -// ReadlinkAt implements FilesystemImpl.ReadlinkAt. -func (fs *FDTestFilesystem) ReadlinkAt(ctx context.Context, rp *ResolvingPath) (string, error) { - return "", syserror.EPERM -} - -// RenameAt implements FilesystemImpl.RenameAt. -func (fs *FDTestFilesystem) RenameAt(ctx context.Context, rp *ResolvingPath, oldParentVD VirtualDentry, oldName string, opts RenameOptions) error { - return syserror.EPERM -} - -// RmdirAt implements FilesystemImpl.RmdirAt. -func (fs *FDTestFilesystem) RmdirAt(ctx context.Context, rp *ResolvingPath) error { - return syserror.EPERM -} - -// SetStatAt implements FilesystemImpl.SetStatAt. -func (fs *FDTestFilesystem) SetStatAt(ctx context.Context, rp *ResolvingPath, opts SetStatOptions) error { - return syserror.EPERM -} - -// StatAt implements FilesystemImpl.StatAt. -func (fs *FDTestFilesystem) StatAt(ctx context.Context, rp *ResolvingPath, opts StatOptions) (linux.Statx, error) { - return linux.Statx{}, syserror.EPERM -} - -// StatFSAt implements FilesystemImpl.StatFSAt. -func (fs *FDTestFilesystem) StatFSAt(ctx context.Context, rp *ResolvingPath) (linux.Statfs, error) { - return linux.Statfs{}, syserror.EPERM -} - -// SymlinkAt implements FilesystemImpl.SymlinkAt. -func (fs *FDTestFilesystem) SymlinkAt(ctx context.Context, rp *ResolvingPath, target string) error { - return syserror.EPERM -} - -// UnlinkAt implements FilesystemImpl.UnlinkAt. -func (fs *FDTestFilesystem) UnlinkAt(ctx context.Context, rp *ResolvingPath) error { - return syserror.EPERM -} - -// ListxattrAt implements FilesystemImpl.ListxattrAt. -func (fs *FDTestFilesystem) ListxattrAt(ctx context.Context, rp *ResolvingPath) ([]string, error) { - return nil, syserror.EPERM -} - -// GetxattrAt implements FilesystemImpl.GetxattrAt. -func (fs *FDTestFilesystem) GetxattrAt(ctx context.Context, rp *ResolvingPath, name string) (string, error) { - return "", syserror.EPERM -} - -// SetxattrAt implements FilesystemImpl.SetxattrAt. -func (fs *FDTestFilesystem) SetxattrAt(ctx context.Context, rp *ResolvingPath, opts SetxattrOptions) error { - return syserror.EPERM -} - -// RemovexattrAt implements FilesystemImpl.RemovexattrAt. -func (fs *FDTestFilesystem) RemovexattrAt(ctx context.Context, rp *ResolvingPath, name string) error { - return syserror.EPERM -} - -// PrependPath implements FilesystemImpl.PrependPath. -func (fs *FDTestFilesystem) PrependPath(ctx context.Context, vfsroot, vd VirtualDentry, b *fspath.Builder) error { - b.PrependComponent(fmt.Sprintf("vfs.fdTestDentry:%p", vd.dentry.impl.(*fdTestDentry))) - return PrependPathSyntheticError{} -} - -type fdTestDentry struct { - vfsd Dentry -} - -// NewDentry returns a new Dentry. -func (fs *FDTestFilesystem) NewDentry() *Dentry { - var d fdTestDentry - d.vfsd.Init(&d) - return &d.vfsd -} - -// IncRef implements DentryImpl.IncRef. -func (d *fdTestDentry) IncRef() { -} - -// TryIncRef implements DentryImpl.TryIncRef. -func (d *fdTestDentry) TryIncRef() bool { - return true -} - -// DecRef implements DentryImpl.DecRef. -func (d *fdTestDentry) DecRef() { -} diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index b2bf48853..d730530b9 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -75,6 +75,14 @@ type VirtualFilesystem struct { // mountpoints is analogous to Linux's mountpoint_hashtable. mountpoints map[*Dentry]map[*Mount]struct{} + // anonMount is a Mount, not included in mounts or mountpoints, + // representing an anonFilesystem. anonMount is used to back + // VirtualDentries returned by VirtualFilesystem.NewAnonVirtualDentry(). + // anonMount is immutable. + // + // anonMount is analogous to Linux's anon_inode_mnt. + anonMount *Mount + // devices contains all registered Devices. devices is protected by // devicesMu. devicesMu sync.RWMutex @@ -110,6 +118,22 @@ func New() *VirtualFilesystem { filesystems: make(map[*Filesystem]struct{}), } vfs.mounts.Init() + + // Construct vfs.anonMount. + anonfsDevMinor, err := vfs.GetAnonBlockDevMinor() + if err != nil { + panic(fmt.Sprintf("VirtualFilesystem.GetAnonBlockDevMinor() failed during VirtualFilesystem construction: %v", err)) + } + anonfs := anonFilesystem{ + devMinor: anonfsDevMinor, + } + anonfs.vfsfs.Init(vfs, &anonfs) + vfs.anonMount = &Mount{ + vfs: vfs, + fs: &anonfs.vfsfs, + refs: 1, + } + return vfs } -- cgit v1.2.3 From d99329e58492ef91b44a0bac346f757e8af2a7ec Mon Sep 17 00:00:00 2001 From: Jianfeng Tan Date: Tue, 28 Jan 2020 12:31:58 -0800 Subject: netlink: add support for RTM_F_LOOKUP_TABLE Test command: $ ip route get 1.1.1.1 Fixes: #1099 Signed-off-by: Jianfeng Tan COPYBARA_INTEGRATE_REVIEW=https://github.com/google/gvisor/pull/1121 from tanjianfeng:fix-1099 e6919f3d4ede5aa51a48b3d2be0d7a4b482dd53d PiperOrigin-RevId: 291990716 --- pkg/abi/linux/netlink_route.go | 13 +++ pkg/sentry/socket/netlink/route/BUILD | 6 +- pkg/sentry/socket/netlink/route/protocol.go | 158 +++++++++++++++++++++++++--- test/syscalls/linux/socket_netlink_route.cc | 84 +++++++++++++++ test/syscalls/linux/socket_netlink_util.cc | 38 +++++++ test/syscalls/linux/socket_netlink_util.h | 11 +- 6 files changed, 295 insertions(+), 15 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/netlink_route.go b/pkg/abi/linux/netlink_route.go index 0e3582ab6..40bec566c 100644 --- a/pkg/abi/linux/netlink_route.go +++ b/pkg/abi/linux/netlink_route.go @@ -205,6 +205,9 @@ type RouteMessage struct { Flags uint32 } +// SizeOfRouteMessage is the size of RouteMessage. +const SizeOfRouteMessage = 12 + // Route types, from uapi/linux/rtnetlink.h. const ( // RTN_UNSPEC represents an unspecified route type. @@ -331,3 +334,13 @@ const ( RTF_GATEWAY = 0x2 RTF_UP = 0x1 ) + +// RtAttr is the header of optional addition route information, as a netlink +// attribute. From include/uapi/linux/rtnetlink.h. +type RtAttr struct { + Len uint16 + Type uint16 +} + +// SizeOfRtAttr is the size of RtAttr. +const SizeOfRtAttr = 4 diff --git a/pkg/sentry/socket/netlink/route/BUILD b/pkg/sentry/socket/netlink/route/BUILD index 0234aadde..622a1eafc 100644 --- a/pkg/sentry/socket/netlink/route/BUILD +++ b/pkg/sentry/socket/netlink/route/BUILD @@ -4,15 +4,19 @@ package(licenses = ["notice"]) go_library( name = "route", - srcs = ["protocol.go"], + srcs = [ + "protocol.go", + ], visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", + "//pkg/binary", "//pkg/context", "//pkg/sentry/inet", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/socket/netlink", "//pkg/syserr", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/socket/netlink/route/protocol.go b/pkg/sentry/socket/netlink/route/protocol.go index 80a15d6cb..2b3c7f5b3 100644 --- a/pkg/sentry/socket/netlink/route/protocol.go +++ b/pkg/sentry/socket/netlink/route/protocol.go @@ -19,12 +19,14 @@ import ( "bytes" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/socket/netlink" "gvisor.dev/gvisor/pkg/syserr" + "gvisor.dev/gvisor/pkg/usermem" ) // commandKind describes the operational class of a message type. @@ -66,8 +68,14 @@ func (p *Protocol) CanSend() bool { return true } -// dumpLinks handles RTM_GETLINK + NLM_F_DUMP requests. +// dumpLinks handles RTM_GETLINK dump requests. func (p *Protocol) dumpLinks(ctx context.Context, hdr linux.NetlinkMessageHeader, data []byte, ms *netlink.MessageSet) *syserr.Error { + // TODO(b/68878065): Only the dump variant of the types below are + // supported. + if hdr.Flags&linux.NLM_F_DUMP != linux.NLM_F_DUMP { + return syserr.ErrNotSupported + } + // NLM_F_DUMP + RTM_GETLINK messages are supposed to include an // ifinfomsg. However, Linux <3.9 only checked for rtgenmsg, and some // userspace applications (including glibc) still include rtgenmsg. @@ -121,8 +129,14 @@ func (p *Protocol) dumpLinks(ctx context.Context, hdr linux.NetlinkMessageHeader return nil } -// dumpAddrs handles RTM_GETADDR + NLM_F_DUMP requests. +// dumpAddrs handles RTM_GETADDR dump requests. func (p *Protocol) dumpAddrs(ctx context.Context, hdr linux.NetlinkMessageHeader, data []byte, ms *netlink.MessageSet) *syserr.Error { + // TODO(b/68878065): Only the dump variant of the types below are + // supported. + if hdr.Flags&linux.NLM_F_DUMP != linux.NLM_F_DUMP { + return syserr.ErrNotSupported + } + // RTM_GETADDR dump requests need not contain anything more than the // netlink header and 1 byte protocol family common to all // NETLINK_ROUTE requests. @@ -163,22 +177,146 @@ func (p *Protocol) dumpAddrs(ctx context.Context, hdr linux.NetlinkMessageHeader return nil } -// dumpRoutes handles RTM_GETROUTE + NLM_F_DUMP requests. +// commonPrefixLen reports the length of the longest IP address prefix. +// This is a simplied version from Golang's src/net/addrselect.go. +func commonPrefixLen(a, b []byte) (cpl int) { + for len(a) > 0 { + if a[0] == b[0] { + cpl += 8 + a = a[1:] + b = b[1:] + continue + } + bits := 8 + ab, bb := a[0], b[0] + for { + ab >>= 1 + bb >>= 1 + bits-- + if ab == bb { + cpl += bits + return + } + } + } + return +} + +// fillRoute returns the Route using LPM algorithm. Refer to Linux's +// net/ipv4/route.c:rt_fill_info(). +func fillRoute(routes []inet.Route, addr []byte) (inet.Route, *syserr.Error) { + family := uint8(linux.AF_INET) + if len(addr) != 4 { + family = linux.AF_INET6 + } + + idx := -1 // Index of the Route rule to be returned. + idxDef := -1 // Index of the default route rule. + prefix := 0 // Current longest prefix. + for i, route := range routes { + if route.Family != family { + continue + } + + if len(route.GatewayAddr) > 0 && route.DstLen == 0 { + idxDef = i + continue + } + + cpl := commonPrefixLen(addr, route.DstAddr) + if cpl < int(route.DstLen) { + continue + } + cpl = int(route.DstLen) + if cpl > prefix { + idx = i + prefix = cpl + } + } + if idx == -1 { + idx = idxDef + } + if idx == -1 { + return inet.Route{}, syserr.ErrNoRoute + } + + route := routes[idx] + if family == linux.AF_INET { + route.DstLen = 32 + } else { + route.DstLen = 128 + } + route.DstAddr = addr + route.Flags |= linux.RTM_F_CLONED // This route is cloned. + return route, nil +} + +// parseForDestination parses a message as format of RouteMessage-RtAttr-dst. +func parseForDestination(data []byte) ([]byte, *syserr.Error) { + var rtMsg linux.RouteMessage + if len(data) < linux.SizeOfRouteMessage { + return nil, syserr.ErrInvalidArgument + } + binary.Unmarshal(data[:linux.SizeOfRouteMessage], usermem.ByteOrder, &rtMsg) + // iproute2 added the RTM_F_LOOKUP_TABLE flag in version v4.4.0. See + // commit bc234301af12. Note we don't check this flag for backward + // compatibility. + if rtMsg.Flags != 0 && rtMsg.Flags != linux.RTM_F_LOOKUP_TABLE { + return nil, syserr.ErrNotSupported + } + + data = data[linux.SizeOfRouteMessage:] + + // TODO(gvisor.dev/issue/1611): Add generic attribute parsing. + var rtAttr linux.RtAttr + if len(data) < linux.SizeOfRtAttr { + return nil, syserr.ErrInvalidArgument + } + binary.Unmarshal(data[:linux.SizeOfRtAttr], usermem.ByteOrder, &rtAttr) + if rtAttr.Type != linux.RTA_DST { + return nil, syserr.ErrInvalidArgument + } + + if len(data) < int(rtAttr.Len) { + return nil, syserr.ErrInvalidArgument + } + return data[linux.SizeOfRtAttr:rtAttr.Len], nil +} + +// dumpRoutes handles RTM_GETROUTE requests. func (p *Protocol) dumpRoutes(ctx context.Context, hdr linux.NetlinkMessageHeader, data []byte, ms *netlink.MessageSet) *syserr.Error { // RTM_GETROUTE dump requests need not contain anything more than the // netlink header and 1 byte protocol family common to all // NETLINK_ROUTE requests. - // We always send back an NLMSG_DONE. - ms.Multi = true - stack := inet.StackFromContext(ctx) if stack == nil { // No network routes. return nil } - for _, rt := range stack.RouteTable() { + routeTables := stack.RouteTable() + + if hdr.Flags == linux.NLM_F_REQUEST { + dst, err := parseForDestination(data) + if err != nil { + return err + } + route, err := fillRoute(routeTables, dst) + if err != nil { + // TODO(gvisor.dev/issue/1237): return NLMSG_ERROR with ENETUNREACH. + return syserr.ErrNotSupported + } + routeTables = append([]inet.Route{}, route) + } else if hdr.Flags&linux.NLM_F_DUMP == linux.NLM_F_DUMP { + // We always send back an NLMSG_DONE. + ms.Multi = true + } else { + // TODO(b/68878065): Only above cases are supported. + return syserr.ErrNotSupported + } + + for _, rt := range routeTables { m := ms.AddMessage(linux.NetlinkMessageHeader{ Type: linux.RTM_NEWROUTE, }) @@ -236,12 +374,6 @@ func (p *Protocol) ProcessMessage(ctx context.Context, hdr linux.NetlinkMessageH } } - // TODO(b/68878065): Only the dump variant of the types below are - // supported. - if hdr.Flags&linux.NLM_F_DUMP != linux.NLM_F_DUMP { - return syserr.ErrNotSupported - } - switch hdr.Type { case linux.RTM_GETLINK: return p.dumpLinks(ctx, hdr, data, ms) diff --git a/test/syscalls/linux/socket_netlink_route.cc b/test/syscalls/linux/socket_netlink_route.cc index ef567f512..1e28e658d 100644 --- a/test/syscalls/linux/socket_netlink_route.cc +++ b/test/syscalls/linux/socket_netlink_route.cc @@ -442,6 +442,90 @@ TEST(NetlinkRouteTest, GetRouteDump) { EXPECT_TRUE(dstFound); } +// GetRouteRequest tests a RTM_GETROUTE request with RTM_F_LOOKUP_TABLE flag. +TEST(NetlinkRouteTest, GetRouteRequest) { + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); + uint32_t port = ASSERT_NO_ERRNO_AND_VALUE(NetlinkPortID(fd.get())); + + struct __attribute__((__packed__)) request { + struct nlmsghdr hdr; + struct rtmsg rtm; + struct nlattr nla; + struct in_addr sin_addr; + }; + + constexpr uint32_t kSeq = 12345; + + struct request req = {}; + req.hdr.nlmsg_len = sizeof(req); + req.hdr.nlmsg_type = RTM_GETROUTE; + req.hdr.nlmsg_flags = NLM_F_REQUEST; + req.hdr.nlmsg_seq = kSeq; + + req.rtm.rtm_family = AF_INET; + req.rtm.rtm_dst_len = 32; + req.rtm.rtm_src_len = 0; + req.rtm.rtm_tos = 0; + req.rtm.rtm_table = RT_TABLE_UNSPEC; + req.rtm.rtm_protocol = RTPROT_UNSPEC; + req.rtm.rtm_scope = RT_SCOPE_UNIVERSE; + req.rtm.rtm_type = RTN_UNSPEC; + req.rtm.rtm_flags = RTM_F_LOOKUP_TABLE; + + req.nla.nla_len = 8; + req.nla.nla_type = RTA_DST; + inet_aton("127.0.0.2", &req.sin_addr); + + bool rtDstFound = false; + ASSERT_NO_ERRNO(NetlinkRequestResponseSingle( + fd, &req, sizeof(req), [&](const struct nlmsghdr* hdr) { + // Validate the reponse to RTM_GETROUTE request with RTM_F_LOOKUP_TABLE + // flag. + EXPECT_THAT(hdr->nlmsg_type, RTM_NEWROUTE); + + EXPECT_TRUE(hdr->nlmsg_flags == 0) << std::hex << hdr->nlmsg_flags; + + EXPECT_EQ(hdr->nlmsg_seq, kSeq); + EXPECT_EQ(hdr->nlmsg_pid, port); + + // RTM_NEWROUTE contains at least the header and rtmsg. + ASSERT_GE(hdr->nlmsg_len, NLMSG_SPACE(sizeof(struct rtmsg))); + const struct rtmsg* msg = + reinterpret_cast(NLMSG_DATA(hdr)); + + // NOTE: rtmsg fields are char fields. + std::cout << "Found route table=" << static_cast(msg->rtm_table) + << ", protocol=" << static_cast(msg->rtm_protocol) + << ", scope=" << static_cast(msg->rtm_scope) + << ", type=" << static_cast(msg->rtm_type); + + EXPECT_EQ(msg->rtm_family, AF_INET); + EXPECT_EQ(msg->rtm_dst_len, 32); + EXPECT_TRUE((msg->rtm_flags & RTM_F_CLONED) == RTM_F_CLONED) + << std::hex << msg->rtm_flags; + + int len = RTM_PAYLOAD(hdr); + std::cout << ", len=" << len; + for (struct rtattr* attr = RTM_RTA(msg); RTA_OK(attr, len); + attr = RTA_NEXT(attr, len)) { + if (attr->rta_type == RTA_DST) { + char address[INET_ADDRSTRLEN] = {}; + inet_ntop(AF_INET, RTA_DATA(attr), address, sizeof(address)); + std::cout << ", dst=" << address; + rtDstFound = true; + } else if (attr->rta_type == RTA_OIF) { + const char* oif = reinterpret_cast(RTA_DATA(attr)); + std::cout << ", oif=" << oif; + } + } + + std::cout << std::endl; + })); + // Found RTA_DST for RTM_F_LOOKUP_TABLE. + EXPECT_TRUE(rtDstFound); +} + // RecvmsgTrunc tests the recvmsg MSG_TRUNC flag with zero length output // buffer. MSG_TRUNC with a zero length buffer should consume subsequent // messages off the socket. diff --git a/test/syscalls/linux/socket_netlink_util.cc b/test/syscalls/linux/socket_netlink_util.cc index 723f5d728..cd2212a1a 100644 --- a/test/syscalls/linux/socket_netlink_util.cc +++ b/test/syscalls/linux/socket_netlink_util.cc @@ -108,5 +108,43 @@ PosixError NetlinkRequestResponse( return NoError(); } +PosixError NetlinkRequestResponseSingle( + const FileDescriptor& fd, void* request, size_t len, + const std::function& fn) { + struct iovec iov = {}; + iov.iov_base = request; + iov.iov_len = len; + + struct msghdr msg = {}; + msg.msg_iov = &iov; + msg.msg_iovlen = 1; + // No destination required; it defaults to pid 0, the kernel. + + RETURN_ERROR_IF_SYSCALL_FAIL(RetryEINTR(sendmsg)(fd.get(), &msg, 0)); + + constexpr size_t kBufferSize = 4096; + std::vector buf(kBufferSize); + iov.iov_base = buf.data(); + iov.iov_len = buf.size(); + + int ret; + RETURN_ERROR_IF_SYSCALL_FAIL(ret = RetryEINTR(recvmsg)(fd.get(), &msg, 0)); + + // We don't bother with the complexity of dealing with truncated messages. + // We must allocate a large enough buffer up front. + if ((msg.msg_flags & MSG_TRUNC) == MSG_TRUNC) { + return PosixError( + EIO, + absl::StrCat("Received truncated message with flags: ", msg.msg_flags)); + } + + for (struct nlmsghdr* hdr = reinterpret_cast(buf.data()); + NLMSG_OK(hdr, ret); hdr = NLMSG_NEXT(hdr, ret)) { + fn(hdr); + } + + return NoError(); +} + } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_netlink_util.h b/test/syscalls/linux/socket_netlink_util.h index 76e772c48..3678c0599 100644 --- a/test/syscalls/linux/socket_netlink_util.h +++ b/test/syscalls/linux/socket_netlink_util.h @@ -32,12 +32,21 @@ PosixErrorOr NetlinkBoundSocket(int protocol); // Returns the port ID of the passed socket. PosixErrorOr NetlinkPortID(int fd); -// Send the passed request and call fn will all response netlink messages. +// Send the passed request and call fn on all response netlink messages. +// +// To be used on requests with NLM_F_MULTI reponses. PosixError NetlinkRequestResponse( const FileDescriptor& fd, void* request, size_t len, const std::function& fn, bool expect_nlmsgerr); +// Send the passed request and call fn on all response netlink messages. +// +// To be used on requests without NLM_F_MULTI reponses. +PosixError NetlinkRequestResponseSingle( + const FileDescriptor& fd, void* request, size_t len, + const std::function& fn); + } // namespace testing } // namespace gvisor -- cgit v1.2.3 From 34fbd8446c386fb0136dad31ab6b173f17049a58 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Tue, 28 Jan 2020 13:10:41 -0800 Subject: Add VFS2 support for epoll. PiperOrigin-RevId: 291997879 --- pkg/abi/linux/epoll.go | 10 +- pkg/sentry/vfs/BUILD | 15 ++ pkg/sentry/vfs/epoll.go | 377 +++++++++++++++++++++++++++++++++++++ pkg/sentry/vfs/file_description.go | 38 ++++ pkg/sentry/vfs/vfs.go | 16 +- 5 files changed, 448 insertions(+), 8 deletions(-) create mode 100644 pkg/sentry/vfs/epoll.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/epoll.go b/pkg/abi/linux/epoll.go index 72083b604..0e881aa3c 100644 --- a/pkg/abi/linux/epoll.go +++ b/pkg/abi/linux/epoll.go @@ -38,8 +38,14 @@ const ( // Per-file descriptor flags. const ( - EPOLLET = 0x80000000 - EPOLLONESHOT = 0x40000000 + EPOLLEXCLUSIVE = 1 << 28 + EPOLLWAKEUP = 1 << 29 + EPOLLONESHOT = 1 << 30 + EPOLLET = 1 << 31 + + // EP_PRIVATE_BITS is fs/eventpoll.c:EP_PRIVATE_BITS, the set of all bits + // in an epoll event mask that correspond to flags rather than I/O events. + EP_PRIVATE_BITS = EPOLLEXCLUSIVE | EPOLLWAKEUP | EPOLLONESHOT | EPOLLET ) // Operation flags. diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index 33516e6f7..ced9d07b1 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -1,7 +1,20 @@ load("//tools:defs.bzl", "go_library", "go_test") +load("//tools/go_generics:defs.bzl", "go_template_instance") licenses(["notice"]) +go_template_instance( + name = "epoll_interest_list", + out = "epoll_interest_list.go", + package = "vfs", + prefix = "epollInterest", + template = "//pkg/ilist:generic_list", + types = { + "Element": "*epollInterest", + "Linker": "*epollInterest", + }, +) + go_library( name = "vfs", srcs = [ @@ -10,6 +23,8 @@ go_library( "debug.go", "dentry.go", "device.go", + "epoll.go", + "epoll_interest_list.go", "file_description.go", "file_description_impl_util.go", "filesystem.go", diff --git a/pkg/sentry/vfs/epoll.go b/pkg/sentry/vfs/epoll.go new file mode 100644 index 000000000..7c83f9a5a --- /dev/null +++ b/pkg/sentry/vfs/epoll.go @@ -0,0 +1,377 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/waiter" +) + +// epollCycleMu serializes attempts to register EpollInstances with other +// EpollInstances in order to check for cycles. +var epollCycleMu sync.Mutex + +// EpollInstance represents an epoll instance, as described by epoll(7). +type EpollInstance struct { + vfsfd FileDescription + FileDescriptionDefaultImpl + DentryMetadataFileDescriptionImpl + + // q holds waiters on this EpollInstance. + q waiter.Queue + + // interest is the set of file descriptors that are registered with the + // EpollInstance for monitoring. interest is protected by interestMu. + interestMu sync.Mutex + interest map[epollInterestKey]*epollInterest + + // mu protects fields in registered epollInterests. + mu sync.Mutex + + // ready is the set of file descriptors that may be "ready" for I/O. Note + // that this must be an ordered list, not a map: "If more than maxevents + // file descriptors are ready when epoll_wait() is called, then successive + // epoll_wait() calls will round robin through the set of ready file + // descriptors. This behavior helps avoid starvation scenarios, where a + // process fails to notice that additional file descriptors are ready + // because it focuses on a set of file descriptors that are already known + // to be ready." - epoll_wait(2) + ready epollInterestList +} + +type epollInterestKey struct { + // file is the registered FileDescription. No reference is held on file; + // instead, when the last reference is dropped, FileDescription.DecRef() + // removes the FileDescription from all EpollInstances. file is immutable. + file *FileDescription + + // num is the file descriptor number with which this entry was registered. + // num is immutable. + num int32 +} + +// epollInterest represents an EpollInstance's interest in a file descriptor. +type epollInterest struct { + // epoll is the owning EpollInstance. epoll is immutable. + epoll *EpollInstance + + // key is the file to which this epollInterest applies. key is immutable. + key epollInterestKey + + // waiter is registered with key.file. entry is protected by epoll.mu. + waiter waiter.Entry + + // mask is the event mask associated with this registration, including + // flags EPOLLET and EPOLLONESHOT. mask is protected by epoll.mu. + mask uint32 + + // ready is true if epollInterestEntry is linked into epoll.ready. ready + // and epollInterestEntry are protected by epoll.mu. + ready bool + epollInterestEntry + + // userData is the epoll_data_t associated with this epollInterest. + // userData is protected by epoll.mu. + userData [2]int32 +} + +// NewEpollInstanceFD returns a FileDescription representing a new epoll +// instance. A reference is taken on the returned FileDescription. +func (vfs *VirtualFilesystem) NewEpollInstanceFD() (*FileDescription, error) { + vd := vfs.NewAnonVirtualDentry("[eventpoll]") + defer vd.DecRef() + ep := &EpollInstance{ + interest: make(map[epollInterestKey]*epollInterest), + } + if err := ep.vfsfd.Init(ep, linux.O_RDWR, vd.Mount(), vd.Dentry(), &FileDescriptionOptions{ + UseDentryMetadata: true, + }); err != nil { + return nil, err + } + return &ep.vfsfd, nil +} + +// Release implements FileDescriptionImpl.Release. +func (ep *EpollInstance) Release() { + // Unregister all polled fds. + ep.interestMu.Lock() + defer ep.interestMu.Unlock() + for key, epi := range ep.interest { + file := key.file + file.epollMu.Lock() + delete(file.epolls, epi) + file.epollMu.Unlock() + file.EventUnregister(&epi.waiter) + } + ep.interest = nil +} + +// Readiness implements waiter.Waitable.Readiness. +func (ep *EpollInstance) Readiness(mask waiter.EventMask) waiter.EventMask { + if mask&waiter.EventIn == 0 { + return 0 + } + ep.mu.Lock() + for epi := ep.ready.Front(); epi != nil; epi = epi.Next() { + wmask := waiter.EventMaskFromLinux(epi.mask) + if epi.key.file.Readiness(wmask)&wmask != 0 { + ep.mu.Unlock() + return waiter.EventIn + } + } + ep.mu.Unlock() + return 0 +} + +// EventRegister implements waiter.Waitable.EventRegister. +func (ep *EpollInstance) EventRegister(e *waiter.Entry, mask waiter.EventMask) { + ep.q.EventRegister(e, mask) +} + +// EventUnregister implements waiter.Waitable.EventUnregister. +func (ep *EpollInstance) EventUnregister(e *waiter.Entry) { + ep.q.EventUnregister(e) +} + +// Seek implements FileDescriptionImpl.Seek. +func (ep *EpollInstance) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + // Linux: fs/eventpoll.c:eventpoll_fops.llseek == noop_llseek + return 0, nil +} + +// AddInterest implements the semantics of EPOLL_CTL_ADD. +// +// Preconditions: A reference must be held on file. +func (ep *EpollInstance) AddInterest(file *FileDescription, num int32, mask uint32, userData [2]int32) error { + // Check for cyclic polling if necessary. + subep, _ := file.impl.(*EpollInstance) + if subep != nil { + epollCycleMu.Lock() + // epollCycleMu must be locked for the rest of AddInterest to ensure + // that cyclic polling is not introduced after the check. + defer epollCycleMu.Unlock() + if subep.mightPoll(ep) { + return syserror.ELOOP + } + } + + ep.interestMu.Lock() + defer ep.interestMu.Unlock() + + // Fail if the key is already registered. + key := epollInterestKey{ + file: file, + num: num, + } + if _, ok := ep.interest[key]; ok { + return syserror.EEXIST + } + + // Register interest in file. + mask |= linux.EPOLLERR | linux.EPOLLRDHUP + epi := &epollInterest{ + epoll: ep, + key: key, + mask: mask, + userData: userData, + } + ep.interest[key] = epi + wmask := waiter.EventMaskFromLinux(mask) + file.EventRegister(&epi.waiter, wmask) + + // Check if the file is already ready. + if file.Readiness(wmask)&wmask != 0 { + epi.Callback(nil) + } + + // Add epi to file.epolls so that it is removed when the last + // FileDescription reference is dropped. + file.epollMu.Lock() + file.epolls[epi] = struct{}{} + file.epollMu.Unlock() + + return nil +} + +func (ep *EpollInstance) mightPoll(ep2 *EpollInstance) bool { + return ep.mightPollRecursive(ep2, 4) // Linux: fs/eventpoll.c:EP_MAX_NESTS +} + +func (ep *EpollInstance) mightPollRecursive(ep2 *EpollInstance, remainingRecursion int) bool { + ep.interestMu.Lock() + defer ep.interestMu.Unlock() + for key := range ep.interest { + nextep, ok := key.file.impl.(*EpollInstance) + if !ok { + continue + } + if nextep == ep2 { + return true + } + if remainingRecursion == 0 { + return true + } + if nextep.mightPollRecursive(ep2, remainingRecursion-1) { + return true + } + } + return false +} + +// ModifyInterest implements the semantics of EPOLL_CTL_MOD. +// +// Preconditions: A reference must be held on file. +func (ep *EpollInstance) ModifyInterest(file *FileDescription, num int32, mask uint32, userData [2]int32) error { + ep.interestMu.Lock() + defer ep.interestMu.Unlock() + + // Fail if the key is not already registered. + epi, ok := ep.interest[epollInterestKey{ + file: file, + num: num, + }] + if !ok { + return syserror.ENOENT + } + + // Update epi for the next call to ep.ReadEvents(). + ep.mu.Lock() + epi.mask = mask + epi.userData = userData + ep.mu.Unlock() + + // Re-register with the new mask. + mask |= linux.EPOLLERR | linux.EPOLLRDHUP + file.EventUnregister(&epi.waiter) + wmask := waiter.EventMaskFromLinux(mask) + file.EventRegister(&epi.waiter, wmask) + + // Check if the file is already ready with the new mask. + if file.Readiness(wmask)&wmask != 0 { + epi.Callback(nil) + } + + return nil +} + +// DeleteInterest implements the semantics of EPOLL_CTL_DEL. +// +// Preconditions: A reference must be held on file. +func (ep *EpollInstance) DeleteInterest(file *FileDescription, num int32) error { + ep.interestMu.Lock() + defer ep.interestMu.Unlock() + + // Fail if the key is not already registered. + epi, ok := ep.interest[epollInterestKey{ + file: file, + num: num, + }] + if !ok { + return syserror.ENOENT + } + + // Unregister from the file so that epi will no longer be readied. + file.EventUnregister(&epi.waiter) + + // Forget about epi. + ep.removeLocked(epi) + + file.epollMu.Lock() + delete(file.epolls, epi) + file.epollMu.Unlock() + + return nil +} + +// Callback implements waiter.EntryCallback.Callback. +func (epi *epollInterest) Callback(*waiter.Entry) { + newReady := false + epi.epoll.mu.Lock() + if !epi.ready { + newReady = true + epi.ready = true + epi.epoll.ready.PushBack(epi) + } + epi.epoll.mu.Unlock() + if newReady { + epi.epoll.q.Notify(waiter.EventIn) + } +} + +// Preconditions: ep.interestMu must be locked. +func (ep *EpollInstance) removeLocked(epi *epollInterest) { + delete(ep.interest, epi.key) + ep.mu.Lock() + if epi.ready { + epi.ready = false + ep.ready.Remove(epi) + } + ep.mu.Unlock() +} + +// ReadEvents reads up to len(events) ready events into events and returns the +// number of events read. +// +// Preconditions: len(events) != 0. +func (ep *EpollInstance) ReadEvents(events []linux.EpollEvent) int { + i := 0 + // Hot path: avoid defer. + ep.mu.Lock() + var next *epollInterest + var requeue epollInterestList + for epi := ep.ready.Front(); epi != nil; epi = next { + next = epi.Next() + // Regardless of what else happens, epi is initially removed from the + // ready list. + ep.ready.Remove(epi) + wmask := waiter.EventMaskFromLinux(epi.mask) + ievents := epi.key.file.Readiness(wmask) & wmask + if ievents == 0 { + // Leave epi off the ready list. + epi.ready = false + continue + } + // Determine what we should do with epi. + switch { + case epi.mask&linux.EPOLLONESHOT != 0: + // Clear all events from the mask; they must be re-added by + // EPOLL_CTL_MOD. + epi.mask &= linux.EP_PRIVATE_BITS + fallthrough + case epi.mask&linux.EPOLLET != 0: + // Leave epi off the ready list. + epi.ready = false + default: + // Queue epi to be moved to the end of the ready list. + requeue.PushBack(epi) + } + // Report ievents. + events[i] = linux.EpollEvent{ + Events: ievents.ToLinux(), + Fd: epi.userData[0], + Data: epi.userData[1], + } + i++ + if i == len(events) { + break + } + } + ep.ready.PushBackList(&requeue) + ep.mu.Unlock() + return i +} diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 225024463..badacb55e 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -22,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/memmap" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" @@ -45,6 +46,11 @@ type FileDescription struct { // memory operations. statusFlags uint32 + // epolls is the set of epollInterests registered for this FileDescription. + // epolls is protected by epollMu. + epollMu sync.Mutex + epolls map[*epollInterest]struct{} + // vd is the filesystem location at which this FileDescription was opened. // A reference is held on vd. vd is immutable. vd VirtualDentry @@ -141,6 +147,23 @@ func (fd *FileDescription) TryIncRef() bool { // DecRef decrements fd's reference count. func (fd *FileDescription) DecRef() { if refs := atomic.AddInt64(&fd.refs, -1); refs == 0 { + // Unregister fd from all epoll instances. + fd.epollMu.Lock() + epolls := fd.epolls + fd.epolls = nil + fd.epollMu.Unlock() + for epi := range epolls { + ep := epi.epoll + ep.interestMu.Lock() + // Check that epi has not been concurrently unregistered by + // EpollInstance.DeleteInterest() or EpollInstance.Release(). + if _, ok := ep.interest[epi.key]; ok { + fd.EventUnregister(&epi.waiter) + ep.removeLocked(epi) + } + ep.interestMu.Unlock() + } + // Release implementation resources. fd.impl.Release() if fd.writable { fd.vd.mount.EndWrite() @@ -453,6 +476,21 @@ func (fd *FileDescription) StatFS(ctx context.Context) (linux.Statfs, error) { return fd.impl.StatFS(ctx) } +// Readiness returns fd's I/O readiness. +func (fd *FileDescription) Readiness(mask waiter.EventMask) waiter.EventMask { + return fd.impl.Readiness(mask) +} + +// EventRegister registers e for I/O readiness events in mask. +func (fd *FileDescription) EventRegister(e *waiter.Entry, mask waiter.EventMask) { + fd.impl.EventRegister(e, mask) +} + +// EventUnregister unregisters e for I/O readiness events. +func (fd *FileDescription) EventUnregister(e *waiter.Entry) { + fd.impl.EventUnregister(e) +} + // PRead reads from the file represented by fd into dst, starting at the given // offset, and returns the number of bytes read. PRead is permitted to return // partial reads with a nil error. diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index d730530b9..908c69f91 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -16,15 +16,19 @@ // // Lock order: // -// FilesystemImpl/FileDescriptionImpl locks -// VirtualFilesystem.mountMu -// Dentry.mu -// Locks acquired by FilesystemImpls between Prepare{Delete,Rename}Dentry and Commit{Delete,Rename*}Dentry -// VirtualFilesystem.filesystemsMu +// EpollInstance.interestMu +// FileDescription.epollMu +// FilesystemImpl/FileDescriptionImpl locks +// VirtualFilesystem.mountMu +// Dentry.mu +// Locks acquired by FilesystemImpls between Prepare{Delete,Rename}Dentry and Commit{Delete,Rename*}Dentry +// VirtualFilesystem.filesystemsMu +// EpollInstance.mu // VirtualFilesystem.fsTypesMu // // Locking Dentry.mu in multiple Dentries requires holding -// VirtualFilesystem.mountMu. +// VirtualFilesystem.mountMu. Locking EpollInstance.interestMu in multiple +// EpollInstances requires holding epollCycleMu. package vfs import ( -- cgit v1.2.3 From f263801a74d4ccac042b068d0928c8738e40af5b Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Tue, 28 Jan 2020 13:36:16 -0800 Subject: fs/splice: don't report partial errors for special files Special files can have additional requirements for granularity. For example, read from eventfd returns EINVAL if a size is less 8 bytes. Reported-by: syzbot+3905f5493bec08eb7b02@syzkaller.appspotmail.com PiperOrigin-RevId: 292002926 --- pkg/sentry/fs/attr.go | 5 +++++ pkg/sentry/fs/file.go | 7 ------- pkg/sentry/fs/splice.go | 5 ----- pkg/sentry/syscalls/linux/sys_splice.go | 19 +++++++++++++++---- test/syscalls/linux/eventfd.cc | 25 +++++++++++++++++++++++++ 5 files changed, 45 insertions(+), 16 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/attr.go b/pkg/sentry/fs/attr.go index fa9e7d517..f60bd423d 100644 --- a/pkg/sentry/fs/attr.go +++ b/pkg/sentry/fs/attr.go @@ -206,6 +206,11 @@ func IsPipe(s StableAttr) bool { return s.Type == Pipe } +// IsAnonymous returns true if StableAttr.Type matches any type of anonymous. +func IsAnonymous(s StableAttr) bool { + return s.Type == Anonymous +} + // IsSocket returns true if StableAttr.Type matches any type of socket. func IsSocket(s StableAttr) bool { return s.Type == Socket diff --git a/pkg/sentry/fs/file.go b/pkg/sentry/fs/file.go index ca3466f4f..78100e448 100644 --- a/pkg/sentry/fs/file.go +++ b/pkg/sentry/fs/file.go @@ -555,10 +555,6 @@ type lockedWriter struct { // // This applies only to Write, not WriteAt. Offset int64 - - // Err contains the first error encountered while copying. This is - // useful to determine whether Writer or Reader failed during io.Copy. - Err error } // Write implements io.Writer.Write. @@ -594,8 +590,5 @@ func (w *lockedWriter) WriteAt(buf []byte, offset int64) (int, error) { break } } - if w.Err == nil { - w.Err = err - } return written, err } diff --git a/pkg/sentry/fs/splice.go b/pkg/sentry/fs/splice.go index 791d1526c..33da82868 100644 --- a/pkg/sentry/fs/splice.go +++ b/pkg/sentry/fs/splice.go @@ -167,11 +167,6 @@ func Splice(ctx context.Context, dst *File, src *File, opts SpliceOpts) (int64, if !srcPipe && !opts.SrcOffset { atomic.StoreInt64(&src.offset, src.offset+n) } - - // Don't report any errors if we have some progress without data loss. - if w.Err == nil { - err = nil - } } // Drop locks. diff --git a/pkg/sentry/syscalls/linux/sys_splice.go b/pkg/sentry/syscalls/linux/sys_splice.go index dd3a5807f..f43d6c155 100644 --- a/pkg/sentry/syscalls/linux/sys_splice.go +++ b/pkg/sentry/syscalls/linux/sys_splice.go @@ -211,8 +211,10 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal opts := fs.SpliceOpts{ Length: count, } + inFileAttr := inFile.Dirent.Inode.StableAttr + outFileAttr := outFile.Dirent.Inode.StableAttr switch { - case fs.IsPipe(inFile.Dirent.Inode.StableAttr) && !fs.IsPipe(outFile.Dirent.Inode.StableAttr): + case fs.IsPipe(inFileAttr) && !fs.IsPipe(outFileAttr): if inOffset != 0 { return 0, nil, syserror.ESPIPE } @@ -229,7 +231,7 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal opts.DstOffset = true opts.DstStart = offset } - case !fs.IsPipe(inFile.Dirent.Inode.StableAttr) && fs.IsPipe(outFile.Dirent.Inode.StableAttr): + case !fs.IsPipe(inFileAttr) && fs.IsPipe(outFileAttr): if outOffset != 0 { return 0, nil, syserror.ESPIPE } @@ -246,13 +248,13 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal opts.SrcOffset = true opts.SrcStart = offset } - case fs.IsPipe(inFile.Dirent.Inode.StableAttr) && fs.IsPipe(outFile.Dirent.Inode.StableAttr): + case fs.IsPipe(inFileAttr) && fs.IsPipe(outFileAttr): if inOffset != 0 || outOffset != 0 { return 0, nil, syserror.ESPIPE } // We may not refer to the same pipe; otherwise it's a continuous loop. - if inFile.Dirent.Inode.StableAttr.InodeID == outFile.Dirent.Inode.StableAttr.InodeID { + if inFileAttr.InodeID == outFileAttr.InodeID { return 0, nil, syserror.EINVAL } default: @@ -262,6 +264,15 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal // Splice data. n, err := doSplice(t, outFile, inFile, opts, nonBlock) + // Special files can have additional requirements for granularity. For + // example, read from eventfd returns EINVAL if a size is less 8 bytes. + // Inotify is another example. read will return EINVAL is a buffer is + // too small to return the next event, but a size of an event isn't + // fixed, it is sizeof(struct inotify_event) + {NAME_LEN} + 1. + if n != 0 && err != nil && (fs.IsAnonymous(inFileAttr) || fs.IsAnonymous(outFileAttr)) { + err = nil + } + // See above; inFile is chosen arbitrarily here. return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "splice", inFile) } diff --git a/test/syscalls/linux/eventfd.cc b/test/syscalls/linux/eventfd.cc index 367682c3d..927001eee 100644 --- a/test/syscalls/linux/eventfd.cc +++ b/test/syscalls/linux/eventfd.cc @@ -132,6 +132,31 @@ TEST(EventfdTest, BigWriteBigRead) { EXPECT_EQ(l[0], 1); } +TEST(EventfdTest, SpliceFromPipePartialSucceeds) { + int pipes[2]; + ASSERT_THAT(pipe2(pipes, O_NONBLOCK), SyscallSucceeds()); + const FileDescriptor pipe_rfd(pipes[0]); + const FileDescriptor pipe_wfd(pipes[1]); + constexpr uint64_t kVal{1}; + + FileDescriptor efd = ASSERT_NO_ERRNO_AND_VALUE(NewEventFD(0, EFD_NONBLOCK)); + + uint64_t event_array[2]; + event_array[0] = kVal; + event_array[1] = kVal; + ASSERT_THAT(write(pipe_wfd.get(), event_array, sizeof(event_array)), + SyscallSucceedsWithValue(sizeof(event_array))); + EXPECT_THAT(splice(pipe_rfd.get(), /*__offin=*/nullptr, efd.get(), + /*__offout=*/nullptr, sizeof(event_array[0]) + 1, + SPLICE_F_NONBLOCK), + SyscallSucceedsWithValue(sizeof(event_array[0]))); + + uint64_t val; + ASSERT_THAT(read(efd.get(), &val, sizeof(val)), + SyscallSucceedsWithValue(sizeof(val))); + EXPECT_EQ(val, kVal); +} + // NotifyNonZero is inherently racy, so random save is disabled. TEST(EventfdTest, NotifyNonZero_NoRandomSave) { // Waits will time out at 10 seconds. -- cgit v1.2.3 From 2862b0b1be9ce821e86877802b9608aad3102916 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Tue, 28 Jan 2020 15:04:34 -0800 Subject: Add //pkg/sentry/fsimpl/devtmpfs. PiperOrigin-RevId: 292021389 --- pkg/sentry/fsimpl/devtmpfs/BUILD | 33 +++++ pkg/sentry/fsimpl/devtmpfs/devtmpfs.go | 187 ++++++++++++++++++++++++++++ pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go | 119 ++++++++++++++++++ pkg/sentry/fsimpl/tmpfs/BUILD | 1 + 4 files changed, 340 insertions(+) create mode 100644 pkg/sentry/fsimpl/devtmpfs/BUILD create mode 100644 pkg/sentry/fsimpl/devtmpfs/devtmpfs.go create mode 100644 pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/devtmpfs/BUILD b/pkg/sentry/fsimpl/devtmpfs/BUILD new file mode 100644 index 000000000..aa0c2ad8c --- /dev/null +++ b/pkg/sentry/fsimpl/devtmpfs/BUILD @@ -0,0 +1,33 @@ +load("//tools:defs.bzl", "go_library", "go_test") + +licenses(["notice"]) + +go_library( + name = "devtmpfs", + srcs = ["devtmpfs.go"], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/abi/linux", + "//pkg/context", + "//pkg/fspath", + "//pkg/sentry/fsimpl/tmpfs", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/vfs", + "//pkg/sync", + ], +) + +go_test( + name = "devtmpfs_test", + size = "small", + srcs = ["devtmpfs_test.go"], + library = ":devtmpfs", + deps = [ + "//pkg/abi/linux", + "//pkg/fspath", + "//pkg/sentry/contexttest", + "//pkg/sentry/fsimpl/tmpfs", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/vfs", + ], +) diff --git a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go new file mode 100644 index 000000000..d36fa74fb --- /dev/null +++ b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go @@ -0,0 +1,187 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package devtmpfs provides an implementation of /dev based on tmpfs, +// analogous to Linux's devtmpfs. +package devtmpfs + +import ( + "fmt" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sync" +) + +// FilesystemType implements vfs.FilesystemType. +type FilesystemType struct { + initOnce sync.Once + initErr error + + // fs is the tmpfs filesystem that backs all mounts of this FilesystemType. + // root is fs' root. fs and root are immutable. + fs *vfs.Filesystem + root *vfs.Dentry +} + +// GetFilesystem implements vfs.FilesystemType.GetFilesystem. +func (fst *FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { + fst.initOnce.Do(func() { + fs, root, err := tmpfs.FilesystemType{}.GetFilesystem(ctx, vfsObj, creds, "" /* source */, vfs.GetFilesystemOptions{ + Data: "mode=0755", // opts from drivers/base/devtmpfs.c:devtmpfs_init() + }) + if err != nil { + fst.initErr = err + return + } + fst.fs = fs + fst.root = root + }) + if fst.initErr != nil { + return nil, nil, fst.initErr + } + fst.fs.IncRef() + fst.root.IncRef() + return fst.fs, fst.root, nil +} + +// Accessor allows devices to create device special files in devtmpfs. +type Accessor struct { + vfsObj *vfs.VirtualFilesystem + mntns *vfs.MountNamespace + root vfs.VirtualDentry + creds *auth.Credentials +} + +// NewAccessor returns an Accessor that supports creation of device special +// files in the devtmpfs instance registered with name fsTypeName in vfsObj. +func NewAccessor(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, fsTypeName string) (*Accessor, error) { + mntns, err := vfsObj.NewMountNamespace(ctx, creds, "devtmpfs" /* source */, fsTypeName, &vfs.GetFilesystemOptions{}) + if err != nil { + return nil, err + } + return &Accessor{ + vfsObj: vfsObj, + mntns: mntns, + root: mntns.Root(), + creds: creds, + }, nil +} + +// Release must be called when a is no longer in use. +func (a *Accessor) Release() { + a.root.DecRef() + a.mntns.DecRef(a.vfsObj) +} + +// accessorContext implements context.Context by extending an existing +// context.Context with an Accessor's values for VFS-relevant state. +type accessorContext struct { + context.Context + a *Accessor +} + +func (a *Accessor) wrapContext(ctx context.Context) *accessorContext { + return &accessorContext{ + Context: ctx, + a: a, + } +} + +// Value implements context.Context.Value. +func (ac *accessorContext) Value(key interface{}) interface{} { + switch key { + case vfs.CtxMountNamespace: + return ac.a.mntns + case vfs.CtxRoot: + ac.a.root.IncRef() + return ac.a.root + default: + return ac.Context.Value(key) + } +} + +func (a *Accessor) pathOperationAt(pathname string) *vfs.PathOperation { + return &vfs.PathOperation{ + Root: a.root, + Start: a.root, + Path: fspath.Parse(pathname), + } +} + +// CreateDeviceFile creates a device special file at the given pathname in the +// devtmpfs instance accessed by the Accessor. +func (a *Accessor) CreateDeviceFile(ctx context.Context, pathname string, kind vfs.DeviceKind, major, minor uint32, perms uint16) error { + mode := (linux.FileMode)(perms) + switch kind { + case vfs.BlockDevice: + mode |= linux.S_IFBLK + case vfs.CharDevice: + mode |= linux.S_IFCHR + default: + panic(fmt.Sprintf("invalid vfs.DeviceKind: %v", kind)) + } + // NOTE: Linux's devtmpfs refuses to automatically delete files it didn't + // create, which it recognizes by storing a pointer to the kdevtmpfs struct + // thread in struct inode::i_private. Accessor doesn't yet support deletion + // of files at all, and probably won't as long as we don't need to support + // kernel modules, so this is moot for now. + return a.vfsObj.MknodAt(a.wrapContext(ctx), a.creds, a.pathOperationAt(pathname), &vfs.MknodOptions{ + Mode: mode, + DevMajor: major, + DevMinor: minor, + }) +} + +// UserspaceInit creates symbolic links and mount points in the devtmpfs +// instance accessed by the Accessor that are created by userspace in Linux. It +// does not create mounts. +func (a *Accessor) UserspaceInit(ctx context.Context) error { + actx := a.wrapContext(ctx) + + // systemd: src/shared/dev-setup.c:dev_setup() + for _, symlink := range []struct { + source string + target string + }{ + // /proc/kcore is not implemented. + {source: "fd", target: "/proc/self/fd"}, + {source: "stdin", target: "/proc/self/fd/0"}, + {source: "stdout", target: "/proc/self/fd/1"}, + {source: "stderr", target: "/proc/self/fd/2"}, + } { + if err := a.vfsObj.SymlinkAt(actx, a.creds, a.pathOperationAt(symlink.source), symlink.target); err != nil { + return fmt.Errorf("failed to create symlink %q => %q: %v", symlink.source, symlink.target, err) + } + } + + // systemd: src/core/mount-setup.c:mount_table + for _, dir := range []string{ + "shm", + "pts", + } { + if err := a.vfsObj.MkdirAt(actx, a.creds, a.pathOperationAt(dir), &vfs.MkdirOptions{ + // systemd: src/core/mount-setup.c:mount_one() + Mode: 0755, + }); err != nil { + return fmt.Errorf("failed to create directory %q: %v", dir, err) + } + } + + return nil +} diff --git a/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go b/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go new file mode 100644 index 000000000..82c58c900 --- /dev/null +++ b/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go @@ -0,0 +1,119 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package devtmpfs + +import ( + "testing" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/contexttest" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" +) + +func TestDevtmpfs(t *testing.T) { + ctx := contexttest.Context(t) + creds := auth.CredentialsFromContext(ctx) + + vfsObj := vfs.New() + // Register tmpfs just so that we can have a root filesystem that isn't + // devtmpfs. + vfsObj.MustRegisterFilesystemType("tmpfs", tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + }) + vfsObj.MustRegisterFilesystemType("devtmpfs", &FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + }) + + // Create a test mount namespace with devtmpfs mounted at "/dev". + const devPath = "/dev" + mntns, err := vfsObj.NewMountNamespace(ctx, creds, "tmpfs" /* source */, "tmpfs" /* fsTypeName */, &vfs.GetFilesystemOptions{}) + if err != nil { + t.Fatalf("failed to create tmpfs root mount: %v", err) + } + defer mntns.DecRef(vfsObj) + root := mntns.Root() + defer root.DecRef() + devpop := vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(devPath), + } + if err := vfsObj.MkdirAt(ctx, creds, &devpop, &vfs.MkdirOptions{ + Mode: 0755, + }); err != nil { + t.Fatalf("failed to create mount point: %v", err) + } + if err := vfsObj.MountAt(ctx, creds, "devtmpfs" /* source */, &devpop, "devtmpfs" /* fsTypeName */, &vfs.MountOptions{}); err != nil { + t.Fatalf("failed to mount devtmpfs: %v", err) + } + + a, err := NewAccessor(ctx, vfsObj, creds, "devtmpfs") + if err != nil { + t.Fatalf("failed to create devtmpfs.Accessor: %v", err) + } + defer a.Release() + + // Create "userspace-initialized" files using a devtmpfs.Accessor. + if err := a.UserspaceInit(ctx); err != nil { + t.Fatalf("failed to userspace-initialize devtmpfs: %v", err) + } + // Created files should be visible in the test mount namespace. + abspath := devPath + "/fd" + target, err := vfsObj.ReadlinkAt(ctx, creds, &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(abspath), + }) + if want := "/proc/self/fd"; err != nil || target != want { + t.Fatalf("readlink(%q): got (%q, %v), wanted (%q, nil)", abspath, target, err, want) + } + + // Create a dummy device special file using a devtmpfs.Accessor. + const ( + pathInDev = "dummy" + kind = vfs.CharDevice + major = 12 + minor = 34 + perms = 0600 + wantMode = linux.S_IFCHR | perms + ) + if err := a.CreateDeviceFile(ctx, pathInDev, kind, major, minor, perms); err != nil { + t.Fatalf("failed to create device file: %v", err) + } + // The device special file should be visible in the test mount namespace. + abspath = devPath + "/" + pathInDev + stat, err := vfsObj.StatAt(ctx, creds, &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(abspath), + }, &vfs.StatOptions{ + Mask: linux.STATX_TYPE | linux.STATX_MODE, + }) + if err != nil { + t.Fatalf("failed to stat device file at %q: %v", abspath, err) + } + if stat.Mode != wantMode { + t.Errorf("device file mode: got %v, wanted %v", stat.Mode, wantMode) + } + if stat.RdevMajor != major { + t.Errorf("major device number: got %v, wanted %v", stat.RdevMajor, major) + } + if stat.RdevMinor != minor { + t.Errorf("minor device number: got %v, wanted %v", stat.RdevMinor, minor) + } +} diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD index fb436860c..c61366224 100644 --- a/pkg/sentry/fsimpl/tmpfs/BUILD +++ b/pkg/sentry/fsimpl/tmpfs/BUILD @@ -27,6 +27,7 @@ go_library( "symlink.go", "tmpfs.go", ], + visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", "//pkg/amutex", -- cgit v1.2.3 From 437c986c6a0ed0e1fccfbfb6706f43d2c801c444 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Tue, 28 Jan 2020 15:13:46 -0800 Subject: Add vfs.FileDescription to FD table FD table now holds both VFS1 and VFS2 types and uses the correct one based on what's set. Parts of this CL are just initial changes (e.g. sys_read.go, runsc/main.go) to serve as a template for the remaining changes. Updates #1487 Updates #1623 PiperOrigin-RevId: 292023223 --- pkg/sentry/kernel/BUILD | 1 + pkg/sentry/kernel/fd_table.go | 166 ++++++++++++++++----- pkg/sentry/kernel/fd_table_test.go | 4 +- pkg/sentry/kernel/fd_table_unsafe.go | 98 ++++++++++-- pkg/sentry/kernel/kernel.go | 31 ++-- pkg/sentry/kernel/task.go | 9 ++ pkg/sentry/kernel/task_exec.go | 3 +- pkg/sentry/syscalls/linux/BUILD | 1 + pkg/sentry/syscalls/linux/error.go | 72 ++++++--- pkg/sentry/syscalls/linux/sys_file.go | 2 +- pkg/sentry/syscalls/linux/vfs2/BUILD | 24 +++ pkg/sentry/syscalls/linux/vfs2/linux64.go | 16 ++ .../syscalls/linux/vfs2/linux64_override_amd64.go | 25 ++++ .../syscalls/linux/vfs2/linux64_override_arm64.go | 25 ++++ pkg/sentry/syscalls/linux/vfs2/sys_read.go | 95 ++++++++++++ runsc/boot/BUILD | 1 + runsc/boot/config.go | 3 + runsc/boot/loader.go | 9 ++ 18 files changed, 496 insertions(+), 89 deletions(-) create mode 100644 pkg/sentry/syscalls/linux/vfs2/BUILD create mode 100644 pkg/sentry/syscalls/linux/vfs2/linux64.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/linux64_override_arm64.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/sys_read.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD index 0738946d9..a27628c0a 100644 --- a/pkg/sentry/kernel/BUILD +++ b/pkg/sentry/kernel/BUILD @@ -188,6 +188,7 @@ go_library( "//pkg/sentry/unimpl:unimplemented_syscall_go_proto", "//pkg/sentry/uniqueid", "//pkg/sentry/usage", + "//pkg/sentry/vfs", "//pkg/state", "//pkg/state/statefile", "//pkg/sync", diff --git a/pkg/sentry/kernel/fd_table.go b/pkg/sentry/kernel/fd_table.go index 9460bb235..56b70ce96 100644 --- a/pkg/sentry/kernel/fd_table.go +++ b/pkg/sentry/kernel/fd_table.go @@ -27,6 +27,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/limits" + "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" ) @@ -62,10 +63,14 @@ func (f FDFlags) ToLinuxFDFlags() (mask uint) { // Note that this is immutable and can only be changed via operations on the // descriptorTable. // +// It contains both VFS1 and VFS2 file types, but only one of them can be set. +// // +stateify savable type descriptor struct { - file *fs.File - flags FDFlags + // TODO(gvisor.dev/issue/1624): Remove fs.File. + file *fs.File + fileVFS2 *vfs.FileDescription + flags FDFlags } // FDTable is used to manage File references and flags. @@ -95,10 +100,11 @@ type FDTable struct { func (f *FDTable) saveDescriptorTable() map[int32]descriptor { m := make(map[int32]descriptor) - f.forEach(func(fd int32, file *fs.File, flags FDFlags) { + f.forEach(func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { m[fd] = descriptor{ - file: file, - flags: flags, + file: file, + fileVFS2: fileVFS2, + flags: flags, } }) return m @@ -107,13 +113,17 @@ func (f *FDTable) saveDescriptorTable() map[int32]descriptor { func (f *FDTable) loadDescriptorTable(m map[int32]descriptor) { f.init() // Initialize table. for fd, d := range m { - f.set(fd, d.file, d.flags) - - // Note that we do _not_ need to acquire a extra table - // reference here. The table reference will already be - // accounted for in the file, so we drop the reference taken by - // set above. - d.file.DecRef() + f.setAll(fd, d.file, d.fileVFS2, d.flags) + + // Note that we do _not_ need to acquire a extra table reference here. The + // table reference will already be accounted for in the file, so we drop the + // reference taken by set above. + switch { + case d.file != nil: + d.file.DecRef() + case d.fileVFS2 != nil: + d.fileVFS2.DecRef() + } } } @@ -139,6 +149,15 @@ func (f *FDTable) drop(file *fs.File) { file.DecRef() } +// dropVFS2 drops the table reference. +func (f *FDTable) dropVFS2(file *vfs.FileDescription) { + // TODO(gvisor.dev/issue/1480): Release locks. + // TODO(gvisor.dev/issue/1479): Send inotify events. + + // Drop the table reference. + file.DecRef() +} + // ID returns a unique identifier for this FDTable. func (f *FDTable) ID() uint64 { return f.uid @@ -156,7 +175,7 @@ func (k *Kernel) NewFDTable() *FDTable { // destroy removes all of the file descriptors from the map. func (f *FDTable) destroy() { - f.RemoveIf(func(*fs.File, FDFlags) bool { + f.RemoveIf(func(*fs.File, *vfs.FileDescription, FDFlags) bool { return true }) } @@ -175,19 +194,26 @@ func (f *FDTable) Size() int { // forEach iterates over all non-nil files. // // It is the caller's responsibility to acquire an appropriate lock. -func (f *FDTable) forEach(fn func(fd int32, file *fs.File, flags FDFlags)) { +func (f *FDTable) forEach(fn func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags)) { fd := int32(0) for { - file, flags, ok := f.get(fd) + file, fileVFS2, flags, ok := f.getAll(fd) if !ok { break } - if file != nil { + switch { + case file != nil: if !file.TryIncRef() { continue // Race caught. } - fn(int32(fd), file, flags) + fn(fd, file, nil, flags) file.DecRef() + case fileVFS2 != nil: + if !fileVFS2.TryIncRef() { + continue // Race caught. + } + fn(fd, nil, fileVFS2, flags) + fileVFS2.DecRef() } fd++ } @@ -196,9 +222,21 @@ func (f *FDTable) forEach(fn func(fd int32, file *fs.File, flags FDFlags)) { // String is a stringer for FDTable. func (f *FDTable) String() string { var b bytes.Buffer - f.forEach(func(fd int32, file *fs.File, flags FDFlags) { - n, _ := file.Dirent.FullName(nil /* root */) - b.WriteString(fmt.Sprintf("\tfd:%d => name %s\n", fd, n)) + f.forEach(func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { + switch { + case file != nil: + n, _ := file.Dirent.FullName(nil /* root */) + b.WriteString(fmt.Sprintf("\tfd:%d => name %s\n", fd, n)) + + case fileVFS2 != nil: + fs := fileVFS2.VirtualDentry().Mount().Filesystem().VirtualFilesystem() + // TODO(gvisor.dev/issue/1623): We have no context nor root. Will this work? + name, err := fs.PathnameWithDeleted(context.Background(), vfs.VirtualDentry{}, fileVFS2.VirtualDentry()) + if err != nil { + b.WriteString(fmt.Sprintf("\n", err)) + } + b.WriteString(fmt.Sprintf("\tfd:%d => name %s\n", fd, name)) + } }) return b.String() } @@ -262,6 +300,17 @@ func (f *FDTable) NewFDs(ctx context.Context, fd int32, files []*fs.File, flags // reference for that FD, the ref count for that existing reference is // decremented. func (f *FDTable) NewFDAt(ctx context.Context, fd int32, file *fs.File, flags FDFlags) error { + return f.newFDAt(ctx, fd, file, nil, flags) +} + +// NewFDAtVFS2 sets the file reference for the given FD. If there is an active +// reference for that FD, the ref count for that existing reference is +// decremented. +func (f *FDTable) NewFDAtVFS2(ctx context.Context, fd int32, file *vfs.FileDescription, flags FDFlags) error { + return f.newFDAt(ctx, fd, nil, file, flags) +} + +func (f *FDTable) newFDAt(ctx context.Context, fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) error { if fd < 0 { // Don't accept negative FDs. return syscall.EBADF @@ -278,7 +327,7 @@ func (f *FDTable) NewFDAt(ctx context.Context, fd int32, file *fs.File, flags FD } // Install the entry. - f.set(fd, file, flags) + f.setAll(fd, file, fileVFS2, flags) return nil } @@ -330,10 +379,35 @@ func (f *FDTable) Get(fd int32) (*fs.File, FDFlags) { } } +// GetVFS2 returns a reference to the file and the flags for the FD or nil if no +// file is defined for the given fd. +// +// N.B. Callers are required to use DecRef when they are done. +// +//go:nosplit +func (f *FDTable) GetVFS2(fd int32) (*vfs.FileDescription, FDFlags) { + if fd < 0 { + return nil, FDFlags{} + } + + for { + file, flags, _ := f.getVFS2(fd) + if file != nil { + if !file.TryIncRef() { + continue // Race caught. + } + // Reference acquired. + return file, flags + } + // No file available. + return nil, FDFlags{} + } +} + // GetFDs returns a list of valid fds. func (f *FDTable) GetFDs() []int32 { fds := make([]int32, 0, int(atomic.LoadInt32(&f.used))) - f.forEach(func(fd int32, file *fs.File, flags FDFlags) { + f.forEach(func(fd int32, _ *fs.File, _ *vfs.FileDescription, _ FDFlags) { fds = append(fds, fd) }) return fds @@ -344,7 +418,19 @@ func (f *FDTable) GetFDs() []int32 { // they're done using the slice. func (f *FDTable) GetRefs() []*fs.File { files := make([]*fs.File, 0, f.Size()) - f.forEach(func(_ int32, file *fs.File, flags FDFlags) { + f.forEach(func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) { + file.IncRef() // Acquire a reference for caller. + files = append(files, file) + }) + return files +} + +// GetRefsVFS2 returns a stable slice of references to all files and bumps the +// reference count on each. The caller must use DecRef on each reference when +// they're done using the slice. +func (f *FDTable) GetRefsVFS2() []*vfs.FileDescription { + files := make([]*vfs.FileDescription, 0, f.Size()) + f.forEach(func(_ int32, _ *fs.File, file *vfs.FileDescription, _ FDFlags) { file.IncRef() // Acquire a reference for caller. files = append(files, file) }) @@ -355,10 +441,15 @@ func (f *FDTable) GetRefs() []*fs.File { func (f *FDTable) Fork() *FDTable { clone := f.k.NewFDTable() - f.forEach(func(fd int32, file *fs.File, flags FDFlags) { + f.forEach(func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { // The set function here will acquire an appropriate table // reference for the clone. We don't need anything else. - clone.set(fd, file, flags) + switch { + case file != nil: + clone.set(fd, file, flags) + case fileVFS2 != nil: + clone.setVFS2(fd, fileVFS2, flags) + } }) return clone } @@ -366,9 +457,9 @@ func (f *FDTable) Fork() *FDTable { // Remove removes an FD from and returns a non-file iff successful. // // N.B. Callers are required to use DecRef when they are done. -func (f *FDTable) Remove(fd int32) *fs.File { +func (f *FDTable) Remove(fd int32) (*fs.File, *vfs.FileDescription) { if fd < 0 { - return nil + return nil, nil } f.mu.Lock() @@ -379,21 +470,26 @@ func (f *FDTable) Remove(fd int32) *fs.File { f.next = fd } - orig, _, _ := f.get(fd) - if orig != nil { - orig.IncRef() // Reference for caller. - f.set(fd, nil, FDFlags{}) // Zap entry. + orig, orig2, _, _ := f.getAll(fd) + + // Add reference for caller. + switch { + case orig != nil: + orig.IncRef() + case orig2 != nil: + orig2.IncRef() } - return orig + f.setAll(fd, nil, nil, FDFlags{}) // Zap entry. + return orig, orig2 } // RemoveIf removes all FDs where cond is true. -func (f *FDTable) RemoveIf(cond func(*fs.File, FDFlags) bool) { +func (f *FDTable) RemoveIf(cond func(*fs.File, *vfs.FileDescription, FDFlags) bool) { f.mu.Lock() defer f.mu.Unlock() - f.forEach(func(fd int32, file *fs.File, flags FDFlags) { - if cond(file, flags) { + f.forEach(func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { + if cond(file, fileVFS2, flags) { f.set(fd, nil, FDFlags{}) // Clear from table. // Update current available position. if fd < f.next { diff --git a/pkg/sentry/kernel/fd_table_test.go b/pkg/sentry/kernel/fd_table_test.go index 261b815f2..29f95a2c4 100644 --- a/pkg/sentry/kernel/fd_table_test.go +++ b/pkg/sentry/kernel/fd_table_test.go @@ -150,13 +150,13 @@ func TestFDTable(t *testing.T) { t.Fatalf("fdTable.Get(2): got a %v, wanted nil", ref) } - ref := fdTable.Remove(1) + ref, _ := fdTable.Remove(1) if ref == nil { t.Fatalf("fdTable.Remove(1) for an existing FD: failed, want success") } ref.DecRef() - if ref := fdTable.Remove(1); ref != nil { + if ref, _ := fdTable.Remove(1); ref != nil { t.Fatalf("r.Remove(1) for a removed FD: got success, want failure") } }) diff --git a/pkg/sentry/kernel/fd_table_unsafe.go b/pkg/sentry/kernel/fd_table_unsafe.go index e9fdb0917..7fd97dc53 100644 --- a/pkg/sentry/kernel/fd_table_unsafe.go +++ b/pkg/sentry/kernel/fd_table_unsafe.go @@ -19,6 +19,7 @@ import ( "unsafe" "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/vfs" ) type descriptorTable struct { @@ -41,15 +42,38 @@ func (f *FDTable) init() { // //go:nosplit func (f *FDTable) get(fd int32) (*fs.File, FDFlags, bool) { + file, _, flags, ok := f.getAll(fd) + return file, flags, ok +} + +// getVFS2 gets a file entry. +// +// The boolean indicates whether this was in range. +// +//go:nosplit +func (f *FDTable) getVFS2(fd int32) (*vfs.FileDescription, FDFlags, bool) { + _, file, flags, ok := f.getAll(fd) + return file, flags, ok +} + +// getAll gets a file entry. +// +// The boolean indicates whether this was in range. +// +//go:nosplit +func (f *FDTable) getAll(fd int32) (*fs.File, *vfs.FileDescription, FDFlags, bool) { slice := *(*[]unsafe.Pointer)(atomic.LoadPointer(&f.slice)) if fd >= int32(len(slice)) { - return nil, FDFlags{}, false + return nil, nil, FDFlags{}, false } d := (*descriptor)(atomic.LoadPointer(&slice[fd])) if d == nil { - return nil, FDFlags{}, true + return nil, nil, FDFlags{}, true } - return d.file, d.flags, true + if d.file != nil && d.fileVFS2 != nil { + panic("VFS1 and VFS2 files set") + } + return d.file, d.fileVFS2, d.flags, true } // set sets an entry. @@ -59,6 +83,30 @@ func (f *FDTable) get(fd int32) (*fs.File, FDFlags, bool) { // // Precondition: mu must be held. func (f *FDTable) set(fd int32, file *fs.File, flags FDFlags) { + f.setAll(fd, file, nil, flags) +} + +// setVFS2 sets an entry. +// +// This handles accounting changes, as well as acquiring and releasing the +// reference needed by the table iff the file is different. +// +// Precondition: mu must be held. +func (f *FDTable) setVFS2(fd int32, file *vfs.FileDescription, flags FDFlags) { + f.setAll(fd, nil, file, flags) +} + +// setAll sets an entry. +// +// This handles accounting changes, as well as acquiring and releasing the +// reference needed by the table iff the file is different. +// +// Precondition: mu must be held. +func (f *FDTable) setAll(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { + if file != nil && fileVFS2 != nil { + panic("VFS1 and VFS2 files set") + } + slice := *(*[]unsafe.Pointer)(atomic.LoadPointer(&f.slice)) // Grow the table as required. @@ -71,33 +119,51 @@ func (f *FDTable) set(fd int32, file *fs.File, flags FDFlags) { atomic.StorePointer(&f.slice, unsafe.Pointer(&slice)) } - // Create the new element. - var d *descriptor - if file != nil { - d = &descriptor{ - file: file, - flags: flags, + var desc *descriptor + if file != nil || fileVFS2 != nil { + desc = &descriptor{ + file: file, + fileVFS2: fileVFS2, + flags: flags, } } // Update the single element. - orig := (*descriptor)(atomic.SwapPointer(&slice[fd], unsafe.Pointer(d))) + orig := (*descriptor)(atomic.SwapPointer(&slice[fd], unsafe.Pointer(desc))) // Acquire a table reference. - if file != nil && (orig == nil || file != orig.file) { - file.IncRef() + if desc != nil { + switch { + case desc.file != nil: + if orig == nil || desc.file != orig.file { + desc.file.IncRef() + } + case desc.fileVFS2 != nil: + if orig == nil || desc.fileVFS2 != orig.fileVFS2 { + desc.fileVFS2.IncRef() + } + } } // Drop the table reference. - if orig != nil && file != orig.file { - f.drop(orig.file) + if orig != nil { + switch { + case orig.file != nil: + if desc == nil || desc.file != orig.file { + f.drop(orig.file) + } + case orig.fileVFS2 != nil: + if desc == nil || desc.fileVFS2 != orig.fileVFS2 { + f.dropVFS2(orig.fileVFS2) + } + } } // Adjust used. switch { - case orig == nil && file != nil: + case orig == nil && desc != nil: atomic.AddInt32(&f.used, 1) - case orig != nil && file == nil: + case orig != nil && desc == nil: atomic.AddInt32(&f.used, -1) } } diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 7b90fac5a..dcd6e91c4 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -65,6 +65,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/unimpl" uspb "gvisor.dev/gvisor/pkg/sentry/unimpl/unimplemented_syscall_go_proto" "gvisor.dev/gvisor/pkg/sentry/uniqueid" + "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/state" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" @@ -435,17 +436,17 @@ func (k *Kernel) flushMountSourceRefs() error { // There may be some open FDs whose filesystems have been unmounted. We // must flush those as well. - return k.tasks.forEachFDPaused(func(file *fs.File) error { + return k.tasks.forEachFDPaused(func(file *fs.File, _ *vfs.FileDescription) error { file.Dirent.Inode.MountSource.FlushDirentRefs() return nil }) } -// forEachFDPaused applies the given function to each open file descriptor in each -// task. +// forEachFDPaused applies the given function to each open file descriptor in +// each task. // // Precondition: Must be called with the kernel paused. -func (ts *TaskSet) forEachFDPaused(f func(*fs.File) error) (err error) { +func (ts *TaskSet) forEachFDPaused(f func(*fs.File, *vfs.FileDescription) error) (err error) { ts.mu.RLock() defer ts.mu.RUnlock() for t := range ts.Root.tids { @@ -453,8 +454,8 @@ func (ts *TaskSet) forEachFDPaused(f func(*fs.File) error) (err error) { if t.fdTable == nil { continue } - t.fdTable.forEach(func(_ int32, file *fs.File, _ FDFlags) { - if lastErr := f(file); lastErr != nil && err == nil { + t.fdTable.forEach(func(_ int32, file *fs.File, fileVFS2 *vfs.FileDescription, _ FDFlags) { + if lastErr := f(file, fileVFS2); lastErr != nil && err == nil { err = lastErr } }) @@ -463,7 +464,8 @@ func (ts *TaskSet) forEachFDPaused(f func(*fs.File) error) (err error) { } func (ts *TaskSet) flushWritesToFiles(ctx context.Context) error { - return ts.forEachFDPaused(func(file *fs.File) error { + // TODO(gvisor.dev/issues/1663): Add save support for VFS2. + return ts.forEachFDPaused(func(file *fs.File, _ *vfs.FileDescription) error { if flags := file.Flags(); !flags.Write { return nil } @@ -474,12 +476,9 @@ func (ts *TaskSet) flushWritesToFiles(ctx context.Context) error { syncErr := file.Fsync(ctx, 0, fs.FileMaxOffset, fs.SyncAll) if err := fs.SaveFileFsyncError(syncErr); err != nil { name, _ := file.Dirent.FullName(nil /* root */) - // Wrap this error in ErrSaveRejection - // so that it will trigger a save - // error, rather than a panic. This - // also allows us to distinguish Fsync - // errors from state file errors in - // state.Save. + // Wrap this error in ErrSaveRejection so that it will trigger a save + // error, rather than a panic. This also allows us to distinguish Fsync + // errors from state file errors in state.Save. return fs.ErrSaveRejection{ Err: fmt.Errorf("%q was not sufficiently synced: %v", name, err), } @@ -519,7 +518,7 @@ func (ts *TaskSet) unregisterEpollWaiters() { for t := range ts.Root.tids { // We can skip locking Task.mu here since the kernel is paused. if t.fdTable != nil { - t.fdTable.forEach(func(_ int32, file *fs.File, _ FDFlags) { + t.fdTable.forEach(func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) { if e, ok := file.FileOperations.(*epoll.EventPoll); ok { e.UnregisterEpollWaiters() } @@ -921,7 +920,7 @@ func (k *Kernel) pauseTimeLocked() { // This means we'll iterate FDTables shared by multiple tasks repeatedly, // but ktime.Timer.Pause is idempotent so this is harmless. if t.fdTable != nil { - t.fdTable.forEach(func(_ int32, file *fs.File, _ FDFlags) { + t.fdTable.forEach(func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) { if tfd, ok := file.FileOperations.(*timerfd.TimerOperations); ok { tfd.PauseTimer() } @@ -951,7 +950,7 @@ func (k *Kernel) resumeTimeLocked() { } } if t.fdTable != nil { - t.fdTable.forEach(func(_ int32, file *fs.File, _ FDFlags) { + t.fdTable.forEach(func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) { if tfd, ok := file.FileOperations.(*timerfd.TimerOperations); ok { tfd.ResumeTimer() } diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index 95adf2778..981e8c7fe 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -35,6 +35,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/unimpl" "gvisor.dev/gvisor/pkg/sentry/uniqueid" "gvisor.dev/gvisor/pkg/sentry/usage" + "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" @@ -743,6 +744,14 @@ func (t *Task) GetFile(fd int32) *fs.File { return f } +// GetFileVFS2 is a convenience wrapper for t.FDTable().GetVFS2. +// +// Precondition: same as FDTable.Get. +func (t *Task) GetFileVFS2(fd int32) *vfs.FileDescription { + f, _ := t.fdTable.GetVFS2(fd) + return f +} + // NewFDs is a convenience wrapper for t.FDTable().NewFDs. // // This automatically passes the task as the context. diff --git a/pkg/sentry/kernel/task_exec.go b/pkg/sentry/kernel/task_exec.go index fa6528386..8f57a34a6 100644 --- a/pkg/sentry/kernel/task_exec.go +++ b/pkg/sentry/kernel/task_exec.go @@ -69,6 +69,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/mm" + "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" ) @@ -198,7 +199,7 @@ func (r *runSyscallAfterExecStop) execute(t *Task) taskRunState { t.tg.pidns.owner.mu.Unlock() // Remove FDs with the CloseOnExec flag set. - t.fdTable.RemoveIf(func(file *fs.File, flags FDFlags) bool { + t.fdTable.RemoveIf(func(_ *fs.File, _ *vfs.FileDescription, flags FDFlags) bool { return flags.CloseOnExec }) diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD index 8d6c52850..be16ee686 100644 --- a/pkg/sentry/syscalls/linux/BUILD +++ b/pkg/sentry/syscalls/linux/BUILD @@ -93,6 +93,7 @@ go_library( "//pkg/sentry/socket/unix/transport", "//pkg/sentry/syscalls", "//pkg/sentry/usage", + "//pkg/sentry/vfs", "//pkg/sync", "//pkg/syserr", "//pkg/syserror", diff --git a/pkg/sentry/syscalls/linux/error.go b/pkg/sentry/syscalls/linux/error.go index 60469549d..64de56ac5 100644 --- a/pkg/sentry/syscalls/linux/error.go +++ b/pkg/sentry/syscalls/linux/error.go @@ -22,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/metric" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) @@ -31,20 +32,58 @@ var ( partialResultOnce sync.Once ) +// HandleIOErrorVFS2 handles special error cases for partial results. For some +// errors, we may consume the error and return only the partial read/write. +// +// op and f are used only for panics. +func HandleIOErrorVFS2(t *kernel.Task, partialResult bool, err, intr error, op string, f *vfs.FileDescription) error { + known, err := handleIOErrorImpl(t, partialResult, err, intr, op) + if err != nil { + return err + } + if !known { + // An unknown error is encountered with a partial read/write. + fs := f.Mount().Filesystem().VirtualFilesystem() + root := vfs.RootFromContext(t) + name, _ := fs.PathnameWithDeleted(t, root, f.VirtualDentry()) + log.Traceback("Invalid request partialResult %v and err (type %T) %v for %s operation on %q", partialResult, err, err, op, name) + partialResultOnce.Do(partialResultMetric.Increment) + } + return nil +} + // handleIOError handles special error cases for partial results. For some // errors, we may consume the error and return only the partial read/write. // // op and f are used only for panics. func handleIOError(t *kernel.Task, partialResult bool, err, intr error, op string, f *fs.File) error { + known, err := handleIOErrorImpl(t, partialResult, err, intr, op) + if err != nil { + return err + } + if !known { + // An unknown error is encountered with a partial read/write. + name, _ := f.Dirent.FullName(nil /* ignore chroot */) + log.Traceback("Invalid request partialResult %v and err (type %T) %v for %s operation on %q, %T", partialResult, err, err, op, name, f.FileOperations) + partialResultOnce.Do(partialResultMetric.Increment) + } + return nil +} + +// handleIOError handles special error cases for partial results. For some +// errors, we may consume the error and return only the partial read/write. +// +// Returns false if error is unknown. +func handleIOErrorImpl(t *kernel.Task, partialResult bool, err, intr error, op string) (bool, error) { switch err { case nil: // Typical successful syscall. - return nil + return true, nil case io.EOF: // EOF is always consumed. If this is a partial read/write // (result != 0), the application will see that, otherwise // they will see 0. - return nil + return true, nil case syserror.ErrExceedsFileSizeLimit: // Ignore partialResult because this error only applies to // normal files, and for those files we cannot accumulate @@ -53,20 +92,20 @@ func handleIOError(t *kernel.Task, partialResult bool, err, intr error, op strin // Do not consume the error and return it as EFBIG. // Simultaneously send a SIGXFSZ per setrlimit(2). t.SendSignal(kernel.SignalInfoNoInfo(linux.SIGXFSZ, t, t)) - return syserror.EFBIG + return true, syserror.EFBIG case syserror.ErrInterrupted: // The syscall was interrupted. Return nil if it completed // partially, otherwise return the error code that the syscall // needs (to indicate to the kernel what it should do). if partialResult { - return nil + return true, nil } - return intr + return true, intr } if !partialResult { // Typical syscall error. - return err + return true, err } switch err { @@ -75,14 +114,14 @@ func handleIOError(t *kernel.Task, partialResult bool, err, intr error, op strin // read/write. Like ErrWouldBlock, since we have a // partial read/write, we consume the error and return // the partial result. - return nil + return true, nil case syserror.EFAULT: // EFAULT is only shown the user if nothing was // read/written. If we read something (this case), they see // a partial read/write. They will then presumably try again // with an incremented buffer, which will EFAULT with // result == 0. - return nil + return true, nil case syserror.EPIPE: // Writes to a pipe or socket will return EPIPE if the other // side is gone. The partial write is returned. EPIPE will be @@ -90,32 +129,29 @@ func handleIOError(t *kernel.Task, partialResult bool, err, intr error, op strin // // TODO(gvisor.dev/issue/161): In some cases SIGPIPE should // also be sent to the application. - return nil + return true, nil case syserror.ENOSPC: // Similar to EPIPE. Return what we wrote this time, and let // ENOSPC be returned on the next call. - return nil + return true, nil case syserror.ECONNRESET: // For TCP sendfile connections, we may have a reset. But we // should just return n as the result. - return nil + return true, nil case syserror.ErrWouldBlock: // Syscall would block, but completed a partial read/write. // This case should only be returned by IssueIO for nonblocking // files. Since we have a partial read/write, we consume // ErrWouldBlock, returning the partial result. - return nil + return true, nil } switch err.(type) { case kernel.SyscallRestartErrno: // Identical to the EINTR case. - return nil + return true, nil } - // An unknown error is encountered with a partial read/write. - name, _ := f.Dirent.FullName(nil /* ignore chroot */) - log.Traceback("Invalid request partialResult %v and err (type %T) %v for %s operation on %q, %T", partialResult, err, err, op, name, f.FileOperations) - partialResultOnce.Do(partialResultMetric.Increment) - return nil + // Error is unknown and cannot be properly handled. + return false, nil } diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go index c54735148..421845ebb 100644 --- a/pkg/sentry/syscalls/linux/sys_file.go +++ b/pkg/sentry/syscalls/linux/sys_file.go @@ -767,7 +767,7 @@ func Close(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // Note that Remove provides a reference on the file that we may use to // flush. It is still active until we drop the final reference below // (and other reference-holding operations complete). - file := t.FDTable().Remove(fd) + file, _ := t.FDTable().Remove(fd) if file == nil { return 0, nil, syserror.EBADF } diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD new file mode 100644 index 000000000..6b8a00b6e --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/BUILD @@ -0,0 +1,24 @@ +load("//tools:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_library( + name = "vfs2", + srcs = [ + "linux64.go", + "linux64_override_amd64.go", + "linux64_override_arm64.go", + "sys_read.go", + ], + visibility = ["//:sandbox"], + deps = [ + "//pkg/sentry/arch", + "//pkg/sentry/kernel", + "//pkg/sentry/syscalls", + "//pkg/sentry/syscalls/linux", + "//pkg/sentry/vfs", + "//pkg/syserror", + "//pkg/usermem", + "//pkg/waiter", + ], +) diff --git a/pkg/sentry/syscalls/linux/vfs2/linux64.go b/pkg/sentry/syscalls/linux/vfs2/linux64.go new file mode 100644 index 000000000..19ee36081 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/linux64.go @@ -0,0 +1,16 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package vfs2 provides syscall implementations that use VFS2. +package vfs2 diff --git a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go new file mode 100644 index 000000000..c134714ee --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go @@ -0,0 +1,25 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/syscalls" +) + +// Override syscall table to add syscalls implementations from this package. +func Override(table map[uintptr]kernel.Syscall) { + table[0] = syscalls.Supported("read", Read) +} diff --git a/pkg/sentry/syscalls/linux/vfs2/linux64_override_arm64.go b/pkg/sentry/syscalls/linux/vfs2/linux64_override_arm64.go new file mode 100644 index 000000000..6af5c400f --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/linux64_override_arm64.go @@ -0,0 +1,25 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/syscalls" +) + +// Override syscall table to add syscalls implementations from this package. +func Override(table map[uintptr]kernel.Syscall) { + table[63] = syscalls.Supported("read", Read) +} diff --git a/pkg/sentry/syscalls/linux/vfs2/sys_read.go b/pkg/sentry/syscalls/linux/vfs2/sys_read.go new file mode 100644 index 000000000..b9fb58464 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/sys_read.go @@ -0,0 +1,95 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +// Read implements linux syscall read(2). Note that we try to get a buffer that +// is exactly the size requested because some applications like qemu expect +// they can do large reads all at once. Bug for bug. Same for other read +// calls below. +func Read(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + addr := args[1].Pointer() + size := args[2].SizeT() + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Check that the file is readable. + if !file.IsReadable() { + return 0, nil, syserror.EBADF + } + + // Check that the size is legitimate. + si := int(size) + if si < 0 { + return 0, nil, syserror.EINVAL + } + + // Get the destination of the read. + dst, err := t.SingleIOSequence(addr, si, usermem.IOOpts{ + AddressSpaceActive: true, + }) + if err != nil { + return 0, nil, err + } + + n, err := read(t, file, dst, vfs.ReadOptions{}) + t.IOUsage().AccountReadSyscall(n) + return uintptr(n), nil, linux.HandleIOErrorVFS2(t, n != 0, err, kernel.ERESTARTSYS, "read", file) +} + +func read(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + n, err := file.Read(t, dst, opts) + if err != syserror.ErrWouldBlock { + return n, err + } + + // Register for notifications. + _, ch := waiter.NewChannelEntry(nil) + // file.EventRegister(&w, EventMaskRead) + + total := n + for { + // Shorten dst to reflect bytes previously read. + dst = dst.DropFirst(int(n)) + + // Issue the request and break out if it completes with anything other than + // "would block". + n, err := file.Read(t, dst, opts) + total += n + if err != syserror.ErrWouldBlock { + break + } + if err := t.Block(ch); err != nil { + break + } + } + //file.EventUnregister(&w) + + return total, err +} diff --git a/runsc/boot/BUILD b/runsc/boot/BUILD index a96c80261..ae4dd102a 100644 --- a/runsc/boot/BUILD +++ b/runsc/boot/BUILD @@ -68,6 +68,7 @@ go_library( "//pkg/sentry/state", "//pkg/sentry/strace", "//pkg/sentry/syscalls/linux", + "//pkg/sentry/syscalls/linux/vfs2", "//pkg/sentry/time", "//pkg/sentry/unimpl:unimplemented_syscall_go_proto", "//pkg/sentry/usage", diff --git a/runsc/boot/config.go b/runsc/boot/config.go index a878bc2ce..35391030f 100644 --- a/runsc/boot/config.go +++ b/runsc/boot/config.go @@ -256,6 +256,9 @@ type Config struct { // // E.g. 0.2 CPU quota will result in 1, and 1.9 in 2. CPUNumFromQuota bool + + // Enables VFS2 (not plumbled through yet). + VFS2 bool } // ToFlags returns a slice of flags that correspond to the given Config. diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index fad72f4ab..9f0d5d7af 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -26,6 +26,7 @@ import ( specs "github.com/opencontainers/runtime-spec/specs-go" "golang.org/x/sys/unix" + "gvisor.dev/gvisor/pkg/abi" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/log" @@ -42,6 +43,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/sighandling" + "gvisor.dev/gvisor/pkg/sentry/syscalls/linux/vfs2" "gvisor.dev/gvisor/pkg/sentry/time" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/sentry/watchdog" @@ -184,6 +186,13 @@ func New(args Args) (*Loader, error) { return nil, fmt.Errorf("setting up memory usage: %v", err) } + if args.Conf.VFS2 { + st, ok := kernel.LookupSyscallTable(abi.Linux, arch.Host) + if ok { + vfs2.Override(st.Table) + } + } + // Create kernel and platform. p, err := createPlatform(args.Conf, args.Device) if err != nil { -- cgit v1.2.3 From 3d046fef06ece6ba20770fa62e0a21569226adaa Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Tue, 28 Jan 2020 16:42:05 -0800 Subject: Changes missing in last submit Updates #1487 Updates #1623 PiperOrigin-RevId: 292040835 --- pkg/sentry/kernel/fd_table.go | 18 +++++++++--------- pkg/sentry/syscalls/linux/sys_read.go | 2 +- pkg/sentry/syscalls/linux/vfs2/sys_read.go | 16 ++++++++-------- 3 files changed, 18 insertions(+), 18 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/fd_table.go b/pkg/sentry/kernel/fd_table.go index 56b70ce96..23b88f7a6 100644 --- a/pkg/sentry/kernel/fd_table.go +++ b/pkg/sentry/kernel/fd_table.go @@ -15,9 +15,9 @@ package kernel import ( - "bytes" "fmt" "math" + "strings" "sync/atomic" "syscall" @@ -221,24 +221,24 @@ func (f *FDTable) forEach(fn func(fd int32, file *fs.File, fileVFS2 *vfs.FileDes // String is a stringer for FDTable. func (f *FDTable) String() string { - var b bytes.Buffer + var buf strings.Builder f.forEach(func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { switch { case file != nil: n, _ := file.Dirent.FullName(nil /* root */) - b.WriteString(fmt.Sprintf("\tfd:%d => name %s\n", fd, n)) + fmt.Fprintf(&buf, "\tfd:%d => name %s\n", fd, n) case fileVFS2 != nil: - fs := fileVFS2.VirtualDentry().Mount().Filesystem().VirtualFilesystem() - // TODO(gvisor.dev/issue/1623): We have no context nor root. Will this work? - name, err := fs.PathnameWithDeleted(context.Background(), vfs.VirtualDentry{}, fileVFS2.VirtualDentry()) + vfsObj := fileVFS2.Mount().Filesystem().VirtualFilesystem() + name, err := vfsObj.PathnameWithDeleted(context.Background(), vfs.VirtualDentry{}, fileVFS2.VirtualDentry()) if err != nil { - b.WriteString(fmt.Sprintf("\n", err)) + fmt.Fprintf(&buf, "\n", err) + return } - b.WriteString(fmt.Sprintf("\tfd:%d => name %s\n", fd, name)) + fmt.Fprintf(&buf, "\tfd:%d => name %s\n", fd, name) } }) - return b.String() + return buf.String() } // NewFDs allocates new FDs guaranteed to be the lowest number available diff --git a/pkg/sentry/syscalls/linux/sys_read.go b/pkg/sentry/syscalls/linux/sys_read.go index f9f594190..227692f06 100644 --- a/pkg/sentry/syscalls/linux/sys_read.go +++ b/pkg/sentry/syscalls/linux/sys_read.go @@ -29,7 +29,7 @@ import ( ) const ( - // EventMaskRead contains events that can be triggerd on reads. + // EventMaskRead contains events that can be triggered on reads. EventMaskRead = waiter.EventIn | waiter.EventHUp | waiter.EventErr ) diff --git a/pkg/sentry/syscalls/linux/vfs2/sys_read.go b/pkg/sentry/syscalls/linux/vfs2/sys_read.go index b9fb58464..7667524c7 100644 --- a/pkg/sentry/syscalls/linux/vfs2/sys_read.go +++ b/pkg/sentry/syscalls/linux/vfs2/sys_read.go @@ -24,6 +24,11 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +const ( + // EventMaskRead contains events that can be triggered on reads. + EventMaskRead = waiter.EventIn | waiter.EventHUp | waiter.EventErr +) + // Read implements linux syscall read(2). Note that we try to get a buffer that // is exactly the size requested because some applications like qemu expect // they can do large reads all at once. Bug for bug. Same for other read @@ -39,11 +44,6 @@ func Read(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC } defer file.DecRef() - // Check that the file is readable. - if !file.IsReadable() { - return 0, nil, syserror.EBADF - } - // Check that the size is legitimate. si := int(size) if si < 0 { @@ -70,8 +70,8 @@ func read(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, opt } // Register for notifications. - _, ch := waiter.NewChannelEntry(nil) - // file.EventRegister(&w, EventMaskRead) + w, ch := waiter.NewChannelEntry(nil) + file.EventRegister(&w, EventMaskRead) total := n for { @@ -89,7 +89,7 @@ func read(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, opt break } } - //file.EventUnregister(&w) + file.EventUnregister(&w) return total, err } -- cgit v1.2.3 From 396c574db276ae1424af7098b5cd917e2bed9921 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Tue, 28 Jan 2020 18:30:36 -0800 Subject: Add support for WritableSource in DynamicBytesFileDescriptionImpl WritableSource is a convenience interface used for files that can be written to, e.g. /proc/net/ipv4/tpc_sack. It reads max of 4KB and only from offset 0 which should cover most cases. It can be extended as neeed. Updates #1195 PiperOrigin-RevId: 292056924 --- pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go | 4 +- pkg/sentry/fsimpl/proc/tasks.go | 2 +- pkg/sentry/fsimpl/proc/tasks_sys.go | 84 +++++++++++-- pkg/sentry/vfs/file_description_impl_util.go | 76 +++++++++--- pkg/sentry/vfs/file_description_impl_util_test.go | 138 +++++++++++++++++----- 5 files changed, 250 insertions(+), 54 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go index 373f801ff..733792c78 100644 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -108,12 +108,12 @@ func (fd *DynamicBytesFD) PRead(ctx context.Context, dst usermem.IOSequence, off // Write implements vfs.FileDescriptionImpl.Write. func (fd *DynamicBytesFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { - return fd.FileDescriptionDefaultImpl.Write(ctx, src, opts) + return fd.DynamicBytesFileDescriptionImpl.Write(ctx, src, opts) } // PWrite implements vfs.FileDescriptionImpl.PWrite. func (fd *DynamicBytesFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - return fd.FileDescriptionDefaultImpl.PWrite(ctx, src, offset, opts) + return fd.DynamicBytesFileDescriptionImpl.PWrite(ctx, src, offset, opts) } // Release implements vfs.FileDescriptionImpl.Release. diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index e0cb9c47b..14bd334e8 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -69,7 +69,7 @@ func newTasksInode(inoGen InoGenerator, k *kernel.Kernel, pidns *kernel.PIDNames "cpuinfo": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(cpuInfoData(k))), //"filesystems": newDentry(root, inoGen.NextIno(), 0444, &filesystemsData{}), "loadavg": newDentry(root, inoGen.NextIno(), 0444, &loadavgData{}), - "sys": newSysDir(root, inoGen), + "sys": newSysDir(root, inoGen, k), "meminfo": newDentry(root, inoGen.NextIno(), 0444, &meminfoData{}), "mounts": kernfs.NewStaticSymlink(root, inoGen.NextIno(), "self/mounts"), "net": newNetDir(root, inoGen, k), diff --git a/pkg/sentry/fsimpl/proc/tasks_sys.go b/pkg/sentry/fsimpl/proc/tasks_sys.go index ad963870b..c7ce74883 100644 --- a/pkg/sentry/fsimpl/proc/tasks_sys.go +++ b/pkg/sentry/fsimpl/proc/tasks_sys.go @@ -21,12 +21,16 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // newSysDir returns the dentry corresponding to /proc/sys directory. -func newSysDir(root *auth.Credentials, inoGen InoGenerator) *kernfs.Dentry { +func newSysDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) *kernfs.Dentry { return kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ "kernel": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ "hostname": newDentry(root, inoGen.NextIno(), 0444, &hostnameData{}), @@ -38,18 +42,18 @@ func newSysDir(root *auth.Credentials, inoGen InoGenerator) *kernfs.Dentry { "mmap_min_addr": newDentry(root, inoGen.NextIno(), 0444, &mmapMinAddrData{}), "overcommit_memory": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0\n")), }), - "net": newSysNetDir(root, inoGen), + "net": newSysNetDir(root, inoGen, k), }) } // newSysNetDir returns the dentry corresponding to /proc/sys/net directory. -func newSysNetDir(root *auth.Credentials, inoGen InoGenerator) *kernfs.Dentry { - return kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "net": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ +func newSysNetDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) *kernfs.Dentry { + var contents map[string]*kernfs.Dentry + + if stack := k.NetworkStack(); stack != nil { + contents = map[string]*kernfs.Dentry{ "ipv4": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - // Add tcp_sack. - // TODO(gvisor.dev/issue/1195): tcp_sack allows write(2) - // "tcp_sack": newTCPSackInode(ctx, msrc, s), + "tcp_sack": newDentry(root, inoGen.NextIno(), 0644, &tcpSackData{stack: stack}), // The following files are simple stubs until they are implemented in // netstack, most of these files are configuration related. We use the @@ -103,7 +107,11 @@ func newSysNetDir(root *auth.Credentials, inoGen InoGenerator) *kernfs.Dentry { "wmem_default": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), "wmem_max": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), }), - }), + } + } + + return kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ + "net": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, contents), }) } @@ -141,3 +149,61 @@ func (*hostnameData) Generate(ctx context.Context, buf *bytes.Buffer) error { buf.WriteString("\n") return nil } + +// tcpSackData implements vfs.WritableDynamicBytesSource for +// /proc/sys/net/tcp_sack. +// +// +stateify savable +type tcpSackData struct { + kernfs.DynamicBytesFile + + stack inet.Stack `state:"wait"` + enabled *bool +} + +var _ vfs.WritableDynamicBytesSource = (*tcpSackData)(nil) + +// Generate implements vfs.DynamicBytesSource. +func (d *tcpSackData) Generate(ctx context.Context, buf *bytes.Buffer) error { + if d.enabled == nil { + sack, err := d.stack.TCPSACKEnabled() + if err != nil { + return err + } + d.enabled = &sack + } + + val := "0\n" + if *d.enabled { + // Technically, this is not quite compatible with Linux. Linux stores these + // as an integer, so if you write "2" into tcp_sack, you should get 2 back. + // Tough luck. + val = "1\n" + } + buf.WriteString(val) + return nil +} + +func (d *tcpSackData) Write(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) { + if offset != 0 { + // No need to handle partial writes thus far. + return 0, syserror.EINVAL + } + if src.NumBytes() == 0 { + return 0, nil + } + + // Limit the amount of memory allocated. + src = src.TakeFirst(usermem.PageSize - 1) + + var v int32 + n, err := usermem.CopyInt32StringInVec(ctx, src.IO, src.Addrs, &v, src.Opts) + if err != nil { + return n, err + } + if d.enabled == nil { + d.enabled = new(bool) + } + *d.enabled = v != 0 + return n, d.stack.SetTCPSACKEnabled(*d.enabled) +} diff --git a/pkg/sentry/vfs/file_description_impl_util.go b/pkg/sentry/vfs/file_description_impl_util.go index fb9b87fdc..a4900c170 100644 --- a/pkg/sentry/vfs/file_description_impl_util.go +++ b/pkg/sentry/vfs/file_description_impl_util.go @@ -192,21 +192,6 @@ func (DentryMetadataFileDescriptionImpl) SetStat(ctx context.Context, opts SetSt panic("illegal call to DentryMetadataFileDescriptionImpl.SetStat") } -// DynamicBytesFileDescriptionImpl may be embedded by implementations of -// FileDescriptionImpl that represent read-only regular files whose contents -// are backed by a bytes.Buffer that is regenerated when necessary, consistent -// with Linux's fs/seq_file.c:single_open(). -// -// DynamicBytesFileDescriptionImpl.SetDataSource() must be called before first -// use. -type DynamicBytesFileDescriptionImpl struct { - data DynamicBytesSource // immutable - mu sync.Mutex // protects the following fields - buf bytes.Buffer - off int64 - lastRead int64 // offset at which the last Read, PRead, or Seek ended -} - // DynamicBytesSource represents a data source for a // DynamicBytesFileDescriptionImpl. type DynamicBytesSource interface { @@ -225,6 +210,30 @@ func (s *StaticData) Generate(ctx context.Context, buf *bytes.Buffer) error { return nil } +// WritableDynamicBytesSource extends DynamicBytesSource to allow writes to the +// underlying source. +type WritableDynamicBytesSource interface { + DynamicBytesSource + + // Write sends writes to the source. + Write(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) +} + +// DynamicBytesFileDescriptionImpl may be embedded by implementations of +// FileDescriptionImpl that represent read-only regular files whose contents +// are backed by a bytes.Buffer that is regenerated when necessary, consistent +// with Linux's fs/seq_file.c:single_open(). +// +// DynamicBytesFileDescriptionImpl.SetDataSource() must be called before first +// use. +type DynamicBytesFileDescriptionImpl struct { + data DynamicBytesSource // immutable + mu sync.Mutex // protects the following fields + buf bytes.Buffer + off int64 + lastRead int64 // offset at which the last Read, PRead, or Seek ended +} + // SetDataSource must be called exactly once on fd before first use. func (fd *DynamicBytesFileDescriptionImpl) SetDataSource(data DynamicBytesSource) { fd.data = data @@ -304,6 +313,43 @@ func (fd *DynamicBytesFileDescriptionImpl) Seek(ctx context.Context, offset int6 return offset, nil } +// Preconditions: fd.mu must be locked. +func (fd *DynamicBytesFileDescriptionImpl) pwriteLocked(ctx context.Context, src usermem.IOSequence, offset int64, opts WriteOptions) (int64, error) { + if opts.Flags&^(linux.RWF_HIPRI|linux.RWF_DSYNC|linux.RWF_SYNC) != 0 { + return 0, syserror.EOPNOTSUPP + } + + writable, ok := fd.data.(WritableDynamicBytesSource) + if !ok { + return 0, syserror.EINVAL + } + n, err := writable.Write(ctx, src, offset) + if err != nil { + return 0, err + } + + // Invalidate cached data that might exist prior to this call. + fd.buf.Reset() + return n, nil +} + +// PWrite implements FileDescriptionImpl.PWrite. +func (fd *DynamicBytesFileDescriptionImpl) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts WriteOptions) (int64, error) { + fd.mu.Lock() + n, err := fd.pwriteLocked(ctx, src, offset, opts) + fd.mu.Unlock() + return n, err +} + +// Write implements FileDescriptionImpl.Write. +func (fd *DynamicBytesFileDescriptionImpl) Write(ctx context.Context, src usermem.IOSequence, opts WriteOptions) (int64, error) { + fd.mu.Lock() + n, err := fd.pwriteLocked(ctx, src, fd.off, opts) + fd.off += n + fd.mu.Unlock() + return n, err +} + // GenericConfigureMMap may be used by most implementations of // FileDescriptionImpl.ConfigureMMap. func GenericConfigureMMap(fd *FileDescription, m memmap.Mappable, opts *memmap.MMapOpts) error { diff --git a/pkg/sentry/vfs/file_description_impl_util_test.go b/pkg/sentry/vfs/file_description_impl_util_test.go index 0f44e7c8c..8fa26418e 100644 --- a/pkg/sentry/vfs/file_description_impl_util_test.go +++ b/pkg/sentry/vfs/file_description_impl_util_test.go @@ -35,61 +35,80 @@ type fileDescription struct { FileDescriptionDefaultImpl } -// genCountFD is a read-only FileDescriptionImpl representing a regular file -// that contains the number of times its DynamicBytesSource.Generate() +// genCount contains the number of times its DynamicBytesSource.Generate() // implementation has been called. -type genCountFD struct { +type genCount struct { + count uint64 // accessed using atomic memory ops +} + +// Generate implements DynamicBytesSource.Generate. +func (g *genCount) Generate(ctx context.Context, buf *bytes.Buffer) error { + fmt.Fprintf(buf, "%d", atomic.AddUint64(&g.count, 1)) + return nil +} + +type storeData struct { + data string +} + +var _ WritableDynamicBytesSource = (*storeData)(nil) + +// Generate implements DynamicBytesSource. +func (d *storeData) Generate(ctx context.Context, buf *bytes.Buffer) error { + buf.WriteString(d.data) + return nil +} + +// Generate implements WritableDynamicBytesSource. +func (d *storeData) Write(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) { + buf := make([]byte, src.NumBytes()) + n, err := src.CopyIn(ctx, buf) + if err != nil { + return 0, err + } + + d.data = string(buf[:n]) + return 0, nil +} + +// testFD is a read-only FileDescriptionImpl representing a regular file. +type testFD struct { fileDescription DynamicBytesFileDescriptionImpl - count uint64 // accessed using atomic memory ops + data DynamicBytesSource } -func newGenCountFD(vfsObj *VirtualFilesystem) *FileDescription { +func newTestFD(vfsObj *VirtualFilesystem, statusFlags uint32, data DynamicBytesSource) *FileDescription { vd := vfsObj.NewAnonVirtualDentry("genCountFD") defer vd.DecRef() - var fd genCountFD - fd.vfsfd.Init(&fd, 0 /* statusFlags */, vd.Mount(), vd.Dentry(), &FileDescriptionOptions{}) - fd.DynamicBytesFileDescriptionImpl.SetDataSource(&fd) + var fd testFD + fd.vfsfd.Init(&fd, statusFlags, vd.Mount(), vd.Dentry(), &FileDescriptionOptions{}) + fd.DynamicBytesFileDescriptionImpl.SetDataSource(data) return &fd.vfsfd } // Release implements FileDescriptionImpl.Release. -func (fd *genCountFD) Release() { -} - -// StatusFlags implements FileDescriptionImpl.StatusFlags. -func (fd *genCountFD) StatusFlags(ctx context.Context) (uint32, error) { - return 0, nil +func (fd *testFD) Release() { } // SetStatusFlags implements FileDescriptionImpl.SetStatusFlags. -func (fd *genCountFD) SetStatusFlags(ctx context.Context, flags uint32) error { - return syserror.EPERM -} - // Stat implements FileDescriptionImpl.Stat. -func (fd *genCountFD) Stat(ctx context.Context, opts StatOptions) (linux.Statx, error) { +func (fd *testFD) Stat(ctx context.Context, opts StatOptions) (linux.Statx, error) { // Note that Statx.Mask == 0 in the return value. return linux.Statx{}, nil } // SetStat implements FileDescriptionImpl.SetStat. -func (fd *genCountFD) SetStat(ctx context.Context, opts SetStatOptions) error { +func (fd *testFD) SetStat(ctx context.Context, opts SetStatOptions) error { return syserror.EPERM } -// Generate implements DynamicBytesSource.Generate. -func (fd *genCountFD) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "%d", atomic.AddUint64(&fd.count, 1)) - return nil -} - func TestGenCountFD(t *testing.T) { ctx := contexttest.Context(t) vfsObj := New() // vfs.New() - fd := newGenCountFD(vfsObj) + fd := newTestFD(vfsObj, linux.O_RDWR, &genCount{}) defer fd.DecRef() // The first read causes Generate to be called to fill the FD's buffer. @@ -130,4 +149,69 @@ func TestGenCountFD(t *testing.T) { if want := byte('3'); buf[0] != want { t.Errorf("PRead: got byte %c, wanted %c", buf[0], want) } + + // Write and PWrite fails. + if _, err := fd.Write(ctx, ioseq, WriteOptions{}); err != syserror.EINVAL { + t.Errorf("Write: got err %v, wanted %v", err, syserror.EINVAL) + } + if _, err := fd.PWrite(ctx, ioseq, 0, WriteOptions{}); err != syserror.EINVAL { + t.Errorf("Write: got err %v, wanted %v", err, syserror.EINVAL) + } +} + +func TestWritable(t *testing.T) { + ctx := contexttest.Context(t) + + vfsObj := New() // vfs.New() + fd := newTestFD(vfsObj, linux.O_RDWR, &storeData{data: "init"}) + defer fd.DecRef() + + buf := make([]byte, 10) + ioseq := usermem.BytesIOSequence(buf) + if n, err := fd.Read(ctx, ioseq, ReadOptions{}); n != 4 && err != io.EOF { + t.Fatalf("Read: got (%v, %v), wanted (4, EOF)", n, err) + } + if want := "init"; want == string(buf) { + t.Fatalf("Read: got %v, wanted %v", string(buf), want) + } + + // Test PWrite. + want := "write" + writeIOSeq := usermem.BytesIOSequence([]byte(want)) + if n, err := fd.PWrite(ctx, writeIOSeq, 0, WriteOptions{}); int(n) != len(want) && err != nil { + t.Errorf("PWrite: got err (%v, %v), wanted (%v, nil)", n, err, len(want)) + } + if n, err := fd.PRead(ctx, ioseq, 0, ReadOptions{}); int(n) != len(want) && err != io.EOF { + t.Fatalf("PRead: got (%v, %v), wanted (%v, EOF)", n, err, len(want)) + } + if want == string(buf) { + t.Fatalf("PRead: got %v, wanted %v", string(buf), want) + } + + // Test Seek to 0 followed by Write. + want = "write2" + writeIOSeq = usermem.BytesIOSequence([]byte(want)) + if n, err := fd.Seek(ctx, 0, linux.SEEK_SET); n != 0 && err != nil { + t.Errorf("Seek: got err (%v, %v), wanted (0, nil)", n, err) + } + if n, err := fd.Write(ctx, writeIOSeq, WriteOptions{}); int(n) != len(want) && err != nil { + t.Errorf("Write: got err (%v, %v), wanted (%v, nil)", n, err, len(want)) + } + if n, err := fd.PRead(ctx, ioseq, 0, ReadOptions{}); int(n) != len(want) && err != io.EOF { + t.Fatalf("PRead: got (%v, %v), wanted (%v, EOF)", n, err, len(want)) + } + if want == string(buf) { + t.Fatalf("PRead: got %v, wanted %v", string(buf), want) + } + + // Test failure if offset != 0. + if n, err := fd.Seek(ctx, 1, linux.SEEK_SET); n != 0 && err != nil { + t.Errorf("Seek: got err (%v, %v), wanted (0, nil)", n, err) + } + if n, err := fd.Write(ctx, writeIOSeq, WriteOptions{}); n != 0 && err != syserror.EINVAL { + t.Errorf("Write: got err (%v, %v), wanted (0, EINVAL)", n, err) + } + if n, err := fd.PWrite(ctx, writeIOSeq, 2, WriteOptions{}); n != 0 && err != syserror.EINVAL { + t.Errorf("PWrite: got err (%v, %v), wanted (0, EINVAL)", n, err) + } } -- cgit v1.2.3 From 4cb55a7a3b09c430fa2b7197fdc7b84b7e88a6ed Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 28 Jan 2020 18:43:24 -0800 Subject: Prevent arbitrary size allocation when sending UDS messages. Currently, Send() will copy data into a new byte slice without regard to the original size. Size checks should be performed before the allocation takes place. Note that for the sake of performance, we avoid putting the buffer allocation into the critical section. As a result, the size checks need to be performed again within Enqueue() in case the limit has changed. PiperOrigin-RevId: 292058147 --- pkg/sentry/socket/unix/transport/queue.go | 40 ++++++++++++++++++++++++------- pkg/sentry/socket/unix/transport/unix.go | 29 +++++++--------------- 2 files changed, 40 insertions(+), 29 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/unix/transport/queue.go b/pkg/sentry/socket/unix/transport/queue.go index 5dcd3d95e..d8f3ad63d 100644 --- a/pkg/sentry/socket/unix/transport/queue.go +++ b/pkg/sentry/socket/unix/transport/queue.go @@ -18,6 +18,8 @@ import ( "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserr" + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/waiter" ) @@ -100,12 +102,16 @@ func (q *queue) IsWritable() bool { // Enqueue adds an entry to the data queue if room is available. // +// If discardEmpty is true and there are zero bytes of data, the packet is +// dropped. +// // If truncate is true, Enqueue may truncate the message before enqueuing it. -// Otherwise, the entire message must fit. If n < e.Length(), err indicates why. +// Otherwise, the entire message must fit. If l is less than the size of data, +// err indicates why. // // If notify is true, ReaderQueue.Notify must be called: // q.ReaderQueue.Notify(waiter.EventIn) -func (q *queue) Enqueue(e *message, truncate bool) (l int64, notify bool, err *syserr.Error) { +func (q *queue) Enqueue(data [][]byte, c ControlMessages, from tcpip.FullAddress, discardEmpty bool, truncate bool) (l int64, notify bool, err *syserr.Error) { q.mu.Lock() if q.closed { @@ -113,9 +119,16 @@ func (q *queue) Enqueue(e *message, truncate bool) (l int64, notify bool, err *s return 0, false, syserr.ErrClosedForSend } - free := q.limit - q.used + for _, d := range data { + l += int64(len(d)) + } + if discardEmpty && l == 0 { + q.mu.Unlock() + c.Release() + return 0, false, nil + } - l = e.Length() + free := q.limit - q.used if l > free && truncate { if free == 0 { @@ -124,8 +137,7 @@ func (q *queue) Enqueue(e *message, truncate bool) (l int64, notify bool, err *s return 0, false, syserr.ErrWouldBlock } - e.Truncate(free) - l = e.Length() + l = free err = syserr.ErrWouldBlock } @@ -136,14 +148,26 @@ func (q *queue) Enqueue(e *message, truncate bool) (l int64, notify bool, err *s } if l > free { - // Message can't fit right now. + // Message can't fit right now, and could not be truncated. q.mu.Unlock() return 0, false, syserr.ErrWouldBlock } + // Aggregate l bytes of data. This will truncate the data if l is less than + // the total bytes held in data. + v := make([]byte, l) + for i, b := 0, v; i < len(data) && len(b) > 0; i++ { + n := copy(b, data[i]) + b = b[n:] + } + notify = q.dataList.Front() == nil q.used += l - q.dataList.PushBack(e) + q.dataList.PushBack(&message{ + Data: buffer.View(v), + Control: c, + Address: from, + }) q.mu.Unlock() diff --git a/pkg/sentry/socket/unix/transport/unix.go b/pkg/sentry/socket/unix/transport/unix.go index dcbafe0e5..2ef654235 100644 --- a/pkg/sentry/socket/unix/transport/unix.go +++ b/pkg/sentry/socket/unix/transport/unix.go @@ -581,7 +581,7 @@ type ConnectedEndpoint interface { // // syserr.ErrWouldBlock can be returned along with a partial write if // the caller should block to send the rest of the data. - Send(data [][]byte, controlMessages ControlMessages, from tcpip.FullAddress) (n int64, notify bool, err *syserr.Error) + Send(data [][]byte, c ControlMessages, from tcpip.FullAddress) (n int64, notify bool, err *syserr.Error) // SendNotify notifies the ConnectedEndpoint of a successful Send. This // must not be called while holding any endpoint locks. @@ -653,35 +653,22 @@ func (e *connectedEndpoint) GetLocalAddress() (tcpip.FullAddress, *tcpip.Error) } // Send implements ConnectedEndpoint.Send. -func (e *connectedEndpoint) Send(data [][]byte, controlMessages ControlMessages, from tcpip.FullAddress) (int64, bool, *syserr.Error) { - var l int64 - for _, d := range data { - l += int64(len(d)) - } - +func (e *connectedEndpoint) Send(data [][]byte, c ControlMessages, from tcpip.FullAddress) (int64, bool, *syserr.Error) { + discardEmpty := false truncate := false if e.endpoint.Type() == linux.SOCK_STREAM { - // Since stream sockets don't preserve message boundaries, we - // can write only as much of the message as fits in the queue. - truncate = true - // Discard empty stream packets. Since stream sockets don't // preserve message boundaries, sending zero bytes is a no-op. // In Linux, the receiver actually uses a zero-length receive // as an indication that the stream was closed. - if l == 0 { - controlMessages.Release() - return 0, false, nil - } - } + discardEmpty = true - v := make([]byte, 0, l) - for _, d := range data { - v = append(v, d...) + // Since stream sockets don't preserve message boundaries, we + // can write only as much of the message as fits in the queue. + truncate = true } - l, notify, err := e.writeQueue.Enqueue(&message{Data: buffer.View(v), Control: controlMessages, Address: from}, truncate) - return int64(l), notify, err + return e.writeQueue.Enqueue(data, c, from, discardEmpty, truncate) } // SendNotify implements ConnectedEndpoint.SendNotify. -- cgit v1.2.3 From 89957c6c87b5ad5c7bac68f93d9472388db57702 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Wed, 29 Jan 2020 04:45:59 -0500 Subject: Lazy-fpsimd support patch series#2: add fpsimd@Arm64 support to kvm module Add fpsimd support to KVM module so that the test case "TestKernelFloatingPoint" can be passed on Arm64 platform. Signed-off-by: Bin Lu --- pkg/sentry/platform/kvm/bluepill_arm64.go | 30 +++++++++++++++++ pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go | 16 +++++++++ pkg/sentry/platform/kvm/kvm_const_arm64.go | 41 ++++++++++++------------ pkg/sentry/platform/kvm/machine_arm64.go | 4 +++ pkg/sentry/platform/kvm/machine_arm64_unsafe.go | 29 +++-------------- pkg/sentry/platform/ring0/defs_arm64.go | 6 ++++ pkg/sentry/platform/ring0/lib_arm64.go | 4 +-- 7 files changed, 83 insertions(+), 47 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/kvm/bluepill_arm64.go b/pkg/sentry/platform/kvm/bluepill_arm64.go index 552341721..c215d443c 100644 --- a/pkg/sentry/platform/kvm/bluepill_arm64.go +++ b/pkg/sentry/platform/kvm/bluepill_arm64.go @@ -54,6 +54,14 @@ func bluepillArchExit(c *vCPU, context *arch.SignalContext64) { context.Pstate = regs.Pstate context.Pstate &^= uint64(ring0.UserFlagsClear) context.Pstate |= ring0.UserFlagsSet + + lazyVfp := c.GetLazyVFP() + if lazyVfp != 0 { + fpsimd := fpsimdPtr((*byte)(c.floatingPointState)) + context.Fpsimd64.Fpsr = fpsimd.Fpsr + context.Fpsimd64.Fpcr = fpsimd.Fpcr + context.Fpsimd64.Vregs = fpsimd.Vregs + } } // KernelSyscall handles kernel syscalls. @@ -64,6 +72,17 @@ func (c *vCPU) KernelSyscall() { if regs.Regs[8] != ^uint64(0) { regs.Pc -= 4 // Rewind. } + + vfpEnable := ring0.CPACREL1() + if vfpEnable != 0 { + fpsimd := fpsimdPtr((*byte)(c.floatingPointState)) + fpcr := ring0.GetFPCR() + fpsr := ring0.GetFPSR() + fpsimd.Fpcr = uint32(fpcr) + fpsimd.Fpsr = uint32(fpsr) + ring0.SaveVRegs((*byte)(c.floatingPointState)) + } + ring0.Halt() } @@ -75,5 +94,16 @@ func (c *vCPU) KernelException(vector ring0.Vector) { if vector == ring0.Vector(bounce) { regs.Pc = 0 } + + vfpEnable := ring0.CPACREL1() + if vfpEnable != 0 { + fpsimd := fpsimdPtr((*byte)(c.floatingPointState)) + fpcr := ring0.GetFPCR() + fpsr := ring0.GetFPSR() + fpsimd.Fpcr = uint32(fpcr) + fpsimd.Fpsr = uint32(fpsr) + ring0.SaveVRegs((*byte)(c.floatingPointState)) + } + ring0.Halt() } diff --git a/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go b/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go index 2f02c03cf..af093c6ec 100644 --- a/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go @@ -17,10 +17,26 @@ package kvm import ( + "unsafe" + "gvisor.dev/gvisor/pkg/sentry/arch" ) +// fpsimdPtr returns a fpsimd64 for the given address. +// +//go:nosplit +func fpsimdPtr(addr *byte) *arch.FpsimdContext { + return (*arch.FpsimdContext)(unsafe.Pointer(addr)) +} + //go:nosplit func dieArchSetup(c *vCPU, context *arch.SignalContext64, guestRegs *userRegs) { // TODO(gvisor.dev/issue/1249): dieTrampoline supporting for Arm64. } + +// bluepillArchFpContext returns the arch-specific fpsimd context. +// +//go:nosplit +func bluepillArchFpContext(context unsafe.Pointer) *arch.FpsimdContext { + return &((*arch.SignalContext64)(context).Fpsimd64) +} diff --git a/pkg/sentry/platform/kvm/kvm_const_arm64.go b/pkg/sentry/platform/kvm/kvm_const_arm64.go index 5a74c6e36..531ae8b1e 100644 --- a/pkg/sentry/platform/kvm/kvm_const_arm64.go +++ b/pkg/sentry/platform/kvm/kvm_const_arm64.go @@ -19,30 +19,31 @@ const ( _KVM_GET_ONE_REG = 0x4010aeab _KVM_SET_ONE_REG = 0x4010aeac - _KVM_ARM_PREFERRED_TARGET = 0x8020aeaf - _KVM_ARM_VCPU_INIT = 0x4020aeae - _KVM_ARM64_REGS_PSTATE = 0x6030000000100042 - _KVM_ARM64_REGS_SP_EL1 = 0x6030000000100044 - _KVM_ARM64_REGS_R0 = 0x6030000000100000 - _KVM_ARM64_REGS_R1 = 0x6030000000100002 - _KVM_ARM64_REGS_R2 = 0x6030000000100004 - _KVM_ARM64_REGS_R3 = 0x6030000000100006 - _KVM_ARM64_REGS_R8 = 0x6030000000100010 - _KVM_ARM64_REGS_R18 = 0x6030000000100024 - _KVM_ARM64_REGS_PC = 0x6030000000100040 - _KVM_ARM64_REGS_MAIR_EL1 = 0x603000000013c510 - _KVM_ARM64_REGS_TCR_EL1 = 0x603000000013c102 - _KVM_ARM64_REGS_TTBR0_EL1 = 0x603000000013c100 - _KVM_ARM64_REGS_TTBR1_EL1 = 0x603000000013c101 - _KVM_ARM64_REGS_SCTLR_EL1 = 0x603000000013c080 - _KVM_ARM64_REGS_CPACR_EL1 = 0x603000000013c082 - _KVM_ARM64_REGS_VBAR_EL1 = 0x603000000013c600 + _KVM_ARM_TARGET_GENERIC_V8 = 5 + _KVM_ARM_PREFERRED_TARGET = 0x8020aeaf + _KVM_ARM_VCPU_INIT = 0x4020aeae + _KVM_ARM64_REGS_PSTATE = 0x6030000000100042 + _KVM_ARM64_REGS_SP_EL1 = 0x6030000000100044 + _KVM_ARM64_REGS_R0 = 0x6030000000100000 + _KVM_ARM64_REGS_R1 = 0x6030000000100002 + _KVM_ARM64_REGS_R2 = 0x6030000000100004 + _KVM_ARM64_REGS_R3 = 0x6030000000100006 + _KVM_ARM64_REGS_R8 = 0x6030000000100010 + _KVM_ARM64_REGS_R18 = 0x6030000000100024 + _KVM_ARM64_REGS_PC = 0x6030000000100040 + _KVM_ARM64_REGS_MAIR_EL1 = 0x603000000013c510 + _KVM_ARM64_REGS_TCR_EL1 = 0x603000000013c102 + _KVM_ARM64_REGS_TTBR0_EL1 = 0x603000000013c100 + _KVM_ARM64_REGS_TTBR1_EL1 = 0x603000000013c101 + _KVM_ARM64_REGS_SCTLR_EL1 = 0x603000000013c080 + _KVM_ARM64_REGS_CPACR_EL1 = 0x603000000013c082 + _KVM_ARM64_REGS_VBAR_EL1 = 0x603000000013c600 ) // Arm64: Architectural Feature Access Control Register EL1. const ( - _FPEN_NOTRAP = 0x3 - _FPEN_SHIFT = 0x20 + _FPEN_NOTRAP = 3 + _FPEN_SHIFT = 20 ) // Arm64: System Control Register EL1. diff --git a/pkg/sentry/platform/kvm/machine_arm64.go b/pkg/sentry/platform/kvm/machine_arm64.go index 09552837a..e42505542 100644 --- a/pkg/sentry/platform/kvm/machine_arm64.go +++ b/pkg/sentry/platform/kvm/machine_arm64.go @@ -28,6 +28,10 @@ type vCPUArchState struct { // // This starts above fixedKernelPCID. PCIDs *pagetables.PCIDs + + // floatingPointState is the floating point state buffer used in guest + // to host transitions. See usage in bluepill_arm64.go. + floatingPointState *arch.FloatingPointData } const ( diff --git a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go index 1c8384e6b..00801dee6 100644 --- a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go @@ -29,30 +29,6 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) -// setMemoryRegion initializes a region. -// -// This may be called from bluepillHandler, and therefore returns an errno -// directly (instead of wrapping in an error) to avoid allocations. -// -//go:nosplit -func (m *machine) setMemoryRegion(slot int, physical, length, virtual uintptr) syscall.Errno { - userRegion := userMemoryRegion{ - slot: uint32(slot), - flags: 0, - guestPhysAddr: uint64(physical), - memorySize: uint64(length), - userspaceAddr: uint64(virtual), - } - - // Set the region. - _, _, errno := syscall.RawSyscall( - syscall.SYS_IOCTL, - uintptr(m.fd), - _KVM_SET_USER_MEMORY_REGION, - uintptr(unsafe.Pointer(&userRegion))) - return errno -} - type kvmVcpuInit struct { target uint32 features [7]uint32 @@ -147,6 +123,7 @@ func (c *vCPU) initArchState() error { reg.addr = uint64(reflect.ValueOf(&data).Pointer()) regGet.addr = uint64(reflect.ValueOf(&dataGet).Pointer()) + vcpuInit.target = _KVM_ARM_TARGET_GENERIC_V8 vcpuInit.features[0] |= (1 << _KVM_ARM_VCPU_PSCI_0_2) if _, _, errno := syscall.RawSyscall( syscall.SYS_IOCTL, @@ -158,7 +135,8 @@ func (c *vCPU) initArchState() error { // cpacr_el1 reg.id = _KVM_ARM64_REGS_CPACR_EL1 - data = (_FPEN_NOTRAP << _FPEN_SHIFT) + // It is off by default, and it is turned on only when in use. + data = 0 // Disable fpsimd. if err := c.setOneRegister(®); err != nil { return err } @@ -250,6 +228,7 @@ func (c *vCPU) initArchState() error { return err } + c.floatingPointState = arch.NewFloatingPointData() return nil } diff --git a/pkg/sentry/platform/ring0/defs_arm64.go b/pkg/sentry/platform/ring0/defs_arm64.go index 1583dda12..0e2ab716c 100644 --- a/pkg/sentry/platform/ring0/defs_arm64.go +++ b/pkg/sentry/platform/ring0/defs_arm64.go @@ -124,6 +124,12 @@ func (c *CPU) SetAppAddr(value uintptr) { c.appAddr = value } +// GetLazyVFP returns the value of cpacr_el1. +//go:nosplit +func (c *CPU) GetLazyVFP() (value uintptr) { + return c.lazyVFP +} + // SwitchArchOpts are embedded in SwitchOpts. type SwitchArchOpts struct { // UserASID indicates that the application ASID to be used on switch, diff --git a/pkg/sentry/platform/ring0/lib_arm64.go b/pkg/sentry/platform/ring0/lib_arm64.go index af075aae4..80922f43d 100644 --- a/pkg/sentry/platform/ring0/lib_arm64.go +++ b/pkg/sentry/platform/ring0/lib_arm64.go @@ -20,13 +20,13 @@ package ring0 func CPACREL1() (value uintptr) // FPCR returns the value of FPCR register. -func FPCR() (value uintptr) +func GetFPCR() (value uintptr) // SetFPCR writes the FPCR value. func SetFPCR(value uintptr) // FPSR returns the value of FPSR register. -func FPSR() (value uintptr) +func GetFPSR() (value uintptr) // SetFPSR writes the FPSR value. func SetFPSR(value uintptr) -- cgit v1.2.3 From 6adbdfe232c3da42a7f6f3a7d882d140196e4068 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Wed, 29 Jan 2020 07:50:34 -0500 Subject: supporting sError in guest kernel on Arm64 For test case 'TestBounce', we use KVM_SET_VCPU_EVENTS to trigger sError to leave guest. Signed-off-by: Bin Lu --- pkg/sentry/platform/ring0/aarch64.go | 6 +++--- pkg/sentry/platform/ring0/entry_arm64.s | 14 +++++++++++++- pkg/sentry/platform/ring0/offsets_arm64.go | 1 + 3 files changed, 17 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/ring0/aarch64.go b/pkg/sentry/platform/ring0/aarch64.go index 6b078cd1e..f6da41c27 100644 --- a/pkg/sentry/platform/ring0/aarch64.go +++ b/pkg/sentry/platform/ring0/aarch64.go @@ -88,14 +88,14 @@ const ( El0Sync_undef El0Sync_dbg El0Sync_inv - VirtualizationException _NR_INTERRUPTS ) // System call vectors. const ( - Syscall Vector = El0Sync_svc - PageFault Vector = El0Sync_da + Syscall Vector = El0Sync_svc + PageFault Vector = El0Sync_da + VirtualizationException Vector = El0Error ) // VirtualAddressBits returns the number bits available for virtual addresses. diff --git a/pkg/sentry/platform/ring0/entry_arm64.s b/pkg/sentry/platform/ring0/entry_arm64.s index 679842288..baa6c4910 100644 --- a/pkg/sentry/platform/ring0/entry_arm64.s +++ b/pkg/sentry/platform/ring0/entry_arm64.s @@ -601,7 +601,19 @@ TEXT ·El0_fiq(SB),NOSPLIT,$0 B ·Shutdown(SB) TEXT ·El0_error(SB),NOSPLIT,$0 - B ·Shutdown(SB) + KERNEL_ENTRY_FROM_EL0 + WORD $0xd538d092 //MRS TPIDR_EL1, R18 + WORD $0xd538601a //MRS FAR_EL1, R26 + + MOVD R26, CPU_FAULT_ADDR(RSV_REG) + + MOVD $1, R3 + MOVD R3, CPU_ERROR_TYPE(RSV_REG) // Set error type to user. + + MOVD $VirtualizationException, R3 + MOVD R3, CPU_VECTOR_CODE(RSV_REG) + + B ·Halt(SB) TEXT ·El0_sync_invalid(SB),NOSPLIT,$0 B ·Shutdown(SB) diff --git a/pkg/sentry/platform/ring0/offsets_arm64.go b/pkg/sentry/platform/ring0/offsets_arm64.go index 8c960c749..057fb5c69 100644 --- a/pkg/sentry/platform/ring0/offsets_arm64.go +++ b/pkg/sentry/platform/ring0/offsets_arm64.go @@ -85,6 +85,7 @@ func Emit(w io.Writer) { fmt.Fprintf(w, "#define PageFault 0x%02x\n", PageFault) fmt.Fprintf(w, "#define Syscall 0x%02x\n", Syscall) + fmt.Fprintf(w, "#define VirtualizationException 0x%02x\n", VirtualizationException) p := &syscall.PtraceRegs{} fmt.Fprintf(w, "\n// Ptrace registers.\n") -- cgit v1.2.3 From 8dcedc953a610b97efe9f68ac8fecf5e15a7e26b Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Wed, 29 Jan 2020 10:08:25 -0800 Subject: Add //pkg/sentry/devices/memdev. PiperOrigin-RevId: 292165063 --- pkg/abi/linux/dev.go | 3 ++ pkg/sentry/devices/memdev/BUILD | 28 +++++++++++ pkg/sentry/devices/memdev/full.go | 75 ++++++++++++++++++++++++++++++ pkg/sentry/devices/memdev/memdev.go | 59 ++++++++++++++++++++++++ pkg/sentry/devices/memdev/null.go | 76 ++++++++++++++++++++++++++++++ pkg/sentry/devices/memdev/random.go | 92 +++++++++++++++++++++++++++++++++++++ pkg/sentry/devices/memdev/zero.go | 88 +++++++++++++++++++++++++++++++++++ 7 files changed, 421 insertions(+) create mode 100644 pkg/sentry/devices/memdev/BUILD create mode 100644 pkg/sentry/devices/memdev/full.go create mode 100644 pkg/sentry/devices/memdev/memdev.go create mode 100644 pkg/sentry/devices/memdev/null.go create mode 100644 pkg/sentry/devices/memdev/random.go create mode 100644 pkg/sentry/devices/memdev/zero.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/dev.go b/pkg/abi/linux/dev.go index 421e11256..89f9a793f 100644 --- a/pkg/abi/linux/dev.go +++ b/pkg/abi/linux/dev.go @@ -36,6 +36,9 @@ func DecodeDeviceID(rdev uint32) (uint16, uint32) { // // See Documentations/devices.txt and uapi/linux/major.h. const ( + // MEM_MAJOR is the major device number for "memory" character devices. + MEM_MAJOR = 1 + // TTYAUX_MAJOR is the major device number for alternate TTY devices. TTYAUX_MAJOR = 5 diff --git a/pkg/sentry/devices/memdev/BUILD b/pkg/sentry/devices/memdev/BUILD new file mode 100644 index 000000000..abe58f818 --- /dev/null +++ b/pkg/sentry/devices/memdev/BUILD @@ -0,0 +1,28 @@ +load("//tools:defs.bzl", "go_library") + +licenses(["notice"]) + +go_library( + name = "memdev", + srcs = [ + "full.go", + "memdev.go", + "null.go", + "random.go", + "zero.go", + ], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/abi/linux", + "//pkg/context", + "//pkg/rand", + "//pkg/safemem", + "//pkg/sentry/fsimpl/devtmpfs", + "//pkg/sentry/memmap", + "//pkg/sentry/mm", + "//pkg/sentry/pgalloc", + "//pkg/sentry/vfs", + "//pkg/syserror", + "//pkg/usermem", + ], +) diff --git a/pkg/sentry/devices/memdev/full.go b/pkg/sentry/devices/memdev/full.go new file mode 100644 index 000000000..c7e197691 --- /dev/null +++ b/pkg/sentry/devices/memdev/full.go @@ -0,0 +1,75 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package memdev + +import ( + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +const fullDevMinor = 7 + +// fullDevice implements vfs.Device for /dev/full. +type fullDevice struct{} + +// Open implements vfs.Device.Open. +func (fullDevice) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd := &fullFD{} + if err := fd.vfsfd.Init(fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{ + UseDentryMetadata: true, + }); err != nil { + return nil, err + } + return &fd.vfsfd, nil +} + +// fullFD implements vfs.FileDescriptionImpl for /dev/full. +type fullFD struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl + vfs.DentryMetadataFileDescriptionImpl +} + +// Release implements vfs.FileDescriptionImpl.Release. +func (fd *fullFD) Release() { + // noop +} + +// PRead implements vfs.FileDescriptionImpl.PRead. +func (fd *fullFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + return dst.ZeroOut(ctx, dst.NumBytes()) +} + +// Read implements vfs.FileDescriptionImpl.Read. +func (fd *fullFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + return dst.ZeroOut(ctx, dst.NumBytes()) +} + +// PWrite implements vfs.FileDescriptionImpl.PWrite. +func (fd *fullFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + return 0, syserror.ENOSPC +} + +// Write implements vfs.FileDescriptionImpl.Write. +func (fd *fullFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + return 0, syserror.ENOSPC +} + +// Seek implements vfs.FileDescriptionImpl.Seek. +func (fd *fullFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + return 0, nil +} diff --git a/pkg/sentry/devices/memdev/memdev.go b/pkg/sentry/devices/memdev/memdev.go new file mode 100644 index 000000000..5759900c4 --- /dev/null +++ b/pkg/sentry/devices/memdev/memdev.go @@ -0,0 +1,59 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package memdev implements "mem" character devices, as implemented in Linux +// by drivers/char/mem.c and drivers/char/random.c. +package memdev + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/devtmpfs" + "gvisor.dev/gvisor/pkg/sentry/vfs" +) + +// Register registers all devices implemented by this package in vfsObj. +func Register(vfsObj *vfs.VirtualFilesystem) error { + for minor, dev := range map[uint32]vfs.Device{ + nullDevMinor: nullDevice{}, + zeroDevMinor: zeroDevice{}, + fullDevMinor: fullDevice{}, + randomDevMinor: randomDevice{}, + urandomDevMinor: randomDevice{}, + } { + if err := vfsObj.RegisterDevice(vfs.CharDevice, linux.MEM_MAJOR, minor, dev, &vfs.RegisterDeviceOptions{ + GroupName: "mem", + }); err != nil { + return err + } + } + return nil +} + +// CreateDevtmpfsFiles creates device special files in dev representing all +// devices implemented by this package. +func CreateDevtmpfsFiles(ctx context.Context, dev *devtmpfs.Accessor) error { + for minor, name := range map[uint32]string{ + nullDevMinor: "null", + zeroDevMinor: "zero", + fullDevMinor: "full", + randomDevMinor: "random", + urandomDevMinor: "urandom", + } { + if err := dev.CreateDeviceFile(ctx, name, vfs.CharDevice, linux.MEM_MAJOR, minor, 0666 /* mode */); err != nil { + return err + } + } + return nil +} diff --git a/pkg/sentry/devices/memdev/null.go b/pkg/sentry/devices/memdev/null.go new file mode 100644 index 000000000..33d060d02 --- /dev/null +++ b/pkg/sentry/devices/memdev/null.go @@ -0,0 +1,76 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package memdev + +import ( + "io" + + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/usermem" +) + +const nullDevMinor = 3 + +// nullDevice implements vfs.Device for /dev/null. +type nullDevice struct{} + +// Open implements vfs.Device.Open. +func (nullDevice) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd := &nullFD{} + if err := fd.vfsfd.Init(fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{ + UseDentryMetadata: true, + }); err != nil { + return nil, err + } + return &fd.vfsfd, nil +} + +// nullFD implements vfs.FileDescriptionImpl for /dev/null. +type nullFD struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl + vfs.DentryMetadataFileDescriptionImpl +} + +// Release implements vfs.FileDescriptionImpl.Release. +func (fd *nullFD) Release() { + // noop +} + +// PRead implements vfs.FileDescriptionImpl.PRead. +func (fd *nullFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + return 0, io.EOF +} + +// Read implements vfs.FileDescriptionImpl.Read. +func (fd *nullFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + return 0, io.EOF +} + +// PWrite implements vfs.FileDescriptionImpl.PWrite. +func (fd *nullFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + return src.NumBytes(), nil +} + +// Write implements vfs.FileDescriptionImpl.Write. +func (fd *nullFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + return src.NumBytes(), nil +} + +// Seek implements vfs.FileDescriptionImpl.Seek. +func (fd *nullFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + return 0, nil +} diff --git a/pkg/sentry/devices/memdev/random.go b/pkg/sentry/devices/memdev/random.go new file mode 100644 index 000000000..acfa23149 --- /dev/null +++ b/pkg/sentry/devices/memdev/random.go @@ -0,0 +1,92 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package memdev + +import ( + "sync/atomic" + + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/rand" + "gvisor.dev/gvisor/pkg/safemem" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/usermem" +) + +const ( + randomDevMinor = 8 + urandomDevMinor = 9 +) + +// randomDevice implements vfs.Device for /dev/random and /dev/urandom. +type randomDevice struct{} + +// Open implements vfs.Device.Open. +func (randomDevice) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd := &randomFD{} + if err := fd.vfsfd.Init(fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{ + UseDentryMetadata: true, + }); err != nil { + return nil, err + } + return &fd.vfsfd, nil +} + +// randomFD implements vfs.FileDescriptionImpl for /dev/random. +type randomFD struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl + vfs.DentryMetadataFileDescriptionImpl + + // off is the "file offset". off is accessed using atomic memory + // operations. + off int64 +} + +// Release implements vfs.FileDescriptionImpl.Release. +func (fd *randomFD) Release() { + // noop +} + +// PRead implements vfs.FileDescriptionImpl.PRead. +func (fd *randomFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + return dst.CopyOutFrom(ctx, safemem.FromIOReader{rand.Reader}) +} + +// Read implements vfs.FileDescriptionImpl.Read. +func (fd *randomFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + n, err := dst.CopyOutFrom(ctx, safemem.FromIOReader{rand.Reader}) + atomic.AddInt64(&fd.off, n) + return n, err +} + +// PWrite implements vfs.FileDescriptionImpl.PWrite. +func (fd *randomFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + // In Linux, this mixes the written bytes into the entropy pool; we just + // throw them away. + return src.NumBytes(), nil +} + +// Write implements vfs.FileDescriptionImpl.Write. +func (fd *randomFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + atomic.AddInt64(&fd.off, src.NumBytes()) + return src.NumBytes(), nil +} + +// Seek implements vfs.FileDescriptionImpl.Seek. +func (fd *randomFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + // Linux: drivers/char/random.c:random_fops.llseek == urandom_fops.llseek + // == noop_llseek + return atomic.LoadInt64(&fd.off), nil +} diff --git a/pkg/sentry/devices/memdev/zero.go b/pkg/sentry/devices/memdev/zero.go new file mode 100644 index 000000000..3b1372b9e --- /dev/null +++ b/pkg/sentry/devices/memdev/zero.go @@ -0,0 +1,88 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package memdev + +import ( + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/memmap" + "gvisor.dev/gvisor/pkg/sentry/mm" + "gvisor.dev/gvisor/pkg/sentry/pgalloc" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/usermem" +) + +const zeroDevMinor = 5 + +// zeroDevice implements vfs.Device for /dev/zero. +type zeroDevice struct{} + +// Open implements vfs.Device.Open. +func (zeroDevice) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd := &zeroFD{} + if err := fd.vfsfd.Init(fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{ + UseDentryMetadata: true, + }); err != nil { + return nil, err + } + return &fd.vfsfd, nil +} + +// zeroFD implements vfs.FileDescriptionImpl for /dev/zero. +type zeroFD struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl + vfs.DentryMetadataFileDescriptionImpl +} + +// Release implements vfs.FileDescriptionImpl.Release. +func (fd *zeroFD) Release() { + // noop +} + +// PRead implements vfs.FileDescriptionImpl.PRead. +func (fd *zeroFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + return dst.ZeroOut(ctx, dst.NumBytes()) +} + +// Read implements vfs.FileDescriptionImpl.Read. +func (fd *zeroFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + return dst.ZeroOut(ctx, dst.NumBytes()) +} + +// PWrite implements vfs.FileDescriptionImpl.PWrite. +func (fd *zeroFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + return src.NumBytes(), nil +} + +// Write implements vfs.FileDescriptionImpl.Write. +func (fd *zeroFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + return src.NumBytes(), nil +} + +// Seek implements vfs.FileDescriptionImpl.Seek. +func (fd *zeroFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + return 0, nil +} + +// ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. +func (fd *zeroFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { + m, err := mm.NewSharedAnonMappable(opts.Length, pgalloc.MemoryFileProviderFromContext(ctx)) + if err != nil { + return err + } + opts.MappingIdentity = m + opts.Mappable = m + return nil +} -- cgit v1.2.3 From 37bb502670caefd4113da062495b4e318ea0f72e Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Wed, 29 Jan 2020 10:35:32 -0800 Subject: sentry: rename SetRSEQInterruptedIP to SetOldRSeqInterruptedIP for arm64 For amd64, this has been done on cl/288342928. PiperOrigin-RevId: 292170856 --- pkg/sentry/arch/arch_arm64.go | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/arch/arch_arm64.go b/pkg/sentry/arch/arch_arm64.go index 94f1a808f..ac98897b5 100644 --- a/pkg/sentry/arch/arch_arm64.go +++ b/pkg/sentry/arch/arch_arm64.go @@ -137,8 +137,8 @@ func (c *context64) SetTLS(value uintptr) bool { return false } -// SetRSEQInterruptedIP implements Context.SetRSEQInterruptedIP. -func (c *context64) SetRSEQInterruptedIP(value uintptr) { +// SetOldRSeqInterruptedIP implements Context.SetOldRSeqInterruptedIP. +func (c *context64) SetOldRSeqInterruptedIP(value uintptr) { c.Regs.Regs[3] = uint64(value) } -- cgit v1.2.3 From 148fda60e8dee29f2df85e3104e3d5de1a225bcf Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Wed, 29 Jan 2020 11:15:59 -0800 Subject: Add plumbing for file locks in VFS2. Updates #1480 PiperOrigin-RevId: 292180192 --- pkg/sentry/vfs/BUILD | 1 + pkg/sentry/vfs/file_description.go | 21 ++++++++++++++++++++- pkg/sentry/vfs/file_description_impl_util.go | 21 +++++++++++++++++++++ 3 files changed, 42 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index ced9d07b1..14b39eb9d 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -44,6 +44,7 @@ go_library( "//pkg/context", "//pkg/fspath", "//pkg/sentry/arch", + "//pkg/sentry/fs/lock", "//pkg/sentry/kernel/auth", "//pkg/sentry/memmap", "//pkg/sync", diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index badacb55e..5bac660c7 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -20,6 +20,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sync" @@ -393,7 +394,25 @@ type FileDescriptionImpl interface { // Removexattr removes the given extended attribute from the file. Removexattr(ctx context.Context, name string) error - // TODO: file locking + // LockBSD tries to acquire a BSD-style advisory file lock. + // + // TODO(gvisor.dev/issue/1480): BSD-style file locking + LockBSD(ctx context.Context, uid lock.UniqueID, t lock.LockType, block lock.Blocker) error + + // LockBSD releases a BSD-style advisory file lock. + // + // TODO(gvisor.dev/issue/1480): BSD-style file locking + UnlockBSD(ctx context.Context, uid lock.UniqueID) error + + // LockPOSIX tries to acquire a POSIX-style advisory file lock. + // + // TODO(gvisor.dev/issue/1480): POSIX-style file locking + LockPOSIX(ctx context.Context, uid lock.UniqueID, t lock.LockType, rng lock.LockRange, block lock.Blocker) error + + // UnlockPOSIX releases a POSIX-style advisory file lock. + // + // TODO(gvisor.dev/issue/1480): POSIX-style file locking + UnlockPOSIX(ctx context.Context, uid lock.UniqueID, rng lock.LockRange) error } // Dirent holds the information contained in struct linux_dirent64. diff --git a/pkg/sentry/vfs/file_description_impl_util.go b/pkg/sentry/vfs/file_description_impl_util.go index a4900c170..c2a52ec1b 100644 --- a/pkg/sentry/vfs/file_description_impl_util.go +++ b/pkg/sentry/vfs/file_description_impl_util.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" @@ -152,6 +153,26 @@ func (FileDescriptionDefaultImpl) Removexattr(ctx context.Context, name string) return syserror.ENOTSUP } +// LockBSD implements FileDescriptionImpl.LockBSD. +func (FileDescriptionDefaultImpl) LockBSD(ctx context.Context, uid lock.UniqueID, t lock.LockType, block lock.Blocker) error { + return syserror.EBADF +} + +// UnlockBSD implements FileDescriptionImpl.UnlockBSD. +func (FileDescriptionDefaultImpl) UnlockBSD(ctx context.Context, uid lock.UniqueID) error { + return syserror.EBADF +} + +// LockPOSIX implements FileDescriptionImpl.LockPOSIX. +func (FileDescriptionDefaultImpl) LockPOSIX(ctx context.Context, uid lock.UniqueID, t lock.LockType, rng lock.LockRange, block lock.Blocker) error { + return syserror.EBADF +} + +// UnlockPOSIX implements FileDescriptionImpl.UnlockPOSIX. +func (FileDescriptionDefaultImpl) UnlockPOSIX(ctx context.Context, uid lock.UniqueID, rng lock.LockRange) error { + return syserror.EBADF +} + // DirectoryFileDescriptionDefaultImpl may be embedded by implementations of // FileDescriptionImpl that always represent directories to obtain // implementations of non-directory I/O methods that return EISDIR. -- cgit v1.2.3 From 51b783505b1ec164b02b48a0fd234509fba01a73 Mon Sep 17 00:00:00 2001 From: Bhasker Hariharan Date: Wed, 29 Jan 2020 15:41:51 -0800 Subject: Add support for TCP_DEFER_ACCEPT. PiperOrigin-RevId: 292233574 --- pkg/sentry/socket/netstack/netstack.go | 22 ++++ pkg/tcpip/tcpip.go | 6 ++ pkg/tcpip/transport/tcp/BUILD | 1 + pkg/tcpip/transport/tcp/accept.go | 25 ++--- pkg/tcpip/transport/tcp/connect.go | 53 +++++++++- pkg/tcpip/transport/tcp/endpoint.go | 26 ++++- pkg/tcpip/transport/tcp/forwarder.go | 4 +- pkg/tcpip/transport/tcp/tcp_test.go | 126 ++++++++++++++++++++++ test/syscalls/linux/socket_inet_loopback.cc | 158 ++++++++++++++++++++++++++++ test/syscalls/linux/tcp_socket.cc | 53 ++++++++++ 10 files changed, 451 insertions(+), 23 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 8619cc506..049d04bf2 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -1260,6 +1260,18 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa return int32(time.Duration(v) / time.Second), nil + case linux.TCP_DEFER_ACCEPT: + if outLen < sizeOfInt32 { + return nil, syserr.ErrInvalidArgument + } + + var v tcpip.TCPDeferAcceptOption + if err := ep.GetSockOpt(&v); err != nil { + return nil, syserr.TranslateNetstackError(err) + } + + return int32(time.Duration(v) / time.Second), nil + default: emitUnimplementedEventTCP(t, name) } @@ -1713,6 +1725,16 @@ func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) * v := usermem.ByteOrder.Uint32(optVal) return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.TCPLingerTimeoutOption(time.Second * time.Duration(v)))) + case linux.TCP_DEFER_ACCEPT: + if len(optVal) < sizeOfInt32 { + return syserr.ErrInvalidArgument + } + v := int32(usermem.ByteOrder.Uint32(optVal)) + if v < 0 { + v = 0 + } + return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.TCPDeferAcceptOption(time.Second * time.Duration(v)))) + case linux.TCP_REPAIR_OPTIONS: t.Kernel().EmitUnimplementedEvent(t) diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 59c9b3fb0..0fa141d58 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -626,6 +626,12 @@ type TCPLingerTimeoutOption time.Duration // before being marked closed. type TCPTimeWaitTimeoutOption time.Duration +// TCPDeferAcceptOption is used by SetSockOpt/GetSockOpt to allow a +// accept to return a completed connection only when there is data to be +// read. This usually means the listening socket will drop the final ACK +// for a handshake till the specified timeout until a segment with data arrives. +type TCPDeferAcceptOption time.Duration + // MulticastTTLOption is used by SetSockOpt/GetSockOpt to control the default // TTL value for multicast messages. The default is 1. type MulticastTTLOption uint8 diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD index 4acd9fb9a..7b4a87a2d 100644 --- a/pkg/tcpip/transport/tcp/BUILD +++ b/pkg/tcpip/transport/tcp/BUILD @@ -57,6 +57,7 @@ go_library( imports = ["gvisor.dev/gvisor/pkg/tcpip/buffer"], visibility = ["//visibility:public"], deps = [ + "//pkg/log", "//pkg/rand", "//pkg/sleep", "//pkg/sync", diff --git a/pkg/tcpip/transport/tcp/accept.go b/pkg/tcpip/transport/tcp/accept.go index d469758eb..6101f2945 100644 --- a/pkg/tcpip/transport/tcp/accept.go +++ b/pkg/tcpip/transport/tcp/accept.go @@ -222,13 +222,13 @@ func (l *listenContext) isCookieValid(id stack.TransportEndpointID, cookie seqnu // createConnectingEndpoint creates a new endpoint in a connecting state, with // the connection parameters given by the arguments. -func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, irs seqnum.Value, rcvdSynOpts *header.TCPSynOptions) (*endpoint, *tcpip.Error) { +func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, irs seqnum.Value, rcvdSynOpts *header.TCPSynOptions, queue *waiter.Queue) (*endpoint, *tcpip.Error) { // Create a new endpoint. netProto := l.netProto if netProto == 0 { netProto = s.route.NetProto } - n := newEndpoint(l.stack, netProto, nil) + n := newEndpoint(l.stack, netProto, queue) n.v6only = l.v6only n.ID = s.id n.boundNICID = s.route.NICID() @@ -273,16 +273,17 @@ func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, i // createEndpoint creates a new endpoint in connected state and then performs // the TCP 3-way handshake. -func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *header.TCPSynOptions) (*endpoint, *tcpip.Error) { +func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *header.TCPSynOptions, queue *waiter.Queue) (*endpoint, *tcpip.Error) { // Create new endpoint. irs := s.sequenceNumber isn := generateSecureISN(s.id, l.stack.Seed()) - ep, err := l.createConnectingEndpoint(s, isn, irs, opts) + ep, err := l.createConnectingEndpoint(s, isn, irs, opts, queue) if err != nil { return nil, err } // listenEP is nil when listenContext is used by tcp.Forwarder. + deferAccept := time.Duration(0) if l.listenEP != nil { l.listenEP.mu.Lock() if l.listenEP.EndpointState() != StateListen { @@ -290,13 +291,12 @@ func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *head return nil, tcpip.ErrConnectionAborted } l.addPendingEndpoint(ep) + deferAccept = l.listenEP.deferAccept l.listenEP.mu.Unlock() } // Perform the 3-way handshake. - h := newHandshake(ep, seqnum.Size(ep.initialReceiveWindow())) - - h.resetToSynRcvd(isn, irs, opts) + h := newPassiveHandshake(ep, seqnum.Size(ep.initialReceiveWindow()), isn, irs, opts, deferAccept) if err := h.execute(); err != nil { ep.Close() if l.listenEP != nil { @@ -377,16 +377,14 @@ func (e *endpoint) handleSynSegment(ctx *listenContext, s *segment, opts *header defer e.decSynRcvdCount() defer s.decRef() - n, err := ctx.createEndpointAndPerformHandshake(s, opts) + n, err := ctx.createEndpointAndPerformHandshake(s, opts, &waiter.Queue{}) if err != nil { e.stack.Stats().TCP.FailedConnectionAttempts.Increment() e.stats.FailedConnectionAttempts.Increment() return } ctx.removePendingEndpoint(n) - // Start the protocol goroutine. - wq := &waiter.Queue{} - n.startAcceptedLoop(wq) + n.startAcceptedLoop() e.stack.Stats().TCP.PassiveConnectionOpenings.Increment() e.deliverAccepted(n) @@ -546,7 +544,7 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { rcvdSynOptions.TSEcr = s.parsedOptions.TSEcr } - n, err := ctx.createConnectingEndpoint(s, s.ackNumber-1, s.sequenceNumber-1, rcvdSynOptions) + n, err := ctx.createConnectingEndpoint(s, s.ackNumber-1, s.sequenceNumber-1, rcvdSynOptions, &waiter.Queue{}) if err != nil { e.stack.Stats().TCP.FailedConnectionAttempts.Increment() e.stats.FailedConnectionAttempts.Increment() @@ -576,8 +574,7 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { // space available in the backlog. // Start the protocol goroutine. - wq := &waiter.Queue{} - n.startAcceptedLoop(wq) + n.startAcceptedLoop() e.stack.Stats().TCP.PassiveConnectionOpenings.Increment() go e.deliverAccepted(n) } diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 4e3c5419c..9ff7ac261 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -86,6 +86,19 @@ type handshake struct { // rcvWndScale is the receive window scale, as defined in RFC 1323. rcvWndScale int + + // startTime is the time at which the first SYN/SYN-ACK was sent. + startTime time.Time + + // deferAccept if non-zero will drop the final ACK for a passive + // handshake till an ACK segment with data is received or the timeout is + // hit. + deferAccept time.Duration + + // acked is true if the the final ACK for a 3-way handshake has + // been received. This is required to stop retransmitting the + // original SYN-ACK when deferAccept is enabled. + acked bool } func newHandshake(ep *endpoint, rcvWnd seqnum.Size) handshake { @@ -112,6 +125,12 @@ func newHandshake(ep *endpoint, rcvWnd seqnum.Size) handshake { return h } +func newPassiveHandshake(ep *endpoint, rcvWnd seqnum.Size, isn, irs seqnum.Value, opts *header.TCPSynOptions, deferAccept time.Duration) handshake { + h := newHandshake(ep, rcvWnd) + h.resetToSynRcvd(isn, irs, opts, deferAccept) + return h +} + // FindWndScale determines the window scale to use for the given maximum window // size. func FindWndScale(wnd seqnum.Size) int { @@ -181,7 +200,7 @@ func (h *handshake) effectiveRcvWndScale() uint8 { // resetToSynRcvd resets the state of the handshake object to the SYN-RCVD // state. -func (h *handshake) resetToSynRcvd(iss seqnum.Value, irs seqnum.Value, opts *header.TCPSynOptions) { +func (h *handshake) resetToSynRcvd(iss seqnum.Value, irs seqnum.Value, opts *header.TCPSynOptions, deferAccept time.Duration) { h.active = false h.state = handshakeSynRcvd h.flags = header.TCPFlagSyn | header.TCPFlagAck @@ -189,6 +208,7 @@ func (h *handshake) resetToSynRcvd(iss seqnum.Value, irs seqnum.Value, opts *hea h.ackNum = irs + 1 h.mss = opts.MSS h.sndWndScale = opts.WS + h.deferAccept = deferAccept h.ep.mu.Lock() h.ep.setEndpointState(StateSynRecv) h.ep.mu.Unlock() @@ -352,6 +372,14 @@ func (h *handshake) synRcvdState(s *segment) *tcpip.Error { // We have previously received (and acknowledged) the peer's SYN. If the // peer acknowledges our SYN, the handshake is completed. if s.flagIsSet(header.TCPFlagAck) { + // If deferAccept is not zero and this is a bare ACK and the + // timeout is not hit then drop the ACK. + if h.deferAccept != 0 && s.data.Size() == 0 && time.Since(h.startTime) < h.deferAccept { + h.acked = true + h.ep.stack.Stats().DroppedPackets.Increment() + return nil + } + // If the timestamp option is negotiated and the segment does // not carry a timestamp option then the segment must be dropped // as per https://tools.ietf.org/html/rfc7323#section-3.2. @@ -365,10 +393,16 @@ func (h *handshake) synRcvdState(s *segment) *tcpip.Error { h.ep.updateRecentTimestamp(s.parsedOptions.TSVal, h.ackNum, s.sequenceNumber) } h.state = handshakeCompleted + h.ep.mu.Lock() h.ep.transitionToStateEstablishedLocked(h) + // If the segment has data then requeue it for the receiver + // to process it again once main loop is started. + if s.data.Size() > 0 { + s.incRef() + h.ep.enqueueSegment(s) + } h.ep.mu.Unlock() - return nil } @@ -471,6 +505,7 @@ func (h *handshake) execute() *tcpip.Error { } } + h.startTime = time.Now() // Initialize the resend timer. resendWaker := sleep.Waker{} timeOut := time.Duration(time.Second) @@ -524,11 +559,21 @@ func (h *handshake) execute() *tcpip.Error { switch index, _ := s.Fetch(true); index { case wakerForResend: timeOut *= 2 - if timeOut > 60*time.Second { + if timeOut > MaxRTO { return tcpip.ErrTimeout } rt.Reset(timeOut) - h.ep.sendSynTCP(&h.ep.route, h.ep.ID, h.ep.ttl, h.ep.sendTOS, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) + // Resend the SYN/SYN-ACK only if the following conditions hold. + // - It's an active handshake (deferAccept does not apply) + // - It's a passive handshake and we have not yet got the final-ACK. + // - It's a passive handshake and we got an ACK but deferAccept is + // enabled and we are now past the deferAccept duration. + // The last is required to provide a way for the peer to complete + // the connection with another ACK or data (as ACKs are never + // retransmitted on their own). + if h.active || !h.acked || h.deferAccept != 0 && time.Since(h.startTime) > h.deferAccept { + h.ep.sendSynTCP(&h.ep.route, h.ep.ID, h.ep.ttl, h.ep.sendTOS, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) + } case wakerForNotification: n := h.ep.fetchNotifications() diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 13718ff55..8d52414b7 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -498,6 +498,13 @@ type endpoint struct { // without any data being acked. userTimeout time.Duration + // deferAccept if non-zero specifies a user specified time during + // which the final ACK of a handshake will be dropped provided the + // ACK is a bare ACK and carries no data. If the timeout is crossed then + // the bare ACK is accepted and the connection is delivered to the + // listener. + deferAccept time.Duration + // pendingAccepted is a synchronization primitive used to track number // of connections that are queued up to be delivered to the accepted // channel. We use this to ensure that all goroutines blocked on writing @@ -1574,6 +1581,15 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.mu.Unlock() return nil + case tcpip.TCPDeferAcceptOption: + e.mu.Lock() + if time.Duration(v) > MaxRTO { + v = tcpip.TCPDeferAcceptOption(MaxRTO) + } + e.deferAccept = time.Duration(v) + e.mu.Unlock() + return nil + default: return nil } @@ -1798,6 +1814,12 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { e.mu.Unlock() return nil + case *tcpip.TCPDeferAcceptOption: + e.mu.Lock() + *o = tcpip.TCPDeferAcceptOption(e.deferAccept) + e.mu.Unlock() + return nil + default: return tcpip.ErrUnknownProtocolOption } @@ -2149,9 +2171,8 @@ func (e *endpoint) listen(backlog int) *tcpip.Error { // startAcceptedLoop sets up required state and starts a goroutine with the // main loop for accepted connections. -func (e *endpoint) startAcceptedLoop(waiterQueue *waiter.Queue) { +func (e *endpoint) startAcceptedLoop() { e.mu.Lock() - e.waiterQueue = waiterQueue e.workerRunning = true e.mu.Unlock() wakerInitDone := make(chan struct{}) @@ -2177,7 +2198,6 @@ func (e *endpoint) Accept() (tcpip.Endpoint, *waiter.Queue, *tcpip.Error) { default: return nil, nil, tcpip.ErrWouldBlock } - return n, n.waiterQueue, nil } diff --git a/pkg/tcpip/transport/tcp/forwarder.go b/pkg/tcpip/transport/tcp/forwarder.go index 7eb613be5..c9ee5bf06 100644 --- a/pkg/tcpip/transport/tcp/forwarder.go +++ b/pkg/tcpip/transport/tcp/forwarder.go @@ -157,13 +157,13 @@ func (r *ForwarderRequest) CreateEndpoint(queue *waiter.Queue) (tcpip.Endpoint, TSVal: r.synOptions.TSVal, TSEcr: r.synOptions.TSEcr, SACKPermitted: r.synOptions.SACKPermitted, - }) + }, queue) if err != nil { return nil, err } // Start the protocol goroutine. - ep.startAcceptedLoop(queue) + ep.startAcceptedLoop() return ep, nil } diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index df2fb1071..a12336d47 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -6787,3 +6787,129 @@ func TestIncreaseWindowOnBufferResize(t *testing.T) { ), ) } + +func TestTCPDeferAccept(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.Create(-1) + + if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatal("Bind failed:", err) + } + + if err := c.EP.Listen(10); err != nil { + t.Fatal("Listen failed:", err) + } + + const tcpDeferAccept = 1 * time.Second + if err := c.EP.SetSockOpt(tcpip.TCPDeferAcceptOption(tcpDeferAccept)); err != nil { + t.Fatalf("c.EP.SetSockOpt(TCPDeferAcceptOption(%s) failed: %v", tcpDeferAccept, err) + } + + irs, iss := executeHandshake(t, c, context.TestPort, false /* synCookiesInUse */) + + if _, _, err := c.EP.Accept(); err != tcpip.ErrWouldBlock { + t.Fatalf("c.EP.Accept() returned unexpected error got: %v, want: %s", err, tcpip.ErrWouldBlock) + } + + // Send data. This should result in an acceptable endpoint. + c.SendPacket([]byte{1, 2, 3, 4}, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: irs + 1, + AckNum: iss + 1, + }) + + // Receive ACK for the data we sent. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagAck), + checker.SeqNum(uint32(iss+1)), + checker.AckNum(uint32(irs+5)))) + + // Give a bit of time for the socket to be delivered to the accept queue. + time.Sleep(50 * time.Millisecond) + aep, _, err := c.EP.Accept() + if err != nil { + t.Fatalf("c.EP.Accept() returned unexpected error got: %v, want: nil", err) + } + + aep.Close() + // Closing aep without reading the data should trigger a RST. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck), + checker.SeqNum(uint32(iss+1)), + checker.AckNum(uint32(irs+5)))) +} + +func TestTCPDeferAcceptTimeout(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.Create(-1) + + if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatal("Bind failed:", err) + } + + if err := c.EP.Listen(10); err != nil { + t.Fatal("Listen failed:", err) + } + + const tcpDeferAccept = 1 * time.Second + if err := c.EP.SetSockOpt(tcpip.TCPDeferAcceptOption(tcpDeferAccept)); err != nil { + t.Fatalf("c.EP.SetSockOpt(TCPDeferAcceptOption(%s) failed: %v", tcpDeferAccept, err) + } + + irs, iss := executeHandshake(t, c, context.TestPort, false /* synCookiesInUse */) + + if _, _, err := c.EP.Accept(); err != tcpip.ErrWouldBlock { + t.Fatalf("c.EP.Accept() returned unexpected error got: %v, want: %s", err, tcpip.ErrWouldBlock) + } + + // Sleep for a little of the tcpDeferAccept timeout. + time.Sleep(tcpDeferAccept + 100*time.Millisecond) + + // On timeout expiry we should get a SYN-ACK retransmission. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagAck|header.TCPFlagSyn), + checker.AckNum(uint32(irs)+1))) + + // Send data. This should result in an acceptable endpoint. + c.SendPacket([]byte{1, 2, 3, 4}, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: irs + 1, + AckNum: iss + 1, + }) + + // Receive ACK for the data we sent. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagAck), + checker.SeqNum(uint32(iss+1)), + checker.AckNum(uint32(irs+5)))) + + // Give sometime for the endpoint to be delivered to the accept queue. + time.Sleep(50 * time.Millisecond) + aep, _, err := c.EP.Accept() + if err != nil { + t.Fatalf("c.EP.Accept() returned unexpected error got: %v, want: nil", err) + } + + aep.Close() + // Closing aep without reading the data should trigger a RST. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck), + checker.SeqNum(uint32(iss+1)), + checker.AckNum(uint32(irs+5)))) +} diff --git a/test/syscalls/linux/socket_inet_loopback.cc b/test/syscalls/linux/socket_inet_loopback.cc index 2f9821555..3bf7081b9 100644 --- a/test/syscalls/linux/socket_inet_loopback.cc +++ b/test/syscalls/linux/socket_inet_loopback.cc @@ -828,6 +828,164 @@ TEST_P(SocketInetLoopbackTest, AcceptedInheritsTCPUserTimeout) { EXPECT_EQ(get, kUserTimeout); } +// TODO(gvisor.dev/issue/1688): Partially completed passive endpoints are not +// saved. Enable S/R once issue is fixed. +TEST_P(SocketInetLoopbackTest, TCPDeferAccept_NoRandomSave) { + // TODO(gvisor.dev/issue/1688): Partially completed passive endpoints are not + // saved. Enable S/R issue is fixed. + DisableSave ds; + + auto const& param = GetParam(); + TestAddress const& listener = param.listener; + TestAddress const& connector = param.connector; + + // Create the listening socket. + const FileDescriptor listen_fd = ASSERT_NO_ERRNO_AND_VALUE( + Socket(listener.family(), SOCK_STREAM, IPPROTO_TCP)); + sockaddr_storage listen_addr = listener.addr; + ASSERT_THAT(bind(listen_fd.get(), reinterpret_cast(&listen_addr), + listener.addr_len), + SyscallSucceeds()); + ASSERT_THAT(listen(listen_fd.get(), SOMAXCONN), SyscallSucceeds()); + + // Get the port bound by the listening socket. + socklen_t addrlen = listener.addr_len; + ASSERT_THAT(getsockname(listen_fd.get(), + reinterpret_cast(&listen_addr), &addrlen), + SyscallSucceeds()); + + const uint16_t port = + ASSERT_NO_ERRNO_AND_VALUE(AddrPort(listener.family(), listen_addr)); + + // Set the TCP_DEFER_ACCEPT on the listening socket. + constexpr int kTCPDeferAccept = 3; + ASSERT_THAT(setsockopt(listen_fd.get(), IPPROTO_TCP, TCP_DEFER_ACCEPT, + &kTCPDeferAccept, sizeof(kTCPDeferAccept)), + SyscallSucceeds()); + + // Connect to the listening socket. + FileDescriptor conn_fd = ASSERT_NO_ERRNO_AND_VALUE( + Socket(connector.family(), SOCK_STREAM, IPPROTO_TCP)); + + sockaddr_storage conn_addr = connector.addr; + ASSERT_NO_ERRNO(SetAddrPort(connector.family(), &conn_addr, port)); + ASSERT_THAT(RetryEINTR(connect)(conn_fd.get(), + reinterpret_cast(&conn_addr), + connector.addr_len), + SyscallSucceeds()); + + // Set the listening socket to nonblock so that we can verify that there is no + // connection in queue despite the connect above succeeding since the peer has + // sent no data and TCP_DEFER_ACCEPT is set on the listening socket. Set the + // FD to O_NONBLOCK. + int opts; + ASSERT_THAT(opts = fcntl(listen_fd.get(), F_GETFL), SyscallSucceeds()); + opts |= O_NONBLOCK; + ASSERT_THAT(fcntl(listen_fd.get(), F_SETFL, opts), SyscallSucceeds()); + + ASSERT_THAT(accept(listen_fd.get(), nullptr, nullptr), + SyscallFailsWithErrno(EWOULDBLOCK)); + + // Set FD back to blocking. + opts &= ~O_NONBLOCK; + ASSERT_THAT(fcntl(listen_fd.get(), F_SETFL, opts), SyscallSucceeds()); + + // Now write some data to the socket. + int data = 0; + ASSERT_THAT(RetryEINTR(write)(conn_fd.get(), &data, sizeof(data)), + SyscallSucceedsWithValue(sizeof(data))); + + // This should now cause the connection to complete and be delivered to the + // accept socket. + + // Accept the connection. + auto accepted = + ASSERT_NO_ERRNO_AND_VALUE(Accept(listen_fd.get(), nullptr, nullptr)); + + // Verify that the accepted socket returns the data written. + int get = -1; + ASSERT_THAT(RetryEINTR(recv)(accepted.get(), &get, sizeof(get), 0), + SyscallSucceedsWithValue(sizeof(get))); + + EXPECT_EQ(get, data); +} + +// TODO(gvisor.dev/issue/1688): Partially completed passive endpoints are not +// saved. Enable S/R once issue is fixed. +TEST_P(SocketInetLoopbackTest, TCPDeferAcceptTimeout_NoRandomSave) { + // TODO(gvisor.dev/issue/1688): Partially completed passive endpoints are not + // saved. Enable S/R once issue is fixed. + DisableSave ds; + + auto const& param = GetParam(); + TestAddress const& listener = param.listener; + TestAddress const& connector = param.connector; + + // Create the listening socket. + const FileDescriptor listen_fd = ASSERT_NO_ERRNO_AND_VALUE( + Socket(listener.family(), SOCK_STREAM, IPPROTO_TCP)); + sockaddr_storage listen_addr = listener.addr; + ASSERT_THAT(bind(listen_fd.get(), reinterpret_cast(&listen_addr), + listener.addr_len), + SyscallSucceeds()); + ASSERT_THAT(listen(listen_fd.get(), SOMAXCONN), SyscallSucceeds()); + + // Get the port bound by the listening socket. + socklen_t addrlen = listener.addr_len; + ASSERT_THAT(getsockname(listen_fd.get(), + reinterpret_cast(&listen_addr), &addrlen), + SyscallSucceeds()); + + const uint16_t port = + ASSERT_NO_ERRNO_AND_VALUE(AddrPort(listener.family(), listen_addr)); + + // Set the TCP_DEFER_ACCEPT on the listening socket. + constexpr int kTCPDeferAccept = 3; + ASSERT_THAT(setsockopt(listen_fd.get(), IPPROTO_TCP, TCP_DEFER_ACCEPT, + &kTCPDeferAccept, sizeof(kTCPDeferAccept)), + SyscallSucceeds()); + + // Connect to the listening socket. + FileDescriptor conn_fd = ASSERT_NO_ERRNO_AND_VALUE( + Socket(connector.family(), SOCK_STREAM, IPPROTO_TCP)); + + sockaddr_storage conn_addr = connector.addr; + ASSERT_NO_ERRNO(SetAddrPort(connector.family(), &conn_addr, port)); + ASSERT_THAT(RetryEINTR(connect)(conn_fd.get(), + reinterpret_cast(&conn_addr), + connector.addr_len), + SyscallSucceeds()); + + // Set the listening socket to nonblock so that we can verify that there is no + // connection in queue despite the connect above succeeding since the peer has + // sent no data and TCP_DEFER_ACCEPT is set on the listening socket. Set the + // FD to O_NONBLOCK. + int opts; + ASSERT_THAT(opts = fcntl(listen_fd.get(), F_GETFL), SyscallSucceeds()); + opts |= O_NONBLOCK; + ASSERT_THAT(fcntl(listen_fd.get(), F_SETFL, opts), SyscallSucceeds()); + + // Verify that there is no acceptable connection before TCP_DEFER_ACCEPT + // timeout is hit. + absl::SleepFor(absl::Seconds(kTCPDeferAccept - 1)); + ASSERT_THAT(accept(listen_fd.get(), nullptr, nullptr), + SyscallFailsWithErrno(EWOULDBLOCK)); + + // Set FD back to blocking. + opts &= ~O_NONBLOCK; + ASSERT_THAT(fcntl(listen_fd.get(), F_SETFL, opts), SyscallSucceeds()); + + // Now sleep for a little over the TCP_DEFER_ACCEPT duration. When the timeout + // is hit a SYN-ACK should be retransmitted by the listener as a last ditch + // attempt to complete the connection with or without data. + absl::SleepFor(absl::Seconds(2)); + + // Verify that we have a connection that can be accepted even though no + // data was written. + auto accepted = + ASSERT_NO_ERRNO_AND_VALUE(Accept(listen_fd.get(), nullptr, nullptr)); +} + INSTANTIATE_TEST_SUITE_P( All, SocketInetLoopbackTest, ::testing::Values( diff --git a/test/syscalls/linux/tcp_socket.cc b/test/syscalls/linux/tcp_socket.cc index 33a5ac66c..525ccbd88 100644 --- a/test/syscalls/linux/tcp_socket.cc +++ b/test/syscalls/linux/tcp_socket.cc @@ -1286,6 +1286,59 @@ TEST_P(SimpleTcpSocketTest, SetTCPUserTimeout) { EXPECT_EQ(get, kTCPUserTimeout); } +TEST_P(SimpleTcpSocketTest, SetTCPDeferAcceptNeg) { + FileDescriptor s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + + // -ve TCP_DEFER_ACCEPT is same as setting it to zero. + constexpr int kNeg = -1; + EXPECT_THAT( + setsockopt(s.get(), IPPROTO_TCP, TCP_DEFER_ACCEPT, &kNeg, sizeof(kNeg)), + SyscallSucceeds()); + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT( + getsockopt(s.get(), IPPROTO_TCP, TCP_USER_TIMEOUT, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, 0); +} + +TEST_P(SimpleTcpSocketTest, GetTCPDeferAcceptDefault) { + FileDescriptor s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT( + getsockopt(s.get(), IPPROTO_TCP, TCP_USER_TIMEOUT, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, 0); +} + +TEST_P(SimpleTcpSocketTest, SetTCPDeferAcceptGreaterThanZero) { + FileDescriptor s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + // kTCPDeferAccept is in seconds. + // NOTE: linux translates seconds to # of retries and back from + // #of retries to seconds. Which means only certain values + // translate back exactly. That's why we use 3 here, a value of + // 5 will result in us getting back 7 instead of 5 in the + // getsockopt. + constexpr int kTCPDeferAccept = 3; + ASSERT_THAT(setsockopt(s.get(), IPPROTO_TCP, TCP_DEFER_ACCEPT, + &kTCPDeferAccept, sizeof(kTCPDeferAccept)), + SyscallSucceeds()); + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT( + getsockopt(s.get(), IPPROTO_TCP, TCP_DEFER_ACCEPT, &get, &get_len), + SyscallSucceeds()); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kTCPDeferAccept); +} + INSTANTIATE_TEST_SUITE_P(AllInetTests, SimpleTcpSocketTest, ::testing::Values(AF_INET, AF_INET6)); -- cgit v1.2.3 From ede8dfab3760afc8063c3418f217e52f7ec70d42 Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Thu, 30 Jan 2020 09:13:36 -0800 Subject: Enforce splice offset limits Splice must not allow negative offsets. Writes also must not allow offset + size to overflow int64. Reads are similarly broken, but not just in splice (b/148095030). Reported-by: syzbot+0e1ff0b95fb2859b4190@syzkaller.appspotmail.com PiperOrigin-RevId: 292361208 --- pkg/sentry/fs/tmpfs/inode_file.go | 10 ++++-- pkg/sentry/syscalls/linux/sys_splice.go | 16 ++++------ test/syscalls/linux/splice.cc | 56 +++++++++++++++++++++++++++++++++ 3 files changed, 70 insertions(+), 12 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/tmpfs/inode_file.go b/pkg/sentry/fs/tmpfs/inode_file.go index dabc10662..25abbc151 100644 --- a/pkg/sentry/fs/tmpfs/inode_file.go +++ b/pkg/sentry/fs/tmpfs/inode_file.go @@ -17,6 +17,7 @@ package tmpfs import ( "fmt" "io" + "math" "time" "gvisor.dev/gvisor/pkg/abi/linux" @@ -444,10 +445,15 @@ func (rw *fileReadWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) defer rw.f.dataMu.Unlock() // Compute the range to write. - end := fs.WriteEndOffset(rw.offset, int64(srcs.NumBytes())) - if end == rw.offset { // srcs.NumBytes() == 0? + if srcs.NumBytes() == 0 { + // Nothing to do. return 0, nil } + end := fs.WriteEndOffset(rw.offset, int64(srcs.NumBytes())) + if end == math.MaxInt64 { + // Overflow. + return 0, syserror.EINVAL + } // Check if seals prevent either file growth or all writes. switch { diff --git a/pkg/sentry/syscalls/linux/sys_splice.go b/pkg/sentry/syscalls/linux/sys_splice.go index f43d6c155..fd642834b 100644 --- a/pkg/sentry/syscalls/linux/sys_splice.go +++ b/pkg/sentry/syscalls/linux/sys_splice.go @@ -25,6 +25,10 @@ import ( // doSplice implements a blocking splice operation. func doSplice(t *kernel.Task, outFile, inFile *fs.File, opts fs.SpliceOpts, nonBlocking bool) (int64, error) { + if opts.Length < 0 || opts.SrcStart < 0 || opts.DstStart < 0 { + return 0, syserror.EINVAL + } + var ( total int64 n int64 @@ -82,11 +86,6 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc offsetAddr := args[2].Pointer() count := int64(args[3].SizeT()) - // Don't send a negative number of bytes. - if count < 0 { - return 0, nil, syserror.EINVAL - } - // Get files. inFile := t.GetFile(inFD) if inFile == nil { @@ -136,11 +135,6 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc return 0, nil, err } - // The offset must be valid. - if offset < 0 { - return 0, nil, syserror.EINVAL - } - // Do the splice. n, err = doSplice(t, outFile, inFile, fs.SpliceOpts{ Length: count, @@ -227,6 +221,7 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if _, err := t.CopyIn(outOffset, &offset); err != nil { return 0, nil, err } + // Use the destination offset. opts.DstOffset = true opts.DstStart = offset @@ -244,6 +239,7 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if _, err := t.CopyIn(inOffset, &offset); err != nil { return 0, nil, err } + // Use the source offset. opts.SrcOffset = true opts.SrcStart = offset diff --git a/test/syscalls/linux/splice.cc b/test/syscalls/linux/splice.cc index 85232cb1f..faa1247f6 100644 --- a/test/syscalls/linux/splice.cc +++ b/test/syscalls/linux/splice.cc @@ -60,6 +60,62 @@ TEST(SpliceTest, TwoRegularFiles) { SyscallFailsWithErrno(EINVAL)); } +int memfd_create(const std::string& name, unsigned int flags) { + return syscall(__NR_memfd_create, name.c_str(), flags); +} + +TEST(SpliceTest, NegativeOffset) { + // Create a new pipe. + int fds[2]; + ASSERT_THAT(pipe(fds), SyscallSucceeds()); + const FileDescriptor rfd(fds[0]); + const FileDescriptor wfd(fds[1]); + + // Fill the pipe. + std::vector buf(kPageSize); + RandomizeBuffer(buf.data(), buf.size()); + ASSERT_THAT(write(wfd.get(), buf.data(), buf.size()), + SyscallSucceedsWithValue(kPageSize)); + + // Open the output file as write only. + int fd; + EXPECT_THAT(fd = memfd_create("negative", 0), SyscallSucceeds()); + const FileDescriptor out_fd(fd); + + loff_t out_offset = 0xffffffffffffffffull; + constexpr int kSize = 2; + EXPECT_THAT(splice(rfd.get(), nullptr, out_fd.get(), &out_offset, kSize, 0), + SyscallFailsWithErrno(EINVAL)); +} + +// Write offset + size overflows int64. +// +// This is a regression test for b/148041624. +TEST(SpliceTest, WriteOverflow) { + // Create a new pipe. + int fds[2]; + ASSERT_THAT(pipe(fds), SyscallSucceeds()); + const FileDescriptor rfd(fds[0]); + const FileDescriptor wfd(fds[1]); + + // Fill the pipe. + std::vector buf(kPageSize); + RandomizeBuffer(buf.data(), buf.size()); + ASSERT_THAT(write(wfd.get(), buf.data(), buf.size()), + SyscallSucceedsWithValue(kPageSize)); + + // Open the output file. + int fd; + EXPECT_THAT(fd = memfd_create("overflow", 0), SyscallSucceeds()); + const FileDescriptor out_fd(fd); + + // out_offset + kSize overflows INT64_MAX. + loff_t out_offset = 0x7ffffffffffffffeull; + constexpr int kSize = 3; + EXPECT_THAT(splice(rfd.get(), nullptr, out_fd.get(), &out_offset, kSize, 0), + SyscallFailsWithErrno(EINVAL)); +} + TEST(SpliceTest, SamePipe) { // Create a new pipe. int fds[2]; -- cgit v1.2.3 From bc3a24d62788bd6b881afeda230f1a5550a5709a Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Fri, 31 Jan 2020 11:49:37 -0800 Subject: Internal change. PiperOrigin-RevId: 292587459 --- kokoro/kythe/generate_xrefs.sh | 2 +- pkg/sentry/fs/g3doc/inotify.md | 16 ++++++++-------- pkg/sentry/mm/README.md | 8 ++++---- 3 files changed, 13 insertions(+), 13 deletions(-) (limited to 'pkg/sentry') diff --git a/kokoro/kythe/generate_xrefs.sh b/kokoro/kythe/generate_xrefs.sh index 7a0fbb3cd..323b0f77b 100644 --- a/kokoro/kythe/generate_xrefs.sh +++ b/kokoro/kythe/generate_xrefs.sh @@ -23,7 +23,7 @@ bazel version python3 -V -readonly KYTHE_VERSION='v0.0.39' +readonly KYTHE_VERSION='v0.0.41' readonly WORKDIR="$(mktemp -d)" readonly KYTHE_DIR="${WORKDIR}/kythe-${KYTHE_VERSION}" if [[ -n "$KOKORO_GIT_COMMIT" ]]; then diff --git a/pkg/sentry/fs/g3doc/inotify.md b/pkg/sentry/fs/g3doc/inotify.md index 71a577d9d..85063d4e6 100644 --- a/pkg/sentry/fs/g3doc/inotify.md +++ b/pkg/sentry/fs/g3doc/inotify.md @@ -112,11 +112,11 @@ attempts to queue a new event, it is already holding `fs.Watches.mu`. If we used `Inotify.mu` to also protect the event queue, this would violate the above lock ordering. -[dirent]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/fs/dirent.go -[event]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/fs/inotify_event.go -[fd_table]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/kernel/fd_table.go -[inode]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/fs/inode.go -[inode_watches]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/fs/inode_inotify.go -[inotify]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/fs/inotify.go -[syscall_dir]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/syscalls/linux/ -[watch]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/fs/inotify_watch.go +[dirent]: https://github.com/google/gvisor/blob/master/pkg/sentry/fs/dirent.go +[event]: https://github.com/google/gvisor/blob/master/pkg/sentry/fs/inotify_event.go +[fd_table]: https://github.com/google/gvisor/blob/master/pkg/sentry/kernel/fd_table.go +[inode]: https://github.com/google/gvisor/blob/master/pkg/sentry/fs/inode.go +[inode_watches]: https://github.com/google/gvisor/blob/master/pkg/sentry/fs/inode_inotify.go +[inotify]: https://github.com/google/gvisor/blob/master/pkg/sentry/fs/inotify.go +[syscall_dir]: https://github.com/google/gvisor/blob/master/pkg/sentry/syscalls/linux/ +[watch]: https://github.com/google/gvisor/blob/master/pkg/sentry/fs/inotify_watch.go diff --git a/pkg/sentry/mm/README.md b/pkg/sentry/mm/README.md index e1322e373..f4d43d927 100644 --- a/pkg/sentry/mm/README.md +++ b/pkg/sentry/mm/README.md @@ -274,7 +274,7 @@ In the sentry: methods [`platform.AddressSpace.MapFile` and `platform.AddressSpace.Unmap`][platform]. -[memmap]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/memmap/memmap.go -[mm]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/mm/mm.go -[pgalloc]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/pgalloc/pgalloc.go -[platform]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/platform/platform.go +[memmap]: https://github.com/google/gvisor/blob/master/pkg/sentry/memmap/memmap.go +[mm]: https://github.com/google/gvisor/blob/master/pkg/sentry/mm/mm.go +[pgalloc]: https://github.com/google/gvisor/blob/master/pkg/sentry/pgalloc/pgalloc.go +[platform]: https://github.com/google/gvisor/blob/master/pkg/sentry/platform/platform.go -- cgit v1.2.3 From eba7bdc24d31388ca81eeab251ed2db108f785dc Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Fri, 31 Jan 2020 13:46:13 -0800 Subject: iptables: enable TCP matching with "-m tcp". A couple other things that changed: - There's a proper extension registration system for matchers. Anyone adding another matcher can use tcp_matcher.go or udp_matcher.go as a template. - All logging and use of syserr.Error in the netfilter package happens at the highest possible level (public functions). Lower-level functions just return normal, descriptive golang errors. --- pkg/abi/linux/netfilter.go | 52 ++++++++ pkg/sentry/socket/netfilter/BUILD | 4 + pkg/sentry/socket/netfilter/extensions.go | 98 +++++++++++++++ pkg/sentry/socket/netfilter/netfilter.go | 187 ++++++++--------------------- pkg/sentry/socket/netfilter/tcp_matcher.go | 143 ++++++++++++++++++++++ pkg/sentry/socket/netfilter/udp_matcher.go | 142 ++++++++++++++++++++++ pkg/tcpip/iptables/BUILD | 1 - pkg/tcpip/iptables/types.go | 3 + pkg/tcpip/iptables/udp_matcher.go | 113 ----------------- 9 files changed, 495 insertions(+), 248 deletions(-) create mode 100644 pkg/sentry/socket/netfilter/extensions.go create mode 100644 pkg/sentry/socket/netfilter/tcp_matcher.go create mode 100644 pkg/sentry/socket/netfilter/udp_matcher.go delete mode 100644 pkg/tcpip/iptables/udp_matcher.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/netfilter.go b/pkg/abi/linux/netfilter.go index 8e40bcc62..e4aabb6bb 100644 --- a/pkg/abi/linux/netfilter.go +++ b/pkg/abi/linux/netfilter.go @@ -348,6 +348,58 @@ func goString(cstring []byte) string { return string(cstring) } +// XTTCP holds data for matching TCP packets. It corresponds to struct xt_tcp +// in include/uapi/linux/netfilter/xt_tcpudp.h. +type XTTCP struct { + // SourcePortStart specifies the inclusive start of the range of source + // ports to which the matcher applies. + SourcePortStart uint16 + + // SourcePortEnd specifies the inclusive end of the range of source ports + // to which the matcher applies. + SourcePortEnd uint16 + + // DestinationPortStart specifies the start of the destination port + // range to which the matcher applies. + DestinationPortStart uint16 + + // DestinationPortEnd specifies the start of the destination port + // range to which the matcher applies. + DestinationPortEnd uint16 + + // Option specifies that a particular TCP option must be set. + Option uint8 + + // FlagMask masks the FlagCompare byte when comparing to the TCP flag + // fields. + FlagMask uint8 + + // FlagCompare is binary and-ed with the TCP flag fields. + FlagCompare uint8 + + // InverseFlags flips the meaning of certain fields. See the + // TX_TCP_INV_* flags. + InverseFlags uint8 +} + +// SizeOfXTTCP is the size of an XTTCP. +const SizeOfXTTCP = 12 + +// Flags in XTTCP.InverseFlags. Corresponding constants are in +// include/uapi/linux/netfilter/xt_tcpudp.h. +const ( + // Invert the meaning of SourcePortStart/End. + XT_TCP_INV_SRCPT = 0x01 + // Invert the meaning of DestinationPortStart/End. + XT_TCP_INV_DSTPT = 0x02 + // Invert the meaning of FlagCompare. + XT_TCP_INV_FLAGS = 0x04 + // Invert the meaning of Option. + XT_TCP_INV_OPTION = 0x08 + // Enable all flags. + XT_TCP_INV_MASK = 0x0F +) + // XTUDP holds data for matching UDP packets. It corresponds to struct xt_udp // in include/uapi/linux/netfilter/xt_tcpudp.h. type XTUDP struct { diff --git a/pkg/sentry/socket/netfilter/BUILD b/pkg/sentry/socket/netfilter/BUILD index fa2a2cb66..c91ec7494 100644 --- a/pkg/sentry/socket/netfilter/BUILD +++ b/pkg/sentry/socket/netfilter/BUILD @@ -5,7 +5,10 @@ package(licenses = ["notice"]) go_library( name = "netfilter", srcs = [ + "extensions.go", "netfilter.go", + "tcp_matcher.go", + "udp_matcher.go", ], # This target depends on netstack and should only be used by epsocket, # which is allowed to depend on netstack. @@ -17,6 +20,7 @@ go_library( "//pkg/sentry/kernel", "//pkg/syserr", "//pkg/tcpip", + "//pkg/tcpip/header", "//pkg/tcpip/iptables", "//pkg/tcpip/stack", "//pkg/usermem", diff --git a/pkg/sentry/socket/netfilter/extensions.go b/pkg/sentry/socket/netfilter/extensions.go new file mode 100644 index 000000000..5a4cac84c --- /dev/null +++ b/pkg/sentry/socket/netfilter/extensions.go @@ -0,0 +1,98 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package netfilter + +import ( + "fmt" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/tcpip/iptables" + "gvisor.dev/gvisor/pkg/usermem" +) + +// TODO(gvisor.dev/issue/170): The following per-matcher params should be +// supported: +// - Table name +// - Match size +// - User size +// - Hooks +// - Proto +// - Family + +// matchMarshaler knows how to (un)marshal the matcher named name(). +type matchMarshaler interface { + // name is the matcher name as stored in the xt_entry_match struct. + name() string + + // marshal converts from an iptables.Matcher to an ABI struct. + marshal(matcher iptables.Matcher) []byte + + // unmarshal converts from the ABI matcher struct to an + // iptables.Matcher. + unmarshal(buf []byte, filter iptables.IPHeaderFilter) (iptables.Matcher, error) +} + +var matchMarshalers = map[string]matchMarshaler{} + +// registerMatchMarshaler should be called by match extensions to register them +// with the netfilter package. +func registerMatchMarshaler(mm matchMarshaler) { + if _, ok := matchMarshalers[mm.name()]; ok { + panic(fmt.Sprintf("Multiple matches registered with name %q.", mm.name())) + } + matchMarshalers[mm.name()] = mm +} + +func marshalMatcher(matcher iptables.Matcher) []byte { + matchMaker, ok := matchMarshalers[matcher.Name()] + if !ok { + panic(fmt.Errorf("Unknown matcher of type %T.", matcher)) + } + return matchMaker.marshal(matcher) +} + +// marshalEntryMatch creates a marshalled XTEntryMatch with the given name and +// data appended at the end. +func marshalEntryMatch(name string, data []byte) []byte { + nflog("marshaling matcher %q", name) + + // We have to pad this struct size to a multiple of 8 bytes. + size := alignUp(linux.SizeOfXTEntryMatch+len(data), 8) + matcher := linux.KernelXTEntryMatch{ + XTEntryMatch: linux.XTEntryMatch{ + MatchSize: uint16(size), + }, + Data: data, + } + copy(matcher.Name[:], name) + + buf := make([]byte, 0, size) + buf = binary.Marshal(buf, usermem.ByteOrder, matcher) + return append(buf, make([]byte, size-len(buf))...) +} + +func unmarshalMatcher(match linux.XTEntryMatch, filter iptables.IPHeaderFilter, buf []byte) (iptables.Matcher, error) { + matchMaker, ok := matchMarshalers[match.Name.String()] + if !ok { + return nil, fmt.Errorf("unsupported matcher with name %q", match.Name.String()) + } + return matchMaker.unmarshal(buf, filter) +} + +// alignUp rounds a length up to an alignment. align must be a power of 2. +func alignUp(length int, align uint) int { + return (length + int(align) - 1) & ^(int(align) - 1) +} diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 3dda6c7a1..8f14643b0 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -17,6 +17,7 @@ package netfilter import ( + "errors" "fmt" "gvisor.dev/gvisor/pkg/abi/linux" @@ -34,10 +35,6 @@ import ( // shouldn't be reached - an error has occurred if we fall through to one. const errorTargetName = "ERROR" -const ( - matcherNameUDP = "udp" -) - // Metadata is used to verify that we are correctly serializing and // deserializing iptables into structs consumable by the iptables tool. We save // a metadata struct when the tables are written, and when they are read out we @@ -68,7 +65,8 @@ func GetInfo(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr) (linux.IPT // Find the appropriate table. table, err := findTable(stack, info.Name) if err != nil { - return linux.IPTGetinfo{}, err + nflog("%v", err) + return linux.IPTGetinfo{}, syserr.ErrInvalidArgument } // Get the hooks that apply to this table. @@ -95,39 +93,40 @@ func GetEntries(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr, outLen // Read in the struct and table name. var userEntries linux.IPTGetEntries if _, err := t.CopyIn(outPtr, &userEntries); err != nil { - log.Warningf("netfilter: couldn't copy in entries %q", userEntries.Name) + nflog("couldn't copy in entries %q", userEntries.Name) return linux.KernelIPTGetEntries{}, syserr.FromError(err) } // Find the appropriate table. table, err := findTable(stack, userEntries.Name) if err != nil { - log.Warningf("netfilter: couldn't find table %q", userEntries.Name) - return linux.KernelIPTGetEntries{}, err + nflog("%v", err) + return linux.KernelIPTGetEntries{}, syserr.ErrInvalidArgument } // Convert netstack's iptables rules to something that the iptables // tool can understand. entries, meta, err := convertNetstackToBinary(userEntries.Name.String(), table) if err != nil { - return linux.KernelIPTGetEntries{}, err + nflog("couldn't read entries: %v", err) + return linux.KernelIPTGetEntries{}, syserr.ErrInvalidArgument } if meta != table.Metadata().(metadata) { panic(fmt.Sprintf("Table %q metadata changed between writing and reading. Was saved as %+v, but is now %+v", userEntries.Name.String(), table.Metadata().(metadata), meta)) } if binary.Size(entries) > uintptr(outLen) { - log.Warningf("Insufficient GetEntries output size: %d", uintptr(outLen)) + nflog("insufficient GetEntries output size: %d", uintptr(outLen)) return linux.KernelIPTGetEntries{}, syserr.ErrInvalidArgument } return entries, nil } -func findTable(stack *stack.Stack, tablename linux.TableName) (iptables.Table, *syserr.Error) { +func findTable(stack *stack.Stack, tablename linux.TableName) (iptables.Table, error) { ipt := stack.IPTables() table, ok := ipt.Tables[tablename.String()] if !ok { - return iptables.Table{}, syserr.ErrInvalidArgument + return iptables.Table{}, fmt.Errorf("couldn't find table %q", tablename) } return table, nil } @@ -151,19 +150,19 @@ func FillDefaultIPTables(stack *stack.Stack) { stack.SetIPTables(ipt) } +// TODO: Return proto. // convertNetstackToBinary converts the iptables as stored in netstack to the // format expected by the iptables tool. Linux stores each table as a binary // blob that can only be traversed by parsing a bit, reading some offsets, // jumping to those offsets, parsing again, etc. -func convertNetstackToBinary(tablename string, table iptables.Table) (linux.KernelIPTGetEntries, metadata, *syserr.Error) { +func convertNetstackToBinary(tablename string, table iptables.Table) (linux.KernelIPTGetEntries, metadata, error) { // Return values. var entries linux.KernelIPTGetEntries var meta metadata // The table name has to fit in the struct. if linux.XT_TABLE_MAXNAMELEN < len(tablename) { - log.Warningf("Table name %q too long.", tablename) - return linux.KernelIPTGetEntries{}, metadata{}, syserr.ErrInvalidArgument + return linux.KernelIPTGetEntries{}, metadata{}, fmt.Errorf("Table name %q too long.", tablename) } copy(entries.Name[:], tablename) @@ -229,46 +228,6 @@ func convertNetstackToBinary(tablename string, table iptables.Table) (linux.Kern return entries, meta, nil } -func marshalMatcher(matcher iptables.Matcher) []byte { - switch m := matcher.(type) { - case *iptables.UDPMatcher: - return marshalUDPMatcher(m) - default: - // TODO(gvisor.dev/issue/170): Support other matchers. - panic(fmt.Errorf("unknown matcher of type %T", matcher)) - } -} - -func marshalUDPMatcher(matcher *iptables.UDPMatcher) []byte { - nflog("convert to binary: marshalling UDP matcher: %+v", matcher) - - // We have to pad this struct size to a multiple of 8 bytes. - size := alignUp(linux.SizeOfXTEntryMatch+linux.SizeOfXTUDP, 8) - - linuxMatcher := linux.KernelXTEntryMatch{ - XTEntryMatch: linux.XTEntryMatch{ - MatchSize: uint16(size), - }, - Data: make([]byte, 0, linux.SizeOfXTUDP), - } - copy(linuxMatcher.Name[:], matcherNameUDP) - - xtudp := linux.XTUDP{ - SourcePortStart: matcher.Data.SourcePortStart, - SourcePortEnd: matcher.Data.SourcePortEnd, - DestinationPortStart: matcher.Data.DestinationPortStart, - DestinationPortEnd: matcher.Data.DestinationPortEnd, - InverseFlags: matcher.Data.InverseFlags, - } - linuxMatcher.Data = binary.Marshal(linuxMatcher.Data, usermem.ByteOrder, xtudp) - - buf := make([]byte, 0, size) - buf = binary.Marshal(buf, usermem.ByteOrder, linuxMatcher) - buf = append(buf, make([]byte, size-len(buf))...) - nflog("convert to binary: marshalled UDP matcher into %v", buf) - return buf[:] -} - func marshalTarget(target iptables.Target) []byte { switch target.(type) { case iptables.UnconditionalAcceptTarget: @@ -332,7 +291,7 @@ func translateFromStandardVerdict(verdict iptables.Verdict) int32 { // translateToStandardVerdict translates from the value in a // linux.XTStandardTarget to an iptables.Verdict. -func translateToStandardVerdict(val int32) (iptables.Verdict, *syserr.Error) { +func translateToStandardVerdict(val int32) (iptables.Verdict, error) { // TODO(gvisor.dev/issue/170): Support other verdicts. switch val { case -linux.NF_ACCEPT - 1: @@ -340,13 +299,12 @@ func translateToStandardVerdict(val int32) (iptables.Verdict, *syserr.Error) { case -linux.NF_DROP - 1: return iptables.Drop, nil case -linux.NF_QUEUE - 1: - log.Warningf("Unsupported iptables verdict QUEUE.") + return iptables.Invalid, errors.New("unsupported iptables verdict QUEUE") case linux.NF_RETURN: - log.Warningf("Unsupported iptables verdict RETURN.") + return iptables.Invalid, errors.New("unsupported iptables verdict RETURN") default: - log.Warningf("Unknown iptables verdict %d.", val) + return iptables.Invalid, fmt.Errorf("unknown iptables verdict %d.", val) } - return iptables.Invalid, syserr.ErrInvalidArgument } // SetEntries sets iptables rules for a single table. See @@ -354,7 +312,7 @@ func translateToStandardVerdict(val int32) (iptables.Verdict, *syserr.Error) { func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { // Get the basic rules data (struct ipt_replace). if len(optVal) < linux.SizeOfIPTReplace { - log.Warningf("netfilter.SetEntries: optVal has insufficient size for replace %d", len(optVal)) + nflog("optVal has insufficient size for replace %d", len(optVal)) return syserr.ErrInvalidArgument } var replace linux.IPTReplace @@ -368,7 +326,7 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { case iptables.TablenameFilter: table = iptables.EmptyFilterTable() default: - log.Warningf("We don't yet support writing to the %q table (gvisor.dev/issue/170)", replace.Name.String()) + nflog("we don't yet support writing to the %q table (gvisor.dev/issue/170)", replace.Name.String()) return syserr.ErrInvalidArgument } @@ -382,7 +340,7 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { // Get the struct ipt_entry. if len(optVal) < linux.SizeOfIPTEntry { - log.Warningf("netfilter: optVal has insufficient size for entry %d", len(optVal)) + nflog("optVal has insufficient size for entry %d", len(optVal)) return syserr.ErrInvalidArgument } var entry linux.IPTEntry @@ -392,7 +350,7 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { optVal = optVal[linux.SizeOfIPTEntry:] if entry.TargetOffset < linux.SizeOfIPTEntry { - log.Warningf("netfilter: entry has too-small target offset %d", entry.TargetOffset) + nflog("entry has too-small target offset %d", entry.TargetOffset) return syserr.ErrInvalidArgument } @@ -400,7 +358,8 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { // filtering fields. filter, err := filterFromIPTIP(entry.IP) if err != nil { - return err + nflog("bad iptip: %v", err) + return syserr.ErrInvalidArgument } // TODO(gvisor.dev/issue/170): Matchers and targets can specify @@ -408,25 +367,26 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { // Get matchers. matchersSize := entry.TargetOffset - linux.SizeOfIPTEntry if len(optVal) < int(matchersSize) { - log.Warningf("netfilter: entry doesn't have enough room for its matchers (only %d bytes remain)", len(optVal)) + nflog("entry doesn't have enough room for its matchers (only %d bytes remain)", len(optVal)) return syserr.ErrInvalidArgument } matchers, err := parseMatchers(filter, optVal[:matchersSize]) if err != nil { - log.Warningf("netfilter: failed to parse matchers: %v", err) - return err + nflog("failed to parse matchers: %v", err) + return syserr.ErrInvalidArgument } optVal = optVal[matchersSize:] // Get the target of the rule. targetSize := entry.NextOffset - entry.TargetOffset if len(optVal) < int(targetSize) { - log.Warningf("netfilter: entry doesn't have enough room for its target (only %d bytes remain)", len(optVal)) + nflog("entry doesn't have enough room for its target (only %d bytes remain)", len(optVal)) return syserr.ErrInvalidArgument } target, err := parseTarget(optVal[:targetSize]) if err != nil { - return err + nflog("failed to parse target: %v", err) + return syserr.ErrInvalidArgument } optVal = optVal[targetSize:] @@ -439,7 +399,7 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { offset += uint32(entry.NextOffset) if initialOptValLen-len(optVal) != int(entry.NextOffset) { - log.Warningf("netfilter: entry NextOffset is %d, but entry took up %d bytes", entry.NextOffset, initialOptValLen-len(optVal)) + nflog("entry NextOffset is %d, but entry took up %d bytes", entry.NextOffset, initialOptValLen-len(optVal)) } } @@ -457,11 +417,11 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { } } if ruleIdx := table.BuiltinChains[hk]; ruleIdx == iptables.HookUnset { - log.Warningf("Hook %v is unset.", hk) + nflog("hook %v is unset.", hk) return syserr.ErrInvalidArgument } if ruleIdx := table.Underflows[hk]; ruleIdx == iptables.HookUnset { - log.Warningf("Underflow %v is unset.", hk) + nflog("underflow %v is unset.", hk) return syserr.ErrInvalidArgument } } @@ -473,7 +433,7 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { for hook, ruleIdx := range table.BuiltinChains { if hook != iptables.Input { if _, ok := table.Rules[ruleIdx].Target.(iptables.UnconditionalAcceptTarget); !ok { - log.Warningf("Hook %d is unsupported.", hook) + nflog("hook %d is unsupported.", hook) return syserr.ErrInvalidArgument } } @@ -499,7 +459,7 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { // parseMatchers parses 0 or more matchers from optVal. optVal should contain // only the matchers. -func parseMatchers(filter iptables.IPHeaderFilter, optVal []byte) ([]iptables.Matcher, *syserr.Error) { +func parseMatchers(filter iptables.IPHeaderFilter, optVal []byte) ([]iptables.Matcher, error) { nflog("set entries: parsing matchers of size %d", len(optVal)) var matchers []iptables.Matcher for len(optVal) > 0 { @@ -507,8 +467,7 @@ func parseMatchers(filter iptables.IPHeaderFilter, optVal []byte) ([]iptables.Ma // Get the XTEntryMatch. if len(optVal) < linux.SizeOfXTEntryMatch { - log.Warningf("netfilter: optVal has insufficient size for entry match: %d", len(optVal)) - return nil, syserr.ErrInvalidArgument + return nil, fmt.Errorf("optVal has insufficient size for entry match: %d", len(optVal)) } var match linux.XTEntryMatch buf := optVal[:linux.SizeOfXTEntryMatch] @@ -517,45 +476,18 @@ func parseMatchers(filter iptables.IPHeaderFilter, optVal []byte) ([]iptables.Ma // Check some invariants. if match.MatchSize < linux.SizeOfXTEntryMatch { - log.Warningf("netfilter: match size is too small, must be at least %d", linux.SizeOfXTEntryMatch) - return nil, syserr.ErrInvalidArgument + + return nil, fmt.Errorf("match size is too small, must be at least %d", linux.SizeOfXTEntryMatch) } if len(optVal) < int(match.MatchSize) { - log.Warningf("netfilter: optVal has insufficient size for match: %d", len(optVal)) - return nil, syserr.ErrInvalidArgument + return nil, fmt.Errorf("optVal has insufficient size for match: %d", len(optVal)) } - buf = optVal[linux.SizeOfXTEntryMatch:match.MatchSize] - var matcher iptables.Matcher - var err error - switch match.Name.String() { - case matcherNameUDP: - if len(buf) < linux.SizeOfXTUDP { - log.Warningf("netfilter: optVal has insufficient size for UDP match: %d", len(optVal)) - return nil, syserr.ErrInvalidArgument - } - // For alignment reasons, the match's total size may - // exceed what's strictly necessary to hold matchData. - var matchData linux.XTUDP - binary.Unmarshal(buf[:linux.SizeOfXTUDP], usermem.ByteOrder, &matchData) - log.Infof("parseMatchers: parsed XTUDP: %+v", matchData) - matcher, err = iptables.NewUDPMatcher(filter, iptables.UDPMatcherParams{ - SourcePortStart: matchData.SourcePortStart, - SourcePortEnd: matchData.SourcePortEnd, - DestinationPortStart: matchData.DestinationPortStart, - DestinationPortEnd: matchData.DestinationPortEnd, - InverseFlags: matchData.InverseFlags, - }) - if err != nil { - log.Warningf("netfilter: failed to create UDP matcher: %v", err) - return nil, syserr.ErrInvalidArgument - } - - default: - log.Warningf("netfilter: unsupported matcher with name %q", match.Name.String()) - return nil, syserr.ErrInvalidArgument + // Parse the specific matcher. + matcher, err := unmarshalMatcher(match, filter, optVal[linux.SizeOfXTEntryMatch:match.MatchSize]) + if err != nil { + return nil, fmt.Errorf("failed to create matcher: %v", err) } - matchers = append(matchers, matcher) // TODO(gvisor.dev/issue/170): Check the revision field. @@ -563,8 +495,7 @@ func parseMatchers(filter iptables.IPHeaderFilter, optVal []byte) ([]iptables.Ma } if len(optVal) != 0 { - log.Warningf("netfilter: optVal should be exhausted after parsing matchers") - return nil, syserr.ErrInvalidArgument + return nil, errors.New("optVal should be exhausted after parsing matchers") } return matchers, nil @@ -572,11 +503,10 @@ func parseMatchers(filter iptables.IPHeaderFilter, optVal []byte) ([]iptables.Ma // parseTarget parses a target from optVal. optVal should contain only the // target. -func parseTarget(optVal []byte) (iptables.Target, *syserr.Error) { +func parseTarget(optVal []byte) (iptables.Target, error) { nflog("set entries: parsing target of size %d", len(optVal)) if len(optVal) < linux.SizeOfXTEntryTarget { - log.Warningf("netfilter: optVal has insufficient size for entry target %d", len(optVal)) - return nil, syserr.ErrInvalidArgument + return nil, fmt.Errorf("optVal has insufficient size for entry target %d", len(optVal)) } var target linux.XTEntryTarget buf := optVal[:linux.SizeOfXTEntryTarget] @@ -585,8 +515,7 @@ func parseTarget(optVal []byte) (iptables.Target, *syserr.Error) { case "": // Standard target. if len(optVal) != linux.SizeOfXTStandardTarget { - log.Warningf("netfilter.SetEntries: optVal has wrong size for standard target %d", len(optVal)) - return nil, syserr.ErrInvalidArgument + return nil, fmt.Errorf("optVal has wrong size for standard target %d", len(optVal)) } var standardTarget linux.XTStandardTarget buf = optVal[:linux.SizeOfXTStandardTarget] @@ -602,15 +531,13 @@ func parseTarget(optVal []byte) (iptables.Target, *syserr.Error) { case iptables.Drop: return iptables.UnconditionalDropTarget{}, nil default: - log.Warningf("Unknown verdict: %v", verdict) - return nil, syserr.ErrInvalidArgument + return nil, fmt.Errorf("Unknown verdict: %v", verdict) } case errorTargetName: // Error target. if len(optVal) != linux.SizeOfXTErrorTarget { - log.Infof("netfilter.SetEntries: optVal has insufficient size for error target %d", len(optVal)) - return nil, syserr.ErrInvalidArgument + return nil, fmt.Errorf("optVal has insufficient size for error target %d", len(optVal)) } var errorTarget linux.XTErrorTarget buf = optVal[:linux.SizeOfXTErrorTarget] @@ -627,20 +554,17 @@ func parseTarget(optVal []byte) (iptables.Target, *syserr.Error) { case errorTargetName: return iptables.ErrorTarget{}, nil default: - log.Infof("Unknown error target %q doesn't exist or isn't supported yet.", errorTarget.Name.String()) - return nil, syserr.ErrInvalidArgument + return nil, fmt.Errorf("Unknown error target %q doesn't exist or isn't supported yet.", errorTarget.Name.String()) } } // Unknown target. - log.Infof("Unknown target %q doesn't exist or isn't supported yet.", target.Name.String()) - return nil, syserr.ErrInvalidArgument + return nil, fmt.Errorf("Unknown target %q doesn't exist or isn't supported yet.", target.Name.String()) } -func filterFromIPTIP(iptip linux.IPTIP) (iptables.IPHeaderFilter, *syserr.Error) { +func filterFromIPTIP(iptip linux.IPTIP) (iptables.IPHeaderFilter, error) { if containsUnsupportedFields(iptip) { - log.Warningf("netfilter: unsupported fields in struct iptip: %+v", iptip) - return iptables.IPHeaderFilter{}, syserr.ErrInvalidArgument + return iptables.IPHeaderFilter{}, fmt.Errorf("unsupported fields in struct iptip: %+v", iptip) } return iptables.IPHeaderFilter{ Protocol: tcpip.TransportProtocolNumber(iptip.Protocol), @@ -678,8 +602,3 @@ func hookFromLinux(hook int) iptables.Hook { } panic(fmt.Sprintf("Unknown hook %d does not correspond to a builtin chain", hook)) } - -// alignUp rounds a length up to an alignment. align must be a power of 2. -func alignUp(length int, align uint) int { - return (length + int(align) - 1) & ^(int(align) - 1) -} diff --git a/pkg/sentry/socket/netfilter/tcp_matcher.go b/pkg/sentry/socket/netfilter/tcp_matcher.go new file mode 100644 index 000000000..1646d22f7 --- /dev/null +++ b/pkg/sentry/socket/netfilter/tcp_matcher.go @@ -0,0 +1,143 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package netfilter + +import ( + "fmt" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/iptables" + "gvisor.dev/gvisor/pkg/usermem" +) + +const matcherNameTCP = "tcp" + +func init() { + registerMatchMarshaler(tcpMarshaler{}) +} + +// tcpMarshaler implements matchMarshaler for TCP matching. +type tcpMarshaler struct{} + +// name implements matchMarshaler.name. +func (tcpMarshaler) name() string { + return matcherNameTCP +} + +// marshal implements matchMarshaler.marshal. +func (tcpMarshaler) marshal(mr iptables.Matcher) []byte { + matcher := mr.(*TCPMatcher) + xttcp := linux.XTTCP{ + SourcePortStart: matcher.sourcePortStart, + SourcePortEnd: matcher.sourcePortEnd, + DestinationPortStart: matcher.destinationPortStart, + DestinationPortEnd: matcher.destinationPortEnd, + } + buf := make([]byte, 0, linux.SizeOfXTUDP) + return marshalEntryMatch(matcherNameTCP, binary.Marshal(buf, usermem.ByteOrder, xttcp)) +} + +// unmarshal implements matchMarshaler.unmarshal. +func (tcpMarshaler) unmarshal(buf []byte, filter iptables.IPHeaderFilter) (iptables.Matcher, error) { + if len(buf) < linux.SizeOfXTTCP { + return nil, fmt.Errorf("buf has insufficient size for TCP match: %d", len(buf)) + } + + // For alignment reasons, the match's total size may + // exceed what's strictly necessary to hold matchData. + var matchData linux.XTTCP + binary.Unmarshal(buf[:linux.SizeOfXTTCP], usermem.ByteOrder, &matchData) + nflog("parseMatchers: parsed XTTCP: %+v", matchData) + + if matchData.Option != 0 || + matchData.FlagMask != 0 || + matchData.FlagCompare != 0 || + matchData.InverseFlags != 0 { + return nil, fmt.Errorf("unsupported TCP matcher flags set") + } + + if filter.Protocol != header.TCPProtocolNumber { + return nil, fmt.Errorf("TCP matching is only valid for protocol %d.", header.TCPProtocolNumber) + } + + return &TCPMatcher{ + sourcePortStart: matchData.SourcePortStart, + sourcePortEnd: matchData.SourcePortEnd, + destinationPortStart: matchData.DestinationPortStart, + destinationPortEnd: matchData.DestinationPortEnd, + }, nil +} + +// TCPMatcher matches TCP packets and their headers. It implements Matcher. +type TCPMatcher struct { + sourcePortStart uint16 + sourcePortEnd uint16 + destinationPortStart uint16 + destinationPortEnd uint16 +} + +// Name implements Matcher.Name. +func (*TCPMatcher) Name() string { + return matcherNameTCP +} + +// Match implements Matcher.Match. +func (tm *TCPMatcher) Match(hook iptables.Hook, pkt tcpip.PacketBuffer, interfaceName string) (bool, bool) { + netHeader := header.IPv4(pkt.NetworkHeader) + + if netHeader.TransportProtocol() != header.TCPProtocolNumber { + return false, false + } + + // We dont't match fragments. + if frag := netHeader.FragmentOffset(); frag != 0 { + if frag == 1 { + return false, true + } + return false, false + } + + // Now we need the transport header. However, this may not have been set + // yet. + // TODO(gvisor.dev/issue/170): Parsing the transport header should + // ultimately be moved into the iptables.Check codepath as matchers are + // added. + var tcpHeader header.TCP + if pkt.TransportHeader != nil { + tcpHeader = header.TCP(pkt.TransportHeader) + } else { + // The TCP header hasn't been parsed yet. We have to do it here. + if len(pkt.Data.First()) < header.TCPMinimumSize { + // There's no valid TCP header here, so we hotdrop the + // packet. + return false, true + } + tcpHeader = header.TCP(pkt.Data.First()) + } + + // Check whether the source and destination ports are within the + // matching range. + if sourcePort := tcpHeader.SourcePort(); sourcePort < tm.sourcePortStart || tm.sourcePortEnd < sourcePort { + return false, false + } + if destinationPort := tcpHeader.DestinationPort(); destinationPort < tm.destinationPortStart || tm.destinationPortEnd < destinationPort { + return false, false + } + + return true, false +} diff --git a/pkg/sentry/socket/netfilter/udp_matcher.go b/pkg/sentry/socket/netfilter/udp_matcher.go new file mode 100644 index 000000000..b6e95bbc5 --- /dev/null +++ b/pkg/sentry/socket/netfilter/udp_matcher.go @@ -0,0 +1,142 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package netfilter + +import ( + "fmt" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/iptables" + "gvisor.dev/gvisor/pkg/usermem" +) + +const matcherNameUDP = "udp" + +func init() { + registerMatchMarshaler(udpMarshaler{}) +} + +// udpMarshaler implements matchMarshaler for UDP matching. +type udpMarshaler struct{} + +// name implements matchMarshaler.name. +func (udpMarshaler) name() string { + return matcherNameUDP +} + +// marshal implements matchMarshaler.marshal. +func (udpMarshaler) marshal(mr iptables.Matcher) []byte { + matcher := mr.(*UDPMatcher) + xtudp := linux.XTUDP{ + SourcePortStart: matcher.sourcePortStart, + SourcePortEnd: matcher.sourcePortEnd, + DestinationPortStart: matcher.destinationPortStart, + DestinationPortEnd: matcher.destinationPortEnd, + } + buf := make([]byte, 0, linux.SizeOfXTUDP) + return marshalEntryMatch(matcherNameUDP, binary.Marshal(buf, usermem.ByteOrder, xtudp)) +} + +// unmarshal implements matchMarshaler.unmarshal. +func (udpMarshaler) unmarshal(buf []byte, filter iptables.IPHeaderFilter) (iptables.Matcher, error) { + if len(buf) < linux.SizeOfXTUDP { + return nil, fmt.Errorf("buf has insufficient size for UDP match: %d", len(buf)) + } + + // For alignment reasons, the match's total size may exceed what's + // strictly necessary to hold matchData. + var matchData linux.XTUDP + binary.Unmarshal(buf[:linux.SizeOfXTUDP], usermem.ByteOrder, &matchData) + nflog("parseMatchers: parsed XTUDP: %+v", matchData) + + if matchData.InverseFlags != 0 { + return nil, fmt.Errorf("unsupported UDP matcher inverse flags set") + } + + if filter.Protocol != header.UDPProtocolNumber { + return nil, fmt.Errorf("UDP matching is only valid for protocol %d.", header.UDPProtocolNumber) + } + + return &UDPMatcher{ + sourcePortStart: matchData.SourcePortStart, + sourcePortEnd: matchData.SourcePortEnd, + destinationPortStart: matchData.DestinationPortStart, + destinationPortEnd: matchData.DestinationPortEnd, + }, nil +} + +// UDPMatcher matches UDP packets and their headers. It implements Matcher. +type UDPMatcher struct { + sourcePortStart uint16 + sourcePortEnd uint16 + destinationPortStart uint16 + destinationPortEnd uint16 +} + +// Name implements Matcher.Name. +func (*UDPMatcher) Name() string { + return matcherNameUDP +} + +// Match implements Matcher.Match. +func (um *UDPMatcher) Match(hook iptables.Hook, pkt tcpip.PacketBuffer, interfaceName string) (bool, bool) { + netHeader := header.IPv4(pkt.NetworkHeader) + + // TODO(gvisor.dev/issue/170): Proto checks should ultimately be moved + // into the iptables.Check codepath as matchers are added. + if netHeader.TransportProtocol() != header.UDPProtocolNumber { + return false, false + } + + // We dont't match fragments. + if frag := netHeader.FragmentOffset(); frag != 0 { + if frag == 1 { + return false, true + } + return false, false + } + + // Now we need the transport header. However, this may not have been set + // yet. + // TODO(gvisor.dev/issue/170): Parsing the transport header should + // ultimately be moved into the iptables.Check codepath as matchers are + // added. + var udpHeader header.UDP + if pkt.TransportHeader != nil { + udpHeader = header.UDP(pkt.TransportHeader) + } else { + // The UDP header hasn't been parsed yet. We have to do it here. + if len(pkt.Data.First()) < header.UDPMinimumSize { + // There's no valid UDP header here, so we hotdrop the + // packet. + return false, true + } + udpHeader = header.UDP(pkt.Data.First()) + } + + // Check whether the source and destination ports are within the + // matching range. + if sourcePort := udpHeader.SourcePort(); sourcePort < um.sourcePortStart || um.sourcePortEnd < sourcePort { + return false, false + } + if destinationPort := udpHeader.DestinationPort(); destinationPort < um.destinationPortStart || um.destinationPortEnd < destinationPort { + return false, false + } + + return true, false +} diff --git a/pkg/tcpip/iptables/BUILD b/pkg/tcpip/iptables/BUILD index bab26580b..d1b73cfdf 100644 --- a/pkg/tcpip/iptables/BUILD +++ b/pkg/tcpip/iptables/BUILD @@ -8,7 +8,6 @@ go_library( "iptables.go", "targets.go", "types.go", - "udp_matcher.go", ], visibility = ["//visibility:public"], deps = [ diff --git a/pkg/tcpip/iptables/types.go b/pkg/tcpip/iptables/types.go index 7f77802a0..d660aab04 100644 --- a/pkg/tcpip/iptables/types.go +++ b/pkg/tcpip/iptables/types.go @@ -171,6 +171,9 @@ type IPHeaderFilter struct { // A Matcher is the interface for matching packets. type Matcher interface { + // Name returns the name of the Matcher. + Name() string + // Match returns whether the packet matches and whether the packet // should be "hotdropped", i.e. dropped immediately. This is usually // used for suspicious packets. diff --git a/pkg/tcpip/iptables/udp_matcher.go b/pkg/tcpip/iptables/udp_matcher.go deleted file mode 100644 index 3bb076f9c..000000000 --- a/pkg/tcpip/iptables/udp_matcher.go +++ /dev/null @@ -1,113 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package iptables - -import ( - "fmt" - - "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/header" -) - -// TODO(gvisor.dev/issue/170): The following per-matcher params should be -// supported: -// - Table name -// - Match size -// - User size -// - Hooks -// - Proto -// - Family - -// UDPMatcher matches UDP packets and their headers. It implements Matcher. -type UDPMatcher struct { - Data UDPMatcherParams -} - -// UDPMatcherParams are the parameters used to create a UDPMatcher. -type UDPMatcherParams struct { - SourcePortStart uint16 - SourcePortEnd uint16 - DestinationPortStart uint16 - DestinationPortEnd uint16 - InverseFlags uint8 -} - -// NewUDPMatcher returns a new instance of UDPMatcher. -func NewUDPMatcher(filter IPHeaderFilter, data UDPMatcherParams) (Matcher, error) { - log.Infof("Adding rule with UDPMatcherParams: %+v", data) - - if data.InverseFlags != 0 { - return nil, fmt.Errorf("unsupported UDP matcher inverse flags set") - } - - if filter.Protocol != header.UDPProtocolNumber { - return nil, fmt.Errorf("UDP matching is only valid for protocol %d.", header.UDPProtocolNumber) - } - - return &UDPMatcher{Data: data}, nil -} - -// Match implements Matcher.Match. -func (um *UDPMatcher) Match(hook Hook, pkt tcpip.PacketBuffer, interfaceName string) (bool, bool) { - netHeader := header.IPv4(pkt.NetworkHeader) - - // TODO(gvisor.dev/issue/170): Proto checks should ultimately be moved - // into the iptables.Check codepath as matchers are added. - if netHeader.TransportProtocol() != header.UDPProtocolNumber { - return false, false - } - - // We dont't match fragments. - if frag := netHeader.FragmentOffset(); frag != 0 { - if frag == 1 { - log.Warningf("Dropping UDP packet: malicious fragmented packet.") - return false, true - } - return false, false - } - - // Now we need the transport header. However, this may not have been set - // yet. - // TODO(gvisor.dev/issue/170): Parsing the transport header should - // ultimately be moved into the iptables.Check codepath as matchers are - // added. - var udpHeader header.UDP - if pkt.TransportHeader != nil { - udpHeader = header.UDP(pkt.TransportHeader) - } else { - // The UDP header hasn't been parsed yet. We have to do it here. - if len(pkt.Data.First()) < header.UDPMinimumSize { - // There's no valid UDP header here, so we hotdrop the - // packet. - log.Warningf("Dropping UDP packet: size too small.") - return false, true - } - udpHeader = header.UDP(pkt.Data.First()) - } - - // Check whether the source and destination ports are within the - // matching range. - sourcePort := udpHeader.SourcePort() - destinationPort := udpHeader.DestinationPort() - if sourcePort < um.Data.SourcePortStart || um.Data.SourcePortEnd < sourcePort { - return false, false - } - if destinationPort < um.Data.DestinationPortStart || um.Data.DestinationPortEnd < destinationPort { - return false, false - } - - return true, false -} -- cgit v1.2.3 From 29ad5762e4549d961f48c65292cfdeb7256524f6 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Fri, 31 Jan 2020 13:53:58 -0800 Subject: Spelling --- pkg/sentry/socket/netfilter/extensions.go | 18 +++++++++--------- pkg/sentry/socket/netfilter/tcp_matcher.go | 10 +++++----- pkg/sentry/socket/netfilter/udp_matcher.go | 10 +++++----- 3 files changed, 19 insertions(+), 19 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/extensions.go b/pkg/sentry/socket/netfilter/extensions.go index 5a4cac84c..b5fbb52e4 100644 --- a/pkg/sentry/socket/netfilter/extensions.go +++ b/pkg/sentry/socket/netfilter/extensions.go @@ -32,8 +32,8 @@ import ( // - Proto // - Family -// matchMarshaler knows how to (un)marshal the matcher named name(). -type matchMarshaler interface { +// matchMaker knows how to (un)marshal the matcher named name(). +type matchMaker interface { // name is the matcher name as stored in the xt_entry_match struct. name() string @@ -45,19 +45,19 @@ type matchMarshaler interface { unmarshal(buf []byte, filter iptables.IPHeaderFilter) (iptables.Matcher, error) } -var matchMarshalers = map[string]matchMarshaler{} +var matchMakers = map[string]matchMaker{} -// registerMatchMarshaler should be called by match extensions to register them +// registermatchMaker should be called by match extensions to register them // with the netfilter package. -func registerMatchMarshaler(mm matchMarshaler) { - if _, ok := matchMarshalers[mm.name()]; ok { +func registerMatchMaker(mm matchMaker) { + if _, ok := matchMakers[mm.name()]; ok { panic(fmt.Sprintf("Multiple matches registered with name %q.", mm.name())) } - matchMarshalers[mm.name()] = mm + matchMakers[mm.name()] = mm } func marshalMatcher(matcher iptables.Matcher) []byte { - matchMaker, ok := matchMarshalers[matcher.Name()] + matchMaker, ok := matchMakers[matcher.Name()] if !ok { panic(fmt.Errorf("Unknown matcher of type %T.", matcher)) } @@ -85,7 +85,7 @@ func marshalEntryMatch(name string, data []byte) []byte { } func unmarshalMatcher(match linux.XTEntryMatch, filter iptables.IPHeaderFilter, buf []byte) (iptables.Matcher, error) { - matchMaker, ok := matchMarshalers[match.Name.String()] + matchMaker, ok := matchMakers[match.Name.String()] if !ok { return nil, fmt.Errorf("unsupported matcher with name %q", match.Name.String()) } diff --git a/pkg/sentry/socket/netfilter/tcp_matcher.go b/pkg/sentry/socket/netfilter/tcp_matcher.go index 1646d22f7..6b2f4c31a 100644 --- a/pkg/sentry/socket/netfilter/tcp_matcher.go +++ b/pkg/sentry/socket/netfilter/tcp_matcher.go @@ -28,18 +28,18 @@ import ( const matcherNameTCP = "tcp" func init() { - registerMatchMarshaler(tcpMarshaler{}) + registerMatchMaker(tcpMarshaler{}) } -// tcpMarshaler implements matchMarshaler for TCP matching. +// tcpMarshaler implements matchMaker for TCP matching. type tcpMarshaler struct{} -// name implements matchMarshaler.name. +// name implements matchMaker.name. func (tcpMarshaler) name() string { return matcherNameTCP } -// marshal implements matchMarshaler.marshal. +// marshal implements matchMaker.marshal. func (tcpMarshaler) marshal(mr iptables.Matcher) []byte { matcher := mr.(*TCPMatcher) xttcp := linux.XTTCP{ @@ -52,7 +52,7 @@ func (tcpMarshaler) marshal(mr iptables.Matcher) []byte { return marshalEntryMatch(matcherNameTCP, binary.Marshal(buf, usermem.ByteOrder, xttcp)) } -// unmarshal implements matchMarshaler.unmarshal. +// unmarshal implements matchMaker.unmarshal. func (tcpMarshaler) unmarshal(buf []byte, filter iptables.IPHeaderFilter) (iptables.Matcher, error) { if len(buf) < linux.SizeOfXTTCP { return nil, fmt.Errorf("buf has insufficient size for TCP match: %d", len(buf)) diff --git a/pkg/sentry/socket/netfilter/udp_matcher.go b/pkg/sentry/socket/netfilter/udp_matcher.go index b6e95bbc5..86aa11696 100644 --- a/pkg/sentry/socket/netfilter/udp_matcher.go +++ b/pkg/sentry/socket/netfilter/udp_matcher.go @@ -28,18 +28,18 @@ import ( const matcherNameUDP = "udp" func init() { - registerMatchMarshaler(udpMarshaler{}) + registerMatchMaker(udpMarshaler{}) } -// udpMarshaler implements matchMarshaler for UDP matching. +// udpMarshaler implements matchMaker for UDP matching. type udpMarshaler struct{} -// name implements matchMarshaler.name. +// name implements matchMaker.name. func (udpMarshaler) name() string { return matcherNameUDP } -// marshal implements matchMarshaler.marshal. +// marshal implements matchMaker.marshal. func (udpMarshaler) marshal(mr iptables.Matcher) []byte { matcher := mr.(*UDPMatcher) xtudp := linux.XTUDP{ @@ -52,7 +52,7 @@ func (udpMarshaler) marshal(mr iptables.Matcher) []byte { return marshalEntryMatch(matcherNameUDP, binary.Marshal(buf, usermem.ByteOrder, xtudp)) } -// unmarshal implements matchMarshaler.unmarshal. +// unmarshal implements matchMaker.unmarshal. func (udpMarshaler) unmarshal(buf []byte, filter iptables.IPHeaderFilter) (iptables.Matcher, error) { if len(buf) < linux.SizeOfXTUDP { return nil, fmt.Errorf("buf has insufficient size for UDP match: %d", len(buf)) -- cgit v1.2.3 From 6c3072243dfbf70062de5f610e14fd6ed2ce5f32 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 31 Jan 2020 14:14:52 -0800 Subject: Implement file locks for regular tmpfs files in VFSv2. Add a file lock implementation that can be embedded into various filesystem implementations. Updates #1480 PiperOrigin-RevId: 292614758 --- pkg/sentry/fsimpl/tmpfs/BUILD | 3 ++ pkg/sentry/fsimpl/tmpfs/regular_file.go | 23 +++++++++ pkg/sentry/fsimpl/tmpfs/regular_file_test.go | 56 ++++++++++++++++++++++ pkg/sentry/fsimpl/tmpfs/tmpfs.go | 43 +++++++++++++++++ pkg/sentry/vfs/lock/BUILD | 13 +++++ pkg/sentry/vfs/lock/lock.go | 72 ++++++++++++++++++++++++++++ 6 files changed, 210 insertions(+) create mode 100644 pkg/sentry/vfs/lock/BUILD create mode 100644 pkg/sentry/vfs/lock/lock.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD index c61366224..57abd5583 100644 --- a/pkg/sentry/fsimpl/tmpfs/BUILD +++ b/pkg/sentry/fsimpl/tmpfs/BUILD @@ -38,6 +38,7 @@ go_library( "//pkg/sentry/arch", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", + "//pkg/sentry/fs/lock", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/pipe", @@ -47,6 +48,7 @@ go_library( "//pkg/sentry/platform", "//pkg/sentry/usage", "//pkg/sentry/vfs", + "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserror", "//pkg/usermem", @@ -86,6 +88,7 @@ go_test( "//pkg/context", "//pkg/fspath", "//pkg/sentry/contexttest", + "//pkg/sentry/fs/lock", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/contexttest", "//pkg/sentry/vfs", diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index e9e6faf67..dab346a41 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -23,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" + "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" @@ -192,6 +193,28 @@ func (fd *regularFileFD) Sync(ctx context.Context) error { return nil } +// LockBSD implements vfs.FileDescriptionImpl.LockBSD. +func (fd *regularFileFD) LockBSD(ctx context.Context, uid lock.UniqueID, t lock.LockType, block lock.Blocker) error { + return fd.inode().lockBSD(uid, t, block) +} + +// UnlockBSD implements vfs.FileDescriptionImpl.UnlockBSD. +func (fd *regularFileFD) UnlockBSD(ctx context.Context, uid lock.UniqueID) error { + fd.inode().unlockBSD(uid) + return nil +} + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (fd *regularFileFD) LockPOSIX(ctx context.Context, uid lock.UniqueID, t lock.LockType, rng lock.LockRange, block lock.Blocker) error { + return fd.inode().lockPOSIX(uid, t, rng, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (fd *regularFileFD) UnlockPOSIX(ctx context.Context, uid lock.UniqueID, rng lock.LockRange) error { + fd.inode().unlockPOSIX(uid, rng) + return nil +} + // regularFileReadWriter implements safemem.Reader and Safemem.Writer. type regularFileReadWriter struct { file *regularFile diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go index 32552e261..2b52992ea 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go @@ -24,9 +24,11 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" ) @@ -260,6 +262,60 @@ func TestPWrite(t *testing.T) { } } +func TestLocks(t *testing.T) { + ctx := contexttest.Context(t) + fd, cleanup, err := newFileFD(ctx, 0644) + if err != nil { + t.Fatal(err) + } + defer cleanup() + + var ( + uid1 lock.UniqueID + uid2 lock.UniqueID + // Non-blocking. + block lock.Blocker + ) + + uid1 = 123 + uid2 = 456 + + if err := fd.Impl().LockBSD(ctx, uid1, lock.ReadLock, block); err != nil { + t.Fatalf("fd.Impl().LockBSD failed: err = %v", err) + } + if err := fd.Impl().LockBSD(ctx, uid2, lock.ReadLock, block); err != nil { + t.Fatalf("fd.Impl().LockBSD failed: err = %v", err) + } + if got, want := fd.Impl().LockBSD(ctx, uid2, lock.WriteLock, block), syserror.ErrWouldBlock; got != want { + t.Fatalf("fd.Impl().LockBSD failed: got = %v, want = %v", got, want) + } + if err := fd.Impl().UnlockBSD(ctx, uid1); err != nil { + t.Fatalf("fd.Impl().UnlockBSD failed: err = %v", err) + } + if err := fd.Impl().LockBSD(ctx, uid2, lock.WriteLock, block); err != nil { + t.Fatalf("fd.Impl().LockBSD failed: err = %v", err) + } + + rng1 := lock.LockRange{0, 1} + rng2 := lock.LockRange{1, 2} + + if err := fd.Impl().LockPOSIX(ctx, uid1, lock.ReadLock, rng1, block); err != nil { + t.Fatalf("fd.Impl().LockPOSIX failed: err = %v", err) + } + if err := fd.Impl().LockPOSIX(ctx, uid2, lock.ReadLock, rng2, block); err != nil { + t.Fatalf("fd.Impl().LockPOSIX failed: err = %v", err) + } + if err := fd.Impl().LockPOSIX(ctx, uid1, lock.WriteLock, rng1, block); err != nil { + t.Fatalf("fd.Impl().LockPOSIX failed: err = %v", err) + } + if got, want := fd.Impl().LockPOSIX(ctx, uid2, lock.ReadLock, rng1, block), syserror.ErrWouldBlock; got != want { + t.Fatalf("fd.Impl().LockPOSIX failed: got = %v, want = %v", got, want) + } + if err := fd.Impl().UnlockPOSIX(ctx, uid1, rng1); err != nil { + t.Fatalf("fd.Impl().UnlockPOSIX failed: err = %v", err) + } +} + func TestPRead(t *testing.T) { ctx := contexttest.Context(t) fd, cleanup, err := newFileFD(ctx, 0644) diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 88dbd6e35..2108d0f4d 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -30,10 +30,12 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) @@ -153,6 +155,9 @@ type inode struct { rdevMajor uint32 rdevMinor uint32 + // Advisory file locks, which lock at the inode level. + locks lock.FileLocks + impl interface{} // immutable } @@ -352,6 +357,44 @@ func (i *inode) setStat(stat linux.Statx) error { return nil } +// TODO(gvisor.dev/issue/1480): support file locking for file types other than regular. +func (i *inode) lockBSD(uid fslock.UniqueID, t fslock.LockType, block fslock.Blocker) error { + switch i.impl.(type) { + case *regularFile: + return i.locks.LockBSD(uid, t, block) + } + return syserror.EBADF +} + +// TODO(gvisor.dev/issue/1480): support file locking for file types other than regular. +func (i *inode) unlockBSD(uid fslock.UniqueID) error { + switch i.impl.(type) { + case *regularFile: + i.locks.UnlockBSD(uid) + return nil + } + return syserror.EBADF +} + +// TODO(gvisor.dev/issue/1480): support file locking for file types other than regular. +func (i *inode) lockPOSIX(uid fslock.UniqueID, t fslock.LockType, rng fslock.LockRange, block fslock.Blocker) error { + switch i.impl.(type) { + case *regularFile: + return i.locks.LockPOSIX(uid, t, rng, block) + } + return syserror.EBADF +} + +// TODO(gvisor.dev/issue/1480): support file locking for file types other than regular. +func (i *inode) unlockPOSIX(uid fslock.UniqueID, rng fslock.LockRange) error { + switch i.impl.(type) { + case *regularFile: + i.locks.UnlockPOSIX(uid, rng) + return nil + } + return syserror.EBADF +} + // allocatedBlocksForSize returns the number of 512B blocks needed to // accommodate the given size in bytes, as appropriate for struct // stat::st_blocks and struct statx::stx_blocks. (Note that this 512B block diff --git a/pkg/sentry/vfs/lock/BUILD b/pkg/sentry/vfs/lock/BUILD new file mode 100644 index 000000000..d9ab063b7 --- /dev/null +++ b/pkg/sentry/vfs/lock/BUILD @@ -0,0 +1,13 @@ +load("//tools:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_library( + name = "lock", + srcs = ["lock.go"], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/sentry/fs/lock", + "//pkg/syserror", + ], +) diff --git a/pkg/sentry/vfs/lock/lock.go b/pkg/sentry/vfs/lock/lock.go new file mode 100644 index 000000000..724dfe743 --- /dev/null +++ b/pkg/sentry/vfs/lock/lock.go @@ -0,0 +1,72 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package lock provides POSIX and BSD style file locking for VFS2 file +// implementations. +// +// The actual implementations can be found in the lock package under +// sentry/fs/lock. +package lock + +import ( + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" + "gvisor.dev/gvisor/pkg/syserror" +) + +// FileLocks supports POSIX and BSD style locks, which correspond to fcntl(2) +// and flock(2) respectively in Linux. It can be embedded into various file +// implementations for VFS2 that support locking. +// +// Note that in Linux these two types of locks are _not_ cooperative, because +// race and deadlock conditions make merging them prohibitive. We do the same +// and keep them oblivious to each other. +type FileLocks struct { + // bsd is a set of BSD-style advisory file wide locks, see flock(2). + bsd fslock.Locks + + // posix is a set of POSIX-style regional advisory locks, see fcntl(2). + posix fslock.Locks +} + +// LockBSD tries to acquire a BSD-style lock on the entire file. +func (fl *FileLocks) LockBSD(uid fslock.UniqueID, t fslock.LockType, block fslock.Blocker) error { + if fl.bsd.LockRegion(uid, t, fslock.LockRange{0, fslock.LockEOF}, block) { + return nil + } + return syserror.ErrWouldBlock +} + +// UnlockBSD releases a BSD-style lock on the entire file. +// +// This operation is always successful, even if there did not exist a lock on +// the requested region held by uid in the first place. +func (fl *FileLocks) UnlockBSD(uid fslock.UniqueID) { + fl.bsd.UnlockRegion(uid, fslock.LockRange{0, fslock.LockEOF}) +} + +// LockPOSIX tries to acquire a POSIX-style lock on a file region. +func (fl *FileLocks) LockPOSIX(uid fslock.UniqueID, t fslock.LockType, rng fslock.LockRange, block fslock.Blocker) error { + if fl.posix.LockRegion(uid, t, rng, block) { + return nil + } + return syserror.ErrWouldBlock +} + +// UnlockPOSIX releases a POSIX-style lock on a file region. +// +// This operation is always successful, even if there did not exist a lock on +// the requested region held by uid in the first place. +func (fl *FileLocks) UnlockPOSIX(uid fslock.UniqueID, rng fslock.LockRange) { + fl.posix.UnlockRegion(uid, rng) +} -- cgit v1.2.3 From d7cd484091543827678f1548b8e5668a7a86e13f Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Tue, 4 Feb 2020 08:20:10 -0800 Subject: Add support for sentry internal pipe for gofer mounts Internal pipes are supported similarly to how internal UDS is done. It is also controlled by the same flag. Fixes #1102 PiperOrigin-RevId: 293150045 --- pkg/sentry/fs/gofer/BUILD | 2 + pkg/sentry/fs/gofer/cache_policy.go | 3 + pkg/sentry/fs/gofer/fifo.go | 40 ++++++++ pkg/sentry/fs/gofer/gofer_test.go | 2 +- pkg/sentry/fs/gofer/path.go | 183 +++++++++++++++++++++++------------ pkg/sentry/fs/gofer/session.go | 183 ++++++++++++++++++++++------------- pkg/sentry/fs/gofer/session_state.go | 12 +-- pkg/sentry/fs/gofer/socket.go | 8 +- test/syscalls/BUILD | 1 - 9 files changed, 293 insertions(+), 141 deletions(-) create mode 100644 pkg/sentry/fs/gofer/fifo.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/gofer/BUILD b/pkg/sentry/fs/gofer/BUILD index 971d3718e..fea135eea 100644 --- a/pkg/sentry/fs/gofer/BUILD +++ b/pkg/sentry/fs/gofer/BUILD @@ -9,6 +9,7 @@ go_library( "cache_policy.go", "context_file.go", "device.go", + "fifo.go", "file.go", "file_state.go", "fs.go", @@ -38,6 +39,7 @@ go_library( "//pkg/sentry/fs/fsutil", "//pkg/sentry/fs/host", "//pkg/sentry/kernel/auth", + "//pkg/sentry/kernel/pipe", "//pkg/sentry/kernel/time", "//pkg/sentry/memmap", "//pkg/sentry/socket/unix/transport", diff --git a/pkg/sentry/fs/gofer/cache_policy.go b/pkg/sentry/fs/gofer/cache_policy.go index ebea03c42..07a564e92 100644 --- a/pkg/sentry/fs/gofer/cache_policy.go +++ b/pkg/sentry/fs/gofer/cache_policy.go @@ -127,6 +127,9 @@ func (cp cachePolicy) revalidate(ctx context.Context, name string, parent, child childIops, ok := child.InodeOperations.(*inodeOperations) if !ok { + if _, ok := child.InodeOperations.(*fifo); ok { + return false + } panic(fmt.Sprintf("revalidating inode operations of unknown type %T", child.InodeOperations)) } parentIops, ok := parent.InodeOperations.(*inodeOperations) diff --git a/pkg/sentry/fs/gofer/fifo.go b/pkg/sentry/fs/gofer/fifo.go new file mode 100644 index 000000000..456557058 --- /dev/null +++ b/pkg/sentry/fs/gofer/fifo.go @@ -0,0 +1,40 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package gofer + +import ( + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/fs" +) + +// +stateify savable +type fifo struct { + fs.InodeOperations + fileIops *inodeOperations +} + +var _ fs.InodeOperations = (*fifo)(nil) + +// Rename implements fs.InodeOperations. It forwards the call to the underlying +// file inode to handle the file rename. Note that file key remains the same +// after the rename to keep the endpoint mapping. +func (i *fifo) Rename(ctx context.Context, inode *fs.Inode, oldParent *fs.Inode, oldName string, newParent *fs.Inode, newName string, replacement bool) error { + return i.fileIops.Rename(ctx, inode, oldParent, oldName, newParent, newName, replacement) +} + +// StatFS implements fs.InodeOperations. +func (i *fifo) StatFS(ctx context.Context) (fs.Info, error) { + return i.fileIops.StatFS(ctx) +} diff --git a/pkg/sentry/fs/gofer/gofer_test.go b/pkg/sentry/fs/gofer/gofer_test.go index 0c2f89ae8..2df2fe889 100644 --- a/pkg/sentry/fs/gofer/gofer_test.go +++ b/pkg/sentry/fs/gofer/gofer_test.go @@ -61,7 +61,7 @@ func rootTest(t *testing.T, name string, cp cachePolicy, fn func(context.Context ctx := contexttest.Context(t) sattr, rootInodeOperations := newInodeOperations(ctx, s, contextFile{ file: rootFile, - }, root.QID, p9.AttrMaskAll(), root.Attr, false /* socket */) + }, root.QID, p9.AttrMaskAll(), root.Attr) m := fs.NewMountSource(ctx, s, &filesystem{}, fs.MountSourceFlags{}) rootInode := fs.NewInode(ctx, rootInodeOperations, m, sattr) diff --git a/pkg/sentry/fs/gofer/path.go b/pkg/sentry/fs/gofer/path.go index 0c1be05ef..a35c3a23d 100644 --- a/pkg/sentry/fs/gofer/path.go +++ b/pkg/sentry/fs/gofer/path.go @@ -23,14 +23,24 @@ import ( "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/sentry/device" "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // maxFilenameLen is the maximum length of a filename. This is dictated by 9P's // encoding of strings, which uses 2 bytes for the length prefix. const maxFilenameLen = (1 << 16) - 1 +func changeType(mode p9.FileMode, newType p9.FileMode) p9.FileMode { + if newType&^p9.FileModeMask != 0 { + panic(fmt.Sprintf("newType contained more bits than just file mode: %x", newType)) + } + clear := mode &^ p9.FileModeMask + return clear | newType +} + // Lookup loads an Inode at name into a Dirent based on the session's cache // policy. func (i *inodeOperations) Lookup(ctx context.Context, dir *fs.Inode, name string) (*fs.Dirent, error) { @@ -69,8 +79,25 @@ func (i *inodeOperations) Lookup(ctx context.Context, dir *fs.Inode, name string return nil, err } + if i.session().overrides != nil { + // Check if file belongs to a internal named pipe. Note that it doesn't need + // to check for sockets because it's done in newInodeOperations below. + deviceKey := device.MultiDeviceKey{ + Device: p9attr.RDev, + SecondaryDevice: i.session().connID, + Inode: qids[0].Path, + } + unlock := i.session().overrides.lock() + if pipeInode := i.session().overrides.getPipe(deviceKey); pipeInode != nil { + unlock() + pipeInode.IncRef() + return fs.NewDirent(ctx, pipeInode, name), nil + } + unlock() + } + // Construct the Inode operations. - sattr, node := newInodeOperations(ctx, i.fileState.s, newFile, qids[0], mask, p9attr, false) + sattr, node := newInodeOperations(ctx, i.fileState.s, newFile, qids[0], mask, p9attr) // Construct a positive Dirent. return fs.NewDirent(ctx, fs.NewInode(ctx, node, dir.MountSource, sattr), name), nil @@ -138,7 +165,7 @@ func (i *inodeOperations) Create(ctx context.Context, dir *fs.Inode, name string qid := qids[0] // Construct the InodeOperations. - sattr, iops := newInodeOperations(ctx, i.fileState.s, unopened, qid, mask, p9attr, false) + sattr, iops := newInodeOperations(ctx, i.fileState.s, unopened, qid, mask, p9attr) // Construct the positive Dirent. d := fs.NewDirent(ctx, fs.NewInode(ctx, iops, dir.MountSource, sattr), name) @@ -223,82 +250,115 @@ func (i *inodeOperations) Bind(ctx context.Context, dir *fs.Inode, name string, return nil, syserror.ENAMETOOLONG } - if i.session().endpoints == nil { + if i.session().overrides == nil { return nil, syscall.EOPNOTSUPP } - // Create replaces the directory fid with the newly created/opened - // file, so clone this directory so it doesn't change out from under - // this node. - _, newFile, err := i.fileState.file.walk(ctx, nil) + // Stabilize the override map while creation is in progress. + unlock := i.session().overrides.lock() + defer unlock() + + sattr, iops, err := i.createEndpointFile(ctx, dir, name, perm, p9.ModeSocket) if err != nil { return nil, err } - // We're not going to use newFile after return. - defer newFile.close(ctx) - // Stabilize the endpoint map while creation is in progress. - unlock := i.session().endpoints.lock() - defer unlock() + // Construct the positive Dirent. + childDir := fs.NewDirent(ctx, fs.NewInode(ctx, iops, dir.MountSource, sattr), name) + i.session().overrides.addBoundEndpoint(iops.fileState.key, childDir, ep) + return childDir, nil +} - // Create a regular file in the gofer and then mark it as a socket by - // adding this inode key in the 'endpoints' map. - owner := fs.FileOwnerFromContext(ctx) - hostFile, err := newFile.create(ctx, name, p9.ReadWrite, p9.FileMode(perm.LinuxMode()), p9.UID(owner.UID), p9.GID(owner.GID)) - if err != nil { - return nil, err +// CreateFifo implements fs.InodeOperations.CreateFifo. +func (i *inodeOperations) CreateFifo(ctx context.Context, dir *fs.Inode, name string, perm fs.FilePermissions) error { + if len(name) > maxFilenameLen { + return syserror.ENAMETOOLONG } - // We're not going to use this file. - hostFile.Close() - i.touchModificationAndStatusChangeTime(ctx, dir) + owner := fs.FileOwnerFromContext(ctx) + mode := p9.FileMode(perm.LinuxMode()) | p9.ModeNamedPipe - // Get the attributes of the file to create inode key. - qid, mask, attr, err := getattr(ctx, newFile) - if err != nil { - return nil, err + // N.B. FIFOs use major/minor numbers 0. + if _, err := i.fileState.file.mknod(ctx, name, mode, 0, 0, p9.UID(owner.UID), p9.GID(owner.GID)); err != nil { + if i.session().overrides == nil || err != syscall.EPERM { + return err + } + // If gofer doesn't support mknod, check if we can create an internal fifo. + return i.createInternalFifo(ctx, dir, name, owner, perm) } - key := device.MultiDeviceKey{ - Device: attr.RDev, - SecondaryDevice: i.session().connID, - Inode: qid.Path, + i.touchModificationAndStatusChangeTime(ctx, dir) + return nil +} + +func (i *inodeOperations) createInternalFifo(ctx context.Context, dir *fs.Inode, name string, owner fs.FileOwner, perm fs.FilePermissions) error { + if i.session().overrides == nil { + return syserror.EPERM } - // Create child dirent. + // Stabilize the override map while creation is in progress. + unlock := i.session().overrides.lock() + defer unlock() - // Get an unopened p9.File for the file we created so that it can be - // cloned and re-opened multiple times after creation. - _, unopened, err := i.fileState.file.walk(ctx, []string{name}) + sattr, fileOps, err := i.createEndpointFile(ctx, dir, name, perm, p9.ModeNamedPipe) if err != nil { - return nil, err + return err } - // Construct the InodeOperations. - sattr, iops := newInodeOperations(ctx, i.fileState.s, unopened, qid, mask, attr, true) + // First create a pipe. + p := pipe.NewPipe(true /* isNamed */, pipe.DefaultPipeSize, usermem.PageSize) + + // Wrap the fileOps with our Fifo. + iops := &fifo{ + InodeOperations: pipe.NewInodeOperations(ctx, perm, p), + fileIops: fileOps, + } + inode := fs.NewInode(ctx, iops, dir.MountSource, sattr) // Construct the positive Dirent. childDir := fs.NewDirent(ctx, fs.NewInode(ctx, iops, dir.MountSource, sattr), name) - i.session().endpoints.add(key, childDir, ep) - return childDir, nil + i.session().overrides.addPipe(fileOps.fileState.key, childDir, inode) + return nil } -// CreateFifo implements fs.InodeOperations.CreateFifo. -func (i *inodeOperations) CreateFifo(ctx context.Context, dir *fs.Inode, name string, perm fs.FilePermissions) error { - if len(name) > maxFilenameLen { - return syserror.ENAMETOOLONG +// Caller must hold Session.endpoint lock. +func (i *inodeOperations) createEndpointFile(ctx context.Context, dir *fs.Inode, name string, perm fs.FilePermissions, fileType p9.FileMode) (fs.StableAttr, *inodeOperations, error) { + _, dirClone, err := i.fileState.file.walk(ctx, nil) + if err != nil { + return fs.StableAttr{}, nil, err } + // We're not going to use dirClone after return. + defer dirClone.close(ctx) + // Create a regular file in the gofer and then mark it as a socket by + // adding this inode key in the 'overrides' map. owner := fs.FileOwnerFromContext(ctx) - mode := p9.FileMode(perm.LinuxMode()) | p9.ModeNamedPipe - - // N.B. FIFOs use major/minor numbers 0. - if _, err := i.fileState.file.mknod(ctx, name, mode, 0, 0, p9.UID(owner.UID), p9.GID(owner.GID)); err != nil { - return err + hostFile, err := dirClone.create(ctx, name, p9.ReadWrite, p9.FileMode(perm.LinuxMode()), p9.UID(owner.UID), p9.GID(owner.GID)) + if err != nil { + return fs.StableAttr{}, nil, err } + // We're not going to use this file. + hostFile.Close() i.touchModificationAndStatusChangeTime(ctx, dir) - return nil + + // Get the attributes of the file to create inode key. + qid, mask, attr, err := getattr(ctx, dirClone) + if err != nil { + return fs.StableAttr{}, nil, err + } + + // Get an unopened p9.File for the file we created so that it can be + // cloned and re-opened multiple times after creation. + _, unopened, err := i.fileState.file.walk(ctx, []string{name}) + if err != nil { + return fs.StableAttr{}, nil, err + } + + // Construct new inode with file type overridden. + attr.Mode = changeType(attr.Mode, fileType) + sattr, iops := newInodeOperations(ctx, i.fileState.s, unopened, qid, mask, attr) + return sattr, iops, nil } // Remove implements InodeOperations.Remove. @@ -307,20 +367,23 @@ func (i *inodeOperations) Remove(ctx context.Context, dir *fs.Inode, name string return syserror.ENAMETOOLONG } - var key device.MultiDeviceKey - removeSocket := false - if i.session().endpoints != nil { - // Find out if file being deleted is a socket that needs to be + var key *device.MultiDeviceKey + if i.session().overrides != nil { + // Find out if file being deleted is a socket or pipe that needs to be // removed from endpoint map. if d, err := i.Lookup(ctx, dir, name); err == nil { defer d.DecRef() - if fs.IsSocket(d.Inode.StableAttr) { - child := d.Inode.InodeOperations.(*inodeOperations) - key = child.fileState.key - removeSocket = true - // Stabilize the endpoint map while deletion is in progress. - unlock := i.session().endpoints.lock() + if fs.IsSocket(d.Inode.StableAttr) || fs.IsPipe(d.Inode.StableAttr) { + switch iops := d.Inode.InodeOperations.(type) { + case *inodeOperations: + key = &iops.fileState.key + case *fifo: + key = &iops.fileIops.fileState.key + } + + // Stabilize the override map while deletion is in progress. + unlock := i.session().overrides.lock() defer unlock() } } @@ -329,8 +392,8 @@ func (i *inodeOperations) Remove(ctx context.Context, dir *fs.Inode, name string if err := i.fileState.file.unlinkAt(ctx, name, 0); err != nil { return err } - if removeSocket { - i.session().endpoints.remove(key) + if key != nil { + i.session().overrides.remove(*key) } i.touchModificationAndStatusChangeTime(ctx, dir) diff --git a/pkg/sentry/fs/gofer/session.go b/pkg/sentry/fs/gofer/session.go index 498c4645a..f6b3ef178 100644 --- a/pkg/sentry/fs/gofer/session.go +++ b/pkg/sentry/fs/gofer/session.go @@ -33,60 +33,107 @@ import ( var DefaultDirentCacheSize uint64 = fs.DefaultDirentCacheSize // +stateify savable -type endpointMaps struct { - // mu protexts the direntMap, the keyMap, and the pathMap below. - mu sync.RWMutex `state:"nosave"` +type overrideInfo struct { + dirent *fs.Dirent + + // endpoint is set when dirent points to a socket. inode must not be set. + endpoint transport.BoundEndpoint + + // inode is set when dirent points to a pipe. endpoint must not be set. + inode *fs.Inode +} - // direntMap links sockets to their dirents. - // It is filled concurrently with the keyMap and is stored upon save. - // Before saving, this map is used to populate the pathMap. - direntMap map[transport.BoundEndpoint]*fs.Dirent +func (l *overrideInfo) inodeType() fs.InodeType { + switch { + case l.endpoint != nil: + return fs.Socket + case l.inode != nil: + return fs.Pipe + } + panic("endpoint or node must be set") +} - // keyMap links MultiDeviceKeys (containing inode IDs) to their sockets. +// +stateify savable +type overrideMaps struct { + // mu protexts the keyMap, and the pathMap below. + mu sync.RWMutex `state:"nosave"` + + // keyMap links MultiDeviceKeys (containing inode IDs) to their sockets/pipes. // It is not stored during save because the inode ID may change upon restore. - keyMap map[device.MultiDeviceKey]transport.BoundEndpoint `state:"nosave"` + keyMap map[device.MultiDeviceKey]*overrideInfo `state:"nosave"` - // pathMap links the sockets to their paths. + // pathMap links the sockets/pipes to their paths. // It is filled before saving from the direntMap and is stored upon save. // Upon restore, this map is used to re-populate the keyMap. - pathMap map[transport.BoundEndpoint]string + pathMap map[*overrideInfo]string +} + +// addBoundEndpoint adds the bound endpoint to the map. +// A reference is taken on the dirent argument. +// +// Precondition: maps must have been locked with 'lock'. +func (e *overrideMaps) addBoundEndpoint(key device.MultiDeviceKey, d *fs.Dirent, ep transport.BoundEndpoint) { + d.IncRef() + e.keyMap[key] = &overrideInfo{dirent: d, endpoint: ep} } -// add adds the endpoint to the maps. +// addPipe adds the pipe inode to the map. // A reference is taken on the dirent argument. // // Precondition: maps must have been locked with 'lock'. -func (e *endpointMaps) add(key device.MultiDeviceKey, d *fs.Dirent, ep transport.BoundEndpoint) { - e.keyMap[key] = ep +func (e *overrideMaps) addPipe(key device.MultiDeviceKey, d *fs.Dirent, inode *fs.Inode) { d.IncRef() - e.direntMap[ep] = d + e.keyMap[key] = &overrideInfo{dirent: d, inode: inode} } // remove deletes the key from the maps. // // Precondition: maps must have been locked with 'lock'. -func (e *endpointMaps) remove(key device.MultiDeviceKey) { - endpoint := e.get(key) +func (e *overrideMaps) remove(key device.MultiDeviceKey) { + endpoint := e.keyMap[key] delete(e.keyMap, key) - - d := e.direntMap[endpoint] - d.DecRef() - delete(e.direntMap, endpoint) + endpoint.dirent.DecRef() } // lock blocks other addition and removal operations from happening while // the backing file is being created or deleted. Returns a function that unlocks // the endpoint map. -func (e *endpointMaps) lock() func() { +func (e *overrideMaps) lock() func() { e.mu.Lock() return func() { e.mu.Unlock() } } -// get returns the endpoint mapped to the given key. +// getBoundEndpoint returns the bound endpoint mapped to the given key. // -// Precondition: maps must have been locked for reading. -func (e *endpointMaps) get(key device.MultiDeviceKey) transport.BoundEndpoint { - return e.keyMap[key] +// Precondition: maps must have been locked. +func (e *overrideMaps) getBoundEndpoint(key device.MultiDeviceKey) transport.BoundEndpoint { + if v := e.keyMap[key]; v != nil { + return v.endpoint + } + return nil +} + +// getPipe returns the pipe inode mapped to the given key. +// +// Precondition: maps must have been locked. +func (e *overrideMaps) getPipe(key device.MultiDeviceKey) *fs.Inode { + if v := e.keyMap[key]; v != nil { + return v.inode + } + return nil +} + +// getType returns the inode type if there is a corresponding endpoint for the +// given key. Returns false otherwise. +func (e *overrideMaps) getType(key device.MultiDeviceKey) (fs.InodeType, bool) { + e.mu.Lock() + v := e.keyMap[key] + e.mu.Unlock() + + if v != nil { + return v.inodeType(), true + } + return 0, false } // session holds state for each 9p session established during sys_mount. @@ -137,16 +184,16 @@ type session struct { // mounter is the EUID/EGID that mounted this file system. mounter fs.FileOwner `state:"wait"` - // endpoints is used to map inodes that represent socket files to their - // corresponding endpoint. Socket files are created as regular files in the - // gofer and their presence in this map indicate that they should indeed be - // socket files. This allows unix domain sockets to be used with paths that - // belong to a gofer. + // overrides is used to map inodes that represent socket/pipes files to their + // corresponding endpoint/iops. These files are created as regular files in + // the gofer and their presence in this map indicate that they should indeed + // be socket/pipe files. This allows unix domain sockets and named pipes to + // be used with paths that belong to a gofer. // // TODO(gvisor.dev/issue/1200): there are few possible races with someone // stat'ing the file and another deleting it concurrently, where the file // will not be reported as socket file. - endpoints *endpointMaps `state:"wait"` + overrides *overrideMaps `state:"wait"` } // Destroy tears down the session. @@ -179,15 +226,21 @@ func (s *session) SaveInodeMapping(inode *fs.Inode, path string) { // This is very unintuitive. We *CANNOT* trust the inode's StableAttrs, // because overlay copyUp may have changed them out from under us. // So much for "immutable". - sattr := inode.InodeOperations.(*inodeOperations).fileState.sattr - s.inodeMappings[sattr.InodeID] = path + switch iops := inode.InodeOperations.(type) { + case *inodeOperations: + s.inodeMappings[iops.fileState.sattr.InodeID] = path + case *fifo: + s.inodeMappings[iops.fileIops.fileState.sattr.InodeID] = path + default: + panic(fmt.Sprintf("Invalid type: %T", iops)) + } } -// newInodeOperations creates a new 9p fs.InodeOperations backed by a p9.File and attributes -// (p9.QID, p9.AttrMask, p9.Attr). +// newInodeOperations creates a new 9p fs.InodeOperations backed by a p9.File +// and attributes (p9.QID, p9.AttrMask, p9.Attr). // // Endpoints lock must not be held if socket == false. -func newInodeOperations(ctx context.Context, s *session, file contextFile, qid p9.QID, valid p9.AttrMask, attr p9.Attr, socket bool) (fs.StableAttr, *inodeOperations) { +func newInodeOperations(ctx context.Context, s *session, file contextFile, qid p9.QID, valid p9.AttrMask, attr p9.Attr) (fs.StableAttr, *inodeOperations) { deviceKey := device.MultiDeviceKey{ Device: attr.RDev, SecondaryDevice: s.connID, @@ -201,17 +254,11 @@ func newInodeOperations(ctx context.Context, s *session, file contextFile, qid p BlockSize: bsize(attr), } - if s.endpoints != nil { - if socket { - sattr.Type = fs.Socket - } else { - // If unix sockets are allowed on this filesystem, check if this file is - // supposed to be a socket file. - unlock := s.endpoints.lock() - if s.endpoints.get(deviceKey) != nil { - sattr.Type = fs.Socket - } - unlock() + if s.overrides != nil && sattr.Type == fs.RegularFile { + // If overrides are allowed on this filesystem, check if this file is + // supposed to be of a different type, e.g. socket. + if t, ok := s.overrides.getType(deviceKey); ok { + sattr.Type = t } } @@ -267,7 +314,7 @@ func Root(ctx context.Context, dev string, filesystem fs.Filesystem, superBlockF s.EnableLeakCheck("gofer.session") if o.privateunixsocket { - s.endpoints = newEndpointMaps() + s.overrides = newOverrideMaps() } // Construct the MountSource with the session and superBlockFlags. @@ -305,26 +352,24 @@ func Root(ctx context.Context, dev string, filesystem fs.Filesystem, superBlockF return nil, err } - sattr, iops := newInodeOperations(ctx, &s, s.attach, qid, valid, attr, false) + sattr, iops := newInodeOperations(ctx, &s, s.attach, qid, valid, attr) return fs.NewInode(ctx, iops, m, sattr), nil } -// newEndpointMaps creates a new endpointMaps. -func newEndpointMaps() *endpointMaps { - return &endpointMaps{ - direntMap: make(map[transport.BoundEndpoint]*fs.Dirent), - keyMap: make(map[device.MultiDeviceKey]transport.BoundEndpoint), - pathMap: make(map[transport.BoundEndpoint]string), +// newOverrideMaps creates a new overrideMaps. +func newOverrideMaps() *overrideMaps { + return &overrideMaps{ + keyMap: make(map[device.MultiDeviceKey]*overrideInfo), + pathMap: make(map[*overrideInfo]string), } } -// fillKeyMap populates key and dirent maps upon restore from saved -// pathmap. +// fillKeyMap populates key and dirent maps upon restore from saved pathmap. func (s *session) fillKeyMap(ctx context.Context) error { - unlock := s.endpoints.lock() + unlock := s.overrides.lock() defer unlock() - for ep, dirPath := range s.endpoints.pathMap { + for ep, dirPath := range s.overrides.pathMap { _, file, err := s.attach.walk(ctx, splitAbsolutePath(dirPath)) if err != nil { return fmt.Errorf("error filling endpointmaps, failed to walk to %q: %v", dirPath, err) @@ -341,25 +386,25 @@ func (s *session) fillKeyMap(ctx context.Context) error { Inode: qid.Path, } - s.endpoints.keyMap[key] = ep + s.overrides.keyMap[key] = ep } return nil } -// fillPathMap populates paths for endpoints from dirents in direntMap +// fillPathMap populates paths for overrides from dirents in direntMap // before save. func (s *session) fillPathMap() error { - unlock := s.endpoints.lock() + unlock := s.overrides.lock() defer unlock() - for ep, dir := range s.endpoints.direntMap { - mountRoot := dir.MountRoot() + for _, endpoint := range s.overrides.keyMap { + mountRoot := endpoint.dirent.MountRoot() defer mountRoot.DecRef() - dirPath, _ := dir.FullName(mountRoot) + dirPath, _ := endpoint.dirent.FullName(mountRoot) if dirPath == "" { return fmt.Errorf("error getting path from dirent") } - s.endpoints.pathMap[ep] = dirPath + s.overrides.pathMap[endpoint] = dirPath } return nil } @@ -368,7 +413,7 @@ func (s *session) fillPathMap() error { func (s *session) restoreEndpointMaps(ctx context.Context) error { // When restoring, only need to create the keyMap because the dirent and path // maps got stored through the save. - s.endpoints.keyMap = make(map[device.MultiDeviceKey]transport.BoundEndpoint) + s.overrides.keyMap = make(map[device.MultiDeviceKey]*overrideInfo) if err := s.fillKeyMap(ctx); err != nil { return fmt.Errorf("failed to insert sockets into endpoint map: %v", err) } @@ -376,6 +421,6 @@ func (s *session) restoreEndpointMaps(ctx context.Context) error { // Re-create pathMap because it can no longer be trusted as socket paths can // change while process continues to run. Empty pathMap will be re-filled upon // next save. - s.endpoints.pathMap = make(map[transport.BoundEndpoint]string) + s.overrides.pathMap = make(map[*overrideInfo]string) return nil } diff --git a/pkg/sentry/fs/gofer/session_state.go b/pkg/sentry/fs/gofer/session_state.go index 0285c5361..111da59f9 100644 --- a/pkg/sentry/fs/gofer/session_state.go +++ b/pkg/sentry/fs/gofer/session_state.go @@ -25,9 +25,9 @@ import ( // beforeSave is invoked by stateify. func (s *session) beforeSave() { - if s.endpoints != nil { + if s.overrides != nil { if err := s.fillPathMap(); err != nil { - panic("failed to save paths to endpoint map before saving" + err.Error()) + panic("failed to save paths to override map before saving" + err.Error()) } } } @@ -74,10 +74,10 @@ func (s *session) afterLoad() { panic(fmt.Sprintf("new attach name %v, want %v", opts.aname, s.aname)) } - // Check if endpointMaps exist when uds sockets are enabled - // (only pathmap will actualy have been saved). - if opts.privateunixsocket != (s.endpoints != nil) { - panic(fmt.Sprintf("new privateunixsocket option %v, want %v", opts.privateunixsocket, s.endpoints != nil)) + // Check if overrideMaps exist when uds sockets are enabled (only pathmaps + // will actually have been saved). + if opts.privateunixsocket != (s.overrides != nil) { + panic(fmt.Sprintf("new privateunixsocket option %v, want %v", opts.privateunixsocket, s.overrides != nil)) } if args.Flags != s.superBlockFlags { panic(fmt.Sprintf("new mount flags %v, want %v", args.Flags, s.superBlockFlags)) diff --git a/pkg/sentry/fs/gofer/socket.go b/pkg/sentry/fs/gofer/socket.go index 376cfce2c..10ba2f5f0 100644 --- a/pkg/sentry/fs/gofer/socket.go +++ b/pkg/sentry/fs/gofer/socket.go @@ -32,15 +32,15 @@ func (i *inodeOperations) BoundEndpoint(inode *fs.Inode, path string) transport. return nil } - if i.session().endpoints != nil { - unlock := i.session().endpoints.lock() + if i.session().overrides != nil { + unlock := i.session().overrides.lock() defer unlock() - ep := i.session().endpoints.get(i.fileState.key) + ep := i.session().overrides.getBoundEndpoint(i.fileState.key) if ep != nil { return ep } - // Not found in endpoints map, it may be a gofer backed unix socket... + // Not found in overrides map, it may be a gofer backed unix socket... } inode.IncRef() diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 40e974314..8f2b75a1c 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -225,7 +225,6 @@ syscall_test( syscall_test( add_overlay = True, test = "//test/syscalls/linux:mknod_test", - use_tmpfs = True, # mknod is not supported over gofer. ) syscall_test( -- cgit v1.2.3 From 492229d0176c1af2ab4ea4cf91bf211e940b5b12 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Tue, 4 Feb 2020 11:28:36 -0800 Subject: VFS2 gofer client Updates #1198 Opening host pipes (by spinning in fdpipe) and host sockets is not yet complete, and will be done in a future CL. Major differences from VFS1 gofer client (sentry/fs/gofer), with varying levels of backportability: - "Cache policies" are replaced by InteropMode, which control the behavior of timestamps in addition to caching. Under InteropModeExclusive (analogous to cacheAll) and InteropModeWritethrough (analogous to cacheAllWritethrough), client timestamps are *not* written back to the server (it is not possible in 9P or Linux for clients to set ctime, so writing back client-authoritative timestamps results in incoherence between atime/mtime and ctime). Under InteropModeShared (analogous to cacheRemoteRevalidating), client timestamps are not used at all (remote filesystem clocks are authoritative). cacheNone is translated to InteropModeShared + new option filesystemOptions.specialRegularFiles. - Under InteropModeShared, "unstable attribute" reloading for permission checks, lookup, and revalidation are fused, which is feasible in VFS2 since gofer.filesystem controls path resolution. This results in a ~33% reduction in RPCs for filesystem operations compared to cacheRemoteRevalidating. For example, consider stat("/foo/bar/baz") where "/foo/bar/baz" fails revalidation, resulting in the instantiation of a new dentry: VFS1 RPCs: getattr("/") // fs.MountNamespace.FindLink() => fs.Inode.CheckPermission() => gofer.inodeOperations.check() => gofer.inodeOperations.UnstableAttr() walkgetattr("/", "foo") = fid1 // fs.Dirent.walk() => gofer.session.Revalidate() => gofer.cachePolicy.Revalidate() clunk(fid1) getattr("/foo") // CheckPermission walkgetattr("/foo", "bar") = fid2 // Revalidate clunk(fid2) getattr("/foo/bar") // CheckPermission walkgetattr("/foo/bar", "baz") = fid3 // Revalidate clunk(fid3) walkgetattr("/foo/bar", "baz") = fid4 // fs.Dirent.walk() => gofer.inodeOperations.Lookup getattr("/foo/bar/baz") // linux.stat() => gofer.inodeOperations.UnstableAttr() VFS2 RPCs: getattr("/") // gofer.filesystem.walkExistingLocked() walkgetattr("/", "foo") = fid1 // gofer.filesystem.stepExistingLocked() clunk(fid1) // No getattr: walkgetattr already updated metadata for permission check walkgetattr("/foo", "bar") = fid2 clunk(fid2) walkgetattr("/foo/bar", "baz") = fid3 // No clunk: fid3 used for new gofer.dentry // No getattr: walkgetattr already updated metadata for stat() - gofer.filesystem.unlinkAt() does not require instantiation of a dentry that represents the file to be deleted. Updates #898. - gofer.regularFileFD.OnClose() skips Tflushf for regular files under InteropModeExclusive, as it's nonsensical to request a remote file flush without flushing locally-buffered writes to that remote file first. - Symlink targets are cached when InteropModeShared is not in effect. - p9.QID.Path (which is already required to be unique for each file within a server, and is accordingly already synthesized from device/inode numbers in all known gofers) is used as-is for inode numbers, rather than being mapped along with attr.RDev in the client to yet another synthetic inode number. - Relevant parts of fsutil.CachingInodeOperations are inlined directly into gofer package code. This avoids having to duplicate part of its functionality in fsutil.HostMappable. PiperOrigin-RevId: 293190213 --- pkg/safemem/seq_unsafe.go | 17 + pkg/sentry/fs/fsutil/BUILD | 4 +- pkg/sentry/fs/fsutil/frame_ref_set.go | 13 +- pkg/sentry/fs/fsutil/inode_cached.go | 2 +- pkg/sentry/fsimpl/gofer/BUILD | 55 ++ pkg/sentry/fsimpl/gofer/directory.go | 190 +++++ pkg/sentry/fsimpl/gofer/filesystem.go | 1087 ++++++++++++++++++++++++++++ pkg/sentry/fsimpl/gofer/gofer.go | 1147 ++++++++++++++++++++++++++++++ pkg/sentry/fsimpl/gofer/handle.go | 135 ++++ pkg/sentry/fsimpl/gofer/handle_unsafe.go | 66 ++ pkg/sentry/fsimpl/gofer/p9file.go | 219 ++++++ pkg/sentry/fsimpl/gofer/pagemath.go | 31 + pkg/sentry/fsimpl/gofer/regular_file.go | 860 ++++++++++++++++++++++ pkg/sentry/fsimpl/gofer/special_file.go | 159 +++++ pkg/sentry/fsimpl/gofer/symlink.go | 47 ++ pkg/sentry/fsimpl/gofer/time.go | 75 ++ pkg/sentry/fsimpl/tmpfs/filesystem.go | 2 +- pkg/sentry/socket/hostinet/socket.go | 23 +- 18 files changed, 4103 insertions(+), 29 deletions(-) create mode 100644 pkg/sentry/fsimpl/gofer/BUILD create mode 100644 pkg/sentry/fsimpl/gofer/directory.go create mode 100644 pkg/sentry/fsimpl/gofer/filesystem.go create mode 100644 pkg/sentry/fsimpl/gofer/gofer.go create mode 100644 pkg/sentry/fsimpl/gofer/handle.go create mode 100644 pkg/sentry/fsimpl/gofer/handle_unsafe.go create mode 100644 pkg/sentry/fsimpl/gofer/p9file.go create mode 100644 pkg/sentry/fsimpl/gofer/pagemath.go create mode 100644 pkg/sentry/fsimpl/gofer/regular_file.go create mode 100644 pkg/sentry/fsimpl/gofer/special_file.go create mode 100644 pkg/sentry/fsimpl/gofer/symlink.go create mode 100644 pkg/sentry/fsimpl/gofer/time.go (limited to 'pkg/sentry') diff --git a/pkg/safemem/seq_unsafe.go b/pkg/safemem/seq_unsafe.go index 354a95dde..dcdfc9600 100644 --- a/pkg/safemem/seq_unsafe.go +++ b/pkg/safemem/seq_unsafe.go @@ -18,6 +18,7 @@ import ( "bytes" "fmt" "reflect" + "syscall" "unsafe" ) @@ -297,3 +298,19 @@ func ZeroSeq(dsts BlockSeq) (uint64, error) { } return done, nil } + +// IovecsFromBlockSeq returns a []syscall.Iovec representing seq. +func IovecsFromBlockSeq(bs BlockSeq) []syscall.Iovec { + iovs := make([]syscall.Iovec, 0, bs.NumBlocks()) + for ; !bs.IsEmpty(); bs = bs.Tail() { + b := bs.Head() + iovs = append(iovs, syscall.Iovec{ + Base: &b.ToSlice()[0], + Len: uint64(b.Len()), + }) + // We don't need to care about b.NeedSafecopy(), because the host + // kernel will handle such address ranges just fine (by returning + // EFAULT). + } + return iovs +} diff --git a/pkg/sentry/fs/fsutil/BUILD b/pkg/sentry/fs/fsutil/BUILD index 4ab2a384f..789369220 100644 --- a/pkg/sentry/fs/fsutil/BUILD +++ b/pkg/sentry/fs/fsutil/BUILD @@ -28,13 +28,13 @@ go_template_instance( "platform": "gvisor.dev/gvisor/pkg/sentry/platform", }, package = "fsutil", - prefix = "frameRef", + prefix = "FrameRef", template = "//pkg/segment:generic_set", types = { "Key": "uint64", "Range": "platform.FileRange", "Value": "uint64", - "Functions": "frameRefSetFunctions", + "Functions": "FrameRefSetFunctions", }, ) diff --git a/pkg/sentry/fs/fsutil/frame_ref_set.go b/pkg/sentry/fs/fsutil/frame_ref_set.go index dd63db32b..6564fd0c6 100644 --- a/pkg/sentry/fs/fsutil/frame_ref_set.go +++ b/pkg/sentry/fs/fsutil/frame_ref_set.go @@ -20,24 +20,25 @@ import ( "gvisor.dev/gvisor/pkg/sentry/platform" ) -type frameRefSetFunctions struct{} +// FrameRefSetFunctions implements segment.Functions for FrameRefSet. +type FrameRefSetFunctions struct{} // MinKey implements segment.Functions.MinKey. -func (frameRefSetFunctions) MinKey() uint64 { +func (FrameRefSetFunctions) MinKey() uint64 { return 0 } // MaxKey implements segment.Functions.MaxKey. -func (frameRefSetFunctions) MaxKey() uint64 { +func (FrameRefSetFunctions) MaxKey() uint64 { return math.MaxUint64 } // ClearValue implements segment.Functions.ClearValue. -func (frameRefSetFunctions) ClearValue(val *uint64) { +func (FrameRefSetFunctions) ClearValue(val *uint64) { } // Merge implements segment.Functions.Merge. -func (frameRefSetFunctions) Merge(_ platform.FileRange, val1 uint64, _ platform.FileRange, val2 uint64) (uint64, bool) { +func (FrameRefSetFunctions) Merge(_ platform.FileRange, val1 uint64, _ platform.FileRange, val2 uint64) (uint64, bool) { if val1 != val2 { return 0, false } @@ -45,6 +46,6 @@ func (frameRefSetFunctions) Merge(_ platform.FileRange, val1 uint64, _ platform. } // Split implements segment.Functions.Split. -func (frameRefSetFunctions) Split(_ platform.FileRange, val uint64, _ uint64) (uint64, uint64) { +func (FrameRefSetFunctions) Split(_ platform.FileRange, val uint64, _ uint64) (uint64, uint64) { return val, val } diff --git a/pkg/sentry/fs/fsutil/inode_cached.go b/pkg/sentry/fs/fsutil/inode_cached.go index 573b8586e..800c8b4e1 100644 --- a/pkg/sentry/fs/fsutil/inode_cached.go +++ b/pkg/sentry/fs/fsutil/inode_cached.go @@ -111,7 +111,7 @@ type CachingInodeOperations struct { // refs tracks active references to data in the cache. // // refs is protected by dataMu. - refs frameRefSet + refs FrameRefSet } // CachingInodeOperationsOptions configures a CachingInodeOperations. diff --git a/pkg/sentry/fsimpl/gofer/BUILD b/pkg/sentry/fsimpl/gofer/BUILD new file mode 100644 index 000000000..4ba76a1e8 --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/BUILD @@ -0,0 +1,55 @@ +load("//tools:defs.bzl", "go_library") +load("//tools/go_generics:defs.bzl", "go_template_instance") + +licenses(["notice"]) + +go_template_instance( + name = "dentry_list", + out = "dentry_list.go", + package = "gofer", + prefix = "dentry", + template = "//pkg/ilist:generic_list", + types = { + "Element": "*dentry", + "Linker": "*dentry", + }, +) + +go_library( + name = "gofer", + srcs = [ + "dentry_list.go", + "directory.go", + "filesystem.go", + "gofer.go", + "handle.go", + "handle_unsafe.go", + "p9file.go", + "pagemath.go", + "regular_file.go", + "special_file.go", + "symlink.go", + "time.go", + ], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/abi/linux", + "//pkg/context", + "//pkg/fd", + "//pkg/fspath", + "//pkg/log", + "//pkg/p9", + "//pkg/safemem", + "//pkg/sentry/fs/fsutil", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/kernel/time", + "//pkg/sentry/memmap", + "//pkg/sentry/pgalloc", + "//pkg/sentry/platform", + "//pkg/sentry/usage", + "//pkg/sentry/vfs", + "//pkg/syserror", + "//pkg/unet", + "//pkg/usermem", + ], +) diff --git a/pkg/sentry/fsimpl/gofer/directory.go b/pkg/sentry/fsimpl/gofer/directory.go new file mode 100644 index 000000000..baa2cdd8e --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/directory.go @@ -0,0 +1,190 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package gofer + +import ( + "sync" + "sync/atomic" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/p9" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +func (d *dentry) isDir() bool { + return d.fileType() == linux.S_IFDIR +} + +// Preconditions: d.dirMu must be locked. d.isDir(). fs.opts.interop != +// InteropModeShared. +func (d *dentry) cacheNegativeChildLocked(name string) { + if d.negativeChildren == nil { + d.negativeChildren = make(map[string]struct{}) + } + d.negativeChildren[name] = struct{}{} +} + +type directoryFD struct { + fileDescription + vfs.DirectoryFileDescriptionDefaultImpl + + mu sync.Mutex + off int64 + dirents []vfs.Dirent +} + +// Release implements vfs.FileDescriptionImpl.Release. +func (fd *directoryFD) Release() { +} + +// IterDirents implements vfs.FileDescriptionImpl.IterDirents. +func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback) error { + fd.mu.Lock() + defer fd.mu.Unlock() + + if fd.dirents == nil { + ds, err := fd.dentry().getDirents(ctx) + if err != nil { + return err + } + fd.dirents = ds + } + + for fd.off < int64(len(fd.dirents)) { + if !cb.Handle(fd.dirents[fd.off]) { + return nil + } + fd.off++ + } + return nil +} + +// Preconditions: d.isDir(). There exists at least one directoryFD representing d. +func (d *dentry) getDirents(ctx context.Context) ([]vfs.Dirent, error) { + // 9P2000.L's readdir does not specify behavior in the presence of + // concurrent mutation of an iterated directory, so implementations may + // duplicate or omit entries in this case, which violates POSIX semantics. + // Thus we read all directory entries while holding d.dirMu to exclude + // directory mutations. (Note that it is impossible for the client to + // exclude concurrent mutation from other remote filesystem users. Since + // there is no way to detect if the server has incorrectly omitted + // directory entries, we simply assume that the server is well-behaved + // under InteropModeShared.) This is inconsistent with Linux (which appears + // to assume that directory fids have the correct semantics, and translates + // struct file_operations::readdir calls directly to readdir RPCs), but is + // consistent with VFS1. + + d.fs.renameMu.RLock() + defer d.fs.renameMu.RUnlock() + d.dirMu.Lock() + defer d.dirMu.Unlock() + if d.dirents != nil { + return d.dirents, nil + } + + // It's not clear if 9P2000.L's readdir is expected to return "." and "..", + // so we generate them here. + parent := d.vfsd.ParentOrSelf().Impl().(*dentry) + dirents := []vfs.Dirent{ + { + Name: ".", + Type: linux.DT_DIR, + Ino: d.ino, + NextOff: 1, + }, + { + Name: "..", + Type: uint8(atomic.LoadUint32(&parent.mode) >> 12), + Ino: parent.ino, + NextOff: 2, + }, + } + off := uint64(0) + const count = 64 * 1024 // for consistency with the vfs1 client + d.handleMu.RLock() + defer d.handleMu.RUnlock() + if !d.handleReadable { + // This should not be possible because a readable handle should have + // been opened when the calling directoryFD was opened. + panic("gofer.dentry.getDirents called without a readable handle") + } + for { + p9ds, err := d.handle.file.readdir(ctx, off, count) + if err != nil { + return nil, err + } + if len(p9ds) == 0 { + // Cache dirents for future directoryFDs if permitted. + if d.fs.opts.interop != InteropModeShared { + d.dirents = dirents + } + return dirents, nil + } + for _, p9d := range p9ds { + if p9d.Name == "." || p9d.Name == ".." { + continue + } + dirent := vfs.Dirent{ + Name: p9d.Name, + Ino: p9d.QID.Path, + NextOff: int64(len(dirents) + 1), + } + // p9 does not expose 9P2000.U's DMDEVICE, DMNAMEDPIPE, or + // DMSOCKET. + switch p9d.Type { + case p9.TypeSymlink: + dirent.Type = linux.DT_LNK + case p9.TypeDir: + dirent.Type = linux.DT_DIR + default: + dirent.Type = linux.DT_REG + } + dirents = append(dirents, dirent) + } + off = p9ds[len(p9ds)-1].Offset + } +} + +// Seek implements vfs.FileDescriptionImpl.Seek. +func (fd *directoryFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + fd.mu.Lock() + defer fd.mu.Unlock() + + switch whence { + case linux.SEEK_SET: + if offset < 0 { + return 0, syserror.EINVAL + } + if offset == 0 { + // Ensure that the next call to fd.IterDirents() calls + // fd.dentry().getDirents(). + fd.dirents = nil + } + fd.off = offset + return fd.off, nil + case linux.SEEK_CUR: + offset += fd.off + if offset < 0 { + return 0, syserror.EINVAL + } + // Don't clear fd.dirents in this case, even if offset == 0. + fd.off = offset + return fd.off, nil + default: + return 0, syserror.EINVAL + } +} diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go new file mode 100644 index 000000000..8eb61debf --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -0,0 +1,1087 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package gofer + +import ( + "sync" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/p9" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// Sync implements vfs.FilesystemImpl.Sync. +func (fs *filesystem) Sync(ctx context.Context) error { + // Snapshot current dentries and special files. + fs.syncMu.Lock() + ds := make([]*dentry, 0, len(fs.dentries)) + for d := range fs.dentries { + ds = append(ds, d) + } + sffds := make([]*specialFileFD, 0, len(fs.specialFileFDs)) + for sffd := range fs.specialFileFDs { + sffds = append(sffds, sffd) + } + fs.syncMu.Unlock() + + // Return the first error we encounter, but sync everything we can + // regardless. + var retErr error + + // Sync regular files. + for _, d := range ds { + if !d.TryIncRef() { + continue + } + err := d.syncSharedHandle(ctx) + d.DecRef() + if err != nil && retErr == nil { + retErr = err + } + } + + // Sync special files, which may be writable but do not use dentry shared + // handles (so they won't be synced by the above). + for _, sffd := range sffds { + if !sffd.vfsfd.TryIncRef() { + continue + } + err := sffd.Sync(ctx) + sffd.vfsfd.DecRef() + if err != nil && retErr == nil { + retErr = err + } + } + + return retErr +} + +// maxFilenameLen is the maximum length of a filename. This is dictated by 9P's +// encoding of strings, which uses 2 bytes for the length prefix. +const maxFilenameLen = (1 << 16) - 1 + +// dentrySlicePool is a pool of *[]*dentry used to store dentries for which +// dentry.checkCachingLocked() must be called. The pool holds pointers to +// slices because Go lacks generics, so sync.Pool operates on interface{}, so +// every call to (what should be) sync.Pool<[]*dentry>.Put() allocates a copy +// of the slice header on the heap. +var dentrySlicePool = sync.Pool{ + New: func() interface{} { + ds := make([]*dentry, 0, 4) // arbitrary non-zero initial capacity + return &ds + }, +} + +func appendDentry(ds *[]*dentry, d *dentry) *[]*dentry { + if ds == nil { + ds = dentrySlicePool.Get().(*[]*dentry) + } + *ds = append(*ds, d) + return ds +} + +// Preconditions: ds != nil. +func putDentrySlice(ds *[]*dentry) { + // Allow dentries to be GC'd. + for i := range *ds { + (*ds)[i] = nil + } + *ds = (*ds)[:0] + dentrySlicePool.Put(ds) +} + +// stepLocked resolves rp.Component() to an existing file, starting from the +// given directory. +// +// Dentries which may become cached as a result of the traversal are appended +// to *ds. +// +// Preconditions: fs.renameMu must be locked. d.dirMu must be locked. +// !rp.Done(). If fs.opts.interop == InteropModeShared, then d's cached +// metadata must be up to date. +func (fs *filesystem) stepLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry, ds **[]*dentry) (*dentry, error) { + if !d.isDir() { + return nil, syserror.ENOTDIR + } + if err := d.checkPermissions(rp.Credentials(), vfs.MayExec, true); err != nil { + return nil, err + } +afterSymlink: + name := rp.Component() + if name == "." { + rp.Advance() + return d, nil + } + if name == ".." { + parentVFSD, err := rp.ResolveParent(&d.vfsd) + if err != nil { + return nil, err + } + parent := parentVFSD.Impl().(*dentry) + if fs.opts.interop == InteropModeShared { + // We must assume that parentVFSD is correct, because if d has been + // moved elsewhere in the remote filesystem so that its parent has + // changed, we have no way of determining its new parent's location + // in the filesystem. Get updated metadata for parentVFSD. + _, attrMask, attr, err := parent.file.getAttr(ctx, dentryAttrMask()) + if err != nil { + return nil, err + } + parent.updateFromP9Attrs(attrMask, &attr) + } + rp.Advance() + return parent, nil + } + childVFSD, err := rp.ResolveChild(&d.vfsd, name) + if err != nil { + return nil, err + } + // FIXME(jamieliu): Linux performs revalidation before mount lookup + // (fs/namei.c:lookup_fast() => __d_lookup_rcu(), d_revalidate(), + // __follow_mount_rcu()). + child, err := fs.revalidateChildLocked(ctx, rp.VirtualFilesystem(), d, name, childVFSD, ds) + if err != nil { + return nil, err + } + if child == nil { + return nil, syserror.ENOENT + } + if child.isSymlink() && rp.ShouldFollowSymlink() { + target, err := child.readlink(ctx, rp.Mount()) + if err != nil { + return nil, err + } + if err := rp.HandleSymlink(target); err != nil { + return nil, err + } + goto afterSymlink // don't check the current directory again + } + rp.Advance() + return child, nil +} + +// revalidateChildLocked must be called after a call to parent.vfsd.Child(name) +// or vfs.ResolvingPath.ResolveChild(name) returns childVFSD (which may be +// nil) to verify that the returned child (or lack thereof) is correct. If no file +// exists at name, revalidateChildLocked returns (nil, nil). +// +// Preconditions: fs.renameMu must be locked. parent.dirMu must be locked. +// parent.isDir(). name is not "." or "..". +// +// Postconditions: If revalidateChildLocked returns a non-nil dentry, its +// cached metadata is up to date. +func (fs *filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.VirtualFilesystem, parent *dentry, name string, childVFSD *vfs.Dentry, ds **[]*dentry) (*dentry, error) { + if childVFSD != nil && fs.opts.interop != InteropModeShared { + // We have a cached dentry that is assumed to be correct. + return childVFSD.Impl().(*dentry), nil + } + // We either don't have a cached dentry or need to verify that it's still + // correct, either of which requires a remote lookup. Check if this name is + // valid before performing the lookup. + if len(name) > maxFilenameLen { + return nil, syserror.ENAMETOOLONG + } + // Check if we've already cached this lookup with a negative result. + if _, ok := parent.negativeChildren[name]; ok { + return nil, nil + } + // Perform the remote lookup. + qid, file, attrMask, attr, err := parent.file.walkGetAttrOne(ctx, name) + if err != nil && err != syserror.ENOENT { + return nil, err + } + if childVFSD != nil { + child := childVFSD.Impl().(*dentry) + if !file.isNil() && qid.Path == child.ino { + // The file at this path hasn't changed. Just update cached + // metadata. + file.close(ctx) + child.updateFromP9Attrs(attrMask, &attr) + return child, nil + } + // The file at this path has changed or no longer exists. Remove + // the stale dentry from the tree, and re-evaluate its caching + // status (i.e. if it has 0 references, drop it). + vfsObj.ForceDeleteDentry(childVFSD) + *ds = appendDentry(*ds, child) + childVFSD = nil + } + if file.isNil() { + // No file exists at this path now. Cache the negative lookup if + // allowed. + if fs.opts.interop != InteropModeShared { + parent.cacheNegativeChildLocked(name) + } + return nil, nil + } + // Create a new dentry representing the file. + child, err := fs.newDentry(ctx, file, qid, attrMask, &attr) + if err != nil { + file.close(ctx) + return nil, err + } + parent.IncRef() // reference held by child on its parent + parent.vfsd.InsertChild(&child.vfsd, name) + // For now, child has 0 references, so our caller should call + // child.checkCachingLocked(). + *ds = appendDentry(*ds, child) + return child, nil +} + +// walkParentDirLocked resolves all but the last path component of rp to an +// existing directory, starting from the given directory (which is usually +// rp.Start().Impl().(*dentry)). It does not check that the returned directory +// is searchable by the provider of rp. +// +// Preconditions: fs.renameMu must be locked. !rp.Done(). If fs.opts.interop == +// InteropModeShared, then d's cached metadata must be up to date. +func (fs *filesystem) walkParentDirLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry, ds **[]*dentry) (*dentry, error) { + for !rp.Final() { + d.dirMu.Lock() + next, err := fs.stepLocked(ctx, rp, d, ds) + d.dirMu.Unlock() + if err != nil { + return nil, err + } + d = next + } + if !d.isDir() { + return nil, syserror.ENOTDIR + } + return d, nil +} + +// resolveLocked resolves rp to an existing file. +// +// Preconditions: fs.renameMu must be locked. +func (fs *filesystem) resolveLocked(ctx context.Context, rp *vfs.ResolvingPath, ds **[]*dentry) (*dentry, error) { + d := rp.Start().Impl().(*dentry) + if fs.opts.interop == InteropModeShared { + // Get updated metadata for rp.Start() as required by fs.stepLocked(). + if err := d.updateFromGetattr(ctx); err != nil { + return nil, err + } + } + for !rp.Done() { + d.dirMu.Lock() + next, err := fs.stepLocked(ctx, rp, d, ds) + d.dirMu.Unlock() + if err != nil { + return nil, err + } + d = next + } + if rp.MustBeDir() && !d.isDir() { + return nil, syserror.ENOTDIR + } + return d, nil +} + +// doCreateAt checks that creating a file at rp is permitted, then invokes +// create to do so. +// +// Preconditions: !rp.Done(). For the final path component in rp, +// !rp.ShouldFollowSymlink(). +func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir bool, create func(parent *dentry, name string) error) error { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + start := rp.Start().Impl().(*dentry) + if fs.opts.interop == InteropModeShared { + // Get updated metadata for start as required by + // fs.walkParentDirLocked(). + if err := start.updateFromGetattr(ctx); err != nil { + return err + } + } + parent, err := fs.walkParentDirLocked(ctx, rp, start, &ds) + if err != nil { + return err + } + if err := parent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true); err != nil { + return err + } + if parent.isDeleted() { + return syserror.ENOENT + } + name := rp.Component() + if name == "." || name == ".." { + return syserror.EEXIST + } + if len(name) > maxFilenameLen { + return syserror.ENAMETOOLONG + } + if !dir && rp.MustBeDir() { + return syserror.ENOENT + } + mnt := rp.Mount() + if err := mnt.CheckBeginWrite(); err != nil { + return err + } + defer mnt.EndWrite() + parent.dirMu.Lock() + defer parent.dirMu.Unlock() + if fs.opts.interop == InteropModeShared { + // The existence of a dentry at name would be inconclusive because the + // file it represents may have been deleted from the remote filesystem, + // so we would need to make an RPC to revalidate the dentry. Just + // attempt the file creation RPC instead. If a file does exist, the RPC + // will fail with EEXIST like we would have. If the RPC succeeds, and a + // stale dentry exists, the dentry will fail revalidation next time + // it's used. + return create(parent, name) + } + if parent.vfsd.Child(name) != nil { + return syserror.EEXIST + } + // No cached dentry exists; however, there might still be an existing file + // at name. As above, we attempt the file creation RPC anyway. + if err := create(parent, name); err != nil { + return err + } + parent.touchCMtime(ctx) + delete(parent.negativeChildren, name) + parent.dirents = nil + return nil +} + +// Preconditions: !rp.Done(). +func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir bool) error { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + start := rp.Start().Impl().(*dentry) + if fs.opts.interop == InteropModeShared { + // Get updated metadata for start as required by + // fs.walkParentDirLocked(). + if err := start.updateFromGetattr(ctx); err != nil { + return err + } + } + parent, err := fs.walkParentDirLocked(ctx, rp, start, &ds) + if err != nil { + return err + } + if err := parent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true); err != nil { + return err + } + if err := rp.Mount().CheckBeginWrite(); err != nil { + return err + } + defer rp.Mount().EndWrite() + + name := rp.Component() + if dir { + if name == "." { + return syserror.EINVAL + } + if name == ".." { + return syserror.ENOTEMPTY + } + } else { + if name == "." || name == ".." { + return syserror.EISDIR + } + } + vfsObj := rp.VirtualFilesystem() + mntns := vfs.MountNamespaceFromContext(ctx) + parent.dirMu.Lock() + defer parent.dirMu.Unlock() + childVFSD := parent.vfsd.Child(name) + var child *dentry + // We only need a dentry representing the file at name if it can be a mount + // point. If childVFSD is nil, then it can't be a mount point. If childVFSD + // is non-nil but stale, the actual file can't be a mount point either; we + // detect this case by just speculatively calling PrepareDeleteDentry and + // only revalidating the dentry if that fails (indicating that the existing + // dentry is a mount point). + if childVFSD != nil { + child = childVFSD.Impl().(*dentry) + if err := vfsObj.PrepareDeleteDentry(mntns, childVFSD); err != nil { + child, err = fs.revalidateChildLocked(ctx, vfsObj, parent, name, childVFSD, &ds) + if err != nil { + return err + } + if child != nil { + childVFSD = &child.vfsd + if err := vfsObj.PrepareDeleteDentry(mntns, childVFSD); err != nil { + return err + } + } else { + childVFSD = nil + } + } + } else if _, ok := parent.negativeChildren[name]; ok { + return syserror.ENOENT + } + flags := uint32(0) + if dir { + if child != nil && !child.isDir() { + return syserror.ENOTDIR + } + flags = linux.AT_REMOVEDIR + } else { + if child != nil && child.isDir() { + return syserror.EISDIR + } + if rp.MustBeDir() { + return syserror.ENOTDIR + } + } + err = parent.file.unlinkAt(ctx, name, flags) + if err != nil { + if childVFSD != nil { + vfsObj.AbortDeleteDentry(childVFSD) + } + return err + } + if fs.opts.interop != InteropModeShared { + parent.touchCMtime(ctx) + parent.cacheNegativeChildLocked(name) + parent.dirents = nil + } + if child != nil { + child.setDeleted() + vfsObj.CommitDeleteDentry(childVFSD) + ds = appendDentry(ds, child) + } + return nil +} + +// renameMuRUnlockAndCheckCaching calls fs.renameMu.RUnlock(), then calls +// dentry.checkCachingLocked on all dentries in *ds with fs.renameMu locked for +// writing. +// +// ds is a pointer-to-pointer since defer evaluates its arguments immediately, +// but dentry slices are allocated lazily, and it's much easier to say "defer +// fs.renameMuRUnlockAndCheckCaching(&ds)" than "defer func() { +// fs.renameMuRUnlockAndCheckCaching(ds) }()" to work around this. +func (fs *filesystem) renameMuRUnlockAndCheckCaching(ds **[]*dentry) { + fs.renameMu.RUnlock() + if *ds == nil { + return + } + if len(**ds) != 0 { + fs.renameMu.Lock() + for _, d := range **ds { + d.checkCachingLocked() + } + fs.renameMu.Unlock() + } + putDentrySlice(*ds) +} + +func (fs *filesystem) renameMuUnlockAndCheckCaching(ds **[]*dentry) { + if *ds == nil { + fs.renameMu.Unlock() + return + } + for _, d := range **ds { + d.checkCachingLocked() + } + fs.renameMu.Unlock() + putDentrySlice(*ds) +} + +// GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. +func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return nil, err + } + if opts.CheckSearchable { + if !d.isDir() { + return nil, syserror.ENOTDIR + } + if err := d.checkPermissions(rp.Credentials(), vfs.MayExec, true); err != nil { + return nil, err + } + } + d.IncRef() + return &d.vfsd, nil +} + +// GetParentDentryAt implements vfs.FilesystemImpl.GetParentDentryAt. +func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + start := rp.Start().Impl().(*dentry) + if fs.opts.interop == InteropModeShared { + // Get updated metadata for start as required by + // fs.walkParentDirLocked(). + if err := start.updateFromGetattr(ctx); err != nil { + return nil, err + } + } + d, err := fs.walkParentDirLocked(ctx, rp, start, &ds) + if err != nil { + return nil, err + } + d.IncRef() + return &d.vfsd, nil +} + +// LinkAt implements vfs.FilesystemImpl.LinkAt. +func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry) error { + return fs.doCreateAt(ctx, rp, false /* dir */, func(parent *dentry, childName string) error { + if rp.Mount() != vd.Mount() { + return syserror.EXDEV + } + // 9P2000.L supports hard links, but we don't. + return syserror.EPERM + }) +} + +// MkdirAt implements vfs.FilesystemImpl.MkdirAt. +func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error { + return fs.doCreateAt(ctx, rp, true /* dir */, func(parent *dentry, name string) error { + creds := rp.Credentials() + _, err := parent.file.mkdir(ctx, name, (p9.FileMode)(opts.Mode), (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)) + return err + }) +} + +// MknodAt implements vfs.FilesystemImpl.MknodAt. +func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MknodOptions) error { + return fs.doCreateAt(ctx, rp, false /* dir */, func(parent *dentry, name string) error { + creds := rp.Credentials() + _, err := parent.file.mknod(ctx, name, (p9.FileMode)(opts.Mode), opts.DevMajor, opts.DevMinor, (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)) + return err + }) +} + +// OpenAt implements vfs.FilesystemImpl.OpenAt. +func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + // Reject O_TMPFILE, which is not supported; supporting it correctly in the + // presence of other remote filesystem users requires remote filesystem + // support, and it isn't clear that there's any way to implement this in + // 9P. + if opts.Flags&linux.O_TMPFILE != 0 { + return nil, syserror.EOPNOTSUPP + } + mayCreate := opts.Flags&linux.O_CREAT != 0 + mustCreate := opts.Flags&(linux.O_CREAT|linux.O_EXCL) == (linux.O_CREAT | linux.O_EXCL) + + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + + start := rp.Start().Impl().(*dentry) + if fs.opts.interop == InteropModeShared { + // Get updated metadata for start as required by fs.stepLocked(). + if err := start.updateFromGetattr(ctx); err != nil { + return nil, err + } + } + if rp.Done() { + return start.openLocked(ctx, rp, opts.Flags) + } + +afterTrailingSymlink: + parent, err := fs.walkParentDirLocked(ctx, rp, start, &ds) + if err != nil { + return nil, err + } + // Check for search permission in the parent directory. + if err := parent.checkPermissions(rp.Credentials(), vfs.MayExec, true); err != nil { + return nil, err + } + // Determine whether or not we need to create a file. + parent.dirMu.Lock() + child, err := fs.stepLocked(ctx, rp, parent, &ds) + if err == syserror.ENOENT && mayCreate { + fd, err := parent.createAndOpenChildLocked(ctx, rp, &opts) + parent.dirMu.Unlock() + return fd, err + } + if err != nil { + parent.dirMu.Unlock() + return nil, err + } + // Open existing child or follow symlink. + parent.dirMu.Unlock() + if mustCreate { + return nil, syserror.EEXIST + } + if child.isSymlink() && rp.ShouldFollowSymlink() { + target, err := child.readlink(ctx, rp.Mount()) + if err != nil { + return nil, err + } + if err := rp.HandleSymlink(target); err != nil { + return nil, err + } + start = parent + goto afterTrailingSymlink + } + return child.openLocked(ctx, rp, opts.Flags) +} + +// Preconditions: fs.renameMu must be locked. +func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, flags uint32) (*vfs.FileDescription, error) { + ats := vfs.AccessTypesForOpenFlags(flags) + if err := d.checkPermissions(rp.Credentials(), ats, d.isDir()); err != nil { + return nil, err + } + mnt := rp.Mount() + filetype := d.fileType() + switch { + case filetype == linux.S_IFREG && !d.fs.opts.regularFilesUseSpecialFileFD: + if err := d.ensureSharedHandle(ctx, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, flags&linux.O_TRUNC != 0); err != nil { + return nil, err + } + fd := ®ularFileFD{} + if err := fd.vfsfd.Init(fd, flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{ + AllowDirectIO: true, + }); err != nil { + return nil, err + } + return &fd.vfsfd, nil + case filetype == linux.S_IFDIR: + // Can't open directories with O_CREAT. + if flags&linux.O_CREAT != 0 { + return nil, syserror.EISDIR + } + // Can't open directories writably. + if ats&vfs.MayWrite != 0 { + return nil, syserror.EISDIR + } + if flags&linux.O_DIRECT != 0 { + return nil, syserror.EINVAL + } + if err := d.ensureSharedHandle(ctx, ats&vfs.MayRead != 0, false /* write */, false /* trunc */); err != nil { + return nil, err + } + fd := &directoryFD{} + if err := fd.vfsfd.Init(fd, flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { + return nil, err + } + return &fd.vfsfd, nil + case filetype == linux.S_IFLNK: + // Can't open symlinks without O_PATH (which is unimplemented). + return nil, syserror.ELOOP + default: + if flags&linux.O_DIRECT != 0 { + return nil, syserror.EINVAL + } + h, err := openHandle(ctx, d.file, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, flags&linux.O_TRUNC != 0) + if err != nil { + return nil, err + } + fd := &specialFileFD{ + handle: h, + } + if err := fd.vfsfd.Init(fd, flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { + h.close(ctx) + return nil, err + } + return &fd.vfsfd, nil + } +} + +// Preconditions: d.fs.renameMu must be locked. d.dirMu must be locked. +func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.OpenOptions) (*vfs.FileDescription, error) { + if err := d.checkPermissions(rp.Credentials(), vfs.MayWrite, true); err != nil { + return nil, err + } + if d.isDeleted() { + return nil, syserror.ENOENT + } + mnt := rp.Mount() + if err := mnt.CheckBeginWrite(); err != nil { + return nil, err + } + defer mnt.EndWrite() + + // 9P2000.L's lcreate takes a fid representing the parent directory, and + // converts it into an open fid representing the created file, so we need + // to duplicate the directory fid first. + _, dirfile, err := d.file.walk(ctx, nil) + if err != nil { + return nil, err + } + creds := rp.Credentials() + name := rp.Component() + fdobj, openFile, createQID, _, err := dirfile.create(ctx, name, (p9.OpenFlags)(opts.Flags), (p9.FileMode)(opts.Mode), (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)) + if err != nil { + dirfile.close(ctx) + return nil, err + } + // Then we need to walk to the file we just created to get a non-open fid + // representing it, and to get its metadata. This must use d.file since, as + // explained above, dirfile was invalidated by dirfile.Create(). + walkQID, nonOpenFile, attrMask, attr, err := d.file.walkGetAttrOne(ctx, name) + if err != nil { + openFile.close(ctx) + if fdobj != nil { + fdobj.Close() + } + return nil, err + } + // Sanity-check that we walked to the file we created. + if createQID.Path != walkQID.Path { + // Probably due to concurrent remote filesystem mutation? + ctx.Warningf("gofer.dentry.createAndOpenChildLocked: created file has QID %v before walk, QID %v after (interop=%v)", createQID, walkQID, d.fs.opts.interop) + nonOpenFile.close(ctx) + openFile.close(ctx) + if fdobj != nil { + fdobj.Close() + } + return nil, syserror.EAGAIN + } + + // Construct the new dentry. + child, err := d.fs.newDentry(ctx, nonOpenFile, createQID, attrMask, &attr) + if err != nil { + nonOpenFile.close(ctx) + openFile.close(ctx) + if fdobj != nil { + fdobj.Close() + } + return nil, err + } + // Incorporate the fid that was opened by lcreate. + useRegularFileFD := child.fileType() == linux.S_IFREG && !d.fs.opts.regularFilesUseSpecialFileFD + if useRegularFileFD { + child.handleMu.Lock() + child.handle.file = openFile + if fdobj != nil { + child.handle.fd = int32(fdobj.Release()) + } + child.handleReadable = vfs.MayReadFileWithOpenFlags(opts.Flags) + child.handleWritable = vfs.MayWriteFileWithOpenFlags(opts.Flags) + child.handleMu.Unlock() + } + // Take a reference on the new dentry to be held by the new file + // description. (This reference also means that the new dentry is not + // eligible for caching yet, so we don't need to append to a dentry slice.) + child.refs = 1 + // Insert the dentry into the tree. + d.IncRef() // reference held by child on its parent d + d.vfsd.InsertChild(&child.vfsd, name) + if d.fs.opts.interop != InteropModeShared { + d.touchCMtime(ctx) + delete(d.negativeChildren, name) + d.dirents = nil + } + + // Finally, construct a file description representing the created file. + var childVFSFD *vfs.FileDescription + mnt.IncRef() + if useRegularFileFD { + fd := ®ularFileFD{} + if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &child.vfsd, &vfs.FileDescriptionOptions{ + AllowDirectIO: true, + }); err != nil { + return nil, err + } + childVFSFD = &fd.vfsfd + } else { + fd := &specialFileFD{ + handle: handle{ + file: openFile, + fd: -1, + }, + } + if fdobj != nil { + fd.handle.fd = int32(fdobj.Release()) + } + if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &child.vfsd, &vfs.FileDescriptionOptions{}); err != nil { + fd.handle.close(ctx) + return nil, err + } + childVFSFD = &fd.vfsfd + } + return childVFSFD, nil +} + +// ReadlinkAt implements vfs.FilesystemImpl.ReadlinkAt. +func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return "", err + } + if !d.isSymlink() { + return "", syserror.EINVAL + } + return d.readlink(ctx, rp.Mount()) +} + +// RenameAt implements vfs.FilesystemImpl.RenameAt. +func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldParentVD vfs.VirtualDentry, oldName string, opts vfs.RenameOptions) error { + if opts.Flags != 0 { + // Requires 9P support. + return syserror.EINVAL + } + + var ds *[]*dentry + fs.renameMu.Lock() + defer fs.renameMuUnlockAndCheckCaching(&ds) + newParent, err := fs.walkParentDirLocked(ctx, rp, rp.Start().Impl().(*dentry), &ds) + if err != nil { + return err + } + newName := rp.Component() + if newName == "." || newName == ".." { + return syserror.EBUSY + } + mnt := rp.Mount() + if mnt != oldParentVD.Mount() { + return syserror.EXDEV + } + if err := mnt.CheckBeginWrite(); err != nil { + return err + } + defer mnt.EndWrite() + + oldParent := oldParentVD.Dentry().Impl().(*dentry) + if fs.opts.interop == InteropModeShared { + if err := oldParent.updateFromGetattr(ctx); err != nil { + return err + } + } + if err := oldParent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true); err != nil { + return err + } + vfsObj := rp.VirtualFilesystem() + // We need a dentry representing the renamed file since, if it's a + // directory, we need to check for write permission on it. + oldParent.dirMu.Lock() + defer oldParent.dirMu.Unlock() + renamed, err := fs.revalidateChildLocked(ctx, vfsObj, oldParent, oldName, oldParent.vfsd.Child(oldName), &ds) + if err != nil { + return err + } + if renamed == nil { + return syserror.ENOENT + } + if renamed.isDir() { + if renamed == newParent || renamed.vfsd.IsAncestorOf(&newParent.vfsd) { + return syserror.EINVAL + } + if oldParent != newParent { + if err := renamed.checkPermissions(rp.Credentials(), vfs.MayWrite, true); err != nil { + return err + } + } + } else { + if opts.MustBeDir || rp.MustBeDir() { + return syserror.ENOTDIR + } + } + + if oldParent != newParent { + if err := newParent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true); err != nil { + return err + } + newParent.dirMu.Lock() + defer newParent.dirMu.Unlock() + } + if newParent.isDeleted() { + return syserror.ENOENT + } + replacedVFSD := newParent.vfsd.Child(newName) + var replaced *dentry + // This is similar to unlinkAt, except: + // + // - We revalidate the replaced dentry unconditionally for simplicity. + // + // - If rp.MustBeDir(), then we need a dentry representing the replaced + // file regardless to confirm that it's a directory. + if replacedVFSD != nil || rp.MustBeDir() { + replaced, err = fs.revalidateChildLocked(ctx, vfsObj, newParent, newName, replacedVFSD, &ds) + if err != nil { + return err + } + if replaced != nil { + if replaced.isDir() { + if !renamed.isDir() { + return syserror.EISDIR + } + } else { + if rp.MustBeDir() || renamed.isDir() { + return syserror.ENOTDIR + } + } + replacedVFSD = &replaced.vfsd + } else { + replacedVFSD = nil + } + } + + if oldParent == newParent && oldName == newName { + return nil + } + if err := vfsObj.PrepareRenameDentry(vfs.MountNamespaceFromContext(ctx), &renamed.vfsd, replacedVFSD); err != nil { + return err + } + if err := renamed.file.rename(ctx, newParent.file, newName); err != nil { + vfsObj.AbortRenameDentry(&renamed.vfsd, replacedVFSD) + return err + } + if fs.opts.interop != InteropModeShared { + oldParent.cacheNegativeChildLocked(oldName) + oldParent.dirents = nil + delete(newParent.negativeChildren, newName) + newParent.dirents = nil + } + vfsObj.CommitRenameReplaceDentry(&renamed.vfsd, &newParent.vfsd, newName, replacedVFSD) + return nil +} + +// RmdirAt implements vfs.FilesystemImpl.RmdirAt. +func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error { + return fs.unlinkAt(ctx, rp, true /* dir */) +} + +// SetStatAt implements vfs.FilesystemImpl.SetStatAt. +func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return err + } + return d.setStat(ctx, rp.Credentials(), &opts.Stat, rp.Mount()) +} + +// StatAt implements vfs.FilesystemImpl.StatAt. +func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return linux.Statx{}, err + } + // Since walking updates metadata for all traversed dentries under + // InteropModeShared, including the returned one, we can return cached + // metadata here regardless of fs.opts.interop. + var stat linux.Statx + d.statTo(&stat) + return stat, nil +} + +// StatFSAt implements vfs.FilesystemImpl.StatFSAt. +func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return linux.Statfs{}, err + } + fsstat, err := d.file.statFS(ctx) + if err != nil { + return linux.Statfs{}, err + } + nameLen := uint64(fsstat.NameLength) + if nameLen > maxFilenameLen { + nameLen = maxFilenameLen + } + return linux.Statfs{ + // This is primarily for distinguishing a gofer file system in + // tests. Testing is important, so instead of defining + // something completely random, use a standard value. + Type: linux.V9FS_MAGIC, + BlockSize: int64(fsstat.BlockSize), + Blocks: fsstat.Blocks, + BlocksFree: fsstat.BlocksFree, + BlocksAvailable: fsstat.BlocksAvailable, + Files: fsstat.Files, + FilesFree: fsstat.FilesFree, + NameLength: nameLen, + }, nil +} + +// SymlinkAt implements vfs.FilesystemImpl.SymlinkAt. +func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, target string) error { + return fs.doCreateAt(ctx, rp, false /* dir */, func(parent *dentry, name string) error { + creds := rp.Credentials() + _, err := parent.file.symlink(ctx, target, name, (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)) + return err + }) +} + +// UnlinkAt implements vfs.FilesystemImpl.UnlinkAt. +func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error { + return fs.unlinkAt(ctx, rp, false /* dir */) +} + +// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. +func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return nil, err + } + return d.listxattr(ctx) +} + +// GetxattrAt implements vfs.FilesystemImpl.GetxattrAt. +func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) (string, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return "", err + } + return d.getxattr(ctx, name) +} + +// SetxattrAt implements vfs.FilesystemImpl.SetxattrAt. +func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return err + } + return d.setxattr(ctx, &opts) +} + +// RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. +func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return err + } + return d.removexattr(ctx, name) +} + +// PrependPath implements vfs.FilesystemImpl.PrependPath. +func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { + fs.renameMu.RLock() + defer fs.renameMu.RUnlock() + return vfs.GenericPrependPath(vfsroot, vd, b) +} diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go new file mode 100644 index 000000000..d0552bd99 --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -0,0 +1,1147 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package gofer provides a filesystem implementation that is backed by a 9p +// server, interchangably referred to as "gofers" throughout this package. +// +// Lock order: +// regularFileFD/directoryFD.mu +// filesystem.renameMu +// dentry.dirMu +// filesystem.syncMu +// dentry.metadataMu +// *** "memmap.Mappable locks" below this point +// dentry.mapsMu +// *** "memmap.Mappable locks taken by Translate" below this point +// dentry.handleMu +// dentry.dataMu +// +// Locking dentry.dirMu in multiple dentries requires holding +// filesystem.renameMu for writing. +package gofer + +import ( + "fmt" + "strconv" + "sync" + "sync/atomic" + "syscall" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/p9" + "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/memmap" + "gvisor.dev/gvisor/pkg/sentry/pgalloc" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/unet" + "gvisor.dev/gvisor/pkg/usermem" +) + +// FilesystemType implements vfs.FilesystemType. +type FilesystemType struct{} + +// filesystem implements vfs.FilesystemImpl. +type filesystem struct { + vfsfs vfs.Filesystem + + // mfp is used to allocate memory that caches regular file contents. mfp is + // immutable. + mfp pgalloc.MemoryFileProvider + + // Immutable options. + opts filesystemOptions + + // client is the client used by this filesystem. client is immutable. + client *p9.Client + + // uid and gid are the effective KUID and KGID of the filesystem's creator, + // and are used as the owner and group for files that don't specify one. + // uid and gid are immutable. + uid auth.KUID + gid auth.KGID + + // renameMu serves two purposes: + // + // - It synchronizes path resolution with renaming initiated by this + // client. + // + // - It is held by path resolution to ensure that reachable dentries remain + // valid. A dentry is reachable by path resolution if it has a non-zero + // reference count (such that it is usable as vfs.ResolvingPath.Start() or + // is reachable from its children), or if it is a child dentry (such that + // it is reachable from its parent). + renameMu sync.RWMutex + + // cachedDentries contains all dentries with 0 references. (Due to race + // conditions, it may also contain dentries with non-zero references.) + // cachedDentriesLen is the number of dentries in cachedDentries. These + // fields are protected by renameMu. + cachedDentries dentryList + cachedDentriesLen uint64 + + // dentries contains all dentries in this filesystem. specialFileFDs + // contains all open specialFileFDs. These fields are protected by syncMu. + syncMu sync.Mutex + dentries map[*dentry]struct{} + specialFileFDs map[*specialFileFD]struct{} +} + +type filesystemOptions struct { + // "Standard" 9P options. + fd int + aname string + interop InteropMode // derived from the "cache" mount option + msize uint32 + version string + + // maxCachedDentries is the maximum number of dentries with 0 references + // retained by the client. + maxCachedDentries uint64 + + // If forcePageCache is true, host FDs may not be used for application + // memory mappings even if available; instead, the client must perform its + // own caching of regular file pages. This is primarily useful for testing. + forcePageCache bool + + // If limitHostFDTranslation is true, apply maxFillRange() constraints to + // host FD mappings returned by dentry.(memmap.Mappable).Translate(). This + // makes memory accounting behavior more consistent between cases where + // host FDs are / are not available, but may increase the frequency of + // sentry-handled page faults on files for which a host FD is available. + limitHostFDTranslation bool + + // If overlayfsStaleRead is true, O_RDONLY host FDs provided by the remote + // filesystem may not be coherent with writable host FDs opened later, so + // mappings of the former must be replaced by mappings of the latter. This + // is usually only the case when the remote filesystem is an overlayfs + // mount on Linux < 4.19. + overlayfsStaleRead bool + + // If regularFilesUseSpecialFileFD is true, application FDs representing + // regular files will use distinct file handles for each FD, in the same + // way that application FDs representing "special files" such as sockets + // do. Note that this disables client caching and mmap for regular files. + regularFilesUseSpecialFileFD bool +} + +// InteropMode controls the client's interaction with other remote filesystem +// users. +type InteropMode uint32 + +const ( + // InteropModeExclusive is appropriate when the filesystem client is the + // only user of the remote filesystem. + // + // - The client may cache arbitrary filesystem state (file data, metadata, + // filesystem structure, etc.). + // + // - Client changes to filesystem state may be sent to the remote + // filesystem asynchronously, except when server permission checks are + // necessary. + // + // - File timestamps are based on client clocks. This ensures that users of + // the client observe timestamps that are coherent with their own clocks + // and consistent with Linux's semantics. However, since it is not always + // possible for clients to set arbitrary atimes and mtimes, and never + // possible for clients to set arbitrary ctimes, file timestamp changes are + // stored in the client only and never sent to the remote filesystem. + InteropModeExclusive InteropMode = iota + + // InteropModeWritethrough is appropriate when there are read-only users of + // the remote filesystem that expect to observe changes made by the + // filesystem client. + // + // - The client may cache arbitrary filesystem state. + // + // - Client changes to filesystem state must be sent to the remote + // filesystem synchronously. + // + // - File timestamps are based on client clocks. As a corollary, access + // timestamp changes from other remote filesystem users will not be visible + // to the client. + InteropModeWritethrough + + // InteropModeShared is appropriate when there are users of the remote + // filesystem that may mutate its state other than the client. + // + // - The client must verify cached filesystem state before using it. + // + // - Client changes to filesystem state must be sent to the remote + // filesystem synchronously. + // + // - File timestamps are based on server clocks. This is necessary to + // ensure that timestamp changes are synchronized between remote filesystem + // users. + // + // Note that the correctness of InteropModeShared depends on the server + // correctly implementing 9P fids (i.e. each fid immutably represents a + // single filesystem object), even in the presence of remote filesystem + // mutations from other users. If this is violated, the behavior of the + // client is undefined. + InteropModeShared +) + +// GetFilesystem implements vfs.FilesystemType.GetFilesystem. +func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { + mfp := pgalloc.MemoryFileProviderFromContext(ctx) + if mfp == nil { + ctx.Warningf("gofer.FilesystemType.GetFilesystem: context does not provide a pgalloc.MemoryFileProvider") + return nil, nil, syserror.EINVAL + } + + mopts := vfs.GenericParseMountOptions(opts.Data) + var fsopts filesystemOptions + + // Check that the transport is "fd". + trans, ok := mopts["trans"] + if !ok { + ctx.Warningf("gofer.FilesystemType.GetFilesystem: transport must be specified as 'trans=fd'") + return nil, nil, syserror.EINVAL + } + delete(mopts, "trans") + if trans != "fd" { + ctx.Warningf("gofer.FilesystemType.GetFilesystem: unsupported transport: trans=%s", trans) + return nil, nil, syserror.EINVAL + } + + // Check that read and write FDs are provided and identical. + rfdstr, ok := mopts["rfdno"] + if !ok { + ctx.Warningf("gofer.FilesystemType.GetFilesystem: read FD must be specified as 'rfdno=") + return nil, nil, syserror.EINVAL + } + delete(mopts, "rfdno") + rfd, err := strconv.Atoi(rfdstr) + if err != nil { + ctx.Warningf("gofer.FilesystemType.GetFilesystem: invalid read FD: rfdno=%s", rfdstr) + return nil, nil, syserror.EINVAL + } + wfdstr, ok := mopts["wfdno"] + if !ok { + ctx.Warningf("gofer.FilesystemType.GetFilesystem: write FD must be specified as 'wfdno=") + return nil, nil, syserror.EINVAL + } + delete(mopts, "wfdno") + wfd, err := strconv.Atoi(wfdstr) + if err != nil { + ctx.Warningf("gofer.FilesystemType.GetFilesystem: invalid write FD: wfdno=%s", wfdstr) + return nil, nil, syserror.EINVAL + } + if rfd != wfd { + ctx.Warningf("gofer.FilesystemType.GetFilesystem: read FD (%d) and write FD (%d) must be equal", rfd, wfd) + return nil, nil, syserror.EINVAL + } + fsopts.fd = rfd + + // Get the attach name. + fsopts.aname = "/" + if aname, ok := mopts["aname"]; ok { + delete(mopts, "aname") + fsopts.aname = aname + } + + // Parse the cache policy. For historical reasons, this defaults to the + // least generally-applicable option, InteropModeExclusive. + fsopts.interop = InteropModeExclusive + if cache, ok := mopts["cache"]; ok { + delete(mopts, "cache") + switch cache { + case "fscache": + fsopts.interop = InteropModeExclusive + case "fscache_writethrough": + fsopts.interop = InteropModeWritethrough + case "none": + fsopts.regularFilesUseSpecialFileFD = true + fallthrough + case "remote_revalidating": + fsopts.interop = InteropModeShared + default: + ctx.Warningf("gofer.FilesystemType.GetFilesystem: invalid cache policy: cache=%s", cache) + return nil, nil, syserror.EINVAL + } + } + + // Parse the 9P message size. + fsopts.msize = 1024 * 1024 // 1M, tested to give good enough performance up to 64M + if msizestr, ok := mopts["msize"]; ok { + delete(mopts, "msize") + msize, err := strconv.ParseUint(msizestr, 10, 32) + if err != nil { + ctx.Warningf("gofer.FilesystemType.GetFilesystem: invalid message size: msize=%s", msizestr) + return nil, nil, syserror.EINVAL + } + fsopts.msize = uint32(msize) + } + + // Parse the 9P protocol version. + fsopts.version = p9.HighestVersionString() + if version, ok := mopts["version"]; ok { + delete(mopts, "version") + fsopts.version = version + } + + // Parse the dentry cache limit. + fsopts.maxCachedDentries = 1000 + if str, ok := mopts["dentry_cache_limit"]; ok { + delete(mopts, "dentry_cache_limit") + maxCachedDentries, err := strconv.ParseUint(str, 10, 64) + if err != nil { + ctx.Warningf("gofer.FilesystemType.GetFilesystem: invalid dentry cache limit: dentry_cache_limit=%s", str) + return nil, nil, syserror.EINVAL + } + fsopts.maxCachedDentries = maxCachedDentries + } + + // Handle simple flags. + if _, ok := mopts["force_page_cache"]; ok { + delete(mopts, "force_page_cache") + fsopts.forcePageCache = true + } + if _, ok := mopts["limit_host_fd_translation"]; ok { + delete(mopts, "limit_host_fd_translation") + fsopts.limitHostFDTranslation = true + } + if _, ok := mopts["overlayfs_stale_read"]; ok { + delete(mopts, "overlayfs_stale_read") + fsopts.overlayfsStaleRead = true + } + // fsopts.regularFilesUseSpecialFileFD can only be enabled by specifying + // "cache=none". + + // Check for unparsed options. + if len(mopts) != 0 { + ctx.Warningf("gofer.FilesystemType.GetFilesystem: unknown options: %v", mopts) + return nil, nil, syserror.EINVAL + } + + // Establish a connection with the server. + conn, err := unet.NewSocket(fsopts.fd) + if err != nil { + return nil, nil, err + } + + // Perform version negotiation with the server. + ctx.UninterruptibleSleepStart(false) + client, err := p9.NewClient(conn, fsopts.msize, fsopts.version) + ctx.UninterruptibleSleepFinish(false) + if err != nil { + conn.Close() + return nil, nil, err + } + // Ownership of conn has been transferred to client. + + // Perform attach to obtain the filesystem root. + ctx.UninterruptibleSleepStart(false) + attached, err := client.Attach(fsopts.aname) + ctx.UninterruptibleSleepFinish(false) + if err != nil { + client.Close() + return nil, nil, err + } + attachFile := p9file{attached} + qid, attrMask, attr, err := attachFile.getAttr(ctx, dentryAttrMask()) + if err != nil { + attachFile.close(ctx) + client.Close() + return nil, nil, err + } + + // Construct the filesystem object. + fs := &filesystem{ + mfp: mfp, + opts: fsopts, + uid: creds.EffectiveKUID, + gid: creds.EffectiveKGID, + client: client, + dentries: make(map[*dentry]struct{}), + specialFileFDs: make(map[*specialFileFD]struct{}), + } + fs.vfsfs.Init(vfsObj, fs) + + // Construct the root dentry. + root, err := fs.newDentry(ctx, attachFile, qid, attrMask, &attr) + if err != nil { + attachFile.close(ctx) + fs.vfsfs.DecRef() + return nil, nil, err + } + // Set the root's reference count to 2. One reference is returned to the + // caller, and the other is deliberately leaked to prevent the root from + // being "cached" and subsequently evicted. Its resources will still be + // cleaned up by fs.Release(). + root.refs = 2 + + return &fs.vfsfs, &root.vfsd, nil +} + +// Release implements vfs.FilesystemImpl.Release. +func (fs *filesystem) Release() { + ctx := context.Background() + mf := fs.mfp.MemoryFile() + + fs.syncMu.Lock() + for d := range fs.dentries { + d.handleMu.Lock() + d.dataMu.Lock() + if d.handleWritable { + // Write dirty cached data to the remote file. + if err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, fs.mfp.MemoryFile(), d.handle.writeFromBlocksAt); err != nil { + log.Warningf("gofer.filesystem.Release: failed to flush dentry: %v", err) + } + // TODO(jamieliu): Do we need to flushf/fsync d? + } + // Discard cached pages. + d.cache.DropAll(mf) + d.dirty.RemoveAll() + d.dataMu.Unlock() + // Close the host fd if one exists. + if d.handle.fd >= 0 { + syscall.Close(int(d.handle.fd)) + d.handle.fd = -1 + } + d.handleMu.Unlock() + } + // There can't be any specialFileFDs still using fs, since each such + // FileDescription would hold a reference on a Mount holding a reference on + // fs. + fs.syncMu.Unlock() + + // Close the connection to the server. This implicitly clunks all fids. + fs.client.Close() +} + +// dentry implements vfs.DentryImpl. +type dentry struct { + vfsd vfs.Dentry + + // refs is the reference count. Each dentry holds a reference on its + // parent, even if disowned. refs is accessed using atomic memory + // operations. + refs int64 + + // fs is the owning filesystem. fs is immutable. + fs *filesystem + + // We don't support hard links, so each dentry maps 1:1 to an inode. + + // file is the unopened p9.File that backs this dentry. file is immutable. + file p9file + + // If deleted is non-zero, the file represented by this dentry has been + // deleted. deleted is accessed using atomic memory operations. + deleted uint32 + + // If cached is true, dentryEntry links dentry into + // filesystem.cachedDentries. cached and dentryEntry are protected by + // filesystem.renameMu. + cached bool + dentryEntry + + dirMu sync.Mutex + + // If this dentry represents a directory, and InteropModeShared is not in + // effect, negativeChildren is a set of child names in this directory that + // are known not to exist. negativeChildren is protected by dirMu. + negativeChildren map[string]struct{} + + // If this dentry represents a directory, InteropModeShared is not in + // effect, and dirents is not nil, it is a cache of all entries in the + // directory, in the order they were returned by the server. dirents is + // protected by dirMu. + dirents []vfs.Dirent + + // Cached metadata; protected by metadataMu and accessed using atomic + // memory operations unless otherwise specified. + metadataMu sync.Mutex + ino uint64 // immutable + mode uint32 // type is immutable, perms are mutable + uid uint32 // auth.KUID, but stored as raw uint32 for sync/atomic + gid uint32 // auth.KGID, but ... + blockSize uint32 // 0 if unknown + // Timestamps, all nsecs from the Unix epoch. + atime int64 + mtime int64 + ctime int64 + btime int64 + // File size, protected by both metadataMu and dataMu (i.e. both must be + // locked to mutate it). + size uint64 + + mapsMu sync.Mutex + + // If this dentry represents a regular file, mappings tracks mappings of + // the file into memmap.MappingSpaces. mappings is protected by mapsMu. + mappings memmap.MappingSet + + // If this dentry represents a regular file or directory: + // + // - handle is the I/O handle used by all regularFileFDs/directoryFDs + // representing this dentry. + // + // - handleReadable is true if handle is readable. + // + // - handleWritable is true if handle is writable. + // + // Invariants: + // + // - If handleReadable == handleWritable == false, then handle.file == nil + // (i.e. there is no open handle). Conversely, if handleReadable || + // handleWritable == true, then handle.file != nil (i.e. there is an open + // handle). + // + // - handleReadable and handleWritable cannot transition from true to false + // (i.e. handles may not be downgraded). + // + // These fields are protected by handleMu. + handleMu sync.RWMutex + handle handle + handleReadable bool + handleWritable bool + + dataMu sync.RWMutex + + // If this dentry represents a regular file that is client-cached, cache + // maps offsets into the cached file to offsets into + // filesystem.mfp.MemoryFile() that store the file's data. cache is + // protected by dataMu. + cache fsutil.FileRangeSet + + // If this dentry represents a regular file that is client-cached, dirty + // tracks dirty segments in cache. dirty is protected by dataMu. + dirty fsutil.DirtySet + + // pf implements platform.File for mappings of handle.fd. + pf dentryPlatformFile + + // If this dentry represents a symbolic link, InteropModeShared is not in + // effect, and haveTarget is true, target is the symlink target. haveTarget + // and target are protected by dataMu. + haveTarget bool + target string +} + +// dentryAttrMask returns a p9.AttrMask enabling all attributes used by the +// gofer client. +func dentryAttrMask() p9.AttrMask { + return p9.AttrMask{ + Mode: true, + UID: true, + GID: true, + ATime: true, + MTime: true, + CTime: true, + Size: true, + BTime: true, + } +} + +// newDentry creates a new dentry representing the given file. The dentry +// initially has no references, but is not cached; it is the caller's +// responsibility to set the dentry's reference count and/or call +// dentry.checkCachingLocked() as appropriate. +func (fs *filesystem) newDentry(ctx context.Context, file p9file, qid p9.QID, mask p9.AttrMask, attr *p9.Attr) (*dentry, error) { + if !mask.Mode { + ctx.Warningf("can't create gofer.dentry without file type") + return nil, syserror.EIO + } + if attr.Mode.FileType() == p9.ModeRegular && !mask.Size { + ctx.Warningf("can't create regular file gofer.dentry without file size") + return nil, syserror.EIO + } + + d := &dentry{ + fs: fs, + file: file, + ino: qid.Path, + mode: uint32(attr.Mode), + uid: uint32(fs.uid), + gid: uint32(fs.gid), + blockSize: usermem.PageSize, + handle: handle{ + fd: -1, + }, + } + d.pf.dentry = d + if mask.UID { + d.uid = uint32(attr.UID) + } + if mask.GID { + d.gid = uint32(attr.GID) + } + if mask.Size { + d.size = attr.Size + } + if attr.BlockSize != 0 { + d.blockSize = uint32(attr.BlockSize) + } + if mask.ATime { + d.atime = dentryTimestampFromP9(attr.ATimeSeconds, attr.ATimeNanoSeconds) + } + if mask.MTime { + d.mtime = dentryTimestampFromP9(attr.MTimeSeconds, attr.MTimeNanoSeconds) + } + if mask.CTime { + d.ctime = dentryTimestampFromP9(attr.CTimeSeconds, attr.CTimeNanoSeconds) + } + if mask.BTime { + d.btime = dentryTimestampFromP9(attr.BTimeSeconds, attr.BTimeNanoSeconds) + } + d.vfsd.Init(d) + + fs.syncMu.Lock() + fs.dentries[d] = struct{}{} + fs.syncMu.Unlock() + return d, nil +} + +// updateFromP9Attrs is called to update d's metadata after an update from the +// remote filesystem. +func (d *dentry) updateFromP9Attrs(mask p9.AttrMask, attr *p9.Attr) { + d.metadataMu.Lock() + if mask.Mode { + if got, want := uint32(attr.Mode.FileType()), d.fileType(); got != want { + d.metadataMu.Unlock() + panic(fmt.Sprintf("gofer.dentry file type changed from %#o to %#o", want, got)) + } + atomic.StoreUint32(&d.mode, uint32(attr.Mode)) + } + if mask.UID { + atomic.StoreUint32(&d.uid, uint32(attr.UID)) + } + if mask.GID { + atomic.StoreUint32(&d.gid, uint32(attr.GID)) + } + // There is no P9_GETATTR_* bit for I/O block size. + if attr.BlockSize != 0 { + atomic.StoreUint32(&d.blockSize, uint32(attr.BlockSize)) + } + if mask.ATime { + atomic.StoreInt64(&d.atime, dentryTimestampFromP9(attr.ATimeSeconds, attr.ATimeNanoSeconds)) + } + if mask.MTime { + atomic.StoreInt64(&d.mtime, dentryTimestampFromP9(attr.MTimeSeconds, attr.MTimeNanoSeconds)) + } + if mask.CTime { + atomic.StoreInt64(&d.ctime, dentryTimestampFromP9(attr.CTimeSeconds, attr.CTimeNanoSeconds)) + } + if mask.BTime { + atomic.StoreInt64(&d.btime, dentryTimestampFromP9(attr.BTimeSeconds, attr.BTimeNanoSeconds)) + } + if mask.Size { + d.dataMu.Lock() + atomic.StoreUint64(&d.size, attr.Size) + d.dataMu.Unlock() + } + d.metadataMu.Unlock() +} + +func (d *dentry) updateFromGetattr(ctx context.Context) error { + // Use d.handle.file, which represents a 9P fid that has been opened, in + // preference to d.file, which represents a 9P fid that has not. This may + // be significantly more efficient in some implementations. + var ( + file p9file + handleMuRLocked bool + ) + d.handleMu.RLock() + if !d.handle.file.isNil() { + file = d.handle.file + handleMuRLocked = true + } else { + file = d.file + d.handleMu.RUnlock() + } + _, attrMask, attr, err := file.getAttr(ctx, dentryAttrMask()) + if handleMuRLocked { + d.handleMu.RUnlock() + } + if err != nil { + return err + } + d.updateFromP9Attrs(attrMask, &attr) + return nil +} + +func (d *dentry) fileType() uint32 { + return atomic.LoadUint32(&d.mode) & linux.S_IFMT +} + +func (d *dentry) statTo(stat *linux.Statx) { + stat.Mask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_NLINK | linux.STATX_UID | linux.STATX_GID | linux.STATX_ATIME | linux.STATX_MTIME | linux.STATX_CTIME | linux.STATX_INO | linux.STATX_SIZE | linux.STATX_BLOCKS | linux.STATX_BTIME + stat.Blksize = atomic.LoadUint32(&d.blockSize) + stat.Nlink = 1 + if d.isDir() { + stat.Nlink = 2 + } + stat.UID = atomic.LoadUint32(&d.uid) + stat.GID = atomic.LoadUint32(&d.gid) + stat.Mode = uint16(atomic.LoadUint32(&d.mode)) + stat.Ino = d.ino + stat.Size = atomic.LoadUint64(&d.size) + // This is consistent with regularFileFD.Seek(), which treats regular files + // as having no holes. + stat.Blocks = (stat.Size + 511) / 512 + stat.Atime = statxTimestampFromDentry(atomic.LoadInt64(&d.atime)) + stat.Btime = statxTimestampFromDentry(atomic.LoadInt64(&d.btime)) + stat.Ctime = statxTimestampFromDentry(atomic.LoadInt64(&d.ctime)) + stat.Mtime = statxTimestampFromDentry(atomic.LoadInt64(&d.mtime)) + // TODO(jamieliu): device number +} + +func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *linux.Statx, mnt *vfs.Mount) error { + if stat.Mask == 0 { + return nil + } + if stat.Mask&^(linux.STATX_MODE|linux.STATX_UID|linux.STATX_GID|linux.STATX_ATIME|linux.STATX_MTIME|linux.STATX_SIZE) != 0 { + return syserror.EPERM + } + if err := vfs.CheckSetStat(creds, stat, uint16(atomic.LoadUint32(&d.mode))&^linux.S_IFMT, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))); err != nil { + return err + } + if err := mnt.CheckBeginWrite(); err != nil { + return err + } + defer mnt.EndWrite() + setLocalAtime := false + setLocalMtime := false + if d.fs.opts.interop != InteropModeShared { + // Timestamp updates will be handled locally. + setLocalAtime = stat.Mask&linux.STATX_ATIME != 0 + setLocalMtime = stat.Mask&linux.STATX_MTIME != 0 + stat.Mask &^= linux.STATX_ATIME | linux.STATX_MTIME + if !setLocalMtime && (stat.Mask&linux.STATX_SIZE != 0) { + // Truncate updates mtime. + setLocalMtime = true + stat.Mtime.Nsec = linux.UTIME_NOW + } + } + d.metadataMu.Lock() + defer d.metadataMu.Unlock() + if stat.Mask != 0 { + if err := d.file.setAttr(ctx, p9.SetAttrMask{ + Permissions: stat.Mask&linux.STATX_MODE != 0, + UID: stat.Mask&linux.STATX_UID != 0, + GID: stat.Mask&linux.STATX_GID != 0, + Size: stat.Mask&linux.STATX_SIZE != 0, + ATime: stat.Mask&linux.STATX_ATIME != 0, + MTime: stat.Mask&linux.STATX_MTIME != 0, + ATimeNotSystemTime: stat.Atime.Nsec != linux.UTIME_NOW, + MTimeNotSystemTime: stat.Mtime.Nsec != linux.UTIME_NOW, + }, p9.SetAttr{ + Permissions: p9.FileMode(stat.Mode), + UID: p9.UID(stat.UID), + GID: p9.GID(stat.GID), + Size: stat.Size, + ATimeSeconds: uint64(stat.Atime.Sec), + ATimeNanoSeconds: uint64(stat.Atime.Nsec), + MTimeSeconds: uint64(stat.Mtime.Sec), + MTimeNanoSeconds: uint64(stat.Mtime.Nsec), + }); err != nil { + return err + } + } + if d.fs.opts.interop == InteropModeShared { + // There's no point to updating d's metadata in this case since it'll + // be overwritten by revalidation before the next time it's used + // anyway. (InteropModeShared inhibits client caching of regular file + // data, so there's no cache to truncate either.) + return nil + } + now, haveNow := nowFromContext(ctx) + if !haveNow { + ctx.Warningf("gofer.dentry.setStat: current time not available") + } + if stat.Mask&linux.STATX_MODE != 0 { + atomic.StoreUint32(&d.mode, d.fileType()|uint32(stat.Mode)) + } + if stat.Mask&linux.STATX_UID != 0 { + atomic.StoreUint32(&d.uid, stat.UID) + } + if stat.Mask&linux.STATX_GID != 0 { + atomic.StoreUint32(&d.gid, stat.GID) + } + if setLocalAtime { + if stat.Atime.Nsec == linux.UTIME_NOW { + if haveNow { + atomic.StoreInt64(&d.atime, now) + } + } else { + atomic.StoreInt64(&d.atime, dentryTimestampFromStatx(stat.Atime)) + } + } + if setLocalMtime { + if stat.Mtime.Nsec == linux.UTIME_NOW { + if haveNow { + atomic.StoreInt64(&d.mtime, now) + } + } else { + atomic.StoreInt64(&d.mtime, dentryTimestampFromStatx(stat.Mtime)) + } + } + if haveNow { + atomic.StoreInt64(&d.ctime, now) + } + if stat.Mask&linux.STATX_SIZE != 0 { + d.dataMu.Lock() + oldSize := d.size + d.size = stat.Size + // d.dataMu must be unlocked to lock d.mapsMu and invalidate mappings + // below. This allows concurrent calls to Read/Translate/etc. These + // functions synchronize with truncation by refusing to use cache + // contents beyond the new d.size. (We are still holding d.metadataMu, + // so we can't race with Write or another truncate.) + d.dataMu.Unlock() + if d.size < oldSize { + oldpgend := pageRoundUp(oldSize) + newpgend := pageRoundUp(d.size) + if oldpgend != newpgend { + d.mapsMu.Lock() + d.mappings.Invalidate(memmap.MappableRange{newpgend, oldpgend}, memmap.InvalidateOpts{ + // Compare Linux's mm/truncate.c:truncate_setsize() => + // truncate_pagecache() => + // mm/memory.c:unmap_mapping_range(evencows=1). + InvalidatePrivate: true, + }) + d.mapsMu.Unlock() + } + // We are now guaranteed that there are no translations of + // truncated pages, and can remove them from the cache. Since + // truncated pages have been removed from the remote file, they + // should be dropped without being written back. + d.dataMu.Lock() + d.cache.Truncate(d.size, d.fs.mfp.MemoryFile()) + d.dirty.KeepClean(memmap.MappableRange{d.size, oldpgend}) + d.dataMu.Unlock() + } + } + return nil +} + +func (d *dentry) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes, isDir bool) error { + return vfs.GenericCheckPermissions(creds, ats, isDir, uint16(atomic.LoadUint32(&d.mode))&0777, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))) +} + +// IncRef implements vfs.DentryImpl.IncRef. +func (d *dentry) IncRef() { + // d.refs may be 0 if d.fs.renameMu is locked, which serializes against + // d.checkCachingLocked(). + atomic.AddInt64(&d.refs, 1) +} + +// TryIncRef implements vfs.DentryImpl.TryIncRef. +func (d *dentry) TryIncRef() bool { + for { + refs := atomic.LoadInt64(&d.refs) + if refs == 0 { + return false + } + if atomic.CompareAndSwapInt64(&d.refs, refs, refs+1) { + return true + } + } +} + +// DecRef implements vfs.DentryImpl.DecRef. +func (d *dentry) DecRef() { + if refs := atomic.AddInt64(&d.refs, -1); refs == 0 { + d.fs.renameMu.Lock() + d.checkCachingLocked() + d.fs.renameMu.Unlock() + } else if refs < 0 { + panic("gofer.dentry.DecRef() called without holding a reference") + } +} + +// checkCachingLocked should be called after d's reference count becomes 0 or it +// becomes disowned. +// +// Preconditions: d.fs.renameMu must be locked for writing. +func (d *dentry) checkCachingLocked() { + // Dentries with a non-zero reference count must be retained. (The only way + // to obtain a reference on a dentry with zero references is via path + // resolution, which requires renameMu, so if d.refs is zero then it will + // remain zero while we hold renameMu for writing.) + if atomic.LoadInt64(&d.refs) != 0 { + if d.cached { + d.fs.cachedDentries.Remove(d) + d.fs.cachedDentriesLen-- + d.cached = false + } + return + } + // Non-child dentries with zero references are no longer reachable by path + // resolution and should be dropped immediately. + if d.vfsd.Parent() == nil || d.vfsd.IsDisowned() { + if d.cached { + d.fs.cachedDentries.Remove(d) + d.fs.cachedDentriesLen-- + d.cached = false + } + d.destroyLocked() + return + } + // If d is already cached, just move it to the front of the LRU. + if d.cached { + d.fs.cachedDentries.Remove(d) + d.fs.cachedDentries.PushFront(d) + return + } + // Cache the dentry, then evict the least recently used cached dentry if + // the cache becomes over-full. + d.fs.cachedDentries.PushFront(d) + d.fs.cachedDentriesLen++ + d.cached = true + if d.fs.cachedDentriesLen > d.fs.opts.maxCachedDentries { + victim := d.fs.cachedDentries.Back() + d.fs.cachedDentries.Remove(victim) + d.fs.cachedDentriesLen-- + victim.cached = false + // victim.refs may have become non-zero from an earlier path + // resolution since it was inserted into fs.cachedDentries; see + // dentry.incRefLocked(). Either way, we brought + // fs.cachedDentriesLen back down to fs.opts.maxCachedDentries, so + // we don't loop. + if atomic.LoadInt64(&victim.refs) == 0 { + if victimParentVFSD := victim.vfsd.Parent(); victimParentVFSD != nil { + victimParent := victimParentVFSD.Impl().(*dentry) + victimParent.dirMu.Lock() + if !victim.vfsd.IsDisowned() { + // victim can't be a mount point (in any mount + // namespace), since VFS holds references on mount + // points. + d.fs.vfsfs.VirtualFilesystem().ForceDeleteDentry(&victim.vfsd) + // We're only deleting the dentry, not the file it + // represents, so we don't need to update + // victimParent.dirents etc. + } + victimParent.dirMu.Unlock() + } + victim.destroyLocked() + } + } +} + +// Preconditions: d.fs.renameMu must be locked for writing. d.refs == 0. d is +// not a child dentry. +func (d *dentry) destroyLocked() { + ctx := context.Background() + d.handleMu.Lock() + if !d.handle.file.isNil() { + mf := d.fs.mfp.MemoryFile() + d.dataMu.Lock() + // Write dirty pages back to the remote filesystem. + if d.handleWritable { + if err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, mf, d.handle.writeFromBlocksAt); err != nil { + log.Warningf("gofer.dentry.DecRef: failed to write dirty data back: %v", err) + } + } + // Discard cached data. + d.cache.DropAll(mf) + d.dirty.RemoveAll() + d.dataMu.Unlock() + // Clunk open fids and close open host FDs. + d.handle.close(ctx) + } + d.handleMu.Unlock() + d.file.close(ctx) + // Remove d from the set of all dentries. + d.fs.syncMu.Lock() + delete(d.fs.dentries, d) + d.fs.syncMu.Unlock() + // Drop the reference held by d on its parent. + if parentVFSD := d.vfsd.Parent(); parentVFSD != nil { + parent := parentVFSD.Impl().(*dentry) + // This is parent.DecRef() without recursive locking of d.fs.renameMu. + if refs := atomic.AddInt64(&parent.refs, -1); refs == 0 { + parent.checkCachingLocked() + } else if refs < 0 { + panic("gofer.dentry.DecRef() called without holding a reference") + } + } +} + +func (d *dentry) isDeleted() bool { + return atomic.LoadUint32(&d.deleted) != 0 +} + +func (d *dentry) setDeleted() { + atomic.StoreUint32(&d.deleted, 1) +} + +func (d *dentry) listxattr(ctx context.Context) ([]string, error) { + return nil, syserror.ENOTSUP +} + +func (d *dentry) getxattr(ctx context.Context, name string) (string, error) { + // TODO(jamieliu): add vfs.GetxattrOptions.Size + return d.file.getXattr(ctx, name, linux.XATTR_SIZE_MAX) +} + +func (d *dentry) setxattr(ctx context.Context, opts *vfs.SetxattrOptions) error { + return d.file.setXattr(ctx, opts.Name, opts.Value, opts.Flags) +} + +func (d *dentry) removexattr(ctx context.Context, name string) error { + return syserror.ENOTSUP +} + +// Preconditions: d.isRegularFile() || d.isDirectory(). +func (d *dentry) ensureSharedHandle(ctx context.Context, read, write, trunc bool) error { + // O_TRUNC unconditionally requires us to obtain a new handle (opened with + // O_TRUNC). + if !trunc { + d.handleMu.RLock() + if (!read || d.handleReadable) && (!write || d.handleWritable) { + // The current handle is sufficient. + d.handleMu.RUnlock() + return nil + } + d.handleMu.RUnlock() + } + + haveOldFD := false + d.handleMu.Lock() + if (read && !d.handleReadable) || (write && !d.handleWritable) || trunc { + // Get a new handle. + wantReadable := d.handleReadable || read + wantWritable := d.handleWritable || write + h, err := openHandle(ctx, d.file, wantReadable, wantWritable, trunc) + if err != nil { + d.handleMu.Unlock() + return err + } + if !d.handle.file.isNil() { + // Check that old and new handles are compatible: If the old handle + // includes a host file descriptor but the new one does not, or + // vice versa, old and new memory mappings may be incoherent. + haveOldFD = d.handle.fd >= 0 + haveNewFD := h.fd >= 0 + if haveOldFD != haveNewFD { + d.handleMu.Unlock() + ctx.Warningf("gofer.dentry.ensureSharedHandle: can't change host FD availability from %v to %v across dentry handle upgrade", haveOldFD, haveNewFD) + h.close(ctx) + return syserror.EIO + } + if haveOldFD { + // We may have raced with callers of d.pf.FD() that are now + // using the old file descriptor, preventing us from safely + // closing it. We could handle this by invalidating existing + // memmap.Translations, but this is expensive. Instead, use + // dup2() to make the old file descriptor refer to the new file + // description, then close the new file descriptor (which is no + // longer needed). Racing callers may use the old or new file + // description, but this doesn't matter since they refer to the + // same file (unless d.fs.opts.overlayfsStaleRead is true, + // which we handle separately). + if err := syscall.Dup2(int(h.fd), int(d.handle.fd)); err != nil { + d.handleMu.Unlock() + ctx.Warningf("gofer.dentry.ensureSharedHandle: failed to dup fd %d to fd %d: %v", h.fd, d.handle.fd, err) + h.close(ctx) + return err + } + syscall.Close(int(h.fd)) + h.fd = d.handle.fd + if d.fs.opts.overlayfsStaleRead { + // Replace sentry mappings of the old FD with mappings of + // the new FD, since the two are not necessarily coherent. + if err := d.pf.hostFileMapper.RegenerateMappings(int(h.fd)); err != nil { + d.handleMu.Unlock() + ctx.Warningf("gofer.dentry.ensureSharedHandle: failed to replace sentry mappings of old FD with mappings of new FD: %v", err) + h.close(ctx) + return err + } + } + // Clunk the old fid before making the new handle visible (by + // unlocking d.handleMu). + d.handle.file.close(ctx) + } + } + // Switch to the new handle. + d.handle = h + d.handleReadable = wantReadable + d.handleWritable = wantWritable + } + d.handleMu.Unlock() + + if d.fs.opts.overlayfsStaleRead && haveOldFD { + // Invalidate application mappings that may be using the old FD; they + // will be replaced with mappings using the new FD after future calls + // to d.Translate(). This requires holding d.mapsMu, which precedes + // d.handleMu in the lock order. + d.mapsMu.Lock() + d.mappings.InvalidateAll(memmap.InvalidateOpts{}) + d.mapsMu.Unlock() + } + + return nil +} + +// fileDescription is embedded by gofer implementations of +// vfs.FileDescriptionImpl. +type fileDescription struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl +} + +func (fd *fileDescription) filesystem() *filesystem { + return fd.vfsfd.Mount().Filesystem().Impl().(*filesystem) +} + +func (fd *fileDescription) dentry() *dentry { + return fd.vfsfd.Dentry().Impl().(*dentry) +} + +// Stat implements vfs.FileDescriptionImpl.Stat. +func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { + d := fd.dentry() + if d.fs.opts.interop == InteropModeShared && opts.Mask&(linux.STATX_MODE|linux.STATX_UID|linux.STATX_GID|linux.STATX_ATIME|linux.STATX_MTIME|linux.STATX_CTIME|linux.STATX_SIZE|linux.STATX_BLOCKS|linux.STATX_BTIME) != 0 && opts.Sync != linux.AT_STATX_DONT_SYNC { + // TODO(jamieliu): Use specialFileFD.handle.file for the getattr if + // available? + if err := d.updateFromGetattr(ctx); err != nil { + return linux.Statx{}, err + } + } + var stat linux.Statx + d.statTo(&stat) + return stat, nil +} + +// SetStat implements vfs.FileDescriptionImpl.SetStat. +func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { + return fd.dentry().setStat(ctx, auth.CredentialsFromContext(ctx), &opts.Stat, fd.vfsfd.Mount()) +} + +// Listxattr implements vfs.FileDescriptionImpl.Listxattr. +func (fd *fileDescription) Listxattr(ctx context.Context) ([]string, error) { + return fd.dentry().listxattr(ctx) +} + +// Getxattr implements vfs.FileDescriptionImpl.Getxattr. +func (fd *fileDescription) Getxattr(ctx context.Context, name string) (string, error) { + return fd.dentry().getxattr(ctx, name) +} + +// Setxattr implements vfs.FileDescriptionImpl.Setxattr. +func (fd *fileDescription) Setxattr(ctx context.Context, opts vfs.SetxattrOptions) error { + return fd.dentry().setxattr(ctx, &opts) +} + +// Removexattr implements vfs.FileDescriptionImpl.Removexattr. +func (fd *fileDescription) Removexattr(ctx context.Context, name string) error { + return fd.dentry().removexattr(ctx, name) +} diff --git a/pkg/sentry/fsimpl/gofer/handle.go b/pkg/sentry/fsimpl/gofer/handle.go new file mode 100644 index 000000000..cfe66f797 --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/handle.go @@ -0,0 +1,135 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package gofer + +import ( + "syscall" + + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/p9" + "gvisor.dev/gvisor/pkg/safemem" +) + +// handle represents a remote "open file descriptor", consisting of an opened +// fid (p9.File) and optionally a host file descriptor. +type handle struct { + file p9file + fd int32 // -1 if unavailable +} + +// Preconditions: read || write. +func openHandle(ctx context.Context, file p9file, read, write, trunc bool) (handle, error) { + _, newfile, err := file.walk(ctx, nil) + if err != nil { + return handle{fd: -1}, err + } + var flags p9.OpenFlags + switch { + case read && !write: + flags = p9.ReadOnly + case !read && write: + flags = p9.WriteOnly + case read && write: + flags = p9.ReadWrite + } + if trunc { + flags |= p9.OpenTruncate + } + fdobj, _, _, err := newfile.open(ctx, flags) + if err != nil { + newfile.close(ctx) + return handle{fd: -1}, err + } + fd := int32(-1) + if fdobj != nil { + fd = int32(fdobj.Release()) + } + return handle{ + file: newfile, + fd: fd, + }, nil +} + +func (h *handle) close(ctx context.Context) { + h.file.close(ctx) + h.file = p9file{} + if h.fd >= 0 { + syscall.Close(int(h.fd)) + h.fd = -1 + } +} + +func (h *handle) readToBlocksAt(ctx context.Context, dsts safemem.BlockSeq, offset uint64) (uint64, error) { + if dsts.IsEmpty() { + return 0, nil + } + if h.fd >= 0 { + ctx.UninterruptibleSleepStart(false) + n, err := hostPreadv(h.fd, dsts, int64(offset)) + ctx.UninterruptibleSleepFinish(false) + return n, err + } + if dsts.NumBlocks() == 1 && !dsts.Head().NeedSafecopy() { + n, err := h.file.readAt(ctx, dsts.Head().ToSlice(), offset) + return uint64(n), err + } + // Buffer the read since p9.File.ReadAt() takes []byte. + buf := make([]byte, dsts.NumBytes()) + n, err := h.file.readAt(ctx, buf, offset) + if n == 0 { + return 0, err + } + if cp, cperr := safemem.CopySeq(dsts, safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf[:n]))); cperr != nil { + return cp, cperr + } + return uint64(n), err +} + +func (h *handle) writeFromBlocksAt(ctx context.Context, srcs safemem.BlockSeq, offset uint64) (uint64, error) { + if srcs.IsEmpty() { + return 0, nil + } + if h.fd >= 0 { + ctx.UninterruptibleSleepStart(false) + n, err := hostPwritev(h.fd, srcs, int64(offset)) + ctx.UninterruptibleSleepFinish(false) + return n, err + } + if srcs.NumBlocks() == 1 && !srcs.Head().NeedSafecopy() { + n, err := h.file.writeAt(ctx, srcs.Head().ToSlice(), offset) + return uint64(n), err + } + // Buffer the write since p9.File.WriteAt() takes []byte. + buf := make([]byte, srcs.NumBytes()) + cp, cperr := safemem.CopySeq(safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf)), srcs) + if cp == 0 { + return 0, cperr + } + n, err := h.file.writeAt(ctx, buf[:cp], offset) + if err != nil { + return uint64(n), err + } + return cp, cperr +} + +func (h *handle) sync(ctx context.Context) error { + if h.fd >= 0 { + ctx.UninterruptibleSleepStart(false) + err := syscall.Fsync(int(h.fd)) + ctx.UninterruptibleSleepFinish(false) + return err + } + return h.file.fsync(ctx) +} diff --git a/pkg/sentry/fsimpl/gofer/handle_unsafe.go b/pkg/sentry/fsimpl/gofer/handle_unsafe.go new file mode 100644 index 000000000..19560ab26 --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/handle_unsafe.go @@ -0,0 +1,66 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package gofer + +import ( + "syscall" + "unsafe" + + "gvisor.dev/gvisor/pkg/safemem" +) + +// Preconditions: !dsts.IsEmpty(). +func hostPreadv(fd int32, dsts safemem.BlockSeq, off int64) (uint64, error) { + // No buffering is necessary regardless of safecopy; host syscalls will + // return EFAULT if appropriate, instead of raising SIGBUS. + if dsts.NumBlocks() == 1 { + // Use pread() instead of preadv() to avoid iovec allocation and + // copying. + dst := dsts.Head() + n, _, e := syscall.Syscall6(syscall.SYS_PREAD64, uintptr(fd), dst.Addr(), uintptr(dst.Len()), uintptr(off), 0, 0) + if e != 0 { + return 0, e + } + return uint64(n), nil + } + iovs := safemem.IovecsFromBlockSeq(dsts) + n, _, e := syscall.Syscall6(syscall.SYS_PREADV, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs)), uintptr(off), 0, 0) + if e != 0 { + return 0, e + } + return uint64(n), nil +} + +// Preconditions: !srcs.IsEmpty(). +func hostPwritev(fd int32, srcs safemem.BlockSeq, off int64) (uint64, error) { + // No buffering is necessary regardless of safecopy; host syscalls will + // return EFAULT if appropriate, instead of raising SIGBUS. + if srcs.NumBlocks() == 1 { + // Use pwrite() instead of pwritev() to avoid iovec allocation and + // copying. + src := srcs.Head() + n, _, e := syscall.Syscall6(syscall.SYS_PWRITE64, uintptr(fd), src.Addr(), uintptr(src.Len()), uintptr(off), 0, 0) + if e != 0 { + return 0, e + } + return uint64(n), nil + } + iovs := safemem.IovecsFromBlockSeq(srcs) + n, _, e := syscall.Syscall6(syscall.SYS_PWRITEV, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs)), uintptr(off), 0, 0) + if e != 0 { + return 0, e + } + return uint64(n), nil +} diff --git a/pkg/sentry/fsimpl/gofer/p9file.go b/pkg/sentry/fsimpl/gofer/p9file.go new file mode 100644 index 000000000..755ac2985 --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/p9file.go @@ -0,0 +1,219 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package gofer + +import ( + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fd" + "gvisor.dev/gvisor/pkg/p9" + "gvisor.dev/gvisor/pkg/syserror" +) + +// p9file is a wrapper around p9.File that provides methods that are +// Context-aware. +type p9file struct { + file p9.File +} + +func (f p9file) isNil() bool { + return f.file == nil +} + +func (f p9file) walk(ctx context.Context, names []string) ([]p9.QID, p9file, error) { + ctx.UninterruptibleSleepStart(false) + qids, newfile, err := f.file.Walk(names) + ctx.UninterruptibleSleepFinish(false) + return qids, p9file{newfile}, err +} + +func (f p9file) walkGetAttr(ctx context.Context, names []string) ([]p9.QID, p9file, p9.AttrMask, p9.Attr, error) { + ctx.UninterruptibleSleepStart(false) + qids, newfile, attrMask, attr, err := f.file.WalkGetAttr(names) + ctx.UninterruptibleSleepFinish(false) + return qids, p9file{newfile}, attrMask, attr, err +} + +// walkGetAttrOne is a wrapper around p9.File.WalkGetAttr that takes a single +// path component and returns a single qid. +func (f p9file) walkGetAttrOne(ctx context.Context, name string) (p9.QID, p9file, p9.AttrMask, p9.Attr, error) { + ctx.UninterruptibleSleepStart(false) + qids, newfile, attrMask, attr, err := f.file.WalkGetAttr([]string{name}) + ctx.UninterruptibleSleepFinish(false) + if err != nil { + return p9.QID{}, p9file{}, p9.AttrMask{}, p9.Attr{}, err + } + if len(qids) != 1 { + ctx.Warningf("p9.File.WalkGetAttr returned %d qids (%v), wanted 1", len(qids), qids) + if newfile != nil { + p9file{newfile}.close(ctx) + } + return p9.QID{}, p9file{}, p9.AttrMask{}, p9.Attr{}, syserror.EIO + } + return qids[0], p9file{newfile}, attrMask, attr, nil +} + +func (f p9file) statFS(ctx context.Context) (p9.FSStat, error) { + ctx.UninterruptibleSleepStart(false) + fsstat, err := f.file.StatFS() + ctx.UninterruptibleSleepFinish(false) + return fsstat, err +} + +func (f p9file) getAttr(ctx context.Context, req p9.AttrMask) (p9.QID, p9.AttrMask, p9.Attr, error) { + ctx.UninterruptibleSleepStart(false) + qid, attrMask, attr, err := f.file.GetAttr(req) + ctx.UninterruptibleSleepFinish(false) + return qid, attrMask, attr, err +} + +func (f p9file) setAttr(ctx context.Context, valid p9.SetAttrMask, attr p9.SetAttr) error { + ctx.UninterruptibleSleepStart(false) + err := f.file.SetAttr(valid, attr) + ctx.UninterruptibleSleepFinish(false) + return err +} + +func (f p9file) getXattr(ctx context.Context, name string, size uint64) (string, error) { + ctx.UninterruptibleSleepStart(false) + val, err := f.file.GetXattr(name, size) + ctx.UninterruptibleSleepFinish(false) + return val, err +} + +func (f p9file) setXattr(ctx context.Context, name, value string, flags uint32) error { + ctx.UninterruptibleSleepStart(false) + err := f.file.SetXattr(name, value, flags) + ctx.UninterruptibleSleepFinish(false) + return err +} + +func (f p9file) allocate(ctx context.Context, mode p9.AllocateMode, offset, length uint64) error { + ctx.UninterruptibleSleepStart(false) + err := f.file.Allocate(mode, offset, length) + ctx.UninterruptibleSleepFinish(false) + return err +} + +func (f p9file) close(ctx context.Context) error { + ctx.UninterruptibleSleepStart(false) + err := f.file.Close() + ctx.UninterruptibleSleepFinish(false) + return err +} + +func (f p9file) open(ctx context.Context, flags p9.OpenFlags) (*fd.FD, p9.QID, uint32, error) { + ctx.UninterruptibleSleepStart(false) + fdobj, qid, iounit, err := f.file.Open(flags) + ctx.UninterruptibleSleepFinish(false) + return fdobj, qid, iounit, err +} + +func (f p9file) readAt(ctx context.Context, p []byte, offset uint64) (int, error) { + ctx.UninterruptibleSleepStart(false) + n, err := f.file.ReadAt(p, offset) + ctx.UninterruptibleSleepFinish(false) + return n, err +} + +func (f p9file) writeAt(ctx context.Context, p []byte, offset uint64) (int, error) { + ctx.UninterruptibleSleepStart(false) + n, err := f.file.WriteAt(p, offset) + ctx.UninterruptibleSleepFinish(false) + return n, err +} + +func (f p9file) fsync(ctx context.Context) error { + ctx.UninterruptibleSleepStart(false) + err := f.file.FSync() + ctx.UninterruptibleSleepFinish(false) + return err +} + +func (f p9file) create(ctx context.Context, name string, flags p9.OpenFlags, permissions p9.FileMode, uid p9.UID, gid p9.GID) (*fd.FD, p9file, p9.QID, uint32, error) { + ctx.UninterruptibleSleepStart(false) + fdobj, newfile, qid, iounit, err := f.file.Create(name, flags, permissions, uid, gid) + ctx.UninterruptibleSleepFinish(false) + return fdobj, p9file{newfile}, qid, iounit, err +} + +func (f p9file) mkdir(ctx context.Context, name string, permissions p9.FileMode, uid p9.UID, gid p9.GID) (p9.QID, error) { + ctx.UninterruptibleSleepStart(false) + qid, err := f.file.Mkdir(name, permissions, uid, gid) + ctx.UninterruptibleSleepFinish(false) + return qid, err +} + +func (f p9file) symlink(ctx context.Context, oldName string, newName string, uid p9.UID, gid p9.GID) (p9.QID, error) { + ctx.UninterruptibleSleepStart(false) + qid, err := f.file.Symlink(oldName, newName, uid, gid) + ctx.UninterruptibleSleepFinish(false) + return qid, err +} + +func (f p9file) link(ctx context.Context, target p9file, newName string) error { + ctx.UninterruptibleSleepStart(false) + err := f.file.Link(target.file, newName) + ctx.UninterruptibleSleepFinish(false) + return err +} + +func (f p9file) mknod(ctx context.Context, name string, mode p9.FileMode, major uint32, minor uint32, uid p9.UID, gid p9.GID) (p9.QID, error) { + ctx.UninterruptibleSleepStart(false) + qid, err := f.file.Mknod(name, mode, major, minor, uid, gid) + ctx.UninterruptibleSleepFinish(false) + return qid, err +} + +func (f p9file) rename(ctx context.Context, newDir p9file, newName string) error { + ctx.UninterruptibleSleepStart(false) + err := f.file.Rename(newDir.file, newName) + ctx.UninterruptibleSleepFinish(false) + return err +} + +func (f p9file) unlinkAt(ctx context.Context, name string, flags uint32) error { + ctx.UninterruptibleSleepStart(false) + err := f.file.UnlinkAt(name, flags) + ctx.UninterruptibleSleepFinish(false) + return err +} + +func (f p9file) readdir(ctx context.Context, offset uint64, count uint32) ([]p9.Dirent, error) { + ctx.UninterruptibleSleepStart(false) + dirents, err := f.file.Readdir(offset, count) + ctx.UninterruptibleSleepFinish(false) + return dirents, err +} + +func (f p9file) readlink(ctx context.Context) (string, error) { + ctx.UninterruptibleSleepStart(false) + target, err := f.file.Readlink() + ctx.UninterruptibleSleepFinish(false) + return target, err +} + +func (f p9file) flush(ctx context.Context) error { + ctx.UninterruptibleSleepStart(false) + err := f.file.Flush() + ctx.UninterruptibleSleepFinish(false) + return err +} + +func (f p9file) connect(ctx context.Context, flags p9.ConnectFlags) (*fd.FD, error) { + ctx.UninterruptibleSleepStart(false) + fdobj, err := f.file.Connect(flags) + ctx.UninterruptibleSleepFinish(false) + return fdobj, err +} diff --git a/pkg/sentry/fsimpl/gofer/pagemath.go b/pkg/sentry/fsimpl/gofer/pagemath.go new file mode 100644 index 000000000..847cb0784 --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/pagemath.go @@ -0,0 +1,31 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package gofer + +import ( + "gvisor.dev/gvisor/pkg/usermem" +) + +// This are equivalent to usermem.Addr.RoundDown/Up, but without the +// potentially truncating conversion to usermem.Addr. This is necessary because +// there is no way to define generic "PageRoundDown/Up" functions in Go. + +func pageRoundDown(x uint64) uint64 { + return x &^ (usermem.PageSize - 1) +} + +func pageRoundUp(x uint64) uint64 { + return pageRoundDown(x + usermem.PageSize - 1) +} diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go new file mode 100644 index 000000000..8e11e06b3 --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/regular_file.go @@ -0,0 +1,860 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package gofer + +import ( + "fmt" + "io" + "math" + "sync" + "sync/atomic" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/safemem" + "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" + "gvisor.dev/gvisor/pkg/sentry/memmap" + "gvisor.dev/gvisor/pkg/sentry/pgalloc" + "gvisor.dev/gvisor/pkg/sentry/platform" + "gvisor.dev/gvisor/pkg/sentry/usage" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +func (d *dentry) isRegularFile() bool { + return d.fileType() == linux.S_IFREG +} + +type regularFileFD struct { + fileDescription + + // off is the file offset. off is protected by mu. + mu sync.Mutex + off int64 +} + +// Release implements vfs.FileDescriptionImpl.Release. +func (fd *regularFileFD) Release() { +} + +// OnClose implements vfs.FileDescriptionImpl.OnClose. +func (fd *regularFileFD) OnClose(ctx context.Context) error { + if !fd.vfsfd.IsWritable() { + return nil + } + // Skip flushing if writes may be buffered by the client, since (as with + // the VFS1 client) we don't flush buffered writes on close anyway. + d := fd.dentry() + if d.fs.opts.interop == InteropModeExclusive { + return nil + } + d.handleMu.RLock() + defer d.handleMu.RUnlock() + return d.handle.file.flush(ctx) +} + +// PRead implements vfs.FileDescriptionImpl.PRead. +func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + if offset < 0 { + return 0, syserror.EINVAL + } + if opts.Flags != 0 { + return 0, syserror.EOPNOTSUPP + } + + // Check for reading at EOF before calling into MM (but not under + // InteropModeShared, which makes d.size unreliable). + d := fd.dentry() + if d.fs.opts.interop != InteropModeShared && uint64(offset) >= atomic.LoadUint64(&d.size) { + return 0, io.EOF + } + + if fd.vfsfd.StatusFlags()&linux.O_DIRECT != 0 { + // Lock d.metadataMu for the rest of the read to prevent d.size from + // changing. + d.metadataMu.Lock() + defer d.metadataMu.Unlock() + // Write dirty cached pages that will be touched by the read back to + // the remote file. + if err := d.writeback(ctx, offset, dst.NumBytes()); err != nil { + return 0, err + } + } + + rw := getDentryReadWriter(ctx, d, offset) + if fd.vfsfd.StatusFlags()&linux.O_DIRECT != 0 { + // Require the read to go to the remote file. + rw.direct = true + } + n, err := dst.CopyOutFrom(ctx, rw) + putDentryReadWriter(rw) + if d.fs.opts.interop != InteropModeShared { + // Compare Linux's mm/filemap.c:do_generic_file_read() => file_accessed(). + d.touchAtime(ctx, fd.vfsfd.Mount()) + } + return n, err +} + +// Read implements vfs.FileDescriptionImpl.Read. +func (fd *regularFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + fd.mu.Lock() + n, err := fd.PRead(ctx, dst, fd.off, opts) + fd.off += n + fd.mu.Unlock() + return n, err +} + +// PWrite implements vfs.FileDescriptionImpl.PWrite. +func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + if offset < 0 { + return 0, syserror.EINVAL + } + if opts.Flags != 0 { + return 0, syserror.EOPNOTSUPP + } + + d := fd.dentry() + d.metadataMu.Lock() + defer d.metadataMu.Unlock() + if d.fs.opts.interop != InteropModeShared { + // Compare Linux's mm/filemap.c:__generic_file_write_iter() => + // file_update_time(). This is d.touchCMtime(), but without locking + // d.metadataMu (recursively). + if now, ok := nowFromContext(ctx); ok { + atomic.StoreInt64(&d.mtime, now) + atomic.StoreInt64(&d.ctime, now) + } + } + if fd.vfsfd.StatusFlags()&linux.O_DIRECT != 0 { + // Write dirty cached pages that will be touched by the write back to + // the remote file. + if err := d.writeback(ctx, offset, src.NumBytes()); err != nil { + return 0, err + } + // Remove touched pages from the cache. + pgstart := pageRoundDown(uint64(offset)) + pgend := pageRoundUp(uint64(offset + src.NumBytes())) + if pgend < pgstart { + return 0, syserror.EINVAL + } + mr := memmap.MappableRange{pgstart, pgend} + var freed []platform.FileRange + d.dataMu.Lock() + cseg := d.cache.LowerBoundSegment(mr.Start) + for cseg.Ok() && cseg.Start() < mr.End { + cseg = d.cache.Isolate(cseg, mr) + freed = append(freed, platform.FileRange{cseg.Value(), cseg.Value() + cseg.Range().Length()}) + cseg = d.cache.Remove(cseg).NextSegment() + } + d.dataMu.Unlock() + // Invalidate mappings of removed pages. + d.mapsMu.Lock() + d.mappings.Invalidate(mr, memmap.InvalidateOpts{}) + d.mapsMu.Unlock() + // Finally free pages removed from the cache. + mf := d.fs.mfp.MemoryFile() + for _, freedFR := range freed { + mf.DecRef(freedFR) + } + } + rw := getDentryReadWriter(ctx, d, offset) + if fd.vfsfd.StatusFlags()&linux.O_DIRECT != 0 { + // Require the write to go to the remote file. + rw.direct = true + } + n, err := src.CopyInTo(ctx, rw) + putDentryReadWriter(rw) + if n != 0 && fd.vfsfd.StatusFlags()&(linux.O_DSYNC|linux.O_SYNC) != 0 { + // Write dirty cached pages touched by the write back to the remote + // file. + if err := d.writeback(ctx, offset, src.NumBytes()); err != nil { + return 0, err + } + // Request the remote filesystem to sync the remote file. + if err := d.handle.file.fsync(ctx); err != nil { + return 0, err + } + } + return n, err +} + +// Write implements vfs.FileDescriptionImpl.Write. +func (fd *regularFileFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + fd.mu.Lock() + n, err := fd.PWrite(ctx, src, fd.off, opts) + fd.off += n + fd.mu.Unlock() + return n, err +} + +type dentryReadWriter struct { + ctx context.Context + d *dentry + off uint64 + direct bool +} + +var dentryReadWriterPool = sync.Pool{ + New: func() interface{} { + return &dentryReadWriter{} + }, +} + +func getDentryReadWriter(ctx context.Context, d *dentry, offset int64) *dentryReadWriter { + rw := dentryReadWriterPool.Get().(*dentryReadWriter) + rw.ctx = ctx + rw.d = d + rw.off = uint64(offset) + rw.direct = false + return rw +} + +func putDentryReadWriter(rw *dentryReadWriter) { + rw.ctx = nil + rw.d = nil + dentryReadWriterPool.Put(rw) +} + +// ReadToBlocks implements safemem.Reader.ReadToBlocks. +func (rw *dentryReadWriter) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) { + if dsts.IsEmpty() { + return 0, nil + } + + // If we have a mmappable host FD (which must be used here to ensure + // coherence with memory-mapped I/O), or if InteropModeShared is in effect + // (which prevents us from caching file contents and makes dentry.size + // unreliable), or if the file was opened O_DIRECT, read directly from + // dentry.handle without locking dentry.dataMu. + rw.d.handleMu.RLock() + if (rw.d.handle.fd >= 0 && !rw.d.fs.opts.forcePageCache) || rw.d.fs.opts.interop == InteropModeShared || rw.direct { + n, err := rw.d.handle.readToBlocksAt(rw.ctx, dsts, rw.off) + rw.d.handleMu.RUnlock() + rw.off += n + return n, err + } + + // Otherwise read from/through the cache. + mf := rw.d.fs.mfp.MemoryFile() + fillCache := mf.ShouldCacheEvictable() + var dataMuUnlock func() + if fillCache { + rw.d.dataMu.Lock() + dataMuUnlock = rw.d.dataMu.Unlock + } else { + rw.d.dataMu.RLock() + dataMuUnlock = rw.d.dataMu.RUnlock + } + + // Compute the range to read (limited by file size and overflow-checked). + if rw.off >= rw.d.size { + dataMuUnlock() + rw.d.handleMu.RUnlock() + return 0, io.EOF + } + end := rw.d.size + if rend := rw.off + dsts.NumBytes(); rend > rw.off && rend < end { + end = rend + } + + var done uint64 + seg, gap := rw.d.cache.Find(rw.off) + for rw.off < end { + mr := memmap.MappableRange{rw.off, end} + switch { + case seg.Ok(): + // Get internal mappings from the cache. + ims, err := mf.MapInternal(seg.FileRangeOf(seg.Range().Intersect(mr)), usermem.Read) + if err != nil { + dataMuUnlock() + rw.d.handleMu.RUnlock() + return done, err + } + + // Copy from internal mappings. + n, err := safemem.CopySeq(dsts, ims) + done += n + rw.off += n + dsts = dsts.DropFirst64(n) + if err != nil { + dataMuUnlock() + rw.d.handleMu.RUnlock() + return done, err + } + + // Continue. + seg, gap = seg.NextNonEmpty() + + case gap.Ok(): + gapMR := gap.Range().Intersect(mr) + if fillCache { + // Read into the cache, then re-enter the loop to read from the + // cache. + reqMR := memmap.MappableRange{ + Start: pageRoundDown(gapMR.Start), + End: pageRoundUp(gapMR.End), + } + optMR := gap.Range() + err := rw.d.cache.Fill(rw.ctx, reqMR, maxFillRange(reqMR, optMR), mf, usage.PageCache, rw.d.handle.readToBlocksAt) + mf.MarkEvictable(rw.d, pgalloc.EvictableRange{optMR.Start, optMR.End}) + seg, gap = rw.d.cache.Find(rw.off) + if !seg.Ok() { + dataMuUnlock() + rw.d.handleMu.RUnlock() + return done, err + } + // err might have occurred in part of gap.Range() outside + // gapMR. Forget about it for now; if the error matters and + // persists, we'll run into it again in a later iteration of + // this loop. + } else { + // Read directly from the file. + gapDsts := dsts.TakeFirst64(gapMR.Length()) + n, err := rw.d.handle.readToBlocksAt(rw.ctx, gapDsts, gapMR.Start) + done += n + rw.off += n + dsts = dsts.DropFirst64(n) + // Partial reads are fine. But we must stop reading. + if n != gapDsts.NumBytes() || err != nil { + dataMuUnlock() + rw.d.handleMu.RUnlock() + return done, err + } + + // Continue. + seg, gap = gap.NextSegment(), fsutil.FileRangeGapIterator{} + } + } + } + dataMuUnlock() + rw.d.handleMu.RUnlock() + return done, nil +} + +// WriteFromBlocks implements safemem.Writer.WriteFromBlocks. +// +// Preconditions: rw.d.metadataMu must be locked. +func (rw *dentryReadWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) { + if srcs.IsEmpty() { + return 0, nil + } + + // If we have a mmappable host FD (which must be used here to ensure + // coherence with memory-mapped I/O), or if InteropModeShared is in effect + // (which prevents us from caching file contents), or if the file was + // opened with O_DIRECT, write directly to dentry.handle without locking + // dentry.dataMu. + rw.d.handleMu.RLock() + if (rw.d.handle.fd >= 0 && !rw.d.fs.opts.forcePageCache) || rw.d.fs.opts.interop == InteropModeShared || rw.direct { + n, err := rw.d.handle.writeFromBlocksAt(rw.ctx, srcs, rw.off) + rw.d.handleMu.RUnlock() + rw.off += n + return n, err + } + + // Otherwise write to/through the cache. + mf := rw.d.fs.mfp.MemoryFile() + rw.d.dataMu.Lock() + + // Compute the range to write (overflow-checked). + start := rw.off + end := rw.off + srcs.NumBytes() + if end <= rw.off { + end = math.MaxInt64 + } + + var ( + done uint64 + retErr error + ) + seg, gap := rw.d.cache.Find(rw.off) + for rw.off < end { + mr := memmap.MappableRange{rw.off, end} + switch { + case seg.Ok(): + // Get internal mappings from the cache. + segMR := seg.Range().Intersect(mr) + ims, err := mf.MapInternal(seg.FileRangeOf(segMR), usermem.Write) + if err != nil { + retErr = err + goto exitLoop + } + + // Copy to internal mappings. + n, err := safemem.CopySeq(ims, srcs) + done += n + rw.off += n + srcs = srcs.DropFirst64(n) + rw.d.dirty.MarkDirty(segMR) + if err != nil { + retErr = err + goto exitLoop + } + + // Continue. + seg, gap = seg.NextNonEmpty() + + case gap.Ok(): + // Write directly to the file. At present, we never fill the cache + // when writing, since doing so can convert small writes into + // inefficient read-modify-write cycles, and we have no mechanism + // for detecting or avoiding this. + gapMR := gap.Range().Intersect(mr) + gapSrcs := srcs.TakeFirst64(gapMR.Length()) + n, err := rw.d.handle.writeFromBlocksAt(rw.ctx, gapSrcs, gapMR.Start) + done += n + rw.off += n + srcs = srcs.DropFirst64(n) + // Partial writes are fine. But we must stop writing. + if n != gapSrcs.NumBytes() || err != nil { + retErr = err + goto exitLoop + } + + // Continue. + seg, gap = gap.NextSegment(), fsutil.FileRangeGapIterator{} + } + } +exitLoop: + if rw.off > rw.d.size { + atomic.StoreUint64(&rw.d.size, rw.off) + // The remote file's size will implicitly be extended to the correct + // value when we write back to it. + } + // If InteropModeWritethrough is in effect, flush written data back to the + // remote filesystem. + if rw.d.fs.opts.interop == InteropModeWritethrough && done != 0 { + if err := fsutil.SyncDirty(rw.ctx, memmap.MappableRange{ + Start: start, + End: rw.off, + }, &rw.d.cache, &rw.d.dirty, rw.d.size, mf, rw.d.handle.writeFromBlocksAt); err != nil { + // We have no idea how many bytes were actually flushed. + rw.off = start + done = 0 + retErr = err + } + } + rw.d.dataMu.Unlock() + rw.d.handleMu.RUnlock() + return done, retErr +} + +func (d *dentry) writeback(ctx context.Context, offset, size int64) error { + if size == 0 { + return nil + } + d.handleMu.RLock() + defer d.handleMu.RUnlock() + d.dataMu.Lock() + defer d.dataMu.Unlock() + // Compute the range of valid bytes (overflow-checked). + if uint64(offset) >= d.size { + return nil + } + end := int64(d.size) + if rend := offset + size; rend > offset && rend < end { + end = rend + } + return fsutil.SyncDirty(ctx, memmap.MappableRange{ + Start: uint64(offset), + End: uint64(end), + }, &d.cache, &d.dirty, d.size, d.fs.mfp.MemoryFile(), d.handle.writeFromBlocksAt) +} + +// Seek implements vfs.FileDescriptionImpl.Seek. +func (fd *regularFileFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + fd.mu.Lock() + defer fd.mu.Unlock() + switch whence { + case linux.SEEK_SET: + // Use offset as specified. + case linux.SEEK_CUR: + offset += fd.off + case linux.SEEK_END, linux.SEEK_DATA, linux.SEEK_HOLE: + // Ensure file size is up to date. + d := fd.dentry() + if fd.filesystem().opts.interop == InteropModeShared { + if err := d.updateFromGetattr(ctx); err != nil { + return 0, err + } + } + size := int64(atomic.LoadUint64(&d.size)) + // For SEEK_DATA and SEEK_HOLE, treat the file as a single contiguous + // block of data. + switch whence { + case linux.SEEK_END: + offset += size + case linux.SEEK_DATA: + if offset > size { + return 0, syserror.ENXIO + } + // Use offset as specified. + case linux.SEEK_HOLE: + if offset > size { + return 0, syserror.ENXIO + } + offset = size + } + default: + return 0, syserror.EINVAL + } + if offset < 0 { + return 0, syserror.EINVAL + } + fd.off = offset + return offset, nil +} + +// Sync implements vfs.FileDescriptionImpl.Sync. +func (fd *regularFileFD) Sync(ctx context.Context) error { + return fd.dentry().syncSharedHandle(ctx) +} + +func (d *dentry) syncSharedHandle(ctx context.Context) error { + d.handleMu.RLock() + if !d.handleWritable { + d.handleMu.RUnlock() + return nil + } + d.dataMu.Lock() + // Write dirty cached data to the remote file. + err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, d.fs.mfp.MemoryFile(), d.handle.writeFromBlocksAt) + d.dataMu.Unlock() + if err == nil { + // Sync the remote file. + err = d.handle.sync(ctx) + } + d.handleMu.RUnlock() + return err +} + +// ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. +func (fd *regularFileFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { + d := fd.dentry() + switch d.fs.opts.interop { + case InteropModeExclusive: + // Any mapping is fine. + case InteropModeWritethrough: + // Shared writable mappings require a host FD, since otherwise we can't + // synchronously flush memory-mapped writes to the remote file. + if opts.Private || !opts.MaxPerms.Write { + break + } + fallthrough + case InteropModeShared: + // All mappings require a host FD to be coherent with other filesystem + // users. + if d.fs.opts.forcePageCache { + // Whether or not we have a host FD, we're not allowed to use it. + return syserror.ENODEV + } + d.handleMu.RLock() + haveFD := d.handle.fd >= 0 + d.handleMu.RUnlock() + if !haveFD { + return syserror.ENODEV + } + default: + panic(fmt.Sprintf("unknown InteropMode %v", d.fs.opts.interop)) + } + return vfs.GenericConfigureMMap(&fd.vfsfd, d, opts) +} + +func (d *dentry) mayCachePages() bool { + if d.fs.opts.interop == InteropModeShared { + return false + } + if d.fs.opts.forcePageCache { + return true + } + d.handleMu.RLock() + haveFD := d.handle.fd >= 0 + d.handleMu.RUnlock() + return haveFD +} + +// AddMapping implements memmap.Mappable.AddMapping. +func (d *dentry) AddMapping(ctx context.Context, ms memmap.MappingSpace, ar usermem.AddrRange, offset uint64, writable bool) error { + d.mapsMu.Lock() + mapped := d.mappings.AddMapping(ms, ar, offset, writable) + // Do this unconditionally since whether we have a host FD can change + // across save/restore. + for _, r := range mapped { + d.pf.hostFileMapper.IncRefOn(r) + } + if d.mayCachePages() { + // d.Evict() will refuse to evict memory-mapped pages, so tell the + // MemoryFile to not bother trying. + mf := d.fs.mfp.MemoryFile() + for _, r := range mapped { + mf.MarkUnevictable(d, pgalloc.EvictableRange{r.Start, r.End}) + } + } + d.mapsMu.Unlock() + return nil +} + +// RemoveMapping implements memmap.Mappable.RemoveMapping. +func (d *dentry) RemoveMapping(ctx context.Context, ms memmap.MappingSpace, ar usermem.AddrRange, offset uint64, writable bool) { + d.mapsMu.Lock() + unmapped := d.mappings.RemoveMapping(ms, ar, offset, writable) + for _, r := range unmapped { + d.pf.hostFileMapper.DecRefOn(r) + } + if d.mayCachePages() { + // Pages that are no longer referenced by any application memory + // mappings are now considered unused; allow MemoryFile to evict them + // when necessary. + mf := d.fs.mfp.MemoryFile() + d.dataMu.Lock() + for _, r := range unmapped { + // Since these pages are no longer mapped, they are no longer + // concurrently dirtyable by a writable memory mapping. + d.dirty.AllowClean(r) + mf.MarkEvictable(d, pgalloc.EvictableRange{r.Start, r.End}) + } + d.dataMu.Unlock() + } + d.mapsMu.Unlock() +} + +// CopyMapping implements memmap.Mappable.CopyMapping. +func (d *dentry) CopyMapping(ctx context.Context, ms memmap.MappingSpace, srcAR, dstAR usermem.AddrRange, offset uint64, writable bool) error { + return d.AddMapping(ctx, ms, dstAR, offset, writable) +} + +// Translate implements memmap.Mappable.Translate. +func (d *dentry) Translate(ctx context.Context, required, optional memmap.MappableRange, at usermem.AccessType) ([]memmap.Translation, error) { + d.handleMu.RLock() + if d.handle.fd >= 0 && !d.fs.opts.forcePageCache { + d.handleMu.RUnlock() + mr := optional + if d.fs.opts.limitHostFDTranslation { + mr = maxFillRange(required, optional) + } + return []memmap.Translation{ + { + Source: mr, + File: &d.pf, + Offset: mr.Start, + Perms: usermem.AnyAccess, + }, + }, nil + } + + d.dataMu.Lock() + + // Constrain translations to d.size (rounded up) to prevent translation to + // pages that may be concurrently truncated. + pgend := pageRoundUp(d.size) + var beyondEOF bool + if required.End > pgend { + if required.Start >= pgend { + d.dataMu.Unlock() + d.handleMu.RUnlock() + return nil, &memmap.BusError{io.EOF} + } + beyondEOF = true + required.End = pgend + } + if optional.End > pgend { + optional.End = pgend + } + + mf := d.fs.mfp.MemoryFile() + cerr := d.cache.Fill(ctx, required, maxFillRange(required, optional), mf, usage.PageCache, d.handle.readToBlocksAt) + + var ts []memmap.Translation + var translatedEnd uint64 + for seg := d.cache.FindSegment(required.Start); seg.Ok() && seg.Start() < required.End; seg, _ = seg.NextNonEmpty() { + segMR := seg.Range().Intersect(optional) + // TODO(jamieliu): Make Translations writable even if writability is + // not required if already kept-dirty by another writable translation. + perms := usermem.AccessType{ + Read: true, + Execute: true, + } + if at.Write { + // From this point forward, this memory can be dirtied through the + // mapping at any time. + d.dirty.KeepDirty(segMR) + perms.Write = true + } + ts = append(ts, memmap.Translation{ + Source: segMR, + File: mf, + Offset: seg.FileRangeOf(segMR).Start, + Perms: perms, + }) + translatedEnd = segMR.End + } + + d.dataMu.Unlock() + d.handleMu.RUnlock() + + // Don't return the error returned by c.cache.Fill if it occurred outside + // of required. + if translatedEnd < required.End && cerr != nil { + return ts, &memmap.BusError{cerr} + } + if beyondEOF { + return ts, &memmap.BusError{io.EOF} + } + return ts, nil +} + +func maxFillRange(required, optional memmap.MappableRange) memmap.MappableRange { + const maxReadahead = 64 << 10 // 64 KB, chosen arbitrarily + if required.Length() >= maxReadahead { + return required + } + if optional.Length() <= maxReadahead { + return optional + } + optional.Start = required.Start + if optional.Length() <= maxReadahead { + return optional + } + optional.End = optional.Start + maxReadahead + return optional +} + +// InvalidateUnsavable implements memmap.Mappable.InvalidateUnsavable. +func (d *dentry) InvalidateUnsavable(ctx context.Context) error { + // Whether we have a host fd (and consequently what platform.File is + // mapped) can change across save/restore, so invalidate all translations + // unconditionally. + d.mapsMu.Lock() + defer d.mapsMu.Unlock() + d.mappings.InvalidateAll(memmap.InvalidateOpts{}) + + // Write the cache's contents back to the remote file so that if we have a + // host fd after restore, the remote file's contents are coherent. + mf := d.fs.mfp.MemoryFile() + d.dataMu.Lock() + defer d.dataMu.Unlock() + if err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, mf, d.handle.writeFromBlocksAt); err != nil { + return err + } + + // Discard the cache so that it's not stored in saved state. This is safe + // because per InvalidateUnsavable invariants, no new translations can have + // been returned after we invalidated all existing translations above. + d.cache.DropAll(mf) + d.dirty.RemoveAll() + + return nil +} + +// Evict implements pgalloc.EvictableMemoryUser.Evict. +func (d *dentry) Evict(ctx context.Context, er pgalloc.EvictableRange) { + d.mapsMu.Lock() + defer d.mapsMu.Unlock() + d.dataMu.Lock() + defer d.dataMu.Unlock() + + mr := memmap.MappableRange{er.Start, er.End} + mf := d.fs.mfp.MemoryFile() + // Only allow pages that are no longer memory-mapped to be evicted. + for mgap := d.mappings.LowerBoundGap(mr.Start); mgap.Ok() && mgap.Start() < mr.End; mgap = mgap.NextGap() { + mgapMR := mgap.Range().Intersect(mr) + if mgapMR.Length() == 0 { + continue + } + if err := fsutil.SyncDirty(ctx, mgapMR, &d.cache, &d.dirty, d.size, mf, d.handle.writeFromBlocksAt); err != nil { + log.Warningf("Failed to writeback cached data %v: %v", mgapMR, err) + } + d.cache.Drop(mgapMR, mf) + d.dirty.KeepClean(mgapMR) + } +} + +// dentryPlatformFile implements platform.File. It exists solely because dentry +// cannot implement both vfs.DentryImpl.IncRef and platform.File.IncRef. +// +// dentryPlatformFile is only used when a host FD representing the remote file +// is available (i.e. dentry.handle.fd >= 0), and that FD is used for +// application memory mappings (i.e. !filesystem.opts.forcePageCache). +type dentryPlatformFile struct { + *dentry + + // fdRefs counts references on platform.File offsets. fdRefs is protected + // by dentry.dataMu. + fdRefs fsutil.FrameRefSet + + // If this dentry represents a regular file, and handle.fd >= 0, + // hostFileMapper caches mappings of handle.fd. + hostFileMapper fsutil.HostFileMapper +} + +// IncRef implements platform.File.IncRef. +func (d *dentryPlatformFile) IncRef(fr platform.FileRange) { + d.dataMu.Lock() + seg, gap := d.fdRefs.Find(fr.Start) + for { + switch { + case seg.Ok() && seg.Start() < fr.End: + seg = d.fdRefs.Isolate(seg, fr) + seg.SetValue(seg.Value() + 1) + seg, gap = seg.NextNonEmpty() + case gap.Ok() && gap.Start() < fr.End: + newRange := gap.Range().Intersect(fr) + usage.MemoryAccounting.Inc(newRange.Length(), usage.Mapped) + seg, gap = d.fdRefs.InsertWithoutMerging(gap, newRange, 1).NextNonEmpty() + default: + d.fdRefs.MergeAdjacent(fr) + d.dataMu.Unlock() + return + } + } +} + +// DecRef implements platform.File.DecRef. +func (d *dentryPlatformFile) DecRef(fr platform.FileRange) { + d.dataMu.Lock() + seg := d.fdRefs.FindSegment(fr.Start) + + for seg.Ok() && seg.Start() < fr.End { + seg = d.fdRefs.Isolate(seg, fr) + if old := seg.Value(); old == 1 { + usage.MemoryAccounting.Dec(seg.Range().Length(), usage.Mapped) + seg = d.fdRefs.Remove(seg).NextSegment() + } else { + seg.SetValue(old - 1) + seg = seg.NextSegment() + } + } + d.fdRefs.MergeAdjacent(fr) + d.dataMu.Unlock() + +} + +// MapInternal implements platform.File.MapInternal. +func (d *dentryPlatformFile) MapInternal(fr platform.FileRange, at usermem.AccessType) (safemem.BlockSeq, error) { + d.handleMu.RLock() + bs, err := d.hostFileMapper.MapInternal(fr, int(d.handle.fd), at.Write) + d.handleMu.RUnlock() + return bs, err +} + +// FD implements platform.File.FD. +func (d *dentryPlatformFile) FD() int { + d.handleMu.RLock() + fd := d.handle.fd + d.handleMu.RUnlock() + return int(fd) +} diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go new file mode 100644 index 000000000..08c691c47 --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/special_file.go @@ -0,0 +1,159 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package gofer + +import ( + "sync" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/safemem" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +// specialFileFD implements vfs.FileDescriptionImpl for files other than +// regular files, directories, and symlinks: pipes, sockets, etc. It is also +// used for regular files when filesystemOptions.specialRegularFiles is in +// effect. specialFileFD differs from regularFileFD by using per-FD handles +// instead of shared per-dentry handles, and never buffering I/O. +type specialFileFD struct { + fileDescription + + // handle is immutable. + handle handle + + // off is the file offset. off is protected by mu. (POSIX 2.9.7 only + // requires operations using the file offset to be atomic for regular files + // and symlinks; however, since specialFileFD may be used for regular + // files, we apply this atomicity unconditionally.) + mu sync.Mutex + off int64 +} + +// Release implements vfs.FileDescriptionImpl.Release. +func (fd *specialFileFD) Release() { + fd.handle.close(context.Background()) + fs := fd.vfsfd.Mount().Filesystem().Impl().(*filesystem) + fs.syncMu.Lock() + delete(fs.specialFileFDs, fd) + fs.syncMu.Unlock() +} + +// OnClose implements vfs.FileDescriptionImpl.OnClose. +func (fd *specialFileFD) OnClose(ctx context.Context) error { + if !fd.vfsfd.IsWritable() { + return nil + } + return fd.handle.file.flush(ctx) +} + +// PRead implements vfs.FileDescriptionImpl.PRead. +func (fd *specialFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + if offset < 0 { + return 0, syserror.EINVAL + } + if opts.Flags != 0 { + return 0, syserror.EOPNOTSUPP + } + + // Going through dst.CopyOutFrom() holds MM locks around file operations of + // unknown duration. For regularFileFD, doing so is necessary to support + // mmap due to lock ordering; MM locks precede dentry.dataMu. That doesn't + // hold here since specialFileFD doesn't client-cache data. Just buffer the + // read instead. + if d := fd.dentry(); d.fs.opts.interop != InteropModeShared { + d.touchAtime(ctx, fd.vfsfd.Mount()) + } + buf := make([]byte, dst.NumBytes()) + n, err := fd.handle.readToBlocksAt(ctx, safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf)), uint64(offset)) + if n == 0 { + return 0, err + } + if cp, cperr := dst.CopyOut(ctx, buf[:n]); cperr != nil { + return int64(cp), cperr + } + return int64(n), err +} + +// Read implements vfs.FileDescriptionImpl.Read. +func (fd *specialFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + fd.mu.Lock() + n, err := fd.PRead(ctx, dst, fd.off, opts) + fd.off += n + fd.mu.Unlock() + return n, err +} + +// PWrite implements vfs.FileDescriptionImpl.PWrite. +func (fd *specialFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + if offset < 0 { + return 0, syserror.EINVAL + } + if opts.Flags != 0 { + return 0, syserror.EOPNOTSUPP + } + + // Do a buffered write. See rationale in PRead. + if d := fd.dentry(); d.fs.opts.interop != InteropModeShared { + d.touchCMtime(ctx) + } + buf := make([]byte, src.NumBytes()) + // Don't do partial writes if we get a partial read from src. + if _, err := src.CopyIn(ctx, buf); err != nil { + return 0, err + } + n, err := fd.handle.writeFromBlocksAt(ctx, safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf)), uint64(offset)) + return int64(n), err +} + +// Write implements vfs.FileDescriptionImpl.Write. +func (fd *specialFileFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + fd.mu.Lock() + n, err := fd.PWrite(ctx, src, fd.off, opts) + fd.off += n + fd.mu.Unlock() + return n, err +} + +// Seek implements vfs.FileDescriptionImpl.Seek. +func (fd *specialFileFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + fd.mu.Lock() + defer fd.mu.Unlock() + switch whence { + case linux.SEEK_SET: + // Use offset as given. + case linux.SEEK_CUR: + offset += fd.off + default: + // SEEK_END, SEEK_DATA, and SEEK_HOLE aren't supported since it's not + // clear that file size is even meaningful for these files. + return 0, syserror.EINVAL + } + if offset < 0 { + return 0, syserror.EINVAL + } + fd.off = offset + return offset, nil +} + +// Sync implements vfs.FileDescriptionImpl.Sync. +func (fd *specialFileFD) Sync(ctx context.Context) error { + if !fd.vfsfd.IsWritable() { + return nil + } + return fd.handle.sync(ctx) +} diff --git a/pkg/sentry/fsimpl/gofer/symlink.go b/pkg/sentry/fsimpl/gofer/symlink.go new file mode 100644 index 000000000..adf43be60 --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/symlink.go @@ -0,0 +1,47 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package gofer + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/vfs" +) + +func (d *dentry) isSymlink() bool { + return d.fileType() == linux.S_IFLNK +} + +// Precondition: d.isSymlink(). +func (d *dentry) readlink(ctx context.Context, mnt *vfs.Mount) (string, error) { + if d.fs.opts.interop != InteropModeShared { + d.touchAtime(ctx, mnt) + d.dataMu.Lock() + if d.haveTarget { + target := d.target + d.dataMu.Unlock() + return target, nil + } + } + target, err := d.file.readlink(ctx) + if d.fs.opts.interop != InteropModeShared { + if err == nil { + d.haveTarget = true + d.target = target + } + d.dataMu.Unlock() + } + return target, err +} diff --git a/pkg/sentry/fsimpl/gofer/time.go b/pkg/sentry/fsimpl/gofer/time.go new file mode 100644 index 000000000..7598ec6a8 --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/time.go @@ -0,0 +1,75 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package gofer + +import ( + "sync/atomic" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" + "gvisor.dev/gvisor/pkg/sentry/vfs" +) + +func dentryTimestampFromP9(s, ns uint64) int64 { + return int64(s*1e9 + ns) +} + +func dentryTimestampFromStatx(ts linux.StatxTimestamp) int64 { + return ts.Sec*1e9 + int64(ts.Nsec) +} + +func statxTimestampFromDentry(ns int64) linux.StatxTimestamp { + return linux.StatxTimestamp{ + Sec: ns / 1e9, + Nsec: uint32(ns % 1e9), + } +} + +func nowFromContext(ctx context.Context) (int64, bool) { + if clock := ktime.RealtimeClockFromContext(ctx); clock != nil { + return clock.Now().Nanoseconds(), true + } + return 0, false +} + +// Preconditions: fs.interop != InteropModeShared. +func (d *dentry) touchAtime(ctx context.Context, mnt *vfs.Mount) { + if err := mnt.CheckBeginWrite(); err != nil { + return + } + now, ok := nowFromContext(ctx) + if !ok { + mnt.EndWrite() + return + } + d.metadataMu.Lock() + atomic.StoreInt64(&d.atime, now) + d.metadataMu.Unlock() + mnt.EndWrite() +} + +// Preconditions: fs.interop != InteropModeShared. The caller has successfully +// called vfs.Mount.CheckBeginWrite(). +func (d *dentry) touchCMtime(ctx context.Context) { + now, ok := nowFromContext(ctx) + if !ok { + return + } + d.metadataMu.Lock() + atomic.StoreInt64(&d.mtime, now) + atomic.StoreInt64(&d.ctime, now) + d.metadataMu.Unlock() +} diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 5ee9cf1e9..72bc15264 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -622,7 +622,7 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error if child.inode.isDir() { return syserror.EISDIR } - if !rp.MustBeDir() { + if rp.MustBeDir() { return syserror.ENOTDIR } mnt := rp.Mount() diff --git a/pkg/sentry/socket/hostinet/socket.go b/pkg/sentry/socket/hostinet/socket.go index bde4c7a1e..34f63986f 100644 --- a/pkg/sentry/socket/hostinet/socket.go +++ b/pkg/sentry/socket/hostinet/socket.go @@ -126,7 +126,7 @@ func (s *socketOperations) Read(ctx context.Context, _ *fs.File, dst usermem.IOS } return uint64(n), nil } - return readv(s.fd, iovecsFromBlockSeq(dsts)) + return readv(s.fd, safemem.IovecsFromBlockSeq(dsts)) })) return int64(n), err } @@ -149,7 +149,7 @@ func (s *socketOperations) Write(ctx context.Context, _ *fs.File, src usermem.IO } return uint64(n), nil } - return writev(s.fd, iovecsFromBlockSeq(srcs)) + return writev(s.fd, safemem.IovecsFromBlockSeq(srcs)) })) return int64(n), err } @@ -402,7 +402,7 @@ func (s *socketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags // We always do a non-blocking recv*(). sysflags := flags | syscall.MSG_DONTWAIT - iovs := iovecsFromBlockSeq(dsts) + iovs := safemem.IovecsFromBlockSeq(dsts) msg := syscall.Msghdr{ Iov: &iovs[0], Iovlen: uint64(len(iovs)), @@ -522,7 +522,7 @@ func (s *socketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to [] return uint64(n), nil } - iovs := iovecsFromBlockSeq(srcs) + iovs := safemem.IovecsFromBlockSeq(srcs) msg := syscall.Msghdr{ Iov: &iovs[0], Iovlen: uint64(len(iovs)), @@ -567,21 +567,6 @@ func (s *socketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to [] return int(n), syserr.FromError(err) } -func iovecsFromBlockSeq(bs safemem.BlockSeq) []syscall.Iovec { - iovs := make([]syscall.Iovec, 0, bs.NumBlocks()) - for ; !bs.IsEmpty(); bs = bs.Tail() { - b := bs.Head() - iovs = append(iovs, syscall.Iovec{ - Base: &b.ToSlice()[0], - Len: uint64(b.Len()), - }) - // We don't need to care about b.NeedSafecopy(), because the host - // kernel will handle such address ranges just fine (by returning - // EFAULT). - } - return iovs -} - func translateIOSyscallError(err error) error { if err == syscall.EAGAIN || err == syscall.EWOULDBLOCK { return syserror.ErrWouldBlock -- cgit v1.2.3 From dcffddf0cae026411e7e678744a1e39dc2b513cf Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Tue, 4 Feb 2020 11:47:41 -0800 Subject: Remove argument from vfs.MountNamespace.DecRef() Updates #1035 PiperOrigin-RevId: 293194631 --- pkg/sentry/fsimpl/devtmpfs/devtmpfs.go | 2 +- pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go | 2 +- pkg/sentry/fsimpl/testutil/testutil.go | 2 +- pkg/sentry/fsimpl/tmpfs/benchmark_test.go | 4 ++-- pkg/sentry/fsimpl/tmpfs/regular_file_test.go | 2 +- pkg/sentry/vfs/mount.go | 3 ++- 6 files changed, 8 insertions(+), 7 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go index d36fa74fb..e03a0c665 100644 --- a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go +++ b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go @@ -86,7 +86,7 @@ func NewAccessor(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth // Release must be called when a is no longer in use. func (a *Accessor) Release() { a.root.DecRef() - a.mntns.DecRef(a.vfsObj) + a.mntns.DecRef() } // accessorContext implements context.Context by extending an existing diff --git a/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go b/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go index 82c58c900..73308a2b5 100644 --- a/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go +++ b/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go @@ -45,7 +45,7 @@ func TestDevtmpfs(t *testing.T) { if err != nil { t.Fatalf("failed to create tmpfs root mount: %v", err) } - defer mntns.DecRef(vfsObj) + defer mntns.DecRef() root := mntns.Root() defer root.DecRef() devpop := vfs.PathOperation{ diff --git a/pkg/sentry/fsimpl/testutil/testutil.go b/pkg/sentry/fsimpl/testutil/testutil.go index 1c98335c1..69fd84ddd 100644 --- a/pkg/sentry/fsimpl/testutil/testutil.go +++ b/pkg/sentry/fsimpl/testutil/testutil.go @@ -98,7 +98,7 @@ func (s *System) WithTemporaryContext(ctx context.Context) *System { // Destroy release resources associated with a test system. func (s *System) Destroy() { s.Root.DecRef() - s.mns.DecRef(s.VFS) // Reference on mns passed to NewSystem. + s.mns.DecRef() // Reference on mns passed to NewSystem. } // ReadToEnd reads the contents of fd until EOF to a string. diff --git a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go index 54241c8e8..9fce5e4b4 100644 --- a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go +++ b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go @@ -183,7 +183,7 @@ func BenchmarkVFS2MemfsStat(b *testing.B) { if err != nil { b.Fatalf("failed to create tmpfs root mount: %v", err) } - defer mntns.DecRef(vfsObj) + defer mntns.DecRef() var filePathBuilder strings.Builder filePathBuilder.WriteByte('/') @@ -374,7 +374,7 @@ func BenchmarkVFS2MemfsMountStat(b *testing.B) { if err != nil { b.Fatalf("failed to create tmpfs root mount: %v", err) } - defer mntns.DecRef(vfsObj) + defer mntns.DecRef() var filePathBuilder strings.Builder filePathBuilder.WriteByte('/') diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go index 2b52992ea..e9f71e334 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go @@ -51,7 +51,7 @@ func newTmpfsRoot(ctx context.Context) (*vfs.VirtualFilesystem, vfs.VirtualDentr root := mntns.Root() return vfsObj, root, func() { root.DecRef() - mntns.DecRef(vfsObj) + mntns.DecRef() }, nil } diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index d39528051..1fbb420f9 100644 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go @@ -423,7 +423,8 @@ func (mntns *MountNamespace) IncRef() { } // DecRef decrements mntns' reference count. -func (mntns *MountNamespace) DecRef(vfs *VirtualFilesystem) { +func (mntns *MountNamespace) DecRef() { + vfs := mntns.root.fs.VirtualFilesystem() if refs := atomic.AddInt64(&mntns.refs, -1); refs == 0 { vfs.mountMu.Lock() vfs.mounts.seq.BeginWrite() -- cgit v1.2.3 From 6823b5e244a5748032130574ae3a25a0a36bbbf5 Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Tue, 4 Feb 2020 13:05:30 -0800 Subject: timer_create(2) should return 0 on success The timer ID is copied out to the argument. Fixes #1738 PiperOrigin-RevId: 293210801 --- pkg/sentry/syscalls/linux/sys_timer.go | 2 +- test/syscalls/linux/timers.cc | 18 +++++++++++++++++- 2 files changed, 18 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/sys_timer.go b/pkg/sentry/syscalls/linux/sys_timer.go index 432351917..a4c400f87 100644 --- a/pkg/sentry/syscalls/linux/sys_timer.go +++ b/pkg/sentry/syscalls/linux/sys_timer.go @@ -146,7 +146,7 @@ func TimerCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S return 0, nil, err } - return uintptr(id), nil, nil + return 0, nil, nil } // TimerSettime implements linux syscall timer_settime(2). diff --git a/test/syscalls/linux/timers.cc b/test/syscalls/linux/timers.cc index 3db18d7ac..2f92c27da 100644 --- a/test/syscalls/linux/timers.cc +++ b/test/syscalls/linux/timers.cc @@ -297,9 +297,13 @@ class IntervalTimer { PosixErrorOr TimerCreate(clockid_t clockid, const struct sigevent& sev) { int timerid; - if (syscall(SYS_timer_create, clockid, &sev, &timerid) < 0) { + int ret = syscall(SYS_timer_create, clockid, &sev, &timerid); + if (ret < 0) { return PosixError(errno, "timer_create"); } + if (ret > 0) { + return PosixError(EINVAL, "timer_create should never return positive"); + } MaybeSave(); return IntervalTimer(timerid); } @@ -317,6 +321,18 @@ TEST(IntervalTimerTest, IsInitiallyStopped) { EXPECT_EQ(0, its.it_value.tv_nsec); } +// Kernel can create multiple timers without issue. +// +// Regression test for gvisor.dev/issue/1738. +TEST(IntervalTimerTest, MultipleTimers) { + struct sigevent sev = {}; + sev.sigev_notify = SIGEV_NONE; + const auto timer1 = + ASSERT_NO_ERRNO_AND_VALUE(TimerCreate(CLOCK_MONOTONIC, sev)); + const auto timer2 = + ASSERT_NO_ERRNO_AND_VALUE(TimerCreate(CLOCK_MONOTONIC, sev)); +} + TEST(IntervalTimerTest, SingleShotSilent) { struct sigevent sev = {}; sev.sigev_notify = SIGEV_NONE; -- cgit v1.2.3 From 665b614e4a6e715bac25bea15c5c29184016e549 Mon Sep 17 00:00:00 2001 From: Ting-Yu Wang Date: Tue, 4 Feb 2020 18:04:26 -0800 Subject: Support RTM_NEWADDR and RTM_GETLINK in (rt)netlink. PiperOrigin-RevId: 293271055 --- pkg/sentry/inet/inet.go | 4 + pkg/sentry/inet/test_stack.go | 6 + pkg/sentry/socket/hostinet/stack.go | 5 + pkg/sentry/socket/netlink/BUILD | 14 +- pkg/sentry/socket/netlink/message.go | 129 +++++++++++ pkg/sentry/socket/netlink/message_test.go | 312 +++++++++++++++++++++++++++ pkg/sentry/socket/netlink/provider.go | 2 +- pkg/sentry/socket/netlink/route/BUILD | 2 - pkg/sentry/socket/netlink/route/protocol.go | 238 ++++++++++++++------ pkg/sentry/socket/netlink/socket.go | 54 ++--- pkg/sentry/socket/netlink/uevent/protocol.go | 2 +- pkg/sentry/socket/netstack/stack.go | 55 +++++ pkg/tcpip/stack/stack.go | 9 + test/syscalls/linux/BUILD | 2 + test/syscalls/linux/socket_netlink_route.cc | 296 ++++++++++++++++++++----- test/syscalls/linux/socket_netlink_util.cc | 45 +++- test/syscalls/linux/socket_netlink_util.h | 9 + 17 files changed, 1022 insertions(+), 162 deletions(-) create mode 100644 pkg/sentry/socket/netlink/message_test.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/inet/inet.go b/pkg/sentry/inet/inet.go index a7dfb78a7..2916a0644 100644 --- a/pkg/sentry/inet/inet.go +++ b/pkg/sentry/inet/inet.go @@ -28,6 +28,10 @@ type Stack interface { // interface indexes to a slice of associated interface address properties. InterfaceAddrs() map[int32][]InterfaceAddr + // AddInterfaceAddr adds an address to the network interface identified by + // index. + AddInterfaceAddr(idx int32, addr InterfaceAddr) error + // SupportsIPv6 returns true if the stack supports IPv6 connectivity. SupportsIPv6() bool diff --git a/pkg/sentry/inet/test_stack.go b/pkg/sentry/inet/test_stack.go index dcfcbd97e..d8961fc94 100644 --- a/pkg/sentry/inet/test_stack.go +++ b/pkg/sentry/inet/test_stack.go @@ -47,6 +47,12 @@ func (s *TestStack) InterfaceAddrs() map[int32][]InterfaceAddr { return s.InterfaceAddrsMap } +// AddInterfaceAddr implements Stack.AddInterfaceAddr. +func (s *TestStack) AddInterfaceAddr(idx int32, addr InterfaceAddr) error { + s.InterfaceAddrsMap[idx] = append(s.InterfaceAddrsMap[idx], addr) + return nil +} + // SupportsIPv6 implements Stack.SupportsIPv6. func (s *TestStack) SupportsIPv6() bool { return s.SupportsIPv6Flag diff --git a/pkg/sentry/socket/hostinet/stack.go b/pkg/sentry/socket/hostinet/stack.go index 034eca676..a48082631 100644 --- a/pkg/sentry/socket/hostinet/stack.go +++ b/pkg/sentry/socket/hostinet/stack.go @@ -310,6 +310,11 @@ func (s *Stack) InterfaceAddrs() map[int32][]inet.InterfaceAddr { return addrs } +// AddInterfaceAddr implements inet.Stack.AddInterfaceAddr. +func (s *Stack) AddInterfaceAddr(idx int32, addr inet.InterfaceAddr) error { + return syserror.EACCES +} + // SupportsIPv6 implements inet.Stack.SupportsIPv6. func (s *Stack) SupportsIPv6() bool { return s.supportsIPv6 diff --git a/pkg/sentry/socket/netlink/BUILD b/pkg/sentry/socket/netlink/BUILD index f8b8e467d..1911cd9b8 100644 --- a/pkg/sentry/socket/netlink/BUILD +++ b/pkg/sentry/socket/netlink/BUILD @@ -1,4 +1,4 @@ -load("//tools:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -33,3 +33,15 @@ go_library( "//pkg/waiter", ], ) + +go_test( + name = "netlink_test", + size = "small", + srcs = [ + "message_test.go", + ], + deps = [ + ":netlink", + "//pkg/abi/linux", + ], +) diff --git a/pkg/sentry/socket/netlink/message.go b/pkg/sentry/socket/netlink/message.go index b21e0ca4b..4ea252ccb 100644 --- a/pkg/sentry/socket/netlink/message.go +++ b/pkg/sentry/socket/netlink/message.go @@ -30,8 +30,16 @@ func alignUp(length int, align uint) int { return (length + int(align) - 1) &^ (int(align) - 1) } +// alignPad returns the length of padding required for alignment. +// +// Preconditions: align is a power of two. +func alignPad(length int, align uint) int { + return alignUp(length, align) - length +} + // Message contains a complete serialized netlink message. type Message struct { + hdr linux.NetlinkMessageHeader buf []byte } @@ -40,10 +48,86 @@ type Message struct { // The header length will be updated by Finalize. func NewMessage(hdr linux.NetlinkMessageHeader) *Message { return &Message{ + hdr: hdr, buf: binary.Marshal(nil, usermem.ByteOrder, hdr), } } +// ParseMessage parses the first message seen at buf, returning the rest of the +// buffer. If message is malformed, ok of false is returned. For last message, +// padding check is loose, if there isn't enought padding, whole buf is consumed +// and ok is set to true. +func ParseMessage(buf []byte) (msg *Message, rest []byte, ok bool) { + b := BytesView(buf) + + hdrBytes, ok := b.Extract(linux.NetlinkMessageHeaderSize) + if !ok { + return + } + var hdr linux.NetlinkMessageHeader + binary.Unmarshal(hdrBytes, usermem.ByteOrder, &hdr) + + // Msg portion. + totalMsgLen := int(hdr.Length) + _, ok = b.Extract(totalMsgLen - linux.NetlinkMessageHeaderSize) + if !ok { + return + } + + // Padding. + numPad := alignPad(totalMsgLen, linux.NLMSG_ALIGNTO) + // Linux permits the last message not being aligned, just consume all of it. + // Ref: net/netlink/af_netlink.c:netlink_rcv_skb + if numPad > len(b) { + numPad = len(b) + } + _, ok = b.Extract(numPad) + if !ok { + return + } + + return &Message{ + hdr: hdr, + buf: buf[:totalMsgLen], + }, []byte(b), true +} + +// Header returns the header of this message. +func (m *Message) Header() linux.NetlinkMessageHeader { + return m.hdr +} + +// GetData unmarshals the payload message header from this netlink message, and +// returns the attributes portion. +func (m *Message) GetData(msg interface{}) (AttrsView, bool) { + b := BytesView(m.buf) + + _, ok := b.Extract(linux.NetlinkMessageHeaderSize) + if !ok { + return nil, false + } + + size := int(binary.Size(msg)) + msgBytes, ok := b.Extract(size) + if !ok { + return nil, false + } + binary.Unmarshal(msgBytes, usermem.ByteOrder, msg) + + numPad := alignPad(linux.NetlinkMessageHeaderSize+size, linux.NLMSG_ALIGNTO) + // Linux permits the last message not being aligned, just consume all of it. + // Ref: net/netlink/af_netlink.c:netlink_rcv_skb + if numPad > len(b) { + numPad = len(b) + } + _, ok = b.Extract(numPad) + if !ok { + return nil, false + } + + return AttrsView(b), true +} + // Finalize returns the []byte containing the entire message, with the total // length set in the message header. The Message must not be modified after // calling Finalize. @@ -157,3 +241,48 @@ func (ms *MessageSet) AddMessage(hdr linux.NetlinkMessageHeader) *Message { ms.Messages = append(ms.Messages, m) return m } + +// AttrsView is a view into the attributes portion of a netlink message. +type AttrsView []byte + +// Empty returns whether there is no attribute left in v. +func (v AttrsView) Empty() bool { + return len(v) == 0 +} + +// ParseFirst parses first netlink attribute at the beginning of v. +func (v AttrsView) ParseFirst() (hdr linux.NetlinkAttrHeader, value []byte, rest AttrsView, ok bool) { + b := BytesView(v) + + hdrBytes, ok := b.Extract(linux.NetlinkAttrHeaderSize) + if !ok { + return + } + binary.Unmarshal(hdrBytes, usermem.ByteOrder, &hdr) + + value, ok = b.Extract(int(hdr.Length) - linux.NetlinkAttrHeaderSize) + if !ok { + return + } + + _, ok = b.Extract(alignPad(int(hdr.Length), linux.NLA_ALIGNTO)) + if !ok { + return + } + + return hdr, value, AttrsView(b), ok +} + +// BytesView supports extracting data from a byte slice with bounds checking. +type BytesView []byte + +// Extract removes the first n bytes from v and returns it. If n is out of +// bounds, it returns false. +func (v *BytesView) Extract(n int) ([]byte, bool) { + if n < 0 || n > len(*v) { + return nil, false + } + extracted := (*v)[:n] + *v = (*v)[n:] + return extracted, true +} diff --git a/pkg/sentry/socket/netlink/message_test.go b/pkg/sentry/socket/netlink/message_test.go new file mode 100644 index 000000000..ef13d9386 --- /dev/null +++ b/pkg/sentry/socket/netlink/message_test.go @@ -0,0 +1,312 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package message_test + +import ( + "bytes" + "reflect" + "testing" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/socket/netlink" +) + +type dummyNetlinkMsg struct { + Foo uint16 +} + +func TestParseMessage(t *testing.T) { + tests := []struct { + desc string + input []byte + + header linux.NetlinkMessageHeader + dataMsg *dummyNetlinkMsg + restLen int + ok bool + }{ + { + desc: "valid", + input: []byte{ + 0x14, 0x00, 0x00, 0x00, // Length + 0x01, 0x00, // Type + 0x02, 0x00, // Flags + 0x03, 0x00, 0x00, 0x00, // Seq + 0x04, 0x00, 0x00, 0x00, // PortID + 0x30, 0x31, 0x00, 0x00, // Data message with 2 bytes padding + }, + header: linux.NetlinkMessageHeader{ + Length: 20, + Type: 1, + Flags: 2, + Seq: 3, + PortID: 4, + }, + dataMsg: &dummyNetlinkMsg{ + Foo: 0x3130, + }, + restLen: 0, + ok: true, + }, + { + desc: "valid with next message", + input: []byte{ + 0x14, 0x00, 0x00, 0x00, // Length + 0x01, 0x00, // Type + 0x02, 0x00, // Flags + 0x03, 0x00, 0x00, 0x00, // Seq + 0x04, 0x00, 0x00, 0x00, // PortID + 0x30, 0x31, 0x00, 0x00, // Data message with 2 bytes padding + 0xFF, // Next message (rest) + }, + header: linux.NetlinkMessageHeader{ + Length: 20, + Type: 1, + Flags: 2, + Seq: 3, + PortID: 4, + }, + dataMsg: &dummyNetlinkMsg{ + Foo: 0x3130, + }, + restLen: 1, + ok: true, + }, + { + desc: "valid for last message without padding", + input: []byte{ + 0x12, 0x00, 0x00, 0x00, // Length + 0x01, 0x00, // Type + 0x02, 0x00, // Flags + 0x03, 0x00, 0x00, 0x00, // Seq + 0x04, 0x00, 0x00, 0x00, // PortID + 0x30, 0x31, // Data message + }, + header: linux.NetlinkMessageHeader{ + Length: 18, + Type: 1, + Flags: 2, + Seq: 3, + PortID: 4, + }, + dataMsg: &dummyNetlinkMsg{ + Foo: 0x3130, + }, + restLen: 0, + ok: true, + }, + { + desc: "valid for last message not to be aligned", + input: []byte{ + 0x13, 0x00, 0x00, 0x00, // Length + 0x01, 0x00, // Type + 0x02, 0x00, // Flags + 0x03, 0x00, 0x00, 0x00, // Seq + 0x04, 0x00, 0x00, 0x00, // PortID + 0x30, 0x31, // Data message + 0x00, // Excessive 1 byte permitted at end + }, + header: linux.NetlinkMessageHeader{ + Length: 19, + Type: 1, + Flags: 2, + Seq: 3, + PortID: 4, + }, + dataMsg: &dummyNetlinkMsg{ + Foo: 0x3130, + }, + restLen: 0, + ok: true, + }, + { + desc: "header.Length too short", + input: []byte{ + 0x04, 0x00, 0x00, 0x00, // Length + 0x01, 0x00, // Type + 0x02, 0x00, // Flags + 0x03, 0x00, 0x00, 0x00, // Seq + 0x04, 0x00, 0x00, 0x00, // PortID + 0x30, 0x31, 0x00, 0x00, // Data message with 2 bytes padding + }, + ok: false, + }, + { + desc: "header.Length too long", + input: []byte{ + 0xFF, 0xFF, 0x00, 0x00, // Length + 0x01, 0x00, // Type + 0x02, 0x00, // Flags + 0x03, 0x00, 0x00, 0x00, // Seq + 0x04, 0x00, 0x00, 0x00, // PortID + 0x30, 0x31, 0x00, 0x00, // Data message with 2 bytes padding + }, + ok: false, + }, + { + desc: "header incomplete", + input: []byte{ + 0x04, 0x00, 0x00, 0x00, // Length + }, + ok: false, + }, + { + desc: "empty message", + input: []byte{}, + ok: false, + }, + } + for _, test := range tests { + msg, rest, ok := netlink.ParseMessage(test.input) + if ok != test.ok { + t.Errorf("%v: got ok = %v, want = %v", test.desc, ok, test.ok) + continue + } + if !test.ok { + continue + } + if !reflect.DeepEqual(msg.Header(), test.header) { + t.Errorf("%v: got hdr = %+v, want = %+v", test.desc, msg.Header(), test.header) + } + + dataMsg := &dummyNetlinkMsg{} + _, dataOk := msg.GetData(dataMsg) + if !dataOk { + t.Errorf("%v: GetData.ok = %v, want = true", test.desc, dataOk) + } else if !reflect.DeepEqual(dataMsg, test.dataMsg) { + t.Errorf("%v: GetData.msg = %+v, want = %+v", test.desc, dataMsg, test.dataMsg) + } + + if got, want := rest, test.input[len(test.input)-test.restLen:]; !bytes.Equal(got, want) { + t.Errorf("%v: got rest = %v, want = %v", test.desc, got, want) + } + } +} + +func TestAttrView(t *testing.T) { + tests := []struct { + desc string + input []byte + + // Outputs for ParseFirst. + hdr linux.NetlinkAttrHeader + value []byte + restLen int + ok bool + + // Outputs for Empty. + isEmpty bool + }{ + { + desc: "valid", + input: []byte{ + 0x06, 0x00, // Length + 0x01, 0x00, // Type + 0x30, 0x31, 0x00, 0x00, // Data with 2 bytes padding + }, + hdr: linux.NetlinkAttrHeader{ + Length: 6, + Type: 1, + }, + value: []byte{0x30, 0x31}, + restLen: 0, + ok: true, + isEmpty: false, + }, + { + desc: "at alignment", + input: []byte{ + 0x08, 0x00, // Length + 0x01, 0x00, // Type + 0x30, 0x31, 0x32, 0x33, // Data + }, + hdr: linux.NetlinkAttrHeader{ + Length: 8, + Type: 1, + }, + value: []byte{0x30, 0x31, 0x32, 0x33}, + restLen: 0, + ok: true, + isEmpty: false, + }, + { + desc: "at alignment with rest data", + input: []byte{ + 0x08, 0x00, // Length + 0x01, 0x00, // Type + 0x30, 0x31, 0x32, 0x33, // Data + 0xFF, 0xFE, // Rest data + }, + hdr: linux.NetlinkAttrHeader{ + Length: 8, + Type: 1, + }, + value: []byte{0x30, 0x31, 0x32, 0x33}, + restLen: 2, + ok: true, + isEmpty: false, + }, + { + desc: "hdr.Length too long", + input: []byte{ + 0xFF, 0x00, // Length + 0x01, 0x00, // Type + 0x30, 0x31, 0x32, 0x33, // Data + }, + ok: false, + isEmpty: false, + }, + { + desc: "hdr.Length too short", + input: []byte{ + 0x01, 0x00, // Length + 0x01, 0x00, // Type + 0x30, 0x31, 0x32, 0x33, // Data + }, + ok: false, + isEmpty: false, + }, + { + desc: "empty", + input: []byte{}, + ok: false, + isEmpty: true, + }, + } + for _, test := range tests { + attrs := netlink.AttrsView(test.input) + + // Test ParseFirst(). + hdr, value, rest, ok := attrs.ParseFirst() + if ok != test.ok { + t.Errorf("%v: got ok = %v, want = %v", test.desc, ok, test.ok) + } else if test.ok { + if !reflect.DeepEqual(hdr, test.hdr) { + t.Errorf("%v: got hdr = %+v, want = %+v", test.desc, hdr, test.hdr) + } + if !bytes.Equal(value, test.value) { + t.Errorf("%v: got value = %v, want = %v", test.desc, value, test.value) + } + if wantRest := test.input[len(test.input)-test.restLen:]; !bytes.Equal(rest, wantRest) { + t.Errorf("%v: got rest = %v, want = %v", test.desc, rest, wantRest) + } + } + + // Test Empty(). + if got, want := attrs.Empty(), test.isEmpty; got != want { + t.Errorf("%v: got empty = %v, want = %v", test.desc, got, want) + } + } +} diff --git a/pkg/sentry/socket/netlink/provider.go b/pkg/sentry/socket/netlink/provider.go index 07f860a49..b0dc70e5c 100644 --- a/pkg/sentry/socket/netlink/provider.go +++ b/pkg/sentry/socket/netlink/provider.go @@ -42,7 +42,7 @@ type Protocol interface { // If err == nil, any messages added to ms will be sent back to the // other end of the socket. Setting ms.Multi will cause an NLMSG_DONE // message to be sent even if ms contains no messages. - ProcessMessage(ctx context.Context, hdr linux.NetlinkMessageHeader, data []byte, ms *MessageSet) *syserr.Error + ProcessMessage(ctx context.Context, msg *Message, ms *MessageSet) *syserr.Error } // Provider is a function that creates a new Protocol for a specific netlink diff --git a/pkg/sentry/socket/netlink/route/BUILD b/pkg/sentry/socket/netlink/route/BUILD index 622a1eafc..93127398d 100644 --- a/pkg/sentry/socket/netlink/route/BUILD +++ b/pkg/sentry/socket/netlink/route/BUILD @@ -10,13 +10,11 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", - "//pkg/binary", "//pkg/context", "//pkg/sentry/inet", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/socket/netlink", "//pkg/syserr", - "//pkg/usermem", ], ) diff --git a/pkg/sentry/socket/netlink/route/protocol.go b/pkg/sentry/socket/netlink/route/protocol.go index 2b3c7f5b3..c84d8bd7c 100644 --- a/pkg/sentry/socket/netlink/route/protocol.go +++ b/pkg/sentry/socket/netlink/route/protocol.go @@ -17,16 +17,15 @@ package route import ( "bytes" + "syscall" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/socket/netlink" "gvisor.dev/gvisor/pkg/syserr" - "gvisor.dev/gvisor/pkg/usermem" ) // commandKind describes the operational class of a message type. @@ -69,13 +68,7 @@ func (p *Protocol) CanSend() bool { } // dumpLinks handles RTM_GETLINK dump requests. -func (p *Protocol) dumpLinks(ctx context.Context, hdr linux.NetlinkMessageHeader, data []byte, ms *netlink.MessageSet) *syserr.Error { - // TODO(b/68878065): Only the dump variant of the types below are - // supported. - if hdr.Flags&linux.NLM_F_DUMP != linux.NLM_F_DUMP { - return syserr.ErrNotSupported - } - +func (p *Protocol) dumpLinks(ctx context.Context, msg *netlink.Message, ms *netlink.MessageSet) *syserr.Error { // NLM_F_DUMP + RTM_GETLINK messages are supposed to include an // ifinfomsg. However, Linux <3.9 only checked for rtgenmsg, and some // userspace applications (including glibc) still include rtgenmsg. @@ -99,44 +92,105 @@ func (p *Protocol) dumpLinks(ctx context.Context, hdr linux.NetlinkMessageHeader return nil } - for id, i := range stack.Interfaces() { - m := ms.AddMessage(linux.NetlinkMessageHeader{ - Type: linux.RTM_NEWLINK, - }) + for idx, i := range stack.Interfaces() { + addNewLinkMessage(ms, idx, i) + } - m.Put(linux.InterfaceInfoMessage{ - Family: linux.AF_UNSPEC, - Type: i.DeviceType, - Index: id, - Flags: i.Flags, - }) + return nil +} - m.PutAttrString(linux.IFLA_IFNAME, i.Name) - m.PutAttr(linux.IFLA_MTU, i.MTU) +// getLinks handles RTM_GETLINK requests. +func (p *Protocol) getLink(ctx context.Context, msg *netlink.Message, ms *netlink.MessageSet) *syserr.Error { + stack := inet.StackFromContext(ctx) + if stack == nil { + // No network devices. + return nil + } - mac := make([]byte, 6) - brd := mac - if len(i.Addr) > 0 { - mac = i.Addr - brd = bytes.Repeat([]byte{0xff}, len(i.Addr)) + // Parse message. + var ifi linux.InterfaceInfoMessage + attrs, ok := msg.GetData(&ifi) + if !ok { + return syserr.ErrInvalidArgument + } + + // Parse attributes. + var byName []byte + for !attrs.Empty() { + ahdr, value, rest, ok := attrs.ParseFirst() + if !ok { + return syserr.ErrInvalidArgument } - m.PutAttr(linux.IFLA_ADDRESS, mac) - m.PutAttr(linux.IFLA_BROADCAST, brd) + attrs = rest - // TODO(gvisor.dev/issue/578): There are many more attributes. + switch ahdr.Type { + case linux.IFLA_IFNAME: + if len(value) < 1 { + return syserr.ErrInvalidArgument + } + byName = value[:len(value)-1] + + // TODO(gvisor.dev/issue/578): Support IFLA_EXT_MASK. + } } + found := false + for idx, i := range stack.Interfaces() { + switch { + case ifi.Index > 0: + if idx != ifi.Index { + continue + } + case byName != nil: + if string(byName) != i.Name { + continue + } + default: + // Criteria not specified. + return syserr.ErrInvalidArgument + } + + addNewLinkMessage(ms, idx, i) + found = true + break + } + if !found { + return syserr.ErrNoDevice + } return nil } -// dumpAddrs handles RTM_GETADDR dump requests. -func (p *Protocol) dumpAddrs(ctx context.Context, hdr linux.NetlinkMessageHeader, data []byte, ms *netlink.MessageSet) *syserr.Error { - // TODO(b/68878065): Only the dump variant of the types below are - // supported. - if hdr.Flags&linux.NLM_F_DUMP != linux.NLM_F_DUMP { - return syserr.ErrNotSupported +// addNewLinkMessage appends RTM_NEWLINK message for the given interface into +// the message set. +func addNewLinkMessage(ms *netlink.MessageSet, idx int32, i inet.Interface) { + m := ms.AddMessage(linux.NetlinkMessageHeader{ + Type: linux.RTM_NEWLINK, + }) + + m.Put(linux.InterfaceInfoMessage{ + Family: linux.AF_UNSPEC, + Type: i.DeviceType, + Index: idx, + Flags: i.Flags, + }) + + m.PutAttrString(linux.IFLA_IFNAME, i.Name) + m.PutAttr(linux.IFLA_MTU, i.MTU) + + mac := make([]byte, 6) + brd := mac + if len(i.Addr) > 0 { + mac = i.Addr + brd = bytes.Repeat([]byte{0xff}, len(i.Addr)) } + m.PutAttr(linux.IFLA_ADDRESS, mac) + m.PutAttr(linux.IFLA_BROADCAST, brd) + + // TODO(gvisor.dev/issue/578): There are many more attributes. +} +// dumpAddrs handles RTM_GETADDR dump requests. +func (p *Protocol) dumpAddrs(ctx context.Context, msg *netlink.Message, ms *netlink.MessageSet) *syserr.Error { // RTM_GETADDR dump requests need not contain anything more than the // netlink header and 1 byte protocol family common to all // NETLINK_ROUTE requests. @@ -168,6 +222,7 @@ func (p *Protocol) dumpAddrs(ctx context.Context, hdr linux.NetlinkMessageHeader Index: uint32(id), }) + m.PutAttr(linux.IFA_LOCAL, []byte(a.Addr)) m.PutAttr(linux.IFA_ADDRESS, []byte(a.Addr)) // TODO(gvisor.dev/issue/578): There are many more attributes. @@ -252,12 +307,12 @@ func fillRoute(routes []inet.Route, addr []byte) (inet.Route, *syserr.Error) { } // parseForDestination parses a message as format of RouteMessage-RtAttr-dst. -func parseForDestination(data []byte) ([]byte, *syserr.Error) { +func parseForDestination(msg *netlink.Message) ([]byte, *syserr.Error) { var rtMsg linux.RouteMessage - if len(data) < linux.SizeOfRouteMessage { + attrs, ok := msg.GetData(&rtMsg) + if !ok { return nil, syserr.ErrInvalidArgument } - binary.Unmarshal(data[:linux.SizeOfRouteMessage], usermem.ByteOrder, &rtMsg) // iproute2 added the RTM_F_LOOKUP_TABLE flag in version v4.4.0. See // commit bc234301af12. Note we don't check this flag for backward // compatibility. @@ -265,26 +320,15 @@ func parseForDestination(data []byte) ([]byte, *syserr.Error) { return nil, syserr.ErrNotSupported } - data = data[linux.SizeOfRouteMessage:] - - // TODO(gvisor.dev/issue/1611): Add generic attribute parsing. - var rtAttr linux.RtAttr - if len(data) < linux.SizeOfRtAttr { - return nil, syserr.ErrInvalidArgument + // Expect first attribute is RTA_DST. + if hdr, value, _, ok := attrs.ParseFirst(); ok && hdr.Type == linux.RTA_DST { + return value, nil } - binary.Unmarshal(data[:linux.SizeOfRtAttr], usermem.ByteOrder, &rtAttr) - if rtAttr.Type != linux.RTA_DST { - return nil, syserr.ErrInvalidArgument - } - - if len(data) < int(rtAttr.Len) { - return nil, syserr.ErrInvalidArgument - } - return data[linux.SizeOfRtAttr:rtAttr.Len], nil + return nil, syserr.ErrInvalidArgument } // dumpRoutes handles RTM_GETROUTE requests. -func (p *Protocol) dumpRoutes(ctx context.Context, hdr linux.NetlinkMessageHeader, data []byte, ms *netlink.MessageSet) *syserr.Error { +func (p *Protocol) dumpRoutes(ctx context.Context, msg *netlink.Message, ms *netlink.MessageSet) *syserr.Error { // RTM_GETROUTE dump requests need not contain anything more than the // netlink header and 1 byte protocol family common to all // NETLINK_ROUTE requests. @@ -295,10 +339,11 @@ func (p *Protocol) dumpRoutes(ctx context.Context, hdr linux.NetlinkMessageHeade return nil } + hdr := msg.Header() routeTables := stack.RouteTable() if hdr.Flags == linux.NLM_F_REQUEST { - dst, err := parseForDestination(data) + dst, err := parseForDestination(msg) if err != nil { return err } @@ -357,10 +402,55 @@ func (p *Protocol) dumpRoutes(ctx context.Context, hdr linux.NetlinkMessageHeade return nil } +// newAddr handles RTM_NEWADDR requests. +func (p *Protocol) newAddr(ctx context.Context, msg *netlink.Message, ms *netlink.MessageSet) *syserr.Error { + stack := inet.StackFromContext(ctx) + if stack == nil { + // No network stack. + return syserr.ErrProtocolNotSupported + } + + var ifa linux.InterfaceAddrMessage + attrs, ok := msg.GetData(&ifa) + if !ok { + return syserr.ErrInvalidArgument + } + + for !attrs.Empty() { + ahdr, value, rest, ok := attrs.ParseFirst() + if !ok { + return syserr.ErrInvalidArgument + } + attrs = rest + + switch ahdr.Type { + case linux.IFA_LOCAL: + err := stack.AddInterfaceAddr(int32(ifa.Index), inet.InterfaceAddr{ + Family: ifa.Family, + PrefixLen: ifa.PrefixLen, + Flags: ifa.Flags, + Addr: value, + }) + if err == syscall.EEXIST { + flags := msg.Header().Flags + if flags&linux.NLM_F_EXCL != 0 { + return syserr.ErrExists + } + } else if err != nil { + return syserr.ErrInvalidArgument + } + } + } + return nil +} + // ProcessMessage implements netlink.Protocol.ProcessMessage. -func (p *Protocol) ProcessMessage(ctx context.Context, hdr linux.NetlinkMessageHeader, data []byte, ms *netlink.MessageSet) *syserr.Error { +func (p *Protocol) ProcessMessage(ctx context.Context, msg *netlink.Message, ms *netlink.MessageSet) *syserr.Error { + hdr := msg.Header() + // All messages start with a 1 byte protocol family. - if len(data) < 1 { + var family uint8 + if _, ok := msg.GetData(&family); !ok { // Linux ignores messages missing the protocol family. See // net/core/rtnetlink.c:rtnetlink_rcv_msg. return nil @@ -374,16 +464,32 @@ func (p *Protocol) ProcessMessage(ctx context.Context, hdr linux.NetlinkMessageH } } - switch hdr.Type { - case linux.RTM_GETLINK: - return p.dumpLinks(ctx, hdr, data, ms) - case linux.RTM_GETADDR: - return p.dumpAddrs(ctx, hdr, data, ms) - case linux.RTM_GETROUTE: - return p.dumpRoutes(ctx, hdr, data, ms) - default: - return syserr.ErrNotSupported + if hdr.Flags&linux.NLM_F_DUMP == linux.NLM_F_DUMP { + // TODO(b/68878065): Only the dump variant of the types below are + // supported. + switch hdr.Type { + case linux.RTM_GETLINK: + return p.dumpLinks(ctx, msg, ms) + case linux.RTM_GETADDR: + return p.dumpAddrs(ctx, msg, ms) + case linux.RTM_GETROUTE: + return p.dumpRoutes(ctx, msg, ms) + default: + return syserr.ErrNotSupported + } + } else if hdr.Flags&linux.NLM_F_REQUEST == linux.NLM_F_REQUEST { + switch hdr.Type { + case linux.RTM_GETLINK: + return p.getLink(ctx, msg, ms) + case linux.RTM_GETROUTE: + return p.dumpRoutes(ctx, msg, ms) + case linux.RTM_NEWADDR: + return p.newAddr(ctx, msg, ms) + default: + return syserr.ErrNotSupported + } } + return syserr.ErrNotSupported } // init registers the NETLINK_ROUTE provider. diff --git a/pkg/sentry/socket/netlink/socket.go b/pkg/sentry/socket/netlink/socket.go index c4b95debb..2ca02567d 100644 --- a/pkg/sentry/socket/netlink/socket.go +++ b/pkg/sentry/socket/netlink/socket.go @@ -644,47 +644,38 @@ func (s *Socket) sendResponse(ctx context.Context, ms *MessageSet) *syserr.Error return nil } -func (s *Socket) dumpErrorMesage(ctx context.Context, hdr linux.NetlinkMessageHeader, ms *MessageSet, err *syserr.Error) *syserr.Error { +func dumpErrorMesage(hdr linux.NetlinkMessageHeader, ms *MessageSet, err *syserr.Error) { m := ms.AddMessage(linux.NetlinkMessageHeader{ Type: linux.NLMSG_ERROR, }) - m.Put(linux.NetlinkErrorMessage{ Error: int32(-err.ToLinux().Number()), Header: hdr, }) - return nil +} +func dumpAckMesage(hdr linux.NetlinkMessageHeader, ms *MessageSet) { + m := ms.AddMessage(linux.NetlinkMessageHeader{ + Type: linux.NLMSG_ERROR, + }) + m.Put(linux.NetlinkErrorMessage{ + Error: 0, + Header: hdr, + }) } // processMessages handles each message in buf, passing it to the protocol // handler for final handling. func (s *Socket) processMessages(ctx context.Context, buf []byte) *syserr.Error { for len(buf) > 0 { - if len(buf) < linux.NetlinkMessageHeaderSize { + msg, rest, ok := ParseMessage(buf) + if !ok { // Linux ignores messages that are too short. See // net/netlink/af_netlink.c:netlink_rcv_skb. break } - - var hdr linux.NetlinkMessageHeader - binary.Unmarshal(buf[:linux.NetlinkMessageHeaderSize], usermem.ByteOrder, &hdr) - - if hdr.Length < linux.NetlinkMessageHeaderSize || uint64(hdr.Length) > uint64(len(buf)) { - // Linux ignores malformed messages. See - // net/netlink/af_netlink.c:netlink_rcv_skb. - break - } - - // Data from this message. - data := buf[linux.NetlinkMessageHeaderSize:hdr.Length] - - // Advance to the next message. - next := alignUp(int(hdr.Length), linux.NLMSG_ALIGNTO) - if next >= len(buf)-1 { - next = len(buf) - 1 - } - buf = buf[next:] + buf = rest + hdr := msg.Header() // Ignore control messages. if hdr.Type < linux.NLMSG_MIN_TYPE { @@ -692,19 +683,10 @@ func (s *Socket) processMessages(ctx context.Context, buf []byte) *syserr.Error } ms := NewMessageSet(s.portID, hdr.Seq) - var err *syserr.Error - // TODO(b/68877377): ACKs not supported yet. - if hdr.Flags&linux.NLM_F_ACK == linux.NLM_F_ACK { - err = syserr.ErrNotSupported - } else { - - err = s.protocol.ProcessMessage(ctx, hdr, data, ms) - } - if err != nil { - ms = NewMessageSet(s.portID, hdr.Seq) - if err := s.dumpErrorMesage(ctx, hdr, ms, err); err != nil { - return err - } + if err := s.protocol.ProcessMessage(ctx, msg, ms); err != nil { + dumpErrorMesage(hdr, ms, err) + } else if hdr.Flags&linux.NLM_F_ACK == linux.NLM_F_ACK { + dumpAckMesage(hdr, ms) } if err := s.sendResponse(ctx, ms); err != nil { diff --git a/pkg/sentry/socket/netlink/uevent/protocol.go b/pkg/sentry/socket/netlink/uevent/protocol.go index 1ee4296bc..029ba21b5 100644 --- a/pkg/sentry/socket/netlink/uevent/protocol.go +++ b/pkg/sentry/socket/netlink/uevent/protocol.go @@ -49,7 +49,7 @@ func (p *Protocol) CanSend() bool { } // ProcessMessage implements netlink.Protocol.ProcessMessage. -func (p *Protocol) ProcessMessage(ctx context.Context, hdr linux.NetlinkMessageHeader, data []byte, ms *netlink.MessageSet) *syserr.Error { +func (p *Protocol) ProcessMessage(ctx context.Context, msg *netlink.Message, ms *netlink.MessageSet) *syserr.Error { // Silently ignore all messages. return nil } diff --git a/pkg/sentry/socket/netstack/stack.go b/pkg/sentry/socket/netstack/stack.go index 31ea66eca..0692482e9 100644 --- a/pkg/sentry/socket/netstack/stack.go +++ b/pkg/sentry/socket/netstack/stack.go @@ -20,6 +20,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/socket/netfilter" "gvisor.dev/gvisor/pkg/syserr" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/iptables" "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" @@ -88,6 +90,59 @@ func (s *Stack) InterfaceAddrs() map[int32][]inet.InterfaceAddr { return nicAddrs } +// AddInterfaceAddr implements inet.Stack.AddInterfaceAddr. +func (s *Stack) AddInterfaceAddr(idx int32, addr inet.InterfaceAddr) error { + var ( + protocol tcpip.NetworkProtocolNumber + address tcpip.Address + ) + switch addr.Family { + case linux.AF_INET: + if len(addr.Addr) < header.IPv4AddressSize { + return syserror.EINVAL + } + if addr.PrefixLen > header.IPv4AddressSize*8 { + return syserror.EINVAL + } + protocol = ipv4.ProtocolNumber + address = tcpip.Address(addr.Addr[:header.IPv4AddressSize]) + + case linux.AF_INET6: + if len(addr.Addr) < header.IPv6AddressSize { + return syserror.EINVAL + } + if addr.PrefixLen > header.IPv6AddressSize*8 { + return syserror.EINVAL + } + protocol = ipv6.ProtocolNumber + address = tcpip.Address(addr.Addr[:header.IPv6AddressSize]) + + default: + return syserror.ENOTSUP + } + + protocolAddress := tcpip.ProtocolAddress{ + Protocol: protocol, + AddressWithPrefix: tcpip.AddressWithPrefix{ + Address: address, + PrefixLen: int(addr.PrefixLen), + }, + } + + // Attach address to interface. + if err := s.Stack.AddProtocolAddressWithOptions(tcpip.NICID(idx), protocolAddress, stack.CanBePrimaryEndpoint); err != nil { + return syserr.TranslateNetstackError(err).ToError() + } + + // Add route for local network. + s.Stack.AddRoute(tcpip.Route{ + Destination: protocolAddress.AddressWithPrefix.Subnet(), + Gateway: "", // No gateway for local network. + NIC: tcpip.NICID(idx), + }) + return nil +} + // TCPReceiveBufferSize implements inet.Stack.TCPReceiveBufferSize. func (s *Stack) TCPReceiveBufferSize() (inet.TCPBufferSize, error) { var rs tcp.ReceiveBufferSizeOption diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index 7057b110e..b793f1d74 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -795,6 +795,8 @@ func (s *Stack) Forwarding() bool { // SetRouteTable assigns the route table to be used by this stack. It // specifies which NIC to use for given destination address ranges. +// +// This method takes ownership of the table. func (s *Stack) SetRouteTable(table []tcpip.Route) { s.mu.Lock() defer s.mu.Unlock() @@ -809,6 +811,13 @@ func (s *Stack) GetRouteTable() []tcpip.Route { return append([]tcpip.Route(nil), s.routeTable...) } +// AddRoute appends a route to the route table. +func (s *Stack) AddRoute(route tcpip.Route) { + s.mu.Lock() + defer s.mu.Unlock() + s.routeTable = append(s.routeTable, route) +} + // NewEndpoint creates a new transport layer endpoint of the given protocol. func (s *Stack) NewEndpoint(transport tcpip.TransportProtocolNumber, network tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { t, ok := s.transportProtocols[transport] diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 273b014d6..f2e3c7072 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -2769,9 +2769,11 @@ cc_binary( deps = [ ":socket_netlink_util", ":socket_test_util", + "//test/util:capability_util", "//test/util:cleanup", "//test/util:file_descriptor", "@com_google_absl//absl/strings:str_format", + "@com_google_absl//absl/types:optional", gtest, "//test/util:test_main", "//test/util:test_util", diff --git a/test/syscalls/linux/socket_netlink_route.cc b/test/syscalls/linux/socket_netlink_route.cc index 1e28e658d..e5aed1eec 100644 --- a/test/syscalls/linux/socket_netlink_route.cc +++ b/test/syscalls/linux/socket_netlink_route.cc @@ -14,6 +14,7 @@ #include #include +#include #include #include #include @@ -25,8 +26,10 @@ #include "gtest/gtest.h" #include "absl/strings/str_format.h" +#include "absl/types/optional.h" #include "test/syscalls/linux/socket_netlink_util.h" #include "test/syscalls/linux/socket_test_util.h" +#include "test/util/capability_util.h" #include "test/util/cleanup.h" #include "test/util/file_descriptor.h" #include "test/util/test_util.h" @@ -38,6 +41,8 @@ namespace testing { namespace { +constexpr uint32_t kSeq = 12345; + using ::testing::AnyOf; using ::testing::Eq; @@ -113,58 +118,224 @@ void CheckGetLinkResponse(const struct nlmsghdr* hdr, int seq, int port) { // TODO(mpratt): Check ifinfomsg contents and following attrs. } +PosixError DumpLinks( + const FileDescriptor& fd, uint32_t seq, + const std::function& fn) { + struct request { + struct nlmsghdr hdr; + struct ifinfomsg ifm; + }; + + struct request req = {}; + req.hdr.nlmsg_len = sizeof(req); + req.hdr.nlmsg_type = RTM_GETLINK; + req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP; + req.hdr.nlmsg_seq = seq; + req.ifm.ifi_family = AF_UNSPEC; + + return NetlinkRequestResponse(fd, &req, sizeof(req), fn, false); +} + TEST(NetlinkRouteTest, GetLinkDump) { FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); uint32_t port = ASSERT_NO_ERRNO_AND_VALUE(NetlinkPortID(fd.get())); + // Loopback is common among all tests, check that it's found. + bool loopbackFound = false; + ASSERT_NO_ERRNO(DumpLinks(fd, kSeq, [&](const struct nlmsghdr* hdr) { + CheckGetLinkResponse(hdr, kSeq, port); + if (hdr->nlmsg_type != RTM_NEWLINK) { + return; + } + ASSERT_GE(hdr->nlmsg_len, NLMSG_SPACE(sizeof(struct ifinfomsg))); + const struct ifinfomsg* msg = + reinterpret_cast(NLMSG_DATA(hdr)); + std::cout << "Found interface idx=" << msg->ifi_index + << ", type=" << std::hex << msg->ifi_type; + if (msg->ifi_type == ARPHRD_LOOPBACK) { + loopbackFound = true; + EXPECT_NE(msg->ifi_flags & IFF_LOOPBACK, 0); + } + })); + EXPECT_TRUE(loopbackFound); +} + +struct Link { + int index; + std::string name; +}; + +PosixErrorOr> FindLoopbackLink() { + ASSIGN_OR_RETURN_ERRNO(FileDescriptor fd, NetlinkBoundSocket(NETLINK_ROUTE)); + + absl::optional link; + RETURN_IF_ERRNO(DumpLinks(fd, kSeq, [&](const struct nlmsghdr* hdr) { + if (hdr->nlmsg_type != RTM_NEWLINK || + hdr->nlmsg_len < NLMSG_SPACE(sizeof(struct ifinfomsg))) { + return; + } + const struct ifinfomsg* msg = + reinterpret_cast(NLMSG_DATA(hdr)); + if (msg->ifi_type == ARPHRD_LOOPBACK) { + const auto* rta = FindRtAttr(hdr, msg, IFLA_IFNAME); + if (rta == nullptr) { + // Ignore links that do not have a name. + return; + } + + link = Link(); + link->index = msg->ifi_index; + link->name = std::string(reinterpret_cast(RTA_DATA(rta))); + } + })); + return link; +} + +// CheckLinkMsg checks a netlink message against an expected link. +void CheckLinkMsg(const struct nlmsghdr* hdr, const Link& link) { + ASSERT_THAT(hdr->nlmsg_type, Eq(RTM_NEWLINK)); + ASSERT_GE(hdr->nlmsg_len, NLMSG_SPACE(sizeof(struct ifinfomsg))); + const struct ifinfomsg* msg = + reinterpret_cast(NLMSG_DATA(hdr)); + EXPECT_EQ(msg->ifi_index, link.index); + + const struct rtattr* rta = FindRtAttr(hdr, msg, IFLA_IFNAME); + EXPECT_NE(nullptr, rta) << "IFLA_IFNAME not found in message."; + if (rta != nullptr) { + std::string name(reinterpret_cast(RTA_DATA(rta))); + EXPECT_EQ(name, link.name); + } +} + +TEST(NetlinkRouteTest, GetLinkByIndex) { + absl::optional loopback_link = + ASSERT_NO_ERRNO_AND_VALUE(FindLoopbackLink()); + ASSERT_TRUE(loopback_link.has_value()); + + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); + struct request { struct nlmsghdr hdr; struct ifinfomsg ifm; }; - constexpr uint32_t kSeq = 12345; - struct request req = {}; req.hdr.nlmsg_len = sizeof(req); req.hdr.nlmsg_type = RTM_GETLINK; - req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP; + req.hdr.nlmsg_flags = NLM_F_REQUEST; req.hdr.nlmsg_seq = kSeq; req.ifm.ifi_family = AF_UNSPEC; + req.ifm.ifi_index = loopback_link->index; - // Loopback is common among all tests, check that it's found. - bool loopbackFound = false; + bool found = false; ASSERT_NO_ERRNO(NetlinkRequestResponse( fd, &req, sizeof(req), [&](const struct nlmsghdr* hdr) { - CheckGetLinkResponse(hdr, kSeq, port); - if (hdr->nlmsg_type != RTM_NEWLINK) { - return; - } - ASSERT_GE(hdr->nlmsg_len, NLMSG_SPACE(sizeof(struct ifinfomsg))); - const struct ifinfomsg* msg = - reinterpret_cast(NLMSG_DATA(hdr)); - std::cout << "Found interface idx=" << msg->ifi_index - << ", type=" << std::hex << msg->ifi_type; - if (msg->ifi_type == ARPHRD_LOOPBACK) { - loopbackFound = true; - EXPECT_NE(msg->ifi_flags & IFF_LOOPBACK, 0); - } + CheckLinkMsg(hdr, *loopback_link); + found = true; }, false)); - EXPECT_TRUE(loopbackFound); + EXPECT_TRUE(found) << "Netlink response does not contain any links."; } -TEST(NetlinkRouteTest, MsgHdrMsgUnsuppType) { +TEST(NetlinkRouteTest, GetLinkByName) { + absl::optional loopback_link = + ASSERT_NO_ERRNO_AND_VALUE(FindLoopbackLink()); + ASSERT_TRUE(loopback_link.has_value()); + FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); struct request { struct nlmsghdr hdr; struct ifinfomsg ifm; + struct rtattr rtattr; + char ifname[IFNAMSIZ]; + char pad[NLMSG_ALIGNTO + RTA_ALIGNTO]; }; - constexpr uint32_t kSeq = 12345; + struct request req = {}; + req.hdr.nlmsg_type = RTM_GETLINK; + req.hdr.nlmsg_flags = NLM_F_REQUEST; + req.hdr.nlmsg_seq = kSeq; + req.ifm.ifi_family = AF_UNSPEC; + req.rtattr.rta_type = IFLA_IFNAME; + req.rtattr.rta_len = RTA_LENGTH(loopback_link->name.size() + 1); + strncpy(req.ifname, loopback_link->name.c_str(), sizeof(req.ifname)); + req.hdr.nlmsg_len = + NLMSG_LENGTH(sizeof(req.ifm)) + NLMSG_ALIGN(req.rtattr.rta_len); + + bool found = false; + ASSERT_NO_ERRNO(NetlinkRequestResponse( + fd, &req, sizeof(req), + [&](const struct nlmsghdr* hdr) { + CheckLinkMsg(hdr, *loopback_link); + found = true; + }, + false)); + EXPECT_TRUE(found) << "Netlink response does not contain any links."; +} + +TEST(NetlinkRouteTest, GetLinkByIndexNotFound) { + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); + + struct request { + struct nlmsghdr hdr; + struct ifinfomsg ifm; + }; + + struct request req = {}; + req.hdr.nlmsg_len = sizeof(req); + req.hdr.nlmsg_type = RTM_GETLINK; + req.hdr.nlmsg_flags = NLM_F_REQUEST; + req.hdr.nlmsg_seq = kSeq; + req.ifm.ifi_family = AF_UNSPEC; + req.ifm.ifi_index = 1234590; + + EXPECT_THAT(NetlinkRequestAckOrError(fd, kSeq, &req, sizeof(req)), + PosixErrorIs(ENODEV, ::testing::_)); +} + +TEST(NetlinkRouteTest, GetLinkByNameNotFound) { + const std::string name = "nodevice?!"; + + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); + + struct request { + struct nlmsghdr hdr; + struct ifinfomsg ifm; + struct rtattr rtattr; + char ifname[IFNAMSIZ]; + char pad[NLMSG_ALIGNTO + RTA_ALIGNTO]; + }; + + struct request req = {}; + req.hdr.nlmsg_type = RTM_GETLINK; + req.hdr.nlmsg_flags = NLM_F_REQUEST; + req.hdr.nlmsg_seq = kSeq; + req.ifm.ifi_family = AF_UNSPEC; + req.rtattr.rta_type = IFLA_IFNAME; + req.rtattr.rta_len = RTA_LENGTH(name.size() + 1); + strncpy(req.ifname, name.c_str(), sizeof(req.ifname)); + req.hdr.nlmsg_len = + NLMSG_LENGTH(sizeof(req.ifm)) + NLMSG_ALIGN(req.rtattr.rta_len); + + EXPECT_THAT(NetlinkRequestAckOrError(fd, kSeq, &req, sizeof(req)), + PosixErrorIs(ENODEV, ::testing::_)); +} + +TEST(NetlinkRouteTest, MsgHdrMsgUnsuppType) { + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); + + struct request { + struct nlmsghdr hdr; + struct ifinfomsg ifm; + }; struct request req = {}; req.hdr.nlmsg_len = sizeof(req); @@ -175,18 +346,8 @@ TEST(NetlinkRouteTest, MsgHdrMsgUnsuppType) { req.hdr.nlmsg_seq = kSeq; req.ifm.ifi_family = AF_UNSPEC; - ASSERT_NO_ERRNO(NetlinkRequestResponse( - fd, &req, sizeof(req), - [&](const struct nlmsghdr* hdr) { - EXPECT_THAT(hdr->nlmsg_type, Eq(NLMSG_ERROR)); - EXPECT_EQ(hdr->nlmsg_seq, kSeq); - EXPECT_GE(hdr->nlmsg_len, sizeof(*hdr) + sizeof(struct nlmsgerr)); - - const struct nlmsgerr* msg = - reinterpret_cast(NLMSG_DATA(hdr)); - EXPECT_EQ(msg->error, -EOPNOTSUPP); - }, - true)); + EXPECT_THAT(NetlinkRequestAckOrError(fd, kSeq, &req, sizeof(req)), + PosixErrorIs(EOPNOTSUPP, ::testing::_)); } TEST(NetlinkRouteTest, MsgHdrMsgTrunc) { @@ -198,8 +359,6 @@ TEST(NetlinkRouteTest, MsgHdrMsgTrunc) { struct ifinfomsg ifm; }; - constexpr uint32_t kSeq = 12345; - struct request req = {}; req.hdr.nlmsg_len = sizeof(req); req.hdr.nlmsg_type = RTM_GETLINK; @@ -238,8 +397,6 @@ TEST(NetlinkRouteTest, MsgTruncMsgHdrMsgTrunc) { struct ifinfomsg ifm; }; - constexpr uint32_t kSeq = 12345; - struct request req = {}; req.hdr.nlmsg_len = sizeof(req); req.hdr.nlmsg_type = RTM_GETLINK; @@ -282,8 +439,6 @@ TEST(NetlinkRouteTest, ControlMessageIgnored) { struct ifinfomsg ifm; }; - constexpr uint32_t kSeq = 12345; - struct request req = {}; // This control message is ignored. We still receive a response for the @@ -317,8 +472,6 @@ TEST(NetlinkRouteTest, GetAddrDump) { struct rtgenmsg rgm; }; - constexpr uint32_t kSeq = 12345; - struct request req; req.hdr.nlmsg_len = sizeof(req); req.hdr.nlmsg_type = RTM_GETADDR; @@ -367,6 +520,57 @@ TEST(NetlinkRouteTest, LookupAll) { ASSERT_GT(count, 0); } +TEST(NetlinkRouteTest, AddAddr) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN))); + + absl::optional loopback_link = + ASSERT_NO_ERRNO_AND_VALUE(FindLoopbackLink()); + ASSERT_TRUE(loopback_link.has_value()); + + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); + + struct request { + struct nlmsghdr hdr; + struct ifaddrmsg ifa; + struct rtattr rtattr; + struct in_addr addr; + char pad[NLMSG_ALIGNTO + RTA_ALIGNTO]; + }; + + struct request req = {}; + req.hdr.nlmsg_type = RTM_NEWADDR; + req.hdr.nlmsg_seq = kSeq; + req.ifa.ifa_family = AF_INET; + req.ifa.ifa_prefixlen = 24; + req.ifa.ifa_flags = 0; + req.ifa.ifa_scope = 0; + req.ifa.ifa_index = loopback_link->index; + req.rtattr.rta_type = IFA_LOCAL; + req.rtattr.rta_len = RTA_LENGTH(sizeof(req.addr)); + inet_pton(AF_INET, "10.0.0.1", &req.addr); + req.hdr.nlmsg_len = + NLMSG_LENGTH(sizeof(req.ifa)) + NLMSG_ALIGN(req.rtattr.rta_len); + + // Create should succeed, as no such address in kernel. + req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_CREATE | NLM_F_ACK; + EXPECT_NO_ERRNO( + NetlinkRequestAckOrError(fd, req.hdr.nlmsg_seq, &req, req.hdr.nlmsg_len)); + + // Replace an existing address should succeed. + req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_REPLACE | NLM_F_ACK; + req.hdr.nlmsg_seq++; + EXPECT_NO_ERRNO( + NetlinkRequestAckOrError(fd, req.hdr.nlmsg_seq, &req, req.hdr.nlmsg_len)); + + // Create exclusive should fail, as we created the address above. + req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_CREATE | NLM_F_EXCL | NLM_F_ACK; + req.hdr.nlmsg_seq++; + EXPECT_THAT( + NetlinkRequestAckOrError(fd, req.hdr.nlmsg_seq, &req, req.hdr.nlmsg_len), + PosixErrorIs(EEXIST, ::testing::_)); +} + // GetRouteDump tests a RTM_GETROUTE + NLM_F_DUMP request. TEST(NetlinkRouteTest, GetRouteDump) { FileDescriptor fd = @@ -378,8 +582,6 @@ TEST(NetlinkRouteTest, GetRouteDump) { struct rtmsg rtm; }; - constexpr uint32_t kSeq = 12345; - struct request req = {}; req.hdr.nlmsg_len = sizeof(req); req.hdr.nlmsg_type = RTM_GETROUTE; @@ -538,8 +740,6 @@ TEST(NetlinkRouteTest, RecvmsgTrunc) { struct rtgenmsg rgm; }; - constexpr uint32_t kSeq = 12345; - struct request req; req.hdr.nlmsg_len = sizeof(req); req.hdr.nlmsg_type = RTM_GETADDR; @@ -615,8 +815,6 @@ TEST(NetlinkRouteTest, RecvmsgTruncPeek) { struct rtgenmsg rgm; }; - constexpr uint32_t kSeq = 12345; - struct request req; req.hdr.nlmsg_len = sizeof(req); req.hdr.nlmsg_type = RTM_GETADDR; @@ -695,8 +893,6 @@ TEST(NetlinkRouteTest, NoPasscredNoCreds) { struct rtgenmsg rgm; }; - constexpr uint32_t kSeq = 12345; - struct request req; req.hdr.nlmsg_len = sizeof(req); req.hdr.nlmsg_type = RTM_GETADDR; @@ -743,8 +939,6 @@ TEST(NetlinkRouteTest, PasscredCreds) { struct rtgenmsg rgm; }; - constexpr uint32_t kSeq = 12345; - struct request req; req.hdr.nlmsg_len = sizeof(req); req.hdr.nlmsg_type = RTM_GETADDR; diff --git a/test/syscalls/linux/socket_netlink_util.cc b/test/syscalls/linux/socket_netlink_util.cc index cd2212a1a..952eecfe8 100644 --- a/test/syscalls/linux/socket_netlink_util.cc +++ b/test/syscalls/linux/socket_netlink_util.cc @@ -16,6 +16,7 @@ #include #include +#include #include #include @@ -71,9 +72,10 @@ PosixError NetlinkRequestResponse( iov.iov_base = buf.data(); iov.iov_len = buf.size(); - // Response is a series of NLM_F_MULTI messages, ending with a NLMSG_DONE - // message. + // If NLM_F_MULTI is set, response is a series of messages that ends with a + // NLMSG_DONE message. int type = -1; + int flags = 0; do { int len; RETURN_ERROR_IF_SYSCALL_FAIL(len = RetryEINTR(recvmsg)(fd.get(), &msg, 0)); @@ -89,6 +91,7 @@ PosixError NetlinkRequestResponse( for (struct nlmsghdr* hdr = reinterpret_cast(buf.data()); NLMSG_OK(hdr, len); hdr = NLMSG_NEXT(hdr, len)) { fn(hdr); + flags = hdr->nlmsg_flags; type = hdr->nlmsg_type; // Done should include an integer payload for dump_done_errno. // See net/netlink/af_netlink.c:netlink_dump @@ -98,11 +101,11 @@ PosixError NetlinkRequestResponse( EXPECT_GE(hdr->nlmsg_len, NLMSG_LENGTH(sizeof(int))); } } - } while (type != NLMSG_DONE && type != NLMSG_ERROR); + } while ((flags & NLM_F_MULTI) && type != NLMSG_DONE && type != NLMSG_ERROR); if (expect_nlmsgerr) { EXPECT_EQ(type, NLMSG_ERROR); - } else { + } else if (flags & NLM_F_MULTI) { EXPECT_EQ(type, NLMSG_DONE); } return NoError(); @@ -146,5 +149,39 @@ PosixError NetlinkRequestResponseSingle( return NoError(); } +PosixError NetlinkRequestAckOrError(const FileDescriptor& fd, uint32_t seq, + void* request, size_t len) { + // Dummy negative number for no error message received. + // We won't get a negative error number so there will be no confusion. + int err = -42; + RETURN_IF_ERRNO(NetlinkRequestResponse( + fd, request, len, + [&](const struct nlmsghdr* hdr) { + EXPECT_EQ(NLMSG_ERROR, hdr->nlmsg_type); + EXPECT_EQ(hdr->nlmsg_seq, seq); + EXPECT_GE(hdr->nlmsg_len, sizeof(*hdr) + sizeof(struct nlmsgerr)); + + const struct nlmsgerr* msg = + reinterpret_cast(NLMSG_DATA(hdr)); + err = -msg->error; + }, + true)); + return PosixError(err); +} + +const struct rtattr* FindRtAttr(const struct nlmsghdr* hdr, + const struct ifinfomsg* msg, int16_t attr) { + const int ifi_space = NLMSG_SPACE(sizeof(*msg)); + int attrlen = hdr->nlmsg_len - ifi_space; + const struct rtattr* rta = reinterpret_cast( + reinterpret_cast(hdr) + NLMSG_ALIGN(ifi_space)); + for (; RTA_OK(rta, attrlen); rta = RTA_NEXT(rta, attrlen)) { + if (rta->rta_type == attr) { + return rta; + } + } + return nullptr; +} + } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_netlink_util.h b/test/syscalls/linux/socket_netlink_util.h index 3678c0599..e13ead406 100644 --- a/test/syscalls/linux/socket_netlink_util.h +++ b/test/syscalls/linux/socket_netlink_util.h @@ -19,6 +19,7 @@ // socket.h has to be included before if_arp.h. #include #include +#include #include "test/util/file_descriptor.h" #include "test/util/posix_error.h" @@ -47,6 +48,14 @@ PosixError NetlinkRequestResponseSingle( const FileDescriptor& fd, void* request, size_t len, const std::function& fn); +// Send the passed request then expect and return an ack or error. +PosixError NetlinkRequestAckOrError(const FileDescriptor& fd, uint32_t seq, + void* request, size_t len); + +// Find rtnetlink attribute in message. +const struct rtattr* FindRtAttr(const struct nlmsghdr* hdr, + const struct ifinfomsg* msg, int16_t attr); + } // namespace testing } // namespace gvisor -- cgit v1.2.3 From eea0eeee933ba8406ae688fce4348271f9513514 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Wed, 5 Feb 2020 11:25:10 -0800 Subject: Disable get/set xattrs until list/remove exist too. PiperOrigin-RevId: 293411655 --- pkg/sentry/syscalls/linux/linux64_amd64.go | 27 ++++--- pkg/sentry/syscalls/linux/linux64_arm64.go | 37 +++++---- test/syscalls/linux/xattr.cc | 124 +++++++++++++++++++++++++++++ 3 files changed, 159 insertions(+), 29 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/linux64_amd64.go b/pkg/sentry/syscalls/linux/linux64_amd64.go index 7435b50bf..588f8b087 100644 --- a/pkg/sentry/syscalls/linux/linux64_amd64.go +++ b/pkg/sentry/syscalls/linux/linux64_amd64.go @@ -228,18 +228,21 @@ var AMD64 = &kernel.SyscallTable{ 185: syscalls.Error("security", syserror.ENOSYS, "Not implemented in Linux.", nil), 186: syscalls.Supported("gettid", Gettid), 187: syscalls.Supported("readahead", Readahead), - 188: syscalls.PartiallySupported("setxattr", SetXattr, "Only supported for tmpfs.", nil), - 189: syscalls.PartiallySupported("lsetxattr", LSetXattr, "Only supported for tmpfs.", nil), - 190: syscalls.PartiallySupported("fsetxattr", FSetXattr, "Only supported for tmpfs.", nil), - 191: syscalls.PartiallySupported("getxattr", GetXattr, "Only supported for tmpfs.", nil), - 192: syscalls.PartiallySupported("lgetxattr", LGetXattr, "Only supported for tmpfs.", nil), - 193: syscalls.PartiallySupported("fgetxattr", FGetXattr, "Only supported for tmpfs.", nil), - 194: syscalls.ErrorWithEvent("listxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 195: syscalls.ErrorWithEvent("llistxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 196: syscalls.ErrorWithEvent("flistxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 197: syscalls.ErrorWithEvent("removexattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 198: syscalls.ErrorWithEvent("lremovexattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 199: syscalls.ErrorWithEvent("fremovexattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + // TODO(b/148303075): Enable set/getxattr (in their various + // forms) once we also have list and removexattr. The JVM + // assumes that if get/set exist, then list and remove do too. + 188: syscalls.ErrorWithEvent("setxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 189: syscalls.ErrorWithEvent("lsetxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 190: syscalls.ErrorWithEvent("fsetxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 191: syscalls.ErrorWithEvent("getxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 192: syscalls.ErrorWithEvent("lgetxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 193: syscalls.ErrorWithEvent("fgetxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 194: syscalls.ErrorWithEvent("listxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 195: syscalls.ErrorWithEvent("llistxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 196: syscalls.ErrorWithEvent("flistxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 197: syscalls.ErrorWithEvent("removexattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 198: syscalls.ErrorWithEvent("lremovexattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 199: syscalls.ErrorWithEvent("fremovexattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), 200: syscalls.Supported("tkill", Tkill), 201: syscalls.Supported("time", Time), 202: syscalls.PartiallySupported("futex", Futex, "Robust futexes not supported.", nil), diff --git a/pkg/sentry/syscalls/linux/linux64_arm64.go b/pkg/sentry/syscalls/linux/linux64_arm64.go index 03a39fe65..06e5ee401 100644 --- a/pkg/sentry/syscalls/linux/linux64_arm64.go +++ b/pkg/sentry/syscalls/linux/linux64_arm64.go @@ -36,23 +36,26 @@ var ARM64 = &kernel.SyscallTable{ }, AuditNumber: linux.AUDIT_ARCH_AARCH64, Table: map[uintptr]kernel.Syscall{ - 0: syscalls.PartiallySupported("io_setup", IoSetup, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 1: syscalls.PartiallySupported("io_destroy", IoDestroy, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 2: syscalls.PartiallySupported("io_submit", IoSubmit, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 3: syscalls.PartiallySupported("io_cancel", IoCancel, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 4: syscalls.PartiallySupported("io_getevents", IoGetevents, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 5: syscalls.PartiallySupported("setxattr", SetXattr, "Only supported for tmpfs.", nil), - 6: syscalls.PartiallySupported("lsetxattr", LSetXattr, "Only supported for tmpfs.", nil), - 7: syscalls.PartiallySupported("fsetxattr", FSetXattr, "Only supported for tmpfs.", nil), - 8: syscalls.PartiallySupported("getxattr", GetXattr, "Only supported for tmpfs.", nil), - 9: syscalls.PartiallySupported("lgetxattr", LGetXattr, "Only supported for tmpfs.", nil), - 10: syscalls.PartiallySupported("fgetxattr", FGetXattr, "Only supported for tmpfs.", nil), - 11: syscalls.ErrorWithEvent("listxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 12: syscalls.ErrorWithEvent("llistxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 13: syscalls.ErrorWithEvent("flistxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 14: syscalls.ErrorWithEvent("removexattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 15: syscalls.ErrorWithEvent("lremovexattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 16: syscalls.ErrorWithEvent("fremovexattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 0: syscalls.PartiallySupported("io_setup", IoSetup, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 1: syscalls.PartiallySupported("io_destroy", IoDestroy, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 2: syscalls.PartiallySupported("io_submit", IoSubmit, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 3: syscalls.PartiallySupported("io_cancel", IoCancel, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 4: syscalls.PartiallySupported("io_getevents", IoGetevents, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + // TODO(b/148303075): Enable set/getxattr (in their various + // forms) once we also have list and removexattr. The JVM + // assumes that if get/set exist, then list and remove do too. + 5: syscalls.ErrorWithEvent("setxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 6: syscalls.ErrorWithEvent("lsetxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 7: syscalls.ErrorWithEvent("fsetxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 8: syscalls.ErrorWithEvent("getxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 9: syscalls.ErrorWithEvent("lgetxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 10: syscalls.ErrorWithEvent("fgetxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 11: syscalls.ErrorWithEvent("listxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 13: syscalls.ErrorWithEvent("llistxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 13: syscalls.ErrorWithEvent("flistxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 14: syscalls.ErrorWithEvent("removexattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 15: syscalls.ErrorWithEvent("lremovexattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 16: syscalls.ErrorWithEvent("fremovexattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), 17: syscalls.Supported("getcwd", Getcwd), 18: syscalls.CapError("lookup_dcookie", linux.CAP_SYS_ADMIN, "", nil), 19: syscalls.Supported("eventfd2", Eventfd2), diff --git a/test/syscalls/linux/xattr.cc b/test/syscalls/linux/xattr.cc index ab21d68c6..85eb31847 100644 --- a/test/syscalls/linux/xattr.cc +++ b/test/syscalls/linux/xattr.cc @@ -39,6 +39,10 @@ namespace { class XattrTest : public FileTest {}; TEST_F(XattrTest, XattrNullName) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); EXPECT_THAT(setxattr(path, nullptr, nullptr, 0, /*flags=*/0), @@ -48,6 +52,10 @@ TEST_F(XattrTest, XattrNullName) { } TEST_F(XattrTest, XattrEmptyName) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); EXPECT_THAT(setxattr(path, "", nullptr, 0, /*flags=*/0), @@ -56,6 +64,10 @@ TEST_F(XattrTest, XattrEmptyName) { } TEST_F(XattrTest, XattrLargeName) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); std::string name = "user."; name += std::string(XATTR_NAME_MAX - name.length(), 'a'); @@ -77,6 +89,10 @@ TEST_F(XattrTest, XattrLargeName) { } TEST_F(XattrTest, XattrInvalidPrefix) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); std::string name(XATTR_NAME_MAX, 'a'); EXPECT_THAT(setxattr(path, name.c_str(), nullptr, 0, /*flags=*/0), @@ -88,6 +104,10 @@ TEST_F(XattrTest, XattrInvalidPrefix) { // Do not allow save/restore cycles after making the test file read-only, as // the restore will fail to open it with r/w permissions. TEST_F(XattrTest, XattrReadOnly_NoRandomSave) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + // Drop capabilities that allow us to override file and directory permissions. ASSERT_NO_ERRNO(SetCapability(CAP_DAC_OVERRIDE, false)); ASSERT_NO_ERRNO(SetCapability(CAP_DAC_READ_SEARCH, false)); @@ -113,6 +133,10 @@ TEST_F(XattrTest, XattrReadOnly_NoRandomSave) { // Do not allow save/restore cycles after making the test file write-only, as // the restore will fail to open it with r/w permissions. TEST_F(XattrTest, XattrWriteOnly_NoRandomSave) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + // Drop capabilities that allow us to override file and directory permissions. ASSERT_NO_ERRNO(SetCapability(CAP_DAC_OVERRIDE, false)); ASSERT_NO_ERRNO(SetCapability(CAP_DAC_READ_SEARCH, false)); @@ -143,6 +167,10 @@ TEST_F(XattrTest, XattrTrustedWithNonadmin) { } TEST_F(XattrTest, XattrOnDirectory) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); const char name[] = "user.test"; EXPECT_THAT(setxattr(dir.path().c_str(), name, NULL, 0, /*flags=*/0), @@ -152,6 +180,10 @@ TEST_F(XattrTest, XattrOnDirectory) { } TEST_F(XattrTest, XattrOnSymlink) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); TempPath link = ASSERT_NO_ERRNO_AND_VALUE( TempPath::CreateSymlinkTo(dir.path(), test_file_name_)); @@ -163,6 +195,10 @@ TEST_F(XattrTest, XattrOnSymlink) { } TEST_F(XattrTest, XattrOnInvalidFileTypes) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char name[] = "user.test"; char char_device[] = "/dev/zero"; @@ -181,6 +217,10 @@ TEST_F(XattrTest, XattrOnInvalidFileTypes) { } TEST_F(XattrTest, SetxattrSizeSmallerThanValue) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; std::vector val = {'a', 'a'}; @@ -196,6 +236,10 @@ TEST_F(XattrTest, SetxattrSizeSmallerThanValue) { } TEST_F(XattrTest, SetxattrZeroSize) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; char val = 'a'; @@ -208,6 +252,10 @@ TEST_F(XattrTest, SetxattrZeroSize) { } TEST_F(XattrTest, SetxattrSizeTooLarge) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; @@ -223,6 +271,10 @@ TEST_F(XattrTest, SetxattrSizeTooLarge) { } TEST_F(XattrTest, SetxattrNullValueAndNonzeroSize) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; EXPECT_THAT(setxattr(path, name, nullptr, 1, /*flags=*/0), @@ -232,6 +284,10 @@ TEST_F(XattrTest, SetxattrNullValueAndNonzeroSize) { } TEST_F(XattrTest, SetxattrNullValueAndZeroSize) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; EXPECT_THAT(setxattr(path, name, nullptr, 0, /*flags=*/0), SyscallSucceeds()); @@ -240,6 +296,10 @@ TEST_F(XattrTest, SetxattrNullValueAndZeroSize) { } TEST_F(XattrTest, SetxattrValueTooLargeButOKSize) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; std::vector val(XATTR_SIZE_MAX + 1); @@ -256,6 +316,10 @@ TEST_F(XattrTest, SetxattrValueTooLargeButOKSize) { } TEST_F(XattrTest, SetxattrReplaceWithSmaller) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; std::vector val = {'a', 'a'}; @@ -271,6 +335,10 @@ TEST_F(XattrTest, SetxattrReplaceWithSmaller) { } TEST_F(XattrTest, SetxattrReplaceWithLarger) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; std::vector val = {'a', 'a'}; @@ -285,6 +353,10 @@ TEST_F(XattrTest, SetxattrReplaceWithLarger) { } TEST_F(XattrTest, SetxattrCreateFlag) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; EXPECT_THAT(setxattr(path, name, nullptr, 0, XATTR_CREATE), @@ -296,6 +368,10 @@ TEST_F(XattrTest, SetxattrCreateFlag) { } TEST_F(XattrTest, SetxattrReplaceFlag) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; EXPECT_THAT(setxattr(path, name, nullptr, 0, XATTR_REPLACE), @@ -308,6 +384,10 @@ TEST_F(XattrTest, SetxattrReplaceFlag) { } TEST_F(XattrTest, SetxattrInvalidFlags) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); int invalid_flags = 0xff; EXPECT_THAT(setxattr(path, nullptr, nullptr, 0, invalid_flags), @@ -315,6 +395,10 @@ TEST_F(XattrTest, SetxattrInvalidFlags) { } TEST_F(XattrTest, Getxattr) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; int val = 1234; @@ -327,6 +411,10 @@ TEST_F(XattrTest, Getxattr) { } TEST_F(XattrTest, GetxattrSizeSmallerThanValue) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; std::vector val = {'a', 'a'}; @@ -339,6 +427,10 @@ TEST_F(XattrTest, GetxattrSizeSmallerThanValue) { } TEST_F(XattrTest, GetxattrSizeLargerThanValue) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; char val = 'a'; @@ -354,6 +446,10 @@ TEST_F(XattrTest, GetxattrSizeLargerThanValue) { } TEST_F(XattrTest, GetxattrZeroSize) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; char val = 'a'; @@ -367,6 +463,10 @@ TEST_F(XattrTest, GetxattrZeroSize) { } TEST_F(XattrTest, GetxattrSizeTooLarge) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; char val = 'a'; @@ -383,6 +483,10 @@ TEST_F(XattrTest, GetxattrSizeTooLarge) { } TEST_F(XattrTest, GetxattrNullValue) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; char val = 'a'; @@ -394,6 +498,10 @@ TEST_F(XattrTest, GetxattrNullValue) { } TEST_F(XattrTest, GetxattrNullValueAndZeroSize) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; char val = 'a'; @@ -410,12 +518,20 @@ TEST_F(XattrTest, GetxattrNullValueAndZeroSize) { } TEST_F(XattrTest, GetxattrNonexistentName) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; EXPECT_THAT(getxattr(path, name, nullptr, 0), SyscallFailsWithErrno(ENODATA)); } TEST_F(XattrTest, LGetSetxattrOnSymlink) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); TempPath link = ASSERT_NO_ERRNO_AND_VALUE( TempPath::CreateSymlinkTo(dir.path(), test_file_name_)); @@ -427,6 +543,10 @@ TEST_F(XattrTest, LGetSetxattrOnSymlink) { } TEST_F(XattrTest, LGetSetxattrOnNonsymlink) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); const char name[] = "user.test"; int val = 1234; @@ -441,6 +561,10 @@ TEST_F(XattrTest, LGetSetxattrOnNonsymlink) { } TEST_F(XattrTest, FGetSetxattr) { + // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are + // supported, and get/set have been added pack to the syscall table. + SKIP_IF(IsRunningOnGvisor()); + const FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(test_file_name_.c_str(), 0)); const char name[] = "user.test"; -- cgit v1.2.3 From f3d95607036b8a502c65aa7b3e8145227274dbbc Mon Sep 17 00:00:00 2001 From: Eyal Soha Date: Wed, 5 Feb 2020 17:56:00 -0800 Subject: recv() on a closed TCP socket returns ENOTCONN From RFC 793 s3.9 p58 Event Processing: If RECEIVE Call arrives in CLOSED state and the user has access to such a connection, the return should be "error: connection does not exist" Fixes #1598 PiperOrigin-RevId: 293494287 --- pkg/sentry/socket/netstack/netstack.go | 7 ++++++- pkg/tcpip/tcpip.go | 4 ++++ pkg/tcpip/transport/tcp/endpoint.go | 4 ++-- pkg/tcpip/transport/tcp/tcp_test.go | 9 ++++----- test/syscalls/linux/tcp_socket.cc | 9 +++++++++ 5 files changed, 25 insertions(+), 8 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 049d04bf2..ed2fbcceb 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -2229,11 +2229,16 @@ func (s *SocketOperations) coalescingRead(ctx context.Context, dst usermem.IOSeq var copied int // Copy as many views as possible into the user-provided buffer. - for dst.NumBytes() != 0 { + for { + // Always do at least one fetchReadView, even if the number of bytes to + // read is 0. err = s.fetchReadView() if err != nil { break } + if dst.NumBytes() == 0 { + break + } var n int var e error diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 0fa141d58..d29d9a704 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -1124,6 +1124,10 @@ type ReadErrors struct { // InvalidEndpointState is the number of times we found the endpoint state // to be unexpected. InvalidEndpointState StatCounter + + // NotConnected is the number of times we tried to read but found that the + // endpoint was not connected. + NotConnected StatCounter } // WriteErrors collects packet write errors from an endpoint write call. diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index b5a8e15ee..e4a6b1b8b 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -1003,8 +1003,8 @@ func (e *endpoint) Read(*tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, if s == StateError { return buffer.View{}, tcpip.ControlMessages{}, he } - e.stats.ReadErrors.InvalidEndpointState.Increment() - return buffer.View{}, tcpip.ControlMessages{}, tcpip.ErrInvalidEndpointState + e.stats.ReadErrors.NotConnected.Increment() + return buffer.View{}, tcpip.ControlMessages{}, tcpip.ErrNotConnected } v, err := e.readLocked() diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index 2c1505067..cc118c993 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -5405,12 +5405,11 @@ func TestEndpointBindListenAcceptState(t *testing.T) { t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) } - // Expect InvalidEndpointState errors on a read at this point. - if _, _, err := ep.Read(nil); err != tcpip.ErrInvalidEndpointState { - t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrInvalidEndpointState) + if _, _, err := ep.Read(nil); err != tcpip.ErrNotConnected { + t.Errorf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrNotConnected) } - if got := ep.Stats().(*tcp.Stats).ReadErrors.InvalidEndpointState.Value(); got != 1 { - t.Fatalf("got EP stats Stats.ReadErrors.InvalidEndpointState got %v want %v", got, 1) + if got := ep.Stats().(*tcp.Stats).ReadErrors.NotConnected.Value(); got != 1 { + t.Errorf("got EP stats Stats.ReadErrors.NotConnected got %v want %v", got, 1) } if err := ep.Listen(10); err != nil { diff --git a/test/syscalls/linux/tcp_socket.cc b/test/syscalls/linux/tcp_socket.cc index 525ccbd88..8a8b68e75 100644 --- a/test/syscalls/linux/tcp_socket.cc +++ b/test/syscalls/linux/tcp_socket.cc @@ -1339,6 +1339,15 @@ TEST_P(SimpleTcpSocketTest, SetTCPDeferAcceptGreaterThanZero) { EXPECT_EQ(get, kTCPDeferAccept); } +TEST_P(SimpleTcpSocketTest, RecvOnClosedSocket) { + auto s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + char buf[1]; + EXPECT_THAT(recv(s.get(), buf, 0, 0), SyscallFailsWithErrno(ENOTCONN)); + EXPECT_THAT(recv(s.get(), buf, sizeof(buf), 0), + SyscallFailsWithErrno(ENOTCONN)); +} + INSTANTIATE_TEST_SUITE_P(AllInetTests, SimpleTcpSocketTest, ::testing::Values(AF_INET, AF_INET6)); -- cgit v1.2.3 From 1b6a12a768216a99a5e0428c42ea4faf79cf3b50 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Wed, 5 Feb 2020 22:45:44 -0800 Subject: Add notes to relevant tests. These were out-of-band notes that can help provide additional context and simplify automated imports. PiperOrigin-RevId: 293525915 --- pkg/metric/metric.go | 1 - pkg/sentry/arch/arch_x86.go | 4 ++ pkg/sentry/arch/signal_amd64.go | 2 +- pkg/sentry/fs/file_overlay_test.go | 1 + pkg/sentry/fs/proc/README.md | 4 ++ pkg/sentry/kernel/BUILD | 1 + pkg/sentry/kernel/kernel.go | 3 ++ pkg/sentry/kernel/kernel_opts.go | 20 +++++++ pkg/sentry/socket/hostinet/BUILD | 1 + pkg/sentry/socket/hostinet/socket.go | 5 +- pkg/sentry/socket/hostinet/sockopt_impl.go | 27 ++++++++++ pkg/tcpip/transport/tcp/endpoint.go | 3 ++ runsc/boot/filter/BUILD | 1 + runsc/boot/filter/config.go | 13 ----- runsc/boot/filter/config_profile.go | 34 ++++++++++++ runsc/container/console_test.go | 5 +- runsc/dockerutil/dockerutil.go | 11 ++-- runsc/testutil/BUILD | 5 +- runsc/testutil/testutil.go | 54 ------------------- runsc/testutil/testutil_runfiles.go | 75 +++++++++++++++++++++++++++ test/image/image_test.go | 8 +-- test/syscalls/build_defs.bzl | 35 +++++++++++-- test/syscalls/linux/chroot.cc | 2 +- test/syscalls/linux/concurrency.cc | 3 +- test/syscalls/linux/exec_proc_exe_workload.cc | 6 +++ test/syscalls/linux/fork.cc | 5 +- test/syscalls/linux/mmap.cc | 8 +-- test/syscalls/linux/open_create.cc | 1 + test/syscalls/linux/preadv.cc | 1 + test/syscalls/linux/proc.cc | 46 +++++++++++++--- test/syscalls/linux/readv.cc | 4 +- test/syscalls/linux/rseq.cc | 2 +- test/syscalls/linux/select.cc | 2 +- test/syscalls/linux/shm.cc | 2 +- test/syscalls/linux/sigprocmask.cc | 2 +- test/syscalls/linux/socket_unix_non_stream.cc | 4 +- test/syscalls/linux/symlink.cc | 2 +- test/syscalls/linux/tcp_socket.cc | 3 +- test/syscalls/linux/time.cc | 1 + test/syscalls/linux/tkill.cc | 2 +- test/util/temp_path.cc | 1 + tools/build/tags.bzl | 4 ++ tools/defs.bzl | 17 +++++- 43 files changed, 318 insertions(+), 113 deletions(-) create mode 100644 pkg/sentry/kernel/kernel_opts.go create mode 100644 pkg/sentry/socket/hostinet/sockopt_impl.go create mode 100644 runsc/boot/filter/config_profile.go create mode 100644 runsc/testutil/testutil_runfiles.go (limited to 'pkg/sentry') diff --git a/pkg/metric/metric.go b/pkg/metric/metric.go index 93d4f2b8c..006fcd9ab 100644 --- a/pkg/metric/metric.go +++ b/pkg/metric/metric.go @@ -46,7 +46,6 @@ var ( // // TODO(b/67298402): Support non-cumulative metrics. // TODO(b/67298427): Support metric fields. -// type Uint64Metric struct { // value is the actual value of the metric. It must be accessed // atomically. diff --git a/pkg/sentry/arch/arch_x86.go b/pkg/sentry/arch/arch_x86.go index a18093155..3db8bd34b 100644 --- a/pkg/sentry/arch/arch_x86.go +++ b/pkg/sentry/arch/arch_x86.go @@ -114,6 +114,10 @@ func newX86FPStateSlice() []byte { size, align := cpuid.HostFeatureSet().ExtendedStateSize() capacity := size // Always use at least 4096 bytes. + // + // For the KVM platform, this state is a fixed 4096 bytes, so make sure + // that the underlying array is at _least_ that size otherwise we will + // corrupt random memory. This is not a pleasant thing to debug. if capacity < 4096 { capacity = 4096 } diff --git a/pkg/sentry/arch/signal_amd64.go b/pkg/sentry/arch/signal_amd64.go index 81b92bb43..6fb756f0e 100644 --- a/pkg/sentry/arch/signal_amd64.go +++ b/pkg/sentry/arch/signal_amd64.go @@ -55,7 +55,7 @@ type SignalContext64 struct { Trapno uint64 Oldmask linux.SignalSet Cr2 uint64 - // Pointer to a struct _fpstate. + // Pointer to a struct _fpstate. See b/33003106#comment8. Fpstate uint64 Reserved [8]uint64 } diff --git a/pkg/sentry/fs/file_overlay_test.go b/pkg/sentry/fs/file_overlay_test.go index 02538bb4f..a76d87e3a 100644 --- a/pkg/sentry/fs/file_overlay_test.go +++ b/pkg/sentry/fs/file_overlay_test.go @@ -177,6 +177,7 @@ func TestReaddirRevalidation(t *testing.T) { // TestReaddirOverlayFrozen tests that calling Readdir on an overlay file with // a frozen dirent tree does not make Readdir calls to the underlying files. +// This is a regression test for b/114808269. func TestReaddirOverlayFrozen(t *testing.T) { ctx := contexttest.Context(t) diff --git a/pkg/sentry/fs/proc/README.md b/pkg/sentry/fs/proc/README.md index 5d4ec6c7b..6667a0916 100644 --- a/pkg/sentry/fs/proc/README.md +++ b/pkg/sentry/fs/proc/README.md @@ -11,6 +11,8 @@ inconsistency, please file a bug. The following files are implemented: + + | File /proc/ | Content | | :------------------------ | :---------------------------------------------------- | | [cpuinfo](#cpuinfo) | Info about the CPU | @@ -22,6 +24,8 @@ The following files are implemented: | [uptime](#uptime) | Wall clock since boot, combined idle time of all cpus | | [version](#version) | Kernel version | + + ### cpuinfo ```bash diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD index a27628c0a..2231d6973 100644 --- a/pkg/sentry/kernel/BUILD +++ b/pkg/sentry/kernel/BUILD @@ -91,6 +91,7 @@ go_library( "fs_context.go", "ipc_namespace.go", "kernel.go", + "kernel_opts.go", "kernel_state.go", "pending_signals.go", "pending_signals_list.go", diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index dcd6e91c4..3ee760ba2 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -235,6 +235,9 @@ type Kernel struct { // events. This is initialized lazily on the first unimplemented // syscall. unimplementedSyscallEmitter eventchannel.Emitter `state:"nosave"` + + // SpecialOpts contains special kernel options. + SpecialOpts } // InitKernelArgs holds arguments to Init. diff --git a/pkg/sentry/kernel/kernel_opts.go b/pkg/sentry/kernel/kernel_opts.go new file mode 100644 index 000000000..2e66ec587 --- /dev/null +++ b/pkg/sentry/kernel/kernel_opts.go @@ -0,0 +1,20 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package kernel + +// SpecialOpts contains non-standard options for the kernel. +// +// +stateify savable +type SpecialOpts struct{} diff --git a/pkg/sentry/socket/hostinet/BUILD b/pkg/sentry/socket/hostinet/BUILD index 5a07d5d0e..023bad156 100644 --- a/pkg/sentry/socket/hostinet/BUILD +++ b/pkg/sentry/socket/hostinet/BUILD @@ -10,6 +10,7 @@ go_library( "save_restore.go", "socket.go", "socket_unsafe.go", + "sockopt_impl.go", "stack.go", ], visibility = ["//pkg/sentry:internal"], diff --git a/pkg/sentry/socket/hostinet/socket.go b/pkg/sentry/socket/hostinet/socket.go index 34f63986f..de76388ac 100644 --- a/pkg/sentry/socket/hostinet/socket.go +++ b/pkg/sentry/socket/hostinet/socket.go @@ -285,7 +285,7 @@ func (s *socketOperations) GetSockOpt(t *kernel.Task, level int, name int, outPt } // Whitelist options and constrain option length. - var optlen int + optlen := getSockOptLen(t, level, name) switch level { case linux.SOL_IP: switch name { @@ -330,7 +330,7 @@ func (s *socketOperations) GetSockOpt(t *kernel.Task, level int, name int, outPt // SetSockOpt implements socket.Socket.SetSockOpt. func (s *socketOperations) SetSockOpt(t *kernel.Task, level int, name int, opt []byte) *syserr.Error { // Whitelist options and constrain option length. - var optlen int + optlen := setSockOptLen(t, level, name) switch level { case linux.SOL_IP: switch name { @@ -353,6 +353,7 @@ func (s *socketOperations) SetSockOpt(t *kernel.Task, level int, name int, opt [ optlen = sizeofInt32 } } + if optlen == 0 { // Pretend to accept socket options we don't understand. This seems // dangerous, but it's what netstack does... diff --git a/pkg/sentry/socket/hostinet/sockopt_impl.go b/pkg/sentry/socket/hostinet/sockopt_impl.go new file mode 100644 index 000000000..8a783712e --- /dev/null +++ b/pkg/sentry/socket/hostinet/sockopt_impl.go @@ -0,0 +1,27 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package hostinet + +import ( + "gvisor.dev/gvisor/pkg/sentry/kernel" +) + +func getSockOptLen(t *kernel.Task, level, name int) int { + return 0 // No custom options. +} + +func setSockOptLen(t *kernel.Task, level, name int) int { + return 0 // No custom options. +} diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index e4a6b1b8b..f2be0e651 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -2166,6 +2166,9 @@ func (e *endpoint) listen(backlog int) *tcpip.Error { e.isRegistered = true e.setEndpointState(StateListen) + // The channel may be non-nil when we're restoring the endpoint, and it + // may be pre-populated with some previously accepted (but not Accepted) + // endpoints. if e.acceptedChan == nil { e.acceptedChan = make(chan *endpoint, backlog) } diff --git a/runsc/boot/filter/BUILD b/runsc/boot/filter/BUILD index ce30f6c53..ed18f0047 100644 --- a/runsc/boot/filter/BUILD +++ b/runsc/boot/filter/BUILD @@ -8,6 +8,7 @@ go_library( "config.go", "config_amd64.go", "config_arm64.go", + "config_profile.go", "extra_filters.go", "extra_filters_msan.go", "extra_filters_race.go", diff --git a/runsc/boot/filter/config.go b/runsc/boot/filter/config.go index f8d351c7b..c69f4c602 100644 --- a/runsc/boot/filter/config.go +++ b/runsc/boot/filter/config.go @@ -536,16 +536,3 @@ func controlServerFilters(fd int) seccomp.SyscallRules { }, } } - -// profileFilters returns extra syscalls made by runtime/pprof package. -func profileFilters() seccomp.SyscallRules { - return seccomp.SyscallRules{ - syscall.SYS_OPENAT: []seccomp.Rule{ - { - seccomp.AllowAny{}, - seccomp.AllowAny{}, - seccomp.AllowValue(syscall.O_RDONLY | syscall.O_LARGEFILE | syscall.O_CLOEXEC), - }, - }, - } -} diff --git a/runsc/boot/filter/config_profile.go b/runsc/boot/filter/config_profile.go new file mode 100644 index 000000000..194952a7b --- /dev/null +++ b/runsc/boot/filter/config_profile.go @@ -0,0 +1,34 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package filter + +import ( + "syscall" + + "gvisor.dev/gvisor/pkg/seccomp" +) + +// profileFilters returns extra syscalls made by runtime/pprof package. +func profileFilters() seccomp.SyscallRules { + return seccomp.SyscallRules{ + syscall.SYS_OPENAT: []seccomp.Rule{ + { + seccomp.AllowAny{}, + seccomp.AllowAny{}, + seccomp.AllowValue(syscall.O_RDONLY | syscall.O_LARGEFILE | syscall.O_CLOEXEC), + }, + }, + } +} diff --git a/runsc/container/console_test.go b/runsc/container/console_test.go index 060b63bf3..c2518d52b 100644 --- a/runsc/container/console_test.go +++ b/runsc/container/console_test.go @@ -196,7 +196,10 @@ func TestJobControlSignalExec(t *testing.T) { defer ptyMaster.Close() defer ptySlave.Close() - // Exec bash and attach a terminal. + // Exec bash and attach a terminal. Note that occasionally /bin/sh + // may be a different shell or have a different configuration (such + // as disabling interactive mode and job control). Since we want to + // explicitly test interactive mode, use /bin/bash. See b/116981926. execArgs := &control.ExecArgs{ Filename: "/bin/bash", // Don't let bash execute from profile or rc files, otherwise diff --git a/runsc/dockerutil/dockerutil.go b/runsc/dockerutil/dockerutil.go index 9b6346ca2..1ff5e8cc3 100644 --- a/runsc/dockerutil/dockerutil.go +++ b/runsc/dockerutil/dockerutil.go @@ -143,8 +143,11 @@ func PrepareFiles(names ...string) (string, error) { return "", fmt.Errorf("os.Chmod(%q, 0777) failed: %v", dir, err) } for _, name := range names { - src := getLocalPath(name) - dst := path.Join(dir, name) + src, err := testutil.FindFile(name) + if err != nil { + return "", fmt.Errorf("testutil.Preparefiles(%q) failed: %v", name, err) + } + dst := path.Join(dir, path.Base(name)) if err := testutil.Copy(src, dst); err != nil { return "", fmt.Errorf("testutil.Copy(%q, %q) failed: %v", src, dst, err) } @@ -152,10 +155,6 @@ func PrepareFiles(names ...string) (string, error) { return dir, nil } -func getLocalPath(file string) string { - return path.Join(".", file) -} - // do executes docker command. func do(args ...string) (string, error) { log.Printf("Running: docker %s\n", args) diff --git a/runsc/testutil/BUILD b/runsc/testutil/BUILD index f845120b0..945405303 100644 --- a/runsc/testutil/BUILD +++ b/runsc/testutil/BUILD @@ -5,7 +5,10 @@ package(licenses = ["notice"]) go_library( name = "testutil", testonly = 1, - srcs = ["testutil.go"], + srcs = [ + "testutil.go", + "testutil_runfiles.go", + ], visibility = ["//:sandbox"], deps = [ "//pkg/log", diff --git a/runsc/testutil/testutil.go b/runsc/testutil/testutil.go index edf2e809a..80c2c9680 100644 --- a/runsc/testutil/testutil.go +++ b/runsc/testutil/testutil.go @@ -79,60 +79,6 @@ func ConfigureExePath() error { return nil } -// FindFile searchs for a file inside the test run environment. It returns the -// full path to the file. It fails if none or more than one file is found. -func FindFile(path string) (string, error) { - wd, err := os.Getwd() - if err != nil { - return "", err - } - - // The test root is demarcated by a path element called "__main__". Search for - // it backwards from the working directory. - root := wd - for { - dir, name := filepath.Split(root) - if name == "__main__" { - break - } - if len(dir) == 0 { - return "", fmt.Errorf("directory __main__ not found in %q", wd) - } - // Remove ending slash to loop around. - root = dir[:len(dir)-1] - } - - // Annoyingly, bazel adds the build type to the directory path for go - // binaries, but not for c++ binaries. We use two different patterns to - // to find our file. - patterns := []string{ - // Try the obvious path first. - filepath.Join(root, path), - // If it was a go binary, use a wildcard to match the build - // type. The pattern is: /test-path/__main__/directories/*/file. - filepath.Join(root, filepath.Dir(path), "*", filepath.Base(path)), - } - - for _, p := range patterns { - matches, err := filepath.Glob(p) - if err != nil { - // "The only possible returned error is ErrBadPattern, - // when pattern is malformed." -godoc - return "", fmt.Errorf("error globbing %q: %v", p, err) - } - switch len(matches) { - case 0: - // Try the next pattern. - case 1: - // We found it. - return matches[0], nil - default: - return "", fmt.Errorf("more than one match found for %q: %s", path, matches) - } - } - return "", fmt.Errorf("file %q not found", path) -} - // TestConfig returns the default configuration to use in tests. Note that // 'RootDir' must be set by caller if required. func TestConfig() *boot.Config { diff --git a/runsc/testutil/testutil_runfiles.go b/runsc/testutil/testutil_runfiles.go new file mode 100644 index 000000000..ece9ea9a1 --- /dev/null +++ b/runsc/testutil/testutil_runfiles.go @@ -0,0 +1,75 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package testutil + +import ( + "fmt" + "os" + "path/filepath" +) + +// FindFile searchs for a file inside the test run environment. It returns the +// full path to the file. It fails if none or more than one file is found. +func FindFile(path string) (string, error) { + wd, err := os.Getwd() + if err != nil { + return "", err + } + + // The test root is demarcated by a path element called "__main__". Search for + // it backwards from the working directory. + root := wd + for { + dir, name := filepath.Split(root) + if name == "__main__" { + break + } + if len(dir) == 0 { + return "", fmt.Errorf("directory __main__ not found in %q", wd) + } + // Remove ending slash to loop around. + root = dir[:len(dir)-1] + } + + // Annoyingly, bazel adds the build type to the directory path for go + // binaries, but not for c++ binaries. We use two different patterns to + // to find our file. + patterns := []string{ + // Try the obvious path first. + filepath.Join(root, path), + // If it was a go binary, use a wildcard to match the build + // type. The pattern is: /test-path/__main__/directories/*/file. + filepath.Join(root, filepath.Dir(path), "*", filepath.Base(path)), + } + + for _, p := range patterns { + matches, err := filepath.Glob(p) + if err != nil { + // "The only possible returned error is ErrBadPattern, + // when pattern is malformed." -godoc + return "", fmt.Errorf("error globbing %q: %v", p, err) + } + switch len(matches) { + case 0: + // Try the next pattern. + case 1: + // We found it. + return matches[0], nil + default: + return "", fmt.Errorf("more than one match found for %q: %s", path, matches) + } + } + return "", fmt.Errorf("file %q not found", path) +} diff --git a/test/image/image_test.go b/test/image/image_test.go index d0dcb1861..0a1e19d6f 100644 --- a/test/image/image_test.go +++ b/test/image/image_test.go @@ -107,7 +107,7 @@ func TestHttpd(t *testing.T) { } d := dockerutil.MakeDocker("http-test") - dir, err := dockerutil.PrepareFiles("latin10k.txt") + dir, err := dockerutil.PrepareFiles("test/image/latin10k.txt") if err != nil { t.Fatalf("PrepareFiles() failed: %v", err) } @@ -139,7 +139,7 @@ func TestNginx(t *testing.T) { } d := dockerutil.MakeDocker("net-test") - dir, err := dockerutil.PrepareFiles("latin10k.txt") + dir, err := dockerutil.PrepareFiles("test/image/latin10k.txt") if err != nil { t.Fatalf("PrepareFiles() failed: %v", err) } @@ -183,7 +183,7 @@ func TestMysql(t *testing.T) { } client := dockerutil.MakeDocker("mysql-client-test") - dir, err := dockerutil.PrepareFiles("mysql.sql") + dir, err := dockerutil.PrepareFiles("test/image/mysql.sql") if err != nil { t.Fatalf("PrepareFiles() failed: %v", err) } @@ -283,7 +283,7 @@ func TestRuby(t *testing.T) { } d := dockerutil.MakeDocker("ruby-test") - dir, err := dockerutil.PrepareFiles("ruby.rb", "ruby.sh") + dir, err := dockerutil.PrepareFiles("test/image/ruby.rb", "test/image/ruby.sh") if err != nil { t.Fatalf("PrepareFiles() failed: %v", err) } diff --git a/test/syscalls/build_defs.bzl b/test/syscalls/build_defs.bzl index 1df761dd0..cbab85ef7 100644 --- a/test/syscalls/build_defs.bzl +++ b/test/syscalls/build_defs.bzl @@ -2,8 +2,6 @@ load("//tools:defs.bzl", "loopback") -# syscall_test is a macro that will create targets to run the given test target -# on the host (native) and runsc. def syscall_test( test, shard_count = 5, @@ -13,6 +11,19 @@ def syscall_test( add_uds_tree = False, add_hostinet = False, tags = None): + """syscall_test is a macro that will create targets for all platforms. + + Args: + test: the test target. + shard_count: shards for defined tests. + size: the defined test size. + use_tmpfs: use tmpfs in the defined tests. + add_overlay: add an overlay test. + add_uds_tree: add a UDS test. + add_hostinet: add a hostinet test. + tags: starting test tags. + """ + _syscall_test( test = test, shard_count = shard_count, @@ -111,6 +122,19 @@ def _syscall_test( # all the tests on a specific flavor. Use --test_tag_filters=ptrace,file_shared. tags += [full_platform, "file_" + file_access] + # Hash this target into one of 15 buckets. This can be used to + # randomly split targets between different workflows. + hash15 = hash(native.package_name() + name) % 15 + tags.append("hash15:" + str(hash15)) + + # TODO(b/139838000): Tests using hostinet must be disabled on Guitar until + # we figure out how to request ipv4 sockets on Guitar machines. + if network == "host": + tags.append("noguitar") + + # Disable off-host networking. + tags.append("requires-net:loopback") + # Add tag to prevent the tests from running in a Bazel sandbox. # TODO(b/120560048): Make the tests run without this tag. tags.append("no-sandbox") @@ -118,8 +142,11 @@ def _syscall_test( # TODO(b/112165693): KVM tests are tagged "manual" to until the platform is # more stable. if platform == "kvm": - tags += ["manual"] - tags += ["requires-kvm"] + tags.append("manual") + tags.append("requires-kvm") + + # TODO(b/112165693): Remove when tests pass reliably. + tags.append("notap") args = [ # Arguments are passed directly to syscall_test_runner binary. diff --git a/test/syscalls/linux/chroot.cc b/test/syscalls/linux/chroot.cc index 0a2d44a2c..85ec013d5 100644 --- a/test/syscalls/linux/chroot.cc +++ b/test/syscalls/linux/chroot.cc @@ -167,7 +167,7 @@ TEST(ChrootTest, DotDotFromOpenFD) { } // Test that link resolution in a chroot can escape the root by following an -// open proc fd. +// open proc fd. Regression test for b/32316719. TEST(ChrootTest, ProcFdLinkResolutionInChroot) { SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SYS_CHROOT))); diff --git a/test/syscalls/linux/concurrency.cc b/test/syscalls/linux/concurrency.cc index f41f99900..7cd6a75bd 100644 --- a/test/syscalls/linux/concurrency.cc +++ b/test/syscalls/linux/concurrency.cc @@ -46,7 +46,8 @@ TEST(ConcurrencyTest, SingleProcessMultithreaded) { } // Test that multiple threads in this process continue to execute in parallel, -// even if an unrelated second process is spawned. +// even if an unrelated second process is spawned. Regression test for +// b/32119508. TEST(ConcurrencyTest, MultiProcessMultithreaded) { // In PID 1, start TIDs 1 and 2, and put both to sleep. // diff --git a/test/syscalls/linux/exec_proc_exe_workload.cc b/test/syscalls/linux/exec_proc_exe_workload.cc index b790fe5be..2989379b7 100644 --- a/test/syscalls/linux/exec_proc_exe_workload.cc +++ b/test/syscalls/linux/exec_proc_exe_workload.cc @@ -21,6 +21,12 @@ #include "test/util/posix_error.h" int main(int argc, char** argv, char** envp) { + // This is annoying. Because remote build systems may put these binaries + // in a content-addressable-store, you may wind up with /proc/self/exe + // pointing to some random path (but with a sensible argv[0]). + // + // Therefore, this test simply checks that the /proc/self/exe + // is absolute and *doesn't* match argv[1]. std::string exe = gvisor::testing::ProcessExePath(getpid()).ValueOrDie(); if (exe[0] != '/') { diff --git a/test/syscalls/linux/fork.cc b/test/syscalls/linux/fork.cc index 906f3358d..ff8bdfeb0 100644 --- a/test/syscalls/linux/fork.cc +++ b/test/syscalls/linux/fork.cc @@ -271,7 +271,7 @@ TEST_F(ForkTest, Alarm) { EXPECT_EQ(0, alarmed); } -// Child cannot affect parent private memory. +// Child cannot affect parent private memory. Regression test for b/24137240. TEST_F(ForkTest, PrivateMemory) { std::atomic local(0); @@ -298,6 +298,9 @@ TEST_F(ForkTest, PrivateMemory) { } // Kernel-accessed buffers should remain coherent across COW. +// +// The buffer must be >= usermem.ZeroCopyMinBytes, as UnsafeAccess operates +// differently. Regression test for b/33811887. TEST_F(ForkTest, COWSegment) { constexpr int kBufSize = 1024; char* read_buf = private_; diff --git a/test/syscalls/linux/mmap.cc b/test/syscalls/linux/mmap.cc index 1c4d9f1c7..11fb1b457 100644 --- a/test/syscalls/linux/mmap.cc +++ b/test/syscalls/linux/mmap.cc @@ -1418,7 +1418,7 @@ TEST_P(MMapFileParamTest, NoSigBusOnPageContainingEOF) { // // On most platforms this is trivial, but when the file is mapped via the sentry // page cache (which does not yet support writing to shared mappings), a bug -// caused reads to fail unnecessarily on such mappings. +// caused reads to fail unnecessarily on such mappings. See b/28913513. TEST_F(MMapFileTest, ReadingWritableSharedFilePageSucceeds) { uintptr_t addr; size_t len = strlen(kFileContents); @@ -1435,7 +1435,7 @@ TEST_F(MMapFileTest, ReadingWritableSharedFilePageSucceeds) { // Tests that EFAULT is returned when invoking a syscall that requires the OS to // read past end of file (resulting in a fault in sentry context in the gVisor -// case). +// case). See b/28913513. TEST_F(MMapFileTest, InternalSigBus) { uintptr_t addr; ASSERT_THAT(addr = Map(0, 2 * kPageSize, PROT_READ | PROT_WRITE, MAP_PRIVATE, @@ -1578,7 +1578,7 @@ TEST_F(MMapFileTest, Bug38498194) { } // Tests that reading from a file to a memory mapping of the same file does not -// deadlock. +// deadlock. See b/34813270. TEST_F(MMapFileTest, SelfRead) { uintptr_t addr; ASSERT_THAT(addr = Map(0, kPageSize, PROT_READ | PROT_WRITE, MAP_SHARED, @@ -1590,7 +1590,7 @@ TEST_F(MMapFileTest, SelfRead) { } // Tests that writing to a file from a memory mapping of the same file does not -// deadlock. +// deadlock. Regression test for b/34813270. TEST_F(MMapFileTest, SelfWrite) { uintptr_t addr; ASSERT_THAT(addr = Map(0, kPageSize, PROT_READ, MAP_SHARED, fd_.get(), 0), diff --git a/test/syscalls/linux/open_create.cc b/test/syscalls/linux/open_create.cc index 431733dbe..902d0a0dc 100644 --- a/test/syscalls/linux/open_create.cc +++ b/test/syscalls/linux/open_create.cc @@ -132,6 +132,7 @@ TEST(CreateTest, CreateFailsOnDirWithoutWritePerms) { } // A file originally created RW, but opened RO can later be opened RW. +// Regression test for b/65385065. TEST(CreateTest, OpenCreateROThenRW) { TempPath file(NewTempAbsPath()); diff --git a/test/syscalls/linux/preadv.cc b/test/syscalls/linux/preadv.cc index f7ea44054..5b0743fe9 100644 --- a/test/syscalls/linux/preadv.cc +++ b/test/syscalls/linux/preadv.cc @@ -37,6 +37,7 @@ namespace testing { namespace { +// Stress copy-on-write. Attempts to reproduce b/38430174. TEST(PreadvTest, MMConcurrencyStress) { // Fill a one-page file with zeroes (the contents don't really matter). const auto f = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileWith( diff --git a/test/syscalls/linux/proc.cc b/test/syscalls/linux/proc.cc index 169b723eb..a23fdb58d 100644 --- a/test/syscalls/linux/proc.cc +++ b/test/syscalls/linux/proc.cc @@ -1352,13 +1352,19 @@ TEST(ProcPidSymlink, SubprocessZombied) { // FIXME(gvisor.dev/issue/164): Inconsistent behavior between gVisor and linux // on proc files. - // 4.17 & gVisor: Syscall succeeds and returns 1 + // + // ~4.3: Syscall fails with EACCES. + // 4.17 & gVisor: Syscall succeeds and returns 1. + // // EXPECT_THAT(ReadlinkWhileZombied("ns/pid", buf, sizeof(buf)), // SyscallFailsWithErrno(EACCES)); // FIXME(gvisor.dev/issue/164): Inconsistent behavior between gVisor and linux // on proc files. - // 4.17 & gVisor: Syscall succeeds and returns 1. + // + // ~4.3: Syscall fails with EACCES. + // 4.17 & gVisor: Syscall succeeds and returns 1. + // // EXPECT_THAT(ReadlinkWhileZombied("ns/user", buf, sizeof(buf)), // SyscallFailsWithErrno(EACCES)); } @@ -1431,8 +1437,12 @@ TEST(ProcPidFile, SubprocessRunning) { TEST(ProcPidFile, SubprocessZombie) { char buf[1]; - // 4.17: Succeeds and returns 1 - // gVisor: Succeeds and returns 0 + // FIXME(gvisor.dev/issue/164): Loosen requirement due to inconsistent + // behavior on different kernels. + // + // ~4.3: Succeds and returns 0. + // 4.17: Succeeds and returns 1. + // gVisor: Succeeds and returns 0. EXPECT_THAT(ReadWhileZombied("auxv", buf, sizeof(buf)), SyscallSucceeds()); EXPECT_THAT(ReadWhileZombied("cmdline", buf, sizeof(buf)), @@ -1458,7 +1468,10 @@ TEST(ProcPidFile, SubprocessZombie) { // FIXME(gvisor.dev/issue/164): Inconsistent behavior between gVisor and linux // on proc files. + // + // ~4.3: Fails and returns EACCES. // gVisor & 4.17: Succeeds and returns 1. + // // EXPECT_THAT(ReadWhileZombied("io", buf, sizeof(buf)), // SyscallFailsWithErrno(EACCES)); } @@ -1467,9 +1480,12 @@ TEST(ProcPidFile, SubprocessZombie) { TEST(ProcPidFile, SubprocessExited) { char buf[1]; - // FIXME(gvisor.dev/issue/164): Inconsistent behavior between kernels + // FIXME(gvisor.dev/issue/164): Inconsistent behavior between kernels. + // + // ~4.3: Fails and returns ESRCH. // gVisor: Fails with ESRCH. // 4.17: Succeeds and returns 1. + // // EXPECT_THAT(ReadWhileExited("auxv", buf, sizeof(buf)), // SyscallFailsWithErrno(ESRCH)); @@ -1641,7 +1657,7 @@ TEST(ProcTask, KilledThreadsDisappear) { EXPECT_NO_ERRNO(DirContainsExactly("/proc/self/task", TaskFiles(initial, {child1.Tid()}))); - // Stat child1's task file. + // Stat child1's task file. Regression test for b/32097707. struct stat statbuf; const std::string child1_task_file = absl::StrCat("/proc/self/task/", child1.Tid()); @@ -1669,7 +1685,7 @@ TEST(ProcTask, KilledThreadsDisappear) { EXPECT_NO_ERRNO(EventuallyDirContainsExactly( "/proc/self/task", TaskFiles(initial, {child3.Tid(), child5.Tid()}))); - // Stat child1's task file again. This time it should fail. + // Stat child1's task file again. This time it should fail. See b/32097707. EXPECT_THAT(stat(child1_task_file.c_str(), &statbuf), SyscallFailsWithErrno(ENOENT)); @@ -1824,7 +1840,7 @@ TEST(ProcSysVmOvercommitMemory, HasNumericValue) { } // Check that link for proc fd entries point the target node, not the -// symlink itself. +// symlink itself. Regression test for b/31155070. TEST(ProcTaskFd, FstatatFollowsSymlink) { const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); const FileDescriptor fd = @@ -1883,6 +1899,20 @@ TEST(ProcMounts, IsSymlink) { EXPECT_EQ(link, "self/mounts"); } +TEST(ProcSelfMountinfo, RequiredFieldsArePresent) { + auto mountinfo = + ASSERT_NO_ERRNO_AND_VALUE(GetContents("/proc/self/mountinfo")); + EXPECT_THAT( + mountinfo, + AllOf( + // Root mount. + ContainsRegex( + R"([0-9]+ [0-9]+ [0-9]+:[0-9]+ / / (rw|ro).*- \S+ \S+ (rw|ro)\S*)"), + // Proc mount - always rw. + ContainsRegex( + R"([0-9]+ [0-9]+ [0-9]+:[0-9]+ / /proc rw.*- \S+ \S+ rw\S*)"))); +} + // Check that /proc/self/mounts looks something like a real mounts file. TEST(ProcSelfMounts, RequiredFieldsArePresent) { auto mounts = ASSERT_NO_ERRNO_AND_VALUE(GetContents("/proc/self/mounts")); diff --git a/test/syscalls/linux/readv.cc b/test/syscalls/linux/readv.cc index 4069cbc7e..baaf9f757 100644 --- a/test/syscalls/linux/readv.cc +++ b/test/syscalls/linux/readv.cc @@ -254,7 +254,9 @@ TEST_F(ReadvTest, IovecOutsideTaskAddressRangeInNonemptyArray) { // This test depends on the maximum extent of a single readv() syscall, so // we can't tolerate interruption from saving. TEST(ReadvTestNoFixture, TruncatedAtMax_NoRandomSave) { - // Ensure that we won't be interrupted by ITIMER_PROF. + // Ensure that we won't be interrupted by ITIMER_PROF. This is particularly + // important in environments where automated profiling tools may start + // ITIMER_PROF automatically. struct itimerval itv = {}; auto const cleanup_itimer = ASSERT_NO_ERRNO_AND_VALUE(ScopedItimer(ITIMER_PROF, itv)); diff --git a/test/syscalls/linux/rseq.cc b/test/syscalls/linux/rseq.cc index 106c045e3..4bfb1ff56 100644 --- a/test/syscalls/linux/rseq.cc +++ b/test/syscalls/linux/rseq.cc @@ -36,7 +36,7 @@ namespace { // We must be very careful about how these tests are written. Each thread may // only have one struct rseq registration, which may be done automatically at // thread start (as of 2019-11-13, glibc does *not* support rseq and thus does -// not do so). +// not do so, but other libraries do). // // Testing of rseq is thus done primarily in a child process with no // registration. This means exec'ing a nostdlib binary, as rseq registration can diff --git a/test/syscalls/linux/select.cc b/test/syscalls/linux/select.cc index 424e2a67f..be2364fb8 100644 --- a/test/syscalls/linux/select.cc +++ b/test/syscalls/linux/select.cc @@ -146,7 +146,7 @@ TEST_F(SelectTest, IgnoreBitsAboveNfds) { // This test illustrates Linux's behavior of 'select' calls passing after // setrlimit RLIMIT_NOFILE is called. In particular, versions of sshd rely on -// this behavior. +// this behavior. See b/122318458. TEST_F(SelectTest, SetrlimitCallNOFILE) { fd_set read_set; FD_ZERO(&read_set); diff --git a/test/syscalls/linux/shm.cc b/test/syscalls/linux/shm.cc index 7ba752599..c7fdbb924 100644 --- a/test/syscalls/linux/shm.cc +++ b/test/syscalls/linux/shm.cc @@ -473,7 +473,7 @@ TEST(ShmTest, PartialUnmap) { } // Check that sentry does not panic when asked for a zero-length private shm -// segment. +// segment. Regression test for b/110694797. TEST(ShmTest, GracefullyFailOnZeroLenSegmentCreation) { EXPECT_THAT(Shmget(IPC_PRIVATE, 0, 0), PosixErrorIs(EINVAL, _)); } diff --git a/test/syscalls/linux/sigprocmask.cc b/test/syscalls/linux/sigprocmask.cc index 654c6a47f..a603fc1d1 100644 --- a/test/syscalls/linux/sigprocmask.cc +++ b/test/syscalls/linux/sigprocmask.cc @@ -237,7 +237,7 @@ TEST_F(SigProcMaskTest, SignalHandler) { } // Check that sigprocmask correctly handles aliasing of the set and oldset -// pointers. +// pointers. Regression test for b/30502311. TEST_F(SigProcMaskTest, AliasedSets) { sigset_t mask; diff --git a/test/syscalls/linux/socket_unix_non_stream.cc b/test/syscalls/linux/socket_unix_non_stream.cc index 276a94eb8..884319e1d 100644 --- a/test/syscalls/linux/socket_unix_non_stream.cc +++ b/test/syscalls/linux/socket_unix_non_stream.cc @@ -109,7 +109,7 @@ PosixErrorOr> CreateFragmentedRegion(const int size, } // A contiguous iov that is heavily fragmented in FileMem can still be sent -// successfully. +// successfully. See b/115833655. TEST_P(UnixNonStreamSocketPairTest, FragmentedSendMsg) { auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); @@ -165,7 +165,7 @@ TEST_P(UnixNonStreamSocketPairTest, FragmentedSendMsg) { } // A contiguous iov that is heavily fragmented in FileMem can still be received -// into successfully. +// into successfully. Regression test for b/115833655. TEST_P(UnixNonStreamSocketPairTest, FragmentedRecvMsg) { auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); diff --git a/test/syscalls/linux/symlink.cc b/test/syscalls/linux/symlink.cc index b249ff91f..03ee1250d 100644 --- a/test/syscalls/linux/symlink.cc +++ b/test/syscalls/linux/symlink.cc @@ -38,7 +38,7 @@ mode_t FilePermission(const std::string& path) { } // Test that name collisions are checked on the new link path, not the source -// path. +// path. Regression test for b/31782115. TEST(SymlinkTest, CanCreateSymlinkWithCachedSourceDirent) { const std::string srcname = NewTempAbsPath(); const std::string newname = NewTempAbsPath(); diff --git a/test/syscalls/linux/tcp_socket.cc b/test/syscalls/linux/tcp_socket.cc index 8a8b68e75..c4591a3b9 100644 --- a/test/syscalls/linux/tcp_socket.cc +++ b/test/syscalls/linux/tcp_socket.cc @@ -244,7 +244,8 @@ TEST_P(TcpSocketTest, ZeroWriteAllowed) { } // Test that a non-blocking write with a buffer that is larger than the send -// buffer size will not actually write the whole thing at once. +// buffer size will not actually write the whole thing at once. Regression test +// for b/64438887. TEST_P(TcpSocketTest, NonblockingLargeWrite) { // Set the FD to O_NONBLOCK. int opts; diff --git a/test/syscalls/linux/time.cc b/test/syscalls/linux/time.cc index c7eead17e..1ccb95733 100644 --- a/test/syscalls/linux/time.cc +++ b/test/syscalls/linux/time.cc @@ -62,6 +62,7 @@ TEST(TimeTest, VsyscallTime_InvalidAddressSIGSEGV) { ::testing::KilledBySignal(SIGSEGV), ""); } +// Mimics the gettimeofday(2) wrapper from the Go runtime <= 1.2. int vsyscall_gettimeofday(struct timeval* tv, struct timezone* tz) { constexpr uint64_t kVsyscallGettimeofdayEntry = 0xffffffffff600000; return reinterpret_cast( diff --git a/test/syscalls/linux/tkill.cc b/test/syscalls/linux/tkill.cc index bae377c69..8d8ebbb24 100644 --- a/test/syscalls/linux/tkill.cc +++ b/test/syscalls/linux/tkill.cc @@ -54,7 +54,7 @@ void SigHandler(int sig, siginfo_t* info, void* context) { TEST_CHECK(info->si_code == SI_TKILL); } -// Test with a real signal. +// Test with a real signal. Regression test for b/24790092. TEST(TkillTest, ValidTIDAndRealSignal) { struct sigaction sa; sa.sa_sigaction = SigHandler; diff --git a/test/util/temp_path.cc b/test/util/temp_path.cc index 35aacb172..9c10b6674 100644 --- a/test/util/temp_path.cc +++ b/test/util/temp_path.cc @@ -77,6 +77,7 @@ std::string NewTempAbsPath() { std::string NewTempRelPath() { return NextTempBasename(); } std::string GetAbsoluteTestTmpdir() { + // Note that TEST_TMPDIR is guaranteed to be set. char* env_tmpdir = getenv("TEST_TMPDIR"); std::string tmp_dir = env_tmpdir != nullptr ? std::string(env_tmpdir) : "/tmp"; diff --git a/tools/build/tags.bzl b/tools/build/tags.bzl index e99c87f81..a6db44e47 100644 --- a/tools/build/tags.bzl +++ b/tools/build/tags.bzl @@ -33,4 +33,8 @@ go_suffixes = [ "_wasm_unsafe", "_linux", "_linux_unsafe", + "_opts", + "_opts_unsafe", + "_impl", + "_impl_unsafe", ] diff --git a/tools/defs.bzl b/tools/defs.bzl index 5d5fa134a..c03b557ae 100644 --- a/tools/defs.bzl +++ b/tools/defs.bzl @@ -73,6 +73,16 @@ def calculate_sets(srcs): result[target].append(file) return result +def go_imports(name, src, out): + """Simplify a single Go source file by eliminating unused imports.""" + native.genrule( + name = name, + srcs = [src], + outs = [out], + tools = ["@org_golang_x_tools//cmd/goimports:goimports"], + cmd = ("$(location @org_golang_x_tools//cmd/goimports:goimports) $(SRCS) > $@"), + ) + def go_library(name, srcs, deps = [], imports = [], stateify = True, marshal = False, **kwargs): """Wraps the standard go_library and does stateification and marshalling. @@ -107,10 +117,15 @@ def go_library(name, srcs, deps = [], imports = [], stateify = True, marshal = F state_sets = calculate_sets(srcs) for (suffix, srcs) in state_sets.items(): go_stateify( - name = name + suffix + "_state_autogen", + name = name + suffix + "_state_autogen_with_imports", srcs = srcs, imports = imports, package = name, + out = name + suffix + "_state_autogen_with_imports.go", + ) + go_imports( + name = name + suffix + "_state_autogen", + src = name + suffix + "_state_autogen_with_imports.go", out = name + suffix + "_state_autogen.go", ) all_srcs = all_srcs + [ -- cgit v1.2.3 From c141eb5f430dc50f6bf90232c369b7b3a542155e Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Fri, 7 Feb 2020 13:47:57 -0800 Subject: Address GH comments. --- pkg/abi/linux/netfilter.go | 2 +- pkg/sentry/socket/netfilter/extensions.go | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/netfilter.go b/pkg/abi/linux/netfilter.go index e4aabb6bb..7363185b7 100644 --- a/pkg/abi/linux/netfilter.go +++ b/pkg/abi/linux/netfilter.go @@ -363,7 +363,7 @@ type XTTCP struct { // range to which the matcher applies. DestinationPortStart uint16 - // DestinationPortEnd specifies the start of the destination port + // DestinationPortEnd specifies the end of the destination port // range to which the matcher applies. DestinationPortEnd uint16 diff --git a/pkg/sentry/socket/netfilter/extensions.go b/pkg/sentry/socket/netfilter/extensions.go index b5fbb52e4..3082976cd 100644 --- a/pkg/sentry/socket/netfilter/extensions.go +++ b/pkg/sentry/socket/netfilter/extensions.go @@ -1,4 +1,4 @@ -// Copyright 2019 The gVisor Authors. +// Copyright 2020 The gVisor Authors. // // Licensed under the Apache License, Version 2.0 (the "License"); // you may not use this file except in compliance with the License. -- cgit v1.2.3 From e1587a28876f8aac689a2cd1b7630f1637655b58 Mon Sep 17 00:00:00 2001 From: Ian Gudger Date: Fri, 7 Feb 2020 14:00:50 -0800 Subject: Log level, optname, optval and optlen in getsockopt/setsockopt in strace. Log 8, 16, and 32 int optvals and dump the memory of other sizes. Updates #1782 PiperOrigin-RevId: 293889388 --- pkg/sentry/strace/linux64_amd64.go | 4 +- pkg/sentry/strace/socket.go | 215 +++++++++++++++++++++++++++++++++++++ pkg/sentry/strace/strace.go | 22 +++- pkg/sentry/strace/syscalls.go | 22 +++- 4 files changed, 258 insertions(+), 5 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/strace/linux64_amd64.go b/pkg/sentry/strace/linux64_amd64.go index 85ec66fd3..a4de545e9 100644 --- a/pkg/sentry/strace/linux64_amd64.go +++ b/pkg/sentry/strace/linux64_amd64.go @@ -78,8 +78,8 @@ var linuxAMD64 = SyscallMap{ 51: makeSyscallInfo("getsockname", FD, PostSockAddr, SockLen), 52: makeSyscallInfo("getpeername", FD, PostSockAddr, SockLen), 53: makeSyscallInfo("socketpair", SockFamily, SockType, SockProtocol, Hex), - 54: makeSyscallInfo("setsockopt", FD, Hex, Hex, Hex, Hex), - 55: makeSyscallInfo("getsockopt", FD, Hex, Hex, Hex, Hex), + 54: makeSyscallInfo("setsockopt", FD, SockOptLevel, SockOptName, SetSockOptVal, Hex /* length by value, not a pointer */), + 55: makeSyscallInfo("getsockopt", FD, SockOptLevel, SockOptName, GetSockOptVal, SockLen), 56: makeSyscallInfo("clone", CloneFlags, Hex, Hex, Hex, Hex), 57: makeSyscallInfo("fork"), 58: makeSyscallInfo("vfork"), diff --git a/pkg/sentry/strace/socket.go b/pkg/sentry/strace/socket.go index d2079c85f..f7ff4573e 100644 --- a/pkg/sentry/strace/socket.go +++ b/pkg/sentry/strace/socket.go @@ -419,3 +419,218 @@ func sockFlags(flags int32) string { } return SocketFlagSet.Parse(uint64(flags)) } + +func getSockOptVal(t *kernel.Task, level, optname uint64, optVal usermem.Addr, optLen usermem.Addr, maximumBlobSize uint, rval uintptr) string { + if int(rval) < 0 { + return hexNum(uint64(optVal)) + } + if optVal == 0 { + return "null" + } + l, err := copySockLen(t, optLen) + if err != nil { + return fmt.Sprintf("%#x {error reading length: %v}", optLen, err) + } + return sockOptVal(t, level, optname, optVal, uint64(l), maximumBlobSize) +} + +func sockOptVal(t *kernel.Task, level, optname uint64, optVal usermem.Addr, optLen uint64, maximumBlobSize uint) string { + switch optLen { + case 1: + var v uint8 + _, err := t.CopyIn(optVal, &v) + if err != nil { + return fmt.Sprintf("%#x {error reading optval: %v}", optVal, err) + } + return fmt.Sprintf("%#x {value=%v}", optVal, v) + case 2: + var v uint16 + _, err := t.CopyIn(optVal, &v) + if err != nil { + return fmt.Sprintf("%#x {error reading optval: %v}", optVal, err) + } + return fmt.Sprintf("%#x {value=%v}", optVal, v) + case 4: + var v uint32 + _, err := t.CopyIn(optVal, &v) + if err != nil { + return fmt.Sprintf("%#x {error reading optval: %v}", optVal, err) + } + return fmt.Sprintf("%#x {value=%v}", optVal, v) + default: + return dump(t, optVal, uint(optLen), maximumBlobSize) + } +} + +var sockOptLevels = abi.ValueSet{ + linux.SOL_IP: "SOL_IP", + linux.SOL_SOCKET: "SOL_SOCKET", + linux.SOL_TCP: "SOL_TCP", + linux.SOL_UDP: "SOL_UDP", + linux.SOL_IPV6: "SOL_IPV6", + linux.SOL_ICMPV6: "SOL_ICMPV6", + linux.SOL_RAW: "SOL_RAW", + linux.SOL_PACKET: "SOL_PACKET", + linux.SOL_NETLINK: "SOL_NETLINK", +} + +var sockOptNames = map[uint64]abi.ValueSet{ + linux.SOL_IP: { + linux.IP_TTL: "IP_TTL", + linux.IP_MULTICAST_TTL: "IP_MULTICAST_TTL", + linux.IP_MULTICAST_IF: "IP_MULTICAST_IF", + linux.IP_MULTICAST_LOOP: "IP_MULTICAST_LOOP", + linux.IP_TOS: "IP_TOS", + linux.IP_RECVTOS: "IP_RECVTOS", + linux.IPT_SO_GET_INFO: "IPT_SO_GET_INFO", + linux.IPT_SO_GET_ENTRIES: "IPT_SO_GET_ENTRIES", + linux.IP_ADD_MEMBERSHIP: "IP_ADD_MEMBERSHIP", + linux.IP_DROP_MEMBERSHIP: "IP_DROP_MEMBERSHIP", + linux.MCAST_JOIN_GROUP: "MCAST_JOIN_GROUP", + linux.IP_ADD_SOURCE_MEMBERSHIP: "IP_ADD_SOURCE_MEMBERSHIP", + linux.IP_BIND_ADDRESS_NO_PORT: "IP_BIND_ADDRESS_NO_PORT", + linux.IP_BLOCK_SOURCE: "IP_BLOCK_SOURCE", + linux.IP_CHECKSUM: "IP_CHECKSUM", + linux.IP_DROP_SOURCE_MEMBERSHIP: "IP_DROP_SOURCE_MEMBERSHIP", + linux.IP_FREEBIND: "IP_FREEBIND", + linux.IP_HDRINCL: "IP_HDRINCL", + linux.IP_IPSEC_POLICY: "IP_IPSEC_POLICY", + linux.IP_MINTTL: "IP_MINTTL", + linux.IP_MSFILTER: "IP_MSFILTER", + linux.IP_MTU_DISCOVER: "IP_MTU_DISCOVER", + linux.IP_MULTICAST_ALL: "IP_MULTICAST_ALL", + linux.IP_NODEFRAG: "IP_NODEFRAG", + linux.IP_OPTIONS: "IP_OPTIONS", + linux.IP_PASSSEC: "IP_PASSSEC", + linux.IP_PKTINFO: "IP_PKTINFO", + linux.IP_RECVERR: "IP_RECVERR", + linux.IP_RECVFRAGSIZE: "IP_RECVFRAGSIZE", + linux.IP_RECVOPTS: "IP_RECVOPTS", + linux.IP_RECVORIGDSTADDR: "IP_RECVORIGDSTADDR", + linux.IP_RECVTTL: "IP_RECVTTL", + linux.IP_RETOPTS: "IP_RETOPTS", + linux.IP_TRANSPARENT: "IP_TRANSPARENT", + linux.IP_UNBLOCK_SOURCE: "IP_UNBLOCK_SOURCE", + linux.IP_UNICAST_IF: "IP_UNICAST_IF", + linux.IP_XFRM_POLICY: "IP_XFRM_POLICY", + linux.MCAST_BLOCK_SOURCE: "MCAST_BLOCK_SOURCE", + linux.MCAST_JOIN_SOURCE_GROUP: "MCAST_JOIN_SOURCE_GROUP", + linux.MCAST_LEAVE_GROUP: "MCAST_LEAVE_GROUP", + linux.MCAST_LEAVE_SOURCE_GROUP: "MCAST_LEAVE_SOURCE_GROUP", + linux.MCAST_MSFILTER: "MCAST_MSFILTER", + linux.MCAST_UNBLOCK_SOURCE: "MCAST_UNBLOCK_SOURCE", + linux.IP_ROUTER_ALERT: "IP_ROUTER_ALERT", + linux.IP_PKTOPTIONS: "IP_PKTOPTIONS", + linux.IP_MTU: "IP_MTU", + }, + linux.SOL_SOCKET: { + linux.SO_ERROR: "SO_ERROR", + linux.SO_PEERCRED: "SO_PEERCRED", + linux.SO_PASSCRED: "SO_PASSCRED", + linux.SO_SNDBUF: "SO_SNDBUF", + linux.SO_RCVBUF: "SO_RCVBUF", + linux.SO_REUSEADDR: "SO_REUSEADDR", + linux.SO_REUSEPORT: "SO_REUSEPORT", + linux.SO_BINDTODEVICE: "SO_BINDTODEVICE", + linux.SO_BROADCAST: "SO_BROADCAST", + linux.SO_KEEPALIVE: "SO_KEEPALIVE", + linux.SO_LINGER: "SO_LINGER", + linux.SO_SNDTIMEO: "SO_SNDTIMEO", + linux.SO_RCVTIMEO: "SO_RCVTIMEO", + linux.SO_OOBINLINE: "SO_OOBINLINE", + linux.SO_TIMESTAMP: "SO_TIMESTAMP", + }, + linux.SOL_TCP: { + linux.TCP_NODELAY: "TCP_NODELAY", + linux.TCP_CORK: "TCP_CORK", + linux.TCP_QUICKACK: "TCP_QUICKACK", + linux.TCP_MAXSEG: "TCP_MAXSEG", + linux.TCP_KEEPIDLE: "TCP_KEEPIDLE", + linux.TCP_KEEPINTVL: "TCP_KEEPINTVL", + linux.TCP_USER_TIMEOUT: "TCP_USER_TIMEOUT", + linux.TCP_INFO: "TCP_INFO", + linux.TCP_CC_INFO: "TCP_CC_INFO", + linux.TCP_NOTSENT_LOWAT: "TCP_NOTSENT_LOWAT", + linux.TCP_ZEROCOPY_RECEIVE: "TCP_ZEROCOPY_RECEIVE", + linux.TCP_CONGESTION: "TCP_CONGESTION", + linux.TCP_LINGER2: "TCP_LINGER2", + linux.TCP_DEFER_ACCEPT: "TCP_DEFER_ACCEPT", + linux.TCP_REPAIR_OPTIONS: "TCP_REPAIR_OPTIONS", + linux.TCP_INQ: "TCP_INQ", + linux.TCP_FASTOPEN: "TCP_FASTOPEN", + linux.TCP_FASTOPEN_CONNECT: "TCP_FASTOPEN_CONNECT", + linux.TCP_FASTOPEN_KEY: "TCP_FASTOPEN_KEY", + linux.TCP_FASTOPEN_NO_COOKIE: "TCP_FASTOPEN_NO_COOKIE", + linux.TCP_KEEPCNT: "TCP_KEEPCNT", + linux.TCP_QUEUE_SEQ: "TCP_QUEUE_SEQ", + linux.TCP_REPAIR: "TCP_REPAIR", + linux.TCP_REPAIR_QUEUE: "TCP_REPAIR_QUEUE", + linux.TCP_REPAIR_WINDOW: "TCP_REPAIR_WINDOW", + linux.TCP_SAVED_SYN: "TCP_SAVED_SYN", + linux.TCP_SAVE_SYN: "TCP_SAVE_SYN", + linux.TCP_SYNCNT: "TCP_SYNCNT", + linux.TCP_THIN_DUPACK: "TCP_THIN_DUPACK", + linux.TCP_THIN_LINEAR_TIMEOUTS: "TCP_THIN_LINEAR_TIMEOUTS", + linux.TCP_TIMESTAMP: "TCP_TIMESTAMP", + linux.TCP_ULP: "TCP_ULP", + linux.TCP_WINDOW_CLAMP: "TCP_WINDOW_CLAMP", + }, + linux.SOL_IPV6: { + linux.IPV6_V6ONLY: "IPV6_V6ONLY", + linux.IPV6_PATHMTU: "IPV6_PATHMTU", + linux.IPV6_TCLASS: "IPV6_TCLASS", + linux.IPV6_ADD_MEMBERSHIP: "IPV6_ADD_MEMBERSHIP", + linux.IPV6_DROP_MEMBERSHIP: "IPV6_DROP_MEMBERSHIP", + linux.IPV6_IPSEC_POLICY: "IPV6_IPSEC_POLICY", + linux.IPV6_JOIN_ANYCAST: "IPV6_JOIN_ANYCAST", + linux.IPV6_LEAVE_ANYCAST: "IPV6_LEAVE_ANYCAST", + linux.IPV6_PKTINFO: "IPV6_PKTINFO", + linux.IPV6_ROUTER_ALERT: "IPV6_ROUTER_ALERT", + linux.IPV6_XFRM_POLICY: "IPV6_XFRM_POLICY", + linux.MCAST_BLOCK_SOURCE: "MCAST_BLOCK_SOURCE", + linux.MCAST_JOIN_GROUP: "MCAST_JOIN_GROUP", + linux.MCAST_JOIN_SOURCE_GROUP: "MCAST_JOIN_SOURCE_GROUP", + linux.MCAST_LEAVE_GROUP: "MCAST_LEAVE_GROUP", + linux.MCAST_LEAVE_SOURCE_GROUP: "MCAST_LEAVE_SOURCE_GROUP", + linux.MCAST_UNBLOCK_SOURCE: "MCAST_UNBLOCK_SOURCE", + linux.IPV6_2292DSTOPTS: "IPV6_2292DSTOPTS", + linux.IPV6_2292HOPLIMIT: "IPV6_2292HOPLIMIT", + linux.IPV6_2292HOPOPTS: "IPV6_2292HOPOPTS", + linux.IPV6_2292PKTINFO: "IPV6_2292PKTINFO", + linux.IPV6_2292PKTOPTIONS: "IPV6_2292PKTOPTIONS", + linux.IPV6_2292RTHDR: "IPV6_2292RTHDR", + linux.IPV6_ADDR_PREFERENCES: "IPV6_ADDR_PREFERENCES", + linux.IPV6_AUTOFLOWLABEL: "IPV6_AUTOFLOWLABEL", + linux.IPV6_DONTFRAG: "IPV6_DONTFRAG", + linux.IPV6_DSTOPTS: "IPV6_DSTOPTS", + linux.IPV6_FLOWINFO: "IPV6_FLOWINFO", + linux.IPV6_FLOWINFO_SEND: "IPV6_FLOWINFO_SEND", + linux.IPV6_FLOWLABEL_MGR: "IPV6_FLOWLABEL_MGR", + linux.IPV6_FREEBIND: "IPV6_FREEBIND", + linux.IPV6_HOPOPTS: "IPV6_HOPOPTS", + linux.IPV6_MINHOPCOUNT: "IPV6_MINHOPCOUNT", + linux.IPV6_MTU: "IPV6_MTU", + linux.IPV6_MTU_DISCOVER: "IPV6_MTU_DISCOVER", + linux.IPV6_MULTICAST_ALL: "IPV6_MULTICAST_ALL", + linux.IPV6_MULTICAST_HOPS: "IPV6_MULTICAST_HOPS", + linux.IPV6_MULTICAST_IF: "IPV6_MULTICAST_IF", + linux.IPV6_MULTICAST_LOOP: "IPV6_MULTICAST_LOOP", + linux.IPV6_RECVDSTOPTS: "IPV6_RECVDSTOPTS", + linux.IPV6_RECVERR: "IPV6_RECVERR", + linux.IPV6_RECVFRAGSIZE: "IPV6_RECVFRAGSIZE", + linux.IPV6_RECVHOPLIMIT: "IPV6_RECVHOPLIMIT", + linux.IPV6_RECVHOPOPTS: "IPV6_RECVHOPOPTS", + linux.IPV6_RECVORIGDSTADDR: "IPV6_RECVORIGDSTADDR", + linux.IPV6_RECVPATHMTU: "IPV6_RECVPATHMTU", + linux.IPV6_RECVPKTINFO: "IPV6_RECVPKTINFO", + linux.IPV6_RECVRTHDR: "IPV6_RECVRTHDR", + linux.IPV6_RECVTCLASS: "IPV6_RECVTCLASS", + linux.IPV6_RTHDR: "IPV6_RTHDR", + linux.IPV6_RTHDRDSTOPTS: "IPV6_RTHDRDSTOPTS", + linux.IPV6_TRANSPARENT: "IPV6_TRANSPARENT", + linux.IPV6_UNICAST_HOPS: "IPV6_UNICAST_HOPS", + linux.IPV6_UNICAST_IF: "IPV6_UNICAST_IF", + linux.MCAST_MSFILTER: "MCAST_MSFILTER", + linux.IPV6_ADDRFORM: "IPV6_ADDRFORM", + }, +} diff --git a/pkg/sentry/strace/strace.go b/pkg/sentry/strace/strace.go index 3fc4a47fc..a796b2396 100644 --- a/pkg/sentry/strace/strace.go +++ b/pkg/sentry/strace/strace.go @@ -55,6 +55,14 @@ var ItimerTypes = abi.ValueSet{ linux.ITIMER_PROF: "ITIMER_PROF", } +func hexNum(num uint64) string { + return "0x" + strconv.FormatUint(num, 16) +} + +func hexArg(arg arch.SyscallArgument) string { + return hexNum(arg.Uint64()) +} + func iovecs(t *kernel.Task, addr usermem.Addr, iovcnt int, printContent bool, maxBytes uint64) string { if iovcnt < 0 || iovcnt > linux.UIO_MAXIOV { return fmt.Sprintf("%#x (error decoding iovecs: invalid iovcnt)", addr) @@ -389,6 +397,12 @@ func (i *SyscallInfo) pre(t *kernel.Task, args arch.SyscallArguments, maximumBlo output = append(output, path(t, args[arg].Pointer())) case ExecveStringVector: output = append(output, stringVector(t, args[arg].Pointer())) + case SetSockOptVal: + output = append(output, sockOptVal(t, args[arg-2].Uint64() /* level */, args[arg-1].Uint64() /* optName */, args[arg].Pointer() /* optVal */, args[arg+1].Uint64() /* optLen */, maximumBlobSize)) + case SockOptLevel: + output = append(output, sockOptLevels.Parse(args[arg].Uint64())) + case SockOptName: + output = append(output, sockOptNames[args[arg-1].Uint64() /* level */].Parse(args[arg].Uint64())) case SockAddr: output = append(output, sockAddr(t, args[arg].Pointer(), uint32(args[arg+1].Uint64()))) case SockLen: @@ -446,7 +460,7 @@ func (i *SyscallInfo) pre(t *kernel.Task, args arch.SyscallArguments, maximumBlo case Hex: fallthrough default: - output = append(output, "0x"+strconv.FormatUint(args[arg].Uint64(), 16)) + output = append(output, hexArg(args[arg])) } } @@ -507,6 +521,12 @@ func (i *SyscallInfo) post(t *kernel.Task, args arch.SyscallArguments, rval uint output[arg] = capData(t, args[arg-1].Pointer(), args[arg].Pointer()) case PollFDs: output[arg] = pollFDs(t, args[arg].Pointer(), uint(args[arg+1].Uint()), true) + case GetSockOptVal: + output[arg] = getSockOptVal(t, args[arg-2].Uint64() /* level */, args[arg-1].Uint64() /* optName */, args[arg].Pointer() /* optVal */, args[arg+1].Pointer() /* optLen */, maximumBlobSize, rval) + case SetSockOptVal: + // No need to print the value again. While it usually + // isn't, the string version of this arg can be long. + output[arg] = hexArg(args[arg]) } } } diff --git a/pkg/sentry/strace/syscalls.go b/pkg/sentry/strace/syscalls.go index 24e29a2ba..446d1e0f6 100644 --- a/pkg/sentry/strace/syscalls.go +++ b/pkg/sentry/strace/syscalls.go @@ -207,9 +207,27 @@ const ( // array is in the next argument. PollFDs - // SelectFDSet is an fd_set argument in select(2)/pselect(2). The number of - // fds represented must be the first argument. + // SelectFDSet is an fd_set argument in select(2)/pselect(2). The + // number of FDs represented must be the first argument. SelectFDSet + + // GetSockOptVal is the optval argument in getsockopt(2). + // + // Formatted after syscall execution. + GetSockOptVal + + // SetSockOptVal is the optval argument in setsockopt(2). + // + // Contents omitted after syscall execution. + SetSockOptVal + + // SockOptLevel is the level argument in getsockopt(2) and + // setsockopt(2). + SockOptLevel + + // SockOptLevel is the optname argument in getsockopt(2) and + // setsockopt(2). + SockOptName ) // defaultFormat is the syscall argument format to use if the actual format is -- cgit v1.2.3 From 17b9f5e66238bde1e4ed3bd9e5fb67342c8b58ec Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 7 Feb 2020 14:46:24 -0800 Subject: Support listxattr and removexattr syscalls. Note that these are only implemented for tmpfs, and other impls will still return EOPNOTSUPP. PiperOrigin-RevId: 293899385 --- pkg/p9/client_file.go | 33 ++++ pkg/p9/file.go | 16 ++ pkg/p9/handlers.go | 33 ++++ pkg/p9/messages.go | 199 +++++++++++++++---- pkg/p9/p9.go | 4 + pkg/p9/version.go | 8 +- pkg/sentry/fs/copy_up.go | 2 +- pkg/sentry/fs/fsutil/inode.go | 20 +- pkg/sentry/fs/gofer/context_file.go | 14 ++ pkg/sentry/fs/gofer/inode.go | 13 +- pkg/sentry/fs/inode.go | 14 +- pkg/sentry/fs/inode_operations.go | 13 +- pkg/sentry/fs/inode_overlay.go | 18 +- pkg/sentry/fs/tmpfs/tmpfs.go | 9 +- pkg/sentry/syscalls/linux/linux64_amd64.go | 27 ++- pkg/sentry/syscalls/linux/linux64_arm64.go | 37 ++-- pkg/sentry/syscalls/linux/sys_xattr.go | 200 +++++++++++++++++++- runsc/fsgofer/fsgofer.go | 16 +- test/syscalls/linux/BUILD | 1 + test/syscalls/linux/xattr.cc | 294 ++++++++++++++++------------- 20 files changed, 733 insertions(+), 238 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/p9/client_file.go b/pkg/p9/client_file.go index 0254e4ccc..2ee07b664 100644 --- a/pkg/p9/client_file.go +++ b/pkg/p9/client_file.go @@ -194,6 +194,39 @@ func (c *clientFile) SetXattr(name, value string, flags uint32) error { return c.client.sendRecv(&Tsetxattr{FID: c.fid, Name: name, Value: value, Flags: flags}, &Rsetxattr{}) } +// ListXattr implements File.ListXattr. +func (c *clientFile) ListXattr(size uint64) (map[string]struct{}, error) { + if atomic.LoadUint32(&c.closed) != 0 { + return nil, syscall.EBADF + } + if !versionSupportsListRemoveXattr(c.client.version) { + return nil, syscall.EOPNOTSUPP + } + + rlistxattr := Rlistxattr{} + if err := c.client.sendRecv(&Tlistxattr{FID: c.fid, Size: size}, &rlistxattr); err != nil { + return nil, err + } + + xattrs := make(map[string]struct{}, len(rlistxattr.Xattrs)) + for _, x := range rlistxattr.Xattrs { + xattrs[x] = struct{}{} + } + return xattrs, nil +} + +// RemoveXattr implements File.RemoveXattr. +func (c *clientFile) RemoveXattr(name string) error { + if atomic.LoadUint32(&c.closed) != 0 { + return syscall.EBADF + } + if !versionSupportsListRemoveXattr(c.client.version) { + return syscall.EOPNOTSUPP + } + + return c.client.sendRecv(&Tremovexattr{FID: c.fid, Name: name}, &Rremovexattr{}) +} + // Allocate implements File.Allocate. func (c *clientFile) Allocate(mode AllocateMode, offset, length uint64) error { if atomic.LoadUint32(&c.closed) != 0 { diff --git a/pkg/p9/file.go b/pkg/p9/file.go index 4607cfcdf..d4ffbc8e3 100644 --- a/pkg/p9/file.go +++ b/pkg/p9/file.go @@ -105,6 +105,22 @@ type File interface { // TODO(b/127675828): Determine concurrency guarantees once implemented. SetXattr(name, value string, flags uint32) error + // ListXattr lists the names of the extended attributes on this node. + // + // Size indicates the size of the buffer that has been allocated to hold the + // attribute list. If the list would be larger than size, implementations may + // return ERANGE to indicate that the buffer is too small, but they are also + // free to ignore the hint entirely (i.e. the value returned may be larger + // than size). All size checking is done independently at the syscall layer. + // + // TODO(b/148303075): Determine concurrency guarantees once implemented. + ListXattr(size uint64) (map[string]struct{}, error) + + // RemoveXattr removes extended attributes on this node. + // + // TODO(b/148303075): Determine concurrency guarantees once implemented. + RemoveXattr(name string) error + // Allocate allows the caller to directly manipulate the allocated disk space // for the file. See fallocate(2) for more details. Allocate(mode AllocateMode, offset, length uint64) error diff --git a/pkg/p9/handlers.go b/pkg/p9/handlers.go index 7d6653a07..2ac45eb80 100644 --- a/pkg/p9/handlers.go +++ b/pkg/p9/handlers.go @@ -941,6 +941,39 @@ func (t *Tsetxattr) handle(cs *connState) message { return &Rsetxattr{} } +// handle implements handler.handle. +func (t *Tlistxattr) handle(cs *connState) message { + ref, ok := cs.LookupFID(t.FID) + if !ok { + return newErr(syscall.EBADF) + } + defer ref.DecRef() + + xattrs, err := ref.file.ListXattr(t.Size) + if err != nil { + return newErr(err) + } + xattrList := make([]string, 0, len(xattrs)) + for x := range xattrs { + xattrList = append(xattrList, x) + } + return &Rlistxattr{Xattrs: xattrList} +} + +// handle implements handler.handle. +func (t *Tremovexattr) handle(cs *connState) message { + ref, ok := cs.LookupFID(t.FID) + if !ok { + return newErr(syscall.EBADF) + } + defer ref.DecRef() + + if err := ref.file.RemoveXattr(t.Name); err != nil { + return newErr(err) + } + return &Rremovexattr{} +} + // handle implements handler.handle. func (t *Treaddir) handle(cs *connState) message { ref, ok := cs.LookupFID(t.Directory) diff --git a/pkg/p9/messages.go b/pkg/p9/messages.go index ceb723d86..b1cede5f5 100644 --- a/pkg/p9/messages.go +++ b/pkg/p9/messages.go @@ -174,11 +174,11 @@ type Rflush struct { } // Decode implements encoder.Decode. -func (*Rflush) Decode(b *buffer) { +func (*Rflush) Decode(*buffer) { } // Encode implements encoder.Encode. -func (*Rflush) Encode(b *buffer) { +func (*Rflush) Encode(*buffer) { } // Type implements message.Type. @@ -188,7 +188,7 @@ func (*Rflush) Type() MsgType { // String implements fmt.Stringer. func (r *Rflush) String() string { - return fmt.Sprintf("RFlush{}") + return "RFlush{}" } // Twalk is a walk request. @@ -300,11 +300,11 @@ type Rclunk struct { } // Decode implements encoder.Decode. -func (*Rclunk) Decode(b *buffer) { +func (*Rclunk) Decode(*buffer) { } // Encode implements encoder.Encode. -func (*Rclunk) Encode(b *buffer) { +func (*Rclunk) Encode(*buffer) { } // Type implements message.Type. @@ -314,7 +314,7 @@ func (*Rclunk) Type() MsgType { // String implements fmt.Stringer. func (r *Rclunk) String() string { - return fmt.Sprintf("Rclunk{}") + return "Rclunk{}" } // Tremove is a remove request. @@ -350,11 +350,11 @@ type Rremove struct { } // Decode implements encoder.Decode. -func (*Rremove) Decode(b *buffer) { +func (*Rremove) Decode(*buffer) { } // Encode implements encoder.Encode. -func (*Rremove) Encode(b *buffer) { +func (*Rremove) Encode(*buffer) { } // Type implements message.Type. @@ -364,7 +364,7 @@ func (*Rremove) Type() MsgType { // String implements fmt.Stringer. func (r *Rremove) String() string { - return fmt.Sprintf("Rremove{}") + return "Rremove{}" } // Rlerror is an error response. @@ -745,16 +745,16 @@ func (*Rlink) Type() MsgType { } // Decode implements encoder.Decode. -func (*Rlink) Decode(b *buffer) { +func (*Rlink) Decode(*buffer) { } // Encode implements encoder.Encode. -func (*Rlink) Encode(b *buffer) { +func (*Rlink) Encode(*buffer) { } // String implements fmt.Stringer. func (r *Rlink) String() string { - return fmt.Sprintf("Rlink{}") + return "Rlink{}" } // Trenameat is a rename request. @@ -803,11 +803,11 @@ type Rrenameat struct { } // Decode implements encoder.Decode. -func (*Rrenameat) Decode(b *buffer) { +func (*Rrenameat) Decode(*buffer) { } // Encode implements encoder.Encode. -func (*Rrenameat) Encode(b *buffer) { +func (*Rrenameat) Encode(*buffer) { } // Type implements message.Type. @@ -817,7 +817,7 @@ func (*Rrenameat) Type() MsgType { // String implements fmt.Stringer. func (r *Rrenameat) String() string { - return fmt.Sprintf("Rrenameat{}") + return "Rrenameat{}" } // Tunlinkat is an unlink request. @@ -861,11 +861,11 @@ type Runlinkat struct { } // Decode implements encoder.Decode. -func (*Runlinkat) Decode(b *buffer) { +func (*Runlinkat) Decode(*buffer) { } // Encode implements encoder.Encode. -func (*Runlinkat) Encode(b *buffer) { +func (*Runlinkat) Encode(*buffer) { } // Type implements message.Type. @@ -875,7 +875,7 @@ func (*Runlinkat) Type() MsgType { // String implements fmt.Stringer. func (r *Runlinkat) String() string { - return fmt.Sprintf("Runlinkat{}") + return "Runlinkat{}" } // Trename is a rename request. @@ -922,11 +922,11 @@ type Rrename struct { } // Decode implements encoder.Decode. -func (*Rrename) Decode(b *buffer) { +func (*Rrename) Decode(*buffer) { } // Encode implements encoder.Encode. -func (*Rrename) Encode(b *buffer) { +func (*Rrename) Encode(*buffer) { } // Type implements message.Type. @@ -936,7 +936,7 @@ func (*Rrename) Type() MsgType { // String implements fmt.Stringer. func (r *Rrename) String() string { - return fmt.Sprintf("Rrename{}") + return "Rrename{}" } // Treadlink is a readlink request. @@ -1409,11 +1409,11 @@ type Rsetattr struct { } // Decode implements encoder.Decode. -func (*Rsetattr) Decode(b *buffer) { +func (*Rsetattr) Decode(*buffer) { } // Encode implements encoder.Encode. -func (*Rsetattr) Encode(b *buffer) { +func (*Rsetattr) Encode(*buffer) { } // Type implements message.Type. @@ -1423,7 +1423,7 @@ func (*Rsetattr) Type() MsgType { // String implements fmt.Stringer. func (r *Rsetattr) String() string { - return fmt.Sprintf("Rsetattr{}") + return "Rsetattr{}" } // Tallocate is an allocate request. This is an extension to 9P protocol, not @@ -1466,11 +1466,11 @@ type Rallocate struct { } // Decode implements encoder.Decode. -func (*Rallocate) Decode(b *buffer) { +func (*Rallocate) Decode(*buffer) { } // Encode implements encoder.Encode. -func (*Rallocate) Encode(b *buffer) { +func (*Rallocate) Encode(*buffer) { } // Type implements message.Type. @@ -1480,7 +1480,71 @@ func (*Rallocate) Type() MsgType { // String implements fmt.Stringer. func (r *Rallocate) String() string { - return fmt.Sprintf("Rallocate{}") + return "Rallocate{}" +} + +// Tlistxattr is a listxattr request. +type Tlistxattr struct { + // FID refers to the file on which to list xattrs. + FID FID + + // Size is the buffer size for the xattr list. + Size uint64 +} + +// Decode implements encoder.Decode. +func (t *Tlistxattr) Decode(b *buffer) { + t.FID = b.ReadFID() + t.Size = b.Read64() +} + +// Encode implements encoder.Encode. +func (t *Tlistxattr) Encode(b *buffer) { + b.WriteFID(t.FID) + b.Write64(t.Size) +} + +// Type implements message.Type. +func (*Tlistxattr) Type() MsgType { + return MsgTlistxattr +} + +// String implements fmt.Stringer. +func (t *Tlistxattr) String() string { + return fmt.Sprintf("Tlistxattr{FID: %d, Size: %d}", t.FID, t.Size) +} + +// Rlistxattr is a listxattr response. +type Rlistxattr struct { + // Xattrs is a list of extended attribute names. + Xattrs []string +} + +// Decode implements encoder.Decode. +func (r *Rlistxattr) Decode(b *buffer) { + n := b.Read16() + r.Xattrs = r.Xattrs[:0] + for i := 0; i < int(n); i++ { + r.Xattrs = append(r.Xattrs, b.ReadString()) + } +} + +// Encode implements encoder.Encode. +func (r *Rlistxattr) Encode(b *buffer) { + b.Write16(uint16(len(r.Xattrs))) + for _, x := range r.Xattrs { + b.WriteString(x) + } +} + +// Type implements message.Type. +func (*Rlistxattr) Type() MsgType { + return MsgRlistxattr +} + +// String implements fmt.Stringer. +func (r *Rlistxattr) String() string { + return fmt.Sprintf("Rlistxattr{Xattrs: %v}", r.Xattrs) } // Txattrwalk walks extended attributes. @@ -1594,11 +1658,11 @@ type Rxattrcreate struct { } // Decode implements encoder.Decode. -func (r *Rxattrcreate) Decode(b *buffer) { +func (r *Rxattrcreate) Decode(*buffer) { } // Encode implements encoder.Encode. -func (r *Rxattrcreate) Encode(b *buffer) { +func (r *Rxattrcreate) Encode(*buffer) { } // Type implements message.Type. @@ -1608,7 +1672,7 @@ func (*Rxattrcreate) Type() MsgType { // String implements fmt.Stringer. func (r *Rxattrcreate) String() string { - return fmt.Sprintf("Rxattrcreate{}") + return "Rxattrcreate{}" } // Tgetxattr is a getxattr request. @@ -1719,11 +1783,11 @@ type Rsetxattr struct { } // Decode implements encoder.Decode. -func (r *Rsetxattr) Decode(b *buffer) { +func (r *Rsetxattr) Decode(*buffer) { } // Encode implements encoder.Encode. -func (r *Rsetxattr) Encode(b *buffer) { +func (r *Rsetxattr) Encode(*buffer) { } // Type implements message.Type. @@ -1733,7 +1797,60 @@ func (*Rsetxattr) Type() MsgType { // String implements fmt.Stringer. func (r *Rsetxattr) String() string { - return fmt.Sprintf("Rsetxattr{}") + return "Rsetxattr{}" +} + +// Tremovexattr is a removexattr request. +type Tremovexattr struct { + // FID refers to the file on which to set xattrs. + FID FID + + // Name is the attribute name. + Name string +} + +// Decode implements encoder.Decode. +func (t *Tremovexattr) Decode(b *buffer) { + t.FID = b.ReadFID() + t.Name = b.ReadString() +} + +// Encode implements encoder.Encode. +func (t *Tremovexattr) Encode(b *buffer) { + b.WriteFID(t.FID) + b.WriteString(t.Name) +} + +// Type implements message.Type. +func (*Tremovexattr) Type() MsgType { + return MsgTremovexattr +} + +// String implements fmt.Stringer. +func (t *Tremovexattr) String() string { + return fmt.Sprintf("Tremovexattr{FID: %d, Name: %s}", t.FID, t.Name) +} + +// Rremovexattr is a removexattr response. +type Rremovexattr struct { +} + +// Decode implements encoder.Decode. +func (r *Rremovexattr) Decode(*buffer) { +} + +// Encode implements encoder.Encode. +func (r *Rremovexattr) Encode(*buffer) { +} + +// Type implements message.Type. +func (*Rremovexattr) Type() MsgType { + return MsgRremovexattr +} + +// String implements fmt.Stringer. +func (r *Rremovexattr) String() string { + return "Rremovexattr{}" } // Treaddir is a readdir request. @@ -1880,11 +1997,11 @@ type Rfsync struct { } // Decode implements encoder.Decode. -func (*Rfsync) Decode(b *buffer) { +func (*Rfsync) Decode(*buffer) { } // Encode implements encoder.Encode. -func (*Rfsync) Encode(b *buffer) { +func (*Rfsync) Encode(*buffer) { } // Type implements message.Type. @@ -1894,7 +2011,7 @@ func (*Rfsync) Type() MsgType { // String implements fmt.Stringer. func (r *Rfsync) String() string { - return fmt.Sprintf("Rfsync{}") + return "Rfsync{}" } // Tstatfs is a stat request. @@ -1980,11 +2097,11 @@ type Rflushf struct { } // Decode implements encoder.Decode. -func (*Rflushf) Decode(b *buffer) { +func (*Rflushf) Decode(*buffer) { } // Encode implements encoder.Encode. -func (*Rflushf) Encode(b *buffer) { +func (*Rflushf) Encode(*buffer) { } // Type implements message.Type. @@ -1994,7 +2111,7 @@ func (*Rflushf) Type() MsgType { // String implements fmt.Stringer. func (*Rflushf) String() string { - return fmt.Sprintf("Rflushf{}") + return "Rflushf{}" } // Twalkgetattr is a walk request. @@ -2484,6 +2601,8 @@ func init() { msgRegistry.register(MsgRgetattr, func() message { return &Rgetattr{} }) msgRegistry.register(MsgTsetattr, func() message { return &Tsetattr{} }) msgRegistry.register(MsgRsetattr, func() message { return &Rsetattr{} }) + msgRegistry.register(MsgTlistxattr, func() message { return &Tlistxattr{} }) + msgRegistry.register(MsgRlistxattr, func() message { return &Rlistxattr{} }) msgRegistry.register(MsgTxattrwalk, func() message { return &Txattrwalk{} }) msgRegistry.register(MsgRxattrwalk, func() message { return &Rxattrwalk{} }) msgRegistry.register(MsgTxattrcreate, func() message { return &Txattrcreate{} }) @@ -2492,6 +2611,8 @@ func init() { msgRegistry.register(MsgRgetxattr, func() message { return &Rgetxattr{} }) msgRegistry.register(MsgTsetxattr, func() message { return &Tsetxattr{} }) msgRegistry.register(MsgRsetxattr, func() message { return &Rsetxattr{} }) + msgRegistry.register(MsgTremovexattr, func() message { return &Tremovexattr{} }) + msgRegistry.register(MsgRremovexattr, func() message { return &Rremovexattr{} }) msgRegistry.register(MsgTreaddir, func() message { return &Treaddir{} }) msgRegistry.register(MsgRreaddir, func() message { return &Rreaddir{} }) msgRegistry.register(MsgTfsync, func() message { return &Tfsync{} }) diff --git a/pkg/p9/p9.go b/pkg/p9/p9.go index 5ab00d625..20ab31f7a 100644 --- a/pkg/p9/p9.go +++ b/pkg/p9/p9.go @@ -335,6 +335,8 @@ const ( MsgRgetattr = 25 MsgTsetattr = 26 MsgRsetattr = 27 + MsgTlistxattr = 28 + MsgRlistxattr = 29 MsgTxattrwalk = 30 MsgRxattrwalk = 31 MsgTxattrcreate = 32 @@ -343,6 +345,8 @@ const ( MsgRgetxattr = 35 MsgTsetxattr = 36 MsgRsetxattr = 37 + MsgTremovexattr = 38 + MsgRremovexattr = 39 MsgTreaddir = 40 MsgRreaddir = 41 MsgTfsync = 50 diff --git a/pkg/p9/version.go b/pkg/p9/version.go index 34a15eb55..09cde9f5a 100644 --- a/pkg/p9/version.go +++ b/pkg/p9/version.go @@ -26,7 +26,7 @@ const ( // // Clients are expected to start requesting this version number and // to continuously decrement it until a Tversion request succeeds. - highestSupportedVersion uint32 = 10 + highestSupportedVersion uint32 = 11 // lowestSupportedVersion is the lowest supported version X in a // version string of the format 9P2000.L.Google.X. @@ -167,3 +167,9 @@ func VersionSupportsOpenTruncateFlag(v uint32) bool { func versionSupportsGetSetXattr(v uint32) bool { return v >= 10 } + +// versionSupportsListRemoveXattr returns true if version v supports +// the Tlistxattr and Tremovexattr messages. +func versionSupportsListRemoveXattr(v uint32) bool { + return v >= 11 +} diff --git a/pkg/sentry/fs/copy_up.go b/pkg/sentry/fs/copy_up.go index f6c79e51b..b060a12ff 100644 --- a/pkg/sentry/fs/copy_up.go +++ b/pkg/sentry/fs/copy_up.go @@ -401,7 +401,7 @@ func copyAttributesLocked(ctx context.Context, upper *Inode, lower *Inode) error if err != nil { return err } - lowerXattr, err := lower.ListXattr(ctx) + lowerXattr, err := lower.ListXattr(ctx, linux.XATTR_SIZE_MAX) if err != nil && err != syserror.EOPNOTSUPP { return err } diff --git a/pkg/sentry/fs/fsutil/inode.go b/pkg/sentry/fs/fsutil/inode.go index 252830572..daecc4ffe 100644 --- a/pkg/sentry/fs/fsutil/inode.go +++ b/pkg/sentry/fs/fsutil/inode.go @@ -247,7 +247,7 @@ func (i *InodeSimpleExtendedAttributes) SetXattr(_ context.Context, _ *fs.Inode, } // ListXattr implements fs.InodeOperations.ListXattr. -func (i *InodeSimpleExtendedAttributes) ListXattr(context.Context, *fs.Inode) (map[string]struct{}, error) { +func (i *InodeSimpleExtendedAttributes) ListXattr(context.Context, *fs.Inode, uint64) (map[string]struct{}, error) { i.mu.RLock() names := make(map[string]struct{}, len(i.xattrs)) for name := range i.xattrs { @@ -257,6 +257,17 @@ func (i *InodeSimpleExtendedAttributes) ListXattr(context.Context, *fs.Inode) (m return names, nil } +// RemoveXattr implements fs.InodeOperations.RemoveXattr. +func (i *InodeSimpleExtendedAttributes) RemoveXattr(_ context.Context, _ *fs.Inode, name string) error { + i.mu.RLock() + defer i.mu.RUnlock() + if _, ok := i.xattrs[name]; ok { + delete(i.xattrs, name) + return nil + } + return syserror.ENOATTR +} + // staticFile is a file with static contents. It is returned by // InodeStaticFileGetter.GetFile. // @@ -460,10 +471,15 @@ func (InodeNoExtendedAttributes) SetXattr(context.Context, *fs.Inode, string, st } // ListXattr implements fs.InodeOperations.ListXattr. -func (InodeNoExtendedAttributes) ListXattr(context.Context, *fs.Inode) (map[string]struct{}, error) { +func (InodeNoExtendedAttributes) ListXattr(context.Context, *fs.Inode, uint64) (map[string]struct{}, error) { return nil, syserror.EOPNOTSUPP } +// RemoveXattr implements fs.InodeOperations.RemoveXattr. +func (InodeNoExtendedAttributes) RemoveXattr(context.Context, *fs.Inode, string) error { + return syserror.EOPNOTSUPP +} + // InodeNoopRelease implements fs.InodeOperations.Release as a noop. type InodeNoopRelease struct{} diff --git a/pkg/sentry/fs/gofer/context_file.go b/pkg/sentry/fs/gofer/context_file.go index 3da818aed..125907d70 100644 --- a/pkg/sentry/fs/gofer/context_file.go +++ b/pkg/sentry/fs/gofer/context_file.go @@ -73,6 +73,20 @@ func (c *contextFile) setXattr(ctx context.Context, name, value string, flags ui return err } +func (c *contextFile) listXattr(ctx context.Context, size uint64) (map[string]struct{}, error) { + ctx.UninterruptibleSleepStart(false) + xattrs, err := c.file.ListXattr(size) + ctx.UninterruptibleSleepFinish(false) + return xattrs, err +} + +func (c *contextFile) removeXattr(ctx context.Context, name string) error { + ctx.UninterruptibleSleepStart(false) + err := c.file.RemoveXattr(name) + ctx.UninterruptibleSleepFinish(false) + return err +} + func (c *contextFile) allocate(ctx context.Context, mode p9.AllocateMode, offset, length uint64) error { ctx.UninterruptibleSleepStart(false) err := c.file.Allocate(mode, offset, length) diff --git a/pkg/sentry/fs/gofer/inode.go b/pkg/sentry/fs/gofer/inode.go index ac28174d2..1c934981b 100644 --- a/pkg/sentry/fs/gofer/inode.go +++ b/pkg/sentry/fs/gofer/inode.go @@ -604,18 +604,23 @@ func (i *inodeOperations) Truncate(ctx context.Context, inode *fs.Inode, length } // GetXattr implements fs.InodeOperations.GetXattr. -func (i *inodeOperations) GetXattr(ctx context.Context, inode *fs.Inode, name string, size uint64) (string, error) { +func (i *inodeOperations) GetXattr(ctx context.Context, _ *fs.Inode, name string, size uint64) (string, error) { return i.fileState.file.getXattr(ctx, name, size) } // SetXattr implements fs.InodeOperations.SetXattr. -func (i *inodeOperations) SetXattr(ctx context.Context, inode *fs.Inode, name string, value string, flags uint32) error { +func (i *inodeOperations) SetXattr(ctx context.Context, _ *fs.Inode, name string, value string, flags uint32) error { return i.fileState.file.setXattr(ctx, name, value, flags) } // ListXattr implements fs.InodeOperations.ListXattr. -func (i *inodeOperations) ListXattr(context.Context, *fs.Inode) (map[string]struct{}, error) { - return nil, syscall.EOPNOTSUPP +func (i *inodeOperations) ListXattr(ctx context.Context, _ *fs.Inode, size uint64) (map[string]struct{}, error) { + return i.fileState.file.listXattr(ctx, size) +} + +// RemoveXattr implements fs.InodeOperations.RemoveXattr. +func (i *inodeOperations) RemoveXattr(ctx context.Context, _ *fs.Inode, name string) error { + return i.fileState.file.removeXattr(ctx, name) } // Allocate implements fs.InodeOperations.Allocate. diff --git a/pkg/sentry/fs/inode.go b/pkg/sentry/fs/inode.go index b66c091ab..55fb71c16 100644 --- a/pkg/sentry/fs/inode.go +++ b/pkg/sentry/fs/inode.go @@ -278,11 +278,19 @@ func (i *Inode) SetXattr(ctx context.Context, d *Dirent, name, value string, fla } // ListXattr calls i.InodeOperations.ListXattr with i as the Inode. -func (i *Inode) ListXattr(ctx context.Context) (map[string]struct{}, error) { +func (i *Inode) ListXattr(ctx context.Context, size uint64) (map[string]struct{}, error) { if i.overlay != nil { - return overlayListXattr(ctx, i.overlay) + return overlayListXattr(ctx, i.overlay, size) } - return i.InodeOperations.ListXattr(ctx, i) + return i.InodeOperations.ListXattr(ctx, i, size) +} + +// RemoveXattr calls i.InodeOperations.RemoveXattr with i as the Inode. +func (i *Inode) RemoveXattr(ctx context.Context, d *Dirent, name string) error { + if i.overlay != nil { + return overlayRemoveXattr(ctx, i.overlay, d, name) + } + return i.InodeOperations.RemoveXattr(ctx, i, name) } // CheckPermission will check if the caller may access this file in the diff --git a/pkg/sentry/fs/inode_operations.go b/pkg/sentry/fs/inode_operations.go index 70f2eae96..2bbfb72ef 100644 --- a/pkg/sentry/fs/inode_operations.go +++ b/pkg/sentry/fs/inode_operations.go @@ -190,7 +190,18 @@ type InodeOperations interface { // ListXattr returns the set of all extended attributes names that // have values. Inodes that do not support extended attributes return // EOPNOTSUPP. - ListXattr(ctx context.Context, inode *Inode) (map[string]struct{}, error) + // + // If this is called through the listxattr(2) syscall, size indicates the + // size of the buffer that the application has allocated to hold the + // attribute list. If the list would be larger than size, implementations may + // return ERANGE to indicate that the buffer is too small, but they are also + // free to ignore the hint entirely. All size checking is done independently + // at the syscall layer. + ListXattr(ctx context.Context, inode *Inode, size uint64) (map[string]struct{}, error) + + // RemoveXattr removes an extended attribute specified by name. Inodes that + // do not support extended attributes return EOPNOTSUPP. + RemoveXattr(ctx context.Context, inode *Inode, name string) error // Check determines whether an Inode can be accessed with the // requested permission mask using the context (which gives access diff --git a/pkg/sentry/fs/inode_overlay.go b/pkg/sentry/fs/inode_overlay.go index 4729b4aac..5ada33a32 100644 --- a/pkg/sentry/fs/inode_overlay.go +++ b/pkg/sentry/fs/inode_overlay.go @@ -564,15 +564,15 @@ func overlaySetxattr(ctx context.Context, o *overlayEntry, d *Dirent, name, valu return o.upper.SetXattr(ctx, d, name, value, flags) } -func overlayListXattr(ctx context.Context, o *overlayEntry) (map[string]struct{}, error) { +func overlayListXattr(ctx context.Context, o *overlayEntry, size uint64) (map[string]struct{}, error) { o.copyMu.RLock() defer o.copyMu.RUnlock() var names map[string]struct{} var err error if o.upper != nil { - names, err = o.upper.ListXattr(ctx) + names, err = o.upper.ListXattr(ctx, size) } else { - names, err = o.lower.ListXattr(ctx) + names, err = o.lower.ListXattr(ctx, size) } for name := range names { // Same as overlayGetXattr, we shouldn't forward along @@ -584,6 +584,18 @@ func overlayListXattr(ctx context.Context, o *overlayEntry) (map[string]struct{} return names, err } +func overlayRemoveXattr(ctx context.Context, o *overlayEntry, d *Dirent, name string) error { + // Don't allow changes to overlay xattrs through a removexattr syscall. + if strings.HasPrefix(XattrOverlayPrefix, name) { + return syserror.EPERM + } + + if err := copyUp(ctx, d); err != nil { + return err + } + return o.upper.RemoveXattr(ctx, d, name) +} + func overlayCheck(ctx context.Context, o *overlayEntry, p PermMask) error { o.copyMu.RLock() // Hot path. Avoid defers. diff --git a/pkg/sentry/fs/tmpfs/tmpfs.go b/pkg/sentry/fs/tmpfs/tmpfs.go index c00cef0a5..3c2b583ae 100644 --- a/pkg/sentry/fs/tmpfs/tmpfs.go +++ b/pkg/sentry/fs/tmpfs/tmpfs.go @@ -159,8 +159,13 @@ func (d *Dir) SetXattr(ctx context.Context, i *fs.Inode, name, value string, fla } // ListXattr implements fs.InodeOperations.ListXattr. -func (d *Dir) ListXattr(ctx context.Context, i *fs.Inode) (map[string]struct{}, error) { - return d.ramfsDir.ListXattr(ctx, i) +func (d *Dir) ListXattr(ctx context.Context, i *fs.Inode, size uint64) (map[string]struct{}, error) { + return d.ramfsDir.ListXattr(ctx, i, size) +} + +// RemoveXattr implements fs.InodeOperations.RemoveXattr. +func (d *Dir) RemoveXattr(ctx context.Context, i *fs.Inode, name string) error { + return d.ramfsDir.RemoveXattr(ctx, i, name) } // Lookup implements fs.InodeOperations.Lookup. diff --git a/pkg/sentry/syscalls/linux/linux64_amd64.go b/pkg/sentry/syscalls/linux/linux64_amd64.go index 588f8b087..79066ad2a 100644 --- a/pkg/sentry/syscalls/linux/linux64_amd64.go +++ b/pkg/sentry/syscalls/linux/linux64_amd64.go @@ -228,21 +228,18 @@ var AMD64 = &kernel.SyscallTable{ 185: syscalls.Error("security", syserror.ENOSYS, "Not implemented in Linux.", nil), 186: syscalls.Supported("gettid", Gettid), 187: syscalls.Supported("readahead", Readahead), - // TODO(b/148303075): Enable set/getxattr (in their various - // forms) once we also have list and removexattr. The JVM - // assumes that if get/set exist, then list and remove do too. - 188: syscalls.ErrorWithEvent("setxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), - 189: syscalls.ErrorWithEvent("lsetxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), - 190: syscalls.ErrorWithEvent("fsetxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), - 191: syscalls.ErrorWithEvent("getxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), - 192: syscalls.ErrorWithEvent("lgetxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), - 193: syscalls.ErrorWithEvent("fgetxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), - 194: syscalls.ErrorWithEvent("listxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), - 195: syscalls.ErrorWithEvent("llistxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), - 196: syscalls.ErrorWithEvent("flistxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), - 197: syscalls.ErrorWithEvent("removexattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), - 198: syscalls.ErrorWithEvent("lremovexattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), - 199: syscalls.ErrorWithEvent("fremovexattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 188: syscalls.PartiallySupported("setxattr", SetXattr, "Only supported for tmpfs.", nil), + 189: syscalls.PartiallySupported("lsetxattr", LSetXattr, "Only supported for tmpfs.", nil), + 190: syscalls.PartiallySupported("fsetxattr", FSetXattr, "Only supported for tmpfs.", nil), + 191: syscalls.PartiallySupported("getxattr", GetXattr, "Only supported for tmpfs.", nil), + 192: syscalls.PartiallySupported("lgetxattr", LGetXattr, "Only supported for tmpfs.", nil), + 193: syscalls.PartiallySupported("fgetxattr", FGetXattr, "Only supported for tmpfs.", nil), + 194: syscalls.PartiallySupported("listxattr", ListXattr, "Only supported for tmpfs", nil), + 195: syscalls.PartiallySupported("llistxattr", LListXattr, "Only supported for tmpfs", nil), + 196: syscalls.PartiallySupported("flistxattr", FListXattr, "Only supported for tmpfs", nil), + 197: syscalls.PartiallySupported("removexattr", RemoveXattr, "Only supported for tmpfs", nil), + 198: syscalls.PartiallySupported("lremovexattr", LRemoveXattr, "Only supported for tmpfs", nil), + 199: syscalls.PartiallySupported("fremovexattr", FRemoveXattr, "Only supported for tmpfs", nil), 200: syscalls.Supported("tkill", Tkill), 201: syscalls.Supported("time", Time), 202: syscalls.PartiallySupported("futex", Futex, "Robust futexes not supported.", nil), diff --git a/pkg/sentry/syscalls/linux/linux64_arm64.go b/pkg/sentry/syscalls/linux/linux64_arm64.go index 06e5ee401..7421619de 100644 --- a/pkg/sentry/syscalls/linux/linux64_arm64.go +++ b/pkg/sentry/syscalls/linux/linux64_arm64.go @@ -36,26 +36,23 @@ var ARM64 = &kernel.SyscallTable{ }, AuditNumber: linux.AUDIT_ARCH_AARCH64, Table: map[uintptr]kernel.Syscall{ - 0: syscalls.PartiallySupported("io_setup", IoSetup, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 1: syscalls.PartiallySupported("io_destroy", IoDestroy, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 2: syscalls.PartiallySupported("io_submit", IoSubmit, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 3: syscalls.PartiallySupported("io_cancel", IoCancel, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 4: syscalls.PartiallySupported("io_getevents", IoGetevents, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - // TODO(b/148303075): Enable set/getxattr (in their various - // forms) once we also have list and removexattr. The JVM - // assumes that if get/set exist, then list and remove do too. - 5: syscalls.ErrorWithEvent("setxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), - 6: syscalls.ErrorWithEvent("lsetxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), - 7: syscalls.ErrorWithEvent("fsetxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), - 8: syscalls.ErrorWithEvent("getxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), - 9: syscalls.ErrorWithEvent("lgetxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), - 10: syscalls.ErrorWithEvent("fgetxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), - 11: syscalls.ErrorWithEvent("listxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), - 13: syscalls.ErrorWithEvent("llistxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), - 13: syscalls.ErrorWithEvent("flistxattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), - 14: syscalls.ErrorWithEvent("removexattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), - 15: syscalls.ErrorWithEvent("lremovexattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), - 16: syscalls.ErrorWithEvent("fremovexattr", syserror.ENOTSUP, "Requires filesystem support.", []string{"gvisor.dev/issue/1636"}), + 0: syscalls.PartiallySupported("io_setup", IoSetup, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 1: syscalls.PartiallySupported("io_destroy", IoDestroy, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 2: syscalls.PartiallySupported("io_submit", IoSubmit, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 3: syscalls.PartiallySupported("io_cancel", IoCancel, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 4: syscalls.PartiallySupported("io_getevents", IoGetevents, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 5: syscalls.PartiallySupported("setxattr", SetXattr, "Only supported for tmpfs.", nil), + 6: syscalls.PartiallySupported("lsetxattr", LSetXattr, "Only supported for tmpfs.", nil), + 7: syscalls.PartiallySupported("fsetxattr", FSetXattr, "Only supported for tmpfs.", nil), + 8: syscalls.PartiallySupported("getxattr", GetXattr, "Only supported for tmpfs.", nil), + 9: syscalls.PartiallySupported("lgetxattr", LGetXattr, "Only supported for tmpfs.", nil), + 10: syscalls.PartiallySupported("fgetxattr", FGetXattr, "Only supported for tmpfs.", nil), + 11: syscalls.PartiallySupported("listxattr", ListXattr, "Only supported for tmpfs", nil), + 12: syscalls.PartiallySupported("llistxattr", LListXattr, "Only supported for tmpfs", nil), + 13: syscalls.PartiallySupported("flistxattr", FListXattr, "Only supported for tmpfs", nil), + 14: syscalls.PartiallySupported("removexattr", RemoveXattr, "Only supported for tmpfs", nil), + 15: syscalls.PartiallySupported("lremovexattr", LRemoveXattr, "Only supported for tmpfs", nil), + 16: syscalls.PartiallySupported("fremovexattr", FRemoveXattr, "Only supported for tmpfs", nil), 17: syscalls.Supported("getcwd", Getcwd), 18: syscalls.CapError("lookup_dcookie", linux.CAP_SYS_ADMIN, "", nil), 19: syscalls.Supported("eventfd2", Eventfd2), diff --git a/pkg/sentry/syscalls/linux/sys_xattr.go b/pkg/sentry/syscalls/linux/sys_xattr.go index efb95555c..342337726 100644 --- a/pkg/sentry/syscalls/linux/sys_xattr.go +++ b/pkg/sentry/syscalls/linux/sys_xattr.go @@ -72,7 +72,7 @@ func getXattrFromPath(t *kernel.Task, args arch.SyscallArguments, resolveSymlink } valueLen := 0 - err = fileOpOn(t, linux.AT_FDCWD, path, resolveSymlink, func(root *fs.Dirent, d *fs.Dirent, _ uint) error { + err = fileOpOn(t, linux.AT_FDCWD, path, resolveSymlink, func(_ *fs.Dirent, d *fs.Dirent, _ uint) error { if dirPath && !fs.IsDir(d.Inode.StableAttr) { return syserror.ENOTDIR } @@ -172,7 +172,7 @@ func setXattrFromPath(t *kernel.Task, args arch.SyscallArguments, resolveSymlink return 0, nil, err } - return 0, nil, fileOpOn(t, linux.AT_FDCWD, path, resolveSymlink, func(root *fs.Dirent, d *fs.Dirent, _ uint) error { + return 0, nil, fileOpOn(t, linux.AT_FDCWD, path, resolveSymlink, func(_ *fs.Dirent, d *fs.Dirent, _ uint) error { if dirPath && !fs.IsDir(d.Inode.StableAttr) { return syserror.ENOTDIR } @@ -187,12 +187,12 @@ func setXattr(t *kernel.Task, d *fs.Dirent, nameAddr, valueAddr usermem.Addr, si return syserror.EINVAL } - if err := checkXattrPermissions(t, d.Inode, fs.PermMask{Write: true}); err != nil { + name, err := copyInXattrName(t, nameAddr) + if err != nil { return err } - name, err := copyInXattrName(t, nameAddr) - if err != nil { + if err := checkXattrPermissions(t, d.Inode, fs.PermMask{Write: true}); err != nil { return err } @@ -226,12 +226,18 @@ func copyInXattrName(t *kernel.Task, nameAddr usermem.Addr) (string, error) { return name, nil } +// Restrict xattrs to regular files and directories. +// +// TODO(b/148380782): In Linux, this restriction technically only applies to +// xattrs in the "user.*" namespace. Make file type checks specific to the +// namespace once we allow other xattr prefixes. +func xattrFileTypeOk(i *fs.Inode) bool { + return fs.IsRegular(i.StableAttr) || fs.IsDir(i.StableAttr) +} + func checkXattrPermissions(t *kernel.Task, i *fs.Inode, perms fs.PermMask) error { // Restrict xattrs to regular files and directories. - // - // In Linux, this restriction technically only applies to xattrs in the - // "user.*" namespace, but we don't allow any other xattr prefixes anyway. - if !fs.IsRegular(i.StableAttr) && !fs.IsDir(i.StableAttr) { + if !xattrFileTypeOk(i) { if perms.Write { return syserror.EPERM } @@ -240,3 +246,179 @@ func checkXattrPermissions(t *kernel.Task, i *fs.Inode, perms fs.PermMask) error return i.CheckPermission(t, perms) } + +// ListXattr implements linux syscall listxattr(2). +func ListXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + return listXattrFromPath(t, args, true) +} + +// LListXattr implements linux syscall llistxattr(2). +func LListXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + return listXattrFromPath(t, args, false) +} + +// FListXattr implements linux syscall flistxattr(2). +func FListXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + listAddr := args[1].Pointer() + size := uint64(args[2].SizeT()) + + // TODO(b/113957122): Return EBADF if the fd was opened with O_PATH. + f := t.GetFile(fd) + if f == nil { + return 0, nil, syserror.EBADF + } + defer f.DecRef() + + n, err := listXattr(t, f.Dirent, listAddr, size) + if err != nil { + return 0, nil, err + } + + return uintptr(n), nil, nil +} + +func listXattrFromPath(t *kernel.Task, args arch.SyscallArguments, resolveSymlink bool) (uintptr, *kernel.SyscallControl, error) { + pathAddr := args[0].Pointer() + listAddr := args[1].Pointer() + size := uint64(args[2].SizeT()) + + path, dirPath, err := copyInPath(t, pathAddr, false /* allowEmpty */) + if err != nil { + return 0, nil, err + } + + n := 0 + err = fileOpOn(t, linux.AT_FDCWD, path, resolveSymlink, func(_ *fs.Dirent, d *fs.Dirent, _ uint) error { + if dirPath && !fs.IsDir(d.Inode.StableAttr) { + return syserror.ENOTDIR + } + + n, err = listXattr(t, d, listAddr, size) + return err + }) + if err != nil { + return 0, nil, err + } + + return uintptr(n), nil, nil +} + +func listXattr(t *kernel.Task, d *fs.Dirent, addr usermem.Addr, size uint64) (int, error) { + if !xattrFileTypeOk(d.Inode) { + return 0, nil + } + + // If listxattr(2) is called with size 0, the buffer size needed to contain + // the xattr list will be returned successfully even if it is nonzero. In + // that case, we need to retrieve the entire list so we can compute and + // return the correct size. + requestedSize := size + if size == 0 || size > linux.XATTR_SIZE_MAX { + requestedSize = linux.XATTR_SIZE_MAX + } + xattrs, err := d.Inode.ListXattr(t, requestedSize) + if err != nil { + return 0, err + } + + // TODO(b/148380782): support namespaces other than "user". + for x := range xattrs { + if !strings.HasPrefix(x, linux.XATTR_USER_PREFIX) { + delete(xattrs, x) + } + } + + listSize := xattrListSize(xattrs) + if listSize > linux.XATTR_SIZE_MAX { + return 0, syserror.E2BIG + } + if uint64(listSize) > requestedSize { + return 0, syserror.ERANGE + } + + // Don't copy out the attributes if size is 0. + if size == 0 { + return listSize, nil + } + + buf := make([]byte, 0, listSize) + for x := range xattrs { + buf = append(buf, []byte(x)...) + buf = append(buf, 0) + } + if _, err := t.CopyOutBytes(addr, buf); err != nil { + return 0, err + } + + return len(buf), nil +} + +func xattrListSize(xattrs map[string]struct{}) int { + size := 0 + for x := range xattrs { + size += len(x) + 1 + } + return size +} + +// RemoveXattr implements linux syscall removexattr(2). +func RemoveXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + return removeXattrFromPath(t, args, true) +} + +// LRemoveXattr implements linux syscall lremovexattr(2). +func LRemoveXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + return removeXattrFromPath(t, args, false) +} + +// FRemoveXattr implements linux syscall fremovexattr(2). +func FRemoveXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + nameAddr := args[1].Pointer() + + // TODO(b/113957122): Return EBADF if the fd was opened with O_PATH. + f := t.GetFile(fd) + if f == nil { + return 0, nil, syserror.EBADF + } + defer f.DecRef() + + return 0, nil, removeXattr(t, f.Dirent, nameAddr) +} + +func removeXattrFromPath(t *kernel.Task, args arch.SyscallArguments, resolveSymlink bool) (uintptr, *kernel.SyscallControl, error) { + pathAddr := args[0].Pointer() + nameAddr := args[1].Pointer() + + path, dirPath, err := copyInPath(t, pathAddr, false /* allowEmpty */) + if err != nil { + return 0, nil, err + } + + return 0, nil, fileOpOn(t, linux.AT_FDCWD, path, resolveSymlink, func(_ *fs.Dirent, d *fs.Dirent, _ uint) error { + if dirPath && !fs.IsDir(d.Inode.StableAttr) { + return syserror.ENOTDIR + } + + return removeXattr(t, d, nameAddr) + }) +} + +// removeXattr implements removexattr(2) from the given *fs.Dirent. +func removeXattr(t *kernel.Task, d *fs.Dirent, nameAddr usermem.Addr) error { + name, err := copyInXattrName(t, nameAddr) + if err != nil { + return err + } + + if err := checkXattrPermissions(t, d.Inode, fs.PermMask{Write: true}); err != nil { + return err + } + + if !strings.HasPrefix(name, linux.XATTR_USER_PREFIX) { + return syserror.EOPNOTSUPP + } + + return d.Inode.RemoveXattr(t, d, name) +} diff --git a/runsc/fsgofer/fsgofer.go b/runsc/fsgofer/fsgofer.go index 4d84ad999..cadd83273 100644 --- a/runsc/fsgofer/fsgofer.go +++ b/runsc/fsgofer/fsgofer.go @@ -768,12 +768,22 @@ func (l *localFile) SetAttr(valid p9.SetAttrMask, attr p9.SetAttr) error { } // TODO(b/127675828): support getxattr. -func (l *localFile) GetXattr(name string, size uint64) (string, error) { +func (*localFile) GetXattr(string, uint64) (string, error) { return "", syscall.EOPNOTSUPP } // TODO(b/127675828): support setxattr. -func (l *localFile) SetXattr(name, value string, flags uint32) error { +func (*localFile) SetXattr(string, string, uint32) error { + return syscall.EOPNOTSUPP +} + +// TODO(b/148303075): support listxattr. +func (*localFile) ListXattr(uint64) (map[string]struct{}, error) { + return nil, syscall.EOPNOTSUPP +} + +// TODO(b/148303075): support removexattr. +func (*localFile) RemoveXattr(string) error { return syscall.EOPNOTSUPP } @@ -790,7 +800,7 @@ func (l *localFile) Allocate(mode p9.AllocateMode, offset, length uint64) error } // Rename implements p9.File; this should never be called. -func (l *localFile) Rename(p9.File, string) error { +func (*localFile) Rename(p9.File, string) error { panic("rename called directly") } diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 12d389c3e..ca1af209a 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -3782,6 +3782,7 @@ cc_binary( "//test/util:capability_util", "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/container:flat_hash_set", "@com_google_absl//absl/strings", gtest, "//test/util:posix_error", diff --git a/test/syscalls/linux/xattr.cc b/test/syscalls/linux/xattr.cc index 85eb31847..8b00ef44c 100644 --- a/test/syscalls/linux/xattr.cc +++ b/test/syscalls/linux/xattr.cc @@ -24,6 +24,7 @@ #include "gmock/gmock.h" #include "gtest/gtest.h" +#include "absl/container/flat_hash_set.h" #include "test/syscalls/linux/file_base.h" #include "test/util/capability_util.h" #include "test/util/file_descriptor.h" @@ -38,36 +39,36 @@ namespace { class XattrTest : public FileTest {}; -TEST_F(XattrTest, XattrNullName) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); +TEST_F(XattrTest, XattrNonexistentFile) { + const char* path = "/does/not/exist"; + EXPECT_THAT(setxattr(path, nullptr, nullptr, 0, /*flags=*/0), + SyscallFailsWithErrno(ENOENT)); + EXPECT_THAT(getxattr(path, nullptr, nullptr, 0), + SyscallFailsWithErrno(ENOENT)); + EXPECT_THAT(listxattr(path, nullptr, 0), SyscallFailsWithErrno(ENOENT)); + EXPECT_THAT(removexattr(path, nullptr), SyscallFailsWithErrno(ENOENT)); +} +TEST_F(XattrTest, XattrNullName) { const char* path = test_file_name_.c_str(); EXPECT_THAT(setxattr(path, nullptr, nullptr, 0, /*flags=*/0), SyscallFailsWithErrno(EFAULT)); EXPECT_THAT(getxattr(path, nullptr, nullptr, 0), SyscallFailsWithErrno(EFAULT)); + EXPECT_THAT(removexattr(path, nullptr), SyscallFailsWithErrno(EFAULT)); } TEST_F(XattrTest, XattrEmptyName) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); EXPECT_THAT(setxattr(path, "", nullptr, 0, /*flags=*/0), SyscallFailsWithErrno(ERANGE)); EXPECT_THAT(getxattr(path, "", nullptr, 0), SyscallFailsWithErrno(ERANGE)); + EXPECT_THAT(removexattr(path, ""), SyscallFailsWithErrno(ERANGE)); } TEST_F(XattrTest, XattrLargeName) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); std::string name = "user."; name += std::string(XATTR_NAME_MAX - name.length(), 'a'); @@ -86,28 +87,23 @@ TEST_F(XattrTest, XattrLargeName) { SyscallFailsWithErrno(ERANGE)); EXPECT_THAT(getxattr(path, name.c_str(), nullptr, 0), SyscallFailsWithErrno(ERANGE)); + EXPECT_THAT(removexattr(path, name.c_str()), SyscallFailsWithErrno(ERANGE)); } TEST_F(XattrTest, XattrInvalidPrefix) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); std::string name(XATTR_NAME_MAX, 'a'); EXPECT_THAT(setxattr(path, name.c_str(), nullptr, 0, /*flags=*/0), SyscallFailsWithErrno(EOPNOTSUPP)); EXPECT_THAT(getxattr(path, name.c_str(), nullptr, 0), SyscallFailsWithErrno(EOPNOTSUPP)); + EXPECT_THAT(removexattr(path, name.c_str()), + SyscallFailsWithErrno(EOPNOTSUPP)); } // Do not allow save/restore cycles after making the test file read-only, as // the restore will fail to open it with r/w permissions. TEST_F(XattrTest, XattrReadOnly_NoRandomSave) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - // Drop capabilities that allow us to override file and directory permissions. ASSERT_NO_ERRNO(SetCapability(CAP_DAC_OVERRIDE, false)); ASSERT_NO_ERRNO(SetCapability(CAP_DAC_READ_SEARCH, false)); @@ -124,19 +120,21 @@ TEST_F(XattrTest, XattrReadOnly_NoRandomSave) { EXPECT_THAT(setxattr(path, name, &val, size, /*flags=*/0), SyscallFailsWithErrno(EACCES)); + EXPECT_THAT(removexattr(path, name), SyscallFailsWithErrno(EACCES)); char buf = '-'; EXPECT_THAT(getxattr(path, name, &buf, size), SyscallSucceedsWithValue(size)); EXPECT_EQ(buf, val); + + char list[sizeof(name)]; + EXPECT_THAT(listxattr(path, list, sizeof(list)), + SyscallSucceedsWithValue(sizeof(name))); + EXPECT_STREQ(list, name); } // Do not allow save/restore cycles after making the test file write-only, as // the restore will fail to open it with r/w permissions. TEST_F(XattrTest, XattrWriteOnly_NoRandomSave) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - // Drop capabilities that allow us to override file and directory permissions. ASSERT_NO_ERRNO(SetCapability(CAP_DAC_OVERRIDE, false)); ASSERT_NO_ERRNO(SetCapability(CAP_DAC_READ_SEARCH, false)); @@ -152,6 +150,14 @@ TEST_F(XattrTest, XattrWriteOnly_NoRandomSave) { EXPECT_THAT(setxattr(path, name, &val, size, /*flags=*/0), SyscallSucceeds()); EXPECT_THAT(getxattr(path, name, nullptr, 0), SyscallFailsWithErrno(EACCES)); + + // listxattr will succeed even without read permissions. + char list[sizeof(name)]; + EXPECT_THAT(listxattr(path, list, sizeof(list)), + SyscallSucceedsWithValue(sizeof(name))); + EXPECT_STREQ(list, name); + + EXPECT_THAT(removexattr(path, name), SyscallSucceeds()); } TEST_F(XattrTest, XattrTrustedWithNonadmin) { @@ -163,64 +169,66 @@ TEST_F(XattrTest, XattrTrustedWithNonadmin) { const char name[] = "trusted.abc"; EXPECT_THAT(setxattr(path, name, nullptr, 0, /*flags=*/0), SyscallFailsWithErrno(EPERM)); + EXPECT_THAT(removexattr(path, name), SyscallFailsWithErrno(EPERM)); EXPECT_THAT(getxattr(path, name, nullptr, 0), SyscallFailsWithErrno(ENODATA)); } TEST_F(XattrTest, XattrOnDirectory) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); const char name[] = "user.test"; - EXPECT_THAT(setxattr(dir.path().c_str(), name, NULL, 0, /*flags=*/0), + EXPECT_THAT(setxattr(dir.path().c_str(), name, nullptr, 0, /*flags=*/0), SyscallSucceeds()); - EXPECT_THAT(getxattr(dir.path().c_str(), name, NULL, 0), + EXPECT_THAT(getxattr(dir.path().c_str(), name, nullptr, 0), SyscallSucceedsWithValue(0)); + + char list[sizeof(name)]; + EXPECT_THAT(listxattr(dir.path().c_str(), list, sizeof(list)), + SyscallSucceedsWithValue(sizeof(name))); + EXPECT_STREQ(list, name); + + EXPECT_THAT(removexattr(dir.path().c_str(), name), SyscallSucceeds()); } TEST_F(XattrTest, XattrOnSymlink) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); TempPath link = ASSERT_NO_ERRNO_AND_VALUE( TempPath::CreateSymlinkTo(dir.path(), test_file_name_)); const char name[] = "user.test"; - EXPECT_THAT(setxattr(link.path().c_str(), name, NULL, 0, /*flags=*/0), + EXPECT_THAT(setxattr(link.path().c_str(), name, nullptr, 0, /*flags=*/0), SyscallSucceeds()); - EXPECT_THAT(getxattr(link.path().c_str(), name, NULL, 0), + EXPECT_THAT(getxattr(link.path().c_str(), name, nullptr, 0), SyscallSucceedsWithValue(0)); + + char list[sizeof(name)]; + EXPECT_THAT(listxattr(link.path().c_str(), list, sizeof(list)), + SyscallSucceedsWithValue(sizeof(name))); + EXPECT_STREQ(list, name); + + EXPECT_THAT(removexattr(link.path().c_str(), name), SyscallSucceeds()); } TEST_F(XattrTest, XattrOnInvalidFileTypes) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - const char name[] = "user.test"; char char_device[] = "/dev/zero"; - EXPECT_THAT(setxattr(char_device, name, NULL, 0, /*flags=*/0), + EXPECT_THAT(setxattr(char_device, name, nullptr, 0, /*flags=*/0), SyscallFailsWithErrno(EPERM)); - EXPECT_THAT(getxattr(char_device, name, NULL, 0), + EXPECT_THAT(getxattr(char_device, name, nullptr, 0), SyscallFailsWithErrno(ENODATA)); + EXPECT_THAT(listxattr(char_device, nullptr, 0), SyscallSucceedsWithValue(0)); // Use tmpfs, where creation of named pipes is supported. const std::string fifo = NewTempAbsPathInDir("/dev/shm"); const char* path = fifo.c_str(); EXPECT_THAT(mknod(path, S_IFIFO | S_IRUSR | S_IWUSR, 0), SyscallSucceeds()); - EXPECT_THAT(setxattr(path, name, NULL, 0, /*flags=*/0), + EXPECT_THAT(setxattr(path, name, nullptr, 0, /*flags=*/0), SyscallFailsWithErrno(EPERM)); - EXPECT_THAT(getxattr(path, name, NULL, 0), SyscallFailsWithErrno(ENODATA)); + EXPECT_THAT(getxattr(path, name, nullptr, 0), SyscallFailsWithErrno(ENODATA)); + EXPECT_THAT(listxattr(path, nullptr, 0), SyscallSucceedsWithValue(0)); + EXPECT_THAT(removexattr(path, name), SyscallFailsWithErrno(EPERM)); } TEST_F(XattrTest, SetxattrSizeSmallerThanValue) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); const char name[] = "user.test"; std::vector val = {'a', 'a'}; @@ -236,10 +244,6 @@ TEST_F(XattrTest, SetxattrSizeSmallerThanValue) { } TEST_F(XattrTest, SetxattrZeroSize) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); const char name[] = "user.test"; char val = 'a'; @@ -252,10 +256,6 @@ TEST_F(XattrTest, SetxattrZeroSize) { } TEST_F(XattrTest, SetxattrSizeTooLarge) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); const char name[] = "user.test"; @@ -271,10 +271,6 @@ TEST_F(XattrTest, SetxattrSizeTooLarge) { } TEST_F(XattrTest, SetxattrNullValueAndNonzeroSize) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); const char name[] = "user.test"; EXPECT_THAT(setxattr(path, name, nullptr, 1, /*flags=*/0), @@ -284,10 +280,6 @@ TEST_F(XattrTest, SetxattrNullValueAndNonzeroSize) { } TEST_F(XattrTest, SetxattrNullValueAndZeroSize) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); const char name[] = "user.test"; EXPECT_THAT(setxattr(path, name, nullptr, 0, /*flags=*/0), SyscallSucceeds()); @@ -296,10 +288,6 @@ TEST_F(XattrTest, SetxattrNullValueAndZeroSize) { } TEST_F(XattrTest, SetxattrValueTooLargeButOKSize) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); const char name[] = "user.test"; std::vector val(XATTR_SIZE_MAX + 1); @@ -316,10 +304,6 @@ TEST_F(XattrTest, SetxattrValueTooLargeButOKSize) { } TEST_F(XattrTest, SetxattrReplaceWithSmaller) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); const char name[] = "user.test"; std::vector val = {'a', 'a'}; @@ -335,10 +319,6 @@ TEST_F(XattrTest, SetxattrReplaceWithSmaller) { } TEST_F(XattrTest, SetxattrReplaceWithLarger) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); const char name[] = "user.test"; std::vector val = {'a', 'a'}; @@ -353,10 +333,6 @@ TEST_F(XattrTest, SetxattrReplaceWithLarger) { } TEST_F(XattrTest, SetxattrCreateFlag) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); const char name[] = "user.test"; EXPECT_THAT(setxattr(path, name, nullptr, 0, XATTR_CREATE), @@ -368,10 +344,6 @@ TEST_F(XattrTest, SetxattrCreateFlag) { } TEST_F(XattrTest, SetxattrReplaceFlag) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); const char name[] = "user.test"; EXPECT_THAT(setxattr(path, name, nullptr, 0, XATTR_REPLACE), @@ -384,10 +356,6 @@ TEST_F(XattrTest, SetxattrReplaceFlag) { } TEST_F(XattrTest, SetxattrInvalidFlags) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); int invalid_flags = 0xff; EXPECT_THAT(setxattr(path, nullptr, nullptr, 0, invalid_flags), @@ -395,10 +363,6 @@ TEST_F(XattrTest, SetxattrInvalidFlags) { } TEST_F(XattrTest, Getxattr) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); const char name[] = "user.test"; int val = 1234; @@ -411,10 +375,6 @@ TEST_F(XattrTest, Getxattr) { } TEST_F(XattrTest, GetxattrSizeSmallerThanValue) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); const char name[] = "user.test"; std::vector val = {'a', 'a'}; @@ -427,10 +387,6 @@ TEST_F(XattrTest, GetxattrSizeSmallerThanValue) { } TEST_F(XattrTest, GetxattrSizeLargerThanValue) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); const char name[] = "user.test"; char val = 'a'; @@ -446,10 +402,6 @@ TEST_F(XattrTest, GetxattrSizeLargerThanValue) { } TEST_F(XattrTest, GetxattrZeroSize) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); const char name[] = "user.test"; char val = 'a'; @@ -463,10 +415,6 @@ TEST_F(XattrTest, GetxattrZeroSize) { } TEST_F(XattrTest, GetxattrSizeTooLarge) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); const char name[] = "user.test"; char val = 'a'; @@ -483,10 +431,6 @@ TEST_F(XattrTest, GetxattrSizeTooLarge) { } TEST_F(XattrTest, GetxattrNullValue) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); const char name[] = "user.test"; char val = 'a'; @@ -498,10 +442,6 @@ TEST_F(XattrTest, GetxattrNullValue) { } TEST_F(XattrTest, GetxattrNullValueAndZeroSize) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - const char* path = test_file_name_.c_str(); const char name[] = "user.test"; char val = 'a'; @@ -518,35 +458,109 @@ TEST_F(XattrTest, GetxattrNullValueAndZeroSize) { } TEST_F(XattrTest, GetxattrNonexistentName) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); + const char* path = test_file_name_.c_str(); + const char name[] = "user.test"; + EXPECT_THAT(getxattr(path, name, nullptr, 0), SyscallFailsWithErrno(ENODATA)); +} + +TEST_F(XattrTest, Listxattr) { + const char* path = test_file_name_.c_str(); + const std::string name = "user.test"; + const std::string name2 = "user.test2"; + const std::string name3 = "user.test3"; + EXPECT_THAT(setxattr(path, name.c_str(), nullptr, 0, /*flags=*/0), + SyscallSucceeds()); + EXPECT_THAT(setxattr(path, name2.c_str(), nullptr, 0, /*flags=*/0), + SyscallSucceeds()); + EXPECT_THAT(setxattr(path, name3.c_str(), nullptr, 0, /*flags=*/0), + SyscallSucceeds()); + std::vector list(name.size() + 1 + name2.size() + 1 + name3.size() + 1); + char* buf = list.data(); + EXPECT_THAT(listxattr(path, buf, XATTR_SIZE_MAX), + SyscallSucceedsWithValue(list.size())); + + absl::flat_hash_set got = {}; + for (char* p = buf; p < buf + list.size(); p += strlen(p) + 1) { + got.insert(std::string{p}); + } + + absl::flat_hash_set expected = {name, name2, name3}; + EXPECT_EQ(got, expected); +} + +TEST_F(XattrTest, ListxattrNoXattrs) { + const char* path = test_file_name_.c_str(); + + std::vector list, expected; + EXPECT_THAT(listxattr(path, list.data(), sizeof(list)), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(list, expected); + + // Listxattr should succeed if there are no attributes, even if the buffer + // passed in is a nullptr. + EXPECT_THAT(listxattr(path, nullptr, sizeof(list)), + SyscallSucceedsWithValue(0)); +} + +TEST_F(XattrTest, ListxattrNullBuffer) { const char* path = test_file_name_.c_str(); const char name[] = "user.test"; + EXPECT_THAT(setxattr(path, name, nullptr, 0, /*flags=*/0), SyscallSucceeds()); + + EXPECT_THAT(listxattr(path, nullptr, sizeof(name)), + SyscallFailsWithErrno(EFAULT)); +} + +TEST_F(XattrTest, ListxattrSizeTooSmall) { + const char* path = test_file_name_.c_str(); + const char name[] = "user.test"; + EXPECT_THAT(setxattr(path, name, nullptr, 0, /*flags=*/0), SyscallSucceeds()); + + char list[sizeof(name) - 1]; + EXPECT_THAT(listxattr(path, list, sizeof(list)), + SyscallFailsWithErrno(ERANGE)); +} + +TEST_F(XattrTest, ListxattrZeroSize) { + const char* path = test_file_name_.c_str(); + const char name[] = "user.test"; + EXPECT_THAT(setxattr(path, name, nullptr, 0, /*flags=*/0), SyscallSucceeds()); + EXPECT_THAT(listxattr(path, nullptr, 0), + SyscallSucceedsWithValue(sizeof(name))); +} + +TEST_F(XattrTest, RemoveXattr) { + const char* path = test_file_name_.c_str(); + const char name[] = "user.test"; + EXPECT_THAT(setxattr(path, name, nullptr, 0, /*flags=*/0), SyscallSucceeds()); + EXPECT_THAT(removexattr(path, name), SyscallSucceeds()); EXPECT_THAT(getxattr(path, name, nullptr, 0), SyscallFailsWithErrno(ENODATA)); } -TEST_F(XattrTest, LGetSetxattrOnSymlink) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); +TEST_F(XattrTest, RemoveXattrNonexistentName) { + const char* path = test_file_name_.c_str(); + const char name[] = "user.test"; + EXPECT_THAT(removexattr(path, name), SyscallFailsWithErrno(ENODATA)); +} +TEST_F(XattrTest, LXattrOnSymlink) { + const char name[] = "user.test"; TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); TempPath link = ASSERT_NO_ERRNO_AND_VALUE( TempPath::CreateSymlinkTo(dir.path(), test_file_name_)); - EXPECT_THAT(lsetxattr(link.path().c_str(), nullptr, nullptr, 0, 0), + EXPECT_THAT(lsetxattr(link.path().c_str(), name, nullptr, 0, 0), SyscallFailsWithErrno(EPERM)); - EXPECT_THAT(lgetxattr(link.path().c_str(), nullptr, nullptr, 0), + EXPECT_THAT(lgetxattr(link.path().c_str(), name, nullptr, 0), SyscallFailsWithErrno(ENODATA)); + EXPECT_THAT(llistxattr(link.path().c_str(), nullptr, 0), + SyscallSucceedsWithValue(0)); + EXPECT_THAT(lremovexattr(link.path().c_str(), name), + SyscallFailsWithErrno(EPERM)); } -TEST_F(XattrTest, LGetSetxattrOnNonsymlink) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); - +TEST_F(XattrTest, LXattrOnNonsymlink) { const char* path = test_file_name_.c_str(); const char name[] = "user.test"; int val = 1234; @@ -558,13 +572,16 @@ TEST_F(XattrTest, LGetSetxattrOnNonsymlink) { EXPECT_THAT(lgetxattr(path, name, &buf, size), SyscallSucceedsWithValue(size)); EXPECT_EQ(buf, val); -} -TEST_F(XattrTest, FGetSetxattr) { - // TODO(gvisor.dev/issue/1636): Re-enable once list/remove xattr are - // supported, and get/set have been added pack to the syscall table. - SKIP_IF(IsRunningOnGvisor()); + char list[sizeof(name)]; + EXPECT_THAT(llistxattr(path, list, sizeof(list)), + SyscallSucceedsWithValue(sizeof(name))); + EXPECT_STREQ(list, name); + EXPECT_THAT(lremovexattr(path, name), SyscallSucceeds()); +} + +TEST_F(XattrTest, XattrWithFD) { const FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(test_file_name_.c_str(), 0)); const char name[] = "user.test"; @@ -577,6 +594,13 @@ TEST_F(XattrTest, FGetSetxattr) { EXPECT_THAT(fgetxattr(fd.get(), name, &buf, size), SyscallSucceedsWithValue(size)); EXPECT_EQ(buf, val); + + char list[sizeof(name)]; + EXPECT_THAT(flistxattr(fd.get(), list, sizeof(list)), + SyscallSucceedsWithValue(sizeof(name))); + EXPECT_STREQ(list, name); + + EXPECT_THAT(fremovexattr(fd.get(), name), SyscallSucceeds()); } } // namespace -- cgit v1.2.3 From 20840bfec087d45853e81d1ac34940f3b2fb920a Mon Sep 17 00:00:00 2001 From: Brad Burlage Date: Mon, 10 Feb 2020 11:57:31 -0800 Subject: Move x86 state definition to its own file. PiperOrigin-RevId: 294271541 --- pkg/sentry/arch/BUILD | 1 + pkg/sentry/arch/arch_state_x86.go | 4 ++-- pkg/sentry/arch/arch_x86.go | 15 -------------- pkg/sentry/arch/arch_x86_impl.go | 43 +++++++++++++++++++++++++++++++++++++++ tools/build/tags.bzl | 24 +++++++++++----------- 5 files changed, 58 insertions(+), 29 deletions(-) create mode 100644 pkg/sentry/arch/arch_x86_impl.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/arch/BUILD b/pkg/sentry/arch/BUILD index 34c0a867d..e27f21e5e 100644 --- a/pkg/sentry/arch/BUILD +++ b/pkg/sentry/arch/BUILD @@ -14,6 +14,7 @@ go_library( "arch_state_aarch64.go", "arch_state_x86.go", "arch_x86.go", + "arch_x86_impl.go", "auxv.go", "signal.go", "signal_act.go", diff --git a/pkg/sentry/arch/arch_state_x86.go b/pkg/sentry/arch/arch_state_x86.go index d388ee9cf..e35c9214a 100644 --- a/pkg/sentry/arch/arch_state_x86.go +++ b/pkg/sentry/arch/arch_state_x86.go @@ -43,8 +43,8 @@ func (e ErrFloatingPoint) Error() string { // and SSE state, so this is the equivalent XSTATE_BV value. const fxsaveBV uint64 = cpuid.XSAVEFeatureX87 | cpuid.XSAVEFeatureSSE -// afterLoad is invoked by stateify. -func (s *State) afterLoad() { +// afterLoadFPState is invoked by afterLoad. +func (s *State) afterLoadFPState() { old := s.x86FPState // Recreate the slice. This is done to ensure that it is aligned diff --git a/pkg/sentry/arch/arch_x86.go b/pkg/sentry/arch/arch_x86.go index 3db8bd34b..88b40a9d1 100644 --- a/pkg/sentry/arch/arch_x86.go +++ b/pkg/sentry/arch/arch_x86.go @@ -155,21 +155,6 @@ func NewFloatingPointData() *FloatingPointData { return (*FloatingPointData)(&(newX86FPState()[0])) } -// State contains the common architecture bits for X86 (the build tag of this -// file ensures it's only built on x86). -// -// +stateify savable -type State struct { - // The system registers. - Regs syscall.PtraceRegs `state:".(syscallPtraceRegs)"` - - // Our floating point state. - x86FPState `state:"wait"` - - // FeatureSet is a pointer to the currently active feature set. - FeatureSet *cpuid.FeatureSet -} - // Proto returns a protobuf representation of the system registers in State. func (s State) Proto() *rpb.Registers { regs := &rpb.AMD64Registers{ diff --git a/pkg/sentry/arch/arch_x86_impl.go b/pkg/sentry/arch/arch_x86_impl.go new file mode 100644 index 000000000..04ac283c6 --- /dev/null +++ b/pkg/sentry/arch/arch_x86_impl.go @@ -0,0 +1,43 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build amd64 i386 + +package arch + +import ( + "syscall" + + "gvisor.dev/gvisor/pkg/cpuid" +) + +// State contains the common architecture bits for X86 (the build tag of this +// file ensures it's only built on x86). +// +// +stateify savable +type State struct { + // The system registers. + Regs syscall.PtraceRegs `state:".(syscallPtraceRegs)"` + + // Our floating point state. + x86FPState `state:"wait"` + + // FeatureSet is a pointer to the currently active feature set. + FeatureSet *cpuid.FeatureSet +} + +// afterLoad is invoked by stateify. +func (s *State) afterLoad() { + s.afterLoadFPState() +} diff --git a/tools/build/tags.bzl b/tools/build/tags.bzl index a6db44e47..558fb53ae 100644 --- a/tools/build/tags.bzl +++ b/tools/build/tags.bzl @@ -3,22 +3,28 @@ go_suffixes = [ "_386", "_386_unsafe", - "_amd64", - "_amd64_unsafe", "_aarch64", "_aarch64_unsafe", + "_amd64", + "_amd64_unsafe", "_arm", - "_arm_unsafe", "_arm64", "_arm64_unsafe", + "_arm_unsafe", + "_impl", + "_impl_unsafe", + "_linux", + "_linux_unsafe", "_mips", - "_mips_unsafe", - "_mipsle", - "_mipsle_unsafe", "_mips64", "_mips64_unsafe", "_mips64le", "_mips64le_unsafe", + "_mips_unsafe", + "_mipsle", + "_mipsle_unsafe", + "_opts", + "_opts_unsafe", "_ppc64", "_ppc64_unsafe", "_ppc64le", @@ -31,10 +37,4 @@ go_suffixes = [ "_sparc64_unsafe", "_wasm", "_wasm_unsafe", - "_linux", - "_linux_unsafe", - "_opts", - "_opts_unsafe", - "_impl", - "_impl_unsafe", ] -- cgit v1.2.3 From bfa0bba72abb69cbc7f4da27d3b4b116c3784495 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Mon, 10 Feb 2020 12:02:47 -0800 Subject: Redirect FIXME to gvisor.dev PiperOrigin-RevId: 294272755 --- pkg/sentry/fsimpl/kernfs/filesystem.go | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 9d65d0179..e49303c26 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -111,10 +111,10 @@ func (fs *Filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.Vir // Dentry isn't cached; it either doesn't exist or failed // revalidation. Attempt to resolve it via Lookup. // - // FIXME(b/144498111): Inode.Lookup() should return *(kernfs.)Dentry, - // not *vfs.Dentry, since (kernfs.)Filesystem assumes that all dentries - // in the filesystem are (kernfs.)Dentry and performs vfs.DentryImpl - // casts accordingly. + // FIXME(gvisor.dev/issue/1193): Inode.Lookup() should return + // *(kernfs.)Dentry, not *vfs.Dentry, since (kernfs.)Filesystem assumes + // that all dentries in the filesystem are (kernfs.)Dentry and performs + // vfs.DentryImpl casts accordingly. var err error childVFSD, err = parent.inode.Lookup(ctx, name) if err != nil { -- cgit v1.2.3 From c9a18b16ade6ec0bc90fc75d0a4ab0621f9d01d6 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Mon, 10 Feb 2020 12:06:06 -0800 Subject: Document MinimumTotalMemoryBytes. PiperOrigin-RevId: 294273559 --- pkg/sentry/usage/memory.go | 4 ++++ 1 file changed, 4 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/usage/memory.go b/pkg/sentry/usage/memory.go index 538c645eb..4320ad17f 100644 --- a/pkg/sentry/usage/memory.go +++ b/pkg/sentry/usage/memory.go @@ -253,6 +253,10 @@ func (m *MemoryLocked) Copy() (MemoryStats, uint64) { } // MinimumTotalMemoryBytes is the minimum reported total system memory. +// +// This can be configured through options provided to the Sentry at start. +// This number is purely synthetic. This is only set before the application +// starts executing, and must not be modified. var MinimumTotalMemoryBytes uint64 = 2 << 30 // 2 GB // TotalMemory returns the "total usable memory" available. -- cgit v1.2.3 From 4d4d47f0c0a21d3404d2edae527b187d62daa3c8 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Mon, 10 Feb 2020 13:04:29 -0800 Subject: Add contextual note. PiperOrigin-RevId: 294285723 --- pkg/sentry/fsimpl/gofer/directory.go | 4 ++++ 1 file changed, 4 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/directory.go b/pkg/sentry/fsimpl/gofer/directory.go index baa2cdd8e..6d4ebc2bf 100644 --- a/pkg/sentry/fsimpl/gofer/directory.go +++ b/pkg/sentry/fsimpl/gofer/directory.go @@ -87,6 +87,10 @@ func (d *dentry) getDirents(ctx context.Context) ([]vfs.Dirent, error) { // to assume that directory fids have the correct semantics, and translates // struct file_operations::readdir calls directly to readdir RPCs), but is // consistent with VFS1. + // + // NOTE(b/135560623): In particular, some gofer implementations may not + // retain state between calls to Readdir, so may not provide a coherent + // directory stream across in the presence of mutation. d.fs.renameMu.RLock() defer d.fs.renameMu.RUnlock() -- cgit v1.2.3 From bb22ebd7fbfc66556b38df669be5c6372daba018 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Mon, 10 Feb 2020 13:20:44 -0800 Subject: Add contextual comment. PiperOrigin-RevId: 294289066 --- pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go | 3 +++ 1 file changed, 3 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go index d1436b943..2015a8871 100644 --- a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go +++ b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go @@ -15,6 +15,9 @@ // These benchmarks emulate memfs benchmarks. Ext4 images must be created // before this benchmark is run using the `make_deep_ext4.sh` script at // /tmp/image-{depth}.ext4 for all the depths tested below. +// +// The benchmark itself cannot run the script because the script requires +// sudo privileges to create the file system images. package benchmark_test import ( -- cgit v1.2.3 From a6f9361c2f7c5b46a200de1dc891a0ce059ad90e Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Mon, 10 Feb 2020 13:51:26 -0800 Subject: Add context to comments. PiperOrigin-RevId: 294295852 --- pkg/sentry/fs/gofer/attr.go | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/gofer/attr.go b/pkg/sentry/fs/gofer/attr.go index 71cccdc34..6db4b762d 100644 --- a/pkg/sentry/fs/gofer/attr.go +++ b/pkg/sentry/fs/gofer/attr.go @@ -88,8 +88,9 @@ func bsize(pattr p9.Attr) int64 { if pattr.BlockSize > 0 { return int64(pattr.BlockSize) } - // Some files may have no clue of their block size. Better not to report - // something misleading or buggy and have a safe default. + // Some files, particularly those that are not on a local file system, + // may have no clue of their block size. Better not to report something + // misleading or buggy and have a safe default. return usermem.PageSize } @@ -149,6 +150,7 @@ func links(valid p9.AttrMask, pattr p9.Attr) uint64 { } // This node is likely backed by a file system that doesn't support links. + // // We could readdir() and count children directories to provide an accurate // link count. However this may be expensive since the gofer may be backed by remote // storage. Instead, simply return 2 links for directories and 1 for everything else -- cgit v1.2.3 From 2889ffa84ec4737b651b14c3ce019c5005d0dd9c Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Mon, 10 Feb 2020 14:09:14 -0800 Subject: Add context to note. PiperOrigin-RevId: 294300040 --- pkg/sentry/arch/arch_amd64.s | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/arch/arch_amd64.s b/pkg/sentry/arch/arch_amd64.s index bd61402cf..6c10336e7 100644 --- a/pkg/sentry/arch/arch_amd64.s +++ b/pkg/sentry/arch/arch_amd64.s @@ -26,10 +26,11 @@ // // func initX86FPState(data *FloatingPointData, useXsave bool) // -// We need to clear out and initialize an empty fp state area since the sentry -// may have left sensitive information in the floating point registers. +// We need to clear out and initialize an empty fp state area since the sentry, +// or any previous loader, may have left sensitive information in the floating +// point registers. // -// Preconditions: data is zeroed +// Preconditions: data is zeroed. TEXT ·initX86FPState(SB), $24-16 // Save MXCSR (callee-save) STMXCSR mxcsr-8(SP) -- cgit v1.2.3 From 475316e87dac806d69bcb06ea4065f3c138bb47e Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Mon, 10 Feb 2020 14:47:00 -0800 Subject: Refactor getxattr. Put most of the logic for getxattr in one place for clarity. This simplifies FGetXattr and getXattrFromPath, which are just wrappers for getXattr. PiperOrigin-RevId: 294308332 --- pkg/sentry/syscalls/linux/sys_xattr.go | 44 ++++++++++++++++------------------ 1 file changed, 20 insertions(+), 24 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/sys_xattr.go b/pkg/sentry/syscalls/linux/sys_xattr.go index 342337726..9d8140b8a 100644 --- a/pkg/sentry/syscalls/linux/sys_xattr.go +++ b/pkg/sentry/syscalls/linux/sys_xattr.go @@ -49,14 +49,11 @@ func FGetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys } defer f.DecRef() - n, value, err := getXattr(t, f.Dirent, nameAddr, size) + n, err := getXattr(t, f.Dirent, nameAddr, valueAddr, size) if err != nil { return 0, nil, err } - if _, err := t.CopyOutBytes(valueAddr, []byte(value)); err != nil { - return 0, nil, err - } return uintptr(n), nil, nil } @@ -71,41 +68,36 @@ func getXattrFromPath(t *kernel.Task, args arch.SyscallArguments, resolveSymlink return 0, nil, err } - valueLen := 0 + n := 0 err = fileOpOn(t, linux.AT_FDCWD, path, resolveSymlink, func(_ *fs.Dirent, d *fs.Dirent, _ uint) error { if dirPath && !fs.IsDir(d.Inode.StableAttr) { return syserror.ENOTDIR } - n, value, err := getXattr(t, d, nameAddr, size) - valueLen = n - if err != nil { - return err - } - - _, err = t.CopyOutBytes(valueAddr, []byte(value)) + n, err = getXattr(t, d, nameAddr, valueAddr, size) return err }) if err != nil { return 0, nil, err } - return uintptr(valueLen), nil, nil + + return uintptr(n), nil, nil } // getXattr implements getxattr(2) from the given *fs.Dirent. -func getXattr(t *kernel.Task, d *fs.Dirent, nameAddr usermem.Addr, size uint64) (int, string, error) { - if err := checkXattrPermissions(t, d.Inode, fs.PermMask{Read: true}); err != nil { - return 0, "", err - } - +func getXattr(t *kernel.Task, d *fs.Dirent, nameAddr, valueAddr usermem.Addr, size uint64) (int, error) { name, err := copyInXattrName(t, nameAddr) if err != nil { - return 0, "", err + return 0, err + } + + if err := checkXattrPermissions(t, d.Inode, fs.PermMask{Read: true}); err != nil { + return 0, err } // TODO(b/148380782): Support xattrs in namespaces other than "user". if !strings.HasPrefix(name, linux.XATTR_USER_PREFIX) { - return 0, "", syserror.EOPNOTSUPP + return 0, syserror.EOPNOTSUPP } // If getxattr(2) is called with size 0, the size of the value will be @@ -118,18 +110,22 @@ func getXattr(t *kernel.Task, d *fs.Dirent, nameAddr usermem.Addr, size uint64) value, err := d.Inode.GetXattr(t, name, requestedSize) if err != nil { - return 0, "", err + return 0, err } n := len(value) if uint64(n) > requestedSize { - return 0, "", syserror.ERANGE + return 0, syserror.ERANGE } // Don't copy out the attribute value if size is 0. if size == 0 { - return n, "", nil + return n, nil + } + + if _, err = t.CopyOutBytes(valueAddr, []byte(value)); err != nil { + return 0, err } - return n, value, nil + return n, nil } // SetXattr implements linux syscall setxattr(2). -- cgit v1.2.3 From 71af006b6fe4504fccb86f0222a8a1864d33fb7d Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Mon, 10 Feb 2020 17:12:03 -0800 Subject: Cleanup internal package group. PiperOrigin-RevId: 294339229 --- pkg/sentry/BUILD | 8 +++----- 1 file changed, 3 insertions(+), 5 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/BUILD b/pkg/sentry/BUILD index e8b794179..e759dc36f 100644 --- a/pkg/sentry/BUILD +++ b/pkg/sentry/BUILD @@ -1,13 +1,11 @@ -# This BUILD file defines a package_group that allows for interdependencies for -# sentry-internal packages. - package(licenses = ["notice"]) +# The "internal" package_group should be used as much as possible by packages +# that should remain Sentry-internal (i.e. not be exposed directly to command +# line tooling or APIs). package_group( name = "internal", packages = [ - "//cloud/gvisor/gopkg/sentry/...", - "//cloud/gvisor/sentry/...", "//pkg/sentry/...", "//runsc/...", # Code generated by go_marshal relies on go_marshal libraries. -- cgit v1.2.3 From 762e4761cc4edd92108f6836ad1933c7158b8be8 Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Tue, 11 Feb 2020 11:08:28 -0800 Subject: Move Align{Up,Down} into binary package. PiperOrigin-RevId: 294477647 --- pkg/binary/binary.go | 10 ++++++++++ pkg/sentry/socket/control/control.go | 26 ++++++++------------------ pkg/sentry/socket/netfilter/extensions.go | 7 +------ pkg/sentry/socket/netlink/message.go | 15 ++++----------- pkg/sentry/strace/BUILD | 1 - pkg/sentry/strace/socket.go | 7 +++---- 6 files changed, 26 insertions(+), 40 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/binary/binary.go b/pkg/binary/binary.go index 631785f7b..25065aef9 100644 --- a/pkg/binary/binary.go +++ b/pkg/binary/binary.go @@ -254,3 +254,13 @@ func WriteUint64(w io.Writer, order binary.ByteOrder, num uint64) error { _, err := w.Write(buf) return err } + +// AlignUp rounds a length up to an alignment. align must be a power of 2. +func AlignUp(length int, align uint) int { + return (length + int(align) - 1) & ^(int(align) - 1) +} + +// AlignDown rounds a length down to an alignment. align must be a power of 2. +func AlignDown(length int, align uint) int { + return length & ^(int(align) - 1) +} diff --git a/pkg/sentry/socket/control/control.go b/pkg/sentry/socket/control/control.go index 00265f15b..6145a7fc3 100644 --- a/pkg/sentry/socket/control/control.go +++ b/pkg/sentry/socket/control/control.go @@ -189,7 +189,7 @@ func putUint32(buf []byte, n uint32) []byte { // putCmsg writes a control message header and as much data as will fit into // the unused capacity of a buffer. func putCmsg(buf []byte, flags int, msgType uint32, align uint, data []int32) ([]byte, int) { - space := AlignDown(cap(buf)-len(buf), 4) + space := binary.AlignDown(cap(buf)-len(buf), 4) // We can't write to space that doesn't exist, so if we are going to align // the available space, we must align down. @@ -282,19 +282,9 @@ func PackCredentials(t *kernel.Task, creds SCMCredentials, buf []byte, flags int return putCmsg(buf, flags, linux.SCM_CREDENTIALS, align, c) } -// AlignUp rounds a length up to an alignment. align must be a power of 2. -func AlignUp(length int, align uint) int { - return (length + int(align) - 1) & ^(int(align) - 1) -} - -// AlignDown rounds a down to an alignment. align must be a power of 2. -func AlignDown(length int, align uint) int { - return length & ^(int(align) - 1) -} - // alignSlice extends a slice's length (up to the capacity) to align it. func alignSlice(buf []byte, align uint) []byte { - aligned := AlignUp(len(buf), align) + aligned := binary.AlignUp(len(buf), align) if aligned > cap(buf) { // Linux allows unaligned data if there isn't room for alignment. // Since there isn't room for alignment, there isn't room for any @@ -377,7 +367,7 @@ func PackControlMessages(t *kernel.Task, cmsgs socket.ControlMessages, buf []byt // cmsgSpace is equivalent to CMSG_SPACE in Linux. func cmsgSpace(t *kernel.Task, dataLen int) int { - return linux.SizeOfControlMessageHeader + AlignUp(dataLen, t.Arch().Width()) + return linux.SizeOfControlMessageHeader + binary.AlignUp(dataLen, t.Arch().Width()) } // CmsgsSpace returns the number of bytes needed to fit the control messages @@ -437,7 +427,7 @@ func Parse(t *kernel.Task, socketOrEndpoint interface{}, buf []byte) (socket.Con case linux.SOL_SOCKET: switch h.Type { case linux.SCM_RIGHTS: - rightsSize := AlignDown(length, linux.SizeOfControlMessageRight) + rightsSize := binary.AlignDown(length, linux.SizeOfControlMessageRight) numRights := rightsSize / linux.SizeOfControlMessageRight if len(fds)+numRights > linux.SCM_MAX_FD { @@ -448,7 +438,7 @@ func Parse(t *kernel.Task, socketOrEndpoint interface{}, buf []byte) (socket.Con fds = append(fds, int32(usermem.ByteOrder.Uint32(buf[j:j+linux.SizeOfControlMessageRight]))) } - i += AlignUp(length, width) + i += binary.AlignUp(length, width) case linux.SCM_CREDENTIALS: if length < linux.SizeOfControlMessageCredentials { @@ -462,7 +452,7 @@ func Parse(t *kernel.Task, socketOrEndpoint interface{}, buf []byte) (socket.Con return socket.ControlMessages{}, err } cmsgs.Unix.Credentials = scmCreds - i += AlignUp(length, width) + i += binary.AlignUp(length, width) default: // Unknown message type. @@ -476,7 +466,7 @@ func Parse(t *kernel.Task, socketOrEndpoint interface{}, buf []byte) (socket.Con } cmsgs.IP.HasTOS = true binary.Unmarshal(buf[i:i+linux.SizeOfControlMessageTOS], usermem.ByteOrder, &cmsgs.IP.TOS) - i += AlignUp(length, width) + i += binary.AlignUp(length, width) default: return socket.ControlMessages{}, syserror.EINVAL @@ -489,7 +479,7 @@ func Parse(t *kernel.Task, socketOrEndpoint interface{}, buf []byte) (socket.Con } cmsgs.IP.HasTClass = true binary.Unmarshal(buf[i:i+linux.SizeOfControlMessageTClass], usermem.ByteOrder, &cmsgs.IP.TClass) - i += AlignUp(length, width) + i += binary.AlignUp(length, width) default: return socket.ControlMessages{}, syserror.EINVAL diff --git a/pkg/sentry/socket/netfilter/extensions.go b/pkg/sentry/socket/netfilter/extensions.go index 22fd0ebe7..b4b244abf 100644 --- a/pkg/sentry/socket/netfilter/extensions.go +++ b/pkg/sentry/socket/netfilter/extensions.go @@ -72,7 +72,7 @@ func marshalEntryMatch(name string, data []byte) []byte { nflog("marshaling matcher %q", name) // We have to pad this struct size to a multiple of 8 bytes. - size := alignUp(linux.SizeOfXTEntryMatch+len(data), 8) + size := binary.AlignUp(linux.SizeOfXTEntryMatch+len(data), 8) matcher := linux.KernelXTEntryMatch{ XTEntryMatch: linux.XTEntryMatch{ MatchSize: uint16(size), @@ -93,8 +93,3 @@ func unmarshalMatcher(match linux.XTEntryMatch, filter iptables.IPHeaderFilter, } return matchMaker.unmarshal(buf, filter) } - -// alignUp rounds a length up to an alignment. align must be a power of 2. -func alignUp(length int, align uint) int { - return (length + int(align) - 1) & ^(int(align) - 1) -} diff --git a/pkg/sentry/socket/netlink/message.go b/pkg/sentry/socket/netlink/message.go index 4ea252ccb..0899c61d1 100644 --- a/pkg/sentry/socket/netlink/message.go +++ b/pkg/sentry/socket/netlink/message.go @@ -23,18 +23,11 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) -// alignUp rounds a length up to an alignment. -// -// Preconditions: align is a power of two. -func alignUp(length int, align uint) int { - return (length + int(align) - 1) &^ (int(align) - 1) -} - // alignPad returns the length of padding required for alignment. // // Preconditions: align is a power of two. func alignPad(length int, align uint) int { - return alignUp(length, align) - length + return binary.AlignUp(length, align) - length } // Message contains a complete serialized netlink message. @@ -138,7 +131,7 @@ func (m *Message) Finalize() []byte { // Align the message. Note that the message length in the header (set // above) is the useful length of the message, not the total aligned // length. See net/netlink/af_netlink.c:__nlmsg_put. - aligned := alignUp(len(m.buf), linux.NLMSG_ALIGNTO) + aligned := binary.AlignUp(len(m.buf), linux.NLMSG_ALIGNTO) m.putZeros(aligned - len(m.buf)) return m.buf } @@ -173,7 +166,7 @@ func (m *Message) PutAttr(atype uint16, v interface{}) { m.Put(v) // Align the attribute. - aligned := alignUp(l, linux.NLA_ALIGNTO) + aligned := binary.AlignUp(l, linux.NLA_ALIGNTO) m.putZeros(aligned - l) } @@ -190,7 +183,7 @@ func (m *Message) PutAttrString(atype uint16, s string) { m.putZeros(1) // Align the attribute. - aligned := alignUp(l, linux.NLA_ALIGNTO) + aligned := binary.AlignUp(l, linux.NLA_ALIGNTO) m.putZeros(aligned - l) } diff --git a/pkg/sentry/strace/BUILD b/pkg/sentry/strace/BUILD index 762a946fe..2f39a6f2b 100644 --- a/pkg/sentry/strace/BUILD +++ b/pkg/sentry/strace/BUILD @@ -30,7 +30,6 @@ go_library( "//pkg/seccomp", "//pkg/sentry/arch", "//pkg/sentry/kernel", - "//pkg/sentry/socket/control", "//pkg/sentry/socket/netlink", "//pkg/sentry/socket/netstack", "//pkg/sentry/syscalls/linux", diff --git a/pkg/sentry/strace/socket.go b/pkg/sentry/strace/socket.go index f7ff4573e..51e6d81b2 100644 --- a/pkg/sentry/strace/socket.go +++ b/pkg/sentry/strace/socket.go @@ -22,7 +22,6 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/socket/control" "gvisor.dev/gvisor/pkg/sentry/socket/netlink" "gvisor.dev/gvisor/pkg/sentry/socket/netstack" slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" @@ -220,13 +219,13 @@ func cmsghdr(t *kernel.Task, addr usermem.Addr, length uint64, maxBytes uint64) if skipData { strs = append(strs, fmt.Sprintf("{level=%s, type=%s, length=%d}", level, typ, h.Length)) - i += control.AlignUp(length, width) + i += binary.AlignUp(length, width) continue } switch h.Type { case linux.SCM_RIGHTS: - rightsSize := control.AlignDown(length, linux.SizeOfControlMessageRight) + rightsSize := binary.AlignDown(length, linux.SizeOfControlMessageRight) numRights := rightsSize / linux.SizeOfControlMessageRight fds := make(linux.ControlMessageRights, numRights) @@ -295,7 +294,7 @@ func cmsghdr(t *kernel.Task, addr usermem.Addr, length uint64, maxBytes uint64) default: panic("unreachable") } - i += control.AlignUp(length, width) + i += binary.AlignUp(length, width) } return fmt.Sprintf("%#x %s", addr, strings.Join(strs, ", ")) -- cgit v1.2.3 From 115898e368e4afe5418a7290d9545fafc7f6f25e Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Tue, 11 Feb 2020 11:37:12 -0800 Subject: Prevent DATA RACE in UnstableAttr. The slaveInodeOperations is currently copying the object when truncate is called (which is a no-op). This may result in a (unconsequential) data race when being modified concurrently. PiperOrigin-RevId: 294484276 --- pkg/sentry/fs/tty/slave.go | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/tty/slave.go b/pkg/sentry/fs/tty/slave.go index db55cdc48..6a2dbc576 100644 --- a/pkg/sentry/fs/tty/slave.go +++ b/pkg/sentry/fs/tty/slave.go @@ -73,7 +73,7 @@ func (si *slaveInodeOperations) Release(ctx context.Context) { } // Truncate implements fs.InodeOperations.Truncate. -func (slaveInodeOperations) Truncate(context.Context, *fs.Inode, int64) error { +func (*slaveInodeOperations) Truncate(context.Context, *fs.Inode, int64) error { return nil } -- cgit v1.2.3 From 6dced977eab69401a114257e386addb9cb03a39d Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Tue, 11 Feb 2020 17:38:04 -0800 Subject: Ensure fsimpl/gofer.dentryPlatformFile.hostFileMapper is initialized. Fixes #1812. (The more direct cause of the deadlock is panic unsafety because the historically high cost of defer means that we avoid it in hot paths, including much of MM; defer is much cheaper as of Go 1.14, but still a measurable overhead.) PiperOrigin-RevId: 294560316 --- pkg/sentry/fs/fsutil/host_file_mapper.go | 17 +++++++++++------ pkg/sentry/fsimpl/gofer/regular_file.go | 5 +++++ 2 files changed, 16 insertions(+), 6 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/fsutil/host_file_mapper.go b/pkg/sentry/fs/fsutil/host_file_mapper.go index 67278aa86..e82afd112 100644 --- a/pkg/sentry/fs/fsutil/host_file_mapper.go +++ b/pkg/sentry/fs/fsutil/host_file_mapper.go @@ -65,13 +65,18 @@ type mapping struct { writable bool } -// NewHostFileMapper returns a HostFileMapper with no references or cached -// mappings. +// Init must be called on zero-value HostFileMappers before first use. +func (f *HostFileMapper) Init() { + f.refs = make(map[uint64]int32) + f.mappings = make(map[uint64]mapping) +} + +// NewHostFileMapper returns an initialized HostFileMapper allocated on the +// heap with no references or cached mappings. func NewHostFileMapper() *HostFileMapper { - return &HostFileMapper{ - refs: make(map[uint64]int32), - mappings: make(map[uint64]mapping), - } + f := &HostFileMapper{} + f.Init() + return f } // IncRefOn increments the reference count on all offsets in mr. diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go index 8e11e06b3..54c1031a7 100644 --- a/pkg/sentry/fsimpl/gofer/regular_file.go +++ b/pkg/sentry/fsimpl/gofer/regular_file.go @@ -571,6 +571,8 @@ func (fd *regularFileFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpt default: panic(fmt.Sprintf("unknown InteropMode %v", d.fs.opts.interop)) } + // After this point, d may be used as a memmap.Mappable. + d.pf.hostFileMapperInitOnce.Do(d.pf.hostFileMapper.Init) return vfs.GenericConfigureMMap(&fd.vfsfd, d, opts) } @@ -799,6 +801,9 @@ type dentryPlatformFile struct { // If this dentry represents a regular file, and handle.fd >= 0, // hostFileMapper caches mappings of handle.fd. hostFileMapper fsutil.HostFileMapper + + // hostFileMapperInitOnce is used to lazily initialize hostFileMapper. + hostFileMapperInitOnce sync.Once } // IncRef implements platform.File.IncRef. -- cgit v1.2.3 From 6fdf2c53a1d084b70602170b660242036fd8fe4f Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Fri, 7 Feb 2020 11:21:07 -0800 Subject: iptables: User chains - Adds creation of user chains via `-N ` - Adds `-j RETURN` support for built-in chains, which triggers the chain's underflow rule (usually the default policy). - Adds tests for chain creation, default policies, and `-j RETURN' from built-in chains. --- pkg/sentry/socket/netfilter/netfilter.go | 115 +++++++++++++++++++----------- pkg/tcpip/iptables/iptables.go | 74 ++++++++++++------- pkg/tcpip/iptables/targets.go | 41 ++++++++--- pkg/tcpip/iptables/types.go | 50 +++++-------- test/iptables/filter_input.go | 117 ++++++++++++++++++++++++++++++- test/iptables/iptables_test.go | 24 +++++++ 6 files changed, 310 insertions(+), 111 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index ea02627de..3fc80e0de 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -50,7 +50,9 @@ type metadata struct { // nflog logs messages related to the writing and reading of iptables. func nflog(format string, args ...interface{}) { - log.Infof("netfilter: "+format, args...) + if log.IsLogging(log.Debug) { + log.Debugf("netfilter: "+format, args...) + } } // GetInfo returns information about iptables. @@ -227,19 +229,23 @@ func convertNetstackToBinary(tablename string, table iptables.Table) (linux.Kern } func marshalTarget(target iptables.Target) []byte { - switch target.(type) { - case iptables.UnconditionalAcceptTarget: - return marshalStandardTarget(iptables.Accept) - case iptables.UnconditionalDropTarget: - return marshalStandardTarget(iptables.Drop) + switch tg := target.(type) { + case iptables.AcceptTarget: + return marshalStandardTarget(iptables.RuleAccept) + case iptables.DropTarget: + return marshalStandardTarget(iptables.RuleDrop) case iptables.ErrorTarget: - return marshalErrorTarget() + return marshalErrorTarget(errorTargetName) + case iptables.UserChainTarget: + return marshalErrorTarget(tg.Name) + case iptables.ReturnTarget: + return marshalStandardTarget(iptables.RuleReturn) default: panic(fmt.Errorf("unknown target of type %T", target)) } } -func marshalStandardTarget(verdict iptables.Verdict) []byte { +func marshalStandardTarget(verdict iptables.RuleVerdict) []byte { nflog("convert to binary: marshalling standard target with size %d", linux.SizeOfXTStandardTarget) // The target's name will be the empty string. @@ -254,14 +260,14 @@ func marshalStandardTarget(verdict iptables.Verdict) []byte { return binary.Marshal(ret, usermem.ByteOrder, target) } -func marshalErrorTarget() []byte { +func marshalErrorTarget(errorName string) []byte { // This is an error target named error target := linux.XTErrorTarget{ Target: linux.XTEntryTarget{ TargetSize: linux.SizeOfXTErrorTarget, }, } - copy(target.Name[:], errorTargetName) + copy(target.Name[:], errorName) copy(target.Target.Name[:], errorTargetName) ret := make([]byte, 0, linux.SizeOfXTErrorTarget) @@ -270,38 +276,35 @@ func marshalErrorTarget() []byte { // translateFromStandardVerdict translates verdicts the same way as the iptables // tool. -func translateFromStandardVerdict(verdict iptables.Verdict) int32 { +func translateFromStandardVerdict(verdict iptables.RuleVerdict) int32 { switch verdict { - case iptables.Accept: + case iptables.RuleAccept: return -linux.NF_ACCEPT - 1 - case iptables.Drop: + case iptables.RuleDrop: return -linux.NF_DROP - 1 - case iptables.Queue: - return -linux.NF_QUEUE - 1 - case iptables.Return: + case iptables.RuleReturn: return linux.NF_RETURN - case iptables.Jump: + default: // TODO(gvisor.dev/issue/170): Support Jump. - panic("Jump isn't supported yet") + panic(fmt.Sprintf("unknown standard verdict: %d", verdict)) } - panic(fmt.Sprintf("unknown standard verdict: %d", verdict)) } -// translateToStandardVerdict translates from the value in a +// translateToStandardTarget translates from the value in a // linux.XTStandardTarget to an iptables.Verdict. -func translateToStandardVerdict(val int32) (iptables.Verdict, error) { +func translateToStandardTarget(val int32) (iptables.Target, error) { // TODO(gvisor.dev/issue/170): Support other verdicts. switch val { case -linux.NF_ACCEPT - 1: - return iptables.Accept, nil + return iptables.AcceptTarget{}, nil case -linux.NF_DROP - 1: - return iptables.Drop, nil + return iptables.DropTarget{}, nil case -linux.NF_QUEUE - 1: - return iptables.Invalid, errors.New("unsupported iptables verdict QUEUE") + return nil, errors.New("unsupported iptables verdict QUEUE") case linux.NF_RETURN: - return iptables.Invalid, errors.New("unsupported iptables verdict RETURN") + return iptables.ReturnTarget{}, nil default: - return iptables.Invalid, fmt.Errorf("unknown iptables verdict %d", val) + return nil, fmt.Errorf("unknown iptables verdict %d", val) } } @@ -411,6 +414,10 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { table.BuiltinChains[hk] = ruleIdx } if offset == replace.Underflow[hook] { + if !validUnderflow(table.Rules[ruleIdx]) { + nflog("underflow for hook %d isn't an unconditional ACCEPT or DROP.") + return syserr.ErrInvalidArgument + } table.Underflows[hk] = ruleIdx } } @@ -425,12 +432,34 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { } } + // Add the user chains. + for ruleIdx, rule := range table.Rules { + target, ok := rule.Target.(iptables.UserChainTarget) + if !ok { + continue + } + + // We found a user chain. Before inserting it into the table, + // check that: + // - There's some other rule after it. + // - There are no matchers. + if ruleIdx == len(table.Rules)-1 { + nflog("user chain must have a rule or default policy.") + return syserr.ErrInvalidArgument + } + if len(table.Rules[ruleIdx].Matchers) != 0 { + nflog("user chain's first node must have no matcheres.") + return syserr.ErrInvalidArgument + } + table.UserChains[target.Name] = ruleIdx + 1 + } + // TODO(gvisor.dev/issue/170): Support other chains. // Since we only support modifying the INPUT chain right now, make sure // all other chains point to ACCEPT rules. for hook, ruleIdx := range table.BuiltinChains { if hook != iptables.Input { - if _, ok := table.Rules[ruleIdx].Target.(iptables.UnconditionalAcceptTarget); !ok { + if _, ok := table.Rules[ruleIdx].Target.(iptables.AcceptTarget); !ok { nflog("hook %d is unsupported.", hook) return syserr.ErrInvalidArgument } @@ -519,18 +548,7 @@ func parseTarget(optVal []byte) (iptables.Target, error) { buf = optVal[:linux.SizeOfXTStandardTarget] binary.Unmarshal(buf, usermem.ByteOrder, &standardTarget) - verdict, err := translateToStandardVerdict(standardTarget.Verdict) - if err != nil { - return nil, err - } - switch verdict { - case iptables.Accept: - return iptables.UnconditionalAcceptTarget{}, nil - case iptables.Drop: - return iptables.UnconditionalDropTarget{}, nil - default: - return nil, fmt.Errorf("Unknown verdict: %v", verdict) - } + return translateToStandardTarget(standardTarget.Verdict) case errorTargetName: // Error target. @@ -548,11 +566,14 @@ func parseTarget(optVal []byte) (iptables.Target, error) { // somehow fall through every rule. // * To mark the start of a user defined chain. These // rules have an error with the name of the chain. - switch errorTarget.Name.String() { + switch name := errorTarget.Name.String(); name { case errorTargetName: + nflog("set entries: error target") return iptables.ErrorTarget{}, nil default: - return nil, fmt.Errorf("unknown error target %q doesn't exist or isn't supported yet.", errorTarget.Name.String()) + // User defined chain. + nflog("set entries: user-defined target %q", name) + return iptables.UserChainTarget{Name: name}, nil } } @@ -585,6 +606,18 @@ func containsUnsupportedFields(iptip linux.IPTIP) bool { iptip.InverseFlags != 0 } +func validUnderflow(rule iptables.Rule) bool { + if len(rule.Matchers) != 0 { + return false + } + switch rule.Target.(type) { + case iptables.AcceptTarget, iptables.DropTarget: + return true + default: + return false + } +} + func hookFromLinux(hook int) iptables.Hook { switch hook { case linux.NF_INET_PRE_ROUTING: diff --git a/pkg/tcpip/iptables/iptables.go b/pkg/tcpip/iptables/iptables.go index 1b9485bbd..75a433a3b 100644 --- a/pkg/tcpip/iptables/iptables.go +++ b/pkg/tcpip/iptables/iptables.go @@ -52,10 +52,10 @@ func DefaultTables() IPTables { Tables: map[string]Table{ TablenameNat: Table{ Rules: []Rule{ - Rule{Target: UnconditionalAcceptTarget{}}, - Rule{Target: UnconditionalAcceptTarget{}}, - Rule{Target: UnconditionalAcceptTarget{}}, - Rule{Target: UnconditionalAcceptTarget{}}, + Rule{Target: AcceptTarget{}}, + Rule{Target: AcceptTarget{}}, + Rule{Target: AcceptTarget{}}, + Rule{Target: AcceptTarget{}}, Rule{Target: ErrorTarget{}}, }, BuiltinChains: map[Hook]int{ @@ -74,8 +74,8 @@ func DefaultTables() IPTables { }, TablenameMangle: Table{ Rules: []Rule{ - Rule{Target: UnconditionalAcceptTarget{}}, - Rule{Target: UnconditionalAcceptTarget{}}, + Rule{Target: AcceptTarget{}}, + Rule{Target: AcceptTarget{}}, Rule{Target: ErrorTarget{}}, }, BuiltinChains: map[Hook]int{ @@ -90,9 +90,9 @@ func DefaultTables() IPTables { }, TablenameFilter: Table{ Rules: []Rule{ - Rule{Target: UnconditionalAcceptTarget{}}, - Rule{Target: UnconditionalAcceptTarget{}}, - Rule{Target: UnconditionalAcceptTarget{}}, + Rule{Target: AcceptTarget{}}, + Rule{Target: AcceptTarget{}}, + Rule{Target: AcceptTarget{}}, Rule{Target: ErrorTarget{}}, }, BuiltinChains: map[Hook]int{ @@ -149,13 +149,11 @@ func (it *IPTables) Check(hook Hook, pkt tcpip.PacketBuffer) bool { for _, tablename := range it.Priorities[hook] { switch verdict := it.checkTable(hook, pkt, tablename); verdict { // If the table returns Accept, move on to the next table. - case Accept: + case TableAccept: continue // The Drop verdict is final. - case Drop: + case TableDrop: return false - case Stolen, Queue, Repeat, None, Jump, Return, Continue: - panic(fmt.Sprintf("Unimplemented verdict %v.", verdict)) default: panic(fmt.Sprintf("Unknown verdict %v.", verdict)) } @@ -166,36 +164,58 @@ func (it *IPTables) Check(hook Hook, pkt tcpip.PacketBuffer) bool { } // Precondition: pkt.NetworkHeader is set. -func (it *IPTables) checkTable(hook Hook, pkt tcpip.PacketBuffer, tablename string) Verdict { +func (it *IPTables) checkTable(hook Hook, pkt tcpip.PacketBuffer, tablename string) TableVerdict { // Start from ruleIdx and walk the list of rules until a rule gives us // a verdict. table := it.Tables[tablename] for ruleIdx := table.BuiltinChains[hook]; ruleIdx < len(table.Rules); ruleIdx++ { switch verdict := it.checkRule(hook, pkt, table, ruleIdx); verdict { - // In either of these cases, this table is done with the packet. - case Accept, Drop: - return verdict - // Continue traversing the rules of the table. - case Continue: + case RuleAccept: + return TableAccept + + case RuleDrop: + return TableDrop + + case RuleContinue: continue - case Stolen, Queue, Repeat, None, Jump, Return: - panic(fmt.Sprintf("Unimplemented verdict %v.", verdict)) + + case RuleReturn: + // TODO(gvisor.dev/issue/170): We don't implement jump + // yet, so any Return is from a built-in chain. That + // means we have to to call the underflow. + underflow := table.Rules[table.Underflows[hook]] + // Underflow is guaranteed to be an unconditional + // ACCEPT or DROP. + switch v, _ := underflow.Target.Action(pkt); v { + case RuleAccept: + return TableAccept + case RuleDrop: + return TableDrop + case RuleContinue, RuleReturn: + panic("Underflows should only return RuleAccept or RuleDrop.") + default: + panic(fmt.Sprintf("Unknown verdict: %d", v)) + } + default: - panic(fmt.Sprintf("Unknown verdict %v.", verdict)) + panic(fmt.Sprintf("Unknown verdict: %d", verdict)) } + } - panic(fmt.Sprintf("Traversed past the entire list of iptables rules in table %q.", tablename)) + // We got through the entire table without a decision. Default to DROP + // for safety. + return TableDrop } // Precondition: pk.NetworkHeader is set. -func (it *IPTables) checkRule(hook Hook, pkt tcpip.PacketBuffer, table Table, ruleIdx int) Verdict { +func (it *IPTables) checkRule(hook Hook, pkt tcpip.PacketBuffer, table Table, ruleIdx int) RuleVerdict { rule := table.Rules[ruleIdx] // First check whether the packet matches the IP header filter. // TODO(gvisor.dev/issue/170): Support other fields of the filter. if rule.Filter.Protocol != 0 && rule.Filter.Protocol != header.IPv4(pkt.NetworkHeader).TransportProtocol() { - return Continue + return RuleContinue } // Go through each rule matcher. If they all match, run @@ -203,10 +223,10 @@ func (it *IPTables) checkRule(hook Hook, pkt tcpip.PacketBuffer, table Table, ru for _, matcher := range rule.Matchers { matches, hotdrop := matcher.Match(hook, pkt, "") if hotdrop { - return Drop + return RuleDrop } if !matches { - return Continue + return RuleContinue } } diff --git a/pkg/tcpip/iptables/targets.go b/pkg/tcpip/iptables/targets.go index 4dd281371..9fc60cfad 100644 --- a/pkg/tcpip/iptables/targets.go +++ b/pkg/tcpip/iptables/targets.go @@ -21,20 +21,20 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" ) -// UnconditionalAcceptTarget accepts all packets. -type UnconditionalAcceptTarget struct{} +// AcceptTarget accepts packets. +type AcceptTarget struct{} // Action implements Target.Action. -func (UnconditionalAcceptTarget) Action(packet tcpip.PacketBuffer) (Verdict, string) { - return Accept, "" +func (AcceptTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, string) { + return RuleAccept, "" } -// UnconditionalDropTarget denies all packets. -type UnconditionalDropTarget struct{} +// DropTarget drops packets. +type DropTarget struct{} // Action implements Target.Action. -func (UnconditionalDropTarget) Action(packet tcpip.PacketBuffer) (Verdict, string) { - return Drop, "" +func (DropTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, string) { + return RuleDrop, "" } // ErrorTarget logs an error and drops the packet. It represents a target that @@ -42,7 +42,26 @@ func (UnconditionalDropTarget) Action(packet tcpip.PacketBuffer) (Verdict, strin type ErrorTarget struct{} // Action implements Target.Action. -func (ErrorTarget) Action(packet tcpip.PacketBuffer) (Verdict, string) { - log.Warningf("ErrorTarget triggered.") - return Drop, "" +func (ErrorTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, string) { + log.Debugf("ErrorTarget triggered.") + return RuleDrop, "" +} + +// UserChainTarget marks a rule as the beginning of a user chain. +type UserChainTarget struct { + Name string +} + +// Action implements Target.Action. +func (UserChainTarget) Action(tcpip.PacketBuffer) (RuleVerdict, string) { + panic("UserChainTarget should never be called.") +} + +// ReturnTarget returns from the current chain. If the chain is a built-in, the +// hook's underflow should be called. +type ReturnTarget struct{} + +// Action implements Target.Action. +func (ReturnTarget) Action(tcpip.PacketBuffer) (RuleVerdict, string) { + return RuleReturn, "" } diff --git a/pkg/tcpip/iptables/types.go b/pkg/tcpip/iptables/types.go index 7d593c35c..5735d001b 100644 --- a/pkg/tcpip/iptables/types.go +++ b/pkg/tcpip/iptables/types.go @@ -56,44 +56,32 @@ const ( NumHooks ) -// A Verdict is returned by a rule's target to indicate how traversal of rules -// should (or should not) continue. -type Verdict int +// A TableVerdict is what a table decides should be done with a packet. +type TableVerdict int const ( - // Invalid indicates an unkonwn or erroneous verdict. - Invalid Verdict = iota + // TableAccept indicates the packet should continue through netstack. + TableAccept TableVerdict = iota - // Accept indicates the packet should continue traversing netstack as - // normal. - Accept - - // Drop inicates the packet should be dropped, stopping traversing - // netstack. - Drop - - // Stolen indicates the packet was co-opted by the target and should - // stop traversing netstack. - Stolen - - // Queue indicates the packet should be queued for userspace processing. - Queue + // TableAccept indicates the packet should be dropped. + TableDrop +) - // Repeat indicates the packet should re-traverse the chains for the - // current hook. - Repeat +// A RuleVerdict is what a rule decides should be done with a packet. +type RuleVerdict int - // None indicates no verdict was reached. - None +const ( + // RuleAccept indicates the packet should continue through netstack. + RuleAccept RuleVerdict = iota - // Jump indicates a jump to another chain. - Jump + // RuleContinue indicates the packet should continue to the next rule. + RuleContinue - // Continue indicates that traversal should continue at the next rule. - Continue + // RuleDrop indicates the packet should be dropped. + RuleDrop - // Return indicates that traversal should return to the calling chain. - Return + // RuleReturn indicates the packet should return to the previous chain. + RuleReturn ) // IPTables holds all the tables for a netstack. @@ -187,5 +175,5 @@ type Target interface { // Action takes an action on the packet and returns a verdict on how // traversal should (or should not) continue. If the return value is // Jump, it also returns the name of the chain to jump to. - Action(packet tcpip.PacketBuffer) (Verdict, string) + Action(packet tcpip.PacketBuffer) (RuleVerdict, string) } diff --git a/test/iptables/filter_input.go b/test/iptables/filter_input.go index bd6059921..e26d6a7d2 100644 --- a/test/iptables/filter_input.go +++ b/test/iptables/filter_input.go @@ -36,6 +36,10 @@ func init() { RegisterTestCase(FilterInputDropTCPSrcPort{}) RegisterTestCase(FilterInputDropUDPPort{}) RegisterTestCase(FilterInputDropUDP{}) + RegisterTestCase(FilterInputCreateUserChain{}) + RegisterTestCase(FilterInputDefaultPolicyAccept{}) + RegisterTestCase(FilterInputDefaultPolicyDrop{}) + RegisterTestCase(FilterInputReturnUnderflow{}) } // FilterInputDropUDP tests that we can drop UDP traffic. @@ -295,8 +299,119 @@ func (FilterInputRequireProtocolUDP) ContainerAction(ip net.IP) error { return nil } -// LocalAction implements TestCase.LocalAction. func (FilterInputRequireProtocolUDP) LocalAction(ip net.IP) error { // No-op. return nil } + +// FilterInputCreateUserChain tests chain creation. +type FilterInputCreateUserChain struct{} + +// Name implements TestCase.Name. +func (FilterInputCreateUserChain) Name() string { + return "FilterInputCreateUserChain" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterInputCreateUserChain) ContainerAction(ip net.IP) error { + // Create a chain. + const chainName = "foochain" + if err := filterTable("-N", chainName); err != nil { + return err + } + + // Add a simple rule to the chain. + return filterTable("-A", chainName, "-j", "DROP") +} + +// LocalAction implements TestCase.LocalAction. +func (FilterInputCreateUserChain) LocalAction(ip net.IP) error { + // No-op. + return nil +} + +// FilterInputDefaultPolicyAccept tests the default ACCEPT policy. +type FilterInputDefaultPolicyAccept struct{} + +// Name implements TestCase.Name. +func (FilterInputDefaultPolicyAccept) Name() string { + return "FilterInputDefaultPolicyAccept" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterInputDefaultPolicyAccept) ContainerAction(ip net.IP) error { + // Set the default policy to accept, then receive a packet. + if err := filterTable("-P", "INPUT", "ACCEPT"); err != nil { + return err + } + return listenUDP(acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (FilterInputDefaultPolicyAccept) LocalAction(ip net.IP) error { + return sendUDPLoop(ip, acceptPort, sendloopDuration) +} + +// FilterInputDefaultPolicyDrop tests the default DROP policy. +type FilterInputDefaultPolicyDrop struct{} + +// Name implements TestCase.Name. +func (FilterInputDefaultPolicyDrop) Name() string { + return "FilterInputDefaultPolicyDrop" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterInputDefaultPolicyDrop) ContainerAction(ip net.IP) error { + if err := filterTable("-P", "INPUT", "DROP"); err != nil { + return err + } + + // Listen for UDP packets on dropPort. + if err := listenUDP(dropPort, sendloopDuration); err == nil { + return fmt.Errorf("packets on port %d should have been dropped, but got a packet", dropPort) + } else if netErr, ok := err.(net.Error); !ok || !netErr.Timeout() { + return fmt.Errorf("error reading: %v", err) + } + + // At this point we know that reading timed out and never received a + // packet. + return nil +} + +// LocalAction implements TestCase.LocalAction. +func (FilterInputDefaultPolicyDrop) LocalAction(ip net.IP) error { + return sendUDPLoop(ip, acceptPort, sendloopDuration) +} + +// FilterInputReturnUnderflow tests that -j RETURN in a built-in chain causes +// the underflow rule (i.e. default policy) to be executed. +type FilterInputReturnUnderflow struct{} + +// Name implements TestCase.Name. +func (FilterInputReturnUnderflow) Name() string { + return "FilterInputReturnUnderflow" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterInputReturnUnderflow) ContainerAction(ip net.IP) error { + // Add a RETURN rule followed by an unconditional accept, and set the + // default policy to DROP. + if err := filterTable("-A", "INPUT", "-j", "RETURN"); err != nil { + return err + } + if err := filterTable("-A", "INPUT", "-j", "DROP"); err != nil { + return err + } + if err := filterTable("-P", "INPUT", "ACCEPT"); err != nil { + return err + } + + // We should receive packets, as the RETURN rule will trigger the default + // ACCEPT policy. + return listenUDP(acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (FilterInputReturnUnderflow) LocalAction(ip net.IP) error { + return sendUDPLoop(ip, acceptPort, sendloopDuration) +} diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go index 41909582a..46a7c99b0 100644 --- a/test/iptables/iptables_test.go +++ b/test/iptables/iptables_test.go @@ -214,6 +214,30 @@ func TestFilterInputDropTCPSrcPort(t *testing.T) { } } +func TestFilterInputCreateUserChain(t *testing.T) { + if err := singleTest(FilterInputCreateUserChain{}); err != nil { + t.Fatal(err) + } +} + +func TestFilterInputDefaultPolicyAccept(t *testing.T) { + if err := singleTest(FilterInputDefaultPolicyAccept{}); err != nil { + t.Fatal(err) + } +} + +func TestFilterInputDefaultPolicyDrop(t *testing.T) { + if err := singleTest(FilterInputDefaultPolicyDrop{}); err != nil { + t.Fatal(err) + } +} + +func TestFilterInputReturnUnderflow(t *testing.T) { + if err := singleTest(FilterInputReturnUnderflow{}); err != nil { + t.Fatal(err) + } +} + func TestFilterOutputDropTCPDestPort(t *testing.T) { if err := singleTest(FilterOutputDropTCPDestPort{}); err != nil { t.Fatal(err) -- cgit v1.2.3 From d30a884775556474f1a893fd30460a6edf0a3039 Mon Sep 17 00:00:00 2001 From: Haibo Xu Date: Tue, 11 Feb 2020 00:52:23 +0000 Subject: Add definition of arch.ARMTrapFlag. Fixes #1708 Signed-off-by: Haibo Xu haibo.xu@arm.com Change-Id: Ib15768692ead17c81c06f7666ca3f0a14064c3a0 --- pkg/sentry/arch/arch_aarch64.go | 3 +++ 1 file changed, 3 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/arch/arch_aarch64.go b/pkg/sentry/arch/arch_aarch64.go index 3b6987665..d7794db63 100644 --- a/pkg/sentry/arch/arch_aarch64.go +++ b/pkg/sentry/arch/arch_aarch64.go @@ -34,6 +34,9 @@ const ( SyscallWidth = 4 ) +// ARMTrapFlag is the mask for the trap flag. +const ARMTrapFlag = uint64(1) << 21 + // aarch64FPState is aarch64 floating point state. type aarch64FPState []byte -- cgit v1.2.3 From 69bf39e8a47d3b4dcbbd04d2e8df476cdfab5e74 Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Thu, 13 Feb 2020 10:58:47 -0800 Subject: Internal change. PiperOrigin-RevId: 294952610 --- pkg/abi/linux/socket.go | 13 ++++ pkg/sentry/socket/control/BUILD | 1 + pkg/sentry/socket/control/control.go | 43 +++++++++++++ pkg/sentry/socket/hostinet/socket.go | 11 +++- pkg/sentry/socket/netstack/netstack.go | 37 ++++++++++-- pkg/tcpip/tcpip.go | 25 ++++++++ pkg/tcpip/transport/udp/endpoint.go | 26 ++++++++ test/syscalls/linux/socket_ip_udp_generic.cc | 44 ++++++++++++++ test/syscalls/linux/socket_ipv4_udp_unbound.cc | 84 ++++++++++++++++++++++++++ test/syscalls/linux/udp_socket_test_cases.cc | 1 - 10 files changed, 278 insertions(+), 7 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/socket.go b/pkg/abi/linux/socket.go index 766ee4014..4a14ef691 100644 --- a/pkg/abi/linux/socket.go +++ b/pkg/abi/linux/socket.go @@ -411,6 +411,15 @@ type ControlMessageCredentials struct { GID uint32 } +// A ControlMessageIPPacketInfo is IP_PKTINFO socket control message. +// +// ControlMessageIPPacketInfo represents struct in_pktinfo from linux/in.h. +type ControlMessageIPPacketInfo struct { + NIC int32 + LocalAddr InetAddr + DestinationAddr InetAddr +} + // SizeOfControlMessageCredentials is the binary size of a // ControlMessageCredentials struct. var SizeOfControlMessageCredentials = int(binary.Size(ControlMessageCredentials{})) @@ -431,6 +440,10 @@ const SizeOfControlMessageTOS = 1 // SizeOfControlMessageTClass is the size of an IPV6_TCLASS control message. const SizeOfControlMessageTClass = 4 +// SizeOfControlMessageIPPacketInfo is the size of an IP_PKTINFO +// control message. +const SizeOfControlMessageIPPacketInfo = 12 + // SCM_MAX_FD is the maximum number of FDs accepted in a single sendmsg call. // From net/scm.h. const SCM_MAX_FD = 253 diff --git a/pkg/sentry/socket/control/BUILD b/pkg/sentry/socket/control/BUILD index 79e16d6e8..4d42d29cb 100644 --- a/pkg/sentry/socket/control/BUILD +++ b/pkg/sentry/socket/control/BUILD @@ -19,6 +19,7 @@ go_library( "//pkg/sentry/socket", "//pkg/sentry/socket/unix/transport", "//pkg/syserror", + "//pkg/tcpip", "//pkg/usermem", ], ) diff --git a/pkg/sentry/socket/control/control.go b/pkg/sentry/socket/control/control.go index 6145a7fc3..4667373d2 100644 --- a/pkg/sentry/socket/control/control.go +++ b/pkg/sentry/socket/control/control.go @@ -26,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/usermem" ) @@ -338,6 +339,22 @@ func PackTClass(t *kernel.Task, tClass int32, buf []byte) []byte { ) } +// PackIPPacketInfo packs an IP_PKTINFO socket control message. +func PackIPPacketInfo(t *kernel.Task, packetInfo tcpip.IPPacketInfo, buf []byte) []byte { + var p linux.ControlMessageIPPacketInfo + p.NIC = int32(packetInfo.NIC) + copy(p.LocalAddr[:], []byte(packetInfo.LocalAddr)) + copy(p.DestinationAddr[:], []byte(packetInfo.DestinationAddr)) + + return putCmsgStruct( + buf, + linux.SOL_IP, + linux.IP_PKTINFO, + t.Arch().Width(), + p, + ) +} + // PackControlMessages packs control messages into the given buffer. // // We skip control messages specific to Unix domain sockets. @@ -362,6 +379,10 @@ func PackControlMessages(t *kernel.Task, cmsgs socket.ControlMessages, buf []byt buf = PackTClass(t, cmsgs.IP.TClass, buf) } + if cmsgs.IP.HasIPPacketInfo { + buf = PackIPPacketInfo(t, cmsgs.IP.PacketInfo, buf) + } + return buf } @@ -394,6 +415,16 @@ func CmsgsSpace(t *kernel.Task, cmsgs socket.ControlMessages) int { return space } +// NewIPPacketInfo returns the IPPacketInfo struct. +func NewIPPacketInfo(packetInfo linux.ControlMessageIPPacketInfo) tcpip.IPPacketInfo { + var p tcpip.IPPacketInfo + p.NIC = tcpip.NICID(packetInfo.NIC) + copy([]byte(p.LocalAddr), packetInfo.LocalAddr[:]) + copy([]byte(p.DestinationAddr), packetInfo.DestinationAddr[:]) + + return p +} + // Parse parses a raw socket control message into portable objects. func Parse(t *kernel.Task, socketOrEndpoint interface{}, buf []byte) (socket.ControlMessages, error) { var ( @@ -468,6 +499,18 @@ func Parse(t *kernel.Task, socketOrEndpoint interface{}, buf []byte) (socket.Con binary.Unmarshal(buf[i:i+linux.SizeOfControlMessageTOS], usermem.ByteOrder, &cmsgs.IP.TOS) i += binary.AlignUp(length, width) + case linux.IP_PKTINFO: + if length < linux.SizeOfControlMessageIPPacketInfo { + return socket.ControlMessages{}, syserror.EINVAL + } + + cmsgs.IP.HasIPPacketInfo = true + var packetInfo linux.ControlMessageIPPacketInfo + binary.Unmarshal(buf[i:i+linux.SizeOfControlMessageIPPacketInfo], usermem.ByteOrder, &packetInfo) + + cmsgs.IP.PacketInfo = NewIPPacketInfo(packetInfo) + i += binary.AlignUp(length, width) + default: return socket.ControlMessages{}, syserror.EINVAL } diff --git a/pkg/sentry/socket/hostinet/socket.go b/pkg/sentry/socket/hostinet/socket.go index de76388ac..22f78d2e2 100644 --- a/pkg/sentry/socket/hostinet/socket.go +++ b/pkg/sentry/socket/hostinet/socket.go @@ -289,7 +289,7 @@ func (s *socketOperations) GetSockOpt(t *kernel.Task, level int, name int, outPt switch level { case linux.SOL_IP: switch name { - case linux.IP_TOS, linux.IP_RECVTOS: + case linux.IP_TOS, linux.IP_RECVTOS, linux.IP_PKTINFO: optlen = sizeofInt32 } case linux.SOL_IPV6: @@ -336,6 +336,8 @@ func (s *socketOperations) SetSockOpt(t *kernel.Task, level int, name int, opt [ switch name { case linux.IP_TOS, linux.IP_RECVTOS: optlen = sizeofInt32 + case linux.IP_PKTINFO: + optlen = linux.SizeOfControlMessageIPPacketInfo } case linux.SOL_IPV6: switch name { @@ -473,7 +475,14 @@ func (s *socketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags case syscall.IP_TOS: controlMessages.IP.HasTOS = true binary.Unmarshal(unixCmsg.Data[:linux.SizeOfControlMessageTOS], usermem.ByteOrder, &controlMessages.IP.TOS) + + case syscall.IP_PKTINFO: + controlMessages.IP.HasIPPacketInfo = true + var packetInfo linux.ControlMessageIPPacketInfo + binary.Unmarshal(unixCmsg.Data[:linux.SizeOfControlMessageIPPacketInfo], usermem.ByteOrder, &packetInfo) + controlMessages.IP.PacketInfo = control.NewIPPacketInfo(packetInfo) } + case syscall.SOL_IPV6: switch unixCmsg.Header.Type { case syscall.IPV6_TCLASS: diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index ed2fbcceb..9757fbfba 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -1414,6 +1414,21 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in } return o, nil + case linux.IP_PKTINFO: + if outLen < sizeOfInt32 { + return nil, syserr.ErrInvalidArgument + } + + v, err := ep.GetSockOptBool(tcpip.ReceiveIPPacketInfoOption) + if err != nil { + return nil, syserr.TranslateNetstackError(err) + } + var o int32 + if v { + o = 1 + } + return o, nil + default: emitUnimplementedEventIP(t, name) } @@ -1762,6 +1777,7 @@ func setSockOptIPv6(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) linux.IPV6_IPSEC_POLICY, linux.IPV6_JOIN_ANYCAST, linux.IPV6_LEAVE_ANYCAST, + // TODO(b/148887420): Add support for IPV6_PKTINFO. linux.IPV6_PKTINFO, linux.IPV6_ROUTER_ALERT, linux.IPV6_XFRM_POLICY, @@ -1949,6 +1965,16 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s } return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.ReceiveTOSOption, v != 0)) + case linux.IP_PKTINFO: + if len(optVal) == 0 { + return nil + } + v, err := parseIntOrChar(optVal) + if err != nil { + return err + } + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.ReceiveIPPacketInfoOption, v != 0)) + case linux.IP_ADD_SOURCE_MEMBERSHIP, linux.IP_BIND_ADDRESS_NO_PORT, linux.IP_BLOCK_SOURCE, @@ -1964,7 +1990,6 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s linux.IP_NODEFRAG, linux.IP_OPTIONS, linux.IP_PASSSEC, - linux.IP_PKTINFO, linux.IP_RECVERR, linux.IP_RECVFRAGSIZE, linux.IP_RECVOPTS, @@ -2395,10 +2420,12 @@ func (s *SocketOperations) nonBlockingRead(ctx context.Context, dst usermem.IOSe func (s *SocketOperations) controlMessages() socket.ControlMessages { return socket.ControlMessages{ IP: tcpip.ControlMessages{ - HasTimestamp: s.readCM.HasTimestamp && s.sockOptTimestamp, - Timestamp: s.readCM.Timestamp, - HasTOS: s.readCM.HasTOS, - TOS: s.readCM.TOS, + HasTimestamp: s.readCM.HasTimestamp && s.sockOptTimestamp, + Timestamp: s.readCM.Timestamp, + HasTOS: s.readCM.HasTOS, + TOS: s.readCM.TOS, + HasIPPacketInfo: s.readCM.HasIPPacketInfo, + PacketInfo: s.readCM.PacketInfo, }, } } diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 0e944712f..9ca39ce40 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -328,6 +328,12 @@ type ControlMessages struct { // Tclass is the IPv6 traffic class of the associated packet. TClass int32 + + // HasIPPacketInfo indicates whether PacketInfo is set. + HasIPPacketInfo bool + + // PacketInfo holds interface and address data on an incoming packet. + PacketInfo IPPacketInfo } // Endpoint is the interface implemented by transport protocols (e.g., tcp, udp) @@ -503,6 +509,11 @@ const ( // V6OnlyOption is used by {G,S}etSockOptBool to specify whether an IPv6 // socket is to be restricted to sending and receiving IPv6 packets only. V6OnlyOption + + // ReceiveIPPacketInfoOption is used by {G,S}etSockOptBool to specify + // if more inforamtion is provided with incoming packets such + // as interface index and address. + ReceiveIPPacketInfoOption ) // SockOptInt represents socket options which values have the int type. @@ -685,6 +696,20 @@ type IPv4TOSOption uint8 // for all subsequent outgoing IPv6 packets from the endpoint. type IPv6TrafficClassOption uint8 +// IPPacketInfo is the message struture for IP_PKTINFO. +// +// +stateify savable +type IPPacketInfo struct { + // NIC is the ID of the NIC to be used. + NIC NICID + + // LocalAddr is the local address. + LocalAddr Address + + // DestinationAddr is the destination address. + DestinationAddr Address +} + // Route is a row in the routing table. It specifies through which NIC (and // gateway) sets of packets should be routed. A row is considered viable if the // masked target address matches the destination address in the row. diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index c9cbed8f4..3fe91cac2 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -29,6 +29,7 @@ import ( type udpPacket struct { udpPacketEntry senderAddress tcpip.FullAddress + packetInfo tcpip.IPPacketInfo data buffer.VectorisedView `state:".(buffer.VectorisedView)"` timestamp int64 tos uint8 @@ -118,6 +119,9 @@ type endpoint struct { // as ancillary data to ControlMessages on Read. receiveTOS bool + // receiveIPPacketInfo determines if the packet info is returned by Read. + receiveIPPacketInfo bool + // shutdownFlags represent the current shutdown state of the endpoint. shutdownFlags tcpip.ShutdownFlags @@ -254,11 +258,17 @@ func (e *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMess } e.mu.RLock() receiveTOS := e.receiveTOS + receiveIPPacketInfo := e.receiveIPPacketInfo e.mu.RUnlock() if receiveTOS { cm.HasTOS = true cm.TOS = p.tos } + + if receiveIPPacketInfo { + cm.HasIPPacketInfo = true + cm.PacketInfo = p.packetInfo + } return p.data.ToView(), cm, nil } @@ -495,6 +505,13 @@ func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { } e.v6only = v + return nil + + case tcpip.ReceiveIPPacketInfoOption: + e.mu.Lock() + e.receiveIPPacketInfo = v + e.mu.Unlock() + return nil } return nil @@ -703,6 +720,12 @@ func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { e.mu.RUnlock() return v, nil + + case tcpip.ReceiveIPPacketInfoOption: + e.mu.RLock() + v := e.receiveIPPacketInfo + e.mu.RUnlock() + return v, nil } return false, tcpip.ErrUnknownProtocolOption @@ -1247,6 +1270,9 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk switch r.NetProto { case header.IPv4ProtocolNumber: packet.tos, _ = header.IPv4(pkt.NetworkHeader).TOS() + packet.packetInfo.LocalAddr = r.LocalAddress + packet.packetInfo.DestinationAddr = r.RemoteAddress + packet.packetInfo.NIC = r.NICID() } packet.timestamp = e.stack.NowNanoseconds() diff --git a/test/syscalls/linux/socket_ip_udp_generic.cc b/test/syscalls/linux/socket_ip_udp_generic.cc index 53290bed7..db5663ecd 100644 --- a/test/syscalls/linux/socket_ip_udp_generic.cc +++ b/test/syscalls/linux/socket_ip_udp_generic.cc @@ -357,5 +357,49 @@ TEST_P(UDPSocketPairTest, SetReuseAddrReusePort) { EXPECT_EQ(get, kSockOptOn); } +// Test getsockopt for a socket which is not set with IP_PKTINFO option. +TEST_P(UDPSocketPairTest, IPPKTINFODefault) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + int get = -1; + socklen_t get_len = sizeof(get); + + ASSERT_THAT( + getsockopt(sockets->first_fd(), SOL_IP, IP_PKTINFO, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kSockOptOff); +} + +// Test setsockopt and getsockopt for a socket with IP_PKTINFO option. +TEST_P(UDPSocketPairTest, SetAndGetIPPKTINFO) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + int level = SOL_IP; + int type = IP_PKTINFO; + + // Check getsockopt before IP_PKTINFO is set. + int get = -1; + socklen_t get_len = sizeof(get); + + ASSERT_THAT(setsockopt(sockets->first_fd(), level, type, &kSockOptOn, + sizeof(kSockOptOn)), + SyscallSucceedsWithValue(0)); + + ASSERT_THAT(getsockopt(sockets->first_fd(), level, type, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get, kSockOptOn); + EXPECT_EQ(get_len, sizeof(get)); + + ASSERT_THAT(setsockopt(sockets->first_fd(), level, type, &kSockOptOff, + sizeof(kSockOptOff)), + SyscallSucceedsWithValue(0)); + + ASSERT_THAT(getsockopt(sockets->first_fd(), level, type, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get, kSockOptOff); + EXPECT_EQ(get_len, sizeof(get)); +} + } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_ipv4_udp_unbound.cc b/test/syscalls/linux/socket_ipv4_udp_unbound.cc index 990ccf23c..bc4b07a62 100644 --- a/test/syscalls/linux/socket_ipv4_udp_unbound.cc +++ b/test/syscalls/linux/socket_ipv4_udp_unbound.cc @@ -15,6 +15,7 @@ #include "test/syscalls/linux/socket_ipv4_udp_unbound.h" #include +#include #include #include #include @@ -2128,5 +2129,88 @@ TEST_P(IPv4UDPUnboundSocketTest, ReuseAddrReusePortDistribution) { SyscallSucceedsWithValue(kMessageSize)); } +// Test that socket will receive packet info control message. +TEST_P(IPv4UDPUnboundSocketTest, SetAndReceiveIPPKTINFO) { + // TODO(gvisor.dev/issue/1202): ioctl() is not supported by hostinet. + SKIP_IF((IsRunningWithHostinet())); + + auto sender = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + auto receiver = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + auto sender_addr = V4Loopback(); + int level = SOL_IP; + int type = IP_PKTINFO; + + ASSERT_THAT( + bind(receiver->get(), reinterpret_cast(&sender_addr.addr), + sender_addr.addr_len), + SyscallSucceeds()); + socklen_t sender_addr_len = sender_addr.addr_len; + ASSERT_THAT(getsockname(receiver->get(), + reinterpret_cast(&sender_addr.addr), + &sender_addr_len), + SyscallSucceeds()); + EXPECT_EQ(sender_addr_len, sender_addr.addr_len); + + auto receiver_addr = V4Loopback(); + reinterpret_cast(&receiver_addr.addr)->sin_port = + reinterpret_cast(&sender_addr.addr)->sin_port; + ASSERT_THAT( + connect(sender->get(), reinterpret_cast(&receiver_addr.addr), + receiver_addr.addr_len), + SyscallSucceeds()); + + // Allow socket to receive control message. + ASSERT_THAT( + setsockopt(receiver->get(), level, type, &kSockOptOn, sizeof(kSockOptOn)), + SyscallSucceeds()); + + // Prepare message to send. + constexpr size_t kDataLength = 1024; + msghdr sent_msg = {}; + iovec sent_iov = {}; + char sent_data[kDataLength]; + sent_iov.iov_base = sent_data; + sent_iov.iov_len = kDataLength; + sent_msg.msg_iov = &sent_iov; + sent_msg.msg_iovlen = 1; + sent_msg.msg_flags = 0; + + ASSERT_THAT(RetryEINTR(sendmsg)(sender->get(), &sent_msg, 0), + SyscallSucceedsWithValue(kDataLength)); + + msghdr received_msg = {}; + iovec received_iov = {}; + char received_data[kDataLength]; + char received_cmsg_buf[CMSG_SPACE(sizeof(in_pktinfo))] = {}; + size_t cmsg_data_len = sizeof(in_pktinfo); + received_iov.iov_base = received_data; + received_iov.iov_len = kDataLength; + received_msg.msg_iov = &received_iov; + received_msg.msg_iovlen = 1; + received_msg.msg_controllen = CMSG_LEN(cmsg_data_len); + received_msg.msg_control = received_cmsg_buf; + + ASSERT_THAT(RetryEINTR(recvmsg)(receiver->get(), &received_msg, 0), + SyscallSucceedsWithValue(kDataLength)); + + cmsghdr* cmsg = CMSG_FIRSTHDR(&received_msg); + ASSERT_NE(cmsg, nullptr); + EXPECT_EQ(cmsg->cmsg_len, CMSG_LEN(cmsg_data_len)); + EXPECT_EQ(cmsg->cmsg_level, level); + EXPECT_EQ(cmsg->cmsg_type, type); + + // Get loopback index. + ifreq ifr = {}; + absl::SNPrintF(ifr.ifr_name, IFNAMSIZ, "lo"); + ASSERT_THAT(ioctl(sender->get(), SIOCGIFINDEX, &ifr), SyscallSucceeds()); + ASSERT_NE(ifr.ifr_ifindex, 0); + + // Check the data + in_pktinfo received_pktinfo = {}; + memcpy(&received_pktinfo, CMSG_DATA(cmsg), sizeof(in_pktinfo)); + EXPECT_EQ(received_pktinfo.ipi_ifindex, ifr.ifr_ifindex); + EXPECT_EQ(received_pktinfo.ipi_spec_dst.s_addr, htonl(INADDR_LOOPBACK)); + EXPECT_EQ(received_pktinfo.ipi_addr.s_addr, htonl(INADDR_LOOPBACK)); +} } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/udp_socket_test_cases.cc b/test/syscalls/linux/udp_socket_test_cases.cc index a2f6ef8cc..9f8de6b48 100644 --- a/test/syscalls/linux/udp_socket_test_cases.cc +++ b/test/syscalls/linux/udp_socket_test_cases.cc @@ -1495,6 +1495,5 @@ TEST_P(UdpSocketTest, SendAndReceiveTOS) { memcpy(&received_tos, CMSG_DATA(cmsg), sizeof(received_tos)); EXPECT_EQ(received_tos, sent_tos); } - } // namespace testing } // namespace gvisor -- cgit v1.2.3 From 6ef63cd7da107d487fda7c48af50fa9802913cd9 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Wed, 12 Feb 2020 16:19:06 -0800 Subject: We can now create and jump in iptables. For example: $ iptables -N foochain $ iptables -A INPUT -j foochain --- pkg/abi/linux/netfilter.go | 9 +- pkg/sentry/socket/netfilter/BUILD | 1 + pkg/sentry/socket/netfilter/netfilter.go | 62 +++++++-- pkg/sentry/socket/netfilter/targets.go | 35 +++++ pkg/tcpip/iptables/iptables.go | 103 +++++++++------ pkg/tcpip/iptables/targets.go | 20 ++- pkg/tcpip/iptables/types.go | 21 +-- test/iptables/filter_input.go | 217 ++++++++++++++++++++++++++++--- test/iptables/iptables_test.go | 36 +++++ test/iptables/iptables_util.go | 10 ++ 10 files changed, 420 insertions(+), 94 deletions(-) create mode 100644 pkg/sentry/socket/netfilter/targets.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/netfilter.go b/pkg/abi/linux/netfilter.go index bbc4df74c..bd2e13ba1 100644 --- a/pkg/abi/linux/netfilter.go +++ b/pkg/abi/linux/netfilter.go @@ -225,11 +225,14 @@ type XTEntryTarget struct { // SizeOfXTEntryTarget is the size of an XTEntryTarget. const SizeOfXTEntryTarget = 32 -// XTStandardTarget is a builtin target, one of ACCEPT, DROP, JUMP, QUEUE, or -// RETURN. It corresponds to struct xt_standard_target in +// XTStandardTarget is a built-in target, one of ACCEPT, DROP, JUMP, QUEUE, +// RETURN, or jump. It corresponds to struct xt_standard_target in // include/uapi/linux/netfilter/x_tables.h. type XTStandardTarget struct { - Target XTEntryTarget + Target XTEntryTarget + // A positive verdict indicates a jump, and is the offset from the + // start of the table to jump to. A negative value means one of the + // other built-in targets. Verdict int32 _ [4]byte } diff --git a/pkg/sentry/socket/netfilter/BUILD b/pkg/sentry/socket/netfilter/BUILD index c91ec7494..7cd2ce55b 100644 --- a/pkg/sentry/socket/netfilter/BUILD +++ b/pkg/sentry/socket/netfilter/BUILD @@ -7,6 +7,7 @@ go_library( srcs = [ "extensions.go", "netfilter.go", + "targets.go", "tcp_matcher.go", "udp_matcher.go", ], diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 3fc80e0de..d322e4144 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -240,13 +240,15 @@ func marshalTarget(target iptables.Target) []byte { return marshalErrorTarget(tg.Name) case iptables.ReturnTarget: return marshalStandardTarget(iptables.RuleReturn) + case JumpTarget: + return marshalJumpTarget(tg) default: panic(fmt.Errorf("unknown target of type %T", target)) } } func marshalStandardTarget(verdict iptables.RuleVerdict) []byte { - nflog("convert to binary: marshalling standard target with size %d", linux.SizeOfXTStandardTarget) + nflog("convert to binary: marshalling standard target") // The target's name will be the empty string. target := linux.XTStandardTarget{ @@ -274,6 +276,23 @@ func marshalErrorTarget(errorName string) []byte { return binary.Marshal(ret, usermem.ByteOrder, target) } +func marshalJumpTarget(jt JumpTarget) []byte { + nflog("convert to binary: marshalling jump target") + + // The target's name will be the empty string. + target := linux.XTStandardTarget{ + Target: linux.XTEntryTarget{ + TargetSize: linux.SizeOfXTStandardTarget, + }, + // Verdict is overloaded by the ABI. When positive, it holds + // the jump offset from the start of the table. + Verdict: int32(jt.Offset), + } + + ret := make([]byte, 0, linux.SizeOfXTStandardTarget) + return binary.Marshal(ret, usermem.ByteOrder, target) +} + // translateFromStandardVerdict translates verdicts the same way as the iptables // tool. func translateFromStandardVerdict(verdict iptables.RuleVerdict) int32 { @@ -335,7 +354,8 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { // Convert input into a list of rules and their offsets. var offset uint32 - var offsets []uint32 + // offsets maps rule byte offsets to their position in table.Rules. + offsets := map[uint32]int{} for entryIdx := uint32(0); entryIdx < replace.NumEntries; entryIdx++ { nflog("set entries: processing entry at offset %d", offset) @@ -396,11 +416,12 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { Target: target, Matchers: matchers, }) - offsets = append(offsets, offset) + offsets[offset] = int(entryIdx) offset += uint32(entry.NextOffset) if initialOptValLen-len(optVal) != int(entry.NextOffset) { nflog("entry NextOffset is %d, but entry took up %d bytes", entry.NextOffset, initialOptValLen-len(optVal)) + return syserr.ErrInvalidArgument } } @@ -409,13 +430,13 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { for hook, _ := range replace.HookEntry { if table.ValidHooks()&(1<= 0 indicates a jump. + return JumpTarget{Offset: uint32(standardTarget.Verdict)}, nil + } case errorTargetName: // Error target. diff --git a/pkg/sentry/socket/netfilter/targets.go b/pkg/sentry/socket/netfilter/targets.go new file mode 100644 index 000000000..c421b87cf --- /dev/null +++ b/pkg/sentry/socket/netfilter/targets.go @@ -0,0 +1,35 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package netfilter + +import ( + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/iptables" +) + +// JumpTarget implements iptables.Target. +type JumpTarget struct { + // Offset is the byte offset of the rule to jump to. It is used for + // marshaling and unmarshaling. + Offset uint32 + + // RuleNum is the rule to jump to. + RuleNum int +} + +// Action implements iptables.Target.Action. +func (jt JumpTarget) Action(tcpip.PacketBuffer) (iptables.RuleVerdict, int) { + return iptables.RuleJump, jt.RuleNum +} diff --git a/pkg/tcpip/iptables/iptables.go b/pkg/tcpip/iptables/iptables.go index 75a433a3b..dbaccbb36 100644 --- a/pkg/tcpip/iptables/iptables.go +++ b/pkg/tcpip/iptables/iptables.go @@ -135,25 +135,53 @@ func EmptyFilterTable() Table { } } +// A chainVerdict is what a table decides should be done with a packet. +type chainVerdict int + +const ( + // chainAccept indicates the packet should continue through netstack. + chainAccept chainVerdict = iota + + // chainAccept indicates the packet should be dropped. + chainDrop + + // chainReturn indicates the packet should return to the calling chain + // or the underflow rule of a builtin chain. + chainReturn +) + // Check runs pkt through the rules for hook. It returns true when the packet // should continue traversing the network stack and false when it should be // dropped. // // Precondition: pkt.NetworkHeader is set. func (it *IPTables) Check(hook Hook, pkt tcpip.PacketBuffer) bool { - // TODO(gvisor.dev/issue/170): A lot of this is uncomplicated because - // we're missing features. Jumps, the call stack, etc. aren't checked - // for yet because we're yet to support them. - // Go through each table containing the hook. for _, tablename := range it.Priorities[hook] { - switch verdict := it.checkTable(hook, pkt, tablename); verdict { + table := it.Tables[tablename] + ruleIdx := table.BuiltinChains[hook] + switch verdict := it.checkChain(hook, pkt, table, ruleIdx); verdict { // If the table returns Accept, move on to the next table. - case TableAccept: + case chainAccept: continue // The Drop verdict is final. - case TableDrop: + case chainDrop: return false + case chainReturn: + // Any Return from a built-in chain means we have to + // call the underflow. + underflow := table.Rules[table.Underflows[hook]] + switch v, _ := underflow.Target.Action(pkt); v { + case RuleAccept: + continue + case RuleDrop: + return false + case RuleJump, RuleReturn: + panic("Underflows should only return RuleAccept or RuleDrop.") + default: + panic(fmt.Sprintf("Unknown verdict: %d", v)) + } + default: panic(fmt.Sprintf("Unknown verdict %v.", verdict)) } @@ -164,37 +192,37 @@ func (it *IPTables) Check(hook Hook, pkt tcpip.PacketBuffer) bool { } // Precondition: pkt.NetworkHeader is set. -func (it *IPTables) checkTable(hook Hook, pkt tcpip.PacketBuffer, tablename string) TableVerdict { +func (it *IPTables) checkChain(hook Hook, pkt tcpip.PacketBuffer, table Table, ruleIdx int) chainVerdict { // Start from ruleIdx and walk the list of rules until a rule gives us // a verdict. - table := it.Tables[tablename] - for ruleIdx := table.BuiltinChains[hook]; ruleIdx < len(table.Rules); ruleIdx++ { - switch verdict := it.checkRule(hook, pkt, table, ruleIdx); verdict { + for ruleIdx < len(table.Rules) { + switch verdict, jumpTo := it.checkRule(hook, pkt, table, ruleIdx); verdict { case RuleAccept: - return TableAccept + return chainAccept case RuleDrop: - return TableDrop - - case RuleContinue: - continue + return chainDrop case RuleReturn: - // TODO(gvisor.dev/issue/170): We don't implement jump - // yet, so any Return is from a built-in chain. That - // means we have to to call the underflow. - underflow := table.Rules[table.Underflows[hook]] - // Underflow is guaranteed to be an unconditional - // ACCEPT or DROP. - switch v, _ := underflow.Target.Action(pkt); v { - case RuleAccept: - return TableAccept - case RuleDrop: - return TableDrop - case RuleContinue, RuleReturn: - panic("Underflows should only return RuleAccept or RuleDrop.") + return chainReturn + + case RuleJump: + // "Jumping" to the next rule just means we're + // continuing on down the list. + if jumpTo == ruleIdx+1 { + ruleIdx++ + continue + } + switch verdict := it.checkChain(hook, pkt, table, jumpTo); verdict { + case chainAccept: + return chainAccept + case chainDrop: + return chainDrop + case chainReturn: + ruleIdx++ + continue default: - panic(fmt.Sprintf("Unknown verdict: %d", v)) + panic(fmt.Sprintf("Unknown verdict: %d", verdict)) } default: @@ -205,17 +233,18 @@ func (it *IPTables) checkTable(hook Hook, pkt tcpip.PacketBuffer, tablename stri // We got through the entire table without a decision. Default to DROP // for safety. - return TableDrop + return chainDrop } // Precondition: pk.NetworkHeader is set. -func (it *IPTables) checkRule(hook Hook, pkt tcpip.PacketBuffer, table Table, ruleIdx int) RuleVerdict { +func (it *IPTables) checkRule(hook Hook, pkt tcpip.PacketBuffer, table Table, ruleIdx int) (RuleVerdict, int) { rule := table.Rules[ruleIdx] // First check whether the packet matches the IP header filter. // TODO(gvisor.dev/issue/170): Support other fields of the filter. if rule.Filter.Protocol != 0 && rule.Filter.Protocol != header.IPv4(pkt.NetworkHeader).TransportProtocol() { - return RuleContinue + // Continue on to the next rule. + return RuleJump, ruleIdx + 1 } // Go through each rule matcher. If they all match, run @@ -223,14 +252,14 @@ func (it *IPTables) checkRule(hook Hook, pkt tcpip.PacketBuffer, table Table, ru for _, matcher := range rule.Matchers { matches, hotdrop := matcher.Match(hook, pkt, "") if hotdrop { - return RuleDrop + return RuleDrop, 0 } if !matches { - return RuleContinue + // Continue on to the next rule. + return RuleJump, ruleIdx + 1 } } // All the matchers matched, so run the target. - verdict, _ := rule.Target.Action(pkt) - return verdict + return rule.Target.Action(pkt) } diff --git a/pkg/tcpip/iptables/targets.go b/pkg/tcpip/iptables/targets.go index 9fc60cfad..81a2e39a2 100644 --- a/pkg/tcpip/iptables/targets.go +++ b/pkg/tcpip/iptables/targets.go @@ -12,8 +12,6 @@ // See the License for the specific language governing permissions and // limitations under the License. -// This file contains various Targets. - package iptables import ( @@ -25,16 +23,16 @@ import ( type AcceptTarget struct{} // Action implements Target.Action. -func (AcceptTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, string) { - return RuleAccept, "" +func (AcceptTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, int) { + return RuleAccept, 0 } // DropTarget drops packets. type DropTarget struct{} // Action implements Target.Action. -func (DropTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, string) { - return RuleDrop, "" +func (DropTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, int) { + return RuleDrop, 0 } // ErrorTarget logs an error and drops the packet. It represents a target that @@ -42,9 +40,9 @@ func (DropTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, string) { type ErrorTarget struct{} // Action implements Target.Action. -func (ErrorTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, string) { +func (ErrorTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, int) { log.Debugf("ErrorTarget triggered.") - return RuleDrop, "" + return RuleDrop, 0 } // UserChainTarget marks a rule as the beginning of a user chain. @@ -53,7 +51,7 @@ type UserChainTarget struct { } // Action implements Target.Action. -func (UserChainTarget) Action(tcpip.PacketBuffer) (RuleVerdict, string) { +func (UserChainTarget) Action(tcpip.PacketBuffer) (RuleVerdict, int) { panic("UserChainTarget should never be called.") } @@ -62,6 +60,6 @@ func (UserChainTarget) Action(tcpip.PacketBuffer) (RuleVerdict, string) { type ReturnTarget struct{} // Action implements Target.Action. -func (ReturnTarget) Action(tcpip.PacketBuffer) (RuleVerdict, string) { - return RuleReturn, "" +func (ReturnTarget) Action(tcpip.PacketBuffer) (RuleVerdict, int) { + return RuleReturn, 0 } diff --git a/pkg/tcpip/iptables/types.go b/pkg/tcpip/iptables/types.go index 5735d001b..7d032fd23 100644 --- a/pkg/tcpip/iptables/types.go +++ b/pkg/tcpip/iptables/types.go @@ -56,17 +56,6 @@ const ( NumHooks ) -// A TableVerdict is what a table decides should be done with a packet. -type TableVerdict int - -const ( - // TableAccept indicates the packet should continue through netstack. - TableAccept TableVerdict = iota - - // TableAccept indicates the packet should be dropped. - TableDrop -) - // A RuleVerdict is what a rule decides should be done with a packet. type RuleVerdict int @@ -74,12 +63,12 @@ const ( // RuleAccept indicates the packet should continue through netstack. RuleAccept RuleVerdict = iota - // RuleContinue indicates the packet should continue to the next rule. - RuleContinue - // RuleDrop indicates the packet should be dropped. RuleDrop + // RuleJump indicates the packet should jump to another chain. + RuleJump + // RuleReturn indicates the packet should return to the previous chain. RuleReturn ) @@ -174,6 +163,6 @@ type Matcher interface { type Target interface { // Action takes an action on the packet and returns a verdict on how // traversal should (or should not) continue. If the return value is - // Jump, it also returns the name of the chain to jump to. - Action(packet tcpip.PacketBuffer) (RuleVerdict, string) + // Jump, it also returns the index of the rule to jump to. + Action(packet tcpip.PacketBuffer) (RuleVerdict, int) } diff --git a/test/iptables/filter_input.go b/test/iptables/filter_input.go index e26d6a7d2..706c09cea 100644 --- a/test/iptables/filter_input.go +++ b/test/iptables/filter_input.go @@ -26,6 +26,7 @@ const ( acceptPort = 2402 sendloopDuration = 2 * time.Second network = "udp4" + chainName = "foochain" ) func init() { @@ -40,6 +41,12 @@ func init() { RegisterTestCase(FilterInputDefaultPolicyAccept{}) RegisterTestCase(FilterInputDefaultPolicyDrop{}) RegisterTestCase(FilterInputReturnUnderflow{}) + RegisterTestCase(FilterInputSerializeJump{}) + RegisterTestCase(FilterInputJumpBasic{}) + RegisterTestCase(FilterInputJumpReturn{}) + RegisterTestCase(FilterInputJumpReturnDrop{}) + RegisterTestCase(FilterInputJumpBuiltin{}) + RegisterTestCase(FilterInputJumpTwice{}) } // FilterInputDropUDP tests that we can drop UDP traffic. @@ -267,13 +274,12 @@ func (FilterInputMultiUDPRules) Name() string { // ContainerAction implements TestCase.ContainerAction. func (FilterInputMultiUDPRules) ContainerAction(ip net.IP) error { - if err := filterTable("-A", "INPUT", "-p", "udp", "-m", "udp", "--destination-port", fmt.Sprintf("%d", dropPort), "-j", "DROP"); err != nil { - return err - } - if err := filterTable("-A", "INPUT", "-p", "udp", "-m", "udp", "--destination-port", fmt.Sprintf("%d", acceptPort), "-j", "ACCEPT"); err != nil { - return err + rules := [][]string{ + {"-A", "INPUT", "-p", "udp", "-m", "udp", "--destination-port", fmt.Sprintf("%d", dropPort), "-j", "DROP"}, + {"-A", "INPUT", "-p", "udp", "-m", "udp", "--destination-port", fmt.Sprintf("%d", acceptPort), "-j", "ACCEPT"}, + {"-L"}, } - return filterTable("-L") + return filterTableRules(rules) } // LocalAction implements TestCase.LocalAction. @@ -314,14 +320,13 @@ func (FilterInputCreateUserChain) Name() string { // ContainerAction implements TestCase.ContainerAction. func (FilterInputCreateUserChain) ContainerAction(ip net.IP) error { - // Create a chain. - const chainName = "foochain" - if err := filterTable("-N", chainName); err != nil { - return err + rules := [][]string{ + // Create a chain. + {"-N", chainName}, + // Add a simple rule to the chain. + {"-A", chainName, "-j", "DROP"}, } - - // Add a simple rule to the chain. - return filterTable("-A", chainName, "-j", "DROP") + return filterTableRules(rules) } // LocalAction implements TestCase.LocalAction. @@ -396,13 +401,12 @@ func (FilterInputReturnUnderflow) Name() string { func (FilterInputReturnUnderflow) ContainerAction(ip net.IP) error { // Add a RETURN rule followed by an unconditional accept, and set the // default policy to DROP. - if err := filterTable("-A", "INPUT", "-j", "RETURN"); err != nil { - return err + rules := [][]string{ + {"-A", "INPUT", "-j", "RETURN"}, + {"-A", "INPUT", "-j", "DROP"}, + {"-P", "INPUT", "ACCEPT"}, } - if err := filterTable("-A", "INPUT", "-j", "DROP"); err != nil { - return err - } - if err := filterTable("-P", "INPUT", "ACCEPT"); err != nil { + if err := filterTableRules(rules); err != nil { return err } @@ -415,3 +419,178 @@ func (FilterInputReturnUnderflow) ContainerAction(ip net.IP) error { func (FilterInputReturnUnderflow) LocalAction(ip net.IP) error { return sendUDPLoop(ip, acceptPort, sendloopDuration) } + +// Verify that we can serialize jumps. +type FilterInputSerializeJump struct{} + +// Name implements TestCase.Name. +func (FilterInputSerializeJump) Name() string { + return "FilterInputSerializeJump" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterInputSerializeJump) ContainerAction(ip net.IP) error { + // Write a JUMP rule, the serialize it with `-L`. + rules := [][]string{ + {"-N", chainName}, + {"-A", "INPUT", "-j", chainName}, + {"-L"}, + } + return filterTableRules(rules) +} + +// LocalAction implements TestCase.LocalAction. +func (FilterInputSerializeJump) LocalAction(ip net.IP) error { + // No-op. + return nil +} + +// Jump to a chain and execute a rule there. +type FilterInputJumpBasic struct{} + +// Name implements TestCase.Name. +func (FilterInputJumpBasic) Name() string { + return "FilterInputJumpBasic" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterInputJumpBasic) ContainerAction(ip net.IP) error { + rules := [][]string{ + {"-P", "INPUT", "DROP"}, + {"-N", chainName}, + {"-A", "INPUT", "-j", chainName}, + {"-A", chainName, "-j", "ACCEPT"}, + } + if err := filterTableRules(rules); err != nil { + return err + } + + // Listen for UDP packets on acceptPort. + return listenUDP(acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (FilterInputJumpBasic) LocalAction(ip net.IP) error { + return sendUDPLoop(ip, acceptPort, sendloopDuration) +} + +// Jump, return, and execute a rule. +type FilterInputJumpReturn struct{} + +// Name implements TestCase.Name. +func (FilterInputJumpReturn) Name() string { + return "FilterInputJumpReturn" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterInputJumpReturn) ContainerAction(ip net.IP) error { + rules := [][]string{ + {"-N", chainName}, + {"-P", "INPUT", "ACCEPT"}, + {"-A", "INPUT", "-j", chainName}, + {"-A", chainName, "-j", "RETURN"}, + {"-A", chainName, "-j", "DROP"}, + } + if err := filterTableRules(rules); err != nil { + return err + } + + // Listen for UDP packets on acceptPort. + return listenUDP(acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (FilterInputJumpReturn) LocalAction(ip net.IP) error { + return sendUDPLoop(ip, acceptPort, sendloopDuration) +} + +type FilterInputJumpReturnDrop struct{} + +// Name implements TestCase.Name. +func (FilterInputJumpReturnDrop) Name() string { + return "FilterInputJumpReturnDrop" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterInputJumpReturnDrop) ContainerAction(ip net.IP) error { + rules := [][]string{ + {"-N", chainName}, + {"-A", "INPUT", "-j", chainName}, + {"-A", "INPUT", "-j", "DROP"}, + {"-A", chainName, "-j", "RETURN"}, + } + if err := filterTableRules(rules); err != nil { + return err + } + + // Listen for UDP packets on dropPort. + if err := listenUDP(dropPort, sendloopDuration); err == nil { + return fmt.Errorf("packets on port %d should have been dropped, but got a packet", dropPort) + } else if netErr, ok := err.(net.Error); !ok || !netErr.Timeout() { + return fmt.Errorf("error reading: %v", err) + } + + // At this point we know that reading timed out and never received a + // packet. + return nil +} + +// LocalAction implements TestCase.LocalAction. +func (FilterInputJumpReturnDrop) LocalAction(ip net.IP) error { + return sendUDPLoop(ip, dropPort, sendloopDuration) +} + +// Jumping to a top-levl chain is illegal. +type FilterInputJumpBuiltin struct{} + +// Name implements TestCase.Name. +func (FilterInputJumpBuiltin) Name() string { + return "FilterInputJumpBuiltin" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterInputJumpBuiltin) ContainerAction(ip net.IP) error { + if err := filterTable("-A", "INPUT", "-j", "OUTPUT"); err == nil { + return fmt.Errorf("iptables should be unable to jump to a built-in chain") + } + return nil +} + +// LocalAction implements TestCase.LocalAction. +func (FilterInputJumpBuiltin) LocalAction(ip net.IP) error { + // No-op. + return nil +} + +// Jump twice, then return twice and execute a rule. +type FilterInputJumpTwice struct{} + +// Name implements TestCase.Name. +func (FilterInputJumpTwice) Name() string { + return "FilterInputJumpTwice" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterInputJumpTwice) ContainerAction(ip net.IP) error { + const chainName2 = chainName + "2" + rules := [][]string{ + {"-P", "INPUT", "DROP"}, + {"-N", chainName}, + {"-N", chainName2}, + {"-A", "INPUT", "-j", chainName}, + {"-A", chainName, "-j", chainName2}, + {"-A", "INPUT", "-j", "ACCEPT"}, + } + if err := filterTableRules(rules); err != nil { + return err + } + + // UDP packets should jump and return twice, eventually hitting the + // ACCEPT rule. + return listenUDP(acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (FilterInputJumpTwice) LocalAction(ip net.IP) error { + return sendUDPLoop(ip, acceptPort, sendloopDuration) +} diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go index 46a7c99b0..0621861eb 100644 --- a/test/iptables/iptables_test.go +++ b/test/iptables/iptables_test.go @@ -249,3 +249,39 @@ func TestFilterOutputDropTCPSrcPort(t *testing.T) { t.Fatal(err) } } + +func TestJumpSerialize(t *testing.T) { + if err := singleTest(FilterInputSerializeJump{}); err != nil { + t.Fatal(err) + } +} + +func TestJumpBasic(t *testing.T) { + if err := singleTest(FilterInputJumpBasic{}); err != nil { + t.Fatal(err) + } +} + +func TestJumpReturn(t *testing.T) { + if err := singleTest(FilterInputJumpReturn{}); err != nil { + t.Fatal(err) + } +} + +func TestJumpReturnDrop(t *testing.T) { + if err := singleTest(FilterInputJumpReturnDrop{}); err != nil { + t.Fatal(err) + } +} + +func TestJumpBuiltin(t *testing.T) { + if err := singleTest(FilterInputJumpBuiltin{}); err != nil { + t.Fatal(err) + } +} + +func TestJumpTwice(t *testing.T) { + if err := singleTest(FilterInputJumpTwice{}); err != nil { + t.Fatal(err) + } +} diff --git a/test/iptables/iptables_util.go b/test/iptables/iptables_util.go index 043114c78..293c4e6ed 100644 --- a/test/iptables/iptables_util.go +++ b/test/iptables/iptables_util.go @@ -35,6 +35,16 @@ func filterTable(args ...string) error { return nil } +// filterTableRules is like filterTable, but runs multiple iptables commands. +func filterTableRules(argsList [][]string) error { + for _, args := range argsList { + if err := filterTable(args...); err != nil { + return err + } + } + return nil +} + // listenUDP listens on a UDP port and returns the value of net.Conn.Read() for // the first read on that port. func listenUDP(port int, timeout time.Duration) error { -- cgit v1.2.3 From a6024f7f5f6f438c11e30be0f93657b1956fd5ba Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Thu, 13 Feb 2020 17:56:34 -0800 Subject: Add FileExec flag to OpenOptions This allow callers to say whether the file is being opened to be executed, so that the proper checks can be done from FilesystemImpl.OpenAt() Updates #1623 PiperOrigin-RevId: 295042595 --- pkg/sentry/fsimpl/ext/filesystem.go | 2 +- pkg/sentry/fsimpl/ext/inode.go | 12 ++++++------ pkg/sentry/fsimpl/gofer/filesystem.go | 24 ++++++++++++------------ pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go | 4 ++-- pkg/sentry/fsimpl/kernfs/fd_impl_util.go | 6 +++--- pkg/sentry/fsimpl/kernfs/filesystem.go | 10 +++++----- pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 6 +++--- pkg/sentry/fsimpl/kernfs/kernfs.go | 2 +- pkg/sentry/fsimpl/kernfs/kernfs_test.go | 8 ++++---- pkg/sentry/fsimpl/proc/subtasks.go | 4 ++-- pkg/sentry/fsimpl/proc/task.go | 4 ++-- pkg/sentry/fsimpl/proc/tasks.go | 4 ++-- pkg/sentry/fsimpl/sys/sys.go | 4 ++-- pkg/sentry/fsimpl/tmpfs/filesystem.go | 2 +- pkg/sentry/vfs/options.go | 5 +++++ pkg/sentry/vfs/permissions.go | 19 ++++++++++++------- pkg/sentry/vfs/vfs.go | 19 +++++++++++++++++++ 17 files changed, 82 insertions(+), 53 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/ext/filesystem.go b/pkg/sentry/fsimpl/ext/filesystem.go index 07bf58953..e05429d41 100644 --- a/pkg/sentry/fsimpl/ext/filesystem.go +++ b/pkg/sentry/fsimpl/ext/filesystem.go @@ -296,7 +296,7 @@ func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf if vfs.MayWriteFileWithOpenFlags(opts.Flags) || opts.Flags&(linux.O_CREAT|linux.O_EXCL|linux.O_TMPFILE) != 0 { return nil, syserror.EROFS } - return inode.open(rp, vfsd, opts.Flags) + return inode.open(rp, vfsd, &opts) } // ReadlinkAt implements vfs.FilesystemImpl.ReadlinkAt. diff --git a/pkg/sentry/fsimpl/ext/inode.go b/pkg/sentry/fsimpl/ext/inode.go index 191b39970..6962083f5 100644 --- a/pkg/sentry/fsimpl/ext/inode.go +++ b/pkg/sentry/fsimpl/ext/inode.go @@ -148,8 +148,8 @@ func newInode(fs *filesystem, inodeNum uint32) (*inode, error) { } // open creates and returns a file description for the dentry passed in. -func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { - ats := vfs.AccessTypesForOpenFlags(flags) +func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts *vfs.OpenOptions) (*vfs.FileDescription, error) { + ats := vfs.AccessTypesForOpenFlags(opts) if err := in.checkPermissions(rp.Credentials(), ats); err != nil { return nil, err } @@ -157,7 +157,7 @@ func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*v switch in.impl.(type) { case *regularFile: var fd regularFileFD - if err := fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{}); err != nil { + if err := fd.vfsfd.Init(&fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{}); err != nil { return nil, err } return &fd.vfsfd, nil @@ -168,17 +168,17 @@ func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*v return nil, syserror.EISDIR } var fd directoryFD - if err := fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{}); err != nil { + if err := fd.vfsfd.Init(&fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{}); err != nil { return nil, err } return &fd.vfsfd, nil case *symlink: - if flags&linux.O_PATH == 0 { + if opts.Flags&linux.O_PATH == 0 { // Can't open symlinks without O_PATH. return nil, syserror.ELOOP } var fd symlinkFD - fd.vfsfd.Init(&fd, flags, mnt, vfsd, &vfs.FileDescriptionOptions{}) + fd.vfsfd.Init(&fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{}) return &fd.vfsfd, nil default: panic(fmt.Sprintf("unknown inode type: %T", in.impl)) diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 8eb61debf..138adb9f7 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -593,7 +593,7 @@ func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf } } if rp.Done() { - return start.openLocked(ctx, rp, opts.Flags) + return start.openLocked(ctx, rp, &opts) } afterTrailingSymlink: @@ -633,12 +633,12 @@ afterTrailingSymlink: start = parent goto afterTrailingSymlink } - return child.openLocked(ctx, rp, opts.Flags) + return child.openLocked(ctx, rp, &opts) } // Preconditions: fs.renameMu must be locked. -func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, flags uint32) (*vfs.FileDescription, error) { - ats := vfs.AccessTypesForOpenFlags(flags) +func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.OpenOptions) (*vfs.FileDescription, error) { + ats := vfs.AccessTypesForOpenFlags(opts) if err := d.checkPermissions(rp.Credentials(), ats, d.isDir()); err != nil { return nil, err } @@ -646,11 +646,11 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, flags ui filetype := d.fileType() switch { case filetype == linux.S_IFREG && !d.fs.opts.regularFilesUseSpecialFileFD: - if err := d.ensureSharedHandle(ctx, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, flags&linux.O_TRUNC != 0); err != nil { + if err := d.ensureSharedHandle(ctx, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, opts.Flags&linux.O_TRUNC != 0); err != nil { return nil, err } fd := ®ularFileFD{} - if err := fd.vfsfd.Init(fd, flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{ + if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{ AllowDirectIO: true, }); err != nil { return nil, err @@ -658,21 +658,21 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, flags ui return &fd.vfsfd, nil case filetype == linux.S_IFDIR: // Can't open directories with O_CREAT. - if flags&linux.O_CREAT != 0 { + if opts.Flags&linux.O_CREAT != 0 { return nil, syserror.EISDIR } // Can't open directories writably. if ats&vfs.MayWrite != 0 { return nil, syserror.EISDIR } - if flags&linux.O_DIRECT != 0 { + if opts.Flags&linux.O_DIRECT != 0 { return nil, syserror.EINVAL } if err := d.ensureSharedHandle(ctx, ats&vfs.MayRead != 0, false /* write */, false /* trunc */); err != nil { return nil, err } fd := &directoryFD{} - if err := fd.vfsfd.Init(fd, flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { + if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { return nil, err } return &fd.vfsfd, nil @@ -680,17 +680,17 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, flags ui // Can't open symlinks without O_PATH (which is unimplemented). return nil, syserror.ELOOP default: - if flags&linux.O_DIRECT != 0 { + if opts.Flags&linux.O_DIRECT != 0 { return nil, syserror.EINVAL } - h, err := openHandle(ctx, d.file, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, flags&linux.O_TRUNC != 0) + h, err := openHandle(ctx, d.file, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, opts.Flags&linux.O_TRUNC != 0) if err != nil { return nil, err } fd := &specialFileFD{ handle: h, } - if err := fd.vfsfd.Init(fd, flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { + if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { h.close(ctx) return nil, err } diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go index 733792c78..d092ccb2a 100644 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -53,9 +53,9 @@ func (f *DynamicBytesFile) Init(creds *auth.Credentials, ino uint64, data vfs.Dy } // Open implements Inode.Open. -func (f *DynamicBytesFile) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { +func (f *DynamicBytesFile) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd := &DynamicBytesFD{} - if err := fd.Init(rp.Mount(), vfsd, f.data, flags); err != nil { + if err := fd.Init(rp.Mount(), vfsd, f.data, opts.Flags); err != nil { return nil, err } return &fd.vfsfd, nil diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index 6104751c8..eda781155 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -43,12 +43,12 @@ type GenericDirectoryFD struct { } // Init initializes a GenericDirectoryFD. -func (fd *GenericDirectoryFD) Init(m *vfs.Mount, d *vfs.Dentry, children *OrderedChildren, flags uint32) error { - if vfs.AccessTypesForOpenFlags(flags)&vfs.MayWrite != 0 { +func (fd *GenericDirectoryFD) Init(m *vfs.Mount, d *vfs.Dentry, children *OrderedChildren, opts *vfs.OpenOptions) error { + if vfs.AccessTypesForOpenFlags(opts)&vfs.MayWrite != 0 { // Can't open directories for writing. return syserror.EISDIR } - if err := fd.vfsfd.Init(fd, flags, m, d, &vfs.FileDescriptionOptions{}); err != nil { + if err := fd.vfsfd.Init(fd, opts.Flags, m, d, &vfs.FileDescriptionOptions{}); err != nil { return err } fd.children = children diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index e49303c26..ee98eb66a 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -365,7 +365,7 @@ func (fs *Filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf // appropriate bits in rp), but are returned by // FileDescriptionImpl.StatusFlags(). opts.Flags &= linux.O_ACCMODE | linux.O_CREAT | linux.O_EXCL | linux.O_TRUNC | linux.O_DIRECTORY | linux.O_NOFOLLOW - ats := vfs.AccessTypesForOpenFlags(opts.Flags) + ats := vfs.AccessTypesForOpenFlags(&opts) // Do not create new file. if opts.Flags&linux.O_CREAT == 0 { @@ -379,7 +379,7 @@ func (fs *Filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf if err := inode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { return nil, err } - return inode.Open(rp, vfsd, opts.Flags) + return inode.Open(rp, vfsd, opts) } // May create new file. @@ -398,7 +398,7 @@ func (fs *Filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf if err := inode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { return nil, err } - return inode.Open(rp, vfsd, opts.Flags) + return inode.Open(rp, vfsd, opts) } afterTrailingSymlink: parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) @@ -438,7 +438,7 @@ afterTrailingSymlink: return nil, err } parentVFSD.Impl().(*Dentry).InsertChild(pc, child) - return child.Impl().(*Dentry).inode.Open(rp, child, opts.Flags) + return child.Impl().(*Dentry).inode.Open(rp, child, opts) } // Open existing file or follow symlink. if mustCreate { @@ -463,7 +463,7 @@ afterTrailingSymlink: if err := childInode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { return nil, err } - return childInode.Open(rp, childVFSD, opts.Flags) + return childInode.Open(rp, childVFSD, opts) } // ReadlinkAt implements vfs.FilesystemImpl.ReadlinkAt. diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index adca2313f..099d70a16 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -507,7 +507,7 @@ type InodeSymlink struct { } // Open implements Inode.Open. -func (InodeSymlink) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { +func (InodeSymlink) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { return nil, syserror.ELOOP } @@ -549,8 +549,8 @@ func (s *StaticDirectory) Init(creds *auth.Credentials, ino uint64, perm linux.F } // Open implements kernfs.Inode. -func (s *StaticDirectory) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { +func (s *StaticDirectory) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd := &GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &s.OrderedChildren, flags) + fd.Init(rp.Mount(), vfsd, &s.OrderedChildren, &opts) return fd.VFSFileDescription(), nil } diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index 79ebea8a5..c74fa999b 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -303,7 +303,7 @@ type Inode interface { // inode for its lifetime. // // Precondition: !rp.Done(). vfsd.Impl() must be a kernfs Dentry. - Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) + Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) } type inodeRefs interface { diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index ee65cf491..96a16e654 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -113,9 +113,9 @@ func (fs *filesystem) newReadonlyDir(creds *auth.Credentials, mode linux.FileMod return &dir.dentry } -func (d *readonlyDir) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { +func (d *readonlyDir) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd := &kernfs.GenericDirectoryFD{} - if err := fd.Init(rp.Mount(), vfsd, &d.OrderedChildren, flags); err != nil { + if err := fd.Init(rp.Mount(), vfsd, &d.OrderedChildren, &opts); err != nil { return nil, err } return fd.VFSFileDescription(), nil @@ -143,9 +143,9 @@ func (fs *filesystem) newDir(creds *auth.Credentials, mode linux.FileMode, conte return &dir.dentry } -func (d *dir) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { +func (d *dir) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &d.OrderedChildren, flags) + fd.Init(rp.Mount(), vfsd, &d.OrderedChildren, &opts) return fd.VFSFileDescription(), nil } diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go index 353e37195..102af0e93 100644 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -114,9 +114,9 @@ func (i *subtasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallb } // Open implements kernfs.Inode. -func (i *subtasksInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { +func (i *subtasksInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, flags) + fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, &opts) return fd.VFSFileDescription(), nil } diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index eb5bc62c0..2d814668a 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -98,9 +98,9 @@ func (i *taskInode) Valid(ctx context.Context) bool { } // Open implements kernfs.Inode. -func (i *taskInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { +func (i *taskInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, flags) + fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, &opts) return fd.VFSFileDescription(), nil } diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index 14bd334e8..ebe21630c 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -205,9 +205,9 @@ func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback } // Open implements kernfs.Inode. -func (i *tasksInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { +func (i *tasksInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, flags) + fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, &opts) return fd.VFSFileDescription(), nil } diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go index e35d52d17..d693fceae 100644 --- a/pkg/sentry/fsimpl/sys/sys.go +++ b/pkg/sentry/fsimpl/sys/sys.go @@ -97,9 +97,9 @@ func (d *dir) SetStat(fs *vfs.Filesystem, opts vfs.SetStatOptions) error { } // Open implements kernfs.Inode.Open. -func (d *dir) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { +func (d *dir) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &d.OrderedChildren, flags) + fd.Init(rp.Mount(), vfsd, &d.OrderedChildren, &opts) return fd.VFSFileDescription(), nil } diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 72bc15264..8785452b6 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -334,7 +334,7 @@ afterTrailingSymlink: } func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.OpenOptions, afterCreate bool) (*vfs.FileDescription, error) { - ats := vfs.AccessTypesForOpenFlags(opts.Flags) + ats := vfs.AccessTypesForOpenFlags(opts) if !afterCreate { if err := d.inode.checkPermissions(rp.Credentials(), ats, d.inode.isDir()); err != nil { return nil, err diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go index b7774bf28..fdf8be157 100644 --- a/pkg/sentry/vfs/options.go +++ b/pkg/sentry/vfs/options.go @@ -72,6 +72,11 @@ type OpenOptions struct { // If FilesystemImpl.OpenAt() creates a file, Mode is the file mode for the // created file. Mode linux.FileMode + + // FileExec is set when the file is being opened to be executed. + // VirtualFilesystem.OpenAt() checks that the caller has execute permissions + // on the file, and that the file is a regular file. + FileExec bool } // ReadOptions contains options to FileDescription.PRead(), diff --git a/pkg/sentry/vfs/permissions.go b/pkg/sentry/vfs/permissions.go index f664581f4..8e250998a 100644 --- a/pkg/sentry/vfs/permissions.go +++ b/pkg/sentry/vfs/permissions.go @@ -103,17 +103,22 @@ func GenericCheckPermissions(creds *auth.Credentials, ats AccessTypes, isDir boo // AccessTypesForOpenFlags returns MayRead|MayWrite in this case. // // Use May{Read,Write}FileWithOpenFlags() for these checks instead. -func AccessTypesForOpenFlags(flags uint32) AccessTypes { - switch flags & linux.O_ACCMODE { +func AccessTypesForOpenFlags(opts *OpenOptions) AccessTypes { + ats := AccessTypes(0) + if opts.FileExec { + ats |= MayExec + } + + switch opts.Flags & linux.O_ACCMODE { case linux.O_RDONLY: - if flags&linux.O_TRUNC != 0 { - return MayRead | MayWrite + if opts.Flags&linux.O_TRUNC != 0 { + return ats | MayRead | MayWrite } - return MayRead + return ats | MayRead case linux.O_WRONLY: - return MayWrite + return ats | MayWrite default: - return MayRead | MayWrite + return ats | MayRead | MayWrite } } diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 908c69f91..9629afee9 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -379,6 +379,25 @@ func (vfs *VirtualFilesystem) OpenAt(ctx context.Context, creds *auth.Credential fd, err := rp.mount.fs.impl.OpenAt(ctx, rp, *opts) if err == nil { vfs.putResolvingPath(rp) + + // TODO(gvisor.dev/issue/1193): Move inside fsimpl to avoid another call + // to FileDescription.Stat(). + if opts.FileExec { + // Only a regular file can be executed. + stat, err := fd.Stat(ctx, StatOptions{Mask: linux.STATX_TYPE}) + if err != nil { + return nil, err + } + if stat.Mask&linux.STATX_TYPE != 0 { + // This shouldn't happen, but if type can't be retrieved, file can't + // be executed. + return nil, syserror.EACCES + } + if linux.FileMode(stat.Mode).FileType() != linux.ModeRegular { + return nil, syserror.EACCES + } + } + return fd, nil } if !rp.handleError(err) { -- cgit v1.2.3 From ebaf29abeb5e6e6cd87f4b0088719abb30e9d4cb Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Fri, 14 Feb 2020 02:04:41 -0500 Subject: passed the kvm test case of "TestKernelSyscall" on Arm64 For kvm test case "TestKernelSyscall", redpill/syscall(-1) in guest kernel level will be trapped in el1_svc. And in el1_svc, we use mmio_exit to leave the guest. Signed-off-by: Bin Lu --- pkg/sentry/platform/ring0/entry_arm64.s | 73 ++++++++++++++++++++++++++----- pkg/sentry/platform/ring0/kernel_arm64.go | 8 ++++ 2 files changed, 70 insertions(+), 11 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/ring0/entry_arm64.s b/pkg/sentry/platform/ring0/entry_arm64.s index baa6c4910..d42eda37b 100644 --- a/pkg/sentry/platform/ring0/entry_arm64.s +++ b/pkg/sentry/platform/ring0/entry_arm64.s @@ -25,10 +25,14 @@ // not available for calls. // +// ERET returns using the ELR and SPSR for the current exception level. #define ERET() \ WORD $0xd69f03e0 +// RSV_REG is a register that holds el1 information temporarily. #define RSV_REG R18_PLATFORM + +// RSV_REG_APP is a register that holds el0 information temporarily. #define RSV_REG_APP R9 #define FPEN_NOTRAP 0x3 @@ -36,6 +40,12 @@ #define FPEN_ENABLE (FPEN_NOTRAP << FPEN_SHIFT) +// Saves a register set. +// +// This is a macro because it may need to executed in contents where a stack is +// not available for calls. +// +// The following registers are not saved: R9, R18. #define REGISTERS_SAVE(reg, offset) \ MOVD R0, offset+PTRACE_R0(reg); \ MOVD R1, offset+PTRACE_R1(reg); \ @@ -67,6 +77,12 @@ MOVD R29, offset+PTRACE_R29(reg); \ MOVD R30, offset+PTRACE_R30(reg); +// Loads a register set. +// +// This is a macro because it may need to executed in contents where a stack is +// not available for calls. +// +// The following registers are not loaded: R9, R18. #define REGISTERS_LOAD(reg, offset) \ MOVD offset+PTRACE_R0(reg), R0; \ MOVD offset+PTRACE_R1(reg), R1; \ @@ -98,7 +114,7 @@ MOVD offset+PTRACE_R29(reg), R29; \ MOVD offset+PTRACE_R30(reg), R30; -//NOP +// NOP-s #define nop31Instructions() \ WORD $0xd503201f; \ WORD $0xd503201f; \ @@ -254,6 +270,7 @@ #define ESR_ELx_WFx_ISS_WFE (UL(1) << 0) #define ESR_ELx_xVC_IMM_MASK ((1UL << 16) - 1) +// LOAD_KERNEL_ADDRESS loads a kernel address. #define LOAD_KERNEL_ADDRESS(from, to) \ MOVD from, to; \ ORR $0xffff000000000000, to, to; @@ -263,15 +280,18 @@ LOAD_KERNEL_ADDRESS(CPU_SELF(from), RSV_REG); \ MOVD $CPU_STACK_TOP(RSV_REG), RSV_REG; \ MOVD RSV_REG, RSP; \ + WORD $0xd538d092; \ //MRS TPIDR_EL1, R18 ISB $15; \ DSB $15; +// SWITCH_TO_APP_PAGETABLE sets a new pagetable for a container application. #define SWITCH_TO_APP_PAGETABLE(from) \ MOVD CPU_TTBR0_APP(from), RSV_REG; \ WORD $0xd5182012; \ // MSR R18, TTBR0_EL1 ISB $15; \ DSB $15; +// SWITCH_TO_KVM_PAGETABLE sets the kvm pagetable. #define SWITCH_TO_KVM_PAGETABLE(from) \ MOVD CPU_TTBR0_KVM(from), RSV_REG; \ WORD $0xd5182012; \ // MSR R18, TTBR0_EL1 @@ -294,6 +314,7 @@ WORD $0xd5181040; \ //MSR R0, CPACR_EL1 ISB $15; +// KERNEL_ENTRY_FROM_EL0 is the entry code of the vcpu from el0 to el1. #define KERNEL_ENTRY_FROM_EL0 \ SUB $16, RSP, RSP; \ // step1, save r18, r9 into kernel temporary stack. STP (RSV_REG, RSV_REG_APP), 16*0(RSP); \ @@ -315,19 +336,22 @@ WORD $0xd5384103; \ // MRS SP_EL0, R3 MOVD R3, PTRACE_SP(RSV_REG_APP); +// KERNEL_ENTRY_FROM_EL1 is the entry code of the vcpu from el1 to el1. #define KERNEL_ENTRY_FROM_EL1 \ WORD $0xd538d092; \ //MRS TPIDR_EL1, R18 - REGISTERS_SAVE(RSV_REG, CPU_REGISTERS); \ // save sentry context + REGISTERS_SAVE(RSV_REG, CPU_REGISTERS); \ // Save sentry context. MOVD RSV_REG_APP, CPU_REGISTERS+PTRACE_R9(RSV_REG); \ WORD $0xd5384004; \ // MRS SPSR_EL1, R4 MOVD R4, CPU_REGISTERS+PTRACE_PSTATE(RSV_REG); \ MRS ELR_EL1, R4; \ MOVD R4, CPU_REGISTERS+PTRACE_PC(RSV_REG); \ MOVD RSP, R4; \ - MOVD R4, CPU_REGISTERS+PTRACE_SP(RSV_REG); + MOVD R4, CPU_REGISTERS+PTRACE_SP(RSV_REG); \ + LOAD_KERNEL_STACK(RSV_REG); // Load the temporary stack. +// Halt halts execution. TEXT ·Halt(SB),NOSPLIT,$0 - // clear bluepill. + // Clear bluepill. WORD $0xd538d092 //MRS TPIDR_EL1, R18 CMP RSV_REG, R9 BNE mmio_exit @@ -341,8 +365,22 @@ mmio_exit: // MMIO_EXIT. MOVD $0, R9 MOVD R0, 0xffff000000001000(R9) - B ·kernelExitToEl1(SB) + RET + +// HaltAndResume halts execution and point the pointer to the resume function. +TEXT ·HaltAndResume(SB),NOSPLIT,$0 + BL ·Halt(SB) + B ·kernelExitToEl1(SB) // Resume. +// HaltEl1SvcAndResume calls Hooks.KernelSyscall and resume. +TEXT ·HaltEl1SvcAndResume(SB),NOSPLIT,$0 + WORD $0xd538d092 // MRS TPIDR_EL1, R18 + MOVD CPU_SELF(RSV_REG), R3 // Load vCPU. + MOVD R3, 8(RSP) // First argument (vCPU). + CALL ·kernelSyscall(SB) // Call the trampoline. + B ·kernelExitToEl1(SB) // Resume. + +// Shutdown stops the guest. TEXT ·Shutdown(SB),NOSPLIT,$0 // PSCI EVENT. MOVD $0x84000009, R0 @@ -429,6 +467,7 @@ TEXT ·kernelExitToEl0(SB),NOSPLIT,$0 TEXT ·kernelExitToEl1(SB),NOSPLIT,$0 ERET() +// Start is the CPU entrypoint. TEXT ·Start(SB),NOSPLIT,$0 IRQ_DISABLE MOVD R8, RSV_REG @@ -437,18 +476,23 @@ TEXT ·Start(SB),NOSPLIT,$0 B ·kernelExitToEl1(SB) +// El1_sync_invalid is the handler for an invalid EL1_sync. TEXT ·El1_sync_invalid(SB),NOSPLIT,$0 B ·Shutdown(SB) +// El1_irq_invalid is the handler for an invalid El1_irq. TEXT ·El1_irq_invalid(SB),NOSPLIT,$0 B ·Shutdown(SB) +// El1_fiq_invalid is the handler for an invalid El1_fiq. TEXT ·El1_fiq_invalid(SB),NOSPLIT,$0 B ·Shutdown(SB) +// El1_error_invalid is the handler for an invalid El1_error. TEXT ·El1_error_invalid(SB),NOSPLIT,$0 B ·Shutdown(SB) +// El1_sync is the handler for El1_sync. TEXT ·El1_sync(SB),NOSPLIT,$0 KERNEL_ENTRY_FROM_EL1 WORD $0xd5385219 // MRS ESR_EL1, R25 @@ -484,10 +528,10 @@ el1_da: MOVD $PageFault, R3 MOVD R3, CPU_VECTOR_CODE(RSV_REG) - B ·Halt(SB) + B ·HaltAndResume(SB) el1_ia: - B ·Halt(SB) + B ·HaltAndResume(SB) el1_sp_pc: B ·Shutdown(SB) @@ -496,7 +540,9 @@ el1_undef: B ·Shutdown(SB) el1_svc: - B ·Halt(SB) + MOVD $0, CPU_ERROR_CODE(RSV_REG) + MOVD $0, CPU_ERROR_TYPE(RSV_REG) + B ·HaltEl1SvcAndResume(SB) el1_dbg: B ·Shutdown(SB) @@ -508,15 +554,19 @@ el1_fpsimd_acc: el1_invalid: B ·Shutdown(SB) +// El1_irq is the handler for El1_irq. TEXT ·El1_irq(SB),NOSPLIT,$0 B ·Shutdown(SB) +// El1_fiq is the handler for El1_fiq. TEXT ·El1_fiq(SB),NOSPLIT,$0 B ·Shutdown(SB) +// El1_error is the handler for El1_error. TEXT ·El1_error(SB),NOSPLIT,$0 B ·Shutdown(SB) +// El0_sync is the handler for El0_sync. TEXT ·El0_sync(SB),NOSPLIT,$0 KERNEL_ENTRY_FROM_EL0 WORD $0xd5385219 // MRS ESR_EL1, R25 @@ -554,7 +604,7 @@ el0_svc: MOVD $Syscall, R3 MOVD R3, CPU_VECTOR_CODE(RSV_REG) - B ·Halt(SB) + B ·HaltAndResume(SB) el0_da: WORD $0xd538d092 //MRS TPIDR_EL1, R18 @@ -568,7 +618,7 @@ el0_da: MOVD $PageFault, R3 MOVD R3, CPU_VECTOR_CODE(RSV_REG) - B ·Halt(SB) + B ·HaltAndResume(SB) el0_ia: B ·Shutdown(SB) @@ -613,7 +663,7 @@ TEXT ·El0_error(SB),NOSPLIT,$0 MOVD $VirtualizationException, R3 MOVD R3, CPU_VECTOR_CODE(RSV_REG) - B ·Halt(SB) + B ·HaltAndResume(SB) TEXT ·El0_sync_invalid(SB),NOSPLIT,$0 B ·Shutdown(SB) @@ -627,6 +677,7 @@ TEXT ·El0_fiq_invalid(SB),NOSPLIT,$0 TEXT ·El0_error_invalid(SB),NOSPLIT,$0 B ·Shutdown(SB) +// Vectors implements exception vector table. TEXT ·Vectors(SB),NOSPLIT,$0 B ·El1_sync_invalid(SB) nop31Instructions() diff --git a/pkg/sentry/platform/ring0/kernel_arm64.go b/pkg/sentry/platform/ring0/kernel_arm64.go index c3d341998..ccacaea6b 100644 --- a/pkg/sentry/platform/ring0/kernel_arm64.go +++ b/pkg/sentry/platform/ring0/kernel_arm64.go @@ -16,6 +16,14 @@ package ring0 +// HaltAndResume halts execution and point the pointer to the resume function. +//go:nosplit +func HaltAndResume() + +// HaltEl1SvcAndResume calls Hooks.KernelSyscall and resume. +//go:nosplit +func HaltEl1SvcAndResume() + // init initializes architecture-specific state. func (k *Kernel) init(opts KernelOpts) { // Save the root page tables. -- cgit v1.2.3 From 4075de11be44372c454aae7f9650cdc814c52229 Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Fri, 14 Feb 2020 11:11:55 -0800 Subject: Plumb VFS2 inside the Sentry - Added fsbridge package with interface that can be used to open and read from VFS1 and VFS2 files. - Converted ELF loader to use fsbridge - Added VFS2 types to FSContext - Added vfs.MountNamespace to ThreadGroup Updates #1623 PiperOrigin-RevId: 295183950 --- pkg/sentry/control/BUILD | 5 + pkg/sentry/control/proc.go | 127 +++++++++++++-- pkg/sentry/fs/proc/BUILD | 1 + pkg/sentry/fs/proc/task.go | 17 +- pkg/sentry/fsbridge/BUILD | 24 +++ pkg/sentry/fsbridge/bridge.go | 54 ++++++ pkg/sentry/fsbridge/fs.go | 181 +++++++++++++++++++++ pkg/sentry/fsbridge/vfs.go | 134 +++++++++++++++ pkg/sentry/fsimpl/devtmpfs/devtmpfs.go | 4 + pkg/sentry/fsimpl/gofer/filesystem.go | 5 +- pkg/sentry/fsimpl/gofer/gofer.go | 3 + pkg/sentry/fsimpl/kernfs/filesystem.go | 10 +- pkg/sentry/fsimpl/proc/BUILD | 1 + pkg/sentry/fsimpl/proc/filesystem.go | 18 +- pkg/sentry/fsimpl/proc/tasks_test.go | 17 +- pkg/sentry/fsimpl/sys/BUILD | 1 + pkg/sentry/fsimpl/sys/sys.go | 3 + pkg/sentry/fsimpl/sys/sys_test.go | 7 +- pkg/sentry/fsimpl/testutil/BUILD | 2 +- pkg/sentry/fsimpl/testutil/kernel.go | 24 +-- pkg/sentry/fsimpl/testutil/testutil.go | 12 +- pkg/sentry/fsimpl/tmpfs/filesystem.go | 12 +- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 3 + pkg/sentry/kernel/BUILD | 2 + pkg/sentry/kernel/fs_context.go | 98 +++++++++-- pkg/sentry/kernel/kernel.go | 145 +++++++++++++---- pkg/sentry/kernel/task.go | 27 +++ pkg/sentry/kernel/task_clone.go | 11 +- pkg/sentry/kernel/task_context.go | 2 +- pkg/sentry/kernel/task_exit.go | 7 + pkg/sentry/kernel/task_log.go | 15 +- pkg/sentry/kernel/task_start.go | 49 +++--- pkg/sentry/kernel/thread_group.go | 6 +- pkg/sentry/loader/BUILD | 2 + pkg/sentry/loader/elf.go | 28 ++-- pkg/sentry/loader/interpreter.go | 6 +- pkg/sentry/loader/loader.go | 179 ++++++-------------- pkg/sentry/loader/vdso.go | 7 +- pkg/sentry/mm/BUILD | 2 +- pkg/sentry/mm/metadata.go | 10 +- pkg/sentry/mm/mm.go | 4 +- pkg/sentry/strace/strace.go | 28 ++++ pkg/sentry/syscalls/linux/BUILD | 1 + pkg/sentry/syscalls/linux/sys_prctl.go | 3 +- pkg/sentry/syscalls/linux/sys_thread.go | 17 +- .../syscalls/linux/vfs2/linux64_override_amd64.go | 106 ++++++++++++ pkg/sentry/vfs/BUILD | 1 + pkg/sentry/vfs/context.go | 7 +- pkg/sentry/vfs/mount.go | 10 +- pkg/sentry/vfs/options.go | 2 +- pkg/sentry/vfs/vfs.go | 5 +- runsc/boot/loader.go | 11 +- 52 files changed, 1134 insertions(+), 322 deletions(-) create mode 100644 pkg/sentry/fsbridge/BUILD create mode 100644 pkg/sentry/fsbridge/bridge.go create mode 100644 pkg/sentry/fsbridge/fs.go create mode 100644 pkg/sentry/fsbridge/vfs.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/control/BUILD b/pkg/sentry/control/BUILD index e69496477..d16d78aa5 100644 --- a/pkg/sentry/control/BUILD +++ b/pkg/sentry/control/BUILD @@ -16,10 +16,13 @@ go_library( ], deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/fd", + "//pkg/fspath", "//pkg/log", "//pkg/sentry/fs", "//pkg/sentry/fs/host", + "//pkg/sentry/fsbridge", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/time", @@ -27,8 +30,10 @@ go_library( "//pkg/sentry/state", "//pkg/sentry/strace", "//pkg/sentry/usage", + "//pkg/sentry/vfs", "//pkg/sentry/watchdog", "//pkg/sync", + "//pkg/syserror", "//pkg/tcpip/link/sniffer", "//pkg/urpc", ], diff --git a/pkg/sentry/control/proc.go b/pkg/sentry/control/proc.go index ced51c66c..8973754c8 100644 --- a/pkg/sentry/control/proc.go +++ b/pkg/sentry/control/proc.go @@ -18,19 +18,26 @@ import ( "bytes" "encoding/json" "fmt" + "path" "sort" "strings" "text/tabwriter" "time" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/host" + "gvisor.dev/gvisor/pkg/sentry/fsbridge" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/usage" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/urpc" ) @@ -60,6 +67,12 @@ type ExecArgs struct { // process's MountNamespace. MountNamespace *fs.MountNamespace + // MountNamespaceVFS2 is the mount namespace to execute the new process in. + // A reference on MountNamespace must be held for the lifetime of the + // ExecArgs. If MountNamespace is nil, it will default to the init + // process's MountNamespace. + MountNamespaceVFS2 *vfs.MountNamespace + // WorkingDirectory defines the working directory for the new process. WorkingDirectory string `json:"wd"` @@ -150,6 +163,7 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI Envv: args.Envv, WorkingDirectory: args.WorkingDirectory, MountNamespace: args.MountNamespace, + MountNamespaceVFS2: args.MountNamespaceVFS2, Credentials: creds, FDTable: fdTable, Umask: 0022, @@ -166,24 +180,53 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI // be donated to the new process in CreateProcess. initArgs.MountNamespace.IncRef() } + if initArgs.MountNamespaceVFS2 != nil { + // initArgs must hold a reference on MountNamespaceVFS2, which will + // be donated to the new process in CreateProcess. + initArgs.MountNamespaceVFS2.IncRef() + } ctx := initArgs.NewContext(proc.Kernel) if initArgs.Filename == "" { - // Get the full path to the filename from the PATH env variable. - paths := fs.GetPath(initArgs.Envv) - mns := initArgs.MountNamespace - if mns == nil { - mns = proc.Kernel.GlobalInit().Leader().MountNamespace() - } - f, err := mns.ResolveExecutablePath(ctx, initArgs.WorkingDirectory, initArgs.Argv[0], paths) - if err != nil { - return nil, 0, nil, fmt.Errorf("error finding executable %q in PATH %v: %v", initArgs.Argv[0], paths, err) + if kernel.VFS2Enabled { + // Get the full path to the filename from the PATH env variable. + if initArgs.MountNamespaceVFS2 == nil { + // Set initArgs so that 'ctx' returns the namespace. + // + // MountNamespaceVFS2 adds a reference to the namespace, which is + // transferred to the new process. + initArgs.MountNamespaceVFS2 = proc.Kernel.GlobalInit().Leader().MountNamespaceVFS2() + } + + paths := fs.GetPath(initArgs.Envv) + vfsObj := proc.Kernel.VFS + file, err := ResolveExecutablePath(ctx, vfsObj, initArgs.WorkingDirectory, initArgs.Argv[0], paths) + if err != nil { + return nil, 0, nil, fmt.Errorf("error finding executable %q in PATH %v: %v", initArgs.Argv[0], paths, err) + } + initArgs.File = fsbridge.NewVFSFile(file) + } else { + // Get the full path to the filename from the PATH env variable. + paths := fs.GetPath(initArgs.Envv) + if initArgs.MountNamespace == nil { + // Set initArgs so that 'ctx' returns the namespace. + initArgs.MountNamespace = proc.Kernel.GlobalInit().Leader().MountNamespace() + + // initArgs must hold a reference on MountNamespace, which will + // be donated to the new process in CreateProcess. + initArgs.MountNamespaceVFS2.IncRef() + } + f, err := initArgs.MountNamespace.ResolveExecutablePath(ctx, initArgs.WorkingDirectory, initArgs.Argv[0], paths) + if err != nil { + return nil, 0, nil, fmt.Errorf("error finding executable %q in PATH %v: %v", initArgs.Argv[0], paths, err) + } + initArgs.Filename = f } - initArgs.Filename = f } mounter := fs.FileOwnerFromContext(ctx) + // TODO(gvisor.dev/issue/1623): Use host FD when supported in VFS2. var ttyFile *fs.File for appFD, hostFile := range args.FilePayload.Files { var appFile *fs.File @@ -411,3 +454,67 @@ func ttyName(tty *kernel.TTY) string { } return fmt.Sprintf("pts/%d", tty.Index) } + +// ResolveExecutablePath resolves the given executable name given a set of +// paths that might contain it. +func ResolveExecutablePath(ctx context.Context, vfsObj *vfs.VirtualFilesystem, wd, name string, paths []string) (*vfs.FileDescription, error) { + root := vfs.RootFromContext(ctx) + defer root.DecRef() + creds := auth.CredentialsFromContext(ctx) + + // Absolute paths can be used directly. + if path.IsAbs(name) { + return openExecutable(ctx, vfsObj, creds, root, name) + } + + // Paths with '/' in them should be joined to the working directory, or + // to the root if working directory is not set. + if strings.IndexByte(name, '/') > 0 { + if len(wd) == 0 { + wd = "/" + } + if !path.IsAbs(wd) { + return nil, fmt.Errorf("working directory %q must be absolute", wd) + } + return openExecutable(ctx, vfsObj, creds, root, path.Join(wd, name)) + } + + // Otherwise, we must lookup the name in the paths, starting from the + // calling context's root directory. + for _, p := range paths { + if !path.IsAbs(p) { + // Relative paths aren't safe, no one should be using them. + log.Warningf("Skipping relative path %q in $PATH", p) + continue + } + + binPath := path.Join(p, name) + f, err := openExecutable(ctx, vfsObj, creds, root, binPath) + if err != nil { + return nil, err + } + if f == nil { + continue // Not found/no access. + } + return f, nil + } + return nil, syserror.ENOENT +} + +func openExecutable(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, root vfs.VirtualDentry, path string) (*vfs.FileDescription, error) { + pop := vfs.PathOperation{ + Root: root, + Start: root, // binPath is absolute, Start can be anything. + Path: fspath.Parse(path), + FollowFinalSymlink: true, + } + opts := &vfs.OpenOptions{ + Flags: linux.O_RDONLY, + FileExec: true, + } + f, err := vfsObj.OpenAt(ctx, creds, &pop, opts) + if err == syserror.ENOENT || err == syserror.EACCES { + return nil, nil + } + return f, err +} diff --git a/pkg/sentry/fs/proc/BUILD b/pkg/sentry/fs/proc/BUILD index 280093c5e..77c2c5c0e 100644 --- a/pkg/sentry/fs/proc/BUILD +++ b/pkg/sentry/fs/proc/BUILD @@ -36,6 +36,7 @@ go_library( "//pkg/sentry/fs/proc/device", "//pkg/sentry/fs/proc/seqfile", "//pkg/sentry/fs/ramfs", + "//pkg/sentry/fsbridge", "//pkg/sentry/inet", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go index ca020e11e..8ab8d8a02 100644 --- a/pkg/sentry/fs/proc/task.go +++ b/pkg/sentry/fs/proc/task.go @@ -28,6 +28,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs/proc/device" "gvisor.dev/gvisor/pkg/sentry/fs/proc/seqfile" "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" + "gvisor.dev/gvisor/pkg/sentry/fsbridge" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/mm" @@ -249,7 +250,7 @@ func newExe(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { return newProcInode(t, exeSymlink, msrc, fs.Symlink, t) } -func (e *exe) executable() (d *fs.Dirent, err error) { +func (e *exe) executable() (file fsbridge.File, err error) { e.t.WithMuLocked(func(t *kernel.Task) { mm := t.MemoryManager() if mm == nil { @@ -262,8 +263,8 @@ func (e *exe) executable() (d *fs.Dirent, err error) { // The MemoryManager may be destroyed, in which case // MemoryManager.destroy will simply set the executable to nil // (with locks held). - d = mm.Executable() - if d == nil { + file = mm.Executable() + if file == nil { err = syserror.ENOENT } }) @@ -283,15 +284,7 @@ func (e *exe) Readlink(ctx context.Context, inode *fs.Inode) (string, error) { } defer exec.DecRef() - root := fs.RootFromContext(ctx) - if root == nil { - // This doesn't correspond to anything in Linux because the vfs is - // global there. - return "", syserror.EINVAL - } - defer root.DecRef() - n, _ := exec.FullName(root) - return n, nil + return exec.PathnameWithDeleted(ctx), nil } // namespaceSymlink represents a symlink in the namespacefs, such as the files diff --git a/pkg/sentry/fsbridge/BUILD b/pkg/sentry/fsbridge/BUILD new file mode 100644 index 000000000..6c798f0bd --- /dev/null +++ b/pkg/sentry/fsbridge/BUILD @@ -0,0 +1,24 @@ +load("//tools:defs.bzl", "go_library") + +licenses(["notice"]) + +go_library( + name = "fsbridge", + srcs = [ + "bridge.go", + "fs.go", + "vfs.go", + ], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/abi/linux", + "//pkg/context", + "//pkg/fspath", + "//pkg/sentry/fs", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/memmap", + "//pkg/sentry/vfs", + "//pkg/syserror", + "//pkg/usermem", + ], +) diff --git a/pkg/sentry/fsbridge/bridge.go b/pkg/sentry/fsbridge/bridge.go new file mode 100644 index 000000000..8e7590721 --- /dev/null +++ b/pkg/sentry/fsbridge/bridge.go @@ -0,0 +1,54 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package fsbridge provides common interfaces to bridge between VFS1 and VFS2 +// files. +package fsbridge + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/memmap" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/usermem" +) + +// File provides a common interface to bridge between VFS1 and VFS2 files. +type File interface { + // PathnameWithDeleted returns an absolute pathname to vd, consistent with + // Linux's d_path(). In particular, if vd.Dentry() has been disowned, + // PathnameWithDeleted appends " (deleted)" to the returned pathname. + PathnameWithDeleted(ctx context.Context) string + + // ReadFull read all contents from the file. + ReadFull(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) + + // ConfigureMMap mutates opts to implement mmap(2) for the file. + ConfigureMMap(context.Context, *memmap.MMapOpts) error + + // Type returns the file type, e.g. linux.S_IFREG. + Type(context.Context) (linux.FileMode, error) + + // IncRef increments reference. + IncRef() + + // DecRef decrements reference. + DecRef() +} + +// Lookup provides a common interface to open files. +type Lookup interface { + // OpenPath opens a file. + OpenPath(ctx context.Context, path string, opts vfs.OpenOptions, remainingTraversals *uint, resolveFinal bool) (File, error) +} diff --git a/pkg/sentry/fsbridge/fs.go b/pkg/sentry/fsbridge/fs.go new file mode 100644 index 000000000..093ce1fb3 --- /dev/null +++ b/pkg/sentry/fsbridge/fs.go @@ -0,0 +1,181 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package fsbridge + +import ( + "io" + "strings" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/memmap" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +// fsFile implements File interface over fs.File. +// +// +stateify savable +type fsFile struct { + file *fs.File +} + +var _ File = (*fsFile)(nil) + +// NewFSFile creates a new File over fs.File. +func NewFSFile(file *fs.File) File { + return &fsFile{file: file} +} + +// PathnameWithDeleted implements File. +func (f *fsFile) PathnameWithDeleted(ctx context.Context) string { + root := fs.RootFromContext(ctx) + if root == nil { + // This doesn't correspond to anything in Linux because the vfs is + // global there. + return "" + } + defer root.DecRef() + + name, _ := f.file.Dirent.FullName(root) + return name +} + +// ReadFull implements File. +func (f *fsFile) ReadFull(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { + var total int64 + for dst.NumBytes() > 0 { + n, err := f.file.Preadv(ctx, dst, offset+total) + total += n + if err == io.EOF && total != 0 { + return total, io.ErrUnexpectedEOF + } else if err != nil { + return total, err + } + dst = dst.DropFirst64(n) + } + return total, nil +} + +// ConfigureMMap implements File. +func (f *fsFile) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { + return f.file.ConfigureMMap(ctx, opts) +} + +// Type implements File. +func (f *fsFile) Type(context.Context) (linux.FileMode, error) { + return linux.FileMode(f.file.Dirent.Inode.StableAttr.Type.LinuxType()), nil +} + +// IncRef implements File. +func (f *fsFile) IncRef() { + f.file.IncRef() +} + +// DecRef implements File. +func (f *fsFile) DecRef() { + f.file.DecRef() +} + +// fsLookup implements Lookup interface using fs.File. +// +// +stateify savable +type fsLookup struct { + mntns *fs.MountNamespace + + root *fs.Dirent + workingDir *fs.Dirent +} + +var _ Lookup = (*fsLookup)(nil) + +// NewFSLookup creates a new Lookup using VFS1. +func NewFSLookup(mntns *fs.MountNamespace, root, workingDir *fs.Dirent) Lookup { + return &fsLookup{ + mntns: mntns, + root: root, + workingDir: workingDir, + } +} + +// OpenPath implements Lookup. +func (l *fsLookup) OpenPath(ctx context.Context, path string, opts vfs.OpenOptions, remainingTraversals *uint, resolveFinal bool) (File, error) { + var d *fs.Dirent + var err error + if resolveFinal { + d, err = l.mntns.FindInode(ctx, l.root, l.workingDir, path, remainingTraversals) + } else { + d, err = l.mntns.FindLink(ctx, l.root, l.workingDir, path, remainingTraversals) + } + if err != nil { + return nil, err + } + defer d.DecRef() + + if !resolveFinal && fs.IsSymlink(d.Inode.StableAttr) { + return nil, syserror.ELOOP + } + + fsPerm := openOptionsToPermMask(&opts) + if err := d.Inode.CheckPermission(ctx, fsPerm); err != nil { + return nil, err + } + + // If they claim it's a directory, then make sure. + if strings.HasSuffix(path, "/") { + if d.Inode.StableAttr.Type != fs.Directory { + return nil, syserror.ENOTDIR + } + } + + if opts.FileExec && d.Inode.StableAttr.Type != fs.RegularFile { + ctx.Infof("%q is not a regular file: %v", path, d.Inode.StableAttr.Type) + return nil, syserror.EACCES + } + + f, err := d.Inode.GetFile(ctx, d, flagsToFileFlags(opts.Flags)) + if err != nil { + return nil, err + } + + return &fsFile{file: f}, nil +} + +func openOptionsToPermMask(opts *vfs.OpenOptions) fs.PermMask { + mode := opts.Flags & linux.O_ACCMODE + return fs.PermMask{ + Read: mode == linux.O_RDONLY || mode == linux.O_RDWR, + Write: mode == linux.O_WRONLY || mode == linux.O_RDWR, + Execute: opts.FileExec, + } +} + +func flagsToFileFlags(flags uint32) fs.FileFlags { + return fs.FileFlags{ + Direct: flags&linux.O_DIRECT != 0, + DSync: flags&(linux.O_DSYNC|linux.O_SYNC) != 0, + Sync: flags&linux.O_SYNC != 0, + NonBlocking: flags&linux.O_NONBLOCK != 0, + Read: (flags & linux.O_ACCMODE) != linux.O_WRONLY, + Write: (flags & linux.O_ACCMODE) != linux.O_RDONLY, + Append: flags&linux.O_APPEND != 0, + Directory: flags&linux.O_DIRECTORY != 0, + Async: flags&linux.O_ASYNC != 0, + LargeFile: flags&linux.O_LARGEFILE != 0, + Truncate: flags&linux.O_TRUNC != 0, + } +} diff --git a/pkg/sentry/fsbridge/vfs.go b/pkg/sentry/fsbridge/vfs.go new file mode 100644 index 000000000..e657c39bc --- /dev/null +++ b/pkg/sentry/fsbridge/vfs.go @@ -0,0 +1,134 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package fsbridge + +import ( + "io" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/memmap" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/usermem" +) + +// fsFile implements File interface over vfs.FileDescription. +// +// +stateify savable +type vfsFile struct { + file *vfs.FileDescription +} + +var _ File = (*vfsFile)(nil) + +// NewVFSFile creates a new File over fs.File. +func NewVFSFile(file *vfs.FileDescription) File { + return &vfsFile{file: file} +} + +// PathnameWithDeleted implements File. +func (f *vfsFile) PathnameWithDeleted(ctx context.Context) string { + root := vfs.RootFromContext(ctx) + defer root.DecRef() + + vfsObj := f.file.VirtualDentry().Mount().Filesystem().VirtualFilesystem() + name, _ := vfsObj.PathnameWithDeleted(ctx, root, f.file.VirtualDentry()) + return name +} + +// ReadFull implements File. +func (f *vfsFile) ReadFull(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { + var total int64 + for dst.NumBytes() > 0 { + n, err := f.file.PRead(ctx, dst, offset+total, vfs.ReadOptions{}) + total += n + if err == io.EOF && total != 0 { + return total, io.ErrUnexpectedEOF + } else if err != nil { + return total, err + } + dst = dst.DropFirst64(n) + } + return total, nil +} + +// ConfigureMMap implements File. +func (f *vfsFile) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { + return f.file.ConfigureMMap(ctx, opts) +} + +// Type implements File. +func (f *vfsFile) Type(ctx context.Context) (linux.FileMode, error) { + stat, err := f.file.Stat(ctx, vfs.StatOptions{}) + if err != nil { + return 0, err + } + return linux.FileMode(stat.Mode).FileType(), nil +} + +// IncRef implements File. +func (f *vfsFile) IncRef() { + f.file.IncRef() +} + +// DecRef implements File. +func (f *vfsFile) DecRef() { + f.file.DecRef() +} + +// fsLookup implements Lookup interface using fs.File. +// +// +stateify savable +type vfsLookup struct { + mntns *vfs.MountNamespace + + root vfs.VirtualDentry + workingDir vfs.VirtualDentry +} + +var _ Lookup = (*vfsLookup)(nil) + +// NewVFSLookup creates a new Lookup using VFS2. +func NewVFSLookup(mntns *vfs.MountNamespace, root, workingDir vfs.VirtualDentry) Lookup { + return &vfsLookup{ + mntns: mntns, + root: root, + workingDir: workingDir, + } +} + +// OpenPath implements Lookup. +// +// remainingTraversals is not configurable in VFS2, all callers are using the +// default anyways. +// +// TODO(gvisor.dev/issue/1623): Check mount has read and exec permission. +func (l *vfsLookup) OpenPath(ctx context.Context, path string, opts vfs.OpenOptions, _ *uint, resolveFinal bool) (File, error) { + vfsObj := l.mntns.Root().Mount().Filesystem().VirtualFilesystem() + creds := auth.CredentialsFromContext(ctx) + pop := &vfs.PathOperation{ + Root: l.root, + Start: l.root, + Path: fspath.Parse(path), + FollowFinalSymlink: resolveFinal, + } + fd, err := vfsObj.OpenAt(ctx, creds, pop, &opts) + if err != nil { + return nil, err + } + return &vfsFile{file: fd}, nil +} diff --git a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go index e03a0c665..abd4f24e7 100644 --- a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go +++ b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go @@ -28,6 +28,9 @@ import ( "gvisor.dev/gvisor/pkg/sync" ) +// Name is the default filesystem name. +const Name = "devtmpfs" + // FilesystemType implements vfs.FilesystemType. type FilesystemType struct { initOnce sync.Once @@ -107,6 +110,7 @@ func (a *Accessor) wrapContext(ctx context.Context) *accessorContext { func (ac *accessorContext) Value(key interface{}) interface{} { switch key { case vfs.CtxMountNamespace: + ac.a.mntns.IncRef() return ac.a.mntns case vfs.CtxRoot: ac.a.root.IncRef() diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 138adb9f7..5cfb0dc4c 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -400,6 +400,7 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b } vfsObj := rp.VirtualFilesystem() mntns := vfs.MountNamespaceFromContext(ctx) + defer mntns.DecRef() parent.dirMu.Lock() defer parent.dirMu.Unlock() childVFSD := parent.vfsd.Child(name) @@ -934,7 +935,9 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa if oldParent == newParent && oldName == newName { return nil } - if err := vfsObj.PrepareRenameDentry(vfs.MountNamespaceFromContext(ctx), &renamed.vfsd, replacedVFSD); err != nil { + mntns := vfs.MountNamespaceFromContext(ctx) + defer mntns.DecRef() + if err := vfsObj.PrepareRenameDentry(mntns, &renamed.vfsd, replacedVFSD); err != nil { return err } if err := renamed.file.rename(ctx, newParent.file, newName); err != nil { diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index d0552bd99..d00850e25 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -52,6 +52,9 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) +// Name is the default filesystem name. +const Name = "9p" + // FilesystemType implements vfs.FilesystemType. type FilesystemType struct{} diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index ee98eb66a..292f58afd 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -544,6 +544,7 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa } mntns := vfs.MountNamespaceFromContext(ctx) + defer mntns.DecRef() virtfs := rp.VirtualFilesystem() srcDirDentry := srcDirVFSD.Impl().(*Dentry) @@ -595,7 +596,10 @@ func (fs *Filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error parentDentry := vfsd.Parent().Impl().(*Dentry) parentDentry.dirMu.Lock() defer parentDentry.dirMu.Unlock() - if err := virtfs.PrepareDeleteDentry(vfs.MountNamespaceFromContext(ctx), vfsd); err != nil { + + mntns := vfs.MountNamespaceFromContext(ctx) + defer mntns.DecRef() + if err := virtfs.PrepareDeleteDentry(mntns, vfsd); err != nil { return err } if err := parentDentry.inode.RmDir(ctx, rp.Component(), vfsd); err != nil { @@ -697,7 +701,9 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error parentDentry := vfsd.Parent().Impl().(*Dentry) parentDentry.dirMu.Lock() defer parentDentry.dirMu.Unlock() - if err := virtfs.PrepareDeleteDentry(vfs.MountNamespaceFromContext(ctx), vfsd); err != nil { + mntns := vfs.MountNamespaceFromContext(ctx) + defer mntns.DecRef() + if err := virtfs.PrepareDeleteDentry(mntns, vfsd); err != nil { return err } if err := parentDentry.inode.Unlink(ctx, rp.Component(), vfsd); err != nil { diff --git a/pkg/sentry/fsimpl/proc/BUILD b/pkg/sentry/fsimpl/proc/BUILD index 12aac2e6a..a83245866 100644 --- a/pkg/sentry/fsimpl/proc/BUILD +++ b/pkg/sentry/fsimpl/proc/BUILD @@ -14,6 +14,7 @@ go_library( "tasks_net.go", "tasks_sys.go", ], + visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", "//pkg/context", diff --git a/pkg/sentry/fsimpl/proc/filesystem.go b/pkg/sentry/fsimpl/proc/filesystem.go index 11477b6a9..5c19d5522 100644 --- a/pkg/sentry/fsimpl/proc/filesystem.go +++ b/pkg/sentry/fsimpl/proc/filesystem.go @@ -26,15 +26,18 @@ import ( "gvisor.dev/gvisor/pkg/sentry/vfs" ) -// procFSType is the factory class for procfs. +// Name is the default filesystem name. +const Name = "proc" + +// FilesystemType is the factory class for procfs. // // +stateify savable -type procFSType struct{} +type FilesystemType struct{} -var _ vfs.FilesystemType = (*procFSType)(nil) +var _ vfs.FilesystemType = (*FilesystemType)(nil) // GetFilesystem implements vfs.FilesystemType. -func (ft *procFSType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { +func (ft *FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { k := kernel.KernelFromContext(ctx) if k == nil { return nil, nil, fmt.Errorf("procfs requires a kernel") @@ -47,12 +50,13 @@ func (ft *procFSType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFile procfs := &kernfs.Filesystem{} procfs.VFSFilesystem().Init(vfsObj, procfs) - var data *InternalData + var cgroups map[string]string if opts.InternalData != nil { - data = opts.InternalData.(*InternalData) + data := opts.InternalData.(*InternalData) + cgroups = data.Cgroups } - _, dentry := newTasksInode(procfs, k, pidns, data.Cgroups) + _, dentry := newTasksInode(procfs, k, pidns, cgroups) return procfs.VFSFilesystem(), dentry.VFSDentry(), nil } diff --git a/pkg/sentry/fsimpl/proc/tasks_test.go b/pkg/sentry/fsimpl/proc/tasks_test.go index 6fc3524db..96c72cbc9 100644 --- a/pkg/sentry/fsimpl/proc/tasks_test.go +++ b/pkg/sentry/fsimpl/proc/tasks_test.go @@ -90,8 +90,7 @@ func setup(t *testing.T) *testutil.System { ctx := k.SupervisorContext() creds := auth.CredentialsFromContext(ctx) - vfsObj := vfs.New() - vfsObj.MustRegisterFilesystemType("procfs", &procFSType{}, &vfs.RegisterFilesystemTypeOptions{ + k.VFS.MustRegisterFilesystemType(Name, &FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, }) fsOpts := vfs.GetFilesystemOptions{ @@ -102,11 +101,11 @@ func setup(t *testing.T) *testutil.System { }, }, } - mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "procfs", &fsOpts) + mntns, err := k.VFS.NewMountNamespace(ctx, creds, "", Name, &fsOpts) if err != nil { t.Fatalf("NewMountNamespace(): %v", err) } - return testutil.NewSystem(ctx, t, vfsObj, mntns) + return testutil.NewSystem(ctx, t, k.VFS, mntns) } func TestTasksEmpty(t *testing.T) { @@ -131,7 +130,7 @@ func TestTasks(t *testing.T) { var tasks []*kernel.Task for i := 0; i < 5; i++ { tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - task, err := testutil.CreateTask(s.Ctx, fmt.Sprintf("name-%d", i), tc) + task, err := testutil.CreateTask(s.Ctx, fmt.Sprintf("name-%d", i), tc, s.MntNs, s.Root, s.Root) if err != nil { t.Fatalf("CreateTask(): %v", err) } @@ -213,7 +212,7 @@ func TestTasksOffset(t *testing.T) { k := kernel.KernelFromContext(s.Ctx) for i := 0; i < 3; i++ { tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - if _, err := testutil.CreateTask(s.Ctx, fmt.Sprintf("name-%d", i), tc); err != nil { + if _, err := testutil.CreateTask(s.Ctx, fmt.Sprintf("name-%d", i), tc, s.MntNs, s.Root, s.Root); err != nil { t.Fatalf("CreateTask(): %v", err) } } @@ -337,7 +336,7 @@ func TestTask(t *testing.T) { k := kernel.KernelFromContext(s.Ctx) tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - _, err := testutil.CreateTask(s.Ctx, "name", tc) + _, err := testutil.CreateTask(s.Ctx, "name", tc, s.MntNs, s.Root, s.Root) if err != nil { t.Fatalf("CreateTask(): %v", err) } @@ -352,7 +351,7 @@ func TestProcSelf(t *testing.T) { k := kernel.KernelFromContext(s.Ctx) tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - task, err := testutil.CreateTask(s.Ctx, "name", tc) + task, err := testutil.CreateTask(s.Ctx, "name", tc, s.MntNs, s.Root, s.Root) if err != nil { t.Fatalf("CreateTask(): %v", err) } @@ -433,7 +432,7 @@ func TestTree(t *testing.T) { var tasks []*kernel.Task for i := 0; i < 5; i++ { tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - task, err := testutil.CreateTask(s.Ctx, fmt.Sprintf("name-%d", i), tc) + task, err := testutil.CreateTask(s.Ctx, fmt.Sprintf("name-%d", i), tc, s.MntNs, s.Root, s.Root) if err != nil { t.Fatalf("CreateTask(): %v", err) } diff --git a/pkg/sentry/fsimpl/sys/BUILD b/pkg/sentry/fsimpl/sys/BUILD index 66c0d8bc8..a741e2bb6 100644 --- a/pkg/sentry/fsimpl/sys/BUILD +++ b/pkg/sentry/fsimpl/sys/BUILD @@ -7,6 +7,7 @@ go_library( srcs = [ "sys.go", ], + visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", "//pkg/context", diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go index d693fceae..c36c4fa11 100644 --- a/pkg/sentry/fsimpl/sys/sys.go +++ b/pkg/sentry/fsimpl/sys/sys.go @@ -28,6 +28,9 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) +// Name is the default filesystem name. +const Name = "sysfs" + // FilesystemType implements vfs.FilesystemType. type FilesystemType struct{} diff --git a/pkg/sentry/fsimpl/sys/sys_test.go b/pkg/sentry/fsimpl/sys/sys_test.go index 8b1cf0bd0..5d1ba5867 100644 --- a/pkg/sentry/fsimpl/sys/sys_test.go +++ b/pkg/sentry/fsimpl/sys/sys_test.go @@ -34,16 +34,15 @@ func newTestSystem(t *testing.T) *testutil.System { } ctx := k.SupervisorContext() creds := auth.CredentialsFromContext(ctx) - v := vfs.New() - v.MustRegisterFilesystemType("sysfs", sys.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + k.VFS.MustRegisterFilesystemType(sys.Name, sys.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, }) - mns, err := v.NewMountNamespace(ctx, creds, "", "sysfs", &vfs.GetFilesystemOptions{}) + mns, err := k.VFS.NewMountNamespace(ctx, creds, "", sys.Name, &vfs.GetFilesystemOptions{}) if err != nil { t.Fatalf("Failed to create new mount namespace: %v", err) } - return testutil.NewSystem(ctx, t, v, mns) + return testutil.NewSystem(ctx, t, k.VFS, mns) } func TestReadCPUFile(t *testing.T) { diff --git a/pkg/sentry/fsimpl/testutil/BUILD b/pkg/sentry/fsimpl/testutil/BUILD index efd5974c4..e4f36f4ae 100644 --- a/pkg/sentry/fsimpl/testutil/BUILD +++ b/pkg/sentry/fsimpl/testutil/BUILD @@ -16,7 +16,7 @@ go_library( "//pkg/cpuid", "//pkg/fspath", "//pkg/memutil", - "//pkg/sentry/fs", + "//pkg/sentry/fsimpl/tmpfs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/sched", diff --git a/pkg/sentry/fsimpl/testutil/kernel.go b/pkg/sentry/fsimpl/testutil/kernel.go index 89f8c4915..a91b3ec4d 100644 --- a/pkg/sentry/fsimpl/testutil/kernel.go +++ b/pkg/sentry/fsimpl/testutil/kernel.go @@ -24,7 +24,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/memutil" - "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/sched" @@ -33,6 +33,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/time" + "gvisor.dev/gvisor/pkg/sentry/vfs" // Platforms are plugable. _ "gvisor.dev/gvisor/pkg/sentry/platform/kvm" @@ -99,26 +100,27 @@ func Boot() (*kernel.Kernel, error) { return nil, fmt.Errorf("initializing kernel: %v", err) } - ctx := k.SupervisorContext() + kernel.VFS2Enabled = true + + vfsObj := vfs.New() + vfsObj.MustRegisterFilesystemType(tmpfs.Name, &tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + AllowUserList: true, + }) + k.VFS = vfsObj - // Create mount namespace without root as it's the minimum required to create - // the global thread group. - mntns, err := fs.NewMountNamespace(ctx, nil) - if err != nil { - return nil, err - } ls, err := limits.NewLinuxLimitSet() if err != nil { return nil, err } - tg := k.NewThreadGroup(mntns, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, ls) + tg := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, ls) k.TestOnly_SetGlobalInit(tg) return k, nil } // CreateTask creates a new bare bones task for tests. -func CreateTask(ctx context.Context, name string, tc *kernel.ThreadGroup) (*kernel.Task, error) { +func CreateTask(ctx context.Context, name string, tc *kernel.ThreadGroup, mntns *vfs.MountNamespace, root, cwd vfs.VirtualDentry) (*kernel.Task, error) { k := kernel.KernelFromContext(ctx) config := &kernel.TaskConfig{ Kernel: k, @@ -129,6 +131,8 @@ func CreateTask(ctx context.Context, name string, tc *kernel.ThreadGroup) (*kern UTSNamespace: kernel.UTSNamespaceFromContext(ctx), IPCNamespace: kernel.IPCNamespaceFromContext(ctx), AbstractSocketNamespace: kernel.NewAbstractSocketNamespace(), + MountNamespaceVFS2: mntns, + FSContext: kernel.NewFSContextVFS2(root, cwd, 0022), } return k.TaskSet().NewTask(config) } diff --git a/pkg/sentry/fsimpl/testutil/testutil.go b/pkg/sentry/fsimpl/testutil/testutil.go index 69fd84ddd..b97e3534a 100644 --- a/pkg/sentry/fsimpl/testutil/testutil.go +++ b/pkg/sentry/fsimpl/testutil/testutil.go @@ -41,12 +41,12 @@ type System struct { Creds *auth.Credentials VFS *vfs.VirtualFilesystem Root vfs.VirtualDentry - mns *vfs.MountNamespace + MntNs *vfs.MountNamespace } // NewSystem constructs a System. // -// Precondition: Caller must hold a reference on mns, whose ownership +// Precondition: Caller must hold a reference on MntNs, whose ownership // is transferred to the new System. func NewSystem(ctx context.Context, t *testing.T, v *vfs.VirtualFilesystem, mns *vfs.MountNamespace) *System { s := &System{ @@ -54,7 +54,7 @@ func NewSystem(ctx context.Context, t *testing.T, v *vfs.VirtualFilesystem, mns Ctx: ctx, Creds: auth.CredentialsFromContext(ctx), VFS: v, - mns: mns, + MntNs: mns, Root: mns.Root(), } return s @@ -75,7 +75,7 @@ func (s *System) WithSubtest(t *testing.T) *System { Ctx: s.Ctx, Creds: s.Creds, VFS: s.VFS, - mns: s.mns, + MntNs: s.MntNs, Root: s.Root, } } @@ -90,7 +90,7 @@ func (s *System) WithTemporaryContext(ctx context.Context) *System { Ctx: ctx, Creds: s.Creds, VFS: s.VFS, - mns: s.mns, + MntNs: s.MntNs, Root: s.Root, } } @@ -98,7 +98,7 @@ func (s *System) WithTemporaryContext(ctx context.Context) *System { // Destroy release resources associated with a test system. func (s *System) Destroy() { s.Root.DecRef() - s.mns.DecRef() // Reference on mns passed to NewSystem. + s.MntNs.DecRef() // Reference on MntNs passed to NewSystem. } // ReadToEnd reads the contents of fd until EOF to a string. diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 8785452b6..7f7b791c4 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -486,7 +486,9 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa vfsObj := rp.VirtualFilesystem() oldParentDir := oldParent.inode.impl.(*directory) newParentDir := newParent.inode.impl.(*directory) - if err := vfsObj.PrepareRenameDentry(vfs.MountNamespaceFromContext(ctx), renamedVFSD, replacedVFSD); err != nil { + mntns := vfs.MountNamespaceFromContext(ctx) + defer mntns.DecRef() + if err := vfsObj.PrepareRenameDentry(mntns, renamedVFSD, replacedVFSD); err != nil { return err } if replaced != nil { @@ -543,7 +545,9 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error } defer mnt.EndWrite() vfsObj := rp.VirtualFilesystem() - if err := vfsObj.PrepareDeleteDentry(vfs.MountNamespaceFromContext(ctx), childVFSD); err != nil { + mntns := vfs.MountNamespaceFromContext(ctx) + defer mntns.DecRef() + if err := vfsObj.PrepareDeleteDentry(mntns, childVFSD); err != nil { return err } parent.inode.impl.(*directory).childList.Remove(child) @@ -631,7 +635,9 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error } defer mnt.EndWrite() vfsObj := rp.VirtualFilesystem() - if err := vfsObj.PrepareDeleteDentry(vfs.MountNamespaceFromContext(ctx), childVFSD); err != nil { + mntns := vfs.MountNamespaceFromContext(ctx) + defer mntns.DecRef() + if err := vfsObj.PrepareDeleteDentry(mntns, childVFSD); err != nil { return err } parent.inode.impl.(*directory).childList.Remove(child) diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 2108d0f4d..c5bb17562 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -40,6 +40,9 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) +// Name is the default filesystem name. +const Name = "tmpfs" + // FilesystemType implements vfs.FilesystemType. type FilesystemType struct{} diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD index 2231d6973..46306945f 100644 --- a/pkg/sentry/kernel/BUILD +++ b/pkg/sentry/kernel/BUILD @@ -157,6 +157,7 @@ go_library( "//pkg/context", "//pkg/cpuid", "//pkg/eventchannel", + "//pkg/fspath", "//pkg/log", "//pkg/metric", "//pkg/refs", @@ -167,6 +168,7 @@ go_library( "//pkg/sentry/fs", "//pkg/sentry/fs/lock", "//pkg/sentry/fs/timerfd", + "//pkg/sentry/fsbridge", "//pkg/sentry/hostcpu", "//pkg/sentry/inet", "//pkg/sentry/kernel/auth", diff --git a/pkg/sentry/kernel/fs_context.go b/pkg/sentry/kernel/fs_context.go index 2448c1d99..7218aa24e 100644 --- a/pkg/sentry/kernel/fs_context.go +++ b/pkg/sentry/kernel/fs_context.go @@ -19,6 +19,7 @@ import ( "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" ) @@ -37,10 +38,16 @@ type FSContext struct { // destroyed. root *fs.Dirent + // rootVFS2 is the filesystem root. + rootVFS2 vfs.VirtualDentry + // cwd is the current working directory. Will be nil iff the FSContext // has been destroyed. cwd *fs.Dirent + // cwdVFS2 is the current working directory. + cwdVFS2 vfs.VirtualDentry + // umask is the current file mode creation mask. When a thread using this // context invokes a syscall that creates a file, bits set in umask are // removed from the permissions that the file is created with. @@ -60,6 +67,19 @@ func newFSContext(root, cwd *fs.Dirent, umask uint) *FSContext { return &f } +// NewFSContextVFS2 returns a new filesystem context. +func NewFSContextVFS2(root, cwd vfs.VirtualDentry, umask uint) *FSContext { + root.IncRef() + cwd.IncRef() + f := FSContext{ + rootVFS2: root, + cwdVFS2: cwd, + umask: umask, + } + f.EnableLeakCheck("kernel.FSContext") + return &f +} + // destroy is the destructor for an FSContext. // // This will call DecRef on both root and cwd Dirents. If either call to @@ -75,11 +95,17 @@ func (f *FSContext) destroy() { f.mu.Lock() defer f.mu.Unlock() - f.root.DecRef() - f.root = nil - - f.cwd.DecRef() - f.cwd = nil + if VFS2Enabled { + f.rootVFS2.DecRef() + f.rootVFS2 = vfs.VirtualDentry{} + f.cwdVFS2.DecRef() + f.cwdVFS2 = vfs.VirtualDentry{} + } else { + f.root.DecRef() + f.root = nil + f.cwd.DecRef() + f.cwd = nil + } } // DecRef implements RefCounter.DecRef with destructor f.destroy. @@ -93,12 +119,21 @@ func (f *FSContext) DecRef() { func (f *FSContext) Fork() *FSContext { f.mu.Lock() defer f.mu.Unlock() - f.cwd.IncRef() - f.root.IncRef() + + if VFS2Enabled { + f.cwdVFS2.IncRef() + f.rootVFS2.IncRef() + } else { + f.cwd.IncRef() + f.root.IncRef() + } + return &FSContext{ - cwd: f.cwd, - root: f.root, - umask: f.umask, + cwd: f.cwd, + root: f.root, + cwdVFS2: f.cwdVFS2, + rootVFS2: f.rootVFS2, + umask: f.umask, } } @@ -109,12 +144,23 @@ func (f *FSContext) Fork() *FSContext { func (f *FSContext) WorkingDirectory() *fs.Dirent { f.mu.Lock() defer f.mu.Unlock() - if f.cwd != nil { - f.cwd.IncRef() - } + + f.cwd.IncRef() return f.cwd } +// WorkingDirectoryVFS2 returns the current working directory. +// +// This will return nil if called after destroy(), otherwise it will return a +// Dirent with a reference taken. +func (f *FSContext) WorkingDirectoryVFS2() vfs.VirtualDentry { + f.mu.Lock() + defer f.mu.Unlock() + + f.cwdVFS2.IncRef() + return f.cwdVFS2 +} + // SetWorkingDirectory sets the current working directory. // This will take an extra reference on the Dirent. // @@ -137,6 +183,20 @@ func (f *FSContext) SetWorkingDirectory(d *fs.Dirent) { old.DecRef() } +// SetWorkingDirectoryVFS2 sets the current working directory. +// This will take an extra reference on the VirtualDentry. +// +// This is not a valid call after destroy. +func (f *FSContext) SetWorkingDirectoryVFS2(d vfs.VirtualDentry) { + f.mu.Lock() + defer f.mu.Unlock() + + old := f.cwdVFS2 + f.cwdVFS2 = d + d.IncRef() + old.DecRef() +} + // RootDirectory returns the current filesystem root. // // This will return nil if called after destroy(), otherwise it will return a @@ -150,6 +210,18 @@ func (f *FSContext) RootDirectory() *fs.Dirent { return f.root } +// RootDirectoryVFS2 returns the current filesystem root. +// +// This will return nil if called after destroy(), otherwise it will return a +// Dirent with a reference taken. +func (f *FSContext) RootDirectoryVFS2() vfs.VirtualDentry { + f.mu.Lock() + defer f.mu.Unlock() + + f.rootVFS2.IncRef() + return f.rootVFS2 +} + // SetRootDirectory sets the root directory. // This will take an extra reference on the Dirent. // diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 3ee760ba2..2665f057c 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -43,11 +43,13 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/eventchannel" + "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/timerfd" + "gvisor.dev/gvisor/pkg/sentry/fsbridge" "gvisor.dev/gvisor/pkg/sentry/hostcpu" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" @@ -71,6 +73,10 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" ) +// VFS2Enabled is set to true when VFS2 is enabled. Added as a global for allow +// easy access everywhere. To be removed once VFS2 becomes the default. +var VFS2Enabled = false + // Kernel represents an emulated Linux kernel. It must be initialized by calling // Init() or LoadFrom(). // @@ -238,6 +244,9 @@ type Kernel struct { // SpecialOpts contains special kernel options. SpecialOpts + + // VFS keeps the filesystem state used across the kernel. + VFS *vfs.VirtualFilesystem } // InitKernelArgs holds arguments to Init. @@ -624,7 +633,7 @@ type CreateProcessArgs struct { // File is a passed host FD pointing to a file to load as the init binary. // // This is checked if and only if Filename is "". - File *fs.File + File fsbridge.File // Argvv is a list of arguments. Argv []string @@ -673,6 +682,13 @@ type CreateProcessArgs struct { // increment it). MountNamespace *fs.MountNamespace + // MountNamespaceVFS2 optionally contains the mount namespace for this + // process. If nil, the init process's mount namespace is used. + // + // Anyone setting MountNamespaceVFS2 must donate a reference (i.e. + // increment it). + MountNamespaceVFS2 *vfs.MountNamespace + // ContainerID is the container that the process belongs to. ContainerID string } @@ -711,11 +727,22 @@ func (ctx *createProcessContext) Value(key interface{}) interface{} { return ctx.args.Credentials case fs.CtxRoot: if ctx.args.MountNamespace != nil { - // MountNamespace.Root() will take a reference on the root - // dirent for us. + // MountNamespace.Root() will take a reference on the root dirent for us. return ctx.args.MountNamespace.Root() } return nil + case vfs.CtxRoot: + if ctx.args.MountNamespaceVFS2 == nil { + return nil + } + // MountNamespaceVFS2.Root() takes a reference on the root dirent for us. + return ctx.args.MountNamespaceVFS2.Root() + case vfs.CtxMountNamespace: + if ctx.k.globalInit == nil { + return nil + } + // MountNamespaceVFS2 takes a reference for us. + return ctx.k.GlobalInit().Leader().MountNamespaceVFS2() case fs.CtxDirentCacheLimiter: return ctx.k.DirentCacheLimiter case ktime.CtxRealtimeClock: @@ -757,34 +784,77 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, defer k.extMu.Unlock() log.Infof("EXEC: %v", args.Argv) - // Grab the mount namespace. - mounts := args.MountNamespace - if mounts == nil { - mounts = k.GlobalInit().Leader().MountNamespace() - mounts.IncRef() - } - - tg := k.NewThreadGroup(mounts, args.PIDNamespace, NewSignalHandlers(), linux.SIGCHLD, args.Limits) ctx := args.NewContext(k) - // Get the root directory from the MountNamespace. - root := mounts.Root() - // The call to newFSContext below will take a reference on root, so we - // don't need to hold this one. - defer root.DecRef() - - // Grab the working directory. - remainingTraversals := uint(args.MaxSymlinkTraversals) - wd := root // Default. - if args.WorkingDirectory != "" { - var err error - wd, err = mounts.FindInode(ctx, root, nil, args.WorkingDirectory, &remainingTraversals) - if err != nil { - return nil, 0, fmt.Errorf("failed to find initial working directory %q: %v", args.WorkingDirectory, err) + var ( + opener fsbridge.Lookup + fsContext *FSContext + mntns *fs.MountNamespace + ) + + if VFS2Enabled { + mntnsVFS2 := args.MountNamespaceVFS2 + if mntnsVFS2 == nil { + // MountNamespaceVFS2 adds a reference to the namespace, which is + // transferred to the new process. + mntnsVFS2 = k.GlobalInit().Leader().MountNamespaceVFS2() + } + // Get the root directory from the MountNamespace. + root := args.MountNamespaceVFS2.Root() + // The call to newFSContext below will take a reference on root, so we + // don't need to hold this one. + defer root.DecRef() + + // Grab the working directory. + wd := root // Default. + if args.WorkingDirectory != "" { + pop := vfs.PathOperation{ + Root: root, + Start: wd, + Path: fspath.Parse(args.WorkingDirectory), + FollowFinalSymlink: true, + } + var err error + wd, err = k.VFS.GetDentryAt(ctx, args.Credentials, &pop, &vfs.GetDentryOptions{ + CheckSearchable: true, + }) + if err != nil { + return nil, 0, fmt.Errorf("failed to find initial working directory %q: %v", args.WorkingDirectory, err) + } + defer wd.DecRef() + } + opener = fsbridge.NewVFSLookup(mntnsVFS2, root, wd) + fsContext = NewFSContextVFS2(root, wd, args.Umask) + + } else { + mntns = args.MountNamespace + if mntns == nil { + mntns = k.GlobalInit().Leader().MountNamespace() + mntns.IncRef() } - defer wd.DecRef() + // Get the root directory from the MountNamespace. + root := mntns.Root() + // The call to newFSContext below will take a reference on root, so we + // don't need to hold this one. + defer root.DecRef() + + // Grab the working directory. + remainingTraversals := args.MaxSymlinkTraversals + wd := root // Default. + if args.WorkingDirectory != "" { + var err error + wd, err = mntns.FindInode(ctx, root, nil, args.WorkingDirectory, &remainingTraversals) + if err != nil { + return nil, 0, fmt.Errorf("failed to find initial working directory %q: %v", args.WorkingDirectory, err) + } + defer wd.DecRef() + } + opener = fsbridge.NewFSLookup(mntns, root, wd) + fsContext = newFSContext(root, wd, args.Umask) } + tg := k.NewThreadGroup(mntns, args.PIDNamespace, NewSignalHandlers(), linux.SIGCHLD, args.Limits) + // Check which file to start from. switch { case args.Filename != "": @@ -805,11 +875,9 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, } // Create a fresh task context. - remainingTraversals = uint(args.MaxSymlinkTraversals) + remainingTraversals := args.MaxSymlinkTraversals loadArgs := loader.LoadArgs{ - Mounts: mounts, - Root: root, - WorkingDirectory: wd, + Opener: opener, RemainingTraversals: &remainingTraversals, ResolveFinal: true, Filename: args.Filename, @@ -834,13 +902,14 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, Kernel: k, ThreadGroup: tg, TaskContext: tc, - FSContext: newFSContext(root, wd, args.Umask), + FSContext: fsContext, FDTable: args.FDTable, Credentials: args.Credentials, AllowedCPUMask: sched.NewFullCPUSet(k.applicationCores), UTSNamespace: args.UTSNamespace, IPCNamespace: args.IPCNamespace, AbstractSocketNamespace: args.AbstractSocketNamespace, + MountNamespaceVFS2: args.MountNamespaceVFS2, ContainerID: args.ContainerID, } t, err := k.tasks.NewTask(config) @@ -1378,6 +1447,20 @@ func (ctx supervisorContext) Value(key interface{}) interface{} { return ctx.k.globalInit.mounts.Root() } return nil + case vfs.CtxRoot: + if ctx.k.globalInit == nil { + return vfs.VirtualDentry{} + } + mntns := ctx.k.GlobalInit().Leader().MountNamespaceVFS2() + defer mntns.DecRef() + // Root() takes a reference on the root dirent for us. + return mntns.Root() + case vfs.CtxMountNamespace: + if ctx.k.globalInit == nil { + return nil + } + // MountNamespaceVFS2() takes a reference for us. + return ctx.k.GlobalInit().Leader().MountNamespaceVFS2() case fs.CtxDirentCacheLimiter: return ctx.k.DirentCacheLimiter case ktime.CtxRealtimeClock: diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index 981e8c7fe..a3443ff21 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -424,6 +424,11 @@ type Task struct { // abstractSockets is protected by mu. abstractSockets *AbstractSocketNamespace + // mountNamespaceVFS2 is the task's mount namespace. + // + // It is protected by mu. It is owned by the task goroutine. + mountNamespaceVFS2 *vfs.MountNamespace + // parentDeathSignal is sent to this task's thread group when its parent exits. // // parentDeathSignal is protected by mu. @@ -638,6 +643,11 @@ func (t *Task) Value(key interface{}) interface{} { return int32(t.ThreadGroup().ID()) case fs.CtxRoot: return t.fsContext.RootDirectory() + case vfs.CtxRoot: + return t.fsContext.RootDirectoryVFS2() + case vfs.CtxMountNamespace: + t.mountNamespaceVFS2.IncRef() + return t.mountNamespaceVFS2 case fs.CtxDirentCacheLimiter: return t.k.DirentCacheLimiter case inet.CtxStack: @@ -701,6 +711,14 @@ func (t *Task) SyscallRestartBlock() SyscallRestartBlock { // Preconditions: The caller must be running on the task goroutine, or t.mu // must be locked. func (t *Task) IsChrooted() bool { + if VFS2Enabled { + realRoot := t.mountNamespaceVFS2.Root() + defer realRoot.DecRef() + root := t.fsContext.RootDirectoryVFS2() + defer root.DecRef() + return root != realRoot + } + realRoot := t.tg.mounts.Root() defer realRoot.DecRef() root := t.fsContext.RootDirectory() @@ -796,6 +814,15 @@ func (t *Task) MountNamespace() *fs.MountNamespace { return t.tg.mounts } +// MountNamespaceVFS2 returns t's MountNamespace. A reference is taken on the +// returned mount namespace. +func (t *Task) MountNamespaceVFS2() *vfs.MountNamespace { + t.mu.Lock() + defer t.mu.Unlock() + t.mountNamespaceVFS2.IncRef() + return t.mountNamespaceVFS2 +} + // AbstractSockets returns t's AbstractSocketNamespace. func (t *Task) AbstractSockets() *AbstractSocketNamespace { return t.abstractSockets diff --git a/pkg/sentry/kernel/task_clone.go b/pkg/sentry/kernel/task_clone.go index 53d4d211b..ba74b4c1c 100644 --- a/pkg/sentry/kernel/task_clone.go +++ b/pkg/sentry/kernel/task_clone.go @@ -199,6 +199,12 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { ipcns = NewIPCNamespace(userns) } + // TODO(b/63601033): Implement CLONE_NEWNS. + mntnsVFS2 := t.mountNamespaceVFS2 + if mntnsVFS2 != nil { + mntnsVFS2.IncRef() + } + tc, err := t.tc.Fork(t, t.k, !opts.NewAddressSpace) if err != nil { return 0, nil, err @@ -241,7 +247,9 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { rseqAddr := usermem.Addr(0) rseqSignature := uint32(0) if opts.NewThreadGroup { - tg.mounts.IncRef() + if tg.mounts != nil { + tg.mounts.IncRef() + } sh := t.tg.signalHandlers if opts.NewSignalHandlers { sh = sh.Fork() @@ -265,6 +273,7 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { UTSNamespace: utsns, IPCNamespace: ipcns, AbstractSocketNamespace: t.abstractSockets, + MountNamespaceVFS2: mntnsVFS2, RSeqAddr: rseqAddr, RSeqSignature: rseqSignature, ContainerID: t.ContainerID(), diff --git a/pkg/sentry/kernel/task_context.go b/pkg/sentry/kernel/task_context.go index 2d6e7733c..2be982684 100644 --- a/pkg/sentry/kernel/task_context.go +++ b/pkg/sentry/kernel/task_context.go @@ -136,7 +136,7 @@ func (t *Task) Stack() *arch.Stack { func (k *Kernel) LoadTaskImage(ctx context.Context, args loader.LoadArgs) (*TaskContext, *syserr.Error) { // If File is not nil, we should load that instead of resolving Filename. if args.File != nil { - args.Filename = args.File.MappedName(ctx) + args.Filename = args.File.PathnameWithDeleted(ctx) } // Prepare a new user address space to load into. diff --git a/pkg/sentry/kernel/task_exit.go b/pkg/sentry/kernel/task_exit.go index 435761e5a..c4ade6e8e 100644 --- a/pkg/sentry/kernel/task_exit.go +++ b/pkg/sentry/kernel/task_exit.go @@ -269,6 +269,13 @@ func (*runExitMain) execute(t *Task) taskRunState { t.fsContext.DecRef() t.fdTable.DecRef() + t.mu.Lock() + if t.mountNamespaceVFS2 != nil { + t.mountNamespaceVFS2.DecRef() + t.mountNamespaceVFS2 = nil + } + t.mu.Unlock() + // If this is the last task to exit from the thread group, release the // thread group's resources. if lastExiter { diff --git a/pkg/sentry/kernel/task_log.go b/pkg/sentry/kernel/task_log.go index 41259210c..6d737d3e5 100644 --- a/pkg/sentry/kernel/task_log.go +++ b/pkg/sentry/kernel/task_log.go @@ -198,18 +198,11 @@ func (t *Task) traceExecEvent(tc *TaskContext) { if !trace.IsEnabled() { return } - d := tc.MemoryManager.Executable() - if d == nil { + file := tc.MemoryManager.Executable() + if file == nil { trace.Logf(t.traceContext, traceCategory, "exec: << unknown >>") return } - defer d.DecRef() - root := t.fsContext.RootDirectory() - if root == nil { - trace.Logf(t.traceContext, traceCategory, "exec: << no root directory >>") - return - } - defer root.DecRef() - n, _ := d.FullName(root) - trace.Logf(t.traceContext, traceCategory, "exec: %s", n) + defer file.DecRef() + trace.Logf(t.traceContext, traceCategory, "exec: %s", file.PathnameWithDeleted(t)) } diff --git a/pkg/sentry/kernel/task_start.go b/pkg/sentry/kernel/task_start.go index de838beef..f9236a842 100644 --- a/pkg/sentry/kernel/task_start.go +++ b/pkg/sentry/kernel/task_start.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/futex" "gvisor.dev/gvisor/pkg/sentry/kernel/sched" "gvisor.dev/gvisor/pkg/sentry/usage" + "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" ) @@ -80,6 +81,9 @@ type TaskConfig struct { // AbstractSocketNamespace is the AbstractSocketNamespace of the new task. AbstractSocketNamespace *AbstractSocketNamespace + // MountNamespaceVFS2 is the MountNamespace of the new task. + MountNamespaceVFS2 *vfs.MountNamespace + // RSeqAddr is a pointer to the the userspace linux.RSeq structure. RSeqAddr usermem.Addr @@ -116,28 +120,29 @@ func (ts *TaskSet) newTask(cfg *TaskConfig) (*Task, error) { parent: cfg.Parent, children: make(map[*Task]struct{}), }, - runState: (*runApp)(nil), - interruptChan: make(chan struct{}, 1), - signalMask: cfg.SignalMask, - signalStack: arch.SignalStack{Flags: arch.SignalStackFlagDisable}, - tc: *tc, - fsContext: cfg.FSContext, - fdTable: cfg.FDTable, - p: cfg.Kernel.Platform.NewContext(), - k: cfg.Kernel, - ptraceTracees: make(map[*Task]struct{}), - allowedCPUMask: cfg.AllowedCPUMask.Copy(), - ioUsage: &usage.IO{}, - niceness: cfg.Niceness, - netns: cfg.NetworkNamespaced, - utsns: cfg.UTSNamespace, - ipcns: cfg.IPCNamespace, - abstractSockets: cfg.AbstractSocketNamespace, - rseqCPU: -1, - rseqAddr: cfg.RSeqAddr, - rseqSignature: cfg.RSeqSignature, - futexWaiter: futex.NewWaiter(), - containerID: cfg.ContainerID, + runState: (*runApp)(nil), + interruptChan: make(chan struct{}, 1), + signalMask: cfg.SignalMask, + signalStack: arch.SignalStack{Flags: arch.SignalStackFlagDisable}, + tc: *tc, + fsContext: cfg.FSContext, + fdTable: cfg.FDTable, + p: cfg.Kernel.Platform.NewContext(), + k: cfg.Kernel, + ptraceTracees: make(map[*Task]struct{}), + allowedCPUMask: cfg.AllowedCPUMask.Copy(), + ioUsage: &usage.IO{}, + niceness: cfg.Niceness, + netns: cfg.NetworkNamespaced, + utsns: cfg.UTSNamespace, + ipcns: cfg.IPCNamespace, + abstractSockets: cfg.AbstractSocketNamespace, + mountNamespaceVFS2: cfg.MountNamespaceVFS2, + rseqCPU: -1, + rseqAddr: cfg.RSeqAddr, + rseqSignature: cfg.RSeqSignature, + futexWaiter: futex.NewWaiter(), + containerID: cfg.ContainerID, } t.creds.Store(cfg.Credentials) t.endStopCond.L = &t.tg.signalHandlers.mu diff --git a/pkg/sentry/kernel/thread_group.go b/pkg/sentry/kernel/thread_group.go index 768e958d2..268f62e9d 100644 --- a/pkg/sentry/kernel/thread_group.go +++ b/pkg/sentry/kernel/thread_group.go @@ -256,7 +256,7 @@ type ThreadGroup struct { tty *TTY } -// NewThreadGroup returns a new, empty thread group in PID namespace ns. The +// NewThreadGroup returns a new, empty thread group in PID namespace pidns. The // thread group leader will send its parent terminationSignal when it exits. // The new thread group isn't visible to the system until a task has been // created inside of it by a successful call to TaskSet.NewTask. @@ -317,7 +317,9 @@ func (tg *ThreadGroup) release() { for _, it := range its { it.DestroyTimer() } - tg.mounts.DecRef() + if tg.mounts != nil { + tg.mounts.DecRef() + } } // forEachChildThreadGroupLocked indicates over all child ThreadGroups. diff --git a/pkg/sentry/loader/BUILD b/pkg/sentry/loader/BUILD index 23790378a..c6aa65f28 100644 --- a/pkg/sentry/loader/BUILD +++ b/pkg/sentry/loader/BUILD @@ -33,6 +33,7 @@ go_library( "//pkg/sentry/fs", "//pkg/sentry/fs/anon", "//pkg/sentry/fs/fsutil", + "//pkg/sentry/fsbridge", "//pkg/sentry/kernel/auth", "//pkg/sentry/limits", "//pkg/sentry/memmap", @@ -40,6 +41,7 @@ go_library( "//pkg/sentry/pgalloc", "//pkg/sentry/uniqueid", "//pkg/sentry/usage", + "//pkg/sentry/vfs", "//pkg/syserr", "//pkg/syserror", "//pkg/usermem", diff --git a/pkg/sentry/loader/elf.go b/pkg/sentry/loader/elf.go index 122ed05c2..616fafa2c 100644 --- a/pkg/sentry/loader/elf.go +++ b/pkg/sentry/loader/elf.go @@ -27,7 +27,7 @@ import ( "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/fsbridge" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/mm" @@ -97,11 +97,11 @@ type elfInfo struct { // accepts from the ELF, and it doesn't parse unnecessary parts of the file. // // ctx may be nil if f does not need it. -func parseHeader(ctx context.Context, f *fs.File) (elfInfo, error) { +func parseHeader(ctx context.Context, f fsbridge.File) (elfInfo, error) { // Check ident first; it will tell us the endianness of the rest of the // structs. var ident [elf.EI_NIDENT]byte - _, err := readFull(ctx, f, usermem.BytesIOSequence(ident[:]), 0) + _, err := f.ReadFull(ctx, usermem.BytesIOSequence(ident[:]), 0) if err != nil { log.Infof("Error reading ELF ident: %v", err) // The entire ident array always exists. @@ -137,7 +137,7 @@ func parseHeader(ctx context.Context, f *fs.File) (elfInfo, error) { var hdr elf.Header64 hdrBuf := make([]byte, header64Size) - _, err = readFull(ctx, f, usermem.BytesIOSequence(hdrBuf), 0) + _, err = f.ReadFull(ctx, usermem.BytesIOSequence(hdrBuf), 0) if err != nil { log.Infof("Error reading ELF header: %v", err) // The entire header always exists. @@ -187,7 +187,7 @@ func parseHeader(ctx context.Context, f *fs.File) (elfInfo, error) { } phdrBuf := make([]byte, totalPhdrSize) - _, err = readFull(ctx, f, usermem.BytesIOSequence(phdrBuf), int64(hdr.Phoff)) + _, err = f.ReadFull(ctx, usermem.BytesIOSequence(phdrBuf), int64(hdr.Phoff)) if err != nil { log.Infof("Error reading ELF phdrs: %v", err) // If phdrs were specified, they should all exist. @@ -227,7 +227,7 @@ func parseHeader(ctx context.Context, f *fs.File) (elfInfo, error) { // mapSegment maps a phdr into the Task. offset is the offset to apply to // phdr.Vaddr. -func mapSegment(ctx context.Context, m *mm.MemoryManager, f *fs.File, phdr *elf.ProgHeader, offset usermem.Addr) error { +func mapSegment(ctx context.Context, m *mm.MemoryManager, f fsbridge.File, phdr *elf.ProgHeader, offset usermem.Addr) error { // We must make a page-aligned mapping. adjust := usermem.Addr(phdr.Vaddr).PageOffset() @@ -395,7 +395,7 @@ type loadedELF struct { // // Preconditions: // * f is an ELF file -func loadParsedELF(ctx context.Context, m *mm.MemoryManager, f *fs.File, info elfInfo, sharedLoadOffset usermem.Addr) (loadedELF, error) { +func loadParsedELF(ctx context.Context, m *mm.MemoryManager, f fsbridge.File, info elfInfo, sharedLoadOffset usermem.Addr) (loadedELF, error) { first := true var start, end usermem.Addr var interpreter string @@ -431,7 +431,7 @@ func loadParsedELF(ctx context.Context, m *mm.MemoryManager, f *fs.File, info el } path := make([]byte, phdr.Filesz) - _, err := readFull(ctx, f, usermem.BytesIOSequence(path), int64(phdr.Off)) + _, err := f.ReadFull(ctx, usermem.BytesIOSequence(path), int64(phdr.Off)) if err != nil { // If an interpreter was specified, it should exist. ctx.Infof("Error reading PT_INTERP path: %v", err) @@ -564,7 +564,7 @@ func loadParsedELF(ctx context.Context, m *mm.MemoryManager, f *fs.File, info el // Preconditions: // * f is an ELF file // * f is the first ELF loaded into m -func loadInitialELF(ctx context.Context, m *mm.MemoryManager, fs *cpuid.FeatureSet, f *fs.File) (loadedELF, arch.Context, error) { +func loadInitialELF(ctx context.Context, m *mm.MemoryManager, fs *cpuid.FeatureSet, f fsbridge.File) (loadedELF, arch.Context, error) { info, err := parseHeader(ctx, f) if err != nil { ctx.Infof("Failed to parse initial ELF: %v", err) @@ -602,7 +602,7 @@ func loadInitialELF(ctx context.Context, m *mm.MemoryManager, fs *cpuid.FeatureS // // Preconditions: // * f is an ELF file -func loadInterpreterELF(ctx context.Context, m *mm.MemoryManager, f *fs.File, initial loadedELF) (loadedELF, error) { +func loadInterpreterELF(ctx context.Context, m *mm.MemoryManager, f fsbridge.File, initial loadedELF) (loadedELF, error) { info, err := parseHeader(ctx, f) if err != nil { if err == syserror.ENOEXEC { @@ -649,16 +649,14 @@ func loadELF(ctx context.Context, args LoadArgs) (loadedELF, arch.Context, error // Refresh the traversal limit. *args.RemainingTraversals = linux.MaxSymlinkTraversals args.Filename = bin.interpreter - d, i, err := openPath(ctx, args) + intFile, err := openPath(ctx, args) if err != nil { ctx.Infof("Error opening interpreter %s: %v", bin.interpreter, err) return loadedELF{}, nil, err } - defer i.DecRef() - // We don't need the Dirent. - d.DecRef() + defer intFile.DecRef() - interp, err = loadInterpreterELF(ctx, args.MemoryManager, i, bin) + interp, err = loadInterpreterELF(ctx, args.MemoryManager, intFile, bin) if err != nil { ctx.Infof("Error loading interpreter: %v", err) return loadedELF{}, nil, err diff --git a/pkg/sentry/loader/interpreter.go b/pkg/sentry/loader/interpreter.go index 098a45d36..3886b4d33 100644 --- a/pkg/sentry/loader/interpreter.go +++ b/pkg/sentry/loader/interpreter.go @@ -19,7 +19,7 @@ import ( "io" "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/fsbridge" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" ) @@ -37,9 +37,9 @@ const ( ) // parseInterpreterScript returns the interpreter path and argv. -func parseInterpreterScript(ctx context.Context, filename string, f *fs.File, argv []string) (newpath string, newargv []string, err error) { +func parseInterpreterScript(ctx context.Context, filename string, f fsbridge.File, argv []string) (newpath string, newargv []string, err error) { line := make([]byte, interpMaxLineLength) - n, err := readFull(ctx, f, usermem.BytesIOSequence(line), 0) + n, err := f.ReadFull(ctx, usermem.BytesIOSequence(line), 0) // Short read is OK. if err != nil && err != io.ErrUnexpectedEOF { if err == io.EOF { diff --git a/pkg/sentry/loader/loader.go b/pkg/sentry/loader/loader.go index 9a613d6b7..d6675b8f0 100644 --- a/pkg/sentry/loader/loader.go +++ b/pkg/sentry/loader/loader.go @@ -20,7 +20,6 @@ import ( "fmt" "io" "path" - "strings" "gvisor.dev/gvisor/pkg/abi" "gvisor.dev/gvisor/pkg/abi/linux" @@ -29,8 +28,10 @@ import ( "gvisor.dev/gvisor/pkg/rand" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/fsbridge" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/mm" + "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -41,16 +42,6 @@ type LoadArgs struct { // MemoryManager is the memory manager to load the executable into. MemoryManager *mm.MemoryManager - // Mounts is the mount namespace in which to look up Filename. - Mounts *fs.MountNamespace - - // Root is the root directory under which to look up Filename. - Root *fs.Dirent - - // WorkingDirectory is the working directory under which to look up - // Filename. - WorkingDirectory *fs.Dirent - // RemainingTraversals is the maximum number of symlinks to follow to // resolve Filename. This counter is passed by reference to keep it // updated throughout the call stack. @@ -65,7 +56,12 @@ type LoadArgs struct { // File is an open fs.File object of the executable. If File is not // nil, then File will be loaded and Filename will be ignored. - File *fs.File + // + // The caller is responsible for checking that the user can execute this file. + File fsbridge.File + + // Opener is used to open the executable file when 'File' is nil. + Opener fsbridge.Lookup // CloseOnExec indicates that the executable (or one of its parent // directories) was opened with O_CLOEXEC. If the executable is an @@ -106,103 +102,32 @@ func readFull(ctx context.Context, f *fs.File, dst usermem.IOSequence, offset in // installed in the Task FDTable. The caller takes ownership of both. // // args.Filename must be a readable, executable, regular file. -func openPath(ctx context.Context, args LoadArgs) (*fs.Dirent, *fs.File, error) { +func openPath(ctx context.Context, args LoadArgs) (fsbridge.File, error) { if args.Filename == "" { ctx.Infof("cannot open empty name") - return nil, nil, syserror.ENOENT - } - - var d *fs.Dirent - var err error - if args.ResolveFinal { - d, err = args.Mounts.FindInode(ctx, args.Root, args.WorkingDirectory, args.Filename, args.RemainingTraversals) - } else { - d, err = args.Mounts.FindLink(ctx, args.Root, args.WorkingDirectory, args.Filename, args.RemainingTraversals) - } - if err != nil { - return nil, nil, err - } - // Defer a DecRef for the sake of failure cases. - defer d.DecRef() - - if !args.ResolveFinal && fs.IsSymlink(d.Inode.StableAttr) { - return nil, nil, syserror.ELOOP - } - - if err := checkPermission(ctx, d); err != nil { - return nil, nil, err - } - - // If they claim it's a directory, then make sure. - // - // N.B. we reject directories below, but we must first reject - // non-directories passed as directories. - if strings.HasSuffix(args.Filename, "/") && !fs.IsDir(d.Inode.StableAttr) { - return nil, nil, syserror.ENOTDIR - } - - if err := checkIsRegularFile(ctx, d, args.Filename); err != nil { - return nil, nil, err - } - - f, err := d.Inode.GetFile(ctx, d, fs.FileFlags{Read: true}) - if err != nil { - return nil, nil, err - } - // Defer a DecRef for the sake of failure cases. - defer f.DecRef() - - if err := checkPread(ctx, f, args.Filename); err != nil { - return nil, nil, err - } - - d.IncRef() - f.IncRef() - return d, f, err -} - -// checkFile performs checks on a file to be executed. -func checkFile(ctx context.Context, f *fs.File, filename string) error { - if err := checkPermission(ctx, f.Dirent); err != nil { - return err - } - - if err := checkIsRegularFile(ctx, f.Dirent, filename); err != nil { - return err + return nil, syserror.ENOENT } - return checkPread(ctx, f, filename) -} - -// checkPermission checks whether the file is readable and executable. -func checkPermission(ctx context.Context, d *fs.Dirent) error { - perms := fs.PermMask{ - // TODO(gvisor.dev/issue/160): Linux requires only execute - // permission, not read. However, our backing filesystems may - // prevent us from reading the file without read permission. - // - // Additionally, a task with a non-readable executable has - // additional constraints on access via ptrace and procfs. - Read: true, - Execute: true, + // TODO(gvisor.dev/issue/160): Linux requires only execute permission, + // not read. However, our backing filesystems may prevent us from reading + // the file without read permission. Additionally, a task with a + // non-readable executable has additional constraints on access via + // ptrace and procfs. + opts := vfs.OpenOptions{ + Flags: linux.O_RDONLY, + FileExec: true, } - return d.Inode.CheckPermission(ctx, perms) + return args.Opener.OpenPath(ctx, args.Filename, opts, args.RemainingTraversals, args.ResolveFinal) } // checkIsRegularFile prevents us from trying to execute a directory, pipe, etc. -func checkIsRegularFile(ctx context.Context, d *fs.Dirent, filename string) error { - attr := d.Inode.StableAttr - if !fs.IsRegular(attr) { - ctx.Infof("%s is not regular: %v", filename, attr) - return syserror.EACCES +func checkIsRegularFile(ctx context.Context, file fsbridge.File, filename string) error { + t, err := file.Type(ctx) + if err != nil { + return err } - return nil -} - -// checkPread checks whether we can read the file at arbitrary offsets. -func checkPread(ctx context.Context, f *fs.File, filename string) error { - if !f.Flags().Pread { - ctx.Infof("%s cannot be read at an offset: %+v", filename, f.Flags()) + if t != linux.ModeRegular { + ctx.Infof("%q is not a regular file: %v", filename, t) return syserror.EACCES } return nil @@ -224,8 +149,10 @@ const ( maxLoaderAttempts = 6 ) -// loadExecutable loads an executable that is pointed to by args.File. If nil, -// the path args.Filename is resolved and loaded. If the executable is an +// loadExecutable loads an executable that is pointed to by args.File. The +// caller is responsible for checking that the user can execute this file. +// If nil, the path args.Filename is resolved and loaded (check that the user +// can execute this file is done here in this case). If the executable is an // interpreter script rather than an ELF, the binary of the corresponding // interpreter will be loaded. // @@ -234,37 +161,27 @@ const ( // * arch.Context matching the binary arch // * fs.Dirent of the binary file // * Possibly updated args.Argv -func loadExecutable(ctx context.Context, args LoadArgs) (loadedELF, arch.Context, *fs.Dirent, []string, error) { +func loadExecutable(ctx context.Context, args LoadArgs) (loadedELF, arch.Context, fsbridge.File, []string, error) { for i := 0; i < maxLoaderAttempts; i++ { - var ( - d *fs.Dirent - err error - ) if args.File == nil { - d, args.File, err = openPath(ctx, args) - // We will return d in the successful case, but defer a DecRef for the - // sake of intermediate loops and failure cases. - if d != nil { - defer d.DecRef() - } - if args.File != nil { - defer args.File.DecRef() + var err error + args.File, err = openPath(ctx, args) + if err != nil { + ctx.Infof("Error opening %s: %v", args.Filename, err) + return loadedELF{}, nil, nil, nil, err } + // Ensure file is release in case the code loops or errors out. + defer args.File.DecRef() } else { - d = args.File.Dirent - d.IncRef() - defer d.DecRef() - err = checkFile(ctx, args.File, args.Filename) - } - if err != nil { - ctx.Infof("Error opening %s: %v", args.Filename, err) - return loadedELF{}, nil, nil, nil, err + if err := checkIsRegularFile(ctx, args.File, args.Filename); err != nil { + return loadedELF{}, nil, nil, nil, err + } } // Check the header. Is this an ELF or interpreter script? var hdr [4]uint8 // N.B. We assume that reading from a regular file cannot block. - _, err = readFull(ctx, args.File, usermem.BytesIOSequence(hdr[:]), 0) + _, err := args.File.ReadFull(ctx, usermem.BytesIOSequence(hdr[:]), 0) // Allow unexpected EOF, as a valid executable could be only three bytes // (e.g., #!a). if err != nil && err != io.ErrUnexpectedEOF { @@ -281,9 +198,10 @@ func loadExecutable(ctx context.Context, args LoadArgs) (loadedELF, arch.Context ctx.Infof("Error loading ELF: %v", err) return loadedELF{}, nil, nil, nil, err } - // An ELF is always terminal. Hold on to d. - d.IncRef() - return loaded, ac, d, args.Argv, err + // An ELF is always terminal. Hold on to file. + args.File.IncRef() + return loaded, ac, args.File, args.Argv, err + case bytes.Equal(hdr[:2], []byte(interpreterScriptMagic)): if args.CloseOnExec { return loadedELF{}, nil, nil, nil, syserror.ENOENT @@ -295,6 +213,7 @@ func loadExecutable(ctx context.Context, args LoadArgs) (loadedELF, arch.Context } // Refresh the traversal limit for the interpreter. *args.RemainingTraversals = linux.MaxSymlinkTraversals + default: ctx.Infof("Unknown magic: %v", hdr) return loadedELF{}, nil, nil, nil, syserror.ENOEXEC @@ -317,11 +236,11 @@ func loadExecutable(ctx context.Context, args LoadArgs) (loadedELF, arch.Context // * Load is called on the Task goroutine. func Load(ctx context.Context, args LoadArgs, extraAuxv []arch.AuxEntry, vdso *VDSO) (abi.OS, arch.Context, string, *syserr.Error) { // Load the executable itself. - loaded, ac, d, newArgv, err := loadExecutable(ctx, args) + loaded, ac, file, newArgv, err := loadExecutable(ctx, args) if err != nil { return 0, nil, "", syserr.NewDynamic(fmt.Sprintf("Failed to load %s: %v", args.Filename, err), syserr.FromError(err).ToLinux()) } - defer d.DecRef() + defer file.DecRef() // Load the VDSO. vdsoAddr, err := loadVDSO(ctx, args.MemoryManager, vdso, loaded) @@ -390,7 +309,7 @@ func Load(ctx context.Context, args LoadArgs, extraAuxv []arch.AuxEntry, vdso *V m.SetEnvvStart(sl.EnvvStart) m.SetEnvvEnd(sl.EnvvEnd) m.SetAuxv(auxv) - m.SetExecutable(d) + m.SetExecutable(file) ac.SetIP(uintptr(loaded.entry)) ac.SetStack(uintptr(stack.Bottom)) diff --git a/pkg/sentry/loader/vdso.go b/pkg/sentry/loader/vdso.go index 52f446ed7..161b28c2c 100644 --- a/pkg/sentry/loader/vdso.go +++ b/pkg/sentry/loader/vdso.go @@ -27,6 +27,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/anon" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" + "gvisor.dev/gvisor/pkg/sentry/fsbridge" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/mm" "gvisor.dev/gvisor/pkg/sentry/pgalloc" @@ -69,6 +70,8 @@ type byteReader struct { var _ fs.FileOperations = (*byteReader)(nil) // newByteReaderFile creates a fake file to read data from. +// +// TODO(gvisor.dev/issue/1623): Convert to VFS2. func newByteReaderFile(ctx context.Context, data []byte) *fs.File { // Create a fake inode. inode := fs.NewInode( @@ -123,7 +126,7 @@ func (b *byteReader) Write(ctx context.Context, file *fs.File, src usermem.IOSeq // * PT_LOAD segments don't extend beyond the end of the file. // // ctx may be nil if f does not need it. -func validateVDSO(ctx context.Context, f *fs.File, size uint64) (elfInfo, error) { +func validateVDSO(ctx context.Context, f fsbridge.File, size uint64) (elfInfo, error) { info, err := parseHeader(ctx, f) if err != nil { log.Infof("Unable to parse VDSO header: %v", err) @@ -221,7 +224,7 @@ type VDSO struct { // PrepareVDSO validates the system VDSO and returns a VDSO, containing the // param page for updating by the kernel. func PrepareVDSO(ctx context.Context, mfp pgalloc.MemoryFileProvider) (*VDSO, error) { - vdsoFile := newByteReaderFile(ctx, vdsoBin) + vdsoFile := fsbridge.NewFSFile(newByteReaderFile(ctx, vdsoBin)) // First make sure the VDSO is valid. vdsoFile does not use ctx, so a // nil context can be passed. diff --git a/pkg/sentry/mm/BUILD b/pkg/sentry/mm/BUILD index e5729ced5..73591dab7 100644 --- a/pkg/sentry/mm/BUILD +++ b/pkg/sentry/mm/BUILD @@ -105,8 +105,8 @@ go_library( "//pkg/safecopy", "//pkg/safemem", "//pkg/sentry/arch", - "//pkg/sentry/fs", "//pkg/sentry/fs/proc/seqfile", + "//pkg/sentry/fsbridge", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/futex", "//pkg/sentry/kernel/shm", diff --git a/pkg/sentry/mm/metadata.go b/pkg/sentry/mm/metadata.go index f550acae0..6a49334f4 100644 --- a/pkg/sentry/mm/metadata.go +++ b/pkg/sentry/mm/metadata.go @@ -16,7 +16,7 @@ package mm import ( "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/fsbridge" "gvisor.dev/gvisor/pkg/usermem" ) @@ -132,7 +132,7 @@ func (mm *MemoryManager) SetAuxv(auxv arch.Auxv) { // // An additional reference will be taken in the case of a non-nil executable, // which must be released by the caller. -func (mm *MemoryManager) Executable() *fs.Dirent { +func (mm *MemoryManager) Executable() fsbridge.File { mm.metadataMu.Lock() defer mm.metadataMu.Unlock() @@ -147,15 +147,15 @@ func (mm *MemoryManager) Executable() *fs.Dirent { // SetExecutable sets the executable. // // This takes a reference on d. -func (mm *MemoryManager) SetExecutable(d *fs.Dirent) { +func (mm *MemoryManager) SetExecutable(file fsbridge.File) { mm.metadataMu.Lock() // Grab a new reference. - d.IncRef() + file.IncRef() // Set the executable. orig := mm.executable - mm.executable = d + mm.executable = file mm.metadataMu.Unlock() diff --git a/pkg/sentry/mm/mm.go b/pkg/sentry/mm/mm.go index 09e582dd3..637383c7a 100644 --- a/pkg/sentry/mm/mm.go +++ b/pkg/sentry/mm/mm.go @@ -37,7 +37,7 @@ package mm import ( "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/fsbridge" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/platform" @@ -215,7 +215,7 @@ type MemoryManager struct { // is not nil, it holds a reference on the Dirent. // // executable is protected by metadataMu. - executable *fs.Dirent + executable fsbridge.File // dumpability describes if and how this MemoryManager may be dumped to // userspace. diff --git a/pkg/sentry/strace/strace.go b/pkg/sentry/strace/strace.go index a796b2396..46cb2a1cc 100644 --- a/pkg/sentry/strace/strace.go +++ b/pkg/sentry/strace/strace.go @@ -141,6 +141,10 @@ func path(t *kernel.Task, addr usermem.Addr) string { } func fd(t *kernel.Task, fd int32) string { + if kernel.VFS2Enabled { + return fdVFS2(t, fd) + } + root := t.FSContext().RootDirectory() if root != nil { defer root.DecRef() @@ -169,6 +173,30 @@ func fd(t *kernel.Task, fd int32) string { return fmt.Sprintf("%#x %s", fd, name) } +func fdVFS2(t *kernel.Task, fd int32) string { + root := t.FSContext().RootDirectoryVFS2() + defer root.DecRef() + + vfsObj := root.Mount().Filesystem().VirtualFilesystem() + if fd == linux.AT_FDCWD { + wd := t.FSContext().WorkingDirectoryVFS2() + defer wd.DecRef() + + name, _ := vfsObj.PathnameWithDeleted(t, root, wd) + return fmt.Sprintf("AT_FDCWD %s", name) + } + + file := t.GetFileVFS2(fd) + if file == nil { + // Cast FD to uint64 to avoid printing negative hex. + return fmt.Sprintf("%#x (bad FD)", uint64(fd)) + } + defer file.DecRef() + + name, _ := vfsObj.PathnameWithDeleted(t, root, file.VirtualDentry()) + return fmt.Sprintf("%#x %s", fd, name) +} + func fdpair(t *kernel.Task, addr usermem.Addr) string { var fds [2]int32 _, err := t.CopyIn(addr, &fds) diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD index be16ee686..0d24fd3c4 100644 --- a/pkg/sentry/syscalls/linux/BUILD +++ b/pkg/sentry/syscalls/linux/BUILD @@ -74,6 +74,7 @@ go_library( "//pkg/sentry/fs/lock", "//pkg/sentry/fs/timerfd", "//pkg/sentry/fs/tmpfs", + "//pkg/sentry/fsbridge", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/epoll", diff --git a/pkg/sentry/syscalls/linux/sys_prctl.go b/pkg/sentry/syscalls/linux/sys_prctl.go index 98db32d77..9c6728530 100644 --- a/pkg/sentry/syscalls/linux/sys_prctl.go +++ b/pkg/sentry/syscalls/linux/sys_prctl.go @@ -20,6 +20,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/fsbridge" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/mm" @@ -135,7 +136,7 @@ func Prctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } // Set the underlying executable. - t.MemoryManager().SetExecutable(file.Dirent) + t.MemoryManager().SetExecutable(fsbridge.NewFSFile(file)) case linux.PR_SET_MM_AUXV, linux.PR_SET_MM_START_CODE, diff --git a/pkg/sentry/syscalls/linux/sys_thread.go b/pkg/sentry/syscalls/linux/sys_thread.go index 0c9e2255d..00915fdde 100644 --- a/pkg/sentry/syscalls/linux/sys_thread.go +++ b/pkg/sentry/syscalls/linux/sys_thread.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/fsbridge" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/sched" "gvisor.dev/gvisor/pkg/sentry/loader" @@ -119,7 +120,7 @@ func execveat(t *kernel.Task, dirFD int32, pathnameAddr, argvAddr, envvAddr user defer root.DecRef() var wd *fs.Dirent - var executable *fs.File + var executable fsbridge.File var closeOnExec bool if dirFD == linux.AT_FDCWD || path.IsAbs(pathname) { // Even if the pathname is absolute, we may still need the wd @@ -136,7 +137,15 @@ func execveat(t *kernel.Task, dirFD int32, pathnameAddr, argvAddr, envvAddr user closeOnExec = fdFlags.CloseOnExec if atEmptyPath && len(pathname) == 0 { - executable = f + // TODO(gvisor.dev/issue/160): Linux requires only execute permission, + // not read. However, our backing filesystems may prevent us from reading + // the file without read permission. Additionally, a task with a + // non-readable executable has additional constraints on access via + // ptrace and procfs. + if err := f.Dirent.Inode.CheckPermission(t, fs.PermMask{Read: true, Execute: true}); err != nil { + return 0, nil, err + } + executable = fsbridge.NewFSFile(f) } else { wd = f.Dirent wd.IncRef() @@ -152,9 +161,7 @@ func execveat(t *kernel.Task, dirFD int32, pathnameAddr, argvAddr, envvAddr user // Load the new TaskContext. remainingTraversals := uint(linux.MaxSymlinkTraversals) loadArgs := loader.LoadArgs{ - Mounts: t.MountNamespace(), - Root: root, - WorkingDirectory: wd, + Opener: fsbridge.NewFSLookup(t.MountNamespace(), root, wd), RemainingTraversals: &remainingTraversals, ResolveFinal: resolveFinal, Filename: pathname, diff --git a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go index c134714ee..e0ac32b33 100644 --- a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go +++ b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go @@ -22,4 +22,110 @@ import ( // Override syscall table to add syscalls implementations from this package. func Override(table map[uintptr]kernel.Syscall) { table[0] = syscalls.Supported("read", Read) + + // Remove syscalls that haven't been converted yet. It's better to get ENOSYS + // rather than a SIGSEGV deep in the stack. + delete(table, 1) // write + delete(table, 2) // open + delete(table, 3) // close + delete(table, 4) // stat + delete(table, 5) // fstat + delete(table, 6) // lstat + delete(table, 7) // poll + delete(table, 8) // lseek + delete(table, 9) // mmap + delete(table, 16) // ioctl + delete(table, 17) // pread64 + delete(table, 18) // pwrite64 + delete(table, 19) // readv + delete(table, 20) // writev + delete(table, 21) // access + delete(table, 22) // pipe + delete(table, 32) // dup + delete(table, 33) // dup2 + delete(table, 40) // sendfile + delete(table, 59) // execve + delete(table, 72) // fcntl + delete(table, 73) // flock + delete(table, 74) // fsync + delete(table, 75) // fdatasync + delete(table, 76) // truncate + delete(table, 77) // ftruncate + delete(table, 78) // getdents + delete(table, 79) // getcwd + delete(table, 80) // chdir + delete(table, 81) // fchdir + delete(table, 82) // rename + delete(table, 83) // mkdir + delete(table, 84) // rmdir + delete(table, 85) // creat + delete(table, 86) // link + delete(table, 87) // unlink + delete(table, 88) // symlink + delete(table, 89) // readlink + delete(table, 90) // chmod + delete(table, 91) // fchmod + delete(table, 92) // chown + delete(table, 93) // fchown + delete(table, 94) // lchown + delete(table, 133) // mknod + delete(table, 137) // statfs + delete(table, 138) // fstatfs + delete(table, 161) // chroot + delete(table, 162) // sync + delete(table, 165) // mount + delete(table, 166) // umount2 + delete(table, 172) // iopl + delete(table, 173) // ioperm + delete(table, 187) // readahead + delete(table, 188) // setxattr + delete(table, 189) // lsetxattr + delete(table, 190) // fsetxattr + delete(table, 191) // getxattr + delete(table, 192) // lgetxattr + delete(table, 193) // fgetxattr + delete(table, 206) // io_setup + delete(table, 207) // io_destroy + delete(table, 208) // io_getevents + delete(table, 209) // io_submit + delete(table, 210) // io_cancel + delete(table, 213) // epoll_create + delete(table, 214) // epoll_ctl_old + delete(table, 215) // epoll_wait_old + delete(table, 216) // remap_file_pages + delete(table, 217) // getdents64 + delete(table, 232) // epoll_wait + delete(table, 233) // epoll_ctl + delete(table, 253) // inotify_init + delete(table, 254) // inotify_add_watch + delete(table, 255) // inotify_rm_watch + delete(table, 257) // openat + delete(table, 258) // mkdirat + delete(table, 259) // mknodat + delete(table, 260) // fchownat + delete(table, 261) // futimesat + delete(table, 262) // fstatat + delete(table, 263) // unlinkat + delete(table, 264) // renameat + delete(table, 265) // linkat + delete(table, 266) // symlinkat + delete(table, 267) // readlinkat + delete(table, 268) // fchmodat + delete(table, 269) // faccessat + delete(table, 270) // pselect + delete(table, 271) // ppoll + delete(table, 285) // fallocate + delete(table, 291) // epoll_create1 + delete(table, 292) // dup3 + delete(table, 293) // pipe2 + delete(table, 294) // inotify_init1 + delete(table, 295) // preadv + delete(table, 296) // pwritev + delete(table, 306) // syncfs + delete(table, 316) // renameat2 + delete(table, 319) // memfd_create + delete(table, 322) // execveat + delete(table, 327) // preadv2 + delete(table, 328) // pwritev2 + delete(table, 332) // statx } diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index 14b39eb9d..0b4f18ab5 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -43,6 +43,7 @@ go_library( "//pkg/abi/linux", "//pkg/context", "//pkg/fspath", + "//pkg/log", "//pkg/sentry/arch", "//pkg/sentry/fs/lock", "//pkg/sentry/kernel/auth", diff --git a/pkg/sentry/vfs/context.go b/pkg/sentry/vfs/context.go index d97362b9a..82781e6d3 100644 --- a/pkg/sentry/vfs/context.go +++ b/pkg/sentry/vfs/context.go @@ -29,9 +29,10 @@ const ( CtxRoot ) -// MountNamespaceFromContext returns the MountNamespace used by ctx. It does -// not take a reference on the returned MountNamespace. If ctx is not -// associated with a MountNamespace, MountNamespaceFromContext returns nil. +// MountNamespaceFromContext returns the MountNamespace used by ctx. If ctx is +// not associated with a MountNamespace, MountNamespaceFromContext returns nil. +// +// A reference is taken on the returned MountNamespace. func MountNamespaceFromContext(ctx context.Context) *MountNamespace { if v := ctx.Value(CtxMountNamespace); v != nil { return v.(*MountNamespace) diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index 1fbb420f9..ad2c9fcf4 100644 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go @@ -114,6 +114,7 @@ type MountNamespace struct { func (vfs *VirtualFilesystem) NewMountNamespace(ctx context.Context, creds *auth.Credentials, source, fsTypeName string, opts *GetFilesystemOptions) (*MountNamespace, error) { rft := vfs.getFilesystemType(fsTypeName) if rft == nil { + ctx.Warningf("Unknown filesystem: %s", fsTypeName) return nil, syserror.ENODEV } fs, root, err := rft.fsType.GetFilesystem(ctx, vfs, creds, source, *opts) @@ -231,9 +232,12 @@ func (vfs *VirtualFilesystem) UmountAt(ctx context.Context, creds *auth.Credenti return syserror.EINVAL } vfs.mountMu.Lock() - if mntns := MountNamespaceFromContext(ctx); mntns != nil && mntns != vd.mount.ns { - vfs.mountMu.Unlock() - return syserror.EINVAL + if mntns := MountNamespaceFromContext(ctx); mntns != nil { + defer mntns.DecRef() + if mntns != vd.mount.ns { + vfs.mountMu.Unlock() + return syserror.EINVAL + } } // TODO(jamieliu): Linux special-cases umount of the caller's root, which diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go index fdf8be157..6af7fdac1 100644 --- a/pkg/sentry/vfs/options.go +++ b/pkg/sentry/vfs/options.go @@ -61,7 +61,7 @@ type MountOptions struct { type OpenOptions struct { // Flags contains access mode and flags as specified for open(2). // - // FilesystemImpls is reponsible for implementing the following flags: + // FilesystemImpls are responsible for implementing the following flags: // O_RDONLY, O_WRONLY, O_RDWR, O_APPEND, O_CREAT, O_DIRECT, O_DSYNC, // O_EXCL, O_NOATIME, O_NOCTTY, O_NONBLOCK, O_PATH, O_SYNC, O_TMPFILE, and // O_TRUNC. VFS is responsible for handling O_DIRECTORY, O_LARGEFILE, and diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 9629afee9..51deae313 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -393,7 +393,8 @@ func (vfs *VirtualFilesystem) OpenAt(ctx context.Context, creds *auth.Credential // be executed. return nil, syserror.EACCES } - if linux.FileMode(stat.Mode).FileType() != linux.ModeRegular { + if t := linux.FileMode(stat.Mode).FileType(); t != linux.ModeRegular { + ctx.Infof("%q is not a regular file: %v", pop.Path, t) return nil, syserror.EACCES } } @@ -743,6 +744,8 @@ func (vfs *VirtualFilesystem) SyncAllFilesystems(ctx context.Context) error { // VirtualDentry methods require that a reference is held on the VirtualDentry. // // VirtualDentry is analogous to Linux's struct path. +// +// +stateify savable type VirtualDentry struct { mount *Mount dentry *Dentry diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index 9f0d5d7af..239ca5302 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -795,16 +795,19 @@ func (l *Loader) executeAsync(args *control.ExecArgs) (kernel.ThreadID, error) { return 0, fmt.Errorf("container %q not started", args.ContainerID) } + // TODO(gvisor.dev/issue/1623): Add VFS2 support + // Get the container MountNamespace from the Task. tg.Leader().WithMuLocked(func(t *kernel.Task) { - // task.MountNamespace() does not take a ref, so we must do so - // ourselves. + // task.MountNamespace() does not take a ref, so we must do so ourselves. args.MountNamespace = t.MountNamespace() args.MountNamespace.IncRef() }) - defer args.MountNamespace.DecRef() + if args.MountNamespace != nil { + defer args.MountNamespace.DecRef() + } - // Add the HOME enviroment varible if it is not already set. + // Add the HOME environment variable if it is not already set. root := args.MountNamespace.Root() defer root.DecRef() ctx := fs.WithRoot(l.k.SupervisorContext(), root) -- cgit v1.2.3 From 3c26f5ecb0087337b1f194b6d429ce68f3af70eb Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Fri, 14 Feb 2020 12:07:08 -0800 Subject: Enable automated marshalling for struct stat. This requires fixing a few build issues for non-am64 platforms. PiperOrigin-RevId: 295196922 --- pkg/abi/linux/BUILD | 1 + pkg/abi/linux/file_amd64.go | 2 + pkg/abi/linux/file_arm64.go | 2 + pkg/abi/linux/time.go | 2 + pkg/sentry/kernel/BUILD | 1 + pkg/sentry/syscalls/linux/BUILD | 2 - pkg/sentry/syscalls/linux/sys_stat.go | 26 +++++++++- pkg/sentry/syscalls/linux/sys_stat_amd64.go | 75 ---------------------------- pkg/sentry/syscalls/linux/sys_stat_arm64.go | 77 ----------------------------- 9 files changed, 32 insertions(+), 156 deletions(-) delete mode 100644 pkg/sentry/syscalls/linux/sys_stat_amd64.go delete mode 100644 pkg/sentry/syscalls/linux/sys_stat_arm64.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/BUILD b/pkg/abi/linux/BUILD index 1f3c0c687..b7015367b 100644 --- a/pkg/abi/linux/BUILD +++ b/pkg/abi/linux/BUILD @@ -59,6 +59,7 @@ go_library( "wait.go", "xattr.go", ], + marshal = True, visibility = ["//visibility:public"], deps = [ "//pkg/abi", diff --git a/pkg/abi/linux/file_amd64.go b/pkg/abi/linux/file_amd64.go index 8693d49c8..6b72364ea 100644 --- a/pkg/abi/linux/file_amd64.go +++ b/pkg/abi/linux/file_amd64.go @@ -25,6 +25,8 @@ const ( ) // Stat represents struct stat. +// +// +marshal type Stat struct { Dev uint64 Ino uint64 diff --git a/pkg/abi/linux/file_arm64.go b/pkg/abi/linux/file_arm64.go index ea3adc5f5..6492c9038 100644 --- a/pkg/abi/linux/file_arm64.go +++ b/pkg/abi/linux/file_arm64.go @@ -25,6 +25,8 @@ const ( ) // Stat represents struct stat. +// +// +marshal type Stat struct { Dev uint64 Ino uint64 diff --git a/pkg/abi/linux/time.go b/pkg/abi/linux/time.go index 5c5a58cd4..e562b46d9 100644 --- a/pkg/abi/linux/time.go +++ b/pkg/abi/linux/time.go @@ -101,6 +101,8 @@ func NsecToTimeT(nsec int64) TimeT { } // Timespec represents struct timespec in . +// +// +marshal type Timespec struct { Sec int64 Nsec int64 diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD index 46306945f..beba29a09 100644 --- a/pkg/sentry/kernel/BUILD +++ b/pkg/sentry/kernel/BUILD @@ -201,6 +201,7 @@ go_library( "//pkg/tcpip/stack", "//pkg/usermem", "//pkg/waiter", + "//tools/go_marshal/marshal", ], ) diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD index 0d24fd3c4..c7883e68e 100644 --- a/pkg/sentry/syscalls/linux/BUILD +++ b/pkg/sentry/syscalls/linux/BUILD @@ -42,8 +42,6 @@ go_library( "sys_socket.go", "sys_splice.go", "sys_stat.go", - "sys_stat_amd64.go", - "sys_stat_arm64.go", "sys_sync.go", "sys_sysinfo.go", "sys_syslog.go", diff --git a/pkg/sentry/syscalls/linux/sys_stat.go b/pkg/sentry/syscalls/linux/sys_stat.go index c841abccb..8b66a9006 100644 --- a/pkg/sentry/syscalls/linux/sys_stat.go +++ b/pkg/sentry/syscalls/linux/sys_stat.go @@ -23,6 +23,24 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) +func statFromAttrs(t *kernel.Task, sattr fs.StableAttr, uattr fs.UnstableAttr) linux.Stat { + return linux.Stat{ + Dev: sattr.DeviceID, + Ino: sattr.InodeID, + Nlink: uattr.Links, + Mode: sattr.Type.LinuxType() | uint32(uattr.Perms.LinuxMode()), + UID: uint32(uattr.Owner.UID.In(t.UserNamespace()).OrOverflow()), + GID: uint32(uattr.Owner.GID.In(t.UserNamespace()).OrOverflow()), + Rdev: uint64(linux.MakeDeviceID(sattr.DeviceFileMajor, sattr.DeviceFileMinor)), + Size: uattr.Size, + Blksize: sattr.BlockSize, + Blocks: uattr.Usage / 512, + ATime: uattr.AccessTime.Timespec(), + MTime: uattr.ModificationTime.Timespec(), + CTime: uattr.StatusChangeTime.Timespec(), + } +} + // Stat implements linux syscall stat(2). func Stat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { addr := args[0].Pointer() @@ -112,7 +130,9 @@ func stat(t *kernel.Task, d *fs.Dirent, dirPath bool, statAddr usermem.Addr) err if err != nil { return err } - return copyOutStat(t, statAddr, d.Inode.StableAttr, uattr) + s := statFromAttrs(t, d.Inode.StableAttr, uattr) + _, err = s.CopyOut(t, statAddr) + return err } // fstat implements fstat for the given *fs.File. @@ -121,7 +141,9 @@ func fstat(t *kernel.Task, f *fs.File, statAddr usermem.Addr) error { if err != nil { return err } - return copyOutStat(t, statAddr, f.Dirent.Inode.StableAttr, uattr) + s := statFromAttrs(t, f.Dirent.Inode.StableAttr, uattr) + _, err = s.CopyOut(t, statAddr) + return err } // Statx implements linux syscall statx(2). diff --git a/pkg/sentry/syscalls/linux/sys_stat_amd64.go b/pkg/sentry/syscalls/linux/sys_stat_amd64.go deleted file mode 100644 index 75a567bd4..000000000 --- a/pkg/sentry/syscalls/linux/sys_stat_amd64.go +++ /dev/null @@ -1,75 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -//+build amd64 - -package linux - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/usermem" -) - -// copyOutStat copies the attributes (sattr, uattr) to the struct stat at -// address dst in t's address space. It encodes the stat struct to bytes -// manually, as stat() is a very common syscall for many applications, and -// t.CopyObjectOut has noticeable performance impact due to its many slice -// allocations and use of reflection. -func copyOutStat(t *kernel.Task, dst usermem.Addr, sattr fs.StableAttr, uattr fs.UnstableAttr) error { - b := t.CopyScratchBuffer(int(linux.SizeOfStat))[:0] - - // Dev (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(sattr.DeviceID)) - // Ino (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(sattr.InodeID)) - // Nlink (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uattr.Links) - // Mode (uint32) - b = binary.AppendUint32(b, usermem.ByteOrder, sattr.Type.LinuxType()|uint32(uattr.Perms.LinuxMode())) - // UID (uint32) - b = binary.AppendUint32(b, usermem.ByteOrder, uint32(uattr.Owner.UID.In(t.UserNamespace()).OrOverflow())) - // GID (uint32) - b = binary.AppendUint32(b, usermem.ByteOrder, uint32(uattr.Owner.GID.In(t.UserNamespace()).OrOverflow())) - // Padding (uint32) - b = binary.AppendUint32(b, usermem.ByteOrder, 0) - // Rdev (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(linux.MakeDeviceID(sattr.DeviceFileMajor, sattr.DeviceFileMinor))) - // Size (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(uattr.Size)) - // Blksize (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(sattr.BlockSize)) - // Blocks (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(uattr.Usage/512)) - - // ATime - atime := uattr.AccessTime.Timespec() - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(atime.Sec)) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(atime.Nsec)) - - // MTime - mtime := uattr.ModificationTime.Timespec() - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(mtime.Sec)) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(mtime.Nsec)) - - // CTime - ctime := uattr.StatusChangeTime.Timespec() - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(ctime.Sec)) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(ctime.Nsec)) - - _, err := t.CopyOutBytes(dst, b) - return err -} diff --git a/pkg/sentry/syscalls/linux/sys_stat_arm64.go b/pkg/sentry/syscalls/linux/sys_stat_arm64.go deleted file mode 100644 index 80c98d05c..000000000 --- a/pkg/sentry/syscalls/linux/sys_stat_arm64.go +++ /dev/null @@ -1,77 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -//+build arm64 - -package linux - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/usermem" -) - -// copyOutStat copies the attributes (sattr, uattr) to the struct stat at -// address dst in t's address space. It encodes the stat struct to bytes -// manually, as stat() is a very common syscall for many applications, and -// t.CopyObjectOut has noticeable performance impact due to its many slice -// allocations and use of reflection. -func copyOutStat(t *kernel.Task, dst usermem.Addr, sattr fs.StableAttr, uattr fs.UnstableAttr) error { - b := t.CopyScratchBuffer(int(linux.SizeOfStat))[:0] - - // Dev (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(sattr.DeviceID)) - // Ino (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(sattr.InodeID)) - // Mode (uint32) - b = binary.AppendUint32(b, usermem.ByteOrder, sattr.Type.LinuxType()|uint32(uattr.Perms.LinuxMode())) - // Nlink (uint32) - b = binary.AppendUint32(b, usermem.ByteOrder, uint32(uattr.Links)) - // UID (uint32) - b = binary.AppendUint32(b, usermem.ByteOrder, uint32(uattr.Owner.UID.In(t.UserNamespace()).OrOverflow())) - // GID (uint32) - b = binary.AppendUint32(b, usermem.ByteOrder, uint32(uattr.Owner.GID.In(t.UserNamespace()).OrOverflow())) - // Rdev (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(linux.MakeDeviceID(sattr.DeviceFileMajor, sattr.DeviceFileMinor))) - // Padding (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, 0) - // Size (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(uattr.Size)) - // Blksize (uint32) - b = binary.AppendUint32(b, usermem.ByteOrder, uint32(sattr.BlockSize)) - // Padding (uint32) - b = binary.AppendUint32(b, usermem.ByteOrder, 0) - // Blocks (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(uattr.Usage/512)) - - // ATime - atime := uattr.AccessTime.Timespec() - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(atime.Sec)) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(atime.Nsec)) - - // MTime - mtime := uattr.ModificationTime.Timespec() - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(mtime.Sec)) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(mtime.Nsec)) - - // CTime - ctime := uattr.StatusChangeTime.Timespec() - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(ctime.Sec)) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(ctime.Nsec)) - - _, err := t.CopyOutBytes(dst, b) - return err -} -- cgit v1.2.3 From e4c7f3e6f6c19f3259820a4c41b69e85c0454379 Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Fri, 14 Feb 2020 13:39:51 -0800 Subject: Inline vfs.VirtualFilesystem in Kernel struct This saves one pointer dereference per VFS access. Updates #1623 PiperOrigin-RevId: 295216176 --- pkg/sentry/control/proc.go | 2 +- pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go | 5 +++- pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go | 5 +++- pkg/sentry/fsimpl/ext/ext_test.go | 5 +++- pkg/sentry/fsimpl/kernfs/kernfs_test.go | 5 +++- pkg/sentry/fsimpl/proc/tasks_test.go | 6 ++-- pkg/sentry/fsimpl/sys/sys_test.go | 6 ++-- pkg/sentry/fsimpl/testutil/kernel.go | 7 +++-- pkg/sentry/fsimpl/tmpfs/benchmark_test.go | 10 +++++-- pkg/sentry/fsimpl/tmpfs/pipe_test.go | 5 +++- pkg/sentry/fsimpl/tmpfs/regular_file_test.go | 6 +++- pkg/sentry/kernel/kernel.go | 9 ++++-- pkg/sentry/vfs/dentry.go | 4 ++- pkg/sentry/vfs/device.go | 3 ++ pkg/sentry/vfs/file_description_impl_util_test.go | 10 +++++-- pkg/sentry/vfs/filesystem.go | 2 ++ pkg/sentry/vfs/filesystem_type.go | 1 + pkg/sentry/vfs/mount.go | 4 +++ pkg/sentry/vfs/mount_unsafe.go | 8 ++++-- pkg/sentry/vfs/vfs.go | 35 +++++++++++------------ 20 files changed, 94 insertions(+), 44 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/control/proc.go b/pkg/sentry/control/proc.go index 8973754c8..5457ba5e7 100644 --- a/pkg/sentry/control/proc.go +++ b/pkg/sentry/control/proc.go @@ -199,7 +199,7 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI } paths := fs.GetPath(initArgs.Envv) - vfsObj := proc.Kernel.VFS + vfsObj := proc.Kernel.VFS() file, err := ResolveExecutablePath(ctx, vfsObj, initArgs.WorkingDirectory, initArgs.Argv[0], paths) if err != nil { return nil, 0, nil, fmt.Errorf("error finding executable %q in PATH %v: %v", initArgs.Argv[0], paths, err) diff --git a/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go b/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go index 73308a2b5..b6d52c015 100644 --- a/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go +++ b/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go @@ -29,7 +29,10 @@ func TestDevtmpfs(t *testing.T) { ctx := contexttest.Context(t) creds := auth.CredentialsFromContext(ctx) - vfsObj := vfs.New() + vfsObj := &vfs.VirtualFilesystem{} + if err := vfsObj.Init(); err != nil { + t.Fatalf("VFS init: %v", err) + } // Register tmpfs just so that we can have a root filesystem that isn't // devtmpfs. vfsObj.MustRegisterFilesystemType("tmpfs", tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ diff --git a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go index 2015a8871..89caee3df 100644 --- a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go +++ b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go @@ -52,7 +52,10 @@ func setUp(b *testing.B, imagePath string) (context.Context, *vfs.VirtualFilesys creds := auth.CredentialsFromContext(ctx) // Create VFS. - vfsObj := vfs.New() + vfsObj := &vfs.VirtualFilesystem{} + if err := vfsObj.Init(); err != nil { + return nil, nil, nil, nil, err + } vfsObj.MustRegisterFilesystemType("extfs", ext.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, }) diff --git a/pkg/sentry/fsimpl/ext/ext_test.go b/pkg/sentry/fsimpl/ext/ext_test.go index 05f992826..ef6127f3c 100644 --- a/pkg/sentry/fsimpl/ext/ext_test.go +++ b/pkg/sentry/fsimpl/ext/ext_test.go @@ -65,7 +65,10 @@ func setUp(t *testing.T, imagePath string) (context.Context, *vfs.VirtualFilesys creds := auth.CredentialsFromContext(ctx) // Create VFS. - vfsObj := vfs.New() + vfsObj := &vfs.VirtualFilesystem{} + if err := vfsObj.Init(); err != nil { + t.Fatalf("VFS init: %v", err) + } vfsObj.MustRegisterFilesystemType("extfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, }) diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index 96a16e654..0459fb305 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -45,7 +45,10 @@ type RootDentryFn func(*auth.Credentials, *filesystem) *kernfs.Dentry func newTestSystem(t *testing.T, rootFn RootDentryFn) *testutil.System { ctx := contexttest.Context(t) creds := auth.CredentialsFromContext(ctx) - v := vfs.New() + v := &vfs.VirtualFilesystem{} + if err := v.Init(); err != nil { + t.Fatalf("VFS init: %v", err) + } v.MustRegisterFilesystemType("testfs", &fsType{rootFn: rootFn}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, }) diff --git a/pkg/sentry/fsimpl/proc/tasks_test.go b/pkg/sentry/fsimpl/proc/tasks_test.go index 96c72cbc9..c5d531fe0 100644 --- a/pkg/sentry/fsimpl/proc/tasks_test.go +++ b/pkg/sentry/fsimpl/proc/tasks_test.go @@ -90,7 +90,7 @@ func setup(t *testing.T) *testutil.System { ctx := k.SupervisorContext() creds := auth.CredentialsFromContext(ctx) - k.VFS.MustRegisterFilesystemType(Name, &FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + k.VFS().MustRegisterFilesystemType(Name, &FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, }) fsOpts := vfs.GetFilesystemOptions{ @@ -101,11 +101,11 @@ func setup(t *testing.T) *testutil.System { }, }, } - mntns, err := k.VFS.NewMountNamespace(ctx, creds, "", Name, &fsOpts) + mntns, err := k.VFS().NewMountNamespace(ctx, creds, "", Name, &fsOpts) if err != nil { t.Fatalf("NewMountNamespace(): %v", err) } - return testutil.NewSystem(ctx, t, k.VFS, mntns) + return testutil.NewSystem(ctx, t, k.VFS(), mntns) } func TestTasksEmpty(t *testing.T) { diff --git a/pkg/sentry/fsimpl/sys/sys_test.go b/pkg/sentry/fsimpl/sys/sys_test.go index 5d1ba5867..4b3602d47 100644 --- a/pkg/sentry/fsimpl/sys/sys_test.go +++ b/pkg/sentry/fsimpl/sys/sys_test.go @@ -34,15 +34,15 @@ func newTestSystem(t *testing.T) *testutil.System { } ctx := k.SupervisorContext() creds := auth.CredentialsFromContext(ctx) - k.VFS.MustRegisterFilesystemType(sys.Name, sys.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + k.VFS().MustRegisterFilesystemType(sys.Name, sys.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, }) - mns, err := k.VFS.NewMountNamespace(ctx, creds, "", sys.Name, &vfs.GetFilesystemOptions{}) + mns, err := k.VFS().NewMountNamespace(ctx, creds, "", sys.Name, &vfs.GetFilesystemOptions{}) if err != nil { t.Fatalf("Failed to create new mount namespace: %v", err) } - return testutil.NewSystem(ctx, t, k.VFS, mns) + return testutil.NewSystem(ctx, t, k.VFS(), mns) } func TestReadCPUFile(t *testing.T) { diff --git a/pkg/sentry/fsimpl/testutil/kernel.go b/pkg/sentry/fsimpl/testutil/kernel.go index a91b3ec4d..d0be32e72 100644 --- a/pkg/sentry/fsimpl/testutil/kernel.go +++ b/pkg/sentry/fsimpl/testutil/kernel.go @@ -102,12 +102,13 @@ func Boot() (*kernel.Kernel, error) { kernel.VFS2Enabled = true - vfsObj := vfs.New() - vfsObj.MustRegisterFilesystemType(tmpfs.Name, &tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + if err := k.VFS().Init(); err != nil { + return nil, fmt.Errorf("VFS init: %v", err) + } + k.VFS().MustRegisterFilesystemType(tmpfs.Name, &tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, AllowUserList: true, }) - k.VFS = vfsObj ls, err := limits.NewLinuxLimitSet() if err != nil { diff --git a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go index 9fce5e4b4..383133e44 100644 --- a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go +++ b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go @@ -175,7 +175,10 @@ func BenchmarkVFS2MemfsStat(b *testing.B) { creds := auth.CredentialsFromContext(ctx) // Create VFS. - vfsObj := vfs.New() + vfsObj := vfs.VirtualFilesystem{} + if err := vfsObj.Init(); err != nil { + b.Fatalf("VFS init: %v", err) + } vfsObj.MustRegisterFilesystemType("tmpfs", tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, }) @@ -366,7 +369,10 @@ func BenchmarkVFS2MemfsMountStat(b *testing.B) { creds := auth.CredentialsFromContext(ctx) // Create VFS. - vfsObj := vfs.New() + vfsObj := vfs.VirtualFilesystem{} + if err := vfsObj.Init(); err != nil { + b.Fatalf("VFS init: %v", err) + } vfsObj.MustRegisterFilesystemType("tmpfs", tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, }) diff --git a/pkg/sentry/fsimpl/tmpfs/pipe_test.go b/pkg/sentry/fsimpl/tmpfs/pipe_test.go index 5ee7f2a72..1614f2c39 100644 --- a/pkg/sentry/fsimpl/tmpfs/pipe_test.go +++ b/pkg/sentry/fsimpl/tmpfs/pipe_test.go @@ -151,7 +151,10 @@ func setup(t *testing.T) (context.Context, *auth.Credentials, *vfs.VirtualFilesy creds := auth.CredentialsFromContext(ctx) // Create VFS. - vfsObj := vfs.New() + vfsObj := &vfs.VirtualFilesystem{} + if err := vfsObj.Init(); err != nil { + t.Fatalf("VFS init: %v", err) + } vfsObj.MustRegisterFilesystemType("tmpfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, }) diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go index e9f71e334..0399725cf 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go @@ -40,7 +40,11 @@ var nextFileID int64 func newTmpfsRoot(ctx context.Context) (*vfs.VirtualFilesystem, vfs.VirtualDentry, func(), error) { creds := auth.CredentialsFromContext(ctx) - vfsObj := vfs.New() + vfsObj := &vfs.VirtualFilesystem{} + if err := vfsObj.Init(); err != nil { + return nil, vfs.VirtualDentry{}, nil, fmt.Errorf("VFS init: %v", err) + } + vfsObj.MustRegisterFilesystemType("tmpfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, }) diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 2665f057c..ea21af33f 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -246,7 +246,7 @@ type Kernel struct { SpecialOpts // VFS keeps the filesystem state used across the kernel. - VFS *vfs.VirtualFilesystem + vfs vfs.VirtualFilesystem } // InitKernelArgs holds arguments to Init. @@ -815,7 +815,7 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, FollowFinalSymlink: true, } var err error - wd, err = k.VFS.GetDentryAt(ctx, args.Credentials, &pop, &vfs.GetDentryOptions{ + wd, err = k.VFS().GetDentryAt(ctx, args.Credentials, &pop, &vfs.GetDentryOptions{ CheckSearchable: true, }) if err != nil { @@ -1506,3 +1506,8 @@ func (k *Kernel) EmitUnimplementedEvent(ctx context.Context) { Registers: t.Arch().StateData().Proto(), }) } + +// VFS returns the virtual filesystem for the kernel. +func (k *Kernel) VFS() *vfs.VirtualFilesystem { + return &k.vfs +} diff --git a/pkg/sentry/vfs/dentry.go b/pkg/sentry/vfs/dentry.go index 486a76475..35b208721 100644 --- a/pkg/sentry/vfs/dentry.go +++ b/pkg/sentry/vfs/dentry.go @@ -71,6 +71,8 @@ import ( // lifetime. Dentry reference counts only indicate the extent to which VFS // requires Dentries to exist; Filesystems may elect to cache or discard // Dentries with zero references. +// +// +stateify savable type Dentry struct { // parent is this Dentry's parent in this Filesystem. If this Dentry is // independent, parent is nil. @@ -89,7 +91,7 @@ type Dentry struct { children map[string]*Dentry // mu synchronizes disowning and mounting over this Dentry. - mu sync.Mutex + mu sync.Mutex `state:"nosave"` // impl is the DentryImpl associated with this Dentry. impl is immutable. // This should be the last field in Dentry. diff --git a/pkg/sentry/vfs/device.go b/pkg/sentry/vfs/device.go index 3af2aa58d..bda5576fa 100644 --- a/pkg/sentry/vfs/device.go +++ b/pkg/sentry/vfs/device.go @@ -56,6 +56,7 @@ type Device interface { Open(ctx context.Context, mnt *Mount, d *Dentry, opts OpenOptions) (*FileDescription, error) } +// +stateify savable type registeredDevice struct { dev Device opts RegisterDeviceOptions @@ -63,6 +64,8 @@ type registeredDevice struct { // RegisterDeviceOptions contains options to // VirtualFilesystem.RegisterDevice(). +// +// +stateify savable type RegisterDeviceOptions struct { // GroupName is the name shown for this device registration in // /proc/devices. If GroupName is empty, this registration will not be diff --git a/pkg/sentry/vfs/file_description_impl_util_test.go b/pkg/sentry/vfs/file_description_impl_util_test.go index 8fa26418e..3a75d4d62 100644 --- a/pkg/sentry/vfs/file_description_impl_util_test.go +++ b/pkg/sentry/vfs/file_description_impl_util_test.go @@ -107,7 +107,10 @@ func (fd *testFD) SetStat(ctx context.Context, opts SetStatOptions) error { func TestGenCountFD(t *testing.T) { ctx := contexttest.Context(t) - vfsObj := New() // vfs.New() + vfsObj := &VirtualFilesystem{} + if err := vfsObj.Init(); err != nil { + t.Fatalf("VFS init: %v", err) + } fd := newTestFD(vfsObj, linux.O_RDWR, &genCount{}) defer fd.DecRef() @@ -162,7 +165,10 @@ func TestGenCountFD(t *testing.T) { func TestWritable(t *testing.T) { ctx := contexttest.Context(t) - vfsObj := New() // vfs.New() + vfsObj := &VirtualFilesystem{} + if err := vfsObj.Init(); err != nil { + t.Fatalf("VFS init: %v", err) + } fd := newTestFD(vfsObj, linux.O_RDWR, &storeData{data: "init"}) defer fd.DecRef() diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index a06a6caf3..556976d0b 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -29,6 +29,8 @@ import ( // Filesystem methods require that a reference is held. // // Filesystem is analogous to Linux's struct super_block. +// +// +stateify savable type Filesystem struct { // refs is the reference count. refs is accessed using atomic memory // operations. diff --git a/pkg/sentry/vfs/filesystem_type.go b/pkg/sentry/vfs/filesystem_type.go index c58b70728..bb9cada81 100644 --- a/pkg/sentry/vfs/filesystem_type.go +++ b/pkg/sentry/vfs/filesystem_type.go @@ -44,6 +44,7 @@ type GetFilesystemOptions struct { InternalData interface{} } +// +stateify savable type registeredFilesystemType struct { fsType FilesystemType opts RegisterFilesystemTypeOptions diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index ad2c9fcf4..9912df799 100644 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go @@ -38,6 +38,8 @@ import ( // // Mount is analogous to Linux's struct mount. (gVisor does not distinguish // between struct mount and struct vfsmount.) +// +// +stateify savable type Mount struct { // vfs, fs, and root are immutable. References are held on fs and root. // @@ -85,6 +87,8 @@ type Mount struct { // MountNamespace methods require that a reference is held. // // MountNamespace is analogous to Linux's struct mnt_namespace. +// +// +stateify savable type MountNamespace struct { // root is the MountNamespace's root mount. root is immutable. root *Mount diff --git a/pkg/sentry/vfs/mount_unsafe.go b/pkg/sentry/vfs/mount_unsafe.go index bd90d36c4..1fe766a44 100644 --- a/pkg/sentry/vfs/mount_unsafe.go +++ b/pkg/sentry/vfs/mount_unsafe.go @@ -64,6 +64,8 @@ func (mnt *Mount) storeKey(vd VirtualDentry) { // (provided mutation is sufficiently uncommon). // // mountTable.Init() must be called on new mountTables before use. +// +// +stateify savable type mountTable struct { // mountTable is implemented as a seqcount-protected hash table that // resolves collisions with linear probing, featuring Robin Hood insertion @@ -75,8 +77,8 @@ type mountTable struct { // intrinsics and inline assembly, limiting the performance of this // approach.) - seq sync.SeqCount - seed uint32 // for hashing keys + seq sync.SeqCount `state:"nosave"` + seed uint32 // for hashing keys // size holds both length (number of elements) and capacity (number of // slots): capacity is stored as its base-2 log (referred to as order) in @@ -89,7 +91,7 @@ type mountTable struct { // length and cap in separate uint32s) for ~free. size uint64 - slots unsafe.Pointer // []mountSlot; never nil after Init + slots unsafe.Pointer `state:"nosave"` // []mountSlot; never nil after Init } type mountSlot struct { diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 51deae313..8f29031b2 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -46,11 +46,13 @@ import ( // // There is no analogue to the VirtualFilesystem type in Linux, as the // equivalent state in Linux is global. +// +// +stateify savable type VirtualFilesystem struct { // mountMu serializes mount mutations. // // mountMu is analogous to Linux's namespace_sem. - mountMu sync.Mutex + mountMu sync.Mutex `state:"nosave"` // mounts maps (mount parent, mount point) pairs to mounts. (Since mounts // are uniquely namespaced, including mount parent in the key correctly @@ -89,44 +91,42 @@ type VirtualFilesystem struct { // devices contains all registered Devices. devices is protected by // devicesMu. - devicesMu sync.RWMutex + devicesMu sync.RWMutex `state:"nosave"` devices map[devTuple]*registeredDevice // anonBlockDevMinor contains all allocated anonymous block device minor // numbers. anonBlockDevMinorNext is a lower bound for the smallest // unallocated anonymous block device number. anonBlockDevMinorNext and // anonBlockDevMinor are protected by anonBlockDevMinorMu. - anonBlockDevMinorMu sync.Mutex + anonBlockDevMinorMu sync.Mutex `state:"nosave"` anonBlockDevMinorNext uint32 anonBlockDevMinor map[uint32]struct{} // fsTypes contains all registered FilesystemTypes. fsTypes is protected by // fsTypesMu. - fsTypesMu sync.RWMutex + fsTypesMu sync.RWMutex `state:"nosave"` fsTypes map[string]*registeredFilesystemType // filesystems contains all Filesystems. filesystems is protected by // filesystemsMu. - filesystemsMu sync.Mutex + filesystemsMu sync.Mutex `state:"nosave"` filesystems map[*Filesystem]struct{} } -// New returns a new VirtualFilesystem with no mounts or FilesystemTypes. -func New() *VirtualFilesystem { - vfs := &VirtualFilesystem{ - mountpoints: make(map[*Dentry]map[*Mount]struct{}), - devices: make(map[devTuple]*registeredDevice), - anonBlockDevMinorNext: 1, - anonBlockDevMinor: make(map[uint32]struct{}), - fsTypes: make(map[string]*registeredFilesystemType), - filesystems: make(map[*Filesystem]struct{}), - } +// Init initializes a new VirtualFilesystem with no mounts or FilesystemTypes. +func (vfs *VirtualFilesystem) Init() error { + vfs.mountpoints = make(map[*Dentry]map[*Mount]struct{}) + vfs.devices = make(map[devTuple]*registeredDevice) + vfs.anonBlockDevMinorNext = 1 + vfs.anonBlockDevMinor = make(map[uint32]struct{}) + vfs.fsTypes = make(map[string]*registeredFilesystemType) + vfs.filesystems = make(map[*Filesystem]struct{}) vfs.mounts.Init() // Construct vfs.anonMount. anonfsDevMinor, err := vfs.GetAnonBlockDevMinor() if err != nil { - panic(fmt.Sprintf("VirtualFilesystem.GetAnonBlockDevMinor() failed during VirtualFilesystem construction: %v", err)) + return err } anonfs := anonFilesystem{ devMinor: anonfsDevMinor, @@ -137,8 +137,7 @@ func New() *VirtualFilesystem { fs: &anonfs.vfsfs, refs: 1, } - - return vfs + return nil } // PathOperation specifies the path operated on by a VFS method. -- cgit v1.2.3 From 87bc2834c97a958d0762833fe8db749ccc6d5d50 Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Fri, 14 Feb 2020 14:23:35 -0800 Subject: Enable automated marshalling for RSeqCriticalSection. PiperOrigin-RevId: 295226468 --- pkg/sentry/kernel/rseq.go | 16 ++-------------- 1 file changed, 2 insertions(+), 14 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/rseq.go b/pkg/sentry/kernel/rseq.go index efebfd872..18416643b 100644 --- a/pkg/sentry/kernel/rseq.go +++ b/pkg/sentry/kernel/rseq.go @@ -303,26 +303,14 @@ func (t *Task) rseqAddrInterrupt() { return } - buf = t.CopyScratchBuffer(linux.SizeOfRSeqCriticalSection) - if _, err := t.CopyInBytes(critAddr, buf); err != nil { + var cs linux.RSeqCriticalSection + if _, err := cs.CopyIn(t, critAddr); err != nil { t.Debugf("Failed to copy critical section from %#x for rseq: %v", critAddr, err) t.forceSignal(linux.SIGSEGV, false /* unconditional */) t.SendSignal(SignalInfoPriv(linux.SIGSEGV)) return } - // Manually marshal RSeqCriticalSection as this is in the hot path when - // rseq is enabled. It must be as fast as possible. - // - // TODO(b/130243041): Replace with go_marshal. - cs := linux.RSeqCriticalSection{ - Version: usermem.ByteOrder.Uint32(buf[0:4]), - Flags: usermem.ByteOrder.Uint32(buf[4:8]), - Start: usermem.ByteOrder.Uint64(buf[8:16]), - PostCommitOffset: usermem.ByteOrder.Uint64(buf[16:24]), - Abort: usermem.ByteOrder.Uint64(buf[24:32]), - } - if cs.Version != 0 { t.Debugf("Unknown version in %+v", cs) t.forceSignal(linux.SIGSEGV, false /* unconditional */) -- cgit v1.2.3 From 3557b2665198b57c04924ad4be8dbf9e42cedf71 Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Fri, 14 Feb 2020 14:39:40 -0800 Subject: Allow vfs.IterDirentsCallback.Handle() to return an error. This is easier than storing errors from e.g. CopyOut in the callback. PiperOrigin-RevId: 295230021 --- pkg/sentry/fsimpl/ext/directory.go | 6 +++--- pkg/sentry/fsimpl/ext/ext_test.go | 4 ++-- pkg/sentry/fsimpl/gofer/directory.go | 4 ++-- pkg/sentry/fsimpl/kernfs/fd_impl_util.go | 12 ++++++------ pkg/sentry/fsimpl/proc/subtasks.go | 4 ++-- pkg/sentry/fsimpl/proc/tasks.go | 12 ++++++------ pkg/sentry/fsimpl/testutil/testutil.go | 4 ++-- pkg/sentry/fsimpl/tmpfs/directory.go | 18 +++++++++--------- pkg/sentry/vfs/file_description.go | 10 +++++----- 9 files changed, 37 insertions(+), 37 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/ext/directory.go b/pkg/sentry/fsimpl/ext/directory.go index ebb72b75e..bd6ede995 100644 --- a/pkg/sentry/fsimpl/ext/directory.go +++ b/pkg/sentry/fsimpl/ext/directory.go @@ -188,14 +188,14 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba childType = fs.ToInodeType(childInode.diskInode.Mode().FileType()) } - if !cb.Handle(vfs.Dirent{ + if err := cb.Handle(vfs.Dirent{ Name: child.diskDirent.FileName(), Type: fs.ToDirentType(childType), Ino: uint64(child.diskDirent.Inode()), NextOff: fd.off + 1, - }) { + }); err != nil { dir.childList.InsertBefore(child, fd.iter) - return nil + return err } fd.off++ } diff --git a/pkg/sentry/fsimpl/ext/ext_test.go b/pkg/sentry/fsimpl/ext/ext_test.go index ef6127f3c..29bb73765 100644 --- a/pkg/sentry/fsimpl/ext/ext_test.go +++ b/pkg/sentry/fsimpl/ext/ext_test.go @@ -499,9 +499,9 @@ func newIterDirentCb() *iterDirentsCb { } // Handle implements vfs.IterDirentsCallback.Handle. -func (cb *iterDirentsCb) Handle(dirent vfs.Dirent) bool { +func (cb *iterDirentsCb) Handle(dirent vfs.Dirent) error { cb.dirents = append(cb.dirents, dirent) - return true + return nil } // TestIterDirents tests the FileDescriptionImpl.IterDirents functionality. diff --git a/pkg/sentry/fsimpl/gofer/directory.go b/pkg/sentry/fsimpl/gofer/directory.go index 6d4ebc2bf..5dbfc6250 100644 --- a/pkg/sentry/fsimpl/gofer/directory.go +++ b/pkg/sentry/fsimpl/gofer/directory.go @@ -65,8 +65,8 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba } for fd.off < int64(len(fd.dirents)) { - if !cb.Handle(fd.dirents[fd.off]) { - return nil + if err := cb.Handle(fd.dirents[fd.off]); err != nil { + return err } fd.off++ } diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index eda781155..5650512e0 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -116,8 +116,8 @@ func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirent Ino: stat.Ino, NextOff: 1, } - if !cb.Handle(dirent) { - return nil + if err := cb.Handle(dirent); err != nil { + return err } fd.off++ } @@ -132,8 +132,8 @@ func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirent Ino: stat.Ino, NextOff: 2, } - if !cb.Handle(dirent) { - return nil + if err := cb.Handle(dirent); err != nil { + return err } fd.off++ } @@ -153,8 +153,8 @@ func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirent Ino: stat.Ino, NextOff: fd.off + 1, } - if !cb.Handle(dirent) { - return nil + if err := cb.Handle(dirent); err != nil { + return err } fd.off++ } diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go index 102af0e93..f3f4e49b4 100644 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -105,8 +105,8 @@ func (i *subtasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallb Ino: i.inoGen.NextIno(), NextOff: offset + 1, } - if !cb.Handle(dirent) { - return offset, nil + if err := cb.Handle(dirent); err != nil { + return offset, err } offset++ } diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index ebe21630c..ce08a7d53 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -151,8 +151,8 @@ func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback Ino: i.inoGen.NextIno(), NextOff: offset + 1, } - if !cb.Handle(dirent) { - return offset, nil + if err := cb.Handle(dirent); err != nil { + return offset, err } offset++ } @@ -163,8 +163,8 @@ func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback Ino: i.inoGen.NextIno(), NextOff: offset + 1, } - if !cb.Handle(dirent) { - return offset, nil + if err := cb.Handle(dirent); err != nil { + return offset, err } offset++ } @@ -196,8 +196,8 @@ func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback Ino: i.inoGen.NextIno(), NextOff: FIRST_PROCESS_ENTRY + 2 + int64(tid) + 1, } - if !cb.Handle(dirent) { - return offset, nil + if err := cb.Handle(dirent); err != nil { + return offset, err } offset++ } diff --git a/pkg/sentry/fsimpl/testutil/testutil.go b/pkg/sentry/fsimpl/testutil/testutil.go index b97e3534a..e16808c63 100644 --- a/pkg/sentry/fsimpl/testutil/testutil.go +++ b/pkg/sentry/fsimpl/testutil/testutil.go @@ -226,7 +226,7 @@ func (d *DirentCollector) SkipDotsChecks(value bool) { } // Handle implements vfs.IterDirentsCallback.Handle. -func (d *DirentCollector) Handle(dirent vfs.Dirent) bool { +func (d *DirentCollector) Handle(dirent vfs.Dirent) error { d.mu.Lock() if d.dirents == nil { d.dirents = make(map[string]*vfs.Dirent) @@ -234,7 +234,7 @@ func (d *DirentCollector) Handle(dirent vfs.Dirent) bool { d.order = append(d.order, &dirent) d.dirents[dirent.Name] = &dirent d.mu.Unlock() - return true + return nil } // Count returns the number of dirents currently in the collector. diff --git a/pkg/sentry/fsimpl/tmpfs/directory.go b/pkg/sentry/fsimpl/tmpfs/directory.go index dc0d27cf9..b4380af38 100644 --- a/pkg/sentry/fsimpl/tmpfs/directory.go +++ b/pkg/sentry/fsimpl/tmpfs/directory.go @@ -74,25 +74,25 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba defer fs.mu.Unlock() if fd.off == 0 { - if !cb.Handle(vfs.Dirent{ + if err := cb.Handle(vfs.Dirent{ Name: ".", Type: linux.DT_DIR, Ino: vfsd.Impl().(*dentry).inode.ino, NextOff: 1, - }) { - return nil + }); err != nil { + return err } fd.off++ } if fd.off == 1 { parentInode := vfsd.ParentOrSelf().Impl().(*dentry).inode - if !cb.Handle(vfs.Dirent{ + if err := cb.Handle(vfs.Dirent{ Name: "..", Type: parentInode.direntType(), Ino: parentInode.ino, NextOff: 2, - }) { - return nil + }); err != nil { + return err } fd.off++ } @@ -111,14 +111,14 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba for child != nil { // Skip other directoryFD iterators. if child.inode != nil { - if !cb.Handle(vfs.Dirent{ + if err := cb.Handle(vfs.Dirent{ Name: child.vfsd.Name(), Type: child.inode.direntType(), Ino: child.inode.ino, NextOff: fd.off + 1, - }) { + }); err != nil { dir.childList.InsertBefore(child, fd.iter) - return nil + return err } fd.off++ } diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 5bac660c7..9a1ad630c 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -435,11 +435,11 @@ type Dirent struct { // IterDirentsCallback receives Dirents from FileDescriptionImpl.IterDirents. type IterDirentsCallback interface { - // Handle handles the given iterated Dirent. It returns true if iteration - // should continue, and false if FileDescriptionImpl.IterDirents should - // terminate now and restart with the same Dirent the next time it is - // called. - Handle(dirent Dirent) bool + // Handle handles the given iterated Dirent. If Handle returns a non-nil + // error, FileDescriptionImpl.IterDirents must stop iteration and return + // the error; the next call to FileDescriptionImpl.IterDirents should + // restart with the same Dirent. + Handle(dirent Dirent) error } // OnClose is called when a file descriptor representing the FileDescription is -- cgit v1.2.3 From 5baf9dc2fbb459828b4102b0a1c5214879434c03 Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Fri, 14 Feb 2020 15:48:09 -0800 Subject: Synchronize signalling with S/R This is to fix a data race between sending an external signal to a ThreadGroup and kernel saving state for S/R. PiperOrigin-RevId: 295244281 --- pkg/sentry/kernel/kernel.go | 8 ++++++++ runsc/boot/loader.go | 8 ++++---- 2 files changed, 12 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index ea21af33f..7da0368f1 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -1169,6 +1169,14 @@ func (k *Kernel) SendExternalSignal(info *arch.SignalInfo, context string) { k.sendExternalSignal(info, context) } +// SendExternalSignalThreadGroup injects a signal into an specific ThreadGroup. +// This function doesn't skip signals like SendExternalSignal does. +func (k *Kernel) SendExternalSignalThreadGroup(tg *ThreadGroup, info *arch.SignalInfo) error { + k.extMu.Lock() + defer k.extMu.Unlock() + return tg.SendSignal(info) +} + // SendContainerSignal sends the given signal to all processes inside the // namespace that match the given container ID. func (k *Kernel) SendContainerSignal(cid string, info *arch.SignalInfo) error { diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index 239ca5302..eef43b9df 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -997,7 +997,7 @@ func (l *Loader) signalProcess(cid string, tgid kernel.ThreadID, signo int32) er execTG, _, err := l.threadGroupFromID(execID{cid: cid, pid: tgid}) if err == nil { // Send signal directly to the identified process. - return execTG.SendSignal(&arch.SignalInfo{Signo: signo}) + return l.k.SendExternalSignalThreadGroup(execTG, &arch.SignalInfo{Signo: signo}) } // The caller may be signaling a process not started directly via exec. @@ -1014,7 +1014,7 @@ func (l *Loader) signalProcess(cid string, tgid kernel.ThreadID, signo int32) er if tg.Leader().ContainerID() != cid { return fmt.Errorf("process %d is part of a different container: %q", tgid, tg.Leader().ContainerID()) } - return tg.SendSignal(&arch.SignalInfo{Signo: signo}) + return l.k.SendExternalSignalThreadGroup(tg, &arch.SignalInfo{Signo: signo}) } func (l *Loader) signalForegrondProcessGroup(cid string, tgid kernel.ThreadID, signo int32) error { @@ -1032,7 +1032,7 @@ func (l *Loader) signalForegrondProcessGroup(cid string, tgid kernel.ThreadID, s // No foreground process group has been set. Signal the // original thread group. log.Warningf("No foreground process group for container %q and PID %d. Sending signal directly to PID %d.", cid, tgid, tgid) - return tg.SendSignal(&arch.SignalInfo{Signo: signo}) + return l.k.SendExternalSignalThreadGroup(tg, &arch.SignalInfo{Signo: signo}) } // Send the signal to all processes in the process group. var lastErr error @@ -1040,7 +1040,7 @@ func (l *Loader) signalForegrondProcessGroup(cid string, tgid kernel.ThreadID, s if tg.ProcessGroup() != pg { continue } - if err := tg.SendSignal(&arch.SignalInfo{Signo: signo}); err != nil { + if err := l.k.SendExternalSignalThreadGroup(tg, &arch.SignalInfo{Signo: signo}); err != nil { lastErr = err } } -- cgit v1.2.3 From a5069f820f22734b6c466068a02bbbe83ba091da Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Fri, 14 Feb 2020 16:19:00 -0800 Subject: Remove linux.EpollEvent.Fd. glibc defines struct epoll_event in such a way that epoll_event.data.fd exists. However, the kernel's definition of struct epoll_event makes epoll_event.data an opaque uint64, so naming half of it "fd" just introduces confusion. Remove the Fd field, and make Data a [2]int32 to compensate. Also add required padding to linux.EpollEvent on ARM64. PiperOrigin-RevId: 295250424 --- pkg/abi/linux/BUILD | 2 ++ pkg/abi/linux/epoll.go | 7 ------- pkg/abi/linux/epoll_amd64.go | 25 +++++++++++++++++++++++++ pkg/abi/linux/epoll_arm64.go | 24 ++++++++++++++++++++++++ pkg/sentry/syscalls/linux/sys_epoll.go | 3 +-- pkg/sentry/vfs/epoll.go | 19 +++++++++---------- 6 files changed, 61 insertions(+), 19 deletions(-) create mode 100644 pkg/abi/linux/epoll_amd64.go create mode 100644 pkg/abi/linux/epoll_arm64.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/BUILD b/pkg/abi/linux/BUILD index b7015367b..a89f34d4b 100644 --- a/pkg/abi/linux/BUILD +++ b/pkg/abi/linux/BUILD @@ -17,6 +17,8 @@ go_library( "dev.go", "elf.go", "epoll.go", + "epoll_amd64.go", + "epoll_arm64.go", "errors.go", "eventfd.go", "exec.go", diff --git a/pkg/abi/linux/epoll.go b/pkg/abi/linux/epoll.go index 0e881aa3c..6e4de69da 100644 --- a/pkg/abi/linux/epoll.go +++ b/pkg/abi/linux/epoll.go @@ -14,13 +14,6 @@ package linux -// EpollEvent is equivalent to struct epoll_event from epoll(2). -type EpollEvent struct { - Events uint32 - Fd int32 - Data int32 -} - // Event masks. const ( EPOLLIN = 0x1 diff --git a/pkg/abi/linux/epoll_amd64.go b/pkg/abi/linux/epoll_amd64.go new file mode 100644 index 000000000..57041491c --- /dev/null +++ b/pkg/abi/linux/epoll_amd64.go @@ -0,0 +1,25 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package linux + +// EpollEvent is equivalent to struct epoll_event from epoll(2). +type EpollEvent struct { + Events uint32 + // Linux makes struct epoll_event::data a __u64. We represent it as + // [2]int32 because, on amd64, Linux also makes struct epoll_event + // __attribute__((packed)), such that there is no padding between Events + // and Data. + Data [2]int32 +} diff --git a/pkg/abi/linux/epoll_arm64.go b/pkg/abi/linux/epoll_arm64.go new file mode 100644 index 000000000..62ef5821e --- /dev/null +++ b/pkg/abi/linux/epoll_arm64.go @@ -0,0 +1,24 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package linux + +// EpollEvent is equivalent to struct epoll_event from epoll(2). +type EpollEvent struct { + Events uint32 + // Linux makes struct epoll_event a __u64, necessitating 4 bytes of padding + // here. + _ int32 + Data [2]int32 +} diff --git a/pkg/sentry/syscalls/linux/sys_epoll.go b/pkg/sentry/syscalls/linux/sys_epoll.go index 5f11b496c..fbef5b376 100644 --- a/pkg/sentry/syscalls/linux/sys_epoll.go +++ b/pkg/sentry/syscalls/linux/sys_epoll.go @@ -83,8 +83,7 @@ func EpollCtl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc } mask = waiter.EventMaskFromLinux(e.Events) - data[0] = e.Fd - data[1] = e.Data + data = e.Data } // Perform the requested operations. diff --git a/pkg/sentry/vfs/epoll.go b/pkg/sentry/vfs/epoll.go index 7c83f9a5a..eed41139b 100644 --- a/pkg/sentry/vfs/epoll.go +++ b/pkg/sentry/vfs/epoll.go @@ -85,8 +85,8 @@ type epollInterest struct { ready bool epollInterestEntry - // userData is the epoll_data_t associated with this epollInterest. - // userData is protected by epoll.mu. + // userData is the struct epoll_event::data associated with this + // epollInterest. userData is protected by epoll.mu. userData [2]int32 } @@ -157,7 +157,7 @@ func (ep *EpollInstance) Seek(ctx context.Context, offset int64, whence int32) ( // AddInterest implements the semantics of EPOLL_CTL_ADD. // // Preconditions: A reference must be held on file. -func (ep *EpollInstance) AddInterest(file *FileDescription, num int32, mask uint32, userData [2]int32) error { +func (ep *EpollInstance) AddInterest(file *FileDescription, num int32, event linux.EpollEvent) error { // Check for cyclic polling if necessary. subep, _ := file.impl.(*EpollInstance) if subep != nil { @@ -183,12 +183,12 @@ func (ep *EpollInstance) AddInterest(file *FileDescription, num int32, mask uint } // Register interest in file. - mask |= linux.EPOLLERR | linux.EPOLLRDHUP + mask := event.Events | linux.EPOLLERR | linux.EPOLLRDHUP epi := &epollInterest{ epoll: ep, key: key, mask: mask, - userData: userData, + userData: event.Data, } ep.interest[key] = epi wmask := waiter.EventMaskFromLinux(mask) @@ -236,7 +236,7 @@ func (ep *EpollInstance) mightPollRecursive(ep2 *EpollInstance, remainingRecursi // ModifyInterest implements the semantics of EPOLL_CTL_MOD. // // Preconditions: A reference must be held on file. -func (ep *EpollInstance) ModifyInterest(file *FileDescription, num int32, mask uint32, userData [2]int32) error { +func (ep *EpollInstance) ModifyInterest(file *FileDescription, num int32, event linux.EpollEvent) error { ep.interestMu.Lock() defer ep.interestMu.Unlock() @@ -250,13 +250,13 @@ func (ep *EpollInstance) ModifyInterest(file *FileDescription, num int32, mask u } // Update epi for the next call to ep.ReadEvents(). + mask := event.Events | linux.EPOLLERR | linux.EPOLLRDHUP ep.mu.Lock() epi.mask = mask - epi.userData = userData + epi.userData = event.Data ep.mu.Unlock() // Re-register with the new mask. - mask |= linux.EPOLLERR | linux.EPOLLRDHUP file.EventUnregister(&epi.waiter) wmask := waiter.EventMaskFromLinux(mask) file.EventRegister(&epi.waiter, wmask) @@ -363,8 +363,7 @@ func (ep *EpollInstance) ReadEvents(events []linux.EpollEvent) int { // Report ievents. events[i] = linux.EpollEvent{ Events: ievents.ToLinux(), - Fd: epi.userData[0], - Data: epi.userData[1], + Data: epi.userData, } i++ if i == len(events) { -- cgit v1.2.3 From fae3de21af7f50266565643c6283912b087b0f5a Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Tue, 18 Feb 2020 10:49:47 -0800 Subject: ring0/pagetables: fix typo PiperOrigin-RevId: 295770717 --- pkg/sentry/platform/ring0/pagetables/BUILD | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/ring0/pagetables/BUILD b/pkg/sentry/platform/ring0/pagetables/BUILD index 971eed7fa..4f2406ce3 100644 --- a/pkg/sentry/platform/ring0/pagetables/BUILD +++ b/pkg/sentry/platform/ring0/pagetables/BUILD @@ -7,7 +7,7 @@ go_template( name = "generic_walker", srcs = select_arch( amd64 = ["walker_amd64.go"], - arm64 = ["walker_amd64.go"], + arm64 = ["walker_arm64.go"], ), opt_types = [ "Visitor", -- cgit v1.2.3 From b30b7f3422202232ad1c385a7ac0d775151fee2f Mon Sep 17 00:00:00 2001 From: Nayana Bidari Date: Tue, 18 Feb 2020 11:30:42 -0800 Subject: Add nat table support for iptables. Add nat table support for Prerouting hook with Redirect option. Add tests to check redirect of ports. --- pkg/abi/linux/netfilter.go | 27 ++++++++++++++ pkg/sentry/socket/netfilter/netfilter.go | 58 +++++++++++++++++++++++++++-- pkg/tcpip/iptables/iptables.go | 21 +++++++++++ pkg/tcpip/iptables/targets.go | 24 ++++++++++++ pkg/tcpip/stack/nic.go | 23 ++++++++++++ test/iptables/iptables_test.go | 12 ++++++ test/iptables/iptables_util.go | 10 +++++ test/iptables/nat.go | 64 +++++++++++++++++++++++++++++++- 8 files changed, 234 insertions(+), 5 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/netfilter.go b/pkg/abi/linux/netfilter.go index bbc4df74c..ba4d84962 100644 --- a/pkg/abi/linux/netfilter.go +++ b/pkg/abi/linux/netfilter.go @@ -250,6 +250,33 @@ type XTErrorTarget struct { // SizeOfXTErrorTarget is the size of an XTErrorTarget. const SizeOfXTErrorTarget = 64 +// NfNATIPV4Range. It corresponds to struct nf_nat_ipv4_range +// in include/uapi/linux/netfilter/nf_nat.h. +type NfNATIPV4Range struct { + Flags uint32 + MinIP [4]byte + MaxIP [4]byte + MinPort uint16 + MaxPort uint16 +} + +// NfNATIPV4MultiRangeCompat. It corresponds to struct +// nf_nat_ipv4_multi_range_compat in include/uapi/linux/netfilter/nf_nat.h. +type NfNATIPV4MultiRangeCompat struct { + Rangesize uint32 + RangeIPV4 [1]NfNATIPV4Range +} + +// XTRedirectTarget triggers a redirect when reached. +type XTRedirectTarget struct { + Target XTEntryTarget + NfRange NfNATIPV4MultiRangeCompat + _ [4]byte +} + +// SizeOfXTRedirectTarget is the size of an XTRedirectTarget. +const SizeOfXTRedirectTarget = 56 + // IPTGetinfo is the argument for the IPT_SO_GET_INFO sockopt. It corresponds // to struct ipt_getinfo in include/uapi/linux/netfilter_ipv4/ip_tables.h. type IPTGetinfo struct { diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 3fc80e0de..512ad624a 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -35,6 +35,11 @@ import ( // shouldn't be reached - an error has occurred if we fall through to one. const errorTargetName = "ERROR" +// redirectTargetName is used to mark targets as redirect targets. Redirect +// targets should be reached for only NAT and Mangle tables. These targets will +// change the destination port/destination IP for packets. +const redirectTargetName = "REDIRECT" + // Metadata is used to verify that we are correctly serializing and // deserializing iptables into structs consumable by the iptables tool. We save // a metadata struct when the tables are written, and when they are read out we @@ -240,6 +245,8 @@ func marshalTarget(target iptables.Target) []byte { return marshalErrorTarget(tg.Name) case iptables.ReturnTarget: return marshalStandardTarget(iptables.RuleReturn) + case iptables.RedirectTarget: + return marshalRedirectTarget() default: panic(fmt.Errorf("unknown target of type %T", target)) } @@ -274,6 +281,18 @@ func marshalErrorTarget(errorName string) []byte { return binary.Marshal(ret, usermem.ByteOrder, target) } +func marshalRedirectTarget() []byte { + // This is a redirect target named redirect + target := linux.XTRedirectTarget{ + Target: linux.XTEntryTarget{ + TargetSize: linux.SizeOfXTRedirectTarget, + }, + } + + ret := make([]byte, 0, linux.SizeOfXTRedirectTarget) + return binary.Marshal(ret, usermem.ByteOrder, target) +} + // translateFromStandardVerdict translates verdicts the same way as the iptables // tool. func translateFromStandardVerdict(verdict iptables.RuleVerdict) int32 { @@ -326,6 +345,8 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { switch replace.Name.String() { case iptables.TablenameFilter: table = iptables.EmptyFilterTable() + case iptables.TablenameNat: + table = iptables.EmptyNatTable() default: nflog("we don't yet support writing to the %q table (gvisor.dev/issue/170)", replace.Name.String()) return syserr.ErrInvalidArgument @@ -455,10 +476,11 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { } // TODO(gvisor.dev/issue/170): Support other chains. - // Since we only support modifying the INPUT chain right now, make sure - // all other chains point to ACCEPT rules. + // Since we only support modifying the INPUT chain and redirect for + // PREROUTING chain right now, make sure all other chains point to + // ACCEPT rules. for hook, ruleIdx := range table.BuiltinChains { - if hook != iptables.Input { + if hook != iptables.Input && hook != iptables.Prerouting { if _, ok := table.Rules[ruleIdx].Target.(iptables.AcceptTarget); !ok { nflog("hook %d is unsupported.", hook) return syserr.ErrInvalidArgument @@ -575,6 +597,36 @@ func parseTarget(optVal []byte) (iptables.Target, error) { nflog("set entries: user-defined target %q", name) return iptables.UserChainTarget{Name: name}, nil } + + case redirectTargetName: + // Redirect target. + if len(optVal) < linux.SizeOfXTRedirectTarget { + return nil, fmt.Errorf("netfilter.SetEntries: optVal has insufficient size for redirect target %d", len(optVal)) + } + + var redirectTarget linux.XTRedirectTarget + buf = optVal[:linux.SizeOfXTRedirectTarget] + binary.Unmarshal(buf, usermem.ByteOrder, &redirectTarget) + + // Copy linux.XTRedirectTarget to iptables.RedirectTarget. + var target iptables.RedirectTarget + nfRange := redirectTarget.NfRange + + target.RangeSize = nfRange.Rangesize + target.Flags = nfRange.RangeIPV4[0].Flags + + target.MinIP = tcpip.Address(nfRange.RangeIPV4[0].MinIP[:]) + target.MaxIP = tcpip.Address(nfRange.RangeIPV4[0].MaxIP[:]) + + // Convert port from big endian to little endian. + port := make([]byte, 2) + binary.BigEndian.PutUint16(port, nfRange.RangeIPV4[0].MinPort) + target.MinPort = binary.LittleEndian.Uint16(port) + + binary.BigEndian.PutUint16(port, nfRange.RangeIPV4[0].MaxPort) + target.MaxPort = binary.LittleEndian.Uint16(port) + return target, nil + } // Unknown target. diff --git a/pkg/tcpip/iptables/iptables.go b/pkg/tcpip/iptables/iptables.go index 75a433a3b..c00d012c0 100644 --- a/pkg/tcpip/iptables/iptables.go +++ b/pkg/tcpip/iptables/iptables.go @@ -135,6 +135,27 @@ func EmptyFilterTable() Table { } } +// EmptyNatTable returns a Table with no rules and the filter table chains +// mapped to HookUnset. +func EmptyNatTable() Table { + return Table{ + Rules: []Rule{}, + BuiltinChains: map[Hook]int{ + Prerouting: HookUnset, + Input: HookUnset, + Output: HookUnset, + Postrouting: HookUnset, + }, + Underflows: map[Hook]int{ + Prerouting: HookUnset, + Input: HookUnset, + Output: HookUnset, + Postrouting: HookUnset, + }, + UserChains: map[string]int{}, + } +} + // Check runs pkt through the rules for hook. It returns true when the packet // should continue traversing the network stack and false when it should be // dropped. diff --git a/pkg/tcpip/iptables/targets.go b/pkg/tcpip/iptables/targets.go index 9fc60cfad..06e65bece 100644 --- a/pkg/tcpip/iptables/targets.go +++ b/pkg/tcpip/iptables/targets.go @@ -19,6 +19,7 @@ package iptables import ( "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/header" ) // AcceptTarget accepts packets. @@ -65,3 +66,26 @@ type ReturnTarget struct{} func (ReturnTarget) Action(tcpip.PacketBuffer) (RuleVerdict, string) { return RuleReturn, "" } + +// RedirectTarget redirects the packet by modifying the destination port/IP. +type RedirectTarget struct { + RangeSize uint32 + Flags uint32 + MinIP tcpip.Address + MaxIP tcpip.Address + MinPort uint16 + MaxPort uint16 +} + +// Action implements Target.Action. +func (rt RedirectTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, string) { + log.Infof("RedirectTarget triggered.") + + // TODO(gvisor.dev/issue/170): Checking only for UDP protocol. + // We're yet to support for TCP protocol. + headerView := packet.Data.First() + h := header.UDP(headerView) + h.SetDestinationPort(rt.MinPort) + + return RuleAccept, "" +} diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index ca3a7a07e..2028f5201 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -25,6 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/iptables" ) // NIC represents a "network interface card" to which the networking stack is @@ -1012,6 +1013,7 @@ func (n *NIC) leaveGroupLocked(addr tcpip.Address) *tcpip.Error { func handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, localLinkAddr, remotelinkAddr tcpip.LinkAddress, ref *referencedNetworkEndpoint, pkt tcpip.PacketBuffer) { r := makeRoute(protocol, dst, src, localLinkAddr, ref, false /* handleLocal */, false /* multicastLoop */) r.RemoteLinkAddress = remotelinkAddr + ref.ep.HandlePacket(&r, pkt) ref.decRef() } @@ -1082,6 +1084,27 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link n.stack.stats.IP.InvalidSourceAddressesReceived.Increment() return } + + // TODO(gvisor.dev/issue/170): Not supporting iptables for IPv6 yet. + if protocol == header.IPv4ProtocolNumber { + newPkt := pkt.Clone() + + headerView := newPkt.Data.First() + h := header.IPv4(headerView) + newPkt.NetworkHeader = headerView[:h.HeaderLength()] + + hlen := int(h.HeaderLength()) + tlen := int(h.TotalLength()) + newPkt.Data.TrimFront(hlen) + newPkt.Data.CapLength(tlen - hlen) + + ipt := n.stack.IPTables() + if ok := ipt.Check(iptables.Prerouting, newPkt); !ok { + // iptables is telling us to drop the packet. + return + } + } + if ref := n.getRef(protocol, dst); ref != nil { handlePacket(protocol, dst, src, linkEP.LinkAddress(), remote, ref, pkt) return diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go index 46a7c99b0..7d061acba 100644 --- a/test/iptables/iptables_test.go +++ b/test/iptables/iptables_test.go @@ -196,12 +196,24 @@ func TestNATRedirectUDPPort(t *testing.T) { } } +func TestNATRedirectTCPPort(t *testing.T) { + if err := singleTest(NATRedirectTCPPort{}); err != nil { + t.Fatal(err) + } +} + func TestNATDropUDP(t *testing.T) { if err := singleTest(NATDropUDP{}); err != nil { t.Fatal(err) } } +func TestNATAcceptAll(t *testing.T) { + if err := singleTest(NATAcceptAll{}); err != nil { + t.Fatal(err) + } +} + func TestFilterInputDropTCPDestPort(t *testing.T) { if err := singleTest(FilterInputDropTCPDestPort{}); err != nil { t.Fatal(err) diff --git a/test/iptables/iptables_util.go b/test/iptables/iptables_util.go index 043114c78..5c9199abf 100644 --- a/test/iptables/iptables_util.go +++ b/test/iptables/iptables_util.go @@ -35,6 +35,16 @@ func filterTable(args ...string) error { return nil } +// natTable calls `iptables -t nat` with the given args. +func natTable(args ...string) error { + args = append([]string{"-t", "nat"}, args...) + cmd := exec.Command(iptablesBinary, args...) + if out, err := cmd.CombinedOutput(); err != nil { + return fmt.Errorf("error running iptables with args %v\nerror: %v\noutput: %s", args, err, string(out)) + } + return nil +} + // listenUDP listens on a UDP port and returns the value of net.Conn.Read() for // the first read on that port. func listenUDP(port int, timeout time.Duration) error { diff --git a/test/iptables/nat.go b/test/iptables/nat.go index b5c6f927e..306cbd1b3 100644 --- a/test/iptables/nat.go +++ b/test/iptables/nat.go @@ -25,7 +25,9 @@ const ( func init() { RegisterTestCase(NATRedirectUDPPort{}) + RegisterTestCase(NATRedirectTCPPort{}) RegisterTestCase(NATDropUDP{}) + RegisterTestCase(NATAcceptAll{}) } // NATRedirectUDPPort tests that packets are redirected to different port. @@ -38,13 +40,14 @@ func (NATRedirectUDPPort) Name() string { // ContainerAction implements TestCase.ContainerAction. func (NATRedirectUDPPort) ContainerAction(ip net.IP) error { - if err := filterTable("-t", "nat", "-A", "PREROUTING", "-p", "udp", "-j", "REDIRECT", "--to-ports", fmt.Sprintf("%d", redirectPort)); err != nil { + if err := natTable("-A", "PREROUTING", "-p", "udp", "-j", "REDIRECT", "--to-ports", fmt.Sprintf("%d", redirectPort)); err != nil { return err } if err := listenUDP(redirectPort, sendloopDuration); err != nil { return fmt.Errorf("packets on port %d should be allowed, but encountered an error: %v", redirectPort, err) } + return nil } @@ -53,6 +56,37 @@ func (NATRedirectUDPPort) LocalAction(ip net.IP) error { return sendUDPLoop(ip, acceptPort, sendloopDuration) } +// NATRedirectTCPPort tests that connections are redirected on specified ports. +type NATRedirectTCPPort struct{} + +// Name implements TestCase.Name. +func (NATRedirectTCPPort) Name() string { + return "NATRedirectTCPPort" +} + +// ContainerAction implements TestCase.ContainerAction. +func (NATRedirectTCPPort) ContainerAction(ip net.IP) error { + if err := natTable("-A", "PREROUTING", "-p", "tcp", "-m", "tcp", "--dport", fmt.Sprintf("%d", dropPort), "-j", "REDIRECT", "--to-ports", fmt.Sprintf("%d", redirectPort)); err != nil { + return err + } + + // Listen for TCP packets on redirect port. + if err := listenTCP(redirectPort, sendloopDuration); err != nil { + return fmt.Errorf("connection on port %d should be accepted, but got error %v", redirectPort, err) + } + + return nil +} + +// LocalAction implements TestCase.LocalAction. +func (NATRedirectTCPPort) LocalAction(ip net.IP) error { + if err := connectTCP(ip, dropPort, acceptPort, sendloopDuration); err != nil { + return fmt.Errorf("connection destined to port %d should be accepted, but got error %v", dropPort, err) + } + + return nil +} + // NATDropUDP tests that packets are not received in ports other than redirect port. type NATDropUDP struct{} @@ -63,7 +97,7 @@ func (NATDropUDP) Name() string { // ContainerAction implements TestCase.ContainerAction. func (NATDropUDP) ContainerAction(ip net.IP) error { - if err := filterTable("-t", "nat", "-A", "PREROUTING", "-p", "udp", "-j", "REDIRECT", "--to-ports", fmt.Sprintf("%d", redirectPort)); err != nil { + if err := natTable("-A", "PREROUTING", "-p", "udp", "-j", "REDIRECT", "--to-ports", fmt.Sprintf("%d", redirectPort)); err != nil { return err } @@ -78,3 +112,29 @@ func (NATDropUDP) ContainerAction(ip net.IP) error { func (NATDropUDP) LocalAction(ip net.IP) error { return sendUDPLoop(ip, acceptPort, sendloopDuration) } + +// NATAcceptAll tests that all UDP packets are accepted. +type NATAcceptAll struct{} + +// Name implements TestCase.Name. +func (NATAcceptAll) Name() string { + return "NATAcceptAll" +} + +// ContainerAction implements TestCase.ContainerAction. +func (NATAcceptAll) ContainerAction(ip net.IP) error { + if err := natTable("-A", "PREROUTING", "-p", "udp", "-j", "ACCEPT"); err != nil { + return err + } + + if err := listenUDP(acceptPort, sendloopDuration); err != nil { + return fmt.Errorf("packets on port %d should be allowed, but encountered an error: %v", acceptPort, err) + } + + return nil +} + +// LocalAction implements TestCase.LocalAction. +func (NATAcceptAll) LocalAction(ip net.IP) error { + return sendUDPLoop(ip, acceptPort, sendloopDuration) +} -- cgit v1.2.3 From 906eb6295d54a05663a223f1dc379a16148de2d1 Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Tue, 18 Feb 2020 13:42:31 -0800 Subject: atomicbitops package cleanups - Redocument memory ordering from "no ordering" to "acquire-release". (No functional change: both LOCK WHATEVER on x86, and LDAXR/STLXR loops on ARM64, already have this property.) - Remove IncUnlessZeroInt32 and DecUnlessOneInt32, which were only faster than the equivalent loops using sync/atomic before the Go compiler inlined non-unsafe.Pointer atomics many releases ago. PiperOrigin-RevId: 295811743 --- pkg/atomicbitops/BUILD | 9 +- pkg/atomicbitops/atomic_bitops.go | 60 ------- pkg/atomicbitops/atomic_bitops_amd64.s | 115 -------------- pkg/atomicbitops/atomic_bitops_arm64.s | 139 ---------------- pkg/atomicbitops/atomic_bitops_common.go | 147 ----------------- pkg/atomicbitops/atomic_bitops_test.go | 262 ------------------------------- pkg/atomicbitops/atomicbitops.go | 47 ++++++ pkg/atomicbitops/atomicbitops_amd64.s | 77 +++++++++ pkg/atomicbitops/atomicbitops_arm64.s | 105 +++++++++++++ pkg/atomicbitops/atomicbitops_noasm.go | 105 +++++++++++++ pkg/atomicbitops/atomicbitops_test.go | 198 +++++++++++++++++++++++ pkg/sentry/mm/address_space.go | 23 ++- pkg/sentry/mm/lifecycle.go | 11 +- 13 files changed, 563 insertions(+), 735 deletions(-) delete mode 100644 pkg/atomicbitops/atomic_bitops.go delete mode 100644 pkg/atomicbitops/atomic_bitops_amd64.s delete mode 100644 pkg/atomicbitops/atomic_bitops_arm64.s delete mode 100644 pkg/atomicbitops/atomic_bitops_common.go delete mode 100644 pkg/atomicbitops/atomic_bitops_test.go create mode 100644 pkg/atomicbitops/atomicbitops.go create mode 100644 pkg/atomicbitops/atomicbitops_amd64.s create mode 100644 pkg/atomicbitops/atomicbitops_arm64.s create mode 100644 pkg/atomicbitops/atomicbitops_noasm.go create mode 100644 pkg/atomicbitops/atomicbitops_test.go (limited to 'pkg/sentry') diff --git a/pkg/atomicbitops/BUILD b/pkg/atomicbitops/BUILD index 3948074ba..ba8b06071 100644 --- a/pkg/atomicbitops/BUILD +++ b/pkg/atomicbitops/BUILD @@ -5,10 +5,9 @@ package(licenses = ["notice"]) go_library( name = "atomicbitops", srcs = [ - "atomic_bitops.go", - "atomic_bitops_amd64.s", - "atomic_bitops_arm64.s", - "atomic_bitops_common.go", + "atomicbitops.go", + "atomicbitops_amd64.s", + "atomicbitops_noasm.go", ], visibility = ["//:sandbox"], ) @@ -16,7 +15,7 @@ go_library( go_test( name = "atomicbitops_test", size = "small", - srcs = ["atomic_bitops_test.go"], + srcs = ["atomicbitops_test.go"], library = ":atomicbitops", deps = ["//pkg/sync"], ) diff --git a/pkg/atomicbitops/atomic_bitops.go b/pkg/atomicbitops/atomic_bitops.go deleted file mode 100644 index fcc41a9ea..000000000 --- a/pkg/atomicbitops/atomic_bitops.go +++ /dev/null @@ -1,60 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build amd64 arm64 - -// Package atomicbitops provides basic bitwise operations in an atomic way. -// The implementation on amd64 leverages the LOCK prefix directly instead of -// relying on the generic cas primitives, and the arm64 leverages the LDAXR -// and STLXR pair primitives. -// -// WARNING: the bitwise ops provided in this package doesn't imply any memory -// ordering. Using them to construct locks must employ proper memory barriers. -package atomicbitops - -// AndUint32 atomically applies bitwise and operation to *addr with val. -func AndUint32(addr *uint32, val uint32) - -// OrUint32 atomically applies bitwise or operation to *addr with val. -func OrUint32(addr *uint32, val uint32) - -// XorUint32 atomically applies bitwise xor operation to *addr with val. -func XorUint32(addr *uint32, val uint32) - -// CompareAndSwapUint32 is like sync/atomic.CompareAndSwapUint32, but returns -// the value previously stored at addr. -func CompareAndSwapUint32(addr *uint32, old, new uint32) uint32 - -// AndUint64 atomically applies bitwise and operation to *addr with val. -func AndUint64(addr *uint64, val uint64) - -// OrUint64 atomically applies bitwise or operation to *addr with val. -func OrUint64(addr *uint64, val uint64) - -// XorUint64 atomically applies bitwise xor operation to *addr with val. -func XorUint64(addr *uint64, val uint64) - -// CompareAndSwapUint64 is like sync/atomic.CompareAndSwapUint64, but returns -// the value previously stored at addr. -func CompareAndSwapUint64(addr *uint64, old, new uint64) uint64 - -// IncUnlessZeroInt32 increments the value stored at the given address and -// returns true; unless the value stored in the pointer is zero, in which case -// it is left unmodified and false is returned. -func IncUnlessZeroInt32(addr *int32) bool - -// DecUnlessOneInt32 decrements the value stored at the given address and -// returns true; unless the value stored in the pointer is 1, in which case it -// is left unmodified and false is returned. -func DecUnlessOneInt32(addr *int32) bool diff --git a/pkg/atomicbitops/atomic_bitops_amd64.s b/pkg/atomicbitops/atomic_bitops_amd64.s deleted file mode 100644 index db0972001..000000000 --- a/pkg/atomicbitops/atomic_bitops_amd64.s +++ /dev/null @@ -1,115 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build amd64 - -#include "textflag.h" - -TEXT ·AndUint32(SB),$0-12 - MOVQ addr+0(FP), BP - MOVL val+8(FP), AX - LOCK - ANDL AX, 0(BP) - RET - -TEXT ·OrUint32(SB),$0-12 - MOVQ addr+0(FP), BP - MOVL val+8(FP), AX - LOCK - ORL AX, 0(BP) - RET - -TEXT ·XorUint32(SB),$0-12 - MOVQ addr+0(FP), BP - MOVL val+8(FP), AX - LOCK - XORL AX, 0(BP) - RET - -TEXT ·CompareAndSwapUint32(SB),$0-20 - MOVQ addr+0(FP), DI - MOVL old+8(FP), AX - MOVL new+12(FP), DX - LOCK - CMPXCHGL DX, 0(DI) - MOVL AX, ret+16(FP) - RET - -TEXT ·AndUint64(SB),$0-16 - MOVQ addr+0(FP), BP - MOVQ val+8(FP), AX - LOCK - ANDQ AX, 0(BP) - RET - -TEXT ·OrUint64(SB),$0-16 - MOVQ addr+0(FP), BP - MOVQ val+8(FP), AX - LOCK - ORQ AX, 0(BP) - RET - -TEXT ·XorUint64(SB),$0-16 - MOVQ addr+0(FP), BP - MOVQ val+8(FP), AX - LOCK - XORQ AX, 0(BP) - RET - -TEXT ·CompareAndSwapUint64(SB),$0-32 - MOVQ addr+0(FP), DI - MOVQ old+8(FP), AX - MOVQ new+16(FP), DX - LOCK - CMPXCHGQ DX, 0(DI) - MOVQ AX, ret+24(FP) - RET - -TEXT ·IncUnlessZeroInt32(SB),NOSPLIT,$0-9 - MOVQ addr+0(FP), DI - MOVL 0(DI), AX - -retry: - TESTL AX, AX - JZ fail - LEAL 1(AX), DX - LOCK - CMPXCHGL DX, 0(DI) - JNZ retry - - SETEQ ret+8(FP) - RET - -fail: - MOVB AX, ret+8(FP) - RET - -TEXT ·DecUnlessOneInt32(SB),NOSPLIT,$0-9 - MOVQ addr+0(FP), DI - MOVL 0(DI), AX - -retry: - LEAL -1(AX), DX - TESTL DX, DX - JZ fail - LOCK - CMPXCHGL DX, 0(DI) - JNZ retry - - SETEQ ret+8(FP) - RET - -fail: - MOVB DX, ret+8(FP) - RET diff --git a/pkg/atomicbitops/atomic_bitops_arm64.s b/pkg/atomicbitops/atomic_bitops_arm64.s deleted file mode 100644 index 97f8808c1..000000000 --- a/pkg/atomicbitops/atomic_bitops_arm64.s +++ /dev/null @@ -1,139 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build arm64 - -#include "textflag.h" - -TEXT ·AndUint32(SB),$0-12 - MOVD ptr+0(FP), R0 - MOVW val+8(FP), R1 -again: - LDAXRW (R0), R2 - ANDW R1, R2 - STLXRW R2, (R0), R3 - CBNZ R3, again - RET - -TEXT ·OrUint32(SB),$0-12 - MOVD ptr+0(FP), R0 - MOVW val+8(FP), R1 -again: - LDAXRW (R0), R2 - ORRW R1, R2 - STLXRW R2, (R0), R3 - CBNZ R3, again - RET - -TEXT ·XorUint32(SB),$0-12 - MOVD ptr+0(FP), R0 - MOVW val+8(FP), R1 -again: - LDAXRW (R0), R2 - EORW R1, R2 - STLXRW R2, (R0), R3 - CBNZ R3, again - RET - -TEXT ·CompareAndSwapUint32(SB),$0-20 - MOVD addr+0(FP), R0 - MOVW old+8(FP), R1 - MOVW new+12(FP), R2 - -again: - LDAXRW (R0), R3 - CMPW R1, R3 - BNE done - STLXRW R2, (R0), R4 - CBNZ R4, again -done: - MOVW R3, prev+16(FP) - RET - -TEXT ·AndUint64(SB),$0-16 - MOVD ptr+0(FP), R0 - MOVD val+8(FP), R1 -again: - LDAXR (R0), R2 - AND R1, R2 - STLXR R2, (R0), R3 - CBNZ R3, again - RET - -TEXT ·OrUint64(SB),$0-16 - MOVD ptr+0(FP), R0 - MOVD val+8(FP), R1 -again: - LDAXR (R0), R2 - ORR R1, R2 - STLXR R2, (R0), R3 - CBNZ R3, again - RET - -TEXT ·XorUint64(SB),$0-16 - MOVD ptr+0(FP), R0 - MOVD val+8(FP), R1 -again: - LDAXR (R0), R2 - EOR R1, R2 - STLXR R2, (R0), R3 - CBNZ R3, again - RET - -TEXT ·CompareAndSwapUint64(SB),$0-32 - MOVD addr+0(FP), R0 - MOVD old+8(FP), R1 - MOVD new+16(FP), R2 - -again: - LDAXR (R0), R3 - CMP R1, R3 - BNE done - STLXR R2, (R0), R4 - CBNZ R4, again -done: - MOVD R3, prev+24(FP) - RET - -TEXT ·IncUnlessZeroInt32(SB),NOSPLIT,$0-9 - MOVD addr+0(FP), R0 - -again: - LDAXRW (R0), R1 - CBZ R1, fail - ADDW $1, R1 - STLXRW R1, (R0), R2 - CBNZ R2, again - MOVW $1, R2 - MOVB R2, ret+8(FP) - RET -fail: - MOVB ZR, ret+8(FP) - RET - -TEXT ·DecUnlessOneInt32(SB),NOSPLIT,$0-9 - MOVD addr+0(FP), R0 - -again: - LDAXRW (R0), R1 - SUBSW $1, R1, R1 - BEQ fail - STLXRW R1, (R0), R2 - CBNZ R2, again - MOVW $1, R2 - MOVB R2, ret+8(FP) - RET -fail: - MOVB ZR, ret+8(FP) - RET diff --git a/pkg/atomicbitops/atomic_bitops_common.go b/pkg/atomicbitops/atomic_bitops_common.go deleted file mode 100644 index 85163ad62..000000000 --- a/pkg/atomicbitops/atomic_bitops_common.go +++ /dev/null @@ -1,147 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build !amd64,!arm64 - -package atomicbitops - -import ( - "sync/atomic" -) - -// AndUint32 atomically applies bitwise and operation to *addr with val. -func AndUint32(addr *uint32, val uint32) { - for { - o := atomic.LoadUint32(addr) - n := o & val - if atomic.CompareAndSwapUint32(addr, o, n) { - break - } - } -} - -// OrUint32 atomically applies bitwise or operation to *addr with val. -func OrUint32(addr *uint32, val uint32) { - for { - o := atomic.LoadUint32(addr) - n := o | val - if atomic.CompareAndSwapUint32(addr, o, n) { - break - } - } -} - -// XorUint32 atomically applies bitwise xor operation to *addr with val. -func XorUint32(addr *uint32, val uint32) { - for { - o := atomic.LoadUint32(addr) - n := o ^ val - if atomic.CompareAndSwapUint32(addr, o, n) { - break - } - } -} - -// CompareAndSwapUint32 is like sync/atomic.CompareAndSwapUint32, but returns -// the value previously stored at addr. -func CompareAndSwapUint32(addr *uint32, old, new uint32) (prev uint32) { - for { - prev = atomic.LoadUint32(addr) - if prev != old { - return - } - if atomic.CompareAndSwapUint32(addr, old, new) { - return - } - } -} - -// AndUint64 atomically applies bitwise and operation to *addr with val. -func AndUint64(addr *uint64, val uint64) { - for { - o := atomic.LoadUint64(addr) - n := o & val - if atomic.CompareAndSwapUint64(addr, o, n) { - break - } - } -} - -// OrUint64 atomically applies bitwise or operation to *addr with val. -func OrUint64(addr *uint64, val uint64) { - for { - o := atomic.LoadUint64(addr) - n := o | val - if atomic.CompareAndSwapUint64(addr, o, n) { - break - } - } -} - -// XorUint64 atomically applies bitwise xor operation to *addr with val. -func XorUint64(addr *uint64, val uint64) { - for { - o := atomic.LoadUint64(addr) - n := o ^ val - if atomic.CompareAndSwapUint64(addr, o, n) { - break - } - } -} - -// CompareAndSwapUint64 is like sync/atomic.CompareAndSwapUint64, but returns -// the value previously stored at addr. -func CompareAndSwapUint64(addr *uint64, old, new uint64) (prev uint64) { - for { - prev = atomic.LoadUint64(addr) - if prev != old { - return - } - if atomic.CompareAndSwapUint64(addr, old, new) { - return - } - } -} - -// IncUnlessZeroInt32 increments the value stored at the given address and -// returns true; unless the value stored in the pointer is zero, in which case -// it is left unmodified and false is returned. -func IncUnlessZeroInt32(addr *int32) bool { - for { - v := atomic.LoadInt32(addr) - if v == 0 { - return false - } - - if atomic.CompareAndSwapInt32(addr, v, v+1) { - return true - } - } -} - -// DecUnlessOneInt32 decrements the value stored at the given address and -// returns true; unless the value stored in the pointer is 1, in which case it -// is left unmodified and false is returned. -func DecUnlessOneInt32(addr *int32) bool { - for { - v := atomic.LoadInt32(addr) - if v == 1 { - return false - } - - if atomic.CompareAndSwapInt32(addr, v, v-1) { - return true - } - } -} diff --git a/pkg/atomicbitops/atomic_bitops_test.go b/pkg/atomicbitops/atomic_bitops_test.go deleted file mode 100644 index 9466d3e23..000000000 --- a/pkg/atomicbitops/atomic_bitops_test.go +++ /dev/null @@ -1,262 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package atomicbitops - -import ( - "runtime" - "testing" - - "gvisor.dev/gvisor/pkg/sync" -) - -const iterations = 100 - -func detectRaces32(val, target uint32, fn func(*uint32, uint32)) bool { - runtime.GOMAXPROCS(100) - for n := 0; n < iterations; n++ { - x := val - var wg sync.WaitGroup - for i := uint32(0); i < 32; i++ { - wg.Add(1) - go func(a *uint32, i uint32) { - defer wg.Done() - fn(a, uint32(1< Date: Tue, 18 Feb 2020 15:44:22 -0800 Subject: Enable IPV6_RECVTCLASS socket option for datagram sockets Added the ability to get/set the IP_RECVTCLASS socket option on UDP endpoints. If enabled, traffic class from the incoming Network Header passed as ancillary data in the ControlMessages. Adding Get/SetSockOptBool to decrease the overhead of getting/setting simple options. (This was absorbed in a CL that will be landing before this one). Test: * Added unit test to udp_test.go that tests getting/setting as well as verifying that we receive expected TOS from incoming packet. * Added a syscall test for verifying getting/setting * Removed test skip for existing syscall test to enable end to end test. PiperOrigin-RevId: 295840218 --- pkg/sentry/socket/control/control.go | 2 +- pkg/sentry/socket/netstack/netstack.go | 27 +++++- pkg/tcpip/checker/checker.go | 14 +++ pkg/tcpip/tcpip.go | 15 ++- pkg/tcpip/transport/udp/endpoint.go | 38 +++++++- pkg/tcpip/transport/udp/udp_test.go | 120 ++++++++++++++---------- test/syscalls/linux/ip_socket_test_util.h | 16 ++-- test/syscalls/linux/socket_ip_udp_generic.cc | 133 +++++++++++++++++++-------- test/syscalls/linux/udp_socket_test_cases.cc | 4 - 9 files changed, 260 insertions(+), 109 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/control/control.go b/pkg/sentry/socket/control/control.go index 4667373d2..8834a1e1a 100644 --- a/pkg/sentry/socket/control/control.go +++ b/pkg/sentry/socket/control/control.go @@ -329,7 +329,7 @@ func PackTOS(t *kernel.Task, tos uint8, buf []byte) []byte { } // PackTClass packs an IPV6_TCLASS socket control message. -func PackTClass(t *kernel.Task, tClass int32, buf []byte) []byte { +func PackTClass(t *kernel.Task, tClass uint32, buf []byte) []byte { return putCmsgStruct( buf, linux.SOL_IPV6, diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 9757fbfba..e187276c5 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -1318,6 +1318,22 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf } return ib, nil + case linux.IPV6_RECVTCLASS: + if outLen < sizeOfInt32 { + return nil, syserr.ErrInvalidArgument + } + + v, err := ep.GetSockOptBool(tcpip.ReceiveTClassOption) + if err != nil { + return nil, syserr.TranslateNetstackError(err) + } + + var o int32 + if v { + o = 1 + } + return o, nil + default: emitUnimplementedEventIPv6(t, name) } @@ -1803,6 +1819,14 @@ func setSockOptIPv6(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) } return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.IPv6TrafficClassOption(v))) + case linux.IPV6_RECVTCLASS: + v, err := parseIntOrChar(optVal) + if err != nil { + return err + } + + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.ReceiveTClassOption, v != 0)) + default: emitUnimplementedEventIPv6(t, name) } @@ -2086,7 +2110,6 @@ func emitUnimplementedEventIPv6(t *kernel.Task, name int) { linux.IPV6_RECVPATHMTU, linux.IPV6_RECVPKTINFO, linux.IPV6_RECVRTHDR, - linux.IPV6_RECVTCLASS, linux.IPV6_RTHDR, linux.IPV6_RTHDRDSTOPTS, linux.IPV6_TCLASS, @@ -2424,6 +2447,8 @@ func (s *SocketOperations) controlMessages() socket.ControlMessages { Timestamp: s.readCM.Timestamp, HasTOS: s.readCM.HasTOS, TOS: s.readCM.TOS, + HasTClass: s.readCM.HasTClass, + TClass: s.readCM.TClass, HasIPPacketInfo: s.readCM.HasIPPacketInfo, PacketInfo: s.readCM.PacketInfo, }, diff --git a/pkg/tcpip/checker/checker.go b/pkg/tcpip/checker/checker.go index 4d6ae0871..c6c160dfc 100644 --- a/pkg/tcpip/checker/checker.go +++ b/pkg/tcpip/checker/checker.go @@ -161,6 +161,20 @@ func FragmentFlags(flags uint8) NetworkChecker { } } +// ReceiveTClass creates a checker that checks the TCLASS field in +// ControlMessages. +func ReceiveTClass(want uint32) ControlMessagesChecker { + return func(t *testing.T, cm tcpip.ControlMessages) { + t.Helper() + if !cm.HasTClass { + t.Fatalf("got cm.HasTClass = %t, want cm.TClass = %d", cm.HasTClass, want) + } + if got := cm.TClass; got != want { + t.Fatalf("got cm.TClass = %d, want %d", got, want) + } + } +} + // ReceiveTOS creates a checker that checks the TOS field in ControlMessages. func ReceiveTOS(want uint8) ControlMessagesChecker { return func(t *testing.T, cm tcpip.ControlMessages) { diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 9ca39ce40..ce5527391 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -323,11 +323,11 @@ type ControlMessages struct { // TOS is the IPv4 type of service of the associated packet. TOS uint8 - // HasTClass indicates whether Tclass is valid/set. + // HasTClass indicates whether TClass is valid/set. HasTClass bool - // Tclass is the IPv6 traffic class of the associated packet. - TClass int32 + // TClass is the IPv6 traffic class of the associated packet. + TClass uint32 // HasIPPacketInfo indicates whether PacketInfo is set. HasIPPacketInfo bool @@ -502,9 +502,13 @@ type WriteOptions struct { type SockOptBool int const ( + // ReceiveTClassOption is used by SetSockOpt/GetSockOpt to specify if the + // IPV6_TCLASS ancillary message is passed with incoming packets. + ReceiveTClassOption SockOptBool = iota + // ReceiveTOSOption is used by SetSockOpt/GetSockOpt to specify if the TOS // ancillary message is passed with incoming packets. - ReceiveTOSOption SockOptBool = iota + ReceiveTOSOption // V6OnlyOption is used by {G,S}etSockOptBool to specify whether an IPv6 // socket is to be restricted to sending and receiving IPv6 packets only. @@ -514,6 +518,9 @@ const ( // if more inforamtion is provided with incoming packets such // as interface index and address. ReceiveIPPacketInfoOption + + // TODO(b/146901447): convert existing bool socket options to be handled via + // Get/SetSockOptBool ) // SockOptInt represents socket options which values have the int type. diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 3fe91cac2..eff7f3600 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -32,7 +32,8 @@ type udpPacket struct { packetInfo tcpip.IPPacketInfo data buffer.VectorisedView `state:".(buffer.VectorisedView)"` timestamp int64 - tos uint8 + // tos stores either the receiveTOS or receiveTClass value. + tos uint8 } // EndpointState represents the state of a UDP endpoint. @@ -119,6 +120,10 @@ type endpoint struct { // as ancillary data to ControlMessages on Read. receiveTOS bool + // receiveTClass determines if the incoming IPv6 TClass header field is + // passed as ancillary data to ControlMessages on Read. + receiveTClass bool + // receiveIPPacketInfo determines if the packet info is returned by Read. receiveIPPacketInfo bool @@ -258,13 +263,18 @@ func (e *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMess } e.mu.RLock() receiveTOS := e.receiveTOS + receiveTClass := e.receiveTClass receiveIPPacketInfo := e.receiveIPPacketInfo e.mu.RUnlock() if receiveTOS { cm.HasTOS = true cm.TOS = p.tos } - + if receiveTClass { + cm.HasTClass = true + // Although TClass is an 8-bit value it's read in the CMsg as a uint32. + cm.TClass = uint32(p.tos) + } if receiveIPPacketInfo { cm.HasIPPacketInfo = true cm.PacketInfo = p.packetInfo @@ -490,6 +500,17 @@ func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { e.mu.Unlock() return nil + case tcpip.ReceiveTClassOption: + // We only support this option on v6 endpoints. + if e.NetProto != header.IPv6ProtocolNumber { + return tcpip.ErrNotSupported + } + + e.mu.Lock() + e.receiveTClass = v + e.mu.Unlock() + return nil + case tcpip.V6OnlyOption: // We only recognize this option on v6 endpoints. if e.NetProto != header.IPv6ProtocolNumber { @@ -709,6 +730,17 @@ func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { e.mu.RUnlock() return v, nil + case tcpip.ReceiveTClassOption: + // We only support this option on v6 endpoints. + if e.NetProto != header.IPv6ProtocolNumber { + return false, tcpip.ErrNotSupported + } + + e.mu.RLock() + v := e.receiveTClass + e.mu.RUnlock() + return v, nil + case tcpip.V6OnlyOption: // We only recognize this option on v6 endpoints. if e.NetProto != header.IPv6ProtocolNumber { @@ -1273,6 +1305,8 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk packet.packetInfo.LocalAddr = r.LocalAddress packet.packetInfo.DestinationAddr = r.RemoteAddress packet.packetInfo.NIC = r.NICID() + case header.IPv6ProtocolNumber: + packet.tos, _ = header.IPv6(pkt.NetworkHeader).TOS() } packet.timestamp = e.stack.NowNanoseconds() diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index f0ff3fe71..34b7c2360 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -409,6 +409,7 @@ func (c *testContext) injectV6Packet(payload []byte, h *header4Tuple, valid bool // Initialize the IP header. ip := header.IPv6(buf) ip.Encode(&header.IPv6Fields{ + TrafficClass: testTOS, PayloadLength: uint16(header.UDPMinimumSize + len(payload)), NextHeader: uint8(udp.ProtocolNumber), HopLimit: 65, @@ -1336,7 +1337,7 @@ func TestSetTTL(t *testing.T) { } } -func TestTOSV4(t *testing.T) { +func TestSetTOS(t *testing.T) { for _, flow := range []testFlow{unicastV4, multicastV4, broadcast} { t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { c := newDualTestContext(t, defaultMTU) @@ -1347,23 +1348,23 @@ func TestTOSV4(t *testing.T) { const tos = testTOS var v tcpip.IPv4TOSOption if err := c.ep.GetSockOpt(&v); err != nil { - c.t.Errorf("GetSockopt failed: %s", err) + c.t.Errorf("GetSockopt(%T) failed: %s", v, err) } // Test for expected default value. if v != 0 { - c.t.Errorf("got GetSockOpt(...) = %#v, want = %#v", v, 0) + c.t.Errorf("got GetSockOpt(%T) = 0x%x, want = 0x%x", v, v, 0) } if err := c.ep.SetSockOpt(tcpip.IPv4TOSOption(tos)); err != nil { - c.t.Errorf("SetSockOpt(%#v) failed: %s", tcpip.IPv4TOSOption(tos), err) + c.t.Errorf("SetSockOpt(%T, 0x%x) failed: %s", v, tcpip.IPv4TOSOption(tos), err) } if err := c.ep.GetSockOpt(&v); err != nil { - c.t.Errorf("GetSockopt failed: %s", err) + c.t.Errorf("GetSockopt(%T) failed: %s", v, err) } if want := tcpip.IPv4TOSOption(tos); v != want { - c.t.Errorf("got GetSockOpt(...) = %#v, want = %#v", v, want) + c.t.Errorf("got GetSockOpt(%T) = 0x%x, want = 0x%x", v, v, want) } testWrite(c, flow, checker.TOS(tos, 0)) @@ -1371,7 +1372,7 @@ func TestTOSV4(t *testing.T) { } } -func TestTOSV6(t *testing.T) { +func TestSetTClass(t *testing.T) { for _, flow := range []testFlow{unicastV4in6, unicastV6, unicastV6Only, multicastV4in6, multicastV6, broadcastIn6} { t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { c := newDualTestContext(t, defaultMTU) @@ -1379,71 +1380,92 @@ func TestTOSV6(t *testing.T) { c.createEndpointForFlow(flow) - const tos = testTOS + const tClass = testTOS var v tcpip.IPv6TrafficClassOption if err := c.ep.GetSockOpt(&v); err != nil { - c.t.Errorf("GetSockopt failed: %s", err) + c.t.Errorf("GetSockopt(%T) failed: %s", v, err) } // Test for expected default value. if v != 0 { - c.t.Errorf("got GetSockOpt(...) = %#v, want = %#v", v, 0) + c.t.Errorf("got GetSockOpt(%T) = 0x%x, want = 0x%x", v, v, 0) } - if err := c.ep.SetSockOpt(tcpip.IPv6TrafficClassOption(tos)); err != nil { - c.t.Errorf("SetSockOpt failed: %s", err) + if err := c.ep.SetSockOpt(tcpip.IPv6TrafficClassOption(tClass)); err != nil { + c.t.Errorf("SetSockOpt(%T, 0x%x) failed: %s", v, tcpip.IPv6TrafficClassOption(tClass), err) } if err := c.ep.GetSockOpt(&v); err != nil { - c.t.Errorf("GetSockopt failed: %s", err) + c.t.Errorf("GetSockopt(%T) failed: %s", v, err) } - if want := tcpip.IPv6TrafficClassOption(tos); v != want { - c.t.Errorf("got GetSockOpt(...) = %#v, want = %#v", v, want) + if want := tcpip.IPv6TrafficClassOption(tClass); v != want { + c.t.Errorf("got GetSockOpt(%T) = 0x%x, want = 0x%x", v, v, want) } - testWrite(c, flow, checker.TOS(tos, 0)) + // The header getter for TClass is called TOS, so use that checker. + testWrite(c, flow, checker.TOS(tClass, 0)) }) } } -func TestReceiveTOSV4(t *testing.T) { - for _, flow := range []testFlow{unicastV4, broadcast} { - t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() +func TestReceiveTosTClass(t *testing.T) { + testCases := []struct { + name string + getReceiveOption tcpip.SockOptBool + tests []testFlow + }{ + {"ReceiveTosOption", tcpip.ReceiveTOSOption, []testFlow{unicastV4, broadcast}}, + {"ReceiveTClassOption", tcpip.ReceiveTClassOption, []testFlow{unicastV4in6, unicastV6, unicastV6Only, broadcastIn6}}, + } + for _, testCase := range testCases { + for _, flow := range testCase.tests { + t.Run(fmt.Sprintf("%s:flow:%s", testCase.name, flow), func(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() - c.createEndpointForFlow(flow) + c.createEndpointForFlow(flow) + option := testCase.getReceiveOption + name := testCase.name - // Verify that setting and reading the option works. - v, err := c.ep.GetSockOptBool(tcpip.ReceiveTOSOption) - if err != nil { - c.t.Fatal("GetSockOptBool(tcpip.ReceiveTOSOption) failed:", err) - } - // Test for expected default value. - if v != false { - c.t.Errorf("got GetSockOptBool(tcpip.ReceiveTOSOption) = %t, want = %t", v, false) - } + // Verify that setting and reading the option works. + v, err := c.ep.GetSockOptBool(option) + if err != nil { + c.t.Errorf("GetSockoptBool(%s) failed: %s", name, err) + } + // Test for expected default value. + if v != false { + c.t.Errorf("got GetSockOptBool(%s) = %t, want = %t", name, v, false) + } - want := true - if err := c.ep.SetSockOptBool(tcpip.ReceiveTOSOption, want); err != nil { - c.t.Fatalf("SetSockOptBool(tcpip.ReceiveTOSOption, %t) failed: %s", want, err) - } + want := true + if err := c.ep.SetSockOptBool(option, want); err != nil { + c.t.Fatalf("SetSockOptBool(%s, %t) failed: %s", name, want, err) + } - got, err := c.ep.GetSockOptBool(tcpip.ReceiveTOSOption) - if err != nil { - c.t.Fatal("GetSockOptBool(tcpip.ReceiveTOSOption) failed:", err) - } - if got != want { - c.t.Fatalf("got GetSockOptBool(tcpip.ReceiveTOSOption) = %t, want = %t", got, want) - } + got, err := c.ep.GetSockOptBool(option) + if err != nil { + c.t.Errorf("GetSockoptBool(%s) failed: %s", name, err) + } - // Verify that the correct received TOS is handed through as - // ancillary data to the ControlMessages struct. - if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { - c.t.Fatal("Bind failed:", err) - } - testRead(c, flow, checker.ReceiveTOS(testTOS)) - }) + if got != want { + c.t.Errorf("got GetSockOptBool(%s) = %t, want = %t", name, got, want) + } + + // Verify that the correct received TOS or TClass is handed through as + // ancillary data to the ControlMessages struct. + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + c.t.Fatalf("Bind failed: %s", err) + } + switch option { + case tcpip.ReceiveTClassOption: + testRead(c, flow, checker.ReceiveTClass(testTOS)) + case tcpip.ReceiveTOSOption: + testRead(c, flow, checker.ReceiveTOS(testTOS)) + default: + t.Fatalf("unknown test variant: %s", name) + } + }) + } } } diff --git a/test/syscalls/linux/ip_socket_test_util.h b/test/syscalls/linux/ip_socket_test_util.h index 083ebbcf0..39fd6709d 100644 --- a/test/syscalls/linux/ip_socket_test_util.h +++ b/test/syscalls/linux/ip_socket_test_util.h @@ -84,20 +84,20 @@ SocketPairKind DualStackUDPBidirectionalBindSocketPair(int type); // SocketPairs created with AF_INET and the given type. SocketPairKind IPv4UDPUnboundSocketPair(int type); -// IPv4UDPUnboundSocketPair returns a SocketKind that represents -// a SimpleSocket created with AF_INET, SOCK_DGRAM, and the given type. +// IPv4UDPUnboundSocket returns a SocketKind that represents a SimpleSocket +// created with AF_INET, SOCK_DGRAM, and the given type. SocketKind IPv4UDPUnboundSocket(int type); -// IPv6UDPUnboundSocketPair returns a SocketKind that represents -// a SimpleSocket created with AF_INET6, SOCK_DGRAM, and the given type. +// IPv6UDPUnboundSocket returns a SocketKind that represents a SimpleSocket +// created with AF_INET6, SOCK_DGRAM, and the given type. SocketKind IPv6UDPUnboundSocket(int type); -// IPv4TCPUnboundSocketPair returns a SocketKind that represents -// a SimpleSocket created with AF_INET, SOCK_STREAM and the given type. +// IPv4TCPUnboundSocket returns a SocketKind that represents a SimpleSocket +// created with AF_INET, SOCK_STREAM and the given type. SocketKind IPv4TCPUnboundSocket(int type); -// IPv6TCPUnboundSocketPair returns a SocketKind that represents -// a SimpleSocket created with AF_INET6, SOCK_STREAM and the given type. +// IPv6TCPUnboundSocket returns a SocketKind that represents a SimpleSocket +// created with AF_INET6, SOCK_STREAM and the given type. SocketKind IPv6TCPUnboundSocket(int type); // IfAddrHelper is a helper class that determines the local interfaces present diff --git a/test/syscalls/linux/socket_ip_udp_generic.cc b/test/syscalls/linux/socket_ip_udp_generic.cc index db5663ecd..1c533fdf2 100644 --- a/test/syscalls/linux/socket_ip_udp_generic.cc +++ b/test/syscalls/linux/socket_ip_udp_generic.cc @@ -14,6 +14,7 @@ #include "test/syscalls/linux/socket_ip_udp_generic.h" +#include #include #include #include @@ -209,46 +210,6 @@ TEST_P(UDPSocketPairTest, SetMulticastLoopChar) { EXPECT_EQ(get, kSockOptOn); } -// Ensure that Receiving TOS is off by default. -TEST_P(UDPSocketPairTest, RecvTosDefault) { - auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); - - int get = -1; - socklen_t get_len = sizeof(get); - ASSERT_THAT( - getsockopt(sockets->first_fd(), IPPROTO_IP, IP_RECVTOS, &get, &get_len), - SyscallSucceedsWithValue(0)); - EXPECT_EQ(get_len, sizeof(get)); - EXPECT_EQ(get, kSockOptOff); -} - -// Test that setting and getting IP_RECVTOS works as expected. -TEST_P(UDPSocketPairTest, SetRecvTos) { - auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); - - ASSERT_THAT(setsockopt(sockets->first_fd(), IPPROTO_IP, IP_RECVTOS, - &kSockOptOff, sizeof(kSockOptOff)), - SyscallSucceeds()); - - int get = -1; - socklen_t get_len = sizeof(get); - ASSERT_THAT( - getsockopt(sockets->first_fd(), IPPROTO_IP, IP_RECVTOS, &get, &get_len), - SyscallSucceedsWithValue(0)); - EXPECT_EQ(get_len, sizeof(get)); - EXPECT_EQ(get, kSockOptOff); - - ASSERT_THAT(setsockopt(sockets->first_fd(), IPPROTO_IP, IP_RECVTOS, - &kSockOptOn, sizeof(kSockOptOn)), - SyscallSucceeds()); - - ASSERT_THAT( - getsockopt(sockets->first_fd(), IPPROTO_IP, IP_RECVTOS, &get, &get_len), - SyscallSucceedsWithValue(0)); - EXPECT_EQ(get_len, sizeof(get)); - EXPECT_EQ(get, kSockOptOn); -} - TEST_P(UDPSocketPairTest, ReuseAddrDefault) { auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); @@ -401,5 +362,97 @@ TEST_P(UDPSocketPairTest, SetAndGetIPPKTINFO) { EXPECT_EQ(get_len, sizeof(get)); } +// Holds TOS or TClass information for IPv4 or IPv6 respectively. +struct RecvTosOption { + int level; + int option; +}; + +RecvTosOption GetRecvTosOption(int domain) { + TEST_CHECK(domain == AF_INET || domain == AF_INET6); + RecvTosOption opt; + switch (domain) { + case AF_INET: + opt.level = IPPROTO_IP; + opt.option = IP_RECVTOS; + break; + case AF_INET6: + opt.level = IPPROTO_IPV6; + opt.option = IPV6_RECVTCLASS; + break; + } + return opt; +} + +// Ensure that Receiving TOS or TCLASS is off by default. +TEST_P(UDPSocketPairTest, RecvTosDefault) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + RecvTosOption t = GetRecvTosOption(GetParam().domain); + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT( + getsockopt(sockets->first_fd(), t.level, t.option, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kSockOptOff); +} + +// Test that setting and getting IP_RECVTOS or IPV6_RECVTCLASS works as +// expected. +TEST_P(UDPSocketPairTest, SetRecvTos) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + RecvTosOption t = GetRecvTosOption(GetParam().domain); + + ASSERT_THAT(setsockopt(sockets->first_fd(), t.level, t.option, &kSockOptOff, + sizeof(kSockOptOff)), + SyscallSucceeds()); + + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT( + getsockopt(sockets->first_fd(), t.level, t.option, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kSockOptOff); + + ASSERT_THAT(setsockopt(sockets->first_fd(), t.level, t.option, &kSockOptOn, + sizeof(kSockOptOn)), + SyscallSucceeds()); + + ASSERT_THAT( + getsockopt(sockets->first_fd(), t.level, t.option, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kSockOptOn); +} + +// Test that any socket (including IPv6 only) accepts the IPv4 TOS option: this +// mirrors behavior in linux. +TEST_P(UDPSocketPairTest, TOSRecvMismatch) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + RecvTosOption t = GetRecvTosOption(AF_INET); + int get = -1; + socklen_t get_len = sizeof(get); + + ASSERT_THAT( + getsockopt(sockets->first_fd(), t.level, t.option, &get, &get_len), + SyscallSucceedsWithValue(0)); +} + +// Test that an IPv4 socket does not support the IPv6 TClass option. +TEST_P(UDPSocketPairTest, TClassRecvMismatch) { + // This should only test AF_INET sockets for the mismatch behavior. + SKIP_IF(GetParam().domain != AF_INET); + + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + int get = -1; + socklen_t get_len = sizeof(get); + + ASSERT_THAT(getsockopt(sockets->first_fd(), IPPROTO_IPV6, IPV6_RECVTCLASS, + &get, &get_len), + SyscallFailsWithErrno(EOPNOTSUPP)); +} + } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/udp_socket_test_cases.cc b/test/syscalls/linux/udp_socket_test_cases.cc index 9f8de6b48..57b1a357c 100644 --- a/test/syscalls/linux/udp_socket_test_cases.cc +++ b/test/syscalls/linux/udp_socket_test_cases.cc @@ -1349,9 +1349,6 @@ TEST_P(UdpSocketTest, TimestampIoctlPersistence) { // outgoing packets, and that a receiving socket with IP_RECVTOS or // IPV6_RECVTCLASS will create the corresponding control message. TEST_P(UdpSocketTest, SetAndReceiveTOS) { - // TODO(b/144868438): IPV6_RECVTCLASS not supported for netstack. - SKIP_IF((GetParam() != AddressFamily::kIpv4) && IsRunningOnGvisor() && - !IsRunningWithHostinet()); ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); ASSERT_THAT(connect(t_, addr_[0], addrlen_), SyscallSucceeds()); @@ -1422,7 +1419,6 @@ TEST_P(UdpSocketTest, SetAndReceiveTOS) { // TOS byte on outgoing packets, and that a receiving socket with IP_RECVTOS or // IPV6_RECVTCLASS will create the corresponding control message. TEST_P(UdpSocketTest, SendAndReceiveTOS) { - // TODO(b/144868438): IPV6_RECVTCLASS not supported for netstack. // TODO(b/146661005): Setting TOS via cmsg not supported for netstack. SKIP_IF(IsRunningOnGvisor() && !IsRunningWithHostinet()); ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); -- cgit v1.2.3 From 2daa21e4d73f2297a8bca32c76100333e9ac4af4 Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Wed, 19 Feb 2020 16:47:58 -0800 Subject: Internal change. PiperOrigin-RevId: 296088213 --- pkg/sentry/socket/netstack/provider.go | 2 ++ 1 file changed, 2 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/provider.go b/pkg/sentry/socket/netstack/provider.go index 5afff2564..5f181f017 100644 --- a/pkg/sentry/socket/netstack/provider.go +++ b/pkg/sentry/socket/netstack/provider.go @@ -75,6 +75,8 @@ func getTransportProtocol(ctx context.Context, stype linux.SockType, protocol in switch protocol { case syscall.IPPROTO_ICMP: return header.ICMPv4ProtocolNumber, true, nil + case syscall.IPPROTO_ICMPV6: + return header.ICMPv6ProtocolNumber, true, nil case syscall.IPPROTO_UDP: return header.UDPProtocolNumber, true, nil case syscall.IPPROTO_TCP: -- cgit v1.2.3 From de68e1d8c437e6234f5e413d7cad6f892a4452d3 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Thu, 20 Feb 2020 01:12:03 -0500 Subject: Code Clean:Move getUserRegisters into dieArchSetup() and other small changes. Consistent with QEMU, getUserRegisters() should be an arch-specific function. So, it should be called in dieArchSetup(). With this patch and the pagetable/pcid patch, the kvm modules on Arm64 can be built successfully. Signed-off-by: Bin Lu --- pkg/sentry/platform/kvm/bluepill.go | 6 ------ pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go | 6 ++++++ pkg/sentry/platform/kvm/machine_arm64_unsafe.go | 24 ------------------------ 3 files changed, 6 insertions(+), 30 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/kvm/bluepill.go b/pkg/sentry/platform/kvm/bluepill.go index 35cd55fef..4b23f7803 100644 --- a/pkg/sentry/platform/kvm/bluepill.go +++ b/pkg/sentry/platform/kvm/bluepill.go @@ -81,12 +81,6 @@ func (c *vCPU) die(context *arch.SignalContext64, msg string) { // Save the death message, which will be thrown. c.dieState.message = msg - // Reload all registers to have an accurate stack trace when we return - // to host mode. This means that the stack should be unwound correctly. - if errno := c.getUserRegisters(&c.dieState.guestRegs); errno != 0 { - throw(msg) - } - // Setup the trampoline. dieArchSetup(c, context, &c.dieState.guestRegs) } diff --git a/pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go b/pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go index a63a6a071..99cac665d 100644 --- a/pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go @@ -31,6 +31,12 @@ import ( // //go:nosplit func dieArchSetup(c *vCPU, context *arch.SignalContext64, guestRegs *userRegs) { + // Reload all registers to have an accurate stack trace when we return + // to host mode. This means that the stack should be unwound correctly. + if errno := c.getUserRegisters(&c.dieState.guestRegs); errno != 0 { + throw(c.dieState.message) + } + // If the vCPU is in user mode, we set the stack to the stored stack // value in the vCPU itself. We don't want to unwind the user stack. if guestRegs.RFLAGS&ring0.UserFlagsSet == ring0.UserFlagsSet { diff --git a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go index 1c8384e6b..b531f2f85 100644 --- a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go @@ -29,30 +29,6 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) -// setMemoryRegion initializes a region. -// -// This may be called from bluepillHandler, and therefore returns an errno -// directly (instead of wrapping in an error) to avoid allocations. -// -//go:nosplit -func (m *machine) setMemoryRegion(slot int, physical, length, virtual uintptr) syscall.Errno { - userRegion := userMemoryRegion{ - slot: uint32(slot), - flags: 0, - guestPhysAddr: uint64(physical), - memorySize: uint64(length), - userspaceAddr: uint64(virtual), - } - - // Set the region. - _, _, errno := syscall.RawSyscall( - syscall.SYS_IOCTL, - uintptr(m.fd), - _KVM_SET_USER_MEMORY_REGION, - uintptr(unsafe.Pointer(&userRegion))) - return errno -} - type kvmVcpuInit struct { target uint32 features [7]uint32 -- cgit v1.2.3 From a369c88c0c4ece5239855000d28df045111c1be7 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Wed, 29 Jan 2020 04:33:06 -0500 Subject: Lazy-fpsimd support patch series#1: add Arm64-fpsimd support to arch module This patch defines the structures and adds the implementations for fpsimd initialization. Signed-off-by: Bin Lu --- pkg/sentry/arch/arch_aarch64.go | 30 +++++++++++++++++++----------- pkg/sentry/arch/arch_arm64.go | 16 +++++++++++++--- pkg/sentry/arch/signal_arm64.go | 19 ++++++++++++++++--- 3 files changed, 48 insertions(+), 17 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/arch/arch_aarch64.go b/pkg/sentry/arch/arch_aarch64.go index 3b6987665..53039465d 100644 --- a/pkg/sentry/arch/arch_aarch64.go +++ b/pkg/sentry/arch/arch_aarch64.go @@ -32,29 +32,35 @@ import ( const ( // SyscallWidth is the width of insturctions. SyscallWidth = 4 + + // fpsimdMagic is the magic number which is used in fpsimd_context. + fpsimdMagic = 0x46508001 + + // fpsimdContextSize is the size of fpsimd_context. + fpsimdContextSize = 0x210 ) // aarch64FPState is aarch64 floating point state. type aarch64FPState []byte -// initAarch64FPState (defined in asm files) sets up initial state. -func initAarch64FPState(data *FloatingPointData) { - // TODO(gvisor.dev/issue/1238): floating-point is not supported. +// initAarch64FPState sets up initial state. +func initAarch64FPState(data aarch64FPState) { + binary.LittleEndian.PutUint32(data, fpsimdMagic) + binary.LittleEndian.PutUint32(data[4:], fpsimdContextSize) } func newAarch64FPStateSlice() []byte { - return alignedBytes(4096, 32)[:4096] + return alignedBytes(4096, 16)[:fpsimdContextSize] } // newAarch64FPState returns an initialized floating point state. // // The returned state is large enough to store all floating point state // supported by host, even if the app won't use much of it due to a restricted -// FeatureSet. Since they may still be able to see state not advertised by -// CPUID we must ensure it does not contain any sentry state. +// FeatureSet. func newAarch64FPState() aarch64FPState { f := aarch64FPState(newAarch64FPStateSlice()) - initAarch64FPState(f.FloatingPointData()) + initAarch64FPState(f) return f } @@ -133,10 +139,10 @@ func (s State) Proto() *rpb.Registers { // Fork creates and returns an identical copy of the state. func (s *State) Fork() State { - // TODO(gvisor.dev/issue/1238): floating-point is not supported. return State{ - Regs: s.Regs, - FeatureSet: s.FeatureSet, + Regs: s.Regs, + aarch64FPState: s.aarch64FPState.fork(), + FeatureSet: s.FeatureSet, } } @@ -285,8 +291,10 @@ func New(arch Arch, fs *cpuid.FeatureSet) Context { case ARM64: return &context64{ State{ - FeatureSet: fs, + aarch64FPState: newAarch64FPState(), + FeatureSet: fs, }, + []aarch64FPState(nil), } } panic(fmt.Sprintf("unknown architecture %v", arch)) diff --git a/pkg/sentry/arch/arch_arm64.go b/pkg/sentry/arch/arch_arm64.go index 94f1a808f..372b650b9 100644 --- a/pkg/sentry/arch/arch_arm64.go +++ b/pkg/sentry/arch/arch_arm64.go @@ -68,6 +68,7 @@ const ( // context64 represents an ARM64 context. type context64 struct { State + sigFPState []aarch64FPState // fpstate to be restored on sigreturn. } // Arch implements Context.Arch. @@ -75,10 +76,19 @@ func (c *context64) Arch() Arch { return ARM64 } +func (c *context64) copySigFPState() []aarch64FPState { + var sigfps []aarch64FPState + for _, s := range c.sigFPState { + sigfps = append(sigfps, s.fork()) + } + return sigfps +} + // Fork returns an exact copy of this context. func (c *context64) Fork() Context { return &context64{ - State: c.State.Fork(), + State: c.State.Fork(), + sigFPState: c.copySigFPState(), } } @@ -137,8 +147,8 @@ func (c *context64) SetTLS(value uintptr) bool { return false } -// SetRSEQInterruptedIP implements Context.SetRSEQInterruptedIP. -func (c *context64) SetRSEQInterruptedIP(value uintptr) { +// SetOldRSeqInterruptedIP implements Context.SetOldRSeqInterruptedIP. +func (c *context64) SetOldRSeqInterruptedIP(value uintptr) { c.Regs.Regs[3] = uint64(value) } diff --git a/pkg/sentry/arch/signal_arm64.go b/pkg/sentry/arch/signal_arm64.go index 4f4cc46a8..28615f97f 100644 --- a/pkg/sentry/arch/signal_arm64.go +++ b/pkg/sentry/arch/signal_arm64.go @@ -30,14 +30,27 @@ type SignalContext64 struct { Sp uint64 Pc uint64 Pstate uint64 - _pad [8]byte // __attribute__((__aligned__(16))) - Reserved [4096]uint8 + _pad [8]byte // __attribute__((__aligned__(16))) + Fpsimd64 FpsimdContext // size = 528 + Reserved [3568]uint8 +} + +type aarch64Ctx struct { + Magic uint32 + Size uint32 +} + +type FpsimdContext struct { + Head aarch64Ctx + Fpsr uint32 + Fpcr uint32 + Vregs [64]uint64 // actually [32]uint128 } // UContext64 is equivalent to ucontext on arm64(arch/arm64/include/uapi/asm/ucontext.h). type UContext64 struct { Flags uint64 - Link *UContext64 + Link uint64 Stack SignalStack Sigset linux.SignalSet // glibc uses a 1024-bit sigset_t -- cgit v1.2.3 From 10ed60e4778e01a2813eea6e3c201826f75982a5 Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Thu, 20 Feb 2020 09:57:06 -0800 Subject: VFS2: Support memory mapping in tmpfs. tmpfs.fileDescription now implements ConfigureMMap. And tmpfs.regularFile implement memmap.Mappable. The methods are mostly unchanged from VFS1 tmpfs. PiperOrigin-RevId: 296234557 --- pkg/sentry/fsimpl/tmpfs/filesystem.go | 8 +- pkg/sentry/fsimpl/tmpfs/regular_file.go | 233 +++++++++++++++++++++++++++----- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 16 ++- 3 files changed, 213 insertions(+), 44 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 7f7b791c4..e1b551422 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -16,7 +16,6 @@ package tmpfs import ( "fmt" - "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" @@ -347,10 +346,9 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.Open return nil, err } if opts.Flags&linux.O_TRUNC != 0 { - impl.mu.Lock() - impl.data.Truncate(0, impl.memFile) - atomic.StoreUint64(&impl.size, 0) - impl.mu.Unlock() + if _, err := impl.truncate(0); err != nil { + return nil, err + } } return &fd.vfsfd, nil case *directory: diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index dab346a41..711442424 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -15,6 +15,7 @@ package tmpfs import ( + "fmt" "io" "math" "sync/atomic" @@ -22,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/safemem" + "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" @@ -34,25 +36,53 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) +// regularFile is a regular (=S_IFREG) tmpfs file. type regularFile struct { inode inode // memFile is a platform.File used to allocate pages to this regularFile. memFile *pgalloc.MemoryFile - // mu protects the fields below. - mu sync.RWMutex + // mapsMu protects mappings. + mapsMu sync.Mutex `state:"nosave"` + + // mappings tracks mappings of the file into memmap.MappingSpaces. + // + // Protected by mapsMu. + mappings memmap.MappingSet + + // writableMappingPages tracks how many pages of virtual memory are mapped + // as potentially writable from this file. If a page has multiple mappings, + // each mapping is counted separately. + // + // This counter is susceptible to overflow as we can potentially count + // mappings from many VMAs. We count pages rather than bytes to slightly + // mitigate this. + // + // Protected by mapsMu. + writableMappingPages uint64 + + // dataMu protects the fields below. + dataMu sync.RWMutex // data maps offsets into the file to offsets into memFile that store // the file's data. + // + // Protected by dataMu. data fsutil.FileRangeSet - // size is the size of data, but accessed using atomic memory - // operations to avoid locking in inode.stat(). - size uint64 - // seals represents file seals on this inode. + // + // Protected by dataMu. seals uint32 + + // size is the size of data. + // + // Protected by both dataMu and inode.mu; reading it requires holding + // either mutex, while writing requires holding both AND using atomics. + // Readers that do not require consistency (like Stat) may read the + // value atomically without holding either lock. + size uint64 } func (fs *filesystem) newRegularFile(creds *auth.Credentials, mode linux.FileMode) *inode { @@ -66,39 +96,170 @@ func (fs *filesystem) newRegularFile(creds *auth.Credentials, mode linux.FileMod // truncate grows or shrinks the file to the given size. It returns true if the // file size was updated. -func (rf *regularFile) truncate(size uint64) (bool, error) { - rf.mu.Lock() - defer rf.mu.Unlock() +func (rf *regularFile) truncate(newSize uint64) (bool, error) { + rf.inode.mu.Lock() + defer rf.inode.mu.Unlock() + return rf.truncateLocked(newSize) +} - if size == rf.size { +// Preconditions: rf.inode.mu must be held. +func (rf *regularFile) truncateLocked(newSize uint64) (bool, error) { + oldSize := rf.size + if newSize == oldSize { // Nothing to do. return false, nil } - if size > rf.size { - // Growing the file. + // Need to hold inode.mu and dataMu while modifying size. + rf.dataMu.Lock() + if newSize > oldSize { + // Can we grow the file? if rf.seals&linux.F_SEAL_GROW != 0 { - // Seal does not allow growth. + rf.dataMu.Unlock() return false, syserror.EPERM } - rf.size = size + // We only need to update the file size. + atomic.StoreUint64(&rf.size, newSize) + rf.dataMu.Unlock() return true, nil } - // Shrinking the file + // We are shrinking the file. First check if this is allowed. if rf.seals&linux.F_SEAL_SHRINK != 0 { - // Seal does not allow shrink. + rf.dataMu.Unlock() return false, syserror.EPERM } - // TODO(gvisor.dev/issues/1197): Invalidate mappings once we have - // mappings. + // Update the file size. + atomic.StoreUint64(&rf.size, newSize) + rf.dataMu.Unlock() + + // Invalidate past translations of truncated pages. + oldpgend := fs.OffsetPageEnd(int64(oldSize)) + newpgend := fs.OffsetPageEnd(int64(newSize)) + if newpgend < oldpgend { + rf.mapsMu.Lock() + rf.mappings.Invalidate(memmap.MappableRange{newpgend, oldpgend}, memmap.InvalidateOpts{ + // Compare Linux's mm/shmem.c:shmem_setattr() => + // mm/memory.c:unmap_mapping_range(evencows=1). + InvalidatePrivate: true, + }) + rf.mapsMu.Unlock() + } - rf.data.Truncate(size, rf.memFile) - rf.size = size + // We are now guaranteed that there are no translations of truncated pages, + // and can remove them. + rf.dataMu.Lock() + rf.data.Truncate(newSize, rf.memFile) + rf.dataMu.Unlock() return true, nil } +// AddMapping implements memmap.Mappable.AddMapping. +func (rf *regularFile) AddMapping(ctx context.Context, ms memmap.MappingSpace, ar usermem.AddrRange, offset uint64, writable bool) error { + rf.mapsMu.Lock() + defer rf.mapsMu.Unlock() + rf.dataMu.RLock() + defer rf.dataMu.RUnlock() + + // Reject writable mapping if F_SEAL_WRITE is set. + if rf.seals&linux.F_SEAL_WRITE != 0 && writable { + return syserror.EPERM + } + + rf.mappings.AddMapping(ms, ar, offset, writable) + if writable { + pagesBefore := rf.writableMappingPages + + // ar is guaranteed to be page aligned per memmap.Mappable. + rf.writableMappingPages += uint64(ar.Length() / usermem.PageSize) + + if rf.writableMappingPages < pagesBefore { + panic(fmt.Sprintf("Overflow while mapping potentially writable pages pointing to a tmpfs file. Before %v, after %v", pagesBefore, rf.writableMappingPages)) + } + } + + return nil +} + +// RemoveMapping implements memmap.Mappable.RemoveMapping. +func (rf *regularFile) RemoveMapping(ctx context.Context, ms memmap.MappingSpace, ar usermem.AddrRange, offset uint64, writable bool) { + rf.mapsMu.Lock() + defer rf.mapsMu.Unlock() + + rf.mappings.RemoveMapping(ms, ar, offset, writable) + + if writable { + pagesBefore := rf.writableMappingPages + + // ar is guaranteed to be page aligned per memmap.Mappable. + rf.writableMappingPages -= uint64(ar.Length() / usermem.PageSize) + + if rf.writableMappingPages > pagesBefore { + panic(fmt.Sprintf("Underflow while unmapping potentially writable pages pointing to a tmpfs file. Before %v, after %v", pagesBefore, rf.writableMappingPages)) + } + } +} + +// CopyMapping implements memmap.Mappable.CopyMapping. +func (rf *regularFile) CopyMapping(ctx context.Context, ms memmap.MappingSpace, srcAR, dstAR usermem.AddrRange, offset uint64, writable bool) error { + return rf.AddMapping(ctx, ms, dstAR, offset, writable) +} + +// Translate implements memmap.Mappable.Translate. +func (rf *regularFile) Translate(ctx context.Context, required, optional memmap.MappableRange, at usermem.AccessType) ([]memmap.Translation, error) { + rf.dataMu.Lock() + defer rf.dataMu.Unlock() + + // Constrain translations to f.attr.Size (rounded up) to prevent + // translation to pages that may be concurrently truncated. + pgend := fs.OffsetPageEnd(int64(rf.size)) + var beyondEOF bool + if required.End > pgend { + if required.Start >= pgend { + return nil, &memmap.BusError{io.EOF} + } + beyondEOF = true + required.End = pgend + } + if optional.End > pgend { + optional.End = pgend + } + + cerr := rf.data.Fill(ctx, required, optional, rf.memFile, usage.Tmpfs, func(_ context.Context, dsts safemem.BlockSeq, _ uint64) (uint64, error) { + // Newly-allocated pages are zeroed, so we don't need to do anything. + return dsts.NumBytes(), nil + }) + + var ts []memmap.Translation + var translatedEnd uint64 + for seg := rf.data.FindSegment(required.Start); seg.Ok() && seg.Start() < required.End; seg, _ = seg.NextNonEmpty() { + segMR := seg.Range().Intersect(optional) + ts = append(ts, memmap.Translation{ + Source: segMR, + File: rf.memFile, + Offset: seg.FileRangeOf(segMR).Start, + Perms: usermem.AnyAccess, + }) + translatedEnd = segMR.End + } + + // Don't return the error returned by f.data.Fill if it occurred outside of + // required. + if translatedEnd < required.End && cerr != nil { + return ts, &memmap.BusError{cerr} + } + if beyondEOF { + return ts, &memmap.BusError{io.EOF} + } + return ts, nil +} + +// InvalidateUnsavable implements memmap.Mappable.InvalidateUnsavable. +func (*regularFile) InvalidateUnsavable(context.Context) error { + return nil +} + type regularFileFD struct { fileDescription @@ -152,8 +313,10 @@ func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off // Overflow. return 0, syserror.EFBIG } + f.inode.mu.Lock() rw := getRegularFileReadWriter(f, offset) n, err := src.CopyInTo(ctx, rw) + f.inode.mu.Unlock() putRegularFileReadWriter(rw) return n, err } @@ -215,6 +378,12 @@ func (fd *regularFileFD) UnlockPOSIX(ctx context.Context, uid lock.UniqueID, rng return nil } +// ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. +func (fd *regularFileFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { + file := fd.inode().impl.(*regularFile) + return vfs.GenericConfigureMMap(&fd.vfsfd, file, opts) +} + // regularFileReadWriter implements safemem.Reader and Safemem.Writer. type regularFileReadWriter struct { file *regularFile @@ -244,14 +413,15 @@ func putRegularFileReadWriter(rw *regularFileReadWriter) { // ReadToBlocks implements safemem.Reader.ReadToBlocks. func (rw *regularFileReadWriter) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) { - rw.file.mu.RLock() + rw.file.dataMu.RLock() + defer rw.file.dataMu.RUnlock() + size := rw.file.size // Compute the range to read (limited by file size and overflow-checked). - if rw.off >= rw.file.size { - rw.file.mu.RUnlock() + if rw.off >= size { return 0, io.EOF } - end := rw.file.size + end := size if rend := rw.off + dsts.NumBytes(); rend > rw.off && rend < end { end = rend } @@ -265,7 +435,6 @@ func (rw *regularFileReadWriter) ReadToBlocks(dsts safemem.BlockSeq) (uint64, er // Get internal mappings. ims, err := rw.file.memFile.MapInternal(seg.FileRangeOf(seg.Range().Intersect(mr)), usermem.Read) if err != nil { - rw.file.mu.RUnlock() return done, err } @@ -275,7 +444,6 @@ func (rw *regularFileReadWriter) ReadToBlocks(dsts safemem.BlockSeq) (uint64, er rw.off += uint64(n) dsts = dsts.DropFirst64(n) if err != nil { - rw.file.mu.RUnlock() return done, err } @@ -291,7 +459,6 @@ func (rw *regularFileReadWriter) ReadToBlocks(dsts safemem.BlockSeq) (uint64, er rw.off += uint64(n) dsts = dsts.DropFirst64(n) if err != nil { - rw.file.mu.RUnlock() return done, err } @@ -299,13 +466,16 @@ func (rw *regularFileReadWriter) ReadToBlocks(dsts safemem.BlockSeq) (uint64, er seg, gap = gap.NextSegment(), fsutil.FileRangeGapIterator{} } } - rw.file.mu.RUnlock() return done, nil } // WriteFromBlocks implements safemem.Writer.WriteFromBlocks. +// +// Preconditions: inode.mu must be held. func (rw *regularFileReadWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) { - rw.file.mu.Lock() + // Hold dataMu so we can modify size. + rw.file.dataMu.Lock() + defer rw.file.dataMu.Unlock() // Compute the range to write (overflow-checked). end := rw.off + srcs.NumBytes() @@ -316,7 +486,6 @@ func (rw *regularFileReadWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, // Check if seals prevent either file growth or all writes. switch { case rw.file.seals&linux.F_SEAL_WRITE != 0: // Write sealed - rw.file.mu.Unlock() return 0, syserror.EPERM case end > rw.file.size && rw.file.seals&linux.F_SEAL_GROW != 0: // Grow sealed // When growth is sealed, Linux effectively allows writes which would @@ -338,7 +507,6 @@ func (rw *regularFileReadWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, } if end <= rw.off { // Truncation would result in no data being written. - rw.file.mu.Unlock() return 0, syserror.EPERM } } @@ -395,9 +563,8 @@ exitLoop: // If the write ends beyond the file's previous size, it causes the // file to grow. if rw.off > rw.file.size { - atomic.StoreUint64(&rw.file.size, rw.off) + rw.file.size = rw.off } - rw.file.mu.Unlock() return done, retErr } diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index c5bb17562..521206305 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -18,9 +18,10 @@ // Lock order: // // filesystem.mu -// regularFileFD.offMu -// regularFile.mu // inode.mu +// regularFileFD.offMu +// regularFile.mapsMu +// regularFile.dataMu package tmpfs import ( @@ -226,12 +227,15 @@ func (i *inode) tryIncRef() bool { func (i *inode) decRef() { if refs := atomic.AddInt64(&i.refs, -1); refs == 0 { - // This is unnecessary; it's mostly to simulate what tmpfs would do. if regFile, ok := i.impl.(*regularFile); ok { - regFile.mu.Lock() + // Hold inode.mu and regFile.dataMu while mutating + // size. + i.mu.Lock() + regFile.dataMu.Lock() regFile.data.DropAll(regFile.memFile) atomic.StoreUint64(®File.size, 0) - regFile.mu.Unlock() + regFile.dataMu.Unlock() + i.mu.Unlock() } } else if refs < 0 { panic("tmpfs.inode.decRef() called without holding a reference") @@ -320,7 +324,7 @@ func (i *inode) setStat(stat linux.Statx) error { if mask&linux.STATX_SIZE != 0 { switch impl := i.impl.(type) { case *regularFile: - updated, err := impl.truncate(stat.Size) + updated, err := impl.truncateLocked(stat.Size) if err != nil { return err } -- cgit v1.2.3 From 9bad87339a10545d267903e7739f8cd978fbd82a Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Thu, 20 Feb 2020 11:29:59 -0800 Subject: Better strace logging for epoll syscalls. MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Example: epoll_ctl(0x3 anon_inode:[eventpoll], EPOLL_CTL_ADD, 0x6 anon_inode:[eventfd], 0x7efe2fd92a80 {events=EPOLLIN|EPOLLOUT data=0x10203040506070a}) = 0x0 (4.411µs) epoll_wait(0x3 anon_inode:[eventpoll], 0x7efe2fd92b50 {{events=EPOLLOUT data=0x102030405060708}{events=EPOLLOUT data=0x102030405060708}{events=EPOLLOUT data=0x102030405060708}}, 0x3, 0xffffffff) = 0x3 (29.891µs) PiperOrigin-RevId: 296258146 --- pkg/abi/linux/epoll.go | 7 +++ pkg/sentry/strace/BUILD | 1 + pkg/sentry/strace/epoll.go | 89 ++++++++++++++++++++++++++++++++++++++ pkg/sentry/strace/linux64_amd64.go | 6 +-- pkg/sentry/strace/linux64_arm64.go | 4 +- pkg/sentry/strace/strace.go | 8 ++++ pkg/sentry/strace/syscalls.go | 10 +++++ 7 files changed, 120 insertions(+), 5 deletions(-) create mode 100644 pkg/sentry/strace/epoll.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/epoll.go b/pkg/abi/linux/epoll.go index 6e4de69da..1121a1a92 100644 --- a/pkg/abi/linux/epoll.go +++ b/pkg/abi/linux/epoll.go @@ -14,6 +14,10 @@ package linux +import ( + "gvisor.dev/gvisor/pkg/binary" +) + // Event masks. const ( EPOLLIN = 0x1 @@ -53,3 +57,6 @@ const ( EPOLL_CTL_DEL = 0x2 EPOLL_CTL_MOD = 0x3 ) + +// SizeOfEpollEvent is the size of EpollEvent struct. +var SizeOfEpollEvent = int(binary.Size(EpollEvent{})) diff --git a/pkg/sentry/strace/BUILD b/pkg/sentry/strace/BUILD index 2f39a6f2b..88d5db9fc 100644 --- a/pkg/sentry/strace/BUILD +++ b/pkg/sentry/strace/BUILD @@ -7,6 +7,7 @@ go_library( srcs = [ "capability.go", "clone.go", + "epoll.go", "futex.go", "linux64_amd64.go", "linux64_arm64.go", diff --git a/pkg/sentry/strace/epoll.go b/pkg/sentry/strace/epoll.go new file mode 100644 index 000000000..a6e48b836 --- /dev/null +++ b/pkg/sentry/strace/epoll.go @@ -0,0 +1,89 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package strace + +import ( + "fmt" + "strings" + + "gvisor.dev/gvisor/pkg/abi" + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/usermem" +) + +func epollEvent(t *kernel.Task, eventAddr usermem.Addr) string { + var e linux.EpollEvent + if _, err := t.CopyIn(eventAddr, &e); err != nil { + return fmt.Sprintf("%#x {error reading event: %v}", eventAddr, err) + } + var sb strings.Builder + fmt.Fprintf(&sb, "%#x ", eventAddr) + writeEpollEvent(&sb, e) + return sb.String() +} + +func epollEvents(t *kernel.Task, eventsAddr usermem.Addr, numEvents, maxBytes uint64) string { + var sb strings.Builder + fmt.Fprintf(&sb, "%#x {", eventsAddr) + addr := eventsAddr + for i := uint64(0); i < numEvents; i++ { + var e linux.EpollEvent + if _, err := t.CopyIn(addr, &e); err != nil { + fmt.Fprintf(&sb, "{error reading event at %#x: %v}", addr, err) + continue + } + writeEpollEvent(&sb, e) + if uint64(sb.Len()) >= maxBytes { + sb.WriteString("...") + break + } + if _, ok := addr.AddLength(uint64(linux.SizeOfEpollEvent)); !ok { + fmt.Fprintf(&sb, "{error reading event at %#x: EFAULT}", addr) + continue + } + } + sb.WriteString("}") + return sb.String() +} + +func writeEpollEvent(sb *strings.Builder, e linux.EpollEvent) { + events := epollEventEvents.Parse(uint64(e.Events)) + fmt.Fprintf(sb, "{events=%s data=[%#x, %#x]}", events, e.Data[0], e.Data[1]) +} + +var epollCtlOps = abi.ValueSet{ + linux.EPOLL_CTL_ADD: "EPOLL_CTL_ADD", + linux.EPOLL_CTL_DEL: "EPOLL_CTL_DEL", + linux.EPOLL_CTL_MOD: "EPOLL_CTL_MOD", +} + +var epollEventEvents = abi.FlagSet{ + {Flag: linux.EPOLLIN, Name: "EPOLLIN"}, + {Flag: linux.EPOLLPRI, Name: "EPOLLPRI"}, + {Flag: linux.EPOLLOUT, Name: "EPOLLOUT"}, + {Flag: linux.EPOLLERR, Name: "EPOLLERR"}, + {Flag: linux.EPOLLHUP, Name: "EPULLHUP"}, + {Flag: linux.EPOLLRDNORM, Name: "EPOLLRDNORM"}, + {Flag: linux.EPOLLRDBAND, Name: "EPOLLRDBAND"}, + {Flag: linux.EPOLLWRNORM, Name: "EPOLLWRNORM"}, + {Flag: linux.EPOLLWRBAND, Name: "EPOLLWRBAND"}, + {Flag: linux.EPOLLMSG, Name: "EPOLLMSG"}, + {Flag: linux.EPOLLRDHUP, Name: "EPOLLRDHUP"}, + {Flag: linux.EPOLLEXCLUSIVE, Name: "EPOLLEXCLUSIVE"}, + {Flag: linux.EPOLLWAKEUP, Name: "EPOLLWAKEUP"}, + {Flag: linux.EPOLLONESHOT, Name: "EPOLLONESHOT"}, + {Flag: linux.EPOLLET, Name: "EPOLLET"}, +} diff --git a/pkg/sentry/strace/linux64_amd64.go b/pkg/sentry/strace/linux64_amd64.go index a4de545e9..71b92eaee 100644 --- a/pkg/sentry/strace/linux64_amd64.go +++ b/pkg/sentry/strace/linux64_amd64.go @@ -256,8 +256,8 @@ var linuxAMD64 = SyscallMap{ 229: makeSyscallInfo("clock_getres", Hex, PostTimespec), 230: makeSyscallInfo("clock_nanosleep", Hex, Hex, Timespec, PostTimespec), 231: makeSyscallInfo("exit_group", Hex), - 232: makeSyscallInfo("epoll_wait", Hex, Hex, Hex, Hex), - 233: makeSyscallInfo("epoll_ctl", Hex, Hex, FD, Hex), + 232: makeSyscallInfo("epoll_wait", FD, EpollEvents, Hex, Hex), + 233: makeSyscallInfo("epoll_ctl", FD, EpollCtlOp, FD, EpollEvent), 234: makeSyscallInfo("tgkill", Hex, Hex, Signal), 235: makeSyscallInfo("utimes", Path, Timeval), // 236: vserver (not implemented in the Linux kernel) @@ -305,7 +305,7 @@ var linuxAMD64 = SyscallMap{ 278: makeSyscallInfo("vmsplice", FD, Hex, Hex, Hex), 279: makeSyscallInfo("move_pages", Hex, Hex, Hex, Hex, Hex, Hex), 280: makeSyscallInfo("utimensat", FD, Path, UTimeTimespec, Hex), - 281: makeSyscallInfo("epoll_pwait", Hex, Hex, Hex, Hex, SigSet, Hex), + 281: makeSyscallInfo("epoll_pwait", FD, EpollEvents, Hex, Hex, SigSet, Hex), 282: makeSyscallInfo("signalfd", Hex, Hex, Hex), 283: makeSyscallInfo("timerfd_create", Hex, Hex), 284: makeSyscallInfo("eventfd", Hex), diff --git a/pkg/sentry/strace/linux64_arm64.go b/pkg/sentry/strace/linux64_arm64.go index 8bc38545f..bd7361a52 100644 --- a/pkg/sentry/strace/linux64_arm64.go +++ b/pkg/sentry/strace/linux64_arm64.go @@ -45,8 +45,8 @@ var linuxARM64 = SyscallMap{ 18: makeSyscallInfo("lookup_dcookie", Hex, Hex, Hex), 19: makeSyscallInfo("eventfd2", Hex, Hex), 20: makeSyscallInfo("epoll_create1", Hex), - 21: makeSyscallInfo("epoll_ctl", Hex, Hex, FD, Hex), - 22: makeSyscallInfo("epoll_pwait", Hex, Hex, Hex, Hex, SigSet, Hex), + 21: makeSyscallInfo("epoll_ctl", FD, EpollCtlOp, FD, EpollEvent), + 22: makeSyscallInfo("epoll_pwait", FD, EpollEvents, Hex, Hex, SigSet, Hex), 23: makeSyscallInfo("dup", FD), 24: makeSyscallInfo("dup3", FD, FD, Hex), 25: makeSyscallInfo("fcntl", FD, Hex, Hex), diff --git a/pkg/sentry/strace/strace.go b/pkg/sentry/strace/strace.go index 46cb2a1cc..77655558e 100644 --- a/pkg/sentry/strace/strace.go +++ b/pkg/sentry/strace/strace.go @@ -481,6 +481,12 @@ func (i *SyscallInfo) pre(t *kernel.Task, args arch.SyscallArguments, maximumBlo output = append(output, capData(t, args[arg-1].Pointer(), args[arg].Pointer())) case PollFDs: output = append(output, pollFDs(t, args[arg].Pointer(), uint(args[arg+1].Uint()), false)) + case EpollCtlOp: + output = append(output, epollCtlOps.Parse(uint64(args[arg].Int()))) + case EpollEvent: + output = append(output, epollEvent(t, args[arg].Pointer())) + case EpollEvents: + output = append(output, epollEvents(t, args[arg].Pointer(), 0 /* numEvents */, uint64(maximumBlobSize))) case SelectFDSet: output = append(output, fdSet(t, int(args[0].Int()), args[arg].Pointer())) case Oct: @@ -549,6 +555,8 @@ func (i *SyscallInfo) post(t *kernel.Task, args arch.SyscallArguments, rval uint output[arg] = capData(t, args[arg-1].Pointer(), args[arg].Pointer()) case PollFDs: output[arg] = pollFDs(t, args[arg].Pointer(), uint(args[arg+1].Uint()), true) + case EpollEvents: + output[arg] = epollEvents(t, args[arg].Pointer(), uint64(rval), uint64(maximumBlobSize)) case GetSockOptVal: output[arg] = getSockOptVal(t, args[arg-2].Uint64() /* level */, args[arg-1].Uint64() /* optName */, args[arg].Pointer() /* optVal */, args[arg+1].Pointer() /* optLen */, maximumBlobSize, rval) case SetSockOptVal: diff --git a/pkg/sentry/strace/syscalls.go b/pkg/sentry/strace/syscalls.go index 446d1e0f6..7e69b9279 100644 --- a/pkg/sentry/strace/syscalls.go +++ b/pkg/sentry/strace/syscalls.go @@ -228,6 +228,16 @@ const ( // SockOptLevel is the optname argument in getsockopt(2) and // setsockopt(2). SockOptName + + // EpollCtlOp is the op argument to epoll_ctl(2). + EpollCtlOp + + // EpollEvent is the event argument in epoll_ctl(2). + EpollEvent + + // EpollEvents is an array of struct epoll_event. It is the events + // argument in epoll_wait(2)/epoll_pwait(2). + EpollEvents ) // defaultFormat is the syscall argument format to use if the actual format is -- cgit v1.2.3 From d90d71474f4c82f742140fdf026821709845cece Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Thu, 20 Feb 2020 14:28:31 -0800 Subject: Remove bytes read/written from marshal.Marshallable API. Users of the API only care about whether the copy in/out succeeds in their entirety, which is already signalled by the returned error. PiperOrigin-RevId: 296297843 --- pkg/sentry/kernel/rseq.go | 2 +- pkg/sentry/syscalls/linux/sys_stat.go | 6 ++---- tools/go_marshal/gomarshal/generator_interfaces.go | 21 +++++++++++---------- tools/go_marshal/gomarshal/generator_tests.go | 2 +- tools/go_marshal/marshal/marshal.go | 4 ++-- 5 files changed, 17 insertions(+), 18 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/rseq.go b/pkg/sentry/kernel/rseq.go index 18416643b..ded95f532 100644 --- a/pkg/sentry/kernel/rseq.go +++ b/pkg/sentry/kernel/rseq.go @@ -304,7 +304,7 @@ func (t *Task) rseqAddrInterrupt() { } var cs linux.RSeqCriticalSection - if _, err := cs.CopyIn(t, critAddr); err != nil { + if err := cs.CopyIn(t, critAddr); err != nil { t.Debugf("Failed to copy critical section from %#x for rseq: %v", critAddr, err) t.forceSignal(linux.SIGSEGV, false /* unconditional */) t.SendSignal(SignalInfoPriv(linux.SIGSEGV)) diff --git a/pkg/sentry/syscalls/linux/sys_stat.go b/pkg/sentry/syscalls/linux/sys_stat.go index 8b66a9006..11f25e00d 100644 --- a/pkg/sentry/syscalls/linux/sys_stat.go +++ b/pkg/sentry/syscalls/linux/sys_stat.go @@ -131,8 +131,7 @@ func stat(t *kernel.Task, d *fs.Dirent, dirPath bool, statAddr usermem.Addr) err return err } s := statFromAttrs(t, d.Inode.StableAttr, uattr) - _, err = s.CopyOut(t, statAddr) - return err + return s.CopyOut(t, statAddr) } // fstat implements fstat for the given *fs.File. @@ -142,8 +141,7 @@ func fstat(t *kernel.Task, f *fs.File, statAddr usermem.Addr) error { return err } s := statFromAttrs(t, f.Dirent.Inode.StableAttr, uattr) - _, err = s.CopyOut(t, statAddr) - return err + return s.CopyOut(t, statAddr) } // Statx implements linux syscall statx(2). diff --git a/tools/go_marshal/gomarshal/generator_interfaces.go b/tools/go_marshal/gomarshal/generator_interfaces.go index 3aa299ccd..834c58cee 100644 --- a/tools/go_marshal/gomarshal/generator_interfaces.go +++ b/tools/go_marshal/gomarshal/generator_interfaces.go @@ -507,13 +507,14 @@ func (g *interfaceGenerator) emitMarshallable() { g.emit("// CopyOut implements marshal.Marshallable.CopyOut.\n") g.recordUsedImport("marshal") g.recordUsedImport("usermem") - g.emit("func (%s *%s) CopyOut(task marshal.Task, addr usermem.Addr) (int, error) {\n", g.r, g.typeName()) + g.emit("func (%s *%s) CopyOut(task marshal.Task, addr usermem.Addr) error {\n", g.r, g.typeName()) g.inIndent(func() { fallback := func() { g.emit("// Type %s doesn't have a packed layout in memory, fall back to MarshalBytes.\n", g.typeName()) g.emit("buf := task.CopyScratchBuffer(%s.SizeBytes())\n", g.r) g.emit("%s.MarshalBytes(buf)\n", g.r) - g.emit("return task.CopyOutBytes(addr, buf)\n") + g.emit("_, err := task.CopyOutBytes(addr, buf)\n") + g.emit("return err\n") } if thisPacked { g.recordUsedImport("reflect") @@ -539,11 +540,11 @@ func (g *interfaceGenerator) emitMarshallable() { g.emit("hdr.Len = %s.SizeBytes()\n", g.r) g.emit("hdr.Cap = %s.SizeBytes()\n\n", g.r) - g.emit("len, err := task.CopyOutBytes(addr, buf)\n") + g.emit("_, err := task.CopyOutBytes(addr, buf)\n") g.emit("// Since we bypassed the compiler's escape analysis, indicate that %s\n", g.r) g.emit("// must live until after the CopyOutBytes.\n") g.emit("runtime.KeepAlive(%s)\n", g.r) - g.emit("return len, err\n") + g.emit("return err\n") } else { fallback() } @@ -553,20 +554,20 @@ func (g *interfaceGenerator) emitMarshallable() { g.emit("// CopyIn implements marshal.Marshallable.CopyIn.\n") g.recordUsedImport("marshal") g.recordUsedImport("usermem") - g.emit("func (%s *%s) CopyIn(task marshal.Task, addr usermem.Addr) (int, error) {\n", g.r, g.typeName()) + g.emit("func (%s *%s) CopyIn(task marshal.Task, addr usermem.Addr) error {\n", g.r, g.typeName()) g.inIndent(func() { fallback := func() { g.emit("// Type %s doesn't have a packed layout in memory, fall back to UnmarshalBytes.\n", g.typeName()) g.emit("buf := task.CopyScratchBuffer(%s.SizeBytes())\n", g.r) - g.emit("n, err := task.CopyInBytes(addr, buf)\n") + g.emit("_, err := task.CopyInBytes(addr, buf)\n") g.emit("if err != nil {\n") g.inIndent(func() { - g.emit("return n, err\n") + g.emit("return err\n") }) g.emit("}\n") g.emit("%s.UnmarshalBytes(buf)\n", g.r) - g.emit("return n, nil\n") + g.emit("return nil\n") } if thisPacked { g.recordUsedImport("reflect") @@ -592,11 +593,11 @@ func (g *interfaceGenerator) emitMarshallable() { g.emit("hdr.Len = %s.SizeBytes()\n", g.r) g.emit("hdr.Cap = %s.SizeBytes()\n\n", g.r) - g.emit("len, err := task.CopyInBytes(addr, buf)\n") + g.emit("_, err := task.CopyInBytes(addr, buf)\n") g.emit("// Since we bypassed the compiler's escape analysis, indicate that %s\n", g.r) g.emit("// must live until after the CopyInBytes.\n") g.emit("runtime.KeepAlive(%s)\n", g.r) - g.emit("return len, err\n") + g.emit("return err\n") } else { fallback() } diff --git a/tools/go_marshal/gomarshal/generator_tests.go b/tools/go_marshal/gomarshal/generator_tests.go index 8c28b00d0..2326e7a07 100644 --- a/tools/go_marshal/gomarshal/generator_tests.go +++ b/tools/go_marshal/gomarshal/generator_tests.go @@ -92,7 +92,7 @@ func (g *testGenerator) emitTestNonZeroSize() { g.emit("x := &%s{}\n", g.typeName()) g.emit("if x.SizeBytes() == 0 {\n") g.inIndent(func() { - g.emit("t.Fatal(\"Marshallable.Size() should not return zero\")\n") + g.emit("t.Fatal(\"Marshallable.SizeBytes() should not return zero\")\n") }) g.emit("}\n") }) diff --git a/tools/go_marshal/marshal/marshal.go b/tools/go_marshal/marshal/marshal.go index 20353850d..f129788e0 100644 --- a/tools/go_marshal/marshal/marshal.go +++ b/tools/go_marshal/marshal/marshal.go @@ -91,12 +91,12 @@ type Marshallable interface { // marshalled does not escape. The implementation should avoid creating // extra copies in memory by directly deserializing to the object's // underlying memory. - CopyIn(task Task, addr usermem.Addr) (int, error) + CopyIn(task Task, addr usermem.Addr) error // CopyOut serializes a Marshallable type to a task's memory. This may only // be called from a task goroutine. This is more efficient than calling // MarshalUnsafe on Marshallable.Packed types, as the type being serialized // does not escape. The implementation should avoid creating extra copies in // memory by directly serializing from the object's underlying memory. - CopyOut(task Task, addr usermem.Addr) (int, error) + CopyOut(task Task, addr usermem.Addr) error } -- cgit v1.2.3 From 4a73bae269ae9f52a962ae3b08a17ccaacf7ba80 Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Thu, 20 Feb 2020 15:19:40 -0800 Subject: Initial network namespace support. TCP/IP will work with netstack networking. hostinet doesn't work, and sockets will have the same behavior as it is now. Before the userspace is able to create device, the default loopback device can be used to test. /proc/net and /sys/net will still be connected to the root network stack; this is the same behavior now. Issue #1833 PiperOrigin-RevId: 296309389 --- pkg/sentry/fs/proc/net.go | 5 +- pkg/sentry/fs/proc/sys_net.go | 4 +- pkg/sentry/fsimpl/proc/tasks_net.go | 5 +- pkg/sentry/fsimpl/proc/tasks_sys.go | 4 +- pkg/sentry/fsimpl/testutil/kernel.go | 1 + pkg/sentry/inet/BUILD | 1 + pkg/sentry/inet/namespace.go | 99 +++++++++++++++++++++++++ pkg/sentry/kernel/kernel.go | 26 ++++--- pkg/sentry/kernel/task.go | 9 +-- pkg/sentry/kernel/task_clone.go | 16 ++-- pkg/sentry/kernel/task_net.go | 19 +++-- pkg/sentry/kernel/task_start.go | 8 +- pkg/tcpip/time_unsafe.go | 2 + runsc/boot/BUILD | 2 +- runsc/boot/controller.go | 11 +-- runsc/boot/loader.go | 121 +++++++++++++++++++++---------- runsc/boot/network.go | 27 +++++++ runsc/boot/pprof.go | 18 ----- runsc/boot/pprof/BUILD | 11 +++ runsc/boot/pprof/pprof.go | 20 +++++ runsc/sandbox/network.go | 25 +------ test/syscalls/BUILD | 2 + test/syscalls/linux/BUILD | 17 +++++ test/syscalls/linux/network_namespace.cc | 121 +++++++++++++++++++++++++++++++ 24 files changed, 451 insertions(+), 123 deletions(-) create mode 100644 pkg/sentry/inet/namespace.go delete mode 100644 runsc/boot/pprof.go create mode 100644 runsc/boot/pprof/BUILD create mode 100644 runsc/boot/pprof/pprof.go create mode 100644 test/syscalls/linux/network_namespace.cc (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/proc/net.go b/pkg/sentry/fs/proc/net.go index 6f2775344..95d5817ff 100644 --- a/pkg/sentry/fs/proc/net.go +++ b/pkg/sentry/fs/proc/net.go @@ -43,7 +43,10 @@ import ( // newNet creates a new proc net entry. func (p *proc) newNetDir(ctx context.Context, k *kernel.Kernel, msrc *fs.MountSource) *fs.Inode { var contents map[string]*fs.Inode - if s := p.k.NetworkStack(); s != nil { + // TODO(gvisor.dev/issue/1833): Support for using the network stack in the + // network namespace of the calling process. We should make this per-process, + // a.k.a. /proc/PID/net, and make /proc/net a symlink to /proc/self/net. + if s := p.k.RootNetworkNamespace().Stack(); s != nil { contents = map[string]*fs.Inode{ "dev": seqfile.NewSeqFileInode(ctx, &netDev{s: s}, msrc), "snmp": seqfile.NewSeqFileInode(ctx, &netSnmp{s: s}, msrc), diff --git a/pkg/sentry/fs/proc/sys_net.go b/pkg/sentry/fs/proc/sys_net.go index 0772d4ae4..d4c4b533d 100644 --- a/pkg/sentry/fs/proc/sys_net.go +++ b/pkg/sentry/fs/proc/sys_net.go @@ -357,7 +357,9 @@ func (p *proc) newSysNetIPv4Dir(ctx context.Context, msrc *fs.MountSource, s ine func (p *proc) newSysNetDir(ctx context.Context, msrc *fs.MountSource) *fs.Inode { var contents map[string]*fs.Inode - if s := p.k.NetworkStack(); s != nil { + // TODO(gvisor.dev/issue/1833): Support for using the network stack in the + // network namespace of the calling process. + if s := p.k.RootNetworkNamespace().Stack(); s != nil { contents = map[string]*fs.Inode{ "ipv4": p.newSysNetIPv4Dir(ctx, msrc, s), "core": p.newSysNetCore(ctx, msrc, s), diff --git a/pkg/sentry/fsimpl/proc/tasks_net.go b/pkg/sentry/fsimpl/proc/tasks_net.go index 608fec017..d4e1812d8 100644 --- a/pkg/sentry/fsimpl/proc/tasks_net.go +++ b/pkg/sentry/fsimpl/proc/tasks_net.go @@ -39,7 +39,10 @@ import ( func newNetDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) *kernfs.Dentry { var contents map[string]*kernfs.Dentry - if stack := k.NetworkStack(); stack != nil { + // TODO(gvisor.dev/issue/1833): Support for using the network stack in the + // network namespace of the calling process. We should make this per-process, + // a.k.a. /proc/PID/net, and make /proc/net a symlink to /proc/self/net. + if stack := k.RootNetworkNamespace().Stack(); stack != nil { const ( arp = "IP address HW type Flags HW address Mask Device\n" netlink = "sk Eth Pid Groups Rmem Wmem Dump Locks Drops Inode\n" diff --git a/pkg/sentry/fsimpl/proc/tasks_sys.go b/pkg/sentry/fsimpl/proc/tasks_sys.go index c7ce74883..3d5dc463c 100644 --- a/pkg/sentry/fsimpl/proc/tasks_sys.go +++ b/pkg/sentry/fsimpl/proc/tasks_sys.go @@ -50,7 +50,9 @@ func newSysDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) *k func newSysNetDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) *kernfs.Dentry { var contents map[string]*kernfs.Dentry - if stack := k.NetworkStack(); stack != nil { + // TODO(gvisor.dev/issue/1833): Support for using the network stack in the + // network namespace of the calling process. + if stack := k.RootNetworkNamespace().Stack(); stack != nil { contents = map[string]*kernfs.Dentry{ "ipv4": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ "tcp_sack": newDentry(root, inoGen.NextIno(), 0644, &tcpSackData{stack: stack}), diff --git a/pkg/sentry/fsimpl/testutil/kernel.go b/pkg/sentry/fsimpl/testutil/kernel.go index d0be32e72..488478e29 100644 --- a/pkg/sentry/fsimpl/testutil/kernel.go +++ b/pkg/sentry/fsimpl/testutil/kernel.go @@ -128,6 +128,7 @@ func CreateTask(ctx context.Context, name string, tc *kernel.ThreadGroup, mntns ThreadGroup: tc, TaskContext: &kernel.TaskContext{Name: name}, Credentials: auth.CredentialsFromContext(ctx), + NetworkNamespace: k.RootNetworkNamespace(), AllowedCPUMask: sched.NewFullCPUSet(k.ApplicationCores()), UTSNamespace: kernel.UTSNamespaceFromContext(ctx), IPCNamespace: kernel.IPCNamespaceFromContext(ctx), diff --git a/pkg/sentry/inet/BUILD b/pkg/sentry/inet/BUILD index 334432abf..07bf39fed 100644 --- a/pkg/sentry/inet/BUILD +++ b/pkg/sentry/inet/BUILD @@ -10,6 +10,7 @@ go_library( srcs = [ "context.go", "inet.go", + "namespace.go", "test_stack.go", ], deps = [ diff --git a/pkg/sentry/inet/namespace.go b/pkg/sentry/inet/namespace.go new file mode 100644 index 000000000..c16667e7f --- /dev/null +++ b/pkg/sentry/inet/namespace.go @@ -0,0 +1,99 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package inet + +// Namespace represents a network namespace. See network_namespaces(7). +// +// +stateify savable +type Namespace struct { + // stack is the network stack implementation of this network namespace. + stack Stack `state:"nosave"` + + // creator allows kernel to create new network stack for network namespaces. + // If nil, no networking will function if network is namespaced. + creator NetworkStackCreator + + // isRoot indicates whether this is the root network namespace. + isRoot bool +} + +// NewRootNamespace creates the root network namespace, with creator +// allowing new network namespaces to be created. If creator is nil, no +// networking will function if the network is namespaced. +func NewRootNamespace(stack Stack, creator NetworkStackCreator) *Namespace { + return &Namespace{ + stack: stack, + creator: creator, + isRoot: true, + } +} + +// NewNamespace creates a new network namespace from the root. +func NewNamespace(root *Namespace) *Namespace { + n := &Namespace{ + creator: root.creator, + } + n.init() + return n +} + +// Stack returns the network stack of n. Stack may return nil if no network +// stack is configured. +func (n *Namespace) Stack() Stack { + return n.stack +} + +// IsRoot returns whether n is the root network namespace. +func (n *Namespace) IsRoot() bool { + return n.isRoot +} + +// RestoreRootStack restores the root network namespace with stack. This should +// only be called when restoring kernel. +func (n *Namespace) RestoreRootStack(stack Stack) { + if !n.isRoot { + panic("RestoreRootStack can only be called on root network namespace") + } + if n.stack != nil { + panic("RestoreRootStack called after a stack has already been set") + } + n.stack = stack +} + +func (n *Namespace) init() { + // Root network namespace will have stack assigned later. + if n.isRoot { + return + } + if n.creator != nil { + var err error + n.stack, err = n.creator.CreateStack() + if err != nil { + panic(err) + } + } +} + +// afterLoad is invoked by stateify. +func (n *Namespace) afterLoad() { + n.init() +} + +// NetworkStackCreator allows new instances of a network stack to be created. It +// is used by the kernel to create new network namespaces when requested. +type NetworkStackCreator interface { + // CreateStack creates a new network stack for a network namespace. + CreateStack() (Stack, error) +} diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 7da0368f1..c62fd6eb1 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -111,7 +111,7 @@ type Kernel struct { timekeeper *Timekeeper tasks *TaskSet rootUserNamespace *auth.UserNamespace - networkStack inet.Stack `state:"nosave"` + rootNetworkNamespace *inet.Namespace applicationCores uint useHostCores bool extraAuxv []arch.AuxEntry @@ -260,8 +260,9 @@ type InitKernelArgs struct { // RootUserNamespace is the root user namespace. RootUserNamespace *auth.UserNamespace - // NetworkStack is the TCP/IP network stack. NetworkStack may be nil. - NetworkStack inet.Stack + // RootNetworkNamespace is the root network namespace. If nil, no networking + // will be available. + RootNetworkNamespace *inet.Namespace // ApplicationCores is the number of logical CPUs visible to sandboxed // applications. The set of logical CPU IDs is [0, ApplicationCores); thus @@ -320,7 +321,10 @@ func (k *Kernel) Init(args InitKernelArgs) error { k.rootUTSNamespace = args.RootUTSNamespace k.rootIPCNamespace = args.RootIPCNamespace k.rootAbstractSocketNamespace = args.RootAbstractSocketNamespace - k.networkStack = args.NetworkStack + k.rootNetworkNamespace = args.RootNetworkNamespace + if k.rootNetworkNamespace == nil { + k.rootNetworkNamespace = inet.NewRootNamespace(nil, nil) + } k.applicationCores = args.ApplicationCores if args.UseHostCores { k.useHostCores = true @@ -543,8 +547,6 @@ func (ts *TaskSet) unregisterEpollWaiters() { func (k *Kernel) LoadFrom(r io.Reader, net inet.Stack, clocks sentrytime.Clocks) error { loadStart := time.Now() - k.networkStack = net - initAppCores := k.applicationCores // Load the pre-saved CPUID FeatureSet. @@ -575,6 +577,10 @@ func (k *Kernel) LoadFrom(r io.Reader, net inet.Stack, clocks sentrytime.Clocks) log.Infof("Kernel load stats: %s", &stats) log.Infof("Kernel load took [%s].", time.Since(kernelStart)) + // rootNetworkNamespace should be populated after loading the state file. + // Restore the root network stack. + k.rootNetworkNamespace.RestoreRootStack(net) + // Load the memory file's state. memoryStart := time.Now() if err := k.mf.LoadFrom(k.SupervisorContext(), r); err != nil { @@ -905,6 +911,7 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, FSContext: fsContext, FDTable: args.FDTable, Credentials: args.Credentials, + NetworkNamespace: k.RootNetworkNamespace(), AllowedCPUMask: sched.NewFullCPUSet(k.applicationCores), UTSNamespace: args.UTSNamespace, IPCNamespace: args.IPCNamespace, @@ -1255,10 +1262,9 @@ func (k *Kernel) RootAbstractSocketNamespace() *AbstractSocketNamespace { return k.rootAbstractSocketNamespace } -// NetworkStack returns the network stack. NetworkStack may return nil if no -// network stack is available. -func (k *Kernel) NetworkStack() inet.Stack { - return k.networkStack +// RootNetworkNamespace returns the root network namespace, always non-nil. +func (k *Kernel) RootNetworkNamespace() *inet.Namespace { + return k.rootNetworkNamespace } // GlobalInit returns the thread group with ID 1 in the root PID namespace, or diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index a3443ff21..e37e23231 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -486,13 +486,10 @@ type Task struct { numaPolicy int32 numaNodeMask uint64 - // If netns is true, the task is in a non-root network namespace. Network - // namespaces aren't currently implemented in full; being in a network - // namespace simply prevents the task from observing any network devices - // (including loopback) or using abstract socket addresses (see unix(7)). + // netns is the task's network namespace. netns is never nil. // - // netns is protected by mu. netns is owned by the task goroutine. - netns bool + // netns is protected by mu. + netns *inet.Namespace // If rseqPreempted is true, before the next call to p.Switch(), // interrupt rseq critical regions as defined by rseqAddr and diff --git a/pkg/sentry/kernel/task_clone.go b/pkg/sentry/kernel/task_clone.go index ba74b4c1c..78866f280 100644 --- a/pkg/sentry/kernel/task_clone.go +++ b/pkg/sentry/kernel/task_clone.go @@ -17,6 +17,7 @@ package kernel import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/bpf" + "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" ) @@ -54,8 +55,7 @@ type SharingOptions struct { NewUserNamespace bool // If NewNetworkNamespace is true, the task should have an independent - // network namespace. (Note that network namespaces are not really - // implemented; see comment on Task.netns for details.) + // network namespace. NewNetworkNamespace bool // If NewFiles is true, the task should use an independent file descriptor @@ -199,6 +199,11 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { ipcns = NewIPCNamespace(userns) } + netns := t.NetworkNamespace() + if opts.NewNetworkNamespace { + netns = inet.NewNamespace(netns) + } + // TODO(b/63601033): Implement CLONE_NEWNS. mntnsVFS2 := t.mountNamespaceVFS2 if mntnsVFS2 != nil { @@ -268,7 +273,7 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { FDTable: fdTable, Credentials: creds, Niceness: t.Niceness(), - NetworkNamespaced: t.netns, + NetworkNamespace: netns, AllowedCPUMask: t.CPUMask(), UTSNamespace: utsns, IPCNamespace: ipcns, @@ -283,9 +288,6 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { } else { cfg.InheritParent = t } - if opts.NewNetworkNamespace { - cfg.NetworkNamespaced = true - } nt, err := t.tg.pidns.owner.NewTask(cfg) if err != nil { if opts.NewThreadGroup { @@ -482,7 +484,7 @@ func (t *Task) Unshare(opts *SharingOptions) error { t.mu.Unlock() return syserror.EPERM } - t.netns = true + t.netns = inet.NewNamespace(t.netns) } if opts.NewUTSNamespace { if !haveCapSysAdmin { diff --git a/pkg/sentry/kernel/task_net.go b/pkg/sentry/kernel/task_net.go index 172a31e1d..f7711232c 100644 --- a/pkg/sentry/kernel/task_net.go +++ b/pkg/sentry/kernel/task_net.go @@ -22,14 +22,23 @@ import ( func (t *Task) IsNetworkNamespaced() bool { t.mu.Lock() defer t.mu.Unlock() - return t.netns + return !t.netns.IsRoot() } // NetworkContext returns the network stack used by the task. NetworkContext // may return nil if no network stack is available. +// +// TODO(gvisor.dev/issue/1833): Migrate callers of this method to +// NetworkNamespace(). func (t *Task) NetworkContext() inet.Stack { - if t.IsNetworkNamespaced() { - return nil - } - return t.k.networkStack + t.mu.Lock() + defer t.mu.Unlock() + return t.netns.Stack() +} + +// NetworkNamespace returns the network namespace observed by the task. +func (t *Task) NetworkNamespace() *inet.Namespace { + t.mu.Lock() + defer t.mu.Unlock() + return t.netns } diff --git a/pkg/sentry/kernel/task_start.go b/pkg/sentry/kernel/task_start.go index f9236a842..a5035bb7f 100644 --- a/pkg/sentry/kernel/task_start.go +++ b/pkg/sentry/kernel/task_start.go @@ -17,6 +17,7 @@ package kernel import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/futex" "gvisor.dev/gvisor/pkg/sentry/kernel/sched" @@ -65,9 +66,8 @@ type TaskConfig struct { // Niceness is the niceness of the new task. Niceness int - // If NetworkNamespaced is true, the new task should observe a non-root - // network namespace. - NetworkNamespaced bool + // NetworkNamespace is the network namespace to be used for the new task. + NetworkNamespace *inet.Namespace // AllowedCPUMask contains the cpus that this task can run on. AllowedCPUMask sched.CPUSet @@ -133,7 +133,7 @@ func (ts *TaskSet) newTask(cfg *TaskConfig) (*Task, error) { allowedCPUMask: cfg.AllowedCPUMask.Copy(), ioUsage: &usage.IO{}, niceness: cfg.Niceness, - netns: cfg.NetworkNamespaced, + netns: cfg.NetworkNamespace, utsns: cfg.UTSNamespace, ipcns: cfg.IPCNamespace, abstractSockets: cfg.AbstractSocketNamespace, diff --git a/pkg/tcpip/time_unsafe.go b/pkg/tcpip/time_unsafe.go index 48764b978..2f98a996f 100644 --- a/pkg/tcpip/time_unsafe.go +++ b/pkg/tcpip/time_unsafe.go @@ -25,6 +25,8 @@ import ( ) // StdClock implements Clock with the time package. +// +// +stateify savable type StdClock struct{} var _ Clock = (*StdClock)(nil) diff --git a/runsc/boot/BUILD b/runsc/boot/BUILD index ae4dd102a..26f68fe3d 100644 --- a/runsc/boot/BUILD +++ b/runsc/boot/BUILD @@ -19,7 +19,6 @@ go_library( "loader_amd64.go", "loader_arm64.go", "network.go", - "pprof.go", "strace.go", "user.go", ], @@ -91,6 +90,7 @@ go_library( "//pkg/usermem", "//runsc/boot/filter", "//runsc/boot/platforms", + "//runsc/boot/pprof", "//runsc/specutils", "@com_github_golang_protobuf//proto:go_default_library", "@com_github_opencontainers_runtime-spec//specs-go:go_default_library", diff --git a/runsc/boot/controller.go b/runsc/boot/controller.go index 9c9e94864..17e774e0c 100644 --- a/runsc/boot/controller.go +++ b/runsc/boot/controller.go @@ -32,6 +32,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/watchdog" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/urpc" + "gvisor.dev/gvisor/runsc/boot/pprof" "gvisor.dev/gvisor/runsc/specutils" ) @@ -142,7 +143,7 @@ func newController(fd int, l *Loader) (*controller, error) { } srv.Register(manager) - if eps, ok := l.k.NetworkStack().(*netstack.Stack); ok { + if eps, ok := l.k.RootNetworkNamespace().Stack().(*netstack.Stack); ok { net := &Network{ Stack: eps.Stack, } @@ -341,7 +342,7 @@ func (cm *containerManager) Restore(o *RestoreOpts, _ *struct{}) error { return fmt.Errorf("creating memory file: %v", err) } k.SetMemoryFile(mf) - networkStack := cm.l.k.NetworkStack() + networkStack := cm.l.k.RootNetworkNamespace().Stack() cm.l.k = k // Set up the restore environment. @@ -365,9 +366,9 @@ func (cm *containerManager) Restore(o *RestoreOpts, _ *struct{}) error { } if cm.l.conf.ProfileEnable { - // initializePProf opens /proc/self/maps, so has to be - // called before installing seccomp filters. - initializePProf() + // pprof.Initialize opens /proc/self/maps, so has to be called before + // installing seccomp filters. + pprof.Initialize() } // Seccomp filters have to be applied before parsing the state file. diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index eef43b9df..e7ca98134 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -49,6 +49,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/watchdog" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/link/loopback" "gvisor.dev/gvisor/pkg/tcpip/link/sniffer" "gvisor.dev/gvisor/pkg/tcpip/network/arp" "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" @@ -60,6 +61,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/transport/udp" "gvisor.dev/gvisor/runsc/boot/filter" _ "gvisor.dev/gvisor/runsc/boot/platforms" // register all platforms. + "gvisor.dev/gvisor/runsc/boot/pprof" "gvisor.dev/gvisor/runsc/specutils" // Include supported socket providers. @@ -230,11 +232,8 @@ func New(args Args) (*Loader, error) { return nil, fmt.Errorf("enabling strace: %v", err) } - // Create an empty network stack because the network namespace may be empty at - // this point. Netns is configured before Run() is called. Netstack is - // configured using a control uRPC message. Host network is configured inside - // Run(). - networkStack, err := newEmptyNetworkStack(args.Conf, k, k) + // Create root network namespace/stack. + netns, err := newRootNetworkNamespace(args.Conf, k, k) if err != nil { return nil, fmt.Errorf("creating network: %v", err) } @@ -277,7 +276,7 @@ func New(args Args) (*Loader, error) { FeatureSet: cpuid.HostFeatureSet(), Timekeeper: tk, RootUserNamespace: creds.UserNamespace, - NetworkStack: networkStack, + RootNetworkNamespace: netns, ApplicationCores: uint(args.NumCPU), Vdso: vdso, RootUTSNamespace: kernel.NewUTSNamespace(args.Spec.Hostname, args.Spec.Hostname, creds.UserNamespace), @@ -466,7 +465,7 @@ func (l *Loader) run() error { // Delay host network configuration to this point because network namespace // is configured after the loader is created and before Run() is called. log.Debugf("Configuring host network") - stack := l.k.NetworkStack().(*hostinet.Stack) + stack := l.k.RootNetworkNamespace().Stack().(*hostinet.Stack) if err := stack.Configure(); err != nil { return err } @@ -485,7 +484,7 @@ func (l *Loader) run() error { // l.restore is set by the container manager when a restore call is made. if !l.restore { if l.conf.ProfileEnable { - initializePProf() + pprof.Initialize() } // Finally done with all configuration. Setup filters before user code @@ -908,48 +907,92 @@ func (l *Loader) WaitExit() kernel.ExitStatus { return l.k.GlobalInit().ExitStatus() } -func newEmptyNetworkStack(conf *Config, clock tcpip.Clock, uniqueID stack.UniqueID) (inet.Stack, error) { +func newRootNetworkNamespace(conf *Config, clock tcpip.Clock, uniqueID stack.UniqueID) (*inet.Namespace, error) { + // Create an empty network stack because the network namespace may be empty at + // this point. Netns is configured before Run() is called. Netstack is + // configured using a control uRPC message. Host network is configured inside + // Run(). switch conf.Network { case NetworkHost: - return hostinet.NewStack(), nil + // No network namespacing support for hostinet yet, hence creator is nil. + return inet.NewRootNamespace(hostinet.NewStack(), nil), nil case NetworkNone, NetworkSandbox: - // NetworkNone sets up loopback using netstack. - netProtos := []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol(), arp.NewProtocol()} - transProtos := []stack.TransportProtocol{tcp.NewProtocol(), udp.NewProtocol(), icmp.NewProtocol4()} - s := netstack.Stack{stack.New(stack.Options{ - NetworkProtocols: netProtos, - TransportProtocols: transProtos, - Clock: clock, - Stats: netstack.Metrics, - HandleLocal: true, - // Enable raw sockets for users with sufficient - // privileges. - RawFactory: raw.EndpointFactory{}, - UniqueID: uniqueID, - })} - - // Enable SACK Recovery. - if err := s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SACKEnabled(true)); err != nil { - return nil, fmt.Errorf("failed to enable SACK: %v", err) + s, err := newEmptySandboxNetworkStack(clock, uniqueID) + if err != nil { + return nil, err } + creator := &sandboxNetstackCreator{ + clock: clock, + uniqueID: uniqueID, + } + return inet.NewRootNamespace(s, creator), nil - // Set default TTLs as required by socket/netstack. - s.Stack.SetNetworkProtocolOption(ipv4.ProtocolNumber, tcpip.DefaultTTLOption(netstack.DefaultTTL)) - s.Stack.SetNetworkProtocolOption(ipv6.ProtocolNumber, tcpip.DefaultTTLOption(netstack.DefaultTTL)) + default: + panic(fmt.Sprintf("invalid network configuration: %v", conf.Network)) + } - // Enable Receive Buffer Auto-Tuning. - if err := s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.ModerateReceiveBufferOption(true)); err != nil { - return nil, fmt.Errorf("SetTransportProtocolOption failed: %v", err) - } +} - s.FillDefaultIPTables() +func newEmptySandboxNetworkStack(clock tcpip.Clock, uniqueID stack.UniqueID) (inet.Stack, error) { + netProtos := []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol(), arp.NewProtocol()} + transProtos := []stack.TransportProtocol{tcp.NewProtocol(), udp.NewProtocol(), icmp.NewProtocol4()} + s := netstack.Stack{stack.New(stack.Options{ + NetworkProtocols: netProtos, + TransportProtocols: transProtos, + Clock: clock, + Stats: netstack.Metrics, + HandleLocal: true, + // Enable raw sockets for users with sufficient + // privileges. + RawFactory: raw.EndpointFactory{}, + UniqueID: uniqueID, + })} - return &s, nil + // Enable SACK Recovery. + if err := s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SACKEnabled(true)); err != nil { + return nil, fmt.Errorf("failed to enable SACK: %v", err) + } - default: - panic(fmt.Sprintf("invalid network configuration: %v", conf.Network)) + // Set default TTLs as required by socket/netstack. + s.Stack.SetNetworkProtocolOption(ipv4.ProtocolNumber, tcpip.DefaultTTLOption(netstack.DefaultTTL)) + s.Stack.SetNetworkProtocolOption(ipv6.ProtocolNumber, tcpip.DefaultTTLOption(netstack.DefaultTTL)) + + // Enable Receive Buffer Auto-Tuning. + if err := s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.ModerateReceiveBufferOption(true)); err != nil { + return nil, fmt.Errorf("SetTransportProtocolOption failed: %v", err) + } + + s.FillDefaultIPTables() + + return &s, nil +} + +// sandboxNetstackCreator implements kernel.NetworkStackCreator. +// +// +stateify savable +type sandboxNetstackCreator struct { + clock tcpip.Clock + uniqueID stack.UniqueID +} + +// CreateStack implements kernel.NetworkStackCreator.CreateStack. +func (f *sandboxNetstackCreator) CreateStack() (inet.Stack, error) { + s, err := newEmptySandboxNetworkStack(f.clock, f.uniqueID) + if err != nil { + return nil, err } + + // Setup loopback. + n := &Network{Stack: s.(*netstack.Stack).Stack} + nicID := tcpip.NICID(f.uniqueID.UniqueID()) + link := DefaultLoopbackLink + linkEP := loopback.New() + if err := n.createNICWithAddrs(nicID, link.Name, linkEP, link.Addresses); err != nil { + return nil, err + } + + return s, nil } // signal sends a signal to one or more processes in a container. If PID is 0, diff --git a/runsc/boot/network.go b/runsc/boot/network.go index 6a8765ec8..bee6ee336 100644 --- a/runsc/boot/network.go +++ b/runsc/boot/network.go @@ -17,6 +17,7 @@ package boot import ( "fmt" "net" + "strings" "syscall" "gvisor.dev/gvisor/pkg/log" @@ -31,6 +32,32 @@ import ( "gvisor.dev/gvisor/pkg/urpc" ) +var ( + // DefaultLoopbackLink contains IP addresses and routes of "127.0.0.1/8" and + // "::1/8" on "lo" interface. + DefaultLoopbackLink = LoopbackLink{ + Name: "lo", + Addresses: []net.IP{ + net.IP("\x7f\x00\x00\x01"), + net.IPv6loopback, + }, + Routes: []Route{ + { + Destination: net.IPNet{ + IP: net.IPv4(0x7f, 0, 0, 0), + Mask: net.IPv4Mask(0xff, 0, 0, 0), + }, + }, + { + Destination: net.IPNet{ + IP: net.IPv6loopback, + Mask: net.IPMask(strings.Repeat("\xff", net.IPv6len)), + }, + }, + }, + } +) + // Network exposes methods that can be used to configure a network stack. type Network struct { Stack *stack.Stack diff --git a/runsc/boot/pprof.go b/runsc/boot/pprof.go deleted file mode 100644 index 463362f02..000000000 --- a/runsc/boot/pprof.go +++ /dev/null @@ -1,18 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package boot - -func initializePProf() { -} diff --git a/runsc/boot/pprof/BUILD b/runsc/boot/pprof/BUILD new file mode 100644 index 000000000..29cb42b2f --- /dev/null +++ b/runsc/boot/pprof/BUILD @@ -0,0 +1,11 @@ +load("//tools:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_library( + name = "pprof", + srcs = ["pprof.go"], + visibility = [ + "//runsc:__subpackages__", + ], +) diff --git a/runsc/boot/pprof/pprof.go b/runsc/boot/pprof/pprof.go new file mode 100644 index 000000000..1ded20dee --- /dev/null +++ b/runsc/boot/pprof/pprof.go @@ -0,0 +1,20 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package pprof provides a stub to initialize custom profilers. +package pprof + +// Initialize will be called at boot for initializing custom profilers. +func Initialize() { +} diff --git a/runsc/sandbox/network.go b/runsc/sandbox/network.go index 99e143696..bc093fba5 100644 --- a/runsc/sandbox/network.go +++ b/runsc/sandbox/network.go @@ -21,7 +21,6 @@ import ( "path/filepath" "runtime" "strconv" - "strings" "syscall" specs "github.com/opencontainers/runtime-spec/specs-go" @@ -75,30 +74,8 @@ func setupNetwork(conn *urpc.Client, pid int, spec *specs.Spec, conf *boot.Confi } func createDefaultLoopbackInterface(conn *urpc.Client) error { - link := boot.LoopbackLink{ - Name: "lo", - Addresses: []net.IP{ - net.IP("\x7f\x00\x00\x01"), - net.IPv6loopback, - }, - Routes: []boot.Route{ - { - Destination: net.IPNet{ - - IP: net.IPv4(0x7f, 0, 0, 0), - Mask: net.IPv4Mask(0xff, 0, 0, 0), - }, - }, - { - Destination: net.IPNet{ - IP: net.IPv6loopback, - Mask: net.IPMask(strings.Repeat("\xff", net.IPv6len)), - }, - }, - }, - } if err := conn.Call(boot.NetworkCreateLinksAndRoutes, &boot.CreateLinksAndRoutesArgs{ - LoopbackLinks: []boot.LoopbackLink{link}, + LoopbackLinks: []boot.LoopbackLink{boot.DefaultLoopbackLink}, }, nil); err != nil { return fmt.Errorf("creating loopback link and routes: %v", err) } diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index d69ac8356..d1977d4de 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -258,6 +258,8 @@ syscall_test( syscall_test(test = "//test/syscalls/linux:munmap_test") +syscall_test(test = "//test/syscalls/linux:network_namespace_test") + syscall_test( add_overlay = True, test = "//test/syscalls/linux:open_create_test", diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 05a818795..aa303af84 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -3639,6 +3639,23 @@ cc_binary( ], ) +cc_binary( + name = "network_namespace_test", + testonly = 1, + srcs = ["network_namespace.cc"], + linkstatic = 1, + deps = [ + ":socket_test_util", + gtest, + "//test/util:capability_util", + "//test/util:memory_util", + "//test/util:test_main", + "//test/util:test_util", + "//test/util:thread_util", + "@com_google_absl//absl/synchronization", + ], +) + cc_binary( name = "semaphore_test", testonly = 1, diff --git a/test/syscalls/linux/network_namespace.cc b/test/syscalls/linux/network_namespace.cc new file mode 100644 index 000000000..6ea48c263 --- /dev/null +++ b/test/syscalls/linux/network_namespace.cc @@ -0,0 +1,121 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include +#include +#include +#include +#include + +#include "gmock/gmock.h" +#include "gtest/gtest.h" +#include "absl/synchronization/notification.h" +#include "test/syscalls/linux/socket_test_util.h" +#include "test/util/capability_util.h" +#include "test/util/memory_util.h" +#include "test/util/test_util.h" +#include "test/util/thread_util.h" + +namespace gvisor { +namespace testing { + +namespace { + +using TestFunc = std::function; +using RunFunc = std::function; + +struct NamespaceStrategy { + RunFunc run; + + static NamespaceStrategy Of(RunFunc run) { + NamespaceStrategy s; + s.run = run; + return s; + } +}; + +PosixError RunWithUnshare(TestFunc fn) { + PosixError err = PosixError(-1, "function did not return a value"); + ScopedThread t([&] { + if (unshare(CLONE_NEWNET) != 0) { + err = PosixError(errno); + return; + } + err = fn(); + }); + t.Join(); + return err; +} + +PosixError RunWithClone(TestFunc fn) { + struct Args { + absl::Notification n; + TestFunc fn; + PosixError err; + }; + Args args; + args.fn = fn; + args.err = PosixError(-1, "function did not return a value"); + + ASSIGN_OR_RETURN_ERRNO( + Mapping child_stack, + MmapAnon(kPageSize, PROT_READ | PROT_WRITE, MAP_PRIVATE)); + pid_t child = clone( + +[](void *arg) { + Args *args = reinterpret_cast(arg); + args->err = args->fn(); + args->n.Notify(); + syscall(SYS_exit, 0); // Exit manually. No return address on stack. + return 0; + }, + reinterpret_cast(child_stack.addr() + kPageSize), + CLONE_NEWNET | CLONE_THREAD | CLONE_SIGHAND | CLONE_VM, &args); + if (child < 0) { + return PosixError(errno, "clone() failed"); + } + args.n.WaitForNotification(); + return args.err; +} + +class NetworkNamespaceTest + : public ::testing::TestWithParam {}; + +TEST_P(NetworkNamespaceTest, LoopbackExists) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN))); + + EXPECT_NO_ERRNO(GetParam().run([]() { + // TODO(gvisor.dev/issue/1833): Update this to test that only "lo" exists. + // Check loopback device exists. + int sock = socket(AF_INET, SOCK_DGRAM, 0); + if (sock < 0) { + return PosixError(errno, "socket() failed"); + } + struct ifreq ifr; + snprintf(ifr.ifr_name, IFNAMSIZ, "lo"); + if (ioctl(sock, SIOCGIFINDEX, &ifr) < 0) { + return PosixError(errno, "ioctl() failed, lo cannot be found"); + } + return NoError(); + })); +} + +INSTANTIATE_TEST_SUITE_P( + AllNetworkNamespaceTest, NetworkNamespaceTest, + ::testing::Values(NamespaceStrategy::Of(RunWithUnshare), + NamespaceStrategy::Of(RunWithClone))); + +} // namespace + +} // namespace testing +} // namespace gvisor -- cgit v1.2.3 From b8f56c79be40d9c75f4e2f279c9d821d1c1c3569 Mon Sep 17 00:00:00 2001 From: Ting-Yu Wang Date: Fri, 21 Feb 2020 15:41:56 -0800 Subject: Implement tap/tun device in vfs. PiperOrigin-RevId: 296526279 --- pkg/abi/linux/BUILD | 1 + pkg/abi/linux/ioctl.go | 26 ++ pkg/abi/linux/ioctl_tun.go | 29 ++ pkg/sentry/fs/dev/BUILD | 5 + pkg/sentry/fs/dev/dev.go | 10 +- pkg/sentry/fs/dev/net_tun.go | 170 +++++++++++ pkg/syserror/syserror.go | 1 + pkg/tcpip/buffer/view.go | 6 + pkg/tcpip/link/channel/BUILD | 1 + pkg/tcpip/link/channel/channel.go | 180 +++++++++--- pkg/tcpip/link/tun/BUILD | 18 +- pkg/tcpip/link/tun/device.go | 352 +++++++++++++++++++++++ pkg/tcpip/link/tun/protocol.go | 56 ++++ pkg/tcpip/stack/nic.go | 32 +++ pkg/tcpip/stack/stack.go | 39 +++ test/syscalls/BUILD | 2 + test/syscalls/linux/BUILD | 30 ++ test/syscalls/linux/dev.cc | 7 + test/syscalls/linux/socket_netlink_route_util.cc | 163 +++++++++++ test/syscalls/linux/socket_netlink_route_util.h | 55 ++++ test/syscalls/linux/tuntap.cc | 346 ++++++++++++++++++++++ 21 files changed, 1490 insertions(+), 39 deletions(-) create mode 100644 pkg/abi/linux/ioctl_tun.go create mode 100644 pkg/sentry/fs/dev/net_tun.go create mode 100644 pkg/tcpip/link/tun/device.go create mode 100644 pkg/tcpip/link/tun/protocol.go create mode 100644 test/syscalls/linux/socket_netlink_route_util.cc create mode 100644 test/syscalls/linux/socket_netlink_route_util.h create mode 100644 test/syscalls/linux/tuntap.cc (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/BUILD b/pkg/abi/linux/BUILD index a89f34d4b..322d1ccc4 100644 --- a/pkg/abi/linux/BUILD +++ b/pkg/abi/linux/BUILD @@ -30,6 +30,7 @@ go_library( "futex.go", "inotify.go", "ioctl.go", + "ioctl_tun.go", "ip.go", "ipc.go", "limits.go", diff --git a/pkg/abi/linux/ioctl.go b/pkg/abi/linux/ioctl.go index 0e18db9ef..2062e6a4b 100644 --- a/pkg/abi/linux/ioctl.go +++ b/pkg/abi/linux/ioctl.go @@ -72,3 +72,29 @@ const ( SIOCGMIIPHY = 0x8947 SIOCGMIIREG = 0x8948 ) + +// ioctl(2) directions. Used to calculate requests number. +// Constants from asm-generic/ioctl.h. +const ( + _IOC_NONE = 0 + _IOC_WRITE = 1 + _IOC_READ = 2 +) + +// Constants from asm-generic/ioctl.h. +const ( + _IOC_NRBITS = 8 + _IOC_TYPEBITS = 8 + _IOC_SIZEBITS = 14 + _IOC_DIRBITS = 2 + + _IOC_NRSHIFT = 0 + _IOC_TYPESHIFT = _IOC_NRSHIFT + _IOC_NRBITS + _IOC_SIZESHIFT = _IOC_TYPESHIFT + _IOC_TYPEBITS + _IOC_DIRSHIFT = _IOC_SIZESHIFT + _IOC_SIZEBITS +) + +// IOC outputs the result of _IOC macro in asm-generic/ioctl.h. +func IOC(dir, typ, nr, size uint32) uint32 { + return uint32(dir)<<_IOC_DIRSHIFT | typ<<_IOC_TYPESHIFT | nr<<_IOC_NRSHIFT | size<<_IOC_SIZESHIFT +} diff --git a/pkg/abi/linux/ioctl_tun.go b/pkg/abi/linux/ioctl_tun.go new file mode 100644 index 000000000..c59c9c136 --- /dev/null +++ b/pkg/abi/linux/ioctl_tun.go @@ -0,0 +1,29 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package linux + +// ioctl(2) request numbers from linux/if_tun.h +var ( + TUNSETIFF = IOC(_IOC_WRITE, 'T', 202, 4) + TUNGETIFF = IOC(_IOC_READ, 'T', 210, 4) +) + +// Flags from net/if_tun.h +const ( + IFF_TUN = 0x0001 + IFF_TAP = 0x0002 + IFF_NO_PI = 0x1000 + IFF_NOFILTER = 0x1000 +) diff --git a/pkg/sentry/fs/dev/BUILD b/pkg/sentry/fs/dev/BUILD index 4c4b7d5cc..9b6bb26d0 100644 --- a/pkg/sentry/fs/dev/BUILD +++ b/pkg/sentry/fs/dev/BUILD @@ -9,6 +9,7 @@ go_library( "device.go", "fs.go", "full.go", + "net_tun.go", "null.go", "random.go", "tty.go", @@ -19,15 +20,19 @@ go_library( "//pkg/context", "//pkg/rand", "//pkg/safemem", + "//pkg/sentry/arch", "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", "//pkg/sentry/fs/ramfs", "//pkg/sentry/fs/tmpfs", + "//pkg/sentry/kernel", "//pkg/sentry/memmap", "//pkg/sentry/mm", "//pkg/sentry/pgalloc", + "//pkg/sentry/socket/netstack", "//pkg/syserror", + "//pkg/tcpip/link/tun", "//pkg/usermem", "//pkg/waiter", ], diff --git a/pkg/sentry/fs/dev/dev.go b/pkg/sentry/fs/dev/dev.go index 35bd23991..7e66c29b0 100644 --- a/pkg/sentry/fs/dev/dev.go +++ b/pkg/sentry/fs/dev/dev.go @@ -66,8 +66,8 @@ func newMemDevice(ctx context.Context, iops fs.InodeOperations, msrc *fs.MountSo }) } -func newDirectory(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - iops := ramfs.NewDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) +func newDirectory(ctx context.Context, contents map[string]*fs.Inode, msrc *fs.MountSource) *fs.Inode { + iops := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) return fs.NewInode(ctx, iops, msrc, fs.StableAttr{ DeviceID: devDevice.DeviceID(), InodeID: devDevice.NextIno(), @@ -111,7 +111,7 @@ func New(ctx context.Context, msrc *fs.MountSource) *fs.Inode { // A devpts is typically mounted at /dev/pts to provide // pseudoterminal support. Place an empty directory there for // the devpts to be mounted over. - "pts": newDirectory(ctx, msrc), + "pts": newDirectory(ctx, nil, msrc), // Similarly, applications expect a ptmx device at /dev/ptmx // connected to the terminals provided by /dev/pts/. Rather // than creating a device directly (which requires a hairy @@ -124,6 +124,10 @@ func New(ctx context.Context, msrc *fs.MountSource) *fs.Inode { "ptmx": newSymlink(ctx, "pts/ptmx", msrc), "tty": newCharacterDevice(ctx, newTTYDevice(ctx, fs.RootOwner, 0666), msrc, ttyDevMajor, ttyDevMinor), + + "net": newDirectory(ctx, map[string]*fs.Inode{ + "tun": newCharacterDevice(ctx, newNetTunDevice(ctx, fs.RootOwner, 0666), msrc, netTunDevMajor, netTunDevMinor), + }, msrc), } iops := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) diff --git a/pkg/sentry/fs/dev/net_tun.go b/pkg/sentry/fs/dev/net_tun.go new file mode 100644 index 000000000..755644488 --- /dev/null +++ b/pkg/sentry/fs/dev/net_tun.go @@ -0,0 +1,170 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package dev + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/socket/netstack" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/tcpip/link/tun" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +const ( + netTunDevMajor = 10 + netTunDevMinor = 200 +) + +// +stateify savable +type netTunInodeOperations struct { + fsutil.InodeGenericChecker `state:"nosave"` + fsutil.InodeNoExtendedAttributes `state:"nosave"` + fsutil.InodeNoopAllocate `state:"nosave"` + fsutil.InodeNoopRelease `state:"nosave"` + fsutil.InodeNoopTruncate `state:"nosave"` + fsutil.InodeNoopWriteOut `state:"nosave"` + fsutil.InodeNotDirectory `state:"nosave"` + fsutil.InodeNotMappable `state:"nosave"` + fsutil.InodeNotSocket `state:"nosave"` + fsutil.InodeNotSymlink `state:"nosave"` + fsutil.InodeVirtual `state:"nosave"` + + fsutil.InodeSimpleAttributes +} + +var _ fs.InodeOperations = (*netTunInodeOperations)(nil) + +func newNetTunDevice(ctx context.Context, owner fs.FileOwner, mode linux.FileMode) *netTunInodeOperations { + return &netTunInodeOperations{ + InodeSimpleAttributes: fsutil.NewInodeSimpleAttributes(ctx, owner, fs.FilePermsFromMode(mode), linux.TMPFS_MAGIC), + } +} + +// GetFile implements fs.InodeOperations.GetFile. +func (iops *netTunInodeOperations) GetFile(ctx context.Context, d *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + return fs.NewFile(ctx, d, flags, &netTunFileOperations{}), nil +} + +// +stateify savable +type netTunFileOperations struct { + fsutil.FileNoSeek `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoSplice `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + fsutil.FileUseInodeUnstableAttr `state:"nosave"` + + device tun.Device +} + +var _ fs.FileOperations = (*netTunFileOperations)(nil) + +// Release implements fs.FileOperations.Release. +func (fops *netTunFileOperations) Release() { + fops.device.Release() +} + +// Ioctl implements fs.FileOperations.Ioctl. +func (fops *netTunFileOperations) Ioctl(ctx context.Context, file *fs.File, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { + request := args[1].Uint() + data := args[2].Pointer() + + switch request { + case linux.TUNSETIFF: + t := kernel.TaskFromContext(ctx) + if t == nil { + panic("Ioctl should be called from a task context") + } + if !t.HasCapability(linux.CAP_NET_ADMIN) { + return 0, syserror.EPERM + } + stack, ok := t.NetworkContext().(*netstack.Stack) + if !ok { + return 0, syserror.EINVAL + } + + var req linux.IFReq + if _, err := usermem.CopyObjectIn(ctx, io, data, &req, usermem.IOOpts{ + AddressSpaceActive: true, + }); err != nil { + return 0, err + } + flags := usermem.ByteOrder.Uint16(req.Data[:]) + return 0, fops.device.SetIff(stack.Stack, req.Name(), flags) + + case linux.TUNGETIFF: + var req linux.IFReq + + copy(req.IFName[:], fops.device.Name()) + + // Linux adds IFF_NOFILTER (the same value as IFF_NO_PI unfortunately) when + // there is no sk_filter. See __tun_chr_ioctl() in net/drivers/tun.c. + flags := fops.device.Flags() | linux.IFF_NOFILTER + usermem.ByteOrder.PutUint16(req.Data[:], flags) + + _, err := usermem.CopyObjectOut(ctx, io, data, &req, usermem.IOOpts{ + AddressSpaceActive: true, + }) + return 0, err + + default: + return 0, syserror.ENOTTY + } +} + +// Write implements fs.FileOperations.Write. +func (fops *netTunFileOperations) Write(ctx context.Context, file *fs.File, src usermem.IOSequence, offset int64) (int64, error) { + data := make([]byte, src.NumBytes()) + if _, err := src.CopyIn(ctx, data); err != nil { + return 0, err + } + return fops.device.Write(data) +} + +// Read implements fs.FileOperations.Read. +func (fops *netTunFileOperations) Read(ctx context.Context, file *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { + data, err := fops.device.Read() + if err != nil { + return 0, err + } + n, err := dst.CopyOut(ctx, data) + if n > 0 && n < len(data) { + // Not an error for partial copying. Packet truncated. + err = nil + } + return int64(n), err +} + +// Readiness implements watier.Waitable.Readiness. +func (fops *netTunFileOperations) Readiness(mask waiter.EventMask) waiter.EventMask { + return fops.device.Readiness(mask) +} + +// EventRegister implements watier.Waitable.EventRegister. +func (fops *netTunFileOperations) EventRegister(e *waiter.Entry, mask waiter.EventMask) { + fops.device.EventRegister(e, mask) +} + +// EventUnregister implements watier.Waitable.EventUnregister. +func (fops *netTunFileOperations) EventUnregister(e *waiter.Entry) { + fops.device.EventUnregister(e) +} diff --git a/pkg/syserror/syserror.go b/pkg/syserror/syserror.go index 2269f6237..4b5a0fca6 100644 --- a/pkg/syserror/syserror.go +++ b/pkg/syserror/syserror.go @@ -29,6 +29,7 @@ var ( EACCES = error(syscall.EACCES) EAGAIN = error(syscall.EAGAIN) EBADF = error(syscall.EBADF) + EBADFD = error(syscall.EBADFD) EBUSY = error(syscall.EBUSY) ECHILD = error(syscall.ECHILD) ECONNREFUSED = error(syscall.ECONNREFUSED) diff --git a/pkg/tcpip/buffer/view.go b/pkg/tcpip/buffer/view.go index 150310c11..17e94c562 100644 --- a/pkg/tcpip/buffer/view.go +++ b/pkg/tcpip/buffer/view.go @@ -156,3 +156,9 @@ func (vv *VectorisedView) Append(vv2 VectorisedView) { vv.views = append(vv.views, vv2.views...) vv.size += vv2.size } + +// AppendView appends the given view into this vectorised view. +func (vv *VectorisedView) AppendView(v View) { + vv.views = append(vv.views, v) + vv.size += len(v) +} diff --git a/pkg/tcpip/link/channel/BUILD b/pkg/tcpip/link/channel/BUILD index 3974c464e..b8b93e78e 100644 --- a/pkg/tcpip/link/channel/BUILD +++ b/pkg/tcpip/link/channel/BUILD @@ -7,6 +7,7 @@ go_library( srcs = ["channel.go"], visibility = ["//visibility:public"], deps = [ + "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/stack", diff --git a/pkg/tcpip/link/channel/channel.go b/pkg/tcpip/link/channel/channel.go index 78d447acd..5944ba190 100644 --- a/pkg/tcpip/link/channel/channel.go +++ b/pkg/tcpip/link/channel/channel.go @@ -20,6 +20,7 @@ package channel import ( "context" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/stack" @@ -33,6 +34,118 @@ type PacketInfo struct { Route stack.Route } +// Notification is the interface for receiving notification from the packet +// queue. +type Notification interface { + // WriteNotify will be called when a write happens to the queue. + WriteNotify() +} + +// NotificationHandle is an opaque handle to the registered notification target. +// It can be used to unregister the notification when no longer interested. +// +// +stateify savable +type NotificationHandle struct { + n Notification +} + +type queue struct { + // mu protects fields below. + mu sync.RWMutex + // c is the outbound packet channel. Sending to c should hold mu. + c chan PacketInfo + numWrite int + numRead int + notify []*NotificationHandle +} + +func (q *queue) Close() { + close(q.c) +} + +func (q *queue) Read() (PacketInfo, bool) { + q.mu.Lock() + defer q.mu.Unlock() + select { + case p := <-q.c: + q.numRead++ + return p, true + default: + return PacketInfo{}, false + } +} + +func (q *queue) ReadContext(ctx context.Context) (PacketInfo, bool) { + // We have to receive from channel without holding the lock, since it can + // block indefinitely. This will cause a window that numWrite - numRead + // produces a larger number, but won't go to negative. numWrite >= numRead + // still holds. + select { + case pkt := <-q.c: + q.mu.Lock() + defer q.mu.Unlock() + q.numRead++ + return pkt, true + case <-ctx.Done(): + return PacketInfo{}, false + } +} + +func (q *queue) Write(p PacketInfo) bool { + wrote := false + + // It's important to make sure nobody can see numWrite until we increment it, + // so numWrite >= numRead holds. + q.mu.Lock() + select { + case q.c <- p: + wrote = true + q.numWrite++ + default: + } + notify := q.notify + q.mu.Unlock() + + if wrote { + // Send notification outside of lock. + for _, h := range notify { + h.n.WriteNotify() + } + } + return wrote +} + +func (q *queue) Num() int { + q.mu.RLock() + defer q.mu.RUnlock() + n := q.numWrite - q.numRead + if n < 0 { + panic("numWrite < numRead") + } + return n +} + +func (q *queue) AddNotify(notify Notification) *NotificationHandle { + q.mu.Lock() + defer q.mu.Unlock() + h := &NotificationHandle{n: notify} + q.notify = append(q.notify, h) + return h +} + +func (q *queue) RemoveNotify(handle *NotificationHandle) { + q.mu.Lock() + defer q.mu.Unlock() + // Make a copy, since we reads the array outside of lock when notifying. + notify := make([]*NotificationHandle, 0, len(q.notify)) + for _, h := range q.notify { + if h != handle { + notify = append(notify, h) + } + } + q.notify = notify +} + // Endpoint is link layer endpoint that stores outbound packets in a channel // and allows injection of inbound packets. type Endpoint struct { @@ -41,14 +154,16 @@ type Endpoint struct { linkAddr tcpip.LinkAddress LinkEPCapabilities stack.LinkEndpointCapabilities - // c is where outbound packets are queued. - c chan PacketInfo + // Outbound packet queue. + q *queue } // New creates a new channel endpoint. func New(size int, mtu uint32, linkAddr tcpip.LinkAddress) *Endpoint { return &Endpoint{ - c: make(chan PacketInfo, size), + q: &queue{ + c: make(chan PacketInfo, size), + }, mtu: mtu, linkAddr: linkAddr, } @@ -57,43 +172,36 @@ func New(size int, mtu uint32, linkAddr tcpip.LinkAddress) *Endpoint { // Close closes e. Further packet injections will panic. Reads continue to // succeed until all packets are read. func (e *Endpoint) Close() { - close(e.c) + e.q.Close() } -// Read does non-blocking read for one packet from the outbound packet queue. +// Read does non-blocking read one packet from the outbound packet queue. func (e *Endpoint) Read() (PacketInfo, bool) { - select { - case pkt := <-e.c: - return pkt, true - default: - return PacketInfo{}, false - } + return e.q.Read() } // ReadContext does blocking read for one packet from the outbound packet queue. // It can be cancelled by ctx, and in this case, it returns false. func (e *Endpoint) ReadContext(ctx context.Context) (PacketInfo, bool) { - select { - case pkt := <-e.c: - return pkt, true - case <-ctx.Done(): - return PacketInfo{}, false - } + return e.q.ReadContext(ctx) } // Drain removes all outbound packets from the channel and counts them. func (e *Endpoint) Drain() int { c := 0 for { - select { - case <-e.c: - c++ - default: + if _, ok := e.Read(); !ok { return c } + c++ } } +// NumQueued returns the number of packet queued for outbound. +func (e *Endpoint) NumQueued() int { + return e.q.Num() +} + // InjectInbound injects an inbound packet. func (e *Endpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { e.InjectLinkAddr(protocol, "", pkt) @@ -155,10 +263,7 @@ func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne Route: route, } - select { - case e.c <- p: - default: - } + e.q.Write(p) return nil } @@ -171,7 +276,6 @@ func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.Pac route.Release() payloadView := pkts[0].Data.ToView() n := 0 -packetLoop: for _, pkt := range pkts { off := pkt.DataOffset size := pkt.DataSize @@ -185,12 +289,10 @@ packetLoop: Route: route, } - select { - case e.c <- p: - n++ - default: - break packetLoop + if !e.q.Write(p) { + break } + n++ } return n, nil @@ -204,13 +306,21 @@ func (e *Endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { GSO: nil, } - select { - case e.c <- p: - default: - } + e.q.Write(p) return nil } // Wait implements stack.LinkEndpoint.Wait. func (*Endpoint) Wait() {} + +// AddNotify adds a notification target for receiving event about outgoing +// packets. +func (e *Endpoint) AddNotify(notify Notification) *NotificationHandle { + return e.q.AddNotify(notify) +} + +// RemoveNotify removes handle from the list of notification targets. +func (e *Endpoint) RemoveNotify(handle *NotificationHandle) { + e.q.RemoveNotify(handle) +} diff --git a/pkg/tcpip/link/tun/BUILD b/pkg/tcpip/link/tun/BUILD index e5096ea38..e0db6cf54 100644 --- a/pkg/tcpip/link/tun/BUILD +++ b/pkg/tcpip/link/tun/BUILD @@ -4,6 +4,22 @@ package(licenses = ["notice"]) go_library( name = "tun", - srcs = ["tun_unsafe.go"], + srcs = [ + "device.go", + "protocol.go", + "tun_unsafe.go", + ], visibility = ["//visibility:public"], + deps = [ + "//pkg/abi/linux", + "//pkg/refs", + "//pkg/sync", + "//pkg/syserror", + "//pkg/tcpip", + "//pkg/tcpip/buffer", + "//pkg/tcpip/header", + "//pkg/tcpip/link/channel", + "//pkg/tcpip/stack", + "//pkg/waiter", + ], ) diff --git a/pkg/tcpip/link/tun/device.go b/pkg/tcpip/link/tun/device.go new file mode 100644 index 000000000..6ff47a742 --- /dev/null +++ b/pkg/tcpip/link/tun/device.go @@ -0,0 +1,352 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tun + +import ( + "fmt" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/refs" + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/link/channel" + "gvisor.dev/gvisor/pkg/tcpip/stack" + "gvisor.dev/gvisor/pkg/waiter" +) + +const ( + // drivers/net/tun.c:tun_net_init() + defaultDevMtu = 1500 + + // Queue length for outbound packet, arriving at fd side for read. Overflow + // causes packet drops. gVisor implementation-specific. + defaultDevOutQueueLen = 1024 +) + +var zeroMAC [6]byte + +// Device is an opened /dev/net/tun device. +// +// +stateify savable +type Device struct { + waiter.Queue + + mu sync.RWMutex `state:"nosave"` + endpoint *tunEndpoint + notifyHandle *channel.NotificationHandle + flags uint16 +} + +// beforeSave is invoked by stateify. +func (d *Device) beforeSave() { + d.mu.Lock() + defer d.mu.Unlock() + // TODO(b/110961832): Restore the device to stack. At this moment, the stack + // is not savable. + if d.endpoint != nil { + panic("/dev/net/tun does not support save/restore when a device is associated with it.") + } +} + +// Release implements fs.FileOperations.Release. +func (d *Device) Release() { + d.mu.Lock() + defer d.mu.Unlock() + + // Decrease refcount if there is an endpoint associated with this file. + if d.endpoint != nil { + d.endpoint.RemoveNotify(d.notifyHandle) + d.endpoint.DecRef() + d.endpoint = nil + } +} + +// SetIff services TUNSETIFF ioctl(2) request. +func (d *Device) SetIff(s *stack.Stack, name string, flags uint16) error { + d.mu.Lock() + defer d.mu.Unlock() + + if d.endpoint != nil { + return syserror.EINVAL + } + + // Input validations. + isTun := flags&linux.IFF_TUN != 0 + isTap := flags&linux.IFF_TAP != 0 + supportedFlags := uint16(linux.IFF_TUN | linux.IFF_TAP | linux.IFF_NO_PI) + if isTap && isTun || !isTap && !isTun || flags&^supportedFlags != 0 { + return syserror.EINVAL + } + + prefix := "tun" + if isTap { + prefix = "tap" + } + + endpoint, err := attachOrCreateNIC(s, name, prefix) + if err != nil { + return syserror.EINVAL + } + + d.endpoint = endpoint + d.notifyHandle = d.endpoint.AddNotify(d) + d.flags = flags + return nil +} + +func attachOrCreateNIC(s *stack.Stack, name, prefix string) (*tunEndpoint, error) { + for { + // 1. Try to attach to an existing NIC. + if name != "" { + if nic, found := s.GetNICByName(name); found { + endpoint, ok := nic.LinkEndpoint().(*tunEndpoint) + if !ok { + // Not a NIC created by tun device. + return nil, syserror.EOPNOTSUPP + } + if !endpoint.TryIncRef() { + // Race detected: NIC got deleted in between. + continue + } + return endpoint, nil + } + } + + // 2. Creating a new NIC. + id := tcpip.NICID(s.UniqueID()) + endpoint := &tunEndpoint{ + Endpoint: channel.New(defaultDevOutQueueLen, defaultDevMtu, ""), + stack: s, + nicID: id, + name: name, + } + if endpoint.name == "" { + endpoint.name = fmt.Sprintf("%s%d", prefix, id) + } + err := s.CreateNICWithOptions(endpoint.nicID, endpoint, stack.NICOptions{ + Name: endpoint.name, + }) + switch err { + case nil: + return endpoint, nil + case tcpip.ErrDuplicateNICID: + // Race detected: A NIC has been created in between. + continue + default: + return nil, syserror.EINVAL + } + } +} + +// Write inject one inbound packet to the network interface. +func (d *Device) Write(data []byte) (int64, error) { + d.mu.RLock() + endpoint := d.endpoint + d.mu.RUnlock() + if endpoint == nil { + return 0, syserror.EBADFD + } + if !endpoint.IsAttached() { + return 0, syserror.EIO + } + + dataLen := int64(len(data)) + + // Packet information. + var pktInfoHdr PacketInfoHeader + if !d.hasFlags(linux.IFF_NO_PI) { + if len(data) < PacketInfoHeaderSize { + // Ignore bad packet. + return dataLen, nil + } + pktInfoHdr = PacketInfoHeader(data[:PacketInfoHeaderSize]) + data = data[PacketInfoHeaderSize:] + } + + // Ethernet header (TAP only). + var ethHdr header.Ethernet + if d.hasFlags(linux.IFF_TAP) { + if len(data) < header.EthernetMinimumSize { + // Ignore bad packet. + return dataLen, nil + } + ethHdr = header.Ethernet(data[:header.EthernetMinimumSize]) + data = data[header.EthernetMinimumSize:] + } + + // Try to determine network protocol number, default zero. + var protocol tcpip.NetworkProtocolNumber + switch { + case pktInfoHdr != nil: + protocol = pktInfoHdr.Protocol() + case ethHdr != nil: + protocol = ethHdr.Type() + } + + // Try to determine remote link address, default zero. + var remote tcpip.LinkAddress + switch { + case ethHdr != nil: + remote = ethHdr.SourceAddress() + default: + remote = tcpip.LinkAddress(zeroMAC[:]) + } + + pkt := tcpip.PacketBuffer{ + Data: buffer.View(data).ToVectorisedView(), + } + if ethHdr != nil { + pkt.LinkHeader = buffer.View(ethHdr) + } + endpoint.InjectLinkAddr(protocol, remote, pkt) + return dataLen, nil +} + +// Read reads one outgoing packet from the network interface. +func (d *Device) Read() ([]byte, error) { + d.mu.RLock() + endpoint := d.endpoint + d.mu.RUnlock() + if endpoint == nil { + return nil, syserror.EBADFD + } + + for { + info, ok := endpoint.Read() + if !ok { + return nil, syserror.ErrWouldBlock + } + + v, ok := d.encodePkt(&info) + if !ok { + // Ignore unsupported packet. + continue + } + return v, nil + } +} + +// encodePkt encodes packet for fd side. +func (d *Device) encodePkt(info *channel.PacketInfo) (buffer.View, bool) { + var vv buffer.VectorisedView + + // Packet information. + if !d.hasFlags(linux.IFF_NO_PI) { + hdr := make(PacketInfoHeader, PacketInfoHeaderSize) + hdr.Encode(&PacketInfoFields{ + Protocol: info.Proto, + }) + vv.AppendView(buffer.View(hdr)) + } + + // If the packet does not already have link layer header, and the route + // does not exist, we can't compute it. This is possibly a raw packet, tun + // device doesn't support this at the moment. + if info.Pkt.LinkHeader == nil && info.Route.RemoteLinkAddress == "" { + return nil, false + } + + // Ethernet header (TAP only). + if d.hasFlags(linux.IFF_TAP) { + // Add ethernet header if not provided. + if info.Pkt.LinkHeader == nil { + hdr := &header.EthernetFields{ + SrcAddr: info.Route.LocalLinkAddress, + DstAddr: info.Route.RemoteLinkAddress, + Type: info.Proto, + } + if hdr.SrcAddr == "" { + hdr.SrcAddr = d.endpoint.LinkAddress() + } + + eth := make(header.Ethernet, header.EthernetMinimumSize) + eth.Encode(hdr) + vv.AppendView(buffer.View(eth)) + } else { + vv.AppendView(info.Pkt.LinkHeader) + } + } + + // Append upper headers. + vv.AppendView(buffer.View(info.Pkt.Header.View()[len(info.Pkt.LinkHeader):])) + // Append data payload. + vv.Append(info.Pkt.Data) + + return vv.ToView(), true +} + +// Name returns the name of the attached network interface. Empty string if +// unattached. +func (d *Device) Name() string { + d.mu.RLock() + defer d.mu.RUnlock() + if d.endpoint != nil { + return d.endpoint.name + } + return "" +} + +// Flags returns the flags set for d. Zero value if unset. +func (d *Device) Flags() uint16 { + d.mu.RLock() + defer d.mu.RUnlock() + return d.flags +} + +func (d *Device) hasFlags(flags uint16) bool { + return d.flags&flags == flags +} + +// Readiness implements watier.Waitable.Readiness. +func (d *Device) Readiness(mask waiter.EventMask) waiter.EventMask { + if mask&waiter.EventIn != 0 { + d.mu.RLock() + endpoint := d.endpoint + d.mu.RUnlock() + if endpoint != nil && endpoint.NumQueued() == 0 { + mask &= ^waiter.EventIn + } + } + return mask & (waiter.EventIn | waiter.EventOut) +} + +// WriteNotify implements channel.Notification.WriteNotify. +func (d *Device) WriteNotify() { + d.Notify(waiter.EventIn) +} + +// tunEndpoint is the link endpoint for the NIC created by the tun device. +// +// It is ref-counted as multiple opening files can attach to the same NIC. +// The last owner is responsible for deleting the NIC. +type tunEndpoint struct { + *channel.Endpoint + + refs.AtomicRefCount + + stack *stack.Stack + nicID tcpip.NICID + name string +} + +// DecRef decrements refcount of e, removes NIC if refcount goes to 0. +func (e *tunEndpoint) DecRef() { + e.DecRefWithDestructor(func() { + e.stack.RemoveNIC(e.nicID) + }) +} diff --git a/pkg/tcpip/link/tun/protocol.go b/pkg/tcpip/link/tun/protocol.go new file mode 100644 index 000000000..89d9d91a9 --- /dev/null +++ b/pkg/tcpip/link/tun/protocol.go @@ -0,0 +1,56 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tun + +import ( + "encoding/binary" + + "gvisor.dev/gvisor/pkg/tcpip" +) + +const ( + // PacketInfoHeaderSize is the size of the packet information header. + PacketInfoHeaderSize = 4 + + offsetFlags = 0 + offsetProtocol = 2 +) + +// PacketInfoFields contains fields sent through the wire if IFF_NO_PI flag is +// not set. +type PacketInfoFields struct { + Flags uint16 + Protocol tcpip.NetworkProtocolNumber +} + +// PacketInfoHeader is the wire representation of the packet information sent if +// IFF_NO_PI flag is not set. +type PacketInfoHeader []byte + +// Encode encodes f into h. +func (h PacketInfoHeader) Encode(f *PacketInfoFields) { + binary.BigEndian.PutUint16(h[offsetFlags:][:2], f.Flags) + binary.BigEndian.PutUint16(h[offsetProtocol:][:2], uint16(f.Protocol)) +} + +// Flags returns the flag field in h. +func (h PacketInfoHeader) Flags() uint16 { + return binary.BigEndian.Uint16(h[offsetFlags:]) +} + +// Protocol returns the protocol field in h. +func (h PacketInfoHeader) Protocol() tcpip.NetworkProtocolNumber { + return tcpip.NetworkProtocolNumber(binary.BigEndian.Uint16(h[offsetProtocol:])) +} diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 862954ab2..46d3a6646 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -298,6 +298,33 @@ func (n *NIC) enable() *tcpip.Error { return nil } +// remove detaches NIC from the link endpoint, and marks existing referenced +// network endpoints expired. This guarantees no packets between this NIC and +// the network stack. +func (n *NIC) remove() *tcpip.Error { + n.mu.Lock() + defer n.mu.Unlock() + + // Detach from link endpoint, so no packet comes in. + n.linkEP.Attach(nil) + + // Remove permanent and permanentTentative addresses, so no packet goes out. + var errs []*tcpip.Error + for nid, ref := range n.mu.endpoints { + switch ref.getKind() { + case permanentTentative, permanent: + if err := n.removePermanentAddressLocked(nid.LocalAddress); err != nil { + errs = append(errs, err) + } + } + } + if len(errs) > 0 { + return errs[0] + } + + return nil +} + // becomeIPv6Router transitions n into an IPv6 router. // // When transitioning into an IPv6 router, host-only state (NDP discovered @@ -1302,6 +1329,11 @@ func (n *NIC) Stack() *Stack { return n.stack } +// LinkEndpoint returns the link endpoint of n. +func (n *NIC) LinkEndpoint() LinkEndpoint { + return n.linkEP +} + // isAddrTentative returns true if addr is tentative on n. // // Note that if addr is not associated with n, then this function will return diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index f0ed76fbe..900dd46c5 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -916,6 +916,18 @@ func (s *Stack) CreateNIC(id tcpip.NICID, ep LinkEndpoint) *tcpip.Error { return s.CreateNICWithOptions(id, ep, NICOptions{}) } +// GetNICByName gets the NIC specified by name. +func (s *Stack) GetNICByName(name string) (*NIC, bool) { + s.mu.RLock() + defer s.mu.RUnlock() + for _, nic := range s.nics { + if nic.Name() == name { + return nic, true + } + } + return nil, false +} + // EnableNIC enables the given NIC so that the link-layer endpoint can start // delivering packets to it. func (s *Stack) EnableNIC(id tcpip.NICID) *tcpip.Error { @@ -956,6 +968,33 @@ func (s *Stack) CheckNIC(id tcpip.NICID) bool { return nic.enabled() } +// RemoveNIC removes NIC and all related routes from the network stack. +func (s *Stack) RemoveNIC(id tcpip.NICID) *tcpip.Error { + s.mu.Lock() + defer s.mu.Unlock() + + nic, ok := s.nics[id] + if !ok { + return tcpip.ErrUnknownNICID + } + delete(s.nics, id) + + // Remove routes in-place. n tracks the number of routes written. + n := 0 + for i, r := range s.routeTable { + if r.NIC != id { + // Keep this route. + if i > n { + s.routeTable[n] = r + } + n++ + } + } + s.routeTable = s.routeTable[:n] + + return nic.remove() +} + // NICAddressRanges returns a map of NICIDs to their associated subnets. func (s *Stack) NICAddressRanges() map[tcpip.NICID][]tcpip.Subnet { s.mu.RLock() diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index d1977d4de..3518e862d 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -678,6 +678,8 @@ syscall_test( test = "//test/syscalls/linux:truncate_test", ) +syscall_test(test = "//test/syscalls/linux:tuntap_test") + syscall_test(test = "//test/syscalls/linux:udp_bind_test") syscall_test( diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index aa303af84..704bae17b 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -131,6 +131,17 @@ cc_library( ], ) +cc_library( + name = "socket_netlink_route_util", + testonly = 1, + srcs = ["socket_netlink_route_util.cc"], + hdrs = ["socket_netlink_route_util.h"], + deps = [ + ":socket_netlink_util", + "@com_google_absl//absl/types:optional", + ], +) + cc_library( name = "socket_test_util", testonly = 1, @@ -3430,6 +3441,25 @@ cc_binary( ], ) +cc_binary( + name = "tuntap_test", + testonly = 1, + srcs = ["tuntap.cc"], + linkstatic = 1, + deps = [ + ":socket_test_util", + gtest, + "//test/syscalls/linux:socket_netlink_route_util", + "//test/util:capability_util", + "//test/util:file_descriptor", + "//test/util:fs_util", + "//test/util:posix_error", + "//test/util:test_main", + "//test/util:test_util", + "@com_google_absl//absl/strings", + ], +) + cc_library( name = "udp_socket_test_cases", testonly = 1, diff --git a/test/syscalls/linux/dev.cc b/test/syscalls/linux/dev.cc index 4dd302eed..4e473268c 100644 --- a/test/syscalls/linux/dev.cc +++ b/test/syscalls/linux/dev.cc @@ -153,6 +153,13 @@ TEST(DevTest, TTYExists) { EXPECT_EQ(statbuf.st_mode, S_IFCHR | 0666); } +TEST(DevTest, NetTunExists) { + struct stat statbuf = {}; + ASSERT_THAT(stat("/dev/net/tun", &statbuf), SyscallSucceeds()); + // Check that it's a character device with rw-rw-rw- permissions. + EXPECT_EQ(statbuf.st_mode, S_IFCHR | 0666); +} + } // namespace } // namespace testing diff --git a/test/syscalls/linux/socket_netlink_route_util.cc b/test/syscalls/linux/socket_netlink_route_util.cc new file mode 100644 index 000000000..53eb3b6b2 --- /dev/null +++ b/test/syscalls/linux/socket_netlink_route_util.cc @@ -0,0 +1,163 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "test/syscalls/linux/socket_netlink_route_util.h" + +#include +#include +#include + +#include "absl/types/optional.h" +#include "test/syscalls/linux/socket_netlink_util.h" + +namespace gvisor { +namespace testing { +namespace { + +constexpr uint32_t kSeq = 12345; + +} // namespace + +PosixError DumpLinks( + const FileDescriptor& fd, uint32_t seq, + const std::function& fn) { + struct request { + struct nlmsghdr hdr; + struct ifinfomsg ifm; + }; + + struct request req = {}; + req.hdr.nlmsg_len = sizeof(req); + req.hdr.nlmsg_type = RTM_GETLINK; + req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP; + req.hdr.nlmsg_seq = seq; + req.ifm.ifi_family = AF_UNSPEC; + + return NetlinkRequestResponse(fd, &req, sizeof(req), fn, false); +} + +PosixErrorOr> DumpLinks() { + ASSIGN_OR_RETURN_ERRNO(FileDescriptor fd, NetlinkBoundSocket(NETLINK_ROUTE)); + + std::vector links; + RETURN_IF_ERRNO(DumpLinks(fd, kSeq, [&](const struct nlmsghdr* hdr) { + if (hdr->nlmsg_type != RTM_NEWLINK || + hdr->nlmsg_len < NLMSG_SPACE(sizeof(struct ifinfomsg))) { + return; + } + const struct ifinfomsg* msg = + reinterpret_cast(NLMSG_DATA(hdr)); + const auto* rta = FindRtAttr(hdr, msg, IFLA_IFNAME); + if (rta == nullptr) { + // Ignore links that do not have a name. + return; + } + + links.emplace_back(); + links.back().index = msg->ifi_index; + links.back().type = msg->ifi_type; + links.back().name = + std::string(reinterpret_cast(RTA_DATA(rta))); + })); + return links; +} + +PosixErrorOr> FindLoopbackLink() { + ASSIGN_OR_RETURN_ERRNO(auto links, DumpLinks()); + for (const auto& link : links) { + if (link.type == ARPHRD_LOOPBACK) { + return absl::optional(link); + } + } + return absl::optional(); +} + +PosixError LinkAddLocalAddr(int index, int family, int prefixlen, + const void* addr, int addrlen) { + ASSIGN_OR_RETURN_ERRNO(FileDescriptor fd, NetlinkBoundSocket(NETLINK_ROUTE)); + + struct request { + struct nlmsghdr hdr; + struct ifaddrmsg ifaddr; + char attrbuf[512]; + }; + + struct request req = {}; + req.hdr.nlmsg_len = NLMSG_LENGTH(sizeof(req.ifaddr)); + req.hdr.nlmsg_type = RTM_NEWADDR; + req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK; + req.hdr.nlmsg_seq = kSeq; + req.ifaddr.ifa_index = index; + req.ifaddr.ifa_family = family; + req.ifaddr.ifa_prefixlen = prefixlen; + + struct rtattr* rta = reinterpret_cast( + reinterpret_cast(&req) + NLMSG_ALIGN(req.hdr.nlmsg_len)); + rta->rta_type = IFA_LOCAL; + rta->rta_len = RTA_LENGTH(addrlen); + req.hdr.nlmsg_len = NLMSG_ALIGN(req.hdr.nlmsg_len) + RTA_LENGTH(addrlen); + memcpy(RTA_DATA(rta), addr, addrlen); + + return NetlinkRequestAckOrError(fd, kSeq, &req, req.hdr.nlmsg_len); +} + +PosixError LinkChangeFlags(int index, unsigned int flags, unsigned int change) { + ASSIGN_OR_RETURN_ERRNO(FileDescriptor fd, NetlinkBoundSocket(NETLINK_ROUTE)); + + struct request { + struct nlmsghdr hdr; + struct ifinfomsg ifinfo; + char pad[NLMSG_ALIGNTO]; + }; + + struct request req = {}; + req.hdr.nlmsg_len = NLMSG_LENGTH(sizeof(req.ifinfo)); + req.hdr.nlmsg_type = RTM_NEWLINK; + req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK; + req.hdr.nlmsg_seq = kSeq; + req.ifinfo.ifi_index = index; + req.ifinfo.ifi_flags = flags; + req.ifinfo.ifi_change = change; + + return NetlinkRequestAckOrError(fd, kSeq, &req, req.hdr.nlmsg_len); +} + +PosixError LinkSetMacAddr(int index, const void* addr, int addrlen) { + ASSIGN_OR_RETURN_ERRNO(FileDescriptor fd, NetlinkBoundSocket(NETLINK_ROUTE)); + + struct request { + struct nlmsghdr hdr; + struct ifinfomsg ifinfo; + char attrbuf[512]; + }; + + struct request req = {}; + req.hdr.nlmsg_len = NLMSG_LENGTH(sizeof(req.ifinfo)); + req.hdr.nlmsg_type = RTM_NEWLINK; + req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK; + req.hdr.nlmsg_seq = kSeq; + req.ifinfo.ifi_index = index; + + struct rtattr* rta = reinterpret_cast( + reinterpret_cast(&req) + NLMSG_ALIGN(req.hdr.nlmsg_len)); + rta->rta_type = IFLA_ADDRESS; + rta->rta_len = RTA_LENGTH(addrlen); + req.hdr.nlmsg_len = NLMSG_ALIGN(req.hdr.nlmsg_len) + RTA_LENGTH(addrlen); + memcpy(RTA_DATA(rta), addr, addrlen); + + return NetlinkRequestAckOrError(fd, kSeq, &req, req.hdr.nlmsg_len); +} + +} // namespace testing +} // namespace gvisor diff --git a/test/syscalls/linux/socket_netlink_route_util.h b/test/syscalls/linux/socket_netlink_route_util.h new file mode 100644 index 000000000..2c018e487 --- /dev/null +++ b/test/syscalls/linux/socket_netlink_route_util.h @@ -0,0 +1,55 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef GVISOR_TEST_SYSCALLS_LINUX_SOCKET_NETLINK_ROUTE_UTIL_H_ +#define GVISOR_TEST_SYSCALLS_LINUX_SOCKET_NETLINK_ROUTE_UTIL_H_ + +#include +#include + +#include + +#include "absl/types/optional.h" +#include "test/syscalls/linux/socket_netlink_util.h" + +namespace gvisor { +namespace testing { + +struct Link { + int index; + int16_t type; + std::string name; +}; + +PosixError DumpLinks(const FileDescriptor& fd, uint32_t seq, + const std::function& fn); + +PosixErrorOr> DumpLinks(); + +PosixErrorOr> FindLoopbackLink(); + +// LinkAddLocalAddr sets IFA_LOCAL attribute on the interface. +PosixError LinkAddLocalAddr(int index, int family, int prefixlen, + const void* addr, int addrlen); + +// LinkChangeFlags changes interface flags. E.g. IFF_UP. +PosixError LinkChangeFlags(int index, unsigned int flags, unsigned int change); + +// LinkSetMacAddr sets IFLA_ADDRESS attribute of the interface. +PosixError LinkSetMacAddr(int index, const void* addr, int addrlen); + +} // namespace testing +} // namespace gvisor + +#endif // GVISOR_TEST_SYSCALLS_LINUX_SOCKET_NETLINK_ROUTE_UTIL_H_ diff --git a/test/syscalls/linux/tuntap.cc b/test/syscalls/linux/tuntap.cc new file mode 100644 index 000000000..f6ac9d7b8 --- /dev/null +++ b/test/syscalls/linux/tuntap.cc @@ -0,0 +1,346 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "gmock/gmock.h" +#include "gtest/gtest.h" +#include "absl/strings/ascii.h" +#include "absl/strings/str_split.h" +#include "test/syscalls/linux/socket_netlink_route_util.h" +#include "test/syscalls/linux/socket_test_util.h" +#include "test/util/capability_util.h" +#include "test/util/file_descriptor.h" +#include "test/util/fs_util.h" +#include "test/util/posix_error.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { +namespace { + +constexpr int kIPLen = 4; + +constexpr const char kDevNetTun[] = "/dev/net/tun"; +constexpr const char kTapName[] = "tap0"; + +constexpr const uint8_t kMacA[ETH_ALEN] = {0xAA, 0xAA, 0xAA, 0xAA, 0xAA, 0xAA}; +constexpr const uint8_t kMacB[ETH_ALEN] = {0xBB, 0xBB, 0xBB, 0xBB, 0xBB, 0xBB}; + +PosixErrorOr> DumpLinkNames() { + ASSIGN_OR_RETURN_ERRNO(auto links, DumpLinks()); + std::set names; + for (const auto& link : links) { + names.emplace(link.name); + } + return names; +} + +PosixErrorOr> GetLinkByName(const std::string& name) { + ASSIGN_OR_RETURN_ERRNO(auto links, DumpLinks()); + for (const auto& link : links) { + if (link.name == name) { + return absl::optional(link); + } + } + return absl::optional(); +} + +struct pihdr { + uint16_t pi_flags; + uint16_t pi_protocol; +} __attribute__((packed)); + +struct ping_pkt { + pihdr pi; + struct ethhdr eth; + struct iphdr ip; + struct icmphdr icmp; + char payload[64]; +} __attribute__((packed)); + +ping_pkt CreatePingPacket(const uint8_t srcmac[ETH_ALEN], const char* srcip, + const uint8_t dstmac[ETH_ALEN], const char* dstip) { + ping_pkt pkt = {}; + + pkt.pi.pi_protocol = htons(ETH_P_IP); + + memcpy(pkt.eth.h_dest, dstmac, sizeof(pkt.eth.h_dest)); + memcpy(pkt.eth.h_source, srcmac, sizeof(pkt.eth.h_source)); + pkt.eth.h_proto = htons(ETH_P_IP); + + pkt.ip.ihl = 5; + pkt.ip.version = 4; + pkt.ip.tos = 0; + pkt.ip.tot_len = htons(sizeof(struct iphdr) + sizeof(struct icmphdr) + + sizeof(pkt.payload)); + pkt.ip.id = 1; + pkt.ip.frag_off = 1 << 6; // Do not fragment + pkt.ip.ttl = 64; + pkt.ip.protocol = IPPROTO_ICMP; + inet_pton(AF_INET, dstip, &pkt.ip.daddr); + inet_pton(AF_INET, srcip, &pkt.ip.saddr); + pkt.ip.check = IPChecksum(pkt.ip); + + pkt.icmp.type = ICMP_ECHO; + pkt.icmp.code = 0; + pkt.icmp.checksum = 0; + pkt.icmp.un.echo.sequence = 1; + pkt.icmp.un.echo.id = 1; + + strncpy(pkt.payload, "abcd", sizeof(pkt.payload)); + pkt.icmp.checksum = ICMPChecksum(pkt.icmp, pkt.payload, sizeof(pkt.payload)); + + return pkt; +} + +struct arp_pkt { + pihdr pi; + struct ethhdr eth; + struct arphdr arp; + uint8_t arp_sha[ETH_ALEN]; + uint8_t arp_spa[kIPLen]; + uint8_t arp_tha[ETH_ALEN]; + uint8_t arp_tpa[kIPLen]; +} __attribute__((packed)); + +std::string CreateArpPacket(const uint8_t srcmac[ETH_ALEN], const char* srcip, + const uint8_t dstmac[ETH_ALEN], const char* dstip) { + std::string buffer; + buffer.resize(sizeof(arp_pkt)); + + arp_pkt* pkt = reinterpret_cast(&buffer[0]); + { + pkt->pi.pi_protocol = htons(ETH_P_ARP); + + memcpy(pkt->eth.h_dest, kMacA, sizeof(pkt->eth.h_dest)); + memcpy(pkt->eth.h_source, kMacB, sizeof(pkt->eth.h_source)); + pkt->eth.h_proto = htons(ETH_P_ARP); + + pkt->arp.ar_hrd = htons(ARPHRD_ETHER); + pkt->arp.ar_pro = htons(ETH_P_IP); + pkt->arp.ar_hln = ETH_ALEN; + pkt->arp.ar_pln = kIPLen; + pkt->arp.ar_op = htons(ARPOP_REPLY); + + memcpy(pkt->arp_sha, srcmac, sizeof(pkt->arp_sha)); + inet_pton(AF_INET, srcip, pkt->arp_spa); + memcpy(pkt->arp_tha, dstmac, sizeof(pkt->arp_tha)); + inet_pton(AF_INET, dstip, pkt->arp_tpa); + } + return buffer; +} + +} // namespace + +class TuntapTest : public ::testing::Test { + protected: + void TearDown() override { + if (ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN))) { + // Bring back capability if we had dropped it in test case. + ASSERT_NO_ERRNO(SetCapability(CAP_NET_ADMIN, true)); + } + } +}; + +TEST_F(TuntapTest, CreateInterfaceNoCap) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN))); + + ASSERT_NO_ERRNO(SetCapability(CAP_NET_ADMIN, false)); + + FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(kDevNetTun, O_RDWR)); + + struct ifreq ifr = {}; + ifr.ifr_flags = IFF_TAP; + strncpy(ifr.ifr_name, kTapName, IFNAMSIZ); + + EXPECT_THAT(ioctl(fd.get(), TUNSETIFF, &ifr), SyscallFailsWithErrno(EPERM)); +} + +TEST_F(TuntapTest, CreateFixedNameInterface) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN))); + + FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(kDevNetTun, O_RDWR)); + + struct ifreq ifr_set = {}; + ifr_set.ifr_flags = IFF_TAP; + strncpy(ifr_set.ifr_name, kTapName, IFNAMSIZ); + EXPECT_THAT(ioctl(fd.get(), TUNSETIFF, &ifr_set), + SyscallSucceedsWithValue(0)); + + struct ifreq ifr_get = {}; + EXPECT_THAT(ioctl(fd.get(), TUNGETIFF, &ifr_get), + SyscallSucceedsWithValue(0)); + + struct ifreq ifr_expect = ifr_set; + // See __tun_chr_ioctl() in net/drivers/tun.c. + ifr_expect.ifr_flags |= IFF_NOFILTER; + + EXPECT_THAT(DumpLinkNames(), + IsPosixErrorOkAndHolds(::testing::Contains(kTapName))); + EXPECT_THAT(memcmp(&ifr_expect, &ifr_get, sizeof(ifr_get)), ::testing::Eq(0)); +} + +TEST_F(TuntapTest, CreateInterface) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN))); + + FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(kDevNetTun, O_RDWR)); + + struct ifreq ifr = {}; + ifr.ifr_flags = IFF_TAP; + // Empty ifr.ifr_name. Let kernel assign. + + EXPECT_THAT(ioctl(fd.get(), TUNSETIFF, &ifr), SyscallSucceedsWithValue(0)); + + struct ifreq ifr_get = {}; + EXPECT_THAT(ioctl(fd.get(), TUNGETIFF, &ifr_get), + SyscallSucceedsWithValue(0)); + + std::string ifname = ifr_get.ifr_name; + EXPECT_THAT(ifname, ::testing::StartsWith("tap")); + EXPECT_THAT(DumpLinkNames(), + IsPosixErrorOkAndHolds(::testing::Contains(ifname))); +} + +TEST_F(TuntapTest, InvalidReadWrite) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN))); + + FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(kDevNetTun, O_RDWR)); + + char buf[128] = {}; + EXPECT_THAT(read(fd.get(), buf, sizeof(buf)), SyscallFailsWithErrno(EBADFD)); + EXPECT_THAT(write(fd.get(), buf, sizeof(buf)), SyscallFailsWithErrno(EBADFD)); +} + +TEST_F(TuntapTest, WriteToDownDevice) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN))); + + // FIXME: gVisor always creates enabled/up'd interfaces. + SKIP_IF(IsRunningOnGvisor()); + + FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(kDevNetTun, O_RDWR)); + + // Device created should be down by default. + struct ifreq ifr = {}; + ifr.ifr_flags = IFF_TAP; + EXPECT_THAT(ioctl(fd.get(), TUNSETIFF, &ifr), SyscallSucceedsWithValue(0)); + + char buf[128] = {}; + EXPECT_THAT(write(fd.get(), buf, sizeof(buf)), SyscallFailsWithErrno(EIO)); +} + +// This test sets up a TAP device and pings kernel by sending ICMP echo request. +// +// It works as the following: +// * Open /dev/net/tun, and create kTapName interface. +// * Use rtnetlink to do initial setup of the interface: +// * Assign IP address 10.0.0.1/24 to kernel. +// * MAC address: kMacA +// * Bring up the interface. +// * Send an ICMP echo reqest (ping) packet from 10.0.0.2 (kMacB) to kernel. +// * Loop to receive packets from TAP device/fd: +// * If packet is an ICMP echo reply, it stops and passes the test. +// * If packet is an ARP request, it responds with canned reply and resends +// the +// ICMP request packet. +TEST_F(TuntapTest, PingKernel) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN))); + + // Interface creation. + FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(kDevNetTun, O_RDWR)); + + struct ifreq ifr_set = {}; + ifr_set.ifr_flags = IFF_TAP; + strncpy(ifr_set.ifr_name, kTapName, IFNAMSIZ); + EXPECT_THAT(ioctl(fd.get(), TUNSETIFF, &ifr_set), + SyscallSucceedsWithValue(0)); + + absl::optional link = + ASSERT_NO_ERRNO_AND_VALUE(GetLinkByName(kTapName)); + ASSERT_TRUE(link.has_value()); + + // Interface setup. + struct in_addr addr; + inet_pton(AF_INET, "10.0.0.1", &addr); + EXPECT_NO_ERRNO(LinkAddLocalAddr(link->index, AF_INET, /*prefixlen=*/24, + &addr, sizeof(addr))); + + if (!IsRunningOnGvisor()) { + // FIXME: gVisor doesn't support setting MAC address on interfaces yet. + EXPECT_NO_ERRNO(LinkSetMacAddr(link->index, kMacA, sizeof(kMacA))); + + // FIXME: gVisor always creates enabled/up'd interfaces. + EXPECT_NO_ERRNO(LinkChangeFlags(link->index, IFF_UP, IFF_UP)); + } + + ping_pkt ping_req = CreatePingPacket(kMacB, "10.0.0.2", kMacA, "10.0.0.1"); + std::string arp_rep = CreateArpPacket(kMacB, "10.0.0.2", kMacA, "10.0.0.1"); + + // Send ping, this would trigger an ARP request on Linux. + EXPECT_THAT(write(fd.get(), &ping_req, sizeof(ping_req)), + SyscallSucceedsWithValue(sizeof(ping_req))); + + // Receive loop to process inbound packets. + struct inpkt { + union { + pihdr pi; + ping_pkt ping; + arp_pkt arp; + }; + }; + while (1) { + inpkt r = {}; + int n = read(fd.get(), &r, sizeof(r)); + EXPECT_THAT(n, SyscallSucceeds()); + + if (n < sizeof(pihdr)) { + std::cerr << "Ignored packet, protocol: " << r.pi.pi_protocol + << " len: " << n << std::endl; + continue; + } + + // Process ARP packet. + if (n >= sizeof(arp_pkt) && r.pi.pi_protocol == htons(ETH_P_ARP)) { + // Respond with canned ARP reply. + EXPECT_THAT(write(fd.get(), arp_rep.data(), arp_rep.size()), + SyscallSucceedsWithValue(arp_rep.size())); + // First ping request might have been dropped due to mac address not in + // ARP cache. Send it again. + EXPECT_THAT(write(fd.get(), &ping_req, sizeof(ping_req)), + SyscallSucceedsWithValue(sizeof(ping_req))); + } + + // Process ping response packet. + if (n >= sizeof(ping_pkt) && r.pi.pi_protocol == ping_req.pi.pi_protocol && + r.ping.ip.protocol == ping_req.ip.protocol && + !memcmp(&r.ping.ip.saddr, &ping_req.ip.daddr, kIPLen) && + !memcmp(&r.ping.ip.daddr, &ping_req.ip.saddr, kIPLen) && + r.ping.icmp.type == 0 && r.ping.icmp.code == 0) { + // Ends and passes the test. + break; + } + } +} + +} // namespace testing +} // namespace gvisor -- cgit v1.2.3 From 93e0c3752981b6a1c5b745faec6506c17480b84b Mon Sep 17 00:00:00 2001 From: Haibo Xu Date: Fri, 21 Feb 2020 10:28:16 +0000 Subject: Enable bluepill dieTrampoline operation on arm64. Signed-off-by: Haibo Xu Change-Id: I9e1bf2513c23bdd8c387e5b3c874c6ad3ca9aab0 --- pkg/sentry/platform/kvm/bluepill_arm64.s | 8 ++++--- pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go | 23 +++++++++++++++++++- pkg/sentry/platform/ring0/aarch64.go | 27 ++++++++++++------------ 3 files changed, 41 insertions(+), 17 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/kvm/bluepill_arm64.s b/pkg/sentry/platform/kvm/bluepill_arm64.s index c61700892..04efa0147 100644 --- a/pkg/sentry/platform/kvm/bluepill_arm64.s +++ b/pkg/sentry/platform/kvm/bluepill_arm64.s @@ -82,6 +82,8 @@ fallback: // dieTrampoline: see bluepill.go, bluepill_arm64_unsafe.go for documentation. TEXT ·dieTrampoline(SB),NOSPLIT,$0 - // TODO(gvisor.dev/issue/1249): dieTrampoline supporting for Arm64. - MOVD R9, 8(RSP) - BL ·dieHandler(SB) + // R0: Fake the old PC as caller + // R1: First argument (vCPU) + MOVD.P R1, 8(RSP) // R1: First argument (vCPU) + MOVD.P R0, 8(RSP) // R0: Fake the old PC as caller + B ·dieHandler(SB) diff --git a/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go b/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go index 2f02c03cf..195331383 100644 --- a/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go @@ -18,9 +18,30 @@ package kvm import ( "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/platform/ring0" ) +// dieArchSetup initialies the state for dieTrampoline. +// +// The arm64 dieTrampoline requires the vCPU to be set in R1, and the last PC +// to be in R0. The trampoline then simulates a call to dieHandler from the +// provided PC. +// //go:nosplit func dieArchSetup(c *vCPU, context *arch.SignalContext64, guestRegs *userRegs) { - // TODO(gvisor.dev/issue/1249): dieTrampoline supporting for Arm64. + // If the vCPU is in user mode, we set the stack to the stored stack + // value in the vCPU itself. We don't want to unwind the user stack. + if guestRegs.Regs.Pstate&ring0.PSR_MODE_MASK == ring0.PSR_MODE_EL0t { + regs := c.CPU.Registers() + context.Regs[0] = regs.Regs[0] + context.Sp = regs.Sp + context.Regs[29] = regs.Regs[29] // stack base address + } else { + context.Regs[0] = guestRegs.Regs.Pc + context.Sp = guestRegs.Regs.Sp + context.Regs[29] = guestRegs.Regs.Regs[29] + context.Pstate = guestRegs.Regs.Pstate + } + context.Regs[1] = uint64(uintptr(unsafe.Pointer(c))) + context.Pc = uint64(dieTrampolineAddr) } diff --git a/pkg/sentry/platform/ring0/aarch64.go b/pkg/sentry/platform/ring0/aarch64.go index f6da41c27..8122ac6e2 100644 --- a/pkg/sentry/platform/ring0/aarch64.go +++ b/pkg/sentry/platform/ring0/aarch64.go @@ -27,26 +27,27 @@ const ( _PTE_PGT_BASE = 0x7000 _PTE_PGT_SIZE = 0x1000 - _PSR_MODE_EL0t = 0x0 - _PSR_MODE_EL1t = 0x4 - _PSR_MODE_EL1h = 0x5 - _PSR_EL_MASK = 0xf - - _PSR_D_BIT = 0x200 - _PSR_A_BIT = 0x100 - _PSR_I_BIT = 0x80 - _PSR_F_BIT = 0x40 + _PSR_D_BIT = 0x00000200 + _PSR_A_BIT = 0x00000100 + _PSR_I_BIT = 0x00000080 + _PSR_F_BIT = 0x00000040 ) const ( + // PSR bits + PSR_MODE_EL0t = 0x00000000 + PSR_MODE_EL1t = 0x00000004 + PSR_MODE_EL1h = 0x00000005 + PSR_MODE_MASK = 0x0000000f + // KernelFlagsSet should always be set in the kernel. - KernelFlagsSet = _PSR_MODE_EL1h + KernelFlagsSet = PSR_MODE_EL1h // UserFlagsSet are always set in userspace. - UserFlagsSet = _PSR_MODE_EL0t + UserFlagsSet = PSR_MODE_EL0t - KernelFlagsClear = _PSR_EL_MASK - UserFlagsClear = _PSR_EL_MASK + KernelFlagsClear = PSR_MODE_MASK + UserFlagsClear = PSR_MODE_MASK PsrDefaultSet = _PSR_D_BIT | _PSR_A_BIT | _PSR_I_BIT | _PSR_F_BIT ) -- cgit v1.2.3 From 53504e29ca27b8dc9e098fbb88983fdbce90cca3 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Tue, 25 Feb 2020 12:16:43 -0800 Subject: Fix mount refcount issue. Each mount is holds a reference on a root Dirent, but the mount itself may live beyond it's own reference. This means that a call to Root() can come after the associated reference has been dropped. Instead of introducing a separate layer of references for mount objects, we simply change the Root() method to use TryIncRef() and allow it to return nil if the mount is already gone. This requires updating a small number of callers and minimizes the change (since VFSv2 will replace this code shortly). PiperOrigin-RevId: 297174230 --- pkg/sentry/fs/mount_test.go | 11 ++++++----- pkg/sentry/fs/mounts.go | 10 +++++++--- pkg/sentry/fs/proc/mounts.go | 16 ++++++++++++---- 3 files changed, 25 insertions(+), 12 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/mount_test.go b/pkg/sentry/fs/mount_test.go index e672a438c..a3d10770b 100644 --- a/pkg/sentry/fs/mount_test.go +++ b/pkg/sentry/fs/mount_test.go @@ -36,11 +36,12 @@ func mountPathsAre(root *Dirent, got []*Mount, want ...string) error { gotPaths := make(map[string]struct{}, len(got)) gotStr := make([]string, len(got)) for i, g := range got { - groot := g.Root() - name, _ := groot.FullName(root) - groot.DecRef() - gotStr[i] = name - gotPaths[name] = struct{}{} + if groot := g.Root(); groot != nil { + name, _ := groot.FullName(root) + groot.DecRef() + gotStr[i] = name + gotPaths[name] = struct{}{} + } } if len(got) != len(want) { return fmt.Errorf("mount paths are different, got: %q, want: %q", gotStr, want) diff --git a/pkg/sentry/fs/mounts.go b/pkg/sentry/fs/mounts.go index 574a2cc91..c7981f66e 100644 --- a/pkg/sentry/fs/mounts.go +++ b/pkg/sentry/fs/mounts.go @@ -100,10 +100,14 @@ func newUndoMount(d *Dirent) *Mount { } } -// Root returns the root dirent of this mount. Callers must call DecRef on the -// returned dirent. +// Root returns the root dirent of this mount. +// +// This may return nil if the mount has already been free. Callers must handle this +// case appropriately. If non-nil, callers must call DecRef on the returned *Dirent. func (m *Mount) Root() *Dirent { - m.root.IncRef() + if !m.root.TryIncRef() { + return nil + } return m.root } diff --git a/pkg/sentry/fs/proc/mounts.go b/pkg/sentry/fs/proc/mounts.go index c10888100..94deb553b 100644 --- a/pkg/sentry/fs/proc/mounts.go +++ b/pkg/sentry/fs/proc/mounts.go @@ -60,13 +60,15 @@ func forEachMount(t *kernel.Task, fn func(string, *fs.Mount)) { }) for _, m := range ms { mroot := m.Root() + if mroot == nil { + continue // No longer valid. + } mountPath, desc := mroot.FullName(rootDir) mroot.DecRef() if !desc { // MountSources that are not descendants of the chroot jail are ignored. continue } - fn(mountPath, m) } } @@ -91,6 +93,12 @@ func (mif *mountInfoFile) ReadSeqFileData(ctx context.Context, handle seqfile.Se var buf bytes.Buffer forEachMount(mif.t, func(mountPath string, m *fs.Mount) { + mroot := m.Root() + if mroot == nil { + return // No longer valid. + } + defer mroot.DecRef() + // Format: // 36 35 98:0 /mnt1 /mnt2 rw,noatime master:1 - ext3 /dev/root rw,errors=continue // (1)(2)(3) (4) (5) (6) (7) (8) (9) (10) (11) @@ -107,9 +115,6 @@ func (mif *mountInfoFile) ReadSeqFileData(ctx context.Context, handle seqfile.Se // (3) Major:Minor device ID. We don't have a superblock, so we // just use the root inode device number. - mroot := m.Root() - defer mroot.DecRef() - sa := mroot.Inode.StableAttr fmt.Fprintf(&buf, "%d:%d ", sa.DeviceFileMajor, sa.DeviceFileMinor) @@ -207,6 +212,9 @@ func (mf *mountsFile) ReadSeqFileData(ctx context.Context, handle seqfile.SeqHan // // The "needs dump"and fsck flags are always 0, which is allowed. root := m.Root() + if root == nil { + return // No longer valid. + } defer root.DecRef() flags := root.Inode.MountSource.Flags -- cgit v1.2.3 From 98b693e61b37a62f7b29ce1cab8b4c4c54fa044e Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Tue, 25 Feb 2020 12:21:27 -0800 Subject: Don't acquire contended lock with the OS thread locked. Fixes #1049 PiperOrigin-RevId: 297175164 --- pkg/sentry/platform/kvm/machine.go | 18 +++++++++++++++++- 1 file changed, 17 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/kvm/machine.go b/pkg/sentry/platform/kvm/machine.go index 8076c7529..f1afc74dc 100644 --- a/pkg/sentry/platform/kvm/machine.go +++ b/pkg/sentry/platform/kvm/machine.go @@ -329,10 +329,12 @@ func (m *machine) Destroy() { } // Get gets an available vCPU. +// +// This will return with the OS thread locked. func (m *machine) Get() *vCPU { + m.mu.RLock() runtime.LockOSThread() tid := procid.Current() - m.mu.RLock() // Check for an exact match. if c := m.vCPUs[tid]; c != nil { @@ -343,8 +345,22 @@ func (m *machine) Get() *vCPU { // The happy path failed. We now proceed to acquire an exclusive lock // (because the vCPU map may change), and scan all available vCPUs. + // In this case, we first unlock the OS thread. Otherwise, if mu is + // not available, the current system thread will be parked and a new + // system thread spawned. We avoid this situation by simply refreshing + // tid after relocking the system thread. m.mu.RUnlock() + runtime.UnlockOSThread() m.mu.Lock() + runtime.LockOSThread() + tid = procid.Current() + + // Recheck for an exact match. + if c := m.vCPUs[tid]; c != nil { + c.lock() + m.mu.Unlock() + return c + } for { // Scan for an available vCPU. -- cgit v1.2.3 From 6def8ea6ac601daa9256a31f818db9f7eb532168 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Tue, 25 Feb 2020 12:22:09 -0800 Subject: Fix nested logging. PiperOrigin-RevId: 297175316 --- pkg/log/glog.go | 6 ++--- pkg/log/json.go | 2 +- pkg/log/json_k8s.go | 2 +- pkg/log/log.go | 60 ++++++++++++++++++++++++++++++++----------- pkg/sentry/kernel/task_log.go | 6 ++--- 5 files changed, 52 insertions(+), 24 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/log/glog.go b/pkg/log/glog.go index cab5fae55..b4f7bb5a4 100644 --- a/pkg/log/glog.go +++ b/pkg/log/glog.go @@ -46,7 +46,7 @@ var pid = os.Getpid() // line The line number // msg The user-supplied message // -func (g *GoogleEmitter) Emit(level Level, timestamp time.Time, format string, args ...interface{}) { +func (g *GoogleEmitter) Emit(depth int, level Level, timestamp time.Time, format string, args ...interface{}) { // Log level. prefix := byte('?') switch level { @@ -64,9 +64,7 @@ func (g *GoogleEmitter) Emit(level Level, timestamp time.Time, format string, ar microsecond := int(timestamp.Nanosecond() / 1000) // 0 = this frame. - // 1 = Debugf, etc. - // 2 = Caller. - _, file, line, ok := runtime.Caller(2) + _, file, line, ok := runtime.Caller(depth + 1) if ok { // Trim any directory path from the file. slash := strings.LastIndexByte(file, byte('/')) diff --git a/pkg/log/json.go b/pkg/log/json.go index a278c8fc8..0943db1cc 100644 --- a/pkg/log/json.go +++ b/pkg/log/json.go @@ -62,7 +62,7 @@ type JSONEmitter struct { } // Emit implements Emitter.Emit. -func (e JSONEmitter) Emit(level Level, timestamp time.Time, format string, v ...interface{}) { +func (e JSONEmitter) Emit(_ int, level Level, timestamp time.Time, format string, v ...interface{}) { j := jsonLog{ Msg: fmt.Sprintf(format, v...), Level: level, diff --git a/pkg/log/json_k8s.go b/pkg/log/json_k8s.go index cee6eb514..6c6fc8b6f 100644 --- a/pkg/log/json_k8s.go +++ b/pkg/log/json_k8s.go @@ -33,7 +33,7 @@ type K8sJSONEmitter struct { } // Emit implements Emitter.Emit. -func (e *K8sJSONEmitter) Emit(level Level, timestamp time.Time, format string, v ...interface{}) { +func (e *K8sJSONEmitter) Emit(_ int, level Level, timestamp time.Time, format string, v ...interface{}) { j := k8sJSONLog{ Log: fmt.Sprintf(format, v...), Level: level, diff --git a/pkg/log/log.go b/pkg/log/log.go index 5056f17e6..a794da1aa 100644 --- a/pkg/log/log.go +++ b/pkg/log/log.go @@ -79,7 +79,7 @@ func (l Level) String() string { type Emitter interface { // Emit emits the given log statement. This allows for control over the // timestamp used for logging. - Emit(level Level, timestamp time.Time, format string, v ...interface{}) + Emit(depth int, level Level, timestamp time.Time, format string, v ...interface{}) } // Writer writes the output to the given writer. @@ -142,7 +142,7 @@ func (l *Writer) Write(data []byte) (int, error) { } // Emit emits the message. -func (l *Writer) Emit(level Level, timestamp time.Time, format string, args ...interface{}) { +func (l *Writer) Emit(_ int, _ Level, _ time.Time, format string, args ...interface{}) { fmt.Fprintf(l, format, args...) } @@ -150,9 +150,9 @@ func (l *Writer) Emit(level Level, timestamp time.Time, format string, args ...i type MultiEmitter []Emitter // Emit emits to all emitters. -func (m *MultiEmitter) Emit(level Level, timestamp time.Time, format string, v ...interface{}) { +func (m *MultiEmitter) Emit(depth int, level Level, timestamp time.Time, format string, v ...interface{}) { for _, e := range *m { - e.Emit(level, timestamp, format, v...) + e.Emit(1+depth, level, timestamp, format, v...) } } @@ -167,7 +167,7 @@ type TestEmitter struct { } // Emit emits to the TestLogger. -func (t *TestEmitter) Emit(level Level, timestamp time.Time, format string, v ...interface{}) { +func (t *TestEmitter) Emit(_ int, level Level, timestamp time.Time, format string, v ...interface{}) { t.Logf(format, v...) } @@ -198,22 +198,37 @@ type BasicLogger struct { // Debugf implements logger.Debugf. func (l *BasicLogger) Debugf(format string, v ...interface{}) { - if l.IsLogging(Debug) { - l.Emit(Debug, time.Now(), format, v...) - } + l.DebugfAtDepth(1, format, v...) } // Infof implements logger.Infof. func (l *BasicLogger) Infof(format string, v ...interface{}) { - if l.IsLogging(Info) { - l.Emit(Info, time.Now(), format, v...) - } + l.InfofAtDepth(1, format, v...) } // Warningf implements logger.Warningf. func (l *BasicLogger) Warningf(format string, v ...interface{}) { + l.WarningfAtDepth(1, format, v...) +} + +// DebugfAtDepth logs at a specific depth. +func (l *BasicLogger) DebugfAtDepth(depth int, format string, v ...interface{}) { + if l.IsLogging(Debug) { + l.Emit(1+depth, Debug, time.Now(), format, v...) + } +} + +// InfofAtDepth logs at a specific depth. +func (l *BasicLogger) InfofAtDepth(depth int, format string, v ...interface{}) { + if l.IsLogging(Info) { + l.Emit(1+depth, Info, time.Now(), format, v...) + } +} + +// WarningfAtDepth logs at a specific depth. +func (l *BasicLogger) WarningfAtDepth(depth int, format string, v ...interface{}) { if l.IsLogging(Warning) { - l.Emit(Warning, time.Now(), format, v...) + l.Emit(1+depth, Warning, time.Now(), format, v...) } } @@ -257,17 +272,32 @@ func SetLevel(newLevel Level) { // Debugf logs to the global logger. func Debugf(format string, v ...interface{}) { - Log().Debugf(format, v...) + Log().DebugfAtDepth(1, format, v...) } // Infof logs to the global logger. func Infof(format string, v ...interface{}) { - Log().Infof(format, v...) + Log().InfofAtDepth(1, format, v...) } // Warningf logs to the global logger. func Warningf(format string, v ...interface{}) { - Log().Warningf(format, v...) + Log().WarningfAtDepth(1, format, v...) +} + +// DebugfAtDepth logs to the global logger. +func DebugfAtDepth(depth int, format string, v ...interface{}) { + Log().DebugfAtDepth(1+depth, format, v...) +} + +// InfofAtDepth logs to the global logger. +func InfofAtDepth(depth int, format string, v ...interface{}) { + Log().InfofAtDepth(1+depth, format, v...) +} + +// WarningfAtDepth logs to the global logger. +func WarningfAtDepth(depth int, format string, v ...interface{}) { + Log().WarningfAtDepth(1+depth, format, v...) } // defaultStackSize is the default buffer size to allocate for stack traces. diff --git a/pkg/sentry/kernel/task_log.go b/pkg/sentry/kernel/task_log.go index 6d737d3e5..eeccaa197 100644 --- a/pkg/sentry/kernel/task_log.go +++ b/pkg/sentry/kernel/task_log.go @@ -32,21 +32,21 @@ const ( // Infof logs an formatted info message by calling log.Infof. func (t *Task) Infof(fmt string, v ...interface{}) { if log.IsLogging(log.Info) { - log.Infof(t.logPrefix.Load().(string)+fmt, v...) + log.InfofAtDepth(1, t.logPrefix.Load().(string)+fmt, v...) } } // Warningf logs a warning string by calling log.Warningf. func (t *Task) Warningf(fmt string, v ...interface{}) { if log.IsLogging(log.Warning) { - log.Warningf(t.logPrefix.Load().(string)+fmt, v...) + log.WarningfAtDepth(1, t.logPrefix.Load().(string)+fmt, v...) } } // Debugf creates a debug string that includes the task ID. func (t *Task) Debugf(fmt string, v ...interface{}) { if log.IsLogging(log.Debug) { - log.Debugf(t.logPrefix.Load().(string)+fmt, v...) + log.DebugfAtDepth(1, t.logPrefix.Load().(string)+fmt, v...) } } -- cgit v1.2.3 From 471b15b212831af31c2fe36cd42cea7ec7b7785b Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Tue, 25 Feb 2020 13:25:36 -0800 Subject: Port most syscalls to VFS2. pipe and pipe2 aren't ported, pending a slight rework of pipe FDs for VFS2. mount and umount2 aren't ported out of temporary laziness. access and faccessat need additional FSImpl methods to implement properly, but are stubbed to prevent googletest from CHECK-failing. Other syscalls require additional plumbing. Updates #1623 PiperOrigin-RevId: 297188448 --- pkg/abi/linux/epoll_amd64.go | 2 + pkg/abi/linux/epoll_arm64.go | 2 + pkg/abi/linux/file.go | 2 + pkg/abi/linux/fs.go | 2 + pkg/abi/linux/signal.go | 2 + pkg/abi/linux/time.go | 6 + pkg/abi/linux/xattr.go | 1 + pkg/fspath/BUILD | 4 +- pkg/fspath/builder.go | 8 + pkg/fspath/builder_unsafe.go | 27 - pkg/fspath/fspath.go | 3 +- pkg/gohacks/BUILD | 11 + pkg/gohacks/gohacks_unsafe.go | 57 ++ pkg/sentry/fsbridge/vfs.go | 10 +- pkg/sentry/fsimpl/proc/tasks.go | 4 +- pkg/sentry/kernel/fd_table.go | 49 +- pkg/sentry/kernel/fs_context.go | 22 + pkg/sentry/kernel/task.go | 18 + pkg/sentry/syscalls/linux/sys_epoll.go | 4 + pkg/sentry/syscalls/linux/sys_file.go | 40 ++ pkg/sentry/syscalls/linux/sys_getdents.go | 4 + pkg/sentry/syscalls/linux/sys_lseek.go | 4 + pkg/sentry/syscalls/linux/sys_mmap.go | 4 + pkg/sentry/syscalls/linux/sys_read.go | 4 + pkg/sentry/syscalls/linux/sys_stat.go | 4 + pkg/sentry/syscalls/linux/sys_sync.go | 4 + pkg/sentry/syscalls/linux/sys_write.go | 4 + pkg/sentry/syscalls/linux/sys_xattr.go | 4 + pkg/sentry/syscalls/linux/vfs2/BUILD | 28 +- pkg/sentry/syscalls/linux/vfs2/epoll.go | 225 ++++++++ pkg/sentry/syscalls/linux/vfs2/epoll_unsafe.go | 44 ++ pkg/sentry/syscalls/linux/vfs2/execve.go | 137 +++++ pkg/sentry/syscalls/linux/vfs2/fd.go | 147 ++++++ pkg/sentry/syscalls/linux/vfs2/filesystem.go | 326 ++++++++++++ pkg/sentry/syscalls/linux/vfs2/fscontext.go | 131 +++++ pkg/sentry/syscalls/linux/vfs2/getdents.go | 149 ++++++ pkg/sentry/syscalls/linux/vfs2/ioctl.go | 35 ++ .../syscalls/linux/vfs2/linux64_override_amd64.go | 216 ++++---- .../syscalls/linux/vfs2/linux64_override_arm64.go | 2 + pkg/sentry/syscalls/linux/vfs2/mmap.go | 92 ++++ pkg/sentry/syscalls/linux/vfs2/path.go | 94 ++++ pkg/sentry/syscalls/linux/vfs2/poll.go | 584 +++++++++++++++++++++ pkg/sentry/syscalls/linux/vfs2/read_write.go | 511 ++++++++++++++++++ pkg/sentry/syscalls/linux/vfs2/setstat.go | 380 ++++++++++++++ pkg/sentry/syscalls/linux/vfs2/stat.go | 346 ++++++++++++ pkg/sentry/syscalls/linux/vfs2/sync.go | 87 +++ pkg/sentry/syscalls/linux/vfs2/sys_read.go | 95 ---- pkg/sentry/syscalls/linux/vfs2/xattr.go | 353 +++++++++++++ pkg/sentry/vfs/BUILD | 1 + pkg/sentry/vfs/epoll.go | 3 + pkg/sentry/vfs/mount_unsafe.go | 12 +- pkg/sentry/vfs/resolving_path.go | 2 +- pkg/sentry/vfs/vfs.go | 10 +- pkg/usermem/BUILD | 2 +- pkg/usermem/usermem.go | 9 +- pkg/usermem/usermem_unsafe.go | 27 - runsc/boot/filter/config.go | 2 + 57 files changed, 4082 insertions(+), 274 deletions(-) delete mode 100644 pkg/fspath/builder_unsafe.go create mode 100644 pkg/gohacks/BUILD create mode 100644 pkg/gohacks/gohacks_unsafe.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/epoll.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/epoll_unsafe.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/execve.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/fd.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/filesystem.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/fscontext.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/getdents.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/ioctl.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/mmap.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/path.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/poll.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/read_write.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/setstat.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/stat.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/sync.go delete mode 100644 pkg/sentry/syscalls/linux/vfs2/sys_read.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/xattr.go delete mode 100644 pkg/usermem/usermem_unsafe.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/epoll_amd64.go b/pkg/abi/linux/epoll_amd64.go index 57041491c..34ff18009 100644 --- a/pkg/abi/linux/epoll_amd64.go +++ b/pkg/abi/linux/epoll_amd64.go @@ -15,6 +15,8 @@ package linux // EpollEvent is equivalent to struct epoll_event from epoll(2). +// +// +marshal type EpollEvent struct { Events uint32 // Linux makes struct epoll_event::data a __u64. We represent it as diff --git a/pkg/abi/linux/epoll_arm64.go b/pkg/abi/linux/epoll_arm64.go index 62ef5821e..f86c35329 100644 --- a/pkg/abi/linux/epoll_arm64.go +++ b/pkg/abi/linux/epoll_arm64.go @@ -15,6 +15,8 @@ package linux // EpollEvent is equivalent to struct epoll_event from epoll(2). +// +// +marshal type EpollEvent struct { Events uint32 // Linux makes struct epoll_event a __u64, necessitating 4 bytes of padding diff --git a/pkg/abi/linux/file.go b/pkg/abi/linux/file.go index c3ab15a4f..e229ac21c 100644 --- a/pkg/abi/linux/file.go +++ b/pkg/abi/linux/file.go @@ -241,6 +241,8 @@ const ( ) // Statx represents struct statx. +// +// +marshal type Statx struct { Mask uint32 Blksize uint32 diff --git a/pkg/abi/linux/fs.go b/pkg/abi/linux/fs.go index 2c652baa2..158d2db5b 100644 --- a/pkg/abi/linux/fs.go +++ b/pkg/abi/linux/fs.go @@ -38,6 +38,8 @@ const ( ) // Statfs is struct statfs, from uapi/asm-generic/statfs.h. +// +// +marshal type Statfs struct { // Type is one of the filesystem magic values, defined above. Type uint64 diff --git a/pkg/abi/linux/signal.go b/pkg/abi/linux/signal.go index c69b04ea9..1c330e763 100644 --- a/pkg/abi/linux/signal.go +++ b/pkg/abi/linux/signal.go @@ -115,6 +115,8 @@ const ( ) // SignalSet is a signal mask with a bit corresponding to each signal. +// +// +marshal type SignalSet uint64 // SignalSetSize is the size in bytes of a SignalSet. diff --git a/pkg/abi/linux/time.go b/pkg/abi/linux/time.go index e562b46d9..e6860ed49 100644 --- a/pkg/abi/linux/time.go +++ b/pkg/abi/linux/time.go @@ -157,6 +157,8 @@ func DurationToTimespec(dur time.Duration) Timespec { const SizeOfTimeval = 16 // Timeval represents struct timeval in . +// +// +marshal type Timeval struct { Sec int64 Usec int64 @@ -230,6 +232,8 @@ type Tms struct { type TimerID int32 // StatxTimestamp represents struct statx_timestamp. +// +// +marshal type StatxTimestamp struct { Sec int64 Nsec uint32 @@ -258,6 +262,8 @@ func NsecToStatxTimestamp(nsec int64) (ts StatxTimestamp) { } // Utime represents struct utimbuf used by utimes(2). +// +// +marshal type Utime struct { Actime int64 Modtime int64 diff --git a/pkg/abi/linux/xattr.go b/pkg/abi/linux/xattr.go index a3b6406fa..99180b208 100644 --- a/pkg/abi/linux/xattr.go +++ b/pkg/abi/linux/xattr.go @@ -18,6 +18,7 @@ package linux const ( XATTR_NAME_MAX = 255 XATTR_SIZE_MAX = 65536 + XATTR_LIST_MAX = 65536 XATTR_CREATE = 1 XATTR_REPLACE = 2 diff --git a/pkg/fspath/BUILD b/pkg/fspath/BUILD index ee84471b2..67dd1e225 100644 --- a/pkg/fspath/BUILD +++ b/pkg/fspath/BUILD @@ -8,9 +8,11 @@ go_library( name = "fspath", srcs = [ "builder.go", - "builder_unsafe.go", "fspath.go", ], + deps = [ + "//pkg/gohacks", + ], ) go_test( diff --git a/pkg/fspath/builder.go b/pkg/fspath/builder.go index 7ddb36826..6318d3874 100644 --- a/pkg/fspath/builder.go +++ b/pkg/fspath/builder.go @@ -16,6 +16,8 @@ package fspath import ( "fmt" + + "gvisor.dev/gvisor/pkg/gohacks" ) // Builder is similar to strings.Builder, but is used to produce pathnames @@ -102,3 +104,9 @@ func (b *Builder) AppendString(str string) { copy(b.buf[b.start:], b.buf[oldStart:]) copy(b.buf[len(b.buf)-len(str):], str) } + +// String returns the accumulated string. No other methods should be called +// after String. +func (b *Builder) String() string { + return gohacks.StringFromImmutableBytes(b.buf[b.start:]) +} diff --git a/pkg/fspath/builder_unsafe.go b/pkg/fspath/builder_unsafe.go deleted file mode 100644 index 75606808d..000000000 --- a/pkg/fspath/builder_unsafe.go +++ /dev/null @@ -1,27 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fspath - -import ( - "unsafe" -) - -// String returns the accumulated string. No other methods should be called -// after String. -func (b *Builder) String() string { - bs := b.buf[b.start:] - // Compare strings.Builder.String(). - return *(*string)(unsafe.Pointer(&bs)) -} diff --git a/pkg/fspath/fspath.go b/pkg/fspath/fspath.go index 9fb3fee24..4c983d5fd 100644 --- a/pkg/fspath/fspath.go +++ b/pkg/fspath/fspath.go @@ -67,7 +67,8 @@ func Parse(pathname string) Path { // Path contains the information contained in a pathname string. // -// Path is copyable by value. +// Path is copyable by value. The zero value for Path is equivalent to +// fspath.Parse(""), i.e. the empty path. type Path struct { // Begin is an iterator to the first path component in the relative part of // the path. diff --git a/pkg/gohacks/BUILD b/pkg/gohacks/BUILD new file mode 100644 index 000000000..798a65eca --- /dev/null +++ b/pkg/gohacks/BUILD @@ -0,0 +1,11 @@ +load("//tools:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_library( + name = "gohacks", + srcs = [ + "gohacks_unsafe.go", + ], + visibility = ["//:sandbox"], +) diff --git a/pkg/gohacks/gohacks_unsafe.go b/pkg/gohacks/gohacks_unsafe.go new file mode 100644 index 000000000..aad675172 --- /dev/null +++ b/pkg/gohacks/gohacks_unsafe.go @@ -0,0 +1,57 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package gohacks contains utilities for subverting the Go compiler. +package gohacks + +import ( + "reflect" + "unsafe" +) + +// Noescape hides a pointer from escape analysis. Noescape is the identity +// function but escape analysis doesn't think the output depends on the input. +// Noescape is inlined and currently compiles down to zero instructions. +// USE CAREFULLY! +// +// (Noescape is copy/pasted from Go's runtime/stubs.go:noescape().) +// +//go:nosplit +func Noescape(p unsafe.Pointer) unsafe.Pointer { + x := uintptr(p) + return unsafe.Pointer(x ^ 0) +} + +// ImmutableBytesFromString is equivalent to []byte(s), except that it uses the +// same memory backing s instead of making a heap-allocated copy. This is only +// valid if the returned slice is never mutated. +func ImmutableBytesFromString(s string) []byte { + shdr := (*reflect.StringHeader)(unsafe.Pointer(&s)) + var bs []byte + bshdr := (*reflect.SliceHeader)(unsafe.Pointer(&bs)) + bshdr.Data = shdr.Data + bshdr.Len = shdr.Len + bshdr.Cap = shdr.Len + return bs +} + +// StringFromImmutableBytes is equivalent to string(bs), except that it uses +// the same memory backing bs instead of making a heap-allocated copy. This is +// only valid if bs is never mutated after StringFromImmutableBytes returns. +func StringFromImmutableBytes(bs []byte) string { + // This is cheaper than messing with reflect.StringHeader and + // reflect.SliceHeader, which as of this writing produces many dead stores + // of zeroes. Compare strings.Builder.String(). + return *(*string)(unsafe.Pointer(&bs)) +} diff --git a/pkg/sentry/fsbridge/vfs.go b/pkg/sentry/fsbridge/vfs.go index e657c39bc..6aa17bfc1 100644 --- a/pkg/sentry/fsbridge/vfs.go +++ b/pkg/sentry/fsbridge/vfs.go @@ -117,15 +117,19 @@ func NewVFSLookup(mntns *vfs.MountNamespace, root, workingDir vfs.VirtualDentry) // default anyways. // // TODO(gvisor.dev/issue/1623): Check mount has read and exec permission. -func (l *vfsLookup) OpenPath(ctx context.Context, path string, opts vfs.OpenOptions, _ *uint, resolveFinal bool) (File, error) { +func (l *vfsLookup) OpenPath(ctx context.Context, pathname string, opts vfs.OpenOptions, _ *uint, resolveFinal bool) (File, error) { vfsObj := l.mntns.Root().Mount().Filesystem().VirtualFilesystem() creds := auth.CredentialsFromContext(ctx) + path := fspath.Parse(pathname) pop := &vfs.PathOperation{ Root: l.root, - Start: l.root, - Path: fspath.Parse(path), + Start: l.workingDir, + Path: path, FollowFinalSymlink: resolveFinal, } + if path.Absolute { + pop.Start = l.root + } fd, err := vfsObj.OpenAt(ctx, creds, pop, &opts) if err != nil { return nil, err diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index ce08a7d53..10c08fa90 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -73,9 +73,9 @@ func newTasksInode(inoGen InoGenerator, k *kernel.Kernel, pidns *kernel.PIDNames "meminfo": newDentry(root, inoGen.NextIno(), 0444, &meminfoData{}), "mounts": kernfs.NewStaticSymlink(root, inoGen.NextIno(), "self/mounts"), "net": newNetDir(root, inoGen, k), - "stat": newDentry(root, inoGen.NextIno(), 0444, &statData{}), + "stat": newDentry(root, inoGen.NextIno(), 0444, &statData{k: k}), "uptime": newDentry(root, inoGen.NextIno(), 0444, &uptimeData{}), - "version": newDentry(root, inoGen.NextIno(), 0444, &versionData{}), + "version": newDentry(root, inoGen.NextIno(), 0444, &versionData{k: k}), } inode := &tasksInode{ diff --git a/pkg/sentry/kernel/fd_table.go b/pkg/sentry/kernel/fd_table.go index 23b88f7a6..58001d56c 100644 --- a/pkg/sentry/kernel/fd_table.go +++ b/pkg/sentry/kernel/fd_table.go @@ -296,6 +296,50 @@ func (f *FDTable) NewFDs(ctx context.Context, fd int32, files []*fs.File, flags return fds, nil } +// NewFDVFS2 allocates a file descriptor greater than or equal to minfd for +// the given file description. If it succeeds, it takes a reference on file. +func (f *FDTable) NewFDVFS2(ctx context.Context, minfd int32, file *vfs.FileDescription, flags FDFlags) (int32, error) { + if minfd < 0 { + // Don't accept negative FDs. + return -1, syscall.EINVAL + } + + // Default limit. + end := int32(math.MaxInt32) + + // Ensure we don't get past the provided limit. + if limitSet := limits.FromContext(ctx); limitSet != nil { + lim := limitSet.Get(limits.NumberOfFiles) + if lim.Cur != limits.Infinity { + end = int32(lim.Cur) + } + if minfd >= end { + return -1, syscall.EMFILE + } + } + + f.mu.Lock() + defer f.mu.Unlock() + + // From f.next to find available fd. + fd := minfd + if fd < f.next { + fd = f.next + } + for fd < end { + if d, _, _ := f.get(fd); d == nil { + f.setVFS2(fd, file, flags) + if fd == f.next { + // Update next search start position. + f.next = fd + 1 + } + return fd, nil + } + fd++ + } + return -1, syscall.EMFILE +} + // NewFDAt sets the file reference for the given FD. If there is an active // reference for that FD, the ref count for that existing reference is // decremented. @@ -316,9 +360,6 @@ func (f *FDTable) newFDAt(ctx context.Context, fd int32, file *fs.File, fileVFS2 return syscall.EBADF } - f.mu.Lock() - defer f.mu.Unlock() - // Check the limit for the provided file. if limitSet := limits.FromContext(ctx); limitSet != nil { if lim := limitSet.Get(limits.NumberOfFiles); lim.Cur != limits.Infinity && uint64(fd) >= lim.Cur { @@ -327,6 +368,8 @@ func (f *FDTable) newFDAt(ctx context.Context, fd int32, file *fs.File, fileVFS2 } // Install the entry. + f.mu.Lock() + defer f.mu.Unlock() f.setAll(fd, file, fileVFS2, flags) return nil } diff --git a/pkg/sentry/kernel/fs_context.go b/pkg/sentry/kernel/fs_context.go index 7218aa24e..47f78df9a 100644 --- a/pkg/sentry/kernel/fs_context.go +++ b/pkg/sentry/kernel/fs_context.go @@ -244,6 +244,28 @@ func (f *FSContext) SetRootDirectory(d *fs.Dirent) { old.DecRef() } +// SetRootDirectoryVFS2 sets the root directory. It takes a reference on vd. +// +// This is not a valid call after free. +func (f *FSContext) SetRootDirectoryVFS2(vd vfs.VirtualDentry) { + if !vd.Ok() { + panic("FSContext.SetRootDirectoryVFS2 called with zero-value VirtualDentry") + } + + f.mu.Lock() + + if !f.rootVFS2.Ok() { + f.mu.Unlock() + panic(fmt.Sprintf("FSContext.SetRootDirectoryVFS2(%v)) called after destroy", vd)) + } + + old := f.rootVFS2 + vd.IncRef() + f.rootVFS2 = vd + f.mu.Unlock() + old.DecRef() +} + // Umask returns the current umask. func (f *FSContext) Umask() uint { f.mu.Lock() diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index e37e23231..2cee2e6ed 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -789,6 +789,15 @@ func (t *Task) NewFDFrom(fd int32, file *fs.File, flags FDFlags) (int32, error) return fds[0], nil } +// NewFDFromVFS2 is a convenience wrapper for t.FDTable().NewFDVFS2. +// +// This automatically passes the task as the context. +// +// Precondition: same as FDTable.Get. +func (t *Task) NewFDFromVFS2(fd int32, file *vfs.FileDescription, flags FDFlags) (int32, error) { + return t.fdTable.NewFDVFS2(t, fd, file, flags) +} + // NewFDAt is a convenience wrapper for t.FDTable().NewFDAt. // // This automatically passes the task as the context. @@ -798,6 +807,15 @@ func (t *Task) NewFDAt(fd int32, file *fs.File, flags FDFlags) error { return t.fdTable.NewFDAt(t, fd, file, flags) } +// NewFDAtVFS2 is a convenience wrapper for t.FDTable().NewFDAtVFS2. +// +// This automatically passes the task as the context. +// +// Precondition: same as FDTable. +func (t *Task) NewFDAtVFS2(fd int32, file *vfs.FileDescription, flags FDFlags) error { + return t.fdTable.NewFDAtVFS2(t, fd, file, flags) +} + // WithMuLocked executes f with t.mu locked. func (t *Task) WithMuLocked(f func(*Task)) { t.mu.Lock() diff --git a/pkg/sentry/syscalls/linux/sys_epoll.go b/pkg/sentry/syscalls/linux/sys_epoll.go index fbef5b376..3ab93fbde 100644 --- a/pkg/sentry/syscalls/linux/sys_epoll.go +++ b/pkg/sentry/syscalls/linux/sys_epoll.go @@ -25,6 +25,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +// LINT.IfChange + // EpollCreate1 implements the epoll_create1(2) linux syscall. func EpollCreate1(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { flags := args[0].Int() @@ -164,3 +166,5 @@ func EpollPwait(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy return EpollWait(t, args) } + +// LINT.ThenChange(vfs2/epoll.go) diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go index 421845ebb..c21f14dc0 100644 --- a/pkg/sentry/syscalls/linux/sys_file.go +++ b/pkg/sentry/syscalls/linux/sys_file.go @@ -130,6 +130,8 @@ func copyInPath(t *kernel.Task, addr usermem.Addr, allowEmpty bool) (path string return path, dirPath, nil } +// LINT.IfChange + func openAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint) (fd uintptr, err error) { path, dirPath, err := copyInPath(t, addr, false /* allowEmpty */) if err != nil { @@ -575,6 +577,10 @@ func Faccessat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys return 0, nil, accessAt(t, dirFD, addr, flags&linux.AT_SYMLINK_NOFOLLOW == 0, mode) } +// LINT.ThenChange(vfs2/filesystem.go) + +// LINT.IfChange + // Ioctl implements linux syscall ioctl(2). func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { fd := args[0].Int() @@ -650,6 +656,10 @@ func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } } +// LINT.ThenChange(vfs2/ioctl.go) + +// LINT.IfChange + // Getcwd implements the linux syscall getcwd(2). func Getcwd(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { addr := args[0].Pointer() @@ -760,6 +770,10 @@ func Fchdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal return 0, nil, nil } +// LINT.ThenChange(vfs2/fscontext.go) + +// LINT.IfChange + // Close implements linux syscall close(2). func Close(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { fd := args[0].Int() @@ -1094,6 +1108,8 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } } +// LINT.ThenChange(vfs2/fd.go) + const ( _FADV_NORMAL = 0 _FADV_RANDOM = 1 @@ -1141,6 +1157,8 @@ func Fadvise64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys return 0, nil, nil } +// LINT.IfChange + func mkdirAt(t *kernel.Task, dirFD int32, addr usermem.Addr, mode linux.FileMode) error { path, _, err := copyInPath(t, addr, false /* allowEmpty */) if err != nil { @@ -1421,6 +1439,10 @@ func Linkat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal return 0, nil, linkAt(t, oldDirFD, oldAddr, newDirFD, newAddr, resolve, allowEmpty) } +// LINT.ThenChange(vfs2/filesystem.go) + +// LINT.IfChange + func readlinkAt(t *kernel.Task, dirFD int32, addr usermem.Addr, bufAddr usermem.Addr, size uint) (copied uintptr, err error) { path, dirPath, err := copyInPath(t, addr, false /* allowEmpty */) if err != nil { @@ -1480,6 +1502,10 @@ func Readlinkat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy return n, nil, err } +// LINT.ThenChange(vfs2/stat.go) + +// LINT.IfChange + func unlinkAt(t *kernel.Task, dirFD int32, addr usermem.Addr) error { path, dirPath, err := copyInPath(t, addr, false /* allowEmpty */) if err != nil { @@ -1516,6 +1542,10 @@ func Unlinkat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc return 0, nil, unlinkAt(t, dirFD, addr) } +// LINT.ThenChange(vfs2/filesystem.go) + +// LINT.IfChange + // Truncate implements linux syscall truncate(2). func Truncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { addr := args[0].Pointer() @@ -1614,6 +1644,8 @@ func Ftruncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys return 0, nil, nil } +// LINT.ThenChange(vfs2/setstat.go) + // Umask implements linux syscall umask(2). func Umask(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { mask := args[0].ModeT() @@ -1621,6 +1653,8 @@ func Umask(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return uintptr(mask), nil, nil } +// LINT.IfChange + // Change ownership of a file. // // uid and gid may be -1, in which case they will not be changed. @@ -1987,6 +2021,10 @@ func Futimesat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys return 0, nil, utimes(t, dirFD, pathnameAddr, ts, true) } +// LINT.ThenChange(vfs2/setstat.go) + +// LINT.IfChange + func renameAt(t *kernel.Task, oldDirFD int32, oldAddr usermem.Addr, newDirFD int32, newAddr usermem.Addr) error { newPath, _, err := copyInPath(t, newAddr, false /* allowEmpty */) if err != nil { @@ -2042,6 +2080,8 @@ func Renameat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc return 0, nil, renameAt(t, oldDirFD, oldPathAddr, newDirFD, newPathAddr) } +// LINT.ThenChange(vfs2/filesystem.go) + // Fallocate implements linux system call fallocate(2). func Fallocate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { fd := args[0].Int() diff --git a/pkg/sentry/syscalls/linux/sys_getdents.go b/pkg/sentry/syscalls/linux/sys_getdents.go index f66f4ffde..b126fecc0 100644 --- a/pkg/sentry/syscalls/linux/sys_getdents.go +++ b/pkg/sentry/syscalls/linux/sys_getdents.go @@ -27,6 +27,8 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) +// LINT.IfChange + // Getdents implements linux syscall getdents(2) for 64bit systems. func Getdents(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { fd := args[0].Int() @@ -244,3 +246,5 @@ func (ds *direntSerializer) CopyOut(name string, attr fs.DentAttr) error { func (ds *direntSerializer) Written() int { return ds.written } + +// LINT.ThenChange(vfs2/getdents.go) diff --git a/pkg/sentry/syscalls/linux/sys_lseek.go b/pkg/sentry/syscalls/linux/sys_lseek.go index 297e920c4..3f7691eae 100644 --- a/pkg/sentry/syscalls/linux/sys_lseek.go +++ b/pkg/sentry/syscalls/linux/sys_lseek.go @@ -21,6 +21,8 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) +// LINT.IfChange + // Lseek implements linux syscall lseek(2). func Lseek(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { fd := args[0].Int() @@ -52,3 +54,5 @@ func Lseek(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } return uintptr(offset), nil, err } + +// LINT.ThenChange(vfs2/read_write.go) diff --git a/pkg/sentry/syscalls/linux/sys_mmap.go b/pkg/sentry/syscalls/linux/sys_mmap.go index 9959f6e61..91694d374 100644 --- a/pkg/sentry/syscalls/linux/sys_mmap.go +++ b/pkg/sentry/syscalls/linux/sys_mmap.go @@ -35,6 +35,8 @@ func Brk(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo return uintptr(addr), nil, nil } +// LINT.IfChange + // Mmap implements linux syscall mmap(2). func Mmap(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { prot := args[2].Int() @@ -104,6 +106,8 @@ func Mmap(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC return uintptr(rv), nil, err } +// LINT.ThenChange(vfs2/mmap.go) + // Munmap implements linux syscall munmap(2). func Munmap(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { return 0, nil, t.MemoryManager().MUnmap(t, args[0].Pointer(), args[1].Uint64()) diff --git a/pkg/sentry/syscalls/linux/sys_read.go b/pkg/sentry/syscalls/linux/sys_read.go index 227692f06..78a2cb750 100644 --- a/pkg/sentry/syscalls/linux/sys_read.go +++ b/pkg/sentry/syscalls/linux/sys_read.go @@ -28,6 +28,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +// LINT.IfChange + const ( // EventMaskRead contains events that can be triggered on reads. EventMaskRead = waiter.EventIn | waiter.EventHUp | waiter.EventErr @@ -388,3 +390,5 @@ func preadv(t *kernel.Task, f *fs.File, dst usermem.IOSequence, offset int64) (i return total, err } + +// LINT.ThenChange(vfs2/read_write.go) diff --git a/pkg/sentry/syscalls/linux/sys_stat.go b/pkg/sentry/syscalls/linux/sys_stat.go index 11f25e00d..701b27b4a 100644 --- a/pkg/sentry/syscalls/linux/sys_stat.go +++ b/pkg/sentry/syscalls/linux/sys_stat.go @@ -23,6 +23,8 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) +// LINT.IfChange + func statFromAttrs(t *kernel.Task, sattr fs.StableAttr, uattr fs.UnstableAttr) linux.Stat { return linux.Stat{ Dev: sattr.DeviceID, @@ -297,3 +299,5 @@ func statfsImpl(t *kernel.Task, d *fs.Dirent, addr usermem.Addr) error { _, err = t.CopyOut(addr, &statfs) return err } + +// LINT.ThenChange(vfs2/stat.go) diff --git a/pkg/sentry/syscalls/linux/sys_sync.go b/pkg/sentry/syscalls/linux/sys_sync.go index 3e55235bd..5ad465ae3 100644 --- a/pkg/sentry/syscalls/linux/sys_sync.go +++ b/pkg/sentry/syscalls/linux/sys_sync.go @@ -22,6 +22,8 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) +// LINT.IfChange + // Sync implements linux system call sync(2). func Sync(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { t.MountNamespace().SyncAll(t) @@ -135,3 +137,5 @@ func SyncFileRange(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel return 0, nil, syserror.ConvertIntr(err, kernel.ERESTARTSYS) } + +// LINT.ThenChange(vfs2/sync.go) diff --git a/pkg/sentry/syscalls/linux/sys_write.go b/pkg/sentry/syscalls/linux/sys_write.go index aba892939..506ee54ce 100644 --- a/pkg/sentry/syscalls/linux/sys_write.go +++ b/pkg/sentry/syscalls/linux/sys_write.go @@ -28,6 +28,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +// LINT.IfChange + const ( // EventMaskWrite contains events that can be triggered on writes. // @@ -358,3 +360,5 @@ func pwritev(t *kernel.Task, f *fs.File, src usermem.IOSequence, offset int64) ( return total, err } + +// LINT.ThenChange(vfs2/read_write.go) diff --git a/pkg/sentry/syscalls/linux/sys_xattr.go b/pkg/sentry/syscalls/linux/sys_xattr.go index 9d8140b8a..2de5e3422 100644 --- a/pkg/sentry/syscalls/linux/sys_xattr.go +++ b/pkg/sentry/syscalls/linux/sys_xattr.go @@ -25,6 +25,8 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) +// LINT.IfChange + // GetXattr implements linux syscall getxattr(2). func GetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { return getXattrFromPath(t, args, true) @@ -418,3 +420,5 @@ func removeXattr(t *kernel.Task, d *fs.Dirent, nameAddr usermem.Addr) error { return d.Inode.RemoveXattr(t, d, name) } + +// LINT.ThenChange(vfs2/xattr.go) diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD index 6b8a00b6e..f51761e81 100644 --- a/pkg/sentry/syscalls/linux/vfs2/BUILD +++ b/pkg/sentry/syscalls/linux/vfs2/BUILD @@ -5,18 +5,44 @@ package(licenses = ["notice"]) go_library( name = "vfs2", srcs = [ + "epoll.go", + "epoll_unsafe.go", + "execve.go", + "fd.go", + "filesystem.go", + "fscontext.go", + "getdents.go", + "ioctl.go", "linux64.go", "linux64_override_amd64.go", "linux64_override_arm64.go", - "sys_read.go", + "mmap.go", + "path.go", + "poll.go", + "read_write.go", + "setstat.go", + "stat.go", + "sync.go", + "xattr.go", ], + marshal = True, visibility = ["//:sandbox"], deps = [ + "//pkg/abi/linux", + "//pkg/fspath", + "//pkg/gohacks", "//pkg/sentry/arch", + "//pkg/sentry/fsbridge", "//pkg/sentry/kernel", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/kernel/time", + "//pkg/sentry/limits", + "//pkg/sentry/loader", + "//pkg/sentry/memmap", "//pkg/sentry/syscalls", "//pkg/sentry/syscalls/linux", "//pkg/sentry/vfs", + "//pkg/sync", "//pkg/syserror", "//pkg/usermem", "//pkg/waiter", diff --git a/pkg/sentry/syscalls/linux/vfs2/epoll.go b/pkg/sentry/syscalls/linux/vfs2/epoll.go new file mode 100644 index 000000000..d6cb0e79a --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/epoll.go @@ -0,0 +1,225 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "math" + "time" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +// EpollCreate1 implements Linux syscall epoll_create1(2). +func EpollCreate1(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + flags := args[0].Int() + if flags&^linux.EPOLL_CLOEXEC != 0 { + return 0, nil, syserror.EINVAL + } + + file, err := t.Kernel().VFS().NewEpollInstanceFD() + if err != nil { + return 0, nil, err + } + defer file.DecRef() + + fd, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{ + CloseOnExec: flags&linux.EPOLL_CLOEXEC != 0, + }) + if err != nil { + return 0, nil, err + } + return uintptr(fd), nil, nil +} + +// EpollCreate implements Linux syscall epoll_create(2). +func EpollCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + size := args[0].Int() + + // "Since Linux 2.6.8, the size argument is ignored, but must be greater + // than zero" - epoll_create(2) + if size <= 0 { + return 0, nil, syserror.EINVAL + } + + file, err := t.Kernel().VFS().NewEpollInstanceFD() + if err != nil { + return 0, nil, err + } + defer file.DecRef() + + fd, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{}) + if err != nil { + return 0, nil, err + } + return uintptr(fd), nil, nil +} + +// EpollCtl implements Linux syscall epoll_ctl(2). +func EpollCtl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + epfd := args[0].Int() + op := args[1].Int() + fd := args[2].Int() + eventAddr := args[3].Pointer() + + epfile := t.GetFileVFS2(epfd) + if epfile == nil { + return 0, nil, syserror.EBADF + } + defer epfile.DecRef() + ep, ok := epfile.Impl().(*vfs.EpollInstance) + if !ok { + return 0, nil, syserror.EINVAL + } + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + if epfile == file { + return 0, nil, syserror.EINVAL + } + + var event linux.EpollEvent + switch op { + case linux.EPOLL_CTL_ADD: + if err := event.CopyIn(t, eventAddr); err != nil { + return 0, nil, err + } + return 0, nil, ep.AddInterest(file, fd, event) + case linux.EPOLL_CTL_DEL: + return 0, nil, ep.DeleteInterest(file, fd) + case linux.EPOLL_CTL_MOD: + if err := event.CopyIn(t, eventAddr); err != nil { + return 0, nil, err + } + return 0, nil, ep.ModifyInterest(file, fd, event) + default: + return 0, nil, syserror.EINVAL + } +} + +// EpollWait implements Linux syscall epoll_wait(2). +func EpollWait(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + epfd := args[0].Int() + eventsAddr := args[1].Pointer() + maxEvents := int(args[2].Int()) + timeout := int(args[3].Int()) + + const _EP_MAX_EVENTS = math.MaxInt32 / sizeofEpollEvent // Linux: fs/eventpoll.c:EP_MAX_EVENTS + if maxEvents <= 0 || maxEvents > _EP_MAX_EVENTS { + return 0, nil, syserror.EINVAL + } + + epfile := t.GetFileVFS2(epfd) + if epfile == nil { + return 0, nil, syserror.EBADF + } + defer epfile.DecRef() + ep, ok := epfile.Impl().(*vfs.EpollInstance) + if !ok { + return 0, nil, syserror.EINVAL + } + + // Use a fixed-size buffer in a loop, instead of make([]linux.EpollEvent, + // maxEvents), so that the buffer can be allocated on the stack. + var ( + events [16]linux.EpollEvent + total int + ch chan struct{} + haveDeadline bool + deadline ktime.Time + ) + for { + batchEvents := len(events) + if batchEvents > maxEvents { + batchEvents = maxEvents + } + n := ep.ReadEvents(events[:batchEvents]) + maxEvents -= n + if n != 0 { + // Copy what we read out. + copiedEvents, err := copyOutEvents(t, eventsAddr, events[:n]) + eventsAddr += usermem.Addr(copiedEvents * sizeofEpollEvent) + total += copiedEvents + if err != nil { + if total != 0 { + return uintptr(total), nil, nil + } + return 0, nil, err + } + // If we've filled the application's event buffer, we're done. + if maxEvents == 0 { + return uintptr(total), nil, nil + } + // Loop if we read a full batch, under the expectation that there + // may be more events to read. + if n == batchEvents { + continue + } + } + // We get here if n != batchEvents. If we read any number of events + // (just now, or in a previous iteration of this loop), or if timeout + // is 0 (such that epoll_wait should be non-blocking), return the + // events we've read so far to the application. + if total != 0 || timeout == 0 { + return uintptr(total), nil, nil + } + // In the first iteration of this loop, register with the epoll + // instance for readability events, but then immediately continue the + // loop since we need to retry ReadEvents() before blocking. In all + // subsequent iterations, block until events are available, the timeout + // expires, or an interrupt arrives. + if ch == nil { + var w waiter.Entry + w, ch = waiter.NewChannelEntry(nil) + epfile.EventRegister(&w, waiter.EventIn) + defer epfile.EventUnregister(&w) + } else { + // Set up the timer if a timeout was specified. + if timeout > 0 && !haveDeadline { + timeoutDur := time.Duration(timeout) * time.Millisecond + deadline = t.Kernel().MonotonicClock().Now().Add(timeoutDur) + haveDeadline = true + } + if err := t.BlockWithDeadline(ch, haveDeadline, deadline); err != nil { + if err == syserror.ETIMEDOUT { + err = nil + } + // total must be 0 since otherwise we would have returned + // above. + return 0, nil, err + } + } + } +} + +// EpollPwait implements Linux syscall epoll_pwait(2). +func EpollPwait(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + maskAddr := args[4].Pointer() + maskSize := uint(args[5].Uint()) + + if err := setTempSignalSet(t, maskAddr, maskSize); err != nil { + return 0, nil, err + } + + return EpollWait(t, args) +} diff --git a/pkg/sentry/syscalls/linux/vfs2/epoll_unsafe.go b/pkg/sentry/syscalls/linux/vfs2/epoll_unsafe.go new file mode 100644 index 000000000..825f325bf --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/epoll_unsafe.go @@ -0,0 +1,44 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "reflect" + "runtime" + "unsafe" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/gohacks" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/usermem" +) + +const sizeofEpollEvent = int(unsafe.Sizeof(linux.EpollEvent{})) + +func copyOutEvents(t *kernel.Task, addr usermem.Addr, events []linux.EpollEvent) (int, error) { + if len(events) == 0 { + return 0, nil + } + // Cast events to a byte slice for copying. + var eventBytes []byte + eventBytesHdr := (*reflect.SliceHeader)(unsafe.Pointer(&eventBytes)) + eventBytesHdr.Data = uintptr(gohacks.Noescape(unsafe.Pointer(&events[0]))) + eventBytesHdr.Len = len(events) * sizeofEpollEvent + eventBytesHdr.Cap = len(events) * sizeofEpollEvent + copiedBytes, err := t.CopyOutBytes(addr, eventBytes) + runtime.KeepAlive(events) + copiedEvents := copiedBytes / sizeofEpollEvent // rounded down + return copiedEvents, err +} diff --git a/pkg/sentry/syscalls/linux/vfs2/execve.go b/pkg/sentry/syscalls/linux/vfs2/execve.go new file mode 100644 index 000000000..aef0078a8 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/execve.go @@ -0,0 +1,137 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fsbridge" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/loader" + slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +// Execve implements linux syscall execve(2). +func Execve(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + pathnameAddr := args[0].Pointer() + argvAddr := args[1].Pointer() + envvAddr := args[2].Pointer() + return execveat(t, linux.AT_FDCWD, pathnameAddr, argvAddr, envvAddr, 0 /* flags */) +} + +// Execveat implements linux syscall execveat(2). +func Execveat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + dirfd := args[0].Int() + pathnameAddr := args[1].Pointer() + argvAddr := args[2].Pointer() + envvAddr := args[3].Pointer() + flags := args[4].Int() + return execveat(t, dirfd, pathnameAddr, argvAddr, envvAddr, flags) +} + +func execveat(t *kernel.Task, dirfd int32, pathnameAddr, argvAddr, envvAddr usermem.Addr, flags int32) (uintptr, *kernel.SyscallControl, error) { + if flags&^(linux.AT_EMPTY_PATH|linux.AT_SYMLINK_NOFOLLOW) != 0 { + return 0, nil, syserror.EINVAL + } + + pathname, err := t.CopyInString(pathnameAddr, linux.PATH_MAX) + if err != nil { + return 0, nil, err + } + var argv, envv []string + if argvAddr != 0 { + var err error + argv, err = t.CopyInVector(argvAddr, slinux.ExecMaxElemSize, slinux.ExecMaxTotalSize) + if err != nil { + return 0, nil, err + } + } + if envvAddr != 0 { + var err error + envv, err = t.CopyInVector(envvAddr, slinux.ExecMaxElemSize, slinux.ExecMaxTotalSize) + if err != nil { + return 0, nil, err + } + } + + root := t.FSContext().RootDirectoryVFS2() + defer root.DecRef() + var executable fsbridge.File + closeOnExec := false + if path := fspath.Parse(pathname); dirfd != linux.AT_FDCWD && !path.Absolute { + // We must open the executable ourselves since dirfd is used as the + // starting point while resolving path, but the task working directory + // is used as the starting point while resolving interpreters (Linux: + // fs/binfmt_script.c:load_script() => fs/exec.c:open_exec() => + // do_open_execat(fd=AT_FDCWD)), and the loader package is currently + // incapable of handling this correctly. + if !path.HasComponents() && flags&linux.AT_EMPTY_PATH == 0 { + return 0, nil, syserror.ENOENT + } + dirfile, dirfileFlags := t.FDTable().GetVFS2(dirfd) + if dirfile == nil { + return 0, nil, syserror.EBADF + } + start := dirfile.VirtualDentry() + start.IncRef() + dirfile.DecRef() + closeOnExec = dirfileFlags.CloseOnExec + file, err := t.Kernel().VFS().OpenAt(t, t.Credentials(), &vfs.PathOperation{ + Root: root, + Start: start, + Path: path, + FollowFinalSymlink: flags&linux.AT_SYMLINK_NOFOLLOW == 0, + }, &vfs.OpenOptions{ + Flags: linux.O_RDONLY, + FileExec: true, + }) + start.DecRef() + if err != nil { + return 0, nil, err + } + defer file.DecRef() + executable = fsbridge.NewVFSFile(file) + } + + // Load the new TaskContext. + mntns := t.MountNamespaceVFS2() // FIXME(jamieliu): useless refcount change + defer mntns.DecRef() + wd := t.FSContext().WorkingDirectoryVFS2() + defer wd.DecRef() + remainingTraversals := uint(linux.MaxSymlinkTraversals) + loadArgs := loader.LoadArgs{ + Opener: fsbridge.NewVFSLookup(mntns, root, wd), + RemainingTraversals: &remainingTraversals, + ResolveFinal: flags&linux.AT_SYMLINK_NOFOLLOW == 0, + Filename: pathname, + File: executable, + CloseOnExec: closeOnExec, + Argv: argv, + Envv: envv, + Features: t.Arch().FeatureSet(), + } + + tc, se := t.Kernel().LoadTaskImage(t, loadArgs) + if se != nil { + return 0, nil, se.ToError() + } + + ctrl, err := t.Execve(tc) + return 0, ctrl, err +} diff --git a/pkg/sentry/syscalls/linux/vfs2/fd.go b/pkg/sentry/syscalls/linux/vfs2/fd.go new file mode 100644 index 000000000..3afcea665 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/fd.go @@ -0,0 +1,147 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" + "gvisor.dev/gvisor/pkg/syserror" +) + +// Close implements Linux syscall close(2). +func Close(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + + // Note that Remove provides a reference on the file that we may use to + // flush. It is still active until we drop the final reference below + // (and other reference-holding operations complete). + _, file := t.FDTable().Remove(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + err := file.OnClose(t) + return 0, nil, slinux.HandleIOErrorVFS2(t, false /* partial */, err, syserror.EINTR, "close", file) +} + +// Dup implements Linux syscall dup(2). +func Dup(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + newFD, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{}) + if err != nil { + return 0, nil, syserror.EMFILE + } + return uintptr(newFD), nil, nil +} + +// Dup2 implements Linux syscall dup2(2). +func Dup2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + oldfd := args[0].Int() + newfd := args[1].Int() + + if oldfd == newfd { + // As long as oldfd is valid, dup2() does nothing and returns newfd. + file := t.GetFileVFS2(oldfd) + if file == nil { + return 0, nil, syserror.EBADF + } + file.DecRef() + return uintptr(newfd), nil, nil + } + + return dup3(t, oldfd, newfd, 0) +} + +// Dup3 implements Linux syscall dup3(2). +func Dup3(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + oldfd := args[0].Int() + newfd := args[1].Int() + flags := args[2].Uint() + + if oldfd == newfd { + return 0, nil, syserror.EINVAL + } + + return dup3(t, oldfd, newfd, flags) +} + +func dup3(t *kernel.Task, oldfd, newfd int32, flags uint32) (uintptr, *kernel.SyscallControl, error) { + if flags&^linux.O_CLOEXEC != 0 { + return 0, nil, syserror.EINVAL + } + + file := t.GetFileVFS2(oldfd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + err := t.NewFDAtVFS2(newfd, file, kernel.FDFlags{ + CloseOnExec: flags&linux.O_CLOEXEC != 0, + }) + if err != nil { + return 0, nil, err + } + return uintptr(newfd), nil, nil +} + +// Fcntl implements linux syscall fcntl(2). +func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + cmd := args[1].Int() + + file, flags := t.FDTable().GetVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + switch cmd { + case linux.F_DUPFD, linux.F_DUPFD_CLOEXEC: + minfd := args[2].Int() + fd, err := t.NewFDFromVFS2(minfd, file, kernel.FDFlags{ + CloseOnExec: cmd == linux.F_DUPFD_CLOEXEC, + }) + if err != nil { + return 0, nil, err + } + return uintptr(fd), nil, nil + case linux.F_GETFD: + return uintptr(flags.ToLinuxFDFlags()), nil, nil + case linux.F_SETFD: + flags := args[2].Uint() + t.FDTable().SetFlags(fd, kernel.FDFlags{ + CloseOnExec: flags&linux.FD_CLOEXEC != 0, + }) + return 0, nil, nil + case linux.F_GETFL: + return uintptr(file.StatusFlags()), nil, nil + case linux.F_SETFL: + return 0, nil, file.SetStatusFlags(t, t.Credentials(), args[2].Uint()) + default: + // TODO(gvisor.dev/issue/1623): Everything else is not yet supported. + return 0, nil, syserror.EINVAL + } +} diff --git a/pkg/sentry/syscalls/linux/vfs2/filesystem.go b/pkg/sentry/syscalls/linux/vfs2/filesystem.go new file mode 100644 index 000000000..fc5ceea4c --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/filesystem.go @@ -0,0 +1,326 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +// Link implements Linux syscall link(2). +func Link(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + oldpathAddr := args[0].Pointer() + newpathAddr := args[1].Pointer() + return 0, nil, linkat(t, linux.AT_FDCWD, oldpathAddr, linux.AT_FDCWD, newpathAddr, 0 /* flags */) +} + +// Linkat implements Linux syscall linkat(2). +func Linkat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + olddirfd := args[0].Int() + oldpathAddr := args[1].Pointer() + newdirfd := args[2].Int() + newpathAddr := args[3].Pointer() + flags := args[4].Int() + return 0, nil, linkat(t, olddirfd, oldpathAddr, newdirfd, newpathAddr, flags) +} + +func linkat(t *kernel.Task, olddirfd int32, oldpathAddr usermem.Addr, newdirfd int32, newpathAddr usermem.Addr, flags int32) error { + if flags&^(linux.AT_EMPTY_PATH|linux.AT_SYMLINK_FOLLOW) != 0 { + return syserror.EINVAL + } + if flags&linux.AT_EMPTY_PATH != 0 && !t.HasCapability(linux.CAP_DAC_READ_SEARCH) { + return syserror.ENOENT + } + + oldpath, err := copyInPath(t, oldpathAddr) + if err != nil { + return err + } + oldtpop, err := getTaskPathOperation(t, olddirfd, oldpath, shouldAllowEmptyPath(flags&linux.AT_EMPTY_PATH != 0), shouldFollowFinalSymlink(flags&linux.AT_SYMLINK_FOLLOW != 0)) + if err != nil { + return err + } + defer oldtpop.Release() + + newpath, err := copyInPath(t, newpathAddr) + if err != nil { + return err + } + newtpop, err := getTaskPathOperation(t, newdirfd, newpath, disallowEmptyPath, nofollowFinalSymlink) + if err != nil { + return err + } + defer newtpop.Release() + + return t.Kernel().VFS().LinkAt(t, t.Credentials(), &oldtpop.pop, &newtpop.pop) +} + +// Mkdir implements Linux syscall mkdir(2). +func Mkdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + addr := args[0].Pointer() + mode := args[1].ModeT() + return 0, nil, mkdirat(t, linux.AT_FDCWD, addr, mode) +} + +// Mkdirat implements Linux syscall mkdirat(2). +func Mkdirat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + dirfd := args[0].Int() + addr := args[1].Pointer() + mode := args[2].ModeT() + return 0, nil, mkdirat(t, dirfd, addr, mode) +} + +func mkdirat(t *kernel.Task, dirfd int32, addr usermem.Addr, mode uint) error { + path, err := copyInPath(t, addr) + if err != nil { + return err + } + tpop, err := getTaskPathOperation(t, dirfd, path, disallowEmptyPath, nofollowFinalSymlink) + if err != nil { + return err + } + defer tpop.Release() + return t.Kernel().VFS().MkdirAt(t, t.Credentials(), &tpop.pop, &vfs.MkdirOptions{ + Mode: linux.FileMode(mode & (0777 | linux.S_ISVTX) &^ t.FSContext().Umask()), + }) +} + +// Mknod implements Linux syscall mknod(2). +func Mknod(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + addr := args[0].Pointer() + mode := args[1].ModeT() + dev := args[2].Uint() + return 0, nil, mknodat(t, linux.AT_FDCWD, addr, mode, dev) +} + +// Mknodat implements Linux syscall mknodat(2). +func Mknodat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + dirfd := args[0].Int() + addr := args[1].Pointer() + mode := args[2].ModeT() + dev := args[3].Uint() + return 0, nil, mknodat(t, dirfd, addr, mode, dev) +} + +func mknodat(t *kernel.Task, dirfd int32, addr usermem.Addr, mode uint, dev uint32) error { + path, err := copyInPath(t, addr) + if err != nil { + return err + } + tpop, err := getTaskPathOperation(t, dirfd, path, disallowEmptyPath, nofollowFinalSymlink) + if err != nil { + return err + } + defer tpop.Release() + major, minor := linux.DecodeDeviceID(dev) + return t.Kernel().VFS().MknodAt(t, t.Credentials(), &tpop.pop, &vfs.MknodOptions{ + Mode: linux.FileMode(mode &^ t.FSContext().Umask()), + DevMajor: uint32(major), + DevMinor: minor, + }) +} + +// Open implements Linux syscall open(2). +func Open(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + addr := args[0].Pointer() + flags := args[1].Uint() + mode := args[2].ModeT() + return openat(t, linux.AT_FDCWD, addr, flags, mode) +} + +// Openat implements Linux syscall openat(2). +func Openat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + dirfd := args[0].Int() + addr := args[1].Pointer() + flags := args[2].Uint() + mode := args[3].ModeT() + return openat(t, dirfd, addr, flags, mode) +} + +// Creat implements Linux syscall creat(2). +func Creat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + addr := args[0].Pointer() + mode := args[1].ModeT() + return openat(t, linux.AT_FDCWD, addr, linux.O_WRONLY|linux.O_CREAT|linux.O_TRUNC, mode) +} + +func openat(t *kernel.Task, dirfd int32, pathAddr usermem.Addr, flags uint32, mode uint) (uintptr, *kernel.SyscallControl, error) { + path, err := copyInPath(t, pathAddr) + if err != nil { + return 0, nil, err + } + tpop, err := getTaskPathOperation(t, dirfd, path, disallowEmptyPath, shouldFollowFinalSymlink(flags&linux.O_NOFOLLOW == 0)) + if err != nil { + return 0, nil, err + } + defer tpop.Release() + + file, err := t.Kernel().VFS().OpenAt(t, t.Credentials(), &tpop.pop, &vfs.OpenOptions{ + Flags: flags, + Mode: linux.FileMode(mode & (0777 | linux.S_ISUID | linux.S_ISGID | linux.S_ISVTX) &^ t.FSContext().Umask()), + }) + if err != nil { + return 0, nil, err + } + defer file.DecRef() + + fd, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{ + CloseOnExec: flags&linux.O_CLOEXEC != 0, + }) + return uintptr(fd), nil, err +} + +// Rename implements Linux syscall rename(2). +func Rename(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + oldpathAddr := args[0].Pointer() + newpathAddr := args[1].Pointer() + return 0, nil, renameat(t, linux.AT_FDCWD, oldpathAddr, linux.AT_FDCWD, newpathAddr, 0 /* flags */) +} + +// Renameat implements Linux syscall renameat(2). +func Renameat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + olddirfd := args[0].Int() + oldpathAddr := args[1].Pointer() + newdirfd := args[2].Int() + newpathAddr := args[3].Pointer() + return 0, nil, renameat(t, olddirfd, oldpathAddr, newdirfd, newpathAddr, 0 /* flags */) +} + +// Renameat2 implements Linux syscall renameat2(2). +func Renameat2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + olddirfd := args[0].Int() + oldpathAddr := args[1].Pointer() + newdirfd := args[2].Int() + newpathAddr := args[3].Pointer() + flags := args[4].Uint() + return 0, nil, renameat(t, olddirfd, oldpathAddr, newdirfd, newpathAddr, flags) +} + +func renameat(t *kernel.Task, olddirfd int32, oldpathAddr usermem.Addr, newdirfd int32, newpathAddr usermem.Addr, flags uint32) error { + oldpath, err := copyInPath(t, oldpathAddr) + if err != nil { + return err + } + // "If oldpath refers to a symbolic link, the link is renamed" - rename(2) + oldtpop, err := getTaskPathOperation(t, olddirfd, oldpath, disallowEmptyPath, nofollowFinalSymlink) + if err != nil { + return err + } + defer oldtpop.Release() + + newpath, err := copyInPath(t, newpathAddr) + if err != nil { + return err + } + newtpop, err := getTaskPathOperation(t, newdirfd, newpath, disallowEmptyPath, nofollowFinalSymlink) + if err != nil { + return err + } + defer newtpop.Release() + + return t.Kernel().VFS().RenameAt(t, t.Credentials(), &oldtpop.pop, &newtpop.pop, &vfs.RenameOptions{ + Flags: flags, + }) +} + +// Rmdir implements Linux syscall rmdir(2). +func Rmdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + pathAddr := args[0].Pointer() + return 0, nil, rmdirat(t, linux.AT_FDCWD, pathAddr) +} + +func rmdirat(t *kernel.Task, dirfd int32, pathAddr usermem.Addr) error { + path, err := copyInPath(t, pathAddr) + if err != nil { + return err + } + tpop, err := getTaskPathOperation(t, dirfd, path, disallowEmptyPath, followFinalSymlink) + if err != nil { + return err + } + defer tpop.Release() + return t.Kernel().VFS().RmdirAt(t, t.Credentials(), &tpop.pop) +} + +// Unlink implements Linux syscall unlink(2). +func Unlink(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + pathAddr := args[0].Pointer() + return 0, nil, unlinkat(t, linux.AT_FDCWD, pathAddr) +} + +func unlinkat(t *kernel.Task, dirfd int32, pathAddr usermem.Addr) error { + path, err := copyInPath(t, pathAddr) + if err != nil { + return err + } + tpop, err := getTaskPathOperation(t, dirfd, path, disallowEmptyPath, nofollowFinalSymlink) + if err != nil { + return err + } + defer tpop.Release() + return t.Kernel().VFS().UnlinkAt(t, t.Credentials(), &tpop.pop) +} + +// Unlinkat implements Linux syscall unlinkat(2). +func Unlinkat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + dirfd := args[0].Int() + pathAddr := args[1].Pointer() + flags := args[2].Int() + + if flags&^linux.AT_REMOVEDIR != 0 { + return 0, nil, syserror.EINVAL + } + + if flags&linux.AT_REMOVEDIR != 0 { + return 0, nil, rmdirat(t, dirfd, pathAddr) + } + return 0, nil, unlinkat(t, dirfd, pathAddr) +} + +// Symlink implements Linux syscall symlink(2). +func Symlink(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + targetAddr := args[0].Pointer() + linkpathAddr := args[1].Pointer() + return 0, nil, symlinkat(t, targetAddr, linux.AT_FDCWD, linkpathAddr) +} + +// Symlinkat implements Linux syscall symlinkat(2). +func Symlinkat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + targetAddr := args[0].Pointer() + newdirfd := args[1].Int() + linkpathAddr := args[2].Pointer() + return 0, nil, symlinkat(t, targetAddr, newdirfd, linkpathAddr) +} + +func symlinkat(t *kernel.Task, targetAddr usermem.Addr, newdirfd int32, linkpathAddr usermem.Addr) error { + target, err := t.CopyInString(targetAddr, linux.PATH_MAX) + if err != nil { + return err + } + linkpath, err := copyInPath(t, linkpathAddr) + if err != nil { + return err + } + tpop, err := getTaskPathOperation(t, newdirfd, linkpath, disallowEmptyPath, nofollowFinalSymlink) + if err != nil { + return err + } + defer tpop.Release() + return t.Kernel().VFS().SymlinkAt(t, t.Credentials(), &tpop.pop, target) +} diff --git a/pkg/sentry/syscalls/linux/vfs2/fscontext.go b/pkg/sentry/syscalls/linux/vfs2/fscontext.go new file mode 100644 index 000000000..317409a18 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/fscontext.go @@ -0,0 +1,131 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// Getcwd implements Linux syscall getcwd(2). +func Getcwd(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + addr := args[0].Pointer() + size := args[1].SizeT() + + root := t.FSContext().RootDirectoryVFS2() + wd := t.FSContext().WorkingDirectoryVFS2() + s, err := t.Kernel().VFS().PathnameForGetcwd(t, root, wd) + root.DecRef() + wd.DecRef() + if err != nil { + return 0, nil, err + } + + // Note this is >= because we need a terminator. + if uint(len(s)) >= size { + return 0, nil, syserror.ERANGE + } + + // Construct a byte slice containing a NUL terminator. + buf := t.CopyScratchBuffer(len(s) + 1) + copy(buf, s) + buf[len(buf)-1] = 0 + + // Write the pathname slice. + n, err := t.CopyOutBytes(addr, buf) + if err != nil { + return 0, nil, err + } + return uintptr(n), nil, nil +} + +// Chdir implements Linux syscall chdir(2). +func Chdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + addr := args[0].Pointer() + + path, err := copyInPath(t, addr) + if err != nil { + return 0, nil, err + } + tpop, err := getTaskPathOperation(t, linux.AT_FDCWD, path, disallowEmptyPath, followFinalSymlink) + if err != nil { + return 0, nil, err + } + defer tpop.Release() + + vd, err := t.Kernel().VFS().GetDentryAt(t, t.Credentials(), &tpop.pop, &vfs.GetDentryOptions{ + CheckSearchable: true, + }) + if err != nil { + return 0, nil, err + } + t.FSContext().SetWorkingDirectoryVFS2(vd) + vd.DecRef() + return 0, nil, nil +} + +// Fchdir implements Linux syscall fchdir(2). +func Fchdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + + tpop, err := getTaskPathOperation(t, fd, fspath.Path{}, allowEmptyPath, nofollowFinalSymlink) + if err != nil { + return 0, nil, err + } + defer tpop.Release() + + vd, err := t.Kernel().VFS().GetDentryAt(t, t.Credentials(), &tpop.pop, &vfs.GetDentryOptions{ + CheckSearchable: true, + }) + if err != nil { + return 0, nil, err + } + t.FSContext().SetWorkingDirectoryVFS2(vd) + vd.DecRef() + return 0, nil, nil +} + +// Chroot implements Linux syscall chroot(2). +func Chroot(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + addr := args[0].Pointer() + + if !t.HasCapability(linux.CAP_SYS_CHROOT) { + return 0, nil, syserror.EPERM + } + + path, err := copyInPath(t, addr) + if err != nil { + return 0, nil, err + } + tpop, err := getTaskPathOperation(t, linux.AT_FDCWD, path, disallowEmptyPath, followFinalSymlink) + if err != nil { + return 0, nil, err + } + defer tpop.Release() + + vd, err := t.Kernel().VFS().GetDentryAt(t, t.Credentials(), &tpop.pop, &vfs.GetDentryOptions{ + CheckSearchable: true, + }) + if err != nil { + return 0, nil, err + } + t.FSContext().SetRootDirectoryVFS2(vd) + vd.DecRef() + return 0, nil, nil +} diff --git a/pkg/sentry/syscalls/linux/vfs2/getdents.go b/pkg/sentry/syscalls/linux/vfs2/getdents.go new file mode 100644 index 000000000..ddc140b65 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/getdents.go @@ -0,0 +1,149 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "fmt" + + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +// Getdents implements Linux syscall getdents(2). +func Getdents(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + return getdents(t, args, false /* isGetdents64 */) +} + +// Getdents64 implements Linux syscall getdents64(2). +func Getdents64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + return getdents(t, args, true /* isGetdents64 */) +} + +func getdents(t *kernel.Task, args arch.SyscallArguments, isGetdents64 bool) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + addr := args[1].Pointer() + size := int(args[2].Uint()) + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + cb := getGetdentsCallback(t, addr, size, isGetdents64) + err := file.IterDirents(t, cb) + n := size - cb.remaining + putGetdentsCallback(cb) + if n == 0 { + return 0, nil, err + } + return uintptr(n), nil, nil +} + +type getdentsCallback struct { + t *kernel.Task + addr usermem.Addr + remaining int + isGetdents64 bool +} + +var getdentsCallbackPool = sync.Pool{ + New: func() interface{} { + return &getdentsCallback{} + }, +} + +func getGetdentsCallback(t *kernel.Task, addr usermem.Addr, size int, isGetdents64 bool) *getdentsCallback { + cb := getdentsCallbackPool.Get().(*getdentsCallback) + *cb = getdentsCallback{ + t: t, + addr: addr, + remaining: size, + isGetdents64: isGetdents64, + } + return cb +} + +func putGetdentsCallback(cb *getdentsCallback) { + cb.t = nil + getdentsCallbackPool.Put(cb) +} + +// Handle implements vfs.IterDirentsCallback.Handle. +func (cb *getdentsCallback) Handle(dirent vfs.Dirent) error { + var buf []byte + if cb.isGetdents64 { + // struct linux_dirent64 { + // ino64_t d_ino; /* 64-bit inode number */ + // off64_t d_off; /* 64-bit offset to next structure */ + // unsigned short d_reclen; /* Size of this dirent */ + // unsigned char d_type; /* File type */ + // char d_name[]; /* Filename (null-terminated) */ + // }; + size := 8 + 8 + 2 + 1 + 1 + len(dirent.Name) + if size < cb.remaining { + return syserror.EINVAL + } + buf = cb.t.CopyScratchBuffer(size) + usermem.ByteOrder.PutUint64(buf[0:8], dirent.Ino) + usermem.ByteOrder.PutUint64(buf[8:16], uint64(dirent.NextOff)) + usermem.ByteOrder.PutUint16(buf[16:18], uint16(size)) + buf[18] = dirent.Type + copy(buf[19:], dirent.Name) + buf[size-1] = 0 // NUL terminator + } else { + // struct linux_dirent { + // unsigned long d_ino; /* Inode number */ + // unsigned long d_off; /* Offset to next linux_dirent */ + // unsigned short d_reclen; /* Length of this linux_dirent */ + // char d_name[]; /* Filename (null-terminated) */ + // /* length is actually (d_reclen - 2 - + // offsetof(struct linux_dirent, d_name)) */ + // /* + // char pad; // Zero padding byte + // char d_type; // File type (only since Linux + // // 2.6.4); offset is (d_reclen - 1) + // */ + // }; + if cb.t.Arch().Width() != 8 { + panic(fmt.Sprintf("unsupported sizeof(unsigned long): %d", cb.t.Arch().Width())) + } + size := 8 + 8 + 2 + 1 + 1 + 1 + len(dirent.Name) + if size < cb.remaining { + return syserror.EINVAL + } + buf = cb.t.CopyScratchBuffer(size) + usermem.ByteOrder.PutUint64(buf[0:8], dirent.Ino) + usermem.ByteOrder.PutUint64(buf[8:16], uint64(dirent.NextOff)) + usermem.ByteOrder.PutUint16(buf[16:18], uint16(size)) + copy(buf[18:], dirent.Name) + buf[size-3] = 0 // NUL terminator + buf[size-2] = 0 // zero padding byte + buf[size-1] = dirent.Type + } + n, err := cb.t.CopyOutBytes(cb.addr, buf) + if err != nil { + // Don't report partially-written dirents by advancing cb.addr or + // cb.remaining. + return err + } + cb.addr += usermem.Addr(n) + cb.remaining -= n + return nil +} diff --git a/pkg/sentry/syscalls/linux/vfs2/ioctl.go b/pkg/sentry/syscalls/linux/vfs2/ioctl.go new file mode 100644 index 000000000..5a2418da9 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/ioctl.go @@ -0,0 +1,35 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/syserror" +) + +// Ioctl implements Linux syscall ioctl(2). +func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + ret, err := file.Ioctl(t, t.MemoryManager(), args) + return ret, nil, err +} diff --git a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go index e0ac32b33..7d220bc20 100644 --- a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go +++ b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go @@ -12,6 +12,8 @@ // See the License for the specific language governing permissions and // limitations under the License. +// +build amd64 + package vfs2 import ( @@ -22,110 +24,142 @@ import ( // Override syscall table to add syscalls implementations from this package. func Override(table map[uintptr]kernel.Syscall) { table[0] = syscalls.Supported("read", Read) - - // Remove syscalls that haven't been converted yet. It's better to get ENOSYS - // rather than a SIGSEGV deep in the stack. - delete(table, 1) // write - delete(table, 2) // open - delete(table, 3) // close - delete(table, 4) // stat - delete(table, 5) // fstat - delete(table, 6) // lstat - delete(table, 7) // poll - delete(table, 8) // lseek - delete(table, 9) // mmap - delete(table, 16) // ioctl - delete(table, 17) // pread64 - delete(table, 18) // pwrite64 - delete(table, 19) // readv - delete(table, 20) // writev - delete(table, 21) // access - delete(table, 22) // pipe - delete(table, 32) // dup - delete(table, 33) // dup2 - delete(table, 40) // sendfile - delete(table, 59) // execve - delete(table, 72) // fcntl - delete(table, 73) // flock - delete(table, 74) // fsync - delete(table, 75) // fdatasync - delete(table, 76) // truncate - delete(table, 77) // ftruncate - delete(table, 78) // getdents - delete(table, 79) // getcwd - delete(table, 80) // chdir - delete(table, 81) // fchdir - delete(table, 82) // rename - delete(table, 83) // mkdir - delete(table, 84) // rmdir - delete(table, 85) // creat - delete(table, 86) // link - delete(table, 87) // unlink - delete(table, 88) // symlink - delete(table, 89) // readlink - delete(table, 90) // chmod - delete(table, 91) // fchmod - delete(table, 92) // chown - delete(table, 93) // fchown - delete(table, 94) // lchown - delete(table, 133) // mknod - delete(table, 137) // statfs - delete(table, 138) // fstatfs - delete(table, 161) // chroot - delete(table, 162) // sync + table[1] = syscalls.Supported("write", Write) + table[2] = syscalls.Supported("open", Open) + table[3] = syscalls.Supported("close", Close) + table[4] = syscalls.Supported("stat", Stat) + table[5] = syscalls.Supported("fstat", Fstat) + table[6] = syscalls.Supported("lstat", Lstat) + table[7] = syscalls.Supported("poll", Poll) + table[8] = syscalls.Supported("lseek", Lseek) + table[9] = syscalls.Supported("mmap", Mmap) + table[16] = syscalls.Supported("ioctl", Ioctl) + table[17] = syscalls.Supported("pread64", Pread64) + table[18] = syscalls.Supported("pwrite64", Pwrite64) + table[19] = syscalls.Supported("readv", Readv) + table[20] = syscalls.Supported("writev", Writev) + table[21] = syscalls.Supported("access", Access) + delete(table, 22) // pipe + table[23] = syscalls.Supported("select", Select) + table[32] = syscalls.Supported("dup", Dup) + table[33] = syscalls.Supported("dup2", Dup2) + delete(table, 40) // sendfile + delete(table, 41) // socket + delete(table, 42) // connect + delete(table, 43) // accept + delete(table, 44) // sendto + delete(table, 45) // recvfrom + delete(table, 46) // sendmsg + delete(table, 47) // recvmsg + delete(table, 48) // shutdown + delete(table, 49) // bind + delete(table, 50) // listen + delete(table, 51) // getsockname + delete(table, 52) // getpeername + delete(table, 53) // socketpair + delete(table, 54) // setsockopt + delete(table, 55) // getsockopt + table[59] = syscalls.Supported("execve", Execve) + table[72] = syscalls.Supported("fcntl", Fcntl) + delete(table, 73) // flock + table[74] = syscalls.Supported("fsync", Fsync) + table[75] = syscalls.Supported("fdatasync", Fdatasync) + table[76] = syscalls.Supported("truncate", Truncate) + table[77] = syscalls.Supported("ftruncate", Ftruncate) + table[78] = syscalls.Supported("getdents", Getdents) + table[79] = syscalls.Supported("getcwd", Getcwd) + table[80] = syscalls.Supported("chdir", Chdir) + table[81] = syscalls.Supported("fchdir", Fchdir) + table[82] = syscalls.Supported("rename", Rename) + table[83] = syscalls.Supported("mkdir", Mkdir) + table[84] = syscalls.Supported("rmdir", Rmdir) + table[85] = syscalls.Supported("creat", Creat) + table[86] = syscalls.Supported("link", Link) + table[87] = syscalls.Supported("unlink", Unlink) + table[88] = syscalls.Supported("symlink", Symlink) + table[89] = syscalls.Supported("readlink", Readlink) + table[90] = syscalls.Supported("chmod", Chmod) + table[91] = syscalls.Supported("fchmod", Fchmod) + table[92] = syscalls.Supported("chown", Chown) + table[93] = syscalls.Supported("fchown", Fchown) + table[94] = syscalls.Supported("lchown", Lchown) + table[132] = syscalls.Supported("utime", Utime) + table[133] = syscalls.Supported("mknod", Mknod) + table[137] = syscalls.Supported("statfs", Statfs) + table[138] = syscalls.Supported("fstatfs", Fstatfs) + table[161] = syscalls.Supported("chroot", Chroot) + table[162] = syscalls.Supported("sync", Sync) delete(table, 165) // mount delete(table, 166) // umount2 - delete(table, 172) // iopl - delete(table, 173) // ioperm delete(table, 187) // readahead - delete(table, 188) // setxattr - delete(table, 189) // lsetxattr - delete(table, 190) // fsetxattr - delete(table, 191) // getxattr - delete(table, 192) // lgetxattr - delete(table, 193) // fgetxattr + table[188] = syscalls.Supported("setxattr", Setxattr) + table[189] = syscalls.Supported("lsetxattr", Lsetxattr) + table[190] = syscalls.Supported("fsetxattr", Fsetxattr) + table[191] = syscalls.Supported("getxattr", Getxattr) + table[192] = syscalls.Supported("lgetxattr", Lgetxattr) + table[193] = syscalls.Supported("fgetxattr", Fgetxattr) + table[194] = syscalls.Supported("listxattr", Listxattr) + table[195] = syscalls.Supported("llistxattr", Llistxattr) + table[196] = syscalls.Supported("flistxattr", Flistxattr) + table[197] = syscalls.Supported("removexattr", Removexattr) + table[198] = syscalls.Supported("lremovexattr", Lremovexattr) + table[199] = syscalls.Supported("fremovexattr", Fremovexattr) delete(table, 206) // io_setup delete(table, 207) // io_destroy delete(table, 208) // io_getevents delete(table, 209) // io_submit delete(table, 210) // io_cancel - delete(table, 213) // epoll_create - delete(table, 214) // epoll_ctl_old - delete(table, 215) // epoll_wait_old - delete(table, 216) // remap_file_pages - delete(table, 217) // getdents64 - delete(table, 232) // epoll_wait - delete(table, 233) // epoll_ctl + table[213] = syscalls.Supported("epoll_create", EpollCreate) + table[217] = syscalls.Supported("getdents64", Getdents64) + delete(table, 221) // fdavise64 + table[232] = syscalls.Supported("epoll_wait", EpollWait) + table[233] = syscalls.Supported("epoll_ctl", EpollCtl) + table[235] = syscalls.Supported("utimes", Utimes) delete(table, 253) // inotify_init delete(table, 254) // inotify_add_watch delete(table, 255) // inotify_rm_watch - delete(table, 257) // openat - delete(table, 258) // mkdirat - delete(table, 259) // mknodat - delete(table, 260) // fchownat - delete(table, 261) // futimesat - delete(table, 262) // fstatat - delete(table, 263) // unlinkat - delete(table, 264) // renameat - delete(table, 265) // linkat - delete(table, 266) // symlinkat - delete(table, 267) // readlinkat - delete(table, 268) // fchmodat - delete(table, 269) // faccessat - delete(table, 270) // pselect - delete(table, 271) // ppoll + table[257] = syscalls.Supported("openat", Openat) + table[258] = syscalls.Supported("mkdirat", Mkdirat) + table[259] = syscalls.Supported("mknodat", Mknodat) + table[260] = syscalls.Supported("fchownat", Fchownat) + table[261] = syscalls.Supported("futimens", Futimens) + table[262] = syscalls.Supported("newfstatat", Newfstatat) + table[263] = syscalls.Supported("unlinkat", Unlinkat) + table[264] = syscalls.Supported("renameat", Renameat) + table[265] = syscalls.Supported("linkat", Linkat) + table[266] = syscalls.Supported("symlinkat", Symlinkat) + table[267] = syscalls.Supported("readlinkat", Readlinkat) + table[268] = syscalls.Supported("fchmodat", Fchmodat) + table[269] = syscalls.Supported("faccessat", Faccessat) + table[270] = syscalls.Supported("pselect", Pselect) + table[271] = syscalls.Supported("ppoll", Ppoll) + delete(table, 275) // splice + delete(table, 276) // tee + table[277] = syscalls.Supported("sync_file_range", SyncFileRange) + table[280] = syscalls.Supported("utimensat", Utimensat) + table[281] = syscalls.Supported("epoll_pwait", EpollPwait) + delete(table, 282) // signalfd + delete(table, 283) // timerfd_create + delete(table, 284) // eventfd delete(table, 285) // fallocate - delete(table, 291) // epoll_create1 - delete(table, 292) // dup3 + delete(table, 286) // timerfd_settime + delete(table, 287) // timerfd_gettime + delete(table, 288) // accept4 + delete(table, 289) // signalfd4 + delete(table, 290) // eventfd2 + table[291] = syscalls.Supported("epoll_create1", EpollCreate1) + table[292] = syscalls.Supported("dup3", Dup3) delete(table, 293) // pipe2 delete(table, 294) // inotify_init1 - delete(table, 295) // preadv - delete(table, 296) // pwritev - delete(table, 306) // syncfs - delete(table, 316) // renameat2 + table[295] = syscalls.Supported("preadv", Preadv) + table[296] = syscalls.Supported("pwritev", Pwritev) + delete(table, 299) // recvmmsg + table[306] = syscalls.Supported("syncfs", Syncfs) + delete(table, 307) // sendmmsg + table[316] = syscalls.Supported("renameat2", Renameat2) delete(table, 319) // memfd_create - delete(table, 322) // execveat - delete(table, 327) // preadv2 - delete(table, 328) // pwritev2 - delete(table, 332) // statx + table[322] = syscalls.Supported("execveat", Execveat) + table[327] = syscalls.Supported("preadv2", Preadv2) + table[328] = syscalls.Supported("pwritev2", Pwritev2) + table[332] = syscalls.Supported("statx", Statx) } diff --git a/pkg/sentry/syscalls/linux/vfs2/linux64_override_arm64.go b/pkg/sentry/syscalls/linux/vfs2/linux64_override_arm64.go index 6af5c400f..a6b367468 100644 --- a/pkg/sentry/syscalls/linux/vfs2/linux64_override_arm64.go +++ b/pkg/sentry/syscalls/linux/vfs2/linux64_override_arm64.go @@ -12,6 +12,8 @@ // See the License for the specific language governing permissions and // limitations under the License. +// +build arm64 + package vfs2 import ( diff --git a/pkg/sentry/syscalls/linux/vfs2/mmap.go b/pkg/sentry/syscalls/linux/vfs2/mmap.go new file mode 100644 index 000000000..60a43f0a0 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/mmap.go @@ -0,0 +1,92 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/memmap" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +// Mmap implements Linux syscall mmap(2). +func Mmap(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + prot := args[2].Int() + flags := args[3].Int() + fd := args[4].Int() + fixed := flags&linux.MAP_FIXED != 0 + private := flags&linux.MAP_PRIVATE != 0 + shared := flags&linux.MAP_SHARED != 0 + anon := flags&linux.MAP_ANONYMOUS != 0 + map32bit := flags&linux.MAP_32BIT != 0 + + // Require exactly one of MAP_PRIVATE and MAP_SHARED. + if private == shared { + return 0, nil, syserror.EINVAL + } + + opts := memmap.MMapOpts{ + Length: args[1].Uint64(), + Offset: args[5].Uint64(), + Addr: args[0].Pointer(), + Fixed: fixed, + Unmap: fixed, + Map32Bit: map32bit, + Private: private, + Perms: usermem.AccessType{ + Read: linux.PROT_READ&prot != 0, + Write: linux.PROT_WRITE&prot != 0, + Execute: linux.PROT_EXEC&prot != 0, + }, + MaxPerms: usermem.AnyAccess, + GrowsDown: linux.MAP_GROWSDOWN&flags != 0, + Precommit: linux.MAP_POPULATE&flags != 0, + } + if linux.MAP_LOCKED&flags != 0 { + opts.MLockMode = memmap.MLockEager + } + defer func() { + if opts.MappingIdentity != nil { + opts.MappingIdentity.DecRef() + } + }() + + if !anon { + // Convert the passed FD to a file reference. + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // mmap unconditionally requires that the FD is readable. + if !file.IsReadable() { + return 0, nil, syserror.EACCES + } + // MAP_SHARED requires that the FD be writable for PROT_WRITE. + if shared && !file.IsWritable() { + opts.MaxPerms.Write = false + } + + if err := file.ConfigureMMap(t, &opts); err != nil { + return 0, nil, err + } + } + + rv, err := t.MemoryManager().MMap(t, opts) + return uintptr(rv), nil, err +} diff --git a/pkg/sentry/syscalls/linux/vfs2/path.go b/pkg/sentry/syscalls/linux/vfs2/path.go new file mode 100644 index 000000000..97da6c647 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/path.go @@ -0,0 +1,94 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +func copyInPath(t *kernel.Task, addr usermem.Addr) (fspath.Path, error) { + pathname, err := t.CopyInString(addr, linux.PATH_MAX) + if err != nil { + return fspath.Path{}, err + } + return fspath.Parse(pathname), nil +} + +type taskPathOperation struct { + pop vfs.PathOperation + haveStartRef bool +} + +func getTaskPathOperation(t *kernel.Task, dirfd int32, path fspath.Path, shouldAllowEmptyPath shouldAllowEmptyPath, shouldFollowFinalSymlink shouldFollowFinalSymlink) (taskPathOperation, error) { + root := t.FSContext().RootDirectoryVFS2() + start := root + haveStartRef := false + if !path.Absolute { + if !path.HasComponents() && !bool(shouldAllowEmptyPath) { + root.DecRef() + return taskPathOperation{}, syserror.ENOENT + } + if dirfd == linux.AT_FDCWD { + start = t.FSContext().WorkingDirectoryVFS2() + haveStartRef = true + } else { + dirfile := t.GetFileVFS2(dirfd) + if dirfile == nil { + root.DecRef() + return taskPathOperation{}, syserror.EBADF + } + start = dirfile.VirtualDentry() + start.IncRef() + haveStartRef = true + dirfile.DecRef() + } + } + return taskPathOperation{ + pop: vfs.PathOperation{ + Root: root, + Start: start, + Path: path, + FollowFinalSymlink: bool(shouldFollowFinalSymlink), + }, + haveStartRef: haveStartRef, + }, nil +} + +func (tpop *taskPathOperation) Release() { + tpop.pop.Root.DecRef() + if tpop.haveStartRef { + tpop.pop.Start.DecRef() + tpop.haveStartRef = false + } +} + +type shouldAllowEmptyPath bool + +const ( + disallowEmptyPath shouldAllowEmptyPath = false + allowEmptyPath shouldAllowEmptyPath = true +) + +type shouldFollowFinalSymlink bool + +const ( + nofollowFinalSymlink shouldFollowFinalSymlink = false + followFinalSymlink shouldFollowFinalSymlink = true +) diff --git a/pkg/sentry/syscalls/linux/vfs2/poll.go b/pkg/sentry/syscalls/linux/vfs2/poll.go new file mode 100644 index 000000000..dbf4882da --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/poll.go @@ -0,0 +1,584 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "fmt" + "time" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" + "gvisor.dev/gvisor/pkg/sentry/limits" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +// fileCap is the maximum allowable files for poll & select. This has no +// equivalent in Linux; it exists in gVisor since allocation failure in Go is +// unrecoverable. +const fileCap = 1024 * 1024 + +// Masks for "readable", "writable", and "exceptional" events as defined by +// select(2). +const ( + // selectReadEvents is analogous to the Linux kernel's + // fs/select.c:POLLIN_SET. + selectReadEvents = linux.POLLIN | linux.POLLHUP | linux.POLLERR + + // selectWriteEvents is analogous to the Linux kernel's + // fs/select.c:POLLOUT_SET. + selectWriteEvents = linux.POLLOUT | linux.POLLERR + + // selectExceptEvents is analogous to the Linux kernel's + // fs/select.c:POLLEX_SET. + selectExceptEvents = linux.POLLPRI +) + +// pollState tracks the associated file description and waiter of a PollFD. +type pollState struct { + file *vfs.FileDescription + waiter waiter.Entry +} + +// initReadiness gets the current ready mask for the file represented by the FD +// stored in pfd.FD. If a channel is passed in, the waiter entry in "state" is +// used to register with the file for event notifications, and a reference to +// the file is stored in "state". +func initReadiness(t *kernel.Task, pfd *linux.PollFD, state *pollState, ch chan struct{}) { + if pfd.FD < 0 { + pfd.REvents = 0 + return + } + + file := t.GetFileVFS2(pfd.FD) + if file == nil { + pfd.REvents = linux.POLLNVAL + return + } + + if ch == nil { + defer file.DecRef() + } else { + state.file = file + state.waiter, _ = waiter.NewChannelEntry(ch) + file.EventRegister(&state.waiter, waiter.EventMaskFromLinux(uint32(pfd.Events))) + } + + r := file.Readiness(waiter.EventMaskFromLinux(uint32(pfd.Events))) + pfd.REvents = int16(r.ToLinux()) & pfd.Events +} + +// releaseState releases all the pollState in "state". +func releaseState(state []pollState) { + for i := range state { + if state[i].file != nil { + state[i].file.EventUnregister(&state[i].waiter) + state[i].file.DecRef() + } + } +} + +// pollBlock polls the PollFDs in "pfd" with a bounded time specified in "timeout" +// when "timeout" is greater than zero. +// +// pollBlock returns the remaining timeout, which is always 0 on a timeout; and 0 or +// positive if interrupted by a signal. +func pollBlock(t *kernel.Task, pfd []linux.PollFD, timeout time.Duration) (time.Duration, uintptr, error) { + var ch chan struct{} + if timeout != 0 { + ch = make(chan struct{}, 1) + } + + // Register for event notification in the files involved if we may + // block (timeout not zero). Once we find a file that has a non-zero + // result, we stop registering for events but still go through all files + // to get their ready masks. + state := make([]pollState, len(pfd)) + defer releaseState(state) + n := uintptr(0) + for i := range pfd { + initReadiness(t, &pfd[i], &state[i], ch) + if pfd[i].REvents != 0 { + n++ + ch = nil + } + } + + if timeout == 0 { + return timeout, n, nil + } + + haveTimeout := timeout >= 0 + + for n == 0 { + var err error + // Wait for a notification. + timeout, err = t.BlockWithTimeout(ch, haveTimeout, timeout) + if err != nil { + if err == syserror.ETIMEDOUT { + err = nil + } + return timeout, 0, err + } + + // We got notified, count how many files are ready. If none, + // then this was a spurious notification, and we just go back + // to sleep with the remaining timeout. + for i := range state { + if state[i].file == nil { + continue + } + + r := state[i].file.Readiness(waiter.EventMaskFromLinux(uint32(pfd[i].Events))) + rl := int16(r.ToLinux()) & pfd[i].Events + if rl != 0 { + pfd[i].REvents = rl + n++ + } + } + } + + return timeout, n, nil +} + +// copyInPollFDs copies an array of struct pollfd unless nfds exceeds the max. +func copyInPollFDs(t *kernel.Task, addr usermem.Addr, nfds uint) ([]linux.PollFD, error) { + if uint64(nfds) > t.ThreadGroup().Limits().GetCapped(limits.NumberOfFiles, fileCap) { + return nil, syserror.EINVAL + } + + pfd := make([]linux.PollFD, nfds) + if nfds > 0 { + if _, err := t.CopyIn(addr, &pfd); err != nil { + return nil, err + } + } + + return pfd, nil +} + +func doPoll(t *kernel.Task, addr usermem.Addr, nfds uint, timeout time.Duration) (time.Duration, uintptr, error) { + pfd, err := copyInPollFDs(t, addr, nfds) + if err != nil { + return timeout, 0, err + } + + // Compatibility warning: Linux adds POLLHUP and POLLERR just before + // polling, in fs/select.c:do_pollfd(). Since pfd is copied out after + // polling, changing event masks here is an application-visible difference. + // (Linux also doesn't copy out event masks at all, only revents.) + for i := range pfd { + pfd[i].Events |= linux.POLLHUP | linux.POLLERR + } + remainingTimeout, n, err := pollBlock(t, pfd, timeout) + err = syserror.ConvertIntr(err, syserror.EINTR) + + // The poll entries are copied out regardless of whether + // any are set or not. This aligns with the Linux behavior. + if nfds > 0 && err == nil { + if _, err := t.CopyOut(addr, pfd); err != nil { + return remainingTimeout, 0, err + } + } + + return remainingTimeout, n, err +} + +// CopyInFDSet copies an fd set from select(2)/pselect(2). +func CopyInFDSet(t *kernel.Task, addr usermem.Addr, nBytes, nBitsInLastPartialByte int) ([]byte, error) { + set := make([]byte, nBytes) + + if addr != 0 { + if _, err := t.CopyIn(addr, &set); err != nil { + return nil, err + } + // If we only use part of the last byte, mask out the extraneous bits. + // + // N.B. This only works on little-endian architectures. + if nBitsInLastPartialByte != 0 { + set[nBytes-1] &^= byte(0xff) << nBitsInLastPartialByte + } + } + return set, nil +} + +func doSelect(t *kernel.Task, nfds int, readFDs, writeFDs, exceptFDs usermem.Addr, timeout time.Duration) (uintptr, error) { + if nfds < 0 || nfds > fileCap { + return 0, syserror.EINVAL + } + + // Calculate the size of the fd sets (one bit per fd). + nBytes := (nfds + 7) / 8 + nBitsInLastPartialByte := nfds % 8 + + // Capture all the provided input vectors. + r, err := CopyInFDSet(t, readFDs, nBytes, nBitsInLastPartialByte) + if err != nil { + return 0, err + } + w, err := CopyInFDSet(t, writeFDs, nBytes, nBitsInLastPartialByte) + if err != nil { + return 0, err + } + e, err := CopyInFDSet(t, exceptFDs, nBytes, nBitsInLastPartialByte) + if err != nil { + return 0, err + } + + // Count how many FDs are actually being requested so that we can build + // a PollFD array. + fdCount := 0 + for i := 0; i < nBytes; i++ { + v := r[i] | w[i] | e[i] + for v != 0 { + v &= (v - 1) + fdCount++ + } + } + + // Build the PollFD array. + pfd := make([]linux.PollFD, 0, fdCount) + var fd int32 + for i := 0; i < nBytes; i++ { + rV, wV, eV := r[i], w[i], e[i] + v := rV | wV | eV + m := byte(1) + for j := 0; j < 8; j++ { + if (v & m) != 0 { + // Make sure the fd is valid and decrement the reference + // immediately to ensure we don't leak. Note, another thread + // might be about to close fd. This is racy, but that's + // OK. Linux is racy in the same way. + file := t.GetFileVFS2(fd) + if file == nil { + return 0, syserror.EBADF + } + file.DecRef() + + var mask int16 + if (rV & m) != 0 { + mask |= selectReadEvents + } + + if (wV & m) != 0 { + mask |= selectWriteEvents + } + + if (eV & m) != 0 { + mask |= selectExceptEvents + } + + pfd = append(pfd, linux.PollFD{ + FD: fd, + Events: mask, + }) + } + + fd++ + m <<= 1 + } + } + + // Do the syscall, then count the number of bits set. + if _, _, err = pollBlock(t, pfd, timeout); err != nil { + return 0, syserror.ConvertIntr(err, syserror.EINTR) + } + + // r, w, and e are currently event mask bitsets; unset bits corresponding + // to events that *didn't* occur. + bitSetCount := uintptr(0) + for idx := range pfd { + events := pfd[idx].REvents + i, j := pfd[idx].FD/8, uint(pfd[idx].FD%8) + m := byte(1) << j + if r[i]&m != 0 { + if (events & selectReadEvents) != 0 { + bitSetCount++ + } else { + r[i] &^= m + } + } + if w[i]&m != 0 { + if (events & selectWriteEvents) != 0 { + bitSetCount++ + } else { + w[i] &^= m + } + } + if e[i]&m != 0 { + if (events & selectExceptEvents) != 0 { + bitSetCount++ + } else { + e[i] &^= m + } + } + } + + // Copy updated vectors back. + if readFDs != 0 { + if _, err := t.CopyOut(readFDs, r); err != nil { + return 0, err + } + } + + if writeFDs != 0 { + if _, err := t.CopyOut(writeFDs, w); err != nil { + return 0, err + } + } + + if exceptFDs != 0 { + if _, err := t.CopyOut(exceptFDs, e); err != nil { + return 0, err + } + } + + return bitSetCount, nil +} + +// timeoutRemaining returns the amount of time remaining for the specified +// timeout or 0 if it has elapsed. +// +// startNs must be from CLOCK_MONOTONIC. +func timeoutRemaining(t *kernel.Task, startNs ktime.Time, timeout time.Duration) time.Duration { + now := t.Kernel().MonotonicClock().Now() + remaining := timeout - now.Sub(startNs) + if remaining < 0 { + remaining = 0 + } + return remaining +} + +// copyOutTimespecRemaining copies the time remaining in timeout to timespecAddr. +// +// startNs must be from CLOCK_MONOTONIC. +func copyOutTimespecRemaining(t *kernel.Task, startNs ktime.Time, timeout time.Duration, timespecAddr usermem.Addr) error { + if timeout <= 0 { + return nil + } + remaining := timeoutRemaining(t, startNs, timeout) + tsRemaining := linux.NsecToTimespec(remaining.Nanoseconds()) + return tsRemaining.CopyOut(t, timespecAddr) +} + +// copyOutTimevalRemaining copies the time remaining in timeout to timevalAddr. +// +// startNs must be from CLOCK_MONOTONIC. +func copyOutTimevalRemaining(t *kernel.Task, startNs ktime.Time, timeout time.Duration, timevalAddr usermem.Addr) error { + if timeout <= 0 { + return nil + } + remaining := timeoutRemaining(t, startNs, timeout) + tvRemaining := linux.NsecToTimeval(remaining.Nanoseconds()) + return tvRemaining.CopyOut(t, timevalAddr) +} + +// pollRestartBlock encapsulates the state required to restart poll(2) via +// restart_syscall(2). +// +// +stateify savable +type pollRestartBlock struct { + pfdAddr usermem.Addr + nfds uint + timeout time.Duration +} + +// Restart implements kernel.SyscallRestartBlock.Restart. +func (p *pollRestartBlock) Restart(t *kernel.Task) (uintptr, error) { + return poll(t, p.pfdAddr, p.nfds, p.timeout) +} + +func poll(t *kernel.Task, pfdAddr usermem.Addr, nfds uint, timeout time.Duration) (uintptr, error) { + remainingTimeout, n, err := doPoll(t, pfdAddr, nfds, timeout) + // On an interrupt poll(2) is restarted with the remaining timeout. + if err == syserror.EINTR { + t.SetSyscallRestartBlock(&pollRestartBlock{ + pfdAddr: pfdAddr, + nfds: nfds, + timeout: remainingTimeout, + }) + return 0, kernel.ERESTART_RESTARTBLOCK + } + return n, err +} + +// Poll implements linux syscall poll(2). +func Poll(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + pfdAddr := args[0].Pointer() + nfds := uint(args[1].Uint()) // poll(2) uses unsigned long. + timeout := time.Duration(args[2].Int()) * time.Millisecond + n, err := poll(t, pfdAddr, nfds, timeout) + return n, nil, err +} + +// Ppoll implements linux syscall ppoll(2). +func Ppoll(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + pfdAddr := args[0].Pointer() + nfds := uint(args[1].Uint()) // poll(2) uses unsigned long. + timespecAddr := args[2].Pointer() + maskAddr := args[3].Pointer() + maskSize := uint(args[4].Uint()) + + timeout, err := copyTimespecInToDuration(t, timespecAddr) + if err != nil { + return 0, nil, err + } + + var startNs ktime.Time + if timeout > 0 { + startNs = t.Kernel().MonotonicClock().Now() + } + + if err := setTempSignalSet(t, maskAddr, maskSize); err != nil { + return 0, nil, err + } + + _, n, err := doPoll(t, pfdAddr, nfds, timeout) + copyErr := copyOutTimespecRemaining(t, startNs, timeout, timespecAddr) + // doPoll returns EINTR if interrupted, but ppoll is normally restartable + // if interrupted by something other than a signal handled by the + // application (i.e. returns ERESTARTNOHAND). However, if + // copyOutTimespecRemaining failed, then the restarted ppoll would use the + // wrong timeout, so the error should be left as EINTR. + // + // Note that this means that if err is nil but copyErr is not, copyErr is + // ignored. This is consistent with Linux. + if err == syserror.EINTR && copyErr == nil { + err = kernel.ERESTARTNOHAND + } + return n, nil, err +} + +// Select implements linux syscall select(2). +func Select(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + nfds := int(args[0].Int()) // select(2) uses an int. + readFDs := args[1].Pointer() + writeFDs := args[2].Pointer() + exceptFDs := args[3].Pointer() + timevalAddr := args[4].Pointer() + + // Use a negative Duration to indicate "no timeout". + timeout := time.Duration(-1) + if timevalAddr != 0 { + var timeval linux.Timeval + if err := timeval.CopyIn(t, timevalAddr); err != nil { + return 0, nil, err + } + if timeval.Sec < 0 || timeval.Usec < 0 { + return 0, nil, syserror.EINVAL + } + timeout = time.Duration(timeval.ToNsecCapped()) + } + startNs := t.Kernel().MonotonicClock().Now() + n, err := doSelect(t, nfds, readFDs, writeFDs, exceptFDs, timeout) + copyErr := copyOutTimevalRemaining(t, startNs, timeout, timevalAddr) + // See comment in Ppoll. + if err == syserror.EINTR && copyErr == nil { + err = kernel.ERESTARTNOHAND + } + return n, nil, err +} + +// Pselect implements linux syscall pselect(2). +func Pselect(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + nfds := int(args[0].Int()) // select(2) uses an int. + readFDs := args[1].Pointer() + writeFDs := args[2].Pointer() + exceptFDs := args[3].Pointer() + timespecAddr := args[4].Pointer() + maskWithSizeAddr := args[5].Pointer() + + timeout, err := copyTimespecInToDuration(t, timespecAddr) + if err != nil { + return 0, nil, err + } + + var startNs ktime.Time + if timeout > 0 { + startNs = t.Kernel().MonotonicClock().Now() + } + + if maskWithSizeAddr != 0 { + if t.Arch().Width() != 8 { + panic(fmt.Sprintf("unsupported sizeof(void*): %d", t.Arch().Width())) + } + var maskStruct sigSetWithSize + if err := maskStruct.CopyIn(t, maskWithSizeAddr); err != nil { + return 0, nil, err + } + if err := setTempSignalSet(t, usermem.Addr(maskStruct.sigsetAddr), uint(maskStruct.sizeofSigset)); err != nil { + return 0, nil, err + } + } + + n, err := doSelect(t, nfds, readFDs, writeFDs, exceptFDs, timeout) + copyErr := copyOutTimespecRemaining(t, startNs, timeout, timespecAddr) + // See comment in Ppoll. + if err == syserror.EINTR && copyErr == nil { + err = kernel.ERESTARTNOHAND + } + return n, nil, err +} + +// +marshal +type sigSetWithSize struct { + sigsetAddr uint64 + sizeofSigset uint64 +} + +// copyTimespecInToDuration copies a Timespec from the untrusted app range, +// validates it and converts it to a Duration. +// +// If the Timespec is larger than what can be represented in a Duration, the +// returned value is the maximum that Duration will allow. +// +// If timespecAddr is NULL, the returned value is negative. +func copyTimespecInToDuration(t *kernel.Task, timespecAddr usermem.Addr) (time.Duration, error) { + // Use a negative Duration to indicate "no timeout". + timeout := time.Duration(-1) + if timespecAddr != 0 { + var timespec linux.Timespec + if err := timespec.CopyIn(t, timespecAddr); err != nil { + return 0, err + } + if !timespec.Valid() { + return 0, syserror.EINVAL + } + timeout = time.Duration(timespec.ToNsecCapped()) + } + return timeout, nil +} + +func setTempSignalSet(t *kernel.Task, maskAddr usermem.Addr, maskSize uint) error { + if maskAddr == 0 { + return nil + } + if maskSize != linux.SignalSetSize { + return syserror.EINVAL + } + var mask linux.SignalSet + if err := mask.CopyIn(t, maskAddr); err != nil { + return err + } + mask &^= kernel.UnblockableSignals + oldmask := t.SignalMask() + t.SetSignalMask(mask) + t.SetSavedSignalMask(oldmask) + return nil +} diff --git a/pkg/sentry/syscalls/linux/vfs2/read_write.go b/pkg/sentry/syscalls/linux/vfs2/read_write.go new file mode 100644 index 000000000..35f6308d6 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/read_write.go @@ -0,0 +1,511 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +const ( + eventMaskRead = waiter.EventIn | waiter.EventHUp | waiter.EventErr + eventMaskWrite = waiter.EventOut | waiter.EventHUp | waiter.EventErr +) + +// Read implements Linux syscall read(2). +func Read(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + addr := args[1].Pointer() + size := args[2].SizeT() + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Check that the size is legitimate. + si := int(size) + if si < 0 { + return 0, nil, syserror.EINVAL + } + + // Get the destination of the read. + dst, err := t.SingleIOSequence(addr, si, usermem.IOOpts{ + AddressSpaceActive: true, + }) + if err != nil { + return 0, nil, err + } + + n, err := read(t, file, dst, vfs.ReadOptions{}) + t.IOUsage().AccountReadSyscall(n) + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, kernel.ERESTARTSYS, "read", file) +} + +// Readv implements Linux syscall readv(2). +func Readv(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + addr := args[1].Pointer() + iovcnt := int(args[2].Int()) + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Get the destination of the read. + dst, err := t.IovecsIOSequence(addr, iovcnt, usermem.IOOpts{ + AddressSpaceActive: true, + }) + if err != nil { + return 0, nil, err + } + + n, err := read(t, file, dst, vfs.ReadOptions{}) + t.IOUsage().AccountReadSyscall(n) + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, kernel.ERESTARTSYS, "readv", file) +} + +func read(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + n, err := file.Read(t, dst, opts) + if err != syserror.ErrWouldBlock || file.StatusFlags()&linux.O_NONBLOCK != 0 { + return n, err + } + + // Register for notifications. + w, ch := waiter.NewChannelEntry(nil) + file.EventRegister(&w, eventMaskRead) + + total := n + for { + // Shorten dst to reflect bytes previously read. + dst = dst.DropFirst(int(n)) + + // Issue the request and break out if it completes with anything other than + // "would block". + n, err := file.Read(t, dst, opts) + total += n + if err != syserror.ErrWouldBlock { + break + } + if err := t.Block(ch); err != nil { + break + } + } + file.EventUnregister(&w) + + return total, err +} + +// Pread64 implements Linux syscall pread64(2). +func Pread64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + addr := args[1].Pointer() + size := args[2].SizeT() + offset := args[3].Int64() + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Check that the offset is legitimate. + if offset < 0 { + return 0, nil, syserror.EINVAL + } + + // Check that the size is legitimate. + si := int(size) + if si < 0 { + return 0, nil, syserror.EINVAL + } + + // Get the destination of the read. + dst, err := t.SingleIOSequence(addr, si, usermem.IOOpts{ + AddressSpaceActive: true, + }) + if err != nil { + return 0, nil, err + } + + n, err := pread(t, file, dst, offset, vfs.ReadOptions{}) + t.IOUsage().AccountReadSyscall(n) + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, kernel.ERESTARTSYS, "pread64", file) +} + +// Preadv implements Linux syscall preadv(2). +func Preadv(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + addr := args[1].Pointer() + iovcnt := int(args[2].Int()) + offset := args[3].Int64() + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Check that the offset is legitimate. + if offset < 0 { + return 0, nil, syserror.EINVAL + } + + // Get the destination of the read. + dst, err := t.IovecsIOSequence(addr, iovcnt, usermem.IOOpts{ + AddressSpaceActive: true, + }) + if err != nil { + return 0, nil, err + } + + n, err := pread(t, file, dst, offset, vfs.ReadOptions{}) + t.IOUsage().AccountReadSyscall(n) + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, kernel.ERESTARTSYS, "preadv", file) +} + +// Preadv2 implements Linux syscall preadv2(2). +func Preadv2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + // While the glibc signature is + // preadv2(int fd, struct iovec* iov, int iov_cnt, off_t offset, int flags) + // the actual syscall + // (https://elixir.bootlin.com/linux/v5.5/source/fs/read_write.c#L1142) + // splits the offset argument into a high/low value for compatibility with + // 32-bit architectures. The flags argument is the 6th argument (index 5). + fd := args[0].Int() + addr := args[1].Pointer() + iovcnt := int(args[2].Int()) + offset := args[3].Int64() + flags := args[5].Int() + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Check that the offset is legitimate. + if offset < -1 { + return 0, nil, syserror.EINVAL + } + + // Get the destination of the read. + dst, err := t.IovecsIOSequence(addr, iovcnt, usermem.IOOpts{ + AddressSpaceActive: true, + }) + if err != nil { + return 0, nil, err + } + + opts := vfs.ReadOptions{ + Flags: uint32(flags), + } + var n int64 + if offset == -1 { + n, err = read(t, file, dst, opts) + } else { + n, err = pread(t, file, dst, offset, opts) + } + t.IOUsage().AccountReadSyscall(n) + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, kernel.ERESTARTSYS, "preadv2", file) +} + +func pread(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + n, err := file.PRead(t, dst, offset, opts) + if err != syserror.ErrWouldBlock || file.StatusFlags()&linux.O_NONBLOCK != 0 { + return n, err + } + + // Register for notifications. + w, ch := waiter.NewChannelEntry(nil) + file.EventRegister(&w, eventMaskRead) + + total := n + for { + // Shorten dst to reflect bytes previously read. + dst = dst.DropFirst(int(n)) + + // Issue the request and break out if it completes with anything other than + // "would block". + n, err := file.PRead(t, dst, offset+total, opts) + total += n + if err != syserror.ErrWouldBlock { + break + } + if err := t.Block(ch); err != nil { + break + } + } + file.EventUnregister(&w) + + return total, err +} + +// Write implements Linux syscall write(2). +func Write(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + addr := args[1].Pointer() + size := args[2].SizeT() + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Check that the size is legitimate. + si := int(size) + if si < 0 { + return 0, nil, syserror.EINVAL + } + + // Get the source of the write. + src, err := t.SingleIOSequence(addr, si, usermem.IOOpts{ + AddressSpaceActive: true, + }) + if err != nil { + return 0, nil, err + } + + n, err := write(t, file, src, vfs.WriteOptions{}) + t.IOUsage().AccountWriteSyscall(n) + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, kernel.ERESTARTSYS, "write", file) +} + +// Writev implements Linux syscall writev(2). +func Writev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + addr := args[1].Pointer() + iovcnt := int(args[2].Int()) + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Get the source of the write. + src, err := t.IovecsIOSequence(addr, iovcnt, usermem.IOOpts{ + AddressSpaceActive: true, + }) + if err != nil { + return 0, nil, err + } + + n, err := write(t, file, src, vfs.WriteOptions{}) + t.IOUsage().AccountWriteSyscall(n) + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, kernel.ERESTARTSYS, "writev", file) +} + +func write(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + n, err := file.Write(t, src, opts) + if err != syserror.ErrWouldBlock || file.StatusFlags()&linux.O_NONBLOCK != 0 { + return n, err + } + + // Register for notifications. + w, ch := waiter.NewChannelEntry(nil) + file.EventRegister(&w, eventMaskWrite) + + total := n + for { + // Shorten src to reflect bytes previously written. + src = src.DropFirst(int(n)) + + // Issue the request and break out if it completes with anything other than + // "would block". + n, err := file.Write(t, src, opts) + total += n + if err != syserror.ErrWouldBlock { + break + } + if err := t.Block(ch); err != nil { + break + } + } + file.EventUnregister(&w) + + return total, err +} + +// Pwrite64 implements Linux syscall pwrite64(2). +func Pwrite64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + addr := args[1].Pointer() + size := args[2].SizeT() + offset := args[3].Int64() + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Check that the offset is legitimate. + if offset < 0 { + return 0, nil, syserror.EINVAL + } + + // Check that the size is legitimate. + si := int(size) + if si < 0 { + return 0, nil, syserror.EINVAL + } + + // Get the source of the write. + src, err := t.SingleIOSequence(addr, si, usermem.IOOpts{ + AddressSpaceActive: true, + }) + if err != nil { + return 0, nil, err + } + + n, err := pwrite(t, file, src, offset, vfs.WriteOptions{}) + t.IOUsage().AccountWriteSyscall(n) + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, kernel.ERESTARTSYS, "pwrite64", file) +} + +// Pwritev implements Linux syscall pwritev(2). +func Pwritev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + addr := args[1].Pointer() + iovcnt := int(args[2].Int()) + offset := args[3].Int64() + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Check that the offset is legitimate. + if offset < 0 { + return 0, nil, syserror.EINVAL + } + + // Get the source of the write. + src, err := t.IovecsIOSequence(addr, iovcnt, usermem.IOOpts{ + AddressSpaceActive: true, + }) + if err != nil { + return 0, nil, err + } + + n, err := pwrite(t, file, src, offset, vfs.WriteOptions{}) + t.IOUsage().AccountReadSyscall(n) + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, kernel.ERESTARTSYS, "pwritev", file) +} + +// Pwritev2 implements Linux syscall pwritev2(2). +func Pwritev2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + // While the glibc signature is + // pwritev2(int fd, struct iovec* iov, int iov_cnt, off_t offset, int flags) + // the actual syscall + // (https://elixir.bootlin.com/linux/v5.5/source/fs/read_write.c#L1162) + // splits the offset argument into a high/low value for compatibility with + // 32-bit architectures. The flags argument is the 6th argument (index 5). + fd := args[0].Int() + addr := args[1].Pointer() + iovcnt := int(args[2].Int()) + offset := args[3].Int64() + flags := args[5].Int() + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Check that the offset is legitimate. + if offset < -1 { + return 0, nil, syserror.EINVAL + } + + // Get the source of the write. + src, err := t.IovecsIOSequence(addr, iovcnt, usermem.IOOpts{ + AddressSpaceActive: true, + }) + if err != nil { + return 0, nil, err + } + + opts := vfs.WriteOptions{ + Flags: uint32(flags), + } + var n int64 + if offset == -1 { + n, err = write(t, file, src, opts) + } else { + n, err = pwrite(t, file, src, offset, opts) + } + t.IOUsage().AccountWriteSyscall(n) + return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, kernel.ERESTARTSYS, "pwritev2", file) +} + +func pwrite(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + n, err := file.PWrite(t, src, offset, opts) + if err != syserror.ErrWouldBlock || file.StatusFlags()&linux.O_NONBLOCK != 0 { + return n, err + } + + // Register for notifications. + w, ch := waiter.NewChannelEntry(nil) + file.EventRegister(&w, eventMaskWrite) + + total := n + for { + // Shorten src to reflect bytes previously written. + src = src.DropFirst(int(n)) + + // Issue the request and break out if it completes with anything other than + // "would block". + n, err := file.PWrite(t, src, offset+total, opts) + total += n + if err != syserror.ErrWouldBlock { + break + } + if err := t.Block(ch); err != nil { + break + } + } + file.EventUnregister(&w) + + return total, err +} + +// Lseek implements Linux syscall lseek(2). +func Lseek(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + offset := args[1].Int64() + whence := args[2].Int() + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + newoff, err := file.Seek(t, offset, whence) + return uintptr(newoff), nil, err +} diff --git a/pkg/sentry/syscalls/linux/vfs2/setstat.go b/pkg/sentry/syscalls/linux/vfs2/setstat.go new file mode 100644 index 000000000..9250659ff --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/setstat.go @@ -0,0 +1,380 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +const chmodMask = 0777 | linux.S_ISUID | linux.S_ISGID | linux.S_ISVTX + +// Chmod implements Linux syscall chmod(2). +func Chmod(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + pathAddr := args[0].Pointer() + mode := args[1].ModeT() + return 0, nil, fchmodat(t, linux.AT_FDCWD, pathAddr, mode) +} + +// Fchmodat implements Linux syscall fchmodat(2). +func Fchmodat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + dirfd := args[0].Int() + pathAddr := args[1].Pointer() + mode := args[2].ModeT() + return 0, nil, fchmodat(t, dirfd, pathAddr, mode) +} + +func fchmodat(t *kernel.Task, dirfd int32, pathAddr usermem.Addr, mode uint) error { + path, err := copyInPath(t, pathAddr) + if err != nil { + return err + } + + return setstatat(t, dirfd, path, disallowEmptyPath, followFinalSymlink, &vfs.SetStatOptions{ + Stat: linux.Statx{ + Mask: linux.STATX_MODE, + Mode: uint16(mode & chmodMask), + }, + }) +} + +// Fchmod implements Linux syscall fchmod(2). +func Fchmod(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + mode := args[1].ModeT() + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + return 0, nil, file.SetStat(t, vfs.SetStatOptions{ + Stat: linux.Statx{ + Mask: linux.STATX_MODE, + Mode: uint16(mode & chmodMask), + }, + }) +} + +// Chown implements Linux syscall chown(2). +func Chown(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + pathAddr := args[0].Pointer() + owner := args[1].Int() + group := args[2].Int() + return 0, nil, fchownat(t, linux.AT_FDCWD, pathAddr, owner, group, 0 /* flags */) +} + +// Lchown implements Linux syscall lchown(2). +func Lchown(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + pathAddr := args[0].Pointer() + owner := args[1].Int() + group := args[2].Int() + return 0, nil, fchownat(t, linux.AT_FDCWD, pathAddr, owner, group, linux.AT_SYMLINK_NOFOLLOW) +} + +// Fchownat implements Linux syscall fchownat(2). +func Fchownat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + dirfd := args[0].Int() + pathAddr := args[1].Pointer() + owner := args[2].Int() + group := args[3].Int() + flags := args[4].Int() + return 0, nil, fchownat(t, dirfd, pathAddr, owner, group, flags) +} + +func fchownat(t *kernel.Task, dirfd int32, pathAddr usermem.Addr, owner, group, flags int32) error { + if flags&^(linux.AT_EMPTY_PATH|linux.AT_SYMLINK_NOFOLLOW) != 0 { + return syserror.EINVAL + } + + path, err := copyInPath(t, pathAddr) + if err != nil { + return err + } + + var opts vfs.SetStatOptions + if err := populateSetStatOptionsForChown(t, owner, group, &opts); err != nil { + return err + } + + return setstatat(t, dirfd, path, shouldAllowEmptyPath(flags&linux.AT_EMPTY_PATH != 0), shouldFollowFinalSymlink(flags&linux.AT_SYMLINK_NOFOLLOW == 0), &opts) +} + +func populateSetStatOptionsForChown(t *kernel.Task, owner, group int32, opts *vfs.SetStatOptions) error { + userns := t.UserNamespace() + if owner != -1 { + kuid := userns.MapToKUID(auth.UID(owner)) + if !kuid.Ok() { + return syserror.EINVAL + } + opts.Stat.Mask |= linux.STATX_UID + opts.Stat.UID = uint32(kuid) + } + if group != -1 { + kgid := userns.MapToKGID(auth.GID(group)) + if !kgid.Ok() { + return syserror.EINVAL + } + opts.Stat.Mask |= linux.STATX_GID + opts.Stat.GID = uint32(kgid) + } + return nil +} + +// Fchown implements Linux syscall fchown(2). +func Fchown(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + owner := args[1].Int() + group := args[2].Int() + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + var opts vfs.SetStatOptions + if err := populateSetStatOptionsForChown(t, owner, group, &opts); err != nil { + return 0, nil, err + } + return 0, nil, file.SetStat(t, opts) +} + +// Truncate implements Linux syscall truncate(2). +func Truncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + addr := args[0].Pointer() + length := args[1].Int64() + + if length < 0 { + return 0, nil, syserror.EINVAL + } + + path, err := copyInPath(t, addr) + if err != nil { + return 0, nil, err + } + + return 0, nil, setstatat(t, linux.AT_FDCWD, path, disallowEmptyPath, followFinalSymlink, &vfs.SetStatOptions{ + Stat: linux.Statx{ + Mask: linux.STATX_SIZE, + Size: uint64(length), + }, + }) +} + +// Ftruncate implements Linux syscall ftruncate(2). +func Ftruncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + length := args[1].Int64() + + if length < 0 { + return 0, nil, syserror.EINVAL + } + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + return 0, nil, file.SetStat(t, vfs.SetStatOptions{ + Stat: linux.Statx{ + Mask: linux.STATX_SIZE, + Size: uint64(length), + }, + }) +} + +// Utime implements Linux syscall utime(2). +func Utime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + pathAddr := args[0].Pointer() + timesAddr := args[1].Pointer() + + path, err := copyInPath(t, pathAddr) + if err != nil { + return 0, nil, err + } + + opts := vfs.SetStatOptions{ + Stat: linux.Statx{ + Mask: linux.STATX_ATIME | linux.STATX_MTIME, + }, + } + if timesAddr == 0 { + opts.Stat.Atime.Nsec = linux.UTIME_NOW + opts.Stat.Mtime.Nsec = linux.UTIME_NOW + } else { + var times linux.Utime + if err := times.CopyIn(t, timesAddr); err != nil { + return 0, nil, err + } + opts.Stat.Atime.Sec = times.Actime + opts.Stat.Mtime.Sec = times.Modtime + } + + return 0, nil, setstatat(t, linux.AT_FDCWD, path, disallowEmptyPath, followFinalSymlink, &opts) +} + +// Utimes implements Linux syscall utimes(2). +func Utimes(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + pathAddr := args[0].Pointer() + timesAddr := args[1].Pointer() + + path, err := copyInPath(t, pathAddr) + if err != nil { + return 0, nil, err + } + + opts := vfs.SetStatOptions{ + Stat: linux.Statx{ + Mask: linux.STATX_ATIME | linux.STATX_MTIME, + }, + } + if timesAddr == 0 { + opts.Stat.Atime.Nsec = linux.UTIME_NOW + opts.Stat.Mtime.Nsec = linux.UTIME_NOW + } else { + var times [2]linux.Timeval + if _, err := t.CopyIn(timesAddr, ×); err != nil { + return 0, nil, err + } + opts.Stat.Atime = linux.StatxTimestamp{ + Sec: times[0].Sec, + Nsec: uint32(times[0].Usec * 1000), + } + opts.Stat.Mtime = linux.StatxTimestamp{ + Sec: times[1].Sec, + Nsec: uint32(times[1].Usec * 1000), + } + } + + return 0, nil, setstatat(t, linux.AT_FDCWD, path, disallowEmptyPath, followFinalSymlink, &opts) +} + +// Utimensat implements Linux syscall utimensat(2). +func Utimensat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + dirfd := args[0].Int() + pathAddr := args[1].Pointer() + timesAddr := args[2].Pointer() + flags := args[3].Int() + + if flags&^linux.AT_SYMLINK_NOFOLLOW != 0 { + return 0, nil, syserror.EINVAL + } + + path, err := copyInPath(t, pathAddr) + if err != nil { + return 0, nil, err + } + + var opts vfs.SetStatOptions + if err := populateSetStatOptionsForUtimens(t, timesAddr, &opts); err != nil { + return 0, nil, err + } + + return 0, nil, setstatat(t, dirfd, path, disallowEmptyPath, followFinalSymlink, &opts) +} + +// Futimens implements Linux syscall futimens(2). +func Futimens(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + timesAddr := args[1].Pointer() + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + var opts vfs.SetStatOptions + if err := populateSetStatOptionsForUtimens(t, timesAddr, &opts); err != nil { + return 0, nil, err + } + + return 0, nil, file.SetStat(t, opts) +} + +func populateSetStatOptionsForUtimens(t *kernel.Task, timesAddr usermem.Addr, opts *vfs.SetStatOptions) error { + if timesAddr == 0 { + opts.Stat.Mask = linux.STATX_ATIME | linux.STATX_MTIME + opts.Stat.Atime.Nsec = linux.UTIME_NOW + opts.Stat.Mtime.Nsec = linux.UTIME_NOW + return nil + } + var times [2]linux.Timespec + if _, err := t.CopyIn(timesAddr, ×); err != nil { + return err + } + if times[0].Nsec != linux.UTIME_OMIT { + opts.Stat.Mask |= linux.STATX_ATIME + opts.Stat.Atime = linux.StatxTimestamp{ + Sec: times[0].Sec, + Nsec: uint32(times[0].Nsec), + } + } + if times[1].Nsec != linux.UTIME_OMIT { + opts.Stat.Mask |= linux.STATX_MTIME + opts.Stat.Mtime = linux.StatxTimestamp{ + Sec: times[1].Sec, + Nsec: uint32(times[1].Nsec), + } + } + return nil +} + +func setstatat(t *kernel.Task, dirfd int32, path fspath.Path, shouldAllowEmptyPath shouldAllowEmptyPath, shouldFollowFinalSymlink shouldFollowFinalSymlink, opts *vfs.SetStatOptions) error { + root := t.FSContext().RootDirectoryVFS2() + defer root.DecRef() + start := root + if !path.Absolute { + if !path.HasComponents() && !bool(shouldAllowEmptyPath) { + return syserror.ENOENT + } + if dirfd == linux.AT_FDCWD { + start = t.FSContext().WorkingDirectoryVFS2() + defer start.DecRef() + } else { + dirfile := t.GetFileVFS2(dirfd) + if dirfile == nil { + return syserror.EBADF + } + if !path.HasComponents() { + // Use FileDescription.SetStat() instead of + // VirtualFilesystem.SetStatAt(), since the former may be able + // to use opened file state to expedite the SetStat. + err := dirfile.SetStat(t, *opts) + dirfile.DecRef() + return err + } + start = dirfile.VirtualDentry() + start.IncRef() + defer start.DecRef() + dirfile.DecRef() + } + } + return t.Kernel().VFS().SetStatAt(t, t.Credentials(), &vfs.PathOperation{ + Root: root, + Start: start, + Path: path, + FollowFinalSymlink: bool(shouldFollowFinalSymlink), + }, opts) +} diff --git a/pkg/sentry/syscalls/linux/vfs2/stat.go b/pkg/sentry/syscalls/linux/vfs2/stat.go new file mode 100644 index 000000000..dca8d7011 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/stat.go @@ -0,0 +1,346 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/gohacks" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +// Stat implements Linux syscall stat(2). +func Stat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + pathAddr := args[0].Pointer() + statAddr := args[1].Pointer() + return 0, nil, fstatat(t, linux.AT_FDCWD, pathAddr, statAddr, 0 /* flags */) +} + +// Lstat implements Linux syscall lstat(2). +func Lstat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + pathAddr := args[0].Pointer() + statAddr := args[1].Pointer() + return 0, nil, fstatat(t, linux.AT_FDCWD, pathAddr, statAddr, linux.AT_SYMLINK_NOFOLLOW) +} + +// Newfstatat implements Linux syscall newfstatat, which backs fstatat(2). +func Newfstatat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + dirfd := args[0].Int() + pathAddr := args[1].Pointer() + statAddr := args[2].Pointer() + flags := args[3].Int() + return 0, nil, fstatat(t, dirfd, pathAddr, statAddr, flags) +} + +func fstatat(t *kernel.Task, dirfd int32, pathAddr, statAddr usermem.Addr, flags int32) error { + if flags&^(linux.AT_EMPTY_PATH|linux.AT_SYMLINK_NOFOLLOW) != 0 { + return syserror.EINVAL + } + + opts := vfs.StatOptions{ + Mask: linux.STATX_BASIC_STATS, + } + + path, err := copyInPath(t, pathAddr) + if err != nil { + return err + } + + root := t.FSContext().RootDirectoryVFS2() + defer root.DecRef() + start := root + if !path.Absolute { + if !path.HasComponents() && flags&linux.AT_EMPTY_PATH == 0 { + return syserror.ENOENT + } + if dirfd == linux.AT_FDCWD { + start = t.FSContext().WorkingDirectoryVFS2() + defer start.DecRef() + } else { + dirfile := t.GetFileVFS2(dirfd) + if dirfile == nil { + return syserror.EBADF + } + if !path.HasComponents() { + // Use FileDescription.Stat() instead of + // VirtualFilesystem.StatAt() for fstatat(fd, ""), since the + // former may be able to use opened file state to expedite the + // Stat. + statx, err := dirfile.Stat(t, opts) + dirfile.DecRef() + if err != nil { + return err + } + var stat linux.Stat + convertStatxToUserStat(t, &statx, &stat) + return stat.CopyOut(t, statAddr) + } + start = dirfile.VirtualDentry() + start.IncRef() + defer start.DecRef() + dirfile.DecRef() + } + } + + statx, err := t.Kernel().VFS().StatAt(t, t.Credentials(), &vfs.PathOperation{ + Root: root, + Start: start, + Path: path, + FollowFinalSymlink: flags&linux.AT_SYMLINK_NOFOLLOW == 0, + }, &opts) + if err != nil { + return err + } + var stat linux.Stat + convertStatxToUserStat(t, &statx, &stat) + return stat.CopyOut(t, statAddr) +} + +// This takes both input and output as pointer arguments to avoid copying large +// structs. +func convertStatxToUserStat(t *kernel.Task, statx *linux.Statx, stat *linux.Stat) { + // Linux just copies fields from struct kstat without regard to struct + // kstat::result_mask (fs/stat.c:cp_new_stat()), so we do too. + userns := t.UserNamespace() + *stat = linux.Stat{ + Dev: uint64(linux.MakeDeviceID(uint16(statx.DevMajor), statx.DevMinor)), + Ino: statx.Ino, + Nlink: uint64(statx.Nlink), + Mode: uint32(statx.Mode), + UID: uint32(auth.KUID(statx.UID).In(userns).OrOverflow()), + GID: uint32(auth.KGID(statx.GID).In(userns).OrOverflow()), + Rdev: uint64(linux.MakeDeviceID(uint16(statx.RdevMajor), statx.RdevMinor)), + Size: int64(statx.Size), + Blksize: int64(statx.Blksize), + Blocks: int64(statx.Blocks), + ATime: timespecFromStatxTimestamp(statx.Atime), + MTime: timespecFromStatxTimestamp(statx.Mtime), + CTime: timespecFromStatxTimestamp(statx.Ctime), + } +} + +func timespecFromStatxTimestamp(sxts linux.StatxTimestamp) linux.Timespec { + return linux.Timespec{ + Sec: sxts.Sec, + Nsec: int64(sxts.Nsec), + } +} + +// Fstat implements Linux syscall fstat(2). +func Fstat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + statAddr := args[1].Pointer() + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + statx, err := file.Stat(t, vfs.StatOptions{ + Mask: linux.STATX_BASIC_STATS, + }) + if err != nil { + return 0, nil, err + } + var stat linux.Stat + convertStatxToUserStat(t, &statx, &stat) + return 0, nil, stat.CopyOut(t, statAddr) +} + +// Statx implements Linux syscall statx(2). +func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + dirfd := args[0].Int() + pathAddr := args[1].Pointer() + flags := args[2].Int() + mask := args[3].Uint() + statxAddr := args[4].Pointer() + + if flags&^(linux.AT_EMPTY_PATH|linux.AT_SYMLINK_NOFOLLOW) != 0 { + return 0, nil, syserror.EINVAL + } + + opts := vfs.StatOptions{ + Mask: mask, + Sync: uint32(flags & linux.AT_STATX_SYNC_TYPE), + } + + path, err := copyInPath(t, pathAddr) + if err != nil { + return 0, nil, err + } + + root := t.FSContext().RootDirectoryVFS2() + defer root.DecRef() + start := root + if !path.Absolute { + if !path.HasComponents() && flags&linux.AT_EMPTY_PATH == 0 { + return 0, nil, syserror.ENOENT + } + if dirfd == linux.AT_FDCWD { + start = t.FSContext().WorkingDirectoryVFS2() + defer start.DecRef() + } else { + dirfile := t.GetFileVFS2(dirfd) + if dirfile == nil { + return 0, nil, syserror.EBADF + } + if !path.HasComponents() { + // Use FileDescription.Stat() instead of + // VirtualFilesystem.StatAt() for statx(fd, ""), since the + // former may be able to use opened file state to expedite the + // Stat. + statx, err := dirfile.Stat(t, opts) + dirfile.DecRef() + if err != nil { + return 0, nil, err + } + userifyStatx(t, &statx) + return 0, nil, statx.CopyOut(t, statxAddr) + } + start = dirfile.VirtualDentry() + start.IncRef() + defer start.DecRef() + dirfile.DecRef() + } + } + + statx, err := t.Kernel().VFS().StatAt(t, t.Credentials(), &vfs.PathOperation{ + Root: root, + Start: start, + Path: path, + FollowFinalSymlink: flags&linux.AT_SYMLINK_NOFOLLOW == 0, + }, &opts) + if err != nil { + return 0, nil, err + } + userifyStatx(t, &statx) + return 0, nil, statx.CopyOut(t, statxAddr) +} + +func userifyStatx(t *kernel.Task, statx *linux.Statx) { + userns := t.UserNamespace() + statx.UID = uint32(auth.KUID(statx.UID).In(userns).OrOverflow()) + statx.GID = uint32(auth.KGID(statx.GID).In(userns).OrOverflow()) +} + +// Readlink implements Linux syscall readlink(2). +func Readlink(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + pathAddr := args[0].Pointer() + bufAddr := args[1].Pointer() + size := args[2].SizeT() + return readlinkat(t, linux.AT_FDCWD, pathAddr, bufAddr, size) +} + +// Access implements Linux syscall access(2). +func Access(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + // FIXME(jamieliu): actually implement + return 0, nil, nil +} + +// Faccessat implements Linux syscall access(2). +func Faccessat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + // FIXME(jamieliu): actually implement + return 0, nil, nil +} + +// Readlinkat implements Linux syscall mknodat(2). +func Readlinkat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + dirfd := args[0].Int() + pathAddr := args[1].Pointer() + bufAddr := args[2].Pointer() + size := args[3].SizeT() + return readlinkat(t, dirfd, pathAddr, bufAddr, size) +} + +func readlinkat(t *kernel.Task, dirfd int32, pathAddr, bufAddr usermem.Addr, size uint) (uintptr, *kernel.SyscallControl, error) { + if int(size) <= 0 { + return 0, nil, syserror.EINVAL + } + + path, err := copyInPath(t, pathAddr) + if err != nil { + return 0, nil, err + } + // "Since Linux 2.6.39, pathname can be an empty string, in which case the + // call operates on the symbolic link referred to by dirfd ..." - + // readlinkat(2) + tpop, err := getTaskPathOperation(t, dirfd, path, allowEmptyPath, nofollowFinalSymlink) + if err != nil { + return 0, nil, err + } + defer tpop.Release() + + target, err := t.Kernel().VFS().ReadlinkAt(t, t.Credentials(), &tpop.pop) + if err != nil { + return 0, nil, err + } + + if len(target) > int(size) { + target = target[:size] + } + n, err := t.CopyOutBytes(bufAddr, gohacks.ImmutableBytesFromString(target)) + if n == 0 { + return 0, nil, err + } + return uintptr(n), nil, nil +} + +// Statfs implements Linux syscall statfs(2). +func Statfs(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + pathAddr := args[0].Pointer() + bufAddr := args[1].Pointer() + + path, err := copyInPath(t, pathAddr) + if err != nil { + return 0, nil, err + } + tpop, err := getTaskPathOperation(t, linux.AT_FDCWD, path, disallowEmptyPath, followFinalSymlink) + if err != nil { + return 0, nil, err + } + defer tpop.Release() + + statfs, err := t.Kernel().VFS().StatFSAt(t, t.Credentials(), &tpop.pop) + if err != nil { + return 0, nil, err + } + + return 0, nil, statfs.CopyOut(t, bufAddr) +} + +// Fstatfs implements Linux syscall fstatfs(2). +func Fstatfs(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + bufAddr := args[1].Pointer() + + tpop, err := getTaskPathOperation(t, fd, fspath.Path{}, allowEmptyPath, nofollowFinalSymlink) + if err != nil { + return 0, nil, err + } + defer tpop.Release() + + statfs, err := t.Kernel().VFS().StatFSAt(t, t.Credentials(), &tpop.pop) + if err != nil { + return 0, nil, err + } + + return 0, nil, statfs.CopyOut(t, bufAddr) +} diff --git a/pkg/sentry/syscalls/linux/vfs2/sync.go b/pkg/sentry/syscalls/linux/vfs2/sync.go new file mode 100644 index 000000000..365250b0b --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/sync.go @@ -0,0 +1,87 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/syserror" +) + +// Sync implements Linux syscall sync(2). +func Sync(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + return 0, nil, t.Kernel().VFS().SyncAllFilesystems(t) +} + +// Syncfs implements Linux syscall syncfs(2). +func Syncfs(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + return 0, nil, file.SyncFS(t) +} + +// Fsync implements Linux syscall fsync(2). +func Fsync(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + return 0, nil, file.Sync(t) +} + +// Fdatasync implements Linux syscall fdatasync(2). +func Fdatasync(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + // TODO(gvisor.dev/issue/1897): Avoid writeback of unnecessary metadata. + return Fsync(t, args) +} + +// SyncFileRange implements Linux syscall sync_file_range(2). +func SyncFileRange(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + offset := args[1].Int64() + nbytes := args[2].Int64() + flags := args[3].Uint() + + if offset < 0 { + return 0, nil, syserror.EINVAL + } + if nbytes < 0 { + return 0, nil, syserror.EINVAL + } + if flags&^(linux.SYNC_FILE_RANGE_WAIT_BEFORE|linux.SYNC_FILE_RANGE_WRITE|linux.SYNC_FILE_RANGE_WAIT_AFTER) != 0 { + return 0, nil, syserror.EINVAL + } + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // TODO(gvisor.dev/issue/1897): Avoid writeback of data ranges outside of + // [offset, offset+nbytes). + return 0, nil, file.Sync(t) +} diff --git a/pkg/sentry/syscalls/linux/vfs2/sys_read.go b/pkg/sentry/syscalls/linux/vfs2/sys_read.go deleted file mode 100644 index 7667524c7..000000000 --- a/pkg/sentry/syscalls/linux/vfs2/sys_read.go +++ /dev/null @@ -1,95 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package vfs2 - -import ( - "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" - "gvisor.dev/gvisor/pkg/waiter" -) - -const ( - // EventMaskRead contains events that can be triggered on reads. - EventMaskRead = waiter.EventIn | waiter.EventHUp | waiter.EventErr -) - -// Read implements linux syscall read(2). Note that we try to get a buffer that -// is exactly the size requested because some applications like qemu expect -// they can do large reads all at once. Bug for bug. Same for other read -// calls below. -func Read(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := args[0].Int() - addr := args[1].Pointer() - size := args[2].SizeT() - - file := t.GetFileVFS2(fd) - if file == nil { - return 0, nil, syserror.EBADF - } - defer file.DecRef() - - // Check that the size is legitimate. - si := int(size) - if si < 0 { - return 0, nil, syserror.EINVAL - } - - // Get the destination of the read. - dst, err := t.SingleIOSequence(addr, si, usermem.IOOpts{ - AddressSpaceActive: true, - }) - if err != nil { - return 0, nil, err - } - - n, err := read(t, file, dst, vfs.ReadOptions{}) - t.IOUsage().AccountReadSyscall(n) - return uintptr(n), nil, linux.HandleIOErrorVFS2(t, n != 0, err, kernel.ERESTARTSYS, "read", file) -} - -func read(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - n, err := file.Read(t, dst, opts) - if err != syserror.ErrWouldBlock { - return n, err - } - - // Register for notifications. - w, ch := waiter.NewChannelEntry(nil) - file.EventRegister(&w, EventMaskRead) - - total := n - for { - // Shorten dst to reflect bytes previously read. - dst = dst.DropFirst(int(n)) - - // Issue the request and break out if it completes with anything other than - // "would block". - n, err := file.Read(t, dst, opts) - total += n - if err != syserror.ErrWouldBlock { - break - } - if err := t.Block(ch); err != nil { - break - } - } - file.EventUnregister(&w) - - return total, err -} diff --git a/pkg/sentry/syscalls/linux/vfs2/xattr.go b/pkg/sentry/syscalls/linux/vfs2/xattr.go new file mode 100644 index 000000000..89e9ff4d7 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/xattr.go @@ -0,0 +1,353 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "bytes" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/gohacks" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +// Listxattr implements Linux syscall listxattr(2). +func Listxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + return listxattr(t, args, followFinalSymlink) +} + +// Llistxattr implements Linux syscall llistxattr(2). +func Llistxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + return listxattr(t, args, nofollowFinalSymlink) +} + +func listxattr(t *kernel.Task, args arch.SyscallArguments, shouldFollowFinalSymlink shouldFollowFinalSymlink) (uintptr, *kernel.SyscallControl, error) { + pathAddr := args[0].Pointer() + listAddr := args[1].Pointer() + size := args[2].SizeT() + + path, err := copyInPath(t, pathAddr) + if err != nil { + return 0, nil, err + } + tpop, err := getTaskPathOperation(t, linux.AT_FDCWD, path, disallowEmptyPath, shouldFollowFinalSymlink) + if err != nil { + return 0, nil, err + } + defer tpop.Release() + + names, err := t.Kernel().VFS().ListxattrAt(t, t.Credentials(), &tpop.pop) + if err != nil { + return 0, nil, err + } + n, err := copyOutXattrNameList(t, listAddr, size, names) + if err != nil { + return 0, nil, err + } + return uintptr(n), nil, nil +} + +// Flistxattr implements Linux syscall flistxattr(2). +func Flistxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + listAddr := args[1].Pointer() + size := args[2].SizeT() + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + names, err := file.Listxattr(t) + if err != nil { + return 0, nil, err + } + n, err := copyOutXattrNameList(t, listAddr, size, names) + if err != nil { + return 0, nil, err + } + return uintptr(n), nil, nil +} + +// Getxattr implements Linux syscall getxattr(2). +func Getxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + return getxattr(t, args, followFinalSymlink) +} + +// Lgetxattr implements Linux syscall lgetxattr(2). +func Lgetxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + return getxattr(t, args, nofollowFinalSymlink) +} + +func getxattr(t *kernel.Task, args arch.SyscallArguments, shouldFollowFinalSymlink shouldFollowFinalSymlink) (uintptr, *kernel.SyscallControl, error) { + pathAddr := args[0].Pointer() + nameAddr := args[1].Pointer() + valueAddr := args[2].Pointer() + size := args[3].SizeT() + + path, err := copyInPath(t, pathAddr) + if err != nil { + return 0, nil, err + } + tpop, err := getTaskPathOperation(t, linux.AT_FDCWD, path, disallowEmptyPath, shouldFollowFinalSymlink) + if err != nil { + return 0, nil, err + } + defer tpop.Release() + + name, err := copyInXattrName(t, nameAddr) + if err != nil { + return 0, nil, err + } + + value, err := t.Kernel().VFS().GetxattrAt(t, t.Credentials(), &tpop.pop, name) + if err != nil { + return 0, nil, err + } + n, err := copyOutXattrValue(t, valueAddr, size, value) + if err != nil { + return 0, nil, err + } + return uintptr(n), nil, nil +} + +// Fgetxattr implements Linux syscall fgetxattr(2). +func Fgetxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + nameAddr := args[1].Pointer() + valueAddr := args[2].Pointer() + size := args[3].SizeT() + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + name, err := copyInXattrName(t, nameAddr) + if err != nil { + return 0, nil, err + } + + value, err := file.Getxattr(t, name) + if err != nil { + return 0, nil, err + } + n, err := copyOutXattrValue(t, valueAddr, size, value) + if err != nil { + return 0, nil, err + } + return uintptr(n), nil, nil +} + +// Setxattr implements Linux syscall setxattr(2). +func Setxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + return 0, nil, setxattr(t, args, followFinalSymlink) +} + +// Lsetxattr implements Linux syscall lsetxattr(2). +func Lsetxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + return 0, nil, setxattr(t, args, nofollowFinalSymlink) +} + +func setxattr(t *kernel.Task, args arch.SyscallArguments, shouldFollowFinalSymlink shouldFollowFinalSymlink) error { + pathAddr := args[0].Pointer() + nameAddr := args[1].Pointer() + valueAddr := args[2].Pointer() + size := args[3].SizeT() + flags := args[4].Int() + + if flags&^(linux.XATTR_CREATE|linux.XATTR_REPLACE) != 0 { + return syserror.EINVAL + } + + path, err := copyInPath(t, pathAddr) + if err != nil { + return err + } + tpop, err := getTaskPathOperation(t, linux.AT_FDCWD, path, disallowEmptyPath, shouldFollowFinalSymlink) + if err != nil { + return err + } + defer tpop.Release() + + name, err := copyInXattrName(t, nameAddr) + if err != nil { + return err + } + value, err := copyInXattrValue(t, valueAddr, size) + if err != nil { + return err + } + + return t.Kernel().VFS().SetxattrAt(t, t.Credentials(), &tpop.pop, &vfs.SetxattrOptions{ + Name: name, + Value: value, + Flags: uint32(flags), + }) +} + +// Fsetxattr implements Linux syscall fsetxattr(2). +func Fsetxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + nameAddr := args[1].Pointer() + valueAddr := args[2].Pointer() + size := args[3].SizeT() + flags := args[4].Int() + + if flags&^(linux.XATTR_CREATE|linux.XATTR_REPLACE) != 0 { + return 0, nil, syserror.EINVAL + } + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + name, err := copyInXattrName(t, nameAddr) + if err != nil { + return 0, nil, err + } + value, err := copyInXattrValue(t, valueAddr, size) + if err != nil { + return 0, nil, err + } + + return 0, nil, file.Setxattr(t, vfs.SetxattrOptions{ + Name: name, + Value: value, + Flags: uint32(flags), + }) +} + +// Removexattr implements Linux syscall removexattr(2). +func Removexattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + return 0, nil, removexattr(t, args, followFinalSymlink) +} + +// Lremovexattr implements Linux syscall lremovexattr(2). +func Lremovexattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + return 0, nil, removexattr(t, args, nofollowFinalSymlink) +} + +func removexattr(t *kernel.Task, args arch.SyscallArguments, shouldFollowFinalSymlink shouldFollowFinalSymlink) error { + pathAddr := args[0].Pointer() + nameAddr := args[1].Pointer() + + path, err := copyInPath(t, pathAddr) + if err != nil { + return err + } + tpop, err := getTaskPathOperation(t, linux.AT_FDCWD, path, disallowEmptyPath, shouldFollowFinalSymlink) + if err != nil { + return err + } + defer tpop.Release() + + name, err := copyInXattrName(t, nameAddr) + if err != nil { + return err + } + + return t.Kernel().VFS().RemovexattrAt(t, t.Credentials(), &tpop.pop, name) +} + +// Fremovexattr implements Linux syscall fremovexattr(2). +func Fremovexattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + nameAddr := args[1].Pointer() + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + name, err := copyInXattrName(t, nameAddr) + if err != nil { + return 0, nil, err + } + + return 0, nil, file.Removexattr(t, name) +} + +func copyInXattrName(t *kernel.Task, nameAddr usermem.Addr) (string, error) { + name, err := t.CopyInString(nameAddr, linux.XATTR_NAME_MAX+1) + if err != nil { + if err == syserror.ENAMETOOLONG { + return "", syserror.ERANGE + } + return "", err + } + if len(name) == 0 { + return "", syserror.ERANGE + } + return name, nil +} + +func copyOutXattrNameList(t *kernel.Task, listAddr usermem.Addr, size uint, names []string) (int, error) { + if size > linux.XATTR_LIST_MAX { + size = linux.XATTR_LIST_MAX + } + var buf bytes.Buffer + for _, name := range names { + buf.WriteString(name) + buf.WriteByte(0) + } + if size == 0 { + // Return the size that would be required to accomodate the list. + return buf.Len(), nil + } + if buf.Len() > int(size) { + if size >= linux.XATTR_LIST_MAX { + return 0, syserror.E2BIG + } + return 0, syserror.ERANGE + } + return t.CopyOutBytes(listAddr, buf.Bytes()) +} + +func copyInXattrValue(t *kernel.Task, valueAddr usermem.Addr, size uint) (string, error) { + if size > linux.XATTR_SIZE_MAX { + return "", syserror.E2BIG + } + buf := make([]byte, size) + if _, err := t.CopyInBytes(valueAddr, buf); err != nil { + return "", err + } + return gohacks.StringFromImmutableBytes(buf), nil +} + +func copyOutXattrValue(t *kernel.Task, valueAddr usermem.Addr, size uint, value string) (int, error) { + if size > linux.XATTR_SIZE_MAX { + size = linux.XATTR_SIZE_MAX + } + if size == 0 { + // Return the size that would be required to accomodate the value. + return len(value), nil + } + if len(value) > int(size) { + if size >= linux.XATTR_SIZE_MAX { + return 0, syserror.E2BIG + } + return 0, syserror.ERANGE + } + return t.CopyOutBytes(valueAddr, gohacks.ImmutableBytesFromString(value)) +} diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index 0b4f18ab5..07c8383e6 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -43,6 +43,7 @@ go_library( "//pkg/abi/linux", "//pkg/context", "//pkg/fspath", + "//pkg/gohacks", "//pkg/log", "//pkg/sentry/arch", "//pkg/sentry/fs/lock", diff --git a/pkg/sentry/vfs/epoll.go b/pkg/sentry/vfs/epoll.go index eed41139b..3da45d744 100644 --- a/pkg/sentry/vfs/epoll.go +++ b/pkg/sentry/vfs/epoll.go @@ -202,6 +202,9 @@ func (ep *EpollInstance) AddInterest(file *FileDescription, num int32, event lin // Add epi to file.epolls so that it is removed when the last // FileDescription reference is dropped. file.epollMu.Lock() + if file.epolls == nil { + file.epolls = make(map[*epollInterest]struct{}) + } file.epolls[epi] = struct{}{} file.epollMu.Unlock() diff --git a/pkg/sentry/vfs/mount_unsafe.go b/pkg/sentry/vfs/mount_unsafe.go index 1fe766a44..bc7581698 100644 --- a/pkg/sentry/vfs/mount_unsafe.go +++ b/pkg/sentry/vfs/mount_unsafe.go @@ -26,6 +26,7 @@ import ( "sync/atomic" "unsafe" + "gvisor.dev/gvisor/pkg/gohacks" "gvisor.dev/gvisor/pkg/sync" ) @@ -160,7 +161,7 @@ func newMountTableSlots(cap uintptr) unsafe.Pointer { // Lookup may be called even if there are concurrent mutators of mt. func (mt *mountTable) Lookup(parent *Mount, point *Dentry) *Mount { key := mountKey{parent: unsafe.Pointer(parent), point: unsafe.Pointer(point)} - hash := memhash(noescape(unsafe.Pointer(&key)), uintptr(mt.seed), mountKeyBytes) + hash := memhash(gohacks.Noescape(unsafe.Pointer(&key)), uintptr(mt.seed), mountKeyBytes) loop: for { @@ -361,12 +362,3 @@ func memhash(p unsafe.Pointer, seed, s uintptr) uintptr //go:linkname rand32 runtime.fastrand func rand32() uint32 - -// This is copy/pasted from runtime.noescape(), and is needed because arguments -// apparently escape from all functions defined by linkname. -// -//go:nosplit -func noescape(p unsafe.Pointer) unsafe.Pointer { - x := uintptr(p) - return unsafe.Pointer(x ^ 0) -} diff --git a/pkg/sentry/vfs/resolving_path.go b/pkg/sentry/vfs/resolving_path.go index 8a0b382f6..eb4ebb511 100644 --- a/pkg/sentry/vfs/resolving_path.go +++ b/pkg/sentry/vfs/resolving_path.go @@ -228,7 +228,7 @@ func (rp *ResolvingPath) Advance() { rp.pit = next } else { // at end of path segment, continue with next one rp.curPart-- - rp.pit = rp.parts[rp.curPart-1] + rp.pit = rp.parts[rp.curPart] } } diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 8f29031b2..73f8043be 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -385,15 +385,11 @@ func (vfs *VirtualFilesystem) OpenAt(ctx context.Context, creds *auth.Credential // Only a regular file can be executed. stat, err := fd.Stat(ctx, StatOptions{Mask: linux.STATX_TYPE}) if err != nil { + fd.DecRef() return nil, err } - if stat.Mask&linux.STATX_TYPE != 0 { - // This shouldn't happen, but if type can't be retrieved, file can't - // be executed. - return nil, syserror.EACCES - } - if t := linux.FileMode(stat.Mode).FileType(); t != linux.ModeRegular { - ctx.Infof("%q is not a regular file: %v", pop.Path, t) + if stat.Mask&linux.STATX_TYPE == 0 || stat.Mode&linux.S_IFMT != linux.S_IFREG { + fd.DecRef() return nil, syserror.EACCES } } diff --git a/pkg/usermem/BUILD b/pkg/usermem/BUILD index ff8b9e91a..6c9ada9c7 100644 --- a/pkg/usermem/BUILD +++ b/pkg/usermem/BUILD @@ -25,7 +25,6 @@ go_library( "bytes_io_unsafe.go", "usermem.go", "usermem_arm64.go", - "usermem_unsafe.go", "usermem_x86.go", ], visibility = ["//:sandbox"], @@ -33,6 +32,7 @@ go_library( "//pkg/atomicbitops", "//pkg/binary", "//pkg/context", + "//pkg/gohacks", "//pkg/log", "//pkg/safemem", "//pkg/syserror", diff --git a/pkg/usermem/usermem.go b/pkg/usermem/usermem.go index 71fd4e155..d2f4403b0 100644 --- a/pkg/usermem/usermem.go +++ b/pkg/usermem/usermem.go @@ -23,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/gohacks" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/syserror" ) @@ -251,7 +252,7 @@ func CopyStringIn(ctx context.Context, uio IO, addr Addr, maxlen int, opts IOOpt } end, ok := addr.AddLength(uint64(readlen)) if !ok { - return stringFromImmutableBytes(buf[:done]), syserror.EFAULT + return gohacks.StringFromImmutableBytes(buf[:done]), syserror.EFAULT } // Shorten the read to avoid crossing page boundaries, since faulting // in a page unnecessarily is expensive. This also ensures that partial @@ -272,16 +273,16 @@ func CopyStringIn(ctx context.Context, uio IO, addr Addr, maxlen int, opts IOOpt // Look for the terminating zero byte, which may have occurred before // hitting err. if i := bytes.IndexByte(buf[done:done+n], byte(0)); i >= 0 { - return stringFromImmutableBytes(buf[:done+i]), nil + return gohacks.StringFromImmutableBytes(buf[:done+i]), nil } done += n if err != nil { - return stringFromImmutableBytes(buf[:done]), err + return gohacks.StringFromImmutableBytes(buf[:done]), err } addr = end } - return stringFromImmutableBytes(buf), syserror.ENAMETOOLONG + return gohacks.StringFromImmutableBytes(buf), syserror.ENAMETOOLONG } // CopyOutVec copies bytes from src to the memory mapped at ars in uio. The diff --git a/pkg/usermem/usermem_unsafe.go b/pkg/usermem/usermem_unsafe.go deleted file mode 100644 index 876783e78..000000000 --- a/pkg/usermem/usermem_unsafe.go +++ /dev/null @@ -1,27 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package usermem - -import ( - "unsafe" -) - -// stringFromImmutableBytes is equivalent to string(bs), except that it never -// copies even if escape analysis can't prove that bs does not escape. This is -// only valid if bs is never mutated after stringFromImmutableBytes returns. -func stringFromImmutableBytes(bs []byte) string { - // Compare strings.Builder.String(). - return *(*string)(unsafe.Pointer(&bs)) -} diff --git a/runsc/boot/filter/config.go b/runsc/boot/filter/config.go index c69f4c602..a4627905e 100644 --- a/runsc/boot/filter/config.go +++ b/runsc/boot/filter/config.go @@ -229,7 +229,9 @@ var allowedSyscalls = seccomp.SyscallRules{ syscall.SYS_NANOSLEEP: {}, syscall.SYS_PPOLL: {}, syscall.SYS_PREAD64: {}, + syscall.SYS_PREADV: {}, syscall.SYS_PWRITE64: {}, + syscall.SYS_PWRITEV: {}, syscall.SYS_READ: {}, syscall.SYS_RECVMSG: []seccomp.Rule{ { -- cgit v1.2.3 From 72e3f3a3eef3a1dc02db0ff71f98a5d7fe89a6e3 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Tue, 25 Feb 2020 13:42:34 -0800 Subject: Add option to skip stuck tasks waiting for address space PiperOrigin-RevId: 297192390 --- pkg/sentry/kernel/kernel.go | 4 ++++ pkg/sentry/kernel/task_context.go | 2 +- pkg/sentry/kernel/task_exec.go | 2 +- pkg/sentry/kernel/task_usermem.go | 2 +- pkg/sentry/mm/address_space.go | 23 ++++++++++++++--------- pkg/sentry/mm/lifecycle.go | 26 ++++++++++++++------------ pkg/sentry/mm/mm.go | 5 +++++ pkg/sentry/mm/mm_test.go | 2 +- 8 files changed, 41 insertions(+), 25 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index c62fd6eb1..8b76750e9 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -247,6 +247,10 @@ type Kernel struct { // VFS keeps the filesystem state used across the kernel. vfs vfs.VirtualFilesystem + + // If set to true, report address space activation waits as if the task is in + // external wait so that the watchdog doesn't report the task stuck. + SleepForAddressSpaceActivation bool } // InitKernelArgs holds arguments to Init. diff --git a/pkg/sentry/kernel/task_context.go b/pkg/sentry/kernel/task_context.go index 2be982684..0158b1788 100644 --- a/pkg/sentry/kernel/task_context.go +++ b/pkg/sentry/kernel/task_context.go @@ -140,7 +140,7 @@ func (k *Kernel) LoadTaskImage(ctx context.Context, args loader.LoadArgs) (*Task } // Prepare a new user address space to load into. - m := mm.NewMemoryManager(k, k) + m := mm.NewMemoryManager(k, k, k.SleepForAddressSpaceActivation) defer m.DecUsers(ctx) args.MemoryManager = m diff --git a/pkg/sentry/kernel/task_exec.go b/pkg/sentry/kernel/task_exec.go index 8f57a34a6..00c425cca 100644 --- a/pkg/sentry/kernel/task_exec.go +++ b/pkg/sentry/kernel/task_exec.go @@ -220,7 +220,7 @@ func (r *runSyscallAfterExecStop) execute(t *Task) taskRunState { t.mu.Unlock() t.unstopVforkParent() // NOTE(b/30316266): All locks must be dropped prior to calling Activate. - t.MemoryManager().Activate() + t.MemoryManager().Activate(t) t.ptraceExec(oldTID) return (*runSyscallExit)(nil) diff --git a/pkg/sentry/kernel/task_usermem.go b/pkg/sentry/kernel/task_usermem.go index 2bf3ce8a8..b02044ad2 100644 --- a/pkg/sentry/kernel/task_usermem.go +++ b/pkg/sentry/kernel/task_usermem.go @@ -30,7 +30,7 @@ var MAX_RW_COUNT = int(usermem.Addr(math.MaxInt32).RoundDown()) // Activate ensures that the task has an active address space. func (t *Task) Activate() { if mm := t.MemoryManager(); mm != nil { - if err := mm.Activate(); err != nil { + if err := mm.Activate(t); err != nil { panic("unable to activate mm: " + err.Error()) } } diff --git a/pkg/sentry/mm/address_space.go b/pkg/sentry/mm/address_space.go index 94d39af60..0332fc71c 100644 --- a/pkg/sentry/mm/address_space.go +++ b/pkg/sentry/mm/address_space.go @@ -18,6 +18,7 @@ import ( "fmt" "sync/atomic" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/usermem" ) @@ -38,7 +39,7 @@ func (mm *MemoryManager) AddressSpace() platform.AddressSpace { // // When this MemoryManager is no longer needed by a task, it should call // Deactivate to release the reference. -func (mm *MemoryManager) Activate() error { +func (mm *MemoryManager) Activate(ctx context.Context) error { // Fast path: the MemoryManager already has an active // platform.AddressSpace, and we just need to indicate that we need it too. for { @@ -91,16 +92,20 @@ func (mm *MemoryManager) Activate() error { if as == nil { // AddressSpace is unavailable, we must wait. // - // activeMu must not be held while waiting, as the user - // of the address space we are waiting on may attempt - // to take activeMu. - // - // Don't call UninterruptibleSleepStart to register the - // wait to allow the watchdog stuck task to trigger in - // case a process is starved waiting for the address - // space. + // activeMu must not be held while waiting, as the user of the address + // space we are waiting on may attempt to take activeMu. mm.activeMu.Unlock() + + sleep := mm.p.CooperativelySchedulesAddressSpace() && mm.sleepForActivation + if sleep { + // Mark this task sleeping while waiting for the address space to + // prevent the watchdog from reporting it as a stuck task. + ctx.UninterruptibleSleepStart(false) + } <-c + if sleep { + ctx.UninterruptibleSleepFinish(false) + } continue } diff --git a/pkg/sentry/mm/lifecycle.go b/pkg/sentry/mm/lifecycle.go index 3c263ebaa..d8a5b9d29 100644 --- a/pkg/sentry/mm/lifecycle.go +++ b/pkg/sentry/mm/lifecycle.go @@ -28,16 +28,17 @@ import ( ) // NewMemoryManager returns a new MemoryManager with no mappings and 1 user. -func NewMemoryManager(p platform.Platform, mfp pgalloc.MemoryFileProvider) *MemoryManager { +func NewMemoryManager(p platform.Platform, mfp pgalloc.MemoryFileProvider, sleepForActivation bool) *MemoryManager { return &MemoryManager{ - p: p, - mfp: mfp, - haveASIO: p.SupportsAddressSpaceIO(), - privateRefs: &privateRefs{}, - users: 1, - auxv: arch.Auxv{}, - dumpability: UserDumpable, - aioManager: aioManager{contexts: make(map[uint64]*AIOContext)}, + p: p, + mfp: mfp, + haveASIO: p.SupportsAddressSpaceIO(), + privateRefs: &privateRefs{}, + users: 1, + auxv: arch.Auxv{}, + dumpability: UserDumpable, + aioManager: aioManager{contexts: make(map[uint64]*AIOContext)}, + sleepForActivation: sleepForActivation, } } @@ -79,9 +80,10 @@ func (mm *MemoryManager) Fork(ctx context.Context) (*MemoryManager, error) { envv: mm.envv, auxv: append(arch.Auxv(nil), mm.auxv...), // IncRef'd below, once we know that there isn't an error. - executable: mm.executable, - dumpability: mm.dumpability, - aioManager: aioManager{contexts: make(map[uint64]*AIOContext)}, + executable: mm.executable, + dumpability: mm.dumpability, + aioManager: aioManager{contexts: make(map[uint64]*AIOContext)}, + sleepForActivation: mm.sleepForActivation, } // Copy vmas. diff --git a/pkg/sentry/mm/mm.go b/pkg/sentry/mm/mm.go index 637383c7a..c2195ae11 100644 --- a/pkg/sentry/mm/mm.go +++ b/pkg/sentry/mm/mm.go @@ -226,6 +226,11 @@ type MemoryManager struct { // aioManager keeps track of AIOContexts used for async IOs. AIOManager // must be cloned when CLONE_VM is used. aioManager aioManager + + // sleepForActivation indicates whether the task should report to be sleeping + // before trying to activate the address space. When set to true, delays in + // activation are not reported as stuck tasks by the watchdog. + sleepForActivation bool } // vma represents a virtual memory area. diff --git a/pkg/sentry/mm/mm_test.go b/pkg/sentry/mm/mm_test.go index edacca741..fdc308542 100644 --- a/pkg/sentry/mm/mm_test.go +++ b/pkg/sentry/mm/mm_test.go @@ -31,7 +31,7 @@ import ( func testMemoryManager(ctx context.Context) *MemoryManager { p := platform.FromContext(ctx) mfp := pgalloc.MemoryFileProviderFromContext(ctx) - mm := NewMemoryManager(p, mfp) + mm := NewMemoryManager(p, mfp, false) mm.layout = arch.MmapLayout{ MinAddr: p.MinUserAddress(), MaxAddr: p.MaxUserAddress(), -- cgit v1.2.3 From acc405ba60834f5dce9ce04cd762d5cda02224cb Mon Sep 17 00:00:00 2001 From: Nayana Bidari Date: Tue, 25 Feb 2020 15:03:51 -0800 Subject: Add nat table support for iptables. - commit the changes for the comments. --- pkg/abi/linux/netfilter.go | 23 +++++++++-- pkg/sentry/socket/netfilter/netfilter.go | 33 +++++++++++----- pkg/tcpip/iptables/iptables.go | 10 ++++- pkg/tcpip/iptables/targets.go | 66 ++++++++++++++++++++++---------- pkg/tcpip/iptables/types.go | 4 +- pkg/tcpip/stack/nic.go | 13 +------ test/iptables/nat.go | 12 +----- 7 files changed, 103 insertions(+), 58 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/netfilter.go b/pkg/abi/linux/netfilter.go index ba4d84962..2179ac995 100644 --- a/pkg/abi/linux/netfilter.go +++ b/pkg/abi/linux/netfilter.go @@ -250,8 +250,24 @@ type XTErrorTarget struct { // SizeOfXTErrorTarget is the size of an XTErrorTarget. const SizeOfXTErrorTarget = 64 +// Flag values for NfNATIPV4Range. The values indicate whether to map +// protocol specific part(ports) or IPs. It corresponds to values in +// include/uapi/linux/netfilter/nf_nat.h. +const ( + NF_NAT_RANGE_MAP_IPS = 1 << 0 + NF_NAT_RANGE_PROTO_SPECIFIED = 1 << 1 + NF_NAT_RANGE_PROTO_RANDOM = 1 << 2 + NF_NAT_RANGE_PERSISTENT = 1 << 3 + NF_NAT_RANGE_PROTO_RANDOM_FULLY = 1 << 4 + NF_NAT_RANGE_PROTO_RANDOM_ALL = (NF_NAT_RANGE_PROTO_RANDOM | NF_NAT_RANGE_PROTO_RANDOM_FULLY) + NF_NAT_RANGE_MASK = (NF_NAT_RANGE_MAP_IPS | + NF_NAT_RANGE_PROTO_SPECIFIED | NF_NAT_RANGE_PROTO_RANDOM | + NF_NAT_RANGE_PERSISTENT | NF_NAT_RANGE_PROTO_RANDOM_FULLY) +) + // NfNATIPV4Range. It corresponds to struct nf_nat_ipv4_range -// in include/uapi/linux/netfilter/nf_nat.h. +// in include/uapi/linux/netfilter/nf_nat.h. The fields are in +// network byte order. type NfNATIPV4Range struct { Flags uint32 MinIP [4]byte @@ -263,11 +279,12 @@ type NfNATIPV4Range struct { // NfNATIPV4MultiRangeCompat. It corresponds to struct // nf_nat_ipv4_multi_range_compat in include/uapi/linux/netfilter/nf_nat.h. type NfNATIPV4MultiRangeCompat struct { - Rangesize uint32 - RangeIPV4 [1]NfNATIPV4Range + RangeSize uint32 + RangeIPV4 NfNATIPV4Range } // XTRedirectTarget triggers a redirect when reached. +// Adding 4 bytes of padding to make the struct 8 byte aligned. type XTRedirectTarget struct { Target XTEntryTarget NfRange NfNATIPV4MultiRangeCompat diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 512ad624a..257cb485b 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -26,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/iptables" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/usermem" @@ -288,6 +289,7 @@ func marshalRedirectTarget() []byte { TargetSize: linux.SizeOfXTRedirectTarget, }, } + copy(target.Target.Name[:], redirectTargetName) ret := make([]byte, 0, linux.SizeOfXTRedirectTarget) return binary.Marshal(ret, usermem.ByteOrder, target) @@ -405,7 +407,7 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { nflog("entry doesn't have enough room for its target (only %d bytes remain)", len(optVal)) return syserr.ErrInvalidArgument } - target, err := parseTarget(optVal[:targetSize]) + target, err := parseTarget(filter, optVal[:targetSize]) if err != nil { nflog("failed to parse target: %v", err) return syserr.ErrInvalidArgument @@ -552,7 +554,7 @@ func parseMatchers(filter iptables.IPHeaderFilter, optVal []byte) ([]iptables.Ma // parseTarget parses a target from optVal. optVal should contain only the // target. -func parseTarget(optVal []byte) (iptables.Target, error) { +func parseTarget(filter iptables.IPHeaderFilter, optVal []byte) (iptables.Target, error) { nflog("set entries: parsing target of size %d", len(optVal)) if len(optVal) < linux.SizeOfXTEntryTarget { return nil, fmt.Errorf("optVal has insufficient size for entry target %d", len(optVal)) @@ -604,6 +606,10 @@ func parseTarget(optVal []byte) (iptables.Target, error) { return nil, fmt.Errorf("netfilter.SetEntries: optVal has insufficient size for redirect target %d", len(optVal)) } + if filter.Protocol != header.TCPProtocolNumber && filter.Protocol != header.UDPProtocolNumber { + return nil, fmt.Errorf("netfilter.SetEntries: invalid argument") + } + var redirectTarget linux.XTRedirectTarget buf = optVal[:linux.SizeOfXTRedirectTarget] binary.Unmarshal(buf, usermem.ByteOrder, &redirectTarget) @@ -612,21 +618,30 @@ func parseTarget(optVal []byte) (iptables.Target, error) { var target iptables.RedirectTarget nfRange := redirectTarget.NfRange - target.RangeSize = nfRange.Rangesize - target.Flags = nfRange.RangeIPV4[0].Flags + // RangeSize should be 1. + if nfRange.RangeSize != 1 { + return nil, fmt.Errorf("netfilter.SetEntries: invalid argument") + } + + // TODO(gvisor.dev/issue/170): Check if the flags are valid. + // Also check if we need to map ports or IP. + // For now, redirect target only supports dest port change. + if nfRange.RangeIPV4.Flags&linux.NF_NAT_RANGE_PROTO_SPECIFIED == 0 { + return nil, fmt.Errorf("netfilter.SetEntries: invalid argument.") + } + target.Flags = nfRange.RangeIPV4.Flags - target.MinIP = tcpip.Address(nfRange.RangeIPV4[0].MinIP[:]) - target.MaxIP = tcpip.Address(nfRange.RangeIPV4[0].MaxIP[:]) + target.MinIP = tcpip.Address(nfRange.RangeIPV4.MinIP[:]) + target.MaxIP = tcpip.Address(nfRange.RangeIPV4.MaxIP[:]) // Convert port from big endian to little endian. port := make([]byte, 2) - binary.BigEndian.PutUint16(port, nfRange.RangeIPV4[0].MinPort) + binary.BigEndian.PutUint16(port, nfRange.RangeIPV4.MinPort) target.MinPort = binary.LittleEndian.Uint16(port) - binary.BigEndian.PutUint16(port, nfRange.RangeIPV4[0].MaxPort) + binary.BigEndian.PutUint16(port, nfRange.RangeIPV4.MaxPort) target.MaxPort = binary.LittleEndian.Uint16(port) return target, nil - } // Unknown target. diff --git a/pkg/tcpip/iptables/iptables.go b/pkg/tcpip/iptables/iptables.go index c00d012c0..f7dc4f720 100644 --- a/pkg/tcpip/iptables/iptables.go +++ b/pkg/tcpip/iptables/iptables.go @@ -207,7 +207,7 @@ func (it *IPTables) checkTable(hook Hook, pkt tcpip.PacketBuffer, tablename stri underflow := table.Rules[table.Underflows[hook]] // Underflow is guaranteed to be an unconditional // ACCEPT or DROP. - switch v, _ := underflow.Target.Action(pkt); v { + switch v, _ := underflow.Target.Action(pkt, underflow.Filter); v { case RuleAccept: return TableAccept case RuleDrop: @@ -233,6 +233,12 @@ func (it *IPTables) checkTable(hook Hook, pkt tcpip.PacketBuffer, tablename stri func (it *IPTables) checkRule(hook Hook, pkt tcpip.PacketBuffer, table Table, ruleIdx int) RuleVerdict { rule := table.Rules[ruleIdx] + // If pkt.NetworkHeader hasn't been set yet, it will be contained in + // pkt.Data.First(). + if pkt.NetworkHeader == nil { + pkt.NetworkHeader = pkt.Data.First() + } + // First check whether the packet matches the IP header filter. // TODO(gvisor.dev/issue/170): Support other fields of the filter. if rule.Filter.Protocol != 0 && rule.Filter.Protocol != header.IPv4(pkt.NetworkHeader).TransportProtocol() { @@ -252,6 +258,6 @@ func (it *IPTables) checkRule(hook Hook, pkt tcpip.PacketBuffer, table Table, ru } // All the matchers matched, so run the target. - verdict, _ := rule.Target.Action(pkt) + verdict, _ := rule.Target.Action(pkt, rule.Filter) return verdict } diff --git a/pkg/tcpip/iptables/targets.go b/pkg/tcpip/iptables/targets.go index 06e65bece..a75938da3 100644 --- a/pkg/tcpip/iptables/targets.go +++ b/pkg/tcpip/iptables/targets.go @@ -26,7 +26,7 @@ import ( type AcceptTarget struct{} // Action implements Target.Action. -func (AcceptTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, string) { +func (AcceptTarget) Action(packet tcpip.PacketBuffer, filter IPHeaderFilter) (RuleVerdict, string) { return RuleAccept, "" } @@ -34,7 +34,7 @@ func (AcceptTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, string) { type DropTarget struct{} // Action implements Target.Action. -func (DropTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, string) { +func (DropTarget) Action(packet tcpip.PacketBuffer, filter IPHeaderFilter) (RuleVerdict, string) { return RuleDrop, "" } @@ -43,7 +43,7 @@ func (DropTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, string) { type ErrorTarget struct{} // Action implements Target.Action. -func (ErrorTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, string) { +func (ErrorTarget) Action(packet tcpip.PacketBuffer, filter IPHeaderFilter) (RuleVerdict, string) { log.Debugf("ErrorTarget triggered.") return RuleDrop, "" } @@ -54,7 +54,7 @@ type UserChainTarget struct { } // Action implements Target.Action. -func (UserChainTarget) Action(tcpip.PacketBuffer) (RuleVerdict, string) { +func (UserChainTarget) Action(tcpip.PacketBuffer, IPHeaderFilter) (RuleVerdict, string) { panic("UserChainTarget should never be called.") } @@ -63,29 +63,55 @@ func (UserChainTarget) Action(tcpip.PacketBuffer) (RuleVerdict, string) { type ReturnTarget struct{} // Action implements Target.Action. -func (ReturnTarget) Action(tcpip.PacketBuffer) (RuleVerdict, string) { +func (ReturnTarget) Action(tcpip.PacketBuffer, IPHeaderFilter) (RuleVerdict, string) { return RuleReturn, "" } // RedirectTarget redirects the packet by modifying the destination port/IP. +// Min and Max values for IP and Ports in the struct indicate the range of +// values which can be used to redirect. type RedirectTarget struct { - RangeSize uint32 - Flags uint32 - MinIP tcpip.Address - MaxIP tcpip.Address - MinPort uint16 - MaxPort uint16 -} + // Flags to check if the redirect is for address or ports. + Flags uint32 -// Action implements Target.Action. -func (rt RedirectTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, string) { - log.Infof("RedirectTarget triggered.") + // Min address used to redirect. + MinIP tcpip.Address + + // Max address used to redirect. + MaxIP tcpip.Address - // TODO(gvisor.dev/issue/170): Checking only for UDP protocol. - // We're yet to support for TCP protocol. - headerView := packet.Data.First() - h := header.UDP(headerView) - h.SetDestinationPort(rt.MinPort) + // Min port used to redirect. + MinPort uint16 + // Max port used to redirect. + MaxPort uint16 +} + +// Action implements Target.Action. +func (rt RedirectTarget) Action(pkt tcpip.PacketBuffer, filter IPHeaderFilter) (RuleVerdict, string) { + headerView := pkt.Data.First() + + // Network header should be set. + netHeader := header.IPv4(headerView) + if netHeader == nil { + return RuleDrop, "" + } + + // TODO(gvisor.dev/issue/170): Check Flags in RedirectTarget if + // we need to change dest address (for OUTPUT chain) or ports. + hlen := int(netHeader.HeaderLength()) + + switch protocol := filter.Protocol; protocol { + case header.UDPProtocolNumber: + udp := header.UDP(headerView[hlen:]) + udp.SetDestinationPort(rt.MinPort) + case header.TCPProtocolNumber: + // TODO(gvisor.dev/issue/170): Need to recompute checksum + // and implement nat connection tracking to support TCP. + tcp := header.TCP(headerView[hlen:]) + tcp.SetDestinationPort(rt.MinPort) + default: + return RuleDrop, "" + } return RuleAccept, "" } diff --git a/pkg/tcpip/iptables/types.go b/pkg/tcpip/iptables/types.go index 5735d001b..0102831d0 100644 --- a/pkg/tcpip/iptables/types.go +++ b/pkg/tcpip/iptables/types.go @@ -63,7 +63,7 @@ const ( // TableAccept indicates the packet should continue through netstack. TableAccept TableVerdict = iota - // TableAccept indicates the packet should be dropped. + // TableDrop indicates the packet should be dropped. TableDrop ) @@ -175,5 +175,5 @@ type Target interface { // Action takes an action on the packet and returns a verdict on how // traversal should (or should not) continue. If the return value is // Jump, it also returns the name of the chain to jump to. - Action(packet tcpip.PacketBuffer) (RuleVerdict, string) + Action(packet tcpip.PacketBuffer, filter IPHeaderFilter) (RuleVerdict, string) } diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 2028f5201..a75dc0322 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -1087,19 +1087,8 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link // TODO(gvisor.dev/issue/170): Not supporting iptables for IPv6 yet. if protocol == header.IPv4ProtocolNumber { - newPkt := pkt.Clone() - - headerView := newPkt.Data.First() - h := header.IPv4(headerView) - newPkt.NetworkHeader = headerView[:h.HeaderLength()] - - hlen := int(h.HeaderLength()) - tlen := int(h.TotalLength()) - newPkt.Data.TrimFront(hlen) - newPkt.Data.CapLength(tlen - hlen) - ipt := n.stack.IPTables() - if ok := ipt.Check(iptables.Prerouting, newPkt); !ok { + if ok := ipt.Check(iptables.Prerouting, pkt); !ok { // iptables is telling us to drop the packet. return } diff --git a/test/iptables/nat.go b/test/iptables/nat.go index 306cbd1b3..899d1c9d3 100644 --- a/test/iptables/nat.go +++ b/test/iptables/nat.go @@ -71,20 +71,12 @@ func (NATRedirectTCPPort) ContainerAction(ip net.IP) error { } // Listen for TCP packets on redirect port. - if err := listenTCP(redirectPort, sendloopDuration); err != nil { - return fmt.Errorf("connection on port %d should be accepted, but got error %v", redirectPort, err) - } - - return nil + return listenTCP(redirectPort, sendloopDuration) } // LocalAction implements TestCase.LocalAction. func (NATRedirectTCPPort) LocalAction(ip net.IP) error { - if err := connectTCP(ip, dropPort, acceptPort, sendloopDuration); err != nil { - return fmt.Errorf("connection destined to port %d should be accepted, but got error %v", dropPort, err) - } - - return nil + return connectTCP(ip, dropPort, acceptPort, sendloopDuration) } // NATDropUDP tests that packets are not received in ports other than redirect port. -- cgit v1.2.3 From 87288b26a1c40776da31c1edcbe9b1f3a6f5c1ed Mon Sep 17 00:00:00 2001 From: Ian Gudger Date: Tue, 25 Feb 2020 15:34:20 -0800 Subject: Add netlink sockopt logging to strace. PiperOrigin-RevId: 297220008 --- pkg/sentry/strace/socket.go | 9 +++++++++ 1 file changed, 9 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/strace/socket.go b/pkg/sentry/strace/socket.go index 51e6d81b2..c0512de89 100644 --- a/pkg/sentry/strace/socket.go +++ b/pkg/sentry/strace/socket.go @@ -632,4 +632,13 @@ var sockOptNames = map[uint64]abi.ValueSet{ linux.MCAST_MSFILTER: "MCAST_MSFILTER", linux.IPV6_ADDRFORM: "IPV6_ADDRFORM", }, + linux.SOL_NETLINK: { + linux.NETLINK_BROADCAST_ERROR: "NETLINK_BROADCAST_ERROR", + linux.NETLINK_CAP_ACK: "NETLINK_CAP_ACK", + linux.NETLINK_DUMP_STRICT_CHK: "NETLINK_DUMP_STRICT_CHK", + linux.NETLINK_EXT_ACK: "NETLINK_EXT_ACK", + linux.NETLINK_LIST_MEMBERSHIPS: "NETLINK_LIST_MEMBERSHIPS", + linux.NETLINK_NO_ENOBUFS: "NETLINK_NO_ENOBUFS", + linux.NETLINK_PKTINFO: "NETLINK_PKTINFO", + }, } -- cgit v1.2.3 From 73201f4c5700ce7af404b968698b86d80989ab1e Mon Sep 17 00:00:00 2001 From: Haibo Xu Date: Tue, 25 Feb 2020 07:30:12 +0000 Subject: Code Clean: Move arch independent codes to common file in kvm pkg. Signed-off-by: Haibo Xu Change-Id: Iefbdf53e8e8d6d23ae75d8a2ff0d2a6e71f414d8 --- pkg/sentry/platform/kvm/kvm.go | 32 ++++++++++++++++++++++++++++++++ pkg/sentry/platform/kvm/kvm_amd64.go | 32 -------------------------------- pkg/sentry/platform/kvm/kvm_arm64.go | 32 -------------------------------- 3 files changed, 32 insertions(+), 64 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/kvm/kvm.go b/pkg/sentry/platform/kvm/kvm.go index 972ba85c3..a9b4af43e 100644 --- a/pkg/sentry/platform/kvm/kvm.go +++ b/pkg/sentry/platform/kvm/kvm.go @@ -27,6 +27,38 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) +// userMemoryRegion is a region of physical memory. +// +// This mirrors kvm_memory_region. +type userMemoryRegion struct { + slot uint32 + flags uint32 + guestPhysAddr uint64 + memorySize uint64 + userspaceAddr uint64 +} + +// runData is the run structure. This may be mapped for synchronous register +// access (although that doesn't appear to be supported by my kernel at least). +// +// This mirrors kvm_run. +type runData struct { + requestInterruptWindow uint8 + _ [7]uint8 + + exitReason uint32 + readyForInterruptInjection uint8 + ifFlag uint8 + _ [2]uint8 + + cr8 uint64 + apicBase uint64 + + // This is the union data for exits. Interpretation depends entirely on + // the exitReason above (see vCPU code for more information). + data [32]uint64 +} + // KVM represents a lightweight VM context. type KVM struct { platform.NoCPUPreemptionDetection diff --git a/pkg/sentry/platform/kvm/kvm_amd64.go b/pkg/sentry/platform/kvm/kvm_amd64.go index c5a6f9c7d..093497bc4 100644 --- a/pkg/sentry/platform/kvm/kvm_amd64.go +++ b/pkg/sentry/platform/kvm/kvm_amd64.go @@ -21,17 +21,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/platform/ring0" ) -// userMemoryRegion is a region of physical memory. -// -// This mirrors kvm_memory_region. -type userMemoryRegion struct { - slot uint32 - flags uint32 - guestPhysAddr uint64 - memorySize uint64 - userspaceAddr uint64 -} - // userRegs represents KVM user registers. // // This mirrors kvm_regs. @@ -169,27 +158,6 @@ type modelControlRegisters struct { entries [16]modelControlRegister } -// runData is the run structure. This may be mapped for synchronous register -// access (although that doesn't appear to be supported by my kernel at least). -// -// This mirrors kvm_run. -type runData struct { - requestInterruptWindow uint8 - _ [7]uint8 - - exitReason uint32 - readyForInterruptInjection uint8 - ifFlag uint8 - _ [2]uint8 - - cr8 uint64 - apicBase uint64 - - // This is the union data for exits. Interpretation depends entirely on - // the exitReason above (see vCPU code for more information). - data [32]uint64 -} - // cpuidEntry is a single CPUID entry. // // This mirrors kvm_cpuid_entry2. diff --git a/pkg/sentry/platform/kvm/kvm_arm64.go b/pkg/sentry/platform/kvm/kvm_arm64.go index 2319c86d3..79045651e 100644 --- a/pkg/sentry/platform/kvm/kvm_arm64.go +++ b/pkg/sentry/platform/kvm/kvm_arm64.go @@ -20,17 +20,6 @@ import ( "syscall" ) -// userMemoryRegion is a region of physical memory. -// -// This mirrors kvm_memory_region. -type userMemoryRegion struct { - slot uint32 - flags uint32 - guestPhysAddr uint64 - memorySize uint64 - userspaceAddr uint64 -} - type kvmOneReg struct { id uint64 addr uint64 @@ -53,27 +42,6 @@ type userRegs struct { fpRegs userFpsimdState } -// runData is the run structure. This may be mapped for synchronous register -// access (although that doesn't appear to be supported by my kernel at least). -// -// This mirrors kvm_run. -type runData struct { - requestInterruptWindow uint8 - _ [7]uint8 - - exitReason uint32 - readyForInterruptInjection uint8 - ifFlag uint8 - _ [2]uint8 - - cr8 uint64 - apicBase uint64 - - // This is the union data for exits. Interpretation depends entirely on - // the exitReason above (see vCPU code for more information). - data [32]uint64 -} - // updateGlobalOnce does global initialization. It has to be called only once. func updateGlobalOnce(fd int) error { physicalInit() -- cgit v1.2.3 From fba479b3c78621cb122af20d1d677fe9193a971c Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Tue, 25 Feb 2020 19:03:23 -0800 Subject: Fix DATA RACE in fs.MayDelete. MayDelete must lock the directory also, otherwise concurrent renames may race. Note that this also changes the methods to be aligned with the actual Remove and RemoveDirectory methods to minimize confusion when reading the code. (It was hard to see that resolution was correct.) PiperOrigin-RevId: 297258304 --- pkg/sentry/fs/dirent.go | 25 ++++++++++++++----------- pkg/sentry/syscalls/linux/sys_file.go | 4 ++-- 2 files changed, 16 insertions(+), 13 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/dirent.go b/pkg/sentry/fs/dirent.go index acab0411a..e0b32e1c1 100644 --- a/pkg/sentry/fs/dirent.go +++ b/pkg/sentry/fs/dirent.go @@ -1438,8 +1438,8 @@ func lockForRename(oldParent *Dirent, oldName string, newParent *Dirent, newName }, nil } -func checkSticky(ctx context.Context, dir *Dirent, victim *Dirent) error { - uattr, err := dir.Inode.UnstableAttr(ctx) +func (d *Dirent) checkSticky(ctx context.Context, victim *Dirent) error { + uattr, err := d.Inode.UnstableAttr(ctx) if err != nil { return syserror.EPERM } @@ -1465,30 +1465,33 @@ func checkSticky(ctx context.Context, dir *Dirent, victim *Dirent) error { return syserror.EPERM } -// MayDelete determines whether `name`, a child of `dir`, can be deleted or +// MayDelete determines whether `name`, a child of `d`, can be deleted or // renamed by `ctx`. // // Compare Linux kernel fs/namei.c:may_delete. -func MayDelete(ctx context.Context, root, dir *Dirent, name string) error { - if err := dir.Inode.CheckPermission(ctx, PermMask{Write: true, Execute: true}); err != nil { +func (d *Dirent) MayDelete(ctx context.Context, root *Dirent, name string) error { + if err := d.Inode.CheckPermission(ctx, PermMask{Write: true, Execute: true}); err != nil { return err } - victim, err := dir.Walk(ctx, root, name) + unlock := d.lockDirectory() + defer unlock() + + victim, err := d.walk(ctx, root, name, true /* may unlock */) if err != nil { return err } defer victim.DecRef() - return mayDelete(ctx, dir, victim) + return d.mayDelete(ctx, victim) } // mayDelete determines whether `victim`, a child of `dir`, can be deleted or // renamed by `ctx`. // // Preconditions: `dir` is writable and executable by `ctx`. -func mayDelete(ctx context.Context, dir, victim *Dirent) error { - if err := checkSticky(ctx, dir, victim); err != nil { +func (d *Dirent) mayDelete(ctx context.Context, victim *Dirent) error { + if err := d.checkSticky(ctx, victim); err != nil { return err } @@ -1542,7 +1545,7 @@ func Rename(ctx context.Context, root *Dirent, oldParent *Dirent, oldName string defer renamed.DecRef() // Check that the renamed dirent is deletable. - if err := mayDelete(ctx, oldParent, renamed); err != nil { + if err := oldParent.mayDelete(ctx, renamed); err != nil { return err } @@ -1580,7 +1583,7 @@ func Rename(ctx context.Context, root *Dirent, oldParent *Dirent, oldName string // across the Rename, so must call DecRef manually (no defer). // Check that we can delete replaced. - if err := mayDelete(ctx, newParent, replaced); err != nil { + if err := newParent.mayDelete(ctx, replaced); err != nil { replaced.DecRef() return err } diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go index c21f14dc0..d10a9bed8 100644 --- a/pkg/sentry/syscalls/linux/sys_file.go +++ b/pkg/sentry/syscalls/linux/sys_file.go @@ -1236,7 +1236,7 @@ func rmdirAt(t *kernel.Task, dirFD int32, addr usermem.Addr) error { return syserror.ENOTEMPTY } - if err := fs.MayDelete(t, root, d, name); err != nil { + if err := d.MayDelete(t, root, name); err != nil { return err } @@ -1517,7 +1517,7 @@ func unlinkAt(t *kernel.Task, dirFD int32, addr usermem.Addr) error { return syserror.ENOTDIR } - if err := fs.MayDelete(t, root, d, name); err != nil { + if err := d.MayDelete(t, root, name); err != nil { return err } -- cgit v1.2.3 From a92087f0f8fe82ce99414ec99ffe33e514cb21f6 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Tue, 25 Feb 2020 19:12:22 -0800 Subject: Add VFS.NewDisconnectedMount(). Analogous to Linux's kern_mount(). PiperOrigin-RevId: 297259580 --- pkg/sentry/vfs/mount.go | 17 +++++++++++++++++ pkg/sentry/vfs/vfs.go | 16 +++++++++++----- 2 files changed, 28 insertions(+), 5 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index 9912df799..31a4e5480 100644 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go @@ -139,6 +139,23 @@ func (vfs *VirtualFilesystem) NewMountNamespace(ctx context.Context, creds *auth return mntns, nil } +// NewDisconnectedMount returns a Mount representing fs with the given root +// (which may be nil). The new Mount is not associated with any MountNamespace +// and is not connected to any other Mounts. References are taken on fs and +// root. +func (vfs *VirtualFilesystem) NewDisconnectedMount(fs *Filesystem, root *Dentry, opts *MountOptions) (*Mount, error) { + fs.IncRef() + if root != nil { + root.IncRef() + } + return &Mount{ + vfs: vfs, + fs: fs, + root: root, + refs: 1, + }, nil +} + // MountAt creates and mounts a Filesystem configured by the given arguments. func (vfs *VirtualFilesystem) MountAt(ctx context.Context, creds *auth.Credentials, source string, target *PathOperation, fsTypeName string, opts *MountOptions) error { rft := vfs.getFilesystemType(fsTypeName) diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 73f8043be..bde81e1ef 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -126,17 +126,23 @@ func (vfs *VirtualFilesystem) Init() error { // Construct vfs.anonMount. anonfsDevMinor, err := vfs.GetAnonBlockDevMinor() if err != nil { - return err + // This shouldn't be possible since anonBlockDevMinorNext was + // initialized to 1 above (no device numbers have been allocated yet). + panic(fmt.Sprintf("VirtualFilesystem.Init: device number allocation for anonfs failed: %v", err)) } anonfs := anonFilesystem{ devMinor: anonfsDevMinor, } anonfs.vfsfs.Init(vfs, &anonfs) - vfs.anonMount = &Mount{ - vfs: vfs, - fs: &anonfs.vfsfs, - refs: 1, + defer anonfs.vfsfs.DecRef() + anonMount, err := vfs.NewDisconnectedMount(&anonfs.vfsfs, nil, &MountOptions{}) + if err != nil { + // We should not be passing any MountOptions that would cause + // construction of this mount to fail. + panic(fmt.Sprintf("VirtualFilesystem.Init: anonfs mount failed: %v", err)) } + vfs.anonMount = anonMount + return nil } -- cgit v1.2.3 From d8ed78431162fcaed0b31b54d939c8a54d4736e7 Mon Sep 17 00:00:00 2001 From: moricho Date: Tue, 25 Feb 2020 16:49:08 +0900 Subject: add profile option --- pkg/sentry/control/pprof.go | 34 +++++++++++++++++++++--- runsc/boot/controller.go | 13 +++++---- runsc/cmd/debug.go | 64 +++++++++++++++++++++++++++++++++++++-------- runsc/sandbox/sandbox.go | 60 ++++++++++++++++++++++++++++++++++++++++++ 4 files changed, 152 insertions(+), 19 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/control/pprof.go b/pkg/sentry/control/pprof.go index 151808911..5d1907c0e 100644 --- a/pkg/sentry/control/pprof.go +++ b/pkg/sentry/control/pprof.go @@ -117,15 +117,43 @@ func (p *Profile) HeapProfile(o *ProfileOpts, _ *struct{}) error { return nil } -// Goroutine is an RPC stub which dumps out the stack trace for all running +// GoroutineProfile is an RPC stub which dumps out the stack trace for all running // goroutines. -func (p *Profile) Goroutine(o *ProfileOpts, _ *struct{}) error { +func (p *Profile) GoroutineProfile(o *ProfileOpts, _ *struct{}) error { if len(o.FilePayload.Files) < 1 { return errNoOutput } output := o.FilePayload.Files[0] defer output.Close() - if err := pprof.Lookup("goroutine").WriteTo(output, 2); err != nil { + if err := pprof.Lookup("goroutine").WriteTo(output, 0); err != nil { + return err + } + return nil +} + +// BlockProfile is an RPC stub which dumps out the stack trace that led to +// blocking on synchronization primitives. +func (p *Profile) BlockProfile(o *ProfileOpts, _ *struct{}) error { + if len(o.FilePayload.Files) < 1 { + return errNoOutput + } + output := o.FilePayload.Files[0] + defer output.Close() + if err := pprof.Lookup("block").WriteTo(output, 0); err != nil { + return err + } + return nil +} + +// MutexProfile is an RPC stub which dumps out the stack trace of holders of +// contended mutexes. +func (p *Profile) MutexProfile(o *ProfileOpts, _ *struct{}) error { + if len(o.FilePayload.Files) < 1 { + return errNoOutput + } + output := o.FilePayload.Files[0] + defer output.Close() + if err := pprof.Lookup("mutex").WriteTo(output, 0); err != nil { return err } return nil diff --git a/runsc/boot/controller.go b/runsc/boot/controller.go index 17e774e0c..8125d5061 100644 --- a/runsc/boot/controller.go +++ b/runsc/boot/controller.go @@ -101,11 +101,14 @@ const ( // Profiling related commands (see pprof.go for more details). const ( - StartCPUProfile = "Profile.StartCPUProfile" - StopCPUProfile = "Profile.StopCPUProfile" - HeapProfile = "Profile.HeapProfile" - StartTrace = "Profile.StartTrace" - StopTrace = "Profile.StopTrace" + StartCPUProfile = "Profile.StartCPUProfile" + StopCPUProfile = "Profile.StopCPUProfile" + HeapProfile = "Profile.HeapProfile" + GoroutineProfile = "Profile.GoroutineProfile" + BlockProfile = "Profile.BlockProfile" + MutexProfile = "Profile.MutexProfile" + StartTrace = "Profile.StartTrace" + StopTrace = "Profile.StopTrace" ) // Logging related commands (see logging.go for more details). diff --git a/runsc/cmd/debug.go b/runsc/cmd/debug.go index 79965460e..b5de2588b 100644 --- a/runsc/cmd/debug.go +++ b/runsc/cmd/debug.go @@ -32,17 +32,20 @@ import ( // Debug implements subcommands.Command for the "debug" command. type Debug struct { - pid int - stacks bool - signal int - profileHeap string - profileCPU string - trace string - strace string - logLevel string - logPackets string - duration time.Duration - ps bool + pid int + stacks bool + signal int + profileHeap string + profileCPU string + profileGoroutine string + profileBlock string + profileMutex string + trace string + strace string + logLevel string + logPackets string + duration time.Duration + ps bool } // Name implements subcommands.Command. @@ -66,6 +69,9 @@ func (d *Debug) SetFlags(f *flag.FlagSet) { f.BoolVar(&d.stacks, "stacks", false, "if true, dumps all sandbox stacks to the log") f.StringVar(&d.profileHeap, "profile-heap", "", "writes heap profile to the given file.") f.StringVar(&d.profileCPU, "profile-cpu", "", "writes CPU profile to the given file.") + f.StringVar(&d.profileGoroutine, "profile-goroutine", "", "writes goroutine profile to the given file.") + f.StringVar(&d.profileBlock, "profile-block", "", "writes block profile to the given file.") + f.StringVar(&d.profileMutex, "profile-mutex", "", "writes mutex profile to the given file.") f.DurationVar(&d.duration, "duration", time.Second, "amount of time to wait for CPU and trace profiles") f.StringVar(&d.trace, "trace", "", "writes an execution trace to the given file.") f.IntVar(&d.signal, "signal", -1, "sends signal to the sandbox") @@ -147,6 +153,42 @@ func (d *Debug) Execute(_ context.Context, f *flag.FlagSet, args ...interface{}) } log.Infof("Heap profile written to %q", d.profileHeap) } + if d.profileGoroutine != "" { + f, err := os.Create(d.profileGoroutine) + if err != nil { + return Errorf(err.Error()) + } + defer f.Close() + + if err := c.Sandbox.GoroutineProfile(f); err != nil { + return Errorf(err.Error()) + } + log.Infof("Goroutine profile written to %q", d.profileGoroutine) + } + if d.profileBlock != "" { + f, err := os.Create(d.profileBlock) + if err != nil { + return Errorf(err.Error()) + } + defer f.Close() + + if err := c.Sandbox.BlockProfile(f); err != nil { + return Errorf(err.Error()) + } + log.Infof("Block profile written to %q", d.profileBlock) + } + if d.profileMutex != "" { + f, err := os.Create(d.profileMutex) + if err != nil { + return Errorf(err.Error()) + } + defer f.Close() + + if err := c.Sandbox.MutexProfile(f); err != nil { + return Errorf(err.Error()) + } + log.Infof("Mutex profile written to %q", d.profileMutex) + } delay := false if d.profileCPU != "" { diff --git a/runsc/sandbox/sandbox.go b/runsc/sandbox/sandbox.go index ec72bdbfd..2e0e2fd66 100644 --- a/runsc/sandbox/sandbox.go +++ b/runsc/sandbox/sandbox.go @@ -972,6 +972,66 @@ func (s *Sandbox) StopCPUProfile() error { return nil } +// GoroutineProfile writes a goroutine profile to the given file. +func (s *Sandbox) GoroutineProfile(f *os.File) error { + log.Debugf("Goroutine profile %q", s.ID) + conn, err := s.sandboxConnect() + if err != nil { + return err + } + defer conn.Close() + + opts := control.ProfileOpts{ + FilePayload: urpc.FilePayload{ + Files: []*os.File{f}, + }, + } + if err := conn.Call(boot.GoroutineProfile, &opts, nil); err != nil { + return fmt.Errorf("getting sandbox %q goroutine profile: %v", s.ID, err) + } + return nil +} + +// BlockProfile writes a block profile to the given file. +func (s *Sandbox) BlockProfile(f *os.File) error { + log.Debugf("Block profile %q", s.ID) + conn, err := s.sandboxConnect() + if err != nil { + return err + } + defer conn.Close() + + opts := control.ProfileOpts{ + FilePayload: urpc.FilePayload{ + Files: []*os.File{f}, + }, + } + if err := conn.Call(boot.BlockProfile, &opts, nil); err != nil { + return fmt.Errorf("getting sandbox %q block profile: %v", s.ID, err) + } + return nil +} + +// MutexProfile writes a mutex profile to the given file. +func (s *Sandbox) MutexProfile(f *os.File) error { + log.Debugf("Mutex profile %q", s.ID) + conn, err := s.sandboxConnect() + if err != nil { + return err + } + defer conn.Close() + + opts := control.ProfileOpts{ + FilePayload: urpc.FilePayload{ + Files: []*os.File{f}, + }, + } + if err := conn.Call(boot.MutexProfile, &opts, nil); err != nil { + return fmt.Errorf("getting sandbox %q mutex profile: %v", s.ID, err) + } + return nil +} + // StartTrace start trace writing to the given file. func (s *Sandbox) StartTrace(f *os.File) error { log.Debugf("Trace start %q", s.ID) -- cgit v1.2.3 From 408979e619c4b5df74503c7a887aaaa06fd0d730 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Fri, 14 Feb 2020 17:19:32 -0800 Subject: iptables: filter by IP address (and range) Enables commands such as: $ iptables -A INPUT -d 127.0.0.1 -j ACCEPT $ iptables -t nat -A PREROUTING ! -d 127.0.0.1 -j REDIRECT Also adds a bunch of REDIRECT+destination tests. --- pkg/abi/linux/netfilter.go | 28 ++++- pkg/sentry/socket/netfilter/netfilter.go | 17 ++- pkg/tcpip/iptables/iptables.go | 28 ++++- pkg/tcpip/iptables/types.go | 12 ++ test/iptables/filter_input.go | 65 ++++++++++ test/iptables/filter_output.go | 56 +++++++++ test/iptables/iptables_test.go | 66 ++++++++++ test/iptables/iptables_util.go | 25 +++- test/iptables/nat.go | 201 ++++++++++++++++++++++++++++++- 9 files changed, 487 insertions(+), 11 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/netfilter.go b/pkg/abi/linux/netfilter.go index bd2e13ba1..aa149afb5 100644 --- a/pkg/abi/linux/netfilter.go +++ b/pkg/abi/linux/netfilter.go @@ -158,10 +158,36 @@ type IPTIP struct { // Flags define matching behavior for the IP header. Flags uint8 - // InverseFlags invert the meaning of fields in struct IPTIP. + // InverseFlags invert the meaning of fields in struct IPTIP. See the + // IPT_INV_* flags. InverseFlags uint8 } +// Inverts the meaning of the Protocol field. Corresponds to a constant in +// include/uapi/linux/netfilter/x_tables.h. +const XT_INV_PROTO = 0x40 + +// Flags in IPTIP.InverseFlags. Corresponding constants are in +// include/uapi/linux/netfilter_ipv4/ip_tables.h. +const ( + // Invert the meaning of InputInterface. + IPT_INV_VIA_IN = 0x01 + // Invert the meaning of OutputInterface. + IPT_INV_VIA_OUT = 0x02 + // Unclear what this is, as no references to it exist in the kernel. + IPT_INV_TOS = 0x04 + // Invert the meaning of Src. + IPT_INV_SRCIP = 0x08 + // Invert the meaning of Dst. + IPT_INV_DSTIP = 0x10 + // Invert the meaning of the IPT_F_FRAG flag. + IPT_INV_FRAG = 0x20 + // Invert the meaning of the Protocol field. + IPT_INV_PROTO = XT_INV_PROTO + // Enable all flags. + IPT_INV_MASK = 0x7F +) + // SizeOfIPTIP is the size of an IPTIP. const SizeOfIPTIP = 84 diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 2ec11f6ac..faa3e892a 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -26,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/iptables" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/usermem" @@ -630,8 +631,14 @@ func filterFromIPTIP(iptip linux.IPTIP) (iptables.IPHeaderFilter, error) { if containsUnsupportedFields(iptip) { return iptables.IPHeaderFilter{}, fmt.Errorf("unsupported fields in struct iptip: %+v", iptip) } + if len(iptip.Dst) != header.IPv4AddressSize || len(iptip.DstMask) != header.IPv4AddressSize { + return iptables.IPHeaderFilter{}, fmt.Errorf("incorrect length of destination (%d) and/or destination mask (%d) fields", len(iptip.Dst), len(iptip.DstMask)) + } return iptables.IPHeaderFilter{ - Protocol: tcpip.TransportProtocolNumber(iptip.Protocol), + Protocol: tcpip.TransportProtocolNumber(iptip.Protocol), + Dst: tcpip.Address(iptip.Dst[:]), + DstMask: tcpip.Address(iptip.DstMask[:]), + DstInvert: iptip.InverseFlags&linux.IPT_INV_DSTIP != 0, }, nil } @@ -639,16 +646,16 @@ func containsUnsupportedFields(iptip linux.IPTIP) bool { // Currently we check that everything except protocol is zeroed. var emptyInetAddr = linux.InetAddr{} var emptyInterface = [linux.IFNAMSIZ]byte{} - return iptip.Dst != emptyInetAddr || - iptip.Src != emptyInetAddr || + // Disable any supported inverse flags. + inverseMask := uint8(linux.IPT_INV_DSTIP) + return iptip.Src != emptyInetAddr || iptip.SrcMask != emptyInetAddr || - iptip.DstMask != emptyInetAddr || iptip.InputInterface != emptyInterface || iptip.OutputInterface != emptyInterface || iptip.InputInterfaceMask != emptyInterface || iptip.OutputInterfaceMask != emptyInterface || iptip.Flags != 0 || - iptip.InverseFlags != 0 + iptip.InverseFlags&^inverseMask != 0 } func validUnderflow(rule iptables.Rule) bool { diff --git a/pkg/tcpip/iptables/iptables.go b/pkg/tcpip/iptables/iptables.go index dbaccbb36..262b6448d 100644 --- a/pkg/tcpip/iptables/iptables.go +++ b/pkg/tcpip/iptables/iptables.go @@ -240,9 +240,8 @@ func (it *IPTables) checkChain(hook Hook, pkt tcpip.PacketBuffer, table Table, r func (it *IPTables) checkRule(hook Hook, pkt tcpip.PacketBuffer, table Table, ruleIdx int) (RuleVerdict, int) { rule := table.Rules[ruleIdx] - // First check whether the packet matches the IP header filter. - // TODO(gvisor.dev/issue/170): Support other fields of the filter. - if rule.Filter.Protocol != 0 && rule.Filter.Protocol != header.IPv4(pkt.NetworkHeader).TransportProtocol() { + // Check whether the packet matches the IP header filter. + if !filterMatch(rule.Filter, header.IPv4(pkt.NetworkHeader)) { // Continue on to the next rule. return RuleJump, ruleIdx + 1 } @@ -263,3 +262,26 @@ func (it *IPTables) checkRule(hook Hook, pkt tcpip.PacketBuffer, table Table, ru // All the matchers matched, so run the target. return rule.Target.Action(pkt) } + +func filterMatch(filter IPHeaderFilter, hdr header.IPv4) bool { + // TODO(gvisor.dev/issue/170): Support other fields of the filter. + // Check the transport protocol. + if filter.Protocol != 0 && filter.Protocol != hdr.TransportProtocol() { + return false + } + + // Check the destination IP. + dest := hdr.DestinationAddress() + matches := true + for i := range filter.Dst { + if dest[i]&filter.DstMask[i] != filter.Dst[i] { + matches = false + break + } + } + if matches == filter.DstInvert { + return false + } + + return true +} diff --git a/pkg/tcpip/iptables/types.go b/pkg/tcpip/iptables/types.go index 7d032fd23..e7fcf6bff 100644 --- a/pkg/tcpip/iptables/types.go +++ b/pkg/tcpip/iptables/types.go @@ -144,6 +144,18 @@ type Rule struct { type IPHeaderFilter struct { // Protocol matches the transport protocol. Protocol tcpip.TransportProtocolNumber + + // Dst matches the destination IP address. + Dst tcpip.Address + + // DstMask masks bits of the destination IP address when comparing with + // Dst. + DstMask tcpip.Address + + // DstInvert inverts the meaning of the destination IP check, i.e. when + // true the filter will match packets that fail the destination + // comparison. + DstInvert bool } // A Matcher is the interface for matching packets. diff --git a/test/iptables/filter_input.go b/test/iptables/filter_input.go index b2fb6401a..0d3350d8a 100644 --- a/test/iptables/filter_input.go +++ b/test/iptables/filter_input.go @@ -47,6 +47,8 @@ func init() { RegisterTestCase(FilterInputJumpReturnDrop{}) RegisterTestCase(FilterInputJumpBuiltin{}) RegisterTestCase(FilterInputJumpTwice{}) + RegisterTestCase(FilterInputDestination{}) + RegisterTestCase(FilterInputInvertDestination{}) } // FilterInputDropUDP tests that we can drop UDP traffic. @@ -595,3 +597,66 @@ func (FilterInputJumpTwice) ContainerAction(ip net.IP) error { func (FilterInputJumpTwice) LocalAction(ip net.IP) error { return sendUDPLoop(ip, acceptPort, sendloopDuration) } + +// FilterInputDestination verifies that we can filter packets via `-d +// `. +type FilterInputDestination struct{} + +// Name implements TestCase.Name. +func (FilterInputDestination) Name() string { + return "FilterInputDestination" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterInputDestination) ContainerAction(ip net.IP) error { + addrs, err := localAddrs() + if err != nil { + return err + } + + // Make INPUT's default action DROP, then ACCEPT all packets bound for + // this machine. + rules := [][]string{{"-P", "INPUT", "DROP"}} + for _, addr := range addrs { + rules = append(rules, []string{"-A", "INPUT", "-d", addr, "-j", "ACCEPT"}) + } + if err := filterTableRules(rules); err != nil { + return err + } + + return listenUDP(acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (FilterInputDestination) LocalAction(ip net.IP) error { + return sendUDPLoop(ip, acceptPort, sendloopDuration) +} + +// FilterInputDestination verifies that we can filter packets via `! -d +// `. +type FilterInputInvertDestination struct{} + +// Name implements TestCase.Name. +func (FilterInputInvertDestination) Name() string { + return "FilterInputInvertDestination" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterInputInvertDestination) ContainerAction(ip net.IP) error { + // Make INPUT's default action DROP, then ACCEPT all packets not bound + // for 127.0.0.1. + rules := [][]string{ + {"-P", "INPUT", "DROP"}, + {"-A", "INPUT", "!", "-d", localIP, "-j", "ACCEPT"}, + } + if err := filterTableRules(rules); err != nil { + return err + } + + return listenUDP(acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (FilterInputInvertDestination) LocalAction(ip net.IP) error { + return sendUDPLoop(ip, acceptPort, sendloopDuration) +} diff --git a/test/iptables/filter_output.go b/test/iptables/filter_output.go index ee2c49f9a..39e648e32 100644 --- a/test/iptables/filter_output.go +++ b/test/iptables/filter_output.go @@ -22,6 +22,8 @@ import ( func init() { RegisterTestCase(FilterOutputDropTCPDestPort{}) RegisterTestCase(FilterOutputDropTCPSrcPort{}) + RegisterTestCase(FilterOutputDestination{}) + RegisterTestCase(FilterOutputInvertDestination{}) } // FilterOutputDropTCPDestPort tests that connections are not accepted on specified source ports. @@ -85,3 +87,57 @@ func (FilterOutputDropTCPSrcPort) LocalAction(ip net.IP) error { return nil } + +// FilterOutputDestination tests that we can selectively allow packets to +// certain destinations. +type FilterOutputDestination struct{} + +// Name implements TestCase.Name. +func (FilterOutputDestination) Name() string { + return "FilterOutputDestination" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterOutputDestination) ContainerAction(ip net.IP) error { + rules := [][]string{ + {"-A", "OUTPUT", "-d", ip.String(), "-j", "ACCEPT"}, + {"-P", "OUTPUT", "DROP"}, + } + if err := filterTableRules(rules); err != nil { + return err + } + + return sendUDPLoop(ip, acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (FilterOutputDestination) LocalAction(ip net.IP) error { + return listenUDP(acceptPort, sendloopDuration) +} + +// FilterOutputInvertDestination tests that we can selectively allow packets +// not headed for a particular destination. +type FilterOutputInvertDestination struct{} + +// Name implements TestCase.Name. +func (FilterOutputInvertDestination) Name() string { + return "FilterOutputInvertDestination" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterOutputInvertDestination) ContainerAction(ip net.IP) error { + rules := [][]string{ + {"-A", "OUTPUT", "!", "-d", localIP, "-j", "ACCEPT"}, + {"-P", "OUTPUT", "DROP"}, + } + if err := filterTableRules(rules); err != nil { + return err + } + + return sendUDPLoop(ip, acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (FilterOutputInvertDestination) LocalAction(ip net.IP) error { + return listenUDP(acceptPort, sendloopDuration) +} diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go index 0621861eb..5eabd2461 100644 --- a/test/iptables/iptables_test.go +++ b/test/iptables/iptables_test.go @@ -285,3 +285,69 @@ func TestJumpTwice(t *testing.T) { t.Fatal(err) } } + +func TestInputDestination(t *testing.T) { + if err := singleTest(FilterInputDestination{}); err != nil { + t.Fatal(err) + } +} + +func TestInputInvertDestination(t *testing.T) { + if err := singleTest(FilterInputInvertDestination{}); err != nil { + t.Fatal(err) + } +} + +func TestOutputDestination(t *testing.T) { + if err := singleTest(FilterOutputDestination{}); err != nil { + t.Fatal(err) + } +} + +func TestOutputInvertDestination(t *testing.T) { + if err := singleTest(FilterOutputInvertDestination{}); err != nil { + t.Fatal(err) + } +} + +func TestNATOutRedirectIP(t *testing.T) { + if err := singleTest(NATOutRedirectIP{}); err != nil { + t.Fatal(err) + } +} + +func TestNATOutDontRedirectIP(t *testing.T) { + if err := singleTest(NATOutDontRedirectIP{}); err != nil { + t.Fatal(err) + } +} + +func TestNATOutRedirectInvert(t *testing.T) { + if err := singleTest(NATOutRedirectInvert{}); err != nil { + t.Fatal(err) + } +} + +func TestNATPreRedirectIP(t *testing.T) { + if err := singleTest(NATPreRedirectIP{}); err != nil { + t.Fatal(err) + } +} + +func TestNATPreDontRedirectIP(t *testing.T) { + if err := singleTest(NATPreDontRedirectIP{}); err != nil { + t.Fatal(err) + } +} + +func TestNATPreRedirectInvert(t *testing.T) { + if err := singleTest(NATPreRedirectInvert{}); err != nil { + t.Fatal(err) + } +} + +func TestNATRedirectRequiresProtocol(t *testing.T) { + if err := singleTest(NATRedirectRequiresProtocol{}); err != nil { + t.Fatal(err) + } +} diff --git a/test/iptables/iptables_util.go b/test/iptables/iptables_util.go index 32cf5a417..178a662e1 100644 --- a/test/iptables/iptables_util.go +++ b/test/iptables/iptables_util.go @@ -24,6 +24,7 @@ import ( ) const iptablesBinary = "iptables" +const localIP = "127.0.0.1" // filterTable calls `iptables -t filter` with the given args. func filterTable(args ...string) error { @@ -46,8 +47,17 @@ func tableCmd(table string, args []string) error { // filterTableRules is like filterTable, but runs multiple iptables commands. func filterTableRules(argsList [][]string) error { + return tableRules("filter", argsList) +} + +// natTableRules is like natTable, but runs multiple iptables commands. +func natTableRules(argsList [][]string) error { + return tableRules("nat", argsList) +} + +func tableRules(table string, argsList [][]string) error { for _, args := range argsList { - if err := filterTable(args...); err != nil { + if err := tableCmd(table, args); err != nil { return err } } @@ -149,3 +159,16 @@ func connectTCP(ip net.IP, remotePort, localPort int, timeout time.Duration) err return nil } + +// localAddrs returns a list of local network interface addresses. +func localAddrs() ([]string, error) { + addrs, err := net.InterfaceAddrs() + if err != nil { + return nil, err + } + addrStrs := make([]string, 0, len(addrs)) + for _, addr := range addrs { + addrStrs = append(addrStrs, addr.String()) + } + return addrStrs, nil +} diff --git a/test/iptables/nat.go b/test/iptables/nat.go index a01117ec8..020c862ad 100644 --- a/test/iptables/nat.go +++ b/test/iptables/nat.go @@ -15,8 +15,10 @@ package iptables import ( + "errors" "fmt" "net" + "time" ) const ( @@ -26,6 +28,13 @@ const ( func init() { RegisterTestCase(NATRedirectUDPPort{}) RegisterTestCase(NATDropUDP{}) + RegisterTestCase(NATPreRedirectIP{}) + RegisterTestCase(NATPreDontRedirectIP{}) + RegisterTestCase(NATPreRedirectInvert{}) + RegisterTestCase(NATOutRedirectIP{}) + RegisterTestCase(NATOutDontRedirectIP{}) + RegisterTestCase(NATOutRedirectInvert{}) + RegisterTestCase(NATRedirectRequiresProtocol{}) } // NATRedirectUDPPort tests that packets are redirected to different port. @@ -53,7 +62,8 @@ func (NATRedirectUDPPort) LocalAction(ip net.IP) error { return sendUDPLoop(ip, acceptPort, sendloopDuration) } -// NATDropUDP tests that packets are not received in ports other than redirect port. +// NATDropUDP tests that packets are not received in ports other than redirect +// port. type NATDropUDP struct{} // Name implements TestCase.Name. @@ -78,3 +88,192 @@ func (NATDropUDP) ContainerAction(ip net.IP) error { func (NATDropUDP) LocalAction(ip net.IP) error { return sendUDPLoop(ip, acceptPort, sendloopDuration) } + +// NATOutRedirectIP uses iptables to select packets based on destination IP and +// redirects them. +type NATOutRedirectIP struct{} + +// Name implements TestCase.Name. +func (NATOutRedirectIP) Name() string { + return "NATOutRedirectIP" +} + +// ContainerAction implements TestCase.ContainerAction. +func (NATOutRedirectIP) ContainerAction(ip net.IP) error { + // Redirect OUTPUT packets to a listening localhost port. + dest := net.IP([]byte{200, 0, 0, 2}) + return loopbackTest(dest, "-A", "OUTPUT", "-d", dest.String(), "-p", "udp", "-j", "REDIRECT", "--to-port", fmt.Sprintf("%d", acceptPort)) +} + +// LocalAction implements TestCase.LocalAction. +func (NATOutRedirectIP) LocalAction(ip net.IP) error { + // No-op. + return nil +} + +// NATOutDontRedirectIP tests that iptables matching with "-d" does not match +// packets it shouldn't. +type NATOutDontRedirectIP struct{} + +// Name implements TestCase.Name. +func (NATOutDontRedirectIP) Name() string { + return "NATOutDontRedirectIP" +} + +// ContainerAction implements TestCase.ContainerAction. +func (NATOutDontRedirectIP) ContainerAction(ip net.IP) error { + if err := natTable("-A", "OUTPUT", "-d", localIP, "-p", "udp", "-j", "REDIRECT", "--to-port", fmt.Sprintf("%d", dropPort)); err != nil { + return err + } + return sendUDPLoop(ip, acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (NATOutDontRedirectIP) LocalAction(ip net.IP) error { + return listenUDP(acceptPort, sendloopDuration) +} + +// NATOutRedirectInvert tests that iptables can match with "! -d". +type NATOutRedirectInvert struct{} + +// Name implements TestCase.Name. +func (NATOutRedirectInvert) Name() string { + return "NATOutRedirectInvert" +} + +// ContainerAction implements TestCase.ContainerAction. +func (NATOutRedirectInvert) ContainerAction(ip net.IP) error { + // Redirect OUTPUT packets to a listening localhost port. + dest := []byte{200, 0, 0, 3} + destStr := "200.0.0.2" + return loopbackTest(dest, "-A", "OUTPUT", "!", "-d", destStr, "-p", "udp", "-j", "REDIRECT", "--to-port", fmt.Sprintf("%d", acceptPort)) +} + +// LocalAction implements TestCase.LocalAction. +func (NATOutRedirectInvert) LocalAction(ip net.IP) error { + // No-op. + return nil +} + +// NATPreRedirectIP tests that we can use iptables to select packets based on +// destination IP and redirect them. +type NATPreRedirectIP struct{} + +// Name implements TestCase.Name. +func (NATPreRedirectIP) Name() string { + return "NATPreRedirectIP" +} + +// ContainerAction implements TestCase.ContainerAction. +func (NATPreRedirectIP) ContainerAction(ip net.IP) error { + addrs, err := localAddrs() + if err != nil { + return err + } + + var rules [][]string + for _, addr := range addrs { + rules = append(rules, []string{"-A", "PREROUTING", "-p", "udp", "-d", addr, "-j", "REDIRECT", "--to-ports", fmt.Sprintf("%d", acceptPort)}) + } + if err := natTableRules(rules); err != nil { + return err + } + return listenUDP(acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (NATPreRedirectIP) LocalAction(ip net.IP) error { + return sendUDPLoop(ip, dropPort, sendloopDuration) +} + +// NATPreDontRedirectIP tests that iptables matching with "-d" does not match +// packets it shouldn't. +type NATPreDontRedirectIP struct{} + +// Name implements TestCase.Name. +func (NATPreDontRedirectIP) Name() string { + return "NATPreDontRedirectIP" +} + +// ContainerAction implements TestCase.ContainerAction. +func (NATPreDontRedirectIP) ContainerAction(ip net.IP) error { + if err := natTable("-A", "PREROUTING", "-p", "udp", "-d", localIP, "-j", "REDIRECT", "--to-ports", fmt.Sprintf("%d", dropPort)); err != nil { + return err + } + return listenUDP(acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (NATPreDontRedirectIP) LocalAction(ip net.IP) error { + return sendUDPLoop(ip, acceptPort, sendloopDuration) +} + +// NATPreRedirectInvert tests that iptables can match with "! -d". +type NATPreRedirectInvert struct{} + +// Name implements TestCase.Name. +func (NATPreRedirectInvert) Name() string { + return "NATPreRedirectInvert" +} + +// ContainerAction implements TestCase.ContainerAction. +func (NATPreRedirectInvert) ContainerAction(ip net.IP) error { + if err := natTable("-A", "PREROUTING", "-p", "udp", "!", "-d", localIP, "-j", "REDIRECT", "--to-ports", fmt.Sprintf("%d", acceptPort)); err != nil { + return err + } + return listenUDP(acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (NATPreRedirectInvert) LocalAction(ip net.IP) error { + return sendUDPLoop(ip, dropPort, sendloopDuration) +} + +// NATRedirectRequiresProtocol tests that use of the --to-ports flag requires a +// protocol to be specified with -p. +type NATRedirectRequiresProtocol struct{} + +// Name implements TestCase.Name. +func (NATRedirectRequiresProtocol) Name() string { + return "NATRedirectRequiresProtocol" +} + +// ContainerAction implements TestCase.ContainerAction. +func (NATRedirectRequiresProtocol) ContainerAction(ip net.IP) error { + if err := natTable("-A", "PREROUTING", "-d", localIP, "-j", "REDIRECT", "--to-ports", fmt.Sprintf("%d", acceptPort)); err == nil { + return errors.New("expected an error using REDIRECT --to-ports without a protocol") + } + return nil +} + +// LocalAction implements TestCase.LocalAction. +func (NATRedirectRequiresProtocol) LocalAction(ip net.IP) error { + // No-op. + return nil +} + +// loopbackTests runs an iptables rule and ensures that packets sent to +// dest:dropPort are received by localhost:acceptPort. +func loopbackTest(dest net.IP, args ...string) error { + if err := natTable(args...); err != nil { + return err + } + sendCh := make(chan error) + listenCh := make(chan error) + go func() { + sendCh <- sendUDPLoop(dest, dropPort, sendloopDuration) + }() + go func() { + listenCh <- listenUDP(acceptPort, sendloopDuration) + }() + select { + case err := <-listenCh: + if err != nil { + return err + } + case <-time.After(sendloopDuration): + return errors.New("timed out") + } + // sendCh will always take the full sendloop time. + return <-sendCh +} -- cgit v1.2.3 From 8fb84f78adfc0dba964ebe97edb51ebf8a80f752 Mon Sep 17 00:00:00 2001 From: Rahat Mahmood Date: Wed, 26 Feb 2020 19:28:20 -0800 Subject: Fix construct of linux.Stat for arm64. PiperOrigin-RevId: 297494373 --- pkg/sentry/syscalls/linux/BUILD | 2 ++ pkg/sentry/syscalls/linux/sys_stat.go | 18 ----------- pkg/sentry/syscalls/linux/sys_stat_amd64.go | 45 +++++++++++++++++++++++++++ pkg/sentry/syscalls/linux/sys_stat_arm64.go | 45 +++++++++++++++++++++++++++ pkg/sentry/syscalls/linux/vfs2/BUILD | 2 ++ pkg/sentry/syscalls/linux/vfs2/stat.go | 23 -------------- pkg/sentry/syscalls/linux/vfs2/stat_amd64.go | 46 ++++++++++++++++++++++++++++ pkg/sentry/syscalls/linux/vfs2/stat_arm64.go | 46 ++++++++++++++++++++++++++++ 8 files changed, 186 insertions(+), 41 deletions(-) create mode 100644 pkg/sentry/syscalls/linux/sys_stat_amd64.go create mode 100644 pkg/sentry/syscalls/linux/sys_stat_arm64.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/stat_amd64.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/stat_arm64.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD index c7883e68e..0d24fd3c4 100644 --- a/pkg/sentry/syscalls/linux/BUILD +++ b/pkg/sentry/syscalls/linux/BUILD @@ -42,6 +42,8 @@ go_library( "sys_socket.go", "sys_splice.go", "sys_stat.go", + "sys_stat_amd64.go", + "sys_stat_arm64.go", "sys_sync.go", "sys_sysinfo.go", "sys_syslog.go", diff --git a/pkg/sentry/syscalls/linux/sys_stat.go b/pkg/sentry/syscalls/linux/sys_stat.go index 701b27b4a..9bd2df104 100644 --- a/pkg/sentry/syscalls/linux/sys_stat.go +++ b/pkg/sentry/syscalls/linux/sys_stat.go @@ -25,24 +25,6 @@ import ( // LINT.IfChange -func statFromAttrs(t *kernel.Task, sattr fs.StableAttr, uattr fs.UnstableAttr) linux.Stat { - return linux.Stat{ - Dev: sattr.DeviceID, - Ino: sattr.InodeID, - Nlink: uattr.Links, - Mode: sattr.Type.LinuxType() | uint32(uattr.Perms.LinuxMode()), - UID: uint32(uattr.Owner.UID.In(t.UserNamespace()).OrOverflow()), - GID: uint32(uattr.Owner.GID.In(t.UserNamespace()).OrOverflow()), - Rdev: uint64(linux.MakeDeviceID(sattr.DeviceFileMajor, sattr.DeviceFileMinor)), - Size: uattr.Size, - Blksize: sattr.BlockSize, - Blocks: uattr.Usage / 512, - ATime: uattr.AccessTime.Timespec(), - MTime: uattr.ModificationTime.Timespec(), - CTime: uattr.StatusChangeTime.Timespec(), - } -} - // Stat implements linux syscall stat(2). func Stat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { addr := args[0].Pointer() diff --git a/pkg/sentry/syscalls/linux/sys_stat_amd64.go b/pkg/sentry/syscalls/linux/sys_stat_amd64.go new file mode 100644 index 000000000..0a04a6113 --- /dev/null +++ b/pkg/sentry/syscalls/linux/sys_stat_amd64.go @@ -0,0 +1,45 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build amd64 + +package linux + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/kernel" +) + +// LINT.IfChange + +func statFromAttrs(t *kernel.Task, sattr fs.StableAttr, uattr fs.UnstableAttr) linux.Stat { + return linux.Stat{ + Dev: sattr.DeviceID, + Ino: sattr.InodeID, + Nlink: uattr.Links, + Mode: sattr.Type.LinuxType() | uint32(uattr.Perms.LinuxMode()), + UID: uint32(uattr.Owner.UID.In(t.UserNamespace()).OrOverflow()), + GID: uint32(uattr.Owner.GID.In(t.UserNamespace()).OrOverflow()), + Rdev: uint64(linux.MakeDeviceID(sattr.DeviceFileMajor, sattr.DeviceFileMinor)), + Size: uattr.Size, + Blksize: sattr.BlockSize, + Blocks: uattr.Usage / 512, + ATime: uattr.AccessTime.Timespec(), + MTime: uattr.ModificationTime.Timespec(), + CTime: uattr.StatusChangeTime.Timespec(), + } +} + +// LINT.ThenChange(vfs2/stat_amd64.go) diff --git a/pkg/sentry/syscalls/linux/sys_stat_arm64.go b/pkg/sentry/syscalls/linux/sys_stat_arm64.go new file mode 100644 index 000000000..5a3b1bfad --- /dev/null +++ b/pkg/sentry/syscalls/linux/sys_stat_arm64.go @@ -0,0 +1,45 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package linux + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/kernel" +) + +// LINT.IfChange + +func statFromAttrs(t *kernel.Task, sattr fs.StableAttr, uattr fs.UnstableAttr) linux.Stat { + return linux.Stat{ + Dev: sattr.DeviceID, + Ino: sattr.InodeID, + Nlink: uint32(uattr.Links), + Mode: sattr.Type.LinuxType() | uint32(uattr.Perms.LinuxMode()), + UID: uint32(uattr.Owner.UID.In(t.UserNamespace()).OrOverflow()), + GID: uint32(uattr.Owner.GID.In(t.UserNamespace()).OrOverflow()), + Rdev: uint64(linux.MakeDeviceID(sattr.DeviceFileMajor, sattr.DeviceFileMinor)), + Size: uattr.Size, + Blksize: int32(sattr.BlockSize), + Blocks: uattr.Usage / 512, + ATime: uattr.AccessTime.Timespec(), + MTime: uattr.ModificationTime.Timespec(), + CTime: uattr.StatusChangeTime.Timespec(), + } +} + +// LINT.ThenChange(vfs2/stat_arm64.go) diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD index f51761e81..e7695e995 100644 --- a/pkg/sentry/syscalls/linux/vfs2/BUILD +++ b/pkg/sentry/syscalls/linux/vfs2/BUILD @@ -22,6 +22,8 @@ go_library( "read_write.go", "setstat.go", "stat.go", + "stat_amd64.go", + "stat_arm64.go", "sync.go", "xattr.go", ], diff --git a/pkg/sentry/syscalls/linux/vfs2/stat.go b/pkg/sentry/syscalls/linux/vfs2/stat.go index dca8d7011..12c532310 100644 --- a/pkg/sentry/syscalls/linux/vfs2/stat.go +++ b/pkg/sentry/syscalls/linux/vfs2/stat.go @@ -113,29 +113,6 @@ func fstatat(t *kernel.Task, dirfd int32, pathAddr, statAddr usermem.Addr, flags return stat.CopyOut(t, statAddr) } -// This takes both input and output as pointer arguments to avoid copying large -// structs. -func convertStatxToUserStat(t *kernel.Task, statx *linux.Statx, stat *linux.Stat) { - // Linux just copies fields from struct kstat without regard to struct - // kstat::result_mask (fs/stat.c:cp_new_stat()), so we do too. - userns := t.UserNamespace() - *stat = linux.Stat{ - Dev: uint64(linux.MakeDeviceID(uint16(statx.DevMajor), statx.DevMinor)), - Ino: statx.Ino, - Nlink: uint64(statx.Nlink), - Mode: uint32(statx.Mode), - UID: uint32(auth.KUID(statx.UID).In(userns).OrOverflow()), - GID: uint32(auth.KGID(statx.GID).In(userns).OrOverflow()), - Rdev: uint64(linux.MakeDeviceID(uint16(statx.RdevMajor), statx.RdevMinor)), - Size: int64(statx.Size), - Blksize: int64(statx.Blksize), - Blocks: int64(statx.Blocks), - ATime: timespecFromStatxTimestamp(statx.Atime), - MTime: timespecFromStatxTimestamp(statx.Mtime), - CTime: timespecFromStatxTimestamp(statx.Ctime), - } -} - func timespecFromStatxTimestamp(sxts linux.StatxTimestamp) linux.Timespec { return linux.Timespec{ Sec: sxts.Sec, diff --git a/pkg/sentry/syscalls/linux/vfs2/stat_amd64.go b/pkg/sentry/syscalls/linux/vfs2/stat_amd64.go new file mode 100644 index 000000000..2da538fc6 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/stat_amd64.go @@ -0,0 +1,46 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build amd64 + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" +) + +// This takes both input and output as pointer arguments to avoid copying large +// structs. +func convertStatxToUserStat(t *kernel.Task, statx *linux.Statx, stat *linux.Stat) { + // Linux just copies fields from struct kstat without regard to struct + // kstat::result_mask (fs/stat.c:cp_new_stat()), so we do too. + userns := t.UserNamespace() + *stat = linux.Stat{ + Dev: uint64(linux.MakeDeviceID(uint16(statx.DevMajor), statx.DevMinor)), + Ino: statx.Ino, + Nlink: uint64(statx.Nlink), + Mode: uint32(statx.Mode), + UID: uint32(auth.KUID(statx.UID).In(userns).OrOverflow()), + GID: uint32(auth.KGID(statx.GID).In(userns).OrOverflow()), + Rdev: uint64(linux.MakeDeviceID(uint16(statx.RdevMajor), statx.RdevMinor)), + Size: int64(statx.Size), + Blksize: int64(statx.Blksize), + Blocks: int64(statx.Blocks), + ATime: timespecFromStatxTimestamp(statx.Atime), + MTime: timespecFromStatxTimestamp(statx.Mtime), + CTime: timespecFromStatxTimestamp(statx.Ctime), + } +} diff --git a/pkg/sentry/syscalls/linux/vfs2/stat_arm64.go b/pkg/sentry/syscalls/linux/vfs2/stat_arm64.go new file mode 100644 index 000000000..88b9c7627 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/stat_arm64.go @@ -0,0 +1,46 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" +) + +// This takes both input and output as pointer arguments to avoid copying large +// structs. +func convertStatxToUserStat(t *kernel.Task, statx *linux.Statx, stat *linux.Stat) { + // Linux just copies fields from struct kstat without regard to struct + // kstat::result_mask (fs/stat.c:cp_new_stat()), so we do too. + userns := t.UserNamespace() + *stat = linux.Stat{ + Dev: uint64(linux.MakeDeviceID(uint16(statx.DevMajor), statx.DevMinor)), + Ino: statx.Ino, + Nlink: uint32(statx.Nlink), + Mode: uint32(statx.Mode), + UID: uint32(auth.KUID(statx.UID).In(userns).OrOverflow()), + GID: uint32(auth.KGID(statx.GID).In(userns).OrOverflow()), + Rdev: uint64(linux.MakeDeviceID(uint16(statx.RdevMajor), statx.RdevMinor)), + Size: int64(statx.Size), + Blksize: int32(statx.Blksize), + Blocks: int64(statx.Blocks), + ATime: timespecFromStatxTimestamp(statx.Atime), + MTime: timespecFromStatxTimestamp(statx.Mtime), + CTime: timespecFromStatxTimestamp(statx.Ctime), + } +} -- cgit v1.2.3 From 5f0e8e6239108c1b597e1cb8cb588bbf09d192b7 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Thu, 27 Feb 2020 01:19:28 -0500 Subject: Prepare the vcpu environment for sentry on Arm64 Signed-off-by: Bin Lu --- pkg/sentry/platform/ring0/entry_arm64.s | 18 ++++++++++++++++++ 1 file changed, 18 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/ring0/entry_arm64.s b/pkg/sentry/platform/ring0/entry_arm64.s index d42eda37b..db6465663 100644 --- a/pkg/sentry/platform/ring0/entry_arm64.s +++ b/pkg/sentry/platform/ring0/entry_arm64.s @@ -394,6 +394,8 @@ TEXT ·Current(SB),NOSPLIT,$0-8 #define STACK_FRAME_SIZE 16 +// kernelExitToEl0 is the entrypoint for application in guest_el0. +// Prepare the vcpu environment for container application. TEXT ·kernelExitToEl0(SB),NOSPLIT,$0 // Step1, save sentry context into memory. REGISTERS_SAVE(RSV_REG, CPU_REGISTERS) @@ -464,7 +466,23 @@ TEXT ·kernelExitToEl0(SB),NOSPLIT,$0 ERET() +// kernelExitToEl1 is the entrypoint for sentry in guest_el1. +// Prepare the vcpu environment for sentry. TEXT ·kernelExitToEl1(SB),NOSPLIT,$0 + WORD $0xd538d092 //MRS TPIDR_EL1, R18 + + MOVD CPU_REGISTERS+PTRACE_PSTATE(RSV_REG), R1 + WORD $0xd5184001 //MSR R1, SPSR_EL1 + + MOVD CPU_REGISTERS+PTRACE_PC(RSV_REG), R1 + MSR R1, ELR_EL1 + + MOVD CPU_REGISTERS+PTRACE_SP(RSV_REG), R1 + MOVD R1, RSP + + REGISTERS_LOAD(RSV_REG, CPU_REGISTERS) + MOVD CPU_REGISTERS+PTRACE_R9(RSV_REG), RSV_REG_APP + ERET() // Start is the CPU entrypoint. -- cgit v1.2.3 From abf7ebcd38e8c2750f4542f29115140bb2b44a9b Mon Sep 17 00:00:00 2001 From: Nayana Bidari Date: Thu, 27 Feb 2020 10:59:32 -0800 Subject: Internal change. PiperOrigin-RevId: 297638665 --- pkg/sentry/socket/netstack/netstack.go | 40 +++++++++-- pkg/tcpip/transport/packet/endpoint.go | 21 +++++- test/syscalls/linux/packet_socket.cc | 124 ++++++++++++++++++++++++++++++--- 3 files changed, 167 insertions(+), 18 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index e187276c5..48c268bfa 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -712,14 +712,40 @@ func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking boo // Bind implements the linux syscall bind(2) for sockets backed by // tcpip.Endpoint. func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { - addr, family, err := AddressAndFamily(sockaddr) - if err != nil { - return err - } - if err := s.checkFamily(family, true /* exact */); err != nil { - return err + family := usermem.ByteOrder.Uint16(sockaddr) + var addr tcpip.FullAddress + + // Bind for AF_PACKET requires only family, protocol and ifindex. + // In function AddressAndFamily, we check the address length which is + // not needed for AF_PACKET bind. + if family == linux.AF_PACKET { + var a linux.SockAddrLink + if len(sockaddr) < sockAddrLinkSize { + return syserr.ErrInvalidArgument + } + binary.Unmarshal(sockaddr[:sockAddrLinkSize], usermem.ByteOrder, &a) + + if a.Protocol != uint16(s.protocol) { + return syserr.ErrInvalidArgument + } + + addr = tcpip.FullAddress{ + NIC: tcpip.NICID(a.InterfaceIndex), + Addr: tcpip.Address(a.HardwareAddr[:header.EthernetAddressSize]), + } + } else { + var err *syserr.Error + addr, family, err = AddressAndFamily(sockaddr) + if err != nil { + return err + } + + if err = s.checkFamily(family, true /* exact */); err != nil { + return err + } + + addr = s.mapFamily(addr, family) } - addr = s.mapFamily(addr, family) // Issue the bind request to the endpoint. return syserr.TranslateNetstackError(s.Endpoint.Bind(addr)) diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go index 5722815e9..09a1cd436 100644 --- a/pkg/tcpip/transport/packet/endpoint.go +++ b/pkg/tcpip/transport/packet/endpoint.go @@ -76,6 +76,7 @@ type endpoint struct { sndBufSize int closed bool stats tcpip.TransportEndpointStats `state:"nosave"` + bound bool } // NewEndpoint returns a new packet endpoint. @@ -125,6 +126,7 @@ func (ep *endpoint) Close() { } ep.closed = true + ep.bound = false ep.waiterQueue.Notify(waiter.EventHUp | waiter.EventErr | waiter.EventIn | waiter.EventOut) } @@ -216,7 +218,24 @@ func (ep *endpoint) Bind(addr tcpip.FullAddress) *tcpip.Error { // sll_family (should be AF_PACKET), sll_protocol, and sll_ifindex." // - packet(7). - return tcpip.ErrNotSupported + ep.mu.Lock() + defer ep.mu.Unlock() + + if ep.bound { + return tcpip.ErrAlreadyBound + } + + // Unregister endpoint with all the nics. + ep.stack.UnregisterPacketEndpoint(0, ep.netProto, ep) + + // Bind endpoint to receive packets from specific interface. + if err := ep.stack.RegisterPacketEndpoint(addr.NIC, ep.netProto, ep); err != nil { + return err + } + + ep.bound = true + + return nil } // GetLocalAddress implements tcpip.Endpoint.GetLocalAddress. diff --git a/test/syscalls/linux/packet_socket.cc b/test/syscalls/linux/packet_socket.cc index 92ae55eec..bc22de788 100644 --- a/test/syscalls/linux/packet_socket.cc +++ b/test/syscalls/linux/packet_socket.cc @@ -13,6 +13,7 @@ // limitations under the License. #include +#include #include #include #include @@ -163,16 +164,11 @@ int CookedPacketTest::GetLoopbackIndex() { return ifr.ifr_ifindex; } -// Receive via a packet socket. -TEST_P(CookedPacketTest, Receive) { - // Let's use a simple IP payload: a UDP datagram. - FileDescriptor udp_sock = - ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_DGRAM, 0)); - SendUDPMessage(udp_sock.get()); - +// Receive and verify the message via packet socket on interface. +void ReceiveMessage(int sock, int ifindex) { // Wait for the socket to become readable. struct pollfd pfd = {}; - pfd.fd = socket_; + pfd.fd = sock; pfd.events = POLLIN; EXPECT_THAT(RetryEINTR(poll)(&pfd, 1, 2000), SyscallSucceedsWithValue(1)); @@ -182,9 +178,10 @@ TEST_P(CookedPacketTest, Receive) { char buf[64]; struct sockaddr_ll src = {}; socklen_t src_len = sizeof(src); - ASSERT_THAT(recvfrom(socket_, buf, sizeof(buf), 0, + ASSERT_THAT(recvfrom(sock, buf, sizeof(buf), 0, reinterpret_cast(&src), &src_len), SyscallSucceedsWithValue(packet_size)); + // sockaddr_ll ends with an 8 byte physical address field, but ethernet // addresses only use 6 bytes. Linux used to return sizeof(sockaddr_ll)-2 // here, but since commit b2cf86e1563e33a14a1c69b3e508d15dc12f804c returns @@ -194,7 +191,7 @@ TEST_P(CookedPacketTest, Receive) { // TODO(b/129292371): Verify protocol once we return it. // Verify the source address. EXPECT_EQ(src.sll_family, AF_PACKET); - EXPECT_EQ(src.sll_ifindex, GetLoopbackIndex()); + EXPECT_EQ(src.sll_ifindex, ifindex); EXPECT_EQ(src.sll_halen, ETH_ALEN); // This came from the loopback device, so the address is all 0s. for (int i = 0; i < src.sll_halen; i++) { @@ -222,6 +219,18 @@ TEST_P(CookedPacketTest, Receive) { EXPECT_EQ(strncmp(payload, kMessage, sizeof(kMessage)), 0); } +// Receive via a packet socket. +TEST_P(CookedPacketTest, Receive) { + // Let's use a simple IP payload: a UDP datagram. + FileDescriptor udp_sock = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_DGRAM, 0)); + SendUDPMessage(udp_sock.get()); + + // Receive and verify the data. + int loopback_index = GetLoopbackIndex(); + ReceiveMessage(socket_, loopback_index); +} + // Send via a packet socket. TEST_P(CookedPacketTest, Send) { // TODO(b/129292371): Remove once we support packet socket writing. @@ -313,6 +322,101 @@ TEST_P(CookedPacketTest, Send) { EXPECT_EQ(src.sin_addr.s_addr, htonl(INADDR_LOOPBACK)); } +// Bind and receive via packet socket. +TEST_P(CookedPacketTest, BindReceive) { + struct sockaddr_ll bind_addr = {}; + bind_addr.sll_family = AF_PACKET; + bind_addr.sll_protocol = htons(GetParam()); + bind_addr.sll_ifindex = GetLoopbackIndex(); + + ASSERT_THAT(bind(socket_, reinterpret_cast(&bind_addr), + sizeof(bind_addr)), + SyscallSucceeds()); + + // Let's use a simple IP payload: a UDP datagram. + FileDescriptor udp_sock = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_DGRAM, 0)); + SendUDPMessage(udp_sock.get()); + + // Receive and verify the data. + ReceiveMessage(socket_, bind_addr.sll_ifindex); +} + +// Double Bind socket. +TEST_P(CookedPacketTest, DoubleBind) { + struct sockaddr_ll bind_addr = {}; + bind_addr.sll_family = AF_PACKET; + bind_addr.sll_protocol = htons(GetParam()); + bind_addr.sll_ifindex = GetLoopbackIndex(); + + ASSERT_THAT(bind(socket_, reinterpret_cast(&bind_addr), + sizeof(bind_addr)), + SyscallSucceeds()); + + // Binding socket again should fail. + ASSERT_THAT( + bind(socket_, reinterpret_cast(&bind_addr), + sizeof(bind_addr)), + // Linux 4.09 returns EINVAL here, but some time before 4.19 it switched + // to EADDRINUSE. + AnyOf(SyscallFailsWithErrno(EADDRINUSE), SyscallFailsWithErrno(EINVAL))); +} + +// Bind and verify we do not receive data on interface which is not bound +TEST_P(CookedPacketTest, BindDrop) { + // Let's use a simple IP payload: a UDP datagram. + FileDescriptor udp_sock = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_DGRAM, 0)); + + struct ifaddrs* if_addr_list = nullptr; + auto cleanup = Cleanup([&if_addr_list]() { freeifaddrs(if_addr_list); }); + + ASSERT_THAT(getifaddrs(&if_addr_list), SyscallSucceeds()); + + // Get interface other than loopback. + struct ifreq ifr = {}; + for (struct ifaddrs* i = if_addr_list; i; i = i->ifa_next) { + if (strcmp(i->ifa_name, "lo") != 0) { + strncpy(ifr.ifr_name, i->ifa_name, sizeof(ifr.ifr_name)); + break; + } + } + + // Skip if no interface is available other than loopback. + if (strlen(ifr.ifr_name) == 0) { + GTEST_SKIP(); + } + + // Get interface index. + EXPECT_THAT(ioctl(socket_, SIOCGIFINDEX, &ifr), SyscallSucceeds()); + EXPECT_NE(ifr.ifr_ifindex, 0); + + // Bind to packet socket requires only family, protocol and ifindex. + struct sockaddr_ll bind_addr = {}; + bind_addr.sll_family = AF_PACKET; + bind_addr.sll_protocol = htons(GetParam()); + bind_addr.sll_ifindex = ifr.ifr_ifindex; + + ASSERT_THAT(bind(socket_, reinterpret_cast(&bind_addr), + sizeof(bind_addr)), + SyscallSucceeds()); + + // Send to loopback interface. + struct sockaddr_in dest = {}; + dest.sin_addr.s_addr = htonl(INADDR_LOOPBACK); + dest.sin_family = AF_INET; + dest.sin_port = kPort; + EXPECT_THAT(sendto(udp_sock.get(), kMessage, sizeof(kMessage), 0, + reinterpret_cast(&dest), sizeof(dest)), + SyscallSucceedsWithValue(sizeof(kMessage))); + + // Wait and make sure the socket never receives any data. + struct pollfd pfd = {}; + pfd.fd = socket_; + pfd.events = POLLIN; + EXPECT_THAT(RetryEINTR(poll)(&pfd, 1, 1000), SyscallSucceedsWithValue(0)); +} + INSTANTIATE_TEST_SUITE_P(AllInetTests, CookedPacketTest, ::testing::Values(ETH_P_IP, ETH_P_ALL)); -- cgit v1.2.3 From af6fab651406c411ef848c360b017713de385880 Mon Sep 17 00:00:00 2001 From: Nayana Bidari Date: Fri, 28 Feb 2020 10:00:38 -0800 Subject: Add nat table support for iptables. - Fix review comments. --- pkg/sentry/socket/netfilter/netfilter.go | 14 +++++--- pkg/tcpip/iptables/targets.go | 55 ++++++++++++++++++++++++-------- pkg/tcpip/iptables/types.go | 2 +- 3 files changed, 53 insertions(+), 18 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index f68a2260d..c8a9e67b8 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -292,9 +292,9 @@ func marshalRedirectTarget() []byte { }, } copy(target.Target.Name[:], redirectTargetName) - + ret := make([]byte, 0, linux.SizeOfXTRedirectTarget) - return binary.Marshal(ret, usermem.ByteOrder, target) + return binary.Marshal(ret, usermem.ByteOrder, target) } func marshalJumpTarget(jt JumpTarget) []byte { @@ -670,15 +670,21 @@ func parseTarget(filter iptables.IPHeaderFilter, optVal []byte) (iptables.Target // TODO(gvisor.dev/issue/170): Check if the flags are valid. // Also check if we need to map ports or IP. - // For now, redirect target only supports dest port change. + // For now, redirect target only supports destination port change. + // Port range and IP range are not supported yet. if nfRange.RangeIPV4.Flags&linux.NF_NAT_RANGE_PROTO_SPECIFIED == 0 { return nil, fmt.Errorf("netfilter.SetEntries: invalid argument.") } - target.Flags = nfRange.RangeIPV4.Flags + target.RangeProtoSpecified = true target.MinIP = tcpip.Address(nfRange.RangeIPV4.MinIP[:]) target.MaxIP = tcpip.Address(nfRange.RangeIPV4.MaxIP[:]) + // TODO(gvisor.dev/issue/170): Port range is not supported yet. + if nfRange.RangeIPV4.MinPort != nfRange.RangeIPV4.MaxPort { + return nil, fmt.Errorf("netfilter.SetEntries: invalid argument.") + } + // Convert port from big endian to little endian. port := make([]byte, 2) binary.BigEndian.PutUint16(port, nfRange.RangeIPV4.MinPort) diff --git a/pkg/tcpip/iptables/targets.go b/pkg/tcpip/iptables/targets.go index 96318118c..ae5af7c53 100644 --- a/pkg/tcpip/iptables/targets.go +++ b/pkg/tcpip/iptables/targets.go @@ -69,8 +69,11 @@ func (ReturnTarget) Action(tcpip.PacketBuffer) (RuleVerdict, int) { // Min and Max values for IP and Ports in the struct indicate the range of // values which can be used to redirect. type RedirectTarget struct { - // Flags to check if the redirect is for address or ports. - Flags uint32 + // TODO(gvisor.dev/issue/170): Other flags need to be aded after + // we support them. + // RangeProtoSpecified flag indicates single port is specified to + // redirect. + RangeProtoSpecified bool // Min address used to redirect. MinIP tcpip.Address @@ -86,30 +89,56 @@ type RedirectTarget struct { } // Action implements Target.Action. +// TODO(gvisor.dev/issue/170): Parse headers without copying. The current +// implementation only works for PREROUTING and calls pkt.Clone(), neither +// of which should be the case. func (rt RedirectTarget) Action(pkt tcpip.PacketBuffer) (RuleVerdict, int) { - headerView := pkt.Data.First() + newPkt := pkt.Clone() - // Network header should be set. + // Set network header. + headerView := newPkt.Data.First() netHeader := header.IPv4(headerView) - if netHeader == nil { - return RuleDrop, 0 - } + newPkt.NetworkHeader = headerView[:netHeader.HeaderLength()] - // TODO(gvisor.dev/issue/170): Check Flags in RedirectTarget if - // we need to change dest address (for OUTPUT chain) or ports. hlen := int(netHeader.HeaderLength()) + tlen := int(netHeader.TotalLength()) + newPkt.Data.TrimFront(hlen) + newPkt.Data.CapLength(tlen - hlen) + // TODO(gvisor.dev/issue/170): Change destination address to + // loopback or interface address on which the packet was + // received. + + // TODO(gvisor.dev/issue/170): Check Flags in RedirectTarget if + // we need to change dest address (for OUTPUT chain) or ports. switch protocol := netHeader.TransportProtocol(); protocol { case header.UDPProtocolNumber: - udp := header.UDP(headerView[hlen:]) - udp.SetDestinationPort(rt.MinPort) + var udpHeader header.UDP + if newPkt.TransportHeader != nil { + udpHeader = header.UDP(newPkt.TransportHeader) + } else { + if len(pkt.Data.First()) < header.UDPMinimumSize { + return RuleDrop, 0 + } + udpHeader = header.UDP(newPkt.Data.First()) + } + udpHeader.SetDestinationPort(rt.MinPort) case header.TCPProtocolNumber: + var tcpHeader header.TCP + if newPkt.TransportHeader != nil { + tcpHeader = header.TCP(newPkt.TransportHeader) + } else { + if len(pkt.Data.First()) < header.TCPMinimumSize { + return RuleDrop, 0 + } + tcpHeader = header.TCP(newPkt.TransportHeader) + } // TODO(gvisor.dev/issue/170): Need to recompute checksum // and implement nat connection tracking to support TCP. - tcp := header.TCP(headerView[hlen:]) - tcp.SetDestinationPort(rt.MinPort) + tcpHeader.SetDestinationPort(rt.MinPort) default: return RuleDrop, 0 } + return RuleAccept, 0 } diff --git a/pkg/tcpip/iptables/types.go b/pkg/tcpip/iptables/types.go index 9c2ad2d46..7d032fd23 100644 --- a/pkg/tcpip/iptables/types.go +++ b/pkg/tcpip/iptables/types.go @@ -163,6 +163,6 @@ type Matcher interface { type Target interface { // Action takes an action on the packet and returns a verdict on how // traversal should (or should not) continue. If the return value is - // Jump, it also returns the name of the chain to jump to. + // Jump, it also returns the index of the rule to jump to. Action(packet tcpip.PacketBuffer) (RuleVerdict, int) } -- cgit v1.2.3 From 0f8a9e362337ee684042331c5bf24a3cb43d6fc4 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Fri, 28 Feb 2020 10:13:59 -0800 Subject: Change dup2 call to dup3 We changed syscalls to allow dup3 for ARM64. Updates #1198 PiperOrigin-RevId: 297870816 --- pkg/sentry/fsimpl/gofer/gofer.go | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index d00850e25..c4a8f0b38 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -1045,13 +1045,13 @@ func (d *dentry) ensureSharedHandle(ctx context.Context, read, write, trunc bool // using the old file descriptor, preventing us from safely // closing it. We could handle this by invalidating existing // memmap.Translations, but this is expensive. Instead, use - // dup2() to make the old file descriptor refer to the new file + // dup3 to make the old file descriptor refer to the new file // description, then close the new file descriptor (which is no // longer needed). Racing callers may use the old or new file // description, but this doesn't matter since they refer to the // same file (unless d.fs.opts.overlayfsStaleRead is true, // which we handle separately). - if err := syscall.Dup2(int(h.fd), int(d.handle.fd)); err != nil { + if err := syscall.Dup3(int(h.fd), int(d.handle.fd), 0); err != nil { d.handleMu.Unlock() ctx.Warningf("gofer.dentry.ensureSharedHandle: failed to dup fd %d to fd %d: %v", h.fd, d.handle.fd, err) h.close(ctx) -- cgit v1.2.3 From 6b4d36e3253238dd72d0861ac1220d147e1de8dd Mon Sep 17 00:00:00 2001 From: Ting-Yu Wang Date: Fri, 28 Feb 2020 10:37:52 -0800 Subject: Hide /dev/net/tun when using hostinet. /dev/net/tun does not currently work with hostinet. This has caused some program starts failing because it thinks the feature exists. PiperOrigin-RevId: 297876196 --- pkg/sentry/fs/dev/BUILD | 1 + pkg/sentry/fs/dev/dev.go | 7 +++++-- pkg/sentry/fs/dev/net_tun.go | 7 +++++++ pkg/sentry/kernel/kernel.go | 4 ++++ test/syscalls/BUILD | 5 +++++ test/syscalls/linux/BUILD | 12 +++++++++++ test/syscalls/linux/dev.cc | 7 ------- test/syscalls/linux/tuntap.cc | 7 +++++++ test/syscalls/linux/tuntap_hostinet.cc | 37 ++++++++++++++++++++++++++++++++++ 9 files changed, 78 insertions(+), 9 deletions(-) create mode 100644 test/syscalls/linux/tuntap_hostinet.cc (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/dev/BUILD b/pkg/sentry/fs/dev/BUILD index 9b6bb26d0..9379a4d7b 100644 --- a/pkg/sentry/fs/dev/BUILD +++ b/pkg/sentry/fs/dev/BUILD @@ -26,6 +26,7 @@ go_library( "//pkg/sentry/fs/fsutil", "//pkg/sentry/fs/ramfs", "//pkg/sentry/fs/tmpfs", + "//pkg/sentry/inet", "//pkg/sentry/kernel", "//pkg/sentry/memmap", "//pkg/sentry/mm", diff --git a/pkg/sentry/fs/dev/dev.go b/pkg/sentry/fs/dev/dev.go index 7e66c29b0..acbd401a0 100644 --- a/pkg/sentry/fs/dev/dev.go +++ b/pkg/sentry/fs/dev/dev.go @@ -22,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" "gvisor.dev/gvisor/pkg/sentry/fs/tmpfs" + "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/usermem" ) @@ -124,10 +125,12 @@ func New(ctx context.Context, msrc *fs.MountSource) *fs.Inode { "ptmx": newSymlink(ctx, "pts/ptmx", msrc), "tty": newCharacterDevice(ctx, newTTYDevice(ctx, fs.RootOwner, 0666), msrc, ttyDevMajor, ttyDevMinor), + } - "net": newDirectory(ctx, map[string]*fs.Inode{ + if isNetTunSupported(inet.StackFromContext(ctx)) { + contents["net"] = newDirectory(ctx, map[string]*fs.Inode{ "tun": newCharacterDevice(ctx, newNetTunDevice(ctx, fs.RootOwner, 0666), msrc, netTunDevMajor, netTunDevMinor), - }, msrc), + }, msrc) } iops := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) diff --git a/pkg/sentry/fs/dev/net_tun.go b/pkg/sentry/fs/dev/net_tun.go index 755644488..dc7ad075a 100644 --- a/pkg/sentry/fs/dev/net_tun.go +++ b/pkg/sentry/fs/dev/net_tun.go @@ -20,6 +20,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" + "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/socket/netstack" "gvisor.dev/gvisor/pkg/syserror" @@ -168,3 +169,9 @@ func (fops *netTunFileOperations) EventRegister(e *waiter.Entry, mask waiter.Eve func (fops *netTunFileOperations) EventUnregister(e *waiter.Entry) { fops.device.EventUnregister(e) } + +// isNetTunSupported returns whether /dev/net/tun device is supported for s. +func isNetTunSupported(s inet.Stack) bool { + _, ok := s.(*netstack.Stack) + return ok +} diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 8b76750e9..1d627564f 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -755,6 +755,8 @@ func (ctx *createProcessContext) Value(key interface{}) interface{} { return ctx.k.GlobalInit().Leader().MountNamespaceVFS2() case fs.CtxDirentCacheLimiter: return ctx.k.DirentCacheLimiter + case inet.CtxStack: + return ctx.k.RootNetworkNamespace().Stack() case ktime.CtxRealtimeClock: return ctx.k.RealtimeClock() case limits.CtxLimits: @@ -1481,6 +1483,8 @@ func (ctx supervisorContext) Value(key interface{}) interface{} { return ctx.k.GlobalInit().Leader().MountNamespaceVFS2() case fs.CtxDirentCacheLimiter: return ctx.k.DirentCacheLimiter + case inet.CtxStack: + return ctx.k.RootNetworkNamespace().Stack() case ktime.CtxRealtimeClock: return ctx.k.RealtimeClock() case limits.CtxLimits: diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 3518e862d..a69b0ce13 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -680,6 +680,11 @@ syscall_test( syscall_test(test = "//test/syscalls/linux:tuntap_test") +syscall_test( + add_hostinet = True, + test = "//test/syscalls/linux:tuntap_hostinet_test", +) + syscall_test(test = "//test/syscalls/linux:udp_bind_test") syscall_test( diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 704bae17b..70c120e42 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -3460,6 +3460,18 @@ cc_binary( ], ) +cc_binary( + name = "tuntap_hostinet_test", + testonly = 1, + srcs = ["tuntap_hostinet.cc"], + linkstatic = 1, + deps = [ + gtest, + "//test/util:test_main", + "//test/util:test_util", + ], +) + cc_library( name = "udp_socket_test_cases", testonly = 1, diff --git a/test/syscalls/linux/dev.cc b/test/syscalls/linux/dev.cc index 4e473268c..4dd302eed 100644 --- a/test/syscalls/linux/dev.cc +++ b/test/syscalls/linux/dev.cc @@ -153,13 +153,6 @@ TEST(DevTest, TTYExists) { EXPECT_EQ(statbuf.st_mode, S_IFCHR | 0666); } -TEST(DevTest, NetTunExists) { - struct stat statbuf = {}; - ASSERT_THAT(stat("/dev/net/tun", &statbuf), SyscallSucceeds()); - // Check that it's a character device with rw-rw-rw- permissions. - EXPECT_EQ(statbuf.st_mode, S_IFCHR | 0666); -} - } // namespace } // namespace testing diff --git a/test/syscalls/linux/tuntap.cc b/test/syscalls/linux/tuntap.cc index f6ac9d7b8..f734511d6 100644 --- a/test/syscalls/linux/tuntap.cc +++ b/test/syscalls/linux/tuntap.cc @@ -153,6 +153,13 @@ std::string CreateArpPacket(const uint8_t srcmac[ETH_ALEN], const char* srcip, } // namespace +TEST(TuntapStaticTest, NetTunExists) { + struct stat statbuf; + ASSERT_THAT(stat(kDevNetTun, &statbuf), SyscallSucceeds()); + // Check that it's a character device with rw-rw-rw- permissions. + EXPECT_EQ(statbuf.st_mode, S_IFCHR | 0666); +} + class TuntapTest : public ::testing::Test { protected: void TearDown() override { diff --git a/test/syscalls/linux/tuntap_hostinet.cc b/test/syscalls/linux/tuntap_hostinet.cc new file mode 100644 index 000000000..0c527419e --- /dev/null +++ b/test/syscalls/linux/tuntap_hostinet.cc @@ -0,0 +1,37 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include +#include +#include + +#include "gtest/gtest.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +namespace { + +TEST(TuntapHostInetTest, NoNetTun) { + SKIP_IF(!IsRunningOnGvisor()); + + struct stat statbuf; + ASSERT_THAT(stat("/dev/net/tun", &statbuf), SyscallFailsWithErrno(ENOENT)); +} + +} // namespace +} // namespace testing + +} // namespace gvisor -- cgit v1.2.3 From 463f4217d109ded8af758fe51a5daf8670da9794 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Fri, 28 Feb 2020 12:28:10 -0800 Subject: Make pipe buffer implementation standard. A follow-up change will convert the networking code to use this standard pipe implementation. PiperOrigin-RevId: 297903206 --- pkg/buffer/BUILD | 39 ++++ pkg/buffer/buffer.go | 67 ++++++ pkg/buffer/safemem.go | 131 ++++++++++++ pkg/buffer/view.go | 382 ++++++++++++++++++++++++++++++++++ pkg/buffer/view_test.go | 233 +++++++++++++++++++++ pkg/buffer/view_unsafe.go | 25 +++ pkg/sentry/kernel/pipe/BUILD | 18 +- pkg/sentry/kernel/pipe/buffer.go | 115 ---------- pkg/sentry/kernel/pipe/buffer_test.go | 32 --- pkg/sentry/kernel/pipe/pipe.go | 118 ++--------- pkg/sentry/kernel/pipe/pipe_util.go | 25 +-- 11 files changed, 912 insertions(+), 273 deletions(-) create mode 100644 pkg/buffer/BUILD create mode 100644 pkg/buffer/buffer.go create mode 100644 pkg/buffer/safemem.go create mode 100644 pkg/buffer/view.go create mode 100644 pkg/buffer/view_test.go create mode 100644 pkg/buffer/view_unsafe.go delete mode 100644 pkg/sentry/kernel/pipe/buffer.go delete mode 100644 pkg/sentry/kernel/pipe/buffer_test.go (limited to 'pkg/sentry') diff --git a/pkg/buffer/BUILD b/pkg/buffer/BUILD new file mode 100644 index 000000000..a77a3beea --- /dev/null +++ b/pkg/buffer/BUILD @@ -0,0 +1,39 @@ +load("//tools:defs.bzl", "go_library", "go_test") +load("//tools/go_generics:defs.bzl", "go_template_instance") + +package(licenses = ["notice"]) + +go_template_instance( + name = "buffer_list", + out = "buffer_list.go", + package = "buffer", + prefix = "buffer", + template = "//pkg/ilist:generic_list", + types = { + "Element": "*Buffer", + "Linker": "*Buffer", + }, +) + +go_library( + name = "buffer", + srcs = [ + "buffer.go", + "buffer_list.go", + "safemem.go", + "view.go", + "view_unsafe.go", + ], + visibility = ["//visibility:public"], + deps = [ + "//pkg/log", + "//pkg/safemem", + ], +) + +go_test( + name = "buffer_test", + size = "small", + srcs = ["view_test.go"], + library = ":buffer", +) diff --git a/pkg/buffer/buffer.go b/pkg/buffer/buffer.go new file mode 100644 index 000000000..d5f64609b --- /dev/null +++ b/pkg/buffer/buffer.go @@ -0,0 +1,67 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package buffer provides the implementation of a buffer view. +package buffer + +import ( + "sync" +) + +const bufferSize = 8144 // See below. + +// Buffer encapsulates a queueable byte buffer. +// +// Note that the total size is slightly less than two pages. This is done +// intentionally to ensure that the buffer object aligns with runtime +// internals. We have no hard size or alignment requirements. This two page +// size will effectively minimize internal fragmentation, but still have a +// large enough chunk to limit excessive segmentation. +// +// +stateify savable +type Buffer struct { + data [bufferSize]byte + read int + write int + bufferEntry +} + +// Reset resets internal data. +// +// This must be called before use. +func (b *Buffer) Reset() { + b.read = 0 + b.write = 0 +} + +// Empty indicates the buffer is empty. +// +// This indicates there is no data left to read. +func (b *Buffer) Empty() bool { + return b.read == b.write +} + +// Full indicates the buffer is full. +// +// This indicates there is no capacity left to write. +func (b *Buffer) Full() bool { + return b.write == len(b.data) +} + +// bufferPool is a pool for buffers. +var bufferPool = sync.Pool{ + New: func() interface{} { + return new(Buffer) + }, +} diff --git a/pkg/buffer/safemem.go b/pkg/buffer/safemem.go new file mode 100644 index 000000000..071aaa488 --- /dev/null +++ b/pkg/buffer/safemem.go @@ -0,0 +1,131 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package buffer + +import ( + "io" + + "gvisor.dev/gvisor/pkg/safemem" +) + +// WriteBlock returns this buffer as a write Block. +func (b *Buffer) WriteBlock() safemem.Block { + return safemem.BlockFromSafeSlice(b.data[b.write:]) +} + +// ReadBlock returns this buffer as a read Block. +func (b *Buffer) ReadBlock() safemem.Block { + return safemem.BlockFromSafeSlice(b.data[b.read:b.write]) +} + +// WriteFromBlocks implements safemem.Writer.WriteFromBlocks. +// +// This will advance the write index. +func (v *View) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) { + need := int(srcs.NumBytes()) + if need == 0 { + return 0, nil + } + + var ( + dst safemem.BlockSeq + blocks []safemem.Block + ) + + // Need at least one buffer. + firstBuf := v.data.Back() + if firstBuf == nil { + firstBuf = bufferPool.Get().(*Buffer) + v.data.PushBack(firstBuf) + } + + // Does the last block have sufficient capacity alone? + if l := len(firstBuf.data) - firstBuf.write; l >= need { + dst = safemem.BlockSeqOf(firstBuf.WriteBlock()) + } else { + // Append blocks until sufficient. + need -= l + blocks = append(blocks, firstBuf.WriteBlock()) + for need > 0 { + emptyBuf := bufferPool.Get().(*Buffer) + v.data.PushBack(emptyBuf) + need -= len(emptyBuf.data) // Full block. + blocks = append(blocks, emptyBuf.WriteBlock()) + } + dst = safemem.BlockSeqFromSlice(blocks) + } + + // Perform the copy. + n, err := safemem.CopySeq(dst, srcs) + v.size += int64(n) + + // Update all indices. + for left := int(n); left > 0; firstBuf = firstBuf.Next() { + if l := len(firstBuf.data) - firstBuf.write; left >= l { + firstBuf.write += l // Whole block. + left -= l + } else { + firstBuf.write += left // Partial block. + left = 0 + } + } + + return n, err +} + +// ReadToBlocks implements safemem.Reader.ReadToBlocks. +// +// This will not advance the read index; the caller should follow +// this call with a call to TrimFront in order to remove the read +// data from the buffer. This is done to support pipe sematics. +func (v *View) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) { + need := int(dsts.NumBytes()) + if need == 0 { + return 0, nil + } + + var ( + src safemem.BlockSeq + blocks []safemem.Block + ) + + firstBuf := v.data.Front() + if firstBuf == nil { + return 0, io.EOF + } + + // Is all the data in a single block? + if l := firstBuf.write - firstBuf.read; l >= need { + src = safemem.BlockSeqOf(firstBuf.ReadBlock()) + } else { + // Build a list of all the buffers. + need -= l + blocks = append(blocks, firstBuf.ReadBlock()) + for buf := firstBuf.Next(); buf != nil && need > 0; buf = buf.Next() { + need -= buf.write - buf.read + blocks = append(blocks, buf.ReadBlock()) + } + src = safemem.BlockSeqFromSlice(blocks) + } + + // Perform the copy. + n, err := safemem.CopySeq(dsts, src) + + // See above: we would normally advance the read index here, but we + // don't do that in order to support pipe semantics. We rely on a + // separate call to TrimFront() in this case. + + return n, err +} diff --git a/pkg/buffer/view.go b/pkg/buffer/view.go new file mode 100644 index 000000000..00fc11e9c --- /dev/null +++ b/pkg/buffer/view.go @@ -0,0 +1,382 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package buffer + +import ( + "fmt" + "io" +) + +// View is a non-linear buffer. +// +// All methods are thread compatible. +// +// +stateify savable +type View struct { + data bufferList + size int64 +} + +// TrimFront removes the first count bytes from the buffer. +func (v *View) TrimFront(count int64) { + if count >= v.size { + v.advanceRead(v.size) + } else { + v.advanceRead(count) + } +} + +// Read implements io.Reader.Read. +// +// Note that reading does not advance the read index. This must be done +// manually using TrimFront or other methods. +func (v *View) Read(p []byte) (int, error) { + return v.ReadAt(p, 0) +} + +// ReadAt implements io.ReaderAt.ReadAt. +func (v *View) ReadAt(p []byte, offset int64) (int, error) { + var ( + skipped int64 + done int64 + ) + for buf := v.data.Front(); buf != nil && done < int64(len(p)); buf = buf.Next() { + needToSkip := int(offset - skipped) + if l := buf.write - buf.read; l <= needToSkip { + skipped += int64(l) + continue + } + + // Actually read data. + n := copy(p[done:], buf.data[buf.read+needToSkip:buf.write]) + skipped += int64(needToSkip) + done += int64(n) + } + if int(done) < len(p) { + return int(done), io.EOF + } + return int(done), nil +} + +// Write implements io.Writer.Write. +func (v *View) Write(p []byte) (int, error) { + v.Append(p) // Does not fail. + return len(p), nil +} + +// advanceRead advances the view's read index. +// +// Precondition: there must be sufficient bytes in the buffer. +func (v *View) advanceRead(count int64) { + for buf := v.data.Front(); buf != nil && count > 0; { + l := int64(buf.write - buf.read) + if l > count { + // There is still data for reading. + buf.read += int(count) + v.size -= count + count = 0 + break + } + + // Read from this buffer. + buf.read += int(l) + count -= l + v.size -= l + + // When all data has been read from a buffer, we push + // it into the empty buffer pool for reuse. + oldBuf := buf + buf = buf.Next() // Iterate. + v.data.Remove(oldBuf) + oldBuf.Reset() + bufferPool.Put(oldBuf) + } + if count > 0 { + panic(fmt.Sprintf("advanceRead still has %d bytes remaining", count)) + } +} + +// Truncate truncates the view to the given bytes. +func (v *View) Truncate(length int64) { + if length < 0 || length >= v.size { + return // Nothing to do. + } + for buf := v.data.Back(); buf != nil && v.size > length; buf = v.data.Back() { + l := int64(buf.write - buf.read) // Local bytes. + switch { + case v.size-l >= length: + // Drop the buffer completely; see above. + v.data.Remove(buf) + v.size -= l + buf.Reset() + bufferPool.Put(buf) + + case v.size > length && v.size-l < length: + // Just truncate the buffer locally. + delta := (length - (v.size - l)) + buf.write = buf.read + int(delta) + v.size = length + + default: + // Should never happen. + panic("invalid buffer during truncation") + } + } + v.size = length // Save the new size. +} + +// Grow grows the given view to the number of bytes. If zero +// is true, all these bytes will be zero. If zero is false, +// then this is the caller's responsibility. +// +// Precondition: length must be >= 0. +func (v *View) Grow(length int64, zero bool) { + if length < 0 { + panic("negative length provided") + } + for v.size < length { + buf := v.data.Back() + + // Is there at least one buffer? + if buf == nil || buf.Full() { + buf = bufferPool.Get().(*Buffer) + v.data.PushBack(buf) + } + + // Write up to length bytes. + l := len(buf.data) - buf.write + if int64(l) > length-v.size { + l = int(length - v.size) + } + + // Zero the written section; note that this pattern is + // specifically recognized and optimized by the compiler. + if zero { + for i := buf.write; i < buf.write+l; i++ { + buf.data[i] = 0 + } + } + + // Advance the index. + buf.write += l + v.size += int64(l) + } +} + +// Prepend prepends the given data. +func (v *View) Prepend(data []byte) { + // Is there any space in the first buffer? + if buf := v.data.Front(); buf != nil && buf.read > 0 { + // Fill up before the first write. + avail := buf.read + copy(buf.data[0:], data[len(data)-avail:]) + data = data[:len(data)-avail] + v.size += int64(avail) + } + + for len(data) > 0 { + // Do we need an empty buffer? + buf := bufferPool.Get().(*Buffer) + v.data.PushFront(buf) + + // The buffer is empty; copy last chunk. + start := len(data) - len(buf.data) + if start < 0 { + start = 0 // Everything. + } + + // We have to put the data at the end of the current + // buffer in order to ensure that the next prepend will + // correctly fill up the beginning of this buffer. + bStart := len(buf.data) - len(data[start:]) + n := copy(buf.data[bStart:], data[start:]) + buf.read = bStart + buf.write = len(buf.data) + data = data[:start] + v.size += int64(n) + } +} + +// Append appends the given data. +func (v *View) Append(data []byte) { + for done := 0; done < len(data); { + buf := v.data.Back() + + // Find the first empty buffer. + if buf == nil || buf.Full() { + buf = bufferPool.Get().(*Buffer) + v.data.PushBack(buf) + } + + // Copy in to the given buffer. + n := copy(buf.data[buf.write:], data[done:]) + done += n + buf.write += n + v.size += int64(n) + } +} + +// Flatten returns a flattened copy of this data. +// +// This method should not be used in any performance-sensitive paths. It may +// allocate a fresh byte slice sufficiently large to contain all the data in +// the buffer. +// +// N.B. Tee data still belongs to this view, as if there is a single buffer +// present, then it will be returned directly. This should be used for +// temporary use only, and a reference to the given slice should not be held. +func (v *View) Flatten() []byte { + if buf := v.data.Front(); buf.Next() == nil { + return buf.data[buf.read:buf.write] // Only one buffer. + } + data := make([]byte, 0, v.size) // Need to flatten. + for buf := v.data.Front(); buf != nil; buf = buf.Next() { + // Copy to the allocated slice. + data = append(data, buf.data[buf.read:buf.write]...) + } + return data +} + +// Size indicates the total amount of data available in this view. +func (v *View) Size() (sz int64) { + sz = v.size // Pre-calculated. + return sz +} + +// Copy makes a strict copy of this view. +func (v *View) Copy() (other View) { + for buf := v.data.Front(); buf != nil; buf = buf.Next() { + other.Append(buf.data[buf.read:buf.write]) + } + return other +} + +// Apply applies the given function across all valid data. +func (v *View) Apply(fn func([]byte)) { + for buf := v.data.Front(); buf != nil; buf = buf.Next() { + if l := int64(buf.write - buf.read); l > 0 { + fn(buf.data[buf.read:buf.write]) + } + } +} + +// Merge merges the provided View with this one. +// +// The other view will be empty after this operation. +func (v *View) Merge(other *View) { + // Copy over all buffers. + for buf := other.data.Front(); buf != nil && !buf.Empty(); buf = other.data.Front() { + other.data.Remove(buf) + v.data.PushBack(buf) + } + + // Adjust sizes. + v.size += other.size + other.size = 0 +} + +// WriteFromReader writes to the buffer from an io.Reader. +func (v *View) WriteFromReader(r io.Reader, count int64) (int64, error) { + var ( + done int64 + n int + err error + ) + for done < count { + buf := v.data.Back() + + // Find the first empty buffer. + if buf == nil || buf.Full() { + buf = bufferPool.Get().(*Buffer) + v.data.PushBack(buf) + } + + // Is this less than the minimum batch? + if len(buf.data[buf.write:]) < minBatch && (count-done) >= int64(minBatch) { + tmp := make([]byte, minBatch) + n, err = r.Read(tmp) + v.Write(tmp[:n]) + done += int64(n) + if err != nil { + break + } + continue + } + + // Limit the read, if necessary. + end := len(buf.data) + if int64(end-buf.write) > (count - done) { + end = buf.write + int(count-done) + } + + // Pass the relevant portion of the buffer. + n, err = r.Read(buf.data[buf.write:end]) + buf.write += n + done += int64(n) + v.size += int64(n) + if err == io.EOF { + err = nil // Short write allowed. + break + } else if err != nil { + break + } + } + return done, err +} + +// ReadToWriter reads from the buffer into an io.Writer. +// +// N.B. This does not consume the bytes read. TrimFront should +// be called appropriately after this call in order to do so. +func (v *View) ReadToWriter(w io.Writer, count int64) (int64, error) { + var ( + done int64 + n int + err error + ) + offset := 0 // Spill-over for batching. + for buf := v.data.Front(); buf != nil && done < count; buf = buf.Next() { + l := buf.write - buf.read - offset + + // Is this less than the minimum batch? + if l < minBatch && (count-done) >= int64(minBatch) && (v.size-done) >= int64(minBatch) { + tmp := make([]byte, minBatch) + n, err = v.ReadAt(tmp, done) + w.Write(tmp[:n]) + done += int64(n) + offset = n - l // Reset below. + if err != nil { + break + } + continue + } + + // Limit the write if necessary. + if int64(l) >= (count - done) { + l = int(count - done) + } + + // Perform the actual write. + n, err = w.Write(buf.data[buf.read+offset : buf.read+offset+l]) + done += int64(n) + if err != nil { + break + } + + // Reset spill-over. + offset = 0 + } + return done, err +} diff --git a/pkg/buffer/view_test.go b/pkg/buffer/view_test.go new file mode 100644 index 000000000..37e652f16 --- /dev/null +++ b/pkg/buffer/view_test.go @@ -0,0 +1,233 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package buffer + +import ( + "bytes" + "strings" + "testing" +) + +func TestView(t *testing.T) { + testCases := []struct { + name string + input string + output string + ops []func(*View) + }{ + // Prepend. + { + name: "prepend", + input: "world", + ops: []func(*View){ + func(v *View) { + v.Prepend([]byte("hello ")) + }, + }, + output: "hello world", + }, + { + name: "prepend fill", + input: strings.Repeat("1", bufferSize-1), + ops: []func(*View){ + func(v *View) { + v.Prepend([]byte("0")) + }, + }, + output: "0" + strings.Repeat("1", bufferSize-1), + }, + { + name: "prepend overflow", + input: strings.Repeat("1", bufferSize), + ops: []func(*View){ + func(v *View) { + v.Prepend([]byte("0")) + }, + }, + output: "0" + strings.Repeat("1", bufferSize), + }, + { + name: "prepend multiple buffers", + input: strings.Repeat("1", bufferSize-1), + ops: []func(*View){ + func(v *View) { + v.Prepend([]byte(strings.Repeat("0", bufferSize*3))) + }, + }, + output: strings.Repeat("0", bufferSize*3) + strings.Repeat("1", bufferSize-1), + }, + + // Append. + { + name: "append", + input: "hello", + ops: []func(*View){ + func(v *View) { + v.Append([]byte(" world")) + }, + }, + output: "hello world", + }, + { + name: "append fill", + input: strings.Repeat("1", bufferSize-1), + ops: []func(*View){ + func(v *View) { + v.Append([]byte("0")) + }, + }, + output: strings.Repeat("1", bufferSize-1) + "0", + }, + { + name: "append overflow", + input: strings.Repeat("1", bufferSize), + ops: []func(*View){ + func(v *View) { + v.Append([]byte("0")) + }, + }, + output: strings.Repeat("1", bufferSize) + "0", + }, + { + name: "append multiple buffers", + input: strings.Repeat("1", bufferSize-1), + ops: []func(*View){ + func(v *View) { + v.Append([]byte(strings.Repeat("0", bufferSize*3))) + }, + }, + output: strings.Repeat("1", bufferSize-1) + strings.Repeat("0", bufferSize*3), + }, + + // Truncate. + { + name: "truncate", + input: "hello world", + ops: []func(*View){ + func(v *View) { + v.Truncate(5) + }, + }, + output: "hello", + }, + { + name: "truncate multiple buffers", + input: strings.Repeat("1", bufferSize*2), + ops: []func(*View){ + func(v *View) { + v.Truncate(bufferSize*2 - 1) + }, + }, + output: strings.Repeat("1", bufferSize*2-1), + }, + { + name: "truncate multiple buffers to one buffer", + input: strings.Repeat("1", bufferSize*2), + ops: []func(*View){ + func(v *View) { + v.Truncate(5) + }, + }, + output: "11111", + }, + + // TrimFront. + { + name: "trim", + input: "hello world", + ops: []func(*View){ + func(v *View) { + v.TrimFront(6) + }, + }, + output: "world", + }, + { + name: "trim multiple buffers", + input: strings.Repeat("1", bufferSize*2), + ops: []func(*View){ + func(v *View) { + v.TrimFront(1) + }, + }, + output: strings.Repeat("1", bufferSize*2-1), + }, + { + name: "trim multiple buffers to one buffer", + input: strings.Repeat("1", bufferSize*2), + ops: []func(*View){ + func(v *View) { + v.TrimFront(bufferSize*2 - 1) + }, + }, + output: "1", + }, + + // Grow. + { + name: "grow", + input: "hello world", + ops: []func(*View){ + func(v *View) { + v.Grow(1, true) + }, + }, + output: "hello world", + }, + { + name: "grow from zero", + ops: []func(*View){ + func(v *View) { + v.Grow(1024, true) + }, + }, + output: strings.Repeat("\x00", 1024), + }, + { + name: "grow from non-zero", + input: strings.Repeat("1", bufferSize), + ops: []func(*View){ + func(v *View) { + v.Grow(bufferSize*2, true) + }, + }, + output: strings.Repeat("1", bufferSize) + strings.Repeat("\x00", bufferSize), + }, + } + + for _, tc := range testCases { + t.Run(tc.name, func(t *testing.T) { + // Construct the new view. + var view View + view.Append([]byte(tc.input)) + + // Run all operations. + for _, op := range tc.ops { + op(&view) + } + + // Flatten and validate. + out := view.Flatten() + if !bytes.Equal([]byte(tc.output), out) { + t.Errorf("expected %q, got %q", tc.output, string(out)) + } + + // Ensure the size is correct. + if len(out) != int(view.Size()) { + t.Errorf("size is wrong: expected %d, got %d", len(out), view.Size()) + } + }) + } +} diff --git a/pkg/buffer/view_unsafe.go b/pkg/buffer/view_unsafe.go new file mode 100644 index 000000000..d1ef39b26 --- /dev/null +++ b/pkg/buffer/view_unsafe.go @@ -0,0 +1,25 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package buffer + +import ( + "unsafe" +) + +// minBatch is the smallest Read or Write operation that the +// WriteFromReader and ReadToWriter functions will use. +// +// This is defined as the size of a native pointer. +const minBatch = int(unsafe.Sizeof(uintptr(0))) diff --git a/pkg/sentry/kernel/pipe/BUILD b/pkg/sentry/kernel/pipe/BUILD index 4c049d5b4..f29dc0472 100644 --- a/pkg/sentry/kernel/pipe/BUILD +++ b/pkg/sentry/kernel/pipe/BUILD @@ -1,25 +1,10 @@ load("//tools:defs.bzl", "go_library", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") package(licenses = ["notice"]) -go_template_instance( - name = "buffer_list", - out = "buffer_list.go", - package = "pipe", - prefix = "buffer", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*buffer", - "Linker": "*buffer", - }, -) - go_library( name = "pipe", srcs = [ - "buffer.go", - "buffer_list.go", "device.go", "node.go", "pipe.go", @@ -33,8 +18,8 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/amutex", + "//pkg/buffer", "//pkg/context", - "//pkg/safemem", "//pkg/sentry/arch", "//pkg/sentry/device", "//pkg/sentry/fs", @@ -51,7 +36,6 @@ go_test( name = "pipe_test", size = "small", srcs = [ - "buffer_test.go", "node_test.go", "pipe_test.go", ], diff --git a/pkg/sentry/kernel/pipe/buffer.go b/pkg/sentry/kernel/pipe/buffer.go deleted file mode 100644 index fe3be5dbd..000000000 --- a/pkg/sentry/kernel/pipe/buffer.go +++ /dev/null @@ -1,115 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package pipe - -import ( - "io" - - "gvisor.dev/gvisor/pkg/safemem" - "gvisor.dev/gvisor/pkg/sync" -) - -// buffer encapsulates a queueable byte buffer. -// -// Note that the total size is slightly less than two pages. This -// is done intentionally to ensure that the buffer object aligns -// with runtime internals. We have no hard size or alignment -// requirements. This two page size will effectively minimize -// internal fragmentation, but still have a large enough chunk -// to limit excessive segmentation. -// -// +stateify savable -type buffer struct { - data [8144]byte - read int - write int - bufferEntry -} - -// Reset resets internal data. -// -// This must be called before use. -func (b *buffer) Reset() { - b.read = 0 - b.write = 0 -} - -// Empty indicates the buffer is empty. -// -// This indicates there is no data left to read. -func (b *buffer) Empty() bool { - return b.read == b.write -} - -// Full indicates the buffer is full. -// -// This indicates there is no capacity left to write. -func (b *buffer) Full() bool { - return b.write == len(b.data) -} - -// WriteFromBlocks implements safemem.Writer.WriteFromBlocks. -func (b *buffer) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) { - dst := safemem.BlockSeqOf(safemem.BlockFromSafeSlice(b.data[b.write:])) - n, err := safemem.CopySeq(dst, srcs) - b.write += int(n) - return n, err -} - -// WriteFromReader writes to the buffer from an io.Reader. -func (b *buffer) WriteFromReader(r io.Reader, count int64) (int64, error) { - dst := b.data[b.write:] - if count < int64(len(dst)) { - dst = b.data[b.write:][:count] - } - n, err := r.Read(dst) - b.write += n - return int64(n), err -} - -// ReadToBlocks implements safemem.Reader.ReadToBlocks. -func (b *buffer) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) { - src := safemem.BlockSeqOf(safemem.BlockFromSafeSlice(b.data[b.read:b.write])) - n, err := safemem.CopySeq(dsts, src) - b.read += int(n) - return n, err -} - -// ReadToWriter reads from the buffer into an io.Writer. -func (b *buffer) ReadToWriter(w io.Writer, count int64, dup bool) (int64, error) { - src := b.data[b.read:b.write] - if count < int64(len(src)) { - src = b.data[b.read:][:count] - } - n, err := w.Write(src) - if !dup { - b.read += n - } - return int64(n), err -} - -// bufferPool is a pool for buffers. -var bufferPool = sync.Pool{ - New: func() interface{} { - return new(buffer) - }, -} - -// newBuffer grabs a new buffer from the pool. -func newBuffer() *buffer { - b := bufferPool.Get().(*buffer) - b.Reset() - return b -} diff --git a/pkg/sentry/kernel/pipe/buffer_test.go b/pkg/sentry/kernel/pipe/buffer_test.go deleted file mode 100644 index 4d54b8b8f..000000000 --- a/pkg/sentry/kernel/pipe/buffer_test.go +++ /dev/null @@ -1,32 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package pipe - -import ( - "testing" - "unsafe" - - "gvisor.dev/gvisor/pkg/usermem" -) - -func TestBufferSize(t *testing.T) { - bufferSize := unsafe.Sizeof(buffer{}) - if bufferSize < usermem.PageSize { - t.Errorf("buffer is less than a page") - } - if bufferSize > (2 * usermem.PageSize) { - t.Errorf("buffer is greater than two pages") - } -} diff --git a/pkg/sentry/kernel/pipe/pipe.go b/pkg/sentry/kernel/pipe/pipe.go index 08410283f..725e9db7d 100644 --- a/pkg/sentry/kernel/pipe/pipe.go +++ b/pkg/sentry/kernel/pipe/pipe.go @@ -20,6 +20,7 @@ import ( "sync/atomic" "syscall" + "gvisor.dev/gvisor/pkg/buffer" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sync" @@ -70,10 +71,10 @@ type Pipe struct { // mu protects all pipe internal state below. mu sync.Mutex `state:"nosave"` - // data is the buffer queue of pipe contents. + // view is the underlying set of buffers. // // This is protected by mu. - data bufferList + view buffer.View // max is the maximum size of the pipe in bytes. When this max has been // reached, writers will get EWOULDBLOCK. @@ -81,11 +82,6 @@ type Pipe struct { // This is protected by mu. max int64 - // size is the current size of the pipe in bytes. - // - // This is protected by mu. - size int64 - // hadWriter indicates if this pipe ever had a writer. Note that this // does not necessarily indicate there is *currently* a writer, just // that there has been a writer at some point since the pipe was @@ -196,7 +192,7 @@ type readOps struct { limit func(int64) // read performs the actual read operation. - read func(*buffer) (int64, error) + read func(*buffer.View) (int64, error) } // read reads data from the pipe into dst and returns the number of bytes @@ -213,7 +209,7 @@ func (p *Pipe) read(ctx context.Context, ops readOps) (int64, error) { defer p.mu.Unlock() // Is the pipe empty? - if p.size == 0 { + if p.view.Size() == 0 { if !p.HasWriters() { // There are no writers, return EOF. return 0, nil @@ -222,71 +218,13 @@ func (p *Pipe) read(ctx context.Context, ops readOps) (int64, error) { } // Limit how much we consume. - if ops.left() > p.size { - ops.limit(p.size) + if ops.left() > p.view.Size() { + ops.limit(p.view.Size()) } - done := int64(0) - for ops.left() > 0 { - // Pop the first buffer. - first := p.data.Front() - if first == nil { - break - } - - // Copy user data. - n, err := ops.read(first) - done += int64(n) - p.size -= n - - // Empty buffer? - if first.Empty() { - // Push to the free list. - p.data.Remove(first) - bufferPool.Put(first) - } - - // Handle errors. - if err != nil { - return done, err - } - } - - return done, nil -} - -// dup duplicates all data from this pipe into the given writer. -// -// There is no blocking behavior implemented here. The writer may propagate -// some blocking error. All the writes must be complete writes. -func (p *Pipe) dup(ctx context.Context, ops readOps) (int64, error) { - p.mu.Lock() - defer p.mu.Unlock() - - // Is the pipe empty? - if p.size == 0 { - if !p.HasWriters() { - // See above. - return 0, nil - } - return 0, syserror.ErrWouldBlock - } - - // Limit how much we consume. - if ops.left() > p.size { - ops.limit(p.size) - } - - done := int64(0) - for buf := p.data.Front(); buf != nil; buf = buf.Next() { - n, err := ops.read(buf) - done += n - if err != nil { - return done, err - } - } - - return done, nil + // Copy user data; the read op is responsible for trimming. + done, err := ops.read(&p.view) + return done, err } type writeOps struct { @@ -297,7 +235,7 @@ type writeOps struct { limit func(int64) // write should write to the provided buffer. - write func(*buffer) (int64, error) + write func(*buffer.View) (int64, error) } // write writes data from sv into the pipe and returns the number of bytes @@ -317,33 +255,19 @@ func (p *Pipe) write(ctx context.Context, ops writeOps) (int64, error) { // POSIX requires that a write smaller than atomicIOBytes (PIPE_BUF) be // atomic, but requires no atomicity for writes larger than this. wanted := ops.left() - if avail := p.max - p.size; wanted > avail { + if avail := p.max - p.view.Size(); wanted > avail { if wanted <= p.atomicIOBytes { return 0, syserror.ErrWouldBlock } ops.limit(avail) } - done := int64(0) - for ops.left() > 0 { - // Need a new buffer? - last := p.data.Back() - if last == nil || last.Full() { - // Add a new buffer to the data list. - last = newBuffer() - p.data.PushBack(last) - } - - // Copy user data. - n, err := ops.write(last) - done += int64(n) - p.size += n - - // Handle errors. - if err != nil { - return done, err - } + // Copy user data. + done, err := ops.write(&p.view) + if err != nil { + return done, err } + if wanted > done { // Partial write due to full pipe. return done, syserror.ErrWouldBlock @@ -396,7 +320,7 @@ func (p *Pipe) HasWriters() bool { // Precondition: mu must be held. func (p *Pipe) rReadinessLocked() waiter.EventMask { ready := waiter.EventMask(0) - if p.HasReaders() && p.data.Front() != nil { + if p.HasReaders() && p.view.Size() != 0 { ready |= waiter.EventIn } if !p.HasWriters() && p.hadWriter { @@ -422,7 +346,7 @@ func (p *Pipe) rReadiness() waiter.EventMask { // Precondition: mu must be held. func (p *Pipe) wReadinessLocked() waiter.EventMask { ready := waiter.EventMask(0) - if p.HasWriters() && p.size < p.max { + if p.HasWriters() && p.view.Size() < p.max { ready |= waiter.EventOut } if !p.HasReaders() { @@ -451,7 +375,7 @@ func (p *Pipe) rwReadiness() waiter.EventMask { func (p *Pipe) queued() int64 { p.mu.Lock() defer p.mu.Unlock() - return p.size + return p.view.Size() } // FifoSize implements fs.FifoSizer.FifoSize. @@ -474,7 +398,7 @@ func (p *Pipe) SetFifoSize(size int64) (int64, error) { } p.mu.Lock() defer p.mu.Unlock() - if size < p.size { + if size < p.view.Size() { return 0, syserror.EBUSY } p.max = size diff --git a/pkg/sentry/kernel/pipe/pipe_util.go b/pkg/sentry/kernel/pipe/pipe_util.go index 80158239e..5a1d4fd57 100644 --- a/pkg/sentry/kernel/pipe/pipe_util.go +++ b/pkg/sentry/kernel/pipe/pipe_util.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/amutex" + "gvisor.dev/gvisor/pkg/buffer" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sync" @@ -49,9 +50,10 @@ func (p *Pipe) Read(ctx context.Context, dst usermem.IOSequence) (int64, error) limit: func(l int64) { dst = dst.TakeFirst64(l) }, - read: func(buf *buffer) (int64, error) { - n, err := dst.CopyOutFrom(ctx, buf) + read: func(view *buffer.View) (int64, error) { + n, err := dst.CopyOutFrom(ctx, view) dst = dst.DropFirst64(n) + view.TrimFront(n) return n, err }, }) @@ -70,16 +72,15 @@ func (p *Pipe) WriteTo(ctx context.Context, w io.Writer, count int64, dup bool) limit: func(l int64) { count = l }, - read: func(buf *buffer) (int64, error) { - n, err := buf.ReadToWriter(w, count, dup) + read: func(view *buffer.View) (int64, error) { + n, err := view.ReadToWriter(w, count) + if !dup { + view.TrimFront(n) + } count -= n return n, err }, } - if dup { - // There is no notification for dup operations. - return p.dup(ctx, ops) - } n, err := p.read(ctx, ops) if n > 0 { p.Notify(waiter.EventOut) @@ -96,8 +97,8 @@ func (p *Pipe) Write(ctx context.Context, src usermem.IOSequence) (int64, error) limit: func(l int64) { src = src.TakeFirst64(l) }, - write: func(buf *buffer) (int64, error) { - n, err := src.CopyInTo(ctx, buf) + write: func(view *buffer.View) (int64, error) { + n, err := src.CopyInTo(ctx, view) src = src.DropFirst64(n) return n, err }, @@ -117,8 +118,8 @@ func (p *Pipe) ReadFrom(ctx context.Context, r io.Reader, count int64) (int64, e limit: func(l int64) { count = l }, - write: func(buf *buffer) (int64, error) { - n, err := buf.WriteFromReader(r, count) + write: func(view *buffer.View) (int64, error) { + n, err := view.WriteFromReader(r, count) count -= n return n, err }, -- cgit v1.2.3 From 837cf6255120fe4577a9fd758ecbe2f52f268ba8 Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Fri, 14 Feb 2020 11:58:07 -0800 Subject: pcids.go isn't arch-specific Signed-off-by: Andrei Vagin --- pkg/sentry/platform/ring0/pagetables/BUILD | 2 +- pkg/sentry/platform/ring0/pagetables/pcids.go | 107 +++++++++++++++++++++ pkg/sentry/platform/ring0/pagetables/pcids_x86.go | 109 ---------------------- 3 files changed, 108 insertions(+), 110 deletions(-) create mode 100644 pkg/sentry/platform/ring0/pagetables/pcids.go delete mode 100644 pkg/sentry/platform/ring0/pagetables/pcids_x86.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/ring0/pagetables/BUILD b/pkg/sentry/platform/ring0/pagetables/BUILD index 4f2406ce3..581841555 100644 --- a/pkg/sentry/platform/ring0/pagetables/BUILD +++ b/pkg/sentry/platform/ring0/pagetables/BUILD @@ -80,7 +80,7 @@ go_library( "pagetables_amd64.go", "pagetables_arm64.go", "pagetables_x86.go", - "pcids_x86.go", + "pcids.go", "walker_amd64.go", "walker_arm64.go", "walker_empty.go", diff --git a/pkg/sentry/platform/ring0/pagetables/pcids.go b/pkg/sentry/platform/ring0/pagetables/pcids.go new file mode 100644 index 000000000..9206030bf --- /dev/null +++ b/pkg/sentry/platform/ring0/pagetables/pcids.go @@ -0,0 +1,107 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package pagetables + +import ( + "gvisor.dev/gvisor/pkg/sync" +) + +// limitPCID is the number of valid PCIDs. +const limitPCID = 4096 + +// PCIDs is a simple PCID database. +// +// This is not protected by locks and is thus suitable for use only with a +// single CPU at a time. +type PCIDs struct { + // mu protects below. + mu sync.Mutex + + // cache are the assigned page tables. + cache map[*PageTables]uint16 + + // avail are available PCIDs. + avail []uint16 +} + +// NewPCIDs returns a new PCID database. +// +// start is the first index to assign. Typically this will be one, as the zero +// pcid will always be flushed on transition (see pagetables_x86.go). This may +// be more than one if specific PCIDs are reserved. +// +// Nil is returned iff the start and size are out of range. +func NewPCIDs(start, size uint16) *PCIDs { + if start+uint16(size) >= limitPCID { + return nil // See comment. + } + p := &PCIDs{ + cache: make(map[*PageTables]uint16), + } + for pcid := start; pcid < start+size; pcid++ { + p.avail = append(p.avail, pcid) + } + return p +} + +// Assign assigns a PCID to the given PageTables. +// +// This may overwrite any previous assignment provided. If this in the case, +// true is returned to indicate that the PCID should be flushed. +func (p *PCIDs) Assign(pt *PageTables) (uint16, bool) { + p.mu.Lock() + if pcid, ok := p.cache[pt]; ok { + p.mu.Unlock() + return pcid, false // No flush. + } + + // Is there something available? + if len(p.avail) > 0 { + pcid := p.avail[len(p.avail)-1] + p.avail = p.avail[:len(p.avail)-1] + p.cache[pt] = pcid + + // We need to flush because while this is in the available + // pool, it may have been used previously. + p.mu.Unlock() + return pcid, true + } + + // Evict an existing table. + for old, pcid := range p.cache { + delete(p.cache, old) + p.cache[pt] = pcid + + // A flush is definitely required in this case, these page + // tables may still be active. (They will just be assigned some + // other PCID if and when they hit the given CPU again.) + p.mu.Unlock() + return pcid, true + } + + // No PCID. + p.mu.Unlock() + return 0, false +} + +// Drop drops references to a set of page tables. +func (p *PCIDs) Drop(pt *PageTables) { + p.mu.Lock() + if pcid, ok := p.cache[pt]; ok { + delete(p.cache, pt) + p.avail = append(p.avail, pcid) + } + p.mu.Unlock() +} diff --git a/pkg/sentry/platform/ring0/pagetables/pcids_x86.go b/pkg/sentry/platform/ring0/pagetables/pcids_x86.go deleted file mode 100644 index e199bae18..000000000 --- a/pkg/sentry/platform/ring0/pagetables/pcids_x86.go +++ /dev/null @@ -1,109 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build i386 amd64 - -package pagetables - -import ( - "gvisor.dev/gvisor/pkg/sync" -) - -// limitPCID is the number of valid PCIDs. -const limitPCID = 4096 - -// PCIDs is a simple PCID database. -// -// This is not protected by locks and is thus suitable for use only with a -// single CPU at a time. -type PCIDs struct { - // mu protects below. - mu sync.Mutex - - // cache are the assigned page tables. - cache map[*PageTables]uint16 - - // avail are available PCIDs. - avail []uint16 -} - -// NewPCIDs returns a new PCID database. -// -// start is the first index to assign. Typically this will be one, as the zero -// pcid will always be flushed on transition (see pagetables_x86.go). This may -// be more than one if specific PCIDs are reserved. -// -// Nil is returned iff the start and size are out of range. -func NewPCIDs(start, size uint16) *PCIDs { - if start+uint16(size) >= limitPCID { - return nil // See comment. - } - p := &PCIDs{ - cache: make(map[*PageTables]uint16), - } - for pcid := start; pcid < start+size; pcid++ { - p.avail = append(p.avail, pcid) - } - return p -} - -// Assign assigns a PCID to the given PageTables. -// -// This may overwrite any previous assignment provided. If this in the case, -// true is returned to indicate that the PCID should be flushed. -func (p *PCIDs) Assign(pt *PageTables) (uint16, bool) { - p.mu.Lock() - if pcid, ok := p.cache[pt]; ok { - p.mu.Unlock() - return pcid, false // No flush. - } - - // Is there something available? - if len(p.avail) > 0 { - pcid := p.avail[len(p.avail)-1] - p.avail = p.avail[:len(p.avail)-1] - p.cache[pt] = pcid - - // We need to flush because while this is in the available - // pool, it may have been used previously. - p.mu.Unlock() - return pcid, true - } - - // Evict an existing table. - for old, pcid := range p.cache { - delete(p.cache, old) - p.cache[pt] = pcid - - // A flush is definitely required in this case, these page - // tables may still be active. (They will just be assigned some - // other PCID if and when they hit the given CPU again.) - p.mu.Unlock() - return pcid, true - } - - // No PCID. - p.mu.Unlock() - return 0, false -} - -// Drop drops references to a set of page tables. -func (p *PCIDs) Drop(pt *PageTables) { - p.mu.Lock() - if pcid, ok := p.cache[pt]; ok { - delete(p.cache, pt) - p.avail = append(p.avail, pcid) - } - p.mu.Unlock() -} -- cgit v1.2.3 From 413a9b7fdc14f8bff660e1988e3ef0355dd4e6c6 Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Fri, 21 Feb 2020 11:00:11 -0800 Subject: Define CPUIDInstruction for arm64 There is no cpuid instruction on arm64, so we need to defined it just to avoid a compile time error. Signed-off-by: Andrei Vagin --- pkg/sentry/arch/arch_arm64.go | 5 +++++ pkg/sentry/kernel/task_run.go | 41 ++++++++++++++++++++++++++++------------- 2 files changed, 33 insertions(+), 13 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/arch/arch_arm64.go b/pkg/sentry/arch/arch_arm64.go index 372b650b9..885115ae2 100644 --- a/pkg/sentry/arch/arch_arm64.go +++ b/pkg/sentry/arch/arch_arm64.go @@ -53,6 +53,11 @@ const ( preferredPIELoadAddr usermem.Addr = maxAddr64 / 6 * 5 ) +var ( + // CPUIDInstruction doesn't exist on ARM64. + CPUIDInstruction = []byte{} +) + // These constants are selected as heuristics to help make the Platform's // potentially limited address space conform as closely to Linux as possible. const ( diff --git a/pkg/sentry/kernel/task_run.go b/pkg/sentry/kernel/task_run.go index 5568c91bc..799cbcd93 100644 --- a/pkg/sentry/kernel/task_run.go +++ b/pkg/sentry/kernel/task_run.go @@ -126,13 +126,39 @@ func (t *Task) doStop() { } } +func (*runApp) handleCPUIDInstruction(t *Task) error { + if len(arch.CPUIDInstruction) == 0 { + // CPUID emulation isn't supported, but this code can be + // executed, because the ptrace platform returns + // ErrContextSignalCPUID on page faults too. Look at + // pkg/sentry/platform/ptrace/ptrace.go:context.Switch for more + // details. + return platform.ErrContextSignal + } + // Is this a CPUID instruction? + region := trace.StartRegion(t.traceContext, cpuidRegion) + expected := arch.CPUIDInstruction[:] + found := make([]byte, len(expected)) + _, err := t.CopyIn(usermem.Addr(t.Arch().IP()), &found) + if err == nil && bytes.Equal(expected, found) { + // Skip the cpuid instruction. + t.Arch().CPUIDEmulate(t) + t.Arch().SetIP(t.Arch().IP() + uintptr(len(expected))) + region.End() + + return nil + } + region.End() // Not an actual CPUID, but required copy-in. + return platform.ErrContextSignal +} + // The runApp state checks for interrupts before executing untrusted // application code. // // +stateify savable type runApp struct{} -func (*runApp) execute(t *Task) taskRunState { +func (app *runApp) execute(t *Task) taskRunState { if t.interrupted() { // Checkpointing instructs tasks to stop by sending an interrupt, so we // must check for stops before entering runInterrupt (instead of @@ -237,21 +263,10 @@ func (*runApp) execute(t *Task) taskRunState { return (*runApp)(nil) case platform.ErrContextSignalCPUID: - // Is this a CPUID instruction? - region := trace.StartRegion(t.traceContext, cpuidRegion) - expected := arch.CPUIDInstruction[:] - found := make([]byte, len(expected)) - _, err := t.CopyIn(usermem.Addr(t.Arch().IP()), &found) - if err == nil && bytes.Equal(expected, found) { - // Skip the cpuid instruction. - t.Arch().CPUIDEmulate(t) - t.Arch().SetIP(t.Arch().IP() + uintptr(len(expected))) - region.End() - + if err := app.handleCPUIDInstruction(t); err == nil { // Resume execution. return (*runApp)(nil) } - region.End() // Not an actual CPUID, but required copy-in. // The instruction at the given RIP was not a CPUID, and we // fallthrough to the default signal deliver behavior below. -- cgit v1.2.3 From ab7ecdd66d2aa4835bbe655ba672bd30813a2a88 Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Fri, 28 Feb 2020 17:53:25 -0800 Subject: watchdog: print panic error message before other messages This is needed for syzkaller to proper classify issues. Right now, all watchdog issues are duped to one with the subject "panic: Sentry detected stuck task(s). See stack trace and message above for more details". PiperOrigin-RevId: 297975363 --- pkg/sentry/watchdog/watchdog.go | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/watchdog/watchdog.go b/pkg/sentry/watchdog/watchdog.go index bfb2fac26..f7d6009a0 100644 --- a/pkg/sentry/watchdog/watchdog.go +++ b/pkg/sentry/watchdog/watchdog.go @@ -221,7 +221,7 @@ func (w *Watchdog) waitForStart() { return } var buf bytes.Buffer - buf.WriteString("Watchdog.Start() not called within %s:\n") + buf.WriteString(fmt.Sprintf("Watchdog.Start() not called within %s", w.StartupTimeout)) w.doAction(w.StartupTimeoutAction, false, &buf) } @@ -325,7 +325,7 @@ func (w *Watchdog) report(offenders map[*kernel.Task]*offender, newTaskFound boo func (w *Watchdog) reportStuckWatchdog() { var buf bytes.Buffer - buf.WriteString("Watchdog goroutine is stuck:\n") + buf.WriteString("Watchdog goroutine is stuck:") w.doAction(w.TaskTimeoutAction, false, &buf) } @@ -359,7 +359,7 @@ func (w *Watchdog) doAction(action Action, skipStack bool, msg *bytes.Buffer) { case <-metricsEmitted: case <-time.After(1 * time.Second): } - panic(fmt.Sprintf("Stack for running G's are skipped while panicking.\n%s", msg.String())) + panic(fmt.Sprintf("%s\nStack for running G's are skipped while panicking.", msg.String())) default: panic(fmt.Sprintf("Unknown watchdog action %v", action)) -- cgit v1.2.3 From 62bd3ca8a37543c060292ca86669e537b9ebf36d Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Mon, 2 Mar 2020 10:05:52 -0800 Subject: Take write lock when removing xattr PiperOrigin-RevId: 298380654 --- pkg/sentry/fs/fsutil/inode.go | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/fsutil/inode.go b/pkg/sentry/fs/fsutil/inode.go index daecc4ffe..1922ff08c 100644 --- a/pkg/sentry/fs/fsutil/inode.go +++ b/pkg/sentry/fs/fsutil/inode.go @@ -259,8 +259,8 @@ func (i *InodeSimpleExtendedAttributes) ListXattr(context.Context, *fs.Inode, ui // RemoveXattr implements fs.InodeOperations.RemoveXattr. func (i *InodeSimpleExtendedAttributes) RemoveXattr(_ context.Context, _ *fs.Inode, name string) error { - i.mu.RLock() - defer i.mu.RUnlock() + i.mu.Lock() + defer i.mu.Unlock() if _, ok := i.xattrs[name]; ok { delete(i.xattrs, name) return nil -- cgit v1.2.3 From 42fb7d349137bd8847e7c3df6493fde3bc8e6e89 Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Mon, 2 Mar 2020 10:32:20 -0800 Subject: socket: take readMu to access readView DATA RACE in netstack.(*SocketOperations).fetchReadView Write at 0x00c001dca138 by goroutine 1001: gvisor.dev/gvisor/pkg/sentry/socket/netstack.(*SocketOperations).fetchReadView() pkg/sentry/socket/netstack/netstack.go:418 +0x85 gvisor.dev/gvisor/pkg/sentry/socket/netstack.(*SocketOperations).coalescingRead() pkg/sentry/socket/netstack/netstack.go:2309 +0x67 gvisor.dev/gvisor/pkg/sentry/socket/netstack.(*SocketOperations).nonBlockingRead() pkg/sentry/socket/netstack/netstack.go:2378 +0x183d Previous read at 0x00c001dca138 by goroutine 1111: gvisor.dev/gvisor/pkg/sentry/socket/netstack.(*SocketOperations).Ioctl() pkg/sentry/socket/netstack/netstack.go:2666 +0x533 gvisor.dev/gvisor/pkg/sentry/syscalls/linux.Ioctl() Reported-by: syzbot+d4c3885fcc346f08deb6@syzkaller.appspotmail.com PiperOrigin-RevId: 298387377 --- pkg/sentry/socket/netstack/netstack.go | 2 ++ 1 file changed, 2 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 48c268bfa..1eeb37446 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -2663,7 +2663,9 @@ func (s *SocketOperations) Ioctl(ctx context.Context, _ *fs.File, io usermem.IO, } // Add bytes removed from the endpoint but not yet sent to the caller. + s.readMu.Lock() v += len(s.readView) + s.readMu.Unlock() if v > math.MaxInt32 { v = math.MaxInt32 -- cgit v1.2.3 From 43abb24657e737dee1108ff0d512b2e1b6d8a3f6 Mon Sep 17 00:00:00 2001 From: Nayana Bidari Date: Mon, 2 Mar 2020 16:30:51 -0800 Subject: Fix panic caused by invalid address for Bind in packet sockets. PiperOrigin-RevId: 298476533 --- pkg/sentry/socket/netstack/netstack.go | 4 ++++ test/syscalls/linux/packet_socket.cc | 13 +++++++++++++ 2 files changed, 17 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 1eeb37446..13a9a60b4 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -712,6 +712,10 @@ func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking boo // Bind implements the linux syscall bind(2) for sockets backed by // tcpip.Endpoint. func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { + if len(sockaddr) < 2 { + return syserr.ErrInvalidArgument + } + family := usermem.ByteOrder.Uint16(sockaddr) var addr tcpip.FullAddress diff --git a/test/syscalls/linux/packet_socket.cc b/test/syscalls/linux/packet_socket.cc index bc22de788..248762ca9 100644 --- a/test/syscalls/linux/packet_socket.cc +++ b/test/syscalls/linux/packet_socket.cc @@ -417,6 +417,19 @@ TEST_P(CookedPacketTest, BindDrop) { EXPECT_THAT(RetryEINTR(poll)(&pfd, 1, 1000), SyscallSucceedsWithValue(0)); } +// Bind with invalid address. +TEST_P(CookedPacketTest, BindFail) { + // Null address. + ASSERT_THAT(bind(socket_, nullptr, sizeof(struct sockaddr)), + SyscallFailsWithErrno(EFAULT)); + + // Address of size 1. + uint8_t addr = 0; + ASSERT_THAT( + bind(socket_, reinterpret_cast(&addr), sizeof(addr)), + SyscallFailsWithErrno(EINVAL)); +} + INSTANTIATE_TEST_SUITE_P(AllInetTests, CookedPacketTest, ::testing::Values(ETH_P_IP, ETH_P_ALL)); -- cgit v1.2.3 From 277a0d5a1fbe8272d4729c01ee4c6e374d047ebc Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Tue, 3 Mar 2020 14:34:36 -0800 Subject: platform/ptrace: don't call probeSeccomp on arm64 The support of PTRACE_SYSEMU on arm64 was added in the 5.3 kernel, so we can be sure that the current version is higher that 5.3. And this change moves vsyscall seccomp rules to the arch specific file, because vsyscall isn't supported on arm64. PiperOrigin-RevId: 298696493 --- pkg/sentry/platform/ptrace/subprocess_amd64.go | 80 +++++++++++++++++++++++--- pkg/sentry/platform/ptrace/subprocess_arm64.go | 11 +++- pkg/sentry/platform/ptrace/subprocess_linux.go | 65 +-------------------- 3 files changed, 84 insertions(+), 72 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/ptrace/subprocess_amd64.go b/pkg/sentry/platform/ptrace/subprocess_amd64.go index e99798c56..cd74945e7 100644 --- a/pkg/sentry/platform/ptrace/subprocess_amd64.go +++ b/pkg/sentry/platform/ptrace/subprocess_amd64.go @@ -21,6 +21,7 @@ import ( "strings" "syscall" + "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/seccomp" "gvisor.dev/gvisor/pkg/sentry/arch" @@ -183,13 +184,76 @@ func enableCpuidFault() { // appendArchSeccompRules append architecture specific seccomp rules when creating BPF program. // Ref attachedThread() for more detail. -func appendArchSeccompRules(rules []seccomp.RuleSet) []seccomp.RuleSet { - return append(rules, seccomp.RuleSet{ - Rules: seccomp.SyscallRules{ - syscall.SYS_ARCH_PRCTL: []seccomp.Rule{ - {seccomp.AllowValue(linux.ARCH_SET_CPUID), seccomp.AllowValue(0)}, +func appendArchSeccompRules(rules []seccomp.RuleSet, defaultAction linux.BPFAction) []seccomp.RuleSet { + rules = append(rules, + // Rules for trapping vsyscall access. + seccomp.RuleSet{ + Rules: seccomp.SyscallRules{ + syscall.SYS_GETTIMEOFDAY: {}, + syscall.SYS_TIME: {}, + unix.SYS_GETCPU: {}, // SYS_GETCPU was not defined in package syscall on amd64. }, - }, - Action: linux.SECCOMP_RET_ALLOW, - }) + Action: linux.SECCOMP_RET_TRAP, + Vsyscall: true, + }) + if defaultAction != linux.SECCOMP_RET_ALLOW { + rules = append(rules, + seccomp.RuleSet{ + Rules: seccomp.SyscallRules{ + syscall.SYS_ARCH_PRCTL: []seccomp.Rule{ + {seccomp.AllowValue(linux.ARCH_SET_CPUID), seccomp.AllowValue(0)}, + }, + }, + Action: linux.SECCOMP_RET_ALLOW, + }) + } + return rules +} + +// probeSeccomp returns true iff seccomp is run after ptrace notifications, +// which is generally the case for kernel version >= 4.8. This check is dynamic +// because kernels have be backported behavior. +// +// See createStub for more information. +// +// Precondition: the runtime OS thread must be locked. +func probeSeccomp() bool { + // Create a completely new, destroyable process. + t, err := attachedThread(0, linux.SECCOMP_RET_ERRNO) + if err != nil { + panic(fmt.Sprintf("seccomp probe failed: %v", err)) + } + defer t.destroy() + + // Set registers to the yield system call. This call is not allowed + // by the filters specified in the attachThread function. + regs := createSyscallRegs(&t.initRegs, syscall.SYS_SCHED_YIELD) + if err := t.setRegs(®s); err != nil { + panic(fmt.Sprintf("ptrace set regs failed: %v", err)) + } + + for { + // Attempt an emulation. + if _, _, errno := syscall.RawSyscall6(syscall.SYS_PTRACE, unix.PTRACE_SYSEMU, uintptr(t.tid), 0, 0, 0, 0); errno != 0 { + panic(fmt.Sprintf("ptrace syscall-enter failed: %v", errno)) + } + + sig := t.wait(stopped) + if sig == (syscallEvent | syscall.SIGTRAP) { + // Did the seccomp errno hook already run? This would + // indicate that seccomp is first in line and we're + // less than 4.8. + if err := t.getRegs(®s); err != nil { + panic(fmt.Sprintf("ptrace get-regs failed: %v", err)) + } + if _, err := syscallReturnValue(®s); err == nil { + // The seccomp errno mode ran first, and reset + // the error in the registers. + return false + } + // The seccomp hook did not run yet, and therefore it + // is safe to use RET_KILL mode for dispatched calls. + return true + } + } } diff --git a/pkg/sentry/platform/ptrace/subprocess_arm64.go b/pkg/sentry/platform/ptrace/subprocess_arm64.go index 7b975137f..7f5c393f0 100644 --- a/pkg/sentry/platform/ptrace/subprocess_arm64.go +++ b/pkg/sentry/platform/ptrace/subprocess_arm64.go @@ -160,6 +160,15 @@ func enableCpuidFault() { // appendArchSeccompRules append architecture specific seccomp rules when creating BPF program. // Ref attachedThread() for more detail. -func appendArchSeccompRules(rules []seccomp.RuleSet) []seccomp.RuleSet { +func appendArchSeccompRules(rules []seccomp.RuleSet, defaultAction linux.BPFAction) []seccomp.RuleSet { return rules } + +// probeSeccomp returns true if seccomp is run after ptrace notifications, +// which is generally the case for kernel version >= 4.8. +// +// On arm64, the support of PTRACE_SYSEMU was added in the 5.3 kernel, so +// probeSeccomp can always return true. +func probeSeccomp() bool { + return true +} diff --git a/pkg/sentry/platform/ptrace/subprocess_linux.go b/pkg/sentry/platform/ptrace/subprocess_linux.go index 74968dfdf..2ce528601 100644 --- a/pkg/sentry/platform/ptrace/subprocess_linux.go +++ b/pkg/sentry/platform/ptrace/subprocess_linux.go @@ -20,7 +20,6 @@ import ( "fmt" "syscall" - "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/procid" @@ -30,54 +29,6 @@ import ( const syscallEvent syscall.Signal = 0x80 -// probeSeccomp returns true iff seccomp is run after ptrace notifications, -// which is generally the case for kernel version >= 4.8. This check is dynamic -// because kernels have be backported behavior. -// -// See createStub for more information. -// -// Precondition: the runtime OS thread must be locked. -func probeSeccomp() bool { - // Create a completely new, destroyable process. - t, err := attachedThread(0, linux.SECCOMP_RET_ERRNO) - if err != nil { - panic(fmt.Sprintf("seccomp probe failed: %v", err)) - } - defer t.destroy() - - // Set registers to the yield system call. This call is not allowed - // by the filters specified in the attachThread function. - regs := createSyscallRegs(&t.initRegs, syscall.SYS_SCHED_YIELD) - if err := t.setRegs(®s); err != nil { - panic(fmt.Sprintf("ptrace set regs failed: %v", err)) - } - - for { - // Attempt an emulation. - if _, _, errno := syscall.RawSyscall6(syscall.SYS_PTRACE, unix.PTRACE_SYSEMU, uintptr(t.tid), 0, 0, 0, 0); errno != 0 { - panic(fmt.Sprintf("ptrace syscall-enter failed: %v", errno)) - } - - sig := t.wait(stopped) - if sig == (syscallEvent | syscall.SIGTRAP) { - // Did the seccomp errno hook already run? This would - // indicate that seccomp is first in line and we're - // less than 4.8. - if err := t.getRegs(®s); err != nil { - panic(fmt.Sprintf("ptrace get-regs failed: %v", err)) - } - if _, err := syscallReturnValue(®s); err == nil { - // The seccomp errno mode ran first, and reset - // the error in the registers. - return false - } - // The seccomp hook did not run yet, and therefore it - // is safe to use RET_KILL mode for dispatched calls. - return true - } - } -} - // createStub creates a fresh stub processes. // // Precondition: the runtime OS thread must be locked. @@ -123,18 +74,7 @@ func attachedThread(flags uintptr, defaultAction linux.BPFAction) (*thread, erro // stub and all its children. This is used to create child stubs // (below), so we must include the ability to fork, but otherwise lock // down available calls only to what is needed. - rules := []seccomp.RuleSet{ - // Rules for trapping vsyscall access. - { - Rules: seccomp.SyscallRules{ - syscall.SYS_GETTIMEOFDAY: {}, - syscall.SYS_TIME: {}, - unix.SYS_GETCPU: {}, // SYS_GETCPU was not defined in package syscall on amd64. - }, - Action: linux.SECCOMP_RET_TRAP, - Vsyscall: true, - }, - } + rules := []seccomp.RuleSet{} if defaultAction != linux.SECCOMP_RET_ALLOW { rules = append(rules, seccomp.RuleSet{ Rules: seccomp.SyscallRules{ @@ -173,9 +113,8 @@ func attachedThread(flags uintptr, defaultAction linux.BPFAction) (*thread, erro }, Action: linux.SECCOMP_RET_ALLOW, }) - - rules = appendArchSeccompRules(rules) } + rules = appendArchSeccompRules(rules, defaultAction) instrs, err := seccomp.BuildProgram(rules, defaultAction) if err != nil { return nil, err -- cgit v1.2.3 From 122d47aed17abf4301596e19fc8ac9cdad8118d9 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Tue, 3 Mar 2020 15:27:23 -0800 Subject: Update cached file size when cache is skipped gofer.dentryReadWriter.WriteFromBlocks was not updating gofer.dentry.size after a write operation that skips the cache. Updates #1198 PiperOrigin-RevId: 298708646 --- pkg/sentry/fsimpl/gofer/regular_file.go | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go index 54c1031a7..e95209661 100644 --- a/pkg/sentry/fsimpl/gofer/regular_file.go +++ b/pkg/sentry/fsimpl/gofer/regular_file.go @@ -361,8 +361,15 @@ func (rw *dentryReadWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, erro rw.d.handleMu.RLock() if (rw.d.handle.fd >= 0 && !rw.d.fs.opts.forcePageCache) || rw.d.fs.opts.interop == InteropModeShared || rw.direct { n, err := rw.d.handle.writeFromBlocksAt(rw.ctx, srcs, rw.off) - rw.d.handleMu.RUnlock() rw.off += n + rw.d.dataMu.Lock() + if rw.off > rw.d.size { + atomic.StoreUint64(&rw.d.size, rw.off) + // The remote file's size will implicitly be extended to the correct + // value when we write back to it. + } + rw.d.dataMu.Unlock() + rw.d.handleMu.RUnlock() return n, err } -- cgit v1.2.3 From 9b64b658c1b9a4986bc5a4ebd9e5ddeb9f52dfa3 Mon Sep 17 00:00:00 2001 From: Ting-Yu Wang Date: Thu, 5 Mar 2020 17:39:11 -0800 Subject: Fix S/R on inet.Namespace. PiperOrigin-RevId: 299238067 --- pkg/sentry/inet/namespace.go | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/inet/namespace.go b/pkg/sentry/inet/namespace.go index c16667e7f..029af3025 100644 --- a/pkg/sentry/inet/namespace.go +++ b/pkg/sentry/inet/namespace.go @@ -23,7 +23,10 @@ type Namespace struct { // creator allows kernel to create new network stack for network namespaces. // If nil, no networking will function if network is namespaced. - creator NetworkStackCreator + // + // At afterLoad(), creator will be used to create network stack. Stateify + // needs to wait for this field to be loaded before calling afterLoad(). + creator NetworkStackCreator `state:"wait"` // isRoot indicates whether this is the root network namespace. isRoot bool -- cgit v1.2.3 From da48fc6cca23a38faef51c5b5f8ae609940773a0 Mon Sep 17 00:00:00 2001 From: Ian Lewis Date: Thu, 5 Mar 2020 18:21:39 -0800 Subject: Stub oom_score_adj and oom_score. Adds an oom_score_adj and oom_score proc file stub. oom_score_adj accepts writes of values -1000 to 1000 and persists the value with the task. New tasks inherit the parent's oom_score_adj. oom_score is a read-only stub that always returns the value '0'. Issue #202 PiperOrigin-RevId: 299245355 --- pkg/sentry/fs/proc/task.go | 126 ++++++++++++++++++++++++++----- pkg/sentry/fsimpl/proc/task.go | 12 +-- pkg/sentry/fsimpl/proc/task_files.go | 43 +++++++++++ pkg/sentry/fsimpl/proc/tasks_test.go | 32 ++++---- pkg/sentry/kernel/task.go | 33 ++++++++ pkg/sentry/kernel/task_clone.go | 6 ++ pkg/sentry/kernel/task_start.go | 4 + test/syscalls/BUILD | 8 +- test/syscalls/linux/BUILD | 13 ++++ test/syscalls/linux/proc.cc | 21 ++++++ test/syscalls/linux/proc_pid_oomscore.cc | 72 ++++++++++++++++++ 11 files changed, 330 insertions(+), 40 deletions(-) create mode 100644 test/syscalls/linux/proc_pid_oomscore.cc (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go index 8ab8d8a02..4e9b0fc00 100644 --- a/pkg/sentry/fs/proc/task.go +++ b/pkg/sentry/fs/proc/task.go @@ -72,24 +72,26 @@ var _ fs.InodeOperations = (*taskDir)(nil) // newTaskDir creates a new proc task entry. func (p *proc) newTaskDir(t *kernel.Task, msrc *fs.MountSource, isThreadGroup bool) *fs.Inode { contents := map[string]*fs.Inode{ - "auxv": newAuxvec(t, msrc), - "cmdline": newExecArgInode(t, msrc, cmdlineExecArg), - "comm": newComm(t, msrc), - "environ": newExecArgInode(t, msrc, environExecArg), - "exe": newExe(t, msrc), - "fd": newFdDir(t, msrc), - "fdinfo": newFdInfoDir(t, msrc), - "gid_map": newGIDMap(t, msrc), - "io": newIO(t, msrc, isThreadGroup), - "maps": newMaps(t, msrc), - "mountinfo": seqfile.NewSeqFileInode(t, &mountInfoFile{t: t}, msrc), - "mounts": seqfile.NewSeqFileInode(t, &mountsFile{t: t}, msrc), - "ns": newNamespaceDir(t, msrc), - "smaps": newSmaps(t, msrc), - "stat": newTaskStat(t, msrc, isThreadGroup, p.pidns), - "statm": newStatm(t, msrc), - "status": newStatus(t, msrc, p.pidns), - "uid_map": newUIDMap(t, msrc), + "auxv": newAuxvec(t, msrc), + "cmdline": newExecArgInode(t, msrc, cmdlineExecArg), + "comm": newComm(t, msrc), + "environ": newExecArgInode(t, msrc, environExecArg), + "exe": newExe(t, msrc), + "fd": newFdDir(t, msrc), + "fdinfo": newFdInfoDir(t, msrc), + "gid_map": newGIDMap(t, msrc), + "io": newIO(t, msrc, isThreadGroup), + "maps": newMaps(t, msrc), + "mountinfo": seqfile.NewSeqFileInode(t, &mountInfoFile{t: t}, msrc), + "mounts": seqfile.NewSeqFileInode(t, &mountsFile{t: t}, msrc), + "ns": newNamespaceDir(t, msrc), + "oom_score": newOOMScore(t, msrc), + "oom_score_adj": newOOMScoreAdj(t, msrc), + "smaps": newSmaps(t, msrc), + "stat": newTaskStat(t, msrc, isThreadGroup, p.pidns), + "statm": newStatm(t, msrc), + "status": newStatus(t, msrc, p.pidns), + "uid_map": newUIDMap(t, msrc), } if isThreadGroup { contents["task"] = p.newSubtasks(t, msrc) @@ -796,4 +798,92 @@ func (f *auxvecFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequenc return int64(n), err } +// newOOMScore returns a oom_score file. It is a stub that always returns 0. +// TODO(gvisor.dev/issue/1967) +func newOOMScore(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { + return newStaticProcInode(t, msrc, []byte("0\n")) +} + +// oomScoreAdj is a file containing the oom_score adjustment for a task. +// +// +stateify savable +type oomScoreAdj struct { + fsutil.SimpleFileInode + + t *kernel.Task +} + +// +stateify savable +type oomScoreAdjFile struct { + fsutil.FileGenericSeek `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoSplice `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + fsutil.FileUseInodeUnstableAttr `state:"nosave"` + waiter.AlwaysReady `state:"nosave"` + + t *kernel.Task +} + +// newOOMScoreAdj returns a oom_score_adj file. +func newOOMScoreAdj(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { + i := &oomScoreAdj{ + SimpleFileInode: *fsutil.NewSimpleFileInode(t, fs.RootOwner, fs.FilePermsFromMode(0644), linux.PROC_SUPER_MAGIC), + t: t, + } + return newProcInode(t, i, msrc, fs.SpecialFile, t) +} + +// Truncate implements fs.InodeOperations.Truncate. Truncate is called when +// O_TRUNC is specified for any kind of existing Dirent but is not called via +// (f)truncate for proc files. +func (*oomScoreAdj) Truncate(context.Context, *fs.Inode, int64) error { + return nil +} + +// GetFile implements fs.InodeOperations.GetFile. +func (o *oomScoreAdj) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + return fs.NewFile(ctx, dirent, flags, &oomScoreAdjFile{t: o.t}), nil +} + +// Read implements fs.FileOperations.Read. +func (f *oomScoreAdjFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { + if offset != 0 { + return 0, io.EOF + } + adj, err := f.t.OOMScoreAdj() + if err != nil { + return 0, err + } + adjBytes := []byte(strconv.FormatInt(int64(adj), 10) + "\n") + n, err := dst.CopyOut(ctx, adjBytes) + return int64(n), err +} + +// Write implements fs.FileOperations.Write. +func (f *oomScoreAdjFile) Write(ctx context.Context, _ *fs.File, src usermem.IOSequence, offset int64) (int64, error) { + if src.NumBytes() == 0 { + return 0, nil + } + + // Limit input size so as not to impact performance if input size is large. + src = src.TakeFirst(usermem.PageSize - 1) + + var v int32 + n, err := usermem.CopyInt32StringInVec(ctx, src.IO, src.Addrs, &v, src.Opts) + if err != nil { + return 0, err + } + + if err := f.t.SetOOMScoreAdj(v); err != nil { + return 0, err + } + + return n, nil +} + // LINT.ThenChange(../../fsimpl/proc/task.go|../../fsimpl/proc/task_files.go) diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index 2d814668a..18e5cd6f6 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -62,11 +62,13 @@ func newTaskInode(inoGen InoGenerator, task *kernel.Task, pidns *kernel.PIDNames "pid": newNamespaceSymlink(task, inoGen.NextIno(), "pid"), "user": newNamespaceSymlink(task, inoGen.NextIno(), "user"), }), - "smaps": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &smapsData{task: task}), - "stat": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &taskStatData{task: task, pidns: pidns, tgstats: isThreadGroup}), - "statm": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &statmData{task: task}), - "status": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &statusData{task: task, pidns: pidns}), - "uid_map": newTaskOwnedFile(task, inoGen.NextIno(), 0644, &idMapData{task: task, gids: false}), + "oom_score": newTaskOwnedFile(task, inoGen.NextIno(), 0444, newStaticFile("0\n")), + "oom_score_adj": newTaskOwnedFile(task, inoGen.NextIno(), 0644, &oomScoreAdj{task: task}), + "smaps": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &smapsData{task: task}), + "stat": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &taskStatData{task: task, pidns: pidns, tgstats: isThreadGroup}), + "statm": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &statmData{task: task}), + "status": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &statusData{task: task, pidns: pidns}), + "uid_map": newTaskOwnedFile(task, inoGen.NextIno(), 0644, &idMapData{task: task, gids: false}), } if isThreadGroup { contents["task"] = newSubtasks(task, pidns, inoGen, cgroupControllers) diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index efd3b3453..5a231ac86 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -525,3 +525,46 @@ func (i *ioData) Generate(ctx context.Context, buf *bytes.Buffer) error { fmt.Fprintf(buf, "cancelled_write_bytes: %d\n", io.BytesWriteCancelled) return nil } + +// oomScoreAdj is a stub of the /proc//oom_score_adj file. +// +// +stateify savable +type oomScoreAdj struct { + kernfs.DynamicBytesFile + + task *kernel.Task +} + +var _ vfs.WritableDynamicBytesSource = (*oomScoreAdj)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (o *oomScoreAdj) Generate(ctx context.Context, buf *bytes.Buffer) error { + adj, err := o.task.OOMScoreAdj() + if err != nil { + return err + } + fmt.Fprintf(buf, "%d\n", adj) + return nil +} + +// Write implements vfs.WritableDynamicBytesSource.Write. +func (o *oomScoreAdj) Write(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) { + if src.NumBytes() == 0 { + return 0, nil + } + + // Limit input size so as not to impact performance if input size is large. + src = src.TakeFirst(usermem.PageSize - 1) + + var v int32 + n, err := usermem.CopyInt32StringInVec(ctx, src.IO, src.Addrs, &v, src.Opts) + if err != nil { + return 0, err + } + + if err := o.task.SetOOMScoreAdj(v); err != nil { + return 0, err + } + + return n, nil +} diff --git a/pkg/sentry/fsimpl/proc/tasks_test.go b/pkg/sentry/fsimpl/proc/tasks_test.go index c5d531fe0..0eb401619 100644 --- a/pkg/sentry/fsimpl/proc/tasks_test.go +++ b/pkg/sentry/fsimpl/proc/tasks_test.go @@ -63,21 +63,23 @@ var ( "thread-self": threadSelfLink.NextOff, } taskStaticFiles = map[string]testutil.DirentType{ - "auxv": linux.DT_REG, - "cgroup": linux.DT_REG, - "cmdline": linux.DT_REG, - "comm": linux.DT_REG, - "environ": linux.DT_REG, - "gid_map": linux.DT_REG, - "io": linux.DT_REG, - "maps": linux.DT_REG, - "ns": linux.DT_DIR, - "smaps": linux.DT_REG, - "stat": linux.DT_REG, - "statm": linux.DT_REG, - "status": linux.DT_REG, - "task": linux.DT_DIR, - "uid_map": linux.DT_REG, + "auxv": linux.DT_REG, + "cgroup": linux.DT_REG, + "cmdline": linux.DT_REG, + "comm": linux.DT_REG, + "environ": linux.DT_REG, + "gid_map": linux.DT_REG, + "io": linux.DT_REG, + "maps": linux.DT_REG, + "ns": linux.DT_DIR, + "oom_score": linux.DT_REG, + "oom_score_adj": linux.DT_REG, + "smaps": linux.DT_REG, + "stat": linux.DT_REG, + "statm": linux.DT_REG, + "status": linux.DT_REG, + "task": linux.DT_DIR, + "uid_map": linux.DT_REG, } ) diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index 2cee2e6ed..c0dbbe890 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -37,6 +37,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) @@ -554,6 +555,13 @@ type Task struct { // // startTime is protected by mu. startTime ktime.Time + + // oomScoreAdj is the task's OOM score adjustment. This is currently not + // used but is maintained for consistency. + // TODO(gvisor.dev/issue/1967) + // + // oomScoreAdj is protected by mu, and is owned by the task goroutine. + oomScoreAdj int32 } func (t *Task) savePtraceTracer() *Task { @@ -847,3 +855,28 @@ func (t *Task) AbstractSockets() *AbstractSocketNamespace { func (t *Task) ContainerID() string { return t.containerID } + +// OOMScoreAdj gets the task's OOM score adjustment. +func (t *Task) OOMScoreAdj() (int32, error) { + t.mu.Lock() + defer t.mu.Unlock() + if t.ExitState() == TaskExitDead { + return 0, syserror.ESRCH + } + return t.oomScoreAdj, nil +} + +// SetOOMScoreAdj sets the task's OOM score adjustment. The value should be +// between -1000 and 1000 inclusive. +func (t *Task) SetOOMScoreAdj(adj int32) error { + t.mu.Lock() + defer t.mu.Unlock() + if t.ExitState() == TaskExitDead { + return syserror.ESRCH + } + if adj > 1000 || adj < -1000 { + return syserror.EINVAL + } + t.oomScoreAdj = adj + return nil +} diff --git a/pkg/sentry/kernel/task_clone.go b/pkg/sentry/kernel/task_clone.go index 78866f280..dda502bb8 100644 --- a/pkg/sentry/kernel/task_clone.go +++ b/pkg/sentry/kernel/task_clone.go @@ -264,6 +264,11 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { rseqSignature = t.rseqSignature } + adj, err := t.OOMScoreAdj() + if err != nil { + return 0, nil, err + } + cfg := &TaskConfig{ Kernel: t.k, ThreadGroup: tg, @@ -282,6 +287,7 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { RSeqAddr: rseqAddr, RSeqSignature: rseqSignature, ContainerID: t.ContainerID(), + OOMScoreAdj: adj, } if opts.NewThreadGroup { cfg.Parent = t diff --git a/pkg/sentry/kernel/task_start.go b/pkg/sentry/kernel/task_start.go index a5035bb7f..2bbf48bb8 100644 --- a/pkg/sentry/kernel/task_start.go +++ b/pkg/sentry/kernel/task_start.go @@ -93,6 +93,9 @@ type TaskConfig struct { // ContainerID is the container the new task belongs to. ContainerID string + + // oomScoreAdj is the task's OOM score adjustment. + OOMScoreAdj int32 } // NewTask creates a new task defined by cfg. @@ -143,6 +146,7 @@ func (ts *TaskSet) newTask(cfg *TaskConfig) (*Task, error) { rseqSignature: cfg.RSeqSignature, futexWaiter: futex.NewWaiter(), containerID: cfg.ContainerID, + oomScoreAdj: cfg.OOMScoreAdj, } t.creds.Store(cfg.Credentials) t.endStopCond.L = &t.tg.signalHandlers.mu diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index a69b0ce13..9800a0cdf 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -318,10 +318,14 @@ syscall_test( test = "//test/syscalls/linux:proc_test", ) -syscall_test(test = "//test/syscalls/linux:proc_pid_uid_gid_map_test") - syscall_test(test = "//test/syscalls/linux:proc_net_test") +syscall_test(test = "//test/syscalls/linux:proc_pid_oomscore_test") + +syscall_test(test = "//test/syscalls/linux:proc_pid_smaps_test") + +syscall_test(test = "//test/syscalls/linux:proc_pid_uid_gid_map_test") + syscall_test( size = "medium", test = "//test/syscalls/linux:pselect_test", diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 0fbd556de..43455f1a3 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -1631,6 +1631,19 @@ cc_binary( ], ) +cc_binary( + name = "proc_pid_oomscore_test", + testonly = 1, + srcs = ["proc_pid_oomscore.cc"], + linkstatic = 1, + deps = [ + "//test/util:fs_util", + "//test/util:test_main", + "//test/util:test_util", + "@com_google_absl//absl/strings", + ], +) + cc_binary( name = "proc_pid_smaps_test", testonly = 1, diff --git a/test/syscalls/linux/proc.cc b/test/syscalls/linux/proc.cc index f91187e75..5a70f6c3b 100644 --- a/test/syscalls/linux/proc.cc +++ b/test/syscalls/linux/proc.cc @@ -1431,6 +1431,12 @@ TEST(ProcPidFile, SubprocessRunning) { EXPECT_THAT(ReadWhileRunning("uid_map", buf, sizeof(buf)), SyscallSucceedsWithValue(sizeof(buf))); + + EXPECT_THAT(ReadWhileRunning("oom_score", buf, sizeof(buf)), + SyscallSucceedsWithValue(sizeof(buf))); + + EXPECT_THAT(ReadWhileRunning("oom_score_adj", buf, sizeof(buf)), + SyscallSucceedsWithValue(sizeof(buf))); } // Test whether /proc/PID/ files can be read for a zombie process. @@ -1466,6 +1472,12 @@ TEST(ProcPidFile, SubprocessZombie) { EXPECT_THAT(ReadWhileZombied("uid_map", buf, sizeof(buf)), SyscallSucceedsWithValue(sizeof(buf))); + EXPECT_THAT(ReadWhileZombied("oom_score", buf, sizeof(buf)), + SyscallSucceedsWithValue(sizeof(buf))); + + EXPECT_THAT(ReadWhileZombied("oom_score_adj", buf, sizeof(buf)), + SyscallSucceedsWithValue(sizeof(buf))); + // FIXME(gvisor.dev/issue/164): Inconsistent behavior between gVisor and linux // on proc files. // @@ -1527,6 +1539,15 @@ TEST(ProcPidFile, SubprocessExited) { EXPECT_THAT(ReadWhileExited("uid_map", buf, sizeof(buf)), SyscallSucceedsWithValue(sizeof(buf))); + + if (!IsRunningOnGvisor()) { + // FIXME(gvisor.dev/issue/164): Succeeds on gVisor. + EXPECT_THAT(ReadWhileExited("oom_score", buf, sizeof(buf)), + SyscallFailsWithErrno(ESRCH)); + } + + EXPECT_THAT(ReadWhileExited("oom_score_adj", buf, sizeof(buf)), + SyscallFailsWithErrno(ESRCH)); } PosixError DirContainsImpl(absl::string_view path, diff --git a/test/syscalls/linux/proc_pid_oomscore.cc b/test/syscalls/linux/proc_pid_oomscore.cc new file mode 100644 index 000000000..707821a3f --- /dev/null +++ b/test/syscalls/linux/proc_pid_oomscore.cc @@ -0,0 +1,72 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include + +#include +#include +#include + +#include "test/util/fs_util.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +namespace { + +PosixErrorOr ReadProcNumber(std::string path) { + ASSIGN_OR_RETURN_ERRNO(std::string contents, GetContents(path)); + EXPECT_EQ(contents[contents.length() - 1], '\n'); + + int num; + if (!absl::SimpleAtoi(contents, &num)) { + return PosixError(EINVAL, "invalid value: " + contents); + } + + return num; +} + +TEST(ProcPidOomscoreTest, BasicRead) { + auto const oom_score = + ASSERT_NO_ERRNO_AND_VALUE(ReadProcNumber("/proc/self/oom_score")); + EXPECT_LE(oom_score, 1000); + EXPECT_GE(oom_score, -1000); +} + +TEST(ProcPidOomscoreAdjTest, BasicRead) { + auto const oom_score = + ASSERT_NO_ERRNO_AND_VALUE(ReadProcNumber("/proc/self/oom_score_adj")); + + // oom_score_adj defaults to 0. + EXPECT_EQ(oom_score, 0); +} + +TEST(ProcPidOomscoreAdjTest, BasicWrite) { + constexpr int test_value = 7; + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open("/proc/self/oom_score_adj", O_WRONLY)); + ASSERT_THAT( + RetryEINTR(write)(fd.get(), std::to_string(test_value).c_str(), 1), + SyscallSucceeds()); + + auto const oom_score = + ASSERT_NO_ERRNO_AND_VALUE(ReadProcNumber("/proc/self/oom_score_adj")); + EXPECT_EQ(oom_score, test_value); +} + +} // namespace + +} // namespace testing +} // namespace gvisor -- cgit v1.2.3 From 6fa5cee82c0f515b001dee5f3840e1f875b2f477 Mon Sep 17 00:00:00 2001 From: Tamir Duberstein Date: Fri, 6 Mar 2020 12:30:37 -0800 Subject: Prevent memory leaks in ilist When list elements are removed from a list but not discarded, it becomes important to invalidate the references they hold to their former neighbors to prevent memory leaks. PiperOrigin-RevId: 299412421 --- pkg/ilist/list.go | 8 ++++++-- pkg/sentry/fs/dirent_cache.go | 2 -- pkg/sentry/fs/inotify.go | 5 ++++- pkg/sentry/kernel/epoll/epoll_state.go | 13 ++++++++----- pkg/tcpip/network/fragmentation/fragmentation.go | 8 +++++--- 5 files changed, 23 insertions(+), 13 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/ilist/list.go b/pkg/ilist/list.go index f3a609b57..8f93e4d6d 100644 --- a/pkg/ilist/list.go +++ b/pkg/ilist/list.go @@ -169,8 +169,9 @@ func (l *List) InsertBefore(a, e Element) { // Remove removes e from l. func (l *List) Remove(e Element) { - prev := ElementMapper{}.linkerFor(e).Prev() - next := ElementMapper{}.linkerFor(e).Next() + linker := ElementMapper{}.linkerFor(e) + prev := linker.Prev() + next := linker.Next() if prev != nil { ElementMapper{}.linkerFor(prev).SetNext(next) @@ -183,6 +184,9 @@ func (l *List) Remove(e Element) { } else { l.tail = prev } + + linker.SetNext(nil) + linker.SetPrev(nil) } // Entry is a default implementation of Linker. Users can add anonymous fields diff --git a/pkg/sentry/fs/dirent_cache.go b/pkg/sentry/fs/dirent_cache.go index 25514ace4..33de32c69 100644 --- a/pkg/sentry/fs/dirent_cache.go +++ b/pkg/sentry/fs/dirent_cache.go @@ -101,8 +101,6 @@ func (c *DirentCache) remove(d *Dirent) { panic(fmt.Sprintf("trying to remove %v, which is not in the dirent cache", d)) } c.list.Remove(d) - d.SetPrev(nil) - d.SetNext(nil) d.DecRef() c.currentSize-- if c.limit != nil { diff --git a/pkg/sentry/fs/inotify.go b/pkg/sentry/fs/inotify.go index 928c90aa0..e3a715c1f 100644 --- a/pkg/sentry/fs/inotify.go +++ b/pkg/sentry/fs/inotify.go @@ -143,7 +143,10 @@ func (i *Inotify) Read(ctx context.Context, _ *File, dst usermem.IOSequence, _ i } var writeLen int64 - for event := i.events.Front(); event != nil; event = event.Next() { + for it := i.events.Front(); it != nil; { + event := it + it = it.Next() + // Does the buffer have enough remaining space to hold the event we're // about to write out? if dst.NumBytes() < int64(event.sizeOf()) { diff --git a/pkg/sentry/kernel/epoll/epoll_state.go b/pkg/sentry/kernel/epoll/epoll_state.go index a0d35d350..8e9f200d0 100644 --- a/pkg/sentry/kernel/epoll/epoll_state.go +++ b/pkg/sentry/kernel/epoll/epoll_state.go @@ -38,11 +38,14 @@ func (e *EventPoll) afterLoad() { } } - for it := e.waitingList.Front(); it != nil; it = it.Next() { - if it.id.File.Readiness(it.mask) != 0 { - e.waitingList.Remove(it) - e.readyList.PushBack(it) - it.curList = &e.readyList + for it := e.waitingList.Front(); it != nil; { + entry := it + it = it.Next() + + if entry.id.File.Readiness(entry.mask) != 0 { + e.waitingList.Remove(entry) + e.readyList.PushBack(entry) + entry.curList = &e.readyList e.Notify(waiter.EventIn) } } diff --git a/pkg/tcpip/network/fragmentation/fragmentation.go b/pkg/tcpip/network/fragmentation/fragmentation.go index 92f2aa13a..f42abc4bb 100644 --- a/pkg/tcpip/network/fragmentation/fragmentation.go +++ b/pkg/tcpip/network/fragmentation/fragmentation.go @@ -115,10 +115,12 @@ func (f *Fragmentation) Process(id uint32, first, last uint16, more bool, vv buf // Evict reassemblers if we are consuming more memory than highLimit until // we reach lowLimit. if f.size > f.highLimit { - tail := f.rList.Back() - for f.size > f.lowLimit && tail != nil { + for f.size > f.lowLimit { + tail := f.rList.Back() + if tail == nil { + break + } f.release(tail) - tail = tail.Prev() } } f.mu.Unlock() -- cgit v1.2.3 From 960f6a975b7e44c0efe8fd38c66b02017c4fe137 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 6 Mar 2020 12:58:45 -0800 Subject: Add plumbing for importing fds in VFS2, along with non-socket, non-TTY impl. In VFS2, imported file descriptors are stored in a kernfs-based filesystem. Upon calling ImportFD, the host fd can be accessed in two ways: 1. a FileDescription that can be added to the FDTable, and 2. a Dentry in the host.filesystem mount, which we will want to access through magic symlinks in /proc/[pid]/fd/. An implementation of the kernfs.Inode interface stores a unique host fd. This inode can be inserted into file descriptions as well as dentries. This change also plumbs in three FileDescriptionImpls corresponding to fds for sockets, TTYs, and other files (only the latter is implemented here). These implementations will mostly make corresponding syscalls to the host. Where possible, the logic is ported over from pkg/sentry/fs/host. Updates #1672 PiperOrigin-RevId: 299417263 --- pkg/sentry/fs/host/util.go | 12 +- pkg/sentry/fsimpl/host/BUILD | 27 +++ pkg/sentry/fsimpl/host/default_file.go | 233 ++++++++++++++++++++ pkg/sentry/fsimpl/host/host.go | 286 +++++++++++++++++++++++++ pkg/sentry/fsimpl/host/util.go | 86 ++++++++ pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go | 2 +- pkg/sentry/fsimpl/kernfs/fd_impl_util.go | 18 +- pkg/sentry/fsimpl/kernfs/filesystem.go | 6 +- pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 54 ++--- pkg/sentry/fsimpl/kernfs/kernfs.go | 7 +- pkg/sentry/fsimpl/proc/subtasks.go | 13 +- pkg/sentry/fsimpl/proc/task.go | 23 +- pkg/sentry/fsimpl/proc/tasks.go | 19 +- pkg/sentry/vfs/BUILD | 4 + pkg/sentry/vfs/file_description_impl_util.go | 4 +- 15 files changed, 731 insertions(+), 63 deletions(-) create mode 100644 pkg/sentry/fsimpl/host/BUILD create mode 100644 pkg/sentry/fsimpl/host/default_file.go create mode 100644 pkg/sentry/fsimpl/host/host.go create mode 100644 pkg/sentry/fsimpl/host/util.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/host/util.go b/pkg/sentry/fs/host/util.go index e37e687c6..7c60dc1db 100644 --- a/pkg/sentry/fs/host/util.go +++ b/pkg/sentry/fs/host/util.go @@ -24,7 +24,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/device" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" + "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/syserror" ) @@ -152,9 +152,9 @@ func unstableAttr(mo *superOperations, s *syscall.Stat_t) fs.UnstableAttr { Usage: s.Blocks * 512, Perms: fs.FilePermsFromMode(linux.FileMode(s.Mode)), Owner: owner(mo, s), - AccessTime: ktime.FromUnix(s.Atim.Sec, s.Atim.Nsec), - ModificationTime: ktime.FromUnix(s.Mtim.Sec, s.Mtim.Nsec), - StatusChangeTime: ktime.FromUnix(s.Ctim.Sec, s.Ctim.Nsec), + AccessTime: time.FromUnix(s.Atim.Sec, s.Atim.Nsec), + ModificationTime: time.FromUnix(s.Mtim.Sec, s.Mtim.Nsec), + StatusChangeTime: time.FromUnix(s.Ctim.Sec, s.Ctim.Nsec), Links: uint64(s.Nlink), } } @@ -165,6 +165,8 @@ type dirInfo struct { bufp int // location of next record in buf. } +// LINT.IfChange + // isBlockError unwraps os errors and checks if they are caused by EAGAIN or // EWOULDBLOCK. This is so they can be transformed into syserror.ErrWouldBlock. func isBlockError(err error) bool { @@ -177,6 +179,8 @@ func isBlockError(err error) bool { return false } +// LINT.ThenChange(../../fsimpl/host/util.go) + func hostEffectiveKIDs() (uint32, []uint32, error) { gids, err := os.Getgroups() if err != nil { diff --git a/pkg/sentry/fsimpl/host/BUILD b/pkg/sentry/fsimpl/host/BUILD new file mode 100644 index 000000000..731f192b3 --- /dev/null +++ b/pkg/sentry/fsimpl/host/BUILD @@ -0,0 +1,27 @@ +load("//tools:defs.bzl", "go_library") + +licenses(["notice"]) + +go_library( + name = "host", + srcs = [ + "default_file.go", + "host.go", + "util.go", + ], + deps = [ + "//pkg/abi/linux", + "//pkg/context", + "//pkg/log", + "//pkg/refs", + "//pkg/safemem", + "//pkg/sentry/fsimpl/kernfs", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/memmap", + "//pkg/sentry/vfs", + "//pkg/sync", + "//pkg/syserror", + "//pkg/usermem", + "@org_golang_x_sys//unix:go_default_library", + ], +) diff --git a/pkg/sentry/fsimpl/host/default_file.go b/pkg/sentry/fsimpl/host/default_file.go new file mode 100644 index 000000000..172cdb161 --- /dev/null +++ b/pkg/sentry/fsimpl/host/default_file.go @@ -0,0 +1,233 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package host + +import ( + "math" + "syscall" + + "golang.org/x/sys/unix" + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/safemem" + "gvisor.dev/gvisor/pkg/sentry/memmap" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +// defaultFileFD implements FileDescriptionImpl for non-socket, non-TTY files. +type defaultFileFD struct { + fileDescription + + // canMap specifies whether we allow the file to be memory mapped. + canMap bool + + // mu protects the fields below. + mu sync.Mutex + + // offset specifies the current file offset. + offset int64 +} + +// TODO(gvisor.dev/issue/1672): Implement Waitable interface. + +// PRead implements FileDescriptionImpl. +func (f *defaultFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null. + if f.inode.isStream { + return 0, syserror.ESPIPE + } + + return readFromHostFD(ctx, f.inode.hostFD, dst, offset, int(opts.Flags)) +} + +// Read implements FileDescriptionImpl. +func (f *defaultFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null. + if f.inode.isStream { + // These files can't be memory mapped, assert this. + if f.canMap { + panic("files that can return EWOULDBLOCK (sockets, pipes, etc.) cannot be memory mapped") + } + + f.mu.Lock() + n, err := readFromHostFD(ctx, f.inode.hostFD, dst, -1, int(opts.Flags)) + f.mu.Unlock() + if isBlockError(err) { + // If we got any data at all, return it as a "completed" partial read + // rather than retrying until complete. + if n != 0 { + err = nil + } else { + err = syserror.ErrWouldBlock + } + } + return n, err + } + // TODO(gvisor.dev/issue/1672): Cache pages, when forced to do so. + f.mu.Lock() + n, err := readFromHostFD(ctx, f.inode.hostFD, dst, f.offset, int(opts.Flags)) + f.offset += n + f.mu.Unlock() + return n, err +} + +func readFromHostFD(ctx context.Context, fd int, dst usermem.IOSequence, offset int64, flags int) (int64, error) { + if flags&^(linux.RWF_VALID) != 0 { + return 0, syserror.EOPNOTSUPP + } + + reader := safemem.FromVecReaderFunc{ + func(srcs [][]byte) (int64, error) { + n, err := unix.Preadv2(fd, srcs, offset, flags) + return int64(n), err + }, + } + n, err := dst.CopyOutFrom(ctx, reader) + return int64(n), err +} + +// PWrite implements FileDescriptionImpl. +func (f *defaultFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null. + if f.inode.isStream { + return 0, syserror.ESPIPE + } + + return writeToHostFD(ctx, f.inode.hostFD, src, offset, int(opts.Flags)) +} + +// Write implements FileDescriptionImpl. +func (f *defaultFileFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null. + if f.inode.isStream { + // These files can't be memory mapped, assert this. + if f.canMap { + panic("files that can return EWOULDBLOCK (sockets, pipes, etc.) cannot be memory mapped") + } + + f.mu.Lock() + n, err := writeToHostFD(ctx, f.inode.hostFD, src, -1, int(opts.Flags)) + f.mu.Unlock() + if isBlockError(err) { + err = syserror.ErrWouldBlock + } + return n, err + } + // TODO(gvisor.dev/issue/1672): Cache pages, when forced to do so. + // TODO(gvisor.dev/issue/1672): Write to end of file and update offset if O_APPEND is set on this file. + f.mu.Lock() + n, err := writeToHostFD(ctx, f.inode.hostFD, src, f.offset, int(opts.Flags)) + f.offset += n + f.mu.Unlock() + return n, err +} + +func writeToHostFD(ctx context.Context, fd int, src usermem.IOSequence, offset int64, flags int) (int64, error) { + if flags&^(linux.RWF_VALID) != 0 { + return 0, syserror.EOPNOTSUPP + } + + writer := safemem.FromVecWriterFunc{ + func(srcs [][]byte) (int64, error) { + n, err := unix.Pwritev2(fd, srcs, offset, flags) + return int64(n), err + }, + } + n, err := src.CopyInTo(ctx, writer) + return int64(n), err +} + +// Seek implements FileDescriptionImpl. +// +// Note that we do not support seeking on directories, since we do not even +// allow directory fds to be imported at all. +func (f *defaultFileFD) Seek(_ context.Context, offset int64, whence int32) (int64, error) { + // TODO(b/34716638): Some char devices do support seeking, e.g. /dev/null. + if f.inode.isStream { + return 0, syserror.ESPIPE + } + + f.mu.Lock() + defer f.mu.Unlock() + + switch whence { + case linux.SEEK_SET: + if offset < 0 { + return f.offset, syserror.EINVAL + } + f.offset = offset + + case linux.SEEK_CUR: + // Check for overflow. Note that underflow cannot occur, since f.offset >= 0. + if offset > math.MaxInt64-f.offset { + return f.offset, syserror.EOVERFLOW + } + if f.offset+offset < 0 { + return f.offset, syserror.EINVAL + } + f.offset += offset + + case linux.SEEK_END: + var s syscall.Stat_t + if err := syscall.Fstat(f.inode.hostFD, &s); err != nil { + return f.offset, err + } + size := s.Size + + // Check for overflow. Note that underflow cannot occur, since size >= 0. + if offset > math.MaxInt64-size { + return f.offset, syserror.EOVERFLOW + } + if size+offset < 0 { + return f.offset, syserror.EINVAL + } + f.offset = size + offset + + case linux.SEEK_DATA, linux.SEEK_HOLE: + // Modifying the offset in the host file table should not matter, since + // this is the only place where we use it. + // + // For reading and writing, we always rely on our internal offset. + n, err := unix.Seek(f.inode.hostFD, offset, int(whence)) + if err != nil { + return f.offset, err + } + f.offset = n + + default: + // Invalid whence. + return f.offset, syserror.EINVAL + } + + return f.offset, nil +} + +// Sync implements FileDescriptionImpl. +func (f *defaultFileFD) Sync(context.Context) error { + // TODO(gvisor.dev/issue/1672): Currently we do not support the SyncData optimization, so we always sync everything. + return unix.Fsync(f.inode.hostFD) +} + +// ConfigureMMap implements FileDescriptionImpl. +func (f *defaultFileFD) ConfigureMMap(_ context.Context, opts *memmap.MMapOpts) error { + if !f.canMap { + return syserror.ENODEV + } + // TODO(gvisor.dev/issue/1672): Implement ConfigureMMap and Mappable interface. + return syserror.ENODEV +} diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go new file mode 100644 index 000000000..c205e6a0b --- /dev/null +++ b/pkg/sentry/fsimpl/host/host.go @@ -0,0 +1,286 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package host provides a filesystem implementation for host files imported as +// file descriptors. +package host + +import ( + "errors" + "fmt" + "syscall" + + "golang.org/x/sys/unix" + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/refs" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/syserror" +) + +// filesystem implements vfs.FilesystemImpl. +type filesystem struct { + kernfs.Filesystem +} + +// ImportFD sets up and returns a vfs.FileDescription from a donated fd. +func ImportFD(mnt *vfs.Mount, hostFD int, ownerUID auth.KUID, ownerGID auth.KGID, isTTY bool) (*vfs.FileDescription, error) { + // Must be importing to a mount of host.filesystem. + fs, ok := mnt.Filesystem().Impl().(*filesystem) + if !ok { + return nil, fmt.Errorf("can't import host FDs into filesystems of type %T", mnt.Filesystem().Impl()) + } + + // Retrieve metadata. + var s syscall.Stat_t + if err := syscall.Fstat(hostFD, &s); err != nil { + return nil, err + } + + fileMode := linux.FileMode(s.Mode) + fileType := fileMode.FileType() + // Pipes, character devices, and sockets can return EWOULDBLOCK for + // operations that would block. + isStream := fileType == syscall.S_IFIFO || fileType == syscall.S_IFCHR || fileType == syscall.S_IFSOCK + + i := &inode{ + hostFD: hostFD, + isStream: isStream, + isTTY: isTTY, + ino: fs.NextIno(), + mode: fileMode, + uid: ownerUID, + gid: ownerGID, + } + + d := &kernfs.Dentry{} + d.Init(i) + // i.open will take a reference on d. + defer d.DecRef() + + return i.open(d.VFSDentry(), mnt) +} + +// inode implements kernfs.Inode. +type inode struct { + kernfs.InodeNotDirectory + kernfs.InodeNotSymlink + + // When the reference count reaches zero, the host fd is closed. + refs.AtomicRefCount + + // hostFD contains the host fd that this file was originally created from, + // which must be available at time of restore. + // + // This field is initialized at creation time and is immutable. + hostFD int + + // isStream is true if the host fd points to a file representing a stream, + // e.g. a socket or a pipe. Such files are not seekable and can return + // EWOULDBLOCK for I/O operations. + // + // This field is initialized at creation time and is immutable. + isStream bool + + // isTTY is true if this file represents a TTY. + // + // This field is initialized at creation time and is immutable. + isTTY bool + + // ino is an inode number unique within this filesystem. + ino uint64 + + // mu protects the inode metadata below. + mu sync.Mutex + + // mode is the file mode of this inode. Note that this value may become out + // of date if the mode is changed on the host, e.g. with chmod. + mode linux.FileMode + + // uid and gid of the file owner. Note that these refer to the owner of the + // file created on import, not the fd on the host. + uid auth.KUID + gid auth.KGID +} + +// Note that these flags may become out of date, since they can be modified +// on the host, e.g. with fcntl. +func fileFlagsFromHostFD(fd int) (int, error) { + flags, err := unix.FcntlInt(uintptr(fd), syscall.F_GETFL, 0) + if err != nil { + log.Warningf("Failed to get file flags for donated FD %d: %v", fd, err) + return 0, err + } + // TODO(gvisor.dev/issue/1672): implement behavior corresponding to these allowed flags. + flags &= syscall.O_ACCMODE | syscall.O_DIRECT | syscall.O_NONBLOCK | syscall.O_DSYNC | syscall.O_SYNC | syscall.O_APPEND + return flags, nil +} + +// CheckPermissions implements kernfs.Inode. +func (i *inode) CheckPermissions(ctx context.Context, creds *auth.Credentials, atx vfs.AccessTypes) error { + return vfs.GenericCheckPermissions(creds, atx, false /* isDir */, uint16(i.mode), i.uid, i.gid) +} + +// Mode implements kernfs.Inode. +func (i *inode) Mode() linux.FileMode { + return i.mode +} + +// Stat implements kernfs.Inode. +func (i *inode) Stat(_ *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { + var s unix.Statx_t + if err := unix.Statx(i.hostFD, "", int(unix.AT_EMPTY_PATH|opts.Sync), int(opts.Mask), &s); err != nil { + return linux.Statx{}, err + } + ls := unixToLinuxStatx(s) + + // Use our own internal inode number and file owner. + // + // TODO(gvisor.dev/issue/1672): Use a kernfs-specific device number as well. + // If we use the device number from the host, it may collide with another + // sentry-internal device number. We handle device/inode numbers without + // relying on the host to prevent collisions. + ls.Ino = i.ino + ls.UID = uint32(i.uid) + ls.GID = uint32(i.gid) + + // Update file mode from the host. + i.mode = linux.FileMode(ls.Mode) + + return ls, nil +} + +// SetStat implements kernfs.Inode. +func (i *inode) SetStat(_ *vfs.Filesystem, opts vfs.SetStatOptions) error { + s := opts.Stat + + m := s.Mask + if m == 0 { + return nil + } + if m&(linux.STATX_UID|linux.STATX_GID) != 0 { + return syserror.EPERM + } + if m&linux.STATX_MODE != 0 { + if err := syscall.Fchmod(i.hostFD, uint32(s.Mode)); err != nil { + return err + } + i.mode = linux.FileMode(s.Mode) + } + if m&linux.STATX_SIZE != 0 { + if err := syscall.Ftruncate(i.hostFD, int64(s.Size)); err != nil { + return err + } + } + if m&(linux.STATX_ATIME|linux.STATX_MTIME) != 0 { + timestamps := []unix.Timespec{ + toTimespec(s.Atime, m&linux.STATX_ATIME == 0), + toTimespec(s.Mtime, m&linux.STATX_MTIME == 0), + } + if err := unix.UtimesNanoAt(i.hostFD, "", timestamps, unix.AT_EMPTY_PATH); err != nil { + return err + } + } + return nil +} + +// DecRef implements kernfs.Inode. +func (i *inode) DecRef() { + i.AtomicRefCount.DecRefWithDestructor(i.Destroy) +} + +// Destroy implements kernfs.Inode. +func (i *inode) Destroy() { + if err := unix.Close(i.hostFD); err != nil { + log.Warningf("failed to close host fd %d: %v", i.hostFD, err) + } +} + +// Open implements kernfs.Inode. +func (i *inode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + return i.open(vfsd, rp.Mount()) +} + +func (i *inode) open(d *vfs.Dentry, mnt *vfs.Mount) (*vfs.FileDescription, error) { + + fileType := i.mode.FileType() + if fileType == syscall.S_IFSOCK { + if i.isTTY { + return nil, errors.New("cannot use host socket as TTY") + } + // TODO(gvisor.dev/issue/1672): support importing sockets. + return nil, errors.New("importing host sockets not supported") + } + + if i.isTTY { + // TODO(gvisor.dev/issue/1672): support importing host fd as TTY. + return nil, errors.New("importing host fd as TTY not supported") + } + + // For simplicity, set offset to 0. Technically, we should + // only set to 0 on files that are not seekable (sockets, pipes, etc.), + // and use the offset from the host fd otherwise. + fd := &defaultFileFD{ + fileDescription: fileDescription{ + inode: i, + }, + canMap: canMap(uint32(fileType)), + mu: sync.Mutex{}, + offset: 0, + } + + vfsfd := &fd.vfsfd + flags, err := fileFlagsFromHostFD(i.hostFD) + if err != nil { + return nil, err + } + + if err := vfsfd.Init(fd, uint32(flags), mnt, d, &vfs.FileDescriptionOptions{}); err != nil { + return nil, err + } + return vfsfd, nil +} + +// fileDescription is embedded by host fd implementations of FileDescriptionImpl. +type fileDescription struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl + + // inode is vfsfd.Dentry().Impl().(*kernfs.Dentry).Inode().(*inode), but + // cached to reduce indirections and casting. fileDescription does not hold + // a reference on the inode through the inode field (since one is already + // held via the Dentry). + // + // inode is immutable after fileDescription creation. + inode *inode +} + +// SetStat implements vfs.FileDescriptionImpl. +func (f *fileDescription) SetStat(_ context.Context, opts vfs.SetStatOptions) error { + return f.inode.SetStat(nil, opts) +} + +// Stat implements vfs.FileDescriptionImpl. +func (f *fileDescription) Stat(_ context.Context, opts vfs.StatOptions) (linux.Statx, error) { + return f.inode.Stat(nil, opts) +} + +// Release implements vfs.FileDescriptionImpl. +func (f *fileDescription) Release() { + // noop +} diff --git a/pkg/sentry/fsimpl/host/util.go b/pkg/sentry/fsimpl/host/util.go new file mode 100644 index 000000000..e1ccacb4d --- /dev/null +++ b/pkg/sentry/fsimpl/host/util.go @@ -0,0 +1,86 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package host + +import ( + "syscall" + + "golang.org/x/sys/unix" + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/syserror" +) + +func toTimespec(ts linux.StatxTimestamp, omit bool) unix.Timespec { + if omit { + return unix.Timespec{ + Sec: 0, + Nsec: unix.UTIME_OMIT, + } + } + return unix.Timespec{ + Sec: int64(ts.Sec), + Nsec: int64(ts.Nsec), + } +} + +func unixToLinuxStatx(s unix.Statx_t) linux.Statx { + return linux.Statx{ + Mask: s.Mask, + Blksize: s.Blksize, + Attributes: s.Attributes, + Nlink: s.Nlink, + UID: s.Uid, + GID: s.Gid, + Mode: s.Mode, + Ino: s.Ino, + Size: s.Size, + Blocks: s.Blocks, + AttributesMask: s.Attributes_mask, + Atime: unixToLinuxStatxTimestamp(s.Atime), + Btime: unixToLinuxStatxTimestamp(s.Btime), + Ctime: unixToLinuxStatxTimestamp(s.Ctime), + Mtime: unixToLinuxStatxTimestamp(s.Mtime), + RdevMajor: s.Rdev_major, + RdevMinor: s.Rdev_minor, + DevMajor: s.Dev_major, + DevMinor: s.Dev_minor, + } +} + +func unixToLinuxStatxTimestamp(ts unix.StatxTimestamp) linux.StatxTimestamp { + return linux.StatxTimestamp{Sec: ts.Sec, Nsec: ts.Nsec} +} + +// wouldBlock returns true for file types that can return EWOULDBLOCK +// for blocking operations, e.g. pipes, character devices, and sockets. +func wouldBlock(fileType uint32) bool { + return fileType == syscall.S_IFIFO || fileType == syscall.S_IFCHR || fileType == syscall.S_IFSOCK +} + +// canMap returns true if a file with fileType is allowed to be memory mapped. +// This is ported over from VFS1, but it's probably not the best way for us +// to check if a file can be memory mapped. +func canMap(fileType uint32) bool { + // TODO(gvisor.dev/issue/1672): Also allow "special files" to be mapped (see fs/host:canMap()). + // + // TODO(b/38213152): Some obscure character devices can be mapped. + return fileType == syscall.S_IFREG +} + +// isBlockError checks if an error is EAGAIN or EWOULDBLOCK. +// If so, they can be transformed into syserror.ErrWouldBlock. +func isBlockError(err error) bool { + return err == syserror.EAGAIN || err == syserror.EWOULDBLOCK +} diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go index d092ccb2a..1c026f4d8 100644 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -122,7 +122,7 @@ func (fd *DynamicBytesFD) Release() {} // Stat implements vfs.FileDescriptionImpl.Stat. func (fd *DynamicBytesFD) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { fs := fd.vfsfd.VirtualDentry().Mount().Filesystem() - return fd.inode.Stat(fs), nil + return fd.inode.Stat(fs, opts) } // SetStat implements vfs.FileDescriptionImpl.SetStat. diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index 5650512e0..da821d524 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -107,9 +107,13 @@ func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirent fs.mu.Lock() defer fs.mu.Unlock() + opts := vfs.StatOptions{Mask: linux.STATX_INO} // Handle ".". if fd.off == 0 { - stat := fd.inode().Stat(vfsFS) + stat, err := fd.inode().Stat(vfsFS, opts) + if err != nil { + return err + } dirent := vfs.Dirent{ Name: ".", Type: linux.DT_DIR, @@ -125,7 +129,10 @@ func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirent // Handle "..". if fd.off == 1 { parentInode := vfsd.ParentOrSelf().Impl().(*Dentry).inode - stat := parentInode.Stat(vfsFS) + stat, err := parentInode.Stat(vfsFS, opts) + if err != nil { + return err + } dirent := vfs.Dirent{ Name: "..", Type: linux.FileMode(stat.Mode).DirentType(), @@ -146,7 +153,10 @@ func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirent childIdx := fd.off - 2 for it := fd.children.nthLocked(childIdx); it != nil; it = it.Next() { inode := it.Dentry.Impl().(*Dentry).inode - stat := inode.Stat(vfsFS) + stat, err := inode.Stat(vfsFS, opts) + if err != nil { + return err + } dirent := vfs.Dirent{ Name: it.Name, Type: linux.FileMode(stat.Mode).DirentType(), @@ -190,7 +200,7 @@ func (fd *GenericDirectoryFD) Seek(ctx context.Context, offset int64, whence int func (fd *GenericDirectoryFD) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { fs := fd.filesystem() inode := fd.inode() - return inode.Stat(fs), nil + return inode.Stat(fs, opts) } // SetStat implements vfs.FileDescriptionImpl.SetStat. diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 292f58afd..1d7e04ad4 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -12,10 +12,10 @@ // See the License for the specific language governing permissions and // limitations under the License. -// This file implements vfs.FilesystemImpl for kernfs. - package kernfs +// This file implements vfs.FilesystemImpl for kernfs. + import ( "fmt" @@ -634,7 +634,7 @@ func (fs *Filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf if err != nil { return linux.Statx{}, err } - return inode.Stat(fs.VFSFilesystem()), nil + return inode.Stat(fs.VFSFilesystem(), opts) } // StatFSAt implements vfs.FilesystemImpl.StatFSAt. diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 099d70a16..d50018b18 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -36,20 +36,20 @@ type InodeNoopRefCount struct { } // IncRef implements Inode.IncRef. -func (n *InodeNoopRefCount) IncRef() { +func (InodeNoopRefCount) IncRef() { } // DecRef implements Inode.DecRef. -func (n *InodeNoopRefCount) DecRef() { +func (InodeNoopRefCount) DecRef() { } // TryIncRef implements Inode.TryIncRef. -func (n *InodeNoopRefCount) TryIncRef() bool { +func (InodeNoopRefCount) TryIncRef() bool { return true } // Destroy implements Inode.Destroy. -func (n *InodeNoopRefCount) Destroy() { +func (InodeNoopRefCount) Destroy() { } // InodeDirectoryNoNewChildren partially implements the Inode interface. @@ -58,27 +58,27 @@ func (n *InodeNoopRefCount) Destroy() { type InodeDirectoryNoNewChildren struct{} // NewFile implements Inode.NewFile. -func (*InodeDirectoryNoNewChildren) NewFile(context.Context, string, vfs.OpenOptions) (*vfs.Dentry, error) { +func (InodeDirectoryNoNewChildren) NewFile(context.Context, string, vfs.OpenOptions) (*vfs.Dentry, error) { return nil, syserror.EPERM } // NewDir implements Inode.NewDir. -func (*InodeDirectoryNoNewChildren) NewDir(context.Context, string, vfs.MkdirOptions) (*vfs.Dentry, error) { +func (InodeDirectoryNoNewChildren) NewDir(context.Context, string, vfs.MkdirOptions) (*vfs.Dentry, error) { return nil, syserror.EPERM } // NewLink implements Inode.NewLink. -func (*InodeDirectoryNoNewChildren) NewLink(context.Context, string, Inode) (*vfs.Dentry, error) { +func (InodeDirectoryNoNewChildren) NewLink(context.Context, string, Inode) (*vfs.Dentry, error) { return nil, syserror.EPERM } // NewSymlink implements Inode.NewSymlink. -func (*InodeDirectoryNoNewChildren) NewSymlink(context.Context, string, string) (*vfs.Dentry, error) { +func (InodeDirectoryNoNewChildren) NewSymlink(context.Context, string, string) (*vfs.Dentry, error) { return nil, syserror.EPERM } // NewNode implements Inode.NewNode. -func (*InodeDirectoryNoNewChildren) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, error) { +func (InodeDirectoryNoNewChildren) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, error) { return nil, syserror.EPERM } @@ -90,62 +90,62 @@ type InodeNotDirectory struct { } // HasChildren implements Inode.HasChildren. -func (*InodeNotDirectory) HasChildren() bool { +func (InodeNotDirectory) HasChildren() bool { return false } // NewFile implements Inode.NewFile. -func (*InodeNotDirectory) NewFile(context.Context, string, vfs.OpenOptions) (*vfs.Dentry, error) { +func (InodeNotDirectory) NewFile(context.Context, string, vfs.OpenOptions) (*vfs.Dentry, error) { panic("NewFile called on non-directory inode") } // NewDir implements Inode.NewDir. -func (*InodeNotDirectory) NewDir(context.Context, string, vfs.MkdirOptions) (*vfs.Dentry, error) { +func (InodeNotDirectory) NewDir(context.Context, string, vfs.MkdirOptions) (*vfs.Dentry, error) { panic("NewDir called on non-directory inode") } // NewLink implements Inode.NewLinkink. -func (*InodeNotDirectory) NewLink(context.Context, string, Inode) (*vfs.Dentry, error) { +func (InodeNotDirectory) NewLink(context.Context, string, Inode) (*vfs.Dentry, error) { panic("NewLink called on non-directory inode") } // NewSymlink implements Inode.NewSymlink. -func (*InodeNotDirectory) NewSymlink(context.Context, string, string) (*vfs.Dentry, error) { +func (InodeNotDirectory) NewSymlink(context.Context, string, string) (*vfs.Dentry, error) { panic("NewSymlink called on non-directory inode") } // NewNode implements Inode.NewNode. -func (*InodeNotDirectory) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, error) { +func (InodeNotDirectory) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, error) { panic("NewNode called on non-directory inode") } // Unlink implements Inode.Unlink. -func (*InodeNotDirectory) Unlink(context.Context, string, *vfs.Dentry) error { +func (InodeNotDirectory) Unlink(context.Context, string, *vfs.Dentry) error { panic("Unlink called on non-directory inode") } // RmDir implements Inode.RmDir. -func (*InodeNotDirectory) RmDir(context.Context, string, *vfs.Dentry) error { +func (InodeNotDirectory) RmDir(context.Context, string, *vfs.Dentry) error { panic("RmDir called on non-directory inode") } // Rename implements Inode.Rename. -func (*InodeNotDirectory) Rename(context.Context, string, string, *vfs.Dentry, *vfs.Dentry) (*vfs.Dentry, error) { +func (InodeNotDirectory) Rename(context.Context, string, string, *vfs.Dentry, *vfs.Dentry) (*vfs.Dentry, error) { panic("Rename called on non-directory inode") } // Lookup implements Inode.Lookup. -func (*InodeNotDirectory) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { +func (InodeNotDirectory) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { panic("Lookup called on non-directory inode") } // IterDirents implements Inode.IterDirents. -func (*InodeNotDirectory) IterDirents(ctx context.Context, callback vfs.IterDirentsCallback, offset, relOffset int64) (newOffset int64, err error) { +func (InodeNotDirectory) IterDirents(ctx context.Context, callback vfs.IterDirentsCallback, offset, relOffset int64) (newOffset int64, err error) { panic("IterDirents called on non-directory inode") } // Valid implements Inode.Valid. -func (*InodeNotDirectory) Valid(context.Context) bool { +func (InodeNotDirectory) Valid(context.Context) bool { return true } @@ -157,17 +157,17 @@ func (*InodeNotDirectory) Valid(context.Context) bool { type InodeNoDynamicLookup struct{} // Lookup implements Inode.Lookup. -func (*InodeNoDynamicLookup) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { +func (InodeNoDynamicLookup) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { return nil, syserror.ENOENT } // IterDirents implements Inode.IterDirents. -func (*InodeNoDynamicLookup) IterDirents(ctx context.Context, callback vfs.IterDirentsCallback, offset, relOffset int64) (int64, error) { +func (InodeNoDynamicLookup) IterDirents(ctx context.Context, callback vfs.IterDirentsCallback, offset, relOffset int64) (int64, error) { return offset, nil } // Valid implements Inode.Valid. -func (*InodeNoDynamicLookup) Valid(ctx context.Context) bool { +func (InodeNoDynamicLookup) Valid(ctx context.Context) bool { return true } @@ -177,7 +177,7 @@ func (*InodeNoDynamicLookup) Valid(ctx context.Context) bool { type InodeNotSymlink struct{} // Readlink implements Inode.Readlink. -func (*InodeNotSymlink) Readlink(context.Context) (string, error) { +func (InodeNotSymlink) Readlink(context.Context) (string, error) { return "", syserror.EINVAL } @@ -219,7 +219,7 @@ func (a *InodeAttrs) Mode() linux.FileMode { // Stat partially implements Inode.Stat. Note that this function doesn't provide // all the stat fields, and the embedder should consider extending the result // with filesystem-specific fields. -func (a *InodeAttrs) Stat(*vfs.Filesystem) linux.Statx { +func (a *InodeAttrs) Stat(*vfs.Filesystem, vfs.StatOptions) (linux.Statx, error) { var stat linux.Statx stat.Mask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_UID | linux.STATX_GID | linux.STATX_INO | linux.STATX_NLINK stat.Ino = atomic.LoadUint64(&a.ino) @@ -230,7 +230,7 @@ func (a *InodeAttrs) Stat(*vfs.Filesystem) linux.Statx { // TODO: Implement other stat fields like timestamps. - return stat + return stat, nil } // SetStat implements Inode.SetStat. diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index c74fa999b..a8ab2a2ba 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -176,8 +176,6 @@ type Dentry struct { vfsd vfs.Dentry inode Inode - refs uint64 - // flags caches useful information about the dentry from the inode. See the // dflags* consts above. Must be accessed by atomic ops. flags uint32 @@ -302,7 +300,8 @@ type Inode interface { // this inode. The returned file description should hold a reference on the // inode for its lifetime. // - // Precondition: !rp.Done(). vfsd.Impl() must be a kernfs Dentry. + // Precondition: rp.Done(). vfsd.Impl() must be the kernfs Dentry containing + // the inode on which Open() is being called. Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) } @@ -328,7 +327,7 @@ type inodeMetadata interface { // Stat returns the metadata for this inode. This corresponds to // vfs.FilesystemImpl.StatAt. - Stat(fs *vfs.Filesystem) linux.Statx + Stat(fs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) // SetStat updates the metadata for this inode. This corresponds to // vfs.FilesystemImpl.SetStatAt. diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go index f3f4e49b4..611645f3f 100644 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -121,8 +121,13 @@ func (i *subtasksInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.O } // Stat implements kernfs.Inode. -func (i *subtasksInode) Stat(vsfs *vfs.Filesystem) linux.Statx { - stat := i.InodeAttrs.Stat(vsfs) - stat.Nlink += uint32(i.task.ThreadGroup().Count()) - return stat +func (i *subtasksInode) Stat(vsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { + stat, err := i.InodeAttrs.Stat(vsfs, opts) + if err != nil { + return linux.Statx{}, err + } + if opts.Mask&linux.STATX_NLINK != 0 { + stat.Nlink += uint32(i.task.ThreadGroup().Count()) + } + return stat, nil } diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index 18e5cd6f6..c0d643f51 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -154,12 +154,21 @@ func newTaskOwnedDir(task *kernel.Task, ino uint64, perm linux.FileMode, childre } // Stat implements kernfs.Inode. -func (i *taskOwnedInode) Stat(fs *vfs.Filesystem) linux.Statx { - stat := i.Inode.Stat(fs) - uid, gid := i.getOwner(linux.FileMode(stat.Mode)) - stat.UID = uint32(uid) - stat.GID = uint32(gid) - return stat +func (i *taskOwnedInode) Stat(fs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { + stat, err := i.Inode.Stat(fs, opts) + if err != nil { + return linux.Statx{}, err + } + if opts.Mask&(linux.STATX_UID|linux.STATX_GID) != 0 { + uid, gid := i.getOwner(linux.FileMode(stat.Mode)) + if opts.Mask&linux.STATX_UID != 0 { + stat.UID = uint32(uid) + } + if opts.Mask&linux.STATX_GID != 0 { + stat.GID = uint32(gid) + } + } + return stat, nil } // CheckPermissions implements kernfs.Inode. @@ -236,7 +245,7 @@ func newNamespaceSymlink(task *kernel.Task, ino uint64, ns string) *kernfs.Dentr // member, there is one entry containing three colon-separated fields: // hierarchy-ID:controller-list:cgroup-path" func newCgroupData(controllers map[string]string) dynamicInode { - buf := bytes.Buffer{} + var buf bytes.Buffer // The hierarchy ids must be positive integers (for cgroup v1), but the // exact number does not matter, so long as they are unique. We can diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index 10c08fa90..b1e39c82f 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -211,17 +211,22 @@ func (i *tasksInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.Open return fd.VFSFileDescription(), nil } -func (i *tasksInode) Stat(vsfs *vfs.Filesystem) linux.Statx { - stat := i.InodeAttrs.Stat(vsfs) +func (i *tasksInode) Stat(vsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { + stat, err := i.InodeAttrs.Stat(vsfs, opts) + if err != nil { + return linux.Statx{}, err + } - // Add dynamic children to link count. - for _, tg := range i.pidns.ThreadGroups() { - if leader := tg.Leader(); leader != nil { - stat.Nlink++ + if opts.Mask&linux.STATX_NLINK != 0 { + // Add dynamic children to link count. + for _, tg := range i.pidns.ThreadGroups() { + if leader := tg.Leader(); leader != nil { + stat.Nlink++ + } } } - return stat + return stat, nil } func cpuInfoData(k *kernel.Kernel) string { diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index 07c8383e6..cb4deb068 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -42,10 +42,13 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/context", + "//pkg/fd", "//pkg/fspath", "//pkg/gohacks", "//pkg/log", + "//pkg/safemem", "//pkg/sentry/arch", + "//pkg/sentry/fs", "//pkg/sentry/fs/lock", "//pkg/sentry/kernel/auth", "//pkg/sentry/memmap", @@ -53,6 +56,7 @@ go_library( "//pkg/syserror", "//pkg/usermem", "//pkg/waiter", + "@org_golang_x_sys//unix:go_default_library", ], ) diff --git a/pkg/sentry/vfs/file_description_impl_util.go b/pkg/sentry/vfs/file_description_impl_util.go index c2a52ec1b..45191d1c3 100644 --- a/pkg/sentry/vfs/file_description_impl_util.go +++ b/pkg/sentry/vfs/file_description_impl_util.go @@ -33,8 +33,8 @@ import ( // implementations to adapt: // - Have a local fileDescription struct (containing FileDescription) which // embeds FileDescriptionDefaultImpl and overrides the default methods -// which are common to all fd implementations for that for that filesystem -// like StatusFlags, SetStatusFlags, Stat, SetStat, StatFS, etc. +// which are common to all fd implementations for that filesystem like +// StatusFlags, SetStatusFlags, Stat, SetStat, StatFS, etc. // - This should be embedded in all file description implementations as the // first field by value. // - Directory FDs would also embed DirectoryFileDescriptionDefaultImpl. -- cgit v1.2.3 From 228813fd26aadea012cd8c39e084e4b0fb23d273 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 6 Mar 2020 15:22:30 -0800 Subject: Update comments and debug level for profiling options. PiperOrigin-RevId: 299448307 --- pkg/sentry/control/pprof.go | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/control/pprof.go b/pkg/sentry/control/pprof.go index 5d1907c0e..663e51989 100644 --- a/pkg/sentry/control/pprof.go +++ b/pkg/sentry/control/pprof.go @@ -117,15 +117,15 @@ func (p *Profile) HeapProfile(o *ProfileOpts, _ *struct{}) error { return nil } -// GoroutineProfile is an RPC stub which dumps out the stack trace for all running -// goroutines. +// GoroutineProfile is an RPC stub which dumps out the stack trace for all +// running goroutines. func (p *Profile) GoroutineProfile(o *ProfileOpts, _ *struct{}) error { if len(o.FilePayload.Files) < 1 { return errNoOutput } output := o.FilePayload.Files[0] defer output.Close() - if err := pprof.Lookup("goroutine").WriteTo(output, 0); err != nil { + if err := pprof.Lookup("goroutine").WriteTo(output, 2); err != nil { return err } return nil -- cgit v1.2.3 From c04958e2fa456587277baef361868bddc0df9e49 Mon Sep 17 00:00:00 2001 From: Haibo Xu Date: Mon, 13 Jan 2020 07:44:58 +0000 Subject: Enable thread local storage support on arm64. Linux use the task.thread.uw.tp_value field to store the TLS pointer on arm64 platform, and we use a similar way in gvisor to store it in the arch/State struct. Signed-off-by: Haibo Xu Change-Id: Ie76b5c6d109bc27ccfd594008a96753806db7764 --- pkg/abi/linux/elf.go | 3 ++ pkg/sentry/arch/arch_aarch64.go | 5 ++ pkg/sentry/arch/arch_arm64.go | 13 ++--- pkg/sentry/platform/ptrace/BUILD | 1 + pkg/sentry/platform/ptrace/ptrace_amd64.go | 14 +++++ pkg/sentry/platform/ptrace/ptrace_arm64_unsafe.go | 62 +++++++++++++++++++++++ pkg/sentry/platform/ptrace/subprocess.go | 12 +++++ 7 files changed, 104 insertions(+), 6 deletions(-) create mode 100644 pkg/sentry/platform/ptrace/ptrace_arm64_unsafe.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/elf.go b/pkg/abi/linux/elf.go index 40f0459a0..7c9a02f20 100644 --- a/pkg/abi/linux/elf.go +++ b/pkg/abi/linux/elf.go @@ -102,4 +102,7 @@ const ( // NT_X86_XSTATE is for x86 extended state using xsave. NT_X86_XSTATE = 0x202 + + // NT_ARM_TLS is for ARM TLS register. + NT_ARM_TLS = 0x401 ) diff --git a/pkg/sentry/arch/arch_aarch64.go b/pkg/sentry/arch/arch_aarch64.go index 5053393c1..b998f84fc 100644 --- a/pkg/sentry/arch/arch_aarch64.go +++ b/pkg/sentry/arch/arch_aarch64.go @@ -95,6 +95,9 @@ type State struct { // Our floating point state. aarch64FPState `state:"wait"` + // TLS pointer + TPValue uint64 + // FeatureSet is a pointer to the currently active feature set. FeatureSet *cpuid.FeatureSet } @@ -145,6 +148,7 @@ func (s *State) Fork() State { return State{ Regs: s.Regs, aarch64FPState: s.aarch64FPState.fork(), + TPValue: s.TPValue, FeatureSet: s.FeatureSet, } } @@ -255,6 +259,7 @@ func (s *State) PtraceSetFPRegs(src io.Reader) (int, error) { const ( _NT_PRSTATUS = 1 _NT_PRFPREG = 2 + _NT_ARM_TLS = 0x401 ) // PtraceGetRegSet implements Context.PtraceGetRegSet. diff --git a/pkg/sentry/arch/arch_arm64.go b/pkg/sentry/arch/arch_arm64.go index 885115ae2..db99c5acb 100644 --- a/pkg/sentry/arch/arch_arm64.go +++ b/pkg/sentry/arch/arch_arm64.go @@ -140,16 +140,17 @@ func (c *context64) SetStack(value uintptr) { // TLS returns the current TLS pointer. func (c *context64) TLS() uintptr { - // TODO(gvisor.dev/issue/1238): TLS is not supported. - // MRS_TPIDR_EL0 - return 0 + return uintptr(c.TPValue) } // SetTLS sets the current TLS pointer. Returns false if value is invalid. func (c *context64) SetTLS(value uintptr) bool { - // TODO(gvisor.dev/issue/1238): TLS is not supported. - // MSR_TPIDR_EL0 - return false + if value >= uintptr(maxAddr64) { + return false + } + + c.TPValue = uint64(value) + return true } // SetOldRSeqInterruptedIP implements Context.SetOldRSeqInterruptedIP. diff --git a/pkg/sentry/platform/ptrace/BUILD b/pkg/sentry/platform/ptrace/BUILD index 95abd321e..30402c2df 100644 --- a/pkg/sentry/platform/ptrace/BUILD +++ b/pkg/sentry/platform/ptrace/BUILD @@ -9,6 +9,7 @@ go_library( "ptrace.go", "ptrace_amd64.go", "ptrace_arm64.go", + "ptrace_arm64_unsafe.go", "ptrace_unsafe.go", "stub_amd64.s", "stub_arm64.s", diff --git a/pkg/sentry/platform/ptrace/ptrace_amd64.go b/pkg/sentry/platform/ptrace/ptrace_amd64.go index db0212538..24fc5dc62 100644 --- a/pkg/sentry/platform/ptrace/ptrace_amd64.go +++ b/pkg/sentry/platform/ptrace/ptrace_amd64.go @@ -31,3 +31,17 @@ func fpRegSet(useXsave bool) uintptr { func stackPointer(r *syscall.PtraceRegs) uintptr { return uintptr(r.Rsp) } + +// x86 use the fs_base register to store the TLS pointer which can be +// get/set in "func (t *thread) get/setRegs(regs *syscall.PtraceRegs)". +// So both of the get/setTLS() operations are noop here. + +// getTLS gets the thread local storage register. +func (t *thread) getTLS(tls *uint64) error { + return nil +} + +// setTLS sets the thread local storage register. +func (t *thread) setTLS(tls *uint64) error { + return nil +} diff --git a/pkg/sentry/platform/ptrace/ptrace_arm64_unsafe.go b/pkg/sentry/platform/ptrace/ptrace_arm64_unsafe.go new file mode 100644 index 000000000..32b8a6be9 --- /dev/null +++ b/pkg/sentry/platform/ptrace/ptrace_arm64_unsafe.go @@ -0,0 +1,62 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package ptrace + +import ( + "syscall" + "unsafe" + + "gvisor.dev/gvisor/pkg/abi/linux" +) + +// getTLS gets the thread local storage register. +func (t *thread) getTLS(tls *uint64) error { + iovec := syscall.Iovec{ + Base: (*byte)(unsafe.Pointer(tls)), + Len: uint64(unsafe.Sizeof(*tls)), + } + _, _, errno := syscall.RawSyscall6( + syscall.SYS_PTRACE, + syscall.PTRACE_GETREGSET, + uintptr(t.tid), + linux.NT_ARM_TLS, + uintptr(unsafe.Pointer(&iovec)), + 0, 0) + if errno != 0 { + return errno + } + return nil +} + +// setTLS sets the thread local storage register. +func (t *thread) setTLS(tls *uint64) error { + iovec := syscall.Iovec{ + Base: (*byte)(unsafe.Pointer(tls)), + Len: uint64(unsafe.Sizeof(*tls)), + } + _, _, errno := syscall.RawSyscall6( + syscall.SYS_PTRACE, + syscall.PTRACE_SETREGSET, + uintptr(t.tid), + linux.NT_ARM_TLS, + uintptr(unsafe.Pointer(&iovec)), + 0, 0) + if errno != 0 { + return errno + } + return nil +} diff --git a/pkg/sentry/platform/ptrace/subprocess.go b/pkg/sentry/platform/ptrace/subprocess.go index 31b7cec53..a644609ef 100644 --- a/pkg/sentry/platform/ptrace/subprocess.go +++ b/pkg/sentry/platform/ptrace/subprocess.go @@ -506,6 +506,9 @@ func (s *subprocess) switchToApp(c *context, ac arch.Context) bool { regs := &ac.StateData().Regs t.resetSysemuRegs(regs) + // Extract TLS register + tls := uint64(ac.TLS()) + // Check for interrupts, and ensure that future interrupts will signal t. if !c.interrupt.Enable(t) { // Pending interrupt; simulate. @@ -526,6 +529,9 @@ func (s *subprocess) switchToApp(c *context, ac arch.Context) bool { if err := t.setFPRegs(fpState, uint64(fpLen), useXsave); err != nil { panic(fmt.Sprintf("ptrace set fpregs (%+v) failed: %v", fpState, err)) } + if err := t.setTLS(&tls); err != nil { + panic(fmt.Sprintf("ptrace set tls (%+v) failed: %v", tls, err)) + } for { // Start running until the next system call. @@ -555,6 +561,12 @@ func (s *subprocess) switchToApp(c *context, ac arch.Context) bool { if err := t.getFPRegs(fpState, uint64(fpLen), useXsave); err != nil { panic(fmt.Sprintf("ptrace get fpregs failed: %v", err)) } + if err := t.getTLS(&tls); err != nil { + panic(fmt.Sprintf("ptrace get tls failed: %v", err)) + } + if !ac.SetTLS(uintptr(tls)) { + panic(fmt.Sprintf("tls value %v is invalid", tls)) + } // Is it a system call? if sig == (syscallEvent | syscall.SIGTRAP) { -- cgit v1.2.3 From b36de6e7be0542b410901d3cbcd1b3c0fc493cf5 Mon Sep 17 00:00:00 2001 From: Ting-Yu Wang Date: Mon, 9 Mar 2020 19:57:35 -0700 Subject: Move /proc/net to /proc/PID/net, and make /proc/net -> /proc/self/net. Issue #1833 PiperOrigin-RevId: 299998105 --- pkg/sentry/fs/proc/net.go | 53 +-- pkg/sentry/fs/proc/proc.go | 2 +- pkg/sentry/fs/proc/task.go | 1 + pkg/sentry/fsimpl/proc/BUILD | 2 +- pkg/sentry/fsimpl/proc/task.go | 1 + pkg/sentry/fsimpl/proc/task_net.go | 790 +++++++++++++++++++++++++++++++++++ pkg/sentry/fsimpl/proc/tasks.go | 2 +- pkg/sentry/fsimpl/proc/tasks_net.go | 787 ---------------------------------- pkg/sentry/fsimpl/proc/tasks_test.go | 3 +- test/syscalls/linux/proc_net.cc | 25 ++ 10 files changed, 849 insertions(+), 817 deletions(-) create mode 100644 pkg/sentry/fsimpl/proc/task_net.go delete mode 100644 pkg/sentry/fsimpl/proc/tasks_net.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/proc/net.go b/pkg/sentry/fs/proc/net.go index 95d5817ff..bd18177d4 100644 --- a/pkg/sentry/fs/proc/net.go +++ b/pkg/sentry/fs/proc/net.go @@ -40,47 +40,48 @@ import ( // LINT.IfChange -// newNet creates a new proc net entry. -func (p *proc) newNetDir(ctx context.Context, k *kernel.Kernel, msrc *fs.MountSource) *fs.Inode { +// newNetDir creates a new proc net entry. +func newNetDir(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { + k := t.Kernel() + var contents map[string]*fs.Inode - // TODO(gvisor.dev/issue/1833): Support for using the network stack in the - // network namespace of the calling process. We should make this per-process, - // a.k.a. /proc/PID/net, and make /proc/net a symlink to /proc/self/net. - if s := p.k.RootNetworkNamespace().Stack(); s != nil { + if s := t.NetworkNamespace().Stack(); s != nil { + // TODO(gvisor.dev/issue/1833): Make sure file contents reflect the task + // network namespace. contents = map[string]*fs.Inode{ - "dev": seqfile.NewSeqFileInode(ctx, &netDev{s: s}, msrc), - "snmp": seqfile.NewSeqFileInode(ctx, &netSnmp{s: s}, msrc), + "dev": seqfile.NewSeqFileInode(t, &netDev{s: s}, msrc), + "snmp": seqfile.NewSeqFileInode(t, &netSnmp{s: s}, msrc), // The following files are simple stubs until they are // implemented in netstack, if the file contains a // header the stub is just the header otherwise it is // an empty file. - "arp": newStaticProcInode(ctx, msrc, []byte("IP address HW type Flags HW address Mask Device\n")), + "arp": newStaticProcInode(t, msrc, []byte("IP address HW type Flags HW address Mask Device\n")), - "netlink": newStaticProcInode(ctx, msrc, []byte("sk Eth Pid Groups Rmem Wmem Dump Locks Drops Inode\n")), - "netstat": newStaticProcInode(ctx, msrc, []byte("TcpExt: SyncookiesSent SyncookiesRecv SyncookiesFailed EmbryonicRsts PruneCalled RcvPruned OfoPruned OutOfWindowIcmps LockDroppedIcmps ArpFilter TW TWRecycled TWKilled PAWSPassive PAWSActive PAWSEstab DelayedACKs DelayedACKLocked DelayedACKLost ListenOverflows ListenDrops TCPPrequeued TCPDirectCopyFromBacklog TCPDirectCopyFromPrequeue TCPPrequeueDropped TCPHPHits TCPHPHitsToUser TCPPureAcks TCPHPAcks TCPRenoRecovery TCPSackRecovery TCPSACKReneging TCPFACKReorder TCPSACKReorder TCPRenoReorder TCPTSReorder TCPFullUndo TCPPartialUndo TCPDSACKUndo TCPLossUndo TCPLostRetransmit TCPRenoFailures TCPSackFailures TCPLossFailures TCPFastRetrans TCPForwardRetrans TCPSlowStartRetrans TCPTimeouts TCPLossProbes TCPLossProbeRecovery TCPRenoRecoveryFail TCPSackRecoveryFail TCPSchedulerFailed TCPRcvCollapsed TCPDSACKOldSent TCPDSACKOfoSent TCPDSACKRecv TCPDSACKOfoRecv TCPAbortOnData TCPAbortOnClose TCPAbortOnMemory TCPAbortOnTimeout TCPAbortOnLinger TCPAbortFailed TCPMemoryPressures TCPSACKDiscard TCPDSACKIgnoredOld TCPDSACKIgnoredNoUndo TCPSpuriousRTOs TCPMD5NotFound TCPMD5Unexpected TCPMD5Failure TCPSackShifted TCPSackMerged TCPSackShiftFallback TCPBacklogDrop TCPMinTTLDrop TCPDeferAcceptDrop IPReversePathFilter TCPTimeWaitOverflow TCPReqQFullDoCookies TCPReqQFullDrop TCPRetransFail TCPRcvCoalesce TCPOFOQueue TCPOFODrop TCPOFOMerge TCPChallengeACK TCPSYNChallenge TCPFastOpenActive TCPFastOpenActiveFail TCPFastOpenPassive TCPFastOpenPassiveFail TCPFastOpenListenOverflow TCPFastOpenCookieReqd TCPSpuriousRtxHostQueues BusyPollRxPackets TCPAutoCorking TCPFromZeroWindowAdv TCPToZeroWindowAdv TCPWantZeroWindowAdv TCPSynRetrans TCPOrigDataSent TCPHystartTrainDetect TCPHystartTrainCwnd TCPHystartDelayDetect TCPHystartDelayCwnd TCPACKSkippedSynRecv TCPACKSkippedPAWS TCPACKSkippedSeq TCPACKSkippedFinWait2 TCPACKSkippedTimeWait TCPACKSkippedChallenge TCPWinProbe TCPKeepAlive TCPMTUPFail TCPMTUPSuccess\n")), - "packet": newStaticProcInode(ctx, msrc, []byte("sk RefCnt Type Proto Iface R Rmem User Inode\n")), - "protocols": newStaticProcInode(ctx, msrc, []byte("protocol size sockets memory press maxhdr slab module cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n")), + "netlink": newStaticProcInode(t, msrc, []byte("sk Eth Pid Groups Rmem Wmem Dump Locks Drops Inode\n")), + "netstat": newStaticProcInode(t, msrc, []byte("TcpExt: SyncookiesSent SyncookiesRecv SyncookiesFailed EmbryonicRsts PruneCalled RcvPruned OfoPruned OutOfWindowIcmps LockDroppedIcmps ArpFilter TW TWRecycled TWKilled PAWSPassive PAWSActive PAWSEstab DelayedACKs DelayedACKLocked DelayedACKLost ListenOverflows ListenDrops TCPPrequeued TCPDirectCopyFromBacklog TCPDirectCopyFromPrequeue TCPPrequeueDropped TCPHPHits TCPHPHitsToUser TCPPureAcks TCPHPAcks TCPRenoRecovery TCPSackRecovery TCPSACKReneging TCPFACKReorder TCPSACKReorder TCPRenoReorder TCPTSReorder TCPFullUndo TCPPartialUndo TCPDSACKUndo TCPLossUndo TCPLostRetransmit TCPRenoFailures TCPSackFailures TCPLossFailures TCPFastRetrans TCPForwardRetrans TCPSlowStartRetrans TCPTimeouts TCPLossProbes TCPLossProbeRecovery TCPRenoRecoveryFail TCPSackRecoveryFail TCPSchedulerFailed TCPRcvCollapsed TCPDSACKOldSent TCPDSACKOfoSent TCPDSACKRecv TCPDSACKOfoRecv TCPAbortOnData TCPAbortOnClose TCPAbortOnMemory TCPAbortOnTimeout TCPAbortOnLinger TCPAbortFailed TCPMemoryPressures TCPSACKDiscard TCPDSACKIgnoredOld TCPDSACKIgnoredNoUndo TCPSpuriousRTOs TCPMD5NotFound TCPMD5Unexpected TCPMD5Failure TCPSackShifted TCPSackMerged TCPSackShiftFallback TCPBacklogDrop TCPMinTTLDrop TCPDeferAcceptDrop IPReversePathFilter TCPTimeWaitOverflow TCPReqQFullDoCookies TCPReqQFullDrop TCPRetransFail TCPRcvCoalesce TCPOFOQueue TCPOFODrop TCPOFOMerge TCPChallengeACK TCPSYNChallenge TCPFastOpenActive TCPFastOpenActiveFail TCPFastOpenPassive TCPFastOpenPassiveFail TCPFastOpenListenOverflow TCPFastOpenCookieReqd TCPSpuriousRtxHostQueues BusyPollRxPackets TCPAutoCorking TCPFromZeroWindowAdv TCPToZeroWindowAdv TCPWantZeroWindowAdv TCPSynRetrans TCPOrigDataSent TCPHystartTrainDetect TCPHystartTrainCwnd TCPHystartDelayDetect TCPHystartDelayCwnd TCPACKSkippedSynRecv TCPACKSkippedPAWS TCPACKSkippedSeq TCPACKSkippedFinWait2 TCPACKSkippedTimeWait TCPACKSkippedChallenge TCPWinProbe TCPKeepAlive TCPMTUPFail TCPMTUPSuccess\n")), + "packet": newStaticProcInode(t, msrc, []byte("sk RefCnt Type Proto Iface R Rmem User Inode\n")), + "protocols": newStaticProcInode(t, msrc, []byte("protocol size sockets memory press maxhdr slab module cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n")), // Linux sets psched values to: nsec per usec, psched // tick in ns, 1000000, high res timer ticks per sec // (ClockGetres returns 1ns resolution). - "psched": newStaticProcInode(ctx, msrc, []byte(fmt.Sprintf("%08x %08x %08x %08x\n", uint64(time.Microsecond/time.Nanosecond), 64, 1000000, uint64(time.Second/time.Nanosecond)))), - "ptype": newStaticProcInode(ctx, msrc, []byte("Type Device Function\n")), - "route": seqfile.NewSeqFileInode(ctx, &netRoute{s: s}, msrc), - "tcp": seqfile.NewSeqFileInode(ctx, &netTCP{k: k}, msrc), - "udp": seqfile.NewSeqFileInode(ctx, &netUDP{k: k}, msrc), - "unix": seqfile.NewSeqFileInode(ctx, &netUnix{k: k}, msrc), + "psched": newStaticProcInode(t, msrc, []byte(fmt.Sprintf("%08x %08x %08x %08x\n", uint64(time.Microsecond/time.Nanosecond), 64, 1000000, uint64(time.Second/time.Nanosecond)))), + "ptype": newStaticProcInode(t, msrc, []byte("Type Device Function\n")), + "route": seqfile.NewSeqFileInode(t, &netRoute{s: s}, msrc), + "tcp": seqfile.NewSeqFileInode(t, &netTCP{k: k}, msrc), + "udp": seqfile.NewSeqFileInode(t, &netUDP{k: k}, msrc), + "unix": seqfile.NewSeqFileInode(t, &netUnix{k: k}, msrc), } if s.SupportsIPv6() { - contents["if_inet6"] = seqfile.NewSeqFileInode(ctx, &ifinet6{s: s}, msrc) - contents["ipv6_route"] = newStaticProcInode(ctx, msrc, []byte("")) - contents["tcp6"] = seqfile.NewSeqFileInode(ctx, &netTCP6{k: k}, msrc) - contents["udp6"] = newStaticProcInode(ctx, msrc, []byte(" sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode\n")) + contents["if_inet6"] = seqfile.NewSeqFileInode(t, &ifinet6{s: s}, msrc) + contents["ipv6_route"] = newStaticProcInode(t, msrc, []byte("")) + contents["tcp6"] = seqfile.NewSeqFileInode(t, &netTCP6{k: k}, msrc) + contents["udp6"] = newStaticProcInode(t, msrc, []byte(" sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode\n")) } } - d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) - return newProcInode(ctx, d, msrc, fs.SpecialDirectory, nil) + d := ramfs.NewDir(t, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) + return newProcInode(t, d, msrc, fs.SpecialDirectory, t) } // ifinet6 implements seqfile.SeqSource for /proc/net/if_inet6. @@ -837,4 +838,4 @@ func (n *netUDP) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]se return data, 0 } -// LINT.ThenChange(../../fsimpl/proc/tasks_net.go) +// LINT.ThenChange(../../fsimpl/proc/task_net.go) diff --git a/pkg/sentry/fs/proc/proc.go b/pkg/sentry/fs/proc/proc.go index c8abb5052..c659224a7 100644 --- a/pkg/sentry/fs/proc/proc.go +++ b/pkg/sentry/fs/proc/proc.go @@ -70,6 +70,7 @@ func New(ctx context.Context, msrc *fs.MountSource, cgroupControllers map[string "loadavg": seqfile.NewSeqFileInode(ctx, &loadavgData{}, msrc), "meminfo": seqfile.NewSeqFileInode(ctx, &meminfoData{k}, msrc), "mounts": newProcInode(ctx, ramfs.NewSymlink(ctx, fs.RootOwner, "self/mounts"), msrc, fs.Symlink, nil), + "net": newProcInode(ctx, ramfs.NewSymlink(ctx, fs.RootOwner, "self/net"), msrc, fs.Symlink, nil), "self": newSelf(ctx, pidns, msrc), "stat": seqfile.NewSeqFileInode(ctx, &statData{k}, msrc), "thread-self": newThreadSelf(ctx, pidns, msrc), @@ -86,7 +87,6 @@ func New(ctx context.Context, msrc *fs.MountSource, cgroupControllers map[string } // Add more contents that need proc to be initialized. - p.AddChild(ctx, "net", p.newNetDir(ctx, k, msrc)) p.AddChild(ctx, "sys", p.newSysDir(ctx, msrc)) return newProcInode(ctx, p, msrc, fs.SpecialDirectory, nil), nil diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go index 4e9b0fc00..03cc788c8 100644 --- a/pkg/sentry/fs/proc/task.go +++ b/pkg/sentry/fs/proc/task.go @@ -84,6 +84,7 @@ func (p *proc) newTaskDir(t *kernel.Task, msrc *fs.MountSource, isThreadGroup bo "maps": newMaps(t, msrc), "mountinfo": seqfile.NewSeqFileInode(t, &mountInfoFile{t: t}, msrc), "mounts": seqfile.NewSeqFileInode(t, &mountsFile{t: t}, msrc), + "net": newNetDir(t, msrc), "ns": newNamespaceDir(t, msrc), "oom_score": newOOMScore(t, msrc), "oom_score_adj": newOOMScoreAdj(t, msrc), diff --git a/pkg/sentry/fsimpl/proc/BUILD b/pkg/sentry/fsimpl/proc/BUILD index a83245866..bb609a305 100644 --- a/pkg/sentry/fsimpl/proc/BUILD +++ b/pkg/sentry/fsimpl/proc/BUILD @@ -9,9 +9,9 @@ go_library( "subtasks.go", "task.go", "task_files.go", + "task_net.go", "tasks.go", "tasks_files.go", - "tasks_net.go", "tasks_sys.go", ], visibility = ["//pkg/sentry:internal"], diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index c0d643f51..493acbd1b 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -57,6 +57,7 @@ func newTaskInode(inoGen InoGenerator, task *kernel.Task, pidns *kernel.PIDNames "maps": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &mapsData{task: task}), //"mountinfo": seqfile.NewSeqFileInode(t, &mountInfoFile{t: t}, msrc), //"mounts": seqfile.NewSeqFileInode(t, &mountsFile{t: t}, msrc), + "net": newTaskNetDir(task, inoGen), "ns": newTaskOwnedDir(task, inoGen.NextIno(), 0511, map[string]*kernfs.Dentry{ "net": newNamespaceSymlink(task, inoGen.NextIno(), "net"), "pid": newNamespaceSymlink(task, inoGen.NextIno(), "pid"), diff --git a/pkg/sentry/fsimpl/proc/task_net.go b/pkg/sentry/fsimpl/proc/task_net.go new file mode 100644 index 000000000..373a7b17d --- /dev/null +++ b/pkg/sentry/fsimpl/proc/task_net.go @@ -0,0 +1,790 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package proc + +import ( + "bytes" + "fmt" + "io" + "reflect" + "time" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/inet" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/socket" + "gvisor.dev/gvisor/pkg/sentry/socket/unix" + "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/usermem" +) + +func newTaskNetDir(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry { + k := task.Kernel() + pidns := task.PIDNamespace() + root := auth.NewRootCredentials(pidns.UserNamespace()) + + var contents map[string]*kernfs.Dentry + if stack := task.NetworkNamespace().Stack(); stack != nil { + const ( + arp = "IP address HW type Flags HW address Mask Device\n" + netlink = "sk Eth Pid Groups Rmem Wmem Dump Locks Drops Inode\n" + packet = "sk RefCnt Type Proto Iface R Rmem User Inode\n" + protocols = "protocol size sockets memory press maxhdr slab module cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n" + ptype = "Type Device Function\n" + upd6 = " sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode\n" + ) + psched := fmt.Sprintf("%08x %08x %08x %08x\n", uint64(time.Microsecond/time.Nanosecond), 64, 1000000, uint64(time.Second/time.Nanosecond)) + + // TODO(gvisor.dev/issue/1833): Make sure file contents reflect the task + // network namespace. + contents = map[string]*kernfs.Dentry{ + "dev": newDentry(root, inoGen.NextIno(), 0444, &netDevData{stack: stack}), + "snmp": newDentry(root, inoGen.NextIno(), 0444, &netSnmpData{stack: stack}), + + // The following files are simple stubs until they are implemented in + // netstack, if the file contains a header the stub is just the header + // otherwise it is an empty file. + "arp": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(arp)), + "netlink": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(netlink)), + "netstat": newDentry(root, inoGen.NextIno(), 0444, &netStatData{}), + "packet": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(packet)), + "protocols": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(protocols)), + + // Linux sets psched values to: nsec per usec, psched tick in ns, 1000000, + // high res timer ticks per sec (ClockGetres returns 1ns resolution). + "psched": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(psched)), + "ptype": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(ptype)), + "route": newDentry(root, inoGen.NextIno(), 0444, &netRouteData{stack: stack}), + "tcp": newDentry(root, inoGen.NextIno(), 0444, &netTCPData{kernel: k}), + "udp": newDentry(root, inoGen.NextIno(), 0444, &netUDPData{kernel: k}), + "unix": newDentry(root, inoGen.NextIno(), 0444, &netUnixData{kernel: k}), + } + + if stack.SupportsIPv6() { + contents["if_inet6"] = newDentry(root, inoGen.NextIno(), 0444, &ifinet6{stack: stack}) + contents["ipv6_route"] = newDentry(root, inoGen.NextIno(), 0444, newStaticFile("")) + contents["tcp6"] = newDentry(root, inoGen.NextIno(), 0444, &netTCP6Data{kernel: k}) + contents["udp6"] = newDentry(root, inoGen.NextIno(), 0444, newStaticFile(upd6)) + } + } + + return newTaskOwnedDir(task, inoGen.NextIno(), 0555, contents) +} + +// ifinet6 implements vfs.DynamicBytesSource for /proc/net/if_inet6. +// +// +stateify savable +type ifinet6 struct { + kernfs.DynamicBytesFile + + stack inet.Stack +} + +var _ dynamicInode = (*ifinet6)(nil) + +func (n *ifinet6) contents() []string { + var lines []string + nics := n.stack.Interfaces() + for id, naddrs := range n.stack.InterfaceAddrs() { + nic, ok := nics[id] + if !ok { + // NIC was added after NICNames was called. We'll just ignore it. + continue + } + + for _, a := range naddrs { + // IPv6 only. + if a.Family != linux.AF_INET6 { + continue + } + + // Fields: + // IPv6 address displayed in 32 hexadecimal chars without colons + // Netlink device number (interface index) in hexadecimal (use nic id) + // Prefix length in hexadecimal + // Scope value (use 0) + // Interface flags + // Device name + lines = append(lines, fmt.Sprintf("%032x %02x %02x %02x %02x %8s\n", a.Addr, id, a.PrefixLen, 0, a.Flags, nic.Name)) + } + } + return lines +} + +// Generate implements vfs.DynamicBytesSource.Generate. +func (n *ifinet6) Generate(ctx context.Context, buf *bytes.Buffer) error { + for _, l := range n.contents() { + buf.WriteString(l) + } + return nil +} + +// netDevData implements vfs.DynamicBytesSource for /proc/net/dev. +// +// +stateify savable +type netDevData struct { + kernfs.DynamicBytesFile + + stack inet.Stack +} + +var _ dynamicInode = (*netDevData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (n *netDevData) Generate(ctx context.Context, buf *bytes.Buffer) error { + interfaces := n.stack.Interfaces() + buf.WriteString("Inter-| Receive | Transmit\n") + buf.WriteString(" face |bytes packets errs drop fifo frame compressed multicast|bytes packets errs drop fifo colls carrier compressed\n") + + for _, i := range interfaces { + // Implements the same format as + // net/core/net-procfs.c:dev_seq_printf_stats. + var stats inet.StatDev + if err := n.stack.Statistics(&stats, i.Name); err != nil { + log.Warningf("Failed to retrieve interface statistics for %v: %v", i.Name, err) + continue + } + fmt.Fprintf( + buf, + "%6s: %7d %7d %4d %4d %4d %5d %10d %9d %8d %7d %4d %4d %4d %5d %7d %10d\n", + i.Name, + // Received + stats[0], // bytes + stats[1], // packets + stats[2], // errors + stats[3], // dropped + stats[4], // fifo + stats[5], // frame + stats[6], // compressed + stats[7], // multicast + // Transmitted + stats[8], // bytes + stats[9], // packets + stats[10], // errors + stats[11], // dropped + stats[12], // fifo + stats[13], // frame + stats[14], // compressed + stats[15], // multicast + ) + } + + return nil +} + +// netUnixData implements vfs.DynamicBytesSource for /proc/net/unix. +// +// +stateify savable +type netUnixData struct { + kernfs.DynamicBytesFile + + kernel *kernel.Kernel +} + +var _ dynamicInode = (*netUnixData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (n *netUnixData) Generate(ctx context.Context, buf *bytes.Buffer) error { + buf.WriteString("Num RefCount Protocol Flags Type St Inode Path\n") + for _, se := range n.kernel.ListSockets() { + s := se.Sock.Get() + if s == nil { + log.Debugf("Couldn't resolve weakref %v in socket table, racing with destruction?", se.Sock) + continue + } + sfile := s.(*fs.File) + if family, _, _ := sfile.FileOperations.(socket.Socket).Type(); family != linux.AF_UNIX { + s.DecRef() + // Not a unix socket. + continue + } + sops := sfile.FileOperations.(*unix.SocketOperations) + + addr, err := sops.Endpoint().GetLocalAddress() + if err != nil { + log.Warningf("Failed to retrieve socket name from %+v: %v", sfile, err) + addr.Addr = "" + } + + sockFlags := 0 + if ce, ok := sops.Endpoint().(transport.ConnectingEndpoint); ok { + if ce.Listening() { + // For unix domain sockets, linux reports a single flag + // value if the socket is listening, of __SO_ACCEPTCON. + sockFlags = linux.SO_ACCEPTCON + } + } + + // In the socket entry below, the value for the 'Num' field requires + // some consideration. Linux prints the address to the struct + // unix_sock representing a socket in the kernel, but may redact the + // value for unprivileged users depending on the kptr_restrict + // sysctl. + // + // One use for this field is to allow a privileged user to + // introspect into the kernel memory to determine information about + // a socket not available through procfs, such as the socket's peer. + // + // In gvisor, returning a pointer to our internal structures would + // be pointless, as it wouldn't match the memory layout for struct + // unix_sock, making introspection difficult. We could populate a + // struct unix_sock with the appropriate data, but even that + // requires consideration for which kernel version to emulate, as + // the definition of this struct changes over time. + // + // For now, we always redact this pointer. + fmt.Fprintf(buf, "%#016p: %08X %08X %08X %04X %02X %5d", + (*unix.SocketOperations)(nil), // Num, pointer to kernel socket struct. + sfile.ReadRefs()-1, // RefCount, don't count our own ref. + 0, // Protocol, always 0 for UDS. + sockFlags, // Flags. + sops.Endpoint().Type(), // Type. + sops.State(), // State. + sfile.InodeID(), // Inode. + ) + + // Path + if len(addr.Addr) != 0 { + if addr.Addr[0] == 0 { + // Abstract path. + fmt.Fprintf(buf, " @%s", string(addr.Addr[1:])) + } else { + fmt.Fprintf(buf, " %s", string(addr.Addr)) + } + } + fmt.Fprintf(buf, "\n") + + s.DecRef() + } + return nil +} + +func networkToHost16(n uint16) uint16 { + // n is in network byte order, so is big-endian. The most-significant byte + // should be stored in the lower address. + // + // We manually inline binary.BigEndian.Uint16() because Go does not support + // non-primitive consts, so binary.BigEndian is a (mutable) var, so calls to + // binary.BigEndian.Uint16() require a read of binary.BigEndian and an + // interface method call, defeating inlining. + buf := [2]byte{byte(n >> 8 & 0xff), byte(n & 0xff)} + return usermem.ByteOrder.Uint16(buf[:]) +} + +func writeInetAddr(w io.Writer, family int, i linux.SockAddr) { + switch family { + case linux.AF_INET: + var a linux.SockAddrInet + if i != nil { + a = *i.(*linux.SockAddrInet) + } + + // linux.SockAddrInet.Port is stored in the network byte order and is + // printed like a number in host byte order. Note that all numbers in host + // byte order are printed with the most-significant byte first when + // formatted with %X. See get_tcp4_sock() and udp4_format_sock() in Linux. + port := networkToHost16(a.Port) + + // linux.SockAddrInet.Addr is stored as a byte slice in big-endian order + // (i.e. most-significant byte in index 0). Linux represents this as a + // __be32 which is a typedef for an unsigned int, and is printed with + // %X. This means that for a little-endian machine, Linux prints the + // least-significant byte of the address first. To emulate this, we first + // invert the byte order for the address using usermem.ByteOrder.Uint32, + // which makes it have the equivalent encoding to a __be32 on a little + // endian machine. Note that this operation is a no-op on a big endian + // machine. Then similar to Linux, we format it with %X, which will print + // the most-significant byte of the __be32 address first, which is now + // actually the least-significant byte of the original address in + // linux.SockAddrInet.Addr on little endian machines, due to the conversion. + addr := usermem.ByteOrder.Uint32(a.Addr[:]) + + fmt.Fprintf(w, "%08X:%04X ", addr, port) + case linux.AF_INET6: + var a linux.SockAddrInet6 + if i != nil { + a = *i.(*linux.SockAddrInet6) + } + + port := networkToHost16(a.Port) + addr0 := usermem.ByteOrder.Uint32(a.Addr[0:4]) + addr1 := usermem.ByteOrder.Uint32(a.Addr[4:8]) + addr2 := usermem.ByteOrder.Uint32(a.Addr[8:12]) + addr3 := usermem.ByteOrder.Uint32(a.Addr[12:16]) + fmt.Fprintf(w, "%08X%08X%08X%08X:%04X ", addr0, addr1, addr2, addr3, port) + } +} + +func commonGenerateTCP(ctx context.Context, buf *bytes.Buffer, k *kernel.Kernel, family int) error { + // t may be nil here if our caller is not part of a task goroutine. This can + // happen for example if we're here for "sentryctl cat". When t is nil, + // degrade gracefully and retrieve what we can. + t := kernel.TaskFromContext(ctx) + + for _, se := range k.ListSockets() { + s := se.Sock.Get() + if s == nil { + log.Debugf("Couldn't resolve weakref with ID %v in socket table, racing with destruction?", se.ID) + continue + } + sfile := s.(*fs.File) + sops, ok := sfile.FileOperations.(socket.Socket) + if !ok { + panic(fmt.Sprintf("Found non-socket file in socket table: %+v", sfile)) + } + if fa, stype, _ := sops.Type(); !(family == fa && stype == linux.SOCK_STREAM) { + s.DecRef() + // Not tcp4 sockets. + continue + } + + // Linux's documentation for the fields below can be found at + // https://www.kernel.org/doc/Documentation/networking/proc_net_tcp.txt. + // For Linux's implementation, see net/ipv4/tcp_ipv4.c:get_tcp4_sock(). + // Note that the header doesn't contain labels for all the fields. + + // Field: sl; entry number. + fmt.Fprintf(buf, "%4d: ", se.ID) + + // Field: local_adddress. + var localAddr linux.SockAddr + if t != nil { + if local, _, err := sops.GetSockName(t); err == nil { + localAddr = local + } + } + writeInetAddr(buf, family, localAddr) + + // Field: rem_address. + var remoteAddr linux.SockAddr + if t != nil { + if remote, _, err := sops.GetPeerName(t); err == nil { + remoteAddr = remote + } + } + writeInetAddr(buf, family, remoteAddr) + + // Field: state; socket state. + fmt.Fprintf(buf, "%02X ", sops.State()) + + // Field: tx_queue, rx_queue; number of packets in the transmit and + // receive queue. Unimplemented. + fmt.Fprintf(buf, "%08X:%08X ", 0, 0) + + // Field: tr, tm->when; timer active state and number of jiffies + // until timer expires. Unimplemented. + fmt.Fprintf(buf, "%02X:%08X ", 0, 0) + + // Field: retrnsmt; number of unrecovered RTO timeouts. + // Unimplemented. + fmt.Fprintf(buf, "%08X ", 0) + + // Field: uid. + uattr, err := sfile.Dirent.Inode.UnstableAttr(ctx) + if err != nil { + log.Warningf("Failed to retrieve unstable attr for socket file: %v", err) + fmt.Fprintf(buf, "%5d ", 0) + } else { + creds := auth.CredentialsFromContext(ctx) + fmt.Fprintf(buf, "%5d ", uint32(uattr.Owner.UID.In(creds.UserNamespace).OrOverflow())) + } + + // Field: timeout; number of unanswered 0-window probes. + // Unimplemented. + fmt.Fprintf(buf, "%8d ", 0) + + // Field: inode. + fmt.Fprintf(buf, "%8d ", sfile.InodeID()) + + // Field: refcount. Don't count the ref we obtain while deferencing + // the weakref to this socket. + fmt.Fprintf(buf, "%d ", sfile.ReadRefs()-1) + + // Field: Socket struct address. Redacted due to the same reason as + // the 'Num' field in /proc/net/unix, see netUnix.ReadSeqFileData. + fmt.Fprintf(buf, "%#016p ", (*socket.Socket)(nil)) + + // Field: retransmit timeout. Unimplemented. + fmt.Fprintf(buf, "%d ", 0) + + // Field: predicted tick of soft clock (delayed ACK control data). + // Unimplemented. + fmt.Fprintf(buf, "%d ", 0) + + // Field: (ack.quick<<1)|ack.pingpong, Unimplemented. + fmt.Fprintf(buf, "%d ", 0) + + // Field: sending congestion window, Unimplemented. + fmt.Fprintf(buf, "%d ", 0) + + // Field: Slow start size threshold, -1 if threshold >= 0xFFFF. + // Unimplemented, report as large threshold. + fmt.Fprintf(buf, "%d", -1) + + fmt.Fprintf(buf, "\n") + + s.DecRef() + } + + return nil +} + +// netTCPData implements vfs.DynamicBytesSource for /proc/net/tcp. +// +// +stateify savable +type netTCPData struct { + kernfs.DynamicBytesFile + + kernel *kernel.Kernel +} + +var _ dynamicInode = (*netTCPData)(nil) + +func (d *netTCPData) Generate(ctx context.Context, buf *bytes.Buffer) error { + buf.WriteString(" sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode \n") + return commonGenerateTCP(ctx, buf, d.kernel, linux.AF_INET) +} + +// netTCP6Data implements vfs.DynamicBytesSource for /proc/net/tcp6. +// +// +stateify savable +type netTCP6Data struct { + kernfs.DynamicBytesFile + + kernel *kernel.Kernel +} + +var _ dynamicInode = (*netTCP6Data)(nil) + +func (d *netTCP6Data) Generate(ctx context.Context, buf *bytes.Buffer) error { + buf.WriteString(" sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode\n") + return commonGenerateTCP(ctx, buf, d.kernel, linux.AF_INET6) +} + +// netUDPData implements vfs.DynamicBytesSource for /proc/net/udp. +// +// +stateify savable +type netUDPData struct { + kernfs.DynamicBytesFile + + kernel *kernel.Kernel +} + +var _ dynamicInode = (*netUDPData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (d *netUDPData) Generate(ctx context.Context, buf *bytes.Buffer) error { + // t may be nil here if our caller is not part of a task goroutine. This can + // happen for example if we're here for "sentryctl cat". When t is nil, + // degrade gracefully and retrieve what we can. + t := kernel.TaskFromContext(ctx) + + for _, se := range d.kernel.ListSockets() { + s := se.Sock.Get() + if s == nil { + log.Debugf("Couldn't resolve weakref with ID %v in socket table, racing with destruction?", se.ID) + continue + } + sfile := s.(*fs.File) + sops, ok := sfile.FileOperations.(socket.Socket) + if !ok { + panic(fmt.Sprintf("Found non-socket file in socket table: %+v", sfile)) + } + if family, stype, _ := sops.Type(); family != linux.AF_INET || stype != linux.SOCK_DGRAM { + s.DecRef() + // Not udp4 socket. + continue + } + + // For Linux's implementation, see net/ipv4/udp.c:udp4_format_sock(). + + // Field: sl; entry number. + fmt.Fprintf(buf, "%5d: ", se.ID) + + // Field: local_adddress. + var localAddr linux.SockAddrInet + if t != nil { + if local, _, err := sops.GetSockName(t); err == nil { + localAddr = *local.(*linux.SockAddrInet) + } + } + writeInetAddr(buf, linux.AF_INET, &localAddr) + + // Field: rem_address. + var remoteAddr linux.SockAddrInet + if t != nil { + if remote, _, err := sops.GetPeerName(t); err == nil { + remoteAddr = *remote.(*linux.SockAddrInet) + } + } + writeInetAddr(buf, linux.AF_INET, &remoteAddr) + + // Field: state; socket state. + fmt.Fprintf(buf, "%02X ", sops.State()) + + // Field: tx_queue, rx_queue; number of packets in the transmit and + // receive queue. Unimplemented. + fmt.Fprintf(buf, "%08X:%08X ", 0, 0) + + // Field: tr, tm->when. Always 0 for UDP. + fmt.Fprintf(buf, "%02X:%08X ", 0, 0) + + // Field: retrnsmt. Always 0 for UDP. + fmt.Fprintf(buf, "%08X ", 0) + + // Field: uid. + uattr, err := sfile.Dirent.Inode.UnstableAttr(ctx) + if err != nil { + log.Warningf("Failed to retrieve unstable attr for socket file: %v", err) + fmt.Fprintf(buf, "%5d ", 0) + } else { + creds := auth.CredentialsFromContext(ctx) + fmt.Fprintf(buf, "%5d ", uint32(uattr.Owner.UID.In(creds.UserNamespace).OrOverflow())) + } + + // Field: timeout. Always 0 for UDP. + fmt.Fprintf(buf, "%8d ", 0) + + // Field: inode. + fmt.Fprintf(buf, "%8d ", sfile.InodeID()) + + // Field: ref; reference count on the socket inode. Don't count the ref + // we obtain while deferencing the weakref to this socket. + fmt.Fprintf(buf, "%d ", sfile.ReadRefs()-1) + + // Field: Socket struct address. Redacted due to the same reason as + // the 'Num' field in /proc/net/unix, see netUnix.ReadSeqFileData. + fmt.Fprintf(buf, "%#016p ", (*socket.Socket)(nil)) + + // Field: drops; number of dropped packets. Unimplemented. + fmt.Fprintf(buf, "%d", 0) + + fmt.Fprintf(buf, "\n") + + s.DecRef() + } + return nil +} + +// netSnmpData implements vfs.DynamicBytesSource for /proc/net/snmp. +// +// +stateify savable +type netSnmpData struct { + kernfs.DynamicBytesFile + + stack inet.Stack +} + +var _ dynamicInode = (*netSnmpData)(nil) + +type snmpLine struct { + prefix string + header string +} + +var snmp = []snmpLine{ + { + prefix: "Ip", + header: "Forwarding DefaultTTL InReceives InHdrErrors InAddrErrors ForwDatagrams InUnknownProtos InDiscards InDelivers OutRequests OutDiscards OutNoRoutes ReasmTimeout ReasmReqds ReasmOKs ReasmFails FragOKs FragFails FragCreates", + }, + { + prefix: "Icmp", + header: "InMsgs InErrors InCsumErrors InDestUnreachs InTimeExcds InParmProbs InSrcQuenchs InRedirects InEchos InEchoReps InTimestamps InTimestampReps InAddrMasks InAddrMaskReps OutMsgs OutErrors OutDestUnreachs OutTimeExcds OutParmProbs OutSrcQuenchs OutRedirects OutEchos OutEchoReps OutTimestamps OutTimestampReps OutAddrMasks OutAddrMaskReps", + }, + { + prefix: "IcmpMsg", + }, + { + prefix: "Tcp", + header: "RtoAlgorithm RtoMin RtoMax MaxConn ActiveOpens PassiveOpens AttemptFails EstabResets CurrEstab InSegs OutSegs RetransSegs InErrs OutRsts InCsumErrors", + }, + { + prefix: "Udp", + header: "InDatagrams NoPorts InErrors OutDatagrams RcvbufErrors SndbufErrors InCsumErrors IgnoredMulti", + }, + { + prefix: "UdpLite", + header: "InDatagrams NoPorts InErrors OutDatagrams RcvbufErrors SndbufErrors InCsumErrors IgnoredMulti", + }, +} + +func toSlice(a interface{}) []uint64 { + v := reflect.Indirect(reflect.ValueOf(a)) + return v.Slice(0, v.Len()).Interface().([]uint64) +} + +func sprintSlice(s []uint64) string { + if len(s) == 0 { + return "" + } + r := fmt.Sprint(s) + return r[1 : len(r)-1] // Remove "[]" introduced by fmt of slice. +} + +// Generate implements vfs.DynamicBytesSource. +func (d *netSnmpData) Generate(ctx context.Context, buf *bytes.Buffer) error { + types := []interface{}{ + &inet.StatSNMPIP{}, + &inet.StatSNMPICMP{}, + nil, // TODO(gvisor.dev/issue/628): Support IcmpMsg stats. + &inet.StatSNMPTCP{}, + &inet.StatSNMPUDP{}, + &inet.StatSNMPUDPLite{}, + } + for i, stat := range types { + line := snmp[i] + if stat == nil { + fmt.Fprintf(buf, "%s:\n", line.prefix) + fmt.Fprintf(buf, "%s:\n", line.prefix) + continue + } + if err := d.stack.Statistics(stat, line.prefix); err != nil { + if err == syserror.EOPNOTSUPP { + log.Infof("Failed to retrieve %s of /proc/net/snmp: %v", line.prefix, err) + } else { + log.Warningf("Failed to retrieve %s of /proc/net/snmp: %v", line.prefix, err) + } + } + + fmt.Fprintf(buf, "%s: %s\n", line.prefix, line.header) + + if line.prefix == "Tcp" { + tcp := stat.(*inet.StatSNMPTCP) + // "Tcp" needs special processing because MaxConn is signed. RFC 2012. + fmt.Sprintf("%s: %s %d %s\n", line.prefix, sprintSlice(tcp[:3]), int64(tcp[3]), sprintSlice(tcp[4:])) + } else { + fmt.Sprintf("%s: %s\n", line.prefix, sprintSlice(toSlice(stat))) + } + } + return nil +} + +// netRouteData implements vfs.DynamicBytesSource for /proc/net/route. +// +// +stateify savable +type netRouteData struct { + kernfs.DynamicBytesFile + + stack inet.Stack +} + +var _ dynamicInode = (*netRouteData)(nil) + +// Generate implements vfs.DynamicBytesSource. +// See Linux's net/ipv4/fib_trie.c:fib_route_seq_show. +func (d *netRouteData) Generate(ctx context.Context, buf *bytes.Buffer) error { + fmt.Fprintf(buf, "%-127s\n", "Iface\tDestination\tGateway\tFlags\tRefCnt\tUse\tMetric\tMask\tMTU\tWindow\tIRTT") + + interfaces := d.stack.Interfaces() + for _, rt := range d.stack.RouteTable() { + // /proc/net/route only includes ipv4 routes. + if rt.Family != linux.AF_INET { + continue + } + + // /proc/net/route does not include broadcast or multicast routes. + if rt.Type == linux.RTN_BROADCAST || rt.Type == linux.RTN_MULTICAST { + continue + } + + iface, ok := interfaces[rt.OutputInterface] + if !ok || iface.Name == "lo" { + continue + } + + var ( + gw uint32 + prefix uint32 + flags = linux.RTF_UP + ) + if len(rt.GatewayAddr) == header.IPv4AddressSize { + flags |= linux.RTF_GATEWAY + gw = usermem.ByteOrder.Uint32(rt.GatewayAddr) + } + if len(rt.DstAddr) == header.IPv4AddressSize { + prefix = usermem.ByteOrder.Uint32(rt.DstAddr) + } + l := fmt.Sprintf( + "%s\t%08X\t%08X\t%04X\t%d\t%d\t%d\t%08X\t%d\t%d\t%d", + iface.Name, + prefix, + gw, + flags, + 0, // RefCnt. + 0, // Use. + 0, // Metric. + (uint32(1)<> 8 & 0xff), byte(n & 0xff)} - return usermem.ByteOrder.Uint16(buf[:]) -} - -func writeInetAddr(w io.Writer, family int, i linux.SockAddr) { - switch family { - case linux.AF_INET: - var a linux.SockAddrInet - if i != nil { - a = *i.(*linux.SockAddrInet) - } - - // linux.SockAddrInet.Port is stored in the network byte order and is - // printed like a number in host byte order. Note that all numbers in host - // byte order are printed with the most-significant byte first when - // formatted with %X. See get_tcp4_sock() and udp4_format_sock() in Linux. - port := networkToHost16(a.Port) - - // linux.SockAddrInet.Addr is stored as a byte slice in big-endian order - // (i.e. most-significant byte in index 0). Linux represents this as a - // __be32 which is a typedef for an unsigned int, and is printed with - // %X. This means that for a little-endian machine, Linux prints the - // least-significant byte of the address first. To emulate this, we first - // invert the byte order for the address using usermem.ByteOrder.Uint32, - // which makes it have the equivalent encoding to a __be32 on a little - // endian machine. Note that this operation is a no-op on a big endian - // machine. Then similar to Linux, we format it with %X, which will print - // the most-significant byte of the __be32 address first, which is now - // actually the least-significant byte of the original address in - // linux.SockAddrInet.Addr on little endian machines, due to the conversion. - addr := usermem.ByteOrder.Uint32(a.Addr[:]) - - fmt.Fprintf(w, "%08X:%04X ", addr, port) - case linux.AF_INET6: - var a linux.SockAddrInet6 - if i != nil { - a = *i.(*linux.SockAddrInet6) - } - - port := networkToHost16(a.Port) - addr0 := usermem.ByteOrder.Uint32(a.Addr[0:4]) - addr1 := usermem.ByteOrder.Uint32(a.Addr[4:8]) - addr2 := usermem.ByteOrder.Uint32(a.Addr[8:12]) - addr3 := usermem.ByteOrder.Uint32(a.Addr[12:16]) - fmt.Fprintf(w, "%08X%08X%08X%08X:%04X ", addr0, addr1, addr2, addr3, port) - } -} - -func commonGenerateTCP(ctx context.Context, buf *bytes.Buffer, k *kernel.Kernel, family int) error { - // t may be nil here if our caller is not part of a task goroutine. This can - // happen for example if we're here for "sentryctl cat". When t is nil, - // degrade gracefully and retrieve what we can. - t := kernel.TaskFromContext(ctx) - - for _, se := range k.ListSockets() { - s := se.Sock.Get() - if s == nil { - log.Debugf("Couldn't resolve weakref with ID %v in socket table, racing with destruction?", se.ID) - continue - } - sfile := s.(*fs.File) - sops, ok := sfile.FileOperations.(socket.Socket) - if !ok { - panic(fmt.Sprintf("Found non-socket file in socket table: %+v", sfile)) - } - if fa, stype, _ := sops.Type(); !(family == fa && stype == linux.SOCK_STREAM) { - s.DecRef() - // Not tcp4 sockets. - continue - } - - // Linux's documentation for the fields below can be found at - // https://www.kernel.org/doc/Documentation/networking/proc_net_tcp.txt. - // For Linux's implementation, see net/ipv4/tcp_ipv4.c:get_tcp4_sock(). - // Note that the header doesn't contain labels for all the fields. - - // Field: sl; entry number. - fmt.Fprintf(buf, "%4d: ", se.ID) - - // Field: local_adddress. - var localAddr linux.SockAddr - if t != nil { - if local, _, err := sops.GetSockName(t); err == nil { - localAddr = local - } - } - writeInetAddr(buf, family, localAddr) - - // Field: rem_address. - var remoteAddr linux.SockAddr - if t != nil { - if remote, _, err := sops.GetPeerName(t); err == nil { - remoteAddr = remote - } - } - writeInetAddr(buf, family, remoteAddr) - - // Field: state; socket state. - fmt.Fprintf(buf, "%02X ", sops.State()) - - // Field: tx_queue, rx_queue; number of packets in the transmit and - // receive queue. Unimplemented. - fmt.Fprintf(buf, "%08X:%08X ", 0, 0) - - // Field: tr, tm->when; timer active state and number of jiffies - // until timer expires. Unimplemented. - fmt.Fprintf(buf, "%02X:%08X ", 0, 0) - - // Field: retrnsmt; number of unrecovered RTO timeouts. - // Unimplemented. - fmt.Fprintf(buf, "%08X ", 0) - - // Field: uid. - uattr, err := sfile.Dirent.Inode.UnstableAttr(ctx) - if err != nil { - log.Warningf("Failed to retrieve unstable attr for socket file: %v", err) - fmt.Fprintf(buf, "%5d ", 0) - } else { - creds := auth.CredentialsFromContext(ctx) - fmt.Fprintf(buf, "%5d ", uint32(uattr.Owner.UID.In(creds.UserNamespace).OrOverflow())) - } - - // Field: timeout; number of unanswered 0-window probes. - // Unimplemented. - fmt.Fprintf(buf, "%8d ", 0) - - // Field: inode. - fmt.Fprintf(buf, "%8d ", sfile.InodeID()) - - // Field: refcount. Don't count the ref we obtain while deferencing - // the weakref to this socket. - fmt.Fprintf(buf, "%d ", sfile.ReadRefs()-1) - - // Field: Socket struct address. Redacted due to the same reason as - // the 'Num' field in /proc/net/unix, see netUnix.ReadSeqFileData. - fmt.Fprintf(buf, "%#016p ", (*socket.Socket)(nil)) - - // Field: retransmit timeout. Unimplemented. - fmt.Fprintf(buf, "%d ", 0) - - // Field: predicted tick of soft clock (delayed ACK control data). - // Unimplemented. - fmt.Fprintf(buf, "%d ", 0) - - // Field: (ack.quick<<1)|ack.pingpong, Unimplemented. - fmt.Fprintf(buf, "%d ", 0) - - // Field: sending congestion window, Unimplemented. - fmt.Fprintf(buf, "%d ", 0) - - // Field: Slow start size threshold, -1 if threshold >= 0xFFFF. - // Unimplemented, report as large threshold. - fmt.Fprintf(buf, "%d", -1) - - fmt.Fprintf(buf, "\n") - - s.DecRef() - } - - return nil -} - -// netTCPData implements vfs.DynamicBytesSource for /proc/net/tcp. -// -// +stateify savable -type netTCPData struct { - kernfs.DynamicBytesFile - - kernel *kernel.Kernel -} - -var _ dynamicInode = (*netTCPData)(nil) - -func (d *netTCPData) Generate(ctx context.Context, buf *bytes.Buffer) error { - buf.WriteString(" sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode \n") - return commonGenerateTCP(ctx, buf, d.kernel, linux.AF_INET) -} - -// netTCP6Data implements vfs.DynamicBytesSource for /proc/net/tcp6. -// -// +stateify savable -type netTCP6Data struct { - kernfs.DynamicBytesFile - - kernel *kernel.Kernel -} - -var _ dynamicInode = (*netTCP6Data)(nil) - -func (d *netTCP6Data) Generate(ctx context.Context, buf *bytes.Buffer) error { - buf.WriteString(" sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode\n") - return commonGenerateTCP(ctx, buf, d.kernel, linux.AF_INET6) -} - -// netUDPData implements vfs.DynamicBytesSource for /proc/net/udp. -// -// +stateify savable -type netUDPData struct { - kernfs.DynamicBytesFile - - kernel *kernel.Kernel -} - -var _ dynamicInode = (*netUDPData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *netUDPData) Generate(ctx context.Context, buf *bytes.Buffer) error { - // t may be nil here if our caller is not part of a task goroutine. This can - // happen for example if we're here for "sentryctl cat". When t is nil, - // degrade gracefully and retrieve what we can. - t := kernel.TaskFromContext(ctx) - - for _, se := range d.kernel.ListSockets() { - s := se.Sock.Get() - if s == nil { - log.Debugf("Couldn't resolve weakref with ID %v in socket table, racing with destruction?", se.ID) - continue - } - sfile := s.(*fs.File) - sops, ok := sfile.FileOperations.(socket.Socket) - if !ok { - panic(fmt.Sprintf("Found non-socket file in socket table: %+v", sfile)) - } - if family, stype, _ := sops.Type(); family != linux.AF_INET || stype != linux.SOCK_DGRAM { - s.DecRef() - // Not udp4 socket. - continue - } - - // For Linux's implementation, see net/ipv4/udp.c:udp4_format_sock(). - - // Field: sl; entry number. - fmt.Fprintf(buf, "%5d: ", se.ID) - - // Field: local_adddress. - var localAddr linux.SockAddrInet - if t != nil { - if local, _, err := sops.GetSockName(t); err == nil { - localAddr = *local.(*linux.SockAddrInet) - } - } - writeInetAddr(buf, linux.AF_INET, &localAddr) - - // Field: rem_address. - var remoteAddr linux.SockAddrInet - if t != nil { - if remote, _, err := sops.GetPeerName(t); err == nil { - remoteAddr = *remote.(*linux.SockAddrInet) - } - } - writeInetAddr(buf, linux.AF_INET, &remoteAddr) - - // Field: state; socket state. - fmt.Fprintf(buf, "%02X ", sops.State()) - - // Field: tx_queue, rx_queue; number of packets in the transmit and - // receive queue. Unimplemented. - fmt.Fprintf(buf, "%08X:%08X ", 0, 0) - - // Field: tr, tm->when. Always 0 for UDP. - fmt.Fprintf(buf, "%02X:%08X ", 0, 0) - - // Field: retrnsmt. Always 0 for UDP. - fmt.Fprintf(buf, "%08X ", 0) - - // Field: uid. - uattr, err := sfile.Dirent.Inode.UnstableAttr(ctx) - if err != nil { - log.Warningf("Failed to retrieve unstable attr for socket file: %v", err) - fmt.Fprintf(buf, "%5d ", 0) - } else { - creds := auth.CredentialsFromContext(ctx) - fmt.Fprintf(buf, "%5d ", uint32(uattr.Owner.UID.In(creds.UserNamespace).OrOverflow())) - } - - // Field: timeout. Always 0 for UDP. - fmt.Fprintf(buf, "%8d ", 0) - - // Field: inode. - fmt.Fprintf(buf, "%8d ", sfile.InodeID()) - - // Field: ref; reference count on the socket inode. Don't count the ref - // we obtain while deferencing the weakref to this socket. - fmt.Fprintf(buf, "%d ", sfile.ReadRefs()-1) - - // Field: Socket struct address. Redacted due to the same reason as - // the 'Num' field in /proc/net/unix, see netUnix.ReadSeqFileData. - fmt.Fprintf(buf, "%#016p ", (*socket.Socket)(nil)) - - // Field: drops; number of dropped packets. Unimplemented. - fmt.Fprintf(buf, "%d", 0) - - fmt.Fprintf(buf, "\n") - - s.DecRef() - } - return nil -} - -// netSnmpData implements vfs.DynamicBytesSource for /proc/net/snmp. -// -// +stateify savable -type netSnmpData struct { - kernfs.DynamicBytesFile - - stack inet.Stack -} - -var _ dynamicInode = (*netSnmpData)(nil) - -type snmpLine struct { - prefix string - header string -} - -var snmp = []snmpLine{ - { - prefix: "Ip", - header: "Forwarding DefaultTTL InReceives InHdrErrors InAddrErrors ForwDatagrams InUnknownProtos InDiscards InDelivers OutRequests OutDiscards OutNoRoutes ReasmTimeout ReasmReqds ReasmOKs ReasmFails FragOKs FragFails FragCreates", - }, - { - prefix: "Icmp", - header: "InMsgs InErrors InCsumErrors InDestUnreachs InTimeExcds InParmProbs InSrcQuenchs InRedirects InEchos InEchoReps InTimestamps InTimestampReps InAddrMasks InAddrMaskReps OutMsgs OutErrors OutDestUnreachs OutTimeExcds OutParmProbs OutSrcQuenchs OutRedirects OutEchos OutEchoReps OutTimestamps OutTimestampReps OutAddrMasks OutAddrMaskReps", - }, - { - prefix: "IcmpMsg", - }, - { - prefix: "Tcp", - header: "RtoAlgorithm RtoMin RtoMax MaxConn ActiveOpens PassiveOpens AttemptFails EstabResets CurrEstab InSegs OutSegs RetransSegs InErrs OutRsts InCsumErrors", - }, - { - prefix: "Udp", - header: "InDatagrams NoPorts InErrors OutDatagrams RcvbufErrors SndbufErrors InCsumErrors IgnoredMulti", - }, - { - prefix: "UdpLite", - header: "InDatagrams NoPorts InErrors OutDatagrams RcvbufErrors SndbufErrors InCsumErrors IgnoredMulti", - }, -} - -func toSlice(a interface{}) []uint64 { - v := reflect.Indirect(reflect.ValueOf(a)) - return v.Slice(0, v.Len()).Interface().([]uint64) -} - -func sprintSlice(s []uint64) string { - if len(s) == 0 { - return "" - } - r := fmt.Sprint(s) - return r[1 : len(r)-1] // Remove "[]" introduced by fmt of slice. -} - -// Generate implements vfs.DynamicBytesSource. -func (d *netSnmpData) Generate(ctx context.Context, buf *bytes.Buffer) error { - types := []interface{}{ - &inet.StatSNMPIP{}, - &inet.StatSNMPICMP{}, - nil, // TODO(gvisor.dev/issue/628): Support IcmpMsg stats. - &inet.StatSNMPTCP{}, - &inet.StatSNMPUDP{}, - &inet.StatSNMPUDPLite{}, - } - for i, stat := range types { - line := snmp[i] - if stat == nil { - fmt.Fprintf(buf, "%s:\n", line.prefix) - fmt.Fprintf(buf, "%s:\n", line.prefix) - continue - } - if err := d.stack.Statistics(stat, line.prefix); err != nil { - if err == syserror.EOPNOTSUPP { - log.Infof("Failed to retrieve %s of /proc/net/snmp: %v", line.prefix, err) - } else { - log.Warningf("Failed to retrieve %s of /proc/net/snmp: %v", line.prefix, err) - } - } - - fmt.Fprintf(buf, "%s: %s\n", line.prefix, line.header) - - if line.prefix == "Tcp" { - tcp := stat.(*inet.StatSNMPTCP) - // "Tcp" needs special processing because MaxConn is signed. RFC 2012. - fmt.Sprintf("%s: %s %d %s\n", line.prefix, sprintSlice(tcp[:3]), int64(tcp[3]), sprintSlice(tcp[4:])) - } else { - fmt.Sprintf("%s: %s\n", line.prefix, sprintSlice(toSlice(stat))) - } - } - return nil -} - -// netRouteData implements vfs.DynamicBytesSource for /proc/net/route. -// -// +stateify savable -type netRouteData struct { - kernfs.DynamicBytesFile - - stack inet.Stack -} - -var _ dynamicInode = (*netRouteData)(nil) - -// Generate implements vfs.DynamicBytesSource. -// See Linux's net/ipv4/fib_trie.c:fib_route_seq_show. -func (d *netRouteData) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "%-127s\n", "Iface\tDestination\tGateway\tFlags\tRefCnt\tUse\tMetric\tMask\tMTU\tWindow\tIRTT") - - interfaces := d.stack.Interfaces() - for _, rt := range d.stack.RouteTable() { - // /proc/net/route only includes ipv4 routes. - if rt.Family != linux.AF_INET { - continue - } - - // /proc/net/route does not include broadcast or multicast routes. - if rt.Type == linux.RTN_BROADCAST || rt.Type == linux.RTN_MULTICAST { - continue - } - - iface, ok := interfaces[rt.OutputInterface] - if !ok || iface.Name == "lo" { - continue - } - - var ( - gw uint32 - prefix uint32 - flags = linux.RTF_UP - ) - if len(rt.GatewayAddr) == header.IPv4AddressSize { - flags |= linux.RTF_GATEWAY - gw = usermem.ByteOrder.Uint32(rt.GatewayAddr) - } - if len(rt.DstAddr) == header.IPv4AddressSize { - prefix = usermem.ByteOrder.Uint32(rt.DstAddr) - } - l := fmt.Sprintf( - "%s\t%08X\t%08X\t%04X\t%d\t%d\t%d\t%08X\t%d\t%d\t%d", - iface.Name, - prefix, - gw, - flags, - 0, // RefCnt. - 0, // Use. - 0, // Metric. - (uint32(1)< Date: Wed, 11 Mar 2020 09:49:06 -0700 Subject: runsc: don't redirect SIGURG which is used by Go's runtime scheduler Go 1.14+ sends SIGURG to Ms to attempt asynchronous preemption of a G. Since it can't guarantee that a SIGURG is only related to preemption, it continues to forward them to signal.Notify (see runtime.sighandler). When runsc is running a container, there are three processes: a parent process and two children (sandbox and gopher). A parent process sets a signal handler for all signals and redirect them to the container init process. This logic should ignore SIGURG signals. We already ignore them in the Sentry, but it will be better to not notify about them when this is possible. PiperOrigin-RevId: 300345286 --- pkg/sentry/sighandling/sighandling.go | 4 ++++ 1 file changed, 4 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/sighandling/sighandling.go b/pkg/sentry/sighandling/sighandling.go index ba1f9043d..959ef7217 100644 --- a/pkg/sentry/sighandling/sighandling.go +++ b/pkg/sentry/sighandling/sighandling.go @@ -83,6 +83,10 @@ func StartSignalForwarding(handler func(linux.Signal)) func() { // for their handling. var sigchans []chan os.Signal for sig := 1; sig <= numSignals+1; sig++ { + // SIGURG is used by Go's runtime scheduler. + if sig == int(linux.SIGURG) { + continue + } sigchan := make(chan os.Signal, 1) sigchans = append(sigchans, sigchan) signal.Notify(sigchan, syscall.Signal(sig)) -- cgit v1.2.3 From 22d89ef5cb7f3321bc7147b130fcff9f083a8aab Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Wed, 11 Mar 2020 12:00:28 -0700 Subject: Import "unsafe" in bluepill_arm64_unsafe.go This fixes a compile time error: pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go:45:35: undefined: unsafe PiperOrigin-RevId: 300375687 --- pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go | 2 ++ 1 file changed, 2 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go b/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go index 195331383..eb5ed574e 100644 --- a/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go @@ -17,6 +17,8 @@ package kvm import ( + "unsafe" + "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/platform/ring0" ) -- cgit v1.2.3 From 7df936f359766618470ae31a7cbf1b761bd19b59 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Wed, 4 Mar 2020 04:20:36 -0500 Subject: passed the syscall test case 'alarm' on Arm64 platform This issue was caused by 'restart_syscall'. The value of Register R0 should be stored after finishing sysemu. So that we can restore the value and restart syscall. Signed-off-by: Bin Lu --- pkg/sentry/arch/arch_aarch64.go | 4 ++++ pkg/sentry/arch/syscalls_arm64.go | 10 +++++++++- 2 files changed, 13 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/arch/arch_aarch64.go b/pkg/sentry/arch/arch_aarch64.go index 5053393c1..01940bca4 100644 --- a/pkg/sentry/arch/arch_aarch64.go +++ b/pkg/sentry/arch/arch_aarch64.go @@ -97,6 +97,9 @@ type State struct { // FeatureSet is a pointer to the currently active feature set. FeatureSet *cpuid.FeatureSet + + // OrigR0 stores the value of register R0. + OrigR0 uint64 } // Proto returns a protobuf representation of the system registers in State. @@ -146,6 +149,7 @@ func (s *State) Fork() State { Regs: s.Regs, aarch64FPState: s.aarch64FPState.fork(), FeatureSet: s.FeatureSet, + OrigR0: s.OrigR0, } } diff --git a/pkg/sentry/arch/syscalls_arm64.go b/pkg/sentry/arch/syscalls_arm64.go index 00d5ef461..dc13b6124 100644 --- a/pkg/sentry/arch/syscalls_arm64.go +++ b/pkg/sentry/arch/syscalls_arm64.go @@ -50,13 +50,21 @@ func (c *context64) SyscallArgs() SyscallArguments { } // RestartSyscall implements Context.RestartSyscall. +// Prepare for system call restart, OrigR0 will be restored to R0. +// Please see the linux code as reference: +// arch/arm64/kernel/signal.c:do_signal() func (c *context64) RestartSyscall() { c.Regs.Pc -= SyscallWidth - c.Regs.Regs[8] = uint64(restartSyscallNr) + // R0 will be backed up into OrigR0 when entering doSyscall(). + // Please see the linux code as reference: + // arch/arm64/kernel/syscall.c:el0_svc_common(). + // Here we restore it back. + c.Regs.Regs[0] = uint64(c.OrigR0) } // RestartSyscallWithRestartBlock implements Context.RestartSyscallWithRestartBlock. func (c *context64) RestartSyscallWithRestartBlock() { c.Regs.Pc -= SyscallWidth + c.Regs.Regs[0] = uint64(c.OrigR0) c.Regs.Regs[8] = uint64(restartSyscallNr) } -- cgit v1.2.3 From 8f8f16efafd48da3c5e4db329a90bb76620b2324 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Fri, 13 Mar 2020 08:56:47 -0700 Subject: Add support for mount flags Plumbs MS_NOEXEC and MS_RDONLY. Others are TODO. Updates #1623 #1193 PiperOrigin-RevId: 300764669 --- pkg/sentry/fsbridge/vfs.go | 2 -- pkg/sentry/vfs/mount.go | 42 ++++++++++++++++++++++-------------------- pkg/sentry/vfs/options.go | 16 +++++++++++++++- pkg/sentry/vfs/vfs.go | 5 +++++ 4 files changed, 42 insertions(+), 23 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsbridge/vfs.go b/pkg/sentry/fsbridge/vfs.go index 6aa17bfc1..79b808359 100644 --- a/pkg/sentry/fsbridge/vfs.go +++ b/pkg/sentry/fsbridge/vfs.go @@ -115,8 +115,6 @@ func NewVFSLookup(mntns *vfs.MountNamespace, root, workingDir vfs.VirtualDentry) // // remainingTraversals is not configurable in VFS2, all callers are using the // default anyways. -// -// TODO(gvisor.dev/issue/1623): Check mount has read and exec permission. func (l *vfsLookup) OpenPath(ctx context.Context, pathname string, opts vfs.OpenOptions, _ *uint, resolveFinal bool) (File, error) { vfsObj := l.mntns.Root().Mount().Filesystem().VirtualFilesystem() creds := auth.CredentialsFromContext(ctx) diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index 31a4e5480..05f6233f9 100644 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go @@ -74,6 +74,10 @@ type Mount struct { // umounted is true. umounted is protected by VirtualFilesystem.mountMu. umounted bool + // flags contains settings as specified for mount(2), e.g. MS_NOEXEC, except + // for MS_RDONLY which is tracked in "writers". + flags MountFlags + // The lower 63 bits of writers is the number of calls to // Mount.CheckBeginWrite() that have not yet been paired with a call to // Mount.EndWrite(). The MSB of writers is set if MS_RDONLY is in effect. @@ -81,6 +85,21 @@ type Mount struct { writers int64 } +func newMount(vfs *VirtualFilesystem, fs *Filesystem, root *Dentry, mntns *MountNamespace, opts *MountOptions) *Mount { + mnt := &Mount{ + vfs: vfs, + fs: fs, + root: root, + flags: opts.Flags, + ns: mntns, + refs: 1, + } + if opts.ReadOnly { + mnt.setReadOnlyLocked(true) + } + return mnt +} + // A MountNamespace is a collection of Mounts. // // MountNamespaces are reference-counted. Unless otherwise specified, all @@ -129,13 +148,7 @@ func (vfs *VirtualFilesystem) NewMountNamespace(ctx context.Context, creds *auth refs: 1, mountpoints: make(map[*Dentry]uint32), } - mntns.root = &Mount{ - vfs: vfs, - fs: fs, - root: root, - ns: mntns, - refs: 1, - } + mntns.root = newMount(vfs, fs, root, mntns, &MountOptions{}) return mntns, nil } @@ -148,12 +161,7 @@ func (vfs *VirtualFilesystem) NewDisconnectedMount(fs *Filesystem, root *Dentry, if root != nil { root.IncRef() } - return &Mount{ - vfs: vfs, - fs: fs, - root: root, - refs: 1, - }, nil + return newMount(vfs, fs, root, nil /* mntns */, opts), nil } // MountAt creates and mounts a Filesystem configured by the given arguments. @@ -218,13 +226,7 @@ func (vfs *VirtualFilesystem) MountAt(ctx context.Context, creds *auth.Credentia // are directories, or neither are, and returns ENOTDIR if this is not the // case. mntns := vd.mount.ns - mnt := &Mount{ - vfs: vfs, - fs: fs, - root: root, - ns: mntns, - refs: 1, - } + mnt := newMount(vfs, fs, root, mntns, opts) vfs.mounts.seq.BeginWrite() vfs.connectLocked(mnt, vd, mntns) vfs.mounts.seq.EndWrite() diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go index 6af7fdac1..3e90dc4ed 100644 --- a/pkg/sentry/vfs/options.go +++ b/pkg/sentry/vfs/options.go @@ -46,8 +46,21 @@ type MknodOptions struct { DevMinor uint32 } +// MountFlags contains flags as specified for mount(2), e.g. MS_NOEXEC. +// MS_RDONLY is not part of MountFlags because it's tracked in Mount.writers. +type MountFlags struct { + // NoExec is equivalent to MS_NOEXEC. + NoExec bool +} + // MountOptions contains options to VirtualFilesystem.MountAt(). type MountOptions struct { + // Flags contains flags as specified for mount(2), e.g. MS_NOEXEC. + Flags MountFlags + + // ReadOnly is equivalent to MS_RDONLY. + ReadOnly bool + // GetFilesystemOptions contains options to FilesystemType.GetFilesystem(). GetFilesystemOptions GetFilesystemOptions @@ -75,7 +88,8 @@ type OpenOptions struct { // FileExec is set when the file is being opened to be executed. // VirtualFilesystem.OpenAt() checks that the caller has execute permissions - // on the file, and that the file is a regular file. + // on the file, that the file is a regular file, and that the mount doesn't + // have MS_NOEXEC set. FileExec bool } diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index bde81e1ef..365e8b30d 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -388,6 +388,11 @@ func (vfs *VirtualFilesystem) OpenAt(ctx context.Context, creds *auth.Credential // TODO(gvisor.dev/issue/1193): Move inside fsimpl to avoid another call // to FileDescription.Stat(). if opts.FileExec { + if fd.Mount().flags.NoExec { + fd.DecRef() + return nil, syserror.EACCES + } + // Only a regular file can be executed. stat, err := fd.Stat(ctx, StatOptions{Mask: linux.STATX_TYPE}) if err != nil { -- cgit v1.2.3 From f458a325e9b6aecf2ee198de19063505c48a14d7 Mon Sep 17 00:00:00 2001 From: Ting-Yu Wang Date: Fri, 13 Mar 2020 11:25:49 -0700 Subject: Fix "application exiting with {Code:0 Signo:27}" during boot. 2aa9514a06a5e34894e606d508ac2df53b082c74 skips SIGURG, but later code expects the sigchans array contains consecutive signal numbers. PiperOrigin-RevId: 300793450 --- pkg/sentry/sighandling/sighandling.go | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/sighandling/sighandling.go b/pkg/sentry/sighandling/sighandling.go index 959ef7217..83195d5a1 100644 --- a/pkg/sentry/sighandling/sighandling.go +++ b/pkg/sentry/sighandling/sighandling.go @@ -83,12 +83,13 @@ func StartSignalForwarding(handler func(linux.Signal)) func() { // for their handling. var sigchans []chan os.Signal for sig := 1; sig <= numSignals+1; sig++ { + sigchan := make(chan os.Signal, 1) + sigchans = append(sigchans, sigchan) + // SIGURG is used by Go's runtime scheduler. if sig == int(linux.SIGURG) { continue } - sigchan := make(chan os.Signal, 1) - sigchans = append(sigchans, sigchan) signal.Notify(sigchan, syscall.Signal(sig)) } // Start up our listener. -- cgit v1.2.3 From 2e38408f20a084de716962d4631e0fec1fd16c16 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 13 Mar 2020 11:40:13 -0700 Subject: Implement access/faccessat for VFS2. Note that the raw faccessat system call does not actually take a flags argument; according to faccessat(2), the glibc wrapper implements the flags by using fstatat(2). Remove the flag argument that we try to extract from vfs1, which would just be a garbage value. Updates #1965 Fixes #2101 PiperOrigin-RevId: 300796067 --- pkg/sentry/fsimpl/ext/filesystem.go | 10 ++++++ pkg/sentry/fsimpl/gofer/filesystem.go | 13 ++++++++ pkg/sentry/fsimpl/kernfs/filesystem.go | 14 ++++++++ pkg/sentry/fsimpl/tmpfs/filesystem.go | 12 +++++++ pkg/sentry/syscalls/linux/sys_file.go | 16 ++++++--- pkg/sentry/syscalls/linux/vfs2/stat.go | 60 +++++++++++++++++++++++++++++++--- pkg/sentry/vfs/anonfs.go | 25 +++++++++++--- pkg/sentry/vfs/filesystem.go | 4 +++ pkg/sentry/vfs/vfs.go | 17 ++++++++++ 9 files changed, 157 insertions(+), 14 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/ext/filesystem.go b/pkg/sentry/fsimpl/ext/filesystem.go index e05429d41..8497be615 100644 --- a/pkg/sentry/fsimpl/ext/filesystem.go +++ b/pkg/sentry/fsimpl/ext/filesystem.go @@ -22,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" @@ -255,6 +256,15 @@ func (fs *filesystem) statTo(stat *linux.Statfs) { // TODO(b/134676337): Set Statfs.Flags and Statfs.FSID. } +// AccessAt implements vfs.Filesystem.Impl.AccessAt. +func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error { + _, inode, err := fs.walk(rp, false) + if err != nil { + return err + } + return inode.checkPermissions(rp.Credentials(), ats) +} + // GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { vfsd, inode, err := fs.walk(rp, false) diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 5cfb0dc4c..38e4cdbc5 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/p9" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" ) @@ -499,6 +500,18 @@ func (fs *filesystem) renameMuUnlockAndCheckCaching(ds **[]*dentry) { putDentrySlice(*ds) } +// AccessAt implements vfs.Filesystem.Impl.AccessAt. +func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return err + } + return d.checkPermissions(creds, ats, d.isDir()) +} + // GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { var ds *[]*dentry diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 1d7e04ad4..3288de290 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -22,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" ) @@ -229,6 +230,19 @@ func (fs *Filesystem) Sync(ctx context.Context) error { return nil } +// AccessAt implements vfs.Filesystem.Impl.AccessAt. +func (fs *Filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error { + fs.mu.RLock() + defer fs.mu.RUnlock() + defer fs.processDeferredDecRefs() + + _, inode, err := fs.walkExistingLocked(ctx, rp) + if err != nil { + return err + } + return inode.CheckPermissions(ctx, creds, ats) +} + // GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. func (fs *Filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { fs.mu.RLock() diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index e1b551422..02637fca6 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -20,6 +20,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" ) @@ -154,6 +155,17 @@ func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(pa return create(parent, name) } +// AccessAt implements vfs.Filesystem.Impl.AccessAt. +func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error { + fs.mu.RLock() + defer fs.mu.RUnlock() + d, err := resolveLocked(rp) + if err != nil { + return err + } + return d.inode.checkPermissions(creds, ats, d.inode.isDir()) +} + // GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { fs.mu.RLock() diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go index d10a9bed8..35a98212a 100644 --- a/pkg/sentry/syscalls/linux/sys_file.go +++ b/pkg/sentry/syscalls/linux/sys_file.go @@ -514,7 +514,7 @@ func (ac accessContext) Value(key interface{}) interface{} { } } -func accessAt(t *kernel.Task, dirFD int32, addr usermem.Addr, resolve bool, mode uint) error { +func accessAt(t *kernel.Task, dirFD int32, addr usermem.Addr, mode uint) error { const rOK = 4 const wOK = 2 const xOK = 1 @@ -529,7 +529,7 @@ func accessAt(t *kernel.Task, dirFD int32, addr usermem.Addr, resolve bool, mode return syserror.EINVAL } - return fileOpOn(t, dirFD, path, resolve, func(root *fs.Dirent, d *fs.Dirent, _ uint) error { + return fileOpOn(t, dirFD, path, true /* resolve */, func(root *fs.Dirent, d *fs.Dirent, _ uint) error { // access(2) and faccessat(2) check permissions using real // UID/GID, not effective UID/GID. // @@ -564,17 +564,23 @@ func Access(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal addr := args[0].Pointer() mode := args[1].ModeT() - return 0, nil, accessAt(t, linux.AT_FDCWD, addr, true, mode) + return 0, nil, accessAt(t, linux.AT_FDCWD, addr, mode) } // Faccessat implements linux syscall faccessat(2). +// +// Note that the faccessat() system call does not take a flags argument: +// "The raw faccessat() system call takes only the first three arguments. The +// AT_EACCESS and AT_SYMLINK_NOFOLLOW flags are actually implemented within +// the glibc wrapper function for faccessat(). If either of these flags is +// specified, then the wrapper function employs fstatat(2) to determine access +// permissions." - faccessat(2) func Faccessat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { dirFD := args[0].Int() addr := args[1].Pointer() mode := args[2].ModeT() - flags := args[3].Int() - return 0, nil, accessAt(t, dirFD, addr, flags&linux.AT_SYMLINK_NOFOLLOW == 0, mode) + return 0, nil, accessAt(t, dirFD, addr, mode) } // LINT.ThenChange(vfs2/filesystem.go) diff --git a/pkg/sentry/syscalls/linux/vfs2/stat.go b/pkg/sentry/syscalls/linux/vfs2/stat.go index 12c532310..a74ea6fd5 100644 --- a/pkg/sentry/syscalls/linux/vfs2/stat.go +++ b/pkg/sentry/syscalls/linux/vfs2/stat.go @@ -228,14 +228,64 @@ func Readlink(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc // Access implements Linux syscall access(2). func Access(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - // FIXME(jamieliu): actually implement - return 0, nil, nil + addr := args[0].Pointer() + mode := args[1].ModeT() + + return 0, nil, accessAt(t, linux.AT_FDCWD, addr, mode) } -// Faccessat implements Linux syscall access(2). +// Faccessat implements Linux syscall faccessat(2). +// +// Note that the faccessat() system call does not take a flags argument: +// "The raw faccessat() system call takes only the first three arguments. The +// AT_EACCESS and AT_SYMLINK_NOFOLLOW flags are actually implemented within +// the glibc wrapper function for faccessat(). If either of these flags is +// specified, then the wrapper function employs fstatat(2) to determine access +// permissions." - faccessat(2) func Faccessat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - // FIXME(jamieliu): actually implement - return 0, nil, nil + dirfd := args[0].Int() + addr := args[1].Pointer() + mode := args[2].ModeT() + + return 0, nil, accessAt(t, dirfd, addr, mode) +} + +func accessAt(t *kernel.Task, dirfd int32, pathAddr usermem.Addr, mode uint) error { + const rOK = 4 + const wOK = 2 + const xOK = 1 + + // Sanity check the mode. + if mode&^(rOK|wOK|xOK) != 0 { + return syserror.EINVAL + } + + path, err := copyInPath(t, pathAddr) + if err != nil { + return err + } + tpop, err := getTaskPathOperation(t, dirfd, path, disallowEmptyPath, followFinalSymlink) + if err != nil { + return err + } + + // access(2) and faccessat(2) check permissions using real + // UID/GID, not effective UID/GID. + // + // "access() needs to use the real uid/gid, not the effective + // uid/gid. We do this by temporarily clearing all FS-related + // capabilities and switching the fsuid/fsgid around to the + // real ones." -fs/open.c:faccessat + creds := t.Credentials().Fork() + creds.EffectiveKUID = creds.RealKUID + creds.EffectiveKGID = creds.RealKGID + if creds.EffectiveKUID.In(creds.UserNamespace) == auth.RootUID { + creds.EffectiveCaps = creds.PermittedCaps + } else { + creds.EffectiveCaps = 0 + } + + return t.Kernel().VFS().AccessAt(t, creds, vfs.AccessTypes(mode), &tpop.pop) } // Readlinkat implements Linux syscall mknodat(2). diff --git a/pkg/sentry/vfs/anonfs.go b/pkg/sentry/vfs/anonfs.go index 2db25be49..925996517 100644 --- a/pkg/sentry/vfs/anonfs.go +++ b/pkg/sentry/vfs/anonfs.go @@ -41,7 +41,14 @@ func (vfs *VirtualFilesystem) NewAnonVirtualDentry(name string) VirtualDentry { } } -const anonfsBlockSize = usermem.PageSize // via fs/libfs.c:pseudo_fs_fill_super() +const ( + anonfsBlockSize = usermem.PageSize // via fs/libfs.c:pseudo_fs_fill_super() + + // Mode, UID, and GID for a generic anonfs file. + anonFileMode = 0600 // no type is correct + anonFileUID = auth.RootKUID + anonFileGID = auth.RootKGID +) // anonFilesystem is the implementation of FilesystemImpl that backs // VirtualDentries returned by VirtualFilesystem.NewAnonVirtualDentry(). @@ -69,6 +76,16 @@ func (fs *anonFilesystem) Sync(ctx context.Context) error { return nil } +// AccessAt implements vfs.Filesystem.Impl.AccessAt. +// +// TODO(gvisor.dev/issue/1965): Implement access permissions. +func (fs *anonFilesystem) AccessAt(ctx context.Context, rp *ResolvingPath, creds *auth.Credentials, ats AccessTypes) error { + if !rp.Done() { + return syserror.ENOTDIR + } + return GenericCheckPermissions(creds, ats, false /* isDir */, anonFileMode, anonFileUID, anonFileGID) +} + // GetDentryAt implements FilesystemImpl.GetDentryAt. func (fs *anonFilesystem) GetDentryAt(ctx context.Context, rp *ResolvingPath, opts GetDentryOptions) (*Dentry, error) { if !rp.Done() { @@ -167,9 +184,9 @@ func (fs *anonFilesystem) StatAt(ctx context.Context, rp *ResolvingPath, opts St Mask: linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_NLINK | linux.STATX_UID | linux.STATX_GID | linux.STATX_INO | linux.STATX_SIZE | linux.STATX_BLOCKS, Blksize: anonfsBlockSize, Nlink: 1, - UID: uint32(auth.RootKUID), - GID: uint32(auth.RootKGID), - Mode: 0600, // no type is correct + UID: uint32(anonFileUID), + GID: uint32(anonFileGID), + Mode: anonFileMode, Ino: 1, Size: 0, Blocks: 0, diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index 556976d0b..c43dcff3d 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -20,6 +20,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ) // A Filesystem is a tree of nodes represented by Dentries, which forms part of @@ -144,6 +145,9 @@ type FilesystemImpl interface { // file data to be written to the underlying [filesystem]", as by syncfs(2). Sync(ctx context.Context) error + // AccessAt checks whether a user with creds can access the file at rp. + AccessAt(ctx context.Context, rp *ResolvingPath, creds *auth.Credentials, ats AccessTypes) error + // GetDentryAt returns a Dentry representing the file at rp. A reference is // taken on the returned Dentry. // diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 365e8b30d..2e2880171 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -174,6 +174,23 @@ type PathOperation struct { FollowFinalSymlink bool } +// AccessAt checks whether a user with creds has access to the file at +// the given path. +func (vfs *VirtualFilesystem) AccessAt(ctx context.Context, creds *auth.Credentials, ats AccessTypes, pop *PathOperation) error { + rp := vfs.getResolvingPath(creds, pop) + for { + err := rp.mount.fs.impl.AccessAt(ctx, rp, creds, ats) + if err == nil { + vfs.putResolvingPath(rp) + return nil + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return err + } + } +} + // GetDentryAt returns a VirtualDentry representing the given path, at which a // file must exist. A reference is taken on the returned VirtualDentry. func (vfs *VirtualFilesystem) GetDentryAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *GetDentryOptions) (VirtualDentry, error) { -- cgit v1.2.3 From b78cee3bae142eb5c602d51874d0cbad274777e2 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Fri, 13 Mar 2020 12:16:59 -0700 Subject: Fix lock recursion in kernel.ProcessGroup.SendSignal(). PiperOrigin-RevId: 300803515 --- pkg/sentry/kernel/sessions.go | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/sessions.go b/pkg/sentry/kernel/sessions.go index 047b5214d..0e19286de 100644 --- a/pkg/sentry/kernel/sessions.go +++ b/pkg/sentry/kernel/sessions.go @@ -246,7 +246,7 @@ func (pg *ProcessGroup) SendSignal(info *arch.SignalInfo) error { var lastErr error for tg := range tasks.Root.tgids { - if tg.ProcessGroup() == pg { + if tg.processGroup == pg { tg.signalHandlers.mu.Lock() infoCopy := *info if err := tg.leader.sendSignalLocked(&infoCopy, true /*group*/); err != nil { -- cgit v1.2.3 From 1c0535297067179a822ba2dd9a6fe13a8be5a666 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Fri, 13 Mar 2020 13:17:59 -0700 Subject: Fix oom_score_adj. - Make oomScoreAdj a ThreadGroup field (Linux: signal_struct::oom_score_adj). - Avoid deadlock caused by Task.OOMScoreAdj()/SetOOMScoreAdj() locking Task.mu and TaskSet.mu in the wrong order (via Task.ExitState()). PiperOrigin-RevId: 300814698 --- pkg/sentry/fs/proc/task.go | 17 ++++++++++------- pkg/sentry/fsimpl/proc/task_files.go | 10 ++++++---- pkg/sentry/kernel/task.go | 29 ++++++----------------------- pkg/sentry/kernel/task_clone.go | 9 +++------ pkg/sentry/kernel/task_start.go | 4 ---- pkg/sentry/kernel/thread_group.go | 7 +++++++ 6 files changed, 32 insertions(+), 44 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go index 03cc788c8..d6c5dd2c1 100644 --- a/pkg/sentry/fs/proc/task.go +++ b/pkg/sentry/fs/proc/task.go @@ -853,15 +853,15 @@ func (o *oomScoreAdj) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.F // Read implements fs.FileOperations.Read. func (f *oomScoreAdjFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { - if offset != 0 { - return 0, io.EOF + if f.t.ExitState() == kernel.TaskExitDead { + return 0, syserror.ESRCH } - adj, err := f.t.OOMScoreAdj() - if err != nil { - return 0, err + var buf bytes.Buffer + fmt.Fprintf(&buf, "%d\n", f.t.OOMScoreAdj()) + if offset >= int64(buf.Len()) { + return 0, io.EOF } - adjBytes := []byte(strconv.FormatInt(int64(adj), 10) + "\n") - n, err := dst.CopyOut(ctx, adjBytes) + n, err := dst.CopyOut(ctx, buf.Bytes()[offset:]) return int64(n), err } @@ -880,6 +880,9 @@ func (f *oomScoreAdjFile) Write(ctx context.Context, _ *fs.File, src usermem.IOS return 0, err } + if f.t.ExitState() == kernel.TaskExitDead { + return 0, syserror.ESRCH + } if err := f.t.SetOOMScoreAdj(v); err != nil { return 0, err } diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index 5a231ac86..4d3332771 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -539,11 +539,10 @@ var _ vfs.WritableDynamicBytesSource = (*oomScoreAdj)(nil) // Generate implements vfs.DynamicBytesSource.Generate. func (o *oomScoreAdj) Generate(ctx context.Context, buf *bytes.Buffer) error { - adj, err := o.task.OOMScoreAdj() - if err != nil { - return err + if o.task.ExitState() == kernel.TaskExitDead { + return syserror.ESRCH } - fmt.Fprintf(buf, "%d\n", adj) + fmt.Fprintf(buf, "%d\n", o.task.OOMScoreAdj()) return nil } @@ -562,6 +561,9 @@ func (o *oomScoreAdj) Write(ctx context.Context, src usermem.IOSequence, offset return 0, err } + if o.task.ExitState() == kernel.TaskExitDead { + return 0, syserror.ESRCH + } if err := o.task.SetOOMScoreAdj(v); err != nil { return 0, err } diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index c0dbbe890..8452ddf5b 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -555,13 +555,6 @@ type Task struct { // // startTime is protected by mu. startTime ktime.Time - - // oomScoreAdj is the task's OOM score adjustment. This is currently not - // used but is maintained for consistency. - // TODO(gvisor.dev/issue/1967) - // - // oomScoreAdj is protected by mu, and is owned by the task goroutine. - oomScoreAdj int32 } func (t *Task) savePtraceTracer() *Task { @@ -856,27 +849,17 @@ func (t *Task) ContainerID() string { return t.containerID } -// OOMScoreAdj gets the task's OOM score adjustment. -func (t *Task) OOMScoreAdj() (int32, error) { - t.mu.Lock() - defer t.mu.Unlock() - if t.ExitState() == TaskExitDead { - return 0, syserror.ESRCH - } - return t.oomScoreAdj, nil +// OOMScoreAdj gets the task's thread group's OOM score adjustment. +func (t *Task) OOMScoreAdj() int32 { + return atomic.LoadInt32(&t.tg.oomScoreAdj) } -// SetOOMScoreAdj sets the task's OOM score adjustment. The value should be -// between -1000 and 1000 inclusive. +// SetOOMScoreAdj sets the task's thread group's OOM score adjustment. The +// value should be between -1000 and 1000 inclusive. func (t *Task) SetOOMScoreAdj(adj int32) error { - t.mu.Lock() - defer t.mu.Unlock() - if t.ExitState() == TaskExitDead { - return syserror.ESRCH - } if adj > 1000 || adj < -1000 { return syserror.EINVAL } - t.oomScoreAdj = adj + atomic.StoreInt32(&t.tg.oomScoreAdj, adj) return nil } diff --git a/pkg/sentry/kernel/task_clone.go b/pkg/sentry/kernel/task_clone.go index dda502bb8..e1ecca99e 100644 --- a/pkg/sentry/kernel/task_clone.go +++ b/pkg/sentry/kernel/task_clone.go @@ -15,6 +15,8 @@ package kernel import ( + "sync/atomic" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/bpf" "gvisor.dev/gvisor/pkg/sentry/inet" @@ -260,15 +262,11 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { sh = sh.Fork() } tg = t.k.NewThreadGroup(tg.mounts, pidns, sh, opts.TerminationSignal, tg.limits.GetCopy()) + tg.oomScoreAdj = atomic.LoadInt32(&t.tg.oomScoreAdj) rseqAddr = t.rseqAddr rseqSignature = t.rseqSignature } - adj, err := t.OOMScoreAdj() - if err != nil { - return 0, nil, err - } - cfg := &TaskConfig{ Kernel: t.k, ThreadGroup: tg, @@ -287,7 +285,6 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { RSeqAddr: rseqAddr, RSeqSignature: rseqSignature, ContainerID: t.ContainerID(), - OOMScoreAdj: adj, } if opts.NewThreadGroup { cfg.Parent = t diff --git a/pkg/sentry/kernel/task_start.go b/pkg/sentry/kernel/task_start.go index 2bbf48bb8..a5035bb7f 100644 --- a/pkg/sentry/kernel/task_start.go +++ b/pkg/sentry/kernel/task_start.go @@ -93,9 +93,6 @@ type TaskConfig struct { // ContainerID is the container the new task belongs to. ContainerID string - - // oomScoreAdj is the task's OOM score adjustment. - OOMScoreAdj int32 } // NewTask creates a new task defined by cfg. @@ -146,7 +143,6 @@ func (ts *TaskSet) newTask(cfg *TaskConfig) (*Task, error) { rseqSignature: cfg.RSeqSignature, futexWaiter: futex.NewWaiter(), containerID: cfg.ContainerID, - oomScoreAdj: cfg.OOMScoreAdj, } t.creds.Store(cfg.Credentials) t.endStopCond.L = &t.tg.signalHandlers.mu diff --git a/pkg/sentry/kernel/thread_group.go b/pkg/sentry/kernel/thread_group.go index 268f62e9d..52849f5b3 100644 --- a/pkg/sentry/kernel/thread_group.go +++ b/pkg/sentry/kernel/thread_group.go @@ -254,6 +254,13 @@ type ThreadGroup struct { // // tty is protected by the signal mutex. tty *TTY + + // oomScoreAdj is the thread group's OOM score adjustment. This is + // currently not used but is maintained for consistency. + // TODO(gvisor.dev/issue/1967) + // + // oomScoreAdj is accessed using atomic memory operations. + oomScoreAdj int32 } // NewThreadGroup returns a new, empty thread group in PID namespace pidns. The -- cgit v1.2.3 From b0f2c3e7646df603156f1b8e8b3382f33353eb04 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Fri, 13 Mar 2020 16:08:06 -0700 Subject: Fix infinite loop in semaphore.sem.wakeWaiters(). PiperOrigin-RevId: 300845134 --- pkg/sentry/kernel/semaphore/semaphore.go | 1 + 1 file changed, 1 insertion(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/semaphore/semaphore.go b/pkg/sentry/kernel/semaphore/semaphore.go index 1000f3287..c00fa1138 100644 --- a/pkg/sentry/kernel/semaphore/semaphore.go +++ b/pkg/sentry/kernel/semaphore/semaphore.go @@ -554,6 +554,7 @@ func (s *sem) wakeWaiters() { for w := s.waiters.Front(); w != nil; { if s.value < w.value { // Still blocked, skip it. + w = w.Next() continue } w.ch <- struct{}{} -- cgit v1.2.3 From 829beebf0b67e20e50dd5ec4a5030636e38cc576 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Fri, 13 Mar 2020 17:16:59 -0700 Subject: Panic if file in FDTable has been destroyed This will give more information about the file to identify where possibly the extra DecRef() would be. PiperOrigin-RevId: 300855874 --- pkg/sentry/kernel/fd_table.go | 11 +++++++++++ 1 file changed, 11 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/fd_table.go b/pkg/sentry/kernel/fd_table.go index 58001d56c..00f914564 100644 --- a/pkg/sentry/kernel/fd_table.go +++ b/pkg/sentry/kernel/fd_table.go @@ -195,6 +195,8 @@ func (f *FDTable) Size() int { // // It is the caller's responsibility to acquire an appropriate lock. func (f *FDTable) forEach(fn func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags)) { + // retries tracks the number of failed TryIncRef attempts for the same FD. + retries := 0 fd := int32(0) for { file, fileVFS2, flags, ok := f.getAll(fd) @@ -204,17 +206,26 @@ func (f *FDTable) forEach(fn func(fd int32, file *fs.File, fileVFS2 *vfs.FileDes switch { case file != nil: if !file.TryIncRef() { + retries++ + if retries > 1000 { + panic(fmt.Sprintf("File in FD table has been destroyed. FD: %d, File: %+v, FileOps: %+v", fd, file, file.FileOperations)) + } continue // Race caught. } fn(fd, file, nil, flags) file.DecRef() case fileVFS2 != nil: if !fileVFS2.TryIncRef() { + retries++ + if retries > 1000 { + panic(fmt.Sprintf("File in FD table has been destroyed. FD: %d, File: %+v, Impl: %+v", fd, fileVFS2, fileVFS2.Impl())) + } continue // Race caught. } fn(fd, nil, fileVFS2, flags) fileVFS2.DecRef() } + retries = 0 fd++ } } -- cgit v1.2.3 From 45a8ae240dd180f1b8b4c56e77ac67e4cd3af96f Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Fri, 13 Mar 2020 18:56:05 -0700 Subject: Add remaining procfs files Closes #1195 PiperOrigin-RevId: 300867055 --- pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 8 + pkg/sentry/fsimpl/kernfs/kernfs.go | 2 +- pkg/sentry/fsimpl/proc/BUILD | 4 + pkg/sentry/fsimpl/proc/subtasks.go | 6 +- pkg/sentry/fsimpl/proc/task.go | 26 +-- pkg/sentry/fsimpl/proc/task_fds.go | 287 ++++++++++++++++++++++++++++ pkg/sentry/fsimpl/proc/task_files.go | 251 +++++++++++++++++++++++- pkg/sentry/fsimpl/proc/tasks.go | 30 ++- pkg/sentry/fsimpl/proc/tasks_files.go | 52 +++-- pkg/sentry/fsimpl/proc/tasks_test.go | 87 +++++++-- pkg/sentry/fsimpl/testutil/BUILD | 2 + pkg/sentry/fsimpl/testutil/kernel.go | 27 ++- pkg/sentry/kernel/fd_table.go | 7 +- 13 files changed, 709 insertions(+), 80 deletions(-) create mode 100644 pkg/sentry/fsimpl/proc/task_fds.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index d50018b18..94ca3dbdd 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -554,3 +554,11 @@ func (s *StaticDirectory) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs fd.Init(rp.Mount(), vfsd, &s.OrderedChildren, &opts) return fd.VFSFileDescription(), nil } + +// AlwaysValid partially implements kernfs.inodeDynamicLookup. +type AlwaysValid struct{} + +// Valid implements kernfs.inodeDynamicLookup. +func (*AlwaysValid) Valid(context.Context) bool { + return true +} diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index a8ab2a2ba..18a34a590 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -319,7 +319,7 @@ type inodeMetadata interface { // CheckPermissions checks that creds may access this inode for the // requested access type, per the the rules of // fs/namei.c:generic_permission(). - CheckPermissions(ctx context.Context, creds *auth.Credentials, atx vfs.AccessTypes) error + CheckPermissions(ctx context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error // Mode returns the (struct stat)::st_mode value for this inode. This is // separated from Stat for performance. diff --git a/pkg/sentry/fsimpl/proc/BUILD b/pkg/sentry/fsimpl/proc/BUILD index bb609a305..8156984eb 100644 --- a/pkg/sentry/fsimpl/proc/BUILD +++ b/pkg/sentry/fsimpl/proc/BUILD @@ -8,6 +8,7 @@ go_library( "filesystem.go", "subtasks.go", "task.go", + "task_fds.go", "task_files.go", "task_net.go", "tasks.go", @@ -19,8 +20,10 @@ go_library( "//pkg/abi/linux", "//pkg/context", "//pkg/log", + "//pkg/refs", "//pkg/safemem", "//pkg/sentry/fs", + "//pkg/sentry/fsbridge", "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/inet", "//pkg/sentry/kernel", @@ -53,6 +56,7 @@ go_test( "//pkg/fspath", "//pkg/sentry/contexttest", "//pkg/sentry/fsimpl/testutil", + "//pkg/sentry/fsimpl/tmpfs", "//pkg/sentry/inet", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go index 611645f3f..a3a7c16a5 100644 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -34,6 +34,7 @@ type subtasksInode struct { kernfs.InodeDirectoryNoNewChildren kernfs.InodeAttrs kernfs.OrderedChildren + kernfs.AlwaysValid task *kernel.Task pidns *kernel.PIDNamespace @@ -61,11 +62,6 @@ func newSubtasks(task *kernel.Task, pidns *kernel.PIDNamespace, inoGen InoGenera return dentry } -// Valid implements kernfs.inodeDynamicLookup. -func (i *subtasksInode) Valid(ctx context.Context) bool { - return true -} - // Lookup implements kernfs.inodeDynamicLookup. func (i *subtasksInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { tid, err := strconv.ParseUint(name, 10, 32) diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index 493acbd1b..4891caab6 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -45,19 +45,19 @@ var _ kernfs.Inode = (*taskInode)(nil) func newTaskInode(inoGen InoGenerator, task *kernel.Task, pidns *kernel.PIDNamespace, isThreadGroup bool, cgroupControllers map[string]string) *kernfs.Dentry { contents := map[string]*kernfs.Dentry{ - "auxv": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &auxvData{task: task}), - "cmdline": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &cmdlineData{task: task, arg: cmdlineDataArg}), - "comm": newComm(task, inoGen.NextIno(), 0444), - "environ": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &cmdlineData{task: task, arg: environDataArg}), - //"exe": newExe(t, msrc), - //"fd": newFdDir(t, msrc), - //"fdinfo": newFdInfoDir(t, msrc), - "gid_map": newTaskOwnedFile(task, inoGen.NextIno(), 0644, &idMapData{task: task, gids: true}), - "io": newTaskOwnedFile(task, inoGen.NextIno(), 0400, newIO(task, isThreadGroup)), - "maps": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &mapsData{task: task}), - //"mountinfo": seqfile.NewSeqFileInode(t, &mountInfoFile{t: t}, msrc), - //"mounts": seqfile.NewSeqFileInode(t, &mountsFile{t: t}, msrc), - "net": newTaskNetDir(task, inoGen), + "auxv": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &auxvData{task: task}), + "cmdline": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &cmdlineData{task: task, arg: cmdlineDataArg}), + "comm": newComm(task, inoGen.NextIno(), 0444), + "environ": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &cmdlineData{task: task, arg: environDataArg}), + "exe": newExeSymlink(task, inoGen.NextIno()), + "fd": newFDDirInode(task, inoGen), + "fdinfo": newFDInfoDirInode(task, inoGen), + "gid_map": newTaskOwnedFile(task, inoGen.NextIno(), 0644, &idMapData{task: task, gids: true}), + "io": newTaskOwnedFile(task, inoGen.NextIno(), 0400, newIO(task, isThreadGroup)), + "maps": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &mapsData{task: task}), + "mountinfo": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &mountInfoData{task: task}), + "mounts": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &mountsData{task: task}), + "net": newTaskNetDir(task, inoGen), "ns": newTaskOwnedDir(task, inoGen.NextIno(), 0511, map[string]*kernfs.Dentry{ "net": newNamespaceSymlink(task, inoGen.NextIno(), "net"), "pid": newNamespaceSymlink(task, inoGen.NextIno(), "pid"), diff --git a/pkg/sentry/fsimpl/proc/task_fds.go b/pkg/sentry/fsimpl/proc/task_fds.go new file mode 100644 index 000000000..76bfc5307 --- /dev/null +++ b/pkg/sentry/fsimpl/proc/task_fds.go @@ -0,0 +1,287 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package proc + +import ( + "bytes" + "fmt" + "sort" + "strconv" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/refs" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +type fdDir struct { + inoGen InoGenerator + task *kernel.Task + + // When produceSymlinks is set, dirents produces for the FDs are reported + // as symlink. Otherwise, they are reported as regular files. + produceSymlink bool +} + +func (i *fdDir) lookup(name string) (*vfs.FileDescription, kernel.FDFlags, error) { + fd, err := strconv.ParseUint(name, 10, 64) + if err != nil { + return nil, kernel.FDFlags{}, syserror.ENOENT + } + + var ( + file *vfs.FileDescription + flags kernel.FDFlags + ) + i.task.WithMuLocked(func(t *kernel.Task) { + if fdTable := t.FDTable(); fdTable != nil { + file, flags = fdTable.GetVFS2(int32(fd)) + } + }) + if file == nil { + return nil, kernel.FDFlags{}, syserror.ENOENT + } + return file, flags, nil +} + +// IterDirents implements kernfs.inodeDynamicLookup. +func (i *fdDir) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, absOffset, relOffset int64) (int64, error) { + var fds []int32 + i.task.WithMuLocked(func(t *kernel.Task) { + if fdTable := t.FDTable(); fdTable != nil { + fds = fdTable.GetFDs() + } + }) + + offset := absOffset + relOffset + typ := uint8(linux.DT_REG) + if i.produceSymlink { + typ = linux.DT_LNK + } + + // Find the appropriate starting point. + idx := sort.Search(len(fds), func(i int) bool { return fds[i] >= int32(relOffset) }) + if idx >= len(fds) { + return offset, nil + } + for _, fd := range fds[idx:] { + dirent := vfs.Dirent{ + Name: strconv.FormatUint(uint64(fd), 10), + Type: typ, + Ino: i.inoGen.NextIno(), + NextOff: offset + 1, + } + if err := cb.Handle(dirent); err != nil { + return offset, err + } + offset++ + } + return offset, nil +} + +// fdDirInode represents the inode for /proc/[pid]/fd directory. +// +// +stateify savable +type fdDirInode struct { + kernfs.InodeNotSymlink + kernfs.InodeDirectoryNoNewChildren + kernfs.InodeAttrs + kernfs.OrderedChildren + kernfs.AlwaysValid + fdDir +} + +var _ kernfs.Inode = (*fdDirInode)(nil) + +func newFDDirInode(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry { + inode := &fdDirInode{ + fdDir: fdDir{ + inoGen: inoGen, + task: task, + produceSymlink: true, + }, + } + inode.InodeAttrs.Init(task.Credentials(), inoGen.NextIno(), linux.ModeDirectory|0555) + + dentry := &kernfs.Dentry{} + dentry.Init(inode) + inode.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) + + return dentry +} + +// Lookup implements kernfs.inodeDynamicLookup. +func (i *fdDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { + file, _, err := i.lookup(name) + if err != nil { + return nil, err + } + taskDentry := newFDSymlink(i.task.Credentials(), file, i.inoGen.NextIno()) + return taskDentry.VFSDentry(), nil +} + +// Open implements kernfs.Inode. +func (i *fdDirInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd := &kernfs.GenericDirectoryFD{} + fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, &opts) + return fd.VFSFileDescription(), nil +} + +// CheckPermissions implements kernfs.Inode. +// +// This is to match Linux, which uses a special permission handler to guarantee +// that a process can still access /proc/self/fd after it has executed +// setuid. See fs/proc/fd.c:proc_fd_permission. +func (i *fdDirInode) CheckPermissions(ctx context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error { + err := i.InodeAttrs.CheckPermissions(ctx, creds, ats) + if err == nil { + // Access granted, no extra check needed. + return nil + } + if t := kernel.TaskFromContext(ctx); t != nil { + // Allow access if the task trying to access it is in the thread group + // corresponding to this directory. + if i.task.ThreadGroup() == t.ThreadGroup() { + // Access granted (overridden). + return nil + } + } + return err +} + +// fdSymlink is an symlink for the /proc/[pid]/fd/[fd] file. +// +// +stateify savable +type fdSymlink struct { + refs.AtomicRefCount + kernfs.InodeAttrs + kernfs.InodeSymlink + + file *vfs.FileDescription +} + +var _ kernfs.Inode = (*fdSymlink)(nil) + +func newFDSymlink(creds *auth.Credentials, file *vfs.FileDescription, ino uint64) *kernfs.Dentry { + file.IncRef() + inode := &fdSymlink{file: file} + inode.Init(creds, ino, linux.ModeSymlink|0777) + + d := &kernfs.Dentry{} + d.Init(inode) + return d +} + +func (s *fdSymlink) Readlink(ctx context.Context) (string, error) { + root := vfs.RootFromContext(ctx) + defer root.DecRef() + + vfsObj := s.file.VirtualDentry().Mount().Filesystem().VirtualFilesystem() + return vfsObj.PathnameWithDeleted(ctx, root, s.file.VirtualDentry()) +} + +func (s *fdSymlink) DecRef() { + s.AtomicRefCount.DecRefWithDestructor(func() { + s.Destroy() + }) +} + +func (s *fdSymlink) Destroy() { + s.file.DecRef() +} + +// fdInfoDirInode represents the inode for /proc/[pid]/fdinfo directory. +// +// +stateify savable +type fdInfoDirInode struct { + kernfs.InodeNotSymlink + kernfs.InodeDirectoryNoNewChildren + kernfs.InodeAttrs + kernfs.OrderedChildren + kernfs.AlwaysValid + fdDir +} + +var _ kernfs.Inode = (*fdInfoDirInode)(nil) + +func newFDInfoDirInode(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry { + inode := &fdInfoDirInode{ + fdDir: fdDir{ + inoGen: inoGen, + task: task, + }, + } + inode.InodeAttrs.Init(task.Credentials(), inoGen.NextIno(), linux.ModeDirectory|0555) + + dentry := &kernfs.Dentry{} + dentry.Init(inode) + inode.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) + + return dentry +} + +// Lookup implements kernfs.inodeDynamicLookup. +func (i *fdInfoDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { + file, flags, err := i.lookup(name) + if err != nil { + return nil, err + } + + data := &fdInfoData{file: file, flags: flags} + dentry := newTaskOwnedFile(i.task, i.inoGen.NextIno(), 0444, data) + return dentry.VFSDentry(), nil +} + +// Open implements kernfs.Inode. +func (i *fdInfoDirInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd := &kernfs.GenericDirectoryFD{} + fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, &opts) + return fd.VFSFileDescription(), nil +} + +// fdInfoData implements vfs.DynamicBytesSource for /proc/[pid]/fdinfo/[fd]. +// +// +stateify savable +type fdInfoData struct { + kernfs.DynamicBytesFile + refs.AtomicRefCount + + file *vfs.FileDescription + flags kernel.FDFlags +} + +var _ dynamicInode = (*fdInfoData)(nil) + +func (d *fdInfoData) DecRef() { + d.AtomicRefCount.DecRefWithDestructor(d.destroy) +} + +func (d *fdInfoData) destroy() { + d.file.DecRef() +} + +// Generate implements vfs.DynamicBytesSource.Generate. +func (d *fdInfoData) Generate(ctx context.Context, buf *bytes.Buffer) error { + // TODO(b/121266871): Include pos, locks, and other data. For now we only + // have flags. + // See https://www.kernel.org/doc/Documentation/filesystems/proc.txt + flags := uint(d.file.StatusFlags()) | d.flags.ToLinuxFileFlags() + fmt.Fprintf(buf, "flags:\t0%o\n", flags) + return nil +} diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index 4d3332771..8c743df8d 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -18,10 +18,14 @@ import ( "bytes" "fmt" "io" + "sort" + "strings" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/safemem" + "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/fsbridge" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" @@ -496,7 +500,7 @@ func (s *statusData) Generate(ctx context.Context, buf *bytes.Buffer) error { return nil } -// ioUsage is the /proc//io and /proc//task//io data provider. +// ioUsage is the /proc/[pid]/io and /proc/[pid]/task/[tid]/io data provider. type ioUsage interface { // IOUsage returns the io usage data. IOUsage() *usage.IO @@ -570,3 +574,248 @@ func (o *oomScoreAdj) Write(ctx context.Context, src usermem.IOSequence, offset return n, nil } + +// exeSymlink is an symlink for the /proc/[pid]/exe file. +// +// +stateify savable +type exeSymlink struct { + kernfs.InodeAttrs + kernfs.InodeNoopRefCount + kernfs.InodeSymlink + + task *kernel.Task +} + +var _ kernfs.Inode = (*exeSymlink)(nil) + +func newExeSymlink(task *kernel.Task, ino uint64) *kernfs.Dentry { + inode := &exeSymlink{task: task} + inode.Init(task.Credentials(), ino, linux.ModeSymlink|0777) + + d := &kernfs.Dentry{} + d.Init(inode) + return d +} + +// Readlink implements kernfs.Inode. +func (s *exeSymlink) Readlink(ctx context.Context) (string, error) { + if !kernel.ContextCanTrace(ctx, s.task, false) { + return "", syserror.EACCES + } + + // Pull out the executable for /proc/[pid]/exe. + exec, err := s.executable() + if err != nil { + return "", err + } + defer exec.DecRef() + + return exec.PathnameWithDeleted(ctx), nil +} + +func (s *exeSymlink) executable() (file fsbridge.File, err error) { + s.task.WithMuLocked(func(t *kernel.Task) { + mm := t.MemoryManager() + if mm == nil { + // TODO(b/34851096): Check shouldn't allow Readlink once the + // Task is zombied. + err = syserror.EACCES + return + } + + // The MemoryManager may be destroyed, in which case + // MemoryManager.destroy will simply set the executable to nil + // (with locks held). + file = mm.Executable() + if file == nil { + err = syserror.ENOENT + } + }) + return +} + +// forEachMountSource runs f for the process root mount and each mount that is +// a descendant of the root. +func forEachMount(t *kernel.Task, fn func(string, *fs.Mount)) { + var fsctx *kernel.FSContext + t.WithMuLocked(func(t *kernel.Task) { + fsctx = t.FSContext() + }) + if fsctx == nil { + // The task has been destroyed. Nothing to show here. + return + } + + // All mount points must be relative to the rootDir, and mounts outside + // will be excluded. + rootDir := fsctx.RootDirectory() + if rootDir == nil { + // The task has been destroyed. Nothing to show here. + return + } + defer rootDir.DecRef() + + mnt := t.MountNamespace().FindMount(rootDir) + if mnt == nil { + // Has it just been unmounted? + return + } + ms := t.MountNamespace().AllMountsUnder(mnt) + sort.Slice(ms, func(i, j int) bool { + return ms[i].ID < ms[j].ID + }) + for _, m := range ms { + mroot := m.Root() + if mroot == nil { + continue // No longer valid. + } + mountPath, desc := mroot.FullName(rootDir) + mroot.DecRef() + if !desc { + // MountSources that are not descendants of the chroot jail are ignored. + continue + } + fn(mountPath, m) + } +} + +// mountInfoData is used to implement /proc/[pid]/mountinfo. +// +// +stateify savable +type mountInfoData struct { + kernfs.DynamicBytesFile + + task *kernel.Task +} + +var _ dynamicInode = (*mountInfoData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (i *mountInfoData) Generate(ctx context.Context, buf *bytes.Buffer) error { + forEachMount(i.task, func(mountPath string, m *fs.Mount) { + mroot := m.Root() + if mroot == nil { + return // No longer valid. + } + defer mroot.DecRef() + + // Format: + // 36 35 98:0 /mnt1 /mnt2 rw,noatime master:1 - ext3 /dev/root rw,errors=continue + // (1)(2)(3) (4) (5) (6) (7) (8) (9) (10) (11) + + // (1) MountSource ID. + fmt.Fprintf(buf, "%d ", m.ID) + + // (2) Parent ID (or this ID if there is no parent). + pID := m.ID + if !m.IsRoot() && !m.IsUndo() { + pID = m.ParentID + } + fmt.Fprintf(buf, "%d ", pID) + + // (3) Major:Minor device ID. We don't have a superblock, so we + // just use the root inode device number. + sa := mroot.Inode.StableAttr + fmt.Fprintf(buf, "%d:%d ", sa.DeviceFileMajor, sa.DeviceFileMinor) + + // (4) Root: the pathname of the directory in the filesystem + // which forms the root of this mount. + // + // NOTE(b/78135857): This will always be "/" until we implement + // bind mounts. + fmt.Fprintf(buf, "/ ") + + // (5) Mount point (relative to process root). + fmt.Fprintf(buf, "%s ", mountPath) + + // (6) Mount options. + flags := mroot.Inode.MountSource.Flags + opts := "rw" + if flags.ReadOnly { + opts = "ro" + } + if flags.NoAtime { + opts += ",noatime" + } + if flags.NoExec { + opts += ",noexec" + } + fmt.Fprintf(buf, "%s ", opts) + + // (7) Optional fields: zero or more fields of the form "tag[:value]". + // (8) Separator: the end of the optional fields is marked by a single hyphen. + fmt.Fprintf(buf, "- ") + + // (9) Filesystem type. + fmt.Fprintf(buf, "%s ", mroot.Inode.MountSource.FilesystemType) + + // (10) Mount source: filesystem-specific information or "none". + fmt.Fprintf(buf, "none ") + + // (11) Superblock options, and final newline. + fmt.Fprintf(buf, "%s\n", superBlockOpts(mountPath, mroot.Inode.MountSource)) + }) + return nil +} + +func superBlockOpts(mountPath string, msrc *fs.MountSource) string { + // gVisor doesn't (yet) have a concept of super block options, so we + // use the ro/rw bit from the mount flag. + opts := "rw" + if msrc.Flags.ReadOnly { + opts = "ro" + } + + // NOTE(b/147673608): If the mount is a cgroup, we also need to include + // the cgroup name in the options. For now we just read that from the + // path. + // TODO(gvisor.dev/issues/190): Once gVisor has full cgroup support, we + // should get this value from the cgroup itself, and not rely on the + // path. + if msrc.FilesystemType == "cgroup" { + splitPath := strings.Split(mountPath, "/") + cgroupType := splitPath[len(splitPath)-1] + opts += "," + cgroupType + } + return opts +} + +// mountsData is used to implement /proc/[pid]/mounts. +// +// +stateify savable +type mountsData struct { + kernfs.DynamicBytesFile + + task *kernel.Task +} + +var _ dynamicInode = (*mountInfoData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (i *mountsData) Generate(ctx context.Context, buf *bytes.Buffer) error { + forEachMount(i.task, func(mountPath string, m *fs.Mount) { + // Format: + // + // + // We use the filesystem name as the first field, since there + // is no real block device we can point to, and we also should + // not expose anything about the remote filesystem. + // + // Only ro/rw option is supported for now. + // + // The "needs dump"and fsck flags are always 0, which is allowed. + root := m.Root() + if root == nil { + return // No longer valid. + } + defer root.DecRef() + + flags := root.Inode.MountSource.Flags + opts := "rw" + if flags.ReadOnly { + opts = "ro" + } + fmt.Fprintf(buf, "%s %s %s %s %d %d\n", "none", mountPath, root.Inode.MountSource.FilesystemType, opts, 0, 0) + }) + return nil +} diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index d203cebd4..07115664c 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -46,6 +46,7 @@ type tasksInode struct { kernfs.InodeDirectoryNoNewChildren kernfs.InodeAttrs kernfs.OrderedChildren + kernfs.AlwaysValid inoGen InoGenerator pidns *kernel.PIDNamespace @@ -66,23 +67,23 @@ var _ kernfs.Inode = (*tasksInode)(nil) func newTasksInode(inoGen InoGenerator, k *kernel.Kernel, pidns *kernel.PIDNamespace, cgroupControllers map[string]string) (*tasksInode, *kernfs.Dentry) { root := auth.NewRootCredentials(pidns.UserNamespace()) contents := map[string]*kernfs.Dentry{ - "cpuinfo": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(cpuInfoData(k))), - //"filesystems": newDentry(root, inoGen.NextIno(), 0444, &filesystemsData{}), - "loadavg": newDentry(root, inoGen.NextIno(), 0444, &loadavgData{}), - "sys": newSysDir(root, inoGen, k), - "meminfo": newDentry(root, inoGen.NextIno(), 0444, &meminfoData{}), - "mounts": kernfs.NewStaticSymlink(root, inoGen.NextIno(), "self/mounts"), - "net": kernfs.NewStaticSymlink(root, inoGen.NextIno(), "self/net"), - "stat": newDentry(root, inoGen.NextIno(), 0444, &statData{k: k}), - "uptime": newDentry(root, inoGen.NextIno(), 0444, &uptimeData{}), - "version": newDentry(root, inoGen.NextIno(), 0444, &versionData{k: k}), + "cpuinfo": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(cpuInfoData(k))), + "filesystems": newDentry(root, inoGen.NextIno(), 0444, &filesystemsData{}), + "loadavg": newDentry(root, inoGen.NextIno(), 0444, &loadavgData{}), + "sys": newSysDir(root, inoGen, k), + "meminfo": newDentry(root, inoGen.NextIno(), 0444, &meminfoData{}), + "mounts": kernfs.NewStaticSymlink(root, inoGen.NextIno(), "self/mounts"), + "net": kernfs.NewStaticSymlink(root, inoGen.NextIno(), "self/net"), + "stat": newDentry(root, inoGen.NextIno(), 0444, &statData{}), + "uptime": newDentry(root, inoGen.NextIno(), 0444, &uptimeData{}), + "version": newDentry(root, inoGen.NextIno(), 0444, &versionData{}), } inode := &tasksInode{ pidns: pidns, inoGen: inoGen, - selfSymlink: newSelfSymlink(root, inoGen.NextIno(), 0444, pidns).VFSDentry(), - threadSelfSymlink: newThreadSelfSymlink(root, inoGen.NextIno(), 0444, pidns).VFSDentry(), + selfSymlink: newSelfSymlink(root, inoGen.NextIno(), pidns).VFSDentry(), + threadSelfSymlink: newThreadSelfSymlink(root, inoGen.NextIno(), pidns).VFSDentry(), cgroupControllers: cgroupControllers, } inode.InodeAttrs.Init(root, inoGen.NextIno(), linux.ModeDirectory|0555) @@ -121,11 +122,6 @@ func (i *tasksInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, erro return taskDentry.VFSDentry(), nil } -// Valid implements kernfs.inodeDynamicLookup. -func (i *tasksInode) Valid(ctx context.Context) bool { - return true -} - // IterDirents implements kernfs.inodeDynamicLookup. func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, offset, _ int64) (int64, error) { // fs/proc/internal.h: #define FIRST_PROCESS_ENTRY 256 diff --git a/pkg/sentry/fsimpl/proc/tasks_files.go b/pkg/sentry/fsimpl/proc/tasks_files.go index 434998910..b99badba8 100644 --- a/pkg/sentry/fsimpl/proc/tasks_files.go +++ b/pkg/sentry/fsimpl/proc/tasks_files.go @@ -40,9 +40,9 @@ type selfSymlink struct { var _ kernfs.Inode = (*selfSymlink)(nil) -func newSelfSymlink(creds *auth.Credentials, ino uint64, perm linux.FileMode, pidns *kernel.PIDNamespace) *kernfs.Dentry { +func newSelfSymlink(creds *auth.Credentials, ino uint64, pidns *kernel.PIDNamespace) *kernfs.Dentry { inode := &selfSymlink{pidns: pidns} - inode.Init(creds, ino, linux.ModeSymlink|perm) + inode.Init(creds, ino, linux.ModeSymlink|0777) d := &kernfs.Dentry{} d.Init(inode) @@ -72,9 +72,9 @@ type threadSelfSymlink struct { var _ kernfs.Inode = (*threadSelfSymlink)(nil) -func newThreadSelfSymlink(creds *auth.Credentials, ino uint64, perm linux.FileMode, pidns *kernel.PIDNamespace) *kernfs.Dentry { +func newThreadSelfSymlink(creds *auth.Credentials, ino uint64, pidns *kernel.PIDNamespace) *kernfs.Dentry { inode := &threadSelfSymlink{pidns: pidns} - inode.Init(creds, ino, linux.ModeSymlink|perm) + inode.Init(creds, ino, linux.ModeSymlink|0777) d := &kernfs.Dentry{} d.Init(inode) @@ -138,21 +138,19 @@ func (c cpuStats) String() string { // +stateify savable type statData struct { kernfs.DynamicBytesFile - - // k is the owning Kernel. - k *kernel.Kernel } var _ dynamicInode = (*statData)(nil) // Generate implements vfs.DynamicBytesSource.Generate. -func (s *statData) Generate(ctx context.Context, buf *bytes.Buffer) error { +func (*statData) Generate(ctx context.Context, buf *bytes.Buffer) error { // TODO(b/37226836): We currently export only zero CPU stats. We could // at least provide some aggregate stats. var cpu cpuStats fmt.Fprintf(buf, "cpu %s\n", cpu) - for c, max := uint(0), s.k.ApplicationCores(); c < max; c++ { + k := kernel.KernelFromContext(ctx) + for c, max := uint(0), k.ApplicationCores(); c < max; c++ { fmt.Fprintf(buf, "cpu%d %s\n", c, cpu) } @@ -176,7 +174,7 @@ func (s *statData) Generate(ctx context.Context, buf *bytes.Buffer) error { fmt.Fprintf(buf, "ctxt 0\n") // CLOCK_REALTIME timestamp from boot, in seconds. - fmt.Fprintf(buf, "btime %d\n", s.k.Timekeeper().BootTime().Seconds()) + fmt.Fprintf(buf, "btime %d\n", k.Timekeeper().BootTime().Seconds()) // Total number of clones. // TODO(b/37226836): Count this. @@ -209,7 +207,7 @@ type loadavgData struct { var _ dynamicInode = (*loadavgData)(nil) // Generate implements vfs.DynamicBytesSource.Generate. -func (d *loadavgData) Generate(ctx context.Context, buf *bytes.Buffer) error { +func (*loadavgData) Generate(ctx context.Context, buf *bytes.Buffer) error { // TODO(b/62345059): Include real data in fields. // Column 1-3: CPU and IO utilization of the last 1, 5, and 10 minute periods. // Column 4-5: currently running processes and the total number of processes. @@ -223,16 +221,14 @@ func (d *loadavgData) Generate(ctx context.Context, buf *bytes.Buffer) error { // +stateify savable type meminfoData struct { kernfs.DynamicBytesFile - - // k is the owning Kernel. - k *kernel.Kernel } var _ dynamicInode = (*meminfoData)(nil) // Generate implements vfs.DynamicBytesSource.Generate. -func (d *meminfoData) Generate(ctx context.Context, buf *bytes.Buffer) error { - mf := d.k.MemoryFile() +func (*meminfoData) Generate(ctx context.Context, buf *bytes.Buffer) error { + k := kernel.KernelFromContext(ctx) + mf := k.MemoryFile() mf.UpdateUsage() snapshot, totalUsage := usage.MemoryAccounting.Copy() totalSize := usage.TotalMemory(mf.TotalSize(), totalUsage) @@ -295,16 +291,14 @@ func (*uptimeData) Generate(ctx context.Context, buf *bytes.Buffer) error { // +stateify savable type versionData struct { kernfs.DynamicBytesFile - - // k is the owning Kernel. - k *kernel.Kernel } var _ dynamicInode = (*versionData)(nil) // Generate implements vfs.DynamicBytesSource.Generate. -func (v *versionData) Generate(ctx context.Context, buf *bytes.Buffer) error { - init := v.k.GlobalInit() +func (*versionData) Generate(ctx context.Context, buf *bytes.Buffer) error { + k := kernel.KernelFromContext(ctx) + init := k.GlobalInit() if init == nil { // Attempted to read before the init Task is created. This can // only occur during startup, which should never need to read @@ -335,3 +329,19 @@ func (v *versionData) Generate(ctx context.Context, buf *bytes.Buffer) error { fmt.Fprintf(buf, "%s version %s %s\n", ver.Sysname, ver.Release, ver.Version) return nil } + +// filesystemsData backs /proc/filesystems. +// +// +stateify savable +type filesystemsData struct { + kernfs.DynamicBytesFile +} + +var _ dynamicInode = (*filesystemsData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (d *filesystemsData) Generate(ctx context.Context, buf *bytes.Buffer) error { + k := kernel.KernelFromContext(ctx) + k.VFS().GenerateProcFilesystems(buf) + return nil +} diff --git a/pkg/sentry/fsimpl/proc/tasks_test.go b/pkg/sentry/fsimpl/proc/tasks_test.go index 1bb9430c0..d0f97c137 100644 --- a/pkg/sentry/fsimpl/proc/tasks_test.go +++ b/pkg/sentry/fsimpl/proc/tasks_test.go @@ -25,6 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/fsimpl/testutil" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" @@ -47,6 +48,7 @@ var ( var ( tasksStaticFiles = map[string]testutil.DirentType{ "cpuinfo": linux.DT_REG, + "filesystems": linux.DT_REG, "loadavg": linux.DT_REG, "meminfo": linux.DT_REG, "mounts": linux.DT_LNK, @@ -68,9 +70,14 @@ var ( "cmdline": linux.DT_REG, "comm": linux.DT_REG, "environ": linux.DT_REG, + "exe": linux.DT_LNK, + "fd": linux.DT_DIR, + "fdinfo": linux.DT_DIR, "gid_map": linux.DT_REG, "io": linux.DT_REG, "maps": linux.DT_REG, + "mountinfo": linux.DT_REG, + "mounts": linux.DT_REG, "net": linux.DT_DIR, "ns": linux.DT_DIR, "oom_score": linux.DT_REG, @@ -96,17 +103,37 @@ func setup(t *testing.T) *testutil.System { k.VFS().MustRegisterFilesystemType(Name, &FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, }) - fsOpts := vfs.GetFilesystemOptions{ - InternalData: &InternalData{ - Cgroups: map[string]string{ - "cpuset": "/foo/cpuset", - "memory": "/foo/memory", + + mntns, err := k.VFS().NewMountNamespace(ctx, creds, "", tmpfs.Name, &vfs.GetFilesystemOptions{}) + if err != nil { + t.Fatalf("NewMountNamespace(): %v", err) + } + pop := &vfs.PathOperation{ + Root: mntns.Root(), + Start: mntns.Root(), + Path: fspath.Parse("/proc"), + } + if err := k.VFS().MkdirAt(ctx, creds, pop, &vfs.MkdirOptions{Mode: 0777}); err != nil { + t.Fatalf("MkDir(/proc): %v", err) + } + + pop = &vfs.PathOperation{ + Root: mntns.Root(), + Start: mntns.Root(), + Path: fspath.Parse("/proc"), + } + mntOpts := &vfs.MountOptions{ + GetFilesystemOptions: vfs.GetFilesystemOptions{ + InternalData: &InternalData{ + Cgroups: map[string]string{ + "cpuset": "/foo/cpuset", + "memory": "/foo/memory", + }, }, }, } - mntns, err := k.VFS().NewMountNamespace(ctx, creds, "", Name, &fsOpts) - if err != nil { - t.Fatalf("NewMountNamespace(): %v", err) + if err := k.VFS().MountAt(ctx, creds, "", pop, Name, mntOpts); err != nil { + t.Fatalf("MountAt(/proc): %v", err) } return testutil.NewSystem(ctx, t, k.VFS(), mntns) } @@ -115,7 +142,7 @@ func TestTasksEmpty(t *testing.T) { s := setup(t) defer s.Destroy() - collector := s.ListDirents(s.PathOpAtRoot("/")) + collector := s.ListDirents(s.PathOpAtRoot("/proc")) s.AssertAllDirentTypes(collector, tasksStaticFiles) s.AssertDirentOffsets(collector, tasksStaticFilesNextOffs) } @@ -141,7 +168,7 @@ func TestTasks(t *testing.T) { expectedDirents[fmt.Sprintf("%d", i+1)] = linux.DT_DIR } - collector := s.ListDirents(s.PathOpAtRoot("/")) + collector := s.ListDirents(s.PathOpAtRoot("/proc")) s.AssertAllDirentTypes(collector, expectedDirents) s.AssertDirentOffsets(collector, tasksStaticFilesNextOffs) @@ -181,7 +208,7 @@ func TestTasks(t *testing.T) { } // Test lookup. - for _, path := range []string{"/1", "/2"} { + for _, path := range []string{"/proc/1", "/proc/2"} { fd, err := s.VFS.OpenAt( s.Ctx, s.Creds, @@ -191,6 +218,7 @@ func TestTasks(t *testing.T) { if err != nil { t.Fatalf("vfsfs.OpenAt(%q) failed: %v", path, err) } + defer fd.DecRef() buf := make([]byte, 1) bufIOSeq := usermem.BytesIOSequence(buf) if _, err := fd.Read(s.Ctx, bufIOSeq, vfs.ReadOptions{}); err != syserror.EISDIR { @@ -201,10 +229,10 @@ func TestTasks(t *testing.T) { if _, err := s.VFS.OpenAt( s.Ctx, s.Creds, - s.PathOpAtRoot("/9999"), + s.PathOpAtRoot("/proc/9999"), &vfs.OpenOptions{}, ); err != syserror.ENOENT { - t.Fatalf("wrong error from vfsfs.OpenAt(/9999): %v", err) + t.Fatalf("wrong error from vfsfs.OpenAt(/proc/9999): %v", err) } } @@ -302,12 +330,13 @@ func TestTasksOffset(t *testing.T) { fd, err := s.VFS.OpenAt( s.Ctx, s.Creds, - s.PathOpAtRoot("/"), + s.PathOpAtRoot("/proc"), &vfs.OpenOptions{}, ) if err != nil { t.Fatalf("vfsfs.OpenAt(/) failed: %v", err) } + defer fd.DecRef() if _, err := fd.Seek(s.Ctx, tc.offset, linux.SEEK_SET); err != nil { t.Fatalf("Seek(%d, SEEK_SET): %v", tc.offset, err) } @@ -344,7 +373,7 @@ func TestTask(t *testing.T) { t.Fatalf("CreateTask(): %v", err) } - collector := s.ListDirents(s.PathOpAtRoot("/1")) + collector := s.ListDirents(s.PathOpAtRoot("/proc/1")) s.AssertAllDirentTypes(collector, taskStaticFiles) } @@ -362,14 +391,14 @@ func TestProcSelf(t *testing.T) { collector := s.WithTemporaryContext(task).ListDirents(&vfs.PathOperation{ Root: s.Root, Start: s.Root, - Path: fspath.Parse("/self/"), + Path: fspath.Parse("/proc/self/"), FollowFinalSymlink: true, }) s.AssertAllDirentTypes(collector, taskStaticFiles) } func iterateDir(ctx context.Context, t *testing.T, s *testutil.System, fd *vfs.FileDescription) { - t.Logf("Iterating: /proc%s", fd.MappedName(ctx)) + t.Logf("Iterating: %s", fd.MappedName(ctx)) var collector testutil.DirentCollector if err := fd.IterDirents(ctx, &collector); err != nil { @@ -412,6 +441,7 @@ func iterateDir(ctx context.Context, t *testing.T, s *testutil.System, fd *vfs.F t.Errorf("vfsfs.OpenAt(%v) failed: %v", childPath, err) continue } + defer child.DecRef() stat, err := child.Stat(ctx, vfs.StatOptions{}) if err != nil { t.Errorf("Stat(%v) failed: %v", childPath, err) @@ -432,6 +462,22 @@ func TestTree(t *testing.T) { defer s.Destroy() k := kernel.KernelFromContext(s.Ctx) + + pop := &vfs.PathOperation{ + Root: s.Root, + Start: s.Root, + Path: fspath.Parse("test-file"), + } + opts := &vfs.OpenOptions{ + Flags: linux.O_RDONLY | linux.O_CREAT, + Mode: 0777, + } + file, err := s.VFS.OpenAt(s.Ctx, s.Creds, pop, opts) + if err != nil { + t.Fatalf("failed to create test file: %v", err) + } + defer file.DecRef() + var tasks []*kernel.Task for i := 0; i < 5; i++ { tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) @@ -439,6 +485,8 @@ func TestTree(t *testing.T) { if err != nil { t.Fatalf("CreateTask(): %v", err) } + // Add file to populate /proc/[pid]/fd and fdinfo directories. + task.FDTable().NewFDVFS2(task, 0, file, kernel.FDFlags{}) tasks = append(tasks, task) } @@ -446,11 +494,12 @@ func TestTree(t *testing.T) { fd, err := s.VFS.OpenAt( ctx, auth.CredentialsFromContext(s.Ctx), - &vfs.PathOperation{Root: s.Root, Start: s.Root, Path: fspath.Parse("/")}, + &vfs.PathOperation{Root: s.Root, Start: s.Root, Path: fspath.Parse("/proc")}, &vfs.OpenOptions{}, ) if err != nil { - t.Fatalf("vfsfs.OpenAt(/) failed: %v", err) + t.Fatalf("vfsfs.OpenAt(/proc) failed: %v", err) } iterateDir(ctx, t, s, fd) + fd.DecRef() } diff --git a/pkg/sentry/fsimpl/testutil/BUILD b/pkg/sentry/fsimpl/testutil/BUILD index e4f36f4ae..0e4053a46 100644 --- a/pkg/sentry/fsimpl/testutil/BUILD +++ b/pkg/sentry/fsimpl/testutil/BUILD @@ -16,12 +16,14 @@ go_library( "//pkg/cpuid", "//pkg/fspath", "//pkg/memutil", + "//pkg/sentry/fsbridge", "//pkg/sentry/fsimpl/tmpfs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/sched", "//pkg/sentry/limits", "//pkg/sentry/loader", + "//pkg/sentry/mm", "//pkg/sentry/pgalloc", "//pkg/sentry/platform", "//pkg/sentry/platform/kvm", diff --git a/pkg/sentry/fsimpl/testutil/kernel.go b/pkg/sentry/fsimpl/testutil/kernel.go index 488478e29..c16a36cdb 100644 --- a/pkg/sentry/fsimpl/testutil/kernel.go +++ b/pkg/sentry/fsimpl/testutil/kernel.go @@ -23,13 +23,16 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/cpuid" + "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/memutil" + "gvisor.dev/gvisor/pkg/sentry/fsbridge" "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/sched" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/loader" + "gvisor.dev/gvisor/pkg/sentry/mm" "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/time" @@ -123,10 +126,17 @@ func Boot() (*kernel.Kernel, error) { // CreateTask creates a new bare bones task for tests. func CreateTask(ctx context.Context, name string, tc *kernel.ThreadGroup, mntns *vfs.MountNamespace, root, cwd vfs.VirtualDentry) (*kernel.Task, error) { k := kernel.KernelFromContext(ctx) + exe, err := newFakeExecutable(ctx, k.VFS(), auth.CredentialsFromContext(ctx), root) + if err != nil { + return nil, err + } + m := mm.NewMemoryManager(k, k, k.SleepForAddressSpaceActivation) + m.SetExecutable(fsbridge.NewVFSFile(exe)) + config := &kernel.TaskConfig{ Kernel: k, ThreadGroup: tc, - TaskContext: &kernel.TaskContext{Name: name}, + TaskContext: &kernel.TaskContext{Name: name, MemoryManager: m}, Credentials: auth.CredentialsFromContext(ctx), NetworkNamespace: k.RootNetworkNamespace(), AllowedCPUMask: sched.NewFullCPUSet(k.ApplicationCores()), @@ -135,10 +145,25 @@ func CreateTask(ctx context.Context, name string, tc *kernel.ThreadGroup, mntns AbstractSocketNamespace: kernel.NewAbstractSocketNamespace(), MountNamespaceVFS2: mntns, FSContext: kernel.NewFSContextVFS2(root, cwd, 0022), + FDTable: k.NewFDTable(), } return k.TaskSet().NewTask(config) } +func newFakeExecutable(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, root vfs.VirtualDentry) (*vfs.FileDescription, error) { + const name = "executable" + pop := &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(name), + } + opts := &vfs.OpenOptions{ + Flags: linux.O_RDONLY | linux.O_CREAT, + Mode: 0777, + } + return vfsObj.OpenAt(ctx, creds, pop, opts) +} + func createMemoryFile() (*pgalloc.MemoryFile, error) { const memfileName = "test-memory" memfd, err := memutil.CreateMemFD(memfileName, 0) diff --git a/pkg/sentry/kernel/fd_table.go b/pkg/sentry/kernel/fd_table.go index 00f914564..7de2e509e 100644 --- a/pkg/sentry/kernel/fd_table.go +++ b/pkg/sentry/kernel/fd_table.go @@ -191,7 +191,7 @@ func (f *FDTable) Size() int { return int(size) } -// forEach iterates over all non-nil files. +// forEach iterates over all non-nil files in sorted order. // // It is the caller's responsibility to acquire an appropriate lock. func (f *FDTable) forEach(fn func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags)) { @@ -458,7 +458,10 @@ func (f *FDTable) GetVFS2(fd int32) (*vfs.FileDescription, FDFlags) { } } -// GetFDs returns a list of valid fds. +// GetFDs returns a sorted list of valid fds. +// +// Precondition: The caller must be running on the task goroutine, or Task.mu +// must be locked. func (f *FDTable) GetFDs() []int32 { fds := make([]int32, 0, int(atomic.LoadInt32(&f.used))) f.forEach(func(fd int32, _ *fs.File, _ *vfs.FileDescription, _ FDFlags) { -- cgit v1.2.3 From 5e413cad10d2358a21dd08216953faee70e62a0b Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Sat, 14 Mar 2020 07:13:15 -0700 Subject: Plumb VFS2 imported fds into virtual filesystem. - When setting up the virtual filesystem, mount a host.filesystem to contain all files that need to be imported. - Make read/preadv syscalls to the host in cases where preadv2 may not be supported yet (likewise for writing). - Make save/restore functions in kernel/kernel.go return early if vfs2 is enabled. PiperOrigin-RevId: 300922353 --- pkg/abi/linux/file.go | 3 + pkg/sentry/fs/host/control.go | 2 + pkg/sentry/fsimpl/host/BUILD | 2 + pkg/sentry/fsimpl/host/default_file.go | 45 +++++++----- pkg/sentry/fsimpl/host/host.go | 124 ++++++++++++++++++++++++++++++--- pkg/sentry/fsimpl/host/util.go | 28 ++------ pkg/sentry/kernel/kernel.go | 40 +++++++---- pkg/sentry/syscalls/linux/sys_stat.go | 5 +- pkg/sentry/syscalls/linux/vfs2/stat.go | 6 +- runsc/boot/filter/config.go | 1 + test/syscalls/linux/stat.cc | 60 ++++++++++++++-- 11 files changed, 246 insertions(+), 70 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/file.go b/pkg/abi/linux/file.go index e229ac21c..dbe58acbe 100644 --- a/pkg/abi/linux/file.go +++ b/pkg/abi/linux/file.go @@ -266,6 +266,9 @@ type Statx struct { DevMinor uint32 } +// SizeOfStatx is the size of a Statx struct. +var SizeOfStatx = binary.Size(Statx{}) + // FileMode represents a mode_t. type FileMode uint16 diff --git a/pkg/sentry/fs/host/control.go b/pkg/sentry/fs/host/control.go index 1658979fc..cd84e1337 100644 --- a/pkg/sentry/fs/host/control.go +++ b/pkg/sentry/fs/host/control.go @@ -32,6 +32,8 @@ func newSCMRights(fds []int) control.SCMRights { } // Files implements control.SCMRights.Files. +// +// TODO(gvisor.dev/issue/2017): Port to VFS2. func (c *scmRights) Files(ctx context.Context, max int) (control.RightsFiles, bool) { n := max var trunc bool diff --git a/pkg/sentry/fsimpl/host/BUILD b/pkg/sentry/fsimpl/host/BUILD index 731f192b3..5d67f88e3 100644 --- a/pkg/sentry/fsimpl/host/BUILD +++ b/pkg/sentry/fsimpl/host/BUILD @@ -9,9 +9,11 @@ go_library( "host.go", "util.go", ], + visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", "//pkg/context", + "//pkg/fd", "//pkg/log", "//pkg/refs", "//pkg/safemem", diff --git a/pkg/sentry/fsimpl/host/default_file.go b/pkg/sentry/fsimpl/host/default_file.go index 172cdb161..98682ba5e 100644 --- a/pkg/sentry/fsimpl/host/default_file.go +++ b/pkg/sentry/fsimpl/host/default_file.go @@ -21,6 +21,7 @@ import ( "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fd" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/vfs" @@ -64,9 +65,7 @@ func (f *defaultFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts v panic("files that can return EWOULDBLOCK (sockets, pipes, etc.) cannot be memory mapped") } - f.mu.Lock() n, err := readFromHostFD(ctx, f.inode.hostFD, dst, -1, int(opts.Flags)) - f.mu.Unlock() if isBlockError(err) { // If we got any data at all, return it as a "completed" partial read // rather than retrying until complete. @@ -86,16 +85,22 @@ func (f *defaultFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts v return n, err } -func readFromHostFD(ctx context.Context, fd int, dst usermem.IOSequence, offset int64, flags int) (int64, error) { - if flags&^(linux.RWF_VALID) != 0 { +func readFromHostFD(ctx context.Context, hostFD int, dst usermem.IOSequence, offset int64, flags int) (int64, error) { + // TODO(gvisor.dev/issue/1672): Support select preadv2 flags. + if flags != 0 { return 0, syserror.EOPNOTSUPP } - reader := safemem.FromVecReaderFunc{ - func(srcs [][]byte) (int64, error) { - n, err := unix.Preadv2(fd, srcs, offset, flags) - return int64(n), err - }, + var reader safemem.Reader + if offset == -1 { + reader = safemem.FromIOReader{fd.NewReadWriter(hostFD)} + } else { + reader = safemem.FromVecReaderFunc{ + func(srcs [][]byte) (int64, error) { + n, err := unix.Preadv(hostFD, srcs, offset) + return int64(n), err + }, + } } n, err := dst.CopyOutFrom(ctx, reader) return int64(n), err @@ -120,9 +125,7 @@ func (f *defaultFileFD) Write(ctx context.Context, src usermem.IOSequence, opts panic("files that can return EWOULDBLOCK (sockets, pipes, etc.) cannot be memory mapped") } - f.mu.Lock() n, err := writeToHostFD(ctx, f.inode.hostFD, src, -1, int(opts.Flags)) - f.mu.Unlock() if isBlockError(err) { err = syserror.ErrWouldBlock } @@ -137,16 +140,22 @@ func (f *defaultFileFD) Write(ctx context.Context, src usermem.IOSequence, opts return n, err } -func writeToHostFD(ctx context.Context, fd int, src usermem.IOSequence, offset int64, flags int) (int64, error) { - if flags&^(linux.RWF_VALID) != 0 { +func writeToHostFD(ctx context.Context, hostFD int, src usermem.IOSequence, offset int64, flags int) (int64, error) { + // TODO(gvisor.dev/issue/1672): Support select pwritev2 flags. + if flags != 0 { return 0, syserror.EOPNOTSUPP } - writer := safemem.FromVecWriterFunc{ - func(srcs [][]byte) (int64, error) { - n, err := unix.Pwritev2(fd, srcs, offset, flags) - return int64(n), err - }, + var writer safemem.Writer + if offset == -1 { + writer = safemem.FromIOWriter{fd.NewReadWriter(hostFD)} + } else { + writer = safemem.FromVecWriterFunc{ + func(srcs [][]byte) (int64, error) { + n, err := unix.Pwritev(hostFD, srcs, offset) + return int64(n), err + }, + } } n, err := src.CopyInTo(ctx, writer) return int64(n), err diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index c205e6a0b..0be812d13 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -38,10 +38,19 @@ type filesystem struct { kernfs.Filesystem } +// NewMount returns a new disconnected mount in vfsObj that may be passed to ImportFD. +func NewMount(vfsObj *vfs.VirtualFilesystem) (*vfs.Mount, error) { + fs := &filesystem{} + fs.Init(vfsObj) + vfsfs := fs.VFSFilesystem() + // NewDisconnectedMount will take an additional reference on vfsfs. + defer vfsfs.DecRef() + return vfsObj.NewDisconnectedMount(vfsfs, nil, &vfs.MountOptions{}) +} + // ImportFD sets up and returns a vfs.FileDescription from a donated fd. func ImportFD(mnt *vfs.Mount, hostFD int, ownerUID auth.KUID, ownerGID auth.KGID, isTTY bool) (*vfs.FileDescription, error) { - // Must be importing to a mount of host.filesystem. - fs, ok := mnt.Filesystem().Impl().(*filesystem) + fs, ok := mnt.Filesystem().Impl().(*kernfs.Filesystem) if !ok { return nil, fmt.Errorf("can't import host FDs into filesystems of type %T", mnt.Filesystem().Impl()) } @@ -54,8 +63,7 @@ func ImportFD(mnt *vfs.Mount, hostFD int, ownerUID auth.KUID, ownerGID auth.KGID fileMode := linux.FileMode(s.Mode) fileType := fileMode.FileType() - // Pipes, character devices, and sockets can return EWOULDBLOCK for - // operations that would block. + // Pipes, character devices, and sockets. isStream := fileType == syscall.S_IFIFO || fileType == syscall.S_IFCHR || fileType == syscall.S_IFSOCK i := &inode{ @@ -143,11 +151,109 @@ func (i *inode) Mode() linux.FileMode { // Stat implements kernfs.Inode. func (i *inode) Stat(_ *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { + if opts.Mask&linux.STATX__RESERVED != 0 { + return linux.Statx{}, syserror.EINVAL + } + if opts.Sync&linux.AT_STATX_SYNC_TYPE == linux.AT_STATX_SYNC_TYPE { + return linux.Statx{}, syserror.EINVAL + } + + // Limit our host call only to known flags. + mask := opts.Mask & linux.STATX_ALL var s unix.Statx_t - if err := unix.Statx(i.hostFD, "", int(unix.AT_EMPTY_PATH|opts.Sync), int(opts.Mask), &s); err != nil { + err := unix.Statx(i.hostFD, "", int(unix.AT_EMPTY_PATH|opts.Sync), int(mask), &s) + // Fallback to fstat(2), if statx(2) is not supported on the host. + // + // TODO(b/151263641): Remove fallback. + if err == syserror.ENOSYS { + return i.fstat(opts) + } else if err != nil { + return linux.Statx{}, err + } + + ls := linux.Statx{Mask: mask} + // Unconditionally fill blksize, attributes, and device numbers, as indicated + // by /include/uapi/linux/stat.h. + // + // RdevMajor/RdevMinor are left as zero, so as not to expose host device + // numbers. + // + // TODO(gvisor.dev/issue/1672): Use kernfs-specific, internally defined + // device numbers. If we use the device number from the host, it may collide + // with another sentry-internal device number. We handle device/inode + // numbers without relying on the host to prevent collisions. + ls.Blksize = s.Blksize + ls.Attributes = s.Attributes + ls.AttributesMask = s.Attributes_mask + + if mask|linux.STATX_TYPE != 0 { + ls.Mode |= s.Mode & linux.S_IFMT + } + if mask|linux.STATX_MODE != 0 { + ls.Mode |= s.Mode &^ linux.S_IFMT + } + if mask|linux.STATX_NLINK != 0 { + ls.Nlink = s.Nlink + } + if mask|linux.STATX_ATIME != 0 { + ls.Atime = unixToLinuxStatxTimestamp(s.Atime) + } + if mask|linux.STATX_BTIME != 0 { + ls.Btime = unixToLinuxStatxTimestamp(s.Btime) + } + if mask|linux.STATX_CTIME != 0 { + ls.Ctime = unixToLinuxStatxTimestamp(s.Ctime) + } + if mask|linux.STATX_MTIME != 0 { + ls.Mtime = unixToLinuxStatxTimestamp(s.Mtime) + } + if mask|linux.STATX_SIZE != 0 { + ls.Size = s.Size + } + if mask|linux.STATX_BLOCKS != 0 { + ls.Blocks = s.Blocks + } + + // Use our own internal inode number and file owner. + if mask|linux.STATX_INO != 0 { + ls.Ino = i.ino + } + if mask|linux.STATX_UID != 0 { + ls.UID = uint32(i.uid) + } + if mask|linux.STATX_GID != 0 { + ls.GID = uint32(i.gid) + } + + return ls, nil +} + +// fstat is a best-effort fallback for inode.Stat() if the host does not +// support statx(2). +// +// We ignore the mask and sync flags in opts and simply supply +// STATX_BASIC_STATS, as fstat(2) itself does not allow the specification +// of a mask or sync flags. fstat(2) does not provide any metadata +// equivalent to Statx.Attributes, Statx.AttributesMask, or Statx.Btime, so +// those fields remain empty. +func (i *inode) fstat(opts vfs.StatOptions) (linux.Statx, error) { + var s unix.Stat_t + if err := unix.Fstat(i.hostFD, &s); err != nil { return linux.Statx{}, err } - ls := unixToLinuxStatx(s) + + // Note that rdev numbers are left as 0; do not expose host device numbers. + ls := linux.Statx{ + Mask: linux.STATX_BASIC_STATS, + Blksize: uint32(s.Blksize), + Nlink: uint32(s.Nlink), + Mode: uint16(s.Mode), + Size: uint64(s.Size), + Blocks: uint64(s.Blocks), + Atime: timespecToStatxTimestamp(s.Atim), + Ctime: timespecToStatxTimestamp(s.Ctim), + Mtime: timespecToStatxTimestamp(s.Mtim), + } // Use our own internal inode number and file owner. // @@ -159,9 +265,6 @@ func (i *inode) Stat(_ *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, erro ls.UID = uint32(i.uid) ls.GID = uint32(i.gid) - // Update file mode from the host. - i.mode = linux.FileMode(ls.Mode) - return ls, nil } @@ -217,7 +320,6 @@ func (i *inode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptio } func (i *inode) open(d *vfs.Dentry, mnt *vfs.Mount) (*vfs.FileDescription, error) { - fileType := i.mode.FileType() if fileType == syscall.S_IFSOCK { if i.isTTY { @@ -227,6 +329,8 @@ func (i *inode) open(d *vfs.Dentry, mnt *vfs.Mount) (*vfs.FileDescription, error return nil, errors.New("importing host sockets not supported") } + // TODO(gvisor.dev/issue/1672): Whitelist specific file types here, so that + // we don't allow importing arbitrary file types without proper support. if i.isTTY { // TODO(gvisor.dev/issue/1672): support importing host fd as TTY. return nil, errors.New("importing host fd as TTY not supported") diff --git a/pkg/sentry/fsimpl/host/util.go b/pkg/sentry/fsimpl/host/util.go index e1ccacb4d..d519feef5 100644 --- a/pkg/sentry/fsimpl/host/util.go +++ b/pkg/sentry/fsimpl/host/util.go @@ -35,34 +35,14 @@ func toTimespec(ts linux.StatxTimestamp, omit bool) unix.Timespec { } } -func unixToLinuxStatx(s unix.Statx_t) linux.Statx { - return linux.Statx{ - Mask: s.Mask, - Blksize: s.Blksize, - Attributes: s.Attributes, - Nlink: s.Nlink, - UID: s.Uid, - GID: s.Gid, - Mode: s.Mode, - Ino: s.Ino, - Size: s.Size, - Blocks: s.Blocks, - AttributesMask: s.Attributes_mask, - Atime: unixToLinuxStatxTimestamp(s.Atime), - Btime: unixToLinuxStatxTimestamp(s.Btime), - Ctime: unixToLinuxStatxTimestamp(s.Ctime), - Mtime: unixToLinuxStatxTimestamp(s.Mtime), - RdevMajor: s.Rdev_major, - RdevMinor: s.Rdev_minor, - DevMajor: s.Dev_major, - DevMinor: s.Dev_minor, - } -} - func unixToLinuxStatxTimestamp(ts unix.StatxTimestamp) linux.StatxTimestamp { return linux.StatxTimestamp{Sec: ts.Sec, Nsec: ts.Nsec} } +func timespecToStatxTimestamp(ts unix.Timespec) linux.StatxTimestamp { + return linux.StatxTimestamp{Sec: int64(ts.Sec), Nsec: uint32(ts.Nsec)} +} + // wouldBlock returns true for file types that can return EWOULDBLOCK // for blocking operations, e.g. pipes, character devices, and sockets. func wouldBlock(fileType uint32) bool { diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 1d627564f..6feda8fa1 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -467,6 +467,11 @@ func (k *Kernel) flushMountSourceRefs() error { // // Precondition: Must be called with the kernel paused. func (ts *TaskSet) forEachFDPaused(f func(*fs.File, *vfs.FileDescription) error) (err error) { + // TODO(gvisor.dev/issue/1663): Add save support for VFS2. + if VFS2Enabled { + return nil + } + ts.mu.RLock() defer ts.mu.RUnlock() for t := range ts.Root.tids { @@ -484,7 +489,7 @@ func (ts *TaskSet) forEachFDPaused(f func(*fs.File, *vfs.FileDescription) error) } func (ts *TaskSet) flushWritesToFiles(ctx context.Context) error { - // TODO(gvisor.dev/issues/1663): Add save support for VFS2. + // TODO(gvisor.dev/issue/1663): Add save support for VFS2. return ts.forEachFDPaused(func(file *fs.File, _ *vfs.FileDescription) error { if flags := file.Flags(); !flags.Write { return nil @@ -533,6 +538,11 @@ func (k *Kernel) invalidateUnsavableMappings(ctx context.Context) error { } func (ts *TaskSet) unregisterEpollWaiters() { + // TODO(gvisor.dev/issue/1663): Add save support for VFS2. + if VFS2Enabled { + return + } + ts.mu.RLock() defer ts.mu.RUnlock() for t := range ts.Root.tids { @@ -1005,11 +1015,14 @@ func (k *Kernel) pauseTimeLocked() { // This means we'll iterate FDTables shared by multiple tasks repeatedly, // but ktime.Timer.Pause is idempotent so this is harmless. if t.fdTable != nil { - t.fdTable.forEach(func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) { - if tfd, ok := file.FileOperations.(*timerfd.TimerOperations); ok { - tfd.PauseTimer() - } - }) + // TODO(gvisor.dev/issue/1663): Add save support for VFS2. + if !VFS2Enabled { + t.fdTable.forEach(func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) { + if tfd, ok := file.FileOperations.(*timerfd.TimerOperations); ok { + tfd.PauseTimer() + } + }) + } } } k.timekeeper.PauseUpdates() @@ -1034,12 +1047,15 @@ func (k *Kernel) resumeTimeLocked() { it.ResumeTimer() } } - if t.fdTable != nil { - t.fdTable.forEach(func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) { - if tfd, ok := file.FileOperations.(*timerfd.TimerOperations); ok { - tfd.ResumeTimer() - } - }) + // TODO(gvisor.dev/issue/1663): Add save support for VFS2. + if !VFS2Enabled { + if t.fdTable != nil { + t.fdTable.forEach(func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) { + if tfd, ok := file.FileOperations.(*timerfd.TimerOperations); ok { + tfd.ResumeTimer() + } + }) + } } } } diff --git a/pkg/sentry/syscalls/linux/sys_stat.go b/pkg/sentry/syscalls/linux/sys_stat.go index 9bd2df104..a11a87cd1 100644 --- a/pkg/sentry/syscalls/linux/sys_stat.go +++ b/pkg/sentry/syscalls/linux/sys_stat.go @@ -136,7 +136,10 @@ func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall mask := args[3].Uint() statxAddr := args[4].Pointer() - if mask&linux.STATX__RESERVED > 0 { + if mask&linux.STATX__RESERVED != 0 { + return 0, nil, syserror.EINVAL + } + if flags&^(linux.AT_SYMLINK_NOFOLLOW|linux.AT_EMPTY_PATH|linux.AT_STATX_SYNC_TYPE) != 0 { return 0, nil, syserror.EINVAL } if flags&linux.AT_STATX_SYNC_TYPE == linux.AT_STATX_SYNC_TYPE { diff --git a/pkg/sentry/syscalls/linux/vfs2/stat.go b/pkg/sentry/syscalls/linux/vfs2/stat.go index a74ea6fd5..97eaedd66 100644 --- a/pkg/sentry/syscalls/linux/vfs2/stat.go +++ b/pkg/sentry/syscalls/linux/vfs2/stat.go @@ -150,7 +150,11 @@ func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall mask := args[3].Uint() statxAddr := args[4].Pointer() - if flags&^(linux.AT_EMPTY_PATH|linux.AT_SYMLINK_NOFOLLOW) != 0 { + if flags&^(linux.AT_EMPTY_PATH|linux.AT_SYMLINK_NOFOLLOW|linux.AT_STATX_SYNC_TYPE) != 0 { + return 0, nil, syserror.EINVAL + } + + if mask&linux.STATX__RESERVED != 0 { return 0, nil, syserror.EINVAL } diff --git a/runsc/boot/filter/config.go b/runsc/boot/filter/config.go index a4627905e..f459d1973 100644 --- a/runsc/boot/filter/config.go +++ b/runsc/boot/filter/config.go @@ -284,6 +284,7 @@ var allowedSyscalls = seccomp.SyscallRules{ {seccomp.AllowAny{}, seccomp.AllowValue(syscall.SHUT_RDWR)}, }, syscall.SYS_SIGALTSTACK: {}, + unix.SYS_STATX: {}, syscall.SYS_SYNC_FILE_RANGE: {}, syscall.SYS_TGKILL: []seccomp.Rule{ { diff --git a/test/syscalls/linux/stat.cc b/test/syscalls/linux/stat.cc index c951ac3b3..513b9cd1c 100644 --- a/test/syscalls/linux/stat.cc +++ b/test/syscalls/linux/stat.cc @@ -607,7 +607,7 @@ int statx(int dirfd, const char* pathname, int flags, unsigned int mask, } TEST_F(StatTest, StatxAbsPath) { - SKIP_IF(!IsRunningOnGvisor() && statx(-1, nullptr, 0, 0, 0) < 0 && + SKIP_IF(!IsRunningOnGvisor() && statx(-1, nullptr, 0, 0, nullptr) < 0 && errno == ENOSYS); struct kernel_statx stx; @@ -617,7 +617,7 @@ TEST_F(StatTest, StatxAbsPath) { } TEST_F(StatTest, StatxRelPathDirFD) { - SKIP_IF(!IsRunningOnGvisor() && statx(-1, nullptr, 0, 0, 0) < 0 && + SKIP_IF(!IsRunningOnGvisor() && statx(-1, nullptr, 0, 0, nullptr) < 0 && errno == ENOSYS); struct kernel_statx stx; @@ -631,7 +631,7 @@ TEST_F(StatTest, StatxRelPathDirFD) { } TEST_F(StatTest, StatxRelPathCwd) { - SKIP_IF(!IsRunningOnGvisor() && statx(-1, nullptr, 0, 0, 0) < 0 && + SKIP_IF(!IsRunningOnGvisor() && statx(-1, nullptr, 0, 0, nullptr) < 0 && errno == ENOSYS); ASSERT_THAT(chdir(GetAbsoluteTestTmpdir().c_str()), SyscallSucceeds()); @@ -643,7 +643,7 @@ TEST_F(StatTest, StatxRelPathCwd) { } TEST_F(StatTest, StatxEmptyPath) { - SKIP_IF(!IsRunningOnGvisor() && statx(-1, nullptr, 0, 0, 0) < 0 && + SKIP_IF(!IsRunningOnGvisor() && statx(-1, nullptr, 0, 0, nullptr) < 0 && errno == ENOSYS); const auto fd = ASSERT_NO_ERRNO_AND_VALUE(Open(test_file_name_, O_RDONLY)); @@ -653,6 +653,58 @@ TEST_F(StatTest, StatxEmptyPath) { EXPECT_TRUE(S_ISREG(stx.stx_mode)); } +TEST_F(StatTest, StatxDoesNotRejectExtraneousMaskBits) { + SKIP_IF(!IsRunningOnGvisor() && statx(-1, nullptr, 0, 0, nullptr) < 0 && + errno == ENOSYS); + + struct kernel_statx stx; + // Set all mask bits except for STATX__RESERVED. + uint mask = 0xffffffff & ~0x80000000; + EXPECT_THAT(statx(-1, test_file_name_.c_str(), 0, mask, &stx), + SyscallSucceeds()); + EXPECT_TRUE(S_ISREG(stx.stx_mode)); +} + +TEST_F(StatTest, StatxRejectsReservedMaskBit) { + SKIP_IF(!IsRunningOnGvisor() && statx(-1, nullptr, 0, 0, nullptr) < 0 && + errno == ENOSYS); + + struct kernel_statx stx; + // Set STATX__RESERVED in the mask. + EXPECT_THAT(statx(-1, test_file_name_.c_str(), 0, 0x80000000, &stx), + SyscallFailsWithErrno(EINVAL)); +} + +TEST_F(StatTest, StatxSymlink) { + SKIP_IF(!IsRunningOnGvisor() && statx(-1, nullptr, 0, 0, nullptr) < 0 && + errno == ENOSYS); + + std::string parent_dir = "/tmp"; + TempPath link = ASSERT_NO_ERRNO_AND_VALUE( + TempPath::CreateSymlinkTo(parent_dir, test_file_name_)); + std::string p = link.path(); + + struct kernel_statx stx; + EXPECT_THAT(statx(AT_FDCWD, p.c_str(), AT_SYMLINK_NOFOLLOW, STATX_ALL, &stx), + SyscallSucceeds()); + EXPECT_TRUE(S_ISLNK(stx.stx_mode)); + EXPECT_THAT(statx(AT_FDCWD, p.c_str(), 0, STATX_ALL, &stx), + SyscallSucceeds()); + EXPECT_TRUE(S_ISREG(stx.stx_mode)); +} + +TEST_F(StatTest, StatxInvalidFlags) { + SKIP_IF(!IsRunningOnGvisor() && statx(-1, nullptr, 0, 0, nullptr) < 0 && + errno == ENOSYS); + + struct kernel_statx stx; + EXPECT_THAT(statx(AT_FDCWD, test_file_name_.c_str(), 12345, 0, &stx), + SyscallFailsWithErrno(EINVAL)); + EXPECT_THAT(statx(AT_FDCWD, test_file_name_.c_str(), + 0x6000 /* AT_STATX_SYNC_TYPE */, 0, &stx), + SyscallFailsWithErrno(EINVAL)); +} + } // namespace } // namespace testing -- cgit v1.2.3 From 97127750289b49dd5e29f8ddb4209137e47fe52d Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Sat, 14 Mar 2020 13:46:55 -0700 Subject: Disallow kernfs.Inode.SetStat for readonly inodes Updates #1195, #1193 PiperOrigin-RevId: 300950993 --- pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go | 7 +++--- pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 9 +++++-- pkg/sentry/fsimpl/kernfs/symlink.go | 7 ++++++ pkg/sentry/fsimpl/proc/subtasks.go | 5 ++++ pkg/sentry/fsimpl/proc/task.go | 10 +++----- pkg/sentry/fsimpl/proc/tasks.go | 16 ++++++++++++- pkg/sentry/fsimpl/proc/tasks_files.go | 33 ++++++++++++++++++++++---- pkg/sentry/fsimpl/sys/sys.go | 8 ++++--- 8 files changed, 74 insertions(+), 21 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go index 1c026f4d8..0d27a8867 100644 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -61,9 +61,10 @@ func (f *DynamicBytesFile) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vf return &fd.vfsfd, nil } -// SetStat implements Inode.SetStat. -func (f *DynamicBytesFile) SetStat(*vfs.Filesystem, vfs.SetStatOptions) error { - // DynamicBytesFiles are immutable. +// SetStat implements Inode.SetStat. By default DynamicBytesFile doesn't allow +// inode attributes to be changed. Override SetStat() making it call +// f.InodeAttrs to allow it. +func (*DynamicBytesFile) SetStat(*vfs.Filesystem, vfs.SetStatOptions) error { return syserror.EPERM } diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 94ca3dbdd..4ed41326d 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -228,7 +228,7 @@ func (a *InodeAttrs) Stat(*vfs.Filesystem, vfs.StatOptions) (linux.Statx, error) stat.GID = atomic.LoadUint32(&a.gid) stat.Nlink = atomic.LoadUint32(&a.nlink) - // TODO: Implement other stat fields like timestamps. + // TODO(gvisor.dev/issue/1193): Implement other stat fields like timestamps. return stat, nil } @@ -256,7 +256,7 @@ func (a *InodeAttrs) SetStat(_ *vfs.Filesystem, opts vfs.SetStatOptions) error { // Note that not all fields are modifiable. For example, the file type and // inode numbers are immutable after node creation. - // TODO: Implement other stat fields like timestamps. + // TODO(gvisor.dev/issue/1193): Implement other stat fields like timestamps. return nil } @@ -555,6 +555,11 @@ func (s *StaticDirectory) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs return fd.VFSFileDescription(), nil } +// SetStat implements Inode.SetStat not allowing inode attributes to be changed. +func (*StaticDirectory) SetStat(*vfs.Filesystem, vfs.SetStatOptions) error { + return syserror.EPERM +} + // AlwaysValid partially implements kernfs.inodeDynamicLookup. type AlwaysValid struct{} diff --git a/pkg/sentry/fsimpl/kernfs/symlink.go b/pkg/sentry/fsimpl/kernfs/symlink.go index 0ee7eb9b7..41c5a3099 100644 --- a/pkg/sentry/fsimpl/kernfs/symlink.go +++ b/pkg/sentry/fsimpl/kernfs/symlink.go @@ -18,6 +18,8 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" ) // StaticSymlink provides an Inode implementation for symlinks that point to @@ -52,3 +54,8 @@ func (s *StaticSymlink) Init(creds *auth.Credentials, ino uint64, target string) func (s *StaticSymlink) Readlink(_ context.Context) (string, error) { return s.target, nil } + +// SetStat implements Inode.SetStat not allowing inode attributes to be changed. +func (*StaticSymlink) SetStat(*vfs.Filesystem, vfs.SetStatOptions) error { + return syserror.EPERM +} diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go index a3a7c16a5..ea6d60f6e 100644 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -127,3 +127,8 @@ func (i *subtasksInode) Stat(vsfs *vfs.Filesystem, opts vfs.StatOptions) (linux. } return stat, nil } + +// SetStat implements Inode.SetStat not allowing inode attributes to be changed. +func (*subtasksInode) SetStat(*vfs.Filesystem, vfs.SetStatOptions) error { + return syserror.EPERM +} diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index 4891caab6..fae3fc5aa 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -107,13 +107,9 @@ func (i *taskInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenO return fd.VFSFileDescription(), nil } -// SetStat implements kernfs.Inode. -func (i *taskInode) SetStat(_ *vfs.Filesystem, opts vfs.SetStatOptions) error { - stat := opts.Stat - if stat.Mask&linux.STATX_MODE != 0 { - return syserror.EPERM - } - return nil +// SetStat implements Inode.SetStat not allowing inode attributes to be changed. +func (*taskInode) SetStat(*vfs.Filesystem, vfs.SetStatOptions) error { + return syserror.EPERM } // taskOwnedInode implements kernfs.Inode and overrides inode owner with task diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index 07115664c..9f2ef8200 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -67,7 +67,7 @@ var _ kernfs.Inode = (*tasksInode)(nil) func newTasksInode(inoGen InoGenerator, k *kernel.Kernel, pidns *kernel.PIDNamespace, cgroupControllers map[string]string) (*tasksInode, *kernfs.Dentry) { root := auth.NewRootCredentials(pidns.UserNamespace()) contents := map[string]*kernfs.Dentry{ - "cpuinfo": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(cpuInfoData(k))), + "cpuinfo": newDentry(root, inoGen.NextIno(), 0444, newStaticFileSetStat(cpuInfoData(k))), "filesystems": newDentry(root, inoGen.NextIno(), 0444, &filesystemsData{}), "loadavg": newDentry(root, inoGen.NextIno(), 0444, &loadavgData{}), "sys": newSysDir(root, inoGen, k), @@ -225,6 +225,20 @@ func (i *tasksInode) Stat(vsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Sta return stat, nil } +// staticFileSetStat implements a special static file that allows inode +// attributes to be set. This is to support /proc files that are readonly, but +// allow attributes to be set. +type staticFileSetStat struct { + dynamicBytesFileSetAttr + vfs.StaticData +} + +var _ dynamicInode = (*staticFileSetStat)(nil) + +func newStaticFileSetStat(data string) *staticFileSetStat { + return &staticFileSetStat{StaticData: vfs.StaticData{Data: data}} +} + func cpuInfoData(k *kernel.Kernel) string { features := k.FeatureSet() if features == nil { diff --git a/pkg/sentry/fsimpl/proc/tasks_files.go b/pkg/sentry/fsimpl/proc/tasks_files.go index b99badba8..20085bb39 100644 --- a/pkg/sentry/fsimpl/proc/tasks_files.go +++ b/pkg/sentry/fsimpl/proc/tasks_files.go @@ -26,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/usage" + "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" ) @@ -62,6 +63,11 @@ func (s *selfSymlink) Readlink(ctx context.Context) (string, error) { return strconv.FormatUint(uint64(tgid), 10), nil } +// SetStat implements Inode.SetStat not allowing inode attributes to be changed. +func (*selfSymlink) SetStat(*vfs.Filesystem, vfs.SetStatOptions) error { + return syserror.EPERM +} + type threadSelfSymlink struct { kernfs.InodeAttrs kernfs.InodeNoopRefCount @@ -95,6 +101,23 @@ func (s *threadSelfSymlink) Readlink(ctx context.Context) (string, error) { return fmt.Sprintf("%d/task/%d", tgid, tid), nil } +// SetStat implements Inode.SetStat not allowing inode attributes to be changed. +func (*threadSelfSymlink) SetStat(*vfs.Filesystem, vfs.SetStatOptions) error { + return syserror.EPERM +} + +// dynamicBytesFileSetAttr implements a special file that allows inode +// attributes to be set. This is to support /proc files that are readonly, but +// allow attributes to be set. +type dynamicBytesFileSetAttr struct { + kernfs.DynamicBytesFile +} + +// SetStat implements Inode.SetStat. +func (d *dynamicBytesFileSetAttr) SetStat(fs *vfs.Filesystem, opts vfs.SetStatOptions) error { + return d.DynamicBytesFile.InodeAttrs.SetStat(fs, opts) +} + // cpuStats contains the breakdown of CPU time for /proc/stat. type cpuStats struct { // user is time spent in userspace tasks with non-positive niceness. @@ -137,7 +160,7 @@ func (c cpuStats) String() string { // // +stateify savable type statData struct { - kernfs.DynamicBytesFile + dynamicBytesFileSetAttr } var _ dynamicInode = (*statData)(nil) @@ -201,7 +224,7 @@ func (*statData) Generate(ctx context.Context, buf *bytes.Buffer) error { // // +stateify savable type loadavgData struct { - kernfs.DynamicBytesFile + dynamicBytesFileSetAttr } var _ dynamicInode = (*loadavgData)(nil) @@ -220,7 +243,7 @@ func (*loadavgData) Generate(ctx context.Context, buf *bytes.Buffer) error { // // +stateify savable type meminfoData struct { - kernfs.DynamicBytesFile + dynamicBytesFileSetAttr } var _ dynamicInode = (*meminfoData)(nil) @@ -271,7 +294,7 @@ func (*meminfoData) Generate(ctx context.Context, buf *bytes.Buffer) error { // // +stateify savable type uptimeData struct { - kernfs.DynamicBytesFile + dynamicBytesFileSetAttr } var _ dynamicInode = (*uptimeData)(nil) @@ -290,7 +313,7 @@ func (*uptimeData) Generate(ctx context.Context, buf *bytes.Buffer) error { // // +stateify savable type versionData struct { - kernfs.DynamicBytesFile + dynamicBytesFileSetAttr } var _ dynamicInode = (*versionData)(nil) diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go index c36c4fa11..3928ff2c8 100644 --- a/pkg/sentry/fsimpl/sys/sys.go +++ b/pkg/sentry/fsimpl/sys/sys.go @@ -94,15 +94,17 @@ func (fs *filesystem) newDir(creds *auth.Credentials, mode linux.FileMode, conte return &d.dentry } -// SetStat implements kernfs.Inode.SetStat. -func (d *dir) SetStat(fs *vfs.Filesystem, opts vfs.SetStatOptions) error { +// SetStat implements Inode.SetStat not allowing inode attributes to be changed. +func (*dir) SetStat(*vfs.Filesystem, vfs.SetStatOptions) error { return syserror.EPERM } // Open implements kernfs.Inode.Open. func (d *dir) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &d.OrderedChildren, &opts) + if err := fd.Init(rp.Mount(), vfsd, &d.OrderedChildren, &opts); err != nil { + return nil, err + } return fd.VFSFileDescription(), nil } -- cgit v1.2.3 From 0f60799a4f8c3db567973574147370fc900df55f Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Mon, 16 Mar 2020 13:28:00 -0700 Subject: Add calls to vfs.CheckSetStat to fsimpls Only gofer filesystem was calling vfs.CheckSetStat for vfs.FilesystemImpl.SetStatAt and vfs.FileDescriptionImpl.SetStat. Updates #1193, #1672, #1197 PiperOrigin-RevId: 301226522 --- pkg/sentry/fsimpl/host/host.go | 16 +++++++++++----- pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go | 2 +- pkg/sentry/fsimpl/kernfs/fd_impl_util.go | 4 +++- pkg/sentry/fsimpl/kernfs/filesystem.go | 2 +- pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 14 ++++++++++++-- pkg/sentry/fsimpl/kernfs/kernfs.go | 6 ++++-- pkg/sentry/fsimpl/kernfs/kernfs_test.go | 2 +- pkg/sentry/fsimpl/kernfs/symlink.go | 2 +- pkg/sentry/fsimpl/proc/subtasks.go | 3 ++- pkg/sentry/fsimpl/proc/task.go | 2 +- pkg/sentry/fsimpl/proc/tasks_files.go | 8 ++++---- pkg/sentry/fsimpl/sys/sys.go | 2 +- pkg/sentry/fsimpl/tmpfs/filesystem.go | 2 +- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 11 +++++++++-- pkg/sentry/vfs/file_description.go | 3 ++- pkg/sentry/vfs/filesystem.go | 4 +++- 16 files changed, 57 insertions(+), 26 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 0be812d13..67c050c30 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -114,7 +114,8 @@ type inode struct { ino uint64 // mu protects the inode metadata below. - mu sync.Mutex + // TODO(gvisor.dev/issue/1672): actually protect fields below. + //mu sync.Mutex // mode is the file mode of this inode. Note that this value may become out // of date if the mode is changed on the host, e.g. with chmod. @@ -269,16 +270,20 @@ func (i *inode) fstat(opts vfs.StatOptions) (linux.Statx, error) { } // SetStat implements kernfs.Inode. -func (i *inode) SetStat(_ *vfs.Filesystem, opts vfs.SetStatOptions) error { +func (i *inode) SetStat(fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { s := opts.Stat m := s.Mask if m == 0 { return nil } - if m&(linux.STATX_UID|linux.STATX_GID) != 0 { + if m&^(linux.STATX_MODE|linux.STATX_SIZE|linux.STATX_ATIME|linux.STATX_MTIME) != 0 { return syserror.EPERM } + if err := vfs.CheckSetStat(creds, &s, uint16(i.Mode().Permissions()), i.uid, i.gid); err != nil { + return err + } + if m&linux.STATX_MODE != 0 { if err := syscall.Fchmod(i.hostFD, uint32(s.Mode)); err != nil { return err @@ -375,8 +380,9 @@ type fileDescription struct { } // SetStat implements vfs.FileDescriptionImpl. -func (f *fileDescription) SetStat(_ context.Context, opts vfs.SetStatOptions) error { - return f.inode.SetStat(nil, opts) +func (f *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { + creds := auth.CredentialsFromContext(ctx) + return f.inode.SetStat(nil, creds, opts) } // Stat implements vfs.FileDescriptionImpl. diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go index 0d27a8867..c788d1d62 100644 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -64,7 +64,7 @@ func (f *DynamicBytesFile) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vf // SetStat implements Inode.SetStat. By default DynamicBytesFile doesn't allow // inode attributes to be changed. Override SetStat() making it call // f.InodeAttrs to allow it. -func (*DynamicBytesFile) SetStat(*vfs.Filesystem, vfs.SetStatOptions) error { +func (*DynamicBytesFile) SetStat(*vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM } diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index da821d524..331c82011 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -17,6 +17,7 @@ package kernfs import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" @@ -206,6 +207,7 @@ func (fd *GenericDirectoryFD) Stat(ctx context.Context, opts vfs.StatOptions) (l // SetStat implements vfs.FileDescriptionImpl.SetStat. func (fd *GenericDirectoryFD) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { fs := fd.filesystem() + creds := auth.CredentialsFromContext(ctx) inode := fd.vfsfd.VirtualDentry().Dentry().Impl().(*Dentry).inode - return inode.SetStat(fs, opts) + return inode.SetStat(fs, creds, opts) } diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 3288de290..37fbe2eea 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -636,7 +636,7 @@ func (fs *Filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts if opts.Stat.Mask == 0 { return nil } - return inode.SetStat(fs.VFSFilesystem(), opts) + return inode.SetStat(fs.VFSFilesystem(), rp.Credentials(), opts) } // StatAt implements vfs.FilesystemImpl.StatAt. diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 4ed41326d..851c61b49 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -234,7 +234,17 @@ func (a *InodeAttrs) Stat(*vfs.Filesystem, vfs.StatOptions) (linux.Statx, error) } // SetStat implements Inode.SetStat. -func (a *InodeAttrs) SetStat(_ *vfs.Filesystem, opts vfs.SetStatOptions) error { +func (a *InodeAttrs) SetStat(fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { + if opts.Stat.Mask == 0 { + return nil + } + if opts.Stat.Mask&^(linux.STATX_MODE|linux.STATX_UID|linux.STATX_GID) != 0 { + return syserror.EPERM + } + if err := vfs.CheckSetStat(creds, &opts.Stat, uint16(a.Mode().Permissions()), auth.KUID(atomic.LoadUint32(&a.uid)), auth.KGID(atomic.LoadUint32(&a.gid))); err != nil { + return err + } + stat := opts.Stat if stat.Mask&linux.STATX_MODE != 0 { for { @@ -556,7 +566,7 @@ func (s *StaticDirectory) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs } // SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*StaticDirectory) SetStat(*vfs.Filesystem, vfs.SetStatOptions) error { +func (*StaticDirectory) SetStat(*vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM } diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index 18a34a590..b12b216d2 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -330,8 +330,10 @@ type inodeMetadata interface { Stat(fs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) // SetStat updates the metadata for this inode. This corresponds to - // vfs.FilesystemImpl.SetStatAt. - SetStat(fs *vfs.Filesystem, opts vfs.SetStatOptions) error + // vfs.FilesystemImpl.SetStatAt. Implementations are responsible for checking + // if the operation can be performed (see vfs.CheckSetStat() for common + // checks). + SetStat(fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error } // Precondition: All methods in this interface may only be called on directory diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index 0459fb305..2875e6ffa 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -91,7 +91,7 @@ type attrs struct { kernfs.InodeAttrs } -func (a *attrs) SetStat(fs *vfs.Filesystem, opt vfs.SetStatOptions) error { +func (*attrs) SetStat(*vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM } diff --git a/pkg/sentry/fsimpl/kernfs/symlink.go b/pkg/sentry/fsimpl/kernfs/symlink.go index 41c5a3099..92f709d29 100644 --- a/pkg/sentry/fsimpl/kernfs/symlink.go +++ b/pkg/sentry/fsimpl/kernfs/symlink.go @@ -56,6 +56,6 @@ func (s *StaticSymlink) Readlink(_ context.Context) (string, error) { } // SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*StaticSymlink) SetStat(*vfs.Filesystem, vfs.SetStatOptions) error { +func (*StaticSymlink) SetStat(*vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM } diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go index ea6d60f6e..eb191aba4 100644 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -22,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" ) @@ -129,6 +130,6 @@ func (i *subtasksInode) Stat(vsfs *vfs.Filesystem, opts vfs.StatOptions) (linux. } // SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*subtasksInode) SetStat(*vfs.Filesystem, vfs.SetStatOptions) error { +func (*subtasksInode) SetStat(*vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM } diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index fae3fc5aa..ceb427ffb 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -108,7 +108,7 @@ func (i *taskInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenO } // SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*taskInode) SetStat(*vfs.Filesystem, vfs.SetStatOptions) error { +func (*taskInode) SetStat(*vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM } diff --git a/pkg/sentry/fsimpl/proc/tasks_files.go b/pkg/sentry/fsimpl/proc/tasks_files.go index 20085bb39..d3d99393f 100644 --- a/pkg/sentry/fsimpl/proc/tasks_files.go +++ b/pkg/sentry/fsimpl/proc/tasks_files.go @@ -64,7 +64,7 @@ func (s *selfSymlink) Readlink(ctx context.Context) (string, error) { } // SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*selfSymlink) SetStat(*vfs.Filesystem, vfs.SetStatOptions) error { +func (*selfSymlink) SetStat(*vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM } @@ -102,7 +102,7 @@ func (s *threadSelfSymlink) Readlink(ctx context.Context) (string, error) { } // SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*threadSelfSymlink) SetStat(*vfs.Filesystem, vfs.SetStatOptions) error { +func (*threadSelfSymlink) SetStat(*vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM } @@ -114,8 +114,8 @@ type dynamicBytesFileSetAttr struct { } // SetStat implements Inode.SetStat. -func (d *dynamicBytesFileSetAttr) SetStat(fs *vfs.Filesystem, opts vfs.SetStatOptions) error { - return d.DynamicBytesFile.InodeAttrs.SetStat(fs, opts) +func (d *dynamicBytesFileSetAttr) SetStat(fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { + return d.DynamicBytesFile.InodeAttrs.SetStat(fs, creds, opts) } // cpuStats contains the breakdown of CPU time for /proc/stat. diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go index 3928ff2c8..9c8e63783 100644 --- a/pkg/sentry/fsimpl/sys/sys.go +++ b/pkg/sentry/fsimpl/sys/sys.go @@ -95,7 +95,7 @@ func (fs *filesystem) newDir(creds *auth.Credentials, mode linux.FileMode, conte } // SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*dir) SetStat(*vfs.Filesystem, vfs.SetStatOptions) error { +func (*dir) SetStat(*vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM } diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 02637fca6..6e8b4cae7 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -575,7 +575,7 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts if err != nil { return err } - return d.inode.setStat(opts.Stat) + return d.inode.setStat(rp.Credentials(), &opts.Stat) } // StatAt implements vfs.FilesystemImpl.StatAt. diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 521206305..c18f1e46e 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -299,10 +299,16 @@ func (i *inode) statTo(stat *linux.Statx) { } } -func (i *inode) setStat(stat linux.Statx) error { +func (i *inode) setStat(creds *auth.Credentials, stat *linux.Statx) error { if stat.Mask == 0 { return nil } + if stat.Mask&^(linux.STATX_MODE|linux.STATX_UID|linux.STATX_GID|linux.STATX_ATIME|linux.STATX_MTIME|linux.STATX_CTIME|linux.STATX_SIZE) != 0 { + return syserror.EPERM + } + if err := vfs.CheckSetStat(creds, stat, uint16(atomic.LoadUint32(&i.mode))&^linux.S_IFMT, auth.KUID(atomic.LoadUint32(&i.uid)), auth.KGID(atomic.LoadUint32(&i.gid))); err != nil { + return err + } i.mu.Lock() var ( needsMtimeBump bool @@ -457,5 +463,6 @@ func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linu // SetStat implements vfs.FileDescriptionImpl.SetStat. func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { - return fd.inode().setStat(opts.Stat) + creds := auth.CredentialsFromContext(ctx) + return fd.inode().setStat(creds, &opts.Stat) } diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 9a1ad630c..8ee549dc2 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -286,7 +286,8 @@ type FileDescriptionImpl interface { Stat(ctx context.Context, opts StatOptions) (linux.Statx, error) // SetStat updates metadata for the file represented by the - // FileDescription. + // FileDescription. Implementations are responsible for checking if the + // operation can be performed (see vfs.CheckSetStat() for common checks). SetStat(ctx context.Context, opts SetStatOptions) error // StatFS returns metadata for the filesystem containing the file diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index c43dcff3d..332decce6 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -366,7 +366,9 @@ type FilesystemImpl interface { // ResolvingPath.Resolve*(), then !rp.Done(). RmdirAt(ctx context.Context, rp *ResolvingPath) error - // SetStatAt updates metadata for the file at the given path. + // SetStatAt updates metadata for the file at the given path. Implementations + // are responsible for checking if the operation can be performed + // (see vfs.CheckSetStat() for common checks). // // Errors: // -- cgit v1.2.3 From 2a6c4369be8d0522a1f439aa02bce0eb21d42ea2 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Mon, 16 Mar 2020 15:59:29 -0700 Subject: Enforce file size rlimits in VFS2 Updates #1035 PiperOrigin-RevId: 301255357 --- pkg/sentry/fsimpl/gofer/gofer.go | 2 +- pkg/sentry/fsimpl/gofer/regular_file.go | 5 ++++ pkg/sentry/fsimpl/gofer/special_file.go | 8 +++++++ pkg/sentry/fsimpl/host/default_file.go | 7 +++++- pkg/sentry/fsimpl/host/host.go | 6 ++--- pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go | 2 +- pkg/sentry/fsimpl/kernfs/fd_impl_util.go | 2 +- pkg/sentry/fsimpl/kernfs/filesystem.go | 2 +- pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 6 ++--- pkg/sentry/fsimpl/kernfs/kernfs.go | 2 +- pkg/sentry/fsimpl/kernfs/kernfs_test.go | 2 +- pkg/sentry/fsimpl/kernfs/symlink.go | 2 +- pkg/sentry/fsimpl/proc/subtasks.go | 2 +- pkg/sentry/fsimpl/proc/task.go | 2 +- pkg/sentry/fsimpl/proc/tasks_files.go | 8 +++---- pkg/sentry/fsimpl/sys/sys.go | 2 +- pkg/sentry/fsimpl/tmpfs/filesystem.go | 2 +- pkg/sentry/fsimpl/tmpfs/regular_file.go | 11 +++++++-- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 6 ++--- pkg/sentry/syscalls/linux/vfs2/setstat.go | 15 ++++++++++-- pkg/sentry/vfs/BUILD | 1 + pkg/sentry/vfs/file_description_impl_util.go | 5 ++++ pkg/sentry/vfs/permissions.go | 33 +++++++++++++++++++++++++- 23 files changed, 103 insertions(+), 30 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index c4a8f0b38..999485492 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -713,7 +713,7 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin if stat.Mask&^(linux.STATX_MODE|linux.STATX_UID|linux.STATX_GID|linux.STATX_ATIME|linux.STATX_MTIME|linux.STATX_SIZE) != 0 { return syserror.EPERM } - if err := vfs.CheckSetStat(creds, stat, uint16(atomic.LoadUint32(&d.mode))&^linux.S_IFMT, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))); err != nil { + if err := vfs.CheckSetStat(ctx, creds, stat, uint16(atomic.LoadUint32(&d.mode))&^linux.S_IFMT, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))); err != nil { return err } if err := mnt.CheckBeginWrite(); err != nil { diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go index e95209661..3593eb1d5 100644 --- a/pkg/sentry/fsimpl/gofer/regular_file.go +++ b/pkg/sentry/fsimpl/gofer/regular_file.go @@ -126,6 +126,11 @@ func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off if opts.Flags != 0 { return 0, syserror.EOPNOTSUPP } + limit, err := vfs.CheckLimit(ctx, offset, src.NumBytes()) + if err != nil { + return 0, err + } + src = src.TakeFirst64(limit) d := fd.dentry() d.metadataMu.Lock() diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go index 08c691c47..274f7346f 100644 --- a/pkg/sentry/fsimpl/gofer/special_file.go +++ b/pkg/sentry/fsimpl/gofer/special_file.go @@ -107,6 +107,14 @@ func (fd *specialFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off return 0, syserror.EOPNOTSUPP } + if fd.dentry().fileType() == linux.S_IFREG { + limit, err := vfs.CheckLimit(ctx, offset, src.NumBytes()) + if err != nil { + return 0, err + } + src = src.TakeFirst64(limit) + } + // Do a buffered write. See rationale in PRead. if d := fd.dentry(); d.fs.opts.interop != InteropModeShared { d.touchCMtime(ctx) diff --git a/pkg/sentry/fsimpl/host/default_file.go b/pkg/sentry/fsimpl/host/default_file.go index 98682ba5e..459238603 100644 --- a/pkg/sentry/fsimpl/host/default_file.go +++ b/pkg/sentry/fsimpl/host/default_file.go @@ -112,7 +112,6 @@ func (f *defaultFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offs if f.inode.isStream { return 0, syserror.ESPIPE } - return writeToHostFD(ctx, f.inode.hostFD, src, offset, int(opts.Flags)) } @@ -146,6 +145,12 @@ func writeToHostFD(ctx context.Context, hostFD int, src usermem.IOSequence, offs return 0, syserror.EOPNOTSUPP } + limit, err := vfs.CheckLimit(ctx, offset, src.NumBytes()) + if err != nil { + return 0, err + } + src = src.TakeFirst64(limit) + var writer safemem.Writer if offset == -1 { writer = safemem.FromIOWriter{fd.NewReadWriter(hostFD)} diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 67c050c30..2eebcd60c 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -270,7 +270,7 @@ func (i *inode) fstat(opts vfs.StatOptions) (linux.Statx, error) { } // SetStat implements kernfs.Inode. -func (i *inode) SetStat(fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { +func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { s := opts.Stat m := s.Mask @@ -280,7 +280,7 @@ func (i *inode) SetStat(fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.Se if m&^(linux.STATX_MODE|linux.STATX_SIZE|linux.STATX_ATIME|linux.STATX_MTIME) != 0 { return syserror.EPERM } - if err := vfs.CheckSetStat(creds, &s, uint16(i.Mode().Permissions()), i.uid, i.gid); err != nil { + if err := vfs.CheckSetStat(ctx, creds, &s, uint16(i.Mode().Permissions()), i.uid, i.gid); err != nil { return err } @@ -382,7 +382,7 @@ type fileDescription struct { // SetStat implements vfs.FileDescriptionImpl. func (f *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { creds := auth.CredentialsFromContext(ctx) - return f.inode.SetStat(nil, creds, opts) + return f.inode.SetStat(ctx, nil, creds, opts) } // Stat implements vfs.FileDescriptionImpl. diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go index c788d1d62..d8bddbafa 100644 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -64,7 +64,7 @@ func (f *DynamicBytesFile) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vf // SetStat implements Inode.SetStat. By default DynamicBytesFile doesn't allow // inode attributes to be changed. Override SetStat() making it call // f.InodeAttrs to allow it. -func (*DynamicBytesFile) SetStat(*vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { +func (*DynamicBytesFile) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM } diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index 331c82011..75c4bab1a 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -209,5 +209,5 @@ func (fd *GenericDirectoryFD) SetStat(ctx context.Context, opts vfs.SetStatOptio fs := fd.filesystem() creds := auth.CredentialsFromContext(ctx) inode := fd.vfsfd.VirtualDentry().Dentry().Impl().(*Dentry).inode - return inode.SetStat(fs, creds, opts) + return inode.SetStat(ctx, fs, creds, opts) } diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 37fbe2eea..31da8b511 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -636,7 +636,7 @@ func (fs *Filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts if opts.Stat.Mask == 0 { return nil } - return inode.SetStat(fs.VFSFilesystem(), rp.Credentials(), opts) + return inode.SetStat(ctx, fs.VFSFilesystem(), rp.Credentials(), opts) } // StatAt implements vfs.FilesystemImpl.StatAt. diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 851c61b49..c612dcf07 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -234,14 +234,14 @@ func (a *InodeAttrs) Stat(*vfs.Filesystem, vfs.StatOptions) (linux.Statx, error) } // SetStat implements Inode.SetStat. -func (a *InodeAttrs) SetStat(fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { +func (a *InodeAttrs) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { if opts.Stat.Mask == 0 { return nil } if opts.Stat.Mask&^(linux.STATX_MODE|linux.STATX_UID|linux.STATX_GID) != 0 { return syserror.EPERM } - if err := vfs.CheckSetStat(creds, &opts.Stat, uint16(a.Mode().Permissions()), auth.KUID(atomic.LoadUint32(&a.uid)), auth.KGID(atomic.LoadUint32(&a.gid))); err != nil { + if err := vfs.CheckSetStat(ctx, creds, &opts.Stat, uint16(a.Mode().Permissions()), auth.KUID(atomic.LoadUint32(&a.uid)), auth.KGID(atomic.LoadUint32(&a.gid))); err != nil { return err } @@ -566,7 +566,7 @@ func (s *StaticDirectory) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs } // SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*StaticDirectory) SetStat(*vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { +func (*StaticDirectory) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM } diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index b12b216d2..794e38908 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -333,7 +333,7 @@ type inodeMetadata interface { // vfs.FilesystemImpl.SetStatAt. Implementations are responsible for checking // if the operation can be performed (see vfs.CheckSetStat() for common // checks). - SetStat(fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error + SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error } // Precondition: All methods in this interface may only be called on directory diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index 2875e6ffa..fb0d25ad7 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -91,7 +91,7 @@ type attrs struct { kernfs.InodeAttrs } -func (*attrs) SetStat(*vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { +func (*attrs) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM } diff --git a/pkg/sentry/fsimpl/kernfs/symlink.go b/pkg/sentry/fsimpl/kernfs/symlink.go index 92f709d29..5918d3309 100644 --- a/pkg/sentry/fsimpl/kernfs/symlink.go +++ b/pkg/sentry/fsimpl/kernfs/symlink.go @@ -56,6 +56,6 @@ func (s *StaticSymlink) Readlink(_ context.Context) (string, error) { } // SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*StaticSymlink) SetStat(*vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { +func (*StaticSymlink) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM } diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go index eb191aba4..a21313666 100644 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -130,6 +130,6 @@ func (i *subtasksInode) Stat(vsfs *vfs.Filesystem, opts vfs.StatOptions) (linux. } // SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*subtasksInode) SetStat(*vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { +func (*subtasksInode) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM } diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index ceb427ffb..49d6efb0e 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -108,7 +108,7 @@ func (i *taskInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenO } // SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*taskInode) SetStat(*vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { +func (*taskInode) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM } diff --git a/pkg/sentry/fsimpl/proc/tasks_files.go b/pkg/sentry/fsimpl/proc/tasks_files.go index d3d99393f..882c1981e 100644 --- a/pkg/sentry/fsimpl/proc/tasks_files.go +++ b/pkg/sentry/fsimpl/proc/tasks_files.go @@ -64,7 +64,7 @@ func (s *selfSymlink) Readlink(ctx context.Context) (string, error) { } // SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*selfSymlink) SetStat(*vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { +func (*selfSymlink) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM } @@ -102,7 +102,7 @@ func (s *threadSelfSymlink) Readlink(ctx context.Context) (string, error) { } // SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*threadSelfSymlink) SetStat(*vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { +func (*threadSelfSymlink) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM } @@ -114,8 +114,8 @@ type dynamicBytesFileSetAttr struct { } // SetStat implements Inode.SetStat. -func (d *dynamicBytesFileSetAttr) SetStat(fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { - return d.DynamicBytesFile.InodeAttrs.SetStat(fs, creds, opts) +func (d *dynamicBytesFileSetAttr) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { + return d.DynamicBytesFile.InodeAttrs.SetStat(ctx, fs, creds, opts) } // cpuStats contains the breakdown of CPU time for /proc/stat. diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go index 9c8e63783..7abfd62f2 100644 --- a/pkg/sentry/fsimpl/sys/sys.go +++ b/pkg/sentry/fsimpl/sys/sys.go @@ -95,7 +95,7 @@ func (fs *filesystem) newDir(creds *auth.Credentials, mode linux.FileMode, conte } // SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*dir) SetStat(*vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { +func (*dir) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM } diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 6e8b4cae7..75d01b853 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -575,7 +575,7 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts if err != nil { return err } - return d.inode.setStat(rp.Credentials(), &opts.Stat) + return d.inode.setStat(ctx, rp.Credentials(), &opts.Stat) } // StatAt implements vfs.FilesystemImpl.StatAt. diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index 711442424..5a2896bf6 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -308,11 +308,18 @@ func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off return 0, nil } f := fd.inode().impl.(*regularFile) - end := offset + srclen - if end < offset { + if end := offset + srclen; end < offset { // Overflow. return 0, syserror.EFBIG } + + var err error + srclen, err = vfs.CheckLimit(ctx, offset, srclen) + if err != nil { + return 0, err + } + src = src.TakeFirst64(srclen) + f.inode.mu.Lock() rw := getRegularFileReadWriter(f, offset) n, err := src.CopyInTo(ctx, rw) diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index c18f1e46e..ff69372b3 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -299,14 +299,14 @@ func (i *inode) statTo(stat *linux.Statx) { } } -func (i *inode) setStat(creds *auth.Credentials, stat *linux.Statx) error { +func (i *inode) setStat(ctx context.Context, creds *auth.Credentials, stat *linux.Statx) error { if stat.Mask == 0 { return nil } if stat.Mask&^(linux.STATX_MODE|linux.STATX_UID|linux.STATX_GID|linux.STATX_ATIME|linux.STATX_MTIME|linux.STATX_CTIME|linux.STATX_SIZE) != 0 { return syserror.EPERM } - if err := vfs.CheckSetStat(creds, stat, uint16(atomic.LoadUint32(&i.mode))&^linux.S_IFMT, auth.KUID(atomic.LoadUint32(&i.uid)), auth.KGID(atomic.LoadUint32(&i.gid))); err != nil { + if err := vfs.CheckSetStat(ctx, creds, stat, uint16(atomic.LoadUint32(&i.mode))&^linux.S_IFMT, auth.KUID(atomic.LoadUint32(&i.uid)), auth.KGID(atomic.LoadUint32(&i.gid))); err != nil { return err } i.mu.Lock() @@ -464,5 +464,5 @@ func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linu // SetStat implements vfs.FileDescriptionImpl.SetStat. func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { creds := auth.CredentialsFromContext(ctx) - return fd.inode().setStat(creds, &opts.Stat) + return fd.inode().setStat(ctx, creds, &opts.Stat) } diff --git a/pkg/sentry/syscalls/linux/vfs2/setstat.go b/pkg/sentry/syscalls/linux/vfs2/setstat.go index 9250659ff..136453ccc 100644 --- a/pkg/sentry/syscalls/linux/vfs2/setstat.go +++ b/pkg/sentry/syscalls/linux/vfs2/setstat.go @@ -173,12 +173,13 @@ func Truncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc return 0, nil, err } - return 0, nil, setstatat(t, linux.AT_FDCWD, path, disallowEmptyPath, followFinalSymlink, &vfs.SetStatOptions{ + err = setstatat(t, linux.AT_FDCWD, path, disallowEmptyPath, followFinalSymlink, &vfs.SetStatOptions{ Stat: linux.Statx{ Mask: linux.STATX_SIZE, Size: uint64(length), }, }) + return 0, nil, handleSetSizeError(t, err) } // Ftruncate implements Linux syscall ftruncate(2). @@ -196,12 +197,13 @@ func Ftruncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys } defer file.DecRef() - return 0, nil, file.SetStat(t, vfs.SetStatOptions{ + err := file.SetStat(t, vfs.SetStatOptions{ Stat: linux.Statx{ Mask: linux.STATX_SIZE, Size: uint64(length), }, }) + return 0, nil, handleSetSizeError(t, err) } // Utime implements Linux syscall utime(2). @@ -378,3 +380,12 @@ func setstatat(t *kernel.Task, dirfd int32, path fspath.Path, shouldAllowEmptyPa FollowFinalSymlink: bool(shouldFollowFinalSymlink), }, opts) } + +func handleSetSizeError(t *kernel.Task, err error) error { + if err == syserror.ErrExceedsFileSizeLimit { + // Convert error to EFBIG and send a SIGXFSZ per setrlimit(2). + t.SendSignal(kernel.SignalInfoNoInfo(linux.SIGXFSZ, t, t)) + return syserror.EFBIG + } + return err +} diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index cb4deb068..a2a06fc8f 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -51,6 +51,7 @@ go_library( "//pkg/sentry/fs", "//pkg/sentry/fs/lock", "//pkg/sentry/kernel/auth", + "//pkg/sentry/limits", "//pkg/sentry/memmap", "//pkg/sync", "//pkg/syserror", diff --git a/pkg/sentry/vfs/file_description_impl_util.go b/pkg/sentry/vfs/file_description_impl_util.go index 45191d1c3..d45e602ce 100644 --- a/pkg/sentry/vfs/file_description_impl_util.go +++ b/pkg/sentry/vfs/file_description_impl_util.go @@ -339,6 +339,11 @@ func (fd *DynamicBytesFileDescriptionImpl) pwriteLocked(ctx context.Context, src if opts.Flags&^(linux.RWF_HIPRI|linux.RWF_DSYNC|linux.RWF_SYNC) != 0 { return 0, syserror.EOPNOTSUPP } + limit, err := CheckLimit(ctx, offset, src.NumBytes()) + if err != nil { + return 0, err + } + src = src.TakeFirst64(limit) writable, ok := fd.data.(WritableDynamicBytesSource) if !ok { diff --git a/pkg/sentry/vfs/permissions.go b/pkg/sentry/vfs/permissions.go index 8e250998a..2c8f23f55 100644 --- a/pkg/sentry/vfs/permissions.go +++ b/pkg/sentry/vfs/permissions.go @@ -15,8 +15,12 @@ package vfs import ( + "math" + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/syserror" ) @@ -147,7 +151,16 @@ func MayWriteFileWithOpenFlags(flags uint32) bool { // CheckSetStat checks that creds has permission to change the metadata of a // file with the given permissions, UID, and GID as specified by stat, subject // to the rules of Linux's fs/attr.c:setattr_prepare(). -func CheckSetStat(creds *auth.Credentials, stat *linux.Statx, mode uint16, kuid auth.KUID, kgid auth.KGID) error { +func CheckSetStat(ctx context.Context, creds *auth.Credentials, stat *linux.Statx, mode uint16, kuid auth.KUID, kgid auth.KGID) error { + if stat.Mask&linux.STATX_SIZE != 0 { + limit, err := CheckLimit(ctx, 0, int64(stat.Size)) + if err != nil { + return err + } + if limit < int64(stat.Size) { + return syserror.ErrExceedsFileSizeLimit + } + } if stat.Mask&linux.STATX_MODE != 0 { if !CanActAsOwner(creds, kuid) { return syserror.EPERM @@ -205,3 +218,21 @@ func CanActAsOwner(creds *auth.Credentials, kuid auth.KUID) bool { func HasCapabilityOnFile(creds *auth.Credentials, cp linux.Capability, kuid auth.KUID, kgid auth.KGID) bool { return creds.HasCapability(cp) && creds.UserNamespace.MapFromKUID(kuid).Ok() && creds.UserNamespace.MapFromKGID(kgid).Ok() } + +// CheckLimit enforces file size rlimits. It returns error if the write +// operation must not proceed. Otherwise it returns the max length allowed to +// without violating the limit. +func CheckLimit(ctx context.Context, offset, size int64) (int64, error) { + fileSizeLimit := limits.FromContext(ctx).Get(limits.FileSize).Cur + if fileSizeLimit > math.MaxInt64 { + return size, nil + } + if offset >= int64(fileSizeLimit) { + return 0, syserror.ErrExceedsFileSizeLimit + } + remaining := int64(fileSizeLimit) - offset + if remaining < size { + return remaining, nil + } + return size, nil +} -- cgit v1.2.3 From b55f0e5d40c17cadf68d6238564d675ed12f8f49 Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Mon, 16 Mar 2020 18:28:29 -0700 Subject: fdtable: don't try to zap fdtable entry if close is called for non-existing fd FDTable.setAll is used to zap entries, but it grows the table up to a specified fd. Reported-by: syzbot+9e281b0750d2d4caa190@syzkaller.appspotmail.com PiperOrigin-RevId: 301280000 --- pkg/sentry/kernel/fd_table.go | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/fd_table.go b/pkg/sentry/kernel/fd_table.go index 7de2e509e..dddc28d5a 100644 --- a/pkg/sentry/kernel/fd_table.go +++ b/pkg/sentry/kernel/fd_table.go @@ -536,7 +536,9 @@ func (f *FDTable) Remove(fd int32) (*fs.File, *vfs.FileDescription) { case orig2 != nil: orig2.IncRef() } - f.setAll(fd, nil, nil, FDFlags{}) // Zap entry. + if orig != nil || orig2 != nil { + f.setAll(fd, nil, nil, FDFlags{}) // Zap entry. + } return orig, orig2 } -- cgit v1.2.3 From 42d78ba61bf93f927ee89099abf5ef27b394b0d9 Mon Sep 17 00:00:00 2001 From: Zach Koopmans Date: Tue, 17 Mar 2020 10:29:12 -0700 Subject: Remove HostFS from Sentry. PiperOrigin-RevId: 301402181 --- pkg/sentry/fs/host/BUILD | 4 +- pkg/sentry/fs/host/descriptor.go | 37 +--- pkg/sentry/fs/host/descriptor_state.go | 2 +- pkg/sentry/fs/host/descriptor_test.go | 4 +- pkg/sentry/fs/host/file.go | 4 +- pkg/sentry/fs/host/fs.go | 339 ----------------------------- pkg/sentry/fs/host/fs_test.go | 380 --------------------------------- pkg/sentry/fs/host/host.go | 59 +++++ pkg/sentry/fs/host/inode.go | 141 ++---------- pkg/sentry/fs/host/inode_state.go | 32 +-- pkg/sentry/fs/host/inode_test.go | 66 ------ pkg/sentry/fs/host/util.go | 84 +------- pkg/sentry/fs/host/util_unsafe.go | 41 ---- 13 files changed, 94 insertions(+), 1099 deletions(-) delete mode 100644 pkg/sentry/fs/host/fs.go delete mode 100644 pkg/sentry/fs/host/fs_test.go create mode 100644 pkg/sentry/fs/host/host.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/host/BUILD b/pkg/sentry/fs/host/BUILD index 21003ea45..011625c80 100644 --- a/pkg/sentry/fs/host/BUILD +++ b/pkg/sentry/fs/host/BUILD @@ -10,7 +10,7 @@ go_library( "descriptor_state.go", "device.go", "file.go", - "fs.go", + "host.go", "inode.go", "inode_state.go", "ioctl_unsafe.go", @@ -62,14 +62,12 @@ go_test( size = "small", srcs = [ "descriptor_test.go", - "fs_test.go", "inode_test.go", "socket_test.go", "wait_test.go", ], library = ":host", deps = [ - "//pkg/context", "//pkg/fd", "//pkg/fdnotifier", "//pkg/sentry/contexttest", diff --git a/pkg/sentry/fs/host/descriptor.go b/pkg/sentry/fs/host/descriptor.go index 2a4d1b291..cfdce6a74 100644 --- a/pkg/sentry/fs/host/descriptor.go +++ b/pkg/sentry/fs/host/descriptor.go @@ -16,7 +16,6 @@ package host import ( "fmt" - "path" "syscall" "gvisor.dev/gvisor/pkg/fdnotifier" @@ -28,12 +27,9 @@ import ( // // +stateify savable type descriptor struct { - // donated is true if the host fd was donated by another process. - donated bool - // If origFD >= 0, it is the host fd that this file was originally created // from, which must be available at time of restore. The FD can be closed - // after descriptor is created. Only set if donated is true. + // after descriptor is created. origFD int // wouldBlock is true if value (below) points to a file that can @@ -41,15 +37,13 @@ type descriptor struct { wouldBlock bool // value is the wrapped host fd. It is never saved or restored - // directly. How it is restored depends on whether it was - // donated and the fs.MountSource it was originally - // opened/created from. + // directly. value int `state:"nosave"` } // newDescriptor returns a wrapped host file descriptor. On success, // the descriptor is registered for event notifications with queue. -func newDescriptor(fd int, donated bool, saveable bool, wouldBlock bool, queue *waiter.Queue) (*descriptor, error) { +func newDescriptor(fd int, saveable bool, wouldBlock bool, queue *waiter.Queue) (*descriptor, error) { ownedFD := fd origFD := -1 if saveable { @@ -69,7 +63,6 @@ func newDescriptor(fd int, donated bool, saveable bool, wouldBlock bool, queue * } } return &descriptor{ - donated: donated, origFD: origFD, wouldBlock: wouldBlock, value: ownedFD, @@ -77,25 +70,11 @@ func newDescriptor(fd int, donated bool, saveable bool, wouldBlock bool, queue * } // initAfterLoad initializes the value of the descriptor after Load. -func (d *descriptor) initAfterLoad(mo *superOperations, id uint64, queue *waiter.Queue) error { - if d.donated { - var err error - d.value, err = syscall.Dup(d.origFD) - if err != nil { - return fmt.Errorf("failed to dup restored fd %d: %v", d.origFD, err) - } - } else { - name, ok := mo.inodeMappings[id] - if !ok { - return fmt.Errorf("failed to find path for inode number %d", id) - } - fullpath := path.Join(mo.root, name) - - var err error - d.value, err = open(nil, fullpath) - if err != nil { - return fmt.Errorf("failed to open %q: %v", fullpath, err) - } +func (d *descriptor) initAfterLoad(id uint64, queue *waiter.Queue) error { + var err error + d.value, err = syscall.Dup(d.origFD) + if err != nil { + return fmt.Errorf("failed to dup restored fd %d: %v", d.origFD, err) } if d.wouldBlock { if err := syscall.SetNonblock(d.value, true); err != nil { diff --git a/pkg/sentry/fs/host/descriptor_state.go b/pkg/sentry/fs/host/descriptor_state.go index 8167390a9..e880582ab 100644 --- a/pkg/sentry/fs/host/descriptor_state.go +++ b/pkg/sentry/fs/host/descriptor_state.go @@ -16,7 +16,7 @@ package host // beforeSave is invoked by stateify. func (d *descriptor) beforeSave() { - if d.donated && d.origFD < 0 { + if d.origFD < 0 { panic("donated file descriptor cannot be saved") } } diff --git a/pkg/sentry/fs/host/descriptor_test.go b/pkg/sentry/fs/host/descriptor_test.go index 4205981f5..d8e4605b6 100644 --- a/pkg/sentry/fs/host/descriptor_test.go +++ b/pkg/sentry/fs/host/descriptor_test.go @@ -47,10 +47,10 @@ func TestDescriptorRelease(t *testing.T) { // FD ownership is transferred to the descritor. queue := &waiter.Queue{} - d, err := newDescriptor(fd, false /* donated*/, tc.saveable, tc.wouldBlock, queue) + d, err := newDescriptor(fd, tc.saveable, tc.wouldBlock, queue) if err != nil { syscall.Close(fd) - t.Fatalf("newDescriptor(%d, %t, false, %t, queue) failed, err: %v", fd, tc.saveable, tc.wouldBlock, err) + t.Fatalf("newDescriptor(%d, %t, %t, queue) failed, err: %v", fd, tc.saveable, tc.wouldBlock, err) } if tc.saveable { if d.origFD < 0 { diff --git a/pkg/sentry/fs/host/file.go b/pkg/sentry/fs/host/file.go index e08f56d04..034862694 100644 --- a/pkg/sentry/fs/host/file.go +++ b/pkg/sentry/fs/host/file.go @@ -101,8 +101,8 @@ func newFileFromDonatedFD(ctx context.Context, donated int, mounter fs.FileOwner }) return s, nil default: - msrc := newMountSource(ctx, "/", mounter, &Filesystem{}, fs.MountSourceFlags{}, false /* dontTranslateOwnership */) - inode, err := newInode(ctx, msrc, donated, saveable, true /* donated */) + msrc := fs.NewNonCachingMountSource(ctx, &filesystem{}, fs.MountSourceFlags{}) + inode, err := newInode(ctx, msrc, donated, saveable) if err != nil { return nil, err } diff --git a/pkg/sentry/fs/host/fs.go b/pkg/sentry/fs/host/fs.go deleted file mode 100644 index d3e8e3a36..000000000 --- a/pkg/sentry/fs/host/fs.go +++ /dev/null @@ -1,339 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package host implements an fs.Filesystem for files backed by host -// file descriptors. -package host - -import ( - "fmt" - "path" - "path/filepath" - "strconv" - "strings" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/fs" -) - -// FilesystemName is the name under which Filesystem is registered. -const FilesystemName = "whitelistfs" - -const ( - // whitelistKey is the mount option containing a comma-separated list - // of host paths to whitelist. - whitelistKey = "whitelist" - - // rootPathKey is the mount option containing the root path of the - // mount. - rootPathKey = "root" - - // dontTranslateOwnershipKey is the key to superOperations.dontTranslateOwnership. - dontTranslateOwnershipKey = "dont_translate_ownership" -) - -// maxTraversals determines link traversals in building the whitelist. -const maxTraversals = 10 - -// Filesystem is a pseudo file system that is only available during the setup -// to lock down the configurations. This filesystem should only be mounted at root. -// -// Think twice before exposing this to applications. -// -// +stateify savable -type Filesystem struct { - // whitelist is a set of host paths to whitelist. - paths []string -} - -var _ fs.Filesystem = (*Filesystem)(nil) - -// Name is the identifier of this file system. -func (*Filesystem) Name() string { - return FilesystemName -} - -// AllowUserMount prohibits users from using mount(2) with this file system. -func (*Filesystem) AllowUserMount() bool { - return false -} - -// AllowUserList allows this filesystem to be listed in /proc/filesystems. -func (*Filesystem) AllowUserList() bool { - return true -} - -// Flags returns that there is nothing special about this file system. -func (*Filesystem) Flags() fs.FilesystemFlags { - return 0 -} - -// Mount returns an fs.Inode exposing the host file system. It is intended to be locked -// down in PreExec below. -func (f *Filesystem) Mount(ctx context.Context, _ string, flags fs.MountSourceFlags, data string, _ interface{}) (*fs.Inode, error) { - // Parse generic comma-separated key=value options. - options := fs.GenericMountSourceOptions(data) - - // Grab the whitelist if one was specified. - // TODO(edahlgren/mpratt/hzy): require another option "testonly" in order to allow - // no whitelist. - if wl, ok := options[whitelistKey]; ok { - f.paths = strings.Split(wl, "|") - delete(options, whitelistKey) - } - - // If the rootPath was set, use it. Othewise default to the root of the - // host fs. - rootPath := "/" - if rp, ok := options[rootPathKey]; ok { - rootPath = rp - delete(options, rootPathKey) - - // We must relativize the whitelisted paths to the new root. - for i, p := range f.paths { - rel, err := filepath.Rel(rootPath, p) - if err != nil { - return nil, fmt.Errorf("whitelist path %q must be a child of root path %q", p, rootPath) - } - f.paths[i] = path.Join("/", rel) - } - } - fd, err := open(nil, rootPath) - if err != nil { - return nil, fmt.Errorf("failed to find root: %v", err) - } - - var dontTranslateOwnership bool - if v, ok := options[dontTranslateOwnershipKey]; ok { - b, err := strconv.ParseBool(v) - if err != nil { - return nil, fmt.Errorf("invalid value for %q: %v", dontTranslateOwnershipKey, err) - } - dontTranslateOwnership = b - delete(options, dontTranslateOwnershipKey) - } - - // Fail if the caller passed us more options than we know about. - if len(options) > 0 { - return nil, fmt.Errorf("unsupported mount options: %v", options) - } - - // The mounting EUID/EGID will be cached by this file system. This will - // be used to assign ownership to files that we own. - owner := fs.FileOwnerFromContext(ctx) - - // Construct the host file system mount and inode. - msrc := newMountSource(ctx, rootPath, owner, f, flags, dontTranslateOwnership) - return newInode(ctx, msrc, fd, false /* saveable */, false /* donated */) -} - -// InstallWhitelist locks down the MountNamespace to only the currently installed -// Dirents and the given paths. -func (f *Filesystem) InstallWhitelist(ctx context.Context, m *fs.MountNamespace) error { - return installWhitelist(ctx, m, f.paths) -} - -func installWhitelist(ctx context.Context, m *fs.MountNamespace, paths []string) error { - if len(paths) == 0 || (len(paths) == 1 && paths[0] == "") { - // Warning will be logged during filter installation if the empty - // whitelist matters (allows for host file access). - return nil - } - - // Done tracks entries already added. - done := make(map[string]bool) - root := m.Root() - defer root.DecRef() - - for i := 0; i < len(paths); i++ { - // Make sure the path is absolute. This is a sanity check. - if !path.IsAbs(paths[i]) { - return fmt.Errorf("path %q is not absolute", paths[i]) - } - - // We need to add all the intermediate paths, in case one of - // them is a symlink that needs to be resolved. - for j := 1; j <= len(paths[i]); j++ { - if j < len(paths[i]) && paths[i][j] != '/' { - continue - } - current := paths[i][:j] - - // Lookup the given component in the tree. - remainingTraversals := uint(maxTraversals) - d, err := m.FindLink(ctx, root, nil, current, &remainingTraversals) - if err != nil { - log.Warningf("populate failed for %q: %v", current, err) - continue - } - - // It's critical that this DecRef happens after the - // freeze below. This ensures that the dentry is in - // place to be frozen. Otherwise, we freeze without - // these entries. - defer d.DecRef() - - // Expand the last component if necessary. - if current == paths[i] { - // Is it a directory or symlink? - sattr := d.Inode.StableAttr - if fs.IsDir(sattr) { - for name := range childDentAttrs(ctx, d) { - paths = append(paths, path.Join(current, name)) - } - } - if fs.IsSymlink(sattr) { - // Only expand symlinks once. The - // folder structure may contain - // recursive symlinks and we don't want - // to end up infinitely expanding this - // symlink. This is safe because this - // is the last component. If a later - // path wants to symlink something - // beneath this symlink that will still - // be handled by the FindLink above. - if done[current] { - continue - } - - s, err := d.Inode.Readlink(ctx) - if err != nil { - log.Warningf("readlink failed for %q: %v", current, err) - continue - } - if path.IsAbs(s) { - paths = append(paths, s) - } else { - target := path.Join(path.Dir(current), s) - paths = append(paths, target) - } - } - } - - // Only report this one once even though we may look - // it up more than once. If we whitelist /a/b,/a then - // /a will be "done" when it is looked up for /a/b, - // however we still need to expand all of its contents - // when whitelisting /a. - if !done[current] { - log.Debugf("whitelisted: %s", current) - } - done[current] = true - } - } - - // Freeze the mount tree in place. This prevents any new paths from - // being opened and any old ones from being removed. If we do provide - // tmpfs mounts, we'll want to freeze/thaw those separately. - m.Freeze() - return nil -} - -func childDentAttrs(ctx context.Context, d *fs.Dirent) map[string]fs.DentAttr { - dirname, _ := d.FullName(nil /* root */) - dir, err := d.Inode.GetFile(ctx, d, fs.FileFlags{Read: true}) - if err != nil { - log.Warningf("failed to open directory %q: %v", dirname, err) - return nil - } - dir.DecRef() - var stubSerializer fs.CollectEntriesSerializer - if err := dir.Readdir(ctx, &stubSerializer); err != nil { - log.Warningf("failed to iterate on host directory %q: %v", dirname, err) - return nil - } - delete(stubSerializer.Entries, ".") - delete(stubSerializer.Entries, "..") - return stubSerializer.Entries -} - -// newMountSource constructs a new host fs.MountSource -// relative to a root path. The root should match the mount point. -func newMountSource(ctx context.Context, root string, mounter fs.FileOwner, filesystem fs.Filesystem, flags fs.MountSourceFlags, dontTranslateOwnership bool) *fs.MountSource { - return fs.NewMountSource(ctx, &superOperations{ - root: root, - inodeMappings: make(map[uint64]string), - mounter: mounter, - dontTranslateOwnership: dontTranslateOwnership, - }, filesystem, flags) -} - -// superOperations implements fs.MountSourceOperations. -// -// +stateify savable -type superOperations struct { - fs.SimpleMountSourceOperations - - // root is the path of the mount point. All inode mappings - // are relative to this root. - root string - - // inodeMappings contains mappings of fs.Inodes associated - // with this MountSource to paths under root. - inodeMappings map[uint64]string - - // mounter is the cached EUID/EGID that mounted this file system. - mounter fs.FileOwner - - // dontTranslateOwnership indicates whether to not translate file - // ownership. - // - // By default, files/directories owned by the sandbox uses UID/GID - // of the mounter. For files/directories that are not owned by the - // sandbox, file UID/GID is translated to a UID/GID which cannot - // be mapped in the sandboxed application's user namespace. The - // UID/GID will look like the nobody UID/GID (65534) but is not - // strictly owned by the user "nobody". - // - // If whitelistfs is a lower filesystem in an overlay, set - // dont_translate_ownership=true in mount options. - dontTranslateOwnership bool -} - -var _ fs.MountSourceOperations = (*superOperations)(nil) - -// ResetInodeMappings implements fs.MountSourceOperations.ResetInodeMappings. -func (m *superOperations) ResetInodeMappings() { - m.inodeMappings = make(map[uint64]string) -} - -// SaveInodeMapping implements fs.MountSourceOperations.SaveInodeMapping. -func (m *superOperations) SaveInodeMapping(inode *fs.Inode, path string) { - // This is very unintuitive. We *CANNOT* trust the inode's StableAttrs, - // because overlay copyUp may have changed them out from under us. - // So much for "immutable". - sattr := inode.InodeOperations.(*inodeOperations).fileState.sattr - m.inodeMappings[sattr.InodeID] = path -} - -// Keep implements fs.MountSourceOperations.Keep. -// -// TODO(b/72455313,b/77596690): It is possible to change the permissions on a -// host file while it is in the dirent cache (say from RO to RW), but it is not -// possible to re-open the file with more relaxed permissions, since the host -// FD is already open and stored in the inode. -// -// Using the dirent LRU cache increases the odds that this bug is encountered. -// Since host file access is relatively fast anyways, we disable the LRU cache -// for host fs files. Once we can properly deal with permissions changes and -// re-opening host files, we should revisit whether or not to make use of the -// LRU cache. -func (*superOperations) Keep(*fs.Dirent) bool { - return false -} - -func init() { - fs.RegisterFilesystem(&Filesystem{}) -} diff --git a/pkg/sentry/fs/host/fs_test.go b/pkg/sentry/fs/host/fs_test.go deleted file mode 100644 index 3111d2df9..000000000 --- a/pkg/sentry/fs/host/fs_test.go +++ /dev/null @@ -1,380 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package host - -import ( - "fmt" - "io/ioutil" - "os" - "path" - "reflect" - "sort" - "testing" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" -) - -// newTestMountNamespace creates a MountNamespace with a ramfs root. -// It returns the host folder created, which should be removed when done. -func newTestMountNamespace(t *testing.T) (*fs.MountNamespace, string, error) { - p, err := ioutil.TempDir("", "root") - if err != nil { - return nil, "", err - } - - fd, err := open(nil, p) - if err != nil { - os.RemoveAll(p) - return nil, "", err - } - ctx := contexttest.Context(t) - root, err := newInode(ctx, newMountSource(ctx, p, fs.RootOwner, &Filesystem{}, fs.MountSourceFlags{}, false), fd, false, false) - if err != nil { - os.RemoveAll(p) - return nil, "", err - } - mm, err := fs.NewMountNamespace(ctx, root) - if err != nil { - os.RemoveAll(p) - return nil, "", err - } - return mm, p, nil -} - -// createTestDirs populates the root with some test files and directories. -// /a/a1.txt -// /a/a2.txt -// /b/b1.txt -// /b/c/c1.txt -// /symlinks/normal.txt -// /symlinks/to_normal.txt -> /symlinks/normal.txt -// /symlinks/recursive -> /symlinks -func createTestDirs(ctx context.Context, t *testing.T, m *fs.MountNamespace) error { - r := m.Root() - defer r.DecRef() - - if err := r.CreateDirectory(ctx, r, "a", fs.FilePermsFromMode(0777)); err != nil { - return err - } - - a, err := r.Walk(ctx, r, "a") - if err != nil { - return err - } - defer a.DecRef() - - a1, err := a.Create(ctx, r, "a1.txt", fs.FileFlags{Read: true, Write: true}, fs.FilePermsFromMode(0666)) - if err != nil { - return err - } - a1.DecRef() - - a2, err := a.Create(ctx, r, "a2.txt", fs.FileFlags{Read: true, Write: true}, fs.FilePermsFromMode(0666)) - if err != nil { - return err - } - a2.DecRef() - - if err := r.CreateDirectory(ctx, r, "b", fs.FilePermsFromMode(0777)); err != nil { - return err - } - - b, err := r.Walk(ctx, r, "b") - if err != nil { - return err - } - defer b.DecRef() - - b1, err := b.Create(ctx, r, "b1.txt", fs.FileFlags{Read: true, Write: true}, fs.FilePermsFromMode(0666)) - if err != nil { - return err - } - b1.DecRef() - - if err := b.CreateDirectory(ctx, r, "c", fs.FilePermsFromMode(0777)); err != nil { - return err - } - - c, err := b.Walk(ctx, r, "c") - if err != nil { - return err - } - defer c.DecRef() - - c1, err := c.Create(ctx, r, "c1.txt", fs.FileFlags{Read: true, Write: true}, fs.FilePermsFromMode(0666)) - if err != nil { - return err - } - c1.DecRef() - - if err := r.CreateDirectory(ctx, r, "symlinks", fs.FilePermsFromMode(0777)); err != nil { - return err - } - - symlinks, err := r.Walk(ctx, r, "symlinks") - if err != nil { - return err - } - defer symlinks.DecRef() - - normal, err := symlinks.Create(ctx, r, "normal.txt", fs.FileFlags{Read: true, Write: true}, fs.FilePermsFromMode(0666)) - if err != nil { - return err - } - normal.DecRef() - - if err := symlinks.CreateLink(ctx, r, "/symlinks/normal.txt", "to_normal.txt"); err != nil { - return err - } - - return symlinks.CreateLink(ctx, r, "/symlinks", "recursive") -} - -// allPaths returns a slice of all paths of entries visible in the rootfs. -func allPaths(ctx context.Context, t *testing.T, m *fs.MountNamespace, base string) ([]string, error) { - var paths []string - root := m.Root() - defer root.DecRef() - - maxTraversals := uint(1) - d, err := m.FindLink(ctx, root, nil, base, &maxTraversals) - if err != nil { - t.Logf("FindLink failed for %q", base) - return paths, err - } - defer d.DecRef() - - if fs.IsDir(d.Inode.StableAttr) { - dir, err := d.Inode.GetFile(ctx, d, fs.FileFlags{Read: true}) - if err != nil { - return nil, fmt.Errorf("failed to open directory %q: %v", base, err) - } - iter, ok := dir.FileOperations.(fs.DirIterator) - if !ok { - return nil, fmt.Errorf("cannot directly iterate on host directory %q", base) - } - dirCtx := &fs.DirCtx{ - Serializer: noopDentrySerializer{}, - } - if _, err := fs.DirentReaddir(ctx, d, iter, root, dirCtx, 0); err != nil { - return nil, err - } - for name := range dirCtx.DentAttrs() { - if name == "." || name == ".." { - continue - } - - fullName := path.Join(base, name) - paths = append(paths, fullName) - - // Recurse. - subpaths, err := allPaths(ctx, t, m, fullName) - if err != nil { - return paths, err - } - paths = append(paths, subpaths...) - } - } - - return paths, nil -} - -type noopDentrySerializer struct{} - -func (noopDentrySerializer) CopyOut(string, fs.DentAttr) error { - return nil -} -func (noopDentrySerializer) Written() int { - return 4096 -} - -// pathsEqual returns true if the two string slices contain the same entries. -func pathsEqual(got, want []string) bool { - sort.Strings(got) - sort.Strings(want) - - if len(got) != len(want) { - return false - } - - for i := range got { - if got[i] != want[i] { - return false - } - } - - return true -} - -func TestWhitelist(t *testing.T) { - for _, test := range []struct { - // description of the test. - desc string - // paths are the paths to whitelist - paths []string - // want are all of the directory entries that should be - // visible (nothing beyond this set should be visible). - want []string - }{ - { - desc: "root", - paths: []string{"/"}, - want: []string{"/a", "/a/a1.txt", "/a/a2.txt", "/b", "/b/b1.txt", "/b/c", "/b/c/c1.txt", "/symlinks", "/symlinks/normal.txt", "/symlinks/to_normal.txt", "/symlinks/recursive"}, - }, - { - desc: "top-level directories", - paths: []string{"/a", "/b"}, - want: []string{"/a", "/a/a1.txt", "/a/a2.txt", "/b", "/b/b1.txt", "/b/c", "/b/c/c1.txt"}, - }, - { - desc: "nested directories (1/2)", - paths: []string{"/b", "/b/c"}, - want: []string{"/b", "/b/b1.txt", "/b/c", "/b/c/c1.txt"}, - }, - { - desc: "nested directories (2/2)", - paths: []string{"/b/c", "/b"}, - want: []string{"/b", "/b/b1.txt", "/b/c", "/b/c/c1.txt"}, - }, - { - desc: "single file", - paths: []string{"/b/c/c1.txt"}, - want: []string{"/b", "/b/c", "/b/c/c1.txt"}, - }, - { - desc: "single file and directory", - paths: []string{"/a/a1.txt", "/b/c"}, - want: []string{"/a", "/a/a1.txt", "/b", "/b/c", "/b/c/c1.txt"}, - }, - { - desc: "symlink", - paths: []string{"/symlinks/to_normal.txt"}, - want: []string{"/symlinks", "/symlinks/normal.txt", "/symlinks/to_normal.txt"}, - }, - { - desc: "recursive symlink", - paths: []string{"/symlinks/recursive/normal.txt"}, - want: []string{"/symlinks", "/symlinks/normal.txt", "/symlinks/recursive"}, - }, - } { - t.Run(test.desc, func(t *testing.T) { - m, p, err := newTestMountNamespace(t) - if err != nil { - t.Errorf("Failed to create MountNamespace: %v", err) - } - defer os.RemoveAll(p) - - ctx := withRoot(contexttest.RootContext(t), m.Root()) - if err := createTestDirs(ctx, t, m); err != nil { - t.Errorf("Failed to create test dirs: %v", err) - } - - if err := installWhitelist(ctx, m, test.paths); err != nil { - t.Errorf("installWhitelist(%v) err got %v want nil", test.paths, err) - } - - got, err := allPaths(ctx, t, m, "/") - if err != nil { - t.Fatalf("Failed to lookup paths (whitelisted: %v): %v", test.paths, err) - } - - if !pathsEqual(got, test.want) { - t.Errorf("For paths %v got %v want %v", test.paths, got, test.want) - } - }) - } -} - -func TestRootPath(t *testing.T) { - // Create a temp dir, which will be the root of our mounted fs. - rootPath, err := ioutil.TempDir(os.TempDir(), "root") - if err != nil { - t.Fatalf("TempDir failed: %v", err) - } - defer os.RemoveAll(rootPath) - - // Create two files inside the new root, one which will be whitelisted - // and one not. - whitelisted, err := ioutil.TempFile(rootPath, "white") - if err != nil { - t.Fatalf("TempFile failed: %v", err) - } - if _, err := ioutil.TempFile(rootPath, "black"); err != nil { - t.Fatalf("TempFile failed: %v", err) - } - - // Create a mount with a root path and single whitelisted file. - hostFS := &Filesystem{} - ctx := contexttest.Context(t) - data := fmt.Sprintf("%s=%s,%s=%s", rootPathKey, rootPath, whitelistKey, whitelisted.Name()) - inode, err := hostFS.Mount(ctx, "", fs.MountSourceFlags{}, data, nil) - if err != nil { - t.Fatalf("Mount failed: %v", err) - } - mm, err := fs.NewMountNamespace(ctx, inode) - if err != nil { - t.Fatalf("NewMountNamespace failed: %v", err) - } - if err := hostFS.InstallWhitelist(ctx, mm); err != nil { - t.Fatalf("InstallWhitelist failed: %v", err) - } - - // Get the contents of the root directory. - rootDir := mm.Root() - rctx := withRoot(ctx, rootDir) - f, err := rootDir.Inode.GetFile(rctx, rootDir, fs.FileFlags{}) - if err != nil { - t.Fatalf("GetFile failed: %v", err) - } - c := &fs.CollectEntriesSerializer{} - if err := f.Readdir(rctx, c); err != nil { - t.Fatalf("Readdir failed: %v", err) - } - - // We should have only our whitelisted file, plus the dots. - want := []string{path.Base(whitelisted.Name()), ".", ".."} - got := c.Order - sort.Strings(want) - sort.Strings(got) - if !reflect.DeepEqual(got, want) { - t.Errorf("Readdir got %v, wanted %v", got, want) - } -} - -type rootContext struct { - context.Context - root *fs.Dirent -} - -// withRoot returns a copy of ctx with the given root. -func withRoot(ctx context.Context, root *fs.Dirent) context.Context { - return &rootContext{ - Context: ctx, - root: root, - } -} - -// Value implements Context.Value. -func (rc rootContext) Value(key interface{}) interface{} { - switch key { - case fs.CtxRoot: - rc.root.IncRef() - return rc.root - default: - return rc.Context.Value(key) - } -} diff --git a/pkg/sentry/fs/host/host.go b/pkg/sentry/fs/host/host.go new file mode 100644 index 000000000..081ba1dd8 --- /dev/null +++ b/pkg/sentry/fs/host/host.go @@ -0,0 +1,59 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package host supports file descriptors imported directly. +package host + +import ( + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// filesystem is a host filesystem. +// +// +stateify savable +type filesystem struct{} + +func init() { + fs.RegisterFilesystem(&filesystem{}) +} + +// FilesystemName is the name under which the filesystem is registered. +const FilesystemName = "host" + +// Name is the name of the filesystem. +func (*filesystem) Name() string { + return FilesystemName +} + +// Mount returns an error. Mounting hostfs is not allowed. +func (*filesystem) Mount(ctx context.Context, device string, flags fs.MountSourceFlags, data string, dataObj interface{}) (*fs.Inode, error) { + return nil, syserror.EPERM +} + +// AllowUserMount prohibits users from using mount(2) with this file system. +func (*filesystem) AllowUserMount() bool { + return false +} + +// AllowUserList prohibits this filesystem to be listed in /proc/filesystems. +func (*filesystem) AllowUserList() bool { + return false +} + +// Flags returns that there is nothing special about this file system. +func (*filesystem) Flags() fs.FilesystemFlags { + return 0 +} diff --git a/pkg/sentry/fs/host/inode.go b/pkg/sentry/fs/host/inode.go index 6fa39caab..1da3c0a17 100644 --- a/pkg/sentry/fs/host/inode.go +++ b/pkg/sentry/fs/host/inode.go @@ -17,12 +17,10 @@ package host import ( "syscall" - "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fd" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/secio" - "gvisor.dev/gvisor/pkg/sentry/device" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/memmap" @@ -69,9 +67,6 @@ type inodeOperations struct { // // +stateify savable type inodeFileState struct { - // Common file system state. - mops *superOperations `state:"wait"` - // descriptor is the backing host FD. descriptor *descriptor `state:"wait"` @@ -160,7 +155,7 @@ func (i *inodeFileState) unstableAttr(ctx context.Context) (fs.UnstableAttr, err if err := syscall.Fstat(i.FD(), &s); err != nil { return fs.UnstableAttr{}, err } - return unstableAttr(i.mops, &s), nil + return unstableAttr(&s), nil } // Allocate implements fsutil.CachedFileObject.Allocate. @@ -172,7 +167,7 @@ func (i *inodeFileState) Allocate(_ context.Context, offset, length int64) error var _ fs.InodeOperations = (*inodeOperations)(nil) // newInode returns a new fs.Inode backed by the host FD. -func newInode(ctx context.Context, msrc *fs.MountSource, fd int, saveable bool, donated bool) (*fs.Inode, error) { +func newInode(ctx context.Context, msrc *fs.MountSource, fd int, saveable bool) (*fs.Inode, error) { // Retrieve metadata. var s syscall.Stat_t err := syscall.Fstat(fd, &s) @@ -181,24 +176,17 @@ func newInode(ctx context.Context, msrc *fs.MountSource, fd int, saveable bool, } fileState := &inodeFileState{ - mops: msrc.MountSourceOperations.(*superOperations), sattr: stableAttr(&s), } // Initialize the wrapped host file descriptor. - fileState.descriptor, err = newDescriptor( - fd, - donated, - saveable, - wouldBlock(&s), - &fileState.queue, - ) + fileState.descriptor, err = newDescriptor(fd, saveable, wouldBlock(&s), &fileState.queue) if err != nil { return nil, err } // Build the fs.InodeOperations. - uattr := unstableAttr(msrc.MountSourceOperations.(*superOperations), &s) + uattr := unstableAttr(&s) iops := &inodeOperations{ fileState: fileState, cachingInodeOps: fsutil.NewCachingInodeOperations(ctx, fileState, uattr, fsutil.CachingInodeOperationsOptions{ @@ -232,54 +220,23 @@ func (i *inodeOperations) Release(context.Context) { // Lookup implements fs.InodeOperations.Lookup. func (i *inodeOperations) Lookup(ctx context.Context, dir *fs.Inode, name string) (*fs.Dirent, error) { - // Get a new FD relative to i at name. - fd, err := open(i, name) - if err != nil { - if err == syserror.ENOENT { - return nil, syserror.ENOENT - } - return nil, err - } - - inode, err := newInode(ctx, dir.MountSource, fd, false /* saveable */, false /* donated */) - if err != nil { - return nil, err - } - - // Return the fs.Dirent. - return fs.NewDirent(ctx, inode, name), nil + return nil, syserror.ENOENT } // Create implements fs.InodeOperations.Create. func (i *inodeOperations) Create(ctx context.Context, dir *fs.Inode, name string, flags fs.FileFlags, perm fs.FilePermissions) (*fs.File, error) { - // Create a file relative to i at name. - // - // N.B. We always open this file O_RDWR regardless of flags because a - // future GetFile might want more access. Open allows this regardless - // of perm. - fd, err := openAt(i, name, syscall.O_RDWR|syscall.O_CREAT|syscall.O_EXCL, perm.LinuxMode()) - if err != nil { - return nil, err - } - - inode, err := newInode(ctx, dir.MountSource, fd, false /* saveable */, false /* donated */) - if err != nil { - return nil, err - } + return nil, syserror.EPERM - d := fs.NewDirent(ctx, inode, name) - defer d.DecRef() - return inode.GetFile(ctx, d, flags) } // CreateDirectory implements fs.InodeOperations.CreateDirectory. func (i *inodeOperations) CreateDirectory(ctx context.Context, dir *fs.Inode, name string, perm fs.FilePermissions) error { - return syscall.Mkdirat(i.fileState.FD(), name, uint32(perm.LinuxMode())) + return syserror.EPERM } // CreateLink implements fs.InodeOperations.CreateLink. func (i *inodeOperations) CreateLink(ctx context.Context, dir *fs.Inode, oldname string, newname string) error { - return createLink(i.fileState.FD(), oldname, newname) + return syserror.EPERM } // CreateHardLink implements fs.InodeOperations.CreateHardLink. @@ -294,25 +251,17 @@ func (*inodeOperations) CreateFifo(context.Context, *fs.Inode, string, fs.FilePe // Remove implements fs.InodeOperations.Remove. func (i *inodeOperations) Remove(ctx context.Context, dir *fs.Inode, name string) error { - return unlinkAt(i.fileState.FD(), name, false /* dir */) + return syserror.EPERM } // RemoveDirectory implements fs.InodeOperations.RemoveDirectory. func (i *inodeOperations) RemoveDirectory(ctx context.Context, dir *fs.Inode, name string) error { - return unlinkAt(i.fileState.FD(), name, true /* dir */) + return syserror.EPERM } // Rename implements fs.InodeOperations.Rename. func (i *inodeOperations) Rename(ctx context.Context, inode *fs.Inode, oldParent *fs.Inode, oldName string, newParent *fs.Inode, newName string, replacement bool) error { - op, ok := oldParent.InodeOperations.(*inodeOperations) - if !ok { - return syscall.EXDEV - } - np, ok := newParent.InodeOperations.(*inodeOperations) - if !ok { - return syscall.EXDEV - } - return syscall.Renameat(op.fileState.FD(), oldName, np.fileState.FD(), newName) + return syserror.EPERM } // Bind implements fs.InodeOperations.Bind. @@ -461,69 +410,7 @@ func (i *inodeOperations) NotifyStatusChange(ctx context.Context) {} // readdirAll returns all of the directory entries in i. func (i *inodeOperations) readdirAll(d *dirInfo) (map[string]fs.DentAttr, error) { - i.readdirMu.Lock() - defer i.readdirMu.Unlock() - - fd := i.fileState.FD() - - // syscall.ReadDirent will use getdents, which will seek the file past - // the last directory entry. To read the directory entries a second - // time, we need to seek back to the beginning. - if _, err := syscall.Seek(fd, 0, 0); err != nil { - if err == syscall.ESPIPE { - // All directories should be seekable. If this file - // isn't seekable, it is not a directory and we should - // return that more sane error. - err = syscall.ENOTDIR - } - return nil, err - } - - names := make([]string, 0, 100) - for { - // Refill the buffer if necessary - if d.bufp >= d.nbuf { - d.bufp = 0 - // ReadDirent will just do a sys_getdents64 to the kernel. - n, err := syscall.ReadDirent(fd, d.buf) - if err != nil { - return nil, err - } - if n == 0 { - break // EOF - } - d.nbuf = n - } - - var nb int - // Parse the dirent buffer we just get and return the directory names along - // with the number of bytes consumed in the buffer. - nb, _, names = syscall.ParseDirent(d.buf[d.bufp:d.nbuf], -1, names) - d.bufp += nb - } - - entries := make(map[string]fs.DentAttr) - for _, filename := range names { - // Lookup the type and host device and inode. - stat, lerr := fstatat(fd, filename, linux.AT_SYMLINK_NOFOLLOW) - if lerr == syscall.ENOENT { - // File disappeared between readdir and lstat. - // Just treat it as if it didn't exist. - continue - } - - // There was a serious problem, we should probably report it. - if lerr != nil { - return nil, lerr - } - - entries[filename] = fs.DentAttr{ - Type: nodeType(&stat), - InodeID: hostFileDevice.Map(device.MultiDeviceKey{ - Device: stat.Dev, - Inode: stat.Ino, - }), - } - } - return entries, nil + // We only support non-directory file descriptors that have been + // imported, so just claim that this isn't a directory, even if it is. + return nil, syscall.ENOTDIR } diff --git a/pkg/sentry/fs/host/inode_state.go b/pkg/sentry/fs/host/inode_state.go index 299e0e0b0..1adbd4562 100644 --- a/pkg/sentry/fs/host/inode_state.go +++ b/pkg/sentry/fs/host/inode_state.go @@ -18,29 +18,14 @@ import ( "fmt" "syscall" - "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/device" "gvisor.dev/gvisor/pkg/sentry/fs" ) -// beforeSave is invoked by stateify. -func (i *inodeFileState) beforeSave() { - if !i.queue.IsEmpty() { - panic("event queue must be empty") - } - if !i.descriptor.donated && i.sattr.Type == fs.RegularFile { - uattr, err := i.unstableAttr(context.Background()) - if err != nil { - panic(fs.ErrSaveRejection{fmt.Errorf("failed to get unstable atttribute of %s: %v", i.mops.inodeMappings[i.sattr.InodeID], err)}) - } - i.savedUAttr = &uattr - } -} - // afterLoad is invoked by stateify. func (i *inodeFileState) afterLoad() { // Initialize the descriptor value. - if err := i.descriptor.initAfterLoad(i.mops, i.sattr.InodeID, &i.queue); err != nil { + if err := i.descriptor.initAfterLoad(i.sattr.InodeID, &i.queue); err != nil { panic(fmt.Sprintf("failed to load value of descriptor: %v", err)) } @@ -61,19 +46,4 @@ func (i *inodeFileState) afterLoad() { // change across save and restore, error out. panic(fs.ErrCorruption{fmt.Errorf("host %s conflict in host device mappings: %s", key, hostFileDevice)}) } - - if !i.descriptor.donated && i.sattr.Type == fs.RegularFile { - env, ok := fs.CurrentRestoreEnvironment() - if !ok { - panic("missing restore environment") - } - uattr := unstableAttr(i.mops, &s) - if env.ValidateFileSize && uattr.Size != i.savedUAttr.Size { - panic(fs.ErrCorruption{fmt.Errorf("file size has changed for %s: previously %d, now %d", i.mops.inodeMappings[i.sattr.InodeID], i.savedUAttr.Size, uattr.Size)}) - } - if env.ValidateFileTimestamp && uattr.ModificationTime != i.savedUAttr.ModificationTime { - panic(fs.ErrCorruption{fmt.Errorf("file modification time has changed for %s: previously %v, now %v", i.mops.inodeMappings[i.sattr.InodeID], i.savedUAttr.ModificationTime, uattr.ModificationTime)}) - } - i.savedUAttr = nil - } } diff --git a/pkg/sentry/fs/host/inode_test.go b/pkg/sentry/fs/host/inode_test.go index 7221bc825..4c374681c 100644 --- a/pkg/sentry/fs/host/inode_test.go +++ b/pkg/sentry/fs/host/inode_test.go @@ -15,9 +15,6 @@ package host import ( - "io/ioutil" - "os" - "path" "syscall" "testing" @@ -25,69 +22,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" ) -// TestMultipleReaddir verifies that multiple Readdir calls return the same -// thing if they use different dir contexts. -func TestMultipleReaddir(t *testing.T) { - p, err := ioutil.TempDir("", "readdir") - if err != nil { - t.Fatalf("Failed to create test dir: %v", err) - } - defer os.RemoveAll(p) - - f, err := os.Create(path.Join(p, "a.txt")) - if err != nil { - t.Fatalf("Failed to create a.txt: %v", err) - } - f.Close() - - f, err = os.Create(path.Join(p, "b.txt")) - if err != nil { - t.Fatalf("Failed to create b.txt: %v", err) - } - f.Close() - - fd, err := open(nil, p) - if err != nil { - t.Fatalf("Failed to open %q: %v", p, err) - } - ctx := contexttest.Context(t) - n, err := newInode(ctx, newMountSource(ctx, p, fs.RootOwner, &Filesystem{}, fs.MountSourceFlags{}, false), fd, false, false) - if err != nil { - t.Fatalf("Failed to create inode: %v", err) - } - - dirent := fs.NewDirent(ctx, n, "readdir") - openFile, err := n.GetFile(ctx, dirent, fs.FileFlags{Read: true}) - if err != nil { - t.Fatalf("Failed to get file: %v", err) - } - defer openFile.DecRef() - - c1 := &fs.DirCtx{DirCursor: new(string)} - if _, err := openFile.FileOperations.(*fileOperations).IterateDir(ctx, dirent, c1, 0); err != nil { - t.Fatalf("First Readdir failed: %v", err) - } - - c2 := &fs.DirCtx{DirCursor: new(string)} - if _, err := openFile.FileOperations.(*fileOperations).IterateDir(ctx, dirent, c2, 0); err != nil { - t.Errorf("Second Readdir failed: %v", err) - } - - if _, ok := c1.DentAttrs()["a.txt"]; !ok { - t.Errorf("want a.txt in first Readdir, got %v", c1.DentAttrs()) - } - if _, ok := c1.DentAttrs()["b.txt"]; !ok { - t.Errorf("want b.txt in first Readdir, got %v", c1.DentAttrs()) - } - - if _, ok := c2.DentAttrs()["a.txt"]; !ok { - t.Errorf("want a.txt in second Readdir, got %v", c2.DentAttrs()) - } - if _, ok := c2.DentAttrs()["b.txt"]; !ok { - t.Errorf("want b.txt in second Readdir, got %v", c2.DentAttrs()) - } -} - // TestCloseFD verifies fds will be closed. func TestCloseFD(t *testing.T) { var p [2]int diff --git a/pkg/sentry/fs/host/util.go b/pkg/sentry/fs/host/util.go index 7c60dc1db..388108fdf 100644 --- a/pkg/sentry/fs/host/util.go +++ b/pkg/sentry/fs/host/util.go @@ -16,7 +16,6 @@ package host import ( "os" - "path" "syscall" "gvisor.dev/gvisor/pkg/abi/linux" @@ -28,45 +27,6 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) -func open(parent *inodeOperations, name string) (int, error) { - if parent == nil && !path.IsAbs(name) { - return -1, syserror.EINVAL - } - name = path.Clean(name) - - // Don't follow through symlinks. - flags := syscall.O_NOFOLLOW - - if fd, err := openAt(parent, name, flags|syscall.O_RDWR, 0); err == nil { - return fd, nil - } - // Retry as read-only. - if fd, err := openAt(parent, name, flags|syscall.O_RDONLY, 0); err == nil { - return fd, nil - } - - // Retry as write-only. - if fd, err := openAt(parent, name, flags|syscall.O_WRONLY, 0); err == nil { - return fd, nil - } - - // Retry as a symlink, by including O_PATH as an option. - fd, err := openAt(parent, name, linux.O_PATH|flags, 0) - if err == nil { - return fd, nil - } - - // Everything failed. - return -1, err -} - -func openAt(parent *inodeOperations, name string, flags int, perm linux.FileMode) (int, error) { - if parent == nil { - return syscall.Open(name, flags, uint32(perm)) - } - return syscall.Openat(parent.fileState.FD(), name, flags, uint32(perm)) -} - func nodeType(s *syscall.Stat_t) fs.InodeType { switch x := (s.Mode & syscall.S_IFMT); x { case syscall.S_IFLNK: @@ -107,51 +67,19 @@ func stableAttr(s *syscall.Stat_t) fs.StableAttr { } } -func owner(mo *superOperations, s *syscall.Stat_t) fs.FileOwner { - // User requested no translation, just return actual owner. - if mo.dontTranslateOwnership { - return fs.FileOwner{auth.KUID(s.Uid), auth.KGID(s.Gid)} - } - - // Show only IDs relevant to the sandboxed task. I.e. if we not own the - // file, no sandboxed task can own the file. In that case, we - // use OverflowID for UID, implying that the IDs are not mapped in the - // "root" user namespace. - // - // E.g. - // sandbox's host EUID/EGID is 1/1. - // some_dir's host UID/GID is 2/1. - // Task that mounted this fs has virtualized EUID/EGID 5/5. - // - // If you executed `ls -n` in the sandboxed task, it would show: - // drwxwrxwrx [...] 65534 5 [...] some_dir - - // Files are owned by OverflowID by default. - owner := fs.FileOwner{auth.KUID(auth.OverflowUID), auth.KGID(auth.OverflowGID)} - - // If we own file on host, let mounting task's initial EUID own - // the file. - if s.Uid == hostUID { - owner.UID = mo.mounter.UID - } - - // If our group matches file's group, make file's group match - // the mounting task's initial EGID. - for _, gid := range hostGIDs { - if s.Gid == gid { - owner.GID = mo.mounter.GID - break - } +func owner(s *syscall.Stat_t) fs.FileOwner { + return fs.FileOwner{ + UID: auth.KUID(s.Uid), + GID: auth.KGID(s.Gid), } - return owner } -func unstableAttr(mo *superOperations, s *syscall.Stat_t) fs.UnstableAttr { +func unstableAttr(s *syscall.Stat_t) fs.UnstableAttr { return fs.UnstableAttr{ Size: s.Size, Usage: s.Blocks * 512, Perms: fs.FilePermsFromMode(linux.FileMode(s.Mode)), - Owner: owner(mo, s), + Owner: owner(s), AccessTime: time.FromUnix(s.Atim.Sec, s.Atim.Nsec), ModificationTime: time.FromUnix(s.Mtim.Sec, s.Mtim.Nsec), StatusChangeTime: time.FromUnix(s.Ctim.Sec, s.Ctim.Nsec), diff --git a/pkg/sentry/fs/host/util_unsafe.go b/pkg/sentry/fs/host/util_unsafe.go index 3ab36b088..23bd35d64 100644 --- a/pkg/sentry/fs/host/util_unsafe.go +++ b/pkg/sentry/fs/host/util_unsafe.go @@ -26,26 +26,6 @@ import ( // NulByte is a single NUL byte. It is passed to readlinkat as an empty string. var NulByte byte = '\x00' -func createLink(fd int, name string, linkName string) error { - namePtr, err := syscall.BytePtrFromString(name) - if err != nil { - return err - } - linkNamePtr, err := syscall.BytePtrFromString(linkName) - if err != nil { - return err - } - _, _, errno := syscall.Syscall( - syscall.SYS_SYMLINKAT, - uintptr(unsafe.Pointer(namePtr)), - uintptr(fd), - uintptr(unsafe.Pointer(linkNamePtr))) - if errno != 0 { - return errno - } - return nil -} - func readLink(fd int) (string, error) { // Buffer sizing copied from os.Readlink. for l := 128; ; l *= 2 { @@ -66,27 +46,6 @@ func readLink(fd int) (string, error) { } } -func unlinkAt(fd int, name string, dir bool) error { - namePtr, err := syscall.BytePtrFromString(name) - if err != nil { - return err - } - var flags uintptr - if dir { - flags = linux.AT_REMOVEDIR - } - _, _, errno := syscall.Syscall( - syscall.SYS_UNLINKAT, - uintptr(fd), - uintptr(unsafe.Pointer(namePtr)), - flags, - ) - if errno != 0 { - return errno - } - return nil -} - func timespecFromTimestamp(t ktime.Time, omit, setSysTime bool) syscall.Timespec { if omit { return syscall.Timespec{0, linux.UTIME_OMIT} -- cgit v1.2.3 From f1d1af2a4ad35dd20a7c56bd9e842e347b126c31 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Wed, 18 Mar 2020 15:12:11 -0700 Subject: Fix FDTable.NewFDVFS2 It was looking at VFS1 table to determine where to allocate the next FD from. Updates #1035 PiperOrigin-RevId: 301678858 --- pkg/sentry/kernel/fd_table.go | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/fd_table.go b/pkg/sentry/kernel/fd_table.go index dddc28d5a..d09d97825 100644 --- a/pkg/sentry/kernel/fd_table.go +++ b/pkg/sentry/kernel/fd_table.go @@ -338,7 +338,7 @@ func (f *FDTable) NewFDVFS2(ctx context.Context, minfd int32, file *vfs.FileDesc fd = f.next } for fd < end { - if d, _, _ := f.get(fd); d == nil { + if d, _, _ := f.getVFS2(fd); d == nil { f.setVFS2(fd, file, flags) if fd == f.next { // Update next search start position. -- cgit v1.2.3 From 3a42638a0b32ceede66d8d593609b424bbdba47e Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Wed, 18 Mar 2020 19:08:46 -0700 Subject: Port imported TTY fds to vfs2. Refactor fs/host.TTYFileOperations so that the relevant functionality can be shared with VFS2 (fsimpl/host.ttyFD). Incorporate host.defaultFileFD into the default host.fileDescription. This way, there is no need for a separate default_file.go. As in vfs1, the TTY file implementation can be built on top of this default and override operations as necessary (PRead/Read/PWrite/Write, Release, Ioctl). Note that these changes still need to be plumbed into runsc, which refers to imported TTYs in control/proc.go:ExecAsync. Updates #1672. PiperOrigin-RevId: 301718157 --- pkg/sentry/fs/host/ioctl_unsafe.go | 4 + pkg/sentry/fs/host/tty.go | 5 + pkg/sentry/fs/host/util.go | 8 +- pkg/sentry/fsimpl/host/BUILD | 6 +- pkg/sentry/fsimpl/host/default_file.go | 247 --------------------- pkg/sentry/fsimpl/host/host.go | 258 ++++++++++++++++++++-- pkg/sentry/fsimpl/host/ioctl_unsafe.go | 56 +++++ pkg/sentry/fsimpl/host/tty.go | 379 +++++++++++++++++++++++++++++++++ 8 files changed, 692 insertions(+), 271 deletions(-) delete mode 100644 pkg/sentry/fsimpl/host/default_file.go create mode 100644 pkg/sentry/fsimpl/host/ioctl_unsafe.go create mode 100644 pkg/sentry/fsimpl/host/tty.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/host/ioctl_unsafe.go b/pkg/sentry/fs/host/ioctl_unsafe.go index 271582e54..150ac8e19 100644 --- a/pkg/sentry/fs/host/ioctl_unsafe.go +++ b/pkg/sentry/fs/host/ioctl_unsafe.go @@ -21,6 +21,8 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" ) +// LINT.IfChange + func ioctlGetTermios(fd int) (*linux.Termios, error) { var t linux.Termios _, _, errno := syscall.Syscall(syscall.SYS_IOCTL, uintptr(fd), linux.TCGETS, uintptr(unsafe.Pointer(&t))) @@ -54,3 +56,5 @@ func ioctlSetWinsize(fd int, w *linux.Winsize) error { } return nil } + +// LINT.ThenChange(../../fsimpl/host/ioctl_unsafe.go) diff --git a/pkg/sentry/fs/host/tty.go b/pkg/sentry/fs/host/tty.go index 3f218b4a7..cb91355ab 100644 --- a/pkg/sentry/fs/host/tty.go +++ b/pkg/sentry/fs/host/tty.go @@ -26,6 +26,8 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) +// LINT.IfChange + // TTYFileOperations implements fs.FileOperations for a host file descriptor // that wraps a TTY FD. // @@ -43,6 +45,7 @@ type TTYFileOperations struct { // connected to this TTY. fgProcessGroup *kernel.ProcessGroup + // termios contains the terminal attributes for this TTY. termios linux.KernelTermios } @@ -357,3 +360,5 @@ func (t *TTYFileOperations) checkChange(ctx context.Context, sig linux.Signal) e _ = pg.SendSignal(kernel.SignalInfoPriv(sig)) return kernel.ERESTARTSYS } + +// LINT.ThenChange(../../fsimpl/host/tty.go) diff --git a/pkg/sentry/fs/host/util.go b/pkg/sentry/fs/host/util.go index 388108fdf..1b0356930 100644 --- a/pkg/sentry/fs/host/util.go +++ b/pkg/sentry/fs/host/util.go @@ -23,7 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/device" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/kernel/time" + ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/syserror" ) @@ -80,9 +80,9 @@ func unstableAttr(s *syscall.Stat_t) fs.UnstableAttr { Usage: s.Blocks * 512, Perms: fs.FilePermsFromMode(linux.FileMode(s.Mode)), Owner: owner(s), - AccessTime: time.FromUnix(s.Atim.Sec, s.Atim.Nsec), - ModificationTime: time.FromUnix(s.Mtim.Sec, s.Mtim.Nsec), - StatusChangeTime: time.FromUnix(s.Ctim.Sec, s.Ctim.Nsec), + AccessTime: ktime.FromUnix(s.Atim.Sec, s.Atim.Nsec), + ModificationTime: ktime.FromUnix(s.Mtim.Sec, s.Mtim.Nsec), + StatusChangeTime: ktime.FromUnix(s.Ctim.Sec, s.Ctim.Nsec), Links: uint64(s.Nlink), } } diff --git a/pkg/sentry/fsimpl/host/BUILD b/pkg/sentry/fsimpl/host/BUILD index 5d67f88e3..0bb4a5c3e 100644 --- a/pkg/sentry/fsimpl/host/BUILD +++ b/pkg/sentry/fsimpl/host/BUILD @@ -5,8 +5,9 @@ licenses(["notice"]) go_library( name = "host", srcs = [ - "default_file.go", "host.go", + "ioctl_unsafe.go", + "tty.go", "util.go", ], visibility = ["//pkg/sentry:internal"], @@ -17,9 +18,12 @@ go_library( "//pkg/log", "//pkg/refs", "//pkg/safemem", + "//pkg/sentry/arch", "//pkg/sentry/fsimpl/kernfs", + "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/memmap", + "//pkg/sentry/unimpl", "//pkg/sentry/vfs", "//pkg/sync", "//pkg/syserror", diff --git a/pkg/sentry/fsimpl/host/default_file.go b/pkg/sentry/fsimpl/host/default_file.go deleted file mode 100644 index 459238603..000000000 --- a/pkg/sentry/fsimpl/host/default_file.go +++ /dev/null @@ -1,247 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package host - -import ( - "math" - "syscall" - - "golang.org/x/sys/unix" - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fd" - "gvisor.dev/gvisor/pkg/safemem" - "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -// defaultFileFD implements FileDescriptionImpl for non-socket, non-TTY files. -type defaultFileFD struct { - fileDescription - - // canMap specifies whether we allow the file to be memory mapped. - canMap bool - - // mu protects the fields below. - mu sync.Mutex - - // offset specifies the current file offset. - offset int64 -} - -// TODO(gvisor.dev/issue/1672): Implement Waitable interface. - -// PRead implements FileDescriptionImpl. -func (f *defaultFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null. - if f.inode.isStream { - return 0, syserror.ESPIPE - } - - return readFromHostFD(ctx, f.inode.hostFD, dst, offset, int(opts.Flags)) -} - -// Read implements FileDescriptionImpl. -func (f *defaultFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null. - if f.inode.isStream { - // These files can't be memory mapped, assert this. - if f.canMap { - panic("files that can return EWOULDBLOCK (sockets, pipes, etc.) cannot be memory mapped") - } - - n, err := readFromHostFD(ctx, f.inode.hostFD, dst, -1, int(opts.Flags)) - if isBlockError(err) { - // If we got any data at all, return it as a "completed" partial read - // rather than retrying until complete. - if n != 0 { - err = nil - } else { - err = syserror.ErrWouldBlock - } - } - return n, err - } - // TODO(gvisor.dev/issue/1672): Cache pages, when forced to do so. - f.mu.Lock() - n, err := readFromHostFD(ctx, f.inode.hostFD, dst, f.offset, int(opts.Flags)) - f.offset += n - f.mu.Unlock() - return n, err -} - -func readFromHostFD(ctx context.Context, hostFD int, dst usermem.IOSequence, offset int64, flags int) (int64, error) { - // TODO(gvisor.dev/issue/1672): Support select preadv2 flags. - if flags != 0 { - return 0, syserror.EOPNOTSUPP - } - - var reader safemem.Reader - if offset == -1 { - reader = safemem.FromIOReader{fd.NewReadWriter(hostFD)} - } else { - reader = safemem.FromVecReaderFunc{ - func(srcs [][]byte) (int64, error) { - n, err := unix.Preadv(hostFD, srcs, offset) - return int64(n), err - }, - } - } - n, err := dst.CopyOutFrom(ctx, reader) - return int64(n), err -} - -// PWrite implements FileDescriptionImpl. -func (f *defaultFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null. - if f.inode.isStream { - return 0, syserror.ESPIPE - } - return writeToHostFD(ctx, f.inode.hostFD, src, offset, int(opts.Flags)) -} - -// Write implements FileDescriptionImpl. -func (f *defaultFileFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { - // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null. - if f.inode.isStream { - // These files can't be memory mapped, assert this. - if f.canMap { - panic("files that can return EWOULDBLOCK (sockets, pipes, etc.) cannot be memory mapped") - } - - n, err := writeToHostFD(ctx, f.inode.hostFD, src, -1, int(opts.Flags)) - if isBlockError(err) { - err = syserror.ErrWouldBlock - } - return n, err - } - // TODO(gvisor.dev/issue/1672): Cache pages, when forced to do so. - // TODO(gvisor.dev/issue/1672): Write to end of file and update offset if O_APPEND is set on this file. - f.mu.Lock() - n, err := writeToHostFD(ctx, f.inode.hostFD, src, f.offset, int(opts.Flags)) - f.offset += n - f.mu.Unlock() - return n, err -} - -func writeToHostFD(ctx context.Context, hostFD int, src usermem.IOSequence, offset int64, flags int) (int64, error) { - // TODO(gvisor.dev/issue/1672): Support select pwritev2 flags. - if flags != 0 { - return 0, syserror.EOPNOTSUPP - } - - limit, err := vfs.CheckLimit(ctx, offset, src.NumBytes()) - if err != nil { - return 0, err - } - src = src.TakeFirst64(limit) - - var writer safemem.Writer - if offset == -1 { - writer = safemem.FromIOWriter{fd.NewReadWriter(hostFD)} - } else { - writer = safemem.FromVecWriterFunc{ - func(srcs [][]byte) (int64, error) { - n, err := unix.Pwritev(hostFD, srcs, offset) - return int64(n), err - }, - } - } - n, err := src.CopyInTo(ctx, writer) - return int64(n), err -} - -// Seek implements FileDescriptionImpl. -// -// Note that we do not support seeking on directories, since we do not even -// allow directory fds to be imported at all. -func (f *defaultFileFD) Seek(_ context.Context, offset int64, whence int32) (int64, error) { - // TODO(b/34716638): Some char devices do support seeking, e.g. /dev/null. - if f.inode.isStream { - return 0, syserror.ESPIPE - } - - f.mu.Lock() - defer f.mu.Unlock() - - switch whence { - case linux.SEEK_SET: - if offset < 0 { - return f.offset, syserror.EINVAL - } - f.offset = offset - - case linux.SEEK_CUR: - // Check for overflow. Note that underflow cannot occur, since f.offset >= 0. - if offset > math.MaxInt64-f.offset { - return f.offset, syserror.EOVERFLOW - } - if f.offset+offset < 0 { - return f.offset, syserror.EINVAL - } - f.offset += offset - - case linux.SEEK_END: - var s syscall.Stat_t - if err := syscall.Fstat(f.inode.hostFD, &s); err != nil { - return f.offset, err - } - size := s.Size - - // Check for overflow. Note that underflow cannot occur, since size >= 0. - if offset > math.MaxInt64-size { - return f.offset, syserror.EOVERFLOW - } - if size+offset < 0 { - return f.offset, syserror.EINVAL - } - f.offset = size + offset - - case linux.SEEK_DATA, linux.SEEK_HOLE: - // Modifying the offset in the host file table should not matter, since - // this is the only place where we use it. - // - // For reading and writing, we always rely on our internal offset. - n, err := unix.Seek(f.inode.hostFD, offset, int(whence)) - if err != nil { - return f.offset, err - } - f.offset = n - - default: - // Invalid whence. - return f.offset, syserror.EINVAL - } - - return f.offset, nil -} - -// Sync implements FileDescriptionImpl. -func (f *defaultFileFD) Sync(context.Context) error { - // TODO(gvisor.dev/issue/1672): Currently we do not support the SyncData optimization, so we always sync everything. - return unix.Fsync(f.inode.hostFD) -} - -// ConfigureMMap implements FileDescriptionImpl. -func (f *defaultFileFD) ConfigureMMap(_ context.Context, opts *memmap.MMapOpts) error { - if !f.canMap { - return syserror.ENODEV - } - // TODO(gvisor.dev/issue/1672): Implement ConfigureMMap and Mappable interface. - return syserror.ENODEV -} diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 2eebcd60c..3afb41395 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -19,18 +19,23 @@ package host import ( "errors" "fmt" + "math" "syscall" "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fd" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/refs" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // filesystem implements vfs.FilesystemImpl. @@ -70,10 +75,20 @@ func ImportFD(mnt *vfs.Mount, hostFD int, ownerUID auth.KUID, ownerGID auth.KGID hostFD: hostFD, isStream: isStream, isTTY: isTTY, + canMap: canMap(uint32(fileType)), ino: fs.NextIno(), mode: fileMode, uid: ownerUID, gid: ownerGID, + // For simplicity, set offset to 0. Technically, we should + // only set to 0 on files that are not seekable (sockets, pipes, etc.), + // and use the offset from the host fd otherwise. + offset: 0, + } + + // These files can't be memory mapped, assert this. + if i.isStream && i.canMap { + panic("files that can return EWOULDBLOCK (sockets, pipes, etc.) cannot be memory mapped") } d := &kernfs.Dentry{} @@ -110,12 +125,17 @@ type inode struct { // This field is initialized at creation time and is immutable. isTTY bool + // canMap specifies whether we allow the file to be memory mapped. + // + // This field is initialized at creation time and is immutable. + canMap bool + // ino is an inode number unique within this filesystem. + // + // This field is initialized at creation time and is immutable. ino uint64 - // mu protects the inode metadata below. - // TODO(gvisor.dev/issue/1672): actually protect fields below. - //mu sync.Mutex + // TODO(gvisor.dev/issue/1672): protect mode, uid, and gid with mutex. // mode is the file mode of this inode. Note that this value may become out // of date if the mode is changed on the host, e.g. with chmod. @@ -125,6 +145,12 @@ type inode struct { // file created on import, not the fd on the host. uid auth.KUID gid auth.KGID + + // offsetMu protects offset. + offsetMu sync.Mutex + + // offset specifies the current file offset. + offset int64 } // Note that these flags may become out of date, since they can be modified @@ -336,36 +362,40 @@ func (i *inode) open(d *vfs.Dentry, mnt *vfs.Mount) (*vfs.FileDescription, error // TODO(gvisor.dev/issue/1672): Whitelist specific file types here, so that // we don't allow importing arbitrary file types without proper support. + var ( + vfsfd *vfs.FileDescription + fdImpl vfs.FileDescriptionImpl + ) if i.isTTY { - // TODO(gvisor.dev/issue/1672): support importing host fd as TTY. - return nil, errors.New("importing host fd as TTY not supported") - } - - // For simplicity, set offset to 0. Technically, we should - // only set to 0 on files that are not seekable (sockets, pipes, etc.), - // and use the offset from the host fd otherwise. - fd := &defaultFileFD{ - fileDescription: fileDescription{ - inode: i, - }, - canMap: canMap(uint32(fileType)), - mu: sync.Mutex{}, - offset: 0, + fd := &ttyFD{ + fileDescription: fileDescription{inode: i}, + termios: linux.DefaultSlaveTermios, + } + vfsfd = &fd.vfsfd + fdImpl = fd + } else { + // For simplicity, set offset to 0. Technically, we should + // only set to 0 on files that are not seekable (sockets, pipes, etc.), + // and use the offset from the host fd otherwise. + fd := &fileDescription{inode: i} + vfsfd = &fd.vfsfd + fdImpl = fd } - vfsfd := &fd.vfsfd flags, err := fileFlagsFromHostFD(i.hostFD) if err != nil { return nil, err } - if err := vfsfd.Init(fd, uint32(flags), mnt, d, &vfs.FileDescriptionOptions{}); err != nil { + if err := vfsfd.Init(fdImpl, uint32(flags), mnt, d, &vfs.FileDescriptionOptions{}); err != nil { return nil, err } return vfsfd, nil } // fileDescription is embedded by host fd implementations of FileDescriptionImpl. +// +// TODO(gvisor.dev/issue/1672): Implement Waitable interface. type fileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl @@ -394,3 +424,193 @@ func (f *fileDescription) Stat(_ context.Context, opts vfs.StatOptions) (linux.S func (f *fileDescription) Release() { // noop } + +// PRead implements FileDescriptionImpl. +func (f *fileDescription) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + i := f.inode + // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null. + if i.isStream { + return 0, syserror.ESPIPE + } + + return readFromHostFD(ctx, i.hostFD, dst, offset, opts.Flags) +} + +// Read implements FileDescriptionImpl. +func (f *fileDescription) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + i := f.inode + // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null. + if i.isStream { + n, err := readFromHostFD(ctx, i.hostFD, dst, -1, opts.Flags) + if isBlockError(err) { + // If we got any data at all, return it as a "completed" partial read + // rather than retrying until complete. + if n != 0 { + err = nil + } else { + err = syserror.ErrWouldBlock + } + } + return n, err + } + // TODO(gvisor.dev/issue/1672): Cache pages, when forced to do so. + i.offsetMu.Lock() + n, err := readFromHostFD(ctx, i.hostFD, dst, i.offset, opts.Flags) + i.offset += n + i.offsetMu.Unlock() + return n, err +} + +func readFromHostFD(ctx context.Context, hostFD int, dst usermem.IOSequence, offset int64, flags uint32) (int64, error) { + // TODO(gvisor.dev/issue/1672): Support select preadv2 flags. + if flags != 0 { + return 0, syserror.EOPNOTSUPP + } + + var reader safemem.Reader + if offset == -1 { + reader = safemem.FromIOReader{fd.NewReadWriter(hostFD)} + } else { + reader = safemem.FromVecReaderFunc{ + func(srcs [][]byte) (int64, error) { + n, err := unix.Preadv(hostFD, srcs, offset) + return int64(n), err + }, + } + } + n, err := dst.CopyOutFrom(ctx, reader) + return int64(n), err +} + +// PWrite implements FileDescriptionImpl. +func (f *fileDescription) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + i := f.inode + // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null. + if i.isStream { + return 0, syserror.ESPIPE + } + + return writeToHostFD(ctx, i.hostFD, src, offset, opts.Flags) +} + +// Write implements FileDescriptionImpl. +func (f *fileDescription) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + i := f.inode + // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null. + if i.isStream { + n, err := writeToHostFD(ctx, i.hostFD, src, -1, opts.Flags) + if isBlockError(err) { + err = syserror.ErrWouldBlock + } + return n, err + } + // TODO(gvisor.dev/issue/1672): Cache pages, when forced to do so. + // TODO(gvisor.dev/issue/1672): Write to end of file and update offset if O_APPEND is set on this file. + i.offsetMu.Lock() + n, err := writeToHostFD(ctx, i.hostFD, src, i.offset, opts.Flags) + i.offset += n + i.offsetMu.Unlock() + return n, err +} + +func writeToHostFD(ctx context.Context, hostFD int, src usermem.IOSequence, offset int64, flags uint32) (int64, error) { + // TODO(gvisor.dev/issue/1672): Support select pwritev2 flags. + if flags != 0 { + return 0, syserror.EOPNOTSUPP + } + + var writer safemem.Writer + if offset == -1 { + writer = safemem.FromIOWriter{fd.NewReadWriter(hostFD)} + } else { + writer = safemem.FromVecWriterFunc{ + func(srcs [][]byte) (int64, error) { + n, err := unix.Pwritev(hostFD, srcs, offset) + return int64(n), err + }, + } + } + n, err := src.CopyInTo(ctx, writer) + return int64(n), err +} + +// Seek implements FileDescriptionImpl. +// +// Note that we do not support seeking on directories, since we do not even +// allow directory fds to be imported at all. +func (f *fileDescription) Seek(_ context.Context, offset int64, whence int32) (int64, error) { + i := f.inode + // TODO(b/34716638): Some char devices do support seeking, e.g. /dev/null. + if i.isStream { + return 0, syserror.ESPIPE + } + + i.offsetMu.Lock() + defer i.offsetMu.Unlock() + + switch whence { + case linux.SEEK_SET: + if offset < 0 { + return i.offset, syserror.EINVAL + } + i.offset = offset + + case linux.SEEK_CUR: + // Check for overflow. Note that underflow cannot occur, since i.offset >= 0. + if offset > math.MaxInt64-i.offset { + return i.offset, syserror.EOVERFLOW + } + if i.offset+offset < 0 { + return i.offset, syserror.EINVAL + } + i.offset += offset + + case linux.SEEK_END: + var s syscall.Stat_t + if err := syscall.Fstat(i.hostFD, &s); err != nil { + return i.offset, err + } + size := s.Size + + // Check for overflow. Note that underflow cannot occur, since size >= 0. + if offset > math.MaxInt64-size { + return i.offset, syserror.EOVERFLOW + } + if size+offset < 0 { + return i.offset, syserror.EINVAL + } + i.offset = size + offset + + case linux.SEEK_DATA, linux.SEEK_HOLE: + // Modifying the offset in the host file table should not matter, since + // this is the only place where we use it. + // + // For reading and writing, we always rely on our internal offset. + n, err := unix.Seek(i.hostFD, offset, int(whence)) + if err != nil { + return i.offset, err + } + i.offset = n + + default: + // Invalid whence. + return i.offset, syserror.EINVAL + } + + return i.offset, nil +} + +// Sync implements FileDescriptionImpl. +func (f *fileDescription) Sync(context.Context) error { + // TODO(gvisor.dev/issue/1672): Currently we do not support the SyncData optimization, so we always sync everything. + return unix.Fsync(f.inode.hostFD) +} + +// ConfigureMMap implements FileDescriptionImpl. +func (f *fileDescription) ConfigureMMap(_ context.Context, opts *memmap.MMapOpts) error { + if !f.inode.canMap { + return syserror.ENODEV + } + // TODO(gvisor.dev/issue/1672): Implement ConfigureMMap and Mappable interface. + return syserror.ENODEV +} diff --git a/pkg/sentry/fsimpl/host/ioctl_unsafe.go b/pkg/sentry/fsimpl/host/ioctl_unsafe.go new file mode 100644 index 000000000..0983bf7d8 --- /dev/null +++ b/pkg/sentry/fsimpl/host/ioctl_unsafe.go @@ -0,0 +1,56 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package host + +import ( + "syscall" + "unsafe" + + "gvisor.dev/gvisor/pkg/abi/linux" +) + +func ioctlGetTermios(fd int) (*linux.Termios, error) { + var t linux.Termios + _, _, errno := syscall.Syscall(syscall.SYS_IOCTL, uintptr(fd), linux.TCGETS, uintptr(unsafe.Pointer(&t))) + if errno != 0 { + return nil, errno + } + return &t, nil +} + +func ioctlSetTermios(fd int, req uint64, t *linux.Termios) error { + _, _, errno := syscall.Syscall(syscall.SYS_IOCTL, uintptr(fd), uintptr(req), uintptr(unsafe.Pointer(t))) + if errno != 0 { + return errno + } + return nil +} + +func ioctlGetWinsize(fd int) (*linux.Winsize, error) { + var w linux.Winsize + _, _, errno := syscall.Syscall(syscall.SYS_IOCTL, uintptr(fd), linux.TIOCGWINSZ, uintptr(unsafe.Pointer(&w))) + if errno != 0 { + return nil, errno + } + return &w, nil +} + +func ioctlSetWinsize(fd int, w *linux.Winsize) error { + _, _, errno := syscall.Syscall(syscall.SYS_IOCTL, uintptr(fd), linux.TIOCSWINSZ, uintptr(unsafe.Pointer(w))) + if errno != 0 { + return errno + } + return nil +} diff --git a/pkg/sentry/fsimpl/host/tty.go b/pkg/sentry/fsimpl/host/tty.go new file mode 100644 index 000000000..8936afb06 --- /dev/null +++ b/pkg/sentry/fsimpl/host/tty.go @@ -0,0 +1,379 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package host + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/unimpl" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +// ttyFD implements vfs.FileDescriptionImpl for a host file descriptor +// that wraps a TTY FD. +type ttyFD struct { + fileDescription + + // mu protects the fields below. + mu sync.Mutex `state:"nosave"` + + // session is the session attached to this ttyFD. + session *kernel.Session + + // fgProcessGroup is the foreground process group that is currently + // connected to this TTY. + fgProcessGroup *kernel.ProcessGroup + + // termios contains the terminal attributes for this TTY. + termios linux.KernelTermios +} + +// InitForegroundProcessGroup sets the foreground process group and session for +// the TTY. This should only be called once, after the foreground process group +// has been created, but before it has started running. +func (t *ttyFD) InitForegroundProcessGroup(pg *kernel.ProcessGroup) { + t.mu.Lock() + defer t.mu.Unlock() + if t.fgProcessGroup != nil { + panic("foreground process group is already set") + } + t.fgProcessGroup = pg + t.session = pg.Session() +} + +// ForegroundProcessGroup returns the foreground process for the TTY. +func (t *ttyFD) ForegroundProcessGroup() *kernel.ProcessGroup { + t.mu.Lock() + defer t.mu.Unlock() + return t.fgProcessGroup +} + +// Release implements fs.FileOperations.Release. +func (t *ttyFD) Release() { + t.mu.Lock() + t.fgProcessGroup = nil + t.mu.Unlock() + + t.fileDescription.Release() +} + +// PRead implements vfs.FileDescriptionImpl. +// +// Reading from a TTY is only allowed for foreground process groups. Background +// process groups will either get EIO or a SIGTTIN. +func (t *ttyFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + t.mu.Lock() + defer t.mu.Unlock() + + // Are we allowed to do the read? + // drivers/tty/n_tty.c:n_tty_read()=>job_control()=>tty_check_change(). + if err := t.checkChange(ctx, linux.SIGTTIN); err != nil { + return 0, err + } + + // Do the read. + return t.fileDescription.PRead(ctx, dst, offset, opts) +} + +// Read implements vfs.FileDescriptionImpl. +// +// Reading from a TTY is only allowed for foreground process groups. Background +// process groups will either get EIO or a SIGTTIN. +func (t *ttyFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + t.mu.Lock() + defer t.mu.Unlock() + + // Are we allowed to do the read? + // drivers/tty/n_tty.c:n_tty_read()=>job_control()=>tty_check_change(). + if err := t.checkChange(ctx, linux.SIGTTIN); err != nil { + return 0, err + } + + // Do the read. + return t.fileDescription.Read(ctx, dst, opts) +} + +// PWrite implements vfs.FileDescriptionImpl. +func (t *ttyFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + t.mu.Lock() + defer t.mu.Unlock() + + // Check whether TOSTOP is enabled. This corresponds to the check in + // drivers/tty/n_tty.c:n_tty_write(). + if t.termios.LEnabled(linux.TOSTOP) { + if err := t.checkChange(ctx, linux.SIGTTOU); err != nil { + return 0, err + } + } + return t.fileDescription.PWrite(ctx, src, offset, opts) +} + +// Write implements vfs.FileDescriptionImpl. +func (t *ttyFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + t.mu.Lock() + defer t.mu.Unlock() + + // Check whether TOSTOP is enabled. This corresponds to the check in + // drivers/tty/n_tty.c:n_tty_write(). + if t.termios.LEnabled(linux.TOSTOP) { + if err := t.checkChange(ctx, linux.SIGTTOU); err != nil { + return 0, err + } + } + return t.fileDescription.Write(ctx, src, opts) +} + +// Ioctl implements vfs.FileDescriptionImpl. +func (t *ttyFD) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { + // Ignore arg[0]. This is the real FD: + fd := t.inode.hostFD + ioctl := args[1].Uint64() + switch ioctl { + case linux.TCGETS: + termios, err := ioctlGetTermios(fd) + if err != nil { + return 0, err + } + _, err = usermem.CopyObjectOut(ctx, io, args[2].Pointer(), termios, usermem.IOOpts{ + AddressSpaceActive: true, + }) + return 0, err + + case linux.TCSETS, linux.TCSETSW, linux.TCSETSF: + t.mu.Lock() + defer t.mu.Unlock() + + if err := t.checkChange(ctx, linux.SIGTTOU); err != nil { + return 0, err + } + + var termios linux.Termios + if _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &termios, usermem.IOOpts{ + AddressSpaceActive: true, + }); err != nil { + return 0, err + } + err := ioctlSetTermios(fd, ioctl, &termios) + if err == nil { + t.termios.FromTermios(termios) + } + return 0, err + + case linux.TIOCGPGRP: + // Args: pid_t *argp + // When successful, equivalent to *argp = tcgetpgrp(fd). + // Get the process group ID of the foreground process group on this + // terminal. + + pidns := kernel.PIDNamespaceFromContext(ctx) + if pidns == nil { + return 0, syserror.ENOTTY + } + + t.mu.Lock() + defer t.mu.Unlock() + + // Map the ProcessGroup into a ProcessGroupID in the task's PID namespace. + pgID := pidns.IDOfProcessGroup(t.fgProcessGroup) + _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), &pgID, usermem.IOOpts{ + AddressSpaceActive: true, + }) + return 0, err + + case linux.TIOCSPGRP: + // Args: const pid_t *argp + // Equivalent to tcsetpgrp(fd, *argp). + // Set the foreground process group ID of this terminal. + + task := kernel.TaskFromContext(ctx) + if task == nil { + return 0, syserror.ENOTTY + } + + t.mu.Lock() + defer t.mu.Unlock() + + // Check that we are allowed to set the process group. + if err := t.checkChange(ctx, linux.SIGTTOU); err != nil { + // drivers/tty/tty_io.c:tiocspgrp() converts -EIO from tty_check_change() + // to -ENOTTY. + if err == syserror.EIO { + return 0, syserror.ENOTTY + } + return 0, err + } + + // Check that calling task's process group is in the TTY session. + if task.ThreadGroup().Session() != t.session { + return 0, syserror.ENOTTY + } + + var pgID kernel.ProcessGroupID + if _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &pgID, usermem.IOOpts{ + AddressSpaceActive: true, + }); err != nil { + return 0, err + } + + // pgID must be non-negative. + if pgID < 0 { + return 0, syserror.EINVAL + } + + // Process group with pgID must exist in this PID namespace. + pidns := task.PIDNamespace() + pg := pidns.ProcessGroupWithID(pgID) + if pg == nil { + return 0, syserror.ESRCH + } + + // Check that new process group is in the TTY session. + if pg.Session() != t.session { + return 0, syserror.EPERM + } + + t.fgProcessGroup = pg + return 0, nil + + case linux.TIOCGWINSZ: + // Args: struct winsize *argp + // Get window size. + winsize, err := ioctlGetWinsize(fd) + if err != nil { + return 0, err + } + _, err = usermem.CopyObjectOut(ctx, io, args[2].Pointer(), winsize, usermem.IOOpts{ + AddressSpaceActive: true, + }) + return 0, err + + case linux.TIOCSWINSZ: + // Args: const struct winsize *argp + // Set window size. + + // Unlike setting the termios, any process group (even background ones) can + // set the winsize. + + var winsize linux.Winsize + if _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &winsize, usermem.IOOpts{ + AddressSpaceActive: true, + }); err != nil { + return 0, err + } + err := ioctlSetWinsize(fd, &winsize) + return 0, err + + // Unimplemented commands. + case linux.TIOCSETD, + linux.TIOCSBRK, + linux.TIOCCBRK, + linux.TCSBRK, + linux.TCSBRKP, + linux.TIOCSTI, + linux.TIOCCONS, + linux.FIONBIO, + linux.TIOCEXCL, + linux.TIOCNXCL, + linux.TIOCGEXCL, + linux.TIOCNOTTY, + linux.TIOCSCTTY, + linux.TIOCGSID, + linux.TIOCGETD, + linux.TIOCVHANGUP, + linux.TIOCGDEV, + linux.TIOCMGET, + linux.TIOCMSET, + linux.TIOCMBIC, + linux.TIOCMBIS, + linux.TIOCGICOUNT, + linux.TCFLSH, + linux.TIOCSSERIAL, + linux.TIOCGPTPEER: + + unimpl.EmitUnimplementedEvent(ctx) + fallthrough + default: + return 0, syserror.ENOTTY + } +} + +// checkChange checks that the process group is allowed to read, write, or +// change the state of the TTY. +// +// This corresponds to Linux drivers/tty/tty_io.c:tty_check_change(). The logic +// is a bit convoluted, but documented inline. +// +// Preconditions: t.mu must be held. +func (t *ttyFD) checkChange(ctx context.Context, sig linux.Signal) error { + task := kernel.TaskFromContext(ctx) + if task == nil { + // No task? Linux does not have an analog for this case, but + // tty_check_change is more of a blacklist of cases than a + // whitelist, and is surprisingly permissive. Allowing the + // change seems most appropriate. + return nil + } + + tg := task.ThreadGroup() + pg := tg.ProcessGroup() + + // If the session for the task is different than the session for the + // controlling TTY, then the change is allowed. Seems like a bad idea, + // but that's exactly what linux does. + if tg.Session() != t.fgProcessGroup.Session() { + return nil + } + + // If we are the foreground process group, then the change is allowed. + if pg == t.fgProcessGroup { + return nil + } + + // We are not the foreground process group. + + // Is the provided signal blocked or ignored? + if (task.SignalMask()&linux.SignalSetOf(sig) != 0) || tg.SignalHandlers().IsIgnored(sig) { + // If the signal is SIGTTIN, then we are attempting to read + // from the TTY. Don't send the signal and return EIO. + if sig == linux.SIGTTIN { + return syserror.EIO + } + + // Otherwise, we are writing or changing terminal state. This is allowed. + return nil + } + + // If the process group is an orphan, return EIO. + if pg.IsOrphan() { + return syserror.EIO + } + + // Otherwise, send the signal to the process group and return ERESTARTSYS. + // + // Note that Linux also unconditionally sets TIF_SIGPENDING on current, + // but this isn't necessary in gVisor because the rationale given in + // 040b6362d58f "tty: fix leakage of -ERESTARTSYS to userland" doesn't + // apply: the sentry will handle -ERESTARTSYS in + // kernel.runApp.execute() even if the kernel.Task isn't interrupted. + // + // Linux ignores the result of kill_pgrp(). + _ = pg.SendSignal(kernel.SignalInfoPriv(sig)) + return kernel.ERESTARTSYS +} -- cgit v1.2.3 From e9e399c25d4fcad2adfe92d73b192b9784774964 Mon Sep 17 00:00:00 2001 From: Bhasker Hariharan Date: Thu, 19 Mar 2020 07:18:47 -0700 Subject: Remove workMu from tcpip.Endpoint. workMu is removed and e.mu is now a mutex that supports TryLock. The packet processing path tries to lock the mutex and if its locked it will just queue the packet and move on. The endpoint.UnlockUser() will process any backlog of packets before unlocking the socket. This simplifies the locking inside tcp endpoints a lot. Further the endpoint.LockUser() implements spinning as long as the lock is not held by another syscall goroutine. This ensures low latency as not spinning leads to the task thread being put to sleep if the lock is held by the packet dispatch path. This is suboptimal as the lower layer rarely holds the lock for long so implementing spinning here helps. If the lock is held by another task goroutine then we just proceed to call LockUser() and the task could be put to sleep. The protocol goroutines themselves just call e.mu.Lock() and block if the lock is currently not available. Updates #231, #357 PiperOrigin-RevId: 301808349 --- pkg/sentry/kernel/epoll/epoll.go | 2 + pkg/sentry/socket/netstack/netstack.go | 24 +- pkg/tcpip/transport/tcp/accept.go | 90 +++--- pkg/tcpip/transport/tcp/connect.go | 78 ++--- pkg/tcpip/transport/tcp/dispatcher.go | 8 +- pkg/tcpip/transport/tcp/endpoint.go | 495 ++++++++++++++++-------------- pkg/tcpip/transport/tcp/endpoint_state.go | 5 +- pkg/tcpip/transport/tcp/protocol.go | 38 +-- pkg/tcpip/transport/tcp/rcv.go | 6 - pkg/tcpip/transport/tcp/segment_queue.go | 8 +- pkg/tcpip/transport/tcp/snd.go | 3 - pkg/tcpip/transport/tcp/tcp_test.go | 41 ++- 12 files changed, 424 insertions(+), 374 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/epoll/epoll.go b/pkg/sentry/kernel/epoll/epoll.go index 8bffb78fc..592650923 100644 --- a/pkg/sentry/kernel/epoll/epoll.go +++ b/pkg/sentry/kernel/epoll/epoll.go @@ -296,8 +296,10 @@ func (*readyCallback) Callback(w *waiter.Entry) { e.waitingList.Remove(entry) e.readyList.PushBack(entry) entry.curList = &e.readyList + e.listsMu.Unlock() e.Notify(waiter.EventIn) + return } e.listsMu.Unlock() diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 13a9a60b4..a2e1da02f 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -29,6 +29,7 @@ import ( "io" "math" "reflect" + "sync/atomic" "syscall" "time" @@ -264,6 +265,12 @@ type SocketOperations struct { skType linux.SockType protocol int + // readViewHasData is 1 iff readView has data to be read, 0 otherwise. + // Must be accessed using atomic operations. It must only be written + // with readMu held but can be read without holding readMu. The latter + // is required to avoid deadlocks in epoll Readiness checks. + readViewHasData uint32 + // readMu protects access to the below fields. readMu sync.Mutex `state:"nosave"` // readView contains the remaining payload from the last packet. @@ -410,21 +417,24 @@ func (s *SocketOperations) isPacketBased() bool { // fetchReadView updates the readView field of the socket if it's currently // empty. It assumes that the socket is locked. +// +// Precondition: s.readMu must be held. func (s *SocketOperations) fetchReadView() *syserr.Error { if len(s.readView) > 0 { return nil } - s.readView = nil s.sender = tcpip.FullAddress{} v, cms, err := s.Endpoint.Read(&s.sender) if err != nil { + atomic.StoreUint32(&s.readViewHasData, 0) return syserr.TranslateNetstackError(err) } s.readView = v s.readCM = cms + atomic.StoreUint32(&s.readViewHasData, 1) return nil } @@ -623,11 +633,9 @@ func (s *SocketOperations) Readiness(mask waiter.EventMask) waiter.EventMask { // Check our cached value iff the caller asked for readability and the // endpoint itself is currently not readable. if (mask & ^r & waiter.EventIn) != 0 { - s.readMu.Lock() - if len(s.readView) > 0 { + if atomic.LoadUint32(&s.readViewHasData) == 1 { r |= waiter.EventIn } - s.readMu.Unlock() } return r @@ -2334,6 +2342,10 @@ func (s *SocketOperations) coalescingRead(ctx context.Context, dst usermem.IOSeq } copied += n s.readView.TrimFront(n) + if len(s.readView) == 0 { + atomic.StoreUint32(&s.readViewHasData, 0) + } + dst = dst.DropFirst(n) if e != nil { err = syserr.FromError(e) @@ -2456,6 +2468,10 @@ func (s *SocketOperations) nonBlockingRead(ctx context.Context, dst usermem.IOSe s.readView.TrimFront(int(n)) } + if len(s.readView) == 0 { + atomic.StoreUint32(&s.readViewHasData, 0) + } + var flags int if msgLen > int(n) { flags |= linux.MSG_TRUNC diff --git a/pkg/tcpip/transport/tcp/accept.go b/pkg/tcpip/transport/tcp/accept.go index 85049e54e..4d7602d54 100644 --- a/pkg/tcpip/transport/tcp/accept.go +++ b/pkg/tcpip/transport/tcp/accept.go @@ -221,7 +221,8 @@ func (l *listenContext) isCookieValid(id stack.TransportEndpointID, cookie seqnu } // createConnectingEndpoint creates a new endpoint in a connecting state, with -// the connection parameters given by the arguments. +// the connection parameters given by the arguments. The endpoint is returned +// with n.mu held. func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, irs seqnum.Value, rcvdSynOpts *header.TCPSynOptions, queue *waiter.Queue) (*endpoint, *tcpip.Error) { // Create a new endpoint. netProto := l.netProto @@ -243,21 +244,6 @@ func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, i n.initGSO() - // Now inherit any socket options that should be inherited from the - // listening endpoint. - // In case of Forwarder listenEP will be nil and hence this check. - if l.listenEP != nil { - l.listenEP.propagateInheritableOptions(n) - } - - // Register new endpoint so that packets are routed to it. - if err := n.stack.RegisterTransportEndpoint(n.boundNICID, n.effectiveNetProtos, ProtocolNumber, n.ID, n, n.reusePort, n.boundBindToDevice); err != nil { - n.Close() - return nil, err - } - - n.isRegistered = true - // Create sender and receiver. // // The receiver at least temporarily has a zero receive window scale, @@ -269,11 +255,27 @@ func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, i // window to grow to a really large value. n.rcvAutoParams.prevCopied = n.initialReceiveWindow() + // Lock the endpoint before registering to ensure that no out of + // band changes are possible due to incoming packets etc till + // the endpoint is done initializing. + n.mu.Lock() + + // Register new endpoint so that packets are routed to it. + if err := n.stack.RegisterTransportEndpoint(n.boundNICID, n.effectiveNetProtos, ProtocolNumber, n.ID, n, n.reusePort, n.boundBindToDevice); err != nil { + n.mu.Unlock() + n.Close() + return nil, err + } + + n.isRegistered = true + return n, nil } // createEndpointAndPerformHandshake creates a new endpoint in connected state // and then performs the TCP 3-way handshake. +// +// The new endpoint is returned with e.mu held. func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *header.TCPSynOptions, queue *waiter.Queue) (*endpoint, *tcpip.Error) { // Create new endpoint. irs := s.sequenceNumber @@ -289,9 +291,25 @@ func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *head l.listenEP.mu.Lock() if l.listenEP.EndpointState() != StateListen { l.listenEP.mu.Unlock() + // Ensure we release any registrations done by the newly + // created endpoint. + ep.mu.Unlock() + ep.Close() + + // Wake up any waiters. This is strictly not required normally + // as a socket that was never accepted can't really have any + // registered waiters except when stack.Wait() is called which + // waits for all registered endpoints to stop and expects an + // EventHUp. + ep.waiterQueue.Notify(waiter.EventHUp | waiter.EventErr | waiter.EventIn | waiter.EventOut) return nil, tcpip.ErrConnectionAborted } l.addPendingEndpoint(ep) + + // Propagate any inheritable options from the listening endpoint + // to the newly created endpoint. + l.listenEP.propagateInheritableOptionsLocked(ep) + deferAccept = l.listenEP.deferAccept l.listenEP.mu.Unlock() } @@ -299,6 +317,7 @@ func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *head // Perform the 3-way handshake. h := newPassiveHandshake(ep, seqnum.Size(ep.initialReceiveWindow()), isn, irs, opts, deferAccept) if err := h.execute(); err != nil { + ep.mu.Unlock() ep.Close() // Wake up any waiters. This is strictly not required normally // as a socket that was never accepted can't really have any @@ -312,9 +331,7 @@ func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *head } return nil, err } - ep.mu.Lock() ep.isConnectNotified = true - ep.mu.Unlock() // Update the receive window scaling. We can't do it before the // handshake because it's possible that the peer doesn't support window @@ -366,12 +383,12 @@ func (e *endpoint) deliverAccepted(n *endpoint) { } } -// propagateInheritableOptions propagates any options set on the listening +// propagateInheritableOptionsLocked propagates any options set on the listening // endpoint to the newly created endpoint. -func (e *endpoint) propagateInheritableOptions(n *endpoint) { - e.mu.Lock() +// +// Precondition: e.mu and n.mu must be held. +func (e *endpoint) propagateInheritableOptionsLocked(n *endpoint) { n.userTimeout = e.userTimeout - e.mu.Unlock() } // handleSynSegment is called in its own goroutine once the listening endpoint @@ -382,7 +399,11 @@ func (e *endpoint) propagateInheritableOptions(n *endpoint) { // cookies to accept connections. func (e *endpoint) handleSynSegment(ctx *listenContext, s *segment, opts *header.TCPSynOptions) { defer decSynRcvdCount() - defer e.decSynRcvdCount() + defer func() { + e.mu.Lock() + e.decSynRcvdCount() + e.mu.Unlock() + }() defer s.decRef() n, err := ctx.createEndpointAndPerformHandshake(s, opts, &waiter.Queue{}) @@ -399,29 +420,21 @@ func (e *endpoint) handleSynSegment(ctx *listenContext, s *segment, opts *header } func (e *endpoint) incSynRcvdCount() bool { - e.mu.Lock() - if e.synRcvdCount >= cap(e.acceptedChan) { - e.mu.Unlock() + if e.synRcvdCount >= (cap(e.acceptedChan)) { return false } e.synRcvdCount++ - e.mu.Unlock() return true } func (e *endpoint) decSynRcvdCount() { - e.mu.Lock() e.synRcvdCount-- - e.mu.Unlock() } func (e *endpoint) acceptQueueIsFull() bool { - e.mu.Lock() if l, c := len(e.acceptedChan)+e.synRcvdCount, cap(e.acceptedChan); l >= c { - e.mu.Unlock() return true } - e.mu.Unlock() return false } @@ -559,6 +572,10 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { return } + // Propagate any inheritable options from the listening endpoint + // to the newly created endpoint. + e.propagateInheritableOptionsLocked(n) + // clear the tsOffset for the newly created // endpoint as the Timestamp was already // randomly offset when the original SYN-ACK was @@ -593,14 +610,12 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { func (e *endpoint) protocolListenLoop(rcvWnd seqnum.Size) *tcpip.Error { e.mu.Lock() v6only := e.v6only - e.mu.Unlock() ctx := newListenContext(e.stack, e, rcvWnd, v6only, e.NetProto) defer func() { // Mark endpoint as closed. This will prevent goroutines running // handleSynSegment() from attempting to queue new connections // to the endpoint. - e.mu.Lock() e.setEndpointState(StateClose) // close any endpoints in SYN-RCVD state. @@ -622,7 +637,10 @@ func (e *endpoint) protocolListenLoop(rcvWnd seqnum.Size) *tcpip.Error { s.AddWaker(&e.notificationWaker, wakerForNotification) s.AddWaker(&e.newSegmentWaker, wakerForNewSegment) for { - switch index, _ := s.Fetch(true); index { + e.mu.Unlock() + index, _ := s.Fetch(true) + e.mu.Lock() + switch index { case wakerForNotification: n := e.fetchNotifications() if n¬ifyClose != 0 { @@ -635,7 +653,9 @@ func (e *endpoint) protocolListenLoop(rcvWnd seqnum.Size) *tcpip.Error { s.decRef() } close(e.drainDone) + e.mu.Unlock() <-e.undrain + e.mu.Lock() } case wakerForNewSegment: diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index be86af502..edb37a549 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -61,6 +61,9 @@ const ( ) // handshake holds the state used during a TCP 3-way handshake. +// +// NOTE: handshake.ep.mu is held during handshake processing. It is released if +// we are going to block and reacquired when we start processing an event. type handshake struct { ep *endpoint state handshakeState @@ -209,9 +212,7 @@ func (h *handshake) resetToSynRcvd(iss seqnum.Value, irs seqnum.Value, opts *hea h.mss = opts.MSS h.sndWndScale = opts.WS h.deferAccept = deferAccept - h.ep.mu.Lock() h.ep.setEndpointState(StateSynRecv) - h.ep.mu.Unlock() } // checkAck checks if the ACK number, if present, of a segment received during @@ -241,9 +242,7 @@ func (h *handshake) synSentState(s *segment) *tcpip.Error { // RFC 793, page 67, states that "If the RST bit is set [and] If the ACK // was acceptable then signal the user "error: connection reset", drop // the segment, enter CLOSED state, delete TCB, and return." - h.ep.mu.Lock() h.ep.workerCleanup = true - h.ep.mu.Unlock() // Although the RFC above calls out ECONNRESET, Linux actually returns // ECONNREFUSED here so we do as well. return tcpip.ErrConnectionRefused @@ -281,9 +280,7 @@ func (h *handshake) synSentState(s *segment) *tcpip.Error { if s.flagIsSet(header.TCPFlagAck) { h.state = handshakeCompleted - h.ep.mu.Lock() h.ep.transitionToStateEstablishedLocked(h) - h.ep.mu.Unlock() h.ep.sendRaw(buffer.VectorisedView{}, header.TCPFlagAck, h.iss+1, h.ackNum, h.rcvWnd>>h.effectiveRcvWndScale()) return nil @@ -293,11 +290,9 @@ func (h *handshake) synSentState(s *segment) *tcpip.Error { // but resend our own SYN and wait for it to be acknowledged in the // SYN-RCVD state. h.state = handshakeSynRcvd - h.ep.mu.Lock() ttl := h.ep.ttl amss := h.ep.amss h.ep.setEndpointState(StateSynRecv) - h.ep.mu.Unlock() synOpts := header.TCPSynOptions{ WS: int(h.effectiveRcvWndScale()), TS: rcvSynOpts.TS, @@ -357,10 +352,6 @@ func (h *handshake) synRcvdState(s *segment) *tcpip.Error { return tcpip.ErrInvalidEndpointState } - h.ep.mu.RLock() - amss := h.ep.amss - h.ep.mu.RUnlock() - h.resetState() synOpts := header.TCPSynOptions{ WS: h.rcvWndScale, @@ -368,7 +359,7 @@ func (h *handshake) synRcvdState(s *segment) *tcpip.Error { TSVal: h.ep.timestamp(), TSEcr: h.ep.recentTimestamp(), SACKPermitted: h.ep.sackPermitted, - MSS: amss, + MSS: h.ep.amss, } h.ep.sendSynTCP(&s.route, h.ep.ID, h.ep.ttl, h.ep.sendTOS, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) return nil @@ -399,15 +390,14 @@ func (h *handshake) synRcvdState(s *segment) *tcpip.Error { } h.state = handshakeCompleted - h.ep.mu.Lock() h.ep.transitionToStateEstablishedLocked(h) + // If the segment has data then requeue it for the receiver // to process it again once main loop is started. if s.data.Size() > 0 { s.incRef() h.ep.enqueueSegment(s) } - h.ep.mu.Unlock() return nil } @@ -493,7 +483,9 @@ func (h *handshake) resolveRoute() *tcpip.Error { } if n¬ifyDrain != 0 { close(h.ep.drainDone) + h.ep.mu.Unlock() <-h.ep.undrain + h.ep.mu.Lock() } } @@ -535,7 +527,6 @@ func (h *handshake) execute() *tcpip.Error { // Send the initial SYN segment and loop until the handshake is // completed. - h.ep.mu.Lock() h.ep.amss = calculateAdvertisedMSS(h.ep.userMSS, h.ep.route) synOpts := header.TCPSynOptions{ @@ -546,7 +537,6 @@ func (h *handshake) execute() *tcpip.Error { SACKPermitted: bool(sackEnabled), MSS: h.ep.amss, } - h.ep.mu.Unlock() // Execute is also called in a listen context so we want to make sure we // only send the TS/SACK option when we received the TS/SACK in the @@ -563,7 +553,11 @@ func (h *handshake) execute() *tcpip.Error { h.ep.sendSynTCP(&h.ep.route, h.ep.ID, h.ep.ttl, h.ep.sendTOS, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) for h.state != handshakeCompleted { - switch index, _ := s.Fetch(true); index { + h.ep.mu.Unlock() + index, _ := s.Fetch(true) + h.ep.mu.Lock() + switch index { + case wakerForResend: timeOut *= 2 if timeOut > MaxRTO { @@ -600,7 +594,9 @@ func (h *handshake) execute() *tcpip.Error { } } close(h.ep.drainDone) + h.ep.mu.Unlock() <-h.ep.undrain + h.ep.mu.Lock() } case wakerForNewSegment: @@ -1016,7 +1012,6 @@ func (e *endpoint) handleReset(s *segment) (ok bool, err *tcpip.Error) { // except SYN-SENT, all reset (RST) segments are // validated by checking their SEQ-fields." So // we only process it if it's acceptable. - e.mu.Lock() switch e.EndpointState() { // In case of a RST in CLOSE-WAIT linux moves // the socket to closed state with an error set @@ -1040,11 +1035,9 @@ func (e *endpoint) handleReset(s *segment) (ok bool, err *tcpip.Error) { case StateCloseWait: e.transitionToStateCloseLocked() e.HardError = tcpip.ErrAborted - e.mu.Unlock() e.notifyProtocolGoroutine(notifyTickleWorker) return false, nil default: - e.mu.Unlock() // RFC 793, page 37 states that "in all states // except SYN-SENT, all reset (RST) segments are // validated by checking their SEQ-fields." So @@ -1157,9 +1150,7 @@ func (e *endpoint) handleSegment(s *segment) (cont bool, err *tcpip.Error) { // Now check if the received segment has caused us to transition // to a CLOSED state, if yes then terminate processing and do // not invoke the sender. - e.mu.RLock() state := e.state - e.mu.RUnlock() if state == StateClose { // When we get into StateClose while processing from the queue, // return immediately and let the protocolMainloop handle it. @@ -1182,9 +1173,7 @@ func (e *endpoint) handleSegment(s *segment) (cont bool, err *tcpip.Error) { // keepalive packets periodically when the connection is idle. If we don't hear // from the other side after a number of tries, we terminate the connection. func (e *endpoint) keepaliveTimerExpired() *tcpip.Error { - e.mu.RLock() userTimeout := e.userTimeout - e.mu.RUnlock() e.keepalive.Lock() if !e.keepalive.enabled || !e.keepalive.timer.checkExpiration() { @@ -1248,6 +1237,7 @@ func (e *endpoint) disableKeepaliveTimer() { // goroutine and is responsible for sending segments and handling received // segments. func (e *endpoint) protocolMainLoop(handshake bool, wakerInitDone chan<- struct{}) *tcpip.Error { + e.mu.Lock() var closeTimer *time.Timer var closeWaker sleep.Waker @@ -1269,7 +1259,6 @@ func (e *endpoint) protocolMainLoop(handshake bool, wakerInitDone chan<- struct{ } e.mu.Unlock() - e.workMu.Unlock() // When the protocol loop exits we should wake up our waiters. e.waiterQueue.Notify(waiter.EventHUp | waiter.EventErr | waiter.EventIn | waiter.EventOut) } @@ -1280,16 +1269,13 @@ func (e *endpoint) protocolMainLoop(handshake bool, wakerInitDone chan<- struct{ // completion. initialRcvWnd := e.initialReceiveWindow() h := newHandshake(e, seqnum.Size(initialRcvWnd)) - e.mu.Lock() h.ep.setEndpointState(StateSynSent) - e.mu.Unlock() if err := h.execute(); err != nil { e.lastErrorMu.Lock() e.lastError = err e.lastErrorMu.Unlock() - e.mu.Lock() e.setEndpointState(StateError) e.HardError = err @@ -1302,9 +1288,7 @@ func (e *endpoint) protocolMainLoop(handshake bool, wakerInitDone chan<- struct{ e.keepalive.timer.init(&e.keepalive.waker) defer e.keepalive.timer.cleanup() - e.mu.Lock() drained := e.drainDone != nil - e.mu.Unlock() if drained { close(e.drainDone) <-e.undrain @@ -1330,10 +1314,8 @@ func (e *endpoint) protocolMainLoop(handshake bool, wakerInitDone chan<- struct{ // This means the socket is being closed due // to the TCP-FIN-WAIT2 timeout was hit. Just // mark the socket as closed. - e.mu.Lock() e.transitionToStateCloseLocked() e.workerCleanup = true - e.mu.Unlock() return nil }, }, @@ -1388,7 +1370,6 @@ func (e *endpoint) protocolMainLoop(handshake bool, wakerInitDone chan<- struct{ } if n¬ifyClose != 0 && closeTimer == nil { - e.mu.Lock() if e.EndpointState() == StateFinWait2 && e.closed { // The socket has been closed and we are in FIN_WAIT2 // so start the FIN_WAIT2 timer. @@ -1397,7 +1378,6 @@ func (e *endpoint) protocolMainLoop(handshake bool, wakerInitDone chan<- struct{ }) e.waiterQueue.Notify(waiter.EventHUp | waiter.EventErr | waiter.EventIn | waiter.EventOut) } - e.mu.Unlock() } if n¬ifyKeepaliveChanged != 0 { @@ -1417,7 +1397,9 @@ func (e *endpoint) protocolMainLoop(handshake bool, wakerInitDone chan<- struct{ // Only block the worker if the endpoint // is not in closed state or error state. close(e.drainDone) + e.mu.Unlock() <-e.undrain + e.mu.Lock() } } @@ -1460,7 +1442,6 @@ func (e *endpoint) protocolMainLoop(handshake bool, wakerInitDone chan<- struct{ } e.rcvListMu.Unlock() - e.mu.Lock() if e.workerCleanup { e.notifyProtocolGoroutine(notifyClose) } @@ -1468,7 +1449,6 @@ func (e *endpoint) protocolMainLoop(handshake bool, wakerInitDone chan<- struct{ // Main loop. Handle segments until both send and receive ends of the // connection have completed. cleanupOnError := func(err *tcpip.Error) { - e.mu.Lock() e.workerCleanup = true if err != nil { e.resetConnectionLocked(err) @@ -1480,16 +1460,11 @@ func (e *endpoint) protocolMainLoop(handshake bool, wakerInitDone chan<- struct{ loop: for e.EndpointState() != StateTimeWait && e.EndpointState() != StateClose && e.EndpointState() != StateError { e.mu.Unlock() - e.workMu.Unlock() v, _ := s.Fetch(true) - e.workMu.Lock() + e.mu.Lock() - // We need to double check here because the notification maybe + // We need to double check here because the notification may be // stale by the time we got around to processing it. - // - // NOTE: since we now hold the workMu the processors cannot - // change the state of the endpoint so it's safe to proceed - // after this check. switch e.EndpointState() { case StateError: // If the endpoint has already transitioned to an ERROR @@ -1502,21 +1477,17 @@ loop: case StateTimeWait: fallthrough case StateClose: - e.mu.Lock() break loop default: if err := funcs[v].f(); err != nil { cleanupOnError(err) return nil } - e.mu.Lock() } } - state := e.EndpointState() - e.mu.Unlock() var reuseTW func() - if state == StateTimeWait { + if e.EndpointState() == StateTimeWait { // Disable close timer as we now entering real TIME_WAIT. if closeTimer != nil { closeTimer.Stop() @@ -1526,14 +1497,11 @@ loop: s.Done() // Wake up any waiters before we enter TIME_WAIT. e.waiterQueue.Notify(waiter.EventHUp | waiter.EventErr | waiter.EventIn | waiter.EventOut) - e.mu.Lock() e.workerCleanup = true - e.mu.Unlock() reuseTW = e.doTimeWait() } // Mark endpoint as closed. - e.mu.Lock() if e.EndpointState() != StateError { e.transitionToStateCloseLocked() } @@ -1649,9 +1617,9 @@ func (e *endpoint) doTimeWait() (twReuse func()) { defer timeWaitTimer.Stop() for { - e.workMu.Unlock() + e.mu.Unlock() v, _ := s.Fetch(true) - e.workMu.Lock() + e.mu.Lock() switch v { case newSegment: extendTimeWait, reuseTW := e.handleTimeWaitSegments() @@ -1674,7 +1642,9 @@ func (e *endpoint) doTimeWait() (twReuse func()) { e.handleTimeWaitSegments() } close(e.drainDone) + e.mu.Unlock() <-e.undrain + e.mu.Lock() return nil } case timeWaitDone: diff --git a/pkg/tcpip/transport/tcp/dispatcher.go b/pkg/tcpip/transport/tcp/dispatcher.go index d792b07d6..90ac956a9 100644 --- a/pkg/tcpip/transport/tcp/dispatcher.go +++ b/pkg/tcpip/transport/tcp/dispatcher.go @@ -128,7 +128,7 @@ func (p *processor) handleSegments() { continue } - if !ep.workMu.TryLock() { + if !ep.mu.TryLock() { ep.newSegmentWaker.Assert() continue } @@ -138,12 +138,10 @@ func (p *processor) handleSegments() { if err := ep.handleSegments(true /* fastPath */); err != nil || ep.EndpointState() == StateClose { // Send any active resets if required. if err != nil { - ep.mu.Lock() ep.resetConnectionLocked(err) - ep.mu.Unlock() } ep.notifyProtocolGoroutine(notifyTickleWorker) - ep.workMu.Unlock() + ep.mu.Unlock() continue } @@ -151,7 +149,7 @@ func (p *processor) handleSegments() { p.epQ.enqueue(ep) } - ep.workMu.Unlock() + ep.mu.Unlock() } } } diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 5187a5e25..eb8a9d73e 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -18,6 +18,7 @@ import ( "encoding/binary" "fmt" "math" + "runtime" "strings" "sync/atomic" "time" @@ -33,7 +34,6 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/ports" "gvisor.dev/gvisor/pkg/tcpip/seqnum" "gvisor.dev/gvisor/pkg/tcpip/stack" - "gvisor.dev/gvisor/pkg/tmutex" "gvisor.dev/gvisor/pkg/waiter" ) @@ -283,6 +283,37 @@ func (*EndpointInfo) IsEndpointInfo() {} // synchronized. The protocol implementation, however, runs in a single // goroutine. // +// Each endpoint has a few mutexes: +// +// e.mu -> Primary mutex for an endpoint must be held for all operations except +// in e.Readiness where acquiring it will result in a deadlock in epoll +// implementation. +// +// The following three mutexes can be acquired independent of e.mu but if +// acquired with e.mu then e.mu must be acquired first. +// +// e.rcvListMu -> Protects the rcvList and associated fields. +// e.sndBufMu -> Protects the sndQueue and associated fields. +// e.lastErrorMu -> Protects the lastError field. +// +// LOCKING/UNLOCKING of the endpoint. The locking of an endpoint is different +// based on the context in which the lock is acquired. In the syscall context +// e.LockUser/e.UnlockUser should be used and when doing background processing +// e.mu.Lock/e.mu.Unlock should be used. The distinction is described below +// in brief. +// +// The reason for this locking behaviour is to avoid wakeups to handle packets. +// In cases where the endpoint is already locked the background processor can +// queue the packet up and go its merry way and the lock owner will eventually +// process the backlog when releasing the lock. Similarly when acquiring the +// lock from say a syscall goroutine we can implement a bit of spinning if we +// know that the lock is not held by another syscall goroutine. Background +// processors should never hold the lock for long and we can avoid an expensive +// sleep/wakeup by spinning for a shortwhile. +// +// For more details please see the detailed documentation on +// e.LockUser/e.UnlockUser methods. +// // +stateify savable type endpoint struct { EndpointInfo @@ -299,12 +330,6 @@ type endpoint struct { // Precondition: epQueue.mu must be held to read/write this field.. pendingProcessing bool `state:"nosave"` - // workMu is used to arbitrate which goroutine may perform protocol - // work. Only the main protocol goroutine is expected to call Lock() on - // it, but other goroutines (e.g., send) may call TryLock() to eagerly - // perform work without having to wait for the main one to wake up. - workMu tmutex.Mutex `state:"nosave"` - // The following fields are initialized at creation time and do not // change throughout the lifetime of the endpoint. stack *stack.Stack `state:"manual"` @@ -330,15 +355,11 @@ type endpoint struct { rcvBufSize int rcvBufUsed int rcvAutoParams rcvBufAutoTuneParams - // zeroWindow indicates that the window was closed due to receive buffer - // space being filled up. This is set by the worker goroutine before - // moving a segment to the rcvList. This setting is cleared by the - // endpoint when a Read() call reads enough data for the new window to - // be non-zero. - zeroWindow bool - // The following fields are protected by the mutex. - mu sync.RWMutex `state:"nosave"` + // mu protects all endpoint fields unless documented otherwise. mu must + // be acquired before interacting with the endpoint fields. + mu sync.Mutex `state:"nosave"` + ownedByUser uint32 // state must be read/set using the EndpointState()/setEndpointState() methods. state EndpointState `state:".(EndpointState)"` @@ -583,14 +604,93 @@ func calculateAdvertisedMSS(userMSS uint16, r stack.Route) uint16 { return maxMSS } +// LockUser tries to lock e.mu and if it fails it will check if the lock is held +// by another syscall goroutine. If yes, then it will goto sleep waiting for the +// lock to be released, if not then it will spin till it acquires the lock or +// another syscall goroutine acquires it in which case it will goto sleep as +// described above. +// +// The assumption behind spinning here being that background packet processing +// should not be holding the lock for long and spinning reduces latency as we +// avoid an expensive sleep/wakeup of of the syscall goroutine). +func (e *endpoint) LockUser() { + for { + // Try first if the sock is locked then check if it's owned + // by another user goroutine if not then we spin, otherwise + // we just goto sleep on the Lock() and wait. + if !e.mu.TryLock() { + // If socket is owned by the user then just goto sleep + // as the lock could be held for a reasonably long time. + if atomic.LoadUint32(&e.ownedByUser) == 1 { + e.mu.Lock() + atomic.StoreUint32(&e.ownedByUser, 1) + return + } + // Spin but yield the processor since the lower half + // should yield the lock soon. + runtime.Gosched() + continue + } + atomic.StoreUint32(&e.ownedByUser, 1) + return + } +} + +// UnlockUser will check if there are any segments already queued for processing +// and process any such segments before unlocking e.mu. This is required because +// we when packets arrive and endpoint lock is already held then such packets +// are queued up to be processed. If the lock is held by the endpoint goroutine +// then it will process these packets but if the lock is instead held by the +// syscall goroutine then we can have the syscall goroutine process the backlog +// before unlocking. +// +// This avoids an unnecessary wakeup of the endpoint protocol goroutine for the +// endpoint. It's also required eventually when we get rid of the endpoint +// protocol goroutine altogether. +// +// Precondition: e.LockUser() must have been called before calling e.UnlockUser() +func (e *endpoint) UnlockUser() { + // Lock segment queue before checking so that we avoid a race where + // segments can be queued between the time we check if queue is empty + // and actually unlock the endpoint mutex. + for { + e.segmentQueue.mu.Lock() + if e.segmentQueue.emptyLocked() { + if atomic.SwapUint32(&e.ownedByUser, 0) != 1 { + panic("e.UnlockUser() called without calling e.LockUser()") + } + e.mu.Unlock() + e.segmentQueue.mu.Unlock() + return + } + e.segmentQueue.mu.Unlock() + + switch e.EndpointState() { + case StateEstablished: + if err := e.handleSegments(true /* fastPath */); err != nil { + e.notifyProtocolGoroutine(notifyTickleWorker) + } + default: + // Since we are waking the endpoint goroutine here just unlock + // and let it process the queued segments. + e.newSegmentWaker.Assert() + if atomic.SwapUint32(&e.ownedByUser, 0) != 1 { + panic("e.UnlockUser() called without calling e.LockUser()") + } + e.mu.Unlock() + return + } + } +} + // StopWork halts packet processing. Only to be used in tests. func (e *endpoint) StopWork() { - e.workMu.Lock() + e.mu.Lock() } // ResumeWork resumes packet processing. Only to be used in tests. func (e *endpoint) ResumeWork() { - e.workMu.Unlock() + e.mu.Unlock() } // setEndpointState updates the state of the endpoint to state atomically. This @@ -709,8 +809,6 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue } e.segmentQueue.setLimit(MaxUnprocessedSegments) - e.workMu.Init() - e.workMu.Lock() e.tsOffset = timeStampOffset() return e @@ -721,9 +819,6 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue func (e *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { result := waiter.EventMask(0) - e.mu.RLock() - defer e.mu.RUnlock() - switch e.EndpointState() { case StateInitial, StateBound, StateConnecting, StateSynSent, StateSynRecv: // Ready for nothing. @@ -823,20 +918,22 @@ func (e *endpoint) Abort() { // with it. It must be called only once and with no other concurrent calls to // the endpoint. func (e *endpoint) Close() { - e.mu.Lock() - closed := e.closed - e.closed = true - e.mu.Unlock() - if closed { + e.LockUser() + defer e.UnlockUser() + if e.closed { return } // Issue a shutdown so that the peer knows we won't send any more data // if we're connected, or stop accepting if we're listening. - e.Shutdown(tcpip.ShutdownWrite | tcpip.ShutdownRead) - - e.mu.Lock() + e.shutdownLocked(tcpip.ShutdownWrite | tcpip.ShutdownRead) + e.closeNoShutdownLocked() +} +// closeNoShutdown closes the endpoint without doing a full shutdown. This is +// used when a connection needs to be aborted with a RST and we want to skip +// a full 4 way TCP shutdown. +func (e *endpoint) closeNoShutdownLocked() { // For listening sockets, we always release ports inline so that they // are immediately available for reuse after Close() is called. If also // registered, we unregister as well otherwise the next user would fail @@ -853,6 +950,8 @@ func (e *endpoint) Close() { e.boundPortFlags = ports.Flags{} } + // Mark endpoint as closed. + e.closed = true // Either perform the local cleanup or kick the worker to make sure it // knows it needs to cleanup. switch e.EndpointState() { @@ -873,8 +972,6 @@ func (e *endpoint) Close() { // goroutine terminates. e.notifyProtocolGoroutine(notifyClose) } - - e.mu.Unlock() } // closePendingAcceptableConnections closes all connections that have completed @@ -909,7 +1006,6 @@ func (e *endpoint) closePendingAcceptableConnectionsLocked() { // after Close() is called and the worker goroutine (if any) is done with its // work. func (e *endpoint) cleanupLocked() { - // Close all endpoints that might have been accepted by TCP but not by // the client. if e.acceptedChan != nil { @@ -954,18 +1050,18 @@ func (e *endpoint) initialReceiveWindow() int { // ModerateRecvBuf adjusts the receive buffer and the advertised window // based on the number of bytes copied to user space. func (e *endpoint) ModerateRecvBuf(copied int) { - e.mu.RLock() + e.LockUser() + defer e.UnlockUser() + e.rcvListMu.Lock() if e.rcvAutoParams.disabled { e.rcvListMu.Unlock() - e.mu.RUnlock() return } now := time.Now() if rtt := e.rcvAutoParams.rtt; rtt == 0 || now.Sub(e.rcvAutoParams.measureTime) < rtt { e.rcvAutoParams.copied += copied e.rcvListMu.Unlock() - e.mu.RUnlock() return } prevRTTCopied := e.rcvAutoParams.copied + copied @@ -1021,7 +1117,6 @@ func (e *endpoint) ModerateRecvBuf(copied int) { e.rcvAutoParams.measureTime = now e.rcvAutoParams.copied = 0 e.rcvListMu.Unlock() - e.mu.RUnlock() } // IPTables implements tcpip.Endpoint.IPTables. @@ -1031,7 +1126,7 @@ func (e *endpoint) IPTables() (iptables.IPTables, error) { // Read reads data from the endpoint. func (e *endpoint) Read(*tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) { - e.mu.RLock() + e.LockUser() // The endpoint can be read if it's connected, or if it's already closed // but has some pending unread data. Also note that a RST being received // would cause the state to become StateError so we should allow the @@ -1041,7 +1136,7 @@ func (e *endpoint) Read(*tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, if s := e.EndpointState(); !s.connected() && s != StateClose && bufUsed == 0 { e.rcvListMu.Unlock() he := e.HardError - e.mu.RUnlock() + e.UnlockUser() if s == StateError { return buffer.View{}, tcpip.ControlMessages{}, he } @@ -1051,7 +1146,7 @@ func (e *endpoint) Read(*tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, v, err := e.readLocked() e.rcvListMu.Unlock() - e.mu.RUnlock() + e.UnlockUser() if err == tcpip.ErrClosedForReceive { e.stats.ReadErrors.ReadClosed.Increment() @@ -1124,13 +1219,13 @@ func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c // (without the MSG_FASTOPEN flag). Corking is unimplemented, so opts.More // and opts.EndOfRecord are also ignored. - e.mu.RLock() + e.LockUser() e.sndBufMu.Lock() avail, err := e.isEndpointWritableLocked() if err != nil { e.sndBufMu.Unlock() - e.mu.RUnlock() + e.UnlockUser() e.stats.WriteErrors.WriteClosed.Increment() return 0, nil, err } @@ -1142,113 +1237,68 @@ func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c // are copying data in. if !opts.Atomic { e.sndBufMu.Unlock() - e.mu.RUnlock() + e.UnlockUser() } // Fetch data. v, perr := p.Payload(avail) if perr != nil || len(v) == 0 { - if opts.Atomic { // See above. + // Note that perr may be nil if len(v) == 0. + if opts.Atomic { e.sndBufMu.Unlock() - e.mu.RUnlock() + e.UnlockUser() } - // Note that perr may be nil if len(v) == 0. return 0, nil, perr } - if opts.Atomic { + queueAndSend := func() (int64, <-chan struct{}, *tcpip.Error) { // Add data to the send queue. s := newSegmentFromView(&e.route, e.ID, v) e.sndBufUsed += len(v) e.sndBufInQueue += seqnum.Size(len(v)) e.sndQueue.PushBack(s) e.sndBufMu.Unlock() - // Release the endpoint lock to prevent deadlocks due to lock - // order inversion when acquiring workMu. - e.mu.RUnlock() - } - if e.workMu.TryLock() { - // Since we released locks in between it's possible that the - // endpoint transitioned to a CLOSED/ERROR states so make - // sure endpoint is still writable before trying to write. - if !opts.Atomic { // See above. - e.mu.RLock() - e.sndBufMu.Lock() - - // Because we released the lock before copying, check state again - // to make sure the endpoint is still in a valid state for a write. - avail, err = e.isEndpointWritableLocked() - if err != nil { - e.sndBufMu.Unlock() - e.mu.RUnlock() - e.stats.WriteErrors.WriteClosed.Increment() - return 0, nil, err - } - - // Discard any excess data copied in due to avail being reduced due - // to a simultaneous write call to the socket. - if avail < len(v) { - v = v[:avail] - } - // Add data to the send queue. - s := newSegmentFromView(&e.route, e.ID, v) - e.sndBufUsed += len(v) - e.sndBufInQueue += seqnum.Size(len(v)) - e.sndQueue.PushBack(s) - e.sndBufMu.Unlock() - // Release the endpoint lock to prevent deadlocks due to lock - // order inversion when acquiring workMu. - e.mu.RUnlock() - - } // Do the work inline. e.handleWrite() - e.workMu.Unlock() - } else { - if !opts.Atomic { // See above. - e.mu.RLock() - e.sndBufMu.Lock() + e.UnlockUser() + return int64(len(v)), nil, nil + } - // Because we released the lock before copying, check state again - // to make sure the endpoint is still in a valid state for a write. - avail, err = e.isEndpointWritableLocked() - if err != nil { - e.sndBufMu.Unlock() - e.mu.RUnlock() - e.stats.WriteErrors.WriteClosed.Increment() - return 0, nil, err - } + if opts.Atomic { + // Locks released in queueAndSend() + return queueAndSend() + } - // Discard any excess data copied in due to avail being reduced due - // to a simultaneous write call to the socket. - if avail < len(v) { - v = v[:avail] - } - // Add data to the send queue. - s := newSegmentFromView(&e.route, e.ID, v) - e.sndBufUsed += len(v) - e.sndBufInQueue += seqnum.Size(len(v)) - e.sndQueue.PushBack(s) - e.sndBufMu.Unlock() - // Release the endpoint lock to prevent deadlocks due to lock - // order inversion when acquiring workMu. - e.mu.RUnlock() + // Since we released locks in between it's possible that the + // endpoint transitioned to a CLOSED/ERROR states so make + // sure endpoint is still writable before trying to write. + e.LockUser() + e.sndBufMu.Lock() + avail, err = e.isEndpointWritableLocked() + if err != nil { + e.sndBufMu.Unlock() + e.UnlockUser() + e.stats.WriteErrors.WriteClosed.Increment() + return 0, nil, err + } - } - // Let the protocol goroutine do the work. - e.sndWaker.Assert() + // Discard any excess data copied in due to avail being reduced due + // to a simultaneous write call to the socket. + if avail < len(v) { + v = v[:avail] } - return int64(len(v)), nil, nil + // Locks released in queueAndSend() + return queueAndSend() } // Peek reads data without consuming it from the endpoint. // // This method does not block if there is no data pending. func (e *endpoint) Peek(vec [][]byte) (int64, tcpip.ControlMessages, *tcpip.Error) { - e.mu.RLock() - defer e.mu.RUnlock() + e.LockUser() + defer e.UnlockUser() // The endpoint can be read if it's connected, or if it's already closed // but has some pending unread data. @@ -1339,6 +1389,9 @@ func (e *endpoint) windowCrossedACKThresholdLocked(deltaBefore int) (crossed boo // SetSockOptBool sets a socket option. func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { + e.LockUser() + defer e.UnlockUser() + switch opt { case tcpip.V6OnlyOption: // We only recognize this option on v6 endpoints. @@ -1346,9 +1399,6 @@ func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { return tcpip.ErrInvalidEndpointState } - e.mu.Lock() - defer e.mu.Unlock() - // We only allow this to be set when we're in the initial state. if e.EndpointState() != StateInitial { return tcpip.ErrInvalidEndpointState @@ -1379,7 +1429,7 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { mask := uint32(notifyReceiveWindowChanged) - e.mu.RLock() + e.LockUser() e.rcvListMu.Lock() // Make sure the receive buffer size allows us to send a @@ -1409,8 +1459,9 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { if crossed, above := e.windowCrossedACKThresholdLocked(availAfter - availBefore); crossed && above { mask |= notifyNonZeroReceiveWindow } + e.rcvListMu.Unlock() - e.mu.RUnlock() + e.UnlockUser() e.notifyProtocolGoroutine(mask) return nil @@ -1466,15 +1517,15 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { return nil case tcpip.ReuseAddressOption: - e.mu.Lock() + e.LockUser() e.reuseAddr = v != 0 - e.mu.Unlock() + e.UnlockUser() return nil case tcpip.ReusePortOption: - e.mu.Lock() + e.LockUser() e.reusePort = v != 0 - e.mu.Unlock() + e.UnlockUser() return nil case tcpip.BindToDeviceOption: @@ -1482,9 +1533,9 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { if id != 0 && !e.stack.HasNIC(id) { return tcpip.ErrUnknownDevice } - e.mu.Lock() + e.LockUser() e.bindToDevice = id - e.mu.Unlock() + e.UnlockUser() return nil case tcpip.QuickAckOption: @@ -1500,16 +1551,16 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { if userMSS < header.TCPMinimumMSS || userMSS > header.TCPMaximumMSS { return tcpip.ErrInvalidOptionValue } - e.mu.Lock() + e.LockUser() e.userMSS = uint16(userMSS) - e.mu.Unlock() + e.UnlockUser() e.notifyProtocolGoroutine(notifyMSSChanged) return nil case tcpip.TTLOption: - e.mu.Lock() + e.LockUser() e.ttl = uint8(v) - e.mu.Unlock() + e.UnlockUser() return nil case tcpip.KeepaliveEnabledOption: @@ -1541,15 +1592,15 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { return nil case tcpip.TCPUserTimeoutOption: - e.mu.Lock() + e.LockUser() e.userTimeout = time.Duration(v) - e.mu.Unlock() + e.UnlockUser() return nil case tcpip.BroadcastOption: - e.mu.Lock() + e.LockUser() e.broadcast = v != 0 - e.mu.Unlock() + e.UnlockUser() return nil case tcpip.CongestionControlOption: @@ -1563,22 +1614,16 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { availCC := strings.Split(string(avail), " ") for _, cc := range availCC { if v == tcpip.CongestionControlOption(cc) { - // Acquire the work mutex as we may need to - // reinitialize the congestion control state. - e.mu.Lock() + e.LockUser() state := e.EndpointState() e.cc = v - e.mu.Unlock() switch state { case StateEstablished: - e.workMu.Lock() - e.mu.Lock() if e.EndpointState() == state { e.snd.cc = e.snd.initCongestionControl(e.cc) } - e.mu.Unlock() - e.workMu.Unlock() } + e.UnlockUser() return nil } } @@ -1588,23 +1633,23 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { return tcpip.ErrNoSuchFile case tcpip.IPv4TOSOption: - e.mu.Lock() + e.LockUser() // TODO(gvisor.dev/issue/995): ECN is not currently supported, // ignore the bits for now. e.sendTOS = uint8(v) & ^uint8(inetECNMask) - e.mu.Unlock() + e.UnlockUser() return nil case tcpip.IPv6TrafficClassOption: - e.mu.Lock() + e.LockUser() // TODO(gvisor.dev/issue/995): ECN is not currently supported, // ignore the bits for now. e.sendTOS = uint8(v) & ^uint8(inetECNMask) - e.mu.Unlock() + e.UnlockUser() return nil case tcpip.TCPLingerTimeoutOption: - e.mu.Lock() + e.LockUser() if v < 0 { // Same as effectively disabling TCPLinger timeout. v = 0 @@ -1622,16 +1667,16 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { v = stkTCPLingerTimeout } e.tcpLingerTimeout = time.Duration(v) - e.mu.Unlock() + e.UnlockUser() return nil case tcpip.TCPDeferAcceptOption: - e.mu.Lock() + e.LockUser() if time.Duration(v) > MaxRTO { v = tcpip.TCPDeferAcceptOption(MaxRTO) } e.deferAccept = time.Duration(v) - e.mu.Unlock() + e.UnlockUser() return nil default: @@ -1641,8 +1686,8 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { // readyReceiveSize returns the number of bytes ready to be received. func (e *endpoint) readyReceiveSize() (int, *tcpip.Error) { - e.mu.RLock() - defer e.mu.RUnlock() + e.LockUser() + defer e.UnlockUser() // The endpoint cannot be in listen state. if e.EndpointState() == StateListen { @@ -1664,9 +1709,9 @@ func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { return false, tcpip.ErrUnknownProtocolOption } - e.mu.Lock() + e.LockUser() v := e.v6only - e.mu.Unlock() + e.UnlockUser() return v, nil } @@ -1730,9 +1775,9 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { return nil case *tcpip.ReuseAddressOption: - e.mu.RLock() + e.LockUser() v := e.reuseAddr - e.mu.RUnlock() + e.UnlockUser() *o = 0 if v { @@ -1741,9 +1786,9 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { return nil case *tcpip.ReusePortOption: - e.mu.RLock() + e.LockUser() v := e.reusePort - e.mu.RUnlock() + e.UnlockUser() *o = 0 if v { @@ -1752,9 +1797,9 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { return nil case *tcpip.BindToDeviceOption: - e.mu.RLock() + e.LockUser() *o = tcpip.BindToDeviceOption(e.bindToDevice) - e.mu.RUnlock() + e.UnlockUser() return nil case *tcpip.QuickAckOption: @@ -1765,16 +1810,16 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { return nil case *tcpip.TTLOption: - e.mu.Lock() + e.LockUser() *o = tcpip.TTLOption(e.ttl) - e.mu.Unlock() + e.UnlockUser() return nil case *tcpip.TCPInfoOption: *o = tcpip.TCPInfoOption{} - e.mu.RLock() + e.LockUser() snd := e.snd - e.mu.RUnlock() + e.UnlockUser() if snd != nil { snd.rtt.Lock() o.RTT = snd.rtt.srtt @@ -1813,9 +1858,9 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { return nil case *tcpip.TCPUserTimeoutOption: - e.mu.Lock() + e.LockUser() *o = tcpip.TCPUserTimeoutOption(e.userTimeout) - e.mu.Unlock() + e.UnlockUser() return nil case *tcpip.OutOfBandInlineOption: @@ -1824,9 +1869,9 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { return nil case *tcpip.BroadcastOption: - e.mu.Lock() + e.LockUser() v := e.broadcast - e.mu.Unlock() + e.UnlockUser() *o = 0 if v { @@ -1835,33 +1880,33 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { return nil case *tcpip.CongestionControlOption: - e.mu.Lock() + e.LockUser() *o = e.cc - e.mu.Unlock() + e.UnlockUser() return nil case *tcpip.IPv4TOSOption: - e.mu.RLock() + e.LockUser() *o = tcpip.IPv4TOSOption(e.sendTOS) - e.mu.RUnlock() + e.UnlockUser() return nil case *tcpip.IPv6TrafficClassOption: - e.mu.RLock() + e.LockUser() *o = tcpip.IPv6TrafficClassOption(e.sendTOS) - e.mu.RUnlock() + e.UnlockUser() return nil case *tcpip.TCPLingerTimeoutOption: - e.mu.Lock() + e.LockUser() *o = tcpip.TCPLingerTimeoutOption(e.tcpLingerTimeout) - e.mu.Unlock() + e.UnlockUser() return nil case *tcpip.TCPDeferAcceptOption: - e.mu.Lock() + e.LockUser() *o = tcpip.TCPDeferAcceptOption(e.deferAccept) - e.mu.Unlock() + e.UnlockUser() return nil default: @@ -1901,8 +1946,8 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { // yet accepted by the app, they are restored without running the main goroutine // here. func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) *tcpip.Error { - e.mu.Lock() - defer e.mu.Unlock() + e.LockUser() + defer e.UnlockUser() connectingAddr := addr.Addr @@ -2071,9 +2116,13 @@ func (*endpoint) ConnectEndpoint(tcpip.Endpoint) *tcpip.Error { // Shutdown closes the read and/or write end of the endpoint connection to its // peer. func (e *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error { - e.mu.Lock() + e.LockUser() + defer e.UnlockUser() + return e.shutdownLocked(flags) +} + +func (e *endpoint) shutdownLocked(flags tcpip.ShutdownFlags) *tcpip.Error { e.shutdownFlags |= flags - finQueued := false switch { case e.EndpointState().connected(): // Close for read. @@ -2087,24 +2136,9 @@ func (e *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error { // If we're fully closed and we have unread data we need to abort // the connection with a RST. if (e.shutdownFlags&tcpip.ShutdownWrite) != 0 && rcvBufUsed > 0 { - e.mu.Unlock() - // Try to send an active reset immediately if the - // work mutex is available. - if e.workMu.TryLock() { - e.mu.Lock() - // We need to double check here to make - // sure worker has not transitioned the - // endpoint out of a connected state - // before trying to send a reset. - if e.EndpointState().connected() { - e.resetConnectionLocked(tcpip.ErrConnectionAborted) - e.notifyProtocolGoroutine(notifyTickleWorker) - } - e.mu.Unlock() - e.workMu.Unlock() - } else { - e.notifyProtocolGoroutine(notifyReset) - } + e.resetConnectionLocked(tcpip.ErrConnectionAborted) + // Wake up worker to terminate loop. + e.notifyProtocolGoroutine(notifyTickleWorker) return nil } } @@ -2116,42 +2150,32 @@ func (e *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error { // Already closed. e.sndBufMu.Unlock() if e.EndpointState() == StateTimeWait { - e.mu.Unlock() return tcpip.ErrNotConnected } - break + return nil } // Queue fin segment. s := newSegmentFromView(&e.route, e.ID, nil) e.sndQueue.PushBack(s) e.sndBufInQueue++ - finQueued = true // Mark endpoint as closed. e.sndClosed = true e.sndBufMu.Unlock() + e.handleClose() } + return nil case e.EndpointState() == StateListen: // Tell protocolListenLoop to stop. if flags&tcpip.ShutdownRead != 0 { e.notifyProtocolGoroutine(notifyClose) } + return nil + default: - e.mu.Unlock() return tcpip.ErrNotConnected } - e.mu.Unlock() - if finQueued { - if e.workMu.TryLock() { - e.handleClose() - e.workMu.Unlock() - } else { - // Tell protocol goroutine to close. - e.sndCloseWaker.Assert() - } - } - return nil } // Listen puts the endpoint in "listen" mode, which allows it to accept @@ -2166,8 +2190,8 @@ func (e *endpoint) Listen(backlog int) *tcpip.Error { } func (e *endpoint) listen(backlog int) *tcpip.Error { - e.mu.Lock() - defer e.mu.Unlock() + e.LockUser() + defer e.UnlockUser() // Allow the backlog to be adjusted if the endpoint is not shutting down. // When the endpoint shuts down, it sets workerCleanup to true, and from @@ -2229,7 +2253,6 @@ func (e *endpoint) listen(backlog int) *tcpip.Error { // startAcceptedLoop sets up required state and starts a goroutine with the // main loop for accepted connections. func (e *endpoint) startAcceptedLoop() { - e.mu.Lock() e.workerRunning = true e.mu.Unlock() wakerInitDone := make(chan struct{}) @@ -2240,8 +2263,8 @@ func (e *endpoint) startAcceptedLoop() { // Accept returns a new endpoint if a peer has established a connection // to an endpoint previously set to listen mode. func (e *endpoint) Accept() (tcpip.Endpoint, *waiter.Queue, *tcpip.Error) { - e.mu.RLock() - defer e.mu.RUnlock() + e.LockUser() + defer e.UnlockUser() // Endpoint must be in listen state before it can accept connections. if e.EndpointState() != StateListen { @@ -2260,8 +2283,8 @@ func (e *endpoint) Accept() (tcpip.Endpoint, *waiter.Queue, *tcpip.Error) { // Bind binds the endpoint to a specific local port and optionally address. func (e *endpoint) Bind(addr tcpip.FullAddress) (err *tcpip.Error) { - e.mu.Lock() - defer e.mu.Unlock() + e.LockUser() + defer e.UnlockUser() return e.bindLocked(addr) } @@ -2339,8 +2362,8 @@ func (e *endpoint) bindLocked(addr tcpip.FullAddress) (err *tcpip.Error) { // GetLocalAddress returns the address to which the endpoint is bound. func (e *endpoint) GetLocalAddress() (tcpip.FullAddress, *tcpip.Error) { - e.mu.RLock() - defer e.mu.RUnlock() + e.LockUser() + defer e.UnlockUser() return tcpip.FullAddress{ Addr: e.ID.LocalAddress, @@ -2351,8 +2374,8 @@ func (e *endpoint) GetLocalAddress() (tcpip.FullAddress, *tcpip.Error) { // GetRemoteAddress returns the address to which the endpoint is connected. func (e *endpoint) GetRemoteAddress() (tcpip.FullAddress, *tcpip.Error) { - e.mu.RLock() - defer e.mu.RUnlock() + e.LockUser() + defer e.UnlockUser() if !e.EndpointState().connected() { return tcpip.FullAddress{}, tcpip.ErrNotConnected @@ -2419,7 +2442,6 @@ func (e *endpoint) updateSndBufferUsage(v int) { // to be read, or when the connection is closed for receiving (in which case // s will be nil). func (e *endpoint) readyToRead(s *segment) { - e.mu.RLock() e.rcvListMu.Lock() if s != nil { s.incRef() @@ -2434,7 +2456,6 @@ func (e *endpoint) readyToRead(s *segment) { e.rcvClosed = true } e.rcvListMu.Unlock() - e.mu.RUnlock() e.waiterQueue.Notify(waiter.EventIn) } @@ -2578,9 +2599,7 @@ func (e *endpoint) completeState() stack.TCPEndpointState { s.SegTime = time.Now() // Copy EndpointID. - e.mu.Lock() s.ID = stack.TCPEndpointID(e.ID) - e.mu.Unlock() // Copy endpoint rcv state. e.rcvListMu.Lock() @@ -2710,10 +2729,10 @@ func (e *endpoint) State() uint32 { // Info returns a copy of the endpoint info. func (e *endpoint) Info() tcpip.EndpointInfo { - e.mu.RLock() + e.LockUser() // Make a copy of the endpoint info. ret := e.EndpointInfo - e.mu.RUnlock() + e.UnlockUser() return &ret } @@ -2728,9 +2747,9 @@ func (e *endpoint) Wait() { e.waiterQueue.EventRegister(&waitEntry, waiter.EventHUp) defer e.waiterQueue.EventUnregister(&waitEntry) for { - e.mu.Lock() + e.LockUser() running := e.workerRunning - e.mu.Unlock() + e.UnlockUser() if !running { break } diff --git a/pkg/tcpip/transport/tcp/endpoint_state.go b/pkg/tcpip/transport/tcp/endpoint_state.go index 4a46f0ec5..9175de441 100644 --- a/pkg/tcpip/transport/tcp/endpoint_state.go +++ b/pkg/tcpip/transport/tcp/endpoint_state.go @@ -162,8 +162,8 @@ func (e *endpoint) loadState(state EndpointState) { connectingLoading.Add(1) } // Directly update the state here rather than using e.setEndpointState - // as the endpoint is still being loaded and the stack reference to increment - // metrics is not yet initialized. + // as the endpoint is still being loaded and the stack reference is not + // yet initialized. atomic.StoreUint32((*uint32)(&e.state), uint32(state)) } @@ -180,7 +180,6 @@ func (e *endpoint) afterLoad() { func (e *endpoint) Resume(s *stack.Stack) { e.stack = s e.segmentQueue.setLimit(MaxUnprocessedSegments) - e.workMu.Init() state := e.origEndpointState switch state { case StateInitial, StateBound, StateListen, StateConnecting, StateEstablished: diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go index 73098d904..b0f918bb4 100644 --- a/pkg/tcpip/transport/tcp/protocol.go +++ b/pkg/tcpip/transport/tcp/protocol.go @@ -95,7 +95,7 @@ const ( ) type protocol struct { - mu sync.Mutex + mu sync.RWMutex sackEnabled bool delayEnabled bool sendBufferSize SendBufferSizeOption @@ -273,57 +273,57 @@ func (p *protocol) SetOption(option interface{}) *tcpip.Error { func (p *protocol) Option(option interface{}) *tcpip.Error { switch v := option.(type) { case *SACKEnabled: - p.mu.Lock() + p.mu.RLock() *v = SACKEnabled(p.sackEnabled) - p.mu.Unlock() + p.mu.RUnlock() return nil case *DelayEnabled: - p.mu.Lock() + p.mu.RLock() *v = DelayEnabled(p.delayEnabled) - p.mu.Unlock() + p.mu.RUnlock() return nil case *SendBufferSizeOption: - p.mu.Lock() + p.mu.RLock() *v = p.sendBufferSize - p.mu.Unlock() + p.mu.RUnlock() return nil case *ReceiveBufferSizeOption: - p.mu.Lock() + p.mu.RLock() *v = p.recvBufferSize - p.mu.Unlock() + p.mu.RUnlock() return nil case *tcpip.CongestionControlOption: - p.mu.Lock() + p.mu.RLock() *v = tcpip.CongestionControlOption(p.congestionControl) - p.mu.Unlock() + p.mu.RUnlock() return nil case *tcpip.AvailableCongestionControlOption: - p.mu.Lock() + p.mu.RLock() *v = tcpip.AvailableCongestionControlOption(strings.Join(p.availableCongestionControl, " ")) - p.mu.Unlock() + p.mu.RUnlock() return nil case *tcpip.ModerateReceiveBufferOption: - p.mu.Lock() + p.mu.RLock() *v = tcpip.ModerateReceiveBufferOption(p.moderateReceiveBuffer) - p.mu.Unlock() + p.mu.RUnlock() return nil case *tcpip.TCPLingerTimeoutOption: - p.mu.Lock() + p.mu.RLock() *v = tcpip.TCPLingerTimeoutOption(p.tcpLingerTimeout) - p.mu.Unlock() + p.mu.RUnlock() return nil case *tcpip.TCPTimeWaitTimeoutOption: - p.mu.Lock() + p.mu.RLock() *v = tcpip.TCPTimeWaitTimeoutOption(p.tcpTimeWaitTimeout) - p.mu.Unlock() + p.mu.RUnlock() return nil default: diff --git a/pkg/tcpip/transport/tcp/rcv.go b/pkg/tcpip/transport/tcp/rcv.go index d80aff1b6..caf8977b3 100644 --- a/pkg/tcpip/transport/tcp/rcv.go +++ b/pkg/tcpip/transport/tcp/rcv.go @@ -168,7 +168,6 @@ func (r *receiver) consumeSegment(s *segment, segSeq seqnum.Value, segLen seqnum // We just received a FIN, our next state depends on whether we sent a // FIN already or not. - r.ep.mu.Lock() switch r.ep.EndpointState() { case StateEstablished: r.ep.setEndpointState(StateCloseWait) @@ -183,7 +182,6 @@ func (r *receiver) consumeSegment(s *segment, segSeq seqnum.Value, segLen seqnum case StateFinWait2: r.ep.setEndpointState(StateTimeWait) } - r.ep.mu.Unlock() // Flush out any pending segments, except the very first one if // it happens to be the one we're handling now because the @@ -208,7 +206,6 @@ func (r *receiver) consumeSegment(s *segment, segSeq seqnum.Value, segLen seqnum // Handle ACK (not FIN-ACK, which we handled above) during one of the // shutdown states. if s.flagIsSet(header.TCPFlagAck) && s.ackNumber == r.ep.snd.sndNxt { - r.ep.mu.Lock() switch r.ep.EndpointState() { case StateFinWait1: r.ep.setEndpointState(StateFinWait2) @@ -222,7 +219,6 @@ func (r *receiver) consumeSegment(s *segment, segSeq seqnum.Value, segLen seqnum case StateLastAck: r.ep.transitionToStateCloseLocked() } - r.ep.mu.Unlock() } return true @@ -336,10 +332,8 @@ func (r *receiver) handleRcvdSegmentClosing(s *segment, state EndpointState, clo // handleRcvdSegment handles TCP segments directed at the connection managed by // r as they arrive. It is called by the protocol main loop. func (r *receiver) handleRcvdSegment(s *segment) (drop bool, err *tcpip.Error) { - r.ep.mu.RLock() state := r.ep.EndpointState() closed := r.ep.closed - r.ep.mu.RUnlock() if state != StateEstablished { drop, err := r.handleRcvdSegmentClosing(s, state, closed) diff --git a/pkg/tcpip/transport/tcp/segment_queue.go b/pkg/tcpip/transport/tcp/segment_queue.go index bd20a7ee9..48a257137 100644 --- a/pkg/tcpip/transport/tcp/segment_queue.go +++ b/pkg/tcpip/transport/tcp/segment_queue.go @@ -28,10 +28,16 @@ type segmentQueue struct { used int } +// emptyLocked determines if the queue is empty. +// Preconditions: q.mu must be held. +func (q *segmentQueue) emptyLocked() bool { + return q.used == 0 +} + // empty determines if the queue is empty. func (q *segmentQueue) empty() bool { q.mu.Lock() - r := q.used == 0 + r := q.emptyLocked() q.mu.Unlock() return r diff --git a/pkg/tcpip/transport/tcp/snd.go b/pkg/tcpip/transport/tcp/snd.go index 657c3146e..17fed4ec5 100644 --- a/pkg/tcpip/transport/tcp/snd.go +++ b/pkg/tcpip/transport/tcp/snd.go @@ -455,9 +455,7 @@ func (s *sender) retransmitTimerExpired() bool { // Give up if we've waited more than a minute since the last resend or // if a user time out is set and we have exceeded the user specified // timeout since the first retransmission. - s.ep.mu.RLock() uto := s.ep.userTimeout - s.ep.mu.RUnlock() if s.firstRetransmittedSegXmitTime.IsZero() { // We store the original xmitTime of the segment that we are @@ -713,7 +711,6 @@ func (s *sender) maybeSendSegment(seg *segment, limit int, end seqnum.Value) (se default: s.ep.setEndpointState(StateFinWait1) } - } else { // We're sending a non-FIN segment. if seg.flags&header.TCPFlagFin != 0 { diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index 5b2b16afa..39d36d2ba 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -2236,9 +2236,17 @@ func TestSegmentMerging(t *testing.T) { c.CreateConnected(789, 30000, -1 /* epRcvBuf */) - // Prevent the endpoint from processing packets. - test.stop(c.EP) + // Send 10 1 byte segments to fill up InitialWindow but don't + // ACK. That should prevent anymore packets from going out. + for i := 0; i < 10; i++ { + view := buffer.NewViewFromBytes([]byte{0}) + if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { + t.Fatalf("Write #%d failed: %v", i+1, err) + } + } + // Now send the segments that should get merged as the congestion + // window is full and we won't be able to send any more packets. var allData []byte for i, data := range [][]byte{{1, 2, 3, 4}, {5, 6, 7}, {8, 9}, {10}, {11}} { allData = append(allData, data...) @@ -2248,8 +2256,29 @@ func TestSegmentMerging(t *testing.T) { } } - // Let the endpoint process the segments that we just sent. - test.resume(c.EP) + // Check that we get 10 packets of 1 byte each. + for i := 0; i < 10; i++ { + b := c.GetPacket() + checker.IPv4(t, b, + checker.PayloadLen(header.TCPMinimumSize+1), + checker.TCP( + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS)+uint32(i)+1), + checker.AckNum(790), + checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), + ), + ) + } + + // Acknowledge the data. + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: c.Port, + Flags: header.TCPFlagAck, + SeqNum: 790, + AckNum: c.IRS.Add(1 + 10), // 10 for the 10 bytes of payload. + RcvWnd: 30000, + }) // Check that data is received. b := c.GetPacket() @@ -2257,7 +2286,7 @@ func TestSegmentMerging(t *testing.T) { checker.PayloadLen(len(allData)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), + checker.SeqNum(uint32(c.IRS)+11), checker.AckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), @@ -2273,7 +2302,7 @@ func TestSegmentMerging(t *testing.T) { DstPort: c.Port, Flags: header.TCPFlagAck, SeqNum: 790, - AckNum: c.IRS.Add(1 + seqnum.Size(len(allData))), + AckNum: c.IRS.Add(11 + seqnum.Size(len(allData))), RcvWnd: 30000, }) }) -- cgit v1.2.3 From 3a37f6791745a26d38b69fbe9d4090f8fd0c7827 Mon Sep 17 00:00:00 2001 From: Bhasker Hariharan Date: Thu, 19 Mar 2020 09:59:21 -0700 Subject: Change SocketOperations.readMu to an RWMutex. Also get rid of the readViewHasData as it's not required anymore. Updates #231, #357 PiperOrigin-RevId: 301837227 --- pkg/sentry/socket/netstack/netstack.go | 22 ++++------------------ 1 file changed, 4 insertions(+), 18 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index a2e1da02f..a6ef7a47e 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -29,7 +29,6 @@ import ( "io" "math" "reflect" - "sync/atomic" "syscall" "time" @@ -265,14 +264,8 @@ type SocketOperations struct { skType linux.SockType protocol int - // readViewHasData is 1 iff readView has data to be read, 0 otherwise. - // Must be accessed using atomic operations. It must only be written - // with readMu held but can be read without holding readMu. The latter - // is required to avoid deadlocks in epoll Readiness checks. - readViewHasData uint32 - // readMu protects access to the below fields. - readMu sync.Mutex `state:"nosave"` + readMu sync.RWMutex `state:"nosave"` // readView contains the remaining payload from the last packet. readView buffer.View // readCM holds control message information for the last packet read @@ -428,13 +421,11 @@ func (s *SocketOperations) fetchReadView() *syserr.Error { v, cms, err := s.Endpoint.Read(&s.sender) if err != nil { - atomic.StoreUint32(&s.readViewHasData, 0) return syserr.TranslateNetstackError(err) } s.readView = v s.readCM = cms - atomic.StoreUint32(&s.readViewHasData, 1) return nil } @@ -633,9 +624,11 @@ func (s *SocketOperations) Readiness(mask waiter.EventMask) waiter.EventMask { // Check our cached value iff the caller asked for readability and the // endpoint itself is currently not readable. if (mask & ^r & waiter.EventIn) != 0 { - if atomic.LoadUint32(&s.readViewHasData) == 1 { + s.readMu.RLock() + if len(s.readView) > 0 { r |= waiter.EventIn } + s.readMu.RUnlock() } return r @@ -2342,9 +2335,6 @@ func (s *SocketOperations) coalescingRead(ctx context.Context, dst usermem.IOSeq } copied += n s.readView.TrimFront(n) - if len(s.readView) == 0 { - atomic.StoreUint32(&s.readViewHasData, 0) - } dst = dst.DropFirst(n) if e != nil { @@ -2468,10 +2458,6 @@ func (s *SocketOperations) nonBlockingRead(ctx context.Context, dst usermem.IOSe s.readView.TrimFront(int(n)) } - if len(s.readView) == 0 { - atomic.StoreUint32(&s.readViewHasData, 0) - } - var flags int if msgLen > int(n) { flags |= linux.MSG_TRUNC -- cgit v1.2.3 From 57d9bd922b4eff922d1a5185529fe5446249d592 Mon Sep 17 00:00:00 2001 From: Zach Koopmans Date: Thu, 19 Mar 2020 15:28:52 -0700 Subject: Remove the "frozen" bit from dirents. Frozen was to lock down changes to the host filesystem for hostFS. Now that hostFS is gone, it can be removed. PiperOrigin-RevId: 301907923 --- pkg/sentry/fs/dirent.go | 133 +------------------------------------ pkg/sentry/fs/file_overlay_test.go | 84 ----------------------- pkg/sentry/fs/mounts.go | 13 ---- 3 files changed, 1 insertion(+), 229 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/dirent.go b/pkg/sentry/fs/dirent.go index e0b32e1c1..0266a5287 100644 --- a/pkg/sentry/fs/dirent.go +++ b/pkg/sentry/fs/dirent.go @@ -17,7 +17,6 @@ package fs import ( "fmt" "path" - "sort" "sync/atomic" "syscall" @@ -121,9 +120,6 @@ type Dirent struct { // deleted may be set atomically when removed. deleted int32 - // frozen indicates this entry can't walk to unknown nodes. - frozen bool - // mounted is true if Dirent is a mount point, similar to include/linux/dcache.h:DCACHE_MOUNTED. mounted bool @@ -253,8 +249,7 @@ func (d *Dirent) IsNegative() bool { return d.Inode == nil } -// hashChild will hash child into the children list of its new parent d, carrying over -// any "frozen" state from d. +// hashChild will hash child into the children list of its new parent d. // // Returns (*WeakRef, true) if hashing child caused a Dirent to be unhashed. The caller must // validate the returned unhashed weak reference. Common cases: @@ -282,9 +277,6 @@ func (d *Dirent) hashChild(child *Dirent) (*refs.WeakRef, bool) { d.IncRef() } - // Carry over parent's frozen state. - child.frozen = d.frozen - return d.hashChildParentSet(child) } @@ -400,38 +392,6 @@ func (d *Dirent) MountRoot() *Dirent { return mountRoot } -// Freeze prevents this dirent from walking to more nodes. Freeze is applied -// recursively to all children. -// -// If this particular Dirent represents a Virtual node, then Walks and Creates -// may proceed as before. -// -// Freeze can only be called before the application starts running, otherwise -// the root it might be out of sync with the application root if modified by -// sys_chroot. -func (d *Dirent) Freeze() { - d.mu.Lock() - defer d.mu.Unlock() - if d.frozen { - // Already frozen. - return - } - d.frozen = true - - // Take a reference when freezing. - for _, w := range d.children { - if child := w.Get(); child != nil { - // NOTE: We would normally drop the reference here. But - // instead we're hanging on to it. - ch := child.(*Dirent) - ch.Freeze() - } - } - - // Drop all expired weak references. - d.flush() -} - // descendantOf returns true if the receiver dirent is equal to, or a // descendant of, the argument dirent. // @@ -524,11 +484,6 @@ func (d *Dirent) walk(ctx context.Context, root *Dirent, name string, walkMayUnl w.Drop() } - // Are we allowed to do the lookup? - if d.frozen && !d.Inode.IsVirtual() { - return nil, syscall.ENOENT - } - // Slow path: load the InodeOperations into memory. Since this is a hot path and the lookup may be // expensive, if possible release the lock and re-acquire it. if walkMayUnlock { @@ -659,11 +614,6 @@ func (d *Dirent) Create(ctx context.Context, root *Dirent, name string, flags Fi return nil, syscall.EEXIST } - // Are we frozen? - if d.frozen && !d.Inode.IsVirtual() { - return nil, syscall.ENOENT - } - // Try the create. We need to trust the file system to return EEXIST (or something // that will translate to EEXIST) if name already exists. file, err := d.Inode.Create(ctx, d, name, flags, perms) @@ -727,11 +677,6 @@ func (d *Dirent) genericCreate(ctx context.Context, root *Dirent, name string, c return syscall.EEXIST } - // Are we frozen? - if d.frozen && !d.Inode.IsVirtual() { - return syscall.ENOENT - } - // Remove any negative Dirent. We've already asserted above with d.exists // that the only thing remaining here can be a negative Dirent. if w, ok := d.children[name]; ok { @@ -862,49 +807,6 @@ func (d *Dirent) GetDotAttrs(root *Dirent) (DentAttr, DentAttr) { return dot, dot } -// readdirFrozen returns readdir results based solely on the frozen children. -func (d *Dirent) readdirFrozen(root *Dirent, offset int64, dirCtx *DirCtx) (int64, error) { - // Collect attrs for "." and "..". - attrs := make(map[string]DentAttr) - names := []string{".", ".."} - attrs["."], attrs[".."] = d.GetDotAttrs(root) - - // Get info from all children. - d.mu.Lock() - defer d.mu.Unlock() - for name, w := range d.children { - if child := w.Get(); child != nil { - defer child.DecRef() - - // Skip negative children. - if child.(*Dirent).IsNegative() { - continue - } - - sattr := child.(*Dirent).Inode.StableAttr - attrs[name] = DentAttr{ - Type: sattr.Type, - InodeID: sattr.InodeID, - } - names = append(names, name) - } - } - - sort.Strings(names) - - if int(offset) >= len(names) { - return offset, nil - } - names = names[int(offset):] - for _, name := range names { - if err := dirCtx.DirEmit(name, attrs[name]); err != nil { - return offset, err - } - offset++ - } - return offset, nil -} - // DirIterator is an open directory containing directory entries that can be read. type DirIterator interface { // IterateDir emits directory entries by calling dirCtx.EmitDir, beginning @@ -964,10 +866,6 @@ func direntReaddir(ctx context.Context, d *Dirent, it DirIterator, root *Dirent, return offset, nil } - if d.frozen { - return d.readdirFrozen(root, offset, dirCtx) - } - // Collect attrs for "." and "..". dot, dotdot := d.GetDotAttrs(root) @@ -1068,11 +966,6 @@ func (d *Dirent) mount(ctx context.Context, inode *Inode) (newChild *Dirent, err return nil, syserror.EINVAL } - // Are we frozen? - if d.parent.frozen && !d.parent.Inode.IsVirtual() { - return nil, syserror.ENOENT - } - // Dirent that'll replace d. // // Note that NewDirent returns with one reference taken; the reference @@ -1101,11 +994,6 @@ func (d *Dirent) unmount(ctx context.Context, replacement *Dirent) error { return syserror.ENOENT } - // Are we frozen? - if d.parent.frozen && !d.parent.Inode.IsVirtual() { - return syserror.ENOENT - } - // Remount our former child in its place. // // As replacement used to be our child, it must already have the right @@ -1135,11 +1023,6 @@ func (d *Dirent) Remove(ctx context.Context, root *Dirent, name string, dirPath unlock := d.lockDirectory() defer unlock() - // Are we frozen? - if d.frozen && !d.Inode.IsVirtual() { - return syscall.ENOENT - } - // Try to walk to the node. child, err := d.walk(ctx, root, name, false /* may unlock */) if err != nil { @@ -1201,11 +1084,6 @@ func (d *Dirent) RemoveDirectory(ctx context.Context, root *Dirent, name string) unlock := d.lockDirectory() defer unlock() - // Are we frozen? - if d.frozen && !d.Inode.IsVirtual() { - return syscall.ENOENT - } - // Check for dots. if name == "." { // Rejected as the last component by rmdir(2). @@ -1519,15 +1397,6 @@ func Rename(ctx context.Context, root *Dirent, oldParent *Dirent, oldName string return err } - // Are we frozen? - // TODO(jamieliu): Is this the right errno? - if oldParent.frozen && !oldParent.Inode.IsVirtual() { - return syscall.ENOENT - } - if newParent.frozen && !newParent.Inode.IsVirtual() { - return syscall.ENOENT - } - // Do we have general permission to remove from oldParent and // create/replace in newParent? if err := oldParent.Inode.CheckPermission(ctx, PermMask{Write: true, Execute: true}); err != nil { diff --git a/pkg/sentry/fs/file_overlay_test.go b/pkg/sentry/fs/file_overlay_test.go index a76d87e3a..1971cc680 100644 --- a/pkg/sentry/fs/file_overlay_test.go +++ b/pkg/sentry/fs/file_overlay_test.go @@ -175,90 +175,6 @@ func TestReaddirRevalidation(t *testing.T) { } } -// TestReaddirOverlayFrozen tests that calling Readdir on an overlay file with -// a frozen dirent tree does not make Readdir calls to the underlying files. -// This is a regression test for b/114808269. -func TestReaddirOverlayFrozen(t *testing.T) { - ctx := contexttest.Context(t) - - // Create an overlay with two directories, each with two files. - upper := newTestRamfsDir(ctx, []dirContent{{name: "upper-file1"}, {name: "upper-file2"}}, nil) - lower := newTestRamfsDir(ctx, []dirContent{{name: "lower-file1"}, {name: "lower-file2"}}, nil) - overlayInode := fs.NewTestOverlayDir(ctx, upper, lower, false) - - // Set that overlay as the root. - root := fs.NewDirent(ctx, overlayInode, "root") - ctx = &rootContext{ - Context: ctx, - root: root, - } - - // Check that calling Readdir on the root now returns all 4 files (2 - // from each layer in the overlay). - rootFile, err := root.Inode.GetFile(ctx, root, fs.FileFlags{Read: true}) - if err != nil { - t.Fatalf("root.Inode.GetFile failed: %v", err) - } - defer rootFile.DecRef() - ser := &fs.CollectEntriesSerializer{} - if err := rootFile.Readdir(ctx, ser); err != nil { - t.Fatalf("rootFile.Readdir failed: %v", err) - } - if got, want := ser.Order, []string{".", "..", "lower-file1", "lower-file2", "upper-file1", "upper-file2"}; !reflect.DeepEqual(got, want) { - t.Errorf("Readdir got names %v, want %v", got, want) - } - - // Readdir should have been called on upper and lower. - upperDir := upper.InodeOperations.(*dir) - lowerDir := lower.InodeOperations.(*dir) - if !upperDir.ReaddirCalled { - t.Errorf("upperDir.ReaddirCalled got %v, want true", upperDir.ReaddirCalled) - } - if !lowerDir.ReaddirCalled { - t.Errorf("lowerDir.ReaddirCalled got %v, want true", lowerDir.ReaddirCalled) - } - - // Reset. - upperDir.ReaddirCalled = false - lowerDir.ReaddirCalled = false - - // Take references on "upper-file1" and "lower-file1", pinning them in - // the dirent tree. - for _, name := range []string{"upper-file1", "lower-file1"} { - if _, err := root.Walk(ctx, root, name); err != nil { - t.Fatalf("root.Walk(%q) failed: %v", name, err) - } - // Don't drop a reference on the returned dirent so that it - // will stay in the tree. - } - - // Freeze the dirent tree. - root.Freeze() - - // Seek back to the beginning of the file. - if _, err := rootFile.Seek(ctx, fs.SeekSet, 0); err != nil { - t.Fatalf("error seeking to beginning of directory: %v", err) - } - - // Calling Readdir on the root now will return only the pinned - // children. - ser = &fs.CollectEntriesSerializer{} - if err := rootFile.Readdir(ctx, ser); err != nil { - t.Fatalf("rootFile.Readdir failed: %v", err) - } - if got, want := ser.Order, []string{".", "..", "lower-file1", "upper-file1"}; !reflect.DeepEqual(got, want) { - t.Errorf("Readdir got names %v, want %v", got, want) - } - - // Readdir should NOT have been called on upper or lower. - if upperDir.ReaddirCalled { - t.Errorf("upperDir.ReaddirCalled got %v, want false", upperDir.ReaddirCalled) - } - if lowerDir.ReaddirCalled { - t.Errorf("lowerDir.ReaddirCalled got %v, want false", lowerDir.ReaddirCalled) - } -} - type rootContext struct { context.Context root *fs.Dirent diff --git a/pkg/sentry/fs/mounts.go b/pkg/sentry/fs/mounts.go index c7981f66e..b414ddaee 100644 --- a/pkg/sentry/fs/mounts.go +++ b/pkg/sentry/fs/mounts.go @@ -273,19 +273,6 @@ func (mns *MountNamespace) DecRef() { mns.DecRefWithDestructor(mns.destroy) } -// Freeze freezes the entire mount tree. -func (mns *MountNamespace) Freeze() { - mns.mu.Lock() - defer mns.mu.Unlock() - - // We only want to freeze Dirents with active references, not Dirents referenced - // by a mount's MountSource. - mns.flushMountSourceRefsLocked() - - // Freeze the entire shebang. - mns.root.Freeze() -} - // withMountLocked prevents further walks to `node`, because `node` is about to // be a mount point. func (mns *MountNamespace) withMountLocked(node *Dirent, fn func() error) error { -- cgit v1.2.3 From 248e46f320525704da917e148a8f69d9b74671a0 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Thu, 19 Mar 2020 23:29:15 -0700 Subject: Whitelist utimensat(2). utimensat is used by hostfs for setting timestamps on imported fds. Previously, this would crash the sandbox since utimensat was not allowed. Correct the VFS2 version of hostfs to match the call in VFS1. PiperOrigin-RevId: 301970121 --- pkg/sentry/fsimpl/host/BUILD | 1 + pkg/sentry/fsimpl/host/host.go | 4 ++-- pkg/sentry/fsimpl/host/util.go | 8 ++++---- pkg/sentry/fsimpl/host/util_unsafe.go | 34 ++++++++++++++++++++++++++++++++++ runsc/boot/filter/config.go | 8 ++++++++ 5 files changed, 49 insertions(+), 6 deletions(-) create mode 100644 pkg/sentry/fsimpl/host/util_unsafe.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/host/BUILD b/pkg/sentry/fsimpl/host/BUILD index 0bb4a5c3e..82e1fb74b 100644 --- a/pkg/sentry/fsimpl/host/BUILD +++ b/pkg/sentry/fsimpl/host/BUILD @@ -9,6 +9,7 @@ go_library( "ioctl_unsafe.go", "tty.go", "util.go", + "util_unsafe.go", ], visibility = ["//pkg/sentry:internal"], deps = [ diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 3afb41395..1f735628f 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -322,11 +322,11 @@ func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Cre } } if m&(linux.STATX_ATIME|linux.STATX_MTIME) != 0 { - timestamps := []unix.Timespec{ + ts := [2]syscall.Timespec{ toTimespec(s.Atime, m&linux.STATX_ATIME == 0), toTimespec(s.Mtime, m&linux.STATX_MTIME == 0), } - if err := unix.UtimesNanoAt(i.hostFD, "", timestamps, unix.AT_EMPTY_PATH); err != nil { + if err := setTimestamps(i.hostFD, &ts); err != nil { return err } } diff --git a/pkg/sentry/fsimpl/host/util.go b/pkg/sentry/fsimpl/host/util.go index d519feef5..2bc757b1a 100644 --- a/pkg/sentry/fsimpl/host/util.go +++ b/pkg/sentry/fsimpl/host/util.go @@ -22,15 +22,15 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) -func toTimespec(ts linux.StatxTimestamp, omit bool) unix.Timespec { +func toTimespec(ts linux.StatxTimestamp, omit bool) syscall.Timespec { if omit { - return unix.Timespec{ + return syscall.Timespec{ Sec: 0, Nsec: unix.UTIME_OMIT, } } - return unix.Timespec{ - Sec: int64(ts.Sec), + return syscall.Timespec{ + Sec: ts.Sec, Nsec: int64(ts.Nsec), } } diff --git a/pkg/sentry/fsimpl/host/util_unsafe.go b/pkg/sentry/fsimpl/host/util_unsafe.go new file mode 100644 index 000000000..5136ac844 --- /dev/null +++ b/pkg/sentry/fsimpl/host/util_unsafe.go @@ -0,0 +1,34 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package host + +import ( + "syscall" + "unsafe" +) + +func setTimestamps(fd int, ts *[2]syscall.Timespec) error { + _, _, errno := syscall.Syscall6( + syscall.SYS_UTIMENSAT, + uintptr(fd), + 0, /* path */ + uintptr(unsafe.Pointer(ts)), + 0, /* flags */ + 0, 0) + if errno != 0 { + return errno + } + return nil +} diff --git a/runsc/boot/filter/config.go b/runsc/boot/filter/config.go index f459d1973..06b9f888a 100644 --- a/runsc/boot/filter/config.go +++ b/runsc/boot/filter/config.go @@ -291,6 +291,14 @@ var allowedSyscalls = seccomp.SyscallRules{ seccomp.AllowValue(uint64(os.Getpid())), }, }, + syscall.SYS_UTIMENSAT: []seccomp.Rule{ + { + seccomp.AllowAny{}, + seccomp.AllowValue(0), /* null pathname */ + seccomp.AllowAny{}, + seccomp.AllowValue(0), /* flags */ + }, + }, syscall.SYS_WRITE: {}, // The only user in rawfile.NonBlockingWrite3 always passes iovcnt with // values 2 or 3. Three iovec-s are passed, when the PACKET_VNET_HDR -- cgit v1.2.3 From 6eebaea94936ffefd3603ff16ad2225856313fa3 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Mon, 23 Mar 2020 14:32:05 -0700 Subject: Correctly release taskPathOperation for accessAt. PiperOrigin-RevId: 302518924 --- pkg/sentry/syscalls/linux/vfs2/stat.go | 1 + 1 file changed, 1 insertion(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/vfs2/stat.go b/pkg/sentry/syscalls/linux/vfs2/stat.go index 97eaedd66..068243132 100644 --- a/pkg/sentry/syscalls/linux/vfs2/stat.go +++ b/pkg/sentry/syscalls/linux/vfs2/stat.go @@ -272,6 +272,7 @@ func accessAt(t *kernel.Task, dirfd int32, pathAddr usermem.Addr, mode uint) err if err != nil { return err } + defer tpop.Release() // access(2) and faccessat(2) check permissions using real // UID/GID, not effective UID/GID. -- cgit v1.2.3 From 369cf38bd7186da97e134538cd4839a8a4d1aa2c Mon Sep 17 00:00:00 2001 From: Bhasker Hariharan Date: Mon, 23 Mar 2020 16:05:29 -0700 Subject: Fix data race in SetSockOpt. PiperOrigin-RevId: 302539171 --- pkg/sentry/socket/netstack/netstack.go | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index a6ef7a47e..c19f5639b 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -2382,9 +2382,9 @@ func (s *SocketOperations) nonBlockingRead(ctx context.Context, dst usermem.IOSe // caller-supplied buffer. s.readMu.Lock() n, err := s.coalescingRead(ctx, dst, trunc) - s.readMu.Unlock() cmsg := s.controlMessages() s.fillCmsgInq(&cmsg) + s.readMu.Unlock() return n, 0, nil, 0, cmsg, err } -- cgit v1.2.3 From a730d74b3230fb32181b9a940c07b61338222874 Mon Sep 17 00:00:00 2001 From: Ian Lewis Date: Mon, 23 Mar 2020 16:11:37 -0700 Subject: Support basic /proc/net/dev metrics for netstack Fixes #506 PiperOrigin-RevId: 302540404 --- pkg/sentry/socket/netstack/stack.go | 73 ++++++++++++++++++++++++++----------- test/syscalls/linux/proc_net.cc | 53 +++++++++++++++++++++++++++ 2 files changed, 105 insertions(+), 21 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/stack.go b/pkg/sentry/socket/netstack/stack.go index 0692482e9..a8e2e8c24 100644 --- a/pkg/sentry/socket/netstack/stack.go +++ b/pkg/sentry/socket/netstack/stack.go @@ -200,36 +200,66 @@ func (s *Stack) SetTCPSACKEnabled(enabled bool) error { // Statistics implements inet.Stack.Statistics. func (s *Stack) Statistics(stat interface{}, arg string) error { switch stats := stat.(type) { + case *inet.StatDev: + for _, ni := range s.Stack.NICInfo() { + if ni.Name != arg { + continue + } + // TODO(gvisor.dev/issue/2103) Support stubbed stats. + *stats = inet.StatDev{ + // Receive section. + ni.Stats.Rx.Bytes.Value(), // bytes. + ni.Stats.Rx.Packets.Value(), // packets. + 0, // errs. + 0, // drop. + 0, // fifo. + 0, // frame. + 0, // compressed. + 0, // multicast. + // Transmit section. + ni.Stats.Tx.Bytes.Value(), // bytes. + ni.Stats.Tx.Packets.Value(), // packets. + 0, // errs. + 0, // drop. + 0, // fifo. + 0, // colls. + 0, // carrier. + 0, // compressed. + } + break + } case *inet.StatSNMPIP: ip := Metrics.IP + // TODO(gvisor.dev/issue/969) Support stubbed stats. *stats = inet.StatSNMPIP{ - 0, // TODO(gvisor.dev/issue/969): Support Ip/Forwarding. - 0, // TODO(gvisor.dev/issue/969): Support Ip/DefaultTTL. + 0, // Ip/Forwarding. + 0, // Ip/DefaultTTL. ip.PacketsReceived.Value(), // InReceives. - 0, // TODO(gvisor.dev/issue/969): Support Ip/InHdrErrors. + 0, // Ip/InHdrErrors. ip.InvalidDestinationAddressesReceived.Value(), // InAddrErrors. - 0, // TODO(gvisor.dev/issue/969): Support Ip/ForwDatagrams. - 0, // TODO(gvisor.dev/issue/969): Support Ip/InUnknownProtos. - 0, // TODO(gvisor.dev/issue/969): Support Ip/InDiscards. + 0, // Ip/ForwDatagrams. + 0, // Ip/InUnknownProtos. + 0, // Ip/InDiscards. ip.PacketsDelivered.Value(), // InDelivers. ip.PacketsSent.Value(), // OutRequests. ip.OutgoingPacketErrors.Value(), // OutDiscards. - 0, // TODO(gvisor.dev/issue/969): Support Ip/OutNoRoutes. - 0, // TODO(gvisor.dev/issue/969): Support Ip/ReasmTimeout. - 0, // TODO(gvisor.dev/issue/969): Support Ip/ReasmReqds. - 0, // TODO(gvisor.dev/issue/969): Support Ip/ReasmOKs. - 0, // TODO(gvisor.dev/issue/969): Support Ip/ReasmFails. - 0, // TODO(gvisor.dev/issue/969): Support Ip/FragOKs. - 0, // TODO(gvisor.dev/issue/969): Support Ip/FragFails. - 0, // TODO(gvisor.dev/issue/969): Support Ip/FragCreates. + 0, // Ip/OutNoRoutes. + 0, // Support Ip/ReasmTimeout. + 0, // Support Ip/ReasmReqds. + 0, // Support Ip/ReasmOKs. + 0, // Support Ip/ReasmFails. + 0, // Support Ip/FragOKs. + 0, // Support Ip/FragFails. + 0, // Support Ip/FragCreates. } case *inet.StatSNMPICMP: in := Metrics.ICMP.V4PacketsReceived.ICMPv4PacketStats out := Metrics.ICMP.V4PacketsSent.ICMPv4PacketStats + // TODO(gvisor.dev/issue/969) Support stubbed stats. *stats = inet.StatSNMPICMP{ - 0, // TODO(gvisor.dev/issue/969): Support Icmp/InMsgs. + 0, // Icmp/InMsgs. Metrics.ICMP.V4PacketsSent.Dropped.Value(), // InErrors. - 0, // TODO(gvisor.dev/issue/969): Support Icmp/InCsumErrors. + 0, // Icmp/InCsumErrors. in.DstUnreachable.Value(), // InDestUnreachs. in.TimeExceeded.Value(), // InTimeExcds. in.ParamProblem.Value(), // InParmProbs. @@ -241,7 +271,7 @@ func (s *Stack) Statistics(stat interface{}, arg string) error { in.TimestampReply.Value(), // InTimestampReps. in.InfoRequest.Value(), // InAddrMasks. in.InfoReply.Value(), // InAddrMaskReps. - 0, // TODO(gvisor.dev/issue/969): Support Icmp/OutMsgs. + 0, // Icmp/OutMsgs. Metrics.ICMP.V4PacketsReceived.Invalid.Value(), // OutErrors. out.DstUnreachable.Value(), // OutDestUnreachs. out.TimeExceeded.Value(), // OutTimeExcds. @@ -277,15 +307,16 @@ func (s *Stack) Statistics(stat interface{}, arg string) error { } case *inet.StatSNMPUDP: udp := Metrics.UDP + // TODO(gvisor.dev/issue/969) Support stubbed stats. *stats = inet.StatSNMPUDP{ udp.PacketsReceived.Value(), // InDatagrams. udp.UnknownPortErrors.Value(), // NoPorts. - 0, // TODO(gvisor.dev/issue/969): Support Udp/InErrors. + 0, // Udp/InErrors. udp.PacketsSent.Value(), // OutDatagrams. udp.ReceiveBufferErrors.Value(), // RcvbufErrors. - 0, // TODO(gvisor.dev/issue/969): Support Udp/SndbufErrors. - 0, // TODO(gvisor.dev/issue/969): Support Udp/InCsumErrors. - 0, // TODO(gvisor.dev/issue/969): Support Udp/IgnoredMulti. + 0, // Udp/SndbufErrors. + 0, // Udp/InCsumErrors. + 0, // Udp/IgnoredMulti. } default: return syserr.ErrEndpointOperation.ToError() diff --git a/test/syscalls/linux/proc_net.cc b/test/syscalls/linux/proc_net.cc index 05c952b99..4e23d1e78 100644 --- a/test/syscalls/linux/proc_net.cc +++ b/test/syscalls/linux/proc_net.cc @@ -92,6 +92,59 @@ TEST(ProcSysNetIpv4Sack, CanReadAndWrite) { EXPECT_EQ(buf, to_write); } +// DeviceEntry is an entry in /proc/net/dev +struct DeviceEntry { + std::string name; + uint64_t stats[16]; +}; + +PosixErrorOr> GetDeviceMetricsFromProc( + const std::string dev) { + std::vector lines = absl::StrSplit(dev, '\n'); + std::vector entries; + + // /proc/net/dev prints 2 lines of headers followed by a line of metrics for + // each network interface. + for (unsigned i = 2; i < lines.size(); i++) { + // Ignore empty lines. + if (lines[i].empty()) { + continue; + } + + std::vector values = + absl::StrSplit(lines[i], ' ', absl::SkipWhitespace()); + + // Interface name + 16 values. + if (values.size() != 17) { + return PosixError(EINVAL, "invalid line: " + lines[i]); + } + + DeviceEntry entry; + entry.name = values[0]; + // Skip the interface name and read only the values. + for (unsigned j = 1; j < 17; j++) { + uint64_t num; + if (!absl::SimpleAtoi(values[j], &num)) { + return PosixError(EINVAL, "invalid value: " + values[j]); + } + entry.stats[j - 1] = num; + } + + entries.push_back(entry); + } + + return entries; +} + +// TEST(ProcNetDev, Format) tests that /proc/net/dev is parsable and +// contains at least one entry. +TEST(ProcNetDev, Format) { + auto dev = ASSERT_NO_ERRNO_AND_VALUE(GetContents("/proc/net/dev")); + auto entries = ASSERT_NO_ERRNO_AND_VALUE(GetDeviceMetricsFromProc(dev)); + + EXPECT_GT(entries.size(), 0); +} + PosixErrorOr GetSNMPMetricFromProc(const std::string snmp, const std::string& type, const std::string& item) { -- cgit v1.2.3 From 7e4073af12bed2c76bc5757ef3e5fbfba75308a0 Mon Sep 17 00:00:00 2001 From: Bhasker Hariharan Date: Tue, 24 Mar 2020 09:05:06 -0700 Subject: Move tcpip.PacketBuffer and IPTables to stack package. This is a precursor to be being able to build an intrusive list of PacketBuffers for use in queuing disciplines being implemented. Updates #2214 PiperOrigin-RevId: 302677662 --- pkg/sentry/socket/netfilter/BUILD | 1 - pkg/sentry/socket/netfilter/extensions.go | 14 +- pkg/sentry/socket/netfilter/netfilter.go | 121 ++++---- pkg/sentry/socket/netfilter/targets.go | 11 +- pkg/sentry/socket/netfilter/tcp_matcher.go | 11 +- pkg/sentry/socket/netfilter/udp_matcher.go | 13 +- pkg/sentry/socket/netstack/BUILD | 1 - pkg/sentry/socket/netstack/stack.go | 3 +- pkg/tcpip/BUILD | 2 - pkg/tcpip/iptables/BUILD | 18 -- pkg/tcpip/iptables/iptables.go | 314 --------------------- pkg/tcpip/iptables/targets.go | 144 ---------- pkg/tcpip/iptables/types.go | 180 ------------ pkg/tcpip/link/channel/channel.go | 14 +- pkg/tcpip/link/fdbased/endpoint.go | 6 +- pkg/tcpip/link/fdbased/endpoint_test.go | 10 +- pkg/tcpip/link/fdbased/mmap.go | 3 +- pkg/tcpip/link/fdbased/packet_dispatchers.go | 4 +- pkg/tcpip/link/loopback/loopback.go | 8 +- pkg/tcpip/link/muxed/injectable.go | 6 +- pkg/tcpip/link/muxed/injectable_test.go | 4 +- pkg/tcpip/link/sharedmem/sharedmem.go | 6 +- pkg/tcpip/link/sharedmem/sharedmem_test.go | 26 +- pkg/tcpip/link/sniffer/sniffer.go | 10 +- pkg/tcpip/link/tun/device.go | 2 +- pkg/tcpip/link/waitable/waitable.go | 6 +- pkg/tcpip/link/waitable/waitable_test.go | 18 +- pkg/tcpip/network/arp/arp.go | 12 +- pkg/tcpip/network/arp/arp_test.go | 2 +- pkg/tcpip/network/ip_test.go | 24 +- pkg/tcpip/network/ipv4/BUILD | 1 - pkg/tcpip/network/ipv4/icmp.go | 9 +- pkg/tcpip/network/ipv4/ipv4.go | 21 +- pkg/tcpip/network/ipv4/ipv4_test.go | 18 +- pkg/tcpip/network/ipv6/icmp.go | 10 +- pkg/tcpip/network/ipv6/icmp_test.go | 14 +- pkg/tcpip/network/ipv6/ipv6.go | 10 +- pkg/tcpip/network/ipv6/ipv6_test.go | 4 +- pkg/tcpip/network/ipv6/ndp_test.go | 8 +- pkg/tcpip/packet_buffer.go | 67 ----- pkg/tcpip/packet_buffer_state.go | 27 -- pkg/tcpip/stack/BUILD | 8 +- pkg/tcpip/stack/forwarder.go | 4 +- pkg/tcpip/stack/forwarder_test.go | 36 +-- pkg/tcpip/stack/iptables.go | 311 ++++++++++++++++++++ pkg/tcpip/stack/iptables_targets.go | 144 ++++++++++ pkg/tcpip/stack/iptables_types.go | 180 ++++++++++++ pkg/tcpip/stack/ndp.go | 4 +- pkg/tcpip/stack/ndp_test.go | 14 +- pkg/tcpip/stack/nic.go | 13 +- pkg/tcpip/stack/nic_test.go | 3 +- pkg/tcpip/stack/packet_buffer.go | 66 +++++ pkg/tcpip/stack/packet_buffer_state.go | 26 ++ pkg/tcpip/stack/registration.go | 32 +-- pkg/tcpip/stack/route.go | 6 +- pkg/tcpip/stack/stack.go | 11 +- pkg/tcpip/stack/stack_test.go | 28 +- pkg/tcpip/stack/transport_demuxer.go | 14 +- pkg/tcpip/stack/transport_demuxer_test.go | 2 +- pkg/tcpip/stack/transport_test.go | 27 +- pkg/tcpip/transport/icmp/BUILD | 1 - pkg/tcpip/transport/icmp/endpoint.go | 11 +- pkg/tcpip/transport/icmp/protocol.go | 2 +- pkg/tcpip/transport/packet/BUILD | 1 - pkg/tcpip/transport/packet/endpoint.go | 9 +- pkg/tcpip/transport/raw/BUILD | 1 - pkg/tcpip/transport/raw/endpoint.go | 9 +- pkg/tcpip/transport/tcp/BUILD | 1 - pkg/tcpip/transport/tcp/connect.go | 6 +- pkg/tcpip/transport/tcp/dispatcher.go | 3 +- pkg/tcpip/transport/tcp/endpoint.go | 7 +- pkg/tcpip/transport/tcp/forwarder.go | 2 +- pkg/tcpip/transport/tcp/protocol.go | 4 +- pkg/tcpip/transport/tcp/segment.go | 3 +- pkg/tcpip/transport/tcp/testing/context/context.go | 10 +- pkg/tcpip/transport/udp/BUILD | 1 - pkg/tcpip/transport/udp/endpoint.go | 9 +- pkg/tcpip/transport/udp/forwarder.go | 4 +- pkg/tcpip/transport/udp/protocol.go | 6 +- pkg/tcpip/transport/udp/udp_test.go | 4 +- 80 files changed, 1080 insertions(+), 1126 deletions(-) delete mode 100644 pkg/tcpip/iptables/BUILD delete mode 100644 pkg/tcpip/iptables/iptables.go delete mode 100644 pkg/tcpip/iptables/targets.go delete mode 100644 pkg/tcpip/iptables/types.go delete mode 100644 pkg/tcpip/packet_buffer.go delete mode 100644 pkg/tcpip/packet_buffer_state.go create mode 100644 pkg/tcpip/stack/iptables.go create mode 100644 pkg/tcpip/stack/iptables_targets.go create mode 100644 pkg/tcpip/stack/iptables_types.go create mode 100644 pkg/tcpip/stack/packet_buffer.go create mode 100644 pkg/tcpip/stack/packet_buffer_state.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/BUILD b/pkg/sentry/socket/netfilter/BUILD index 7cd2ce55b..e801abeb8 100644 --- a/pkg/sentry/socket/netfilter/BUILD +++ b/pkg/sentry/socket/netfilter/BUILD @@ -22,7 +22,6 @@ go_library( "//pkg/syserr", "//pkg/tcpip", "//pkg/tcpip/header", - "//pkg/tcpip/iptables", "//pkg/tcpip/stack", "//pkg/usermem", ], diff --git a/pkg/sentry/socket/netfilter/extensions.go b/pkg/sentry/socket/netfilter/extensions.go index b4b244abf..0336a32d8 100644 --- a/pkg/sentry/socket/netfilter/extensions.go +++ b/pkg/sentry/socket/netfilter/extensions.go @@ -19,7 +19,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/tcpip/iptables" + "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/usermem" ) @@ -37,12 +37,12 @@ type matchMaker interface { // name is the matcher name as stored in the xt_entry_match struct. name() string - // marshal converts from an iptables.Matcher to an ABI struct. - marshal(matcher iptables.Matcher) []byte + // marshal converts from an stack.Matcher to an ABI struct. + marshal(matcher stack.Matcher) []byte // unmarshal converts from the ABI matcher struct to an - // iptables.Matcher. - unmarshal(buf []byte, filter iptables.IPHeaderFilter) (iptables.Matcher, error) + // stack.Matcher. + unmarshal(buf []byte, filter stack.IPHeaderFilter) (stack.Matcher, error) } // matchMakers maps the name of supported matchers to the matchMaker that @@ -58,7 +58,7 @@ func registerMatchMaker(mm matchMaker) { matchMakers[mm.name()] = mm } -func marshalMatcher(matcher iptables.Matcher) []byte { +func marshalMatcher(matcher stack.Matcher) []byte { matchMaker, ok := matchMakers[matcher.Name()] if !ok { panic(fmt.Sprintf("Unknown matcher of type %T.", matcher)) @@ -86,7 +86,7 @@ func marshalEntryMatch(name string, data []byte) []byte { return append(buf, make([]byte, size-len(buf))...) } -func unmarshalMatcher(match linux.XTEntryMatch, filter iptables.IPHeaderFilter, buf []byte) (iptables.Matcher, error) { +func unmarshalMatcher(match linux.XTEntryMatch, filter stack.IPHeaderFilter, buf []byte) (stack.Matcher, error) { matchMaker, ok := matchMakers[match.Name.String()] if !ok { return nil, fmt.Errorf("unsupported matcher with name %q", match.Name.String()) diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index b5b9be46f..55bcc3ace 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -27,7 +27,6 @@ import ( "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/header" - "gvisor.dev/gvisor/pkg/tcpip/iptables" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/usermem" ) @@ -129,19 +128,19 @@ func GetEntries(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr, outLen return entries, nil } -func findTable(stack *stack.Stack, tablename linux.TableName) (iptables.Table, error) { - ipt := stack.IPTables() +func findTable(stk *stack.Stack, tablename linux.TableName) (stack.Table, error) { + ipt := stk.IPTables() table, ok := ipt.Tables[tablename.String()] if !ok { - return iptables.Table{}, fmt.Errorf("couldn't find table %q", tablename) + return stack.Table{}, fmt.Errorf("couldn't find table %q", tablename) } return table, nil } // FillDefaultIPTables sets stack's IPTables to the default tables and // populates them with metadata. -func FillDefaultIPTables(stack *stack.Stack) { - ipt := iptables.DefaultTables() +func FillDefaultIPTables(stk *stack.Stack) { + ipt := stack.DefaultTables() // In order to fill in the metadata, we have to translate ipt from its // netstack format to Linux's giant-binary-blob format. @@ -154,14 +153,14 @@ func FillDefaultIPTables(stack *stack.Stack) { ipt.Tables[name] = table } - stack.SetIPTables(ipt) + stk.SetIPTables(ipt) } // convertNetstackToBinary converts the iptables as stored in netstack to the // format expected by the iptables tool. Linux stores each table as a binary // blob that can only be traversed by parsing a bit, reading some offsets, // jumping to those offsets, parsing again, etc. -func convertNetstackToBinary(tablename string, table iptables.Table) (linux.KernelIPTGetEntries, metadata, error) { +func convertNetstackToBinary(tablename string, table stack.Table) (linux.KernelIPTGetEntries, metadata, error) { // Return values. var entries linux.KernelIPTGetEntries var meta metadata @@ -234,19 +233,19 @@ func convertNetstackToBinary(tablename string, table iptables.Table) (linux.Kern return entries, meta, nil } -func marshalTarget(target iptables.Target) []byte { +func marshalTarget(target stack.Target) []byte { switch tg := target.(type) { - case iptables.AcceptTarget: - return marshalStandardTarget(iptables.RuleAccept) - case iptables.DropTarget: - return marshalStandardTarget(iptables.RuleDrop) - case iptables.ErrorTarget: + case stack.AcceptTarget: + return marshalStandardTarget(stack.RuleAccept) + case stack.DropTarget: + return marshalStandardTarget(stack.RuleDrop) + case stack.ErrorTarget: return marshalErrorTarget(errorTargetName) - case iptables.UserChainTarget: + case stack.UserChainTarget: return marshalErrorTarget(tg.Name) - case iptables.ReturnTarget: - return marshalStandardTarget(iptables.RuleReturn) - case iptables.RedirectTarget: + case stack.ReturnTarget: + return marshalStandardTarget(stack.RuleReturn) + case stack.RedirectTarget: return marshalRedirectTarget() case JumpTarget: return marshalJumpTarget(tg) @@ -255,7 +254,7 @@ func marshalTarget(target iptables.Target) []byte { } } -func marshalStandardTarget(verdict iptables.RuleVerdict) []byte { +func marshalStandardTarget(verdict stack.RuleVerdict) []byte { nflog("convert to binary: marshalling standard target") // The target's name will be the empty string. @@ -316,13 +315,13 @@ func marshalJumpTarget(jt JumpTarget) []byte { // translateFromStandardVerdict translates verdicts the same way as the iptables // tool. -func translateFromStandardVerdict(verdict iptables.RuleVerdict) int32 { +func translateFromStandardVerdict(verdict stack.RuleVerdict) int32 { switch verdict { - case iptables.RuleAccept: + case stack.RuleAccept: return -linux.NF_ACCEPT - 1 - case iptables.RuleDrop: + case stack.RuleDrop: return -linux.NF_DROP - 1 - case iptables.RuleReturn: + case stack.RuleReturn: return linux.NF_RETURN default: // TODO(gvisor.dev/issue/170): Support Jump. @@ -331,18 +330,18 @@ func translateFromStandardVerdict(verdict iptables.RuleVerdict) int32 { } // translateToStandardTarget translates from the value in a -// linux.XTStandardTarget to an iptables.Verdict. -func translateToStandardTarget(val int32) (iptables.Target, error) { +// linux.XTStandardTarget to an stack.Verdict. +func translateToStandardTarget(val int32) (stack.Target, error) { // TODO(gvisor.dev/issue/170): Support other verdicts. switch val { case -linux.NF_ACCEPT - 1: - return iptables.AcceptTarget{}, nil + return stack.AcceptTarget{}, nil case -linux.NF_DROP - 1: - return iptables.DropTarget{}, nil + return stack.DropTarget{}, nil case -linux.NF_QUEUE - 1: return nil, errors.New("unsupported iptables verdict QUEUE") case linux.NF_RETURN: - return iptables.ReturnTarget{}, nil + return stack.ReturnTarget{}, nil default: return nil, fmt.Errorf("unknown iptables verdict %d", val) } @@ -350,7 +349,7 @@ func translateToStandardTarget(val int32) (iptables.Target, error) { // SetEntries sets iptables rules for a single table. See // net/ipv4/netfilter/ip_tables.c:translate_table for reference. -func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { +func SetEntries(stk *stack.Stack, optVal []byte) *syserr.Error { // Get the basic rules data (struct ipt_replace). if len(optVal) < linux.SizeOfIPTReplace { nflog("optVal has insufficient size for replace %d", len(optVal)) @@ -362,12 +361,12 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { binary.Unmarshal(replaceBuf, usermem.ByteOrder, &replace) // TODO(gvisor.dev/issue/170): Support other tables. - var table iptables.Table + var table stack.Table switch replace.Name.String() { - case iptables.TablenameFilter: - table = iptables.EmptyFilterTable() - case iptables.TablenameNat: - table = iptables.EmptyNatTable() + case stack.TablenameFilter: + table = stack.EmptyFilterTable() + case stack.TablenameNat: + table = stack.EmptyNatTable() default: nflog("we don't yet support writing to the %q table (gvisor.dev/issue/170)", replace.Name.String()) return syserr.ErrInvalidArgument @@ -434,7 +433,7 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { } optVal = optVal[targetSize:] - table.Rules = append(table.Rules, iptables.Rule{ + table.Rules = append(table.Rules, stack.Rule{ Filter: filter, Target: target, Matchers: matchers, @@ -465,11 +464,11 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { table.Underflows[hk] = ruleIdx } } - if ruleIdx := table.BuiltinChains[hk]; ruleIdx == iptables.HookUnset { + if ruleIdx := table.BuiltinChains[hk]; ruleIdx == stack.HookUnset { nflog("hook %v is unset.", hk) return syserr.ErrInvalidArgument } - if ruleIdx := table.Underflows[hk]; ruleIdx == iptables.HookUnset { + if ruleIdx := table.Underflows[hk]; ruleIdx == stack.HookUnset { nflog("underflow %v is unset.", hk) return syserr.ErrInvalidArgument } @@ -478,7 +477,7 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { // Add the user chains. for ruleIdx, rule := range table.Rules { - target, ok := rule.Target.(iptables.UserChainTarget) + target, ok := rule.Target.(stack.UserChainTarget) if !ok { continue } @@ -522,8 +521,8 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { // PREROUTING chain right now, make sure all other chains point to // ACCEPT rules. for hook, ruleIdx := range table.BuiltinChains { - if hook != iptables.Input && hook != iptables.Prerouting { - if _, ok := table.Rules[ruleIdx].Target.(iptables.AcceptTarget); !ok { + if hook != stack.Input && hook != stack.Prerouting { + if _, ok := table.Rules[ruleIdx].Target.(stack.AcceptTarget); !ok { nflog("hook %d is unsupported.", hook) return syserr.ErrInvalidArgument } @@ -535,7 +534,7 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { // - There are no chains without an unconditional final rule. // - There are no chains without an unconditional underflow rule. - ipt := stack.IPTables() + ipt := stk.IPTables() table.SetMetadata(metadata{ HookEntry: replace.HookEntry, Underflow: replace.Underflow, @@ -543,16 +542,16 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { Size: replace.Size, }) ipt.Tables[replace.Name.String()] = table - stack.SetIPTables(ipt) + stk.SetIPTables(ipt) return nil } // parseMatchers parses 0 or more matchers from optVal. optVal should contain // only the matchers. -func parseMatchers(filter iptables.IPHeaderFilter, optVal []byte) ([]iptables.Matcher, error) { +func parseMatchers(filter stack.IPHeaderFilter, optVal []byte) ([]stack.Matcher, error) { nflog("set entries: parsing matchers of size %d", len(optVal)) - var matchers []iptables.Matcher + var matchers []stack.Matcher for len(optVal) > 0 { nflog("set entries: optVal has len %d", len(optVal)) @@ -594,7 +593,7 @@ func parseMatchers(filter iptables.IPHeaderFilter, optVal []byte) ([]iptables.Ma // parseTarget parses a target from optVal. optVal should contain only the // target. -func parseTarget(filter iptables.IPHeaderFilter, optVal []byte) (iptables.Target, error) { +func parseTarget(filter stack.IPHeaderFilter, optVal []byte) (stack.Target, error) { nflog("set entries: parsing target of size %d", len(optVal)) if len(optVal) < linux.SizeOfXTEntryTarget { return nil, fmt.Errorf("optVal has insufficient size for entry target %d", len(optVal)) @@ -638,11 +637,11 @@ func parseTarget(filter iptables.IPHeaderFilter, optVal []byte) (iptables.Target switch name := errorTarget.Name.String(); name { case errorTargetName: nflog("set entries: error target") - return iptables.ErrorTarget{}, nil + return stack.ErrorTarget{}, nil default: // User defined chain. nflog("set entries: user-defined target %q", name) - return iptables.UserChainTarget{Name: name}, nil + return stack.UserChainTarget{Name: name}, nil } case redirectTargetName: @@ -659,8 +658,8 @@ func parseTarget(filter iptables.IPHeaderFilter, optVal []byte) (iptables.Target buf = optVal[:linux.SizeOfXTRedirectTarget] binary.Unmarshal(buf, usermem.ByteOrder, &redirectTarget) - // Copy linux.XTRedirectTarget to iptables.RedirectTarget. - var target iptables.RedirectTarget + // Copy linux.XTRedirectTarget to stack.RedirectTarget. + var target stack.RedirectTarget nfRange := redirectTarget.NfRange // RangeSize should be 1. @@ -699,14 +698,14 @@ func parseTarget(filter iptables.IPHeaderFilter, optVal []byte) (iptables.Target return nil, fmt.Errorf("unknown target %q doesn't exist or isn't supported yet.", target.Name.String()) } -func filterFromIPTIP(iptip linux.IPTIP) (iptables.IPHeaderFilter, error) { +func filterFromIPTIP(iptip linux.IPTIP) (stack.IPHeaderFilter, error) { if containsUnsupportedFields(iptip) { - return iptables.IPHeaderFilter{}, fmt.Errorf("unsupported fields in struct iptip: %+v", iptip) + return stack.IPHeaderFilter{}, fmt.Errorf("unsupported fields in struct iptip: %+v", iptip) } if len(iptip.Dst) != header.IPv4AddressSize || len(iptip.DstMask) != header.IPv4AddressSize { - return iptables.IPHeaderFilter{}, fmt.Errorf("incorrect length of destination (%d) and/or destination mask (%d) fields", len(iptip.Dst), len(iptip.DstMask)) + return stack.IPHeaderFilter{}, fmt.Errorf("incorrect length of destination (%d) and/or destination mask (%d) fields", len(iptip.Dst), len(iptip.DstMask)) } - return iptables.IPHeaderFilter{ + return stack.IPHeaderFilter{ Protocol: tcpip.TransportProtocolNumber(iptip.Protocol), Dst: tcpip.Address(iptip.Dst[:]), DstMask: tcpip.Address(iptip.DstMask[:]), @@ -733,30 +732,30 @@ func containsUnsupportedFields(iptip linux.IPTIP) bool { iptip.InverseFlags&^inverseMask != 0 } -func validUnderflow(rule iptables.Rule) bool { +func validUnderflow(rule stack.Rule) bool { if len(rule.Matchers) != 0 { return false } switch rule.Target.(type) { - case iptables.AcceptTarget, iptables.DropTarget: + case stack.AcceptTarget, stack.DropTarget: return true default: return false } } -func hookFromLinux(hook int) iptables.Hook { +func hookFromLinux(hook int) stack.Hook { switch hook { case linux.NF_INET_PRE_ROUTING: - return iptables.Prerouting + return stack.Prerouting case linux.NF_INET_LOCAL_IN: - return iptables.Input + return stack.Input case linux.NF_INET_FORWARD: - return iptables.Forward + return stack.Forward case linux.NF_INET_LOCAL_OUT: - return iptables.Output + return stack.Output case linux.NF_INET_POST_ROUTING: - return iptables.Postrouting + return stack.Postrouting } panic(fmt.Sprintf("Unknown hook %d does not correspond to a builtin chain", hook)) } diff --git a/pkg/sentry/socket/netfilter/targets.go b/pkg/sentry/socket/netfilter/targets.go index c421b87cf..c948de876 100644 --- a/pkg/sentry/socket/netfilter/targets.go +++ b/pkg/sentry/socket/netfilter/targets.go @@ -15,11 +15,10 @@ package netfilter import ( - "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/iptables" + "gvisor.dev/gvisor/pkg/tcpip/stack" ) -// JumpTarget implements iptables.Target. +// JumpTarget implements stack.Target. type JumpTarget struct { // Offset is the byte offset of the rule to jump to. It is used for // marshaling and unmarshaling. @@ -29,7 +28,7 @@ type JumpTarget struct { RuleNum int } -// Action implements iptables.Target.Action. -func (jt JumpTarget) Action(tcpip.PacketBuffer) (iptables.RuleVerdict, int) { - return iptables.RuleJump, jt.RuleNum +// Action implements stack.Target.Action. +func (jt JumpTarget) Action(stack.PacketBuffer) (stack.RuleVerdict, int) { + return stack.RuleJump, jt.RuleNum } diff --git a/pkg/sentry/socket/netfilter/tcp_matcher.go b/pkg/sentry/socket/netfilter/tcp_matcher.go index f9945e214..ff1cfd8f6 100644 --- a/pkg/sentry/socket/netfilter/tcp_matcher.go +++ b/pkg/sentry/socket/netfilter/tcp_matcher.go @@ -19,9 +19,8 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/header" - "gvisor.dev/gvisor/pkg/tcpip/iptables" + "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/usermem" ) @@ -40,7 +39,7 @@ func (tcpMarshaler) name() string { } // marshal implements matchMaker.marshal. -func (tcpMarshaler) marshal(mr iptables.Matcher) []byte { +func (tcpMarshaler) marshal(mr stack.Matcher) []byte { matcher := mr.(*TCPMatcher) xttcp := linux.XTTCP{ SourcePortStart: matcher.sourcePortStart, @@ -53,7 +52,7 @@ func (tcpMarshaler) marshal(mr iptables.Matcher) []byte { } // unmarshal implements matchMaker.unmarshal. -func (tcpMarshaler) unmarshal(buf []byte, filter iptables.IPHeaderFilter) (iptables.Matcher, error) { +func (tcpMarshaler) unmarshal(buf []byte, filter stack.IPHeaderFilter) (stack.Matcher, error) { if len(buf) < linux.SizeOfXTTCP { return nil, fmt.Errorf("buf has insufficient size for TCP match: %d", len(buf)) } @@ -97,7 +96,7 @@ func (*TCPMatcher) Name() string { } // Match implements Matcher.Match. -func (tm *TCPMatcher) Match(hook iptables.Hook, pkt tcpip.PacketBuffer, interfaceName string) (bool, bool) { +func (tm *TCPMatcher) Match(hook stack.Hook, pkt stack.PacketBuffer, interfaceName string) (bool, bool) { netHeader := header.IPv4(pkt.NetworkHeader) if netHeader.TransportProtocol() != header.TCPProtocolNumber { @@ -115,7 +114,7 @@ func (tm *TCPMatcher) Match(hook iptables.Hook, pkt tcpip.PacketBuffer, interfac // Now we need the transport header. However, this may not have been set // yet. // TODO(gvisor.dev/issue/170): Parsing the transport header should - // ultimately be moved into the iptables.Check codepath as matchers are + // ultimately be moved into the stack.Check codepath as matchers are // added. var tcpHeader header.TCP if pkt.TransportHeader != nil { diff --git a/pkg/sentry/socket/netfilter/udp_matcher.go b/pkg/sentry/socket/netfilter/udp_matcher.go index 86aa11696..3359418c1 100644 --- a/pkg/sentry/socket/netfilter/udp_matcher.go +++ b/pkg/sentry/socket/netfilter/udp_matcher.go @@ -19,9 +19,8 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/header" - "gvisor.dev/gvisor/pkg/tcpip/iptables" + "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/usermem" ) @@ -40,7 +39,7 @@ func (udpMarshaler) name() string { } // marshal implements matchMaker.marshal. -func (udpMarshaler) marshal(mr iptables.Matcher) []byte { +func (udpMarshaler) marshal(mr stack.Matcher) []byte { matcher := mr.(*UDPMatcher) xtudp := linux.XTUDP{ SourcePortStart: matcher.sourcePortStart, @@ -53,7 +52,7 @@ func (udpMarshaler) marshal(mr iptables.Matcher) []byte { } // unmarshal implements matchMaker.unmarshal. -func (udpMarshaler) unmarshal(buf []byte, filter iptables.IPHeaderFilter) (iptables.Matcher, error) { +func (udpMarshaler) unmarshal(buf []byte, filter stack.IPHeaderFilter) (stack.Matcher, error) { if len(buf) < linux.SizeOfXTUDP { return nil, fmt.Errorf("buf has insufficient size for UDP match: %d", len(buf)) } @@ -94,11 +93,11 @@ func (*UDPMatcher) Name() string { } // Match implements Matcher.Match. -func (um *UDPMatcher) Match(hook iptables.Hook, pkt tcpip.PacketBuffer, interfaceName string) (bool, bool) { +func (um *UDPMatcher) Match(hook stack.Hook, pkt stack.PacketBuffer, interfaceName string) (bool, bool) { netHeader := header.IPv4(pkt.NetworkHeader) // TODO(gvisor.dev/issue/170): Proto checks should ultimately be moved - // into the iptables.Check codepath as matchers are added. + // into the stack.Check codepath as matchers are added. if netHeader.TransportProtocol() != header.UDPProtocolNumber { return false, false } @@ -114,7 +113,7 @@ func (um *UDPMatcher) Match(hook iptables.Hook, pkt tcpip.PacketBuffer, interfac // Now we need the transport header. However, this may not have been set // yet. // TODO(gvisor.dev/issue/170): Parsing the transport header should - // ultimately be moved into the iptables.Check codepath as matchers are + // ultimately be moved into the stack.Check codepath as matchers are // added. var udpHeader header.UDP if pkt.TransportHeader != nil { diff --git a/pkg/sentry/socket/netstack/BUILD b/pkg/sentry/socket/netstack/BUILD index ab01cb4fa..cbf46b1e9 100644 --- a/pkg/sentry/socket/netstack/BUILD +++ b/pkg/sentry/socket/netstack/BUILD @@ -38,7 +38,6 @@ go_library( "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/header", - "//pkg/tcpip/iptables", "//pkg/tcpip/network/ipv4", "//pkg/tcpip/network/ipv6", "//pkg/tcpip/stack", diff --git a/pkg/sentry/socket/netstack/stack.go b/pkg/sentry/socket/netstack/stack.go index a8e2e8c24..f5fa18136 100644 --- a/pkg/sentry/socket/netstack/stack.go +++ b/pkg/sentry/socket/netstack/stack.go @@ -23,7 +23,6 @@ import ( "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/header" - "gvisor.dev/gvisor/pkg/tcpip/iptables" "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" "gvisor.dev/gvisor/pkg/tcpip/network/ipv6" "gvisor.dev/gvisor/pkg/tcpip/stack" @@ -363,7 +362,7 @@ func (s *Stack) RouteTable() []inet.Route { } // IPTables returns the stack's iptables. -func (s *Stack) IPTables() (iptables.IPTables, error) { +func (s *Stack) IPTables() (stack.IPTables, error) { return s.Stack.IPTables(), nil } diff --git a/pkg/tcpip/BUILD b/pkg/tcpip/BUILD index 26f7ba86b..454e07662 100644 --- a/pkg/tcpip/BUILD +++ b/pkg/tcpip/BUILD @@ -5,8 +5,6 @@ package(licenses = ["notice"]) go_library( name = "tcpip", srcs = [ - "packet_buffer.go", - "packet_buffer_state.go", "tcpip.go", "time_unsafe.go", "timer.go", diff --git a/pkg/tcpip/iptables/BUILD b/pkg/tcpip/iptables/BUILD deleted file mode 100644 index d1b73cfdf..000000000 --- a/pkg/tcpip/iptables/BUILD +++ /dev/null @@ -1,18 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "iptables", - srcs = [ - "iptables.go", - "targets.go", - "types.go", - ], - visibility = ["//visibility:public"], - deps = [ - "//pkg/log", - "//pkg/tcpip", - "//pkg/tcpip/header", - ], -) diff --git a/pkg/tcpip/iptables/iptables.go b/pkg/tcpip/iptables/iptables.go deleted file mode 100644 index d30571c74..000000000 --- a/pkg/tcpip/iptables/iptables.go +++ /dev/null @@ -1,314 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package iptables supports packet filtering and manipulation via the iptables -// tool. -package iptables - -import ( - "fmt" - - "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/header" -) - -// Table names. -const ( - TablenameNat = "nat" - TablenameMangle = "mangle" - TablenameFilter = "filter" -) - -// Chain names as defined by net/ipv4/netfilter/ip_tables.c. -const ( - ChainNamePrerouting = "PREROUTING" - ChainNameInput = "INPUT" - ChainNameForward = "FORWARD" - ChainNameOutput = "OUTPUT" - ChainNamePostrouting = "POSTROUTING" -) - -// HookUnset indicates that there is no hook set for an entrypoint or -// underflow. -const HookUnset = -1 - -// DefaultTables returns a default set of tables. Each chain is set to accept -// all packets. -func DefaultTables() IPTables { - // TODO(gvisor.dev/issue/170): We may be able to swap out some strings for - // iotas. - return IPTables{ - Tables: map[string]Table{ - TablenameNat: Table{ - Rules: []Rule{ - Rule{Target: AcceptTarget{}}, - Rule{Target: AcceptTarget{}}, - Rule{Target: AcceptTarget{}}, - Rule{Target: AcceptTarget{}}, - Rule{Target: ErrorTarget{}}, - }, - BuiltinChains: map[Hook]int{ - Prerouting: 0, - Input: 1, - Output: 2, - Postrouting: 3, - }, - Underflows: map[Hook]int{ - Prerouting: 0, - Input: 1, - Output: 2, - Postrouting: 3, - }, - UserChains: map[string]int{}, - }, - TablenameMangle: Table{ - Rules: []Rule{ - Rule{Target: AcceptTarget{}}, - Rule{Target: AcceptTarget{}}, - Rule{Target: ErrorTarget{}}, - }, - BuiltinChains: map[Hook]int{ - Prerouting: 0, - Output: 1, - }, - Underflows: map[Hook]int{ - Prerouting: 0, - Output: 1, - }, - UserChains: map[string]int{}, - }, - TablenameFilter: Table{ - Rules: []Rule{ - Rule{Target: AcceptTarget{}}, - Rule{Target: AcceptTarget{}}, - Rule{Target: AcceptTarget{}}, - Rule{Target: ErrorTarget{}}, - }, - BuiltinChains: map[Hook]int{ - Input: 0, - Forward: 1, - Output: 2, - }, - Underflows: map[Hook]int{ - Input: 0, - Forward: 1, - Output: 2, - }, - UserChains: map[string]int{}, - }, - }, - Priorities: map[Hook][]string{ - Input: []string{TablenameNat, TablenameFilter}, - Prerouting: []string{TablenameMangle, TablenameNat}, - Output: []string{TablenameMangle, TablenameNat, TablenameFilter}, - }, - } -} - -// EmptyFilterTable returns a Table with no rules and the filter table chains -// mapped to HookUnset. -func EmptyFilterTable() Table { - return Table{ - Rules: []Rule{}, - BuiltinChains: map[Hook]int{ - Input: HookUnset, - Forward: HookUnset, - Output: HookUnset, - }, - Underflows: map[Hook]int{ - Input: HookUnset, - Forward: HookUnset, - Output: HookUnset, - }, - UserChains: map[string]int{}, - } -} - -// EmptyNatTable returns a Table with no rules and the filter table chains -// mapped to HookUnset. -func EmptyNatTable() Table { - return Table{ - Rules: []Rule{}, - BuiltinChains: map[Hook]int{ - Prerouting: HookUnset, - Input: HookUnset, - Output: HookUnset, - Postrouting: HookUnset, - }, - Underflows: map[Hook]int{ - Prerouting: HookUnset, - Input: HookUnset, - Output: HookUnset, - Postrouting: HookUnset, - }, - UserChains: map[string]int{}, - } -} - -// A chainVerdict is what a table decides should be done with a packet. -type chainVerdict int - -const ( - // chainAccept indicates the packet should continue through netstack. - chainAccept chainVerdict = iota - - // chainAccept indicates the packet should be dropped. - chainDrop - - // chainReturn indicates the packet should return to the calling chain - // or the underflow rule of a builtin chain. - chainReturn -) - -// Check runs pkt through the rules for hook. It returns true when the packet -// should continue traversing the network stack and false when it should be -// dropped. -// -// Precondition: pkt.NetworkHeader is set. -func (it *IPTables) Check(hook Hook, pkt tcpip.PacketBuffer) bool { - // Go through each table containing the hook. - for _, tablename := range it.Priorities[hook] { - table := it.Tables[tablename] - ruleIdx := table.BuiltinChains[hook] - switch verdict := it.checkChain(hook, pkt, table, ruleIdx); verdict { - // If the table returns Accept, move on to the next table. - case chainAccept: - continue - // The Drop verdict is final. - case chainDrop: - return false - case chainReturn: - // Any Return from a built-in chain means we have to - // call the underflow. - underflow := table.Rules[table.Underflows[hook]] - switch v, _ := underflow.Target.Action(pkt); v { - case RuleAccept: - continue - case RuleDrop: - return false - case RuleJump, RuleReturn: - panic("Underflows should only return RuleAccept or RuleDrop.") - default: - panic(fmt.Sprintf("Unknown verdict: %d", v)) - } - - default: - panic(fmt.Sprintf("Unknown verdict %v.", verdict)) - } - } - - // Every table returned Accept. - return true -} - -// Precondition: pkt.NetworkHeader is set. -func (it *IPTables) checkChain(hook Hook, pkt tcpip.PacketBuffer, table Table, ruleIdx int) chainVerdict { - // Start from ruleIdx and walk the list of rules until a rule gives us - // a verdict. - for ruleIdx < len(table.Rules) { - switch verdict, jumpTo := it.checkRule(hook, pkt, table, ruleIdx); verdict { - case RuleAccept: - return chainAccept - - case RuleDrop: - return chainDrop - - case RuleReturn: - return chainReturn - - case RuleJump: - // "Jumping" to the next rule just means we're - // continuing on down the list. - if jumpTo == ruleIdx+1 { - ruleIdx++ - continue - } - switch verdict := it.checkChain(hook, pkt, table, jumpTo); verdict { - case chainAccept: - return chainAccept - case chainDrop: - return chainDrop - case chainReturn: - ruleIdx++ - continue - default: - panic(fmt.Sprintf("Unknown verdict: %d", verdict)) - } - - default: - panic(fmt.Sprintf("Unknown verdict: %d", verdict)) - } - - } - - // We got through the entire table without a decision. Default to DROP - // for safety. - return chainDrop -} - -// Precondition: pk.NetworkHeader is set. -func (it *IPTables) checkRule(hook Hook, pkt tcpip.PacketBuffer, table Table, ruleIdx int) (RuleVerdict, int) { - rule := table.Rules[ruleIdx] - - // If pkt.NetworkHeader hasn't been set yet, it will be contained in - // pkt.Data.First(). - if pkt.NetworkHeader == nil { - pkt.NetworkHeader = pkt.Data.First() - } - - // Check whether the packet matches the IP header filter. - if !filterMatch(rule.Filter, header.IPv4(pkt.NetworkHeader)) { - // Continue on to the next rule. - return RuleJump, ruleIdx + 1 - } - - // Go through each rule matcher. If they all match, run - // the rule target. - for _, matcher := range rule.Matchers { - matches, hotdrop := matcher.Match(hook, pkt, "") - if hotdrop { - return RuleDrop, 0 - } - if !matches { - // Continue on to the next rule. - return RuleJump, ruleIdx + 1 - } - } - - // All the matchers matched, so run the target. - return rule.Target.Action(pkt) -} - -func filterMatch(filter IPHeaderFilter, hdr header.IPv4) bool { - // TODO(gvisor.dev/issue/170): Support other fields of the filter. - // Check the transport protocol. - if filter.Protocol != 0 && filter.Protocol != hdr.TransportProtocol() { - return false - } - - // Check the destination IP. - dest := hdr.DestinationAddress() - matches := true - for i := range filter.Dst { - if dest[i]&filter.DstMask[i] != filter.Dst[i] { - matches = false - break - } - } - if matches == filter.DstInvert { - return false - } - - return true -} diff --git a/pkg/tcpip/iptables/targets.go b/pkg/tcpip/iptables/targets.go deleted file mode 100644 index e457f2349..000000000 --- a/pkg/tcpip/iptables/targets.go +++ /dev/null @@ -1,144 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package iptables - -import ( - "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/header" -) - -// AcceptTarget accepts packets. -type AcceptTarget struct{} - -// Action implements Target.Action. -func (AcceptTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, int) { - return RuleAccept, 0 -} - -// DropTarget drops packets. -type DropTarget struct{} - -// Action implements Target.Action. -func (DropTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, int) { - return RuleDrop, 0 -} - -// ErrorTarget logs an error and drops the packet. It represents a target that -// should be unreachable. -type ErrorTarget struct{} - -// Action implements Target.Action. -func (ErrorTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, int) { - log.Debugf("ErrorTarget triggered.") - return RuleDrop, 0 -} - -// UserChainTarget marks a rule as the beginning of a user chain. -type UserChainTarget struct { - Name string -} - -// Action implements Target.Action. -func (UserChainTarget) Action(tcpip.PacketBuffer) (RuleVerdict, int) { - panic("UserChainTarget should never be called.") -} - -// ReturnTarget returns from the current chain. If the chain is a built-in, the -// hook's underflow should be called. -type ReturnTarget struct{} - -// Action implements Target.Action. -func (ReturnTarget) Action(tcpip.PacketBuffer) (RuleVerdict, int) { - return RuleReturn, 0 -} - -// RedirectTarget redirects the packet by modifying the destination port/IP. -// Min and Max values for IP and Ports in the struct indicate the range of -// values which can be used to redirect. -type RedirectTarget struct { - // TODO(gvisor.dev/issue/170): Other flags need to be added after - // we support them. - // RangeProtoSpecified flag indicates single port is specified to - // redirect. - RangeProtoSpecified bool - - // Min address used to redirect. - MinIP tcpip.Address - - // Max address used to redirect. - MaxIP tcpip.Address - - // Min port used to redirect. - MinPort uint16 - - // Max port used to redirect. - MaxPort uint16 -} - -// Action implements Target.Action. -// TODO(gvisor.dev/issue/170): Parse headers without copying. The current -// implementation only works for PREROUTING and calls pkt.Clone(), neither -// of which should be the case. -func (rt RedirectTarget) Action(pkt tcpip.PacketBuffer) (RuleVerdict, int) { - newPkt := pkt.Clone() - - // Set network header. - headerView := newPkt.Data.First() - netHeader := header.IPv4(headerView) - newPkt.NetworkHeader = headerView[:header.IPv4MinimumSize] - - hlen := int(netHeader.HeaderLength()) - tlen := int(netHeader.TotalLength()) - newPkt.Data.TrimFront(hlen) - newPkt.Data.CapLength(tlen - hlen) - - // TODO(gvisor.dev/issue/170): Change destination address to - // loopback or interface address on which the packet was - // received. - - // TODO(gvisor.dev/issue/170): Check Flags in RedirectTarget if - // we need to change dest address (for OUTPUT chain) or ports. - switch protocol := netHeader.TransportProtocol(); protocol { - case header.UDPProtocolNumber: - var udpHeader header.UDP - if newPkt.TransportHeader != nil { - udpHeader = header.UDP(newPkt.TransportHeader) - } else { - if len(pkt.Data.First()) < header.UDPMinimumSize { - return RuleDrop, 0 - } - udpHeader = header.UDP(newPkt.Data.First()) - } - udpHeader.SetDestinationPort(rt.MinPort) - case header.TCPProtocolNumber: - var tcpHeader header.TCP - if newPkt.TransportHeader != nil { - tcpHeader = header.TCP(newPkt.TransportHeader) - } else { - if len(pkt.Data.First()) < header.TCPMinimumSize { - return RuleDrop, 0 - } - tcpHeader = header.TCP(newPkt.TransportHeader) - } - // TODO(gvisor.dev/issue/170): Need to recompute checksum - // and implement nat connection tracking to support TCP. - tcpHeader.SetDestinationPort(rt.MinPort) - default: - return RuleDrop, 0 - } - - return RuleAccept, 0 -} diff --git a/pkg/tcpip/iptables/types.go b/pkg/tcpip/iptables/types.go deleted file mode 100644 index e7fcf6bff..000000000 --- a/pkg/tcpip/iptables/types.go +++ /dev/null @@ -1,180 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package iptables - -import ( - "gvisor.dev/gvisor/pkg/tcpip" -) - -// A Hook specifies one of the hooks built into the network stack. -// -// Userspace app Userspace app -// ^ | -// | v -// [Input] [Output] -// ^ | -// | v -// | routing -// | | -// | v -// ----->[Prerouting]----->routing----->[Forward]---------[Postrouting]-----> -type Hook uint - -// These values correspond to values in include/uapi/linux/netfilter.h. -const ( - // Prerouting happens before a packet is routed to applications or to - // be forwarded. - Prerouting Hook = iota - - // Input happens before a packet reaches an application. - Input - - // Forward happens once it's decided that a packet should be forwarded - // to another host. - Forward - - // Output happens after a packet is written by an application to be - // sent out. - Output - - // Postrouting happens just before a packet goes out on the wire. - Postrouting - - // The total number of hooks. - NumHooks -) - -// A RuleVerdict is what a rule decides should be done with a packet. -type RuleVerdict int - -const ( - // RuleAccept indicates the packet should continue through netstack. - RuleAccept RuleVerdict = iota - - // RuleDrop indicates the packet should be dropped. - RuleDrop - - // RuleJump indicates the packet should jump to another chain. - RuleJump - - // RuleReturn indicates the packet should return to the previous chain. - RuleReturn -) - -// IPTables holds all the tables for a netstack. -type IPTables struct { - // Tables maps table names to tables. User tables have arbitrary names. - Tables map[string]Table - - // Priorities maps each hook to a list of table names. The order of the - // list is the order in which each table should be visited for that - // hook. - Priorities map[Hook][]string -} - -// A Table defines a set of chains and hooks into the network stack. It is -// really just a list of rules with some metadata for entrypoints and such. -type Table struct { - // Rules holds the rules that make up the table. - Rules []Rule - - // BuiltinChains maps builtin chains to their entrypoint rule in Rules. - BuiltinChains map[Hook]int - - // Underflows maps builtin chains to their underflow rule in Rules - // (i.e. the rule to execute if the chain returns without a verdict). - Underflows map[Hook]int - - // UserChains holds user-defined chains for the keyed by name. Users - // can give their chains arbitrary names. - UserChains map[string]int - - // Metadata holds information about the Table that is useful to users - // of IPTables, but not to the netstack IPTables code itself. - metadata interface{} -} - -// ValidHooks returns a bitmap of the builtin hooks for the given table. -func (table *Table) ValidHooks() uint32 { - hooks := uint32(0) - for hook := range table.BuiltinChains { - hooks |= 1 << hook - } - return hooks -} - -// Metadata returns the metadata object stored in table. -func (table *Table) Metadata() interface{} { - return table.metadata -} - -// SetMetadata sets the metadata object stored in table. -func (table *Table) SetMetadata(metadata interface{}) { - table.metadata = metadata -} - -// A Rule is a packet processing rule. It consists of two pieces. First it -// contains zero or more matchers, each of which is a specification of which -// packets this rule applies to. If there are no matchers in the rule, it -// applies to any packet. -type Rule struct { - // Filter holds basic IP filtering fields common to every rule. - Filter IPHeaderFilter - - // Matchers is the list of matchers for this rule. - Matchers []Matcher - - // Target is the action to invoke if all the matchers match the packet. - Target Target -} - -// IPHeaderFilter holds basic IP filtering data common to every rule. -type IPHeaderFilter struct { - // Protocol matches the transport protocol. - Protocol tcpip.TransportProtocolNumber - - // Dst matches the destination IP address. - Dst tcpip.Address - - // DstMask masks bits of the destination IP address when comparing with - // Dst. - DstMask tcpip.Address - - // DstInvert inverts the meaning of the destination IP check, i.e. when - // true the filter will match packets that fail the destination - // comparison. - DstInvert bool -} - -// A Matcher is the interface for matching packets. -type Matcher interface { - // Name returns the name of the Matcher. - Name() string - - // Match returns whether the packet matches and whether the packet - // should be "hotdropped", i.e. dropped immediately. This is usually - // used for suspicious packets. - // - // Precondition: packet.NetworkHeader is set. - Match(hook Hook, packet tcpip.PacketBuffer, interfaceName string) (matches bool, hotdrop bool) -} - -// A Target is the interface for taking an action for a packet. -type Target interface { - // Action takes an action on the packet and returns a verdict on how - // traversal should (or should not) continue. If the return value is - // Jump, it also returns the index of the rule to jump to. - Action(packet tcpip.PacketBuffer) (RuleVerdict, int) -} diff --git a/pkg/tcpip/link/channel/channel.go b/pkg/tcpip/link/channel/channel.go index 5944ba190..a8d6653ce 100644 --- a/pkg/tcpip/link/channel/channel.go +++ b/pkg/tcpip/link/channel/channel.go @@ -28,7 +28,7 @@ import ( // PacketInfo holds all the information about an outbound packet. type PacketInfo struct { - Pkt tcpip.PacketBuffer + Pkt stack.PacketBuffer Proto tcpip.NetworkProtocolNumber GSO *stack.GSO Route stack.Route @@ -203,12 +203,12 @@ func (e *Endpoint) NumQueued() int { } // InjectInbound injects an inbound packet. -func (e *Endpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { +func (e *Endpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { e.InjectLinkAddr(protocol, "", pkt) } // InjectLinkAddr injects an inbound packet with a remote link address. -func (e *Endpoint) InjectLinkAddr(protocol tcpip.NetworkProtocolNumber, remote tcpip.LinkAddress, pkt tcpip.PacketBuffer) { +func (e *Endpoint) InjectLinkAddr(protocol tcpip.NetworkProtocolNumber, remote tcpip.LinkAddress, pkt stack.PacketBuffer) { e.dispatcher.DeliverNetworkPacket(e, remote, "" /* local */, protocol, pkt) } @@ -251,7 +251,7 @@ func (e *Endpoint) LinkAddress() tcpip.LinkAddress { } // WritePacket stores outbound packets into the channel. -func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { +func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { // Clone r then release its resource so we only get the relevant fields from // stack.Route without holding a reference to a NIC's endpoint. route := r.Clone() @@ -269,7 +269,7 @@ func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne } // WritePackets stores outbound packets into the channel. -func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []stack.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { // Clone r then release its resource so we only get the relevant fields from // stack.Route without holding a reference to a NIC's endpoint. route := r.Clone() @@ -280,7 +280,7 @@ func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.Pac off := pkt.DataOffset size := pkt.DataSize p := PacketInfo{ - Pkt: tcpip.PacketBuffer{ + Pkt: stack.PacketBuffer{ Header: pkt.Header, Data: buffer.NewViewFromBytes(payloadView[off : off+size]).ToVectorisedView(), }, @@ -301,7 +301,7 @@ func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.Pac // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. func (e *Endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { p := PacketInfo{ - Pkt: tcpip.PacketBuffer{Data: vv}, + Pkt: stack.PacketBuffer{Data: vv}, Proto: 0, GSO: nil, } diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go index 3b36b9673..235e647ff 100644 --- a/pkg/tcpip/link/fdbased/endpoint.go +++ b/pkg/tcpip/link/fdbased/endpoint.go @@ -386,7 +386,7 @@ const ( // WritePacket writes outbound packets to the file descriptor. If it is not // currently writable, the packet is dropped. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { if e.hdrSize > 0 { // Add ethernet header if needed. eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize)) @@ -440,7 +440,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne // WritePackets writes outbound packets to the file descriptor. If it is not // currently writable, the packet is dropped. -func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []stack.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { var ethHdrBuf []byte // hdr + data iovLen := 2 @@ -610,7 +610,7 @@ func (e *InjectableEndpoint) Attach(dispatcher stack.NetworkDispatcher) { } // InjectInbound injects an inbound packet. -func (e *InjectableEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { +func (e *InjectableEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, protocol, pkt) } diff --git a/pkg/tcpip/link/fdbased/endpoint_test.go b/pkg/tcpip/link/fdbased/endpoint_test.go index 2066987eb..c7dbbbc6b 100644 --- a/pkg/tcpip/link/fdbased/endpoint_test.go +++ b/pkg/tcpip/link/fdbased/endpoint_test.go @@ -45,7 +45,7 @@ const ( type packetInfo struct { raddr tcpip.LinkAddress proto tcpip.NetworkProtocolNumber - contents tcpip.PacketBuffer + contents stack.PacketBuffer } type context struct { @@ -92,7 +92,7 @@ func (c *context) cleanup() { syscall.Close(c.fds[1]) } -func (c *context) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote tcpip.LinkAddress, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { +func (c *context) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote tcpip.LinkAddress, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { c.ch <- packetInfo{remote, protocol, pkt} } @@ -168,7 +168,7 @@ func testWritePacket(t *testing.T, plen int, eth bool, gsoMaxSize uint32) { L3HdrLen: header.IPv4MaximumHeaderSize, } } - if err := c.ep.WritePacket(r, gso, proto, tcpip.PacketBuffer{ + if err := c.ep.WritePacket(r, gso, proto, stack.PacketBuffer{ Header: hdr, Data: payload.ToVectorisedView(), }); err != nil { @@ -261,7 +261,7 @@ func TestPreserveSrcAddress(t *testing.T) { // WritePacket panics given a prependable with anything less than // the minimum size of the ethernet header. hdr := buffer.NewPrependable(header.EthernetMinimumSize) - if err := c.ep.WritePacket(r, nil /* gso */, proto, tcpip.PacketBuffer{ + if err := c.ep.WritePacket(r, nil /* gso */, proto, stack.PacketBuffer{ Header: hdr, Data: buffer.VectorisedView{}, }); err != nil { @@ -324,7 +324,7 @@ func TestDeliverPacket(t *testing.T) { want := packetInfo{ raddr: raddr, proto: proto, - contents: tcpip.PacketBuffer{ + contents: stack.PacketBuffer{ Data: buffer.View(b).ToVectorisedView(), LinkHeader: buffer.View(hdr), }, diff --git a/pkg/tcpip/link/fdbased/mmap.go b/pkg/tcpip/link/fdbased/mmap.go index 62ed1e569..fe2bf3b0b 100644 --- a/pkg/tcpip/link/fdbased/mmap.go +++ b/pkg/tcpip/link/fdbased/mmap.go @@ -25,6 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/link/rawfile" + "gvisor.dev/gvisor/pkg/tcpip/stack" ) const ( @@ -190,7 +191,7 @@ func (d *packetMMapDispatcher) dispatch() (bool, *tcpip.Error) { } pkt = pkt[d.e.hdrSize:] - d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, tcpip.PacketBuffer{ + d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, stack.PacketBuffer{ Data: buffer.View(pkt).ToVectorisedView(), LinkHeader: buffer.View(eth), }) diff --git a/pkg/tcpip/link/fdbased/packet_dispatchers.go b/pkg/tcpip/link/fdbased/packet_dispatchers.go index c67d684ce..cb4cbea69 100644 --- a/pkg/tcpip/link/fdbased/packet_dispatchers.go +++ b/pkg/tcpip/link/fdbased/packet_dispatchers.go @@ -139,7 +139,7 @@ func (d *readVDispatcher) dispatch() (bool, *tcpip.Error) { } used := d.capViews(n, BufConfig) - pkt := tcpip.PacketBuffer{ + pkt := stack.PacketBuffer{ Data: buffer.NewVectorisedView(n, append([]buffer.View(nil), d.views[:used]...)), LinkHeader: buffer.View(eth), } @@ -296,7 +296,7 @@ func (d *recvMMsgDispatcher) dispatch() (bool, *tcpip.Error) { } used := d.capViews(k, int(n), BufConfig) - pkt := tcpip.PacketBuffer{ + pkt := stack.PacketBuffer{ Data: buffer.NewVectorisedView(int(n), append([]buffer.View(nil), d.views[k][:used]...)), LinkHeader: buffer.View(eth), } diff --git a/pkg/tcpip/link/loopback/loopback.go b/pkg/tcpip/link/loopback/loopback.go index 499cc608f..4039753b7 100644 --- a/pkg/tcpip/link/loopback/loopback.go +++ b/pkg/tcpip/link/loopback/loopback.go @@ -76,7 +76,7 @@ func (*endpoint) Wait() {} // WritePacket implements stack.LinkEndpoint.WritePacket. It delivers outbound // packets to the network-layer dispatcher. -func (e *endpoint) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { +func (e *endpoint) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { views := make([]buffer.View, 1, 1+len(pkt.Data.Views())) views[0] = pkt.Header.View() views = append(views, pkt.Data.Views()...) @@ -84,7 +84,7 @@ func (e *endpoint) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.Netw // Because we're immediately turning around and writing the packet back // to the rx path, we intentionally don't preserve the remote and local // link addresses from the stack.Route we're passed. - e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, protocol, tcpip.PacketBuffer{ + e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, protocol, stack.PacketBuffer{ Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views), }) @@ -92,7 +92,7 @@ func (e *endpoint) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.Netw } // WritePackets implements stack.LinkEndpoint.WritePackets. -func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, []tcpip.PacketBuffer, tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, []stack.PacketBuffer, tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { panic("not implemented") } @@ -106,7 +106,7 @@ func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { // There should be an ethernet header at the beginning of vv. linkHeader := header.Ethernet(vv.First()[:header.EthernetMinimumSize]) vv.TrimFront(len(linkHeader)) - e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, linkHeader.Type(), tcpip.PacketBuffer{ + e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, linkHeader.Type(), stack.PacketBuffer{ Data: vv, LinkHeader: buffer.View(linkHeader), }) diff --git a/pkg/tcpip/link/muxed/injectable.go b/pkg/tcpip/link/muxed/injectable.go index 445b22c17..f5973066d 100644 --- a/pkg/tcpip/link/muxed/injectable.go +++ b/pkg/tcpip/link/muxed/injectable.go @@ -80,14 +80,14 @@ func (m *InjectableEndpoint) IsAttached() bool { } // InjectInbound implements stack.InjectableLinkEndpoint. -func (m *InjectableEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { +func (m *InjectableEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { m.dispatcher.DeliverNetworkPacket(m, "" /* remote */, "" /* local */, protocol, pkt) } // WritePackets writes outbound packets to the appropriate // LinkInjectableEndpoint based on the RemoteAddress. HandleLocal only works if // r.RemoteAddress has a route registered in this endpoint. -func (m *InjectableEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (m *InjectableEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []stack.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { endpoint, ok := m.routes[r.RemoteAddress] if !ok { return 0, tcpip.ErrNoRoute @@ -98,7 +98,7 @@ func (m *InjectableEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts [ // WritePacket writes outbound packets to the appropriate LinkInjectableEndpoint // based on the RemoteAddress. HandleLocal only works if r.RemoteAddress has a // route registered in this endpoint. -func (m *InjectableEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { +func (m *InjectableEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { if endpoint, ok := m.routes[r.RemoteAddress]; ok { return endpoint.WritePacket(r, gso, protocol, pkt) } diff --git a/pkg/tcpip/link/muxed/injectable_test.go b/pkg/tcpip/link/muxed/injectable_test.go index 63b249837..87c734c1f 100644 --- a/pkg/tcpip/link/muxed/injectable_test.go +++ b/pkg/tcpip/link/muxed/injectable_test.go @@ -50,7 +50,7 @@ func TestInjectableEndpointDispatch(t *testing.T) { hdr.Prepend(1)[0] = 0xFA packetRoute := stack.Route{RemoteAddress: dstIP} - endpoint.WritePacket(&packetRoute, nil /* gso */, ipv4.ProtocolNumber, tcpip.PacketBuffer{ + endpoint.WritePacket(&packetRoute, nil /* gso */, ipv4.ProtocolNumber, stack.PacketBuffer{ Header: hdr, Data: buffer.NewViewFromBytes([]byte{0xFB}).ToVectorisedView(), }) @@ -70,7 +70,7 @@ func TestInjectableEndpointDispatchHdrOnly(t *testing.T) { hdr := buffer.NewPrependable(1) hdr.Prepend(1)[0] = 0xFA packetRoute := stack.Route{RemoteAddress: dstIP} - endpoint.WritePacket(&packetRoute, nil /* gso */, ipv4.ProtocolNumber, tcpip.PacketBuffer{ + endpoint.WritePacket(&packetRoute, nil /* gso */, ipv4.ProtocolNumber, stack.PacketBuffer{ Header: hdr, Data: buffer.NewView(0).ToVectorisedView(), }) diff --git a/pkg/tcpip/link/sharedmem/sharedmem.go b/pkg/tcpip/link/sharedmem/sharedmem.go index 655e537c4..6461d0108 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem.go +++ b/pkg/tcpip/link/sharedmem/sharedmem.go @@ -185,7 +185,7 @@ func (e *endpoint) LinkAddress() tcpip.LinkAddress { // WritePacket writes outbound packets to the file descriptor. If it is not // currently writable, the packet is dropped. -func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { +func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { // Add the ethernet header here. eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize)) pkt.LinkHeader = buffer.View(eth) @@ -214,7 +214,7 @@ func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.Netw } // WritePackets implements stack.LinkEndpoint.WritePackets. -func (e *endpoint) WritePackets(r *stack.Route, _ *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (e *endpoint) WritePackets(r *stack.Route, _ *stack.GSO, pkts []stack.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { panic("not implemented") } @@ -275,7 +275,7 @@ func (e *endpoint) dispatchLoop(d stack.NetworkDispatcher) { // Send packet up the stack. eth := header.Ethernet(b[:header.EthernetMinimumSize]) - d.DeliverNetworkPacket(e, eth.SourceAddress(), eth.DestinationAddress(), eth.Type(), tcpip.PacketBuffer{ + d.DeliverNetworkPacket(e, eth.SourceAddress(), eth.DestinationAddress(), eth.Type(), stack.PacketBuffer{ Data: buffer.View(b[header.EthernetMinimumSize:]).ToVectorisedView(), LinkHeader: buffer.View(eth), }) diff --git a/pkg/tcpip/link/sharedmem/sharedmem_test.go b/pkg/tcpip/link/sharedmem/sharedmem_test.go index 5c729a439..27ea3f531 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem_test.go +++ b/pkg/tcpip/link/sharedmem/sharedmem_test.go @@ -131,7 +131,7 @@ func newTestContext(t *testing.T, mtu, bufferSize uint32, addr tcpip.LinkAddress return c } -func (c *testContext) DeliverNetworkPacket(_ stack.LinkEndpoint, remoteLinkAddr, localLinkAddr tcpip.LinkAddress, proto tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { +func (c *testContext) DeliverNetworkPacket(_ stack.LinkEndpoint, remoteLinkAddr, localLinkAddr tcpip.LinkAddress, proto tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { c.mu.Lock() c.packets = append(c.packets, packetInfo{ addr: remoteLinkAddr, @@ -273,7 +273,7 @@ func TestSimpleSend(t *testing.T) { randomFill(buf) proto := tcpip.NetworkProtocolNumber(rand.Intn(0x10000)) - if err := c.ep.WritePacket(&r, nil /* gso */, proto, tcpip.PacketBuffer{ + if err := c.ep.WritePacket(&r, nil /* gso */, proto, stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != nil { @@ -345,7 +345,7 @@ func TestPreserveSrcAddressInSend(t *testing.T) { hdr := buffer.NewPrependable(header.EthernetMinimumSize) proto := tcpip.NetworkProtocolNumber(rand.Intn(0x10000)) - if err := c.ep.WritePacket(&r, nil /* gso */, proto, tcpip.PacketBuffer{ + if err := c.ep.WritePacket(&r, nil /* gso */, proto, stack.PacketBuffer{ Header: hdr, }); err != nil { t.Fatalf("WritePacket failed: %v", err) @@ -401,7 +401,7 @@ func TestFillTxQueue(t *testing.T) { for i := queuePipeSize / 40; i > 0; i-- { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != nil { @@ -419,7 +419,7 @@ func TestFillTxQueue(t *testing.T) { // Next attempt to write must fail. hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != want { @@ -447,7 +447,7 @@ func TestFillTxQueueAfterBadCompletion(t *testing.T) { // Send two packets so that the id slice has at least two slots. for i := 2; i > 0; i-- { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != nil { @@ -470,7 +470,7 @@ func TestFillTxQueueAfterBadCompletion(t *testing.T) { ids := make(map[uint64]struct{}) for i := queuePipeSize / 40; i > 0; i-- { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != nil { @@ -488,7 +488,7 @@ func TestFillTxQueueAfterBadCompletion(t *testing.T) { // Next attempt to write must fail. hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != want { @@ -514,7 +514,7 @@ func TestFillTxMemory(t *testing.T) { ids := make(map[uint64]struct{}) for i := queueDataSize / bufferSize; i > 0; i-- { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != nil { @@ -533,7 +533,7 @@ func TestFillTxMemory(t *testing.T) { // Next attempt to write must fail. hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }) @@ -561,7 +561,7 @@ func TestFillTxMemoryWithMultiBuffer(t *testing.T) { // until there is only one buffer left. for i := queueDataSize/bufferSize - 1; i > 0; i-- { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != nil { @@ -577,7 +577,7 @@ func TestFillTxMemoryWithMultiBuffer(t *testing.T) { { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) uu := buffer.NewView(bufferSize).ToVectorisedView() - if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ Header: hdr, Data: uu, }); err != want { @@ -588,7 +588,7 @@ func TestFillTxMemoryWithMultiBuffer(t *testing.T) { // Attempt to write the one-buffer packet again. It must succeed. { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, tcpip.PacketBuffer{ + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != nil { diff --git a/pkg/tcpip/link/sniffer/sniffer.go b/pkg/tcpip/link/sniffer/sniffer.go index 3392b7edd..0a6b8945c 100644 --- a/pkg/tcpip/link/sniffer/sniffer.go +++ b/pkg/tcpip/link/sniffer/sniffer.go @@ -123,7 +123,7 @@ func NewWithFile(lower stack.LinkEndpoint, file *os.File, snapLen uint32) (stack // DeliverNetworkPacket implements the stack.NetworkDispatcher interface. It is // called by the link-layer endpoint being wrapped when a packet arrives, and // logs the packet before forwarding to the actual dispatcher. -func (e *endpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { +func (e *endpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { if atomic.LoadUint32(&LogPackets) == 1 && e.file == nil { logPacket("recv", protocol, pkt.Data.First(), nil) } @@ -200,7 +200,7 @@ func (e *endpoint) GSOMaxSize() uint32 { return 0 } -func (e *endpoint) dumpPacket(gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { +func (e *endpoint) dumpPacket(gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { if atomic.LoadUint32(&LogPackets) == 1 && e.file == nil { logPacket("send", protocol, pkt.Header.View(), gso) } @@ -232,7 +232,7 @@ func (e *endpoint) dumpPacket(gso *stack.GSO, protocol tcpip.NetworkProtocolNumb // WritePacket implements the stack.LinkEndpoint interface. It is called by // higher-level protocols to write packets; it just logs the packet and // forwards the request to the lower endpoint. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { e.dumpPacket(gso, protocol, pkt) return e.lower.WritePacket(r, gso, protocol, pkt) } @@ -240,10 +240,10 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne // WritePackets implements the stack.LinkEndpoint interface. It is called by // higher-level protocols to write packets; it just logs the packet and // forwards the request to the lower endpoint. -func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []stack.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { view := pkts[0].Data.ToView() for _, pkt := range pkts { - e.dumpPacket(gso, protocol, tcpip.PacketBuffer{ + e.dumpPacket(gso, protocol, stack.PacketBuffer{ Header: pkt.Header, Data: view[pkt.DataOffset:][:pkt.DataSize].ToVectorisedView(), }) diff --git a/pkg/tcpip/link/tun/device.go b/pkg/tcpip/link/tun/device.go index f6e301304..617446ea2 100644 --- a/pkg/tcpip/link/tun/device.go +++ b/pkg/tcpip/link/tun/device.go @@ -213,7 +213,7 @@ func (d *Device) Write(data []byte) (int64, error) { remote = tcpip.LinkAddress(zeroMAC[:]) } - pkt := tcpip.PacketBuffer{ + pkt := stack.PacketBuffer{ Data: buffer.View(data).ToVectorisedView(), } if ethHdr != nil { diff --git a/pkg/tcpip/link/waitable/waitable.go b/pkg/tcpip/link/waitable/waitable.go index a8de38979..52fe397bf 100644 --- a/pkg/tcpip/link/waitable/waitable.go +++ b/pkg/tcpip/link/waitable/waitable.go @@ -50,7 +50,7 @@ func New(lower stack.LinkEndpoint) *Endpoint { // It is called by the link-layer endpoint being wrapped when a packet arrives, // and only forwards to the actual dispatcher if Wait or WaitDispatch haven't // been called. -func (e *Endpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { +func (e *Endpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { if !e.dispatchGate.Enter() { return } @@ -99,7 +99,7 @@ func (e *Endpoint) LinkAddress() tcpip.LinkAddress { // WritePacket implements stack.LinkEndpoint.WritePacket. It is called by // higher-level protocols to write packets. It only forwards packets to the // lower endpoint if Wait or WaitWrite haven't been called. -func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { +func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { if !e.writeGate.Enter() { return nil } @@ -112,7 +112,7 @@ func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne // WritePackets implements stack.LinkEndpoint.WritePackets. It is called by // higher-level protocols to write packets. It only forwards packets to the // lower endpoint if Wait or WaitWrite haven't been called. -func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []stack.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { if !e.writeGate.Enter() { return len(pkts), nil } diff --git a/pkg/tcpip/link/waitable/waitable_test.go b/pkg/tcpip/link/waitable/waitable_test.go index 31b11a27a..88224e494 100644 --- a/pkg/tcpip/link/waitable/waitable_test.go +++ b/pkg/tcpip/link/waitable/waitable_test.go @@ -35,7 +35,7 @@ type countedEndpoint struct { dispatcher stack.NetworkDispatcher } -func (e *countedEndpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { +func (e *countedEndpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { e.dispatchCount++ } @@ -65,13 +65,13 @@ func (e *countedEndpoint) LinkAddress() tcpip.LinkAddress { return e.linkAddr } -func (e *countedEndpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { +func (e *countedEndpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { e.writeCount++ return nil } // WritePackets implements stack.LinkEndpoint.WritePackets. -func (e *countedEndpoint) WritePackets(r *stack.Route, _ *stack.GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (e *countedEndpoint) WritePackets(r *stack.Route, _ *stack.GSO, pkts []stack.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { e.writeCount += len(pkts) return len(pkts), nil } @@ -89,21 +89,21 @@ func TestWaitWrite(t *testing.T) { wep := New(ep) // Write and check that it goes through. - wep.WritePacket(nil, nil /* gso */, 0, tcpip.PacketBuffer{}) + wep.WritePacket(nil, nil /* gso */, 0, stack.PacketBuffer{}) if want := 1; ep.writeCount != want { t.Fatalf("Unexpected writeCount: got=%v, want=%v", ep.writeCount, want) } // Wait on dispatches, then try to write. It must go through. wep.WaitDispatch() - wep.WritePacket(nil, nil /* gso */, 0, tcpip.PacketBuffer{}) + wep.WritePacket(nil, nil /* gso */, 0, stack.PacketBuffer{}) if want := 2; ep.writeCount != want { t.Fatalf("Unexpected writeCount: got=%v, want=%v", ep.writeCount, want) } // Wait on writes, then try to write. It must not go through. wep.WaitWrite() - wep.WritePacket(nil, nil /* gso */, 0, tcpip.PacketBuffer{}) + wep.WritePacket(nil, nil /* gso */, 0, stack.PacketBuffer{}) if want := 2; ep.writeCount != want { t.Fatalf("Unexpected writeCount: got=%v, want=%v", ep.writeCount, want) } @@ -120,21 +120,21 @@ func TestWaitDispatch(t *testing.T) { } // Dispatch and check that it goes through. - ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, tcpip.PacketBuffer{}) + ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, stack.PacketBuffer{}) if want := 1; ep.dispatchCount != want { t.Fatalf("Unexpected dispatchCount: got=%v, want=%v", ep.dispatchCount, want) } // Wait on writes, then try to dispatch. It must go through. wep.WaitWrite() - ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, tcpip.PacketBuffer{}) + ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, stack.PacketBuffer{}) if want := 2; ep.dispatchCount != want { t.Fatalf("Unexpected dispatchCount: got=%v, want=%v", ep.dispatchCount, want) } // Wait on dispatches, then try to dispatch. It must not go through. wep.WaitDispatch() - ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, tcpip.PacketBuffer{}) + ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, stack.PacketBuffer{}) if want := 2; ep.dispatchCount != want { t.Fatalf("Unexpected dispatchCount: got=%v, want=%v", ep.dispatchCount, want) } diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index e9fcc89a8..255098372 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -79,20 +79,20 @@ func (e *endpoint) MaxHeaderLength() uint16 { func (e *endpoint) Close() {} -func (e *endpoint) WritePacket(*stack.Route, *stack.GSO, stack.NetworkHeaderParams, tcpip.PacketBuffer) *tcpip.Error { +func (e *endpoint) WritePacket(*stack.Route, *stack.GSO, stack.NetworkHeaderParams, stack.PacketBuffer) *tcpip.Error { return tcpip.ErrNotSupported } // WritePackets implements stack.NetworkEndpoint.WritePackets. -func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, []tcpip.PacketBuffer, stack.NetworkHeaderParams) (int, *tcpip.Error) { +func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, []stack.PacketBuffer, stack.NetworkHeaderParams) (int, *tcpip.Error) { return 0, tcpip.ErrNotSupported } -func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt tcpip.PacketBuffer) *tcpip.Error { +func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuffer) *tcpip.Error { return tcpip.ErrNotSupported } -func (e *endpoint) HandlePacket(r *stack.Route, pkt tcpip.PacketBuffer) { +func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { v := pkt.Data.First() h := header.ARP(v) if !h.IsValid() { @@ -113,7 +113,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt tcpip.PacketBuffer) { copy(packet.ProtocolAddressSender(), h.ProtocolAddressTarget()) copy(packet.HardwareAddressTarget(), h.HardwareAddressSender()) copy(packet.ProtocolAddressTarget(), h.ProtocolAddressSender()) - e.linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, tcpip.PacketBuffer{ + e.linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, stack.PacketBuffer{ Header: hdr, }) fallthrough // also fill the cache from requests @@ -167,7 +167,7 @@ func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, linkEP stack. copy(h.ProtocolAddressSender(), localAddr) copy(h.ProtocolAddressTarget(), addr) - return linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, tcpip.PacketBuffer{ + return linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, stack.PacketBuffer{ Header: hdr, }) } diff --git a/pkg/tcpip/network/arp/arp_test.go b/pkg/tcpip/network/arp/arp_test.go index 03cf03b6d..b3e239ac7 100644 --- a/pkg/tcpip/network/arp/arp_test.go +++ b/pkg/tcpip/network/arp/arp_test.go @@ -103,7 +103,7 @@ func TestDirectRequest(t *testing.T) { inject := func(addr tcpip.Address) { copy(h.ProtocolAddressTarget(), addr) - c.linkEP.InjectInbound(arp.ProtocolNumber, tcpip.PacketBuffer{ + c.linkEP.InjectInbound(arp.ProtocolNumber, stack.PacketBuffer{ Data: v.ToVectorisedView(), }) } diff --git a/pkg/tcpip/network/ip_test.go b/pkg/tcpip/network/ip_test.go index f4d78f8c6..4950d69fc 100644 --- a/pkg/tcpip/network/ip_test.go +++ b/pkg/tcpip/network/ip_test.go @@ -96,7 +96,7 @@ func (t *testObject) checkValues(protocol tcpip.TransportProtocolNumber, vv buff // DeliverTransportPacket is called by network endpoints after parsing incoming // packets. This is used by the test object to verify that the results of the // parsing are expected. -func (t *testObject) DeliverTransportPacket(r *stack.Route, protocol tcpip.TransportProtocolNumber, pkt tcpip.PacketBuffer) { +func (t *testObject) DeliverTransportPacket(r *stack.Route, protocol tcpip.TransportProtocolNumber, pkt stack.PacketBuffer) { t.checkValues(protocol, pkt.Data, r.RemoteAddress, r.LocalAddress) t.dataCalls++ } @@ -104,7 +104,7 @@ func (t *testObject) DeliverTransportPacket(r *stack.Route, protocol tcpip.Trans // DeliverTransportControlPacket is called by network endpoints after parsing // incoming control (ICMP) packets. This is used by the test object to verify // that the results of the parsing are expected. -func (t *testObject) DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ stack.ControlType, extra uint32, pkt tcpip.PacketBuffer) { +func (t *testObject) DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ stack.ControlType, extra uint32, pkt stack.PacketBuffer) { t.checkValues(trans, pkt.Data, remote, local) if typ != t.typ { t.t.Errorf("typ = %v, want %v", typ, t.typ) @@ -150,7 +150,7 @@ func (*testObject) Wait() {} // WritePacket is called by network endpoints after producing a packet and // writing it to the link endpoint. This is used by the test object to verify // that the produced packet is as expected. -func (t *testObject) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { +func (t *testObject) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { var prot tcpip.TransportProtocolNumber var srcAddr tcpip.Address var dstAddr tcpip.Address @@ -172,7 +172,7 @@ func (t *testObject) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.Ne } // WritePackets implements stack.LinkEndpoint.WritePackets. -func (t *testObject) WritePackets(_ *stack.Route, _ *stack.GSO, pkt []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (t *testObject) WritePackets(_ *stack.Route, _ *stack.GSO, pkt []stack.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { panic("not implemented") } @@ -246,7 +246,7 @@ func TestIPv4Send(t *testing.T) { if err != nil { t.Fatalf("could not find route: %v", err) } - if err := ep.WritePacket(&r, nil /* gso */, stack.NetworkHeaderParams{Protocol: 123, TTL: 123, TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + if err := ep.WritePacket(&r, nil /* gso */, stack.NetworkHeaderParams{Protocol: 123, TTL: 123, TOS: stack.DefaultTOS}, stack.PacketBuffer{ Header: hdr, Data: payload.ToVectorisedView(), }); err != nil { @@ -289,7 +289,7 @@ func TestIPv4Receive(t *testing.T) { if err != nil { t.Fatalf("could not find route: %v", err) } - ep.HandlePacket(&r, tcpip.PacketBuffer{ + ep.HandlePacket(&r, stack.PacketBuffer{ Data: view.ToVectorisedView(), }) if o.dataCalls != 1 { @@ -379,7 +379,7 @@ func TestIPv4ReceiveControl(t *testing.T) { o.extra = c.expectedExtra vv := view[:len(view)-c.trunc].ToVectorisedView() - ep.HandlePacket(&r, tcpip.PacketBuffer{ + ep.HandlePacket(&r, stack.PacketBuffer{ Data: vv, }) if want := c.expectedCount; o.controlCalls != want { @@ -444,7 +444,7 @@ func TestIPv4FragmentationReceive(t *testing.T) { } // Send first segment. - ep.HandlePacket(&r, tcpip.PacketBuffer{ + ep.HandlePacket(&r, stack.PacketBuffer{ Data: frag1.ToVectorisedView(), }) if o.dataCalls != 0 { @@ -452,7 +452,7 @@ func TestIPv4FragmentationReceive(t *testing.T) { } // Send second segment. - ep.HandlePacket(&r, tcpip.PacketBuffer{ + ep.HandlePacket(&r, stack.PacketBuffer{ Data: frag2.ToVectorisedView(), }) if o.dataCalls != 1 { @@ -487,7 +487,7 @@ func TestIPv6Send(t *testing.T) { if err != nil { t.Fatalf("could not find route: %v", err) } - if err := ep.WritePacket(&r, nil /* gso */, stack.NetworkHeaderParams{Protocol: 123, TTL: 123, TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + if err := ep.WritePacket(&r, nil /* gso */, stack.NetworkHeaderParams{Protocol: 123, TTL: 123, TOS: stack.DefaultTOS}, stack.PacketBuffer{ Header: hdr, Data: payload.ToVectorisedView(), }); err != nil { @@ -530,7 +530,7 @@ func TestIPv6Receive(t *testing.T) { t.Fatalf("could not find route: %v", err) } - ep.HandlePacket(&r, tcpip.PacketBuffer{ + ep.HandlePacket(&r, stack.PacketBuffer{ Data: view.ToVectorisedView(), }) if o.dataCalls != 1 { @@ -644,7 +644,7 @@ func TestIPv6ReceiveControl(t *testing.T) { // Set ICMPv6 checksum. icmp.SetChecksum(header.ICMPv6Checksum(icmp, outerSrcAddr, localIpv6Addr, buffer.VectorisedView{})) - ep.HandlePacket(&r, tcpip.PacketBuffer{ + ep.HandlePacket(&r, stack.PacketBuffer{ Data: view[:len(view)-c.trunc].ToVectorisedView(), }) if want := c.expectedCount; o.controlCalls != want { diff --git a/pkg/tcpip/network/ipv4/BUILD b/pkg/tcpip/network/ipv4/BUILD index 0fef2b1f1..880ea7de2 100644 --- a/pkg/tcpip/network/ipv4/BUILD +++ b/pkg/tcpip/network/ipv4/BUILD @@ -13,7 +13,6 @@ go_library( "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/header", - "//pkg/tcpip/iptables", "//pkg/tcpip/network/fragmentation", "//pkg/tcpip/network/hash", "//pkg/tcpip/stack", diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go index 32bf39e43..c4bf1ba5c 100644 --- a/pkg/tcpip/network/ipv4/icmp.go +++ b/pkg/tcpip/network/ipv4/icmp.go @@ -15,7 +15,6 @@ package ipv4 import ( - "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" @@ -25,7 +24,7 @@ import ( // the original packet that caused the ICMP one to be sent. This information is // used to find out which transport endpoint must be notified about the ICMP // packet. -func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt tcpip.PacketBuffer) { +func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack.PacketBuffer) { h := header.IPv4(pkt.Data.First()) // We don't use IsValid() here because ICMP only requires that the IP @@ -53,7 +52,7 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt tcpip. e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, h.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) } -func (e *endpoint) handleICMP(r *stack.Route, pkt tcpip.PacketBuffer) { +func (e *endpoint) handleICMP(r *stack.Route, pkt stack.PacketBuffer) { stats := r.Stats() received := stats.ICMP.V4PacketsReceived v := pkt.Data.First() @@ -85,7 +84,7 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt tcpip.PacketBuffer) { // It's possible that a raw socket expects to receive this. h.SetChecksum(wantChecksum) - e.dispatcher.DeliverTransportPacket(r, header.ICMPv4ProtocolNumber, tcpip.PacketBuffer{ + e.dispatcher.DeliverTransportPacket(r, header.ICMPv4ProtocolNumber, stack.PacketBuffer{ Data: pkt.Data.Clone(nil), NetworkHeader: append(buffer.View(nil), pkt.NetworkHeader...), }) @@ -99,7 +98,7 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt tcpip.PacketBuffer) { pkt.SetChecksum(0) pkt.SetChecksum(^header.Checksum(pkt, header.ChecksumVV(vv, 0))) sent := stats.ICMP.V4PacketsSent - if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, stack.PacketBuffer{ Header: hdr, Data: vv, TransportHeader: buffer.View(pkt), diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index 4f1742938..b3ee6000e 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -26,7 +26,6 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" - "gvisor.dev/gvisor/pkg/tcpip/iptables" "gvisor.dev/gvisor/pkg/tcpip/network/fragmentation" "gvisor.dev/gvisor/pkg/tcpip/network/hash" "gvisor.dev/gvisor/pkg/tcpip/stack" @@ -125,7 +124,7 @@ func (e *endpoint) GSOMaxSize() uint32 { // packet's stated length matches the length of the header+payload. mtu // includes the IP header and options. This does not support the DontFragment // IP flag. -func (e *endpoint) writePacketFragments(r *stack.Route, gso *stack.GSO, mtu int, pkt tcpip.PacketBuffer) *tcpip.Error { +func (e *endpoint) writePacketFragments(r *stack.Route, gso *stack.GSO, mtu int, pkt stack.PacketBuffer) *tcpip.Error { // This packet is too big, it needs to be fragmented. ip := header.IPv4(pkt.Header.View()) flags := ip.Flags() @@ -165,7 +164,7 @@ func (e *endpoint) writePacketFragments(r *stack.Route, gso *stack.GSO, mtu int, if i > 0 { newPayload := pkt.Data.Clone(nil) newPayload.CapLength(innerMTU) - if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, tcpip.PacketBuffer{ + if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, stack.PacketBuffer{ Header: pkt.Header, Data: newPayload, NetworkHeader: buffer.View(h), @@ -184,7 +183,7 @@ func (e *endpoint) writePacketFragments(r *stack.Route, gso *stack.GSO, mtu int, newPayload := pkt.Data.Clone(nil) newPayloadLength := outerMTU - pkt.Header.UsedLength() newPayload.CapLength(newPayloadLength) - if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, tcpip.PacketBuffer{ + if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, stack.PacketBuffer{ Header: pkt.Header, Data: newPayload, NetworkHeader: buffer.View(h), @@ -198,7 +197,7 @@ func (e *endpoint) writePacketFragments(r *stack.Route, gso *stack.GSO, mtu int, startOfHdr := pkt.Header startOfHdr.TrimBack(pkt.Header.UsedLength() - outerMTU) emptyVV := buffer.NewVectorisedView(0, []buffer.View{}) - if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, tcpip.PacketBuffer{ + if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, stack.PacketBuffer{ Header: startOfHdr, Data: emptyVV, NetworkHeader: buffer.View(h), @@ -241,7 +240,7 @@ func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadS } // WritePacket writes a packet to the given destination address and protocol. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, pkt tcpip.PacketBuffer) *tcpip.Error { +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, pkt stack.PacketBuffer) *tcpip.Error { ip := e.addIPHeader(r, &pkt.Header, pkt.Data.Size(), params) pkt.NetworkHeader = buffer.View(ip) @@ -253,7 +252,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw views = append(views, pkt.Data.Views()...) loopedR := r.MakeLoopedRoute() - e.HandlePacket(&loopedR, tcpip.PacketBuffer{ + e.HandlePacket(&loopedR, stack.PacketBuffer{ Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views), }) @@ -273,7 +272,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw } // WritePackets implements stack.NetworkEndpoint.WritePackets. -func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, params stack.NetworkHeaderParams) (int, *tcpip.Error) { +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []stack.PacketBuffer, params stack.NetworkHeaderParams) (int, *tcpip.Error) { if r.Loop&stack.PacketLoop != 0 { panic("multiple packets in local loop") } @@ -292,7 +291,7 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.Pac // WriteHeaderIncludedPacket writes a packet already containing a network // header through the given route. -func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt tcpip.PacketBuffer) *tcpip.Error { +func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuffer) *tcpip.Error { // The packet already has an IP header, but there are a few required // checks. ip := header.IPv4(pkt.Data.First()) @@ -344,7 +343,7 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt tcpip.PacketBuf // HandlePacket is called by the link layer when new ipv4 packets arrive for // this endpoint. -func (e *endpoint) HandlePacket(r *stack.Route, pkt tcpip.PacketBuffer) { +func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { headerView := pkt.Data.First() h := header.IPv4(headerView) if !h.IsValid(pkt.Data.Size()) { @@ -361,7 +360,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt tcpip.PacketBuffer) { // iptables filtering. All packets that reach here are intended for // this machine and will not be forwarded. ipt := e.stack.IPTables() - if ok := ipt.Check(iptables.Input, pkt); !ok { + if ok := ipt.Check(stack.Input, pkt); !ok { // iptables is telling us to drop the packet. return } diff --git a/pkg/tcpip/network/ipv4/ipv4_test.go b/pkg/tcpip/network/ipv4/ipv4_test.go index e900f1b45..5a864d832 100644 --- a/pkg/tcpip/network/ipv4/ipv4_test.go +++ b/pkg/tcpip/network/ipv4/ipv4_test.go @@ -113,7 +113,7 @@ func makeHdrAndPayload(hdrLength int, extraLength int, viewSizes []int) (buffer. // comparePayloads compared the contents of all the packets against the contents // of the source packet. -func compareFragments(t *testing.T, packets []tcpip.PacketBuffer, sourcePacketInfo tcpip.PacketBuffer, mtu uint32) { +func compareFragments(t *testing.T, packets []stack.PacketBuffer, sourcePacketInfo stack.PacketBuffer, mtu uint32) { t.Helper() // Make a complete array of the sourcePacketInfo packet. source := header.IPv4(packets[0].Header.View()[:header.IPv4MinimumSize]) @@ -173,7 +173,7 @@ func compareFragments(t *testing.T, packets []tcpip.PacketBuffer, sourcePacketIn type errorChannel struct { *channel.Endpoint - Ch chan tcpip.PacketBuffer + Ch chan stack.PacketBuffer packetCollectorErrors []*tcpip.Error } @@ -183,7 +183,7 @@ type errorChannel struct { func newErrorChannel(size int, mtu uint32, linkAddr tcpip.LinkAddress, packetCollectorErrors []*tcpip.Error) *errorChannel { return &errorChannel{ Endpoint: channel.New(size, mtu, linkAddr), - Ch: make(chan tcpip.PacketBuffer, size), + Ch: make(chan stack.PacketBuffer, size), packetCollectorErrors: packetCollectorErrors, } } @@ -202,7 +202,7 @@ func (e *errorChannel) Drain() int { } // WritePacket stores outbound packets into the channel. -func (e *errorChannel) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { +func (e *errorChannel) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { select { case e.Ch <- pkt: default: @@ -281,13 +281,13 @@ func TestFragmentation(t *testing.T) { for _, ft := range fragTests { t.Run(ft.description, func(t *testing.T) { hdr, payload := makeHdrAndPayload(ft.hdrLength, ft.extraLength, ft.payloadViewsSizes) - source := tcpip.PacketBuffer{ + source := stack.PacketBuffer{ Header: hdr, // Save the source payload because WritePacket will modify it. Data: payload.Clone(nil), } c := buildContext(t, nil, ft.mtu) - err := c.Route.WritePacket(ft.gso, stack.NetworkHeaderParams{Protocol: tcp.ProtocolNumber, TTL: 42, TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + err := c.Route.WritePacket(ft.gso, stack.NetworkHeaderParams{Protocol: tcp.ProtocolNumber, TTL: 42, TOS: stack.DefaultTOS}, stack.PacketBuffer{ Header: hdr, Data: payload, }) @@ -295,7 +295,7 @@ func TestFragmentation(t *testing.T) { t.Errorf("err got %v, want %v", err, nil) } - var results []tcpip.PacketBuffer + var results []stack.PacketBuffer L: for { select { @@ -337,7 +337,7 @@ func TestFragmentationErrors(t *testing.T) { t.Run(ft.description, func(t *testing.T) { hdr, payload := makeHdrAndPayload(ft.hdrLength, header.IPv4MinimumSize, ft.payloadViewsSizes) c := buildContext(t, ft.packetCollectorErrors, ft.mtu) - err := c.Route.WritePacket(&stack.GSO{}, stack.NetworkHeaderParams{Protocol: tcp.ProtocolNumber, TTL: 42, TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + err := c.Route.WritePacket(&stack.GSO{}, stack.NetworkHeaderParams{Protocol: tcp.ProtocolNumber, TTL: 42, TOS: stack.DefaultTOS}, stack.PacketBuffer{ Header: hdr, Data: payload, }) @@ -459,7 +459,7 @@ func TestInvalidFragments(t *testing.T) { s.CreateNIC(nicID, sniffer.New(ep)) for _, pkt := range tc.packets { - ep.InjectLinkAddr(header.IPv4ProtocolNumber, remoteLinkAddr, tcpip.PacketBuffer{ + ep.InjectLinkAddr(header.IPv4ProtocolNumber, remoteLinkAddr, stack.PacketBuffer{ Data: buffer.NewVectorisedView(len(pkt), []buffer.View{pkt}), }) } diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index 45dc757c7..8640feffc 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -27,7 +27,7 @@ import ( // the original packet that caused the ICMP one to be sent. This information is // used to find out which transport endpoint must be notified about the ICMP // packet. -func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt tcpip.PacketBuffer) { +func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack.PacketBuffer) { h := header.IPv6(pkt.Data.First()) // We don't use IsValid() here because ICMP only requires that up to @@ -62,7 +62,7 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt tcpip. e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, h.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) } -func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt tcpip.PacketBuffer) { +func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.PacketBuffer) { stats := r.Stats().ICMP sent := stats.V6PacketsSent received := stats.V6PacketsReceived @@ -243,7 +243,7 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt tcpip.P // // The IP Hop Limit field has a value of 255, i.e., the packet // could not possibly have been forwarded by a router. - if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, TOS: stack.DefaultTOS}, stack.PacketBuffer{ Header: hdr, }); err != nil { sent.Dropped.Increment() @@ -330,7 +330,7 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt tcpip.P copy(packet, h) packet.SetType(header.ICMPv6EchoReply) packet.SetChecksum(header.ICMPv6Checksum(packet, r.LocalAddress, r.RemoteAddress, pkt.Data)) - if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, stack.PacketBuffer{ Header: hdr, Data: pkt.Data, }); err != nil { @@ -463,7 +463,7 @@ func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, linkEP stack. }) // TODO(stijlist): count this in ICMP stats. - return linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, tcpip.PacketBuffer{ + return linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, stack.PacketBuffer{ Header: hdr, }) } diff --git a/pkg/tcpip/network/ipv6/icmp_test.go b/pkg/tcpip/network/ipv6/icmp_test.go index 50c4b6474..bae09ed94 100644 --- a/pkg/tcpip/network/ipv6/icmp_test.go +++ b/pkg/tcpip/network/ipv6/icmp_test.go @@ -56,7 +56,7 @@ func (*stubLinkEndpoint) LinkAddress() tcpip.LinkAddress { return "" } -func (*stubLinkEndpoint) WritePacket(*stack.Route, *stack.GSO, tcpip.NetworkProtocolNumber, tcpip.PacketBuffer) *tcpip.Error { +func (*stubLinkEndpoint) WritePacket(*stack.Route, *stack.GSO, tcpip.NetworkProtocolNumber, stack.PacketBuffer) *tcpip.Error { return nil } @@ -66,7 +66,7 @@ type stubDispatcher struct { stack.TransportDispatcher } -func (*stubDispatcher) DeliverTransportPacket(*stack.Route, tcpip.TransportProtocolNumber, tcpip.PacketBuffer) { +func (*stubDispatcher) DeliverTransportPacket(*stack.Route, tcpip.TransportProtocolNumber, stack.PacketBuffer) { } type stubLinkAddressCache struct { @@ -187,7 +187,7 @@ func TestICMPCounts(t *testing.T) { SrcAddr: r.LocalAddress, DstAddr: r.RemoteAddress, }) - ep.HandlePacket(&r, tcpip.PacketBuffer{ + ep.HandlePacket(&r, stack.PacketBuffer{ Data: hdr.View().ToVectorisedView(), }) } @@ -326,7 +326,7 @@ func routeICMPv6Packet(t *testing.T, args routeArgs, fn func(*testing.T, header. views := []buffer.View{pi.Pkt.Header.View(), pi.Pkt.Data.ToView()} size := pi.Pkt.Header.UsedLength() + pi.Pkt.Data.Size() vv := buffer.NewVectorisedView(size, views) - args.dst.InjectLinkAddr(pi.Proto, args.dst.LinkAddress(), tcpip.PacketBuffer{ + args.dst.InjectLinkAddr(pi.Proto, args.dst.LinkAddress(), stack.PacketBuffer{ Data: vv, }) } @@ -561,7 +561,7 @@ func TestICMPChecksumValidationSimple(t *testing.T) { SrcAddr: lladdr1, DstAddr: lladdr0, }) - e.InjectInbound(ProtocolNumber, tcpip.PacketBuffer{ + e.InjectInbound(ProtocolNumber, stack.PacketBuffer{ Data: hdr.View().ToVectorisedView(), }) } @@ -738,7 +738,7 @@ func TestICMPChecksumValidationWithPayload(t *testing.T) { SrcAddr: lladdr1, DstAddr: lladdr0, }) - e.InjectInbound(ProtocolNumber, tcpip.PacketBuffer{ + e.InjectInbound(ProtocolNumber, stack.PacketBuffer{ Data: hdr.View().ToVectorisedView(), }) } @@ -916,7 +916,7 @@ func TestICMPChecksumValidationWithPayloadMultipleViews(t *testing.T) { SrcAddr: lladdr1, DstAddr: lladdr0, }) - e.InjectInbound(ProtocolNumber, tcpip.PacketBuffer{ + e.InjectInbound(ProtocolNumber, stack.PacketBuffer{ Data: buffer.NewVectorisedView(header.IPv6MinimumSize+size+payloadSize, []buffer.View{hdr.View(), payload}), }) } diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index 9aef5234b..29e597002 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -112,7 +112,7 @@ func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadS } // WritePacket writes a packet to the given destination address and protocol. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, pkt tcpip.PacketBuffer) *tcpip.Error { +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, pkt stack.PacketBuffer) *tcpip.Error { ip := e.addIPHeader(r, &pkt.Header, pkt.Data.Size(), params) pkt.NetworkHeader = buffer.View(ip) @@ -124,7 +124,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw views = append(views, pkt.Data.Views()...) loopedR := r.MakeLoopedRoute() - e.HandlePacket(&loopedR, tcpip.PacketBuffer{ + e.HandlePacket(&loopedR, stack.PacketBuffer{ Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views), }) @@ -139,7 +139,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw } // WritePackets implements stack.LinkEndpoint.WritePackets. -func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, params stack.NetworkHeaderParams) (int, *tcpip.Error) { +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []stack.PacketBuffer, params stack.NetworkHeaderParams) (int, *tcpip.Error) { if r.Loop&stack.PacketLoop != 0 { panic("not implemented") } @@ -161,14 +161,14 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.Pac // WriteHeaderIncludedPacker implements stack.NetworkEndpoint. It is not yet // supported by IPv6. -func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt tcpip.PacketBuffer) *tcpip.Error { +func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuffer) *tcpip.Error { // TODO(b/146666412): Support IPv6 header-included packets. return tcpip.ErrNotSupported } // HandlePacket is called by the link layer when new ipv6 packets arrive for // this endpoint. -func (e *endpoint) HandlePacket(r *stack.Route, pkt tcpip.PacketBuffer) { +func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { headerView := pkt.Data.First() h := header.IPv6(headerView) if !h.IsValid(pkt.Data.Size()) { diff --git a/pkg/tcpip/network/ipv6/ipv6_test.go b/pkg/tcpip/network/ipv6/ipv6_test.go index 1cbfa7278..ed98ef22a 100644 --- a/pkg/tcpip/network/ipv6/ipv6_test.go +++ b/pkg/tcpip/network/ipv6/ipv6_test.go @@ -55,7 +55,7 @@ func testReceiveICMP(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst DstAddr: dst, }) - e.InjectInbound(ProtocolNumber, tcpip.PacketBuffer{ + e.InjectInbound(ProtocolNumber, stack.PacketBuffer{ Data: hdr.View().ToVectorisedView(), }) @@ -113,7 +113,7 @@ func testReceiveUDP(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst DstAddr: dst, }) - e.InjectInbound(ProtocolNumber, tcpip.PacketBuffer{ + e.InjectInbound(ProtocolNumber, stack.PacketBuffer{ Data: hdr.View().ToVectorisedView(), }) diff --git a/pkg/tcpip/network/ipv6/ndp_test.go b/pkg/tcpip/network/ipv6/ndp_test.go index c9395de52..f924ed9e1 100644 --- a/pkg/tcpip/network/ipv6/ndp_test.go +++ b/pkg/tcpip/network/ipv6/ndp_test.go @@ -135,7 +135,7 @@ func TestNeighorSolicitationWithSourceLinkLayerOption(t *testing.T) { t.Fatalf("got invalid = %d, want = 0", got) } - e.InjectInbound(ProtocolNumber, tcpip.PacketBuffer{ + e.InjectInbound(ProtocolNumber, stack.PacketBuffer{ Data: hdr.View().ToVectorisedView(), }) @@ -238,7 +238,7 @@ func TestNeighorAdvertisementWithTargetLinkLayerOption(t *testing.T) { t.Fatalf("got invalid = %d, want = 0", got) } - e.InjectInbound(ProtocolNumber, tcpip.PacketBuffer{ + e.InjectInbound(ProtocolNumber, stack.PacketBuffer{ Data: hdr.View().ToVectorisedView(), }) @@ -304,7 +304,7 @@ func TestHopLimitValidation(t *testing.T) { SrcAddr: r.LocalAddress, DstAddr: r.RemoteAddress, }) - ep.HandlePacket(r, tcpip.PacketBuffer{ + ep.HandlePacket(r, stack.PacketBuffer{ Data: hdr.View().ToVectorisedView(), }) } @@ -588,7 +588,7 @@ func TestRouterAdvertValidation(t *testing.T) { t.Fatalf("got rxRA = %d, want = 0", got) } - e.InjectInbound(header.IPv6ProtocolNumber, tcpip.PacketBuffer{ + e.InjectInbound(header.IPv6ProtocolNumber, stack.PacketBuffer{ Data: hdr.View().ToVectorisedView(), }) diff --git a/pkg/tcpip/packet_buffer.go b/pkg/tcpip/packet_buffer.go deleted file mode 100644 index ab24372e7..000000000 --- a/pkg/tcpip/packet_buffer.go +++ /dev/null @@ -1,67 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at // -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package tcpip - -import "gvisor.dev/gvisor/pkg/tcpip/buffer" - -// A PacketBuffer contains all the data of a network packet. -// -// As a PacketBuffer traverses up the stack, it may be necessary to pass it to -// multiple endpoints. Clone() should be called in such cases so that -// modifications to the Data field do not affect other copies. -// -// +stateify savable -type PacketBuffer struct { - // Data holds the payload of the packet. For inbound packets, it also - // holds the headers, which are consumed as the packet moves up the - // stack. Headers are guaranteed not to be split across views. - // - // The bytes backing Data are immutable, but Data itself may be trimmed - // or otherwise modified. - Data buffer.VectorisedView - - // DataOffset is used for GSO output. It is the offset into the Data - // field where the payload of this packet starts. - DataOffset int - - // DataSize is used for GSO output. It is the size of this packet's - // payload. - DataSize int - - // Header holds the headers of outbound packets. As a packet is passed - // down the stack, each layer adds to Header. - Header buffer.Prependable - - // These fields are used by both inbound and outbound packets. They - // typically overlap with the Data and Header fields. - // - // The bytes backing these views are immutable. Each field may be nil - // if either it has not been set yet or no such header exists (e.g. - // packets sent via loopback may not have a link header). - // - // These fields may be Views into other slices (either Data or Header). - // SR dosen't support this, so deep copies are necessary in some cases. - LinkHeader buffer.View - NetworkHeader buffer.View - TransportHeader buffer.View -} - -// Clone makes a copy of pk. It clones the Data field, which creates a new -// VectorisedView but does not deep copy the underlying bytes. -// -// Clone also does not deep copy any of its other fields. -func (pk PacketBuffer) Clone() PacketBuffer { - pk.Data = pk.Data.Clone(nil) - return pk -} diff --git a/pkg/tcpip/packet_buffer_state.go b/pkg/tcpip/packet_buffer_state.go deleted file mode 100644 index ad3cc24fa..000000000 --- a/pkg/tcpip/packet_buffer_state.go +++ /dev/null @@ -1,27 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package tcpip - -import "gvisor.dev/gvisor/pkg/tcpip/buffer" - -// beforeSave is invoked by stateify. -func (pk *PacketBuffer) beforeSave() { - // Non-Data fields may be slices of the Data field. This causes - // problems for SR, so during save we make each header independent. - pk.Header = pk.Header.DeepCopy() - pk.LinkHeader = append(buffer.View(nil), pk.LinkHeader...) - pk.NetworkHeader = append(buffer.View(nil), pk.NetworkHeader...) - pk.TransportHeader = append(buffer.View(nil), pk.TransportHeader...) -} diff --git a/pkg/tcpip/stack/BUILD b/pkg/tcpip/stack/BUILD index 6c029b2fb..7a43a1d4e 100644 --- a/pkg/tcpip/stack/BUILD +++ b/pkg/tcpip/stack/BUILD @@ -21,10 +21,15 @@ go_library( "dhcpv6configurationfromndpra_string.go", "forwarder.go", "icmp_rate_limit.go", + "iptables.go", + "iptables_targets.go", + "iptables_types.go", "linkaddrcache.go", "linkaddrentry_list.go", "ndp.go", "nic.go", + "packet_buffer.go", + "packet_buffer_state.go", "registration.go", "route.go", "stack.go", @@ -34,6 +39,7 @@ go_library( visibility = ["//visibility:public"], deps = [ "//pkg/ilist", + "//pkg/log", "//pkg/rand", "//pkg/sleep", "//pkg/sync", @@ -41,7 +47,6 @@ go_library( "//pkg/tcpip/buffer", "//pkg/tcpip/hash/jenkins", "//pkg/tcpip/header", - "//pkg/tcpip/iptables", "//pkg/tcpip/ports", "//pkg/tcpip/seqnum", "//pkg/waiter", @@ -65,7 +70,6 @@ go_test( "//pkg/tcpip/buffer", "//pkg/tcpip/checker", "//pkg/tcpip/header", - "//pkg/tcpip/iptables", "//pkg/tcpip/link/channel", "//pkg/tcpip/link/loopback", "//pkg/tcpip/network/ipv4", diff --git a/pkg/tcpip/stack/forwarder.go b/pkg/tcpip/stack/forwarder.go index 631953935..6b64cd37f 100644 --- a/pkg/tcpip/stack/forwarder.go +++ b/pkg/tcpip/stack/forwarder.go @@ -32,7 +32,7 @@ type pendingPacket struct { nic *NIC route *Route proto tcpip.NetworkProtocolNumber - pkt tcpip.PacketBuffer + pkt PacketBuffer } type forwardQueue struct { @@ -50,7 +50,7 @@ func newForwardQueue() *forwardQueue { return &forwardQueue{packets: make(map[<-chan struct{}][]*pendingPacket)} } -func (f *forwardQueue) enqueue(ch <-chan struct{}, n *NIC, r *Route, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { +func (f *forwardQueue) enqueue(ch <-chan struct{}, n *NIC, r *Route, protocol tcpip.NetworkProtocolNumber, pkt PacketBuffer) { shouldWait := false f.Lock() diff --git a/pkg/tcpip/stack/forwarder_test.go b/pkg/tcpip/stack/forwarder_test.go index 321b7524d..c45c43d21 100644 --- a/pkg/tcpip/stack/forwarder_test.go +++ b/pkg/tcpip/stack/forwarder_test.go @@ -68,7 +68,7 @@ func (f *fwdTestNetworkEndpoint) ID() *NetworkEndpointID { return &f.id } -func (f *fwdTestNetworkEndpoint) HandlePacket(r *Route, pkt tcpip.PacketBuffer) { +func (f *fwdTestNetworkEndpoint) HandlePacket(r *Route, pkt PacketBuffer) { // Consume the network header. b := pkt.Data.First() pkt.Data.TrimFront(fwdTestNetHeaderLen) @@ -89,7 +89,7 @@ func (f *fwdTestNetworkEndpoint) Capabilities() LinkEndpointCapabilities { return f.ep.Capabilities() } -func (f *fwdTestNetworkEndpoint) WritePacket(r *Route, gso *GSO, params NetworkHeaderParams, pkt tcpip.PacketBuffer) *tcpip.Error { +func (f *fwdTestNetworkEndpoint) WritePacket(r *Route, gso *GSO, params NetworkHeaderParams, pkt PacketBuffer) *tcpip.Error { // Add the protocol's header to the packet and send it to the link // endpoint. b := pkt.Header.Prepend(fwdTestNetHeaderLen) @@ -101,11 +101,11 @@ func (f *fwdTestNetworkEndpoint) WritePacket(r *Route, gso *GSO, params NetworkH } // WritePackets implements LinkEndpoint.WritePackets. -func (f *fwdTestNetworkEndpoint) WritePackets(r *Route, gso *GSO, pkts []tcpip.PacketBuffer, params NetworkHeaderParams) (int, *tcpip.Error) { +func (f *fwdTestNetworkEndpoint) WritePackets(r *Route, gso *GSO, pkts []PacketBuffer, params NetworkHeaderParams) (int, *tcpip.Error) { panic("not implemented") } -func (*fwdTestNetworkEndpoint) WriteHeaderIncludedPacket(r *Route, pkt tcpip.PacketBuffer) *tcpip.Error { +func (*fwdTestNetworkEndpoint) WriteHeaderIncludedPacket(r *Route, pkt PacketBuffer) *tcpip.Error { return tcpip.ErrNotSupported } @@ -183,7 +183,7 @@ func (f *fwdTestNetworkProtocol) LinkAddressProtocol() tcpip.NetworkProtocolNumb type fwdTestPacketInfo struct { RemoteLinkAddress tcpip.LinkAddress LocalLinkAddress tcpip.LinkAddress - Pkt tcpip.PacketBuffer + Pkt PacketBuffer } type fwdTestLinkEndpoint struct { @@ -196,12 +196,12 @@ type fwdTestLinkEndpoint struct { } // InjectInbound injects an inbound packet. -func (e *fwdTestLinkEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { +func (e *fwdTestLinkEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt PacketBuffer) { e.InjectLinkAddr(protocol, "", pkt) } // InjectLinkAddr injects an inbound packet with a remote link address. -func (e *fwdTestLinkEndpoint) InjectLinkAddr(protocol tcpip.NetworkProtocolNumber, remote tcpip.LinkAddress, pkt tcpip.PacketBuffer) { +func (e *fwdTestLinkEndpoint) InjectLinkAddr(protocol tcpip.NetworkProtocolNumber, remote tcpip.LinkAddress, pkt PacketBuffer) { e.dispatcher.DeliverNetworkPacket(e, remote, "" /* local */, protocol, pkt) } @@ -244,7 +244,7 @@ func (e *fwdTestLinkEndpoint) LinkAddress() tcpip.LinkAddress { return e.linkAddr } -func (e fwdTestLinkEndpoint) WritePacket(r *Route, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error { +func (e fwdTestLinkEndpoint) WritePacket(r *Route, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkt PacketBuffer) *tcpip.Error { p := fwdTestPacketInfo{ RemoteLinkAddress: r.RemoteLinkAddress, LocalLinkAddress: r.LocalLinkAddress, @@ -260,7 +260,7 @@ func (e fwdTestLinkEndpoint) WritePacket(r *Route, gso *GSO, protocol tcpip.Netw } // WritePackets stores outbound packets into the channel. -func (e *fwdTestLinkEndpoint) WritePackets(r *Route, gso *GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (e *fwdTestLinkEndpoint) WritePackets(r *Route, gso *GSO, pkts []PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { n := 0 for _, pkt := range pkts { e.WritePacket(r, gso, protocol, pkt) @@ -273,7 +273,7 @@ func (e *fwdTestLinkEndpoint) WritePackets(r *Route, gso *GSO, pkts []tcpip.Pack // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. func (e *fwdTestLinkEndpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { p := fwdTestPacketInfo{ - Pkt: tcpip.PacketBuffer{Data: vv}, + Pkt: PacketBuffer{Data: vv}, } select { @@ -355,7 +355,7 @@ func TestForwardingWithStaticResolver(t *testing.T) { // forwarded to NIC 2. buf := buffer.NewView(30) buf[0] = 3 - ep1.InjectInbound(fwdTestNetNumber, tcpip.PacketBuffer{ + ep1.InjectInbound(fwdTestNetNumber, PacketBuffer{ Data: buf.ToVectorisedView(), }) @@ -392,7 +392,7 @@ func TestForwardingWithFakeResolver(t *testing.T) { // forwarded to NIC 2. buf := buffer.NewView(30) buf[0] = 3 - ep1.InjectInbound(fwdTestNetNumber, tcpip.PacketBuffer{ + ep1.InjectInbound(fwdTestNetNumber, PacketBuffer{ Data: buf.ToVectorisedView(), }) @@ -423,7 +423,7 @@ func TestForwardingWithNoResolver(t *testing.T) { // forwarded to NIC 2. buf := buffer.NewView(30) buf[0] = 3 - ep1.InjectInbound(fwdTestNetNumber, tcpip.PacketBuffer{ + ep1.InjectInbound(fwdTestNetNumber, PacketBuffer{ Data: buf.ToVectorisedView(), }) @@ -453,7 +453,7 @@ func TestForwardingWithFakeResolverPartialTimeout(t *testing.T) { // not be forwarded. buf := buffer.NewView(30) buf[0] = 4 - ep1.InjectInbound(fwdTestNetNumber, tcpip.PacketBuffer{ + ep1.InjectInbound(fwdTestNetNumber, PacketBuffer{ Data: buf.ToVectorisedView(), }) @@ -461,7 +461,7 @@ func TestForwardingWithFakeResolverPartialTimeout(t *testing.T) { // forwarded to NIC 2. buf = buffer.NewView(30) buf[0] = 3 - ep1.InjectInbound(fwdTestNetNumber, tcpip.PacketBuffer{ + ep1.InjectInbound(fwdTestNetNumber, PacketBuffer{ Data: buf.ToVectorisedView(), }) @@ -503,7 +503,7 @@ func TestForwardingWithFakeResolverTwoPackets(t *testing.T) { for i := 0; i < 2; i++ { buf := buffer.NewView(30) buf[0] = 3 - ep1.InjectInbound(fwdTestNetNumber, tcpip.PacketBuffer{ + ep1.InjectInbound(fwdTestNetNumber, PacketBuffer{ Data: buf.ToVectorisedView(), }) } @@ -550,7 +550,7 @@ func TestForwardingWithFakeResolverManyPackets(t *testing.T) { buf[0] = 3 // Set the packet sequence number. binary.BigEndian.PutUint16(buf[fwdTestNetHeaderLen:], uint16(i)) - ep1.InjectInbound(fwdTestNetNumber, tcpip.PacketBuffer{ + ep1.InjectInbound(fwdTestNetNumber, PacketBuffer{ Data: buf.ToVectorisedView(), }) } @@ -603,7 +603,7 @@ func TestForwardingWithFakeResolverManyResolutions(t *testing.T) { // maxPendingResolutions + 7). buf := buffer.NewView(30) buf[0] = byte(3 + i) - ep1.InjectInbound(fwdTestNetNumber, tcpip.PacketBuffer{ + ep1.InjectInbound(fwdTestNetNumber, PacketBuffer{ Data: buf.ToVectorisedView(), }) } diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go new file mode 100644 index 000000000..37907ae24 --- /dev/null +++ b/pkg/tcpip/stack/iptables.go @@ -0,0 +1,311 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package stack + +import ( + "fmt" + + "gvisor.dev/gvisor/pkg/tcpip/header" +) + +// Table names. +const ( + TablenameNat = "nat" + TablenameMangle = "mangle" + TablenameFilter = "filter" +) + +// Chain names as defined by net/ipv4/netfilter/ip_tables.c. +const ( + ChainNamePrerouting = "PREROUTING" + ChainNameInput = "INPUT" + ChainNameForward = "FORWARD" + ChainNameOutput = "OUTPUT" + ChainNamePostrouting = "POSTROUTING" +) + +// HookUnset indicates that there is no hook set for an entrypoint or +// underflow. +const HookUnset = -1 + +// DefaultTables returns a default set of tables. Each chain is set to accept +// all packets. +func DefaultTables() IPTables { + // TODO(gvisor.dev/issue/170): We may be able to swap out some strings for + // iotas. + return IPTables{ + Tables: map[string]Table{ + TablenameNat: Table{ + Rules: []Rule{ + Rule{Target: AcceptTarget{}}, + Rule{Target: AcceptTarget{}}, + Rule{Target: AcceptTarget{}}, + Rule{Target: AcceptTarget{}}, + Rule{Target: ErrorTarget{}}, + }, + BuiltinChains: map[Hook]int{ + Prerouting: 0, + Input: 1, + Output: 2, + Postrouting: 3, + }, + Underflows: map[Hook]int{ + Prerouting: 0, + Input: 1, + Output: 2, + Postrouting: 3, + }, + UserChains: map[string]int{}, + }, + TablenameMangle: Table{ + Rules: []Rule{ + Rule{Target: AcceptTarget{}}, + Rule{Target: AcceptTarget{}}, + Rule{Target: ErrorTarget{}}, + }, + BuiltinChains: map[Hook]int{ + Prerouting: 0, + Output: 1, + }, + Underflows: map[Hook]int{ + Prerouting: 0, + Output: 1, + }, + UserChains: map[string]int{}, + }, + TablenameFilter: Table{ + Rules: []Rule{ + Rule{Target: AcceptTarget{}}, + Rule{Target: AcceptTarget{}}, + Rule{Target: AcceptTarget{}}, + Rule{Target: ErrorTarget{}}, + }, + BuiltinChains: map[Hook]int{ + Input: 0, + Forward: 1, + Output: 2, + }, + Underflows: map[Hook]int{ + Input: 0, + Forward: 1, + Output: 2, + }, + UserChains: map[string]int{}, + }, + }, + Priorities: map[Hook][]string{ + Input: []string{TablenameNat, TablenameFilter}, + Prerouting: []string{TablenameMangle, TablenameNat}, + Output: []string{TablenameMangle, TablenameNat, TablenameFilter}, + }, + } +} + +// EmptyFilterTable returns a Table with no rules and the filter table chains +// mapped to HookUnset. +func EmptyFilterTable() Table { + return Table{ + Rules: []Rule{}, + BuiltinChains: map[Hook]int{ + Input: HookUnset, + Forward: HookUnset, + Output: HookUnset, + }, + Underflows: map[Hook]int{ + Input: HookUnset, + Forward: HookUnset, + Output: HookUnset, + }, + UserChains: map[string]int{}, + } +} + +// EmptyNatTable returns a Table with no rules and the filter table chains +// mapped to HookUnset. +func EmptyNatTable() Table { + return Table{ + Rules: []Rule{}, + BuiltinChains: map[Hook]int{ + Prerouting: HookUnset, + Input: HookUnset, + Output: HookUnset, + Postrouting: HookUnset, + }, + Underflows: map[Hook]int{ + Prerouting: HookUnset, + Input: HookUnset, + Output: HookUnset, + Postrouting: HookUnset, + }, + UserChains: map[string]int{}, + } +} + +// A chainVerdict is what a table decides should be done with a packet. +type chainVerdict int + +const ( + // chainAccept indicates the packet should continue through netstack. + chainAccept chainVerdict = iota + + // chainAccept indicates the packet should be dropped. + chainDrop + + // chainReturn indicates the packet should return to the calling chain + // or the underflow rule of a builtin chain. + chainReturn +) + +// Check runs pkt through the rules for hook. It returns true when the packet +// should continue traversing the network stack and false when it should be +// dropped. +// +// Precondition: pkt.NetworkHeader is set. +func (it *IPTables) Check(hook Hook, pkt PacketBuffer) bool { + // Go through each table containing the hook. + for _, tablename := range it.Priorities[hook] { + table := it.Tables[tablename] + ruleIdx := table.BuiltinChains[hook] + switch verdict := it.checkChain(hook, pkt, table, ruleIdx); verdict { + // If the table returns Accept, move on to the next table. + case chainAccept: + continue + // The Drop verdict is final. + case chainDrop: + return false + case chainReturn: + // Any Return from a built-in chain means we have to + // call the underflow. + underflow := table.Rules[table.Underflows[hook]] + switch v, _ := underflow.Target.Action(pkt); v { + case RuleAccept: + continue + case RuleDrop: + return false + case RuleJump, RuleReturn: + panic("Underflows should only return RuleAccept or RuleDrop.") + default: + panic(fmt.Sprintf("Unknown verdict: %d", v)) + } + + default: + panic(fmt.Sprintf("Unknown verdict %v.", verdict)) + } + } + + // Every table returned Accept. + return true +} + +// Precondition: pkt.NetworkHeader is set. +func (it *IPTables) checkChain(hook Hook, pkt PacketBuffer, table Table, ruleIdx int) chainVerdict { + // Start from ruleIdx and walk the list of rules until a rule gives us + // a verdict. + for ruleIdx < len(table.Rules) { + switch verdict, jumpTo := it.checkRule(hook, pkt, table, ruleIdx); verdict { + case RuleAccept: + return chainAccept + + case RuleDrop: + return chainDrop + + case RuleReturn: + return chainReturn + + case RuleJump: + // "Jumping" to the next rule just means we're + // continuing on down the list. + if jumpTo == ruleIdx+1 { + ruleIdx++ + continue + } + switch verdict := it.checkChain(hook, pkt, table, jumpTo); verdict { + case chainAccept: + return chainAccept + case chainDrop: + return chainDrop + case chainReturn: + ruleIdx++ + continue + default: + panic(fmt.Sprintf("Unknown verdict: %d", verdict)) + } + + default: + panic(fmt.Sprintf("Unknown verdict: %d", verdict)) + } + + } + + // We got through the entire table without a decision. Default to DROP + // for safety. + return chainDrop +} + +// Precondition: pk.NetworkHeader is set. +func (it *IPTables) checkRule(hook Hook, pkt PacketBuffer, table Table, ruleIdx int) (RuleVerdict, int) { + rule := table.Rules[ruleIdx] + + // If pkt.NetworkHeader hasn't been set yet, it will be contained in + // pkt.Data.First(). + if pkt.NetworkHeader == nil { + pkt.NetworkHeader = pkt.Data.First() + } + + // Check whether the packet matches the IP header filter. + if !filterMatch(rule.Filter, header.IPv4(pkt.NetworkHeader)) { + // Continue on to the next rule. + return RuleJump, ruleIdx + 1 + } + + // Go through each rule matcher. If they all match, run + // the rule target. + for _, matcher := range rule.Matchers { + matches, hotdrop := matcher.Match(hook, pkt, "") + if hotdrop { + return RuleDrop, 0 + } + if !matches { + // Continue on to the next rule. + return RuleJump, ruleIdx + 1 + } + } + + // All the matchers matched, so run the target. + return rule.Target.Action(pkt) +} + +func filterMatch(filter IPHeaderFilter, hdr header.IPv4) bool { + // TODO(gvisor.dev/issue/170): Support other fields of the filter. + // Check the transport protocol. + if filter.Protocol != 0 && filter.Protocol != hdr.TransportProtocol() { + return false + } + + // Check the destination IP. + dest := hdr.DestinationAddress() + matches := true + for i := range filter.Dst { + if dest[i]&filter.DstMask[i] != filter.Dst[i] { + matches = false + break + } + } + if matches == filter.DstInvert { + return false + } + + return true +} diff --git a/pkg/tcpip/stack/iptables_targets.go b/pkg/tcpip/stack/iptables_targets.go new file mode 100644 index 000000000..7b4543caf --- /dev/null +++ b/pkg/tcpip/stack/iptables_targets.go @@ -0,0 +1,144 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package stack + +import ( + "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/header" +) + +// AcceptTarget accepts packets. +type AcceptTarget struct{} + +// Action implements Target.Action. +func (AcceptTarget) Action(packet PacketBuffer) (RuleVerdict, int) { + return RuleAccept, 0 +} + +// DropTarget drops packets. +type DropTarget struct{} + +// Action implements Target.Action. +func (DropTarget) Action(packet PacketBuffer) (RuleVerdict, int) { + return RuleDrop, 0 +} + +// ErrorTarget logs an error and drops the packet. It represents a target that +// should be unreachable. +type ErrorTarget struct{} + +// Action implements Target.Action. +func (ErrorTarget) Action(packet PacketBuffer) (RuleVerdict, int) { + log.Debugf("ErrorTarget triggered.") + return RuleDrop, 0 +} + +// UserChainTarget marks a rule as the beginning of a user chain. +type UserChainTarget struct { + Name string +} + +// Action implements Target.Action. +func (UserChainTarget) Action(PacketBuffer) (RuleVerdict, int) { + panic("UserChainTarget should never be called.") +} + +// ReturnTarget returns from the current chain. If the chain is a built-in, the +// hook's underflow should be called. +type ReturnTarget struct{} + +// Action implements Target.Action. +func (ReturnTarget) Action(PacketBuffer) (RuleVerdict, int) { + return RuleReturn, 0 +} + +// RedirectTarget redirects the packet by modifying the destination port/IP. +// Min and Max values for IP and Ports in the struct indicate the range of +// values which can be used to redirect. +type RedirectTarget struct { + // TODO(gvisor.dev/issue/170): Other flags need to be added after + // we support them. + // RangeProtoSpecified flag indicates single port is specified to + // redirect. + RangeProtoSpecified bool + + // Min address used to redirect. + MinIP tcpip.Address + + // Max address used to redirect. + MaxIP tcpip.Address + + // Min port used to redirect. + MinPort uint16 + + // Max port used to redirect. + MaxPort uint16 +} + +// Action implements Target.Action. +// TODO(gvisor.dev/issue/170): Parse headers without copying. The current +// implementation only works for PREROUTING and calls pkt.Clone(), neither +// of which should be the case. +func (rt RedirectTarget) Action(pkt PacketBuffer) (RuleVerdict, int) { + newPkt := pkt.Clone() + + // Set network header. + headerView := newPkt.Data.First() + netHeader := header.IPv4(headerView) + newPkt.NetworkHeader = headerView[:header.IPv4MinimumSize] + + hlen := int(netHeader.HeaderLength()) + tlen := int(netHeader.TotalLength()) + newPkt.Data.TrimFront(hlen) + newPkt.Data.CapLength(tlen - hlen) + + // TODO(gvisor.dev/issue/170): Change destination address to + // loopback or interface address on which the packet was + // received. + + // TODO(gvisor.dev/issue/170): Check Flags in RedirectTarget if + // we need to change dest address (for OUTPUT chain) or ports. + switch protocol := netHeader.TransportProtocol(); protocol { + case header.UDPProtocolNumber: + var udpHeader header.UDP + if newPkt.TransportHeader != nil { + udpHeader = header.UDP(newPkt.TransportHeader) + } else { + if len(pkt.Data.First()) < header.UDPMinimumSize { + return RuleDrop, 0 + } + udpHeader = header.UDP(newPkt.Data.First()) + } + udpHeader.SetDestinationPort(rt.MinPort) + case header.TCPProtocolNumber: + var tcpHeader header.TCP + if newPkt.TransportHeader != nil { + tcpHeader = header.TCP(newPkt.TransportHeader) + } else { + if len(pkt.Data.First()) < header.TCPMinimumSize { + return RuleDrop, 0 + } + tcpHeader = header.TCP(newPkt.TransportHeader) + } + // TODO(gvisor.dev/issue/170): Need to recompute checksum + // and implement nat connection tracking to support TCP. + tcpHeader.SetDestinationPort(rt.MinPort) + default: + return RuleDrop, 0 + } + + return RuleAccept, 0 +} diff --git a/pkg/tcpip/stack/iptables_types.go b/pkg/tcpip/stack/iptables_types.go new file mode 100644 index 000000000..2ffb55f2a --- /dev/null +++ b/pkg/tcpip/stack/iptables_types.go @@ -0,0 +1,180 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package stack + +import ( + "gvisor.dev/gvisor/pkg/tcpip" +) + +// A Hook specifies one of the hooks built into the network stack. +// +// Userspace app Userspace app +// ^ | +// | v +// [Input] [Output] +// ^ | +// | v +// | routing +// | | +// | v +// ----->[Prerouting]----->routing----->[Forward]---------[Postrouting]-----> +type Hook uint + +// These values correspond to values in include/uapi/linux/netfilter.h. +const ( + // Prerouting happens before a packet is routed to applications or to + // be forwarded. + Prerouting Hook = iota + + // Input happens before a packet reaches an application. + Input + + // Forward happens once it's decided that a packet should be forwarded + // to another host. + Forward + + // Output happens after a packet is written by an application to be + // sent out. + Output + + // Postrouting happens just before a packet goes out on the wire. + Postrouting + + // The total number of hooks. + NumHooks +) + +// A RuleVerdict is what a rule decides should be done with a packet. +type RuleVerdict int + +const ( + // RuleAccept indicates the packet should continue through netstack. + RuleAccept RuleVerdict = iota + + // RuleDrop indicates the packet should be dropped. + RuleDrop + + // RuleJump indicates the packet should jump to another chain. + RuleJump + + // RuleReturn indicates the packet should return to the previous chain. + RuleReturn +) + +// IPTables holds all the tables for a netstack. +type IPTables struct { + // Tables maps table names to tables. User tables have arbitrary names. + Tables map[string]Table + + // Priorities maps each hook to a list of table names. The order of the + // list is the order in which each table should be visited for that + // hook. + Priorities map[Hook][]string +} + +// A Table defines a set of chains and hooks into the network stack. It is +// really just a list of rules with some metadata for entrypoints and such. +type Table struct { + // Rules holds the rules that make up the table. + Rules []Rule + + // BuiltinChains maps builtin chains to their entrypoint rule in Rules. + BuiltinChains map[Hook]int + + // Underflows maps builtin chains to their underflow rule in Rules + // (i.e. the rule to execute if the chain returns without a verdict). + Underflows map[Hook]int + + // UserChains holds user-defined chains for the keyed by name. Users + // can give their chains arbitrary names. + UserChains map[string]int + + // Metadata holds information about the Table that is useful to users + // of IPTables, but not to the netstack IPTables code itself. + metadata interface{} +} + +// ValidHooks returns a bitmap of the builtin hooks for the given table. +func (table *Table) ValidHooks() uint32 { + hooks := uint32(0) + for hook := range table.BuiltinChains { + hooks |= 1 << hook + } + return hooks +} + +// Metadata returns the metadata object stored in table. +func (table *Table) Metadata() interface{} { + return table.metadata +} + +// SetMetadata sets the metadata object stored in table. +func (table *Table) SetMetadata(metadata interface{}) { + table.metadata = metadata +} + +// A Rule is a packet processing rule. It consists of two pieces. First it +// contains zero or more matchers, each of which is a specification of which +// packets this rule applies to. If there are no matchers in the rule, it +// applies to any packet. +type Rule struct { + // Filter holds basic IP filtering fields common to every rule. + Filter IPHeaderFilter + + // Matchers is the list of matchers for this rule. + Matchers []Matcher + + // Target is the action to invoke if all the matchers match the packet. + Target Target +} + +// IPHeaderFilter holds basic IP filtering data common to every rule. +type IPHeaderFilter struct { + // Protocol matches the transport protocol. + Protocol tcpip.TransportProtocolNumber + + // Dst matches the destination IP address. + Dst tcpip.Address + + // DstMask masks bits of the destination IP address when comparing with + // Dst. + DstMask tcpip.Address + + // DstInvert inverts the meaning of the destination IP check, i.e. when + // true the filter will match packets that fail the destination + // comparison. + DstInvert bool +} + +// A Matcher is the interface for matching packets. +type Matcher interface { + // Name returns the name of the Matcher. + Name() string + + // Match returns whether the packet matches and whether the packet + // should be "hotdropped", i.e. dropped immediately. This is usually + // used for suspicious packets. + // + // Precondition: packet.NetworkHeader is set. + Match(hook Hook, packet PacketBuffer, interfaceName string) (matches bool, hotdrop bool) +} + +// A Target is the interface for taking an action for a packet. +type Target interface { + // Action takes an action on the packet and returns a verdict on how + // traversal should (or should not) continue. If the return value is + // Jump, it also returns the index of the rule to jump to. + Action(packet PacketBuffer) (RuleVerdict, int) +} diff --git a/pkg/tcpip/stack/ndp.go b/pkg/tcpip/stack/ndp.go index d689a006d..630fdefc5 100644 --- a/pkg/tcpip/stack/ndp.go +++ b/pkg/tcpip/stack/ndp.go @@ -564,7 +564,7 @@ func (ndp *ndpState) sendDADPacket(addr tcpip.Address) *tcpip.Error { Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, TOS: DefaultTOS, - }, tcpip.PacketBuffer{Header: hdr}, + }, PacketBuffer{Header: hdr}, ); err != nil { sent.Dropped.Increment() return err @@ -1283,7 +1283,7 @@ func (ndp *ndpState) startSolicitingRouters() { Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, TOS: DefaultTOS, - }, tcpip.PacketBuffer{Header: hdr}, + }, PacketBuffer{Header: hdr}, ); err != nil { sent.Dropped.Increment() log.Printf("startSolicitingRouters: error writing NDP router solicit message on NIC(%d); err = %s", ndp.nic.ID(), err) diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go index 4368c236c..06edd05b6 100644 --- a/pkg/tcpip/stack/ndp_test.go +++ b/pkg/tcpip/stack/ndp_test.go @@ -602,7 +602,7 @@ func TestDADFail(t *testing.T) { // Receive a packet to simulate multiple nodes owning or // attempting to own the same address. hdr := test.makeBuf(addr1) - e.InjectInbound(header.IPv6ProtocolNumber, tcpip.PacketBuffer{ + e.InjectInbound(header.IPv6ProtocolNumber, stack.PacketBuffer{ Data: hdr.View().ToVectorisedView(), }) @@ -918,7 +918,7 @@ func TestSetNDPConfigurations(t *testing.T) { // raBufWithOptsAndDHCPv6 returns a valid NDP Router Advertisement with options // and DHCPv6 configurations specified. -func raBufWithOptsAndDHCPv6(ip tcpip.Address, rl uint16, managedAddress, otherConfigurations bool, optSer header.NDPOptionsSerializer) tcpip.PacketBuffer { +func raBufWithOptsAndDHCPv6(ip tcpip.Address, rl uint16, managedAddress, otherConfigurations bool, optSer header.NDPOptionsSerializer) stack.PacketBuffer { icmpSize := header.ICMPv6HeaderSize + header.NDPRAMinimumSize + int(optSer.Length()) hdr := buffer.NewPrependable(header.IPv6MinimumSize + icmpSize) pkt := header.ICMPv6(hdr.Prepend(icmpSize)) @@ -953,14 +953,14 @@ func raBufWithOptsAndDHCPv6(ip tcpip.Address, rl uint16, managedAddress, otherCo DstAddr: header.IPv6AllNodesMulticastAddress, }) - return tcpip.PacketBuffer{Data: hdr.View().ToVectorisedView()} + return stack.PacketBuffer{Data: hdr.View().ToVectorisedView()} } // raBufWithOpts returns a valid NDP Router Advertisement with options. // // Note, raBufWithOpts does not populate any of the RA fields other than the // Router Lifetime. -func raBufWithOpts(ip tcpip.Address, rl uint16, optSer header.NDPOptionsSerializer) tcpip.PacketBuffer { +func raBufWithOpts(ip tcpip.Address, rl uint16, optSer header.NDPOptionsSerializer) stack.PacketBuffer { return raBufWithOptsAndDHCPv6(ip, rl, false, false, optSer) } @@ -969,7 +969,7 @@ func raBufWithOpts(ip tcpip.Address, rl uint16, optSer header.NDPOptionsSerializ // // Note, raBufWithDHCPv6 does not populate any of the RA fields other than the // DHCPv6 related ones. -func raBufWithDHCPv6(ip tcpip.Address, managedAddresses, otherConfiguratiosns bool) tcpip.PacketBuffer { +func raBufWithDHCPv6(ip tcpip.Address, managedAddresses, otherConfiguratiosns bool) stack.PacketBuffer { return raBufWithOptsAndDHCPv6(ip, 0, managedAddresses, otherConfiguratiosns, header.NDPOptionsSerializer{}) } @@ -977,7 +977,7 @@ func raBufWithDHCPv6(ip tcpip.Address, managedAddresses, otherConfiguratiosns bo // // Note, raBuf does not populate any of the RA fields other than the // Router Lifetime. -func raBuf(ip tcpip.Address, rl uint16) tcpip.PacketBuffer { +func raBuf(ip tcpip.Address, rl uint16) stack.PacketBuffer { return raBufWithOpts(ip, rl, header.NDPOptionsSerializer{}) } @@ -986,7 +986,7 @@ func raBuf(ip tcpip.Address, rl uint16) tcpip.PacketBuffer { // // Note, raBufWithPI does not populate any of the RA fields other than the // Router Lifetime. -func raBufWithPI(ip tcpip.Address, rl uint16, prefix tcpip.AddressWithPrefix, onLink, auto bool, vl, pl uint32) tcpip.PacketBuffer { +func raBufWithPI(ip tcpip.Address, rl uint16, prefix tcpip.AddressWithPrefix, onLink, auto bool, vl, pl uint32) stack.PacketBuffer { flags := uint8(0) if onLink { // The OnLink flag is the 7th bit in the flags byte. diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 9dcb1d52c..b6fa647ea 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -26,7 +26,6 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" - "gvisor.dev/gvisor/pkg/tcpip/iptables" ) var ipv4BroadcastAddr = tcpip.ProtocolAddress{ @@ -1144,7 +1143,7 @@ func (n *NIC) isInGroup(addr tcpip.Address) bool { return joins != 0 } -func handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, localLinkAddr, remotelinkAddr tcpip.LinkAddress, ref *referencedNetworkEndpoint, pkt tcpip.PacketBuffer) { +func handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, localLinkAddr, remotelinkAddr tcpip.LinkAddress, ref *referencedNetworkEndpoint, pkt PacketBuffer) { r := makeRoute(protocol, dst, src, localLinkAddr, ref, false /* handleLocal */, false /* multicastLoop */) r.RemoteLinkAddress = remotelinkAddr @@ -1158,7 +1157,7 @@ func handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, // Note that the ownership of the slice backing vv is retained by the caller. // This rule applies only to the slice itself, not to the items of the slice; // the ownership of the items is not retained by the caller. -func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { +func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt PacketBuffer) { n.mu.RLock() enabled := n.mu.enabled // If the NIC is not yet enabled, don't receive any packets. @@ -1222,7 +1221,7 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link // TODO(gvisor.dev/issue/170): Not supporting iptables for IPv6 yet. if protocol == header.IPv4ProtocolNumber { ipt := n.stack.IPTables() - if ok := ipt.Check(iptables.Prerouting, pkt); !ok { + if ok := ipt.Check(Prerouting, pkt); !ok { // iptables is telling us to drop the packet. return } @@ -1287,7 +1286,7 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link } } -func (n *NIC) forwardPacket(r *Route, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { +func (n *NIC) forwardPacket(r *Route, protocol tcpip.NetworkProtocolNumber, pkt PacketBuffer) { // TODO(b/143425874) Decrease the TTL field in forwarded packets. firstData := pkt.Data.First() @@ -1318,7 +1317,7 @@ func (n *NIC) forwardPacket(r *Route, protocol tcpip.NetworkProtocolNumber, pkt // DeliverTransportPacket delivers the packets to the appropriate transport // protocol endpoint. -func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt tcpip.PacketBuffer) { +func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt PacketBuffer) { state, ok := n.stack.transportProtocols[protocol] if !ok { n.stack.stats.UnknownProtocolRcvdPackets.Increment() @@ -1364,7 +1363,7 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN // DeliverTransportControlPacket delivers control packets to the appropriate // transport protocol endpoint. -func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, pkt tcpip.PacketBuffer) { +func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, pkt PacketBuffer) { state, ok := n.stack.transportProtocols[trans] if !ok { return diff --git a/pkg/tcpip/stack/nic_test.go b/pkg/tcpip/stack/nic_test.go index edaee3b86..d672fc157 100644 --- a/pkg/tcpip/stack/nic_test.go +++ b/pkg/tcpip/stack/nic_test.go @@ -17,7 +17,6 @@ package stack import ( "testing" - "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" ) @@ -45,7 +44,7 @@ func TestDisabledRxStatsWhenNICDisabled(t *testing.T) { t.FailNow() } - nic.DeliverNetworkPacket(nil, "", "", 0, tcpip.PacketBuffer{Data: buffer.View([]byte{1, 2, 3, 4}).ToVectorisedView()}) + nic.DeliverNetworkPacket(nil, "", "", 0, PacketBuffer{Data: buffer.View([]byte{1, 2, 3, 4}).ToVectorisedView()}) if got := nic.stats.DisabledRx.Packets.Value(); got != 1 { t.Errorf("got DisabledRx.Packets = %d, want = 1", got) diff --git a/pkg/tcpip/stack/packet_buffer.go b/pkg/tcpip/stack/packet_buffer.go new file mode 100644 index 000000000..1850fa8c3 --- /dev/null +++ b/pkg/tcpip/stack/packet_buffer.go @@ -0,0 +1,66 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at // +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +package stack + +import "gvisor.dev/gvisor/pkg/tcpip/buffer" + +// A PacketBuffer contains all the data of a network packet. +// +// As a PacketBuffer traverses up the stack, it may be necessary to pass it to +// multiple endpoints. Clone() should be called in such cases so that +// modifications to the Data field do not affect other copies. +// +// +stateify savable +type PacketBuffer struct { + // Data holds the payload of the packet. For inbound packets, it also + // holds the headers, which are consumed as the packet moves up the + // stack. Headers are guaranteed not to be split across views. + // + // The bytes backing Data are immutable, but Data itself may be trimmed + // or otherwise modified. + Data buffer.VectorisedView + + // DataOffset is used for GSO output. It is the offset into the Data + // field where the payload of this packet starts. + DataOffset int + + // DataSize is used for GSO output. It is the size of this packet's + // payload. + DataSize int + + // Header holds the headers of outbound packets. As a packet is passed + // down the stack, each layer adds to Header. + Header buffer.Prependable + + // These fields are used by both inbound and outbound packets. They + // typically overlap with the Data and Header fields. + // + // The bytes backing these views are immutable. Each field may be nil + // if either it has not been set yet or no such header exists (e.g. + // packets sent via loopback may not have a link header). + // + // These fields may be Views into other slices (either Data or Header). + // SR dosen't support this, so deep copies are necessary in some cases. + LinkHeader buffer.View + NetworkHeader buffer.View + TransportHeader buffer.View +} + +// Clone makes a copy of pk. It clones the Data field, which creates a new +// VectorisedView but does not deep copy the underlying bytes. +// +// Clone also does not deep copy any of its other fields. +func (pk PacketBuffer) Clone() PacketBuffer { + pk.Data = pk.Data.Clone(nil) + return pk +} diff --git a/pkg/tcpip/stack/packet_buffer_state.go b/pkg/tcpip/stack/packet_buffer_state.go new file mode 100644 index 000000000..76602549e --- /dev/null +++ b/pkg/tcpip/stack/packet_buffer_state.go @@ -0,0 +1,26 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +package stack + +import "gvisor.dev/gvisor/pkg/tcpip/buffer" + +// beforeSave is invoked by stateify. +func (pk *PacketBuffer) beforeSave() { + // Non-Data fields may be slices of the Data field. This causes + // problems for SR, so during save we make each header independent. + pk.Header = pk.Header.DeepCopy() + pk.LinkHeader = append(buffer.View(nil), pk.LinkHeader...) + pk.NetworkHeader = append(buffer.View(nil), pk.NetworkHeader...) + pk.TransportHeader = append(buffer.View(nil), pk.TransportHeader...) +} diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go index fa28b46b1..ac043b722 100644 --- a/pkg/tcpip/stack/registration.go +++ b/pkg/tcpip/stack/registration.go @@ -67,12 +67,12 @@ type TransportEndpoint interface { // this transport endpoint. It sets pkt.TransportHeader. // // HandlePacket takes ownership of pkt. - HandlePacket(r *Route, id TransportEndpointID, pkt tcpip.PacketBuffer) + HandlePacket(r *Route, id TransportEndpointID, pkt PacketBuffer) // HandleControlPacket is called by the stack when new control (e.g. // ICMP) packets arrive to this transport endpoint. // HandleControlPacket takes ownership of pkt. - HandleControlPacket(id TransportEndpointID, typ ControlType, extra uint32, pkt tcpip.PacketBuffer) + HandleControlPacket(id TransportEndpointID, typ ControlType, extra uint32, pkt PacketBuffer) // Abort initiates an expedited endpoint teardown. It puts the endpoint // in a closed state and frees all resources associated with it. This @@ -100,7 +100,7 @@ type RawTransportEndpoint interface { // layer up. // // HandlePacket takes ownership of pkt. - HandlePacket(r *Route, pkt tcpip.PacketBuffer) + HandlePacket(r *Route, pkt PacketBuffer) } // PacketEndpoint is the interface that needs to be implemented by packet @@ -118,7 +118,7 @@ type PacketEndpoint interface { // should construct its own ethernet header for applications. // // HandlePacket takes ownership of pkt. - HandlePacket(nicID tcpip.NICID, addr tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) + HandlePacket(nicID tcpip.NICID, addr tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, pkt PacketBuffer) } // TransportProtocol is the interface that needs to be implemented by transport @@ -150,7 +150,7 @@ type TransportProtocol interface { // stats purposes only). // // HandleUnknownDestinationPacket takes ownership of pkt. - HandleUnknownDestinationPacket(r *Route, id TransportEndpointID, pkt tcpip.PacketBuffer) bool + HandleUnknownDestinationPacket(r *Route, id TransportEndpointID, pkt PacketBuffer) bool // SetOption allows enabling/disabling protocol specific features. // SetOption returns an error if the option is not supported or the @@ -180,7 +180,7 @@ type TransportDispatcher interface { // pkt.NetworkHeader must be set before calling DeliverTransportPacket. // // DeliverTransportPacket takes ownership of pkt. - DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt tcpip.PacketBuffer) + DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt PacketBuffer) // DeliverTransportControlPacket delivers control packets to the // appropriate transport protocol endpoint. @@ -189,7 +189,7 @@ type TransportDispatcher interface { // DeliverTransportControlPacket. // // DeliverTransportControlPacket takes ownership of pkt. - DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, pkt tcpip.PacketBuffer) + DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, pkt PacketBuffer) } // PacketLooping specifies where an outbound packet should be sent. @@ -242,15 +242,15 @@ type NetworkEndpoint interface { // WritePacket writes a packet to the given destination address and // protocol. It sets pkt.NetworkHeader. pkt.TransportHeader must have // already been set. - WritePacket(r *Route, gso *GSO, params NetworkHeaderParams, pkt tcpip.PacketBuffer) *tcpip.Error + WritePacket(r *Route, gso *GSO, params NetworkHeaderParams, pkt PacketBuffer) *tcpip.Error // WritePackets writes packets to the given destination address and // protocol. pkts must not be zero length. - WritePackets(r *Route, gso *GSO, pkts []tcpip.PacketBuffer, params NetworkHeaderParams) (int, *tcpip.Error) + WritePackets(r *Route, gso *GSO, pkts []PacketBuffer, params NetworkHeaderParams) (int, *tcpip.Error) // WriteHeaderIncludedPacket writes a packet that includes a network // header to the given destination address. - WriteHeaderIncludedPacket(r *Route, pkt tcpip.PacketBuffer) *tcpip.Error + WriteHeaderIncludedPacket(r *Route, pkt PacketBuffer) *tcpip.Error // ID returns the network protocol endpoint ID. ID() *NetworkEndpointID @@ -265,7 +265,7 @@ type NetworkEndpoint interface { // this network endpoint. It sets pkt.NetworkHeader. // // HandlePacket takes ownership of pkt. - HandlePacket(r *Route, pkt tcpip.PacketBuffer) + HandlePacket(r *Route, pkt PacketBuffer) // Close is called when the endpoint is reomved from a stack. Close() @@ -322,7 +322,7 @@ type NetworkDispatcher interface { // packets sent via loopback), and won't have the field set. // // DeliverNetworkPacket takes ownership of pkt. - DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) + DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt PacketBuffer) } // LinkEndpointCapabilities is the type associated with the capabilities @@ -354,7 +354,7 @@ const ( // LinkEndpoint is the interface implemented by data link layer protocols (e.g., // ethernet, loopback, raw) and used by network layer protocols to send packets // out through the implementer's data link endpoint. When a link header exists, -// it sets each tcpip.PacketBuffer's LinkHeader field before passing it up the +// it sets each PacketBuffer's LinkHeader field before passing it up the // stack. type LinkEndpoint interface { // MTU is the maximum transmission unit for this endpoint. This is @@ -385,7 +385,7 @@ type LinkEndpoint interface { // To participate in transparent bridging, a LinkEndpoint implementation // should call eth.Encode with header.EthernetFields.SrcAddr set to // r.LocalLinkAddress if it is provided. - WritePacket(r *Route, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) *tcpip.Error + WritePacket(r *Route, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkt PacketBuffer) *tcpip.Error // WritePackets writes packets with the given protocol through the // given route. pkts must not be zero length. @@ -393,7 +393,7 @@ type LinkEndpoint interface { // Right now, WritePackets is used only when the software segmentation // offload is enabled. If it will be used for something else, it may // require to change syscall filters. - WritePackets(r *Route, gso *GSO, pkts []tcpip.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) + WritePackets(r *Route, gso *GSO, pkts []PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) // WriteRawPacket writes a packet directly to the link. The packet // should already have an ethernet header. @@ -426,7 +426,7 @@ type InjectableLinkEndpoint interface { LinkEndpoint // InjectInbound injects an inbound packet. - InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) + InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt PacketBuffer) // InjectOutbound writes a fully formed outbound packet directly to the // link. diff --git a/pkg/tcpip/stack/route.go b/pkg/tcpip/stack/route.go index f565aafb2..9fbe8a411 100644 --- a/pkg/tcpip/stack/route.go +++ b/pkg/tcpip/stack/route.go @@ -153,7 +153,7 @@ func (r *Route) IsResolutionRequired() bool { } // WritePacket writes the packet through the given route. -func (r *Route) WritePacket(gso *GSO, params NetworkHeaderParams, pkt tcpip.PacketBuffer) *tcpip.Error { +func (r *Route) WritePacket(gso *GSO, params NetworkHeaderParams, pkt PacketBuffer) *tcpip.Error { if !r.ref.isValidForOutgoing() { return tcpip.ErrInvalidEndpointState } @@ -169,7 +169,7 @@ func (r *Route) WritePacket(gso *GSO, params NetworkHeaderParams, pkt tcpip.Pack } // WritePackets writes the set of packets through the given route. -func (r *Route) WritePackets(gso *GSO, pkts []tcpip.PacketBuffer, params NetworkHeaderParams) (int, *tcpip.Error) { +func (r *Route) WritePackets(gso *GSO, pkts []PacketBuffer, params NetworkHeaderParams) (int, *tcpip.Error) { if !r.ref.isValidForOutgoing() { return 0, tcpip.ErrInvalidEndpointState } @@ -190,7 +190,7 @@ func (r *Route) WritePackets(gso *GSO, pkts []tcpip.PacketBuffer, params Network // WriteHeaderIncludedPacket writes a packet already containing a network // header through the given route. -func (r *Route) WriteHeaderIncludedPacket(pkt tcpip.PacketBuffer) *tcpip.Error { +func (r *Route) WriteHeaderIncludedPacket(pkt PacketBuffer) *tcpip.Error { if !r.ref.isValidForOutgoing() { return tcpip.ErrInvalidEndpointState } diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index 6f423874a..a9584d636 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -31,7 +31,6 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" - "gvisor.dev/gvisor/pkg/tcpip/iptables" "gvisor.dev/gvisor/pkg/tcpip/ports" "gvisor.dev/gvisor/pkg/tcpip/seqnum" "gvisor.dev/gvisor/pkg/waiter" @@ -51,7 +50,7 @@ const ( type transportProtocolState struct { proto TransportProtocol - defaultHandler func(r *Route, id TransportEndpointID, pkt tcpip.PacketBuffer) bool + defaultHandler func(r *Route, id TransportEndpointID, pkt PacketBuffer) bool } // TCPProbeFunc is the expected function type for a TCP probe function to be @@ -428,7 +427,7 @@ type Stack struct { // tables are the iptables packet filtering and manipulation rules. The are // protected by tablesMu.` - tables iptables.IPTables + tables IPTables // resumableEndpoints is a list of endpoints that need to be resumed if the // stack is being restored. @@ -738,7 +737,7 @@ func (s *Stack) TransportProtocolOption(transport tcpip.TransportProtocolNumber, // // It must be called only during initialization of the stack. Changing it as the // stack is operating is not supported. -func (s *Stack) SetTransportProtocolHandler(p tcpip.TransportProtocolNumber, h func(*Route, TransportEndpointID, tcpip.PacketBuffer) bool) { +func (s *Stack) SetTransportProtocolHandler(p tcpip.TransportProtocolNumber, h func(*Route, TransportEndpointID, PacketBuffer) bool) { state := s.transportProtocols[p] if state != nil { state.defaultHandler = h @@ -1701,7 +1700,7 @@ func (s *Stack) IsInGroup(nicID tcpip.NICID, multicastAddr tcpip.Address) (bool, } // IPTables returns the stack's iptables. -func (s *Stack) IPTables() iptables.IPTables { +func (s *Stack) IPTables() IPTables { s.tablesMu.RLock() t := s.tables s.tablesMu.RUnlock() @@ -1709,7 +1708,7 @@ func (s *Stack) IPTables() iptables.IPTables { } // SetIPTables sets the stack's iptables. -func (s *Stack) SetIPTables(ipt iptables.IPTables) { +func (s *Stack) SetIPTables(ipt IPTables) { s.tablesMu.Lock() s.tables = ipt s.tablesMu.Unlock() diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index 9836b340f..555fcd92f 100644 --- a/pkg/tcpip/stack/stack_test.go +++ b/pkg/tcpip/stack/stack_test.go @@ -90,7 +90,7 @@ func (f *fakeNetworkEndpoint) ID() *stack.NetworkEndpointID { return &f.id } -func (f *fakeNetworkEndpoint) HandlePacket(r *stack.Route, pkt tcpip.PacketBuffer) { +func (f *fakeNetworkEndpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { // Increment the received packet count in the protocol descriptor. f.proto.packetCount[int(f.id.LocalAddress[0])%len(f.proto.packetCount)]++ @@ -126,7 +126,7 @@ func (f *fakeNetworkEndpoint) Capabilities() stack.LinkEndpointCapabilities { return f.ep.Capabilities() } -func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, pkt tcpip.PacketBuffer) *tcpip.Error { +func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, pkt stack.PacketBuffer) *tcpip.Error { // Increment the sent packet count in the protocol descriptor. f.proto.sendPacketCount[int(r.RemoteAddress[0])%len(f.proto.sendPacketCount)]++ @@ -141,7 +141,7 @@ func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, params views := make([]buffer.View, 1, 1+len(pkt.Data.Views())) views[0] = pkt.Header.View() views = append(views, pkt.Data.Views()...) - f.HandlePacket(r, tcpip.PacketBuffer{ + f.HandlePacket(r, stack.PacketBuffer{ Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views), }) } @@ -153,11 +153,11 @@ func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, params } // WritePackets implements stack.LinkEndpoint.WritePackets. -func (f *fakeNetworkEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []tcpip.PacketBuffer, params stack.NetworkHeaderParams) (int, *tcpip.Error) { +func (f *fakeNetworkEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []stack.PacketBuffer, params stack.NetworkHeaderParams) (int, *tcpip.Error) { panic("not implemented") } -func (*fakeNetworkEndpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt tcpip.PacketBuffer) *tcpip.Error { +func (*fakeNetworkEndpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuffer) *tcpip.Error { return tcpip.ErrNotSupported } @@ -287,7 +287,7 @@ func TestNetworkReceive(t *testing.T) { // Make sure packet with wrong address is not delivered. buf[0] = 3 - ep.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + ep.InjectInbound(fakeNetNumber, stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if fakeNet.packetCount[1] != 0 { @@ -299,7 +299,7 @@ func TestNetworkReceive(t *testing.T) { // Make sure packet is delivered to first endpoint. buf[0] = 1 - ep.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + ep.InjectInbound(fakeNetNumber, stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if fakeNet.packetCount[1] != 1 { @@ -311,7 +311,7 @@ func TestNetworkReceive(t *testing.T) { // Make sure packet is delivered to second endpoint. buf[0] = 2 - ep.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + ep.InjectInbound(fakeNetNumber, stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if fakeNet.packetCount[1] != 1 { @@ -322,7 +322,7 @@ func TestNetworkReceive(t *testing.T) { } // Make sure packet is not delivered if protocol number is wrong. - ep.InjectInbound(fakeNetNumber-1, tcpip.PacketBuffer{ + ep.InjectInbound(fakeNetNumber-1, stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if fakeNet.packetCount[1] != 1 { @@ -334,7 +334,7 @@ func TestNetworkReceive(t *testing.T) { // Make sure packet that is too small is dropped. buf.CapLength(2) - ep.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + ep.InjectInbound(fakeNetNumber, stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if fakeNet.packetCount[1] != 1 { @@ -356,7 +356,7 @@ func sendTo(s *stack.Stack, addr tcpip.Address, payload buffer.View) *tcpip.Erro func send(r stack.Route, payload buffer.View) *tcpip.Error { hdr := buffer.NewPrependable(int(r.MaxHeaderLength())) - return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}, stack.PacketBuffer{ Header: hdr, Data: payload.ToVectorisedView(), }) @@ -414,7 +414,7 @@ func testFailingRecv(t *testing.T, fakeNet *fakeNetworkProtocol, localAddrByte b func testRecvInternal(t *testing.T, fakeNet *fakeNetworkProtocol, localAddrByte byte, ep *channel.Endpoint, buf buffer.View, want int) { t.Helper() - ep.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + ep.InjectInbound(fakeNetNumber, stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if got := fakeNet.PacketCount(localAddrByte); got != want { @@ -2257,7 +2257,7 @@ func TestNICStats(t *testing.T) { // Send a packet to address 1. buf := buffer.NewView(30) - ep1.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + ep1.InjectInbound(fakeNetNumber, stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if got, want := s.NICInfo()[1].Stats.Rx.Packets.Value(), uint64(1); got != want { @@ -2339,7 +2339,7 @@ func TestNICForwarding(t *testing.T) { // Send a packet to dstAddr. buf := buffer.NewView(30) buf[0] = dstAddr[0] - ep1.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + ep1.InjectInbound(fakeNetNumber, stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) diff --git a/pkg/tcpip/stack/transport_demuxer.go b/pkg/tcpip/stack/transport_demuxer.go index d4c0359e8..c55e3e8bc 100644 --- a/pkg/tcpip/stack/transport_demuxer.go +++ b/pkg/tcpip/stack/transport_demuxer.go @@ -85,7 +85,7 @@ func (epsByNic *endpointsByNic) transportEndpoints() []TransportEndpoint { // HandlePacket is called by the stack when new packets arrive to this transport // endpoint. -func (epsByNic *endpointsByNic) handlePacket(r *Route, id TransportEndpointID, pkt tcpip.PacketBuffer) { +func (epsByNic *endpointsByNic) handlePacket(r *Route, id TransportEndpointID, pkt PacketBuffer) { epsByNic.mu.RLock() mpep, ok := epsByNic.endpoints[r.ref.nic.ID()] @@ -116,7 +116,7 @@ func (epsByNic *endpointsByNic) handlePacket(r *Route, id TransportEndpointID, p } // HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket. -func (epsByNic *endpointsByNic) handleControlPacket(n *NIC, id TransportEndpointID, typ ControlType, extra uint32, pkt tcpip.PacketBuffer) { +func (epsByNic *endpointsByNic) handleControlPacket(n *NIC, id TransportEndpointID, typ ControlType, extra uint32, pkt PacketBuffer) { epsByNic.mu.RLock() defer epsByNic.mu.RUnlock() @@ -184,7 +184,7 @@ type transportDemuxer struct { // the dispatcher to delivery packets to the QueuePacket method instead of // calling HandlePacket directly on the endpoint. type queuedTransportProtocol interface { - QueuePacket(r *Route, ep TransportEndpoint, id TransportEndpointID, pkt tcpip.PacketBuffer) + QueuePacket(r *Route, ep TransportEndpoint, id TransportEndpointID, pkt PacketBuffer) } func newTransportDemuxer(stack *Stack) *transportDemuxer { @@ -312,7 +312,7 @@ func selectEndpoint(id TransportEndpointID, mpep *multiPortEndpoint, seed uint32 return mpep.endpoints[idx] } -func (ep *multiPortEndpoint) handlePacketAll(r *Route, id TransportEndpointID, pkt tcpip.PacketBuffer) { +func (ep *multiPortEndpoint) handlePacketAll(r *Route, id TransportEndpointID, pkt PacketBuffer) { ep.mu.RLock() queuedProtocol, mustQueue := ep.demux.queuedProtocols[protocolIDs{ep.netProto, ep.transProto}] // HandlePacket takes ownership of pkt, so each endpoint needs @@ -403,7 +403,7 @@ func (d *transportDemuxer) unregisterEndpoint(netProtos []tcpip.NetworkProtocolN // deliverPacket attempts to find one or more matching transport endpoints, and // then, if matches are found, delivers the packet to them. Returns true if // the packet no longer needs to be handled. -func (d *transportDemuxer) deliverPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt tcpip.PacketBuffer, id TransportEndpointID) bool { +func (d *transportDemuxer) deliverPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt PacketBuffer, id TransportEndpointID) bool { eps, ok := d.protocol[protocolIDs{r.NetProto, protocol}] if !ok { return false @@ -453,7 +453,7 @@ func (d *transportDemuxer) deliverPacket(r *Route, protocol tcpip.TransportProto // deliverRawPacket attempts to deliver the given packet and returns whether it // was delivered successfully. -func (d *transportDemuxer) deliverRawPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt tcpip.PacketBuffer) bool { +func (d *transportDemuxer) deliverRawPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt PacketBuffer) bool { eps, ok := d.protocol[protocolIDs{r.NetProto, protocol}] if !ok { return false @@ -477,7 +477,7 @@ func (d *transportDemuxer) deliverRawPacket(r *Route, protocol tcpip.TransportPr // deliverControlPacket attempts to deliver the given control packet. Returns // true if it found an endpoint, false otherwise. -func (d *transportDemuxer) deliverControlPacket(n *NIC, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, pkt tcpip.PacketBuffer, id TransportEndpointID) bool { +func (d *transportDemuxer) deliverControlPacket(n *NIC, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, pkt PacketBuffer, id TransportEndpointID) bool { eps, ok := d.protocol[protocolIDs{net, trans}] if !ok { return false diff --git a/pkg/tcpip/stack/transport_demuxer_test.go b/pkg/tcpip/stack/transport_demuxer_test.go index 0e3e239c5..84311bcc8 100644 --- a/pkg/tcpip/stack/transport_demuxer_test.go +++ b/pkg/tcpip/stack/transport_demuxer_test.go @@ -150,7 +150,7 @@ func (c *testContext) sendV6Packet(payload []byte, h *headers, linkEpID tcpip.NI u.SetChecksum(^u.CalculateChecksum(xsum)) // Inject packet. - c.linkEps[linkEpID].InjectInbound(ipv6.ProtocolNumber, tcpip.PacketBuffer{ + c.linkEps[linkEpID].InjectInbound(ipv6.ProtocolNumber, stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) } diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go index 5d1da2f8b..8ca9ac3cf 100644 --- a/pkg/tcpip/stack/transport_test.go +++ b/pkg/tcpip/stack/transport_test.go @@ -19,7 +19,6 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" - "gvisor.dev/gvisor/pkg/tcpip/iptables" "gvisor.dev/gvisor/pkg/tcpip/link/channel" "gvisor.dev/gvisor/pkg/tcpip/link/loopback" "gvisor.dev/gvisor/pkg/tcpip/stack" @@ -87,7 +86,7 @@ func (f *fakeTransportEndpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions if err != nil { return 0, nil, err } - if err := f.route.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + if err := f.route.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}, stack.PacketBuffer{ Header: hdr, Data: buffer.View(v).ToVectorisedView(), }); err != nil { @@ -214,7 +213,7 @@ func (*fakeTransportEndpoint) GetRemoteAddress() (tcpip.FullAddress, *tcpip.Erro return tcpip.FullAddress{}, nil } -func (f *fakeTransportEndpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, _ tcpip.PacketBuffer) { +func (f *fakeTransportEndpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, _ stack.PacketBuffer) { // Increment the number of received packets. f.proto.packetCount++ if f.acceptQueue != nil { @@ -231,7 +230,7 @@ func (f *fakeTransportEndpoint) HandlePacket(r *stack.Route, id stack.TransportE } } -func (f *fakeTransportEndpoint) HandleControlPacket(stack.TransportEndpointID, stack.ControlType, uint32, tcpip.PacketBuffer) { +func (f *fakeTransportEndpoint) HandleControlPacket(stack.TransportEndpointID, stack.ControlType, uint32, stack.PacketBuffer) { // Increment the number of received control packets. f.proto.controlCount++ } @@ -242,8 +241,8 @@ func (f *fakeTransportEndpoint) State() uint32 { func (f *fakeTransportEndpoint) ModerateRecvBuf(copied int) {} -func (f *fakeTransportEndpoint) IPTables() (iptables.IPTables, error) { - return iptables.IPTables{}, nil +func (f *fakeTransportEndpoint) IPTables() (stack.IPTables, error) { + return stack.IPTables{}, nil } func (f *fakeTransportEndpoint) Resume(*stack.Stack) {} @@ -288,7 +287,7 @@ func (*fakeTransportProtocol) ParsePorts(buffer.View) (src, dst uint16, err *tcp return 0, 0, nil } -func (*fakeTransportProtocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, tcpip.PacketBuffer) bool { +func (*fakeTransportProtocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, stack.PacketBuffer) bool { return true } @@ -368,7 +367,7 @@ func TestTransportReceive(t *testing.T) { // Make sure packet with wrong protocol is not delivered. buf[0] = 1 buf[2] = 0 - linkEP.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + linkEP.InjectInbound(fakeNetNumber, stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if fakeTrans.packetCount != 0 { @@ -379,7 +378,7 @@ func TestTransportReceive(t *testing.T) { buf[0] = 1 buf[1] = 3 buf[2] = byte(fakeTransNumber) - linkEP.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + linkEP.InjectInbound(fakeNetNumber, stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if fakeTrans.packetCount != 0 { @@ -390,7 +389,7 @@ func TestTransportReceive(t *testing.T) { buf[0] = 1 buf[1] = 2 buf[2] = byte(fakeTransNumber) - linkEP.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + linkEP.InjectInbound(fakeNetNumber, stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if fakeTrans.packetCount != 1 { @@ -445,7 +444,7 @@ func TestTransportControlReceive(t *testing.T) { buf[fakeNetHeaderLen+0] = 0 buf[fakeNetHeaderLen+1] = 1 buf[fakeNetHeaderLen+2] = 0 - linkEP.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + linkEP.InjectInbound(fakeNetNumber, stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if fakeTrans.controlCount != 0 { @@ -456,7 +455,7 @@ func TestTransportControlReceive(t *testing.T) { buf[fakeNetHeaderLen+0] = 3 buf[fakeNetHeaderLen+1] = 1 buf[fakeNetHeaderLen+2] = byte(fakeTransNumber) - linkEP.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + linkEP.InjectInbound(fakeNetNumber, stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if fakeTrans.controlCount != 0 { @@ -467,7 +466,7 @@ func TestTransportControlReceive(t *testing.T) { buf[fakeNetHeaderLen+0] = 2 buf[fakeNetHeaderLen+1] = 1 buf[fakeNetHeaderLen+2] = byte(fakeTransNumber) - linkEP.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + linkEP.InjectInbound(fakeNetNumber, stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if fakeTrans.controlCount != 1 { @@ -622,7 +621,7 @@ func TestTransportForwarding(t *testing.T) { req[0] = 1 req[1] = 3 req[2] = byte(fakeTransNumber) - ep2.InjectInbound(fakeNetNumber, tcpip.PacketBuffer{ + ep2.InjectInbound(fakeNetNumber, stack.PacketBuffer{ Data: req.ToVectorisedView(), }) diff --git a/pkg/tcpip/transport/icmp/BUILD b/pkg/tcpip/transport/icmp/BUILD index ac18ec5b1..9ce625c17 100644 --- a/pkg/tcpip/transport/icmp/BUILD +++ b/pkg/tcpip/transport/icmp/BUILD @@ -31,7 +31,6 @@ go_library( "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/header", - "//pkg/tcpip/iptables", "//pkg/tcpip/stack", "//pkg/tcpip/transport/raw", "//pkg/tcpip/transport/tcp", diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index 2a396e9bc..613b12ead 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -19,7 +19,6 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" - "gvisor.dev/gvisor/pkg/tcpip/iptables" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/waiter" ) @@ -135,7 +134,7 @@ func (e *endpoint) Close() { func (e *endpoint) ModerateRecvBuf(copied int) {} // IPTables implements tcpip.Endpoint.IPTables. -func (e *endpoint) IPTables() (iptables.IPTables, error) { +func (e *endpoint) IPTables() (stack.IPTables, error) { return e.stack.IPTables(), nil } @@ -441,7 +440,7 @@ func send4(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Err if ttl == 0 { ttl = r.DefaultTTL() } - return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}, stack.PacketBuffer{ Header: hdr, Data: data.ToVectorisedView(), TransportHeader: buffer.View(icmpv4), @@ -471,7 +470,7 @@ func send6(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Err if ttl == 0 { ttl = r.DefaultTTL() } - return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}, stack.PacketBuffer{ Header: hdr, Data: dataVV, TransportHeader: buffer.View(icmpv6), @@ -733,7 +732,7 @@ func (e *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { // HandlePacket is called by the stack when new packets arrive to this transport // endpoint. -func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt tcpip.PacketBuffer) { +func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) { // Only accept echo replies. switch e.NetProto { case header.IPv4ProtocolNumber: @@ -795,7 +794,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk } // HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket. -func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.ControlType, extra uint32, pkt tcpip.PacketBuffer) { +func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.ControlType, extra uint32, pkt stack.PacketBuffer) { } // State implements tcpip.Endpoint.State. The ICMP endpoint currently doesn't diff --git a/pkg/tcpip/transport/icmp/protocol.go b/pkg/tcpip/transport/icmp/protocol.go index 113d92901..3c47692b2 100644 --- a/pkg/tcpip/transport/icmp/protocol.go +++ b/pkg/tcpip/transport/icmp/protocol.go @@ -104,7 +104,7 @@ func (p *protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) // HandleUnknownDestinationPacket handles packets targeted at this protocol but // that don't match any existing endpoint. -func (*protocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, tcpip.PacketBuffer) bool { +func (*protocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, stack.PacketBuffer) bool { return true } diff --git a/pkg/tcpip/transport/packet/BUILD b/pkg/tcpip/transport/packet/BUILD index d22de6b26..b989b1209 100644 --- a/pkg/tcpip/transport/packet/BUILD +++ b/pkg/tcpip/transport/packet/BUILD @@ -31,7 +31,6 @@ go_library( "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/header", - "//pkg/tcpip/iptables", "//pkg/tcpip/stack", "//pkg/waiter", ], diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go index 09a1cd436..df49d0995 100644 --- a/pkg/tcpip/transport/packet/endpoint.go +++ b/pkg/tcpip/transport/packet/endpoint.go @@ -29,7 +29,6 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" - "gvisor.dev/gvisor/pkg/tcpip/iptables" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/waiter" ) @@ -100,8 +99,8 @@ func NewEndpoint(s *stack.Stack, cooked bool, netProto tcpip.NetworkProtocolNumb } // Abort implements stack.TransportEndpoint.Abort. -func (e *endpoint) Abort() { - e.Close() +func (ep *endpoint) Abort() { + ep.Close() } // Close implements tcpip.Endpoint.Close. @@ -134,7 +133,7 @@ func (ep *endpoint) Close() { func (ep *endpoint) ModerateRecvBuf(copied int) {} // IPTables implements tcpip.Endpoint.IPTables. -func (ep *endpoint) IPTables() (iptables.IPTables, error) { +func (ep *endpoint) IPTables() (stack.IPTables, error) { return ep.stack.IPTables(), nil } @@ -299,7 +298,7 @@ func (ep *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { } // HandlePacket implements stack.PacketEndpoint.HandlePacket. -func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, pkt tcpip.PacketBuffer) { +func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { ep.rcvMu.Lock() // Drop the packet if our buffer is currently full. diff --git a/pkg/tcpip/transport/raw/BUILD b/pkg/tcpip/transport/raw/BUILD index c9baf4600..2eab09088 100644 --- a/pkg/tcpip/transport/raw/BUILD +++ b/pkg/tcpip/transport/raw/BUILD @@ -32,7 +32,6 @@ go_library( "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/header", - "//pkg/tcpip/iptables", "//pkg/tcpip/stack", "//pkg/tcpip/transport/packet", "//pkg/waiter", diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index 2ef5fac76..536dafd1e 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -30,7 +30,6 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" - "gvisor.dev/gvisor/pkg/tcpip/iptables" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/waiter" ) @@ -161,7 +160,7 @@ func (e *endpoint) Close() { func (e *endpoint) ModerateRecvBuf(copied int) {} // IPTables implements tcpip.Endpoint.IPTables. -func (e *endpoint) IPTables() (iptables.IPTables, error) { +func (e *endpoint) IPTables() (stack.IPTables, error) { return e.stack.IPTables(), nil } @@ -342,7 +341,7 @@ func (e *endpoint) finishWrite(payloadBytes []byte, route *stack.Route) (int64, switch e.NetProto { case header.IPv4ProtocolNumber: if !e.associated { - if err := route.WriteHeaderIncludedPacket(tcpip.PacketBuffer{ + if err := route.WriteHeaderIncludedPacket(stack.PacketBuffer{ Data: buffer.View(payloadBytes).ToVectorisedView(), }); err != nil { return 0, nil, err @@ -350,7 +349,7 @@ func (e *endpoint) finishWrite(payloadBytes []byte, route *stack.Route) (int64, break } hdr := buffer.NewPrependable(len(payloadBytes) + int(route.MaxHeaderLength())) - if err := route.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: e.TransProto, TTL: route.DefaultTTL(), TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + if err := route.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: e.TransProto, TTL: route.DefaultTTL(), TOS: stack.DefaultTOS}, stack.PacketBuffer{ Header: hdr, Data: buffer.View(payloadBytes).ToVectorisedView(), }); err != nil { @@ -574,7 +573,7 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { } // HandlePacket implements stack.RawTransportEndpoint.HandlePacket. -func (e *endpoint) HandlePacket(route *stack.Route, pkt tcpip.PacketBuffer) { +func (e *endpoint) HandlePacket(route *stack.Route, pkt stack.PacketBuffer) { e.rcvMu.Lock() // Drop the packet if our buffer is currently full. diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD index 2fdf6c0a5..7f94f9646 100644 --- a/pkg/tcpip/transport/tcp/BUILD +++ b/pkg/tcpip/transport/tcp/BUILD @@ -66,7 +66,6 @@ go_library( "//pkg/tcpip/buffer", "//pkg/tcpip/hash/jenkins", "//pkg/tcpip/header", - "//pkg/tcpip/iptables", "//pkg/tcpip/ports", "//pkg/tcpip/seqnum", "//pkg/tcpip/stack", diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 53193afc6..79552fc61 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -705,7 +705,7 @@ func (e *endpoint) sendTCP(r *stack.Route, id stack.TransportEndpointID, data bu return nil } -func buildTCPHdr(r *stack.Route, id stack.TransportEndpointID, pkt *tcpip.PacketBuffer, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts []byte, gso *stack.GSO) { +func buildTCPHdr(r *stack.Route, id stack.TransportEndpointID, pkt *stack.PacketBuffer, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts []byte, gso *stack.GSO) { optLen := len(opts) hdr := &pkt.Header packetSize := pkt.DataSize @@ -752,7 +752,7 @@ func sendTCPBatch(r *stack.Route, id stack.TransportEndpointID, data buffer.Vect // Allocate one big slice for all the headers. hdrSize := header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen buf := make([]byte, n*hdrSize) - pkts := make([]tcpip.PacketBuffer, n) + pkts := make([]stack.PacketBuffer, n) for i := range pkts { pkts[i].Header = buffer.NewEmptyPrependableFromView(buf[i*hdrSize:][:hdrSize]) } @@ -795,7 +795,7 @@ func sendTCP(r *stack.Route, id stack.TransportEndpointID, data buffer.Vectorise return sendTCPBatch(r, id, data, ttl, tos, flags, seq, ack, rcvWnd, opts, gso) } - pkt := tcpip.PacketBuffer{ + pkt := stack.PacketBuffer{ Header: buffer.NewPrependable(header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen), DataOffset: 0, DataSize: data.Size(), diff --git a/pkg/tcpip/transport/tcp/dispatcher.go b/pkg/tcpip/transport/tcp/dispatcher.go index 90ac956a9..6062ca916 100644 --- a/pkg/tcpip/transport/tcp/dispatcher.go +++ b/pkg/tcpip/transport/tcp/dispatcher.go @@ -18,7 +18,6 @@ import ( "gvisor.dev/gvisor/pkg/rand" "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/hash/jenkins" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" @@ -187,7 +186,7 @@ func (d *dispatcher) wait() { } } -func (d *dispatcher) queuePacket(r *stack.Route, stackEP stack.TransportEndpoint, id stack.TransportEndpointID, pkt tcpip.PacketBuffer) { +func (d *dispatcher) queuePacket(r *stack.Route, stackEP stack.TransportEndpoint, id stack.TransportEndpointID, pkt stack.PacketBuffer) { ep := stackEP.(*endpoint) s := newSegment(r, id, pkt) if !s.parse() { diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index eb8a9d73e..594efaa11 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -30,7 +30,6 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/hash/jenkins" "gvisor.dev/gvisor/pkg/tcpip/header" - "gvisor.dev/gvisor/pkg/tcpip/iptables" "gvisor.dev/gvisor/pkg/tcpip/ports" "gvisor.dev/gvisor/pkg/tcpip/seqnum" "gvisor.dev/gvisor/pkg/tcpip/stack" @@ -1120,7 +1119,7 @@ func (e *endpoint) ModerateRecvBuf(copied int) { } // IPTables implements tcpip.Endpoint.IPTables. -func (e *endpoint) IPTables() (iptables.IPTables, error) { +func (e *endpoint) IPTables() (stack.IPTables, error) { return e.stack.IPTables(), nil } @@ -2388,7 +2387,7 @@ func (e *endpoint) GetRemoteAddress() (tcpip.FullAddress, *tcpip.Error) { }, nil } -func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt tcpip.PacketBuffer) { +func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) { // TCP HandlePacket is not required anymore as inbound packets first // land at the Dispatcher which then can either delivery using the // worker go routine or directly do the invoke the tcp processing inline @@ -2407,7 +2406,7 @@ func (e *endpoint) enqueueSegment(s *segment) bool { } // HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket. -func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.ControlType, extra uint32, pkt tcpip.PacketBuffer) { +func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.ControlType, extra uint32, pkt stack.PacketBuffer) { switch typ { case stack.ControlPacketTooBig: e.sndBufMu.Lock() diff --git a/pkg/tcpip/transport/tcp/forwarder.go b/pkg/tcpip/transport/tcp/forwarder.go index c9ee5bf06..a094471b8 100644 --- a/pkg/tcpip/transport/tcp/forwarder.go +++ b/pkg/tcpip/transport/tcp/forwarder.go @@ -61,7 +61,7 @@ func NewForwarder(s *stack.Stack, rcvWnd, maxInFlight int, handler func(*Forward // // This function is expected to be passed as an argument to the // stack.SetTransportProtocolHandler function. -func (f *Forwarder) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt tcpip.PacketBuffer) bool { +func (f *Forwarder) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) bool { s := newSegment(r, id, pkt) defer s.decRef() diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go index b0f918bb4..57985b85d 100644 --- a/pkg/tcpip/transport/tcp/protocol.go +++ b/pkg/tcpip/transport/tcp/protocol.go @@ -140,7 +140,7 @@ func (*protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) { // to a specific processing queue. Each queue is serviced by its own processor // goroutine which is responsible for dequeuing and doing full TCP dispatch of // the packet. -func (p *protocol) QueuePacket(r *stack.Route, ep stack.TransportEndpoint, id stack.TransportEndpointID, pkt tcpip.PacketBuffer) { +func (p *protocol) QueuePacket(r *stack.Route, ep stack.TransportEndpoint, id stack.TransportEndpointID, pkt stack.PacketBuffer) { p.dispatcher.queuePacket(r, ep, id, pkt) } @@ -151,7 +151,7 @@ func (p *protocol) QueuePacket(r *stack.Route, ep stack.TransportEndpoint, id st // a reset is sent in response to any incoming segment except another reset. In // particular, SYNs addressed to a non-existent connection are rejected by this // means." -func (*protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, pkt tcpip.PacketBuffer) bool { +func (*protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) bool { s := newSegment(r, id, pkt) defer s.decRef() diff --git a/pkg/tcpip/transport/tcp/segment.go b/pkg/tcpip/transport/tcp/segment.go index 5d0bc4f72..e6fe7985d 100644 --- a/pkg/tcpip/transport/tcp/segment.go +++ b/pkg/tcpip/transport/tcp/segment.go @@ -18,7 +18,6 @@ import ( "sync/atomic" "time" - "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/seqnum" @@ -61,7 +60,7 @@ type segment struct { xmitCount uint32 } -func newSegment(r *stack.Route, id stack.TransportEndpointID, pkt tcpip.PacketBuffer) *segment { +func newSegment(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) *segment { s := &segment{ refCnt: 1, id: id, diff --git a/pkg/tcpip/transport/tcp/testing/context/context.go b/pkg/tcpip/transport/tcp/testing/context/context.go index 8cea20fb5..d4f6bc635 100644 --- a/pkg/tcpip/transport/tcp/testing/context/context.go +++ b/pkg/tcpip/transport/tcp/testing/context/context.go @@ -307,7 +307,7 @@ func (c *Context) SendICMPPacket(typ header.ICMPv4Type, code uint8, p1, p2 []byt copy(icmp[header.ICMPv4PayloadOffset:], p2) // Inject packet. - c.linkEP.InjectInbound(ipv4.ProtocolNumber, tcpip.PacketBuffer{ + c.linkEP.InjectInbound(ipv4.ProtocolNumber, stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) } @@ -363,7 +363,7 @@ func (c *Context) BuildSegmentWithAddrs(payload []byte, h *Headers, src, dst tcp // SendSegment sends a TCP segment that has already been built and written to a // buffer.VectorisedView. func (c *Context) SendSegment(s buffer.VectorisedView) { - c.linkEP.InjectInbound(ipv4.ProtocolNumber, tcpip.PacketBuffer{ + c.linkEP.InjectInbound(ipv4.ProtocolNumber, stack.PacketBuffer{ Data: s, }) } @@ -371,7 +371,7 @@ func (c *Context) SendSegment(s buffer.VectorisedView) { // SendPacket builds and sends a TCP segment(with the provided payload & TCP // headers) in an IPv4 packet via the link layer endpoint. func (c *Context) SendPacket(payload []byte, h *Headers) { - c.linkEP.InjectInbound(ipv4.ProtocolNumber, tcpip.PacketBuffer{ + c.linkEP.InjectInbound(ipv4.ProtocolNumber, stack.PacketBuffer{ Data: c.BuildSegment(payload, h), }) } @@ -380,7 +380,7 @@ func (c *Context) SendPacket(payload []byte, h *Headers) { // & TCPheaders) in an IPv4 packet via the link layer endpoint using the // provided source and destination IPv4 addresses. func (c *Context) SendPacketWithAddrs(payload []byte, h *Headers, src, dst tcpip.Address) { - c.linkEP.InjectInbound(ipv4.ProtocolNumber, tcpip.PacketBuffer{ + c.linkEP.InjectInbound(ipv4.ProtocolNumber, stack.PacketBuffer{ Data: c.BuildSegmentWithAddrs(payload, h, src, dst), }) } @@ -548,7 +548,7 @@ func (c *Context) SendV6PacketWithAddrs(payload []byte, h *Headers, src, dst tcp t.SetChecksum(^t.CalculateChecksum(xsum)) // Inject packet. - c.linkEP.InjectInbound(ipv6.ProtocolNumber, tcpip.PacketBuffer{ + c.linkEP.InjectInbound(ipv6.ProtocolNumber, stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) } diff --git a/pkg/tcpip/transport/udp/BUILD b/pkg/tcpip/transport/udp/BUILD index adc908e24..b5d2d0ba6 100644 --- a/pkg/tcpip/transport/udp/BUILD +++ b/pkg/tcpip/transport/udp/BUILD @@ -32,7 +32,6 @@ go_library( "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/header", - "//pkg/tcpip/iptables", "//pkg/tcpip/ports", "//pkg/tcpip/stack", "//pkg/tcpip/transport/raw", diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 0af4514e1..a3372ac58 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -19,7 +19,6 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" - "gvisor.dev/gvisor/pkg/tcpip/iptables" "gvisor.dev/gvisor/pkg/tcpip/ports" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/waiter" @@ -234,7 +233,7 @@ func (e *endpoint) Close() { func (e *endpoint) ModerateRecvBuf(copied int) {} // IPTables implements tcpip.Endpoint.IPTables. -func (e *endpoint) IPTables() (iptables.IPTables, error) { +func (e *endpoint) IPTables() (stack.IPTables, error) { return e.stack.IPTables(), nil } @@ -913,7 +912,7 @@ func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort u if useDefaultTTL { ttl = r.DefaultTTL() } - if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}, tcpip.PacketBuffer{ + if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}, stack.PacketBuffer{ Header: hdr, Data: data, TransportHeader: buffer.View(udp), @@ -1260,7 +1259,7 @@ func (e *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { // HandlePacket is called by the stack when new packets arrive to this transport // endpoint. -func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt tcpip.PacketBuffer) { +func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) { // Get the header then trim it from the view. hdr := header.UDP(pkt.Data.First()) if int(hdr.Length()) > pkt.Data.Size() { @@ -1327,7 +1326,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk } // HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket. -func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.ControlType, extra uint32, pkt tcpip.PacketBuffer) { +func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.ControlType, extra uint32, pkt stack.PacketBuffer) { } // State implements tcpip.Endpoint.State. diff --git a/pkg/tcpip/transport/udp/forwarder.go b/pkg/tcpip/transport/udp/forwarder.go index fc706ede2..a674ceb68 100644 --- a/pkg/tcpip/transport/udp/forwarder.go +++ b/pkg/tcpip/transport/udp/forwarder.go @@ -43,7 +43,7 @@ func NewForwarder(s *stack.Stack, handler func(*ForwarderRequest)) *Forwarder { // // This function is expected to be passed as an argument to the // stack.SetTransportProtocolHandler function. -func (f *Forwarder) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt tcpip.PacketBuffer) bool { +func (f *Forwarder) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) bool { f.handler(&ForwarderRequest{ stack: f.stack, route: r, @@ -61,7 +61,7 @@ type ForwarderRequest struct { stack *stack.Stack route *stack.Route id stack.TransportEndpointID - pkt tcpip.PacketBuffer + pkt stack.PacketBuffer } // ID returns the 4-tuple (src address, src port, dst address, dst port) that diff --git a/pkg/tcpip/transport/udp/protocol.go b/pkg/tcpip/transport/udp/protocol.go index 8df089d22..6e31a9bac 100644 --- a/pkg/tcpip/transport/udp/protocol.go +++ b/pkg/tcpip/transport/udp/protocol.go @@ -66,7 +66,7 @@ func (*protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) { // HandleUnknownDestinationPacket handles packets targeted at this protocol but // that don't match any existing endpoint. -func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, pkt tcpip.PacketBuffer) bool { +func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) bool { // Get the header then trim it from the view. hdr := header.UDP(pkt.Data.First()) if int(hdr.Length()) > pkt.Data.Size() { @@ -135,7 +135,7 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans pkt.SetType(header.ICMPv4DstUnreachable) pkt.SetCode(header.ICMPv4PortUnreachable) pkt.SetChecksum(header.ICMPv4Checksum(pkt, payload)) - r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, stack.PacketBuffer{ Header: hdr, Data: payload, }) @@ -172,7 +172,7 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans pkt.SetType(header.ICMPv6DstUnreachable) pkt.SetCode(header.ICMPv6PortUnreachable) pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, payload)) - r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, tcpip.PacketBuffer{ + r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, stack.PacketBuffer{ Header: hdr, Data: payload, }) diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index 34b7c2360..0905726c1 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -439,7 +439,7 @@ func (c *testContext) injectV6Packet(payload []byte, h *header4Tuple, valid bool u.SetChecksum(^u.CalculateChecksum(xsum)) // Inject packet. - c.linkEP.InjectInbound(ipv6.ProtocolNumber, tcpip.PacketBuffer{ + c.linkEP.InjectInbound(ipv6.ProtocolNumber, stack.PacketBuffer{ Data: buf.ToVectorisedView(), NetworkHeader: buffer.View(ip), TransportHeader: buffer.View(u), @@ -486,7 +486,7 @@ func (c *testContext) injectV4Packet(payload []byte, h *header4Tuple, valid bool // Inject packet. - c.linkEP.InjectInbound(ipv4.ProtocolNumber, tcpip.PacketBuffer{ + c.linkEP.InjectInbound(ipv4.ProtocolNumber, stack.PacketBuffer{ Data: buf.ToVectorisedView(), NetworkHeader: buffer.View(ip), TransportHeader: buffer.View(u), -- cgit v1.2.3 From d8c4eff3f77b2a5dde34389033fe0ce4589b2e82 Mon Sep 17 00:00:00 2001 From: Bhasker Hariharan Date: Wed, 25 Mar 2020 08:10:11 -0700 Subject: Automated rollback of changelist 301837227 PiperOrigin-RevId: 302891559 --- pkg/sentry/socket/netstack/netstack.go | 22 ++++++++++++++++++---- 1 file changed, 18 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index c19f5639b..f14c336b9 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -29,6 +29,7 @@ import ( "io" "math" "reflect" + "sync/atomic" "syscall" "time" @@ -264,8 +265,14 @@ type SocketOperations struct { skType linux.SockType protocol int + // readViewHasData is 1 iff readView has data to be read, 0 otherwise. + // Must be accessed using atomic operations. It must only be written + // with readMu held but can be read without holding readMu. The latter + // is required to avoid deadlocks in epoll Readiness checks. + readViewHasData uint32 + // readMu protects access to the below fields. - readMu sync.RWMutex `state:"nosave"` + readMu sync.Mutex `state:"nosave"` // readView contains the remaining payload from the last packet. readView buffer.View // readCM holds control message information for the last packet read @@ -421,11 +428,13 @@ func (s *SocketOperations) fetchReadView() *syserr.Error { v, cms, err := s.Endpoint.Read(&s.sender) if err != nil { + atomic.StoreUint32(&s.readViewHasData, 0) return syserr.TranslateNetstackError(err) } s.readView = v s.readCM = cms + atomic.StoreUint32(&s.readViewHasData, 1) return nil } @@ -624,11 +633,9 @@ func (s *SocketOperations) Readiness(mask waiter.EventMask) waiter.EventMask { // Check our cached value iff the caller asked for readability and the // endpoint itself is currently not readable. if (mask & ^r & waiter.EventIn) != 0 { - s.readMu.RLock() - if len(s.readView) > 0 { + if atomic.LoadUint32(&s.readViewHasData) == 1 { r |= waiter.EventIn } - s.readMu.RUnlock() } return r @@ -2335,6 +2342,9 @@ func (s *SocketOperations) coalescingRead(ctx context.Context, dst usermem.IOSeq } copied += n s.readView.TrimFront(n) + if len(s.readView) == 0 { + atomic.StoreUint32(&s.readViewHasData, 0) + } dst = dst.DropFirst(n) if e != nil { @@ -2458,6 +2468,10 @@ func (s *SocketOperations) nonBlockingRead(ctx context.Context, dst usermem.IOSe s.readView.TrimFront(int(n)) } + if len(s.readView) == 0 { + atomic.StoreUint32(&s.readViewHasData, 0) + } + var flags int if msgLen > int(n) { flags |= linux.MSG_TRUNC -- cgit v1.2.3 From c7f5673529af758c9f7c95523535174c7929dab5 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Wed, 25 Mar 2020 14:44:18 -0700 Subject: Set file mode and type to attribute Makes less error prone to find file type. Updates #1197 PiperOrigin-RevId: 302974244 --- pkg/sentry/fsimpl/tmpfs/device_file.go | 10 ++++++++++ pkg/sentry/fsimpl/tmpfs/directory.go | 7 +------ pkg/sentry/fsimpl/tmpfs/named_pipe.go | 2 +- pkg/sentry/fsimpl/tmpfs/regular_file.go | 2 +- pkg/sentry/fsimpl/tmpfs/symlink.go | 3 ++- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 26 ++++++++++++-------------- 6 files changed, 27 insertions(+), 23 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/tmpfs/device_file.go b/pkg/sentry/fsimpl/tmpfs/device_file.go index 84b181b90..83bf885ee 100644 --- a/pkg/sentry/fsimpl/tmpfs/device_file.go +++ b/pkg/sentry/fsimpl/tmpfs/device_file.go @@ -15,6 +15,8 @@ package tmpfs import ( + "fmt" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" @@ -33,6 +35,14 @@ func (fs *filesystem) newDeviceFile(creds *auth.Credentials, mode linux.FileMode major: major, minor: minor, } + switch kind { + case vfs.BlockDevice: + mode |= linux.S_IFBLK + case vfs.CharDevice: + mode |= linux.S_IFCHR + default: + panic(fmt.Sprintf("invalid DeviceKind: %v", kind)) + } file.inode.init(file, fs, creds, mode) file.inode.nlink = 1 // from parent directory return &file.inode diff --git a/pkg/sentry/fsimpl/tmpfs/directory.go b/pkg/sentry/fsimpl/tmpfs/directory.go index b4380af38..37c75ab64 100644 --- a/pkg/sentry/fsimpl/tmpfs/directory.go +++ b/pkg/sentry/fsimpl/tmpfs/directory.go @@ -34,16 +34,11 @@ type directory struct { func (fs *filesystem) newDirectory(creds *auth.Credentials, mode linux.FileMode) *inode { dir := &directory{} - dir.inode.init(dir, fs, creds, mode) + dir.inode.init(dir, fs, creds, linux.S_IFDIR|mode) dir.inode.nlink = 2 // from "." and parent directory or ".." for root return &dir.inode } -func (i *inode) isDir() bool { - _, ok := i.impl.(*directory) - return ok -} - type directoryFD struct { fileDescription vfs.DirectoryFileDescriptionDefaultImpl diff --git a/pkg/sentry/fsimpl/tmpfs/named_pipe.go b/pkg/sentry/fsimpl/tmpfs/named_pipe.go index 0c57fdca3..2c5c739df 100644 --- a/pkg/sentry/fsimpl/tmpfs/named_pipe.go +++ b/pkg/sentry/fsimpl/tmpfs/named_pipe.go @@ -34,7 +34,7 @@ type namedPipe struct { // * rp.Mount().CheckBeginWrite() has been called successfully. func (fs *filesystem) newNamedPipe(creds *auth.Credentials, mode linux.FileMode) *inode { file := &namedPipe{pipe: pipe.NewVFSPipe(pipe.DefaultPipeSize, usermem.PageSize)} - file.inode.init(file, fs, creds, mode) + file.inode.init(file, fs, creds, linux.S_IFIFO|mode) file.inode.nlink = 1 // Only the parent has a link. return &file.inode } diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index 5a2896bf6..26cd65605 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -89,7 +89,7 @@ func (fs *filesystem) newRegularFile(creds *auth.Credentials, mode linux.FileMod file := ®ularFile{ memFile: fs.memFile, } - file.inode.init(file, fs, creds, mode) + file.inode.init(file, fs, creds, linux.S_IFREG|mode) file.inode.nlink = 1 // from parent directory return &file.inode } diff --git a/pkg/sentry/fsimpl/tmpfs/symlink.go b/pkg/sentry/fsimpl/tmpfs/symlink.go index 5246aca84..47e075ed4 100644 --- a/pkg/sentry/fsimpl/tmpfs/symlink.go +++ b/pkg/sentry/fsimpl/tmpfs/symlink.go @@ -15,6 +15,7 @@ package tmpfs import ( + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ) @@ -27,7 +28,7 @@ func (fs *filesystem) newSymlink(creds *auth.Credentials, target string) *inode link := &symlink{ target: target, } - link.inode.init(link, fs, creds, 0777) + link.inode.init(link, fs, creds, linux.S_IFLNK|0777) link.inode.nlink = 1 // from parent directory return &link.inode } diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index ff69372b3..2d5070a46 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -144,7 +144,7 @@ type inode struct { // Inode metadata. Writing multiple fields atomically requires holding // mu, othewise atomic operations can be used. mu sync.Mutex - mode uint32 // excluding file type bits, which are based on impl + mode uint32 // file type and mode nlink uint32 // protected by filesystem.mu instead of inode.mu uid uint32 // auth.KUID, but stored as raw uint32 for sync/atomic gid uint32 // auth.KGID, but ... @@ -168,6 +168,9 @@ type inode struct { const maxLinks = math.MaxUint32 func (i *inode) init(impl interface{}, fs *filesystem, creds *auth.Credentials, mode linux.FileMode) { + if mode.FileType() == 0 { + panic("file type is required in FileMode") + } i.clock = fs.clock i.refs = 1 i.mode = uint32(mode) @@ -269,31 +272,21 @@ func (i *inode) statTo(stat *linux.Statx) { // TODO(gvisor.dev/issues/1197): Device number. switch impl := i.impl.(type) { case *regularFile: - stat.Mode |= linux.S_IFREG stat.Mask |= linux.STATX_SIZE | linux.STATX_BLOCKS stat.Size = uint64(atomic.LoadUint64(&impl.size)) // In tmpfs, this will be FileRangeSet.Span() / 512 (but also cached in // a uint64 accessed using atomic memory operations to avoid taking // locks). stat.Blocks = allocatedBlocksForSize(stat.Size) - case *directory: - stat.Mode |= linux.S_IFDIR case *symlink: - stat.Mode |= linux.S_IFLNK stat.Mask |= linux.STATX_SIZE | linux.STATX_BLOCKS stat.Size = uint64(len(impl.target)) stat.Blocks = allocatedBlocksForSize(stat.Size) - case *namedPipe: - stat.Mode |= linux.S_IFIFO case *deviceFile: - switch impl.kind { - case vfs.BlockDevice: - stat.Mode |= linux.S_IFBLK - case vfs.CharDevice: - stat.Mode |= linux.S_IFCHR - } stat.RdevMajor = impl.major stat.RdevMinor = impl.minor + case *directory, *namedPipe: + // Nothing to do. default: panic(fmt.Sprintf("unknown inode type: %T", i.impl)) } @@ -316,7 +309,8 @@ func (i *inode) setStat(ctx context.Context, creds *auth.Credentials, stat *linu ) mask := stat.Mask if mask&linux.STATX_MODE != 0 { - atomic.StoreUint32(&i.mode, uint32(stat.Mode)) + ft := atomic.LoadUint32(&i.mode) & linux.S_IFMT + atomic.StoreUint32(&i.mode, ft|uint32(stat.Mode&^linux.S_IFMT)) needsCtimeBump = true } if mask&linux.STATX_UID != 0 { @@ -439,6 +433,10 @@ func (i *inode) direntType() uint8 { } } +func (i *inode) isDir() bool { + return linux.FileMode(i.mode).FileType() == linux.S_IFDIR +} + // fileDescription is embedded by tmpfs implementations of // vfs.FileDescriptionImpl. type fileDescription struct { -- cgit v1.2.3 From e541ebec2fdb5b29209cb3fc8235b77edcaebb6a Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Wed, 25 Mar 2020 14:54:10 -0700 Subject: Misc fixes to make stat_test pass (almost) The only test failing now requires socket which is not available in VFS2 yet. Updates #1198 PiperOrigin-RevId: 302976572 --- pkg/bits/bits_template.go | 8 ++++++ pkg/bits/uint64_test.go | 18 ++++++++++++ pkg/sentry/fsimpl/gofer/filesystem.go | 16 +++++++++-- pkg/sentry/fsimpl/gofer/gofer.go | 41 ++++++++++++++++++++++++---- pkg/sentry/syscalls/linux/vfs2/BUILD | 1 + pkg/sentry/syscalls/linux/vfs2/filesystem.go | 2 +- pkg/sentry/syscalls/linux/vfs2/getdents.go | 4 +-- pkg/sentry/syscalls/linux/vfs2/stat.go | 7 ++++- pkg/sentry/vfs/resolving_path.go | 16 +++++++++-- test/syscalls/linux/stat.cc | 11 +++++++- 10 files changed, 109 insertions(+), 15 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/bits/bits_template.go b/pkg/bits/bits_template.go index 93a435b80..998645388 100644 --- a/pkg/bits/bits_template.go +++ b/pkg/bits/bits_template.go @@ -42,3 +42,11 @@ func Mask(is ...int) T { func MaskOf(i int) T { return T(1) << T(i) } + +// IsPowerOfTwo returns true if v is power of 2. +func IsPowerOfTwo(v T) bool { + if v == 0 { + return false + } + return v&(v-1) == 0 +} diff --git a/pkg/bits/uint64_test.go b/pkg/bits/uint64_test.go index 1b018d808..193d1ebcd 100644 --- a/pkg/bits/uint64_test.go +++ b/pkg/bits/uint64_test.go @@ -114,3 +114,21 @@ func TestIsOn(t *testing.T) { } } } + +func TestIsPowerOfTwo(t *testing.T) { + for _, tc := range []struct { + v uint64 + want bool + }{ + {v: 0, want: false}, + {v: 1, want: true}, + {v: 2, want: true}, + {v: 3, want: false}, + {v: 4, want: true}, + {v: 5, want: false}, + } { + if got := IsPowerOfTwo64(tc.v); got != tc.want { + t.Errorf("IsPowerOfTwo(%d) = %t, want: %t", tc.v, got, tc.want) + } + } +} diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 38e4cdbc5..26b492185 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -454,6 +454,9 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b } if fs.opts.interop != InteropModeShared { parent.touchCMtime(ctx) + if dir { + parent.decLinks() + } parent.cacheNegativeChildLocked(name) parent.dirents = nil } @@ -569,8 +572,13 @@ func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error { return fs.doCreateAt(ctx, rp, true /* dir */, func(parent *dentry, name string) error { creds := rp.Credentials() - _, err := parent.file.mkdir(ctx, name, (p9.FileMode)(opts.Mode), (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)) - return err + if _, err := parent.file.mkdir(ctx, name, (p9.FileMode)(opts.Mode), (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)); err != nil { + return err + } + if fs.opts.interop != InteropModeShared { + parent.incLinks() + } + return nil }) } @@ -962,6 +970,10 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa oldParent.dirents = nil delete(newParent.negativeChildren, newName) newParent.dirents = nil + if renamed.isDir() { + oldParent.decLinks() + newParent.incLinks() + } } vfsObj.CommitRenameReplaceDentry(&renamed.vfsd, &newParent.vfsd, newName, replacedVFSD) return nil diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 999485492..13928ce36 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -485,6 +485,11 @@ type dentry struct { // locked to mutate it). size uint64 + // nlink counts the number of hard links to this dentry. It's updated and + // accessed using atomic operations. It's not protected by metadataMu like the + // other metadata fields. + nlink uint32 + mapsMu sync.Mutex // If this dentry represents a regular file, mappings tracks mappings of @@ -604,6 +609,9 @@ func (fs *filesystem) newDentry(ctx context.Context, file p9file, qid p9.QID, ma if mask.BTime { d.btime = dentryTimestampFromP9(attr.BTimeSeconds, attr.BTimeNanoSeconds) } + if mask.NLink { + d.nlink = uint32(attr.NLink) + } d.vfsd.Init(d) fs.syncMu.Lock() @@ -645,6 +653,9 @@ func (d *dentry) updateFromP9Attrs(mask p9.AttrMask, attr *p9.Attr) { if mask.BTime { atomic.StoreInt64(&d.btime, dentryTimestampFromP9(attr.BTimeSeconds, attr.BTimeNanoSeconds)) } + if mask.NLink { + atomic.StoreUint32(&d.nlink, uint32(attr.NLink)) + } if mask.Size { d.dataMu.Lock() atomic.StoreUint64(&d.size, attr.Size) @@ -687,10 +698,7 @@ func (d *dentry) fileType() uint32 { func (d *dentry) statTo(stat *linux.Statx) { stat.Mask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_NLINK | linux.STATX_UID | linux.STATX_GID | linux.STATX_ATIME | linux.STATX_MTIME | linux.STATX_CTIME | linux.STATX_INO | linux.STATX_SIZE | linux.STATX_BLOCKS | linux.STATX_BTIME stat.Blksize = atomic.LoadUint32(&d.blockSize) - stat.Nlink = 1 - if d.isDir() { - stat.Nlink = 2 - } + stat.Nlink = atomic.LoadUint32(&d.nlink) stat.UID = atomic.LoadUint32(&d.uid) stat.GID = atomic.LoadUint32(&d.gid) stat.Mode = uint16(atomic.LoadUint32(&d.mode)) @@ -703,7 +711,7 @@ func (d *dentry) statTo(stat *linux.Statx) { stat.Btime = statxTimestampFromDentry(atomic.LoadInt64(&d.btime)) stat.Ctime = statxTimestampFromDentry(atomic.LoadInt64(&d.ctime)) stat.Mtime = statxTimestampFromDentry(atomic.LoadInt64(&d.mtime)) - // TODO(jamieliu): device number + // TODO(gvisor.dev/issue/1198): device number } func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *linux.Statx, mnt *vfs.Mount) error { @@ -1094,6 +1102,26 @@ func (d *dentry) ensureSharedHandle(ctx context.Context, read, write, trunc bool return nil } +// incLinks increments link count. +// +// Preconditions: d.nlink != 0 && d.nlink < math.MaxUint32. +func (d *dentry) incLinks() { + v := atomic.AddUint32(&d.nlink, 1) + if v < 2 { + panic(fmt.Sprintf("dentry.nlink is invalid (was 0 or overflowed): %d", v)) + } +} + +// decLinks decrements link count. +// +// Preconditions: d.nlink > 1. +func (d *dentry) decLinks() { + v := atomic.AddUint32(&d.nlink, ^uint32(0)) + if v == 0 { + panic(fmt.Sprintf("dentry.nlink must be greater than 0: %d", v)) + } +} + // fileDescription is embedded by gofer implementations of // vfs.FileDescriptionImpl. type fileDescription struct { @@ -1112,7 +1140,8 @@ func (fd *fileDescription) dentry() *dentry { // Stat implements vfs.FileDescriptionImpl.Stat. func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { d := fd.dentry() - if d.fs.opts.interop == InteropModeShared && opts.Mask&(linux.STATX_MODE|linux.STATX_UID|linux.STATX_GID|linux.STATX_ATIME|linux.STATX_MTIME|linux.STATX_CTIME|linux.STATX_SIZE|linux.STATX_BLOCKS|linux.STATX_BTIME) != 0 && opts.Sync != linux.AT_STATX_DONT_SYNC { + const validMask = uint32(linux.STATX_MODE | linux.STATX_UID | linux.STATX_GID | linux.STATX_ATIME | linux.STATX_MTIME | linux.STATX_CTIME | linux.STATX_SIZE | linux.STATX_BLOCKS | linux.STATX_BTIME) + if d.fs.opts.interop == InteropModeShared && opts.Mask&(validMask) != 0 && opts.Sync != linux.AT_STATX_DONT_SYNC { // TODO(jamieliu): Use specialFileFD.handle.file for the getattr if // available? if err := d.updateFromGetattr(ctx); err != nil { diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD index e7695e995..2eb210014 100644 --- a/pkg/sentry/syscalls/linux/vfs2/BUILD +++ b/pkg/sentry/syscalls/linux/vfs2/BUILD @@ -31,6 +31,7 @@ go_library( visibility = ["//:sandbox"], deps = [ "//pkg/abi/linux", + "//pkg/bits", "//pkg/fspath", "//pkg/gohacks", "//pkg/sentry/arch", diff --git a/pkg/sentry/syscalls/linux/vfs2/filesystem.go b/pkg/sentry/syscalls/linux/vfs2/filesystem.go index fc5ceea4c..a859095e2 100644 --- a/pkg/sentry/syscalls/linux/vfs2/filesystem.go +++ b/pkg/sentry/syscalls/linux/vfs2/filesystem.go @@ -250,7 +250,7 @@ func rmdirat(t *kernel.Task, dirfd int32, pathAddr usermem.Addr) error { if err != nil { return err } - tpop, err := getTaskPathOperation(t, dirfd, path, disallowEmptyPath, followFinalSymlink) + tpop, err := getTaskPathOperation(t, dirfd, path, disallowEmptyPath, nofollowFinalSymlink) if err != nil { return err } diff --git a/pkg/sentry/syscalls/linux/vfs2/getdents.go b/pkg/sentry/syscalls/linux/vfs2/getdents.go index ddc140b65..a61cc5059 100644 --- a/pkg/sentry/syscalls/linux/vfs2/getdents.go +++ b/pkg/sentry/syscalls/linux/vfs2/getdents.go @@ -97,7 +97,7 @@ func (cb *getdentsCallback) Handle(dirent vfs.Dirent) error { // char d_name[]; /* Filename (null-terminated) */ // }; size := 8 + 8 + 2 + 1 + 1 + len(dirent.Name) - if size < cb.remaining { + if size > cb.remaining { return syserror.EINVAL } buf = cb.t.CopyScratchBuffer(size) @@ -125,7 +125,7 @@ func (cb *getdentsCallback) Handle(dirent vfs.Dirent) error { panic(fmt.Sprintf("unsupported sizeof(unsigned long): %d", cb.t.Arch().Width())) } size := 8 + 8 + 2 + 1 + 1 + 1 + len(dirent.Name) - if size < cb.remaining { + if size > cb.remaining { return syserror.EINVAL } buf = cb.t.CopyScratchBuffer(size) diff --git a/pkg/sentry/syscalls/linux/vfs2/stat.go b/pkg/sentry/syscalls/linux/vfs2/stat.go index 068243132..fdfe49243 100644 --- a/pkg/sentry/syscalls/linux/vfs2/stat.go +++ b/pkg/sentry/syscalls/linux/vfs2/stat.go @@ -16,6 +16,7 @@ package vfs2 import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/bits" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/gohacks" "gvisor.dev/gvisor/pkg/sentry/arch" @@ -153,7 +154,11 @@ func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if flags&^(linux.AT_EMPTY_PATH|linux.AT_SYMLINK_NOFOLLOW|linux.AT_STATX_SYNC_TYPE) != 0 { return 0, nil, syserror.EINVAL } - + // Make sure that only one sync type option is set. + syncType := uint32(flags & linux.AT_STATX_SYNC_TYPE) + if syncType != 0 && !bits.IsPowerOfTwo32(syncType) { + return 0, nil, syserror.EINVAL + } if mask&linux.STATX__RESERVED != 0 { return 0, nil, syserror.EINVAL } diff --git a/pkg/sentry/vfs/resolving_path.go b/pkg/sentry/vfs/resolving_path.go index eb4ebb511..8f31495da 100644 --- a/pkg/sentry/vfs/resolving_path.go +++ b/pkg/sentry/vfs/resolving_path.go @@ -329,10 +329,22 @@ func (rp *ResolvingPath) ResolveComponent(d *Dentry) (*Dentry, error) { // component in pcs represents a symbolic link, the symbolic link should be // followed. // +// If path is terminated with '/', the '/' is considered the last element and +// any symlink before that is followed: +// - For most non-creating walks, the last path component is handled by +// fs/namei.c:lookup_last(), which sets LOOKUP_FOLLOW if the first byte +// after the path component is non-NULL (which is only possible if it's '/') +// and the path component is of type LAST_NORM. +// +// - For open/openat/openat2 without O_CREAT, the last path component is +// handled by fs/namei.c:do_last(), which does the same, though without the +// LAST_NORM check. +// // Preconditions: !rp.Done(). func (rp *ResolvingPath) ShouldFollowSymlink() bool { - // Non-final symlinks are always followed. - return rp.flags&rpflagsFollowFinalSymlink != 0 || !rp.Final() + // Non-final symlinks are always followed. Paths terminated with '/' are also + // always followed. + return rp.flags&rpflagsFollowFinalSymlink != 0 || !rp.Final() || rp.MustBeDir() } // HandleSymlink is called when the current path component is a symbolic link diff --git a/test/syscalls/linux/stat.cc b/test/syscalls/linux/stat.cc index 513b9cd1c..2503960f3 100644 --- a/test/syscalls/linux/stat.cc +++ b/test/syscalls/linux/stat.cc @@ -34,6 +34,13 @@ #include "test/util/temp_path.h" #include "test/util/test_util.h" +#ifndef AT_STATX_FORCE_SYNC +#define AT_STATX_FORCE_SYNC 0x2000 +#endif +#ifndef AT_STATX_DONT_SYNC +#define AT_STATX_DONT_SYNC 0x4000 +#endif + namespace gvisor { namespace testing { @@ -700,8 +707,10 @@ TEST_F(StatTest, StatxInvalidFlags) { struct kernel_statx stx; EXPECT_THAT(statx(AT_FDCWD, test_file_name_.c_str(), 12345, 0, &stx), SyscallFailsWithErrno(EINVAL)); + + // Sync flags are mutually exclusive. EXPECT_THAT(statx(AT_FDCWD, test_file_name_.c_str(), - 0x6000 /* AT_STATX_SYNC_TYPE */, 0, &stx), + AT_STATX_FORCE_SYNC | AT_STATX_DONT_SYNC, 0, &stx), SyscallFailsWithErrno(EINVAL)); } -- cgit v1.2.3 From f2eba940159b4ed55359f99a3422ad899a8647ee Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Wed, 25 Mar 2020 15:23:48 -0700 Subject: Remove TODO to push down exec permission check Pushing it down requires all implementation to check for exec individualy which is not maintanable. Making it part of GenericCheckPermissions add extra cost to everyone that calls it. So it's better to keep is in VirtualFilesystem.OpenAt. Updates #1193 PiperOrigin-RevId: 302982993 --- pkg/sentry/vfs/vfs.go | 2 -- 1 file changed, 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 2e2880171..03d1fb943 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -402,8 +402,6 @@ func (vfs *VirtualFilesystem) OpenAt(ctx context.Context, creds *auth.Credential if err == nil { vfs.putResolvingPath(rp) - // TODO(gvisor.dev/issue/1193): Move inside fsimpl to avoid another call - // to FileDescription.Stat(). if opts.FileExec { if fd.Mount().flags.NoExec { fd.DecRef() -- cgit v1.2.3 From de694e5484502d53166d70b36141e62fcdf07803 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Wed, 25 Mar 2020 19:12:25 -0700 Subject: Combine file mode and isDir arguments Updates #1035 PiperOrigin-RevId: 303021328 --- pkg/abi/linux/file.go | 5 +++++ pkg/sentry/fsimpl/ext/inode.go | 2 +- pkg/sentry/fsimpl/gofer/filesystem.go | 22 +++++++++++----------- pkg/sentry/fsimpl/gofer/gofer.go | 7 ++++--- pkg/sentry/fsimpl/host/host.go | 6 +++--- pkg/sentry/fsimpl/kernfs/fd_impl_util.go | 3 +-- pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 6 ++---- pkg/sentry/fsimpl/proc/task.go | 9 +-------- pkg/sentry/fsimpl/tmpfs/filesystem.go | 24 ++++++++++++------------ pkg/sentry/fsimpl/tmpfs/tmpfs.go | 8 +++++--- pkg/sentry/vfs/anonfs.go | 2 +- pkg/sentry/vfs/permissions.go | 23 ++++++++++------------- 12 files changed, 56 insertions(+), 61 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/file.go b/pkg/abi/linux/file.go index dbe58acbe..055ac1d7c 100644 --- a/pkg/abi/linux/file.go +++ b/pkg/abi/linux/file.go @@ -287,6 +287,11 @@ func (m FileMode) ExtraBits() FileMode { return m &^ (PermissionsMask | FileTypeMask) } +// IsDir returns true if file type represents a directory. +func (m FileMode) IsDir() bool { + return m.FileType() == S_IFDIR +} + // String returns a string representation of m. func (m FileMode) String() string { var s []string diff --git a/pkg/sentry/fsimpl/ext/inode.go b/pkg/sentry/fsimpl/ext/inode.go index 6962083f5..a39a37318 100644 --- a/pkg/sentry/fsimpl/ext/inode.go +++ b/pkg/sentry/fsimpl/ext/inode.go @@ -186,7 +186,7 @@ func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts *vfs.OpenOpt } func (in *inode) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes) error { - return vfs.GenericCheckPermissions(creds, ats, in.isDir(), uint16(in.diskInode.Mode()), in.diskInode.UID(), in.diskInode.GID()) + return vfs.GenericCheckPermissions(creds, ats, in.diskInode.Mode(), in.diskInode.UID(), in.diskInode.GID()) } // statTo writes the statx fields to the output parameter. diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 26b492185..1e43df9ec 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -119,7 +119,7 @@ func (fs *filesystem) stepLocked(ctx context.Context, rp *vfs.ResolvingPath, d * if !d.isDir() { return nil, syserror.ENOTDIR } - if err := d.checkPermissions(rp.Credentials(), vfs.MayExec, true); err != nil { + if err := d.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { return nil, err } afterSymlink: @@ -314,7 +314,7 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir if err != nil { return err } - if err := parent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true); err != nil { + if err := parent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { return err } if parent.isDeleted() { @@ -378,7 +378,7 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b if err != nil { return err } - if err := parent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true); err != nil { + if err := parent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { return err } if err := rp.Mount().CheckBeginWrite(); err != nil { @@ -512,7 +512,7 @@ func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds if err != nil { return err } - return d.checkPermissions(creds, ats, d.isDir()) + return d.checkPermissions(creds, ats) } // GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. @@ -528,7 +528,7 @@ func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, op if !d.isDir() { return nil, syserror.ENOTDIR } - if err := d.checkPermissions(rp.Credentials(), vfs.MayExec, true); err != nil { + if err := d.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { return nil, err } } @@ -624,7 +624,7 @@ afterTrailingSymlink: return nil, err } // Check for search permission in the parent directory. - if err := parent.checkPermissions(rp.Credentials(), vfs.MayExec, true); err != nil { + if err := parent.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { return nil, err } // Determine whether or not we need to create a file. @@ -661,7 +661,7 @@ afterTrailingSymlink: // Preconditions: fs.renameMu must be locked. func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.OpenOptions) (*vfs.FileDescription, error) { ats := vfs.AccessTypesForOpenFlags(opts) - if err := d.checkPermissions(rp.Credentials(), ats, d.isDir()); err != nil { + if err := d.checkPermissions(rp.Credentials(), ats); err != nil { return nil, err } mnt := rp.Mount() @@ -722,7 +722,7 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vf // Preconditions: d.fs.renameMu must be locked. d.dirMu must be locked. func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.OpenOptions) (*vfs.FileDescription, error) { - if err := d.checkPermissions(rp.Credentials(), vfs.MayWrite, true); err != nil { + if err := d.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil { return nil, err } if d.isDeleted() { @@ -884,7 +884,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa return err } } - if err := oldParent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true); err != nil { + if err := oldParent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { return err } vfsObj := rp.VirtualFilesystem() @@ -904,7 +904,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa return syserror.EINVAL } if oldParent != newParent { - if err := renamed.checkPermissions(rp.Credentials(), vfs.MayWrite, true); err != nil { + if err := renamed.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil { return err } } @@ -915,7 +915,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa } if oldParent != newParent { - if err := newParent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true); err != nil { + if err := newParent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { return err } newParent.dirMu.Lock() diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 13928ce36..cf276a417 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -721,7 +721,8 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin if stat.Mask&^(linux.STATX_MODE|linux.STATX_UID|linux.STATX_GID|linux.STATX_ATIME|linux.STATX_MTIME|linux.STATX_SIZE) != 0 { return syserror.EPERM } - if err := vfs.CheckSetStat(ctx, creds, stat, uint16(atomic.LoadUint32(&d.mode))&^linux.S_IFMT, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))); err != nil { + mode := linux.FileMode(atomic.LoadUint32(&d.mode)) + if err := vfs.CheckSetStat(ctx, creds, stat, mode, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))); err != nil { return err } if err := mnt.CheckBeginWrite(); err != nil { @@ -843,8 +844,8 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin return nil } -func (d *dentry) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes, isDir bool) error { - return vfs.GenericCheckPermissions(creds, ats, isDir, uint16(atomic.LoadUint32(&d.mode))&0777, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))) +func (d *dentry) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes) error { + return vfs.GenericCheckPermissions(creds, ats, linux.FileMode(atomic.LoadUint32(&d.mode)), auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))) } // IncRef implements vfs.DentryImpl.IncRef. diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 1f735628f..a54985ef5 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -167,8 +167,8 @@ func fileFlagsFromHostFD(fd int) (int, error) { } // CheckPermissions implements kernfs.Inode. -func (i *inode) CheckPermissions(ctx context.Context, creds *auth.Credentials, atx vfs.AccessTypes) error { - return vfs.GenericCheckPermissions(creds, atx, false /* isDir */, uint16(i.mode), i.uid, i.gid) +func (i *inode) CheckPermissions(ctx context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error { + return vfs.GenericCheckPermissions(creds, ats, i.mode, i.uid, i.gid) } // Mode implements kernfs.Inode. @@ -306,7 +306,7 @@ func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Cre if m&^(linux.STATX_MODE|linux.STATX_SIZE|linux.STATX_ATIME|linux.STATX_MTIME) != 0 { return syserror.EPERM } - if err := vfs.CheckSetStat(ctx, creds, &s, uint16(i.Mode().Permissions()), i.uid, i.gid); err != nil { + if err := vfs.CheckSetStat(ctx, creds, &s, i.Mode(), i.uid, i.gid); err != nil { return err } diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index 75c4bab1a..bfa786c88 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -206,8 +206,7 @@ func (fd *GenericDirectoryFD) Stat(ctx context.Context, opts vfs.StatOptions) (l // SetStat implements vfs.FileDescriptionImpl.SetStat. func (fd *GenericDirectoryFD) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { - fs := fd.filesystem() creds := auth.CredentialsFromContext(ctx) inode := fd.vfsfd.VirtualDentry().Dentry().Impl().(*Dentry).inode - return inode.SetStat(ctx, fs, creds, opts) + return inode.SetStat(ctx, fd.filesystem(), creds, opts) } diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index c612dcf07..5c84b10c9 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -241,7 +241,7 @@ func (a *InodeAttrs) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *aut if opts.Stat.Mask&^(linux.STATX_MODE|linux.STATX_UID|linux.STATX_GID) != 0 { return syserror.EPERM } - if err := vfs.CheckSetStat(ctx, creds, &opts.Stat, uint16(a.Mode().Permissions()), auth.KUID(atomic.LoadUint32(&a.uid)), auth.KGID(atomic.LoadUint32(&a.gid))); err != nil { + if err := vfs.CheckSetStat(ctx, creds, &opts.Stat, a.Mode(), auth.KUID(atomic.LoadUint32(&a.uid)), auth.KGID(atomic.LoadUint32(&a.gid))); err != nil { return err } @@ -273,12 +273,10 @@ func (a *InodeAttrs) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *aut // CheckPermissions implements Inode.CheckPermissions. func (a *InodeAttrs) CheckPermissions(_ context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error { - mode := a.Mode() return vfs.GenericCheckPermissions( creds, ats, - mode.FileType() == linux.ModeDirectory, - uint16(mode), + a.Mode(), auth.KUID(atomic.LoadUint32(&a.uid)), auth.KGID(atomic.LoadUint32(&a.gid)), ) diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index 49d6efb0e..aee2a4392 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -172,14 +172,7 @@ func (i *taskOwnedInode) Stat(fs *vfs.Filesystem, opts vfs.StatOptions) (linux.S func (i *taskOwnedInode) CheckPermissions(_ context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error { mode := i.Mode() uid, gid := i.getOwner(mode) - return vfs.GenericCheckPermissions( - creds, - ats, - mode.FileType() == linux.ModeDirectory, - uint16(mode), - uid, - gid, - ) + return vfs.GenericCheckPermissions(creds, ats, mode, uid, gid) } func (i *taskOwnedInode) getOwner(mode linux.FileMode) (auth.KUID, auth.KGID) { diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 75d01b853..12cc64385 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -41,7 +41,7 @@ func stepLocked(rp *vfs.ResolvingPath, d *dentry) (*dentry, error) { if !d.inode.isDir() { return nil, syserror.ENOTDIR } - if err := d.inode.checkPermissions(rp.Credentials(), vfs.MayExec, true); err != nil { + if err := d.inode.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { return nil, err } afterSymlink: @@ -125,7 +125,7 @@ func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(pa if err != nil { return err } - if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true /* isDir */); err != nil { + if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { return err } name := rp.Component() @@ -163,7 +163,7 @@ func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds if err != nil { return err } - return d.inode.checkPermissions(creds, ats, d.inode.isDir()) + return d.inode.checkPermissions(creds, ats) } // GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. @@ -178,7 +178,7 @@ func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, op if !d.inode.isDir() { return nil, syserror.ENOTDIR } - if err := d.inode.checkPermissions(rp.Credentials(), vfs.MayExec, true /* isDir */); err != nil { + if err := d.inode.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { return nil, err } } @@ -301,7 +301,7 @@ afterTrailingSymlink: return nil, err } // Check for search permission in the parent directory. - if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayExec, true); err != nil { + if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { return nil, err } // Reject attempts to open directories with O_CREAT. @@ -316,7 +316,7 @@ afterTrailingSymlink: child, err := stepLocked(rp, parent) if err == syserror.ENOENT { // Already checked for searchability above; now check for writability. - if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite, true); err != nil { + if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil { return nil, err } if err := rp.Mount().CheckBeginWrite(); err != nil { @@ -347,7 +347,7 @@ afterTrailingSymlink: func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.OpenOptions, afterCreate bool) (*vfs.FileDescription, error) { ats := vfs.AccessTypesForOpenFlags(opts) if !afterCreate { - if err := d.inode.checkPermissions(rp.Credentials(), ats, d.inode.isDir()); err != nil { + if err := d.inode.checkPermissions(rp.Credentials(), ats); err != nil { return nil, err } } @@ -428,7 +428,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa defer mnt.EndWrite() oldParent := oldParentVD.Dentry().Impl().(*dentry) - if err := oldParent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true /* isDir */); err != nil { + if err := oldParent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { return err } // Call vfs.Dentry.Child() instead of stepLocked() or rp.ResolveChild(), @@ -445,7 +445,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa } if oldParent != newParent { // Writability is needed to change renamed's "..". - if err := renamed.inode.checkPermissions(rp.Credentials(), vfs.MayWrite, true /* isDir */); err != nil { + if err := renamed.inode.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil { return err } } @@ -455,7 +455,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa } } - if err := newParent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true /* isDir */); err != nil { + if err := newParent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { return err } replacedVFSD := newParent.vfsd.Child(newName) @@ -528,7 +528,7 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error if err != nil { return err } - if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true /* isDir */); err != nil { + if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { return err } name := rp.Component() @@ -621,7 +621,7 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error if err != nil { return err } - if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec, true /* isDir */); err != nil { + if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { return err } name := rp.Component() diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 2d5070a46..2f9e6c876 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -245,8 +245,9 @@ func (i *inode) decRef() { } } -func (i *inode) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes, isDir bool) error { - return vfs.GenericCheckPermissions(creds, ats, isDir, uint16(atomic.LoadUint32(&i.mode)), auth.KUID(atomic.LoadUint32(&i.uid)), auth.KGID(atomic.LoadUint32(&i.gid))) +func (i *inode) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes) error { + mode := linux.FileMode(atomic.LoadUint32(&i.mode)) + return vfs.GenericCheckPermissions(creds, ats, mode, auth.KUID(atomic.LoadUint32(&i.uid)), auth.KGID(atomic.LoadUint32(&i.gid))) } // Go won't inline this function, and returning linux.Statx (which is quite @@ -299,7 +300,8 @@ func (i *inode) setStat(ctx context.Context, creds *auth.Credentials, stat *linu if stat.Mask&^(linux.STATX_MODE|linux.STATX_UID|linux.STATX_GID|linux.STATX_ATIME|linux.STATX_MTIME|linux.STATX_CTIME|linux.STATX_SIZE) != 0 { return syserror.EPERM } - if err := vfs.CheckSetStat(ctx, creds, stat, uint16(atomic.LoadUint32(&i.mode))&^linux.S_IFMT, auth.KUID(atomic.LoadUint32(&i.uid)), auth.KGID(atomic.LoadUint32(&i.gid))); err != nil { + mode := linux.FileMode(atomic.LoadUint32(&i.mode)) + if err := vfs.CheckSetStat(ctx, creds, stat, mode, auth.KUID(atomic.LoadUint32(&i.uid)), auth.KGID(atomic.LoadUint32(&i.gid))); err != nil { return err } i.mu.Lock() diff --git a/pkg/sentry/vfs/anonfs.go b/pkg/sentry/vfs/anonfs.go index 925996517..a62e43589 100644 --- a/pkg/sentry/vfs/anonfs.go +++ b/pkg/sentry/vfs/anonfs.go @@ -83,7 +83,7 @@ func (fs *anonFilesystem) AccessAt(ctx context.Context, rp *ResolvingPath, creds if !rp.Done() { return syserror.ENOTDIR } - return GenericCheckPermissions(creds, ats, false /* isDir */, anonFileMode, anonFileUID, anonFileGID) + return GenericCheckPermissions(creds, ats, anonFileMode, anonFileUID, anonFileGID) } // GetDentryAt implements FilesystemImpl.GetDentryAt. diff --git a/pkg/sentry/vfs/permissions.go b/pkg/sentry/vfs/permissions.go index 2c8f23f55..f9647f90e 100644 --- a/pkg/sentry/vfs/permissions.go +++ b/pkg/sentry/vfs/permissions.go @@ -29,9 +29,9 @@ type AccessTypes uint16 // Bits in AccessTypes. const ( + MayExec AccessTypes = 1 + MayWrite AccessTypes = 2 MayRead AccessTypes = 4 - MayWrite = 2 - MayExec = 1 ) // OnlyRead returns true if access _only_ allows read. @@ -56,16 +56,17 @@ func (a AccessTypes) MayExec() bool { // GenericCheckPermissions checks that creds has the given access rights on a // file with the given permissions, UID, and GID, subject to the rules of -// fs/namei.c:generic_permission(). isDir is true if the file is a directory. -func GenericCheckPermissions(creds *auth.Credentials, ats AccessTypes, isDir bool, mode uint16, kuid auth.KUID, kgid auth.KGID) error { +// fs/namei.c:generic_permission(). +func GenericCheckPermissions(creds *auth.Credentials, ats AccessTypes, mode linux.FileMode, kuid auth.KUID, kgid auth.KGID) error { // Check permission bits. - perms := mode + perms := uint16(mode.Permissions()) if creds.EffectiveKUID == kuid { perms >>= 6 } else if creds.InGroup(kgid) { perms >>= 3 } if uint16(ats)&perms == uint16(ats) { + // All permission bits match, access granted. return nil } @@ -77,7 +78,7 @@ func GenericCheckPermissions(creds *auth.Credentials, ats AccessTypes, isDir boo } // CAP_DAC_READ_SEARCH allows the caller to read and search arbitrary // directories, and read arbitrary non-directory files. - if (isDir && !ats.MayWrite()) || ats.OnlyRead() { + if (mode.IsDir() && !ats.MayWrite()) || ats.OnlyRead() { if creds.HasCapability(linux.CAP_DAC_READ_SEARCH) { return nil } @@ -85,7 +86,7 @@ func GenericCheckPermissions(creds *auth.Credentials, ats AccessTypes, isDir boo // CAP_DAC_OVERRIDE allows arbitrary access to directories, read/write // access to non-directory files, and execute access to non-directory files // for which at least one execute bit is set. - if isDir || !ats.MayExec() || (mode&0111 != 0) { + if mode.IsDir() || !ats.MayExec() || (mode.Permissions()&0111 != 0) { if creds.HasCapability(linux.CAP_DAC_OVERRIDE) { return nil } @@ -151,7 +152,7 @@ func MayWriteFileWithOpenFlags(flags uint32) bool { // CheckSetStat checks that creds has permission to change the metadata of a // file with the given permissions, UID, and GID as specified by stat, subject // to the rules of Linux's fs/attr.c:setattr_prepare(). -func CheckSetStat(ctx context.Context, creds *auth.Credentials, stat *linux.Statx, mode uint16, kuid auth.KUID, kgid auth.KGID) error { +func CheckSetStat(ctx context.Context, creds *auth.Credentials, stat *linux.Statx, mode linux.FileMode, kuid auth.KUID, kgid auth.KGID) error { if stat.Mask&linux.STATX_SIZE != 0 { limit, err := CheckLimit(ctx, 0, int64(stat.Size)) if err != nil { @@ -190,11 +191,7 @@ func CheckSetStat(ctx context.Context, creds *auth.Credentials, stat *linux.Stat (stat.Mask&linux.STATX_CTIME != 0 && stat.Ctime.Nsec != linux.UTIME_NOW) { return syserror.EPERM } - // isDir is irrelevant in the following call to - // GenericCheckPermissions since ats == MayWrite means that - // CAP_DAC_READ_SEARCH does not apply, and CAP_DAC_OVERRIDE - // applies, regardless of isDir. - if err := GenericCheckPermissions(creds, MayWrite, false /* isDir */, mode, kuid, kgid); err != nil { + if err := GenericCheckPermissions(creds, MayWrite, mode, kuid, kgid); err != nil { return err } } -- cgit v1.2.3 From 92b9069b67b927cef25a1490ebd142ad6d65690d Mon Sep 17 00:00:00 2001 From: Nayana Bidari Date: Fri, 20 Mar 2020 12:00:21 -0700 Subject: Support owner matching for iptables. This feature will match UID and GID of the packet creator, for locally generated packets. This match is only valid in the OUTPUT and POSTROUTING chains. Forwarded packets do not have any socket associated with them. Packets from kernel threads do have a socket, but usually no owner. --- pkg/abi/linux/netfilter.go | 41 ++++++++ pkg/abi/linux/netfilter_test.go | 1 + pkg/sentry/kernel/task.go | 12 +++ pkg/sentry/socket/netfilter/BUILD | 1 + pkg/sentry/socket/netfilter/netfilter.go | 7 +- pkg/sentry/socket/netfilter/owner_matcher.go | 128 ++++++++++++++++++++++++ pkg/sentry/socket/netstack/provider.go | 6 ++ pkg/tcpip/network/ipv4/ipv4.go | 15 +++ pkg/tcpip/stack/packet_buffer.go | 9 +- pkg/tcpip/stack/transport_test.go | 2 + pkg/tcpip/tcpip.go | 12 +++ pkg/tcpip/transport/icmp/endpoint.go | 12 ++- pkg/tcpip/transport/packet/endpoint.go | 2 + pkg/tcpip/transport/raw/endpoint.go | 9 ++ pkg/tcpip/transport/tcp/accept.go | 5 +- pkg/tcpip/transport/tcp/connect.go | 10 +- pkg/tcpip/transport/tcp/endpoint.go | 7 ++ pkg/tcpip/transport/tcp/forwarder.go | 2 +- pkg/tcpip/transport/tcp/protocol.go | 2 +- pkg/tcpip/transport/udp/endpoint.go | 12 ++- test/iptables/filter_output.go | 143 +++++++++++++++++++++++++++ test/iptables/iptables_test.go | 30 ++++++ 22 files changed, 451 insertions(+), 17 deletions(-) create mode 100644 pkg/sentry/socket/netfilter/owner_matcher.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/netfilter.go b/pkg/abi/linux/netfilter.go index 80dc09aa9..a8d4f9d69 100644 --- a/pkg/abi/linux/netfilter.go +++ b/pkg/abi/linux/netfilter.go @@ -509,3 +509,44 @@ const ( // Enable all flags. XT_UDP_INV_MASK = 0x03 ) + +// IPTOwnerInfo holds data for matching packets with owner. It corresponds +// to struct ipt_owner_info in libxt_owner.c of iptables binary. +type IPTOwnerInfo struct { + // UID is user id which created the packet. + UID uint32 + + // GID is group id which created the packet. + GID uint32 + + // PID is process id of the process which created the packet. + PID uint32 + + // SID is session id which created the packet. + SID uint32 + + // Comm is the command name which created the packet. + Comm [16]byte + + // Match is used to match UID/GID of the socket. See the + // XT_OWNER_* flags below. + Match uint8 + + // Invert flips the meaning of Match field. + Invert uint8 +} + +// SizeOfIPTOwnerInfo is the size of an XTOwnerMatchInfo. +const SizeOfIPTOwnerInfo = 34 + +// Flags in IPTOwnerInfo.Match. Corresponding constants are in +// include/uapi/linux/netfilter/xt_owner.h. +const ( + // Match the UID of the packet. + XT_OWNER_UID = 1 << 0 + // Match the GID of the packet. + XT_OWNER_GID = 1 << 1 + // Match if the socket exists for the packet. Forwarded + // packets do not have an associated socket. + XT_OWNER_SOCKET = 1 << 2 +) diff --git a/pkg/abi/linux/netfilter_test.go b/pkg/abi/linux/netfilter_test.go index 21e237f92..565dd550e 100644 --- a/pkg/abi/linux/netfilter_test.go +++ b/pkg/abi/linux/netfilter_test.go @@ -29,6 +29,7 @@ func TestSizes(t *testing.T) { {IPTGetEntries{}, SizeOfIPTGetEntries}, {IPTGetinfo{}, SizeOfIPTGetinfo}, {IPTIP{}, SizeOfIPTIP}, + {IPTOwnerInfo{}, SizeOfIPTOwnerInfo}, {IPTReplace{}, SizeOfIPTReplace}, {XTCounters{}, SizeOfXTCounters}, {XTEntryMatch{}, SizeOfXTEntryMatch}, diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index 8452ddf5b..d6546735e 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -863,3 +863,15 @@ func (t *Task) SetOOMScoreAdj(adj int32) error { atomic.StoreInt32(&t.tg.oomScoreAdj, adj) return nil } + +// UID returns t's uid. +// TODO(gvisor.dev/issue/170): This method is not namespaced yet. +func (t *Task) UID() uint32 { + return uint32(t.Credentials().EffectiveKUID) +} + +// GID returns t's gid. +// TODO(gvisor.dev/issue/170): This method is not namespaced yet. +func (t *Task) GID() uint32 { + return uint32(t.Credentials().EffectiveKGID) +} diff --git a/pkg/sentry/socket/netfilter/BUILD b/pkg/sentry/socket/netfilter/BUILD index e801abeb8..721094bbf 100644 --- a/pkg/sentry/socket/netfilter/BUILD +++ b/pkg/sentry/socket/netfilter/BUILD @@ -7,6 +7,7 @@ go_library( srcs = [ "extensions.go", "netfilter.go", + "owner_matcher.go", "targets.go", "tcp_matcher.go", "udp_matcher.go", diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 55bcc3ace..878f81fd5 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -517,11 +517,10 @@ func SetEntries(stk *stack.Stack, optVal []byte) *syserr.Error { } // TODO(gvisor.dev/issue/170): Support other chains. - // Since we only support modifying the INPUT chain and redirect for - // PREROUTING chain right now, make sure all other chains point to - // ACCEPT rules. + // Since we only support modifying the INPUT, PREROUTING and OUTPUT chain right now, + // make sure all other chains point to ACCEPT rules. for hook, ruleIdx := range table.BuiltinChains { - if hook != stack.Input && hook != stack.Prerouting { + if hook == stack.Forward || hook == stack.Postrouting { if _, ok := table.Rules[ruleIdx].Target.(stack.AcceptTarget); !ok { nflog("hook %d is unsupported.", hook) return syserr.ErrInvalidArgument diff --git a/pkg/sentry/socket/netfilter/owner_matcher.go b/pkg/sentry/socket/netfilter/owner_matcher.go new file mode 100644 index 000000000..5949a7c29 --- /dev/null +++ b/pkg/sentry/socket/netfilter/owner_matcher.go @@ -0,0 +1,128 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package netfilter + +import ( + "fmt" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/tcpip/stack" + "gvisor.dev/gvisor/pkg/usermem" +) + +const matcherNameOwner = "owner" + +func init() { + registerMatchMaker(ownerMarshaler{}) +} + +// ownerMarshaler implements matchMaker for owner matching. +type ownerMarshaler struct{} + +// name implements matchMaker.name. +func (ownerMarshaler) name() string { + return matcherNameOwner +} + +// marshal implements matchMaker.marshal. +func (ownerMarshaler) marshal(mr stack.Matcher) []byte { + matcher := mr.(*OwnerMatcher) + iptOwnerInfo := linux.IPTOwnerInfo{ + UID: matcher.uid, + GID: matcher.gid, + } + + // Support for UID match. + // TODO(gvisor.dev/issue/170): Need to support gid match. + if matcher.matchUID { + iptOwnerInfo.Match = linux.XT_OWNER_UID + } else if matcher.matchGID { + panic("GID match is not supported.") + } else { + panic("UID match is not set.") + } + + buf := make([]byte, 0, linux.SizeOfIPTOwnerInfo) + return marshalEntryMatch(matcherNameOwner, binary.Marshal(buf, usermem.ByteOrder, iptOwnerInfo)) +} + +// unmarshal implements matchMaker.unmarshal. +func (ownerMarshaler) unmarshal(buf []byte, filter stack.IPHeaderFilter) (stack.Matcher, error) { + if len(buf) < linux.SizeOfIPTOwnerInfo { + return nil, fmt.Errorf("buf has insufficient size for owner match: %d", len(buf)) + } + + // For alignment reasons, the match's total size may + // exceed what's strictly necessary to hold matchData. + var matchData linux.IPTOwnerInfo + binary.Unmarshal(buf[:linux.SizeOfIPTOwnerInfo], usermem.ByteOrder, &matchData) + nflog("parseMatchers: parsed IPTOwnerInfo: %+v", matchData) + + if matchData.Invert != 0 { + return nil, fmt.Errorf("invert flag is not supported for owner match") + } + + // Support for UID match. + // TODO(gvisor.dev/issue/170): Need to support gid match. + if matchData.Match&linux.XT_OWNER_UID != linux.XT_OWNER_UID { + return nil, fmt.Errorf("owner match is only supported for uid") + } + + // Check Flags. + var owner OwnerMatcher + owner.uid = matchData.UID + owner.gid = matchData.GID + owner.matchUID = true + + return &owner, nil +} + +type OwnerMatcher struct { + uid uint32 + gid uint32 + matchUID bool + matchGID bool + invert uint8 +} + +// Name implements Matcher.Name. +func (*OwnerMatcher) Name() string { + return matcherNameOwner +} + +// Match implements Matcher.Match. +func (om *OwnerMatcher) Match(hook stack.Hook, pkt stack.PacketBuffer, interfaceName string) (bool, bool) { + // Support only for OUTPUT chain. + // TODO(gvisor.dev/issue/170): Need to support for POSTROUTING chain also. + if hook != stack.Output { + return false, true + } + + // If the packet owner is not set, drop the packet. + // Support for uid match. + // TODO(gvisor.dev/issue/170): Need to support gid match. + if pkt.Owner == nil || !om.matchUID { + return false, true + } + + // TODO(gvisor.dev/issue/170): Need to add tests to verify + // drop rule when packet UID does not match owner matcher UID. + if pkt.Owner.UID() != om.uid { + return false, false + } + + return true, false +} diff --git a/pkg/sentry/socket/netstack/provider.go b/pkg/sentry/socket/netstack/provider.go index 5f181f017..eb090e79b 100644 --- a/pkg/sentry/socket/netstack/provider.go +++ b/pkg/sentry/socket/netstack/provider.go @@ -126,6 +126,12 @@ func (p *provider) Socket(t *kernel.Task, stype linux.SockType, protocol int) (* ep, e = eps.Stack.NewRawEndpoint(transProto, p.netProto, wq, associated) } else { ep, e = eps.Stack.NewEndpoint(transProto, p.netProto, wq) + + // Assign task to PacketOwner interface to get the UID and GID for + // iptables owner matching. + if e == nil { + ep.SetOwner(t) + } } if e != nil { return nil, syserr.TranslateNetstackError(e) diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index b3ee6000e..a7d9a8b25 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -244,6 +244,14 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw ip := e.addIPHeader(r, &pkt.Header, pkt.Data.Size(), params) pkt.NetworkHeader = buffer.View(ip) + // iptables filtering. All packets that reach here are locally + // generated. + ipt := e.stack.IPTables() + if ok := ipt.Check(stack.Output, pkt); !ok { + // iptables is telling us to drop the packet. + return nil + } + if r.Loop&stack.PacketLoop != 0 { // The inbound path expects the network header to still be in // the PacketBuffer's Data field. @@ -280,7 +288,14 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []stack.Pac return len(pkts), nil } + // iptables filtering. All packets that reach here are locally + // generated. + ipt := e.stack.IPTables() for i := range pkts { + if ok := ipt.Check(stack.Output, pkts[i]); !ok { + // iptables is telling us to drop the packet. + continue + } ip := e.addIPHeader(r, &pkts[i].Header, pkts[i].DataSize, params) pkts[i].NetworkHeader = buffer.View(ip) } diff --git a/pkg/tcpip/stack/packet_buffer.go b/pkg/tcpip/stack/packet_buffer.go index 9505a4e92..9367de180 100644 --- a/pkg/tcpip/stack/packet_buffer.go +++ b/pkg/tcpip/stack/packet_buffer.go @@ -13,7 +13,10 @@ package stack -import "gvisor.dev/gvisor/pkg/tcpip/buffer" +import ( + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/buffer" +) // A PacketBuffer contains all the data of a network packet. // @@ -59,6 +62,10 @@ type PacketBuffer struct { // Hash is the transport layer hash of this packet. A value of zero // indicates no valid hash has been set. Hash uint32 + + // Owner is implemented by task to get the uid and gid. + // Only set for locally generated packets. + Owner tcpip.PacketOwner } // Clone makes a copy of pk. It clones the Data field, which creates a new diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go index 8ca9ac3cf..3084e6593 100644 --- a/pkg/tcpip/stack/transport_test.go +++ b/pkg/tcpip/stack/transport_test.go @@ -56,6 +56,8 @@ func (f *fakeTransportEndpoint) Stats() tcpip.EndpointStats { return nil } +func (f *fakeTransportEndpoint) SetOwner(owner tcpip.PacketOwner) {} + func newFakeTransportEndpoint(s *stack.Stack, proto *fakeTransportProtocol, netProto tcpip.NetworkProtocolNumber, uniqueID uint64) tcpip.Endpoint { return &fakeTransportEndpoint{stack: s, TransportEndpointInfo: stack.TransportEndpointInfo{NetProto: netProto}, proto: proto, uniqueID: uniqueID} } diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 3dc5d87d6..2ef3271f1 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -336,6 +336,15 @@ type ControlMessages struct { PacketInfo IPPacketInfo } +// PacketOwner is used to get UID and GID of the packet. +type PacketOwner interface { + // UID returns UID of the packet. + UID() uint32 + + // GID returns GID of the packet. + GID() uint32 +} + // Endpoint is the interface implemented by transport protocols (e.g., tcp, udp) // that exposes functionality like read, write, connect, etc. to users of the // networking stack. @@ -470,6 +479,9 @@ type Endpoint interface { // Stats returns a reference to the endpoint stats. Stats() EndpointStats + + // SetOwner sets the task owner to the endpoint owner. + SetOwner(owner PacketOwner) } // EndpointInfo is the interface implemented by each endpoint info struct. diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index 613b12ead..b007302fb 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -73,6 +73,9 @@ type endpoint struct { route stack.Route `state:"manual"` ttl uint8 stats tcpip.TransportEndpointStats `state:"nosave"` + + // owner is used to get uid and gid of the packet. + owner tcpip.PacketOwner } func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { @@ -133,6 +136,10 @@ func (e *endpoint) Close() { // ModerateRecvBuf implements tcpip.Endpoint.ModerateRecvBuf. func (e *endpoint) ModerateRecvBuf(copied int) {} +func (e *endpoint) SetOwner(owner tcpip.PacketOwner) { + e.owner = owner +} + // IPTables implements tcpip.Endpoint.IPTables. func (e *endpoint) IPTables() (stack.IPTables, error) { return e.stack.IPTables(), nil @@ -321,7 +328,7 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c switch e.NetProto { case header.IPv4ProtocolNumber: - err = send4(route, e.ID.LocalPort, v, e.ttl) + err = send4(route, e.ID.LocalPort, v, e.ttl, e.owner) case header.IPv6ProtocolNumber: err = send6(route, e.ID.LocalPort, v, e.ttl) @@ -415,7 +422,7 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { } } -func send4(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Error { +func send4(r *stack.Route, ident uint16, data buffer.View, ttl uint8, owner tcpip.PacketOwner) *tcpip.Error { if len(data) < header.ICMPv4MinimumSize { return tcpip.ErrInvalidEndpointState } @@ -444,6 +451,7 @@ func send4(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Err Header: hdr, Data: data.ToVectorisedView(), TransportHeader: buffer.View(icmpv4), + Owner: owner, }) } diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go index df49d0995..23158173d 100644 --- a/pkg/tcpip/transport/packet/endpoint.go +++ b/pkg/tcpip/transport/packet/endpoint.go @@ -392,3 +392,5 @@ func (ep *endpoint) Info() tcpip.EndpointInfo { func (ep *endpoint) Stats() tcpip.EndpointStats { return &ep.stats } + +func (ep *endpoint) SetOwner(owner tcpip.PacketOwner) {} diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index 536dafd1e..337bc1c71 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -80,6 +80,9 @@ type endpoint struct { // Connect(), and is valid only when conneted is true. route stack.Route `state:"manual"` stats tcpip.TransportEndpointStats `state:"nosave"` + + // owner is used to get uid and gid of the packet. + owner tcpip.PacketOwner } // NewEndpoint returns a raw endpoint for the given protocols. @@ -159,6 +162,10 @@ func (e *endpoint) Close() { // ModerateRecvBuf implements tcpip.Endpoint.ModerateRecvBuf. func (e *endpoint) ModerateRecvBuf(copied int) {} +func (e *endpoint) SetOwner(owner tcpip.PacketOwner) { + e.owner = owner +} + // IPTables implements tcpip.Endpoint.IPTables. func (e *endpoint) IPTables() (stack.IPTables, error) { return e.stack.IPTables(), nil @@ -348,10 +355,12 @@ func (e *endpoint) finishWrite(payloadBytes []byte, route *stack.Route) (int64, } break } + hdr := buffer.NewPrependable(len(payloadBytes) + int(route.MaxHeaderLength())) if err := route.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: e.TransProto, TTL: route.DefaultTTL(), TOS: stack.DefaultTOS}, stack.PacketBuffer{ Header: hdr, Data: buffer.View(payloadBytes).ToVectorisedView(), + Owner: e.owner, }); err != nil { return 0, nil, err } diff --git a/pkg/tcpip/transport/tcp/accept.go b/pkg/tcpip/transport/tcp/accept.go index 375ca21f6..7a9dea4ac 100644 --- a/pkg/tcpip/transport/tcp/accept.go +++ b/pkg/tcpip/transport/tcp/accept.go @@ -276,7 +276,7 @@ func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, i // and then performs the TCP 3-way handshake. // // The new endpoint is returned with e.mu held. -func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *header.TCPSynOptions, queue *waiter.Queue) (*endpoint, *tcpip.Error) { +func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *header.TCPSynOptions, queue *waiter.Queue, owner tcpip.PacketOwner) (*endpoint, *tcpip.Error) { // Create new endpoint. irs := s.sequenceNumber isn := generateSecureISN(s.id, l.stack.Seed()) @@ -284,6 +284,7 @@ func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *head if err != nil { return nil, err } + ep.owner = owner // listenEP is nil when listenContext is used by tcp.Forwarder. deferAccept := time.Duration(0) @@ -414,7 +415,7 @@ func (e *endpoint) handleSynSegment(ctx *listenContext, s *segment, opts *header }() defer s.decRef() - n, err := ctx.createEndpointAndPerformHandshake(s, opts, &waiter.Queue{}) + n, err := ctx.createEndpointAndPerformHandshake(s, opts, &waiter.Queue{}, e.owner) if err != nil { e.stack.Stats().TCP.FailedConnectionAttempts.Increment() e.stats.FailedConnectionAttempts.Increment() diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 1d245c2c6..3239a5911 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -745,7 +745,7 @@ func (e *endpoint) sendSynTCP(r *stack.Route, tf tcpFields, opts header.TCPSynOp func (e *endpoint) sendTCP(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso *stack.GSO) *tcpip.Error { tf.txHash = e.txHash - if err := sendTCP(r, tf, data, gso); err != nil { + if err := sendTCP(r, tf, data, gso, e.owner); err != nil { e.stats.SendErrors.SegmentSendToNetworkFailed.Increment() return err } @@ -787,7 +787,7 @@ func buildTCPHdr(r *stack.Route, tf tcpFields, pkt *stack.PacketBuffer, gso *sta } } -func sendTCPBatch(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso *stack.GSO) *tcpip.Error { +func sendTCPBatch(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso *stack.GSO, owner tcpip.PacketOwner) *tcpip.Error { optLen := len(tf.opts) if tf.rcvWnd > 0xffff { tf.rcvWnd = 0xffff @@ -816,6 +816,7 @@ func sendTCPBatch(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso pkts[i].DataSize = packetSize pkts[i].Data = data pkts[i].Hash = tf.txHash + pkts[i].Owner = owner buildTCPHdr(r, tf, &pkts[i], gso) off += packetSize tf.seq = tf.seq.Add(seqnum.Size(packetSize)) @@ -833,14 +834,14 @@ func sendTCPBatch(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso // sendTCP sends a TCP segment with the provided options via the provided // network endpoint and under the provided identity. -func sendTCP(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso *stack.GSO) *tcpip.Error { +func sendTCP(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso *stack.GSO, owner tcpip.PacketOwner) *tcpip.Error { optLen := len(tf.opts) if tf.rcvWnd > 0xffff { tf.rcvWnd = 0xffff } if r.Loop&stack.PacketLoop == 0 && gso != nil && gso.Type == stack.GSOSW && int(gso.MSS) < data.Size() { - return sendTCPBatch(r, tf, data, gso) + return sendTCPBatch(r, tf, data, gso, owner) } pkt := stack.PacketBuffer{ @@ -849,6 +850,7 @@ func sendTCP(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso *stac DataSize: data.Size(), Data: data, Hash: tf.txHash, + Owner: owner, } buildTCPHdr(r, tf, &pkt, gso) diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 1ebee0cfe..9b123e968 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -603,6 +603,9 @@ type endpoint struct { // txHash is the transport layer hash to be set on outbound packets // emitted by this endpoint. txHash uint32 + + // owner is used to get uid and gid of the packet. + owner tcpip.PacketOwner } // UniqueID implements stack.TransportEndpoint.UniqueID. @@ -1132,6 +1135,10 @@ func (e *endpoint) ModerateRecvBuf(copied int) { e.rcvListMu.Unlock() } +func (e *endpoint) SetOwner(owner tcpip.PacketOwner) { + e.owner = owner +} + // IPTables implements tcpip.Endpoint.IPTables. func (e *endpoint) IPTables() (stack.IPTables, error) { return e.stack.IPTables(), nil diff --git a/pkg/tcpip/transport/tcp/forwarder.go b/pkg/tcpip/transport/tcp/forwarder.go index a094471b8..808410c92 100644 --- a/pkg/tcpip/transport/tcp/forwarder.go +++ b/pkg/tcpip/transport/tcp/forwarder.go @@ -157,7 +157,7 @@ func (r *ForwarderRequest) CreateEndpoint(queue *waiter.Queue) (tcpip.Endpoint, TSVal: r.synOptions.TSVal, TSEcr: r.synOptions.TSEcr, SACKPermitted: r.synOptions.SACKPermitted, - }, queue) + }, queue, nil) if err != nil { return nil, err } diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go index 1377107ca..dce9a1652 100644 --- a/pkg/tcpip/transport/tcp/protocol.go +++ b/pkg/tcpip/transport/tcp/protocol.go @@ -199,7 +199,7 @@ func replyWithReset(s *segment) { seq: seq, ack: ack, rcvWnd: 0, - }, buffer.VectorisedView{}, nil /* gso */) + }, buffer.VectorisedView{}, nil /* gso */, nil /* PacketOwner */) } // SetOption implements stack.TransportProtocol.SetOption. diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index a3372ac58..120d3baa3 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -143,6 +143,9 @@ type endpoint struct { // TODO(b/142022063): Add ability to save and restore per endpoint stats. stats tcpip.TransportEndpointStats `state:"nosave"` + + // owner is used to get uid and gid of the packet. + owner tcpip.PacketOwner } // +stateify savable @@ -484,7 +487,7 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c useDefaultTTL = false } - if err := sendUDP(route, buffer.View(v).ToVectorisedView(), e.ID.LocalPort, dstPort, ttl, useDefaultTTL, e.sendTOS); err != nil { + if err := sendUDP(route, buffer.View(v).ToVectorisedView(), e.ID.LocalPort, dstPort, ttl, useDefaultTTL, e.sendTOS, e.owner); err != nil { return 0, nil, err } return int64(len(v)), nil, nil @@ -886,7 +889,7 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { // sendUDP sends a UDP segment via the provided network endpoint and under the // provided identity. -func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort uint16, ttl uint8, useDefaultTTL bool, tos uint8) *tcpip.Error { +func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort uint16, ttl uint8, useDefaultTTL bool, tos uint8, owner tcpip.PacketOwner) *tcpip.Error { // Allocate a buffer for the UDP header. hdr := buffer.NewPrependable(header.UDPMinimumSize + int(r.MaxHeaderLength())) @@ -916,6 +919,7 @@ func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort u Header: hdr, Data: data, TransportHeader: buffer.View(udp), + Owner: owner, }); err != nil { r.Stats().UDP.PacketSendErrors.Increment() return err @@ -1356,3 +1360,7 @@ func (*endpoint) Wait() {} func isBroadcastOrMulticast(a tcpip.Address) bool { return a == header.IPv4Broadcast || header.IsV4MulticastAddress(a) || header.IsV6MulticastAddress(a) } + +func (e *endpoint) SetOwner(owner tcpip.PacketOwner) { + e.owner = owner +} diff --git a/test/iptables/filter_output.go b/test/iptables/filter_output.go index 4582d514c..f6d974b85 100644 --- a/test/iptables/filter_output.go +++ b/test/iptables/filter_output.go @@ -24,6 +24,11 @@ func init() { RegisterTestCase(FilterOutputDropTCPSrcPort{}) RegisterTestCase(FilterOutputDestination{}) RegisterTestCase(FilterOutputInvertDestination{}) + RegisterTestCase(FilterOutputAcceptTCPOwner{}) + RegisterTestCase(FilterOutputDropTCPOwner{}) + RegisterTestCase(FilterOutputAcceptUDPOwner{}) + RegisterTestCase(FilterOutputDropUDPOwner{}) + RegisterTestCase(FilterOutputOwnerFail{}) } // FilterOutputDropTCPDestPort tests that connections are not accepted on @@ -90,6 +95,144 @@ func (FilterOutputDropTCPSrcPort) LocalAction(ip net.IP) error { return nil } +// FilterOutputAcceptTCPOwner tests that TCP connections from uid owner are accepted. +type FilterOutputAcceptTCPOwner struct{} + +// Name implements TestCase.Name. +func (FilterOutputAcceptTCPOwner) Name() string { + return "FilterOutputAcceptTCPOwner" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterOutputAcceptTCPOwner) ContainerAction(ip net.IP) error { + if err := filterTable("-A", "OUTPUT", "-p", "tcp", "-m", "owner", "--uid-owner", "root", "-j", "ACCEPT"); err != nil { + return err + } + + // Listen for TCP packets on accept port. + if err := listenTCP(acceptPort, sendloopDuration); err != nil { + return fmt.Errorf("connection on port %d should be accepted, but got dropped", acceptPort) + } + + return nil +} + +// LocalAction implements TestCase.LocalAction. +func (FilterOutputAcceptTCPOwner) LocalAction(ip net.IP) error { + if err := connectTCP(ip, acceptPort, sendloopDuration); err != nil { + return fmt.Errorf("connection destined to port %d should be accepted, but got dropped", acceptPort) + } + + return nil +} + +// FilterOutputDropTCPOwner tests that TCP connections from uid owner are dropped. +type FilterOutputDropTCPOwner struct{} + +// Name implements TestCase.Name. +func (FilterOutputDropTCPOwner) Name() string { + return "FilterOutputDropTCPOwner" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterOutputDropTCPOwner) ContainerAction(ip net.IP) error { + if err := filterTable("-A", "OUTPUT", "-p", "tcp", "-m", "owner", "--uid-owner", "root", "-j", "DROP"); err != nil { + return err + } + + // Listen for TCP packets on accept port. + if err := listenTCP(acceptPort, sendloopDuration); err == nil { + return fmt.Errorf("connection on port %d should be dropped, but got accepted", acceptPort) + } + + return nil +} + +// LocalAction implements TestCase.LocalAction. +func (FilterOutputDropTCPOwner) LocalAction(ip net.IP) error { + if err := connectTCP(ip, acceptPort, sendloopDuration); err == nil { + return fmt.Errorf("connection destined to port %d should be dropped, but got accepted", acceptPort) + } + + return nil +} + +// FilterOutputAcceptUDPOwner tests that UDP packets from uid owner are accepted. +type FilterOutputAcceptUDPOwner struct{} + +// Name implements TestCase.Name. +func (FilterOutputAcceptUDPOwner) Name() string { + return "FilterOutputAcceptUDPOwner" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterOutputAcceptUDPOwner) ContainerAction(ip net.IP) error { + if err := filterTable("-A", "OUTPUT", "-p", "udp", "-m", "owner", "--uid-owner", "root", "-j", "ACCEPT"); err != nil { + return err + } + + // Send UDP packets on acceptPort. + return sendUDPLoop(ip, acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (FilterOutputAcceptUDPOwner) LocalAction(ip net.IP) error { + // Listen for UDP packets on acceptPort. + return listenUDP(acceptPort, sendloopDuration) +} + +// FilterOutputDropUDPOwner tests that UDP packets from uid owner are dropped. +type FilterOutputDropUDPOwner struct{} + +// Name implements TestCase.Name. +func (FilterOutputDropUDPOwner) Name() string { + return "FilterOutputDropUDPOwner" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterOutputDropUDPOwner) ContainerAction(ip net.IP) error { + if err := filterTable("-A", "OUTPUT", "-p", "udp", "-m", "owner", "--uid-owner", "root", "-j", "DROP"); err != nil { + return err + } + + // Send UDP packets on dropPort. + return sendUDPLoop(ip, dropPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (FilterOutputDropUDPOwner) LocalAction(ip net.IP) error { + // Listen for UDP packets on dropPort. + if err := listenUDP(dropPort, sendloopDuration); err == nil { + return fmt.Errorf("packets should not be received") + } + + return nil +} + +// FilterOutputOwnerFail tests that without uid/gid option, owner rule +// will fail. +type FilterOutputOwnerFail struct{} + +// Name implements TestCase.Name. +func (FilterOutputOwnerFail) Name() string { + return "FilterOutputOwnerFail" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterOutputOwnerFail) ContainerAction(ip net.IP) error { + if err := filterTable("-A", "OUTPUT", "-p", "udp", "-m", "owner", "-j", "ACCEPT"); err == nil { + return fmt.Errorf("Invalid argument") + } + + return nil +} + +// LocalAction implements TestCase.LocalAction. +func (FilterOutputOwnerFail) LocalAction(ip net.IP) error { + // no-op. + return nil +} + // FilterOutputDestination tests that we can selectively allow packets to // certain destinations. type FilterOutputDestination struct{} diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go index 7f1f70606..493d69052 100644 --- a/test/iptables/iptables_test.go +++ b/test/iptables/iptables_test.go @@ -274,6 +274,36 @@ func TestFilterOutputDropTCPSrcPort(t *testing.T) { } } +func TestFilterOutputAcceptTCPOwner(t *testing.T) { + if err := singleTest(FilterOutputAcceptTCPOwner{}); err != nil { + t.Fatal(err) + } +} + +func TestFilterOutputDropTCPOwner(t *testing.T) { + if err := singleTest(FilterOutputDropTCPOwner{}); err != nil { + t.Fatal(err) + } +} + +func TestFilterOutputAcceptUDPOwner(t *testing.T) { + if err := singleTest(FilterOutputAcceptUDPOwner{}); err != nil { + t.Fatal(err) + } +} + +func TestFilterOutputDropUDPOwner(t *testing.T) { + if err := singleTest(FilterOutputDropUDPOwner{}); err != nil { + t.Fatal(err) + } +} + +func TestFilterOutputOwnerFail(t *testing.T) { + if err := singleTest(FilterOutputOwnerFail{}); err != nil { + t.Fatal(err) + } +} + func TestJumpSerialize(t *testing.T) { if err := singleTest(FilterInputSerializeJump{}); err != nil { t.Fatal(err) -- cgit v1.2.3 From e466ab04a20731ebeb8a9725d808def975d4c88d Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Thu, 26 Mar 2020 13:45:02 -0700 Subject: Add unique ID to Mount type. Analagous to Linux's mount.mnt_id. This ID is displayed in /proc/[pid]/mountinfo. PiperOrigin-RevId: 303185564 --- pkg/sentry/vfs/mount.go | 7 +++++++ 1 file changed, 7 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index 05f6233f9..4b68cabda 100644 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go @@ -24,6 +24,9 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) +// lastMountID is used to allocate mount ids. Must be accessed atomically. +var lastMountID uint64 + // A Mount is a replacement of a Dentry (Mount.key.point) from one Filesystem // (Mount.key.parent.fs) with a Dentry (Mount.root) from another Filesystem // (Mount.fs), which applies to path resolution in the context of a particular @@ -48,6 +51,9 @@ type Mount struct { fs *Filesystem root *Dentry + // ID is the immutable mount ID. + ID uint64 + // key is protected by VirtualFilesystem.mountMu and // VirtualFilesystem.mounts.seq, and may be nil. References are held on // key.parent and key.point if they are not nil. @@ -87,6 +93,7 @@ type Mount struct { func newMount(vfs *VirtualFilesystem, fs *Filesystem, root *Dentry, mntns *MountNamespace, opts *MountOptions) *Mount { mnt := &Mount{ + ID: atomic.AddUint64(&lastMountID, 1), vfs: vfs, fs: fs, root: root, -- cgit v1.2.3 From 137f3614009b0ef931c1d00a083b4ae8e6a39bc9 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Thu, 26 Mar 2020 16:46:15 -0700 Subject: Use host-defined file owner and mode, when possible, for imported fds. Using the host-defined file owner matches VFS1. It is more correct to use the host-defined mode, since the cached value may become out of date. However, kernfs.Inode.Mode() does not return an error--other filesystems on kernfs are in-memory so retrieving mode should not fail. Therefore, if the host syscall fails, we rely on a cached value instead. Updates #1672. PiperOrigin-RevId: 303220864 --- pkg/sentry/control/proc.go | 6 +-- pkg/sentry/fs/host/BUILD | 1 - pkg/sentry/fs/host/control.go | 2 +- pkg/sentry/fs/host/file.go | 10 ++-- pkg/sentry/fs/host/inode_test.go | 3 +- pkg/sentry/fs/host/wait_test.go | 3 +- pkg/sentry/fsimpl/host/host.go | 110 ++++++++++++++++++++++++++------------- runsc/boot/fds.go | 5 +- 8 files changed, 87 insertions(+), 53 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/control/proc.go b/pkg/sentry/control/proc.go index 5457ba5e7..b51fb3959 100644 --- a/pkg/sentry/control/proc.go +++ b/pkg/sentry/control/proc.go @@ -224,8 +224,6 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI } } - mounter := fs.FileOwnerFromContext(ctx) - // TODO(gvisor.dev/issue/1623): Use host FD when supported in VFS2. var ttyFile *fs.File for appFD, hostFile := range args.FilePayload.Files { @@ -235,7 +233,7 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI // Import the file as a host TTY file. if ttyFile == nil { var err error - appFile, err = host.ImportFile(ctx, int(hostFile.Fd()), mounter, true /* isTTY */) + appFile, err = host.ImportFile(ctx, int(hostFile.Fd()), true /* isTTY */) if err != nil { return nil, 0, nil, err } @@ -254,7 +252,7 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI } else { // Import the file as a regular host file. var err error - appFile, err = host.ImportFile(ctx, int(hostFile.Fd()), mounter, false /* isTTY */) + appFile, err = host.ImportFile(ctx, int(hostFile.Fd()), false /* isTTY */) if err != nil { return nil, 0, nil, err } diff --git a/pkg/sentry/fs/host/BUILD b/pkg/sentry/fs/host/BUILD index 011625c80..aabce6cc9 100644 --- a/pkg/sentry/fs/host/BUILD +++ b/pkg/sentry/fs/host/BUILD @@ -71,7 +71,6 @@ go_test( "//pkg/fd", "//pkg/fdnotifier", "//pkg/sentry/contexttest", - "//pkg/sentry/fs", "//pkg/sentry/kernel/time", "//pkg/sentry/socket", "//pkg/sentry/socket/unix/transport", diff --git a/pkg/sentry/fs/host/control.go b/pkg/sentry/fs/host/control.go index cd84e1337..52c0504b6 100644 --- a/pkg/sentry/fs/host/control.go +++ b/pkg/sentry/fs/host/control.go @@ -78,7 +78,7 @@ func fdsToFiles(ctx context.Context, fds []int) []*fs.File { } // Create the file backed by hostFD. - file, err := NewFile(ctx, fd, fs.FileOwnerFromContext(ctx)) + file, err := NewFile(ctx, fd) if err != nil { ctx.Warningf("Error creating file from host FD: %v", err) break diff --git a/pkg/sentry/fs/host/file.go b/pkg/sentry/fs/host/file.go index 034862694..3e48b8b2c 100644 --- a/pkg/sentry/fs/host/file.go +++ b/pkg/sentry/fs/host/file.go @@ -60,8 +60,8 @@ var _ fs.FileOperations = (*fileOperations)(nil) // The returned File cannot be saved, since there is no guarantee that the same // FD will exist or represent the same file at time of restore. If such a // guarantee does exist, use ImportFile instead. -func NewFile(ctx context.Context, fd int, mounter fs.FileOwner) (*fs.File, error) { - return newFileFromDonatedFD(ctx, fd, mounter, false, false) +func NewFile(ctx context.Context, fd int) (*fs.File, error) { + return newFileFromDonatedFD(ctx, fd, false, false) } // ImportFile creates a new File backed by the provided host file descriptor. @@ -71,13 +71,13 @@ func NewFile(ctx context.Context, fd int, mounter fs.FileOwner) (*fs.File, error // If the returned file is saved, it will be restored by re-importing the FD // originally passed to ImportFile. It is the restorer's responsibility to // ensure that the FD represents the same file. -func ImportFile(ctx context.Context, fd int, mounter fs.FileOwner, isTTY bool) (*fs.File, error) { - return newFileFromDonatedFD(ctx, fd, mounter, true, isTTY) +func ImportFile(ctx context.Context, fd int, isTTY bool) (*fs.File, error) { + return newFileFromDonatedFD(ctx, fd, true, isTTY) } // newFileFromDonatedFD returns an fs.File from a donated FD. If the FD is // saveable, then saveable is true. -func newFileFromDonatedFD(ctx context.Context, donated int, mounter fs.FileOwner, saveable, isTTY bool) (*fs.File, error) { +func newFileFromDonatedFD(ctx context.Context, donated int, saveable, isTTY bool) (*fs.File, error) { var s syscall.Stat_t if err := syscall.Fstat(donated, &s); err != nil { return nil, err diff --git a/pkg/sentry/fs/host/inode_test.go b/pkg/sentry/fs/host/inode_test.go index 4c374681c..c507f57eb 100644 --- a/pkg/sentry/fs/host/inode_test.go +++ b/pkg/sentry/fs/host/inode_test.go @@ -19,7 +19,6 @@ import ( "testing" "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" ) // TestCloseFD verifies fds will be closed. @@ -33,7 +32,7 @@ func TestCloseFD(t *testing.T) { // Use the write-end because we will detect if it's closed on the read end. ctx := contexttest.Context(t) - file, err := NewFile(ctx, p[1], fs.RootOwner) + file, err := NewFile(ctx, p[1]) if err != nil { t.Fatalf("Failed to create File: %v", err) } diff --git a/pkg/sentry/fs/host/wait_test.go b/pkg/sentry/fs/host/wait_test.go index d49c3a635..ce397a5e3 100644 --- a/pkg/sentry/fs/host/wait_test.go +++ b/pkg/sentry/fs/host/wait_test.go @@ -20,7 +20,6 @@ import ( "time" "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/waiter" ) @@ -34,7 +33,7 @@ func TestWait(t *testing.T) { defer syscall.Close(fds[1]) ctx := contexttest.Context(t) - file, err := NewFile(ctx, fds[0], fs.RootOwner) + file, err := NewFile(ctx, fds[0]) if err != nil { syscall.Close(fds[0]) t.Fatalf("NewFile failed: %v", err) diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index a54985ef5..17e3d6e9d 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -54,7 +54,7 @@ func NewMount(vfsObj *vfs.VirtualFilesystem) (*vfs.Mount, error) { } // ImportFD sets up and returns a vfs.FileDescription from a donated fd. -func ImportFD(mnt *vfs.Mount, hostFD int, ownerUID auth.KUID, ownerGID auth.KGID, isTTY bool) (*vfs.FileDescription, error) { +func ImportFD(mnt *vfs.Mount, hostFD int, isTTY bool) (*vfs.FileDescription, error) { fs, ok := mnt.Filesystem().Impl().(*kernfs.Filesystem) if !ok { return nil, fmt.Errorf("can't import host FDs into filesystems of type %T", mnt.Filesystem().Impl()) @@ -78,8 +78,6 @@ func ImportFD(mnt *vfs.Mount, hostFD int, ownerUID auth.KUID, ownerGID auth.KGID canMap: canMap(uint32(fileType)), ino: fs.NextIno(), mode: fileMode, - uid: ownerUID, - gid: ownerGID, // For simplicity, set offset to 0. Technically, we should // only set to 0 on files that are not seekable (sockets, pipes, etc.), // and use the offset from the host fd otherwise. @@ -135,17 +133,20 @@ type inode struct { // This field is initialized at creation time and is immutable. ino uint64 - // TODO(gvisor.dev/issue/1672): protect mode, uid, and gid with mutex. + // modeMu protects mode. + modeMu sync.Mutex - // mode is the file mode of this inode. Note that this value may become out - // of date if the mode is changed on the host, e.g. with chmod. + // mode is a cached version of the file mode on the host. Note that it may + // become out of date if the mode is changed on the host, e.g. with chmod. + // + // Generally, it is better to retrieve the mode from the host through an + // fstat syscall. We only use this value in inode.Mode(), which cannot + // return an error, if the syscall to host fails. + // + // FIXME(b/152294168): Plumb error into Inode.Mode() return value so we + // can get rid of this. mode linux.FileMode - // uid and gid of the file owner. Note that these refer to the owner of the - // file created on import, not the fd on the host. - uid auth.KUID - gid auth.KGID - // offsetMu protects offset. offsetMu sync.Mutex @@ -168,12 +169,35 @@ func fileFlagsFromHostFD(fd int) (int, error) { // CheckPermissions implements kernfs.Inode. func (i *inode) CheckPermissions(ctx context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error { - return vfs.GenericCheckPermissions(creds, ats, i.mode, i.uid, i.gid) + mode, uid, gid, err := i.getPermissions() + if err != nil { + return err + } + return vfs.GenericCheckPermissions(creds, ats, mode, uid, gid) } // Mode implements kernfs.Inode. func (i *inode) Mode() linux.FileMode { - return i.mode + mode, _, _, err := i.getPermissions() + if err != nil { + return i.mode + } + + return linux.FileMode(mode) +} + +func (i *inode) getPermissions() (linux.FileMode, auth.KUID, auth.KGID, error) { + // Retrieve metadata. + var s syscall.Stat_t + if err := syscall.Fstat(i.hostFD, &s); err != nil { + return 0, 0, 0, err + } + + // Update cached mode. + i.modeMu.Lock() + i.mode = linux.FileMode(s.Mode) + i.modeMu.Unlock() + return linux.FileMode(s.Mode), auth.KUID(s.Uid), auth.KGID(s.Gid), nil } // Stat implements kernfs.Inode. @@ -213,45 +237,51 @@ func (i *inode) Stat(_ *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, erro ls.Attributes = s.Attributes ls.AttributesMask = s.Attributes_mask - if mask|linux.STATX_TYPE != 0 { + if mask&linux.STATX_TYPE != 0 { ls.Mode |= s.Mode & linux.S_IFMT } - if mask|linux.STATX_MODE != 0 { + if mask&linux.STATX_MODE != 0 { ls.Mode |= s.Mode &^ linux.S_IFMT } - if mask|linux.STATX_NLINK != 0 { + if mask&linux.STATX_NLINK != 0 { ls.Nlink = s.Nlink } - if mask|linux.STATX_ATIME != 0 { + if mask&linux.STATX_UID != 0 { + ls.UID = s.Uid + } + if mask&linux.STATX_GID != 0 { + ls.GID = s.Gid + } + if mask&linux.STATX_ATIME != 0 { ls.Atime = unixToLinuxStatxTimestamp(s.Atime) } - if mask|linux.STATX_BTIME != 0 { + if mask&linux.STATX_BTIME != 0 { ls.Btime = unixToLinuxStatxTimestamp(s.Btime) } - if mask|linux.STATX_CTIME != 0 { + if mask&linux.STATX_CTIME != 0 { ls.Ctime = unixToLinuxStatxTimestamp(s.Ctime) } - if mask|linux.STATX_MTIME != 0 { + if mask&linux.STATX_MTIME != 0 { ls.Mtime = unixToLinuxStatxTimestamp(s.Mtime) } - if mask|linux.STATX_SIZE != 0 { + if mask&linux.STATX_SIZE != 0 { ls.Size = s.Size } - if mask|linux.STATX_BLOCKS != 0 { + if mask&linux.STATX_BLOCKS != 0 { ls.Blocks = s.Blocks } - // Use our own internal inode number and file owner. - if mask|linux.STATX_INO != 0 { + // Use our own internal inode number. + if mask&linux.STATX_INO != 0 { ls.Ino = i.ino } - if mask|linux.STATX_UID != 0 { - ls.UID = uint32(i.uid) - } - if mask|linux.STATX_GID != 0 { - ls.GID = uint32(i.gid) - } + // Update cached mode. + if (mask&linux.STATX_TYPE != 0) && (mask&linux.STATX_MODE != 0) { + i.modeMu.Lock() + i.mode = linux.FileMode(s.Mode) + i.modeMu.Unlock() + } return ls, nil } @@ -274,6 +304,8 @@ func (i *inode) fstat(opts vfs.StatOptions) (linux.Statx, error) { Mask: linux.STATX_BASIC_STATS, Blksize: uint32(s.Blksize), Nlink: uint32(s.Nlink), + UID: s.Uid, + GID: s.Gid, Mode: uint16(s.Mode), Size: uint64(s.Size), Blocks: uint64(s.Blocks), @@ -282,15 +314,13 @@ func (i *inode) fstat(opts vfs.StatOptions) (linux.Statx, error) { Mtime: timespecToStatxTimestamp(s.Mtim), } - // Use our own internal inode number and file owner. + // Use our own internal inode number. // // TODO(gvisor.dev/issue/1672): Use a kernfs-specific device number as well. // If we use the device number from the host, it may collide with another // sentry-internal device number. We handle device/inode numbers without // relying on the host to prevent collisions. ls.Ino = i.ino - ls.UID = uint32(i.uid) - ls.GID = uint32(i.gid) return ls, nil } @@ -306,7 +336,11 @@ func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Cre if m&^(linux.STATX_MODE|linux.STATX_SIZE|linux.STATX_ATIME|linux.STATX_MTIME) != 0 { return syserror.EPERM } - if err := vfs.CheckSetStat(ctx, creds, &s, i.Mode(), i.uid, i.gid); err != nil { + mode, uid, gid, err := i.getPermissions() + if err != nil { + return err + } + if err := vfs.CheckSetStat(ctx, creds, &s, mode.Permissions(), uid, gid); err != nil { return err } @@ -314,7 +348,9 @@ func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Cre if err := syscall.Fchmod(i.hostFD, uint32(s.Mode)); err != nil { return err } + i.modeMu.Lock() i.mode = linux.FileMode(s.Mode) + i.modeMu.Unlock() } if m&linux.STATX_SIZE != 0 { if err := syscall.Ftruncate(i.hostFD, int64(s.Size)); err != nil { @@ -351,7 +387,11 @@ func (i *inode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptio } func (i *inode) open(d *vfs.Dentry, mnt *vfs.Mount) (*vfs.FileDescription, error) { - fileType := i.mode.FileType() + mode, _, _, err := i.getPermissions() + if err != nil { + return nil, err + } + fileType := mode.FileType() if fileType == syscall.S_IFSOCK { if i.isTTY { return nil, errors.New("cannot use host socket as TTY") diff --git a/runsc/boot/fds.go b/runsc/boot/fds.go index 417d2d5fb..5314b0f2a 100644 --- a/runsc/boot/fds.go +++ b/runsc/boot/fds.go @@ -34,7 +34,6 @@ func createFDTable(ctx context.Context, console bool, stdioFDs []int) (*kernel.F k := kernel.KernelFromContext(ctx) fdTable := k.NewFDTable() defer fdTable.DecRef() - mounter := fs.FileOwnerFromContext(ctx) var ttyFile *fs.File for appFD, hostFD := range stdioFDs { @@ -44,7 +43,7 @@ func createFDTable(ctx context.Context, console bool, stdioFDs []int) (*kernel.F // Import the file as a host TTY file. if ttyFile == nil { var err error - appFile, err = host.ImportFile(ctx, hostFD, mounter, true /* isTTY */) + appFile, err = host.ImportFile(ctx, hostFD, true /* isTTY */) if err != nil { return nil, err } @@ -63,7 +62,7 @@ func createFDTable(ctx context.Context, console bool, stdioFDs []int) (*kernel.F } else { // Import the file as a regular host file. var err error - appFile, err = host.ImportFile(ctx, hostFD, mounter, false /* isTTY */) + appFile, err = host.ImportFile(ctx, hostFD, false /* isTTY */) if err != nil { return nil, err } -- cgit v1.2.3 From 76a7ace751bfd4b16411edbc0a2b06d0308b8832 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Thu, 26 Mar 2020 21:47:46 -0700 Subject: Add BoundEndpointAt filesystem operation. BoundEndpointAt() is needed to support Unix sockets bound at a file path, corresponding to BoundEndpoint() in VFS1. Updates #1476. PiperOrigin-RevId: 303258251 --- pkg/sentry/fsimpl/ext/BUILD | 1 + pkg/sentry/fsimpl/ext/filesystem.go | 12 ++++++++++ pkg/sentry/fsimpl/gofer/BUILD | 1 + pkg/sentry/fsimpl/gofer/filesystem.go | 8 +++++++ pkg/sentry/fsimpl/kernfs/BUILD | 1 + pkg/sentry/fsimpl/kernfs/filesystem.go | 13 ++++++++++ pkg/sentry/fsimpl/tmpfs/BUILD | 1 + pkg/sentry/fsimpl/tmpfs/filesystem.go | 8 +++++++ pkg/sentry/vfs/BUILD | 1 + pkg/sentry/vfs/anonfs.go | 9 +++++++ pkg/sentry/vfs/filesystem.go | 8 ++++++- pkg/sentry/vfs/vfs.go | 44 +++++++++++++++++++++++++++------- pkg/syserror/syserror.go | 1 + 13 files changed, 99 insertions(+), 9 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/ext/BUILD b/pkg/sentry/fsimpl/ext/BUILD index 6f78f478f..d83d75b3d 100644 --- a/pkg/sentry/fsimpl/ext/BUILD +++ b/pkg/sentry/fsimpl/ext/BUILD @@ -45,6 +45,7 @@ go_library( "//pkg/sentry/fsimpl/ext/disklayout", "//pkg/sentry/kernel/auth", "//pkg/sentry/memmap", + "//pkg/sentry/socket/unix/transport", "//pkg/sentry/syscalls/linux", "//pkg/sentry/vfs", "//pkg/sync", diff --git a/pkg/sentry/fsimpl/ext/filesystem.go b/pkg/sentry/fsimpl/ext/filesystem.go index 8497be615..48eaccdbc 100644 --- a/pkg/sentry/fsimpl/ext/filesystem.go +++ b/pkg/sentry/fsimpl/ext/filesystem.go @@ -23,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" @@ -463,6 +464,17 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error return syserror.EROFS } +// BoundEndpointAt implements FilesystemImpl.BoundEndpointAt. +func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath) (transport.BoundEndpoint, error) { + _, _, err := fs.walk(rp, false) + if err != nil { + return nil, err + } + + // TODO(b/134676337): Support sockets. + return nil, syserror.ECONNREFUSED +} + // ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) { _, _, err := fs.walk(rp, false) diff --git a/pkg/sentry/fsimpl/gofer/BUILD b/pkg/sentry/fsimpl/gofer/BUILD index 4ba76a1e8..d15a36709 100644 --- a/pkg/sentry/fsimpl/gofer/BUILD +++ b/pkg/sentry/fsimpl/gofer/BUILD @@ -46,6 +46,7 @@ go_library( "//pkg/sentry/memmap", "//pkg/sentry/pgalloc", "//pkg/sentry/platform", + "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usage", "//pkg/sentry/vfs", "//pkg/syserror", diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 1e43df9ec..269624362 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -22,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" ) @@ -1059,6 +1060,13 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error return fs.unlinkAt(ctx, rp, false /* dir */) } +// BoundEndpointAt implements FilesystemImpl.BoundEndpointAt. +// +// TODO(gvisor.dev/issue/1476): Implement BoundEndpointAt. +func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath) (transport.BoundEndpoint, error) { + return nil, syserror.ECONNREFUSED +} + // ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) { var ds *[]*dentry diff --git a/pkg/sentry/fsimpl/kernfs/BUILD b/pkg/sentry/fsimpl/kernfs/BUILD index e73f1f857..b3d6299d0 100644 --- a/pkg/sentry/fsimpl/kernfs/BUILD +++ b/pkg/sentry/fsimpl/kernfs/BUILD @@ -35,6 +35,7 @@ go_library( "//pkg/refs", "//pkg/sentry/kernel/auth", "//pkg/sentry/memmap", + "//pkg/sentry/socket/unix/transport", "//pkg/sentry/vfs", "//pkg/sync", "//pkg/syserror", diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 31da8b511..a429fa23d 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -23,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" ) @@ -728,6 +729,18 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error return nil } +// BoundEndpointAt implements FilesystemImpl.BoundEndpointAt. +func (fs *Filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath) (transport.BoundEndpoint, error) { + fs.mu.RLock() + _, _, err := fs.walkExistingLocked(ctx, rp) + fs.mu.RUnlock() + fs.processDeferredDecRefs() + if err != nil { + return nil, err + } + return nil, syserror.ECONNREFUSED +} + // ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. func (fs *Filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) { fs.mu.RLock() diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD index 57abd5583..6ea35affb 100644 --- a/pkg/sentry/fsimpl/tmpfs/BUILD +++ b/pkg/sentry/fsimpl/tmpfs/BUILD @@ -46,6 +46,7 @@ go_library( "//pkg/sentry/memmap", "//pkg/sentry/pgalloc", "//pkg/sentry/platform", + "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usage", "//pkg/sentry/vfs", "//pkg/sentry/vfs/lock", diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 12cc64385..e678ecc37 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" ) @@ -656,6 +657,13 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error return nil } +// BoundEndpointAt implements FilesystemImpl.BoundEndpointAt. +// +// TODO(gvisor.dev/issue/1476): Implement BoundEndpointAt. +func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath) (transport.BoundEndpoint, error) { + return nil, syserror.ECONNREFUSED +} + // ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) { fs.mu.RLock() diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index a2a06fc8f..bf4d27c7d 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -53,6 +53,7 @@ go_library( "//pkg/sentry/kernel/auth", "//pkg/sentry/limits", "//pkg/sentry/memmap", + "//pkg/sentry/socket/unix/transport", "//pkg/sync", "//pkg/syserror", "//pkg/usermem", diff --git a/pkg/sentry/vfs/anonfs.go b/pkg/sentry/vfs/anonfs.go index a62e43589..f58867066 100644 --- a/pkg/sentry/vfs/anonfs.go +++ b/pkg/sentry/vfs/anonfs.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" ) @@ -222,6 +223,14 @@ func (fs *anonFilesystem) UnlinkAt(ctx context.Context, rp *ResolvingPath) error return syserror.EPERM } +// BoundEndpointAt implements FilesystemImpl.BoundEndpointAt. +func (fs *anonFilesystem) BoundEndpointAt(ctx context.Context, rp *ResolvingPath) (transport.BoundEndpoint, error) { + if !rp.Final() { + return nil, syserror.ENOTDIR + } + return nil, syserror.ECONNREFUSED +} + // ListxattrAt implements FilesystemImpl.ListxattrAt. func (fs *anonFilesystem) ListxattrAt(ctx context.Context, rp *ResolvingPath) ([]string, error) { if !rp.Done() { diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index 332decce6..7b7d233f9 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" ) // A Filesystem is a tree of nodes represented by Dentries, which forms part of @@ -460,6 +461,11 @@ type FilesystemImpl interface { // RemovexattrAt returns ENOTSUP. RemovexattrAt(ctx context.Context, rp *ResolvingPath, name string) error + // BoundEndpointAt returns the Unix socket endpoint bound at the path rp. + // + // - If a non-socket file exists at rp, then BoundEndpointAt returns ECONNREFUSED. + BoundEndpointAt(ctx context.Context, rp *ResolvingPath) (transport.BoundEndpoint, error) + // PrependPath prepends a path from vd to vd.Mount().Root() to b. // // If vfsroot.Ok(), it is the contextual VFS root; if it is encountered @@ -482,7 +488,7 @@ type FilesystemImpl interface { // Preconditions: vd.Mount().Filesystem().Impl() == this FilesystemImpl. PrependPath(ctx context.Context, vfsroot, vd VirtualDentry, b *fspath.Builder) error - // TODO: inotify_add_watch(); bind() + // TODO: inotify_add_watch() } // PrependPathAtVFSRootError is returned by implementations of diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 03d1fb943..1708c1a53 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -38,6 +38,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) @@ -230,7 +231,7 @@ func (vfs *VirtualFilesystem) getParentDirAndName(ctx context.Context, creds *au } if checkInvariants { if rp.canHandleError(err) && rp.Done() { - panic(fmt.Sprintf("%T.GetParentDentryAt() consumed all path components and returned %T", rp.mount.fs.impl, err)) + panic(fmt.Sprintf("%T.GetParentDentryAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } if !rp.handleError(err) { @@ -271,7 +272,7 @@ func (vfs *VirtualFilesystem) LinkAt(ctx context.Context, creds *auth.Credential } if checkInvariants { if rp.canHandleError(err) && rp.Done() { - panic(fmt.Sprintf("%T.LinkAt() consumed all path components and returned %T", rp.mount.fs.impl, err)) + panic(fmt.Sprintf("%T.LinkAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } if !rp.handleError(err) { @@ -307,7 +308,7 @@ func (vfs *VirtualFilesystem) MkdirAt(ctx context.Context, creds *auth.Credentia } if checkInvariants { if rp.canHandleError(err) && rp.Done() { - panic(fmt.Sprintf("%T.MkdirAt() consumed all path components and returned %T", rp.mount.fs.impl, err)) + panic(fmt.Sprintf("%T.MkdirAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } if !rp.handleError(err) { @@ -340,7 +341,7 @@ func (vfs *VirtualFilesystem) MknodAt(ctx context.Context, creds *auth.Credentia } if checkInvariants { if rp.canHandleError(err) && rp.Done() { - panic(fmt.Sprintf("%T.MknodAt() consumed all path components and returned %T", rp.mount.fs.impl, err)) + panic(fmt.Sprintf("%T.MknodAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } if !rp.handleError(err) { @@ -350,6 +351,33 @@ func (vfs *VirtualFilesystem) MknodAt(ctx context.Context, creds *auth.Credentia } } +// BoundEndpointAt gets the bound endpoint at the given path, if one exists. +func (vfs *VirtualFilesystem) BoundEndpointAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation) (transport.BoundEndpoint, error) { + if !pop.Path.Begin.Ok() { + if pop.Path.Absolute { + return nil, syserror.ECONNREFUSED + } + return nil, syserror.ENOENT + } + rp := vfs.getResolvingPath(creds, pop) + for { + bep, err := rp.mount.fs.impl.BoundEndpointAt(ctx, rp) + if err == nil { + vfs.putResolvingPath(rp) + return bep, nil + } + if checkInvariants { + if rp.canHandleError(err) && rp.Done() { + panic(fmt.Sprintf("%T.BoundEndpointAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) + } + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return nil, err + } + } +} + // OpenAt returns a FileDescription providing access to the file at the given // path. A reference is taken on the returned FileDescription. func (vfs *VirtualFilesystem) OpenAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *OpenOptions) (*FileDescription, error) { @@ -494,7 +522,7 @@ func (vfs *VirtualFilesystem) RenameAt(ctx context.Context, creds *auth.Credenti } if checkInvariants { if rp.canHandleError(err) && rp.Done() { - panic(fmt.Sprintf("%T.RenameAt() consumed all path components and returned %T", rp.mount.fs.impl, err)) + panic(fmt.Sprintf("%T.RenameAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } if !rp.handleError(err) { @@ -527,7 +555,7 @@ func (vfs *VirtualFilesystem) RmdirAt(ctx context.Context, creds *auth.Credentia } if checkInvariants { if rp.canHandleError(err) && rp.Done() { - panic(fmt.Sprintf("%T.RmdirAt() consumed all path components and returned %T", rp.mount.fs.impl, err)) + panic(fmt.Sprintf("%T.RmdirAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } if !rp.handleError(err) { @@ -608,7 +636,7 @@ func (vfs *VirtualFilesystem) SymlinkAt(ctx context.Context, creds *auth.Credent } if checkInvariants { if rp.canHandleError(err) && rp.Done() { - panic(fmt.Sprintf("%T.SymlinkAt() consumed all path components and returned %T", rp.mount.fs.impl, err)) + panic(fmt.Sprintf("%T.SymlinkAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } if !rp.handleError(err) { @@ -640,7 +668,7 @@ func (vfs *VirtualFilesystem) UnlinkAt(ctx context.Context, creds *auth.Credenti } if checkInvariants { if rp.canHandleError(err) && rp.Done() { - panic(fmt.Sprintf("%T.UnlinkAt() consumed all path components and returned %T", rp.mount.fs.impl, err)) + panic(fmt.Sprintf("%T.UnlinkAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } if !rp.handleError(err) { diff --git a/pkg/syserror/syserror.go b/pkg/syserror/syserror.go index 4b5a0fca6..f86db0999 100644 --- a/pkg/syserror/syserror.go +++ b/pkg/syserror/syserror.go @@ -27,6 +27,7 @@ import ( var ( E2BIG = error(syscall.E2BIG) EACCES = error(syscall.EACCES) + EADDRINUSE = error(syscall.EADDRINUSE) EAGAIN = error(syscall.EAGAIN) EBADF = error(syscall.EBADF) EBADFD = error(syscall.EBADFD) -- cgit v1.2.3 From be415eedcb3d5f61ed9f6a90bba8df2ecd54d8c8 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Wed, 26 Feb 2020 04:17:54 -0500 Subject: add arch-specific feature into mm Signed-off-by: Bin Lu --- pkg/sentry/mm/lifecycle.go | 1 + pkg/sentry/mm/metadata.go | 14 ++++++++++++++ pkg/sentry/mm/mm.go | 3 +++ 3 files changed, 18 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/mm/lifecycle.go b/pkg/sentry/mm/lifecycle.go index d8a5b9d29..e6269107c 100644 --- a/pkg/sentry/mm/lifecycle.go +++ b/pkg/sentry/mm/lifecycle.go @@ -84,6 +84,7 @@ func (mm *MemoryManager) Fork(ctx context.Context) (*MemoryManager, error) { dumpability: mm.dumpability, aioManager: aioManager{contexts: make(map[uint64]*AIOContext)}, sleepForActivation: mm.sleepForActivation, + vdsoSigReturnAddr : mm.vdsoSigReturnAddr, } // Copy vmas. diff --git a/pkg/sentry/mm/metadata.go b/pkg/sentry/mm/metadata.go index 6a49334f4..28e5057f7 100644 --- a/pkg/sentry/mm/metadata.go +++ b/pkg/sentry/mm/metadata.go @@ -167,3 +167,17 @@ func (mm *MemoryManager) SetExecutable(file fsbridge.File) { orig.DecRef() } } + +// VDSOSigReturn returns the address of vdso_sigreturn. +func (mm *MemoryManager) VDSOSigReturn() uint64 { + mm.metadataMu.Lock() + defer mm.metadataMu.Unlock() + return mm.vdsoSigReturnAddr +} + +// SetVDSOSigReturn sets the address of vdso_sigreturn. +func (mm *MemoryManager) SetVDSOSigReturn(addr uint64) { + mm.metadataMu.Lock() + defer mm.metadataMu.Unlock() + mm.vdsoSigReturnAddr = addr +} diff --git a/pkg/sentry/mm/mm.go b/pkg/sentry/mm/mm.go index c2195ae11..34d3bde7a 100644 --- a/pkg/sentry/mm/mm.go +++ b/pkg/sentry/mm/mm.go @@ -231,6 +231,9 @@ type MemoryManager struct { // before trying to activate the address space. When set to true, delays in // activation are not reported as stuck tasks by the watchdog. sleepForActivation bool + + // vdsoSigReturnAddr is the address of 'vdso_sigreturn'. + vdsoSigReturnAddr uint64 } // vma represents a virtual memory area. -- cgit v1.2.3 From 10f2c8db915df14102e3f4d9efcfce372c90707a Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Fri, 27 Mar 2020 16:53:28 -0700 Subject: Add FilesystemType.Name method, and FilesystemType field to Filesystem struct. Both have analogues in Linux: * struct file_system_type has a char *name field. * struct super_block keeps a pointer to the file_system_type. These fields are necessary to support the `filesystem type` field in /proc/[pid]/mountinfo. PiperOrigin-RevId: 303434063 --- pkg/sentry/fsimpl/devtmpfs/devtmpfs.go | 5 +++++ pkg/sentry/fsimpl/ext/ext.go | 12 ++++++++++-- pkg/sentry/fsimpl/gofer/gofer.go | 7 ++++++- pkg/sentry/fsimpl/host/host.go | 15 ++++++++++++++- pkg/sentry/fsimpl/kernfs/kernfs.go | 7 ++----- pkg/sentry/fsimpl/kernfs/kernfs_test.go | 8 ++++++-- pkg/sentry/fsimpl/proc/filesystem.go | 11 ++++++++--- pkg/sentry/fsimpl/sys/sys.go | 9 +++++++-- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 7 ++++++- pkg/sentry/vfs/anonfs.go | 13 +++++++++++++ pkg/sentry/vfs/filesystem.go | 11 ++++++++++- pkg/sentry/vfs/filesystem_type.go | 3 +++ pkg/sentry/vfs/vfs.go | 2 +- 13 files changed, 91 insertions(+), 19 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go index abd4f24e7..64f1b142c 100644 --- a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go +++ b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go @@ -42,6 +42,11 @@ type FilesystemType struct { root *vfs.Dentry } +// Name implements vfs.FilesystemType.Name. +func (*FilesystemType) Name() string { + return Name +} + // GetFilesystem implements vfs.FilesystemType.GetFilesystem. func (fst *FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { fst.initOnce.Do(func() { diff --git a/pkg/sentry/fsimpl/ext/ext.go b/pkg/sentry/fsimpl/ext/ext.go index 373d23b74..7176af6d1 100644 --- a/pkg/sentry/fsimpl/ext/ext.go +++ b/pkg/sentry/fsimpl/ext/ext.go @@ -30,6 +30,9 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) +// Name is the name of this filesystem. +const Name = "ext" + // FilesystemType implements vfs.FilesystemType. type FilesystemType struct{} @@ -91,8 +94,13 @@ func isCompatible(sb disklayout.SuperBlock) bool { return true } +// Name implements vfs.FilesystemType.Name. +func (FilesystemType) Name() string { + return Name +} + // GetFilesystem implements vfs.FilesystemType.GetFilesystem. -func (FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { +func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { // TODO(b/134676337): Ensure that the user is mounting readonly. If not, // EACCESS should be returned according to mount(2). Filesystem independent // flags (like readonly) are currently not available in pkg/sentry/vfs. @@ -103,7 +111,7 @@ func (FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFile } fs := filesystem{dev: dev, inodeCache: make(map[uint32]*inode)} - fs.vfsfs.Init(vfsObj, &fs) + fs.vfsfs.Init(vfsObj, &fsType, &fs) fs.sb, err = readSuperBlock(dev) if err != nil { return nil, nil, err diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index cf276a417..8e41b6b1c 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -199,6 +199,11 @@ const ( InteropModeShared ) +// Name implements vfs.FilesystemType.Name. +func (FilesystemType) Name() string { + return Name +} + // GetFilesystem implements vfs.FilesystemType.GetFilesystem. func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { mfp := pgalloc.MemoryFileProviderFromContext(ctx) @@ -374,7 +379,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt dentries: make(map[*dentry]struct{}), specialFileFDs: make(map[*specialFileFD]struct{}), } - fs.vfsfs.Init(vfsObj, fs) + fs.vfsfs.Init(vfsObj, &fstype, fs) // Construct the root dentry. root, err := fs.newDentry(ctx, attachFile, qid, attrMask, &attr) diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 17e3d6e9d..7d9dcd4c9 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -38,6 +38,19 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) +// filesystemType implements vfs.FilesystemType. +type filesystemType struct{} + +// GetFilesystem implements FilesystemType.GetFilesystem. +func (filesystemType) GetFilesystem(context.Context, *vfs.VirtualFilesystem, *auth.Credentials, string, vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { + panic("cannot instaniate a host filesystem") +} + +// Name implements FilesystemType.Name. +func (filesystemType) Name() string { + return "none" +} + // filesystem implements vfs.FilesystemImpl. type filesystem struct { kernfs.Filesystem @@ -46,7 +59,7 @@ type filesystem struct { // NewMount returns a new disconnected mount in vfsObj that may be passed to ImportFD. func NewMount(vfsObj *vfs.VirtualFilesystem) (*vfs.Mount, error) { fs := &filesystem{} - fs.Init(vfsObj) + fs.Init(vfsObj, &filesystemType{}) vfsfs := fs.VFSFilesystem() // NewDisconnectedMount will take an additional reference on vfsfs. defer vfsfs.DecRef() diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index 794e38908..2cefef020 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -63,9 +63,6 @@ import ( "gvisor.dev/gvisor/pkg/sync" ) -// FilesystemType implements vfs.FilesystemType. -type FilesystemType struct{} - // Filesystem mostly implements vfs.FilesystemImpl for a generic in-memory // filesystem. Concrete implementations are expected to embed this in their own // Filesystem type. @@ -138,8 +135,8 @@ func (fs *Filesystem) processDeferredDecRefsLocked() { // Init initializes a kernfs filesystem. This should be called from during // vfs.FilesystemType.NewFilesystem for the concrete filesystem embedding // kernfs. -func (fs *Filesystem) Init(vfsObj *vfs.VirtualFilesystem) { - fs.vfsfs.Init(vfsObj, fs) +func (fs *Filesystem) Init(vfsObj *vfs.VirtualFilesystem, fsType vfs.FilesystemType) { + fs.vfsfs.Init(vfsObj, fsType, fs) } // VFSFilesystem returns the generic vfs filesystem object. diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index fb0d25ad7..465451f35 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -187,9 +187,13 @@ func (*dir) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, err return nil, syserror.EPERM } -func (fst *fsType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opt vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { +func (fsType) Name() string { + return "kernfs" +} + +func (fst fsType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opt vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { fs := &filesystem{} - fs.Init(vfsObj) + fs.Init(vfsObj, &fst) root := fst.rootFn(creds, fs) return fs.VFSFilesystem(), root.VFSDentry(), nil } diff --git a/pkg/sentry/fsimpl/proc/filesystem.go b/pkg/sentry/fsimpl/proc/filesystem.go index 5c19d5522..104fc9030 100644 --- a/pkg/sentry/fsimpl/proc/filesystem.go +++ b/pkg/sentry/fsimpl/proc/filesystem.go @@ -36,8 +36,13 @@ type FilesystemType struct{} var _ vfs.FilesystemType = (*FilesystemType)(nil) -// GetFilesystem implements vfs.FilesystemType. -func (ft *FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { +// Name implements vfs.FilesystemType.Name. +func (FilesystemType) Name() string { + return Name +} + +// GetFilesystem implements vfs.FilesystemType.GetFilesystem. +func (ft FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { k := kernel.KernelFromContext(ctx) if k == nil { return nil, nil, fmt.Errorf("procfs requires a kernel") @@ -48,7 +53,7 @@ func (ft *FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virtual } procfs := &kernfs.Filesystem{} - procfs.VFSFilesystem().Init(vfsObj, procfs) + procfs.VFSFilesystem().Init(vfsObj, &ft, procfs) var cgroups map[string]string if opts.InternalData != nil { diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go index 7abfd62f2..5c617270e 100644 --- a/pkg/sentry/fsimpl/sys/sys.go +++ b/pkg/sentry/fsimpl/sys/sys.go @@ -39,10 +39,15 @@ type filesystem struct { kernfs.Filesystem } +// Name implements vfs.FilesystemType.Name. +func (FilesystemType) Name() string { + return Name +} + // GetFilesystem implements vfs.FilesystemType.GetFilesystem. -func (FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { +func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { fs := &filesystem{} - fs.Filesystem.Init(vfsObj) + fs.Filesystem.Init(vfsObj, &fsType) k := kernel.KernelFromContext(ctx) maxCPUCores := k.ApplicationCores() defaultSysDirMode := linux.FileMode(0755) diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 2f9e6c876..b07b0dbae 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -63,6 +63,11 @@ type filesystem struct { nextInoMinusOne uint64 // accessed using atomic memory operations } +// Name implements vfs.FilesystemType.Name. +func (FilesystemType) Name() string { + return Name +} + // GetFilesystem implements vfs.FilesystemType.GetFilesystem. func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { memFileProvider := pgalloc.MemoryFileProviderFromContext(ctx) @@ -74,7 +79,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt memFile: memFileProvider.MemoryFile(), clock: clock, } - fs.vfsfs.Init(vfsObj, &fs) + fs.vfsfs.Init(vfsObj, &fstype, &fs) root := fs.newDentry(fs.newDirectory(creds, 01777)) return &fs.vfsfs, &root.vfsd, nil } diff --git a/pkg/sentry/vfs/anonfs.go b/pkg/sentry/vfs/anonfs.go index f58867066..d1f6dfb45 100644 --- a/pkg/sentry/vfs/anonfs.go +++ b/pkg/sentry/vfs/anonfs.go @@ -51,6 +51,19 @@ const ( anonFileGID = auth.RootKGID ) +// anonFilesystemType implements FilesystemType. +type anonFilesystemType struct{} + +// GetFilesystem implements FilesystemType.GetFilesystem. +func (anonFilesystemType) GetFilesystem(context.Context, *VirtualFilesystem, *auth.Credentials, string, GetFilesystemOptions) (*Filesystem, *Dentry, error) { + panic("cannot instaniate an anon filesystem") +} + +// Name implemenents FilesystemType.Name. +func (anonFilesystemType) Name() string { + return "none" +} + // anonFilesystem is the implementation of FilesystemImpl that backs // VirtualDentries returned by VirtualFilesystem.NewAnonVirtualDentry(). // diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index 7b7d233f9..cd34782ff 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -42,21 +42,30 @@ type Filesystem struct { // immutable. vfs *VirtualFilesystem + // fsType is the FilesystemType of this Filesystem. + fsType FilesystemType + // impl is the FilesystemImpl associated with this Filesystem. impl is // immutable. This should be the last field in Dentry. impl FilesystemImpl } // Init must be called before first use of fs. -func (fs *Filesystem) Init(vfsObj *VirtualFilesystem, impl FilesystemImpl) { +func (fs *Filesystem) Init(vfsObj *VirtualFilesystem, fsType FilesystemType, impl FilesystemImpl) { fs.refs = 1 fs.vfs = vfsObj + fs.fsType = fsType fs.impl = impl vfsObj.filesystemsMu.Lock() vfsObj.filesystems[fs] = struct{}{} vfsObj.filesystemsMu.Unlock() } +// FilesystemType returns the FilesystemType for this Filesystem. +func (fs *Filesystem) FilesystemType() FilesystemType { + return fs.fsType +} + // VirtualFilesystem returns the containing VirtualFilesystem. func (fs *Filesystem) VirtualFilesystem() *VirtualFilesystem { return fs.vfs diff --git a/pkg/sentry/vfs/filesystem_type.go b/pkg/sentry/vfs/filesystem_type.go index bb9cada81..f2298f7f6 100644 --- a/pkg/sentry/vfs/filesystem_type.go +++ b/pkg/sentry/vfs/filesystem_type.go @@ -30,6 +30,9 @@ type FilesystemType interface { // along with its mount root. A reference is taken on the returned // Filesystem and Dentry. GetFilesystem(ctx context.Context, vfsObj *VirtualFilesystem, creds *auth.Credentials, source string, opts GetFilesystemOptions) (*Filesystem, *Dentry, error) + + // Name returns the name of this FilesystemType. + Name() string } // GetFilesystemOptions contains options to FilesystemType.GetFilesystem. diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 1708c1a53..720b90d8f 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -134,7 +134,7 @@ func (vfs *VirtualFilesystem) Init() error { anonfs := anonFilesystem{ devMinor: anonfsDevMinor, } - anonfs.vfsfs.Init(vfs, &anonfs) + anonfs.vfsfs.Init(vfs, &anonFilesystemType{}, &anonfs) defer anonfs.vfsfs.DecRef() anonMount, err := vfs.NewDisconnectedMount(&anonfs.vfsfs, nil, &MountOptions{}) if err != nil { -- cgit v1.2.3 From f6e4daa67ad5f07ac1bcff33476b4d13f49a69bc Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Fri, 27 Mar 2020 16:54:45 -0700 Subject: Add vfs.PathnameReachable(). /proc/[pid]/mount* omit mounts whose mount point is outside the chroot, which is checked (indirectly) via __d_path(). PiperOrigin-RevId: 303434226 --- pkg/sentry/vfs/pathname.go | 43 +++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 43 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/vfs/pathname.go b/pkg/sentry/vfs/pathname.go index b318c681a..f21a88034 100644 --- a/pkg/sentry/vfs/pathname.go +++ b/pkg/sentry/vfs/pathname.go @@ -90,6 +90,49 @@ loop: return b.String(), nil } +// PathnameReachable returns an absolute pathname to vd, consistent with +// Linux's __d_path() (as used by seq_path_root()). If vfsroot.Ok() and vd is +// not reachable from vfsroot, such that seq_path_root() would return SEQ_SKIP +// (causing the entire containing entry to be skipped), PathnameReachable +// returns ("", nil). +func (vfs *VirtualFilesystem) PathnameReachable(ctx context.Context, vfsroot, vd VirtualDentry) (string, error) { + b := getFSPathBuilder() + defer putFSPathBuilder(b) + haveRef := false + defer func() { + if haveRef { + vd.DecRef() + } + }() +loop: + for { + err := vd.mount.fs.impl.PrependPath(ctx, vfsroot, vd, b) + switch err.(type) { + case nil: + if vd.mount == vfsroot.mount && vd.mount.root == vfsroot.dentry { + break loop + } + nextVD := vfs.getMountpointAt(vd.mount, vfsroot) + if !nextVD.Ok() { + return "", nil + } + if haveRef { + vd.DecRef() + } + vd = nextVD + haveRef = true + case PrependPathAtVFSRootError: + break loop + case PrependPathAtNonMountRootError, PrependPathSyntheticError: + return "", nil + default: + return "", err + } + } + b.PrependByte('/') + return b.String(), nil +} + // PathnameForGetcwd returns an absolute pathname to vd, consistent with // Linux's sys_getcwd(). func (vfs *VirtualFilesystem) PathnameForGetcwd(ctx context.Context, vfsroot, vd VirtualDentry) (string, error) { -- cgit v1.2.3 From 3fac85da951f9f56d0232718ea7584250cf11f31 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Mon, 30 Mar 2020 12:36:30 -0700 Subject: kvm: handle exit reasons even under EINTR. In the case of other signals (preemption), inject a normal bounce and defer the signal until the vCPU has been returned from guest mode. PiperOrigin-RevId: 303799678 --- pkg/atomicbitops/atomicbitops_amd64.s | 16 ++-- pkg/atomicbitops/atomicbitops_arm64.s | 16 ++-- pkg/atomicbitops/atomicbitops_noasm.go | 8 ++ pkg/safecopy/safecopy.go | 4 +- pkg/safecopy/safecopy_unsafe.go | 6 +- pkg/sentry/platform/kvm/BUILD | 1 + pkg/sentry/platform/kvm/bluepill.go | 12 ++- pkg/sentry/platform/kvm/bluepill_unsafe.go | 97 ++++++++++++++++--------- pkg/sentry/platform/kvm/kvm_const.go | 1 + pkg/sentry/platform/kvm/kvm_test.go | 64 ++++++++++++++-- pkg/sentry/platform/kvm/machine.go | 9 +-- pkg/sentry/platform/kvm/machine_amd64_unsafe.go | 25 ------- pkg/sentry/platform/kvm/machine_arm64_unsafe.go | 26 ------- pkg/sentry/platform/kvm/machine_unsafe.go | 41 +++++++++++ runsc/sandbox/sandbox.go | 6 -- 15 files changed, 207 insertions(+), 125 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/atomicbitops/atomicbitops_amd64.s b/pkg/atomicbitops/atomicbitops_amd64.s index 54c887ee5..f0edd4de7 100644 --- a/pkg/atomicbitops/atomicbitops_amd64.s +++ b/pkg/atomicbitops/atomicbitops_amd64.s @@ -16,28 +16,28 @@ #include "textflag.h" -TEXT ·AndUint32(SB),$0-12 +TEXT ·AndUint32(SB),NOSPLIT,$0-12 MOVQ addr+0(FP), BP MOVL val+8(FP), AX LOCK ANDL AX, 0(BP) RET -TEXT ·OrUint32(SB),$0-12 +TEXT ·OrUint32(SB),NOSPLIT,$0-12 MOVQ addr+0(FP), BP MOVL val+8(FP), AX LOCK ORL AX, 0(BP) RET -TEXT ·XorUint32(SB),$0-12 +TEXT ·XorUint32(SB),NOSPLIT,$0-12 MOVQ addr+0(FP), BP MOVL val+8(FP), AX LOCK XORL AX, 0(BP) RET -TEXT ·CompareAndSwapUint32(SB),$0-20 +TEXT ·CompareAndSwapUint32(SB),NOSPLIT,$0-20 MOVQ addr+0(FP), DI MOVL old+8(FP), AX MOVL new+12(FP), DX @@ -46,28 +46,28 @@ TEXT ·CompareAndSwapUint32(SB),$0-20 MOVL AX, ret+16(FP) RET -TEXT ·AndUint64(SB),$0-16 +TEXT ·AndUint64(SB),NOSPLIT,$0-16 MOVQ addr+0(FP), BP MOVQ val+8(FP), AX LOCK ANDQ AX, 0(BP) RET -TEXT ·OrUint64(SB),$0-16 +TEXT ·OrUint64(SB),NOSPLIT,$0-16 MOVQ addr+0(FP), BP MOVQ val+8(FP), AX LOCK ORQ AX, 0(BP) RET -TEXT ·XorUint64(SB),$0-16 +TEXT ·XorUint64(SB),NOSPLIT,$0-16 MOVQ addr+0(FP), BP MOVQ val+8(FP), AX LOCK XORQ AX, 0(BP) RET -TEXT ·CompareAndSwapUint64(SB),$0-32 +TEXT ·CompareAndSwapUint64(SB),NOSPLIT,$0-32 MOVQ addr+0(FP), DI MOVQ old+8(FP), AX MOVQ new+16(FP), DX diff --git a/pkg/atomicbitops/atomicbitops_arm64.s b/pkg/atomicbitops/atomicbitops_arm64.s index 5c780851b..644a6bca5 100644 --- a/pkg/atomicbitops/atomicbitops_arm64.s +++ b/pkg/atomicbitops/atomicbitops_arm64.s @@ -16,7 +16,7 @@ #include "textflag.h" -TEXT ·AndUint32(SB),$0-12 +TEXT ·AndUint32(SB),NOSPLIT,$0-12 MOVD ptr+0(FP), R0 MOVW val+8(FP), R1 again: @@ -26,7 +26,7 @@ again: CBNZ R3, again RET -TEXT ·OrUint32(SB),$0-12 +TEXT ·OrUint32(SB),NOSPLIT,$0-12 MOVD ptr+0(FP), R0 MOVW val+8(FP), R1 again: @@ -36,7 +36,7 @@ again: CBNZ R3, again RET -TEXT ·XorUint32(SB),$0-12 +TEXT ·XorUint32(SB),NOSPLIT,$0-12 MOVD ptr+0(FP), R0 MOVW val+8(FP), R1 again: @@ -46,7 +46,7 @@ again: CBNZ R3, again RET -TEXT ·CompareAndSwapUint32(SB),$0-20 +TEXT ·CompareAndSwapUint32(SB),NOSPLIT,$0-20 MOVD addr+0(FP), R0 MOVW old+8(FP), R1 MOVW new+12(FP), R2 @@ -60,7 +60,7 @@ done: MOVW R3, prev+16(FP) RET -TEXT ·AndUint64(SB),$0-16 +TEXT ·AndUint64(SB),NOSPLIT,$0-16 MOVD ptr+0(FP), R0 MOVD val+8(FP), R1 again: @@ -70,7 +70,7 @@ again: CBNZ R3, again RET -TEXT ·OrUint64(SB),$0-16 +TEXT ·OrUint64(SB),NOSPLIT,$0-16 MOVD ptr+0(FP), R0 MOVD val+8(FP), R1 again: @@ -80,7 +80,7 @@ again: CBNZ R3, again RET -TEXT ·XorUint64(SB),$0-16 +TEXT ·XorUint64(SB),NOSPLIT,$0-16 MOVD ptr+0(FP), R0 MOVD val+8(FP), R1 again: @@ -90,7 +90,7 @@ again: CBNZ R3, again RET -TEXT ·CompareAndSwapUint64(SB),$0-32 +TEXT ·CompareAndSwapUint64(SB),NOSPLIT,$0-32 MOVD addr+0(FP), R0 MOVD old+8(FP), R1 MOVD new+16(FP), R2 diff --git a/pkg/atomicbitops/atomicbitops_noasm.go b/pkg/atomicbitops/atomicbitops_noasm.go index 3b2898256..4e9c27b98 100644 --- a/pkg/atomicbitops/atomicbitops_noasm.go +++ b/pkg/atomicbitops/atomicbitops_noasm.go @@ -20,6 +20,7 @@ import ( "sync/atomic" ) +//go:nosplit func AndUint32(addr *uint32, val uint32) { for { o := atomic.LoadUint32(addr) @@ -30,6 +31,7 @@ func AndUint32(addr *uint32, val uint32) { } } +//go:nosplit func OrUint32(addr *uint32, val uint32) { for { o := atomic.LoadUint32(addr) @@ -40,6 +42,7 @@ func OrUint32(addr *uint32, val uint32) { } } +//go:nosplit func XorUint32(addr *uint32, val uint32) { for { o := atomic.LoadUint32(addr) @@ -50,6 +53,7 @@ func XorUint32(addr *uint32, val uint32) { } } +//go:nosplit func CompareAndSwapUint32(addr *uint32, old, new uint32) (prev uint32) { for { prev = atomic.LoadUint32(addr) @@ -62,6 +66,7 @@ func CompareAndSwapUint32(addr *uint32, old, new uint32) (prev uint32) { } } +//go:nosplit func AndUint64(addr *uint64, val uint64) { for { o := atomic.LoadUint64(addr) @@ -72,6 +77,7 @@ func AndUint64(addr *uint64, val uint64) { } } +//go:nosplit func OrUint64(addr *uint64, val uint64) { for { o := atomic.LoadUint64(addr) @@ -82,6 +88,7 @@ func OrUint64(addr *uint64, val uint64) { } } +//go:nosplit func XorUint64(addr *uint64, val uint64) { for { o := atomic.LoadUint64(addr) @@ -92,6 +99,7 @@ func XorUint64(addr *uint64, val uint64) { } } +//go:nosplit func CompareAndSwapUint64(addr *uint64, old, new uint64) (prev uint64) { for { prev = atomic.LoadUint64(addr) diff --git a/pkg/safecopy/safecopy.go b/pkg/safecopy/safecopy.go index 2fb7e5809..521f1a82d 100644 --- a/pkg/safecopy/safecopy.go +++ b/pkg/safecopy/safecopy.go @@ -127,10 +127,10 @@ func initializeAddresses() { func init() { initializeAddresses() - if err := ReplaceSignalHandler(syscall.SIGSEGV, reflect.ValueOf(signalHandler).Pointer(), &savedSigSegVHandler); err != nil { + if err := ReplaceSignalHandler(syscall.SIGSEGV, reflect.ValueOf(signalHandler).Pointer(), &savedSigSegVHandler, 0); err != nil { panic(fmt.Sprintf("Unable to set handler for SIGSEGV: %v", err)) } - if err := ReplaceSignalHandler(syscall.SIGBUS, reflect.ValueOf(signalHandler).Pointer(), &savedSigBusHandler); err != nil { + if err := ReplaceSignalHandler(syscall.SIGBUS, reflect.ValueOf(signalHandler).Pointer(), &savedSigBusHandler, 0); err != nil { panic(fmt.Sprintf("Unable to set handler for SIGBUS: %v", err)) } syserror.AddErrorUnwrapper(func(e error) (syscall.Errno, bool) { diff --git a/pkg/safecopy/safecopy_unsafe.go b/pkg/safecopy/safecopy_unsafe.go index 41dd567f3..b15b920fe 100644 --- a/pkg/safecopy/safecopy_unsafe.go +++ b/pkg/safecopy/safecopy_unsafe.go @@ -324,11 +324,13 @@ func errorFromFaultSignal(addr uintptr, sig int32) error { // // It stores the value of the previously set handler in previous. // +// The extraMask parameter is OR'ed into the existing signal handler mask. +// // This function will be called on initialization in order to install safecopy // handlers for appropriate signals. These handlers will call the previous // handler however, and if this is function is being used externally then the // same courtesy is expected. -func ReplaceSignalHandler(sig syscall.Signal, handler uintptr, previous *uintptr) error { +func ReplaceSignalHandler(sig syscall.Signal, handler uintptr, previous *uintptr, extraMask uint64) error { var sa struct { handler uintptr flags uint64 @@ -348,10 +350,10 @@ func ReplaceSignalHandler(sig syscall.Signal, handler uintptr, previous *uintptr if sa.handler == 0 { return fmt.Errorf("previous handler for signal %x isn't set", sig) } - *previous = sa.handler // Install our own handler. + sa.mask |= extraMask sa.handler = handler if _, _, e := syscall.RawSyscall6(syscall.SYS_RT_SIGACTION, uintptr(sig), uintptr(unsafe.Pointer(&sa)), 0, maskLen, 0, 0); e != 0 { return e diff --git a/pkg/sentry/platform/kvm/BUILD b/pkg/sentry/platform/kvm/BUILD index 159f7eafd..e27f57536 100644 --- a/pkg/sentry/platform/kvm/BUILD +++ b/pkg/sentry/platform/kvm/BUILD @@ -70,6 +70,7 @@ go_test( "requires-kvm", ], deps = [ + "//pkg/procid", "//pkg/sentry/arch", "//pkg/sentry/platform", "//pkg/sentry/platform/kvm/testutil", diff --git a/pkg/sentry/platform/kvm/bluepill.go b/pkg/sentry/platform/kvm/bluepill.go index 4b23f7803..555b5fa96 100644 --- a/pkg/sentry/platform/kvm/bluepill.go +++ b/pkg/sentry/platform/kvm/bluepill.go @@ -46,6 +46,14 @@ var ( // bounceSignalMask has only bounceSignal set. bounceSignalMask = uint64(1 << (uint64(bounceSignal) - 1)) + // otherSignalsMask includes all other signals that will be cause the + // vCPU to exit during execution. + // + // Currently, this includes the preemption signal and the profiling + // signal. In general, these should be signals whose delivery actually + // influences the way the program executes as the switch can be costly. + otherSignalsMask = uint64(1<<(uint64(syscall.SIGURG)-1)) | uint64(1<<(uint64(syscall.SIGPROF)-1)) + // bounce is the interrupt vector used to return to the kernel. bounce = uint32(ring0.VirtualizationException) @@ -86,8 +94,8 @@ func (c *vCPU) die(context *arch.SignalContext64, msg string) { } func init() { - // Install the handler. - if err := safecopy.ReplaceSignalHandler(bluepillSignal, reflect.ValueOf(sighandler).Pointer(), &savedHandler); err != nil { + // Install the handler, masking all signals. + if err := safecopy.ReplaceSignalHandler(bluepillSignal, reflect.ValueOf(sighandler).Pointer(), &savedHandler, ^uint64(0)); err != nil { panic(fmt.Sprintf("Unable to set handler for signal %d: %v", bluepillSignal, err)) } diff --git a/pkg/sentry/platform/kvm/bluepill_unsafe.go b/pkg/sentry/platform/kvm/bluepill_unsafe.go index 9add7c944..4e9d80765 100644 --- a/pkg/sentry/platform/kvm/bluepill_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_unsafe.go @@ -24,6 +24,7 @@ import ( "syscall" "unsafe" + "gvisor.dev/gvisor/pkg/atomicbitops" "gvisor.dev/gvisor/pkg/sentry/arch" ) @@ -58,6 +59,19 @@ func bluepillArchContext(context unsafe.Pointer) *arch.SignalContext64 { return &((*arch.UContext64)(context).MContext) } +// injectInterrupt is a helper to inject an interrupt. +// +//go:nosplit +func injectInterrupt(c *vCPU) { + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(c.fd), + _KVM_INTERRUPT, + uintptr(unsafe.Pointer(&bounce))); errno != 0 { + throw("interrupt injection failed") + } +} + // bluepillHandler is called from the signal stub. // // The world may be stopped while this is executing, and it executes on the @@ -69,6 +83,9 @@ func bluepillHandler(context unsafe.Pointer) { // Sanitize the registers; interrupts must always be disabled. c := bluepillArchEnter(bluepillArchContext(context)) + // Enable preemption. + c.setSignalMask(true) + // Increment the number of switches. atomic.AddUint32(&c.switches, 1) @@ -89,6 +106,9 @@ func bluepillHandler(context unsafe.Pointer) { // interrupted KVM. Since we're in a signal handler // currently, all signals are masked and the signal // must have been delivered directly to this thread. + // + // We will not be able to actually do subsequent + // KVM_RUNs until this signal is processed. timeout := syscall.Timespec{} sig, _, errno := syscall.RawSyscall6( syscall.SYS_RT_SIGTIMEDWAIT, @@ -98,12 +118,24 @@ func bluepillHandler(context unsafe.Pointer) { 8, // sigset size. 0, 0) if errno == syscall.EAGAIN { - continue - } - if errno != 0 { + // If weren't able to process this signal, then + // it must not have been in the bounceMask. By + // elimination, it must have been the + // preemption signal. We can't process this + // signal right now, so we need to disable + // preemption until the interrupt is actually + // handled. + c.setSignalMask(false) + // Note that there is a waiter for this vCPU. + // This will cause the vCPU to exit at some + // point in the future (releasing the user lock + // and guest mode). + atomicbitops.OrUint32(&c.state, vCPUWaiter) + } else if errno != 0 { + // We only expect success or a timeout. throw("error waiting for pending signal") - } - if sig != uintptr(bounceSignal) { + } else if sig != uintptr(bounceSignal) { + // Only the bounce should be processed. throw("unexpected signal") } @@ -114,11 +146,10 @@ func bluepillHandler(context unsafe.Pointer) { // ready. if c.runData.readyForInterruptInjection == 0 { c.runData.requestInterruptWindow = 1 - continue // Rerun vCPU. } else { - // Force injection below; the vCPU is ready. - c.runData.exitReason = _KVM_EXIT_IRQ_WINDOW_OPEN + injectInterrupt(c) } + continue // Rerun vCPU. case syscall.EFAULT: // If a fault is not serviceable due to the host // backing pages having page permissions, instead of an @@ -137,6 +168,30 @@ func bluepillHandler(context unsafe.Pointer) { } switch c.runData.exitReason { + case _KVM_EXIT_HLT: + // Copy out registers. + bluepillArchExit(c, bluepillArchContext(context)) + + // Return to the vCPUReady state; notify any waiters. + user := atomic.LoadUint32(&c.state) & vCPUUser + switch atomic.SwapUint32(&c.state, user) { + case user | vCPUGuest: // Expected case. + case user | vCPUGuest | vCPUWaiter: + c.notify() + default: + throw("invalid state") + } + return + case _KVM_EXIT_IRQ_WINDOW_OPEN: + // Inject an interrupt now. + injectInterrupt(c) + // Clear previous injection request. + c.runData.requestInterruptWindow = 0 + case _KVM_EXIT_INTR: + // This is fine, it is the normal exit reason during + // signal delivery. However, we still need to handle + // other potential exit reasons *combined* with EINTR, + // so this switch must be hit even after the above. case _KVM_EXIT_EXCEPTION: c.die(bluepillArchContext(context), "exception") return @@ -155,20 +210,6 @@ func bluepillHandler(context unsafe.Pointer) { case _KVM_EXIT_DEBUG: c.die(bluepillArchContext(context), "debug") return - case _KVM_EXIT_HLT: - // Copy out registers. - bluepillArchExit(c, bluepillArchContext(context)) - - // Return to the vCPUReady state; notify any waiters. - user := atomic.LoadUint32(&c.state) & vCPUUser - switch atomic.SwapUint32(&c.state, user) { - case user | vCPUGuest: // Expected case. - case user | vCPUGuest | vCPUWaiter: - c.notify() - default: - throw("invalid state") - } - return case _KVM_EXIT_MMIO: // Increment the fault count. atomic.AddUint32(&c.faults, 1) @@ -200,18 +241,6 @@ func bluepillHandler(context unsafe.Pointer) { data[i] = *b } } - case _KVM_EXIT_IRQ_WINDOW_OPEN: - // Interrupt: we must have requested an interrupt - // window; set the interrupt line. - if _, _, errno := syscall.RawSyscall( - syscall.SYS_IOCTL, - uintptr(c.fd), - _KVM_INTERRUPT, - uintptr(unsafe.Pointer(&bounce))); errno != 0 { - throw("interrupt injection failed") - } - // Clear previous injection request. - c.runData.requestInterruptWindow = 0 case _KVM_EXIT_SHUTDOWN: c.die(bluepillArchContext(context), "shutdown") return diff --git a/pkg/sentry/platform/kvm/kvm_const.go b/pkg/sentry/platform/kvm/kvm_const.go index 1d5c77ff4..07d9c9a98 100644 --- a/pkg/sentry/platform/kvm/kvm_const.go +++ b/pkg/sentry/platform/kvm/kvm_const.go @@ -48,6 +48,7 @@ const ( _KVM_EXIT_IRQ_WINDOW_OPEN = 0x7 _KVM_EXIT_SHUTDOWN = 0x8 _KVM_EXIT_FAIL_ENTRY = 0x9 + _KVM_EXIT_INTR = 0xa _KVM_EXIT_INTERNAL_ERROR = 0x11 _KVM_EXIT_SYSTEM_EVENT = 0x18 ) diff --git a/pkg/sentry/platform/kvm/kvm_test.go b/pkg/sentry/platform/kvm/kvm_test.go index c42752d50..d42ba3f24 100644 --- a/pkg/sentry/platform/kvm/kvm_test.go +++ b/pkg/sentry/platform/kvm/kvm_test.go @@ -16,12 +16,15 @@ package kvm import ( "math/rand" + "os" "reflect" + "runtime" "sync/atomic" "syscall" "testing" "time" + "gvisor.dev/gvisor/pkg/procid" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/kvm/testutil" @@ -320,15 +323,18 @@ func TestBounce(t *testing.T) { }) } +// randomSleep is used by some race tests below. +// +// O(hundreds of microseconds) is appropriate to ensure different overlaps and +// different schedules. +func randomSleep() { + if n := rand.Intn(1000); n > 100 { + time.Sleep(time.Duration(n) * time.Microsecond) + } +} + func TestBounceStress(t *testing.T) { applicationTest(t, true, testutil.SpinLoop, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { - randomSleep := func() { - // O(hundreds of microseconds) is appropriate to ensure - // different overlaps and different schedules. - if n := rand.Intn(1000); n > 100 { - time.Sleep(time.Duration(n) * time.Microsecond) - } - } for i := 0; i < 1000; i++ { // Start an asynchronously executing goroutine that // calls Bounce at pseudo-random point in time. @@ -355,6 +361,50 @@ func TestBounceStress(t *testing.T) { }) } +func TestPreemption(t *testing.T) { + applicationTest(t, true, testutil.SpinLoop, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { + // Lock the main vCPU thread. + runtime.LockOSThread() + pid := os.Getpid() + tid := procid.Current() + running := uint32(1) + defer atomic.StoreUint32(&running, 0) + + // Start generating "preemptions". + go func() { + for atomic.LoadUint32(&running) != 0 { + // Kick via a preemption: best effort. + syscall.Tgkill(pid, int(tid), syscall.SIGURG) + randomSleep() + } + }() + + for i := 0; i < 1000; i++ { + randomSleep() + var si arch.SignalInfo + if _, err := c.SwitchToUser(ring0.SwitchOpts{ + Registers: regs, + FloatingPointState: dummyFPState, + PageTables: pt, + }, &si); err != platform.ErrContextInterrupt { + t.Errorf("application partial restore: got %v, wanted %v", err, platform.ErrContextInterrupt) + } + // Was this caused by a preemption signal? + if got := atomic.LoadUint32(&c.state); got&vCPUGuest != 0 && got&vCPUWaiter == 0 { + continue + } + c.unlock() + // Should have dropped from guest mode, processed preemption. + if got := atomic.LoadUint32(&c.state); got != vCPUReady { + t.Errorf("vCPU not in ready state: got %v", got) + } + randomSleep() + c.lock() + } + return false + }) +} + func TestInvalidate(t *testing.T) { var data uintptr // Used below. applicationTest(t, true, testutil.Touch, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { diff --git a/pkg/sentry/platform/kvm/machine.go b/pkg/sentry/platform/kvm/machine.go index f1afc74dc..345b71e8f 100644 --- a/pkg/sentry/platform/kvm/machine.go +++ b/pkg/sentry/platform/kvm/machine.go @@ -108,6 +108,9 @@ type vCPU struct { // This is a bitmask of the three fields (vCPU*) described above. state uint32 + // signalMask is the vCPU signal mask. + signalMask uint64 + // runData for this vCPU. runData *runData @@ -121,6 +124,7 @@ type vCPU struct { // vCPUArchState is the architecture-specific state. vCPUArchState + // dieState is the temporary state associated with throwing exceptions. dieState dieState } @@ -153,11 +157,6 @@ func (m *machine) newVCPU() *vCPU { c.CPU.Init(&m.kernel, c) m.vCPUsByID[c.id] = c - // Ensure the signal mask is correct. - if err := c.setSignalMask(); err != nil { - panic(fmt.Sprintf("error setting signal mask: %v", err)) - } - // Map the run data. runData, err := mapRunData(int(fd)) if err != nil { diff --git a/pkg/sentry/platform/kvm/machine_amd64_unsafe.go b/pkg/sentry/platform/kvm/machine_amd64_unsafe.go index 7156c245f..52286e56d 100644 --- a/pkg/sentry/platform/kvm/machine_amd64_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_amd64_unsafe.go @@ -111,31 +111,6 @@ func (c *vCPU) setSystemTime() error { return nil } -// setSignalMask sets the vCPU signal mask. -// -// This must be called prior to running the vCPU. -func (c *vCPU) setSignalMask() error { - // The layout of this structure implies that it will not necessarily be - // the same layout chosen by the Go compiler. It gets fudged here. - var data struct { - length uint32 - mask1 uint32 - mask2 uint32 - _ uint32 - } - data.length = 8 // Fixed sigset size. - data.mask1 = ^uint32(bounceSignalMask & 0xffffffff) - data.mask2 = ^uint32(bounceSignalMask >> 32) - if _, _, errno := syscall.RawSyscall( - syscall.SYS_IOCTL, - uintptr(c.fd), - _KVM_SET_SIGNAL_MASK, - uintptr(unsafe.Pointer(&data))); errno != 0 { - return fmt.Errorf("error setting signal mask: %v", errno) - } - return nil -} - // setUserRegisters sets user registers in the vCPU. func (c *vCPU) setUserRegisters(uregs *userRegs) error { if _, _, errno := syscall.RawSyscall( diff --git a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go index b531f2f85..185eeb4f0 100644 --- a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go @@ -268,32 +268,6 @@ func (c *vCPU) setSystemTime() error { return nil } -// setSignalMask sets the vCPU signal mask. -// -// This must be called prior to running the vCPU. -func (c *vCPU) setSignalMask() error { - // The layout of this structure implies that it will not necessarily be - // the same layout chosen by the Go compiler. It gets fudged here. - var data struct { - length uint32 - mask1 uint32 - mask2 uint32 - _ uint32 - } - data.length = 8 // Fixed sigset size. - data.mask1 = ^uint32(bounceSignalMask & 0xffffffff) - data.mask2 = ^uint32(bounceSignalMask >> 32) - if _, _, errno := syscall.RawSyscall( - syscall.SYS_IOCTL, - uintptr(c.fd), - _KVM_SET_SIGNAL_MASK, - uintptr(unsafe.Pointer(&data))); errno != 0 { - return fmt.Errorf("error setting signal mask: %v", errno) - } - - return nil -} - // SwitchToUser unpacks architectural-details. func (c *vCPU) SwitchToUser(switchOpts ring0.SwitchOpts, info *arch.SignalInfo) (usermem.AccessType, error) { // Check for canonical addresses. diff --git a/pkg/sentry/platform/kvm/machine_unsafe.go b/pkg/sentry/platform/kvm/machine_unsafe.go index f04be2ab5..e4de0a889 100644 --- a/pkg/sentry/platform/kvm/machine_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_unsafe.go @@ -87,6 +87,47 @@ func unmapRunData(r *runData) error { return nil } +// setSignalMask sets the vCPU signal mask. +// +// This will be called from the bluepill handler, and therefore must not +// perform any allocation. +// +//go:nosplit +func (c *vCPU) setSignalMask(enableOthers bool) { + // The signal mask is either: + // *) Only the bounce signal, which we need to use to execute the + // machine state up until the bounce interrupt can be processed. + // or + // *) All signals, which is the default state unless we need to + // continue execution to exit guest mode (the case above). + mask := bounceSignalMask + if enableOthers { + mask |= otherSignalsMask + } + if c.signalMask == mask { + return // Already set. + } + + // The layout of this structure implies that it will not necessarily be + // the same layout chosen by the Go compiler. It gets fudged here. + var data struct { + length uint32 + mask1 uint32 + mask2 uint32 + _ uint32 + } + data.length = 8 // Fixed sigset size. + data.mask1 = ^uint32(mask & 0xffffffff) + data.mask2 = ^uint32(mask >> 32) + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(c.fd), + _KVM_SET_SIGNAL_MASK, + uintptr(unsafe.Pointer(&data))); errno != 0 { + throw("setSignal mask failed") + } +} + // atomicAddressSpace is an atomic address space pointer. type atomicAddressSpace struct { pointer unsafe.Pointer diff --git a/runsc/sandbox/sandbox.go b/runsc/sandbox/sandbox.go index 8de75ae57..6c15727fa 100644 --- a/runsc/sandbox/sandbox.go +++ b/runsc/sandbox/sandbox.go @@ -444,12 +444,6 @@ func (s *Sandbox) createSandboxProcess(conf *boot.Config, args *Args, startSyncF nextFD++ } - // TODO(b/151157106): syscall tests fail by timeout if asyncpreemptoff - // isn't set. - if conf.Platform == "kvm" { - cmd.Env = append(cmd.Env, "GODEBUG=asyncpreemptoff=1") - } - // The current process' stdio must be passed to the application via the // --stdio-fds flag. The stdio of the sandbox process itself must not // be connected to the same FDs, otherwise we risk leaking sandbox -- cgit v1.2.3 From 9de982ea790ffe56eca07b6535e9420b669b7c0c Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Tue, 31 Mar 2020 15:00:25 -0700 Subject: Allow passing root file type to tmpfs. PiperOrigin-RevId: 304053357 --- pkg/sentry/fsimpl/testutil/testutil.go | 3 +++ pkg/sentry/fsimpl/tmpfs/tmpfs.go | 32 ++++++++++++++++++++++++++++++-- 2 files changed, 33 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/testutil/testutil.go b/pkg/sentry/fsimpl/testutil/testutil.go index e16808c63..0556af877 100644 --- a/pkg/sentry/fsimpl/testutil/testutil.go +++ b/pkg/sentry/fsimpl/testutil/testutil.go @@ -162,6 +162,9 @@ func (s *System) ListDirents(pop *vfs.PathOperation) *DirentCollector { // exactly the specified set of expected entries. AssertAllDirentTypes respects // collector.skipDots, and implicitly checks for "." and ".." accordingly. func (s *System) AssertAllDirentTypes(collector *DirentCollector, expected map[string]DirentType) { + if expected == nil { + expected = make(map[string]DirentType) + } // Also implicitly check for "." and "..", if enabled. if !collector.skipDots { expected["."] = linux.DT_DIR diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index b07b0dbae..afd9f8533 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -68,6 +68,17 @@ func (FilesystemType) Name() string { return Name } +// FilesystemOpts is used to pass configuration data to tmpfs. +type FilesystemOpts struct { + // RootFileType is the FileType of the filesystem root. Valid values + // are: S_IFDIR, S_IFREG, and S_IFLNK. Defaults to S_IFDIR. + RootFileType uint16 + + // RootSymlinkTarget is the target of the root symlink. Only valid if + // RootFileType == S_IFLNK. + RootSymlinkTarget string +} + // GetFilesystem implements vfs.FilesystemType.GetFilesystem. func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { memFileProvider := pgalloc.MemoryFileProviderFromContext(ctx) @@ -79,9 +90,26 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt memFile: memFileProvider.MemoryFile(), clock: clock, } + fs.vfsfs.Init(vfsObj, &fstype, &fs) - root := fs.newDentry(fs.newDirectory(creds, 01777)) - return &fs.vfsfs, &root.vfsd, nil + + typ := uint16(linux.S_IFDIR) + tmpfsOpts, ok := opts.InternalData.(FilesystemOpts) + if ok && tmpfsOpts.RootFileType != 0 { + typ = tmpfsOpts.RootFileType + } + var root *inode + switch typ { + case linux.S_IFREG: + root = fs.newRegularFile(creds, 0777) + case linux.S_IFLNK: + root = fs.newSymlink(creds, tmpfsOpts.RootSymlinkTarget) + case linux.S_IFDIR: + root = fs.newDirectory(creds, 01777) + default: + return nil, nil, fmt.Errorf("invalid tmpfs root file type: %#o", typ) + } + return &fs.vfsfs, &fs.newDentry(root).vfsd, nil } // Release implements vfs.FilesystemImpl.Release. -- cgit v1.2.3 From e1c8eaca8f8413b17dab8f01b2e123e9d4b9ddbc Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Tue, 31 Mar 2020 15:00:26 -0700 Subject: Fix /proc/self/mounts and /proc/self/mountinfo in VFS2. Some extra fields were added to the Mount type to expose necessary data to the proc filesystem. PiperOrigin-RevId: 304053361 --- pkg/sentry/fsimpl/proc/task_files.go | 183 +++++---------------------------- pkg/sentry/vfs/mount.go | 192 ++++++++++++++++++++++++++++++++++- 2 files changed, 219 insertions(+), 156 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index 8c743df8d..df0d1bcc5 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -18,13 +18,10 @@ import ( "bytes" "fmt" "io" - "sort" - "strings" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/safemem" - "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fsbridge" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -634,51 +631,6 @@ func (s *exeSymlink) executable() (file fsbridge.File, err error) { return } -// forEachMountSource runs f for the process root mount and each mount that is -// a descendant of the root. -func forEachMount(t *kernel.Task, fn func(string, *fs.Mount)) { - var fsctx *kernel.FSContext - t.WithMuLocked(func(t *kernel.Task) { - fsctx = t.FSContext() - }) - if fsctx == nil { - // The task has been destroyed. Nothing to show here. - return - } - - // All mount points must be relative to the rootDir, and mounts outside - // will be excluded. - rootDir := fsctx.RootDirectory() - if rootDir == nil { - // The task has been destroyed. Nothing to show here. - return - } - defer rootDir.DecRef() - - mnt := t.MountNamespace().FindMount(rootDir) - if mnt == nil { - // Has it just been unmounted? - return - } - ms := t.MountNamespace().AllMountsUnder(mnt) - sort.Slice(ms, func(i, j int) bool { - return ms[i].ID < ms[j].ID - }) - for _, m := range ms { - mroot := m.Root() - if mroot == nil { - continue // No longer valid. - } - mountPath, desc := mroot.FullName(rootDir) - mroot.DecRef() - if !desc { - // MountSources that are not descendants of the chroot jail are ignored. - continue - } - fn(mountPath, m) - } -} - // mountInfoData is used to implement /proc/[pid]/mountinfo. // // +stateify savable @@ -692,92 +644,22 @@ var _ dynamicInode = (*mountInfoData)(nil) // Generate implements vfs.DynamicBytesSource.Generate. func (i *mountInfoData) Generate(ctx context.Context, buf *bytes.Buffer) error { - forEachMount(i.task, func(mountPath string, m *fs.Mount) { - mroot := m.Root() - if mroot == nil { - return // No longer valid. - } - defer mroot.DecRef() - - // Format: - // 36 35 98:0 /mnt1 /mnt2 rw,noatime master:1 - ext3 /dev/root rw,errors=continue - // (1)(2)(3) (4) (5) (6) (7) (8) (9) (10) (11) - - // (1) MountSource ID. - fmt.Fprintf(buf, "%d ", m.ID) - - // (2) Parent ID (or this ID if there is no parent). - pID := m.ID - if !m.IsRoot() && !m.IsUndo() { - pID = m.ParentID - } - fmt.Fprintf(buf, "%d ", pID) - - // (3) Major:Minor device ID. We don't have a superblock, so we - // just use the root inode device number. - sa := mroot.Inode.StableAttr - fmt.Fprintf(buf, "%d:%d ", sa.DeviceFileMajor, sa.DeviceFileMinor) - - // (4) Root: the pathname of the directory in the filesystem - // which forms the root of this mount. - // - // NOTE(b/78135857): This will always be "/" until we implement - // bind mounts. - fmt.Fprintf(buf, "/ ") - - // (5) Mount point (relative to process root). - fmt.Fprintf(buf, "%s ", mountPath) - - // (6) Mount options. - flags := mroot.Inode.MountSource.Flags - opts := "rw" - if flags.ReadOnly { - opts = "ro" - } - if flags.NoAtime { - opts += ",noatime" - } - if flags.NoExec { - opts += ",noexec" - } - fmt.Fprintf(buf, "%s ", opts) - - // (7) Optional fields: zero or more fields of the form "tag[:value]". - // (8) Separator: the end of the optional fields is marked by a single hyphen. - fmt.Fprintf(buf, "- ") - - // (9) Filesystem type. - fmt.Fprintf(buf, "%s ", mroot.Inode.MountSource.FilesystemType) - - // (10) Mount source: filesystem-specific information or "none". - fmt.Fprintf(buf, "none ") - - // (11) Superblock options, and final newline. - fmt.Fprintf(buf, "%s\n", superBlockOpts(mountPath, mroot.Inode.MountSource)) + var fsctx *kernel.FSContext + i.task.WithMuLocked(func(t *kernel.Task) { + fsctx = t.FSContext() }) - return nil -} - -func superBlockOpts(mountPath string, msrc *fs.MountSource) string { - // gVisor doesn't (yet) have a concept of super block options, so we - // use the ro/rw bit from the mount flag. - opts := "rw" - if msrc.Flags.ReadOnly { - opts = "ro" + if fsctx == nil { + // The task has been destroyed. Nothing to show here. + return nil } - - // NOTE(b/147673608): If the mount is a cgroup, we also need to include - // the cgroup name in the options. For now we just read that from the - // path. - // TODO(gvisor.dev/issues/190): Once gVisor has full cgroup support, we - // should get this value from the cgroup itself, and not rely on the - // path. - if msrc.FilesystemType == "cgroup" { - splitPath := strings.Split(mountPath, "/") - cgroupType := splitPath[len(splitPath)-1] - opts += "," + cgroupType + rootDir := fsctx.RootDirectoryVFS2() + if !rootDir.Ok() { + // Root has been destroyed. Don't try to read mounts. + return nil } - return opts + defer rootDir.DecRef() + i.task.Kernel().VFS().GenerateProcMountInfo(ctx, rootDir, buf) + return nil } // mountsData is used to implement /proc/[pid]/mounts. @@ -789,33 +671,24 @@ type mountsData struct { task *kernel.Task } -var _ dynamicInode = (*mountInfoData)(nil) +var _ dynamicInode = (*mountsData)(nil) // Generate implements vfs.DynamicBytesSource.Generate. func (i *mountsData) Generate(ctx context.Context, buf *bytes.Buffer) error { - forEachMount(i.task, func(mountPath string, m *fs.Mount) { - // Format: - // - // - // We use the filesystem name as the first field, since there - // is no real block device we can point to, and we also should - // not expose anything about the remote filesystem. - // - // Only ro/rw option is supported for now. - // - // The "needs dump"and fsck flags are always 0, which is allowed. - root := m.Root() - if root == nil { - return // No longer valid. - } - defer root.DecRef() - - flags := root.Inode.MountSource.Flags - opts := "rw" - if flags.ReadOnly { - opts = "ro" - } - fmt.Fprintf(buf, "%s %s %s %s %d %d\n", "none", mountPath, root.Inode.MountSource.FilesystemType, opts, 0, 0) + var fsctx *kernel.FSContext + i.task.WithMuLocked(func(t *kernel.Task) { + fsctx = t.FSContext() }) + if fsctx == nil { + // The task has been destroyed. Nothing to show here. + return nil + } + rootDir := fsctx.RootDirectoryVFS2() + if !rootDir.Ok() { + // Root has been destroyed. Don't try to read mounts. + return nil + } + defer rootDir.DecRef() + i.task.Kernel().VFS().GenerateProcMounts(ctx, rootDir, buf) return nil } diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index 4b68cabda..7792eb1a0 100644 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go @@ -15,7 +15,11 @@ package vfs import ( + "bytes" + "fmt" "math" + "sort" + "strings" "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" @@ -44,7 +48,7 @@ var lastMountID uint64 // // +stateify savable type Mount struct { - // vfs, fs, and root are immutable. References are held on fs and root. + // vfs, fs, root are immutable. References are held on fs and root. // // Invariant: root belongs to fs. vfs *VirtualFilesystem @@ -639,12 +643,28 @@ func (mnt *Mount) setReadOnlyLocked(ro bool) error { return nil } +func (mnt *Mount) readOnly() bool { + return atomic.LoadInt64(&mnt.writers) < 0 +} + // Filesystem returns the mounted Filesystem. It does not take a reference on // the returned Filesystem. func (mnt *Mount) Filesystem() *Filesystem { return mnt.fs } +// submountsLocked returns this Mount and all Mounts that are descendents of +// it. +// +// Precondition: mnt.vfs.mountMu must be held. +func (mnt *Mount) submountsLocked() []*Mount { + mounts := []*Mount{mnt} + for m := range mnt.children { + mounts = append(mounts, m.submountsLocked()...) + } + return mounts +} + // Root returns mntns' root. A reference is taken on the returned // VirtualDentry. func (mntns *MountNamespace) Root() VirtualDentry { @@ -655,3 +675,173 @@ func (mntns *MountNamespace) Root() VirtualDentry { vd.IncRef() return vd } + +// GenerateProcMounts emits the contents of /proc/[pid]/mounts for vfs to buf. +// +// Preconditions: taskRootDir.Ok(). +func (vfs *VirtualFilesystem) GenerateProcMounts(ctx context.Context, taskRootDir VirtualDentry, buf *bytes.Buffer) { + vfs.mountMu.Lock() + defer vfs.mountMu.Unlock() + rootMnt := taskRootDir.mount + mounts := rootMnt.submountsLocked() + sort.Slice(mounts, func(i, j int) bool { return mounts[i].ID < mounts[j].ID }) + for _, mnt := range mounts { + // Get the path to this mount relative to task root. + mntRootVD := VirtualDentry{ + mount: mnt, + dentry: mnt.root, + } + path, err := vfs.PathnameReachable(ctx, taskRootDir, mntRootVD) + if err != nil { + // For some reason we didn't get a path. Log a warning + // and run with empty path. + ctx.Warningf("Error getting pathname for mount root %+v: %v", mnt.root, err) + path = "" + } + if path == "" { + // Either an error occurred, or path is not reachable + // from root. + break + } + + opts := "rw" + if mnt.readOnly() { + opts = "ro" + } + if mnt.flags.NoExec { + opts += ",noexec" + } + + // Format: + // + // + // The "needs dump" and "fsck order" flags are always 0, which + // is allowed. + fmt.Fprintf(buf, "%s %s %s %s %d %d\n", "none", path, mnt.fs.FilesystemType().Name(), opts, 0, 0) + } +} + +// GenerateProcMountInfo emits the contents of /proc/[pid]/mountinfo for vfs to +// buf. +// +// Preconditions: taskRootDir.Ok(). +func (vfs *VirtualFilesystem) GenerateProcMountInfo(ctx context.Context, taskRootDir VirtualDentry, buf *bytes.Buffer) { + vfs.mountMu.Lock() + defer vfs.mountMu.Unlock() + rootMnt := taskRootDir.mount + mounts := rootMnt.submountsLocked() + sort.Slice(mounts, func(i, j int) bool { return mounts[i].ID < mounts[j].ID }) + for _, mnt := range mounts { + // Get the path to this mount relative to task root. + mntRootVD := VirtualDentry{ + mount: mnt, + dentry: mnt.root, + } + path, err := vfs.PathnameReachable(ctx, taskRootDir, mntRootVD) + if err != nil { + // For some reason we didn't get a path. Log a warning + // and run with empty path. + ctx.Warningf("Error getting pathname for mount root %+v: %v", mnt.root, err) + path = "" + } + if path == "" { + // Either an error occurred, or path is not reachable + // from root. + break + } + // Stat the mount root to get the major/minor device numbers. + pop := &PathOperation{ + Root: mntRootVD, + Start: mntRootVD, + } + statx, err := vfs.StatAt(ctx, auth.NewAnonymousCredentials(), pop, &StatOptions{}) + if err != nil { + // Well that's not good. Ignore this mount. + break + } + + // Format: + // 36 35 98:0 /mnt1 /mnt2 rw,noatime master:1 - ext3 /dev/root rw,errors=continue + // (1)(2)(3) (4) (5) (6) (7) (8) (9) (10) (11) + + // (1) Mount ID. + fmt.Fprintf(buf, "%d ", mnt.ID) + + // (2) Parent ID (or this ID if there is no parent). + pID := mnt.ID + if p := mnt.parent(); p != nil { + pID = p.ID + } + fmt.Fprintf(buf, "%d ", pID) + + // (3) Major:Minor device ID. We don't have a superblock, so we + // just use the root inode device number. + fmt.Fprintf(buf, "%d:%d ", statx.DevMajor, statx.DevMinor) + + // (4) Root: the pathname of the directory in the filesystem + // which forms the root of this mount. + // + // NOTE(b/78135857): This will always be "/" until we implement + // bind mounts. + fmt.Fprintf(buf, "/ ") + + // (5) Mount point (relative to process root). + fmt.Fprintf(buf, "%s ", manglePath(path)) + + // (6) Mount options. + opts := "rw" + if mnt.readOnly() { + opts = "ro" + } + if mnt.flags.NoExec { + opts += ",noexec" + } + // TODO(gvisor.dev/issue/1193): Add "noatime" if MS_NOATIME is + // set. + fmt.Fprintf(buf, "%s ", opts) + + // (7) Optional fields: zero or more fields of the form "tag[:value]". + // (8) Separator: the end of the optional fields is marked by a single hyphen. + fmt.Fprintf(buf, "- ") + + // (9) Filesystem type. + fmt.Fprintf(buf, "%s ", mnt.fs.FilesystemType().Name()) + + // (10) Mount source: filesystem-specific information or "none". + fmt.Fprintf(buf, "none ") + + // (11) Superblock options, and final newline. + fmt.Fprintf(buf, "%s\n", superBlockOpts(path, mnt)) + } +} + +// manglePath replaces ' ', '\t', '\n', and '\\' with their octal equivalents. +// See Linux fs/seq_file.c:mangle_path. +func manglePath(p string) string { + r := strings.NewReplacer(" ", "\\040", "\t", "\\011", "\n", "\\012", "\\", "\\134") + return r.Replace(p) +} + +// superBlockOpts returns the super block options string for the the mount at +// the given path. +func superBlockOpts(mountPath string, mnt *Mount) string { + // gVisor doesn't (yet) have a concept of super block options, so we + // use the ro/rw bit from the mount flag. + opts := "rw" + if mnt.readOnly() { + opts = "ro" + } + + // NOTE(b/147673608): If the mount is a cgroup, we also need to include + // the cgroup name in the options. For now we just read that from the + // path. + // TODO(gvisor.dev/issues/190): Once gVisor has full cgroup support, we + // should get this value from the cgroup itself, and not rely on the + // path. + if mnt.fs.FilesystemType().Name() == "cgroup" { + splitPath := strings.Split(mountPath, "/") + cgroupType := splitPath[len(splitPath)-1] + opts += "," + cgroupType + } + return opts +} -- cgit v1.2.3 From b6639f77e59d885cb092c15a7a0c5a988e149b40 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Tue, 31 Mar 2020 15:00:30 -0700 Subject: Include original copyUp error in panic if cleanupUpper fails. When copyUp fails, we attempt to clean up the upper filesystem by removing any files that have already been copied-up. If the cleanup fails, we panic because the "overlay filesystem is in an inconsistent state". This CL adds the original copy-up error to the panic information, to hopefully make it easier to track down how the overlay filesystem got into the inconsistent state. PiperOrigin-RevId: 304053370 --- pkg/sentry/fs/copy_up.go | 28 +++++++++++++++------------- pkg/sentry/fs/inode_overlay.go | 3 ++- 2 files changed, 17 insertions(+), 14 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/copy_up.go b/pkg/sentry/fs/copy_up.go index b060a12ff..ab1424c95 100644 --- a/pkg/sentry/fs/copy_up.go +++ b/pkg/sentry/fs/copy_up.go @@ -222,8 +222,8 @@ func copyUpLocked(ctx context.Context, parent *Dirent, next *Dirent) error { } childUpper, err := parentUpper.Lookup(ctx, next.name) if err != nil { - log.Warningf("copy up failed to lookup directory: %v", err) - cleanupUpper(ctx, parentUpper, next.name) + werr := fmt.Errorf("copy up failed to lookup directory: %v", err) + cleanupUpper(ctx, parentUpper, next.name, werr) return syserror.EIO } defer childUpper.DecRef() @@ -242,8 +242,8 @@ func copyUpLocked(ctx context.Context, parent *Dirent, next *Dirent) error { } childUpper, err := parentUpper.Lookup(ctx, next.name) if err != nil { - log.Warningf("copy up failed to lookup symlink: %v", err) - cleanupUpper(ctx, parentUpper, next.name) + werr := fmt.Errorf("copy up failed to lookup symlink: %v", err) + cleanupUpper(ctx, parentUpper, next.name, werr) return syserror.EIO } defer childUpper.DecRef() @@ -256,23 +256,23 @@ func copyUpLocked(ctx context.Context, parent *Dirent, next *Dirent) error { // Bring file attributes up to date. This does not include size, which will be // brought up to date with copyContentsLocked. if err := copyAttributesLocked(ctx, childUpperInode, next.Inode.overlay.lower); err != nil { - log.Warningf("copy up failed to copy up attributes: %v", err) - cleanupUpper(ctx, parentUpper, next.name) + werr := fmt.Errorf("copy up failed to copy up attributes: %v", err) + cleanupUpper(ctx, parentUpper, next.name, werr) return syserror.EIO } // Copy the entire file. if err := copyContentsLocked(ctx, childUpperInode, next.Inode.overlay.lower, attrs.Size); err != nil { - log.Warningf("copy up failed to copy up contents: %v", err) - cleanupUpper(ctx, parentUpper, next.name) + werr := fmt.Errorf("copy up failed to copy up contents: %v", err) + cleanupUpper(ctx, parentUpper, next.name, werr) return syserror.EIO } lowerMappable := next.Inode.overlay.lower.Mappable() upperMappable := childUpperInode.Mappable() if lowerMappable != nil && upperMappable == nil { - log.Warningf("copy up failed: cannot ensure memory mapping coherence") - cleanupUpper(ctx, parentUpper, next.name) + werr := fmt.Errorf("copy up failed: cannot ensure memory mapping coherence") + cleanupUpper(ctx, parentUpper, next.name, werr) return syserror.EIO } @@ -324,12 +324,14 @@ func copyUpLocked(ctx context.Context, parent *Dirent, next *Dirent) error { return nil } -// cleanupUpper removes name from parent, and panics if it is unsuccessful. -func cleanupUpper(ctx context.Context, parent *Inode, name string) { +// cleanupUpper is called when copy-up fails. It logs the copy-up error and +// attempts to remove name from parent. If that fails, then it panics. +func cleanupUpper(ctx context.Context, parent *Inode, name string, copyUpErr error) { + log.Warningf(copyUpErr.Error()) if err := parent.InodeOperations.Remove(ctx, parent, name); err != nil { // Unfortunately we don't have much choice. We shouldn't // willingly give the caller access to a nonsense filesystem. - panic(fmt.Sprintf("overlay filesystem is in an inconsistent state: failed to remove %q from upper filesystem: %v", name, err)) + panic(fmt.Sprintf("overlay filesystem is in an inconsistent state: copyUp got error: %v; then cleanup failed to remove %q from upper filesystem: %v.", copyUpErr, name, err)) } } diff --git a/pkg/sentry/fs/inode_overlay.go b/pkg/sentry/fs/inode_overlay.go index 5ada33a32..537c8d257 100644 --- a/pkg/sentry/fs/inode_overlay.go +++ b/pkg/sentry/fs/inode_overlay.go @@ -231,7 +231,8 @@ func overlayCreate(ctx context.Context, o *overlayEntry, parent *Dirent, name st upperFile.Dirent.Inode.IncRef() entry, err := newOverlayEntry(ctx, upperFile.Dirent.Inode, nil, false) if err != nil { - cleanupUpper(ctx, o.upper, name) + werr := fmt.Errorf("newOverlayEntry failed: %v", err) + cleanupUpper(ctx, o.upper, name, werr) return nil, err } -- cgit v1.2.3 From 639d94f9f71b43e86320a6e9157c932f5d7936a7 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 31 Mar 2020 19:15:55 -0700 Subject: Add socket filesystem and global disconnected socket mount for VFS2. A socket mount where anonymous sockets will reside is added to the VirtualFilesystem. Socketfs is built on top of kernfs. Updates #1476, #1478, #1484, #1485. PiperOrigin-RevId: 304095251 --- pkg/sentry/fsimpl/sockfs/BUILD | 16 +++++++++ pkg/sentry/fsimpl/sockfs/sockfs.go | 73 ++++++++++++++++++++++++++++++++++++++ pkg/sentry/kernel/BUILD | 1 + pkg/sentry/kernel/kernel.go | 24 +++++++++++++ test/syscalls/linux/socket_unix.cc | 2 ++ 5 files changed, 116 insertions(+) create mode 100644 pkg/sentry/fsimpl/sockfs/BUILD create mode 100644 pkg/sentry/fsimpl/sockfs/sockfs.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/sockfs/BUILD b/pkg/sentry/fsimpl/sockfs/BUILD new file mode 100644 index 000000000..790d50e65 --- /dev/null +++ b/pkg/sentry/fsimpl/sockfs/BUILD @@ -0,0 +1,16 @@ +load("//tools:defs.bzl", "go_library") + +licenses(["notice"]) + +go_library( + name = "sockfs", + srcs = ["sockfs.go"], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/context", + "//pkg/sentry/fsimpl/kernfs", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/vfs", + "//pkg/syserror", + ], +) diff --git a/pkg/sentry/fsimpl/sockfs/sockfs.go b/pkg/sentry/fsimpl/sockfs/sockfs.go new file mode 100644 index 000000000..c13511de2 --- /dev/null +++ b/pkg/sentry/fsimpl/sockfs/sockfs.go @@ -0,0 +1,73 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package sockfs provides a filesystem implementation for anonymous sockets. +package sockfs + +import ( + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// NewFilesystem creates a new sockfs filesystem. +// +// Note that there should only ever be one instance of sockfs.Filesystem, +// backing a global socket mount. +func NewFilesystem(vfsObj *vfs.VirtualFilesystem) *vfs.Filesystem { + fs, _, err := filesystemType{}.GetFilesystem(nil, vfsObj, nil, "", vfs.GetFilesystemOptions{}) + if err != nil { + panic("failed to create sockfs filesystem") + } + return fs +} + +// filesystemType implements vfs.FilesystemType. +type filesystemType struct{} + +// GetFilesystem implements FilesystemType.GetFilesystem. +func (fsType filesystemType) GetFilesystem(_ context.Context, vfsObj *vfs.VirtualFilesystem, _ *auth.Credentials, _ string, _ vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { + fs := &filesystem{} + fs.Init(vfsObj, fsType) + return fs.VFSFilesystem(), nil, nil +} + +// Name implements FilesystemType.Name. +// +// Note that registering sockfs is unnecessary, except for the fact that it +// will not show up under /proc/filesystems as a result. This is a very minor +// discrepancy from Linux. +func (filesystemType) Name() string { + return "sockfs" +} + +// filesystem implements vfs.FilesystemImpl. +type filesystem struct { + kernfs.Filesystem +} + +// inode implements kernfs.Inode. +type inode struct { + kernfs.InodeNotDirectory + kernfs.InodeNotSymlink + kernfs.InodeAttrs + kernfs.InodeNoopRefCount +} + +// Open implements kernfs.Inode.Open. +func (i *inode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + return nil, syserror.ENXIO +} diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD index beba29a09..bb7e3cbc3 100644 --- a/pkg/sentry/kernel/BUILD +++ b/pkg/sentry/kernel/BUILD @@ -169,6 +169,7 @@ go_library( "//pkg/sentry/fs/lock", "//pkg/sentry/fs/timerfd", "//pkg/sentry/fsbridge", + "//pkg/sentry/fsimpl/sockfs", "//pkg/sentry/hostcpu", "//pkg/sentry/inet", "//pkg/sentry/kernel/auth", diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 6feda8fa1..0a448b57c 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -50,6 +50,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/timerfd" "gvisor.dev/gvisor/pkg/sentry/fsbridge" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" "gvisor.dev/gvisor/pkg/sentry/hostcpu" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" @@ -225,6 +226,11 @@ type Kernel struct { // by extMu. nextSocketEntry uint64 + // socketMount is a disconnected vfs.Mount, not included in k.vfs, + // representing a sockfs.filesystem. socketMount is used to back + // VirtualDentries representing anonymous sockets. + socketMount *vfs.Mount + // deviceRegistry is used to save/restore device.SimpleDevices. deviceRegistry struct{} `state:".(*device.Registry)"` @@ -348,6 +354,19 @@ func (k *Kernel) Init(args InitKernelArgs) error { k.monotonicClock = &timekeeperClock{tk: args.Timekeeper, c: sentrytime.Monotonic} k.futexes = futex.NewManager() k.netlinkPorts = port.New() + if VFS2Enabled { + if err := k.vfs.Init(); err != nil { + return fmt.Errorf("failed to initialize VFS: %v", err) + } + fs := sockfs.NewFilesystem(&k.vfs) + // NewDisconnectedMount will take an additional reference on fs. + defer fs.DecRef() + sm, err := k.vfs.NewDisconnectedMount(fs, nil, &vfs.MountOptions{}) + if err != nil { + return fmt.Errorf("failed to initialize socket mount: %v", err) + } + k.socketMount = sm + } return nil } @@ -1452,6 +1471,11 @@ func (k *Kernel) ListSockets() []*SocketEntry { return socks } +// SocketMount returns the global socket mount. +func (k *Kernel) SocketMount() *vfs.Mount { + return k.socketMount +} + // supervisorContext is a privileged context. type supervisorContext struct { context.NoopSleeper diff --git a/test/syscalls/linux/socket_unix.cc b/test/syscalls/linux/socket_unix.cc index 4cf1f76f1..8bf663e8b 100644 --- a/test/syscalls/linux/socket_unix.cc +++ b/test/syscalls/linux/socket_unix.cc @@ -257,6 +257,8 @@ TEST_P(UnixSocketPairTest, ShutdownWrite) { TEST_P(UnixSocketPairTest, SocketReopenFromProcfs) { // TODO(b/122310852): We should be returning ENXIO and NOT EIO. + // TODO(github.dev/issue/1624): This should be resolved in VFS2. Verify + // that this is the case and delete the SKIP_IF once we delete VFS1. SKIP_IF(IsRunningOnGvisor()); auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); -- cgit v1.2.3 From 5eb41c8fbabac090251fbfb43bd9c814124aa575 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Tue, 11 Feb 2020 02:55:51 -0500 Subject: Arm64 signal#2: signal support in arch module SA_RESTORER is always used on Intel platform. But this flag is optional on other platforms. The vdso is enabled, so we can use the sigreturn trampolines the vdso provides instead on Arm platform. Signed-off-by: Bin Lu --- pkg/sentry/arch/signal_arm64.go | 30 ++++++++++++++++++++++++------ pkg/sentry/kernel/task_signals.go | 13 +++++++++++++ 2 files changed, 37 insertions(+), 6 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/arch/signal_arm64.go b/pkg/sentry/arch/signal_arm64.go index 4f4cc46a8..b57d6a17d 100644 --- a/pkg/sentry/arch/signal_arm64.go +++ b/pkg/sentry/arch/signal_arm64.go @@ -83,9 +83,12 @@ func (c *context64) SignalSetup(st *Stack, act *SignalAct, info *SignalInfo, alt if ucSize < 0 { panic("can't get size of UContext64") } - // st.Arch.Width() is for the restorer address. sizeof(siginfo) == 128. - frameSize := int(st.Arch.Width()) + ucSize + 128 - frameBottom := (sp-usermem.Addr(frameSize)) & ^usermem.Addr(15) - 8 + + // frameSize = ucSize + sizeof(siginfo). + // sizeof(siginfo) == 128. + // R30 stores the restorer address. + frameSize := ucSize + 128 + frameBottom := (sp - usermem.Addr(frameSize)) & ^usermem.Addr(15) sp = frameBottom + usermem.Addr(frameSize) st.Bottom = sp @@ -115,12 +118,27 @@ func (c *context64) SignalSetup(st *Stack, act *SignalAct, info *SignalInfo, alt c.Regs.Regs[0] = uint64(info.Signo) c.Regs.Regs[1] = uint64(infoAddr) c.Regs.Regs[2] = uint64(ucAddr) - + c.Regs.Regs[30] = uint64(act.Restorer) return nil } // SignalRestore implements Context.SignalRestore. -// Only used on intel. func (c *context64) SignalRestore(st *Stack, rt bool) (linux.SignalSet, SignalStack, error) { - return 0, SignalStack{}, nil + // Copy out the stack frame. + var uc UContext64 + if _, err := st.Pop(&uc); err != nil { + return 0, SignalStack{}, err + } + var info SignalInfo + if _, err := st.Pop(&info); err != nil { + return 0, SignalStack{}, err + } + + // Restore registers. + c.Regs.Regs = uc.MContext.Regs + c.Regs.Pc = uc.MContext.Pc + c.Regs.Sp = uc.MContext.Sp + c.Regs.Pstate = uc.MContext.Pstate + + return uc.Sigset, uc.Stack, nil } diff --git a/pkg/sentry/kernel/task_signals.go b/pkg/sentry/kernel/task_signals.go index 8802db142..0e74236c9 100644 --- a/pkg/sentry/kernel/task_signals.go +++ b/pkg/sentry/kernel/task_signals.go @@ -263,6 +263,19 @@ func (t *Task) deliverSignalToHandler(info *arch.SignalInfo, act arch.SignalAct) if t.haveSavedSignalMask { mask = t.savedSignalMask } + + // Set up the restorer. + // x86-64 should always uses SA_RESTORER, but this flag is optional on other platforms. + // Please see the linux code as reference: + // linux/arch/x86/kernel/signal.c:__setup_rt_frame() + // If SA_RESTORER is not configured, we can use the sigreturn trampolines + // the vdso provides instead. + // Please see the linux code as reference: + // linux/arch/arm64/kernel/signal.c:setup_return() + if act.Flags&linux.SA_RESTORER == 0 { + act.Restorer = t.MemoryManager().VDSOSigReturn() + } + if err := t.Arch().SignalSetup(st, &act, info, &alt, mask); err != nil { return err } -- cgit v1.2.3 From 840980aeba0b5224b13bcaadf5785ac5305a5230 Mon Sep 17 00:00:00 2001 From: Rahat Mahmood Date: Tue, 31 Mar 2020 22:54:50 -0700 Subject: Implement automated marshalling for slices of Marshallable types. PiperOrigin-RevId: 304119255 --- pkg/sentry/kernel/rseq.go | 2 +- pkg/sentry/syscalls/linux/sys_stat.go | 6 +- pkg/sentry/syscalls/linux/vfs2/epoll.go | 4 +- pkg/sentry/syscalls/linux/vfs2/poll.go | 14 +- pkg/sentry/syscalls/linux/vfs2/setstat.go | 2 +- pkg/sentry/syscalls/linux/vfs2/stat.go | 23 +- tools/go_marshal/analysis/analysis_unsafe.go | 4 + tools/go_marshal/defs.bzl | 3 +- tools/go_marshal/gomarshal/generator.go | 130 ++++-- tools/go_marshal/gomarshal/generator_interfaces.go | 62 +++ .../generator_interfaces_array_newtype.go | 84 +--- .../generator_interfaces_primitive_newtype.go | 173 ++++--- .../gomarshal/generator_interfaces_struct.go | 308 +++++++++--- tools/go_marshal/gomarshal/generator_tests.go | 52 ++- tools/go_marshal/gomarshal/util.go | 21 +- tools/go_marshal/marshal/marshal.go | 103 ++++- tools/go_marshal/primitive/BUILD | 18 + tools/go_marshal/primitive/primitive.go | 175 +++++++ tools/go_marshal/test/BUILD | 14 + tools/go_marshal/test/benchmark_test.go | 42 ++ tools/go_marshal/test/external/external.go | 8 + tools/go_marshal/test/marshal_test.go | 515 +++++++++++++++++++++ tools/go_marshal/test/test.go | 36 +- 23 files changed, 1525 insertions(+), 274 deletions(-) create mode 100644 tools/go_marshal/primitive/BUILD create mode 100644 tools/go_marshal/primitive/primitive.go create mode 100644 tools/go_marshal/test/marshal_test.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/rseq.go b/pkg/sentry/kernel/rseq.go index ded95f532..18416643b 100644 --- a/pkg/sentry/kernel/rseq.go +++ b/pkg/sentry/kernel/rseq.go @@ -304,7 +304,7 @@ func (t *Task) rseqAddrInterrupt() { } var cs linux.RSeqCriticalSection - if err := cs.CopyIn(t, critAddr); err != nil { + if _, err := cs.CopyIn(t, critAddr); err != nil { t.Debugf("Failed to copy critical section from %#x for rseq: %v", critAddr, err) t.forceSignal(linux.SIGSEGV, false /* unconditional */) t.SendSignal(SignalInfoPriv(linux.SIGSEGV)) diff --git a/pkg/sentry/syscalls/linux/sys_stat.go b/pkg/sentry/syscalls/linux/sys_stat.go index a11a87cd1..46ebf27a2 100644 --- a/pkg/sentry/syscalls/linux/sys_stat.go +++ b/pkg/sentry/syscalls/linux/sys_stat.go @@ -115,7 +115,8 @@ func stat(t *kernel.Task, d *fs.Dirent, dirPath bool, statAddr usermem.Addr) err return err } s := statFromAttrs(t, d.Inode.StableAttr, uattr) - return s.CopyOut(t, statAddr) + _, err = s.CopyOut(t, statAddr) + return err } // fstat implements fstat for the given *fs.File. @@ -125,7 +126,8 @@ func fstat(t *kernel.Task, f *fs.File, statAddr usermem.Addr) error { return err } s := statFromAttrs(t, f.Dirent.Inode.StableAttr, uattr) - return s.CopyOut(t, statAddr) + _, err = s.CopyOut(t, statAddr) + return err } // Statx implements linux syscall statx(2). diff --git a/pkg/sentry/syscalls/linux/vfs2/epoll.go b/pkg/sentry/syscalls/linux/vfs2/epoll.go index d6cb0e79a..5a938cee2 100644 --- a/pkg/sentry/syscalls/linux/vfs2/epoll.go +++ b/pkg/sentry/syscalls/linux/vfs2/epoll.go @@ -101,14 +101,14 @@ func EpollCtl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc var event linux.EpollEvent switch op { case linux.EPOLL_CTL_ADD: - if err := event.CopyIn(t, eventAddr); err != nil { + if _, err := event.CopyIn(t, eventAddr); err != nil { return 0, nil, err } return 0, nil, ep.AddInterest(file, fd, event) case linux.EPOLL_CTL_DEL: return 0, nil, ep.DeleteInterest(file, fd) case linux.EPOLL_CTL_MOD: - if err := event.CopyIn(t, eventAddr); err != nil { + if _, err := event.CopyIn(t, eventAddr); err != nil { return 0, nil, err } return 0, nil, ep.ModifyInterest(file, fd, event) diff --git a/pkg/sentry/syscalls/linux/vfs2/poll.go b/pkg/sentry/syscalls/linux/vfs2/poll.go index dbf4882da..ff1b25d7b 100644 --- a/pkg/sentry/syscalls/linux/vfs2/poll.go +++ b/pkg/sentry/syscalls/linux/vfs2/poll.go @@ -374,7 +374,8 @@ func copyOutTimespecRemaining(t *kernel.Task, startNs ktime.Time, timeout time.D } remaining := timeoutRemaining(t, startNs, timeout) tsRemaining := linux.NsecToTimespec(remaining.Nanoseconds()) - return tsRemaining.CopyOut(t, timespecAddr) + _, err := tsRemaining.CopyOut(t, timespecAddr) + return err } // copyOutTimevalRemaining copies the time remaining in timeout to timevalAddr. @@ -386,7 +387,8 @@ func copyOutTimevalRemaining(t *kernel.Task, startNs ktime.Time, timeout time.Du } remaining := timeoutRemaining(t, startNs, timeout) tvRemaining := linux.NsecToTimeval(remaining.Nanoseconds()) - return tvRemaining.CopyOut(t, timevalAddr) + _, err := tvRemaining.CopyOut(t, timevalAddr) + return err } // pollRestartBlock encapsulates the state required to restart poll(2) via @@ -477,7 +479,7 @@ func Select(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal timeout := time.Duration(-1) if timevalAddr != 0 { var timeval linux.Timeval - if err := timeval.CopyIn(t, timevalAddr); err != nil { + if _, err := timeval.CopyIn(t, timevalAddr); err != nil { return 0, nil, err } if timeval.Sec < 0 || timeval.Usec < 0 { @@ -519,7 +521,7 @@ func Pselect(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca panic(fmt.Sprintf("unsupported sizeof(void*): %d", t.Arch().Width())) } var maskStruct sigSetWithSize - if err := maskStruct.CopyIn(t, maskWithSizeAddr); err != nil { + if _, err := maskStruct.CopyIn(t, maskWithSizeAddr); err != nil { return 0, nil, err } if err := setTempSignalSet(t, usermem.Addr(maskStruct.sigsetAddr), uint(maskStruct.sizeofSigset)); err != nil { @@ -554,7 +556,7 @@ func copyTimespecInToDuration(t *kernel.Task, timespecAddr usermem.Addr) (time.D timeout := time.Duration(-1) if timespecAddr != 0 { var timespec linux.Timespec - if err := timespec.CopyIn(t, timespecAddr); err != nil { + if _, err := timespec.CopyIn(t, timespecAddr); err != nil { return 0, err } if !timespec.Valid() { @@ -573,7 +575,7 @@ func setTempSignalSet(t *kernel.Task, maskAddr usermem.Addr, maskSize uint) erro return syserror.EINVAL } var mask linux.SignalSet - if err := mask.CopyIn(t, maskAddr); err != nil { + if _, err := mask.CopyIn(t, maskAddr); err != nil { return err } mask &^= kernel.UnblockableSignals diff --git a/pkg/sentry/syscalls/linux/vfs2/setstat.go b/pkg/sentry/syscalls/linux/vfs2/setstat.go index 136453ccc..4e61f1452 100644 --- a/pkg/sentry/syscalls/linux/vfs2/setstat.go +++ b/pkg/sentry/syscalls/linux/vfs2/setstat.go @@ -226,7 +226,7 @@ func Utime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall opts.Stat.Mtime.Nsec = linux.UTIME_NOW } else { var times linux.Utime - if err := times.CopyIn(t, timesAddr); err != nil { + if _, err := times.CopyIn(t, timesAddr); err != nil { return 0, nil, err } opts.Stat.Atime.Sec = times.Actime diff --git a/pkg/sentry/syscalls/linux/vfs2/stat.go b/pkg/sentry/syscalls/linux/vfs2/stat.go index fdfe49243..bb1d5cac4 100644 --- a/pkg/sentry/syscalls/linux/vfs2/stat.go +++ b/pkg/sentry/syscalls/linux/vfs2/stat.go @@ -91,7 +91,8 @@ func fstatat(t *kernel.Task, dirfd int32, pathAddr, statAddr usermem.Addr, flags } var stat linux.Stat convertStatxToUserStat(t, &statx, &stat) - return stat.CopyOut(t, statAddr) + _, err = stat.CopyOut(t, statAddr) + return err } start = dirfile.VirtualDentry() start.IncRef() @@ -111,7 +112,8 @@ func fstatat(t *kernel.Task, dirfd int32, pathAddr, statAddr usermem.Addr, flags } var stat linux.Stat convertStatxToUserStat(t, &statx, &stat) - return stat.CopyOut(t, statAddr) + _, err = stat.CopyOut(t, statAddr) + return err } func timespecFromStatxTimestamp(sxts linux.StatxTimestamp) linux.Timespec { @@ -140,7 +142,8 @@ func Fstat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } var stat linux.Stat convertStatxToUserStat(t, &statx, &stat) - return 0, nil, stat.CopyOut(t, statAddr) + _, err = stat.CopyOut(t, statAddr) + return 0, nil, err } // Statx implements Linux syscall statx(2). @@ -199,7 +202,8 @@ func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return 0, nil, err } userifyStatx(t, &statx) - return 0, nil, statx.CopyOut(t, statxAddr) + _, err = statx.CopyOut(t, statxAddr) + return 0, nil, err } start = dirfile.VirtualDentry() start.IncRef() @@ -218,7 +222,8 @@ func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return 0, nil, err } userifyStatx(t, &statx) - return 0, nil, statx.CopyOut(t, statxAddr) + _, err = statx.CopyOut(t, statxAddr) + return 0, nil, err } func userifyStatx(t *kernel.Task, statx *linux.Statx) { @@ -359,8 +364,8 @@ func Statfs(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, err } - - return 0, nil, statfs.CopyOut(t, bufAddr) + _, err = statfs.CopyOut(t, bufAddr) + return 0, nil, err } // Fstatfs implements Linux syscall fstatfs(2). @@ -378,6 +383,6 @@ func Fstatfs(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if err != nil { return 0, nil, err } - - return 0, nil, statfs.CopyOut(t, bufAddr) + _, err = statfs.CopyOut(t, bufAddr) + return 0, nil, err } diff --git a/tools/go_marshal/analysis/analysis_unsafe.go b/tools/go_marshal/analysis/analysis_unsafe.go index 9a9a4f298..cd55cf5cb 100644 --- a/tools/go_marshal/analysis/analysis_unsafe.go +++ b/tools/go_marshal/analysis/analysis_unsafe.go @@ -161,6 +161,10 @@ func AlignmentCheck(t *testing.T, typ reflect.Type) (ok bool, delta uint64) { if typ.NumField() > 0 && nextXOff != int(typ.Size()) { implicitPad := int(typ.Size()) - nextXOff f := typ.Field(typ.NumField() - 1) // Final field + if tag, ok := f.Tag.Lookup("marshal"); ok && tag == "unaligned" { + // Final field explicitly marked unaligned. + break + } t.Fatalf("Suspect offset for field %s.%s at the end of %s, detected an implicit %d byte padding from offset %d to %d at the end of the struct; either add %d bytes of explict padding at end of the struct or tag the final field %s as `marshal:\"unaligned\"`.", typ.Name(), f.Name, typ.Name(), implicitPad, nextXOff, typ.Size(), implicitPad, f.Name) } diff --git a/tools/go_marshal/defs.bzl b/tools/go_marshal/defs.bzl index d79786a68..323e33882 100644 --- a/tools/go_marshal/defs.bzl +++ b/tools/go_marshal/defs.bzl @@ -53,9 +53,10 @@ go_marshal = rule( # marshal_deps are the dependencies requied by generated code. marshal_deps = [ - "//tools/go_marshal/marshal", + "//pkg/gohacks", "//pkg/safecopy", "//pkg/usermem", + "//tools/go_marshal/marshal", ] # marshal_test_deps are required by test targets. diff --git a/tools/go_marshal/gomarshal/generator.go b/tools/go_marshal/gomarshal/generator.go index 82983804c..935a36b25 100644 --- a/tools/go_marshal/gomarshal/generator.go +++ b/tools/go_marshal/gomarshal/generator.go @@ -28,12 +28,6 @@ import ( "gvisor.dev/gvisor/tools/tags" ) -const ( - marshalImport = "gvisor.dev/gvisor/tools/go_marshal/marshal" - safecopyImport = "gvisor.dev/gvisor/pkg/safecopy" - usermemImport = "gvisor.dev/gvisor/pkg/usermem" -) - // List of identifiers we use in generated code that may conflict with a // similarly-named source identifier. Abort gracefully when we see these to // avoid potentially confusing compilation failures in generated code. @@ -44,8 +38,8 @@ const ( // All recievers are single letters, so we don't allow import aliases to be a // single letter. var badIdents = []string{ - "addr", "blk", "buf", "dst", "dsts", "err", "hdr", "idx", "inner", "len", - "ptr", "src", "srcs", "task", "val", + "addr", "blk", "buf", "dst", "dsts", "count", "err", "hdr", "idx", "inner", + "length", "limit", "ptr", "size", "src", "srcs", "task", "val", // All single-letter identifiers. } @@ -110,9 +104,10 @@ func NewGenerator(srcs []string, out, outTest, pkg string, imports []string) (*G g.imports.add("reflect") g.imports.add("runtime") g.imports.add("unsafe") - g.imports.add(marshalImport) - g.imports.add(safecopyImport) - g.imports.add(usermemImport) + g.imports.add("gvisor.dev/gvisor/pkg/gohacks") + g.imports.add("gvisor.dev/gvisor/pkg/safecopy") + g.imports.add("gvisor.dev/gvisor/pkg/usermem") + g.imports.add("gvisor.dev/gvisor/tools/go_marshal/marshal") return &g, nil } @@ -194,10 +189,73 @@ func (g *Generator) parse() ([]*ast.File, []*token.FileSet, error) { return files, fsets, nil } +// sliceAPI carries information about the '+marshal slice' directive. +type sliceAPI struct { + // Comment node in the AST containing the +marshal tag. + comment *ast.Comment + // Identifier fragment to use when naming generated functions for the slice + // API. + ident string + // Whether the generated functions should reference the newtype name, or the + // inner type name. Only meaningful on newtype declarations on primitives. + inner bool +} + +// marshallableType carries information about a type marked with the '+marshal' +// directive. +type marshallableType struct { + spec *ast.TypeSpec + slice *sliceAPI +} + +func newMarshallableType(fset *token.FileSet, tagLine *ast.Comment, spec *ast.TypeSpec) marshallableType { + mt := marshallableType{ + spec: spec, + slice: nil, + } + + var unhandledTags []string + + for _, tag := range strings.Fields(strings.TrimPrefix(tagLine.Text, "// +marshal")) { + if strings.HasPrefix(tag, "slice:") { + tokens := strings.Split(tag, ":") + if len(tokens) < 2 || len(tokens) > 3 { + abortAt(fset.Position(tagLine.Slash), fmt.Sprintf("+marshal directive has invalid 'slice' clause. Expecting format 'slice:[:inner]', got '%v'", tag)) + } + if len(tokens[1]) == 0 { + abortAt(fset.Position(tagLine.Slash), "+marshal slice directive has empty identifier argument. Expecting '+marshal slice:identifier'") + } + + sa := &sliceAPI{ + comment: tagLine, + ident: tokens[1], + } + mt.slice = sa + + if len(tokens) == 3 { + if tokens[2] != "inner" { + abortAt(fset.Position(tagLine.Slash), "+marshal slice directive has an invalid argument. Expecting '+marshal slice:[:inner]'") + } + sa.inner = true + } + + continue + } + + unhandledTags = append(unhandledTags, tag) + } + + if len(unhandledTags) > 0 { + abortAt(fset.Position(tagLine.Slash), fmt.Sprintf("+marshal directive contained the following unknown clauses: %v", strings.Join(unhandledTags, " "))) + } + + return mt +} + // collectMarshallableTypes walks the parsed AST and collects a list of type // declarations for which we need to generate the Marshallable interface. -func (g *Generator) collectMarshallableTypes(a *ast.File, f *token.FileSet) []*ast.TypeSpec { - var types []*ast.TypeSpec +func (g *Generator) collectMarshallableTypes(a *ast.File, f *token.FileSet) []marshallableType { + var types []marshallableType for _, decl := range a.Decls { gdecl, ok := decl.(*ast.GenDecl) // Type declaration? @@ -212,9 +270,11 @@ func (g *Generator) collectMarshallableTypes(a *ast.File, f *token.FileSet) []*a } // Does the comment contain a "+marshal" line? marked := false + var tagLine *ast.Comment for _, c := range gdecl.Doc.List { - if c.Text == "// +marshal" { + if strings.HasPrefix(c.Text, "// +marshal") { marked = true + tagLine = c break } } @@ -229,20 +289,17 @@ func (g *Generator) collectMarshallableTypes(a *ast.File, f *token.FileSet) []*a switch t.Type.(type) { case *ast.StructType: debugfAt(f.Position(t.Pos()), "Collected marshallable struct %s.\n", t.Name.Name) - types = append(types, t) - continue case *ast.Ident: // Newtype on primitive. debugfAt(f.Position(t.Pos()), "Collected marshallable newtype on primitive %s.\n", t.Name.Name) - types = append(types, t) - continue case *ast.ArrayType: // Newtype on array. debugfAt(f.Position(t.Pos()), "Collected marshallable newtype on array %s.\n", t.Name.Name) - types = append(types, t) - continue + default: + // A user specifically requested marshalling on this type, but we + // don't support it. + abortAt(f.Position(t.Pos()), fmt.Sprintf("Marshalling codegen was requested on type '%s', but go-marshal doesn't support this kind of declaration.\n", t.Name)) } - // A user specifically requested marshalling on this type, but we - // don't support it. - abortAt(f.Position(t.Pos()), fmt.Sprintf("Marshalling codegen was requested on type '%s', but go-marshal doesn't support this kind of declaration.\n", t.Name)) + types = append(types, newMarshallableType(f, tagLine, t)) + } } return types @@ -281,19 +338,28 @@ func (g *Generator) collectImports(a *ast.File, f *token.FileSet) map[string]imp } -func (g *Generator) generateOne(t *ast.TypeSpec, fset *token.FileSet) *interfaceGenerator { - i := newInterfaceGenerator(t, fset) - switch ty := t.Type.(type) { +func (g *Generator) generateOne(t marshallableType, fset *token.FileSet) *interfaceGenerator { + i := newInterfaceGenerator(t.spec, fset) + switch ty := t.spec.Type.(type) { case *ast.StructType: - i.validateStruct(t, ty) + i.validateStruct(t.spec, ty) i.emitMarshallableForStruct(ty) + if t.slice != nil { + i.emitMarshallableSliceForStruct(ty, t.slice) + } case *ast.Ident: i.validatePrimitiveNewtype(ty) i.emitMarshallableForPrimitiveNewtype(ty) + if t.slice != nil { + i.emitMarshallableSliceForPrimitiveNewtype(ty, t.slice) + } case *ast.ArrayType: - i.validateArrayNewtype(t.Name, ty) + i.validateArrayNewtype(t.spec.Name, ty) // After validate, we can safely call arrayLen. - i.emitMarshallableForArrayNewtype(t.Name, ty.Elt.(*ast.Ident), arrayLen(ty)) + i.emitMarshallableForArrayNewtype(t.spec.Name, ty.Elt.(*ast.Ident), arrayLen(ty)) + if t.slice != nil { + abortAt(fset.Position(t.slice.comment.Slash), fmt.Sprintf("Array type marked as '+marshal slice:...', but this is not supported. Perhaps fold one of the dimensions?")) + } default: // This should've been filtered out by collectMarshallabeTypes. panic(fmt.Sprintf("Unexpected type %+v", ty)) @@ -303,9 +369,9 @@ func (g *Generator) generateOne(t *ast.TypeSpec, fset *token.FileSet) *interface // generateOneTestSuite generates a test suite for the automatically generated // implementations type t. -func (g *Generator) generateOneTestSuite(t *ast.TypeSpec) *testGenerator { - i := newTestGenerator(t) - i.emitTests() +func (g *Generator) generateOneTestSuite(t marshallableType) *testGenerator { + i := newTestGenerator(t.spec) + i.emitTests(t.slice) return i } diff --git a/tools/go_marshal/gomarshal/generator_interfaces.go b/tools/go_marshal/gomarshal/generator_interfaces.go index 8babf61d2..8812c6878 100644 --- a/tools/go_marshal/gomarshal/generator_interfaces.go +++ b/tools/go_marshal/gomarshal/generator_interfaces.go @@ -163,3 +163,65 @@ func (g *interfaceGenerator) unmarshalScalar(accessor, typ, bufVar string) { g.recordPotentiallyNonPackedField(accessor) } } + +// emitCastToByteSlice unsafely casts an arbitrary type's underlying memory to a +// byte slice, bypassing escape analysis. The caller is responsible for ensuring +// srcPtr lives until they're done with dstVar, the runtime does not consider +// dstVar dependent on srcPtr due to the escape analysis bypass. +// +// srcPtr must be a pointer. +// +// This function uses internally uses the identifier "hdr", and cannot be used +// in a context where it is already bound. +func (g *interfaceGenerator) emitCastToByteSlice(srcPtr, dstVar, lenExpr string) { + g.recordUsedImport("gohacks") + g.emit("// Construct a slice backed by dst's underlying memory.\n") + g.emit("var %s []byte\n", dstVar) + g.emit("hdr := (*reflect.SliceHeader)(unsafe.Pointer(&%s))\n", dstVar) + g.emit("hdr.Data = uintptr(gohacks.Noescape(unsafe.Pointer(%s)))\n", srcPtr) + g.emit("hdr.Len = %s\n", lenExpr) + g.emit("hdr.Cap = %s\n\n", lenExpr) +} + +// emitCastToByteSlice unsafely casts a slice with elements of an abitrary type +// to a byte slice. As part of the cast, the byte slice is made to look +// independent of the src slice by bypassing escape analysis. This means the +// byte slice can be used without causing the source to escape. The caller is +// responsible for ensuring srcPtr lives until they're done with dstVar, as the +// runtime no longer considers dstVar dependent on srcPtr and is free to GC it. +// +// srcPtr must be a pointer. +// +// This function uses internally uses the identifiers "ptr", "val" and "hdr", +// and cannot be used in a context where these identifiers are already bound. +func (g *interfaceGenerator) emitCastSliceToByteSlice(srcPtr, dstVar, lenExpr string) { + g.emitNoEscapeSliceDataPointer(srcPtr, "val") + + g.emit("// Construct a slice backed by dst's underlying memory.\n") + g.emit("var %s []byte\n", dstVar) + g.emit("hdr := (*reflect.SliceHeader)(unsafe.Pointer(&%s))\n", dstVar) + g.emit("hdr.Data = uintptr(val)\n") + g.emit("hdr.Len = %s\n", lenExpr) + g.emit("hdr.Cap = %s\n\n", lenExpr) +} + +// emitNoEscapeSliceDataPointer unsafely casts a slice's data pointer to an +// unsafe.Pointer, bypassing escape analysis. The caller is responsible for +// ensuring srcPtr lives until they're done with dstVar, as the runtime no +// longer considers dstVar dependent on srcPtr and is free to GC it. +// +// srcPtr must be a pointer. +// +// This function uses internally uses the identifier "ptr" cannot be used in a +// context where this identifier is already bound. +func (g *interfaceGenerator) emitNoEscapeSliceDataPointer(srcPtr, dstVar string) { + g.recordUsedImport("gohacks") + g.emit("ptr := unsafe.Pointer(%s)\n", srcPtr) + g.emit("%s := gohacks.Noescape(unsafe.Pointer((*reflect.SliceHeader)(ptr).Data))\n\n", dstVar) +} + +func (g *interfaceGenerator) emitKeepAlive(ptrVar string) { + g.emit("// Since we bypassed the compiler's escape analysis, indicate that %s\n", ptrVar) + g.emit("// must live until the use above.\n") + g.emit("runtime.KeepAlive(%s)\n", ptrVar) +} diff --git a/tools/go_marshal/gomarshal/generator_interfaces_array_newtype.go b/tools/go_marshal/gomarshal/generator_interfaces_array_newtype.go index da36d9305..5ba74a606 100644 --- a/tools/go_marshal/gomarshal/generator_interfaces_array_newtype.go +++ b/tools/go_marshal/gomarshal/generator_interfaces_array_newtype.go @@ -104,79 +104,43 @@ func (g *interfaceGenerator) emitMarshallableForArrayNewtype(n, elt *ast.Ident, }) g.emit("}\n\n") + g.emit("// CopyOutN implements marshal.Marshallable.CopyOutN.\n") + g.emit("func (%s *%s) CopyOutN(task marshal.Task, addr usermem.Addr, limit int) (int, error) {\n", g.r, g.typeName()) + g.inIndent(func() { + g.emitCastToByteSlice(g.r, "buf", fmt.Sprintf("%s.SizeBytes()", g.r)) + + g.emit("length, err := task.CopyOutBytes(addr, buf[:limit])\n") + g.emitKeepAlive(g.r) + g.emit("return length, err\n") + }) + g.emit("}\n\n") + g.emit("// CopyOut implements marshal.Marshallable.CopyOut.\n") - g.emit("func (%s *%s) CopyOut(task marshal.Task, addr usermem.Addr) error {\n", g.r, g.typeName()) + g.emit("func (%s *%s) CopyOut(task marshal.Task, addr usermem.Addr) (int, error) {\n", g.r, g.typeName()) g.inIndent(func() { - // Fast serialization. - g.emit("// Bypass escape analysis on %s. The no-op arithmetic operation on the\n", g.r) - g.emit("// pointer makes the compiler think val doesn't depend on %s.\n", g.r) - g.emit("// See src/runtime/stubs.go:noescape() in the golang toolchain.\n") - g.emit("ptr := unsafe.Pointer(%s)\n", g.r) - g.emit("val := uintptr(ptr)\n") - g.emit("val = val^0\n\n") - - g.emit("// Construct a slice backed by %s's underlying memory.\n", g.r) - g.emit("var buf []byte\n") - g.emit("hdr := (*reflect.SliceHeader)(unsafe.Pointer(&buf))\n") - g.emit("hdr.Data = val\n") - g.emit("hdr.Len = %s.SizeBytes()\n", g.r) - g.emit("hdr.Cap = %s.SizeBytes()\n\n", g.r) - - g.emit("_, err := task.CopyOutBytes(addr, buf)\n") - g.emit("// Since we bypassed the compiler's escape analysis, indicate that %s\n", g.r) - g.emit("// must live until after the CopyOutBytes.\n") - g.emit("runtime.KeepAlive(%s)\n", g.r) - g.emit("return err\n") + g.emit("return %s.CopyOutN(task, addr, %s.SizeBytes())\n", g.r, g.r) }) g.emit("}\n\n") g.emit("// CopyIn implements marshal.Marshallable.CopyIn.\n") - g.emit("func (%s *%s) CopyIn(task marshal.Task, addr usermem.Addr) error {\n", g.r, g.typeName()) + g.emit("func (%s *%s) CopyIn(task marshal.Task, addr usermem.Addr) (int, error) {\n", g.r, g.typeName()) g.inIndent(func() { - g.emit("// Bypass escape analysis on %s. The no-op arithmetic operation on the\n", g.r) - g.emit("// pointer makes the compiler think val doesn't depend on %s.\n", g.r) - g.emit("// See src/runtime/stubs.go:noescape() in the golang toolchain.\n") - g.emit("ptr := unsafe.Pointer(%s)\n", g.r) - g.emit("val := uintptr(ptr)\n") - g.emit("val = val^0\n\n") - - g.emit("// Construct a slice backed by %s's underlying memory.\n", g.r) - g.emit("var buf []byte\n") - g.emit("hdr := (*reflect.SliceHeader)(unsafe.Pointer(&buf))\n") - g.emit("hdr.Data = val\n") - g.emit("hdr.Len = %s.SizeBytes()\n", g.r) - g.emit("hdr.Cap = %s.SizeBytes()\n\n", g.r) - - g.emit("_, err := task.CopyInBytes(addr, buf)\n") - g.emit("// Since we bypassed the compiler's escape analysis, indicate that %s\n", g.r) - g.emit("// must live until after the CopyInBytes.\n") - g.emit("runtime.KeepAlive(%s)\n", g.r) - g.emit("return err\n") + g.emitCastToByteSlice(g.r, "buf", fmt.Sprintf("%s.SizeBytes()", g.r)) + + g.emit("length, err := task.CopyInBytes(addr, buf)\n") + g.emitKeepAlive(g.r) + g.emit("return length, err\n") }) g.emit("}\n\n") g.emit("// WriteTo implements io.WriterTo.WriteTo.\n") g.emit("func (%s *%s) WriteTo(w io.Writer) (int64, error) {\n", g.r, g.typeName()) g.inIndent(func() { - g.emit("// Bypass escape analysis on %s. The no-op arithmetic operation on the\n", g.r) - g.emit("// pointer makes the compiler think val doesn't depend on %s.\n", g.r) - g.emit("// See src/runtime/stubs.go:noescape() in the golang toolchain.\n") - g.emit("ptr := unsafe.Pointer(%s)\n", g.r) - g.emit("val := uintptr(ptr)\n") - g.emit("val = val^0\n\n") - - g.emit("// Construct a slice backed by %s's underlying memory.\n", g.r) - g.emit("var buf []byte\n") - g.emit("hdr := (*reflect.SliceHeader)(unsafe.Pointer(&buf))\n") - g.emit("hdr.Data = val\n") - g.emit("hdr.Len = %s.SizeBytes()\n", g.r) - g.emit("hdr.Cap = %s.SizeBytes()\n\n", g.r) - - g.emit("len, err := w.Write(buf)\n") - g.emit("// Since we bypassed the compiler's escape analysis, indicate that %s\n", g.r) - g.emit("// must live until after the Write.\n") - g.emit("runtime.KeepAlive(%s)\n", g.r) - g.emit("return int64(len), err\n") + g.emitCastToByteSlice(g.r, "buf", fmt.Sprintf("%s.SizeBytes()", g.r)) + + g.emit("length, err := w.Write(buf)\n") + g.emitKeepAlive(g.r) + g.emit("return int64(length), err\n") }) g.emit("}\n\n") diff --git a/tools/go_marshal/gomarshal/generator_interfaces_primitive_newtype.go b/tools/go_marshal/gomarshal/generator_interfaces_primitive_newtype.go index 159397825..ef9bb903d 100644 --- a/tools/go_marshal/gomarshal/generator_interfaces_primitive_newtype.go +++ b/tools/go_marshal/gomarshal/generator_interfaces_primitive_newtype.go @@ -150,80 +150,133 @@ func (g *interfaceGenerator) emitMarshallableForPrimitiveNewtype(nt *ast.Ident) }) g.emit("}\n\n") + g.emit("// CopyOutN implements marshal.Marshallable.CopyOutN.\n") + g.emit("func (%s *%s) CopyOutN(task marshal.Task, addr usermem.Addr, limit int) (int, error) {\n", g.r, g.typeName()) + g.inIndent(func() { + g.emitCastToByteSlice(g.r, "buf", fmt.Sprintf("%s.SizeBytes()", g.r)) + + g.emit("length, err := task.CopyOutBytes(addr, buf[:limit])\n") + g.emitKeepAlive(g.r) + g.emit("return length, err\n") + }) + g.emit("}\n\n") + g.emit("// CopyOut implements marshal.Marshallable.CopyOut.\n") - g.emit("func (%s *%s) CopyOut(task marshal.Task, addr usermem.Addr) error {\n", g.r, g.typeName()) + g.emit("func (%s *%s) CopyOut(task marshal.Task, addr usermem.Addr) (int, error) {\n", g.r, g.typeName()) g.inIndent(func() { - // Fast serialization. - g.emit("// Bypass escape analysis on %s. The no-op arithmetic operation on the\n", g.r) - g.emit("// pointer makes the compiler think val doesn't depend on %s.\n", g.r) - g.emit("// See src/runtime/stubs.go:noescape() in the golang toolchain.\n") - g.emit("ptr := unsafe.Pointer(%s)\n", g.r) - g.emit("val := uintptr(ptr)\n") - g.emit("val = val^0\n\n") - - g.emit("// Construct a slice backed by %s's underlying memory.\n", g.r) - g.emit("var buf []byte\n") - g.emit("hdr := (*reflect.SliceHeader)(unsafe.Pointer(&buf))\n") - g.emit("hdr.Data = val\n") - g.emit("hdr.Len = %s.SizeBytes()\n", g.r) - g.emit("hdr.Cap = %s.SizeBytes()\n\n", g.r) - - g.emit("_, err := task.CopyOutBytes(addr, buf)\n") - g.emit("// Since we bypassed the compiler's escape analysis, indicate that %s\n", g.r) - g.emit("// must live until after the CopyOutBytes.\n") - g.emit("runtime.KeepAlive(%s)\n", g.r) - g.emit("return err\n") + g.emit("return %s.CopyOutN(task, addr, %s.SizeBytes())\n", g.r, g.r) }) g.emit("}\n\n") g.emit("// CopyIn implements marshal.Marshallable.CopyIn.\n") - g.emit("func (%s *%s) CopyIn(task marshal.Task, addr usermem.Addr) error {\n", g.r, g.typeName()) + g.emit("func (%s *%s) CopyIn(task marshal.Task, addr usermem.Addr) (int, error) {\n", g.r, g.typeName()) g.inIndent(func() { - g.emit("// Bypass escape analysis on %s. The no-op arithmetic operation on the\n", g.r) - g.emit("// pointer makes the compiler think val doesn't depend on %s.\n", g.r) - g.emit("// See src/runtime/stubs.go:noescape() in the golang toolchain.\n") - g.emit("ptr := unsafe.Pointer(%s)\n", g.r) - g.emit("val := uintptr(ptr)\n") - g.emit("val = val^0\n\n") - - g.emit("// Construct a slice backed by %s's underlying memory.\n", g.r) - g.emit("var buf []byte\n") - g.emit("hdr := (*reflect.SliceHeader)(unsafe.Pointer(&buf))\n") - g.emit("hdr.Data = val\n") - g.emit("hdr.Len = %s.SizeBytes()\n", g.r) - g.emit("hdr.Cap = %s.SizeBytes()\n\n", g.r) - - g.emit("_, err := task.CopyInBytes(addr, buf)\n") - g.emit("// Since we bypassed the compiler's escape analysis, indicate that %s\n", g.r) - g.emit("// must live until after the CopyInBytes.\n") - g.emit("runtime.KeepAlive(%s)\n", g.r) - g.emit("return err\n") + g.emitCastToByteSlice(g.r, "buf", fmt.Sprintf("%s.SizeBytes()", g.r)) + + g.emit("length, err := task.CopyInBytes(addr, buf)\n") + g.emitKeepAlive(g.r) + g.emit("return length, err\n") }) g.emit("}\n\n") g.emit("// WriteTo implements io.WriterTo.WriteTo.\n") g.emit("func (%s *%s) WriteTo(w io.Writer) (int64, error) {\n", g.r, g.typeName()) g.inIndent(func() { - g.emit("// Bypass escape analysis on %s. The no-op arithmetic operation on the\n", g.r) - g.emit("// pointer makes the compiler think val doesn't depend on %s.\n", g.r) - g.emit("// See src/runtime/stubs.go:noescape() in the golang toolchain.\n") - g.emit("ptr := unsafe.Pointer(%s)\n", g.r) - g.emit("val := uintptr(ptr)\n") - g.emit("val = val^0\n\n") - - g.emit("// Construct a slice backed by %s's underlying memory.\n", g.r) - g.emit("var buf []byte\n") - g.emit("hdr := (*reflect.SliceHeader)(unsafe.Pointer(&buf))\n") - g.emit("hdr.Data = val\n") - g.emit("hdr.Len = %s.SizeBytes()\n", g.r) - g.emit("hdr.Cap = %s.SizeBytes()\n\n", g.r) - - g.emit("len, err := w.Write(buf)\n") - g.emit("// Since we bypassed the compiler's escape analysis, indicate that %s\n", g.r) - g.emit("// must live until after the Write.\n") - g.emit("runtime.KeepAlive(%s)\n", g.r) - g.emit("return int64(len), err\n") + g.emitCastToByteSlice(g.r, "buf", fmt.Sprintf("%s.SizeBytes()", g.r)) + + g.emit("length, err := w.Write(buf)\n") + g.emitKeepAlive(g.r) + g.emit("return int64(length), err\n") + + }) + g.emit("}\n\n") +} + +func (g *interfaceGenerator) emitMarshallableSliceForPrimitiveNewtype(nt *ast.Ident, slice *sliceAPI) { + g.recordUsedImport("marshal") + g.recordUsedImport("usermem") + g.recordUsedImport("reflect") + g.recordUsedImport("runtime") + g.recordUsedImport("unsafe") + + eltType := g.typeName() + if slice.inner { + eltType = nt.Name + } + + g.emit("// Copy%sIn copies in a slice of %s objects from the task's memory.\n", slice.ident, eltType) + g.emit("func Copy%sIn(task marshal.Task, addr usermem.Addr, dst []%s) (int, error) {\n", slice.ident, eltType) + g.inIndent(func() { + g.emit("count := len(dst)\n") + g.emit("if count == 0 {\n") + g.inIndent(func() { + g.emit("return 0, nil\n") + }) + g.emit("}\n") + g.emit("size := (*%s)(nil).SizeBytes()\n\n", g.typeName()) + + g.emitCastSliceToByteSlice("&dst", "buf", "size * count") + + g.emit("length, err := task.CopyInBytes(addr, buf)\n") + g.emitKeepAlive("dst") + g.emit("return length, err\n") + }) + g.emit("}\n\n") + + g.emit("// Copy%sOut copies a slice of %s objects to the task's memory.\n", slice.ident, eltType) + g.emit("func Copy%sOut(task marshal.Task, addr usermem.Addr, src []%s) (int, error) {\n", slice.ident, eltType) + g.inIndent(func() { + g.emit("count := len(src)\n") + g.emit("if count == 0 {\n") + g.inIndent(func() { + g.emit("return 0, nil\n") + }) + g.emit("}\n") + g.emit("size := (*%s)(nil).SizeBytes()\n\n", g.typeName()) + + g.emitCastSliceToByteSlice("&src", "buf", "size * count") + + g.emit("length, err := task.CopyOutBytes(addr, buf)\n") + g.emitKeepAlive("src") + g.emit("return length, err\n") + }) + g.emit("}\n\n") + + g.emit("// MarshalUnsafe%s is like %s.MarshalUnsafe, but for a []%s.\n", slice.ident, g.typeName(), g.typeName()) + g.emit("func MarshalUnsafe%s(src []%s, dst []byte) (int, error) {\n", slice.ident, g.typeName()) + g.inIndent(func() { + g.emit("count := len(src)\n") + g.emit("if count == 0 {\n") + g.inIndent(func() { + g.emit("return 0, nil\n") + }) + g.emit("}\n") + g.emit("size := (*%s)(nil).SizeBytes()\n\n", g.typeName()) + + g.emitNoEscapeSliceDataPointer("&src", "val") + + g.emit("length, err := safecopy.CopyIn(dst[:(size*count)], val)\n") + g.emitKeepAlive("src") + g.emit("return length, err\n") + }) + g.emit("}\n\n") + + g.emit("// UnmarshalUnsafe%s is like %s.UnmarshalUnsafe, but for a []%s.\n", slice.ident, g.typeName(), g.typeName()) + g.emit("func UnmarshalUnsafe%s(dst []%s, src []byte) (int, error) {\n", slice.ident, g.typeName()) + g.inIndent(func() { + g.emit("count := len(dst)\n") + g.emit("if count == 0 {\n") + g.inIndent(func() { + g.emit("return 0, nil\n") + }) + g.emit("}\n") + g.emit("size := (*%s)(nil).SizeBytes()\n\n", g.typeName()) + + g.emitNoEscapeSliceDataPointer("&dst", "val") + g.emit("length, err := safecopy.CopyOut(val, src[:(size*count)])\n") + g.emitKeepAlive("dst") + g.emit("return length, err\n") }) g.emit("}\n\n") } diff --git a/tools/go_marshal/gomarshal/generator_interfaces_struct.go b/tools/go_marshal/gomarshal/generator_interfaces_struct.go index e66a38b2e..bd57eae0e 100644 --- a/tools/go_marshal/gomarshal/generator_interfaces_struct.go +++ b/tools/go_marshal/gomarshal/generator_interfaces_struct.go @@ -72,20 +72,24 @@ func (g *interfaceGenerator) validateStruct(ts *ast.TypeSpec, st *ast.StructType }) } -func (g *interfaceGenerator) emitMarshallableForStruct(st *ast.StructType) { - // Is g.t a packed struct without consideing field types? - thisPacked := true +func (g *interfaceGenerator) isStructPacked(st *ast.StructType) bool { + packed := true forEachStructField(st, func(f *ast.Field) { if f.Tag != nil { if f.Tag.Value == "`marshal:\"unaligned\"`" { - if thisPacked { + if packed { debugfAt(g.f.Position(g.t.Pos()), fmt.Sprintf("Marking type '%s' as not packed due to tag `marshal:\"unaligned\"`.\n", g.t.Name)) - thisPacked = false + packed = false } } } }) + return packed +} + +func (g *interfaceGenerator) emitMarshallableForStruct(st *ast.StructType) { + thisPacked := g.isStructPacked(st) g.emit("// SizeBytes implements marshal.Marshallable.SizeBytes.\n") g.emit("func (%s *%s) SizeBytes() int {\n", g.r, g.typeName()) @@ -302,17 +306,16 @@ func (g *interfaceGenerator) emitMarshallableForStruct(st *ast.StructType) { }) g.emit("}\n\n") - g.emit("// CopyOut implements marshal.Marshallable.CopyOut.\n") + g.emit("// CopyOutN implements marshal.Marshallable.CopyOutN.\n") g.recordUsedImport("marshal") g.recordUsedImport("usermem") - g.emit("func (%s *%s) CopyOut(task marshal.Task, addr usermem.Addr) error {\n", g.r, g.typeName()) + g.emit("func (%s *%s) CopyOutN(task marshal.Task, addr usermem.Addr, limit int) (int, error) {\n", g.r, g.typeName()) g.inIndent(func() { fallback := func() { g.emit("// Type %s doesn't have a packed layout in memory, fall back to MarshalBytes.\n", g.typeName()) g.emit("buf := task.CopyScratchBuffer(%s.SizeBytes())\n", g.r) g.emit("%s.MarshalBytes(buf)\n", g.r) - g.emit("_, err := task.CopyOutBytes(addr, buf)\n") - g.emit("return err\n") + g.emit("return task.CopyOutBytes(addr, buf[:limit])\n") } if thisPacked { g.recordUsedImport("reflect") @@ -324,48 +327,39 @@ func (g *interfaceGenerator) emitMarshallableForStruct(st *ast.StructType) { g.emit("}\n\n") } // Fast serialization. - g.emit("// Bypass escape analysis on %s. The no-op arithmetic operation on the\n", g.r) - g.emit("// pointer makes the compiler think val doesn't depend on %s.\n", g.r) - g.emit("// See src/runtime/stubs.go:noescape() in the golang toolchain.\n") - g.emit("ptr := unsafe.Pointer(%s)\n", g.r) - g.emit("val := uintptr(ptr)\n") - g.emit("val = val^0\n\n") - - g.emit("// Construct a slice backed by %s's underlying memory.\n", g.r) - g.emit("var buf []byte\n") - g.emit("hdr := (*reflect.SliceHeader)(unsafe.Pointer(&buf))\n") - g.emit("hdr.Data = val\n") - g.emit("hdr.Len = %s.SizeBytes()\n", g.r) - g.emit("hdr.Cap = %s.SizeBytes()\n\n", g.r) - - g.emit("_, err := task.CopyOutBytes(addr, buf)\n") - g.emit("// Since we bypassed the compiler's escape analysis, indicate that %s\n", g.r) - g.emit("// must live until after the CopyOutBytes.\n") - g.emit("runtime.KeepAlive(%s)\n", g.r) - g.emit("return err\n") + g.emitCastToByteSlice(g.r, "buf", fmt.Sprintf("%s.SizeBytes()", g.r)) + + g.emit("length, err := task.CopyOutBytes(addr, buf[:limit])\n") + g.emitKeepAlive(g.r) + g.emit("return length, err\n") } else { fallback() } }) g.emit("}\n\n") + g.emit("// CopyOut implements marshal.Marshallable.CopyOut.\n") + g.recordUsedImport("marshal") + g.recordUsedImport("usermem") + g.emit("func (%s *%s) CopyOut(task marshal.Task, addr usermem.Addr) (int, error) {\n", g.r, g.typeName()) + g.inIndent(func() { + g.emit("return %s.CopyOutN(task, addr, %s.SizeBytes())\n", g.r, g.r) + }) + g.emit("}\n\n") + g.emit("// CopyIn implements marshal.Marshallable.CopyIn.\n") g.recordUsedImport("marshal") g.recordUsedImport("usermem") - g.emit("func (%s *%s) CopyIn(task marshal.Task, addr usermem.Addr) error {\n", g.r, g.typeName()) + g.emit("func (%s *%s) CopyIn(task marshal.Task, addr usermem.Addr) (int, error) {\n", g.r, g.typeName()) g.inIndent(func() { fallback := func() { g.emit("// Type %s doesn't have a packed layout in memory, fall back to UnmarshalBytes.\n", g.typeName()) g.emit("buf := task.CopyScratchBuffer(%s.SizeBytes())\n", g.r) - g.emit("_, err := task.CopyInBytes(addr, buf)\n") - g.emit("if err != nil {\n") - g.inIndent(func() { - g.emit("return err\n") - }) - g.emit("}\n") - + g.emit("length, err := task.CopyInBytes(addr, buf)\n") + g.emit("// Unmarshal unconditionally. If we had a short copy-in, this results in a\n") + g.emit("// partially unmarshalled struct.\n") g.emit("%s.UnmarshalBytes(buf)\n", g.r) - g.emit("return nil\n") + g.emit("return length, err\n") } if thisPacked { g.recordUsedImport("reflect") @@ -377,25 +371,11 @@ func (g *interfaceGenerator) emitMarshallableForStruct(st *ast.StructType) { g.emit("}\n\n") } // Fast deserialization. - g.emit("// Bypass escape analysis on %s. The no-op arithmetic operation on the\n", g.r) - g.emit("// pointer makes the compiler think val doesn't depend on %s.\n", g.r) - g.emit("// See src/runtime/stubs.go:noescape() in the golang toolchain.\n") - g.emit("ptr := unsafe.Pointer(%s)\n", g.r) - g.emit("val := uintptr(ptr)\n") - g.emit("val = val^0\n\n") - - g.emit("// Construct a slice backed by %s's underlying memory.\n", g.r) - g.emit("var buf []byte\n") - g.emit("hdr := (*reflect.SliceHeader)(unsafe.Pointer(&buf))\n") - g.emit("hdr.Data = val\n") - g.emit("hdr.Len = %s.SizeBytes()\n", g.r) - g.emit("hdr.Cap = %s.SizeBytes()\n\n", g.r) - - g.emit("_, err := task.CopyInBytes(addr, buf)\n") - g.emit("// Since we bypassed the compiler's escape analysis, indicate that %s\n", g.r) - g.emit("// must live until after the CopyInBytes.\n") - g.emit("runtime.KeepAlive(%s)\n", g.r) - g.emit("return err\n") + g.emitCastToByteSlice(g.r, "buf", fmt.Sprintf("%s.SizeBytes()", g.r)) + + g.emit("length, err := task.CopyInBytes(addr, buf)\n") + g.emitKeepAlive(g.r) + g.emit("return length, err\n") } else { fallback() } @@ -410,8 +390,8 @@ func (g *interfaceGenerator) emitMarshallableForStruct(st *ast.StructType) { g.emit("// Type %s doesn't have a packed layout in memory, fall back to MarshalBytes.\n", g.typeName()) g.emit("buf := make([]byte, %s.SizeBytes())\n", g.r) g.emit("%s.MarshalBytes(buf)\n", g.r) - g.emit("n, err := w.Write(buf)\n") - g.emit("return int64(n), err\n") + g.emit("length, err := w.Write(buf)\n") + g.emit("return int64(length), err\n") } if thisPacked { g.recordUsedImport("reflect") @@ -423,25 +403,199 @@ func (g *interfaceGenerator) emitMarshallableForStruct(st *ast.StructType) { g.emit("}\n\n") } // Fast serialization. - g.emit("// Bypass escape analysis on %s. The no-op arithmetic operation on the\n", g.r) - g.emit("// pointer makes the compiler think val doesn't depend on %s.\n", g.r) - g.emit("// See src/runtime/stubs.go:noescape() in the golang toolchain.\n") - g.emit("ptr := unsafe.Pointer(%s)\n", g.r) - g.emit("val := uintptr(ptr)\n") - g.emit("val = val^0\n\n") - - g.emit("// Construct a slice backed by %s's underlying memory.\n", g.r) - g.emit("var buf []byte\n") - g.emit("hdr := (*reflect.SliceHeader)(unsafe.Pointer(&buf))\n") - g.emit("hdr.Data = val\n") - g.emit("hdr.Len = %s.SizeBytes()\n", g.r) - g.emit("hdr.Cap = %s.SizeBytes()\n\n", g.r) - - g.emit("len, err := w.Write(buf)\n") - g.emit("// Since we bypassed the compiler's escape analysis, indicate that %s\n", g.r) - g.emit("// must live until after the Write.\n") - g.emit("runtime.KeepAlive(%s)\n", g.r) - g.emit("return int64(len), err\n") + g.emitCastToByteSlice(g.r, "buf", fmt.Sprintf("%s.SizeBytes()", g.r)) + + g.emit("length, err := w.Write(buf)\n") + g.emitKeepAlive(g.r) + g.emit("return int64(length), err\n") + } else { + fallback() + } + }) + g.emit("}\n\n") +} + +func (g *interfaceGenerator) emitMarshallableSliceForStruct(st *ast.StructType, slice *sliceAPI) { + thisPacked := g.isStructPacked(st) + + if slice.inner { + abortAt(g.f.Position(slice.comment.Slash), fmt.Sprintf("The ':inner' argument to '+marshal slice:%s:inner' is only applicable to newtypes on primitives. Remove it from this struct declaration.", slice.ident)) + } + + g.recordUsedImport("marshal") + g.recordUsedImport("usermem") + + g.emit("// Copy%sIn copies in a slice of %s objects from the task's memory.\n", slice.ident, g.typeName()) + g.emit("func Copy%sIn(task marshal.Task, addr usermem.Addr, dst []%s) (int, error) {\n", slice.ident, g.typeName()) + g.inIndent(func() { + g.emit("count := len(dst)\n") + g.emit("if count == 0 {\n") + g.inIndent(func() { + g.emit("return 0, nil\n") + }) + g.emit("}\n") + g.emit("size := (*%s)(nil).SizeBytes()\n\n", g.typeName()) + + fallback := func() { + g.emit("// Type %s doesn't have a packed layout in memory, fall back to UnmarshalBytes.\n", g.typeName()) + g.emit("buf := task.CopyScratchBuffer(size * count)\n") + g.emit("length, err := task.CopyInBytes(addr, buf)\n\n") + + g.emit("// Unmarshal as much as possible, even on error. First handle full objects.\n") + g.emit("limit := length/size\n") + g.emit("for idx := 0; idx < limit; idx++ {\n") + g.inIndent(func() { + g.emit("dst[idx].UnmarshalBytes(buf[size*idx:size*(idx+1)])\n") + }) + g.emit("}\n\n") + + g.emit("// Handle any final partial object.\n") + g.emit("if length < size*count && length%size != 0 {\n") + g.inIndent(func() { + g.emit("idx := limit\n") + g.emit("dst[idx].UnmarshalBytes(buf[size*idx:size*(idx+1)])\n") + }) + g.emit("}\n\n") + + g.emit("return length, err\n") + } + if thisPacked { + g.recordUsedImport("reflect") + g.recordUsedImport("runtime") + g.recordUsedImport("unsafe") + if _, ok := g.areFieldsPackedExpression(); ok { + g.emit("if !dst[0].Packed() {\n") + g.inIndent(fallback) + g.emit("}\n\n") + } + // Fast deserialization. + g.emitCastSliceToByteSlice("&dst", "buf", "size * count") + + g.emit("length, err := task.CopyInBytes(addr, buf)\n") + g.emitKeepAlive("dst") + g.emit("return length, err\n") + } else { + fallback() + } + }) + g.emit("}\n\n") + + g.emit("// Copy%sOut copies a slice of %s objects to the task's memory.\n", slice.ident, g.typeName()) + g.emit("func Copy%sOut(task marshal.Task, addr usermem.Addr, src []%s) (int, error) {\n", slice.ident, g.typeName()) + g.inIndent(func() { + g.emit("count := len(src)\n") + g.emit("if count == 0 {\n") + g.inIndent(func() { + g.emit("return 0, nil\n") + }) + g.emit("}\n") + g.emit("size := (*%s)(nil).SizeBytes()\n\n", g.typeName()) + + fallback := func() { + g.emit("// Type %s doesn't have a packed layout in memory, fall back to MarshalBytes.\n", g.typeName()) + g.emit("buf := task.CopyScratchBuffer(size * count)\n") + g.emit("for idx := 0; idx < count; idx++ {\n") + g.inIndent(func() { + g.emit("src[idx].MarshalBytes(buf[size*idx:size*(idx+1)])\n") + }) + g.emit("}\n") + g.emit("return task.CopyOutBytes(addr, buf)\n") + } + if thisPacked { + g.recordUsedImport("reflect") + g.recordUsedImport("runtime") + g.recordUsedImport("unsafe") + if _, ok := g.areFieldsPackedExpression(); ok { + g.emit("if !src[0].Packed() {\n") + g.inIndent(fallback) + g.emit("}\n\n") + } + // Fast serialization. + g.emitCastSliceToByteSlice("&src", "buf", "size * count") + + g.emit("length, err := task.CopyOutBytes(addr, buf)\n") + g.emitKeepAlive("src") + g.emit("return length, err\n") + } else { + fallback() + } + }) + g.emit("}\n\n") + + g.emit("// MarshalUnsafe%s is like %s.MarshalUnsafe, but for a []%s.\n", slice.ident, g.typeName(), g.typeName()) + g.emit("func MarshalUnsafe%s(src []%s, dst []byte) (int, error) {\n", slice.ident, g.typeName()) + g.inIndent(func() { + g.emit("count := len(src)\n") + g.emit("if count == 0 {\n") + g.inIndent(func() { + g.emit("return 0, nil\n") + }) + g.emit("}\n") + g.emit("size := (*%s)(nil).SizeBytes()\n\n", g.typeName()) + + fallback := func() { + g.emit("// Type %s doesn't have a packed layout in memory, fall back to MarshalBytes.\n", g.typeName()) + g.emit("for idx := 0; idx < count; idx++ {\n") + g.inIndent(func() { + g.emit("src[idx].MarshalBytes(dst[size*idx:(size)*(idx+1)])\n") + }) + g.emit("}\n") + g.emit("return size * count, nil\n") + } + if thisPacked { + g.recordUsedImport("reflect") + g.recordUsedImport("runtime") + g.recordUsedImport("unsafe") + if _, ok := g.areFieldsPackedExpression(); ok { + g.emit("if !src[0].Packed() {\n") + g.inIndent(fallback) + g.emit("}\n\n") + } + g.emitNoEscapeSliceDataPointer("&src", "val") + + g.emit("length, err := safecopy.CopyIn(dst[:(size*count)], val)\n") + g.emitKeepAlive("src") + g.emit("return length, err\n") + } else { + fallback() + } + }) + g.emit("}\n\n") + + g.emit("// UnmarshalUnsafe%s is like %s.UnmarshalUnsafe, but for a []%s.\n", slice.ident, g.typeName(), g.typeName()) + g.emit("func UnmarshalUnsafe%s(dst []%s, src []byte) (int, error) {\n", slice.ident, g.typeName()) + g.inIndent(func() { + g.emit("count := len(dst)\n") + g.emit("if count == 0 {\n") + g.inIndent(func() { + g.emit("return 0, nil\n") + }) + g.emit("}\n") + g.emit("size := (*%s)(nil).SizeBytes()\n\n", g.typeName()) + + fallback := func() { + g.emit("// Type %s doesn't have a packed layout in memory, fall back to UnmarshalBytes.\n", g.typeName()) + g.emit("for idx := 0; idx < count; idx++ {\n") + g.inIndent(func() { + g.emit("dst[idx].UnmarshalBytes(src[size*idx:size*(idx+1)])\n") + }) + g.emit("}\n") + g.emit("return size * count, nil\n") + } + if thisPacked { + g.recordUsedImport("reflect") + g.recordUsedImport("runtime") + g.recordUsedImport("unsafe") + if _, ok := g.areFieldsPackedExpression(); ok { + g.emit("if !dst[0].Packed() {\n") + g.inIndent(fallback) + g.emit("}\n\n") + } + g.emitNoEscapeSliceDataPointer("&dst", "val") + + g.emit("length, err := safecopy.CopyOut(val, src[:(size*count)])\n") + g.emitKeepAlive("dst") + g.emit("return length, err\n") } else { fallback() } diff --git a/tools/go_marshal/gomarshal/generator_tests.go b/tools/go_marshal/gomarshal/generator_tests.go index fd992e44a..631295373 100644 --- a/tools/go_marshal/gomarshal/generator_tests.go +++ b/tools/go_marshal/gomarshal/generator_tests.go @@ -30,6 +30,11 @@ var standardImports = []string{ "gvisor.dev/gvisor/tools/go_marshal/analysis", } +var sliceAPIImports = []string{ + "encoding/binary", + "gvisor.dev/gvisor/pkg/usermem", +} + type testGenerator struct { sourceBuffer @@ -58,6 +63,11 @@ func newTestGenerator(t *ast.TypeSpec) *testGenerator { for _, i := range standardImports { g.imports.add(i).markUsed() } + // These imports are used if a type requests the slice API. Don't + // mark them as used by default. + for _, i := range sliceAPIImports { + g.imports.add(i) + } return g } @@ -132,6 +142,42 @@ func (g *testGenerator) emitTestMarshalUnmarshalPreservesData() { }) } +func (g *testGenerator) emitTestMarshalUnmarshalSlicePreservesData(slice *sliceAPI) { + for _, name := range []string{"binary", "usermem"} { + if !g.imports.markUsed(name) { + panic(fmt.Sprintf("Generated test for '%s' referenced a non-existent import with local name '%s'", g.typeName(), name)) + } + } + + g.inTestFunction("TestSafeMarshalUnmarshalSlicePreservesData", func() { + g.emit("var x, y, yUnsafe [8]%s\n", g.typeName()) + g.emit("analysis.RandomizeValue(&x)\n\n") + g.emit("size := (*%s)(nil).SizeBytes() * len(x)\n", g.typeName()) + g.emit("buf := bytes.NewBuffer(make([]byte, size))\n") + g.emit("buf.Reset()\n") + g.emit("if err := binary.Write(buf, usermem.ByteOrder, x[:]); err != nil {\n") + g.inIndent(func() { + g.emit("t.Fatal(fmt.Sprintf(\"binary.Write failed: %v\", err))\n") + }) + g.emit("}\n") + g.emit("bufUnsafe := make([]byte, size)\n") + g.emit("MarshalUnsafe%s(x[:], bufUnsafe)\n\n", slice.ident) + + g.emit("UnmarshalUnsafe%s(y[:], buf.Bytes())\n", slice.ident) + g.emit("if !reflect.DeepEqual(x, y) {\n") + g.inIndent(func() { + g.emit("t.Fatal(fmt.Sprintf(\"Data corrupted across binary.Write/UnmarshalUnsafeSlice cycle:\\nBefore: %+v\\nAfter: %+v\\n\", x, y))\n") + }) + g.emit("}\n") + g.emit("UnmarshalUnsafe%s(yUnsafe[:], bufUnsafe)\n", slice.ident) + g.emit("if !reflect.DeepEqual(x, yUnsafe) {\n") + g.inIndent(func() { + g.emit("t.Fatal(fmt.Sprintf(\"Data corrupted across MarshalUnsafeSlice/UnmarshalUnsafeSlice cycle:\\nBefore: %+v\\nAfter: %+v\\n\", x, yUnsafe))\n") + }) + g.emit("}\n\n") + }) +} + func (g *testGenerator) emitTestWriteToUnmarshalPreservesData() { g.inTestFunction("TestWriteToUnmarshalPreservesData", func() { g.emit("var x, y, yUnsafe %s\n", g.typeName()) @@ -170,12 +216,16 @@ func (g *testGenerator) emitTestSizeBytesOnTypedNilPtr() { }) } -func (g *testGenerator) emitTests() { +func (g *testGenerator) emitTests(slice *sliceAPI) { g.emitTestNonZeroSize() g.emitTestSuspectAlignment() g.emitTestMarshalUnmarshalPreservesData() g.emitTestWriteToUnmarshalPreservesData() g.emitTestSizeBytesOnTypedNilPtr() + + if slice != nil { + g.emitTestMarshalUnmarshalSlicePreservesData(slice) + } } func (g *testGenerator) write(out io.Writer) error { diff --git a/tools/go_marshal/gomarshal/util.go b/tools/go_marshal/gomarshal/util.go index a0936e013..4cb22dd2d 100644 --- a/tools/go_marshal/gomarshal/util.go +++ b/tools/go_marshal/gomarshal/util.go @@ -344,22 +344,25 @@ func newImportTable() *importTable { // result in a panic. func (i *importTable) merge(other *importTable) { for name, im := range other.is { - if dup, ok := i.is[name]; ok && !dup.equivalent(im) { - panic(fmt.Sprintf("Found colliding import statements: ours: %+v, other's: %+v", dup, im)) + dup, ok := i.is[name] + if ok { + // When merging two imports, if either are marked used, the merged entry + // should also be marked used. + im.used = im.used || dup.used + + if !dup.equivalent(im) { + panic(fmt.Sprintf("Found colliding import statements: ours: %+v, other's: %+v", dup, im)) + } } - i.is[name] = im } } func (i *importTable) addStmt(s *importStmt) *importStmt { if old, ok := i.is[s.name]; ok && !old.equivalent(s) { - // A collision should always be between an import inserted by the - // go-marshal tool and an import from the original source file (assuming - // the original source file was valid). We could theoretically handle - // the collision by assigning a local name to our import. However, this - // would need to be plumbed throughout the generator. Given that - // collisions should be rare, simply panic on collision. + // We could theoretically handle the collision by assigning a local name + // to one of the imports. However, this is a non-trivial transformation. + // Given that collisions should be rare, simply panic on collision. panic(fmt.Sprintf("Import collision: old: %s as %v; new: %v as %v", old.path, old.name, s.path, s.name)) } i.is[s.name] = s diff --git a/tools/go_marshal/marshal/marshal.go b/tools/go_marshal/marshal/marshal.go index f129788e0..cb2166252 100644 --- a/tools/go_marshal/marshal/marshal.go +++ b/tools/go_marshal/marshal/marshal.go @@ -42,7 +42,11 @@ type Task interface { CopyInBytes(addr usermem.Addr, b []byte) (int, error) } -// Marshallable represents a type that can be marshalled to and from memory. +// Marshallable represents operations on a type that can be marshalled to and +// from memory. +// +// go-marshal automatically generates implementations for this interface for +// types marked as '+marshal'. type Marshallable interface { io.WriterTo @@ -54,12 +58,18 @@ type Marshallable interface { // likely make use of the type of these fields). SizeBytes() int - // MarshalBytes serializes a copy of a type to dst. dst must be at least - // SizeBytes() long. + // MarshalBytes serializes a copy of a type to dst. dst may be smaller than + // SizeBytes(), which results in a part of the struct being marshalled. Note + // that this may have unexpected results for non-packed types, as implicit + // padding needs to be taken into account when reasoning about how much of + // the type is serialized. MarshalBytes(dst []byte) - // UnmarshalBytes deserializes a type from src. src must be at least - // SizeBytes() long. + // UnmarshalBytes deserializes a type from src. src may be smaller than + // SizeBytes(), which results in a partially deserialized struct. Note that + // this may have unexpected results for non-packed types, as implicit + // padding needs to be taken into account when reasoning about how much of + // the type is deserialized. UnmarshalBytes(src []byte) // Packed returns true if the marshalled size of the type is the same as the @@ -67,13 +77,20 @@ type Marshallable interface { // starting at unaligned addresses (should always be true by default for ABI // structs, verified by automatically generated tests when using // go_marshal), and has no fields marked `marshal:"unaligned"`. + // + // Packed must return the same result for all possible values of the type + // implementing it. Violating this constraint implies the type doesn't have + // a static memory layout, and will lead to memory corruption. + // Go-marshal-generated code reuses the result of Packed for multiple values + // of the same type. Packed() bool // MarshalUnsafe serializes a type by bulk copying its in-memory // representation to the dst buffer. This is only safe to do when the type // has no implicit padding, see Marshallable.Packed. When Packed would // return false, MarshalUnsafe should fall back to the safer but slower - // MarshalBytes. + // MarshalBytes. dst may be smaller than SizeBytes(), see comment for + // MarshalBytes for implications. MarshalUnsafe(dst []byte) // UnmarshalUnsafe deserializes a type by directly copying to the underlying @@ -82,7 +99,8 @@ type Marshallable interface { // This allows much faster unmarshalling of types which have no implicit // padding, see Marshallable.Packed. When Packed would return false, // UnmarshalUnsafe should fall back to the safer but slower unmarshal - // mechanism implemented in UnmarshalBytes. + // mechanism implemented in UnmarshalBytes. src may be smaller than + // SizeBytes(), see comment for UnmarshalBytes for implications. UnmarshalUnsafe(src []byte) // CopyIn deserializes a Marshallable type from a task's memory. This may @@ -91,12 +109,79 @@ type Marshallable interface { // marshalled does not escape. The implementation should avoid creating // extra copies in memory by directly deserializing to the object's // underlying memory. - CopyIn(task Task, addr usermem.Addr) error + // + // If the copy-in from the task memory is only partially successful, CopyIn + // should still attempt to deserialize as much data as possible. See comment + // for UnmarshalBytes. + CopyIn(task Task, addr usermem.Addr) (int, error) // CopyOut serializes a Marshallable type to a task's memory. This may only // be called from a task goroutine. This is more efficient than calling // MarshalUnsafe on Marshallable.Packed types, as the type being serialized // does not escape. The implementation should avoid creating extra copies in // memory by directly serializing from the object's underlying memory. - CopyOut(task Task, addr usermem.Addr) error + // + // The copy-out to the task memory may be partially successful, in which + // case CopyOut returns how much data was serialized. See comment for + // MarshalBytes for implications. + CopyOut(task Task, addr usermem.Addr) (int, error) + + // CopyOutN is like CopyOut, but explicitly requests a partial + // copy-out. Note that this may yield unexpected results for non-packed + // types and the caller may only want to allow this for packed types. See + // comment on MarshalBytes. + // + // The limit must be less than or equal to SizeBytes(). + CopyOutN(task Task, addr usermem.Addr, limit int) (int, error) } + +// go-marshal generates additional functions for a type based on additional +// clauses to the +marshal directive. They are documented below. +// +// Slice API +// ========= +// +// Adding a "slice" clause to the +marshal directive for structs or newtypes on +// primitives like this: +// +// // +marshal slice:FooSlice +// type Foo struct { ... } +// +// Generates four additional functions for marshalling slices of Foos like this: +// +// // MarshalUnsafeFooSlice is like Foo.MarshalUnsafe, buf for a []Foo. It's +// // more efficient that repeatedly calling calling Foo.MarshalUnsafe over a +// // []Foo in a loop. +// func MarshalUnsafeFooSlice(src []Foo, dst []byte) (int, error) { ... } +// +// // UnmarshalUnsafeFooSlice is like Foo.UnmarshalUnsafe, buf for a []Foo. It's +// // more efficient that repeatedly calling calling Foo.UnmarshalUnsafe over a +// // []Foo in a loop. +// func UnmarshalUnsafeFooSlice(dst []Foo, src []byte) (int, error) { ... } +// +// // CopyFooSliceIn copies in a slice of Foo objects from the task's memory. +// func CopyFooSliceIn(task marshal.Task, addr usermem.Addr, dst []Foo) (int, error) { ... } +// +// // CopyFooSliceIn copies out a slice of Foo objects to the task's memory. +// func CopyFooSliceOut(task marshal.Task, addr usermem.Addr, src []Foo) (int, error) { ... } +// +// The name of the functions are of the format "Copy%sIn" and "Copy%sOut", where +// %s is the first argument to the slice clause. This directive is not supported +// for newtypes on arrays. +// +// The slice clause also takes an optional second argument, which must be the +// value "inner": +// +// // +marshal slice:Int32Slice:inner +// type Int32 int32 +// +// This is only valid on newtypes on primitives, and causes the generated +// functions to accept slices of the inner type instead: +// +// func CopyInt32SliceIn(task marshal.Task, addr usermem.Addr, dst []int32) (int, error) { ... } +// +// Without "inner", they would instead be: +// +// func CopyInt32SliceIn(task marshal.Task, addr usermem.Addr, dst []Int32) (int, error) { ... } +// +// This may help avoid a cast depending on how the generated functions are used. diff --git a/tools/go_marshal/primitive/BUILD b/tools/go_marshal/primitive/BUILD new file mode 100644 index 000000000..cc08ba63a --- /dev/null +++ b/tools/go_marshal/primitive/BUILD @@ -0,0 +1,18 @@ +load("//tools:defs.bzl", "go_library") + +licenses(["notice"]) + +go_library( + name = "primitive", + srcs = [ + "primitive.go", + ], + marshal = True, + visibility = [ + "//:sandbox", + ], + deps = [ + "//pkg/usermem", + "//tools/go_marshal/marshal", + ], +) diff --git a/tools/go_marshal/primitive/primitive.go b/tools/go_marshal/primitive/primitive.go new file mode 100644 index 000000000..ebcf130ae --- /dev/null +++ b/tools/go_marshal/primitive/primitive.go @@ -0,0 +1,175 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package primitive defines marshal.Marshallable implementations for primitive +// types. +package primitive + +import ( + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/tools/go_marshal/marshal" +) + +// Int16 is a marshal.Marshallable implementation for int16. +// +// +marshal slice:Int16Slice:inner +type Int16 int16 + +// Uint16 is a marshal.Marshallable implementation for uint16. +// +// +marshal slice:Uint16Slice:inner +type Uint16 uint16 + +// Int32 is a marshal.Marshallable implementation for int32. +// +// +marshal slice:Int32Slice:inner +type Int32 int32 + +// Uint32 is a marshal.Marshallable implementation for uint32. +// +// +marshal slice:Uint32Slice:inner +type Uint32 uint32 + +// Int64 is a marshal.Marshallable implementation for int64. +// +// +marshal slice:Int64Slice:inner +type Int64 int64 + +// Uint64 is a marshal.Marshallable implementation for uint64. +// +// +marshal slice:Uint64Slice:inner +type Uint64 uint64 + +// Below, we define some convenience functions for marshalling primitive types +// using the newtypes above, without requiring superfluous casts. + +// 16-bit integers + +// CopyInt16In is a convenient wrapper for copying in an int16 from the task's +// memory. +func CopyInt16In(task marshal.Task, addr usermem.Addr, dst *int16) (int, error) { + var buf Int16 + n, err := buf.CopyIn(task, addr) + if err != nil { + return n, err + } + *dst = int16(buf) + return n, nil +} + +// CopyInt16Out is a convenient wrapper for copying out an int16 to the task's +// memory. +func CopyInt16Out(task marshal.Task, addr usermem.Addr, src int16) (int, error) { + srcP := Int16(src) + return srcP.CopyOut(task, addr) +} + +// CopyUint16In is a convenient wrapper for copying in a uint16 from the task's +// memory. +func CopyUint16In(task marshal.Task, addr usermem.Addr, dst *uint16) (int, error) { + var buf Uint16 + n, err := buf.CopyIn(task, addr) + if err != nil { + return n, err + } + *dst = uint16(buf) + return n, nil +} + +// CopyUint16Out is a convenient wrapper for copying out a uint16 to the task's +// memory. +func CopyUint16Out(task marshal.Task, addr usermem.Addr, src uint16) (int, error) { + srcP := Uint16(src) + return srcP.CopyOut(task, addr) +} + +// 32-bit integers + +// CopyInt32In is a convenient wrapper for copying in an int32 from the task's +// memory. +func CopyInt32In(task marshal.Task, addr usermem.Addr, dst *int32) (int, error) { + var buf Int32 + n, err := buf.CopyIn(task, addr) + if err != nil { + return n, err + } + *dst = int32(buf) + return n, nil +} + +// CopyInt32Out is a convenient wrapper for copying out an int32 to the task's +// memory. +func CopyInt32Out(task marshal.Task, addr usermem.Addr, src int32) (int, error) { + srcP := Int32(src) + return srcP.CopyOut(task, addr) +} + +// CopyUint32In is a convenient wrapper for copying in a uint32 from the task's +// memory. +func CopyUint32In(task marshal.Task, addr usermem.Addr, dst *uint32) (int, error) { + var buf Uint32 + n, err := buf.CopyIn(task, addr) + if err != nil { + return n, err + } + *dst = uint32(buf) + return n, nil +} + +// CopyUint32Out is a convenient wrapper for copying out a uint32 to the task's +// memory. +func CopyUint32Out(task marshal.Task, addr usermem.Addr, src uint32) (int, error) { + srcP := Uint32(src) + return srcP.CopyOut(task, addr) +} + +// 64-bit integers + +// CopyInt64In is a convenient wrapper for copying in an int64 from the task's +// memory. +func CopyInt64In(task marshal.Task, addr usermem.Addr, dst *int64) (int, error) { + var buf Int64 + n, err := buf.CopyIn(task, addr) + if err != nil { + return n, err + } + *dst = int64(buf) + return n, nil +} + +// CopyInt64Out is a convenient wrapper for copying out an int64 to the task's +// memory. +func CopyInt64Out(task marshal.Task, addr usermem.Addr, src int64) (int, error) { + srcP := Int64(src) + return srcP.CopyOut(task, addr) +} + +// CopyUint64In is a convenient wrapper for copying in a uint64 from the task's +// memory. +func CopyUint64In(task marshal.Task, addr usermem.Addr, dst *uint64) (int, error) { + var buf Uint64 + n, err := buf.CopyIn(task, addr) + if err != nil { + return n, err + } + *dst = uint64(buf) + return n, nil +} + +// CopyUint64Out is a convenient wrapper for copying out a uint64 to the task's +// memory. +func CopyUint64Out(task marshal.Task, addr usermem.Addr, src uint64) (int, error) { + srcP := Uint64(src) + return srcP.CopyOut(task, addr) +} diff --git a/tools/go_marshal/test/BUILD b/tools/go_marshal/test/BUILD index f27c5ce52..3b839799d 100644 --- a/tools/go_marshal/test/BUILD +++ b/tools/go_marshal/test/BUILD @@ -39,3 +39,17 @@ go_binary( "//tools/go_marshal/marshal", ], ) + +go_test( + name = "marshal_test", + size = "small", + srcs = ["marshal_test.go"], + deps = [ + ":test", + "//pkg/syserror", + "//pkg/usermem", + "//tools/go_marshal/analysis", + "//tools/go_marshal/marshal", + "@com_github_google_go-cmp//cmp:go_default_library", + ], +) diff --git a/tools/go_marshal/test/benchmark_test.go b/tools/go_marshal/test/benchmark_test.go index c79defe9e..224d308c7 100644 --- a/tools/go_marshal/test/benchmark_test.go +++ b/tools/go_marshal/test/benchmark_test.go @@ -176,3 +176,45 @@ func BenchmarkGoMarshalUnsafe(b *testing.B) { panic(fmt.Sprintf("Data corruption across marshal/unmarshal cycle:\nBefore: %+v\nAfter: %+v\n", s1, s2)) } } + +func BenchmarkBinarySlice(b *testing.B) { + var s1, s2 [64]test.Stat + analysis.RandomizeValue(&s1) + + size := binary.Size(s1) + + b.ResetTimer() + + for n := 0; n < b.N; n++ { + buf := make([]byte, 0, size) + buf = binary.Marshal(buf, usermem.ByteOrder, &s1) + binary.Unmarshal(buf, usermem.ByteOrder, &s2) + } + + b.StopTimer() + + // Sanity check, make sure the values were preserved. + if !reflect.DeepEqual(s1, s2) { + panic(fmt.Sprintf("Data corruption across marshal/unmarshal cycle:\nBefore: %+v\nAfter: %+v\n", s1, s2)) + } +} + +func BenchmarkGoMarshalUnsafeSlice(b *testing.B) { + var s1, s2 [64]test.Stat + analysis.RandomizeValue(&s1) + + b.ResetTimer() + + for n := 0; n < b.N; n++ { + buf := make([]byte, (*test.Stat)(nil).SizeBytes()*len(s1)) + test.MarshalUnsafeStatSlice(s1[:], buf) + test.UnmarshalUnsafeStatSlice(s2[:], buf) + } + + b.StopTimer() + + // Sanity check, make sure the values were preserved. + if !reflect.DeepEqual(s1, s2) { + panic(fmt.Sprintf("Data corruption across marshal/unmarshal cycle:\nBefore: %+v\nAfter: %+v\n", s1, s2)) + } +} diff --git a/tools/go_marshal/test/external/external.go b/tools/go_marshal/test/external/external.go index 4be3722f3..26fe8e0c8 100644 --- a/tools/go_marshal/test/external/external.go +++ b/tools/go_marshal/test/external/external.go @@ -21,3 +21,11 @@ package external type External struct { j int64 } + +// NotPacked is an unaligned Marshallable type for use in testing. +// +// +marshal +type NotPacked struct { + a int32 + b byte `marshal:"unaligned"` +} diff --git a/tools/go_marshal/test/marshal_test.go b/tools/go_marshal/test/marshal_test.go new file mode 100644 index 000000000..16829ee45 --- /dev/null +++ b/tools/go_marshal/test/marshal_test.go @@ -0,0 +1,515 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package marshal_test contains manual tests for the marshal interface. These +// are intended to test behaviour not covered by the automatically generated +// tests. +package marshal_test + +import ( + "bytes" + "encoding/binary" + "fmt" + "reflect" + "runtime" + "testing" + "unsafe" + + "github.com/google/go-cmp/cmp" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/tools/go_marshal/analysis" + "gvisor.dev/gvisor/tools/go_marshal/marshal" + "gvisor.dev/gvisor/tools/go_marshal/test" +) + +var simulatedErr error = syserror.EFAULT + +// mockTask implements marshal.Task. +type mockTask struct { + taskMem usermem.BytesIO +} + +// populate fills the task memory with the contents of val. +func (t *mockTask) populate(val interface{}) { + var buf bytes.Buffer + // Use binary.Write so we aren't testing go-marshal against its own + // potentially buggy implementation. + if err := binary.Write(&buf, usermem.ByteOrder, val); err != nil { + panic(err) + } + t.taskMem.Bytes = buf.Bytes() +} + +func (t *mockTask) setLimit(n int) { + if len(t.taskMem.Bytes) < n { + grown := make([]byte, n) + copy(grown, t.taskMem.Bytes) + t.taskMem.Bytes = grown + return + } + t.taskMem.Bytes = t.taskMem.Bytes[:n] +} + +// CopyScratchBuffer implements marshal.Task.CopyScratchBuffer. +func (t *mockTask) CopyScratchBuffer(size int) []byte { + return make([]byte, size) +} + +// CopyOutBytes implements marshal.Task.CopyOutBytes. The implementation +// completely ignores the target address and stores a copy of b in its +// internally buffer, overriding any previous contents. +func (t *mockTask) CopyOutBytes(_ usermem.Addr, b []byte) (int, error) { + return t.taskMem.CopyOut(nil, 0, b, usermem.IOOpts{}) +} + +// CopyInBytes implements marshal.Task.CopyInBytes. The implementation +// completely ignores the source address and always fills b from the begining of +// its internal buffer. +func (t *mockTask) CopyInBytes(_ usermem.Addr, b []byte) (int, error) { + return t.taskMem.CopyIn(nil, 0, b, usermem.IOOpts{}) +} + +// unsafeMemory returns the underlying memory for m. The returned slice is only +// valid for the lifetime for m. The garbage collector isn't aware that the +// returned slice is related to m, the caller must ensure m lives long enough. +func unsafeMemory(m marshal.Marshallable) []byte { + if !m.Packed() { + // We can't return a slice pointing to the underlying memory + // since the layout isn't packed. Allocate a temporary buffer + // and marshal instead. + var buf bytes.Buffer + if err := binary.Write(&buf, usermem.ByteOrder, m); err != nil { + panic(err) + } + return buf.Bytes() + } + + // reflect.ValueOf(m) + // .Elem() // Unwrap interface to inner concrete object + // .Addr() // Pointer value to object + // .Pointer() // Actual address from the pointer value + ptr := reflect.ValueOf(m).Elem().Addr().Pointer() + + size := m.SizeBytes() + + var mem []byte + hdr := (*reflect.SliceHeader)(unsafe.Pointer(&mem)) + hdr.Data = ptr + hdr.Len = size + hdr.Cap = size + + return mem +} + +// unsafeMemorySlice returns the underlying memory for m. The returned slice is +// only valid for the lifetime for m. The garbage collector isn't aware that the +// returned slice is related to m, the caller must ensure m lives long enough. +// +// Precondition: m must be a slice. +func unsafeMemorySlice(m interface{}, elt marshal.Marshallable) []byte { + kind := reflect.TypeOf(m).Kind() + if kind != reflect.Slice { + panic("unsafeMemorySlice called on non-slice") + } + + if !elt.Packed() { + // We can't return a slice pointing to the underlying memory + // since the layout isn't packed. Allocate a temporary buffer + // and marshal instead. + var buf bytes.Buffer + if err := binary.Write(&buf, usermem.ByteOrder, m); err != nil { + panic(err) + } + return buf.Bytes() + } + + v := reflect.ValueOf(m) + length := v.Len() * elt.SizeBytes() + + var mem []byte + hdr := (*reflect.SliceHeader)(unsafe.Pointer(&mem)) + hdr.Data = v.Pointer() // This is a pointer to the first elem for slices. + hdr.Len = length + hdr.Cap = length + + return mem +} + +func isZeroes(buf []byte) bool { + for _, b := range buf { + if b != 0 { + return false + } + } + return true +} + +// compareMemory compares the first n bytes of two chuncks of memory represented +// by expected and actual. +func compareMemory(t *testing.T, expected, actual []byte, n int) { + t.Logf("Expected (%d): %v (%d) + (%d) %v\n", len(expected), expected[:n], n, len(expected)-n, expected[n:]) + t.Logf("Actual (%d): %v (%d) + (%d) %v\n", len(actual), actual[:n], n, len(actual)-n, actual[n:]) + + if diff := cmp.Diff(expected[:n], actual[:n]); diff != "" { + t.Errorf("Memory buffers don't match:\n--- expected only\n+++ actual only\n%v", diff) + } +} + +// limitedCopyIn populates task memory with src, then unmarshals task memory to +// dst. The task signals an error at limit bytes during copy-in, which should +// result in a truncated unmarshalling. +func limitedCopyIn(t *testing.T, src, dst marshal.Marshallable, limit int) { + var task mockTask + task.populate(src) + task.setLimit(limit) + + n, err := dst.CopyIn(&task, usermem.Addr(0)) + if n != limit { + t.Errorf("CopyIn copied unexpected number of bytes, expected %d, got %d", limit, n) + } + if err != simulatedErr { + t.Errorf("CopyIn returned unexpected error, expected %v, got %v", simulatedErr, err) + } + + expectedMem := unsafeMemory(src) + defer runtime.KeepAlive(src) + actualMem := unsafeMemory(dst) + defer runtime.KeepAlive(dst) + + compareMemory(t, expectedMem, actualMem, n) + + // The last n bytes should be zero for actual, since actual was + // zero-initialized, and CopyIn shouldn't have touched those bytes. However + // we can only guarantee we didn't touch anything in the last n bytes if the + // layout is packed. + if dst.Packed() && !isZeroes(actualMem[n:]) { + t.Errorf("Expected the last %d bytes of copied in object to be zeroes, got %v\n", dst.SizeBytes()-n, actualMem) + } +} + +// limitedCopyOut marshals src to task memory. The task signals an error at +// limit bytes during copy-out, which should result in a truncated marshalling. +func limitedCopyOut(t *testing.T, src marshal.Marshallable, limit int) { + var task mockTask + task.setLimit(limit) + + n, err := src.CopyOut(&task, usermem.Addr(0)) + if n != limit { + t.Errorf("CopyOut copied unexpected number of bytes, expected %d, got %d", limit, n) + } + if err != simulatedErr { + t.Errorf("CopyOut returned unexpected error, expected %v, got %v", simulatedErr, err) + } + + expectedMem := unsafeMemory(src) + defer runtime.KeepAlive(src) + actualMem := task.taskMem.Bytes + + compareMemory(t, expectedMem, actualMem, n) +} + +// copyOutN marshals src to task memory, requesting the marshalling to be +// limited to limit bytes. +func copyOutN(t *testing.T, src marshal.Marshallable, limit int) { + var task mockTask + task.setLimit(limit) + + n, err := src.CopyOutN(&task, usermem.Addr(0), limit) + if err != nil { + t.Errorf("CopyOut returned unexpected error: %v", err) + } + if n != limit { + t.Errorf("CopyOut copied unexpected number of bytes, expected %d, got %d", limit, n) + } + + expectedMem := unsafeMemory(src) + defer runtime.KeepAlive(src) + actualMem := task.taskMem.Bytes + + t.Logf("Expected: %v + %v\n", expectedMem[:n], expectedMem[n:]) + t.Logf("Actual : %v + %v\n", actualMem[:n], actualMem[n:]) + + compareMemory(t, expectedMem, actualMem, n) +} + +// TestLimitedMarshalling verifies marshalling/unmarshalling succeeds when the +// underyling copy in/out operations partially succeed. +func TestLimitedMarshalling(t *testing.T) { + types := []reflect.Type{ + // Packed types. + reflect.TypeOf((*test.Type2)(nil)), + reflect.TypeOf((*test.Type3)(nil)), + reflect.TypeOf((*test.Timespec)(nil)), + reflect.TypeOf((*test.Stat)(nil)), + reflect.TypeOf((*test.InetAddr)(nil)), + reflect.TypeOf((*test.SignalSet)(nil)), + reflect.TypeOf((*test.SignalSetAlias)(nil)), + // Non-packed types. + reflect.TypeOf((*test.Type1)(nil)), + reflect.TypeOf((*test.Type4)(nil)), + reflect.TypeOf((*test.Type5)(nil)), + reflect.TypeOf((*test.Type6)(nil)), + reflect.TypeOf((*test.Type7)(nil)), + reflect.TypeOf((*test.Type8)(nil)), + } + + for _, tyPtr := range types { + // Remove one level of pointer-indirection from the type. We get this + // back when we pass the type to reflect.New. + ty := tyPtr.Elem() + + // Partial copy-in. + t.Run(fmt.Sprintf("PartialCopyIn_%v", ty), func(t *testing.T) { + expected := reflect.New(ty).Interface().(marshal.Marshallable) + actual := reflect.New(ty).Interface().(marshal.Marshallable) + analysis.RandomizeValue(expected) + + limitedCopyIn(t, expected, actual, expected.SizeBytes()/2) + }) + + // Partial copy-out. + t.Run(fmt.Sprintf("PartialCopyOut_%v", ty), func(t *testing.T) { + expected := reflect.New(ty).Interface().(marshal.Marshallable) + analysis.RandomizeValue(expected) + + limitedCopyOut(t, expected, expected.SizeBytes()/2) + }) + + // Explicitly request partial copy-out. + t.Run(fmt.Sprintf("PartialCopyOutN_%v", ty), func(t *testing.T) { + expected := reflect.New(ty).Interface().(marshal.Marshallable) + analysis.RandomizeValue(expected) + + copyOutN(t, expected, expected.SizeBytes()/2) + }) + } +} + +// TestLimitedMarshalling verifies marshalling/unmarshalling of slices of +// marshallable types succeed when the underyling copy in/out operations +// partially succeed. +func TestLimitedSliceMarshalling(t *testing.T) { + types := []struct { + arrayPtrType reflect.Type + copySliceIn func(task marshal.Task, addr usermem.Addr, dstSlice interface{}) (int, error) + copySliceOut func(task marshal.Task, addr usermem.Addr, srcSlice interface{}) (int, error) + unsafeMemory func(arrPtr interface{}) []byte + }{ + // Packed types. + { + reflect.TypeOf((*[20]test.Stat)(nil)), + func(task marshal.Task, addr usermem.Addr, dst interface{}) (int, error) { + slice := dst.(*[20]test.Stat)[:] + return test.CopyStatSliceIn(task, addr, slice) + }, + func(task marshal.Task, addr usermem.Addr, src interface{}) (int, error) { + slice := src.(*[20]test.Stat)[:] + return test.CopyStatSliceOut(task, addr, slice) + }, + func(a interface{}) []byte { + slice := a.(*[20]test.Stat)[:] + return unsafeMemorySlice(slice, &slice[0]) + }, + }, + { + reflect.TypeOf((*[1]test.Stat)(nil)), + func(task marshal.Task, addr usermem.Addr, dst interface{}) (int, error) { + slice := dst.(*[1]test.Stat)[:] + return test.CopyStatSliceIn(task, addr, slice) + }, + func(task marshal.Task, addr usermem.Addr, src interface{}) (int, error) { + slice := src.(*[1]test.Stat)[:] + return test.CopyStatSliceOut(task, addr, slice) + }, + func(a interface{}) []byte { + slice := a.(*[1]test.Stat)[:] + return unsafeMemorySlice(slice, &slice[0]) + }, + }, + { + reflect.TypeOf((*[5]test.SignalSetAlias)(nil)), + func(task marshal.Task, addr usermem.Addr, dst interface{}) (int, error) { + slice := dst.(*[5]test.SignalSetAlias)[:] + return test.CopySignalSetAliasSliceIn(task, addr, slice) + }, + func(task marshal.Task, addr usermem.Addr, src interface{}) (int, error) { + slice := src.(*[5]test.SignalSetAlias)[:] + return test.CopySignalSetAliasSliceOut(task, addr, slice) + }, + func(a interface{}) []byte { + slice := a.(*[5]test.SignalSetAlias)[:] + return unsafeMemorySlice(slice, &slice[0]) + }, + }, + // Non-packed types. + { + reflect.TypeOf((*[20]test.Type1)(nil)), + func(task marshal.Task, addr usermem.Addr, dst interface{}) (int, error) { + slice := dst.(*[20]test.Type1)[:] + return test.CopyType1SliceIn(task, addr, slice) + }, + func(task marshal.Task, addr usermem.Addr, src interface{}) (int, error) { + slice := src.(*[20]test.Type1)[:] + return test.CopyType1SliceOut(task, addr, slice) + }, + func(a interface{}) []byte { + slice := a.(*[20]test.Type1)[:] + return unsafeMemorySlice(slice, &slice[0]) + }, + }, + { + reflect.TypeOf((*[1]test.Type1)(nil)), + func(task marshal.Task, addr usermem.Addr, dst interface{}) (int, error) { + slice := dst.(*[1]test.Type1)[:] + return test.CopyType1SliceIn(task, addr, slice) + }, + func(task marshal.Task, addr usermem.Addr, src interface{}) (int, error) { + slice := src.(*[1]test.Type1)[:] + return test.CopyType1SliceOut(task, addr, slice) + }, + func(a interface{}) []byte { + slice := a.(*[1]test.Type1)[:] + return unsafeMemorySlice(slice, &slice[0]) + }, + }, + { + reflect.TypeOf((*[7]test.Type8)(nil)), + func(task marshal.Task, addr usermem.Addr, dst interface{}) (int, error) { + slice := dst.(*[7]test.Type8)[:] + return test.CopyType8SliceIn(task, addr, slice) + }, + func(task marshal.Task, addr usermem.Addr, src interface{}) (int, error) { + slice := src.(*[7]test.Type8)[:] + return test.CopyType8SliceOut(task, addr, slice) + }, + func(a interface{}) []byte { + slice := a.(*[7]test.Type8)[:] + return unsafeMemorySlice(slice, &slice[0]) + }, + }, + } + + for _, tt := range types { + // The body of this loop is generic over the type tt.arrayPtrType, with + // the help of reflection. To aid in readability, comments below show + // the equivalent go code assuming + // tt.arrayPtrType = typeof(*[20]test.Stat). + + // Equivalent: + // var x *[20]test.Stat + // arrayTy := reflect.TypeOf(*x) + arrayTy := tt.arrayPtrType.Elem() + + // Partial copy-in of slices. + t.Run(fmt.Sprintf("PartialCopySliceIn_%v", arrayTy), func(t *testing.T) { + // Equivalent: + // var x [20]test.Stat + // length := len(x) + length := arrayTy.Len() + if length < 1 { + panic("Test type can't be zero-length array") + } + // Equivalent: + // elem := new(test.Stat).(marshal.Marshallable) + elem := reflect.New(arrayTy.Elem()).Interface().(marshal.Marshallable) + + // Equivalent: + // var expected, actual interface{} + // expected = new([20]test.Stat) + // actual = new([20]test.Stat) + expected := reflect.New(arrayTy).Interface() + actual := reflect.New(arrayTy).Interface() + + analysis.RandomizeValue(expected) + + limit := (length * elem.SizeBytes()) / 2 + // Also make sure the limit is partially inside one of the elements. + limit += elem.SizeBytes() / 2 + analysis.RandomizeValue(expected) + + var task mockTask + task.populate(expected) + task.setLimit(limit) + + n, err := tt.copySliceIn(&task, usermem.Addr(0), actual) + if n != limit { + t.Errorf("CopyIn copied unexpected number of bytes, expected %d, got %d", limit, n) + } + if n < length*elem.SizeBytes() && err != simulatedErr { + t.Errorf("CopyIn returned unexpected error, expected %v, got %v", simulatedErr, err) + } + + expectedMem := tt.unsafeMemory(expected) + defer runtime.KeepAlive(expected) + actualMem := tt.unsafeMemory(actual) + defer runtime.KeepAlive(actual) + + compareMemory(t, expectedMem, actualMem, n) + + // The last n bytes should be zero for actual, since actual was + // zero-initialized, and CopyIn shouldn't have touched those bytes. However + // we can only guarantee we didn't touch anything in the last n bytes if the + // layout is packed. + if elem.Packed() && !isZeroes(actualMem[n:]) { + t.Errorf("Expected the last %d bytes of copied in object to be zeroes, got %v\n", (elem.SizeBytes()*length)-n, actualMem) + } + }) + + // Partial copy-out of slices. + t.Run(fmt.Sprintf("PartialCopySliceOut_%v", arrayTy), func(t *testing.T) { + // Equivalent: + // var x [20]test.Stat + // length := len(x) + length := arrayTy.Len() + if length < 1 { + panic("Test type can't be zero-length array") + } + // Equivalent: + // elem := new(test.Stat).(marshal.Marshallable) + elem := reflect.New(arrayTy.Elem()).Interface().(marshal.Marshallable) + + // Equivalent: + // var expected, actual interface{} + // expected = new([20]test.Stat) + // actual = new([20]test.Stat) + expected := reflect.New(arrayTy).Interface() + + analysis.RandomizeValue(expected) + + limit := (length * elem.SizeBytes()) / 2 + // Also make sure the limit is partially inside one of the elements. + limit += elem.SizeBytes() / 2 + analysis.RandomizeValue(expected) + + var task mockTask + task.populate(expected) + task.setLimit(limit) + + n, err := tt.copySliceOut(&task, usermem.Addr(0), expected) + if n != limit { + t.Errorf("CopyIn copied unexpected number of bytes, expected %d, got %d", limit, n) + } + if n < length*elem.SizeBytes() && err != simulatedErr { + t.Errorf("CopyIn returned unexpected error, expected %v, got %v", simulatedErr, err) + } + + expectedMem := tt.unsafeMemory(expected) + defer runtime.KeepAlive(expected) + actualMem := task.taskMem.Bytes + + compareMemory(t, expectedMem, actualMem, n) + }) + } +} diff --git a/tools/go_marshal/test/test.go b/tools/go_marshal/test/test.go index c829db6da..43df73545 100644 --- a/tools/go_marshal/test/test.go +++ b/tools/go_marshal/test/test.go @@ -23,7 +23,7 @@ import ( // Type1 is a test data type. // -// +marshal +// +marshal slice:Type1Slice type Type1 struct { a Type2 x, y int64 // Multiple field names. @@ -75,6 +75,34 @@ type Type5 struct { m int64 } +// Type6 is a test data type ends mid-word. +// +// +marshal +type Type6 struct { + a int64 + b int64 + // If c isn't marked unaligned, analysis fails (as it should, since + // the unsafe API corrupts Type7). + c byte `marshal:"unaligned"` +} + +// Type7 is a test data type that contains a child struct that ends +// mid-word. +// +marshal +type Type7 struct { + x Type6 + y int64 +} + +// Type8 is a test data type which contains an external non-packed field. +// +// +marshal slice:Type8Slice +type Type8 struct { + a int64 + np ex.NotPacked + b int64 +} + // Timespec represents struct timespec in . // // +marshal @@ -85,7 +113,7 @@ type Timespec struct { // Stat represents struct stat. // -// +marshal +// +marshal slice:StatSlice type Stat struct { Dev uint64 Ino uint64 @@ -111,10 +139,10 @@ type InetAddr [4]byte // SignalSet is an example marshallable newtype on a primitive. // -// +marshal +// +marshal slice:SignalSetSlice:inner type SignalSet uint64 // SignalSetAlias is an example newtype on another marshallable type. // -// +marshal +// +marshal slice:SignalSetAliasSlice type SignalSetAlias SignalSet -- cgit v1.2.3 From d01a8ca3473bfbc7a5eb8da5ea662925e5f2673d Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Wed, 1 Apr 2020 07:56:05 -0700 Subject: Add FileDescription interface for socket files. Refactor the existing socket interface to share methods between VFS1 and VFS2. The method signatures do not contain anything filesystem-related, so they don't need to be re-defined for VFS2. Updates #1476, #1478, #1484, #1485. PiperOrigin-RevId: 304184545 --- pkg/sentry/socket/BUILD | 1 + pkg/sentry/socket/socket.go | 87 +++++++++++++++++++++++++++++++++++++++++++-- 2 files changed, 86 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/BUILD b/pkg/sentry/socket/BUILD index 611fa22c3..c40c6d673 100644 --- a/pkg/sentry/socket/BUILD +++ b/pkg/sentry/socket/BUILD @@ -16,6 +16,7 @@ go_library( "//pkg/sentry/kernel", "//pkg/sentry/kernel/time", "//pkg/sentry/socket/unix/transport", + "//pkg/sentry/vfs", "//pkg/syserr", "//pkg/tcpip", "//pkg/usermem", diff --git a/pkg/sentry/socket/socket.go b/pkg/sentry/socket/socket.go index 50d9744e6..b5ba4a56b 100644 --- a/pkg/sentry/socket/socket.go +++ b/pkg/sentry/socket/socket.go @@ -31,6 +31,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" + "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/usermem" @@ -48,11 +49,25 @@ func (c *ControlMessages) Release() { c.Unix.Release() } -// Socket is the interface containing socket syscalls used by the syscall layer -// to redirect them to the appropriate implementation. +// Socket is an interface combining fs.FileOperations and SocketOps, +// representing a VFS1 socket file. type Socket interface { fs.FileOperations + SocketOps +} + +// SocketVFS2 is an interface combining vfs.FileDescription and SocketOps, +// representing a VFS2 socket file. +type SocketVFS2 interface { + vfs.FileDescriptionImpl + SocketOps +} +// SocketOps is the interface containing socket syscalls used by the syscall +// layer to redirect them to the appropriate implementation. +// +// It is implemented by both Socket and SocketVFS2. +type SocketOps interface { // Connect implements the connect(2) linux syscall. Connect(t *kernel.Task, sockaddr []byte, blocking bool) *syserr.Error @@ -153,6 +168,8 @@ var families = make(map[int][]Provider) // RegisterProvider registers the provider of a given address family so that // sockets of that type can be created via socket() and/or socketpair() // syscalls. +// +// This should only be called during the initialization of the address family. func RegisterProvider(family int, provider Provider) { families[family] = append(families[family], provider) } @@ -216,6 +233,72 @@ func NewDirent(ctx context.Context, d *device.Device) *fs.Dirent { return fs.NewDirent(ctx, inode, fmt.Sprintf("socket:[%d]", ino)) } +// ProviderVFS2 is the vfs2 interface implemented by providers of sockets for +// specific address families (e.g., AF_INET). +type ProviderVFS2 interface { + // Socket creates a new socket. + // + // If a nil Socket _and_ a nil error is returned, it means that the + // protocol is not supported. A non-nil error should only be returned + // if the protocol is supported, but an error occurs during creation. + Socket(t *kernel.Task, stype linux.SockType, protocol int) (*vfs.FileDescription, *syserr.Error) + + // Pair creates a pair of connected sockets. + // + // See Socket for error information. + Pair(t *kernel.Task, stype linux.SockType, protocol int) (*vfs.FileDescription, *vfs.FileDescription, *syserr.Error) +} + +// familiesVFS2 holds a map of all known address families and their providers. +var familiesVFS2 = make(map[int][]ProviderVFS2) + +// RegisterProviderVFS2 registers the provider of a given address family so that +// sockets of that type can be created via socket() and/or socketpair() +// syscalls. +// +// This should only be called during the initialization of the address family. +func RegisterProviderVFS2(family int, provider ProviderVFS2) { + familiesVFS2[family] = append(familiesVFS2[family], provider) +} + +// NewVFS2 creates a new socket with the given family, type and protocol. +func NewVFS2(t *kernel.Task, family int, stype linux.SockType, protocol int) (*vfs.FileDescription, *syserr.Error) { + for _, p := range familiesVFS2[family] { + s, err := p.Socket(t, stype, protocol) + if err != nil { + return nil, err + } + if s != nil { + // TODO: Add vfs2 sockets to global socket table. + return s, nil + } + } + + return nil, syserr.ErrAddressFamilyNotSupported +} + +// PairVFS2 creates a new connected socket pair with the given family, type and +// protocol. +func PairVFS2(t *kernel.Task, family int, stype linux.SockType, protocol int) (*vfs.FileDescription, *vfs.FileDescription, *syserr.Error) { + providers, ok := familiesVFS2[family] + if !ok { + return nil, nil, syserr.ErrAddressFamilyNotSupported + } + + for _, p := range providers { + s1, s2, err := p.Pair(t, stype, protocol) + if err != nil { + return nil, nil, err + } + if s1 != nil && s2 != nil { + // TODO: Add vfs2 sockets to global socket table. + return s1, s2, nil + } + } + + return nil, nil, syserr.ErrSocketNotSupported +} + // SendReceiveTimeout stores timeouts for send and receive calls. // // It is meant to be embedded into Socket implementations to help satisfy the -- cgit v1.2.3 From db7917556a7e4bd2cd6d183c68f04a4787dec493 Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Wed, 1 Apr 2020 09:58:51 -0700 Subject: Fix 386 build tags The build tag for 32-bit x86 is 386, not i386. Updates #2298 PiperOrigin-RevId: 304206373 --- pkg/cpuid/cpuid_parse_x86_test.go | 2 +- pkg/cpuid/cpuid_x86.go | 2 +- pkg/cpuid/cpuid_x86_test.go | 2 +- pkg/sentry/arch/arch_state_x86.go | 2 +- pkg/sentry/arch/arch_x86.go | 2 +- pkg/sentry/arch/arch_x86_impl.go | 2 +- pkg/sentry/arch/signal_stack.go | 2 +- pkg/sentry/platform/ring0/pagetables/pagetables_x86.go | 2 +- pkg/sentry/platform/ring0/x86.go | 2 +- pkg/usermem/usermem_x86.go | 2 +- 10 files changed, 10 insertions(+), 10 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/cpuid/cpuid_parse_x86_test.go b/pkg/cpuid/cpuid_parse_x86_test.go index d48418e69..c9bd40e1b 100644 --- a/pkg/cpuid/cpuid_parse_x86_test.go +++ b/pkg/cpuid/cpuid_parse_x86_test.go @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -// +build i386 amd64 +// +build 386 amd64 package cpuid diff --git a/pkg/cpuid/cpuid_x86.go b/pkg/cpuid/cpuid_x86.go index 9abf6914d..562f8f405 100644 --- a/pkg/cpuid/cpuid_x86.go +++ b/pkg/cpuid/cpuid_x86.go @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -// +build i386 amd64 +// +build 386 amd64 package cpuid diff --git a/pkg/cpuid/cpuid_x86_test.go b/pkg/cpuid/cpuid_x86_test.go index 0fe20c213..bacf345c8 100644 --- a/pkg/cpuid/cpuid_x86_test.go +++ b/pkg/cpuid/cpuid_x86_test.go @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -// +build i386 amd64 +// +build 386 amd64 package cpuid diff --git a/pkg/sentry/arch/arch_state_x86.go b/pkg/sentry/arch/arch_state_x86.go index e35c9214a..aa31169e0 100644 --- a/pkg/sentry/arch/arch_state_x86.go +++ b/pkg/sentry/arch/arch_state_x86.go @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -// +build amd64 i386 +// +build amd64 386 package arch diff --git a/pkg/sentry/arch/arch_x86.go b/pkg/sentry/arch/arch_x86.go index 88b40a9d1..7fc4c0473 100644 --- a/pkg/sentry/arch/arch_x86.go +++ b/pkg/sentry/arch/arch_x86.go @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -// +build amd64 i386 +// +build amd64 386 package arch diff --git a/pkg/sentry/arch/arch_x86_impl.go b/pkg/sentry/arch/arch_x86_impl.go index 04ac283c6..3edf40764 100644 --- a/pkg/sentry/arch/arch_x86_impl.go +++ b/pkg/sentry/arch/arch_x86_impl.go @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -// +build amd64 i386 +// +build amd64 386 package arch diff --git a/pkg/sentry/arch/signal_stack.go b/pkg/sentry/arch/signal_stack.go index 1a6056171..e58f055c7 100644 --- a/pkg/sentry/arch/signal_stack.go +++ b/pkg/sentry/arch/signal_stack.go @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -// +build i386 amd64 arm64 +// +build 386 amd64 arm64 package arch diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_x86.go b/pkg/sentry/platform/ring0/pagetables/pagetables_x86.go index dcf061df9..157438d9b 100644 --- a/pkg/sentry/platform/ring0/pagetables/pagetables_x86.go +++ b/pkg/sentry/platform/ring0/pagetables/pagetables_x86.go @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -// +build i386 amd64 +// +build 386 amd64 package pagetables diff --git a/pkg/sentry/platform/ring0/x86.go b/pkg/sentry/platform/ring0/x86.go index 5f80d64e8..9da0ea685 100644 --- a/pkg/sentry/platform/ring0/x86.go +++ b/pkg/sentry/platform/ring0/x86.go @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -// +build i386 amd64 +// +build 386 amd64 package ring0 diff --git a/pkg/usermem/usermem_x86.go b/pkg/usermem/usermem_x86.go index 8059b72d2..d96f829fb 100644 --- a/pkg/usermem/usermem_x86.go +++ b/pkg/usermem/usermem_x86.go @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -// +build amd64 i386 +// +build amd64 386 package usermem -- cgit v1.2.3 From 4e6a1a5adb5607423c180089d8b464ef7dfdd1ae Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Wed, 1 Apr 2020 11:05:05 -0700 Subject: Automated rollback of changelist 303799678 PiperOrigin-RevId: 304221302 --- pkg/atomicbitops/atomicbitops_amd64.s | 16 ++-- pkg/atomicbitops/atomicbitops_arm64.s | 16 ++-- pkg/atomicbitops/atomicbitops_noasm.go | 8 -- pkg/safecopy/safecopy.go | 4 +- pkg/safecopy/safecopy_unsafe.go | 6 +- pkg/sentry/platform/kvm/BUILD | 1 - pkg/sentry/platform/kvm/bluepill.go | 12 +-- pkg/sentry/platform/kvm/bluepill_unsafe.go | 97 +++++++++---------------- pkg/sentry/platform/kvm/kvm_const.go | 1 - pkg/sentry/platform/kvm/kvm_test.go | 64 ++-------------- pkg/sentry/platform/kvm/machine.go | 9 ++- pkg/sentry/platform/kvm/machine_amd64_unsafe.go | 25 +++++++ pkg/sentry/platform/kvm/machine_arm64_unsafe.go | 26 +++++++ pkg/sentry/platform/kvm/machine_unsafe.go | 41 ----------- runsc/sandbox/sandbox.go | 6 ++ 15 files changed, 125 insertions(+), 207 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/atomicbitops/atomicbitops_amd64.s b/pkg/atomicbitops/atomicbitops_amd64.s index f0edd4de7..54c887ee5 100644 --- a/pkg/atomicbitops/atomicbitops_amd64.s +++ b/pkg/atomicbitops/atomicbitops_amd64.s @@ -16,28 +16,28 @@ #include "textflag.h" -TEXT ·AndUint32(SB),NOSPLIT,$0-12 +TEXT ·AndUint32(SB),$0-12 MOVQ addr+0(FP), BP MOVL val+8(FP), AX LOCK ANDL AX, 0(BP) RET -TEXT ·OrUint32(SB),NOSPLIT,$0-12 +TEXT ·OrUint32(SB),$0-12 MOVQ addr+0(FP), BP MOVL val+8(FP), AX LOCK ORL AX, 0(BP) RET -TEXT ·XorUint32(SB),NOSPLIT,$0-12 +TEXT ·XorUint32(SB),$0-12 MOVQ addr+0(FP), BP MOVL val+8(FP), AX LOCK XORL AX, 0(BP) RET -TEXT ·CompareAndSwapUint32(SB),NOSPLIT,$0-20 +TEXT ·CompareAndSwapUint32(SB),$0-20 MOVQ addr+0(FP), DI MOVL old+8(FP), AX MOVL new+12(FP), DX @@ -46,28 +46,28 @@ TEXT ·CompareAndSwapUint32(SB),NOSPLIT,$0-20 MOVL AX, ret+16(FP) RET -TEXT ·AndUint64(SB),NOSPLIT,$0-16 +TEXT ·AndUint64(SB),$0-16 MOVQ addr+0(FP), BP MOVQ val+8(FP), AX LOCK ANDQ AX, 0(BP) RET -TEXT ·OrUint64(SB),NOSPLIT,$0-16 +TEXT ·OrUint64(SB),$0-16 MOVQ addr+0(FP), BP MOVQ val+8(FP), AX LOCK ORQ AX, 0(BP) RET -TEXT ·XorUint64(SB),NOSPLIT,$0-16 +TEXT ·XorUint64(SB),$0-16 MOVQ addr+0(FP), BP MOVQ val+8(FP), AX LOCK XORQ AX, 0(BP) RET -TEXT ·CompareAndSwapUint64(SB),NOSPLIT,$0-32 +TEXT ·CompareAndSwapUint64(SB),$0-32 MOVQ addr+0(FP), DI MOVQ old+8(FP), AX MOVQ new+16(FP), DX diff --git a/pkg/atomicbitops/atomicbitops_arm64.s b/pkg/atomicbitops/atomicbitops_arm64.s index 644a6bca5..5c780851b 100644 --- a/pkg/atomicbitops/atomicbitops_arm64.s +++ b/pkg/atomicbitops/atomicbitops_arm64.s @@ -16,7 +16,7 @@ #include "textflag.h" -TEXT ·AndUint32(SB),NOSPLIT,$0-12 +TEXT ·AndUint32(SB),$0-12 MOVD ptr+0(FP), R0 MOVW val+8(FP), R1 again: @@ -26,7 +26,7 @@ again: CBNZ R3, again RET -TEXT ·OrUint32(SB),NOSPLIT,$0-12 +TEXT ·OrUint32(SB),$0-12 MOVD ptr+0(FP), R0 MOVW val+8(FP), R1 again: @@ -36,7 +36,7 @@ again: CBNZ R3, again RET -TEXT ·XorUint32(SB),NOSPLIT,$0-12 +TEXT ·XorUint32(SB),$0-12 MOVD ptr+0(FP), R0 MOVW val+8(FP), R1 again: @@ -46,7 +46,7 @@ again: CBNZ R3, again RET -TEXT ·CompareAndSwapUint32(SB),NOSPLIT,$0-20 +TEXT ·CompareAndSwapUint32(SB),$0-20 MOVD addr+0(FP), R0 MOVW old+8(FP), R1 MOVW new+12(FP), R2 @@ -60,7 +60,7 @@ done: MOVW R3, prev+16(FP) RET -TEXT ·AndUint64(SB),NOSPLIT,$0-16 +TEXT ·AndUint64(SB),$0-16 MOVD ptr+0(FP), R0 MOVD val+8(FP), R1 again: @@ -70,7 +70,7 @@ again: CBNZ R3, again RET -TEXT ·OrUint64(SB),NOSPLIT,$0-16 +TEXT ·OrUint64(SB),$0-16 MOVD ptr+0(FP), R0 MOVD val+8(FP), R1 again: @@ -80,7 +80,7 @@ again: CBNZ R3, again RET -TEXT ·XorUint64(SB),NOSPLIT,$0-16 +TEXT ·XorUint64(SB),$0-16 MOVD ptr+0(FP), R0 MOVD val+8(FP), R1 again: @@ -90,7 +90,7 @@ again: CBNZ R3, again RET -TEXT ·CompareAndSwapUint64(SB),NOSPLIT,$0-32 +TEXT ·CompareAndSwapUint64(SB),$0-32 MOVD addr+0(FP), R0 MOVD old+8(FP), R1 MOVD new+16(FP), R2 diff --git a/pkg/atomicbitops/atomicbitops_noasm.go b/pkg/atomicbitops/atomicbitops_noasm.go index 4e9c27b98..3b2898256 100644 --- a/pkg/atomicbitops/atomicbitops_noasm.go +++ b/pkg/atomicbitops/atomicbitops_noasm.go @@ -20,7 +20,6 @@ import ( "sync/atomic" ) -//go:nosplit func AndUint32(addr *uint32, val uint32) { for { o := atomic.LoadUint32(addr) @@ -31,7 +30,6 @@ func AndUint32(addr *uint32, val uint32) { } } -//go:nosplit func OrUint32(addr *uint32, val uint32) { for { o := atomic.LoadUint32(addr) @@ -42,7 +40,6 @@ func OrUint32(addr *uint32, val uint32) { } } -//go:nosplit func XorUint32(addr *uint32, val uint32) { for { o := atomic.LoadUint32(addr) @@ -53,7 +50,6 @@ func XorUint32(addr *uint32, val uint32) { } } -//go:nosplit func CompareAndSwapUint32(addr *uint32, old, new uint32) (prev uint32) { for { prev = atomic.LoadUint32(addr) @@ -66,7 +62,6 @@ func CompareAndSwapUint32(addr *uint32, old, new uint32) (prev uint32) { } } -//go:nosplit func AndUint64(addr *uint64, val uint64) { for { o := atomic.LoadUint64(addr) @@ -77,7 +72,6 @@ func AndUint64(addr *uint64, val uint64) { } } -//go:nosplit func OrUint64(addr *uint64, val uint64) { for { o := atomic.LoadUint64(addr) @@ -88,7 +82,6 @@ func OrUint64(addr *uint64, val uint64) { } } -//go:nosplit func XorUint64(addr *uint64, val uint64) { for { o := atomic.LoadUint64(addr) @@ -99,7 +92,6 @@ func XorUint64(addr *uint64, val uint64) { } } -//go:nosplit func CompareAndSwapUint64(addr *uint64, old, new uint64) (prev uint64) { for { prev = atomic.LoadUint64(addr) diff --git a/pkg/safecopy/safecopy.go b/pkg/safecopy/safecopy.go index 521f1a82d..2fb7e5809 100644 --- a/pkg/safecopy/safecopy.go +++ b/pkg/safecopy/safecopy.go @@ -127,10 +127,10 @@ func initializeAddresses() { func init() { initializeAddresses() - if err := ReplaceSignalHandler(syscall.SIGSEGV, reflect.ValueOf(signalHandler).Pointer(), &savedSigSegVHandler, 0); err != nil { + if err := ReplaceSignalHandler(syscall.SIGSEGV, reflect.ValueOf(signalHandler).Pointer(), &savedSigSegVHandler); err != nil { panic(fmt.Sprintf("Unable to set handler for SIGSEGV: %v", err)) } - if err := ReplaceSignalHandler(syscall.SIGBUS, reflect.ValueOf(signalHandler).Pointer(), &savedSigBusHandler, 0); err != nil { + if err := ReplaceSignalHandler(syscall.SIGBUS, reflect.ValueOf(signalHandler).Pointer(), &savedSigBusHandler); err != nil { panic(fmt.Sprintf("Unable to set handler for SIGBUS: %v", err)) } syserror.AddErrorUnwrapper(func(e error) (syscall.Errno, bool) { diff --git a/pkg/safecopy/safecopy_unsafe.go b/pkg/safecopy/safecopy_unsafe.go index b15b920fe..41dd567f3 100644 --- a/pkg/safecopy/safecopy_unsafe.go +++ b/pkg/safecopy/safecopy_unsafe.go @@ -324,13 +324,11 @@ func errorFromFaultSignal(addr uintptr, sig int32) error { // // It stores the value of the previously set handler in previous. // -// The extraMask parameter is OR'ed into the existing signal handler mask. -// // This function will be called on initialization in order to install safecopy // handlers for appropriate signals. These handlers will call the previous // handler however, and if this is function is being used externally then the // same courtesy is expected. -func ReplaceSignalHandler(sig syscall.Signal, handler uintptr, previous *uintptr, extraMask uint64) error { +func ReplaceSignalHandler(sig syscall.Signal, handler uintptr, previous *uintptr) error { var sa struct { handler uintptr flags uint64 @@ -350,10 +348,10 @@ func ReplaceSignalHandler(sig syscall.Signal, handler uintptr, previous *uintptr if sa.handler == 0 { return fmt.Errorf("previous handler for signal %x isn't set", sig) } + *previous = sa.handler // Install our own handler. - sa.mask |= extraMask sa.handler = handler if _, _, e := syscall.RawSyscall6(syscall.SYS_RT_SIGACTION, uintptr(sig), uintptr(unsafe.Pointer(&sa)), 0, maskLen, 0, 0); e != 0 { return e diff --git a/pkg/sentry/platform/kvm/BUILD b/pkg/sentry/platform/kvm/BUILD index e27f57536..159f7eafd 100644 --- a/pkg/sentry/platform/kvm/BUILD +++ b/pkg/sentry/platform/kvm/BUILD @@ -70,7 +70,6 @@ go_test( "requires-kvm", ], deps = [ - "//pkg/procid", "//pkg/sentry/arch", "//pkg/sentry/platform", "//pkg/sentry/platform/kvm/testutil", diff --git a/pkg/sentry/platform/kvm/bluepill.go b/pkg/sentry/platform/kvm/bluepill.go index 555b5fa96..4b23f7803 100644 --- a/pkg/sentry/platform/kvm/bluepill.go +++ b/pkg/sentry/platform/kvm/bluepill.go @@ -46,14 +46,6 @@ var ( // bounceSignalMask has only bounceSignal set. bounceSignalMask = uint64(1 << (uint64(bounceSignal) - 1)) - // otherSignalsMask includes all other signals that will be cause the - // vCPU to exit during execution. - // - // Currently, this includes the preemption signal and the profiling - // signal. In general, these should be signals whose delivery actually - // influences the way the program executes as the switch can be costly. - otherSignalsMask = uint64(1<<(uint64(syscall.SIGURG)-1)) | uint64(1<<(uint64(syscall.SIGPROF)-1)) - // bounce is the interrupt vector used to return to the kernel. bounce = uint32(ring0.VirtualizationException) @@ -94,8 +86,8 @@ func (c *vCPU) die(context *arch.SignalContext64, msg string) { } func init() { - // Install the handler, masking all signals. - if err := safecopy.ReplaceSignalHandler(bluepillSignal, reflect.ValueOf(sighandler).Pointer(), &savedHandler, ^uint64(0)); err != nil { + // Install the handler. + if err := safecopy.ReplaceSignalHandler(bluepillSignal, reflect.ValueOf(sighandler).Pointer(), &savedHandler); err != nil { panic(fmt.Sprintf("Unable to set handler for signal %d: %v", bluepillSignal, err)) } diff --git a/pkg/sentry/platform/kvm/bluepill_unsafe.go b/pkg/sentry/platform/kvm/bluepill_unsafe.go index 4e9d80765..9add7c944 100644 --- a/pkg/sentry/platform/kvm/bluepill_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_unsafe.go @@ -24,7 +24,6 @@ import ( "syscall" "unsafe" - "gvisor.dev/gvisor/pkg/atomicbitops" "gvisor.dev/gvisor/pkg/sentry/arch" ) @@ -59,19 +58,6 @@ func bluepillArchContext(context unsafe.Pointer) *arch.SignalContext64 { return &((*arch.UContext64)(context).MContext) } -// injectInterrupt is a helper to inject an interrupt. -// -//go:nosplit -func injectInterrupt(c *vCPU) { - if _, _, errno := syscall.RawSyscall( - syscall.SYS_IOCTL, - uintptr(c.fd), - _KVM_INTERRUPT, - uintptr(unsafe.Pointer(&bounce))); errno != 0 { - throw("interrupt injection failed") - } -} - // bluepillHandler is called from the signal stub. // // The world may be stopped while this is executing, and it executes on the @@ -83,9 +69,6 @@ func bluepillHandler(context unsafe.Pointer) { // Sanitize the registers; interrupts must always be disabled. c := bluepillArchEnter(bluepillArchContext(context)) - // Enable preemption. - c.setSignalMask(true) - // Increment the number of switches. atomic.AddUint32(&c.switches, 1) @@ -106,9 +89,6 @@ func bluepillHandler(context unsafe.Pointer) { // interrupted KVM. Since we're in a signal handler // currently, all signals are masked and the signal // must have been delivered directly to this thread. - // - // We will not be able to actually do subsequent - // KVM_RUNs until this signal is processed. timeout := syscall.Timespec{} sig, _, errno := syscall.RawSyscall6( syscall.SYS_RT_SIGTIMEDWAIT, @@ -118,24 +98,12 @@ func bluepillHandler(context unsafe.Pointer) { 8, // sigset size. 0, 0) if errno == syscall.EAGAIN { - // If weren't able to process this signal, then - // it must not have been in the bounceMask. By - // elimination, it must have been the - // preemption signal. We can't process this - // signal right now, so we need to disable - // preemption until the interrupt is actually - // handled. - c.setSignalMask(false) - // Note that there is a waiter for this vCPU. - // This will cause the vCPU to exit at some - // point in the future (releasing the user lock - // and guest mode). - atomicbitops.OrUint32(&c.state, vCPUWaiter) - } else if errno != 0 { - // We only expect success or a timeout. + continue + } + if errno != 0 { throw("error waiting for pending signal") - } else if sig != uintptr(bounceSignal) { - // Only the bounce should be processed. + } + if sig != uintptr(bounceSignal) { throw("unexpected signal") } @@ -146,10 +114,11 @@ func bluepillHandler(context unsafe.Pointer) { // ready. if c.runData.readyForInterruptInjection == 0 { c.runData.requestInterruptWindow = 1 + continue // Rerun vCPU. } else { - injectInterrupt(c) + // Force injection below; the vCPU is ready. + c.runData.exitReason = _KVM_EXIT_IRQ_WINDOW_OPEN } - continue // Rerun vCPU. case syscall.EFAULT: // If a fault is not serviceable due to the host // backing pages having page permissions, instead of an @@ -168,30 +137,6 @@ func bluepillHandler(context unsafe.Pointer) { } switch c.runData.exitReason { - case _KVM_EXIT_HLT: - // Copy out registers. - bluepillArchExit(c, bluepillArchContext(context)) - - // Return to the vCPUReady state; notify any waiters. - user := atomic.LoadUint32(&c.state) & vCPUUser - switch atomic.SwapUint32(&c.state, user) { - case user | vCPUGuest: // Expected case. - case user | vCPUGuest | vCPUWaiter: - c.notify() - default: - throw("invalid state") - } - return - case _KVM_EXIT_IRQ_WINDOW_OPEN: - // Inject an interrupt now. - injectInterrupt(c) - // Clear previous injection request. - c.runData.requestInterruptWindow = 0 - case _KVM_EXIT_INTR: - // This is fine, it is the normal exit reason during - // signal delivery. However, we still need to handle - // other potential exit reasons *combined* with EINTR, - // so this switch must be hit even after the above. case _KVM_EXIT_EXCEPTION: c.die(bluepillArchContext(context), "exception") return @@ -210,6 +155,20 @@ func bluepillHandler(context unsafe.Pointer) { case _KVM_EXIT_DEBUG: c.die(bluepillArchContext(context), "debug") return + case _KVM_EXIT_HLT: + // Copy out registers. + bluepillArchExit(c, bluepillArchContext(context)) + + // Return to the vCPUReady state; notify any waiters. + user := atomic.LoadUint32(&c.state) & vCPUUser + switch atomic.SwapUint32(&c.state, user) { + case user | vCPUGuest: // Expected case. + case user | vCPUGuest | vCPUWaiter: + c.notify() + default: + throw("invalid state") + } + return case _KVM_EXIT_MMIO: // Increment the fault count. atomic.AddUint32(&c.faults, 1) @@ -241,6 +200,18 @@ func bluepillHandler(context unsafe.Pointer) { data[i] = *b } } + case _KVM_EXIT_IRQ_WINDOW_OPEN: + // Interrupt: we must have requested an interrupt + // window; set the interrupt line. + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(c.fd), + _KVM_INTERRUPT, + uintptr(unsafe.Pointer(&bounce))); errno != 0 { + throw("interrupt injection failed") + } + // Clear previous injection request. + c.runData.requestInterruptWindow = 0 case _KVM_EXIT_SHUTDOWN: c.die(bluepillArchContext(context), "shutdown") return diff --git a/pkg/sentry/platform/kvm/kvm_const.go b/pkg/sentry/platform/kvm/kvm_const.go index 07d9c9a98..1d5c77ff4 100644 --- a/pkg/sentry/platform/kvm/kvm_const.go +++ b/pkg/sentry/platform/kvm/kvm_const.go @@ -48,7 +48,6 @@ const ( _KVM_EXIT_IRQ_WINDOW_OPEN = 0x7 _KVM_EXIT_SHUTDOWN = 0x8 _KVM_EXIT_FAIL_ENTRY = 0x9 - _KVM_EXIT_INTR = 0xa _KVM_EXIT_INTERNAL_ERROR = 0x11 _KVM_EXIT_SYSTEM_EVENT = 0x18 ) diff --git a/pkg/sentry/platform/kvm/kvm_test.go b/pkg/sentry/platform/kvm/kvm_test.go index d42ba3f24..c42752d50 100644 --- a/pkg/sentry/platform/kvm/kvm_test.go +++ b/pkg/sentry/platform/kvm/kvm_test.go @@ -16,15 +16,12 @@ package kvm import ( "math/rand" - "os" "reflect" - "runtime" "sync/atomic" "syscall" "testing" "time" - "gvisor.dev/gvisor/pkg/procid" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/kvm/testutil" @@ -323,18 +320,15 @@ func TestBounce(t *testing.T) { }) } -// randomSleep is used by some race tests below. -// -// O(hundreds of microseconds) is appropriate to ensure different overlaps and -// different schedules. -func randomSleep() { - if n := rand.Intn(1000); n > 100 { - time.Sleep(time.Duration(n) * time.Microsecond) - } -} - func TestBounceStress(t *testing.T) { applicationTest(t, true, testutil.SpinLoop, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { + randomSleep := func() { + // O(hundreds of microseconds) is appropriate to ensure + // different overlaps and different schedules. + if n := rand.Intn(1000); n > 100 { + time.Sleep(time.Duration(n) * time.Microsecond) + } + } for i := 0; i < 1000; i++ { // Start an asynchronously executing goroutine that // calls Bounce at pseudo-random point in time. @@ -361,50 +355,6 @@ func TestBounceStress(t *testing.T) { }) } -func TestPreemption(t *testing.T) { - applicationTest(t, true, testutil.SpinLoop, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { - // Lock the main vCPU thread. - runtime.LockOSThread() - pid := os.Getpid() - tid := procid.Current() - running := uint32(1) - defer atomic.StoreUint32(&running, 0) - - // Start generating "preemptions". - go func() { - for atomic.LoadUint32(&running) != 0 { - // Kick via a preemption: best effort. - syscall.Tgkill(pid, int(tid), syscall.SIGURG) - randomSleep() - } - }() - - for i := 0; i < 1000; i++ { - randomSleep() - var si arch.SignalInfo - if _, err := c.SwitchToUser(ring0.SwitchOpts{ - Registers: regs, - FloatingPointState: dummyFPState, - PageTables: pt, - }, &si); err != platform.ErrContextInterrupt { - t.Errorf("application partial restore: got %v, wanted %v", err, platform.ErrContextInterrupt) - } - // Was this caused by a preemption signal? - if got := atomic.LoadUint32(&c.state); got&vCPUGuest != 0 && got&vCPUWaiter == 0 { - continue - } - c.unlock() - // Should have dropped from guest mode, processed preemption. - if got := atomic.LoadUint32(&c.state); got != vCPUReady { - t.Errorf("vCPU not in ready state: got %v", got) - } - randomSleep() - c.lock() - } - return false - }) -} - func TestInvalidate(t *testing.T) { var data uintptr // Used below. applicationTest(t, true, testutil.Touch, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { diff --git a/pkg/sentry/platform/kvm/machine.go b/pkg/sentry/platform/kvm/machine.go index 345b71e8f..f1afc74dc 100644 --- a/pkg/sentry/platform/kvm/machine.go +++ b/pkg/sentry/platform/kvm/machine.go @@ -108,9 +108,6 @@ type vCPU struct { // This is a bitmask of the three fields (vCPU*) described above. state uint32 - // signalMask is the vCPU signal mask. - signalMask uint64 - // runData for this vCPU. runData *runData @@ -124,7 +121,6 @@ type vCPU struct { // vCPUArchState is the architecture-specific state. vCPUArchState - // dieState is the temporary state associated with throwing exceptions. dieState dieState } @@ -157,6 +153,11 @@ func (m *machine) newVCPU() *vCPU { c.CPU.Init(&m.kernel, c) m.vCPUsByID[c.id] = c + // Ensure the signal mask is correct. + if err := c.setSignalMask(); err != nil { + panic(fmt.Sprintf("error setting signal mask: %v", err)) + } + // Map the run data. runData, err := mapRunData(int(fd)) if err != nil { diff --git a/pkg/sentry/platform/kvm/machine_amd64_unsafe.go b/pkg/sentry/platform/kvm/machine_amd64_unsafe.go index 52286e56d..7156c245f 100644 --- a/pkg/sentry/platform/kvm/machine_amd64_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_amd64_unsafe.go @@ -111,6 +111,31 @@ func (c *vCPU) setSystemTime() error { return nil } +// setSignalMask sets the vCPU signal mask. +// +// This must be called prior to running the vCPU. +func (c *vCPU) setSignalMask() error { + // The layout of this structure implies that it will not necessarily be + // the same layout chosen by the Go compiler. It gets fudged here. + var data struct { + length uint32 + mask1 uint32 + mask2 uint32 + _ uint32 + } + data.length = 8 // Fixed sigset size. + data.mask1 = ^uint32(bounceSignalMask & 0xffffffff) + data.mask2 = ^uint32(bounceSignalMask >> 32) + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(c.fd), + _KVM_SET_SIGNAL_MASK, + uintptr(unsafe.Pointer(&data))); errno != 0 { + return fmt.Errorf("error setting signal mask: %v", errno) + } + return nil +} + // setUserRegisters sets user registers in the vCPU. func (c *vCPU) setUserRegisters(uregs *userRegs) error { if _, _, errno := syscall.RawSyscall( diff --git a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go index 185eeb4f0..b531f2f85 100644 --- a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go @@ -268,6 +268,32 @@ func (c *vCPU) setSystemTime() error { return nil } +// setSignalMask sets the vCPU signal mask. +// +// This must be called prior to running the vCPU. +func (c *vCPU) setSignalMask() error { + // The layout of this structure implies that it will not necessarily be + // the same layout chosen by the Go compiler. It gets fudged here. + var data struct { + length uint32 + mask1 uint32 + mask2 uint32 + _ uint32 + } + data.length = 8 // Fixed sigset size. + data.mask1 = ^uint32(bounceSignalMask & 0xffffffff) + data.mask2 = ^uint32(bounceSignalMask >> 32) + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(c.fd), + _KVM_SET_SIGNAL_MASK, + uintptr(unsafe.Pointer(&data))); errno != 0 { + return fmt.Errorf("error setting signal mask: %v", errno) + } + + return nil +} + // SwitchToUser unpacks architectural-details. func (c *vCPU) SwitchToUser(switchOpts ring0.SwitchOpts, info *arch.SignalInfo) (usermem.AccessType, error) { // Check for canonical addresses. diff --git a/pkg/sentry/platform/kvm/machine_unsafe.go b/pkg/sentry/platform/kvm/machine_unsafe.go index e4de0a889..f04be2ab5 100644 --- a/pkg/sentry/platform/kvm/machine_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_unsafe.go @@ -87,47 +87,6 @@ func unmapRunData(r *runData) error { return nil } -// setSignalMask sets the vCPU signal mask. -// -// This will be called from the bluepill handler, and therefore must not -// perform any allocation. -// -//go:nosplit -func (c *vCPU) setSignalMask(enableOthers bool) { - // The signal mask is either: - // *) Only the bounce signal, which we need to use to execute the - // machine state up until the bounce interrupt can be processed. - // or - // *) All signals, which is the default state unless we need to - // continue execution to exit guest mode (the case above). - mask := bounceSignalMask - if enableOthers { - mask |= otherSignalsMask - } - if c.signalMask == mask { - return // Already set. - } - - // The layout of this structure implies that it will not necessarily be - // the same layout chosen by the Go compiler. It gets fudged here. - var data struct { - length uint32 - mask1 uint32 - mask2 uint32 - _ uint32 - } - data.length = 8 // Fixed sigset size. - data.mask1 = ^uint32(mask & 0xffffffff) - data.mask2 = ^uint32(mask >> 32) - if _, _, errno := syscall.RawSyscall( - syscall.SYS_IOCTL, - uintptr(c.fd), - _KVM_SET_SIGNAL_MASK, - uintptr(unsafe.Pointer(&data))); errno != 0 { - throw("setSignal mask failed") - } -} - // atomicAddressSpace is an atomic address space pointer. type atomicAddressSpace struct { pointer unsafe.Pointer diff --git a/runsc/sandbox/sandbox.go b/runsc/sandbox/sandbox.go index 6c15727fa..8de75ae57 100644 --- a/runsc/sandbox/sandbox.go +++ b/runsc/sandbox/sandbox.go @@ -444,6 +444,12 @@ func (s *Sandbox) createSandboxProcess(conf *boot.Config, args *Args, startSyncF nextFD++ } + // TODO(b/151157106): syscall tests fail by timeout if asyncpreemptoff + // isn't set. + if conf.Platform == "kvm" { + cmd.Env = append(cmd.Env, "GODEBUG=asyncpreemptoff=1") + } + // The current process' stdio must be passed to the application via the // --stdio-fds flag. The stdio of the sandbox process itself must not // be connected to the same FDs, otherwise we risk leaking sandbox -- cgit v1.2.3 From 0d1e299079392043fae24c2be524a0eefe7d8085 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Wed, 1 Apr 2020 12:05:17 -0700 Subject: Pass configurable FilesystemType to tmpfs. PiperOrigin-RevId: 304234086 --- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 26 +++++++++++++++++++------- 1 file changed, 19 insertions(+), 7 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index afd9f8533..8bc8818c0 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -77,6 +77,11 @@ type FilesystemOpts struct { // RootSymlinkTarget is the target of the root symlink. Only valid if // RootFileType == S_IFLNK. RootSymlinkTarget string + + // FilesystemType allows setting a different FilesystemType for this + // tmpfs filesystem. This allows tmpfs to "impersonate" other + // filesystems, like ramdiskfs and cgroupfs. + FilesystemType vfs.FilesystemType } // GetFilesystem implements vfs.FilesystemType.GetFilesystem. @@ -91,15 +96,22 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt clock: clock, } - fs.vfsfs.Init(vfsObj, &fstype, &fs) - - typ := uint16(linux.S_IFDIR) + rootFileType := uint16(linux.S_IFDIR) + newFSType := vfs.FilesystemType(&fstype) tmpfsOpts, ok := opts.InternalData.(FilesystemOpts) - if ok && tmpfsOpts.RootFileType != 0 { - typ = tmpfsOpts.RootFileType + if ok { + if tmpfsOpts.RootFileType != 0 { + rootFileType = tmpfsOpts.RootFileType + } + if tmpfsOpts.FilesystemType != nil { + newFSType = tmpfsOpts.FilesystemType + } } + + fs.vfsfs.Init(vfsObj, newFSType, &fs) + var root *inode - switch typ { + switch rootFileType { case linux.S_IFREG: root = fs.newRegularFile(creds, 0777) case linux.S_IFLNK: @@ -107,7 +119,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt case linux.S_IFDIR: root = fs.newDirectory(creds, 01777) default: - return nil, nil, fmt.Errorf("invalid tmpfs root file type: %#o", typ) + return nil, nil, fmt.Errorf("invalid tmpfs root file type: %#o", rootFileType) } return &fs.vfsfs, &fs.newDentry(root).vfsd, nil } -- cgit v1.2.3 From dbc507dc5cfde2f69a94a58fcb2744ef0899ce7e Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Thu, 2 Apr 2020 16:57:08 -0700 Subject: Add equivalents to FMODE_PREAD/PWRITE to VFS2. This is mostly required for PipeTest_OffsetCalls. The options are DenyPRead/PWrite rather than AllowPRead/PWrite since, in Linux terms, fs/open.c:do_dentry_open sets FMODE_PREAD|FMODE_PWRITE unconditionally (although it allows filesystem implementations of open to unset these flags), so they're set for most FDs; it's usually FDs created outside of open(2) that don't get them, e.g.: - Syscall-created pipes (fs/pipe.c:create_pipe_files => fs/file_table.c:alloc_file_pseudo) - Epoll instances (fs/eventpoll.c:do_epoll_create => fs/anon_inodes.c:anon_inode_getfile => alloc_file_pseudo) - Sockets (net/socket.c:sock_alloc_file => alloc_file_pseudo) This CL adds the flags to epoll instances; a subsequent CL reworks the VFS2 implementation of pipe FDs to be filesystem-independent and adds the flags there, and sockets aren't implemented yet. Updates #1035 PiperOrigin-RevId: 304506434 --- pkg/sentry/vfs/epoll.go | 2 ++ pkg/sentry/vfs/file_description.go | 15 +++++++++++++++ 2 files changed, 17 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/vfs/epoll.go b/pkg/sentry/vfs/epoll.go index 3da45d744..8e0b40841 100644 --- a/pkg/sentry/vfs/epoll.go +++ b/pkg/sentry/vfs/epoll.go @@ -99,6 +99,8 @@ func (vfs *VirtualFilesystem) NewEpollInstanceFD() (*FileDescription, error) { interest: make(map[epollInterestKey]*epollInterest), } if err := ep.vfsfd.Init(ep, linux.O_RDWR, vd.Mount(), vd.Dentry(), &FileDescriptionOptions{ + DenyPRead: true, + DenyPWrite: true, UseDentryMetadata: true, }); err != nil { return nil, err diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 8ee549dc2..5df4bbf45 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -84,6 +84,13 @@ type FileDescriptionOptions struct { // usually only the case if O_DIRECT would actually have an effect. AllowDirectIO bool + // If DenyPRead is true, calls to FileDescription.PRead() return ESPIPE. + DenyPRead bool + + // If DenyPWrite is true, calls to FileDescription.PWrite() return + // ESPIPE. + DenyPWrite bool + // If UseDentryMetadata is true, calls to FileDescription methods that // interact with file and filesystem metadata (Stat, SetStat, StatFS, // Listxattr, Getxattr, Setxattr, Removexattr) are implemented by calling @@ -306,6 +313,7 @@ type FileDescriptionImpl interface { // - If opts.Flags specifies unsupported options, PRead returns EOPNOTSUPP. // // Preconditions: The FileDescription was opened for reading. + // FileDescriptionOptions.DenyPRead == false. PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts ReadOptions) (int64, error) // Read is similar to PRead, but does not specify an offset. @@ -337,6 +345,7 @@ type FileDescriptionImpl interface { // EOPNOTSUPP. // // Preconditions: The FileDescription was opened for writing. + // FileDescriptionOptions.DenyPWrite == false. PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts WriteOptions) (int64, error) // Write is similar to PWrite, but does not specify an offset, which is @@ -515,6 +524,9 @@ func (fd *FileDescription) EventUnregister(e *waiter.Entry) { // offset, and returns the number of bytes read. PRead is permitted to return // partial reads with a nil error. func (fd *FileDescription) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts ReadOptions) (int64, error) { + if fd.opts.DenyPRead { + return 0, syserror.ESPIPE + } if !fd.readable { return 0, syserror.EBADF } @@ -533,6 +545,9 @@ func (fd *FileDescription) Read(ctx context.Context, dst usermem.IOSequence, opt // offset, and returns the number of bytes written. PWrite is permitted to // return partial writes with a nil error. func (fd *FileDescription) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts WriteOptions) (int64, error) { + if fd.opts.DenyPWrite { + return 0, syserror.ESPIPE + } if !fd.writable { return 0, syserror.EBADF } -- cgit v1.2.3 From 5b2396d244ed6283d928a72bdd4cc58d78ef3175 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Thu, 2 Apr 2020 17:06:19 -0700 Subject: Fix typo in TODO comments. PiperOrigin-RevId: 304508083 --- pkg/sentry/fs/proc/mounts.go | 3 ++- pkg/sentry/fsimpl/tmpfs/filesystem.go | 6 +++--- pkg/sentry/fsimpl/tmpfs/stat_test.go | 4 ++-- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 2 +- pkg/sentry/socket/netstack/netstack.go | 2 +- pkg/sentry/vfs/mount.go | 3 ++- test/syscalls/linux/socket_inet_loopback.cc | 2 +- 7 files changed, 12 insertions(+), 10 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/proc/mounts.go b/pkg/sentry/fs/proc/mounts.go index 94deb553b..1fc9c703c 100644 --- a/pkg/sentry/fs/proc/mounts.go +++ b/pkg/sentry/fs/proc/mounts.go @@ -170,7 +170,8 @@ func superBlockOpts(mountPath string, msrc *fs.MountSource) string { // NOTE(b/147673608): If the mount is a cgroup, we also need to include // the cgroup name in the options. For now we just read that from the // path. - // TODO(gvisor.dev/issues/190): Once gVisor has full cgroup support, we + // + // TODO(gvisor.dev/issue/190): Once gVisor has full cgroup support, we // should get this value from the cgroup itself, and not rely on the // path. if msrc.FilesystemType == "cgroup" { diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index e678ecc37..4cf27bf13 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -57,7 +57,7 @@ afterSymlink: } next := nextVFSD.Impl().(*dentry) if symlink, ok := next.inode.impl.(*symlink); ok && rp.ShouldFollowSymlink() { - // TODO(gvisor.dev/issues/1197): Symlink traversals updates + // TODO(gvisor.dev/issue/1197): Symlink traversals updates // access time. if err := rp.HandleSymlink(symlink.target); err != nil { return nil, err @@ -515,7 +515,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa oldParent.inode.decLinksLocked() newParent.inode.incLinksLocked() } - // TODO(gvisor.dev/issues/1197): Update timestamps and parent directory + // TODO(gvisor.dev/issue/1197): Update timestamps and parent directory // sizes. vfsObj.CommitRenameReplaceDentry(renamedVFSD, &newParent.vfsd, newName, replacedVFSD) return nil @@ -600,7 +600,7 @@ func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linu if err != nil { return linux.Statfs{}, err } - // TODO(gvisor.dev/issues/1197): Actually implement statfs. + // TODO(gvisor.dev/issue/1197): Actually implement statfs. return linux.Statfs{}, syserror.ENOSYS } diff --git a/pkg/sentry/fsimpl/tmpfs/stat_test.go b/pkg/sentry/fsimpl/tmpfs/stat_test.go index ebe035dee..3e02e7190 100644 --- a/pkg/sentry/fsimpl/tmpfs/stat_test.go +++ b/pkg/sentry/fsimpl/tmpfs/stat_test.go @@ -29,7 +29,7 @@ func TestStatAfterCreate(t *testing.T) { mode := linux.FileMode(0644) // Run with different file types. - // TODO(gvisor.dev/issues/1197): Also test symlinks and sockets. + // TODO(gvisor.dev/issue/1197): Also test symlinks and sockets. for _, typ := range []string{"file", "dir", "pipe"} { t.Run(fmt.Sprintf("type=%q", typ), func(t *testing.T) { var ( @@ -169,7 +169,7 @@ func TestSetStat(t *testing.T) { mode := linux.FileMode(0644) // Run with different file types. - // TODO(gvisor.dev/issues/1197): Also test symlinks and sockets. + // TODO(gvisor.dev/issue/1197): Also test symlinks and sockets. for _, typ := range []string{"file", "dir", "pipe"} { t.Run(fmt.Sprintf("type=%q", typ), func(t *testing.T) { var ( diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 8bc8818c0..54da15849 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -315,7 +315,7 @@ func (i *inode) statTo(stat *linux.Statx) { stat.Atime = linux.NsecToStatxTimestamp(i.atime) stat.Ctime = linux.NsecToStatxTimestamp(i.ctime) stat.Mtime = linux.NsecToStatxTimestamp(i.mtime) - // TODO(gvisor.dev/issues/1197): Device number. + // TODO(gvisor.dev/issue/1197): Device number. switch impl := i.impl.(type) { case *regularFile: stat.Mask |= linux.STATX_SIZE | linux.STATX_BLOCKS diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index f14c336b9..06a5b53bc 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -663,7 +663,7 @@ func (s *SocketOperations) checkFamily(family uint16, exact bool) *syserr.Error // This is a hack to work around the fact that both IPv4 and IPv6 ANY are // represented by the empty string. // -// TODO(gvisor.dev/issues/1556): remove this function. +// TODO(gvisor.dev/issue/1556): remove this function. func (s *SocketOperations) mapFamily(addr tcpip.FullAddress, family uint16) tcpip.FullAddress { if len(addr.Addr) == 0 && s.family == linux.AF_INET6 && family == linux.AF_INET { addr.Addr = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\xff\xff\x00\x00\x00\x00" diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index 7792eb1a0..1b8ecc415 100644 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go @@ -835,7 +835,8 @@ func superBlockOpts(mountPath string, mnt *Mount) string { // NOTE(b/147673608): If the mount is a cgroup, we also need to include // the cgroup name in the options. For now we just read that from the // path. - // TODO(gvisor.dev/issues/190): Once gVisor has full cgroup support, we + // + // TODO(gvisor.dev/issue/190): Once gVisor has full cgroup support, we // should get this value from the cgroup itself, and not rely on the // path. if mnt.fs.FilesystemType().Name() == "cgroup" { diff --git a/test/syscalls/linux/socket_inet_loopback.cc b/test/syscalls/linux/socket_inet_loopback.cc index 16888de2a..2ffc86382 100644 --- a/test/syscalls/linux/socket_inet_loopback.cc +++ b/test/syscalls/linux/socket_inet_loopback.cc @@ -234,7 +234,7 @@ TEST_P(DualStackSocketTest, AddressOperations) { } } -// TODO(gvisor.dev/issues/1556): uncomment V4MappedAny. +// TODO(gvisor.dev/issue/1556): uncomment V4MappedAny. INSTANTIATE_TEST_SUITE_P( All, DualStackSocketTest, ::testing::Combine( -- cgit v1.2.3 From dd3bc499970c22ebbd270030b4564e6b8e4e929e Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Thu, 2 Apr 2020 19:37:41 -0700 Subject: Add NAME_MAX checks and update file times NAME_MAX should be enforced per filesystem implementation because other file systems may not have the same restriction. Gofer filesystem now keeps a reference to the kernel clock to avoid lookup in the Context on file access to update atime. Update access, modification, and status change times in tmpfs. Updates #1197, #1198. PiperOrigin-RevId: 304527148 --- pkg/sentry/fsimpl/gofer/directory.go | 7 +++- pkg/sentry/fsimpl/gofer/filesystem.go | 13 ++++++-- pkg/sentry/fsimpl/gofer/gofer.go | 22 ++++++------- pkg/sentry/fsimpl/gofer/regular_file.go | 7 ++-- pkg/sentry/fsimpl/gofer/special_file.go | 4 +-- pkg/sentry/fsimpl/gofer/symlink.go | 2 +- pkg/sentry/fsimpl/gofer/time.go | 39 +++++++++++----------- pkg/sentry/fsimpl/kernfs/filesystem.go | 9 ++++++ pkg/sentry/fsimpl/tmpfs/directory.go | 2 ++ pkg/sentry/fsimpl/tmpfs/filesystem.go | 25 +++++++++++++-- pkg/sentry/fsimpl/tmpfs/regular_file.go | 4 ++- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 57 ++++++++++++++++++++++++++++++--- 12 files changed, 139 insertions(+), 52 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/directory.go b/pkg/sentry/fsimpl/gofer/directory.go index 5dbfc6250..49d9f859b 100644 --- a/pkg/sentry/fsimpl/gofer/directory.go +++ b/pkg/sentry/fsimpl/gofer/directory.go @@ -56,14 +56,19 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba fd.mu.Lock() defer fd.mu.Unlock() + d := fd.dentry() if fd.dirents == nil { - ds, err := fd.dentry().getDirents(ctx) + ds, err := d.getDirents(ctx) if err != nil { return err } fd.dirents = ds } + if d.fs.opts.interop != InteropModeShared { + d.touchAtime(fd.vfsfd.Mount()) + } + for fd.off < int64(len(fd.dirents)) { if err := cb.Handle(fd.dirents[fd.off]); err != nil { return err diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 269624362..305228bda 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -356,7 +356,9 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir if err := create(parent, name); err != nil { return err } - parent.touchCMtime(ctx) + if fs.opts.interop != InteropModeShared { + parent.touchCMtime() + } delete(parent.negativeChildren, name) parent.dirents = nil return nil @@ -454,7 +456,7 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b return err } if fs.opts.interop != InteropModeShared { - parent.touchCMtime(ctx) + parent.touchCMtime() if dir { parent.decLinks() } @@ -802,7 +804,6 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving d.IncRef() // reference held by child on its parent d d.vfsd.InsertChild(&child.vfsd, name) if d.fs.opts.interop != InteropModeShared { - d.touchCMtime(ctx) delete(d.negativeChildren, name) d.dirents = nil } @@ -834,6 +835,9 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving } childVFSFD = &fd.vfsfd } + if d.fs.opts.interop != InteropModeShared { + d.touchCMtime() + } return childVFSFD, nil } @@ -975,6 +979,9 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa oldParent.decLinks() newParent.incLinks() } + oldParent.touchCMtime() + newParent.touchCMtime() + renamed.touchCtime() } vfsObj.CommitRenameReplaceDentry(&renamed.vfsd, &newParent.vfsd, newName, replacedVFSD) return nil diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 8e41b6b1c..adee8bb60 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -44,6 +44,7 @@ import ( "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/vfs" @@ -72,6 +73,9 @@ type filesystem struct { // client is the client used by this filesystem. client is immutable. client *p9.Client + // clock is a realtime clock used to set timestamps in file operations. + clock ktime.Clock + // uid and gid are the effective KUID and KGID of the filesystem's creator, // and are used as the owner and group for files that don't specify one. // uid and gid are immutable. @@ -376,6 +380,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt uid: creds.EffectiveKUID, gid: creds.EffectiveKGID, client: client, + clock: ktime.RealtimeClockFromContext(ctx), dentries: make(map[*dentry]struct{}), specialFileFDs: make(map[*specialFileFD]struct{}), } @@ -779,10 +784,7 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin // data, so there's no cache to truncate either.) return nil } - now, haveNow := nowFromContext(ctx) - if !haveNow { - ctx.Warningf("gofer.dentry.setStat: current time not available") - } + now := d.fs.clock.Now().Nanoseconds() if stat.Mask&linux.STATX_MODE != 0 { atomic.StoreUint32(&d.mode, d.fileType()|uint32(stat.Mode)) } @@ -794,25 +796,19 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin } if setLocalAtime { if stat.Atime.Nsec == linux.UTIME_NOW { - if haveNow { - atomic.StoreInt64(&d.atime, now) - } + atomic.StoreInt64(&d.atime, now) } else { atomic.StoreInt64(&d.atime, dentryTimestampFromStatx(stat.Atime)) } } if setLocalMtime { if stat.Mtime.Nsec == linux.UTIME_NOW { - if haveNow { - atomic.StoreInt64(&d.mtime, now) - } + atomic.StoreInt64(&d.mtime, now) } else { atomic.StoreInt64(&d.mtime, dentryTimestampFromStatx(stat.Mtime)) } } - if haveNow { - atomic.StoreInt64(&d.ctime, now) - } + atomic.StoreInt64(&d.ctime, now) if stat.Mask&linux.STATX_SIZE != 0 { d.dataMu.Lock() oldSize := d.size diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go index 3593eb1d5..857f7c74e 100644 --- a/pkg/sentry/fsimpl/gofer/regular_file.go +++ b/pkg/sentry/fsimpl/gofer/regular_file.go @@ -104,7 +104,7 @@ func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offs putDentryReadWriter(rw) if d.fs.opts.interop != InteropModeShared { // Compare Linux's mm/filemap.c:do_generic_file_read() => file_accessed(). - d.touchAtime(ctx, fd.vfsfd.Mount()) + d.touchAtime(fd.vfsfd.Mount()) } return n, err } @@ -139,10 +139,7 @@ func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off // Compare Linux's mm/filemap.c:__generic_file_write_iter() => // file_update_time(). This is d.touchCMtime(), but without locking // d.metadataMu (recursively). - if now, ok := nowFromContext(ctx); ok { - atomic.StoreInt64(&d.mtime, now) - atomic.StoreInt64(&d.ctime, now) - } + d.touchCMtimeLocked() } if fd.vfsfd.StatusFlags()&linux.O_DIRECT != 0 { // Write dirty cached pages that will be touched by the write back to diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go index 274f7346f..507e0e276 100644 --- a/pkg/sentry/fsimpl/gofer/special_file.go +++ b/pkg/sentry/fsimpl/gofer/special_file.go @@ -76,7 +76,7 @@ func (fd *specialFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offs // hold here since specialFileFD doesn't client-cache data. Just buffer the // read instead. if d := fd.dentry(); d.fs.opts.interop != InteropModeShared { - d.touchAtime(ctx, fd.vfsfd.Mount()) + d.touchAtime(fd.vfsfd.Mount()) } buf := make([]byte, dst.NumBytes()) n, err := fd.handle.readToBlocksAt(ctx, safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf)), uint64(offset)) @@ -117,7 +117,7 @@ func (fd *specialFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off // Do a buffered write. See rationale in PRead. if d := fd.dentry(); d.fs.opts.interop != InteropModeShared { - d.touchCMtime(ctx) + d.touchCMtime() } buf := make([]byte, src.NumBytes()) // Don't do partial writes if we get a partial read from src. diff --git a/pkg/sentry/fsimpl/gofer/symlink.go b/pkg/sentry/fsimpl/gofer/symlink.go index adf43be60..2ec819f86 100644 --- a/pkg/sentry/fsimpl/gofer/symlink.go +++ b/pkg/sentry/fsimpl/gofer/symlink.go @@ -27,7 +27,7 @@ func (d *dentry) isSymlink() bool { // Precondition: d.isSymlink(). func (d *dentry) readlink(ctx context.Context, mnt *vfs.Mount) (string, error) { if d.fs.opts.interop != InteropModeShared { - d.touchAtime(ctx, mnt) + d.touchAtime(mnt) d.dataMu.Lock() if d.haveTarget { target := d.target diff --git a/pkg/sentry/fsimpl/gofer/time.go b/pkg/sentry/fsimpl/gofer/time.go index 7598ec6a8..2608e7e1d 100644 --- a/pkg/sentry/fsimpl/gofer/time.go +++ b/pkg/sentry/fsimpl/gofer/time.go @@ -18,8 +18,6 @@ import ( "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/vfs" ) @@ -38,23 +36,12 @@ func statxTimestampFromDentry(ns int64) linux.StatxTimestamp { } } -func nowFromContext(ctx context.Context) (int64, bool) { - if clock := ktime.RealtimeClockFromContext(ctx); clock != nil { - return clock.Now().Nanoseconds(), true - } - return 0, false -} - // Preconditions: fs.interop != InteropModeShared. -func (d *dentry) touchAtime(ctx context.Context, mnt *vfs.Mount) { +func (d *dentry) touchAtime(mnt *vfs.Mount) { if err := mnt.CheckBeginWrite(); err != nil { return } - now, ok := nowFromContext(ctx) - if !ok { - mnt.EndWrite() - return - } + now := d.fs.clock.Now().Nanoseconds() d.metadataMu.Lock() atomic.StoreInt64(&d.atime, now) d.metadataMu.Unlock() @@ -63,13 +50,25 @@ func (d *dentry) touchAtime(ctx context.Context, mnt *vfs.Mount) { // Preconditions: fs.interop != InteropModeShared. The caller has successfully // called vfs.Mount.CheckBeginWrite(). -func (d *dentry) touchCMtime(ctx context.Context) { - now, ok := nowFromContext(ctx) - if !ok { - return - } +func (d *dentry) touchCtime() { + now := d.fs.clock.Now().Nanoseconds() + d.metadataMu.Lock() + atomic.StoreInt64(&d.ctime, now) + d.metadataMu.Unlock() +} + +// Preconditions: fs.interop != InteropModeShared. The caller has successfully +// called vfs.Mount.CheckBeginWrite(). +func (d *dentry) touchCMtime() { + now := d.fs.clock.Now().Nanoseconds() d.metadataMu.Lock() atomic.StoreInt64(&d.mtime, now) atomic.StoreInt64(&d.ctime, now) d.metadataMu.Unlock() } + +func (d *dentry) touchCMtimeLocked() { + now := d.fs.clock.Now().Nanoseconds() + atomic.StoreInt64(&d.mtime, now) + atomic.StoreInt64(&d.ctime, now) +} diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index a429fa23d..89f5da3d4 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -63,6 +63,9 @@ afterSymlink: rp.Advance() return nextVFSD, nil } + if len(name) > linux.NAME_MAX { + return nil, syserror.ENAMETOOLONG + } d.dirMu.Lock() nextVFSD, err := rp.ResolveChild(vfsd, name) if err != nil { @@ -191,6 +194,9 @@ func checkCreateLocked(ctx context.Context, rp *vfs.ResolvingPath, parentVFSD *v if pc == "." || pc == ".." { return "", syserror.EEXIST } + if len(pc) > linux.NAME_MAX { + return "", syserror.ENAMETOOLONG + } childVFSD, err := rp.ResolveChild(parentVFSD, pc) if err != nil { return "", err @@ -433,6 +439,9 @@ afterTrailingSymlink: if pc == "." || pc == ".." { return nil, syserror.EISDIR } + if len(pc) > linux.NAME_MAX { + return nil, syserror.ENAMETOOLONG + } // Determine whether or not we need to create a file. childVFSD, err := rp.ResolveChild(parentVFSD, pc) if err != nil { diff --git a/pkg/sentry/fsimpl/tmpfs/directory.go b/pkg/sentry/fsimpl/tmpfs/directory.go index 37c75ab64..45712c9b9 100644 --- a/pkg/sentry/fsimpl/tmpfs/directory.go +++ b/pkg/sentry/fsimpl/tmpfs/directory.go @@ -68,6 +68,8 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba fs.mu.Lock() defer fs.mu.Unlock() + fd.inode().touchAtime(fd.vfsfd.Mount()) + if fd.off == 0 { if err := cb.Handle(vfs.Dirent{ Name: ".", diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 4cf27bf13..1978af69c 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -46,6 +46,9 @@ func stepLocked(rp *vfs.ResolvingPath, d *dentry) (*dentry, error) { return nil, err } afterSymlink: + if len(rp.Component()) > linux.NAME_MAX { + return nil, syserror.ENAMETOOLONG + } nextVFSD, err := rp.ResolveComponent(&d.vfsd) if err != nil { return nil, err @@ -133,6 +136,9 @@ func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(pa if name == "." || name == ".." { return syserror.EEXIST } + if len(name) > linux.NAME_MAX { + return syserror.ENAMETOOLONG + } // Call parent.vfsd.Child() instead of stepLocked() or rp.ResolveChild(), // because if the child exists we want to return EEXIST immediately instead // of attempting symlink/mount traversal. @@ -153,7 +159,11 @@ func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(pa return err } defer mnt.EndWrite() - return create(parent, name) + if err := create(parent, name); err != nil { + return err + } + parent.inode.touchCMtime() + return nil } // AccessAt implements vfs.Filesystem.Impl.AccessAt. @@ -328,7 +338,12 @@ afterTrailingSymlink: child := fs.newDentry(fs.newRegularFile(rp.Credentials(), opts.Mode)) parent.vfsd.InsertChild(&child.vfsd, name) parent.inode.impl.(*directory).childList.PushBack(child) - return child.open(ctx, rp, &opts, true) + fd, err := child.open(ctx, rp, &opts, true) + if err != nil { + return nil, err + } + parent.inode.touchCMtime() + return fd, nil } if err != nil { return nil, err @@ -398,6 +413,7 @@ func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (st if !ok { return "", syserror.EINVAL } + symlink.inode.touchAtime(rp.Mount()) return symlink.target, nil } @@ -515,6 +531,9 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa oldParent.inode.decLinksLocked() newParent.inode.incLinksLocked() } + oldParent.inode.touchCMtime() + newParent.inode.touchCMtime() + renamed.inode.touchCtime() // TODO(gvisor.dev/issue/1197): Update timestamps and parent directory // sizes. vfsObj.CommitRenameReplaceDentry(renamedVFSD, &newParent.vfsd, newName, replacedVFSD) @@ -565,6 +584,7 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error parent.inode.decLinksLocked() // from child's ".." child.inode.decLinksLocked() vfsObj.CommitDeleteDentry(childVFSD) + parent.inode.touchCMtime() return nil } @@ -654,6 +674,7 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error parent.inode.impl.(*directory).childList.Remove(child) child.inode.decLinksLocked() vfsObj.CommitDeleteDentry(childVFSD) + parent.inode.touchCMtime() return nil } diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index 26cd65605..57e5e28ec 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -286,7 +286,8 @@ func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offs rw := getRegularFileReadWriter(f, offset) n, err := dst.CopyOutFrom(ctx, rw) putRegularFileReadWriter(rw) - return int64(n), err + fd.inode().touchAtime(fd.vfsfd.Mount()) + return n, err } // Read implements vfs.FileDescriptionImpl.Read. @@ -323,6 +324,7 @@ func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off f.inode.mu.Lock() rw := getRegularFileReadWriter(f, offset) n, err := src.CopyInTo(ctx, rw) + fd.inode().touchCMtimeLocked() f.inode.mu.Unlock() putRegularFileReadWriter(rw) return n, err diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 54da15849..ad47288f8 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -385,28 +385,41 @@ func (i *inode) setStat(ctx context.Context, creds *auth.Credentials, stat *linu return syserror.EINVAL } } + now := i.clock.Now().Nanoseconds() if mask&linux.STATX_ATIME != 0 { - atomic.StoreInt64(&i.atime, stat.Atime.ToNsecCapped()) + if stat.Atime.Nsec == linux.UTIME_NOW { + atomic.StoreInt64(&i.atime, now) + } else { + atomic.StoreInt64(&i.atime, stat.Atime.ToNsecCapped()) + } needsCtimeBump = true } if mask&linux.STATX_MTIME != 0 { - atomic.StoreInt64(&i.mtime, stat.Mtime.ToNsecCapped()) + if stat.Mtime.Nsec == linux.UTIME_NOW { + atomic.StoreInt64(&i.mtime, now) + } else { + atomic.StoreInt64(&i.mtime, stat.Mtime.ToNsecCapped()) + } needsCtimeBump = true // Ignore the mtime bump, since we just set it ourselves. needsMtimeBump = false } if mask&linux.STATX_CTIME != 0 { - atomic.StoreInt64(&i.ctime, stat.Ctime.ToNsecCapped()) + if stat.Ctime.Nsec == linux.UTIME_NOW { + atomic.StoreInt64(&i.ctime, now) + } else { + atomic.StoreInt64(&i.ctime, stat.Ctime.ToNsecCapped()) + } // Ignore the ctime bump, since we just set it ourselves. needsCtimeBump = false } - now := i.clock.Now().Nanoseconds() if needsMtimeBump { atomic.StoreInt64(&i.mtime, now) } if needsCtimeBump { atomic.StoreInt64(&i.ctime, now) } + i.mu.Unlock() return nil } @@ -484,6 +497,42 @@ func (i *inode) isDir() bool { return linux.FileMode(i.mode).FileType() == linux.S_IFDIR } +func (i *inode) touchAtime(mnt *vfs.Mount) { + if err := mnt.CheckBeginWrite(); err != nil { + return + } + now := i.clock.Now().Nanoseconds() + i.mu.Lock() + atomic.StoreInt64(&i.atime, now) + i.mu.Unlock() + mnt.EndWrite() +} + +// Preconditions: The caller has called vfs.Mount.CheckBeginWrite(). +func (i *inode) touchCtime() { + now := i.clock.Now().Nanoseconds() + i.mu.Lock() + atomic.StoreInt64(&i.ctime, now) + i.mu.Unlock() +} + +// Preconditions: The caller has called vfs.Mount.CheckBeginWrite(). +func (i *inode) touchCMtime() { + now := i.clock.Now().Nanoseconds() + i.mu.Lock() + atomic.StoreInt64(&i.mtime, now) + atomic.StoreInt64(&i.ctime, now) + i.mu.Unlock() +} + +// Preconditions: The caller has called vfs.Mount.CheckBeginWrite() and holds +// inode.mu. +func (i *inode) touchCMtimeLocked() { + now := i.clock.Now().Nanoseconds() + atomic.StoreInt64(&i.mtime, now) + atomic.StoreInt64(&i.ctime, now) +} + // fileDescription is embedded by tmpfs implementations of // vfs.FileDescriptionImpl. type fileDescription struct { -- cgit v1.2.3 From a94309628ebbc2e6c4997890f1b966fa7a16be20 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Fri, 3 Apr 2020 13:39:45 -0700 Subject: Ensure EOF is handled propertly during splice. PiperOrigin-RevId: 304684417 --- pkg/sentry/kernel/pipe/pipe.go | 13 ++++++++++--- test/syscalls/linux/sendfile.cc | 28 ++++++++++++++++++++++++++++ 2 files changed, 38 insertions(+), 3 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/pipe/pipe.go b/pkg/sentry/kernel/pipe/pipe.go index 725e9db7d..62c8691f1 100644 --- a/pkg/sentry/kernel/pipe/pipe.go +++ b/pkg/sentry/kernel/pipe/pipe.go @@ -255,7 +255,8 @@ func (p *Pipe) write(ctx context.Context, ops writeOps) (int64, error) { // POSIX requires that a write smaller than atomicIOBytes (PIPE_BUF) be // atomic, but requires no atomicity for writes larger than this. wanted := ops.left() - if avail := p.max - p.view.Size(); wanted > avail { + avail := p.max - p.view.Size() + if wanted > avail { if wanted <= p.atomicIOBytes { return 0, syserror.ErrWouldBlock } @@ -268,8 +269,14 @@ func (p *Pipe) write(ctx context.Context, ops writeOps) (int64, error) { return done, err } - if wanted > done { - // Partial write due to full pipe. + if done < avail { + // Non-failure, but short write. + return done, nil + } + if done < wanted { + // Partial write due to full pipe. Note that this could also be + // the short write case above, we would expect a second call + // and the write to return zero bytes in this case. return done, syserror.ErrWouldBlock } diff --git a/test/syscalls/linux/sendfile.cc b/test/syscalls/linux/sendfile.cc index 580ab5193..ebaafe47e 100644 --- a/test/syscalls/linux/sendfile.cc +++ b/test/syscalls/linux/sendfile.cc @@ -530,6 +530,34 @@ TEST(SendFileTest, SendToSpecialFile) { SyscallSucceedsWithValue(kSize & (~7))); } +TEST(SendFileTest, SendFileToPipe) { + // Create temp file. + constexpr char kData[] = ""; + constexpr int kDataSize = sizeof(kData) - 1; + const TempPath in_file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileWith( + GetAbsoluteTestTmpdir(), kData, TempPath::kDefaultFileMode)); + const FileDescriptor inf = + ASSERT_NO_ERRNO_AND_VALUE(Open(in_file.path(), O_RDONLY)); + + // Create a pipe for sending to a pipe. + int fds[2]; + ASSERT_THAT(pipe(fds), SyscallSucceeds()); + const FileDescriptor rfd(fds[0]); + const FileDescriptor wfd(fds[1]); + + // Expect to read up to the given size. + std::vector buf(kDataSize); + ScopedThread t([&]() { + absl::SleepFor(absl::Milliseconds(100)); + ASSERT_THAT(read(rfd.get(), buf.data(), buf.size()), + SyscallSucceedsWithValue(kDataSize)); + }); + + // Send with twice the size of the file, which should hit EOF. + EXPECT_THAT(sendfile(wfd.get(), inf.get(), nullptr, kDataSize * 2), + SyscallSucceedsWithValue(kDataSize)); +} + } // namespace } // namespace testing -- cgit v1.2.3 From 5818663ebe26857845685702d99db41c7aa2cf3d Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 3 Apr 2020 14:07:42 -0700 Subject: Add FileDescriptionImpl for Unix sockets. This change involves several steps: - Refactor the VFS1 unix socket implementation to share methods between VFS1 and VFS2 where possible. Re-implement the rest. - Override the default PRead, Read, PWrite, Write, Ioctl, Release methods in FileDescriptionDefaultImpl. - Add functions to create and initialize a new Dentry/Inode and FileDescription for a Unix socket file. Updates #1476 PiperOrigin-RevId: 304689796 --- pkg/sentry/fsimpl/sockfs/BUILD | 1 + pkg/sentry/fsimpl/sockfs/sockfs.go | 29 +++ pkg/sentry/kernel/BUILD | 1 + pkg/sentry/socket/netstack/netstack.go | 8 +- pkg/sentry/socket/unix/BUILD | 4 + pkg/sentry/socket/unix/unix.go | 89 ++++++--- pkg/sentry/socket/unix/unix_vfs2.go | 348 +++++++++++++++++++++++++++++++++ pkg/sentry/vfs/options.go | 5 + 8 files changed, 456 insertions(+), 29 deletions(-) create mode 100644 pkg/sentry/socket/unix/unix_vfs2.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/sockfs/BUILD b/pkg/sentry/fsimpl/sockfs/BUILD index 790d50e65..52084ddb5 100644 --- a/pkg/sentry/fsimpl/sockfs/BUILD +++ b/pkg/sentry/fsimpl/sockfs/BUILD @@ -7,6 +7,7 @@ go_library( srcs = ["sockfs.go"], visibility = ["//pkg/sentry:internal"], deps = [ + "//pkg/abi/linux", "//pkg/context", "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/kernel/auth", diff --git a/pkg/sentry/fsimpl/sockfs/sockfs.go b/pkg/sentry/fsimpl/sockfs/sockfs.go index c13511de2..3f7ad1d65 100644 --- a/pkg/sentry/fsimpl/sockfs/sockfs.go +++ b/pkg/sentry/fsimpl/sockfs/sockfs.go @@ -16,6 +16,7 @@ package sockfs import ( + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" @@ -60,6 +61,10 @@ type filesystem struct { } // inode implements kernfs.Inode. +// +// TODO(gvisor.dev/issue/1476): Add device numbers to this inode (which are +// not included in InodeAttrs) to store the numbers of the appropriate +// socket device. Override InodeAttrs.Stat() accordingly. type inode struct { kernfs.InodeNotDirectory kernfs.InodeNotSymlink @@ -71,3 +76,27 @@ type inode struct { func (i *inode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { return nil, syserror.ENXIO } + +// InitSocket initializes a socket FileDescription, with a corresponding +// Dentry in mnt. +// +// fd should be the FileDescription associated with socketImpl, i.e. its first +// field. mnt should be the global socket mount, Kernel.socketMount. +func InitSocket(socketImpl vfs.FileDescriptionImpl, fd *vfs.FileDescription, mnt *vfs.Mount, creds *auth.Credentials) error { + fsimpl := mnt.Filesystem().Impl() + fs := fsimpl.(*kernfs.Filesystem) + + // File mode matches net/socket.c:sock_alloc. + filemode := linux.FileMode(linux.S_IFSOCK | 0600) + i := &inode{} + i.Init(creds, fs.NextIno(), filemode) + + d := &kernfs.Dentry{} + d.Init(i) + + opts := &vfs.FileDescriptionOptions{UseDentryMetadata: true} + if err := fd.Init(socketImpl, linux.O_RDWR, mnt, d.VFSDentry(), opts); err != nil { + return err + } + return nil +} diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD index bb7e3cbc3..e0ff58d8c 100644 --- a/pkg/sentry/kernel/BUILD +++ b/pkg/sentry/kernel/BUILD @@ -169,6 +169,7 @@ go_library( "//pkg/sentry/fs/lock", "//pkg/sentry/fs/timerfd", "//pkg/sentry/fsbridge", + "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/fsimpl/sockfs", "//pkg/sentry/hostcpu", "//pkg/sentry/inet", diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 06a5b53bc..5d0085462 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -940,7 +940,7 @@ func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr us // GetSockOpt can be used to implement the linux syscall getsockopt(2) for // sockets backed by a commonEndpoint. -func GetSockOpt(t *kernel.Task, s socket.Socket, ep commonEndpoint, family int, skType linux.SockType, level, name, outLen int) (interface{}, *syserr.Error) { +func GetSockOpt(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, family int, skType linux.SockType, level, name, outLen int) (interface{}, *syserr.Error) { switch level { case linux.SOL_SOCKET: return getSockOptSocket(t, s, ep, family, skType, name, outLen) @@ -966,7 +966,7 @@ func GetSockOpt(t *kernel.Task, s socket.Socket, ep commonEndpoint, family int, } // getSockOptSocket implements GetSockOpt when level is SOL_SOCKET. -func getSockOptSocket(t *kernel.Task, s socket.Socket, ep commonEndpoint, family int, skType linux.SockType, name, outLen int) (interface{}, *syserr.Error) { +func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, family int, skType linux.SockType, name, outLen int) (interface{}, *syserr.Error) { // TODO(b/124056281): Stop rejecting short optLen values in getsockopt. switch name { case linux.SO_ERROR: @@ -1541,7 +1541,7 @@ func (s *SocketOperations) SetSockOpt(t *kernel.Task, level int, name int, optVa // SetSockOpt can be used to implement the linux syscall setsockopt(2) for // sockets backed by a commonEndpoint. -func SetSockOpt(t *kernel.Task, s socket.Socket, ep commonEndpoint, level int, name int, optVal []byte) *syserr.Error { +func SetSockOpt(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, level int, name int, optVal []byte) *syserr.Error { switch level { case linux.SOL_SOCKET: return setSockOptSocket(t, s, ep, name, optVal) @@ -1568,7 +1568,7 @@ func SetSockOpt(t *kernel.Task, s socket.Socket, ep commonEndpoint, level int, n } // setSockOptSocket implements SetSockOpt when level is SOL_SOCKET. -func setSockOptSocket(t *kernel.Task, s socket.Socket, ep commonEndpoint, name int, optVal []byte) *syserr.Error { +func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name int, optVal []byte) *syserr.Error { switch name { case linux.SO_SNDBUF: if len(optVal) < sizeOfInt32 { diff --git a/pkg/sentry/socket/unix/BUILD b/pkg/sentry/socket/unix/BUILD index 08743deba..de2cc4bdf 100644 --- a/pkg/sentry/socket/unix/BUILD +++ b/pkg/sentry/socket/unix/BUILD @@ -8,23 +8,27 @@ go_library( "device.go", "io.go", "unix.go", + "unix_vfs2.go", ], visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", "//pkg/context", + "//pkg/fspath", "//pkg/refs", "//pkg/safemem", "//pkg/sentry/arch", "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", + "//pkg/sentry/fsimpl/sockfs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/time", "//pkg/sentry/socket", "//pkg/sentry/socket/control", "//pkg/sentry/socket/netstack", "//pkg/sentry/socket/unix/transport", + "//pkg/sentry/vfs", "//pkg/syserr", "//pkg/syserror", "//pkg/tcpip", diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go index 4d30aa714..7c64f30fa 100644 --- a/pkg/sentry/socket/unix/unix.go +++ b/pkg/sentry/socket/unix/unix.go @@ -23,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" @@ -33,6 +34,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/socket/control" "gvisor.dev/gvisor/pkg/sentry/socket/netstack" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" + "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip" @@ -52,11 +54,8 @@ type SocketOperations struct { fsutil.FileNoSplice `state:"nosave"` fsutil.FileNoopFlush `state:"nosave"` fsutil.FileUseInodeUnstableAttr `state:"nosave"` - refs.AtomicRefCount - socket.SendReceiveTimeout - ep transport.Endpoint - stype linux.SockType + socketOpsCommon } // New creates a new unix socket. @@ -75,16 +74,29 @@ func NewWithDirent(ctx context.Context, d *fs.Dirent, ep transport.Endpoint, sty } s := SocketOperations{ - ep: ep, - stype: stype, + socketOpsCommon: socketOpsCommon{ + ep: ep, + stype: stype, + }, } s.EnableLeakCheck("unix.SocketOperations") return fs.NewFile(ctx, d, flags, &s) } +// socketOpsCommon contains the socket operations common to VFS1 and VFS2. +// +// +stateify savable +type socketOpsCommon struct { + refs.AtomicRefCount + socket.SendReceiveTimeout + + ep transport.Endpoint + stype linux.SockType +} + // DecRef implements RefCounter.DecRef. -func (s *SocketOperations) DecRef() { +func (s *socketOpsCommon) DecRef() { s.DecRefWithDestructor(func() { s.ep.Close() }) @@ -97,7 +109,7 @@ func (s *SocketOperations) Release() { s.DecRef() } -func (s *SocketOperations) isPacket() bool { +func (s *socketOpsCommon) isPacket() bool { switch s.stype { case linux.SOCK_DGRAM, linux.SOCK_SEQPACKET: return true @@ -110,7 +122,7 @@ func (s *SocketOperations) isPacket() bool { } // Endpoint extracts the transport.Endpoint. -func (s *SocketOperations) Endpoint() transport.Endpoint { +func (s *socketOpsCommon) Endpoint() transport.Endpoint { return s.ep } @@ -143,7 +155,7 @@ func extractPath(sockaddr []byte) (string, *syserr.Error) { // GetPeerName implements the linux syscall getpeername(2) for sockets backed by // a transport.Endpoint. -func (s *SocketOperations) GetPeerName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) { +func (s *socketOpsCommon) GetPeerName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) { addr, err := s.ep.GetRemoteAddress() if err != nil { return nil, 0, syserr.TranslateNetstackError(err) @@ -155,7 +167,7 @@ func (s *SocketOperations) GetPeerName(t *kernel.Task) (linux.SockAddr, uint32, // GetSockName implements the linux syscall getsockname(2) for sockets backed by // a transport.Endpoint. -func (s *SocketOperations) GetSockName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) { +func (s *socketOpsCommon) GetSockName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) { addr, err := s.ep.GetLocalAddress() if err != nil { return nil, 0, syserr.TranslateNetstackError(err) @@ -178,7 +190,7 @@ func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr us // Listen implements the linux syscall listen(2) for sockets backed by // a transport.Endpoint. -func (s *SocketOperations) Listen(t *kernel.Task, backlog int) *syserr.Error { +func (s *socketOpsCommon) Listen(t *kernel.Task, backlog int) *syserr.Error { return s.ep.Listen(backlog) } @@ -310,6 +322,8 @@ func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { } // Create the socket. + // + // TODO(gvisor.dev/issue/2324): Correctly set file permissions. childDir, err := d.Bind(t, t.FSContext().RootDirectory(), name, bep, fs.FilePermissions{User: fs.PermMask{Read: true}}) if err != nil { return syserr.ErrPortInUse @@ -345,6 +359,31 @@ func extractEndpoint(t *kernel.Task, sockaddr []byte) (transport.BoundEndpoint, return ep, nil } + if kernel.VFS2Enabled { + p := fspath.Parse(path) + root := t.FSContext().RootDirectoryVFS2() + start := root + relPath := !p.Absolute + if relPath { + start = t.FSContext().WorkingDirectoryVFS2() + } + pop := vfs.PathOperation{ + Root: root, + Start: start, + Path: p, + FollowFinalSymlink: true, + } + ep, e := t.Kernel().VFS().BoundEndpointAt(t, t.Credentials(), &pop) + root.DecRef() + if relPath { + start.DecRef() + } + if e != nil { + return nil, syserr.FromError(e) + } + return ep, nil + } + // Find the node in the filesystem. root := t.FSContext().RootDirectory() cwd := t.FSContext().WorkingDirectory() @@ -363,12 +402,11 @@ func extractEndpoint(t *kernel.Task, sockaddr []byte) (transport.BoundEndpoint, // No socket! return nil, syserr.ErrConnectionRefused } - return ep, nil } // Connect implements the linux syscall connect(2) for unix sockets. -func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking bool) *syserr.Error { +func (s *socketOpsCommon) Connect(t *kernel.Task, sockaddr []byte, blocking bool) *syserr.Error { ep, err := extractEndpoint(t, sockaddr) if err != nil { return err @@ -379,7 +417,7 @@ func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking boo return s.ep.Connect(t, ep) } -// Writev implements fs.FileOperations.Write. +// Write implements fs.FileOperations.Write. func (s *SocketOperations) Write(ctx context.Context, _ *fs.File, src usermem.IOSequence, _ int64) (int64, error) { t := kernel.TaskFromContext(ctx) ctrl := control.New(t, s.ep, nil) @@ -399,7 +437,7 @@ func (s *SocketOperations) Write(ctx context.Context, _ *fs.File, src usermem.IO // SendMsg implements the linux syscall sendmsg(2) for unix sockets backed by // a transport.Endpoint. -func (s *SocketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to []byte, flags int, haveDeadline bool, deadline ktime.Time, controlMessages socket.ControlMessages) (int, *syserr.Error) { +func (s *socketOpsCommon) SendMsg(t *kernel.Task, src usermem.IOSequence, to []byte, flags int, haveDeadline bool, deadline ktime.Time, controlMessages socket.ControlMessages) (int, *syserr.Error) { w := EndpointWriter{ Ctx: t, Endpoint: s.ep, @@ -453,27 +491,27 @@ func (s *SocketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to [] } // Passcred implements transport.Credentialer.Passcred. -func (s *SocketOperations) Passcred() bool { +func (s *socketOpsCommon) Passcred() bool { return s.ep.Passcred() } // ConnectedPasscred implements transport.Credentialer.ConnectedPasscred. -func (s *SocketOperations) ConnectedPasscred() bool { +func (s *socketOpsCommon) ConnectedPasscred() bool { return s.ep.ConnectedPasscred() } // Readiness implements waiter.Waitable.Readiness. -func (s *SocketOperations) Readiness(mask waiter.EventMask) waiter.EventMask { +func (s *socketOpsCommon) Readiness(mask waiter.EventMask) waiter.EventMask { return s.ep.Readiness(mask) } // EventRegister implements waiter.Waitable.EventRegister. -func (s *SocketOperations) EventRegister(e *waiter.Entry, mask waiter.EventMask) { +func (s *socketOpsCommon) EventRegister(e *waiter.Entry, mask waiter.EventMask) { s.ep.EventRegister(e, mask) } // EventUnregister implements waiter.Waitable.EventUnregister. -func (s *SocketOperations) EventUnregister(e *waiter.Entry) { +func (s *socketOpsCommon) EventUnregister(e *waiter.Entry) { s.ep.EventUnregister(e) } @@ -485,7 +523,7 @@ func (s *SocketOperations) SetSockOpt(t *kernel.Task, level int, name int, optVa // Shutdown implements the linux syscall shutdown(2) for sockets backed by // a transport.Endpoint. -func (s *SocketOperations) Shutdown(t *kernel.Task, how int) *syserr.Error { +func (s *socketOpsCommon) Shutdown(t *kernel.Task, how int) *syserr.Error { f, err := netstack.ConvertShutdown(how) if err != nil { return err @@ -511,7 +549,7 @@ func (s *SocketOperations) Read(ctx context.Context, _ *fs.File, dst usermem.IOS // RecvMsg implements the linux syscall recvmsg(2) for sockets backed by // a transport.Endpoint. -func (s *SocketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, haveDeadline bool, deadline ktime.Time, senderRequested bool, controlDataLen uint64) (n int, msgFlags int, senderAddr linux.SockAddr, senderAddrLen uint32, controlMessages socket.ControlMessages, err *syserr.Error) { +func (s *socketOpsCommon) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, haveDeadline bool, deadline ktime.Time, senderRequested bool, controlDataLen uint64) (n int, msgFlags int, senderAddr linux.SockAddr, senderAddrLen uint32, controlMessages socket.ControlMessages, err *syserr.Error) { trunc := flags&linux.MSG_TRUNC != 0 peek := flags&linux.MSG_PEEK != 0 dontWait := flags&linux.MSG_DONTWAIT != 0 @@ -648,12 +686,12 @@ func (s *SocketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags } // State implements socket.Socket.State. -func (s *SocketOperations) State() uint32 { +func (s *socketOpsCommon) State() uint32 { return s.ep.State() } // Type implements socket.Socket.Type. -func (s *SocketOperations) Type() (family int, skType linux.SockType, protocol int) { +func (s *socketOpsCommon) Type() (family int, skType linux.SockType, protocol int) { // Unix domain sockets always have a protocol of 0. return linux.AF_UNIX, s.stype, 0 } @@ -706,4 +744,5 @@ func (*provider) Pair(t *kernel.Task, stype linux.SockType, protocol int) (*fs.F func init() { socket.RegisterProvider(linux.AF_UNIX, &provider{}) + socket.RegisterProviderVFS2(linux.AF_UNIX, &providerVFS2{}) } diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go new file mode 100644 index 000000000..ca1388e2c --- /dev/null +++ b/pkg/sentry/socket/unix/unix_vfs2.go @@ -0,0 +1,348 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package unix + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/socket/control" + "gvisor.dev/gvisor/pkg/sentry/socket/netstack" + "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserr" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +// SocketVFS2 implements socket.SocketVFS2 (and by extension, +// vfs.FileDescriptionImpl) for Unix sockets. +type SocketVFS2 struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl + vfs.DentryMetadataFileDescriptionImpl + + socketOpsCommon +} + +// NewVFS2File creates and returns a new vfs.FileDescription for a unix socket. +func NewVFS2File(t *kernel.Task, ep transport.Endpoint, stype linux.SockType) (*vfs.FileDescription, *syserr.Error) { + sock := NewFDImpl(ep, stype) + vfsfd := &sock.vfsfd + if err := sockfs.InitSocket(sock, vfsfd, t.Kernel().SocketMount(), t.Credentials()); err != nil { + return nil, syserr.FromError(err) + } + return vfsfd, nil +} + +// NewFDImpl creates and returns a new SocketVFS2. +func NewFDImpl(ep transport.Endpoint, stype linux.SockType) *SocketVFS2 { + // You can create AF_UNIX, SOCK_RAW sockets. They're the same as + // SOCK_DGRAM and don't require CAP_NET_RAW. + if stype == linux.SOCK_RAW { + stype = linux.SOCK_DGRAM + } + + return &SocketVFS2{ + socketOpsCommon: socketOpsCommon{ + ep: ep, + stype: stype, + }, + } +} + +// GetSockOpt implements the linux syscall getsockopt(2) for sockets backed by +// a transport.Endpoint. +func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) { + return netstack.GetSockOpt(t, s, s.ep, linux.AF_UNIX, s.ep.Type(), level, name, outLen) +} + +// blockingAccept implements a blocking version of accept(2), that is, if no +// connections are ready to be accept, it will block until one becomes ready. +func (s *SocketVFS2) blockingAccept(t *kernel.Task) (transport.Endpoint, *syserr.Error) { + // Register for notifications. + e, ch := waiter.NewChannelEntry(nil) + s.socketOpsCommon.EventRegister(&e, waiter.EventIn) + defer s.socketOpsCommon.EventUnregister(&e) + + // Try to accept the connection; if it fails, then wait until we get a + // notification. + for { + if ep, err := s.ep.Accept(); err != syserr.ErrWouldBlock { + return ep, err + } + + if err := t.Block(ch); err != nil { + return nil, syserr.FromError(err) + } + } +} + +// Accept implements the linux syscall accept(2) for sockets backed by +// a transport.Endpoint. +func (s *SocketVFS2) Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (int32, linux.SockAddr, uint32, *syserr.Error) { + // Issue the accept request to get the new endpoint. + ep, err := s.ep.Accept() + if err != nil { + if err != syserr.ErrWouldBlock || !blocking { + return 0, nil, 0, err + } + + var err *syserr.Error + ep, err = s.blockingAccept(t) + if err != nil { + return 0, nil, 0, err + } + } + + // We expect this to be a FileDescription here. + ns, err := NewVFS2File(t, ep, s.stype) + if err != nil { + return 0, nil, 0, err + } + defer ns.DecRef() + + if flags&linux.SOCK_NONBLOCK != 0 { + ns.SetStatusFlags(t, t.Credentials(), linux.SOCK_NONBLOCK) + } + + var addr linux.SockAddr + var addrLen uint32 + if peerRequested { + // Get address of the peer. + var err *syserr.Error + addr, addrLen, err = ns.Impl().(*SocketVFS2).GetPeerName(t) + if err != nil { + return 0, nil, 0, err + } + } + + fd, e := t.NewFDFromVFS2(0, ns, kernel.FDFlags{ + CloseOnExec: flags&linux.SOCK_CLOEXEC != 0, + }) + if e != nil { + return 0, nil, 0, syserr.FromError(e) + } + + // TODO: add vfs2 sockets to global table. + return fd, addr, addrLen, nil +} + +// Bind implements the linux syscall bind(2) for unix sockets. +func (s *SocketVFS2) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { + p, e := extractPath(sockaddr) + if e != nil { + return e + } + + bep, ok := s.ep.(transport.BoundEndpoint) + if !ok { + // This socket can't be bound. + return syserr.ErrInvalidArgument + } + + return s.ep.Bind(tcpip.FullAddress{Addr: tcpip.Address(p)}, func() *syserr.Error { + // Is it abstract? + if p[0] == 0 { + if t.IsNetworkNamespaced() { + return syserr.ErrInvalidEndpointState + } + if err := t.AbstractSockets().Bind(p[1:], bep, s); err != nil { + // syserr.ErrPortInUse corresponds to EADDRINUSE. + return syserr.ErrPortInUse + } + } else { + path := fspath.Parse(p) + root := t.FSContext().RootDirectoryVFS2() + defer root.DecRef() + start := root + relPath := !path.Absolute + if relPath { + start = t.FSContext().WorkingDirectoryVFS2() + defer start.DecRef() + } + pop := vfs.PathOperation{ + Root: root, + Start: start, + Path: path, + } + err := t.Kernel().VFS().MknodAt(t, t.Credentials(), &pop, &vfs.MknodOptions{ + // TODO(gvisor.dev/issue/2324): The file permissions should be taken + // from s and t.FSContext().Umask() (see net/unix/af_unix.c:unix_bind), + // but VFS1 just always uses 0400. Resolve this inconsistency. + Mode: linux.S_IFSOCK | 0400, + Endpoint: bep, + }) + if err == syserror.EEXIST { + return syserr.ErrAddressInUse + } + return syserr.FromError(err) + } + + return nil + }) +} + +// Ioctl implements vfs.FileDescriptionImpl. +func (s *SocketVFS2) Ioctl(ctx context.Context, uio usermem.IO, args arch.SyscallArguments) (uintptr, error) { + return netstack.Ioctl(ctx, s.ep, uio, args) +} + +// PRead implements vfs.FileDescriptionImpl. +func (s *SocketVFS2) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + return 0, syserror.ESPIPE +} + +// Read implements vfs.FileDescriptionImpl. +func (s *SocketVFS2) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + // All flags other than RWF_NOWAIT should be ignored. + // TODO(gvisor.dev/issue/1476): Support RWF_NOWAIT. + if opts.Flags != 0 { + return 0, syserror.EOPNOTSUPP + } + + if dst.NumBytes() == 0 { + return 0, nil + } + return dst.CopyOutFrom(ctx, &EndpointReader{ + Ctx: ctx, + Endpoint: s.ep, + NumRights: 0, + Peek: false, + From: nil, + }) +} + +// PWrite implements vfs.FileDescriptionImpl. +func (s *SocketVFS2) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + return 0, syserror.ESPIPE +} + +// Write implements vfs.FileDescriptionImpl. +func (s *SocketVFS2) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + // All flags other than RWF_NOWAIT should be ignored. + // TODO(gvisor.dev/issue/1476): Support RWF_NOWAIT. + if opts.Flags != 0 { + return 0, syserror.EOPNOTSUPP + } + + t := kernel.TaskFromContext(ctx) + ctrl := control.New(t, s.ep, nil) + + if src.NumBytes() == 0 { + nInt, err := s.ep.SendMsg(ctx, [][]byte{}, ctrl, nil) + return int64(nInt), err.ToError() + } + + return src.CopyInTo(ctx, &EndpointWriter{ + Ctx: ctx, + Endpoint: s.ep, + Control: ctrl, + To: nil, + }) +} + +// Release implements vfs.FileDescriptionImpl. +func (s *SocketVFS2) Release() { + // Release only decrements a reference on s because s may be referenced in + // the abstract socket namespace. + s.DecRef() +} + +// Readiness implements waiter.Waitable.Readiness. +func (s *SocketVFS2) Readiness(mask waiter.EventMask) waiter.EventMask { + return s.socketOpsCommon.Readiness(mask) +} + +// EventRegister implements waiter.Waitable.EventRegister. +func (s *SocketVFS2) EventRegister(e *waiter.Entry, mask waiter.EventMask) { + s.socketOpsCommon.EventRegister(e, mask) +} + +// EventUnregister implements waiter.Waitable.EventUnregister. +func (s *SocketVFS2) EventUnregister(e *waiter.Entry) { + s.socketOpsCommon.EventUnregister(e) +} + +// SetSockOpt implements the linux syscall setsockopt(2) for sockets backed by +// a transport.Endpoint. +func (s *SocketVFS2) SetSockOpt(t *kernel.Task, level int, name int, optVal []byte) *syserr.Error { + return netstack.SetSockOpt(t, s, s.ep, level, name, optVal) +} + +// providerVFS2 is a unix domain socket provider for VFS2. +type providerVFS2 struct{} + +func (*providerVFS2) Socket(t *kernel.Task, stype linux.SockType, protocol int) (*vfs.FileDescription, *syserr.Error) { + // Check arguments. + if protocol != 0 && protocol != linux.AF_UNIX /* PF_UNIX */ { + return nil, syserr.ErrProtocolNotSupported + } + + // Create the endpoint and socket. + var ep transport.Endpoint + switch stype { + case linux.SOCK_DGRAM, linux.SOCK_RAW: + ep = transport.NewConnectionless(t) + case linux.SOCK_SEQPACKET, linux.SOCK_STREAM: + ep = transport.NewConnectioned(t, stype, t.Kernel()) + default: + return nil, syserr.ErrInvalidArgument + } + + f, err := NewVFS2File(t, ep, stype) + if err != nil { + ep.Close() + return nil, err + } + return f, nil +} + +// Pair creates a new pair of AF_UNIX connected sockets. +func (*providerVFS2) Pair(t *kernel.Task, stype linux.SockType, protocol int) (*vfs.FileDescription, *vfs.FileDescription, *syserr.Error) { + // Check arguments. + if protocol != 0 && protocol != linux.AF_UNIX /* PF_UNIX */ { + return nil, nil, syserr.ErrProtocolNotSupported + } + + switch stype { + case linux.SOCK_STREAM, linux.SOCK_DGRAM, linux.SOCK_SEQPACKET, linux.SOCK_RAW: + // Ok + default: + return nil, nil, syserr.ErrInvalidArgument + } + + // Create the endpoints and sockets. + ep1, ep2 := transport.NewPair(t, stype, t.Kernel()) + s1, err := NewVFS2File(t, ep1, stype) + if err != nil { + ep1.Close() + ep2.Close() + return nil, nil, err + } + s2, err := NewVFS2File(t, ep2, stype) + if err != nil { + s1.DecRef() + ep2.Close() + return nil, nil, err + } + + return s1, s2, nil +} diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go index 3e90dc4ed..2f04bf882 100644 --- a/pkg/sentry/vfs/options.go +++ b/pkg/sentry/vfs/options.go @@ -16,6 +16,7 @@ package vfs import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" ) // GetDentryOptions contains options to VirtualFilesystem.GetDentryAt() and @@ -44,6 +45,10 @@ type MknodOptions struct { // DevMinor are the major and minor device numbers for the created device. DevMajor uint32 DevMinor uint32 + + // Endpoint is the endpoint to bind to the created file, if a socket file is + // being created for bind(2) on a Unix domain socket. + Endpoint transport.BoundEndpoint } // MountFlags contains flags as specified for mount(2), e.g. MS_NOEXEC. -- cgit v1.2.3 From fc99a7ebf0c24b6f7b3cfd6351436373ed54548b Mon Sep 17 00:00:00 2001 From: Bhasker Hariharan Date: Fri, 3 Apr 2020 18:34:48 -0700 Subject: Refactor software GSO code. Software GSO implementation currently has a complicated code path with implicit assumptions that all packets to WritePackets carry same Data and it does this to avoid allocations on the path etc. But this makes it hard to reuse the WritePackets API. This change breaks all such assumptions by introducing a new Vectorised View API ReadToVV which can be used to cleanly split a VV into multiple independent VVs. Further this change also makes packet buffers linkable to form an intrusive list. This allows us to get rid of the array of packet buffers that are passed in the WritePackets API call and replace it with a list of packet buffers. While this code does introduce some more allocations in the benchmarks it doesn't cause any degradation. Updates #231 PiperOrigin-RevId: 304731742 --- pkg/ilist/list.go | 13 ++- pkg/sentry/kernel/kernel.go | 22 +++-- pkg/tcpip/buffer/view.go | 53 +++++++++- pkg/tcpip/buffer/view_test.go | 137 ++++++++++++++++++++++++++ pkg/tcpip/link/channel/channel.go | 18 ++-- pkg/tcpip/link/fdbased/endpoint.go | 162 ++++++++++++++----------------- pkg/tcpip/link/loopback/loopback.go | 2 +- pkg/tcpip/link/muxed/injectable.go | 2 +- pkg/tcpip/link/sharedmem/sharedmem.go | 2 +- pkg/tcpip/link/sniffer/sniffer.go | 14 +-- pkg/tcpip/link/waitable/waitable.go | 4 +- pkg/tcpip/link/waitable/waitable_test.go | 6 +- pkg/tcpip/network/arp/arp.go | 2 +- pkg/tcpip/network/ip_test.go | 2 +- pkg/tcpip/network/ipv4/ipv4.go | 37 +++++-- pkg/tcpip/network/ipv6/icmp.go | 2 +- pkg/tcpip/network/ipv6/ipv6.go | 12 +-- pkg/tcpip/stack/BUILD | 14 ++- pkg/tcpip/stack/forwarder_test.go | 8 +- pkg/tcpip/stack/iptables.go | 17 ++++ pkg/tcpip/stack/ndp_test.go | 2 +- pkg/tcpip/stack/packet_buffer.go | 14 +-- pkg/tcpip/stack/packet_buffer_state.go | 27 ------ pkg/tcpip/stack/registration.go | 4 +- pkg/tcpip/stack/route.go | 19 ++-- pkg/tcpip/stack/stack_test.go | 2 +- pkg/tcpip/transport/tcp/connect.go | 47 +++++---- pkg/tcpip/transport/tcp/segment.go | 6 +- 28 files changed, 420 insertions(+), 230 deletions(-) delete mode 100644 pkg/tcpip/stack/packet_buffer_state.go (limited to 'pkg/sentry') diff --git a/pkg/ilist/list.go b/pkg/ilist/list.go index 8f93e4d6d..0d07da3b1 100644 --- a/pkg/ilist/list.go +++ b/pkg/ilist/list.go @@ -86,12 +86,21 @@ func (l *List) Back() Element { return l.tail } +// Len returns the number of elements in the list. +// +// NOTE: This is an O(n) operation. +func (l *List) Len() (count int) { + for e := l.Front(); e != nil; e = e.Next() { + count++ + } + return count +} + // PushFront inserts the element e at the front of list l. func (l *List) PushFront(e Element) { linker := ElementMapper{}.linkerFor(e) linker.SetNext(l.head) linker.SetPrev(nil) - if l.head != nil { ElementMapper{}.linkerFor(l.head).SetPrev(e) } else { @@ -106,7 +115,6 @@ func (l *List) PushBack(e Element) { linker := ElementMapper{}.linkerFor(e) linker.SetNext(nil) linker.SetPrev(l.tail) - if l.tail != nil { ElementMapper{}.linkerFor(l.tail).SetNext(e) } else { @@ -127,7 +135,6 @@ func (l *List) PushBackList(m *List) { l.tail = m.tail } - m.head = nil m.tail = nil } diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 0a448b57c..2e6f42b92 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -564,15 +564,25 @@ func (ts *TaskSet) unregisterEpollWaiters() { ts.mu.RLock() defer ts.mu.RUnlock() + + // Tasks that belong to the same process could potentially point to the + // same FDTable. So we retain a map of processed ones to avoid + // processing the same FDTable multiple times. + processed := make(map[*FDTable]struct{}) for t := range ts.Root.tids { // We can skip locking Task.mu here since the kernel is paused. - if t.fdTable != nil { - t.fdTable.forEach(func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) { - if e, ok := file.FileOperations.(*epoll.EventPoll); ok { - e.UnregisterEpollWaiters() - } - }) + if t.fdTable == nil { + continue + } + if _, ok := processed[t.fdTable]; ok { + continue } + t.fdTable.forEach(func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) { + if e, ok := file.FileOperations.(*epoll.EventPoll); ok { + e.UnregisterEpollWaiters() + } + }) + processed[t.fdTable] = struct{}{} } } diff --git a/pkg/tcpip/buffer/view.go b/pkg/tcpip/buffer/view.go index 8d42cd066..8ec5d5d5c 100644 --- a/pkg/tcpip/buffer/view.go +++ b/pkg/tcpip/buffer/view.go @@ -17,6 +17,7 @@ package buffer import ( "bytes" + "io" ) // View is a slice of a buffer, with convenience methods. @@ -89,6 +90,47 @@ func (vv *VectorisedView) TrimFront(count int) { } } +// Read implements io.Reader. +func (vv *VectorisedView) Read(v View) (copied int, err error) { + count := len(v) + for count > 0 && len(vv.views) > 0 { + if count < len(vv.views[0]) { + vv.size -= count + copy(v[copied:], vv.views[0][:count]) + vv.views[0].TrimFront(count) + copied += count + return copied, nil + } + count -= len(vv.views[0]) + copy(v[copied:], vv.views[0]) + copied += len(vv.views[0]) + vv.RemoveFirst() + } + if copied == 0 { + return 0, io.EOF + } + return copied, nil +} + +// ReadToVV reads up to n bytes from vv to dstVV and removes them from vv. It +// returns the number of bytes copied. +func (vv *VectorisedView) ReadToVV(dstVV *VectorisedView, count int) (copied int) { + for count > 0 && len(vv.views) > 0 { + if count < len(vv.views[0]) { + vv.size -= count + dstVV.AppendView(vv.views[0][:count]) + vv.views[0].TrimFront(count) + copied += count + return + } + count -= len(vv.views[0]) + dstVV.AppendView(vv.views[0]) + copied += len(vv.views[0]) + vv.RemoveFirst() + } + return copied +} + // CapLength irreversibly reduces the length of the vectorised view. func (vv *VectorisedView) CapLength(length int) { if length < 0 { @@ -116,12 +158,12 @@ func (vv *VectorisedView) CapLength(length int) { // Clone returns a clone of this VectorisedView. // If the buffer argument is large enough to contain all the Views of this VectorisedView, // the method will avoid allocations and use the buffer to store the Views of the clone. -func (vv VectorisedView) Clone(buffer []View) VectorisedView { +func (vv *VectorisedView) Clone(buffer []View) VectorisedView { return VectorisedView{views: append(buffer[:0], vv.views...), size: vv.size} } // First returns the first view of the vectorised view. -func (vv VectorisedView) First() View { +func (vv *VectorisedView) First() View { if len(vv.views) == 0 { return nil } @@ -134,11 +176,12 @@ func (vv *VectorisedView) RemoveFirst() { return } vv.size -= len(vv.views[0]) + vv.views[0] = nil vv.views = vv.views[1:] } // Size returns the size in bytes of the entire content stored in the vectorised view. -func (vv VectorisedView) Size() int { +func (vv *VectorisedView) Size() int { return vv.size } @@ -146,7 +189,7 @@ func (vv VectorisedView) Size() int { // // If the vectorised view contains a single view, that view will be returned // directly. -func (vv VectorisedView) ToView() View { +func (vv *VectorisedView) ToView() View { if len(vv.views) == 1 { return vv.views[0] } @@ -158,7 +201,7 @@ func (vv VectorisedView) ToView() View { } // Views returns the slice containing the all views. -func (vv VectorisedView) Views() []View { +func (vv *VectorisedView) Views() []View { return vv.views } diff --git a/pkg/tcpip/buffer/view_test.go b/pkg/tcpip/buffer/view_test.go index ebc3a17b7..106e1994c 100644 --- a/pkg/tcpip/buffer/view_test.go +++ b/pkg/tcpip/buffer/view_test.go @@ -233,3 +233,140 @@ func TestToClone(t *testing.T) { }) } } + +func TestVVReadToVV(t *testing.T) { + testCases := []struct { + comment string + vv VectorisedView + bytesToRead int + wantBytes string + leftVV VectorisedView + }{ + { + comment: "large VV, short read", + vv: vv(30, "012345678901234567890123456789"), + bytesToRead: 10, + wantBytes: "0123456789", + leftVV: vv(20, "01234567890123456789"), + }, + { + comment: "largeVV, multiple views, short read", + vv: vv(13, "123", "345", "567", "8910"), + bytesToRead: 6, + wantBytes: "123345", + leftVV: vv(7, "567", "8910"), + }, + { + comment: "smallVV (multiple views), large read", + vv: vv(3, "1", "2", "3"), + bytesToRead: 10, + wantBytes: "123", + leftVV: vv(0, ""), + }, + { + comment: "smallVV (single view), large read", + vv: vv(1, "1"), + bytesToRead: 10, + wantBytes: "1", + leftVV: vv(0, ""), + }, + { + comment: "emptyVV, large read", + vv: vv(0, ""), + bytesToRead: 10, + wantBytes: "", + leftVV: vv(0, ""), + }, + } + + for _, tc := range testCases { + t.Run(tc.comment, func(t *testing.T) { + var readTo VectorisedView + inSize := tc.vv.Size() + copied := tc.vv.ReadToVV(&readTo, tc.bytesToRead) + if got, want := copied, len(tc.wantBytes); got != want { + t.Errorf("incorrect number of bytes copied returned in ReadToVV got: %d, want: %d, tc: %+v", got, want, tc) + } + if got, want := string(readTo.ToView()), tc.wantBytes; got != want { + t.Errorf("unexpected content in readTo got: %s, want: %s", got, want) + } + if got, want := tc.vv.Size(), inSize-copied; got != want { + t.Errorf("test VV has incorrect size after reading got: %d, want: %d, tc.vv: %+v", got, want, tc.vv) + } + if got, want := string(tc.vv.ToView()), string(tc.leftVV.ToView()); got != want { + t.Errorf("unexpected data left in vv after read got: %+v, want: %+v", got, want) + } + }) + } +} + +func TestVVRead(t *testing.T) { + testCases := []struct { + comment string + vv VectorisedView + bytesToRead int + readBytes string + leftBytes string + wantError bool + }{ + { + comment: "large VV, short read", + vv: vv(30, "012345678901234567890123456789"), + bytesToRead: 10, + readBytes: "0123456789", + leftBytes: "01234567890123456789", + }, + { + comment: "largeVV, multiple buffers, short read", + vv: vv(13, "123", "345", "567", "8910"), + bytesToRead: 6, + readBytes: "123345", + leftBytes: "5678910", + }, + { + comment: "smallVV, large read", + vv: vv(3, "1", "2", "3"), + bytesToRead: 10, + readBytes: "123", + leftBytes: "", + }, + { + comment: "smallVV, large read", + vv: vv(1, "1"), + bytesToRead: 10, + readBytes: "1", + leftBytes: "", + }, + { + comment: "emptyVV, large read", + vv: vv(0, ""), + bytesToRead: 10, + readBytes: "", + wantError: true, + }, + } + + for _, tc := range testCases { + t.Run(tc.comment, func(t *testing.T) { + readTo := NewView(tc.bytesToRead) + inSize := tc.vv.Size() + copied, err := tc.vv.Read(readTo) + if !tc.wantError && err != nil { + t.Fatalf("unexpected error in tc.vv.Read(..) = %s", err) + } + readTo = readTo[:copied] + if got, want := copied, len(tc.readBytes); got != want { + t.Errorf("incorrect number of bytes copied returned in ReadToVV got: %d, want: %d, tc.vv: %+v", got, want, tc.vv) + } + if got, want := string(readTo), tc.readBytes; got != want { + t.Errorf("unexpected data in readTo got: %s, want: %s", got, want) + } + if got, want := tc.vv.Size(), inSize-copied; got != want { + t.Errorf("test VV has incorrect size after reading got: %d, want: %d, tc.vv: %+v", got, want, tc.vv) + } + if got, want := string(tc.vv.ToView()), tc.leftBytes; got != want { + t.Errorf("vv has incorrect data after Read got: %s, want: %s", got, want) + } + }) + } +} diff --git a/pkg/tcpip/link/channel/channel.go b/pkg/tcpip/link/channel/channel.go index a8d6653ce..b4a0ae53d 100644 --- a/pkg/tcpip/link/channel/channel.go +++ b/pkg/tcpip/link/channel/channel.go @@ -28,7 +28,7 @@ import ( // PacketInfo holds all the information about an outbound packet. type PacketInfo struct { - Pkt stack.PacketBuffer + Pkt *stack.PacketBuffer Proto tcpip.NetworkProtocolNumber GSO *stack.GSO Route stack.Route @@ -257,7 +257,7 @@ func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne route := r.Clone() route.Release() p := PacketInfo{ - Pkt: pkt, + Pkt: &pkt, Proto: protocol, GSO: gso, Route: route, @@ -269,21 +269,15 @@ func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne } // WritePackets stores outbound packets into the channel. -func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []stack.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { // Clone r then release its resource so we only get the relevant fields from // stack.Route without holding a reference to a NIC's endpoint. route := r.Clone() route.Release() - payloadView := pkts[0].Data.ToView() n := 0 - for _, pkt := range pkts { - off := pkt.DataOffset - size := pkt.DataSize + for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() { p := PacketInfo{ - Pkt: stack.PacketBuffer{ - Header: pkt.Header, - Data: buffer.NewViewFromBytes(payloadView[off : off+size]).ToVectorisedView(), - }, + Pkt: pkt, Proto: protocol, GSO: gso, Route: route, @@ -301,7 +295,7 @@ func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []stack.Pac // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. func (e *Endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { p := PacketInfo{ - Pkt: stack.PacketBuffer{Data: vv}, + Pkt: &stack.PacketBuffer{Data: vv}, Proto: 0, GSO: nil, } diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go index 3b3b6909b..7198742b7 100644 --- a/pkg/tcpip/link/fdbased/endpoint.go +++ b/pkg/tcpip/link/fdbased/endpoint.go @@ -441,118 +441,106 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne // WritePackets writes outbound packets to the file descriptor. If it is not // currently writable, the packet is dropped. -func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []stack.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { - var ethHdrBuf []byte - // hdr + data - iovLen := 2 - if e.hdrSize > 0 { - // Add ethernet header if needed. - ethHdrBuf = make([]byte, header.EthernetMinimumSize) - eth := header.Ethernet(ethHdrBuf) - ethHdr := &header.EthernetFields{ - DstAddr: r.RemoteLinkAddress, - Type: protocol, - } - - // Preserve the src address if it's set in the route. - if r.LocalLinkAddress != "" { - ethHdr.SrcAddr = r.LocalLinkAddress - } else { - ethHdr.SrcAddr = e.addr - } - eth.Encode(ethHdr) - iovLen++ - } +// +// NOTE: This API uses sendmmsg to batch packets. As a result the underlying FD +// picked to write the packet out has to be the same for all packets in the +// list. In other words all packets in the batch should belong to the same +// flow. +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + n := pkts.Len() - n := len(pkts) - - views := pkts[0].Data.Views() - /* - * Each boundary in views can add one more iovec. - * - * payload | | | | - * ----------------------------- - * packets | | | | | | | - * ----------------------------- - * iovecs | | | | | | | | | - */ - iovec := make([]syscall.Iovec, n*iovLen+len(views)-1) mmsgHdrs := make([]rawfile.MMsgHdr, n) + i := 0 + for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() { + var ethHdrBuf []byte + iovLen := 0 + if e.hdrSize > 0 { + // Add ethernet header if needed. + ethHdrBuf = make([]byte, header.EthernetMinimumSize) + eth := header.Ethernet(ethHdrBuf) + ethHdr := &header.EthernetFields{ + DstAddr: r.RemoteLinkAddress, + Type: protocol, + } - iovecIdx := 0 - viewIdx := 0 - viewOff := 0 - off := 0 - nextOff := 0 - for i := range pkts { - // TODO(b/134618279): Different packets may have different data - // in the future. We should handle this. - if !viewsEqual(pkts[i].Data.Views(), views) { - panic("All packets in pkts should have the same Data.") + // Preserve the src address if it's set in the route. + if r.LocalLinkAddress != "" { + ethHdr.SrcAddr = r.LocalLinkAddress + } else { + ethHdr.SrcAddr = e.addr + } + eth.Encode(ethHdr) + iovLen++ } - prevIovecIdx := iovecIdx - mmsgHdr := &mmsgHdrs[i] - mmsgHdr.Msg.Iov = &iovec[iovecIdx] - packetSize := pkts[i].DataSize - hdr := &pkts[i].Header - - off = pkts[i].DataOffset - if off != nextOff { - // We stop in a different point last time. - size := packetSize - viewIdx = 0 - viewOff = 0 - for size > 0 { - if size >= len(views[viewIdx]) { - viewIdx++ - viewOff = 0 - size -= len(views[viewIdx]) - } else { - viewOff = size - size = 0 + var vnetHdrBuf []byte + vnetHdr := virtioNetHdr{} + if e.Capabilities()&stack.CapabilityHardwareGSO != 0 { + if gso != nil { + vnetHdr.hdrLen = uint16(pkt.Header.UsedLength()) + if gso.NeedsCsum { + vnetHdr.flags = _VIRTIO_NET_HDR_F_NEEDS_CSUM + vnetHdr.csumStart = header.EthernetMinimumSize + gso.L3HdrLen + vnetHdr.csumOffset = gso.CsumOffset + } + if gso.Type != stack.GSONone && uint16(pkt.Data.Size()) > gso.MSS { + switch gso.Type { + case stack.GSOTCPv4: + vnetHdr.gsoType = _VIRTIO_NET_HDR_GSO_TCPV4 + case stack.GSOTCPv6: + vnetHdr.gsoType = _VIRTIO_NET_HDR_GSO_TCPV6 + default: + panic(fmt.Sprintf("Unknown gso type: %v", gso.Type)) + } + vnetHdr.gsoSize = gso.MSS } } + vnetHdrBuf = vnetHdrToByteSlice(&vnetHdr) + iovLen++ } - nextOff = off + packetSize + iovecs := make([]syscall.Iovec, iovLen+1+len(pkt.Data.Views())) + mmsgHdr := &mmsgHdrs[i] + mmsgHdr.Msg.Iov = &iovecs[0] + iovecIdx := 0 + if vnetHdrBuf != nil { + v := &iovecs[iovecIdx] + v.Base = &vnetHdrBuf[0] + v.Len = uint64(len(vnetHdrBuf)) + iovecIdx++ + } if ethHdrBuf != nil { - v := &iovec[iovecIdx] + v := &iovecs[iovecIdx] v.Base = ðHdrBuf[0] v.Len = uint64(len(ethHdrBuf)) iovecIdx++ } - - v := &iovec[iovecIdx] + pktSize := uint64(0) + // Encode L3 Header + v := &iovecs[iovecIdx] + hdr := &pkt.Header hdrView := hdr.View() v.Base = &hdrView[0] v.Len = uint64(len(hdrView)) + pktSize += v.Len iovecIdx++ - for packetSize > 0 { - vec := &iovec[iovecIdx] + // Now encode the Transport Payload. + pktViews := pkt.Data.Views() + for i := range pktViews { + vec := &iovecs[iovecIdx] iovecIdx++ - - v := views[viewIdx] - vec.Base = &v[viewOff] - s := len(v) - viewOff - if s <= packetSize { - viewIdx++ - viewOff = 0 - } else { - s = packetSize - viewOff += s - } - vec.Len = uint64(s) - packetSize -= s + vec.Base = &pktViews[i][0] + vec.Len = uint64(len(pktViews[i])) + pktSize += vec.Len } - - mmsgHdr.Msg.Iovlen = uint64(iovecIdx - prevIovecIdx) + mmsgHdr.Msg.Iovlen = uint64(iovecIdx) + i++ } packets := 0 for packets < n { - fd := e.fds[pkts[packets].Hash%uint32(len(e.fds))] + fd := e.fds[pkts.Front().Hash%uint32(len(e.fds))] sent, err := rawfile.NonBlockingSendMMsg(fd, mmsgHdrs) if err != nil { return packets, err diff --git a/pkg/tcpip/link/loopback/loopback.go b/pkg/tcpip/link/loopback/loopback.go index 4039753b7..1e2255bfa 100644 --- a/pkg/tcpip/link/loopback/loopback.go +++ b/pkg/tcpip/link/loopback/loopback.go @@ -92,7 +92,7 @@ func (e *endpoint) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.Netw } // WritePackets implements stack.LinkEndpoint.WritePackets. -func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, []stack.PacketBuffer, tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, stack.PacketBufferList, tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { panic("not implemented") } diff --git a/pkg/tcpip/link/muxed/injectable.go b/pkg/tcpip/link/muxed/injectable.go index f5973066d..a5478ce17 100644 --- a/pkg/tcpip/link/muxed/injectable.go +++ b/pkg/tcpip/link/muxed/injectable.go @@ -87,7 +87,7 @@ func (m *InjectableEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, // WritePackets writes outbound packets to the appropriate // LinkInjectableEndpoint based on the RemoteAddress. HandleLocal only works if // r.RemoteAddress has a route registered in this endpoint. -func (m *InjectableEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []stack.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (m *InjectableEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { endpoint, ok := m.routes[r.RemoteAddress] if !ok { return 0, tcpip.ErrNoRoute diff --git a/pkg/tcpip/link/sharedmem/sharedmem.go b/pkg/tcpip/link/sharedmem/sharedmem.go index 6461d0108..0796d717e 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem.go +++ b/pkg/tcpip/link/sharedmem/sharedmem.go @@ -214,7 +214,7 @@ func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.Netw } // WritePackets implements stack.LinkEndpoint.WritePackets. -func (e *endpoint) WritePackets(r *stack.Route, _ *stack.GSO, pkts []stack.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (e *endpoint) WritePackets(r *stack.Route, _ *stack.GSO, pkts stack.PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { panic("not implemented") } diff --git a/pkg/tcpip/link/sniffer/sniffer.go b/pkg/tcpip/link/sniffer/sniffer.go index 0a6b8945c..062388f4d 100644 --- a/pkg/tcpip/link/sniffer/sniffer.go +++ b/pkg/tcpip/link/sniffer/sniffer.go @@ -200,7 +200,7 @@ func (e *endpoint) GSOMaxSize() uint32 { return 0 } -func (e *endpoint) dumpPacket(gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { +func (e *endpoint) dumpPacket(gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { if atomic.LoadUint32(&LogPackets) == 1 && e.file == nil { logPacket("send", protocol, pkt.Header.View(), gso) } @@ -233,20 +233,16 @@ func (e *endpoint) dumpPacket(gso *stack.GSO, protocol tcpip.NetworkProtocolNumb // higher-level protocols to write packets; it just logs the packet and // forwards the request to the lower endpoint. func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { - e.dumpPacket(gso, protocol, pkt) + e.dumpPacket(gso, protocol, &pkt) return e.lower.WritePacket(r, gso, protocol, pkt) } // WritePackets implements the stack.LinkEndpoint interface. It is called by // higher-level protocols to write packets; it just logs the packet and // forwards the request to the lower endpoint. -func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []stack.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { - view := pkts[0].Data.ToView() - for _, pkt := range pkts { - e.dumpPacket(gso, protocol, stack.PacketBuffer{ - Header: pkt.Header, - Data: view[pkt.DataOffset:][:pkt.DataSize].ToVectorisedView(), - }) +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() { + e.dumpPacket(gso, protocol, pkt) } return e.lower.WritePackets(r, gso, pkts, protocol) } diff --git a/pkg/tcpip/link/waitable/waitable.go b/pkg/tcpip/link/waitable/waitable.go index 52fe397bf..2b3741276 100644 --- a/pkg/tcpip/link/waitable/waitable.go +++ b/pkg/tcpip/link/waitable/waitable.go @@ -112,9 +112,9 @@ func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne // WritePackets implements stack.LinkEndpoint.WritePackets. It is called by // higher-level protocols to write packets. It only forwards packets to the // lower endpoint if Wait or WaitWrite haven't been called. -func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []stack.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { if !e.writeGate.Enter() { - return len(pkts), nil + return pkts.Len(), nil } n, err := e.lower.WritePackets(r, gso, pkts, protocol) diff --git a/pkg/tcpip/link/waitable/waitable_test.go b/pkg/tcpip/link/waitable/waitable_test.go index 88224e494..54eb5322b 100644 --- a/pkg/tcpip/link/waitable/waitable_test.go +++ b/pkg/tcpip/link/waitable/waitable_test.go @@ -71,9 +71,9 @@ func (e *countedEndpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcp } // WritePackets implements stack.LinkEndpoint.WritePackets. -func (e *countedEndpoint) WritePackets(r *stack.Route, _ *stack.GSO, pkts []stack.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { - e.writeCount += len(pkts) - return len(pkts), nil +func (e *countedEndpoint) WritePackets(r *stack.Route, _ *stack.GSO, pkts stack.PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + e.writeCount += pkts.Len() + return pkts.Len(), nil } func (e *countedEndpoint) WriteRawPacket(buffer.VectorisedView) *tcpip.Error { diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index 255098372..7acbfa0a8 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -84,7 +84,7 @@ func (e *endpoint) WritePacket(*stack.Route, *stack.GSO, stack.NetworkHeaderPara } // WritePackets implements stack.NetworkEndpoint.WritePackets. -func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, []stack.PacketBuffer, stack.NetworkHeaderParams) (int, *tcpip.Error) { +func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, stack.PacketBufferList, stack.NetworkHeaderParams) (int, *tcpip.Error) { return 0, tcpip.ErrNotSupported } diff --git a/pkg/tcpip/network/ip_test.go b/pkg/tcpip/network/ip_test.go index 4950d69fc..4c20301c6 100644 --- a/pkg/tcpip/network/ip_test.go +++ b/pkg/tcpip/network/ip_test.go @@ -172,7 +172,7 @@ func (t *testObject) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.Ne } // WritePackets implements stack.LinkEndpoint.WritePackets. -func (t *testObject) WritePackets(_ *stack.Route, _ *stack.GSO, pkt []stack.PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (t *testObject) WritePackets(_ *stack.Route, _ *stack.GSO, pkt stack.PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { panic("not implemented") } diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index a7d9a8b25..104aafbed 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -280,28 +280,47 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw } // WritePackets implements stack.NetworkEndpoint.WritePackets. -func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []stack.PacketBuffer, params stack.NetworkHeaderParams) (int, *tcpip.Error) { +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.PacketBufferList, params stack.NetworkHeaderParams) (int, *tcpip.Error) { if r.Loop&stack.PacketLoop != 0 { panic("multiple packets in local loop") } if r.Loop&stack.PacketOut == 0 { - return len(pkts), nil + return pkts.Len(), nil + } + + for pkt := pkts.Front(); pkt != nil; { + ip := e.addIPHeader(r, &pkt.Header, pkt.Data.Size(), params) + pkt.NetworkHeader = buffer.View(ip) + pkt = pkt.Next() } // iptables filtering. All packets that reach here are locally // generated. ipt := e.stack.IPTables() - for i := range pkts { - if ok := ipt.Check(stack.Output, pkts[i]); !ok { - // iptables is telling us to drop the packet. + dropped := ipt.CheckPackets(stack.Output, pkts) + if len(dropped) == 0 { + // Fast path: If no packets are to be dropped then we can just invoke the + // faster WritePackets API directly. + n, err := e.linkEP.WritePackets(r, gso, pkts, ProtocolNumber) + r.Stats().IP.PacketsSent.IncrementBy(uint64(n)) + return n, err + } + + // Slow Path as we are dropping some packets in the batch degrade to + // emitting one packet at a time. + n := 0 + for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() { + if _, ok := dropped[pkt]; ok { continue } - ip := e.addIPHeader(r, &pkts[i].Header, pkts[i].DataSize, params) - pkts[i].NetworkHeader = buffer.View(ip) + if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, *pkt); err != nil { + r.Stats().IP.PacketsSent.IncrementBy(uint64(n)) + return n, err + } + n++ } - n, err := e.linkEP.WritePackets(r, gso, pkts, ProtocolNumber) r.Stats().IP.PacketsSent.IncrementBy(uint64(n)) - return n, err + return n, nil } // WriteHeaderIncludedPacket writes a packet already containing a network diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index 6d2d2c034..f91180aa3 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -79,7 +79,7 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P // Only the first view in vv is accounted for by h. To account for the // rest of vv, a shallow copy is made and the first view is removed. // This copy is used as extra payload during the checksum calculation. - payload := pkt.Data + payload := pkt.Data.Clone(nil) payload.RemoveFirst() if got, want := h.Checksum(), header.ICMPv6Checksum(h, iph.SourceAddress(), iph.DestinationAddress(), payload); got != want { received.Invalid.Increment() diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index b462b8604..a815b4d9b 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -143,19 +143,17 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw } // WritePackets implements stack.LinkEndpoint.WritePackets. -func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []stack.PacketBuffer, params stack.NetworkHeaderParams) (int, *tcpip.Error) { +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.PacketBufferList, params stack.NetworkHeaderParams) (int, *tcpip.Error) { if r.Loop&stack.PacketLoop != 0 { panic("not implemented") } if r.Loop&stack.PacketOut == 0 { - return len(pkts), nil + return pkts.Len(), nil } - for i := range pkts { - hdr := &pkts[i].Header - size := pkts[i].DataSize - ip := e.addIPHeader(r, hdr, size, params) - pkts[i].NetworkHeader = buffer.View(ip) + for pb := pkts.Front(); pb != nil; pb = pb.Next() { + ip := e.addIPHeader(r, &pb.Header, pb.Data.Size(), params) + pb.NetworkHeader = buffer.View(ip) } n, err := e.linkEP.WritePackets(r, gso, pkts, ProtocolNumber) diff --git a/pkg/tcpip/stack/BUILD b/pkg/tcpip/stack/BUILD index 8d80e9cee..5e963a4af 100644 --- a/pkg/tcpip/stack/BUILD +++ b/pkg/tcpip/stack/BUILD @@ -15,6 +15,18 @@ go_template_instance( }, ) +go_template_instance( + name = "packet_buffer_list", + out = "packet_buffer_list.go", + package = "stack", + prefix = "PacketBuffer", + template = "//pkg/ilist:generic_list", + types = { + "Element": "*PacketBuffer", + "Linker": "*PacketBuffer", + }, +) + go_library( name = "stack", srcs = [ @@ -29,7 +41,7 @@ go_library( "ndp.go", "nic.go", "packet_buffer.go", - "packet_buffer_state.go", + "packet_buffer_list.go", "rand.go", "registration.go", "route.go", diff --git a/pkg/tcpip/stack/forwarder_test.go b/pkg/tcpip/stack/forwarder_test.go index c45c43d21..e9c652042 100644 --- a/pkg/tcpip/stack/forwarder_test.go +++ b/pkg/tcpip/stack/forwarder_test.go @@ -101,7 +101,7 @@ func (f *fwdTestNetworkEndpoint) WritePacket(r *Route, gso *GSO, params NetworkH } // WritePackets implements LinkEndpoint.WritePackets. -func (f *fwdTestNetworkEndpoint) WritePackets(r *Route, gso *GSO, pkts []PacketBuffer, params NetworkHeaderParams) (int, *tcpip.Error) { +func (f *fwdTestNetworkEndpoint) WritePackets(r *Route, gso *GSO, pkts PacketBufferList, params NetworkHeaderParams) (int, *tcpip.Error) { panic("not implemented") } @@ -260,10 +260,10 @@ func (e fwdTestLinkEndpoint) WritePacket(r *Route, gso *GSO, protocol tcpip.Netw } // WritePackets stores outbound packets into the channel. -func (e *fwdTestLinkEndpoint) WritePackets(r *Route, gso *GSO, pkts []PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (e *fwdTestLinkEndpoint) WritePackets(r *Route, gso *GSO, pkts PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { n := 0 - for _, pkt := range pkts { - e.WritePacket(r, gso, protocol, pkt) + for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() { + e.WritePacket(r, gso, protocol, *pkt) n++ } diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go index 37907ae24..6c0a4b24d 100644 --- a/pkg/tcpip/stack/iptables.go +++ b/pkg/tcpip/stack/iptables.go @@ -209,6 +209,23 @@ func (it *IPTables) Check(hook Hook, pkt PacketBuffer) bool { return true } +// CheckPackets runs pkts through the rules for hook and returns a map of packets that +// should not go forward. +// +// NOTE: unlike the Check API the returned map contains packets that should be +// dropped. +func (it *IPTables) CheckPackets(hook Hook, pkts PacketBufferList) (drop map[*PacketBuffer]struct{}) { + for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() { + if ok := it.Check(hook, *pkt); !ok { + if drop == nil { + drop = make(map[*PacketBuffer]struct{}) + } + drop[pkt] = struct{}{} + } + } + return drop +} + // Precondition: pkt.NetworkHeader is set. func (it *IPTables) checkChain(hook Hook, pkt PacketBuffer, table Table, ruleIdx int) chainVerdict { // Start from ruleIdx and walk the list of rules until a rule gives us diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go index 598468bdd..27dc8baf9 100644 --- a/pkg/tcpip/stack/ndp_test.go +++ b/pkg/tcpip/stack/ndp_test.go @@ -468,7 +468,7 @@ func TestDADResolve(t *testing.T) { // As per RFC 4861 section 4.3, a possible option is the Source Link // Layer option, but this option MUST NOT be included when the source // address of the packet is the unspecified address. - checker.IPv6(t, p.Pkt.Header.View().ToVectorisedView().First(), + checker.IPv6(t, p.Pkt.Header.View(), checker.SrcAddr(header.IPv6Any), checker.DstAddr(snmc), checker.TTL(header.NDPHopLimit), diff --git a/pkg/tcpip/stack/packet_buffer.go b/pkg/tcpip/stack/packet_buffer.go index 9367de180..dc125f25e 100644 --- a/pkg/tcpip/stack/packet_buffer.go +++ b/pkg/tcpip/stack/packet_buffer.go @@ -23,9 +23,11 @@ import ( // As a PacketBuffer traverses up the stack, it may be necessary to pass it to // multiple endpoints. Clone() should be called in such cases so that // modifications to the Data field do not affect other copies. -// -// +stateify savable type PacketBuffer struct { + // PacketBufferEntry is used to build an intrusive list of + // PacketBuffers. + PacketBufferEntry + // Data holds the payload of the packet. For inbound packets, it also // holds the headers, which are consumed as the packet moves up the // stack. Headers are guaranteed not to be split across views. @@ -34,14 +36,6 @@ type PacketBuffer struct { // or otherwise modified. Data buffer.VectorisedView - // DataOffset is used for GSO output. It is the offset into the Data - // field where the payload of this packet starts. - DataOffset int - - // DataSize is used for GSO output. It is the size of this packet's - // payload. - DataSize int - // Header holds the headers of outbound packets. As a packet is passed // down the stack, each layer adds to Header. Header buffer.Prependable diff --git a/pkg/tcpip/stack/packet_buffer_state.go b/pkg/tcpip/stack/packet_buffer_state.go deleted file mode 100644 index 0c6b7924c..000000000 --- a/pkg/tcpip/stack/packet_buffer_state.go +++ /dev/null @@ -1,27 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package stack - -import "gvisor.dev/gvisor/pkg/tcpip/buffer" - -// beforeSave is invoked by stateify. -func (pk *PacketBuffer) beforeSave() { - // Non-Data fields may be slices of the Data field. This causes - // problems for SR, so during save we make each header independent. - pk.Header = pk.Header.DeepCopy() - pk.LinkHeader = append(buffer.View(nil), pk.LinkHeader...) - pk.NetworkHeader = append(buffer.View(nil), pk.NetworkHeader...) - pk.TransportHeader = append(buffer.View(nil), pk.TransportHeader...) -} diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go index ac043b722..23ca9ee03 100644 --- a/pkg/tcpip/stack/registration.go +++ b/pkg/tcpip/stack/registration.go @@ -246,7 +246,7 @@ type NetworkEndpoint interface { // WritePackets writes packets to the given destination address and // protocol. pkts must not be zero length. - WritePackets(r *Route, gso *GSO, pkts []PacketBuffer, params NetworkHeaderParams) (int, *tcpip.Error) + WritePackets(r *Route, gso *GSO, pkts PacketBufferList, params NetworkHeaderParams) (int, *tcpip.Error) // WriteHeaderIncludedPacket writes a packet that includes a network // header to the given destination address. @@ -393,7 +393,7 @@ type LinkEndpoint interface { // Right now, WritePackets is used only when the software segmentation // offload is enabled. If it will be used for something else, it may // require to change syscall filters. - WritePackets(r *Route, gso *GSO, pkts []PacketBuffer, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) + WritePackets(r *Route, gso *GSO, pkts PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) // WriteRawPacket writes a packet directly to the link. The packet // should already have an ethernet header. diff --git a/pkg/tcpip/stack/route.go b/pkg/tcpip/stack/route.go index 9fbe8a411..a0e5e0300 100644 --- a/pkg/tcpip/stack/route.go +++ b/pkg/tcpip/stack/route.go @@ -168,23 +168,26 @@ func (r *Route) WritePacket(gso *GSO, params NetworkHeaderParams, pkt PacketBuff return err } -// WritePackets writes the set of packets through the given route. -func (r *Route) WritePackets(gso *GSO, pkts []PacketBuffer, params NetworkHeaderParams) (int, *tcpip.Error) { +// WritePackets writes a list of n packets through the given route and returns +// the number of packets written. +func (r *Route) WritePackets(gso *GSO, pkts PacketBufferList, params NetworkHeaderParams) (int, *tcpip.Error) { if !r.ref.isValidForOutgoing() { return 0, tcpip.ErrInvalidEndpointState } n, err := r.ref.ep.WritePackets(r, gso, pkts, params) if err != nil { - r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(len(pkts) - n)) + r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(pkts.Len() - n)) } r.ref.nic.stats.Tx.Packets.IncrementBy(uint64(n)) - payloadSize := 0 - for i := 0; i < n; i++ { - r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(pkts[i].Header.UsedLength())) - payloadSize += pkts[i].DataSize + + writtenBytes := 0 + for i, pb := 0, pkts.Front(); i < n && pb != nil; i, pb = i+1, pb.Next() { + writtenBytes += pb.Header.UsedLength() + writtenBytes += pb.Data.Size() } - r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(payloadSize)) + + r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(writtenBytes)) return n, err } diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index b8543b71e..3f8a2a095 100644 --- a/pkg/tcpip/stack/stack_test.go +++ b/pkg/tcpip/stack/stack_test.go @@ -153,7 +153,7 @@ func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, params } // WritePackets implements stack.LinkEndpoint.WritePackets. -func (f *fakeNetworkEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []stack.PacketBuffer, params stack.NetworkHeaderParams) (int, *tcpip.Error) { +func (f *fakeNetworkEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.PacketBufferList, params stack.NetworkHeaderParams) (int, *tcpip.Error) { panic("not implemented") } diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 3239a5911..2ca3fb809 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -756,8 +756,7 @@ func (e *endpoint) sendTCP(r *stack.Route, tf tcpFields, data buffer.VectorisedV func buildTCPHdr(r *stack.Route, tf tcpFields, pkt *stack.PacketBuffer, gso *stack.GSO) { optLen := len(tf.opts) hdr := &pkt.Header - packetSize := pkt.DataSize - off := pkt.DataOffset + packetSize := pkt.Data.Size() // Initialize the header. tcp := header.TCP(hdr.Prepend(header.TCPMinimumSize + optLen)) pkt.TransportHeader = buffer.View(tcp) @@ -782,12 +781,18 @@ func buildTCPHdr(r *stack.Route, tf tcpFields, pkt *stack.PacketBuffer, gso *sta // header and data and get the right sum of the TCP packet. tcp.SetChecksum(xsum) } else if r.Capabilities()&stack.CapabilityTXChecksumOffload == 0 { - xsum = header.ChecksumVVWithOffset(pkt.Data, xsum, off, packetSize) + xsum = header.ChecksumVV(pkt.Data, xsum) tcp.SetChecksum(^tcp.CalculateChecksum(xsum)) } } func sendTCPBatch(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso *stack.GSO, owner tcpip.PacketOwner) *tcpip.Error { + // We need to shallow clone the VectorisedView here as ReadToView will + // split the VectorisedView and Trim underlying views as it splits. Not + // doing the clone here will cause the underlying views of data itself + // to be altered. + data = data.Clone(nil) + optLen := len(tf.opts) if tf.rcvWnd > 0xffff { tf.rcvWnd = 0xffff @@ -796,31 +801,25 @@ func sendTCPBatch(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso mss := int(gso.MSS) n := (data.Size() + mss - 1) / mss - // Allocate one big slice for all the headers. - hdrSize := header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen - buf := make([]byte, n*hdrSize) - pkts := make([]stack.PacketBuffer, n) - for i := range pkts { - pkts[i].Header = buffer.NewEmptyPrependableFromView(buf[i*hdrSize:][:hdrSize]) - } - size := data.Size() - off := 0 + hdrSize := header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen + var pkts stack.PacketBufferList for i := 0; i < n; i++ { packetSize := mss if packetSize > size { packetSize = size } size -= packetSize - pkts[i].DataOffset = off - pkts[i].DataSize = packetSize - pkts[i].Data = data - pkts[i].Hash = tf.txHash - pkts[i].Owner = owner - buildTCPHdr(r, tf, &pkts[i], gso) - off += packetSize + var pkt stack.PacketBuffer + pkt.Header = buffer.NewPrependable(hdrSize) + pkt.Hash = tf.txHash + pkt.Owner = owner + data.ReadToVV(&pkt.Data, packetSize) + buildTCPHdr(r, tf, &pkt, gso) tf.seq = tf.seq.Add(seqnum.Size(packetSize)) + pkts.PushBack(&pkt) } + if tf.ttl == 0 { tf.ttl = r.DefaultTTL() } @@ -845,12 +844,10 @@ func sendTCP(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso *stac } pkt := stack.PacketBuffer{ - Header: buffer.NewPrependable(header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen), - DataOffset: 0, - DataSize: data.Size(), - Data: data, - Hash: tf.txHash, - Owner: owner, + Header: buffer.NewPrependable(header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen), + Data: data, + Hash: tf.txHash, + Owner: owner, } buildTCPHdr(r, tf, &pkt, gso) diff --git a/pkg/tcpip/transport/tcp/segment.go b/pkg/tcpip/transport/tcp/segment.go index e6fe7985d..40461fd31 100644 --- a/pkg/tcpip/transport/tcp/segment.go +++ b/pkg/tcpip/transport/tcp/segment.go @@ -77,9 +77,11 @@ func newSegmentFromView(r *stack.Route, id stack.TransportEndpointID, v buffer.V id: id, route: r.Clone(), } - s.views[0] = v - s.data = buffer.NewVectorisedView(len(v), s.views[:1]) s.rcvdTime = time.Now() + if len(v) != 0 { + s.views[0] = v + s.data = buffer.NewVectorisedView(len(v), s.views[:1]) + } return s } -- cgit v1.2.3 From 24bee1c1813a691072cff5bad7a528690a99eb5e Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Sat, 4 Apr 2020 21:01:42 -0700 Subject: Record VFS2 sockets in global socket map. Updates #1476, #1478, #1484, #1485. PiperOrigin-RevId: 304845354 --- pkg/sentry/fsimpl/proc/BUILD | 1 - pkg/sentry/fsimpl/proc/task_net.go | 88 ++++++++++++++++++++++--------------- pkg/sentry/kernel/kernel.go | 30 +++++++++++-- pkg/sentry/socket/socket.go | 6 ++- pkg/sentry/socket/unix/unix_vfs2.go | 2 +- pkg/sentry/vfs/file_description.go | 6 +++ 6 files changed, 91 insertions(+), 42 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/proc/BUILD b/pkg/sentry/fsimpl/proc/BUILD index 8156984eb..17c1342b5 100644 --- a/pkg/sentry/fsimpl/proc/BUILD +++ b/pkg/sentry/fsimpl/proc/BUILD @@ -22,7 +22,6 @@ go_library( "//pkg/log", "//pkg/refs", "//pkg/safemem", - "//pkg/sentry/fs", "//pkg/sentry/fsbridge", "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/inet", diff --git a/pkg/sentry/fsimpl/proc/task_net.go b/pkg/sentry/fsimpl/proc/task_net.go index 373a7b17d..6b2a77328 100644 --- a/pkg/sentry/fsimpl/proc/task_net.go +++ b/pkg/sentry/fsimpl/proc/task_net.go @@ -24,7 +24,6 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -32,6 +31,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/unix" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" + "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/usermem" @@ -206,22 +206,21 @@ var _ dynamicInode = (*netUnixData)(nil) func (n *netUnixData) Generate(ctx context.Context, buf *bytes.Buffer) error { buf.WriteString("Num RefCount Protocol Flags Type St Inode Path\n") for _, se := range n.kernel.ListSockets() { - s := se.Sock.Get() - if s == nil { - log.Debugf("Couldn't resolve weakref %v in socket table, racing with destruction?", se.Sock) + s := se.SockVFS2 + if !s.TryIncRef() { + log.Debugf("Couldn't get reference on %v in socket table, racing with destruction?", s) continue } - sfile := s.(*fs.File) - if family, _, _ := sfile.FileOperations.(socket.Socket).Type(); family != linux.AF_UNIX { + if family, _, _ := s.Impl().(socket.SocketVFS2).Type(); family != linux.AF_UNIX { s.DecRef() // Not a unix socket. continue } - sops := sfile.FileOperations.(*unix.SocketOperations) + sops := s.Impl().(*unix.SocketVFS2) addr, err := sops.Endpoint().GetLocalAddress() if err != nil { - log.Warningf("Failed to retrieve socket name from %+v: %v", sfile, err) + log.Warningf("Failed to retrieve socket name from %+v: %v", s, err) addr.Addr = "" } @@ -234,6 +233,15 @@ func (n *netUnixData) Generate(ctx context.Context, buf *bytes.Buffer) error { } } + // Get inode number. + var ino uint64 + stat, statErr := s.Stat(ctx, vfs.StatOptions{Mask: linux.STATX_INO}) + if statErr != nil || stat.Mask&linux.STATX_INO == 0 { + log.Warningf("Failed to retrieve ino for socket file: %v", statErr) + } else { + ino = stat.Ino + } + // In the socket entry below, the value for the 'Num' field requires // some consideration. Linux prints the address to the struct // unix_sock representing a socket in the kernel, but may redact the @@ -252,14 +260,14 @@ func (n *netUnixData) Generate(ctx context.Context, buf *bytes.Buffer) error { // the definition of this struct changes over time. // // For now, we always redact this pointer. - fmt.Fprintf(buf, "%#016p: %08X %08X %08X %04X %02X %5d", + fmt.Fprintf(buf, "%#016p: %08X %08X %08X %04X %02X %8d", (*unix.SocketOperations)(nil), // Num, pointer to kernel socket struct. - sfile.ReadRefs()-1, // RefCount, don't count our own ref. + s.Refs()-1, // RefCount, don't count our own ref. 0, // Protocol, always 0 for UDS. sockFlags, // Flags. sops.Endpoint().Type(), // Type. sops.State(), // State. - sfile.InodeID(), // Inode. + ino, // Inode. ) // Path @@ -341,15 +349,14 @@ func commonGenerateTCP(ctx context.Context, buf *bytes.Buffer, k *kernel.Kernel, t := kernel.TaskFromContext(ctx) for _, se := range k.ListSockets() { - s := se.Sock.Get() - if s == nil { - log.Debugf("Couldn't resolve weakref with ID %v in socket table, racing with destruction?", se.ID) + s := se.SockVFS2 + if !s.TryIncRef() { + log.Debugf("Couldn't get reference on %v in socket table, racing with destruction?", s) continue } - sfile := s.(*fs.File) - sops, ok := sfile.FileOperations.(socket.Socket) + sops, ok := s.Impl().(socket.SocketVFS2) if !ok { - panic(fmt.Sprintf("Found non-socket file in socket table: %+v", sfile)) + panic(fmt.Sprintf("Found non-socket file in socket table: %+v", s)) } if fa, stype, _ := sops.Type(); !(family == fa && stype == linux.SOCK_STREAM) { s.DecRef() @@ -398,14 +405,15 @@ func commonGenerateTCP(ctx context.Context, buf *bytes.Buffer, k *kernel.Kernel, // Unimplemented. fmt.Fprintf(buf, "%08X ", 0) + stat, statErr := s.Stat(ctx, vfs.StatOptions{Mask: linux.STATX_UID | linux.STATX_INO}) + // Field: uid. - uattr, err := sfile.Dirent.Inode.UnstableAttr(ctx) - if err != nil { - log.Warningf("Failed to retrieve unstable attr for socket file: %v", err) + if statErr != nil || stat.Mask&linux.STATX_UID == 0 { + log.Warningf("Failed to retrieve uid for socket file: %v", statErr) fmt.Fprintf(buf, "%5d ", 0) } else { creds := auth.CredentialsFromContext(ctx) - fmt.Fprintf(buf, "%5d ", uint32(uattr.Owner.UID.In(creds.UserNamespace).OrOverflow())) + fmt.Fprintf(buf, "%5d ", uint32(auth.KUID(stat.UID).In(creds.UserNamespace).OrOverflow())) } // Field: timeout; number of unanswered 0-window probes. @@ -413,11 +421,16 @@ func commonGenerateTCP(ctx context.Context, buf *bytes.Buffer, k *kernel.Kernel, fmt.Fprintf(buf, "%8d ", 0) // Field: inode. - fmt.Fprintf(buf, "%8d ", sfile.InodeID()) + if statErr != nil || stat.Mask&linux.STATX_INO == 0 { + log.Warningf("Failed to retrieve inode for socket file: %v", statErr) + fmt.Fprintf(buf, "%8d ", 0) + } else { + fmt.Fprintf(buf, "%8d ", stat.Ino) + } // Field: refcount. Don't count the ref we obtain while deferencing // the weakref to this socket. - fmt.Fprintf(buf, "%d ", sfile.ReadRefs()-1) + fmt.Fprintf(buf, "%d ", s.Refs()-1) // Field: Socket struct address. Redacted due to the same reason as // the 'Num' field in /proc/net/unix, see netUnix.ReadSeqFileData. @@ -499,15 +512,14 @@ func (d *netUDPData) Generate(ctx context.Context, buf *bytes.Buffer) error { t := kernel.TaskFromContext(ctx) for _, se := range d.kernel.ListSockets() { - s := se.Sock.Get() - if s == nil { - log.Debugf("Couldn't resolve weakref with ID %v in socket table, racing with destruction?", se.ID) + s := se.SockVFS2 + if !s.TryIncRef() { + log.Debugf("Couldn't get reference on %v in socket table, racing with destruction?", s) continue } - sfile := s.(*fs.File) - sops, ok := sfile.FileOperations.(socket.Socket) + sops, ok := s.Impl().(socket.SocketVFS2) if !ok { - panic(fmt.Sprintf("Found non-socket file in socket table: %+v", sfile)) + panic(fmt.Sprintf("Found non-socket file in socket table: %+v", s)) } if family, stype, _ := sops.Type(); family != linux.AF_INET || stype != linux.SOCK_DGRAM { s.DecRef() @@ -551,25 +563,31 @@ func (d *netUDPData) Generate(ctx context.Context, buf *bytes.Buffer) error { // Field: retrnsmt. Always 0 for UDP. fmt.Fprintf(buf, "%08X ", 0) + stat, statErr := s.Stat(ctx, vfs.StatOptions{Mask: linux.STATX_UID | linux.STATX_INO}) + // Field: uid. - uattr, err := sfile.Dirent.Inode.UnstableAttr(ctx) - if err != nil { - log.Warningf("Failed to retrieve unstable attr for socket file: %v", err) + if statErr != nil || stat.Mask&linux.STATX_UID == 0 { + log.Warningf("Failed to retrieve uid for socket file: %v", statErr) fmt.Fprintf(buf, "%5d ", 0) } else { creds := auth.CredentialsFromContext(ctx) - fmt.Fprintf(buf, "%5d ", uint32(uattr.Owner.UID.In(creds.UserNamespace).OrOverflow())) + fmt.Fprintf(buf, "%5d ", uint32(auth.KUID(stat.UID).In(creds.UserNamespace).OrOverflow())) } // Field: timeout. Always 0 for UDP. fmt.Fprintf(buf, "%8d ", 0) // Field: inode. - fmt.Fprintf(buf, "%8d ", sfile.InodeID()) + if statErr != nil || stat.Mask&linux.STATX_INO == 0 { + log.Warningf("Failed to retrieve inode for socket file: %v", statErr) + fmt.Fprintf(buf, "%8d ", 0) + } else { + fmt.Fprintf(buf, "%8d ", stat.Ino) + } // Field: ref; reference count on the socket inode. Don't count the ref // we obtain while deferencing the weakref to this socket. - fmt.Fprintf(buf, "%d ", sfile.ReadRefs()-1) + fmt.Fprintf(buf, "%d ", s.Refs()-1) // Field: Socket struct address. Redacted due to the same reason as // the 'Num' field in /proc/net/unix, see netUnix.ReadSeqFileData. diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 2e6f42b92..ba8935a82 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -1445,9 +1445,10 @@ func (k *Kernel) SupervisorContext() context.Context { // +stateify savable type SocketEntry struct { socketEntry - k *Kernel - Sock *refs.WeakRef - ID uint64 // Socket table entry number. + k *Kernel + Sock *refs.WeakRef + SockVFS2 *vfs.FileDescription + ID uint64 // Socket table entry number. } // WeakRefGone implements refs.WeakRefUser.WeakRefGone. @@ -1470,7 +1471,30 @@ func (k *Kernel) RecordSocket(sock *fs.File) { k.extMu.Unlock() } +// RecordSocketVFS2 adds a VFS2 socket to the system-wide socket table for +// tracking. +// +// Precondition: Caller must hold a reference to sock. +// +// Note that the socket table will not hold a reference on the +// vfs.FileDescription, because we do not support weak refs on VFS2 files. +func (k *Kernel) RecordSocketVFS2(sock *vfs.FileDescription) { + k.extMu.Lock() + id := k.nextSocketEntry + k.nextSocketEntry++ + s := &SocketEntry{ + k: k, + ID: id, + SockVFS2: sock, + } + k.sockets.PushBack(s) + k.extMu.Unlock() +} + // ListSockets returns a snapshot of all sockets. +// +// Callers of ListSockets() in VFS2 should use SocketEntry.SockVFS2.TryIncRef() +// to get a reference on a socket in the table. func (k *Kernel) ListSockets() []*SocketEntry { k.extMu.Lock() var socks []*SocketEntry diff --git a/pkg/sentry/socket/socket.go b/pkg/sentry/socket/socket.go index b5ba4a56b..6580bd6e9 100644 --- a/pkg/sentry/socket/socket.go +++ b/pkg/sentry/socket/socket.go @@ -269,7 +269,7 @@ func NewVFS2(t *kernel.Task, family int, stype linux.SockType, protocol int) (*v return nil, err } if s != nil { - // TODO: Add vfs2 sockets to global socket table. + t.Kernel().RecordSocketVFS2(s) return s, nil } } @@ -291,7 +291,9 @@ func PairVFS2(t *kernel.Task, family int, stype linux.SockType, protocol int) (* return nil, nil, err } if s1 != nil && s2 != nil { - // TODO: Add vfs2 sockets to global socket table. + k := t.Kernel() + k.RecordSocketVFS2(s1) + k.RecordSocketVFS2(s2) return s1, s2, nil } } diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go index ca1388e2c..3e54d49c4 100644 --- a/pkg/sentry/socket/unix/unix_vfs2.go +++ b/pkg/sentry/socket/unix/unix_vfs2.go @@ -141,7 +141,7 @@ func (s *SocketVFS2) Accept(t *kernel.Task, peerRequested bool, flags int, block return 0, nil, 0, syserr.FromError(e) } - // TODO: add vfs2 sockets to global table. + t.Kernel().RecordSocketVFS2(ns) return fd, addr, addrLen, nil } diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 5df4bbf45..28e93a441 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -182,6 +182,12 @@ func (fd *FileDescription) DecRef() { } } +// Refs returns the current number of references. The returned count +// is inherently racy and is unsafe to use without external synchronization. +func (fd *FileDescription) Refs() int64 { + return atomic.LoadInt64(&fd.refs) +} + // Mount returns the mount on which fd was opened. It does not take a reference // on the returned Mount. func (fd *FileDescription) Mount() *Mount { -- cgit v1.2.3 From 00d9776a4bb1cc1d7125af7d3e54a939a4f3847a Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Mon, 6 Apr 2020 07:30:20 -0700 Subject: Add socket files to tmpfs VFS2. Updates #1476. PiperOrigin-RevId: 305024274 --- pkg/sentry/fsimpl/tmpfs/BUILD | 1 + pkg/sentry/fsimpl/tmpfs/filesystem.go | 20 +++++++++++++++----- pkg/sentry/fsimpl/tmpfs/socket_file.go | 34 ++++++++++++++++++++++++++++++++++ pkg/sentry/fsimpl/tmpfs/tmpfs.go | 4 +++- 4 files changed, 53 insertions(+), 6 deletions(-) create mode 100644 pkg/sentry/fsimpl/tmpfs/socket_file.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD index 6ea35affb..f2ac23c88 100644 --- a/pkg/sentry/fsimpl/tmpfs/BUILD +++ b/pkg/sentry/fsimpl/tmpfs/BUILD @@ -24,6 +24,7 @@ go_library( "filesystem.go", "named_pipe.go", "regular_file.go", + "socket_file.go", "symlink.go", "tmpfs.go", ], diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 1978af69c..5339d7072 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -261,8 +261,7 @@ func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v case linux.S_IFCHR: childInode = fs.newDeviceFile(rp.Credentials(), opts.Mode, vfs.CharDevice, opts.DevMajor, opts.DevMinor) case linux.S_IFSOCK: - // Not yet supported. - return syserror.EPERM + childInode = fs.newSocketFile(rp.Credentials(), opts.Mode, opts.Endpoint) default: return syserror.EINVAL } @@ -396,6 +395,8 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.Open return newNamedPipeFD(ctx, impl, rp, &d.vfsd, opts.Flags) case *deviceFile: return rp.VirtualFilesystem().OpenDeviceSpecialFile(ctx, rp.Mount(), &d.vfsd, impl.kind, impl.major, impl.minor, opts) + case *socketFile: + return nil, syserror.ENXIO default: panic(fmt.Sprintf("unknown inode type: %T", d.inode.impl)) } @@ -679,10 +680,19 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error } // BoundEndpointAt implements FilesystemImpl.BoundEndpointAt. -// -// TODO(gvisor.dev/issue/1476): Implement BoundEndpointAt. func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath) (transport.BoundEndpoint, error) { - return nil, syserror.ECONNREFUSED + fs.mu.RLock() + defer fs.mu.RUnlock() + d, err := resolveLocked(rp) + if err != nil { + return nil, err + } + switch impl := d.inode.impl.(type) { + case *socketFile: + return impl.ep, nil + default: + return nil, syserror.ECONNREFUSED + } } // ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. diff --git a/pkg/sentry/fsimpl/tmpfs/socket_file.go b/pkg/sentry/fsimpl/tmpfs/socket_file.go new file mode 100644 index 000000000..25c2321af --- /dev/null +++ b/pkg/sentry/fsimpl/tmpfs/socket_file.go @@ -0,0 +1,34 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tmpfs + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" +) + +// socketFile is a socket (=S_IFSOCK) tmpfs file. +type socketFile struct { + inode inode + ep transport.BoundEndpoint +} + +func (fs *filesystem) newSocketFile(creds *auth.Credentials, mode linux.FileMode, ep transport.BoundEndpoint) *inode { + file := &socketFile{ep: ep} + file.inode.init(file, fs, creds, mode) + file.inode.nlink = 1 // from parent directory + return &file.inode +} diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index ad47288f8..654e788e3 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -331,7 +331,7 @@ func (i *inode) statTo(stat *linux.Statx) { case *deviceFile: stat.RdevMajor = impl.major stat.RdevMinor = impl.minor - case *directory, *namedPipe: + case *socketFile, *directory, *namedPipe: // Nothing to do. default: panic(fmt.Sprintf("unknown inode type: %T", i.impl)) @@ -479,6 +479,8 @@ func (i *inode) direntType() uint8 { return linux.DT_DIR case *symlink: return linux.DT_LNK + case *socketFile: + return linux.DT_SOCK case *deviceFile: switch impl.kind { case vfs.BlockDevice: -- cgit v1.2.3 From f332a864e8cc7799332838deffab37244ff8ffc7 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Mon, 6 Apr 2020 10:51:54 -0700 Subject: Port timerfd to VFS2. PiperOrigin-RevId: 305067208 --- pkg/sentry/kernel/kernel.go | 28 ++-- pkg/sentry/syscalls/linux/vfs2/BUILD | 1 + .../syscalls/linux/vfs2/linux64_override_amd64.go | 6 +- pkg/sentry/syscalls/linux/vfs2/sys_timerfd.go | 123 ++++++++++++++++++ pkg/sentry/vfs/BUILD | 2 + pkg/sentry/vfs/file_description.go | 7 + pkg/sentry/vfs/timerfd.go | 142 +++++++++++++++++++++ 7 files changed, 295 insertions(+), 14 deletions(-) create mode 100644 pkg/sentry/syscalls/linux/vfs2/sys_timerfd.go create mode 100644 pkg/sentry/vfs/timerfd.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index ba8935a82..de8a95854 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -1044,14 +1044,17 @@ func (k *Kernel) pauseTimeLocked() { // This means we'll iterate FDTables shared by multiple tasks repeatedly, // but ktime.Timer.Pause is idempotent so this is harmless. if t.fdTable != nil { - // TODO(gvisor.dev/issue/1663): Add save support for VFS2. - if !VFS2Enabled { - t.fdTable.forEach(func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) { + t.fdTable.forEach(func(_ int32, file *fs.File, fd *vfs.FileDescription, _ FDFlags) { + if VFS2Enabled { + if tfd, ok := fd.Impl().(*vfs.TimerFileDescription); ok { + tfd.PauseTimer() + } + } else { if tfd, ok := file.FileOperations.(*timerfd.TimerOperations); ok { tfd.PauseTimer() } - }) - } + } + }) } } k.timekeeper.PauseUpdates() @@ -1076,15 +1079,18 @@ func (k *Kernel) resumeTimeLocked() { it.ResumeTimer() } } - // TODO(gvisor.dev/issue/1663): Add save support for VFS2. - if !VFS2Enabled { - if t.fdTable != nil { - t.fdTable.forEach(func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) { + if t.fdTable != nil { + t.fdTable.forEach(func(_ int32, file *fs.File, fd *vfs.FileDescription, _ FDFlags) { + if VFS2Enabled { + if tfd, ok := fd.Impl().(*vfs.TimerFileDescription); ok { + tfd.ResumeTimer() + } + } else { if tfd, ok := file.FileOperations.(*timerfd.TimerOperations); ok { tfd.ResumeTimer() } - }) - } + } + }) } } } diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD index 2eb210014..0004e60d9 100644 --- a/pkg/sentry/syscalls/linux/vfs2/BUILD +++ b/pkg/sentry/syscalls/linux/vfs2/BUILD @@ -25,6 +25,7 @@ go_library( "stat_amd64.go", "stat_arm64.go", "sync.go", + "sys_timerfd.go", "xattr.go", ], marshal = True, diff --git a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go index 7d220bc20..63febc2f7 100644 --- a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go +++ b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go @@ -139,11 +139,11 @@ func Override(table map[uintptr]kernel.Syscall) { table[280] = syscalls.Supported("utimensat", Utimensat) table[281] = syscalls.Supported("epoll_pwait", EpollPwait) delete(table, 282) // signalfd - delete(table, 283) // timerfd_create + table[283] = syscalls.Supported("timerfd_create", TimerfdCreate) delete(table, 284) // eventfd delete(table, 285) // fallocate - delete(table, 286) // timerfd_settime - delete(table, 287) // timerfd_gettime + table[286] = syscalls.Supported("timerfd_settime", TimerfdSettime) + table[287] = syscalls.Supported("timerfd_gettime", TimerfdGettime) delete(table, 288) // accept4 delete(table, 289) // signalfd4 delete(table, 290) // eventfd2 diff --git a/pkg/sentry/syscalls/linux/vfs2/sys_timerfd.go b/pkg/sentry/syscalls/linux/vfs2/sys_timerfd.go new file mode 100644 index 000000000..7938a5249 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/sys_timerfd.go @@ -0,0 +1,123 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// TimerfdCreate implements Linux syscall timerfd_create(2). +func TimerfdCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + clockID := args[0].Int() + flags := args[1].Int() + + if flags&^(linux.TFD_CLOEXEC|linux.TFD_NONBLOCK) != 0 { + return 0, nil, syserror.EINVAL + } + + var fileFlags uint32 + if flags&linux.TFD_NONBLOCK != 0 { + fileFlags = linux.O_NONBLOCK + } + + var clock ktime.Clock + switch clockID { + case linux.CLOCK_REALTIME: + clock = t.Kernel().RealtimeClock() + case linux.CLOCK_MONOTONIC, linux.CLOCK_BOOTTIME: + clock = t.Kernel().MonotonicClock() + default: + return 0, nil, syserror.EINVAL + } + file, err := t.Kernel().VFS().NewTimerFD(clock, fileFlags) + if err != nil { + return 0, nil, err + } + defer file.DecRef() + fd, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{ + CloseOnExec: flags&linux.TFD_CLOEXEC != 0, + }) + if err != nil { + return 0, nil, err + } + return uintptr(fd), nil, nil +} + +// TimerfdSettime implements Linux syscall timerfd_settime(2). +func TimerfdSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + flags := args[1].Int() + newValAddr := args[2].Pointer() + oldValAddr := args[3].Pointer() + + if flags&^(linux.TFD_TIMER_ABSTIME) != 0 { + return 0, nil, syserror.EINVAL + } + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + tfd, ok := file.Impl().(*vfs.TimerFileDescription) + if !ok { + return 0, nil, syserror.EINVAL + } + + var newVal linux.Itimerspec + if _, err := t.CopyIn(newValAddr, &newVal); err != nil { + return 0, nil, err + } + newS, err := ktime.SettingFromItimerspec(newVal, flags&linux.TFD_TIMER_ABSTIME != 0, tfd.Clock()) + if err != nil { + return 0, nil, err + } + tm, oldS := tfd.SetTime(newS) + if oldValAddr != 0 { + oldVal := ktime.ItimerspecFromSetting(tm, oldS) + if _, err := t.CopyOut(oldValAddr, &oldVal); err != nil { + return 0, nil, err + } + } + return 0, nil, nil +} + +// TimerfdGettime implements Linux syscall timerfd_gettime(2). +func TimerfdGettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + curValAddr := args[1].Pointer() + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + tfd, ok := file.Impl().(*vfs.TimerFileDescription) + if !ok { + return 0, nil, syserror.EINVAL + } + + tm, s := tfd.GetTime() + curVal := ktime.ItimerspecFromSetting(tm, s) + _, err := t.CopyOut(curValAddr, &curVal) + return 0, nil, err +} diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index bf4d27c7d..9aeb83fb0 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -36,6 +36,7 @@ go_library( "pathname.go", "permissions.go", "resolving_path.go", + "timerfd.go", "vfs.go", ], visibility = ["//pkg/sentry:internal"], @@ -51,6 +52,7 @@ go_library( "//pkg/sentry/fs", "//pkg/sentry/fs/lock", "//pkg/sentry/kernel/auth", + "//pkg/sentry/kernel/time", "//pkg/sentry/limits", "//pkg/sentry/memmap", "//pkg/sentry/socket/unix/transport", diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 28e93a441..20c545fca 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -91,6 +91,10 @@ type FileDescriptionOptions struct { // ESPIPE. DenyPWrite bool + // if InvalidWrite is true, calls to FileDescription.Write() return + // EINVAL. + InvalidWrite bool + // If UseDentryMetadata is true, calls to FileDescription methods that // interact with file and filesystem metadata (Stat, SetStat, StatFS, // Listxattr, Getxattr, Setxattr, Removexattr) are implemented by calling @@ -562,6 +566,9 @@ func (fd *FileDescription) PWrite(ctx context.Context, src usermem.IOSequence, o // Write is similar to PWrite, but does not specify an offset. func (fd *FileDescription) Write(ctx context.Context, src usermem.IOSequence, opts WriteOptions) (int64, error) { + if fd.opts.InvalidWrite { + return 0, syserror.EINVAL + } if !fd.writable { return 0, syserror.EBADF } diff --git a/pkg/sentry/vfs/timerfd.go b/pkg/sentry/vfs/timerfd.go new file mode 100644 index 000000000..42b880656 --- /dev/null +++ b/pkg/sentry/vfs/timerfd.go @@ -0,0 +1,142 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs + +import ( + "sync/atomic" + + "gvisor.dev/gvisor/pkg/context" + ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +// TimerFileDescription implements FileDescriptionImpl for timer fds. It also +// implements ktime.TimerListener. +type TimerFileDescription struct { + vfsfd FileDescription + FileDescriptionDefaultImpl + DentryMetadataFileDescriptionImpl + + events waiter.Queue + timer *ktime.Timer + + // val is the number of timer expirations since the last successful + // call to PRead, or SetTime. val must be accessed using atomic memory + // operations. + val uint64 +} + +var _ FileDescriptionImpl = (*TimerFileDescription)(nil) +var _ ktime.TimerListener = (*TimerFileDescription)(nil) + +// NewTimerFD returns a new timer fd. +func (vfs *VirtualFilesystem) NewTimerFD(clock ktime.Clock, flags uint32) (*FileDescription, error) { + vd := vfs.NewAnonVirtualDentry("[timerfd]") + defer vd.DecRef() + tfd := &TimerFileDescription{} + tfd.timer = ktime.NewTimer(clock, tfd) + if err := tfd.vfsfd.Init(tfd, flags, vd.Mount(), vd.Dentry(), &FileDescriptionOptions{ + UseDentryMetadata: true, + DenyPRead: true, + DenyPWrite: true, + InvalidWrite: true, + }); err != nil { + return nil, err + } + return &tfd.vfsfd, nil +} + +// Read implements FileDescriptionImpl.Read. +func (tfd *TimerFileDescription) Read(ctx context.Context, dst usermem.IOSequence, opts ReadOptions) (int64, error) { + const sizeofUint64 = 8 + if dst.NumBytes() < sizeofUint64 { + return 0, syserror.EINVAL + } + if val := atomic.SwapUint64(&tfd.val, 0); val != 0 { + var buf [sizeofUint64]byte + usermem.ByteOrder.PutUint64(buf[:], val) + if _, err := dst.CopyOut(ctx, buf[:]); err != nil { + // Linux does not undo consuming the number of + // expirations even if writing to userspace fails. + return 0, err + } + return sizeofUint64, nil + } + return 0, syserror.ErrWouldBlock +} + +// Clock returns the timer fd's Clock. +func (tfd *TimerFileDescription) Clock() ktime.Clock { + return tfd.timer.Clock() +} + +// GetTime returns the associated Timer's setting and the time at which it was +// observed. +func (tfd *TimerFileDescription) GetTime() (ktime.Time, ktime.Setting) { + return tfd.timer.Get() +} + +// SetTime atomically changes the associated Timer's setting, resets the number +// of expirations to 0, and returns the previous setting and the time at which +// it was observed. +func (tfd *TimerFileDescription) SetTime(s ktime.Setting) (ktime.Time, ktime.Setting) { + return tfd.timer.SwapAnd(s, func() { atomic.StoreUint64(&tfd.val, 0) }) +} + +// Readiness implements waiter.Waitable.Readiness. +func (tfd *TimerFileDescription) Readiness(mask waiter.EventMask) waiter.EventMask { + var ready waiter.EventMask + if atomic.LoadUint64(&tfd.val) != 0 { + ready |= waiter.EventIn + } + return ready +} + +// EventRegister implements waiter.Waitable.EventRegister. +func (tfd *TimerFileDescription) EventRegister(e *waiter.Entry, mask waiter.EventMask) { + tfd.events.EventRegister(e, mask) +} + +// EventUnregister implements waiter.Waitable.EventUnregister. +func (tfd *TimerFileDescription) EventUnregister(e *waiter.Entry) { + tfd.events.EventUnregister(e) +} + +// PauseTimer pauses the associated Timer. +func (tfd *TimerFileDescription) PauseTimer() { + tfd.timer.Pause() +} + +// ResumeTimer resumes the associated Timer. +func (tfd *TimerFileDescription) ResumeTimer() { + tfd.timer.Resume() +} + +// Release implements FileDescriptionImpl.Release() +func (tfd *TimerFileDescription) Release() { + tfd.timer.Destroy() +} + +// Notify implements ktime.TimerListener.Notify. +func (tfd *TimerFileDescription) Notify(exp uint64, setting ktime.Setting) (ktime.Setting, bool) { + atomic.AddUint64(&tfd.val, exp) + tfd.events.Notify(waiter.EventIn) + return ktime.Setting{}, false +} + +// Destroy implements ktime.TimerListener.Destroy. +func (tfd *TimerFileDescription) Destroy() {} -- cgit v1.2.3 From dd98fdd5beb7f02e7c7b3aeb4f07f5d00ffc41e7 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Mon, 6 Apr 2020 16:31:27 -0700 Subject: Correctly implement magic symlinks in VFS2 procfs. Updates #1195 PiperOrigin-RevId: 305143567 --- pkg/sentry/fsbridge/vfs.go | 28 +++++++++++++--------- pkg/sentry/fsimpl/kernfs/filesystem.go | 36 +++++++++++++++++++---------- pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 5 ++++ pkg/sentry/fsimpl/kernfs/kernfs.go | 16 ++++++++++++- pkg/sentry/fsimpl/kernfs/symlink.go | 5 ++++ pkg/sentry/fsimpl/proc/task_fds.go | 6 +++++ pkg/sentry/fsimpl/proc/task_files.go | 17 ++++++++++++++ pkg/sentry/fsimpl/proc/tasks_files.go | 10 ++++++++ 8 files changed, 99 insertions(+), 24 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsbridge/vfs.go b/pkg/sentry/fsbridge/vfs.go index 79b808359..89168220a 100644 --- a/pkg/sentry/fsbridge/vfs.go +++ b/pkg/sentry/fsbridge/vfs.go @@ -26,22 +26,22 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) -// fsFile implements File interface over vfs.FileDescription. +// VFSFile implements File interface over vfs.FileDescription. // // +stateify savable -type vfsFile struct { +type VFSFile struct { file *vfs.FileDescription } -var _ File = (*vfsFile)(nil) +var _ File = (*VFSFile)(nil) // NewVFSFile creates a new File over fs.File. func NewVFSFile(file *vfs.FileDescription) File { - return &vfsFile{file: file} + return &VFSFile{file: file} } // PathnameWithDeleted implements File. -func (f *vfsFile) PathnameWithDeleted(ctx context.Context) string { +func (f *VFSFile) PathnameWithDeleted(ctx context.Context) string { root := vfs.RootFromContext(ctx) defer root.DecRef() @@ -51,7 +51,7 @@ func (f *vfsFile) PathnameWithDeleted(ctx context.Context) string { } // ReadFull implements File. -func (f *vfsFile) ReadFull(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { +func (f *VFSFile) ReadFull(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { var total int64 for dst.NumBytes() > 0 { n, err := f.file.PRead(ctx, dst, offset+total, vfs.ReadOptions{}) @@ -67,12 +67,12 @@ func (f *vfsFile) ReadFull(ctx context.Context, dst usermem.IOSequence, offset i } // ConfigureMMap implements File. -func (f *vfsFile) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { +func (f *VFSFile) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { return f.file.ConfigureMMap(ctx, opts) } // Type implements File. -func (f *vfsFile) Type(ctx context.Context) (linux.FileMode, error) { +func (f *VFSFile) Type(ctx context.Context) (linux.FileMode, error) { stat, err := f.file.Stat(ctx, vfs.StatOptions{}) if err != nil { return 0, err @@ -81,15 +81,21 @@ func (f *vfsFile) Type(ctx context.Context) (linux.FileMode, error) { } // IncRef implements File. -func (f *vfsFile) IncRef() { +func (f *VFSFile) IncRef() { f.file.IncRef() } // DecRef implements File. -func (f *vfsFile) DecRef() { +func (f *VFSFile) DecRef() { f.file.DecRef() } +// FileDescription returns the FileDescription represented by f. It does not +// take an additional reference on the returned FileDescription. +func (f *VFSFile) FileDescription() *vfs.FileDescription { + return f.file +} + // fsLookup implements Lookup interface using fs.File. // // +stateify savable @@ -132,5 +138,5 @@ func (l *vfsLookup) OpenPath(ctx context.Context, pathname string, opts vfs.Open if err != nil { return nil, err } - return &vfsFile{file: fd}, nil + return &VFSFile{file: fd}, nil } diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 89f5da3d4..16a3c18ae 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -79,16 +79,22 @@ afterSymlink: } // Resolve any symlink at current path component. if rp.ShouldFollowSymlink() && next.isSymlink() { - // TODO: VFS2 needs something extra for /proc/[pid]/fd/ "magic symlinks". - target, err := next.inode.Readlink(ctx) + targetVD, targetPathname, err := next.inode.Getlink(ctx) if err != nil { return nil, err } - if err := rp.HandleSymlink(target); err != nil { - return nil, err + if targetVD.Ok() { + err := rp.HandleJump(targetVD) + targetVD.DecRef() + if err != nil { + return nil, err + } + } else { + if err := rp.HandleSymlink(targetPathname); err != nil { + return nil, err + } } goto afterSymlink - } rp.Advance() return &next.vfsd, nil @@ -470,19 +476,25 @@ afterTrailingSymlink: } childDentry := childVFSD.Impl().(*Dentry) childInode := childDentry.inode - if rp.ShouldFollowSymlink() { - if childDentry.isSymlink() { - target, err := childInode.Readlink(ctx) + if rp.ShouldFollowSymlink() && childDentry.isSymlink() { + targetVD, targetPathname, err := childInode.Getlink(ctx) + if err != nil { + return nil, err + } + if targetVD.Ok() { + err := rp.HandleJump(targetVD) + targetVD.DecRef() if err != nil { return nil, err } - if err := rp.HandleSymlink(target); err != nil { + } else { + if err := rp.HandleSymlink(targetPathname); err != nil { return nil, err } - // rp.Final() may no longer be true since we now need to resolve the - // symlink target. - goto afterTrailingSymlink } + // rp.Final() may no longer be true since we now need to resolve the + // symlink target. + goto afterTrailingSymlink } if err := childInode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { return nil, err diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 5c84b10c9..65f09af5d 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -181,6 +181,11 @@ func (InodeNotSymlink) Readlink(context.Context) (string, error) { return "", syserror.EINVAL } +// Getlink implements Inode.Getlink. +func (InodeNotSymlink) Getlink(context.Context) (vfs.VirtualDentry, string, error) { + return vfs.VirtualDentry{}, "", syserror.EINVAL +} + // InodeAttrs partially implements the Inode interface, specifically the // inodeMetadata sub interface. InodeAttrs provides functionality related to // inode attributes. diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index 2cefef020..ad76b9f64 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -414,7 +414,21 @@ type inodeDynamicLookup interface { } type inodeSymlink interface { - // Readlink resolves the target of a symbolic link. If an inode is not a + // Readlink returns the target of a symbolic link. If an inode is not a // symlink, the implementation should return EINVAL. Readlink(ctx context.Context) (string, error) + + // Getlink returns the target of a symbolic link, as used by path + // resolution: + // + // - If the inode is a "magic link" (a link whose target is most accurately + // represented as a VirtualDentry), Getlink returns (ok VirtualDentry, "", + // nil). A reference is taken on the returned VirtualDentry. + // + // - If the inode is an ordinary symlink, Getlink returns (zero-value + // VirtualDentry, symlink target, nil). + // + // - If the inode is not a symlink, Getlink returns (zero-value + // VirtualDentry, "", EINVAL). + Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) } diff --git a/pkg/sentry/fsimpl/kernfs/symlink.go b/pkg/sentry/fsimpl/kernfs/symlink.go index 5918d3309..018aa503c 100644 --- a/pkg/sentry/fsimpl/kernfs/symlink.go +++ b/pkg/sentry/fsimpl/kernfs/symlink.go @@ -55,6 +55,11 @@ func (s *StaticSymlink) Readlink(_ context.Context) (string, error) { return s.target, nil } +// Getlink implements Inode.Getlink. +func (s *StaticSymlink) Getlink(_ context.Context) (vfs.VirtualDentry, string, error) { + return vfs.VirtualDentry{}, s.target, nil +} + // SetStat implements Inode.SetStat not allowing inode attributes to be changed. func (*StaticSymlink) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM diff --git a/pkg/sentry/fsimpl/proc/task_fds.go b/pkg/sentry/fsimpl/proc/task_fds.go index 76bfc5307..9c8656b28 100644 --- a/pkg/sentry/fsimpl/proc/task_fds.go +++ b/pkg/sentry/fsimpl/proc/task_fds.go @@ -196,6 +196,12 @@ func (s *fdSymlink) Readlink(ctx context.Context) (string, error) { return vfsObj.PathnameWithDeleted(ctx, root, s.file.VirtualDentry()) } +func (s *fdSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) { + vd := s.file.VirtualDentry() + vd.IncRef() + return vd, "", nil +} + func (s *fdSymlink) DecRef() { s.AtomicRefCount.DecRefWithDestructor(func() { s.Destroy() diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index df0d1bcc5..88ea6a6d8 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -610,6 +610,23 @@ func (s *exeSymlink) Readlink(ctx context.Context) (string, error) { return exec.PathnameWithDeleted(ctx), nil } +// Getlink implements kernfs.Inode.Getlink. +func (s *exeSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) { + if !kernel.ContextCanTrace(ctx, s.task, false) { + return vfs.VirtualDentry{}, "", syserror.EACCES + } + + exec, err := s.executable() + if err != nil { + return vfs.VirtualDentry{}, "", err + } + defer exec.DecRef() + + vd := exec.(*fsbridge.VFSFile).FileDescription().VirtualDentry() + vd.IncRef() + return vd, "", nil +} + func (s *exeSymlink) executable() (file fsbridge.File, err error) { s.task.WithMuLocked(func(t *kernel.Task) { mm := t.MemoryManager() diff --git a/pkg/sentry/fsimpl/proc/tasks_files.go b/pkg/sentry/fsimpl/proc/tasks_files.go index 882c1981e..4621e2de0 100644 --- a/pkg/sentry/fsimpl/proc/tasks_files.go +++ b/pkg/sentry/fsimpl/proc/tasks_files.go @@ -63,6 +63,11 @@ func (s *selfSymlink) Readlink(ctx context.Context) (string, error) { return strconv.FormatUint(uint64(tgid), 10), nil } +func (s *selfSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) { + target, err := s.Readlink(ctx) + return vfs.VirtualDentry{}, target, err +} + // SetStat implements Inode.SetStat not allowing inode attributes to be changed. func (*selfSymlink) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM @@ -101,6 +106,11 @@ func (s *threadSelfSymlink) Readlink(ctx context.Context) (string, error) { return fmt.Sprintf("%d/task/%d", tgid, tid), nil } +func (s *threadSelfSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) { + target, err := s.Readlink(ctx) + return vfs.VirtualDentry{}, target, err +} + // SetStat implements Inode.SetStat not allowing inode attributes to be changed. func (*threadSelfSymlink) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { return syserror.EPERM -- cgit v1.2.3 From 94319a8241cb299edc812024d6132b7a3819a4dc Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Tue, 7 Apr 2020 09:40:38 -0700 Subject: Make gofer.dentry.destroyLocked idempotent gofer operations accumulate dentries touched in a slice to call checkCachingLocked on them when the operation is over. In case the same dentry is touched multiple times during the operation, checkCachingLocked, and consequently destroyLocked, may be called more than once for the same dentry. Updates #1198 PiperOrigin-RevId: 305276819 --- pkg/sentry/fsimpl/gofer/BUILD | 12 ++++++- pkg/sentry/fsimpl/gofer/gofer.go | 36 +++++++++++++++++--- pkg/sentry/fsimpl/gofer/gofer_test.go | 64 +++++++++++++++++++++++++++++++++++ test/syscalls/linux/open.cc | 22 ++++++++++++ 4 files changed, 129 insertions(+), 5 deletions(-) create mode 100644 pkg/sentry/fsimpl/gofer/gofer_test.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/BUILD b/pkg/sentry/fsimpl/gofer/BUILD index d15a36709..99d1e3f8f 100644 --- a/pkg/sentry/fsimpl/gofer/BUILD +++ b/pkg/sentry/fsimpl/gofer/BUILD @@ -1,4 +1,4 @@ -load("//tools:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") licenses(["notice"]) @@ -54,3 +54,13 @@ go_library( "//pkg/usermem", ], ) + +go_test( + name = "gofer_test", + srcs = ["gofer_test.go"], + library = ":gofer", + deps = [ + "//pkg/p9", + "//pkg/sentry/contexttest", + ], +) diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index adee8bb60..20edaf643 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -444,7 +444,8 @@ type dentry struct { // refs is the reference count. Each dentry holds a reference on its // parent, even if disowned. refs is accessed using atomic memory - // operations. + // operations. When refs reaches 0, the dentry may be added to the cache or + // destroyed. If refs==-1 the dentry has already been destroyed. refs int64 // fs is the owning filesystem. fs is immutable. @@ -860,7 +861,7 @@ func (d *dentry) IncRef() { func (d *dentry) TryIncRef() bool { for { refs := atomic.LoadInt64(&d.refs) - if refs == 0 { + if refs <= 0 { return false } if atomic.CompareAndSwapInt64(&d.refs, refs, refs+1) { @@ -883,13 +884,20 @@ func (d *dentry) DecRef() { // checkCachingLocked should be called after d's reference count becomes 0 or it // becomes disowned. // +// It may be called on a destroyed dentry. For example, +// renameMu[R]UnlockAndCheckCaching may call checkCachingLocked multiple times +// for the same dentry when the dentry is visited more than once in the same +// operation. One of the calls may destroy the dentry, so subsequent calls will +// do nothing. +// // Preconditions: d.fs.renameMu must be locked for writing. func (d *dentry) checkCachingLocked() { // Dentries with a non-zero reference count must be retained. (The only way // to obtain a reference on a dentry with zero references is via path // resolution, which requires renameMu, so if d.refs is zero then it will // remain zero while we hold renameMu for writing.) - if atomic.LoadInt64(&d.refs) != 0 { + refs := atomic.LoadInt64(&d.refs) + if refs > 0 { if d.cached { d.fs.cachedDentries.Remove(d) d.fs.cachedDentriesLen-- @@ -897,6 +905,10 @@ func (d *dentry) checkCachingLocked() { } return } + if refs == -1 { + // Dentry has already been destroyed. + return + } // Non-child dentries with zero references are no longer reachable by path // resolution and should be dropped immediately. if d.vfsd.Parent() == nil || d.vfsd.IsDisowned() { @@ -949,9 +961,22 @@ func (d *dentry) checkCachingLocked() { } } +// destroyLocked destroys the dentry. It may flushes dirty pages from cache, +// close p9 file and remove reference on parent dentry. +// // Preconditions: d.fs.renameMu must be locked for writing. d.refs == 0. d is // not a child dentry. func (d *dentry) destroyLocked() { + switch atomic.LoadInt64(&d.refs) { + case 0: + // Mark the dentry destroyed. + atomic.StoreInt64(&d.refs, -1) + case -1: + panic("dentry.destroyLocked() called on already destroyed dentry") + default: + panic("dentry.destroyLocked() called with references on the dentry") + } + ctx := context.Background() d.handleMu.Lock() if !d.handle.file.isNil() { @@ -971,7 +996,10 @@ func (d *dentry) destroyLocked() { d.handle.close(ctx) } d.handleMu.Unlock() - d.file.close(ctx) + if !d.file.isNil() { + d.file.close(ctx) + d.file = p9file{} + } // Remove d from the set of all dentries. d.fs.syncMu.Lock() delete(d.fs.dentries, d) diff --git a/pkg/sentry/fsimpl/gofer/gofer_test.go b/pkg/sentry/fsimpl/gofer/gofer_test.go new file mode 100644 index 000000000..82bc239db --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/gofer_test.go @@ -0,0 +1,64 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package gofer + +import ( + "sync/atomic" + "testing" + + "gvisor.dev/gvisor/pkg/p9" + "gvisor.dev/gvisor/pkg/sentry/contexttest" +) + +func TestDestroyIdempotent(t *testing.T) { + fs := filesystem{ + dentries: make(map[*dentry]struct{}), + opts: filesystemOptions{ + // Test relies on no dentry being held in the cache. + maxCachedDentries: 0, + }, + } + + ctx := contexttest.Context(t) + attr := &p9.Attr{ + Mode: p9.ModeRegular, + } + mask := p9.AttrMask{ + Mode: true, + Size: true, + } + parent, err := fs.newDentry(ctx, p9file{}, p9.QID{}, mask, attr) + if err != nil { + t.Fatalf("fs.newDentry(): %v", err) + } + + child, err := fs.newDentry(ctx, p9file{}, p9.QID{}, mask, attr) + if err != nil { + t.Fatalf("fs.newDentry(): %v", err) + } + parent.IncRef() // reference held by child on its parent. + parent.vfsd.InsertChild(&child.vfsd, "child") + + child.checkCachingLocked() + if got := atomic.LoadInt64(&child.refs); got != -1 { + t.Fatalf("child.refs=%d, want: -1", got) + } + // Parent will also be destroyed when child reference is removed. + if got := atomic.LoadInt64(&parent.refs); got != -1 { + t.Fatalf("parent.refs=%d, want: -1", got) + } + child.checkCachingLocked() + child.checkCachingLocked() +} diff --git a/test/syscalls/linux/open.cc b/test/syscalls/linux/open.cc index 267ae19f6..640fe6bfc 100644 --- a/test/syscalls/linux/open.cc +++ b/test/syscalls/linux/open.cc @@ -186,6 +186,28 @@ TEST_F(OpenTest, OpenNoFollowStillFollowsLinksInPath) { ASSERT_NO_ERRNO_AND_VALUE(Open(path_via_symlink, O_RDONLY | O_NOFOLLOW)); } +// Test that open(2) can follow symlinks that point back to the same tree. +// Test sets up files as follows: +// root/child/symlink => redirects to ../.. +// root/child/target => regular file +// +// open("root/child/symlink/root/child/file") +TEST_F(OpenTest, SymlinkRecurse) { + auto root = + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDirIn(GetAbsoluteTestTmpdir())); + auto child = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDirIn(root.path())); + auto symlink = ASSERT_NO_ERRNO_AND_VALUE( + TempPath::CreateSymlinkTo(child.path(), "../..")); + auto target = ASSERT_NO_ERRNO_AND_VALUE( + TempPath::CreateFileWith(child.path(), "abc", 0644)); + auto path_via_symlink = + JoinPath(symlink.path(), Basename(root.path()), Basename(child.path()), + Basename(target.path())); + const auto contents = + ASSERT_NO_ERRNO_AND_VALUE(GetContents(path_via_symlink)); + ASSERT_EQ(contents, "abc"); +} + TEST_F(OpenTest, Fault) { char* totally_not_null = nullptr; ASSERT_THAT(open(totally_not_null, O_RDONLY), SyscallFailsWithErrno(EFAULT)); -- cgit v1.2.3 From d5ddb5365086b13c0688c40fc74fa4cc4c5528db Mon Sep 17 00:00:00 2001 From: Ting-Yu Wang Date: Tue, 7 Apr 2020 14:32:24 -0700 Subject: Remove out-of-date TODOs. We already have network namespace for netstack. PiperOrigin-RevId: 305341954 --- pkg/sentry/socket/netstack/provider.go | 8 -------- 1 file changed, 8 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/provider.go b/pkg/sentry/socket/netstack/provider.go index eb090e79b..c3f04b613 100644 --- a/pkg/sentry/socket/netstack/provider.go +++ b/pkg/sentry/socket/netstack/provider.go @@ -62,10 +62,6 @@ func getTransportProtocol(ctx context.Context, stype linux.SockType, protocol in } case linux.SOCK_RAW: - // TODO(b/142504697): "In order to create a raw socket, a - // process must have the CAP_NET_RAW capability in the user - // namespace that governs its network namespace." - raw(7) - // Raw sockets require CAP_NET_RAW. creds := auth.CredentialsFromContext(ctx) if !creds.HasCapability(linux.CAP_NET_RAW) { @@ -141,10 +137,6 @@ func (p *provider) Socket(t *kernel.Task, stype linux.SockType, protocol int) (* } func packetSocket(t *kernel.Task, epStack *Stack, stype linux.SockType, protocol int) (*fs.File, *syserr.Error) { - // TODO(b/142504697): "In order to create a packet socket, a process - // must have the CAP_NET_RAW capability in the user namespace that - // governs its network namespace." - packet(7) - // Packet sockets require CAP_NET_RAW. creds := auth.CredentialsFromContext(t) if !creds.HasCapability(linux.CAP_NET_RAW) { -- cgit v1.2.3 From 693b6bdda9206a5910c552a3997b2df5480d6947 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 7 Apr 2020 16:16:31 -0700 Subject: Correctly distinguish between seekable and non-seekable host fds. Check whether an fd is seekable by calling the seek syscall and examining the return value, instead of checking the file type, which is inaccurate. PiperOrigin-RevId: 305361593 --- pkg/sentry/fsimpl/host/host.go | 42 ++++++++++++++++++++---------------------- 1 file changed, 20 insertions(+), 22 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 7d9dcd4c9..97fa7f7ab 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -74,31 +74,34 @@ func ImportFD(mnt *vfs.Mount, hostFD int, isTTY bool) (*vfs.FileDescription, err } // Retrieve metadata. - var s syscall.Stat_t - if err := syscall.Fstat(hostFD, &s); err != nil { + var s unix.Stat_t + if err := unix.Fstat(hostFD, &s); err != nil { return nil, err } fileMode := linux.FileMode(s.Mode) fileType := fileMode.FileType() - // Pipes, character devices, and sockets. - isStream := fileType == syscall.S_IFIFO || fileType == syscall.S_IFCHR || fileType == syscall.S_IFSOCK + + // Determine if hostFD is seekable. If not, this syscall will return ESPIPE + // (see fs/read_write.c:llseek), e.g. for pipes, sockets, and some character + // devices. + _, err := unix.Seek(hostFD, 0, linux.SEEK_CUR) + seekable := err != syserror.ESPIPE i := &inode{ hostFD: hostFD, - isStream: isStream, + seekable: seekable, isTTY: isTTY, canMap: canMap(uint32(fileType)), ino: fs.NextIno(), mode: fileMode, - // For simplicity, set offset to 0. Technically, we should - // only set to 0 on files that are not seekable (sockets, pipes, etc.), - // and use the offset from the host fd otherwise. + // For simplicity, set offset to 0. Technically, we should use the existing + // offset on the host if the file is seekable. offset: 0, } - // These files can't be memory mapped, assert this. - if i.isStream && i.canMap { + // Non-seekable files can't be memory mapped, assert this. + if !i.seekable && i.canMap { panic("files that can return EWOULDBLOCK (sockets, pipes, etc.) cannot be memory mapped") } @@ -124,12 +127,12 @@ type inode struct { // This field is initialized at creation time and is immutable. hostFD int - // isStream is true if the host fd points to a file representing a stream, + // seekable is false if the host fd points to a file representing a stream, // e.g. a socket or a pipe. Such files are not seekable and can return // EWOULDBLOCK for I/O operations. // // This field is initialized at creation time and is immutable. - isStream bool + seekable bool // isTTY is true if this file represents a TTY. // @@ -481,8 +484,7 @@ func (f *fileDescription) Release() { // PRead implements FileDescriptionImpl. func (f *fileDescription) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { i := f.inode - // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null. - if i.isStream { + if !i.seekable { return 0, syserror.ESPIPE } @@ -492,8 +494,7 @@ func (f *fileDescription) PRead(ctx context.Context, dst usermem.IOSequence, off // Read implements FileDescriptionImpl. func (f *fileDescription) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { i := f.inode - // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null. - if i.isStream { + if !i.seekable { n, err := readFromHostFD(ctx, i.hostFD, dst, -1, opts.Flags) if isBlockError(err) { // If we got any data at all, return it as a "completed" partial read @@ -538,8 +539,7 @@ func readFromHostFD(ctx context.Context, hostFD int, dst usermem.IOSequence, off // PWrite implements FileDescriptionImpl. func (f *fileDescription) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { i := f.inode - // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null. - if i.isStream { + if !i.seekable { return 0, syserror.ESPIPE } @@ -549,8 +549,7 @@ func (f *fileDescription) PWrite(ctx context.Context, src usermem.IOSequence, of // Write implements FileDescriptionImpl. func (f *fileDescription) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { i := f.inode - // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null. - if i.isStream { + if !i.seekable { n, err := writeToHostFD(ctx, i.hostFD, src, -1, opts.Flags) if isBlockError(err) { err = syserror.ErrWouldBlock @@ -593,8 +592,7 @@ func writeToHostFD(ctx context.Context, hostFD int, src usermem.IOSequence, offs // allow directory fds to be imported at all. func (f *fileDescription) Seek(_ context.Context, offset int64, whence int32) (int64, error) { i := f.inode - // TODO(b/34716638): Some char devices do support seeking, e.g. /dev/null. - if i.isStream { + if !i.seekable { return 0, syserror.ESPIPE } -- cgit v1.2.3 From 5a1324625f1d0d9ea2d4874f9d6d1008ec12f45e Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Tue, 7 Apr 2020 18:26:34 -0700 Subject: Make unlink tests pass with goferfs Required directory checks were being skipped when there was no child cached. Now the code always loads the child file before unlinking it. Updates #1198 PiperOrigin-RevId: 305382323 --- pkg/sentry/fsimpl/gofer/filesystem.go | 5 +++++ 1 file changed, 5 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 305228bda..137260898 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -437,14 +437,19 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b flags := uint32(0) if dir { if child != nil && !child.isDir() { + vfsObj.AbortDeleteDentry(childVFSD) return syserror.ENOTDIR } flags = linux.AT_REMOVEDIR } else { if child != nil && child.isDir() { + vfsObj.AbortDeleteDentry(childVFSD) return syserror.EISDIR } if rp.MustBeDir() { + if childVFSD != nil { + vfsObj.AbortDeleteDentry(childVFSD) + } return syserror.ENOTDIR } } -- cgit v1.2.3 From b574c715a799e476ac788e5f5b2c68f1a00b3538 Mon Sep 17 00:00:00 2001 From: Haibo Xu Date: Wed, 4 Mar 2020 05:44:46 +0000 Subject: Move pagetables.limitPCID to arch-specific file. X86 provide 12 bits for PCID while arm64 support 8/16 bits ASID. Signed-off-by: Haibo Xu Change-Id: I0bd9236e44e6b6c4c88eb6e9adc5ac27b918bf6c --- pkg/sentry/platform/ring0/pagetables/BUILD | 3 ++ pkg/sentry/platform/ring0/pagetables/pcids.go | 5 +-- .../platform/ring0/pagetables/pcids_aarch64.go | 32 +++++++++++++++ .../platform/ring0/pagetables/pcids_aarch64.s | 45 ++++++++++++++++++++++ pkg/sentry/platform/ring0/pagetables/pcids_x86.go | 20 ++++++++++ 5 files changed, 101 insertions(+), 4 deletions(-) create mode 100644 pkg/sentry/platform/ring0/pagetables/pcids_aarch64.go create mode 100644 pkg/sentry/platform/ring0/pagetables/pcids_aarch64.s create mode 100644 pkg/sentry/platform/ring0/pagetables/pcids_x86.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/ring0/pagetables/BUILD b/pkg/sentry/platform/ring0/pagetables/BUILD index 581841555..16d5f478b 100644 --- a/pkg/sentry/platform/ring0/pagetables/BUILD +++ b/pkg/sentry/platform/ring0/pagetables/BUILD @@ -81,6 +81,9 @@ go_library( "pagetables_arm64.go", "pagetables_x86.go", "pcids.go", + "pcids_aarch64.go", + "pcids_aarch64.s", + "pcids_x86.go", "walker_amd64.go", "walker_arm64.go", "walker_empty.go", diff --git a/pkg/sentry/platform/ring0/pagetables/pcids.go b/pkg/sentry/platform/ring0/pagetables/pcids.go index 9206030bf..964496aac 100644 --- a/pkg/sentry/platform/ring0/pagetables/pcids.go +++ b/pkg/sentry/platform/ring0/pagetables/pcids.go @@ -18,9 +18,6 @@ import ( "gvisor.dev/gvisor/pkg/sync" ) -// limitPCID is the number of valid PCIDs. -const limitPCID = 4096 - // PCIDs is a simple PCID database. // // This is not protected by locks and is thus suitable for use only with a @@ -44,7 +41,7 @@ type PCIDs struct { // // Nil is returned iff the start and size are out of range. func NewPCIDs(start, size uint16) *PCIDs { - if start+uint16(size) >= limitPCID { + if start+uint16(size) > limitPCID { return nil // See comment. } p := &PCIDs{ diff --git a/pkg/sentry/platform/ring0/pagetables/pcids_aarch64.go b/pkg/sentry/platform/ring0/pagetables/pcids_aarch64.go new file mode 100644 index 000000000..fbfd41d83 --- /dev/null +++ b/pkg/sentry/platform/ring0/pagetables/pcids_aarch64.go @@ -0,0 +1,32 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package pagetables + +// limitPCID is the maximum value of PCIDs. +// +// In VMSAv8-64, the PCID(ASID) size is an IMPLEMENTATION DEFINED choice +// of 8 bits or 16 bits, and ID_AA64MMFR0_EL1.ASIDBits identifies the +// supported size. When an implementation supports a 16-bit ASID, TCR_ELx.AS +// selects whether the top 8 bits of the ASID are used. +var limitPCID uint16 + +// GetASIDBits return the system ASID bits, 8 or 16 bits. +func GetASIDBits() uint8 + +func init() { + limitPCID = uint16(1)< Date: Wed, 25 Mar 2020 16:55:02 -0700 Subject: Fix all copy locks violations. This required minor restructuring of how system call tables were saved and restored, but it makes way more sense this way. Updates #2243 --- pkg/log/glog.go | 6 +++--- pkg/log/json.go | 2 +- pkg/log/json_k8s.go | 4 ++-- pkg/log/log.go | 2 +- pkg/log/log_test.go | 6 +++--- pkg/sentry/contexttest/contexttest.go | 4 ++-- pkg/sentry/fs/host/socket_test.go | 6 +++--- pkg/sentry/fs/proc/sys_net.go | 4 ++-- pkg/sentry/kernel/syscalls.go | 33 ++++++++++++++++---------------- pkg/sentry/kernel/syscalls_state.go | 36 ++++++++++++++++++++++++++--------- pkg/sentry/kernel/task_context.go | 2 +- pkg/sentry/kernel/time/time.go | 10 +++++----- pkg/state/state.go | 5 +---- runsc/boot/compat.go | 2 +- runsc/main.go | 6 +++--- tools/go_stateify/main.go | 2 +- tools/nogo.json | 13 ------------- 17 files changed, 72 insertions(+), 71 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/log/glog.go b/pkg/log/glog.go index b4f7bb5a4..f57c4427b 100644 --- a/pkg/log/glog.go +++ b/pkg/log/glog.go @@ -25,7 +25,7 @@ import ( // GoogleEmitter is a wrapper that emits logs in a format compatible with // package github.com/golang/glog. type GoogleEmitter struct { - Writer + *Writer } // pid is used for the threadid component of the header. @@ -46,7 +46,7 @@ var pid = os.Getpid() // line The line number // msg The user-supplied message // -func (g *GoogleEmitter) Emit(depth int, level Level, timestamp time.Time, format string, args ...interface{}) { +func (g GoogleEmitter) Emit(depth int, level Level, timestamp time.Time, format string, args ...interface{}) { // Log level. prefix := byte('?') switch level { @@ -81,5 +81,5 @@ func (g *GoogleEmitter) Emit(depth int, level Level, timestamp time.Time, format message := fmt.Sprintf(format, args...) // Emit the formatted result. - fmt.Fprintf(&g.Writer, "%c%02d%02d %02d:%02d:%02d.%06d % 7d %s:%d] %s\n", prefix, int(month), day, hour, minute, second, microsecond, pid, file, line, message) + fmt.Fprintf(g.Writer, "%c%02d%02d %02d:%02d:%02d.%06d % 7d %s:%d] %s\n", prefix, int(month), day, hour, minute, second, microsecond, pid, file, line, message) } diff --git a/pkg/log/json.go b/pkg/log/json.go index 0943db1cc..bdf9d691e 100644 --- a/pkg/log/json.go +++ b/pkg/log/json.go @@ -58,7 +58,7 @@ func (lv *Level) UnmarshalJSON(b []byte) error { // JSONEmitter logs messages in json format. type JSONEmitter struct { - Writer + *Writer } // Emit implements Emitter.Emit. diff --git a/pkg/log/json_k8s.go b/pkg/log/json_k8s.go index 6c6fc8b6f..5883e95e1 100644 --- a/pkg/log/json_k8s.go +++ b/pkg/log/json_k8s.go @@ -29,11 +29,11 @@ type k8sJSONLog struct { // K8sJSONEmitter logs messages in json format that is compatible with // Kubernetes fluent configuration. type K8sJSONEmitter struct { - Writer + *Writer } // Emit implements Emitter.Emit. -func (e *K8sJSONEmitter) Emit(_ int, level Level, timestamp time.Time, format string, v ...interface{}) { +func (e K8sJSONEmitter) Emit(_ int, level Level, timestamp time.Time, format string, v ...interface{}) { j := k8sJSONLog{ Log: fmt.Sprintf(format, v...), Level: level, diff --git a/pkg/log/log.go b/pkg/log/log.go index a794da1aa..37e0605ad 100644 --- a/pkg/log/log.go +++ b/pkg/log/log.go @@ -374,5 +374,5 @@ func CopyStandardLogTo(l Level) error { func init() { // Store the initial value for the log. - log.Store(&BasicLogger{Level: Info, Emitter: &GoogleEmitter{Writer{Next: os.Stderr}}}) + log.Store(&BasicLogger{Level: Info, Emitter: GoogleEmitter{&Writer{Next: os.Stderr}}}) } diff --git a/pkg/log/log_test.go b/pkg/log/log_test.go index 402cc29ae..9ff18559b 100644 --- a/pkg/log/log_test.go +++ b/pkg/log/log_test.go @@ -52,7 +52,7 @@ func TestDropMessages(t *testing.T) { t.Fatalf("Write should have failed") } - fmt.Printf("writer: %+v\n", w) + fmt.Printf("writer: %#v\n", &w) tw.fail = false if _, err := w.Write([]byte("line 2\n")); err != nil { @@ -76,7 +76,7 @@ func TestDropMessages(t *testing.T) { func TestCaller(t *testing.T) { tw := &testWriter{} - e := &GoogleEmitter{Writer: Writer{Next: tw}} + e := GoogleEmitter{Writer: &Writer{Next: tw}} bl := &BasicLogger{ Emitter: e, Level: Debug, @@ -94,7 +94,7 @@ func BenchmarkGoogleLogging(b *testing.B) { tw := &testWriter{ limit: 1, // Only record one message. } - e := &GoogleEmitter{Writer: Writer{Next: tw}} + e := GoogleEmitter{Writer: &Writer{Next: tw}} bl := &BasicLogger{ Emitter: e, Level: Debug, diff --git a/pkg/sentry/contexttest/contexttest.go b/pkg/sentry/contexttest/contexttest.go index 031fc64ec..8e5658c7a 100644 --- a/pkg/sentry/contexttest/contexttest.go +++ b/pkg/sentry/contexttest/contexttest.go @@ -97,7 +97,7 @@ type hostClock struct { } // Now implements ktime.Clock.Now. -func (hostClock) Now() ktime.Time { +func (*hostClock) Now() ktime.Time { return ktime.FromNanoseconds(time.Now().UnixNano()) } @@ -127,7 +127,7 @@ func (t *TestContext) Value(key interface{}) interface{} { case uniqueid.CtxInotifyCookie: return atomic.AddUint32(&lastInotifyCookie, 1) case ktime.CtxRealtimeClock: - return hostClock{} + return &hostClock{} default: if val, ok := t.otherValues[key]; ok { return val diff --git a/pkg/sentry/fs/host/socket_test.go b/pkg/sentry/fs/host/socket_test.go index eb4afe520..affdbcacb 100644 --- a/pkg/sentry/fs/host/socket_test.go +++ b/pkg/sentry/fs/host/socket_test.go @@ -199,14 +199,14 @@ func TestListen(t *testing.T) { } func TestPasscred(t *testing.T) { - e := ConnectedEndpoint{} + e := &ConnectedEndpoint{} if got, want := e.Passcred(), false; got != want { t.Errorf("Got %#v.Passcred() = %t, want = %t", e, got, want) } } func TestGetLocalAddress(t *testing.T) { - e := ConnectedEndpoint{path: "foo"} + e := &ConnectedEndpoint{path: "foo"} want := tcpip.FullAddress{Addr: tcpip.Address("foo")} if got, err := e.GetLocalAddress(); err != nil || got != want { t.Errorf("Got %#v.GetLocalAddress() = %#v, %v, want = %#v, %v", e, got, err, want, nil) @@ -214,7 +214,7 @@ func TestGetLocalAddress(t *testing.T) { } func TestQueuedSize(t *testing.T) { - e := ConnectedEndpoint{} + e := &ConnectedEndpoint{} tests := []struct { name string f func() int64 diff --git a/pkg/sentry/fs/proc/sys_net.go b/pkg/sentry/fs/proc/sys_net.go index d4c4b533d..702fdd392 100644 --- a/pkg/sentry/fs/proc/sys_net.go +++ b/pkg/sentry/fs/proc/sys_net.go @@ -80,7 +80,7 @@ func newTCPMemInode(ctx context.Context, msrc *fs.MountSource, s inet.Stack, dir } // Truncate implements fs.InodeOperations.Truncate. -func (tcpMemInode) Truncate(context.Context, *fs.Inode, int64) error { +func (*tcpMemInode) Truncate(context.Context, *fs.Inode, int64) error { return nil } @@ -196,7 +196,7 @@ func newTCPSackInode(ctx context.Context, msrc *fs.MountSource, s inet.Stack) *f } // Truncate implements fs.InodeOperations.Truncate. -func (tcpSack) Truncate(context.Context, *fs.Inode, int64) error { +func (*tcpSack) Truncate(context.Context, *fs.Inode, int64) error { return nil } diff --git a/pkg/sentry/kernel/syscalls.go b/pkg/sentry/kernel/syscalls.go index 93c4fe969..c9a2321b8 100644 --- a/pkg/sentry/kernel/syscalls.go +++ b/pkg/sentry/kernel/syscalls.go @@ -218,56 +218,55 @@ type Stracer interface { SyscallExit(context interface{}, t *Task, sysno, rval uintptr, err error) } -// SyscallTable is a lookup table of system calls. Critically, a SyscallTable -// is *immutable*. In order to make supporting suspend and resume sane, they -// must be uniquely registered and may not change during operation. +// SyscallTable is a lookup table of system calls. // -// +stateify savable +// Note that a SyscallTable is not savable directly. Instead, they are saved as +// an OS/Arch pair and lookup happens again on restore. type SyscallTable struct { // OS is the operating system that this syscall table implements. - OS abi.OS `state:"wait"` + OS abi.OS // Arch is the architecture that this syscall table targets. - Arch arch.Arch `state:"wait"` + Arch arch.Arch // The OS version that this syscall table implements. - Version Version `state:"manual"` + Version Version // AuditNumber is a numeric constant that represents the syscall table. If // non-zero, auditNumber must be one of the AUDIT_ARCH_* values defined by // linux/audit.h. - AuditNumber uint32 `state:"manual"` + AuditNumber uint32 // Table is the collection of functions. - Table map[uintptr]Syscall `state:"manual"` + Table map[uintptr]Syscall // lookup is a fixed-size array that holds the syscalls (indexed by // their numbers). It is used for fast look ups. - lookup []SyscallFn `state:"manual"` + lookup []SyscallFn // Emulate is a collection of instruction addresses to emulate. The // keys are addresses, and the values are system call numbers. - Emulate map[usermem.Addr]uintptr `state:"manual"` + Emulate map[usermem.Addr]uintptr // The function to call in case of a missing system call. - Missing MissingFn `state:"manual"` + Missing MissingFn // Stracer traces this syscall table. - Stracer Stracer `state:"manual"` + Stracer Stracer // External is used to handle an external callback. - External func(*Kernel) `state:"manual"` + External func(*Kernel) // ExternalFilterBefore is called before External is called before the syscall is executed. // External is not called if it returns false. - ExternalFilterBefore func(*Task, uintptr, arch.SyscallArguments) bool `state:"manual"` + ExternalFilterBefore func(*Task, uintptr, arch.SyscallArguments) bool // ExternalFilterAfter is called before External is called after the syscall is executed. // External is not called if it returns false. - ExternalFilterAfter func(*Task, uintptr, arch.SyscallArguments) bool `state:"manual"` + ExternalFilterAfter func(*Task, uintptr, arch.SyscallArguments) bool // FeatureEnable stores the strace and one-shot enable bits. - FeatureEnable SyscallFlagsTable `state:"manual"` + FeatureEnable SyscallFlagsTable } // allSyscallTables contains all known tables. diff --git a/pkg/sentry/kernel/syscalls_state.go b/pkg/sentry/kernel/syscalls_state.go index 00358326b..90f890495 100644 --- a/pkg/sentry/kernel/syscalls_state.go +++ b/pkg/sentry/kernel/syscalls_state.go @@ -14,16 +14,34 @@ package kernel -import "fmt" +import ( + "fmt" -// afterLoad is invoked by stateify. -func (s *SyscallTable) afterLoad() { - otherTable, ok := LookupSyscallTable(s.OS, s.Arch) - if !ok { - // Couldn't find a reference? - panic(fmt.Sprintf("syscall table not found for OS %v Arch %v", s.OS, s.Arch)) + "gvisor.dev/gvisor/pkg/abi" + "gvisor.dev/gvisor/pkg/sentry/arch" +) + +// syscallTableInfo is used to reload the SyscallTable. +// +// +stateify savable +type syscallTableInfo struct { + OS abi.OS + Arch arch.Arch +} + +// saveSt saves the SyscallTable. +func (tc *TaskContext) saveSt() syscallTableInfo { + return syscallTableInfo{ + OS: tc.st.OS, + Arch: tc.st.Arch, } +} - // Copy the table. - *s = *otherTable +// loadSt loads the SyscallTable. +func (tc *TaskContext) loadSt(sti syscallTableInfo) { + st, ok := LookupSyscallTable(sti.OS, sti.Arch) + if !ok { + panic(fmt.Sprintf("syscall table not found for OS %v, Arch %v", sti.OS, sti.Arch)) + } + tc.st = st // Save the table reference. } diff --git a/pkg/sentry/kernel/task_context.go b/pkg/sentry/kernel/task_context.go index 0158b1788..c115e8d1f 100644 --- a/pkg/sentry/kernel/task_context.go +++ b/pkg/sentry/kernel/task_context.go @@ -49,7 +49,7 @@ type TaskContext struct { fu *futex.Manager // st is the task's syscall table. - st *SyscallTable + st *SyscallTable `state:".(syscallTableInfo)"` } // release releases all resources held by the TaskContext. release is called by diff --git a/pkg/sentry/kernel/time/time.go b/pkg/sentry/kernel/time/time.go index 706de83ef..e959700f2 100644 --- a/pkg/sentry/kernel/time/time.go +++ b/pkg/sentry/kernel/time/time.go @@ -245,7 +245,7 @@ type Clock interface { type WallRateClock struct{} // WallTimeUntil implements Clock.WallTimeUntil. -func (WallRateClock) WallTimeUntil(t, now Time) time.Duration { +func (*WallRateClock) WallTimeUntil(t, now Time) time.Duration { return t.Sub(now) } @@ -254,16 +254,16 @@ func (WallRateClock) WallTimeUntil(t, now Time) time.Duration { type NoClockEvents struct{} // Readiness implements waiter.Waitable.Readiness. -func (NoClockEvents) Readiness(mask waiter.EventMask) waiter.EventMask { +func (*NoClockEvents) Readiness(mask waiter.EventMask) waiter.EventMask { return 0 } // EventRegister implements waiter.Waitable.EventRegister. -func (NoClockEvents) EventRegister(e *waiter.Entry, mask waiter.EventMask) { +func (*NoClockEvents) EventRegister(e *waiter.Entry, mask waiter.EventMask) { } // EventUnregister implements waiter.Waitable.EventUnregister. -func (NoClockEvents) EventUnregister(e *waiter.Entry) { +func (*NoClockEvents) EventUnregister(e *waiter.Entry) { } // ClockEventsQueue implements waiter.Waitable by wrapping waiter.Queue and @@ -273,7 +273,7 @@ type ClockEventsQueue struct { } // Readiness implements waiter.Waitable.Readiness. -func (ClockEventsQueue) Readiness(mask waiter.EventMask) waiter.EventMask { +func (*ClockEventsQueue) Readiness(mask waiter.EventMask) waiter.EventMask { return 0 } diff --git a/pkg/state/state.go b/pkg/state/state.go index dbe507ab4..03ae2dbb0 100644 --- a/pkg/state/state.go +++ b/pkg/state/state.go @@ -241,10 +241,7 @@ func Register(name string, instance interface{}, fns Fns) { // // This function is used by the stateify tool. func IsZeroValue(val interface{}) bool { - if val == nil { - return true - } - return reflect.DeepEqual(val, reflect.Zero(reflect.TypeOf(val)).Interface()) + return val == nil || reflect.ValueOf(val).Elem().IsZero() } // step captures one encoding / decoding step. On each step, there is up to one diff --git a/runsc/boot/compat.go b/runsc/boot/compat.go index 8995d678e..b7cfb35bf 100644 --- a/runsc/boot/compat.go +++ b/runsc/boot/compat.go @@ -65,7 +65,7 @@ func newCompatEmitter(logFD int) (*compatEmitter, error) { if logFD > 0 { f := os.NewFile(uintptr(logFD), "user log file") - target := &log.MultiEmitter{c.sink, &log.K8sJSONEmitter{log.Writer{Next: f}}} + target := &log.MultiEmitter{c.sink, log.K8sJSONEmitter{&log.Writer{Next: f}}} c.sink = &log.BasicLogger{Level: log.Info, Emitter: target} } return c, nil diff --git a/runsc/main.go b/runsc/main.go index 62e184ec9..c1c78529c 100644 --- a/runsc/main.go +++ b/runsc/main.go @@ -342,11 +342,11 @@ func main() { func newEmitter(format string, logFile io.Writer) log.Emitter { switch format { case "text": - return &log.GoogleEmitter{log.Writer{Next: logFile}} + return log.GoogleEmitter{&log.Writer{Next: logFile}} case "json": - return &log.JSONEmitter{log.Writer{Next: logFile}} + return log.JSONEmitter{&log.Writer{Next: logFile}} case "json-k8s": - return &log.K8sJSONEmitter{log.Writer{Next: logFile}} + return log.K8sJSONEmitter{&log.Writer{Next: logFile}} } cmd.Fatalf("invalid log format %q, must be 'text', 'json', or 'json-k8s'", format) panic("unreachable") diff --git a/tools/go_stateify/main.go b/tools/go_stateify/main.go index 3437aa476..309ee9c21 100644 --- a/tools/go_stateify/main.go +++ b/tools/go_stateify/main.go @@ -206,7 +206,7 @@ func main() { initCalls = append(initCalls, fmt.Sprintf("%sRegister(\"%s.%s\", (*%s)(nil), state.Fns{Save: (*%s).save, Load: (*%s).load})", statePrefix, *fullPkg, name, name, name, name)) } emitZeroCheck := func(name string) { - fmt.Fprintf(outputFile, " if !%sIsZeroValue(x.%s) { m.Failf(\"%s is %%v, expected zero\", x.%s) }\n", statePrefix, name, name, name) + fmt.Fprintf(outputFile, " if !%sIsZeroValue(&x.%s) { m.Failf(\"%s is %%#v, expected zero\", &x.%s) }\n", statePrefix, name, name, name) } emitLoadValue := func(name, typName string) { fmt.Fprintf(outputFile, " m.LoadValue(\"%s\", new(%s), func(y interface{}) { x.load%s(y.(%s)) })\n", name, typName, camelCased(name), typName) diff --git a/tools/nogo.json b/tools/nogo.json index 83cb76b93..cc05ba027 100644 --- a/tools/nogo.json +++ b/tools/nogo.json @@ -9,19 +9,6 @@ "/external/": "allowed: not subject to unsafe naming rules" } }, - "copylocks": { - "exclude_files": { - ".*_state_autogen.go": "fix: m.Failf copies by value", - "/pkg/log/json.go": "fix: Emit passes lock by value: gvisor.dev/gvisor/pkg/log.JSONEmitter contains gvisor.dev/gvisor/pkg/log.Writer contains gvisor.dev/gvisor/pkg/sync.Mutex", - "/pkg/log/log_test.go": "fix: call of fmt.Printf copies lock value: gvisor.dev/gvisor/pkg/log.Writer contains gvisor.dev/gvisor/pkg/sync.Mutex", - "/pkg/sentry/fs/host/socket_test.go": "fix: call of t.Errorf copies lock value: gvisor.dev/gvisor/pkg/sentry/fs/host.ConnectedEndpoint contains gvisor.dev/gvisor/pkg/refs.AtomicRefCount contains gvisor.dev/gvisor/pkg/sync.Mutex", - "/pkg/sentry/fs/proc/sys_net.go": "fix: Truncate passes lock by value: gvisor.dev/gvisor/pkg/sentry/fs/proc.tcpMemInode contains gvisor.dev/gvisor/pkg/sentry/fs/fsutil.SimpleFileInode contains gvisor.dev/gvisor/pkg/sentry/fs/fsutil.InodeSimpleAttributes contains gvisor.dev/gvisor/pkg/sync.RWMutex", - "/pkg/sentry/fs/proc/sys_net.go": "fix: Truncate passes lock by value: gvisor.dev/gvisor/pkg/sentry/fs/proc.tcpSack contains gvisor.dev/gvisor/pkg/sentry/fs/fsutil.SimpleFileInode contains gvisor.dev/gvisor/pkg/sentry/fs/fsutil.InodeSimpleAttributes contains gvisor.dev/gvisor/pkg/sync.RWMutex", - "/pkg/sentry/fs/tty/slave.go": "fix: Truncate passes lock by value: gvisor.dev/gvisor/pkg/sentry/fs/tty.slaveInodeOperations contains gvisor.dev/gvisor/pkg/sentry/fs/fsutil.SimpleFileInode contains gvisor.dev/gvisor/pkg/sentry/fs/fsutil.InodeSimpleAttributes contains gvisor.dev/gvisor/pkg/sync.RWMutex", - "/pkg/sentry/kernel/time/time.go": "fix: Readiness passes lock by value: gvisor.dev/gvisor/pkg/sentry/kernel/time.ClockEventsQueue contains gvisor.dev/gvisor/pkg/waiter.Queue contains gvisor.dev/gvisor/pkg/sync.RWMutex", - "/pkg/sentry/kernel/syscalls_state.go": "fix: assignment copies lock value to *s: gvisor.dev/gvisor/pkg/sentry/kernel.SyscallTable contains gvisor.dev/gvisor/pkg/sentry/kernel.SyscallFlagsTable contains gvisor.dev/gvisor/pkg/sync.Mutex" - } - }, "lostcancel": { "exclude_files": { "/pkg/tcpip/network/arp/arp_test.go": "fix: the cancel function returned by context.WithTimeout should be called, not discarded, to avoid a context leak", -- cgit v1.2.3 From 928a7c60b8f02811e9c0fcbed0077efd55471cc4 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Wed, 25 Mar 2020 17:15:05 -0700 Subject: Fix all printf formatting errors. Updates #2243 --- pkg/eventchannel/event_test.go | 4 ++-- pkg/segment/test/segment_test.go | 2 +- pkg/sentry/fs/fdpipe/pipe_test.go | 4 ++-- pkg/sentry/fsimpl/tmpfs/stat_test.go | 8 ++++---- pkg/tcpip/header/eth_test.go | 2 +- pkg/tcpip/header/ndp_test.go | 2 +- pkg/tcpip/link/fdbased/endpoint.go | 2 +- pkg/tcpip/stack/ndp_test.go | 12 ++++++------ pkg/tcpip/stack/stack_test.go | 32 ++++++++++++++++---------------- pkg/tcpip/tcpip_test.go | 2 +- pkg/tcpip/transport/tcp/tcp_test.go | 4 ++-- pkg/tcpip/transport/udp/udp_test.go | 2 +- runsc/container/test_app/test_app.go | 2 +- test/root/cgroup_test.go | 4 ++-- test/runtimes/blacklist_test.go | 2 +- test/runtimes/runner.go | 2 +- tools/nogo.json | 31 ------------------------------- 17 files changed, 43 insertions(+), 74 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/eventchannel/event_test.go b/pkg/eventchannel/event_test.go index 7f41b4a27..43750360b 100644 --- a/pkg/eventchannel/event_test.go +++ b/pkg/eventchannel/event_test.go @@ -78,7 +78,7 @@ func TestMultiEmitter(t *testing.T) { for _, name := range names { m := testMessage{name: name} if _, err := me.Emit(m); err != nil { - t.Fatal("me.Emit(%v) failed: %v", m, err) + t.Fatalf("me.Emit(%v) failed: %v", m, err) } } @@ -96,7 +96,7 @@ func TestMultiEmitter(t *testing.T) { // Close multiEmitter. if err := me.Close(); err != nil { - t.Fatal("me.Close() failed: %v", err) + t.Fatalf("me.Close() failed: %v", err) } // All testEmitters should be closed. diff --git a/pkg/segment/test/segment_test.go b/pkg/segment/test/segment_test.go index f19a005f3..97b16c158 100644 --- a/pkg/segment/test/segment_test.go +++ b/pkg/segment/test/segment_test.go @@ -63,7 +63,7 @@ func checkSet(s *Set, expectedSegments int) error { return fmt.Errorf("incorrect order: key %d (segment %d) >= key %d (segment %d)", prev, nrSegments-1, next, nrSegments) } if got, want := seg.Value(), seg.Start()+valueOffset; got != want { - return fmt.Errorf("segment %d has key %d, value %d (expected %d)", nrSegments, seg.Start, got, want) + return fmt.Errorf("segment %d has key %d, value %d (expected %d)", nrSegments, seg.Start(), got, want) } prev = next havePrev = true diff --git a/pkg/sentry/fs/fdpipe/pipe_test.go b/pkg/sentry/fs/fdpipe/pipe_test.go index 5aff0cc95..a0082ecca 100644 --- a/pkg/sentry/fs/fdpipe/pipe_test.go +++ b/pkg/sentry/fs/fdpipe/pipe_test.go @@ -119,7 +119,7 @@ func TestNewPipe(t *testing.T) { continue } if flags := p.flags; test.flags != flags { - t.Errorf("%s: got file flags %s, want %s", test.desc, flags, test.flags) + t.Errorf("%s: got file flags %v, want %v", test.desc, flags, test.flags) continue } if len(test.readAheadBuffer) != len(p.readAheadBuffer) { @@ -136,7 +136,7 @@ func TestNewPipe(t *testing.T) { continue } if !fdnotifier.HasFD(int32(f.FD())) { - t.Errorf("%s: pipe fd %d is not registered for events", test.desc, f.FD) + t.Errorf("%s: pipe fd %d is not registered for events", test.desc, f.FD()) } } } diff --git a/pkg/sentry/fsimpl/tmpfs/stat_test.go b/pkg/sentry/fsimpl/tmpfs/stat_test.go index 3e02e7190..d4f59ee5b 100644 --- a/pkg/sentry/fsimpl/tmpfs/stat_test.go +++ b/pkg/sentry/fsimpl/tmpfs/stat_test.go @@ -140,7 +140,7 @@ func TestSetStatAtime(t *testing.T) { Mask: 0, Atime: linux.NsecToStatxTimestamp(100), }}); err != nil { - t.Errorf("SetStat atime without mask failed: %v") + t.Errorf("SetStat atime without mask failed: %v", err) } // Atime should be unchanged. if gotStat, err := fd.Stat(ctx, allStatOptions); err != nil { @@ -155,7 +155,7 @@ func TestSetStatAtime(t *testing.T) { Atime: linux.NsecToStatxTimestamp(100), } if err := fd.SetStat(ctx, vfs.SetStatOptions{Stat: setStat}); err != nil { - t.Errorf("SetStat atime with mask failed: %v") + t.Errorf("SetStat atime with mask failed: %v", err) } if gotStat, err := fd.Stat(ctx, allStatOptions); err != nil { t.Errorf("Stat got error: %v", err) @@ -205,7 +205,7 @@ func TestSetStat(t *testing.T) { Mask: 0, Atime: linux.NsecToStatxTimestamp(100), }}); err != nil { - t.Errorf("SetStat atime without mask failed: %v") + t.Errorf("SetStat atime without mask failed: %v", err) } // Atime should be unchanged. if gotStat, err := fd.Stat(ctx, allStatOptions); err != nil { @@ -220,7 +220,7 @@ func TestSetStat(t *testing.T) { Atime: linux.NsecToStatxTimestamp(100), } if err := fd.SetStat(ctx, vfs.SetStatOptions{Stat: setStat}); err != nil { - t.Errorf("SetStat atime with mask failed: %v") + t.Errorf("SetStat atime with mask failed: %v", err) } if gotStat, err := fd.Stat(ctx, allStatOptions); err != nil { t.Errorf("Stat got error: %v", err) diff --git a/pkg/tcpip/header/eth_test.go b/pkg/tcpip/header/eth_test.go index 7a0014ad9..14413f2ce 100644 --- a/pkg/tcpip/header/eth_test.go +++ b/pkg/tcpip/header/eth_test.go @@ -88,7 +88,7 @@ func TestEthernetAddressFromMulticastIPv4Address(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { if got := EthernetAddressFromMulticastIPv4Address(test.addr); got != test.expectedLinkAddr { - t.Fatalf("got EthernetAddressFromMulticastIPv4Address(%s) = %s, want = %s", got, test.expectedLinkAddr) + t.Fatalf("got EthernetAddressFromMulticastIPv4Address(%s) = %s, want = %s", test.addr, got, test.expectedLinkAddr) } }) } diff --git a/pkg/tcpip/header/ndp_test.go b/pkg/tcpip/header/ndp_test.go index 1cb9f5dc8..969f8f1e8 100644 --- a/pkg/tcpip/header/ndp_test.go +++ b/pkg/tcpip/header/ndp_test.go @@ -115,7 +115,7 @@ func TestNDPNeighborAdvert(t *testing.T) { // Make sure flags got updated in the backing buffer. if got := b[ndpNAFlagsOffset]; got != 64 { - t.Errorf("got flags byte = %d, want = 64") + t.Errorf("got flags byte = %d, want = 64", got) } } diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go index 7198742b7..b857ce9d0 100644 --- a/pkg/tcpip/link/fdbased/endpoint.go +++ b/pkg/tcpip/link/fdbased/endpoint.go @@ -91,7 +91,7 @@ func (p PacketDispatchMode) String() string { case PacketMMap: return "PacketMMap" default: - return fmt.Sprintf("unknown packet dispatch mode %v", p) + return fmt.Sprintf("unknown packet dispatch mode '%d'", p) } } diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go index 27dc8baf9..bc822e3b1 100644 --- a/pkg/tcpip/stack/ndp_test.go +++ b/pkg/tcpip/stack/ndp_test.go @@ -1959,7 +1959,7 @@ func TestAutoGenAddrDeprecateFromPI(t *testing.T) { // addr2 is deprecated but if explicitly requested, it should be used. fullAddr2 := tcpip.FullAddress{Addr: addr2.Address, NIC: nicID} if got := addrForNewConnectionWithAddr(t, s, fullAddr2); got != addr2.Address { - t.Errorf("got addrForNewConnectionWithAddr(_, _, %+v) = %s, want = %s", got, addr2.Address) + t.Errorf("got addrForNewConnectionWithAddr(_, _, %+v) = %s, want = %s", fullAddr2, got, addr2.Address) } // Another PI w/ 0 preferred lifetime should not result in a deprecation @@ -1972,7 +1972,7 @@ func TestAutoGenAddrDeprecateFromPI(t *testing.T) { } expectPrimaryAddr(addr1) if got := addrForNewConnectionWithAddr(t, s, fullAddr2); got != addr2.Address { - t.Errorf("got addrForNewConnectionWithAddr(_, _, %+v) = %s, want = %s", got, addr2.Address) + t.Errorf("got addrForNewConnectionWithAddr(_, _, %+v) = %s, want = %s", fullAddr2, got, addr2.Address) } // Refresh lifetimes of addr generated from prefix2. @@ -2084,7 +2084,7 @@ func TestAutoGenAddrTimerDeprecation(t *testing.T) { // addr1 is deprecated but if explicitly requested, it should be used. fullAddr1 := tcpip.FullAddress{Addr: addr1.Address, NIC: nicID} if got := addrForNewConnectionWithAddr(t, s, fullAddr1); got != addr1.Address { - t.Errorf("got addrForNewConnectionWithAddr(_, _, %+v) = %s, want = %s", got, addr1.Address) + t.Errorf("got addrForNewConnectionWithAddr(_, _, %+v) = %s, want = %s", fullAddr1, got, addr1.Address) } // Refresh valid lifetime for addr of prefix1, w/ 0 preferred lifetime to make @@ -2097,7 +2097,7 @@ func TestAutoGenAddrTimerDeprecation(t *testing.T) { } expectPrimaryAddr(addr2) if got := addrForNewConnectionWithAddr(t, s, fullAddr1); got != addr1.Address { - t.Errorf("got addrForNewConnectionWithAddr(_, _, %+v) = %s, want = %s", got, addr1.Address) + t.Errorf("got addrForNewConnectionWithAddr(_, _, %+v) = %s, want = %s", fullAddr1, got, addr1.Address) } // Refresh lifetimes for addr of prefix1. @@ -2121,7 +2121,7 @@ func TestAutoGenAddrTimerDeprecation(t *testing.T) { // addr2 should be the primary endpoint now since it is not deprecated. expectPrimaryAddr(addr2) if got := addrForNewConnectionWithAddr(t, s, fullAddr1); got != addr1.Address { - t.Errorf("got addrForNewConnectionWithAddr(_, _, %+v) = %s, want = %s", got, addr1.Address) + t.Errorf("got addrForNewConnectionWithAddr(_, _, %+v) = %s, want = %s", fullAddr1, got, addr1.Address) } // Wait for addr of prefix1 to be invalidated. @@ -2564,7 +2564,7 @@ func TestAutoGenAddrAfterRemoval(t *testing.T) { AddressWithPrefix: addr2, } if err := s.AddProtocolAddressWithOptions(nicID, protoAddr2, stack.FirstPrimaryEndpoint); err != nil { - t.Fatalf("AddProtocolAddressWithOptions(%d, %+v, %d, %s) = %s", nicID, protoAddr2, stack.FirstPrimaryEndpoint, err) + t.Fatalf("AddProtocolAddressWithOptions(%d, %+v, %d) = %s", nicID, protoAddr2, stack.FirstPrimaryEndpoint, err) } // addr2 should be more preferred now since it is at the front of the primary // list. diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index 3f8a2a095..c7634ceb1 100644 --- a/pkg/tcpip/stack/stack_test.go +++ b/pkg/tcpip/stack/stack_test.go @@ -1445,19 +1445,19 @@ func TestOutgoingBroadcastWithEmptyRouteTable(t *testing.T) { protoAddr := tcpip.ProtocolAddress{Protocol: fakeNetNumber, AddressWithPrefix: tcpip.AddressWithPrefix{header.IPv4Any, 0}} if err := s.AddProtocolAddress(1, protoAddr); err != nil { - t.Fatalf("AddProtocolAddress(1, %s) failed: %s", protoAddr, err) + t.Fatalf("AddProtocolAddress(1, %v) failed: %v", protoAddr, err) } r, err := s.FindRoute(1, header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, false /* multicastLoop */) if err != nil { - t.Fatalf("FindRoute(1, %s, %s, %d) failed: %s", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err) + t.Fatalf("FindRoute(1, %v, %v, %d) failed: %v", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err) } if err := verifyRoute(r, stack.Route{LocalAddress: header.IPv4Any, RemoteAddress: header.IPv4Broadcast}); err != nil { - t.Errorf("FindRoute(1, %s, %s, %d) returned unexpected Route: %s)", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err) + t.Errorf("FindRoute(1, %v, %v, %d) returned unexpected Route: %v", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err) } // If the NIC doesn't exist, it won't work. if _, err := s.FindRoute(2, header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, false /* multicastLoop */); err != tcpip.ErrNetworkUnreachable { - t.Fatalf("got FindRoute(2, %s, %s, %d) = %s want = %s", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err, tcpip.ErrNetworkUnreachable) + t.Fatalf("got FindRoute(2, %v, %v, %d) = %v want = %v", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err, tcpip.ErrNetworkUnreachable) } } @@ -1483,12 +1483,12 @@ func TestOutgoingBroadcastWithRouteTable(t *testing.T) { } nic1ProtoAddr := tcpip.ProtocolAddress{fakeNetNumber, nic1Addr} if err := s.AddProtocolAddress(1, nic1ProtoAddr); err != nil { - t.Fatalf("AddProtocolAddress(1, %s) failed: %s", nic1ProtoAddr, err) + t.Fatalf("AddProtocolAddress(1, %v) failed: %v", nic1ProtoAddr, err) } nic2ProtoAddr := tcpip.ProtocolAddress{fakeNetNumber, nic2Addr} if err := s.AddProtocolAddress(2, nic2ProtoAddr); err != nil { - t.Fatalf("AddAddress(2, %s) failed: %s", nic2ProtoAddr, err) + t.Fatalf("AddAddress(2, %v) failed: %v", nic2ProtoAddr, err) } // Set the initial route table. @@ -1503,10 +1503,10 @@ func TestOutgoingBroadcastWithRouteTable(t *testing.T) { // When an interface is given, the route for a broadcast goes through it. r, err := s.FindRoute(1, nic1Addr.Address, header.IPv4Broadcast, fakeNetNumber, false /* multicastLoop */) if err != nil { - t.Fatalf("FindRoute(1, %s, %s, %d) failed: %s", nic1Addr.Address, header.IPv4Broadcast, fakeNetNumber, err) + t.Fatalf("FindRoute(1, %v, %v, %d) failed: %v", nic1Addr.Address, header.IPv4Broadcast, fakeNetNumber, err) } if err := verifyRoute(r, stack.Route{LocalAddress: nic1Addr.Address, RemoteAddress: header.IPv4Broadcast}); err != nil { - t.Errorf("FindRoute(1, %s, %s, %d) returned unexpected Route: %s)", nic1Addr.Address, header.IPv4Broadcast, fakeNetNumber, err) + t.Errorf("FindRoute(1, %v, %v, %d) returned unexpected Route: %v", nic1Addr.Address, header.IPv4Broadcast, fakeNetNumber, err) } // When an interface is not given, it consults the route table. @@ -2399,7 +2399,7 @@ func TestNICContextPreservation(t *testing.T) { t.Fatalf("got nicinfos[%d] = _, %t, want _, true; nicinfos = %+v", id, ok, nicinfos) } if got, want := nicinfo.Context == test.want, true; got != want { - t.Fatal("got nicinfo.Context == ctx = %t, want %t; nicinfo.Context = %p, ctx = %p", got, want, nicinfo.Context, test.want) + t.Fatalf("got nicinfo.Context == ctx = %t, want %t; nicinfo.Context = %p, ctx = %p", got, want, nicinfo.Context, test.want) } }) } @@ -2768,7 +2768,7 @@ func TestNewPEBOnPromotionToPermanent(t *testing.T) { { subnet, err := tcpip.NewSubnet("\x00", "\x00") if err != nil { - t.Fatalf("NewSubnet failed:", err) + t.Fatalf("NewSubnet failed: %v", err) } s.SetRouteTable([]tcpip.Route{{Destination: subnet, Gateway: "\x00", NIC: 1}}) } @@ -2782,11 +2782,11 @@ func TestNewPEBOnPromotionToPermanent(t *testing.T) { // permanentExpired kind. r, err := s.FindRoute(1, "\x01", "\x02", fakeNetNumber, false) if err != nil { - t.Fatal("FindRoute failed:", err) + t.Fatalf("FindRoute failed: %v", err) } defer r.Release() if err := s.RemoveAddress(1, "\x01"); err != nil { - t.Fatalf("RemoveAddress failed:", err) + t.Fatalf("RemoveAddress failed: %v", err) } // @@ -2798,7 +2798,7 @@ func TestNewPEBOnPromotionToPermanent(t *testing.T) { // Add some other address with peb set to // FirstPrimaryEndpoint. if err := s.AddAddressWithOptions(1, fakeNetNumber, "\x03", stack.FirstPrimaryEndpoint); err != nil { - t.Fatal("AddAddressWithOptions failed:", err) + t.Fatalf("AddAddressWithOptions failed: %v", err) } @@ -2806,7 +2806,7 @@ func TestNewPEBOnPromotionToPermanent(t *testing.T) { // make sure the new peb was respected. // (The address should just be promoted now). if err := s.AddAddressWithOptions(1, fakeNetNumber, "\x01", ps); err != nil { - t.Fatal("AddAddressWithOptions failed:", err) + t.Fatalf("AddAddressWithOptions failed: %v", err) } var primaryAddrs []tcpip.Address for _, pa := range s.NICInfo()[1].ProtocolAddresses { @@ -2839,11 +2839,11 @@ func TestNewPEBOnPromotionToPermanent(t *testing.T) { // GetMainNICAddress; else, our original address // should be returned. if err := s.RemoveAddress(1, "\x03"); err != nil { - t.Fatalf("RemoveAddress failed:", err) + t.Fatalf("RemoveAddress failed: %v", err) } addr, err = s.GetMainNICAddress(1, fakeNetNumber) if err != nil { - t.Fatal("s.GetMainNICAddress failed:", err) + t.Fatalf("s.GetMainNICAddress failed: %v", err) } if ps == stack.NeverPrimaryEndpoint { if want := (tcpip.AddressWithPrefix{}); addr != want { diff --git a/pkg/tcpip/tcpip_test.go b/pkg/tcpip/tcpip_test.go index 8c0aacffa..1c8e2bc34 100644 --- a/pkg/tcpip/tcpip_test.go +++ b/pkg/tcpip/tcpip_test.go @@ -218,7 +218,7 @@ func TestAddressWithPrefixSubnet(t *testing.T) { gotSubnet := ap.Subnet() wantSubnet, err := NewSubnet(tt.subnetAddr, tt.subnetMask) if err != nil { - t.Error("NewSubnet(%q, %q) failed: %s", tt.subnetAddr, tt.subnetMask, err) + t.Errorf("NewSubnet(%q, %q) failed: %s", tt.subnetAddr, tt.subnetMask, err) continue } if gotSubnet != wantSubnet { diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index ce3df7478..8a87fa0a8 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -284,7 +284,7 @@ func TestTCPResetSentForACKWhenNotUsingSynCookies(t *testing.T) { // are released instantly on Close. tcpTW := tcpip.TCPTimeWaitTimeoutOption(1 * time.Millisecond) if err := c.Stack().SetTransportProtocolOption(tcp.ProtocolNumber, tcpTW); err != nil { - t.Fatalf("e.stack.SetTransportProtocolOption(%d, %s) = %s", tcp.ProtocolNumber, tcpTW, err) + t.Fatalf("e.stack.SetTransportProtocolOption(%d, %v) = %v", tcp.ProtocolNumber, tcpTW, err) } c.EP.Close() @@ -5609,7 +5609,7 @@ func TestReceiveBufferAutoTuningApplicationLimited(t *testing.T) { return } if w := tcp.WindowSize(); w == 0 || w > uint16(wantRcvWnd) { - t.Errorf("expected a non-zero window: got %d, want <= wantRcvWnd", w, wantRcvWnd) + t.Errorf("expected a non-zero window: got %d, want <= wantRcvWnd", w) } }, )) diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index 0905726c1..b3fe557ca 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -607,7 +607,7 @@ func testReadInternal(c *testContext, flow testFlow, packetShouldBeDropped, expe // Check the peer address. h := flow.header4Tuple(incoming) if addr.Addr != h.srcAddr.Addr { - c.t.Fatalf("unexpected remote address: got %s, want %s", addr.Addr, h.srcAddr) + c.t.Fatalf("unexpected remote address: got %s, want %v", addr.Addr, h.srcAddr) } // Check the payload. diff --git a/runsc/container/test_app/test_app.go b/runsc/container/test_app/test_app.go index 01c47c79f..5f1c4b7d6 100644 --- a/runsc/container/test_app/test_app.go +++ b/runsc/container/test_app/test_app.go @@ -96,7 +96,7 @@ func (c *uds) Execute(ctx context.Context, f *flag.FlagSet, args ...interface{}) listener, err := net.Listen("unix", c.socketPath) if err != nil { - log.Fatal("error listening on socket %q:", c.socketPath, err) + log.Fatalf("error listening on socket %q: %v", c.socketPath, err) } go server(listener, outputFile) diff --git a/test/root/cgroup_test.go b/test/root/cgroup_test.go index 4038661cb..679342def 100644 --- a/test/root/cgroup_test.go +++ b/test/root/cgroup_test.go @@ -53,7 +53,7 @@ func verifyPid(pid int, path string) error { if scanner.Err() != nil { return scanner.Err() } - return fmt.Errorf("got: %s, want: %d", gots, pid) + return fmt.Errorf("got: %v, want: %d", gots, pid) } // TestCgroup sets cgroup options and checks that cgroup was properly configured. @@ -106,7 +106,7 @@ func TestMemCGroup(t *testing.T) { time.Sleep(100 * time.Millisecond) } - t.Fatalf("%vMB is less than %vMB: %v", memUsage>>20, allocMemSize>>20) + t.Fatalf("%vMB is less than %vMB", memUsage>>20, allocMemSize>>20) } // TestCgroup sets cgroup options and checks that cgroup was properly configured. diff --git a/test/runtimes/blacklist_test.go b/test/runtimes/blacklist_test.go index 52f49b984..0ff69ab18 100644 --- a/test/runtimes/blacklist_test.go +++ b/test/runtimes/blacklist_test.go @@ -32,6 +32,6 @@ func TestBlacklists(t *testing.T) { t.Fatalf("error parsing blacklist: %v", err) } if *blacklistFile != "" && len(bl) == 0 { - t.Errorf("got empty blacklist for file %q", blacklistFile) + t.Errorf("got empty blacklist for file %q", *blacklistFile) } } diff --git a/test/runtimes/runner.go b/test/runtimes/runner.go index ddb890dbc..3c98f4570 100644 --- a/test/runtimes/runner.go +++ b/test/runtimes/runner.go @@ -114,7 +114,7 @@ func getTests(d dockerutil.Docker, blacklist map[string]struct{}) ([]testing.Int F: func(t *testing.T) { // Is the test blacklisted? if _, ok := blacklist[tc]; ok { - t.Skip("SKIP: blacklisted test %q", tc) + t.Skipf("SKIP: blacklisted test %q", tc) } var ( diff --git a/tools/nogo.json b/tools/nogo.json index cc05ba027..f69999e50 100644 --- a/tools/nogo.json +++ b/tools/nogo.json @@ -27,37 +27,6 @@ "/external/io_opencensus_go/tag/map_codec.go": "allowed: false positive" } }, - "printf": { - "exclude_files": { - ".*_abi_autogen_test.go": "fix: Sprintf format has insufficient args", - "/pkg/segment/test/segment_test.go": "fix: Errorf format %d arg seg.Start is a func value, not called", - "/pkg/tcpip/tcpip_test.go": "fix: Error call has possible formatting directive %q", - "/pkg/tcpip/header/eth_test.go": "fix: Fatalf format %s reads arg #3, but call has 2 args", - "/pkg/tcpip/header/ndp_test.go": "fix: Errorf format %d reads arg #1, but call has 0 args", - "/pkg/eventchannel/event_test.go": "fix: Fatal call has possible formatting directive %v", - "/pkg/tcpip/stack/ndp.go": "fix: Fatalf format %s has arg protocolAddr of wrong type gvisor.dev/gvisor/pkg/tcpip.ProtocolAddress", - "/pkg/sentry/fs/fdpipe/pipe_test.go": "fix: Errorf format %s has arg flags of wrong type gvisor.dev/gvisor/pkg/sentry/fs.FileFlags", - "/pkg/sentry/fs/fdpipe/pipe_test.go": "fix: Errorf format %d arg f.FD is a func value, not called", - "/pkg/tcpip/link/fdbased/endpoint.go": "fix: Sprintf format %v with arg p causes recursive String method call", - "/pkg/tcpip/transport/udp/udp_test.go": "fix: Fatalf format %s has arg h.srcAddr of wrong type gvisor.dev/gvisor/pkg/tcpip.FullAddress", - "/pkg/tcpip/transport/tcp/tcp_test.go": "fix: Fatalf format %s has arg tcpTW of wrong type gvisor.dev/gvisor/pkg/tcpip.TCPTimeWaitTimeoutOption", - "/pkg/tcpip/transport/tcp/tcp_test.go": "fix: Errorf call needs 1 arg but has 2 args", - "/pkg/tcpip/stack/ndp_test.go": "fix: Errorf format %s reads arg #3, but call has 2 args", - "/pkg/tcpip/stack/ndp_test.go": "fix: Fatalf format %s reads arg #5, but call has 4 args", - "/pkg/tcpip/stack/stack_test.go": "fix: Fatalf format %s has arg protoAddr of wrong type gvisor.dev/gvisor/pkg/tcpip.ProtocolAddress", - "/pkg/tcpip/stack/stack_test.go": "fix: Fatalf format %s has arg nic1ProtoAddr of wrong type gvisor.dev/gvisor/pkg/tcpip.ProtocolAddress", - "/pkg/tcpip/stack/stack_test.go": "fix: Fatalf format %s has arg nic2ProtoAddr of wrong type gvisor.dev/gvisor/pkg/tcpip.ProtocolAddress", - "/pkg/tcpip/stack/stack_test.go": "fix: Fatal call has possible formatting directive %t", - "/pkg/tcpip/stack/stack_test.go": "fix: Fatalf call has arguments but no formatting directives", - "/pkg/tcpip/link/fdbased/endpoint.go": "fix: Sprintf format %v with arg p causes recursive String method call", - "/pkg/sentry/fsimpl/tmpfs/stat_test.go": "fix: Errorf format %v reads arg #1, but call has 0 args", - "/runsc/container/test_app/test_app.go": "fix: Fatal call has possible formatting directive %q", - "/test/root/cgroup_test.go": "fix: Errorf format %s has arg gots of wrong type []int", - "/test/root/cgroup_test.go": "fix: Fatalf format %v reads arg #3, but call has 2 args", - "/test/runtimes/runner.go": "fix: Skip call has possible formatting directive %q", - "/test/runtimes/blacklist_test.go": "fix: Errorf format %q has arg blacklistFile of wrong type *string" - } - }, "structtag": { "exclude_files": { "/external/": "allowed: may use arbitrary tags" -- cgit v1.2.3 From f888b9ce83c202bb77c1e29c3ee60cc485906536 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Wed, 25 Mar 2020 16:57:37 -0700 Subject: Fix unused result errors. This fixes a bug in the proc net directory. Updates #2243 --- pkg/sentry/fsimpl/proc/task_net.go | 4 ++-- tools/nogo.json | 6 ------ 2 files changed, 2 insertions(+), 8 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/proc/task_net.go b/pkg/sentry/fsimpl/proc/task_net.go index 6b2a77328..6595fcee6 100644 --- a/pkg/sentry/fsimpl/proc/task_net.go +++ b/pkg/sentry/fsimpl/proc/task_net.go @@ -688,9 +688,9 @@ func (d *netSnmpData) Generate(ctx context.Context, buf *bytes.Buffer) error { if line.prefix == "Tcp" { tcp := stat.(*inet.StatSNMPTCP) // "Tcp" needs special processing because MaxConn is signed. RFC 2012. - fmt.Sprintf("%s: %s %d %s\n", line.prefix, sprintSlice(tcp[:3]), int64(tcp[3]), sprintSlice(tcp[4:])) + fmt.Fprintf(buf, "%s: %s %d %s\n", line.prefix, sprintSlice(tcp[:3]), int64(tcp[3]), sprintSlice(tcp[4:])) } else { - fmt.Sprintf("%s: %s\n", line.prefix, sprintSlice(toSlice(stat))) + fmt.Fprintf(buf, "%s: %s\n", line.prefix, sprintSlice(toSlice(stat))) } } return nil diff --git a/tools/nogo.json b/tools/nogo.json index f69999e50..09bda9212 100644 --- a/tools/nogo.json +++ b/tools/nogo.json @@ -44,11 +44,5 @@ "/pkg/sentry/platform/safecopy/safecopy_unsafe.go": "allowed: special case", "/pkg/sentry/vfs/mount_unsafe.go": "allowed: special case" } - }, - "unusedresult": { - "exclude_files": { - "/pkg/sentry/fsimpl/proc/task_net.go": "fix: result of fmt.Sprintf call not used", - "/pkg/sentry/fsimpl/proc/tasks_net.go": "fix: result of fmt.Sprintf call not used" - } } } -- cgit v1.2.3 From b30130567d81157e39b692e0116f86015f0bcc71 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Wed, 8 Apr 2020 13:33:44 -0700 Subject: Enable SubprocessExited and SubprocessZombie for gVisor Updates #164 PiperOrigin-RevId: 305544029 --- pkg/sentry/fs/proc/task.go | 28 ++++++++++++++++-- pkg/sentry/fsimpl/proc/task.go | 16 ----------- pkg/sentry/fsimpl/proc/task_files.go | 56 ++++++++++++++++++++++++++++++++++-- test/syscalls/linux/proc.cc | 31 ++++++++------------ 4 files changed, 90 insertions(+), 41 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go index d6c5dd2c1..4d42eac83 100644 --- a/pkg/sentry/fs/proc/task.go +++ b/pkg/sentry/fs/proc/task.go @@ -57,6 +57,16 @@ func getTaskMM(t *kernel.Task) (*mm.MemoryManager, error) { return m, nil } +func checkTaskState(t *kernel.Task) error { + switch t.ExitState() { + case kernel.TaskExitZombie: + return syserror.EACCES + case kernel.TaskExitDead: + return syserror.ESRCH + } + return nil +} + // taskDir represents a task-level directory. // // +stateify savable @@ -254,11 +264,12 @@ func newExe(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { } func (e *exe) executable() (file fsbridge.File, err error) { + if err := checkTaskState(e.t); err != nil { + return nil, err + } e.t.WithMuLocked(func(t *kernel.Task) { mm := t.MemoryManager() if mm == nil { - // TODO(b/34851096): Check shouldn't allow Readlink once the - // Task is zombied. err = syserror.EACCES return } @@ -268,7 +279,7 @@ func (e *exe) executable() (file fsbridge.File, err error) { // (with locks held). file = mm.Executable() if file == nil { - err = syserror.ENOENT + err = syserror.ESRCH } }) return @@ -313,11 +324,22 @@ func newNamespaceSymlink(t *kernel.Task, msrc *fs.MountSource, name string) *fs. return newProcInode(t, n, msrc, fs.Symlink, t) } +// Readlink reads the symlink value. +func (n *namespaceSymlink) Readlink(ctx context.Context, inode *fs.Inode) (string, error) { + if err := checkTaskState(n.t); err != nil { + return "", err + } + return n.Symlink.Readlink(ctx, inode) +} + // Getlink implements fs.InodeOperations.Getlink. func (n *namespaceSymlink) Getlink(ctx context.Context, inode *fs.Inode) (*fs.Dirent, error) { if !kernel.ContextCanTrace(ctx, n.t, false) { return nil, syserror.EACCES } + if err := checkTaskState(n.t); err != nil { + return nil, err + } // Create a new regular file to fake the namespace file. iops := fsutil.NewNoReadWriteFileInode(ctx, fs.RootOwner, fs.FilePermsFromMode(0777), linux.PROC_SUPER_MAGIC) diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index aee2a4392..888afc0fd 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -214,22 +214,6 @@ func newIO(t *kernel.Task, isThreadGroup bool) *ioData { return &ioData{ioUsage: t} } -func newNamespaceSymlink(task *kernel.Task, ino uint64, ns string) *kernfs.Dentry { - // Namespace symlinks should contain the namespace name and the inode number - // for the namespace instance, so for example user:[123456]. We currently fake - // the inode number by sticking the symlink inode in its place. - target := fmt.Sprintf("%s:[%d]", ns, ino) - - inode := &kernfs.StaticSymlink{} - // Note: credentials are overridden by taskOwnedInode. - inode.Init(task.Credentials(), ino, target) - - taskInode := &taskOwnedInode{Inode: inode, owner: task} - d := &kernfs.Dentry{} - d.Init(taskInode) - return d -} - // newCgroupData creates inode that shows cgroup information. // From man 7 cgroups: "For each cgroup hierarchy of which the process is a // member, there is one entry containing three colon-separated fields: diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index 88ea6a6d8..2c6f8bdfc 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -64,6 +64,16 @@ func getMMIncRef(task *kernel.Task) (*mm.MemoryManager, error) { return m, nil } +func checkTaskState(t *kernel.Task) error { + switch t.ExitState() { + case kernel.TaskExitZombie: + return syserror.EACCES + case kernel.TaskExitDead: + return syserror.ESRCH + } + return nil +} + type bufferWriter struct { buf *bytes.Buffer } @@ -628,11 +638,13 @@ func (s *exeSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, er } func (s *exeSymlink) executable() (file fsbridge.File, err error) { + if err := checkTaskState(s.task); err != nil { + return nil, err + } + s.task.WithMuLocked(func(t *kernel.Task) { mm := t.MemoryManager() if mm == nil { - // TODO(b/34851096): Check shouldn't allow Readlink once the - // Task is zombied. err = syserror.EACCES return } @@ -642,7 +654,7 @@ func (s *exeSymlink) executable() (file fsbridge.File, err error) { // (with locks held). file = mm.Executable() if file == nil { - err = syserror.ENOENT + err = syserror.ESRCH } }) return @@ -709,3 +721,41 @@ func (i *mountsData) Generate(ctx context.Context, buf *bytes.Buffer) error { i.task.Kernel().VFS().GenerateProcMounts(ctx, rootDir, buf) return nil } + +type namespaceSymlink struct { + kernfs.StaticSymlink + + task *kernel.Task +} + +func newNamespaceSymlink(task *kernel.Task, ino uint64, ns string) *kernfs.Dentry { + // Namespace symlinks should contain the namespace name and the inode number + // for the namespace instance, so for example user:[123456]. We currently fake + // the inode number by sticking the symlink inode in its place. + target := fmt.Sprintf("%s:[%d]", ns, ino) + + inode := &namespaceSymlink{task: task} + // Note: credentials are overridden by taskOwnedInode. + inode.Init(task.Credentials(), ino, target) + + taskInode := &taskOwnedInode{Inode: inode, owner: task} + d := &kernfs.Dentry{} + d.Init(taskInode) + return d +} + +// Readlink implements Inode. +func (s *namespaceSymlink) Readlink(ctx context.Context) (string, error) { + if err := checkTaskState(s.task); err != nil { + return "", err + } + return s.StaticSymlink.Readlink(ctx) +} + +// Getlink implements Inode.Getlink. +func (s *namespaceSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) { + if err := checkTaskState(s.task); err != nil { + return vfs.VirtualDentry{}, "", err + } + return s.StaticSymlink.Getlink(ctx) +} diff --git a/test/syscalls/linux/proc.cc b/test/syscalls/linux/proc.cc index 5a70f6c3b..da98e1f66 100644 --- a/test/syscalls/linux/proc.cc +++ b/test/syscalls/linux/proc.cc @@ -1326,8 +1326,6 @@ TEST(ProcPidSymlink, SubprocessRunning) { SyscallSucceedsWithValue(sizeof(buf))); } -// FIXME(gvisor.dev/issue/164): Inconsistent behavior between gVisor and linux -// on proc files. TEST(ProcPidSymlink, SubprocessZombied) { ASSERT_NO_ERRNO(SetCapability(CAP_DAC_OVERRIDE, false)); ASSERT_NO_ERRNO(SetCapability(CAP_DAC_READ_SEARCH, false)); @@ -1337,7 +1335,7 @@ TEST(ProcPidSymlink, SubprocessZombied) { int want = EACCES; if (!IsRunningOnGvisor()) { auto version = ASSERT_NO_ERRNO_AND_VALUE(GetKernelVersion()); - if (version.major == 4 && version.minor > 3) { + if (version.major > 4 || (version.major == 4 && version.minor > 3)) { want = ENOENT; } } @@ -1350,30 +1348,25 @@ TEST(ProcPidSymlink, SubprocessZombied) { SyscallFailsWithErrno(want)); } - // FIXME(gvisor.dev/issue/164): Inconsistent behavior between gVisor and linux - // on proc files. + // FIXME(gvisor.dev/issue/164): Inconsistent behavior between linux on proc + // files. // // ~4.3: Syscall fails with EACCES. - // 4.17 & gVisor: Syscall succeeds and returns 1. + // 4.17: Syscall succeeds and returns 1. // - // EXPECT_THAT(ReadlinkWhileZombied("ns/pid", buf, sizeof(buf)), - // SyscallFailsWithErrno(EACCES)); + if (!IsRunningOnGvisor()) { + return; + } - // FIXME(gvisor.dev/issue/164): Inconsistent behavior between gVisor and linux - // on proc files. - // - // ~4.3: Syscall fails with EACCES. - // 4.17 & gVisor: Syscall succeeds and returns 1. - // - // EXPECT_THAT(ReadlinkWhileZombied("ns/user", buf, sizeof(buf)), - // SyscallFailsWithErrno(EACCES)); + EXPECT_THAT(ReadlinkWhileZombied("ns/pid", buf, sizeof(buf)), + SyscallFailsWithErrno(want)); + + EXPECT_THAT(ReadlinkWhileZombied("ns/user", buf, sizeof(buf)), + SyscallFailsWithErrno(want)); } // Test whether /proc/PID/ symlinks can be read for an exited process. TEST(ProcPidSymlink, SubprocessExited) { - // FIXME(gvisor.dev/issue/164): These all succeed on gVisor. - SKIP_IF(IsRunningOnGvisor()); - char buf[1]; EXPECT_THAT(ReadlinkWhileExited("exe", buf, sizeof(buf)), -- cgit v1.2.3 From d8c0c1d9d5f30f4cfe8e0adedd264b10aff793f7 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Wed, 8 Apr 2020 13:39:02 -0700 Subject: Do not hold FileDescription references in VFS2 procfs inodes. FileDescription references are side-effectual; for example, holding a reference on the write end of a pipe prevents reads from the read end from returning EOF. This change is consistent with Linux, but not VFS1; while VFS1 also has this bug, it's less visible there since VFS1 procfs disables caching. Updates #1195 PiperOrigin-RevId: 305545099 --- pkg/sentry/fsimpl/proc/task_fds.go | 125 ++++++++++++++++++++----------------- 1 file changed, 67 insertions(+), 58 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/proc/task_fds.go b/pkg/sentry/fsimpl/proc/task_fds.go index 9c8656b28..046265eca 100644 --- a/pkg/sentry/fsimpl/proc/task_fds.go +++ b/pkg/sentry/fsimpl/proc/task_fds.go @@ -30,34 +30,35 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) -type fdDir struct { - inoGen InoGenerator - task *kernel.Task - - // When produceSymlinks is set, dirents produces for the FDs are reported - // as symlink. Otherwise, they are reported as regular files. - produceSymlink bool -} - -func (i *fdDir) lookup(name string) (*vfs.FileDescription, kernel.FDFlags, error) { - fd, err := strconv.ParseUint(name, 10, 64) - if err != nil { - return nil, kernel.FDFlags{}, syserror.ENOENT - } - +func getTaskFD(t *kernel.Task, fd int32) (*vfs.FileDescription, kernel.FDFlags) { var ( file *vfs.FileDescription flags kernel.FDFlags ) - i.task.WithMuLocked(func(t *kernel.Task) { - if fdTable := t.FDTable(); fdTable != nil { - file, flags = fdTable.GetVFS2(int32(fd)) + t.WithMuLocked(func(t *kernel.Task) { + if fdt := t.FDTable(); fdt != nil { + file, flags = fdt.GetVFS2(fd) } }) + return file, flags +} + +func taskFDExists(t *kernel.Task, fd int32) bool { + file, _ := getTaskFD(t, fd) if file == nil { - return nil, kernel.FDFlags{}, syserror.ENOENT + return false } - return file, flags, nil + file.DecRef() + return true +} + +type fdDir struct { + inoGen InoGenerator + task *kernel.Task + + // When produceSymlinks is set, dirents produces for the FDs are reported + // as symlink. Otherwise, they are reported as regular files. + produceSymlink bool } // IterDirents implements kernfs.inodeDynamicLookup. @@ -128,11 +129,15 @@ func newFDDirInode(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry { // Lookup implements kernfs.inodeDynamicLookup. func (i *fdDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { - file, _, err := i.lookup(name) + fdInt, err := strconv.ParseInt(name, 10, 32) if err != nil { - return nil, err + return nil, syserror.ENOENT + } + fd := int32(fdInt) + if !taskFDExists(i.task, fd) { + return nil, syserror.ENOENT } - taskDentry := newFDSymlink(i.task.Credentials(), file, i.inoGen.NextIno()) + taskDentry := newFDSymlink(i.task, fd, i.inoGen.NextIno()) return taskDentry.VFSDentry(), nil } @@ -169,19 +174,22 @@ func (i *fdDirInode) CheckPermissions(ctx context.Context, creds *auth.Credentia // // +stateify savable type fdSymlink struct { - refs.AtomicRefCount kernfs.InodeAttrs + kernfs.InodeNoopRefCount kernfs.InodeSymlink - file *vfs.FileDescription + task *kernel.Task + fd int32 } var _ kernfs.Inode = (*fdSymlink)(nil) -func newFDSymlink(creds *auth.Credentials, file *vfs.FileDescription, ino uint64) *kernfs.Dentry { - file.IncRef() - inode := &fdSymlink{file: file} - inode.Init(creds, ino, linux.ModeSymlink|0777) +func newFDSymlink(task *kernel.Task, fd int32, ino uint64) *kernfs.Dentry { + inode := &fdSymlink{ + task: task, + fd: fd, + } + inode.Init(task.Credentials(), ino, linux.ModeSymlink|0777) d := &kernfs.Dentry{} d.Init(inode) @@ -189,29 +197,27 @@ func newFDSymlink(creds *auth.Credentials, file *vfs.FileDescription, ino uint64 } func (s *fdSymlink) Readlink(ctx context.Context) (string, error) { + file, _ := getTaskFD(s.task, s.fd) + if file == nil { + return "", syserror.ENOENT + } + defer file.DecRef() root := vfs.RootFromContext(ctx) defer root.DecRef() - - vfsObj := s.file.VirtualDentry().Mount().Filesystem().VirtualFilesystem() - return vfsObj.PathnameWithDeleted(ctx, root, s.file.VirtualDentry()) + return s.task.Kernel().VFS().PathnameWithDeleted(ctx, root, file.VirtualDentry()) } func (s *fdSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) { - vd := s.file.VirtualDentry() + file, _ := getTaskFD(s.task, s.fd) + if file == nil { + return vfs.VirtualDentry{}, "", syserror.ENOENT + } + defer file.DecRef() + vd := file.VirtualDentry() vd.IncRef() return vd, "", nil } -func (s *fdSymlink) DecRef() { - s.AtomicRefCount.DecRefWithDestructor(func() { - s.Destroy() - }) -} - -func (s *fdSymlink) Destroy() { - s.file.DecRef() -} - // fdInfoDirInode represents the inode for /proc/[pid]/fdinfo directory. // // +stateify savable @@ -244,12 +250,18 @@ func newFDInfoDirInode(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry { // Lookup implements kernfs.inodeDynamicLookup. func (i *fdInfoDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { - file, flags, err := i.lookup(name) + fdInt, err := strconv.ParseInt(name, 10, 32) if err != nil { - return nil, err + return nil, syserror.ENOENT + } + fd := int32(fdInt) + if !taskFDExists(i.task, fd) { + return nil, syserror.ENOENT + } + data := &fdInfoData{ + task: i.task, + fd: fd, } - - data := &fdInfoData{file: file, flags: flags} dentry := newTaskOwnedFile(i.task, i.inoGen.NextIno(), 0444, data) return dentry.VFSDentry(), nil } @@ -268,26 +280,23 @@ type fdInfoData struct { kernfs.DynamicBytesFile refs.AtomicRefCount - file *vfs.FileDescription - flags kernel.FDFlags + task *kernel.Task + fd int32 } var _ dynamicInode = (*fdInfoData)(nil) -func (d *fdInfoData) DecRef() { - d.AtomicRefCount.DecRefWithDestructor(d.destroy) -} - -func (d *fdInfoData) destroy() { - d.file.DecRef() -} - // Generate implements vfs.DynamicBytesSource.Generate. func (d *fdInfoData) Generate(ctx context.Context, buf *bytes.Buffer) error { + file, descriptorFlags := getTaskFD(d.task, d.fd) + if file == nil { + return syserror.ENOENT + } + defer file.DecRef() // TODO(b/121266871): Include pos, locks, and other data. For now we only // have flags. // See https://www.kernel.org/doc/Documentation/filesystems/proc.txt - flags := uint(d.file.StatusFlags()) | d.flags.ToLinuxFileFlags() + flags := uint(file.StatusFlags()) | descriptorFlags.ToLinuxFileFlags() fmt.Fprintf(buf, "flags:\t0%o\n", flags) return nil } -- cgit v1.2.3 From 357f136e42de81b033b65b7f39a4a555275a17e3 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Wed, 8 Apr 2020 14:38:09 -0700 Subject: Handle utimes correctly for shared gofer filesystems. Determine system time from within the sentry rather than relying on the remote filesystem to prevent inconsistencies. Resolve related TODOs; the time discrepancies in question don't exist anymore. PiperOrigin-RevId: 305557099 --- pkg/sentry/fs/gofer/util.go | 16 ++++++++++++++-- test/syscalls/linux/utimes.cc | 18 +----------------- 2 files changed, 15 insertions(+), 19 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/gofer/util.go b/pkg/sentry/fs/gofer/util.go index 2d8d3a2ea..47a6c69bf 100644 --- a/pkg/sentry/fs/gofer/util.go +++ b/pkg/sentry/fs/gofer/util.go @@ -20,17 +20,29 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/sentry/fs" + ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" ) func utimes(ctx context.Context, file contextFile, ts fs.TimeSpec) error { if ts.ATimeOmit && ts.MTimeOmit { return nil } + + // Replace requests to use the "system time" with the current time to + // ensure that timestamps remain consistent with the remote + // filesystem. + now := ktime.NowFromContext(ctx) + if ts.ATimeSetSystemTime { + ts.ATime = now + } + if ts.MTimeSetSystemTime { + ts.MTime = now + } mask := p9.SetAttrMask{ ATime: !ts.ATimeOmit, - ATimeNotSystemTime: !ts.ATimeSetSystemTime, + ATimeNotSystemTime: true, MTime: !ts.MTimeOmit, - MTimeNotSystemTime: !ts.MTimeSetSystemTime, + MTimeNotSystemTime: true, } as, ans := ts.ATime.Unix() ms, mns := ts.MTime.Unix() diff --git a/test/syscalls/linux/utimes.cc b/test/syscalls/linux/utimes.cc index 3a927a430..22e6d1a85 100644 --- a/test/syscalls/linux/utimes.cc +++ b/test/syscalls/linux/utimes.cc @@ -34,17 +34,10 @@ namespace testing { namespace { -// TODO(b/36516566): utimes(nullptr) does not pick the "now" time in the -// application's time domain, so when asserting that times are within a window, -// we expand the window to allow for differences between the time domains. -constexpr absl::Duration kClockSlack = absl::Milliseconds(100); - // TimeBoxed runs fn, setting before and after to (coarse realtime) times // guaranteed* to come before and after fn started and completed, respectively. // // fn may be called more than once if the clock is adjusted. -// -// * See the comment on kClockSlack. gVisor breaks this guarantee. void TimeBoxed(absl::Time* before, absl::Time* after, std::function const& fn) { do { @@ -69,12 +62,6 @@ void TimeBoxed(absl::Time* before, absl::Time* after, // which could lead to test failures, but that is very unlikely to happen. continue; } - - if (IsRunningOnGvisor()) { - // See comment on kClockSlack. - *before -= kClockSlack; - *after += kClockSlack; - } } while (*after < *before); } @@ -235,10 +222,7 @@ void TestUtimensat(int dirFd, std::string const& path) { EXPECT_GE(mtime3, before); EXPECT_LE(mtime3, after); - if (!IsRunningOnGvisor()) { - // FIXME(b/36516566): Gofers set atime and mtime to different "now" times. - EXPECT_EQ(atime3, mtime3); - } + EXPECT_EQ(atime3, mtime3); } TEST(UtimensatTest, OnAbsPath) { -- cgit v1.2.3 From 981a587476e11e49cf49edb31705d8727b0db556 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Wed, 8 Apr 2020 17:32:57 -0700 Subject: Remove InodeOperations FIXMEs that will be obsoleted by VFS2. PiperOrigin-RevId: 305588941 --- pkg/sentry/fs/gofer/inode.go | 3 --- pkg/sentry/fs/host/inode.go | 3 --- pkg/sentry/fs/inode.go | 2 -- 3 files changed, 8 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/gofer/inode.go b/pkg/sentry/fs/gofer/inode.go index 1c934981b..811e8ea30 100644 --- a/pkg/sentry/fs/gofer/inode.go +++ b/pkg/sentry/fs/gofer/inode.go @@ -710,13 +710,10 @@ func init() { } // AddLink implements InodeOperations.AddLink, but is currently a noop. -// FIXME(b/63117438): Remove this from InodeOperations altogether. func (*inodeOperations) AddLink() {} // DropLink implements InodeOperations.DropLink, but is currently a noop. -// FIXME(b/63117438): Remove this from InodeOperations altogether. func (*inodeOperations) DropLink() {} // NotifyStatusChange implements fs.InodeOperations.NotifyStatusChange. -// FIXME(b/63117438): Remove this from InodeOperations altogether. func (i *inodeOperations) NotifyStatusChange(ctx context.Context) {} diff --git a/pkg/sentry/fs/host/inode.go b/pkg/sentry/fs/host/inode.go index 1da3c0a17..62f1246aa 100644 --- a/pkg/sentry/fs/host/inode.go +++ b/pkg/sentry/fs/host/inode.go @@ -397,15 +397,12 @@ func (i *inodeOperations) StatFS(context.Context) (fs.Info, error) { } // AddLink implements fs.InodeOperations.AddLink. -// FIXME(b/63117438): Remove this from InodeOperations altogether. func (i *inodeOperations) AddLink() {} // DropLink implements fs.InodeOperations.DropLink. -// FIXME(b/63117438): Remove this from InodeOperations altogether. func (i *inodeOperations) DropLink() {} // NotifyStatusChange implements fs.InodeOperations.NotifyStatusChange. -// FIXME(b/63117438): Remove this from InodeOperations altogether. func (i *inodeOperations) NotifyStatusChange(ctx context.Context) {} // readdirAll returns all of the directory entries in i. diff --git a/pkg/sentry/fs/inode.go b/pkg/sentry/fs/inode.go index 55fb71c16..73f89abcc 100644 --- a/pkg/sentry/fs/inode.go +++ b/pkg/sentry/fs/inode.go @@ -397,8 +397,6 @@ func (i *Inode) Getlink(ctx context.Context) (*Dirent, error) { // AddLink calls i.InodeOperations.AddLink. func (i *Inode) AddLink() { if i.overlay != nil { - // FIXME(b/63117438): Remove this from InodeOperations altogether. - // // This interface is only used by ramfs to update metadata of // children. These filesystems should _never_ have overlay // Inodes cached as children. So explicitly disallow this -- cgit v1.2.3 From 6dd5a1f3fe55daa8510b1ee5e3a59219aad92af6 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Wed, 8 Apr 2020 17:56:55 -0700 Subject: Clean up TODOs PiperOrigin-RevId: 305592245 --- pkg/sentry/fs/tmpfs/fs.go | 3 --- pkg/sentry/fsimpl/kernfs/filesystem.go | 2 +- pkg/sentry/kernel/ptrace.go | 1 - pkg/sentry/vfs/filesystem.go | 2 +- pkg/sentry/vfs/mount.go | 12 ++++++------ pkg/sentry/vfs/mount_test.go | 2 +- runsc/cmd/gofer.go | 5 ++--- test/syscalls/linux/epoll.cc | 4 ---- test/syscalls/linux/file_base.h | 1 + test/syscalls/linux/pwrite64.cc | 9 +-------- test/syscalls/linux/tuntap.cc | 7 ++++--- test/syscalls/linux/write.cc | 10 ++-------- tools/go_generics/defs.bzl | 1 - 13 files changed, 19 insertions(+), 40 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/tmpfs/fs.go b/pkg/sentry/fs/tmpfs/fs.go index d5be56c3f..bc117ca6a 100644 --- a/pkg/sentry/fs/tmpfs/fs.go +++ b/pkg/sentry/fs/tmpfs/fs.go @@ -44,9 +44,6 @@ const ( // lookup. cacheRevalidate = "revalidate" - // TODO(edahlgren/mpratt): support a tmpfs size limit. - // size = "size" - // Permissions that exceed modeMask will be rejected. modeMask = 01777 diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 16a3c18ae..4433071aa 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -682,7 +682,7 @@ func (fs *Filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linu if err != nil { return linux.Statfs{}, err } - // TODO: actually implement statfs + // TODO(gvisor.dev/issue/1193): actually implement statfs. return linux.Statfs{}, syserror.ENOSYS } diff --git a/pkg/sentry/kernel/ptrace.go b/pkg/sentry/kernel/ptrace.go index 35ad97d5d..e23e796ef 100644 --- a/pkg/sentry/kernel/ptrace.go +++ b/pkg/sentry/kernel/ptrace.go @@ -184,7 +184,6 @@ func (t *Task) CanTrace(target *Task, attach bool) bool { if targetCreds.PermittedCaps&^callerCreds.PermittedCaps != 0 { return false } - // TODO: Yama LSM return true } diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index cd34782ff..bef1bd312 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -497,7 +497,7 @@ type FilesystemImpl interface { // Preconditions: vd.Mount().Filesystem().Impl() == this FilesystemImpl. PrependPath(ctx context.Context, vfsroot, vd VirtualDentry, b *fspath.Builder) error - // TODO: inotify_add_watch() + // TODO(gvisor.dev/issue/1479): inotify_add_watch() } // PrependPathAtVFSRootError is returned by implementations of diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index 1b8ecc415..f06946103 100644 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go @@ -233,9 +233,9 @@ func (vfs *VirtualFilesystem) MountAt(ctx context.Context, creds *auth.Credentia } vd.dentry.mu.Lock() } - // TODO: Linux requires that either both the mount point and the mount root - // are directories, or neither are, and returns ENOTDIR if this is not the - // case. + // TODO(gvisor.dev/issue/1035): Linux requires that either both the mount + // point and the mount root are directories, or neither are, and returns + // ENOTDIR if this is not the case. mntns := vd.mount.ns mnt := newMount(vfs, fs, root, mntns, opts) vfs.mounts.seq.BeginWrite() @@ -274,9 +274,9 @@ func (vfs *VirtualFilesystem) UmountAt(ctx context.Context, creds *auth.Credenti } } - // TODO(jamieliu): Linux special-cases umount of the caller's root, which - // we don't implement yet (we'll just fail it since the caller holds a - // reference on it). + // TODO(gvisor.dev/issue/1035): Linux special-cases umount of the caller's + // root, which we don't implement yet (we'll just fail it since the caller + // holds a reference on it). vfs.mounts.seq.BeginWrite() if opts.Flags&linux.MNT_DETACH == 0 { diff --git a/pkg/sentry/vfs/mount_test.go b/pkg/sentry/vfs/mount_test.go index 3b933468d..3335e4057 100644 --- a/pkg/sentry/vfs/mount_test.go +++ b/pkg/sentry/vfs/mount_test.go @@ -55,7 +55,7 @@ func TestMountTableInsertLookup(t *testing.T) { } } -// TODO: concurrent lookup/insertion/removal +// TODO(gvisor.dev/issue/1035): concurrent lookup/insertion/removal. // must be powers of 2 var benchNumMounts = []int{1 << 2, 1 << 5, 1 << 8} diff --git a/runsc/cmd/gofer.go b/runsc/cmd/gofer.go index 02e5af3d3..28f0d54b9 100644 --- a/runsc/cmd/gofer.go +++ b/runsc/cmd/gofer.go @@ -272,9 +272,8 @@ func setupRootFS(spec *specs.Spec, conf *boot.Config) error { root := spec.Root.Path if !conf.TestOnlyAllowRunAsCurrentUserWithoutChroot { - // FIXME: runsc can't be re-executed without - // /proc, so we create a tmpfs mount, mount ./proc and ./root - // there, then move this mount to the root and after + // runsc can't be re-executed without /proc, so we create a tmpfs mount, + // mount ./proc and ./root there, then move this mount to the root and after // setCapsAndCallSelf, runsc will chroot into /root. // // We need a directory to construct a new root and we know that diff --git a/test/syscalls/linux/epoll.cc b/test/syscalls/linux/epoll.cc index a4f8f3cec..f57d38dc7 100644 --- a/test/syscalls/linux/epoll.cc +++ b/test/syscalls/linux/epoll.cc @@ -56,10 +56,6 @@ TEST(EpollTest, AllWritable) { struct epoll_event result[kFDsPerEpoll]; ASSERT_THAT(RetryEINTR(epoll_wait)(epollfd.get(), result, kFDsPerEpoll, -1), SyscallSucceedsWithValue(kFDsPerEpoll)); - // TODO(edahlgren): Why do some tests check epoll_event::data, and others - // don't? Does Linux actually guarantee that, in any of these test cases, - // epoll_wait will necessarily write out the epoll_events in the order that - // they were registered? for (int i = 0; i < kFDsPerEpoll; i++) { ASSERT_EQ(result[i].events, EPOLLOUT); } diff --git a/test/syscalls/linux/file_base.h b/test/syscalls/linux/file_base.h index 25fdd7106..fb418e052 100644 --- a/test/syscalls/linux/file_base.h +++ b/test/syscalls/linux/file_base.h @@ -87,6 +87,7 @@ class FileTest : public ::testing::Test { ClosePipes(); } + protected: std::string test_file_name_; FileDescriptor test_file_fd_; diff --git a/test/syscalls/linux/pwrite64.cc b/test/syscalls/linux/pwrite64.cc index b48fe540d..c2f72e010 100644 --- a/test/syscalls/linux/pwrite64.cc +++ b/test/syscalls/linux/pwrite64.cc @@ -27,14 +27,7 @@ namespace testing { namespace { -// This test is currently very rudimentary. -// -// TODO(edahlgren): -// * bad buffer states (EFAULT). -// * bad fds (wrong permission, wrong type of file, EBADF). -// * check offset is not incremented. -// * check for EOF. -// * writing to pipes, symlinks, special files. +// TODO(gvisor.dev/issue/2370): This test is currently very rudimentary. class Pwrite64 : public ::testing::Test { void SetUp() override { name_ = NewTempAbsPath(); diff --git a/test/syscalls/linux/tuntap.cc b/test/syscalls/linux/tuntap.cc index 53ad2dda3..3a8ba37eb 100644 --- a/test/syscalls/linux/tuntap.cc +++ b/test/syscalls/linux/tuntap.cc @@ -242,7 +242,7 @@ TEST_F(TuntapTest, InvalidReadWrite) { TEST_F(TuntapTest, WriteToDownDevice) { SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN))); - // FIXME: gVisor always creates enabled/up'd interfaces. + // FIXME(b/110961832): gVisor always creates enabled/up'd interfaces. SKIP_IF(IsRunningOnGvisor()); FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(kDevNetTun, O_RDWR)); @@ -280,10 +280,11 @@ PosixErrorOr OpenAndAttachTap( &addr, sizeof(addr))); if (!IsRunningOnGvisor()) { - // FIXME: gVisor doesn't support setting MAC address on interfaces yet. + // FIXME(b/110961832): gVisor doesn't support setting MAC address on + // interfaces yet. RETURN_IF_ERRNO(LinkSetMacAddr(link->index, kMacA, sizeof(kMacA))); - // FIXME: gVisor always creates enabled/up'd interfaces. + // FIXME(b/110961832): gVisor always creates enabled/up'd interfaces. RETURN_IF_ERRNO(LinkChangeFlags(link->index, IFF_UP, IFF_UP)); } diff --git a/test/syscalls/linux/write.cc b/test/syscalls/linux/write.cc index 9b219cfd6..39b5b2f56 100644 --- a/test/syscalls/linux/write.cc +++ b/test/syscalls/linux/write.cc @@ -31,14 +31,8 @@ namespace gvisor { namespace testing { namespace { -// This test is currently very rudimentary. -// -// TODO(edahlgren): -// * bad buffer states (EFAULT). -// * bad fds (wrong permission, wrong type of file, EBADF). -// * check offset is incremented. -// * check for EOF. -// * writing to pipes, symlinks, special files. + +// TODO(gvisor.dev/issue/2370): This test is currently very rudimentary. class WriteTest : public ::testing::Test { public: ssize_t WriteBytes(int fd, int bytes) { diff --git a/tools/go_generics/defs.bzl b/tools/go_generics/defs.bzl index c5be52ecd..8c9995fd4 100644 --- a/tools/go_generics/defs.bzl +++ b/tools/go_generics/defs.bzl @@ -105,7 +105,6 @@ def _go_template_instance_impl(ctx): executable = ctx.executable._tool, ) - # TODO: How can we get the dependencies out? return struct( files = depset([output]), ) -- cgit v1.2.3 From 0f75f7273d8c4ace73d93b6b00f81d53a5cf76ea Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Wed, 8 Apr 2020 18:40:46 -0700 Subject: Don't call platform.AddressSpace.MapFile with no permissions. PiperOrigin-RevId: 305598136 --- pkg/sentry/mm/address_space.go | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/mm/address_space.go b/pkg/sentry/mm/address_space.go index 0332fc71c..5c667117c 100644 --- a/pkg/sentry/mm/address_space.go +++ b/pkg/sentry/mm/address_space.go @@ -201,8 +201,10 @@ func (mm *MemoryManager) mapASLocked(pseg pmaIterator, ar usermem.AddrRange, pre if pma.needCOW { perms.Write = false } - if err := mm.as.MapFile(pmaMapAR.Start, pma.file, pseg.fileRangeOf(pmaMapAR), perms, precommit); err != nil { - return err + if perms.Any() { // MapFile precondition + if err := mm.as.MapFile(pmaMapAR.Start, pma.file, pseg.fileRangeOf(pmaMapAR), perms, precommit); err != nil { + return err + } } pseg = pseg.NextSegment() } -- cgit v1.2.3 From a10389e783aab5f530641394ef44c8a1dede9372 Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Wed, 8 Apr 2020 23:02:09 -0700 Subject: splice: cap splice calls to MAX_RW_COUNT The Linux does the same. Reported-by: syzbot+e81716e8956e92e9d56b@syzkaller.appspotmail.com PiperOrigin-RevId: 305625439 --- pkg/sentry/syscalls/linux/sys_splice.go | 4 ++ test/syscalls/linux/BUILD | 2 + test/syscalls/linux/sendfile_socket.cc | 105 ++++++++++++++------------------ 3 files changed, 53 insertions(+), 58 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/sys_splice.go b/pkg/sentry/syscalls/linux/sys_splice.go index fd642834b..fbc6cf15f 100644 --- a/pkg/sentry/syscalls/linux/sys_splice.go +++ b/pkg/sentry/syscalls/linux/sys_splice.go @@ -29,6 +29,10 @@ func doSplice(t *kernel.Task, outFile, inFile *fs.File, opts fs.SpliceOpts, nonB return 0, syserror.EINVAL } + if opts.Length > int64(kernel.MAX_RW_COUNT) { + opts.Length = int64(kernel.MAX_RW_COUNT) + } + var ( total int64 n int64 diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index d0c431234..ae3017608 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -2026,6 +2026,8 @@ cc_binary( "//test/util:file_descriptor", "@com_google_absl//absl/strings", gtest, + ":ip_socket_test_util", + ":unix_domain_socket_test_util", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", diff --git a/test/syscalls/linux/sendfile_socket.cc b/test/syscalls/linux/sendfile_socket.cc index e94672679..c101fe9d2 100644 --- a/test/syscalls/linux/sendfile_socket.cc +++ b/test/syscalls/linux/sendfile_socket.cc @@ -23,6 +23,7 @@ #include "gtest/gtest.h" #include "absl/strings/string_view.h" +#include "test/syscalls/linux/ip_socket_test_util.h" #include "test/syscalls/linux/socket_test_util.h" #include "test/util/file_descriptor.h" #include "test/util/temp_path.h" @@ -35,61 +36,39 @@ namespace { class SendFileTest : public ::testing::TestWithParam { protected: - PosixErrorOr> Sockets() { + PosixErrorOr> Sockets(int type) { // Bind a server socket. int family = GetParam(); - struct sockaddr server_addr = {}; switch (family) { case AF_INET: { - struct sockaddr_in* server_addr_in = - reinterpret_cast(&server_addr); - server_addr_in->sin_family = family; - server_addr_in->sin_addr.s_addr = INADDR_ANY; - break; + if (type == SOCK_STREAM) { + return SocketPairKind{ + "TCP", AF_INET, type, 0, + TCPAcceptBindSocketPairCreator(AF_INET, type, 0, false)} + .Create(); + } else { + return SocketPairKind{ + "UDP", AF_INET, type, 0, + UDPBidirectionalBindSocketPairCreator(AF_INET, type, 0, false)} + .Create(); + } } case AF_UNIX: { - struct sockaddr_un* server_addr_un = - reinterpret_cast(&server_addr); - server_addr_un->sun_family = family; - server_addr_un->sun_path[0] = '\0'; - break; + if (type == SOCK_STREAM) { + return SocketPairKind{ + "UNIX", AF_UNIX, type, 0, + FilesystemAcceptBindSocketPairCreator(AF_UNIX, type, 0)} + .Create(); + } else { + return SocketPairKind{ + "UNIX", AF_UNIX, type, 0, + FilesystemBidirectionalBindSocketPairCreator(AF_UNIX, type, 0)} + .Create(); + } } default: return PosixError(EINVAL); } - int server = socket(family, SOCK_STREAM, 0); - if (bind(server, &server_addr, sizeof(server_addr)) < 0) { - return PosixError(errno); - } - if (listen(server, 1) < 0) { - close(server); - return PosixError(errno); - } - - // Fetch the address; both are anonymous. - socklen_t length = sizeof(server_addr); - if (getsockname(server, &server_addr, &length) < 0) { - close(server); - return PosixError(errno); - } - - // Connect the client. - int client = socket(family, SOCK_STREAM, 0); - if (connect(client, &server_addr, length) < 0) { - close(server); - close(client); - return PosixError(errno); - } - - // Accept on the server. - int server_client = accept(server, nullptr, 0); - if (server_client < 0) { - close(server); - close(client); - return PosixError(errno); - } - close(server); - return std::make_tuple(client, server_client); } }; @@ -106,9 +85,7 @@ TEST_P(SendFileTest, SendMultiple) { const TempPath out_file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); // Create sockets. - std::tuple fds = ASSERT_NO_ERRNO_AND_VALUE(Sockets()); - const FileDescriptor server(std::get<0>(fds)); - FileDescriptor client(std::get<1>(fds)); // non-const, reset is used. + auto socks = ASSERT_NO_ERRNO_AND_VALUE(Sockets(SOCK_STREAM)); // Thread that reads data from socket and dumps to a file. ScopedThread th([&] { @@ -118,7 +95,7 @@ TEST_P(SendFileTest, SendMultiple) { // Read until socket is closed. char buf[10240]; for (int cnt = 0;; cnt++) { - int r = RetryEINTR(read)(server.get(), buf, sizeof(buf)); + int r = RetryEINTR(read)(socks->first_fd(), buf, sizeof(buf)); // We cannot afford to save on every read() call. if (cnt % 1000 == 0) { ASSERT_THAT(r, SyscallSucceeds()); @@ -152,7 +129,7 @@ TEST_P(SendFileTest, SendMultiple) { << ", remain=" << remain << std::endl; // Send data and verify that sendfile returns the correct value. - int res = sendfile(client.get(), inf.get(), nullptr, remain); + int res = sendfile(socks->second_fd(), inf.get(), nullptr, remain); // We cannot afford to save on every sendfile() call. if (cnt % 120 == 0) { MaybeSave(); @@ -169,7 +146,7 @@ TEST_P(SendFileTest, SendMultiple) { } // Close socket to stop thread. - client.reset(); + close(socks->release_second_fd()); th.Join(); // Verify that the output file has the correct data. @@ -183,9 +160,7 @@ TEST_P(SendFileTest, SendMultiple) { TEST_P(SendFileTest, Shutdown) { // Create a socket. - std::tuple fds = ASSERT_NO_ERRNO_AND_VALUE(Sockets()); - const FileDescriptor client(std::get<0>(fds)); - FileDescriptor server(std::get<1>(fds)); // non-const, reset below. + auto socks = ASSERT_NO_ERRNO_AND_VALUE(Sockets(SOCK_STREAM)); // If this is a TCP socket, then turn off linger. if (GetParam() == AF_INET) { @@ -193,7 +168,7 @@ TEST_P(SendFileTest, Shutdown) { sl.l_onoff = 1; sl.l_linger = 0; ASSERT_THAT( - setsockopt(server.get(), SOL_SOCKET, SO_LINGER, &sl, sizeof(sl)), + setsockopt(socks->first_fd(), SOL_SOCKET, SO_LINGER, &sl, sizeof(sl)), SyscallSucceeds()); } @@ -212,12 +187,12 @@ TEST_P(SendFileTest, Shutdown) { ScopedThread t([&]() { size_t done = 0; while (done < data.size()) { - int n = RetryEINTR(read)(server.get(), data.data(), data.size()); + int n = RetryEINTR(read)(socks->first_fd(), data.data(), data.size()); ASSERT_THAT(n, SyscallSucceeds()); done += n; } // Close the server side socket. - server.reset(); + close(socks->release_first_fd()); }); // Continuously stream from the file to the socket. Note we do not assert @@ -225,7 +200,7 @@ TEST_P(SendFileTest, Shutdown) { // data is written. Eventually, we should get a connection reset error. while (1) { off_t offset = 0; // Always read from the start. - int n = sendfile(client.get(), inf.get(), &offset, data.size()); + int n = sendfile(socks->second_fd(), inf.get(), &offset, data.size()); EXPECT_THAT(n, AnyOf(SyscallFailsWithErrno(ECONNRESET), SyscallFailsWithErrno(EPIPE), SyscallSucceeds())); if (n <= 0) { @@ -234,6 +209,20 @@ TEST_P(SendFileTest, Shutdown) { } } +TEST_P(SendFileTest, SendpageFromEmptyFileToUDP) { + auto socks = ASSERT_NO_ERRNO_AND_VALUE(Sockets(SOCK_DGRAM)); + + TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(file.path(), O_RDWR)); + + // The value to the count argument has to be so that it is impossible to + // allocate a buffer of this size. In Linux, sendfile transfer at most + // 0x7ffff000 (MAX_RW_COUNT) bytes. + EXPECT_THAT(sendfile(socks->first_fd(), fd.get(), 0x0, 0x8000000000004), + SyscallSucceedsWithValue(0)); +} + INSTANTIATE_TEST_SUITE_P(AddressFamily, SendFileTest, ::testing::Values(AF_UNIX, AF_INET)); -- cgit v1.2.3 From ab54d4f496dc59721066e295e162b3be903795d5 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Thu, 2 Apr 2020 01:04:58 -0400 Subject: remove nogo exemption for machine_arm64_unsafe.go Minimize the use of unsafe. Signed-off-by: Bin Lu --- pkg/sentry/platform/kvm/kvm_arm64.go | 4 +- pkg/sentry/platform/kvm/machine_arm64_unsafe.go | 63 -------------- pkg/sentry/platform/ring0/BUILD | 2 + pkg/sentry/platform/ring0/lib_arm64.go | 7 ++ pkg/sentry/platform/ring0/lib_arm64_unsafe.go | 108 ++++++++++++++++++++++++ tools/nogo.json | 1 - 6 files changed, 120 insertions(+), 65 deletions(-) create mode 100644 pkg/sentry/platform/ring0/lib_arm64_unsafe.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/kvm/kvm_arm64.go b/pkg/sentry/platform/kvm/kvm_arm64.go index 79045651e..716198712 100644 --- a/pkg/sentry/platform/kvm/kvm_arm64.go +++ b/pkg/sentry/platform/kvm/kvm_arm64.go @@ -18,6 +18,8 @@ package kvm import ( "syscall" + + "gvisor.dev/gvisor/pkg/sentry/platform/ring0" ) type kvmOneReg struct { @@ -46,6 +48,6 @@ type userRegs struct { func updateGlobalOnce(fd int) error { physicalInit() err := updateSystemValues(int(fd)) - updateVectorTable() + ring0.Init() return err } diff --git a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go index b531f2f85..3b35858ae 100644 --- a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go @@ -48,69 +48,6 @@ func (m *machine) initArchState() error { return nil } -func getPageWithReflect(p uintptr) []byte { - return (*(*[0xFFFFFF]byte)(unsafe.Pointer(p & ^uintptr(syscall.Getpagesize()-1))))[:syscall.Getpagesize()] -} - -// Work around: move ring0.Vectors() into a specific address with 11-bits alignment. -// -// According to the design documentation of Arm64, -// the start address of exception vector table should be 11-bits aligned. -// Please see the code in linux kernel as reference: arch/arm64/kernel/entry.S -// But, we can't align a function's start address to a specific address by using golang. -// We have raised this question in golang community: -// https://groups.google.com/forum/m/#!topic/golang-dev/RPj90l5x86I -// This function will be removed when golang supports this feature. -// -// There are 2 jobs were implemented in this function: -// 1, move the start address of exception vector table into the specific address. -// 2, modify the offset of each instruction. -func updateVectorTable() { - fromLocation := reflect.ValueOf(ring0.Vectors).Pointer() - offset := fromLocation & (1<<11 - 1) - if offset != 0 { - offset = 1<<11 - offset - } - - toLocation := fromLocation + offset - page := getPageWithReflect(toLocation) - if err := syscall.Mprotect(page, syscall.PROT_READ|syscall.PROT_WRITE|syscall.PROT_EXEC); err != nil { - panic(err) - } - - page = getPageWithReflect(toLocation + 4096) - if err := syscall.Mprotect(page, syscall.PROT_READ|syscall.PROT_WRITE|syscall.PROT_EXEC); err != nil { - panic(err) - } - - // Move exception-vector-table into the specific address. - var entry *uint32 - var entryFrom *uint32 - for i := 1; i <= 0x800; i++ { - entry = (*uint32)(unsafe.Pointer(toLocation + 0x800 - uintptr(i))) - entryFrom = (*uint32)(unsafe.Pointer(fromLocation + 0x800 - uintptr(i))) - *entry = *entryFrom - } - - // The offset from the address of each unconditionally branch is changed. - // We should modify the offset of each instruction. - nums := []uint32{0x0, 0x80, 0x100, 0x180, 0x200, 0x280, 0x300, 0x380, 0x400, 0x480, 0x500, 0x580, 0x600, 0x680, 0x700, 0x780} - for _, num := range nums { - entry = (*uint32)(unsafe.Pointer(toLocation + uintptr(num))) - *entry = *entry - (uint32)(offset/4) - } - - page = getPageWithReflect(toLocation) - if err := syscall.Mprotect(page, syscall.PROT_READ|syscall.PROT_EXEC); err != nil { - panic(err) - } - - page = getPageWithReflect(toLocation + 4096) - if err := syscall.Mprotect(page, syscall.PROT_READ|syscall.PROT_EXEC); err != nil { - panic(err) - } -} - // initArchState initializes architecture-specific state. func (c *vCPU) initArchState() error { var ( diff --git a/pkg/sentry/platform/ring0/BUILD b/pkg/sentry/platform/ring0/BUILD index 934b6fbcd..cdcad0fdc 100644 --- a/pkg/sentry/platform/ring0/BUILD +++ b/pkg/sentry/platform/ring0/BUILD @@ -71,12 +71,14 @@ go_library( "lib_amd64.go", "lib_amd64.s", "lib_arm64.go", + "lib_arm64_unsafe.go", "lib_arm64.s", "ring0.go", ], visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/cpuid", + "//pkg/safecopy", "//pkg/sentry/platform/ring0/pagetables", "//pkg/usermem", ], diff --git a/pkg/sentry/platform/ring0/lib_arm64.go b/pkg/sentry/platform/ring0/lib_arm64.go index af075aae4..242b9305c 100644 --- a/pkg/sentry/platform/ring0/lib_arm64.go +++ b/pkg/sentry/platform/ring0/lib_arm64.go @@ -37,3 +37,10 @@ func SaveVRegs(*byte) // LoadVRegs loads V0-V31 registers. func LoadVRegs(*byte) + +// Init sets function pointers based on architectural features. +// +// This must be called prior to using ring0. +func Init() { + rewriteVectors() +} diff --git a/pkg/sentry/platform/ring0/lib_arm64_unsafe.go b/pkg/sentry/platform/ring0/lib_arm64_unsafe.go new file mode 100644 index 000000000..c05166fea --- /dev/null +++ b/pkg/sentry/platform/ring0/lib_arm64_unsafe.go @@ -0,0 +1,108 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package ring0 + +import ( + "reflect" + "syscall" + "unsafe" + + "gvisor.dev/gvisor/pkg/safecopy" + "gvisor.dev/gvisor/pkg/usermem" +) + +const ( + nopInstruction = 0xd503201f + instSize = unsafe.Sizeof(uint32(0)) + vectorsRawLen = 0x800 +) + +func unsafeSlice(addr uintptr, length int) (slice []uint32) { + hdr := (*reflect.SliceHeader)(unsafe.Pointer(&slice)) + hdr.Data = addr + hdr.Len = length / int(instSize) + hdr.Cap = length / int(instSize) + return slice +} + +// Work around: move ring0.Vectors() into a specific address with 11-bits alignment. +// +// According to the design documentation of Arm64, +// the start address of exception vector table should be 11-bits aligned. +// Please see the code in linux kernel as reference: arch/arm64/kernel/entry.S +// But, we can't align a function's start address to a specific address by using golang. +// We have raised this question in golang community: +// https://groups.google.com/forum/m/#!topic/golang-dev/RPj90l5x86I +// This function will be removed when golang supports this feature. +// +// There are 2 jobs were implemented in this function: +// 1, move the start address of exception vector table into the specific address. +// 2, modify the offset of each instruction. +func rewriteVectors() { + vectorsBegin := reflect.ValueOf(Vectors).Pointer() + + // The exception-vector-table is required to be 11-bits aligned. + // And the size is 0x800. + // Please see the documentation as reference: + // https://developer.arm.com/docs/100933/0100/aarch64-exception-vector-table + // + // But, golang does not allow to set a function's address to a specific value. + // So, for gvisor, I defined the size of exception-vector-table as 4K, + // filled the 2nd 2K part with NOP-s. + // So that, I can safely move the 1st 2K part into the address with 11-bits alignment. + // + // So, the prerequisite for this function to work correctly is: + // vectorsSafeLen >= 0x1000 + // vectorsRawLen = 0x800 + vectorsSafeLen := int(safecopy.FindEndAddress(vectorsBegin) - vectorsBegin) + if vectorsSafeLen < 2*vectorsRawLen { + panic("Can't update vectors") + } + + vectorsSafeTable := unsafeSlice(vectorsBegin, vectorsSafeLen) // Now a []uint32 + vectorsRawLen32 := vectorsRawLen / int(instSize) + + offset := vectorsBegin & (1<<11 - 1) + if offset != 0 { + offset = 1<<11 - offset + } + + pageBegin := (vectorsBegin + offset) & ^uintptr(usermem.PageSize-1) + + _, _, errno := syscall.Syscall(syscall.SYS_MPROTECT, uintptr(pageBegin), uintptr(usermem.PageSize), uintptr(syscall.PROT_READ|syscall.PROT_WRITE|syscall.PROT_EXEC)) + if errno != 0 { + panic(errno.Error()) + } + + offset = offset / instSize // By index, not bytes. + // Move exception-vector-table into the specific address, should uses memmove here. + for i := 1; i <= vectorsRawLen32; i++ { + vectorsSafeTable[int(offset)+vectorsRawLen32-i] = vectorsSafeTable[vectorsRawLen32-i] + } + + // Adjust branch since instruction was moved forward. + for i := 0; i < vectorsRawLen32; i++ { + if vectorsSafeTable[int(offset)+i] != nopInstruction { + vectorsSafeTable[int(offset)+i] -= uint32(offset) + } + } + + _, _, errno = syscall.Syscall(syscall.SYS_MPROTECT, uintptr(pageBegin), uintptr(usermem.PageSize), uintptr(syscall.PROT_READ|syscall.PROT_EXEC)) + if errno != 0 { + panic(errno.Error()) + } +} diff --git a/tools/nogo.json b/tools/nogo.json index 83cb76b93..2e354e160 100644 --- a/tools/nogo.json +++ b/tools/nogo.json @@ -83,7 +83,6 @@ "/pkg/gohacks/gohacks_unsafe.go": "allowed: special case", "/pkg/sentry/fs/fsutil/host_file_mapper_unsafe.go": "allowed: special case", "/pkg/sentry/platform/kvm/(bluepill|machine)_unsafe.go": "allowed: special case", - "/pkg/sentry/platform/kvm/machine_arm64_unsafe.go": "fix: gvisor.dev/issue/22464", "/pkg/sentry/platform/ring0/pagetables/allocator_unsafe.go": "allowed: special case", "/pkg/sentry/platform/safecopy/safecopy_unsafe.go": "allowed: special case", "/pkg/sentry/vfs/mount_unsafe.go": "allowed: special case" -- cgit v1.2.3 From 7928aa345e334f2c68f8f03b71d8cabe79e8db7e Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Thu, 9 Apr 2020 09:30:39 -0700 Subject: Convert int and bool socket options to use GetSockOptInt and GetSockOptBool PiperOrigin-RevId: 305699233 --- pkg/sentry/socket/netstack/netstack.go | 155 +++++------- pkg/sentry/socket/unix/transport/BUILD | 1 + pkg/sentry/socket/unix/transport/unix.go | 50 ++-- pkg/tcpip/stack/transport_demuxer_test.go | 35 ++- pkg/tcpip/tcpip.go | 145 ++++++----- pkg/tcpip/transport/icmp/endpoint.go | 47 ++-- pkg/tcpip/transport/raw/endpoint.go | 18 +- pkg/tcpip/transport/tcp/endpoint.go | 387 +++++++++++++----------------- pkg/tcpip/transport/tcp/tcp_test.go | 110 +++++---- pkg/tcpip/transport/udp/endpoint.go | 231 +++++++++--------- pkg/tcpip/transport/udp/udp_test.go | 60 ++--- 11 files changed, 583 insertions(+), 656 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 5d0085462..20e3fa0d2 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -300,7 +300,7 @@ type SocketOperations struct { // New creates a new endpoint socket. func New(t *kernel.Task, family int, skType linux.SockType, protocol int, queue *waiter.Queue, endpoint tcpip.Endpoint) (*fs.File, *syserr.Error) { if skType == linux.SOCK_STREAM { - if err := endpoint.SetSockOptInt(tcpip.DelayOption, 1); err != nil { + if err := endpoint.SetSockOptBool(tcpip.DelayOption, true); err != nil { return nil, syserr.TranslateNetstackError(err) } } @@ -965,6 +965,13 @@ func GetSockOpt(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, family in return nil, syserr.ErrProtocolNotAvailable } +func boolToInt32(v bool) int32 { + if v { + return 1 + } + return 0 +} + // getSockOptSocket implements GetSockOpt when level is SOL_SOCKET. func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, family int, skType linux.SockType, name, outLen int) (interface{}, *syserr.Error) { // TODO(b/124056281): Stop rejecting short optLen values in getsockopt. @@ -998,12 +1005,11 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam return nil, syserr.ErrInvalidArgument } - var v tcpip.PasscredOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptBool(tcpip.PasscredOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(v), nil + return boolToInt32(v), nil case linux.SO_SNDBUF: if outLen < sizeOfInt32 { @@ -1042,24 +1048,22 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam return nil, syserr.ErrInvalidArgument } - var v tcpip.ReuseAddressOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptBool(tcpip.ReuseAddressOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(v), nil + return boolToInt32(v), nil case linux.SO_REUSEPORT: if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - var v tcpip.ReusePortOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptBool(tcpip.ReusePortOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(v), nil + return boolToInt32(v), nil case linux.SO_BINDTODEVICE: var v tcpip.BindToDeviceOption @@ -1089,24 +1093,22 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam return nil, syserr.ErrInvalidArgument } - var v tcpip.BroadcastOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptBool(tcpip.BroadcastOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(v), nil + return boolToInt32(v), nil case linux.SO_KEEPALIVE: if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - var v tcpip.KeepaliveEnabledOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptBool(tcpip.KeepaliveEnabledOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(v), nil + return boolToInt32(v), nil case linux.SO_LINGER: if outLen < linux.SizeOfLinger { @@ -1156,47 +1158,41 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa return nil, syserr.ErrInvalidArgument } - v, err := ep.GetSockOptInt(tcpip.DelayOption) + v, err := ep.GetSockOptBool(tcpip.DelayOption) if err != nil { return nil, syserr.TranslateNetstackError(err) } - - if v == 0 { - return int32(1), nil - } - return int32(0), nil + return boolToInt32(!v), nil case linux.TCP_CORK: if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - var v tcpip.CorkOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptBool(tcpip.CorkOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(v), nil + return boolToInt32(v), nil case linux.TCP_QUICKACK: if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - var v tcpip.QuickAckOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptBool(tcpip.QuickAckOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(v), nil + return boolToInt32(v), nil case linux.TCP_MAXSEG: if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - var v tcpip.MaxSegOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptInt(tcpip.MaxSegOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } @@ -1328,11 +1324,7 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf if err != nil { return nil, syserr.TranslateNetstackError(err) } - var o int32 - if v { - o = 1 - } - return o, nil + return boolToInt32(v), nil case linux.IPV6_PATHMTU: t.Kernel().EmitUnimplementedEvent(t) @@ -1342,8 +1334,8 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf if outLen == 0 { return make([]byte, 0), nil } - var v tcpip.IPv6TrafficClassOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptInt(tcpip.IPv6TrafficClassOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } @@ -1365,12 +1357,7 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf if err != nil { return nil, syserr.TranslateNetstackError(err) } - - var o int32 - if v { - o = 1 - } - return o, nil + return boolToInt32(v), nil default: emitUnimplementedEventIPv6(t, name) @@ -1386,8 +1373,8 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in return nil, syserr.ErrInvalidArgument } - var v tcpip.TTLOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptInt(tcpip.TTLOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } @@ -1403,8 +1390,8 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in return nil, syserr.ErrInvalidArgument } - var v tcpip.MulticastTTLOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptInt(tcpip.MulticastTTLOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } @@ -1429,23 +1416,19 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in return nil, syserr.ErrInvalidArgument } - var v tcpip.MulticastLoopOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptBool(tcpip.MulticastLoopOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } - - if v { - return int32(1), nil - } - return int32(0), nil + return boolToInt32(v), nil case linux.IP_TOS: // Length handling for parity with Linux. if outLen == 0 { return []byte(nil), nil } - var v tcpip.IPv4TOSOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptInt(tcpip.IPv4TOSOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } if outLen < sizeOfInt32 { @@ -1462,11 +1445,7 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in if err != nil { return nil, syserr.TranslateNetstackError(err) } - var o int32 - if v { - o = 1 - } - return o, nil + return boolToInt32(v), nil case linux.IP_PKTINFO: if outLen < sizeOfInt32 { @@ -1477,11 +1456,7 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in if err != nil { return nil, syserr.TranslateNetstackError(err) } - var o int32 - if v { - o = 1 - } - return o, nil + return boolToInt32(v), nil default: emitUnimplementedEventIP(t, name) @@ -1592,7 +1567,7 @@ func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, nam } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.ReuseAddressOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.ReuseAddressOption, v != 0)) case linux.SO_REUSEPORT: if len(optVal) < sizeOfInt32 { @@ -1600,7 +1575,7 @@ func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, nam } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.ReusePortOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.ReusePortOption, v != 0)) case linux.SO_BINDTODEVICE: n := bytes.IndexByte(optVal, 0) @@ -1628,7 +1603,7 @@ func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, nam } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.BroadcastOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.BroadcastOption, v != 0)) case linux.SO_PASSCRED: if len(optVal) < sizeOfInt32 { @@ -1636,7 +1611,7 @@ func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, nam } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.PasscredOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.PasscredOption, v != 0)) case linux.SO_KEEPALIVE: if len(optVal) < sizeOfInt32 { @@ -1644,7 +1619,7 @@ func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, nam } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.KeepaliveEnabledOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.KeepaliveEnabledOption, v != 0)) case linux.SO_SNDTIMEO: if len(optVal) < linux.SizeOfTimeval { @@ -1716,11 +1691,7 @@ func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) * } v := usermem.ByteOrder.Uint32(optVal) - var o int - if v == 0 { - o = 1 - } - return syserr.TranslateNetstackError(ep.SetSockOptInt(tcpip.DelayOption, o)) + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.DelayOption, v == 0)) case linux.TCP_CORK: if len(optVal) < sizeOfInt32 { @@ -1728,7 +1699,7 @@ func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) * } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.CorkOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.CorkOption, v != 0)) case linux.TCP_QUICKACK: if len(optVal) < sizeOfInt32 { @@ -1736,7 +1707,7 @@ func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) * } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.QuickAckOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.QuickAckOption, v != 0)) case linux.TCP_MAXSEG: if len(optVal) < sizeOfInt32 { @@ -1744,7 +1715,7 @@ func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) * } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.MaxSegOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptInt(tcpip.MaxSegOption, int(v))) case linux.TCP_KEEPIDLE: if len(optVal) < sizeOfInt32 { @@ -1855,7 +1826,7 @@ func setSockOptIPv6(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) if v == -1 { v = 0 } - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.IPv6TrafficClassOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptInt(tcpip.IPv6TrafficClassOption, int(v))) case linux.IPV6_RECVTCLASS: v, err := parseIntOrChar(optVal) @@ -1940,7 +1911,7 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s if v < 0 || v > 255 { return syserr.ErrInvalidArgument } - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.MulticastTTLOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptInt(tcpip.MulticastTTLOption, int(v))) case linux.IP_ADD_MEMBERSHIP: req, err := copyInMulticastRequest(optVal, false /* allowAddr */) @@ -1987,9 +1958,7 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s return err } - return syserr.TranslateNetstackError(ep.SetSockOpt( - tcpip.MulticastLoopOption(v != 0), - )) + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.MulticastLoopOption, v != 0)) case linux.MCAST_JOIN_GROUP: // FIXME(b/124219304): Implement MCAST_JOIN_GROUP. @@ -2008,7 +1977,7 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s } else if v < 1 || v > 255 { return syserr.ErrInvalidArgument } - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.TTLOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptInt(tcpip.TTLOption, int(v))) case linux.IP_TOS: if len(optVal) == 0 { @@ -2018,7 +1987,7 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s if err != nil { return err } - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.IPv4TOSOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptInt(tcpip.IPv4TOSOption, int(v))) case linux.IP_RECVTOS: v, err := parseIntOrChar(optVal) diff --git a/pkg/sentry/socket/unix/transport/BUILD b/pkg/sentry/socket/unix/transport/BUILD index 74bcd6300..c708b6030 100644 --- a/pkg/sentry/socket/unix/transport/BUILD +++ b/pkg/sentry/socket/unix/transport/BUILD @@ -30,6 +30,7 @@ go_library( "//pkg/abi/linux", "//pkg/context", "//pkg/ilist", + "//pkg/log", "//pkg/refs", "//pkg/sync", "//pkg/syserr", diff --git a/pkg/sentry/socket/unix/transport/unix.go b/pkg/sentry/socket/unix/transport/unix.go index 2ef654235..1f3880cc5 100644 --- a/pkg/sentry/socket/unix/transport/unix.go +++ b/pkg/sentry/socket/unix/transport/unix.go @@ -20,6 +20,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/tcpip" @@ -838,24 +839,45 @@ func (e *baseEndpoint) SendMsg(ctx context.Context, data [][]byte, c ControlMess // SetSockOpt sets a socket option. Currently not supported. func (e *baseEndpoint) SetSockOpt(opt interface{}) *tcpip.Error { - switch v := opt.(type) { - case tcpip.PasscredOption: - e.setPasscred(v != 0) - return nil - } return nil } func (e *baseEndpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { + switch opt { + case tcpip.BroadcastOption: + case tcpip.PasscredOption: + e.setPasscred(v) + case tcpip.ReuseAddressOption: + default: + log.Warningf("Unsupported socket option: %d", opt) + return tcpip.ErrUnknownProtocolOption + } return nil } func (e *baseEndpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { + switch opt { + case tcpip.SendBufferSizeOption: + case tcpip.ReceiveBufferSizeOption: + default: + log.Warningf("Unsupported socket option: %d", opt) + return tcpip.ErrUnknownProtocolOption + } return nil } func (e *baseEndpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { - return false, tcpip.ErrUnknownProtocolOption + switch opt { + case tcpip.KeepaliveEnabledOption: + return false, nil + + case tcpip.PasscredOption: + return e.Passcred(), nil + + default: + log.Warningf("Unsupported socket option: %d", opt) + return false, tcpip.ErrUnknownProtocolOption + } } func (e *baseEndpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { @@ -914,29 +936,19 @@ func (e *baseEndpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { return int(v), nil default: + log.Warningf("Unsupported socket option: %d", opt) return -1, tcpip.ErrUnknownProtocolOption } } // GetSockOpt implements tcpip.Endpoint.GetSockOpt. func (e *baseEndpoint) GetSockOpt(opt interface{}) *tcpip.Error { - switch o := opt.(type) { + switch opt.(type) { case tcpip.ErrorOption: return nil - case *tcpip.PasscredOption: - if e.Passcred() { - *o = tcpip.PasscredOption(1) - } else { - *o = tcpip.PasscredOption(0) - } - return nil - - case *tcpip.KeepaliveEnabledOption: - *o = 0 - return nil - default: + log.Warningf("Unsupported socket option: %T", opt) return tcpip.ErrUnknownProtocolOption } } diff --git a/pkg/tcpip/stack/transport_demuxer_test.go b/pkg/tcpip/stack/transport_demuxer_test.go index c65b0c632..2474a7db3 100644 --- a/pkg/tcpip/stack/transport_demuxer_test.go +++ b/pkg/tcpip/stack/transport_demuxer_test.go @@ -206,7 +206,7 @@ func TestTransportDemuxerRegister(t *testing.T) { // the distribution of packets received matches expectations. func TestBindToDeviceDistribution(t *testing.T) { type endpointSockopts struct { - reuse int + reuse bool bindToDevice tcpip.NICID } for _, test := range []struct { @@ -221,11 +221,11 @@ func TestBindToDeviceDistribution(t *testing.T) { "BindPortReuse", // 5 endpoints that all have reuse set. []endpointSockopts{ - {reuse: 1, bindToDevice: 0}, - {reuse: 1, bindToDevice: 0}, - {reuse: 1, bindToDevice: 0}, - {reuse: 1, bindToDevice: 0}, - {reuse: 1, bindToDevice: 0}, + {reuse: true, bindToDevice: 0}, + {reuse: true, bindToDevice: 0}, + {reuse: true, bindToDevice: 0}, + {reuse: true, bindToDevice: 0}, + {reuse: true, bindToDevice: 0}, }, map[tcpip.NICID][]float64{ // Injected packets on dev0 get distributed evenly. @@ -236,9 +236,9 @@ func TestBindToDeviceDistribution(t *testing.T) { "BindToDevice", // 3 endpoints with various bindings. []endpointSockopts{ - {reuse: 0, bindToDevice: 1}, - {reuse: 0, bindToDevice: 2}, - {reuse: 0, bindToDevice: 3}, + {reuse: false, bindToDevice: 1}, + {reuse: false, bindToDevice: 2}, + {reuse: false, bindToDevice: 3}, }, map[tcpip.NICID][]float64{ // Injected packets on dev0 go only to the endpoint bound to dev0. @@ -253,12 +253,12 @@ func TestBindToDeviceDistribution(t *testing.T) { "ReuseAndBindToDevice", // 6 endpoints with various bindings. []endpointSockopts{ - {reuse: 1, bindToDevice: 1}, - {reuse: 1, bindToDevice: 1}, - {reuse: 1, bindToDevice: 2}, - {reuse: 1, bindToDevice: 2}, - {reuse: 1, bindToDevice: 2}, - {reuse: 1, bindToDevice: 0}, + {reuse: true, bindToDevice: 1}, + {reuse: true, bindToDevice: 1}, + {reuse: true, bindToDevice: 2}, + {reuse: true, bindToDevice: 2}, + {reuse: true, bindToDevice: 2}, + {reuse: true, bindToDevice: 0}, }, map[tcpip.NICID][]float64{ // Injected packets on dev0 get distributed among endpoints bound to @@ -309,9 +309,8 @@ func TestBindToDeviceDistribution(t *testing.T) { }(ep) defer ep.Close() - reusePortOption := tcpip.ReusePortOption(endpoint.reuse) - if err := ep.SetSockOpt(reusePortOption); err != nil { - t.Fatalf("SetSockOpt(%#v) on endpoint %d failed: %s", reusePortOption, i, err) + if err := ep.SetSockOptBool(tcpip.ReusePortOption, endpoint.reuse); err != nil { + t.Fatalf("SetSockOptBool(ReusePortOption, %t) on endpoint %d failed: %s", endpoint.reuse, i, err) } bindToDeviceOption := tcpip.BindToDeviceOption(endpoint.bindToDevice) if err := ep.SetSockOpt(bindToDeviceOption); err != nil { diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 2ef3271f1..aec7126ff 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -520,34 +520,90 @@ type WriteOptions struct { type SockOptBool int const ( + // BroadcastOption is used by SetSockOpt/GetSockOpt to specify whether + // datagram sockets are allowed to send packets to a broadcast address. + BroadcastOption SockOptBool = iota + + // CorkOption is used by SetSockOpt/GetSockOpt to specify if data should be + // held until segments are full by the TCP transport protocol. + CorkOption + + // DelayOption is used by SetSockOpt/GetSockOpt to specify if data + // should be sent out immediately by the transport protocol. For TCP, + // it determines if the Nagle algorithm is on or off. + DelayOption + + // KeepaliveEnabledOption is used by SetSockOpt/GetSockOpt to specify whether + // TCP keepalive is enabled for this socket. + KeepaliveEnabledOption + + // MulticastLoopOption is used by SetSockOpt/GetSockOpt to specify whether + // multicast packets sent over a non-loopback interface will be looped back. + MulticastLoopOption + + // PasscredOption is used by SetSockOpt/GetSockOpt to specify whether + // SCM_CREDENTIALS socket control messages are enabled. + // + // Only supported on Unix sockets. + PasscredOption + + // QuickAckOption is stubbed out in SetSockOpt/GetSockOpt. + QuickAckOption + // ReceiveTClassOption is used by SetSockOpt/GetSockOpt to specify if the // IPV6_TCLASS ancillary message is passed with incoming packets. - ReceiveTClassOption SockOptBool = iota + ReceiveTClassOption // ReceiveTOSOption is used by SetSockOpt/GetSockOpt to specify if the TOS // ancillary message is passed with incoming packets. ReceiveTOSOption - // V6OnlyOption is used by {G,S}etSockOptBool to specify whether an IPv6 - // socket is to be restricted to sending and receiving IPv6 packets only. - V6OnlyOption - // ReceiveIPPacketInfoOption is used by {G,S}etSockOptBool to specify // if more inforamtion is provided with incoming packets such // as interface index and address. ReceiveIPPacketInfoOption - // TODO(b/146901447): convert existing bool socket options to be handled via - // Get/SetSockOptBool + // ReuseAddressOption is used by SetSockOpt/GetSockOpt to specify whether Bind() + // should allow reuse of local address. + ReuseAddressOption + + // ReusePortOption is used by SetSockOpt/GetSockOpt to permit multiple sockets + // to be bound to an identical socket address. + ReusePortOption + + // V6OnlyOption is used by {G,S}etSockOptBool to specify whether an IPv6 + // socket is to be restricted to sending and receiving IPv6 packets only. + V6OnlyOption ) // SockOptInt represents socket options which values have the int type. type SockOptInt int const ( + // KeepaliveCountOption is used by SetSockOpt/GetSockOpt to specify the number + // of un-ACKed TCP keepalives that will be sent before the connection is + // closed. + KeepaliveCountOption SockOptInt = iota + + // IPv4TOSOption is used by SetSockOpt/GetSockOpt to specify TOS + // for all subsequent outgoing IPv4 packets from the endpoint. + IPv4TOSOption + + // IPv6TrafficClassOption is used by SetSockOpt/GetSockOpt to specify TOS + // for all subsequent outgoing IPv6 packets from the endpoint. + IPv6TrafficClassOption + + // MaxSegOption is used by SetSockOpt/GetSockOpt to set/get the current + // Maximum Segment Size(MSS) value as specified using the TCP_MAXSEG option. + MaxSegOption + + // MulticastTTLOption is used by SetSockOpt/GetSockOpt to control the default + // TTL value for multicast messages. The default is 1. + MulticastTTLOption + // ReceiveQueueSizeOption is used in GetSockOptInt to specify that the // number of unread bytes in the input buffer should be returned. - ReceiveQueueSizeOption SockOptInt = iota + ReceiveQueueSizeOption // SendBufferSizeOption is used by SetSockOptInt/GetSockOptInt to // specify the send buffer size option. @@ -561,44 +617,21 @@ const ( // number of unread bytes in the output buffer should be returned. SendQueueSizeOption - // DelayOption is used by SetSockOpt/GetSockOpt to specify if data - // should be sent out immediately by the transport protocol. For TCP, - // it determines if the Nagle algorithm is on or off. - DelayOption - - // TODO(b/137664753): convert all int socket options to be handled via - // GetSockOptInt. + // TTLOption is used by SetSockOpt/GetSockOpt to control the default TTL/hop + // limit value for unicast messages. The default is protocol specific. + // + // A zero value indicates the default. + TTLOption ) // ErrorOption is used in GetSockOpt to specify that the last error reported by // the endpoint should be cleared and returned. type ErrorOption struct{} -// CorkOption is used by SetSockOpt/GetSockOpt to specify if data should be -// held until segments are full by the TCP transport protocol. -type CorkOption int - -// ReuseAddressOption is used by SetSockOpt/GetSockOpt to specify whether Bind() -// should allow reuse of local address. -type ReuseAddressOption int - -// ReusePortOption is used by SetSockOpt/GetSockOpt to permit multiple sockets -// to be bound to an identical socket address. -type ReusePortOption int - // BindToDeviceOption is used by SetSockOpt/GetSockOpt to specify that sockets // should bind only on a specific NIC. type BindToDeviceOption NICID -// QuickAckOption is stubbed out in SetSockOpt/GetSockOpt. -type QuickAckOption int - -// PasscredOption is used by SetSockOpt/GetSockOpt to specify whether -// SCM_CREDENTIALS socket control messages are enabled. -// -// Only supported on Unix sockets. -type PasscredOption int - // TCPInfoOption is used by GetSockOpt to expose TCP statistics. // // TODO(b/64800844): Add and populate stat fields. @@ -607,10 +640,6 @@ type TCPInfoOption struct { RTTVar time.Duration } -// KeepaliveEnabledOption is used by SetSockOpt/GetSockOpt to specify whether -// TCP keepalive is enabled for this socket. -type KeepaliveEnabledOption int - // KeepaliveIdleOption is used by SetSockOpt/GetSockOpt to specify the time a // connection must remain idle before the first TCP keepalive packet is sent. // Once this time is reached, KeepaliveIntervalOption is used instead. @@ -620,11 +649,6 @@ type KeepaliveIdleOption time.Duration // interval between sending TCP keepalive packets. type KeepaliveIntervalOption time.Duration -// KeepaliveCountOption is used by SetSockOpt/GetSockOpt to specify the number -// of un-ACKed TCP keepalives that will be sent before the connection is -// closed. -type KeepaliveCountOption int - // TCPUserTimeoutOption is used by SetSockOpt/GetSockOpt to specify a user // specified timeout for a given TCP connection. // See: RFC5482 for details. @@ -638,20 +662,9 @@ type CongestionControlOption string // control algorithms. type AvailableCongestionControlOption string -// ModerateReceiveBufferOption allows the caller to enable/disable TCP receive // buffer moderation. type ModerateReceiveBufferOption bool -// MaxSegOption is used by SetSockOpt/GetSockOpt to set/get the current -// Maximum Segment Size(MSS) value as specified using the TCP_MAXSEG option. -type MaxSegOption int - -// TTLOption is used by SetSockOpt/GetSockOpt to control the default TTL/hop -// limit value for unicast messages. The default is protocol specific. -// -// A zero value indicates the default. -type TTLOption uint8 - // TCPLingerTimeoutOption is used by SetSockOpt/GetSockOpt to set/get the // maximum duration for which a socket lingers in the TCP_FIN_WAIT_2 state // before being marked closed. @@ -668,10 +681,6 @@ type TCPTimeWaitTimeoutOption time.Duration // for a handshake till the specified timeout until a segment with data arrives. type TCPDeferAcceptOption time.Duration -// MulticastTTLOption is used by SetSockOpt/GetSockOpt to control the default -// TTL value for multicast messages. The default is 1. -type MulticastTTLOption uint8 - // MulticastInterfaceOption is used by SetSockOpt/GetSockOpt to specify a // default interface for multicast. type MulticastInterfaceOption struct { @@ -679,10 +688,6 @@ type MulticastInterfaceOption struct { InterfaceAddr Address } -// MulticastLoopOption is used by SetSockOpt/GetSockOpt to specify whether -// multicast packets sent over a non-loopback interface will be looped back. -type MulticastLoopOption bool - // MembershipOption is used by SetSockOpt/GetSockOpt as an argument to // AddMembershipOption and RemoveMembershipOption. type MembershipOption struct { @@ -705,22 +710,10 @@ type RemoveMembershipOption MembershipOption // TCP out-of-band data is delivered along with the normal in-band data. type OutOfBandInlineOption int -// BroadcastOption is used by SetSockOpt/GetSockOpt to specify whether -// datagram sockets are allowed to send packets to a broadcast address. -type BroadcastOption int - // DefaultTTLOption is used by stack.(*Stack).NetworkProtocolOption to specify // a default TTL. type DefaultTTLOption uint8 -// IPv4TOSOption is used by SetSockOpt/GetSockOpt to specify TOS -// for all subsequent outgoing IPv4 packets from the endpoint. -type IPv4TOSOption uint8 - -// IPv6TrafficClassOption is used by SetSockOpt/GetSockOpt to specify TOS -// for all subsequent outgoing IPv6 packets from the endpoint. -type IPv6TrafficClassOption uint8 - // IPPacketInfo is the message struture for IP_PKTINFO. // // +stateify savable diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index b007302fb..3a133eef9 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -348,29 +348,37 @@ func (e *endpoint) Peek([][]byte) (int64, tcpip.ControlMessages, *tcpip.Error) { // SetSockOpt sets a socket option. func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { - switch o := opt.(type) { - case tcpip.TTLOption: - e.mu.Lock() - e.ttl = uint8(o) - e.mu.Unlock() - } - return nil } // SetSockOptBool sets a socket option. Currently not supported. func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { - return nil + return tcpip.ErrUnknownProtocolOption } // SetSockOptInt sets a socket option. Currently not supported. func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { + switch opt { + case tcpip.TTLOption: + e.mu.Lock() + e.ttl = uint8(v) + e.mu.Unlock() + + default: + return tcpip.ErrUnknownProtocolOption + } return nil } // GetSockOptBool implements tcpip.Endpoint.GetSockOptBool. func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { - return false, tcpip.ErrUnknownProtocolOption + switch opt { + case tcpip.KeepaliveEnabledOption: + return false, nil + + default: + return false, tcpip.ErrUnknownProtocolOption + } } // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. @@ -397,26 +405,23 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { e.rcvMu.Unlock() return v, nil + case tcpip.TTLOption: + e.rcvMu.Lock() + v := int(e.ttl) + e.rcvMu.Unlock() + return v, nil + + default: + return -1, tcpip.ErrUnknownProtocolOption } - return -1, tcpip.ErrUnknownProtocolOption } // GetSockOpt implements tcpip.Endpoint.GetSockOpt. func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { - switch o := opt.(type) { + switch opt.(type) { case tcpip.ErrorOption: return nil - case *tcpip.KeepaliveEnabledOption: - *o = 0 - return nil - - case *tcpip.TTLOption: - e.rcvMu.Lock() - *o = tcpip.TTLOption(e.ttl) - e.rcvMu.Unlock() - return nil - default: return tcpip.ErrUnknownProtocolOption } diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index 337bc1c71..eee754a5a 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -533,14 +533,10 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { // GetSockOpt implements tcpip.Endpoint.GetSockOpt. func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { - switch o := opt.(type) { + switch opt.(type) { case tcpip.ErrorOption: return nil - case *tcpip.KeepaliveEnabledOption: - *o = 0 - return nil - default: return tcpip.ErrUnknownProtocolOption } @@ -548,7 +544,13 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { // GetSockOptBool implements tcpip.Endpoint.GetSockOptBool. func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { - return false, tcpip.ErrUnknownProtocolOption + switch opt { + case tcpip.KeepaliveEnabledOption: + return false, nil + + default: + return false, tcpip.ErrUnknownProtocolOption + } } // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. @@ -576,9 +578,9 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { e.rcvMu.Unlock() return v, nil + default: + return -1, tcpip.ErrUnknownProtocolOption } - - return -1, tcpip.ErrUnknownProtocolOption } // HandlePacket implements stack.RawTransportEndpoint.HandlePacket. diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 9b123e968..a8d443f73 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -821,7 +821,7 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue var de DelayEnabled if err := s.TransportProtocolOption(ProtocolNumber, &de); err == nil && de { - e.SetSockOptInt(tcpip.DelayOption, 1) + e.SetSockOptBool(tcpip.DelayOption, true) } var tcpLT tcpip.TCPLingerTimeoutOption @@ -1409,10 +1409,60 @@ func (e *endpoint) windowCrossedACKThresholdLocked(deltaBefore int) (crossed boo // SetSockOptBool sets a socket option. func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { - e.LockUser() - defer e.UnlockUser() - switch opt { + + case tcpip.BroadcastOption: + e.LockUser() + e.broadcast = v + e.UnlockUser() + + case tcpip.CorkOption: + e.LockUser() + if !v { + atomic.StoreUint32(&e.cork, 0) + + // Handle the corked data. + e.sndWaker.Assert() + } else { + atomic.StoreUint32(&e.cork, 1) + } + e.UnlockUser() + + case tcpip.DelayOption: + if v { + atomic.StoreUint32(&e.delay, 1) + } else { + atomic.StoreUint32(&e.delay, 0) + + // Handle delayed data. + e.sndWaker.Assert() + } + + case tcpip.KeepaliveEnabledOption: + e.keepalive.Lock() + e.keepalive.enabled = v + e.keepalive.Unlock() + e.notifyProtocolGoroutine(notifyKeepaliveChanged) + + case tcpip.QuickAckOption: + o := uint32(1) + if v { + o = 0 + } + atomic.StoreUint32(&e.slowAck, o) + + case tcpip.ReuseAddressOption: + e.LockUser() + e.reuseAddr = v + e.UnlockUser() + return nil + + case tcpip.ReusePortOption: + e.LockUser() + e.reusePort = v + e.UnlockUser() + return nil + case tcpip.V6OnlyOption: // We only recognize this option on v6 endpoints. if e.NetProto != header.IPv6ProtocolNumber { @@ -1424,7 +1474,11 @@ func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { return tcpip.ErrInvalidEndpointState } + e.LockUser() e.v6only = v + e.UnlockUser() + default: + return tcpip.ErrUnknownProtocolOption } return nil @@ -1432,7 +1486,40 @@ func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { // SetSockOptInt sets a socket option. func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { + // Lower 2 bits represents ECN bits. RFC 3168, section 23.1 + const inetECNMask = 3 + switch opt { + case tcpip.KeepaliveCountOption: + e.keepalive.Lock() + e.keepalive.count = int(v) + e.keepalive.Unlock() + e.notifyProtocolGoroutine(notifyKeepaliveChanged) + + case tcpip.IPv4TOSOption: + e.LockUser() + // TODO(gvisor.dev/issue/995): ECN is not currently supported, + // ignore the bits for now. + e.sendTOS = uint8(v) & ^uint8(inetECNMask) + e.UnlockUser() + + case tcpip.IPv6TrafficClassOption: + e.LockUser() + // TODO(gvisor.dev/issue/995): ECN is not currently supported, + // ignore the bits for now. + e.sendTOS = uint8(v) & ^uint8(inetECNMask) + e.UnlockUser() + + case tcpip.MaxSegOption: + userMSS := v + if userMSS < header.TCPMinimumMSS || userMSS > header.TCPMaximumMSS { + return tcpip.ErrInvalidOptionValue + } + e.LockUser() + e.userMSS = uint16(userMSS) + e.UnlockUser() + e.notifyProtocolGoroutine(notifyMSSChanged) + case tcpip.ReceiveBufferSizeOption: // Make sure the receive buffer size is within the min and max // allowed. @@ -1483,7 +1570,6 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { e.rcvListMu.Unlock() e.UnlockUser() e.notifyProtocolGoroutine(mask) - return nil case tcpip.SendBufferSizeOption: // Make sure the send buffer size is within the min and max @@ -1502,52 +1588,21 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { e.sndBufMu.Lock() e.sndBufSize = size e.sndBufMu.Unlock() - return nil - case tcpip.DelayOption: - if v == 0 { - atomic.StoreUint32(&e.delay, 0) - - // Handle delayed data. - e.sndWaker.Assert() - } else { - atomic.StoreUint32(&e.delay, 1) - } - return nil + case tcpip.TTLOption: + e.LockUser() + e.ttl = uint8(v) + e.UnlockUser() default: - return nil + return tcpip.ErrUnknownProtocolOption } + return nil } // SetSockOpt sets a socket option. func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { - // Lower 2 bits represents ECN bits. RFC 3168, section 23.1 - const inetECNMask = 3 switch v := opt.(type) { - case tcpip.CorkOption: - if v == 0 { - atomic.StoreUint32(&e.cork, 0) - - // Handle the corked data. - e.sndWaker.Assert() - } else { - atomic.StoreUint32(&e.cork, 1) - } - return nil - - case tcpip.ReuseAddressOption: - e.LockUser() - e.reuseAddr = v != 0 - e.UnlockUser() - return nil - - case tcpip.ReusePortOption: - e.LockUser() - e.reusePort = v != 0 - e.UnlockUser() - return nil - case tcpip.BindToDeviceOption: id := tcpip.NICID(v) if id != 0 && !e.stack.HasNIC(id) { @@ -1556,72 +1611,26 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.LockUser() e.bindToDevice = id e.UnlockUser() - return nil - - case tcpip.QuickAckOption: - if v == 0 { - atomic.StoreUint32(&e.slowAck, 1) - } else { - atomic.StoreUint32(&e.slowAck, 0) - } - return nil - - case tcpip.MaxSegOption: - userMSS := v - if userMSS < header.TCPMinimumMSS || userMSS > header.TCPMaximumMSS { - return tcpip.ErrInvalidOptionValue - } - e.LockUser() - e.userMSS = uint16(userMSS) - e.UnlockUser() - e.notifyProtocolGoroutine(notifyMSSChanged) - return nil - - case tcpip.TTLOption: - e.LockUser() - e.ttl = uint8(v) - e.UnlockUser() - return nil - - case tcpip.KeepaliveEnabledOption: - e.keepalive.Lock() - e.keepalive.enabled = v != 0 - e.keepalive.Unlock() - e.notifyProtocolGoroutine(notifyKeepaliveChanged) - return nil case tcpip.KeepaliveIdleOption: e.keepalive.Lock() e.keepalive.idle = time.Duration(v) e.keepalive.Unlock() e.notifyProtocolGoroutine(notifyKeepaliveChanged) - return nil case tcpip.KeepaliveIntervalOption: e.keepalive.Lock() e.keepalive.interval = time.Duration(v) e.keepalive.Unlock() e.notifyProtocolGoroutine(notifyKeepaliveChanged) - return nil - case tcpip.KeepaliveCountOption: - e.keepalive.Lock() - e.keepalive.count = int(v) - e.keepalive.Unlock() - e.notifyProtocolGoroutine(notifyKeepaliveChanged) - return nil + case tcpip.OutOfBandInlineOption: + // We don't currently support disabling this option. case tcpip.TCPUserTimeoutOption: e.LockUser() e.userTimeout = time.Duration(v) e.UnlockUser() - return nil - - case tcpip.BroadcastOption: - e.LockUser() - e.broadcast = v != 0 - e.UnlockUser() - return nil case tcpip.CongestionControlOption: // Query the available cc algorithms in the stack and @@ -1652,22 +1661,6 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { // control algorithm is specified. return tcpip.ErrNoSuchFile - case tcpip.IPv4TOSOption: - e.LockUser() - // TODO(gvisor.dev/issue/995): ECN is not currently supported, - // ignore the bits for now. - e.sendTOS = uint8(v) & ^uint8(inetECNMask) - e.UnlockUser() - return nil - - case tcpip.IPv6TrafficClassOption: - e.LockUser() - // TODO(gvisor.dev/issue/995): ECN is not currently supported, - // ignore the bits for now. - e.sendTOS = uint8(v) & ^uint8(inetECNMask) - e.UnlockUser() - return nil - case tcpip.TCPLingerTimeoutOption: e.LockUser() if v < 0 { @@ -1688,7 +1681,6 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { } e.tcpLingerTimeout = time.Duration(v) e.UnlockUser() - return nil case tcpip.TCPDeferAcceptOption: e.LockUser() @@ -1697,11 +1689,11 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { } e.deferAccept = time.Duration(v) e.UnlockUser() - return nil default: return nil } + return nil } // readyReceiveSize returns the number of bytes ready to be received. @@ -1723,6 +1715,43 @@ func (e *endpoint) readyReceiveSize() (int, *tcpip.Error) { // GetSockOptBool implements tcpip.Endpoint.GetSockOptBool. func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { switch opt { + case tcpip.BroadcastOption: + e.LockUser() + v := e.broadcast + e.UnlockUser() + return v, nil + + case tcpip.CorkOption: + return atomic.LoadUint32(&e.cork) != 0, nil + + case tcpip.DelayOption: + return atomic.LoadUint32(&e.delay) != 0, nil + + case tcpip.KeepaliveEnabledOption: + e.keepalive.Lock() + v := e.keepalive.enabled + e.keepalive.Unlock() + + return v, nil + + case tcpip.QuickAckOption: + v := atomic.LoadUint32(&e.slowAck) == 0 + return v, nil + + case tcpip.ReuseAddressOption: + e.LockUser() + v := e.reuseAddr + e.UnlockUser() + + return v, nil + + case tcpip.ReusePortOption: + e.LockUser() + v := e.reusePort + e.UnlockUser() + + return v, nil + case tcpip.V6OnlyOption: // We only recognize this option on v6 endpoints. if e.NetProto != header.IPv6ProtocolNumber { @@ -1734,14 +1763,41 @@ func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { e.UnlockUser() return v, nil - } - return false, tcpip.ErrUnknownProtocolOption + default: + return false, tcpip.ErrUnknownProtocolOption + } } // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { switch opt { + case tcpip.KeepaliveCountOption: + e.keepalive.Lock() + v := e.keepalive.count + e.keepalive.Unlock() + return v, nil + + case tcpip.IPv4TOSOption: + e.LockUser() + v := int(e.sendTOS) + e.UnlockUser() + return v, nil + + case tcpip.IPv6TrafficClassOption: + e.LockUser() + v := int(e.sendTOS) + e.UnlockUser() + return v, nil + + case tcpip.MaxSegOption: + // This is just stubbed out. Linux never returns the user_mss + // value as it either returns the defaultMSS or returns the + // actual current MSS. Netstack just returns the defaultMSS + // always for now. + v := header.TCPDefaultMSS + return v, nil + case tcpip.ReceiveQueueSizeOption: return e.readyReceiveSize() @@ -1757,12 +1813,11 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { e.rcvListMu.Unlock() return v, nil - case tcpip.DelayOption: - var o int - if v := atomic.LoadUint32(&e.delay); v != 0 { - o = 1 - } - return o, nil + case tcpip.TTLOption: + e.LockUser() + v := int(e.ttl) + e.UnlockUser() + return v, nil default: return -1, tcpip.ErrUnknownProtocolOption @@ -1779,61 +1834,10 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { e.lastErrorMu.Unlock() return err - case *tcpip.MaxSegOption: - // This is just stubbed out. Linux never returns the user_mss - // value as it either returns the defaultMSS or returns the - // actual current MSS. Netstack just returns the defaultMSS - // always for now. - *o = header.TCPDefaultMSS - return nil - - case *tcpip.CorkOption: - *o = 0 - if v := atomic.LoadUint32(&e.cork); v != 0 { - *o = 1 - } - return nil - - case *tcpip.ReuseAddressOption: - e.LockUser() - v := e.reuseAddr - e.UnlockUser() - - *o = 0 - if v { - *o = 1 - } - return nil - - case *tcpip.ReusePortOption: - e.LockUser() - v := e.reusePort - e.UnlockUser() - - *o = 0 - if v { - *o = 1 - } - return nil - case *tcpip.BindToDeviceOption: e.LockUser() *o = tcpip.BindToDeviceOption(e.bindToDevice) e.UnlockUser() - return nil - - case *tcpip.QuickAckOption: - *o = 1 - if v := atomic.LoadUint32(&e.slowAck); v != 0 { - *o = 0 - } - return nil - - case *tcpip.TTLOption: - e.LockUser() - *o = tcpip.TTLOption(e.ttl) - e.UnlockUser() - return nil case *tcpip.TCPInfoOption: *o = tcpip.TCPInfoOption{} @@ -1846,92 +1850,45 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { o.RTTVar = snd.rtt.rttvar snd.rtt.Unlock() } - return nil - - case *tcpip.KeepaliveEnabledOption: - e.keepalive.Lock() - v := e.keepalive.enabled - e.keepalive.Unlock() - - *o = 0 - if v { - *o = 1 - } - return nil case *tcpip.KeepaliveIdleOption: e.keepalive.Lock() *o = tcpip.KeepaliveIdleOption(e.keepalive.idle) e.keepalive.Unlock() - return nil case *tcpip.KeepaliveIntervalOption: e.keepalive.Lock() *o = tcpip.KeepaliveIntervalOption(e.keepalive.interval) e.keepalive.Unlock() - return nil - - case *tcpip.KeepaliveCountOption: - e.keepalive.Lock() - *o = tcpip.KeepaliveCountOption(e.keepalive.count) - e.keepalive.Unlock() - return nil case *tcpip.TCPUserTimeoutOption: e.LockUser() *o = tcpip.TCPUserTimeoutOption(e.userTimeout) e.UnlockUser() - return nil case *tcpip.OutOfBandInlineOption: // We don't currently support disabling this option. *o = 1 - return nil - - case *tcpip.BroadcastOption: - e.LockUser() - v := e.broadcast - e.UnlockUser() - - *o = 0 - if v { - *o = 1 - } - return nil case *tcpip.CongestionControlOption: e.LockUser() *o = e.cc e.UnlockUser() - return nil - - case *tcpip.IPv4TOSOption: - e.LockUser() - *o = tcpip.IPv4TOSOption(e.sendTOS) - e.UnlockUser() - return nil - - case *tcpip.IPv6TrafficClassOption: - e.LockUser() - *o = tcpip.IPv6TrafficClassOption(e.sendTOS) - e.UnlockUser() - return nil case *tcpip.TCPLingerTimeoutOption: e.LockUser() *o = tcpip.TCPLingerTimeoutOption(e.tcpLingerTimeout) e.UnlockUser() - return nil case *tcpip.TCPDeferAcceptOption: e.LockUser() *o = tcpip.TCPDeferAcceptOption(e.deferAccept) e.UnlockUser() - return nil default: return tcpip.ErrUnknownProtocolOption } + return nil } // checkV4MappedLocked determines the effective network protocol and converts diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index ce3df7478..32d0af6c4 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -728,7 +728,7 @@ func TestUserSuppliedMSSOnConnectV4(t *testing.T) { const maxMSS = mtu - header.IPv4MinimumSize - header.TCPMinimumSize tests := []struct { name string - setMSS uint16 + setMSS int expMSS uint16 }{ { @@ -756,15 +756,14 @@ func TestUserSuppliedMSSOnConnectV4(t *testing.T) { c.Create(-1) // Set the MSS socket option. - opt := tcpip.MaxSegOption(test.setMSS) - if err := c.EP.SetSockOpt(opt); err != nil { - t.Fatalf("SetSockOpt(%#v) failed: %s", opt, err) + if err := c.EP.SetSockOptInt(tcpip.MaxSegOption, test.setMSS); err != nil { + t.Fatalf("SetSockOptInt(MaxSegOption, %d) failed: %s", test.setMSS, err) } // Get expected window size. rcvBufSize, err := c.EP.GetSockOptInt(tcpip.ReceiveBufferSizeOption) if err != nil { - t.Fatalf("GetSockOpt(%v) failed: %s", tcpip.ReceiveBufferSizeOption, err) + t.Fatalf("GetSockOptInt(ReceiveBufferSizeOption) failed: %s", err) } ws := tcp.FindWndScale(seqnum.Size(rcvBufSize)) @@ -818,15 +817,14 @@ func TestUserSuppliedMSSOnConnectV6(t *testing.T) { c.CreateV6Endpoint(true) // Set the MSS socket option. - opt := tcpip.MaxSegOption(test.setMSS) - if err := c.EP.SetSockOpt(opt); err != nil { - t.Fatalf("SetSockOpt(%#v) failed: %s", opt, err) + if err := c.EP.SetSockOptInt(tcpip.MaxSegOption, int(test.setMSS)); err != nil { + t.Fatalf("SetSockOptInt(MaxSegOption, %d) failed: %s", test.setMSS, err) } // Get expected window size. rcvBufSize, err := c.EP.GetSockOptInt(tcpip.ReceiveBufferSizeOption) if err != nil { - t.Fatalf("GetSockOpt(%v) failed: %s", tcpip.ReceiveBufferSizeOption, err) + t.Fatalf("GetSockOptInt(ReceiveBufferSizeOption) failed: %s", err) } ws := tcp.FindWndScale(seqnum.Size(rcvBufSize)) @@ -1077,17 +1075,17 @@ func TestTOSV4(t *testing.T) { c.EP = ep const tos = 0xC0 - if err := c.EP.SetSockOpt(tcpip.IPv4TOSOption(tos)); err != nil { - t.Errorf("SetSockOpt(%#v) failed: %s", tcpip.IPv4TOSOption(tos), err) + if err := c.EP.SetSockOptInt(tcpip.IPv4TOSOption, tos); err != nil { + t.Errorf("SetSockOptInt(IPv4TOSOption, %d) failed: %s", tos, err) } - var v tcpip.IPv4TOSOption - if err := c.EP.GetSockOpt(&v); err != nil { - t.Errorf("GetSockopt failed: %s", err) + v, err := c.EP.GetSockOptInt(tcpip.IPv4TOSOption) + if err != nil { + t.Errorf("GetSockoptInt(IPv4TOSOption) failed: %s", err) } - if want := tcpip.IPv4TOSOption(tos); v != want { - t.Errorf("got GetSockOpt(...) = %#v, want = %#v", v, want) + if v != tos { + t.Errorf("got GetSockOptInt(IPv4TOSOption) = %d, want = %d", v, tos) } testV4Connect(t, c, checker.TOS(tos, 0)) @@ -1125,17 +1123,17 @@ func TestTrafficClassV6(t *testing.T) { c.CreateV6Endpoint(false) const tos = 0xC0 - if err := c.EP.SetSockOpt(tcpip.IPv6TrafficClassOption(tos)); err != nil { - t.Errorf("SetSockOpt(%#v) failed: %s", tcpip.IPv6TrafficClassOption(tos), err) + if err := c.EP.SetSockOptInt(tcpip.IPv6TrafficClassOption, tos); err != nil { + t.Errorf("SetSockOpInt(IPv6TrafficClassOption, %d) failed: %s", tos, err) } - var v tcpip.IPv6TrafficClassOption - if err := c.EP.GetSockOpt(&v); err != nil { - t.Fatalf("GetSockopt failed: %s", err) + v, err := c.EP.GetSockOptInt(tcpip.IPv6TrafficClassOption) + if err != nil { + t.Fatalf("GetSockoptInt(IPv6TrafficClassOption) failed: %s", err) } - if want := tcpip.IPv6TrafficClassOption(tos); v != want { - t.Errorf("got GetSockOpt(...) = %#v, want = %#v", v, want) + if v != tos { + t.Errorf("got GetSockOptInt(IPv6TrafficClassOption) = %d, want = %d", v, tos) } // Test the connection request. @@ -1711,7 +1709,7 @@ func TestNoWindowShrinking(t *testing.T) { c.CreateConnected(789, 30000, 10) if err := c.EP.SetSockOptInt(tcpip.ReceiveBufferSizeOption, 5); err != nil { - t.Fatalf("SetSockOpt failed: %v", err) + t.Fatalf("SetSockOptInt(ReceiveBufferSizeOption, 5) failed: %v", err) } we, ch := waiter.NewChannelEntry(nil) @@ -1984,7 +1982,7 @@ func TestScaledWindowAccept(t *testing.T) { // Set the window size greater than the maximum non-scaled window. if err := ep.SetSockOptInt(tcpip.ReceiveBufferSizeOption, 65535*3); err != nil { - t.Fatalf("SetSockOpt failed failed: %v", err) + t.Fatalf("SetSockOptInt(ReceiveBufferSizeOption, 65535*3) failed failed: %v", err) } if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { @@ -2057,7 +2055,7 @@ func TestNonScaledWindowAccept(t *testing.T) { // Set the window size greater than the maximum non-scaled window. if err := ep.SetSockOptInt(tcpip.ReceiveBufferSizeOption, 65535*3); err != nil { - t.Fatalf("SetSockOpt failed failed: %v", err) + t.Fatalf("SetSockOptInt(ReceiveBufferSizeOption, 65535*3) failed failed: %v", err) } if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { @@ -2221,10 +2219,10 @@ func TestSegmentMerging(t *testing.T) { { "cork", func(ep tcpip.Endpoint) { - ep.SetSockOpt(tcpip.CorkOption(1)) + ep.SetSockOptBool(tcpip.CorkOption, true) }, func(ep tcpip.Endpoint) { - ep.SetSockOpt(tcpip.CorkOption(0)) + ep.SetSockOptBool(tcpip.CorkOption, false) }, }, } @@ -2316,7 +2314,7 @@ func TestDelay(t *testing.T) { c.CreateConnected(789, 30000, -1 /* epRcvBuf */) - c.EP.SetSockOptInt(tcpip.DelayOption, 1) + c.EP.SetSockOptBool(tcpip.DelayOption, true) var allData []byte for i, data := range [][]byte{{0}, {1, 2, 3, 4}, {5, 6, 7}, {8, 9}, {10}, {11}} { @@ -2364,7 +2362,7 @@ func TestUndelay(t *testing.T) { c.CreateConnected(789, 30000, -1 /* epRcvBuf */) - c.EP.SetSockOptInt(tcpip.DelayOption, 1) + c.EP.SetSockOptBool(tcpip.DelayOption, true) allData := [][]byte{{0}, {1, 2, 3}} for i, data := range allData { @@ -2397,7 +2395,7 @@ func TestUndelay(t *testing.T) { // Check that we don't get the second packet yet. c.CheckNoPacketTimeout("delayed second packet transmitted", 100*time.Millisecond) - c.EP.SetSockOptInt(tcpip.DelayOption, 0) + c.EP.SetSockOptBool(tcpip.DelayOption, false) // Check that data is received. second := c.GetPacket() @@ -2434,8 +2432,8 @@ func TestMSSNotDelayed(t *testing.T) { fn func(tcpip.Endpoint) }{ {"no-op", func(tcpip.Endpoint) {}}, - {"delay", func(ep tcpip.Endpoint) { ep.SetSockOptInt(tcpip.DelayOption, 1) }}, - {"cork", func(ep tcpip.Endpoint) { ep.SetSockOpt(tcpip.CorkOption(1)) }}, + {"delay", func(ep tcpip.Endpoint) { ep.SetSockOptBool(tcpip.DelayOption, true) }}, + {"cork", func(ep tcpip.Endpoint) { ep.SetSockOptBool(tcpip.CorkOption, true) }}, } for _, test := range tests { @@ -2576,12 +2574,12 @@ func TestSetTTL(t *testing.T) { t.Fatalf("NewEndpoint failed: %v", err) } - if err := c.EP.SetSockOpt(tcpip.TTLOption(wantTTL)); err != nil { - t.Fatalf("SetSockOpt failed: %v", err) + if err := c.EP.SetSockOptInt(tcpip.TTLOption, int(wantTTL)); err != nil { + t.Fatalf("SetSockOptInt(TTLOption, %d) failed: %s", wantTTL, err) } if err := c.EP.Connect(tcpip.FullAddress{Addr: context.TestAddr, Port: context.TestPort}); err != tcpip.ErrConnectStarted { - t.Fatalf("Unexpected return value from Connect: %v", err) + t.Fatalf("Unexpected return value from Connect: %s", err) } // Receive SYN packet. @@ -2621,7 +2619,7 @@ func TestPassiveSendMSSLessThanMTU(t *testing.T) { // window scaling option. const rcvBufferSize = 0x20000 if err := ep.SetSockOptInt(tcpip.ReceiveBufferSizeOption, rcvBufferSize); err != nil { - t.Fatalf("SetSockOpt failed failed: %v", err) + t.Fatalf("SetSockOptInt(ReceiveBufferSizeOption, %d) failed failed: %s", rcvBufferSize, err) } if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { @@ -2765,7 +2763,7 @@ func TestSynOptionsOnActiveConnect(t *testing.T) { const rcvBufferSize = 0x20000 const wndScale = 2 if err := c.EP.SetSockOptInt(tcpip.ReceiveBufferSizeOption, rcvBufferSize); err != nil { - t.Fatalf("SetSockOpt failed failed: %v", err) + t.Fatalf("SetSockOptInt(ReceiveBufferSizeOption, %d) failed failed: %s", rcvBufferSize, err) } // Start connection attempt. @@ -3882,26 +3880,26 @@ func TestMinMaxBufferSizes(t *testing.T) { // Set values below the min. if err := ep.SetSockOptInt(tcpip.ReceiveBufferSizeOption, 199); err != nil { - t.Fatalf("GetSockOpt failed: %v", err) + t.Fatalf("SetSockOptInt(ReceiveBufferSizeOption, 199) failed: %s", err) } checkRecvBufferSize(t, ep, 200) if err := ep.SetSockOptInt(tcpip.SendBufferSizeOption, 299); err != nil { - t.Fatalf("GetSockOpt failed: %v", err) + t.Fatalf("SetSockOptInt(SendBufferSizeOption, 299) failed: %s", err) } checkSendBufferSize(t, ep, 300) // Set values above the max. if err := ep.SetSockOptInt(tcpip.ReceiveBufferSizeOption, 1+tcp.DefaultReceiveBufferSize*20); err != nil { - t.Fatalf("GetSockOpt failed: %v", err) + t.Fatalf("SetSockOptInt(ReceiveBufferSizeOption) failed: %s", err) } checkRecvBufferSize(t, ep, tcp.DefaultReceiveBufferSize*20) if err := ep.SetSockOptInt(tcpip.SendBufferSizeOption, 1+tcp.DefaultSendBufferSize*30); err != nil { - t.Fatalf("GetSockOpt failed: %v", err) + t.Fatalf("SetSockOptInt(SendBufferSizeOption) failed: %s", err) } checkSendBufferSize(t, ep, tcp.DefaultSendBufferSize*30) @@ -4147,11 +4145,11 @@ func TestConnectAvoidsBoundPorts(t *testing.T) { case "ipv4": case "ipv6": if err := ep.SetSockOptBool(tcpip.V6OnlyOption, true); err != nil { - t.Fatalf("SetSockOpt(V6OnlyOption(true)) failed: %v", err) + t.Fatalf("SetSockOptBool(V6OnlyOption(true)) failed: %s", err) } case "dual": if err := ep.SetSockOptBool(tcpip.V6OnlyOption, false); err != nil { - t.Fatalf("SetSockOpt(V6OnlyOption(false)) failed: %v", err) + t.Fatalf("SetSockOptBool(V6OnlyOption(false)) failed: %s", err) } default: t.Fatalf("unknown network: '%s'", network) @@ -4477,8 +4475,8 @@ func TestKeepalive(t *testing.T) { const keepAliveInterval = 10 * time.Millisecond c.EP.SetSockOpt(tcpip.KeepaliveIdleOption(10 * time.Millisecond)) c.EP.SetSockOpt(tcpip.KeepaliveIntervalOption(keepAliveInterval)) - c.EP.SetSockOpt(tcpip.KeepaliveCountOption(5)) - c.EP.SetSockOpt(tcpip.KeepaliveEnabledOption(1)) + c.EP.SetSockOptInt(tcpip.KeepaliveCountOption, 5) + c.EP.SetSockOptBool(tcpip.KeepaliveEnabledOption, true) // 5 unacked keepalives are sent. ACK each one, and check that the // connection stays alive after 5. @@ -5770,14 +5768,14 @@ func TestReceiveBufferAutoTuning(t *testing.T) { func TestDelayEnabled(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - checkDelayOption(t, c, false, 0) // Delay is disabled by default. + checkDelayOption(t, c, false, false) // Delay is disabled by default. for _, v := range []struct { delayEnabled tcp.DelayEnabled - wantDelayOption int + wantDelayOption bool }{ - {delayEnabled: false, wantDelayOption: 0}, - {delayEnabled: true, wantDelayOption: 1}, + {delayEnabled: false, wantDelayOption: false}, + {delayEnabled: true, wantDelayOption: true}, } { c := context.New(t, defaultMTU) defer c.Cleanup() @@ -5788,7 +5786,7 @@ func TestDelayEnabled(t *testing.T) { } } -func checkDelayOption(t *testing.T, c *context.Context, wantDelayEnabled tcp.DelayEnabled, wantDelayOption int) { +func checkDelayOption(t *testing.T, c *context.Context, wantDelayEnabled tcp.DelayEnabled, wantDelayOption bool) { t.Helper() var gotDelayEnabled tcp.DelayEnabled @@ -5803,12 +5801,12 @@ func checkDelayOption(t *testing.T, c *context.Context, wantDelayEnabled tcp.Del if err != nil { t.Fatalf("NewEndPoint(tcp, ipv4, new(waiter.Queue)) failed: %v", err) } - gotDelayOption, err := ep.GetSockOptInt(tcpip.DelayOption) + gotDelayOption, err := ep.GetSockOptBool(tcpip.DelayOption) if err != nil { - t.Fatalf("ep.GetSockOptInt(tcpip.DelayOption) failed: %v", err) + t.Fatalf("ep.GetSockOptBool(tcpip.DelayOption) failed: %s", err) } if gotDelayOption != wantDelayOption { - t.Errorf("ep.GetSockOptInt(tcpip.DelayOption) got: %d, want: %d", gotDelayOption, wantDelayOption) + t.Errorf("ep.GetSockOptBool(tcpip.DelayOption) got: %t, want: %t", gotDelayOption, wantDelayOption) } } @@ -6620,8 +6618,8 @@ func TestKeepaliveWithUserTimeout(t *testing.T) { const keepAliveInterval = 10 * time.Millisecond c.EP.SetSockOpt(tcpip.KeepaliveIdleOption(10 * time.Millisecond)) c.EP.SetSockOpt(tcpip.KeepaliveIntervalOption(keepAliveInterval)) - c.EP.SetSockOpt(tcpip.KeepaliveCountOption(10)) - c.EP.SetSockOpt(tcpip.KeepaliveEnabledOption(1)) + c.EP.SetSockOptInt(tcpip.KeepaliveCountOption, 10) + c.EP.SetSockOptBool(tcpip.KeepaliveEnabledOption, true) // Set userTimeout to be the duration for 3 keepalive probes. userTimeout := 30 * time.Millisecond diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 120d3baa3..492cc1fcb 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -501,11 +501,20 @@ func (e *endpoint) Peek([][]byte) (int64, tcpip.ControlMessages, *tcpip.Error) { // SetSockOptBool implements tcpip.Endpoint.SetSockOptBool. func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { switch opt { + case tcpip.BroadcastOption: + e.mu.Lock() + e.broadcast = v + e.mu.Unlock() + + case tcpip.MulticastLoopOption: + e.mu.Lock() + e.multicastLoop = v + e.mu.Unlock() + case tcpip.ReceiveTOSOption: e.mu.Lock() e.receiveTOS = v e.mu.Unlock() - return nil case tcpip.ReceiveTClassOption: // We only support this option on v6 endpoints. @@ -516,7 +525,18 @@ func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { e.mu.Lock() e.receiveTClass = v e.mu.Unlock() - return nil + + case tcpip.ReceiveIPPacketInfoOption: + e.mu.Lock() + e.receiveIPPacketInfo = v + e.mu.Unlock() + + case tcpip.ReuseAddressOption: + + case tcpip.ReusePortOption: + e.mu.Lock() + e.reusePort = v + e.mu.Unlock() case tcpip.V6OnlyOption: // We only recognize this option on v6 endpoints. @@ -533,13 +553,8 @@ func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { } e.v6only = v - return nil - - case tcpip.ReceiveIPPacketInfoOption: - e.mu.Lock() - e.receiveIPPacketInfo = v - e.mu.Unlock() - return nil + default: + return tcpip.ErrUnknownProtocolOption } return nil @@ -547,22 +562,40 @@ func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { // SetSockOptInt implements tcpip.Endpoint.SetSockOptInt. func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { - return nil -} + switch opt { + case tcpip.MulticastTTLOption: + e.mu.Lock() + e.multicastTTL = uint8(v) + e.mu.Unlock() -// SetSockOpt implements tcpip.Endpoint.SetSockOpt. -func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { - switch v := opt.(type) { case tcpip.TTLOption: e.mu.Lock() e.ttl = uint8(v) e.mu.Unlock() - case tcpip.MulticastTTLOption: + case tcpip.IPv4TOSOption: e.mu.Lock() - e.multicastTTL = uint8(v) + e.sendTOS = uint8(v) + e.mu.Unlock() + + case tcpip.IPv6TrafficClassOption: + e.mu.Lock() + e.sendTOS = uint8(v) e.mu.Unlock() + case tcpip.ReceiveBufferSizeOption: + case tcpip.SendBufferSizeOption: + + default: + return tcpip.ErrUnknownProtocolOption + } + + return nil +} + +// SetSockOpt implements tcpip.Endpoint.SetSockOpt. +func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { + switch v := opt.(type) { case tcpip.MulticastInterfaceOption: e.mu.Lock() defer e.mu.Unlock() @@ -686,16 +719,6 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.multicastMemberships[memToRemoveIndex] = e.multicastMemberships[len(e.multicastMemberships)-1] e.multicastMemberships = e.multicastMemberships[:len(e.multicastMemberships)-1] - case tcpip.MulticastLoopOption: - e.mu.Lock() - e.multicastLoop = bool(v) - e.mu.Unlock() - - case tcpip.ReusePortOption: - e.mu.Lock() - e.reusePort = v != 0 - e.mu.Unlock() - case tcpip.BindToDeviceOption: id := tcpip.NICID(v) if id != 0 && !e.stack.HasNIC(id) { @@ -704,26 +727,6 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.mu.Lock() e.bindToDevice = id e.mu.Unlock() - return nil - - case tcpip.BroadcastOption: - e.mu.Lock() - e.broadcast = v != 0 - e.mu.Unlock() - - return nil - - case tcpip.IPv4TOSOption: - e.mu.Lock() - e.sendTOS = uint8(v) - e.mu.Unlock() - return nil - - case tcpip.IPv6TrafficClassOption: - e.mu.Lock() - e.sendTOS = uint8(v) - e.mu.Unlock() - return nil } return nil } @@ -731,6 +734,21 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { // GetSockOptBool implements tcpip.Endpoint.GetSockOptBool. func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { switch opt { + case tcpip.BroadcastOption: + e.mu.RLock() + v := e.broadcast + e.mu.RUnlock() + return v, nil + + case tcpip.KeepaliveEnabledOption: + return false, nil + + case tcpip.MulticastLoopOption: + e.mu.RLock() + v := e.multicastLoop + e.mu.RUnlock() + return v, nil + case tcpip.ReceiveTOSOption: e.mu.RLock() v := e.receiveTOS @@ -748,6 +766,22 @@ func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { e.mu.RUnlock() return v, nil + case tcpip.ReceiveIPPacketInfoOption: + e.mu.RLock() + v := e.receiveIPPacketInfo + e.mu.RUnlock() + return v, nil + + case tcpip.ReuseAddressOption: + return false, nil + + case tcpip.ReusePortOption: + e.mu.RLock() + v := e.reusePort + e.mu.RUnlock() + + return v, nil + case tcpip.V6OnlyOption: // We only recognize this option on v6 endpoints. if e.NetProto != header.IPv6ProtocolNumber { @@ -760,19 +794,32 @@ func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { return v, nil - case tcpip.ReceiveIPPacketInfoOption: - e.mu.RLock() - v := e.receiveIPPacketInfo - e.mu.RUnlock() - return v, nil + default: + return false, tcpip.ErrUnknownProtocolOption } - - return false, tcpip.ErrUnknownProtocolOption } // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { switch opt { + case tcpip.IPv4TOSOption: + e.mu.RLock() + v := int(e.sendTOS) + e.mu.RUnlock() + return v, nil + + case tcpip.IPv6TrafficClassOption: + e.mu.RLock() + v := int(e.sendTOS) + e.mu.RUnlock() + return v, nil + + case tcpip.MulticastTTLOption: + e.mu.Lock() + v := int(e.multicastTTL) + e.mu.Unlock() + return v, nil + case tcpip.ReceiveQueueSizeOption: v := 0 e.rcvMu.Lock() @@ -794,29 +841,22 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { v := e.rcvBufSizeMax e.rcvMu.Unlock() return v, nil - } - return -1, tcpip.ErrUnknownProtocolOption + case tcpip.TTLOption: + e.mu.Lock() + v := int(e.ttl) + e.mu.Unlock() + return v, nil + + default: + return -1, tcpip.ErrUnknownProtocolOption + } } // GetSockOpt implements tcpip.Endpoint.GetSockOpt. func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { switch o := opt.(type) { case tcpip.ErrorOption: - return nil - - case *tcpip.TTLOption: - e.mu.Lock() - *o = tcpip.TTLOption(e.ttl) - e.mu.Unlock() - return nil - - case *tcpip.MulticastTTLOption: - e.mu.Lock() - *o = tcpip.MulticastTTLOption(e.multicastTTL) - e.mu.Unlock() - return nil - case *tcpip.MulticastInterfaceOption: e.mu.Lock() *o = tcpip.MulticastInterfaceOption{ @@ -824,67 +864,16 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { e.multicastAddr, } e.mu.Unlock() - return nil - - case *tcpip.MulticastLoopOption: - e.mu.RLock() - v := e.multicastLoop - e.mu.RUnlock() - - *o = tcpip.MulticastLoopOption(v) - return nil - - case *tcpip.ReuseAddressOption: - *o = 0 - return nil - - case *tcpip.ReusePortOption: - e.mu.RLock() - v := e.reusePort - e.mu.RUnlock() - - *o = 0 - if v { - *o = 1 - } - return nil case *tcpip.BindToDeviceOption: e.mu.RLock() *o = tcpip.BindToDeviceOption(e.bindToDevice) e.mu.RUnlock() - return nil - - case *tcpip.KeepaliveEnabledOption: - *o = 0 - return nil - - case *tcpip.BroadcastOption: - e.mu.RLock() - v := e.broadcast - e.mu.RUnlock() - - *o = 0 - if v { - *o = 1 - } - return nil - - case *tcpip.IPv4TOSOption: - e.mu.RLock() - *o = tcpip.IPv4TOSOption(e.sendTOS) - e.mu.RUnlock() - return nil - - case *tcpip.IPv6TrafficClassOption: - e.mu.RLock() - *o = tcpip.IPv6TrafficClassOption(e.sendTOS) - e.mu.RUnlock() - return nil default: return tcpip.ErrUnknownProtocolOption } + return nil } // sendUDP sends a UDP segment via the provided network endpoint and under the diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index 0905726c1..b3ee688b7 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -343,11 +343,11 @@ func (c *testContext) createEndpointForFlow(flow testFlow) { c.createEndpoint(flow.sockProto()) if flow.isV6Only() { if err := c.ep.SetSockOptBool(tcpip.V6OnlyOption, true); err != nil { - c.t.Fatalf("SetSockOpt failed: %v", err) + c.t.Fatalf("SetSockOptBool failed: %s", err) } } else if flow.isBroadcast() { - if err := c.ep.SetSockOpt(tcpip.BroadcastOption(1)); err != nil { - c.t.Fatal("SetSockOpt failed:", err) + if err := c.ep.SetSockOptBool(tcpip.BroadcastOption, true); err != nil { + c.t.Fatalf("SetSockOptBool failed: %s", err) } } } @@ -1271,8 +1271,8 @@ func TestTTL(t *testing.T) { c.createEndpointForFlow(flow) const multicastTTL = 42 - if err := c.ep.SetSockOpt(tcpip.MulticastTTLOption(multicastTTL)); err != nil { - c.t.Fatalf("SetSockOpt failed: %v", err) + if err := c.ep.SetSockOptInt(tcpip.MulticastTTLOption, multicastTTL); err != nil { + c.t.Fatalf("SetSockOptInt failed: %s", err) } var wantTTL uint8 @@ -1311,8 +1311,8 @@ func TestSetTTL(t *testing.T) { c.createEndpointForFlow(flow) - if err := c.ep.SetSockOpt(tcpip.TTLOption(wantTTL)); err != nil { - c.t.Fatalf("SetSockOpt failed: %v", err) + if err := c.ep.SetSockOptInt(tcpip.TTLOption, int(wantTTL)); err != nil { + c.t.Fatalf("SetSockOptInt(TTLOption, %d) failed: %s", wantTTL, err) } var p stack.NetworkProtocol @@ -1346,25 +1346,26 @@ func TestSetTOS(t *testing.T) { c.createEndpointForFlow(flow) const tos = testTOS - var v tcpip.IPv4TOSOption - if err := c.ep.GetSockOpt(&v); err != nil { - c.t.Errorf("GetSockopt(%T) failed: %s", v, err) + v, err := c.ep.GetSockOptInt(tcpip.IPv4TOSOption) + if err != nil { + c.t.Errorf("GetSockOptInt(IPv4TOSOption) failed: %s", err) } // Test for expected default value. if v != 0 { - c.t.Errorf("got GetSockOpt(%T) = 0x%x, want = 0x%x", v, v, 0) + c.t.Errorf("got GetSockOpt(IPv4TOSOption) = 0x%x, want = 0x%x", v, 0) } - if err := c.ep.SetSockOpt(tcpip.IPv4TOSOption(tos)); err != nil { - c.t.Errorf("SetSockOpt(%T, 0x%x) failed: %s", v, tcpip.IPv4TOSOption(tos), err) + if err := c.ep.SetSockOptInt(tcpip.IPv4TOSOption, tos); err != nil { + c.t.Errorf("SetSockOptInt(IPv4TOSOption, 0x%x) failed: %s", tos, err) } - if err := c.ep.GetSockOpt(&v); err != nil { - c.t.Errorf("GetSockopt(%T) failed: %s", v, err) + v, err = c.ep.GetSockOptInt(tcpip.IPv4TOSOption) + if err != nil { + c.t.Errorf("GetSockOptInt(IPv4TOSOption) failed: %s", err) } - if want := tcpip.IPv4TOSOption(tos); v != want { - c.t.Errorf("got GetSockOpt(%T) = 0x%x, want = 0x%x", v, v, want) + if v != tos { + c.t.Errorf("got GetSockOptInt(IPv4TOSOption) = 0x%x, want = 0x%x", v, tos) } testWrite(c, flow, checker.TOS(tos, 0)) @@ -1381,25 +1382,26 @@ func TestSetTClass(t *testing.T) { c.createEndpointForFlow(flow) const tClass = testTOS - var v tcpip.IPv6TrafficClassOption - if err := c.ep.GetSockOpt(&v); err != nil { - c.t.Errorf("GetSockopt(%T) failed: %s", v, err) + v, err := c.ep.GetSockOptInt(tcpip.IPv6TrafficClassOption) + if err != nil { + c.t.Errorf("GetSockOptInt(IPv6TrafficClassOption) failed: %s", err) } // Test for expected default value. if v != 0 { - c.t.Errorf("got GetSockOpt(%T) = 0x%x, want = 0x%x", v, v, 0) + c.t.Errorf("got GetSockOptInt(IPv6TrafficClassOption) = 0x%x, want = 0x%x", v, 0) } - if err := c.ep.SetSockOpt(tcpip.IPv6TrafficClassOption(tClass)); err != nil { - c.t.Errorf("SetSockOpt(%T, 0x%x) failed: %s", v, tcpip.IPv6TrafficClassOption(tClass), err) + if err := c.ep.SetSockOptInt(tcpip.IPv6TrafficClassOption, tClass); err != nil { + c.t.Errorf("SetSockOptInt(IPv6TrafficClassOption, 0x%x) failed: %s", tClass, err) } - if err := c.ep.GetSockOpt(&v); err != nil { - c.t.Errorf("GetSockopt(%T) failed: %s", v, err) + v, err = c.ep.GetSockOptInt(tcpip.IPv6TrafficClassOption) + if err != nil { + c.t.Errorf("GetSockOptInt(IPv6TrafficClassOption) failed: %s", err) } - if want := tcpip.IPv6TrafficClassOption(tClass); v != want { - c.t.Errorf("got GetSockOpt(%T) = 0x%x, want = 0x%x", v, v, want) + if v != tClass { + c.t.Errorf("got GetSockOptInt(IPv6TrafficClassOption) = 0x%x, want = 0x%x", v, tClass) } // The header getter for TClass is called TOS, so use that checker. @@ -1430,7 +1432,7 @@ func TestReceiveTosTClass(t *testing.T) { // Verify that setting and reading the option works. v, err := c.ep.GetSockOptBool(option) if err != nil { - c.t.Errorf("GetSockoptBool(%s) failed: %s", name, err) + c.t.Errorf("GetSockOptBool(%s) failed: %s", name, err) } // Test for expected default value. if v != false { @@ -1444,7 +1446,7 @@ func TestReceiveTosTClass(t *testing.T) { got, err := c.ep.GetSockOptBool(option) if err != nil { - c.t.Errorf("GetSockoptBool(%s) failed: %s", name, err) + c.t.Errorf("GetSockOptBool(%s) failed: %s", name, err) } if got != want { -- cgit v1.2.3 From 9f87502b4619b60779ce19c41ea0e6bd6582e8e4 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Thu, 9 Apr 2020 16:40:12 -0700 Subject: Remove TODOs from Async IO Block and drain requests in io_destroy(2). Note the reason to create read-only mapping. PiperOrigin-RevId: 305786312 --- pkg/sentry/mm/aio_context.go | 101 ++++++++++++++++++++++++----------- pkg/sentry/mm/aio_context_state.go | 2 +- pkg/sentry/syscalls/linux/sys_aio.go | 34 +++++++++--- test/syscalls/linux/aio.cc | 12 +++-- 4 files changed, 107 insertions(+), 42 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/mm/aio_context.go b/pkg/sentry/mm/aio_context.go index cb29d94b0..379148903 100644 --- a/pkg/sentry/mm/aio_context.go +++ b/pkg/sentry/mm/aio_context.go @@ -59,25 +59,27 @@ func (a *aioManager) newAIOContext(events uint32, id uint64) bool { } a.contexts[id] = &AIOContext{ - done: make(chan struct{}, 1), + requestReady: make(chan struct{}, 1), maxOutstanding: events, } return true } -// destroyAIOContext destroys an asynchronous I/O context. +// destroyAIOContext destroys an asynchronous I/O context. It doesn't wait for +// for pending requests to complete. Returns the destroyed AIOContext so it can +// be drained. // -// False is returned if the context does not exist. -func (a *aioManager) destroyAIOContext(id uint64) bool { +// Nil is returned if the context does not exist. +func (a *aioManager) destroyAIOContext(id uint64) *AIOContext { a.mu.Lock() defer a.mu.Unlock() ctx, ok := a.contexts[id] if !ok { - return false + return nil } delete(a.contexts, id) ctx.destroy() - return true + return ctx } // lookupAIOContext looks up the given context. @@ -102,8 +104,8 @@ type ioResult struct { // // +stateify savable type AIOContext struct { - // done is the notification channel used for all requests. - done chan struct{} `state:"nosave"` + // requestReady is the notification channel used for all requests. + requestReady chan struct{} `state:"nosave"` // mu protects below. mu sync.Mutex `state:"nosave"` @@ -129,8 +131,14 @@ func (ctx *AIOContext) destroy() { ctx.mu.Lock() defer ctx.mu.Unlock() ctx.dead = true - if ctx.outstanding == 0 { - close(ctx.done) + ctx.checkForDone() +} + +// Preconditions: ctx.mu must be held by caller. +func (ctx *AIOContext) checkForDone() { + if ctx.dead && ctx.outstanding == 0 { + close(ctx.requestReady) + ctx.requestReady = nil } } @@ -154,11 +162,12 @@ func (ctx *AIOContext) PopRequest() (interface{}, bool) { // Is there anything ready? if e := ctx.results.Front(); e != nil { - ctx.results.Remove(e) - ctx.outstanding-- - if ctx.outstanding == 0 && ctx.dead { - close(ctx.done) + if ctx.outstanding == 0 { + panic("AIOContext outstanding is going negative") } + ctx.outstanding-- + ctx.results.Remove(e) + ctx.checkForDone() return e.data, true } return nil, false @@ -172,26 +181,58 @@ func (ctx *AIOContext) FinishRequest(data interface{}) { // Push to the list and notify opportunistically. The channel notify // here is guaranteed to be safe because outstanding must be non-zero. - // The done channel is only closed when outstanding reaches zero. + // The requestReady channel is only closed when outstanding reaches zero. ctx.results.PushBack(&ioResult{data: data}) select { - case ctx.done <- struct{}{}: + case ctx.requestReady <- struct{}{}: default: } } // WaitChannel returns a channel that is notified when an AIO request is -// completed. -// -// The boolean return value indicates whether or not the context is active. -func (ctx *AIOContext) WaitChannel() (chan struct{}, bool) { +// completed. Returns nil if the context is destroyed and there are no more +// outstanding requests. +func (ctx *AIOContext) WaitChannel() chan struct{} { ctx.mu.Lock() defer ctx.mu.Unlock() - if ctx.outstanding == 0 && ctx.dead { - return nil, false + return ctx.requestReady +} + +// Dead returns true if the context has been destroyed. +func (ctx *AIOContext) Dead() bool { + ctx.mu.Lock() + defer ctx.mu.Unlock() + return ctx.dead +} + +// CancelPendingRequest forgets about a request that hasn't yet completed. +func (ctx *AIOContext) CancelPendingRequest() { + ctx.mu.Lock() + defer ctx.mu.Unlock() + + if ctx.outstanding == 0 { + panic("AIOContext outstanding is going negative") } - return ctx.done, true + ctx.outstanding-- + ctx.checkForDone() +} + +// Drain drops all completed requests. Pending requests remain untouched. +func (ctx *AIOContext) Drain() { + ctx.mu.Lock() + defer ctx.mu.Unlock() + + if ctx.outstanding == 0 { + return + } + size := uint32(ctx.results.Len()) + if ctx.outstanding < size { + panic("AIOContext outstanding is going negative") + } + ctx.outstanding -= size + ctx.results.Reset() + ctx.checkForDone() } // aioMappable implements memmap.MappingIdentity and memmap.Mappable for AIO @@ -332,9 +373,9 @@ func (mm *MemoryManager) NewAIOContext(ctx context.Context, events uint32) (uint Length: aioRingBufferSize, MappingIdentity: m, Mappable: m, - // TODO(fvoznika): Linux does "do_mmap_pgoff(..., PROT_READ | - // PROT_WRITE, ...)" in fs/aio.c:aio_setup_ring(); why do we make this - // mapping read-only? + // Linux uses "do_mmap_pgoff(..., PROT_READ | PROT_WRITE, ...)" in + // fs/aio.c:aio_setup_ring(). Since we don't implement AIO_RING_MAGIC, + // user mode should not write to this page. Perms: usermem.Read, MaxPerms: usermem.Read, }) @@ -349,11 +390,11 @@ func (mm *MemoryManager) NewAIOContext(ctx context.Context, events uint32) (uint return id, nil } -// DestroyAIOContext destroys an asynchronous I/O context. It returns false if -// the context does not exist. -func (mm *MemoryManager) DestroyAIOContext(ctx context.Context, id uint64) bool { +// DestroyAIOContext destroys an asynchronous I/O context. It returns the +// destroyed context. nil if the context does not exist. +func (mm *MemoryManager) DestroyAIOContext(ctx context.Context, id uint64) *AIOContext { if _, ok := mm.LookupAIOContext(ctx, id); !ok { - return false + return nil } // Only unmaps after it assured that the address is a valid aio context to diff --git a/pkg/sentry/mm/aio_context_state.go b/pkg/sentry/mm/aio_context_state.go index c37fc9f7b..3dabac1af 100644 --- a/pkg/sentry/mm/aio_context_state.go +++ b/pkg/sentry/mm/aio_context_state.go @@ -16,5 +16,5 @@ package mm // afterLoad is invoked by stateify. func (a *AIOContext) afterLoad() { - a.done = make(chan struct{}, 1) + a.requestReady = make(chan struct{}, 1) } diff --git a/pkg/sentry/syscalls/linux/sys_aio.go b/pkg/sentry/syscalls/linux/sys_aio.go index b401978db..38cbeba5a 100644 --- a/pkg/sentry/syscalls/linux/sys_aio.go +++ b/pkg/sentry/syscalls/linux/sys_aio.go @@ -114,14 +114,28 @@ func IoSetup(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca func IoDestroy(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { id := args[0].Uint64() - // Destroy the given context. - if !t.MemoryManager().DestroyAIOContext(t, id) { + ctx := t.MemoryManager().DestroyAIOContext(t, id) + if ctx == nil { // Does not exist. return 0, nil, syserror.EINVAL } - // FIXME(fvoznika): Linux blocks until all AIO to the destroyed context is - // done. - return 0, nil, nil + + // Drain completed requests amd wait for pending requests until there are no + // more. + for { + ctx.Drain() + + ch := ctx.WaitChannel() + if ch == nil { + // No more requests, we're done. + return 0, nil, nil + } + // The task cannot be interrupted during the wait. Equivalent to + // TASK_UNINTERRUPTIBLE in Linux. + t.UninterruptibleSleepStart(true /* deactivate */) + <-ch + t.UninterruptibleSleepFinish(true /* activate */) + } } // IoGetevents implements linux syscall io_getevents(2). @@ -200,13 +214,13 @@ func IoGetevents(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S func waitForRequest(ctx *mm.AIOContext, t *kernel.Task, haveDeadline bool, deadline ktime.Time) (interface{}, error) { for { if v, ok := ctx.PopRequest(); ok { - // Request was readly available. Just return it. + // Request was readily available. Just return it. return v, nil } // Need to wait for request completion. - done, active := ctx.WaitChannel() - if !active { + done := ctx.WaitChannel() + if done == nil { // Context has been destroyed. return nil, syserror.EINVAL } @@ -248,6 +262,10 @@ func memoryFor(t *kernel.Task, cb *ioCallback) (usermem.IOSequence, error) { } func performCallback(t *kernel.Task, file *fs.File, cbAddr usermem.Addr, cb *ioCallback, ioseq usermem.IOSequence, ctx *mm.AIOContext, eventFile *fs.File) { + if ctx.Dead() { + ctx.CancelPendingRequest() + return + } ev := &ioEvent{ Data: cb.Data, Obj: uint64(cbAddr), diff --git a/test/syscalls/linux/aio.cc b/test/syscalls/linux/aio.cc index a33daff17..806d5729e 100644 --- a/test/syscalls/linux/aio.cc +++ b/test/syscalls/linux/aio.cc @@ -89,6 +89,7 @@ class AIOTest : public FileTest { FileTest::TearDown(); if (ctx_ != 0) { ASSERT_THAT(DestroyContext(), SyscallSucceeds()); + ctx_ = 0; } } @@ -188,14 +189,19 @@ TEST_F(AIOTest, BadWrite) { } TEST_F(AIOTest, ExitWithPendingIo) { - // Setup a context that is 5 entries deep. - ASSERT_THAT(SetupContext(5), SyscallSucceeds()); + // Setup a context that is 100 entries deep. + ASSERT_THAT(SetupContext(100), SyscallSucceeds()); struct iocb cb = CreateCallback(); struct iocb* cbs[] = {&cb}; // Submit a request but don't complete it to make it pending. - EXPECT_THAT(Submit(1, cbs), SyscallSucceeds()); + for (int i = 0; i < 100; ++i) { + EXPECT_THAT(Submit(1, cbs), SyscallSucceeds()); + } + + ASSERT_THAT(DestroyContext(), SyscallSucceeds()); + ctx_ = 0; } int Submitter(void* arg) { -- cgit v1.2.3 From 257225c34b81ff0d0b5ce8ae333f5905f9e86cce Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Thu, 9 Apr 2020 17:29:43 -0700 Subject: Downgrade VFS1-specific FIXME to a NOTE. PiperOrigin-RevId: 305794509 --- pkg/sentry/fs/dirent.go | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/dirent.go b/pkg/sentry/fs/dirent.go index 0266a5287..65be12175 100644 --- a/pkg/sentry/fs/dirent.go +++ b/pkg/sentry/fs/dirent.go @@ -312,9 +312,9 @@ func (d *Dirent) SyncAll(ctx context.Context) { // There is nothing to sync for a read-only filesystem. if !d.Inode.MountSource.Flags.ReadOnly { - // FIXME(b/34856369): This should be a mount traversal, not a - // Dirent traversal, because some Inodes that need to be synced - // may no longer be reachable by name (after sys_unlink). + // NOTE(b/34856369): This should be a mount traversal, not a Dirent + // traversal, because some Inodes that need to be synced may no longer + // be reachable by name (after sys_unlink). // // Write out metadata, dirty page cached pages, and sync disk/remote // caches. -- cgit v1.2.3 From c9195349c9ac24ccb538e92b308225dfa4184c42 Mon Sep 17 00:00:00 2001 From: Ting-Yu Wang Date: Thu, 9 Apr 2020 17:59:30 -0700 Subject: Replace type assertion with TaskFromContext. This should fix panic at aio callback. PiperOrigin-RevId: 305798549 --- pkg/sentry/socket/netstack/netstack.go | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 20e3fa0d2..7ac38764d 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -535,7 +535,7 @@ func (s *SocketOperations) Write(ctx context.Context, _ *fs.File, src usermem.IO } if resCh != nil { - t := ctx.(*kernel.Task) + t := kernel.TaskFromContext(ctx) if err := t.Block(resCh); err != nil { return 0, syserr.FromError(err).ToError() } @@ -608,7 +608,7 @@ func (s *SocketOperations) ReadFrom(ctx context.Context, _ *fs.File, r io.Reader } if resCh != nil { - t := ctx.(*kernel.Task) + t := kernel.TaskFromContext(ctx) if err := t.Block(resCh); err != nil { return 0, syserr.FromError(err).ToError() } -- cgit v1.2.3 From 7aa5caae71c29b0be9047a7c156a9daaa435ebb8 Mon Sep 17 00:00:00 2001 From: Haibo Xu Date: Wed, 11 Mar 2020 03:21:34 +0000 Subject: Enable syscall ptrace test on arm64. Signed-off-by: Haibo Xu Change-Id: I5bb8fa7d580d173b1438d6465e1adb442216c8fa --- pkg/sentry/arch/arch.go | 3 +++ pkg/sentry/arch/syscalls_amd64.go | 7 +++++++ pkg/sentry/arch/syscalls_arm64.go | 13 ++++++++++++- pkg/sentry/kernel/task_syscall.go | 14 ++++++++++++++ test/syscalls/linux/ptrace.cc | 31 ++++++++++++++++++++++++------- 5 files changed, 60 insertions(+), 8 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/arch/arch.go b/pkg/sentry/arch/arch.go index 1d11cc472..a903d031c 100644 --- a/pkg/sentry/arch/arch.go +++ b/pkg/sentry/arch/arch.go @@ -88,6 +88,9 @@ type Context interface { // SyscallNo returns the syscall number. SyscallNo() uintptr + // SyscallSaveOrig save orignal register value. + SyscallSaveOrig() + // SyscallArgs returns the syscall arguments in an array. SyscallArgs() SyscallArguments diff --git a/pkg/sentry/arch/syscalls_amd64.go b/pkg/sentry/arch/syscalls_amd64.go index 8b4f23007..3859f41ee 100644 --- a/pkg/sentry/arch/syscalls_amd64.go +++ b/pkg/sentry/arch/syscalls_amd64.go @@ -18,6 +18,13 @@ package arch const restartSyscallNr = uintptr(219) +// SyscallSaveOrig save the value of the register which is clobbered in +// syscall handler(doSyscall()). +// +// Noop on x86. +func (c *context64) SyscallSaveOrig() { +} + // SyscallNo returns the syscall number according to the 64-bit convention. func (c *context64) SyscallNo() uintptr { return uintptr(c.Regs.Orig_rax) diff --git a/pkg/sentry/arch/syscalls_arm64.go b/pkg/sentry/arch/syscalls_arm64.go index dc13b6124..92d062513 100644 --- a/pkg/sentry/arch/syscalls_arm64.go +++ b/pkg/sentry/arch/syscalls_arm64.go @@ -18,6 +18,17 @@ package arch const restartSyscallNr = uintptr(128) +// SyscallSaveOrig save the value of the register R0 which is clobbered in +// syscall handler(doSyscall()). +// +// In linux, at the entry of the syscall handler(el0_svc_common()), value of R0 +// is saved to the pt_regs.orig_x0 in kernel code. But currently, the orig_x0 +// was not accessible to the user space application, so we have to do the same +// operation in the sentry code to save the R0 value into the App context. +func (c *context64) SyscallSaveOrig() { + c.OrigR0 = c.Regs.Regs[0] +} + // SyscallNo returns the syscall number according to the 64-bit convention. func (c *context64) SyscallNo() uintptr { return uintptr(c.Regs.Regs[8]) @@ -40,7 +51,7 @@ func (c *context64) SyscallNo() uintptr { // R30: the link register. func (c *context64) SyscallArgs() SyscallArguments { return SyscallArguments{ - SyscallArgument{Value: uintptr(c.Regs.Regs[0])}, + SyscallArgument{Value: uintptr(c.OrigR0)}, SyscallArgument{Value: uintptr(c.Regs.Regs[1])}, SyscallArgument{Value: uintptr(c.Regs.Regs[2])}, SyscallArgument{Value: uintptr(c.Regs.Regs[3])}, diff --git a/pkg/sentry/kernel/task_syscall.go b/pkg/sentry/kernel/task_syscall.go index d555d69a8..3d7a734ef 100644 --- a/pkg/sentry/kernel/task_syscall.go +++ b/pkg/sentry/kernel/task_syscall.go @@ -194,6 +194,19 @@ func (t *Task) executeSyscall(sysno uintptr, args arch.SyscallArguments) (rval u // // The syscall path is very hot; avoid defer. func (t *Task) doSyscall() taskRunState { + // Save value of the register which is clobbered in the following + // t.Arch().SetReturn(-ENOSYS) operation. This is dedicated to arm64. + // + // On x86, register rax was shared by syscall number and return + // value, and at the entry of the syscall handler, the rax was + // saved to regs.orig_rax which was exposed to user space. + // But on arm64, syscall number was passed through X8, and the X0 + // was shared by the first syscall argument and return value. The + // X0 was saved to regs.orig_x0 which was not exposed to user space. + // So we have to do the same operation here to save the X0 value + // into the task context. + t.Arch().SyscallSaveOrig() + sysno := t.Arch().SyscallNo() args := t.Arch().SyscallArgs() @@ -269,6 +282,7 @@ func (*runSyscallAfterSyscallEnterStop) execute(t *Task) taskRunState { return (*runSyscallExit)(nil) } args := t.Arch().SyscallArgs() + return t.doSyscallInvoke(sysno, args) } diff --git a/test/syscalls/linux/ptrace.cc b/test/syscalls/linux/ptrace.cc index cb828ff88..926690eb8 100644 --- a/test/syscalls/linux/ptrace.cc +++ b/test/syscalls/linux/ptrace.cc @@ -400,9 +400,11 @@ TEST(PtraceTest, GetRegSet) { // Read exactly the full register set. EXPECT_EQ(iov.iov_len, sizeof(regs)); -#ifdef __x86_64__ +#if defined(__x86_64__) // Child called kill(2), with SIGSTOP as arg 2. EXPECT_EQ(regs.rsi, SIGSTOP); +#elif defined(__aarch64__) + EXPECT_EQ(regs.regs[1], SIGSTOP); #endif // Suppress SIGSTOP and resume the child. @@ -752,15 +754,23 @@ TEST(PtraceTest, SyscallSucceeds()); EXPECT_TRUE(siginfo.si_code == SIGTRAP || siginfo.si_code == (SIGTRAP | 0x80)) << "si_code = " << siginfo.si_code; -#ifdef __x86_64__ + { struct user_regs_struct regs = {}; - ASSERT_THAT(ptrace(PTRACE_GETREGS, child_pid, 0, ®s), SyscallSucceeds()); + struct iovec iov; + iov.iov_base = ®s; + iov.iov_len = sizeof(regs); + EXPECT_THAT(ptrace(PTRACE_GETREGSET, child_pid, NT_PRSTATUS, &iov), + SyscallSucceeds()); +#if defined(__x86_64__) EXPECT_TRUE(regs.orig_rax == SYS_vfork || regs.orig_rax == SYS_clone) << "orig_rax = " << regs.orig_rax; EXPECT_EQ(grandchild_pid, regs.rax); - } +#elif defined(__aarch64__) + EXPECT_TRUE(regs.regs[8] == SYS_clone) << "regs[8] = " << regs.regs[8]; + EXPECT_EQ(grandchild_pid, regs.regs[0]); #endif // defined(__x86_64__) + } // After this point, the child will be making wait4 syscalls that will be // interrupted by saving, so saving is not permitted. Note that this is @@ -805,14 +815,21 @@ TEST(PtraceTest, SyscallSucceedsWithValue(child_pid)); EXPECT_TRUE(WIFSTOPPED(status) && WSTOPSIG(status) == (SIGTRAP | 0x80)) << " status " << status; -#ifdef __x86_64__ { struct user_regs_struct regs = {}; - ASSERT_THAT(ptrace(PTRACE_GETREGS, child_pid, 0, ®s), SyscallSucceeds()); + struct iovec iov; + iov.iov_base = ®s; + iov.iov_len = sizeof(regs); + EXPECT_THAT(ptrace(PTRACE_GETREGSET, child_pid, NT_PRSTATUS, &iov), + SyscallSucceeds()); +#if defined(__x86_64__) EXPECT_EQ(SYS_wait4, regs.orig_rax); EXPECT_EQ(grandchild_pid, regs.rax); - } +#elif defined(__aarch64__) + EXPECT_EQ(SYS_wait4, regs.regs[8]); + EXPECT_EQ(grandchild_pid, regs.regs[0]); #endif // defined(__x86_64__) + } // Detach from the child and wait for it to exit. ASSERT_THAT(ptrace(PTRACE_DETACH, child_pid, 0, 0), SyscallSucceeds()); -- cgit v1.2.3 From 1798d6cbee3360b09d3736069e15fd746e863bd2 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Fri, 10 Apr 2020 11:17:59 -0700 Subject: Remove TODO from kernel.Stracer The dependency strace=>kernel grew over time. strace also depends on task's FD table and FSContext. It could be fixed with some interfaces the other way, but then we're trading an interface for another, and kernel.Stracer is likely cleaner. Closes #155 PiperOrigin-RevId: 305909678 --- pkg/sentry/kernel/syscalls.go | 3 --- pkg/sentry/strace/strace.go | 3 --- 2 files changed, 6 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/syscalls.go b/pkg/sentry/kernel/syscalls.go index c9a2321b8..2e3565747 100644 --- a/pkg/sentry/kernel/syscalls.go +++ b/pkg/sentry/kernel/syscalls.go @@ -209,9 +209,6 @@ type Stracer interface { // SyscallEnter is called on syscall entry. // // The returned private data is passed to SyscallExit. - // - // TODO(gvisor.dev/issue/155): remove kernel imports from the strace - // package so that the type can be used directly. SyscallEnter(t *Task, sysno uintptr, args arch.SyscallArguments, flags uint32) interface{} // SyscallExit is called on syscall exit. diff --git a/pkg/sentry/strace/strace.go b/pkg/sentry/strace/strace.go index 77655558e..b94c4fbf5 100644 --- a/pkg/sentry/strace/strace.go +++ b/pkg/sentry/strace/strace.go @@ -778,9 +778,6 @@ func (s SyscallMap) Name(sysno uintptr) string { // // N.B. This is not in an init function because we can't be sure all syscall // tables are registered with the kernel when init runs. -// -// TODO(gvisor.dev/issue/155): remove kernel package dependencies from this -// package and have the kernel package self-initialize all syscall tables. func Initialize() { for _, table := range kernel.SyscallTables() { // Is this known? -- cgit v1.2.3 From 96f914295920404e7c5c97553771e09b31f6900a Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Fri, 10 Apr 2020 15:46:16 -0700 Subject: Use O_CLOEXEC when dup'ing FDs The sentry doesn't allow execve, but it's a good defense in-depth measure. PiperOrigin-RevId: 305958737 --- pkg/sentry/fs/gofer/inode.go | 2 +- pkg/sentry/fsimpl/gofer/gofer.go | 2 +- runsc/boot/filter/config.go | 2 +- runsc/main.go | 2 +- 4 files changed, 4 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/gofer/inode.go b/pkg/sentry/fs/gofer/inode.go index 811e8ea30..a016c896e 100644 --- a/pkg/sentry/fs/gofer/inode.go +++ b/pkg/sentry/fs/gofer/inode.go @@ -273,7 +273,7 @@ func (i *inodeFileState) recreateReadHandles(ctx context.Context, writer *handle // operations on the old will see the new data. Then, make the new handle take // ownereship of the old FD and mark the old readHandle to not close the FD // when done. - if err := syscall.Dup3(h.Host.FD(), i.readHandles.Host.FD(), 0); err != nil { + if err := syscall.Dup3(h.Host.FD(), i.readHandles.Host.FD(), syscall.O_CLOEXEC); err != nil { return err } diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 20edaf643..bdf11fa65 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -1089,7 +1089,7 @@ func (d *dentry) ensureSharedHandle(ctx context.Context, read, write, trunc bool // description, but this doesn't matter since they refer to the // same file (unless d.fs.opts.overlayfsStaleRead is true, // which we handle separately). - if err := syscall.Dup3(int(h.fd), int(d.handle.fd), 0); err != nil { + if err := syscall.Dup3(int(h.fd), int(d.handle.fd), syscall.O_CLOEXEC); err != nil { d.handleMu.Unlock() ctx.Warningf("gofer.dentry.ensureSharedHandle: failed to dup fd %d to fd %d: %v", h.fd, d.handle.fd, err) h.close(ctx) diff --git a/runsc/boot/filter/config.go b/runsc/boot/filter/config.go index 06b9f888a..1828d116a 100644 --- a/runsc/boot/filter/config.go +++ b/runsc/boot/filter/config.go @@ -44,7 +44,7 @@ var allowedSyscalls = seccomp.SyscallRules{ { seccomp.AllowAny{}, seccomp.AllowAny{}, - seccomp.AllowValue(0), + seccomp.AllowValue(syscall.O_CLOEXEC), }, }, syscall.SYS_EPOLL_CREATE1: {}, diff --git a/runsc/main.go b/runsc/main.go index c1c78529c..59f624842 100644 --- a/runsc/main.go +++ b/runsc/main.go @@ -291,7 +291,7 @@ func main() { // want with them. Since Docker and Containerd both eat boot's stderr, we // dup our stderr to the provided log FD so that panics will appear in the // logs, rather than just disappear. - if err := syscall.Dup3(fd, int(os.Stderr.Fd()), 0); err != nil { + if err := syscall.Dup3(fd, int(os.Stderr.Fd()), syscall.O_CLOEXEC); err != nil { cmd.Fatalf("error dup'ing fd %d to stderr: %v", fd, err) } } -- cgit v1.2.3 From 09ddb5a4262c39744643b612109dd12dcce176a8 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 10 Apr 2020 19:01:39 -0700 Subject: Port extended attributes to VFS2. As in VFS1, we only support the user.* namespace. Plumbing is added to tmpfs and goferfs. Note that because of the slightly different order of checks between VFS2 and Linux, one of the xattr tests needs to be relaxed slightly. Fixes #2363. PiperOrigin-RevId: 305985121 --- pkg/sentry/fsimpl/ext/filesystem.go | 4 +- pkg/sentry/fsimpl/gofer/filesystem.go | 12 ++-- pkg/sentry/fsimpl/gofer/gofer.go | 58 +++++++++++---- pkg/sentry/fsimpl/gofer/p9file.go | 14 ++++ pkg/sentry/fsimpl/kernfs/filesystem.go | 4 +- pkg/sentry/fsimpl/tmpfs/BUILD | 1 + pkg/sentry/fsimpl/tmpfs/filesystem.go | 24 +++---- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 77 ++++++++++++++++++++ pkg/sentry/syscalls/linux/vfs2/xattr.go | 13 ++-- pkg/sentry/vfs/anonfs.go | 4 +- pkg/sentry/vfs/file_description.go | 32 ++++++--- pkg/sentry/vfs/file_description_impl_util.go | 4 +- pkg/sentry/vfs/filesystem.go | 24 ++++++- pkg/sentry/vfs/memxattr/BUILD | 15 ++++ pkg/sentry/vfs/memxattr/xattr.go | 102 +++++++++++++++++++++++++++ pkg/sentry/vfs/options.go | 14 ++++ pkg/sentry/vfs/vfs.go | 8 +-- test/syscalls/linux/xattr.cc | 8 +-- 18 files changed, 350 insertions(+), 68 deletions(-) create mode 100644 pkg/sentry/vfs/memxattr/BUILD create mode 100644 pkg/sentry/vfs/memxattr/xattr.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/ext/filesystem.go b/pkg/sentry/fsimpl/ext/filesystem.go index 48eaccdbc..afea58f65 100644 --- a/pkg/sentry/fsimpl/ext/filesystem.go +++ b/pkg/sentry/fsimpl/ext/filesystem.go @@ -476,7 +476,7 @@ func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath } // ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. -func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) { +func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, size uint64) ([]string, error) { _, _, err := fs.walk(rp, false) if err != nil { return nil, err @@ -485,7 +485,7 @@ func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([ } // GetxattrAt implements vfs.FilesystemImpl.GetxattrAt. -func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) (string, error) { +func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetxattrOptions) (string, error) { _, _, err := fs.walk(rp, false) if err != nil { return "", err diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 137260898..cd744bf5e 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -1080,7 +1080,7 @@ func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath } // ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. -func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) { +func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, size uint64) ([]string, error) { var ds *[]*dentry fs.renameMu.RLock() defer fs.renameMuRUnlockAndCheckCaching(&ds) @@ -1088,11 +1088,11 @@ func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([ if err != nil { return nil, err } - return d.listxattr(ctx) + return d.listxattr(ctx, rp.Credentials(), size) } // GetxattrAt implements vfs.FilesystemImpl.GetxattrAt. -func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) (string, error) { +func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetxattrOptions) (string, error) { var ds *[]*dentry fs.renameMu.RLock() defer fs.renameMuRUnlockAndCheckCaching(&ds) @@ -1100,7 +1100,7 @@ func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, nam if err != nil { return "", err } - return d.getxattr(ctx, name) + return d.getxattr(ctx, rp.Credentials(), &opts) } // SetxattrAt implements vfs.FilesystemImpl.SetxattrAt. @@ -1112,7 +1112,7 @@ func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt if err != nil { return err } - return d.setxattr(ctx, &opts) + return d.setxattr(ctx, rp.Credentials(), &opts) } // RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. @@ -1124,7 +1124,7 @@ func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, if err != nil { return err } - return d.removexattr(ctx, name) + return d.removexattr(ctx, rp.Credentials(), name) } // PrependPath implements vfs.FilesystemImpl.PrependPath. diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index bdf11fa65..2485cdb53 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -34,6 +34,7 @@ package gofer import ( "fmt" "strconv" + "strings" "sync" "sync/atomic" "syscall" @@ -1024,21 +1025,50 @@ func (d *dentry) setDeleted() { atomic.StoreUint32(&d.deleted, 1) } -func (d *dentry) listxattr(ctx context.Context) ([]string, error) { - return nil, syserror.ENOTSUP +// We only support xattrs prefixed with "user." (see b/148380782). Currently, +// there is no need to expose any other xattrs through a gofer. +func (d *dentry) listxattr(ctx context.Context, creds *auth.Credentials, size uint64) ([]string, error) { + xattrMap, err := d.file.listXattr(ctx, size) + if err != nil { + return nil, err + } + xattrs := make([]string, 0, len(xattrMap)) + for x := range xattrMap { + if strings.HasPrefix(x, linux.XATTR_USER_PREFIX) { + xattrs = append(xattrs, x) + } + } + return xattrs, nil } -func (d *dentry) getxattr(ctx context.Context, name string) (string, error) { - // TODO(jamieliu): add vfs.GetxattrOptions.Size - return d.file.getXattr(ctx, name, linux.XATTR_SIZE_MAX) +func (d *dentry) getxattr(ctx context.Context, creds *auth.Credentials, opts *vfs.GetxattrOptions) (string, error) { + if err := d.checkPermissions(creds, vfs.MayRead); err != nil { + return "", err + } + if !strings.HasPrefix(opts.Name, linux.XATTR_USER_PREFIX) { + return "", syserror.EOPNOTSUPP + } + return d.file.getXattr(ctx, opts.Name, opts.Size) } -func (d *dentry) setxattr(ctx context.Context, opts *vfs.SetxattrOptions) error { +func (d *dentry) setxattr(ctx context.Context, creds *auth.Credentials, opts *vfs.SetxattrOptions) error { + if err := d.checkPermissions(creds, vfs.MayWrite); err != nil { + return err + } + if !strings.HasPrefix(opts.Name, linux.XATTR_USER_PREFIX) { + return syserror.EOPNOTSUPP + } return d.file.setXattr(ctx, opts.Name, opts.Value, opts.Flags) } -func (d *dentry) removexattr(ctx context.Context, name string) error { - return syserror.ENOTSUP +func (d *dentry) removexattr(ctx context.Context, creds *auth.Credentials, name string) error { + if err := d.checkPermissions(creds, vfs.MayWrite); err != nil { + return err + } + if !strings.HasPrefix(name, linux.XATTR_USER_PREFIX) { + return syserror.EOPNOTSUPP + } + return d.file.removeXattr(ctx, name) } // Preconditions: d.isRegularFile() || d.isDirectory(). @@ -1189,21 +1219,21 @@ func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) } // Listxattr implements vfs.FileDescriptionImpl.Listxattr. -func (fd *fileDescription) Listxattr(ctx context.Context) ([]string, error) { - return fd.dentry().listxattr(ctx) +func (fd *fileDescription) Listxattr(ctx context.Context, size uint64) ([]string, error) { + return fd.dentry().listxattr(ctx, auth.CredentialsFromContext(ctx), size) } // Getxattr implements vfs.FileDescriptionImpl.Getxattr. -func (fd *fileDescription) Getxattr(ctx context.Context, name string) (string, error) { - return fd.dentry().getxattr(ctx, name) +func (fd *fileDescription) Getxattr(ctx context.Context, opts vfs.GetxattrOptions) (string, error) { + return fd.dentry().getxattr(ctx, auth.CredentialsFromContext(ctx), &opts) } // Setxattr implements vfs.FileDescriptionImpl.Setxattr. func (fd *fileDescription) Setxattr(ctx context.Context, opts vfs.SetxattrOptions) error { - return fd.dentry().setxattr(ctx, &opts) + return fd.dentry().setxattr(ctx, auth.CredentialsFromContext(ctx), &opts) } // Removexattr implements vfs.FileDescriptionImpl.Removexattr. func (fd *fileDescription) Removexattr(ctx context.Context, name string) error { - return fd.dentry().removexattr(ctx, name) + return fd.dentry().removexattr(ctx, auth.CredentialsFromContext(ctx), name) } diff --git a/pkg/sentry/fsimpl/gofer/p9file.go b/pkg/sentry/fsimpl/gofer/p9file.go index 755ac2985..87f0b877f 100644 --- a/pkg/sentry/fsimpl/gofer/p9file.go +++ b/pkg/sentry/fsimpl/gofer/p9file.go @@ -85,6 +85,13 @@ func (f p9file) setAttr(ctx context.Context, valid p9.SetAttrMask, attr p9.SetAt return err } +func (f p9file) listXattr(ctx context.Context, size uint64) (map[string]struct{}, error) { + ctx.UninterruptibleSleepStart(false) + xattrs, err := f.file.ListXattr(size) + ctx.UninterruptibleSleepFinish(false) + return xattrs, err +} + func (f p9file) getXattr(ctx context.Context, name string, size uint64) (string, error) { ctx.UninterruptibleSleepStart(false) val, err := f.file.GetXattr(name, size) @@ -99,6 +106,13 @@ func (f p9file) setXattr(ctx context.Context, name, value string, flags uint32) return err } +func (f p9file) removeXattr(ctx context.Context, name string) error { + ctx.UninterruptibleSleepStart(false) + err := f.file.RemoveXattr(name) + ctx.UninterruptibleSleepFinish(false) + return err +} + func (f p9file) allocate(ctx context.Context, mode p9.AllocateMode, offset, length uint64) error { ctx.UninterruptibleSleepStart(false) err := f.file.Allocate(mode, offset, length) diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 4433071aa..baf81b4db 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -763,7 +763,7 @@ func (fs *Filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath } // ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. -func (fs *Filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) { +func (fs *Filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, size uint64) ([]string, error) { fs.mu.RLock() _, _, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() @@ -776,7 +776,7 @@ func (fs *Filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([ } // GetxattrAt implements vfs.FilesystemImpl.GetxattrAt. -func (fs *Filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) (string, error) { +func (fs *Filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetxattrOptions) (string, error) { fs.mu.RLock() _, _, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD index f2ac23c88..4e6cd3491 100644 --- a/pkg/sentry/fsimpl/tmpfs/BUILD +++ b/pkg/sentry/fsimpl/tmpfs/BUILD @@ -51,6 +51,7 @@ go_library( "//pkg/sentry/usage", "//pkg/sentry/vfs", "//pkg/sentry/vfs/lock", + "//pkg/sentry/vfs/memxattr", "//pkg/sync", "//pkg/syserror", "//pkg/usermem", diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 5339d7072..f4d50d64f 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -696,51 +696,47 @@ func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath } // ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. -func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) { +func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, size uint64) ([]string, error) { fs.mu.RLock() defer fs.mu.RUnlock() - _, err := resolveLocked(rp) + d, err := resolveLocked(rp) if err != nil { return nil, err } - // TODO(b/127675828): support extended attributes - return nil, syserror.ENOTSUP + return d.inode.listxattr(size) } // GetxattrAt implements vfs.FilesystemImpl.GetxattrAt. -func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) (string, error) { +func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetxattrOptions) (string, error) { fs.mu.RLock() defer fs.mu.RUnlock() - _, err := resolveLocked(rp) + d, err := resolveLocked(rp) if err != nil { return "", err } - // TODO(b/127675828): support extended attributes - return "", syserror.ENOTSUP + return d.inode.getxattr(rp.Credentials(), &opts) } // SetxattrAt implements vfs.FilesystemImpl.SetxattrAt. func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { fs.mu.RLock() defer fs.mu.RUnlock() - _, err := resolveLocked(rp) + d, err := resolveLocked(rp) if err != nil { return err } - // TODO(b/127675828): support extended attributes - return syserror.ENOTSUP + return d.inode.setxattr(rp.Credentials(), &opts) } // RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { fs.mu.RLock() defer fs.mu.RUnlock() - _, err := resolveLocked(rp) + d, err := resolveLocked(rp) if err != nil { return err } - // TODO(b/127675828): support extended attributes - return syserror.ENOTSUP + return d.inode.removexattr(rp.Credentials(), name) } // PrependPath implements vfs.FilesystemImpl.PrependPath. diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 654e788e3..9fa8637d5 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -27,6 +27,7 @@ package tmpfs import ( "fmt" "math" + "strings" "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" @@ -37,6 +38,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sentry/vfs/lock" + "gvisor.dev/gvisor/pkg/sentry/vfs/memxattr" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) @@ -186,6 +188,11 @@ type inode struct { // filesystem.RmdirAt() drops the reference. refs int64 + // xattrs implements extended attributes. + // + // TODO(b/148380782): Support xattrs other than user.* + xattrs memxattr.SimpleExtendedAttributes + // Inode metadata. Writing multiple fields atomically requires holding // mu, othewise atomic operations can be used. mu sync.Mutex @@ -535,6 +542,56 @@ func (i *inode) touchCMtimeLocked() { atomic.StoreInt64(&i.ctime, now) } +func (i *inode) listxattr(size uint64) ([]string, error) { + return i.xattrs.Listxattr(size) +} + +func (i *inode) getxattr(creds *auth.Credentials, opts *vfs.GetxattrOptions) (string, error) { + if err := i.checkPermissions(creds, vfs.MayRead); err != nil { + return "", err + } + if !strings.HasPrefix(opts.Name, linux.XATTR_USER_PREFIX) { + return "", syserror.EOPNOTSUPP + } + if !i.userXattrSupported() { + return "", syserror.ENODATA + } + return i.xattrs.Getxattr(opts) +} + +func (i *inode) setxattr(creds *auth.Credentials, opts *vfs.SetxattrOptions) error { + if err := i.checkPermissions(creds, vfs.MayWrite); err != nil { + return err + } + if !strings.HasPrefix(opts.Name, linux.XATTR_USER_PREFIX) { + return syserror.EOPNOTSUPP + } + if !i.userXattrSupported() { + return syserror.EPERM + } + return i.xattrs.Setxattr(opts) +} + +func (i *inode) removexattr(creds *auth.Credentials, name string) error { + if err := i.checkPermissions(creds, vfs.MayWrite); err != nil { + return err + } + if !strings.HasPrefix(name, linux.XATTR_USER_PREFIX) { + return syserror.EOPNOTSUPP + } + if !i.userXattrSupported() { + return syserror.EPERM + } + return i.xattrs.Removexattr(name) +} + +// Extended attributes in the user.* namespace are only supported for regular +// files and directories. +func (i *inode) userXattrSupported() bool { + filetype := linux.S_IFMT & atomic.LoadUint32(&i.mode) + return filetype == linux.S_IFREG || filetype == linux.S_IFDIR +} + // fileDescription is embedded by tmpfs implementations of // vfs.FileDescriptionImpl. type fileDescription struct { @@ -562,3 +619,23 @@ func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) creds := auth.CredentialsFromContext(ctx) return fd.inode().setStat(ctx, creds, &opts.Stat) } + +// Listxattr implements vfs.FileDescriptionImpl.Listxattr. +func (fd *fileDescription) Listxattr(ctx context.Context, size uint64) ([]string, error) { + return fd.inode().listxattr(size) +} + +// Getxattr implements vfs.FileDescriptionImpl.Getxattr. +func (fd *fileDescription) Getxattr(ctx context.Context, opts vfs.GetxattrOptions) (string, error) { + return fd.inode().getxattr(auth.CredentialsFromContext(ctx), &opts) +} + +// Setxattr implements vfs.FileDescriptionImpl.Setxattr. +func (fd *fileDescription) Setxattr(ctx context.Context, opts vfs.SetxattrOptions) error { + return fd.inode().setxattr(auth.CredentialsFromContext(ctx), &opts) +} + +// Removexattr implements vfs.FileDescriptionImpl.Removexattr. +func (fd *fileDescription) Removexattr(ctx context.Context, name string) error { + return fd.inode().removexattr(auth.CredentialsFromContext(ctx), name) +} diff --git a/pkg/sentry/syscalls/linux/vfs2/xattr.go b/pkg/sentry/syscalls/linux/vfs2/xattr.go index 89e9ff4d7..af455d5c1 100644 --- a/pkg/sentry/syscalls/linux/vfs2/xattr.go +++ b/pkg/sentry/syscalls/linux/vfs2/xattr.go @@ -51,7 +51,7 @@ func listxattr(t *kernel.Task, args arch.SyscallArguments, shouldFollowFinalSyml } defer tpop.Release() - names, err := t.Kernel().VFS().ListxattrAt(t, t.Credentials(), &tpop.pop) + names, err := t.Kernel().VFS().ListxattrAt(t, t.Credentials(), &tpop.pop, uint64(size)) if err != nil { return 0, nil, err } @@ -74,7 +74,7 @@ func Flistxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy } defer file.DecRef() - names, err := file.Listxattr(t) + names, err := file.Listxattr(t, uint64(size)) if err != nil { return 0, nil, err } @@ -116,7 +116,10 @@ func getxattr(t *kernel.Task, args arch.SyscallArguments, shouldFollowFinalSymli return 0, nil, err } - value, err := t.Kernel().VFS().GetxattrAt(t, t.Credentials(), &tpop.pop, name) + value, err := t.Kernel().VFS().GetxattrAt(t, t.Credentials(), &tpop.pop, &vfs.GetxattrOptions{ + Name: name, + Size: uint64(size), + }) if err != nil { return 0, nil, err } @@ -145,7 +148,7 @@ func Fgetxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys return 0, nil, err } - value, err := file.Getxattr(t, name) + value, err := file.Getxattr(t, &vfs.GetxattrOptions{Name: name, Size: uint64(size)}) if err != nil { return 0, nil, err } @@ -230,7 +233,7 @@ func Fsetxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys return 0, nil, err } - return 0, nil, file.Setxattr(t, vfs.SetxattrOptions{ + return 0, nil, file.Setxattr(t, &vfs.SetxattrOptions{ Name: name, Value: value, Flags: uint32(flags), diff --git a/pkg/sentry/vfs/anonfs.go b/pkg/sentry/vfs/anonfs.go index d1f6dfb45..a64d86122 100644 --- a/pkg/sentry/vfs/anonfs.go +++ b/pkg/sentry/vfs/anonfs.go @@ -245,7 +245,7 @@ func (fs *anonFilesystem) BoundEndpointAt(ctx context.Context, rp *ResolvingPath } // ListxattrAt implements FilesystemImpl.ListxattrAt. -func (fs *anonFilesystem) ListxattrAt(ctx context.Context, rp *ResolvingPath) ([]string, error) { +func (fs *anonFilesystem) ListxattrAt(ctx context.Context, rp *ResolvingPath, size uint64) ([]string, error) { if !rp.Done() { return nil, syserror.ENOTDIR } @@ -253,7 +253,7 @@ func (fs *anonFilesystem) ListxattrAt(ctx context.Context, rp *ResolvingPath) ([ } // GetxattrAt implements FilesystemImpl.GetxattrAt. -func (fs *anonFilesystem) GetxattrAt(ctx context.Context, rp *ResolvingPath, name string) (string, error) { +func (fs *anonFilesystem) GetxattrAt(ctx context.Context, rp *ResolvingPath, opts GetxattrOptions) (string, error) { if !rp.Done() { return "", syserror.ENOTDIR } diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 20c545fca..4fb9aea87 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -401,11 +401,11 @@ type FileDescriptionImpl interface { Ioctl(ctx context.Context, uio usermem.IO, args arch.SyscallArguments) (uintptr, error) // Listxattr returns all extended attribute names for the file. - Listxattr(ctx context.Context) ([]string, error) + Listxattr(ctx context.Context, size uint64) ([]string, error) // Getxattr returns the value associated with the given extended attribute // for the file. - Getxattr(ctx context.Context, name string) (string, error) + Getxattr(ctx context.Context, opts GetxattrOptions) (string, error) // Setxattr changes the value associated with the given extended attribute // for the file. @@ -605,18 +605,23 @@ func (fd *FileDescription) Ioctl(ctx context.Context, uio usermem.IO, args arch. // Listxattr returns all extended attribute names for the file represented by // fd. -func (fd *FileDescription) Listxattr(ctx context.Context) ([]string, error) { +// +// If the size of the list (including a NUL terminating byte after every entry) +// would exceed size, ERANGE may be returned. Note that implementations +// are free to ignore size entirely and return without error). In all cases, +// if size is 0, the list should be returned without error, regardless of size. +func (fd *FileDescription) Listxattr(ctx context.Context, size uint64) ([]string, error) { if fd.opts.UseDentryMetadata { vfsObj := fd.vd.mount.vfs rp := vfsObj.getResolvingPath(auth.CredentialsFromContext(ctx), &PathOperation{ Root: fd.vd, Start: fd.vd, }) - names, err := fd.vd.mount.fs.impl.ListxattrAt(ctx, rp) + names, err := fd.vd.mount.fs.impl.ListxattrAt(ctx, rp, size) vfsObj.putResolvingPath(rp) return names, err } - names, err := fd.impl.Listxattr(ctx) + names, err := fd.impl.Listxattr(ctx, size) if err == syserror.ENOTSUP { // Linux doesn't actually return ENOTSUP in this case; instead, // fs/xattr.c:vfs_listxattr() falls back to allowing the security @@ -629,34 +634,39 @@ func (fd *FileDescription) Listxattr(ctx context.Context) ([]string, error) { // Getxattr returns the value associated with the given extended attribute for // the file represented by fd. -func (fd *FileDescription) Getxattr(ctx context.Context, name string) (string, error) { +// +// If the size of the return value exceeds opts.Size, ERANGE may be returned +// (note that implementations are free to ignore opts.Size entirely and return +// without error). In all cases, if opts.Size is 0, the value should be +// returned without error, regardless of size. +func (fd *FileDescription) Getxattr(ctx context.Context, opts *GetxattrOptions) (string, error) { if fd.opts.UseDentryMetadata { vfsObj := fd.vd.mount.vfs rp := vfsObj.getResolvingPath(auth.CredentialsFromContext(ctx), &PathOperation{ Root: fd.vd, Start: fd.vd, }) - val, err := fd.vd.mount.fs.impl.GetxattrAt(ctx, rp, name) + val, err := fd.vd.mount.fs.impl.GetxattrAt(ctx, rp, *opts) vfsObj.putResolvingPath(rp) return val, err } - return fd.impl.Getxattr(ctx, name) + return fd.impl.Getxattr(ctx, *opts) } // Setxattr changes the value associated with the given extended attribute for // the file represented by fd. -func (fd *FileDescription) Setxattr(ctx context.Context, opts SetxattrOptions) error { +func (fd *FileDescription) Setxattr(ctx context.Context, opts *SetxattrOptions) error { if fd.opts.UseDentryMetadata { vfsObj := fd.vd.mount.vfs rp := vfsObj.getResolvingPath(auth.CredentialsFromContext(ctx), &PathOperation{ Root: fd.vd, Start: fd.vd, }) - err := fd.vd.mount.fs.impl.SetxattrAt(ctx, rp, opts) + err := fd.vd.mount.fs.impl.SetxattrAt(ctx, rp, *opts) vfsObj.putResolvingPath(rp) return err } - return fd.impl.Setxattr(ctx, opts) + return fd.impl.Setxattr(ctx, *opts) } // Removexattr removes the given extended attribute from the file represented diff --git a/pkg/sentry/vfs/file_description_impl_util.go b/pkg/sentry/vfs/file_description_impl_util.go index d45e602ce..f4c111926 100644 --- a/pkg/sentry/vfs/file_description_impl_util.go +++ b/pkg/sentry/vfs/file_description_impl_util.go @@ -130,14 +130,14 @@ func (FileDescriptionDefaultImpl) Ioctl(ctx context.Context, uio usermem.IO, arg // Listxattr implements FileDescriptionImpl.Listxattr analogously to // inode_operations::listxattr == NULL in Linux. -func (FileDescriptionDefaultImpl) Listxattr(ctx context.Context) ([]string, error) { +func (FileDescriptionDefaultImpl) Listxattr(ctx context.Context, size uint64) ([]string, error) { // This isn't exactly accurate; see FileDescription.Listxattr. return nil, syserror.ENOTSUP } // Getxattr implements FileDescriptionImpl.Getxattr analogously to // inode::i_opflags & IOP_XATTR == 0 in Linux. -func (FileDescriptionDefaultImpl) Getxattr(ctx context.Context, name string) (string, error) { +func (FileDescriptionDefaultImpl) Getxattr(ctx context.Context, opts GetxattrOptions) (string, error) { return "", syserror.ENOTSUP } diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index bef1bd312..a537a29d1 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -442,7 +442,13 @@ type FilesystemImpl interface { // - If extended attributes are not supported by the filesystem, // ListxattrAt returns nil. (See FileDescription.Listxattr for an // explanation.) - ListxattrAt(ctx context.Context, rp *ResolvingPath) ([]string, error) + // + // - If the size of the list (including a NUL terminating byte after every + // entry) would exceed size, ERANGE may be returned. Note that + // implementations are free to ignore size entirely and return without + // error). In all cases, if size is 0, the list should be returned without + // error, regardless of size. + ListxattrAt(ctx context.Context, rp *ResolvingPath, size uint64) ([]string, error) // GetxattrAt returns the value associated with the given extended // attribute for the file at rp. @@ -451,7 +457,15 @@ type FilesystemImpl interface { // // - If extended attributes are not supported by the filesystem, GetxattrAt // returns ENOTSUP. - GetxattrAt(ctx context.Context, rp *ResolvingPath, name string) (string, error) + // + // - If an extended attribute named opts.Name does not exist, ENODATA is + // returned. + // + // - If the size of the return value exceeds opts.Size, ERANGE may be + // returned (note that implementations are free to ignore opts.Size entirely + // and return without error). In all cases, if opts.Size is 0, the value + // should be returned without error, regardless of size. + GetxattrAt(ctx context.Context, rp *ResolvingPath, opts GetxattrOptions) (string, error) // SetxattrAt changes the value associated with the given extended // attribute for the file at rp. @@ -460,6 +474,10 @@ type FilesystemImpl interface { // // - If extended attributes are not supported by the filesystem, SetxattrAt // returns ENOTSUP. + // + // - If XATTR_CREATE is set in opts.Flag and opts.Name already exists, + // EEXIST is returned. If XATTR_REPLACE is set and opts.Name does not exist, + // ENODATA is returned. SetxattrAt(ctx context.Context, rp *ResolvingPath, opts SetxattrOptions) error // RemovexattrAt removes the given extended attribute from the file at rp. @@ -468,6 +486,8 @@ type FilesystemImpl interface { // // - If extended attributes are not supported by the filesystem, // RemovexattrAt returns ENOTSUP. + // + // - If name does not exist, ENODATA is returned. RemovexattrAt(ctx context.Context, rp *ResolvingPath, name string) error // BoundEndpointAt returns the Unix socket endpoint bound at the path rp. diff --git a/pkg/sentry/vfs/memxattr/BUILD b/pkg/sentry/vfs/memxattr/BUILD new file mode 100644 index 000000000..d8c4d27b9 --- /dev/null +++ b/pkg/sentry/vfs/memxattr/BUILD @@ -0,0 +1,15 @@ +load("//tools:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_library( + name = "memxattr", + srcs = ["xattr.go"], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/abi/linux", + "//pkg/sentry/vfs", + "//pkg/sync", + "//pkg/syserror", + ], +) diff --git a/pkg/sentry/vfs/memxattr/xattr.go b/pkg/sentry/vfs/memxattr/xattr.go new file mode 100644 index 000000000..cc1e7d764 --- /dev/null +++ b/pkg/sentry/vfs/memxattr/xattr.go @@ -0,0 +1,102 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package memxattr provides a default, in-memory extended attribute +// implementation. +package memxattr + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/syserror" +) + +// SimpleExtendedAttributes implements extended attributes using a map of +// names to values. +// +// +stateify savable +type SimpleExtendedAttributes struct { + // mu protects the below fields. + mu sync.RWMutex `state:"nosave"` + xattrs map[string]string +} + +// Getxattr returns the value at 'name'. +func (x *SimpleExtendedAttributes) Getxattr(opts *vfs.GetxattrOptions) (string, error) { + x.mu.RLock() + value, ok := x.xattrs[opts.Name] + x.mu.RUnlock() + if !ok { + return "", syserror.ENODATA + } + // Check that the size of the buffer provided in getxattr(2) is large enough + // to contain the value. + if opts.Size != 0 && uint64(len(value)) > opts.Size { + return "", syserror.ERANGE + } + return value, nil +} + +// Setxattr sets 'value' at 'name'. +func (x *SimpleExtendedAttributes) Setxattr(opts *vfs.SetxattrOptions) error { + x.mu.Lock() + defer x.mu.Unlock() + if x.xattrs == nil { + if opts.Flags&linux.XATTR_REPLACE != 0 { + return syserror.ENODATA + } + x.xattrs = make(map[string]string) + } + + _, ok := x.xattrs[opts.Name] + if ok && opts.Flags&linux.XATTR_CREATE != 0 { + return syserror.EEXIST + } + if !ok && opts.Flags&linux.XATTR_REPLACE != 0 { + return syserror.ENODATA + } + + x.xattrs[opts.Name] = opts.Value + return nil +} + +// Listxattr returns all names in xattrs. +func (x *SimpleExtendedAttributes) Listxattr(size uint64) ([]string, error) { + // Keep track of the size of the buffer needed in listxattr(2) for the list. + listSize := 0 + x.mu.RLock() + names := make([]string, 0, len(x.xattrs)) + for n := range x.xattrs { + names = append(names, n) + // Add one byte per null terminator. + listSize += len(n) + 1 + } + x.mu.RUnlock() + if size != 0 && uint64(listSize) > size { + return nil, syserror.ERANGE + } + return names, nil +} + +// Removexattr removes the xattr at 'name'. +func (x *SimpleExtendedAttributes) Removexattr(name string) error { + x.mu.Lock() + defer x.mu.Unlock() + if _, ok := x.xattrs[name]; !ok { + return syserror.ENODATA + } + delete(x.xattrs, name) + return nil +} diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go index 2f04bf882..534528ce6 100644 --- a/pkg/sentry/vfs/options.go +++ b/pkg/sentry/vfs/options.go @@ -132,6 +132,20 @@ type SetStatOptions struct { Stat linux.Statx } +// GetxattrOptions contains options to VirtualFilesystem.GetxattrAt(), +// FilesystemImpl.GetxattrAt(), FileDescription.Getxattr(), and +// FileDescriptionImpl.Getxattr(). +type GetxattrOptions struct { + // Name is the name of the extended attribute to retrieve. + Name string + + // Size is the maximum value size that the caller will tolerate. If the value + // is larger than size, getxattr methods may return ERANGE, but they are also + // free to ignore the hint entirely (i.e. the value returned may be larger + // than size). All size checking is done independently at the syscall layer. + Size uint64 +} + // SetxattrOptions contains options to VirtualFilesystem.SetxattrAt(), // FilesystemImpl.SetxattrAt(), FileDescription.Setxattr(), and // FileDescriptionImpl.Setxattr(). diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 720b90d8f..f592913d5 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -680,10 +680,10 @@ func (vfs *VirtualFilesystem) UnlinkAt(ctx context.Context, creds *auth.Credenti // ListxattrAt returns all extended attribute names for the file at the given // path. -func (vfs *VirtualFilesystem) ListxattrAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation) ([]string, error) { +func (vfs *VirtualFilesystem) ListxattrAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, size uint64) ([]string, error) { rp := vfs.getResolvingPath(creds, pop) for { - names, err := rp.mount.fs.impl.ListxattrAt(ctx, rp) + names, err := rp.mount.fs.impl.ListxattrAt(ctx, rp, size) if err == nil { vfs.putResolvingPath(rp) return names, nil @@ -705,10 +705,10 @@ func (vfs *VirtualFilesystem) ListxattrAt(ctx context.Context, creds *auth.Crede // GetxattrAt returns the value associated with the given extended attribute // for the file at the given path. -func (vfs *VirtualFilesystem) GetxattrAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, name string) (string, error) { +func (vfs *VirtualFilesystem) GetxattrAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *GetxattrOptions) (string, error) { rp := vfs.getResolvingPath(creds, pop) for { - val, err := rp.mount.fs.impl.GetxattrAt(ctx, rp, name) + val, err := rp.mount.fs.impl.GetxattrAt(ctx, rp, *opts) if err == nil { vfs.putResolvingPath(rp) return val, nil diff --git a/test/syscalls/linux/xattr.cc b/test/syscalls/linux/xattr.cc index 8b00ef44c..3231732ec 100644 --- a/test/syscalls/linux/xattr.cc +++ b/test/syscalls/linux/xattr.cc @@ -41,12 +41,12 @@ class XattrTest : public FileTest {}; TEST_F(XattrTest, XattrNonexistentFile) { const char* path = "/does/not/exist"; - EXPECT_THAT(setxattr(path, nullptr, nullptr, 0, /*flags=*/0), - SyscallFailsWithErrno(ENOENT)); - EXPECT_THAT(getxattr(path, nullptr, nullptr, 0), + const char* name = "user.test"; + EXPECT_THAT(setxattr(path, name, nullptr, 0, /*flags=*/0), SyscallFailsWithErrno(ENOENT)); + EXPECT_THAT(getxattr(path, name, nullptr, 0), SyscallFailsWithErrno(ENOENT)); EXPECT_THAT(listxattr(path, nullptr, 0), SyscallFailsWithErrno(ENOENT)); - EXPECT_THAT(removexattr(path, nullptr), SyscallFailsWithErrno(ENOENT)); + EXPECT_THAT(removexattr(path, name), SyscallFailsWithErrno(ENOENT)); } TEST_F(XattrTest, XattrNullName) { -- cgit v1.2.3 From daf3322498b698518a3c8545ad05f790deb3848c Mon Sep 17 00:00:00 2001 From: Ian Lewis Date: Fri, 10 Apr 2020 20:31:07 -0700 Subject: Add logging message for noNewPrivileges OCI option. noNewPrivileges is ignored if set to false since gVisor assumes that PR_SET_NO_NEW_PRIVS is always enabled. PiperOrigin-RevId: 305991947 --- pkg/sentry/kernel/task_identity.go | 2 +- pkg/sentry/syscalls/linux/sys_prctl.go | 4 ++-- runsc/specutils/specutils.go | 6 ++++++ 3 files changed, 9 insertions(+), 3 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/task_identity.go b/pkg/sentry/kernel/task_identity.go index ce3e6ef28..0325967e4 100644 --- a/pkg/sentry/kernel/task_identity.go +++ b/pkg/sentry/kernel/task_identity.go @@ -455,7 +455,7 @@ func (t *Task) SetKeepCaps(k bool) { t.creds.Store(creds) } -// updateCredsForExec updates t.creds to reflect an execve(). +// updateCredsForExecLocked updates t.creds to reflect an execve(). // // NOTE(b/30815691): We currently do not implement privileged executables // (set-user/group-ID bits and file capabilities). This allows us to make a lot diff --git a/pkg/sentry/syscalls/linux/sys_prctl.go b/pkg/sentry/syscalls/linux/sys_prctl.go index 9c6728530..f92bf8096 100644 --- a/pkg/sentry/syscalls/linux/sys_prctl.go +++ b/pkg/sentry/syscalls/linux/sys_prctl.go @@ -161,8 +161,8 @@ func Prctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if args[1].Int() != 1 || args[2].Int() != 0 || args[3].Int() != 0 || args[4].Int() != 0 { return 0, nil, syserror.EINVAL } - // no_new_privs is assumed to always be set. See - // kernel.Task.updateCredsForExec. + // PR_SET_NO_NEW_PRIVS is assumed to always be set. + // See kernel.Task.updateCredsForExecLocked. return 0, nil, nil case linux.PR_GET_NO_NEW_PRIVS: diff --git a/runsc/specutils/specutils.go b/runsc/specutils/specutils.go index 0f4a9cf6d..837d5e238 100644 --- a/runsc/specutils/specutils.go +++ b/runsc/specutils/specutils.go @@ -92,6 +92,12 @@ func ValidateSpec(spec *specs.Spec) error { log.Warningf("AppArmor profile %q is being ignored", spec.Process.ApparmorProfile) } + // PR_SET_NO_NEW_PRIVS is assumed to always be set. + // See kernel.Task.updateCredsForExecLocked. + if !spec.Process.NoNewPrivileges { + log.Warningf("noNewPrivileges ignored. PR_SET_NO_NEW_PRIVS is assumed to always be set.") + } + // TODO(gvisor.dev/issue/510): Apply seccomp to application inside sandbox. if spec.Linux != nil && spec.Linux.Seccomp != nil { log.Warningf("Seccomp spec is being ignored") -- cgit v1.2.3 From 445c366581637b64336a18d69519faee5a444f5d Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Mon, 13 Apr 2020 10:51:08 -0700 Subject: Fix VFS2 getdents()/getdents64() alignment. PiperOrigin-RevId: 306263615 --- pkg/sentry/syscalls/linux/vfs2/getdents.go | 20 ++++++++++++++++---- 1 file changed, 16 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/vfs2/getdents.go b/pkg/sentry/syscalls/linux/vfs2/getdents.go index a61cc5059..62e98817d 100644 --- a/pkg/sentry/syscalls/linux/vfs2/getdents.go +++ b/pkg/sentry/syscalls/linux/vfs2/getdents.go @@ -97,6 +97,7 @@ func (cb *getdentsCallback) Handle(dirent vfs.Dirent) error { // char d_name[]; /* Filename (null-terminated) */ // }; size := 8 + 8 + 2 + 1 + 1 + len(dirent.Name) + size = (size + 7) &^ 7 // round up to multiple of 8 if size > cb.remaining { return syserror.EINVAL } @@ -106,7 +107,12 @@ func (cb *getdentsCallback) Handle(dirent vfs.Dirent) error { usermem.ByteOrder.PutUint16(buf[16:18], uint16(size)) buf[18] = dirent.Type copy(buf[19:], dirent.Name) - buf[size-1] = 0 // NUL terminator + // Zero out all remaining bytes in buf, including the NUL terminator + // after dirent.Name. + bufTail := buf[19+len(dirent.Name):] + for i := range bufTail { + bufTail[i] = 0 + } } else { // struct linux_dirent { // unsigned long d_ino; /* Inode number */ @@ -125,6 +131,7 @@ func (cb *getdentsCallback) Handle(dirent vfs.Dirent) error { panic(fmt.Sprintf("unsupported sizeof(unsigned long): %d", cb.t.Arch().Width())) } size := 8 + 8 + 2 + 1 + 1 + 1 + len(dirent.Name) + size = (size + 7) &^ 7 // round up to multiple of sizeof(long) if size > cb.remaining { return syserror.EINVAL } @@ -133,9 +140,14 @@ func (cb *getdentsCallback) Handle(dirent vfs.Dirent) error { usermem.ByteOrder.PutUint64(buf[8:16], uint64(dirent.NextOff)) usermem.ByteOrder.PutUint16(buf[16:18], uint16(size)) copy(buf[18:], dirent.Name) - buf[size-3] = 0 // NUL terminator - buf[size-2] = 0 // zero padding byte - buf[size-1] = dirent.Type + // Zero out all remaining bytes in buf, including the NUL terminator + // after dirent.Name and the zero padding byte between the name and + // dirent type. + bufTail := buf[18+len(dirent.Name):] + for i := range bufTail { + bufTail[i] = 0 + } + bufTail[2] = dirent.Type } n, err := cb.t.CopyOutBytes(cb.addr, buf) if err != nil { -- cgit v1.2.3 From 6a4d17a31dc209afbbca66e871a7c6dc299c167b Mon Sep 17 00:00:00 2001 From: Jon Budd Date: Mon, 13 Apr 2020 11:01:02 -0700 Subject: Remove obsolete TODOs for b/38173783 The comments in the ticket indicate that this behavior is fine and that the ticket should be closed, so we shouldn't need pointers to the ticket. PiperOrigin-RevId: 306266071 --- pkg/context/context.go | 4 ---- pkg/sentry/arch/stack.go | 3 --- pkg/sentry/fs/gofer/file_state.go | 1 - pkg/sentry/fs/gofer/handles.go | 1 - pkg/sentry/fs/gofer/inode_state.go | 1 - pkg/sentry/fs/gofer/session_state.go | 1 - pkg/sentry/fs/inode.go | 1 - pkg/sentry/kernel/shm/shm.go | 2 +- pkg/sentry/kernel/task_context.go | 1 - pkg/sentry/kernel/task_signals.go | 2 -- pkg/usermem/usermem.go | 3 --- 11 files changed, 1 insertion(+), 19 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/context/context.go b/pkg/context/context.go index 23e009ef3..5319b6d8d 100644 --- a/pkg/context/context.go +++ b/pkg/context/context.go @@ -127,10 +127,6 @@ func (logContext) Value(key interface{}) interface{} { var bgContext = &logContext{Logger: log.Log()} // Background returns an empty context using the default logger. -// -// Users should be wary of using a Background context. Please tag any use with -// FIXME(b/38173783) and a note to remove this use. -// // Generally, one should use the Task as their context when available, or avoid // having to use a context in places where a Task is unavailable. // diff --git a/pkg/sentry/arch/stack.go b/pkg/sentry/arch/stack.go index 09bceabc9..1108fa0bd 100644 --- a/pkg/sentry/arch/stack.go +++ b/pkg/sentry/arch/stack.go @@ -97,7 +97,6 @@ func (s *Stack) Push(vals ...interface{}) (usermem.Addr, error) { if c < 0 { return 0, fmt.Errorf("bad binary.Size for %T", v) } - // TODO(b/38173783): Use a real context.Context. n, err := usermem.CopyObjectOut(context.Background(), s.IO, s.Bottom-usermem.Addr(c), norm, usermem.IOOpts{}) if err != nil || c != n { return 0, err @@ -121,11 +120,9 @@ func (s *Stack) Pop(vals ...interface{}) (usermem.Addr, error) { var err error if isVaddr { value := s.Arch.Native(uintptr(0)) - // TODO(b/38173783): Use a real context.Context. n, err = usermem.CopyObjectIn(context.Background(), s.IO, s.Bottom, value, usermem.IOOpts{}) *vaddr = usermem.Addr(s.Arch.Value(value)) } else { - // TODO(b/38173783): Use a real context.Context. n, err = usermem.CopyObjectIn(context.Background(), s.IO, s.Bottom, v, usermem.IOOpts{}) } if err != nil { diff --git a/pkg/sentry/fs/gofer/file_state.go b/pkg/sentry/fs/gofer/file_state.go index ff96b28ba..edd6576aa 100644 --- a/pkg/sentry/fs/gofer/file_state.go +++ b/pkg/sentry/fs/gofer/file_state.go @@ -34,7 +34,6 @@ func (f *fileOperations) afterLoad() { flags := f.flags flags.Truncate = false - // TODO(b/38173783): Context is not plumbed to save/restore. f.handles, err = f.inodeOperations.fileState.getHandles(context.Background(), flags, f.inodeOperations.cachingInodeOps) if err != nil { return fmt.Errorf("failed to re-open handle: %v", err) diff --git a/pkg/sentry/fs/gofer/handles.go b/pkg/sentry/fs/gofer/handles.go index 9f7c3e89f..fc14249be 100644 --- a/pkg/sentry/fs/gofer/handles.go +++ b/pkg/sentry/fs/gofer/handles.go @@ -57,7 +57,6 @@ func (h *handles) DecRef() { } } } - // FIXME(b/38173783): Context is not plumbed here. if err := h.File.close(context.Background()); err != nil { log.Warningf("error closing p9 file: %v", err) } diff --git a/pkg/sentry/fs/gofer/inode_state.go b/pkg/sentry/fs/gofer/inode_state.go index 238f7804c..a3402e343 100644 --- a/pkg/sentry/fs/gofer/inode_state.go +++ b/pkg/sentry/fs/gofer/inode_state.go @@ -123,7 +123,6 @@ func (i *inodeFileState) afterLoad() { // beforeSave. return fmt.Errorf("failed to find path for inode number %d. Device %s contains %s", i.sattr.InodeID, i.s.connID, fs.InodeMappings(i.s.inodeMappings)) } - // TODO(b/38173783): Context is not plumbed to save/restore. ctx := &dummyClockContext{context.Background()} _, i.file, err = i.s.attach.walk(ctx, splitAbsolutePath(name)) diff --git a/pkg/sentry/fs/gofer/session_state.go b/pkg/sentry/fs/gofer/session_state.go index 111da59f9..2d398b753 100644 --- a/pkg/sentry/fs/gofer/session_state.go +++ b/pkg/sentry/fs/gofer/session_state.go @@ -104,7 +104,6 @@ func (s *session) afterLoad() { // If private unix sockets are enabled, create and fill the session's endpoint // maps. if opts.privateunixsocket { - // TODO(b/38173783): Context is not plumbed to save/restore. ctx := &dummyClockContext{context.Background()} if err = s.restoreEndpointMaps(ctx); err != nil { diff --git a/pkg/sentry/fs/inode.go b/pkg/sentry/fs/inode.go index 73f89abcc..a34fbc946 100644 --- a/pkg/sentry/fs/inode.go +++ b/pkg/sentry/fs/inode.go @@ -102,7 +102,6 @@ func (i *Inode) DecRef() { // destroy releases the Inode and releases the msrc reference taken. func (i *Inode) destroy() { - // FIXME(b/38173783): Context is not plumbed here. ctx := context.Background() if err := i.WriteOut(ctx); err != nil { // FIXME(b/65209558): Mark as warning again once noatime is diff --git a/pkg/sentry/kernel/shm/shm.go b/pkg/sentry/kernel/shm/shm.go index 208569057..f66cfcc7f 100644 --- a/pkg/sentry/kernel/shm/shm.go +++ b/pkg/sentry/kernel/shm/shm.go @@ -461,7 +461,7 @@ func (s *Shm) AddMapping(ctx context.Context, _ memmap.MappingSpace, _ usermem.A func (s *Shm) RemoveMapping(ctx context.Context, _ memmap.MappingSpace, _ usermem.AddrRange, _ uint64, _ bool) { s.mu.Lock() defer s.mu.Unlock() - // TODO(b/38173783): RemoveMapping may be called during task exit, when ctx + // RemoveMapping may be called during task exit, when ctx // is context.Background. Gracefully handle missing clocks. Failing to // update the detach time in these cases is ok, since no one can observe the // omission. diff --git a/pkg/sentry/kernel/task_context.go b/pkg/sentry/kernel/task_context.go index c115e8d1f..9fa528384 100644 --- a/pkg/sentry/kernel/task_context.go +++ b/pkg/sentry/kernel/task_context.go @@ -58,7 +58,6 @@ func (tc *TaskContext) release() { // Nil out pointers so that if the task is saved after release, it doesn't // follow the pointers to possibly now-invalid objects. if tc.MemoryManager != nil { - // TODO(b/38173783) tc.MemoryManager.DecUsers(context.Background()) tc.MemoryManager = nil } diff --git a/pkg/sentry/kernel/task_signals.go b/pkg/sentry/kernel/task_signals.go index 8802db142..6aa798346 100644 --- a/pkg/sentry/kernel/task_signals.go +++ b/pkg/sentry/kernel/task_signals.go @@ -513,8 +513,6 @@ func (t *Task) canReceiveSignalLocked(sig linux.Signal) bool { if t.stop != nil { return false } - // - TODO(b/38173783): No special case for when t is also the sending task, - // because the identity of the sender is unknown. // - Do not choose tasks that have already been interrupted, as they may be // busy handling another signal. if len(t.interruptChan) != 0 { diff --git a/pkg/usermem/usermem.go b/pkg/usermem/usermem.go index d2f4403b0..cd6a0ea6b 100644 --- a/pkg/usermem/usermem.go +++ b/pkg/usermem/usermem.go @@ -29,9 +29,6 @@ import ( ) // IO provides access to the contents of a virtual memory space. -// -// FIXME(b/38173783): Implementations of IO cannot expect ctx to contain any -// meaningful data. type IO interface { // CopyOut copies len(src) bytes from src to the memory mapped at addr. It // returns the number of bytes copied. If the number of bytes copied is < -- cgit v1.2.3 From 5d885d7fb21414d903d57ffe2b95bcc62c098d6a Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Mon, 13 Apr 2020 13:01:28 -0700 Subject: Port socket-related syscalls to VFS2. Note that most kinds of sockets are not yet supported in VFS2 (only Unix sockets are partially supported at the moment), so these syscalls will still generally fail. Enabling them allows us to begin running socket tests for VFS2 as more features are ported over. Updates #1476, #1478, #1484, #1485. PiperOrigin-RevId: 306292294 --- pkg/sentry/kernel/fd_table.go | 55 + pkg/sentry/kernel/task.go | 9 + pkg/sentry/syscalls/linux/sys_socket.go | 9 +- pkg/sentry/syscalls/linux/vfs2/BUILD | 6 + .../syscalls/linux/vfs2/linux64_override_amd64.go | 40 +- pkg/sentry/syscalls/linux/vfs2/socket.go | 1138 ++++++++++++++++++++ 6 files changed, 1238 insertions(+), 19 deletions(-) create mode 100644 pkg/sentry/syscalls/linux/vfs2/socket.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/fd_table.go b/pkg/sentry/kernel/fd_table.go index d09d97825..ed40b5303 100644 --- a/pkg/sentry/kernel/fd_table.go +++ b/pkg/sentry/kernel/fd_table.go @@ -307,6 +307,61 @@ func (f *FDTable) NewFDs(ctx context.Context, fd int32, files []*fs.File, flags return fds, nil } +// NewFDsVFS2 allocates new FDs guaranteed to be the lowest number available +// greater than or equal to the fd parameter. All files will share the set +// flags. Success is guaranteed to be all or none. +func (f *FDTable) NewFDsVFS2(ctx context.Context, fd int32, files []*vfs.FileDescription, flags FDFlags) (fds []int32, err error) { + if fd < 0 { + // Don't accept negative FDs. + return nil, syscall.EINVAL + } + + // Default limit. + end := int32(math.MaxInt32) + + // Ensure we don't get past the provided limit. + if limitSet := limits.FromContext(ctx); limitSet != nil { + lim := limitSet.Get(limits.NumberOfFiles) + if lim.Cur != limits.Infinity { + end = int32(lim.Cur) + } + if fd >= end { + return nil, syscall.EMFILE + } + } + + f.mu.Lock() + defer f.mu.Unlock() + + // From f.next to find available fd. + if fd < f.next { + fd = f.next + } + + // Install all entries. + for i := fd; i < end && len(fds) < len(files); i++ { + if d, _, _ := f.getVFS2(i); d == nil { + f.setVFS2(i, files[len(fds)], flags) // Set the descriptor. + fds = append(fds, i) // Record the file descriptor. + } + } + + // Failure? Unwind existing FDs. + if len(fds) < len(files) { + for _, i := range fds { + f.setVFS2(i, nil, FDFlags{}) // Zap entry. + } + return nil, syscall.EMFILE + } + + if fd == f.next { + // Update next search start position. + f.next = fds[len(fds)-1] + 1 + } + + return fds, nil +} + // NewFDVFS2 allocates a file descriptor greater than or equal to minfd for // the given file description. If it succeeds, it takes a reference on file. func (f *FDTable) NewFDVFS2(ctx context.Context, minfd int32, file *vfs.FileDescription, flags FDFlags) (int32, error) { diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index d6546735e..e5d133d6c 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -777,6 +777,15 @@ func (t *Task) NewFDs(fd int32, files []*fs.File, flags FDFlags) ([]int32, error return t.fdTable.NewFDs(t, fd, files, flags) } +// NewFDsVFS2 is a convenience wrapper for t.FDTable().NewFDsVFS2. +// +// This automatically passes the task as the context. +// +// Precondition: same as FDTable. +func (t *Task) NewFDsVFS2(fd int32, files []*vfs.FileDescription, flags FDFlags) ([]int32, error) { + return t.fdTable.NewFDsVFS2(t, fd, files, flags) +} + // NewFDFrom is a convenience wrapper for t.FDTable().NewFDs with a single file. // // This automatically passes the task as the context. diff --git a/pkg/sentry/syscalls/linux/sys_socket.go b/pkg/sentry/syscalls/linux/sys_socket.go index 2919228d0..61b2576ac 100644 --- a/pkg/sentry/syscalls/linux/sys_socket.go +++ b/pkg/sentry/syscalls/linux/sys_socket.go @@ -31,6 +31,8 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) +// LINT.IfChange + // minListenBacklog is the minimum reasonable backlog for listening sockets. const minListenBacklog = 8 @@ -244,7 +246,10 @@ func SocketPair(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy // Copy the file descriptors out. if _, err := t.CopyOut(socks, fds); err != nil { - // Note that we don't close files here; see pipe(2) also. + for _, fd := range fds { + _, file := t.FDTable().Remove(fd) + file.DecRef() + } return 0, nil, err } @@ -1128,3 +1133,5 @@ func SendTo(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal n, err := sendTo(t, fd, bufPtr, bufLen, flags, namePtr, nameLen) return n, nil, err } + +// LINT.ThenChange(./vfs2/socket.go) diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD index 0004e60d9..b32abfe59 100644 --- a/pkg/sentry/syscalls/linux/vfs2/BUILD +++ b/pkg/sentry/syscalls/linux/vfs2/BUILD @@ -21,6 +21,7 @@ go_library( "poll.go", "read_write.go", "setstat.go", + "socket.go", "stat.go", "stat_amd64.go", "stat_arm64.go", @@ -32,6 +33,7 @@ go_library( visibility = ["//:sandbox"], deps = [ "//pkg/abi/linux", + "//pkg/binary", "//pkg/bits", "//pkg/fspath", "//pkg/gohacks", @@ -43,10 +45,14 @@ go_library( "//pkg/sentry/limits", "//pkg/sentry/loader", "//pkg/sentry/memmap", + "//pkg/sentry/socket", + "//pkg/sentry/socket/control", + "//pkg/sentry/socket/unix/transport", "//pkg/sentry/syscalls", "//pkg/sentry/syscalls/linux", "//pkg/sentry/vfs", "//pkg/sync", + "//pkg/syserr", "//pkg/syserror", "//pkg/usermem", "//pkg/waiter", diff --git a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go index 63febc2f7..645e0bcb8 100644 --- a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go +++ b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go @@ -44,21 +44,22 @@ func Override(table map[uintptr]kernel.Syscall) { table[32] = syscalls.Supported("dup", Dup) table[33] = syscalls.Supported("dup2", Dup2) delete(table, 40) // sendfile - delete(table, 41) // socket - delete(table, 42) // connect - delete(table, 43) // accept - delete(table, 44) // sendto - delete(table, 45) // recvfrom - delete(table, 46) // sendmsg - delete(table, 47) // recvmsg - delete(table, 48) // shutdown - delete(table, 49) // bind - delete(table, 50) // listen - delete(table, 51) // getsockname - delete(table, 52) // getpeername - delete(table, 53) // socketpair - delete(table, 54) // setsockopt - delete(table, 55) // getsockopt + // TODO(gvisor.dev/issue/1485): Port all socket variants to VFS2. + table[41] = syscalls.PartiallySupported("socket", Socket, "In process of porting socket syscalls to VFS2.", nil) + table[42] = syscalls.PartiallySupported("connect", Connect, "In process of porting socket syscalls to VFS2.", nil) + table[43] = syscalls.PartiallySupported("accept", Accept, "In process of porting socket syscalls to VFS2.", nil) + table[44] = syscalls.PartiallySupported("sendto", SendTo, "In process of porting socket syscalls to VFS2.", nil) + table[45] = syscalls.PartiallySupported("recvfrom", RecvFrom, "In process of porting socket syscalls to VFS2.", nil) + table[46] = syscalls.PartiallySupported("sendmsg", SendMsg, "In process of porting socket syscalls to VFS2.", nil) + table[47] = syscalls.PartiallySupported("recvmsg", RecvMsg, "In process of porting socket syscalls to VFS2.", nil) + table[48] = syscalls.PartiallySupported("shutdown", Shutdown, "In process of porting socket syscalls to VFS2.", nil) + table[49] = syscalls.PartiallySupported("bind", Bind, "In process of porting socket syscalls to VFS2.", nil) + table[50] = syscalls.PartiallySupported("listen", Listen, "In process of porting socket syscalls to VFS2.", nil) + table[51] = syscalls.PartiallySupported("getsockname", GetSockName, "In process of porting socket syscalls to VFS2.", nil) + table[52] = syscalls.PartiallySupported("getpeername", GetPeerName, "In process of porting socket syscalls to VFS2.", nil) + table[53] = syscalls.PartiallySupported("socketpair", SocketPair, "In process of porting socket syscalls to VFS2.", nil) + table[54] = syscalls.PartiallySupported("getsockopt", GetSockOpt, "In process of porting socket syscalls to VFS2.", nil) + table[55] = syscalls.PartiallySupported("setsockopt", SetSockOpt, "In process of porting socket syscalls to VFS2.", nil) table[59] = syscalls.Supported("execve", Execve) table[72] = syscalls.Supported("fcntl", Fcntl) delete(table, 73) // flock @@ -144,7 +145,8 @@ func Override(table map[uintptr]kernel.Syscall) { delete(table, 285) // fallocate table[286] = syscalls.Supported("timerfd_settime", TimerfdSettime) table[287] = syscalls.Supported("timerfd_gettime", TimerfdGettime) - delete(table, 288) // accept4 + // TODO(gvisor.dev/issue/1485): Port all socket variants to VFS2. + table[288] = syscalls.PartiallySupported("accept4", Accept4, "In process of porting socket syscalls to VFS2.", nil) delete(table, 289) // signalfd4 delete(table, 290) // eventfd2 table[291] = syscalls.Supported("epoll_create1", EpollCreate1) @@ -153,9 +155,11 @@ func Override(table map[uintptr]kernel.Syscall) { delete(table, 294) // inotify_init1 table[295] = syscalls.Supported("preadv", Preadv) table[296] = syscalls.Supported("pwritev", Pwritev) - delete(table, 299) // recvmmsg + // TODO(gvisor.dev/issue/1485): Port all socket variants to VFS2. + table[299] = syscalls.PartiallySupported("recvmmsg", RecvMMsg, "In process of porting socket syscalls to VFS2.", nil) table[306] = syscalls.Supported("syncfs", Syncfs) - delete(table, 307) // sendmmsg + // TODO(gvisor.dev/issue/1485): Port all socket variants to VFS2. + table[307] = syscalls.PartiallySupported("sendmmsg", SendMMsg, "In process of porting socket syscalls to VFS2.", nil) table[316] = syscalls.Supported("renameat2", Renameat2) delete(table, 319) // memfd_create table[322] = syscalls.Supported("execveat", Execveat) diff --git a/pkg/sentry/syscalls/linux/vfs2/socket.go b/pkg/sentry/syscalls/linux/vfs2/socket.go new file mode 100644 index 000000000..79a4a7ada --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/socket.go @@ -0,0 +1,1138 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "time" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" + "gvisor.dev/gvisor/pkg/sentry/socket" + "gvisor.dev/gvisor/pkg/sentry/socket/control" + "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" + slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserr" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +// minListenBacklog is the minimum reasonable backlog for listening sockets. +const minListenBacklog = 8 + +// maxListenBacklog is the maximum allowed backlog for listening sockets. +const maxListenBacklog = 1024 + +// maxAddrLen is the maximum socket address length we're willing to accept. +const maxAddrLen = 200 + +// maxOptLen is the maximum sockopt parameter length we're willing to accept. +const maxOptLen = 1024 * 8 + +// maxControlLen is the maximum length of the msghdr.msg_control buffer we're +// willing to accept. Note that this limit is smaller than Linux, which allows +// buffers upto INT_MAX. +const maxControlLen = 10 * 1024 * 1024 + +// nameLenOffset is the offset from the start of the MessageHeader64 struct to +// the NameLen field. +const nameLenOffset = 8 + +// controlLenOffset is the offset form the start of the MessageHeader64 struct +// to the ControlLen field. +const controlLenOffset = 40 + +// flagsOffset is the offset form the start of the MessageHeader64 struct +// to the Flags field. +const flagsOffset = 48 + +const sizeOfInt32 = 4 + +// messageHeader64Len is the length of a MessageHeader64 struct. +var messageHeader64Len = uint64(binary.Size(MessageHeader64{})) + +// multipleMessageHeader64Len is the length of a multipeMessageHeader64 struct. +var multipleMessageHeader64Len = uint64(binary.Size(multipleMessageHeader64{})) + +// baseRecvFlags are the flags that are accepted across recvmsg(2), +// recvmmsg(2), and recvfrom(2). +const baseRecvFlags = linux.MSG_OOB | linux.MSG_DONTROUTE | linux.MSG_DONTWAIT | linux.MSG_NOSIGNAL | linux.MSG_WAITALL | linux.MSG_TRUNC | linux.MSG_CTRUNC + +// MessageHeader64 is the 64-bit representation of the msghdr struct used in +// the recvmsg and sendmsg syscalls. +type MessageHeader64 struct { + // Name is the optional pointer to a network address buffer. + Name uint64 + + // NameLen is the length of the buffer pointed to by Name. + NameLen uint32 + _ uint32 + + // Iov is a pointer to an array of io vectors that describe the memory + // locations involved in the io operation. + Iov uint64 + + // IovLen is the length of the array pointed to by Iov. + IovLen uint64 + + // Control is the optional pointer to ancillary control data. + Control uint64 + + // ControlLen is the length of the data pointed to by Control. + ControlLen uint64 + + // Flags on the sent/received message. + Flags int32 + _ int32 +} + +// multipleMessageHeader64 is the 64-bit representation of the mmsghdr struct used in +// the recvmmsg and sendmmsg syscalls. +type multipleMessageHeader64 struct { + msgHdr MessageHeader64 + msgLen uint32 + _ int32 +} + +// CopyInMessageHeader64 copies a message header from user to kernel memory. +func CopyInMessageHeader64(t *kernel.Task, addr usermem.Addr, msg *MessageHeader64) error { + b := t.CopyScratchBuffer(52) + if _, err := t.CopyInBytes(addr, b); err != nil { + return err + } + + msg.Name = usermem.ByteOrder.Uint64(b[0:]) + msg.NameLen = usermem.ByteOrder.Uint32(b[8:]) + msg.Iov = usermem.ByteOrder.Uint64(b[16:]) + msg.IovLen = usermem.ByteOrder.Uint64(b[24:]) + msg.Control = usermem.ByteOrder.Uint64(b[32:]) + msg.ControlLen = usermem.ByteOrder.Uint64(b[40:]) + msg.Flags = int32(usermem.ByteOrder.Uint32(b[48:])) + + return nil +} + +// CaptureAddress allocates memory for and copies a socket address structure +// from the untrusted address space range. +func CaptureAddress(t *kernel.Task, addr usermem.Addr, addrlen uint32) ([]byte, error) { + if addrlen > maxAddrLen { + return nil, syserror.EINVAL + } + + addrBuf := make([]byte, addrlen) + if _, err := t.CopyInBytes(addr, addrBuf); err != nil { + return nil, err + } + + return addrBuf, nil +} + +// writeAddress writes a sockaddr structure and its length to an output buffer +// in the unstrusted address space range. If the address is bigger than the +// buffer, it is truncated. +func writeAddress(t *kernel.Task, addr interface{}, addrLen uint32, addrPtr usermem.Addr, addrLenPtr usermem.Addr) error { + // Get the buffer length. + var bufLen uint32 + if _, err := t.CopyIn(addrLenPtr, &bufLen); err != nil { + return err + } + + if int32(bufLen) < 0 { + return syserror.EINVAL + } + + // Write the length unconditionally. + if _, err := t.CopyOut(addrLenPtr, addrLen); err != nil { + return err + } + + if addr == nil { + return nil + } + + if bufLen > addrLen { + bufLen = addrLen + } + + // Copy as much of the address as will fit in the buffer. + encodedAddr := binary.Marshal(nil, usermem.ByteOrder, addr) + if bufLen > uint32(len(encodedAddr)) { + bufLen = uint32(len(encodedAddr)) + } + _, err := t.CopyOutBytes(addrPtr, encodedAddr[:int(bufLen)]) + return err +} + +// Socket implements the linux syscall socket(2). +func Socket(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + domain := int(args[0].Int()) + stype := args[1].Int() + protocol := int(args[2].Int()) + + // Check and initialize the flags. + if stype & ^(0xf|linux.SOCK_NONBLOCK|linux.SOCK_CLOEXEC) != 0 { + return 0, nil, syserror.EINVAL + } + + // Create the new socket. + s, e := socket.NewVFS2(t, domain, linux.SockType(stype&0xf), protocol) + if e != nil { + return 0, nil, e.ToError() + } + defer s.DecRef() + + if err := s.SetStatusFlags(t, t.Credentials(), uint32(stype&linux.SOCK_NONBLOCK)); err != nil { + return 0, nil, err + } + + fd, err := t.NewFDFromVFS2(0, s, kernel.FDFlags{ + CloseOnExec: stype&linux.SOCK_CLOEXEC != 0, + }) + if err != nil { + return 0, nil, err + } + + return uintptr(fd), nil, nil +} + +// SocketPair implements the linux syscall socketpair(2). +func SocketPair(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + domain := int(args[0].Int()) + stype := args[1].Int() + protocol := int(args[2].Int()) + addr := args[3].Pointer() + + // Check and initialize the flags. + if stype & ^(0xf|linux.SOCK_NONBLOCK|linux.SOCK_CLOEXEC) != 0 { + return 0, nil, syserror.EINVAL + } + + // Create the socket pair. + s1, s2, e := socket.PairVFS2(t, domain, linux.SockType(stype&0xf), protocol) + if e != nil { + return 0, nil, e.ToError() + } + // Adding to the FD table will cause an extra reference to be acquired. + defer s1.DecRef() + defer s2.DecRef() + + nonblocking := uint32(stype & linux.SOCK_NONBLOCK) + if err := s1.SetStatusFlags(t, t.Credentials(), nonblocking); err != nil { + return 0, nil, err + } + if err := s2.SetStatusFlags(t, t.Credentials(), nonblocking); err != nil { + return 0, nil, err + } + + // Create the FDs for the sockets. + flags := kernel.FDFlags{ + CloseOnExec: stype&linux.SOCK_CLOEXEC != 0, + } + fds, err := t.NewFDsVFS2(0, []*vfs.FileDescription{s1, s2}, flags) + if err != nil { + return 0, nil, err + } + + if _, err := t.CopyOut(addr, fds); err != nil { + for _, fd := range fds { + _, file := t.FDTable().Remove(fd) + file.DecRef() + } + return 0, nil, err + } + + return 0, nil, nil +} + +// Connect implements the linux syscall connect(2). +func Connect(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + addr := args[1].Pointer() + addrlen := args[2].Uint() + + // Get socket from the file descriptor. + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Extract the socket. + s, ok := file.Impl().(socket.SocketVFS2) + if !ok { + return 0, nil, syserror.ENOTSOCK + } + + // Capture address and call syscall implementation. + a, err := CaptureAddress(t, addr, addrlen) + if err != nil { + return 0, nil, err + } + + blocking := (file.StatusFlags() & linux.SOCK_NONBLOCK) == 0 + return 0, nil, syserror.ConvertIntr(s.Connect(t, a, blocking).ToError(), kernel.ERESTARTSYS) +} + +// accept is the implementation of the accept syscall. It is called by accept +// and accept4 syscall handlers. +func accept(t *kernel.Task, fd int32, addr usermem.Addr, addrLen usermem.Addr, flags int) (uintptr, error) { + // Check that no unsupported flags are passed in. + if flags & ^(linux.SOCK_NONBLOCK|linux.SOCK_CLOEXEC) != 0 { + return 0, syserror.EINVAL + } + + // Get socket from the file descriptor. + file := t.GetFileVFS2(fd) + if file == nil { + return 0, syserror.EBADF + } + defer file.DecRef() + + // Extract the socket. + s, ok := file.Impl().(socket.SocketVFS2) + if !ok { + return 0, syserror.ENOTSOCK + } + + // Call the syscall implementation for this socket, then copy the + // output address if one is specified. + blocking := (file.StatusFlags() & linux.SOCK_NONBLOCK) == 0 + + peerRequested := addrLen != 0 + nfd, peer, peerLen, e := s.Accept(t, peerRequested, flags, blocking) + if e != nil { + return 0, syserror.ConvertIntr(e.ToError(), kernel.ERESTARTSYS) + } + if peerRequested { + // NOTE(magi): Linux does not give you an error if it can't + // write the data back out so neither do we. + if err := writeAddress(t, peer, peerLen, addr, addrLen); err == syserror.EINVAL { + return 0, err + } + } + return uintptr(nfd), nil +} + +// Accept4 implements the linux syscall accept4(2). +func Accept4(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + addr := args[1].Pointer() + addrlen := args[2].Pointer() + flags := int(args[3].Int()) + + n, err := accept(t, fd, addr, addrlen, flags) + return n, nil, err +} + +// Accept implements the linux syscall accept(2). +func Accept(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + addr := args[1].Pointer() + addrlen := args[2].Pointer() + + n, err := accept(t, fd, addr, addrlen, 0) + return n, nil, err +} + +// Bind implements the linux syscall bind(2). +func Bind(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + addr := args[1].Pointer() + addrlen := args[2].Uint() + + // Get socket from the file descriptor. + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Extract the socket. + s, ok := file.Impl().(socket.SocketVFS2) + if !ok { + return 0, nil, syserror.ENOTSOCK + } + + // Capture address and call syscall implementation. + a, err := CaptureAddress(t, addr, addrlen) + if err != nil { + return 0, nil, err + } + + return 0, nil, s.Bind(t, a).ToError() +} + +// Listen implements the linux syscall listen(2). +func Listen(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + backlog := args[1].Int() + + // Get socket from the file descriptor. + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Extract the socket. + s, ok := file.Impl().(socket.SocketVFS2) + if !ok { + return 0, nil, syserror.ENOTSOCK + } + + // Per Linux, the backlog is silently capped to reasonable values. + if backlog <= 0 { + backlog = minListenBacklog + } + if backlog > maxListenBacklog { + backlog = maxListenBacklog + } + + return 0, nil, s.Listen(t, int(backlog)).ToError() +} + +// Shutdown implements the linux syscall shutdown(2). +func Shutdown(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + how := args[1].Int() + + // Get socket from the file descriptor. + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Extract the socket. + s, ok := file.Impl().(socket.SocketVFS2) + if !ok { + return 0, nil, syserror.ENOTSOCK + } + + // Validate how, then call syscall implementation. + switch how { + case linux.SHUT_RD, linux.SHUT_WR, linux.SHUT_RDWR: + default: + return 0, nil, syserror.EINVAL + } + + return 0, nil, s.Shutdown(t, int(how)).ToError() +} + +// GetSockOpt implements the linux syscall getsockopt(2). +func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + level := args[1].Int() + name := args[2].Int() + optValAddr := args[3].Pointer() + optLenAddr := args[4].Pointer() + + // Get socket from the file descriptor. + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Extract the socket. + s, ok := file.Impl().(socket.SocketVFS2) + if !ok { + return 0, nil, syserror.ENOTSOCK + } + + // Read the length. Reject negative values. + optLen := int32(0) + if _, err := t.CopyIn(optLenAddr, &optLen); err != nil { + return 0, nil, err + } + if optLen < 0 { + return 0, nil, syserror.EINVAL + } + + // Call syscall implementation then copy both value and value len out. + v, e := getSockOpt(t, s, int(level), int(name), optValAddr, int(optLen)) + if e != nil { + return 0, nil, e.ToError() + } + + vLen := int32(binary.Size(v)) + if _, err := t.CopyOut(optLenAddr, vLen); err != nil { + return 0, nil, err + } + + if v != nil { + if _, err := t.CopyOut(optValAddr, v); err != nil { + return 0, nil, err + } + } + + return 0, nil, nil +} + +// getSockOpt tries to handle common socket options, or dispatches to a specific +// socket implementation. +func getSockOpt(t *kernel.Task, s socket.SocketVFS2, level, name int, optValAddr usermem.Addr, len int) (interface{}, *syserr.Error) { + if level == linux.SOL_SOCKET { + switch name { + case linux.SO_TYPE, linux.SO_DOMAIN, linux.SO_PROTOCOL: + if len < sizeOfInt32 { + return nil, syserr.ErrInvalidArgument + } + } + + switch name { + case linux.SO_TYPE: + _, skType, _ := s.Type() + return int32(skType), nil + case linux.SO_DOMAIN: + family, _, _ := s.Type() + return int32(family), nil + case linux.SO_PROTOCOL: + _, _, protocol := s.Type() + return int32(protocol), nil + } + } + + return s.GetSockOpt(t, level, name, optValAddr, len) +} + +// SetSockOpt implements the linux syscall setsockopt(2). +// +// Note that unlike Linux, enabling SO_PASSCRED does not autobind the socket. +func SetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + level := args[1].Int() + name := args[2].Int() + optValAddr := args[3].Pointer() + optLen := args[4].Int() + + // Get socket from the file descriptor. + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Extract the socket. + s, ok := file.Impl().(socket.SocketVFS2) + if !ok { + return 0, nil, syserror.ENOTSOCK + } + + if optLen < 0 { + return 0, nil, syserror.EINVAL + } + if optLen > maxOptLen { + return 0, nil, syserror.EINVAL + } + buf := t.CopyScratchBuffer(int(optLen)) + if _, err := t.CopyIn(optValAddr, &buf); err != nil { + return 0, nil, err + } + + // Call syscall implementation. + if err := s.SetSockOpt(t, int(level), int(name), buf); err != nil { + return 0, nil, err.ToError() + } + + return 0, nil, nil +} + +// GetSockName implements the linux syscall getsockname(2). +func GetSockName(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + addr := args[1].Pointer() + addrlen := args[2].Pointer() + + // Get socket from the file descriptor. + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Extract the socket. + s, ok := file.Impl().(socket.SocketVFS2) + if !ok { + return 0, nil, syserror.ENOTSOCK + } + + // Get the socket name and copy it to the caller. + v, vl, err := s.GetSockName(t) + if err != nil { + return 0, nil, err.ToError() + } + + return 0, nil, writeAddress(t, v, vl, addr, addrlen) +} + +// GetPeerName implements the linux syscall getpeername(2). +func GetPeerName(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + addr := args[1].Pointer() + addrlen := args[2].Pointer() + + // Get socket from the file descriptor. + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Extract the socket. + s, ok := file.Impl().(socket.SocketVFS2) + if !ok { + return 0, nil, syserror.ENOTSOCK + } + + // Get the socket peer name and copy it to the caller. + v, vl, err := s.GetPeerName(t) + if err != nil { + return 0, nil, err.ToError() + } + + return 0, nil, writeAddress(t, v, vl, addr, addrlen) +} + +// RecvMsg implements the linux syscall recvmsg(2). +func RecvMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + msgPtr := args[1].Pointer() + flags := args[2].Int() + + if t.Arch().Width() != 8 { + // We only handle 64-bit for now. + return 0, nil, syserror.EINVAL + } + + // Get socket from the file descriptor. + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Extract the socket. + s, ok := file.Impl().(socket.SocketVFS2) + if !ok { + return 0, nil, syserror.ENOTSOCK + } + + // Reject flags that we don't handle yet. + if flags & ^(baseRecvFlags|linux.MSG_PEEK|linux.MSG_CMSG_CLOEXEC|linux.MSG_ERRQUEUE) != 0 { + return 0, nil, syserror.EINVAL + } + + if (file.StatusFlags() & linux.SOCK_NONBLOCK) != 0 { + flags |= linux.MSG_DONTWAIT + } + + var haveDeadline bool + var deadline ktime.Time + if dl := s.RecvTimeout(); dl > 0 { + deadline = t.Kernel().MonotonicClock().Now().Add(time.Duration(dl) * time.Nanosecond) + haveDeadline = true + } else if dl < 0 { + flags |= linux.MSG_DONTWAIT + } + + n, err := recvSingleMsg(t, s, msgPtr, flags, haveDeadline, deadline) + return n, nil, err +} + +// RecvMMsg implements the linux syscall recvmmsg(2). +func RecvMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + msgPtr := args[1].Pointer() + vlen := args[2].Uint() + flags := args[3].Int() + toPtr := args[4].Pointer() + + if t.Arch().Width() != 8 { + // We only handle 64-bit for now. + return 0, nil, syserror.EINVAL + } + + // Reject flags that we don't handle yet. + if flags & ^(baseRecvFlags|linux.MSG_CMSG_CLOEXEC|linux.MSG_ERRQUEUE) != 0 { + return 0, nil, syserror.EINVAL + } + + // Get socket from the file descriptor. + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Extract the socket. + s, ok := file.Impl().(socket.SocketVFS2) + if !ok { + return 0, nil, syserror.ENOTSOCK + } + + if (file.StatusFlags() & linux.SOCK_NONBLOCK) != 0 { + flags |= linux.MSG_DONTWAIT + } + + var haveDeadline bool + var deadline ktime.Time + if toPtr != 0 { + var ts linux.Timespec + if _, err := ts.CopyIn(t, toPtr); err != nil { + return 0, nil, err + } + if !ts.Valid() { + return 0, nil, syserror.EINVAL + } + deadline = t.Kernel().MonotonicClock().Now().Add(ts.ToDuration()) + haveDeadline = true + } + + if !haveDeadline { + if dl := s.RecvTimeout(); dl > 0 { + deadline = t.Kernel().MonotonicClock().Now().Add(time.Duration(dl) * time.Nanosecond) + haveDeadline = true + } else if dl < 0 { + flags |= linux.MSG_DONTWAIT + } + } + + var count uint32 + var err error + for i := uint64(0); i < uint64(vlen); i++ { + mp, ok := msgPtr.AddLength(i * multipleMessageHeader64Len) + if !ok { + return 0, nil, syserror.EFAULT + } + var n uintptr + if n, err = recvSingleMsg(t, s, mp, flags, haveDeadline, deadline); err != nil { + break + } + + // Copy the received length to the caller. + lp, ok := mp.AddLength(messageHeader64Len) + if !ok { + return 0, nil, syserror.EFAULT + } + if _, err = t.CopyOut(lp, uint32(n)); err != nil { + break + } + count++ + } + + if count == 0 { + return 0, nil, err + } + return uintptr(count), nil, nil +} + +func recvSingleMsg(t *kernel.Task, s socket.SocketVFS2, msgPtr usermem.Addr, flags int32, haveDeadline bool, deadline ktime.Time) (uintptr, error) { + // Capture the message header and io vectors. + var msg MessageHeader64 + if err := CopyInMessageHeader64(t, msgPtr, &msg); err != nil { + return 0, err + } + + if msg.IovLen > linux.UIO_MAXIOV { + return 0, syserror.EMSGSIZE + } + dst, err := t.IovecsIOSequence(usermem.Addr(msg.Iov), int(msg.IovLen), usermem.IOOpts{ + AddressSpaceActive: true, + }) + if err != nil { + return 0, err + } + + // FIXME(b/63594852): Pretend we have an empty error queue. + if flags&linux.MSG_ERRQUEUE != 0 { + return 0, syserror.EAGAIN + } + + // Fast path when no control message nor name buffers are provided. + if msg.ControlLen == 0 && msg.NameLen == 0 { + n, mflags, _, _, cms, err := s.RecvMsg(t, dst, int(flags), haveDeadline, deadline, false, 0) + if err != nil { + return 0, syserror.ConvertIntr(err.ToError(), kernel.ERESTARTSYS) + } + if !cms.Unix.Empty() { + mflags |= linux.MSG_CTRUNC + cms.Release() + } + + if int(msg.Flags) != mflags { + // Copy out the flags to the caller. + if _, err := t.CopyOut(msgPtr+flagsOffset, int32(mflags)); err != nil { + return 0, err + } + } + + return uintptr(n), nil + } + + if msg.ControlLen > maxControlLen { + return 0, syserror.ENOBUFS + } + n, mflags, sender, senderLen, cms, e := s.RecvMsg(t, dst, int(flags), haveDeadline, deadline, msg.NameLen != 0, msg.ControlLen) + if e != nil { + return 0, syserror.ConvertIntr(e.ToError(), kernel.ERESTARTSYS) + } + defer cms.Release() + + controlData := make([]byte, 0, msg.ControlLen) + controlData = control.PackControlMessages(t, cms, controlData) + + if cr, ok := s.(transport.Credentialer); ok && cr.Passcred() { + creds, _ := cms.Unix.Credentials.(control.SCMCredentials) + controlData, mflags = control.PackCredentials(t, creds, controlData, mflags) + } + + if cms.Unix.Rights != nil { + controlData, mflags = control.PackRights(t, cms.Unix.Rights.(control.SCMRights), flags&linux.MSG_CMSG_CLOEXEC != 0, controlData, mflags) + } + + // Copy the address to the caller. + if msg.NameLen != 0 { + if err := writeAddress(t, sender, senderLen, usermem.Addr(msg.Name), usermem.Addr(msgPtr+nameLenOffset)); err != nil { + return 0, err + } + } + + // Copy the control data to the caller. + if _, err := t.CopyOut(msgPtr+controlLenOffset, uint64(len(controlData))); err != nil { + return 0, err + } + if len(controlData) > 0 { + if _, err := t.CopyOut(usermem.Addr(msg.Control), controlData); err != nil { + return 0, err + } + } + + // Copy out the flags to the caller. + if _, err := t.CopyOut(msgPtr+flagsOffset, int32(mflags)); err != nil { + return 0, err + } + + return uintptr(n), nil +} + +// recvFrom is the implementation of the recvfrom syscall. It is called by +// recvfrom and recv syscall handlers. +func recvFrom(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flags int32, namePtr usermem.Addr, nameLenPtr usermem.Addr) (uintptr, error) { + if int(bufLen) < 0 { + return 0, syserror.EINVAL + } + + // Reject flags that we don't handle yet. + if flags & ^(baseRecvFlags|linux.MSG_PEEK|linux.MSG_CONFIRM) != 0 { + return 0, syserror.EINVAL + } + + // Get socket from the file descriptor. + file := t.GetFileVFS2(fd) + if file == nil { + return 0, syserror.EBADF + } + defer file.DecRef() + + // Extract the socket. + s, ok := file.Impl().(socket.SocketVFS2) + if !ok { + return 0, syserror.ENOTSOCK + } + + if (file.StatusFlags() & linux.SOCK_NONBLOCK) != 0 { + flags |= linux.MSG_DONTWAIT + } + + dst, err := t.SingleIOSequence(bufPtr, int(bufLen), usermem.IOOpts{ + AddressSpaceActive: true, + }) + if err != nil { + return 0, err + } + + var haveDeadline bool + var deadline ktime.Time + if dl := s.RecvTimeout(); dl > 0 { + deadline = t.Kernel().MonotonicClock().Now().Add(time.Duration(dl) * time.Nanosecond) + haveDeadline = true + } else if dl < 0 { + flags |= linux.MSG_DONTWAIT + } + + n, _, sender, senderLen, cm, e := s.RecvMsg(t, dst, int(flags), haveDeadline, deadline, nameLenPtr != 0, 0) + cm.Release() + if e != nil { + return 0, syserror.ConvertIntr(e.ToError(), kernel.ERESTARTSYS) + } + + // Copy the address to the caller. + if nameLenPtr != 0 { + if err := writeAddress(t, sender, senderLen, namePtr, nameLenPtr); err != nil { + return 0, err + } + } + + return uintptr(n), nil +} + +// RecvFrom implements the linux syscall recvfrom(2). +func RecvFrom(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + bufPtr := args[1].Pointer() + bufLen := args[2].Uint64() + flags := args[3].Int() + namePtr := args[4].Pointer() + nameLenPtr := args[5].Pointer() + + n, err := recvFrom(t, fd, bufPtr, bufLen, flags, namePtr, nameLenPtr) + return n, nil, err +} + +// SendMsg implements the linux syscall sendmsg(2). +func SendMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + msgPtr := args[1].Pointer() + flags := args[2].Int() + + if t.Arch().Width() != 8 { + // We only handle 64-bit for now. + return 0, nil, syserror.EINVAL + } + + // Get socket from the file descriptor. + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Extract the socket. + s, ok := file.Impl().(socket.SocketVFS2) + if !ok { + return 0, nil, syserror.ENOTSOCK + } + + // Reject flags that we don't handle yet. + if flags & ^(linux.MSG_DONTWAIT|linux.MSG_EOR|linux.MSG_MORE|linux.MSG_NOSIGNAL) != 0 { + return 0, nil, syserror.EINVAL + } + + if (file.StatusFlags() & linux.SOCK_NONBLOCK) != 0 { + flags |= linux.MSG_DONTWAIT + } + + n, err := sendSingleMsg(t, s, file, msgPtr, flags) + return n, nil, err +} + +// SendMMsg implements the linux syscall sendmmsg(2). +func SendMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + msgPtr := args[1].Pointer() + vlen := args[2].Uint() + flags := args[3].Int() + + if t.Arch().Width() != 8 { + // We only handle 64-bit for now. + return 0, nil, syserror.EINVAL + } + + // Get socket from the file descriptor. + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Extract the socket. + s, ok := file.Impl().(socket.SocketVFS2) + if !ok { + return 0, nil, syserror.ENOTSOCK + } + + // Reject flags that we don't handle yet. + if flags & ^(linux.MSG_DONTWAIT|linux.MSG_EOR|linux.MSG_MORE|linux.MSG_NOSIGNAL) != 0 { + return 0, nil, syserror.EINVAL + } + + if (file.StatusFlags() & linux.SOCK_NONBLOCK) != 0 { + flags |= linux.MSG_DONTWAIT + } + + var count uint32 + var err error + for i := uint64(0); i < uint64(vlen); i++ { + mp, ok := msgPtr.AddLength(i * multipleMessageHeader64Len) + if !ok { + return 0, nil, syserror.EFAULT + } + var n uintptr + if n, err = sendSingleMsg(t, s, file, mp, flags); err != nil { + break + } + + // Copy the received length to the caller. + lp, ok := mp.AddLength(messageHeader64Len) + if !ok { + return 0, nil, syserror.EFAULT + } + if _, err = t.CopyOut(lp, uint32(n)); err != nil { + break + } + count++ + } + + if count == 0 { + return 0, nil, err + } + return uintptr(count), nil, nil +} + +func sendSingleMsg(t *kernel.Task, s socket.SocketVFS2, file *vfs.FileDescription, msgPtr usermem.Addr, flags int32) (uintptr, error) { + // Capture the message header. + var msg MessageHeader64 + if err := CopyInMessageHeader64(t, msgPtr, &msg); err != nil { + return 0, err + } + + var controlData []byte + if msg.ControlLen > 0 { + // Put an upper bound to prevent large allocations. + if msg.ControlLen > maxControlLen { + return 0, syserror.ENOBUFS + } + controlData = make([]byte, msg.ControlLen) + if _, err := t.CopyIn(usermem.Addr(msg.Control), &controlData); err != nil { + return 0, err + } + } + + // Read the destination address if one is specified. + var to []byte + if msg.NameLen != 0 { + var err error + to, err = CaptureAddress(t, usermem.Addr(msg.Name), msg.NameLen) + if err != nil { + return 0, err + } + } + + // Read data then call the sendmsg implementation. + if msg.IovLen > linux.UIO_MAXIOV { + return 0, syserror.EMSGSIZE + } + src, err := t.IovecsIOSequence(usermem.Addr(msg.Iov), int(msg.IovLen), usermem.IOOpts{ + AddressSpaceActive: true, + }) + if err != nil { + return 0, err + } + + controlMessages, err := control.Parse(t, s, controlData) + if err != nil { + return 0, err + } + + var haveDeadline bool + var deadline ktime.Time + if dl := s.SendTimeout(); dl > 0 { + deadline = t.Kernel().MonotonicClock().Now().Add(time.Duration(dl) * time.Nanosecond) + haveDeadline = true + } else if dl < 0 { + flags |= linux.MSG_DONTWAIT + } + + // Call the syscall implementation. + n, e := s.SendMsg(t, src, to, int(flags), haveDeadline, deadline, controlMessages) + err = slinux.HandleIOErrorVFS2(t, n != 0, e.ToError(), kernel.ERESTARTSYS, "sendmsg", file) + if err != nil { + controlMessages.Release() + } + return uintptr(n), err +} + +// sendTo is the implementation of the sendto syscall. It is called by sendto +// and send syscall handlers. +func sendTo(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flags int32, namePtr usermem.Addr, nameLen uint32) (uintptr, error) { + bl := int(bufLen) + if bl < 0 { + return 0, syserror.EINVAL + } + + // Get socket from the file descriptor. + file := t.GetFileVFS2(fd) + if file == nil { + return 0, syserror.EBADF + } + defer file.DecRef() + + // Extract the socket. + s, ok := file.Impl().(socket.SocketVFS2) + if !ok { + return 0, syserror.ENOTSOCK + } + + if (file.StatusFlags() & linux.SOCK_NONBLOCK) != 0 { + flags |= linux.MSG_DONTWAIT + } + + // Read the destination address if one is specified. + var to []byte + var err error + if namePtr != 0 { + to, err = CaptureAddress(t, namePtr, nameLen) + if err != nil { + return 0, err + } + } + + src, err := t.SingleIOSequence(bufPtr, bl, usermem.IOOpts{ + AddressSpaceActive: true, + }) + if err != nil { + return 0, err + } + + var haveDeadline bool + var deadline ktime.Time + if dl := s.SendTimeout(); dl > 0 { + deadline = t.Kernel().MonotonicClock().Now().Add(time.Duration(dl) * time.Nanosecond) + haveDeadline = true + } else if dl < 0 { + flags |= linux.MSG_DONTWAIT + } + + // Call the syscall implementation. + n, e := s.SendMsg(t, src, to, int(flags), haveDeadline, deadline, socket.ControlMessages{Unix: control.New(t, s, nil)}) + return uintptr(n), slinux.HandleIOErrorVFS2(t, n != 0, e.ToError(), kernel.ERESTARTSYS, "sendto", file) +} + +// SendTo implements the linux syscall sendto(2). +func SendTo(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + bufPtr := args[1].Pointer() + bufLen := args[2].Uint64() + flags := args[3].Int() + namePtr := args[4].Pointer() + nameLen := args[5].Uint() + + n, err := sendTo(t, fd, bufPtr, bufLen, flags, namePtr, nameLen) + return n, nil, err +} -- cgit v1.2.3 From 71e6ac3e1f551cf52166bf501de114f06502b994 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Mon, 13 Apr 2020 17:58:52 -0700 Subject: Don't allow read/write when offset+size overflows. PiperOrigin-RevId: 306348346 --- pkg/sentry/syscalls/linux/sys_read.go | 8 ++++---- pkg/sentry/syscalls/linux/sys_splice.go | 4 +++- pkg/sentry/syscalls/linux/sys_write.go | 4 ++-- pkg/sentry/syscalls/linux/vfs2/read_write.go | 8 ++++---- test/syscalls/linux/memfd.cc | 1 + test/syscalls/linux/pread64.cc | 16 ++++++++++++++++ test/syscalls/linux/pwrite64.cc | 12 ++++++++++++ test/syscalls/linux/sendfile.cc | 23 +++++++++++++++++++++++ test/syscalls/linux/splice.cc | 1 + 9 files changed, 66 insertions(+), 11 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/sys_read.go b/pkg/sentry/syscalls/linux/sys_read.go index 78a2cb750..071b4bacc 100644 --- a/pkg/sentry/syscalls/linux/sys_read.go +++ b/pkg/sentry/syscalls/linux/sys_read.go @@ -96,8 +96,8 @@ func Readahead(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys return 0, nil, syserror.EINVAL } - // Check that the offset is legitimate. - if offset < 0 { + // Check that the offset is legitimate and does not overflow. + if offset < 0 || offset+int64(size) < 0 { return 0, nil, syserror.EINVAL } @@ -120,8 +120,8 @@ func Pread64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca } defer file.DecRef() - // Check that the offset is legitimate. - if offset < 0 { + // Check that the offset is legitimate and does not overflow. + if offset < 0 || offset+int64(size) < 0 { return 0, nil, syserror.EINVAL } diff --git a/pkg/sentry/syscalls/linux/sys_splice.go b/pkg/sentry/syscalls/linux/sys_splice.go index fbc6cf15f..df0d0f461 100644 --- a/pkg/sentry/syscalls/linux/sys_splice.go +++ b/pkg/sentry/syscalls/linux/sys_splice.go @@ -16,6 +16,7 @@ package linux import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -25,7 +26,8 @@ import ( // doSplice implements a blocking splice operation. func doSplice(t *kernel.Task, outFile, inFile *fs.File, opts fs.SpliceOpts, nonBlocking bool) (int64, error) { - if opts.Length < 0 || opts.SrcStart < 0 || opts.DstStart < 0 { + log.Infof("NLAC: doSplice opts: %+v", opts) + if opts.Length < 0 || opts.SrcStart < 0 || opts.DstStart < 0 || (opts.SrcStart+opts.Length < 0) { return 0, syserror.EINVAL } diff --git a/pkg/sentry/syscalls/linux/sys_write.go b/pkg/sentry/syscalls/linux/sys_write.go index 506ee54ce..6ec0de96e 100644 --- a/pkg/sentry/syscalls/linux/sys_write.go +++ b/pkg/sentry/syscalls/linux/sys_write.go @@ -87,8 +87,8 @@ func Pwrite64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc } defer file.DecRef() - // Check that the offset is legitimate. - if offset < 0 { + // Check that the offset is legitimate and does not overflow. + if offset < 0 || offset+int64(size) < 0 { return 0, nil, syserror.EINVAL } diff --git a/pkg/sentry/syscalls/linux/vfs2/read_write.go b/pkg/sentry/syscalls/linux/vfs2/read_write.go index 35f6308d6..898b190fd 100644 --- a/pkg/sentry/syscalls/linux/vfs2/read_write.go +++ b/pkg/sentry/syscalls/linux/vfs2/read_write.go @@ -130,8 +130,8 @@ func Pread64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca } defer file.DecRef() - // Check that the offset is legitimate. - if offset < 0 { + // Check that the offset is legitimate and does not overflow. + if offset < 0 || offset+int64(size) < 0 { return 0, nil, syserror.EINVAL } @@ -362,8 +362,8 @@ func Pwrite64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc } defer file.DecRef() - // Check that the offset is legitimate. - if offset < 0 { + // Check that the offset is legitimate and does not overflow. + if offset < 0 || offset+int64(size) < 0 { return 0, nil, syserror.EINVAL } diff --git a/test/syscalls/linux/memfd.cc b/test/syscalls/linux/memfd.cc index e57b49a4a..f8b7f7938 100644 --- a/test/syscalls/linux/memfd.cc +++ b/test/syscalls/linux/memfd.cc @@ -16,6 +16,7 @@ #include #include #include +#include #include #include #include diff --git a/test/syscalls/linux/pread64.cc b/test/syscalls/linux/pread64.cc index 2cecf2e5f..bcdbbb044 100644 --- a/test/syscalls/linux/pread64.cc +++ b/test/syscalls/linux/pread64.cc @@ -14,6 +14,7 @@ #include #include +#include #include #include #include @@ -118,6 +119,21 @@ TEST_F(Pread64Test, EndOfFile) { EXPECT_THAT(pread64(fd.get(), buf, 1024, 0), SyscallSucceedsWithValue(0)); } +int memfd_create(const std::string& name, unsigned int flags) { + return syscall(__NR_memfd_create, name.c_str(), flags); +} + +TEST_F(Pread64Test, Overflow) { + int f = memfd_create("negative", 0); + const FileDescriptor fd(f); + + EXPECT_THAT(ftruncate(fd.get(), 0x7fffffffffffffffull), SyscallSucceeds()); + + char buf[10]; + EXPECT_THAT(pread64(fd.get(), buf, sizeof(buf), 0x7fffffffffffffffull), + SyscallFailsWithErrno(EINVAL)); +} + TEST(Pread64TestNoTempFile, CantReadSocketPair_NoRandomSave) { int sock_fds[2]; EXPECT_THAT(socketpair(AF_UNIX, SOCK_STREAM, 0, sock_fds), SyscallSucceeds()); diff --git a/test/syscalls/linux/pwrite64.cc b/test/syscalls/linux/pwrite64.cc index c2f72e010..e69794910 100644 --- a/test/syscalls/linux/pwrite64.cc +++ b/test/syscalls/linux/pwrite64.cc @@ -14,6 +14,7 @@ #include #include +#include #include #include #include @@ -65,6 +66,17 @@ TEST_F(Pwrite64, InvalidArgs) { EXPECT_THAT(close(fd), SyscallSucceeds()); } +TEST_F(Pwrite64, Overflow) { + int fd; + ASSERT_THAT(fd = open(name_.c_str(), O_APPEND | O_RDWR), SyscallSucceeds()); + constexpr int64_t kBufSize = 1024; + std::vector buf(kBufSize); + std::fill(buf.begin(), buf.end(), 'a'); + EXPECT_THAT(PwriteFd(fd, buf.data(), buf.size(), 0x7fffffffffffffffull), + SyscallFailsWithErrno(EINVAL)); + EXPECT_THAT(close(fd), SyscallSucceeds()); +} + } // namespace } // namespace testing diff --git a/test/syscalls/linux/sendfile.cc b/test/syscalls/linux/sendfile.cc index ebaafe47e..64123e904 100644 --- a/test/syscalls/linux/sendfile.cc +++ b/test/syscalls/linux/sendfile.cc @@ -13,6 +13,7 @@ // limitations under the License. #include +#include #include #include #include @@ -70,6 +71,28 @@ TEST(SendFileTest, InvalidOffset) { SyscallFailsWithErrno(EINVAL)); } +int memfd_create(const std::string& name, unsigned int flags) { + return syscall(__NR_memfd_create, name.c_str(), flags); +} + +TEST(SendFileTest, Overflow) { + // Create input file. + const TempPath in_file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); + const FileDescriptor inf = + ASSERT_NO_ERRNO_AND_VALUE(Open(in_file.path(), O_RDONLY)); + + // Open the output file. + int fd; + EXPECT_THAT(fd = memfd_create("overflow", 0), SyscallSucceeds()); + const FileDescriptor outf(fd); + + // out_offset + kSize overflows INT64_MAX. + loff_t out_offset = 0x7ffffffffffffffeull; + constexpr int kSize = 3; + EXPECT_THAT(sendfile(outf.get(), inf.get(), &out_offset, kSize), + SyscallFailsWithErrno(EINVAL)); +} + TEST(SendFileTest, SendTrivially) { // Create temp files. constexpr char kData[] = "To be, or not to be, that is the question:"; diff --git a/test/syscalls/linux/splice.cc b/test/syscalls/linux/splice.cc index faa1247f6..f103e2e56 100644 --- a/test/syscalls/linux/splice.cc +++ b/test/syscalls/linux/splice.cc @@ -13,6 +13,7 @@ // limitations under the License. #include +#include #include #include #include -- cgit v1.2.3 From 52b4b19249adfeba65fe6f0ef27111f2ed887888 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Tue, 14 Apr 2020 13:36:36 -0700 Subject: Pass O_LARGEFILE in syscalls/linux/vfs2.openat. Needed for PipeTest_Flags: files opened by open() and openat() get O_LARGEFILE (on architectures with 64-bit off_t), but not FDs created by other syscalls such as pipe(). Updates #1035 PiperOrigin-RevId: 306504788 --- pkg/sentry/syscalls/linux/vfs2/filesystem.go | 2 +- pkg/sentry/vfs/file_description.go | 2 +- pkg/sentry/vfs/vfs.go | 5 +---- 3 files changed, 3 insertions(+), 6 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/vfs2/filesystem.go b/pkg/sentry/syscalls/linux/vfs2/filesystem.go index a859095e2..46d3e189c 100644 --- a/pkg/sentry/syscalls/linux/vfs2/filesystem.go +++ b/pkg/sentry/syscalls/linux/vfs2/filesystem.go @@ -172,7 +172,7 @@ func openat(t *kernel.Task, dirfd int32, pathAddr usermem.Addr, flags uint32, mo defer tpop.Release() file, err := t.Kernel().VFS().OpenAt(t, t.Credentials(), &tpop.pop, &vfs.OpenOptions{ - Flags: flags, + Flags: flags | linux.O_LARGEFILE, Mode: linux.FileMode(mode & (0777 | linux.S_ISUID | linux.S_ISGID | linux.S_ISVTX) &^ t.FSContext().Umask()), }) if err != nil { diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 4fb9aea87..5976b5ccd 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -122,7 +122,7 @@ func (fd *FileDescription) Init(impl FileDescriptionImpl, statusFlags uint32, mn } fd.refs = 1 - fd.statusFlags = statusFlags | linux.O_LARGEFILE + fd.statusFlags = statusFlags fd.vd = VirtualDentry{ mount: mnt, dentry: d, diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index f592913d5..053c6e1d1 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -383,14 +383,11 @@ func (vfs *VirtualFilesystem) BoundEndpointAt(ctx context.Context, creds *auth.C func (vfs *VirtualFilesystem) OpenAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *OpenOptions) (*FileDescription, error) { // Remove: // - // - O_LARGEFILE, which we always report in FileDescription status flags - // since only 64-bit architectures are supported at this time. - // // - O_CLOEXEC, which affects file descriptors and therefore must be // handled outside of VFS. // // - Unknown flags. - opts.Flags &= linux.O_ACCMODE | linux.O_CREAT | linux.O_EXCL | linux.O_NOCTTY | linux.O_TRUNC | linux.O_APPEND | linux.O_NONBLOCK | linux.O_DSYNC | linux.O_ASYNC | linux.O_DIRECT | linux.O_DIRECTORY | linux.O_NOFOLLOW | linux.O_NOATIME | linux.O_SYNC | linux.O_PATH | linux.O_TMPFILE + opts.Flags &= linux.O_ACCMODE | linux.O_CREAT | linux.O_EXCL | linux.O_NOCTTY | linux.O_TRUNC | linux.O_APPEND | linux.O_NONBLOCK | linux.O_DSYNC | linux.O_ASYNC | linux.O_DIRECT | linux.O_LARGEFILE | linux.O_DIRECTORY | linux.O_NOFOLLOW | linux.O_NOATIME | linux.O_SYNC | linux.O_PATH | linux.O_TMPFILE // Linux's __O_SYNC (which we call linux.O_SYNC) implies O_DSYNC. if opts.Flags&linux.O_SYNC != 0 { opts.Flags |= linux.O_DSYNC -- cgit v1.2.3 From 2dd6384de89a866bddb9184b8d7ab85b5b8f7100 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Tue, 14 Apr 2020 14:40:08 -0700 Subject: Fix cleanup around socketpair() failure to copy out FDs. - Use the fs.File, rather than the vfs.FileDescription, in the VFS1 version. - Check for a nil fs.File/vfs.FileDescription before calling DecRef, which is possible if a racing dup2() or dup3() replaces the file descriptor between when it is installed and when it is returned. (This is not possible in Linux because Linux separates allocation of a file descriptor from binding an allocated file descriptor to a struct file, and dup2/dup3 return EBUSY if asked to replace an allocated but unbound file descriptor.) PiperOrigin-RevId: 306517101 --- pkg/sentry/syscalls/linux/sys_socket.go | 5 +++-- pkg/sentry/syscalls/linux/vfs2/socket.go | 5 +++-- 2 files changed, 6 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/sys_socket.go b/pkg/sentry/syscalls/linux/sys_socket.go index 61b2576ac..0760af77b 100644 --- a/pkg/sentry/syscalls/linux/sys_socket.go +++ b/pkg/sentry/syscalls/linux/sys_socket.go @@ -247,8 +247,9 @@ func SocketPair(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy // Copy the file descriptors out. if _, err := t.CopyOut(socks, fds); err != nil { for _, fd := range fds { - _, file := t.FDTable().Remove(fd) - file.DecRef() + if file, _ := t.FDTable().Remove(fd); file != nil { + file.DecRef() + } } return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/vfs2/socket.go b/pkg/sentry/syscalls/linux/vfs2/socket.go index 79a4a7ada..b1ede32f0 100644 --- a/pkg/sentry/syscalls/linux/vfs2/socket.go +++ b/pkg/sentry/syscalls/linux/vfs2/socket.go @@ -250,8 +250,9 @@ func SocketPair(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy if _, err := t.CopyOut(addr, fds); err != nil { for _, fd := range fds { - _, file := t.FDTable().Remove(fd) - file.DecRef() + if _, file := t.FDTable().Remove(fd); file != nil { + file.DecRef() + } } return 0, nil, err } -- cgit v1.2.3 From 28399818fc1e5d294cc93ddd4a1ac7e31c375fbf Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Thu, 16 Apr 2020 11:48:14 -0700 Subject: Make ExtractErrno a function PiperOrigin-RevId: 306891171 --- pkg/sentry/kernel/task_run.go | 2 +- pkg/sentry/kernel/task_signals.go | 2 +- pkg/sentry/kernel/task_syscall.go | 12 ++++++------ pkg/sentry/strace/strace.go | 2 +- pkg/sentry/syscalls/linux/sys_aio.go | 2 +- 5 files changed, 10 insertions(+), 10 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/task_run.go b/pkg/sentry/kernel/task_run.go index 799cbcd93..2ba8d7e63 100644 --- a/pkg/sentry/kernel/task_run.go +++ b/pkg/sentry/kernel/task_run.go @@ -353,7 +353,7 @@ func (app *runApp) execute(t *Task) taskRunState { default: // What happened? Can't continue. t.Warningf("Unexpected SwitchToApp error: %v", err) - t.PrepareExit(ExitStatus{Code: t.ExtractErrno(err, -1)}) + t.PrepareExit(ExitStatus{Code: ExtractErrno(err, -1)}) return (*runExit)(nil) } } diff --git a/pkg/sentry/kernel/task_signals.go b/pkg/sentry/kernel/task_signals.go index 6aa798346..f07de2089 100644 --- a/pkg/sentry/kernel/task_signals.go +++ b/pkg/sentry/kernel/task_signals.go @@ -174,7 +174,7 @@ func (t *Task) deliverSignal(info *arch.SignalInfo, act arch.SignalAct) taskRunS fallthrough case (sre == ERESTARTSYS && !act.IsRestart()): t.Debugf("Not restarting syscall %d after errno %d: interrupted by signal %d", t.Arch().SyscallNo(), sre, info.Signo) - t.Arch().SetReturn(uintptr(-t.ExtractErrno(syserror.EINTR, -1))) + t.Arch().SetReturn(uintptr(-ExtractErrno(syserror.EINTR, -1))) default: t.Debugf("Restarting syscall %d after errno %d: interrupted by signal %d", t.Arch().SyscallNo(), sre, info.Signo) t.Arch().RestartSyscall() diff --git a/pkg/sentry/kernel/task_syscall.go b/pkg/sentry/kernel/task_syscall.go index 3d7a734ef..c9db78e06 100644 --- a/pkg/sentry/kernel/task_syscall.go +++ b/pkg/sentry/kernel/task_syscall.go @@ -312,7 +312,7 @@ func (t *Task) doSyscallInvoke(sysno uintptr, args arch.SyscallArguments) taskRu return ctrl.next } } else if err != nil { - t.Arch().SetReturn(uintptr(-t.ExtractErrno(err, int(sysno)))) + t.Arch().SetReturn(uintptr(-ExtractErrno(err, int(sysno)))) t.haveSyscallReturn = true } else { t.Arch().SetReturn(rval) @@ -431,7 +431,7 @@ func (t *Task) doVsyscallInvoke(sysno uintptr, args arch.SyscallArguments, calle // A return is not emulated in this case. return (*runApp)(nil) } - t.Arch().SetReturn(uintptr(-t.ExtractErrno(err, int(sysno)))) + t.Arch().SetReturn(uintptr(-ExtractErrno(err, int(sysno)))) } t.Arch().SetIP(t.Arch().Value(caller)) t.Arch().SetStack(t.Arch().Stack() + uintptr(t.Arch().Width())) @@ -441,7 +441,7 @@ func (t *Task) doVsyscallInvoke(sysno uintptr, args arch.SyscallArguments, calle // ExtractErrno extracts an integer error number from the error. // The syscall number is purely for context in the error case. Use -1 if // syscall number is unknown. -func (t *Task) ExtractErrno(err error, sysno int) int { +func ExtractErrno(err error, sysno int) int { switch err := err.(type) { case nil: return 0 @@ -455,11 +455,11 @@ func (t *Task) ExtractErrno(err error, sysno int) int { // handled (and the SIGBUS is delivered). return int(syscall.EFAULT) case *os.PathError: - return t.ExtractErrno(err.Err, sysno) + return ExtractErrno(err.Err, sysno) case *os.LinkError: - return t.ExtractErrno(err.Err, sysno) + return ExtractErrno(err.Err, sysno) case *os.SyscallError: - return t.ExtractErrno(err.Err, sysno) + return ExtractErrno(err.Err, sysno) default: if errno, ok := syserror.TranslateError(err); ok { return int(errno) diff --git a/pkg/sentry/strace/strace.go b/pkg/sentry/strace/strace.go index b94c4fbf5..68ca537c8 100644 --- a/pkg/sentry/strace/strace.go +++ b/pkg/sentry/strace/strace.go @@ -719,7 +719,7 @@ func (s SyscallMap) SyscallEnter(t *kernel.Task, sysno uintptr, args arch.Syscal // SyscallExit implements kernel.Stracer.SyscallExit. It logs the syscall // exit trace. func (s SyscallMap) SyscallExit(context interface{}, t *kernel.Task, sysno, rval uintptr, err error) { - errno := t.ExtractErrno(err, int(sysno)) + errno := kernel.ExtractErrno(err, int(sysno)) c := context.(*syscallContext) elapsed := time.Since(c.start) diff --git a/pkg/sentry/syscalls/linux/sys_aio.go b/pkg/sentry/syscalls/linux/sys_aio.go index 38cbeba5a..d781d6a04 100644 --- a/pkg/sentry/syscalls/linux/sys_aio.go +++ b/pkg/sentry/syscalls/linux/sys_aio.go @@ -290,7 +290,7 @@ func performCallback(t *kernel.Task, file *fs.File, cbAddr usermem.Addr, cb *ioC // Update the result. if err != nil { err = handleIOError(t, ev.Result != 0 /* partial */, err, nil /* never interrupted */, "aio", file) - ev.Result = -int64(t.ExtractErrno(err, 0)) + ev.Result = -int64(kernel.ExtractErrno(err, 0)) } file.DecRef() -- cgit v1.2.3 From f03996c5e9803934226e4b3a10827501cb936ab9 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Thu, 16 Apr 2020 19:26:02 -0700 Subject: Implement pipe(2) and pipe2(2) for VFS2. Updates #1035 PiperOrigin-RevId: 306968644 --- pkg/sentry/fsimpl/pipefs/BUILD | 20 +++ pkg/sentry/fsimpl/pipefs/pipefs.go | 148 +++++++++++++++++++ pkg/sentry/fsimpl/tmpfs/filesystem.go | 2 +- pkg/sentry/fsimpl/tmpfs/named_pipe.go | 23 +-- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 2 +- pkg/sentry/kernel/BUILD | 1 + pkg/sentry/kernel/kernel.go | 30 +++- pkg/sentry/kernel/pipe/vfs.go | 162 ++++++++++++--------- pkg/sentry/syscalls/linux/sys_pipe.go | 14 +- pkg/sentry/syscalls/linux/vfs2/BUILD | 3 + pkg/sentry/syscalls/linux/vfs2/fd.go | 17 +++ .../syscalls/linux/vfs2/linux64_override_amd64.go | 4 +- pkg/sentry/syscalls/linux/vfs2/pipe.go | 63 ++++++++ pkg/sentry/syscalls/linux/vfs2/read_write.go | 8 +- pkg/sentry/vfs/vfs.go | 2 +- test/syscalls/linux/pipe.cc | 2 + 16 files changed, 389 insertions(+), 112 deletions(-) create mode 100644 pkg/sentry/fsimpl/pipefs/BUILD create mode 100644 pkg/sentry/fsimpl/pipefs/pipefs.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/pipe.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/pipefs/BUILD b/pkg/sentry/fsimpl/pipefs/BUILD new file mode 100644 index 000000000..0d411606f --- /dev/null +++ b/pkg/sentry/fsimpl/pipefs/BUILD @@ -0,0 +1,20 @@ +load("//tools:defs.bzl", "go_library") + +licenses(["notice"]) + +go_library( + name = "pipefs", + srcs = ["pipefs.go"], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/abi/linux", + "//pkg/context", + "//pkg/sentry/fsimpl/kernfs", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/kernel/pipe", + "//pkg/sentry/kernel/time", + "//pkg/sentry/vfs", + "//pkg/syserror", + "//pkg/usermem", + ], +) diff --git a/pkg/sentry/fsimpl/pipefs/pipefs.go b/pkg/sentry/fsimpl/pipefs/pipefs.go new file mode 100644 index 000000000..faf3179bc --- /dev/null +++ b/pkg/sentry/fsimpl/pipefs/pipefs.go @@ -0,0 +1,148 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package pipefs provides the filesystem implementation backing +// Kernel.PipeMount. +package pipefs + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" + ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +type filesystemType struct{} + +// Name implements vfs.FilesystemType.Name. +func (filesystemType) Name() string { + return "pipefs" +} + +// GetFilesystem implements vfs.FilesystemType.GetFilesystem. +func (filesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { + panic("pipefs.filesystemType.GetFilesystem should never be called") +} + +// filesystem implements vfs.FilesystemImpl. +type filesystem struct { + kernfs.Filesystem + + // TODO(gvisor.dev/issue/1193): + // + // - kernfs does not provide a way to implement statfs, from which we + // should indicate PIPEFS_MAGIC. + // + // - kernfs does not provide a way to override names for + // vfs.FilesystemImpl.PrependPath(); pipefs inodes should use synthetic + // name fmt.Sprintf("pipe:[%d]", inode.ino). +} + +// NewFilesystem sets up and returns a new vfs.Filesystem implemented by +// pipefs. +func NewFilesystem(vfsObj *vfs.VirtualFilesystem) *vfs.Filesystem { + fs := &filesystem{} + fs.Init(vfsObj, filesystemType{}) + return fs.VFSFilesystem() +} + +// inode implements kernfs.Inode. +type inode struct { + kernfs.InodeNotDirectory + kernfs.InodeNotSymlink + kernfs.InodeNoopRefCount + + pipe *pipe.VFSPipe + + ino uint64 + uid auth.KUID + gid auth.KGID + // We use the creation timestamp for all of atime, mtime, and ctime. + ctime ktime.Time +} + +func newInode(ctx context.Context, fs *kernfs.Filesystem) *inode { + creds := auth.CredentialsFromContext(ctx) + return &inode{ + pipe: pipe.NewVFSPipe(false /* isNamed */, pipe.DefaultPipeSize, usermem.PageSize), + ino: fs.NextIno(), + uid: creds.EffectiveKUID, + gid: creds.EffectiveKGID, + ctime: ktime.NowFromContext(ctx), + } +} + +const pipeMode = 0600 | linux.S_IFIFO + +// CheckPermissions implements kernfs.Inode.CheckPermissions. +func (i *inode) CheckPermissions(ctx context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error { + return vfs.GenericCheckPermissions(creds, ats, pipeMode, i.uid, i.gid) +} + +// Mode implements kernfs.Inode.Mode. +func (i *inode) Mode() linux.FileMode { + return pipeMode +} + +// Stat implements kernfs.Inode.Stat. +func (i *inode) Stat(vfsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { + ts := linux.NsecToStatxTimestamp(i.ctime.Nanoseconds()) + return linux.Statx{ + Mask: linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_NLINK | linux.STATX_UID | linux.STATX_GID | linux.STATX_ATIME | linux.STATX_MTIME | linux.STATX_CTIME | linux.STATX_INO | linux.STATX_SIZE | linux.STATX_BLOCKS, + Blksize: usermem.PageSize, + Nlink: 1, + UID: uint32(i.uid), + GID: uint32(i.gid), + Mode: pipeMode, + Ino: i.ino, + Size: 0, + Blocks: 0, + Atime: ts, + Ctime: ts, + Mtime: ts, + // TODO(gvisor.dev/issue/1197): Device number. + }, nil +} + +// SetStat implements kernfs.Inode.SetStat. +func (i *inode) SetStat(ctx context.Context, vfsfs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { + if opts.Stat.Mask == 0 { + return nil + } + return syserror.EPERM +} + +// Open implements kernfs.Inode.Open. +func (i *inode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + // FIXME(b/38173783): kernfs does not plumb Context here. + return i.pipe.Open(context.Background(), rp.Mount(), vfsd, opts.Flags) +} + +// NewConnectedPipeFDs returns a pair of FileDescriptions representing the read +// and write ends of a newly-created pipe, as for pipe(2) and pipe2(2). +// +// Preconditions: mnt.Filesystem() must have been returned by NewFilesystem(). +func NewConnectedPipeFDs(ctx context.Context, mnt *vfs.Mount, flags uint32) (*vfs.FileDescription, *vfs.FileDescription) { + fs := mnt.Filesystem().Impl().(*kernfs.Filesystem) + inode := newInode(ctx, fs) + var d kernfs.Dentry + d.Init(inode) + defer d.DecRef() + return inode.pipe.ReaderWriterPair(mnt, d.VFSDentry(), flags) +} diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index f4d50d64f..660f5a29b 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -392,7 +392,7 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.Open // Can't open symlinks without O_PATH (which is unimplemented). return nil, syserror.ELOOP case *namedPipe: - return newNamedPipeFD(ctx, impl, rp, &d.vfsd, opts.Flags) + return impl.pipe.Open(ctx, rp.Mount(), &d.vfsd, opts.Flags) case *deviceFile: return rp.VirtualFilesystem().OpenDeviceSpecialFile(ctx, rp.Mount(), &d.vfsd, impl.kind, impl.major, impl.minor, opts) case *socketFile: diff --git a/pkg/sentry/fsimpl/tmpfs/named_pipe.go b/pkg/sentry/fsimpl/tmpfs/named_pipe.go index 2c5c739df..8d77b3fa8 100644 --- a/pkg/sentry/fsimpl/tmpfs/named_pipe.go +++ b/pkg/sentry/fsimpl/tmpfs/named_pipe.go @@ -16,10 +16,8 @@ package tmpfs import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" - "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/usermem" ) @@ -33,27 +31,8 @@ type namedPipe struct { // * fs.mu must be locked. // * rp.Mount().CheckBeginWrite() has been called successfully. func (fs *filesystem) newNamedPipe(creds *auth.Credentials, mode linux.FileMode) *inode { - file := &namedPipe{pipe: pipe.NewVFSPipe(pipe.DefaultPipeSize, usermem.PageSize)} + file := &namedPipe{pipe: pipe.NewVFSPipe(true /* isNamed */, pipe.DefaultPipeSize, usermem.PageSize)} file.inode.init(file, fs, creds, linux.S_IFIFO|mode) file.inode.nlink = 1 // Only the parent has a link. return &file.inode } - -// namedPipeFD implements vfs.FileDescriptionImpl. Methods are implemented -// entirely via struct embedding. -type namedPipeFD struct { - fileDescription - - *pipe.VFSPipeFD -} - -func newNamedPipeFD(ctx context.Context, np *namedPipe, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { - var err error - var fd namedPipeFD - fd.VFSPipeFD, err = np.pipe.NewVFSPipeFD(ctx, vfsd, &fd.vfsfd, flags) - if err != nil { - return nil, err - } - fd.vfsfd.Init(&fd, flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}) - return &fd.vfsfd, nil -} diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 9fa8637d5..a59b24d45 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -357,6 +357,7 @@ func (i *inode) setStat(ctx context.Context, creds *auth.Credentials, stat *linu return err } i.mu.Lock() + defer i.mu.Unlock() var ( needsMtimeBump bool needsCtimeBump bool @@ -427,7 +428,6 @@ func (i *inode) setStat(ctx context.Context, creds *auth.Credentials, stat *linu atomic.StoreInt64(&i.ctime, now) } - i.mu.Unlock() return nil } diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD index e0ff58d8c..e47af66d6 100644 --- a/pkg/sentry/kernel/BUILD +++ b/pkg/sentry/kernel/BUILD @@ -170,6 +170,7 @@ go_library( "//pkg/sentry/fs/timerfd", "//pkg/sentry/fsbridge", "//pkg/sentry/fsimpl/kernfs", + "//pkg/sentry/fsimpl/pipefs", "//pkg/sentry/fsimpl/sockfs", "//pkg/sentry/hostcpu", "//pkg/sentry/inet", diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index de8a95854..fef60e636 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -50,6 +50,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/timerfd" "gvisor.dev/gvisor/pkg/sentry/fsbridge" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/pipefs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" "gvisor.dev/gvisor/pkg/sentry/hostcpu" "gvisor.dev/gvisor/pkg/sentry/inet" @@ -254,6 +255,10 @@ type Kernel struct { // VFS keeps the filesystem state used across the kernel. vfs vfs.VirtualFilesystem + // pipeMount is the Mount used for pipes created by the pipe() and pipe2() + // syscalls (as opposed to named pipes created by mknod()). + pipeMount *vfs.Mount + // If set to true, report address space activation waits as if the task is in // external wait so that the watchdog doesn't report the task stuck. SleepForAddressSpaceActivation bool @@ -354,19 +359,29 @@ func (k *Kernel) Init(args InitKernelArgs) error { k.monotonicClock = &timekeeperClock{tk: args.Timekeeper, c: sentrytime.Monotonic} k.futexes = futex.NewManager() k.netlinkPorts = port.New() + if VFS2Enabled { if err := k.vfs.Init(); err != nil { return fmt.Errorf("failed to initialize VFS: %v", err) } - fs := sockfs.NewFilesystem(&k.vfs) - // NewDisconnectedMount will take an additional reference on fs. - defer fs.DecRef() - sm, err := k.vfs.NewDisconnectedMount(fs, nil, &vfs.MountOptions{}) + + pipeFilesystem := pipefs.NewFilesystem(&k.vfs) + defer pipeFilesystem.DecRef() + pipeMount, err := k.vfs.NewDisconnectedMount(pipeFilesystem, nil, &vfs.MountOptions{}) + if err != nil { + return fmt.Errorf("failed to create pipefs mount: %v", err) + } + k.pipeMount = pipeMount + + socketFilesystem := sockfs.NewFilesystem(&k.vfs) + defer socketFilesystem.DecRef() + socketMount, err := k.vfs.NewDisconnectedMount(socketFilesystem, nil, &vfs.MountOptions{}) if err != nil { return fmt.Errorf("failed to initialize socket mount: %v", err) } - k.socketMount = sm + k.socketMount = socketMount } + return nil } @@ -1613,3 +1628,8 @@ func (k *Kernel) EmitUnimplementedEvent(ctx context.Context) { func (k *Kernel) VFS() *vfs.VirtualFilesystem { return &k.vfs } + +// PipeMount returns the pipefs mount. +func (k *Kernel) PipeMount() *vfs.Mount { + return k.pipeMount +} diff --git a/pkg/sentry/kernel/pipe/vfs.go b/pkg/sentry/kernel/pipe/vfs.go index a5675bd70..b54f08a30 100644 --- a/pkg/sentry/kernel/pipe/vfs.go +++ b/pkg/sentry/kernel/pipe/vfs.go @@ -49,38 +49,42 @@ type VFSPipe struct { } // NewVFSPipe returns an initialized VFSPipe. -func NewVFSPipe(sizeBytes, atomicIOBytes int64) *VFSPipe { +func NewVFSPipe(isNamed bool, sizeBytes, atomicIOBytes int64) *VFSPipe { var vp VFSPipe - initPipe(&vp.pipe, true /* isNamed */, sizeBytes, atomicIOBytes) + initPipe(&vp.pipe, isNamed, sizeBytes, atomicIOBytes) return &vp } -// NewVFSPipeFD opens a named pipe. Named pipes have special blocking semantics -// during open: +// ReaderWriterPair returns read-only and write-only FDs for vp. // -// "Normally, opening the FIFO blocks until the other end is opened also. A -// process can open a FIFO in nonblocking mode. In this case, opening for -// read-only will succeed even if no-one has opened on the write side yet, -// opening for write-only will fail with ENXIO (no such device or address) -// unless the other end has already been opened. Under Linux, opening a FIFO -// for read and write will succeed both in blocking and nonblocking mode. POSIX -// leaves this behavior undefined. This can be used to open a FIFO for writing -// while there are no readers available." - fifo(7) -func (vp *VFSPipe) NewVFSPipeFD(ctx context.Context, vfsd *vfs.Dentry, vfsfd *vfs.FileDescription, flags uint32) (*VFSPipeFD, error) { +// Preconditions: statusFlags should not contain an open access mode. +func (vp *VFSPipe) ReaderWriterPair(mnt *vfs.Mount, vfsd *vfs.Dentry, statusFlags uint32) (*vfs.FileDescription, *vfs.FileDescription) { + return vp.newFD(mnt, vfsd, linux.O_RDONLY|statusFlags), vp.newFD(mnt, vfsd, linux.O_WRONLY|statusFlags) +} + +// Open opens the pipe represented by vp. +func (vp *VFSPipe) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, statusFlags uint32) (*vfs.FileDescription, error) { vp.mu.Lock() defer vp.mu.Unlock() - readable := vfs.MayReadFileWithOpenFlags(flags) - writable := vfs.MayWriteFileWithOpenFlags(flags) + readable := vfs.MayReadFileWithOpenFlags(statusFlags) + writable := vfs.MayWriteFileWithOpenFlags(statusFlags) if !readable && !writable { return nil, syserror.EINVAL } - vfd, err := vp.open(vfsd, vfsfd, flags) - if err != nil { - return nil, err - } + fd := vp.newFD(mnt, vfsd, statusFlags) + // Named pipes have special blocking semantics during open: + // + // "Normally, opening the FIFO blocks until the other end is opened also. A + // process can open a FIFO in nonblocking mode. In this case, opening for + // read-only will succeed even if no-one has opened on the write side yet, + // opening for write-only will fail with ENXIO (no such device or address) + // unless the other end has already been opened. Under Linux, opening a + // FIFO for read and write will succeed both in blocking and nonblocking + // mode. POSIX leaves this behavior undefined. This can be used to open a + // FIFO for writing while there are no readers available." - fifo(7) switch { case readable && writable: // Pipes opened for read-write always succeed without blocking. @@ -89,23 +93,26 @@ func (vp *VFSPipe) NewVFSPipeFD(ctx context.Context, vfsd *vfs.Dentry, vfsfd *vf case readable: newHandleLocked(&vp.rWakeup) - // If this pipe is being opened as nonblocking and there's no + // If this pipe is being opened as blocking and there's no // writer, we have to wait for a writer to open the other end. - if flags&linux.O_NONBLOCK == 0 && !vp.pipe.HasWriters() && !waitFor(&vp.mu, &vp.wWakeup, ctx) { + if vp.pipe.isNamed && statusFlags&linux.O_NONBLOCK == 0 && !vp.pipe.HasWriters() && !waitFor(&vp.mu, &vp.wWakeup, ctx) { + fd.DecRef() return nil, syserror.EINTR } case writable: newHandleLocked(&vp.wWakeup) - if !vp.pipe.HasReaders() { - // Nonblocking, write-only opens fail with ENXIO when - // the read side isn't open yet. - if flags&linux.O_NONBLOCK != 0 { + if vp.pipe.isNamed && !vp.pipe.HasReaders() { + // Non-blocking, write-only opens fail with ENXIO when the read + // side isn't open yet. + if statusFlags&linux.O_NONBLOCK != 0 { + fd.DecRef() return nil, syserror.ENXIO } // Wait for a reader to open the other end. if !waitFor(&vp.mu, &vp.rWakeup, ctx) { + fd.DecRef() return nil, syserror.EINTR } } @@ -114,96 +121,93 @@ func (vp *VFSPipe) NewVFSPipeFD(ctx context.Context, vfsd *vfs.Dentry, vfsfd *vf panic("invalid pipe flags: must be readable, writable, or both") } - return vfd, nil + return fd, nil } // Preconditions: vp.mu must be held. -func (vp *VFSPipe) open(vfsd *vfs.Dentry, vfsfd *vfs.FileDescription, flags uint32) (*VFSPipeFD, error) { - var fd VFSPipeFD - fd.flags = flags - fd.readable = vfs.MayReadFileWithOpenFlags(flags) - fd.writable = vfs.MayWriteFileWithOpenFlags(flags) - fd.vfsfd = vfsfd - fd.pipe = &vp.pipe +func (vp *VFSPipe) newFD(mnt *vfs.Mount, vfsd *vfs.Dentry, statusFlags uint32) *vfs.FileDescription { + fd := &VFSPipeFD{ + pipe: &vp.pipe, + } + fd.vfsfd.Init(fd, statusFlags, mnt, vfsd, &vfs.FileDescriptionOptions{ + DenyPRead: true, + DenyPWrite: true, + UseDentryMetadata: true, + }) switch { - case fd.readable && fd.writable: + case fd.vfsfd.IsReadable() && fd.vfsfd.IsWritable(): vp.pipe.rOpen() vp.pipe.wOpen() - case fd.readable: + case fd.vfsfd.IsReadable(): vp.pipe.rOpen() - case fd.writable: + case fd.vfsfd.IsWritable(): vp.pipe.wOpen() default: panic("invalid pipe flags: must be readable, writable, or both") } - return &fd, nil + return &fd.vfsfd } -// VFSPipeFD implements a subset of vfs.FileDescriptionImpl for pipes. It is -// expected that filesystesm will use this in a struct implementing -// vfs.FileDescriptionImpl. +// VFSPipeFD implements vfs.FileDescriptionImpl for pipes. type VFSPipeFD struct { - pipe *Pipe - flags uint32 - readable bool - writable bool - vfsfd *vfs.FileDescription + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl + vfs.DentryMetadataFileDescriptionImpl + + pipe *Pipe } // Release implements vfs.FileDescriptionImpl.Release. func (fd *VFSPipeFD) Release() { var event waiter.EventMask - if fd.readable { + if fd.vfsfd.IsReadable() { fd.pipe.rClose() - event |= waiter.EventIn + event |= waiter.EventOut } - if fd.writable { + if fd.vfsfd.IsWritable() { fd.pipe.wClose() - event |= waiter.EventOut + event |= waiter.EventIn | waiter.EventHUp } if event == 0 { panic("invalid pipe flags: must be readable, writable, or both") } - if fd.writable { - fd.vfsfd.VirtualDentry().Mount().EndWrite() - } - fd.pipe.Notify(event) } -// OnClose implements vfs.FileDescriptionImpl.OnClose. -func (fd *VFSPipeFD) OnClose(_ context.Context) error { - return nil +// Readiness implements waiter.Waitable.Readiness. +func (fd *VFSPipeFD) Readiness(mask waiter.EventMask) waiter.EventMask { + switch { + case fd.vfsfd.IsReadable() && fd.vfsfd.IsWritable(): + return fd.pipe.rwReadiness() + case fd.vfsfd.IsReadable(): + return fd.pipe.rReadiness() + case fd.vfsfd.IsWritable(): + return fd.pipe.wReadiness() + default: + panic("pipe FD is neither readable nor writable") + } } -// PRead implements vfs.FileDescriptionImpl.PRead. -func (fd *VFSPipeFD) PRead(_ context.Context, _ usermem.IOSequence, _ int64, _ vfs.ReadOptions) (int64, error) { - return 0, syserror.ESPIPE +// EventRegister implements waiter.Waitable.EventRegister. +func (fd *VFSPipeFD) EventRegister(e *waiter.Entry, mask waiter.EventMask) { + fd.pipe.EventRegister(e, mask) +} + +// EventUnregister implements waiter.Waitable.EventUnregister. +func (fd *VFSPipeFD) EventUnregister(e *waiter.Entry) { + fd.pipe.EventUnregister(e) } // Read implements vfs.FileDescriptionImpl.Read. func (fd *VFSPipeFD) Read(ctx context.Context, dst usermem.IOSequence, _ vfs.ReadOptions) (int64, error) { - if !fd.readable { - return 0, syserror.EINVAL - } - return fd.pipe.Read(ctx, dst) } -// PWrite implements vfs.FileDescriptionImpl.PWrite. -func (fd *VFSPipeFD) PWrite(_ context.Context, _ usermem.IOSequence, _ int64, _ vfs.WriteOptions) (int64, error) { - return 0, syserror.ESPIPE -} - // Write implements vfs.FileDescriptionImpl.Write. func (fd *VFSPipeFD) Write(ctx context.Context, src usermem.IOSequence, _ vfs.WriteOptions) (int64, error) { - if !fd.writable { - return 0, syserror.EINVAL - } - return fd.pipe.Write(ctx, src) } @@ -211,3 +215,17 @@ func (fd *VFSPipeFD) Write(ctx context.Context, src usermem.IOSequence, _ vfs.Wr func (fd *VFSPipeFD) Ioctl(ctx context.Context, uio usermem.IO, args arch.SyscallArguments) (uintptr, error) { return fd.pipe.Ioctl(ctx, uio, args) } + +// PipeSize implements fcntl(F_GETPIPE_SZ). +func (fd *VFSPipeFD) PipeSize() int64 { + // Inline Pipe.FifoSize() rather than calling it with nil Context and + // fs.File and ignoring the returned error (which is always nil). + fd.pipe.mu.Lock() + defer fd.pipe.mu.Unlock() + return fd.pipe.max +} + +// SetPipeSize implements fcntl(F_SETPIPE_SZ). +func (fd *VFSPipeFD) SetPipeSize(size int64) (int64, error) { + return fd.pipe.SetFifoSize(size) +} diff --git a/pkg/sentry/syscalls/linux/sys_pipe.go b/pkg/sentry/syscalls/linux/sys_pipe.go index 798344042..43c510930 100644 --- a/pkg/sentry/syscalls/linux/sys_pipe.go +++ b/pkg/sentry/syscalls/linux/sys_pipe.go @@ -24,6 +24,8 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) +// LINT.IfChange + // pipe2 implements the actual system call with flags. func pipe2(t *kernel.Task, addr usermem.Addr, flags uint) (uintptr, error) { if flags&^(linux.O_NONBLOCK|linux.O_CLOEXEC) != 0 { @@ -45,10 +47,12 @@ func pipe2(t *kernel.Task, addr usermem.Addr, flags uint) (uintptr, error) { } if _, err := t.CopyOut(addr, fds); err != nil { - // The files are not closed in this case, the exact semantics - // of this error case are not well defined, but they could have - // already been observed by user space. - return 0, syserror.EFAULT + for _, fd := range fds { + if file, _ := t.FDTable().Remove(fd); file != nil { + file.DecRef() + } + } + return 0, err } return 0, nil } @@ -69,3 +73,5 @@ func Pipe2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall n, err := pipe2(t, addr, flags) return n, nil, err } + +// LINT.ThenChange(vfs2/pipe.go) diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD index b32abfe59..6ff2d84d2 100644 --- a/pkg/sentry/syscalls/linux/vfs2/BUILD +++ b/pkg/sentry/syscalls/linux/vfs2/BUILD @@ -18,6 +18,7 @@ go_library( "linux64_override_arm64.go", "mmap.go", "path.go", + "pipe.go", "poll.go", "read_write.go", "setstat.go", @@ -39,8 +40,10 @@ go_library( "//pkg/gohacks", "//pkg/sentry/arch", "//pkg/sentry/fsbridge", + "//pkg/sentry/fsimpl/pipefs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", + "//pkg/sentry/kernel/pipe", "//pkg/sentry/kernel/time", "//pkg/sentry/limits", "//pkg/sentry/loader", diff --git a/pkg/sentry/syscalls/linux/vfs2/fd.go b/pkg/sentry/syscalls/linux/vfs2/fd.go index 3afcea665..8181d80f4 100644 --- a/pkg/sentry/syscalls/linux/vfs2/fd.go +++ b/pkg/sentry/syscalls/linux/vfs2/fd.go @@ -18,6 +18,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" "gvisor.dev/gvisor/pkg/syserror" ) @@ -140,6 +141,22 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return uintptr(file.StatusFlags()), nil, nil case linux.F_SETFL: return 0, nil, file.SetStatusFlags(t, t.Credentials(), args[2].Uint()) + case linux.F_SETPIPE_SZ: + pipefile, ok := file.Impl().(*pipe.VFSPipeFD) + if !ok { + return 0, nil, syserror.EBADF + } + n, err := pipefile.SetPipeSize(int64(args[2].Int())) + if err != nil { + return 0, nil, err + } + return uintptr(n), nil, nil + case linux.F_GETPIPE_SZ: + pipefile, ok := file.Impl().(*pipe.VFSPipeFD) + if !ok { + return 0, nil, syserror.EBADF + } + return uintptr(pipefile.PipeSize()), nil, nil default: // TODO(gvisor.dev/issue/1623): Everything else is not yet supported. return 0, nil, syserror.EINVAL diff --git a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go index 645e0bcb8..21eb98444 100644 --- a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go +++ b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go @@ -39,7 +39,7 @@ func Override(table map[uintptr]kernel.Syscall) { table[19] = syscalls.Supported("readv", Readv) table[20] = syscalls.Supported("writev", Writev) table[21] = syscalls.Supported("access", Access) - delete(table, 22) // pipe + table[22] = syscalls.Supported("pipe", Pipe) table[23] = syscalls.Supported("select", Select) table[32] = syscalls.Supported("dup", Dup) table[33] = syscalls.Supported("dup2", Dup2) @@ -151,7 +151,7 @@ func Override(table map[uintptr]kernel.Syscall) { delete(table, 290) // eventfd2 table[291] = syscalls.Supported("epoll_create1", EpollCreate1) table[292] = syscalls.Supported("dup3", Dup3) - delete(table, 293) // pipe2 + table[293] = syscalls.Supported("pipe2", Pipe2) delete(table, 294) // inotify_init1 table[295] = syscalls.Supported("preadv", Preadv) table[296] = syscalls.Supported("pwritev", Pwritev) diff --git a/pkg/sentry/syscalls/linux/vfs2/pipe.go b/pkg/sentry/syscalls/linux/vfs2/pipe.go new file mode 100644 index 000000000..4a01e4209 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/pipe.go @@ -0,0 +1,63 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/pipefs" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +// Pipe implements Linux syscall pipe(2). +func Pipe(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + addr := args[0].Pointer() + return 0, nil, pipe2(t, addr, 0) +} + +// Pipe2 implements Linux syscall pipe2(2). +func Pipe2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + addr := args[0].Pointer() + flags := args[1].Int() + return 0, nil, pipe2(t, addr, flags) +} + +func pipe2(t *kernel.Task, addr usermem.Addr, flags int32) error { + if flags&^(linux.O_NONBLOCK|linux.O_CLOEXEC) != 0 { + return syserror.EINVAL + } + r, w := pipefs.NewConnectedPipeFDs(t, t.Kernel().PipeMount(), uint32(flags&linux.O_NONBLOCK)) + defer r.DecRef() + defer w.DecRef() + + fds, err := t.NewFDsVFS2(0, []*vfs.FileDescription{r, w}, kernel.FDFlags{ + CloseOnExec: flags&linux.O_CLOEXEC != 0, + }) + if err != nil { + return err + } + if _, err := t.CopyOut(addr, fds); err != nil { + for _, fd := range fds { + if _, file := t.FDTable().Remove(fd); file != nil { + file.DecRef() + } + } + return err + } + return nil +} diff --git a/pkg/sentry/syscalls/linux/vfs2/read_write.go b/pkg/sentry/syscalls/linux/vfs2/read_write.go index 898b190fd..6c6998f45 100644 --- a/pkg/sentry/syscalls/linux/vfs2/read_write.go +++ b/pkg/sentry/syscalls/linux/vfs2/read_write.go @@ -103,7 +103,7 @@ func read(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, opt // Issue the request and break out if it completes with anything other than // "would block". - n, err := file.Read(t, dst, opts) + n, err = file.Read(t, dst, opts) total += n if err != syserror.ErrWouldBlock { break @@ -248,7 +248,7 @@ func pread(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, of // Issue the request and break out if it completes with anything other than // "would block". - n, err := file.PRead(t, dst, offset+total, opts) + n, err = file.PRead(t, dst, offset+total, opts) total += n if err != syserror.ErrWouldBlock { break @@ -335,7 +335,7 @@ func write(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, op // Issue the request and break out if it completes with anything other than // "would block". - n, err := file.Write(t, src, opts) + n, err = file.Write(t, src, opts) total += n if err != syserror.ErrWouldBlock { break @@ -480,7 +480,7 @@ func pwrite(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, o // Issue the request and break out if it completes with anything other than // "would block". - n, err := file.PWrite(t, src, offset+total, opts) + n, err = file.PWrite(t, src, offset+total, opts) total += n if err != syserror.ErrWouldBlock { break diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 053c6e1d1..cb5bbd781 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -335,7 +335,7 @@ func (vfs *VirtualFilesystem) MknodAt(ctx context.Context, creds *auth.Credentia rp := vfs.getResolvingPath(creds, pop) for { err := rp.mount.fs.impl.MknodAt(ctx, rp, *opts) - if err != nil { + if err == nil { vfs.putResolvingPath(rp) return nil } diff --git a/test/syscalls/linux/pipe.cc b/test/syscalls/linux/pipe.cc index d8e19e910..67228b66b 100644 --- a/test/syscalls/linux/pipe.cc +++ b/test/syscalls/linux/pipe.cc @@ -265,6 +265,8 @@ TEST_P(PipeTest, OffsetCalls) { SyscallFailsWithErrno(ESPIPE)); struct iovec iov; + iov.iov_base = &buf; + iov.iov_len = sizeof(buf); EXPECT_THAT(preadv(wfd_.get(), &iov, 1, 0), SyscallFailsWithErrno(ESPIPE)); EXPECT_THAT(pwritev(rfd_.get(), &iov, 1, 0), SyscallFailsWithErrno(ESPIPE)); } -- cgit v1.2.3 From fe001edb14e6e879ab4ebca0d2ac71d770ac8cce Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Tue, 11 Feb 2020 02:35:39 -0500 Subject: Arm64: VDSO support for signal The vdso is enabled, so we can use the sigreturn trampolines the vdso provides in arch module. Signed-off-by: Bin Lu --- pkg/sentry/loader/loader.go | 9 +++++++++ pkg/sentry/loader/vdso.go | 25 +++++++++++++++++++++++++ vdso/syscalls.h | 25 +++++++++++++------------ vdso/vdso.cc | 12 ++++++------ vdso/vdso_amd64.lds | 1 + 5 files changed, 54 insertions(+), 18 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/loader/loader.go b/pkg/sentry/loader/loader.go index 9a613d6b7..57422706d 100644 --- a/pkg/sentry/loader/loader.go +++ b/pkg/sentry/loader/loader.go @@ -392,6 +392,15 @@ func Load(ctx context.Context, args LoadArgs, extraAuxv []arch.AuxEntry, vdso *V m.SetAuxv(auxv) m.SetExecutable(d) + symbolValue, err := getSymbolValueFromVDSO("rt_sigreturn") + if err != nil { + return 0, nil, "", syserr.NewDynamic(fmt.Sprintf("Failed to find rt_sigreturn in vdso: %v", err), syserr.FromError(err).ToLinux()) + } + + // Found rt_sigretrun. + addr := uint64(vdsoAddr) + symbolValue - vdsoPrelink + m.SetVDSOSigReturn(addr) + ac.SetIP(uintptr(loaded.entry)) ac.SetStack(uintptr(stack.Bottom)) diff --git a/pkg/sentry/loader/vdso.go b/pkg/sentry/loader/vdso.go index 52f446ed7..01eaefd26 100644 --- a/pkg/sentry/loader/vdso.go +++ b/pkg/sentry/loader/vdso.go @@ -15,9 +15,11 @@ package loader import ( + "bytes" "debug/elf" "fmt" "io" + "strings" "gvisor.dev/gvisor/pkg/abi" "gvisor.dev/gvisor/pkg/context" @@ -37,6 +39,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +const vdsoPrelink = 0xffffffffff700000 + type fileContext struct { context.Context } @@ -218,6 +222,27 @@ type VDSO struct { phdrs []elf.ProgHeader `state:".([]elfProgHeader)"` } +// getSymbolValueFromVDSO returns the specific symbol value in vdso.so. +func getSymbolValueFromVDSO(symbol string) (uint64, error) { + f, err := elf.NewFile(bytes.NewReader(vdsoBin)) + if err != nil { + return 0, err + } + syms, err := f.Symbols() + if err != nil { + return 0, err + } + + for _, sym := range syms { + if elf.ST_BIND(sym.Info) != elf.STB_LOCAL && sym.Section != elf.SHN_UNDEF { + if strings.Contains(sym.Name, symbol) { + return sym.Value, nil + } + } + } + return 0, fmt.Errorf("no %v in vdso.so", symbol) +} + // PrepareVDSO validates the system VDSO and returns a VDSO, containing the // param page for updating by the kernel. func PrepareVDSO(ctx context.Context, mfp pgalloc.MemoryFileProvider) (*VDSO, error) { diff --git a/vdso/syscalls.h b/vdso/syscalls.h index b6d15a7d3..f630ae563 100644 --- a/vdso/syscalls.h +++ b/vdso/syscalls.h @@ -26,6 +26,9 @@ #include #include +#define __stringify_1(x...) #x +#define __stringify(x...) __stringify_1(x) + namespace vdso { #if __x86_64__ @@ -51,20 +54,13 @@ static inline int sys_getcpu(unsigned* cpu, unsigned* node, return num; } -#elif __aarch64__ - -static inline int sys_rt_sigreturn(void) { - int num = __NR_rt_sigreturn; - - asm volatile( - "mov x8, %0\n" - "svc #0 \n" - : "+r"(num) - : - :); - return num; +static inline void sys_rt_sigreturn(void) { + asm volatile("movl $" __stringify(__NR_rt_sigreturn)", %eax \n" + "syscall \n"); } +#elif __aarch64__ + static inline int sys_clock_gettime(clockid_t _clkid, struct timespec* _ts) { register struct timespec* ts asm("x1") = _ts; register clockid_t clkid asm("x0") = _clkid; @@ -91,6 +87,11 @@ static inline int sys_clock_getres(clockid_t _clkid, struct timespec* _ts) { return ret; } +static inline void sys_rt_sigreturn(void) { + asm volatile("mov x8, #" __stringify(__NR_rt_sigreturn)" \n" + "svc #0 \n"); +} + #else #error "unsupported architecture" #endif diff --git a/vdso/vdso.cc b/vdso/vdso.cc index 8bb80a7a4..62f59766d 100644 --- a/vdso/vdso.cc +++ b/vdso/vdso.cc @@ -69,6 +69,12 @@ int __common_gettimeofday(struct timeval* tv, struct timezone* tz) { } } // namespace +// __kernel_rt_sigreturn() implements rt_sigreturn() +extern "C" void __kernel_rt_sigreturn(unsigned long unused) { + // No optimizations yet, just make the real system call. + sys_rt_sigreturn(); +} + #if __x86_64__ // __vdso_clock_gettime() implements clock_gettime() @@ -139,12 +145,6 @@ extern "C" int __kernel_clock_getres(clockid_t clock, struct timespec* res) { return ret; } -// __kernel_rt_sigreturn() implements gettimeofday() -extern "C" int __kernel_rt_sigreturn(unsigned long unused) { - // No optimizations yet, just make the real system call. - return sys_rt_sigreturn(); -} - #else #error "unsupported architecture" #endif diff --git a/vdso/vdso_amd64.lds b/vdso/vdso_amd64.lds index e2615ae9e..d114290da 100644 --- a/vdso/vdso_amd64.lds +++ b/vdso/vdso_amd64.lds @@ -95,6 +95,7 @@ VERSION { __vdso_getcpu; time; __vdso_time; + __kernel_rt_sigreturn; local: *; }; -- cgit v1.2.3 From b4de018a67f5b5cb5ffc782c915107e1402ed833 Mon Sep 17 00:00:00 2001 From: Tamir Duberstein Date: Fri, 17 Apr 2020 06:40:23 -0700 Subject: Permit setting unknown options This previously changed in 305699233, but this behaviour turned out to be load bearing. PiperOrigin-RevId: 307033802 --- pkg/sentry/socket/unix/transport/unix.go | 2 -- pkg/tcpip/transport/icmp/endpoint.go | 4 +--- pkg/tcpip/transport/tcp/endpoint.go | 4 ---- pkg/tcpip/transport/udp/endpoint.go | 4 ---- 4 files changed, 1 insertion(+), 13 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/unix/transport/unix.go b/pkg/sentry/socket/unix/transport/unix.go index 1f3880cc5..2f1b127df 100644 --- a/pkg/sentry/socket/unix/transport/unix.go +++ b/pkg/sentry/socket/unix/transport/unix.go @@ -850,7 +850,6 @@ func (e *baseEndpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Erro case tcpip.ReuseAddressOption: default: log.Warningf("Unsupported socket option: %d", opt) - return tcpip.ErrUnknownProtocolOption } return nil } @@ -861,7 +860,6 @@ func (e *baseEndpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { case tcpip.ReceiveBufferSizeOption: default: log.Warningf("Unsupported socket option: %d", opt) - return tcpip.ErrUnknownProtocolOption } return nil } diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index 3a133eef9..feef8dca0 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -353,7 +353,7 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { // SetSockOptBool sets a socket option. Currently not supported. func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { - return tcpip.ErrUnknownProtocolOption + return nil } // SetSockOptInt sets a socket option. Currently not supported. @@ -364,8 +364,6 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { e.ttl = uint8(v) e.mu.Unlock() - default: - return tcpip.ErrUnknownProtocolOption } return nil } diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 5d0ea9e93..7e8def82d 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -1477,8 +1477,6 @@ func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { e.LockUser() e.v6only = v e.UnlockUser() - default: - return tcpip.ErrUnknownProtocolOption } return nil @@ -1592,8 +1590,6 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { e.ttl = uint8(v) e.UnlockUser() - default: - return tcpip.ErrUnknownProtocolOption } return nil } diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 492cc1fcb..edb54f0be 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -553,8 +553,6 @@ func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { } e.v6only = v - default: - return tcpip.ErrUnknownProtocolOption } return nil @@ -586,8 +584,6 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { case tcpip.ReceiveBufferSizeOption: case tcpip.SendBufferSizeOption: - default: - return tcpip.ErrUnknownProtocolOption } return nil -- cgit v1.2.3 From 12bde95635ac266aab8087b4705372bb177638f3 Mon Sep 17 00:00:00 2001 From: Zach Koopmans Date: Fri, 17 Apr 2020 10:38:04 -0700 Subject: Get /bin/true to run on VFS2 Included: - loader_test.go RunTest and TestStartSignal VFS2 - container_test.go TestAppExitStatus on VFS2 - experimental flag added to runsc to turn on VFS2 Note: shared mounts are not yet supported. PiperOrigin-RevId: 307070753 --- pkg/sentry/kernel/syscalls.go | 7 + runsc/boot/BUILD | 11 ++ runsc/boot/config.go | 5 + runsc/boot/fds.go | 33 ++++ runsc/boot/fs.go | 9 +- runsc/boot/loader.go | 31 +++- runsc/boot/loader_amd64.go | 5 +- runsc/boot/loader_arm64.go | 5 +- runsc/boot/loader_test.go | 37 ++++- runsc/boot/user.go | 64 ++++++++ runsc/boot/vfs.go | 310 ++++++++++++++++++++++++++++++++++++++ runsc/container/container_test.go | 14 +- runsc/main.go | 3 + 13 files changed, 513 insertions(+), 21 deletions(-) create mode 100644 runsc/boot/vfs.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/syscalls.go b/pkg/sentry/kernel/syscalls.go index 2e3565747..84156d5a1 100644 --- a/pkg/sentry/kernel/syscalls.go +++ b/pkg/sentry/kernel/syscalls.go @@ -326,6 +326,13 @@ func RegisterSyscallTable(s *SyscallTable) { allSyscallTables = append(allSyscallTables, s) } +// FlushSyscallTablesTestOnly flushes the syscall tables for tests. Used for +// parameterized VFSv2 tests. +// TODO(gvisor.dv/issue/1624): Remove when VFS1 is no longer supported. +func FlushSyscallTablesTestOnly() { + allSyscallTables = nil +} + // Lookup returns the syscall implementation, if one exists. func (s *SyscallTable) Lookup(sysno uintptr) SyscallFn { if sysno < uintptr(len(s.lookup)) { diff --git a/runsc/boot/BUILD b/runsc/boot/BUILD index 23f42382f..5451f1eba 100644 --- a/runsc/boot/BUILD +++ b/runsc/boot/BUILD @@ -21,6 +21,7 @@ go_library( "network.go", "strace.go", "user.go", + "vfs.go", ], visibility = [ "//runsc:__subpackages__", @@ -33,6 +34,7 @@ go_library( "//pkg/control/server", "//pkg/cpuid", "//pkg/eventchannel", + "//pkg/fspath", "//pkg/log", "//pkg/memutil", "//pkg/rand", @@ -40,6 +42,7 @@ go_library( "//pkg/sentry/arch", "//pkg/sentry/arch:registers_go_proto", "//pkg/sentry/control", + "//pkg/sentry/devices/memdev", "//pkg/sentry/fs", "//pkg/sentry/fs/dev", "//pkg/sentry/fs/gofer", @@ -49,6 +52,12 @@ go_library( "//pkg/sentry/fs/sys", "//pkg/sentry/fs/tmpfs", "//pkg/sentry/fs/tty", + "//pkg/sentry/fsimpl/devtmpfs", + "//pkg/sentry/fsimpl/gofer", + "//pkg/sentry/fsimpl/host", + "//pkg/sentry/fsimpl/proc", + "//pkg/sentry/fsimpl/sys", + "//pkg/sentry/fsimpl/tmpfs", "//pkg/sentry/inet", "//pkg/sentry/kernel", "//pkg/sentry/kernel:uncaught_signal_go_proto", @@ -71,6 +80,7 @@ go_library( "//pkg/sentry/time", "//pkg/sentry/unimpl:unimplemented_syscall_go_proto", "//pkg/sentry/usage", + "//pkg/sentry/vfs", "//pkg/sentry/watchdog", "//pkg/sync", "//pkg/syserror", @@ -114,6 +124,7 @@ go_test( "//pkg/p9", "//pkg/sentry/contexttest", "//pkg/sentry/fs", + "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sync", "//pkg/unet", diff --git a/runsc/boot/config.go b/runsc/boot/config.go index 7ea5bfade..715a19112 100644 --- a/runsc/boot/config.go +++ b/runsc/boot/config.go @@ -305,5 +305,10 @@ func (c *Config) ToFlags() []string { if len(c.TestOnlyTestNameEnv) != 0 { f = append(f, "--TESTONLY-test-name-env="+c.TestOnlyTestNameEnv) } + + if c.VFS2 { + f = append(f, "--vfs2=true") + } + return f } diff --git a/runsc/boot/fds.go b/runsc/boot/fds.go index 5314b0f2a..7e49f6f9f 100644 --- a/runsc/boot/fds.go +++ b/runsc/boot/fds.go @@ -20,6 +20,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/host" + vfshost "gvisor.dev/gvisor/pkg/sentry/fsimpl/host" "gvisor.dev/gvisor/pkg/sentry/kernel" ) @@ -31,6 +32,10 @@ func createFDTable(ctx context.Context, console bool, stdioFDs []int) (*kernel.F return nil, fmt.Errorf("stdioFDs should contain exactly 3 FDs (stdin, stdout, and stderr), but %d FDs received", len(stdioFDs)) } + if kernel.VFS2Enabled { + return createFDTableVFS2(ctx, console, stdioFDs) + } + k := kernel.KernelFromContext(ctx) fdTable := k.NewFDTable() defer fdTable.DecRef() @@ -78,3 +83,31 @@ func createFDTable(ctx context.Context, console bool, stdioFDs []int) (*kernel.F fdTable.IncRef() return fdTable, nil } + +func createFDTableVFS2(ctx context.Context, console bool, stdioFDs []int) (*kernel.FDTable, error) { + k := kernel.KernelFromContext(ctx) + fdTable := k.NewFDTable() + defer fdTable.DecRef() + + hostMount, err := vfshost.NewMount(k.VFS()) + if err != nil { + return nil, fmt.Errorf("creating host mount: %w", err) + } + + for appFD, hostFD := range stdioFDs { + // TODO(gvisor.dev/issue/1482): Add TTY support. + appFile, err := vfshost.ImportFD(hostMount, hostFD, false) + if err != nil { + return nil, err + } + + if err := fdTable.NewFDAtVFS2(ctx, int32(appFD), appFile, kernel.FDFlags{}); err != nil { + appFile.DecRef() + return nil, err + } + appFile.DecRef() + } + + fdTable.IncRef() + return fdTable, nil +} diff --git a/runsc/boot/fs.go b/runsc/boot/fs.go index 82cc612d2..98cce60af 100644 --- a/runsc/boot/fs.go +++ b/runsc/boot/fs.go @@ -278,6 +278,9 @@ func subtargets(root string, mnts []specs.Mount) []string { } func setupContainerFS(ctx context.Context, conf *Config, mntr *containerMounter, procArgs *kernel.CreateProcessArgs) error { + if conf.VFS2 { + return setupContainerVFS2(ctx, conf, mntr, procArgs) + } mns, err := mntr.setupFS(conf, procArgs) if err != nil { return err @@ -573,6 +576,9 @@ func newContainerMounter(spec *specs.Spec, goferFDs []int, k *kernel.Kernel, hin // should be mounted (e.g. a volume shared between containers). It must be // called for the root container only. func (c *containerMounter) processHints(conf *Config) error { + if conf.VFS2 { + return nil + } ctx := c.k.SupervisorContext() for _, hint := range c.hints.mounts { // TODO(b/142076984): Only support tmpfs for now. Bind mounts require a @@ -781,9 +787,6 @@ func (c *containerMounter) getMountNameAndOptions(conf *Config, m specs.Mount) ( useOverlay = conf.Overlay && !mountFlags(m.Options).ReadOnly default: - // TODO(nlacasse): Support all the mount types and make this a fatal error. - // Most applications will "just work" without them, so this is a warning - // for now. log.Warningf("ignoring unknown filesystem type %q", m.Type) } return fsName, opts, useOverlay, nil diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index 654441f65..cf1f47bc7 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -26,7 +26,6 @@ import ( specs "github.com/opencontainers/runtime-spec/specs-go" "golang.org/x/sys/unix" - "gvisor.dev/gvisor/pkg/abi" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/log" @@ -73,6 +72,8 @@ import ( _ "gvisor.dev/gvisor/pkg/sentry/socket/unix" ) +var syscallTable *kernel.SyscallTable + // Loader keeps state needed to start the kernel and run the container.. type Loader struct { // k is the kernel. @@ -195,13 +196,14 @@ func New(args Args) (*Loader, error) { return nil, fmt.Errorf("setting up memory usage: %v", err) } - if args.Conf.VFS2 { - st, ok := kernel.LookupSyscallTable(abi.Linux, arch.Host) - if ok { - vfs2.Override(st.Table) - } + // Patch the syscall table. + kernel.VFS2Enabled = args.Conf.VFS2 + if kernel.VFS2Enabled { + vfs2.Override(syscallTable.Table) } + kernel.RegisterSyscallTable(syscallTable) + // Create kernel and platform. p, err := createPlatform(args.Conf, args.Device) if err != nil { @@ -392,11 +394,16 @@ func newProcess(id string, spec *specs.Spec, creds *auth.Credentials, k *kernel. return kernel.CreateProcessArgs{}, fmt.Errorf("creating limits: %v", err) } + wd := spec.Process.Cwd + if wd == "" { + wd = "/" + } + // Create the process arguments. procArgs := kernel.CreateProcessArgs{ Argv: spec.Process.Args, Envv: spec.Process.Env, - WorkingDirectory: spec.Process.Cwd, // Defaults to '/' if empty. + WorkingDirectory: wd, Credentials: creds, Umask: 0022, Limits: ls, @@ -541,7 +548,15 @@ func (l *Loader) run() error { } // Add the HOME enviroment variable if it is not already set. - envv, err := maybeAddExecUserHome(ctx, l.rootProcArgs.MountNamespace, l.rootProcArgs.Credentials.RealKUID, l.rootProcArgs.Envv) + var envv []string + if kernel.VFS2Enabled { + envv, err = maybeAddExecUserHomeVFS2(ctx, l.rootProcArgs.MountNamespaceVFS2, + l.rootProcArgs.Credentials.RealKUID, l.rootProcArgs.Envv) + + } else { + envv, err = maybeAddExecUserHome(ctx, l.rootProcArgs.MountNamespace, + l.rootProcArgs.Credentials.RealKUID, l.rootProcArgs.Envv) + } if err != nil { return err } diff --git a/runsc/boot/loader_amd64.go b/runsc/boot/loader_amd64.go index b9669f2ac..78df86611 100644 --- a/runsc/boot/loader_amd64.go +++ b/runsc/boot/loader_amd64.go @@ -17,11 +17,10 @@ package boot import ( - "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" ) func init() { - // Register the global syscall table. - kernel.RegisterSyscallTable(linux.AMD64) + // Set the global syscall table. + syscallTable = linux.AMD64 } diff --git a/runsc/boot/loader_arm64.go b/runsc/boot/loader_arm64.go index cf64d28c8..250785010 100644 --- a/runsc/boot/loader_arm64.go +++ b/runsc/boot/loader_arm64.go @@ -17,11 +17,10 @@ package boot import ( - "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" ) func init() { - // Register the global syscall table. - kernel.RegisterSyscallTable(linux.ARM64) + // Set the global syscall table. + syscallTable = linux.ARM64 } diff --git a/runsc/boot/loader_test.go b/runsc/boot/loader_test.go index c9a75b76d..e7c71734f 100644 --- a/runsc/boot/loader_test.go +++ b/runsc/boot/loader_test.go @@ -30,6 +30,7 @@ import ( "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/unet" "gvisor.dev/gvisor/runsc/fsgofer" @@ -66,6 +67,11 @@ func testSpec() *specs.Spec { } } +func resetSyscallTable() { + kernel.VFS2Enabled = false + kernel.FlushSyscallTablesTestOnly() +} + // startGofer starts a new gofer routine serving 'root' path. It returns the // sandbox side of the connection, and a function that when called will stop the // gofer. @@ -101,7 +107,7 @@ func startGofer(root string) (int, func(), error) { return sandboxEnd, cleanup, nil } -func createLoader() (*Loader, func(), error) { +func createLoader(vfsEnabled bool) (*Loader, func(), error) { fd, err := server.CreateSocket(ControlSocketAddr(fmt.Sprintf("%010d", rand.Int())[:10])) if err != nil { return nil, nil, err @@ -109,6 +115,8 @@ func createLoader() (*Loader, func(), error) { conf := testConfig() spec := testSpec() + conf.VFS2 = vfsEnabled + sandEnd, cleanup, err := startGofer(spec.Root.Path) if err != nil { return nil, nil, err @@ -142,10 +150,22 @@ func createLoader() (*Loader, func(), error) { // TestRun runs a simple application in a sandbox and checks that it succeeds. func TestRun(t *testing.T) { - l, cleanup, err := createLoader() + defer resetSyscallTable() + doRun(t, false) +} + +// TestRunVFS2 runs TestRun in VFSv2. +func TestRunVFS2(t *testing.T) { + defer resetSyscallTable() + doRun(t, true) +} + +func doRun(t *testing.T, vfsEnabled bool) { + l, cleanup, err := createLoader(vfsEnabled) if err != nil { t.Fatalf("error creating loader: %v", err) } + defer l.Destroy() defer cleanup() @@ -179,7 +199,18 @@ func TestRun(t *testing.T) { // TestStartSignal tests that the controller Start message will cause // WaitForStartSignal to return. func TestStartSignal(t *testing.T) { - l, cleanup, err := createLoader() + defer resetSyscallTable() + doStartSignal(t, false) +} + +// TestStartSignalVFS2 does TestStartSignal with VFS2. +func TestStartSignalVFS2(t *testing.T) { + defer resetSyscallTable() + doStartSignal(t, true) +} + +func doStartSignal(t *testing.T, vfsEnabled bool) { + l, cleanup, err := createLoader(vfsEnabled) if err != nil { t.Fatalf("error creating loader: %v", err) } diff --git a/runsc/boot/user.go b/runsc/boot/user.go index f0aa52135..332e4fce5 100644 --- a/runsc/boot/user.go +++ b/runsc/boot/user.go @@ -23,8 +23,10 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/usermem" ) @@ -84,6 +86,48 @@ func getExecUserHome(ctx context.Context, rootMns *fs.MountNamespace, uid auth.K File: f, } + return findHomeInPasswd(uint32(uid), r, defaultHome) +} + +type fileReaderVFS2 struct { + ctx context.Context + fd *vfs.FileDescription +} + +func (r *fileReaderVFS2) Read(buf []byte) (int, error) { + n, err := r.fd.Read(r.ctx, usermem.BytesIOSequence(buf), vfs.ReadOptions{}) + return int(n), err +} + +func getExecUserHomeVFS2(ctx context.Context, mns *vfs.MountNamespace, uid auth.KUID) (string, error) { + const defaultHome = "/" + + root := mns.Root() + defer root.DecRef() + + creds := auth.CredentialsFromContext(ctx) + + target := &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse("/etc/passwd"), + } + + opts := &vfs.OpenOptions{ + Flags: linux.O_RDONLY, + } + + fd, err := root.Mount().Filesystem().VirtualFilesystem().OpenAt(ctx, creds, target, opts) + if err != nil { + return defaultHome, nil + } + defer fd.DecRef() + + r := &fileReaderVFS2{ + ctx: ctx, + fd: fd, + } + homeDir, err := findHomeInPasswd(uint32(uid), r, defaultHome) if err != nil { return "", err @@ -111,6 +155,26 @@ func maybeAddExecUserHome(ctx context.Context, mns *fs.MountNamespace, uid auth. if err != nil { return nil, fmt.Errorf("error reading exec user: %v", err) } + + return append(envv, "HOME="+homeDir), nil +} + +func maybeAddExecUserHomeVFS2(ctx context.Context, vmns *vfs.MountNamespace, uid auth.KUID, envv []string) ([]string, error) { + // Check if the envv already contains HOME. + for _, env := range envv { + if strings.HasPrefix(env, "HOME=") { + // We have it. Return the original slice unmodified. + return envv, nil + } + } + + // Read /etc/passwd for the user's HOME directory and set the HOME + // environment variable as required by POSIX if it is not overridden by + // the user. + homeDir, err := getExecUserHomeVFS2(ctx, vmns, uid) + if err != nil { + return nil, fmt.Errorf("error reading exec user: %v", err) + } return append(envv, "HOME="+homeDir), nil } diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go new file mode 100644 index 000000000..82083c57d --- /dev/null +++ b/runsc/boot/vfs.go @@ -0,0 +1,310 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package boot + +import ( + "fmt" + "path" + "strconv" + "strings" + + specs "github.com/opencontainers/runtime-spec/specs-go" + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/devices/memdev" + "gvisor.dev/gvisor/pkg/sentry/fs" + devtmpfsimpl "gvisor.dev/gvisor/pkg/sentry/fsimpl/devtmpfs" + goferimpl "gvisor.dev/gvisor/pkg/sentry/fsimpl/gofer" + procimpl "gvisor.dev/gvisor/pkg/sentry/fsimpl/proc" + sysimpl "gvisor.dev/gvisor/pkg/sentry/fsimpl/sys" + tmpfsimpl "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" + "gvisor.dev/gvisor/pkg/syserror" + + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" +) + +func registerFilesystems(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials) error { + + vfsObj.MustRegisterFilesystemType(rootFsName, &goferimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserList: true, + }) + + vfsObj.MustRegisterFilesystemType(bind, &goferimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserList: true, + }) + + vfsObj.MustRegisterFilesystemType(devpts, &devtmpfsimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + AllowUserList: true, + }) + + vfsObj.MustRegisterFilesystemType(devtmpfs, &devtmpfsimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + AllowUserList: true, + }) + vfsObj.MustRegisterFilesystemType(proc, &procimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + AllowUserList: true, + }) + vfsObj.MustRegisterFilesystemType(sysfs, &sysimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + AllowUserList: true, + }) + vfsObj.MustRegisterFilesystemType(tmpfs, &tmpfsimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + AllowUserList: true, + }) + vfsObj.MustRegisterFilesystemType(nonefs, &sysimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + AllowUserList: true, + }) + + // Setup files in devtmpfs. + if err := memdev.Register(vfsObj); err != nil { + return fmt.Errorf("registering memdev: %w", err) + } + a, err := devtmpfsimpl.NewAccessor(ctx, vfsObj, creds, devtmpfsimpl.Name) + if err != nil { + return fmt.Errorf("creating devtmpfs accessor: %w", err) + } + defer a.Release() + + if err := a.UserspaceInit(ctx); err != nil { + return fmt.Errorf("initializing userspace: %w", err) + } + if err := memdev.CreateDevtmpfsFiles(ctx, a); err != nil { + return fmt.Errorf("creating devtmpfs files: %w", err) + } + return nil +} + +func setupContainerVFS2(ctx context.Context, conf *Config, mntr *containerMounter, procArgs *kernel.CreateProcessArgs) error { + if err := mntr.k.VFS().Init(); err != nil { + return fmt.Errorf("failed to initialize VFS: %w", err) + } + mns, err := mntr.setupVFS2(ctx, conf, procArgs) + if err != nil { + return fmt.Errorf("failed to setupFS: %w", err) + } + procArgs.MountNamespaceVFS2 = mns + return setExecutablePathVFS2(ctx, procArgs) +} + +func setExecutablePathVFS2(ctx context.Context, procArgs *kernel.CreateProcessArgs) error { + + exe := procArgs.Argv[0] + + // Absolute paths can be used directly. + if path.IsAbs(exe) { + procArgs.Filename = exe + return nil + } + + // Paths with '/' in them should be joined to the working directory, or + // to the root if working directory is not set. + if strings.IndexByte(exe, '/') > 0 { + + if !path.IsAbs(procArgs.WorkingDirectory) { + return fmt.Errorf("working directory %q must be absolute", procArgs.WorkingDirectory) + } + + procArgs.Filename = path.Join(procArgs.WorkingDirectory, exe) + return nil + } + + // Paths with a '/' are relative to the CWD. + if strings.IndexByte(exe, '/') > 0 { + procArgs.Filename = path.Join(procArgs.WorkingDirectory, exe) + return nil + } + + // Otherwise, We must lookup the name in the paths, starting from the + // root directory. + root := procArgs.MountNamespaceVFS2.Root() + defer root.DecRef() + + paths := fs.GetPath(procArgs.Envv) + creds := procArgs.Credentials + + for _, p := range paths { + + binPath := path.Join(p, exe) + + pop := &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(binPath), + FollowFinalSymlink: true, + } + + opts := &vfs.OpenOptions{ + FileExec: true, + Flags: linux.O_RDONLY, + } + + dentry, err := root.Mount().Filesystem().VirtualFilesystem().OpenAt(ctx, creds, pop, opts) + if err == syserror.ENOENT || err == syserror.EACCES { + // Didn't find it here. + continue + } + if err != nil { + return err + } + dentry.DecRef() + + procArgs.Filename = binPath + return nil + } + + return fmt.Errorf("executable %q not found in $PATH=%q", exe, strings.Join(paths, ":")) +} + +func (c *containerMounter) setupVFS2(ctx context.Context, conf *Config, procArgs *kernel.CreateProcessArgs) (*vfs.MountNamespace, error) { + log.Infof("Configuring container's file system with VFS2") + + // Create context with root credentials to mount the filesystem (the current + // user may not be privileged enough). + rootProcArgs := *procArgs + rootProcArgs.WorkingDirectory = "/" + rootProcArgs.Credentials = auth.NewRootCredentials(procArgs.Credentials.UserNamespace) + rootProcArgs.Umask = 0022 + rootProcArgs.MaxSymlinkTraversals = linux.MaxSymlinkTraversals + rootCtx := procArgs.NewContext(c.k) + + creds := procArgs.Credentials + if err := registerFilesystems(rootCtx, c.k.VFS(), creds); err != nil { + return nil, fmt.Errorf("register filesystems: %w", err) + } + + fd := c.fds.remove() + + opts := strings.Join(p9MountOptionsVFS2(fd, conf.FileAccess), ",") + + log.Infof("Mounting root over 9P, ioFD: %d", fd) + mns, err := c.k.VFS().NewMountNamespace(ctx, creds, "", rootFsName, &vfs.GetFilesystemOptions{Data: opts}) + if err != nil { + return nil, fmt.Errorf("setting up mountnamespace: %w", err) + } + + rootProcArgs.MountNamespaceVFS2 = mns + + // Mount submounts. + if err := c.mountSubmountsVFS2(rootCtx, conf, mns, creds); err != nil { + return nil, fmt.Errorf("mounting submounts vfs2: %w", err) + } + + return mns, nil +} + +func (c *containerMounter) mountSubmountsVFS2(ctx context.Context, conf *Config, mns *vfs.MountNamespace, creds *auth.Credentials) error { + + for _, submount := range c.mounts { + log.Debugf("Mounting %q to %q, type: %s, options: %s", submount.Source, submount.Destination, submount.Type, submount.Options) + if err := c.mountSubmountVFS2(ctx, conf, mns, creds, &submount); err != nil { + return err + } + } + + // TODO(gvisor.dev/issue/1487): implement mountTmp from fs.go. + + return c.checkDispenser() +} + +// TODO(gvisor.dev/issue/1487): Implement submount options similar to the VFS1 version. +func (c *containerMounter) mountSubmountVFS2(ctx context.Context, conf *Config, mns *vfs.MountNamespace, creds *auth.Credentials, submount *specs.Mount) error { + root := mns.Root() + defer root.DecRef() + target := &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(submount.Destination), + } + + _, options, useOverlay, err := c.getMountNameAndOptionsVFS2(conf, *submount) + if err != nil { + return fmt.Errorf("mountOptions failed: %w", err) + } + + opts := &vfs.MountOptions{ + GetFilesystemOptions: vfs.GetFilesystemOptions{ + Data: strings.Join(options, ","), + }, + InternalMount: true, + } + + // All writes go to upper, be paranoid and make lower readonly. + opts.ReadOnly = useOverlay + + if err := c.k.VFS().MountAt(ctx, creds, "", target, submount.Type, opts); err != nil { + return fmt.Errorf("failed to mount %q (type: %s): %w, opts: %v", submount.Destination, submount.Type, err, opts) + } + log.Infof("Mounted %q to %q type: %s, internal-options: %q", submount.Source, submount.Destination, submount.Type, opts) + return nil +} + +// getMountNameAndOptionsVFS2 retrieves the fsName, opts, and useOverlay values +// used for mounts. +func (c *containerMounter) getMountNameAndOptionsVFS2(conf *Config, m specs.Mount) (string, []string, bool, error) { + var ( + fsName string + opts []string + useOverlay bool + ) + + switch m.Type { + case devpts, devtmpfs, proc, sysfs: + fsName = m.Type + case nonefs: + fsName = sysfs + case tmpfs: + fsName = m.Type + + var err error + opts, err = parseAndFilterOptions(m.Options, tmpfsAllowedOptions...) + if err != nil { + return "", nil, false, err + } + + case bind: + fd := c.fds.remove() + fsName = "9p" + opts = p9MountOptionsVFS2(fd, c.getMountAccessType(m)) + // If configured, add overlay to all writable mounts. + useOverlay = conf.Overlay && !mountFlags(m.Options).ReadOnly + + default: + log.Warningf("ignoring unknown filesystem type %q", m.Type) + } + return fsName, opts, useOverlay, nil +} + +// p9MountOptions creates a slice of options for a p9 mount. +// TODO(gvisor.dev/issue/1200): Remove this version in favor of the one in +// fs.go when privateunixsocket lands. +func p9MountOptionsVFS2(fd int, fa FileAccessType) []string { + opts := []string{ + "trans=fd", + "rfdno=" + strconv.Itoa(fd), + "wfdno=" + strconv.Itoa(fd), + } + if fa == FileAccessShared { + opts = append(opts, "cache=remote_revalidating") + } + return opts +} diff --git a/runsc/container/container_test.go b/runsc/container/container_test.go index 442e80ac0..24f9ecc35 100644 --- a/runsc/container/container_test.go +++ b/runsc/container/container_test.go @@ -521,9 +521,21 @@ func TestExePath(t *testing.T) { // Test the we can retrieve the application exit status from the container. func TestAppExitStatus(t *testing.T) { + conf := testutil.TestConfig() + conf.VFS2 = false + doAppExitStatus(t, conf) +} + +// This is TestAppExitStatus for VFSv2. +func TestAppExitStatusVFS2(t *testing.T) { + conf := testutil.TestConfig() + conf.VFS2 = true + doAppExitStatus(t, conf) +} + +func doAppExitStatus(t *testing.T, conf *boot.Config) { // First container will succeed. succSpec := testutil.NewSpecWithArgs("true") - conf := testutil.TestConfig() rootDir, bundleDir, err := testutil.SetupContainer(succSpec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) diff --git a/runsc/main.go b/runsc/main.go index c1c78529c..9d52f3006 100644 --- a/runsc/main.go +++ b/runsc/main.go @@ -84,6 +84,7 @@ var ( rootless = flag.Bool("rootless", false, "it allows the sandbox to be started with a user that is not root. Sandbox and Gofer processes may run with same privileges as current user.") referenceLeakMode = flag.String("ref-leak-mode", "disabled", "sets reference leak check mode: disabled (default), log-names, log-traces.") cpuNumFromQuota = flag.Bool("cpu-num-from-quota", false, "set cpu number to cpu quota (least integer greater or equal to quota value, but not less than 2)") + vfs2Enabled = flag.Bool("vfs2", false, "TEST ONLY; use while VFSv2 is landing. This uses the new experimental VFS layer.") // Test flags, not to be used outside tests, ever. testOnlyAllowRunAsCurrentUserWithoutChroot = flag.Bool("TESTONLY-unsafe-nonroot", false, "TEST ONLY; do not ever use! This skips many security measures that isolate the host from the sandbox.") @@ -230,6 +231,7 @@ func main() { ReferenceLeakMode: refsLeakMode, OverlayfsStaleRead: *overlayfsStaleRead, CPUNumFromQuota: *cpuNumFromQuota, + VFS2: *vfs2Enabled, TestOnlyAllowRunAsCurrentUserWithoutChroot: *testOnlyAllowRunAsCurrentUserWithoutChroot, TestOnlyTestNameEnv: *testOnlyTestNameEnv, @@ -313,6 +315,7 @@ func main() { log.Infof("\t\tFileAccess: %v, overlay: %t", conf.FileAccess, conf.Overlay) log.Infof("\t\tNetwork: %v, logging: %t", conf.Network, conf.LogPackets) log.Infof("\t\tStrace: %t, max size: %d, syscalls: %s", conf.Strace, conf.StraceLogSize, conf.StraceSyscalls) + log.Infof("\t\tVFS2 enabled: %v", conf.VFS2) log.Infof("***************************") if *testOnlyAllowRunAsCurrentUserWithoutChroot { -- cgit v1.2.3 From a551add5d8a5bf631cd9859c761e579fdb33ec82 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Mon, 13 Apr 2020 17:37:21 -0700 Subject: Remove View.First() and View.RemoveFirst() These methods let users eaily break the VectorisedView abstraction, and allowed netstack to slip into pseudo-enforcement of the "all headers are in the first View" invariant. Removing them and replacing with PullUp(n) breaks this reliance and will make it easier to add iptables support and rework network buffer management. The new View.PullUp(n) method is low cost in the common case, when when all the headers fit in the first View. --- pkg/sentry/socket/netfilter/tcp_matcher.go | 5 +- pkg/sentry/socket/netfilter/udp_matcher.go | 5 +- pkg/tcpip/buffer/view.go | 55 ++++++++++---- pkg/tcpip/buffer/view_test.go | 113 +++++++++++++++++++++++++++++ pkg/tcpip/link/loopback/loopback.go | 10 +-- pkg/tcpip/link/sharedmem/sharedmem_test.go | 2 +- pkg/tcpip/link/sniffer/sniffer.go | 65 +++++++++++++---- pkg/tcpip/network/arp/arp.go | 5 +- pkg/tcpip/network/ipv4/icmp.go | 20 +++-- pkg/tcpip/network/ipv4/ipv4.go | 12 ++- pkg/tcpip/network/ipv6/icmp.go | 74 ++++++++++++------- pkg/tcpip/network/ipv6/icmp_test.go | 3 +- pkg/tcpip/network/ipv6/ipv6.go | 6 +- pkg/tcpip/stack/forwarder_test.go | 13 ++-- pkg/tcpip/stack/iptables.go | 22 +++++- pkg/tcpip/stack/iptables_targets.go | 23 ++++-- pkg/tcpip/stack/nic.go | 34 +++------ pkg/tcpip/stack/packet_buffer.go | 4 +- pkg/tcpip/stack/stack_test.go | 10 ++- pkg/tcpip/stack/transport_test.go | 5 +- pkg/tcpip/transport/icmp/endpoint.go | 8 +- pkg/tcpip/transport/tcp/segment.go | 29 +++++--- pkg/tcpip/transport/tcp/tcp_test.go | 4 +- pkg/tcpip/transport/udp/endpoint.go | 6 +- pkg/tcpip/transport/udp/protocol.go | 9 ++- 25 files changed, 395 insertions(+), 147 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/tcp_matcher.go b/pkg/sentry/socket/netfilter/tcp_matcher.go index ff1cfd8f6..55c0f04f3 100644 --- a/pkg/sentry/socket/netfilter/tcp_matcher.go +++ b/pkg/sentry/socket/netfilter/tcp_matcher.go @@ -121,12 +121,13 @@ func (tm *TCPMatcher) Match(hook stack.Hook, pkt stack.PacketBuffer, interfaceNa tcpHeader = header.TCP(pkt.TransportHeader) } else { // The TCP header hasn't been parsed yet. We have to do it here. - if len(pkt.Data.First()) < header.TCPMinimumSize { + hdr, ok := pkt.Data.PullUp(header.TCPMinimumSize) + if !ok { // There's no valid TCP header here, so we hotdrop the // packet. return false, true } - tcpHeader = header.TCP(pkt.Data.First()) + tcpHeader = header.TCP(hdr) } // Check whether the source and destination ports are within the diff --git a/pkg/sentry/socket/netfilter/udp_matcher.go b/pkg/sentry/socket/netfilter/udp_matcher.go index 3359418c1..04d03d494 100644 --- a/pkg/sentry/socket/netfilter/udp_matcher.go +++ b/pkg/sentry/socket/netfilter/udp_matcher.go @@ -120,12 +120,13 @@ func (um *UDPMatcher) Match(hook stack.Hook, pkt stack.PacketBuffer, interfaceNa udpHeader = header.UDP(pkt.TransportHeader) } else { // The UDP header hasn't been parsed yet. We have to do it here. - if len(pkt.Data.First()) < header.UDPMinimumSize { + hdr, ok := pkt.Data.PullUp(header.UDPMinimumSize) + if !ok { // There's no valid UDP header here, so we hotdrop the // packet. return false, true } - udpHeader = header.UDP(pkt.Data.First()) + udpHeader = header.UDP(hdr) } // Check whether the source and destination ports are within the diff --git a/pkg/tcpip/buffer/view.go b/pkg/tcpip/buffer/view.go index 8ec5d5d5c..f01217c91 100644 --- a/pkg/tcpip/buffer/view.go +++ b/pkg/tcpip/buffer/view.go @@ -77,7 +77,8 @@ func NewVectorisedView(size int, views []View) VectorisedView { return VectorisedView{views: views, size: size} } -// TrimFront removes the first "count" bytes of the vectorised view. +// TrimFront removes the first "count" bytes of the vectorised view. It panics +// if count > vv.Size(). func (vv *VectorisedView) TrimFront(count int) { for count > 0 && len(vv.views) > 0 { if count < len(vv.views[0]) { @@ -86,7 +87,7 @@ func (vv *VectorisedView) TrimFront(count int) { return } count -= len(vv.views[0]) - vv.RemoveFirst() + vv.removeFirst() } } @@ -104,7 +105,7 @@ func (vv *VectorisedView) Read(v View) (copied int, err error) { count -= len(vv.views[0]) copy(v[copied:], vv.views[0]) copied += len(vv.views[0]) - vv.RemoveFirst() + vv.removeFirst() } if copied == 0 { return 0, io.EOF @@ -126,7 +127,7 @@ func (vv *VectorisedView) ReadToVV(dstVV *VectorisedView, count int) (copied int count -= len(vv.views[0]) dstVV.AppendView(vv.views[0]) copied += len(vv.views[0]) - vv.RemoveFirst() + vv.removeFirst() } return copied } @@ -162,22 +163,37 @@ func (vv *VectorisedView) Clone(buffer []View) VectorisedView { return VectorisedView{views: append(buffer[:0], vv.views...), size: vv.size} } -// First returns the first view of the vectorised view. -func (vv *VectorisedView) First() View { +// PullUp returns the first "count" bytes of the vectorised view. If those +// bytes aren't already contiguous inside the vectorised view, PullUp will +// reallocate as needed to make them contiguous. PullUp fails and returns false +// when count > vv.Size(). +func (vv *VectorisedView) PullUp(count int) (View, bool) { if len(vv.views) == 0 { - return nil + return nil, count == 0 + } + if count <= len(vv.views[0]) { + return vv.views[0][:count], true + } + if count > vv.size { + return nil, false } - return vv.views[0] -} -// RemoveFirst removes the first view of the vectorised view. -func (vv *VectorisedView) RemoveFirst() { - if len(vv.views) == 0 { - return + newFirst := NewView(count) + i := 0 + for offset := 0; offset < count; i++ { + copy(newFirst[offset:], vv.views[i]) + if count-offset < len(vv.views[i]) { + vv.views[i].TrimFront(count - offset) + break + } + offset += len(vv.views[i]) + vv.views[i] = nil } - vv.size -= len(vv.views[0]) - vv.views[0] = nil - vv.views = vv.views[1:] + // We're guaranteed that i > 0, since count is too large for the first + // view. + vv.views[i-1] = newFirst + vv.views = vv.views[i-1:] + return newFirst, true } // Size returns the size in bytes of the entire content stored in the vectorised view. @@ -225,3 +241,10 @@ func (vv *VectorisedView) Readers() []bytes.Reader { } return readers } + +// removeFirst panics when len(vv.views) < 1. +func (vv *VectorisedView) removeFirst() { + vv.size -= len(vv.views[0]) + vv.views[0] = nil + vv.views = vv.views[1:] +} diff --git a/pkg/tcpip/buffer/view_test.go b/pkg/tcpip/buffer/view_test.go index 106e1994c..c56795c7b 100644 --- a/pkg/tcpip/buffer/view_test.go +++ b/pkg/tcpip/buffer/view_test.go @@ -16,6 +16,7 @@ package buffer import ( + "bytes" "reflect" "testing" ) @@ -370,3 +371,115 @@ func TestVVRead(t *testing.T) { }) } } + +var pullUpTestCases = []struct { + comment string + in VectorisedView + count int + want []byte + result VectorisedView + ok bool +}{ + { + comment: "simple case", + in: vv(2, "12"), + count: 1, + want: []byte("1"), + result: vv(2, "12"), + ok: true, + }, + { + comment: "entire View", + in: vv(2, "1", "2"), + count: 1, + want: []byte("1"), + result: vv(2, "1", "2"), + ok: true, + }, + { + comment: "spanning across two Views", + in: vv(3, "1", "23"), + count: 2, + want: []byte("12"), + result: vv(3, "12", "3"), + ok: true, + }, + { + comment: "spanning across all Views", + in: vv(5, "1", "23", "45"), + count: 5, + want: []byte("12345"), + result: vv(5, "12345"), + ok: true, + }, + { + comment: "count = 0", + in: vv(1, "1"), + count: 0, + want: []byte{}, + result: vv(1, "1"), + ok: true, + }, + { + comment: "count = size", + in: vv(1, "1"), + count: 1, + want: []byte("1"), + result: vv(1, "1"), + ok: true, + }, + { + comment: "count too large", + in: vv(3, "1", "23"), + count: 4, + want: nil, + result: vv(3, "1", "23"), + ok: false, + }, + { + comment: "empty vv", + in: vv(0, ""), + count: 1, + want: nil, + result: vv(0, ""), + ok: false, + }, + { + comment: "empty vv, count = 0", + in: vv(0, ""), + count: 0, + want: nil, + result: vv(0, ""), + ok: true, + }, + { + comment: "empty views", + in: vv(3, "", "1", "", "23"), + count: 2, + want: []byte("12"), + result: vv(3, "12", "3"), + ok: true, + }, +} + +func TestPullUp(t *testing.T) { + for _, c := range pullUpTestCases { + got, ok := c.in.PullUp(c.count) + + // Is the return value right? + if ok != c.ok { + t.Errorf("Test %q failed when calling PullUp(%d) on %v. Got an ok of %t. Want %t", + c.comment, c.count, c.in, ok, c.ok) + } + if bytes.Compare(got, View(c.want)) != 0 { + t.Errorf("Test %q failed when calling PullUp(%d) on %v. Got %v. Want %v", + c.comment, c.count, c.in, got, c.want) + } + + // Is the underlying structure right? + if !reflect.DeepEqual(c.in, c.result) { + t.Errorf("Test %q failed when calling PullUp(%d). Got vv with structure %v. Wanted %v", + c.comment, c.count, c.in, c.result) + } + } +} diff --git a/pkg/tcpip/link/loopback/loopback.go b/pkg/tcpip/link/loopback/loopback.go index 1e2255bfa..073c84ef9 100644 --- a/pkg/tcpip/link/loopback/loopback.go +++ b/pkg/tcpip/link/loopback/loopback.go @@ -98,13 +98,13 @@ func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, stack.PacketBufferList // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { - // Reject the packet if it's shorter than an ethernet header. - if vv.Size() < header.EthernetMinimumSize { + // There should be an ethernet header at the beginning of vv. + hdr, ok := vv.PullUp(header.EthernetMinimumSize) + if !ok { + // Reject the packet if it's shorter than an ethernet header. return tcpip.ErrBadAddress } - - // There should be an ethernet header at the beginning of vv. - linkHeader := header.Ethernet(vv.First()[:header.EthernetMinimumSize]) + linkHeader := header.Ethernet(hdr) vv.TrimFront(len(linkHeader)) e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, linkHeader.Type(), stack.PacketBuffer{ Data: vv, diff --git a/pkg/tcpip/link/sharedmem/sharedmem_test.go b/pkg/tcpip/link/sharedmem/sharedmem_test.go index 27ea3f531..33f640b85 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem_test.go +++ b/pkg/tcpip/link/sharedmem/sharedmem_test.go @@ -674,7 +674,7 @@ func TestSimpleReceive(t *testing.T) { // Wait for packet to be received, then check it. c.waitForPackets(1, time.After(5*time.Second), "Timeout waiting for packet") c.mu.Lock() - rcvd := []byte(c.packets[0].vv.First()) + rcvd := []byte(c.packets[0].vv.ToView()) c.packets = c.packets[:0] c.mu.Unlock() diff --git a/pkg/tcpip/link/sniffer/sniffer.go b/pkg/tcpip/link/sniffer/sniffer.go index be2537a82..0799c8f4d 100644 --- a/pkg/tcpip/link/sniffer/sniffer.go +++ b/pkg/tcpip/link/sniffer/sniffer.go @@ -171,11 +171,7 @@ func (e *endpoint) GSOMaxSize() uint32 { func (e *endpoint) dumpPacket(prefix string, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { writer := e.writer if writer == nil && atomic.LoadUint32(&LogPackets) == 1 { - first := pkt.Header.View() - if len(first) == 0 { - first = pkt.Data.First() - } - logPacket(prefix, protocol, first, gso) + logPacket(prefix, protocol, pkt, gso) } if writer != nil && atomic.LoadUint32(&LogPacketsToPCAP) == 1 { totalLength := pkt.Header.UsedLength() + pkt.Data.Size() @@ -238,7 +234,7 @@ func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { // Wait implements stack.LinkEndpoint.Wait. func (e *endpoint) Wait() { e.lower.Wait() } -func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.View, gso *stack.GSO) { +func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer, gso *stack.GSO) { // Figure out the network layer info. var transProto uint8 src := tcpip.Address("unknown") @@ -247,28 +243,49 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie size := uint16(0) var fragmentOffset uint16 var moreFragments bool + + // Create a clone of pkt, including any headers if present. Avoid allocating + // backing memory for the clone. + views := [8]buffer.View{} + vv := buffer.NewVectorisedView(0, views[:0]) + vv.AppendView(pkt.Header.View()) + vv.Append(pkt.Data) + switch protocol { case header.IPv4ProtocolNumber: - ipv4 := header.IPv4(b) + hdr, ok := vv.PullUp(header.IPv4MinimumSize) + if !ok { + return + } + ipv4 := header.IPv4(hdr) fragmentOffset = ipv4.FragmentOffset() moreFragments = ipv4.Flags()&header.IPv4FlagMoreFragments == header.IPv4FlagMoreFragments src = ipv4.SourceAddress() dst = ipv4.DestinationAddress() transProto = ipv4.Protocol() size = ipv4.TotalLength() - uint16(ipv4.HeaderLength()) - b = b[ipv4.HeaderLength():] + vv.TrimFront(int(ipv4.HeaderLength())) id = int(ipv4.ID()) case header.IPv6ProtocolNumber: - ipv6 := header.IPv6(b) + hdr, ok := vv.PullUp(header.IPv6MinimumSize) + if !ok { + return + } + ipv6 := header.IPv6(hdr) src = ipv6.SourceAddress() dst = ipv6.DestinationAddress() transProto = ipv6.NextHeader() size = ipv6.PayloadLength() - b = b[header.IPv6MinimumSize:] + vv.TrimFront(header.IPv6MinimumSize) case header.ARPProtocolNumber: - arp := header.ARP(b) + hdr, ok := vv.PullUp(header.ARPSize) + if !ok { + return + } + vv.TrimFront(header.ARPSize) + arp := header.ARP(hdr) log.Infof( "%s arp %v (%v) -> %v (%v) valid:%v", prefix, @@ -284,7 +301,7 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie // We aren't guaranteed to have a transport header - it's possible for // writes via raw endpoints to contain only network headers. - if minSize, ok := transportProtocolMinSizes[tcpip.TransportProtocolNumber(transProto)]; ok && len(b) < minSize { + if minSize, ok := transportProtocolMinSizes[tcpip.TransportProtocolNumber(transProto)]; ok && vv.Size() < minSize { log.Infof("%s %v -> %v transport protocol: %d, but no transport header found (possible raw packet)", prefix, src, dst, transProto) return } @@ -297,7 +314,11 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie switch tcpip.TransportProtocolNumber(transProto) { case header.ICMPv4ProtocolNumber: transName = "icmp" - icmp := header.ICMPv4(b) + hdr, ok := vv.PullUp(header.ICMPv4MinimumSize) + if !ok { + break + } + icmp := header.ICMPv4(hdr) icmpType := "unknown" if fragmentOffset == 0 { switch icmp.Type() { @@ -330,7 +351,11 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie case header.ICMPv6ProtocolNumber: transName = "icmp" - icmp := header.ICMPv6(b) + hdr, ok := vv.PullUp(header.ICMPv6MinimumSize) + if !ok { + break + } + icmp := header.ICMPv6(hdr) icmpType := "unknown" switch icmp.Type() { case header.ICMPv6DstUnreachable: @@ -361,7 +386,11 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie case header.UDPProtocolNumber: transName = "udp" - udp := header.UDP(b) + hdr, ok := vv.PullUp(header.UDPMinimumSize) + if !ok { + break + } + udp := header.UDP(hdr) if fragmentOffset == 0 && len(udp) >= header.UDPMinimumSize { srcPort = udp.SourcePort() dstPort = udp.DestinationPort() @@ -371,7 +400,11 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie case header.TCPProtocolNumber: transName = "tcp" - tcp := header.TCP(b) + hdr, ok := vv.PullUp(header.TCPMinimumSize) + if !ok { + break + } + tcp := header.TCP(hdr) if fragmentOffset == 0 && len(tcp) >= header.TCPMinimumSize { offset := int(tcp.DataOffset()) if offset < header.TCPMinimumSize { diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index 7acbfa0a8..cf73a939e 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -93,7 +93,10 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuf } func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { - v := pkt.Data.First() + v, ok := pkt.Data.PullUp(header.ARPSize) + if !ok { + return + } h := header.ARP(v) if !h.IsValid() { return diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go index c4bf1ba5c..4cbefe5ab 100644 --- a/pkg/tcpip/network/ipv4/icmp.go +++ b/pkg/tcpip/network/ipv4/icmp.go @@ -25,7 +25,11 @@ import ( // used to find out which transport endpoint must be notified about the ICMP // packet. func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack.PacketBuffer) { - h := header.IPv4(pkt.Data.First()) + h, ok := pkt.Data.PullUp(header.IPv4MinimumSize) + if !ok { + return + } + hdr := header.IPv4(h) // We don't use IsValid() here because ICMP only requires that the IP // header plus 8 bytes of the transport header be included. So it's @@ -34,12 +38,12 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. // // Drop packet if it doesn't have the basic IPv4 header or if the // original source address doesn't match the endpoint's address. - if len(h) < header.IPv4MinimumSize || h.SourceAddress() != e.id.LocalAddress { + if hdr.SourceAddress() != e.id.LocalAddress { return } - hlen := int(h.HeaderLength()) - if pkt.Data.Size() < hlen || h.FragmentOffset() != 0 { + hlen := int(hdr.HeaderLength()) + if pkt.Data.Size() < hlen || hdr.FragmentOffset() != 0 { // We won't be able to handle this if it doesn't contain the // full IPv4 header, or if it's a fragment not at offset 0 // (because it won't have the transport header). @@ -48,15 +52,15 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. // Skip the ip header, then deliver control message. pkt.Data.TrimFront(hlen) - p := h.TransportProtocol() - e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, h.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) + p := hdr.TransportProtocol() + e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, hdr.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) } func (e *endpoint) handleICMP(r *stack.Route, pkt stack.PacketBuffer) { stats := r.Stats() received := stats.ICMP.V4PacketsReceived - v := pkt.Data.First() - if len(v) < header.ICMPv4MinimumSize { + v, ok := pkt.Data.PullUp(header.ICMPv4MinimumSize) + if !ok { received.Invalid.Increment() return } diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index 104aafbed..17202cc7a 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -328,7 +328,11 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuffer) *tcpip.Error { // The packet already has an IP header, but there are a few required // checks. - ip := header.IPv4(pkt.Data.First()) + h, ok := pkt.Data.PullUp(header.IPv4MinimumSize) + if !ok { + return tcpip.ErrInvalidOptionValue + } + ip := header.IPv4(h) if !ip.IsValid(pkt.Data.Size()) { return tcpip.ErrInvalidOptionValue } @@ -378,7 +382,11 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuf // HandlePacket is called by the link layer when new ipv4 packets arrive for // this endpoint. func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { - headerView := pkt.Data.First() + headerView, ok := pkt.Data.PullUp(header.IPv4MinimumSize) + if !ok { + r.Stats().IP.MalformedPacketsReceived.Increment() + return + } h := header.IPv4(headerView) if !h.IsValid(pkt.Data.Size()) { r.Stats().IP.MalformedPacketsReceived.Increment() diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index b68983d10..bdf3a0d25 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -28,7 +28,11 @@ import ( // used to find out which transport endpoint must be notified about the ICMP // packet. func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack.PacketBuffer) { - h := header.IPv6(pkt.Data.First()) + h, ok := pkt.Data.PullUp(header.IPv6MinimumSize) + if !ok { + return + } + hdr := header.IPv6(h) // We don't use IsValid() here because ICMP only requires that up to // 1280 bytes of the original packet be included. So it's likely that it @@ -36,17 +40,21 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. // // Drop packet if it doesn't have the basic IPv6 header or if the // original source address doesn't match the endpoint's address. - if len(h) < header.IPv6MinimumSize || h.SourceAddress() != e.id.LocalAddress { + if hdr.SourceAddress() != e.id.LocalAddress { return } // Skip the IP header, then handle the fragmentation header if there // is one. pkt.Data.TrimFront(header.IPv6MinimumSize) - p := h.TransportProtocol() + p := hdr.TransportProtocol() if p == header.IPv6FragmentHeader { - f := header.IPv6Fragment(pkt.Data.First()) - if !f.IsValid() || f.FragmentOffset() != 0 { + f, ok := pkt.Data.PullUp(header.IPv6FragmentHeaderSize) + if !ok { + return + } + fragHdr := header.IPv6Fragment(f) + if !fragHdr.IsValid() || fragHdr.FragmentOffset() != 0 { // We can't handle fragments that aren't at offset 0 // because they don't have the transport headers. return @@ -55,19 +63,19 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. // Skip fragmentation header and find out the actual protocol // number. pkt.Data.TrimFront(header.IPv6FragmentHeaderSize) - p = f.TransportProtocol() + p = fragHdr.TransportProtocol() } // Deliver the control packet to the transport endpoint. - e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, h.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) + e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, hdr.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) } func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.PacketBuffer, hasFragmentHeader bool) { stats := r.Stats().ICMP sent := stats.V6PacketsSent received := stats.V6PacketsReceived - v := pkt.Data.First() - if len(v) < header.ICMPv6MinimumSize { + v, ok := pkt.Data.PullUp(header.ICMPv6HeaderSize) + if !ok { received.Invalid.Increment() return } @@ -76,11 +84,9 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P // Validate ICMPv6 checksum before processing the packet. // - // Only the first view in vv is accounted for by h. To account for the - // rest of vv, a shallow copy is made and the first view is removed. // This copy is used as extra payload during the checksum calculation. payload := pkt.Data.Clone(nil) - payload.RemoveFirst() + payload.TrimFront(len(h)) if got, want := h.Checksum(), header.ICMPv6Checksum(h, iph.SourceAddress(), iph.DestinationAddress(), payload); got != want { received.Invalid.Increment() return @@ -101,34 +107,40 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P switch h.Type() { case header.ICMPv6PacketTooBig: received.PacketTooBig.Increment() - if len(v) < header.ICMPv6PacketTooBigMinimumSize { + hdr, ok := pkt.Data.PullUp(header.ICMPv6PacketTooBigMinimumSize) + if !ok { received.Invalid.Increment() return } pkt.Data.TrimFront(header.ICMPv6PacketTooBigMinimumSize) - mtu := h.MTU() + mtu := header.ICMPv6(hdr).MTU() e.handleControl(stack.ControlPacketTooBig, calculateMTU(mtu), pkt) case header.ICMPv6DstUnreachable: received.DstUnreachable.Increment() - if len(v) < header.ICMPv6DstUnreachableMinimumSize { + hdr, ok := pkt.Data.PullUp(header.ICMPv6DstUnreachableMinimumSize) + if !ok { received.Invalid.Increment() return } pkt.Data.TrimFront(header.ICMPv6DstUnreachableMinimumSize) - switch h.Code() { + switch header.ICMPv6(hdr).Code() { case header.ICMPv6PortUnreachable: e.handleControl(stack.ControlPortUnreachable, 0, pkt) } case header.ICMPv6NeighborSolicit: received.NeighborSolicit.Increment() - if len(v) < header.ICMPv6NeighborSolicitMinimumSize || !isNDPValid() { + if pkt.Data.Size() < header.ICMPv6NeighborSolicitMinimumSize || !isNDPValid() { received.Invalid.Increment() return } - ns := header.NDPNeighborSolicit(h.NDPPayload()) + // The remainder of payload must be only the neighbor solicitation, so + // payload.ToView() always returns the solicitation. Per RFC 6980 section 5, + // NDP messages cannot be fragmented. Also note that in the common case NDP + // datagrams are very small and ToView() will not incur allocations. + ns := header.NDPNeighborSolicit(payload.ToView()) it, err := ns.Options().Iter(true) if err != nil { // If we have a malformed NDP NS option, drop the packet. @@ -286,12 +298,16 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P case header.ICMPv6NeighborAdvert: received.NeighborAdvert.Increment() - if len(v) < header.ICMPv6NeighborAdvertSize || !isNDPValid() { + if pkt.Data.Size() < header.ICMPv6NeighborAdvertSize || !isNDPValid() { received.Invalid.Increment() return } - na := header.NDPNeighborAdvert(h.NDPPayload()) + // The remainder of payload must be only the neighbor advertisement, so + // payload.ToView() always returns the advertisement. Per RFC 6980 section + // 5, NDP messages cannot be fragmented. Also note that in the common case + // NDP datagrams are very small and ToView() will not incur allocations. + na := header.NDPNeighborAdvert(payload.ToView()) it, err := na.Options().Iter(true) if err != nil { // If we have a malformed NDP NA option, drop the packet. @@ -363,14 +379,15 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P case header.ICMPv6EchoRequest: received.EchoRequest.Increment() - if len(v) < header.ICMPv6EchoMinimumSize { + icmpHdr, ok := pkt.Data.PullUp(header.ICMPv6EchoMinimumSize) + if !ok { received.Invalid.Increment() return } pkt.Data.TrimFront(header.ICMPv6EchoMinimumSize) hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv6EchoMinimumSize) packet := header.ICMPv6(hdr.Prepend(header.ICMPv6EchoMinimumSize)) - copy(packet, h) + copy(packet, icmpHdr) packet.SetType(header.ICMPv6EchoReply) packet.SetChecksum(header.ICMPv6Checksum(packet, r.LocalAddress, r.RemoteAddress, pkt.Data)) if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, stack.PacketBuffer{ @@ -384,7 +401,7 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P case header.ICMPv6EchoReply: received.EchoReply.Increment() - if len(v) < header.ICMPv6EchoMinimumSize { + if pkt.Data.Size() < header.ICMPv6EchoMinimumSize { received.Invalid.Increment() return } @@ -406,8 +423,9 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P case header.ICMPv6RouterAdvert: received.RouterAdvert.Increment() - p := h.NDPPayload() - if len(p) < header.NDPRAMinimumSize || !isNDPValid() { + // Is the NDP payload of sufficient size to hold a Router + // Advertisement? + if pkt.Data.Size()-header.ICMPv6HeaderSize < header.NDPRAMinimumSize || !isNDPValid() { received.Invalid.Increment() return } @@ -425,7 +443,11 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P return } - ra := header.NDPRouterAdvert(p) + // The remainder of payload must be only the router advertisement, so + // payload.ToView() always returns the advertisement. Per RFC 6980 section + // 5, NDP messages cannot be fragmented. Also note that in the common case + // NDP datagrams are very small and ToView() will not incur allocations. + ra := header.NDPRouterAdvert(payload.ToView()) opts := ra.Options() // Are options valid as per the wire format? diff --git a/pkg/tcpip/network/ipv6/icmp_test.go b/pkg/tcpip/network/ipv6/icmp_test.go index bd099a7f8..d412ff688 100644 --- a/pkg/tcpip/network/ipv6/icmp_test.go +++ b/pkg/tcpip/network/ipv6/icmp_test.go @@ -166,7 +166,8 @@ func TestICMPCounts(t *testing.T) { }, { typ: header.ICMPv6NeighborSolicit, - size: header.ICMPv6NeighborSolicitMinimumSize}, + size: header.ICMPv6NeighborSolicitMinimumSize, + }, { typ: header.ICMPv6NeighborAdvert, size: header.ICMPv6NeighborAdvertMinimumSize, diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index 331b0817b..486725131 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -171,7 +171,11 @@ func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuffe // HandlePacket is called by the link layer when new ipv6 packets arrive for // this endpoint. func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { - headerView := pkt.Data.First() + headerView, ok := pkt.Data.PullUp(header.IPv6MinimumSize) + if !ok { + r.Stats().IP.MalformedPacketsReceived.Increment() + return + } h := header.IPv6(headerView) if !h.IsValid(pkt.Data.Size()) { r.Stats().IP.MalformedPacketsReceived.Increment() diff --git a/pkg/tcpip/stack/forwarder_test.go b/pkg/tcpip/stack/forwarder_test.go index e9c652042..c7c663498 100644 --- a/pkg/tcpip/stack/forwarder_test.go +++ b/pkg/tcpip/stack/forwarder_test.go @@ -70,7 +70,10 @@ func (f *fwdTestNetworkEndpoint) ID() *NetworkEndpointID { func (f *fwdTestNetworkEndpoint) HandlePacket(r *Route, pkt PacketBuffer) { // Consume the network header. - b := pkt.Data.First() + b, ok := pkt.Data.PullUp(fwdTestNetHeaderLen) + if !ok { + return + } pkt.Data.TrimFront(fwdTestNetHeaderLen) // Dispatch the packet to the transport protocol. @@ -473,7 +476,7 @@ func TestForwardingWithFakeResolverPartialTimeout(t *testing.T) { t.Fatal("packet not forwarded") } - b := p.Pkt.Header.View() + b := p.Pkt.Data.ToView() if b[0] != 3 { t.Fatalf("got b[0] = %d, want = 3", b[0]) } @@ -517,7 +520,7 @@ func TestForwardingWithFakeResolverTwoPackets(t *testing.T) { t.Fatal("packet not forwarded") } - b := p.Pkt.Header.View() + b := p.Pkt.Data.ToView() if b[0] != 3 { t.Fatalf("got b[0] = %d, want = 3", b[0]) } @@ -564,7 +567,7 @@ func TestForwardingWithFakeResolverManyPackets(t *testing.T) { t.Fatal("packet not forwarded") } - b := p.Pkt.Header.View() + b := p.Pkt.Data.ToView() if b[0] != 3 { t.Fatalf("got b[0] = %d, want = 3", b[0]) } @@ -619,7 +622,7 @@ func TestForwardingWithFakeResolverManyResolutions(t *testing.T) { // The first 5 packets (address 3 to 7) should not be forwarded // because their address resolutions are interrupted. - b := p.Pkt.Header.View() + b := p.Pkt.Data.ToView() if b[0] < 8 { t.Fatalf("got b[0] = %d, want b[0] >= 8", b[0]) } diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go index 6c0a4b24d..6b91159d4 100644 --- a/pkg/tcpip/stack/iptables.go +++ b/pkg/tcpip/stack/iptables.go @@ -212,6 +212,11 @@ func (it *IPTables) Check(hook Hook, pkt PacketBuffer) bool { // CheckPackets runs pkts through the rules for hook and returns a map of packets that // should not go forward. // +// Precondition: pkt is a IPv4 packet of at least length header.IPv4MinimumSize. +// +// TODO(gvisor.dev/issue/170): pk.NetworkHeader will always be set as a +// precondition. +// // NOTE: unlike the Check API the returned map contains packets that should be // dropped. func (it *IPTables) CheckPackets(hook Hook, pkts PacketBufferList) (drop map[*PacketBuffer]struct{}) { @@ -226,7 +231,9 @@ func (it *IPTables) CheckPackets(hook Hook, pkts PacketBufferList) (drop map[*Pa return drop } -// Precondition: pkt.NetworkHeader is set. +// Precondition: pkt is a IPv4 packet of at least length header.IPv4MinimumSize. +// TODO(gvisor.dev/issue/170): pk.NetworkHeader will always be set as a +// precondition. func (it *IPTables) checkChain(hook Hook, pkt PacketBuffer, table Table, ruleIdx int) chainVerdict { // Start from ruleIdx and walk the list of rules until a rule gives us // a verdict. @@ -271,14 +278,21 @@ func (it *IPTables) checkChain(hook Hook, pkt PacketBuffer, table Table, ruleIdx return chainDrop } -// Precondition: pk.NetworkHeader is set. +// Precondition: pkt is a IPv4 packet of at least length header.IPv4MinimumSize. +// TODO(gvisor.dev/issue/170): pk.NetworkHeader will always be set as a +// precondition. func (it *IPTables) checkRule(hook Hook, pkt PacketBuffer, table Table, ruleIdx int) (RuleVerdict, int) { rule := table.Rules[ruleIdx] // If pkt.NetworkHeader hasn't been set yet, it will be contained in - // pkt.Data.First(). + // pkt.Data. if pkt.NetworkHeader == nil { - pkt.NetworkHeader = pkt.Data.First() + var ok bool + pkt.NetworkHeader, ok = pkt.Data.PullUp(header.IPv4MinimumSize) + if !ok { + // Precondition has been violated. + panic(fmt.Sprintf("iptables checks require IPv4 headers of at least %d bytes", header.IPv4MinimumSize)) + } } // Check whether the packet matches the IP header filter. diff --git a/pkg/tcpip/stack/iptables_targets.go b/pkg/tcpip/stack/iptables_targets.go index 7b4543caf..8be61f4b1 100644 --- a/pkg/tcpip/stack/iptables_targets.go +++ b/pkg/tcpip/stack/iptables_targets.go @@ -96,9 +96,12 @@ func (rt RedirectTarget) Action(pkt PacketBuffer) (RuleVerdict, int) { newPkt := pkt.Clone() // Set network header. - headerView := newPkt.Data.First() + headerView, ok := newPkt.Data.PullUp(header.IPv4MinimumSize) + if !ok { + return RuleDrop, 0 + } netHeader := header.IPv4(headerView) - newPkt.NetworkHeader = headerView[:header.IPv4MinimumSize] + newPkt.NetworkHeader = headerView hlen := int(netHeader.HeaderLength()) tlen := int(netHeader.TotalLength()) @@ -117,10 +120,14 @@ func (rt RedirectTarget) Action(pkt PacketBuffer) (RuleVerdict, int) { if newPkt.TransportHeader != nil { udpHeader = header.UDP(newPkt.TransportHeader) } else { - if len(pkt.Data.First()) < header.UDPMinimumSize { + if pkt.Data.Size() < header.UDPMinimumSize { + return RuleDrop, 0 + } + hdr, ok := newPkt.Data.PullUp(header.UDPMinimumSize) + if !ok { return RuleDrop, 0 } - udpHeader = header.UDP(newPkt.Data.First()) + udpHeader = header.UDP(hdr) } udpHeader.SetDestinationPort(rt.MinPort) case header.TCPProtocolNumber: @@ -128,10 +135,14 @@ func (rt RedirectTarget) Action(pkt PacketBuffer) (RuleVerdict, int) { if newPkt.TransportHeader != nil { tcpHeader = header.TCP(newPkt.TransportHeader) } else { - if len(pkt.Data.First()) < header.TCPMinimumSize { + if pkt.Data.Size() < header.TCPMinimumSize { return RuleDrop, 0 } - tcpHeader = header.TCP(newPkt.TransportHeader) + hdr, ok := newPkt.Data.PullUp(header.TCPMinimumSize) + if !ok { + return RuleDrop, 0 + } + tcpHeader = header.TCP(hdr) } // TODO(gvisor.dev/issue/170): Need to recompute checksum // and implement nat connection tracking to support TCP. diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 016dbe15e..0c2b1f36a 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -1203,12 +1203,12 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link n.stack.stats.IP.PacketsReceived.Increment() } - if len(pkt.Data.First()) < netProto.MinimumPacketSize() { + netHeader, ok := pkt.Data.PullUp(netProto.MinimumPacketSize()) + if !ok { n.stack.stats.MalformedRcvdPackets.Increment() return } - - src, dst := netProto.ParseAddresses(pkt.Data.First()) + src, dst := netProto.ParseAddresses(netHeader) if n.stack.handleLocal && !n.isLoopback() && n.getRef(protocol, src) != nil { // The source address is one of our own, so we never should have gotten a @@ -1289,22 +1289,8 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link func (n *NIC) forwardPacket(r *Route, protocol tcpip.NetworkProtocolNumber, pkt PacketBuffer) { // TODO(b/143425874) Decrease the TTL field in forwarded packets. - - firstData := pkt.Data.First() - pkt.Data.RemoveFirst() - - if linkHeaderLen := int(n.linkEP.MaxHeaderLength()); linkHeaderLen == 0 { - pkt.Header = buffer.NewPrependableFromView(firstData) - } else { - firstDataLen := len(firstData) - - // pkt.Header should have enough capacity to hold n.linkEP's headers. - pkt.Header = buffer.NewPrependable(firstDataLen + linkHeaderLen) - - // TODO(b/151227689): avoid copying the packet when forwarding - if n := copy(pkt.Header.Prepend(firstDataLen), firstData); n != firstDataLen { - panic(fmt.Sprintf("copied %d bytes, expected %d", n, firstDataLen)) - } + if linkHeaderLen := int(n.linkEP.MaxHeaderLength()); linkHeaderLen != 0 { + pkt.Header = buffer.NewPrependable(linkHeaderLen) } if err := n.linkEP.WritePacket(r, nil /* gso */, protocol, pkt); err != nil { @@ -1332,12 +1318,13 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN // validly formed. n.stack.demux.deliverRawPacket(r, protocol, pkt) - if len(pkt.Data.First()) < transProto.MinimumPacketSize() { + transHeader, ok := pkt.Data.PullUp(transProto.MinimumPacketSize()) + if !ok { n.stack.stats.MalformedRcvdPackets.Increment() return } - srcPort, dstPort, err := transProto.ParsePorts(pkt.Data.First()) + srcPort, dstPort, err := transProto.ParsePorts(transHeader) if err != nil { n.stack.stats.MalformedRcvdPackets.Increment() return @@ -1375,11 +1362,12 @@ func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcp // ICMPv4 only guarantees that 8 bytes of the transport protocol will // be present in the payload. We know that the ports are within the // first 8 bytes for all known transport protocols. - if len(pkt.Data.First()) < 8 { + transHeader, ok := pkt.Data.PullUp(8) + if !ok { return } - srcPort, dstPort, err := transProto.ParsePorts(pkt.Data.First()) + srcPort, dstPort, err := transProto.ParsePorts(transHeader) if err != nil { return } diff --git a/pkg/tcpip/stack/packet_buffer.go b/pkg/tcpip/stack/packet_buffer.go index dc125f25e..e954a8b7e 100644 --- a/pkg/tcpip/stack/packet_buffer.go +++ b/pkg/tcpip/stack/packet_buffer.go @@ -37,7 +37,9 @@ type PacketBuffer struct { Data buffer.VectorisedView // Header holds the headers of outbound packets. As a packet is passed - // down the stack, each layer adds to Header. + // down the stack, each layer adds to Header. Note that forwarded + // packets don't populate Headers on their way out -- their headers and + // payload are never parsed out and remain in Data. Header buffer.Prependable // These fields are used by both inbound and outbound packets. They diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index c7634ceb1..d45d2cc1f 100644 --- a/pkg/tcpip/stack/stack_test.go +++ b/pkg/tcpip/stack/stack_test.go @@ -95,16 +95,18 @@ func (f *fakeNetworkEndpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffe f.proto.packetCount[int(f.id.LocalAddress[0])%len(f.proto.packetCount)]++ // Consume the network header. - b := pkt.Data.First() + b, ok := pkt.Data.PullUp(fakeNetHeaderLen) + if !ok { + return + } pkt.Data.TrimFront(fakeNetHeaderLen) // Handle control packets. if b[2] == uint8(fakeControlProtocol) { - nb := pkt.Data.First() - if len(nb) < fakeNetHeaderLen { + nb, ok := pkt.Data.PullUp(fakeNetHeaderLen) + if !ok { return } - pkt.Data.TrimFront(fakeNetHeaderLen) f.dispatcher.DeliverTransportControlPacket(tcpip.Address(nb[1:2]), tcpip.Address(nb[0:1]), fakeNetNumber, tcpip.TransportProtocolNumber(nb[2]), stack.ControlPortUnreachable, 0, pkt) return diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go index 3084e6593..a611e44ab 100644 --- a/pkg/tcpip/stack/transport_test.go +++ b/pkg/tcpip/stack/transport_test.go @@ -642,10 +642,11 @@ func TestTransportForwarding(t *testing.T) { t.Fatal("Response packet not forwarded") } - if dst := p.Pkt.Header.View()[0]; dst != 3 { + hdrs := p.Pkt.Data.ToView() + if dst := hdrs[0]; dst != 3 { t.Errorf("Response packet has incorrect destination addresss: got = %d, want = 3", dst) } - if src := p.Pkt.Header.View()[1]; src != 1 { + if src := hdrs[1]; src != 1 { t.Errorf("Response packet has incorrect source addresss: got = %d, want = 3", src) } } diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index feef8dca0..b1d820372 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -747,15 +747,15 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk // Only accept echo replies. switch e.NetProto { case header.IPv4ProtocolNumber: - h := header.ICMPv4(pkt.Data.First()) - if h.Type() != header.ICMPv4EchoReply { + h, ok := pkt.Data.PullUp(header.ICMPv4MinimumSize) + if !ok || header.ICMPv4(h).Type() != header.ICMPv4EchoReply { e.stack.Stats().DroppedPackets.Increment() e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() return } case header.IPv6ProtocolNumber: - h := header.ICMPv6(pkt.Data.First()) - if h.Type() != header.ICMPv6EchoReply { + h, ok := pkt.Data.PullUp(header.ICMPv6MinimumSize) + if !ok || header.ICMPv6(h).Type() != header.ICMPv6EchoReply { e.stack.Stats().DroppedPackets.Increment() e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() return diff --git a/pkg/tcpip/transport/tcp/segment.go b/pkg/tcpip/transport/tcp/segment.go index 40461fd31..7712ce652 100644 --- a/pkg/tcpip/transport/tcp/segment.go +++ b/pkg/tcpip/transport/tcp/segment.go @@ -144,7 +144,11 @@ func (s *segment) logicalLen() seqnum.Size { // TCP checksum and stores the checksum and result of checksum verification in // the csum and csumValid fields of the segment. func (s *segment) parse() bool { - h := header.TCP(s.data.First()) + h, ok := s.data.PullUp(header.TCPMinimumSize) + if !ok { + return false + } + hdr := header.TCP(h) // h is the header followed by the payload. We check that the offset to // the data respects the following constraints: @@ -156,12 +160,16 @@ func (s *segment) parse() bool { // N.B. The segment has already been validated as having at least the // minimum TCP size before reaching here, so it's safe to read the // fields. - offset := int(h.DataOffset()) - if offset < header.TCPMinimumSize || offset > len(h) { + offset := int(hdr.DataOffset()) + if offset < header.TCPMinimumSize { + return false + } + hdrWithOpts, ok := s.data.PullUp(offset) + if !ok { return false } - s.options = []byte(h[header.TCPMinimumSize:offset]) + s.options = []byte(hdrWithOpts[header.TCPMinimumSize:]) s.parsedOptions = header.ParseTCPOptions(s.options) // Query the link capabilities to decide if checksum validation is @@ -173,18 +181,19 @@ func (s *segment) parse() bool { s.data.TrimFront(offset) } if verifyChecksum { - s.csum = h.Checksum() + hdr = header.TCP(hdrWithOpts) + s.csum = hdr.Checksum() xsum := s.route.PseudoHeaderChecksum(ProtocolNumber, uint16(s.data.Size())) - xsum = h.CalculateChecksum(xsum) + xsum = hdr.CalculateChecksum(xsum) s.data.TrimFront(offset) xsum = header.ChecksumVV(s.data, xsum) s.csumValid = xsum == 0xffff } - s.sequenceNumber = seqnum.Value(h.SequenceNumber()) - s.ackNumber = seqnum.Value(h.AckNumber()) - s.flags = h.Flags() - s.window = seqnum.Size(h.WindowSize()) + s.sequenceNumber = seqnum.Value(hdr.SequenceNumber()) + s.ackNumber = seqnum.Value(hdr.AckNumber()) + s.flags = hdr.Flags() + s.window = seqnum.Size(hdr.WindowSize()) return true } diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index ab1014c7f..286c66cf5 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -3548,7 +3548,7 @@ func TestReceivedInvalidSegmentCountIncrement(t *testing.T) { AckNum: c.IRS.Add(1), RcvWnd: 30000, }) - tcpbuf := vv.First()[header.IPv4MinimumSize:] + tcpbuf := vv.ToView()[header.IPv4MinimumSize:] tcpbuf[header.TCPDataOffset] = ((header.TCPMinimumSize - 1) / 4) << 4 c.SendSegment(vv) @@ -3575,7 +3575,7 @@ func TestReceivedIncorrectChecksumIncrement(t *testing.T) { AckNum: c.IRS.Add(1), RcvWnd: 30000, }) - tcpbuf := vv.First()[header.IPv4MinimumSize:] + tcpbuf := vv.ToView()[header.IPv4MinimumSize:] // Overwrite a byte in the payload which should cause checksum // verification to fail. tcpbuf[(tcpbuf[header.TCPDataOffset]>>4)*4] = 0x4 diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index edb54f0be..756ab913a 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -1250,8 +1250,8 @@ func (e *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { // endpoint. func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) { // Get the header then trim it from the view. - hdr := header.UDP(pkt.Data.First()) - if int(hdr.Length()) > pkt.Data.Size() { + hdr, ok := pkt.Data.PullUp(header.UDPMinimumSize) + if !ok || int(header.UDP(hdr).Length()) > pkt.Data.Size() { // Malformed packet. e.stack.Stats().UDP.MalformedPacketsReceived.Increment() e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() @@ -1286,7 +1286,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk senderAddress: tcpip.FullAddress{ NIC: r.NICID(), Addr: id.RemoteAddress, - Port: hdr.SourcePort(), + Port: header.UDP(hdr).SourcePort(), }, } packet.data = pkt.Data diff --git a/pkg/tcpip/transport/udp/protocol.go b/pkg/tcpip/transport/udp/protocol.go index 6e31a9bac..52af6de22 100644 --- a/pkg/tcpip/transport/udp/protocol.go +++ b/pkg/tcpip/transport/udp/protocol.go @@ -68,8 +68,13 @@ func (*protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) { // that don't match any existing endpoint. func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) bool { // Get the header then trim it from the view. - hdr := header.UDP(pkt.Data.First()) - if int(hdr.Length()) > pkt.Data.Size() { + h, ok := pkt.Data.PullUp(header.UDPMinimumSize) + if !ok { + // Malformed packet. + r.Stack().Stats().UDP.MalformedPacketsReceived.Increment() + return true + } + if int(header.UDP(h).Length()) > pkt.Data.Size() { // Malformed packet. r.Stack().Stats().UDP.MalformedPacketsReceived.Increment() return true -- cgit v1.2.3 From e838290e671c9d72dbaa3aba13bf0c35f1147de4 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Fri, 17 Apr 2020 15:31:51 -0700 Subject: prlimit: don't check credentials on self prlimit was erroneously comparing UIDs and GIDs when getting/setting a process' own limits. From the manpage: To set or get the resources of a process other than itself, the caller must have the CAP_SYS_RESOURCE capability, or the real, effective, and saved set user IDs of the target process must match the real user ID of the caller and the real, effective, and saved set group IDs of the target process must match the real group ID of the caller. PiperOrigin-RevId: 307127266 --- pkg/sentry/syscalls/linux/sys_rlimit.go | 2 +- test/syscalls/linux/uidgid.cc | 12 ++++++++++++ 2 files changed, 13 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/sys_rlimit.go b/pkg/sentry/syscalls/linux/sys_rlimit.go index e08c333d6..d5d5b6959 100644 --- a/pkg/sentry/syscalls/linux/sys_rlimit.go +++ b/pkg/sentry/syscalls/linux/sys_rlimit.go @@ -197,7 +197,7 @@ func Prlimit64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys // saved set user IDs of the target process must match the real user ID of // the caller and the real, effective, and saved set group IDs of the // target process must match the real group ID of the caller." - if !t.HasCapabilityIn(linux.CAP_SYS_RESOURCE, t.PIDNamespace().UserNamespace()) { + if ot != t && !t.HasCapabilityIn(linux.CAP_SYS_RESOURCE, t.PIDNamespace().UserNamespace()) { cred, tcred := t.Credentials(), ot.Credentials() if cred.RealKUID != tcred.RealKUID || cred.RealKUID != tcred.EffectiveKUID || diff --git a/test/syscalls/linux/uidgid.cc b/test/syscalls/linux/uidgid.cc index 6218fbce1..ff66a79f4 100644 --- a/test/syscalls/linux/uidgid.cc +++ b/test/syscalls/linux/uidgid.cc @@ -14,6 +14,7 @@ #include #include +#include #include #include @@ -249,6 +250,17 @@ TEST(UidGidRootTest, Setgroups) { SyscallFailsWithErrno(EFAULT)); } +TEST(UidGidRootTest, Setuid_prlimit) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(IsRoot())); + + // Change our UID. + EXPECT_THAT(seteuid(65534), SyscallSucceeds()); + + // Despite the UID change, we should be able to get our own limits. + struct rlimit rl = {}; + ASSERT_THAT(prlimit(0, RLIMIT_NOFILE, NULL, &rl), SyscallSucceeds()); +} + } // namespace } // namespace testing -- cgit v1.2.3 From 9a233c94f1bd54c7c3c11f7166a09e2eacd179c5 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Fri, 17 Apr 2020 22:18:36 -0700 Subject: Fix watchdog skipStack: the meaning was reversed. PiperOrigin-RevId: 307166317 --- pkg/sentry/watchdog/watchdog.go | 13 ++++++------- 1 file changed, 6 insertions(+), 7 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/watchdog/watchdog.go b/pkg/sentry/watchdog/watchdog.go index f7d6009a0..fcc46420f 100644 --- a/pkg/sentry/watchdog/watchdog.go +++ b/pkg/sentry/watchdog/watchdog.go @@ -319,8 +319,8 @@ func (w *Watchdog) report(offenders map[*kernel.Task]*offender, newTaskFound boo // Dump stack only if a new task is detected or if it sometime has // passed since the last time a stack dump was generated. - skipStack := newTaskFound || time.Since(w.lastStackDump) >= stackDumpSameTaskPeriod - w.doAction(w.TaskTimeoutAction, skipStack, &buf) + showStack := newTaskFound || time.Since(w.lastStackDump) >= stackDumpSameTaskPeriod + w.doAction(w.TaskTimeoutAction, showStack, &buf) } func (w *Watchdog) reportStuckWatchdog() { @@ -329,16 +329,15 @@ func (w *Watchdog) reportStuckWatchdog() { w.doAction(w.TaskTimeoutAction, false, &buf) } -// doAction will take the given action. If the action is LogWarnind and -// skipStack is true, then the stack printing will be skipped. -func (w *Watchdog) doAction(action Action, skipStack bool, msg *bytes.Buffer) { +// doAction will take the given action. If the action is LogWarning and +// showStack is false, then the stack printing will be skipped. +func (w *Watchdog) doAction(action Action, showStack bool, msg *bytes.Buffer) { switch action { case LogWarning: - if skipStack { + if !showStack { msg.WriteString("\n...[stack dump skipped]...") log.Warningf(msg.String()) return - } log.TracebackAll(msg.String()) w.lastStackDump = time.Now() -- cgit v1.2.3 From 1a940f2b6c834693d9f85e3b648a3be3d986129d Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Mon, 20 Apr 2020 08:50:27 -0700 Subject: Resolve issue with file mode for host fds. Instead of plumbing error through kernfs.Inode.Mode, panic if err != nil. The errors that can result from an fstat syscall all indicate that something is fundamentally wrong, and panicking should be acceptable. PiperOrigin-RevId: 307406847 --- pkg/sentry/fsimpl/host/host.go | 34 +++------------------------------- 1 file changed, 3 insertions(+), 31 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 97fa7f7ab..fe14476f1 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -94,7 +94,6 @@ func ImportFD(mnt *vfs.Mount, hostFD int, isTTY bool) (*vfs.FileDescription, err isTTY: isTTY, canMap: canMap(uint32(fileType)), ino: fs.NextIno(), - mode: fileMode, // For simplicity, set offset to 0. Technically, we should use the existing // offset on the host if the file is seekable. offset: 0, @@ -149,20 +148,6 @@ type inode struct { // This field is initialized at creation time and is immutable. ino uint64 - // modeMu protects mode. - modeMu sync.Mutex - - // mode is a cached version of the file mode on the host. Note that it may - // become out of date if the mode is changed on the host, e.g. with chmod. - // - // Generally, it is better to retrieve the mode from the host through an - // fstat syscall. We only use this value in inode.Mode(), which cannot - // return an error, if the syscall to host fails. - // - // FIXME(b/152294168): Plumb error into Inode.Mode() return value so we - // can get rid of this. - mode linux.FileMode - // offsetMu protects offset. offsetMu sync.Mutex @@ -195,10 +180,11 @@ func (i *inode) CheckPermissions(ctx context.Context, creds *auth.Credentials, a // Mode implements kernfs.Inode. func (i *inode) Mode() linux.FileMode { mode, _, _, err := i.getPermissions() + // Retrieving the mode from the host fd using fstat(2) should not fail. + // If the syscall does not succeed, something is fundamentally wrong. if err != nil { - return i.mode + panic(fmt.Sprintf("failed to retrieve mode from host fd %d: %v", i.hostFD, err)) } - return linux.FileMode(mode) } @@ -208,11 +194,6 @@ func (i *inode) getPermissions() (linux.FileMode, auth.KUID, auth.KGID, error) { if err := syscall.Fstat(i.hostFD, &s); err != nil { return 0, 0, 0, err } - - // Update cached mode. - i.modeMu.Lock() - i.mode = linux.FileMode(s.Mode) - i.modeMu.Unlock() return linux.FileMode(s.Mode), auth.KUID(s.Uid), auth.KGID(s.Gid), nil } @@ -292,12 +273,6 @@ func (i *inode) Stat(_ *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, erro ls.Ino = i.ino } - // Update cached mode. - if (mask&linux.STATX_TYPE != 0) && (mask&linux.STATX_MODE != 0) { - i.modeMu.Lock() - i.mode = linux.FileMode(s.Mode) - i.modeMu.Unlock() - } return ls, nil } @@ -364,9 +339,6 @@ func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Cre if err := syscall.Fchmod(i.hostFD, uint32(s.Mode)); err != nil { return err } - i.modeMu.Lock() - i.mode = linux.FileMode(s.Mode) - i.modeMu.Unlock() } if m&linux.STATX_SIZE != 0 { if err := syscall.Ftruncate(i.hostFD, int64(s.Size)); err != nil { -- cgit v1.2.3 From e72ce8cce46ed48af306df50d252853c8790924d Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Mon, 20 Apr 2020 10:09:23 -0700 Subject: Change lingering uses of "memfs" in fsimpl/tmpfs to "tmpfs". PiperOrigin-RevId: 307422746 --- pkg/sentry/fsimpl/tmpfs/benchmark_test.go | 4 ++-- pkg/sentry/fsimpl/tmpfs/filesystem.go | 2 +- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 2 +- 3 files changed, 4 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go index 383133e44..651912169 100644 --- a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go +++ b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go @@ -168,7 +168,7 @@ func BenchmarkVFS1TmpfsStat(b *testing.B) { } } -func BenchmarkVFS2MemfsStat(b *testing.B) { +func BenchmarkVFS2TmpfsStat(b *testing.B) { for _, depth := range depths { b.Run(fmt.Sprintf("%d", depth), func(b *testing.B) { ctx := contexttest.Context(b) @@ -362,7 +362,7 @@ func BenchmarkVFS1TmpfsMountStat(b *testing.B) { } } -func BenchmarkVFS2MemfsMountStat(b *testing.B) { +func BenchmarkVFS2TmpfsMountStat(b *testing.B) { for _, depth := range depths { b.Run(fmt.Sprintf("%d", depth), func(b *testing.B) { ctx := contexttest.Context(b) diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 660f5a29b..452c4e2e0 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -148,7 +148,7 @@ func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(pa if !dir && rp.MustBeDir() { return syserror.ENOENT } - // In memfs, the only way to cause a dentry to be disowned is by removing + // In tmpfs, the only way to cause a dentry to be disowned is by removing // it from the filesystem, so this check is equivalent to checking if // parent has been removed. if parent.vfsd.IsDisowned() { diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index a59b24d45..82c709b43 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -247,7 +247,7 @@ func (i *inode) incLinksLocked() { panic("tmpfs.inode.incLinksLocked() called with no existing links") } if i.nlink == maxLinks { - panic("memfs.inode.incLinksLocked() called with maximum link count") + panic("tmpfs.inode.incLinksLocked() called with maximum link count") } atomic.AddUint32(&i.nlink, 1) } -- cgit v1.2.3 From 120d3b50f4875824ec69f0cc39a09ac84fced35c Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Tue, 21 Apr 2020 07:15:25 -0700 Subject: Automated rollback of changelist 307477185 PiperOrigin-RevId: 307598974 --- pkg/sentry/socket/netfilter/tcp_matcher.go | 5 +- pkg/sentry/socket/netfilter/udp_matcher.go | 5 +- pkg/tcpip/buffer/view.go | 55 ++++---------- pkg/tcpip/buffer/view_test.go | 113 ----------------------------- pkg/tcpip/link/loopback/loopback.go | 10 +-- pkg/tcpip/link/sharedmem/sharedmem_test.go | 2 +- pkg/tcpip/link/sniffer/sniffer.go | 65 ++++------------- pkg/tcpip/network/arp/arp.go | 5 +- pkg/tcpip/network/ipv4/icmp.go | 20 ++--- pkg/tcpip/network/ipv4/ipv4.go | 12 +-- pkg/tcpip/network/ipv6/icmp.go | 74 +++++++------------ pkg/tcpip/network/ipv6/icmp_test.go | 3 +- pkg/tcpip/network/ipv6/ipv6.go | 6 +- pkg/tcpip/stack/forwarder_test.go | 13 ++-- pkg/tcpip/stack/iptables.go | 22 +----- pkg/tcpip/stack/iptables_targets.go | 23 ++---- pkg/tcpip/stack/nic.go | 34 ++++++--- pkg/tcpip/stack/packet_buffer.go | 4 +- pkg/tcpip/stack/stack_test.go | 10 +-- pkg/tcpip/stack/transport_test.go | 5 +- pkg/tcpip/transport/icmp/endpoint.go | 8 +- pkg/tcpip/transport/tcp/segment.go | 29 +++----- pkg/tcpip/transport/tcp/tcp_test.go | 4 +- pkg/tcpip/transport/udp/endpoint.go | 6 +- pkg/tcpip/transport/udp/protocol.go | 9 +-- 25 files changed, 147 insertions(+), 395 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/tcp_matcher.go b/pkg/sentry/socket/netfilter/tcp_matcher.go index 55c0f04f3..ff1cfd8f6 100644 --- a/pkg/sentry/socket/netfilter/tcp_matcher.go +++ b/pkg/sentry/socket/netfilter/tcp_matcher.go @@ -121,13 +121,12 @@ func (tm *TCPMatcher) Match(hook stack.Hook, pkt stack.PacketBuffer, interfaceNa tcpHeader = header.TCP(pkt.TransportHeader) } else { // The TCP header hasn't been parsed yet. We have to do it here. - hdr, ok := pkt.Data.PullUp(header.TCPMinimumSize) - if !ok { + if len(pkt.Data.First()) < header.TCPMinimumSize { // There's no valid TCP header here, so we hotdrop the // packet. return false, true } - tcpHeader = header.TCP(hdr) + tcpHeader = header.TCP(pkt.Data.First()) } // Check whether the source and destination ports are within the diff --git a/pkg/sentry/socket/netfilter/udp_matcher.go b/pkg/sentry/socket/netfilter/udp_matcher.go index 04d03d494..3359418c1 100644 --- a/pkg/sentry/socket/netfilter/udp_matcher.go +++ b/pkg/sentry/socket/netfilter/udp_matcher.go @@ -120,13 +120,12 @@ func (um *UDPMatcher) Match(hook stack.Hook, pkt stack.PacketBuffer, interfaceNa udpHeader = header.UDP(pkt.TransportHeader) } else { // The UDP header hasn't been parsed yet. We have to do it here. - hdr, ok := pkt.Data.PullUp(header.UDPMinimumSize) - if !ok { + if len(pkt.Data.First()) < header.UDPMinimumSize { // There's no valid UDP header here, so we hotdrop the // packet. return false, true } - udpHeader = header.UDP(hdr) + udpHeader = header.UDP(pkt.Data.First()) } // Check whether the source and destination ports are within the diff --git a/pkg/tcpip/buffer/view.go b/pkg/tcpip/buffer/view.go index f01217c91..8ec5d5d5c 100644 --- a/pkg/tcpip/buffer/view.go +++ b/pkg/tcpip/buffer/view.go @@ -77,8 +77,7 @@ func NewVectorisedView(size int, views []View) VectorisedView { return VectorisedView{views: views, size: size} } -// TrimFront removes the first "count" bytes of the vectorised view. It panics -// if count > vv.Size(). +// TrimFront removes the first "count" bytes of the vectorised view. func (vv *VectorisedView) TrimFront(count int) { for count > 0 && len(vv.views) > 0 { if count < len(vv.views[0]) { @@ -87,7 +86,7 @@ func (vv *VectorisedView) TrimFront(count int) { return } count -= len(vv.views[0]) - vv.removeFirst() + vv.RemoveFirst() } } @@ -105,7 +104,7 @@ func (vv *VectorisedView) Read(v View) (copied int, err error) { count -= len(vv.views[0]) copy(v[copied:], vv.views[0]) copied += len(vv.views[0]) - vv.removeFirst() + vv.RemoveFirst() } if copied == 0 { return 0, io.EOF @@ -127,7 +126,7 @@ func (vv *VectorisedView) ReadToVV(dstVV *VectorisedView, count int) (copied int count -= len(vv.views[0]) dstVV.AppendView(vv.views[0]) copied += len(vv.views[0]) - vv.removeFirst() + vv.RemoveFirst() } return copied } @@ -163,37 +162,22 @@ func (vv *VectorisedView) Clone(buffer []View) VectorisedView { return VectorisedView{views: append(buffer[:0], vv.views...), size: vv.size} } -// PullUp returns the first "count" bytes of the vectorised view. If those -// bytes aren't already contiguous inside the vectorised view, PullUp will -// reallocate as needed to make them contiguous. PullUp fails and returns false -// when count > vv.Size(). -func (vv *VectorisedView) PullUp(count int) (View, bool) { +// First returns the first view of the vectorised view. +func (vv *VectorisedView) First() View { if len(vv.views) == 0 { - return nil, count == 0 - } - if count <= len(vv.views[0]) { - return vv.views[0][:count], true - } - if count > vv.size { - return nil, false + return nil } + return vv.views[0] +} - newFirst := NewView(count) - i := 0 - for offset := 0; offset < count; i++ { - copy(newFirst[offset:], vv.views[i]) - if count-offset < len(vv.views[i]) { - vv.views[i].TrimFront(count - offset) - break - } - offset += len(vv.views[i]) - vv.views[i] = nil +// RemoveFirst removes the first view of the vectorised view. +func (vv *VectorisedView) RemoveFirst() { + if len(vv.views) == 0 { + return } - // We're guaranteed that i > 0, since count is too large for the first - // view. - vv.views[i-1] = newFirst - vv.views = vv.views[i-1:] - return newFirst, true + vv.size -= len(vv.views[0]) + vv.views[0] = nil + vv.views = vv.views[1:] } // Size returns the size in bytes of the entire content stored in the vectorised view. @@ -241,10 +225,3 @@ func (vv *VectorisedView) Readers() []bytes.Reader { } return readers } - -// removeFirst panics when len(vv.views) < 1. -func (vv *VectorisedView) removeFirst() { - vv.size -= len(vv.views[0]) - vv.views[0] = nil - vv.views = vv.views[1:] -} diff --git a/pkg/tcpip/buffer/view_test.go b/pkg/tcpip/buffer/view_test.go index c56795c7b..106e1994c 100644 --- a/pkg/tcpip/buffer/view_test.go +++ b/pkg/tcpip/buffer/view_test.go @@ -16,7 +16,6 @@ package buffer import ( - "bytes" "reflect" "testing" ) @@ -371,115 +370,3 @@ func TestVVRead(t *testing.T) { }) } } - -var pullUpTestCases = []struct { - comment string - in VectorisedView - count int - want []byte - result VectorisedView - ok bool -}{ - { - comment: "simple case", - in: vv(2, "12"), - count: 1, - want: []byte("1"), - result: vv(2, "12"), - ok: true, - }, - { - comment: "entire View", - in: vv(2, "1", "2"), - count: 1, - want: []byte("1"), - result: vv(2, "1", "2"), - ok: true, - }, - { - comment: "spanning across two Views", - in: vv(3, "1", "23"), - count: 2, - want: []byte("12"), - result: vv(3, "12", "3"), - ok: true, - }, - { - comment: "spanning across all Views", - in: vv(5, "1", "23", "45"), - count: 5, - want: []byte("12345"), - result: vv(5, "12345"), - ok: true, - }, - { - comment: "count = 0", - in: vv(1, "1"), - count: 0, - want: []byte{}, - result: vv(1, "1"), - ok: true, - }, - { - comment: "count = size", - in: vv(1, "1"), - count: 1, - want: []byte("1"), - result: vv(1, "1"), - ok: true, - }, - { - comment: "count too large", - in: vv(3, "1", "23"), - count: 4, - want: nil, - result: vv(3, "1", "23"), - ok: false, - }, - { - comment: "empty vv", - in: vv(0, ""), - count: 1, - want: nil, - result: vv(0, ""), - ok: false, - }, - { - comment: "empty vv, count = 0", - in: vv(0, ""), - count: 0, - want: nil, - result: vv(0, ""), - ok: true, - }, - { - comment: "empty views", - in: vv(3, "", "1", "", "23"), - count: 2, - want: []byte("12"), - result: vv(3, "12", "3"), - ok: true, - }, -} - -func TestPullUp(t *testing.T) { - for _, c := range pullUpTestCases { - got, ok := c.in.PullUp(c.count) - - // Is the return value right? - if ok != c.ok { - t.Errorf("Test %q failed when calling PullUp(%d) on %v. Got an ok of %t. Want %t", - c.comment, c.count, c.in, ok, c.ok) - } - if bytes.Compare(got, View(c.want)) != 0 { - t.Errorf("Test %q failed when calling PullUp(%d) on %v. Got %v. Want %v", - c.comment, c.count, c.in, got, c.want) - } - - // Is the underlying structure right? - if !reflect.DeepEqual(c.in, c.result) { - t.Errorf("Test %q failed when calling PullUp(%d). Got vv with structure %v. Wanted %v", - c.comment, c.count, c.in, c.result) - } - } -} diff --git a/pkg/tcpip/link/loopback/loopback.go b/pkg/tcpip/link/loopback/loopback.go index 073c84ef9..1e2255bfa 100644 --- a/pkg/tcpip/link/loopback/loopback.go +++ b/pkg/tcpip/link/loopback/loopback.go @@ -98,13 +98,13 @@ func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, stack.PacketBufferList // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { - // There should be an ethernet header at the beginning of vv. - hdr, ok := vv.PullUp(header.EthernetMinimumSize) - if !ok { - // Reject the packet if it's shorter than an ethernet header. + // Reject the packet if it's shorter than an ethernet header. + if vv.Size() < header.EthernetMinimumSize { return tcpip.ErrBadAddress } - linkHeader := header.Ethernet(hdr) + + // There should be an ethernet header at the beginning of vv. + linkHeader := header.Ethernet(vv.First()[:header.EthernetMinimumSize]) vv.TrimFront(len(linkHeader)) e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, linkHeader.Type(), stack.PacketBuffer{ Data: vv, diff --git a/pkg/tcpip/link/sharedmem/sharedmem_test.go b/pkg/tcpip/link/sharedmem/sharedmem_test.go index 33f640b85..27ea3f531 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem_test.go +++ b/pkg/tcpip/link/sharedmem/sharedmem_test.go @@ -674,7 +674,7 @@ func TestSimpleReceive(t *testing.T) { // Wait for packet to be received, then check it. c.waitForPackets(1, time.After(5*time.Second), "Timeout waiting for packet") c.mu.Lock() - rcvd := []byte(c.packets[0].vv.ToView()) + rcvd := []byte(c.packets[0].vv.First()) c.packets = c.packets[:0] c.mu.Unlock() diff --git a/pkg/tcpip/link/sniffer/sniffer.go b/pkg/tcpip/link/sniffer/sniffer.go index 0799c8f4d..be2537a82 100644 --- a/pkg/tcpip/link/sniffer/sniffer.go +++ b/pkg/tcpip/link/sniffer/sniffer.go @@ -171,7 +171,11 @@ func (e *endpoint) GSOMaxSize() uint32 { func (e *endpoint) dumpPacket(prefix string, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { writer := e.writer if writer == nil && atomic.LoadUint32(&LogPackets) == 1 { - logPacket(prefix, protocol, pkt, gso) + first := pkt.Header.View() + if len(first) == 0 { + first = pkt.Data.First() + } + logPacket(prefix, protocol, first, gso) } if writer != nil && atomic.LoadUint32(&LogPacketsToPCAP) == 1 { totalLength := pkt.Header.UsedLength() + pkt.Data.Size() @@ -234,7 +238,7 @@ func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { // Wait implements stack.LinkEndpoint.Wait. func (e *endpoint) Wait() { e.lower.Wait() } -func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer, gso *stack.GSO) { +func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.View, gso *stack.GSO) { // Figure out the network layer info. var transProto uint8 src := tcpip.Address("unknown") @@ -243,49 +247,28 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.P size := uint16(0) var fragmentOffset uint16 var moreFragments bool - - // Create a clone of pkt, including any headers if present. Avoid allocating - // backing memory for the clone. - views := [8]buffer.View{} - vv := buffer.NewVectorisedView(0, views[:0]) - vv.AppendView(pkt.Header.View()) - vv.Append(pkt.Data) - switch protocol { case header.IPv4ProtocolNumber: - hdr, ok := vv.PullUp(header.IPv4MinimumSize) - if !ok { - return - } - ipv4 := header.IPv4(hdr) + ipv4 := header.IPv4(b) fragmentOffset = ipv4.FragmentOffset() moreFragments = ipv4.Flags()&header.IPv4FlagMoreFragments == header.IPv4FlagMoreFragments src = ipv4.SourceAddress() dst = ipv4.DestinationAddress() transProto = ipv4.Protocol() size = ipv4.TotalLength() - uint16(ipv4.HeaderLength()) - vv.TrimFront(int(ipv4.HeaderLength())) + b = b[ipv4.HeaderLength():] id = int(ipv4.ID()) case header.IPv6ProtocolNumber: - hdr, ok := vv.PullUp(header.IPv6MinimumSize) - if !ok { - return - } - ipv6 := header.IPv6(hdr) + ipv6 := header.IPv6(b) src = ipv6.SourceAddress() dst = ipv6.DestinationAddress() transProto = ipv6.NextHeader() size = ipv6.PayloadLength() - vv.TrimFront(header.IPv6MinimumSize) + b = b[header.IPv6MinimumSize:] case header.ARPProtocolNumber: - hdr, ok := vv.PullUp(header.ARPSize) - if !ok { - return - } - vv.TrimFront(header.ARPSize) - arp := header.ARP(hdr) + arp := header.ARP(b) log.Infof( "%s arp %v (%v) -> %v (%v) valid:%v", prefix, @@ -301,7 +284,7 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.P // We aren't guaranteed to have a transport header - it's possible for // writes via raw endpoints to contain only network headers. - if minSize, ok := transportProtocolMinSizes[tcpip.TransportProtocolNumber(transProto)]; ok && vv.Size() < minSize { + if minSize, ok := transportProtocolMinSizes[tcpip.TransportProtocolNumber(transProto)]; ok && len(b) < minSize { log.Infof("%s %v -> %v transport protocol: %d, but no transport header found (possible raw packet)", prefix, src, dst, transProto) return } @@ -314,11 +297,7 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.P switch tcpip.TransportProtocolNumber(transProto) { case header.ICMPv4ProtocolNumber: transName = "icmp" - hdr, ok := vv.PullUp(header.ICMPv4MinimumSize) - if !ok { - break - } - icmp := header.ICMPv4(hdr) + icmp := header.ICMPv4(b) icmpType := "unknown" if fragmentOffset == 0 { switch icmp.Type() { @@ -351,11 +330,7 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.P case header.ICMPv6ProtocolNumber: transName = "icmp" - hdr, ok := vv.PullUp(header.ICMPv6MinimumSize) - if !ok { - break - } - icmp := header.ICMPv6(hdr) + icmp := header.ICMPv6(b) icmpType := "unknown" switch icmp.Type() { case header.ICMPv6DstUnreachable: @@ -386,11 +361,7 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.P case header.UDPProtocolNumber: transName = "udp" - hdr, ok := vv.PullUp(header.UDPMinimumSize) - if !ok { - break - } - udp := header.UDP(hdr) + udp := header.UDP(b) if fragmentOffset == 0 && len(udp) >= header.UDPMinimumSize { srcPort = udp.SourcePort() dstPort = udp.DestinationPort() @@ -400,11 +371,7 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.P case header.TCPProtocolNumber: transName = "tcp" - hdr, ok := vv.PullUp(header.TCPMinimumSize) - if !ok { - break - } - tcp := header.TCP(hdr) + tcp := header.TCP(b) if fragmentOffset == 0 && len(tcp) >= header.TCPMinimumSize { offset := int(tcp.DataOffset()) if offset < header.TCPMinimumSize { diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index cf73a939e..7acbfa0a8 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -93,10 +93,7 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuf } func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { - v, ok := pkt.Data.PullUp(header.ARPSize) - if !ok { - return - } + v := pkt.Data.First() h := header.ARP(v) if !h.IsValid() { return diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go index 4cbefe5ab..c4bf1ba5c 100644 --- a/pkg/tcpip/network/ipv4/icmp.go +++ b/pkg/tcpip/network/ipv4/icmp.go @@ -25,11 +25,7 @@ import ( // used to find out which transport endpoint must be notified about the ICMP // packet. func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack.PacketBuffer) { - h, ok := pkt.Data.PullUp(header.IPv4MinimumSize) - if !ok { - return - } - hdr := header.IPv4(h) + h := header.IPv4(pkt.Data.First()) // We don't use IsValid() here because ICMP only requires that the IP // header plus 8 bytes of the transport header be included. So it's @@ -38,12 +34,12 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. // // Drop packet if it doesn't have the basic IPv4 header or if the // original source address doesn't match the endpoint's address. - if hdr.SourceAddress() != e.id.LocalAddress { + if len(h) < header.IPv4MinimumSize || h.SourceAddress() != e.id.LocalAddress { return } - hlen := int(hdr.HeaderLength()) - if pkt.Data.Size() < hlen || hdr.FragmentOffset() != 0 { + hlen := int(h.HeaderLength()) + if pkt.Data.Size() < hlen || h.FragmentOffset() != 0 { // We won't be able to handle this if it doesn't contain the // full IPv4 header, or if it's a fragment not at offset 0 // (because it won't have the transport header). @@ -52,15 +48,15 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. // Skip the ip header, then deliver control message. pkt.Data.TrimFront(hlen) - p := hdr.TransportProtocol() - e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, hdr.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) + p := h.TransportProtocol() + e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, h.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) } func (e *endpoint) handleICMP(r *stack.Route, pkt stack.PacketBuffer) { stats := r.Stats() received := stats.ICMP.V4PacketsReceived - v, ok := pkt.Data.PullUp(header.ICMPv4MinimumSize) - if !ok { + v := pkt.Data.First() + if len(v) < header.ICMPv4MinimumSize { received.Invalid.Increment() return } diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index 17202cc7a..104aafbed 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -328,11 +328,7 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuffer) *tcpip.Error { // The packet already has an IP header, but there are a few required // checks. - h, ok := pkt.Data.PullUp(header.IPv4MinimumSize) - if !ok { - return tcpip.ErrInvalidOptionValue - } - ip := header.IPv4(h) + ip := header.IPv4(pkt.Data.First()) if !ip.IsValid(pkt.Data.Size()) { return tcpip.ErrInvalidOptionValue } @@ -382,11 +378,7 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuf // HandlePacket is called by the link layer when new ipv4 packets arrive for // this endpoint. func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { - headerView, ok := pkt.Data.PullUp(header.IPv4MinimumSize) - if !ok { - r.Stats().IP.MalformedPacketsReceived.Increment() - return - } + headerView := pkt.Data.First() h := header.IPv4(headerView) if !h.IsValid(pkt.Data.Size()) { r.Stats().IP.MalformedPacketsReceived.Increment() diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index bdf3a0d25..b68983d10 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -28,11 +28,7 @@ import ( // used to find out which transport endpoint must be notified about the ICMP // packet. func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack.PacketBuffer) { - h, ok := pkt.Data.PullUp(header.IPv6MinimumSize) - if !ok { - return - } - hdr := header.IPv6(h) + h := header.IPv6(pkt.Data.First()) // We don't use IsValid() here because ICMP only requires that up to // 1280 bytes of the original packet be included. So it's likely that it @@ -40,21 +36,17 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. // // Drop packet if it doesn't have the basic IPv6 header or if the // original source address doesn't match the endpoint's address. - if hdr.SourceAddress() != e.id.LocalAddress { + if len(h) < header.IPv6MinimumSize || h.SourceAddress() != e.id.LocalAddress { return } // Skip the IP header, then handle the fragmentation header if there // is one. pkt.Data.TrimFront(header.IPv6MinimumSize) - p := hdr.TransportProtocol() + p := h.TransportProtocol() if p == header.IPv6FragmentHeader { - f, ok := pkt.Data.PullUp(header.IPv6FragmentHeaderSize) - if !ok { - return - } - fragHdr := header.IPv6Fragment(f) - if !fragHdr.IsValid() || fragHdr.FragmentOffset() != 0 { + f := header.IPv6Fragment(pkt.Data.First()) + if !f.IsValid() || f.FragmentOffset() != 0 { // We can't handle fragments that aren't at offset 0 // because they don't have the transport headers. return @@ -63,19 +55,19 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. // Skip fragmentation header and find out the actual protocol // number. pkt.Data.TrimFront(header.IPv6FragmentHeaderSize) - p = fragHdr.TransportProtocol() + p = f.TransportProtocol() } // Deliver the control packet to the transport endpoint. - e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, hdr.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) + e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, h.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) } func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.PacketBuffer, hasFragmentHeader bool) { stats := r.Stats().ICMP sent := stats.V6PacketsSent received := stats.V6PacketsReceived - v, ok := pkt.Data.PullUp(header.ICMPv6HeaderSize) - if !ok { + v := pkt.Data.First() + if len(v) < header.ICMPv6MinimumSize { received.Invalid.Increment() return } @@ -84,9 +76,11 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P // Validate ICMPv6 checksum before processing the packet. // + // Only the first view in vv is accounted for by h. To account for the + // rest of vv, a shallow copy is made and the first view is removed. // This copy is used as extra payload during the checksum calculation. payload := pkt.Data.Clone(nil) - payload.TrimFront(len(h)) + payload.RemoveFirst() if got, want := h.Checksum(), header.ICMPv6Checksum(h, iph.SourceAddress(), iph.DestinationAddress(), payload); got != want { received.Invalid.Increment() return @@ -107,40 +101,34 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P switch h.Type() { case header.ICMPv6PacketTooBig: received.PacketTooBig.Increment() - hdr, ok := pkt.Data.PullUp(header.ICMPv6PacketTooBigMinimumSize) - if !ok { + if len(v) < header.ICMPv6PacketTooBigMinimumSize { received.Invalid.Increment() return } pkt.Data.TrimFront(header.ICMPv6PacketTooBigMinimumSize) - mtu := header.ICMPv6(hdr).MTU() + mtu := h.MTU() e.handleControl(stack.ControlPacketTooBig, calculateMTU(mtu), pkt) case header.ICMPv6DstUnreachable: received.DstUnreachable.Increment() - hdr, ok := pkt.Data.PullUp(header.ICMPv6DstUnreachableMinimumSize) - if !ok { + if len(v) < header.ICMPv6DstUnreachableMinimumSize { received.Invalid.Increment() return } pkt.Data.TrimFront(header.ICMPv6DstUnreachableMinimumSize) - switch header.ICMPv6(hdr).Code() { + switch h.Code() { case header.ICMPv6PortUnreachable: e.handleControl(stack.ControlPortUnreachable, 0, pkt) } case header.ICMPv6NeighborSolicit: received.NeighborSolicit.Increment() - if pkt.Data.Size() < header.ICMPv6NeighborSolicitMinimumSize || !isNDPValid() { + if len(v) < header.ICMPv6NeighborSolicitMinimumSize || !isNDPValid() { received.Invalid.Increment() return } - // The remainder of payload must be only the neighbor solicitation, so - // payload.ToView() always returns the solicitation. Per RFC 6980 section 5, - // NDP messages cannot be fragmented. Also note that in the common case NDP - // datagrams are very small and ToView() will not incur allocations. - ns := header.NDPNeighborSolicit(payload.ToView()) + ns := header.NDPNeighborSolicit(h.NDPPayload()) it, err := ns.Options().Iter(true) if err != nil { // If we have a malformed NDP NS option, drop the packet. @@ -298,16 +286,12 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P case header.ICMPv6NeighborAdvert: received.NeighborAdvert.Increment() - if pkt.Data.Size() < header.ICMPv6NeighborAdvertSize || !isNDPValid() { + if len(v) < header.ICMPv6NeighborAdvertSize || !isNDPValid() { received.Invalid.Increment() return } - // The remainder of payload must be only the neighbor advertisement, so - // payload.ToView() always returns the advertisement. Per RFC 6980 section - // 5, NDP messages cannot be fragmented. Also note that in the common case - // NDP datagrams are very small and ToView() will not incur allocations. - na := header.NDPNeighborAdvert(payload.ToView()) + na := header.NDPNeighborAdvert(h.NDPPayload()) it, err := na.Options().Iter(true) if err != nil { // If we have a malformed NDP NA option, drop the packet. @@ -379,15 +363,14 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P case header.ICMPv6EchoRequest: received.EchoRequest.Increment() - icmpHdr, ok := pkt.Data.PullUp(header.ICMPv6EchoMinimumSize) - if !ok { + if len(v) < header.ICMPv6EchoMinimumSize { received.Invalid.Increment() return } pkt.Data.TrimFront(header.ICMPv6EchoMinimumSize) hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv6EchoMinimumSize) packet := header.ICMPv6(hdr.Prepend(header.ICMPv6EchoMinimumSize)) - copy(packet, icmpHdr) + copy(packet, h) packet.SetType(header.ICMPv6EchoReply) packet.SetChecksum(header.ICMPv6Checksum(packet, r.LocalAddress, r.RemoteAddress, pkt.Data)) if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, stack.PacketBuffer{ @@ -401,7 +384,7 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P case header.ICMPv6EchoReply: received.EchoReply.Increment() - if pkt.Data.Size() < header.ICMPv6EchoMinimumSize { + if len(v) < header.ICMPv6EchoMinimumSize { received.Invalid.Increment() return } @@ -423,9 +406,8 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P case header.ICMPv6RouterAdvert: received.RouterAdvert.Increment() - // Is the NDP payload of sufficient size to hold a Router - // Advertisement? - if pkt.Data.Size()-header.ICMPv6HeaderSize < header.NDPRAMinimumSize || !isNDPValid() { + p := h.NDPPayload() + if len(p) < header.NDPRAMinimumSize || !isNDPValid() { received.Invalid.Increment() return } @@ -443,11 +425,7 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P return } - // The remainder of payload must be only the router advertisement, so - // payload.ToView() always returns the advertisement. Per RFC 6980 section - // 5, NDP messages cannot be fragmented. Also note that in the common case - // NDP datagrams are very small and ToView() will not incur allocations. - ra := header.NDPRouterAdvert(payload.ToView()) + ra := header.NDPRouterAdvert(p) opts := ra.Options() // Are options valid as per the wire format? diff --git a/pkg/tcpip/network/ipv6/icmp_test.go b/pkg/tcpip/network/ipv6/icmp_test.go index d412ff688..bd099a7f8 100644 --- a/pkg/tcpip/network/ipv6/icmp_test.go +++ b/pkg/tcpip/network/ipv6/icmp_test.go @@ -166,8 +166,7 @@ func TestICMPCounts(t *testing.T) { }, { typ: header.ICMPv6NeighborSolicit, - size: header.ICMPv6NeighborSolicitMinimumSize, - }, + size: header.ICMPv6NeighborSolicitMinimumSize}, { typ: header.ICMPv6NeighborAdvert, size: header.ICMPv6NeighborAdvertMinimumSize, diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index 486725131..331b0817b 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -171,11 +171,7 @@ func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuffe // HandlePacket is called by the link layer when new ipv6 packets arrive for // this endpoint. func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { - headerView, ok := pkt.Data.PullUp(header.IPv6MinimumSize) - if !ok { - r.Stats().IP.MalformedPacketsReceived.Increment() - return - } + headerView := pkt.Data.First() h := header.IPv6(headerView) if !h.IsValid(pkt.Data.Size()) { r.Stats().IP.MalformedPacketsReceived.Increment() diff --git a/pkg/tcpip/stack/forwarder_test.go b/pkg/tcpip/stack/forwarder_test.go index c7c663498..e9c652042 100644 --- a/pkg/tcpip/stack/forwarder_test.go +++ b/pkg/tcpip/stack/forwarder_test.go @@ -70,10 +70,7 @@ func (f *fwdTestNetworkEndpoint) ID() *NetworkEndpointID { func (f *fwdTestNetworkEndpoint) HandlePacket(r *Route, pkt PacketBuffer) { // Consume the network header. - b, ok := pkt.Data.PullUp(fwdTestNetHeaderLen) - if !ok { - return - } + b := pkt.Data.First() pkt.Data.TrimFront(fwdTestNetHeaderLen) // Dispatch the packet to the transport protocol. @@ -476,7 +473,7 @@ func TestForwardingWithFakeResolverPartialTimeout(t *testing.T) { t.Fatal("packet not forwarded") } - b := p.Pkt.Data.ToView() + b := p.Pkt.Header.View() if b[0] != 3 { t.Fatalf("got b[0] = %d, want = 3", b[0]) } @@ -520,7 +517,7 @@ func TestForwardingWithFakeResolverTwoPackets(t *testing.T) { t.Fatal("packet not forwarded") } - b := p.Pkt.Data.ToView() + b := p.Pkt.Header.View() if b[0] != 3 { t.Fatalf("got b[0] = %d, want = 3", b[0]) } @@ -567,7 +564,7 @@ func TestForwardingWithFakeResolverManyPackets(t *testing.T) { t.Fatal("packet not forwarded") } - b := p.Pkt.Data.ToView() + b := p.Pkt.Header.View() if b[0] != 3 { t.Fatalf("got b[0] = %d, want = 3", b[0]) } @@ -622,7 +619,7 @@ func TestForwardingWithFakeResolverManyResolutions(t *testing.T) { // The first 5 packets (address 3 to 7) should not be forwarded // because their address resolutions are interrupted. - b := p.Pkt.Data.ToView() + b := p.Pkt.Header.View() if b[0] < 8 { t.Fatalf("got b[0] = %d, want b[0] >= 8", b[0]) } diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go index 6b91159d4..6c0a4b24d 100644 --- a/pkg/tcpip/stack/iptables.go +++ b/pkg/tcpip/stack/iptables.go @@ -212,11 +212,6 @@ func (it *IPTables) Check(hook Hook, pkt PacketBuffer) bool { // CheckPackets runs pkts through the rules for hook and returns a map of packets that // should not go forward. // -// Precondition: pkt is a IPv4 packet of at least length header.IPv4MinimumSize. -// -// TODO(gvisor.dev/issue/170): pk.NetworkHeader will always be set as a -// precondition. -// // NOTE: unlike the Check API the returned map contains packets that should be // dropped. func (it *IPTables) CheckPackets(hook Hook, pkts PacketBufferList) (drop map[*PacketBuffer]struct{}) { @@ -231,9 +226,7 @@ func (it *IPTables) CheckPackets(hook Hook, pkts PacketBufferList) (drop map[*Pa return drop } -// Precondition: pkt is a IPv4 packet of at least length header.IPv4MinimumSize. -// TODO(gvisor.dev/issue/170): pk.NetworkHeader will always be set as a -// precondition. +// Precondition: pkt.NetworkHeader is set. func (it *IPTables) checkChain(hook Hook, pkt PacketBuffer, table Table, ruleIdx int) chainVerdict { // Start from ruleIdx and walk the list of rules until a rule gives us // a verdict. @@ -278,21 +271,14 @@ func (it *IPTables) checkChain(hook Hook, pkt PacketBuffer, table Table, ruleIdx return chainDrop } -// Precondition: pkt is a IPv4 packet of at least length header.IPv4MinimumSize. -// TODO(gvisor.dev/issue/170): pk.NetworkHeader will always be set as a -// precondition. +// Precondition: pk.NetworkHeader is set. func (it *IPTables) checkRule(hook Hook, pkt PacketBuffer, table Table, ruleIdx int) (RuleVerdict, int) { rule := table.Rules[ruleIdx] // If pkt.NetworkHeader hasn't been set yet, it will be contained in - // pkt.Data. + // pkt.Data.First(). if pkt.NetworkHeader == nil { - var ok bool - pkt.NetworkHeader, ok = pkt.Data.PullUp(header.IPv4MinimumSize) - if !ok { - // Precondition has been violated. - panic(fmt.Sprintf("iptables checks require IPv4 headers of at least %d bytes", header.IPv4MinimumSize)) - } + pkt.NetworkHeader = pkt.Data.First() } // Check whether the packet matches the IP header filter. diff --git a/pkg/tcpip/stack/iptables_targets.go b/pkg/tcpip/stack/iptables_targets.go index 8be61f4b1..7b4543caf 100644 --- a/pkg/tcpip/stack/iptables_targets.go +++ b/pkg/tcpip/stack/iptables_targets.go @@ -96,12 +96,9 @@ func (rt RedirectTarget) Action(pkt PacketBuffer) (RuleVerdict, int) { newPkt := pkt.Clone() // Set network header. - headerView, ok := newPkt.Data.PullUp(header.IPv4MinimumSize) - if !ok { - return RuleDrop, 0 - } + headerView := newPkt.Data.First() netHeader := header.IPv4(headerView) - newPkt.NetworkHeader = headerView + newPkt.NetworkHeader = headerView[:header.IPv4MinimumSize] hlen := int(netHeader.HeaderLength()) tlen := int(netHeader.TotalLength()) @@ -120,14 +117,10 @@ func (rt RedirectTarget) Action(pkt PacketBuffer) (RuleVerdict, int) { if newPkt.TransportHeader != nil { udpHeader = header.UDP(newPkt.TransportHeader) } else { - if pkt.Data.Size() < header.UDPMinimumSize { - return RuleDrop, 0 - } - hdr, ok := newPkt.Data.PullUp(header.UDPMinimumSize) - if !ok { + if len(pkt.Data.First()) < header.UDPMinimumSize { return RuleDrop, 0 } - udpHeader = header.UDP(hdr) + udpHeader = header.UDP(newPkt.Data.First()) } udpHeader.SetDestinationPort(rt.MinPort) case header.TCPProtocolNumber: @@ -135,14 +128,10 @@ func (rt RedirectTarget) Action(pkt PacketBuffer) (RuleVerdict, int) { if newPkt.TransportHeader != nil { tcpHeader = header.TCP(newPkt.TransportHeader) } else { - if pkt.Data.Size() < header.TCPMinimumSize { + if len(pkt.Data.First()) < header.TCPMinimumSize { return RuleDrop, 0 } - hdr, ok := newPkt.Data.PullUp(header.TCPMinimumSize) - if !ok { - return RuleDrop, 0 - } - tcpHeader = header.TCP(hdr) + tcpHeader = header.TCP(newPkt.TransportHeader) } // TODO(gvisor.dev/issue/170): Need to recompute checksum // and implement nat connection tracking to support TCP. diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 0c2b1f36a..016dbe15e 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -1203,12 +1203,12 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link n.stack.stats.IP.PacketsReceived.Increment() } - netHeader, ok := pkt.Data.PullUp(netProto.MinimumPacketSize()) - if !ok { + if len(pkt.Data.First()) < netProto.MinimumPacketSize() { n.stack.stats.MalformedRcvdPackets.Increment() return } - src, dst := netProto.ParseAddresses(netHeader) + + src, dst := netProto.ParseAddresses(pkt.Data.First()) if n.stack.handleLocal && !n.isLoopback() && n.getRef(protocol, src) != nil { // The source address is one of our own, so we never should have gotten a @@ -1289,8 +1289,22 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link func (n *NIC) forwardPacket(r *Route, protocol tcpip.NetworkProtocolNumber, pkt PacketBuffer) { // TODO(b/143425874) Decrease the TTL field in forwarded packets. - if linkHeaderLen := int(n.linkEP.MaxHeaderLength()); linkHeaderLen != 0 { - pkt.Header = buffer.NewPrependable(linkHeaderLen) + + firstData := pkt.Data.First() + pkt.Data.RemoveFirst() + + if linkHeaderLen := int(n.linkEP.MaxHeaderLength()); linkHeaderLen == 0 { + pkt.Header = buffer.NewPrependableFromView(firstData) + } else { + firstDataLen := len(firstData) + + // pkt.Header should have enough capacity to hold n.linkEP's headers. + pkt.Header = buffer.NewPrependable(firstDataLen + linkHeaderLen) + + // TODO(b/151227689): avoid copying the packet when forwarding + if n := copy(pkt.Header.Prepend(firstDataLen), firstData); n != firstDataLen { + panic(fmt.Sprintf("copied %d bytes, expected %d", n, firstDataLen)) + } } if err := n.linkEP.WritePacket(r, nil /* gso */, protocol, pkt); err != nil { @@ -1318,13 +1332,12 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN // validly formed. n.stack.demux.deliverRawPacket(r, protocol, pkt) - transHeader, ok := pkt.Data.PullUp(transProto.MinimumPacketSize()) - if !ok { + if len(pkt.Data.First()) < transProto.MinimumPacketSize() { n.stack.stats.MalformedRcvdPackets.Increment() return } - srcPort, dstPort, err := transProto.ParsePorts(transHeader) + srcPort, dstPort, err := transProto.ParsePorts(pkt.Data.First()) if err != nil { n.stack.stats.MalformedRcvdPackets.Increment() return @@ -1362,12 +1375,11 @@ func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcp // ICMPv4 only guarantees that 8 bytes of the transport protocol will // be present in the payload. We know that the ports are within the // first 8 bytes for all known transport protocols. - transHeader, ok := pkt.Data.PullUp(8) - if !ok { + if len(pkt.Data.First()) < 8 { return } - srcPort, dstPort, err := transProto.ParsePorts(transHeader) + srcPort, dstPort, err := transProto.ParsePorts(pkt.Data.First()) if err != nil { return } diff --git a/pkg/tcpip/stack/packet_buffer.go b/pkg/tcpip/stack/packet_buffer.go index e954a8b7e..dc125f25e 100644 --- a/pkg/tcpip/stack/packet_buffer.go +++ b/pkg/tcpip/stack/packet_buffer.go @@ -37,9 +37,7 @@ type PacketBuffer struct { Data buffer.VectorisedView // Header holds the headers of outbound packets. As a packet is passed - // down the stack, each layer adds to Header. Note that forwarded - // packets don't populate Headers on their way out -- their headers and - // payload are never parsed out and remain in Data. + // down the stack, each layer adds to Header. Header buffer.Prependable // These fields are used by both inbound and outbound packets. They diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index d45d2cc1f..c7634ceb1 100644 --- a/pkg/tcpip/stack/stack_test.go +++ b/pkg/tcpip/stack/stack_test.go @@ -95,18 +95,16 @@ func (f *fakeNetworkEndpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffe f.proto.packetCount[int(f.id.LocalAddress[0])%len(f.proto.packetCount)]++ // Consume the network header. - b, ok := pkt.Data.PullUp(fakeNetHeaderLen) - if !ok { - return - } + b := pkt.Data.First() pkt.Data.TrimFront(fakeNetHeaderLen) // Handle control packets. if b[2] == uint8(fakeControlProtocol) { - nb, ok := pkt.Data.PullUp(fakeNetHeaderLen) - if !ok { + nb := pkt.Data.First() + if len(nb) < fakeNetHeaderLen { return } + pkt.Data.TrimFront(fakeNetHeaderLen) f.dispatcher.DeliverTransportControlPacket(tcpip.Address(nb[1:2]), tcpip.Address(nb[0:1]), fakeNetNumber, tcpip.TransportProtocolNumber(nb[2]), stack.ControlPortUnreachable, 0, pkt) return diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go index a611e44ab..3084e6593 100644 --- a/pkg/tcpip/stack/transport_test.go +++ b/pkg/tcpip/stack/transport_test.go @@ -642,11 +642,10 @@ func TestTransportForwarding(t *testing.T) { t.Fatal("Response packet not forwarded") } - hdrs := p.Pkt.Data.ToView() - if dst := hdrs[0]; dst != 3 { + if dst := p.Pkt.Header.View()[0]; dst != 3 { t.Errorf("Response packet has incorrect destination addresss: got = %d, want = 3", dst) } - if src := hdrs[1]; src != 1 { + if src := p.Pkt.Header.View()[1]; src != 1 { t.Errorf("Response packet has incorrect source addresss: got = %d, want = 3", src) } } diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index b1d820372..feef8dca0 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -747,15 +747,15 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk // Only accept echo replies. switch e.NetProto { case header.IPv4ProtocolNumber: - h, ok := pkt.Data.PullUp(header.ICMPv4MinimumSize) - if !ok || header.ICMPv4(h).Type() != header.ICMPv4EchoReply { + h := header.ICMPv4(pkt.Data.First()) + if h.Type() != header.ICMPv4EchoReply { e.stack.Stats().DroppedPackets.Increment() e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() return } case header.IPv6ProtocolNumber: - h, ok := pkt.Data.PullUp(header.ICMPv6MinimumSize) - if !ok || header.ICMPv6(h).Type() != header.ICMPv6EchoReply { + h := header.ICMPv6(pkt.Data.First()) + if h.Type() != header.ICMPv6EchoReply { e.stack.Stats().DroppedPackets.Increment() e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() return diff --git a/pkg/tcpip/transport/tcp/segment.go b/pkg/tcpip/transport/tcp/segment.go index 7712ce652..40461fd31 100644 --- a/pkg/tcpip/transport/tcp/segment.go +++ b/pkg/tcpip/transport/tcp/segment.go @@ -144,11 +144,7 @@ func (s *segment) logicalLen() seqnum.Size { // TCP checksum and stores the checksum and result of checksum verification in // the csum and csumValid fields of the segment. func (s *segment) parse() bool { - h, ok := s.data.PullUp(header.TCPMinimumSize) - if !ok { - return false - } - hdr := header.TCP(h) + h := header.TCP(s.data.First()) // h is the header followed by the payload. We check that the offset to // the data respects the following constraints: @@ -160,16 +156,12 @@ func (s *segment) parse() bool { // N.B. The segment has already been validated as having at least the // minimum TCP size before reaching here, so it's safe to read the // fields. - offset := int(hdr.DataOffset()) - if offset < header.TCPMinimumSize { - return false - } - hdrWithOpts, ok := s.data.PullUp(offset) - if !ok { + offset := int(h.DataOffset()) + if offset < header.TCPMinimumSize || offset > len(h) { return false } - s.options = []byte(hdrWithOpts[header.TCPMinimumSize:]) + s.options = []byte(h[header.TCPMinimumSize:offset]) s.parsedOptions = header.ParseTCPOptions(s.options) // Query the link capabilities to decide if checksum validation is @@ -181,19 +173,18 @@ func (s *segment) parse() bool { s.data.TrimFront(offset) } if verifyChecksum { - hdr = header.TCP(hdrWithOpts) - s.csum = hdr.Checksum() + s.csum = h.Checksum() xsum := s.route.PseudoHeaderChecksum(ProtocolNumber, uint16(s.data.Size())) - xsum = hdr.CalculateChecksum(xsum) + xsum = h.CalculateChecksum(xsum) s.data.TrimFront(offset) xsum = header.ChecksumVV(s.data, xsum) s.csumValid = xsum == 0xffff } - s.sequenceNumber = seqnum.Value(hdr.SequenceNumber()) - s.ackNumber = seqnum.Value(hdr.AckNumber()) - s.flags = hdr.Flags() - s.window = seqnum.Size(hdr.WindowSize()) + s.sequenceNumber = seqnum.Value(h.SequenceNumber()) + s.ackNumber = seqnum.Value(h.AckNumber()) + s.flags = h.Flags() + s.window = seqnum.Size(h.WindowSize()) return true } diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index 286c66cf5..ab1014c7f 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -3548,7 +3548,7 @@ func TestReceivedInvalidSegmentCountIncrement(t *testing.T) { AckNum: c.IRS.Add(1), RcvWnd: 30000, }) - tcpbuf := vv.ToView()[header.IPv4MinimumSize:] + tcpbuf := vv.First()[header.IPv4MinimumSize:] tcpbuf[header.TCPDataOffset] = ((header.TCPMinimumSize - 1) / 4) << 4 c.SendSegment(vv) @@ -3575,7 +3575,7 @@ func TestReceivedIncorrectChecksumIncrement(t *testing.T) { AckNum: c.IRS.Add(1), RcvWnd: 30000, }) - tcpbuf := vv.ToView()[header.IPv4MinimumSize:] + tcpbuf := vv.First()[header.IPv4MinimumSize:] // Overwrite a byte in the payload which should cause checksum // verification to fail. tcpbuf[(tcpbuf[header.TCPDataOffset]>>4)*4] = 0x4 diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 756ab913a..edb54f0be 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -1250,8 +1250,8 @@ func (e *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { // endpoint. func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) { // Get the header then trim it from the view. - hdr, ok := pkt.Data.PullUp(header.UDPMinimumSize) - if !ok || int(header.UDP(hdr).Length()) > pkt.Data.Size() { + hdr := header.UDP(pkt.Data.First()) + if int(hdr.Length()) > pkt.Data.Size() { // Malformed packet. e.stack.Stats().UDP.MalformedPacketsReceived.Increment() e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() @@ -1286,7 +1286,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk senderAddress: tcpip.FullAddress{ NIC: r.NICID(), Addr: id.RemoteAddress, - Port: header.UDP(hdr).SourcePort(), + Port: hdr.SourcePort(), }, } packet.data = pkt.Data diff --git a/pkg/tcpip/transport/udp/protocol.go b/pkg/tcpip/transport/udp/protocol.go index 52af6de22..6e31a9bac 100644 --- a/pkg/tcpip/transport/udp/protocol.go +++ b/pkg/tcpip/transport/udp/protocol.go @@ -68,13 +68,8 @@ func (*protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) { // that don't match any existing endpoint. func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) bool { // Get the header then trim it from the view. - h, ok := pkt.Data.PullUp(header.UDPMinimumSize) - if !ok { - // Malformed packet. - r.Stack().Stats().UDP.MalformedPacketsReceived.Increment() - return true - } - if int(header.UDP(h).Length()) > pkt.Data.Size() { + hdr := header.UDP(pkt.Data.First()) + if int(hdr.Length()) > pkt.Data.Size() { // Malformed packet. r.Stack().Stats().UDP.MalformedPacketsReceived.Increment() return true -- cgit v1.2.3 From 7c0f3bc8576addbec001095d754a756691d26df3 Mon Sep 17 00:00:00 2001 From: Dave Bailey Date: Tue, 21 Apr 2020 09:34:42 -0700 Subject: Sentry metrics updates. Sentry metrics with nanoseconds units are labeled as such, and non-cumulative sentry metrics are supported. PiperOrigin-RevId: 307621080 --- pkg/metric/metric.go | 41 +++++++++++++++++----------------- pkg/metric/metric.proto | 10 ++++++++- pkg/metric/metric_test.go | 22 +++++++++++------- pkg/sentry/fs/file.go | 2 +- pkg/sentry/fs/gofer/file.go | 4 ++-- pkg/sentry/fs/tmpfs/inode_file.go | 2 +- pkg/sentry/socket/netstack/netstack.go | 14 ++++++++---- 7 files changed, 58 insertions(+), 37 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/metric/metric.go b/pkg/metric/metric.go index 895253625..64aa365ce 100644 --- a/pkg/metric/metric.go +++ b/pkg/metric/metric.go @@ -39,16 +39,11 @@ var ( // Uint64Metric encapsulates a uint64 that represents some kind of metric to be // monitored. // -// All metrics must be cumulative, meaning that their values will only increase -// over time. -// // Metrics are not saved across save/restore and thus reset to zero on restore. // -// TODO(b/67298402): Support non-cumulative metrics. // TODO(b/67298427): Support metric fields. type Uint64Metric struct { - // value is the actual value of the metric. It must be accessed - // atomically. + // value is the actual value of the metric. It must be accessed atomically. value uint64 } @@ -110,13 +105,10 @@ type customUint64Metric struct { // Register must only be called at init and will return and error if called // after Initialized. // -// All metrics must be cumulative, meaning that the return values of value must -// only increase over time. -// // Preconditions: // * name must be globally unique. // * Initialize/Disable have not been called. -func RegisterCustomUint64Metric(name string, sync bool, description string, value func() uint64) error { +func RegisterCustomUint64Metric(name string, cumulative, sync bool, units pb.MetricMetadata_Units, description string, value func() uint64) error { if initialized { return ErrInitializationDone } @@ -129,9 +121,10 @@ func RegisterCustomUint64Metric(name string, sync bool, description string, valu metadata: &pb.MetricMetadata{ Name: name, Description: description, - Cumulative: true, + Cumulative: cumulative, Sync: sync, - Type: pb.MetricMetadata_UINT64, + Type: pb.MetricMetadata_TYPE_UINT64, + Units: units, }, value: value, } @@ -140,24 +133,32 @@ func RegisterCustomUint64Metric(name string, sync bool, description string, valu // MustRegisterCustomUint64Metric calls RegisterCustomUint64Metric and panics // if it returns an error. -func MustRegisterCustomUint64Metric(name string, sync bool, description string, value func() uint64) { - if err := RegisterCustomUint64Metric(name, sync, description, value); err != nil { +func MustRegisterCustomUint64Metric(name string, cumulative, sync bool, description string, value func() uint64) { + if err := RegisterCustomUint64Metric(name, cumulative, sync, pb.MetricMetadata_UNITS_NONE, description, value); err != nil { panic(fmt.Sprintf("Unable to register metric %q: %v", name, err)) } } -// NewUint64Metric creates and registers a new metric with the given name. +// NewUint64Metric creates and registers a new cumulative metric with the given name. // // Metrics must be statically defined (i.e., at init). -func NewUint64Metric(name string, sync bool, description string) (*Uint64Metric, error) { +func NewUint64Metric(name string, sync bool, units pb.MetricMetadata_Units, description string) (*Uint64Metric, error) { var m Uint64Metric - return &m, RegisterCustomUint64Metric(name, sync, description, m.Value) + return &m, RegisterCustomUint64Metric(name, true /* cumulative */, sync, units, description, m.Value) } -// MustCreateNewUint64Metric calls NewUint64Metric and panics if it returns an -// error. +// MustCreateNewUint64Metric calls NewUint64Metric and panics if it returns an error. func MustCreateNewUint64Metric(name string, sync bool, description string) *Uint64Metric { - m, err := NewUint64Metric(name, sync, description) + m, err := NewUint64Metric(name, sync, pb.MetricMetadata_UNITS_NONE, description) + if err != nil { + panic(fmt.Sprintf("Unable to create metric %q: %v", name, err)) + } + return m +} + +// MustCreateNewUint64NanosecondsMetric calls NewUint64Metric and panics if it returns an error. +func MustCreateNewUint64NanosecondsMetric(name string, sync bool, description string) *Uint64Metric { + m, err := NewUint64Metric(name, sync, pb.MetricMetadata_UNITS_NANOSECONDS, description) if err != nil { panic(fmt.Sprintf("Unable to create metric %q: %v", name, err)) } diff --git a/pkg/metric/metric.proto b/pkg/metric/metric.proto index a2c2bd1ba..3cc89047d 100644 --- a/pkg/metric/metric.proto +++ b/pkg/metric/metric.proto @@ -36,10 +36,18 @@ message MetricMetadata { // the monitoring system. bool sync = 4; - enum Type { UINT64 = 0; } + enum Type { TYPE_UINT64 = 0; } // type is the type of the metric value. Type type = 5; + + enum Units { + UNITS_NONE = 0; + UNITS_NANOSECONDS = 1; + } + + // units is the units of the metric value. + Units units = 6; } // MetricRegistration contains the metadata for all metrics that will be in diff --git a/pkg/metric/metric_test.go b/pkg/metric/metric_test.go index 34969385a..c425ea532 100644 --- a/pkg/metric/metric_test.go +++ b/pkg/metric/metric_test.go @@ -66,12 +66,12 @@ const ( func TestInitialize(t *testing.T) { defer reset() - _, err := NewUint64Metric("/foo", false, fooDescription) + _, err := NewUint64Metric("/foo", false, pb.MetricMetadata_UNITS_NONE, fooDescription) if err != nil { t.Fatalf("NewUint64Metric got err %v want nil", err) } - _, err = NewUint64Metric("/bar", true, barDescription) + _, err = NewUint64Metric("/bar", true, pb.MetricMetadata_UNITS_NANOSECONDS, barDescription) if err != nil { t.Fatalf("NewUint64Metric got err %v want nil", err) } @@ -94,8 +94,8 @@ func TestInitialize(t *testing.T) { foundFoo := false foundBar := false for _, m := range mr.Metrics { - if m.Type != pb.MetricMetadata_UINT64 { - t.Errorf("Metadata %+v Type got %v want %v", m, m.Type, pb.MetricMetadata_UINT64) + if m.Type != pb.MetricMetadata_TYPE_UINT64 { + t.Errorf("Metadata %+v Type got %v want %v", m, m.Type, pb.MetricMetadata_TYPE_UINT64) } if !m.Cumulative { t.Errorf("Metadata %+v Cumulative got false want true", m) @@ -110,6 +110,9 @@ func TestInitialize(t *testing.T) { if m.Sync { t.Errorf("/foo %+v Sync got true want false", m) } + if m.Units != pb.MetricMetadata_UNITS_NONE { + t.Errorf("/foo %+v Units got %v want %v", m, m.Units, pb.MetricMetadata_UNITS_NONE) + } case "/bar": foundBar = true if m.Description != barDescription { @@ -118,6 +121,9 @@ func TestInitialize(t *testing.T) { if !m.Sync { t.Errorf("/bar %+v Sync got true want false", m) } + if m.Units != pb.MetricMetadata_UNITS_NANOSECONDS { + t.Errorf("/bar %+v Units got %v want %v", m, m.Units, pb.MetricMetadata_UNITS_NANOSECONDS) + } } } @@ -132,12 +138,12 @@ func TestInitialize(t *testing.T) { func TestDisable(t *testing.T) { defer reset() - _, err := NewUint64Metric("/foo", false, fooDescription) + _, err := NewUint64Metric("/foo", false, pb.MetricMetadata_UNITS_NONE, fooDescription) if err != nil { t.Fatalf("NewUint64Metric got err %v want nil", err) } - _, err = NewUint64Metric("/bar", true, barDescription) + _, err = NewUint64Metric("/bar", true, pb.MetricMetadata_UNITS_NONE, barDescription) if err != nil { t.Fatalf("NewUint64Metric got err %v want nil", err) } @@ -161,12 +167,12 @@ func TestDisable(t *testing.T) { func TestEmitMetricUpdate(t *testing.T) { defer reset() - foo, err := NewUint64Metric("/foo", false, fooDescription) + foo, err := NewUint64Metric("/foo", false, pb.MetricMetadata_UNITS_NONE, fooDescription) if err != nil { t.Fatalf("NewUint64Metric got err %v want nil", err) } - _, err = NewUint64Metric("/bar", true, barDescription) + _, err = NewUint64Metric("/bar", true, pb.MetricMetadata_UNITS_NONE, barDescription) if err != nil { t.Fatalf("NewUint64Metric got err %v want nil", err) } diff --git a/pkg/sentry/fs/file.go b/pkg/sentry/fs/file.go index 78100e448..846252c89 100644 --- a/pkg/sentry/fs/file.go +++ b/pkg/sentry/fs/file.go @@ -44,7 +44,7 @@ var ( RecordWaitTime = false reads = metric.MustCreateNewUint64Metric("/fs/reads", false /* sync */, "Number of file reads.") - readWait = metric.MustCreateNewUint64Metric("/fs/read_wait", false /* sync */, "Time waiting on file reads, in nanoseconds.") + readWait = metric.MustCreateNewUint64NanosecondsMetric("/fs/read_wait", false /* sync */, "Time waiting on file reads, in nanoseconds.") ) // IncrementWait increments the given wait time metric, if enabled. diff --git a/pkg/sentry/fs/gofer/file.go b/pkg/sentry/fs/gofer/file.go index 23296f246..b2fcab127 100644 --- a/pkg/sentry/fs/gofer/file.go +++ b/pkg/sentry/fs/gofer/file.go @@ -37,9 +37,9 @@ var ( opens9P = metric.MustCreateNewUint64Metric("/gofer/opens_9p", false /* sync */, "Number of times a 9P file was opened from a gofer.") opensHost = metric.MustCreateNewUint64Metric("/gofer/opens_host", false /* sync */, "Number of times a host file was opened from a gofer.") reads9P = metric.MustCreateNewUint64Metric("/gofer/reads_9p", false /* sync */, "Number of 9P file reads from a gofer.") - readWait9P = metric.MustCreateNewUint64Metric("/gofer/read_wait_9p", false /* sync */, "Time waiting on 9P file reads from a gofer, in nanoseconds.") + readWait9P = metric.MustCreateNewUint64NanosecondsMetric("/gofer/read_wait_9p", false /* sync */, "Time waiting on 9P file reads from a gofer, in nanoseconds.") readsHost = metric.MustCreateNewUint64Metric("/gofer/reads_host", false /* sync */, "Number of host file reads from a gofer.") - readWaitHost = metric.MustCreateNewUint64Metric("/gofer/read_wait_host", false /* sync */, "Time waiting on host file reads from a gofer, in nanoseconds.") + readWaitHost = metric.MustCreateNewUint64NanosecondsMetric("/gofer/read_wait_host", false /* sync */, "Time waiting on host file reads from a gofer, in nanoseconds.") ) // fileOperations implements fs.FileOperations for a remote file system. diff --git a/pkg/sentry/fs/tmpfs/inode_file.go b/pkg/sentry/fs/tmpfs/inode_file.go index 25abbc151..1dc75291d 100644 --- a/pkg/sentry/fs/tmpfs/inode_file.go +++ b/pkg/sentry/fs/tmpfs/inode_file.go @@ -39,7 +39,7 @@ var ( opensRO = metric.MustCreateNewUint64Metric("/in_memory_file/opens_ro", false /* sync */, "Number of times an in-memory file was opened in read-only mode.") opensW = metric.MustCreateNewUint64Metric("/in_memory_file/opens_w", false /* sync */, "Number of times an in-memory file was opened in write mode.") reads = metric.MustCreateNewUint64Metric("/in_memory_file/reads", false /* sync */, "Number of in-memory file reads.") - readWait = metric.MustCreateNewUint64Metric("/in_memory_file/read_wait", false /* sync */, "Time waiting on in-memory file reads, in nanoseconds.") + readWait = metric.MustCreateNewUint64NanosecondsMetric("/in_memory_file/read_wait", false /* sync */, "Time waiting on in-memory file reads, in nanoseconds.") ) // fileInodeOperations implements fs.InodeOperations for a regular tmpfs file. diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 7ac38764d..d5879c10f 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -63,7 +63,13 @@ import ( func mustCreateMetric(name, description string) *tcpip.StatCounter { var cm tcpip.StatCounter - metric.MustRegisterCustomUint64Metric(name, false /* sync */, description, cm.Value) + metric.MustRegisterCustomUint64Metric(name, true /* cumulative */, false /* sync */, description, cm.Value) + return &cm +} + +func mustCreateGauge(name, description string) *tcpip.StatCounter { + var cm tcpip.StatCounter + metric.MustRegisterCustomUint64Metric(name, false /* cumulative */, false /* sync */, description, cm.Value) return &cm } @@ -151,10 +157,10 @@ var Metrics = tcpip.Stats{ TCP: tcpip.TCPStats{ ActiveConnectionOpenings: mustCreateMetric("/netstack/tcp/active_connection_openings", "Number of connections opened successfully via Connect."), PassiveConnectionOpenings: mustCreateMetric("/netstack/tcp/passive_connection_openings", "Number of connections opened successfully via Listen."), - CurrentEstablished: mustCreateMetric("/netstack/tcp/current_established", "Number of connections in ESTABLISHED state now."), - CurrentConnected: mustCreateMetric("/netstack/tcp/current_open", "Number of connections that are in connected state."), + CurrentEstablished: mustCreateGauge("/netstack/tcp/current_established", "Number of connections in ESTABLISHED state now."), + CurrentConnected: mustCreateGauge("/netstack/tcp/current_open", "Number of connections that are in connected state."), EstablishedResets: mustCreateMetric("/netstack/tcp/established_resets", "Number of times TCP connections have made a direct transition to the CLOSED state from either the ESTABLISHED state or the CLOSE-WAIT state"), - EstablishedClosed: mustCreateMetric("/netstack/tcp/established_closed", "number of times established TCP connections made a transition to CLOSED state."), + EstablishedClosed: mustCreateMetric("/netstack/tcp/established_closed", "Number of times established TCP connections made a transition to CLOSED state."), EstablishedTimedout: mustCreateMetric("/netstack/tcp/established_timedout", "Number of times an established connection was reset because of keep-alive time out."), ListenOverflowSynDrop: mustCreateMetric("/netstack/tcp/listen_overflow_syn_drop", "Number of times the listen queue overflowed and a SYN was dropped."), ListenOverflowAckDrop: mustCreateMetric("/netstack/tcp/listen_overflow_ack_drop", "Number of times the listen queue overflowed and the final ACK in the handshake was dropped."), -- cgit v1.2.3 From 9b5e305e05ef3ad51778981062d6152cea1cd4fb Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Tue, 21 Apr 2020 12:16:42 -0700 Subject: Remove filesystem structure from vfs.Dentry. MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This change: - Drastically simplifies the synchronization model: filesystem structure is both implementation-defined and implementation-synchronized. - Allows implementations of vfs.DentryImpl to use implementation-specific dentry types, reducing casts during path traversal. - Doesn't require dentries representing non-directory files to waste space on a map of children. - Allows dentry revalidation and mount lookup to be correctly ordered (fixed FIXME in fsimpl/gofer/filesystem.go). - Removes the need to have two separate maps in gofer.dentry (dentry.vfsd.children and dentry.negativeChildren) for positive and negative lookups respectively. //pkg/sentry/fsimpl/tmpfs/benchmark_test.go: name old time/op new time/op delta VFS2TmpfsStat/1-112 172ns ± 4% 165ns ± 3% -4.08% (p=0.002 n=9+9) VFS2TmpfsStat/2-112 199ns ± 3% 195ns ±10% ~ (p=0.132 n=8+9) VFS2TmpfsStat/3-112 230ns ± 2% 216ns ± 2% -6.15% (p=0.000 n=8+8) VFS2TmpfsStat/8-112 390ns ± 2% 358ns ± 4% -8.33% (p=0.000 n=9+8) VFS2TmpfsStat/64-112 2.20µs ± 3% 2.01µs ± 3% -8.48% (p=0.000 n=10+8) VFS2TmpfsStat/100-112 3.42µs ± 9% 3.08µs ± 2% -9.82% (p=0.000 n=9+8) VFS2TmpfsMountStat/1-112 278ns ± 1% 286ns ±15% ~ (p=0.712 n=8+10) VFS2TmpfsMountStat/2-112 311ns ± 4% 298ns ± 2% -4.27% (p=0.000 n=9+8) VFS2TmpfsMountStat/3-112 339ns ± 3% 330ns ± 9% ~ (p=0.070 n=8+9) VFS2TmpfsMountStat/8-112 503ns ± 3% 466ns ± 3% -7.38% (p=0.000 n=8+8) VFS2TmpfsMountStat/64-112 2.53µs ±16% 2.17µs ± 7% -14.19% (p=0.000 n=10+9) VFS2TmpfsMountStat/100-112 3.60µs ± 4% 3.30µs ± 8% -8.33% (p=0.001 n=8+9) Updates #1035 PiperOrigin-RevId: 307655892 --- pkg/sentry/fsimpl/ext/BUILD | 12 ++ pkg/sentry/fsimpl/ext/dentry.go | 4 + pkg/sentry/fsimpl/ext/directory.go | 21 ++- pkg/sentry/fsimpl/ext/filesystem.go | 54 ++++-- pkg/sentry/fsimpl/ext/inode.go | 2 +- pkg/sentry/fsimpl/gofer/BUILD | 12 ++ pkg/sentry/fsimpl/gofer/directory.go | 55 +++--- pkg/sentry/fsimpl/gofer/filesystem.go | 202 +++++++++++--------- pkg/sentry/fsimpl/gofer/gofer.go | 66 ++++--- pkg/sentry/fsimpl/gofer/gofer_test.go | 3 +- pkg/sentry/fsimpl/kernfs/BUILD | 12 ++ pkg/sentry/fsimpl/kernfs/fd_impl_util.go | 2 +- pkg/sentry/fsimpl/kernfs/filesystem.go | 159 +++++++++------- pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 2 +- pkg/sentry/fsimpl/kernfs/kernfs.go | 33 ++-- pkg/sentry/fsimpl/proc/tasks_test.go | 16 +- pkg/sentry/fsimpl/tmpfs/BUILD | 12 ++ pkg/sentry/fsimpl/tmpfs/benchmark_test.go | 7 - pkg/sentry/fsimpl/tmpfs/directory.go | 84 ++++++--- pkg/sentry/fsimpl/tmpfs/filesystem.go | 249 +++++++++++++------------ pkg/sentry/fsimpl/tmpfs/stat_test.go | 12 +- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 82 ++++---- pkg/sentry/vfs/dentry.go | 259 +++++--------------------- pkg/sentry/vfs/file_description.go | 3 +- pkg/sentry/vfs/filesystem.go | 5 +- pkg/sentry/vfs/filesystem_impl_util.go | 26 --- pkg/sentry/vfs/genericfstree/BUILD | 16 ++ pkg/sentry/vfs/genericfstree/genericfstree.go | 80 ++++++++ pkg/sentry/vfs/mount.go | 9 +- pkg/sentry/vfs/pathname.go | 6 +- pkg/sentry/vfs/resolving_path.go | 85 +++------ 31 files changed, 836 insertions(+), 754 deletions(-) create mode 100644 pkg/sentry/vfs/genericfstree/BUILD create mode 100644 pkg/sentry/vfs/genericfstree/genericfstree.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/ext/BUILD b/pkg/sentry/fsimpl/ext/BUILD index d83d75b3d..a4947c480 100644 --- a/pkg/sentry/fsimpl/ext/BUILD +++ b/pkg/sentry/fsimpl/ext/BUILD @@ -15,6 +15,17 @@ go_template_instance( }, ) +go_template_instance( + name = "fstree", + out = "fstree.go", + package = "ext", + prefix = "generic", + template = "//pkg/sentry/vfs/genericfstree:generic_fstree", + types = { + "Dentry": "dentry", + }, +) + go_library( name = "ext", srcs = [ @@ -26,6 +37,7 @@ go_library( "extent_file.go", "file_description.go", "filesystem.go", + "fstree.go", "inode.go", "regular_file.go", "symlink.go", diff --git a/pkg/sentry/fsimpl/ext/dentry.go b/pkg/sentry/fsimpl/ext/dentry.go index a080cb189..bfbd7c3d4 100644 --- a/pkg/sentry/fsimpl/ext/dentry.go +++ b/pkg/sentry/fsimpl/ext/dentry.go @@ -22,6 +22,10 @@ import ( type dentry struct { vfsd vfs.Dentry + // Protected by filesystem.mu. + parent *dentry + name string + // inode is the inode represented by this dentry. Multiple Dentries may // share a single non-directory Inode (with hard links). inode is // immutable. diff --git a/pkg/sentry/fsimpl/ext/directory.go b/pkg/sentry/fsimpl/ext/directory.go index bd6ede995..12b875c8f 100644 --- a/pkg/sentry/fsimpl/ext/directory.go +++ b/pkg/sentry/fsimpl/ext/directory.go @@ -21,7 +21,6 @@ import ( "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" - "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" @@ -31,6 +30,10 @@ import ( type directory struct { inode inode + // childCache maps filenames to dentries for children for which dentries + // have been instantiated. childCache is protected by filesystem.mu. + childCache map[string]*dentry + // mu serializes the changes to childList. // Lock Order (outermost locks must be taken first): // directory.mu @@ -50,9 +53,13 @@ type directory struct { childMap map[string]*dirent } -// newDirectroy is the directory constructor. -func newDirectroy(inode inode, newDirent bool) (*directory, error) { - file := &directory{inode: inode, childMap: make(map[string]*dirent)} +// newDirectory is the directory constructor. +func newDirectory(inode inode, newDirent bool) (*directory, error) { + file := &directory{ + inode: inode, + childCache: make(map[string]*dentry), + childMap: make(map[string]*dirent), + } file.inode.impl = file // Initialize childList by reading dirents from the underlying file. @@ -299,9 +306,3 @@ func (fd *directoryFD) Seek(ctx context.Context, offset int64, whence int32) (in fd.off = offset return offset, nil } - -// ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. -func (fd *directoryFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { - // mmap(2) specifies that EACCESS should be returned for non-regular file fds. - return syserror.EACCES -} diff --git a/pkg/sentry/fsimpl/ext/filesystem.go b/pkg/sentry/fsimpl/ext/filesystem.go index afea58f65..2c22a04af 100644 --- a/pkg/sentry/fsimpl/ext/filesystem.go +++ b/pkg/sentry/fsimpl/ext/filesystem.go @@ -89,14 +89,33 @@ func stepLocked(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, inode *inode, write boo } for { - nextVFSD, err := rp.ResolveComponent(vfsd) - if err != nil { - return nil, nil, err + name := rp.Component() + if name == "." { + rp.Advance() + return vfsd, inode, nil } - if nextVFSD == nil { - // Since the Dentry tree is not the sole source of truth for extfs, if it's - // not in the Dentry tree, it might need to be pulled from disk. - childDirent, ok := inode.impl.(*directory).childMap[rp.Component()] + d := vfsd.Impl().(*dentry) + if name == ".." { + isRoot, err := rp.CheckRoot(vfsd) + if err != nil { + return nil, nil, err + } + if isRoot || d.parent == nil { + rp.Advance() + return vfsd, inode, nil + } + if err := rp.CheckMount(&d.parent.vfsd); err != nil { + return nil, nil, err + } + rp.Advance() + return &d.parent.vfsd, d.parent.inode, nil + } + + dir := inode.impl.(*directory) + child, ok := dir.childCache[name] + if !ok { + // We may need to instantiate a new dentry for this child. + childDirent, ok := dir.childMap[name] if !ok { // The underlying inode does not exist on disk. return nil, nil, syserror.ENOENT @@ -115,21 +134,22 @@ func stepLocked(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, inode *inode, write boo } // incRef because this is being added to the dentry tree. childInode.incRef() - child := newDentry(childInode) - vfsd.InsertChild(&child.vfsd, rp.Component()) - - // Continue as usual now that nextVFSD is not nil. - nextVFSD = &child.vfsd + child = newDentry(childInode) + child.parent = d + child.name = name + dir.childCache[name] = child + } + if err := rp.CheckMount(&child.vfsd); err != nil { + return nil, nil, err } - nextInode := nextVFSD.Impl().(*dentry).inode - if nextInode.isSymlink() && rp.ShouldFollowSymlink() { - if err := rp.HandleSymlink(inode.impl.(*symlink).target); err != nil { + if child.inode.isSymlink() && rp.ShouldFollowSymlink() { + if err := rp.HandleSymlink(child.inode.impl.(*symlink).target); err != nil { return nil, nil, err } continue } rp.Advance() - return nextVFSD, nextInode, nil + return &child.vfsd, child.inode, nil } } @@ -515,5 +535,5 @@ func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { fs.mu.RLock() defer fs.mu.RUnlock() - return vfs.GenericPrependPath(vfsroot, vd, b) + return genericPrependPath(vfsroot, vd.Mount(), vd.Dentry().Impl().(*dentry), b) } diff --git a/pkg/sentry/fsimpl/ext/inode.go b/pkg/sentry/fsimpl/ext/inode.go index a39a37318..a98512350 100644 --- a/pkg/sentry/fsimpl/ext/inode.go +++ b/pkg/sentry/fsimpl/ext/inode.go @@ -136,7 +136,7 @@ func newInode(fs *filesystem, inodeNum uint32) (*inode, error) { } return &f.inode, nil case linux.ModeDirectory: - f, err := newDirectroy(inode, fs.sb.IncompatibleFeatures().DirentFileType) + f, err := newDirectory(inode, fs.sb.IncompatibleFeatures().DirentFileType) if err != nil { return nil, err } diff --git a/pkg/sentry/fsimpl/gofer/BUILD b/pkg/sentry/fsimpl/gofer/BUILD index 99d1e3f8f..acd061905 100644 --- a/pkg/sentry/fsimpl/gofer/BUILD +++ b/pkg/sentry/fsimpl/gofer/BUILD @@ -15,12 +15,24 @@ go_template_instance( }, ) +go_template_instance( + name = "fstree", + out = "fstree.go", + package = "gofer", + prefix = "generic", + template = "//pkg/sentry/vfs/genericfstree:generic_fstree", + types = { + "Dentry": "dentry", + }, +) + go_library( name = "gofer", srcs = [ "dentry_list.go", "directory.go", "filesystem.go", + "fstree.go", "gofer.go", "handle.go", "handle_unsafe.go", diff --git a/pkg/sentry/fsimpl/gofer/directory.go b/pkg/sentry/fsimpl/gofer/directory.go index 49d9f859b..d02691232 100644 --- a/pkg/sentry/fsimpl/gofer/directory.go +++ b/pkg/sentry/fsimpl/gofer/directory.go @@ -29,13 +29,25 @@ func (d *dentry) isDir() bool { return d.fileType() == linux.S_IFDIR } +// Preconditions: filesystem.renameMu must be locked. d.dirMu must be locked. +// d.isDir(). child must be a newly-created dentry that has never had a parent. +func (d *dentry) cacheNewChildLocked(child *dentry, name string) { + d.IncRef() // reference held by child on its parent + child.parent = d + child.name = name + if d.children == nil { + d.children = make(map[string]*dentry) + } + d.children[name] = child +} + // Preconditions: d.dirMu must be locked. d.isDir(). fs.opts.interop != // InteropModeShared. func (d *dentry) cacheNegativeChildLocked(name string) { - if d.negativeChildren == nil { - d.negativeChildren = make(map[string]struct{}) + if d.children == nil { + d.children = make(map[string]*dentry) } - d.negativeChildren[name] = struct{}{} + d.children[name] = nil } type directoryFD struct { @@ -80,34 +92,32 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba // Preconditions: d.isDir(). There exists at least one directoryFD representing d. func (d *dentry) getDirents(ctx context.Context) ([]vfs.Dirent, error) { - // 9P2000.L's readdir does not specify behavior in the presence of - // concurrent mutation of an iterated directory, so implementations may - // duplicate or omit entries in this case, which violates POSIX semantics. - // Thus we read all directory entries while holding d.dirMu to exclude - // directory mutations. (Note that it is impossible for the client to - // exclude concurrent mutation from other remote filesystem users. Since - // there is no way to detect if the server has incorrectly omitted - // directory entries, we simply assume that the server is well-behaved - // under InteropModeShared.) This is inconsistent with Linux (which appears - // to assume that directory fids have the correct semantics, and translates - // struct file_operations::readdir calls directly to readdir RPCs), but is - // consistent with VFS1. - // - // NOTE(b/135560623): In particular, some gofer implementations may not - // retain state between calls to Readdir, so may not provide a coherent - // directory stream across in the presence of mutation. - + // NOTE(b/135560623): 9P2000.L's readdir does not specify behavior in the + // presence of concurrent mutation of an iterated directory, so + // implementations may duplicate or omit entries in this case, which + // violates POSIX semantics. Thus we read all directory entries while + // holding d.dirMu to exclude directory mutations. (Note that it is + // impossible for the client to exclude concurrent mutation from other + // remote filesystem users. Since there is no way to detect if the server + // has incorrectly omitted directory entries, we simply assume that the + // server is well-behaved under InteropModeShared.) This is inconsistent + // with Linux (which appears to assume that directory fids have the correct + // semantics, and translates struct file_operations::readdir calls directly + // to readdir RPCs), but is consistent with VFS1. + + // filesystem.renameMu is needed for d.parent, and must be locked before + // dentry.dirMu. d.fs.renameMu.RLock() - defer d.fs.renameMu.RUnlock() d.dirMu.Lock() defer d.dirMu.Unlock() if d.dirents != nil { + d.fs.renameMu.RUnlock() return d.dirents, nil } // It's not clear if 9P2000.L's readdir is expected to return "." and "..", // so we generate them here. - parent := d.vfsd.ParentOrSelf().Impl().(*dentry) + parent := genericParentOrSelf(d) dirents := []vfs.Dirent{ { Name: ".", @@ -122,6 +132,7 @@ func (d *dentry) getDirents(ctx context.Context) ([]vfs.Dirent, error) { NextOff: 2, }, } + d.fs.renameMu.RUnlock() off := uint64(0) const count = 64 * 1024 // for consistency with the vfs1 client d.handleMu.RLock() diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index cd744bf5e..43e863c61 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -116,6 +116,8 @@ func putDentrySlice(ds *[]*dentry) { // Preconditions: fs.renameMu must be locked. d.dirMu must be locked. // !rp.Done(). If fs.opts.interop == InteropModeShared, then d's cached // metadata must be up to date. +// +// Postconditions: The returned dentry's cached metadata is up to date. func (fs *filesystem) stepLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry, ds **[]*dentry) (*dentry, error) { if !d.isDir() { return nil, syserror.ENOTDIR @@ -130,39 +132,42 @@ afterSymlink: return d, nil } if name == ".." { - parentVFSD, err := rp.ResolveParent(&d.vfsd) - if err != nil { + if isRoot, err := rp.CheckRoot(&d.vfsd); err != nil { + return nil, err + } else if isRoot || d.parent == nil { + rp.Advance() + return d, nil + } + // We must assume that d.parent is correct, because if d has been moved + // elsewhere in the remote filesystem so that its parent has changed, + // we have no way of determining its new parent's location in the + // filesystem. + // + // Call rp.CheckMount() before updating d.parent's metadata, since if + // we traverse to another mount then d.parent's metadata is irrelevant. + if err := rp.CheckMount(&d.parent.vfsd); err != nil { return nil, err } - parent := parentVFSD.Impl().(*dentry) - if fs.opts.interop == InteropModeShared { - // We must assume that parentVFSD is correct, because if d has been - // moved elsewhere in the remote filesystem so that its parent has - // changed, we have no way of determining its new parent's location - // in the filesystem. Get updated metadata for parentVFSD. - _, attrMask, attr, err := parent.file.getAttr(ctx, dentryAttrMask()) + if fs.opts.interop == InteropModeShared && d != d.parent { + _, attrMask, attr, err := d.parent.file.getAttr(ctx, dentryAttrMask()) if err != nil { return nil, err } - parent.updateFromP9Attrs(attrMask, &attr) + d.parent.updateFromP9Attrs(attrMask, &attr) } rp.Advance() - return parent, nil + return d.parent, nil } - childVFSD, err := rp.ResolveChild(&d.vfsd, name) - if err != nil { - return nil, err - } - // FIXME(jamieliu): Linux performs revalidation before mount lookup - // (fs/namei.c:lookup_fast() => __d_lookup_rcu(), d_revalidate(), - // __follow_mount_rcu()). - child, err := fs.revalidateChildLocked(ctx, rp.VirtualFilesystem(), d, name, childVFSD, ds) + child, err := fs.getChildLocked(ctx, rp.VirtualFilesystem(), d, name, ds) if err != nil { return nil, err } if child == nil { return nil, syserror.ENOENT } + if err := rp.CheckMount(&child.vfsd); err != nil { + return nil, err + } if child.isSymlink() && rp.ShouldFollowSymlink() { target, err := child.readlink(ctx, rp.Mount()) if err != nil { @@ -177,38 +182,37 @@ afterSymlink: return child, nil } -// revalidateChildLocked must be called after a call to parent.vfsd.Child(name) -// or vfs.ResolvingPath.ResolveChild(name) returns childVFSD (which may be -// nil) to verify that the returned child (or lack thereof) is correct. If no file -// exists at name, revalidateChildLocked returns (nil, nil). +// getChildLocked returns a dentry representing the child of parent with the +// given name. If no such child exists, getChildLocked returns (nil, nil). // // Preconditions: fs.renameMu must be locked. parent.dirMu must be locked. // parent.isDir(). name is not "." or "..". // -// Postconditions: If revalidateChildLocked returns a non-nil dentry, its -// cached metadata is up to date. -func (fs *filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.VirtualFilesystem, parent *dentry, name string, childVFSD *vfs.Dentry, ds **[]*dentry) (*dentry, error) { - if childVFSD != nil && fs.opts.interop != InteropModeShared { - // We have a cached dentry that is assumed to be correct. - return childVFSD.Impl().(*dentry), nil - } - // We either don't have a cached dentry or need to verify that it's still - // correct, either of which requires a remote lookup. Check if this name is - // valid before performing the lookup. +// Postconditions: If getChildLocked returns a non-nil dentry, its cached +// metadata is up to date. +func (fs *filesystem) getChildLocked(ctx context.Context, vfsObj *vfs.VirtualFilesystem, parent *dentry, name string, ds **[]*dentry) (*dentry, error) { if len(name) > maxFilenameLen { return nil, syserror.ENAMETOOLONG } - // Check if we've already cached this lookup with a negative result. - if _, ok := parent.negativeChildren[name]; ok { - return nil, nil + child, ok := parent.children[name] + if ok && fs.opts.interop != InteropModeShared { + // Whether child is nil or not, it is cached information that is + // assumed to be correct. + return child, nil } - // Perform the remote lookup. + // We either don't have cached information or need to verify that it's + // still correct, either of which requires a remote lookup. Check if this + // name is valid before performing the lookup. + return fs.revalidateChildLocked(ctx, vfsObj, parent, name, child, ds) +} + +// Preconditions: As for getChildLocked. +func (fs *filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.VirtualFilesystem, parent *dentry, name string, child *dentry, ds **[]*dentry) (*dentry, error) { qid, file, attrMask, attr, err := parent.file.walkGetAttrOne(ctx, name) if err != nil && err != syserror.ENOENT { return nil, err } - if childVFSD != nil { - child := childVFSD.Impl().(*dentry) + if child != nil { if !file.isNil() && qid.Path == child.ino { // The file at this path hasn't changed. Just update cached // metadata. @@ -219,9 +223,8 @@ func (fs *filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.Vir // The file at this path has changed or no longer exists. Remove // the stale dentry from the tree, and re-evaluate its caching // status (i.e. if it has 0 references, drop it). - vfsObj.ForceDeleteDentry(childVFSD) + vfsObj.InvalidateDentry(&child.vfsd) *ds = appendDentry(*ds, child) - childVFSD = nil } if file.isNil() { // No file exists at this path now. Cache the negative lookup if @@ -232,13 +235,12 @@ func (fs *filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.Vir return nil, nil } // Create a new dentry representing the file. - child, err := fs.newDentry(ctx, file, qid, attrMask, &attr) + child, err = fs.newDentry(ctx, file, qid, attrMask, &attr) if err != nil { file.close(ctx) return nil, err } - parent.IncRef() // reference held by child on its parent - parent.vfsd.InsertChild(&child.vfsd, name) + parent.cacheNewChildLocked(child, name) // For now, child has 0 references, so our caller should call // child.checkCachingLocked(). *ds = appendDentry(*ds, child) @@ -318,9 +320,6 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir if err := parent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { return err } - if parent.isDeleted() { - return syserror.ENOENT - } name := rp.Component() if name == "." || name == ".." { return syserror.EEXIST @@ -331,6 +330,9 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir if !dir && rp.MustBeDir() { return syserror.ENOENT } + if parent.isDeleted() { + return syserror.ENOENT + } mnt := rp.Mount() if err := mnt.CheckBeginWrite(); err != nil { return err @@ -348,7 +350,7 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir // it's used. return create(parent, name) } - if parent.vfsd.Child(name) != nil { + if child := parent.children[name]; child != nil { return syserror.EEXIST } // No cached dentry exists; however, there might still be an existing file @@ -356,10 +358,11 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir if err := create(parent, name); err != nil { return err } - if fs.opts.interop != InteropModeShared { - parent.touchCMtime() - } - delete(parent.negativeChildren, name) + parent.touchCMtime() + // Either parent.children[name] doesn't exist (in which case this is a + // no-op) or is nil (in which case this erases the now-stale information + // that the file doesn't exist). + delete(parent.children, name) parent.dirents = nil return nil } @@ -407,56 +410,55 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b defer mntns.DecRef() parent.dirMu.Lock() defer parent.dirMu.Unlock() - childVFSD := parent.vfsd.Child(name) - var child *dentry + child, ok := parent.children[name] + if ok && child == nil { + return syserror.ENOENT + } // We only need a dentry representing the file at name if it can be a mount - // point. If childVFSD is nil, then it can't be a mount point. If childVFSD - // is non-nil but stale, the actual file can't be a mount point either; we + // point. If child is nil, then it can't be a mount point. If child is + // non-nil but stale, the actual file can't be a mount point either; we // detect this case by just speculatively calling PrepareDeleteDentry and // only revalidating the dentry if that fails (indicating that the existing // dentry is a mount point). - if childVFSD != nil { - child = childVFSD.Impl().(*dentry) - if err := vfsObj.PrepareDeleteDentry(mntns, childVFSD); err != nil { - child, err = fs.revalidateChildLocked(ctx, vfsObj, parent, name, childVFSD, &ds) + if child != nil { + if err := vfsObj.PrepareDeleteDentry(mntns, &child.vfsd); err != nil { + if fs.opts.interop != InteropModeShared { + return err + } + child, err = fs.revalidateChildLocked(ctx, vfsObj, parent, name, child, &ds) if err != nil { return err } if child != nil { - childVFSD = &child.vfsd - if err := vfsObj.PrepareDeleteDentry(mntns, childVFSD); err != nil { + if err := vfsObj.PrepareDeleteDentry(mntns, &child.vfsd); err != nil { return err } - } else { - childVFSD = nil } } - } else if _, ok := parent.negativeChildren[name]; ok { - return syserror.ENOENT } flags := uint32(0) if dir { if child != nil && !child.isDir() { - vfsObj.AbortDeleteDentry(childVFSD) + vfsObj.AbortDeleteDentry(&child.vfsd) return syserror.ENOTDIR } flags = linux.AT_REMOVEDIR } else { if child != nil && child.isDir() { - vfsObj.AbortDeleteDentry(childVFSD) + vfsObj.AbortDeleteDentry(&child.vfsd) return syserror.EISDIR } if rp.MustBeDir() { - if childVFSD != nil { - vfsObj.AbortDeleteDentry(childVFSD) + if child != nil { + vfsObj.AbortDeleteDentry(&child.vfsd) } return syserror.ENOTDIR } } err = parent.file.unlinkAt(ctx, name, flags) if err != nil { - if childVFSD != nil { - vfsObj.AbortDeleteDentry(childVFSD) + if child != nil { + vfsObj.AbortDeleteDentry(&child.vfsd) } return err } @@ -467,10 +469,12 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b } parent.cacheNegativeChildLocked(name) parent.dirents = nil + } else { + delete(parent.children, name) } if child != nil { child.setDeleted() - vfsObj.CommitDeleteDentry(childVFSD) + vfsObj.CommitDeleteDentry(&child.vfsd) ds = appendDentry(ds, child) } return nil @@ -806,16 +810,14 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving // eligible for caching yet, so we don't need to append to a dentry slice.) child.refs = 1 // Insert the dentry into the tree. - d.IncRef() // reference held by child on its parent d - d.vfsd.InsertChild(&child.vfsd, name) + d.cacheNewChildLocked(child, name) if d.fs.opts.interop != InteropModeShared { - delete(d.negativeChildren, name) + d.touchCMtime() d.dirents = nil } // Finally, construct a file description representing the created file. var childVFSFD *vfs.FileDescription - mnt.IncRef() if useRegularFileFD { fd := ®ularFileFD{} if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &child.vfsd, &vfs.FileDescriptionOptions{ @@ -840,9 +842,6 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving } childVFSFD = &fd.vfsfd } - if d.fs.opts.interop != InteropModeShared { - d.touchCMtime() - } return childVFSFD, nil } @@ -902,7 +901,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa // directory, we need to check for write permission on it. oldParent.dirMu.Lock() defer oldParent.dirMu.Unlock() - renamed, err := fs.revalidateChildLocked(ctx, vfsObj, oldParent, oldName, oldParent.vfsd.Child(oldName), &ds) + renamed, err := fs.getChildLocked(ctx, vfsObj, oldParent, oldName, &ds) if err != nil { return err } @@ -910,7 +909,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa return syserror.ENOENT } if renamed.isDir() { - if renamed == newParent || renamed.vfsd.IsAncestorOf(&newParent.vfsd) { + if renamed == newParent || genericIsAncestorDentry(renamed, newParent) { return syserror.EINVAL } if oldParent != newParent { @@ -934,16 +933,17 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa if newParent.isDeleted() { return syserror.ENOENT } - replacedVFSD := newParent.vfsd.Child(newName) - var replaced *dentry + replaced := newParent.children[newName] // This is similar to unlinkAt, except: // - // - We revalidate the replaced dentry unconditionally for simplicity. + // - If a dentry exists for the file to be replaced, we revalidate it + // unconditionally (instead of only if PrepareRenameDentry fails) for + // simplicity. // // - If rp.MustBeDir(), then we need a dentry representing the replaced // file regardless to confirm that it's a directory. - if replacedVFSD != nil || rp.MustBeDir() { - replaced, err = fs.revalidateChildLocked(ctx, vfsObj, newParent, newName, replacedVFSD, &ds) + if replaced != nil || rp.MustBeDir() { + replaced, err = fs.getChildLocked(ctx, rp.VirtualFilesystem(), newParent, newName, &ds) if err != nil { return err } @@ -957,11 +957,12 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa return syserror.ENOTDIR } } - replacedVFSD = &replaced.vfsd - } else { - replacedVFSD = nil } } + var replacedVFSD *vfs.Dentry + if replaced != nil { + replacedVFSD = &replaced.vfsd + } if oldParent == newParent && oldName == newName { return nil @@ -978,7 +979,6 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa if fs.opts.interop != InteropModeShared { oldParent.cacheNegativeChildLocked(oldName) oldParent.dirents = nil - delete(newParent.negativeChildren, newName) newParent.dirents = nil if renamed.isDir() { oldParent.decLinks() @@ -987,8 +987,24 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa oldParent.touchCMtime() newParent.touchCMtime() renamed.touchCtime() + } else { + delete(oldParent.children, oldName) + } + if oldParent != newParent { + appendDentry(ds, oldParent) + newParent.IncRef() + } + renamed.parent = newParent + renamed.name = newName + if newParent.children == nil { + newParent.children = make(map[string]*dentry) + } + newParent.children[newName] = renamed + if replaced != nil { + replaced.setDeleted() + appendDentry(ds, replaced) } - vfsObj.CommitRenameReplaceDentry(&renamed.vfsd, &newParent.vfsd, newName, replacedVFSD) + vfsObj.CommitRenameReplaceDentry(&renamed.vfsd, replacedVFSD) return nil } @@ -1131,5 +1147,5 @@ func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { fs.renameMu.RLock() defer fs.renameMu.RUnlock() - return vfs.GenericPrependPath(vfsroot, vd, b) + return genericPrependPath(vfsroot, vd.Mount(), vd.Dentry().Impl().(*dentry), b) } diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 2485cdb53..293df2545 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -452,6 +452,16 @@ type dentry struct { // fs is the owning filesystem. fs is immutable. fs *filesystem + // parent is this dentry's parent directory. Each dentry holds a reference + // on its parent. If this dentry is a filesystem root, parent is nil. + // parent is protected by filesystem.renameMu. + parent *dentry + + // name is the name of this dentry in its parent. If this dentry is a + // filesystem root, name is the empty string. name is protected by + // filesystem.renameMu. + name string + // We don't support hard links, so each dentry maps 1:1 to an inode. // file is the unopened p9.File that backs this dentry. file is immutable. @@ -469,10 +479,15 @@ type dentry struct { dirMu sync.Mutex - // If this dentry represents a directory, and InteropModeShared is not in - // effect, negativeChildren is a set of child names in this directory that - // are known not to exist. negativeChildren is protected by dirMu. - negativeChildren map[string]struct{} + // If this dentry represents a directory, children contains: + // + // - Mappings of child filenames to dentries representing those children. + // + // - Mappings of child filenames that are known not to exist to nil + // dentries (only if InteropModeShared is not in effect). + // + // children is protected by dirMu. + children map[string]*dentry // If this dentry represents a directory, InteropModeShared is not in // effect, and dirents is not nil, it is a cache of all entries in the @@ -910,9 +925,9 @@ func (d *dentry) checkCachingLocked() { // Dentry has already been destroyed. return } - // Non-child dentries with zero references are no longer reachable by path - // resolution and should be dropped immediately. - if d.vfsd.Parent() == nil || d.vfsd.IsDisowned() { + // Deleted and invalidated dentries with zero references are no longer + // reachable by path resolution and should be dropped immediately. + if d.vfsd.IsDead() { if d.cached { d.fs.cachedDentries.Remove(d) d.fs.cachedDentriesLen-- @@ -937,28 +952,26 @@ func (d *dentry) checkCachingLocked() { d.fs.cachedDentries.Remove(victim) d.fs.cachedDentriesLen-- victim.cached = false - // victim.refs may have become non-zero from an earlier path - // resolution since it was inserted into fs.cachedDentries; see - // dentry.incRefLocked(). Either way, we brought - // fs.cachedDentriesLen back down to fs.opts.maxCachedDentries, so - // we don't loop. + // victim.refs may have become non-zero from an earlier path resolution + // since it was inserted into fs.cachedDentries. if atomic.LoadInt64(&victim.refs) == 0 { - if victimParentVFSD := victim.vfsd.Parent(); victimParentVFSD != nil { - victimParent := victimParentVFSD.Impl().(*dentry) - victimParent.dirMu.Lock() - if !victim.vfsd.IsDisowned() { - // victim can't be a mount point (in any mount - // namespace), since VFS holds references on mount - // points. - d.fs.vfsfs.VirtualFilesystem().ForceDeleteDentry(&victim.vfsd) + if victim.parent != nil { + victim.parent.dirMu.Lock() + if !victim.vfsd.IsDead() { + // Note that victim can't be a mount point (in any mount + // namespace), since VFS holds references on mount points. + d.fs.vfsfs.VirtualFilesystem().InvalidateDentry(&victim.vfsd) + delete(victim.parent.children, victim.name) // We're only deleting the dentry, not the file it // represents, so we don't need to update // victimParent.dirents etc. } - victimParent.dirMu.Unlock() + victim.parent.dirMu.Unlock() } victim.destroyLocked() } + // Whether or not victim was destroyed, we brought fs.cachedDentriesLen + // back down to fs.opts.maxCachedDentries, so we don't loop. } } @@ -1005,12 +1018,11 @@ func (d *dentry) destroyLocked() { d.fs.syncMu.Lock() delete(d.fs.dentries, d) d.fs.syncMu.Unlock() - // Drop the reference held by d on its parent. - if parentVFSD := d.vfsd.Parent(); parentVFSD != nil { - parent := parentVFSD.Impl().(*dentry) - // This is parent.DecRef() without recursive locking of d.fs.renameMu. - if refs := atomic.AddInt64(&parent.refs, -1); refs == 0 { - parent.checkCachingLocked() + // Drop the reference held by d on its parent without recursively locking + // d.fs.renameMu. + if d.parent != nil { + if refs := atomic.AddInt64(&d.parent.refs, -1); refs == 0 { + d.parent.checkCachingLocked() } else if refs < 0 { panic("gofer.dentry.DecRef() called without holding a reference") } diff --git a/pkg/sentry/fsimpl/gofer/gofer_test.go b/pkg/sentry/fsimpl/gofer/gofer_test.go index 82bc239db..4041fb252 100644 --- a/pkg/sentry/fsimpl/gofer/gofer_test.go +++ b/pkg/sentry/fsimpl/gofer/gofer_test.go @@ -48,8 +48,7 @@ func TestDestroyIdempotent(t *testing.T) { if err != nil { t.Fatalf("fs.newDentry(): %v", err) } - parent.IncRef() // reference held by child on its parent. - parent.vfsd.InsertChild(&child.vfsd, "child") + parent.cacheNewChildLocked(child, "child") child.checkCachingLocked() if got := atomic.LoadInt64(&child.refs); got != -1 { diff --git a/pkg/sentry/fsimpl/kernfs/BUILD b/pkg/sentry/fsimpl/kernfs/BUILD index b3d6299d0..ef34cb28a 100644 --- a/pkg/sentry/fsimpl/kernfs/BUILD +++ b/pkg/sentry/fsimpl/kernfs/BUILD @@ -3,6 +3,17 @@ load("//tools/go_generics:defs.bzl", "go_template_instance") licenses(["notice"]) +go_template_instance( + name = "fstree", + out = "fstree.go", + package = "kernfs", + prefix = "generic", + template = "//pkg/sentry/vfs/genericfstree:generic_fstree", + types = { + "Dentry": "Dentry", + }, +) + go_template_instance( name = "slot_list", out = "slot_list.go", @@ -21,6 +32,7 @@ go_library( "dynamic_bytes_file.go", "fd_impl_util.go", "filesystem.go", + "fstree.go", "inode_impl_util.go", "kernfs.go", "slot_list.go", diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index bfa786c88..e8a4670b8 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -129,7 +129,7 @@ func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirent // Handle "..". if fd.off == 1 { - parentInode := vfsd.ParentOrSelf().Impl().(*Dentry).inode + parentInode := genericParentOrSelf(vfsd.Impl().(*Dentry)).inode stat, err := parentInode.Stat(vfsFS, opts) if err != nil { return err diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index baf81b4db..01c23d192 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -56,25 +56,28 @@ afterSymlink: return vfsd, nil } if name == ".." { - nextVFSD, err := rp.ResolveParent(vfsd) - if err != nil { + if isRoot, err := rp.CheckRoot(vfsd); err != nil { + return nil, err + } else if isRoot || d.parent == nil { + rp.Advance() + return vfsd, nil + } + if err := rp.CheckMount(&d.parent.vfsd); err != nil { return nil, err } rp.Advance() - return nextVFSD, nil + return &d.parent.vfsd, nil } if len(name) > linux.NAME_MAX { return nil, syserror.ENAMETOOLONG } d.dirMu.Lock() - nextVFSD, err := rp.ResolveChild(vfsd, name) + next, err := fs.revalidateChildLocked(ctx, rp.VirtualFilesystem(), d, name, d.children[name]) + d.dirMu.Unlock() if err != nil { - d.dirMu.Unlock() return nil, err } - next, err := fs.revalidateChildLocked(ctx, rp.VirtualFilesystem(), d, name, nextVFSD) - d.dirMu.Unlock() - if err != nil { + if err := rp.CheckMount(&next.vfsd); err != nil { return nil, err } // Resolve any symlink at current path component. @@ -108,17 +111,17 @@ afterSymlink: // parent.dirMu must be locked. parent.isDir(). name is not "." or "..". // // Postconditions: Caller must call fs.processDeferredDecRefs*. -func (fs *Filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.VirtualFilesystem, parent *Dentry, name string, childVFSD *vfs.Dentry) (*Dentry, error) { - if childVFSD != nil { +func (fs *Filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.VirtualFilesystem, parent *Dentry, name string, child *Dentry) (*Dentry, error) { + if child != nil { // Cached dentry exists, revalidate. - child := childVFSD.Impl().(*Dentry) if !child.inode.Valid(ctx) { - vfsObj.ForceDeleteDentry(childVFSD) - fs.deferDecRef(childVFSD) // Reference from Lookup. - childVFSD = nil + delete(parent.children, name) + vfsObj.InvalidateDentry(&child.vfsd) + fs.deferDecRef(&child.vfsd) // Reference from Lookup. + child = nil } } - if childVFSD == nil { + if child == nil { // Dentry isn't cached; it either doesn't exist or failed // revalidation. Attempt to resolve it via Lookup. // @@ -126,15 +129,15 @@ func (fs *Filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.Vir // *(kernfs.)Dentry, not *vfs.Dentry, since (kernfs.)Filesystem assumes // that all dentries in the filesystem are (kernfs.)Dentry and performs // vfs.DentryImpl casts accordingly. - var err error - childVFSD, err = parent.inode.Lookup(ctx, name) + childVFSD, err := parent.inode.Lookup(ctx, name) if err != nil { return nil, err } // Reference on childVFSD dropped by a corresponding Valid. - parent.insertChildLocked(name, childVFSD) + child = childVFSD.Impl().(*Dentry) + parent.insertChildLocked(name, child) } - return childVFSD.Impl().(*Dentry), nil + return child, nil } // walkExistingLocked resolves rp to an existing file. @@ -203,14 +206,11 @@ func checkCreateLocked(ctx context.Context, rp *vfs.ResolvingPath, parentVFSD *v if len(pc) > linux.NAME_MAX { return "", syserror.ENAMETOOLONG } - childVFSD, err := rp.ResolveChild(parentVFSD, pc) - if err != nil { - return "", err - } - if childVFSD != nil { + // FIXME(gvisor.dev/issue/1193): Data race due to not holding dirMu. + if _, ok := parentVFSD.Impl().(*Dentry).children[pc]; ok { return "", syserror.EEXIST } - if parentVFSD.IsDisowned() { + if parentVFSD.IsDead() { return "", syserror.ENOENT } return pc, nil @@ -220,14 +220,14 @@ func checkCreateLocked(ctx context.Context, rp *vfs.ResolvingPath, parentVFSD *v // // Preconditions: Filesystem.mu must be locked for at least reading. func checkDeleteLocked(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry) error { - parentVFSD := vfsd.Parent() - if parentVFSD == nil { + parent := vfsd.Impl().(*Dentry).parent + if parent == nil { return syserror.EBUSY } - if parentVFSD.IsDisowned() { + if parent.vfsd.IsDead() { return syserror.ENOENT } - if err := parentVFSD.Impl().(*Dentry).inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { + if err := parent.inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { return err } return nil @@ -321,11 +321,11 @@ func (fs *Filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. return syserror.EPERM } - child, err := parentInode.NewLink(ctx, pc, d.inode) + childVFSD, err := parentInode.NewLink(ctx, pc, d.inode) if err != nil { return err } - parentVFSD.Impl().(*Dentry).InsertChild(pc, child) + parentVFSD.Impl().(*Dentry).InsertChild(pc, childVFSD.Impl().(*Dentry)) return nil } @@ -349,11 +349,11 @@ func (fs *Filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts v return err } defer rp.Mount().EndWrite() - child, err := parentInode.NewDir(ctx, pc, opts) + childVFSD, err := parentInode.NewDir(ctx, pc, opts) if err != nil { return err } - parentVFSD.Impl().(*Dentry).InsertChild(pc, child) + parentVFSD.Impl().(*Dentry).InsertChild(pc, childVFSD.Impl().(*Dentry)) return nil } @@ -377,11 +377,11 @@ func (fs *Filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v return err } defer rp.Mount().EndWrite() - new, err := parentInode.NewNode(ctx, pc, opts) + newVFSD, err := parentInode.NewNode(ctx, pc, opts) if err != nil { return err } - parentVFSD.Impl().(*Dentry).InsertChild(pc, new) + parentVFSD.Impl().(*Dentry).InsertChild(pc, newVFSD.Impl().(*Dentry)) return nil } @@ -449,11 +449,8 @@ afterTrailingSymlink: return nil, syserror.ENAMETOOLONG } // Determine whether or not we need to create a file. - childVFSD, err := rp.ResolveChild(parentVFSD, pc) - if err != nil { - return nil, err - } - if childVFSD == nil { + childVFSD, err := fs.stepExistingLocked(ctx, rp, parentVFSD) + if err == syserror.ENOENT { // Already checked for searchability above; now check for writability. if err := parentInode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite); err != nil { return nil, err @@ -463,21 +460,24 @@ afterTrailingSymlink: } defer rp.Mount().EndWrite() // Create and open the child. - child, err := parentInode.NewFile(ctx, pc, opts) + childVFSD, err = parentInode.NewFile(ctx, pc, opts) if err != nil { return nil, err } + child := childVFSD.Impl().(*Dentry) parentVFSD.Impl().(*Dentry).InsertChild(pc, child) - return child.Impl().(*Dentry).inode.Open(rp, child, opts) + return child.inode.Open(rp, childVFSD, opts) + } + if err != nil { + return nil, err } // Open existing file or follow symlink. if mustCreate { return nil, syserror.EEXIST } - childDentry := childVFSD.Impl().(*Dentry) - childInode := childDentry.inode - if rp.ShouldFollowSymlink() && childDentry.isSymlink() { - targetVD, targetPathname, err := childInode.Getlink(ctx) + child := childVFSD.Impl().(*Dentry) + if rp.ShouldFollowSymlink() && child.isSymlink() { + targetVD, targetPathname, err := child.inode.Getlink(ctx) if err != nil { return nil, err } @@ -496,10 +496,10 @@ afterTrailingSymlink: // symlink target. goto afterTrailingSymlink } - if err := childInode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { + if err := child.inode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { return nil, err } - return childInode.Open(rp, childVFSD, opts) + return child.inode.Open(rp, &child.vfsd, opts) } // ReadlinkAt implements vfs.FilesystemImpl.ReadlinkAt. @@ -526,15 +526,16 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa noReplace := opts.Flags&linux.RENAME_NOREPLACE != 0 fs.mu.Lock() - defer fs.mu.Lock() + defer fs.processDeferredDecRefsLocked() + defer fs.mu.Unlock() // Resolve the destination directory first to verify that it's on this // Mount. dstDirVFSD, dstDirInode, err := fs.walkParentDirLocked(ctx, rp) - fs.processDeferredDecRefsLocked() if err != nil { return err } + dstDir := dstDirVFSD.Impl().(*Dentry) mnt := rp.Mount() if mnt != oldParentVD.Mount() { return syserror.EXDEV @@ -547,9 +548,8 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa srcDirVFSD := oldParentVD.Dentry() srcDir := srcDirVFSD.Impl().(*Dentry) srcDir.dirMu.Lock() - src, err := fs.revalidateChildLocked(ctx, rp.VirtualFilesystem(), srcDir, oldName, srcDirVFSD.Child(oldName)) + src, err := fs.revalidateChildLocked(ctx, rp.VirtualFilesystem(), srcDir, oldName, srcDir.children[oldName]) srcDir.dirMu.Unlock() - fs.processDeferredDecRefsLocked() if err != nil { return err } @@ -561,7 +561,7 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa } // Can we create the dst dentry? - var dstVFSD *vfs.Dentry + var dst *Dentry pc, err := checkCreateLocked(ctx, rp, dstDirVFSD, dstDirInode) switch err { case nil: @@ -571,38 +571,51 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa // Won't overwrite existing node since RENAME_NOREPLACE was requested. return syserror.EEXIST } - dstVFSD, err = rp.ResolveChild(dstDirVFSD, pc) - if err != nil { + dst = dstDir.children[pc] + if dst == nil { panic(fmt.Sprintf("Child %q for parent Dentry %+v disappeared inside atomic section?", pc, dstDirVFSD)) } default: return err } + var dstVFSD *vfs.Dentry + if dst != nil { + dstVFSD = &dst.vfsd + } mntns := vfs.MountNamespaceFromContext(ctx) defer mntns.DecRef() virtfs := rp.VirtualFilesystem() - srcDirDentry := srcDirVFSD.Impl().(*Dentry) - dstDirDentry := dstDirVFSD.Impl().(*Dentry) - // We can't deadlock here due to lock ordering because we're protected from // concurrent renames by fs.mu held for writing. - srcDirDentry.dirMu.Lock() - defer srcDirDentry.dirMu.Unlock() - dstDirDentry.dirMu.Lock() - defer dstDirDentry.dirMu.Unlock() + srcDir.dirMu.Lock() + defer srcDir.dirMu.Unlock() + if srcDir != dstDir { + dstDir.dirMu.Lock() + defer dstDir.dirMu.Unlock() + } if err := virtfs.PrepareRenameDentry(mntns, srcVFSD, dstVFSD); err != nil { return err } - srcDirInode := srcDirDentry.inode - replaced, err := srcDirInode.Rename(ctx, srcVFSD.Name(), pc, srcVFSD, dstDirVFSD) + replaced, err := srcDir.inode.Rename(ctx, src.name, pc, srcVFSD, dstDirVFSD) if err != nil { virtfs.AbortRenameDentry(srcVFSD, dstVFSD) return err } - virtfs.CommitRenameReplaceDentry(srcVFSD, dstDirVFSD, pc, replaced) + delete(srcDir.children, src.name) + if srcDir != dstDir { + fs.deferDecRef(srcDirVFSD) + dstDir.IncRef() + } + src.parent = dstDir + src.name = pc + if dstDir.children == nil { + dstDir.children = make(map[string]*Dentry) + } + dstDir.children[pc] = src + virtfs.CommitRenameReplaceDentry(srcVFSD, replaced) return nil } @@ -622,14 +635,15 @@ func (fs *Filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error if err := checkDeleteLocked(ctx, rp, vfsd); err != nil { return err } - if !vfsd.Impl().(*Dentry).isDir() { + d := vfsd.Impl().(*Dentry) + if !d.isDir() { return syserror.ENOTDIR } if inode.HasChildren() { return syserror.ENOTEMPTY } virtfs := rp.VirtualFilesystem() - parentDentry := vfsd.Parent().Impl().(*Dentry) + parentDentry := d.parent parentDentry.dirMu.Lock() defer parentDentry.dirMu.Unlock() @@ -706,11 +720,11 @@ func (fs *Filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, targ return err } defer rp.Mount().EndWrite() - child, err := parentInode.NewSymlink(ctx, pc, target) + childVFSD, err := parentInode.NewSymlink(ctx, pc, target) if err != nil { return err } - parentVFSD.Impl().(*Dentry).InsertChild(pc, child) + parentVFSD.Impl().(*Dentry).InsertChild(pc, childVFSD.Impl().(*Dentry)) return nil } @@ -730,11 +744,12 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error if err := checkDeleteLocked(ctx, rp, vfsd); err != nil { return err } - if vfsd.Impl().(*Dentry).isDir() { + d := vfsd.Impl().(*Dentry) + if d.isDir() { return syserror.EISDIR } virtfs := rp.VirtualFilesystem() - parentDentry := vfsd.Parent().Impl().(*Dentry) + parentDentry := d.parent parentDentry.dirMu.Lock() defer parentDentry.dirMu.Unlock() mntns := vfs.MountNamespaceFromContext(ctx) @@ -818,5 +833,5 @@ func (fs *Filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, func (fs *Filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { fs.mu.RLock() defer fs.mu.RUnlock() - return vfs.GenericPrependPath(vfsroot, vd, b) + return genericPrependPath(vfsroot, vd.Mount(), vd.Dentry().Impl().(*Dentry), b) } diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 65f09af5d..9f526359e 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -370,7 +370,7 @@ func (o *OrderedChildren) Populate(d *Dentry, children map[string]*Dentry) uint3 if err := o.Insert(name, child.VFSDentry()); err != nil { panic(fmt.Sprintf("Collision when attempting to insert child %q (%+v) into %+v", name, child, d)) } - d.InsertChild(name, child.VFSDentry()) + d.InsertChild(name, child) } return links } diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index ad76b9f64..f5041824f 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -168,17 +168,22 @@ const ( // // Must be initialized by Init prior to first use. type Dentry struct { - refs.AtomicRefCount + vfsd vfs.Dentry - vfsd vfs.Dentry - inode Inode + refs.AtomicRefCount // flags caches useful information about the dentry from the inode. See the // dflags* consts above. Must be accessed by atomic ops. flags uint32 - // dirMu protects vfsd.children for directory dentries. - dirMu sync.Mutex + parent *Dentry + name string + + // dirMu protects children and the names of child Dentries. + dirMu sync.Mutex + children map[string]*Dentry + + inode Inode } // Init initializes this dentry. @@ -222,8 +227,8 @@ func (d *Dentry) DecRef() { func (d *Dentry) destroy() { d.inode.DecRef() // IncRef from Init. d.inode = nil - if parent := d.vfsd.Parent(); parent != nil { - parent.DecRef() // IncRef from Dentry.InsertChild. + if d.parent != nil { + d.parent.DecRef() // IncRef from Dentry.InsertChild. } } @@ -233,7 +238,7 @@ func (d *Dentry) destroy() { // updates the link count on d if required. // // Precondition: d must represent a directory inode. -func (d *Dentry) InsertChild(name string, child *vfs.Dentry) { +func (d *Dentry) InsertChild(name string, child *Dentry) { d.dirMu.Lock() d.insertChildLocked(name, child) d.dirMu.Unlock() @@ -243,13 +248,17 @@ func (d *Dentry) InsertChild(name string, child *vfs.Dentry) { // preconditions. // // Precondition: d.dirMu must be locked. -func (d *Dentry) insertChildLocked(name string, child *vfs.Dentry) { +func (d *Dentry) insertChildLocked(name string, child *Dentry) { if !d.isDir() { panic(fmt.Sprintf("InsertChild called on non-directory Dentry: %+v.", d)) } - vfsDentry := d.VFSDentry() - vfsDentry.IncRef() // DecRef in child's Dentry.destroy. - vfsDentry.InsertChild(child, name) + d.IncRef() // DecRef in child's Dentry.destroy. + child.parent = d + child.name = name + if d.children == nil { + d.children = make(map[string]*Dentry) + } + d.children[name] = child } // The Inode interface maps filesystem-level operations that operate on paths to diff --git a/pkg/sentry/fsimpl/proc/tasks_test.go b/pkg/sentry/fsimpl/proc/tasks_test.go index d0f97c137..19abb5034 100644 --- a/pkg/sentry/fsimpl/proc/tasks_test.go +++ b/pkg/sentry/fsimpl/proc/tasks_test.go @@ -415,36 +415,36 @@ func iterateDir(ctx context.Context, t *testing.T, s *testutil.System, fd *vfs.F if d.Name == "." || d.Name == ".." { continue } - childPath := path.Join(fd.MappedName(ctx), d.Name) + absPath := path.Join(fd.MappedName(ctx), d.Name) if d.Type == linux.DT_LNK { link, err := s.VFS.ReadlinkAt( ctx, auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: s.Root, Start: s.Root, Path: fspath.Parse(childPath)}, + &vfs.PathOperation{Root: s.Root, Start: s.Root, Path: fspath.Parse(absPath)}, ) if err != nil { - t.Errorf("vfsfs.ReadlinkAt(%v) failed: %v", childPath, err) + t.Errorf("vfsfs.ReadlinkAt(%v) failed: %v", absPath, err) } else { - t.Logf("Skipping symlink: /proc%s => %s", childPath, link) + t.Logf("Skipping symlink: %s => %s", absPath, link) } continue } - t.Logf("Opening: /proc%s", childPath) + t.Logf("Opening: %s", absPath) child, err := s.VFS.OpenAt( ctx, auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: s.Root, Start: s.Root, Path: fspath.Parse(childPath)}, + &vfs.PathOperation{Root: s.Root, Start: s.Root, Path: fspath.Parse(absPath)}, &vfs.OpenOptions{}, ) if err != nil { - t.Errorf("vfsfs.OpenAt(%v) failed: %v", childPath, err) + t.Errorf("vfsfs.OpenAt(%v) failed: %v", absPath, err) continue } defer child.DecRef() stat, err := child.Stat(ctx, vfs.StatOptions{}) if err != nil { - t.Errorf("Stat(%v) failed: %v", childPath, err) + t.Errorf("Stat(%v) failed: %v", absPath, err) } if got := linux.FileMode(stat.Mode).DirentType(); got != d.Type { t.Errorf("wrong file mode, stat: %v, dirent: %v", got, d.Type) diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD index 4e6cd3491..a2d9649e7 100644 --- a/pkg/sentry/fsimpl/tmpfs/BUILD +++ b/pkg/sentry/fsimpl/tmpfs/BUILD @@ -15,6 +15,17 @@ go_template_instance( }, ) +go_template_instance( + name = "fstree", + out = "fstree.go", + package = "tmpfs", + prefix = "generic", + template = "//pkg/sentry/vfs/genericfstree:generic_fstree", + types = { + "Dentry": "dentry", + }, +) + go_library( name = "tmpfs", srcs = [ @@ -22,6 +33,7 @@ go_library( "device_file.go", "directory.go", "filesystem.go", + "fstree.go", "named_pipe.go", "regular_file.go", "socket_file.go", diff --git a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go index 651912169..2fb5c4d84 100644 --- a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go +++ b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go @@ -438,13 +438,6 @@ func BenchmarkVFS2TmpfsMountStat(b *testing.B) { filePathBuilder.WriteByte('/') } - // Verify that we didn't create any directories under the mount - // point (i.e. they were all created on the submount). - firstDirName := fmt.Sprintf("%d", depth) - if child := mountPoint.Dentry().Child(firstDirName); child != nil { - b.Fatalf("created directory %q under root mount, not submount", firstDirName) - } - // Create the file that will be stat'd. fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ Root: root, diff --git a/pkg/sentry/fsimpl/tmpfs/directory.go b/pkg/sentry/fsimpl/tmpfs/directory.go index 45712c9b9..f2399981b 100644 --- a/pkg/sentry/fsimpl/tmpfs/directory.go +++ b/pkg/sentry/fsimpl/tmpfs/directory.go @@ -15,35 +15,77 @@ package tmpfs import ( + "sync/atomic" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) type directory struct { - inode inode + // Since directories can't be hard-linked, each directory can only be + // associated with a single dentry, which we can store in the directory + // struct. + dentry dentry + inode inode + + // childMap maps the names of the directory's children to their dentries. + // childMap is protected by filesystem.mu. + childMap map[string]*dentry - // childList is a list containing (1) child Dentries and (2) fake Dentries + // numChildren is len(childMap), but accessed using atomic memory + // operations to avoid locking in inode.statTo(). + numChildren int64 + + // childList is a list containing (1) child dentries and (2) fake dentries // (with inode == nil) that represent the iteration position of // directoryFDs. childList is used to support directoryFD.IterDirents() - // efficiently. childList is protected by filesystem.mu. + // efficiently. childList is protected by iterMu. + iterMu sync.Mutex childList dentryList } -func (fs *filesystem) newDirectory(creds *auth.Credentials, mode linux.FileMode) *inode { +func (fs *filesystem) newDirectory(creds *auth.Credentials, mode linux.FileMode) *directory { dir := &directory{} dir.inode.init(dir, fs, creds, linux.S_IFDIR|mode) dir.inode.nlink = 2 // from "." and parent directory or ".." for root - return &dir.inode + dir.dentry.inode = &dir.inode + dir.dentry.vfsd.Init(&dir.dentry) + return dir +} + +// Preconditions: filesystem.mu must be locked for writing. dir must not +// already contain a child with the given name. +func (dir *directory) insertChildLocked(child *dentry, name string) { + child.parent = &dir.dentry + child.name = name + if dir.childMap == nil { + dir.childMap = make(map[string]*dentry) + } + dir.childMap[name] = child + atomic.AddInt64(&dir.numChildren, 1) + dir.iterMu.Lock() + dir.childList.PushBack(child) + dir.iterMu.Unlock() +} + +// Preconditions: filesystem.mu must be locked for writing. +func (dir *directory) removeChildLocked(child *dentry) { + delete(dir.childMap, child.name) + atomic.AddInt64(&dir.numChildren, -1) + dir.iterMu.Lock() + dir.childList.Remove(child) + dir.iterMu.Unlock() } type directoryFD struct { fileDescription vfs.DirectoryFileDescriptionDefaultImpl - // Protected by filesystem.mu. + // Protected by directory.iterMu. iter *dentry off int64 } @@ -51,11 +93,10 @@ type directoryFD struct { // Release implements vfs.FileDescriptionImpl.Release. func (fd *directoryFD) Release() { if fd.iter != nil { - fs := fd.filesystem() dir := fd.inode().impl.(*directory) - fs.mu.Lock() + dir.iterMu.Lock() dir.childList.Remove(fd.iter) - fs.mu.Unlock() + dir.iterMu.Unlock() fd.iter = nil } } @@ -63,10 +104,13 @@ func (fd *directoryFD) Release() { // IterDirents implements vfs.FileDescriptionImpl.IterDirents. func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback) error { fs := fd.filesystem() - vfsd := fd.vfsfd.VirtualDentry().Dentry() + dir := fd.inode().impl.(*directory) - fs.mu.Lock() - defer fs.mu.Unlock() + // fs.mu is required to read d.parent and dentry.name. + fs.mu.RLock() + defer fs.mu.RUnlock() + dir.iterMu.Lock() + defer dir.iterMu.Unlock() fd.inode().touchAtime(fd.vfsfd.Mount()) @@ -74,15 +118,16 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba if err := cb.Handle(vfs.Dirent{ Name: ".", Type: linux.DT_DIR, - Ino: vfsd.Impl().(*dentry).inode.ino, + Ino: dir.inode.ino, NextOff: 1, }); err != nil { return err } fd.off++ } + if fd.off == 1 { - parentInode := vfsd.ParentOrSelf().Impl().(*dentry).inode + parentInode := genericParentOrSelf(&dir.dentry).inode if err := cb.Handle(vfs.Dirent{ Name: "..", Type: parentInode.direntType(), @@ -94,7 +139,6 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba fd.off++ } - dir := vfsd.Impl().(*dentry).inode.impl.(*directory) var child *dentry if fd.iter == nil { // Start iteration at the beginning of dir. @@ -109,7 +153,7 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba // Skip other directoryFD iterators. if child.inode != nil { if err := cb.Handle(vfs.Dirent{ - Name: child.vfsd.Name(), + Name: child.name, Type: child.inode.direntType(), Ino: child.inode.ino, NextOff: fd.off + 1, @@ -127,9 +171,9 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba // Seek implements vfs.FileDescriptionImpl.Seek. func (fd *directoryFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - fs := fd.filesystem() - fs.mu.Lock() - defer fs.mu.Unlock() + dir := fd.inode().impl.(*directory) + dir.iterMu.Lock() + defer dir.iterMu.Unlock() switch whence { case linux.SEEK_SET: @@ -157,8 +201,6 @@ func (fd *directoryFD) Seek(ctx context.Context, offset int64, whence int32) (in remChildren = offset - 2 } - dir := fd.inode().impl.(*directory) - // Ensure that fd.iter exists and is not linked into dir.childList. if fd.iter == nil { fd.iter = &dentry{} diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 452c4e2e0..5b62f9ebb 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -39,27 +39,43 @@ func (fs *filesystem) Sync(ctx context.Context) error { // // Preconditions: filesystem.mu must be locked. !rp.Done(). func stepLocked(rp *vfs.ResolvingPath, d *dentry) (*dentry, error) { - if !d.inode.isDir() { + dir, ok := d.inode.impl.(*directory) + if !ok { return nil, syserror.ENOTDIR } if err := d.inode.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { return nil, err } afterSymlink: - if len(rp.Component()) > linux.NAME_MAX { - return nil, syserror.ENAMETOOLONG + name := rp.Component() + if name == "." { + rp.Advance() + return d, nil } - nextVFSD, err := rp.ResolveComponent(&d.vfsd) - if err != nil { - return nil, err + if name == ".." { + if isRoot, err := rp.CheckRoot(&d.vfsd); err != nil { + return nil, err + } else if isRoot || d.parent == nil { + rp.Advance() + return d, nil + } + if err := rp.CheckMount(&d.parent.vfsd); err != nil { + return nil, err + } + rp.Advance() + return d.parent, nil } - if nextVFSD == nil { - // Since the Dentry tree is the sole source of truth for tmpfs, if it's - // not in the Dentry tree, it doesn't exist. + if len(name) > linux.NAME_MAX { + return nil, syserror.ENAMETOOLONG + } + child, ok := dir.childMap[name] + if !ok { return nil, syserror.ENOENT } - next := nextVFSD.Impl().(*dentry) - if symlink, ok := next.inode.impl.(*symlink); ok && rp.ShouldFollowSymlink() { + if err := rp.CheckMount(&child.vfsd); err != nil { + return nil, err + } + if symlink, ok := child.inode.impl.(*symlink); ok && rp.ShouldFollowSymlink() { // TODO(gvisor.dev/issue/1197): Symlink traversals updates // access time. if err := rp.HandleSymlink(symlink.target); err != nil { @@ -68,7 +84,7 @@ afterSymlink: goto afterSymlink // don't check the current directory again } rp.Advance() - return next, nil + return child, nil } // walkParentDirLocked resolves all but the last path component of rp to an @@ -80,7 +96,7 @@ afterSymlink: // fs/namei.c:path_parentat(). // // Preconditions: filesystem.mu must be locked. !rp.Done(). -func walkParentDirLocked(rp *vfs.ResolvingPath, d *dentry) (*dentry, error) { +func walkParentDirLocked(rp *vfs.ResolvingPath, d *dentry) (*directory, error) { for !rp.Final() { next, err := stepLocked(rp, d) if err != nil { @@ -88,10 +104,11 @@ func walkParentDirLocked(rp *vfs.ResolvingPath, d *dentry) (*dentry, error) { } d = next } - if !d.inode.isDir() { + dir, ok := d.inode.impl.(*directory) + if !ok { return nil, syserror.ENOTDIR } - return d, nil + return dir, nil } // resolveLocked resolves rp to an existing file. @@ -122,14 +139,14 @@ func resolveLocked(rp *vfs.ResolvingPath) (*dentry, error) { // // Preconditions: !rp.Done(). For the final path component in rp, // !rp.ShouldFollowSymlink(). -func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(parent *dentry, name string) error) error { +func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(parentDir *directory, name string) error) error { fs.mu.Lock() defer fs.mu.Unlock() - parent, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) + parentDir, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) if err != nil { return err } - if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { + if err := parentDir.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { return err } name := rp.Component() @@ -139,19 +156,15 @@ func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(pa if len(name) > linux.NAME_MAX { return syserror.ENAMETOOLONG } - // Call parent.vfsd.Child() instead of stepLocked() or rp.ResolveChild(), - // because if the child exists we want to return EEXIST immediately instead - // of attempting symlink/mount traversal. - if parent.vfsd.Child(name) != nil { + if _, ok := parentDir.childMap[name]; ok { return syserror.EEXIST } if !dir && rp.MustBeDir() { return syserror.ENOENT } - // In tmpfs, the only way to cause a dentry to be disowned is by removing - // it from the filesystem, so this check is equivalent to checking if - // parent has been removed. - if parent.vfsd.IsDisowned() { + // tmpfs never calls VFS.InvalidateDentry(), so parentDir.dentry can only + // be dead if it was deleted. + if parentDir.dentry.vfsd.IsDead() { return syserror.ENOENT } mnt := rp.Mount() @@ -159,10 +172,10 @@ func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(pa return err } defer mnt.EndWrite() - if err := create(parent, name); err != nil { + if err := create(parentDir, name); err != nil { return err } - parent.inode.touchCMtime() + parentDir.inode.touchCMtime() return nil } @@ -201,17 +214,17 @@ func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, op func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { fs.mu.RLock() defer fs.mu.RUnlock() - d, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) + dir, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) if err != nil { return nil, err } - d.IncRef() - return &d.vfsd, nil + dir.dentry.IncRef() + return &dir.dentry.vfsd, nil } // LinkAt implements vfs.FilesystemImpl.LinkAt. func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry) error { - return fs.doCreateAt(rp, false /* dir */, func(parent *dentry, name string) error { + return fs.doCreateAt(rp, false /* dir */, func(parentDir *directory, name string) error { if rp.Mount() != vd.Mount() { return syserror.EXDEV } @@ -226,30 +239,27 @@ func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. return syserror.EMLINK } d.inode.incLinksLocked() - child := fs.newDentry(d.inode) - parent.vfsd.InsertChild(&child.vfsd, name) - parent.inode.impl.(*directory).childList.PushBack(child) + parentDir.insertChildLocked(fs.newDentry(d.inode), name) return nil }) } // MkdirAt implements vfs.FilesystemImpl.MkdirAt. func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error { - return fs.doCreateAt(rp, true /* dir */, func(parent *dentry, name string) error { - if parent.inode.nlink == maxLinks { + return fs.doCreateAt(rp, true /* dir */, func(parentDir *directory, name string) error { + if parentDir.inode.nlink == maxLinks { return syserror.EMLINK } - parent.inode.incLinksLocked() // from child's ".." - child := fs.newDentry(fs.newDirectory(rp.Credentials(), opts.Mode)) - parent.vfsd.InsertChild(&child.vfsd, name) - parent.inode.impl.(*directory).childList.PushBack(child) + parentDir.inode.incLinksLocked() // from child's ".." + childDir := fs.newDirectory(rp.Credentials(), opts.Mode) + parentDir.insertChildLocked(&childDir.dentry, name) return nil }) } // MknodAt implements vfs.FilesystemImpl.MknodAt. func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MknodOptions) error { - return fs.doCreateAt(rp, false /* dir */, func(parent *dentry, name string) error { + return fs.doCreateAt(rp, false /* dir */, func(parentDir *directory, name string) error { var childInode *inode switch opts.Mode.FileType() { case 0, linux.S_IFREG: @@ -266,8 +276,7 @@ func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v return syserror.EINVAL } child := fs.newDentry(childInode) - parent.vfsd.InsertChild(&child.vfsd, name) - parent.inode.impl.(*directory).childList.PushBack(child) + parentDir.insertChildLocked(child, name) return nil }) } @@ -306,12 +315,12 @@ func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf return start.open(ctx, rp, &opts, false /* afterCreate */) } afterTrailingSymlink: - parent, err := walkParentDirLocked(rp, start) + parentDir, err := walkParentDirLocked(rp, start) if err != nil { return nil, err } // Check for search permission in the parent directory. - if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { + if err := parentDir.inode.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { return nil, err } // Reject attempts to open directories with O_CREAT. @@ -322,11 +331,14 @@ afterTrailingSymlink: if name == "." || name == ".." { return nil, syserror.EISDIR } + if len(name) > linux.NAME_MAX { + return nil, syserror.ENAMETOOLONG + } // Determine whether or not we need to create a file. - child, err := stepLocked(rp, parent) - if err == syserror.ENOENT { + child, ok := parentDir.childMap[name] + if !ok { // Already checked for searchability above; now check for writability. - if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil { + if err := parentDir.inode.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil { return nil, err } if err := rp.Mount().CheckBeginWrite(); err != nil { @@ -335,21 +347,26 @@ afterTrailingSymlink: defer rp.Mount().EndWrite() // Create and open the child. child := fs.newDentry(fs.newRegularFile(rp.Credentials(), opts.Mode)) - parent.vfsd.InsertChild(&child.vfsd, name) - parent.inode.impl.(*directory).childList.PushBack(child) + parentDir.insertChildLocked(child, name) fd, err := child.open(ctx, rp, &opts, true) if err != nil { return nil, err } - parent.inode.touchCMtime() + parentDir.inode.touchCMtime() return fd, nil } - if err != nil { + // Is the file mounted over? + if err := rp.CheckMount(&child.vfsd); err != nil { return nil, err } // Do we need to resolve a trailing symlink? - if !rp.Done() { - start = parent + if symlink, ok := child.inode.impl.(*symlink); ok && rp.ShouldFollowSymlink() { + // TODO(gvisor.dev/issue/1197): Symlink traversals updates + // access time. + if err := rp.HandleSymlink(symlink.target); err != nil { + return nil, err + } + start = &parentDir.dentry goto afterTrailingSymlink } // Open existing file. @@ -428,7 +445,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa // Resolve newParent first to verify that it's on this Mount. fs.mu.Lock() defer fs.mu.Unlock() - newParent, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) + newParentDir, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) if err != nil { return err } @@ -445,23 +462,22 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa } defer mnt.EndWrite() - oldParent := oldParentVD.Dentry().Impl().(*dentry) - if err := oldParent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { + oldParentDir := oldParentVD.Dentry().Impl().(*dentry).inode.impl.(*directory) + if err := oldParentDir.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { return err } - // Call vfs.Dentry.Child() instead of stepLocked() or rp.ResolveChild(), - // because if the existing child is a symlink or mount point then we want - // to rename over it rather than follow it. - renamedVFSD := oldParent.vfsd.Child(oldName) - if renamedVFSD == nil { + renamed, ok := oldParentDir.childMap[oldName] + if !ok { return syserror.ENOENT } - renamed := renamedVFSD.Impl().(*dentry) + // Note that we don't need to call rp.CheckMount(), since if renamed is a + // mount point then we want to rename the mount point, not anything in the + // mounted filesystem. if renamed.inode.isDir() { - if renamed == newParent || renamedVFSD.IsAncestorOf(&newParent.vfsd) { + if renamed == &newParentDir.dentry || genericIsAncestorDentry(renamed, &newParentDir.dentry) { return syserror.EINVAL } - if oldParent != newParent { + if oldParentDir != newParentDir { // Writability is needed to change renamed's "..". if err := renamed.inode.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil { return err @@ -473,18 +489,17 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa } } - if err := newParent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { + if err := newParentDir.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { return err } - replacedVFSD := newParent.vfsd.Child(newName) - var replaced *dentry - if replacedVFSD != nil { - replaced = replacedVFSD.Impl().(*dentry) - if replaced.inode.isDir() { + replaced, ok := newParentDir.childMap[newName] + if ok { + replacedDir, ok := replaced.inode.impl.(*directory) + if ok { if !renamed.inode.isDir() { return syserror.EISDIR } - if replaced.vfsd.HasChildren() { + if len(replacedDir.childMap) != 0 { return syserror.ENOTEMPTY } } else { @@ -496,11 +511,13 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa } } } else { - if renamed.inode.isDir() && newParent.inode.nlink == maxLinks { + if renamed.inode.isDir() && newParentDir.inode.nlink == maxLinks { return syserror.EMLINK } } - if newParent.vfsd.IsDisowned() { + // tmpfs never calls VFS.InvalidateDentry(), so newParentDir.dentry can + // only be dead if it was deleted. + if newParentDir.dentry.vfsd.IsDead() { return syserror.ENOENT } @@ -508,36 +525,38 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa // simplicity, under the assumption that applications are not intentionally // doing noop renames expecting them to succeed where non-noop renames // would fail. - if renamedVFSD == replacedVFSD { + if renamed == replaced { return nil } vfsObj := rp.VirtualFilesystem() - oldParentDir := oldParent.inode.impl.(*directory) - newParentDir := newParent.inode.impl.(*directory) mntns := vfs.MountNamespaceFromContext(ctx) defer mntns.DecRef() - if err := vfsObj.PrepareRenameDentry(mntns, renamedVFSD, replacedVFSD); err != nil { + var replacedVFSD *vfs.Dentry + if replaced != nil { + replacedVFSD = &replaced.vfsd + } + if err := vfsObj.PrepareRenameDentry(mntns, &renamed.vfsd, replacedVFSD); err != nil { return err } if replaced != nil { - newParentDir.childList.Remove(replaced) + newParentDir.removeChildLocked(replaced) if replaced.inode.isDir() { - newParent.inode.decLinksLocked() // from replaced's ".." + newParentDir.inode.decLinksLocked() // from replaced's ".." } replaced.inode.decLinksLocked() } - oldParentDir.childList.Remove(renamed) - newParentDir.childList.PushBack(renamed) - if renamed.inode.isDir() { - oldParent.inode.decLinksLocked() - newParent.inode.incLinksLocked() + oldParentDir.removeChildLocked(renamed) + newParentDir.insertChildLocked(renamed, newName) + vfsObj.CommitRenameReplaceDentry(&renamed.vfsd, replacedVFSD) + oldParentDir.inode.touchCMtime() + if oldParentDir != newParentDir { + if renamed.inode.isDir() { + oldParentDir.inode.decLinksLocked() + newParentDir.inode.incLinksLocked() + } + newParentDir.inode.touchCMtime() } - oldParent.inode.touchCMtime() - newParent.inode.touchCMtime() renamed.inode.touchCtime() - // TODO(gvisor.dev/issue/1197): Update timestamps and parent directory - // sizes. - vfsObj.CommitRenameReplaceDentry(renamedVFSD, &newParent.vfsd, newName, replacedVFSD) return nil } @@ -545,11 +564,11 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error { fs.mu.Lock() defer fs.mu.Unlock() - parent, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) + parentDir, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) if err != nil { return err } - if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { + if err := parentDir.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { return err } name := rp.Component() @@ -559,15 +578,15 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error if name == ".." { return syserror.ENOTEMPTY } - childVFSD := parent.vfsd.Child(name) - if childVFSD == nil { + child, ok := parentDir.childMap[name] + if !ok { return syserror.ENOENT } - child := childVFSD.Impl().(*dentry) - if !child.inode.isDir() { + childDir, ok := child.inode.impl.(*directory) + if !ok { return syserror.ENOTDIR } - if childVFSD.HasChildren() { + if len(childDir.childMap) != 0 { return syserror.ENOTEMPTY } mnt := rp.Mount() @@ -578,14 +597,14 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error vfsObj := rp.VirtualFilesystem() mntns := vfs.MountNamespaceFromContext(ctx) defer mntns.DecRef() - if err := vfsObj.PrepareDeleteDentry(mntns, childVFSD); err != nil { + if err := vfsObj.PrepareDeleteDentry(mntns, &child.vfsd); err != nil { return err } - parent.inode.impl.(*directory).childList.Remove(child) - parent.inode.decLinksLocked() // from child's ".." + parentDir.removeChildLocked(child) + parentDir.inode.decLinksLocked() // from child's ".." child.inode.decLinksLocked() - vfsObj.CommitDeleteDentry(childVFSD) - parent.inode.touchCMtime() + vfsObj.CommitDeleteDentry(&child.vfsd) + parentDir.inode.touchCMtime() return nil } @@ -627,10 +646,9 @@ func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linu // SymlinkAt implements vfs.FilesystemImpl.SymlinkAt. func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, target string) error { - return fs.doCreateAt(rp, false /* dir */, func(parent *dentry, name string) error { + return fs.doCreateAt(rp, false /* dir */, func(parentDir *directory, name string) error { child := fs.newDentry(fs.newSymlink(rp.Credentials(), target)) - parent.vfsd.InsertChild(&child.vfsd, name) - parent.inode.impl.(*directory).childList.PushBack(child) + parentDir.insertChildLocked(child, name) return nil }) } @@ -639,22 +657,21 @@ func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, targ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error { fs.mu.Lock() defer fs.mu.Unlock() - parent, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) + parentDir, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) if err != nil { return err } - if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { + if err := parentDir.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { return err } name := rp.Component() if name == "." || name == ".." { return syserror.EISDIR } - childVFSD := parent.vfsd.Child(name) - if childVFSD == nil { + child, ok := parentDir.childMap[name] + if !ok { return syserror.ENOENT } - child := childVFSD.Impl().(*dentry) if child.inode.isDir() { return syserror.EISDIR } @@ -669,13 +686,13 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error vfsObj := rp.VirtualFilesystem() mntns := vfs.MountNamespaceFromContext(ctx) defer mntns.DecRef() - if err := vfsObj.PrepareDeleteDentry(mntns, childVFSD); err != nil { + if err := vfsObj.PrepareDeleteDentry(mntns, &child.vfsd); err != nil { return err } - parent.inode.impl.(*directory).childList.Remove(child) + parentDir.removeChildLocked(child) child.inode.decLinksLocked() - vfsObj.CommitDeleteDentry(childVFSD) - parent.inode.touchCMtime() + vfsObj.CommitDeleteDentry(&child.vfsd) + parentDir.inode.touchCMtime() return nil } @@ -743,5 +760,5 @@ func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { fs.mu.RLock() defer fs.mu.RUnlock() - return vfs.GenericPrependPath(vfsroot, vd, b) + return genericPrependPath(vfsroot, vd.Mount(), vd.Dentry().Impl().(*dentry), b) } diff --git a/pkg/sentry/fsimpl/tmpfs/stat_test.go b/pkg/sentry/fsimpl/tmpfs/stat_test.go index d4f59ee5b..60c2c980e 100644 --- a/pkg/sentry/fsimpl/tmpfs/stat_test.go +++ b/pkg/sentry/fsimpl/tmpfs/stat_test.go @@ -71,9 +71,15 @@ func TestStatAfterCreate(t *testing.T) { t.Errorf("got btime %d, want 0", got.Btime.ToNsec()) } - // Size should be 0. - if got.Size != 0 { - t.Errorf("got size %d, want 0", got.Size) + // Size should be 0 (except for directories, which make up a size + // of 20 per entry, including the "." and ".." entries present in + // otherwise-empty directories). + wantSize := uint64(0) + if typ == "dir" { + wantSize = 40 + } + if got.Size != wantSize { + t.Errorf("got size %d, want %d", got.Size, wantSize) } // Nlink should be 1 for files, 2 for dirs. diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 82c709b43..efc931468 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -12,16 +12,19 @@ // See the License for the specific language governing permissions and // limitations under the License. -// Package tmpfs provides a filesystem implementation that behaves like tmpfs: -// the Dentry tree is the sole source of truth for the state of the filesystem. +// Package tmpfs provides an in-memory filesystem whose contents are +// application-mutable, consistent with Linux's tmpfs. // // Lock order: // // filesystem.mu // inode.mu // regularFileFD.offMu +// *** "memmap.Mappable locks" below this point // regularFile.mapsMu +// *** "memmap.Mappable locks taken by Translate" below this point // regularFile.dataMu +// directory.iterMu package tmpfs import ( @@ -41,6 +44,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/vfs/memxattr" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // Name is the default filesystem name. @@ -112,18 +116,18 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt fs.vfsfs.Init(vfsObj, newFSType, &fs) - var root *inode + var root *dentry switch rootFileType { case linux.S_IFREG: - root = fs.newRegularFile(creds, 0777) + root = fs.newDentry(fs.newRegularFile(creds, 0777)) case linux.S_IFLNK: - root = fs.newSymlink(creds, tmpfsOpts.RootSymlinkTarget) + root = fs.newDentry(fs.newSymlink(creds, tmpfsOpts.RootSymlinkTarget)) case linux.S_IFDIR: - root = fs.newDirectory(creds, 01777) + root = &fs.newDirectory(creds, 01777).dentry default: return nil, nil, fmt.Errorf("invalid tmpfs root file type: %#o", rootFileType) } - return &fs.vfsfs, &fs.newDentry(root).vfsd, nil + return &fs.vfsfs, &root.vfsd, nil } // Release implements vfs.FilesystemImpl.Release. @@ -134,20 +138,29 @@ func (fs *filesystem) Release() { type dentry struct { vfsd vfs.Dentry + // parent is this dentry's parent directory. Each referenced dentry holds a + // reference on parent.dentry. If this dentry is a filesystem root, parent + // is nil. parent is protected by filesystem.mu. + parent *dentry + + // name is the name of this dentry in its parent. If this dentry is a + // filesystem root, name is the empty string. name is protected by + // filesystem.mu. + name string + + // dentryEntry (ugh) links dentries into their parent directory.childList. + dentryEntry + // inode is the inode represented by this dentry. Multiple Dentries may // share a single non-directory inode (with hard links). inode is // immutable. - inode *inode - + // // tmpfs doesn't count references on dentries; because the dentry tree is // the sole source of truth, it is by definition always consistent with the // state of the filesystem. However, it does count references on inodes, // because inode resources are released when all references are dropped. - // (tmpfs doesn't really have resources to release, but we implement - // reference counting because tmpfs regular files will.) - - // dentryEntry (ugh) links dentries into their parent directory.childList. - dentryEntry + // dentry therefore forwards reference counting directly to inode. + inode *inode } func (fs *filesystem) newDentry(inode *inode) *dentry { @@ -207,10 +220,6 @@ type inode struct { ctime int64 // nanoseconds mtime int64 // nanoseconds - // Only meaningful for device special files. - rdevMajor uint32 - rdevMinor uint32 - // Advisory file locks, which lock at the inode level. locks lock.FileLocks @@ -230,7 +239,7 @@ func (i *inode) init(impl interface{}, fs *filesystem, creds *auth.Credentials, i.gid = uint32(creds.EffectiveKGID) i.ino = atomic.AddUint64(&fs.nextInoMinusOne, 1) // Tmpfs creation sets atime, ctime, and mtime to current time. - now := i.clock.Now().Nanoseconds() + now := fs.clock.Now().Nanoseconds() i.atime = now i.ctime = now i.mtime = now @@ -283,14 +292,10 @@ func (i *inode) tryIncRef() bool { func (i *inode) decRef() { if refs := atomic.AddInt64(&i.refs, -1); refs == 0 { if regFile, ok := i.impl.(*regularFile); ok { - // Hold inode.mu and regFile.dataMu while mutating - // size. - i.mu.Lock() - regFile.dataMu.Lock() + // Release memory used by regFile to store data. Since regFile is + // no longer usable, we don't need to grab any locks or update any + // metadata. regFile.data.DropAll(regFile.memFile) - atomic.StoreUint64(®File.size, 0) - regFile.dataMu.Unlock() - i.mu.Unlock() } } else if refs < 0 { panic("tmpfs.inode.decRef() called without holding a reference") @@ -310,15 +315,15 @@ func (i *inode) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes) e // a concurrent modification), so we do not require holding inode.mu. func (i *inode) statTo(stat *linux.Statx) { stat.Mask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_NLINK | - linux.STATX_UID | linux.STATX_GID | linux.STATX_INO | linux.STATX_ATIME | - linux.STATX_BTIME | linux.STATX_CTIME | linux.STATX_MTIME - stat.Blksize = 1 // usermem.PageSize in tmpfs + linux.STATX_UID | linux.STATX_GID | linux.STATX_INO | linux.STATX_SIZE | + linux.STATX_BLOCKS | linux.STATX_ATIME | linux.STATX_CTIME | + linux.STATX_MTIME + stat.Blksize = usermem.PageSize stat.Nlink = atomic.LoadUint32(&i.nlink) stat.UID = atomic.LoadUint32(&i.uid) stat.GID = atomic.LoadUint32(&i.gid) stat.Mode = uint16(atomic.LoadUint32(&i.mode)) stat.Ino = i.ino - // Linux's tmpfs has no concept of btime, so zero-value is returned. stat.Atime = linux.NsecToStatxTimestamp(i.atime) stat.Ctime = linux.NsecToStatxTimestamp(i.ctime) stat.Mtime = linux.NsecToStatxTimestamp(i.mtime) @@ -327,19 +332,22 @@ func (i *inode) statTo(stat *linux.Statx) { case *regularFile: stat.Mask |= linux.STATX_SIZE | linux.STATX_BLOCKS stat.Size = uint64(atomic.LoadUint64(&impl.size)) - // In tmpfs, this will be FileRangeSet.Span() / 512 (but also cached in - // a uint64 accessed using atomic memory operations to avoid taking - // locks). + // TODO(jamieliu): This should be impl.data.Span() / 512, but this is + // too expensive to compute here. Cache it in regularFile. stat.Blocks = allocatedBlocksForSize(stat.Size) + case *directory: + // "20" is mm/shmem.c:BOGO_DIRENT_SIZE. + stat.Size = 20 * (2 + uint64(atomic.LoadInt64(&impl.numChildren))) + // stat.Blocks is 0. case *symlink: - stat.Mask |= linux.STATX_SIZE | linux.STATX_BLOCKS stat.Size = uint64(len(impl.target)) - stat.Blocks = allocatedBlocksForSize(stat.Size) + // stat.Blocks is 0. + case *namedPipe, *socketFile: + // stat.Size and stat.Blocks are 0. case *deviceFile: + // stat.Size and stat.Blocks are 0. stat.RdevMajor = impl.major stat.RdevMinor = impl.minor - case *socketFile, *directory, *namedPipe: - // Nothing to do. default: panic(fmt.Sprintf("unknown inode type: %T", i.impl)) } diff --git a/pkg/sentry/vfs/dentry.go b/pkg/sentry/vfs/dentry.go index 35b208721..8624dbd5d 100644 --- a/pkg/sentry/vfs/dentry.go +++ b/pkg/sentry/vfs/dentry.go @@ -15,34 +15,17 @@ package vfs import ( - "fmt" "sync/atomic" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) -// Dentry represents a node in a Filesystem tree which may represent a file. +// Dentry represents a node in a Filesystem tree at which a file exists. // // Dentries are reference-counted. Unless otherwise specified, all Dentry // methods require that a reference is held. // -// A Dentry transitions through up to 3 different states through its lifetime: -// -// - Dentries are initially "independent". Independent Dentries have no parent, -// and consequently no name. -// -// - Dentry.InsertChild() causes an independent Dentry to become a "child" of -// another Dentry. A child node has a parent node, and a name in that parent, -// both of which are mutable by DentryMoveChild(). Each child Dentry's name is -// unique within its parent. -// -// - Dentry.RemoveChild() causes a child Dentry to become "disowned". A -// disowned Dentry can still refer to its former parent and its former name in -// said parent, but the disowned Dentry is no longer reachable from its parent, -// and a new Dentry with the same name may become a child of the parent. (This -// is analogous to a struct dentry being "unhashed" in Linux.) -// // Dentry is loosely analogous to Linux's struct dentry, but: // // - VFS does not associate Dentries with inodes. gVisor interacts primarily @@ -57,9 +40,6 @@ import ( // and/or FileDescriptionImpl methods in gVisor's VFS. Filesystems that do // support inodes may store appropriate state in implementations of DentryImpl. // -// - VFS does not provide synchronization for mutable Dentry fields, other than -// mount-related ones. -// // - VFS does not require that Dentries are instantiated for all paths accessed // through VFS, only those that are tracked beyond the scope of a single // Filesystem operation. This includes file descriptions, mount points, mount @@ -67,6 +47,10 @@ import ( // of Dentries for operations on mutable remote filesystems that can't actually // cache any state in the Dentry. // +// - VFS does not track filesystem structure (i.e. relationships between +// Dentries), since both the relevant state and synchronization are +// filesystem-specific. +// // - For the reasons above, VFS is not directly responsible for managing Dentry // lifetime. Dentry reference counts only indicate the extent to which VFS // requires Dentries to exist; Filesystems may elect to cache or discard @@ -74,36 +58,23 @@ import ( // // +stateify savable type Dentry struct { - // parent is this Dentry's parent in this Filesystem. If this Dentry is - // independent, parent is nil. - parent *Dentry - - // name is this Dentry's name in parent. - name string + // mu synchronizes deletion/invalidation and mounting over this Dentry. + mu sync.Mutex `state:"nosave"` - flags uint32 + // dead is true if the file represented by this Dentry has been deleted (by + // CommitDeleteDentry or CommitRenameReplaceDentry) or invalidated (by + // InvalidateDentry). dead is protected by mu. + dead bool // mounts is the number of Mounts for which this Dentry is Mount.point. // mounts is accessed using atomic memory operations. mounts uint32 - // children are child Dentries. - children map[string]*Dentry - - // mu synchronizes disowning and mounting over this Dentry. - mu sync.Mutex `state:"nosave"` - // impl is the DentryImpl associated with this Dentry. impl is immutable. // This should be the last field in Dentry. impl DentryImpl } -const ( - // dflagsDisownedMask is set in Dentry.flags if the Dentry has been - // disowned. - dflagsDisownedMask = 1 << iota -) - // Init must be called before first use of d. func (d *Dentry) Init(impl DentryImpl) { d.impl = impl @@ -134,20 +105,6 @@ type DentryImpl interface { DecRef() } -// IsDisowned returns true if d is disowned. -func (d *Dentry) IsDisowned() bool { - return atomic.LoadUint32(&d.flags)&dflagsDisownedMask != 0 -} - -// Preconditions: !d.IsDisowned(). -func (d *Dentry) setDisowned() { - atomic.AddUint32(&d.flags, dflagsDisownedMask) -} - -func (d *Dentry) isMounted() bool { - return atomic.LoadUint32(&d.mounts) != 0 -} - // IncRef increments d's reference count. func (d *Dentry) IncRef() { d.impl.IncRef() @@ -164,104 +121,26 @@ func (d *Dentry) DecRef() { d.impl.DecRef() } -// These functions are exported so that filesystem implementations can use -// them. The vfs package, and users of VFS, should not call these functions. -// Unless otherwise specified, these methods require that there are no -// concurrent mutators of d. - -// Name returns d's name in its parent in its owning Filesystem. If d is -// independent, Name returns an empty string. -func (d *Dentry) Name() string { - return d.name -} - -// Parent returns d's parent in its owning Filesystem. It does not take a -// reference on the returned Dentry. If d is independent, Parent returns nil. -func (d *Dentry) Parent() *Dentry { - return d.parent -} - -// ParentOrSelf is equivalent to Parent, but returns d if d is independent. -func (d *Dentry) ParentOrSelf() *Dentry { - if d.parent == nil { - return d - } - return d.parent -} - -// Child returns d's child with the given name in its owning Filesystem. It -// does not take a reference on the returned Dentry. If no such child exists, -// Child returns nil. -func (d *Dentry) Child(name string) *Dentry { - return d.children[name] -} - -// HasChildren returns true if d has any children. -func (d *Dentry) HasChildren() bool { - return len(d.children) != 0 -} - -// Children returns a map containing all of d's children. -func (d *Dentry) Children() map[string]*Dentry { - if !d.HasChildren() { - return nil - } - m := make(map[string]*Dentry) - for name, child := range d.children { - m[name] = child - } - return m +// IsDead returns true if d has been deleted or invalidated by its owning +// filesystem. +func (d *Dentry) IsDead() bool { + d.mu.Lock() + defer d.mu.Unlock() + return d.dead } -// InsertChild makes child a child of d with the given name. -// -// InsertChild is a mutator of d and child. -// -// Preconditions: child must be an independent Dentry. d and child must be from -// the same Filesystem. d must not already have a child with the given name. -func (d *Dentry) InsertChild(child *Dentry, name string) { - if checkInvariants { - if _, ok := d.children[name]; ok { - panic(fmt.Sprintf("parent already contains a child named %q", name)) - } - if child.parent != nil || child.name != "" { - panic(fmt.Sprintf("child is not independent: parent = %v, name = %q", child.parent, child.name)) - } - } - if d.children == nil { - d.children = make(map[string]*Dentry) - } - d.children[name] = child - child.parent = d - child.name = name +func (d *Dentry) isMounted() bool { + return atomic.LoadUint32(&d.mounts) != 0 } -// IsAncestorOf returns true if d is an ancestor of d2; that is, d is either -// d2's parent or an ancestor of d2's parent. -func (d *Dentry) IsAncestorOf(d2 *Dentry) bool { - for d2.parent != nil { - if d2.parent == d { - return true - } - d2 = d2.parent - } - return false -} +// The following functions are exported so that filesystem implementations can +// use them. The vfs package, and users of VFS, should not call these +// functions. // PrepareDeleteDentry must be called before attempting to delete the file // represented by d. If PrepareDeleteDentry succeeds, the caller must call // AbortDeleteDentry or CommitDeleteDentry depending on the deletion's outcome. -// -// Preconditions: d is a child Dentry. func (vfs *VirtualFilesystem) PrepareDeleteDentry(mntns *MountNamespace, d *Dentry) error { - if checkInvariants { - if d.parent == nil { - panic("d is independent") - } - if d.IsDisowned() { - panic("d is already disowned") - } - } vfs.mountMu.Lock() if mntns.mountpoints[d] != 0 { vfs.mountMu.Unlock() @@ -280,42 +159,27 @@ func (vfs *VirtualFilesystem) AbortDeleteDentry(d *Dentry) { d.mu.Unlock() } -// CommitDeleteDentry must be called after the file represented by d is -// deleted, and causes d to become disowned. -// -// CommitDeleteDentry is a mutator of d and d.Parent(). -// -// Preconditions: PrepareDeleteDentry was previously called on d. +// CommitDeleteDentry must be called after PrepareDeleteDentry if the deletion +// succeeds. func (vfs *VirtualFilesystem) CommitDeleteDentry(d *Dentry) { - if d.parent != nil { - delete(d.parent.children, d.name) - } - d.setDisowned() + d.dead = true d.mu.Unlock() if d.isMounted() { - vfs.forgetDisownedMountpoint(d) + vfs.forgetDeadMountpoint(d) } } -// ForceDeleteDentry causes d to become disowned. It should only be used in -// cases where VFS has no ability to stop the deletion (e.g. d represents the -// local state of a file on a remote filesystem on which the file has already -// been deleted). -// -// ForceDeleteDentry is a mutator of d and d.Parent(). -// -// Preconditions: d is a child Dentry. -func (vfs *VirtualFilesystem) ForceDeleteDentry(d *Dentry) { - if checkInvariants { - if d.parent == nil { - panic("d is independent") - } - if d.IsDisowned() { - panic("d is already disowned") - } - } +// InvalidateDentry is called when d ceases to represent the file it formerly +// did for reasons outside of VFS' control (e.g. d represents the local state +// of a file on a remote filesystem on which the file has already been +// deleted). +func (vfs *VirtualFilesystem) InvalidateDentry(d *Dentry) { d.mu.Lock() - vfs.CommitDeleteDentry(d) + d.dead = true + d.mu.Unlock() + if d.isMounted() { + vfs.forgetDeadMountpoint(d) + } } // PrepareRenameDentry must be called before attempting to rename the file @@ -324,25 +188,9 @@ func (vfs *VirtualFilesystem) ForceDeleteDentry(d *Dentry) { // caller must call AbortRenameDentry, CommitRenameReplaceDentry, or // CommitRenameExchangeDentry depending on the rename's outcome. // -// Preconditions: from is a child Dentry. If to is not nil, it must be a child -// Dentry from the same Filesystem. from != to. +// Preconditions: If to is not nil, it must be a child Dentry from the same +// Filesystem. from != to. func (vfs *VirtualFilesystem) PrepareRenameDentry(mntns *MountNamespace, from, to *Dentry) error { - if checkInvariants { - if from.parent == nil { - panic("from is independent") - } - if from.IsDisowned() { - panic("from is already disowned") - } - if to != nil { - if to.parent == nil { - panic("to is independent") - } - if to.IsDisowned() { - panic("to is already disowned") - } - } - } vfs.mountMu.Lock() if mntns.mountpoints[from] != 0 { vfs.mountMu.Unlock() @@ -376,24 +224,14 @@ func (vfs *VirtualFilesystem) AbortRenameDentry(from, to *Dentry) { // is renamed without RENAME_EXCHANGE. If to is not nil, it represents the file // that was replaced by from. // -// CommitRenameReplaceDentry is a mutator of from, to, from.Parent(), and -// to.Parent(). -// // Preconditions: PrepareRenameDentry was previously called on from and to. -// newParent.Child(newName) == to. -func (vfs *VirtualFilesystem) CommitRenameReplaceDentry(from, newParent *Dentry, newName string, to *Dentry) { - if newParent.children == nil { - newParent.children = make(map[string]*Dentry) - } - newParent.children[newName] = from - from.parent = newParent - from.name = newName +func (vfs *VirtualFilesystem) CommitRenameReplaceDentry(from, to *Dentry) { from.mu.Unlock() if to != nil { - to.setDisowned() + to.dead = true to.mu.Unlock() if to.isMounted() { - vfs.forgetDisownedMountpoint(to) + vfs.forgetDeadMountpoint(to) } } } @@ -401,25 +239,18 @@ func (vfs *VirtualFilesystem) CommitRenameReplaceDentry(from, newParent *Dentry, // CommitRenameExchangeDentry must be called after the files represented by // from and to are exchanged by rename(RENAME_EXCHANGE). // -// CommitRenameExchangeDentry is a mutator of from, to, from.Parent(), and -// to.Parent(). -// // Preconditions: PrepareRenameDentry was previously called on from and to. func (vfs *VirtualFilesystem) CommitRenameExchangeDentry(from, to *Dentry) { - from.parent, to.parent = to.parent, from.parent - from.name, to.name = to.name, from.name - from.parent.children[from.name] = from - to.parent.children[to.name] = to from.mu.Unlock() to.mu.Unlock() } -// forgetDisownedMountpoint is called when a mount point is deleted to umount -// all mounts using it in all other mount namespaces. +// forgetDeadMountpoint is called when a mount point is deleted or invalidated +// to umount all mounts using it in all other mount namespaces. // -// forgetDisownedMountpoint is analogous to Linux's +// forgetDeadMountpoint is analogous to Linux's // fs/namespace.c:__detach_mounts(). -func (vfs *VirtualFilesystem) forgetDisownedMountpoint(d *Dentry) { +func (vfs *VirtualFilesystem) forgetDeadMountpoint(d *Dentry) { var ( vdsToDecRef []VirtualDentry mountsToDecRef []*Mount diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 5976b5ccd..15cc091e2 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -127,7 +127,8 @@ func (fd *FileDescription) Init(impl FileDescriptionImpl, statusFlags uint32, mn mount: mnt, dentry: d, } - fd.vd.IncRef() + mnt.IncRef() + d.IncRef() fd.opts = *opts fd.readable = MayReadFileWithOpenFlags(statusFlags) fd.writable = writable diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index a537a29d1..74577bc2f 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -346,7 +346,10 @@ type FilesystemImpl interface { // ENOTEMPTY. // // Preconditions: !rp.Done(). For the final path component in rp, - // !rp.ShouldFollowSymlink(). oldName is not "." or "..". + // !rp.ShouldFollowSymlink(). oldParentVD.Dentry() was obtained from a + // previous call to + // oldParentVD.Mount().Filesystem().Impl().GetParentDentryAt(). oldName is + // not "." or "..". // // Postconditions: If RenameAt returns an error returned by // ResolvingPath.Resolve*(), then !rp.Done(). diff --git a/pkg/sentry/vfs/filesystem_impl_util.go b/pkg/sentry/vfs/filesystem_impl_util.go index 7315a588e..465e610e0 100644 --- a/pkg/sentry/vfs/filesystem_impl_util.go +++ b/pkg/sentry/vfs/filesystem_impl_util.go @@ -16,8 +16,6 @@ package vfs import ( "strings" - - "gvisor.dev/gvisor/pkg/fspath" ) // GenericParseMountOptions parses a comma-separated list of options of the @@ -43,27 +41,3 @@ func GenericParseMountOptions(str string) map[string]string { } return m } - -// GenericPrependPath may be used by implementations of -// FilesystemImpl.PrependPath() for which a single statically-determined lock -// or set of locks is sufficient to ensure its preconditions (as opposed to -// e.g. per-Dentry locks). -// -// Preconditions: Dentry.Name() and Dentry.Parent() must be held constant for -// vd.Dentry() and all of its ancestors. -func GenericPrependPath(vfsroot, vd VirtualDentry, b *fspath.Builder) error { - mnt, d := vd.mount, vd.dentry - for { - if mnt == vfsroot.mount && d == vfsroot.dentry { - return PrependPathAtVFSRootError{} - } - if d == mnt.root { - return nil - } - if d.parent == nil { - return PrependPathAtNonMountRootError{} - } - b.PrependComponent(d.name) - d = d.parent - } -} diff --git a/pkg/sentry/vfs/genericfstree/BUILD b/pkg/sentry/vfs/genericfstree/BUILD new file mode 100644 index 000000000..d8fd92677 --- /dev/null +++ b/pkg/sentry/vfs/genericfstree/BUILD @@ -0,0 +1,16 @@ +load("//tools/go_generics:defs.bzl", "go_template") + +package( + default_visibility = ["//:sandbox"], + licenses = ["notice"], +) + +go_template( + name = "generic_fstree", + srcs = [ + "genericfstree.go", + ], + types = [ + "Dentry", + ], +) diff --git a/pkg/sentry/vfs/genericfstree/genericfstree.go b/pkg/sentry/vfs/genericfstree/genericfstree.go new file mode 100644 index 000000000..286510195 --- /dev/null +++ b/pkg/sentry/vfs/genericfstree/genericfstree.go @@ -0,0 +1,80 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package genericfstree provides tools for implementing vfs.FilesystemImpls +// where a single statically-determined lock or set of locks is sufficient to +// ensure that a Dentry's name and parent are contextually immutable. +// +// Clients using this package must use the go_template_instance rule in +// tools/go_generics/defs.bzl to create an instantiation of this template +// package, providing types to use in place of Dentry. +package genericfstree + +import ( + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/vfs" +) + +// Dentry is a required type parameter that is a struct with the given fields. +type Dentry struct { + // vfsd is the embedded vfs.Dentry corresponding to this vfs.DentryImpl. + vfsd vfs.Dentry + + // parent is the parent of this Dentry in the filesystem's tree. If this + // Dentry is a filesystem root, parent is nil. + parent *Dentry + + // name is the name of this Dentry in its parent. If this Dentry is a + // filesystem root, name is unspecified. + name string +} + +// IsAncestorDentry returns true if d is an ancestor of d2; that is, d is +// either d2's parent or an ancestor of d2's parent. +func IsAncestorDentry(d, d2 *Dentry) bool { + for { + if d2.parent == d { + return true + } + if d2.parent == d2 { + return false + } + d2 = d2.parent + } +} + +// ParentOrSelf returns d.parent. If d.parent is nil, ParentOrSelf returns d. +func ParentOrSelf(d *Dentry) *Dentry { + if d.parent != nil { + return d.parent + } + return d +} + +// PrependPath is a generic implementation of FilesystemImpl.PrependPath(). +func PrependPath(vfsroot vfs.VirtualDentry, mnt *vfs.Mount, d *Dentry, b *fspath.Builder) error { + for { + if mnt == vfsroot.Mount() && &d.vfsd == vfsroot.Dentry() { + return vfs.PrependPathAtVFSRootError{} + } + if &d.vfsd == mnt.Root() { + return nil + } + if d.parent == nil { + return vfs.PrependPathAtNonMountRootError{} + } + b.PrependComponent(d.name) + d = d.parent + } +} diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index f06946103..02850b65c 100644 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go @@ -188,6 +188,7 @@ func (vfs *VirtualFilesystem) MountAt(ctx context.Context, creds *auth.Credentia if err != nil { return err } + // We can't hold vfs.mountMu while calling FilesystemImpl methods due to // lock ordering. vd, err := vfs.GetDentryAt(ctx, creds, target, &GetDentryOptions{}) @@ -199,7 +200,7 @@ func (vfs *VirtualFilesystem) MountAt(ctx context.Context, creds *auth.Credentia vfs.mountMu.Lock() vd.dentry.mu.Lock() for { - if vd.dentry.IsDisowned() { + if vd.dentry.dead { vd.dentry.mu.Unlock() vfs.mountMu.Unlock() vd.DecRef() @@ -665,6 +666,12 @@ func (mnt *Mount) submountsLocked() []*Mount { return mounts } +// Root returns the mount's root. It does not take a reference on the returned +// Dentry. +func (mnt *Mount) Root() *Dentry { + return mnt.root +} + // Root returns mntns' root. A reference is taken on the returned // VirtualDentry. func (mntns *MountNamespace) Root() VirtualDentry { diff --git a/pkg/sentry/vfs/pathname.go b/pkg/sentry/vfs/pathname.go index f21a88034..cd78d66bc 100644 --- a/pkg/sentry/vfs/pathname.go +++ b/pkg/sentry/vfs/pathname.go @@ -58,7 +58,7 @@ loop: switch err.(type) { case nil: if vd.mount == vfsroot.mount && vd.mount.root == vfsroot.dentry { - // GenericPrependPath() will have returned + // genericfstree.PrependPath() will have returned // PrependPathAtVFSRootError in this case since it checks // against vfsroot before mnt.root, but other implementations // of FilesystemImpl.PrependPath() may return nil instead. @@ -84,7 +84,7 @@ loop: } } b.PrependByte('/') - if origD.IsDisowned() { + if origD.IsDead() { b.AppendString(" (deleted)") } return b.String(), nil @@ -136,7 +136,7 @@ loop: // PathnameForGetcwd returns an absolute pathname to vd, consistent with // Linux's sys_getcwd(). func (vfs *VirtualFilesystem) PathnameForGetcwd(ctx context.Context, vfsroot, vd VirtualDentry) (string, error) { - if vd.dentry.IsDisowned() { + if vd.dentry.IsDead() { return "", syserror.ENOENT } diff --git a/pkg/sentry/vfs/resolving_path.go b/pkg/sentry/vfs/resolving_path.go index 8f31495da..9d047ff88 100644 --- a/pkg/sentry/vfs/resolving_path.go +++ b/pkg/sentry/vfs/resolving_path.go @@ -29,7 +29,9 @@ import ( // // From the perspective of FilesystemImpl methods, a ResolvingPath represents a // starting Dentry on the associated Filesystem (on which a reference is -// already held) and a stream of path components relative to that Dentry. +// already held), a stream of path components relative to that Dentry, and +// elements of the invoking Context that are commonly required by +// FilesystemImpl methods. // // ResolvingPath is loosely analogous to Linux's struct nameidata. type ResolvingPath struct { @@ -251,18 +253,17 @@ func (rp *ResolvingPath) relpathCommit() { rp.origParts[rp.curPart] = rp.pit } -// ResolveParent returns the VFS parent of d. It does not take a reference on -// the returned Dentry. -// -// Preconditions: There are no concurrent mutators of d. -// -// Postconditions: If the returned error is nil, then the returned Dentry is -// not nil. -func (rp *ResolvingPath) ResolveParent(d *Dentry) (*Dentry, error) { - var parent *Dentry +// CheckRoot is called before resolving the parent of the Dentry d. If the +// Dentry is contextually a VFS root, such that path resolution should treat +// d's parent as itself, CheckRoot returns (true, nil). If the Dentry is the +// root of a non-root mount, such that path resolution should switch to another +// Mount, CheckRoot returns (unspecified, non-nil error). Otherwise, path +// resolution should resolve d's parent normally, and CheckRoot returns (false, +// nil). +func (rp *ResolvingPath) CheckRoot(d *Dentry) (bool, error) { if d == rp.root.dentry && rp.mount == rp.root.mount { - // At contextual VFS root. - parent = d + // At contextual VFS root (due to e.g. chroot(2)). + return true, nil } else if d == rp.mount.root { // At mount root ... vd := rp.vfs.getMountpointAt(rp.mount, rp.root) @@ -270,59 +271,27 @@ func (rp *ResolvingPath) ResolveParent(d *Dentry) (*Dentry, error) { // ... of non-root mount. rp.nextMount = vd.mount rp.nextStart = vd.dentry - return nil, resolveMountRootOrJumpError{} + return false, resolveMountRootOrJumpError{} } // ... of root mount. - parent = d - } else if d.parent == nil { - // At filesystem root. - parent = d - } else { - parent = d.parent + return true, nil } - if parent.isMounted() { - if mnt := rp.vfs.getMountAt(rp.mount, parent); mnt != nil { - rp.nextMount = mnt - return nil, resolveMountPointError{} - } - } - return parent, nil + return false, nil } -// ResolveChild returns the VFS child of d with the given name. It does not -// take a reference on the returned Dentry. If no such child exists, -// ResolveChild returns (nil, nil). -// -// Preconditions: There are no concurrent mutators of d. -func (rp *ResolvingPath) ResolveChild(d *Dentry, name string) (*Dentry, error) { - child := d.children[name] - if child == nil { - return nil, nil +// CheckMount is called after resolving the parent or child of another Dentry +// to d. If d is a mount point, such that path resolution should switch to +// another Mount, CheckMount returns a non-nil error. Otherwise, CheckMount +// returns nil. +func (rp *ResolvingPath) CheckMount(d *Dentry) error { + if !d.isMounted() { + return nil } - if child.isMounted() { - if mnt := rp.vfs.getMountAt(rp.mount, child); mnt != nil { - rp.nextMount = mnt - return nil, resolveMountPointError{} - } - } - return child, nil -} - -// ResolveComponent returns the Dentry reached by starting at d and resolving -// the current path component in the stream represented by rp. It does not -// advance the stream. It does not take a reference on the returned Dentry. If -// no such Dentry exists, ResolveComponent returns (nil, nil). -// -// Preconditions: !rp.Done(). There are no concurrent mutators of d. -func (rp *ResolvingPath) ResolveComponent(d *Dentry) (*Dentry, error) { - switch pc := rp.Component(); pc { - case ".": - return d, nil - case "..": - return rp.ResolveParent(d) - default: - return rp.ResolveChild(d, pc) + if mnt := rp.vfs.getMountAt(rp.mount, d); mnt != nil { + rp.nextMount = mnt + return resolveMountPointError{} } + return nil } // ShouldFollowSymlink returns true if, supposing that the current path -- cgit v1.2.3 From 37e01fd2ea6a0e67637975863317be9aae1b02f0 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Tue, 21 Apr 2020 16:30:26 -0700 Subject: Misc VFS2 fixes - Fix defer operation ordering in kernfs.Filesystem.AccessAt() - Add AT_NULL entry in proc/pid/auvx - Fix line padding in /proc/pid/maps - Fix linux_dirent serialization for getdents(2) - Remove file creation flags from vfs.FileDescription.statusFlags() Updates #1193, #1035 PiperOrigin-RevId: 307704159 --- pkg/sentry/fs/proc/task.go | 3 +-- pkg/sentry/fsimpl/kernfs/filesystem.go | 2 +- pkg/sentry/fsimpl/proc/task_files.go | 13 +++++++------ pkg/sentry/fsimpl/proc/tasks_sys.go | 2 +- pkg/sentry/mm/procfs.go | 4 ++-- pkg/sentry/syscalls/linux/vfs2/getdents.go | 6 +++--- pkg/sentry/vfs/file_description.go | 15 +++++++++------ 7 files changed, 24 insertions(+), 21 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go index 4d42eac83..4bbe90198 100644 --- a/pkg/sentry/fs/proc/task.go +++ b/pkg/sentry/fs/proc/task.go @@ -73,8 +73,7 @@ func checkTaskState(t *kernel.Task) error { type taskDir struct { ramfs.Dir - t *kernel.Task - pidns *kernel.PIDNamespace + t *kernel.Task } var _ fs.InodeOperations = (*taskDir)(nil) diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 01c23d192..3164d022c 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -246,8 +246,8 @@ func (fs *Filesystem) Sync(ctx context.Context) error { // AccessAt implements vfs.Filesystem.Impl.AccessAt. func (fs *Filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error { fs.mu.RLock() - defer fs.mu.RUnlock() defer fs.processDeferredDecRefs() + defer fs.mu.RUnlock() _, inode, err := fs.walkExistingLocked(ctx, rp) if err != nil { diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index 2c6f8bdfc..f3173e197 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -111,17 +111,18 @@ func (d *auxvData) Generate(ctx context.Context, buf *bytes.Buffer) error { } defer m.DecUsers(ctx) - // Space for buffer with AT_NULL (0) terminator at the end. auxv := m.Auxv() + // Space for buffer with AT_NULL (0) terminator at the end. buf.Grow((len(auxv) + 1) * 16) for _, e := range auxv { - var tmp [8]byte - usermem.ByteOrder.PutUint64(tmp[:], e.Key) - buf.Write(tmp[:]) - - usermem.ByteOrder.PutUint64(tmp[:], uint64(e.Value)) + var tmp [16]byte + usermem.ByteOrder.PutUint64(tmp[:8], e.Key) + usermem.ByteOrder.PutUint64(tmp[8:], uint64(e.Value)) buf.Write(tmp[:]) } + var atNull [16]byte + buf.Write(atNull[:]) + return nil } diff --git a/pkg/sentry/fsimpl/proc/tasks_sys.go b/pkg/sentry/fsimpl/proc/tasks_sys.go index 3d5dc463c..f08668ca2 100644 --- a/pkg/sentry/fsimpl/proc/tasks_sys.go +++ b/pkg/sentry/fsimpl/proc/tasks_sys.go @@ -39,7 +39,7 @@ func newSysDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) *k "shmmni": newDentry(root, inoGen.NextIno(), 0444, shmData(linux.SHMMNI)), }), "vm": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "mmap_min_addr": newDentry(root, inoGen.NextIno(), 0444, &mmapMinAddrData{}), + "mmap_min_addr": newDentry(root, inoGen.NextIno(), 0444, &mmapMinAddrData{k: k}), "overcommit_memory": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0\n")), }), "net": newSysNetDir(root, inoGen, k), diff --git a/pkg/sentry/mm/procfs.go b/pkg/sentry/mm/procfs.go index 1ab92f046..6efe5102b 100644 --- a/pkg/sentry/mm/procfs.go +++ b/pkg/sentry/mm/procfs.go @@ -148,7 +148,7 @@ func (mm *MemoryManager) appendVMAMapsEntryLocked(ctx context.Context, vseg vmaI // Do not include the guard page: fs/proc/task_mmu.c:show_map_vma() => // stack_guard_page_start(). - fmt.Fprintf(b, "%08x-%08x %s%s %08x %02x:%02x %d ", + lineLen, _ := fmt.Fprintf(b, "%08x-%08x %s%s %08x %02x:%02x %d ", vseg.Start(), vseg.End(), vma.realPerms, private, vma.off, devMajor, devMinor, ino) // Figure out our filename or hint. @@ -165,7 +165,7 @@ func (mm *MemoryManager) appendVMAMapsEntryLocked(ctx context.Context, vseg vmaI } if s != "" { // Per linux, we pad until the 74th character. - if pad := 73 - b.Len(); pad > 0 { + if pad := 73 - lineLen; pad > 0 { b.WriteString(strings.Repeat(" ", pad)) } b.WriteString(s) diff --git a/pkg/sentry/syscalls/linux/vfs2/getdents.go b/pkg/sentry/syscalls/linux/vfs2/getdents.go index 62e98817d..c7c7bf7ce 100644 --- a/pkg/sentry/syscalls/linux/vfs2/getdents.go +++ b/pkg/sentry/syscalls/linux/vfs2/getdents.go @@ -130,7 +130,7 @@ func (cb *getdentsCallback) Handle(dirent vfs.Dirent) error { if cb.t.Arch().Width() != 8 { panic(fmt.Sprintf("unsupported sizeof(unsigned long): %d", cb.t.Arch().Width())) } - size := 8 + 8 + 2 + 1 + 1 + 1 + len(dirent.Name) + size := 8 + 8 + 2 + 1 + 1 + len(dirent.Name) size = (size + 7) &^ 7 // round up to multiple of sizeof(long) if size > cb.remaining { return syserror.EINVAL @@ -143,11 +143,11 @@ func (cb *getdentsCallback) Handle(dirent vfs.Dirent) error { // Zero out all remaining bytes in buf, including the NUL terminator // after dirent.Name and the zero padding byte between the name and // dirent type. - bufTail := buf[18+len(dirent.Name):] + bufTail := buf[18+len(dirent.Name) : size-1] for i := range bufTail { bufTail[i] = 0 } - bufTail[2] = dirent.Type + buf[size-1] = dirent.Type } n, err := cb.t.CopyOutBytes(cb.addr, buf) if err != nil { diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 15cc091e2..418d69b96 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -111,10 +111,10 @@ type FileDescriptionOptions struct { } // Init must be called before first use of fd. If it succeeds, it takes -// references on mnt and d. statusFlags is the initial file description status -// flags, which is usually the full set of flags passed to open(2). -func (fd *FileDescription) Init(impl FileDescriptionImpl, statusFlags uint32, mnt *Mount, d *Dentry, opts *FileDescriptionOptions) error { - writable := MayWriteFileWithOpenFlags(statusFlags) +// references on mnt and d. flags is the initial file description flags, which +// is usually the full set of flags passed to open(2). +func (fd *FileDescription) Init(impl FileDescriptionImpl, flags uint32, mnt *Mount, d *Dentry, opts *FileDescriptionOptions) error { + writable := MayWriteFileWithOpenFlags(flags) if writable { if err := mnt.CheckBeginWrite(); err != nil { return err @@ -122,7 +122,10 @@ func (fd *FileDescription) Init(impl FileDescriptionImpl, statusFlags uint32, mn } fd.refs = 1 - fd.statusFlags = statusFlags + + // Remove "file creation flags" to mirror the behavior from file.f_flags in + // fs/open.c:do_dentry_open + fd.statusFlags = flags &^ (linux.O_CREAT | linux.O_EXCL | linux.O_NOCTTY | linux.O_TRUNC) fd.vd = VirtualDentry{ mount: mnt, dentry: d, @@ -130,7 +133,7 @@ func (fd *FileDescription) Init(impl FileDescriptionImpl, statusFlags uint32, mn mnt.IncRef() d.IncRef() fd.opts = *opts - fd.readable = MayReadFileWithOpenFlags(statusFlags) + fd.readable = MayReadFileWithOpenFlags(flags) fd.writable = writable fd.impl = impl return nil -- cgit v1.2.3 From 80d0a958199cc6095e2d580e403d50ac1c3b5206 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 21 Apr 2020 17:58:43 -0700 Subject: Update gofer.filesystem.BoundEndpointAt() to allow path resolution. Even though BoundEndpointAt is not yet implemented for gofer fs, allow path resolution errors to be returned so that we can jump to tmpfs, where it is implemented. Updates #1476. PiperOrigin-RevId: 307718335 --- pkg/sentry/fsimpl/gofer/filesystem.go | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 43e863c61..eba4aabe8 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -1089,9 +1089,15 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error } // BoundEndpointAt implements FilesystemImpl.BoundEndpointAt. -// -// TODO(gvisor.dev/issue/1476): Implement BoundEndpointAt. func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath) (transport.BoundEndpoint, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckCaching(&ds) + _, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return nil, err + } + // TODO(gvisor.dev/issue/1476): Implement BoundEndpointAt. return nil, syserror.ECONNREFUSED } -- cgit v1.2.3 From 5e3596a6b8abb4c7ee8253be447b86a7b0fad7ad Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 21 Apr 2020 19:01:51 -0700 Subject: Fix set/getsockopt in vfs2 override. Updates #1476. PiperOrigin-RevId: 307726055 --- pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go index 21eb98444..74920f785 100644 --- a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go +++ b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go @@ -58,8 +58,8 @@ func Override(table map[uintptr]kernel.Syscall) { table[51] = syscalls.PartiallySupported("getsockname", GetSockName, "In process of porting socket syscalls to VFS2.", nil) table[52] = syscalls.PartiallySupported("getpeername", GetPeerName, "In process of porting socket syscalls to VFS2.", nil) table[53] = syscalls.PartiallySupported("socketpair", SocketPair, "In process of porting socket syscalls to VFS2.", nil) - table[54] = syscalls.PartiallySupported("getsockopt", GetSockOpt, "In process of porting socket syscalls to VFS2.", nil) - table[55] = syscalls.PartiallySupported("setsockopt", SetSockOpt, "In process of porting socket syscalls to VFS2.", nil) + table[54] = syscalls.PartiallySupported("setsockopt", SetSockOpt, "In process of porting socket syscalls to VFS2.", nil) + table[55] = syscalls.PartiallySupported("getsockopt", GetSockOpt, "In process of porting socket syscalls to VFS2.", nil) table[59] = syscalls.Supported("execve", Execve) table[72] = syscalls.Supported("fcntl", Fcntl) delete(table, 73) // flock -- cgit v1.2.3 From 0c586946ea26610b87c4ff7bda783a5a9ca11ec0 Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Wed, 22 Apr 2020 17:48:59 -0700 Subject: Specify a memory file in platform.New(). PiperOrigin-RevId: 307941984 --- pkg/abi/linux/BUILD | 1 + pkg/abi/linux/arch_amd64.go | 23 +++++++++++++++++++++++ pkg/abi/linux/seccomp.go | 7 +++++++ pkg/flipcall/packet_window_allocator.go | 4 ++-- pkg/seccomp/seccomp_unsafe.go | 9 +-------- pkg/sentry/kernel/task_run.go | 1 + pkg/sentry/platform/kvm/context.go | 3 +++ pkg/sentry/platform/kvm/kvm.go | 5 +++++ pkg/sentry/platform/platform.go | 21 +++++++++++++++++++++ pkg/sentry/platform/ptrace/ptrace.go | 13 +++++++++++++ pkg/sentry/platform/ptrace/subprocess.go | 2 +- runsc/cmd/BUILD | 2 +- runsc/cmd/boot.go | 9 +++++++-- runsc/sandbox/sandbox.go | 10 +++++++--- tools/nogo/config.go | 3 +++ 15 files changed, 96 insertions(+), 17 deletions(-) create mode 100644 pkg/abi/linux/arch_amd64.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/BUILD b/pkg/abi/linux/BUILD index 322d1ccc4..59b0e138a 100644 --- a/pkg/abi/linux/BUILD +++ b/pkg/abi/linux/BUILD @@ -10,6 +10,7 @@ go_library( name = "linux", srcs = [ "aio.go", + "arch_amd64.go", "audit.go", "bpf.go", "capability.go", diff --git a/pkg/abi/linux/arch_amd64.go b/pkg/abi/linux/arch_amd64.go new file mode 100644 index 000000000..0be31e755 --- /dev/null +++ b/pkg/abi/linux/arch_amd64.go @@ -0,0 +1,23 @@ +// Copyright 2020 The gVisor Authors. + +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build amd64 + +package linux + +// Start and end addresses of the vsyscall page. +const ( + VSyscallStartAddr uint64 = 0xffffffffff600000 + VSyscallEndAddr uint64 = 0xffffffffff601000 +) diff --git a/pkg/abi/linux/seccomp.go b/pkg/abi/linux/seccomp.go index 4eeb5cd7a..d0607e256 100644 --- a/pkg/abi/linux/seccomp.go +++ b/pkg/abi/linux/seccomp.go @@ -63,3 +63,10 @@ func (a BPFAction) String() string { func (a BPFAction) Data() uint16 { return uint16(a & SECCOMP_RET_DATA) } + +// SockFprog is sock_fprog taken from . +type SockFprog struct { + Len uint16 + pad [6]byte + Filter *BPFInstruction +} diff --git a/pkg/flipcall/packet_window_allocator.go b/pkg/flipcall/packet_window_allocator.go index ccb918fab..af9cc3d21 100644 --- a/pkg/flipcall/packet_window_allocator.go +++ b/pkg/flipcall/packet_window_allocator.go @@ -134,7 +134,7 @@ func (pwa *PacketWindowAllocator) Allocate(size int) (PacketWindowDescriptor, er start := pwa.nextAlloc pwa.nextAlloc = end return PacketWindowDescriptor{ - FD: pwa.fd, + FD: pwa.FD(), Offset: start, Length: size, }, nil @@ -158,7 +158,7 @@ func (pwa *PacketWindowAllocator) ensureFileSize(min int64) error { } newSize = newNewSize } - if err := syscall.Ftruncate(pwa.fd, newSize); err != nil { + if err := syscall.Ftruncate(pwa.FD(), newSize); err != nil { return fmt.Errorf("ftruncate failed: %v", err) } pwa.fileSize = newSize diff --git a/pkg/seccomp/seccomp_unsafe.go b/pkg/seccomp/seccomp_unsafe.go index be328db12..f7e986589 100644 --- a/pkg/seccomp/seccomp_unsafe.go +++ b/pkg/seccomp/seccomp_unsafe.go @@ -21,13 +21,6 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" ) -// sockFprog is sock_fprog taken from . -type sockFprog struct { - Len uint16 - pad [6]byte - Filter *linux.BPFInstruction -} - // SetFilter installs the given BPF program. // // This is safe to call from an afterFork context. @@ -39,7 +32,7 @@ func SetFilter(instrs []linux.BPFInstruction) syscall.Errno { return errno } - sockProg := sockFprog{ + sockProg := linux.SockFprog{ Len: uint16(len(instrs)), Filter: (*linux.BPFInstruction)(unsafe.Pointer(&instrs[0])), } diff --git a/pkg/sentry/kernel/task_run.go b/pkg/sentry/kernel/task_run.go index 2ba8d7e63..d654dd997 100644 --- a/pkg/sentry/kernel/task_run.go +++ b/pkg/sentry/kernel/task_run.go @@ -96,6 +96,7 @@ func (t *Task) run(threadID uintptr) { t.tg.liveGoroutines.Done() t.tg.pidns.owner.liveGoroutines.Done() t.tg.pidns.owner.runningGoroutines.Done() + t.p.Release() // Keep argument alive because stack trace for dead variables may not be correct. runtime.KeepAlive(threadID) diff --git a/pkg/sentry/platform/kvm/context.go b/pkg/sentry/platform/kvm/context.go index c769ac7b4..6507121ea 100644 --- a/pkg/sentry/platform/kvm/context.go +++ b/pkg/sentry/platform/kvm/context.go @@ -85,3 +85,6 @@ func (c *context) Switch(as platform.AddressSpace, ac arch.Context, _ int32) (*a func (c *context) Interrupt() { c.interrupt.NotifyInterrupt() } + +// Release implements platform.Context.Release(). +func (c *context) Release() {} diff --git a/pkg/sentry/platform/kvm/kvm.go b/pkg/sentry/platform/kvm/kvm.go index a9b4af43e..ae813e24e 100644 --- a/pkg/sentry/platform/kvm/kvm.go +++ b/pkg/sentry/platform/kvm/kvm.go @@ -191,6 +191,11 @@ func (*constructor) OpenDevice() (*os.File, error) { return OpenDevice() } +// Flags implements platform.Constructor.Flags(). +func (*constructor) Requirements() platform.Requirements { + return platform.Requirements{} +} + func init() { platform.Register("kvm", &constructor{}) } diff --git a/pkg/sentry/platform/platform.go b/pkg/sentry/platform/platform.go index 2ca696382..171513f3f 100644 --- a/pkg/sentry/platform/platform.go +++ b/pkg/sentry/platform/platform.go @@ -148,6 +148,9 @@ type Context interface { // Interrupt interrupts a concurrent call to Switch(), causing it to return // ErrContextInterrupt. Interrupt() + + // Release() releases any resources associated with this context. + Release() } var ( @@ -353,10 +356,28 @@ func (fr FileRange) String() string { return fmt.Sprintf("[%#x, %#x)", fr.Start, fr.End) } +// Requirements is used to specify platform specific requirements. +type Requirements struct { + // RequiresCurrentPIDNS indicates that the sandbox has to be started in the + // current pid namespace. + RequiresCurrentPIDNS bool + // RequiresCapSysPtrace indicates that the sandbox has to be started with + // the CAP_SYS_PTRACE capability. + RequiresCapSysPtrace bool +} + // Constructor represents a platform type. type Constructor interface { + // New returns a new platform instance. + // + // Arguments: + // + // * deviceFile - the device file (e.g. /dev/kvm for the KVM platform). New(deviceFile *os.File) (Platform, error) OpenDevice() (*os.File, error) + + // Requirements returns platform specific requirements. + Requirements() Requirements } // platforms contains all available platform types. diff --git a/pkg/sentry/platform/ptrace/ptrace.go b/pkg/sentry/platform/ptrace/ptrace.go index 03adb624b..08d055e05 100644 --- a/pkg/sentry/platform/ptrace/ptrace.go +++ b/pkg/sentry/platform/ptrace/ptrace.go @@ -177,6 +177,9 @@ func (c *context) Interrupt() { c.interrupt.NotifyInterrupt() } +// Release implements platform.Context.Release(). +func (c *context) Release() {} + // PTrace represents a collection of ptrace subprocesses. type PTrace struct { platform.MMapMinAddr @@ -248,6 +251,16 @@ func (*constructor) OpenDevice() (*os.File, error) { return nil, nil } +// Flags implements platform.Constructor.Flags(). +func (*constructor) Requirements() platform.Requirements { + // TODO(b/75837838): Also set a new PID namespace so that we limit + // access to other host processes. + return platform.Requirements{ + RequiresCapSysPtrace: true, + RequiresCurrentPIDNS: true, + } +} + func init() { platform.Register("ptrace", &constructor{}) } diff --git a/pkg/sentry/platform/ptrace/subprocess.go b/pkg/sentry/platform/ptrace/subprocess.go index a644609ef..773ddb1ed 100644 --- a/pkg/sentry/platform/ptrace/subprocess.go +++ b/pkg/sentry/platform/ptrace/subprocess.go @@ -332,7 +332,7 @@ func (t *thread) unexpectedStubExit() { msg, err := t.getEventMessage() status := syscall.WaitStatus(msg) if status.Signaled() && status.Signal() == syscall.SIGKILL { - // SIGKILL can be only sent by an user or OOM-killer. In both + // SIGKILL can be only sent by a user or OOM-killer. In both // these cases, we don't need to panic. There is no reasons to // think that something wrong in gVisor. log.Warningf("The ptrace stub process %v has been killed by SIGKILL.", t.tgid) diff --git a/runsc/cmd/BUILD b/runsc/cmd/BUILD index d0bb4613a..4900fbe16 100644 --- a/runsc/cmd/BUILD +++ b/runsc/cmd/BUILD @@ -44,13 +44,13 @@ go_library( "//pkg/sentry/control", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", + "//pkg/sentry/platform", "//pkg/state", "//pkg/state/statefile", "//pkg/sync", "//pkg/unet", "//pkg/urpc", "//runsc/boot", - "//runsc/boot/platforms", "//runsc/console", "//runsc/container", "//runsc/flag", diff --git a/runsc/cmd/boot.go b/runsc/cmd/boot.go index 0938944a6..4c2ac6ff0 100644 --- a/runsc/cmd/boot.go +++ b/runsc/cmd/boot.go @@ -25,8 +25,8 @@ import ( specs "github.com/opencontainers/runtime-spec/specs-go" "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/runsc/boot" - "gvisor.dev/gvisor/runsc/boot/platforms" "gvisor.dev/gvisor/runsc/flag" "gvisor.dev/gvisor/runsc/specutils" ) @@ -183,7 +183,12 @@ func (b *Boot) Execute(_ context.Context, f *flag.FlagSet, args ...interface{}) if caps == nil { caps = &specs.LinuxCapabilities{} } - if conf.Platform == platforms.Ptrace { + + gPlatform, err := platform.Lookup(conf.Platform) + if err != nil { + Fatalf("loading platform: %v", err) + } + if gPlatform.Requirements().RequiresCapSysPtrace { // Ptrace platform requires extra capabilities. const c = "CAP_SYS_PTRACE" caps.Bounding = append(caps.Bounding, c) diff --git a/runsc/sandbox/sandbox.go b/runsc/sandbox/sandbox.go index e82bcef6f..e4ec16e2f 100644 --- a/runsc/sandbox/sandbox.go +++ b/runsc/sandbox/sandbox.go @@ -446,9 +446,13 @@ func (s *Sandbox) createSandboxProcess(conf *boot.Config, args *Args, startSyncF nextFD++ } - // If the platform needs a device FD we must pass it in. - if deviceFile, err := deviceFileForPlatform(conf.Platform); err != nil { + gPlatform, err := platform.Lookup(conf.Platform) + if err != nil { return err + } + + if deviceFile, err := gPlatform.OpenDevice(); err != nil { + return fmt.Errorf("opening device file for platform %q: %v", gPlatform, err) } else if deviceFile != nil { defer deviceFile.Close() cmd.ExtraFiles = append(cmd.ExtraFiles, deviceFile) @@ -539,7 +543,7 @@ func (s *Sandbox) createSandboxProcess(conf *boot.Config, args *Args, startSyncF {Type: specs.UTSNamespace}, } - if conf.Platform == platforms.Ptrace { + if gPlatform.Requirements().RequiresCurrentPIDNS { // TODO(b/75837838): Also set a new PID namespace so that we limit // access to other host processes. log.Infof("Sandbox will be started in the current PID namespace") diff --git a/tools/nogo/config.go b/tools/nogo/config.go index 0c4b7dd40..6958fca69 100644 --- a/tools/nogo/config.go +++ b/tools/nogo/config.go @@ -103,6 +103,9 @@ var analyzerConfig = map[*analysis.Analyzer]matcher{ "pkg/sentry/platform/ring0/pagetables/allocator_unsafe.go", // Special case. "pkg/sentry/platform/safecopy/safecopy_unsafe.go", // Special case. "pkg/sentry/vfs/mount_unsafe.go", // Special case. + "pkg/sentry/platform/systrap/stub_unsafe.go", // Special case. + "pkg/sentry/platform/systrap/switchto_google_unsafe.go", // Special case. + "pkg/sentry/platform/systrap/sysmsg_thread_unsafe.go", // Special case. ), ), unusedresult.Analyzer: alwaysMatches(), -- cgit v1.2.3 From e69a871c7bd4e4859b0acd8b875171f3ebbaec29 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Wed, 22 Apr 2020 22:17:01 -0700 Subject: Move user home detection to its own library. PiperOrigin-RevId: 307977689 --- pkg/sentry/fs/user/BUILD | 34 ++++++ pkg/sentry/fs/user/user.go | 237 +++++++++++++++++++++++++++++++++++++ pkg/sentry/fs/user/user_test.go | 198 +++++++++++++++++++++++++++++++ runsc/boot/BUILD | 5 +- runsc/boot/loader.go | 7 +- runsc/boot/user.go | 234 ------------------------------------ runsc/boot/user_test.go | 254 ---------------------------------------- 7 files changed, 474 insertions(+), 495 deletions(-) create mode 100644 pkg/sentry/fs/user/BUILD create mode 100644 pkg/sentry/fs/user/user.go create mode 100644 pkg/sentry/fs/user/user_test.go delete mode 100644 runsc/boot/user.go delete mode 100644 runsc/boot/user_test.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/user/BUILD b/pkg/sentry/fs/user/BUILD new file mode 100644 index 000000000..f37f979f1 --- /dev/null +++ b/pkg/sentry/fs/user/BUILD @@ -0,0 +1,34 @@ +load("//tools:defs.bzl", "go_library", "go_test") + +package(licenses = ["notice"]) + +go_library( + name = "user", + srcs = ["user.go"], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/abi/linux", + "//pkg/context", + "//pkg/fspath", + "//pkg/sentry/fs", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/vfs", + "//pkg/usermem", + ], +) + +go_test( + name = "user_test", + size = "small", + srcs = ["user_test.go"], + library = ":user", + deps = [ + "//pkg/abi/linux", + "//pkg/context", + "//pkg/sentry/fs", + "//pkg/sentry/fs/tmpfs", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/kernel/contexttest", + "//pkg/usermem", + ], +) diff --git a/pkg/sentry/fs/user/user.go b/pkg/sentry/fs/user/user.go new file mode 100644 index 000000000..fe7f67c00 --- /dev/null +++ b/pkg/sentry/fs/user/user.go @@ -0,0 +1,237 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package user + +import ( + "bufio" + "fmt" + "io" + "strconv" + "strings" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/usermem" +) + +type fileReader struct { + // Ctx is the context for the file reader. + Ctx context.Context + + // File is the file to read from. + File *fs.File +} + +// Read implements io.Reader.Read. +func (r *fileReader) Read(buf []byte) (int, error) { + n, err := r.File.Readv(r.Ctx, usermem.BytesIOSequence(buf)) + return int(n), err +} + +// getExecUserHome returns the home directory of the executing user read from +// /etc/passwd as read from the container filesystem. +func getExecUserHome(ctx context.Context, rootMns *fs.MountNamespace, uid auth.KUID) (string, error) { + // The default user home directory to return if no user matching the user + // if found in the /etc/passwd found in the image. + const defaultHome = "/" + + // Open the /etc/passwd file from the dirent via the root mount namespace. + mnsRoot := rootMns.Root() + maxTraversals := uint(linux.MaxSymlinkTraversals) + dirent, err := rootMns.FindInode(ctx, mnsRoot, nil, "/etc/passwd", &maxTraversals) + if err != nil { + // NOTE: Ignore errors opening the passwd file. If the passwd file + // doesn't exist we will return the default home directory. + return defaultHome, nil + } + defer dirent.DecRef() + + // Check read permissions on the file. + if err := dirent.Inode.CheckPermission(ctx, fs.PermMask{Read: true}); err != nil { + // NOTE: Ignore permissions errors here and return default root dir. + return defaultHome, nil + } + + // Only open regular files. We don't open other files like named pipes as + // they may block and might present some attack surface to the container. + // Note that runc does not seem to do this kind of checking. + if !fs.IsRegular(dirent.Inode.StableAttr) { + return defaultHome, nil + } + + f, err := dirent.Inode.GetFile(ctx, dirent, fs.FileFlags{Read: true, Directory: false}) + if err != nil { + return "", err + } + defer f.DecRef() + + r := &fileReader{ + Ctx: ctx, + File: f, + } + + return findHomeInPasswd(uint32(uid), r, defaultHome) +} + +type fileReaderVFS2 struct { + ctx context.Context + fd *vfs.FileDescription +} + +func (r *fileReaderVFS2) Read(buf []byte) (int, error) { + n, err := r.fd.Read(r.ctx, usermem.BytesIOSequence(buf), vfs.ReadOptions{}) + return int(n), err +} + +func getExecUserHomeVFS2(ctx context.Context, mns *vfs.MountNamespace, uid auth.KUID) (string, error) { + const defaultHome = "/" + + root := mns.Root() + defer root.DecRef() + + creds := auth.CredentialsFromContext(ctx) + + target := &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse("/etc/passwd"), + } + + opts := &vfs.OpenOptions{ + Flags: linux.O_RDONLY, + } + + fd, err := root.Mount().Filesystem().VirtualFilesystem().OpenAt(ctx, creds, target, opts) + if err != nil { + return defaultHome, nil + } + defer fd.DecRef() + + r := &fileReaderVFS2{ + ctx: ctx, + fd: fd, + } + + homeDir, err := findHomeInPasswd(uint32(uid), r, defaultHome) + if err != nil { + return "", err + } + + return homeDir, nil +} + +// MaybeAddExecUserHome returns a new slice with the HOME enviroment variable +// set if the slice does not already contain it, otherwise it returns the +// original slice unmodified. +func MaybeAddExecUserHome(ctx context.Context, mns *fs.MountNamespace, uid auth.KUID, envv []string) ([]string, error) { + // Check if the envv already contains HOME. + for _, env := range envv { + if strings.HasPrefix(env, "HOME=") { + // We have it. Return the original slice unmodified. + return envv, nil + } + } + + // Read /etc/passwd for the user's HOME directory and set the HOME + // environment variable as required by POSIX if it is not overridden by + // the user. + homeDir, err := getExecUserHome(ctx, mns, uid) + if err != nil { + return nil, fmt.Errorf("error reading exec user: %v", err) + } + + return append(envv, "HOME="+homeDir), nil +} + +// MaybeAddExecUserHomeVFS2 returns a new slice with the HOME enviroment +// variable set if the slice does not already contain it, otherwise it returns +// the original slice unmodified. +func MaybeAddExecUserHomeVFS2(ctx context.Context, vmns *vfs.MountNamespace, uid auth.KUID, envv []string) ([]string, error) { + // Check if the envv already contains HOME. + for _, env := range envv { + if strings.HasPrefix(env, "HOME=") { + // We have it. Return the original slice unmodified. + return envv, nil + } + } + + // Read /etc/passwd for the user's HOME directory and set the HOME + // environment variable as required by POSIX if it is not overridden by + // the user. + homeDir, err := getExecUserHomeVFS2(ctx, vmns, uid) + if err != nil { + return nil, fmt.Errorf("error reading exec user: %v", err) + } + return append(envv, "HOME="+homeDir), nil +} + +// findHomeInPasswd parses a passwd file and returns the given user's home +// directory. This function does it's best to replicate the runc's behavior. +func findHomeInPasswd(uid uint32, passwd io.Reader, defaultHome string) (string, error) { + s := bufio.NewScanner(passwd) + + for s.Scan() { + if err := s.Err(); err != nil { + return "", err + } + + line := strings.TrimSpace(s.Text()) + if line == "" { + continue + } + + // Pull out part of passwd entry. Loosely parse the passwd entry as some + // passwd files could be poorly written and for compatibility with runc. + // + // Per 'man 5 passwd' + // /etc/passwd contains one line for each user account, with seven + // fields delimited by colons (“:”). These fields are: + // + // - login name + // - optional encrypted password + // - numerical user ID + // - numerical group ID + // - user name or comment field + // - user home directory + // - optional user command interpreter + parts := strings.Split(line, ":") + + found := false + homeDir := "" + for i, p := range parts { + switch i { + case 2: + parsedUID, err := strconv.ParseUint(p, 10, 32) + if err == nil && parsedUID == uint64(uid) { + found = true + } + case 5: + homeDir = p + } + } + if found { + // NOTE: If the uid is present but the home directory is not + // present in the /etc/passwd entry we return an empty string. This + // is, for better or worse, what runc does. + return homeDir, nil + } + } + + return defaultHome, nil +} diff --git a/pkg/sentry/fs/user/user_test.go b/pkg/sentry/fs/user/user_test.go new file mode 100644 index 000000000..7d8e9ac7c --- /dev/null +++ b/pkg/sentry/fs/user/user_test.go @@ -0,0 +1,198 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package user + +import ( + "fmt" + "strings" + "testing" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/fs/tmpfs" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" + "gvisor.dev/gvisor/pkg/usermem" +) + +// createEtcPasswd creates /etc/passwd with the given contents and mode. If +// mode is empty, then no file will be created. If mode is not a regular file +// mode, then contents is ignored. +func createEtcPasswd(ctx context.Context, root *fs.Dirent, contents string, mode linux.FileMode) error { + if err := root.CreateDirectory(ctx, root, "etc", fs.FilePermsFromMode(0755)); err != nil { + return err + } + etc, err := root.Walk(ctx, root, "etc") + if err != nil { + return err + } + defer etc.DecRef() + switch mode.FileType() { + case 0: + // Don't create anything. + return nil + case linux.S_IFREG: + passwd, err := etc.Create(ctx, root, "passwd", fs.FileFlags{Write: true}, fs.FilePermsFromMode(mode)) + if err != nil { + return err + } + defer passwd.DecRef() + if _, err := passwd.Writev(ctx, usermem.BytesIOSequence([]byte(contents))); err != nil { + return err + } + return nil + case linux.S_IFDIR: + return etc.CreateDirectory(ctx, root, "passwd", fs.FilePermsFromMode(mode)) + case linux.S_IFIFO: + return etc.CreateFifo(ctx, root, "passwd", fs.FilePermsFromMode(mode)) + default: + return fmt.Errorf("unknown file type %x", mode.FileType()) + } +} + +// TestGetExecUserHome tests the getExecUserHome function. +func TestGetExecUserHome(t *testing.T) { + tests := map[string]struct { + uid auth.KUID + passwdContents string + passwdMode linux.FileMode + expected string + }{ + "success": { + uid: 1000, + passwdContents: "adin::1000:1111::/home/adin:/bin/sh", + passwdMode: linux.S_IFREG | 0666, + expected: "/home/adin", + }, + "no_perms": { + uid: 1000, + passwdContents: "adin::1000:1111::/home/adin:/bin/sh", + passwdMode: linux.S_IFREG, + expected: "/", + }, + "no_passwd": { + uid: 1000, + expected: "/", + }, + "directory": { + uid: 1000, + passwdMode: linux.S_IFDIR | 0666, + expected: "/", + }, + // Currently we don't allow named pipes. + "named_pipe": { + uid: 1000, + passwdMode: linux.S_IFIFO | 0666, + expected: "/", + }, + } + + for name, tc := range tests { + t.Run(name, func(t *testing.T) { + ctx := contexttest.Context(t) + msrc := fs.NewPseudoMountSource(ctx) + rootInode := tmpfs.NewDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0777), msrc) + + mns, err := fs.NewMountNamespace(ctx, rootInode) + if err != nil { + t.Fatalf("NewMountNamespace failed: %v", err) + } + defer mns.DecRef() + root := mns.Root() + defer root.DecRef() + ctx = fs.WithRoot(ctx, root) + + if err := createEtcPasswd(ctx, root, tc.passwdContents, tc.passwdMode); err != nil { + t.Fatalf("createEtcPasswd failed: %v", err) + } + + got, err := getExecUserHome(ctx, mns, tc.uid) + if err != nil { + t.Fatalf("failed to get user home: %v", err) + } + + if got != tc.expected { + t.Fatalf("expected %v, got: %v", tc.expected, got) + } + }) + } +} + +// TestFindHomeInPasswd tests the findHomeInPasswd function's passwd file parsing. +func TestFindHomeInPasswd(t *testing.T) { + tests := map[string]struct { + uid uint32 + passwd string + expected string + def string + }{ + "empty": { + uid: 1000, + passwd: "", + expected: "/", + def: "/", + }, + "whitespace": { + uid: 1000, + passwd: " ", + expected: "/", + def: "/", + }, + "full": { + uid: 1000, + passwd: "adin::1000:1111::/home/adin:/bin/sh", + expected: "/home/adin", + def: "/", + }, + // For better or worse, this is how runc works. + "partial": { + uid: 1000, + passwd: "adin::1000:1111:", + expected: "", + def: "/", + }, + "multiple": { + uid: 1001, + passwd: "adin::1000:1111::/home/adin:/bin/sh\nian::1001:1111::/home/ian:/bin/sh", + expected: "/home/ian", + def: "/", + }, + "duplicate": { + uid: 1000, + passwd: "adin::1000:1111::/home/adin:/bin/sh\nian::1000:1111::/home/ian:/bin/sh", + expected: "/home/adin", + def: "/", + }, + "empty_lines": { + uid: 1001, + passwd: "adin::1000:1111::/home/adin:/bin/sh\n\n\nian::1001:1111::/home/ian:/bin/sh", + expected: "/home/ian", + def: "/", + }, + } + + for name, tc := range tests { + t.Run(name, func(t *testing.T) { + got, err := findHomeInPasswd(tc.uid, strings.NewReader(tc.passwd), tc.def) + if err != nil { + t.Fatalf("error parsing passwd: %v", err) + } + if tc.expected != got { + t.Fatalf("expected %v, got: %v", tc.expected, got) + } + }) + } +} diff --git a/runsc/boot/BUILD b/runsc/boot/BUILD index 5451f1eba..72c2fe381 100644 --- a/runsc/boot/BUILD +++ b/runsc/boot/BUILD @@ -20,7 +20,6 @@ go_library( "loader_arm64.go", "network.go", "strace.go", - "user.go", "vfs.go", ], visibility = [ @@ -52,6 +51,7 @@ go_library( "//pkg/sentry/fs/sys", "//pkg/sentry/fs/tmpfs", "//pkg/sentry/fs/tty", + "//pkg/sentry/fs/user", "//pkg/sentry/fsimpl/devtmpfs", "//pkg/sentry/fsimpl/gofer", "//pkg/sentry/fsimpl/host", @@ -97,7 +97,6 @@ go_library( "//pkg/tcpip/transport/tcp", "//pkg/tcpip/transport/udp", "//pkg/urpc", - "//pkg/usermem", "//runsc/boot/filter", "//runsc/boot/platforms", "//runsc/boot/pprof", @@ -115,7 +114,6 @@ go_test( "compat_test.go", "fs_test.go", "loader_test.go", - "user_test.go", ], library = ":boot", deps = [ @@ -125,7 +123,6 @@ go_test( "//pkg/sentry/contexttest", "//pkg/sentry/fs", "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", "//pkg/sync", "//pkg/unet", "//runsc/fsgofer", diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index cf1f47bc7..096b0e9f0 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -35,6 +35,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/control" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/host" + "gvisor.dev/gvisor/pkg/sentry/fs/user" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" @@ -550,11 +551,11 @@ func (l *Loader) run() error { // Add the HOME enviroment variable if it is not already set. var envv []string if kernel.VFS2Enabled { - envv, err = maybeAddExecUserHomeVFS2(ctx, l.rootProcArgs.MountNamespaceVFS2, + envv, err = user.MaybeAddExecUserHomeVFS2(ctx, l.rootProcArgs.MountNamespaceVFS2, l.rootProcArgs.Credentials.RealKUID, l.rootProcArgs.Envv) } else { - envv, err = maybeAddExecUserHome(ctx, l.rootProcArgs.MountNamespace, + envv, err = user.MaybeAddExecUserHome(ctx, l.rootProcArgs.MountNamespace, l.rootProcArgs.Credentials.RealKUID, l.rootProcArgs.Envv) } if err != nil { @@ -860,7 +861,7 @@ func (l *Loader) executeAsync(args *control.ExecArgs) (kernel.ThreadID, error) { root := args.MountNamespace.Root() defer root.DecRef() ctx := fs.WithRoot(l.k.SupervisorContext(), root) - envv, err := maybeAddExecUserHome(ctx, args.MountNamespace, args.KUID, args.Envv) + envv, err := user.MaybeAddExecUserHome(ctx, args.MountNamespace, args.KUID, args.Envv) if err != nil { return 0, err } diff --git a/runsc/boot/user.go b/runsc/boot/user.go deleted file mode 100644 index 332e4fce5..000000000 --- a/runsc/boot/user.go +++ /dev/null @@ -1,234 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package boot - -import ( - "bufio" - "fmt" - "io" - "strconv" - "strings" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/usermem" -) - -type fileReader struct { - // Ctx is the context for the file reader. - Ctx context.Context - - // File is the file to read from. - File *fs.File -} - -// Read implements io.Reader.Read. -func (r *fileReader) Read(buf []byte) (int, error) { - n, err := r.File.Readv(r.Ctx, usermem.BytesIOSequence(buf)) - return int(n), err -} - -// getExecUserHome returns the home directory of the executing user read from -// /etc/passwd as read from the container filesystem. -func getExecUserHome(ctx context.Context, rootMns *fs.MountNamespace, uid auth.KUID) (string, error) { - // The default user home directory to return if no user matching the user - // if found in the /etc/passwd found in the image. - const defaultHome = "/" - - // Open the /etc/passwd file from the dirent via the root mount namespace. - mnsRoot := rootMns.Root() - maxTraversals := uint(linux.MaxSymlinkTraversals) - dirent, err := rootMns.FindInode(ctx, mnsRoot, nil, "/etc/passwd", &maxTraversals) - if err != nil { - // NOTE: Ignore errors opening the passwd file. If the passwd file - // doesn't exist we will return the default home directory. - return defaultHome, nil - } - defer dirent.DecRef() - - // Check read permissions on the file. - if err := dirent.Inode.CheckPermission(ctx, fs.PermMask{Read: true}); err != nil { - // NOTE: Ignore permissions errors here and return default root dir. - return defaultHome, nil - } - - // Only open regular files. We don't open other files like named pipes as - // they may block and might present some attack surface to the container. - // Note that runc does not seem to do this kind of checking. - if !fs.IsRegular(dirent.Inode.StableAttr) { - return defaultHome, nil - } - - f, err := dirent.Inode.GetFile(ctx, dirent, fs.FileFlags{Read: true, Directory: false}) - if err != nil { - return "", err - } - defer f.DecRef() - - r := &fileReader{ - Ctx: ctx, - File: f, - } - - return findHomeInPasswd(uint32(uid), r, defaultHome) -} - -type fileReaderVFS2 struct { - ctx context.Context - fd *vfs.FileDescription -} - -func (r *fileReaderVFS2) Read(buf []byte) (int, error) { - n, err := r.fd.Read(r.ctx, usermem.BytesIOSequence(buf), vfs.ReadOptions{}) - return int(n), err -} - -func getExecUserHomeVFS2(ctx context.Context, mns *vfs.MountNamespace, uid auth.KUID) (string, error) { - const defaultHome = "/" - - root := mns.Root() - defer root.DecRef() - - creds := auth.CredentialsFromContext(ctx) - - target := &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse("/etc/passwd"), - } - - opts := &vfs.OpenOptions{ - Flags: linux.O_RDONLY, - } - - fd, err := root.Mount().Filesystem().VirtualFilesystem().OpenAt(ctx, creds, target, opts) - if err != nil { - return defaultHome, nil - } - defer fd.DecRef() - - r := &fileReaderVFS2{ - ctx: ctx, - fd: fd, - } - - homeDir, err := findHomeInPasswd(uint32(uid), r, defaultHome) - if err != nil { - return "", err - } - - return homeDir, nil -} - -// maybeAddExecUserHome returns a new slice with the HOME enviroment variable -// set if the slice does not already contain it, otherwise it returns the -// original slice unmodified. -func maybeAddExecUserHome(ctx context.Context, mns *fs.MountNamespace, uid auth.KUID, envv []string) ([]string, error) { - // Check if the envv already contains HOME. - for _, env := range envv { - if strings.HasPrefix(env, "HOME=") { - // We have it. Return the original slice unmodified. - return envv, nil - } - } - - // Read /etc/passwd for the user's HOME directory and set the HOME - // environment variable as required by POSIX if it is not overridden by - // the user. - homeDir, err := getExecUserHome(ctx, mns, uid) - if err != nil { - return nil, fmt.Errorf("error reading exec user: %v", err) - } - - return append(envv, "HOME="+homeDir), nil -} - -func maybeAddExecUserHomeVFS2(ctx context.Context, vmns *vfs.MountNamespace, uid auth.KUID, envv []string) ([]string, error) { - // Check if the envv already contains HOME. - for _, env := range envv { - if strings.HasPrefix(env, "HOME=") { - // We have it. Return the original slice unmodified. - return envv, nil - } - } - - // Read /etc/passwd for the user's HOME directory and set the HOME - // environment variable as required by POSIX if it is not overridden by - // the user. - homeDir, err := getExecUserHomeVFS2(ctx, vmns, uid) - if err != nil { - return nil, fmt.Errorf("error reading exec user: %v", err) - } - return append(envv, "HOME="+homeDir), nil -} - -// findHomeInPasswd parses a passwd file and returns the given user's home -// directory. This function does it's best to replicate the runc's behavior. -func findHomeInPasswd(uid uint32, passwd io.Reader, defaultHome string) (string, error) { - s := bufio.NewScanner(passwd) - - for s.Scan() { - if err := s.Err(); err != nil { - return "", err - } - - line := strings.TrimSpace(s.Text()) - if line == "" { - continue - } - - // Pull out part of passwd entry. Loosely parse the passwd entry as some - // passwd files could be poorly written and for compatibility with runc. - // - // Per 'man 5 passwd' - // /etc/passwd contains one line for each user account, with seven - // fields delimited by colons (“:”). These fields are: - // - // - login name - // - optional encrypted password - // - numerical user ID - // - numerical group ID - // - user name or comment field - // - user home directory - // - optional user command interpreter - parts := strings.Split(line, ":") - - found := false - homeDir := "" - for i, p := range parts { - switch i { - case 2: - parsedUID, err := strconv.ParseUint(p, 10, 32) - if err == nil && parsedUID == uint64(uid) { - found = true - } - case 5: - homeDir = p - } - } - if found { - // NOTE: If the uid is present but the home directory is not - // present in the /etc/passwd entry we return an empty string. This - // is, for better or worse, what runc does. - return homeDir, nil - } - } - - return defaultHome, nil -} diff --git a/runsc/boot/user_test.go b/runsc/boot/user_test.go deleted file mode 100644 index fb4e13dfb..000000000 --- a/runsc/boot/user_test.go +++ /dev/null @@ -1,254 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package boot - -import ( - "io/ioutil" - "os" - "path/filepath" - "strings" - "syscall" - "testing" - - specs "github.com/opencontainers/runtime-spec/specs-go" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" -) - -func setupTempDir() (string, error) { - tmpDir, err := ioutil.TempDir(os.TempDir(), "exec-user-test") - if err != nil { - return "", err - } - return tmpDir, nil -} - -func setupPasswd(contents string, perms os.FileMode) func() (string, error) { - return func() (string, error) { - tmpDir, err := setupTempDir() - if err != nil { - return "", err - } - - if err := os.Mkdir(filepath.Join(tmpDir, "etc"), 0777); err != nil { - return "", err - } - - f, err := os.Create(filepath.Join(tmpDir, "etc", "passwd")) - if err != nil { - return "", err - } - defer f.Close() - - _, err = f.WriteString(contents) - if err != nil { - return "", err - } - - err = f.Chmod(perms) - if err != nil { - return "", err - } - return tmpDir, nil - } -} - -// TestGetExecUserHome tests the getExecUserHome function. -func TestGetExecUserHome(t *testing.T) { - tests := map[string]struct { - uid auth.KUID - createRoot func() (string, error) - expected string - }{ - "success": { - uid: 1000, - createRoot: setupPasswd("adin::1000:1111::/home/adin:/bin/sh", 0666), - expected: "/home/adin", - }, - "no_passwd": { - uid: 1000, - createRoot: setupTempDir, - expected: "/", - }, - "no_perms": { - uid: 1000, - createRoot: setupPasswd("adin::1000:1111::/home/adin:/bin/sh", 0000), - expected: "/", - }, - "directory": { - uid: 1000, - createRoot: func() (string, error) { - tmpDir, err := setupTempDir() - if err != nil { - return "", err - } - - if err := os.Mkdir(filepath.Join(tmpDir, "etc"), 0777); err != nil { - return "", err - } - - if err := syscall.Mkdir(filepath.Join(tmpDir, "etc", "passwd"), 0666); err != nil { - return "", err - } - - return tmpDir, nil - }, - expected: "/", - }, - // Currently we don't allow named pipes. - "named_pipe": { - uid: 1000, - createRoot: func() (string, error) { - tmpDir, err := setupTempDir() - if err != nil { - return "", err - } - - if err := os.Mkdir(filepath.Join(tmpDir, "etc"), 0777); err != nil { - return "", err - } - - if err := syscall.Mkfifo(filepath.Join(tmpDir, "etc", "passwd"), 0666); err != nil { - return "", err - } - - return tmpDir, nil - }, - expected: "/", - }, - } - - for name, tc := range tests { - t.Run(name, func(t *testing.T) { - tmpDir, err := tc.createRoot() - if err != nil { - t.Fatalf("failed to create root dir: %v", err) - } - - sandEnd, cleanup, err := startGofer(tmpDir) - if err != nil { - t.Fatalf("failed to create gofer: %v", err) - } - defer cleanup() - - ctx := contexttest.Context(t) - conf := &Config{ - RootDir: "unused_root_dir", - Network: NetworkNone, - DisableSeccomp: true, - } - - spec := &specs.Spec{ - Root: &specs.Root{ - Path: tmpDir, - Readonly: true, - }, - // Add /proc mount as tmpfs to avoid needing a kernel. - Mounts: []specs.Mount{ - { - Destination: "/proc", - Type: "tmpfs", - }, - }, - } - - mntr := newContainerMounter(spec, []int{sandEnd}, nil, &podMountHints{}) - mns, err := mntr.createMountNamespace(ctx, conf) - if err != nil { - t.Fatalf("failed to create mount namespace: %v", err) - } - ctx = fs.WithRoot(ctx, mns.Root()) - if err := mntr.mountSubmounts(ctx, conf, mns); err != nil { - t.Fatalf("failed to create mount namespace: %v", err) - } - - got, err := getExecUserHome(ctx, mns, tc.uid) - if err != nil { - t.Fatalf("failed to get user home: %v", err) - } - - if got != tc.expected { - t.Fatalf("expected %v, got: %v", tc.expected, got) - } - }) - } -} - -// TestFindHomeInPasswd tests the findHomeInPasswd function's passwd file parsing. -func TestFindHomeInPasswd(t *testing.T) { - tests := map[string]struct { - uid uint32 - passwd string - expected string - def string - }{ - "empty": { - uid: 1000, - passwd: "", - expected: "/", - def: "/", - }, - "whitespace": { - uid: 1000, - passwd: " ", - expected: "/", - def: "/", - }, - "full": { - uid: 1000, - passwd: "adin::1000:1111::/home/adin:/bin/sh", - expected: "/home/adin", - def: "/", - }, - // For better or worse, this is how runc works. - "partial": { - uid: 1000, - passwd: "adin::1000:1111:", - expected: "", - def: "/", - }, - "multiple": { - uid: 1001, - passwd: "adin::1000:1111::/home/adin:/bin/sh\nian::1001:1111::/home/ian:/bin/sh", - expected: "/home/ian", - def: "/", - }, - "duplicate": { - uid: 1000, - passwd: "adin::1000:1111::/home/adin:/bin/sh\nian::1000:1111::/home/ian:/bin/sh", - expected: "/home/adin", - def: "/", - }, - "empty_lines": { - uid: 1001, - passwd: "adin::1000:1111::/home/adin:/bin/sh\n\n\nian::1001:1111::/home/ian:/bin/sh", - expected: "/home/ian", - def: "/", - }, - } - - for name, tc := range tests { - t.Run(name, func(t *testing.T) { - got, err := findHomeInPasswd(tc.uid, strings.NewReader(tc.passwd), tc.def) - if err != nil { - t.Fatalf("error parsing passwd: %v", err) - } - if tc.expected != got { - t.Fatalf("expected %v, got: %v", tc.expected, got) - } - }) - } -} -- cgit v1.2.3 From e0c67014cb2200ad58cd28b12fddb3f55652a21b Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Thu, 23 Apr 2020 11:06:59 -0700 Subject: Factor fsimpl/gofer.host{Preadv,Pwritev} out of fsimpl/gofer. Also fix returning EOF when 0 bytes are read. PiperOrigin-RevId: 308089875 --- pkg/sentry/fsimpl/gofer/BUILD | 2 +- pkg/sentry/fsimpl/gofer/handle.go | 5 +- pkg/sentry/fsimpl/gofer/handle_unsafe.go | 66 ------------------- pkg/sentry/fsimpl/host/BUILD | 3 +- pkg/sentry/fsimpl/host/host.go | 31 ++------- pkg/sentry/hostfd/BUILD | 17 +++++ pkg/sentry/hostfd/hostfd.go | 84 ++++++++++++++++++++++++ pkg/sentry/hostfd/hostfd_unsafe.go | 107 +++++++++++++++++++++++++++++++ 8 files changed, 218 insertions(+), 97 deletions(-) delete mode 100644 pkg/sentry/fsimpl/gofer/handle_unsafe.go create mode 100644 pkg/sentry/hostfd/BUILD create mode 100644 pkg/sentry/hostfd/hostfd.go create mode 100644 pkg/sentry/hostfd/hostfd_unsafe.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/BUILD b/pkg/sentry/fsimpl/gofer/BUILD index acd061905..b9c4beee4 100644 --- a/pkg/sentry/fsimpl/gofer/BUILD +++ b/pkg/sentry/fsimpl/gofer/BUILD @@ -35,7 +35,6 @@ go_library( "fstree.go", "gofer.go", "handle.go", - "handle_unsafe.go", "p9file.go", "pagemath.go", "regular_file.go", @@ -53,6 +52,7 @@ go_library( "//pkg/p9", "//pkg/safemem", "//pkg/sentry/fs/fsutil", + "//pkg/sentry/hostfd", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/time", "//pkg/sentry/memmap", diff --git a/pkg/sentry/fsimpl/gofer/handle.go b/pkg/sentry/fsimpl/gofer/handle.go index cfe66f797..724a3f1f7 100644 --- a/pkg/sentry/fsimpl/gofer/handle.go +++ b/pkg/sentry/fsimpl/gofer/handle.go @@ -20,6 +20,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/safemem" + "gvisor.dev/gvisor/pkg/sentry/hostfd" ) // handle represents a remote "open file descriptor", consisting of an opened @@ -77,7 +78,7 @@ func (h *handle) readToBlocksAt(ctx context.Context, dsts safemem.BlockSeq, offs } if h.fd >= 0 { ctx.UninterruptibleSleepStart(false) - n, err := hostPreadv(h.fd, dsts, int64(offset)) + n, err := hostfd.Preadv2(h.fd, dsts, int64(offset), 0 /* flags */) ctx.UninterruptibleSleepFinish(false) return n, err } @@ -103,7 +104,7 @@ func (h *handle) writeFromBlocksAt(ctx context.Context, srcs safemem.BlockSeq, o } if h.fd >= 0 { ctx.UninterruptibleSleepStart(false) - n, err := hostPwritev(h.fd, srcs, int64(offset)) + n, err := hostfd.Pwritev2(h.fd, srcs, int64(offset), 0 /* flags */) ctx.UninterruptibleSleepFinish(false) return n, err } diff --git a/pkg/sentry/fsimpl/gofer/handle_unsafe.go b/pkg/sentry/fsimpl/gofer/handle_unsafe.go deleted file mode 100644 index 19560ab26..000000000 --- a/pkg/sentry/fsimpl/gofer/handle_unsafe.go +++ /dev/null @@ -1,66 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package gofer - -import ( - "syscall" - "unsafe" - - "gvisor.dev/gvisor/pkg/safemem" -) - -// Preconditions: !dsts.IsEmpty(). -func hostPreadv(fd int32, dsts safemem.BlockSeq, off int64) (uint64, error) { - // No buffering is necessary regardless of safecopy; host syscalls will - // return EFAULT if appropriate, instead of raising SIGBUS. - if dsts.NumBlocks() == 1 { - // Use pread() instead of preadv() to avoid iovec allocation and - // copying. - dst := dsts.Head() - n, _, e := syscall.Syscall6(syscall.SYS_PREAD64, uintptr(fd), dst.Addr(), uintptr(dst.Len()), uintptr(off), 0, 0) - if e != 0 { - return 0, e - } - return uint64(n), nil - } - iovs := safemem.IovecsFromBlockSeq(dsts) - n, _, e := syscall.Syscall6(syscall.SYS_PREADV, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs)), uintptr(off), 0, 0) - if e != 0 { - return 0, e - } - return uint64(n), nil -} - -// Preconditions: !srcs.IsEmpty(). -func hostPwritev(fd int32, srcs safemem.BlockSeq, off int64) (uint64, error) { - // No buffering is necessary regardless of safecopy; host syscalls will - // return EFAULT if appropriate, instead of raising SIGBUS. - if srcs.NumBlocks() == 1 { - // Use pwrite() instead of pwritev() to avoid iovec allocation and - // copying. - src := srcs.Head() - n, _, e := syscall.Syscall6(syscall.SYS_PWRITE64, uintptr(fd), src.Addr(), uintptr(src.Len()), uintptr(off), 0, 0) - if e != 0 { - return 0, e - } - return uint64(n), nil - } - iovs := safemem.IovecsFromBlockSeq(srcs) - n, _, e := syscall.Syscall6(syscall.SYS_PWRITEV, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs)), uintptr(off), 0, 0) - if e != 0 { - return 0, e - } - return uint64(n), nil -} diff --git a/pkg/sentry/fsimpl/host/BUILD b/pkg/sentry/fsimpl/host/BUILD index 82e1fb74b..44dd9f672 100644 --- a/pkg/sentry/fsimpl/host/BUILD +++ b/pkg/sentry/fsimpl/host/BUILD @@ -15,12 +15,11 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/context", - "//pkg/fd", "//pkg/log", "//pkg/refs", - "//pkg/safemem", "//pkg/sentry/arch", "//pkg/sentry/fsimpl/kernfs", + "//pkg/sentry/hostfd", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/memmap", diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index fe14476f1..ae94cfa6e 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -25,11 +25,10 @@ import ( "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fd" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/hostfd" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/vfs" @@ -492,19 +491,9 @@ func readFromHostFD(ctx context.Context, hostFD int, dst usermem.IOSequence, off if flags != 0 { return 0, syserror.EOPNOTSUPP } - - var reader safemem.Reader - if offset == -1 { - reader = safemem.FromIOReader{fd.NewReadWriter(hostFD)} - } else { - reader = safemem.FromVecReaderFunc{ - func(srcs [][]byte) (int64, error) { - n, err := unix.Preadv(hostFD, srcs, offset) - return int64(n), err - }, - } - } + reader := hostfd.GetReadWriterAt(int32(hostFD), offset, flags) n, err := dst.CopyOutFrom(ctx, reader) + hostfd.PutReadWriterAt(reader) return int64(n), err } @@ -542,19 +531,9 @@ func writeToHostFD(ctx context.Context, hostFD int, src usermem.IOSequence, offs if flags != 0 { return 0, syserror.EOPNOTSUPP } - - var writer safemem.Writer - if offset == -1 { - writer = safemem.FromIOWriter{fd.NewReadWriter(hostFD)} - } else { - writer = safemem.FromVecWriterFunc{ - func(srcs [][]byte) (int64, error) { - n, err := unix.Pwritev(hostFD, srcs, offset) - return int64(n), err - }, - } - } + writer := hostfd.GetReadWriterAt(int32(hostFD), offset, flags) n, err := src.CopyInTo(ctx, writer) + hostfd.PutReadWriterAt(writer) return int64(n), err } diff --git a/pkg/sentry/hostfd/BUILD b/pkg/sentry/hostfd/BUILD new file mode 100644 index 000000000..364a78306 --- /dev/null +++ b/pkg/sentry/hostfd/BUILD @@ -0,0 +1,17 @@ +load("//tools:defs.bzl", "go_library") + +licenses(["notice"]) + +go_library( + name = "hostfd", + srcs = [ + "hostfd.go", + "hostfd_unsafe.go", + ], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/safemem", + "//pkg/sync", + "@org_golang_x_sys//unix:go_default_library", + ], +) diff --git a/pkg/sentry/hostfd/hostfd.go b/pkg/sentry/hostfd/hostfd.go new file mode 100644 index 000000000..70dd9cafb --- /dev/null +++ b/pkg/sentry/hostfd/hostfd.go @@ -0,0 +1,84 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package hostfd provides efficient I/O with host file descriptors. +package hostfd + +import ( + "gvisor.dev/gvisor/pkg/safemem" + "gvisor.dev/gvisor/pkg/sync" +) + +// ReadWriterAt implements safemem.Reader and safemem.Writer by reading from +// and writing to a host file descriptor respectively. ReadWriterAts should be +// obtained by calling GetReadWriterAt. +// +// Clients should usually prefer to use Preadv2 and Pwritev2 directly. +type ReadWriterAt struct { + fd int32 + offset int64 + flags uint32 +} + +var rwpool = sync.Pool{ + New: func() interface{} { + return &ReadWriterAt{} + }, +} + +// GetReadWriterAt returns a ReadWriterAt that reads from / writes to the given +// host file descriptor, starting at the given offset and using the given +// preadv2(2)/pwritev2(2) flags. If offset is -1, the host file descriptor's +// offset is used instead. Users are responsible for ensuring that fd remains +// valid for the lifetime of the returned ReadWriterAt, and must call +// PutReadWriterAt when it is no longer needed. +func GetReadWriterAt(fd int32, offset int64, flags uint32) *ReadWriterAt { + rw := rwpool.Get().(*ReadWriterAt) + *rw = ReadWriterAt{ + fd: fd, + offset: offset, + flags: flags, + } + return rw +} + +// PutReadWriterAt releases a ReadWriterAt returned by a previous call to +// GetReadWriterAt that is no longer in use. +func PutReadWriterAt(rw *ReadWriterAt) { + rwpool.Put(rw) +} + +// ReadToBlocks implements safemem.Reader.ReadToBlocks. +func (rw *ReadWriterAt) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) { + if dsts.IsEmpty() { + return 0, nil + } + n, err := Preadv2(rw.fd, dsts, rw.offset, rw.flags) + if rw.offset >= 0 { + rw.offset += int64(n) + } + return n, err +} + +// WriteFromBlocks implements safemem.Writer.WriteFromBlocks. +func (rw *ReadWriterAt) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) { + if srcs.IsEmpty() { + return 0, nil + } + n, err := Pwritev2(rw.fd, srcs, rw.offset, rw.flags) + if rw.offset >= 0 { + rw.offset += int64(n) + } + return n, err +} diff --git a/pkg/sentry/hostfd/hostfd_unsafe.go b/pkg/sentry/hostfd/hostfd_unsafe.go new file mode 100644 index 000000000..5e9e60fc4 --- /dev/null +++ b/pkg/sentry/hostfd/hostfd_unsafe.go @@ -0,0 +1,107 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package hostfd + +import ( + "io" + "syscall" + "unsafe" + + "golang.org/x/sys/unix" + "gvisor.dev/gvisor/pkg/safemem" +) + +// Preadv2 reads up to dsts.NumBytes() bytes from host file descriptor fd into +// dsts. offset and flags are interpreted as for preadv2(2). +// +// Preconditions: !dsts.IsEmpty(). +func Preadv2(fd int32, dsts safemem.BlockSeq, offset int64, flags uint32) (uint64, error) { + // No buffering is necessary regardless of safecopy; host syscalls will + // return EFAULT if appropriate, instead of raising SIGBUS. + var ( + n uintptr + e syscall.Errno + ) + // Avoid preadv2(2) if possible, since it's relatively new and thus least + // likely to be supported by the host kernel. + if flags == 0 { + if dsts.NumBlocks() == 1 { + // Use read() or pread() to avoid iovec allocation and copying. + dst := dsts.Head() + if offset == -1 { + n, _, e = syscall.Syscall(unix.SYS_READ, uintptr(fd), dst.Addr(), uintptr(dst.Len())) + } else { + n, _, e = syscall.Syscall6(unix.SYS_PREAD64, uintptr(fd), dst.Addr(), uintptr(dst.Len()), uintptr(offset), 0 /* pos_h */, 0 /* unused */) + } + } else { + iovs := safemem.IovecsFromBlockSeq(dsts) + if offset == -1 { + n, _, e = syscall.Syscall(unix.SYS_READV, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs))) + } else { + n, _, e = syscall.Syscall6(unix.SYS_PREADV, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs)), uintptr(offset), 0 /* pos_h */, 0 /* unused */) + } + } + } else { + iovs := safemem.IovecsFromBlockSeq(dsts) + n, _, e = syscall.Syscall6(unix.SYS_PREADV2, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs)), uintptr(offset), 0 /* pos_h */, uintptr(flags)) + } + if e != 0 { + return 0, e + } + if n == 0 { + return 0, io.EOF + } + return uint64(n), nil +} + +// Pwritev2 writes up to srcs.NumBytes() from srcs into host file descriptor +// fd. offset and flags are interpreted as for pwritev2(2). +// +// Preconditions: !srcs.IsEmpty(). +func Pwritev2(fd int32, srcs safemem.BlockSeq, offset int64, flags uint32) (uint64, error) { + // No buffering is necessary regardless of safecopy; host syscalls will + // return EFAULT if appropriate, instead of raising SIGBUS. + var ( + n uintptr + e syscall.Errno + ) + // Avoid pwritev2(2) if possible, since it's relatively new and thus least + // likely to be supported by the host kernel. + if flags == 0 { + if srcs.NumBlocks() == 1 { + // Use write() or pwrite() to avoid iovec allocation and copying. + src := srcs.Head() + if offset == -1 { + n, _, e = syscall.Syscall(unix.SYS_WRITE, uintptr(fd), src.Addr(), uintptr(src.Len())) + } else { + n, _, e = syscall.Syscall6(unix.SYS_PWRITE64, uintptr(fd), src.Addr(), uintptr(src.Len()), uintptr(offset), 0 /* pos_h */, 0 /* unused */) + } + } else { + iovs := safemem.IovecsFromBlockSeq(srcs) + if offset == -1 { + n, _, e = syscall.Syscall(unix.SYS_WRITEV, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs))) + } else { + n, _, e = syscall.Syscall6(unix.SYS_PWRITEV, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs)), uintptr(offset), 0 /* pos_h */, 0 /* unused */) + } + } + } else { + iovs := safemem.IovecsFromBlockSeq(srcs) + n, _, e = syscall.Syscall6(unix.SYS_PWRITEV2, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs)), uintptr(offset), 0 /* pos_h */, uintptr(flags)) + } + if e != 0 { + return 0, e + } + return uint64(n), nil +} -- cgit v1.2.3 From 1481499fe27157ad2716c00682f6ad819115a6c7 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Thu, 23 Apr 2020 11:32:08 -0700 Subject: Simplify Docker test infrastructure. This change adds a layer of abstraction around the internal Docker APIs, and eliminates all direct dependencies on Dockerfiles in the infrastructure. A subsequent change will automated the generation of local images (with efficient caching). Note that this change drops the use of bazel container rules, as that experiment does not seem to be viable. PiperOrigin-RevId: 308095430 --- WORKSPACE | 39 - pkg/sentry/fsimpl/ext/BUILD | 2 +- pkg/sentry/fsimpl/ext/ext_test.go | 3 +- pkg/tcpip/transport/tcp/BUILD | 2 +- pkg/tcpip/transport/tcp/tcp_noracedetector_test.go | 2 +- pkg/test/criutil/BUILD | 14 + pkg/test/criutil/criutil.go | 306 +++ pkg/test/dockerutil/BUILD | 14 + pkg/test/dockerutil/dockerutil.go | 581 ++++++ pkg/test/testutil/BUILD | 20 + pkg/test/testutil/testutil.go | 550 ++++++ pkg/test/testutil/testutil_runfiles.go | 75 + runsc/boot/BUILD | 1 + runsc/cmd/BUILD | 2 +- runsc/cmd/capability_test.go | 9 +- runsc/container/BUILD | 6 +- runsc/container/console_test.go | 115 +- runsc/container/container.go | 2 +- runsc/container/container_norace_test.go | 20 + runsc/container/container_race_test.go | 20 + runsc/container/container_test.go | 2046 ++++++++++---------- runsc/container/multi_container_test.go | 1161 +++++------ runsc/container/shared_volume_test.go | 18 +- runsc/container/test_app/BUILD | 21 - runsc/container/test_app/fds.go | 185 -- runsc/container/test_app/test_app.go | 394 ---- runsc/criutil/BUILD | 11 - runsc/criutil/criutil.go | 277 --- runsc/dockerutil/BUILD | 14 - runsc/dockerutil/dockerutil.go | 486 ----- runsc/testutil/BUILD | 21 - runsc/testutil/testutil.go | 433 ----- runsc/testutil/testutil_runfiles.go | 75 - scripts/iptables_tests.sh | 13 +- test/cmd/test_app/BUILD | 21 + test/cmd/test_app/fds.go | 185 ++ test/cmd/test_app/test_app.go | 394 ++++ test/e2e/BUILD | 4 +- test/e2e/exec_test.go | 193 +- test/e2e/integration_test.go | 233 ++- test/e2e/regression_test.go | 18 +- test/image/BUILD | 4 +- test/image/image_test.go | 195 +- test/image/ruby.sh | 0 test/iptables/BUILD | 8 +- test/iptables/README.md | 2 +- test/iptables/iptables.go | 7 + test/iptables/iptables_test.go | 271 +-- test/iptables/iptables_util.go | 2 +- test/iptables/runner/BUILD | 17 +- test/iptables/runner/main.go | 3 + test/packetdrill/packetdrill_test.sh | 25 +- test/packetimpact/testbench/dut.go | 2 +- test/packetimpact/tests/test_runner.sh | 24 +- test/root/BUILD | 8 +- test/root/cgroup_test.go | 114 +- test/root/chroot_test.go | 20 +- test/root/crictl_test.go | 192 +- test/root/main_test.go | 2 +- test/root/oom_score_adj_test.go | 78 +- test/root/runsc_test.go | 2 +- test/root/testdata/BUILD | 18 - test/root/testdata/busybox.go | 32 - test/root/testdata/containerd_config.go | 39 - test/root/testdata/httpd.go | 32 - test/root/testdata/httpd_mount_paths.go | 53 - test/root/testdata/sandbox.go | 30 - test/root/testdata/simple.go | 41 - test/runner/BUILD | 2 +- test/runner/runner.go | 12 +- test/runtimes/BUILD | 22 +- test/runtimes/README.md | 56 - test/runtimes/blacklist_test.go | 37 - test/runtimes/build_defs.bzl | 75 - test/runtimes/defs.bzl | 79 + test/runtimes/images/proctor/BUILD | 26 - test/runtimes/images/proctor/go.go | 90 - test/runtimes/images/proctor/java.go | 71 - test/runtimes/images/proctor/nodejs.go | 46 - test/runtimes/images/proctor/php.go | 42 - test/runtimes/images/proctor/proctor.go | 163 -- test/runtimes/images/proctor/proctor_test.go | 127 -- test/runtimes/images/proctor/python.go | 49 - test/runtimes/proctor/BUILD | 27 + test/runtimes/proctor/go.go | 90 + test/runtimes/proctor/java.go | 71 + test/runtimes/proctor/nodejs.go | 46 + test/runtimes/proctor/php.go | 42 + test/runtimes/proctor/proctor.go | 163 ++ test/runtimes/proctor/proctor_test.go | 127 ++ test/runtimes/proctor/python.go | 49 + test/runtimes/runner.go | 196 -- test/runtimes/runner.sh | 35 - test/runtimes/runner/BUILD | 21 + test/runtimes/runner/blacklist_test.go | 37 + test/runtimes/runner/main.go | 189 ++ tools/bazeldefs/defs.bzl | 4 - tools/defs.bzl | 4 +- 98 files changed, 5512 insertions(+), 5693 deletions(-) create mode 100644 pkg/test/criutil/BUILD create mode 100644 pkg/test/criutil/criutil.go create mode 100644 pkg/test/dockerutil/BUILD create mode 100644 pkg/test/dockerutil/dockerutil.go create mode 100644 pkg/test/testutil/BUILD create mode 100644 pkg/test/testutil/testutil.go create mode 100644 pkg/test/testutil/testutil_runfiles.go create mode 100644 runsc/container/container_norace_test.go create mode 100644 runsc/container/container_race_test.go delete mode 100644 runsc/container/test_app/BUILD delete mode 100644 runsc/container/test_app/fds.go delete mode 100644 runsc/container/test_app/test_app.go delete mode 100644 runsc/criutil/BUILD delete mode 100644 runsc/criutil/criutil.go delete mode 100644 runsc/dockerutil/BUILD delete mode 100644 runsc/dockerutil/dockerutil.go delete mode 100644 runsc/testutil/BUILD delete mode 100644 runsc/testutil/testutil.go delete mode 100644 runsc/testutil/testutil_runfiles.go create mode 100644 test/cmd/test_app/BUILD create mode 100644 test/cmd/test_app/fds.go create mode 100644 test/cmd/test_app/test_app.go mode change 100644 => 100755 test/image/ruby.sh delete mode 100644 test/root/testdata/BUILD delete mode 100644 test/root/testdata/busybox.go delete mode 100644 test/root/testdata/containerd_config.go delete mode 100644 test/root/testdata/httpd.go delete mode 100644 test/root/testdata/httpd_mount_paths.go delete mode 100644 test/root/testdata/sandbox.go delete mode 100644 test/root/testdata/simple.go delete mode 100644 test/runtimes/README.md delete mode 100644 test/runtimes/blacklist_test.go delete mode 100644 test/runtimes/build_defs.bzl create mode 100644 test/runtimes/defs.bzl delete mode 100644 test/runtimes/images/proctor/BUILD delete mode 100644 test/runtimes/images/proctor/go.go delete mode 100644 test/runtimes/images/proctor/java.go delete mode 100644 test/runtimes/images/proctor/nodejs.go delete mode 100644 test/runtimes/images/proctor/php.go delete mode 100644 test/runtimes/images/proctor/proctor.go delete mode 100644 test/runtimes/images/proctor/proctor_test.go delete mode 100644 test/runtimes/images/proctor/python.go create mode 100644 test/runtimes/proctor/BUILD create mode 100644 test/runtimes/proctor/go.go create mode 100644 test/runtimes/proctor/java.go create mode 100644 test/runtimes/proctor/nodejs.go create mode 100644 test/runtimes/proctor/php.go create mode 100644 test/runtimes/proctor/proctor.go create mode 100644 test/runtimes/proctor/proctor_test.go create mode 100644 test/runtimes/proctor/python.go delete mode 100644 test/runtimes/runner.go delete mode 100755 test/runtimes/runner.sh create mode 100644 test/runtimes/runner/BUILD create mode 100644 test/runtimes/runner/blacklist_test.go create mode 100644 test/runtimes/runner/main.go (limited to 'pkg/sentry') diff --git a/WORKSPACE b/WORKSPACE index b895647fb..3bf5cc9c1 100644 --- a/WORKSPACE +++ b/WORKSPACE @@ -127,45 +127,6 @@ load("@rules_pkg//:deps.bzl", "rules_pkg_dependencies") rules_pkg_dependencies() -# Container rules. -http_archive( - name = "io_bazel_rules_docker", - sha256 = "14ac30773fdb393ddec90e158c9ec7ebb3f8a4fd533ec2abbfd8789ad81a284b", - strip_prefix = "rules_docker-0.12.1", - urls = ["https://github.com/bazelbuild/rules_docker/releases/download/v0.12.1/rules_docker-v0.12.1.tar.gz"], -) - -load( - "@io_bazel_rules_docker//repositories:repositories.bzl", - container_repositories = "repositories", -) - -container_repositories() - -load("@io_bazel_rules_docker//repositories:deps.bzl", container_deps = "deps") - -container_deps() - -load( - "@io_bazel_rules_docker//container:container.bzl", - "container_pull", -) - -# This container is built from the Dockerfile in test/iptables/runner. -container_pull( - name = "iptables-test", - digest = "sha256:a137d692a2eb9fc7bf95c5f4a568da090e2c31098e93634421ed88f3a3f1db65", - registry = "gcr.io", - repository = "gvisor-presubmit/iptables-test", -) - -load( - "@io_bazel_rules_docker//go:image.bzl", - _go_image_repos = "repositories", -) - -_go_image_repos() - # Load C++ grpc rules. http_archive( name = "com_github_grpc_grpc", diff --git a/pkg/sentry/fsimpl/ext/BUILD b/pkg/sentry/fsimpl/ext/BUILD index a4947c480..ff861d0fe 100644 --- a/pkg/sentry/fsimpl/ext/BUILD +++ b/pkg/sentry/fsimpl/ext/BUILD @@ -93,8 +93,8 @@ go_test( "//pkg/sentry/kernel/auth", "//pkg/sentry/vfs", "//pkg/syserror", + "//pkg/test/testutil", "//pkg/usermem", - "//runsc/testutil", "@com_github_google_go-cmp//cmp:go_default_library", "@com_github_google_go-cmp//cmp/cmpopts:go_default_library", ], diff --git a/pkg/sentry/fsimpl/ext/ext_test.go b/pkg/sentry/fsimpl/ext/ext_test.go index 29bb73765..64e9a579f 100644 --- a/pkg/sentry/fsimpl/ext/ext_test.go +++ b/pkg/sentry/fsimpl/ext/ext_test.go @@ -32,9 +32,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/test/testutil" "gvisor.dev/gvisor/pkg/usermem" - - "gvisor.dev/gvisor/runsc/testutil" ) const ( diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD index 61426623c..f2aa69069 100644 --- a/pkg/tcpip/transport/tcp/BUILD +++ b/pkg/tcpip/transport/tcp/BUILD @@ -105,8 +105,8 @@ go_test( "//pkg/tcpip/seqnum", "//pkg/tcpip/stack", "//pkg/tcpip/transport/tcp/testing/context", + "//pkg/test/testutil", "//pkg/waiter", - "//runsc/testutil", ], ) diff --git a/pkg/tcpip/transport/tcp/tcp_noracedetector_test.go b/pkg/tcpip/transport/tcp/tcp_noracedetector_test.go index 359a75e73..5fe23113b 100644 --- a/pkg/tcpip/transport/tcp/tcp_noracedetector_test.go +++ b/pkg/tcpip/transport/tcp/tcp_noracedetector_test.go @@ -31,7 +31,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/transport/tcp" "gvisor.dev/gvisor/pkg/tcpip/transport/tcp/testing/context" - "gvisor.dev/gvisor/runsc/testutil" + "gvisor.dev/gvisor/pkg/test/testutil" ) func TestFastRecovery(t *testing.T) { diff --git a/pkg/test/criutil/BUILD b/pkg/test/criutil/BUILD new file mode 100644 index 000000000..a7b082cee --- /dev/null +++ b/pkg/test/criutil/BUILD @@ -0,0 +1,14 @@ +load("//tools:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_library( + name = "criutil", + testonly = 1, + srcs = ["criutil.go"], + visibility = ["//:sandbox"], + deps = [ + "//pkg/test/dockerutil", + "//pkg/test/testutil", + ], +) diff --git a/pkg/test/criutil/criutil.go b/pkg/test/criutil/criutil.go new file mode 100644 index 000000000..bebebb48e --- /dev/null +++ b/pkg/test/criutil/criutil.go @@ -0,0 +1,306 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package criutil contains utility functions for interacting with the +// Container Runtime Interface (CRI), principally via the crictl command line +// tool. This requires critools to be installed on the local system. +package criutil + +import ( + "encoding/json" + "fmt" + "os" + "os/exec" + "strings" + "time" + + "gvisor.dev/gvisor/pkg/test/dockerutil" + "gvisor.dev/gvisor/pkg/test/testutil" +) + +// Crictl contains information required to run the crictl utility. +type Crictl struct { + logger testutil.Logger + endpoint string + cleanup []func() +} + +// resolvePath attempts to find binary paths. It may set the path to invalid, +// which will cause the execution to fail with a sensible error. +func resolvePath(executable string) string { + guess, err := exec.LookPath(executable) + if err != nil { + guess = fmt.Sprintf("/usr/local/bin/%s", executable) + } + return guess +} + +// NewCrictl returns a Crictl configured with a timeout and an endpoint over +// which it will talk to containerd. +func NewCrictl(logger testutil.Logger, endpoint string) *Crictl { + // Attempt to find the executable, but don't bother propagating the + // error at this point. The first command executed will return with a + // binary not found error. + return &Crictl{ + logger: logger, + endpoint: endpoint, + } +} + +// CleanUp executes cleanup functions. +func (cc *Crictl) CleanUp() { + for _, c := range cc.cleanup { + c() + } + cc.cleanup = nil +} + +// RunPod creates a sandbox. It corresponds to `crictl runp`. +func (cc *Crictl) RunPod(sbSpecFile string) (string, error) { + podID, err := cc.run("runp", sbSpecFile) + if err != nil { + return "", fmt.Errorf("runp failed: %v", err) + } + // Strip the trailing newline from crictl output. + return strings.TrimSpace(podID), nil +} + +// Create creates a container within a sandbox. It corresponds to `crictl +// create`. +func (cc *Crictl) Create(podID, contSpecFile, sbSpecFile string) (string, error) { + podID, err := cc.run("create", podID, contSpecFile, sbSpecFile) + if err != nil { + return "", fmt.Errorf("create failed: %v", err) + } + // Strip the trailing newline from crictl output. + return strings.TrimSpace(podID), nil +} + +// Start starts a container. It corresponds to `crictl start`. +func (cc *Crictl) Start(contID string) (string, error) { + output, err := cc.run("start", contID) + if err != nil { + return "", fmt.Errorf("start failed: %v", err) + } + return output, nil +} + +// Stop stops a container. It corresponds to `crictl stop`. +func (cc *Crictl) Stop(contID string) error { + _, err := cc.run("stop", contID) + return err +} + +// Exec execs a program inside a container. It corresponds to `crictl exec`. +func (cc *Crictl) Exec(contID string, args ...string) (string, error) { + a := []string{"exec", contID} + a = append(a, args...) + output, err := cc.run(a...) + if err != nil { + return "", fmt.Errorf("exec failed: %v", err) + } + return output, nil +} + +// Rm removes a container. It corresponds to `crictl rm`. +func (cc *Crictl) Rm(contID string) error { + _, err := cc.run("rm", contID) + return err +} + +// StopPod stops a pod. It corresponds to `crictl stopp`. +func (cc *Crictl) StopPod(podID string) error { + _, err := cc.run("stopp", podID) + return err +} + +// containsConfig is a minimal copy of +// https://github.com/kubernetes/kubernetes/blob/master/pkg/kubelet/apis/cri/runtime/v1alpha2/api.proto +// It only contains fields needed for testing. +type containerConfig struct { + Status containerStatus +} + +type containerStatus struct { + Network containerNetwork +} + +type containerNetwork struct { + IP string +} + +// PodIP returns a pod's IP address. +func (cc *Crictl) PodIP(podID string) (string, error) { + output, err := cc.run("inspectp", podID) + if err != nil { + return "", err + } + conf := &containerConfig{} + if err := json.Unmarshal([]byte(output), conf); err != nil { + return "", fmt.Errorf("failed to unmarshal JSON: %v, %s", err, output) + } + if conf.Status.Network.IP == "" { + return "", fmt.Errorf("no IP found in config: %s", output) + } + return conf.Status.Network.IP, nil +} + +// RmPod removes a container. It corresponds to `crictl rmp`. +func (cc *Crictl) RmPod(podID string) error { + _, err := cc.run("rmp", podID) + return err +} + +// Import imports the given container from the local Docker instance. +func (cc *Crictl) Import(image string) error { + // Note that we provide a 10 minute timeout after connect because we may + // be pushing a lot of bytes in order to import the image. The connect + // timeout stays the same and is inherited from the Crictl instance. + cmd := testutil.Command(cc.logger, + resolvePath("ctr"), + fmt.Sprintf("--connect-timeout=%s", 30*time.Second), + fmt.Sprintf("--address=%s", cc.endpoint), + "-n", "k8s.io", "images", "import", "-") + cmd.Stderr = os.Stderr // Pass through errors. + + // Create a pipe and start the program. + w, err := cmd.StdinPipe() + if err != nil { + return err + } + if err := cmd.Start(); err != nil { + return err + } + + // Save the image on the other end. + if err := dockerutil.Save(cc.logger, image, w); err != nil { + cmd.Wait() + return err + } + + // Close our pipe reference & see if it was loaded. + if err := w.Close(); err != nil { + return w.Close() + } + + return cmd.Wait() +} + +// StartContainer pulls the given image ands starts the container in the +// sandbox with the given podID. +// +// Note that the image will always be imported from the local docker daemon. +func (cc *Crictl) StartContainer(podID, image, sbSpec, contSpec string) (string, error) { + if err := cc.Import(image); err != nil { + return "", err + } + + // Write the specs to files that can be read by crictl. + sbSpecFile, cleanup, err := testutil.WriteTmpFile("sbSpec", sbSpec) + if err != nil { + return "", fmt.Errorf("failed to write sandbox spec: %v", err) + } + cc.cleanup = append(cc.cleanup, cleanup) + contSpecFile, cleanup, err := testutil.WriteTmpFile("contSpec", contSpec) + if err != nil { + return "", fmt.Errorf("failed to write container spec: %v", err) + } + cc.cleanup = append(cc.cleanup, cleanup) + + return cc.startContainer(podID, image, sbSpecFile, contSpecFile) +} + +func (cc *Crictl) startContainer(podID, image, sbSpecFile, contSpecFile string) (string, error) { + contID, err := cc.Create(podID, contSpecFile, sbSpecFile) + if err != nil { + return "", fmt.Errorf("failed to create container in pod %q: %v", podID, err) + } + + if _, err := cc.Start(contID); err != nil { + return "", fmt.Errorf("failed to start container %q in pod %q: %v", contID, podID, err) + } + + return contID, nil +} + +// StopContainer stops and deletes the container with the given container ID. +func (cc *Crictl) StopContainer(contID string) error { + if err := cc.Stop(contID); err != nil { + return fmt.Errorf("failed to stop container %q: %v", contID, err) + } + + if err := cc.Rm(contID); err != nil { + return fmt.Errorf("failed to remove container %q: %v", contID, err) + } + + return nil +} + +// StartPodAndContainer starts a sandbox and container in that sandbox. It +// returns the pod ID and container ID. +func (cc *Crictl) StartPodAndContainer(image, sbSpec, contSpec string) (string, string, error) { + if err := cc.Import(image); err != nil { + return "", "", err + } + + // Write the specs to files that can be read by crictl. + sbSpecFile, cleanup, err := testutil.WriteTmpFile("sbSpec", sbSpec) + if err != nil { + return "", "", fmt.Errorf("failed to write sandbox spec: %v", err) + } + cc.cleanup = append(cc.cleanup, cleanup) + contSpecFile, cleanup, err := testutil.WriteTmpFile("contSpec", contSpec) + if err != nil { + return "", "", fmt.Errorf("failed to write container spec: %v", err) + } + cc.cleanup = append(cc.cleanup, cleanup) + + podID, err := cc.RunPod(sbSpecFile) + if err != nil { + return "", "", err + } + + contID, err := cc.startContainer(podID, image, sbSpecFile, contSpecFile) + + return podID, contID, err +} + +// StopPodAndContainer stops a container and pod. +func (cc *Crictl) StopPodAndContainer(podID, contID string) error { + if err := cc.StopContainer(contID); err != nil { + return fmt.Errorf("failed to stop container %q in pod %q: %v", contID, podID, err) + } + + if err := cc.StopPod(podID); err != nil { + return fmt.Errorf("failed to stop pod %q: %v", podID, err) + } + + if err := cc.RmPod(podID); err != nil { + return fmt.Errorf("failed to remove pod %q: %v", podID, err) + } + + return nil +} + +// run runs crictl with the given args. +func (cc *Crictl) run(args ...string) (string, error) { + defaultArgs := []string{ + resolvePath("crictl"), + "--image-endpoint", fmt.Sprintf("unix://%s", cc.endpoint), + "--runtime-endpoint", fmt.Sprintf("unix://%s", cc.endpoint), + } + fullArgs := append(defaultArgs, args...) + out, err := testutil.Command(cc.logger, fullArgs...).CombinedOutput() + return string(out), err +} diff --git a/pkg/test/dockerutil/BUILD b/pkg/test/dockerutil/BUILD new file mode 100644 index 000000000..7c8758e35 --- /dev/null +++ b/pkg/test/dockerutil/BUILD @@ -0,0 +1,14 @@ +load("//tools:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_library( + name = "dockerutil", + testonly = 1, + srcs = ["dockerutil.go"], + visibility = ["//:sandbox"], + deps = [ + "//pkg/test/testutil", + "@com_github_kr_pty//:go_default_library", + ], +) diff --git a/pkg/test/dockerutil/dockerutil.go b/pkg/test/dockerutil/dockerutil.go new file mode 100644 index 000000000..baa8fc2f2 --- /dev/null +++ b/pkg/test/dockerutil/dockerutil.go @@ -0,0 +1,581 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package dockerutil is a collection of utility functions. +package dockerutil + +import ( + "encoding/json" + "flag" + "fmt" + "io" + "io/ioutil" + "log" + "net" + "os" + "os/exec" + "path" + "regexp" + "strconv" + "strings" + "syscall" + "time" + + "github.com/kr/pty" + "gvisor.dev/gvisor/pkg/test/testutil" +) + +var ( + // runtime is the runtime to use for tests. This will be applied to all + // containers. Note that the default here ("runsc") corresponds to the + // default used by the installations. This is important, because the + // default installer for vm_tests (in tools/installers:head, invoked + // via tools/vm:defs.bzl) will install with this name. So without + // changing anything, tests should have a runsc runtime available to + // them. Otherwise installers should update the existing runtime + // instead of installing a new one. + runtime = flag.String("runtime", "runsc", "specify which runtime to use") + + // config is the default Docker daemon configuration path. + config = flag.String("config_path", "/etc/docker/daemon.json", "configuration file for reading paths") +) + +// EnsureSupportedDockerVersion checks if correct docker is installed. +// +// This logs directly to stderr, as it is typically called from a Main wrapper. +func EnsureSupportedDockerVersion() { + cmd := exec.Command("docker", "version") + out, err := cmd.CombinedOutput() + if err != nil { + log.Fatalf("error running %q: %v", "docker version", err) + } + re := regexp.MustCompile(`Version:\s+(\d+)\.(\d+)\.\d.*`) + matches := re.FindStringSubmatch(string(out)) + if len(matches) != 3 { + log.Fatalf("Invalid docker output: %s", out) + } + major, _ := strconv.Atoi(matches[1]) + minor, _ := strconv.Atoi(matches[2]) + if major < 17 || (major == 17 && minor < 9) { + log.Fatalf("Docker version 17.09.0 or greater is required, found: %02d.%02d", major, minor) + } +} + +// RuntimePath returns the binary path for the current runtime. +func RuntimePath() (string, error) { + // Read the configuration data; the file must exist. + configBytes, err := ioutil.ReadFile(*config) + if err != nil { + return "", err + } + + // Unmarshal the configuration. + c := make(map[string]interface{}) + if err := json.Unmarshal(configBytes, &c); err != nil { + return "", err + } + + // Decode the expected configuration. + r, ok := c["runtimes"] + if !ok { + return "", fmt.Errorf("no runtimes declared: %v", c) + } + rs, ok := r.(map[string]interface{}) + if !ok { + // The runtimes are not a map. + return "", fmt.Errorf("unexpected format: %v", c) + } + r, ok = rs[*runtime] + if !ok { + // The expected runtime is not declared. + return "", fmt.Errorf("runtime %q not found: %v", *runtime, c) + } + rs, ok = r.(map[string]interface{}) + if !ok { + // The runtime is not a map. + return "", fmt.Errorf("unexpected format: %v", c) + } + p, ok := rs["path"].(string) + if !ok { + // The runtime does not declare a path. + return "", fmt.Errorf("unexpected format: %v", c) + } + return p, nil +} + +// Save exports a container image to the given Writer. +// +// Note that the writer should be actively consuming the output, otherwise it +// is not guaranteed that the Save will make any progress and the call may +// stall indefinitely. +// +// This is called by criutil in order to import imports. +func Save(logger testutil.Logger, image string, w io.Writer) error { + cmd := testutil.Command(logger, "docker", "save", testutil.ImageByName(image)) + cmd.Stdout = w // Send directly to the writer. + return cmd.Run() +} + +// MountMode describes if the mount should be ro or rw. +type MountMode int + +const ( + // ReadOnly is what the name says. + ReadOnly MountMode = iota + // ReadWrite is what the name says. + ReadWrite +) + +// String returns the mount mode argument for this MountMode. +func (m MountMode) String() string { + switch m { + case ReadOnly: + return "ro" + case ReadWrite: + return "rw" + } + panic(fmt.Sprintf("invalid mode: %d", m)) +} + +// Docker contains the name and the runtime of a docker container. +type Docker struct { + logger testutil.Logger + Runtime string + Name string + copyErr error + mounts []string + cleanups []func() +} + +// MakeDocker sets up the struct for a Docker container. +// +// Names of containers will be unique. +func MakeDocker(logger testutil.Logger) *Docker { + return &Docker{ + logger: logger, + Name: testutil.RandomID(logger.Name()), + Runtime: *runtime, + } +} + +// Mount mounts the given source and makes it available in the container. +func (d *Docker) Mount(target, source string, mode MountMode) { + d.mounts = append(d.mounts, fmt.Sprintf("-v=%s:%s:%v", source, target, mode)) +} + +// CopyFiles copies in and mounts the given files. They are always ReadOnly. +func (d *Docker) CopyFiles(target string, sources ...string) { + dir, err := ioutil.TempDir("", d.Name) + if err != nil { + d.copyErr = fmt.Errorf("ioutil.TempDir failed: %v", err) + return + } + d.cleanups = append(d.cleanups, func() { os.RemoveAll(dir) }) + if err := os.Chmod(dir, 0755); err != nil { + d.copyErr = fmt.Errorf("os.Chmod(%q, 0755) failed: %v", dir, err) + return + } + for _, name := range sources { + src, err := testutil.FindFile(name) + if err != nil { + d.copyErr = fmt.Errorf("testutil.FindFile(%q) failed: %v", name, err) + return + } + dst := path.Join(dir, path.Base(name)) + if err := testutil.Copy(src, dst); err != nil { + d.copyErr = fmt.Errorf("testutil.Copy(%q, %q) failed: %v", src, dst, err) + return + } + d.logger.Logf("copy: %s -> %s", src, dst) + } + d.Mount(target, dir, ReadOnly) +} + +// Link links the given target. +func (d *Docker) Link(target string, source *Docker) { + d.mounts = append(d.mounts, fmt.Sprintf("--link=%s:%s", source.Name, target)) +} + +// RunOpts are options for running a container. +type RunOpts struct { + // Image is the image relative to images/. This will be mangled + // appropriately, to ensure that only first-party images are used. + Image string + + // Memory is the memory limit in kB. + Memory int + + // Ports are the ports to be allocated. + Ports []int + + // WorkDir sets the working directory. + WorkDir string + + // ReadOnly sets the read-only flag. + ReadOnly bool + + // Env are additional environment variables. + Env []string + + // User is the user to use. + User string + + // Privileged enables privileged mode. + Privileged bool + + // CapAdd are the extra set of capabilities to add. + CapAdd []string + + // CapDrop are the extra set of capabilities to drop. + CapDrop []string + + // Pty indicates that a pty will be allocated. If this is non-nil, then + // this will run after start-up with the *exec.Command and Pty file + // passed in to the function. + Pty func(*exec.Cmd, *os.File) + + // Foreground indicates that the container should be run in the + // foreground. If this is true, then the output will be available as a + // return value from the Run function. + Foreground bool + + // Extra are extra arguments that may be passed. + Extra []string +} + +// args returns common arguments. +// +// Note that this does not define the complete behavior. +func (d *Docker) argsFor(r *RunOpts, command string, p []string) (rv []string) { + isExec := command == "exec" + isRun := command == "run" + + if isRun || isExec { + rv = append(rv, "-i") + } + if r.Pty != nil { + rv = append(rv, "-t") + } + if r.User != "" { + rv = append(rv, fmt.Sprintf("--user=%s", r.User)) + } + if r.Privileged { + rv = append(rv, "--privileged") + } + for _, c := range r.CapAdd { + rv = append(rv, fmt.Sprintf("--cap-add=%s", c)) + } + for _, c := range r.CapDrop { + rv = append(rv, fmt.Sprintf("--cap-drop=%s", c)) + } + for _, e := range r.Env { + rv = append(rv, fmt.Sprintf("--env=%s", e)) + } + if r.WorkDir != "" { + rv = append(rv, fmt.Sprintf("--workdir=%s", r.WorkDir)) + } + if !isExec { + if r.Memory != 0 { + rv = append(rv, fmt.Sprintf("--memory=%dk", r.Memory)) + } + for _, p := range r.Ports { + rv = append(rv, fmt.Sprintf("--publish=%d", p)) + } + if r.ReadOnly { + rv = append(rv, fmt.Sprintf("--read-only")) + } + if len(p) > 0 { + rv = append(rv, "--entrypoint=") + } + } + + // Always attach the test environment & Extra. + rv = append(rv, fmt.Sprintf("--env=RUNSC_TEST_NAME=%s", d.Name)) + rv = append(rv, r.Extra...) + + // Attach necessary bits. + if isExec { + rv = append(rv, d.Name) + } else { + rv = append(rv, d.mounts...) + rv = append(rv, fmt.Sprintf("--runtime=%s", d.Runtime)) + rv = append(rv, fmt.Sprintf("--name=%s", d.Name)) + rv = append(rv, testutil.ImageByName(r.Image)) + } + + // Attach other arguments. + rv = append(rv, p...) + return rv +} + +// run runs a complete command. +func (d *Docker) run(r RunOpts, command string, p ...string) (string, error) { + if d.copyErr != nil { + return "", d.copyErr + } + basicArgs := []string{"docker"} + if command == "spawn" { + command = "run" + basicArgs = append(basicArgs, command) + basicArgs = append(basicArgs, "-d") + } else { + basicArgs = append(basicArgs, command) + } + customArgs := d.argsFor(&r, command, p) + cmd := testutil.Command(d.logger, append(basicArgs, customArgs...)...) + if r.Pty != nil { + // If allocating a terminal, then we just ignore the output + // from the command. + ptmx, err := pty.Start(cmd.Cmd) + if err != nil { + return "", err + } + defer cmd.Wait() // Best effort. + r.Pty(cmd.Cmd, ptmx) + } else { + // Can't support PTY or streaming. + out, err := cmd.CombinedOutput() + return string(out), err + } + return "", nil +} + +// Create calls 'docker create' with the arguments provided. +func (d *Docker) Create(r RunOpts, args ...string) error { + _, err := d.run(r, "create", args...) + return err +} + +// Start calls 'docker start'. +func (d *Docker) Start() error { + return testutil.Command(d.logger, "docker", "start", d.Name).Run() +} + +// Stop calls 'docker stop'. +func (d *Docker) Stop() error { + return testutil.Command(d.logger, "docker", "stop", d.Name).Run() +} + +// Run calls 'docker run' with the arguments provided. +func (d *Docker) Run(r RunOpts, args ...string) (string, error) { + return d.run(r, "run", args...) +} + +// Spawn starts the container and detaches. +func (d *Docker) Spawn(r RunOpts, args ...string) error { + _, err := d.run(r, "spawn", args...) + return err +} + +// Logs calls 'docker logs'. +func (d *Docker) Logs() (string, error) { + // Don't capture the output; since it will swamp the logs. + out, err := exec.Command("docker", "logs", d.Name).CombinedOutput() + return string(out), err +} + +// Exec calls 'docker exec' with the arguments provided. +func (d *Docker) Exec(r RunOpts, args ...string) (string, error) { + return d.run(r, "exec", args...) +} + +// Pause calls 'docker pause'. +func (d *Docker) Pause() error { + return testutil.Command(d.logger, "docker", "pause", d.Name).Run() +} + +// Unpause calls 'docker pause'. +func (d *Docker) Unpause() error { + return testutil.Command(d.logger, "docker", "unpause", d.Name).Run() +} + +// Checkpoint calls 'docker checkpoint'. +func (d *Docker) Checkpoint(name string) error { + return testutil.Command(d.logger, "docker", "checkpoint", "create", d.Name, name).Run() +} + +// Restore calls 'docker start --checkname [name]'. +func (d *Docker) Restore(name string) error { + return testutil.Command(d.logger, "docker", "start", fmt.Sprintf("--checkpoint=%s", name), d.Name).Run() +} + +// Kill calls 'docker kill'. +func (d *Docker) Kill() error { + // Skip logging this command, it will likely be an error. + out, err := exec.Command("docker", "kill", d.Name).CombinedOutput() + if err != nil && !strings.Contains(string(out), "is not running") { + return err + } + return nil +} + +// Remove calls 'docker rm'. +func (d *Docker) Remove() error { + return testutil.Command(d.logger, "docker", "rm", d.Name).Run() +} + +// CleanUp kills and deletes the container (best effort). +func (d *Docker) CleanUp() { + // Kill the container. + if err := d.Kill(); err != nil { + // Just log; can't do anything here. + d.logger.Logf("error killing container %q: %v", d.Name, err) + } + // Remove the image. + if err := d.Remove(); err != nil { + d.logger.Logf("error removing container %q: %v", d.Name, err) + } + // Forget all mounts. + d.mounts = nil + // Execute all cleanups. + for _, c := range d.cleanups { + c() + } + d.cleanups = nil +} + +// FindPort returns the host port that is mapped to 'sandboxPort'. This calls +// docker to allocate a free port in the host and prevent conflicts. +func (d *Docker) FindPort(sandboxPort int) (int, error) { + format := fmt.Sprintf(`{{ (index (index .NetworkSettings.Ports "%d/tcp") 0).HostPort }}`, sandboxPort) + out, err := testutil.Command(d.logger, "docker", "inspect", "-f", format, d.Name).CombinedOutput() + if err != nil { + return -1, fmt.Errorf("error retrieving port: %v", err) + } + port, err := strconv.Atoi(strings.TrimSuffix(string(out), "\n")) + if err != nil { + return -1, fmt.Errorf("error parsing port %q: %v", out, err) + } + return port, nil +} + +// FindIP returns the IP address of the container. +func (d *Docker) FindIP() (net.IP, error) { + const format = `{{range .NetworkSettings.Networks}}{{.IPAddress}}{{end}}` + out, err := testutil.Command(d.logger, "docker", "inspect", "-f", format, d.Name).CombinedOutput() + if err != nil { + return net.IP{}, fmt.Errorf("error retrieving IP: %v", err) + } + ip := net.ParseIP(strings.TrimSpace(string(out))) + if ip == nil { + return net.IP{}, fmt.Errorf("invalid IP: %q", string(out)) + } + return ip, nil +} + +// SandboxPid returns the PID to the sandbox process. +func (d *Docker) SandboxPid() (int, error) { + out, err := testutil.Command(d.logger, "docker", "inspect", "-f={{.State.Pid}}", d.Name).CombinedOutput() + if err != nil { + return -1, fmt.Errorf("error retrieving pid: %v", err) + } + pid, err := strconv.Atoi(strings.TrimSuffix(string(out), "\n")) + if err != nil { + return -1, fmt.Errorf("error parsing pid %q: %v", out, err) + } + return pid, nil +} + +// ID returns the container ID. +func (d *Docker) ID() (string, error) { + out, err := testutil.Command(d.logger, "docker", "inspect", "-f={{.Id}}", d.Name).CombinedOutput() + if err != nil { + return "", fmt.Errorf("error retrieving ID: %v", err) + } + return strings.TrimSpace(string(out)), nil +} + +// Wait waits for container to exit, up to the given timeout. Returns error if +// wait fails or timeout is hit. Returns the application return code otherwise. +// Note that the application may have failed even if err == nil, always check +// the exit code. +func (d *Docker) Wait(timeout time.Duration) (syscall.WaitStatus, error) { + timeoutChan := time.After(timeout) + waitChan := make(chan (syscall.WaitStatus)) + errChan := make(chan (error)) + + go func() { + out, err := testutil.Command(d.logger, "docker", "wait", d.Name).CombinedOutput() + if err != nil { + errChan <- fmt.Errorf("error waiting for container %q: %v", d.Name, err) + } + exit, err := strconv.Atoi(strings.TrimSuffix(string(out), "\n")) + if err != nil { + errChan <- fmt.Errorf("error parsing exit code %q: %v", out, err) + } + waitChan <- syscall.WaitStatus(uint32(exit)) + }() + + select { + case ws := <-waitChan: + return ws, nil + case err := <-errChan: + return syscall.WaitStatus(1), err + case <-timeoutChan: + return syscall.WaitStatus(1), fmt.Errorf("timeout waiting for container %q", d.Name) + } +} + +// WaitForOutput calls 'docker logs' to retrieve containers output and searches +// for the given pattern. +func (d *Docker) WaitForOutput(pattern string, timeout time.Duration) (string, error) { + matches, err := d.WaitForOutputSubmatch(pattern, timeout) + if err != nil { + return "", err + } + if len(matches) == 0 { + return "", nil + } + return matches[0], nil +} + +// WaitForOutputSubmatch calls 'docker logs' to retrieve containers output and +// searches for the given pattern. It returns any regexp submatches as well. +func (d *Docker) WaitForOutputSubmatch(pattern string, timeout time.Duration) ([]string, error) { + re := regexp.MustCompile(pattern) + var ( + lastOut string + stopped bool + ) + for exp := time.Now().Add(timeout); time.Now().Before(exp); { + out, err := d.Logs() + if err != nil { + return nil, err + } + if out != lastOut { + if lastOut == "" { + d.logger.Logf("output (start): %s", out) + } else if strings.HasPrefix(out, lastOut) { + d.logger.Logf("output (contn): %s", out[len(lastOut):]) + } else { + d.logger.Logf("output (trunc): %s", out) + } + lastOut = out // Save for future. + if matches := re.FindStringSubmatch(lastOut); matches != nil { + return matches, nil // Success! + } + } else if stopped { + // The sandbox stopped and we looked at the + // logs at least once since determining that. + return nil, fmt.Errorf("no longer running: %v", err) + } else if pid, err := d.SandboxPid(); pid == 0 || err != nil { + // The sandbox may have stopped, but it's + // possible that it has emitted the terminal + // line between the last call to Logs and here. + stopped = true + } + time.Sleep(100 * time.Millisecond) + } + return nil, fmt.Errorf("timeout waiting for output %q: %s", re.String(), lastOut) +} diff --git a/pkg/test/testutil/BUILD b/pkg/test/testutil/BUILD new file mode 100644 index 000000000..03b1b4677 --- /dev/null +++ b/pkg/test/testutil/BUILD @@ -0,0 +1,20 @@ +load("//tools:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_library( + name = "testutil", + testonly = 1, + srcs = [ + "testutil.go", + "testutil_runfiles.go", + ], + visibility = ["//:sandbox"], + deps = [ + "//pkg/sync", + "//runsc/boot", + "//runsc/specutils", + "@com_github_cenkalti_backoff//:go_default_library", + "@com_github_opencontainers_runtime-spec//specs-go:go_default_library", + ], +) diff --git a/pkg/test/testutil/testutil.go b/pkg/test/testutil/testutil.go new file mode 100644 index 000000000..d75ceca3d --- /dev/null +++ b/pkg/test/testutil/testutil.go @@ -0,0 +1,550 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package testutil contains utility functions for runsc tests. +package testutil + +import ( + "bufio" + "context" + "debug/elf" + "encoding/base32" + "encoding/json" + "flag" + "fmt" + "io" + "io/ioutil" + "log" + "math" + "math/rand" + "net/http" + "os" + "os/exec" + "os/signal" + "path" + "path/filepath" + "strconv" + "strings" + "sync/atomic" + "syscall" + "testing" + "time" + + "github.com/cenkalti/backoff" + specs "github.com/opencontainers/runtime-spec/specs-go" + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/runsc/boot" + "gvisor.dev/gvisor/runsc/specutils" +) + +var ( + checkpoint = flag.Bool("checkpoint", true, "control checkpoint/restore support") +) + +// IsCheckpointSupported returns the relevant command line flag. +func IsCheckpointSupported() bool { + return *checkpoint +} + +// nameToActual is used by ImageByName (for now). +var nameToActual = map[string]string{ + "basic/alpine": "alpine", + "basic/busybox": "busybox:1.31.1", + "basic/httpd": "httpd", + "basic/mysql": "mysql", + "basic/nginx": "nginx", + "basic/python": "gcr.io/gvisor-presubmit/python-hello", + "basic/resolv": "k8s.gcr.io/busybox", + "basic/ruby": "ruby", + "basic/tomcat": "tomcat:8.0", + "basic/ubuntu": "ubuntu:trusty", + "iptables": "gcr.io/gvisor-presubmit/iptables-test", + "packetdrill": "gcr.io/gvisor-presubmit/packetdrill", + "packetimpact": "gcr.io/gvisor-presubmit/packetimpact", + "runtimes/go1.12": "gcr.io/gvisor-presubmit/go1.12", + "runtimes/java11": "gcr.io/gvisor-presubmit/java11", + "runtimes/nodejs12.4.0": "gcr.io/gvisor-presubmit/nodejs12.4.0", + "runtimes/php7.3.6": "gcr.io/gvisor-presubmit/php7.3.6", + "runtimes/python3.7.3": "gcr.io/gvisor-presubmit/python3.7.3", +} + +// ImageByName mangles the image name used locally. +// +// For now, this is implemented as a static lookup table. In a subsequent +// change, this will be used to reference a locally-generated image. +func ImageByName(name string) string { + actual, ok := nameToActual[name] + if !ok { + panic(fmt.Sprintf("unknown image: %v", name)) + } + // A terrible hack, for now execute a manual pull. + if out, err := exec.Command("docker", "pull", actual).CombinedOutput(); err != nil { + panic(fmt.Sprintf("error pulling image %q -> %q: %v, out: %s", name, actual, err, string(out))) + } + return actual +} + +// ConfigureExePath configures the executable for runsc in the test environment. +func ConfigureExePath() error { + path, err := FindFile("runsc/runsc") + if err != nil { + return err + } + specutils.ExePath = path + return nil +} + +// TmpDir returns the absolute path to a writable directory that can be used as +// scratch by the test. +func TmpDir() string { + dir := os.Getenv("TEST_TMPDIR") + if dir == "" { + dir = "/tmp" + } + return dir +} + +// Logger is a simple logging wrapper. +// +// This is designed to be implemented by *testing.T. +type Logger interface { + Name() string + Logf(fmt string, args ...interface{}) +} + +// DefaultLogger logs using the log package. +type DefaultLogger string + +// Name implements Logger.Name. +func (d DefaultLogger) Name() string { + return string(d) +} + +// Logf implements Logger.Logf. +func (d DefaultLogger) Logf(fmt string, args ...interface{}) { + log.Printf(fmt, args...) +} + +// Cmd is a simple wrapper. +type Cmd struct { + logger Logger + *exec.Cmd +} + +// CombinedOutput returns the output and logs. +func (c *Cmd) CombinedOutput() ([]byte, error) { + out, err := c.Cmd.CombinedOutput() + if len(out) > 0 { + c.logger.Logf("output: %s", string(out)) + } + if err != nil { + c.logger.Logf("error: %v", err) + } + return out, err +} + +// Command is a simple wrapper around exec.Command, that logs. +func Command(logger Logger, args ...string) *Cmd { + logger.Logf("command: %s", strings.Join(args, " ")) + return &Cmd{ + logger: logger, + Cmd: exec.Command(args[0], args[1:]...), + } +} + +// TestConfig returns the default configuration to use in tests. Note that +// 'RootDir' must be set by caller if required. +func TestConfig(t *testing.T) *boot.Config { + logDir := os.TempDir() + if dir, ok := os.LookupEnv("TEST_UNDECLARED_OUTPUTS_DIR"); ok { + logDir = dir + "/" + } + return &boot.Config{ + Debug: true, + DebugLog: path.Join(logDir, "runsc.log."+t.Name()+".%TIMESTAMP%.%COMMAND%"), + LogFormat: "text", + DebugLogFormat: "text", + LogPackets: true, + Network: boot.NetworkNone, + Strace: true, + Platform: "ptrace", + FileAccess: boot.FileAccessExclusive, + NumNetworkChannels: 1, + + TestOnlyAllowRunAsCurrentUserWithoutChroot: true, + } +} + +// NewSpecWithArgs creates a simple spec with the given args suitable for use +// in tests. +func NewSpecWithArgs(args ...string) *specs.Spec { + return &specs.Spec{ + // The host filesystem root is the container root. + Root: &specs.Root{ + Path: "/", + Readonly: true, + }, + Process: &specs.Process{ + Args: args, + Env: []string{ + "PATH=" + os.Getenv("PATH"), + }, + Capabilities: specutils.AllCapabilities(), + }, + Mounts: []specs.Mount{ + // Hide the host /etc to avoid any side-effects. + // For example, bash reads /etc/passwd and if it is + // very big, tests can fail by timeout. + { + Type: "tmpfs", + Destination: "/etc", + }, + // Root is readonly, but many tests want to write to tmpdir. + // This creates a writable mount inside the root. Also, when tmpdir points + // to "/tmp", it makes the the actual /tmp to be mounted and not a tmpfs + // inside the sentry. + { + Type: "bind", + Destination: TmpDir(), + Source: TmpDir(), + }, + }, + Hostname: "runsc-test-hostname", + } +} + +// SetupRootDir creates a root directory for containers. +func SetupRootDir() (string, func(), error) { + rootDir, err := ioutil.TempDir(TmpDir(), "containers") + if err != nil { + return "", nil, fmt.Errorf("error creating root dir: %v", err) + } + return rootDir, func() { os.RemoveAll(rootDir) }, nil +} + +// SetupContainer creates a bundle and root dir for the container, generates a +// test config, and writes the spec to config.json in the bundle dir. +func SetupContainer(spec *specs.Spec, conf *boot.Config) (rootDir, bundleDir string, cleanup func(), err error) { + rootDir, rootCleanup, err := SetupRootDir() + if err != nil { + return "", "", nil, err + } + conf.RootDir = rootDir + bundleDir, bundleCleanup, err := SetupBundleDir(spec) + if err != nil { + rootCleanup() + return "", "", nil, err + } + return rootDir, bundleDir, func() { + bundleCleanup() + rootCleanup() + }, err +} + +// SetupBundleDir creates a bundle dir and writes the spec to config.json. +func SetupBundleDir(spec *specs.Spec) (string, func(), error) { + bundleDir, err := ioutil.TempDir(TmpDir(), "bundle") + if err != nil { + return "", nil, fmt.Errorf("error creating bundle dir: %v", err) + } + cleanup := func() { os.RemoveAll(bundleDir) } + if err := writeSpec(bundleDir, spec); err != nil { + cleanup() + return "", nil, fmt.Errorf("error writing spec: %v", err) + } + return bundleDir, cleanup, nil +} + +// writeSpec writes the spec to disk in the given directory. +func writeSpec(dir string, spec *specs.Spec) error { + b, err := json.Marshal(spec) + if err != nil { + return err + } + return ioutil.WriteFile(filepath.Join(dir, "config.json"), b, 0755) +} + +// RandomID returns 20 random bytes following the given prefix. +func RandomID(prefix string) string { + // Read 20 random bytes. + b := make([]byte, 20) + // "[Read] always returns len(p) and a nil error." --godoc + if _, err := rand.Read(b); err != nil { + panic("rand.Read failed: " + err.Error()) + } + return fmt.Sprintf("%s-%s", prefix, base32.StdEncoding.EncodeToString(b)) +} + +// RandomContainerID generates a random container id for each test. +// +// The container id is used to create an abstract unix domain socket, which +// must be unique. While the container forbids creating two containers with the +// same name, sometimes between test runs the socket does not get cleaned up +// quickly enough, causing container creation to fail. +func RandomContainerID() string { + return RandomID("test-container-") +} + +// Copy copies file from src to dst. +func Copy(src, dst string) error { + in, err := os.Open(src) + if err != nil { + return err + } + defer in.Close() + + st, err := in.Stat() + if err != nil { + return err + } + + out, err := os.OpenFile(dst, os.O_CREATE|os.O_TRUNC|os.O_WRONLY, st.Mode().Perm()) + if err != nil { + return err + } + defer out.Close() + + // Mirror the local user's permissions across all users. This is + // because as we inject things into the container, the UID/GID will + // change. Also, the build system may generate artifacts with different + // modes. At the top-level (volume mapping) we have a big read-only + // knob that can be applied to prevent modifications. + // + // Note that this must be done via a separate Chmod call, otherwise the + // current process's umask will get in the way. + var mode os.FileMode + if st.Mode()&0100 != 0 { + mode |= 0111 + } + if st.Mode()&0200 != 0 { + mode |= 0222 + } + if st.Mode()&0400 != 0 { + mode |= 0444 + } + if err := os.Chmod(dst, mode); err != nil { + return err + } + + _, err = io.Copy(out, in) + return err +} + +// Poll is a shorthand function to poll for something with given timeout. +func Poll(cb func() error, timeout time.Duration) error { + ctx, cancel := context.WithTimeout(context.Background(), timeout) + defer cancel() + b := backoff.WithContext(backoff.NewConstantBackOff(100*time.Millisecond), ctx) + return backoff.Retry(cb, b) +} + +// WaitForHTTP tries GET requests on a port until the call succeeds or timeout. +func WaitForHTTP(port int, timeout time.Duration) error { + cb := func() error { + c := &http.Client{ + // Calculate timeout to be able to do minimum 5 attempts. + Timeout: timeout / 5, + } + url := fmt.Sprintf("http://localhost:%d/", port) + resp, err := c.Get(url) + if err != nil { + log.Printf("Waiting %s: %v", url, err) + return err + } + resp.Body.Close() + return nil + } + return Poll(cb, timeout) +} + +// Reaper reaps child processes. +type Reaper struct { + // mu protects ch, which will be nil if the reaper is not running. + mu sync.Mutex + ch chan os.Signal +} + +// Start starts reaping child processes. +func (r *Reaper) Start() { + r.mu.Lock() + defer r.mu.Unlock() + + if r.ch != nil { + panic("reaper.Start called on a running reaper") + } + + r.ch = make(chan os.Signal, 1) + signal.Notify(r.ch, syscall.SIGCHLD) + + go func() { + for { + r.mu.Lock() + ch := r.ch + r.mu.Unlock() + if ch == nil { + return + } + + _, ok := <-ch + if !ok { + // Channel closed. + return + } + for { + cpid, _ := syscall.Wait4(-1, nil, syscall.WNOHANG, nil) + if cpid < 1 { + break + } + } + } + }() +} + +// Stop stops reaping child processes. +func (r *Reaper) Stop() { + r.mu.Lock() + defer r.mu.Unlock() + + if r.ch == nil { + panic("reaper.Stop called on a stopped reaper") + } + + signal.Stop(r.ch) + close(r.ch) + r.ch = nil +} + +// StartReaper is a helper that starts a new Reaper and returns a function to +// stop it. +func StartReaper() func() { + r := &Reaper{} + r.Start() + return r.Stop +} + +// WaitUntilRead reads from the given reader until the wanted string is found +// or until timeout. +func WaitUntilRead(r io.Reader, want string, split bufio.SplitFunc, timeout time.Duration) error { + sc := bufio.NewScanner(r) + if split != nil { + sc.Split(split) + } + // done must be accessed atomically. A value greater than 0 indicates + // that the read loop can exit. + var done uint32 + doneCh := make(chan struct{}) + go func() { + for sc.Scan() { + t := sc.Text() + if strings.Contains(t, want) { + atomic.StoreUint32(&done, 1) + close(doneCh) + break + } + if atomic.LoadUint32(&done) > 0 { + break + } + } + }() + select { + case <-time.After(timeout): + atomic.StoreUint32(&done, 1) + return fmt.Errorf("timeout waiting to read %q", want) + case <-doneCh: + return nil + } +} + +// KillCommand kills the process running cmd unless it hasn't been started. It +// returns an error if it cannot kill the process unless the reason is that the +// process has already exited. +// +// KillCommand will also reap the process. +func KillCommand(cmd *exec.Cmd) error { + if cmd.Process == nil { + return nil + } + if err := cmd.Process.Kill(); err != nil { + if !strings.Contains(err.Error(), "process already finished") { + return fmt.Errorf("failed to kill process %v: %v", cmd, err) + } + } + return cmd.Wait() +} + +// WriteTmpFile writes text to a temporary file, closes the file, and returns +// the name of the file. A cleanup function is also returned. +func WriteTmpFile(pattern, text string) (string, func(), error) { + file, err := ioutil.TempFile(TmpDir(), pattern) + if err != nil { + return "", nil, err + } + defer file.Close() + if _, err := file.Write([]byte(text)); err != nil { + return "", nil, err + } + return file.Name(), func() { os.RemoveAll(file.Name()) }, nil +} + +// IsStatic returns true iff the given file is a static binary. +func IsStatic(filename string) (bool, error) { + f, err := elf.Open(filename) + if err != nil { + return false, err + } + for _, prog := range f.Progs { + if prog.Type == elf.PT_INTERP { + return false, nil // Has interpreter. + } + } + return true, nil +} + +// TestIndicesForShard returns indices for this test shard based on the +// TEST_SHARD_INDEX and TEST_TOTAL_SHARDS environment vars. +// +// If either of the env vars are not present, then the function will return all +// tests. If there are more shards than there are tests, then the returned list +// may be empty. +func TestIndicesForShard(numTests int) ([]int, error) { + var ( + shardIndex = 0 + shardTotal = 1 + ) + + indexStr, totalStr := os.Getenv("TEST_SHARD_INDEX"), os.Getenv("TEST_TOTAL_SHARDS") + if indexStr != "" && totalStr != "" { + // Parse index and total to ints. + var err error + shardIndex, err = strconv.Atoi(indexStr) + if err != nil { + return nil, fmt.Errorf("invalid TEST_SHARD_INDEX %q: %v", indexStr, err) + } + shardTotal, err = strconv.Atoi(totalStr) + if err != nil { + return nil, fmt.Errorf("invalid TEST_TOTAL_SHARDS %q: %v", totalStr, err) + } + } + + // Calculate! + var indices []int + numBlocks := int(math.Ceil(float64(numTests) / float64(shardTotal))) + for i := 0; i < numBlocks; i++ { + pick := i*shardTotal + shardIndex + if pick < numTests { + indices = append(indices, pick) + } + } + return indices, nil +} diff --git a/pkg/test/testutil/testutil_runfiles.go b/pkg/test/testutil/testutil_runfiles.go new file mode 100644 index 000000000..ece9ea9a1 --- /dev/null +++ b/pkg/test/testutil/testutil_runfiles.go @@ -0,0 +1,75 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package testutil + +import ( + "fmt" + "os" + "path/filepath" +) + +// FindFile searchs for a file inside the test run environment. It returns the +// full path to the file. It fails if none or more than one file is found. +func FindFile(path string) (string, error) { + wd, err := os.Getwd() + if err != nil { + return "", err + } + + // The test root is demarcated by a path element called "__main__". Search for + // it backwards from the working directory. + root := wd + for { + dir, name := filepath.Split(root) + if name == "__main__" { + break + } + if len(dir) == 0 { + return "", fmt.Errorf("directory __main__ not found in %q", wd) + } + // Remove ending slash to loop around. + root = dir[:len(dir)-1] + } + + // Annoyingly, bazel adds the build type to the directory path for go + // binaries, but not for c++ binaries. We use two different patterns to + // to find our file. + patterns := []string{ + // Try the obvious path first. + filepath.Join(root, path), + // If it was a go binary, use a wildcard to match the build + // type. The pattern is: /test-path/__main__/directories/*/file. + filepath.Join(root, filepath.Dir(path), "*", filepath.Base(path)), + } + + for _, p := range patterns { + matches, err := filepath.Glob(p) + if err != nil { + // "The only possible returned error is ErrBadPattern, + // when pattern is malformed." -godoc + return "", fmt.Errorf("error globbing %q: %v", p, err) + } + switch len(matches) { + case 0: + // Try the next pattern. + case 1: + // We found it. + return matches[0], nil + default: + return "", fmt.Errorf("more than one match found for %q: %s", path, matches) + } + } + return "", fmt.Errorf("file %q not found", path) +} diff --git a/runsc/boot/BUILD b/runsc/boot/BUILD index 72c2fe381..69dcc74f2 100644 --- a/runsc/boot/BUILD +++ b/runsc/boot/BUILD @@ -23,6 +23,7 @@ go_library( "vfs.go", ], visibility = [ + "//pkg/test:__subpackages__", "//runsc:__subpackages__", "//test:__subpackages__", ], diff --git a/runsc/cmd/BUILD b/runsc/cmd/BUILD index 4900fbe16..af3538ef0 100644 --- a/runsc/cmd/BUILD +++ b/runsc/cmd/BUILD @@ -82,11 +82,11 @@ go_test( "//pkg/log", "//pkg/sentry/control", "//pkg/sentry/kernel/auth", + "//pkg/test/testutil", "//pkg/urpc", "//runsc/boot", "//runsc/container", "//runsc/specutils", - "//runsc/testutil", "@com_github_google_go-cmp//cmp:go_default_library", "@com_github_google_go-cmp//cmp/cmpopts:go_default_library", "@com_github_opencontainers_runtime-spec//specs-go:go_default_library", diff --git a/runsc/cmd/capability_test.go b/runsc/cmd/capability_test.go index 9360d7442..a84067112 100644 --- a/runsc/cmd/capability_test.go +++ b/runsc/cmd/capability_test.go @@ -23,10 +23,10 @@ import ( specs "github.com/opencontainers/runtime-spec/specs-go" "github.com/syndtr/gocapability/capability" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/test/testutil" "gvisor.dev/gvisor/runsc/boot" "gvisor.dev/gvisor/runsc/container" "gvisor.dev/gvisor/runsc/specutils" - "gvisor.dev/gvisor/runsc/testutil" ) func init() { @@ -90,16 +90,15 @@ func TestCapabilities(t *testing.T) { // Use --network=host to make sandbox use spec's capabilities. conf.Network = boot.NetworkHost - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() // Create and start the container. args := container.Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, } diff --git a/runsc/container/BUILD b/runsc/container/BUILD index 0aaeea3a8..331b8e866 100644 --- a/runsc/container/BUILD +++ b/runsc/container/BUILD @@ -33,13 +33,15 @@ go_test( size = "large", srcs = [ "console_test.go", + "container_norace_test.go", + "container_race_test.go", "container_test.go", "multi_container_test.go", "shared_volume_test.go", ], data = [ "//runsc", - "//runsc/container/test_app", + "//test/cmd/test_app", ], library = ":container", shard_count = 5, @@ -54,12 +56,12 @@ go_test( "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sync", + "//pkg/test/testutil", "//pkg/unet", "//pkg/urpc", "//runsc/boot", "//runsc/boot/platforms", "//runsc/specutils", - "//runsc/testutil", "@com_github_cenkalti_backoff//:go_default_library", "@com_github_kr_pty//:go_default_library", "@com_github_opencontainers_runtime-spec//specs-go:go_default_library", diff --git a/runsc/container/console_test.go b/runsc/container/console_test.go index af245b6d8..294dca5e7 100644 --- a/runsc/container/console_test.go +++ b/runsc/container/console_test.go @@ -29,9 +29,9 @@ import ( "gvisor.dev/gvisor/pkg/sentry/control" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/test/testutil" "gvisor.dev/gvisor/pkg/unet" "gvisor.dev/gvisor/pkg/urpc" - "gvisor.dev/gvisor/runsc/testutil" ) // socketPath creates a path inside bundleDir and ensures that the returned @@ -58,25 +58,26 @@ func socketPath(bundleDir string) (string, error) { } // createConsoleSocket creates a socket at the given path that will receive a -// console fd from the sandbox. If no error occurs, it returns the server -// socket and a cleanup function. -func createConsoleSocket(path string) (*unet.ServerSocket, func() error, error) { +// console fd from the sandbox. If an error occurs, t.Fatalf will be called. +// The function returning should be deferred as cleanup. +func createConsoleSocket(t *testing.T, path string) (*unet.ServerSocket, func()) { + t.Helper() srv, err := unet.BindAndListen(path, false) if err != nil { - return nil, nil, fmt.Errorf("error binding and listening to socket %q: %v", path, err) + t.Fatalf("error binding and listening to socket %q: %v", path, err) } - cleanup := func() error { + cleanup := func() { + // Log errors; nothing can be done. if err := srv.Close(); err != nil { - return fmt.Errorf("error closing socket %q: %v", path, err) + t.Logf("error closing socket %q: %v", path, err) } if err := os.Remove(path); err != nil { - return fmt.Errorf("error removing socket %q: %v", path, err) + t.Logf("error removing socket %q: %v", path, err) } - return nil } - return srv, cleanup, nil + return srv, cleanup } // receiveConsolePTY accepts a connection on the server socket and reads fds. @@ -118,45 +119,42 @@ func receiveConsolePTY(srv *unet.ServerSocket) (*os.File, error) { // Test that an pty FD is sent over the console socket if one is provided. func TestConsoleSocket(t *testing.T) { - for _, conf := range configs(t, all...) { - t.Logf("Running test with conf: %+v", conf) - spec := testutil.NewSpecWithArgs("true") - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) - if err != nil { - t.Fatalf("error setting up container: %v", err) - } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + for name, conf := range configs(t, all...) { + t.Run(name, func(t *testing.T) { + spec := testutil.NewSpecWithArgs("true") + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() - sock, err := socketPath(bundleDir) - if err != nil { - t.Fatalf("error getting socket path: %v", err) - } - srv, cleanup, err := createConsoleSocket(sock) - if err != nil { - t.Fatalf("error creating socket at %q: %v", sock, err) - } - defer cleanup() - - // Create the container and pass the socket name. - args := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - ConsoleSocket: sock, - } - c, err := New(conf, args) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - defer c.Destroy() + sock, err := socketPath(bundleDir) + if err != nil { + t.Fatalf("error getting socket path: %v", err) + } + srv, cleanup := createConsoleSocket(t, sock) + defer cleanup() + + // Create the container and pass the socket name. + args := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + ConsoleSocket: sock, + } + c, err := New(conf, args) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer c.Destroy() - // Make sure we get a console PTY. - ptyMaster, err := receiveConsolePTY(srv) - if err != nil { - t.Fatalf("error receiving console FD: %v", err) - } - ptyMaster.Close() + // Make sure we get a console PTY. + ptyMaster, err := receiveConsolePTY(srv) + if err != nil { + t.Fatalf("error receiving console FD: %v", err) + } + ptyMaster.Close() + }) } } @@ -165,16 +163,15 @@ func TestJobControlSignalExec(t *testing.T) { spec := testutil.NewSpecWithArgs("/bin/sleep", "10000") conf := testutil.TestConfig(t) - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() // Create and start the container. args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, } @@ -292,26 +289,22 @@ func TestJobControlSignalRootContainer(t *testing.T) { spec := testutil.NewSpecWithArgs("/bin/bash", "--noprofile", "--norc") spec.Process.Terminal = true - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() sock, err := socketPath(bundleDir) if err != nil { t.Fatalf("error getting socket path: %v", err) } - srv, cleanup, err := createConsoleSocket(sock) - if err != nil { - t.Fatalf("error creating socket at %q: %v", sock, err) - } + srv, cleanup := createConsoleSocket(t, sock) defer cleanup() // Create the container and pass the socket name. args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, ConsoleSocket: sock, @@ -368,7 +361,7 @@ func TestJobControlSignalRootContainer(t *testing.T) { {PID: 1, Cmd: "bash", Threads: []kernel.ThreadID{1}}, } if err := waitForProcessList(c, expectedPL); err != nil { - t.Fatal(err) + t.Fatalf("error waiting for processes: %v", err) } // Execute sleep via the terminal. @@ -377,7 +370,7 @@ func TestJobControlSignalRootContainer(t *testing.T) { // Wait for sleep to start. expectedPL = append(expectedPL, &control.Process{PID: 2, PPID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{2}}) if err := waitForProcessList(c, expectedPL); err != nil { - t.Fatal(err) + t.Fatalf("error waiting for processes: %v", err) } // Reset the pty buffer, so there is less output for us to scan later. diff --git a/runsc/container/container.go b/runsc/container/container.go index 7233659b1..117ea7d7b 100644 --- a/runsc/container/container.go +++ b/runsc/container/container.go @@ -274,7 +274,7 @@ func New(conf *boot.Config, args Args) (*Container, error) { } if err := os.MkdirAll(conf.RootDir, 0711); err != nil { - return nil, fmt.Errorf("creating container root directory: %v", err) + return nil, fmt.Errorf("creating container root directory %q: %v", conf.RootDir, err) } c := &Container{ diff --git a/runsc/container/container_norace_test.go b/runsc/container/container_norace_test.go new file mode 100644 index 000000000..838c1e20a --- /dev/null +++ b/runsc/container/container_norace_test.go @@ -0,0 +1,20 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build !race + +package container + +// Allow both kvm and ptrace for non-race builds. +var platformOptions = []configOption{ptrace, kvm} diff --git a/runsc/container/container_race_test.go b/runsc/container/container_race_test.go new file mode 100644 index 000000000..9fb4c4fc0 --- /dev/null +++ b/runsc/container/container_race_test.go @@ -0,0 +1,20 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build race + +package container + +// Only enabled ptrace with race builds. +var platformOptions = []configOption{ptrace} diff --git a/runsc/container/container_test.go b/runsc/container/container_test.go index 5db6d64aa..3ff89f38c 100644 --- a/runsc/container/container_test.go +++ b/runsc/container/container_test.go @@ -39,10 +39,10 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/test/testutil" "gvisor.dev/gvisor/runsc/boot" "gvisor.dev/gvisor/runsc/boot/platforms" "gvisor.dev/gvisor/runsc/specutils" - "gvisor.dev/gvisor/runsc/testutil" ) // waitForProcessList waits for the given process list to show up in the container. @@ -215,16 +215,15 @@ func readOutputNum(file string, position int) (int, error) { // run starts the sandbox and waits for it to exit, checking that the // application succeeded. func run(spec *specs.Spec, conf *boot.Config) error { - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { return fmt.Errorf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() // Create, start and wait for the container. args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, Attached: true, @@ -243,35 +242,41 @@ type configOption int const ( overlay configOption = iota + ptrace kvm nonExclusiveFS ) -var noOverlay = []configOption{kvm, nonExclusiveFS} -var all = append(noOverlay, overlay) +var ( + noOverlay = append(platformOptions, nonExclusiveFS) + all = append(noOverlay, overlay) +) // configs generates different configurations to run tests. -func configs(t *testing.T, opts ...configOption) []*boot.Config { +func configs(t *testing.T, opts ...configOption) map[string]*boot.Config { // Always load the default config. - cs := []*boot.Config{testutil.TestConfig(t)} - + cs := make(map[string]*boot.Config) for _, o := range opts { - c := testutil.TestConfig(t) switch o { case overlay: + c := testutil.TestConfig(t) c.Overlay = true + cs["overlay"] = c + case ptrace: + c := testutil.TestConfig(t) + c.Platform = platforms.Ptrace + cs["ptrace"] = c case kvm: - // TODO(b/112165693): KVM tests are flaky. Disable until fixed. - continue - + c := testutil.TestConfig(t) c.Platform = platforms.KVM + cs["kvm"] = c case nonExclusiveFS: + c := testutil.TestConfig(t) c.FileAccess = boot.FileAccessShared + cs["non-exclusive"] = c default: panic(fmt.Sprintf("unknown config option %v", o)) - } - cs = append(cs, c) } return cs } @@ -285,133 +290,133 @@ func TestLifecycle(t *testing.T) { childReaper.Start() defer childReaper.Stop() - for _, conf := range configs(t, all...) { - t.Logf("Running test with conf: %+v", conf) - // The container will just sleep for a long time. We will kill it before - // it finishes sleeping. - spec := testutil.NewSpecWithArgs("sleep", "100") - - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) - if err != nil { - t.Fatalf("error setting up container: %v", err) - } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) - - // expectedPL lists the expected process state of the container. - expectedPL := []*control.Process{ - { - UID: 0, - PID: 1, - PPID: 0, - C: 0, - Cmd: "sleep", - Threads: []kernel.ThreadID{1}, - }, - } - // Create the container. - args := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - } - c, err := New(conf, args) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - defer c.Destroy() + for name, conf := range configs(t, all...) { + t.Run(name, func(t *testing.T) { + // The container will just sleep for a long time. We will kill it before + // it finishes sleeping. + spec := testutil.NewSpecWithArgs("sleep", "100") - // Load the container from disk and check the status. - c, err = Load(rootDir, args.ID) - if err != nil { - t.Fatalf("error loading container: %v", err) - } - if got, want := c.Status, Created; got != want { - t.Errorf("container status got %v, want %v", got, want) - } + rootDir, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() - // List should return the container id. - ids, err := List(rootDir) - if err != nil { - t.Fatalf("error listing containers: %v", err) - } - if got, want := ids, []string{args.ID}; !reflect.DeepEqual(got, want) { - t.Errorf("container list got %v, want %v", got, want) - } + // expectedPL lists the expected process state of the container. + expectedPL := []*control.Process{ + { + UID: 0, + PID: 1, + PPID: 0, + C: 0, + Cmd: "sleep", + Threads: []kernel.ThreadID{1}, + }, + } + // Create the container. + args := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + c, err := New(conf, args) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer c.Destroy() - // Start the container. - if err := c.Start(conf); err != nil { - t.Fatalf("error starting container: %v", err) - } + // Load the container from disk and check the status. + c, err = Load(rootDir, args.ID) + if err != nil { + t.Fatalf("error loading container: %v", err) + } + if got, want := c.Status, Created; got != want { + t.Errorf("container status got %v, want %v", got, want) + } - // Load the container from disk and check the status. - c, err = Load(rootDir, args.ID) - if err != nil { - t.Fatalf("error loading container: %v", err) - } - if got, want := c.Status, Running; got != want { - t.Errorf("container status got %v, want %v", got, want) - } + // List should return the container id. + ids, err := List(rootDir) + if err != nil { + t.Fatalf("error listing containers: %v", err) + } + if got, want := ids, []string{args.ID}; !reflect.DeepEqual(got, want) { + t.Errorf("container list got %v, want %v", got, want) + } - // Verify that "sleep 100" is running. - if err := waitForProcessList(c, expectedPL); err != nil { - t.Error(err) - } + // Start the container. + if err := c.Start(conf); err != nil { + t.Fatalf("error starting container: %v", err) + } - // Wait on the container. - var wg sync.WaitGroup - wg.Add(1) - ch := make(chan struct{}) - go func() { - ch <- struct{}{} - ws, err := c.Wait() + // Load the container from disk and check the status. + c, err = Load(rootDir, args.ID) if err != nil { - t.Fatalf("error waiting on container: %v", err) + t.Fatalf("error loading container: %v", err) } - if got, want := ws.Signal(), syscall.SIGTERM; got != want { - t.Fatalf("got signal %v, want %v", got, want) + if got, want := c.Status, Running; got != want { + t.Errorf("container status got %v, want %v", got, want) } - wg.Done() - }() - // Wait a bit to ensure that we've started waiting on the - // container before we signal. - <-ch - time.Sleep(100 * time.Millisecond) - // Send the container a SIGTERM which will cause it to stop. - if err := c.SignalContainer(syscall.SIGTERM, false); err != nil { - t.Fatalf("error sending signal %v to container: %v", syscall.SIGTERM, err) - } - // Wait for it to die. - wg.Wait() + // Verify that "sleep 100" is running. + if err := waitForProcessList(c, expectedPL); err != nil { + t.Error(err) + } - // Load the container from disk and check the status. - c, err = Load(rootDir, args.ID) - if err != nil { - t.Fatalf("error loading container: %v", err) - } - if got, want := c.Status, Stopped; got != want { - t.Errorf("container status got %v, want %v", got, want) - } + // Wait on the container. + ch := make(chan error) + go func() { + ws, err := c.Wait() + if err != nil { + ch <- err + } + if got, want := ws.Signal(), syscall.SIGTERM; got != want { + ch <- fmt.Errorf("got signal %v, want %v", got, want) + } + ch <- nil + }() - // Destroy the container. - if err := c.Destroy(); err != nil { - t.Fatalf("error destroying container: %v", err) - } + // Wait a bit to ensure that we've started waiting on + // the container before we signal. + time.Sleep(time.Second) - // List should not return the container id. - ids, err = List(rootDir) - if err != nil { - t.Fatalf("error listing containers: %v", err) - } - if len(ids) != 0 { - t.Errorf("expected container list to be empty, but got %v", ids) - } + // Send the container a SIGTERM which will cause it to stop. + if err := c.SignalContainer(syscall.SIGTERM, false); err != nil { + t.Fatalf("error sending signal %v to container: %v", syscall.SIGTERM, err) + } - // Loading the container by id should fail. - if _, err = Load(rootDir, args.ID); err == nil { - t.Errorf("expected loading destroyed container to fail, but it did not") - } + // Wait for it to die. + if err := <-ch; err != nil { + t.Fatalf("error waiting for container: %v", err) + } + + // Load the container from disk and check the status. + c, err = Load(rootDir, args.ID) + if err != nil { + t.Fatalf("error loading container: %v", err) + } + if got, want := c.Status, Stopped; got != want { + t.Errorf("container status got %v, want %v", got, want) + } + + // Destroy the container. + if err := c.Destroy(); err != nil { + t.Fatalf("error destroying container: %v", err) + } + + // List should not return the container id. + ids, err = List(rootDir) + if err != nil { + t.Fatalf("error listing containers: %v", err) + } + if len(ids) != 0 { + t.Errorf("expected container list to be empty, but got %v", ids) + } + + // Loading the container by id should fail. + if _, err = Load(rootDir, args.ID); err == nil { + t.Errorf("expected loading destroyed container to fail, but it did not") + } + }) } } @@ -420,12 +425,14 @@ func TestExePath(t *testing.T) { // Create two directories that will be prepended to PATH. firstPath, err := ioutil.TempDir(testutil.TmpDir(), "first") if err != nil { - t.Fatal(err) + t.Fatalf("error creating temporary directory: %v", err) } + defer os.RemoveAll(firstPath) secondPath, err := ioutil.TempDir(testutil.TmpDir(), "second") if err != nil { - t.Fatal(err) + t.Fatalf("error creating temporary directory: %v", err) } + defer os.RemoveAll(secondPath) // Create two minimal executables in the second path, two of which // will be masked by files in first path. @@ -433,11 +440,11 @@ func TestExePath(t *testing.T) { path := filepath.Join(secondPath, p) f, err := os.OpenFile(path, os.O_CREATE|os.O_EXCL|os.O_RDWR, 0777) if err != nil { - t.Fatal(err) + t.Fatalf("error opening path: %v", err) } defer f.Close() if _, err := io.WriteString(f, "#!/bin/true\n"); err != nil { - t.Fatal(err) + t.Fatalf("error writing contents: %v", err) } } @@ -446,7 +453,7 @@ func TestExePath(t *testing.T) { nonExecutable := filepath.Join(firstPath, "masked1") f2, err := os.OpenFile(nonExecutable, os.O_CREATE|os.O_EXCL, 0666) if err != nil { - t.Fatal(err) + t.Fatalf("error opening file: %v", err) } f2.Close() @@ -454,68 +461,69 @@ func TestExePath(t *testing.T) { // executable in the second. nonRegular := filepath.Join(firstPath, "masked2") if err := os.Mkdir(nonRegular, 0777); err != nil { - t.Fatal(err) - } - - for _, conf := range configs(t, overlay) { - t.Logf("Running test with conf: %+v", conf) - for _, test := range []struct { - path string - success bool - }{ - {path: "true", success: true}, - {path: "bin/true", success: true}, - {path: "/bin/true", success: true}, - {path: "thisfiledoesntexit", success: false}, - {path: "bin/thisfiledoesntexit", success: false}, - {path: "/bin/thisfiledoesntexit", success: false}, - - {path: "unmasked", success: true}, - {path: filepath.Join(firstPath, "unmasked"), success: false}, - {path: filepath.Join(secondPath, "unmasked"), success: true}, - - {path: "masked1", success: true}, - {path: filepath.Join(firstPath, "masked1"), success: false}, - {path: filepath.Join(secondPath, "masked1"), success: true}, - - {path: "masked2", success: true}, - {path: filepath.Join(firstPath, "masked2"), success: false}, - {path: filepath.Join(secondPath, "masked2"), success: true}, - } { - spec := testutil.NewSpecWithArgs(test.path) - spec.Process.Env = []string{ - fmt.Sprintf("PATH=%s:%s:%s", firstPath, secondPath, os.Getenv("PATH")), - } - - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) - if err != nil { - t.Fatalf("exec: %s, error setting up container: %v", test.path, err) - } - - args := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - Attached: true, - } - ws, err := Run(conf, args) + t.Fatalf("error making directory: %v", err) + } + + for name, conf := range configs(t, overlay) { + t.Run(name, func(t *testing.T) { + for _, test := range []struct { + path string + success bool + }{ + {path: "true", success: true}, + {path: "bin/true", success: true}, + {path: "/bin/true", success: true}, + {path: "thisfiledoesntexit", success: false}, + {path: "bin/thisfiledoesntexit", success: false}, + {path: "/bin/thisfiledoesntexit", success: false}, + + {path: "unmasked", success: true}, + {path: filepath.Join(firstPath, "unmasked"), success: false}, + {path: filepath.Join(secondPath, "unmasked"), success: true}, + + {path: "masked1", success: true}, + {path: filepath.Join(firstPath, "masked1"), success: false}, + {path: filepath.Join(secondPath, "masked1"), success: true}, + + {path: "masked2", success: true}, + {path: filepath.Join(firstPath, "masked2"), success: false}, + {path: filepath.Join(secondPath, "masked2"), success: true}, + } { + t.Run(fmt.Sprintf("path=%s,success=%t", test.path, test.success), func(t *testing.T) { + spec := testutil.NewSpecWithArgs(test.path) + spec.Process.Env = []string{ + fmt.Sprintf("PATH=%s:%s:%s", firstPath, secondPath, os.Getenv("PATH")), + } - os.RemoveAll(rootDir) - os.RemoveAll(bundleDir) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("exec: error setting up container: %v", err) + } + defer cleanup() - if test.success { - if err != nil { - t.Errorf("exec: %s, error running container: %v", test.path, err) - } - if ws.ExitStatus() != 0 { - t.Errorf("exec: %s, got exit status %v want %v", test.path, ws.ExitStatus(), 0) - } - } else { - if err == nil { - t.Errorf("exec: %s, got: no error, want: error", test.path) - } + args := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + Attached: true, + } + ws, err := Run(conf, args) + + if test.success { + if err != nil { + t.Errorf("exec: error running container: %v", err) + } + if ws.ExitStatus() != 0 { + t.Errorf("exec: got exit status %v want %v", ws.ExitStatus(), 0) + } + } else { + if err == nil { + t.Errorf("exec: got: no error, want: error") + } + } + }) } - } + }) } } @@ -534,15 +542,14 @@ func doAppExitStatus(t *testing.T, vfs2 bool) { succSpec := testutil.NewSpecWithArgs("true") conf := testutil.TestConfig(t) conf.VFS2 = vfs2 - rootDir, bundleDir, err := testutil.SetupContainer(succSpec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(succSpec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: succSpec, BundleDir: bundleDir, Attached: true, @@ -559,15 +566,14 @@ func doAppExitStatus(t *testing.T, vfs2 bool) { wantStatus := 123 errSpec := testutil.NewSpecWithArgs("bash", "-c", fmt.Sprintf("exit %d", wantStatus)) - rootDir2, bundleDir2, err := testutil.SetupContainer(errSpec, conf) + _, bundleDir2, cleanup2, err := testutil.SetupContainer(errSpec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir2) - defer os.RemoveAll(bundleDir2) + defer cleanup2() args2 := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: errSpec, BundleDir: bundleDir2, Attached: true, @@ -583,166 +589,163 @@ func doAppExitStatus(t *testing.T, vfs2 bool) { // TestExec verifies that a container can exec a new program. func TestExec(t *testing.T) { - for _, conf := range configs(t, overlay) { - t.Logf("Running test with conf: %+v", conf) + for name, conf := range configs(t, overlay) { + t.Run(name, func(t *testing.T) { + const uid = 343 + spec := testutil.NewSpecWithArgs("sleep", "100") - const uid = 343 - spec := testutil.NewSpecWithArgs("sleep", "100") + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) - if err != nil { - t.Fatalf("error setting up container: %v", err) - } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + // Create and start the container. + args := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + cont, err := New(conf, args) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer cont.Destroy() + if err := cont.Start(conf); err != nil { + t.Fatalf("error starting container: %v", err) + } - // Create and start the container. - args := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - } - cont, err := New(conf, args) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - defer cont.Destroy() - if err := cont.Start(conf); err != nil { - t.Fatalf("error starting container: %v", err) - } + // expectedPL lists the expected process state of the container. + expectedPL := []*control.Process{ + { + UID: 0, + PID: 1, + PPID: 0, + C: 0, + Cmd: "sleep", + Threads: []kernel.ThreadID{1}, + }, + { + UID: uid, + PID: 2, + PPID: 0, + C: 0, + Cmd: "sleep", + Threads: []kernel.ThreadID{2}, + }, + } - // expectedPL lists the expected process state of the container. - expectedPL := []*control.Process{ - { - UID: 0, - PID: 1, - PPID: 0, - C: 0, - Cmd: "sleep", - Threads: []kernel.ThreadID{1}, - }, - { - UID: uid, - PID: 2, - PPID: 0, - C: 0, - Cmd: "sleep", - Threads: []kernel.ThreadID{2}, - }, - } + // Verify that "sleep 100" is running. + if err := waitForProcessList(cont, expectedPL[:1]); err != nil { + t.Error(err) + } - // Verify that "sleep 100" is running. - if err := waitForProcessList(cont, expectedPL[:1]); err != nil { - t.Error(err) - } + execArgs := &control.ExecArgs{ + Filename: "/bin/sleep", + Argv: []string{"/bin/sleep", "5"}, + WorkingDirectory: "/", + KUID: uid, + } - execArgs := &control.ExecArgs{ - Filename: "/bin/sleep", - Argv: []string{"/bin/sleep", "5"}, - WorkingDirectory: "/", - KUID: uid, - } + // Verify that "sleep 100" and "sleep 5" are running + // after exec. First, start running exec (whick + // blocks). + ch := make(chan error) + go func() { + exitStatus, err := cont.executeSync(execArgs) + if err != nil { + ch <- err + } else if exitStatus != 0 { + ch <- fmt.Errorf("failed with exit status: %v", exitStatus) + } else { + ch <- nil + } + }() - // Verify that "sleep 100" and "sleep 5" are running after exec. - // First, start running exec (whick blocks). - status := make(chan error, 1) - go func() { - exitStatus, err := cont.executeSync(execArgs) - if err != nil { - log.Debugf("error executing: %v", err) - status <- err - } else if exitStatus != 0 { - log.Debugf("bad status: %d", exitStatus) - status <- fmt.Errorf("failed with exit status: %v", exitStatus) - } else { - status <- nil + if err := waitForProcessList(cont, expectedPL); err != nil { + t.Fatalf("error waiting for processes: %v", err) } - }() - - if err := waitForProcessList(cont, expectedPL); err != nil { - t.Fatal(err) - } - // Ensure that exec finished without error. - select { - case <-time.After(10 * time.Second): - t.Fatalf("container timed out waiting for exec to finish.") - case st := <-status: - if st != nil { - t.Errorf("container failed to exec %v: %v", args, err) + // Ensure that exec finished without error. + select { + case <-time.After(10 * time.Second): + t.Fatalf("container timed out waiting for exec to finish.") + case err := <-ch: + if err != nil { + t.Errorf("container failed to exec %v: %v", args, err) + } } - } + }) } } // TestKillPid verifies that we can signal individual exec'd processes. func TestKillPid(t *testing.T) { - for _, conf := range configs(t, overlay) { - t.Logf("Running test with conf: %+v", conf) - - app, err := testutil.FindFile("runsc/container/test_app/test_app") - if err != nil { - t.Fatal("error finding test_app:", err) - } + for name, conf := range configs(t, overlay) { + t.Run(name, func(t *testing.T) { + app, err := testutil.FindFile("test/cmd/test_app/test_app") + if err != nil { + t.Fatal("error finding test_app:", err) + } - const nProcs = 4 - spec := testutil.NewSpecWithArgs(app, "task-tree", "--depth", strconv.Itoa(nProcs-1), "--width=1", "--pause=true") - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) - if err != nil { - t.Fatalf("error setting up container: %v", err) - } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + const nProcs = 4 + spec := testutil.NewSpecWithArgs(app, "task-tree", "--depth", strconv.Itoa(nProcs-1), "--width=1", "--pause=true") + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() - // Create and start the container. - args := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - } - cont, err := New(conf, args) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - defer cont.Destroy() - if err := cont.Start(conf); err != nil { - t.Fatalf("error starting container: %v", err) - } + // Create and start the container. + args := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + cont, err := New(conf, args) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer cont.Destroy() + if err := cont.Start(conf); err != nil { + t.Fatalf("error starting container: %v", err) + } - // Verify that all processes are running. - if err := waitForProcessCount(cont, nProcs); err != nil { - t.Fatalf("timed out waiting for processes to start: %v", err) - } + // Verify that all processes are running. + if err := waitForProcessCount(cont, nProcs); err != nil { + t.Fatalf("timed out waiting for processes to start: %v", err) + } - // Kill the child process with the largest PID. - procs, err := cont.Processes() - if err != nil { - t.Fatalf("failed to get process list: %v", err) - } - var pid int32 - for _, p := range procs { - if pid < int32(p.PID) { - pid = int32(p.PID) + // Kill the child process with the largest PID. + procs, err := cont.Processes() + if err != nil { + t.Fatalf("failed to get process list: %v", err) + } + var pid int32 + for _, p := range procs { + if pid < int32(p.PID) { + pid = int32(p.PID) + } + } + if err := cont.SignalProcess(syscall.SIGKILL, pid); err != nil { + t.Fatalf("failed to signal process %d: %v", pid, err) } - } - if err := cont.SignalProcess(syscall.SIGKILL, pid); err != nil { - t.Fatalf("failed to signal process %d: %v", pid, err) - } - // Verify that one process is gone. - if err := waitForProcessCount(cont, nProcs-1); err != nil { - t.Fatal(err) - } + // Verify that one process is gone. + if err := waitForProcessCount(cont, nProcs-1); err != nil { + t.Fatalf("error waiting for processes: %v", err) + } - procs, err = cont.Processes() - if err != nil { - t.Fatalf("failed to get process list: %v", err) - } - for _, p := range procs { - if pid == int32(p.PID) { - t.Fatalf("pid %d is still alive, which should be killed", pid) + procs, err = cont.Processes() + if err != nil { + t.Fatalf("failed to get process list: %v", err) } - } + for _, p := range procs { + if pid == int32(p.PID) { + t.Fatalf("pid %d is still alive, which should be killed", pid) + } + } + }) } } @@ -753,160 +756,160 @@ func TestKillPid(t *testing.T) { // be the next consecutive number after the last number from the checkpointed container. func TestCheckpointRestore(t *testing.T) { // Skip overlay because test requires writing to host file. - for _, conf := range configs(t, noOverlay...) { - t.Logf("Running test with conf: %+v", conf) - - dir, err := ioutil.TempDir(testutil.TmpDir(), "checkpoint-test") - if err != nil { - t.Fatalf("ioutil.TempDir failed: %v", err) - } - if err := os.Chmod(dir, 0777); err != nil { - t.Fatalf("error chmoding file: %q, %v", dir, err) - } + for name, conf := range configs(t, noOverlay...) { + t.Run(name, func(t *testing.T) { + dir, err := ioutil.TempDir(testutil.TmpDir(), "checkpoint-test") + if err != nil { + t.Fatalf("ioutil.TempDir failed: %v", err) + } + defer os.RemoveAll(dir) + if err := os.Chmod(dir, 0777); err != nil { + t.Fatalf("error chmoding file: %q, %v", dir, err) + } - outputPath := filepath.Join(dir, "output") - outputFile, err := createWriteableOutputFile(outputPath) - if err != nil { - t.Fatalf("error creating output file: %v", err) - } - defer outputFile.Close() + outputPath := filepath.Join(dir, "output") + outputFile, err := createWriteableOutputFile(outputPath) + if err != nil { + t.Fatalf("error creating output file: %v", err) + } + defer outputFile.Close() - script := fmt.Sprintf("for ((i=0; ;i++)); do echo $i >> %q; sleep 1; done", outputPath) - spec := testutil.NewSpecWithArgs("bash", "-c", script) - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) - if err != nil { - t.Fatalf("error setting up container: %v", err) - } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + script := fmt.Sprintf("for ((i=0; ;i++)); do echo $i >> %q; sleep 1; done", outputPath) + spec := testutil.NewSpecWithArgs("bash", "-c", script) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() - // Create and start the container. - args := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - } - cont, err := New(conf, args) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - defer cont.Destroy() - if err := cont.Start(conf); err != nil { - t.Fatalf("error starting container: %v", err) - } + // Create and start the container. + args := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + cont, err := New(conf, args) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer cont.Destroy() + if err := cont.Start(conf); err != nil { + t.Fatalf("error starting container: %v", err) + } - // Set the image path, which is where the checkpoint image will be saved. - imagePath := filepath.Join(dir, "test-image-file") + // Set the image path, which is where the checkpoint image will be saved. + imagePath := filepath.Join(dir, "test-image-file") - // Create the image file and open for writing. - file, err := os.OpenFile(imagePath, os.O_CREATE|os.O_EXCL|os.O_RDWR, 0644) - if err != nil { - t.Fatalf("error opening new file at imagePath: %v", err) - } - defer file.Close() + // Create the image file and open for writing. + file, err := os.OpenFile(imagePath, os.O_CREATE|os.O_EXCL|os.O_RDWR, 0644) + if err != nil { + t.Fatalf("error opening new file at imagePath: %v", err) + } + defer file.Close() - // Wait until application has ran. - if err := waitForFileNotEmpty(outputFile); err != nil { - t.Fatalf("Failed to wait for output file: %v", err) - } + // Wait until application has ran. + if err := waitForFileNotEmpty(outputFile); err != nil { + t.Fatalf("Failed to wait for output file: %v", err) + } - // Checkpoint running container; save state into new file. - if err := cont.Checkpoint(file); err != nil { - t.Fatalf("error checkpointing container to empty file: %v", err) - } - defer os.RemoveAll(imagePath) + // Checkpoint running container; save state into new file. + if err := cont.Checkpoint(file); err != nil { + t.Fatalf("error checkpointing container to empty file: %v", err) + } + defer os.RemoveAll(imagePath) - lastNum, err := readOutputNum(outputPath, -1) - if err != nil { - t.Fatalf("error with outputFile: %v", err) - } + lastNum, err := readOutputNum(outputPath, -1) + if err != nil { + t.Fatalf("error with outputFile: %v", err) + } - // Delete and recreate file before restoring. - if err := os.Remove(outputPath); err != nil { - t.Fatalf("error removing file") - } - outputFile2, err := createWriteableOutputFile(outputPath) - if err != nil { - t.Fatalf("error creating output file: %v", err) - } - defer outputFile2.Close() + // Delete and recreate file before restoring. + if err := os.Remove(outputPath); err != nil { + t.Fatalf("error removing file") + } + outputFile2, err := createWriteableOutputFile(outputPath) + if err != nil { + t.Fatalf("error creating output file: %v", err) + } + defer outputFile2.Close() - // Restore into a new container. - args2 := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - } - cont2, err := New(conf, args2) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - defer cont2.Destroy() + // Restore into a new container. + args2 := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + cont2, err := New(conf, args2) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer cont2.Destroy() - if err := cont2.Restore(spec, conf, imagePath); err != nil { - t.Fatalf("error restoring container: %v", err) - } + if err := cont2.Restore(spec, conf, imagePath); err != nil { + t.Fatalf("error restoring container: %v", err) + } - // Wait until application has ran. - if err := waitForFileNotEmpty(outputFile2); err != nil { - t.Fatalf("Failed to wait for output file: %v", err) - } + // Wait until application has ran. + if err := waitForFileNotEmpty(outputFile2); err != nil { + t.Fatalf("Failed to wait for output file: %v", err) + } - firstNum, err := readOutputNum(outputPath, 0) - if err != nil { - t.Fatalf("error with outputFile: %v", err) - } + firstNum, err := readOutputNum(outputPath, 0) + if err != nil { + t.Fatalf("error with outputFile: %v", err) + } - // Check that lastNum is one less than firstNum and that the container picks - // up from where it left off. - if lastNum+1 != firstNum { - t.Errorf("error numbers not in order, previous: %d, next: %d", lastNum, firstNum) - } - cont2.Destroy() + // Check that lastNum is one less than firstNum and that the container picks + // up from where it left off. + if lastNum+1 != firstNum { + t.Errorf("error numbers not in order, previous: %d, next: %d", lastNum, firstNum) + } + cont2.Destroy() - // Restore into another container! - // Delete and recreate file before restoring. - if err := os.Remove(outputPath); err != nil { - t.Fatalf("error removing file") - } - outputFile3, err := createWriteableOutputFile(outputPath) - if err != nil { - t.Fatalf("error creating output file: %v", err) - } - defer outputFile3.Close() + // Restore into another container! + // Delete and recreate file before restoring. + if err := os.Remove(outputPath); err != nil { + t.Fatalf("error removing file") + } + outputFile3, err := createWriteableOutputFile(outputPath) + if err != nil { + t.Fatalf("error creating output file: %v", err) + } + defer outputFile3.Close() - // Restore into a new container. - args3 := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - } - cont3, err := New(conf, args3) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - defer cont3.Destroy() + // Restore into a new container. + args3 := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + cont3, err := New(conf, args3) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer cont3.Destroy() - if err := cont3.Restore(spec, conf, imagePath); err != nil { - t.Fatalf("error restoring container: %v", err) - } + if err := cont3.Restore(spec, conf, imagePath); err != nil { + t.Fatalf("error restoring container: %v", err) + } - // Wait until application has ran. - if err := waitForFileNotEmpty(outputFile3); err != nil { - t.Fatalf("Failed to wait for output file: %v", err) - } + // Wait until application has ran. + if err := waitForFileNotEmpty(outputFile3); err != nil { + t.Fatalf("Failed to wait for output file: %v", err) + } - firstNum2, err := readOutputNum(outputPath, 0) - if err != nil { - t.Fatalf("error with outputFile: %v", err) - } + firstNum2, err := readOutputNum(outputPath, 0) + if err != nil { + t.Fatalf("error with outputFile: %v", err) + } - // Check that lastNum is one less than firstNum and that the container picks - // up from where it left off. - if lastNum+1 != firstNum2 { - t.Errorf("error numbers not in order, previous: %d, next: %d", lastNum, firstNum2) - } - cont3.Destroy() + // Check that lastNum is one less than firstNum and that the container picks + // up from where it left off. + if lastNum+1 != firstNum2 { + t.Errorf("error numbers not in order, previous: %d, next: %d", lastNum, firstNum2) + } + cont3.Destroy() + }) } } @@ -914,135 +917,134 @@ func TestCheckpointRestore(t *testing.T) { // with filesystem Unix Domain Socket use. func TestUnixDomainSockets(t *testing.T) { // Skip overlay because test requires writing to host file. - for _, conf := range configs(t, noOverlay...) { - t.Logf("Running test with conf: %+v", conf) - - // UDS path is limited to 108 chars for compatibility with older systems. - // Use '/tmp' (instead of testutil.TmpDir) to ensure the size limit is - // not exceeded. Assumes '/tmp' exists in the system. - dir, err := ioutil.TempDir("/tmp", "uds-test") - if err != nil { - t.Fatalf("ioutil.TempDir failed: %v", err) - } - defer os.RemoveAll(dir) + for name, conf := range configs(t, noOverlay...) { + t.Run(name, func(t *testing.T) { + // UDS path is limited to 108 chars for compatibility with older systems. + // Use '/tmp' (instead of testutil.TmpDir) to ensure the size limit is + // not exceeded. Assumes '/tmp' exists in the system. + dir, err := ioutil.TempDir("/tmp", "uds-test") + if err != nil { + t.Fatalf("ioutil.TempDir failed: %v", err) + } + defer os.RemoveAll(dir) - outputPath := filepath.Join(dir, "uds_output") - outputFile, err := os.OpenFile(outputPath, os.O_CREATE|os.O_EXCL|os.O_RDWR, 0666) - if err != nil { - t.Fatalf("error creating output file: %v", err) - } - defer outputFile.Close() + outputPath := filepath.Join(dir, "uds_output") + outputFile, err := os.OpenFile(outputPath, os.O_CREATE|os.O_EXCL|os.O_RDWR, 0666) + if err != nil { + t.Fatalf("error creating output file: %v", err) + } + defer outputFile.Close() - app, err := testutil.FindFile("runsc/container/test_app/test_app") - if err != nil { - t.Fatal("error finding test_app:", err) - } + app, err := testutil.FindFile("test/cmd/test_app/test_app") + if err != nil { + t.Fatal("error finding test_app:", err) + } - socketPath := filepath.Join(dir, "uds_socket") - defer os.Remove(socketPath) + socketPath := filepath.Join(dir, "uds_socket") + defer os.Remove(socketPath) - spec := testutil.NewSpecWithArgs(app, "uds", "--file", outputPath, "--socket", socketPath) - spec.Process.User = specs.User{ - UID: uint32(os.Getuid()), - GID: uint32(os.Getgid()), - } - spec.Mounts = []specs.Mount{{ - Type: "bind", - Destination: dir, - Source: dir, - }} + spec := testutil.NewSpecWithArgs(app, "uds", "--file", outputPath, "--socket", socketPath) + spec.Process.User = specs.User{ + UID: uint32(os.Getuid()), + GID: uint32(os.Getgid()), + } + spec.Mounts = []specs.Mount{{ + Type: "bind", + Destination: dir, + Source: dir, + }} - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) - if err != nil { - t.Fatalf("error setting up container: %v", err) - } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() - // Create and start the container. - args := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - } - cont, err := New(conf, args) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - defer cont.Destroy() - if err := cont.Start(conf); err != nil { - t.Fatalf("error starting container: %v", err) - } + // Create and start the container. + args := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + cont, err := New(conf, args) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer cont.Destroy() + if err := cont.Start(conf); err != nil { + t.Fatalf("error starting container: %v", err) + } - // Set the image path, the location where the checkpoint image will be saved. - imagePath := filepath.Join(dir, "test-image-file") + // Set the image path, the location where the checkpoint image will be saved. + imagePath := filepath.Join(dir, "test-image-file") - // Create the image file and open for writing. - file, err := os.OpenFile(imagePath, os.O_CREATE|os.O_EXCL|os.O_RDWR, 0644) - if err != nil { - t.Fatalf("error opening new file at imagePath: %v", err) - } - defer file.Close() - defer os.RemoveAll(imagePath) + // Create the image file and open for writing. + file, err := os.OpenFile(imagePath, os.O_CREATE|os.O_EXCL|os.O_RDWR, 0644) + if err != nil { + t.Fatalf("error opening new file at imagePath: %v", err) + } + defer file.Close() + defer os.RemoveAll(imagePath) - // Wait until application has ran. - if err := waitForFileNotEmpty(outputFile); err != nil { - t.Fatalf("Failed to wait for output file: %v", err) - } + // Wait until application has ran. + if err := waitForFileNotEmpty(outputFile); err != nil { + t.Fatalf("Failed to wait for output file: %v", err) + } - // Checkpoint running container; save state into new file. - if err := cont.Checkpoint(file); err != nil { - t.Fatalf("error checkpointing container to empty file: %v", err) - } + // Checkpoint running container; save state into new file. + if err := cont.Checkpoint(file); err != nil { + t.Fatalf("error checkpointing container to empty file: %v", err) + } - // Read last number outputted before checkpoint. - lastNum, err := readOutputNum(outputPath, -1) - if err != nil { - t.Fatalf("error with outputFile: %v", err) - } + // Read last number outputted before checkpoint. + lastNum, err := readOutputNum(outputPath, -1) + if err != nil { + t.Fatalf("error with outputFile: %v", err) + } - // Delete and recreate file before restoring. - if err := os.Remove(outputPath); err != nil { - t.Fatalf("error removing file") - } - outputFile2, err := os.OpenFile(outputPath, os.O_CREATE|os.O_EXCL|os.O_RDWR, 0666) - if err != nil { - t.Fatalf("error creating output file: %v", err) - } - defer outputFile2.Close() + // Delete and recreate file before restoring. + if err := os.Remove(outputPath); err != nil { + t.Fatalf("error removing file") + } + outputFile2, err := os.OpenFile(outputPath, os.O_CREATE|os.O_EXCL|os.O_RDWR, 0666) + if err != nil { + t.Fatalf("error creating output file: %v", err) + } + defer outputFile2.Close() - // Restore into a new container. - argsRestore := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - } - contRestore, err := New(conf, argsRestore) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - defer contRestore.Destroy() + // Restore into a new container. + argsRestore := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + contRestore, err := New(conf, argsRestore) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer contRestore.Destroy() - if err := contRestore.Restore(spec, conf, imagePath); err != nil { - t.Fatalf("error restoring container: %v", err) - } + if err := contRestore.Restore(spec, conf, imagePath); err != nil { + t.Fatalf("error restoring container: %v", err) + } - // Wait until application has ran. - if err := waitForFileNotEmpty(outputFile2); err != nil { - t.Fatalf("Failed to wait for output file: %v", err) - } + // Wait until application has ran. + if err := waitForFileNotEmpty(outputFile2); err != nil { + t.Fatalf("Failed to wait for output file: %v", err) + } - // Read first number outputted after restore. - firstNum, err := readOutputNum(outputPath, 0) - if err != nil { - t.Fatalf("error with outputFile: %v", err) - } + // Read first number outputted after restore. + firstNum, err := readOutputNum(outputPath, 0) + if err != nil { + t.Fatalf("error with outputFile: %v", err) + } - // Check that lastNum is one less than firstNum. - if lastNum+1 != firstNum { - t.Errorf("error numbers not consecutive, previous: %d, next: %d", lastNum, firstNum) - } - contRestore.Destroy() + // Check that lastNum is one less than firstNum. + if lastNum+1 != firstNum { + t.Errorf("error numbers not consecutive, previous: %d, next: %d", lastNum, firstNum) + } + contRestore.Destroy() + }) } } @@ -1052,10 +1054,8 @@ func TestUnixDomainSockets(t *testing.T) { // recreated. Then it resumes the container, verify that the file gets created // again. func TestPauseResume(t *testing.T) { - for _, conf := range configs(t, noOverlay...) { - t.Run(fmt.Sprintf("conf: %+v", conf), func(t *testing.T) { - t.Logf("Running test with conf: %+v", conf) - + for name, conf := range configs(t, noOverlay...) { + t.Run(name, func(t *testing.T) { tmpDir, err := ioutil.TempDir(testutil.TmpDir(), "lock") if err != nil { t.Fatalf("error creating temp dir: %v", err) @@ -1066,16 +1066,15 @@ func TestPauseResume(t *testing.T) { script := fmt.Sprintf("while [[ true ]]; do touch %q; sleep 0.1; done", running) spec := testutil.NewSpecWithArgs("/bin/bash", "-c", script) - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() // Create and start the container. args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, } @@ -1134,16 +1133,15 @@ func TestPauseResume(t *testing.T) { func TestPauseResumeStatus(t *testing.T) { spec := testutil.NewSpecWithArgs("sleep", "20") conf := testutil.TestConfig(t) - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() // Create and start the container. args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, } @@ -1199,359 +1197,356 @@ func TestCapabilities(t *testing.T) { uid := auth.KUID(os.Getuid() + 1) gid := auth.KGID(os.Getgid() + 1) - for _, conf := range configs(t, all...) { - t.Logf("Running test with conf: %+v", conf) - - spec := testutil.NewSpecWithArgs("sleep", "100") - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) - if err != nil { - t.Fatalf("error setting up container: %v", err) - } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + for name, conf := range configs(t, all...) { + t.Run(name, func(t *testing.T) { + spec := testutil.NewSpecWithArgs("sleep", "100") + rootDir, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() - // Create and start the container. - args := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - } - cont, err := New(conf, args) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - defer cont.Destroy() - if err := cont.Start(conf); err != nil { - t.Fatalf("error starting container: %v", err) - } + // Create and start the container. + args := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + cont, err := New(conf, args) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer cont.Destroy() + if err := cont.Start(conf); err != nil { + t.Fatalf("error starting container: %v", err) + } - // expectedPL lists the expected process state of the container. - expectedPL := []*control.Process{ - { - UID: 0, - PID: 1, - PPID: 0, - C: 0, - Cmd: "sleep", - Threads: []kernel.ThreadID{1}, - }, - { - UID: uid, - PID: 2, - PPID: 0, - C: 0, - Cmd: "exe", - Threads: []kernel.ThreadID{2}, - }, - } - if err := waitForProcessList(cont, expectedPL[:1]); err != nil { - t.Fatalf("Failed to wait for sleep to start, err: %v", err) - } + // expectedPL lists the expected process state of the container. + expectedPL := []*control.Process{ + { + UID: 0, + PID: 1, + PPID: 0, + C: 0, + Cmd: "sleep", + Threads: []kernel.ThreadID{1}, + }, + { + UID: uid, + PID: 2, + PPID: 0, + C: 0, + Cmd: "exe", + Threads: []kernel.ThreadID{2}, + }, + } + if err := waitForProcessList(cont, expectedPL[:1]); err != nil { + t.Fatalf("Failed to wait for sleep to start, err: %v", err) + } - // Create an executable that can't be run with the specified UID:GID. - // This shouldn't be callable within the container until we add the - // CAP_DAC_OVERRIDE capability to skip the access check. - exePath := filepath.Join(rootDir, "exe") - if err := ioutil.WriteFile(exePath, []byte("#!/bin/sh\necho hello"), 0770); err != nil { - t.Fatalf("couldn't create executable: %v", err) - } - defer os.Remove(exePath) - - // Need to traverse the intermediate directory. - os.Chmod(rootDir, 0755) - - execArgs := &control.ExecArgs{ - Filename: exePath, - Argv: []string{exePath}, - WorkingDirectory: "/", - KUID: uid, - KGID: gid, - Capabilities: &auth.TaskCapabilities{}, - } + // Create an executable that can't be run with the specified UID:GID. + // This shouldn't be callable within the container until we add the + // CAP_DAC_OVERRIDE capability to skip the access check. + exePath := filepath.Join(rootDir, "exe") + if err := ioutil.WriteFile(exePath, []byte("#!/bin/sh\necho hello"), 0770); err != nil { + t.Fatalf("couldn't create executable: %v", err) + } + defer os.Remove(exePath) + + // Need to traverse the intermediate directory. + os.Chmod(rootDir, 0755) + + execArgs := &control.ExecArgs{ + Filename: exePath, + Argv: []string{exePath}, + WorkingDirectory: "/", + KUID: uid, + KGID: gid, + Capabilities: &auth.TaskCapabilities{}, + } - // "exe" should fail because we don't have the necessary permissions. - if _, err := cont.executeSync(execArgs); err == nil { - t.Fatalf("container executed without error, but an error was expected") - } + // "exe" should fail because we don't have the necessary permissions. + if _, err := cont.executeSync(execArgs); err == nil { + t.Fatalf("container executed without error, but an error was expected") + } - // Now we run with the capability enabled and should succeed. - execArgs.Capabilities = &auth.TaskCapabilities{ - EffectiveCaps: auth.CapabilitySetOf(linux.CAP_DAC_OVERRIDE), - } - // "exe" should not fail this time. - if _, err := cont.executeSync(execArgs); err != nil { - t.Fatalf("container failed to exec %v: %v", args, err) - } + // Now we run with the capability enabled and should succeed. + execArgs.Capabilities = &auth.TaskCapabilities{ + EffectiveCaps: auth.CapabilitySetOf(linux.CAP_DAC_OVERRIDE), + } + // "exe" should not fail this time. + if _, err := cont.executeSync(execArgs); err != nil { + t.Fatalf("container failed to exec %v: %v", args, err) + } + }) } } // TestRunNonRoot checks that sandbox can be configured when running as // non-privileged user. func TestRunNonRoot(t *testing.T) { - for _, conf := range configs(t, noOverlay...) { - t.Logf("Running test with conf: %+v", conf) - - spec := testutil.NewSpecWithArgs("/bin/true") - - // Set a random user/group with no access to "blocked" dir. - spec.Process.User.UID = 343 - spec.Process.User.GID = 2401 - spec.Process.Capabilities = nil + for name, conf := range configs(t, noOverlay...) { + t.Run(name, func(t *testing.T) { + spec := testutil.NewSpecWithArgs("/bin/true") + + // Set a random user/group with no access to "blocked" dir. + spec.Process.User.UID = 343 + spec.Process.User.GID = 2401 + spec.Process.Capabilities = nil + + // User running inside container can't list '$TMP/blocked' and would fail to + // mount it. + dir, err := ioutil.TempDir(testutil.TmpDir(), "blocked") + if err != nil { + t.Fatalf("ioutil.TempDir() failed: %v", err) + } + if err := os.Chmod(dir, 0700); err != nil { + t.Fatalf("os.MkDir(%q) failed: %v", dir, err) + } + dir = path.Join(dir, "test") + if err := os.Mkdir(dir, 0755); err != nil { + t.Fatalf("os.MkDir(%q) failed: %v", dir, err) + } - // User running inside container can't list '$TMP/blocked' and would fail to - // mount it. - dir, err := ioutil.TempDir(testutil.TmpDir(), "blocked") - if err != nil { - t.Fatalf("ioutil.TempDir() failed: %v", err) - } - if err := os.Chmod(dir, 0700); err != nil { - t.Fatalf("os.MkDir(%q) failed: %v", dir, err) - } - dir = path.Join(dir, "test") - if err := os.Mkdir(dir, 0755); err != nil { - t.Fatalf("os.MkDir(%q) failed: %v", dir, err) - } + src, err := ioutil.TempDir(testutil.TmpDir(), "src") + if err != nil { + t.Fatalf("ioutil.TempDir() failed: %v", err) + } - src, err := ioutil.TempDir(testutil.TmpDir(), "src") - if err != nil { - t.Fatalf("ioutil.TempDir() failed: %v", err) - } + spec.Mounts = append(spec.Mounts, specs.Mount{ + Destination: dir, + Source: src, + Type: "bind", + }) - spec.Mounts = append(spec.Mounts, specs.Mount{ - Destination: dir, - Source: src, - Type: "bind", + if err := run(spec, conf); err != nil { + t.Fatalf("error running sandbox: %v", err) + } }) - - if err := run(spec, conf); err != nil { - t.Fatalf("error running sandbox: %v", err) - } } } // TestMountNewDir checks that runsc will create destination directory if it // doesn't exit. func TestMountNewDir(t *testing.T) { - for _, conf := range configs(t, overlay) { - t.Logf("Running test with conf: %+v", conf) + for name, conf := range configs(t, overlay) { + t.Run(name, func(t *testing.T) { + root, err := ioutil.TempDir(testutil.TmpDir(), "root") + if err != nil { + t.Fatal("ioutil.TempDir() failed:", err) + } - root, err := ioutil.TempDir(testutil.TmpDir(), "root") - if err != nil { - t.Fatal("ioutil.TempDir() failed:", err) - } + srcDir := path.Join(root, "src", "dir", "anotherdir") + if err := os.MkdirAll(srcDir, 0755); err != nil { + t.Fatalf("os.MkDir(%q) failed: %v", srcDir, err) + } - srcDir := path.Join(root, "src", "dir", "anotherdir") - if err := os.MkdirAll(srcDir, 0755); err != nil { - t.Fatalf("os.MkDir(%q) failed: %v", srcDir, err) - } + mountDir := path.Join(root, "dir", "anotherdir") - mountDir := path.Join(root, "dir", "anotherdir") + spec := testutil.NewSpecWithArgs("/bin/ls", mountDir) + spec.Mounts = append(spec.Mounts, specs.Mount{ + Destination: mountDir, + Source: srcDir, + Type: "bind", + }) - spec := testutil.NewSpecWithArgs("/bin/ls", mountDir) - spec.Mounts = append(spec.Mounts, specs.Mount{ - Destination: mountDir, - Source: srcDir, - Type: "bind", + if err := run(spec, conf); err != nil { + t.Fatalf("error running sandbox: %v", err) + } }) - - if err := run(spec, conf); err != nil { - t.Fatalf("error running sandbox: %v", err) - } } } func TestReadonlyRoot(t *testing.T) { - for _, conf := range configs(t, overlay) { - t.Logf("Running test with conf: %+v", conf) - - spec := testutil.NewSpecWithArgs("/bin/touch", "/foo") - spec.Root.Readonly = true - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) - if err != nil { - t.Fatalf("error setting up container: %v", err) - } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + for name, conf := range configs(t, overlay) { + t.Run(name, func(t *testing.T) { + spec := testutil.NewSpecWithArgs("/bin/touch", "/foo") + spec.Root.Readonly = true + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() - // Create, start and wait for the container. - args := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - } - c, err := New(conf, args) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - defer c.Destroy() - if err := c.Start(conf); err != nil { - t.Fatalf("error starting container: %v", err) - } + // Create, start and wait for the container. + args := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + c, err := New(conf, args) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer c.Destroy() + if err := c.Start(conf); err != nil { + t.Fatalf("error starting container: %v", err) + } - ws, err := c.Wait() - if err != nil { - t.Fatalf("error waiting on container: %v", err) - } - if !ws.Exited() || syscall.Errno(ws.ExitStatus()) != syscall.EPERM { - t.Fatalf("container failed, waitStatus: %v", ws) - } + ws, err := c.Wait() + if err != nil { + t.Fatalf("error waiting on container: %v", err) + } + if !ws.Exited() || syscall.Errno(ws.ExitStatus()) != syscall.EPERM { + t.Fatalf("container failed, waitStatus: %v", ws) + } + }) } } func TestUIDMap(t *testing.T) { - for _, conf := range configs(t, noOverlay...) { - t.Logf("Running test with conf: %+v", conf) - testDir, err := ioutil.TempDir(testutil.TmpDir(), "test-mount") - if err != nil { - t.Fatal(err) - } - defer os.RemoveAll(testDir) - testFile := path.Join(testDir, "testfile") - - spec := testutil.NewSpecWithArgs("touch", "/tmp/testfile") - uid := os.Getuid() - gid := os.Getgid() - spec.Linux = &specs.Linux{ - Namespaces: []specs.LinuxNamespace{ - {Type: specs.UserNamespace}, - {Type: specs.PIDNamespace}, - {Type: specs.MountNamespace}, - }, - UIDMappings: []specs.LinuxIDMapping{ - { - ContainerID: 0, - HostID: uint32(uid), - Size: 1, + for name, conf := range configs(t, noOverlay...) { + t.Run(name, func(t *testing.T) { + testDir, err := ioutil.TempDir(testutil.TmpDir(), "test-mount") + if err != nil { + t.Fatalf("ioutil.TempDir() failed: %v", err) + } + defer os.RemoveAll(testDir) + testFile := path.Join(testDir, "testfile") + + spec := testutil.NewSpecWithArgs("touch", "/tmp/testfile") + uid := os.Getuid() + gid := os.Getgid() + spec.Linux = &specs.Linux{ + Namespaces: []specs.LinuxNamespace{ + {Type: specs.UserNamespace}, + {Type: specs.PIDNamespace}, + {Type: specs.MountNamespace}, }, - }, - GIDMappings: []specs.LinuxIDMapping{ - { - ContainerID: 0, - HostID: uint32(gid), - Size: 1, + UIDMappings: []specs.LinuxIDMapping{ + { + ContainerID: 0, + HostID: uint32(uid), + Size: 1, + }, }, - }, - } + GIDMappings: []specs.LinuxIDMapping{ + { + ContainerID: 0, + HostID: uint32(gid), + Size: 1, + }, + }, + } - spec.Mounts = append(spec.Mounts, specs.Mount{ - Destination: "/tmp", - Source: testDir, - Type: "bind", - }) + spec.Mounts = append(spec.Mounts, specs.Mount{ + Destination: "/tmp", + Source: testDir, + Type: "bind", + }) - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) - if err != nil { - t.Fatalf("error setting up container: %v", err) - } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() - // Create, start and wait for the container. - args := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - } - c, err := New(conf, args) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - defer c.Destroy() - if err := c.Start(conf); err != nil { - t.Fatalf("error starting container: %v", err) - } + // Create, start and wait for the container. + args := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + c, err := New(conf, args) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer c.Destroy() + if err := c.Start(conf); err != nil { + t.Fatalf("error starting container: %v", err) + } - ws, err := c.Wait() - if err != nil { - t.Fatalf("error waiting on container: %v", err) - } - if !ws.Exited() || ws.ExitStatus() != 0 { - t.Fatalf("container failed, waitStatus: %v", ws) - } - st := syscall.Stat_t{} - if err := syscall.Stat(testFile, &st); err != nil { - t.Fatalf("error stat /testfile: %v", err) - } + ws, err := c.Wait() + if err != nil { + t.Fatalf("error waiting on container: %v", err) + } + if !ws.Exited() || ws.ExitStatus() != 0 { + t.Fatalf("container failed, waitStatus: %v", ws) + } + st := syscall.Stat_t{} + if err := syscall.Stat(testFile, &st); err != nil { + t.Fatalf("error stat /testfile: %v", err) + } - if st.Uid != uint32(uid) || st.Gid != uint32(gid) { - t.Fatalf("UID: %d (%d) GID: %d (%d)", st.Uid, uid, st.Gid, gid) - } + if st.Uid != uint32(uid) || st.Gid != uint32(gid) { + t.Fatalf("UID: %d (%d) GID: %d (%d)", st.Uid, uid, st.Gid, gid) + } + }) } } func TestReadonlyMount(t *testing.T) { - for _, conf := range configs(t, overlay) { - t.Logf("Running test with conf: %+v", conf) - - dir, err := ioutil.TempDir(testutil.TmpDir(), "ro-mount") - spec := testutil.NewSpecWithArgs("/bin/touch", path.Join(dir, "file")) - if err != nil { - t.Fatalf("ioutil.TempDir() failed: %v", err) - } - spec.Mounts = append(spec.Mounts, specs.Mount{ - Destination: dir, - Source: dir, - Type: "bind", - Options: []string{"ro"}, - }) - spec.Root.Readonly = false - - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) - if err != nil { - t.Fatalf("error setting up container: %v", err) - } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + for name, conf := range configs(t, overlay) { + t.Run(name, func(t *testing.T) { + dir, err := ioutil.TempDir(testutil.TmpDir(), "ro-mount") + spec := testutil.NewSpecWithArgs("/bin/touch", path.Join(dir, "file")) + if err != nil { + t.Fatalf("ioutil.TempDir() failed: %v", err) + } + spec.Mounts = append(spec.Mounts, specs.Mount{ + Destination: dir, + Source: dir, + Type: "bind", + Options: []string{"ro"}, + }) + spec.Root.Readonly = false + + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() - // Create, start and wait for the container. - args := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - } - c, err := New(conf, args) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - defer c.Destroy() - if err := c.Start(conf); err != nil { - t.Fatalf("error starting container: %v", err) - } + // Create, start and wait for the container. + args := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + c, err := New(conf, args) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer c.Destroy() + if err := c.Start(conf); err != nil { + t.Fatalf("error starting container: %v", err) + } - ws, err := c.Wait() - if err != nil { - t.Fatalf("error waiting on container: %v", err) - } - if !ws.Exited() || syscall.Errno(ws.ExitStatus()) != syscall.EPERM { - t.Fatalf("container failed, waitStatus: %v", ws) - } + ws, err := c.Wait() + if err != nil { + t.Fatalf("error waiting on container: %v", err) + } + if !ws.Exited() || syscall.Errno(ws.ExitStatus()) != syscall.EPERM { + t.Fatalf("container failed, waitStatus: %v", ws) + } + }) } } // TestAbbreviatedIDs checks that runsc supports using abbreviated container // IDs in place of full IDs. func TestAbbreviatedIDs(t *testing.T) { - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { t.Fatalf("error creating root dir: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() conf := testutil.TestConfig(t) conf.RootDir = rootDir cids := []string{ - "foo-" + testutil.UniqueContainerID(), - "bar-" + testutil.UniqueContainerID(), - "baz-" + testutil.UniqueContainerID(), + "foo-" + testutil.RandomContainerID(), + "bar-" + testutil.RandomContainerID(), + "baz-" + testutil.RandomContainerID(), } for _, cid := range cids { spec := testutil.NewSpecWithArgs("sleep", "100") - bundleDir, err := testutil.SetupBundleDir(spec) + bundleDir, cleanup, err := testutil.SetupBundleDir(spec) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(bundleDir) + defer cleanup() // Create and start the container. args := Args{ @@ -1596,16 +1591,15 @@ func TestAbbreviatedIDs(t *testing.T) { func TestGoferExits(t *testing.T) { spec := testutil.NewSpecWithArgs("/bin/sleep", "10000") conf := testutil.TestConfig(t) - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() // Create and start the container. args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, } @@ -1634,7 +1628,7 @@ func TestGoferExits(t *testing.T) { } func TestRootNotMount(t *testing.T) { - appSym, err := testutil.FindFile("runsc/container/test_app/test_app") + appSym, err := testutil.FindFile("test/cmd/test_app/test_app") if err != nil { t.Fatal("error finding test_app:", err) } @@ -1671,7 +1665,7 @@ func TestRootNotMount(t *testing.T) { } func TestUserLog(t *testing.T) { - app, err := testutil.FindFile("runsc/container/test_app/test_app") + app, err := testutil.FindFile("test/cmd/test_app/test_app") if err != nil { t.Fatal("error finding test_app:", err) } @@ -1679,12 +1673,11 @@ func TestUserLog(t *testing.T) { // sched_rr_get_interval = 148 - not implemented in gvisor. spec := testutil.NewSpecWithArgs(app, "syscall", "--syscall=148") conf := testutil.TestConfig(t) - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() dir, err := ioutil.TempDir(testutil.TmpDir(), "user_log_test") if err != nil { @@ -1694,7 +1687,7 @@ func TestUserLog(t *testing.T) { // Create, start and wait for the container. args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, UserLog: userLog, @@ -1718,72 +1711,70 @@ func TestUserLog(t *testing.T) { } func TestWaitOnExitedSandbox(t *testing.T) { - for _, conf := range configs(t, all...) { - t.Logf("Running test with conf: %+v", conf) - - // Run a shell that sleeps for 1 second and then exits with a - // non-zero code. - const wantExit = 17 - cmd := fmt.Sprintf("sleep 1; exit %d", wantExit) - spec := testutil.NewSpecWithArgs("/bin/sh", "-c", cmd) - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) - if err != nil { - t.Fatalf("error setting up container: %v", err) - } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + for name, conf := range configs(t, all...) { + t.Run(name, func(t *testing.T) { + // Run a shell that sleeps for 1 second and then exits with a + // non-zero code. + const wantExit = 17 + cmd := fmt.Sprintf("sleep 1; exit %d", wantExit) + spec := testutil.NewSpecWithArgs("/bin/sh", "-c", cmd) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() - // Create and Start the container. - args := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - } - c, err := New(conf, args) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - defer c.Destroy() - if err := c.Start(conf); err != nil { - t.Fatalf("error starting container: %v", err) - } + // Create and Start the container. + args := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + c, err := New(conf, args) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer c.Destroy() + if err := c.Start(conf); err != nil { + t.Fatalf("error starting container: %v", err) + } - // Wait on the sandbox. This will make an RPC to the sandbox - // and get the actual exit status of the application. - ws, err := c.Wait() - if err != nil { - t.Fatalf("error waiting on container: %v", err) - } - if got := ws.ExitStatus(); got != wantExit { - t.Errorf("got exit status %d, want %d", got, wantExit) - } + // Wait on the sandbox. This will make an RPC to the sandbox + // and get the actual exit status of the application. + ws, err := c.Wait() + if err != nil { + t.Fatalf("error waiting on container: %v", err) + } + if got := ws.ExitStatus(); got != wantExit { + t.Errorf("got exit status %d, want %d", got, wantExit) + } - // Now the sandbox has exited, but the zombie sandbox process - // still exists. Calling Wait() now will return the sandbox - // exit status. - ws, err = c.Wait() - if err != nil { - t.Fatalf("error waiting on container: %v", err) - } - if got := ws.ExitStatus(); got != wantExit { - t.Errorf("got exit status %d, want %d", got, wantExit) - } + // Now the sandbox has exited, but the zombie sandbox process + // still exists. Calling Wait() now will return the sandbox + // exit status. + ws, err = c.Wait() + if err != nil { + t.Fatalf("error waiting on container: %v", err) + } + if got := ws.ExitStatus(); got != wantExit { + t.Errorf("got exit status %d, want %d", got, wantExit) + } + }) } } func TestDestroyNotStarted(t *testing.T) { spec := testutil.NewSpecWithArgs("/bin/sleep", "100") conf := testutil.TestConfig(t) - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() // Create the container and check that it can be destroyed. args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, } @@ -1801,16 +1792,15 @@ func TestDestroyStarting(t *testing.T) { for i := 0; i < 10; i++ { spec := testutil.NewSpecWithArgs("/bin/sleep", "100") conf := testutil.TestConfig(t) - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + rootDir, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() // Create the container and check that it can be destroyed. args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, } @@ -1845,23 +1835,23 @@ func TestDestroyStarting(t *testing.T) { } func TestCreateWorkingDir(t *testing.T) { - for _, conf := range configs(t, overlay) { - t.Logf("Running test with conf: %+v", conf) - - tmpDir, err := ioutil.TempDir(testutil.TmpDir(), "cwd-create") - if err != nil { - t.Fatalf("ioutil.TempDir() failed: %v", err) - } - dir := path.Join(tmpDir, "new/working/dir") + for name, conf := range configs(t, overlay) { + t.Run(name, func(t *testing.T) { + tmpDir, err := ioutil.TempDir(testutil.TmpDir(), "cwd-create") + if err != nil { + t.Fatalf("ioutil.TempDir() failed: %v", err) + } + dir := path.Join(tmpDir, "new/working/dir") - // touch will fail if the directory doesn't exist. - spec := testutil.NewSpecWithArgs("/bin/touch", path.Join(dir, "file")) - spec.Process.Cwd = dir - spec.Root.Readonly = true + // touch will fail if the directory doesn't exist. + spec := testutil.NewSpecWithArgs("/bin/touch", path.Join(dir, "file")) + spec.Process.Cwd = dir + spec.Root.Readonly = true - if err := run(spec, conf); err != nil { - t.Fatalf("Error running container: %v", err) - } + if err := run(spec, conf); err != nil { + t.Fatalf("Error running container: %v", err) + } + }) } } @@ -1919,15 +1909,14 @@ func TestMountPropagation(t *testing.T) { } conf := testutil.TestConfig(t) - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, } @@ -1969,81 +1958,81 @@ func TestMountPropagation(t *testing.T) { } func TestMountSymlink(t *testing.T) { - for _, conf := range configs(t, overlay) { - t.Logf("Running test with conf: %+v", conf) - - dir, err := ioutil.TempDir(testutil.TmpDir(), "mount-symlink") - if err != nil { - t.Fatalf("ioutil.TempDir() failed: %v", err) - } + for name, conf := range configs(t, overlay) { + t.Run(name, func(t *testing.T) { + dir, err := ioutil.TempDir(testutil.TmpDir(), "mount-symlink") + if err != nil { + t.Fatalf("ioutil.TempDir() failed: %v", err) + } + defer os.RemoveAll(dir) - source := path.Join(dir, "source") - target := path.Join(dir, "target") - for _, path := range []string{source, target} { - if err := os.MkdirAll(path, 0777); err != nil { - t.Fatalf("os.MkdirAll(): %v", err) + source := path.Join(dir, "source") + target := path.Join(dir, "target") + for _, path := range []string{source, target} { + if err := os.MkdirAll(path, 0777); err != nil { + t.Fatalf("os.MkdirAll(): %v", err) + } } - } - f, err := os.Create(path.Join(source, "file")) - if err != nil { - t.Fatalf("os.Create(): %v", err) - } - f.Close() + f, err := os.Create(path.Join(source, "file")) + if err != nil { + t.Fatalf("os.Create(): %v", err) + } + f.Close() - link := path.Join(dir, "link") - if err := os.Symlink(target, link); err != nil { - t.Fatalf("os.Symlink(%q, %q): %v", target, link, err) - } + link := path.Join(dir, "link") + if err := os.Symlink(target, link); err != nil { + t.Fatalf("os.Symlink(%q, %q): %v", target, link, err) + } - spec := testutil.NewSpecWithArgs("/bin/sleep", "1000") + spec := testutil.NewSpecWithArgs("/bin/sleep", "1000") - // Mount to a symlink to ensure the mount code will follow it and mount - // at the symlink target. - spec.Mounts = append(spec.Mounts, specs.Mount{ - Type: "bind", - Destination: link, - Source: source, - }) + // Mount to a symlink to ensure the mount code will follow it and mount + // at the symlink target. + spec.Mounts = append(spec.Mounts, specs.Mount{ + Type: "bind", + Destination: link, + Source: source, + }) - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) - if err != nil { - t.Fatalf("error setting up container: %v", err) - } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() - args := Args{ - ID: testutil.UniqueContainerID(), - Spec: spec, - BundleDir: bundleDir, - } - cont, err := New(conf, args) - if err != nil { - t.Fatalf("creating container: %v", err) - } - defer cont.Destroy() + args := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + cont, err := New(conf, args) + if err != nil { + t.Fatalf("creating container: %v", err) + } + defer cont.Destroy() - if err := cont.Start(conf); err != nil { - t.Fatalf("starting container: %v", err) - } + if err := cont.Start(conf); err != nil { + t.Fatalf("starting container: %v", err) + } - // Check that symlink was resolved and mount was created where the symlink - // is pointing to. - file := path.Join(target, "file") - execArgs := &control.ExecArgs{ - Filename: "/usr/bin/test", - Argv: []string{"test", "-f", file}, - } - if ws, err := cont.executeSync(execArgs); err != nil || ws != 0 { - t.Fatalf("exec: test -f %q, ws: %v, err: %v", file, ws, err) - } + // Check that symlink was resolved and mount was created where the symlink + // is pointing to. + file := path.Join(target, "file") + execArgs := &control.ExecArgs{ + Filename: "/usr/bin/test", + Argv: []string{"test", "-f", file}, + } + if ws, err := cont.executeSync(execArgs); err != nil || ws != 0 { + t.Fatalf("exec: test -f %q, ws: %v, err: %v", file, ws, err) + } + }) } } // Check that --net-raw disables the CAP_NET_RAW capability. func TestNetRaw(t *testing.T) { capNetRaw := strconv.FormatUint(bits.MaskOf64(int(linux.CAP_NET_RAW)), 10) - app, err := testutil.FindFile("runsc/container/test_app/test_app") + app, err := testutil.FindFile("test/cmd/test_app/test_app") if err != nil { t.Fatal("error finding test_app:", err) } @@ -2106,7 +2095,7 @@ func TestTTYField(t *testing.T) { stop := testutil.StartReaper() defer stop() - testApp, err := testutil.FindFile("runsc/container/test_app/test_app") + testApp, err := testutil.FindFile("test/cmd/test_app/test_app") if err != nil { t.Fatal("error finding test_app:", err) } @@ -2140,16 +2129,15 @@ func TestTTYField(t *testing.T) { } spec := testutil.NewSpecWithArgs(cmd...) - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() // Create and start the container. args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, } diff --git a/runsc/container/multi_container_test.go b/runsc/container/multi_container_test.go index dc2fb42ce..e3704b453 100644 --- a/runsc/container/multi_container_test.go +++ b/runsc/container/multi_container_test.go @@ -30,15 +30,15 @@ import ( "gvisor.dev/gvisor/pkg/sentry/control" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/test/testutil" "gvisor.dev/gvisor/runsc/boot" "gvisor.dev/gvisor/runsc/specutils" - "gvisor.dev/gvisor/runsc/testutil" ) func createSpecs(cmds ...[]string) ([]*specs.Spec, []string) { var specs []*specs.Spec var ids []string - rootID := testutil.UniqueContainerID() + rootID := testutil.RandomContainerID() for i, cmd := range cmds { spec := testutil.NewSpecWithArgs(cmd...) @@ -52,7 +52,7 @@ func createSpecs(cmds ...[]string) ([]*specs.Spec, []string) { specutils.ContainerdContainerTypeAnnotation: specutils.ContainerdContainerTypeContainer, specutils.ContainerdSandboxIDAnnotation: rootID, } - ids = append(ids, testutil.UniqueContainerID()) + ids = append(ids, testutil.RandomContainerID()) } specs = append(specs, spec) } @@ -64,23 +64,29 @@ func startContainers(conf *boot.Config, specs []*specs.Spec, ids []string) ([]*C panic("conf.RootDir not set. Call testutil.SetupRootDir() to set.") } - var containers []*Container - var bundles []string - cleanup := func() { + var ( + containers []*Container + cleanups []func() + ) + cleanups = append(cleanups, func() { for _, c := range containers { c.Destroy() } - for _, b := range bundles { - os.RemoveAll(b) + }) + cleanupAll := func() { + for _, c := range cleanups { + c() } } + localClean := specutils.MakeCleanup(cleanupAll) + defer localClean.Clean() + for i, spec := range specs { - bundleDir, err := testutil.SetupBundleDir(spec) + bundleDir, cleanup, err := testutil.SetupBundleDir(spec) if err != nil { - cleanup() return nil, nil, fmt.Errorf("error setting up container: %v", err) } - bundles = append(bundles, bundleDir) + cleanups = append(cleanups, cleanup) args := Args{ ID: ids[i], @@ -89,17 +95,17 @@ func startContainers(conf *boot.Config, specs []*specs.Spec, ids []string) ([]*C } cont, err := New(conf, args) if err != nil { - cleanup() return nil, nil, fmt.Errorf("error creating container: %v", err) } containers = append(containers, cont) if err := cont.Start(conf); err != nil { - cleanup() return nil, nil, fmt.Errorf("error starting container: %v", err) } } - return containers, cleanup, nil + + localClean.Release() + return containers, cleanupAll, nil } type execDesc struct { @@ -135,159 +141,159 @@ func createSharedMount(mount specs.Mount, name string, pod ...*specs.Spec) { // TestMultiContainerSanity checks that it is possible to run 2 dead-simple // containers in the same sandbox. func TestMultiContainerSanity(t *testing.T) { - for _, conf := range configs(t, all...) { - t.Logf("Running test with conf: %+v", conf) - - rootDir, err := testutil.SetupRootDir() - if err != nil { - t.Fatalf("error creating root dir: %v", err) - } - defer os.RemoveAll(rootDir) - conf.RootDir = rootDir + for name, conf := range configs(t, all...) { + t.Run(name, func(t *testing.T) { + rootDir, cleanup, err := testutil.SetupRootDir() + if err != nil { + t.Fatalf("error creating root dir: %v", err) + } + defer cleanup() + conf.RootDir = rootDir - // Setup the containers. - sleep := []string{"sleep", "100"} - specs, ids := createSpecs(sleep, sleep) - containers, cleanup, err := startContainers(conf, specs, ids) - if err != nil { - t.Fatalf("error starting containers: %v", err) - } - defer cleanup() + // Setup the containers. + sleep := []string{"sleep", "100"} + specs, ids := createSpecs(sleep, sleep) + containers, cleanup, err := startContainers(conf, specs, ids) + if err != nil { + t.Fatalf("error starting containers: %v", err) + } + defer cleanup() - // Check via ps that multiple processes are running. - expectedPL := []*control.Process{ - {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, - } - if err := waitForProcessList(containers[0], expectedPL); err != nil { - t.Errorf("failed to wait for sleep to start: %v", err) - } - expectedPL = []*control.Process{ - {PID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{2}}, - } - if err := waitForProcessList(containers[1], expectedPL); err != nil { - t.Errorf("failed to wait for sleep to start: %v", err) - } + // Check via ps that multiple processes are running. + expectedPL := []*control.Process{ + {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, + } + if err := waitForProcessList(containers[0], expectedPL); err != nil { + t.Errorf("failed to wait for sleep to start: %v", err) + } + expectedPL = []*control.Process{ + {PID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{2}}, + } + if err := waitForProcessList(containers[1], expectedPL); err != nil { + t.Errorf("failed to wait for sleep to start: %v", err) + } + }) } } // TestMultiPIDNS checks that it is possible to run 2 dead-simple // containers in the same sandbox with different pidns. func TestMultiPIDNS(t *testing.T) { - for _, conf := range configs(t, all...) { - t.Logf("Running test with conf: %+v", conf) - - rootDir, err := testutil.SetupRootDir() - if err != nil { - t.Fatalf("error creating root dir: %v", err) - } - defer os.RemoveAll(rootDir) - conf.RootDir = rootDir - - // Setup the containers. - sleep := []string{"sleep", "100"} - testSpecs, ids := createSpecs(sleep, sleep) - testSpecs[1].Linux = &specs.Linux{ - Namespaces: []specs.LinuxNamespace{ - { - Type: "pid", + for name, conf := range configs(t, all...) { + t.Run(name, func(t *testing.T) { + rootDir, cleanup, err := testutil.SetupRootDir() + if err != nil { + t.Fatalf("error creating root dir: %v", err) + } + defer cleanup() + conf.RootDir = rootDir + + // Setup the containers. + sleep := []string{"sleep", "100"} + testSpecs, ids := createSpecs(sleep, sleep) + testSpecs[1].Linux = &specs.Linux{ + Namespaces: []specs.LinuxNamespace{ + { + Type: "pid", + }, }, - }, - } + } - containers, cleanup, err := startContainers(conf, testSpecs, ids) - if err != nil { - t.Fatalf("error starting containers: %v", err) - } - defer cleanup() + containers, cleanup, err := startContainers(conf, testSpecs, ids) + if err != nil { + t.Fatalf("error starting containers: %v", err) + } + defer cleanup() - // Check via ps that multiple processes are running. - expectedPL := []*control.Process{ - {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, - } - if err := waitForProcessList(containers[0], expectedPL); err != nil { - t.Errorf("failed to wait for sleep to start: %v", err) - } - expectedPL = []*control.Process{ - {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, - } - if err := waitForProcessList(containers[1], expectedPL); err != nil { - t.Errorf("failed to wait for sleep to start: %v", err) - } + // Check via ps that multiple processes are running. + expectedPL := []*control.Process{ + {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, + } + if err := waitForProcessList(containers[0], expectedPL); err != nil { + t.Errorf("failed to wait for sleep to start: %v", err) + } + expectedPL = []*control.Process{ + {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, + } + if err := waitForProcessList(containers[1], expectedPL); err != nil { + t.Errorf("failed to wait for sleep to start: %v", err) + } + }) } } // TestMultiPIDNSPath checks the pidns path. func TestMultiPIDNSPath(t *testing.T) { - for _, conf := range configs(t, all...) { - t.Logf("Running test with conf: %+v", conf) - - rootDir, err := testutil.SetupRootDir() - if err != nil { - t.Fatalf("error creating root dir: %v", err) - } - defer os.RemoveAll(rootDir) - conf.RootDir = rootDir - - // Setup the containers. - sleep := []string{"sleep", "100"} - testSpecs, ids := createSpecs(sleep, sleep, sleep) - testSpecs[0].Linux = &specs.Linux{ - Namespaces: []specs.LinuxNamespace{ - { - Type: "pid", - Path: "/proc/1/ns/pid", + for name, conf := range configs(t, all...) { + t.Run(name, func(t *testing.T) { + rootDir, cleanup, err := testutil.SetupRootDir() + if err != nil { + t.Fatalf("error creating root dir: %v", err) + } + defer cleanup() + conf.RootDir = rootDir + + // Setup the containers. + sleep := []string{"sleep", "100"} + testSpecs, ids := createSpecs(sleep, sleep, sleep) + testSpecs[0].Linux = &specs.Linux{ + Namespaces: []specs.LinuxNamespace{ + { + Type: "pid", + Path: "/proc/1/ns/pid", + }, }, - }, - } - testSpecs[1].Linux = &specs.Linux{ - Namespaces: []specs.LinuxNamespace{ - { - Type: "pid", - Path: "/proc/1/ns/pid", + } + testSpecs[1].Linux = &specs.Linux{ + Namespaces: []specs.LinuxNamespace{ + { + Type: "pid", + Path: "/proc/1/ns/pid", + }, }, - }, - } - testSpecs[2].Linux = &specs.Linux{ - Namespaces: []specs.LinuxNamespace{ - { - Type: "pid", - Path: "/proc/2/ns/pid", + } + testSpecs[2].Linux = &specs.Linux{ + Namespaces: []specs.LinuxNamespace{ + { + Type: "pid", + Path: "/proc/2/ns/pid", + }, }, - }, - } + } - containers, cleanup, err := startContainers(conf, testSpecs, ids) - if err != nil { - t.Fatalf("error starting containers: %v", err) - } - defer cleanup() + containers, cleanup, err := startContainers(conf, testSpecs, ids) + if err != nil { + t.Fatalf("error starting containers: %v", err) + } + defer cleanup() - // Check via ps that multiple processes are running. - expectedPL := []*control.Process{ - {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, - } - if err := waitForProcessList(containers[0], expectedPL); err != nil { - t.Errorf("failed to wait for sleep to start: %v", err) - } - if err := waitForProcessList(containers[2], expectedPL); err != nil { - t.Errorf("failed to wait for sleep to start: %v", err) - } + // Check via ps that multiple processes are running. + expectedPL := []*control.Process{ + {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, + } + if err := waitForProcessList(containers[0], expectedPL); err != nil { + t.Errorf("failed to wait for sleep to start: %v", err) + } + if err := waitForProcessList(containers[2], expectedPL); err != nil { + t.Errorf("failed to wait for sleep to start: %v", err) + } - expectedPL = []*control.Process{ - {PID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{2}}, - } - if err := waitForProcessList(containers[1], expectedPL); err != nil { - t.Errorf("failed to wait for sleep to start: %v", err) - } + expectedPL = []*control.Process{ + {PID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{2}}, + } + if err := waitForProcessList(containers[1], expectedPL); err != nil { + t.Errorf("failed to wait for sleep to start: %v", err) + } + }) } } func TestMultiContainerWait(t *testing.T) { - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { t.Fatalf("error creating root dir: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() conf := testutil.TestConfig(t) conf.RootDir = rootDir @@ -361,11 +367,11 @@ func TestMultiContainerWait(t *testing.T) { // TestExecWait ensures what we can wait containers and individual processes in the // sandbox that have already exited. func TestExecWait(t *testing.T) { - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { t.Fatalf("error creating root dir: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() conf := testutil.TestConfig(t) conf.RootDir = rootDir @@ -457,11 +463,11 @@ func TestMultiContainerMount(t *testing.T) { }) // Setup the containers. - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { t.Fatalf("error creating root dir: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() conf := testutil.TestConfig(t) conf.RootDir = rootDir @@ -484,174 +490,174 @@ func TestMultiContainerMount(t *testing.T) { // TestMultiContainerSignal checks that it is possible to signal individual // containers without killing the entire sandbox. func TestMultiContainerSignal(t *testing.T) { - for _, conf := range configs(t, all...) { - t.Logf("Running test with conf: %+v", conf) - - rootDir, err := testutil.SetupRootDir() - if err != nil { - t.Fatalf("error creating root dir: %v", err) - } - defer os.RemoveAll(rootDir) - conf.RootDir = rootDir + for name, conf := range configs(t, all...) { + t.Run(name, func(t *testing.T) { + rootDir, cleanup, err := testutil.SetupRootDir() + if err != nil { + t.Fatalf("error creating root dir: %v", err) + } + defer cleanup() + conf.RootDir = rootDir - // Setup the containers. - sleep := []string{"sleep", "100"} - specs, ids := createSpecs(sleep, sleep) - containers, cleanup, err := startContainers(conf, specs, ids) - if err != nil { - t.Fatalf("error starting containers: %v", err) - } - defer cleanup() + // Setup the containers. + sleep := []string{"sleep", "100"} + specs, ids := createSpecs(sleep, sleep) + containers, cleanup, err := startContainers(conf, specs, ids) + if err != nil { + t.Fatalf("error starting containers: %v", err) + } + defer cleanup() - // Check via ps that container 1 process is running. - expectedPL := []*control.Process{ - {PID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{2}}, - } + // Check via ps that container 1 process is running. + expectedPL := []*control.Process{ + {PID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{2}}, + } - if err := waitForProcessList(containers[1], expectedPL); err != nil { - t.Errorf("failed to wait for sleep to start: %v", err) - } + if err := waitForProcessList(containers[1], expectedPL); err != nil { + t.Errorf("failed to wait for sleep to start: %v", err) + } - // Kill process 2. - if err := containers[1].SignalContainer(syscall.SIGKILL, false); err != nil { - t.Errorf("failed to kill process 2: %v", err) - } + // Kill process 2. + if err := containers[1].SignalContainer(syscall.SIGKILL, false); err != nil { + t.Errorf("failed to kill process 2: %v", err) + } - // Make sure process 1 is still running. - expectedPL = []*control.Process{ - {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, - } - if err := waitForProcessList(containers[0], expectedPL); err != nil { - t.Errorf("failed to wait for sleep to start: %v", err) - } + // Make sure process 1 is still running. + expectedPL = []*control.Process{ + {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, + } + if err := waitForProcessList(containers[0], expectedPL); err != nil { + t.Errorf("failed to wait for sleep to start: %v", err) + } - // goferPid is reset when container is destroyed. - goferPid := containers[1].GoferPid + // goferPid is reset when container is destroyed. + goferPid := containers[1].GoferPid - // Destroy container and ensure container's gofer process has exited. - if err := containers[1].Destroy(); err != nil { - t.Errorf("failed to destroy container: %v", err) - } - _, _, err = specutils.RetryEintr(func() (uintptr, uintptr, error) { - cpid, err := syscall.Wait4(goferPid, nil, 0, nil) - return uintptr(cpid), 0, err - }) - if err != syscall.ECHILD { - t.Errorf("error waiting for gofer to exit: %v", err) - } - // Make sure process 1 is still running. - if err := waitForProcessList(containers[0], expectedPL); err != nil { - t.Errorf("failed to wait for sleep to start: %v", err) - } + // Destroy container and ensure container's gofer process has exited. + if err := containers[1].Destroy(); err != nil { + t.Errorf("failed to destroy container: %v", err) + } + _, _, err = specutils.RetryEintr(func() (uintptr, uintptr, error) { + cpid, err := syscall.Wait4(goferPid, nil, 0, nil) + return uintptr(cpid), 0, err + }) + if err != syscall.ECHILD { + t.Errorf("error waiting for gofer to exit: %v", err) + } + // Make sure process 1 is still running. + if err := waitForProcessList(containers[0], expectedPL); err != nil { + t.Errorf("failed to wait for sleep to start: %v", err) + } - // Now that process 2 is gone, ensure we get an error trying to - // signal it again. - if err := containers[1].SignalContainer(syscall.SIGKILL, false); err == nil { - t.Errorf("container %q shouldn't exist, but we were able to signal it", containers[1].ID) - } + // Now that process 2 is gone, ensure we get an error trying to + // signal it again. + if err := containers[1].SignalContainer(syscall.SIGKILL, false); err == nil { + t.Errorf("container %q shouldn't exist, but we were able to signal it", containers[1].ID) + } - // Kill process 1. - if err := containers[0].SignalContainer(syscall.SIGKILL, false); err != nil { - t.Errorf("failed to kill process 1: %v", err) - } + // Kill process 1. + if err := containers[0].SignalContainer(syscall.SIGKILL, false); err != nil { + t.Errorf("failed to kill process 1: %v", err) + } - // Ensure that container's gofer and sandbox process are no more. - err = blockUntilWaitable(containers[0].GoferPid) - if err != nil && err != syscall.ECHILD { - t.Errorf("error waiting for gofer to exit: %v", err) - } + // Ensure that container's gofer and sandbox process are no more. + err = blockUntilWaitable(containers[0].GoferPid) + if err != nil && err != syscall.ECHILD { + t.Errorf("error waiting for gofer to exit: %v", err) + } - err = blockUntilWaitable(containers[0].Sandbox.Pid) - if err != nil && err != syscall.ECHILD { - t.Errorf("error waiting for sandbox to exit: %v", err) - } + err = blockUntilWaitable(containers[0].Sandbox.Pid) + if err != nil && err != syscall.ECHILD { + t.Errorf("error waiting for sandbox to exit: %v", err) + } - // The sentry should be gone, so signaling should yield an error. - if err := containers[0].SignalContainer(syscall.SIGKILL, false); err == nil { - t.Errorf("sandbox %q shouldn't exist, but we were able to signal it", containers[0].Sandbox.ID) - } + // The sentry should be gone, so signaling should yield an error. + if err := containers[0].SignalContainer(syscall.SIGKILL, false); err == nil { + t.Errorf("sandbox %q shouldn't exist, but we were able to signal it", containers[0].Sandbox.ID) + } - if err := containers[0].Destroy(); err != nil { - t.Errorf("failed to destroy container: %v", err) - } + if err := containers[0].Destroy(); err != nil { + t.Errorf("failed to destroy container: %v", err) + } + }) } } // TestMultiContainerDestroy checks that container are properly cleaned-up when // they are destroyed. func TestMultiContainerDestroy(t *testing.T) { - app, err := testutil.FindFile("runsc/container/test_app/test_app") + app, err := testutil.FindFile("test/cmd/test_app/test_app") if err != nil { t.Fatal("error finding test_app:", err) } - for _, conf := range configs(t, all...) { - t.Logf("Running test with conf: %+v", conf) - - rootDir, err := testutil.SetupRootDir() - if err != nil { - t.Fatalf("error creating root dir: %v", err) - } - defer os.RemoveAll(rootDir) - conf.RootDir = rootDir - - // First container will remain intact while the second container is killed. - podSpecs, ids := createSpecs( - []string{"sleep", "100"}, - []string{app, "fork-bomb"}) - - // Run the fork bomb in a PID namespace to prevent processes to be - // re-parented to PID=1 in the root container. - podSpecs[1].Linux = &specs.Linux{ - Namespaces: []specs.LinuxNamespace{{Type: "pid"}}, - } - containers, cleanup, err := startContainers(conf, podSpecs, ids) - if err != nil { - t.Fatalf("error starting containers: %v", err) - } - defer cleanup() + for name, conf := range configs(t, all...) { + t.Run(name, func(t *testing.T) { + rootDir, cleanup, err := testutil.SetupRootDir() + if err != nil { + t.Fatalf("error creating root dir: %v", err) + } + defer cleanup() + conf.RootDir = rootDir + + // First container will remain intact while the second container is killed. + podSpecs, ids := createSpecs( + []string{"sleep", "100"}, + []string{app, "fork-bomb"}) + + // Run the fork bomb in a PID namespace to prevent processes to be + // re-parented to PID=1 in the root container. + podSpecs[1].Linux = &specs.Linux{ + Namespaces: []specs.LinuxNamespace{{Type: "pid"}}, + } + containers, cleanup, err := startContainers(conf, podSpecs, ids) + if err != nil { + t.Fatalf("error starting containers: %v", err) + } + defer cleanup() - // Exec more processes to ensure signal all works for exec'd processes too. - args := &control.ExecArgs{ - Filename: app, - Argv: []string{app, "fork-bomb"}, - } - if _, err := containers[1].Execute(args); err != nil { - t.Fatalf("error exec'ing: %v", err) - } + // Exec more processes to ensure signal all works for exec'd processes too. + args := &control.ExecArgs{ + Filename: app, + Argv: []string{app, "fork-bomb"}, + } + if _, err := containers[1].Execute(args); err != nil { + t.Fatalf("error exec'ing: %v", err) + } - // Let it brew... - time.Sleep(500 * time.Millisecond) + // Let it brew... + time.Sleep(500 * time.Millisecond) - if err := containers[1].Destroy(); err != nil { - t.Fatalf("error destroying container: %v", err) - } + if err := containers[1].Destroy(); err != nil { + t.Fatalf("error destroying container: %v", err) + } - // Check that destroy killed all processes belonging to the container and - // waited for them to exit before returning. - pss, err := containers[0].Sandbox.Processes("") - if err != nil { - t.Fatalf("error getting process data from sandbox: %v", err) - } - expectedPL := []*control.Process{{PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}} - if r, err := procListsEqual(pss, expectedPL); !r { - t.Errorf("container got process list: %s, want: %s: error: %v", - procListToString(pss), procListToString(expectedPL), err) - } + // Check that destroy killed all processes belonging to the container and + // waited for them to exit before returning. + pss, err := containers[0].Sandbox.Processes("") + if err != nil { + t.Fatalf("error getting process data from sandbox: %v", err) + } + expectedPL := []*control.Process{{PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}} + if r, err := procListsEqual(pss, expectedPL); !r { + t.Errorf("container got process list: %s, want: %s: error: %v", + procListToString(pss), procListToString(expectedPL), err) + } - // Check that cont.Destroy is safe to call multiple times. - if err := containers[1].Destroy(); err != nil { - t.Errorf("error destroying container: %v", err) - } + // Check that cont.Destroy is safe to call multiple times. + if err := containers[1].Destroy(); err != nil { + t.Errorf("error destroying container: %v", err) + } + }) } } func TestMultiContainerProcesses(t *testing.T) { - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { t.Fatalf("error creating root dir: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() conf := testutil.TestConfig(t) conf.RootDir = rootDir @@ -706,11 +712,11 @@ func TestMultiContainerProcesses(t *testing.T) { // TestMultiContainerKillAll checks that all process that belong to a container // are killed when SIGKILL is sent to *all* processes in that container. func TestMultiContainerKillAll(t *testing.T) { - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { t.Fatalf("error creating root dir: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() conf := testutil.TestConfig(t) conf.RootDir = rootDir @@ -721,7 +727,7 @@ func TestMultiContainerKillAll(t *testing.T) { {killContainer: true}, {killContainer: false}, } { - app, err := testutil.FindFile("runsc/container/test_app/test_app") + app, err := testutil.FindFile("test/cmd/test_app/test_app") if err != nil { t.Fatal("error finding test_app:", err) } @@ -739,11 +745,11 @@ func TestMultiContainerKillAll(t *testing.T) { // Wait until all processes are created. rootProcCount := int(math.Pow(2, 3) - 1) if err := waitForProcessCount(containers[0], rootProcCount); err != nil { - t.Fatal(err) + t.Fatalf("error waitting for processes: %v", err) } procCount := int(math.Pow(2, 5) - 1) if err := waitForProcessCount(containers[1], procCount); err != nil { - t.Fatal(err) + t.Fatalf("error waiting for processes: %v", err) } // Exec more processes to ensure signal works for exec'd processes too. @@ -757,7 +763,7 @@ func TestMultiContainerKillAll(t *testing.T) { // Wait for these new processes to start. procCount += int(math.Pow(2, 3) - 1) if err := waitForProcessCount(containers[1], procCount); err != nil { - t.Fatal(err) + t.Fatalf("error waiting for processes: %v", err) } if tc.killContainer { @@ -790,11 +796,11 @@ func TestMultiContainerKillAll(t *testing.T) { // Check that all processes are gone. if err := waitForProcessCount(containers[1], 0); err != nil { - t.Fatal(err) + t.Fatalf("error waiting for processes: %v", err) } // Check that root container was not affected. if err := waitForProcessCount(containers[0], rootProcCount); err != nil { - t.Fatal(err) + t.Fatalf("error waiting for processes: %v", err) } } } @@ -805,17 +811,16 @@ func TestMultiContainerDestroyNotStarted(t *testing.T) { []string{"/bin/sleep", "100"}) conf := testutil.TestConfig(t) - rootDir, rootBundleDir, err := testutil.SetupContainer(specs[0], conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(specs[0], conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(rootBundleDir) + defer cleanup() rootArgs := Args{ ID: ids[0], Spec: specs[0], - BundleDir: rootBundleDir, + BundleDir: bundleDir, } root, err := New(conf, rootArgs) if err != nil { @@ -827,11 +832,11 @@ func TestMultiContainerDestroyNotStarted(t *testing.T) { } // Create and destroy sub-container. - bundleDir, err := testutil.SetupBundleDir(specs[1]) + bundleDir, cleanupSub, err := testutil.SetupBundleDir(specs[1]) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(bundleDir) + defer cleanupSub() args := Args{ ID: ids[1], @@ -859,17 +864,16 @@ func TestMultiContainerDestroyStarting(t *testing.T) { specs, ids := createSpecs(cmds...) conf := testutil.TestConfig(t) - rootDir, rootBundleDir, err := testutil.SetupContainer(specs[0], conf) + rootDir, bundleDir, cleanup, err := testutil.SetupContainer(specs[0], conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(rootBundleDir) + defer cleanup() rootArgs := Args{ ID: ids[0], Spec: specs[0], - BundleDir: rootBundleDir, + BundleDir: bundleDir, } root, err := New(conf, rootArgs) if err != nil { @@ -886,16 +890,16 @@ func TestMultiContainerDestroyStarting(t *testing.T) { continue // skip root container } - bundleDir, err := testutil.SetupBundleDir(specs[i]) + bundleDir, cleanup, err := testutil.SetupBundleDir(specs[i]) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(bundleDir) + defer cleanup() rootArgs := Args{ ID: ids[i], Spec: specs[i], - BundleDir: rootBundleDir, + BundleDir: bundleDir, } cont, err := New(conf, rootArgs) if err != nil { @@ -937,11 +941,11 @@ func TestMultiContainerDifferentFilesystems(t *testing.T) { script := fmt.Sprintf("if [ -f %q ]; then exit 1; else touch %q; fi", filename, filename) cmd := []string{"sh", "-c", script} - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { t.Fatalf("error creating root dir: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() conf := testutil.TestConfig(t) conf.RootDir = rootDir @@ -977,7 +981,7 @@ func TestMultiContainerDifferentFilesystems(t *testing.T) { // TestMultiContainerContainerDestroyStress tests that IO operations continue // to work after containers have been stopped and gofers killed. func TestMultiContainerContainerDestroyStress(t *testing.T) { - app, err := testutil.FindFile("runsc/container/test_app/test_app") + app, err := testutil.FindFile("test/cmd/test_app/test_app") if err != nil { t.Fatal("error finding test_app:", err) } @@ -1007,12 +1011,11 @@ func TestMultiContainerContainerDestroyStress(t *testing.T) { childrenIDs := allIDs[1:] conf := testutil.TestConfig(t) - rootDir, bundleDir, err := testutil.SetupContainer(rootSpec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(rootSpec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() // Start root container. rootArgs := Args{ @@ -1038,11 +1041,11 @@ func TestMultiContainerContainerDestroyStress(t *testing.T) { var children []*Container for j, spec := range specs { - bundleDir, err := testutil.SetupBundleDir(spec) + bundleDir, cleanup, err := testutil.SetupBundleDir(spec) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(bundleDir) + defer cleanup() args := Args{ ID: ids[j], @@ -1080,306 +1083,306 @@ func TestMultiContainerContainerDestroyStress(t *testing.T) { // Test that pod shared mounts are properly mounted in 2 containers and that // changes from one container is reflected in the other. func TestMultiContainerSharedMount(t *testing.T) { - for _, conf := range configs(t, all...) { - t.Logf("Running test with conf: %+v", conf) - - rootDir, err := testutil.SetupRootDir() - if err != nil { - t.Fatalf("error creating root dir: %v", err) - } - defer os.RemoveAll(rootDir) - conf.RootDir = rootDir - - // Setup the containers. - sleep := []string{"sleep", "100"} - podSpec, ids := createSpecs(sleep, sleep) - mnt0 := specs.Mount{ - Destination: "/mydir/test", - Source: "/some/dir", - Type: "tmpfs", - Options: nil, - } - podSpec[0].Mounts = append(podSpec[0].Mounts, mnt0) + for name, conf := range configs(t, all...) { + t.Run(name, func(t *testing.T) { + rootDir, cleanup, err := testutil.SetupRootDir() + if err != nil { + t.Fatalf("error creating root dir: %v", err) + } + defer cleanup() + conf.RootDir = rootDir + + // Setup the containers. + sleep := []string{"sleep", "100"} + podSpec, ids := createSpecs(sleep, sleep) + mnt0 := specs.Mount{ + Destination: "/mydir/test", + Source: "/some/dir", + Type: "tmpfs", + Options: nil, + } + podSpec[0].Mounts = append(podSpec[0].Mounts, mnt0) - mnt1 := mnt0 - mnt1.Destination = "/mydir2/test2" - podSpec[1].Mounts = append(podSpec[1].Mounts, mnt1) + mnt1 := mnt0 + mnt1.Destination = "/mydir2/test2" + podSpec[1].Mounts = append(podSpec[1].Mounts, mnt1) - createSharedMount(mnt0, "test-mount", podSpec...) + createSharedMount(mnt0, "test-mount", podSpec...) - containers, cleanup, err := startContainers(conf, podSpec, ids) - if err != nil { - t.Fatalf("error starting containers: %v", err) - } - defer cleanup() + containers, cleanup, err := startContainers(conf, podSpec, ids) + if err != nil { + t.Fatalf("error starting containers: %v", err) + } + defer cleanup() - file0 := path.Join(mnt0.Destination, "abc") - file1 := path.Join(mnt1.Destination, "abc") - execs := []execDesc{ - { - c: containers[0], - cmd: []string{"/usr/bin/test", "-d", mnt0.Destination}, - desc: "directory is mounted in container0", - }, - { - c: containers[1], - cmd: []string{"/usr/bin/test", "-d", mnt1.Destination}, - desc: "directory is mounted in container1", - }, - { - c: containers[0], - cmd: []string{"/usr/bin/touch", file0}, - desc: "create file in container0", - }, - { - c: containers[0], - cmd: []string{"/usr/bin/test", "-f", file0}, - desc: "file appears in container0", - }, - { - c: containers[1], - cmd: []string{"/usr/bin/test", "-f", file1}, - desc: "file appears in container1", - }, - { - c: containers[1], - cmd: []string{"/bin/rm", file1}, - desc: "file removed from container1", - }, - { - c: containers[0], - cmd: []string{"/usr/bin/test", "!", "-f", file0}, - desc: "file removed from container0", - }, - { - c: containers[1], - cmd: []string{"/usr/bin/test", "!", "-f", file1}, - desc: "file removed from container1", - }, - { - c: containers[1], - cmd: []string{"/bin/mkdir", file1}, - desc: "create directory in container1", - }, - { - c: containers[0], - cmd: []string{"/usr/bin/test", "-d", file0}, - desc: "dir appears in container0", - }, - { - c: containers[1], - cmd: []string{"/usr/bin/test", "-d", file1}, - desc: "dir appears in container1", - }, - { - c: containers[0], - cmd: []string{"/bin/rmdir", file0}, - desc: "create directory in container0", - }, - { - c: containers[0], - cmd: []string{"/usr/bin/test", "!", "-d", file0}, - desc: "dir removed from container0", - }, - { - c: containers[1], - cmd: []string{"/usr/bin/test", "!", "-d", file1}, - desc: "dir removed from container1", - }, - } - if err := execMany(execs); err != nil { - t.Fatal(err.Error()) - } + file0 := path.Join(mnt0.Destination, "abc") + file1 := path.Join(mnt1.Destination, "abc") + execs := []execDesc{ + { + c: containers[0], + cmd: []string{"/usr/bin/test", "-d", mnt0.Destination}, + desc: "directory is mounted in container0", + }, + { + c: containers[1], + cmd: []string{"/usr/bin/test", "-d", mnt1.Destination}, + desc: "directory is mounted in container1", + }, + { + c: containers[0], + cmd: []string{"/usr/bin/touch", file0}, + desc: "create file in container0", + }, + { + c: containers[0], + cmd: []string{"/usr/bin/test", "-f", file0}, + desc: "file appears in container0", + }, + { + c: containers[1], + cmd: []string{"/usr/bin/test", "-f", file1}, + desc: "file appears in container1", + }, + { + c: containers[1], + cmd: []string{"/bin/rm", file1}, + desc: "file removed from container1", + }, + { + c: containers[0], + cmd: []string{"/usr/bin/test", "!", "-f", file0}, + desc: "file removed from container0", + }, + { + c: containers[1], + cmd: []string{"/usr/bin/test", "!", "-f", file1}, + desc: "file removed from container1", + }, + { + c: containers[1], + cmd: []string{"/bin/mkdir", file1}, + desc: "create directory in container1", + }, + { + c: containers[0], + cmd: []string{"/usr/bin/test", "-d", file0}, + desc: "dir appears in container0", + }, + { + c: containers[1], + cmd: []string{"/usr/bin/test", "-d", file1}, + desc: "dir appears in container1", + }, + { + c: containers[0], + cmd: []string{"/bin/rmdir", file0}, + desc: "create directory in container0", + }, + { + c: containers[0], + cmd: []string{"/usr/bin/test", "!", "-d", file0}, + desc: "dir removed from container0", + }, + { + c: containers[1], + cmd: []string{"/usr/bin/test", "!", "-d", file1}, + desc: "dir removed from container1", + }, + } + if err := execMany(execs); err != nil { + t.Fatal(err.Error()) + } + }) } } // Test that pod mounts are mounted as readonly when requested. func TestMultiContainerSharedMountReadonly(t *testing.T) { - for _, conf := range configs(t, all...) { - t.Logf("Running test with conf: %+v", conf) - - rootDir, err := testutil.SetupRootDir() - if err != nil { - t.Fatalf("error creating root dir: %v", err) - } - defer os.RemoveAll(rootDir) - conf.RootDir = rootDir - - // Setup the containers. - sleep := []string{"sleep", "100"} - podSpec, ids := createSpecs(sleep, sleep) - mnt0 := specs.Mount{ - Destination: "/mydir/test", - Source: "/some/dir", - Type: "tmpfs", - Options: []string{"ro"}, - } - podSpec[0].Mounts = append(podSpec[0].Mounts, mnt0) + for name, conf := range configs(t, all...) { + t.Run(name, func(t *testing.T) { + rootDir, cleanup, err := testutil.SetupRootDir() + if err != nil { + t.Fatalf("error creating root dir: %v", err) + } + defer cleanup() + conf.RootDir = rootDir + + // Setup the containers. + sleep := []string{"sleep", "100"} + podSpec, ids := createSpecs(sleep, sleep) + mnt0 := specs.Mount{ + Destination: "/mydir/test", + Source: "/some/dir", + Type: "tmpfs", + Options: []string{"ro"}, + } + podSpec[0].Mounts = append(podSpec[0].Mounts, mnt0) - mnt1 := mnt0 - mnt1.Destination = "/mydir2/test2" - podSpec[1].Mounts = append(podSpec[1].Mounts, mnt1) + mnt1 := mnt0 + mnt1.Destination = "/mydir2/test2" + podSpec[1].Mounts = append(podSpec[1].Mounts, mnt1) - createSharedMount(mnt0, "test-mount", podSpec...) + createSharedMount(mnt0, "test-mount", podSpec...) - containers, cleanup, err := startContainers(conf, podSpec, ids) - if err != nil { - t.Fatalf("error starting containers: %v", err) - } - defer cleanup() + containers, cleanup, err := startContainers(conf, podSpec, ids) + if err != nil { + t.Fatalf("error starting containers: %v", err) + } + defer cleanup() - file0 := path.Join(mnt0.Destination, "abc") - file1 := path.Join(mnt1.Destination, "abc") - execs := []execDesc{ - { - c: containers[0], - cmd: []string{"/usr/bin/test", "-d", mnt0.Destination}, - desc: "directory is mounted in container0", - }, - { - c: containers[1], - cmd: []string{"/usr/bin/test", "-d", mnt1.Destination}, - desc: "directory is mounted in container1", - }, - { - c: containers[0], - cmd: []string{"/usr/bin/touch", file0}, - want: 1, - desc: "fails to write to container0", - }, - { - c: containers[1], - cmd: []string{"/usr/bin/touch", file1}, - want: 1, - desc: "fails to write to container1", - }, - } - if err := execMany(execs); err != nil { - t.Fatal(err.Error()) - } + file0 := path.Join(mnt0.Destination, "abc") + file1 := path.Join(mnt1.Destination, "abc") + execs := []execDesc{ + { + c: containers[0], + cmd: []string{"/usr/bin/test", "-d", mnt0.Destination}, + desc: "directory is mounted in container0", + }, + { + c: containers[1], + cmd: []string{"/usr/bin/test", "-d", mnt1.Destination}, + desc: "directory is mounted in container1", + }, + { + c: containers[0], + cmd: []string{"/usr/bin/touch", file0}, + want: 1, + desc: "fails to write to container0", + }, + { + c: containers[1], + cmd: []string{"/usr/bin/touch", file1}, + want: 1, + desc: "fails to write to container1", + }, + } + if err := execMany(execs); err != nil { + t.Fatal(err.Error()) + } + }) } } // Test that shared pod mounts continue to work after container is restarted. func TestMultiContainerSharedMountRestart(t *testing.T) { - for _, conf := range configs(t, all...) { - t.Logf("Running test with conf: %+v", conf) - - rootDir, err := testutil.SetupRootDir() - if err != nil { - t.Fatalf("error creating root dir: %v", err) - } - defer os.RemoveAll(rootDir) - conf.RootDir = rootDir - - // Setup the containers. - sleep := []string{"sleep", "100"} - podSpec, ids := createSpecs(sleep, sleep) - mnt0 := specs.Mount{ - Destination: "/mydir/test", - Source: "/some/dir", - Type: "tmpfs", - Options: nil, - } - podSpec[0].Mounts = append(podSpec[0].Mounts, mnt0) + for name, conf := range configs(t, all...) { + t.Run(name, func(t *testing.T) { + rootDir, cleanup, err := testutil.SetupRootDir() + if err != nil { + t.Fatalf("error creating root dir: %v", err) + } + defer cleanup() + conf.RootDir = rootDir + + // Setup the containers. + sleep := []string{"sleep", "100"} + podSpec, ids := createSpecs(sleep, sleep) + mnt0 := specs.Mount{ + Destination: "/mydir/test", + Source: "/some/dir", + Type: "tmpfs", + Options: nil, + } + podSpec[0].Mounts = append(podSpec[0].Mounts, mnt0) - mnt1 := mnt0 - mnt1.Destination = "/mydir2/test2" - podSpec[1].Mounts = append(podSpec[1].Mounts, mnt1) + mnt1 := mnt0 + mnt1.Destination = "/mydir2/test2" + podSpec[1].Mounts = append(podSpec[1].Mounts, mnt1) - createSharedMount(mnt0, "test-mount", podSpec...) + createSharedMount(mnt0, "test-mount", podSpec...) - containers, cleanup, err := startContainers(conf, podSpec, ids) - if err != nil { - t.Fatalf("error starting containers: %v", err) - } - defer cleanup() + containers, cleanup, err := startContainers(conf, podSpec, ids) + if err != nil { + t.Fatalf("error starting containers: %v", err) + } + defer cleanup() - file0 := path.Join(mnt0.Destination, "abc") - file1 := path.Join(mnt1.Destination, "abc") - execs := []execDesc{ - { - c: containers[0], - cmd: []string{"/usr/bin/touch", file0}, - desc: "create file in container0", - }, - { - c: containers[0], - cmd: []string{"/usr/bin/test", "-f", file0}, - desc: "file appears in container0", - }, - { - c: containers[1], - cmd: []string{"/usr/bin/test", "-f", file1}, - desc: "file appears in container1", - }, - } - if err := execMany(execs); err != nil { - t.Fatal(err.Error()) - } + file0 := path.Join(mnt0.Destination, "abc") + file1 := path.Join(mnt1.Destination, "abc") + execs := []execDesc{ + { + c: containers[0], + cmd: []string{"/usr/bin/touch", file0}, + desc: "create file in container0", + }, + { + c: containers[0], + cmd: []string{"/usr/bin/test", "-f", file0}, + desc: "file appears in container0", + }, + { + c: containers[1], + cmd: []string{"/usr/bin/test", "-f", file1}, + desc: "file appears in container1", + }, + } + if err := execMany(execs); err != nil { + t.Fatal(err.Error()) + } - containers[1].Destroy() + containers[1].Destroy() - bundleDir, err := testutil.SetupBundleDir(podSpec[1]) - if err != nil { - t.Fatalf("error restarting container: %v", err) - } - defer os.RemoveAll(bundleDir) + bundleDir, cleanup, err := testutil.SetupBundleDir(podSpec[1]) + if err != nil { + t.Fatalf("error restarting container: %v", err) + } + defer cleanup() - args := Args{ - ID: ids[1], - Spec: podSpec[1], - BundleDir: bundleDir, - } - containers[1], err = New(conf, args) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - if err := containers[1].Start(conf); err != nil { - t.Fatalf("error starting container: %v", err) - } + args := Args{ + ID: ids[1], + Spec: podSpec[1], + BundleDir: bundleDir, + } + containers[1], err = New(conf, args) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + if err := containers[1].Start(conf); err != nil { + t.Fatalf("error starting container: %v", err) + } - execs = []execDesc{ - { - c: containers[0], - cmd: []string{"/usr/bin/test", "-f", file0}, - desc: "file is still in container0", - }, - { - c: containers[1], - cmd: []string{"/usr/bin/test", "-f", file1}, - desc: "file is still in container1", - }, - { - c: containers[1], - cmd: []string{"/bin/rm", file1}, - desc: "file removed from container1", - }, - { - c: containers[0], - cmd: []string{"/usr/bin/test", "!", "-f", file0}, - desc: "file removed from container0", - }, - { - c: containers[1], - cmd: []string{"/usr/bin/test", "!", "-f", file1}, - desc: "file removed from container1", - }, - } - if err := execMany(execs); err != nil { - t.Fatal(err.Error()) - } + execs = []execDesc{ + { + c: containers[0], + cmd: []string{"/usr/bin/test", "-f", file0}, + desc: "file is still in container0", + }, + { + c: containers[1], + cmd: []string{"/usr/bin/test", "-f", file1}, + desc: "file is still in container1", + }, + { + c: containers[1], + cmd: []string{"/bin/rm", file1}, + desc: "file removed from container1", + }, + { + c: containers[0], + cmd: []string{"/usr/bin/test", "!", "-f", file0}, + desc: "file removed from container0", + }, + { + c: containers[1], + cmd: []string{"/usr/bin/test", "!", "-f", file1}, + desc: "file removed from container1", + }, + } + if err := execMany(execs); err != nil { + t.Fatal(err.Error()) + } + }) } } // Test that unsupported pod mounts options are ignored when matching master and // slave mounts. func TestMultiContainerSharedMountUnsupportedOptions(t *testing.T) { - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { t.Fatalf("error creating root dir: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() conf := testutil.TestConfig(t) conf.RootDir = rootDir @@ -1428,7 +1431,7 @@ func TestMultiContainerSharedMountUnsupportedOptions(t *testing.T) { // Test that one container can send an FD to another container, even though // they have distinct MountNamespaces. func TestMultiContainerMultiRootCanHandleFDs(t *testing.T) { - app, err := testutil.FindFile("runsc/container/test_app/test_app") + app, err := testutil.FindFile("test/cmd/test_app/test_app") if err != nil { t.Fatal("error finding test_app:", err) } @@ -1457,11 +1460,11 @@ func TestMultiContainerMultiRootCanHandleFDs(t *testing.T) { Type: "tmpfs", } - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { t.Fatalf("error creating root dir: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() conf := testutil.TestConfig(t) conf.RootDir = rootDir @@ -1494,11 +1497,11 @@ func TestMultiContainerMultiRootCanHandleFDs(t *testing.T) { // Test that container is destroyed when Gofer is killed. func TestMultiContainerGoferKilled(t *testing.T) { - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { t.Fatalf("error creating root dir: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() conf := testutil.TestConfig(t) conf.RootDir = rootDir @@ -1581,11 +1584,11 @@ func TestMultiContainerLoadSandbox(t *testing.T) { sleep := []string{"sleep", "100"} specs, ids := createSpecs(sleep, sleep, sleep) - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { t.Fatalf("error creating root dir: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() conf := testutil.TestConfig(t) conf.RootDir = rootDir @@ -1614,7 +1617,7 @@ func TestMultiContainerLoadSandbox(t *testing.T) { } // Create a valid but empty container directory. - randomCID := testutil.UniqueContainerID() + randomCID := testutil.RandomContainerID() dir = filepath.Join(conf.RootDir, randomCID) if err := os.MkdirAll(dir, 0755); err != nil { t.Fatalf("os.MkdirAll(%q)=%v", dir, err) @@ -1681,11 +1684,11 @@ func TestMultiContainerRunNonRoot(t *testing.T) { Type: "bind", }) - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { t.Fatalf("error creating root dir: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() conf := testutil.TestConfig(t) conf.RootDir = rootDir diff --git a/runsc/container/shared_volume_test.go b/runsc/container/shared_volume_test.go index f80852414..bac177a88 100644 --- a/runsc/container/shared_volume_test.go +++ b/runsc/container/shared_volume_test.go @@ -24,8 +24,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/control" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/test/testutil" "gvisor.dev/gvisor/runsc/boot" - "gvisor.dev/gvisor/runsc/testutil" ) // TestSharedVolume checks that modifications to a volume mount are propagated @@ -33,7 +33,6 @@ import ( func TestSharedVolume(t *testing.T) { conf := testutil.TestConfig(t) conf.FileAccess = boot.FileAccessShared - t.Logf("Running test with conf: %+v", conf) // Main process just sleeps. We will use "exec" to probe the state of // the filesystem. @@ -44,16 +43,15 @@ func TestSharedVolume(t *testing.T) { t.Fatalf("TempDir failed: %v", err) } - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() // Create and start the container. args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, } @@ -192,7 +190,6 @@ func checkFile(c *Container, filename string, want []byte) error { func TestSharedVolumeFile(t *testing.T) { conf := testutil.TestConfig(t) conf.FileAccess = boot.FileAccessShared - t.Logf("Running test with conf: %+v", conf) // Main process just sleeps. We will use "exec" to probe the state of // the filesystem. @@ -203,16 +200,15 @@ func TestSharedVolumeFile(t *testing.T) { t.Fatalf("TempDir failed: %v", err) } - rootDir, bundleDir, err := testutil.SetupContainer(spec, conf) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { t.Fatalf("error setting up container: %v", err) } - defer os.RemoveAll(rootDir) - defer os.RemoveAll(bundleDir) + defer cleanup() // Create and start the container. args := Args{ - ID: testutil.UniqueContainerID(), + ID: testutil.RandomContainerID(), Spec: spec, BundleDir: bundleDir, } diff --git a/runsc/container/test_app/BUILD b/runsc/container/test_app/BUILD deleted file mode 100644 index 0defbd9fc..000000000 --- a/runsc/container/test_app/BUILD +++ /dev/null @@ -1,21 +0,0 @@ -load("//tools:defs.bzl", "go_binary") - -package(licenses = ["notice"]) - -go_binary( - name = "test_app", - testonly = 1, - srcs = [ - "fds.go", - "test_app.go", - ], - pure = True, - visibility = ["//runsc/container:__pkg__"], - deps = [ - "//pkg/unet", - "//runsc/flag", - "//runsc/testutil", - "@com_github_google_subcommands//:go_default_library", - "@com_github_kr_pty//:go_default_library", - ], -) diff --git a/runsc/container/test_app/fds.go b/runsc/container/test_app/fds.go deleted file mode 100644 index 2a146a2c3..000000000 --- a/runsc/container/test_app/fds.go +++ /dev/null @@ -1,185 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package main - -import ( - "context" - "io/ioutil" - "log" - "os" - "time" - - "github.com/google/subcommands" - "gvisor.dev/gvisor/pkg/unet" - "gvisor.dev/gvisor/runsc/flag" - "gvisor.dev/gvisor/runsc/testutil" -) - -const fileContents = "foobarbaz" - -// fdSender will open a file and send the FD over a unix domain socket. -type fdSender struct { - socketPath string -} - -// Name implements subcommands.Command.Name. -func (*fdSender) Name() string { - return "fd_sender" -} - -// Synopsis implements subcommands.Command.Synopsys. -func (*fdSender) Synopsis() string { - return "creates a file and sends the FD over the socket" -} - -// Usage implements subcommands.Command.Usage. -func (*fdSender) Usage() string { - return "fd_sender " -} - -// SetFlags implements subcommands.Command.SetFlags. -func (fds *fdSender) SetFlags(f *flag.FlagSet) { - f.StringVar(&fds.socketPath, "socket", "", "path to socket") -} - -// Execute implements subcommands.Command.Execute. -func (fds *fdSender) Execute(ctx context.Context, f *flag.FlagSet, args ...interface{}) subcommands.ExitStatus { - if fds.socketPath == "" { - log.Fatalf("socket flag must be set") - } - - dir, err := ioutil.TempDir("", "") - if err != nil { - log.Fatalf("TempDir failed: %v", err) - } - - fileToSend, err := ioutil.TempFile(dir, "") - if err != nil { - log.Fatalf("TempFile failed: %v", err) - } - defer fileToSend.Close() - - if _, err := fileToSend.WriteString(fileContents); err != nil { - log.Fatalf("Write(%q) failed: %v", fileContents, err) - } - - // Receiver may not be started yet, so try connecting in a poll loop. - var s *unet.Socket - if err := testutil.Poll(func() error { - var err error - s, err = unet.Connect(fds.socketPath, true /* SEQPACKET, so we can send empty message with FD */) - return err - }, 10*time.Second); err != nil { - log.Fatalf("Error connecting to socket %q: %v", fds.socketPath, err) - } - defer s.Close() - - w := s.Writer(true) - w.ControlMessage.PackFDs(int(fileToSend.Fd())) - if _, err := w.WriteVec([][]byte{[]byte{'a'}}); err != nil { - log.Fatalf("Error sending FD %q over socket %q: %v", fileToSend.Fd(), fds.socketPath, err) - } - - log.Print("FD SENDER exiting successfully") - return subcommands.ExitSuccess -} - -// fdReceiver receives an FD from a unix domain socket and does things to it. -type fdReceiver struct { - socketPath string -} - -// Name implements subcommands.Command.Name. -func (*fdReceiver) Name() string { - return "fd_receiver" -} - -// Synopsis implements subcommands.Command.Synopsys. -func (*fdReceiver) Synopsis() string { - return "reads an FD from a unix socket, and then does things to it" -} - -// Usage implements subcommands.Command.Usage. -func (*fdReceiver) Usage() string { - return "fd_receiver " -} - -// SetFlags implements subcommands.Command.SetFlags. -func (fdr *fdReceiver) SetFlags(f *flag.FlagSet) { - f.StringVar(&fdr.socketPath, "socket", "", "path to socket") -} - -// Execute implements subcommands.Command.Execute. -func (fdr *fdReceiver) Execute(ctx context.Context, f *flag.FlagSet, args ...interface{}) subcommands.ExitStatus { - if fdr.socketPath == "" { - log.Fatalf("Flags cannot be empty, given: socket: %q", fdr.socketPath) - } - - ss, err := unet.BindAndListen(fdr.socketPath, true /* packet */) - if err != nil { - log.Fatalf("BindAndListen(%q) failed: %v", fdr.socketPath, err) - } - defer ss.Close() - - var s *unet.Socket - c := make(chan error, 1) - go func() { - var err error - s, err = ss.Accept() - c <- err - }() - - select { - case err := <-c: - if err != nil { - log.Fatalf("Accept() failed: %v", err) - } - case <-time.After(10 * time.Second): - log.Fatalf("Timeout waiting for accept") - } - - r := s.Reader(true) - r.EnableFDs(1) - b := [][]byte{{'a'}} - if n, err := r.ReadVec(b); n != 1 || err != nil { - log.Fatalf("ReadVec got n=%d err %v (wanted 0, nil)", n, err) - } - - fds, err := r.ExtractFDs() - if err != nil { - log.Fatalf("ExtractFD() got err %v", err) - } - if len(fds) != 1 { - log.Fatalf("ExtractFD() got %d FDs, wanted 1", len(fds)) - } - fd := fds[0] - - file := os.NewFile(uintptr(fd), "received file") - defer file.Close() - if _, err := file.Seek(0, os.SEEK_SET); err != nil { - log.Fatalf("Seek(0, 0) failed: %v", err) - } - - got, err := ioutil.ReadAll(file) - if err != nil { - log.Fatalf("ReadAll failed: %v", err) - } - if string(got) != fileContents { - log.Fatalf("ReadAll got %q want %q", string(got), fileContents) - } - - log.Print("FD RECEIVER exiting successfully") - return subcommands.ExitSuccess -} diff --git a/runsc/container/test_app/test_app.go b/runsc/container/test_app/test_app.go deleted file mode 100644 index 5f1c4b7d6..000000000 --- a/runsc/container/test_app/test_app.go +++ /dev/null @@ -1,394 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Binary test_app is like a swiss knife for tests that need to run anything -// inside the sandbox. New functionality can be added with new commands. -package main - -import ( - "context" - "fmt" - "io" - "io/ioutil" - "log" - "net" - "os" - "os/exec" - "regexp" - "strconv" - sys "syscall" - "time" - - "github.com/google/subcommands" - "github.com/kr/pty" - "gvisor.dev/gvisor/runsc/flag" - "gvisor.dev/gvisor/runsc/testutil" -) - -func main() { - subcommands.Register(subcommands.HelpCommand(), "") - subcommands.Register(subcommands.FlagsCommand(), "") - subcommands.Register(new(capability), "") - subcommands.Register(new(fdReceiver), "") - subcommands.Register(new(fdSender), "") - subcommands.Register(new(forkBomb), "") - subcommands.Register(new(ptyRunner), "") - subcommands.Register(new(reaper), "") - subcommands.Register(new(syscall), "") - subcommands.Register(new(taskTree), "") - subcommands.Register(new(uds), "") - - flag.Parse() - - exitCode := subcommands.Execute(context.Background()) - os.Exit(int(exitCode)) -} - -type uds struct { - fileName string - socketPath string -} - -// Name implements subcommands.Command.Name. -func (*uds) Name() string { - return "uds" -} - -// Synopsis implements subcommands.Command.Synopsys. -func (*uds) Synopsis() string { - return "creates unix domain socket client and server. Client sends a contant flow of sequential numbers. Server prints them to --file" -} - -// Usage implements subcommands.Command.Usage. -func (*uds) Usage() string { - return "uds " -} - -// SetFlags implements subcommands.Command.SetFlags. -func (c *uds) SetFlags(f *flag.FlagSet) { - f.StringVar(&c.fileName, "file", "", "name of output file") - f.StringVar(&c.socketPath, "socket", "", "path to socket") -} - -// Execute implements subcommands.Command.Execute. -func (c *uds) Execute(ctx context.Context, f *flag.FlagSet, args ...interface{}) subcommands.ExitStatus { - if c.fileName == "" || c.socketPath == "" { - log.Fatalf("Flags cannot be empty, given: fileName: %q, socketPath: %q", c.fileName, c.socketPath) - return subcommands.ExitFailure - } - outputFile, err := os.OpenFile(c.fileName, os.O_WRONLY|os.O_CREATE, 0666) - if err != nil { - log.Fatal("error opening output file:", err) - } - - defer os.Remove(c.socketPath) - - listener, err := net.Listen("unix", c.socketPath) - if err != nil { - log.Fatalf("error listening on socket %q: %v", c.socketPath, err) - } - - go server(listener, outputFile) - for i := 0; ; i++ { - conn, err := net.Dial("unix", c.socketPath) - if err != nil { - log.Fatal("error dialing:", err) - } - if _, err := conn.Write([]byte(strconv.Itoa(i))); err != nil { - log.Fatal("error writing:", err) - } - conn.Close() - time.Sleep(100 * time.Millisecond) - } -} - -func server(listener net.Listener, out *os.File) { - buf := make([]byte, 16) - - for { - c, err := listener.Accept() - if err != nil { - log.Fatal("error accepting connection:", err) - } - nr, err := c.Read(buf) - if err != nil { - log.Fatal("error reading from buf:", err) - } - data := buf[0:nr] - fmt.Fprint(out, string(data)+"\n") - } -} - -type taskTree struct { - depth int - width int - pause bool -} - -// Name implements subcommands.Command. -func (*taskTree) Name() string { - return "task-tree" -} - -// Synopsis implements subcommands.Command. -func (*taskTree) Synopsis() string { - return "creates a tree of tasks" -} - -// Usage implements subcommands.Command. -func (*taskTree) Usage() string { - return "task-tree " -} - -// SetFlags implements subcommands.Command. -func (c *taskTree) SetFlags(f *flag.FlagSet) { - f.IntVar(&c.depth, "depth", 1, "number of levels to create") - f.IntVar(&c.width, "width", 1, "number of tasks at each level") - f.BoolVar(&c.pause, "pause", false, "whether the tasks should pause perpetually") -} - -// Execute implements subcommands.Command. -func (c *taskTree) Execute(ctx context.Context, f *flag.FlagSet, args ...interface{}) subcommands.ExitStatus { - stop := testutil.StartReaper() - defer stop() - - if c.depth == 0 { - log.Printf("Child sleeping, PID: %d\n", os.Getpid()) - select {} - } - log.Printf("Parent %d sleeping, PID: %d\n", c.depth, os.Getpid()) - - var cmds []*exec.Cmd - for i := 0; i < c.width; i++ { - cmd := exec.Command( - "/proc/self/exe", c.Name(), - "--depth", strconv.Itoa(c.depth-1), - "--width", strconv.Itoa(c.width), - "--pause", strconv.FormatBool(c.pause)) - cmd.Stdout = os.Stdout - cmd.Stderr = os.Stderr - - if err := cmd.Start(); err != nil { - log.Fatal("failed to call self:", err) - } - cmds = append(cmds, cmd) - } - - for _, c := range cmds { - c.Wait() - } - - if c.pause { - select {} - } - - return subcommands.ExitSuccess -} - -type forkBomb struct { - delay time.Duration -} - -// Name implements subcommands.Command. -func (*forkBomb) Name() string { - return "fork-bomb" -} - -// Synopsis implements subcommands.Command. -func (*forkBomb) Synopsis() string { - return "creates child process until the end of times" -} - -// Usage implements subcommands.Command. -func (*forkBomb) Usage() string { - return "fork-bomb " -} - -// SetFlags implements subcommands.Command. -func (c *forkBomb) SetFlags(f *flag.FlagSet) { - f.DurationVar(&c.delay, "delay", 100*time.Millisecond, "amount of time to delay creation of child") -} - -// Execute implements subcommands.Command. -func (c *forkBomb) Execute(ctx context.Context, f *flag.FlagSet, args ...interface{}) subcommands.ExitStatus { - time.Sleep(c.delay) - - cmd := exec.Command("/proc/self/exe", c.Name()) - cmd.Stdout = os.Stdout - cmd.Stderr = os.Stderr - if err := cmd.Run(); err != nil { - log.Fatal("failed to call self:", err) - } - return subcommands.ExitSuccess -} - -type reaper struct{} - -// Name implements subcommands.Command. -func (*reaper) Name() string { - return "reaper" -} - -// Synopsis implements subcommands.Command. -func (*reaper) Synopsis() string { - return "reaps all children in a loop" -} - -// Usage implements subcommands.Command. -func (*reaper) Usage() string { - return "reaper " -} - -// SetFlags implements subcommands.Command. -func (*reaper) SetFlags(*flag.FlagSet) {} - -// Execute implements subcommands.Command. -func (c *reaper) Execute(ctx context.Context, f *flag.FlagSet, args ...interface{}) subcommands.ExitStatus { - stop := testutil.StartReaper() - defer stop() - select {} -} - -type syscall struct { - sysno uint64 -} - -// Name implements subcommands.Command. -func (*syscall) Name() string { - return "syscall" -} - -// Synopsis implements subcommands.Command. -func (*syscall) Synopsis() string { - return "syscall makes a syscall" -} - -// Usage implements subcommands.Command. -func (*syscall) Usage() string { - return "syscall " -} - -// SetFlags implements subcommands.Command. -func (s *syscall) SetFlags(f *flag.FlagSet) { - f.Uint64Var(&s.sysno, "syscall", 0, "syscall to call") -} - -// Execute implements subcommands.Command. -func (s *syscall) Execute(ctx context.Context, f *flag.FlagSet, args ...interface{}) subcommands.ExitStatus { - if _, _, errno := sys.Syscall(uintptr(s.sysno), 0, 0, 0); errno != 0 { - fmt.Printf("syscall(%d, 0, 0...) failed: %v\n", s.sysno, errno) - } else { - fmt.Printf("syscall(%d, 0, 0...) success\n", s.sysno) - } - return subcommands.ExitSuccess -} - -type capability struct { - enabled uint64 - disabled uint64 -} - -// Name implements subcommands.Command. -func (*capability) Name() string { - return "capability" -} - -// Synopsis implements subcommands.Command. -func (*capability) Synopsis() string { - return "checks if effective capabilities are set/unset" -} - -// Usage implements subcommands.Command. -func (*capability) Usage() string { - return "capability [--enabled=number] [--disabled=number]" -} - -// SetFlags implements subcommands.Command. -func (c *capability) SetFlags(f *flag.FlagSet) { - f.Uint64Var(&c.enabled, "enabled", 0, "") - f.Uint64Var(&c.disabled, "disabled", 0, "") -} - -// Execute implements subcommands.Command. -func (c *capability) Execute(ctx context.Context, f *flag.FlagSet, args ...interface{}) subcommands.ExitStatus { - if c.enabled == 0 && c.disabled == 0 { - fmt.Println("One of the flags must be set") - return subcommands.ExitUsageError - } - - status, err := ioutil.ReadFile("/proc/self/status") - if err != nil { - fmt.Printf("Error reading %q: %v\n", "proc/self/status", err) - return subcommands.ExitFailure - } - re := regexp.MustCompile("CapEff:\t([0-9a-f]+)\n") - matches := re.FindStringSubmatch(string(status)) - if matches == nil || len(matches) != 2 { - fmt.Printf("Effective capabilities not found in\n%s\n", status) - return subcommands.ExitFailure - } - caps, err := strconv.ParseUint(matches[1], 16, 64) - if err != nil { - fmt.Printf("failed to convert capabilities %q: %v\n", matches[1], err) - return subcommands.ExitFailure - } - - if c.enabled != 0 && (caps&c.enabled) != c.enabled { - fmt.Printf("Missing capabilities, want: %#x: got: %#x\n", c.enabled, caps) - return subcommands.ExitFailure - } - if c.disabled != 0 && (caps&c.disabled) != 0 { - fmt.Printf("Extra capabilities found, dont_want: %#x: got: %#x\n", c.disabled, caps) - return subcommands.ExitFailure - } - - return subcommands.ExitSuccess -} - -type ptyRunner struct{} - -// Name implements subcommands.Command. -func (*ptyRunner) Name() string { - return "pty-runner" -} - -// Synopsis implements subcommands.Command. -func (*ptyRunner) Synopsis() string { - return "runs the given command with an open pty terminal" -} - -// Usage implements subcommands.Command. -func (*ptyRunner) Usage() string { - return "pty-runner [command]" -} - -// SetFlags implements subcommands.Command.SetFlags. -func (*ptyRunner) SetFlags(f *flag.FlagSet) {} - -// Execute implements subcommands.Command. -func (*ptyRunner) Execute(_ context.Context, fs *flag.FlagSet, _ ...interface{}) subcommands.ExitStatus { - c := exec.Command(fs.Args()[0], fs.Args()[1:]...) - f, err := pty.Start(c) - if err != nil { - fmt.Printf("pty.Start failed: %v", err) - return subcommands.ExitFailure - } - defer f.Close() - - // Copy stdout from the command to keep this process alive until the - // subprocess exits. - io.Copy(os.Stdout, f) - - return subcommands.ExitSuccess -} diff --git a/runsc/criutil/BUILD b/runsc/criutil/BUILD deleted file mode 100644 index 8a571a000..000000000 --- a/runsc/criutil/BUILD +++ /dev/null @@ -1,11 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "criutil", - testonly = 1, - srcs = ["criutil.go"], - visibility = ["//:sandbox"], - deps = ["//runsc/testutil"], -) diff --git a/runsc/criutil/criutil.go b/runsc/criutil/criutil.go deleted file mode 100644 index 773f5a1c4..000000000 --- a/runsc/criutil/criutil.go +++ /dev/null @@ -1,277 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package criutil contains utility functions for interacting with the -// Container Runtime Interface (CRI), principally via the crictl command line -// tool. This requires critools to be installed on the local system. -package criutil - -import ( - "encoding/json" - "fmt" - "os" - "os/exec" - "strings" - "time" - - "gvisor.dev/gvisor/runsc/testutil" -) - -const endpointPrefix = "unix://" - -// Crictl contains information required to run the crictl utility. -type Crictl struct { - executable string - timeout time.Duration - imageEndpoint string - runtimeEndpoint string -} - -// NewCrictl returns a Crictl configured with a timeout and an endpoint over -// which it will talk to containerd. -func NewCrictl(timeout time.Duration, endpoint string) *Crictl { - // Bazel doesn't pass PATH through, assume the location of crictl - // unless specified by environment variable. - executable := os.Getenv("CRICTL_PATH") - if executable == "" { - executable = "/usr/local/bin/crictl" - } - return &Crictl{ - executable: executable, - timeout: timeout, - imageEndpoint: endpointPrefix + endpoint, - runtimeEndpoint: endpointPrefix + endpoint, - } -} - -// Pull pulls an container image. It corresponds to `crictl pull`. -func (cc *Crictl) Pull(imageName string) error { - _, err := cc.run("pull", imageName) - return err -} - -// RunPod creates a sandbox. It corresponds to `crictl runp`. -func (cc *Crictl) RunPod(sbSpecFile string) (string, error) { - podID, err := cc.run("runp", sbSpecFile) - if err != nil { - return "", fmt.Errorf("runp failed: %v", err) - } - // Strip the trailing newline from crictl output. - return strings.TrimSpace(podID), nil -} - -// Create creates a container within a sandbox. It corresponds to `crictl -// create`. -func (cc *Crictl) Create(podID, contSpecFile, sbSpecFile string) (string, error) { - podID, err := cc.run("create", podID, contSpecFile, sbSpecFile) - if err != nil { - return "", fmt.Errorf("create failed: %v", err) - } - // Strip the trailing newline from crictl output. - return strings.TrimSpace(podID), nil -} - -// Start starts a container. It corresponds to `crictl start`. -func (cc *Crictl) Start(contID string) (string, error) { - output, err := cc.run("start", contID) - if err != nil { - return "", fmt.Errorf("start failed: %v", err) - } - return output, nil -} - -// Stop stops a container. It corresponds to `crictl stop`. -func (cc *Crictl) Stop(contID string) error { - _, err := cc.run("stop", contID) - return err -} - -// Exec execs a program inside a container. It corresponds to `crictl exec`. -func (cc *Crictl) Exec(contID string, args ...string) (string, error) { - a := []string{"exec", contID} - a = append(a, args...) - output, err := cc.run(a...) - if err != nil { - return "", fmt.Errorf("exec failed: %v", err) - } - return output, nil -} - -// Rm removes a container. It corresponds to `crictl rm`. -func (cc *Crictl) Rm(contID string) error { - _, err := cc.run("rm", contID) - return err -} - -// StopPod stops a pod. It corresponds to `crictl stopp`. -func (cc *Crictl) StopPod(podID string) error { - _, err := cc.run("stopp", podID) - return err -} - -// containsConfig is a minimal copy of -// https://github.com/kubernetes/kubernetes/blob/master/pkg/kubelet/apis/cri/runtime/v1alpha2/api.proto -// It only contains fields needed for testing. -type containerConfig struct { - Status containerStatus -} - -type containerStatus struct { - Network containerNetwork -} - -type containerNetwork struct { - IP string -} - -// PodIP returns a pod's IP address. -func (cc *Crictl) PodIP(podID string) (string, error) { - output, err := cc.run("inspectp", podID) - if err != nil { - return "", err - } - conf := &containerConfig{} - if err := json.Unmarshal([]byte(output), conf); err != nil { - return "", fmt.Errorf("failed to unmarshal JSON: %v, %s", err, output) - } - if conf.Status.Network.IP == "" { - return "", fmt.Errorf("no IP found in config: %s", output) - } - return conf.Status.Network.IP, nil -} - -// RmPod removes a container. It corresponds to `crictl rmp`. -func (cc *Crictl) RmPod(podID string) error { - _, err := cc.run("rmp", podID) - return err -} - -// StartContainer pulls the given image ands starts the container in the -// sandbox with the given podID. -func (cc *Crictl) StartContainer(podID, image, sbSpec, contSpec string) (string, error) { - // Write the specs to files that can be read by crictl. - sbSpecFile, err := testutil.WriteTmpFile("sbSpec", sbSpec) - if err != nil { - return "", fmt.Errorf("failed to write sandbox spec: %v", err) - } - contSpecFile, err := testutil.WriteTmpFile("contSpec", contSpec) - if err != nil { - return "", fmt.Errorf("failed to write container spec: %v", err) - } - - return cc.startContainer(podID, image, sbSpecFile, contSpecFile) -} - -func (cc *Crictl) startContainer(podID, image, sbSpecFile, contSpecFile string) (string, error) { - if err := cc.Pull(image); err != nil { - return "", fmt.Errorf("failed to pull %s: %v", image, err) - } - - contID, err := cc.Create(podID, contSpecFile, sbSpecFile) - if err != nil { - return "", fmt.Errorf("failed to create container in pod %q: %v", podID, err) - } - - if _, err := cc.Start(contID); err != nil { - return "", fmt.Errorf("failed to start container %q in pod %q: %v", contID, podID, err) - } - - return contID, nil -} - -// StopContainer stops and deletes the container with the given container ID. -func (cc *Crictl) StopContainer(contID string) error { - if err := cc.Stop(contID); err != nil { - return fmt.Errorf("failed to stop container %q: %v", contID, err) - } - - if err := cc.Rm(contID); err != nil { - return fmt.Errorf("failed to remove container %q: %v", contID, err) - } - - return nil -} - -// StartPodAndContainer pulls an image, then starts a sandbox and container in -// that sandbox. It returns the pod ID and container ID. -func (cc *Crictl) StartPodAndContainer(image, sbSpec, contSpec string) (string, string, error) { - // Write the specs to files that can be read by crictl. - sbSpecFile, err := testutil.WriteTmpFile("sbSpec", sbSpec) - if err != nil { - return "", "", fmt.Errorf("failed to write sandbox spec: %v", err) - } - contSpecFile, err := testutil.WriteTmpFile("contSpec", contSpec) - if err != nil { - return "", "", fmt.Errorf("failed to write container spec: %v", err) - } - - podID, err := cc.RunPod(sbSpecFile) - if err != nil { - return "", "", err - } - - contID, err := cc.startContainer(podID, image, sbSpecFile, contSpecFile) - - return podID, contID, err -} - -// StopPodAndContainer stops a container and pod. -func (cc *Crictl) StopPodAndContainer(podID, contID string) error { - if err := cc.StopContainer(contID); err != nil { - return fmt.Errorf("failed to stop container %q in pod %q: %v", contID, podID, err) - } - - if err := cc.StopPod(podID); err != nil { - return fmt.Errorf("failed to stop pod %q: %v", podID, err) - } - - if err := cc.RmPod(podID); err != nil { - return fmt.Errorf("failed to remove pod %q: %v", podID, err) - } - - return nil -} - -// run runs crictl with the given args and returns an error if it takes longer -// than cc.Timeout to run. -func (cc *Crictl) run(args ...string) (string, error) { - defaultArgs := []string{ - "--image-endpoint", cc.imageEndpoint, - "--runtime-endpoint", cc.runtimeEndpoint, - } - cmd := exec.Command(cc.executable, append(defaultArgs, args...)...) - - // Run the command with a timeout. - done := make(chan string) - errCh := make(chan error) - go func() { - output, err := cmd.CombinedOutput() - if err != nil { - errCh <- fmt.Errorf("error: \"%v\", output: %s", err, string(output)) - return - } - done <- string(output) - }() - select { - case output := <-done: - return output, nil - case err := <-errCh: - return "", err - case <-time.After(cc.timeout): - if err := testutil.KillCommand(cmd); err != nil { - return "", fmt.Errorf("timed out, then couldn't kill process %+v: %v", cmd, err) - } - return "", fmt.Errorf("timed out: %+v", cmd) - } -} diff --git a/runsc/dockerutil/BUILD b/runsc/dockerutil/BUILD deleted file mode 100644 index 8621af901..000000000 --- a/runsc/dockerutil/BUILD +++ /dev/null @@ -1,14 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "dockerutil", - testonly = 1, - srcs = ["dockerutil.go"], - visibility = ["//:sandbox"], - deps = [ - "//runsc/testutil", - "@com_github_kr_pty//:go_default_library", - ], -) diff --git a/runsc/dockerutil/dockerutil.go b/runsc/dockerutil/dockerutil.go deleted file mode 100644 index f009486bc..000000000 --- a/runsc/dockerutil/dockerutil.go +++ /dev/null @@ -1,486 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package dockerutil is a collection of utility functions, primarily for -// testing. -package dockerutil - -import ( - "encoding/json" - "flag" - "fmt" - "io/ioutil" - "log" - "os" - "os/exec" - "path" - "regexp" - "strconv" - "strings" - "syscall" - "time" - - "github.com/kr/pty" - "gvisor.dev/gvisor/runsc/testutil" -) - -var ( - // runtime is the runtime to use for tests. This will be applied to all - // containers. Note that the default here ("runsc") corresponds to the - // default used by the installations. This is important, because the - // default installer for vm_tests (in tools/installers:head, invoked - // via tools/vm:defs.bzl) will install with this name. So without - // changing anything, tests should have a runsc runtime available to - // them. Otherwise installers should update the existing runtime - // instead of installing a new one. - runtime = flag.String("runtime", "runsc", "specify which runtime to use") - - // config is the default Docker daemon configuration path. - config = flag.String("config_path", "/etc/docker/daemon.json", "configuration file for reading paths") -) - -// EnsureSupportedDockerVersion checks if correct docker is installed. -func EnsureSupportedDockerVersion() { - cmd := exec.Command("docker", "version") - out, err := cmd.CombinedOutput() - if err != nil { - log.Fatalf("Error running %q: %v", "docker version", err) - } - re := regexp.MustCompile(`Version:\s+(\d+)\.(\d+)\.\d.*`) - matches := re.FindStringSubmatch(string(out)) - if len(matches) != 3 { - log.Fatalf("Invalid docker output: %s", out) - } - major, _ := strconv.Atoi(matches[1]) - minor, _ := strconv.Atoi(matches[2]) - if major < 17 || (major == 17 && minor < 9) { - log.Fatalf("Docker version 17.09.0 or greater is required, found: %02d.%02d", major, minor) - } -} - -// RuntimePath returns the binary path for the current runtime. -func RuntimePath() (string, error) { - // Read the configuration data; the file must exist. - configBytes, err := ioutil.ReadFile(*config) - if err != nil { - return "", err - } - - // Unmarshal the configuration. - c := make(map[string]interface{}) - if err := json.Unmarshal(configBytes, &c); err != nil { - return "", err - } - - // Decode the expected configuration. - r, ok := c["runtimes"] - if !ok { - return "", fmt.Errorf("no runtimes declared: %v", c) - } - rs, ok := r.(map[string]interface{}) - if !ok { - // The runtimes are not a map. - return "", fmt.Errorf("unexpected format: %v", c) - } - r, ok = rs[*runtime] - if !ok { - // The expected runtime is not declared. - return "", fmt.Errorf("runtime %q not found: %v", *runtime, c) - } - rs, ok = r.(map[string]interface{}) - if !ok { - // The runtime is not a map. - return "", fmt.Errorf("unexpected format: %v", c) - } - p, ok := rs["path"].(string) - if !ok { - // The runtime does not declare a path. - return "", fmt.Errorf("unexpected format: %v", c) - } - return p, nil -} - -// MountMode describes if the mount should be ro or rw. -type MountMode int - -const ( - // ReadOnly is what the name says. - ReadOnly MountMode = iota - // ReadWrite is what the name says. - ReadWrite -) - -// String returns the mount mode argument for this MountMode. -func (m MountMode) String() string { - switch m { - case ReadOnly: - return "ro" - case ReadWrite: - return "rw" - } - panic(fmt.Sprintf("invalid mode: %d", m)) -} - -// MountArg formats the volume argument to mount in the container. -func MountArg(source, target string, mode MountMode) string { - return fmt.Sprintf("-v=%s:%s:%v", source, target, mode) -} - -// LinkArg formats the link argument. -func LinkArg(source *Docker, target string) string { - return fmt.Sprintf("--link=%s:%s", source.Name, target) -} - -// PrepareFiles creates temp directory to copy files there. The sandbox doesn't -// have access to files in the test dir. -func PrepareFiles(names ...string) (string, error) { - dir, err := ioutil.TempDir("", "image-test") - if err != nil { - return "", fmt.Errorf("ioutil.TempDir failed: %v", err) - } - if err := os.Chmod(dir, 0777); err != nil { - return "", fmt.Errorf("os.Chmod(%q, 0777) failed: %v", dir, err) - } - for _, name := range names { - src, err := testutil.FindFile(name) - if err != nil { - return "", fmt.Errorf("testutil.Preparefiles(%q) failed: %v", name, err) - } - dst := path.Join(dir, path.Base(name)) - if err := testutil.Copy(src, dst); err != nil { - return "", fmt.Errorf("testutil.Copy(%q, %q) failed: %v", src, dst, err) - } - } - return dir, nil -} - -// do executes docker command. -func do(args ...string) (string, error) { - log.Printf("Running: docker %s\n", args) - cmd := exec.Command("docker", args...) - out, err := cmd.CombinedOutput() - if err != nil { - return "", fmt.Errorf("error executing docker %s: %v\nout: %s", args, err, out) - } - return string(out), nil -} - -// doWithPty executes docker command with stdio attached to a pty. -func doWithPty(args ...string) (*exec.Cmd, *os.File, error) { - log.Printf("Running with pty: docker %s\n", args) - cmd := exec.Command("docker", args...) - ptmx, err := pty.Start(cmd) - if err != nil { - return nil, nil, fmt.Errorf("error executing docker %s with a pty: %v", args, err) - } - return cmd, ptmx, nil -} - -// Pull pulls a docker image. This is used in tests to isolate the -// time to pull the image off the network from the time to actually -// start the container, to avoid timeouts over slow networks. -func Pull(image string) error { - _, err := do("pull", image) - return err -} - -// Docker contains the name and the runtime of a docker container. -type Docker struct { - Runtime string - Name string -} - -// MakeDocker sets up the struct for a Docker container. -// Names of containers will be unique. -func MakeDocker(namePrefix string) Docker { - return Docker{ - Name: testutil.RandomName(namePrefix), - Runtime: *runtime, - } -} - -// logDockerID logs a container id, which is needed to find container runsc logs. -func (d *Docker) logDockerID() { - id, err := d.ID() - if err != nil { - log.Printf("%v\n", err) - } - log.Printf("Name: %s ID: %v\n", d.Name, id) -} - -// Create calls 'docker create' with the arguments provided. -func (d *Docker) Create(args ...string) error { - a := []string{"create", "--runtime", d.Runtime, "--name", d.Name} - a = append(a, args...) - _, err := do(a...) - if err == nil { - d.logDockerID() - } - return err -} - -// Start calls 'docker start'. -func (d *Docker) Start() error { - if _, err := do("start", d.Name); err != nil { - return fmt.Errorf("error starting container %q: %v", d.Name, err) - } - return nil -} - -// Stop calls 'docker stop'. -func (d *Docker) Stop() error { - if _, err := do("stop", d.Name); err != nil { - return fmt.Errorf("error stopping container %q: %v", d.Name, err) - } - return nil -} - -// Run calls 'docker run' with the arguments provided. The container starts -// running in the background and the call returns immediately. -func (d *Docker) Run(args ...string) error { - a := d.runArgs("-d") - a = append(a, args...) - _, err := do(a...) - if err == nil { - d.logDockerID() - } - return err -} - -// RunWithPty is like Run but with an attached pty. -func (d *Docker) RunWithPty(args ...string) (*exec.Cmd, *os.File, error) { - a := d.runArgs("-it") - a = append(a, args...) - return doWithPty(a...) -} - -// RunFg calls 'docker run' with the arguments provided in the foreground. It -// blocks until the container exits and returns the output. -func (d *Docker) RunFg(args ...string) (string, error) { - a := d.runArgs(args...) - out, err := do(a...) - if err == nil { - d.logDockerID() - } - return string(out), err -} - -func (d *Docker) runArgs(args ...string) []string { - // Environment variable RUNSC_TEST_NAME is picked up by the runtime and added - // to the log name, so one can easily identify the corresponding logs for - // this test. - rv := []string{"run", "--runtime", d.Runtime, "--name", d.Name, "-e", "RUNSC_TEST_NAME=" + d.Name} - return append(rv, args...) -} - -// Logs calls 'docker logs'. -func (d *Docker) Logs() (string, error) { - return do("logs", d.Name) -} - -// Exec calls 'docker exec' with the arguments provided. -func (d *Docker) Exec(args ...string) (string, error) { - return d.ExecWithFlags(nil, args...) -} - -// ExecWithFlags calls 'docker exec name '. -func (d *Docker) ExecWithFlags(flags []string, args ...string) (string, error) { - a := []string{"exec"} - a = append(a, flags...) - a = append(a, d.Name) - a = append(a, args...) - return do(a...) -} - -// ExecAsUser calls 'docker exec' as the given user with the arguments -// provided. -func (d *Docker) ExecAsUser(user string, args ...string) (string, error) { - a := []string{"exec", "--user", user, d.Name} - a = append(a, args...) - return do(a...) -} - -// ExecWithTerminal calls 'docker exec -it' with the arguments provided and -// attaches a pty to stdio. -func (d *Docker) ExecWithTerminal(args ...string) (*exec.Cmd, *os.File, error) { - a := []string{"exec", "-it", d.Name} - a = append(a, args...) - return doWithPty(a...) -} - -// Pause calls 'docker pause'. -func (d *Docker) Pause() error { - if _, err := do("pause", d.Name); err != nil { - return fmt.Errorf("error pausing container %q: %v", d.Name, err) - } - return nil -} - -// Unpause calls 'docker pause'. -func (d *Docker) Unpause() error { - if _, err := do("unpause", d.Name); err != nil { - return fmt.Errorf("error unpausing container %q: %v", d.Name, err) - } - return nil -} - -// Checkpoint calls 'docker checkpoint'. -func (d *Docker) Checkpoint(name string) error { - if _, err := do("checkpoint", "create", d.Name, name); err != nil { - return fmt.Errorf("error pausing container %q: %v", d.Name, err) - } - return nil -} - -// Restore calls 'docker start --checkname [name]'. -func (d *Docker) Restore(name string) error { - if _, err := do("start", "--checkpoint", name, d.Name); err != nil { - return fmt.Errorf("error starting container %q: %v", d.Name, err) - } - return nil -} - -// Remove calls 'docker rm'. -func (d *Docker) Remove() error { - if _, err := do("rm", d.Name); err != nil { - return fmt.Errorf("error deleting container %q: %v", d.Name, err) - } - return nil -} - -// CleanUp kills and deletes the container (best effort). -func (d *Docker) CleanUp() { - d.logDockerID() - if _, err := do("kill", d.Name); err != nil { - if strings.Contains(err.Error(), "is not running") { - // Nothing to kill. Don't log the error in this case. - } else { - log.Printf("error killing container %q: %v", d.Name, err) - } - } - if err := d.Remove(); err != nil { - log.Print(err) - } -} - -// FindPort returns the host port that is mapped to 'sandboxPort'. This calls -// docker to allocate a free port in the host and prevent conflicts. -func (d *Docker) FindPort(sandboxPort int) (int, error) { - format := fmt.Sprintf(`{{ (index (index .NetworkSettings.Ports "%d/tcp") 0).HostPort }}`, sandboxPort) - out, err := do("inspect", "-f", format, d.Name) - if err != nil { - return -1, fmt.Errorf("error retrieving port: %v", err) - } - port, err := strconv.Atoi(strings.TrimSuffix(string(out), "\n")) - if err != nil { - return -1, fmt.Errorf("error parsing port %q: %v", out, err) - } - return port, nil -} - -// FindIP returns the IP address of the container as a string. -func (d *Docker) FindIP() (string, error) { - const format = `{{range .NetworkSettings.Networks}}{{.IPAddress}}{{end}}` - out, err := do("inspect", "-f", format, d.Name) - if err != nil { - return "", fmt.Errorf("error retrieving IP: %v", err) - } - return strings.TrimSpace(out), nil -} - -// SandboxPid returns the PID to the sandbox process. -func (d *Docker) SandboxPid() (int, error) { - out, err := do("inspect", "-f={{.State.Pid}}", d.Name) - if err != nil { - return -1, fmt.Errorf("error retrieving pid: %v", err) - } - pid, err := strconv.Atoi(strings.TrimSuffix(string(out), "\n")) - if err != nil { - return -1, fmt.Errorf("error parsing pid %q: %v", out, err) - } - return pid, nil -} - -// ID returns the container ID. -func (d *Docker) ID() (string, error) { - out, err := do("inspect", "-f={{.Id}}", d.Name) - if err != nil { - return "", fmt.Errorf("error retrieving ID: %v", err) - } - return strings.TrimSpace(string(out)), nil -} - -// Wait waits for container to exit, up to the given timeout. Returns error if -// wait fails or timeout is hit. Returns the application return code otherwise. -// Note that the application may have failed even if err == nil, always check -// the exit code. -func (d *Docker) Wait(timeout time.Duration) (syscall.WaitStatus, error) { - timeoutChan := time.After(timeout) - waitChan := make(chan (syscall.WaitStatus)) - errChan := make(chan (error)) - - go func() { - out, err := do("wait", d.Name) - if err != nil { - errChan <- fmt.Errorf("error waiting for container %q: %v", d.Name, err) - } - exit, err := strconv.Atoi(strings.TrimSuffix(string(out), "\n")) - if err != nil { - errChan <- fmt.Errorf("error parsing exit code %q: %v", out, err) - } - waitChan <- syscall.WaitStatus(uint32(exit)) - }() - - select { - case ws := <-waitChan: - return ws, nil - case err := <-errChan: - return syscall.WaitStatus(1), err - case <-timeoutChan: - return syscall.WaitStatus(1), fmt.Errorf("timeout waiting for container %q", d.Name) - } -} - -// WaitForOutput calls 'docker logs' to retrieve containers output and searches -// for the given pattern. -func (d *Docker) WaitForOutput(pattern string, timeout time.Duration) (string, error) { - matches, err := d.WaitForOutputSubmatch(pattern, timeout) - if err != nil { - return "", err - } - if len(matches) == 0 { - return "", nil - } - return matches[0], nil -} - -// WaitForOutputSubmatch calls 'docker logs' to retrieve containers output and -// searches for the given pattern. It returns any regexp submatches as well. -func (d *Docker) WaitForOutputSubmatch(pattern string, timeout time.Duration) ([]string, error) { - re := regexp.MustCompile(pattern) - var out string - for exp := time.Now().Add(timeout); time.Now().Before(exp); { - var err error - out, err = d.Logs() - if err != nil { - return nil, err - } - if matches := re.FindStringSubmatch(out); matches != nil { - // Success! - return matches, nil - } - time.Sleep(100 * time.Millisecond) - } - return nil, fmt.Errorf("timeout waiting for output %q: %s", re.String(), out) -} diff --git a/runsc/testutil/BUILD b/runsc/testutil/BUILD deleted file mode 100644 index 945405303..000000000 --- a/runsc/testutil/BUILD +++ /dev/null @@ -1,21 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "testutil", - testonly = 1, - srcs = [ - "testutil.go", - "testutil_runfiles.go", - ], - visibility = ["//:sandbox"], - deps = [ - "//pkg/log", - "//pkg/sync", - "//runsc/boot", - "//runsc/specutils", - "@com_github_cenkalti_backoff//:go_default_library", - "@com_github_opencontainers_runtime-spec//specs-go:go_default_library", - ], -) diff --git a/runsc/testutil/testutil.go b/runsc/testutil/testutil.go deleted file mode 100644 index 5e09f8f16..000000000 --- a/runsc/testutil/testutil.go +++ /dev/null @@ -1,433 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package testutil contains utility functions for runsc tests. -package testutil - -import ( - "bufio" - "context" - "debug/elf" - "encoding/base32" - "encoding/json" - "flag" - "fmt" - "io" - "io/ioutil" - "math" - "math/rand" - "net/http" - "os" - "os/exec" - "os/signal" - "path" - "path/filepath" - "strconv" - "strings" - "sync/atomic" - "syscall" - "testing" - "time" - - "github.com/cenkalti/backoff" - specs "github.com/opencontainers/runtime-spec/specs-go" - "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/runsc/boot" - "gvisor.dev/gvisor/runsc/specutils" -) - -var ( - checkpoint = flag.Bool("checkpoint", true, "control checkpoint/restore support") -) - -func init() { - rand.Seed(time.Now().UnixNano()) -} - -// IsCheckpointSupported returns the relevant command line flag. -func IsCheckpointSupported() bool { - return *checkpoint -} - -// TmpDir returns the absolute path to a writable directory that can be used as -// scratch by the test. -func TmpDir() string { - dir := os.Getenv("TEST_TMPDIR") - if dir == "" { - dir = "/tmp" - } - return dir -} - -// ConfigureExePath configures the executable for runsc in the test environment. -func ConfigureExePath() error { - path, err := FindFile("runsc/runsc") - if err != nil { - return err - } - specutils.ExePath = path - return nil -} - -// TestConfig returns the default configuration to use in tests. Note that -// 'RootDir' must be set by caller if required. -func TestConfig(t *testing.T) *boot.Config { - logDir := "" - if dir, ok := os.LookupEnv("TEST_UNDECLARED_OUTPUTS_DIR"); ok { - logDir = dir + "/" - } - return &boot.Config{ - Debug: true, - DebugLog: path.Join(logDir, "runsc.log."+t.Name()+".%TIMESTAMP%.%COMMAND%"), - LogFormat: "text", - DebugLogFormat: "text", - LogPackets: true, - Network: boot.NetworkNone, - Strace: true, - Platform: "ptrace", - FileAccess: boot.FileAccessExclusive, - NumNetworkChannels: 1, - - TestOnlyAllowRunAsCurrentUserWithoutChroot: true, - } -} - -// NewSpecWithArgs creates a simple spec with the given args suitable for use -// in tests. -func NewSpecWithArgs(args ...string) *specs.Spec { - return &specs.Spec{ - // The host filesystem root is the container root. - Root: &specs.Root{ - Path: "/", - Readonly: true, - }, - Process: &specs.Process{ - Args: args, - Env: []string{ - "PATH=" + os.Getenv("PATH"), - }, - Capabilities: specutils.AllCapabilities(), - }, - Mounts: []specs.Mount{ - // Hide the host /etc to avoid any side-effects. - // For example, bash reads /etc/passwd and if it is - // very big, tests can fail by timeout. - { - Type: "tmpfs", - Destination: "/etc", - }, - // Root is readonly, but many tests want to write to tmpdir. - // This creates a writable mount inside the root. Also, when tmpdir points - // to "/tmp", it makes the the actual /tmp to be mounted and not a tmpfs - // inside the sentry. - { - Type: "bind", - Destination: TmpDir(), - Source: TmpDir(), - }, - }, - Hostname: "runsc-test-hostname", - } -} - -// SetupRootDir creates a root directory for containers. -func SetupRootDir() (string, error) { - rootDir, err := ioutil.TempDir(TmpDir(), "containers") - if err != nil { - return "", fmt.Errorf("error creating root dir: %v", err) - } - return rootDir, nil -} - -// SetupContainer creates a bundle and root dir for the container, generates a -// test config, and writes the spec to config.json in the bundle dir. -func SetupContainer(spec *specs.Spec, conf *boot.Config) (rootDir, bundleDir string, err error) { - rootDir, err = SetupRootDir() - if err != nil { - return "", "", err - } - conf.RootDir = rootDir - bundleDir, err = SetupBundleDir(spec) - return rootDir, bundleDir, err -} - -// SetupBundleDir creates a bundle dir and writes the spec to config.json. -func SetupBundleDir(spec *specs.Spec) (bundleDir string, err error) { - bundleDir, err = ioutil.TempDir(TmpDir(), "bundle") - if err != nil { - return "", fmt.Errorf("error creating bundle dir: %v", err) - } - - if err = writeSpec(bundleDir, spec); err != nil { - return "", fmt.Errorf("error writing spec: %v", err) - } - return bundleDir, nil -} - -// writeSpec writes the spec to disk in the given directory. -func writeSpec(dir string, spec *specs.Spec) error { - b, err := json.Marshal(spec) - if err != nil { - return err - } - return ioutil.WriteFile(filepath.Join(dir, "config.json"), b, 0755) -} - -// UniqueContainerID generates a unique container id for each test. -// -// The container id is used to create an abstract unix domain socket, which must -// be unique. While the container forbids creating two containers with the same -// name, sometimes between test runs the socket does not get cleaned up quickly -// enough, causing container creation to fail. -func UniqueContainerID() string { - // Read 20 random bytes. - b := make([]byte, 20) - // "[Read] always returns len(p) and a nil error." --godoc - if _, err := rand.Read(b); err != nil { - panic("rand.Read failed: " + err.Error()) - } - // base32 encode the random bytes, so that the name is a valid - // container id and can be used as a socket name in the filesystem. - return fmt.Sprintf("test-container-%s", base32.StdEncoding.EncodeToString(b)) -} - -// Copy copies file from src to dst. -func Copy(src, dst string) error { - in, err := os.Open(src) - if err != nil { - return err - } - defer in.Close() - - out, err := os.Create(dst) - if err != nil { - return err - } - defer out.Close() - - _, err = io.Copy(out, in) - return err -} - -// Poll is a shorthand function to poll for something with given timeout. -func Poll(cb func() error, timeout time.Duration) error { - ctx, cancel := context.WithTimeout(context.Background(), timeout) - defer cancel() - b := backoff.WithContext(backoff.NewConstantBackOff(100*time.Millisecond), ctx) - return backoff.Retry(cb, b) -} - -// WaitForHTTP tries GET requests on a port until the call succeeds or timeout. -func WaitForHTTP(port int, timeout time.Duration) error { - cb := func() error { - c := &http.Client{ - // Calculate timeout to be able to do minimum 5 attempts. - Timeout: timeout / 5, - } - url := fmt.Sprintf("http://localhost:%d/", port) - resp, err := c.Get(url) - if err != nil { - log.Infof("Waiting %s: %v", url, err) - return err - } - resp.Body.Close() - return nil - } - return Poll(cb, timeout) -} - -// Reaper reaps child processes. -type Reaper struct { - // mu protects ch, which will be nil if the reaper is not running. - mu sync.Mutex - ch chan os.Signal -} - -// Start starts reaping child processes. -func (r *Reaper) Start() { - r.mu.Lock() - defer r.mu.Unlock() - - if r.ch != nil { - panic("reaper.Start called on a running reaper") - } - - r.ch = make(chan os.Signal, 1) - signal.Notify(r.ch, syscall.SIGCHLD) - - go func() { - for { - r.mu.Lock() - ch := r.ch - r.mu.Unlock() - if ch == nil { - return - } - - _, ok := <-ch - if !ok { - // Channel closed. - return - } - for { - cpid, _ := syscall.Wait4(-1, nil, syscall.WNOHANG, nil) - if cpid < 1 { - break - } - } - } - }() -} - -// Stop stops reaping child processes. -func (r *Reaper) Stop() { - r.mu.Lock() - defer r.mu.Unlock() - - if r.ch == nil { - panic("reaper.Stop called on a stopped reaper") - } - - signal.Stop(r.ch) - close(r.ch) - r.ch = nil -} - -// StartReaper is a helper that starts a new Reaper and returns a function to -// stop it. -func StartReaper() func() { - r := &Reaper{} - r.Start() - return r.Stop -} - -// WaitUntilRead reads from the given reader until the wanted string is found -// or until timeout. -func WaitUntilRead(r io.Reader, want string, split bufio.SplitFunc, timeout time.Duration) error { - sc := bufio.NewScanner(r) - if split != nil { - sc.Split(split) - } - // done must be accessed atomically. A value greater than 0 indicates - // that the read loop can exit. - var done uint32 - doneCh := make(chan struct{}) - go func() { - for sc.Scan() { - t := sc.Text() - if strings.Contains(t, want) { - atomic.StoreUint32(&done, 1) - close(doneCh) - break - } - if atomic.LoadUint32(&done) > 0 { - break - } - } - }() - select { - case <-time.After(timeout): - atomic.StoreUint32(&done, 1) - return fmt.Errorf("timeout waiting to read %q", want) - case <-doneCh: - return nil - } -} - -// KillCommand kills the process running cmd unless it hasn't been started. It -// returns an error if it cannot kill the process unless the reason is that the -// process has already exited. -func KillCommand(cmd *exec.Cmd) error { - if cmd.Process == nil { - return nil - } - if err := cmd.Process.Kill(); err != nil { - if !strings.Contains(err.Error(), "process already finished") { - return fmt.Errorf("failed to kill process %v: %v", cmd, err) - } - } - return nil -} - -// WriteTmpFile writes text to a temporary file, closes the file, and returns -// the name of the file. -func WriteTmpFile(pattern, text string) (string, error) { - file, err := ioutil.TempFile(TmpDir(), pattern) - if err != nil { - return "", err - } - defer file.Close() - if _, err := file.Write([]byte(text)); err != nil { - return "", err - } - return file.Name(), nil -} - -// RandomName create a name with a 6 digit random number appended to it. -func RandomName(prefix string) string { - return fmt.Sprintf("%s-%06d", prefix, rand.Int31n(1000000)) -} - -// IsStatic returns true iff the given file is a static binary. -func IsStatic(filename string) (bool, error) { - f, err := elf.Open(filename) - if err != nil { - return false, err - } - for _, prog := range f.Progs { - if prog.Type == elf.PT_INTERP { - return false, nil // Has interpreter. - } - } - return true, nil -} - -// TestIndicesForShard returns indices for this test shard based on the -// TEST_SHARD_INDEX and TEST_TOTAL_SHARDS environment vars. -// -// If either of the env vars are not present, then the function will return all -// tests. If there are more shards than there are tests, then the returned list -// may be empty. -func TestIndicesForShard(numTests int) ([]int, error) { - var ( - shardIndex = 0 - shardTotal = 1 - ) - - indexStr, totalStr := os.Getenv("TEST_SHARD_INDEX"), os.Getenv("TEST_TOTAL_SHARDS") - if indexStr != "" && totalStr != "" { - // Parse index and total to ints. - var err error - shardIndex, err = strconv.Atoi(indexStr) - if err != nil { - return nil, fmt.Errorf("invalid TEST_SHARD_INDEX %q: %v", indexStr, err) - } - shardTotal, err = strconv.Atoi(totalStr) - if err != nil { - return nil, fmt.Errorf("invalid TEST_TOTAL_SHARDS %q: %v", totalStr, err) - } - } - - // Calculate! - var indices []int - numBlocks := int(math.Ceil(float64(numTests) / float64(shardTotal))) - for i := 0; i < numBlocks; i++ { - pick := i*shardTotal + shardIndex - if pick < numTests { - indices = append(indices, pick) - } - } - return indices, nil -} diff --git a/runsc/testutil/testutil_runfiles.go b/runsc/testutil/testutil_runfiles.go deleted file mode 100644 index ece9ea9a1..000000000 --- a/runsc/testutil/testutil_runfiles.go +++ /dev/null @@ -1,75 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package testutil - -import ( - "fmt" - "os" - "path/filepath" -) - -// FindFile searchs for a file inside the test run environment. It returns the -// full path to the file. It fails if none or more than one file is found. -func FindFile(path string) (string, error) { - wd, err := os.Getwd() - if err != nil { - return "", err - } - - // The test root is demarcated by a path element called "__main__". Search for - // it backwards from the working directory. - root := wd - for { - dir, name := filepath.Split(root) - if name == "__main__" { - break - } - if len(dir) == 0 { - return "", fmt.Errorf("directory __main__ not found in %q", wd) - } - // Remove ending slash to loop around. - root = dir[:len(dir)-1] - } - - // Annoyingly, bazel adds the build type to the directory path for go - // binaries, but not for c++ binaries. We use two different patterns to - // to find our file. - patterns := []string{ - // Try the obvious path first. - filepath.Join(root, path), - // If it was a go binary, use a wildcard to match the build - // type. The pattern is: /test-path/__main__/directories/*/file. - filepath.Join(root, filepath.Dir(path), "*", filepath.Base(path)), - } - - for _, p := range patterns { - matches, err := filepath.Glob(p) - if err != nil { - // "The only possible returned error is ErrBadPattern, - // when pattern is malformed." -godoc - return "", fmt.Errorf("error globbing %q: %v", p, err) - } - switch len(matches) { - case 0: - // Try the next pattern. - case 1: - // We found it. - return matches[0], nil - default: - return "", fmt.Errorf("more than one match found for %q: %s", path, matches) - } - } - return "", fmt.Errorf("file %q not found", path) -} diff --git a/scripts/iptables_tests.sh b/scripts/iptables_tests.sh index 0f46909ac..c8da1f32d 100755 --- a/scripts/iptables_tests.sh +++ b/scripts/iptables_tests.sh @@ -17,14 +17,5 @@ source $(dirname $0)/common.sh install_runsc_for_test iptables --net-raw - -# Build the docker image for the test. -run //test/iptables/runner:runner-image --norun - -test //test/iptables:iptables_test \ - "--test_arg=--runtime=runc" \ - "--test_arg=--image=bazel/test/iptables/runner:runner-image" - -test //test/iptables:iptables_test \ - "--test_arg=--runtime=${RUNTIME}" \ - "--test_arg=--image=bazel/test/iptables/runner:runner-image" +test //test/iptables:iptables_test --test_arg=--runtime=runc +test //test/iptables:iptables_test --test_arg=--runtime=${RUNTIME} diff --git a/test/cmd/test_app/BUILD b/test/cmd/test_app/BUILD new file mode 100644 index 000000000..98ba5a3d9 --- /dev/null +++ b/test/cmd/test_app/BUILD @@ -0,0 +1,21 @@ +load("//tools:defs.bzl", "go_binary") + +package(licenses = ["notice"]) + +go_binary( + name = "test_app", + testonly = 1, + srcs = [ + "fds.go", + "test_app.go", + ], + pure = True, + visibility = ["//runsc/container:__pkg__"], + deps = [ + "//pkg/test/testutil", + "//pkg/unet", + "//runsc/flag", + "@com_github_google_subcommands//:go_default_library", + "@com_github_kr_pty//:go_default_library", + ], +) diff --git a/test/cmd/test_app/fds.go b/test/cmd/test_app/fds.go new file mode 100644 index 000000000..a7658eefd --- /dev/null +++ b/test/cmd/test_app/fds.go @@ -0,0 +1,185 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package main + +import ( + "context" + "io/ioutil" + "log" + "os" + "time" + + "github.com/google/subcommands" + "gvisor.dev/gvisor/pkg/test/testutil" + "gvisor.dev/gvisor/pkg/unet" + "gvisor.dev/gvisor/runsc/flag" +) + +const fileContents = "foobarbaz" + +// fdSender will open a file and send the FD over a unix domain socket. +type fdSender struct { + socketPath string +} + +// Name implements subcommands.Command.Name. +func (*fdSender) Name() string { + return "fd_sender" +} + +// Synopsis implements subcommands.Command.Synopsys. +func (*fdSender) Synopsis() string { + return "creates a file and sends the FD over the socket" +} + +// Usage implements subcommands.Command.Usage. +func (*fdSender) Usage() string { + return "fd_sender " +} + +// SetFlags implements subcommands.Command.SetFlags. +func (fds *fdSender) SetFlags(f *flag.FlagSet) { + f.StringVar(&fds.socketPath, "socket", "", "path to socket") +} + +// Execute implements subcommands.Command.Execute. +func (fds *fdSender) Execute(ctx context.Context, f *flag.FlagSet, args ...interface{}) subcommands.ExitStatus { + if fds.socketPath == "" { + log.Fatalf("socket flag must be set") + } + + dir, err := ioutil.TempDir("", "") + if err != nil { + log.Fatalf("TempDir failed: %v", err) + } + + fileToSend, err := ioutil.TempFile(dir, "") + if err != nil { + log.Fatalf("TempFile failed: %v", err) + } + defer fileToSend.Close() + + if _, err := fileToSend.WriteString(fileContents); err != nil { + log.Fatalf("Write(%q) failed: %v", fileContents, err) + } + + // Receiver may not be started yet, so try connecting in a poll loop. + var s *unet.Socket + if err := testutil.Poll(func() error { + var err error + s, err = unet.Connect(fds.socketPath, true /* SEQPACKET, so we can send empty message with FD */) + return err + }, 10*time.Second); err != nil { + log.Fatalf("Error connecting to socket %q: %v", fds.socketPath, err) + } + defer s.Close() + + w := s.Writer(true) + w.ControlMessage.PackFDs(int(fileToSend.Fd())) + if _, err := w.WriteVec([][]byte{[]byte{'a'}}); err != nil { + log.Fatalf("Error sending FD %q over socket %q: %v", fileToSend.Fd(), fds.socketPath, err) + } + + log.Print("FD SENDER exiting successfully") + return subcommands.ExitSuccess +} + +// fdReceiver receives an FD from a unix domain socket and does things to it. +type fdReceiver struct { + socketPath string +} + +// Name implements subcommands.Command.Name. +func (*fdReceiver) Name() string { + return "fd_receiver" +} + +// Synopsis implements subcommands.Command.Synopsys. +func (*fdReceiver) Synopsis() string { + return "reads an FD from a unix socket, and then does things to it" +} + +// Usage implements subcommands.Command.Usage. +func (*fdReceiver) Usage() string { + return "fd_receiver " +} + +// SetFlags implements subcommands.Command.SetFlags. +func (fdr *fdReceiver) SetFlags(f *flag.FlagSet) { + f.StringVar(&fdr.socketPath, "socket", "", "path to socket") +} + +// Execute implements subcommands.Command.Execute. +func (fdr *fdReceiver) Execute(ctx context.Context, f *flag.FlagSet, args ...interface{}) subcommands.ExitStatus { + if fdr.socketPath == "" { + log.Fatalf("Flags cannot be empty, given: socket: %q", fdr.socketPath) + } + + ss, err := unet.BindAndListen(fdr.socketPath, true /* packet */) + if err != nil { + log.Fatalf("BindAndListen(%q) failed: %v", fdr.socketPath, err) + } + defer ss.Close() + + var s *unet.Socket + c := make(chan error, 1) + go func() { + var err error + s, err = ss.Accept() + c <- err + }() + + select { + case err := <-c: + if err != nil { + log.Fatalf("Accept() failed: %v", err) + } + case <-time.After(10 * time.Second): + log.Fatalf("Timeout waiting for accept") + } + + r := s.Reader(true) + r.EnableFDs(1) + b := [][]byte{{'a'}} + if n, err := r.ReadVec(b); n != 1 || err != nil { + log.Fatalf("ReadVec got n=%d err %v (wanted 0, nil)", n, err) + } + + fds, err := r.ExtractFDs() + if err != nil { + log.Fatalf("ExtractFD() got err %v", err) + } + if len(fds) != 1 { + log.Fatalf("ExtractFD() got %d FDs, wanted 1", len(fds)) + } + fd := fds[0] + + file := os.NewFile(uintptr(fd), "received file") + defer file.Close() + if _, err := file.Seek(0, os.SEEK_SET); err != nil { + log.Fatalf("Seek(0, 0) failed: %v", err) + } + + got, err := ioutil.ReadAll(file) + if err != nil { + log.Fatalf("ReadAll failed: %v", err) + } + if string(got) != fileContents { + log.Fatalf("ReadAll got %q want %q", string(got), fileContents) + } + + log.Print("FD RECEIVER exiting successfully") + return subcommands.ExitSuccess +} diff --git a/test/cmd/test_app/test_app.go b/test/cmd/test_app/test_app.go new file mode 100644 index 000000000..3ba4f38f8 --- /dev/null +++ b/test/cmd/test_app/test_app.go @@ -0,0 +1,394 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Binary test_app is like a swiss knife for tests that need to run anything +// inside the sandbox. New functionality can be added with new commands. +package main + +import ( + "context" + "fmt" + "io" + "io/ioutil" + "log" + "net" + "os" + "os/exec" + "regexp" + "strconv" + sys "syscall" + "time" + + "github.com/google/subcommands" + "github.com/kr/pty" + "gvisor.dev/gvisor/pkg/test/testutil" + "gvisor.dev/gvisor/runsc/flag" +) + +func main() { + subcommands.Register(subcommands.HelpCommand(), "") + subcommands.Register(subcommands.FlagsCommand(), "") + subcommands.Register(new(capability), "") + subcommands.Register(new(fdReceiver), "") + subcommands.Register(new(fdSender), "") + subcommands.Register(new(forkBomb), "") + subcommands.Register(new(ptyRunner), "") + subcommands.Register(new(reaper), "") + subcommands.Register(new(syscall), "") + subcommands.Register(new(taskTree), "") + subcommands.Register(new(uds), "") + + flag.Parse() + + exitCode := subcommands.Execute(context.Background()) + os.Exit(int(exitCode)) +} + +type uds struct { + fileName string + socketPath string +} + +// Name implements subcommands.Command.Name. +func (*uds) Name() string { + return "uds" +} + +// Synopsis implements subcommands.Command.Synopsys. +func (*uds) Synopsis() string { + return "creates unix domain socket client and server. Client sends a contant flow of sequential numbers. Server prints them to --file" +} + +// Usage implements subcommands.Command.Usage. +func (*uds) Usage() string { + return "uds " +} + +// SetFlags implements subcommands.Command.SetFlags. +func (c *uds) SetFlags(f *flag.FlagSet) { + f.StringVar(&c.fileName, "file", "", "name of output file") + f.StringVar(&c.socketPath, "socket", "", "path to socket") +} + +// Execute implements subcommands.Command.Execute. +func (c *uds) Execute(ctx context.Context, f *flag.FlagSet, args ...interface{}) subcommands.ExitStatus { + if c.fileName == "" || c.socketPath == "" { + log.Fatalf("Flags cannot be empty, given: fileName: %q, socketPath: %q", c.fileName, c.socketPath) + return subcommands.ExitFailure + } + outputFile, err := os.OpenFile(c.fileName, os.O_WRONLY|os.O_CREATE, 0666) + if err != nil { + log.Fatal("error opening output file:", err) + } + + defer os.Remove(c.socketPath) + + listener, err := net.Listen("unix", c.socketPath) + if err != nil { + log.Fatalf("error listening on socket %q: %v", c.socketPath, err) + } + + go server(listener, outputFile) + for i := 0; ; i++ { + conn, err := net.Dial("unix", c.socketPath) + if err != nil { + log.Fatal("error dialing:", err) + } + if _, err := conn.Write([]byte(strconv.Itoa(i))); err != nil { + log.Fatal("error writing:", err) + } + conn.Close() + time.Sleep(100 * time.Millisecond) + } +} + +func server(listener net.Listener, out *os.File) { + buf := make([]byte, 16) + + for { + c, err := listener.Accept() + if err != nil { + log.Fatal("error accepting connection:", err) + } + nr, err := c.Read(buf) + if err != nil { + log.Fatal("error reading from buf:", err) + } + data := buf[0:nr] + fmt.Fprint(out, string(data)+"\n") + } +} + +type taskTree struct { + depth int + width int + pause bool +} + +// Name implements subcommands.Command. +func (*taskTree) Name() string { + return "task-tree" +} + +// Synopsis implements subcommands.Command. +func (*taskTree) Synopsis() string { + return "creates a tree of tasks" +} + +// Usage implements subcommands.Command. +func (*taskTree) Usage() string { + return "task-tree " +} + +// SetFlags implements subcommands.Command. +func (c *taskTree) SetFlags(f *flag.FlagSet) { + f.IntVar(&c.depth, "depth", 1, "number of levels to create") + f.IntVar(&c.width, "width", 1, "number of tasks at each level") + f.BoolVar(&c.pause, "pause", false, "whether the tasks should pause perpetually") +} + +// Execute implements subcommands.Command. +func (c *taskTree) Execute(ctx context.Context, f *flag.FlagSet, args ...interface{}) subcommands.ExitStatus { + stop := testutil.StartReaper() + defer stop() + + if c.depth == 0 { + log.Printf("Child sleeping, PID: %d\n", os.Getpid()) + select {} + } + log.Printf("Parent %d sleeping, PID: %d\n", c.depth, os.Getpid()) + + var cmds []*exec.Cmd + for i := 0; i < c.width; i++ { + cmd := exec.Command( + "/proc/self/exe", c.Name(), + "--depth", strconv.Itoa(c.depth-1), + "--width", strconv.Itoa(c.width), + "--pause", strconv.FormatBool(c.pause)) + cmd.Stdout = os.Stdout + cmd.Stderr = os.Stderr + + if err := cmd.Start(); err != nil { + log.Fatal("failed to call self:", err) + } + cmds = append(cmds, cmd) + } + + for _, c := range cmds { + c.Wait() + } + + if c.pause { + select {} + } + + return subcommands.ExitSuccess +} + +type forkBomb struct { + delay time.Duration +} + +// Name implements subcommands.Command. +func (*forkBomb) Name() string { + return "fork-bomb" +} + +// Synopsis implements subcommands.Command. +func (*forkBomb) Synopsis() string { + return "creates child process until the end of times" +} + +// Usage implements subcommands.Command. +func (*forkBomb) Usage() string { + return "fork-bomb " +} + +// SetFlags implements subcommands.Command. +func (c *forkBomb) SetFlags(f *flag.FlagSet) { + f.DurationVar(&c.delay, "delay", 100*time.Millisecond, "amount of time to delay creation of child") +} + +// Execute implements subcommands.Command. +func (c *forkBomb) Execute(ctx context.Context, f *flag.FlagSet, args ...interface{}) subcommands.ExitStatus { + time.Sleep(c.delay) + + cmd := exec.Command("/proc/self/exe", c.Name()) + cmd.Stdout = os.Stdout + cmd.Stderr = os.Stderr + if err := cmd.Run(); err != nil { + log.Fatal("failed to call self:", err) + } + return subcommands.ExitSuccess +} + +type reaper struct{} + +// Name implements subcommands.Command. +func (*reaper) Name() string { + return "reaper" +} + +// Synopsis implements subcommands.Command. +func (*reaper) Synopsis() string { + return "reaps all children in a loop" +} + +// Usage implements subcommands.Command. +func (*reaper) Usage() string { + return "reaper " +} + +// SetFlags implements subcommands.Command. +func (*reaper) SetFlags(*flag.FlagSet) {} + +// Execute implements subcommands.Command. +func (c *reaper) Execute(ctx context.Context, f *flag.FlagSet, args ...interface{}) subcommands.ExitStatus { + stop := testutil.StartReaper() + defer stop() + select {} +} + +type syscall struct { + sysno uint64 +} + +// Name implements subcommands.Command. +func (*syscall) Name() string { + return "syscall" +} + +// Synopsis implements subcommands.Command. +func (*syscall) Synopsis() string { + return "syscall makes a syscall" +} + +// Usage implements subcommands.Command. +func (*syscall) Usage() string { + return "syscall " +} + +// SetFlags implements subcommands.Command. +func (s *syscall) SetFlags(f *flag.FlagSet) { + f.Uint64Var(&s.sysno, "syscall", 0, "syscall to call") +} + +// Execute implements subcommands.Command. +func (s *syscall) Execute(ctx context.Context, f *flag.FlagSet, args ...interface{}) subcommands.ExitStatus { + if _, _, errno := sys.Syscall(uintptr(s.sysno), 0, 0, 0); errno != 0 { + fmt.Printf("syscall(%d, 0, 0...) failed: %v\n", s.sysno, errno) + } else { + fmt.Printf("syscall(%d, 0, 0...) success\n", s.sysno) + } + return subcommands.ExitSuccess +} + +type capability struct { + enabled uint64 + disabled uint64 +} + +// Name implements subcommands.Command. +func (*capability) Name() string { + return "capability" +} + +// Synopsis implements subcommands.Command. +func (*capability) Synopsis() string { + return "checks if effective capabilities are set/unset" +} + +// Usage implements subcommands.Command. +func (*capability) Usage() string { + return "capability [--enabled=number] [--disabled=number]" +} + +// SetFlags implements subcommands.Command. +func (c *capability) SetFlags(f *flag.FlagSet) { + f.Uint64Var(&c.enabled, "enabled", 0, "") + f.Uint64Var(&c.disabled, "disabled", 0, "") +} + +// Execute implements subcommands.Command. +func (c *capability) Execute(ctx context.Context, f *flag.FlagSet, args ...interface{}) subcommands.ExitStatus { + if c.enabled == 0 && c.disabled == 0 { + fmt.Println("One of the flags must be set") + return subcommands.ExitUsageError + } + + status, err := ioutil.ReadFile("/proc/self/status") + if err != nil { + fmt.Printf("Error reading %q: %v\n", "proc/self/status", err) + return subcommands.ExitFailure + } + re := regexp.MustCompile("CapEff:\t([0-9a-f]+)\n") + matches := re.FindStringSubmatch(string(status)) + if matches == nil || len(matches) != 2 { + fmt.Printf("Effective capabilities not found in\n%s\n", status) + return subcommands.ExitFailure + } + caps, err := strconv.ParseUint(matches[1], 16, 64) + if err != nil { + fmt.Printf("failed to convert capabilities %q: %v\n", matches[1], err) + return subcommands.ExitFailure + } + + if c.enabled != 0 && (caps&c.enabled) != c.enabled { + fmt.Printf("Missing capabilities, want: %#x: got: %#x\n", c.enabled, caps) + return subcommands.ExitFailure + } + if c.disabled != 0 && (caps&c.disabled) != 0 { + fmt.Printf("Extra capabilities found, dont_want: %#x: got: %#x\n", c.disabled, caps) + return subcommands.ExitFailure + } + + return subcommands.ExitSuccess +} + +type ptyRunner struct{} + +// Name implements subcommands.Command. +func (*ptyRunner) Name() string { + return "pty-runner" +} + +// Synopsis implements subcommands.Command. +func (*ptyRunner) Synopsis() string { + return "runs the given command with an open pty terminal" +} + +// Usage implements subcommands.Command. +func (*ptyRunner) Usage() string { + return "pty-runner [command]" +} + +// SetFlags implements subcommands.Command.SetFlags. +func (*ptyRunner) SetFlags(f *flag.FlagSet) {} + +// Execute implements subcommands.Command. +func (*ptyRunner) Execute(_ context.Context, fs *flag.FlagSet, _ ...interface{}) subcommands.ExitStatus { + c := exec.Command(fs.Args()[0], fs.Args()[1:]...) + f, err := pty.Start(c) + if err != nil { + fmt.Printf("pty.Start failed: %v", err) + return subcommands.ExitFailure + } + defer f.Close() + + // Copy stdout from the command to keep this process alive until the + // subprocess exits. + io.Copy(os.Stdout, f) + + return subcommands.ExitSuccess +} diff --git a/test/e2e/BUILD b/test/e2e/BUILD index 76e04f878..44cce0e3b 100644 --- a/test/e2e/BUILD +++ b/test/e2e/BUILD @@ -20,9 +20,9 @@ go_test( deps = [ "//pkg/abi/linux", "//pkg/bits", - "//runsc/dockerutil", + "//pkg/test/dockerutil", + "//pkg/test/testutil", "//runsc/specutils", - "//runsc/testutil", ], ) diff --git a/test/e2e/exec_test.go b/test/e2e/exec_test.go index 594c8e752..6a63b1232 100644 --- a/test/e2e/exec_test.go +++ b/test/e2e/exec_test.go @@ -23,6 +23,8 @@ package integration import ( "fmt" + "os" + "os/exec" "strconv" "strings" "syscall" @@ -31,23 +33,23 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/bits" - "gvisor.dev/gvisor/runsc/dockerutil" + "gvisor.dev/gvisor/pkg/test/dockerutil" "gvisor.dev/gvisor/runsc/specutils" ) // Test that exec uses the exact same capability set as the container. func TestExecCapabilities(t *testing.T) { - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("exec-capabilities-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() // Start the container. - if err := d.Run("alpine", "sh", "-c", "cat /proc/self/status; sleep 100"); err != nil { + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/alpine", + }, "sh", "-c", "cat /proc/self/status; sleep 100"); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() + // Check that capability. matches, err := d.WaitForOutputSubmatch("CapEff:\t([0-9a-f]+)\n", 5*time.Second) if err != nil { t.Fatalf("WaitForOutputSubmatch() timeout: %v", err) @@ -59,7 +61,7 @@ func TestExecCapabilities(t *testing.T) { t.Log("Root capabilities:", want) // Now check that exec'd process capabilities match the root. - got, err := d.Exec("grep", "CapEff:", "/proc/self/status") + got, err := d.Exec(dockerutil.RunOpts{}, "grep", "CapEff:", "/proc/self/status") if err != nil { t.Fatalf("docker exec failed: %v", err) } @@ -72,16 +74,16 @@ func TestExecCapabilities(t *testing.T) { // Test that 'exec --privileged' adds all capabilities, except for CAP_NET_RAW // which is removed from the container when --net-raw=false. func TestExecPrivileged(t *testing.T) { - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("exec-privileged-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() // Start the container with all capabilities dropped. - if err := d.Run("--cap-drop=all", "alpine", "sh", "-c", "cat /proc/self/status; sleep 100"); err != nil { + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/alpine", + CapDrop: []string{"all"}, + }, "sh", "-c", "cat /proc/self/status; sleep 100"); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // Check that all capabilities where dropped from container. matches, err := d.WaitForOutputSubmatch("CapEff:\t([0-9a-f]+)\n", 5*time.Second) @@ -100,9 +102,11 @@ func TestExecPrivileged(t *testing.T) { t.Fatalf("Container should have no capabilities: %x", containerCaps) } - // Check that 'exec --privileged' adds all capabilities, except - // for CAP_NET_RAW. - got, err := d.ExecWithFlags([]string{"--privileged"}, "grep", "CapEff:", "/proc/self/status") + // Check that 'exec --privileged' adds all capabilities, except for + // CAP_NET_RAW. + got, err := d.Exec(dockerutil.RunOpts{ + Privileged: true, + }, "grep", "CapEff:", "/proc/self/status") if err != nil { t.Fatalf("docker exec failed: %v", err) } @@ -114,97 +118,99 @@ func TestExecPrivileged(t *testing.T) { } func TestExecJobControl(t *testing.T) { - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("exec-job-control-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() // Start the container. - if err := d.Run("alpine", "sleep", "1000"); err != nil { + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/alpine", + }, "sleep", "1000"); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // Exec 'sh' with an attached pty. - cmd, ptmx, err := d.ExecWithTerminal("sh") - if err != nil { + if _, err := d.Exec(dockerutil.RunOpts{ + Pty: func(cmd *exec.Cmd, ptmx *os.File) { + // Call "sleep 100 | cat" in the shell. We pipe to cat + // so that there will be two processes in the + // foreground process group. + if _, err := ptmx.Write([]byte("sleep 100 | cat\n")); err != nil { + t.Fatalf("error writing to pty: %v", err) + } + + // Give shell a few seconds to start executing the sleep. + time.Sleep(2 * time.Second) + + // Send a ^C to the pty, which should kill sleep and + // cat, but not the shell. \x03 is ASCII "end of + // text", which is the same as ^C. + if _, err := ptmx.Write([]byte{'\x03'}); err != nil { + t.Fatalf("error writing to pty: %v", err) + } + + // The shell should still be alive at this point. Sleep + // should have exited with code 2+128=130. We'll exit + // with 10 plus that number, so that we can be sure + // that the shell did not get signalled. + if _, err := ptmx.Write([]byte("exit $(expr $? + 10)\n")); err != nil { + t.Fatalf("error writing to pty: %v", err) + } + + // Exec process should exit with code 10+130=140. + ps, err := cmd.Process.Wait() + if err != nil { + t.Fatalf("error waiting for exec process: %v", err) + } + ws := ps.Sys().(syscall.WaitStatus) + if !ws.Exited() { + t.Errorf("ws.Exited got false, want true") + } + if got, want := ws.ExitStatus(), 140; got != want { + t.Errorf("ws.ExitedStatus got %d, want %d", got, want) + } + }, + }, "sh"); err != nil { t.Fatalf("docker exec failed: %v", err) } - defer ptmx.Close() - - // Call "sleep 100 | cat" in the shell. We pipe to cat so that there - // will be two processes in the foreground process group. - if _, err := ptmx.Write([]byte("sleep 100 | cat\n")); err != nil { - t.Fatalf("error writing to pty: %v", err) - } - - // Give shell a few seconds to start executing the sleep. - time.Sleep(2 * time.Second) - - // Send a ^C to the pty, which should kill sleep and cat, but not the - // shell. \x03 is ASCII "end of text", which is the same as ^C. - if _, err := ptmx.Write([]byte{'\x03'}); err != nil { - t.Fatalf("error writing to pty: %v", err) - } - - // The shell should still be alive at this point. Sleep should have - // exited with code 2+128=130. We'll exit with 10 plus that number, so - // that we can be sure that the shell did not get signalled. - if _, err := ptmx.Write([]byte("exit $(expr $? + 10)\n")); err != nil { - t.Fatalf("error writing to pty: %v", err) - } - - // Exec process should exit with code 10+130=140. - ps, err := cmd.Process.Wait() - if err != nil { - t.Fatalf("error waiting for exec process: %v", err) - } - ws := ps.Sys().(syscall.WaitStatus) - if !ws.Exited() { - t.Errorf("ws.Exited got false, want true") - } - if got, want := ws.ExitStatus(), 140; got != want { - t.Errorf("ws.ExitedStatus got %d, want %d", got, want) - } } // Test that failure to exec returns proper error message. func TestExecError(t *testing.T) { - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("exec-error-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() // Start the container. - if err := d.Run("alpine", "sleep", "1000"); err != nil { + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/alpine", + }, "sleep", "1000"); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() - _, err := d.Exec("no_can_find") + // Attempt to exec a binary that doesn't exist. + out, err := d.Exec(dockerutil.RunOpts{}, "no_can_find") if err == nil { t.Fatalf("docker exec didn't fail") } - if want := `error finding executable "no_can_find" in PATH`; !strings.Contains(err.Error(), want) { - t.Fatalf("docker exec wrong error, got: %s, want: .*%s.*", err.Error(), want) + if want := `error finding executable "no_can_find" in PATH`; !strings.Contains(out, want) { + t.Fatalf("docker exec wrong error, got: %s, want: .*%s.*", out, want) } } // Test that exec inherits environment from run. func TestExecEnv(t *testing.T) { - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("exec-env-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() // Start the container with env FOO=BAR. - if err := d.Run("-e", "FOO=BAR", "alpine", "sleep", "1000"); err != nil { + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/alpine", + Env: []string{"FOO=BAR"}, + }, "sleep", "1000"); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // Exec "echo $FOO". - got, err := d.Exec("/bin/sh", "-c", "echo $FOO") + got, err := d.Exec(dockerutil.RunOpts{}, "/bin/sh", "-c", "echo $FOO") if err != nil { t.Fatalf("docker exec failed: %v", err) } @@ -216,17 +222,19 @@ func TestExecEnv(t *testing.T) { // TestRunEnvHasHome tests that run always has HOME environment set. func TestRunEnvHasHome(t *testing.T) { // Base alpine image does not have any environment variables set. - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("run-env-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() // Exec "echo $HOME". The 'bin' user's home dir is '/bin'. - got, err := d.RunFg("--user", "bin", "alpine", "/bin/sh", "-c", "echo $HOME") + got, err := d.Run(dockerutil.RunOpts{ + Image: "basic/alpine", + User: "bin", + }, "/bin/sh", "-c", "echo $HOME") if err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() + + // Check that the directory matches. if got, want := strings.TrimSpace(got), "/bin"; got != want { t.Errorf("bad output from 'docker run'. Got %q; Want %q.", got, want) } @@ -235,18 +243,17 @@ func TestRunEnvHasHome(t *testing.T) { // Test that exec always has HOME environment set, even when not set in run. func TestExecEnvHasHome(t *testing.T) { // Base alpine image does not have any environment variables set. - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("exec-env-home-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() - if err := d.Run("alpine", "sleep", "1000"); err != nil { + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/alpine", + }, "sleep", "1000"); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // Exec "echo $HOME", and expect to see "/root". - got, err := d.Exec("/bin/sh", "-c", "echo $HOME") + got, err := d.Exec(dockerutil.RunOpts{}, "/bin/sh", "-c", "echo $HOME") if err != nil { t.Fatalf("docker exec failed: %v", err) } @@ -258,12 +265,14 @@ func TestExecEnvHasHome(t *testing.T) { newUID := 1234 newHome := "/foo/bar" cmd := fmt.Sprintf("mkdir -p -m 777 %q && adduser foo -D -u %d -h %q", newHome, newUID, newHome) - if _, err := d.Exec("/bin/sh", "-c", cmd); err != nil { + if _, err := d.Exec(dockerutil.RunOpts{}, "/bin/sh", "-c", cmd); err != nil { t.Fatalf("docker exec failed: %v", err) } // Execute the same as the new user and expect newHome. - got, err = d.ExecAsUser(strconv.Itoa(newUID), "/bin/sh", "-c", "echo $HOME") + got, err = d.Exec(dockerutil.RunOpts{ + User: strconv.Itoa(newUID), + }, "/bin/sh", "-c", "echo $HOME") if err != nil { t.Fatalf("docker exec failed: %v", err) } diff --git a/test/e2e/integration_test.go b/test/e2e/integration_test.go index cc4fbbaed..404e37689 100644 --- a/test/e2e/integration_test.go +++ b/test/e2e/integration_test.go @@ -27,14 +27,15 @@ import ( "net" "net/http" "os" + "os/exec" "strconv" "strings" "syscall" "testing" "time" - "gvisor.dev/gvisor/runsc/dockerutil" - "gvisor.dev/gvisor/runsc/testutil" + "gvisor.dev/gvisor/pkg/test/dockerutil" + "gvisor.dev/gvisor/pkg/test/testutil" ) // httpRequestSucceeds sends a request to a given url and checks that the status is OK. @@ -53,65 +54,66 @@ func httpRequestSucceeds(client http.Client, server string, port int) error { // TestLifeCycle tests a basic Create/Start/Stop docker container life cycle. func TestLifeCycle(t *testing.T) { - if err := dockerutil.Pull("nginx"); err != nil { - t.Fatal("docker pull failed:", err) - } - d := dockerutil.MakeDocker("lifecycle-test") - if err := d.Create("-p", "80", "nginx"); err != nil { - t.Fatal("docker create failed:", err) + d := dockerutil.MakeDocker(t) + defer d.CleanUp() + + // Start the container. + if err := d.Create(dockerutil.RunOpts{ + Image: "basic/nginx", + Ports: []int{80}, + }); err != nil { + t.Fatalf("docker create failed: %v", err) } if err := d.Start(); err != nil { - d.CleanUp() - t.Fatal("docker start failed:", err) + t.Fatalf("docker start failed: %v", err) } - // Test that container is working + // Test that container is working. port, err := d.FindPort(80) if err != nil { - t.Fatal("docker.FindPort(80) failed: ", err) + t.Fatalf("docker.FindPort(80) failed: %v", err) } if err := testutil.WaitForHTTP(port, 30*time.Second); err != nil { - t.Fatal("WaitForHTTP() timeout:", err) + t.Fatalf("WaitForHTTP() timeout: %v", err) } client := http.Client{Timeout: time.Duration(2 * time.Second)} if err := httpRequestSucceeds(client, "localhost", port); err != nil { - t.Error("http request failed:", err) + t.Errorf("http request failed: %v", err) } if err := d.Stop(); err != nil { - d.CleanUp() - t.Fatal("docker stop failed:", err) + t.Fatalf("docker stop failed: %v", err) } if err := d.Remove(); err != nil { - t.Fatal("docker rm failed:", err) + t.Fatalf("docker rm failed: %v", err) } } func TestPauseResume(t *testing.T) { - const img = "gcr.io/gvisor-presubmit/python-hello" if !testutil.IsCheckpointSupported() { - t.Log("Checkpoint is not supported, skipping test.") - return + t.Skip("Checkpoint is not supported.") } - if err := dockerutil.Pull(img); err != nil { - t.Fatal("docker pull failed:", err) - } - d := dockerutil.MakeDocker("pause-resume-test") - if err := d.Run("-p", "8080", img); err != nil { + d := dockerutil.MakeDocker(t) + defer d.CleanUp() + + // Start the container. + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/python", + Ports: []int{8080}, // See Dockerfile. + }); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // Find where port 8080 is mapped to. port, err := d.FindPort(8080) if err != nil { - t.Fatal("docker.FindPort(8080) failed:", err) + t.Fatalf("docker.FindPort(8080) failed: %v", err) } // Wait until it's up and running. if err := testutil.WaitForHTTP(port, 30*time.Second); err != nil { - t.Fatal("WaitForHTTP() timeout:", err) + t.Fatalf("WaitForHTTP() timeout: %v", err) } // Check that container is working. @@ -121,7 +123,7 @@ func TestPauseResume(t *testing.T) { } if err := d.Pause(); err != nil { - t.Fatal("docker pause failed:", err) + t.Fatalf("docker pause failed: %v", err) } // Check if container is paused. @@ -137,12 +139,12 @@ func TestPauseResume(t *testing.T) { } if err := d.Unpause(); err != nil { - t.Fatal("docker unpause failed:", err) + t.Fatalf("docker unpause failed: %v", err) } // Wait until it's up and running. if err := testutil.WaitForHTTP(port, 30*time.Second); err != nil { - t.Fatal("WaitForHTTP() timeout:", err) + t.Fatalf("WaitForHTTP() timeout: %v", err) } // Check if container is working again. @@ -152,43 +154,43 @@ func TestPauseResume(t *testing.T) { } func TestCheckpointRestore(t *testing.T) { - const img = "gcr.io/gvisor-presubmit/python-hello" if !testutil.IsCheckpointSupported() { - t.Log("Pause/resume is not supported, skipping test.") - return + t.Skip("Pause/resume is not supported.") } - if err := dockerutil.Pull(img); err != nil { - t.Fatal("docker pull failed:", err) - } - d := dockerutil.MakeDocker("save-restore-test") - if err := d.Run("-p", "8080", img); err != nil { + d := dockerutil.MakeDocker(t) + defer d.CleanUp() + + // Start the container. + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/python", + Ports: []int{8080}, // See Dockerfile. + }); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() + // Create a snapshot. if err := d.Checkpoint("test"); err != nil { - t.Fatal("docker checkpoint failed:", err) + t.Fatalf("docker checkpoint failed: %v", err) } - if _, err := d.Wait(30 * time.Second); err != nil { - t.Fatal(err) + t.Fatalf("wait failed: %v", err) } // TODO(b/143498576): Remove Poll after github.com/moby/moby/issues/38963 is fixed. if err := testutil.Poll(func() error { return d.Restore("test") }, 15*time.Second); err != nil { - t.Fatal("docker restore failed:", err) + t.Fatalf("docker restore failed: %v", err) } // Find where port 8080 is mapped to. port, err := d.FindPort(8080) if err != nil { - t.Fatal("docker.FindPort(8080) failed:", err) + t.Fatalf("docker.FindPort(8080) failed: %v", err) } // Wait until it's up and running. if err := testutil.WaitForHTTP(port, 30*time.Second); err != nil { - t.Fatal("WaitForHTTP() timeout:", err) + t.Fatalf("WaitForHTTP() timeout: %v", err) } // Check if container is working again. @@ -200,26 +202,28 @@ func TestCheckpointRestore(t *testing.T) { // Create client and server that talk to each other using the local IP. func TestConnectToSelf(t *testing.T) { - d := dockerutil.MakeDocker("connect-to-self-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() // Creates server that replies "server" and exists. Sleeps at the end because // 'docker exec' gets killed if the init process exists before it can finish. - if err := d.Run("ubuntu:trusty", "/bin/sh", "-c", "echo server | nc -l -p 8080 && sleep 1"); err != nil { - t.Fatal("docker run failed:", err) + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/ubuntu", + }, "/bin/sh", "-c", "echo server | nc -l -p 8080 && sleep 1"); err != nil { + t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // Finds IP address for host. - ip, err := d.Exec("/bin/sh", "-c", "cat /etc/hosts | grep ${HOSTNAME} | awk '{print $1}'") + ip, err := d.Exec(dockerutil.RunOpts{}, "/bin/sh", "-c", "cat /etc/hosts | grep ${HOSTNAME} | awk '{print $1}'") if err != nil { - t.Fatal("docker exec failed:", err) + t.Fatalf("docker exec failed: %v", err) } ip = strings.TrimRight(ip, "\n") // Runs client that sends "client" to the server and exits. - reply, err := d.Exec("/bin/sh", "-c", fmt.Sprintf("echo client | nc %s 8080", ip)) + reply, err := d.Exec(dockerutil.RunOpts{}, "/bin/sh", "-c", fmt.Sprintf("echo client | nc %s 8080", ip)) if err != nil { - t.Fatal("docker exec failed:", err) + t.Fatalf("docker exec failed: %v", err) } // Ensure both client and server got the message from each other. @@ -227,21 +231,22 @@ func TestConnectToSelf(t *testing.T) { t.Errorf("Error on server, want: %q, got: %q", want, reply) } if _, err := d.WaitForOutput("^client\n$", 1*time.Second); err != nil { - t.Fatal("docker.WaitForOutput(client) timeout:", err) + t.Fatalf("docker.WaitForOutput(client) timeout: %v", err) } } func TestMemLimit(t *testing.T) { - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatal("docker pull failed:", err) - } - d := dockerutil.MakeDocker("cgroup-test") - cmd := "cat /proc/meminfo | grep MemTotal: | awk '{print $2}'" - out, err := d.RunFg("--memory=500MB", "alpine", "sh", "-c", cmd) + d := dockerutil.MakeDocker(t) + defer d.CleanUp() + + allocMemory := 500 * 1024 + out, err := d.Run(dockerutil.RunOpts{ + Image: "basic/alpine", + Memory: allocMemory, // In kB. + }, "sh", "-c", "cat /proc/meminfo | grep MemTotal: | awk '{print $2}'") if err != nil { - t.Fatal("docker run failed:", err) + t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // Remove warning message that swap isn't present. if strings.HasPrefix(out, "WARNING") { @@ -252,27 +257,30 @@ func TestMemLimit(t *testing.T) { out = lines[1] } + // Ensure the memory matches what we want. got, err := strconv.ParseUint(strings.TrimSpace(out), 10, 64) if err != nil { t.Fatalf("failed to parse %q: %v", out, err) } - if want := uint64(500 * 1024); got != want { + if want := uint64(allocMemory); got != want { t.Errorf("MemTotal got: %d, want: %d", got, want) } } func TestNumCPU(t *testing.T) { - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatal("docker pull failed:", err) - } - d := dockerutil.MakeDocker("cgroup-test") - cmd := "cat /proc/cpuinfo | grep 'processor.*:' | wc -l" - out, err := d.RunFg("--cpuset-cpus=0", "alpine", "sh", "-c", cmd) + d := dockerutil.MakeDocker(t) + defer d.CleanUp() + + // Read how many cores are in the container. + out, err := d.Run(dockerutil.RunOpts{ + Image: "basic/alpine", + Extra: []string{"--cpuset-cpus=0"}, + }, "sh", "-c", "cat /proc/cpuinfo | grep 'processor.*:' | wc -l") if err != nil { - t.Fatal("docker run failed:", err) + t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() + // Ensure it matches what we want. got, err := strconv.Atoi(strings.TrimSpace(out)) if err != nil { t.Fatalf("failed to parse %q: %v", out, err) @@ -284,39 +292,39 @@ func TestNumCPU(t *testing.T) { // TestJobControl tests that job control characters are handled properly. func TestJobControl(t *testing.T) { - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("job-control-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() // Start the container with an attached PTY. - _, ptmx, err := d.RunWithPty("alpine", "sh") - if err != nil { + if _, err := d.Run(dockerutil.RunOpts{ + Image: "basic/alpine", + Pty: func(_ *exec.Cmd, ptmx *os.File) { + // Call "sleep 100" in the shell. + if _, err := ptmx.Write([]byte("sleep 100\n")); err != nil { + t.Fatalf("error writing to pty: %v", err) + } + + // Give shell a few seconds to start executing the sleep. + time.Sleep(2 * time.Second) + + // Send a ^C to the pty, which should kill sleep, but + // not the shell. \x03 is ASCII "end of text", which + // is the same as ^C. + if _, err := ptmx.Write([]byte{'\x03'}); err != nil { + t.Fatalf("error writing to pty: %v", err) + } + + // The shell should still be alive at this point. Sleep + // should have exited with code 2+128=130. We'll exit + // with 10 plus that number, so that we can be sure + // that the shell did not get signalled. + if _, err := ptmx.Write([]byte("exit $(expr $? + 10)\n")); err != nil { + t.Fatalf("error writing to pty: %v", err) + } + }, + }, "sh"); err != nil { t.Fatalf("docker run failed: %v", err) } - defer ptmx.Close() - defer d.CleanUp() - - // Call "sleep 100" in the shell. - if _, err := ptmx.Write([]byte("sleep 100\n")); err != nil { - t.Fatalf("error writing to pty: %v", err) - } - - // Give shell a few seconds to start executing the sleep. - time.Sleep(2 * time.Second) - - // Send a ^C to the pty, which should kill sleep, but not the shell. - // \x03 is ASCII "end of text", which is the same as ^C. - if _, err := ptmx.Write([]byte{'\x03'}); err != nil { - t.Fatalf("error writing to pty: %v", err) - } - - // The shell should still be alive at this point. Sleep should have - // exited with code 2+128=130. We'll exit with 10 plus that number, so - // that we can be sure that the shell did not get signalled. - if _, err := ptmx.Write([]byte("exit $(expr $? + 10)\n")); err != nil { - t.Fatalf("error writing to pty: %v", err) - } // Wait for the container to exit. got, err := d.Wait(5 * time.Second) @@ -332,14 +340,25 @@ func TestJobControl(t *testing.T) { // TestTmpFile checks that files inside '/tmp' are not overridden. In addition, // it checks that working dir is created if it doesn't exit. func TestTmpFile(t *testing.T) { - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatal("docker pull failed:", err) + d := dockerutil.MakeDocker(t) + defer d.CleanUp() + + // Should work without ReadOnly + if _, err := d.Run(dockerutil.RunOpts{ + Image: "basic/alpine", + WorkDir: "/tmp/foo/bar", + }, "touch", "/tmp/foo/bar/file"); err != nil { + t.Fatalf("docker run failed: %v", err) } - d := dockerutil.MakeDocker("tmp-file-test") - if err := d.Run("-w=/tmp/foo/bar", "--read-only", "alpine", "touch", "/tmp/foo/bar/file"); err != nil { - t.Fatal("docker run failed:", err) + + // Expect failure. + if _, err := d.Run(dockerutil.RunOpts{ + Image: "basic/alpine", + WorkDir: "/tmp/foo/bar", + ReadOnly: true, + }, "touch", "/tmp/foo/bar/file"); err == nil { + t.Fatalf("docker run expected failure, but succeeded") } - defer d.CleanUp() } func TestMain(m *testing.M) { diff --git a/test/e2e/regression_test.go b/test/e2e/regression_test.go index 2488be383..327a2174c 100644 --- a/test/e2e/regression_test.go +++ b/test/e2e/regression_test.go @@ -18,7 +18,7 @@ import ( "strings" "testing" - "gvisor.dev/gvisor/runsc/dockerutil" + "gvisor.dev/gvisor/pkg/test/dockerutil" ) // Test that UDS can be created using overlay when parent directory is in lower @@ -27,19 +27,19 @@ import ( // Prerequisite: the directory where the socket file is created must not have // been open for write before bind(2) is called. func TestBindOverlay(t *testing.T) { - if err := dockerutil.Pull("ubuntu:trusty"); err != nil { - t.Fatal("docker pull failed:", err) - } - d := dockerutil.MakeDocker("bind-overlay-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() - cmd := "nc -l -U /var/run/sock & p=$! && sleep 1 && echo foobar-asdf | nc -U /var/run/sock && wait $p" - got, err := d.RunFg("ubuntu:trusty", "bash", "-c", cmd) + // Run the container. + got, err := d.Run(dockerutil.RunOpts{ + Image: "basic/ubuntu", + }, "bash", "-c", "nc -l -U /var/run/sock & p=$! && sleep 1 && echo foobar-asdf | nc -U /var/run/sock && wait $p") if err != nil { - t.Fatal("docker run failed:", err) + t.Fatalf("docker run failed: %v", err) } + // Check the output contains what we want. if want := "foobar-asdf"; !strings.Contains(got, want) { t.Fatalf("docker run output is missing %q: %s", want, got) } - defer d.CleanUp() } diff --git a/test/image/BUILD b/test/image/BUILD index 7392ac54e..e749e47d4 100644 --- a/test/image/BUILD +++ b/test/image/BUILD @@ -22,8 +22,8 @@ go_test( ], visibility = ["//:sandbox"], deps = [ - "//runsc/dockerutil", - "//runsc/testutil", + "//pkg/test/dockerutil", + "//pkg/test/testutil", ], ) diff --git a/test/image/image_test.go b/test/image/image_test.go index 0a1e19d6f..2e3543109 100644 --- a/test/image/image_test.go +++ b/test/image/image_test.go @@ -28,24 +28,29 @@ import ( "log" "net/http" "os" - "path/filepath" "strings" "testing" "time" - "gvisor.dev/gvisor/runsc/dockerutil" - "gvisor.dev/gvisor/runsc/testutil" + "gvisor.dev/gvisor/pkg/test/dockerutil" + "gvisor.dev/gvisor/pkg/test/testutil" ) func TestHelloWorld(t *testing.T) { - d := dockerutil.MakeDocker("hello-test") - if err := d.Run("hello-world"); err != nil { + d := dockerutil.MakeDocker(t) + defer d.CleanUp() + + // Run the basic container. + out, err := d.Run(dockerutil.RunOpts{ + Image: "basic/alpine", + }, "echo", "Hello world!") + if err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() - if _, err := d.WaitForOutput("Hello from Docker!", 5*time.Second); err != nil { - t.Fatalf("docker didn't say hello: %v", err) + // Check the output. + if !strings.Contains(out, "Hello world!") { + t.Fatalf("docker didn't say hello: got %s", out) } } @@ -102,27 +107,22 @@ func testHTTPServer(t *testing.T, port int) { } func TestHttpd(t *testing.T) { - if err := dockerutil.Pull("httpd"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("http-test") - - dir, err := dockerutil.PrepareFiles("test/image/latin10k.txt") - if err != nil { - t.Fatalf("PrepareFiles() failed: %v", err) - } + d := dockerutil.MakeDocker(t) + defer d.CleanUp() // Start the container. - mountArg := dockerutil.MountArg(dir, "/usr/local/apache2/htdocs", dockerutil.ReadOnly) - if err := d.Run("-p", "80", mountArg, "httpd"); err != nil { + d.CopyFiles("/usr/local/apache2/htdocs", "test/image/latin10k.txt") + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/httpd", + Ports: []int{80}, + }); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // Find where port 80 is mapped to. port, err := d.FindPort(80) if err != nil { - t.Fatalf("docker.FindPort(80) failed: %v", err) + t.Fatalf("FindPort(80) failed: %v", err) } // Wait until it's up and running. @@ -134,27 +134,22 @@ func TestHttpd(t *testing.T) { } func TestNginx(t *testing.T) { - if err := dockerutil.Pull("nginx"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("net-test") - - dir, err := dockerutil.PrepareFiles("test/image/latin10k.txt") - if err != nil { - t.Fatalf("PrepareFiles() failed: %v", err) - } + d := dockerutil.MakeDocker(t) + defer d.CleanUp() // Start the container. - mountArg := dockerutil.MountArg(dir, "/usr/share/nginx/html", dockerutil.ReadOnly) - if err := d.Run("-p", "80", mountArg, "nginx"); err != nil { + d.CopyFiles("/usr/share/nginx/html", "test/image/latin10k.txt") + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/nginx", + Ports: []int{80}, + }); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // Find where port 80 is mapped to. port, err := d.FindPort(80) if err != nil { - t.Fatalf("docker.FindPort(80) failed: %v", err) + t.Fatalf("FindPort(80) failed: %v", err) } // Wait until it's up and running. @@ -166,99 +161,58 @@ func TestNginx(t *testing.T) { } func TestMysql(t *testing.T) { - if err := dockerutil.Pull("mysql"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("mysql-test") + server := dockerutil.MakeDocker(t) + defer server.CleanUp() // Start the container. - if err := d.Run("-e", "MYSQL_ROOT_PASSWORD=foobar123", "mysql"); err != nil { + if err := server.Spawn(dockerutil.RunOpts{ + Image: "basic/mysql", + Env: []string{"MYSQL_ROOT_PASSWORD=foobar123"}, + }); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // Wait until it's up and running. - if _, err := d.WaitForOutput("port: 3306 MySQL Community Server", 3*time.Minute); err != nil { - t.Fatalf("docker.WaitForOutput() timeout: %v", err) + if _, err := server.WaitForOutput("port: 3306 MySQL Community Server", 3*time.Minute); err != nil { + t.Fatalf("WaitForOutput() timeout: %v", err) } - client := dockerutil.MakeDocker("mysql-client-test") - dir, err := dockerutil.PrepareFiles("test/image/mysql.sql") - if err != nil { - t.Fatalf("PrepareFiles() failed: %v", err) - } + // Generate the client and copy in the SQL payload. + client := dockerutil.MakeDocker(t) + defer client.CleanUp() - // Tell mysql client to connect to the server and execute the file in verbose - // mode to verify the output. - args := []string{ - dockerutil.LinkArg(&d, "mysql"), - dockerutil.MountArg(dir, "/sql", dockerutil.ReadWrite), - "mysql", - "mysql", "-hmysql", "-uroot", "-pfoobar123", "-v", "-e", "source /sql/mysql.sql", - } - if err := client.Run(args...); err != nil { + // Tell mysql client to connect to the server and execute the file in + // verbose mode to verify the output. + client.CopyFiles("/sql", "test/image/mysql.sql") + client.Link("mysql", server) + if _, err := client.Run(dockerutil.RunOpts{ + Image: "basic/mysql", + }, "mysql", "-hmysql", "-uroot", "-pfoobar123", "-v", "-e", "source /sql/mysql.sql"); err != nil { t.Fatalf("docker run failed: %v", err) } - defer client.CleanUp() // Ensure file executed to the end and shutdown mysql. - if _, err := client.WaitForOutput("--------------\nshutdown\n--------------", 15*time.Second); err != nil { - t.Fatalf("docker.WaitForOutput() timeout: %v", err) - } - if _, err := d.WaitForOutput("mysqld: Shutdown complete", 30*time.Second); err != nil { - t.Fatalf("docker.WaitForOutput() timeout: %v", err) + if _, err := server.WaitForOutput("mysqld: Shutdown complete", 30*time.Second); err != nil { + t.Fatalf("WaitForOutput() timeout: %v", err) } } -func TestPythonHello(t *testing.T) { - // TODO(b/136503277): Once we have more complete python runtime tests, - // we can drop this one. - const img = "gcr.io/gvisor-presubmit/python-hello" - if err := dockerutil.Pull(img); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("python-hello-test") - if err := d.Run("-p", "8080", img); err != nil { - t.Fatalf("docker run failed: %v", err) - } +func TestTomcat(t *testing.T) { + d := dockerutil.MakeDocker(t) defer d.CleanUp() - // Find where port 8080 is mapped to. - port, err := d.FindPort(8080) - if err != nil { - t.Fatalf("docker.FindPort(8080) failed: %v", err) - } - - // Wait until it's up and running. - if err := testutil.WaitForHTTP(port, 30*time.Second); err != nil { - t.Fatalf("WaitForHTTP() timeout: %v", err) - } - - // Ensure that content is being served. - url := fmt.Sprintf("http://localhost:%d", port) - resp, err := http.Get(url) - if err != nil { - t.Errorf("Error reaching http server: %v", err) - } - if want := http.StatusOK; resp.StatusCode != want { - t.Errorf("Wrong response code, got: %d, want: %d", resp.StatusCode, want) - } -} - -func TestTomcat(t *testing.T) { - if err := dockerutil.Pull("tomcat:8.0"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("tomcat-test") - if err := d.Run("-p", "8080", "tomcat:8.0"); err != nil { + // Start the server. + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/tomcat", + Ports: []int{8080}, + }); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // Find where port 8080 is mapped to. port, err := d.FindPort(8080) if err != nil { - t.Fatalf("docker.FindPort(8080) failed: %v", err) + t.Fatalf("FindPort(8080) failed: %v", err) } // Wait until it's up and running. @@ -278,28 +232,22 @@ func TestTomcat(t *testing.T) { } func TestRuby(t *testing.T) { - if err := dockerutil.Pull("ruby"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("ruby-test") - - dir, err := dockerutil.PrepareFiles("test/image/ruby.rb", "test/image/ruby.sh") - if err != nil { - t.Fatalf("PrepareFiles() failed: %v", err) - } - if err := os.Chmod(filepath.Join(dir, "ruby.sh"), 0333); err != nil { - t.Fatalf("os.Chmod(%q, 0333) failed: %v", dir, err) - } + d := dockerutil.MakeDocker(t) + defer d.CleanUp() - if err := d.Run("-p", "8080", dockerutil.MountArg(dir, "/src", dockerutil.ReadOnly), "ruby", "/src/ruby.sh"); err != nil { + // Execute the ruby workload. + d.CopyFiles("/src", "test/image/ruby.rb", "test/image/ruby.sh") + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/ruby", + Ports: []int{8080}, + }, "/src/ruby.sh"); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // Find where port 8080 is mapped to. port, err := d.FindPort(8080) if err != nil { - t.Fatalf("docker.FindPort(8080) failed: %v", err) + t.Fatalf("FindPort(8080) failed: %v", err) } // Wait until it's up and running, 'gem install' can take some time. @@ -326,18 +274,17 @@ func TestRuby(t *testing.T) { } func TestStdio(t *testing.T) { - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatalf("docker pull failed: %v", err) - } - d := dockerutil.MakeDocker("stdio-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() wantStdout := "hello stdout" wantStderr := "bonjour stderr" cmd := fmt.Sprintf("echo %q; echo %q 1>&2;", wantStdout, wantStderr) - if err := d.Run("alpine", "/bin/sh", "-c", cmd); err != nil { + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/alpine", + }, "/bin/sh", "-c", cmd); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() for _, want := range []string{wantStdout, wantStderr} { if _, err := d.WaitForOutput(want, 5*time.Second); err != nil { diff --git a/test/image/ruby.sh b/test/image/ruby.sh old mode 100644 new mode 100755 diff --git a/test/iptables/BUILD b/test/iptables/BUILD index 6bb3b82b5..3e29ca90d 100644 --- a/test/iptables/BUILD +++ b/test/iptables/BUILD @@ -14,7 +14,7 @@ go_library( ], visibility = ["//test/iptables:__subpackages__"], deps = [ - "//runsc/testutil", + "//pkg/test/testutil", ], ) @@ -23,14 +23,14 @@ go_test( srcs = [ "iptables_test.go", ], + data = ["//test/iptables/runner"], library = ":iptables", tags = [ "local", "manual", ], deps = [ - "//pkg/log", - "//runsc/dockerutil", - "//runsc/testutil", + "//pkg/test/dockerutil", + "//pkg/test/testutil", ], ) diff --git a/test/iptables/README.md b/test/iptables/README.md index cc8a2fcac..b9f44bd40 100644 --- a/test/iptables/README.md +++ b/test/iptables/README.md @@ -38,7 +38,7 @@ Build the testing Docker container. Re-run this when you modify the test code in this directory: ```bash -$ bazel run //test/iptables/runner:runner-image -- --norun +$ make load-iptables ``` Run an individual test via: diff --git a/test/iptables/iptables.go b/test/iptables/iptables.go index 2e565d988..16cb4f4da 100644 --- a/test/iptables/iptables.go +++ b/test/iptables/iptables.go @@ -18,12 +18,19 @@ package iptables import ( "fmt" "net" + "time" ) // IPExchangePort is the port the container listens on to receive the IP // address of the local process. const IPExchangePort = 2349 +// TerminalStatement is the last statement in the test runner. +const TerminalStatement = "Finished!" + +// TestTimeout is the timeout used for all tests. +const TestTimeout = 10 * time.Minute + // A TestCase contains one action to run in the container and one to run // locally. The actions run concurrently and each must succeed for the test // pass. diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go index 493d69052..334d8e676 100644 --- a/test/iptables/iptables_test.go +++ b/test/iptables/iptables_test.go @@ -15,28 +15,14 @@ package iptables import ( - "flag" "fmt" "net" - "os" - "path" "testing" - "time" - "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/runsc/dockerutil" - "gvisor.dev/gvisor/runsc/testutil" + "gvisor.dev/gvisor/pkg/test/dockerutil" + "gvisor.dev/gvisor/pkg/test/testutil" ) -const timeout = 18 * time.Second - -var image = flag.String("image", "bazel/test/iptables/runner:runner-image", "image to run tests in") - -type result struct { - output string - err error -} - // singleTest runs a TestCase. Each test follows a pattern: // - Create a container. // - Get the container's IP. @@ -46,77 +32,45 @@ type result struct { // // Container output is logged to $TEST_UNDECLARED_OUTPUTS_DIR if it exists, or // to stderr. -func singleTest(test TestCase) error { +func singleTest(t *testing.T, test TestCase) { if _, ok := Tests[test.Name()]; !ok { - return fmt.Errorf("no test found with name %q. Has it been registered?", test.Name()) + t.Fatalf("no test found with name %q. Has it been registered?", test.Name()) } + d := dockerutil.MakeDocker(t) + defer d.CleanUp() + // Create and start the container. - cont := dockerutil.MakeDocker("gvisor-iptables") - defer cont.CleanUp() - resultChan := make(chan *result) - go func() { - output, err := cont.RunFg("--cap-add=NET_ADMIN", *image, "-name", test.Name()) - logContainer(output, err) - resultChan <- &result{output, err} - }() + d.CopyFiles("/runner", "test/iptables/runner/runner") + if err := d.Spawn(dockerutil.RunOpts{ + Image: "iptables", + CapAdd: []string{"NET_ADMIN"}, + }, "/runner/runner", "-name", test.Name()); err != nil { + t.Fatalf("docker run failed: %v", err) + } // Get the container IP. - ip, err := getIP(cont) + ip, err := d.FindIP() if err != nil { - return fmt.Errorf("failed to get container IP: %v", err) + t.Fatalf("failed to get container IP: %v", err) } // Give the container our IP. if err := sendIP(ip); err != nil { - return fmt.Errorf("failed to send IP to container: %v", err) + t.Fatalf("failed to send IP to container: %v", err) } // Run our side of the test. - errChan := make(chan error) - go func() { - errChan <- test.LocalAction(ip) - }() - - // Wait for both the container and local tests to finish. - var res *result - to := time.After(timeout) - for localDone := false; res == nil || !localDone; { - select { - case res = <-resultChan: - log.Infof("Container finished.") - case err, localDone = <-errChan: - log.Infof("Local finished.") - if err != nil { - return fmt.Errorf("local test failed: %v", err) - } - case <-to: - return fmt.Errorf("timed out after %f seconds", timeout.Seconds()) - } + if err := test.LocalAction(ip); err != nil { + t.Fatalf("LocalAction failed: %v", err) } - return res.err -} - -func getIP(cont dockerutil.Docker) (net.IP, error) { - // The container might not have started yet, so retry a few times. - var ipStr string - to := time.After(timeout) - for ipStr == "" { - ipStr, _ = cont.FindIP() - select { - case <-to: - return net.IP{}, fmt.Errorf("timed out getting IP after %f seconds", timeout.Seconds()) - default: - time.Sleep(250 * time.Millisecond) - } - } - ip := net.ParseIP(ipStr) - if ip == nil { - return net.IP{}, fmt.Errorf("invalid IP: %q", ipStr) + // Wait for the final statement. This structure has the side effect + // that all container logs will appear within the individual test + // context. + if _, err := d.WaitForOutput(TerminalStatement, TestTimeout); err != nil { + t.Fatalf("test failed: %v", err) } - log.Infof("Container has IP of %s", ipStr) - return ip, nil } func sendIP(ip net.IP) error { @@ -132,7 +86,7 @@ func sendIP(ip net.IP) error { conn = c return err } - if err := testutil.Poll(cb, timeout); err != nil { + if err := testutil.Poll(cb, TestTimeout); err != nil { return fmt.Errorf("timed out waiting to send IP, most recent error: %v", err) } if _, err := conn.Write([]byte{0}); err != nil { @@ -141,281 +95,184 @@ func sendIP(ip net.IP) error { return nil } -func logContainer(output string, err error) { - msg := fmt.Sprintf("Container error: %v\nContainer output:\n%v", err, output) - if artifactsDir := os.Getenv("TEST_UNDECLARED_OUTPUTS_DIR"); artifactsDir != "" { - fpath := path.Join(artifactsDir, "container.log") - if file, err := os.OpenFile(fpath, os.O_WRONLY|os.O_CREATE, 0644); err != nil { - log.Warningf("Failed to open log file %q: %v", fpath, err) - } else { - defer file.Close() - if _, err := file.Write([]byte(msg)); err == nil { - return - } - log.Warningf("Failed to write to log file %s: %v", fpath, err) - } - } - - // We couldn't write to the output directory -- just log to stderr. - log.Infof(msg) -} - func TestFilterInputDropUDP(t *testing.T) { - if err := singleTest(FilterInputDropUDP{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputDropUDP{}) } func TestFilterInputDropUDPPort(t *testing.T) { - if err := singleTest(FilterInputDropUDPPort{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputDropUDPPort{}) } func TestFilterInputDropDifferentUDPPort(t *testing.T) { - if err := singleTest(FilterInputDropDifferentUDPPort{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputDropDifferentUDPPort{}) } func TestFilterInputDropAll(t *testing.T) { - if err := singleTest(FilterInputDropAll{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputDropAll{}) } func TestFilterInputDropOnlyUDP(t *testing.T) { - if err := singleTest(FilterInputDropOnlyUDP{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputDropOnlyUDP{}) } func TestNATRedirectUDPPort(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(NATRedirectUDPPort{}); err != nil { - t.Fatal(err) - } + singleTest(t, NATRedirectUDPPort{}) } func TestNATRedirectTCPPort(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(NATRedirectTCPPort{}); err != nil { - t.Fatal(err) - } + singleTest(t, NATRedirectTCPPort{}) } func TestNATDropUDP(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(NATDropUDP{}); err != nil { - t.Fatal(err) - } + singleTest(t, NATDropUDP{}) } func TestNATAcceptAll(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(NATAcceptAll{}); err != nil { - t.Fatal(err) - } + singleTest(t, NATAcceptAll{}) } func TestFilterInputDropTCPDestPort(t *testing.T) { - if err := singleTest(FilterInputDropTCPDestPort{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputDropTCPDestPort{}) } func TestFilterInputDropTCPSrcPort(t *testing.T) { - if err := singleTest(FilterInputDropTCPSrcPort{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputDropTCPSrcPort{}) } func TestFilterInputCreateUserChain(t *testing.T) { - if err := singleTest(FilterInputCreateUserChain{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputCreateUserChain{}) } func TestFilterInputDefaultPolicyAccept(t *testing.T) { - if err := singleTest(FilterInputDefaultPolicyAccept{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputDefaultPolicyAccept{}) } func TestFilterInputDefaultPolicyDrop(t *testing.T) { - if err := singleTest(FilterInputDefaultPolicyDrop{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputDefaultPolicyDrop{}) } func TestFilterInputReturnUnderflow(t *testing.T) { - if err := singleTest(FilterInputReturnUnderflow{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputReturnUnderflow{}) } func TestFilterOutputDropTCPDestPort(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("filter OUTPUT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(FilterOutputDropTCPDestPort{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterOutputDropTCPDestPort{}) } func TestFilterOutputDropTCPSrcPort(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("filter OUTPUT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(FilterOutputDropTCPSrcPort{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterOutputDropTCPSrcPort{}) } func TestFilterOutputAcceptTCPOwner(t *testing.T) { - if err := singleTest(FilterOutputAcceptTCPOwner{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterOutputAcceptTCPOwner{}) } func TestFilterOutputDropTCPOwner(t *testing.T) { - if err := singleTest(FilterOutputDropTCPOwner{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterOutputDropTCPOwner{}) } func TestFilterOutputAcceptUDPOwner(t *testing.T) { - if err := singleTest(FilterOutputAcceptUDPOwner{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterOutputAcceptUDPOwner{}) } func TestFilterOutputDropUDPOwner(t *testing.T) { - if err := singleTest(FilterOutputDropUDPOwner{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterOutputDropUDPOwner{}) } func TestFilterOutputOwnerFail(t *testing.T) { - if err := singleTest(FilterOutputOwnerFail{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterOutputOwnerFail{}) } func TestJumpSerialize(t *testing.T) { - if err := singleTest(FilterInputSerializeJump{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputSerializeJump{}) } func TestJumpBasic(t *testing.T) { - if err := singleTest(FilterInputJumpBasic{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputJumpBasic{}) } func TestJumpReturn(t *testing.T) { - if err := singleTest(FilterInputJumpReturn{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputJumpReturn{}) } func TestJumpReturnDrop(t *testing.T) { - if err := singleTest(FilterInputJumpReturnDrop{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputJumpReturnDrop{}) } func TestJumpBuiltin(t *testing.T) { - if err := singleTest(FilterInputJumpBuiltin{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputJumpBuiltin{}) } func TestJumpTwice(t *testing.T) { - if err := singleTest(FilterInputJumpTwice{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputJumpTwice{}) } func TestInputDestination(t *testing.T) { - if err := singleTest(FilterInputDestination{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputDestination{}) } func TestInputInvertDestination(t *testing.T) { - if err := singleTest(FilterInputInvertDestination{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterInputInvertDestination{}) } func TestOutputDestination(t *testing.T) { - if err := singleTest(FilterOutputDestination{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterOutputDestination{}) } func TestOutputInvertDestination(t *testing.T) { - if err := singleTest(FilterOutputInvertDestination{}); err != nil { - t.Fatal(err) - } + singleTest(t, FilterOutputInvertDestination{}) } func TestNATOutRedirectIP(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(NATOutRedirectIP{}); err != nil { - t.Fatal(err) - } + singleTest(t, NATOutRedirectIP{}) } func TestNATOutDontRedirectIP(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(NATOutDontRedirectIP{}); err != nil { - t.Fatal(err) - } + singleTest(t, NATOutDontRedirectIP{}) } func TestNATOutRedirectInvert(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(NATOutRedirectInvert{}); err != nil { - t.Fatal(err) - } + singleTest(t, NATOutRedirectInvert{}) } func TestNATPreRedirectIP(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(NATPreRedirectIP{}); err != nil { - t.Fatal(err) - } + singleTest(t, NATPreRedirectIP{}) } func TestNATPreDontRedirectIP(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(NATPreDontRedirectIP{}); err != nil { - t.Fatal(err) - } + singleTest(t, NATPreDontRedirectIP{}) } func TestNATPreRedirectInvert(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(NATPreRedirectInvert{}); err != nil { - t.Fatal(err) - } + singleTest(t, NATPreRedirectInvert{}) } func TestNATRedirectRequiresProtocol(t *testing.T) { // TODO(gvisor.dev/issue/170): Enable when supported. t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - if err := singleTest(NATRedirectRequiresProtocol{}); err != nil { - t.Fatal(err) - } + singleTest(t, NATRedirectRequiresProtocol{}) } diff --git a/test/iptables/iptables_util.go b/test/iptables/iptables_util.go index 134391e8d..2a00677be 100644 --- a/test/iptables/iptables_util.go +++ b/test/iptables/iptables_util.go @@ -20,7 +20,7 @@ import ( "os/exec" "time" - "gvisor.dev/gvisor/runsc/testutil" + "gvisor.dev/gvisor/pkg/test/testutil" ) const iptablesBinary = "iptables" diff --git a/test/iptables/runner/BUILD b/test/iptables/runner/BUILD index b9199387a..24504a1b9 100644 --- a/test/iptables/runner/BUILD +++ b/test/iptables/runner/BUILD @@ -1,4 +1,4 @@ -load("//tools:defs.bzl", "container_image", "go_binary", "go_image") +load("//tools:defs.bzl", "go_binary") package(licenses = ["notice"]) @@ -6,18 +6,7 @@ go_binary( name = "runner", testonly = 1, srcs = ["main.go"], - deps = ["//test/iptables"], -) - -container_image( - name = "iptables-base", - base = "@iptables-test//image", -) - -go_image( - name = "runner-image", - testonly = 1, - srcs = ["main.go"], - base = ":iptables-base", + pure = True, + visibility = ["//test/iptables:__subpackages__"], deps = ["//test/iptables"], ) diff --git a/test/iptables/runner/main.go b/test/iptables/runner/main.go index 3c794114e..6f77c0684 100644 --- a/test/iptables/runner/main.go +++ b/test/iptables/runner/main.go @@ -46,6 +46,9 @@ func main() { if err := test.ContainerAction(ip); err != nil { log.Fatalf("Failed running test %q: %v", *name, err) } + + // Emit the final line. + log.Printf("%s", iptables.TerminalStatement) } // getIP listens for a connection from the local process and returns the source diff --git a/test/packetdrill/packetdrill_test.sh b/test/packetdrill/packetdrill_test.sh index c8268170f..922547d65 100755 --- a/test/packetdrill/packetdrill_test.sh +++ b/test/packetdrill/packetdrill_test.sh @@ -85,23 +85,26 @@ if [[ ! -x "${INIT_SCRIPT-}" ]]; then exit 2 fi +function new_net_prefix() { + # Class C, 192.0.0.0 to 223.255.255.255, transitionally has mask 24. + echo "$(shuf -i 192-223 -n 1).$(shuf -i 0-255 -n 1).$(shuf -i 0-255 -n 1)" +} + # Variables specific to the control network and interface start with CTRL_. # Variables specific to the test network and interface start with TEST_. # Variables specific to the DUT start with DUT_. # Variables specific to the test runner start with TEST_RUNNER_. declare -r PACKETDRILL="/packetdrill/gtests/net/packetdrill/packetdrill" # Use random numbers so that test networks don't collide. -declare -r CTRL_NET="ctrl_net-$(shuf -i 0-99999999 -n 1)" -declare -r TEST_NET="test_net-$(shuf -i 0-99999999 -n 1)" +declare CTRL_NET="ctrl_net-$(shuf -i 0-99999999 -n 1)" +declare CTRL_NET_PREFIX=$(new_net_prefix) +declare TEST_NET="test_net-$(shuf -i 0-99999999 -n 1)" +declare TEST_NET_PREFIX=$(new_net_prefix) declare -r tolerance_usecs=100000 # On both DUT and test runner, testing packets are on the eth2 interface. declare -r TEST_DEVICE="eth2" # Number of bits in the *_NET_PREFIX variables. declare -r NET_MASK="24" -function new_net_prefix() { - # Class C, 192.0.0.0 to 223.255.255.255, transitionally has mask 24. - echo "$(shuf -i 192-223 -n 1).$(shuf -i 0-255 -n 1).$(shuf -i 0-255 -n 1)" -} # Last bits of the DUT's IP address. declare -r DUT_NET_SUFFIX=".10" # Control port. @@ -137,23 +140,21 @@ function finish { trap finish EXIT # Subnet for control packets between test runner and DUT. -declare CTRL_NET_PREFIX=$(new_net_prefix) while ! docker network create \ "--subnet=${CTRL_NET_PREFIX}.0/${NET_MASK}" "${CTRL_NET}"; do sleep 0.1 - declare CTRL_NET_PREFIX=$(new_net_prefix) + CTRL_NET_PREFIX=$(new_net_prefix) + CTRL_NET="ctrl_net-$(shuf -i 0-99999999 -n 1)" done # Subnet for the packets that are part of the test. -declare TEST_NET_PREFIX=$(new_net_prefix) while ! docker network create \ "--subnet=${TEST_NET_PREFIX}.0/${NET_MASK}" "${TEST_NET}"; do sleep 0.1 - declare TEST_NET_PREFIX=$(new_net_prefix) + TEST_NET_PREFIX=$(new_net_prefix) + TEST_NET="test_net-$(shuf -i 0-99999999 -n 1)" done -docker pull "${IMAGE_TAG}" - # Create the DUT container and connect to network. DUT=$(docker create ${RUNTIME_ARG} --privileged --rm \ --stop-timeout ${TIMEOUT} -it ${IMAGE_TAG}) diff --git a/test/packetimpact/testbench/dut.go b/test/packetimpact/testbench/dut.go index 9335909c0..3f340c6bc 100644 --- a/test/packetimpact/testbench/dut.go +++ b/test/packetimpact/testbench/dut.go @@ -132,7 +132,7 @@ func (dut *DUT) CreateBoundSocket(typ, proto int32, addr net.IP) (int32, uint16) copy(sa.Addr[:], addr.To16()) dut.Bind(fd, &sa) } else { - dut.t.Fatal("unknown ip addr type for remoteIP") + dut.t.Fatalf("unknown ip addr type for remoteIP") } sa := dut.GetSockName(fd) var port int diff --git a/test/packetimpact/tests/test_runner.sh b/test/packetimpact/tests/test_runner.sh index 2be3c17c3..46d63d5e5 100755 --- a/test/packetimpact/tests/test_runner.sh +++ b/test/packetimpact/tests/test_runner.sh @@ -107,21 +107,24 @@ if [[ ! -f "${TESTBENCH_BINARY-}" ]]; then exit 2 fi +function new_net_prefix() { + # Class C, 192.0.0.0 to 223.255.255.255, transitionally has mask 24. + echo "$(shuf -i 192-223 -n 1).$(shuf -i 0-255 -n 1).$(shuf -i 0-255 -n 1)" +} + # Variables specific to the control network and interface start with CTRL_. # Variables specific to the test network and interface start with TEST_. # Variables specific to the DUT start with DUT_. # Variables specific to the test bench start with TESTBENCH_. # Use random numbers so that test networks don't collide. -declare -r CTRL_NET="ctrl_net-${RANDOM}${RANDOM}" -declare -r TEST_NET="test_net-${RANDOM}${RANDOM}" +declare CTRL_NET="ctrl_net-${RANDOM}${RANDOM}" +declare CTRL_NET_PREFIX=$(new_net_prefix) +declare TEST_NET="test_net-${RANDOM}${RANDOM}" +declare TEST_NET_PREFIX=$(new_net_prefix) # On both DUT and test bench, testing packets are on the eth2 interface. declare -r TEST_DEVICE="eth2" # Number of bits in the *_NET_PREFIX variables. declare -r NET_MASK="24" -function new_net_prefix() { - # Class C, 192.0.0.0 to 223.255.255.255, transitionally has mask 24. - echo "$(shuf -i 192-223 -n 1).$(shuf -i 0-255 -n 1).$(shuf -i 0-255 -n 1)" -} # Last bits of the DUT's IP address. declare -r DUT_NET_SUFFIX=".10" # Control port. @@ -130,6 +133,7 @@ declare -r CTRL_PORT="40000" declare -r TESTBENCH_NET_SUFFIX=".20" declare -r TIMEOUT="60" declare -r IMAGE_TAG="gcr.io/gvisor-presubmit/packetimpact" + # Make sure that docker is installed. docker --version @@ -169,19 +173,19 @@ function finish { trap finish EXIT # Subnet for control packets between test bench and DUT. -declare CTRL_NET_PREFIX=$(new_net_prefix) while ! docker network create \ "--subnet=${CTRL_NET_PREFIX}.0/${NET_MASK}" "${CTRL_NET}"; do sleep 0.1 - declare CTRL_NET_PREFIX=$(new_net_prefix) + CTRL_NET_PREFIX=$(new_net_prefix) + CTRL_NET="ctrl_net-${RANDOM}${RANDOM}" done # Subnet for the packets that are part of the test. -declare TEST_NET_PREFIX=$(new_net_prefix) while ! docker network create \ "--subnet=${TEST_NET_PREFIX}.0/${NET_MASK}" "${TEST_NET}"; do sleep 0.1 - declare TEST_NET_PREFIX=$(new_net_prefix) + TEST_NET_PREFIX=$(new_net_prefix) + TEST_NET="test_net-${RANDOM}${RANDOM}" done docker pull "${IMAGE_TAG}" diff --git a/test/root/BUILD b/test/root/BUILD index 05166673a..17e51e66e 100644 --- a/test/root/BUILD +++ b/test/root/BUILD @@ -33,14 +33,12 @@ go_test( ], visibility = ["//:sandbox"], deps = [ - "//runsc/boot", + "//pkg/test/criutil", + "//pkg/test/dockerutil", + "//pkg/test/testutil", "//runsc/cgroup", "//runsc/container", - "//runsc/criutil", - "//runsc/dockerutil", "//runsc/specutils", - "//runsc/testutil", - "//test/root/testdata", "@com_github_cenkalti_backoff//:go_default_library", "@com_github_opencontainers_runtime-spec//specs-go:go_default_library", "@com_github_syndtr_gocapability//capability:go_default_library", diff --git a/test/root/cgroup_test.go b/test/root/cgroup_test.go index 679342def..8876d0d61 100644 --- a/test/root/cgroup_test.go +++ b/test/root/cgroup_test.go @@ -26,9 +26,9 @@ import ( "testing" "time" + "gvisor.dev/gvisor/pkg/test/dockerutil" + "gvisor.dev/gvisor/pkg/test/testutil" "gvisor.dev/gvisor/runsc/cgroup" - "gvisor.dev/gvisor/runsc/dockerutil" - "gvisor.dev/gvisor/runsc/testutil" ) func verifyPid(pid int, path string) error { @@ -56,54 +56,70 @@ func verifyPid(pid int, path string) error { return fmt.Errorf("got: %v, want: %d", gots, pid) } -// TestCgroup sets cgroup options and checks that cgroup was properly configured. func TestMemCGroup(t *testing.T) { - allocMemSize := 128 << 20 - if err := dockerutil.Pull("python"); err != nil { - t.Fatal("docker pull failed:", err) - } - d := dockerutil.MakeDocker("memusage-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() // Start a new container and allocate the specified about of memory. - args := []string{ - "--memory=256MB", - "python", - "python", - "-c", - fmt.Sprintf("import time; s = 'a' * %d; time.sleep(100)", allocMemSize), - } - if err := d.Run(args...); err != nil { - t.Fatal("docker create failed:", err) + allocMemSize := 128 << 20 + allocMemLimit := 2 * allocMemSize + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/python", + Memory: allocMemLimit / 1024, // Must be in Kb. + }, "python", "-c", fmt.Sprintf("import time; s = 'a' * %d; time.sleep(100)", allocMemSize)); err != nil { + t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() + // Extract the ID to lookup the cgroup. gid, err := d.ID() if err != nil { t.Fatalf("Docker.ID() failed: %v", err) } t.Logf("cgroup ID: %s", gid) - path := filepath.Join("/sys/fs/cgroup/memory/docker", gid, "memory.usage_in_bytes") - memUsage := 0 - // Wait when the container will allocate memory. + memUsage := 0 start := time.Now() - for time.Now().Sub(start) < 30*time.Second { + for time.Since(start) < 30*time.Second { + // Sleep for a brief period of time after spawning the + // container (so that Docker can create the cgroup etc. + // or after looping below (so the application can start). + time.Sleep(100 * time.Millisecond) + + // Read the cgroup memory limit. + path := filepath.Join("/sys/fs/cgroup/memory/docker", gid, "memory.limit_in_bytes") outRaw, err := ioutil.ReadFile(path) if err != nil { - t.Fatalf("failed to read %q: %v", path, err) + // It's possible that the container does not exist yet. + continue } out := strings.TrimSpace(string(outRaw)) + memLimit, err := strconv.Atoi(out) + if err != nil { + t.Fatalf("Atoi(%v): %v", out, err) + } + if memLimit != allocMemLimit { + // The group may not have had the correct limit set yet. + continue + } + + // Read the cgroup memory usage. + path = filepath.Join("/sys/fs/cgroup/memory/docker", gid, "memory.max_usage_in_bytes") + outRaw, err = ioutil.ReadFile(path) + if err != nil { + t.Fatalf("error reading usage: %v", err) + } + out = strings.TrimSpace(string(outRaw)) memUsage, err = strconv.Atoi(out) if err != nil { t.Fatalf("Atoi(%v): %v", out, err) } + t.Logf("read usage: %v, wanted: %v", memUsage, allocMemSize) - if memUsage > allocMemSize { + // Are we done? + if memUsage >= allocMemSize { return } - - time.Sleep(100 * time.Millisecond) } t.Fatalf("%vMB is less than %vMB", memUsage>>20, allocMemSize>>20) @@ -111,10 +127,8 @@ func TestMemCGroup(t *testing.T) { // TestCgroup sets cgroup options and checks that cgroup was properly configured. func TestCgroup(t *testing.T) { - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatal("docker pull failed:", err) - } - d := dockerutil.MakeDocker("cgroup-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() // This is not a comprehensive list of attributes. // @@ -179,10 +193,11 @@ func TestCgroup(t *testing.T) { want: "5", }, { - arg: "--blkio-weight=750", - ctrl: "blkio", - file: "blkio.weight", - want: "750", + arg: "--blkio-weight=750", + ctrl: "blkio", + file: "blkio.weight", + want: "750", + skipIfNotFound: true, // blkio groups may not be available. }, } @@ -191,12 +206,15 @@ func TestCgroup(t *testing.T) { args = append(args, attr.arg) } - args = append(args, "alpine", "sleep", "10000") - if err := d.Run(args...); err != nil { - t.Fatal("docker create failed:", err) + // Start the container. + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/alpine", + Extra: args, // Cgroup arguments. + }, "sleep", "10000"); err != nil { + t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() + // Lookup the relevant cgroup ID. gid, err := d.ID() if err != nil { t.Fatalf("Docker.ID() failed: %v", err) @@ -245,17 +263,21 @@ func TestCgroup(t *testing.T) { } } +// TestCgroup sets cgroup options and checks that cgroup was properly configured. func TestCgroupParent(t *testing.T) { - if err := dockerutil.Pull("alpine"); err != nil { - t.Fatal("docker pull failed:", err) - } - d := dockerutil.MakeDocker("cgroup-test") + d := dockerutil.MakeDocker(t) + defer d.CleanUp() - parent := testutil.RandomName("runsc") - if err := d.Run("--cgroup-parent", parent, "alpine", "sleep", "10000"); err != nil { - t.Fatal("docker create failed:", err) + // Construct a known cgroup name. + parent := testutil.RandomID("runsc-") + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/alpine", + Extra: []string{fmt.Sprintf("--cgroup-parent=%s", parent)}, + }, "sleep", "10000"); err != nil { + t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() + + // Extract the ID to look up the cgroup. gid, err := d.ID() if err != nil { t.Fatalf("Docker.ID() failed: %v", err) diff --git a/test/root/chroot_test.go b/test/root/chroot_test.go index be0f63d18..a306132a4 100644 --- a/test/root/chroot_test.go +++ b/test/root/chroot_test.go @@ -24,17 +24,20 @@ import ( "strings" "testing" - "gvisor.dev/gvisor/runsc/dockerutil" + "gvisor.dev/gvisor/pkg/test/dockerutil" ) // TestChroot verifies that the sandbox is chroot'd and that mounts are cleaned // up after the sandbox is destroyed. func TestChroot(t *testing.T) { - d := dockerutil.MakeDocker("chroot-test") - if err := d.Run("alpine", "sleep", "10000"); err != nil { + d := dockerutil.MakeDocker(t) + defer d.CleanUp() + + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/alpine", + }, "sleep", "10000"); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() pid, err := d.SandboxPid() if err != nil { @@ -76,11 +79,14 @@ func TestChroot(t *testing.T) { } func TestChrootGofer(t *testing.T) { - d := dockerutil.MakeDocker("chroot-test") - if err := d.Run("alpine", "sleep", "10000"); err != nil { + d := dockerutil.MakeDocker(t) + defer d.CleanUp() + + if err := d.Spawn(dockerutil.RunOpts{ + Image: "basic/alpine", + }, "sleep", "10000"); err != nil { t.Fatalf("docker run failed: %v", err) } - defer d.CleanUp() // It's tricky to find gofers. Get sandbox PID first, then find parent. From // parent get all immediate children, remove the sandbox, and everything else diff --git a/test/root/crictl_test.go b/test/root/crictl_test.go index 3f90c4c6a..85007dcce 100644 --- a/test/root/crictl_test.go +++ b/test/root/crictl_test.go @@ -16,6 +16,7 @@ package root import ( "bytes" + "encoding/json" "fmt" "io" "io/ioutil" @@ -29,16 +30,58 @@ import ( "testing" "time" - "gvisor.dev/gvisor/runsc/criutil" - "gvisor.dev/gvisor/runsc/dockerutil" + "gvisor.dev/gvisor/pkg/test/criutil" + "gvisor.dev/gvisor/pkg/test/dockerutil" + "gvisor.dev/gvisor/pkg/test/testutil" "gvisor.dev/gvisor/runsc/specutils" - "gvisor.dev/gvisor/runsc/testutil" - "gvisor.dev/gvisor/test/root/testdata" ) // Tests for crictl have to be run as root (rather than in a user namespace) // because crictl creates named network namespaces in /var/run/netns/. +// SimpleSpec returns a JSON config for a simple container that runs the +// specified command in the specified image. +func SimpleSpec(name, image string, cmd []string, extra map[string]interface{}) string { + s := map[string]interface{}{ + "metadata": map[string]string{ + "name": name, + }, + "image": map[string]string{ + "image": testutil.ImageByName(image), + }, + "log_path": fmt.Sprintf("%s.log", name), + } + if len(cmd) > 0 { // Omit if empty. + s["command"] = cmd + } + for k, v := range extra { + s[k] = v // Extra settings. + } + v, err := json.Marshal(s) + if err != nil { + // This shouldn't happen. + panic(err) + } + return string(v) +} + +// Sandbox is a default JSON config for a sandbox. +var Sandbox = `{ + "metadata": { + "name": "default-sandbox", + "namespace": "default", + "attempt": 1, + "uid": "hdishd83djaidwnduwk28bcsb" + }, + "linux": { + }, + "log_directory": "/tmp" +} +` + +// Httpd is a JSON config for an httpd container. +var Httpd = SimpleSpec("httpd", "basic/httpd", nil, nil) + // TestCrictlSanity refers to b/112433158. func TestCrictlSanity(t *testing.T) { // Setup containerd and crictl. @@ -47,9 +90,9 @@ func TestCrictlSanity(t *testing.T) { t.Fatalf("failed to setup crictl: %v", err) } defer cleanup() - podID, contID, err := crictl.StartPodAndContainer("httpd", testdata.Sandbox, testdata.Httpd) + podID, contID, err := crictl.StartPodAndContainer("basic/httpd", Sandbox, Httpd) if err != nil { - t.Fatal(err) + t.Fatalf("start failed: %v", err) } // Look for the httpd page. @@ -59,10 +102,38 @@ func TestCrictlSanity(t *testing.T) { // Stop everything. if err := crictl.StopPodAndContainer(podID, contID); err != nil { - t.Fatal(err) + t.Fatalf("stop failed: %v", err) } } +// HttpdMountPaths is a JSON config for an httpd container with additional +// mounts. +var HttpdMountPaths = SimpleSpec("httpd", "basic/httpd", nil, map[string]interface{}{ + "mounts": []map[string]interface{}{ + map[string]interface{}{ + "container_path": "/var/run/secrets/kubernetes.io/serviceaccount", + "host_path": "/var/lib/kubelet/pods/82bae206-cdf5-11e8-b245-8cdcd43ac064/volumes/kubernetes.io~secret/default-token-2rpfx", + "readonly": true, + }, + map[string]interface{}{ + "container_path": "/etc/hosts", + "host_path": "/var/lib/kubelet/pods/82bae206-cdf5-11e8-b245-8cdcd43ac064/etc-hosts", + "readonly": false, + }, + map[string]interface{}{ + "container_path": "/dev/termination-log", + "host_path": "/var/lib/kubelet/pods/82bae206-cdf5-11e8-b245-8cdcd43ac064/containers/httpd/d1709580", + "readonly": false, + }, + map[string]interface{}{ + "container_path": "/usr/local/apache2/htdocs/test", + "host_path": "/var/lib/kubelet/pods/82bae206-cdf5-11e8-b245-8cdcd43ac064", + "readonly": true, + }, + }, + "linux": map[string]interface{}{}, +}) + // TestMountPaths refers to b/117635704. func TestMountPaths(t *testing.T) { // Setup containerd and crictl. @@ -71,9 +142,9 @@ func TestMountPaths(t *testing.T) { t.Fatalf("failed to setup crictl: %v", err) } defer cleanup() - podID, contID, err := crictl.StartPodAndContainer("httpd", testdata.Sandbox, testdata.HttpdMountPaths) + podID, contID, err := crictl.StartPodAndContainer("basic/httpd", Sandbox, HttpdMountPaths) if err != nil { - t.Fatal(err) + t.Fatalf("start failed: %v", err) } // Look for the directory available at /test. @@ -83,7 +154,7 @@ func TestMountPaths(t *testing.T) { // Stop everything. if err := crictl.StopPodAndContainer(podID, contID); err != nil { - t.Fatal(err) + t.Fatalf("stop failed: %v", err) } } @@ -95,14 +166,16 @@ func TestMountOverSymlinks(t *testing.T) { t.Fatalf("failed to setup crictl: %v", err) } defer cleanup() - podID, contID, err := crictl.StartPodAndContainer("k8s.gcr.io/busybox", testdata.Sandbox, testdata.MountOverSymlink) + + spec := SimpleSpec("busybox", "basic/resolv", []string{"sleep", "1000"}, nil) + podID, contID, err := crictl.StartPodAndContainer("basic/resolv", Sandbox, spec) if err != nil { - t.Fatal(err) + t.Fatalf("start failed: %v", err) } out, err := crictl.Exec(contID, "readlink", "/etc/resolv.conf") if err != nil { - t.Fatal(err) + t.Fatalf("readlink failed: %v, out: %s", err, out) } if want := "/tmp/resolv.conf"; !strings.Contains(string(out), want) { t.Fatalf("/etc/resolv.conf is not pointing to %q: %q", want, string(out)) @@ -110,11 +183,11 @@ func TestMountOverSymlinks(t *testing.T) { etc, err := crictl.Exec(contID, "cat", "/etc/resolv.conf") if err != nil { - t.Fatal(err) + t.Fatalf("cat failed: %v, out: %s", err, etc) } tmp, err := crictl.Exec(contID, "cat", "/tmp/resolv.conf") if err != nil { - t.Fatal(err) + t.Fatalf("cat failed: %v, out: %s", err, out) } if tmp != etc { t.Fatalf("file content doesn't match:\n\t/etc/resolv.conf: %s\n\t/tmp/resolv.conf: %s", string(etc), string(tmp)) @@ -122,7 +195,7 @@ func TestMountOverSymlinks(t *testing.T) { // Stop everything. if err := crictl.StopPodAndContainer(podID, contID); err != nil { - t.Fatal(err) + t.Fatalf("stop failed: %v", err) } } @@ -135,16 +208,16 @@ func TestHomeDir(t *testing.T) { t.Fatalf("failed to setup crictl: %v", err) } defer cleanup() - contSpec := testdata.SimpleSpec("root", "k8s.gcr.io/busybox", []string{"sleep", "1000"}) - podID, contID, err := crictl.StartPodAndContainer("k8s.gcr.io/busybox", testdata.Sandbox, contSpec) + contSpec := SimpleSpec("root", "basic/busybox", []string{"sleep", "1000"}, nil) + podID, contID, err := crictl.StartPodAndContainer("basic/busybox", Sandbox, contSpec) if err != nil { - t.Fatal(err) + t.Fatalf("start failed: %v", err) } t.Run("root container", func(t *testing.T) { out, err := crictl.Exec(contID, "sh", "-c", "echo $HOME") if err != nil { - t.Fatal(err) + t.Fatalf("exec failed: %v, out: %s", err, out) } if got, want := strings.TrimSpace(string(out)), "/root"; got != want { t.Fatalf("Home directory invalid. Got %q, Want : %q", got, want) @@ -153,32 +226,47 @@ func TestHomeDir(t *testing.T) { t.Run("sub-container", func(t *testing.T) { // Create a sub container in the same pod. - subContSpec := testdata.SimpleSpec("subcontainer", "k8s.gcr.io/busybox", []string{"sleep", "1000"}) - subContID, err := crictl.StartContainer(podID, "k8s.gcr.io/busybox", testdata.Sandbox, subContSpec) + subContSpec := SimpleSpec("subcontainer", "basic/busybox", []string{"sleep", "1000"}, nil) + subContID, err := crictl.StartContainer(podID, "basic/busybox", Sandbox, subContSpec) if err != nil { - t.Fatal(err) + t.Fatalf("start failed: %v", err) } out, err := crictl.Exec(subContID, "sh", "-c", "echo $HOME") if err != nil { - t.Fatal(err) + t.Fatalf("exec failed: %v, out: %s", err, out) } if got, want := strings.TrimSpace(string(out)), "/root"; got != want { t.Fatalf("Home directory invalid. Got %q, Want: %q", got, want) } if err := crictl.StopContainer(subContID); err != nil { - t.Fatal(err) + t.Fatalf("stop failed: %v", err) } }) // Stop everything. if err := crictl.StopPodAndContainer(podID, contID); err != nil { - t.Fatal(err) + t.Fatalf("stop failed: %v", err) } } +// containerdConfigTemplate is a .toml config for containerd. It contains a +// formatting verb so the runtime field can be set via fmt.Sprintf. +const containerdConfigTemplate = ` +disabled_plugins = ["restart"] +[plugins.linux] + runtime = "%s" + runtime_root = "/tmp/test-containerd/runsc" + shim = "/usr/local/bin/gvisor-containerd-shim" + shim_debug = true + +[plugins.cri.containerd.runtimes.runsc] + runtime_type = "io.containerd.runtime.v1.linux" + runtime_engine = "%s" +` + // setup sets up before a test. Specifically it: // * Creates directories and a socket for containerd to utilize. // * Runs containerd and waits for it to reach a "ready" state for testing. @@ -213,50 +301,52 @@ func setup(t *testing.T) (*criutil.Crictl, func(), error) { if err != nil { t.Fatalf("error discovering runtime path: %v", err) } - config, err := testutil.WriteTmpFile("containerd-config", testdata.ContainerdConfig(runtime)) + config, configCleanup, err := testutil.WriteTmpFile("containerd-config", fmt.Sprintf(containerdConfigTemplate, runtime, runtime)) if err != nil { t.Fatalf("failed to write containerd config") } - cleanups = append(cleanups, func() { os.RemoveAll(config) }) + cleanups = append(cleanups, configCleanup) // Start containerd. - containerd := exec.Command(getContainerd(), + cmd := exec.Command(getContainerd(), "--config", config, "--log-level", "debug", "--root", containerdRoot, "--state", containerdState, "--address", sockAddr) + startupR, startupW := io.Pipe() + defer startupR.Close() + defer startupW.Close() + stderr := &bytes.Buffer{} + stdout := &bytes.Buffer{} + cmd.Stderr = io.MultiWriter(startupW, stderr) + cmd.Stdout = io.MultiWriter(startupW, stdout) cleanups = append(cleanups, func() { - if err := testutil.KillCommand(containerd); err != nil { - log.Printf("error killing containerd: %v", err) - } + t.Logf("containerd stdout: %s", stdout.String()) + t.Logf("containerd stderr: %s", stderr.String()) }) - containerdStderr, err := containerd.StderrPipe() - if err != nil { - t.Fatalf("failed to get containerd stderr: %v", err) - } - containerdStdout, err := containerd.StdoutPipe() - if err != nil { - t.Fatalf("failed to get containerd stdout: %v", err) - } - if err := containerd.Start(); err != nil { + + // Start the process. + if err := cmd.Start(); err != nil { t.Fatalf("failed running containerd: %v", err) } - // Wait for containerd to boot. Then put all containerd output into a - // buffer to be logged at the end of the test. - testutil.WaitUntilRead(containerdStderr, "Start streaming server", nil, 10*time.Second) - stdoutBuf := &bytes.Buffer{} - stderrBuf := &bytes.Buffer{} - go func() { io.Copy(stdoutBuf, containerdStdout) }() - go func() { io.Copy(stderrBuf, containerdStderr) }() + // Wait for containerd to boot. + if err := testutil.WaitUntilRead(startupR, "Start streaming server", nil, 10*time.Second); err != nil { + t.Fatalf("failed to start containerd: %v", err) + } + + // Kill must be the last cleanup (as it will be executed first). + cc := criutil.NewCrictl(t, sockAddr) cleanups = append(cleanups, func() { - t.Logf("containerd stdout: %s", string(stdoutBuf.Bytes())) - t.Logf("containerd stderr: %s", string(stderrBuf.Bytes())) + cc.CleanUp() // Remove tmp files, etc. + if err := testutil.KillCommand(cmd); err != nil { + log.Printf("error killing containerd: %v", err) + } }) cleanup.Release() - return criutil.NewCrictl(20*time.Second, sockAddr), cleanupFunc, nil + return cc, cleanupFunc, nil } // httpGet GETs the contents of a file served from a pod on port 80. diff --git a/test/root/main_test.go b/test/root/main_test.go index d74dec85f..9fb17e0dd 100644 --- a/test/root/main_test.go +++ b/test/root/main_test.go @@ -21,7 +21,7 @@ import ( "testing" "github.com/syndtr/gocapability/capability" - "gvisor.dev/gvisor/runsc/dockerutil" + "gvisor.dev/gvisor/pkg/test/dockerutil" "gvisor.dev/gvisor/runsc/specutils" ) diff --git a/test/root/oom_score_adj_test.go b/test/root/oom_score_adj_test.go index 22488b05d..9a3cecd97 100644 --- a/test/root/oom_score_adj_test.go +++ b/test/root/oom_score_adj_test.go @@ -20,10 +20,9 @@ import ( "testing" specs "github.com/opencontainers/runtime-spec/specs-go" - "gvisor.dev/gvisor/runsc/boot" + "gvisor.dev/gvisor/pkg/test/testutil" "gvisor.dev/gvisor/runsc/container" "gvisor.dev/gvisor/runsc/specutils" - "gvisor.dev/gvisor/runsc/testutil" ) var ( @@ -40,15 +39,6 @@ var ( // TestOOMScoreAdjSingle tests that oom_score_adj is set properly in a // single container sandbox. func TestOOMScoreAdjSingle(t *testing.T) { - rootDir, err := testutil.SetupRootDir() - if err != nil { - t.Fatalf("error creating root dir: %v", err) - } - defer os.RemoveAll(rootDir) - - conf := testutil.TestConfig(t) - conf.RootDir = rootDir - ppid, err := specutils.GetParentPid(os.Getpid()) if err != nil { t.Fatalf("getting parent pid: %v", err) @@ -89,11 +79,11 @@ func TestOOMScoreAdjSingle(t *testing.T) { for _, testCase := range testCases { t.Run(testCase.Name, func(t *testing.T) { - id := testutil.UniqueContainerID() + id := testutil.RandomContainerID() s := testutil.NewSpecWithArgs("sleep", "1000") s.Process.OOMScoreAdj = testCase.OOMScoreAdj - containers, cleanup, err := startContainers(conf, []*specs.Spec{s}, []string{id}) + containers, cleanup, err := startContainers(t, []*specs.Spec{s}, []string{id}) if err != nil { t.Fatalf("error starting containers: %v", err) } @@ -131,15 +121,6 @@ func TestOOMScoreAdjSingle(t *testing.T) { // TestOOMScoreAdjMulti tests that oom_score_adj is set properly in a // multi-container sandbox. func TestOOMScoreAdjMulti(t *testing.T) { - rootDir, err := testutil.SetupRootDir() - if err != nil { - t.Fatalf("error creating root dir: %v", err) - } - defer os.RemoveAll(rootDir) - - conf := testutil.TestConfig(t) - conf.RootDir = rootDir - ppid, err := specutils.GetParentPid(os.Getpid()) if err != nil { t.Fatalf("getting parent pid: %v", err) @@ -257,7 +238,7 @@ func TestOOMScoreAdjMulti(t *testing.T) { } } - containers, cleanup, err := startContainers(conf, specs, ids) + containers, cleanup, err := startContainers(t, specs, ids) if err != nil { t.Fatalf("error starting containers: %v", err) } @@ -321,7 +302,7 @@ func TestOOMScoreAdjMulti(t *testing.T) { func createSpecs(cmds ...[]string) ([]*specs.Spec, []string) { var specs []*specs.Spec var ids []string - rootID := testutil.UniqueContainerID() + rootID := testutil.RandomContainerID() for i, cmd := range cmds { spec := testutil.NewSpecWithArgs(cmd...) @@ -335,35 +316,48 @@ func createSpecs(cmds ...[]string) ([]*specs.Spec, []string) { specutils.ContainerdContainerTypeAnnotation: specutils.ContainerdContainerTypeContainer, specutils.ContainerdSandboxIDAnnotation: rootID, } - ids = append(ids, testutil.UniqueContainerID()) + ids = append(ids, testutil.RandomContainerID()) } specs = append(specs, spec) } return specs, ids } -func startContainers(conf *boot.Config, specs []*specs.Spec, ids []string) ([]*container.Container, func(), error) { - if len(conf.RootDir) == 0 { - panic("conf.RootDir not set. Call testutil.SetupRootDir() to set.") - } - - var containers []*container.Container - var bundles []string - cleanup := func() { +func startContainers(t *testing.T, specs []*specs.Spec, ids []string) ([]*container.Container, func(), error) { + var ( + containers []*container.Container + cleanups []func() + ) + cleanups = append(cleanups, func() { for _, c := range containers { c.Destroy() } - for _, b := range bundles { - os.RemoveAll(b) + }) + cleanupAll := func() { + for _, c := range cleanups { + c() } } + localClean := specutils.MakeCleanup(cleanupAll) + defer localClean.Clean() + + // All containers must share the same root. + rootDir, cleanup, err := testutil.SetupRootDir() + if err != nil { + t.Fatalf("error creating root dir: %v", err) + } + cleanups = append(cleanups, cleanup) + + // Point this to from the configuration. + conf := testutil.TestConfig(t) + conf.RootDir = rootDir + for i, spec := range specs { - bundleDir, err := testutil.SetupBundleDir(spec) + bundleDir, cleanup, err := testutil.SetupBundleDir(spec) if err != nil { - cleanup() - return nil, nil, fmt.Errorf("error setting up container: %v", err) + return nil, nil, fmt.Errorf("error setting up bundle: %v", err) } - bundles = append(bundles, bundleDir) + cleanups = append(cleanups, cleanup) args := container.Args{ ID: ids[i], @@ -372,15 +366,15 @@ func startContainers(conf *boot.Config, specs []*specs.Spec, ids []string) ([]*c } cont, err := container.New(conf, args) if err != nil { - cleanup() return nil, nil, fmt.Errorf("error creating container: %v", err) } containers = append(containers, cont) if err := cont.Start(conf); err != nil { - cleanup() return nil, nil, fmt.Errorf("error starting container: %v", err) } } - return containers, cleanup, nil + + localClean.Release() + return containers, cleanupAll, nil } diff --git a/test/root/runsc_test.go b/test/root/runsc_test.go index 90373e2db..25204bebb 100644 --- a/test/root/runsc_test.go +++ b/test/root/runsc_test.go @@ -28,8 +28,8 @@ import ( "github.com/cenkalti/backoff" "golang.org/x/sys/unix" + "gvisor.dev/gvisor/pkg/test/testutil" "gvisor.dev/gvisor/runsc/specutils" - "gvisor.dev/gvisor/runsc/testutil" ) // TestDoKill checks that when "runsc do..." is killed, the sandbox process is diff --git a/test/root/testdata/BUILD b/test/root/testdata/BUILD deleted file mode 100644 index 6859541ad..000000000 --- a/test/root/testdata/BUILD +++ /dev/null @@ -1,18 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "testdata", - srcs = [ - "busybox.go", - "containerd_config.go", - "httpd.go", - "httpd_mount_paths.go", - "sandbox.go", - "simple.go", - ], - visibility = [ - "//:sandbox", - ], -) diff --git a/test/root/testdata/busybox.go b/test/root/testdata/busybox.go deleted file mode 100644 index e4dbd2843..000000000 --- a/test/root/testdata/busybox.go +++ /dev/null @@ -1,32 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package testdata - -// MountOverSymlink is a JSON config for a container that /etc/resolv.conf is a -// symlink to /tmp/resolv.conf. -var MountOverSymlink = ` -{ - "metadata": { - "name": "busybox" - }, - "image": { - "image": "k8s.gcr.io/busybox" - }, - "command": [ - "sleep", - "1000" - ] -} -` diff --git a/test/root/testdata/containerd_config.go b/test/root/testdata/containerd_config.go deleted file mode 100644 index e12f1ec88..000000000 --- a/test/root/testdata/containerd_config.go +++ /dev/null @@ -1,39 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package testdata contains data required for root tests. -package testdata - -import "fmt" - -// containerdConfigTemplate is a .toml config for containerd. It contains a -// formatting verb so the runtime field can be set via fmt.Sprintf. -const containerdConfigTemplate = ` -disabled_plugins = ["restart"] -[plugins.linux] - runtime = "%s" - runtime_root = "/tmp/test-containerd/runsc" - shim = "/usr/local/bin/gvisor-containerd-shim" - shim_debug = true - -[plugins.cri.containerd.runtimes.runsc] - runtime_type = "io.containerd.runtime.v1.linux" - runtime_engine = "%s" -` - -// ContainerdConfig returns a containerd config file with the specified -// runtime. -func ContainerdConfig(runtime string) string { - return fmt.Sprintf(containerdConfigTemplate, runtime, runtime) -} diff --git a/test/root/testdata/httpd.go b/test/root/testdata/httpd.go deleted file mode 100644 index 45d5e33d4..000000000 --- a/test/root/testdata/httpd.go +++ /dev/null @@ -1,32 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package testdata - -// Httpd is a JSON config for an httpd container. -const Httpd = ` -{ - "metadata": { - "name": "httpd" - }, - "image":{ - "image": "httpd" - }, - "mounts": [ - ], - "linux": { - }, - "log_path": "httpd.log" -} -` diff --git a/test/root/testdata/httpd_mount_paths.go b/test/root/testdata/httpd_mount_paths.go deleted file mode 100644 index ac3f4446a..000000000 --- a/test/root/testdata/httpd_mount_paths.go +++ /dev/null @@ -1,53 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package testdata - -// HttpdMountPaths is a JSON config for an httpd container with additional -// mounts. -const HttpdMountPaths = ` -{ - "metadata": { - "name": "httpd" - }, - "image":{ - "image": "httpd" - }, - "mounts": [ - { - "container_path": "/var/run/secrets/kubernetes.io/serviceaccount", - "host_path": "/var/lib/kubelet/pods/82bae206-cdf5-11e8-b245-8cdcd43ac064/volumes/kubernetes.io~secret/default-token-2rpfx", - "readonly": true - }, - { - "container_path": "/etc/hosts", - "host_path": "/var/lib/kubelet/pods/82bae206-cdf5-11e8-b245-8cdcd43ac064/etc-hosts", - "readonly": false - }, - { - "container_path": "/dev/termination-log", - "host_path": "/var/lib/kubelet/pods/82bae206-cdf5-11e8-b245-8cdcd43ac064/containers/httpd/d1709580", - "readonly": false - }, - { - "container_path": "/usr/local/apache2/htdocs/test", - "host_path": "/var/lib/kubelet/pods/82bae206-cdf5-11e8-b245-8cdcd43ac064", - "readonly": true - } - ], - "linux": { - }, - "log_path": "httpd.log" -} -` diff --git a/test/root/testdata/sandbox.go b/test/root/testdata/sandbox.go deleted file mode 100644 index 0db210370..000000000 --- a/test/root/testdata/sandbox.go +++ /dev/null @@ -1,30 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package testdata - -// Sandbox is a default JSON config for a sandbox. -const Sandbox = ` -{ - "metadata": { - "name": "default-sandbox", - "namespace": "default", - "attempt": 1, - "uid": "hdishd83djaidwnduwk28bcsb" - }, - "linux": { - }, - "log_directory": "/tmp" -} -` diff --git a/test/root/testdata/simple.go b/test/root/testdata/simple.go deleted file mode 100644 index 1cca53f0c..000000000 --- a/test/root/testdata/simple.go +++ /dev/null @@ -1,41 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package testdata - -import ( - "encoding/json" - "fmt" -) - -// SimpleSpec returns a JSON config for a simple container that runs the -// specified command in the specified image. -func SimpleSpec(name, image string, cmd []string) string { - cmds, err := json.Marshal(cmd) - if err != nil { - // This shouldn't happen. - panic(err) - } - return fmt.Sprintf(` -{ - "metadata": { - "name": %q - }, - "image": { - "image": %q - }, - "command": %s - } -`, name, image, cmds) -} diff --git a/test/runner/BUILD b/test/runner/BUILD index 9959ef9b0..6833c9986 100644 --- a/test/runner/BUILD +++ b/test/runner/BUILD @@ -12,8 +12,8 @@ go_binary( visibility = ["//:sandbox"], deps = [ "//pkg/log", + "//pkg/test/testutil", "//runsc/specutils", - "//runsc/testutil", "//test/runner/gtest", "//test/uds", "@com_github_opencontainers_runtime-spec//specs-go:go_default_library", diff --git a/test/runner/runner.go b/test/runner/runner.go index 0d3742f71..14c9cbc47 100644 --- a/test/runner/runner.go +++ b/test/runner/runner.go @@ -32,8 +32,8 @@ import ( specs "github.com/opencontainers/runtime-spec/specs-go" "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/test/testutil" "gvisor.dev/gvisor/runsc/specutils" - "gvisor.dev/gvisor/runsc/testutil" "gvisor.dev/gvisor/test/runner/gtest" "gvisor.dev/gvisor/test/uds" ) @@ -115,20 +115,20 @@ func runTestCaseNative(testBin string, tc gtest.TestCase, t *testing.T) { // // Returns an error if the sandboxed application exits non-zero. func runRunsc(tc gtest.TestCase, spec *specs.Spec) error { - bundleDir, err := testutil.SetupBundleDir(spec) + bundleDir, cleanup, err := testutil.SetupBundleDir(spec) if err != nil { return fmt.Errorf("SetupBundleDir failed: %v", err) } - defer os.RemoveAll(bundleDir) + defer cleanup() - rootDir, err := testutil.SetupRootDir() + rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { return fmt.Errorf("SetupRootDir failed: %v", err) } - defer os.RemoveAll(rootDir) + defer cleanup() name := tc.FullName() - id := testutil.UniqueContainerID() + id := testutil.RandomContainerID() log.Infof("Running test %q in container %q", name, id) specutils.LogSpec(spec) diff --git a/test/runtimes/BUILD b/test/runtimes/BUILD index 2c472bf8d..4cd627222 100644 --- a/test/runtimes/BUILD +++ b/test/runtimes/BUILD @@ -1,20 +1,7 @@ -# These packages are used to run language runtime tests inside gVisor sandboxes. - -load("//tools:defs.bzl", "go_binary", "go_test") -load("//test/runtimes:build_defs.bzl", "runtime_test") +load("//test/runtimes:defs.bzl", "runtime_test") package(licenses = ["notice"]) -go_binary( - name = "runner", - testonly = 1, - srcs = ["runner.go"], - deps = [ - "//runsc/dockerutil", - "//runsc/testutil", - ], -) - runtime_test( name = "go1.12", blacklist_file = "blacklist_go1.12.csv", @@ -44,10 +31,3 @@ runtime_test( blacklist_file = "blacklist_python3.7.3.csv", lang = "python", ) - -go_test( - name = "blacklist_test", - size = "small", - srcs = ["blacklist_test.go"], - library = ":runner", -) diff --git a/test/runtimes/README.md b/test/runtimes/README.md deleted file mode 100644 index 42d722553..000000000 --- a/test/runtimes/README.md +++ /dev/null @@ -1,56 +0,0 @@ -# Runtimes Tests Dockerfiles - -The Dockerfiles defined under this path are configured to host the execution of -the runtimes language tests. Each Dockerfile can support the language indicated -by its directory. - -The following runtimes are currently supported: - -- Go 1.12 -- Java 11 -- Node.js 12 -- PHP 7.3 -- Python 3.7 - -### Building and pushing the images: - -The canonical source of images is the -[gvisor-presubmit container registry](https://gcr.io/gvisor-presubmit/). You can -build new images with the following command: - -```bash -$ cd images -$ docker build -f Dockerfile_$LANG [-t $NAME] . -``` - -To push them to our container registry, set the tag in the command above to -`gcr.io/gvisor-presubmit/$LANG`, then push them. (Note that you will need -appropriate permissions to the `gvisor-presubmit` GCP project.) - -```bash -gcloud docker -- push gcr.io/gvisor-presubmit/$LANG -``` - -#### Running in Docker locally: - -1) [Install and configure Docker](https://docs.docker.com/install/) - -2) Pull the image you want to run: - -```bash -$ docker pull gcr.io/gvisor-presubmit/$LANG -``` - -3) Run docker with the image. - -```bash -$ docker run [--runtime=runsc] --rm -it $NAME [FLAG] -``` - -Running the command with no flags will cause all the available tests to execute. - -Flags can be added for additional functionality: - -- --list: Print a list of all available tests -- --test <name>: Run a single test from the list of available tests -- --v: Print the language version diff --git a/test/runtimes/blacklist_test.go b/test/runtimes/blacklist_test.go deleted file mode 100644 index 0ff69ab18..000000000 --- a/test/runtimes/blacklist_test.go +++ /dev/null @@ -1,37 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package main - -import ( - "flag" - "os" - "testing" -) - -func TestMain(m *testing.M) { - flag.Parse() - os.Exit(m.Run()) -} - -// Test that the blacklist parses without error. -func TestBlacklists(t *testing.T) { - bl, err := getBlacklist() - if err != nil { - t.Fatalf("error parsing blacklist: %v", err) - } - if *blacklistFile != "" && len(bl) == 0 { - t.Errorf("got empty blacklist for file %q", *blacklistFile) - } -} diff --git a/test/runtimes/build_defs.bzl b/test/runtimes/build_defs.bzl deleted file mode 100644 index 92e275a76..000000000 --- a/test/runtimes/build_defs.bzl +++ /dev/null @@ -1,75 +0,0 @@ -"""Defines a rule for runtime test targets.""" - -load("//tools:defs.bzl", "go_test", "loopback") - -def runtime_test( - name, - lang, - image_repo = "gcr.io/gvisor-presubmit", - image_name = None, - blacklist_file = None, - shard_count = 50, - size = "enormous"): - """Generates sh_test and blacklist test targets for a given runtime. - - Args: - name: The name of the runtime being tested. Typically, the lang + version. - This is used in the names of the generated test targets. - lang: The language being tested. - image_repo: The docker repository containing the proctor image to run. - i.e., the prefix to the fully qualified docker image id. - image_name: The name of the image in the image_repo. - Defaults to the test name. - blacklist_file: A test blacklist to pass to the runtime test's runner. - shard_count: See Bazel common test attributes. - size: See Bazel common test attributes. - """ - if image_name == None: - image_name = name - args = [ - "--lang", - lang, - "--image", - "/".join([image_repo, image_name]), - ] - data = [ - ":runner", - loopback, - ] - if blacklist_file: - args += ["--blacklist_file", "test/runtimes/" + blacklist_file] - data += [blacklist_file] - - # Add a test that the blacklist parses correctly. - blacklist_test(name, blacklist_file) - - sh_test( - name = name + "_test", - srcs = ["runner.sh"], - args = args, - data = data, - size = size, - shard_count = shard_count, - tags = [ - # Requires docker and runsc to be configured before the test runs. - "local", - # Don't include test target in wildcard target patterns. - "manual", - ], - ) - -def blacklist_test(name, blacklist_file): - """Test that a blacklist parses correctly.""" - go_test( - name = name + "_blacklist_test", - library = ":runner", - srcs = ["blacklist_test.go"], - args = ["--blacklist_file", "test/runtimes/" + blacklist_file], - data = [blacklist_file], - ) - -def sh_test(**kwargs): - """Wraps the standard sh_test.""" - native.sh_test( - **kwargs - ) diff --git a/test/runtimes/defs.bzl b/test/runtimes/defs.bzl new file mode 100644 index 000000000..f836dd952 --- /dev/null +++ b/test/runtimes/defs.bzl @@ -0,0 +1,79 @@ +"""Defines a rule for runtime test targets.""" + +load("//tools:defs.bzl", "go_test") + +def _runtime_test_impl(ctx): + # Construct arguments. + args = [ + "--lang", + ctx.attr.lang, + "--image", + ctx.attr.image, + ] + if ctx.attr.blacklist_file: + args += [ + "--blacklist_file", + ctx.files.blacklist_file[0].short_path, + ] + + # Build a runner. + runner = ctx.actions.declare_file("%s-executer" % ctx.label.name) + runner_content = "\n".join([ + "#!/bin/bash", + "%s %s\n" % (ctx.files._runner[0].short_path, " ".join(args)), + ]) + ctx.actions.write(runner, runner_content, is_executable = True) + + # Return the runner. + return [DefaultInfo( + executable = runner, + runfiles = ctx.runfiles( + files = ctx.files._runner + ctx.files.blacklist_file + ctx.files._proctor, + collect_default = True, + collect_data = True, + ), + )] + +_runtime_test = rule( + implementation = _runtime_test_impl, + attrs = { + "image": attr.string( + mandatory = False, + ), + "lang": attr.string( + mandatory = True, + ), + "blacklist_file": attr.label( + mandatory = False, + allow_single_file = True, + ), + "_runner": attr.label( + default = "//test/runtimes/runner:runner", + ), + "_proctor": attr.label( + default = "//test/runtimes/proctor:proctor", + ), + }, + test = True, +) + +def runtime_test(name, **kwargs): + _runtime_test( + name = name, + image = name, # Resolved as images/runtimes/%s. + tags = [ + "local", + "manual", + ], + **kwargs + ) + +def blacklist_test(name, blacklist_file): + """Test that a blacklist parses correctly.""" + go_test( + name = name + "_blacklist_test", + library = ":runner", + srcs = ["blacklist_test.go"], + args = ["--blacklist_file", "test/runtimes/" + blacklist_file], + data = [blacklist_file], + ) diff --git a/test/runtimes/images/proctor/BUILD b/test/runtimes/images/proctor/BUILD deleted file mode 100644 index 85e004c45..000000000 --- a/test/runtimes/images/proctor/BUILD +++ /dev/null @@ -1,26 +0,0 @@ -load("//tools:defs.bzl", "go_binary", "go_test") - -package(licenses = ["notice"]) - -go_binary( - name = "proctor", - srcs = [ - "go.go", - "java.go", - "nodejs.go", - "php.go", - "proctor.go", - "python.go", - ], - visibility = ["//test/runtimes/images:__subpackages__"], -) - -go_test( - name = "proctor_test", - size = "small", - srcs = ["proctor_test.go"], - library = ":proctor", - deps = [ - "//runsc/testutil", - ], -) diff --git a/test/runtimes/images/proctor/go.go b/test/runtimes/images/proctor/go.go deleted file mode 100644 index 3e2d5d8db..000000000 --- a/test/runtimes/images/proctor/go.go +++ /dev/null @@ -1,90 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package main - -import ( - "fmt" - "os" - "os/exec" - "regexp" - "strings" -) - -var ( - goTestRegEx = regexp.MustCompile(`^.+\.go$`) - - // Directories with .dir contain helper files for tests. - // Exclude benchmarks and stress tests. - goDirFilter = regexp.MustCompile(`^(bench|stress)\/.+$|^.+\.dir.+$`) -) - -// Location of Go tests on disk. -const goTestDir = "/usr/local/go/test" - -// goRunner implements TestRunner for Go. -// -// There are two types of Go tests: "Go tool tests" and "Go tests on disk". -// "Go tool tests" are found and executed using `go tool dist test`. "Go tests -// on disk" are found in the /usr/local/go/test directory and are executed -// using `go run run.go`. -type goRunner struct{} - -var _ TestRunner = goRunner{} - -// ListTests implements TestRunner.ListTests. -func (goRunner) ListTests() ([]string, error) { - // Go tool dist test tests. - args := []string{"tool", "dist", "test", "-list"} - cmd := exec.Command("go", args...) - cmd.Stderr = os.Stderr - out, err := cmd.Output() - if err != nil { - return nil, fmt.Errorf("failed to list: %v", err) - } - var toolSlice []string - for _, test := range strings.Split(string(out), "\n") { - toolSlice = append(toolSlice, test) - } - - // Go tests on disk. - diskSlice, err := search(goTestDir, goTestRegEx) - if err != nil { - return nil, err - } - // Remove items from /bench/, /stress/ and .dir files - diskFiltered := diskSlice[:0] - for _, file := range diskSlice { - if !goDirFilter.MatchString(file) { - diskFiltered = append(diskFiltered, file) - } - } - - return append(toolSlice, diskFiltered...), nil -} - -// TestCmd implements TestRunner.TestCmd. -func (goRunner) TestCmd(test string) *exec.Cmd { - // Check if test exists on disk by searching for file of the same name. - // This will determine whether or not it is a Go test on disk. - if strings.HasSuffix(test, ".go") { - // Test has suffix ".go" which indicates a disk test, run it as such. - cmd := exec.Command("go", "run", "run.go", "-v", "--", test) - cmd.Dir = goTestDir - return cmd - } - - // No ".go" suffix, run as a tool test. - return exec.Command("go", "tool", "dist", "test", "-run", test) -} diff --git a/test/runtimes/images/proctor/java.go b/test/runtimes/images/proctor/java.go deleted file mode 100644 index 8b362029d..000000000 --- a/test/runtimes/images/proctor/java.go +++ /dev/null @@ -1,71 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package main - -import ( - "fmt" - "os" - "os/exec" - "regexp" - "strings" -) - -// Directories to exclude from tests. -var javaExclDirs = regexp.MustCompile(`(^(sun\/security)|(java\/util\/stream)|(java\/time)| )`) - -// Location of java tests. -const javaTestDir = "/root/test/jdk" - -// javaRunner implements TestRunner for Java. -type javaRunner struct{} - -var _ TestRunner = javaRunner{} - -// ListTests implements TestRunner.ListTests. -func (javaRunner) ListTests() ([]string, error) { - args := []string{ - "-dir:" + javaTestDir, - "-ignore:quiet", - "-a", - "-listtests", - ":jdk_core", - ":jdk_svc", - ":jdk_sound", - ":jdk_imageio", - } - cmd := exec.Command("jtreg", args...) - cmd.Stderr = os.Stderr - out, err := cmd.Output() - if err != nil { - return nil, fmt.Errorf("jtreg -listtests : %v", err) - } - var testSlice []string - for _, test := range strings.Split(string(out), "\n") { - if !javaExclDirs.MatchString(test) { - testSlice = append(testSlice, test) - } - } - return testSlice, nil -} - -// TestCmd implements TestRunner.TestCmd. -func (javaRunner) TestCmd(test string) *exec.Cmd { - args := []string{ - "-noreport", - "-dir:" + javaTestDir, - test, - } - return exec.Command("jtreg", args...) -} diff --git a/test/runtimes/images/proctor/nodejs.go b/test/runtimes/images/proctor/nodejs.go deleted file mode 100644 index bd57db444..000000000 --- a/test/runtimes/images/proctor/nodejs.go +++ /dev/null @@ -1,46 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package main - -import ( - "os/exec" - "path/filepath" - "regexp" -) - -var nodejsTestRegEx = regexp.MustCompile(`^test-[^-].+\.js$`) - -// Location of nodejs tests relative to working dir. -const nodejsTestDir = "test" - -// nodejsRunner implements TestRunner for NodeJS. -type nodejsRunner struct{} - -var _ TestRunner = nodejsRunner{} - -// ListTests implements TestRunner.ListTests. -func (nodejsRunner) ListTests() ([]string, error) { - testSlice, err := search(nodejsTestDir, nodejsTestRegEx) - if err != nil { - return nil, err - } - return testSlice, nil -} - -// TestCmd implements TestRunner.TestCmd. -func (nodejsRunner) TestCmd(test string) *exec.Cmd { - args := []string{filepath.Join("tools", "test.py"), test} - return exec.Command("/usr/bin/python", args...) -} diff --git a/test/runtimes/images/proctor/php.go b/test/runtimes/images/proctor/php.go deleted file mode 100644 index 9115040e1..000000000 --- a/test/runtimes/images/proctor/php.go +++ /dev/null @@ -1,42 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package main - -import ( - "os/exec" - "regexp" -) - -var phpTestRegEx = regexp.MustCompile(`^.+\.phpt$`) - -// phpRunner implements TestRunner for PHP. -type phpRunner struct{} - -var _ TestRunner = phpRunner{} - -// ListTests implements TestRunner.ListTests. -func (phpRunner) ListTests() ([]string, error) { - testSlice, err := search(".", phpTestRegEx) - if err != nil { - return nil, err - } - return testSlice, nil -} - -// TestCmd implements TestRunner.TestCmd. -func (phpRunner) TestCmd(test string) *exec.Cmd { - args := []string{"test", "TESTS=" + test} - return exec.Command("make", args...) -} diff --git a/test/runtimes/images/proctor/proctor.go b/test/runtimes/images/proctor/proctor.go deleted file mode 100644 index b54abe434..000000000 --- a/test/runtimes/images/proctor/proctor.go +++ /dev/null @@ -1,163 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Binary proctor runs the test for a particular runtime. It is meant to be -// included in Docker images for all runtime tests. -package main - -import ( - "flag" - "fmt" - "log" - "os" - "os/exec" - "os/signal" - "path/filepath" - "regexp" - "syscall" -) - -// TestRunner is an interface that must be implemented for each runtime -// integrated with proctor. -type TestRunner interface { - // ListTests returns a string slice of tests available to run. - ListTests() ([]string, error) - - // TestCmd returns an *exec.Cmd that will run the given test. - TestCmd(test string) *exec.Cmd -} - -var ( - runtime = flag.String("runtime", "", "name of runtime") - list = flag.Bool("list", false, "list all available tests") - testName = flag.String("test", "", "run a single test from the list of available tests") - pause = flag.Bool("pause", false, "cause container to pause indefinitely, reaping any zombie children") -) - -func main() { - flag.Parse() - - if *pause { - pauseAndReap() - panic("pauseAndReap should never return") - } - - if *runtime == "" { - log.Fatalf("runtime flag must be provided") - } - - tr, err := testRunnerForRuntime(*runtime) - if err != nil { - log.Fatalf("%v", err) - } - - // List tests. - if *list { - tests, err := tr.ListTests() - if err != nil { - log.Fatalf("failed to list tests: %v", err) - } - for _, test := range tests { - fmt.Println(test) - } - return - } - - var tests []string - if *testName == "" { - // Run every test. - tests, err = tr.ListTests() - if err != nil { - log.Fatalf("failed to get all tests: %v", err) - } - } else { - // Run a single test. - tests = []string{*testName} - } - for _, test := range tests { - cmd := tr.TestCmd(test) - cmd.Stdout, cmd.Stderr = os.Stdout, os.Stderr - if err := cmd.Run(); err != nil { - log.Fatalf("FAIL: %v", err) - } - } -} - -// testRunnerForRuntime returns a new TestRunner for the given runtime. -func testRunnerForRuntime(runtime string) (TestRunner, error) { - switch runtime { - case "go": - return goRunner{}, nil - case "java": - return javaRunner{}, nil - case "nodejs": - return nodejsRunner{}, nil - case "php": - return phpRunner{}, nil - case "python": - return pythonRunner{}, nil - } - return nil, fmt.Errorf("invalid runtime %q", runtime) -} - -// pauseAndReap is like init. It runs forever and reaps any children. -func pauseAndReap() { - // Get notified of any new children. - ch := make(chan os.Signal, 1) - signal.Notify(ch, syscall.SIGCHLD) - - for { - if _, ok := <-ch; !ok { - // Channel closed. This should not happen. - panic("signal channel closed") - } - - // Reap the child. - for { - if cpid, _ := syscall.Wait4(-1, nil, syscall.WNOHANG, nil); cpid < 1 { - break - } - } - } -} - -// search is a helper function to find tests in the given directory that match -// the regex. -func search(root string, testFilter *regexp.Regexp) ([]string, error) { - var testSlice []string - - err := filepath.Walk(root, func(path string, info os.FileInfo, err error) error { - if err != nil { - return err - } - - name := filepath.Base(path) - - if info.IsDir() || !testFilter.MatchString(name) { - return nil - } - - relPath, err := filepath.Rel(root, path) - if err != nil { - return err - } - testSlice = append(testSlice, relPath) - return nil - }) - if err != nil { - return nil, fmt.Errorf("walking %q: %v", root, err) - } - - return testSlice, nil -} diff --git a/test/runtimes/images/proctor/proctor_test.go b/test/runtimes/images/proctor/proctor_test.go deleted file mode 100644 index 6bb61d142..000000000 --- a/test/runtimes/images/proctor/proctor_test.go +++ /dev/null @@ -1,127 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package main - -import ( - "io/ioutil" - "os" - "path/filepath" - "reflect" - "regexp" - "strings" - "testing" - - "gvisor.dev/gvisor/runsc/testutil" -) - -func touch(t *testing.T, name string) { - t.Helper() - f, err := os.Create(name) - if err != nil { - t.Fatal(err) - } - if err := f.Close(); err != nil { - t.Fatal(err) - } -} - -func TestSearchEmptyDir(t *testing.T) { - td, err := ioutil.TempDir(testutil.TmpDir(), "searchtest") - if err != nil { - t.Fatal(err) - } - defer os.RemoveAll(td) - - var want []string - - testFilter := regexp.MustCompile(`^test-[^-].+\.tc$`) - got, err := search(td, testFilter) - if err != nil { - t.Errorf("search error: %v", err) - } - - if !reflect.DeepEqual(got, want) { - t.Errorf("Found %#v; want %#v", got, want) - } -} - -func TestSearch(t *testing.T) { - td, err := ioutil.TempDir(testutil.TmpDir(), "searchtest") - if err != nil { - t.Fatal(err) - } - defer os.RemoveAll(td) - - // Creating various files similar to the test filter regex. - files := []string{ - "emp/", - "tee/", - "test-foo.tc", - "test-foo.tc", - "test-bar.tc", - "test-sam.tc", - "Test-que.tc", - "test-brett", - "test--abc.tc", - "test---xyz.tc", - "test-bool.TC", - "--test-gvs.tc", - " test-pew.tc", - "dir/test_baz.tc", - "dir/testsnap.tc", - "dir/test-luk.tc", - "dir/nest/test-ok.tc", - "dir/dip/diz/goog/test-pack.tc", - "dir/dip/diz/wobble/thud/test-cas.e", - "dir/dip/diz/wobble/thud/test-cas.tc", - } - want := []string{ - "dir/dip/diz/goog/test-pack.tc", - "dir/dip/diz/wobble/thud/test-cas.tc", - "dir/nest/test-ok.tc", - "dir/test-luk.tc", - "test-bar.tc", - "test-foo.tc", - "test-sam.tc", - } - - for _, item := range files { - if strings.HasSuffix(item, "/") { - // This item is a directory, create it. - if err := os.MkdirAll(filepath.Join(td, item), 0755); err != nil { - t.Fatal(err) - } - } else { - // This item is a file, create the directory and touch file. - // Create directory in which file should be created - fullDirPath := filepath.Join(td, filepath.Dir(item)) - if err := os.MkdirAll(fullDirPath, 0755); err != nil { - t.Fatal(err) - } - // Create file with full path to file. - touch(t, filepath.Join(td, item)) - } - } - - testFilter := regexp.MustCompile(`^test-[^-].+\.tc$`) - got, err := search(td, testFilter) - if err != nil { - t.Errorf("search error: %v", err) - } - - if !reflect.DeepEqual(got, want) { - t.Errorf("Found %#v; want %#v", got, want) - } -} diff --git a/test/runtimes/images/proctor/python.go b/test/runtimes/images/proctor/python.go deleted file mode 100644 index b9e0fbe6f..000000000 --- a/test/runtimes/images/proctor/python.go +++ /dev/null @@ -1,49 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package main - -import ( - "fmt" - "os" - "os/exec" - "strings" -) - -// pythonRunner implements TestRunner for Python. -type pythonRunner struct{} - -var _ TestRunner = pythonRunner{} - -// ListTests implements TestRunner.ListTests. -func (pythonRunner) ListTests() ([]string, error) { - args := []string{"-m", "test", "--list-tests"} - cmd := exec.Command("./python", args...) - cmd.Stderr = os.Stderr - out, err := cmd.Output() - if err != nil { - return nil, fmt.Errorf("failed to list: %v", err) - } - var toolSlice []string - for _, test := range strings.Split(string(out), "\n") { - toolSlice = append(toolSlice, test) - } - return toolSlice, nil -} - -// TestCmd implements TestRunner.TestCmd. -func (pythonRunner) TestCmd(test string) *exec.Cmd { - args := []string{"-m", "test", test} - return exec.Command("./python", args...) -} diff --git a/test/runtimes/proctor/BUILD b/test/runtimes/proctor/BUILD new file mode 100644 index 000000000..50a26d182 --- /dev/null +++ b/test/runtimes/proctor/BUILD @@ -0,0 +1,27 @@ +load("//tools:defs.bzl", "go_binary", "go_test") + +package(licenses = ["notice"]) + +go_binary( + name = "proctor", + srcs = [ + "go.go", + "java.go", + "nodejs.go", + "php.go", + "proctor.go", + "python.go", + ], + pure = True, + visibility = ["//test/runtimes:__pkg__"], +) + +go_test( + name = "proctor_test", + size = "small", + srcs = ["proctor_test.go"], + library = ":proctor", + deps = [ + "//pkg/test/testutil", + ], +) diff --git a/test/runtimes/proctor/go.go b/test/runtimes/proctor/go.go new file mode 100644 index 000000000..3e2d5d8db --- /dev/null +++ b/test/runtimes/proctor/go.go @@ -0,0 +1,90 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package main + +import ( + "fmt" + "os" + "os/exec" + "regexp" + "strings" +) + +var ( + goTestRegEx = regexp.MustCompile(`^.+\.go$`) + + // Directories with .dir contain helper files for tests. + // Exclude benchmarks and stress tests. + goDirFilter = regexp.MustCompile(`^(bench|stress)\/.+$|^.+\.dir.+$`) +) + +// Location of Go tests on disk. +const goTestDir = "/usr/local/go/test" + +// goRunner implements TestRunner for Go. +// +// There are two types of Go tests: "Go tool tests" and "Go tests on disk". +// "Go tool tests" are found and executed using `go tool dist test`. "Go tests +// on disk" are found in the /usr/local/go/test directory and are executed +// using `go run run.go`. +type goRunner struct{} + +var _ TestRunner = goRunner{} + +// ListTests implements TestRunner.ListTests. +func (goRunner) ListTests() ([]string, error) { + // Go tool dist test tests. + args := []string{"tool", "dist", "test", "-list"} + cmd := exec.Command("go", args...) + cmd.Stderr = os.Stderr + out, err := cmd.Output() + if err != nil { + return nil, fmt.Errorf("failed to list: %v", err) + } + var toolSlice []string + for _, test := range strings.Split(string(out), "\n") { + toolSlice = append(toolSlice, test) + } + + // Go tests on disk. + diskSlice, err := search(goTestDir, goTestRegEx) + if err != nil { + return nil, err + } + // Remove items from /bench/, /stress/ and .dir files + diskFiltered := diskSlice[:0] + for _, file := range diskSlice { + if !goDirFilter.MatchString(file) { + diskFiltered = append(diskFiltered, file) + } + } + + return append(toolSlice, diskFiltered...), nil +} + +// TestCmd implements TestRunner.TestCmd. +func (goRunner) TestCmd(test string) *exec.Cmd { + // Check if test exists on disk by searching for file of the same name. + // This will determine whether or not it is a Go test on disk. + if strings.HasSuffix(test, ".go") { + // Test has suffix ".go" which indicates a disk test, run it as such. + cmd := exec.Command("go", "run", "run.go", "-v", "--", test) + cmd.Dir = goTestDir + return cmd + } + + // No ".go" suffix, run as a tool test. + return exec.Command("go", "tool", "dist", "test", "-run", test) +} diff --git a/test/runtimes/proctor/java.go b/test/runtimes/proctor/java.go new file mode 100644 index 000000000..8b362029d --- /dev/null +++ b/test/runtimes/proctor/java.go @@ -0,0 +1,71 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package main + +import ( + "fmt" + "os" + "os/exec" + "regexp" + "strings" +) + +// Directories to exclude from tests. +var javaExclDirs = regexp.MustCompile(`(^(sun\/security)|(java\/util\/stream)|(java\/time)| )`) + +// Location of java tests. +const javaTestDir = "/root/test/jdk" + +// javaRunner implements TestRunner for Java. +type javaRunner struct{} + +var _ TestRunner = javaRunner{} + +// ListTests implements TestRunner.ListTests. +func (javaRunner) ListTests() ([]string, error) { + args := []string{ + "-dir:" + javaTestDir, + "-ignore:quiet", + "-a", + "-listtests", + ":jdk_core", + ":jdk_svc", + ":jdk_sound", + ":jdk_imageio", + } + cmd := exec.Command("jtreg", args...) + cmd.Stderr = os.Stderr + out, err := cmd.Output() + if err != nil { + return nil, fmt.Errorf("jtreg -listtests : %v", err) + } + var testSlice []string + for _, test := range strings.Split(string(out), "\n") { + if !javaExclDirs.MatchString(test) { + testSlice = append(testSlice, test) + } + } + return testSlice, nil +} + +// TestCmd implements TestRunner.TestCmd. +func (javaRunner) TestCmd(test string) *exec.Cmd { + args := []string{ + "-noreport", + "-dir:" + javaTestDir, + test, + } + return exec.Command("jtreg", args...) +} diff --git a/test/runtimes/proctor/nodejs.go b/test/runtimes/proctor/nodejs.go new file mode 100644 index 000000000..bd57db444 --- /dev/null +++ b/test/runtimes/proctor/nodejs.go @@ -0,0 +1,46 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package main + +import ( + "os/exec" + "path/filepath" + "regexp" +) + +var nodejsTestRegEx = regexp.MustCompile(`^test-[^-].+\.js$`) + +// Location of nodejs tests relative to working dir. +const nodejsTestDir = "test" + +// nodejsRunner implements TestRunner for NodeJS. +type nodejsRunner struct{} + +var _ TestRunner = nodejsRunner{} + +// ListTests implements TestRunner.ListTests. +func (nodejsRunner) ListTests() ([]string, error) { + testSlice, err := search(nodejsTestDir, nodejsTestRegEx) + if err != nil { + return nil, err + } + return testSlice, nil +} + +// TestCmd implements TestRunner.TestCmd. +func (nodejsRunner) TestCmd(test string) *exec.Cmd { + args := []string{filepath.Join("tools", "test.py"), test} + return exec.Command("/usr/bin/python", args...) +} diff --git a/test/runtimes/proctor/php.go b/test/runtimes/proctor/php.go new file mode 100644 index 000000000..9115040e1 --- /dev/null +++ b/test/runtimes/proctor/php.go @@ -0,0 +1,42 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package main + +import ( + "os/exec" + "regexp" +) + +var phpTestRegEx = regexp.MustCompile(`^.+\.phpt$`) + +// phpRunner implements TestRunner for PHP. +type phpRunner struct{} + +var _ TestRunner = phpRunner{} + +// ListTests implements TestRunner.ListTests. +func (phpRunner) ListTests() ([]string, error) { + testSlice, err := search(".", phpTestRegEx) + if err != nil { + return nil, err + } + return testSlice, nil +} + +// TestCmd implements TestRunner.TestCmd. +func (phpRunner) TestCmd(test string) *exec.Cmd { + args := []string{"test", "TESTS=" + test} + return exec.Command("make", args...) +} diff --git a/test/runtimes/proctor/proctor.go b/test/runtimes/proctor/proctor.go new file mode 100644 index 000000000..b54abe434 --- /dev/null +++ b/test/runtimes/proctor/proctor.go @@ -0,0 +1,163 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Binary proctor runs the test for a particular runtime. It is meant to be +// included in Docker images for all runtime tests. +package main + +import ( + "flag" + "fmt" + "log" + "os" + "os/exec" + "os/signal" + "path/filepath" + "regexp" + "syscall" +) + +// TestRunner is an interface that must be implemented for each runtime +// integrated with proctor. +type TestRunner interface { + // ListTests returns a string slice of tests available to run. + ListTests() ([]string, error) + + // TestCmd returns an *exec.Cmd that will run the given test. + TestCmd(test string) *exec.Cmd +} + +var ( + runtime = flag.String("runtime", "", "name of runtime") + list = flag.Bool("list", false, "list all available tests") + testName = flag.String("test", "", "run a single test from the list of available tests") + pause = flag.Bool("pause", false, "cause container to pause indefinitely, reaping any zombie children") +) + +func main() { + flag.Parse() + + if *pause { + pauseAndReap() + panic("pauseAndReap should never return") + } + + if *runtime == "" { + log.Fatalf("runtime flag must be provided") + } + + tr, err := testRunnerForRuntime(*runtime) + if err != nil { + log.Fatalf("%v", err) + } + + // List tests. + if *list { + tests, err := tr.ListTests() + if err != nil { + log.Fatalf("failed to list tests: %v", err) + } + for _, test := range tests { + fmt.Println(test) + } + return + } + + var tests []string + if *testName == "" { + // Run every test. + tests, err = tr.ListTests() + if err != nil { + log.Fatalf("failed to get all tests: %v", err) + } + } else { + // Run a single test. + tests = []string{*testName} + } + for _, test := range tests { + cmd := tr.TestCmd(test) + cmd.Stdout, cmd.Stderr = os.Stdout, os.Stderr + if err := cmd.Run(); err != nil { + log.Fatalf("FAIL: %v", err) + } + } +} + +// testRunnerForRuntime returns a new TestRunner for the given runtime. +func testRunnerForRuntime(runtime string) (TestRunner, error) { + switch runtime { + case "go": + return goRunner{}, nil + case "java": + return javaRunner{}, nil + case "nodejs": + return nodejsRunner{}, nil + case "php": + return phpRunner{}, nil + case "python": + return pythonRunner{}, nil + } + return nil, fmt.Errorf("invalid runtime %q", runtime) +} + +// pauseAndReap is like init. It runs forever and reaps any children. +func pauseAndReap() { + // Get notified of any new children. + ch := make(chan os.Signal, 1) + signal.Notify(ch, syscall.SIGCHLD) + + for { + if _, ok := <-ch; !ok { + // Channel closed. This should not happen. + panic("signal channel closed") + } + + // Reap the child. + for { + if cpid, _ := syscall.Wait4(-1, nil, syscall.WNOHANG, nil); cpid < 1 { + break + } + } + } +} + +// search is a helper function to find tests in the given directory that match +// the regex. +func search(root string, testFilter *regexp.Regexp) ([]string, error) { + var testSlice []string + + err := filepath.Walk(root, func(path string, info os.FileInfo, err error) error { + if err != nil { + return err + } + + name := filepath.Base(path) + + if info.IsDir() || !testFilter.MatchString(name) { + return nil + } + + relPath, err := filepath.Rel(root, path) + if err != nil { + return err + } + testSlice = append(testSlice, relPath) + return nil + }) + if err != nil { + return nil, fmt.Errorf("walking %q: %v", root, err) + } + + return testSlice, nil +} diff --git a/test/runtimes/proctor/proctor_test.go b/test/runtimes/proctor/proctor_test.go new file mode 100644 index 000000000..6ef2de085 --- /dev/null +++ b/test/runtimes/proctor/proctor_test.go @@ -0,0 +1,127 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package main + +import ( + "io/ioutil" + "os" + "path/filepath" + "reflect" + "regexp" + "strings" + "testing" + + "gvisor.dev/gvisor/pkg/test/testutil" +) + +func touch(t *testing.T, name string) { + t.Helper() + f, err := os.Create(name) + if err != nil { + t.Fatalf("error creating file %q: %v", name, err) + } + if err := f.Close(); err != nil { + t.Fatalf("error closing file %q: %v", name, err) + } +} + +func TestSearchEmptyDir(t *testing.T) { + td, err := ioutil.TempDir(testutil.TmpDir(), "searchtest") + if err != nil { + t.Fatalf("error creating searchtest: %v", err) + } + defer os.RemoveAll(td) + + var want []string + + testFilter := regexp.MustCompile(`^test-[^-].+\.tc$`) + got, err := search(td, testFilter) + if err != nil { + t.Errorf("search error: %v", err) + } + + if !reflect.DeepEqual(got, want) { + t.Errorf("Found %#v; want %#v", got, want) + } +} + +func TestSearch(t *testing.T) { + td, err := ioutil.TempDir(testutil.TmpDir(), "searchtest") + if err != nil { + t.Fatalf("error creating searchtest: %v", err) + } + defer os.RemoveAll(td) + + // Creating various files similar to the test filter regex. + files := []string{ + "emp/", + "tee/", + "test-foo.tc", + "test-foo.tc", + "test-bar.tc", + "test-sam.tc", + "Test-que.tc", + "test-brett", + "test--abc.tc", + "test---xyz.tc", + "test-bool.TC", + "--test-gvs.tc", + " test-pew.tc", + "dir/test_baz.tc", + "dir/testsnap.tc", + "dir/test-luk.tc", + "dir/nest/test-ok.tc", + "dir/dip/diz/goog/test-pack.tc", + "dir/dip/diz/wobble/thud/test-cas.e", + "dir/dip/diz/wobble/thud/test-cas.tc", + } + want := []string{ + "dir/dip/diz/goog/test-pack.tc", + "dir/dip/diz/wobble/thud/test-cas.tc", + "dir/nest/test-ok.tc", + "dir/test-luk.tc", + "test-bar.tc", + "test-foo.tc", + "test-sam.tc", + } + + for _, item := range files { + if strings.HasSuffix(item, "/") { + // This item is a directory, create it. + if err := os.MkdirAll(filepath.Join(td, item), 0755); err != nil { + t.Fatalf("error making directory: %v", err) + } + } else { + // This item is a file, create the directory and touch file. + // Create directory in which file should be created + fullDirPath := filepath.Join(td, filepath.Dir(item)) + if err := os.MkdirAll(fullDirPath, 0755); err != nil { + t.Fatalf("error making directory: %v", err) + } + // Create file with full path to file. + touch(t, filepath.Join(td, item)) + } + } + + testFilter := regexp.MustCompile(`^test-[^-].+\.tc$`) + got, err := search(td, testFilter) + if err != nil { + t.Errorf("search error: %v", err) + } + + if !reflect.DeepEqual(got, want) { + t.Errorf("Found %#v; want %#v", got, want) + } +} diff --git a/test/runtimes/proctor/python.go b/test/runtimes/proctor/python.go new file mode 100644 index 000000000..b9e0fbe6f --- /dev/null +++ b/test/runtimes/proctor/python.go @@ -0,0 +1,49 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package main + +import ( + "fmt" + "os" + "os/exec" + "strings" +) + +// pythonRunner implements TestRunner for Python. +type pythonRunner struct{} + +var _ TestRunner = pythonRunner{} + +// ListTests implements TestRunner.ListTests. +func (pythonRunner) ListTests() ([]string, error) { + args := []string{"-m", "test", "--list-tests"} + cmd := exec.Command("./python", args...) + cmd.Stderr = os.Stderr + out, err := cmd.Output() + if err != nil { + return nil, fmt.Errorf("failed to list: %v", err) + } + var toolSlice []string + for _, test := range strings.Split(string(out), "\n") { + toolSlice = append(toolSlice, test) + } + return toolSlice, nil +} + +// TestCmd implements TestRunner.TestCmd. +func (pythonRunner) TestCmd(test string) *exec.Cmd { + args := []string{"-m", "test", test} + return exec.Command("./python", args...) +} diff --git a/test/runtimes/runner.go b/test/runtimes/runner.go deleted file mode 100644 index 3c98f4570..000000000 --- a/test/runtimes/runner.go +++ /dev/null @@ -1,196 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Binary runner runs the runtime tests in a Docker container. -package main - -import ( - "encoding/csv" - "flag" - "fmt" - "io" - "os" - "sort" - "strings" - "testing" - "time" - - "gvisor.dev/gvisor/runsc/dockerutil" - "gvisor.dev/gvisor/runsc/testutil" -) - -var ( - lang = flag.String("lang", "", "language runtime to test") - image = flag.String("image", "", "docker image with runtime tests") - blacklistFile = flag.String("blacklist_file", "", "file containing blacklist of tests to exclude, in CSV format with fields: test name, bug id, comment") -) - -// Wait time for each test to run. -const timeout = 5 * time.Minute - -func main() { - flag.Parse() - if *lang == "" || *image == "" { - fmt.Fprintf(os.Stderr, "lang and image flags must not be empty\n") - os.Exit(1) - } - - os.Exit(runTests()) -} - -// runTests is a helper that is called by main. It exists so that we can run -// defered functions before exiting. It returns an exit code that should be -// passed to os.Exit. -func runTests() int { - // Get tests to blacklist. - blacklist, err := getBlacklist() - if err != nil { - fmt.Fprintf(os.Stderr, "Error getting blacklist: %s\n", err.Error()) - return 1 - } - - // Create a single docker container that will be used for all tests. - d := dockerutil.MakeDocker("gvisor-" + *lang) - defer d.CleanUp() - - // Get a slice of tests to run. This will also start a single Docker - // container that will be used to run each test. The final test will - // stop the Docker container. - tests, err := getTests(d, blacklist) - if err != nil { - fmt.Fprintf(os.Stderr, "%s\n", err.Error()) - return 1 - } - - m := testing.MainStart(testDeps{}, tests, nil, nil) - return m.Run() -} - -// getTests returns a slice of tests to run, subject to the shard size and -// index. -func getTests(d dockerutil.Docker, blacklist map[string]struct{}) ([]testing.InternalTest, error) { - // Pull the image. - if err := dockerutil.Pull(*image); err != nil { - return nil, fmt.Errorf("docker pull %q failed: %v", *image, err) - } - - // Run proctor with --pause flag to keep container alive forever. - if err := d.Run(*image, "--pause"); err != nil { - return nil, fmt.Errorf("docker run failed: %v", err) - } - - // Get a list of all tests in the image. - list, err := d.Exec("/proctor", "--runtime", *lang, "--list") - if err != nil { - return nil, fmt.Errorf("docker exec failed: %v", err) - } - - // Calculate a subset of tests to run corresponding to the current - // shard. - tests := strings.Fields(list) - sort.Strings(tests) - indices, err := testutil.TestIndicesForShard(len(tests)) - if err != nil { - return nil, fmt.Errorf("TestsForShard() failed: %v", err) - } - - var itests []testing.InternalTest - for _, tci := range indices { - // Capture tc in this scope. - tc := tests[tci] - itests = append(itests, testing.InternalTest{ - Name: tc, - F: func(t *testing.T) { - // Is the test blacklisted? - if _, ok := blacklist[tc]; ok { - t.Skipf("SKIP: blacklisted test %q", tc) - } - - var ( - now = time.Now() - done = make(chan struct{}) - output string - err error - ) - - go func() { - fmt.Printf("RUNNING %s...\n", tc) - output, err = d.Exec("/proctor", "--runtime", *lang, "--test", tc) - close(done) - }() - - select { - case <-done: - if err == nil { - fmt.Printf("PASS: %s (%v)\n\n", tc, time.Since(now)) - return - } - t.Errorf("FAIL: %s (%v):\n%s\n", tc, time.Since(now), output) - case <-time.After(timeout): - t.Errorf("TIMEOUT: %s (%v):\n%s\n", tc, time.Since(now), output) - } - }, - }) - } - return itests, nil -} - -// getBlacklist reads the blacklist file and returns a set of test names to -// exclude. -func getBlacklist() (map[string]struct{}, error) { - blacklist := make(map[string]struct{}) - if *blacklistFile == "" { - return blacklist, nil - } - file, err := testutil.FindFile(*blacklistFile) - if err != nil { - return nil, err - } - f, err := os.Open(file) - if err != nil { - return nil, err - } - defer f.Close() - - r := csv.NewReader(f) - - // First line is header. Skip it. - if _, err := r.Read(); err != nil { - return nil, err - } - - for { - record, err := r.Read() - if err == io.EOF { - break - } - if err != nil { - return nil, err - } - blacklist[record[0]] = struct{}{} - } - return blacklist, nil -} - -// testDeps implements testing.testDeps (an unexported interface), and is -// required to use testing.MainStart. -type testDeps struct{} - -func (f testDeps) MatchString(a, b string) (bool, error) { return a == b, nil } -func (f testDeps) StartCPUProfile(io.Writer) error { return nil } -func (f testDeps) StopCPUProfile() {} -func (f testDeps) WriteProfileTo(string, io.Writer, int) error { return nil } -func (f testDeps) ImportPath() string { return "" } -func (f testDeps) StartTestLog(io.Writer) {} -func (f testDeps) StopTestLog() error { return nil } diff --git a/test/runtimes/runner.sh b/test/runtimes/runner.sh deleted file mode 100755 index a8d9a3460..000000000 --- a/test/runtimes/runner.sh +++ /dev/null @@ -1,35 +0,0 @@ -#!/bin/bash - -# Copyright 2018 The gVisor Authors. -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -set -euf -x -o pipefail - -echo -- "$@" - -# Create outputs dir if it does not exist. -if [[ -n "${TEST_UNDECLARED_OUTPUTS_DIR}" ]]; then - mkdir -p "${TEST_UNDECLARED_OUTPUTS_DIR}" - chmod a+rwx "${TEST_UNDECLARED_OUTPUTS_DIR}" -fi - -# Update the timestamp on the shard status file. Bazel looks for this. -touch "${TEST_SHARD_STATUS_FILE}" - -# Get location of runner binary. -readonly runner=$(find "${TEST_SRCDIR}" -name runner) - -# Pass the arguments of this script directly to the runner. -exec "${runner}" "$@" - diff --git a/test/runtimes/runner/BUILD b/test/runtimes/runner/BUILD new file mode 100644 index 000000000..63924b9c5 --- /dev/null +++ b/test/runtimes/runner/BUILD @@ -0,0 +1,21 @@ +load("//tools:defs.bzl", "go_binary", "go_test") + +package(licenses = ["notice"]) + +go_binary( + name = "runner", + testonly = 1, + srcs = ["main.go"], + visibility = ["//test/runtimes:__pkg__"], + deps = [ + "//pkg/test/dockerutil", + "//pkg/test/testutil", + ], +) + +go_test( + name = "blacklist_test", + size = "small", + srcs = ["blacklist_test.go"], + library = ":runner", +) diff --git a/test/runtimes/runner/blacklist_test.go b/test/runtimes/runner/blacklist_test.go new file mode 100644 index 000000000..0ff69ab18 --- /dev/null +++ b/test/runtimes/runner/blacklist_test.go @@ -0,0 +1,37 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package main + +import ( + "flag" + "os" + "testing" +) + +func TestMain(m *testing.M) { + flag.Parse() + os.Exit(m.Run()) +} + +// Test that the blacklist parses without error. +func TestBlacklists(t *testing.T) { + bl, err := getBlacklist() + if err != nil { + t.Fatalf("error parsing blacklist: %v", err) + } + if *blacklistFile != "" && len(bl) == 0 { + t.Errorf("got empty blacklist for file %q", *blacklistFile) + } +} diff --git a/test/runtimes/runner/main.go b/test/runtimes/runner/main.go new file mode 100644 index 000000000..57540e00e --- /dev/null +++ b/test/runtimes/runner/main.go @@ -0,0 +1,189 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Binary runner runs the runtime tests in a Docker container. +package main + +import ( + "encoding/csv" + "flag" + "fmt" + "io" + "os" + "sort" + "strings" + "testing" + "time" + + "gvisor.dev/gvisor/pkg/test/dockerutil" + "gvisor.dev/gvisor/pkg/test/testutil" +) + +var ( + lang = flag.String("lang", "", "language runtime to test") + image = flag.String("image", "", "docker image with runtime tests") + blacklistFile = flag.String("blacklist_file", "", "file containing blacklist of tests to exclude, in CSV format with fields: test name, bug id, comment") +) + +// Wait time for each test to run. +const timeout = 5 * time.Minute + +func main() { + flag.Parse() + if *lang == "" || *image == "" { + fmt.Fprintf(os.Stderr, "lang and image flags must not be empty\n") + os.Exit(1) + } + os.Exit(runTests()) +} + +// runTests is a helper that is called by main. It exists so that we can run +// defered functions before exiting. It returns an exit code that should be +// passed to os.Exit. +func runTests() int { + // Get tests to blacklist. + blacklist, err := getBlacklist() + if err != nil { + fmt.Fprintf(os.Stderr, "Error getting blacklist: %s\n", err.Error()) + return 1 + } + + // Construct the shared docker instance. + d := dockerutil.MakeDocker(testutil.DefaultLogger(*lang)) + defer d.CleanUp() + + // Get a slice of tests to run. This will also start a single Docker + // container that will be used to run each test. The final test will + // stop the Docker container. + tests, err := getTests(d, blacklist) + if err != nil { + fmt.Fprintf(os.Stderr, "%s\n", err.Error()) + return 1 + } + + m := testing.MainStart(testDeps{}, tests, nil, nil) + return m.Run() +} + +// getTests executes all tests as table tests. +func getTests(d *dockerutil.Docker, blacklist map[string]struct{}) ([]testing.InternalTest, error) { + // Start the container. + d.CopyFiles("/proctor", "test/runtimes/proctor/proctor") + if err := d.Spawn(dockerutil.RunOpts{ + Image: fmt.Sprintf("runtimes/%s", *image), + }, "/proctor/proctor", "--pause"); err != nil { + return nil, fmt.Errorf("docker run failed: %v", err) + } + + // Get a list of all tests in the image. + list, err := d.Exec(dockerutil.RunOpts{}, "/proctor/proctor", "--runtime", *lang, "--list") + if err != nil { + return nil, fmt.Errorf("docker exec failed: %v", err) + } + + // Calculate a subset of tests to run corresponding to the current + // shard. + tests := strings.Fields(list) + sort.Strings(tests) + indices, err := testutil.TestIndicesForShard(len(tests)) + if err != nil { + return nil, fmt.Errorf("TestsForShard() failed: %v", err) + } + + var itests []testing.InternalTest + for _, tci := range indices { + // Capture tc in this scope. + tc := tests[tci] + itests = append(itests, testing.InternalTest{ + Name: tc, + F: func(t *testing.T) { + // Is the test blacklisted? + if _, ok := blacklist[tc]; ok { + t.Skipf("SKIP: blacklisted test %q", tc) + } + + var ( + now = time.Now() + done = make(chan struct{}) + output string + err error + ) + + go func() { + fmt.Printf("RUNNING %s...\n", tc) + output, err = d.Exec(dockerutil.RunOpts{}, "/proctor/proctor", "--runtime", *lang, "--test", tc) + close(done) + }() + + select { + case <-done: + if err == nil { + fmt.Printf("PASS: %s (%v)\n\n", tc, time.Since(now)) + return + } + t.Errorf("FAIL: %s (%v):\n%s\n", tc, time.Since(now), output) + case <-time.After(timeout): + t.Errorf("TIMEOUT: %s (%v):\n%s\n", tc, time.Since(now), output) + } + }, + }) + } + + return itests, nil +} + +// getBlacklist reads the blacklist file and returns a set of test names to +// exclude. +func getBlacklist() (map[string]struct{}, error) { + blacklist := make(map[string]struct{}) + if *blacklistFile == "" { + return blacklist, nil + } + f, err := os.Open(*blacklistFile) + if err != nil { + return nil, err + } + defer f.Close() + + r := csv.NewReader(f) + + // First line is header. Skip it. + if _, err := r.Read(); err != nil { + return nil, err + } + + for { + record, err := r.Read() + if err == io.EOF { + break + } + if err != nil { + return nil, err + } + blacklist[record[0]] = struct{}{} + } + return blacklist, nil +} + +// testDeps implements testing.testDeps (an unexported interface), and is +// required to use testing.MainStart. +type testDeps struct{} + +func (f testDeps) MatchString(a, b string) (bool, error) { return a == b, nil } +func (f testDeps) StartCPUProfile(io.Writer) error { return nil } +func (f testDeps) StopCPUProfile() {} +func (f testDeps) WriteProfileTo(string, io.Writer, int) error { return nil } +func (f testDeps) ImportPath() string { return "" } +func (f testDeps) StartTestLog(io.Writer) {} +func (f testDeps) StopTestLog() error { return nil } diff --git a/tools/bazeldefs/defs.bzl b/tools/bazeldefs/defs.bzl index 2207b9b34..3c22aec24 100644 --- a/tools/bazeldefs/defs.bzl +++ b/tools/bazeldefs/defs.bzl @@ -5,18 +5,14 @@ load("@io_bazel_rules_go//go:def.bzl", "GoLibrary", _go_binary = "go_binary", _g load("@io_bazel_rules_go//proto:def.bzl", _go_grpc_library = "go_grpc_library", _go_proto_library = "go_proto_library") load("@rules_cc//cc:defs.bzl", _cc_binary = "cc_binary", _cc_library = "cc_library", _cc_proto_library = "cc_proto_library", _cc_test = "cc_test") load("@rules_pkg//:pkg.bzl", _pkg_deb = "pkg_deb", _pkg_tar = "pkg_tar") -load("@io_bazel_rules_docker//go:image.bzl", _go_image = "go_image") -load("@io_bazel_rules_docker//container:container.bzl", _container_image = "container_image") load("@pydeps//:requirements.bzl", _py_requirement = "requirement") load("@com_github_grpc_grpc//bazel:cc_grpc_library.bzl", _cc_grpc_library = "cc_grpc_library") -container_image = _container_image cc_library = _cc_library cc_flags_supplier = _cc_flags_supplier cc_proto_library = _cc_proto_library cc_test = _cc_test cc_toolchain = "@bazel_tools//tools/cpp:current_cc_toolchain" -go_image = _go_image go_embed_data = _go_embed_data gtest = "@com_google_googletest//:gtest" grpcpp = "@com_github_grpc_grpc//:grpc++" diff --git a/tools/defs.bzl b/tools/defs.bzl index 33240e7f4..cdaf281f3 100644 --- a/tools/defs.bzl +++ b/tools/defs.bzl @@ -7,7 +7,7 @@ change for Google-internal and bazel-compatible rules. load("//tools/go_stateify:defs.bzl", "go_stateify") load("//tools/go_marshal:defs.bzl", "go_marshal", "marshal_deps", "marshal_test_deps") -load("//tools/bazeldefs:defs.bzl", _cc_binary = "cc_binary", _cc_flags_supplier = "cc_flags_supplier", _cc_grpc_library = "cc_grpc_library", _cc_library = "cc_library", _cc_proto_library = "cc_proto_library", _cc_test = "cc_test", _cc_toolchain = "cc_toolchain", _container_image = "container_image", _default_installer = "default_installer", _default_net_util = "default_net_util", _gbenchmark = "gbenchmark", _go_binary = "go_binary", _go_embed_data = "go_embed_data", _go_grpc_and_proto_libraries = "go_grpc_and_proto_libraries", _go_image = "go_image", _go_library = "go_library", _go_proto_library = "go_proto_library", _go_test = "go_test", _grpcpp = "grpcpp", _gtest = "gtest", _loopback = "loopback", _pkg_deb = "pkg_deb", _pkg_tar = "pkg_tar", _proto_library = "proto_library", _py_binary = "py_binary", _py_library = "py_library", _py_requirement = "py_requirement", _py_test = "py_test", _select_arch = "select_arch", _select_system = "select_system") +load("//tools/bazeldefs:defs.bzl", _cc_binary = "cc_binary", _cc_flags_supplier = "cc_flags_supplier", _cc_grpc_library = "cc_grpc_library", _cc_library = "cc_library", _cc_proto_library = "cc_proto_library", _cc_test = "cc_test", _cc_toolchain = "cc_toolchain", _default_installer = "default_installer", _default_net_util = "default_net_util", _gbenchmark = "gbenchmark", _go_binary = "go_binary", _go_embed_data = "go_embed_data", _go_grpc_and_proto_libraries = "go_grpc_and_proto_libraries", _go_library = "go_library", _go_proto_library = "go_proto_library", _go_test = "go_test", _grpcpp = "grpcpp", _gtest = "gtest", _loopback = "loopback", _pkg_deb = "pkg_deb", _pkg_tar = "pkg_tar", _proto_library = "proto_library", _py_binary = "py_binary", _py_library = "py_library", _py_requirement = "py_requirement", _py_test = "py_test", _select_arch = "select_arch", _select_system = "select_system") load("//tools/bazeldefs:platforms.bzl", _default_platform = "default_platform", _platforms = "platforms") load("//tools/bazeldefs:tags.bzl", "go_suffixes") load("//tools/nogo:defs.bzl", "nogo_test") @@ -19,12 +19,10 @@ cc_grpc_library = _cc_grpc_library cc_library = _cc_library cc_test = _cc_test cc_toolchain = _cc_toolchain -container_image = _container_image default_installer = _default_installer default_net_util = _default_net_util gbenchmark = _gbenchmark go_embed_data = _go_embed_data -go_image = _go_image go_test = _go_test gtest = _gtest grpcpp = _grpcpp -- cgit v1.2.3 From 5042ea7e2cbdc0c04fd454583589a3b1e152f95d Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Thu, 23 Apr 2020 15:35:56 -0700 Subject: Add vfs.MkdirOptions.ForSyntheticMountpoint. PiperOrigin-RevId: 308143529 --- pkg/sentry/fsimpl/gofer/directory.go | 147 +++++++++++----- pkg/sentry/fsimpl/gofer/filesystem.go | 323 ++++++++++++++++++++++------------ pkg/sentry/fsimpl/gofer/gofer.go | 177 ++++++++++++------- pkg/sentry/fsimpl/gofer/gofer_test.go | 2 +- pkg/sentry/vfs/filesystem.go | 3 +- pkg/sentry/vfs/options.go | 19 ++ runsc/boot/vfs.go | 6 + 7 files changed, 461 insertions(+), 216 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/directory.go b/pkg/sentry/fsimpl/gofer/directory.go index d02691232..c67766ab2 100644 --- a/pkg/sentry/fsimpl/gofer/directory.go +++ b/pkg/sentry/fsimpl/gofer/directory.go @@ -21,8 +21,10 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/p9" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) func (d *dentry) isDir() bool { @@ -41,15 +43,46 @@ func (d *dentry) cacheNewChildLocked(child *dentry, name string) { d.children[name] = child } -// Preconditions: d.dirMu must be locked. d.isDir(). fs.opts.interop != -// InteropModeShared. -func (d *dentry) cacheNegativeChildLocked(name string) { +// Preconditions: d.dirMu must be locked. d.isDir(). +func (d *dentry) cacheNegativeLookupLocked(name string) { + // Don't cache negative lookups if InteropModeShared is in effect (since + // this makes remote lookup unavoidable), or if d.isSynthetic() (in which + // case the only files in the directory are those for which a dentry exists + // in d.children). Instead, just delete any previously-cached dentry. + if d.fs.opts.interop == InteropModeShared || d.isSynthetic() { + delete(d.children, name) + return + } if d.children == nil { d.children = make(map[string]*dentry) } d.children[name] = nil } +// createSyntheticDirectory creates a synthetic directory with the given name +// in d. +// +// Preconditions: d.dirMu must be locked. d.isDir(). d does not already contain +// a child with the given name. +func (d *dentry) createSyntheticDirectoryLocked(name string, mode linux.FileMode, kuid auth.KUID, kgid auth.KGID) { + d2 := &dentry{ + refs: 1, // held by d + fs: d.fs, + mode: uint32(mode) | linux.S_IFDIR, + uid: uint32(kuid), + gid: uint32(kgid), + blockSize: usermem.PageSize, // arbitrary + handle: handle{ + fd: -1, + }, + } + d2.pf.dentry = d2 + d2.vfsd.Init(d2) + + d.cacheNewChildLocked(d2, name) + d.syntheticChildren++ +} + type directoryFD struct { fileDescription vfs.DirectoryFileDescriptionDefaultImpl @@ -77,7 +110,7 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba fd.dirents = ds } - if d.fs.opts.interop != InteropModeShared { + if d.cachedMetadataAuthoritative() { d.touchAtime(fd.vfsfd.Mount()) } @@ -108,10 +141,10 @@ func (d *dentry) getDirents(ctx context.Context) ([]vfs.Dirent, error) { // filesystem.renameMu is needed for d.parent, and must be locked before // dentry.dirMu. d.fs.renameMu.RLock() + defer d.fs.renameMu.RUnlock() d.dirMu.Lock() defer d.dirMu.Unlock() if d.dirents != nil { - d.fs.renameMu.RUnlock() return d.dirents, nil } @@ -132,51 +165,81 @@ func (d *dentry) getDirents(ctx context.Context) ([]vfs.Dirent, error) { NextOff: 2, }, } - d.fs.renameMu.RUnlock() - off := uint64(0) - const count = 64 * 1024 // for consistency with the vfs1 client - d.handleMu.RLock() - defer d.handleMu.RUnlock() - if !d.handleReadable { - // This should not be possible because a readable handle should have - // been opened when the calling directoryFD was opened. - panic("gofer.dentry.getDirents called without a readable handle") - } - for { - p9ds, err := d.handle.file.readdir(ctx, off, count) - if err != nil { - return nil, err + var realChildren map[string]struct{} + if !d.isSynthetic() { + if d.syntheticChildren != 0 && d.fs.opts.interop == InteropModeShared { + // Record the set of children d actually has so that we don't emit + // duplicate entries for synthetic children. + realChildren = make(map[string]struct{}) } - if len(p9ds) == 0 { - // Cache dirents for future directoryFDs if permitted. - if d.fs.opts.interop != InteropModeShared { - d.dirents = dirents + off := uint64(0) + const count = 64 * 1024 // for consistency with the vfs1 client + d.handleMu.RLock() + if !d.handleReadable { + // This should not be possible because a readable handle should + // have been opened when the calling directoryFD was opened. + d.handleMu.RUnlock() + panic("gofer.dentry.getDirents called without a readable handle") + } + for { + p9ds, err := d.handle.file.readdir(ctx, off, count) + if err != nil { + d.handleMu.RUnlock() + return nil, err + } + if len(p9ds) == 0 { + d.handleMu.RUnlock() + break + } + for _, p9d := range p9ds { + if p9d.Name == "." || p9d.Name == ".." { + continue + } + dirent := vfs.Dirent{ + Name: p9d.Name, + Ino: p9d.QID.Path, + NextOff: int64(len(dirents) + 1), + } + // p9 does not expose 9P2000.U's DMDEVICE, DMNAMEDPIPE, or + // DMSOCKET. + switch p9d.Type { + case p9.TypeSymlink: + dirent.Type = linux.DT_LNK + case p9.TypeDir: + dirent.Type = linux.DT_DIR + default: + dirent.Type = linux.DT_REG + } + dirents = append(dirents, dirent) + if realChildren != nil { + realChildren[p9d.Name] = struct{}{} + } } - return dirents, nil + off = p9ds[len(p9ds)-1].Offset } - for _, p9d := range p9ds { - if p9d.Name == "." || p9d.Name == ".." { + } + // Emit entries for synthetic children. + if d.syntheticChildren != 0 { + for _, child := range d.children { + if child == nil || !child.isSynthetic() { continue } - dirent := vfs.Dirent{ - Name: p9d.Name, - Ino: p9d.QID.Path, - NextOff: int64(len(dirents) + 1), - } - // p9 does not expose 9P2000.U's DMDEVICE, DMNAMEDPIPE, or - // DMSOCKET. - switch p9d.Type { - case p9.TypeSymlink: - dirent.Type = linux.DT_LNK - case p9.TypeDir: - dirent.Type = linux.DT_DIR - default: - dirent.Type = linux.DT_REG + if _, ok := realChildren[child.name]; ok { + continue } - dirents = append(dirents, dirent) + dirents = append(dirents, vfs.Dirent{ + Name: child.name, + Type: uint8(atomic.LoadUint32(&child.mode) >> 12), + Ino: child.ino, + NextOff: int64(len(dirents) + 1), + }) } - off = p9ds[len(p9ds)-1].Offset } + // Cache dirents for future directoryFDs if permitted. + if d.cachedMetadataAuthoritative() { + d.dirents = dirents + } + return dirents, nil } // Seek implements vfs.FileDescriptionImpl.Seek. diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index eba4aabe8..98ccb42fd 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -29,14 +29,16 @@ import ( // Sync implements vfs.FilesystemImpl.Sync. func (fs *filesystem) Sync(ctx context.Context) error { - // Snapshot current dentries and special files. + // Snapshot current syncable dentries and special files. fs.syncMu.Lock() - ds := make([]*dentry, 0, len(fs.dentries)) - for d := range fs.dentries { + ds := make([]*dentry, 0, len(fs.syncableDentries)) + for d := range fs.syncableDentries { + d.IncRef() ds = append(ds, d) } sffds := make([]*specialFileFD, 0, len(fs.specialFileFDs)) for sffd := range fs.specialFileFDs { + sffd.vfsfd.IncRef() sffds = append(sffds, sffd) } fs.syncMu.Unlock() @@ -47,9 +49,6 @@ func (fs *filesystem) Sync(ctx context.Context) error { // Sync regular files. for _, d := range ds { - if !d.TryIncRef() { - continue - } err := d.syncSharedHandle(ctx) d.DecRef() if err != nil && retErr == nil { @@ -60,9 +59,6 @@ func (fs *filesystem) Sync(ctx context.Context) error { // Sync special files, which may be writable but do not use dentry shared // handles (so they won't be synced by the above). for _, sffd := range sffds { - if !sffd.vfsfd.TryIncRef() { - continue - } err := sffd.Sync(ctx) sffd.vfsfd.DecRef() if err != nil && retErr == nil { @@ -114,8 +110,8 @@ func putDentrySlice(ds *[]*dentry) { // to *ds. // // Preconditions: fs.renameMu must be locked. d.dirMu must be locked. -// !rp.Done(). If fs.opts.interop == InteropModeShared, then d's cached -// metadata must be up to date. +// !rp.Done(). If !d.cachedMetadataAuthoritative(), then d's cached metadata +// must be up to date. // // Postconditions: The returned dentry's cached metadata is up to date. func (fs *filesystem) stepLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry, ds **[]*dentry) (*dentry, error) { @@ -148,7 +144,7 @@ afterSymlink: if err := rp.CheckMount(&d.parent.vfsd); err != nil { return nil, err } - if fs.opts.interop == InteropModeShared && d != d.parent { + if d != d.parent && !d.cachedMetadataAuthoritative() { _, attrMask, attr, err := d.parent.file.getAttr(ctx, dentryAttrMask()) if err != nil { return nil, err @@ -195,7 +191,7 @@ func (fs *filesystem) getChildLocked(ctx context.Context, vfsObj *vfs.VirtualFil return nil, syserror.ENAMETOOLONG } child, ok := parent.children[name] - if ok && fs.opts.interop != InteropModeShared { + if (ok && fs.opts.interop != InteropModeShared) || parent.isSynthetic() { // Whether child is nil or not, it is cached information that is // assumed to be correct. return child, nil @@ -206,7 +202,7 @@ func (fs *filesystem) getChildLocked(ctx context.Context, vfsObj *vfs.VirtualFil return fs.revalidateChildLocked(ctx, vfsObj, parent, name, child, ds) } -// Preconditions: As for getChildLocked. +// Preconditions: As for getChildLocked. !parent.isSynthetic(). func (fs *filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.VirtualFilesystem, parent *dentry, name string, child *dentry, ds **[]*dentry) (*dentry, error) { qid, file, attrMask, attr, err := parent.file.walkGetAttrOne(ctx, name) if err != nil && err != syserror.ENOENT { @@ -220,24 +216,41 @@ func (fs *filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.Vir child.updateFromP9Attrs(attrMask, &attr) return child, nil } - // The file at this path has changed or no longer exists. Remove - // the stale dentry from the tree, and re-evaluate its caching - // status (i.e. if it has 0 references, drop it). + if file.isNil() && child.isSynthetic() { + // We have a synthetic file, and no remote file has arisen to + // replace it. + return child, nil + } + // The file at this path has changed or no longer exists. Mark the + // dentry invalidated, and re-evaluate its caching status (i.e. if it + // has 0 references, drop it). Wait to update parent.children until we + // know what to replace the existing dentry with (i.e. one of the + // returns below), to avoid a redundant map access. vfsObj.InvalidateDentry(&child.vfsd) + if child.isSynthetic() { + // Normally we don't mark invalidated dentries as deleted since + // they may still exist (but at a different path), and also for + // consistency with Linux. However, synthetic files are guaranteed + // to become unreachable if their dentries are invalidated, so + // treat their invalidation as deletion. + child.setDeleted() + parent.syntheticChildren-- + child.decRefLocked() + parent.dirents = nil + } *ds = appendDentry(*ds, child) } if file.isNil() { // No file exists at this path now. Cache the negative lookup if // allowed. - if fs.opts.interop != InteropModeShared { - parent.cacheNegativeChildLocked(name) - } + parent.cacheNegativeLookupLocked(name) return nil, nil } // Create a new dentry representing the file. child, err = fs.newDentry(ctx, file, qid, attrMask, &attr) if err != nil { file.close(ctx) + delete(parent.children, name) return nil, err } parent.cacheNewChildLocked(child, name) @@ -252,8 +265,9 @@ func (fs *filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.Vir // rp.Start().Impl().(*dentry)). It does not check that the returned directory // is searchable by the provider of rp. // -// Preconditions: fs.renameMu must be locked. !rp.Done(). If fs.opts.interop == -// InteropModeShared, then d's cached metadata must be up to date. +// Preconditions: fs.renameMu must be locked. !rp.Done(). If +// !d.cachedMetadataAuthoritative(), then d's cached metadata must be up to +// date. func (fs *filesystem) walkParentDirLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry, ds **[]*dentry) (*dentry, error) { for !rp.Final() { d.dirMu.Lock() @@ -275,7 +289,7 @@ func (fs *filesystem) walkParentDirLocked(ctx context.Context, rp *vfs.Resolving // Preconditions: fs.renameMu must be locked. func (fs *filesystem) resolveLocked(ctx context.Context, rp *vfs.ResolvingPath, ds **[]*dentry) (*dentry, error) { d := rp.Start().Impl().(*dentry) - if fs.opts.interop == InteropModeShared { + if !d.cachedMetadataAuthoritative() { // Get updated metadata for rp.Start() as required by fs.stepLocked(). if err := d.updateFromGetattr(ctx); err != nil { return nil, err @@ -297,16 +311,17 @@ func (fs *filesystem) resolveLocked(ctx context.Context, rp *vfs.ResolvingPath, } // doCreateAt checks that creating a file at rp is permitted, then invokes -// create to do so. +// createInRemoteDir (if the parent directory is a real remote directory) or +// createInSyntheticDir (if the parent directory is synthetic) to do so. // // Preconditions: !rp.Done(). For the final path component in rp, // !rp.ShouldFollowSymlink(). -func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir bool, create func(parent *dentry, name string) error) error { +func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir bool, createInRemoteDir func(parent *dentry, name string) error, createInSyntheticDir func(parent *dentry, name string) error) error { var ds *[]*dentry fs.renameMu.RLock() defer fs.renameMuRUnlockAndCheckCaching(&ds) start := rp.Start().Impl().(*dentry) - if fs.opts.interop == InteropModeShared { + if !start.cachedMetadataAuthoritative() { // Get updated metadata for start as required by // fs.walkParentDirLocked(). if err := start.updateFromGetattr(ctx); err != nil { @@ -340,6 +355,20 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir defer mnt.EndWrite() parent.dirMu.Lock() defer parent.dirMu.Unlock() + if parent.isSynthetic() { + if child := parent.children[name]; child != nil { + return syserror.EEXIST + } + if createInSyntheticDir == nil { + return syserror.EPERM + } + if err := createInSyntheticDir(parent, name); err != nil { + return err + } + parent.touchCMtime() + parent.dirents = nil + return nil + } if fs.opts.interop == InteropModeShared { // The existence of a dentry at name would be inconclusive because the // file it represents may have been deleted from the remote filesystem, @@ -348,21 +377,21 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir // will fail with EEXIST like we would have. If the RPC succeeds, and a // stale dentry exists, the dentry will fail revalidation next time // it's used. - return create(parent, name) + return createInRemoteDir(parent, name) } if child := parent.children[name]; child != nil { return syserror.EEXIST } // No cached dentry exists; however, there might still be an existing file // at name. As above, we attempt the file creation RPC anyway. - if err := create(parent, name); err != nil { + if err := createInRemoteDir(parent, name); err != nil { return err } + if child, ok := parent.children[name]; ok && child == nil { + // Delete the now-stale negative dentry. + delete(parent.children, name) + } parent.touchCMtime() - // Either parent.children[name] doesn't exist (in which case this is a - // no-op) or is nil (in which case this erases the now-stale information - // that the file doesn't exist). - delete(parent.children, name) parent.dirents = nil return nil } @@ -373,7 +402,7 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b fs.renameMu.RLock() defer fs.renameMuRUnlockAndCheckCaching(&ds) start := rp.Start().Impl().(*dentry) - if fs.opts.interop == InteropModeShared { + if !start.cachedMetadataAuthoritative() { // Get updated metadata for start as required by // fs.walkParentDirLocked(). if err := start.updateFromGetattr(ctx); err != nil { @@ -421,8 +450,10 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b // only revalidating the dentry if that fails (indicating that the existing // dentry is a mount point). if child != nil { + child.dirMu.Lock() + defer child.dirMu.Unlock() if err := vfsObj.PrepareDeleteDentry(mntns, &child.vfsd); err != nil { - if fs.opts.interop != InteropModeShared { + if parent.cachedMetadataAuthoritative() { return err } child, err = fs.revalidateChildLocked(ctx, vfsObj, parent, name, child, &ds) @@ -437,13 +468,37 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b } } flags := uint32(0) + // If a dentry exists, use it for best-effort checks on its deletability. if dir { - if child != nil && !child.isDir() { - vfsObj.AbortDeleteDentry(&child.vfsd) - return syserror.ENOTDIR + if child != nil { + // child must be an empty directory. + if child.syntheticChildren != 0 { + // This is definitely not an empty directory, irrespective of + // fs.opts.interop. + vfsObj.AbortDeleteDentry(&child.vfsd) + return syserror.ENOTEMPTY + } + // If InteropModeShared is in effect and the first call to + // PrepareDeleteDentry above succeeded, then child wasn't + // revalidated (so we can't expect its file type to be correct) and + // individually revalidating its children (to confirm that they + // still exist) would be a waste of time. + if child.cachedMetadataAuthoritative() { + if !child.isDir() { + vfsObj.AbortDeleteDentry(&child.vfsd) + return syserror.ENOTDIR + } + for _, grandchild := range child.children { + if grandchild != nil { + vfsObj.AbortDeleteDentry(&child.vfsd) + return syserror.ENOTEMPTY + } + } + } } flags = linux.AT_REMOVEDIR } else { + // child must be a non-directory file. if child != nil && child.isDir() { vfsObj.AbortDeleteDentry(&child.vfsd) return syserror.EISDIR @@ -455,28 +510,36 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b return syserror.ENOTDIR } } - err = parent.file.unlinkAt(ctx, name, flags) - if err != nil { - if child != nil { - vfsObj.AbortDeleteDentry(&child.vfsd) - } - return err - } - if fs.opts.interop != InteropModeShared { - parent.touchCMtime() - if dir { - parent.decLinks() + if parent.isSynthetic() { + if child == nil { + return syserror.ENOENT } - parent.cacheNegativeChildLocked(name) - parent.dirents = nil } else { - delete(parent.children, name) + err = parent.file.unlinkAt(ctx, name, flags) + if err != nil { + if child != nil { + vfsObj.AbortDeleteDentry(&child.vfsd) + } + return err + } } if child != nil { - child.setDeleted() vfsObj.CommitDeleteDentry(&child.vfsd) + child.setDeleted() + if child.isSynthetic() { + parent.syntheticChildren-- + child.decRefLocked() + } ds = appendDentry(ds, child) } + parent.cacheNegativeLookupLocked(name) + if parent.cachedMetadataAuthoritative() { + parent.dirents = nil + parent.touchCMtime() + if dir { + parent.decLinks() + } + } return nil } @@ -554,7 +617,7 @@ func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPa fs.renameMu.RLock() defer fs.renameMuRUnlockAndCheckCaching(&ds) start := rp.Start().Impl().(*dentry) - if fs.opts.interop == InteropModeShared { + if !start.cachedMetadataAuthoritative() { // Get updated metadata for start as required by // fs.walkParentDirLocked(). if err := start.updateFromGetattr(ctx); err != nil { @@ -577,20 +640,32 @@ func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. } // 9P2000.L supports hard links, but we don't. return syserror.EPERM - }) + }, nil) } // MkdirAt implements vfs.FilesystemImpl.MkdirAt. func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error { + creds := rp.Credentials() return fs.doCreateAt(ctx, rp, true /* dir */, func(parent *dentry, name string) error { - creds := rp.Credentials() if _, err := parent.file.mkdir(ctx, name, (p9.FileMode)(opts.Mode), (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)); err != nil { - return err + if !opts.ForSyntheticMountpoint || err == syserror.EEXIST { + return err + } + ctx.Infof("Failed to create remote directory %q: %v; falling back to synthetic directory", name, err) + parent.createSyntheticDirectoryLocked(name, opts.Mode, creds.EffectiveKUID, creds.EffectiveKGID) } if fs.opts.interop != InteropModeShared { parent.incLinks() } return nil + }, func(parent *dentry, name string) error { + if !opts.ForSyntheticMountpoint { + // Can't create non-synthetic files in synthetic directories. + return syserror.EPERM + } + parent.createSyntheticDirectoryLocked(name, opts.Mode, creds.EffectiveKUID, creds.EffectiveKGID) + parent.incLinks() + return nil }) } @@ -600,7 +675,7 @@ func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v creds := rp.Credentials() _, err := parent.file.mknod(ctx, name, (p9.FileMode)(opts.Mode), opts.DevMajor, opts.DevMinor, (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)) return err - }) + }, nil) } // OpenAt implements vfs.FilesystemImpl.OpenAt. @@ -620,7 +695,7 @@ func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf defer fs.renameMuRUnlockAndCheckCaching(&ds) start := rp.Start().Impl().(*dentry) - if fs.opts.interop == InteropModeShared { + if !start.cachedMetadataAuthoritative() { // Get updated metadata for start as required by fs.stepLocked(). if err := start.updateFromGetattr(ctx); err != nil { return nil, err @@ -643,6 +718,10 @@ afterTrailingSymlink: parent.dirMu.Lock() child, err := fs.stepLocked(ctx, rp, parent, &ds) if err == syserror.ENOENT && mayCreate { + if parent.isSynthetic() { + parent.dirMu.Unlock() + return nil, syserror.EPERM + } fd, err := parent.createAndOpenChildLocked(ctx, rp, &opts) parent.dirMu.Unlock() return fd, err @@ -702,8 +781,10 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vf if opts.Flags&linux.O_DIRECT != 0 { return nil, syserror.EINVAL } - if err := d.ensureSharedHandle(ctx, ats&vfs.MayRead != 0, false /* write */, false /* trunc */); err != nil { - return nil, err + if !d.isSynthetic() { + if err := d.ensureSharedHandle(ctx, ats&vfs.MayRead != 0, false /* write */, false /* trunc */); err != nil { + return nil, err + } } fd := &directoryFD{} if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { @@ -733,6 +814,7 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vf } // Preconditions: d.fs.renameMu must be locked. d.dirMu must be locked. +// !d.isSynthetic(). func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.OpenOptions) (*vfs.FileDescription, error) { if err := d.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil { return nil, err @@ -811,7 +893,7 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving child.refs = 1 // Insert the dentry into the tree. d.cacheNewChildLocked(child, name) - if d.fs.opts.interop != InteropModeShared { + if d.cachedMetadataAuthoritative() { d.touchCMtime() d.dirents = nil } @@ -888,7 +970,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa defer mnt.EndWrite() oldParent := oldParentVD.Dentry().Impl().(*dentry) - if fs.opts.interop == InteropModeShared { + if !oldParent.cachedMetadataAuthoritative() { if err := oldParent.updateFromGetattr(ctx); err != nil { return err } @@ -933,35 +1015,22 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa if newParent.isDeleted() { return syserror.ENOENT } - replaced := newParent.children[newName] - // This is similar to unlinkAt, except: - // - // - If a dentry exists for the file to be replaced, we revalidate it - // unconditionally (instead of only if PrepareRenameDentry fails) for - // simplicity. - // - // - If rp.MustBeDir(), then we need a dentry representing the replaced - // file regardless to confirm that it's a directory. - if replaced != nil || rp.MustBeDir() { - replaced, err = fs.getChildLocked(ctx, rp.VirtualFilesystem(), newParent, newName, &ds) - if err != nil { - return err - } - if replaced != nil { - if replaced.isDir() { - if !renamed.isDir() { - return syserror.EISDIR - } - } else { - if rp.MustBeDir() || renamed.isDir() { - return syserror.ENOTDIR - } - } - } + replaced, err := fs.getChildLocked(ctx, rp.VirtualFilesystem(), newParent, newName, &ds) + if err != nil { + return err } var replacedVFSD *vfs.Dentry if replaced != nil { replacedVFSD = &replaced.vfsd + if replaced.isDir() { + if !renamed.isDir() { + return syserror.EISDIR + } + } else { + if rp.MustBeDir() || renamed.isDir() { + return syserror.ENOTDIR + } + } } if oldParent == newParent && oldName == newName { @@ -972,27 +1041,47 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa if err := vfsObj.PrepareRenameDentry(mntns, &renamed.vfsd, replacedVFSD); err != nil { return err } - if err := renamed.file.rename(ctx, newParent.file, newName); err != nil { - vfsObj.AbortRenameDentry(&renamed.vfsd, replacedVFSD) - return err + + // Update the remote filesystem. + if !renamed.isSynthetic() { + if err := renamed.file.rename(ctx, newParent.file, newName); err != nil { + vfsObj.AbortRenameDentry(&renamed.vfsd, replacedVFSD) + return err + } + } else if replaced != nil && !replaced.isSynthetic() { + // We are replacing an existing real file with a synthetic one, so we + // need to unlink the former. + flags := uint32(0) + if replaced.isDir() { + flags = linux.AT_REMOVEDIR + } + if err := newParent.file.unlinkAt(ctx, newName, flags); err != nil { + vfsObj.AbortRenameDentry(&renamed.vfsd, replacedVFSD) + return err + } } - if fs.opts.interop != InteropModeShared { - oldParent.cacheNegativeChildLocked(oldName) - oldParent.dirents = nil - newParent.dirents = nil - if renamed.isDir() { - oldParent.decLinks() - newParent.incLinks() + + // Update the dentry tree. + vfsObj.CommitRenameReplaceDentry(&renamed.vfsd, replacedVFSD) + if replaced != nil { + replaced.setDeleted() + if replaced.isSynthetic() { + newParent.syntheticChildren-- + replaced.decRefLocked() } - oldParent.touchCMtime() - newParent.touchCMtime() - renamed.touchCtime() - } else { - delete(oldParent.children, oldName) + ds = appendDentry(ds, replaced) } + oldParent.cacheNegativeLookupLocked(oldName) + // We don't use newParent.cacheNewChildLocked() since we don't want to mess + // with reference counts and queue oldParent for checkCachingLocked if the + // parent isn't actually changing. if oldParent != newParent { - appendDentry(ds, oldParent) + ds = appendDentry(ds, oldParent) newParent.IncRef() + if renamed.isSynthetic() { + oldParent.syntheticChildren-- + newParent.syntheticChildren++ + } } renamed.parent = newParent renamed.name = newName @@ -1000,11 +1089,25 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa newParent.children = make(map[string]*dentry) } newParent.children[newName] = renamed - if replaced != nil { - replaced.setDeleted() - appendDentry(ds, replaced) + + // Update metadata. + if renamed.cachedMetadataAuthoritative() { + renamed.touchCtime() + } + if oldParent.cachedMetadataAuthoritative() { + oldParent.dirents = nil + oldParent.touchCMtime() + if renamed.isDir() { + oldParent.decLinks() + } + } + if newParent.cachedMetadataAuthoritative() { + newParent.dirents = nil + newParent.touchCMtime() + if renamed.isDir() { + newParent.incLinks() + } } - vfsObj.CommitRenameReplaceDentry(&renamed.vfsd, replacedVFSD) return nil } @@ -1051,6 +1154,10 @@ func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linu if err != nil { return linux.Statfs{}, err } + // If d is synthetic, invoke statfs on the first ancestor of d that isn't. + for d.isSynthetic() { + d = d.parent + } fsstat, err := d.file.statFS(ctx) if err != nil { return linux.Statfs{}, err @@ -1080,7 +1187,7 @@ func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, targ creds := rp.Credentials() _, err := parent.file.symlink(ctx, target, name, (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)) return err - }) + }, nil) } // UnlinkAt implements vfs.FilesystemImpl.UnlinkAt. diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 293df2545..8b4e91d17 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -27,8 +27,9 @@ // dentry.handleMu // dentry.dataMu // -// Locking dentry.dirMu in multiple dentries requires holding -// filesystem.renameMu for writing. +// Locking dentry.dirMu in multiple dentries requires that either ancestor +// dentries are locked before descendant dentries, or that filesystem.renameMu +// is locked for writing. package gofer import ( @@ -102,11 +103,12 @@ type filesystem struct { cachedDentries dentryList cachedDentriesLen uint64 - // dentries contains all dentries in this filesystem. specialFileFDs - // contains all open specialFileFDs. These fields are protected by syncMu. - syncMu sync.Mutex - dentries map[*dentry]struct{} - specialFileFDs map[*specialFileFD]struct{} + // syncableDentries contains all dentries in this filesystem for which + // !dentry.file.isNil(). specialFileFDs contains all open specialFileFDs. + // These fields are protected by syncMu. + syncMu sync.Mutex + syncableDentries map[*dentry]struct{} + specialFileFDs map[*specialFileFD]struct{} } type filesystemOptions struct { @@ -187,7 +189,8 @@ const ( // InteropModeShared is appropriate when there are users of the remote // filesystem that may mutate its state other than the client. // - // - The client must verify cached filesystem state before using it. + // - The client must verify ("revalidate") cached filesystem state before + // using it. // // - Client changes to filesystem state must be sent to the remote // filesystem synchronously. @@ -376,14 +379,14 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt // Construct the filesystem object. fs := &filesystem{ - mfp: mfp, - opts: fsopts, - uid: creds.EffectiveKUID, - gid: creds.EffectiveKGID, - client: client, - clock: ktime.RealtimeClockFromContext(ctx), - dentries: make(map[*dentry]struct{}), - specialFileFDs: make(map[*specialFileFD]struct{}), + mfp: mfp, + opts: fsopts, + uid: creds.EffectiveKUID, + gid: creds.EffectiveKGID, + client: client, + clock: ktime.RealtimeClockFromContext(ctx), + syncableDentries: make(map[*dentry]struct{}), + specialFileFDs: make(map[*specialFileFD]struct{}), } fs.vfsfs.Init(vfsObj, &fstype, fs) @@ -409,7 +412,7 @@ func (fs *filesystem) Release() { mf := fs.mfp.MemoryFile() fs.syncMu.Lock() - for d := range fs.dentries { + for d := range fs.syncableDentries { d.handleMu.Lock() d.dataMu.Lock() if d.handleWritable { @@ -444,9 +447,11 @@ type dentry struct { vfsd vfs.Dentry // refs is the reference count. Each dentry holds a reference on its - // parent, even if disowned. refs is accessed using atomic memory - // operations. When refs reaches 0, the dentry may be added to the cache or - // destroyed. If refs==-1 the dentry has already been destroyed. + // parent, even if disowned. An additional reference is held on all + // synthetic dentries until they are unlinked or invalidated. When refs + // reaches 0, the dentry may be added to the cache or destroyed. If refs == + // -1, the dentry has already been destroyed. refs is accessed using atomic + // memory operations. refs int64 // fs is the owning filesystem. fs is immutable. @@ -465,6 +470,12 @@ type dentry struct { // We don't support hard links, so each dentry maps 1:1 to an inode. // file is the unopened p9.File that backs this dentry. file is immutable. + // + // If file.isNil(), this dentry represents a synthetic file, i.e. a file + // that does not exist on the remote filesystem. As of this writing, this + // is only possible for a directory created with + // MkdirOptions.ForSyntheticMountpoint == true. + // TODO(gvisor.dev/issue/1476): Support synthetic sockets (and pipes). file p9file // If deleted is non-zero, the file represented by this dentry has been @@ -484,15 +495,21 @@ type dentry struct { // - Mappings of child filenames to dentries representing those children. // // - Mappings of child filenames that are known not to exist to nil - // dentries (only if InteropModeShared is not in effect). + // dentries (only if InteropModeShared is not in effect and the directory + // is not synthetic). // // children is protected by dirMu. children map[string]*dentry - // If this dentry represents a directory, InteropModeShared is not in - // effect, and dirents is not nil, it is a cache of all entries in the - // directory, in the order they were returned by the server. dirents is - // protected by dirMu. + // If this dentry represents a directory, syntheticChildren is the number + // of child dentries for which dentry.isSynthetic() == true. + // syntheticChildren is protected by dirMu. + syntheticChildren int + + // If this dentry represents a directory, + // dentry.cachedMetadataAuthoritative() == true, and dirents is not nil, it + // is a cache of all entries in the directory, in the order they were + // returned by the server. dirents is protected by dirMu. dirents []vfs.Dirent // Cached metadata; protected by metadataMu and accessed using atomic @@ -589,6 +606,8 @@ func dentryAttrMask() p9.AttrMask { // initially has no references, but is not cached; it is the caller's // responsibility to set the dentry's reference count and/or call // dentry.checkCachingLocked() as appropriate. +// +// Preconditions: !file.isNil(). func (fs *filesystem) newDentry(ctx context.Context, file p9file, qid p9.QID, mask p9.AttrMask, attr *p9.Attr) (*dentry, error) { if !mask.Mode { ctx.Warningf("can't create gofer.dentry without file type") @@ -612,10 +631,10 @@ func (fs *filesystem) newDentry(ctx context.Context, file p9file, qid p9.QID, ma }, } d.pf.dentry = d - if mask.UID { + if mask.UID && attr.UID != auth.NoID { d.uid = uint32(attr.UID) } - if mask.GID { + if mask.GID && attr.GID != auth.NoID { d.gid = uint32(attr.GID) } if mask.Size { @@ -642,11 +661,19 @@ func (fs *filesystem) newDentry(ctx context.Context, file p9file, qid p9.QID, ma d.vfsd.Init(d) fs.syncMu.Lock() - fs.dentries[d] = struct{}{} + fs.syncableDentries[d] = struct{}{} fs.syncMu.Unlock() return d, nil } +func (d *dentry) isSynthetic() bool { + return d.file.isNil() +} + +func (d *dentry) cachedMetadataAuthoritative() bool { + return d.fs.opts.interop != InteropModeShared || d.isSynthetic() +} + // updateFromP9Attrs is called to update d's metadata after an update from the // remote filesystem. func (d *dentry) updateFromP9Attrs(mask p9.AttrMask, attr *p9.Attr) { @@ -691,6 +718,7 @@ func (d *dentry) updateFromP9Attrs(mask p9.AttrMask, attr *p9.Attr) { d.metadataMu.Unlock() } +// Preconditions: !d.isSynthetic() func (d *dentry) updateFromGetattr(ctx context.Context) error { // Use d.handle.file, which represents a 9P fid that has been opened, in // preference to d.file, which represents a 9P fid that has not. This may @@ -758,7 +786,7 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin defer mnt.EndWrite() setLocalAtime := false setLocalMtime := false - if d.fs.opts.interop != InteropModeShared { + if d.cachedMetadataAuthoritative() { // Timestamp updates will be handled locally. setLocalAtime = stat.Mask&linux.STATX_ATIME != 0 setLocalMtime = stat.Mask&linux.STATX_MTIME != 0 @@ -771,35 +799,37 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin } d.metadataMu.Lock() defer d.metadataMu.Unlock() - if stat.Mask != 0 { - if err := d.file.setAttr(ctx, p9.SetAttrMask{ - Permissions: stat.Mask&linux.STATX_MODE != 0, - UID: stat.Mask&linux.STATX_UID != 0, - GID: stat.Mask&linux.STATX_GID != 0, - Size: stat.Mask&linux.STATX_SIZE != 0, - ATime: stat.Mask&linux.STATX_ATIME != 0, - MTime: stat.Mask&linux.STATX_MTIME != 0, - ATimeNotSystemTime: stat.Atime.Nsec != linux.UTIME_NOW, - MTimeNotSystemTime: stat.Mtime.Nsec != linux.UTIME_NOW, - }, p9.SetAttr{ - Permissions: p9.FileMode(stat.Mode), - UID: p9.UID(stat.UID), - GID: p9.GID(stat.GID), - Size: stat.Size, - ATimeSeconds: uint64(stat.Atime.Sec), - ATimeNanoSeconds: uint64(stat.Atime.Nsec), - MTimeSeconds: uint64(stat.Mtime.Sec), - MTimeNanoSeconds: uint64(stat.Mtime.Nsec), - }); err != nil { - return err + if !d.isSynthetic() { + if stat.Mask != 0 { + if err := d.file.setAttr(ctx, p9.SetAttrMask{ + Permissions: stat.Mask&linux.STATX_MODE != 0, + UID: stat.Mask&linux.STATX_UID != 0, + GID: stat.Mask&linux.STATX_GID != 0, + Size: stat.Mask&linux.STATX_SIZE != 0, + ATime: stat.Mask&linux.STATX_ATIME != 0, + MTime: stat.Mask&linux.STATX_MTIME != 0, + ATimeNotSystemTime: stat.Atime.Nsec != linux.UTIME_NOW, + MTimeNotSystemTime: stat.Mtime.Nsec != linux.UTIME_NOW, + }, p9.SetAttr{ + Permissions: p9.FileMode(stat.Mode), + UID: p9.UID(stat.UID), + GID: p9.GID(stat.GID), + Size: stat.Size, + ATimeSeconds: uint64(stat.Atime.Sec), + ATimeNanoSeconds: uint64(stat.Atime.Nsec), + MTimeSeconds: uint64(stat.Mtime.Sec), + MTimeNanoSeconds: uint64(stat.Mtime.Nsec), + }); err != nil { + return err + } + } + if d.fs.opts.interop == InteropModeShared { + // There's no point to updating d's metadata in this case since + // it'll be overwritten by revalidation before the next time it's + // used anyway. (InteropModeShared inhibits client caching of + // regular file data, so there's no cache to truncate either.) + return nil } - } - if d.fs.opts.interop == InteropModeShared { - // There's no point to updating d's metadata in this case since it'll - // be overwritten by revalidation before the next time it's used - // anyway. (InteropModeShared inhibits client caching of regular file - // data, so there's no cache to truncate either.) - return nil } now := d.fs.clock.Now().Nanoseconds() if stat.Mask&linux.STATX_MODE != 0 { @@ -897,6 +927,15 @@ func (d *dentry) DecRef() { } } +// decRefLocked decrements d's reference count without calling +// d.checkCachingLocked, even if d's reference count reaches 0; callers are +// responsible for ensuring that d.checkCachingLocked will be called later. +func (d *dentry) decRefLocked() { + if refs := atomic.AddInt64(&d.refs, -1); refs < 0 { + panic("gofer.dentry.decRefLocked() called without holding a reference") + } +} + // checkCachingLocked should be called after d's reference count becomes 0 or it // becomes disowned. // @@ -1013,11 +1052,11 @@ func (d *dentry) destroyLocked() { if !d.file.isNil() { d.file.close(ctx) d.file = p9file{} + // Remove d from the set of syncable dentries. + d.fs.syncMu.Lock() + delete(d.fs.syncableDentries, d) + d.fs.syncMu.Unlock() } - // Remove d from the set of all dentries. - d.fs.syncMu.Lock() - delete(d.fs.dentries, d) - d.fs.syncMu.Unlock() // Drop the reference held by d on its parent without recursively locking // d.fs.renameMu. if d.parent != nil { @@ -1040,6 +1079,9 @@ func (d *dentry) setDeleted() { // We only support xattrs prefixed with "user." (see b/148380782). Currently, // there is no need to expose any other xattrs through a gofer. func (d *dentry) listxattr(ctx context.Context, creds *auth.Credentials, size uint64) ([]string, error) { + if d.file.isNil() { + return nil, nil + } xattrMap, err := d.file.listXattr(ctx, size) if err != nil { return nil, err @@ -1054,6 +1096,9 @@ func (d *dentry) listxattr(ctx context.Context, creds *auth.Credentials, size ui } func (d *dentry) getxattr(ctx context.Context, creds *auth.Credentials, opts *vfs.GetxattrOptions) (string, error) { + if d.file.isNil() { + return "", syserror.ENODATA + } if err := d.checkPermissions(creds, vfs.MayRead); err != nil { return "", err } @@ -1064,6 +1109,9 @@ func (d *dentry) getxattr(ctx context.Context, creds *auth.Credentials, opts *vf } func (d *dentry) setxattr(ctx context.Context, creds *auth.Credentials, opts *vfs.SetxattrOptions) error { + if d.file.isNil() { + return syserror.EPERM + } if err := d.checkPermissions(creds, vfs.MayWrite); err != nil { return err } @@ -1074,6 +1122,9 @@ func (d *dentry) setxattr(ctx context.Context, creds *auth.Credentials, opts *vf } func (d *dentry) removexattr(ctx context.Context, creds *auth.Credentials, name string) error { + if d.file.isNil() { + return syserror.EPERM + } if err := d.checkPermissions(creds, vfs.MayWrite); err != nil { return err } @@ -1083,7 +1134,7 @@ func (d *dentry) removexattr(ctx context.Context, creds *auth.Credentials, name return d.file.removeXattr(ctx, name) } -// Preconditions: d.isRegularFile() || d.isDirectory(). +// Preconditions: !d.file.isNil(). d.isRegularFile() || d.isDirectory(). func (d *dentry) ensureSharedHandle(ctx context.Context, read, write, trunc bool) error { // O_TRUNC unconditionally requires us to obtain a new handle (opened with // O_TRUNC). @@ -1213,7 +1264,7 @@ func (fd *fileDescription) dentry() *dentry { func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { d := fd.dentry() const validMask = uint32(linux.STATX_MODE | linux.STATX_UID | linux.STATX_GID | linux.STATX_ATIME | linux.STATX_MTIME | linux.STATX_CTIME | linux.STATX_SIZE | linux.STATX_BLOCKS | linux.STATX_BTIME) - if d.fs.opts.interop == InteropModeShared && opts.Mask&(validMask) != 0 && opts.Sync != linux.AT_STATX_DONT_SYNC { + if !d.cachedMetadataAuthoritative() && opts.Mask&validMask != 0 && opts.Sync != linux.AT_STATX_DONT_SYNC { // TODO(jamieliu): Use specialFileFD.handle.file for the getattr if // available? if err := d.updateFromGetattr(ctx); err != nil { diff --git a/pkg/sentry/fsimpl/gofer/gofer_test.go b/pkg/sentry/fsimpl/gofer/gofer_test.go index 4041fb252..adff39490 100644 --- a/pkg/sentry/fsimpl/gofer/gofer_test.go +++ b/pkg/sentry/fsimpl/gofer/gofer_test.go @@ -24,7 +24,7 @@ import ( func TestDestroyIdempotent(t *testing.T) { fs := filesystem{ - dentries: make(map[*dentry]struct{}), + syncableDentries: make(map[*dentry]struct{}), opts: filesystemOptions{ // Test relies on no dentry being held in the cache. maxCachedDentries: 0, diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index 74577bc2f..20e5bb072 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -443,8 +443,7 @@ type FilesystemImpl interface { // Errors: // // - If extended attributes are not supported by the filesystem, - // ListxattrAt returns nil. (See FileDescription.Listxattr for an - // explanation.) + // ListxattrAt returns ENOTSUP. // // - If the size of the list (including a NUL terminating byte after every // entry) would exceed size, ERANGE may be returned. Note that diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go index 534528ce6..022bac127 100644 --- a/pkg/sentry/vfs/options.go +++ b/pkg/sentry/vfs/options.go @@ -33,6 +33,25 @@ type GetDentryOptions struct { type MkdirOptions struct { // Mode is the file mode bits for the created directory. Mode linux.FileMode + + // If ForSyntheticMountpoint is true, FilesystemImpl.MkdirAt() may create + // the given directory in memory only (as opposed to persistent storage). + // The created directory should be able to support the creation of + // subdirectories with ForSyntheticMountpoint == true. It does not need to + // support the creation of subdirectories with ForSyntheticMountpoint == + // false, or files of other types. + // + // FilesystemImpls are permitted to ignore the ForSyntheticMountpoint + // option. + // + // The ForSyntheticMountpoint option exists because, unlike mount(2), the + // OCI Runtime Specification permits the specification of mount points that + // do not exist, under the expectation that container runtimes will create + // them. (More accurately, the OCI Runtime Specification completely fails + // to document this feature, but it's implemented by runc.) + // ForSyntheticMountpoint allows such mount points to be created even when + // the underlying persistent filesystem is immutable. + ForSyntheticMountpoint bool } // MknodOptions contains options to VirtualFilesystem.MknodAt() and diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go index 82083c57d..bce3a3593 100644 --- a/runsc/boot/vfs.go +++ b/runsc/boot/vfs.go @@ -251,6 +251,12 @@ func (c *containerMounter) mountSubmountVFS2(ctx context.Context, conf *Config, // All writes go to upper, be paranoid and make lower readonly. opts.ReadOnly = useOverlay + if err := c.k.VFS().MkdirAt(ctx, creds, target, &vfs.MkdirOptions{ + ForSyntheticMountpoint: true, + }); err != nil && err != syserror.EEXIST { + // Log a warning, but attempt the mount anyway. + log.Warningf("Failed to create mount point at %q: %v", submount.Destination, err) + } if err := c.k.VFS().MountAt(ctx, creds, "", target, submount.Type, opts); err != nil { return fmt.Errorf("failed to mount %q (type: %s): %w, opts: %v", submount.Destination, submount.Type, err, opts) } -- cgit v1.2.3 From 93dd47146185ec7004f514e23bad9f225f55efb1 Mon Sep 17 00:00:00 2001 From: Rahat Mahmood Date: Thu, 23 Apr 2020 15:47:59 -0700 Subject: Enable automated marshalling for epoll events. Ensure we use the correct architecture-specific defintion of epoll event, and use go-marshal for serialization. PiperOrigin-RevId: 308145677 --- pkg/abi/linux/epoll_amd64.go | 4 ++- pkg/abi/linux/epoll_arm64.go | 4 ++- pkg/sentry/kernel/epoll/BUILD | 1 + pkg/sentry/kernel/epoll/epoll.go | 20 +++--------- pkg/sentry/syscalls/epoll.go | 3 +- pkg/sentry/syscalls/linux/sys_epoll.go | 27 ++-------------- pkg/sentry/syscalls/linux/vfs2/BUILD | 1 - pkg/sentry/syscalls/linux/vfs2/epoll.go | 7 ++-- pkg/sentry/syscalls/linux/vfs2/epoll_unsafe.go | 44 -------------------------- 9 files changed, 20 insertions(+), 91 deletions(-) delete mode 100644 pkg/sentry/syscalls/linux/vfs2/epoll_unsafe.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/epoll_amd64.go b/pkg/abi/linux/epoll_amd64.go index 34ff18009..7e74b1143 100644 --- a/pkg/abi/linux/epoll_amd64.go +++ b/pkg/abi/linux/epoll_amd64.go @@ -12,11 +12,13 @@ // See the License for the specific language governing permissions and // limitations under the License. +// +build amd64 + package linux // EpollEvent is equivalent to struct epoll_event from epoll(2). // -// +marshal +// +marshal slice:EpollEventSlice type EpollEvent struct { Events uint32 // Linux makes struct epoll_event::data a __u64. We represent it as diff --git a/pkg/abi/linux/epoll_arm64.go b/pkg/abi/linux/epoll_arm64.go index f86c35329..a35939cc9 100644 --- a/pkg/abi/linux/epoll_arm64.go +++ b/pkg/abi/linux/epoll_arm64.go @@ -12,11 +12,13 @@ // See the License for the specific language governing permissions and // limitations under the License. +// +build arm64 + package linux // EpollEvent is equivalent to struct epoll_event from epoll(2). // -// +marshal +// +marshal slice:EpollEventSlice type EpollEvent struct { Events uint32 // Linux makes struct epoll_event a __u64, necessitating 4 bytes of padding diff --git a/pkg/sentry/kernel/epoll/BUILD b/pkg/sentry/kernel/epoll/BUILD index dedf0fa15..75eedd5a2 100644 --- a/pkg/sentry/kernel/epoll/BUILD +++ b/pkg/sentry/kernel/epoll/BUILD @@ -24,6 +24,7 @@ go_library( ], visibility = ["//pkg/sentry:internal"], deps = [ + "//pkg/abi/linux", "//pkg/context", "//pkg/refs", "//pkg/sentry/fs", diff --git a/pkg/sentry/kernel/epoll/epoll.go b/pkg/sentry/kernel/epoll/epoll.go index 592650923..3d78cd48f 100644 --- a/pkg/sentry/kernel/epoll/epoll.go +++ b/pkg/sentry/kernel/epoll/epoll.go @@ -20,6 +20,7 @@ import ( "fmt" "syscall" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/fs" @@ -30,19 +31,6 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) -// Event describes the event mask that was observed and the user data to be -// returned when one of the events occurs. It has this format to match the linux -// format to avoid extra copying/allocation when writing events to userspace. -type Event struct { - // Events is the event mask containing the set of events that have been - // observed on an entry. - Events uint32 - - // Data is an opaque 64-bit value provided by the caller when adding the - // entry, and returned to the caller when the entry reports an event. - Data [2]int32 -} - // EntryFlags is a bitmask that holds an entry's flags. type EntryFlags int @@ -227,9 +215,9 @@ func (e *EventPoll) Readiness(mask waiter.EventMask) waiter.EventMask { } // ReadEvents returns up to max available events. -func (e *EventPoll) ReadEvents(max int) []Event { +func (e *EventPoll) ReadEvents(max int) []linux.EpollEvent { var local pollEntryList - var ret []Event + var ret []linux.EpollEvent e.listsMu.Lock() @@ -251,7 +239,7 @@ func (e *EventPoll) ReadEvents(max int) []Event { } // Add event to the array that will be returned to caller. - ret = append(ret, Event{ + ret = append(ret, linux.EpollEvent{ Events: uint32(ready), Data: entry.userData, }) diff --git a/pkg/sentry/syscalls/epoll.go b/pkg/sentry/syscalls/epoll.go index 87dcad18b..d9fb808c0 100644 --- a/pkg/sentry/syscalls/epoll.go +++ b/pkg/sentry/syscalls/epoll.go @@ -17,6 +17,7 @@ package syscalls import ( "time" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/epoll" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" @@ -118,7 +119,7 @@ func RemoveEpoll(t *kernel.Task, epfd int32, fd int32) error { } // WaitEpoll implements the epoll_wait(2) linux syscall. -func WaitEpoll(t *kernel.Task, fd int32, max int, timeout int) ([]epoll.Event, error) { +func WaitEpoll(t *kernel.Task, fd int32, max int, timeout int) ([]linux.EpollEvent, error) { // Get epoll from the file descriptor. epollfile := t.GetFile(fd) if epollfile == nil { diff --git a/pkg/sentry/syscalls/linux/sys_epoll.go b/pkg/sentry/syscalls/linux/sys_epoll.go index 3ab93fbde..51bf205cf 100644 --- a/pkg/sentry/syscalls/linux/sys_epoll.go +++ b/pkg/sentry/syscalls/linux/sys_epoll.go @@ -21,7 +21,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/epoll" "gvisor.dev/gvisor/pkg/sentry/syscalls" "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) @@ -72,7 +71,7 @@ func EpollCtl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc var data [2]int32 if op != linux.EPOLL_CTL_DEL { var e linux.EpollEvent - if _, err := t.CopyIn(eventAddr, &e); err != nil { + if _, err := e.CopyIn(t, eventAddr); err != nil { return 0, nil, err } @@ -105,28 +104,6 @@ func EpollCtl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc } } -// copyOutEvents copies epoll events from the kernel to user memory. -func copyOutEvents(t *kernel.Task, addr usermem.Addr, e []epoll.Event) error { - const itemLen = 12 - buffLen := len(e) * itemLen - if _, ok := addr.AddLength(uint64(buffLen)); !ok { - return syserror.EFAULT - } - - b := t.CopyScratchBuffer(buffLen) - for i := range e { - usermem.ByteOrder.PutUint32(b[i*itemLen:], e[i].Events) - usermem.ByteOrder.PutUint32(b[i*itemLen+4:], uint32(e[i].Data[0])) - usermem.ByteOrder.PutUint32(b[i*itemLen+8:], uint32(e[i].Data[1])) - } - - if _, err := t.CopyOutBytes(addr, b); err != nil { - return err - } - - return nil -} - // EpollWait implements the epoll_wait(2) linux syscall. func EpollWait(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { epfd := args[0].Int() @@ -140,7 +117,7 @@ func EpollWait(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys } if len(r) != 0 { - if err := copyOutEvents(t, eventsAddr, r); err != nil { + if _, err := linux.CopyEpollEventSliceOut(t, eventsAddr, r); err != nil { return 0, nil, err } } diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD index 6ff2d84d2..f6fb0f219 100644 --- a/pkg/sentry/syscalls/linux/vfs2/BUILD +++ b/pkg/sentry/syscalls/linux/vfs2/BUILD @@ -6,7 +6,6 @@ go_library( name = "vfs2", srcs = [ "epoll.go", - "epoll_unsafe.go", "execve.go", "fd.go", "filesystem.go", diff --git a/pkg/sentry/syscalls/linux/vfs2/epoll.go b/pkg/sentry/syscalls/linux/vfs2/epoll.go index 5a938cee2..34c90ae3e 100644 --- a/pkg/sentry/syscalls/linux/vfs2/epoll.go +++ b/pkg/sentry/syscalls/linux/vfs2/epoll.go @@ -28,6 +28,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +var sizeofEpollEvent = (*linux.EpollEvent)(nil).SizeBytes() + // EpollCreate1 implements Linux syscall epoll_create1(2). func EpollCreate1(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { flags := args[0].Int() @@ -124,7 +126,7 @@ func EpollWait(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys maxEvents := int(args[2].Int()) timeout := int(args[3].Int()) - const _EP_MAX_EVENTS = math.MaxInt32 / sizeofEpollEvent // Linux: fs/eventpoll.c:EP_MAX_EVENTS + var _EP_MAX_EVENTS = math.MaxInt32 / sizeofEpollEvent // Linux: fs/eventpoll.c:EP_MAX_EVENTS if maxEvents <= 0 || maxEvents > _EP_MAX_EVENTS { return 0, nil, syserror.EINVAL } @@ -157,7 +159,8 @@ func EpollWait(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys maxEvents -= n if n != 0 { // Copy what we read out. - copiedEvents, err := copyOutEvents(t, eventsAddr, events[:n]) + copiedBytes, err := linux.CopyEpollEventSliceOut(t, eventsAddr, events[:n]) + copiedEvents := copiedBytes / sizeofEpollEvent // rounded down eventsAddr += usermem.Addr(copiedEvents * sizeofEpollEvent) total += copiedEvents if err != nil { diff --git a/pkg/sentry/syscalls/linux/vfs2/epoll_unsafe.go b/pkg/sentry/syscalls/linux/vfs2/epoll_unsafe.go deleted file mode 100644 index 825f325bf..000000000 --- a/pkg/sentry/syscalls/linux/vfs2/epoll_unsafe.go +++ /dev/null @@ -1,44 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package vfs2 - -import ( - "reflect" - "runtime" - "unsafe" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/gohacks" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/usermem" -) - -const sizeofEpollEvent = int(unsafe.Sizeof(linux.EpollEvent{})) - -func copyOutEvents(t *kernel.Task, addr usermem.Addr, events []linux.EpollEvent) (int, error) { - if len(events) == 0 { - return 0, nil - } - // Cast events to a byte slice for copying. - var eventBytes []byte - eventBytesHdr := (*reflect.SliceHeader)(unsafe.Pointer(&eventBytes)) - eventBytesHdr.Data = uintptr(gohacks.Noescape(unsafe.Pointer(&events[0]))) - eventBytesHdr.Len = len(events) * sizeofEpollEvent - eventBytesHdr.Cap = len(events) * sizeofEpollEvent - copiedBytes, err := t.CopyOutBytes(addr, eventBytes) - runtime.KeepAlive(events) - copiedEvents := copiedBytes / sizeofEpollEvent // rounded down - return copiedEvents, err -} -- cgit v1.2.3 From eccae0f77d3708d591119488f427eca90de7c711 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Thu, 23 Apr 2020 17:27:24 -0700 Subject: Remove View.First() and View.RemoveFirst() These methods let users eaily break the VectorisedView abstraction, and allowed netstack to slip into pseudo-enforcement of the "all headers are in the first View" invariant. Removing them and replacing with PullUp(n) breaks this reliance and will make it easier to add iptables support and rework network buffer management. The new View.PullUp(n) method is low cost in the common case, when when all the headers fit in the first View. PiperOrigin-RevId: 308163542 --- pkg/sentry/socket/netfilter/tcp_matcher.go | 5 +- pkg/sentry/socket/netfilter/udp_matcher.go | 5 +- pkg/tcpip/buffer/view.go | 55 ++++++++++---- pkg/tcpip/buffer/view_test.go | 113 +++++++++++++++++++++++++++++ pkg/tcpip/link/loopback/loopback.go | 10 +-- pkg/tcpip/link/rawfile/BUILD | 9 ++- pkg/tcpip/link/rawfile/rawfile_test.go | 46 ++++++++++++ pkg/tcpip/link/rawfile/rawfile_unsafe.go | 6 +- pkg/tcpip/link/sharedmem/sharedmem_test.go | 2 +- pkg/tcpip/link/sniffer/sniffer.go | 65 +++++++++++++---- pkg/tcpip/network/arp/arp.go | 5 +- pkg/tcpip/network/ipv4/icmp.go | 20 +++-- pkg/tcpip/network/ipv4/ipv4.go | 12 ++- pkg/tcpip/network/ipv6/icmp.go | 74 ++++++++++++------- pkg/tcpip/network/ipv6/icmp_test.go | 3 +- pkg/tcpip/network/ipv6/ipv6.go | 6 +- pkg/tcpip/stack/forwarder_test.go | 13 ++-- pkg/tcpip/stack/iptables.go | 22 +++++- pkg/tcpip/stack/iptables_targets.go | 23 ++++-- pkg/tcpip/stack/nic.go | 34 +++------ pkg/tcpip/stack/packet_buffer.go | 8 +- pkg/tcpip/stack/stack_test.go | 10 ++- pkg/tcpip/stack/transport_test.go | 5 +- pkg/tcpip/transport/icmp/endpoint.go | 8 +- pkg/tcpip/transport/tcp/segment.go | 29 +++++--- pkg/tcpip/transport/tcp/tcp_test.go | 4 +- pkg/tcpip/transport/udp/endpoint.go | 6 +- pkg/tcpip/transport/udp/protocol.go | 9 ++- 28 files changed, 458 insertions(+), 149 deletions(-) create mode 100644 pkg/tcpip/link/rawfile/rawfile_test.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/tcp_matcher.go b/pkg/sentry/socket/netfilter/tcp_matcher.go index ff1cfd8f6..55c0f04f3 100644 --- a/pkg/sentry/socket/netfilter/tcp_matcher.go +++ b/pkg/sentry/socket/netfilter/tcp_matcher.go @@ -121,12 +121,13 @@ func (tm *TCPMatcher) Match(hook stack.Hook, pkt stack.PacketBuffer, interfaceNa tcpHeader = header.TCP(pkt.TransportHeader) } else { // The TCP header hasn't been parsed yet. We have to do it here. - if len(pkt.Data.First()) < header.TCPMinimumSize { + hdr, ok := pkt.Data.PullUp(header.TCPMinimumSize) + if !ok { // There's no valid TCP header here, so we hotdrop the // packet. return false, true } - tcpHeader = header.TCP(pkt.Data.First()) + tcpHeader = header.TCP(hdr) } // Check whether the source and destination ports are within the diff --git a/pkg/sentry/socket/netfilter/udp_matcher.go b/pkg/sentry/socket/netfilter/udp_matcher.go index 3359418c1..04d03d494 100644 --- a/pkg/sentry/socket/netfilter/udp_matcher.go +++ b/pkg/sentry/socket/netfilter/udp_matcher.go @@ -120,12 +120,13 @@ func (um *UDPMatcher) Match(hook stack.Hook, pkt stack.PacketBuffer, interfaceNa udpHeader = header.UDP(pkt.TransportHeader) } else { // The UDP header hasn't been parsed yet. We have to do it here. - if len(pkt.Data.First()) < header.UDPMinimumSize { + hdr, ok := pkt.Data.PullUp(header.UDPMinimumSize) + if !ok { // There's no valid UDP header here, so we hotdrop the // packet. return false, true } - udpHeader = header.UDP(pkt.Data.First()) + udpHeader = header.UDP(hdr) } // Check whether the source and destination ports are within the diff --git a/pkg/tcpip/buffer/view.go b/pkg/tcpip/buffer/view.go index 8ec5d5d5c..f01217c91 100644 --- a/pkg/tcpip/buffer/view.go +++ b/pkg/tcpip/buffer/view.go @@ -77,7 +77,8 @@ func NewVectorisedView(size int, views []View) VectorisedView { return VectorisedView{views: views, size: size} } -// TrimFront removes the first "count" bytes of the vectorised view. +// TrimFront removes the first "count" bytes of the vectorised view. It panics +// if count > vv.Size(). func (vv *VectorisedView) TrimFront(count int) { for count > 0 && len(vv.views) > 0 { if count < len(vv.views[0]) { @@ -86,7 +87,7 @@ func (vv *VectorisedView) TrimFront(count int) { return } count -= len(vv.views[0]) - vv.RemoveFirst() + vv.removeFirst() } } @@ -104,7 +105,7 @@ func (vv *VectorisedView) Read(v View) (copied int, err error) { count -= len(vv.views[0]) copy(v[copied:], vv.views[0]) copied += len(vv.views[0]) - vv.RemoveFirst() + vv.removeFirst() } if copied == 0 { return 0, io.EOF @@ -126,7 +127,7 @@ func (vv *VectorisedView) ReadToVV(dstVV *VectorisedView, count int) (copied int count -= len(vv.views[0]) dstVV.AppendView(vv.views[0]) copied += len(vv.views[0]) - vv.RemoveFirst() + vv.removeFirst() } return copied } @@ -162,22 +163,37 @@ func (vv *VectorisedView) Clone(buffer []View) VectorisedView { return VectorisedView{views: append(buffer[:0], vv.views...), size: vv.size} } -// First returns the first view of the vectorised view. -func (vv *VectorisedView) First() View { +// PullUp returns the first "count" bytes of the vectorised view. If those +// bytes aren't already contiguous inside the vectorised view, PullUp will +// reallocate as needed to make them contiguous. PullUp fails and returns false +// when count > vv.Size(). +func (vv *VectorisedView) PullUp(count int) (View, bool) { if len(vv.views) == 0 { - return nil + return nil, count == 0 + } + if count <= len(vv.views[0]) { + return vv.views[0][:count], true + } + if count > vv.size { + return nil, false } - return vv.views[0] -} -// RemoveFirst removes the first view of the vectorised view. -func (vv *VectorisedView) RemoveFirst() { - if len(vv.views) == 0 { - return + newFirst := NewView(count) + i := 0 + for offset := 0; offset < count; i++ { + copy(newFirst[offset:], vv.views[i]) + if count-offset < len(vv.views[i]) { + vv.views[i].TrimFront(count - offset) + break + } + offset += len(vv.views[i]) + vv.views[i] = nil } - vv.size -= len(vv.views[0]) - vv.views[0] = nil - vv.views = vv.views[1:] + // We're guaranteed that i > 0, since count is too large for the first + // view. + vv.views[i-1] = newFirst + vv.views = vv.views[i-1:] + return newFirst, true } // Size returns the size in bytes of the entire content stored in the vectorised view. @@ -225,3 +241,10 @@ func (vv *VectorisedView) Readers() []bytes.Reader { } return readers } + +// removeFirst panics when len(vv.views) < 1. +func (vv *VectorisedView) removeFirst() { + vv.size -= len(vv.views[0]) + vv.views[0] = nil + vv.views = vv.views[1:] +} diff --git a/pkg/tcpip/buffer/view_test.go b/pkg/tcpip/buffer/view_test.go index 106e1994c..c56795c7b 100644 --- a/pkg/tcpip/buffer/view_test.go +++ b/pkg/tcpip/buffer/view_test.go @@ -16,6 +16,7 @@ package buffer import ( + "bytes" "reflect" "testing" ) @@ -370,3 +371,115 @@ func TestVVRead(t *testing.T) { }) } } + +var pullUpTestCases = []struct { + comment string + in VectorisedView + count int + want []byte + result VectorisedView + ok bool +}{ + { + comment: "simple case", + in: vv(2, "12"), + count: 1, + want: []byte("1"), + result: vv(2, "12"), + ok: true, + }, + { + comment: "entire View", + in: vv(2, "1", "2"), + count: 1, + want: []byte("1"), + result: vv(2, "1", "2"), + ok: true, + }, + { + comment: "spanning across two Views", + in: vv(3, "1", "23"), + count: 2, + want: []byte("12"), + result: vv(3, "12", "3"), + ok: true, + }, + { + comment: "spanning across all Views", + in: vv(5, "1", "23", "45"), + count: 5, + want: []byte("12345"), + result: vv(5, "12345"), + ok: true, + }, + { + comment: "count = 0", + in: vv(1, "1"), + count: 0, + want: []byte{}, + result: vv(1, "1"), + ok: true, + }, + { + comment: "count = size", + in: vv(1, "1"), + count: 1, + want: []byte("1"), + result: vv(1, "1"), + ok: true, + }, + { + comment: "count too large", + in: vv(3, "1", "23"), + count: 4, + want: nil, + result: vv(3, "1", "23"), + ok: false, + }, + { + comment: "empty vv", + in: vv(0, ""), + count: 1, + want: nil, + result: vv(0, ""), + ok: false, + }, + { + comment: "empty vv, count = 0", + in: vv(0, ""), + count: 0, + want: nil, + result: vv(0, ""), + ok: true, + }, + { + comment: "empty views", + in: vv(3, "", "1", "", "23"), + count: 2, + want: []byte("12"), + result: vv(3, "12", "3"), + ok: true, + }, +} + +func TestPullUp(t *testing.T) { + for _, c := range pullUpTestCases { + got, ok := c.in.PullUp(c.count) + + // Is the return value right? + if ok != c.ok { + t.Errorf("Test %q failed when calling PullUp(%d) on %v. Got an ok of %t. Want %t", + c.comment, c.count, c.in, ok, c.ok) + } + if bytes.Compare(got, View(c.want)) != 0 { + t.Errorf("Test %q failed when calling PullUp(%d) on %v. Got %v. Want %v", + c.comment, c.count, c.in, got, c.want) + } + + // Is the underlying structure right? + if !reflect.DeepEqual(c.in, c.result) { + t.Errorf("Test %q failed when calling PullUp(%d). Got vv with structure %v. Wanted %v", + c.comment, c.count, c.in, c.result) + } + } +} diff --git a/pkg/tcpip/link/loopback/loopback.go b/pkg/tcpip/link/loopback/loopback.go index 1e2255bfa..073c84ef9 100644 --- a/pkg/tcpip/link/loopback/loopback.go +++ b/pkg/tcpip/link/loopback/loopback.go @@ -98,13 +98,13 @@ func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, stack.PacketBufferList // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { - // Reject the packet if it's shorter than an ethernet header. - if vv.Size() < header.EthernetMinimumSize { + // There should be an ethernet header at the beginning of vv. + hdr, ok := vv.PullUp(header.EthernetMinimumSize) + if !ok { + // Reject the packet if it's shorter than an ethernet header. return tcpip.ErrBadAddress } - - // There should be an ethernet header at the beginning of vv. - linkHeader := header.Ethernet(vv.First()[:header.EthernetMinimumSize]) + linkHeader := header.Ethernet(hdr) vv.TrimFront(len(linkHeader)) e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, linkHeader.Type(), stack.PacketBuffer{ Data: vv, diff --git a/pkg/tcpip/link/rawfile/BUILD b/pkg/tcpip/link/rawfile/BUILD index 14b527bc2..9cc08d0e2 100644 --- a/pkg/tcpip/link/rawfile/BUILD +++ b/pkg/tcpip/link/rawfile/BUILD @@ -1,4 +1,4 @@ -load("//tools:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -18,3 +18,10 @@ go_library( "@org_golang_x_sys//unix:go_default_library", ], ) + +go_test( + name = "rawfile_test", + size = "small", + srcs = ["rawfile_test.go"], + library = ":rawfile", +) diff --git a/pkg/tcpip/link/rawfile/rawfile_test.go b/pkg/tcpip/link/rawfile/rawfile_test.go new file mode 100644 index 000000000..8f14ba761 --- /dev/null +++ b/pkg/tcpip/link/rawfile/rawfile_test.go @@ -0,0 +1,46 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build linux + +package rawfile + +import ( + "syscall" + "testing" +) + +func TestNonBlockingWrite3ZeroLength(t *testing.T) { + fd, err := syscall.Open("/dev/null", syscall.O_WRONLY, 0) + if err != nil { + t.Fatalf("failed to open /dev/null: %v", err) + } + defer syscall.Close(fd) + + if err := NonBlockingWrite3(fd, []byte{}, []byte{0}, nil); err != nil { + t.Fatalf("failed to write: %v", err) + } +} + +func TestNonBlockingWrite3Nil(t *testing.T) { + fd, err := syscall.Open("/dev/null", syscall.O_WRONLY, 0) + if err != nil { + t.Fatalf("failed to open /dev/null: %v", err) + } + defer syscall.Close(fd) + + if err := NonBlockingWrite3(fd, nil, []byte{0}, nil); err != nil { + t.Fatalf("failed to write: %v", err) + } +} diff --git a/pkg/tcpip/link/rawfile/rawfile_unsafe.go b/pkg/tcpip/link/rawfile/rawfile_unsafe.go index 44e25d475..92efd0bf8 100644 --- a/pkg/tcpip/link/rawfile/rawfile_unsafe.go +++ b/pkg/tcpip/link/rawfile/rawfile_unsafe.go @@ -76,9 +76,13 @@ func NonBlockingWrite3(fd int, b1, b2, b3 []byte) *tcpip.Error { // We have two buffers. Build the iovec that represents them and issue // a writev syscall. + var base *byte + if len(b1) > 0 { + base = &b1[0] + } iovec := [3]syscall.Iovec{ { - Base: &b1[0], + Base: base, Len: uint64(len(b1)), }, { diff --git a/pkg/tcpip/link/sharedmem/sharedmem_test.go b/pkg/tcpip/link/sharedmem/sharedmem_test.go index 27ea3f531..33f640b85 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem_test.go +++ b/pkg/tcpip/link/sharedmem/sharedmem_test.go @@ -674,7 +674,7 @@ func TestSimpleReceive(t *testing.T) { // Wait for packet to be received, then check it. c.waitForPackets(1, time.After(5*time.Second), "Timeout waiting for packet") c.mu.Lock() - rcvd := []byte(c.packets[0].vv.First()) + rcvd := []byte(c.packets[0].vv.ToView()) c.packets = c.packets[:0] c.mu.Unlock() diff --git a/pkg/tcpip/link/sniffer/sniffer.go b/pkg/tcpip/link/sniffer/sniffer.go index be2537a82..0799c8f4d 100644 --- a/pkg/tcpip/link/sniffer/sniffer.go +++ b/pkg/tcpip/link/sniffer/sniffer.go @@ -171,11 +171,7 @@ func (e *endpoint) GSOMaxSize() uint32 { func (e *endpoint) dumpPacket(prefix string, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { writer := e.writer if writer == nil && atomic.LoadUint32(&LogPackets) == 1 { - first := pkt.Header.View() - if len(first) == 0 { - first = pkt.Data.First() - } - logPacket(prefix, protocol, first, gso) + logPacket(prefix, protocol, pkt, gso) } if writer != nil && atomic.LoadUint32(&LogPacketsToPCAP) == 1 { totalLength := pkt.Header.UsedLength() + pkt.Data.Size() @@ -238,7 +234,7 @@ func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { // Wait implements stack.LinkEndpoint.Wait. func (e *endpoint) Wait() { e.lower.Wait() } -func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.View, gso *stack.GSO) { +func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer, gso *stack.GSO) { // Figure out the network layer info. var transProto uint8 src := tcpip.Address("unknown") @@ -247,28 +243,49 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie size := uint16(0) var fragmentOffset uint16 var moreFragments bool + + // Create a clone of pkt, including any headers if present. Avoid allocating + // backing memory for the clone. + views := [8]buffer.View{} + vv := buffer.NewVectorisedView(0, views[:0]) + vv.AppendView(pkt.Header.View()) + vv.Append(pkt.Data) + switch protocol { case header.IPv4ProtocolNumber: - ipv4 := header.IPv4(b) + hdr, ok := vv.PullUp(header.IPv4MinimumSize) + if !ok { + return + } + ipv4 := header.IPv4(hdr) fragmentOffset = ipv4.FragmentOffset() moreFragments = ipv4.Flags()&header.IPv4FlagMoreFragments == header.IPv4FlagMoreFragments src = ipv4.SourceAddress() dst = ipv4.DestinationAddress() transProto = ipv4.Protocol() size = ipv4.TotalLength() - uint16(ipv4.HeaderLength()) - b = b[ipv4.HeaderLength():] + vv.TrimFront(int(ipv4.HeaderLength())) id = int(ipv4.ID()) case header.IPv6ProtocolNumber: - ipv6 := header.IPv6(b) + hdr, ok := vv.PullUp(header.IPv6MinimumSize) + if !ok { + return + } + ipv6 := header.IPv6(hdr) src = ipv6.SourceAddress() dst = ipv6.DestinationAddress() transProto = ipv6.NextHeader() size = ipv6.PayloadLength() - b = b[header.IPv6MinimumSize:] + vv.TrimFront(header.IPv6MinimumSize) case header.ARPProtocolNumber: - arp := header.ARP(b) + hdr, ok := vv.PullUp(header.ARPSize) + if !ok { + return + } + vv.TrimFront(header.ARPSize) + arp := header.ARP(hdr) log.Infof( "%s arp %v (%v) -> %v (%v) valid:%v", prefix, @@ -284,7 +301,7 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie // We aren't guaranteed to have a transport header - it's possible for // writes via raw endpoints to contain only network headers. - if minSize, ok := transportProtocolMinSizes[tcpip.TransportProtocolNumber(transProto)]; ok && len(b) < minSize { + if minSize, ok := transportProtocolMinSizes[tcpip.TransportProtocolNumber(transProto)]; ok && vv.Size() < minSize { log.Infof("%s %v -> %v transport protocol: %d, but no transport header found (possible raw packet)", prefix, src, dst, transProto) return } @@ -297,7 +314,11 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie switch tcpip.TransportProtocolNumber(transProto) { case header.ICMPv4ProtocolNumber: transName = "icmp" - icmp := header.ICMPv4(b) + hdr, ok := vv.PullUp(header.ICMPv4MinimumSize) + if !ok { + break + } + icmp := header.ICMPv4(hdr) icmpType := "unknown" if fragmentOffset == 0 { switch icmp.Type() { @@ -330,7 +351,11 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie case header.ICMPv6ProtocolNumber: transName = "icmp" - icmp := header.ICMPv6(b) + hdr, ok := vv.PullUp(header.ICMPv6MinimumSize) + if !ok { + break + } + icmp := header.ICMPv6(hdr) icmpType := "unknown" switch icmp.Type() { case header.ICMPv6DstUnreachable: @@ -361,7 +386,11 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie case header.UDPProtocolNumber: transName = "udp" - udp := header.UDP(b) + hdr, ok := vv.PullUp(header.UDPMinimumSize) + if !ok { + break + } + udp := header.UDP(hdr) if fragmentOffset == 0 && len(udp) >= header.UDPMinimumSize { srcPort = udp.SourcePort() dstPort = udp.DestinationPort() @@ -371,7 +400,11 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie case header.TCPProtocolNumber: transName = "tcp" - tcp := header.TCP(b) + hdr, ok := vv.PullUp(header.TCPMinimumSize) + if !ok { + break + } + tcp := header.TCP(hdr) if fragmentOffset == 0 && len(tcp) >= header.TCPMinimumSize { offset := int(tcp.DataOffset()) if offset < header.TCPMinimumSize { diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index 7acbfa0a8..cf73a939e 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -93,7 +93,10 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuf } func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { - v := pkt.Data.First() + v, ok := pkt.Data.PullUp(header.ARPSize) + if !ok { + return + } h := header.ARP(v) if !h.IsValid() { return diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go index c4bf1ba5c..4cbefe5ab 100644 --- a/pkg/tcpip/network/ipv4/icmp.go +++ b/pkg/tcpip/network/ipv4/icmp.go @@ -25,7 +25,11 @@ import ( // used to find out which transport endpoint must be notified about the ICMP // packet. func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack.PacketBuffer) { - h := header.IPv4(pkt.Data.First()) + h, ok := pkt.Data.PullUp(header.IPv4MinimumSize) + if !ok { + return + } + hdr := header.IPv4(h) // We don't use IsValid() here because ICMP only requires that the IP // header plus 8 bytes of the transport header be included. So it's @@ -34,12 +38,12 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. // // Drop packet if it doesn't have the basic IPv4 header or if the // original source address doesn't match the endpoint's address. - if len(h) < header.IPv4MinimumSize || h.SourceAddress() != e.id.LocalAddress { + if hdr.SourceAddress() != e.id.LocalAddress { return } - hlen := int(h.HeaderLength()) - if pkt.Data.Size() < hlen || h.FragmentOffset() != 0 { + hlen := int(hdr.HeaderLength()) + if pkt.Data.Size() < hlen || hdr.FragmentOffset() != 0 { // We won't be able to handle this if it doesn't contain the // full IPv4 header, or if it's a fragment not at offset 0 // (because it won't have the transport header). @@ -48,15 +52,15 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. // Skip the ip header, then deliver control message. pkt.Data.TrimFront(hlen) - p := h.TransportProtocol() - e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, h.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) + p := hdr.TransportProtocol() + e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, hdr.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) } func (e *endpoint) handleICMP(r *stack.Route, pkt stack.PacketBuffer) { stats := r.Stats() received := stats.ICMP.V4PacketsReceived - v := pkt.Data.First() - if len(v) < header.ICMPv4MinimumSize { + v, ok := pkt.Data.PullUp(header.ICMPv4MinimumSize) + if !ok { received.Invalid.Increment() return } diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index 104aafbed..17202cc7a 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -328,7 +328,11 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuffer) *tcpip.Error { // The packet already has an IP header, but there are a few required // checks. - ip := header.IPv4(pkt.Data.First()) + h, ok := pkt.Data.PullUp(header.IPv4MinimumSize) + if !ok { + return tcpip.ErrInvalidOptionValue + } + ip := header.IPv4(h) if !ip.IsValid(pkt.Data.Size()) { return tcpip.ErrInvalidOptionValue } @@ -378,7 +382,11 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuf // HandlePacket is called by the link layer when new ipv4 packets arrive for // this endpoint. func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { - headerView := pkt.Data.First() + headerView, ok := pkt.Data.PullUp(header.IPv4MinimumSize) + if !ok { + r.Stats().IP.MalformedPacketsReceived.Increment() + return + } h := header.IPv4(headerView) if !h.IsValid(pkt.Data.Size()) { r.Stats().IP.MalformedPacketsReceived.Increment() diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index b68983d10..bdf3a0d25 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -28,7 +28,11 @@ import ( // used to find out which transport endpoint must be notified about the ICMP // packet. func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack.PacketBuffer) { - h := header.IPv6(pkt.Data.First()) + h, ok := pkt.Data.PullUp(header.IPv6MinimumSize) + if !ok { + return + } + hdr := header.IPv6(h) // We don't use IsValid() here because ICMP only requires that up to // 1280 bytes of the original packet be included. So it's likely that it @@ -36,17 +40,21 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. // // Drop packet if it doesn't have the basic IPv6 header or if the // original source address doesn't match the endpoint's address. - if len(h) < header.IPv6MinimumSize || h.SourceAddress() != e.id.LocalAddress { + if hdr.SourceAddress() != e.id.LocalAddress { return } // Skip the IP header, then handle the fragmentation header if there // is one. pkt.Data.TrimFront(header.IPv6MinimumSize) - p := h.TransportProtocol() + p := hdr.TransportProtocol() if p == header.IPv6FragmentHeader { - f := header.IPv6Fragment(pkt.Data.First()) - if !f.IsValid() || f.FragmentOffset() != 0 { + f, ok := pkt.Data.PullUp(header.IPv6FragmentHeaderSize) + if !ok { + return + } + fragHdr := header.IPv6Fragment(f) + if !fragHdr.IsValid() || fragHdr.FragmentOffset() != 0 { // We can't handle fragments that aren't at offset 0 // because they don't have the transport headers. return @@ -55,19 +63,19 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. // Skip fragmentation header and find out the actual protocol // number. pkt.Data.TrimFront(header.IPv6FragmentHeaderSize) - p = f.TransportProtocol() + p = fragHdr.TransportProtocol() } // Deliver the control packet to the transport endpoint. - e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, h.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) + e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, hdr.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) } func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.PacketBuffer, hasFragmentHeader bool) { stats := r.Stats().ICMP sent := stats.V6PacketsSent received := stats.V6PacketsReceived - v := pkt.Data.First() - if len(v) < header.ICMPv6MinimumSize { + v, ok := pkt.Data.PullUp(header.ICMPv6HeaderSize) + if !ok { received.Invalid.Increment() return } @@ -76,11 +84,9 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P // Validate ICMPv6 checksum before processing the packet. // - // Only the first view in vv is accounted for by h. To account for the - // rest of vv, a shallow copy is made and the first view is removed. // This copy is used as extra payload during the checksum calculation. payload := pkt.Data.Clone(nil) - payload.RemoveFirst() + payload.TrimFront(len(h)) if got, want := h.Checksum(), header.ICMPv6Checksum(h, iph.SourceAddress(), iph.DestinationAddress(), payload); got != want { received.Invalid.Increment() return @@ -101,34 +107,40 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P switch h.Type() { case header.ICMPv6PacketTooBig: received.PacketTooBig.Increment() - if len(v) < header.ICMPv6PacketTooBigMinimumSize { + hdr, ok := pkt.Data.PullUp(header.ICMPv6PacketTooBigMinimumSize) + if !ok { received.Invalid.Increment() return } pkt.Data.TrimFront(header.ICMPv6PacketTooBigMinimumSize) - mtu := h.MTU() + mtu := header.ICMPv6(hdr).MTU() e.handleControl(stack.ControlPacketTooBig, calculateMTU(mtu), pkt) case header.ICMPv6DstUnreachable: received.DstUnreachable.Increment() - if len(v) < header.ICMPv6DstUnreachableMinimumSize { + hdr, ok := pkt.Data.PullUp(header.ICMPv6DstUnreachableMinimumSize) + if !ok { received.Invalid.Increment() return } pkt.Data.TrimFront(header.ICMPv6DstUnreachableMinimumSize) - switch h.Code() { + switch header.ICMPv6(hdr).Code() { case header.ICMPv6PortUnreachable: e.handleControl(stack.ControlPortUnreachable, 0, pkt) } case header.ICMPv6NeighborSolicit: received.NeighborSolicit.Increment() - if len(v) < header.ICMPv6NeighborSolicitMinimumSize || !isNDPValid() { + if pkt.Data.Size() < header.ICMPv6NeighborSolicitMinimumSize || !isNDPValid() { received.Invalid.Increment() return } - ns := header.NDPNeighborSolicit(h.NDPPayload()) + // The remainder of payload must be only the neighbor solicitation, so + // payload.ToView() always returns the solicitation. Per RFC 6980 section 5, + // NDP messages cannot be fragmented. Also note that in the common case NDP + // datagrams are very small and ToView() will not incur allocations. + ns := header.NDPNeighborSolicit(payload.ToView()) it, err := ns.Options().Iter(true) if err != nil { // If we have a malformed NDP NS option, drop the packet. @@ -286,12 +298,16 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P case header.ICMPv6NeighborAdvert: received.NeighborAdvert.Increment() - if len(v) < header.ICMPv6NeighborAdvertSize || !isNDPValid() { + if pkt.Data.Size() < header.ICMPv6NeighborAdvertSize || !isNDPValid() { received.Invalid.Increment() return } - na := header.NDPNeighborAdvert(h.NDPPayload()) + // The remainder of payload must be only the neighbor advertisement, so + // payload.ToView() always returns the advertisement. Per RFC 6980 section + // 5, NDP messages cannot be fragmented. Also note that in the common case + // NDP datagrams are very small and ToView() will not incur allocations. + na := header.NDPNeighborAdvert(payload.ToView()) it, err := na.Options().Iter(true) if err != nil { // If we have a malformed NDP NA option, drop the packet. @@ -363,14 +379,15 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P case header.ICMPv6EchoRequest: received.EchoRequest.Increment() - if len(v) < header.ICMPv6EchoMinimumSize { + icmpHdr, ok := pkt.Data.PullUp(header.ICMPv6EchoMinimumSize) + if !ok { received.Invalid.Increment() return } pkt.Data.TrimFront(header.ICMPv6EchoMinimumSize) hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv6EchoMinimumSize) packet := header.ICMPv6(hdr.Prepend(header.ICMPv6EchoMinimumSize)) - copy(packet, h) + copy(packet, icmpHdr) packet.SetType(header.ICMPv6EchoReply) packet.SetChecksum(header.ICMPv6Checksum(packet, r.LocalAddress, r.RemoteAddress, pkt.Data)) if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, stack.PacketBuffer{ @@ -384,7 +401,7 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P case header.ICMPv6EchoReply: received.EchoReply.Increment() - if len(v) < header.ICMPv6EchoMinimumSize { + if pkt.Data.Size() < header.ICMPv6EchoMinimumSize { received.Invalid.Increment() return } @@ -406,8 +423,9 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P case header.ICMPv6RouterAdvert: received.RouterAdvert.Increment() - p := h.NDPPayload() - if len(p) < header.NDPRAMinimumSize || !isNDPValid() { + // Is the NDP payload of sufficient size to hold a Router + // Advertisement? + if pkt.Data.Size()-header.ICMPv6HeaderSize < header.NDPRAMinimumSize || !isNDPValid() { received.Invalid.Increment() return } @@ -425,7 +443,11 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P return } - ra := header.NDPRouterAdvert(p) + // The remainder of payload must be only the router advertisement, so + // payload.ToView() always returns the advertisement. Per RFC 6980 section + // 5, NDP messages cannot be fragmented. Also note that in the common case + // NDP datagrams are very small and ToView() will not incur allocations. + ra := header.NDPRouterAdvert(payload.ToView()) opts := ra.Options() // Are options valid as per the wire format? diff --git a/pkg/tcpip/network/ipv6/icmp_test.go b/pkg/tcpip/network/ipv6/icmp_test.go index bd099a7f8..d412ff688 100644 --- a/pkg/tcpip/network/ipv6/icmp_test.go +++ b/pkg/tcpip/network/ipv6/icmp_test.go @@ -166,7 +166,8 @@ func TestICMPCounts(t *testing.T) { }, { typ: header.ICMPv6NeighborSolicit, - size: header.ICMPv6NeighborSolicitMinimumSize}, + size: header.ICMPv6NeighborSolicitMinimumSize, + }, { typ: header.ICMPv6NeighborAdvert, size: header.ICMPv6NeighborAdvertMinimumSize, diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index 331b0817b..486725131 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -171,7 +171,11 @@ func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuffe // HandlePacket is called by the link layer when new ipv6 packets arrive for // this endpoint. func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { - headerView := pkt.Data.First() + headerView, ok := pkt.Data.PullUp(header.IPv6MinimumSize) + if !ok { + r.Stats().IP.MalformedPacketsReceived.Increment() + return + } h := header.IPv6(headerView) if !h.IsValid(pkt.Data.Size()) { r.Stats().IP.MalformedPacketsReceived.Increment() diff --git a/pkg/tcpip/stack/forwarder_test.go b/pkg/tcpip/stack/forwarder_test.go index e9c652042..c7c663498 100644 --- a/pkg/tcpip/stack/forwarder_test.go +++ b/pkg/tcpip/stack/forwarder_test.go @@ -70,7 +70,10 @@ func (f *fwdTestNetworkEndpoint) ID() *NetworkEndpointID { func (f *fwdTestNetworkEndpoint) HandlePacket(r *Route, pkt PacketBuffer) { // Consume the network header. - b := pkt.Data.First() + b, ok := pkt.Data.PullUp(fwdTestNetHeaderLen) + if !ok { + return + } pkt.Data.TrimFront(fwdTestNetHeaderLen) // Dispatch the packet to the transport protocol. @@ -473,7 +476,7 @@ func TestForwardingWithFakeResolverPartialTimeout(t *testing.T) { t.Fatal("packet not forwarded") } - b := p.Pkt.Header.View() + b := p.Pkt.Data.ToView() if b[0] != 3 { t.Fatalf("got b[0] = %d, want = 3", b[0]) } @@ -517,7 +520,7 @@ func TestForwardingWithFakeResolverTwoPackets(t *testing.T) { t.Fatal("packet not forwarded") } - b := p.Pkt.Header.View() + b := p.Pkt.Data.ToView() if b[0] != 3 { t.Fatalf("got b[0] = %d, want = 3", b[0]) } @@ -564,7 +567,7 @@ func TestForwardingWithFakeResolverManyPackets(t *testing.T) { t.Fatal("packet not forwarded") } - b := p.Pkt.Header.View() + b := p.Pkt.Data.ToView() if b[0] != 3 { t.Fatalf("got b[0] = %d, want = 3", b[0]) } @@ -619,7 +622,7 @@ func TestForwardingWithFakeResolverManyResolutions(t *testing.T) { // The first 5 packets (address 3 to 7) should not be forwarded // because their address resolutions are interrupted. - b := p.Pkt.Header.View() + b := p.Pkt.Data.ToView() if b[0] < 8 { t.Fatalf("got b[0] = %d, want b[0] >= 8", b[0]) } diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go index 6c0a4b24d..6b91159d4 100644 --- a/pkg/tcpip/stack/iptables.go +++ b/pkg/tcpip/stack/iptables.go @@ -212,6 +212,11 @@ func (it *IPTables) Check(hook Hook, pkt PacketBuffer) bool { // CheckPackets runs pkts through the rules for hook and returns a map of packets that // should not go forward. // +// Precondition: pkt is a IPv4 packet of at least length header.IPv4MinimumSize. +// +// TODO(gvisor.dev/issue/170): pk.NetworkHeader will always be set as a +// precondition. +// // NOTE: unlike the Check API the returned map contains packets that should be // dropped. func (it *IPTables) CheckPackets(hook Hook, pkts PacketBufferList) (drop map[*PacketBuffer]struct{}) { @@ -226,7 +231,9 @@ func (it *IPTables) CheckPackets(hook Hook, pkts PacketBufferList) (drop map[*Pa return drop } -// Precondition: pkt.NetworkHeader is set. +// Precondition: pkt is a IPv4 packet of at least length header.IPv4MinimumSize. +// TODO(gvisor.dev/issue/170): pk.NetworkHeader will always be set as a +// precondition. func (it *IPTables) checkChain(hook Hook, pkt PacketBuffer, table Table, ruleIdx int) chainVerdict { // Start from ruleIdx and walk the list of rules until a rule gives us // a verdict. @@ -271,14 +278,21 @@ func (it *IPTables) checkChain(hook Hook, pkt PacketBuffer, table Table, ruleIdx return chainDrop } -// Precondition: pk.NetworkHeader is set. +// Precondition: pkt is a IPv4 packet of at least length header.IPv4MinimumSize. +// TODO(gvisor.dev/issue/170): pk.NetworkHeader will always be set as a +// precondition. func (it *IPTables) checkRule(hook Hook, pkt PacketBuffer, table Table, ruleIdx int) (RuleVerdict, int) { rule := table.Rules[ruleIdx] // If pkt.NetworkHeader hasn't been set yet, it will be contained in - // pkt.Data.First(). + // pkt.Data. if pkt.NetworkHeader == nil { - pkt.NetworkHeader = pkt.Data.First() + var ok bool + pkt.NetworkHeader, ok = pkt.Data.PullUp(header.IPv4MinimumSize) + if !ok { + // Precondition has been violated. + panic(fmt.Sprintf("iptables checks require IPv4 headers of at least %d bytes", header.IPv4MinimumSize)) + } } // Check whether the packet matches the IP header filter. diff --git a/pkg/tcpip/stack/iptables_targets.go b/pkg/tcpip/stack/iptables_targets.go index 7b4543caf..8be61f4b1 100644 --- a/pkg/tcpip/stack/iptables_targets.go +++ b/pkg/tcpip/stack/iptables_targets.go @@ -96,9 +96,12 @@ func (rt RedirectTarget) Action(pkt PacketBuffer) (RuleVerdict, int) { newPkt := pkt.Clone() // Set network header. - headerView := newPkt.Data.First() + headerView, ok := newPkt.Data.PullUp(header.IPv4MinimumSize) + if !ok { + return RuleDrop, 0 + } netHeader := header.IPv4(headerView) - newPkt.NetworkHeader = headerView[:header.IPv4MinimumSize] + newPkt.NetworkHeader = headerView hlen := int(netHeader.HeaderLength()) tlen := int(netHeader.TotalLength()) @@ -117,10 +120,14 @@ func (rt RedirectTarget) Action(pkt PacketBuffer) (RuleVerdict, int) { if newPkt.TransportHeader != nil { udpHeader = header.UDP(newPkt.TransportHeader) } else { - if len(pkt.Data.First()) < header.UDPMinimumSize { + if pkt.Data.Size() < header.UDPMinimumSize { + return RuleDrop, 0 + } + hdr, ok := newPkt.Data.PullUp(header.UDPMinimumSize) + if !ok { return RuleDrop, 0 } - udpHeader = header.UDP(newPkt.Data.First()) + udpHeader = header.UDP(hdr) } udpHeader.SetDestinationPort(rt.MinPort) case header.TCPProtocolNumber: @@ -128,10 +135,14 @@ func (rt RedirectTarget) Action(pkt PacketBuffer) (RuleVerdict, int) { if newPkt.TransportHeader != nil { tcpHeader = header.TCP(newPkt.TransportHeader) } else { - if len(pkt.Data.First()) < header.TCPMinimumSize { + if pkt.Data.Size() < header.TCPMinimumSize { return RuleDrop, 0 } - tcpHeader = header.TCP(newPkt.TransportHeader) + hdr, ok := newPkt.Data.PullUp(header.TCPMinimumSize) + if !ok { + return RuleDrop, 0 + } + tcpHeader = header.TCP(hdr) } // TODO(gvisor.dev/issue/170): Need to recompute checksum // and implement nat connection tracking to support TCP. diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 016dbe15e..0c2b1f36a 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -1203,12 +1203,12 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link n.stack.stats.IP.PacketsReceived.Increment() } - if len(pkt.Data.First()) < netProto.MinimumPacketSize() { + netHeader, ok := pkt.Data.PullUp(netProto.MinimumPacketSize()) + if !ok { n.stack.stats.MalformedRcvdPackets.Increment() return } - - src, dst := netProto.ParseAddresses(pkt.Data.First()) + src, dst := netProto.ParseAddresses(netHeader) if n.stack.handleLocal && !n.isLoopback() && n.getRef(protocol, src) != nil { // The source address is one of our own, so we never should have gotten a @@ -1289,22 +1289,8 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link func (n *NIC) forwardPacket(r *Route, protocol tcpip.NetworkProtocolNumber, pkt PacketBuffer) { // TODO(b/143425874) Decrease the TTL field in forwarded packets. - - firstData := pkt.Data.First() - pkt.Data.RemoveFirst() - - if linkHeaderLen := int(n.linkEP.MaxHeaderLength()); linkHeaderLen == 0 { - pkt.Header = buffer.NewPrependableFromView(firstData) - } else { - firstDataLen := len(firstData) - - // pkt.Header should have enough capacity to hold n.linkEP's headers. - pkt.Header = buffer.NewPrependable(firstDataLen + linkHeaderLen) - - // TODO(b/151227689): avoid copying the packet when forwarding - if n := copy(pkt.Header.Prepend(firstDataLen), firstData); n != firstDataLen { - panic(fmt.Sprintf("copied %d bytes, expected %d", n, firstDataLen)) - } + if linkHeaderLen := int(n.linkEP.MaxHeaderLength()); linkHeaderLen != 0 { + pkt.Header = buffer.NewPrependable(linkHeaderLen) } if err := n.linkEP.WritePacket(r, nil /* gso */, protocol, pkt); err != nil { @@ -1332,12 +1318,13 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN // validly formed. n.stack.demux.deliverRawPacket(r, protocol, pkt) - if len(pkt.Data.First()) < transProto.MinimumPacketSize() { + transHeader, ok := pkt.Data.PullUp(transProto.MinimumPacketSize()) + if !ok { n.stack.stats.MalformedRcvdPackets.Increment() return } - srcPort, dstPort, err := transProto.ParsePorts(pkt.Data.First()) + srcPort, dstPort, err := transProto.ParsePorts(transHeader) if err != nil { n.stack.stats.MalformedRcvdPackets.Increment() return @@ -1375,11 +1362,12 @@ func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcp // ICMPv4 only guarantees that 8 bytes of the transport protocol will // be present in the payload. We know that the ports are within the // first 8 bytes for all known transport protocols. - if len(pkt.Data.First()) < 8 { + transHeader, ok := pkt.Data.PullUp(8) + if !ok { return } - srcPort, dstPort, err := transProto.ParsePorts(pkt.Data.First()) + srcPort, dstPort, err := transProto.ParsePorts(transHeader) if err != nil { return } diff --git a/pkg/tcpip/stack/packet_buffer.go b/pkg/tcpip/stack/packet_buffer.go index dc125f25e..7d36f8e84 100644 --- a/pkg/tcpip/stack/packet_buffer.go +++ b/pkg/tcpip/stack/packet_buffer.go @@ -37,7 +37,13 @@ type PacketBuffer struct { Data buffer.VectorisedView // Header holds the headers of outbound packets. As a packet is passed - // down the stack, each layer adds to Header. + // down the stack, each layer adds to Header. Note that forwarded + // packets don't populate Headers on their way out -- their headers and + // payload are never parsed out and remain in Data. + // + // TODO(gvisor.dev/issue/170): Forwarded packets don't currently + // populate Header, but should. This will be doable once early parsing + // (https://github.com/google/gvisor/pull/1995) is supported. Header buffer.Prependable // These fields are used by both inbound and outbound packets. They diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index c7634ceb1..d45d2cc1f 100644 --- a/pkg/tcpip/stack/stack_test.go +++ b/pkg/tcpip/stack/stack_test.go @@ -95,16 +95,18 @@ func (f *fakeNetworkEndpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffe f.proto.packetCount[int(f.id.LocalAddress[0])%len(f.proto.packetCount)]++ // Consume the network header. - b := pkt.Data.First() + b, ok := pkt.Data.PullUp(fakeNetHeaderLen) + if !ok { + return + } pkt.Data.TrimFront(fakeNetHeaderLen) // Handle control packets. if b[2] == uint8(fakeControlProtocol) { - nb := pkt.Data.First() - if len(nb) < fakeNetHeaderLen { + nb, ok := pkt.Data.PullUp(fakeNetHeaderLen) + if !ok { return } - pkt.Data.TrimFront(fakeNetHeaderLen) f.dispatcher.DeliverTransportControlPacket(tcpip.Address(nb[1:2]), tcpip.Address(nb[0:1]), fakeNetNumber, tcpip.TransportProtocolNumber(nb[2]), stack.ControlPortUnreachable, 0, pkt) return diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go index 3084e6593..a611e44ab 100644 --- a/pkg/tcpip/stack/transport_test.go +++ b/pkg/tcpip/stack/transport_test.go @@ -642,10 +642,11 @@ func TestTransportForwarding(t *testing.T) { t.Fatal("Response packet not forwarded") } - if dst := p.Pkt.Header.View()[0]; dst != 3 { + hdrs := p.Pkt.Data.ToView() + if dst := hdrs[0]; dst != 3 { t.Errorf("Response packet has incorrect destination addresss: got = %d, want = 3", dst) } - if src := p.Pkt.Header.View()[1]; src != 1 { + if src := hdrs[1]; src != 1 { t.Errorf("Response packet has incorrect source addresss: got = %d, want = 3", src) } } diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index feef8dca0..b1d820372 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -747,15 +747,15 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk // Only accept echo replies. switch e.NetProto { case header.IPv4ProtocolNumber: - h := header.ICMPv4(pkt.Data.First()) - if h.Type() != header.ICMPv4EchoReply { + h, ok := pkt.Data.PullUp(header.ICMPv4MinimumSize) + if !ok || header.ICMPv4(h).Type() != header.ICMPv4EchoReply { e.stack.Stats().DroppedPackets.Increment() e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() return } case header.IPv6ProtocolNumber: - h := header.ICMPv6(pkt.Data.First()) - if h.Type() != header.ICMPv6EchoReply { + h, ok := pkt.Data.PullUp(header.ICMPv6MinimumSize) + if !ok || header.ICMPv6(h).Type() != header.ICMPv6EchoReply { e.stack.Stats().DroppedPackets.Increment() e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() return diff --git a/pkg/tcpip/transport/tcp/segment.go b/pkg/tcpip/transport/tcp/segment.go index 40461fd31..7712ce652 100644 --- a/pkg/tcpip/transport/tcp/segment.go +++ b/pkg/tcpip/transport/tcp/segment.go @@ -144,7 +144,11 @@ func (s *segment) logicalLen() seqnum.Size { // TCP checksum and stores the checksum and result of checksum verification in // the csum and csumValid fields of the segment. func (s *segment) parse() bool { - h := header.TCP(s.data.First()) + h, ok := s.data.PullUp(header.TCPMinimumSize) + if !ok { + return false + } + hdr := header.TCP(h) // h is the header followed by the payload. We check that the offset to // the data respects the following constraints: @@ -156,12 +160,16 @@ func (s *segment) parse() bool { // N.B. The segment has already been validated as having at least the // minimum TCP size before reaching here, so it's safe to read the // fields. - offset := int(h.DataOffset()) - if offset < header.TCPMinimumSize || offset > len(h) { + offset := int(hdr.DataOffset()) + if offset < header.TCPMinimumSize { + return false + } + hdrWithOpts, ok := s.data.PullUp(offset) + if !ok { return false } - s.options = []byte(h[header.TCPMinimumSize:offset]) + s.options = []byte(hdrWithOpts[header.TCPMinimumSize:]) s.parsedOptions = header.ParseTCPOptions(s.options) // Query the link capabilities to decide if checksum validation is @@ -173,18 +181,19 @@ func (s *segment) parse() bool { s.data.TrimFront(offset) } if verifyChecksum { - s.csum = h.Checksum() + hdr = header.TCP(hdrWithOpts) + s.csum = hdr.Checksum() xsum := s.route.PseudoHeaderChecksum(ProtocolNumber, uint16(s.data.Size())) - xsum = h.CalculateChecksum(xsum) + xsum = hdr.CalculateChecksum(xsum) s.data.TrimFront(offset) xsum = header.ChecksumVV(s.data, xsum) s.csumValid = xsum == 0xffff } - s.sequenceNumber = seqnum.Value(h.SequenceNumber()) - s.ackNumber = seqnum.Value(h.AckNumber()) - s.flags = h.Flags() - s.window = seqnum.Size(h.WindowSize()) + s.sequenceNumber = seqnum.Value(hdr.SequenceNumber()) + s.ackNumber = seqnum.Value(hdr.AckNumber()) + s.flags = hdr.Flags() + s.window = seqnum.Size(hdr.WindowSize()) return true } diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index ab1014c7f..286c66cf5 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -3548,7 +3548,7 @@ func TestReceivedInvalidSegmentCountIncrement(t *testing.T) { AckNum: c.IRS.Add(1), RcvWnd: 30000, }) - tcpbuf := vv.First()[header.IPv4MinimumSize:] + tcpbuf := vv.ToView()[header.IPv4MinimumSize:] tcpbuf[header.TCPDataOffset] = ((header.TCPMinimumSize - 1) / 4) << 4 c.SendSegment(vv) @@ -3575,7 +3575,7 @@ func TestReceivedIncorrectChecksumIncrement(t *testing.T) { AckNum: c.IRS.Add(1), RcvWnd: 30000, }) - tcpbuf := vv.First()[header.IPv4MinimumSize:] + tcpbuf := vv.ToView()[header.IPv4MinimumSize:] // Overwrite a byte in the payload which should cause checksum // verification to fail. tcpbuf[(tcpbuf[header.TCPDataOffset]>>4)*4] = 0x4 diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index edb54f0be..756ab913a 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -1250,8 +1250,8 @@ func (e *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { // endpoint. func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) { // Get the header then trim it from the view. - hdr := header.UDP(pkt.Data.First()) - if int(hdr.Length()) > pkt.Data.Size() { + hdr, ok := pkt.Data.PullUp(header.UDPMinimumSize) + if !ok || int(header.UDP(hdr).Length()) > pkt.Data.Size() { // Malformed packet. e.stack.Stats().UDP.MalformedPacketsReceived.Increment() e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() @@ -1286,7 +1286,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk senderAddress: tcpip.FullAddress{ NIC: r.NICID(), Addr: id.RemoteAddress, - Port: hdr.SourcePort(), + Port: header.UDP(hdr).SourcePort(), }, } packet.data = pkt.Data diff --git a/pkg/tcpip/transport/udp/protocol.go b/pkg/tcpip/transport/udp/protocol.go index 6e31a9bac..52af6de22 100644 --- a/pkg/tcpip/transport/udp/protocol.go +++ b/pkg/tcpip/transport/udp/protocol.go @@ -68,8 +68,13 @@ func (*protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) { // that don't match any existing endpoint. func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) bool { // Get the header then trim it from the view. - hdr := header.UDP(pkt.Data.First()) - if int(hdr.Length()) > pkt.Data.Size() { + h, ok := pkt.Data.PullUp(header.UDPMinimumSize) + if !ok { + // Malformed packet. + r.Stack().Stats().UDP.MalformedPacketsReceived.Increment() + return true + } + if int(header.UDP(h).Length()) > pkt.Data.Size() { // Malformed packet. r.Stack().Stats().UDP.MalformedPacketsReceived.Increment() return true -- cgit v1.2.3 From 696feaf10c9339a57d177a913e847ddb488ece69 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Thu, 23 Apr 2020 17:32:59 -0700 Subject: Port devpts to VFS2. PiperOrigin-RevId: 308164359 --- pkg/sentry/fs/tty/line_discipline.go | 4 + pkg/sentry/fs/tty/master.go | 4 + pkg/sentry/fs/tty/queue.go | 4 + pkg/sentry/fs/tty/slave.go | 4 + pkg/sentry/fs/tty/terminal.go | 4 + pkg/sentry/fsimpl/devpts/BUILD | 43 +++ pkg/sentry/fsimpl/devpts/devpts.go | 207 +++++++++++++ pkg/sentry/fsimpl/devpts/devpts_test.go | 56 ++++ pkg/sentry/fsimpl/devpts/line_discipline.go | 449 ++++++++++++++++++++++++++++ pkg/sentry/fsimpl/devpts/master.go | 226 ++++++++++++++ pkg/sentry/fsimpl/devpts/queue.go | 240 +++++++++++++++ pkg/sentry/fsimpl/devpts/slave.go | 186 ++++++++++++ pkg/sentry/fsimpl/devpts/terminal.go | 124 ++++++++ pkg/sentry/fsimpl/devtmpfs/devtmpfs.go | 13 +- pkg/sentry/fsimpl/kernfs/filesystem.go | 2 +- pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 9 +- 16 files changed, 1570 insertions(+), 5 deletions(-) create mode 100644 pkg/sentry/fsimpl/devpts/BUILD create mode 100644 pkg/sentry/fsimpl/devpts/devpts.go create mode 100644 pkg/sentry/fsimpl/devpts/devpts_test.go create mode 100644 pkg/sentry/fsimpl/devpts/line_discipline.go create mode 100644 pkg/sentry/fsimpl/devpts/master.go create mode 100644 pkg/sentry/fsimpl/devpts/queue.go create mode 100644 pkg/sentry/fsimpl/devpts/slave.go create mode 100644 pkg/sentry/fsimpl/devpts/terminal.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/tty/line_discipline.go b/pkg/sentry/fs/tty/line_discipline.go index 12b1c6097..2e9dd2d55 100644 --- a/pkg/sentry/fs/tty/line_discipline.go +++ b/pkg/sentry/fs/tty/line_discipline.go @@ -27,6 +27,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +// LINT.IfChange + const ( // canonMaxBytes is the number of bytes that fit into a single line of // terminal input in canonical mode. This corresponds to N_TTY_BUF_SIZE @@ -443,3 +445,5 @@ func (l *lineDiscipline) peek(b []byte) int { } return size } + +// LINT.ThenChange(../../fsimpl/devpts/line_discipline.go) diff --git a/pkg/sentry/fs/tty/master.go b/pkg/sentry/fs/tty/master.go index f62da49bd..fe07fa929 100644 --- a/pkg/sentry/fs/tty/master.go +++ b/pkg/sentry/fs/tty/master.go @@ -26,6 +26,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +// LINT.IfChange + // masterInodeOperations are the fs.InodeOperations for the master end of the // Terminal (ptmx file). // @@ -232,3 +234,5 @@ func maybeEmitUnimplementedEvent(ctx context.Context, cmd uint32) { unimpl.EmitUnimplementedEvent(ctx) } } + +// LINT.ThenChange(../../fsimpl/devpts/master.go) diff --git a/pkg/sentry/fs/tty/queue.go b/pkg/sentry/fs/tty/queue.go index 1ca79c0b2..ceabb9b1e 100644 --- a/pkg/sentry/fs/tty/queue.go +++ b/pkg/sentry/fs/tty/queue.go @@ -25,6 +25,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +// LINT.IfChange + // waitBufMaxBytes is the maximum size of a wait buffer. It is based on // TTYB_DEFAULT_MEM_LIMIT. const waitBufMaxBytes = 131072 @@ -234,3 +236,5 @@ func (q *queue) waitBufAppend(b []byte) { q.waitBuf = append(q.waitBuf, b) q.waitBufLen += uint64(len(b)) } + +// LINT.ThenChange(../../fsimpl/devpts/queue.go) diff --git a/pkg/sentry/fs/tty/slave.go b/pkg/sentry/fs/tty/slave.go index 6a2dbc576..9871f6fc6 100644 --- a/pkg/sentry/fs/tty/slave.go +++ b/pkg/sentry/fs/tty/slave.go @@ -25,6 +25,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +// LINT.IfChange + // slaveInodeOperations are the fs.InodeOperations for the slave end of the // Terminal (pts file). // @@ -172,3 +174,5 @@ func (sf *slaveFileOperations) Ioctl(ctx context.Context, _ *fs.File, io usermem return 0, syserror.ENOTTY } } + +// LINT.ThenChange(../../fsimpl/devpts/slave.go) diff --git a/pkg/sentry/fs/tty/terminal.go b/pkg/sentry/fs/tty/terminal.go index 5883f26db..ddcccf4da 100644 --- a/pkg/sentry/fs/tty/terminal.go +++ b/pkg/sentry/fs/tty/terminal.go @@ -23,6 +23,8 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) +// LINT.IfChange + // Terminal is a pseudoterminal. // // +stateify savable @@ -126,3 +128,5 @@ func (tm *Terminal) tty(isMaster bool) *kernel.TTY { } return tm.slaveKTTY } + +// LINT.ThenChange(../../fsimpl/devpts/terminal.go) diff --git a/pkg/sentry/fsimpl/devpts/BUILD b/pkg/sentry/fsimpl/devpts/BUILD new file mode 100644 index 000000000..585764223 --- /dev/null +++ b/pkg/sentry/fsimpl/devpts/BUILD @@ -0,0 +1,43 @@ +load("//tools:defs.bzl", "go_library", "go_test") + +licenses(["notice"]) + +go_library( + name = "devpts", + srcs = [ + "devpts.go", + "line_discipline.go", + "master.go", + "queue.go", + "slave.go", + "terminal.go", + ], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/abi/linux", + "//pkg/context", + "//pkg/safemem", + "//pkg/sentry/arch", + "//pkg/sentry/fsimpl/kernfs", + "//pkg/sentry/kernel", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/unimpl", + "//pkg/sentry/vfs", + "//pkg/sync", + "//pkg/syserror", + "//pkg/usermem", + "//pkg/waiter", + ], +) + +go_test( + name = "devpts_test", + size = "small", + srcs = ["devpts_test.go"], + library = ":devpts", + deps = [ + "//pkg/abi/linux", + "//pkg/sentry/contexttest", + "//pkg/usermem", + ], +) diff --git a/pkg/sentry/fsimpl/devpts/devpts.go b/pkg/sentry/fsimpl/devpts/devpts.go new file mode 100644 index 000000000..07a69b940 --- /dev/null +++ b/pkg/sentry/fsimpl/devpts/devpts.go @@ -0,0 +1,207 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package devpts provides a filesystem implementation that behaves like +// devpts. +package devpts + +import ( + "fmt" + "math" + "sort" + "strconv" + "sync" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// Name is the filesystem name. +const Name = "devpts" + +// FilesystemType implements vfs.FilesystemType. +type FilesystemType struct{} + +// Name implements vfs.FilesystemType.Name. +func (FilesystemType) Name() string { + return Name +} + +var _ vfs.FilesystemType = (*FilesystemType)(nil) + +// GetFilesystem implements vfs.FilesystemType.GetFilesystem. +func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { + // No data allowed. + if opts.Data != "" { + return nil, nil, syserror.EINVAL + } + + fs, root := fstype.newFilesystem(vfsObj, creds) + return fs.VFSFilesystem(), root.VFSDentry(), nil +} + +// newFilesystem creates a new devpts filesystem with root directory and ptmx +// master inode. It returns the filesystem and root Dentry. +func (fstype FilesystemType) newFilesystem(vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials) (*kernfs.Filesystem, *kernfs.Dentry) { + fs := &kernfs.Filesystem{} + fs.Init(vfsObj, fstype) + + // Construct the root directory. This is always inode id 1. + root := &rootInode{ + slaves: make(map[uint32]*slaveInode), + } + root.InodeAttrs.Init(creds, 1, linux.ModeDirectory|0555) + root.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) + root.dentry.Init(root) + + // Construct the pts master inode and dentry. Linux always uses inode + // id 2 for ptmx. See fs/devpts/inode.c:mknod_ptmx. + master := &masterInode{ + root: root, + } + master.InodeAttrs.Init(creds, 2, linux.ModeCharacterDevice|0666) + master.dentry.Init(master) + + // Add the master as a child of the root. + links := root.OrderedChildren.Populate(&root.dentry, map[string]*kernfs.Dentry{ + "ptmx": &master.dentry, + }) + root.IncLinks(links) + + return fs, &root.dentry +} + +// rootInode is the root directory inode for the devpts mounts. +type rootInode struct { + kernfs.AlwaysValid + kernfs.InodeAttrs + kernfs.InodeDirectoryNoNewChildren + kernfs.InodeNotSymlink + kernfs.OrderedChildren + + // Keep a reference to this inode's dentry. + dentry kernfs.Dentry + + // master is the master pty inode. Immutable. + master *masterInode + + // root is the root directory inode for this filesystem. Immutable. + root *rootInode + + // mu protects the fields below. + mu sync.Mutex + + // slaves maps pty ids to slave inodes. + slaves map[uint32]*slaveInode + + // nextIdx is the next pty index to use. Must be accessed atomically. + // + // TODO(b/29356795): reuse indices when ptys are closed. + nextIdx uint32 +} + +var _ kernfs.Inode = (*rootInode)(nil) + +// allocateTerminal creates a new Terminal and installs a pts node for it. +func (i *rootInode) allocateTerminal(creds *auth.Credentials) (*Terminal, error) { + i.mu.Lock() + defer i.mu.Unlock() + if i.nextIdx == math.MaxUint32 { + return nil, syserror.ENOMEM + } + idx := i.nextIdx + i.nextIdx++ + + // Sanity check that slave with idx does not exist. + if _, ok := i.slaves[idx]; ok { + panic(fmt.Sprintf("pty index collision; index %d already exists", idx)) + } + + // Create the new terminal and slave. + t := newTerminal(idx) + slave := &slaveInode{ + root: i, + t: t, + } + // Linux always uses pty index + 3 as the inode id. See + // fs/devpts/inode.c:devpts_pty_new(). + slave.InodeAttrs.Init(creds, uint64(idx+3), linux.ModeCharacterDevice|0600) + slave.dentry.Init(slave) + i.slaves[idx] = slave + + return t, nil +} + +// masterClose is called when the master end of t is closed. +func (i *rootInode) masterClose(t *Terminal) { + i.mu.Lock() + defer i.mu.Unlock() + + // Sanity check that slave with idx exists. + if _, ok := i.slaves[t.n]; !ok { + panic(fmt.Sprintf("pty with index %d does not exist", t.n)) + } + delete(i.slaves, t.n) +} + +// Open implements kernfs.Inode.Open. +func (i *rootInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd := &kernfs.GenericDirectoryFD{} + fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, &opts) + return fd.VFSFileDescription(), nil +} + +// Lookup implements kernfs.Inode.Lookup. +func (i *rootInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { + idx, err := strconv.ParseUint(name, 10, 32) + if err != nil { + return nil, syserror.ENOENT + } + i.mu.Lock() + defer i.mu.Unlock() + if si, ok := i.slaves[uint32(idx)]; ok { + si.dentry.IncRef() + return si.dentry.VFSDentry(), nil + + } + return nil, syserror.ENOENT +} + +// IterDirents implements kernfs.Inode.IterDirents. +func (i *rootInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, offset, relOffset int64) (int64, error) { + i.mu.Lock() + defer i.mu.Unlock() + ids := make([]int, 0, len(i.slaves)) + for id := range i.slaves { + ids = append(ids, int(id)) + } + sort.Ints(ids) + for _, id := range ids[relOffset:] { + dirent := vfs.Dirent{ + Name: strconv.FormatUint(uint64(id), 10), + Type: linux.DT_CHR, + Ino: i.slaves[uint32(id)].InodeAttrs.Ino(), + NextOff: offset + 1, + } + if err := cb.Handle(dirent); err != nil { + return offset, err + } + offset++ + } + return offset, nil +} diff --git a/pkg/sentry/fsimpl/devpts/devpts_test.go b/pkg/sentry/fsimpl/devpts/devpts_test.go new file mode 100644 index 000000000..b7c149047 --- /dev/null +++ b/pkg/sentry/fsimpl/devpts/devpts_test.go @@ -0,0 +1,56 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package devpts + +import ( + "testing" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/contexttest" + "gvisor.dev/gvisor/pkg/usermem" +) + +func TestSimpleMasterToSlave(t *testing.T) { + ld := newLineDiscipline(linux.DefaultSlaveTermios) + ctx := contexttest.Context(t) + inBytes := []byte("hello, tty\n") + src := usermem.BytesIOSequence(inBytes) + outBytes := make([]byte, 32) + dst := usermem.BytesIOSequence(outBytes) + + // Write to the input queue. + nw, err := ld.inputQueueWrite(ctx, src) + if err != nil { + t.Fatalf("error writing to input queue: %v", err) + } + if nw != int64(len(inBytes)) { + t.Fatalf("wrote wrong length: got %d, want %d", nw, len(inBytes)) + } + + // Read from the input queue. + nr, err := ld.inputQueueRead(ctx, dst) + if err != nil { + t.Fatalf("error reading from input queue: %v", err) + } + if nr != int64(len(inBytes)) { + t.Fatalf("read wrong length: got %d, want %d", nr, len(inBytes)) + } + + outStr := string(outBytes[:nr]) + inStr := string(inBytes) + if outStr != inStr { + t.Fatalf("written and read strings do not match: got %q, want %q", outStr, inStr) + } +} diff --git a/pkg/sentry/fsimpl/devpts/line_discipline.go b/pkg/sentry/fsimpl/devpts/line_discipline.go new file mode 100644 index 000000000..e201801d6 --- /dev/null +++ b/pkg/sentry/fsimpl/devpts/line_discipline.go @@ -0,0 +1,449 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package devpts + +import ( + "bytes" + "unicode/utf8" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +// LINT.IfChange + +const ( + // canonMaxBytes is the number of bytes that fit into a single line of + // terminal input in canonical mode. This corresponds to N_TTY_BUF_SIZE + // in include/linux/tty.h. + canonMaxBytes = 4096 + + // nonCanonMaxBytes is the maximum number of bytes that can be read at + // a time in noncanonical mode. + nonCanonMaxBytes = canonMaxBytes - 1 + + spacesPerTab = 8 +) + +// lineDiscipline dictates how input and output are handled between the +// pseudoterminal (pty) master and slave. It can be configured to alter I/O, +// modify control characters (e.g. Ctrl-C for SIGINT), etc. The following man +// pages are good resources for how to affect the line discipline: +// +// * termios(3) +// * tty_ioctl(4) +// +// This file corresponds most closely to drivers/tty/n_tty.c. +// +// lineDiscipline has a simple structure but supports a multitude of options +// (see the above man pages). It consists of two queues of bytes: one from the +// terminal master to slave (the input queue) and one from slave to master (the +// output queue). When bytes are written to one end of the pty, the line +// discipline reads the bytes, modifies them or takes special action if +// required, and enqueues them to be read by the other end of the pty: +// +// input from terminal +-------------+ input to process (e.g. bash) +// +------------------------>| input queue |---------------------------+ +// | (inputQueueWrite) +-------------+ (inputQueueRead) | +// | | +// | v +// masterFD slaveFD +// ^ | +// | | +// | output to terminal +--------------+ output from process | +// +------------------------| output queue |<--------------------------+ +// (outputQueueRead) +--------------+ (outputQueueWrite) +// +// Lock order: +// termiosMu +// inQueue.mu +// outQueue.mu +// +// +stateify savable +type lineDiscipline struct { + // sizeMu protects size. + sizeMu sync.Mutex `state:"nosave"` + + // size is the terminal size (width and height). + size linux.WindowSize + + // inQueue is the input queue of the terminal. + inQueue queue + + // outQueue is the output queue of the terminal. + outQueue queue + + // termiosMu protects termios. + termiosMu sync.RWMutex `state:"nosave"` + + // termios is the terminal configuration used by the lineDiscipline. + termios linux.KernelTermios + + // column is the location in a row of the cursor. This is important for + // handling certain special characters like backspace. + column int + + // masterWaiter is used to wait on the master end of the TTY. + masterWaiter waiter.Queue `state:"zerovalue"` + + // slaveWaiter is used to wait on the slave end of the TTY. + slaveWaiter waiter.Queue `state:"zerovalue"` +} + +func newLineDiscipline(termios linux.KernelTermios) *lineDiscipline { + ld := lineDiscipline{termios: termios} + ld.inQueue.transformer = &inputQueueTransformer{} + ld.outQueue.transformer = &outputQueueTransformer{} + return &ld +} + +// getTermios gets the linux.Termios for the tty. +func (l *lineDiscipline) getTermios(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { + l.termiosMu.RLock() + defer l.termiosMu.RUnlock() + // We must copy a Termios struct, not KernelTermios. + t := l.termios.ToTermios() + _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), t, usermem.IOOpts{ + AddressSpaceActive: true, + }) + return 0, err +} + +// setTermios sets a linux.Termios for the tty. +func (l *lineDiscipline) setTermios(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { + l.termiosMu.Lock() + defer l.termiosMu.Unlock() + oldCanonEnabled := l.termios.LEnabled(linux.ICANON) + // We must copy a Termios struct, not KernelTermios. + var t linux.Termios + _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &t, usermem.IOOpts{ + AddressSpaceActive: true, + }) + l.termios.FromTermios(t) + + // If canonical mode is turned off, move bytes from inQueue's wait + // buffer to its read buffer. Anything already in the read buffer is + // now readable. + if oldCanonEnabled && !l.termios.LEnabled(linux.ICANON) { + l.inQueue.mu.Lock() + l.inQueue.pushWaitBufLocked(l) + l.inQueue.readable = true + l.inQueue.mu.Unlock() + l.slaveWaiter.Notify(waiter.EventIn) + } + + return 0, err +} + +func (l *lineDiscipline) windowSize(ctx context.Context, io usermem.IO, args arch.SyscallArguments) error { + l.sizeMu.Lock() + defer l.sizeMu.Unlock() + _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), l.size, usermem.IOOpts{ + AddressSpaceActive: true, + }) + return err +} + +func (l *lineDiscipline) setWindowSize(ctx context.Context, io usermem.IO, args arch.SyscallArguments) error { + l.sizeMu.Lock() + defer l.sizeMu.Unlock() + _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &l.size, usermem.IOOpts{ + AddressSpaceActive: true, + }) + return err +} + +func (l *lineDiscipline) masterReadiness() waiter.EventMask { + // We don't have to lock a termios because the default master termios + // is immutable. + return l.inQueue.writeReadiness(&linux.MasterTermios) | l.outQueue.readReadiness(&linux.MasterTermios) +} + +func (l *lineDiscipline) slaveReadiness() waiter.EventMask { + l.termiosMu.RLock() + defer l.termiosMu.RUnlock() + return l.outQueue.writeReadiness(&l.termios) | l.inQueue.readReadiness(&l.termios) +} + +func (l *lineDiscipline) inputQueueReadSize(ctx context.Context, io usermem.IO, args arch.SyscallArguments) error { + return l.inQueue.readableSize(ctx, io, args) +} + +func (l *lineDiscipline) inputQueueRead(ctx context.Context, dst usermem.IOSequence) (int64, error) { + l.termiosMu.RLock() + defer l.termiosMu.RUnlock() + n, pushed, err := l.inQueue.read(ctx, dst, l) + if err != nil { + return 0, err + } + if n > 0 { + l.masterWaiter.Notify(waiter.EventOut) + if pushed { + l.slaveWaiter.Notify(waiter.EventIn) + } + return n, nil + } + return 0, syserror.ErrWouldBlock +} + +func (l *lineDiscipline) inputQueueWrite(ctx context.Context, src usermem.IOSequence) (int64, error) { + l.termiosMu.RLock() + defer l.termiosMu.RUnlock() + n, err := l.inQueue.write(ctx, src, l) + if err != nil { + return 0, err + } + if n > 0 { + l.slaveWaiter.Notify(waiter.EventIn) + return n, nil + } + return 0, syserror.ErrWouldBlock +} + +func (l *lineDiscipline) outputQueueReadSize(ctx context.Context, io usermem.IO, args arch.SyscallArguments) error { + return l.outQueue.readableSize(ctx, io, args) +} + +func (l *lineDiscipline) outputQueueRead(ctx context.Context, dst usermem.IOSequence) (int64, error) { + l.termiosMu.RLock() + defer l.termiosMu.RUnlock() + n, pushed, err := l.outQueue.read(ctx, dst, l) + if err != nil { + return 0, err + } + if n > 0 { + l.slaveWaiter.Notify(waiter.EventOut) + if pushed { + l.masterWaiter.Notify(waiter.EventIn) + } + return n, nil + } + return 0, syserror.ErrWouldBlock +} + +func (l *lineDiscipline) outputQueueWrite(ctx context.Context, src usermem.IOSequence) (int64, error) { + l.termiosMu.RLock() + defer l.termiosMu.RUnlock() + n, err := l.outQueue.write(ctx, src, l) + if err != nil { + return 0, err + } + if n > 0 { + l.masterWaiter.Notify(waiter.EventIn) + return n, nil + } + return 0, syserror.ErrWouldBlock +} + +// transformer is a helper interface to make it easier to stateify queue. +type transformer interface { + // transform functions require queue's mutex to be held. + transform(*lineDiscipline, *queue, []byte) int +} + +// outputQueueTransformer implements transformer. It performs line discipline +// transformations on the output queue. +// +// +stateify savable +type outputQueueTransformer struct{} + +// transform does output processing for one end of the pty. See +// drivers/tty/n_tty.c:do_output_char for an analogous kernel function. +// +// Preconditions: +// * l.termiosMu must be held for reading. +// * q.mu must be held. +func (*outputQueueTransformer) transform(l *lineDiscipline, q *queue, buf []byte) int { + // transformOutput is effectively always in noncanonical mode, as the + // master termios never has ICANON set. + + if !l.termios.OEnabled(linux.OPOST) { + q.readBuf = append(q.readBuf, buf...) + if len(q.readBuf) > 0 { + q.readable = true + } + return len(buf) + } + + var ret int + for len(buf) > 0 { + size := l.peek(buf) + cBytes := append([]byte{}, buf[:size]...) + ret += size + buf = buf[size:] + // We're guaranteed that cBytes has at least one element. + switch cBytes[0] { + case '\n': + if l.termios.OEnabled(linux.ONLRET) { + l.column = 0 + } + if l.termios.OEnabled(linux.ONLCR) { + q.readBuf = append(q.readBuf, '\r', '\n') + continue + } + case '\r': + if l.termios.OEnabled(linux.ONOCR) && l.column == 0 { + continue + } + if l.termios.OEnabled(linux.OCRNL) { + cBytes[0] = '\n' + if l.termios.OEnabled(linux.ONLRET) { + l.column = 0 + } + break + } + l.column = 0 + case '\t': + spaces := spacesPerTab - l.column%spacesPerTab + if l.termios.OutputFlags&linux.TABDLY == linux.XTABS { + l.column += spaces + q.readBuf = append(q.readBuf, bytes.Repeat([]byte{' '}, spacesPerTab)...) + continue + } + l.column += spaces + case '\b': + if l.column > 0 { + l.column-- + } + default: + l.column++ + } + q.readBuf = append(q.readBuf, cBytes...) + } + if len(q.readBuf) > 0 { + q.readable = true + } + return ret +} + +// inputQueueTransformer implements transformer. It performs line discipline +// transformations on the input queue. +// +// +stateify savable +type inputQueueTransformer struct{} + +// transform does input processing for one end of the pty. Characters read are +// transformed according to flags set in the termios struct. See +// drivers/tty/n_tty.c:n_tty_receive_char_special for an analogous kernel +// function. +// +// Preconditions: +// * l.termiosMu must be held for reading. +// * q.mu must be held. +func (*inputQueueTransformer) transform(l *lineDiscipline, q *queue, buf []byte) int { + // If there's a line waiting to be read in canonical mode, don't write + // anything else to the read buffer. + if l.termios.LEnabled(linux.ICANON) && q.readable { + return 0 + } + + maxBytes := nonCanonMaxBytes + if l.termios.LEnabled(linux.ICANON) { + maxBytes = canonMaxBytes + } + + var ret int + for len(buf) > 0 && len(q.readBuf) < canonMaxBytes { + size := l.peek(buf) + cBytes := append([]byte{}, buf[:size]...) + // We're guaranteed that cBytes has at least one element. + switch cBytes[0] { + case '\r': + if l.termios.IEnabled(linux.IGNCR) { + buf = buf[size:] + ret += size + continue + } + if l.termios.IEnabled(linux.ICRNL) { + cBytes[0] = '\n' + } + case '\n': + if l.termios.IEnabled(linux.INLCR) { + cBytes[0] = '\r' + } + } + + // In canonical mode, we discard non-terminating characters + // after the first 4095. + if l.shouldDiscard(q, cBytes) { + buf = buf[size:] + ret += size + continue + } + + // Stop if the buffer would be overfilled. + if len(q.readBuf)+size > maxBytes { + break + } + buf = buf[size:] + ret += size + + // If we get EOF, make the buffer available for reading. + if l.termios.LEnabled(linux.ICANON) && l.termios.IsEOF(cBytes[0]) { + q.readable = true + break + } + + q.readBuf = append(q.readBuf, cBytes...) + + // Anything written to the readBuf will have to be echoed. + if l.termios.LEnabled(linux.ECHO) { + l.outQueue.writeBytes(cBytes, l) + l.masterWaiter.Notify(waiter.EventIn) + } + + // If we finish a line, make it available for reading. + if l.termios.LEnabled(linux.ICANON) && l.termios.IsTerminating(cBytes) { + q.readable = true + break + } + } + + // In noncanonical mode, everything is readable. + if !l.termios.LEnabled(linux.ICANON) && len(q.readBuf) > 0 { + q.readable = true + } + + return ret +} + +// shouldDiscard returns whether c should be discarded. In canonical mode, if +// too many bytes are enqueued, we keep reading input and discarding it until +// we find a terminating character. Signal/echo processing still occurs. +// +// Precondition: +// * l.termiosMu must be held for reading. +// * q.mu must be held. +func (l *lineDiscipline) shouldDiscard(q *queue, cBytes []byte) bool { + return l.termios.LEnabled(linux.ICANON) && len(q.readBuf)+len(cBytes) >= canonMaxBytes && !l.termios.IsTerminating(cBytes) +} + +// peek returns the size in bytes of the next character to process. As long as +// b isn't empty, peek returns a value of at least 1. +func (l *lineDiscipline) peek(b []byte) int { + size := 1 + // If UTF-8 support is enabled, runes might be multiple bytes. + if l.termios.IEnabled(linux.IUTF8) { + _, size = utf8.DecodeRune(b) + } + return size +} + +// LINT.ThenChange(../../fs/tty/line_discipline.go) diff --git a/pkg/sentry/fsimpl/devpts/master.go b/pkg/sentry/fsimpl/devpts/master.go new file mode 100644 index 000000000..60340c28e --- /dev/null +++ b/pkg/sentry/fsimpl/devpts/master.go @@ -0,0 +1,226 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package devpts + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/unimpl" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +// LINT.IfChange + +// masterInode is the inode for the master end of the Terminal. +type masterInode struct { + kernfs.InodeAttrs + kernfs.InodeNoopRefCount + kernfs.InodeNotDirectory + kernfs.InodeNotSymlink + + // Keep a reference to this inode's dentry. + dentry kernfs.Dentry + + // root is the devpts root inode. + root *rootInode +} + +var _ kernfs.Inode = (*masterInode)(nil) + +// Open implements kernfs.Inode.Open. +func (mi *masterInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + t, err := mi.root.allocateTerminal(rp.Credentials()) + if err != nil { + return nil, err + } + + mi.IncRef() + fd := &masterFileDescription{ + inode: mi, + t: t, + } + if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}); err != nil { + mi.DecRef() + return nil, err + } + return &fd.vfsfd, nil +} + +// Stat implements kernfs.Inode.Stat. +func (mi *masterInode) Stat(vfsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { + statx, err := mi.InodeAttrs.Stat(vfsfs, opts) + if err != nil { + return linux.Statx{}, err + } + statx.Blksize = 1024 + statx.RdevMajor = linux.TTYAUX_MAJOR + statx.RdevMinor = linux.PTMX_MINOR + return statx, nil +} + +// SetStat implements kernfs.Inode.SetStat +func (mi *masterInode) SetStat(ctx context.Context, vfsfs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { + if opts.Stat.Mask&linux.STATX_SIZE != 0 { + return syserror.EINVAL + } + return mi.InodeAttrs.SetStat(ctx, vfsfs, creds, opts) +} + +type masterFileDescription struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl + + inode *masterInode + t *Terminal +} + +var _ vfs.FileDescriptionImpl = (*masterFileDescription)(nil) + +// Release implements vfs.FileDescriptionImpl.Release. +func (mfd *masterFileDescription) Release() { + mfd.inode.root.masterClose(mfd.t) + mfd.inode.DecRef() +} + +// EventRegister implements waiter.Waitable.EventRegister. +func (mfd *masterFileDescription) EventRegister(e *waiter.Entry, mask waiter.EventMask) { + mfd.t.ld.masterWaiter.EventRegister(e, mask) +} + +// EventUnregister implements waiter.Waitable.EventUnregister. +func (mfd *masterFileDescription) EventUnregister(e *waiter.Entry) { + mfd.t.ld.masterWaiter.EventUnregister(e) +} + +// Readiness implements waiter.Waitable.Readiness. +func (mfd *masterFileDescription) Readiness(mask waiter.EventMask) waiter.EventMask { + return mfd.t.ld.masterReadiness() +} + +// Read implements vfs.FileDescriptionImpl.Read. +func (mfd *masterFileDescription) Read(ctx context.Context, dst usermem.IOSequence, _ vfs.ReadOptions) (int64, error) { + return mfd.t.ld.outputQueueRead(ctx, dst) +} + +// Write implements vfs.FileDescriptionImpl.Write. +func (mfd *masterFileDescription) Write(ctx context.Context, src usermem.IOSequence, _ vfs.WriteOptions) (int64, error) { + return mfd.t.ld.inputQueueWrite(ctx, src) +} + +// Ioctl implements vfs.FileDescriptionImpl.Ioctl. +func (mfd *masterFileDescription) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { + switch cmd := args[1].Uint(); cmd { + case linux.FIONREAD: // linux.FIONREAD == linux.TIOCINQ + // Get the number of bytes in the output queue read buffer. + return 0, mfd.t.ld.outputQueueReadSize(ctx, io, args) + case linux.TCGETS: + // N.B. TCGETS on the master actually returns the configuration + // of the slave end. + return mfd.t.ld.getTermios(ctx, io, args) + case linux.TCSETS: + // N.B. TCSETS on the master actually affects the configuration + // of the slave end. + return mfd.t.ld.setTermios(ctx, io, args) + case linux.TCSETSW: + // TODO(b/29356795): This should drain the output queue first. + return mfd.t.ld.setTermios(ctx, io, args) + case linux.TIOCGPTN: + _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), uint32(mfd.t.n), usermem.IOOpts{ + AddressSpaceActive: true, + }) + return 0, err + case linux.TIOCSPTLCK: + // TODO(b/29356795): Implement pty locking. For now just pretend we do. + return 0, nil + case linux.TIOCGWINSZ: + return 0, mfd.t.ld.windowSize(ctx, io, args) + case linux.TIOCSWINSZ: + return 0, mfd.t.ld.setWindowSize(ctx, io, args) + case linux.TIOCSCTTY: + // Make the given terminal the controlling terminal of the + // calling process. + return 0, mfd.t.setControllingTTY(ctx, io, args, true /* isMaster */) + case linux.TIOCNOTTY: + // Release this process's controlling terminal. + return 0, mfd.t.releaseControllingTTY(ctx, io, args, true /* isMaster */) + case linux.TIOCGPGRP: + // Get the foreground process group. + return mfd.t.foregroundProcessGroup(ctx, io, args, true /* isMaster */) + case linux.TIOCSPGRP: + // Set the foreground process group. + return mfd.t.setForegroundProcessGroup(ctx, io, args, true /* isMaster */) + default: + maybeEmitUnimplementedEvent(ctx, cmd) + return 0, syserror.ENOTTY + } +} + +// SetStat implements vfs.FileDescriptionImpl.SetStat. +func (mfd *masterFileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { + creds := auth.CredentialsFromContext(ctx) + fs := mfd.vfsfd.VirtualDentry().Mount().Filesystem() + return mfd.inode.SetStat(ctx, fs, creds, opts) +} + +// Stat implements vfs.FileDescriptionImpl.Stat. +func (mfd *masterFileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { + fs := mfd.vfsfd.VirtualDentry().Mount().Filesystem() + return mfd.inode.Stat(fs, opts) +} + +// maybeEmitUnimplementedEvent emits unimplemented event if cmd is valid. +func maybeEmitUnimplementedEvent(ctx context.Context, cmd uint32) { + switch cmd { + case linux.TCGETS, + linux.TCSETS, + linux.TCSETSW, + linux.TCSETSF, + linux.TIOCGWINSZ, + linux.TIOCSWINSZ, + linux.TIOCSETD, + linux.TIOCSBRK, + linux.TIOCCBRK, + linux.TCSBRK, + linux.TCSBRKP, + linux.TIOCSTI, + linux.TIOCCONS, + linux.FIONBIO, + linux.TIOCEXCL, + linux.TIOCNXCL, + linux.TIOCGEXCL, + linux.TIOCGSID, + linux.TIOCGETD, + linux.TIOCVHANGUP, + linux.TIOCGDEV, + linux.TIOCMGET, + linux.TIOCMSET, + linux.TIOCMBIC, + linux.TIOCMBIS, + linux.TIOCGICOUNT, + linux.TCFLSH, + linux.TIOCSSERIAL, + linux.TIOCGPTPEER: + + unimpl.EmitUnimplementedEvent(ctx) + } +} + +// LINT.ThenChange(../../fs/tty/master.go) diff --git a/pkg/sentry/fsimpl/devpts/queue.go b/pkg/sentry/fsimpl/devpts/queue.go new file mode 100644 index 000000000..29a6be858 --- /dev/null +++ b/pkg/sentry/fsimpl/devpts/queue.go @@ -0,0 +1,240 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package devpts + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/safemem" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +// LINT.IfChange + +// waitBufMaxBytes is the maximum size of a wait buffer. It is based on +// TTYB_DEFAULT_MEM_LIMIT. +const waitBufMaxBytes = 131072 + +// queue represents one of the input or output queues between a pty master and +// slave. Bytes written to a queue are added to the read buffer until it is +// full, at which point they are written to the wait buffer. Bytes are +// processed (i.e. undergo termios transformations) as they are added to the +// read buffer. The read buffer is readable when its length is nonzero and +// readable is true. +// +// +stateify savable +type queue struct { + // mu protects everything in queue. + mu sync.Mutex `state:"nosave"` + + // readBuf is buffer of data ready to be read when readable is true. + // This data has been processed. + readBuf []byte + + // waitBuf contains data that can't fit into readBuf. It is put here + // until it can be loaded into the read buffer. waitBuf contains data + // that hasn't been processed. + waitBuf [][]byte + waitBufLen uint64 + + // readable indicates whether the read buffer can be read from. In + // canonical mode, there can be an unterminated line in the read buffer, + // so readable must be checked. + readable bool + + // transform is the the queue's function for transforming bytes + // entering the queue. For example, transform might convert all '\r's + // entering the queue to '\n's. + transformer +} + +// readReadiness returns whether q is ready to be read from. +func (q *queue) readReadiness(t *linux.KernelTermios) waiter.EventMask { + q.mu.Lock() + defer q.mu.Unlock() + if len(q.readBuf) > 0 && q.readable { + return waiter.EventIn + } + return waiter.EventMask(0) +} + +// writeReadiness returns whether q is ready to be written to. +func (q *queue) writeReadiness(t *linux.KernelTermios) waiter.EventMask { + q.mu.Lock() + defer q.mu.Unlock() + if q.waitBufLen < waitBufMaxBytes { + return waiter.EventOut + } + return waiter.EventMask(0) +} + +// readableSize writes the number of readable bytes to userspace. +func (q *queue) readableSize(ctx context.Context, io usermem.IO, args arch.SyscallArguments) error { + q.mu.Lock() + defer q.mu.Unlock() + var size int32 + if q.readable { + size = int32(len(q.readBuf)) + } + + _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), size, usermem.IOOpts{ + AddressSpaceActive: true, + }) + return err + +} + +// read reads from q to userspace. It returns the number of bytes read as well +// as whether the read caused more readable data to become available (whether +// data was pushed from the wait buffer to the read buffer). +// +// Preconditions: +// * l.termiosMu must be held for reading. +func (q *queue) read(ctx context.Context, dst usermem.IOSequence, l *lineDiscipline) (int64, bool, error) { + q.mu.Lock() + defer q.mu.Unlock() + + if !q.readable { + return 0, false, syserror.ErrWouldBlock + } + + if dst.NumBytes() > canonMaxBytes { + dst = dst.TakeFirst(canonMaxBytes) + } + + n, err := dst.CopyOutFrom(ctx, safemem.ReaderFunc(func(dst safemem.BlockSeq) (uint64, error) { + src := safemem.BlockSeqOf(safemem.BlockFromSafeSlice(q.readBuf)) + n, err := safemem.CopySeq(dst, src) + if err != nil { + return 0, err + } + q.readBuf = q.readBuf[n:] + + // If we read everything, this queue is no longer readable. + if len(q.readBuf) == 0 { + q.readable = false + } + + return n, nil + })) + if err != nil { + return 0, false, err + } + + // Move data from the queue's wait buffer to its read buffer. + nPushed := q.pushWaitBufLocked(l) + + return int64(n), nPushed > 0, nil +} + +// write writes to q from userspace. +// +// Preconditions: +// * l.termiosMu must be held for reading. +func (q *queue) write(ctx context.Context, src usermem.IOSequence, l *lineDiscipline) (int64, error) { + q.mu.Lock() + defer q.mu.Unlock() + + // Copy data into the wait buffer. + n, err := src.CopyInTo(ctx, safemem.WriterFunc(func(src safemem.BlockSeq) (uint64, error) { + copyLen := src.NumBytes() + room := waitBufMaxBytes - q.waitBufLen + // If out of room, return EAGAIN. + if room == 0 && copyLen > 0 { + return 0, syserror.ErrWouldBlock + } + // Cap the size of the wait buffer. + if copyLen > room { + copyLen = room + src = src.TakeFirst64(room) + } + buf := make([]byte, copyLen) + + // Copy the data into the wait buffer. + dst := safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf)) + n, err := safemem.CopySeq(dst, src) + if err != nil { + return 0, err + } + q.waitBufAppend(buf) + + return n, nil + })) + if err != nil { + return 0, err + } + + // Push data from the wait to the read buffer. + q.pushWaitBufLocked(l) + + return n, nil +} + +// writeBytes writes to q from b. +// +// Preconditions: +// * l.termiosMu must be held for reading. +func (q *queue) writeBytes(b []byte, l *lineDiscipline) { + q.mu.Lock() + defer q.mu.Unlock() + + // Write to the wait buffer. + q.waitBufAppend(b) + q.pushWaitBufLocked(l) +} + +// pushWaitBufLocked fills the queue's read buffer with data from the wait +// buffer. +// +// Preconditions: +// * l.termiosMu must be held for reading. +// * q.mu must be locked. +func (q *queue) pushWaitBufLocked(l *lineDiscipline) int { + if q.waitBufLen == 0 { + return 0 + } + + // Move data from the wait to the read buffer. + var total int + var i int + for i = 0; i < len(q.waitBuf); i++ { + n := q.transform(l, q, q.waitBuf[i]) + total += n + if n != len(q.waitBuf[i]) { + // The read buffer filled up without consuming the + // entire buffer. + q.waitBuf[i] = q.waitBuf[i][n:] + break + } + } + + // Update wait buffer based on consumed data. + q.waitBuf = q.waitBuf[i:] + q.waitBufLen -= uint64(total) + + return total +} + +// Precondition: q.mu must be locked. +func (q *queue) waitBufAppend(b []byte) { + q.waitBuf = append(q.waitBuf, b) + q.waitBufLen += uint64(len(b)) +} + +// LINT.ThenChange(../../fs/tty/queue.go) diff --git a/pkg/sentry/fsimpl/devpts/slave.go b/pkg/sentry/fsimpl/devpts/slave.go new file mode 100644 index 000000000..e7e50d51e --- /dev/null +++ b/pkg/sentry/fsimpl/devpts/slave.go @@ -0,0 +1,186 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package devpts + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +// LINT.IfChange + +// slaveInode is the inode for the slave end of the Terminal. +type slaveInode struct { + kernfs.InodeAttrs + kernfs.InodeNoopRefCount + kernfs.InodeNotDirectory + kernfs.InodeNotSymlink + + // Keep a reference to this inode's dentry. + dentry kernfs.Dentry + + // root is the devpts root inode. + root *rootInode + + // t is the connected Terminal. + t *Terminal +} + +var _ kernfs.Inode = (*slaveInode)(nil) + +// Open implements kernfs.Inode.Open. +func (si *slaveInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + si.IncRef() + fd := &slaveFileDescription{ + inode: si, + } + if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}); err != nil { + si.DecRef() + return nil, err + } + return &fd.vfsfd, nil + +} + +// Valid implements kernfs.Inode.Valid. +func (si *slaveInode) Valid(context.Context) bool { + // Return valid if the slave still exists. + si.root.mu.Lock() + defer si.root.mu.Unlock() + _, ok := si.root.slaves[si.t.n] + return ok +} + +// Stat implements kernfs.Inode.Stat. +func (si *slaveInode) Stat(vfsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { + statx, err := si.InodeAttrs.Stat(vfsfs, opts) + if err != nil { + return linux.Statx{}, err + } + statx.Blksize = 1024 + statx.RdevMajor = linux.UNIX98_PTY_SLAVE_MAJOR + statx.RdevMinor = si.t.n + return statx, nil +} + +// SetStat implements kernfs.Inode.SetStat +func (si *slaveInode) SetStat(ctx context.Context, vfsfs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { + if opts.Stat.Mask&linux.STATX_SIZE != 0 { + return syserror.EINVAL + } + return si.InodeAttrs.SetStat(ctx, vfsfs, creds, opts) +} + +type slaveFileDescription struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl + + inode *slaveInode +} + +var _ vfs.FileDescriptionImpl = (*slaveFileDescription)(nil) + +// Release implements fs.FileOperations.Release. +func (sfd *slaveFileDescription) Release() { + sfd.inode.DecRef() +} + +// EventRegister implements waiter.Waitable.EventRegister. +func (sfd *slaveFileDescription) EventRegister(e *waiter.Entry, mask waiter.EventMask) { + sfd.inode.t.ld.slaveWaiter.EventRegister(e, mask) +} + +// EventUnregister implements waiter.Waitable.EventUnregister. +func (sfd *slaveFileDescription) EventUnregister(e *waiter.Entry) { + sfd.inode.t.ld.slaveWaiter.EventUnregister(e) +} + +// Readiness implements waiter.Waitable.Readiness. +func (sfd *slaveFileDescription) Readiness(mask waiter.EventMask) waiter.EventMask { + return sfd.inode.t.ld.slaveReadiness() +} + +// Read implements vfs.FileDescriptionImpl.Read. +func (sfd *slaveFileDescription) Read(ctx context.Context, dst usermem.IOSequence, _ vfs.ReadOptions) (int64, error) { + return sfd.inode.t.ld.inputQueueRead(ctx, dst) +} + +// Write implements vfs.FileDescriptionImpl.Write. +func (sfd *slaveFileDescription) Write(ctx context.Context, src usermem.IOSequence, _ vfs.WriteOptions) (int64, error) { + return sfd.inode.t.ld.outputQueueWrite(ctx, src) +} + +// Ioctl implements vfs.FileDescripionImpl.Ioctl. +func (sfd *slaveFileDescription) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { + switch cmd := args[1].Uint(); cmd { + case linux.FIONREAD: // linux.FIONREAD == linux.TIOCINQ + // Get the number of bytes in the input queue read buffer. + return 0, sfd.inode.t.ld.inputQueueReadSize(ctx, io, args) + case linux.TCGETS: + return sfd.inode.t.ld.getTermios(ctx, io, args) + case linux.TCSETS: + return sfd.inode.t.ld.setTermios(ctx, io, args) + case linux.TCSETSW: + // TODO(b/29356795): This should drain the output queue first. + return sfd.inode.t.ld.setTermios(ctx, io, args) + case linux.TIOCGPTN: + _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), uint32(sfd.inode.t.n), usermem.IOOpts{ + AddressSpaceActive: true, + }) + return 0, err + case linux.TIOCGWINSZ: + return 0, sfd.inode.t.ld.windowSize(ctx, io, args) + case linux.TIOCSWINSZ: + return 0, sfd.inode.t.ld.setWindowSize(ctx, io, args) + case linux.TIOCSCTTY: + // Make the given terminal the controlling terminal of the + // calling process. + return 0, sfd.inode.t.setControllingTTY(ctx, io, args, false /* isMaster */) + case linux.TIOCNOTTY: + // Release this process's controlling terminal. + return 0, sfd.inode.t.releaseControllingTTY(ctx, io, args, false /* isMaster */) + case linux.TIOCGPGRP: + // Get the foreground process group. + return sfd.inode.t.foregroundProcessGroup(ctx, io, args, false /* isMaster */) + case linux.TIOCSPGRP: + // Set the foreground process group. + return sfd.inode.t.setForegroundProcessGroup(ctx, io, args, false /* isMaster */) + default: + maybeEmitUnimplementedEvent(ctx, cmd) + return 0, syserror.ENOTTY + } +} + +// SetStat implements vfs.FileDescriptionImpl.SetStat. +func (sfd *slaveFileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { + creds := auth.CredentialsFromContext(ctx) + fs := sfd.vfsfd.VirtualDentry().Mount().Filesystem() + return sfd.inode.SetStat(ctx, fs, creds, opts) +} + +// Stat implements vfs.FileDescriptionImpl.Stat. +func (sfd *slaveFileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { + fs := sfd.vfsfd.VirtualDentry().Mount().Filesystem() + return sfd.inode.Stat(fs, opts) +} + +// LINT.ThenChange(../../fs/tty/slave.go) diff --git a/pkg/sentry/fsimpl/devpts/terminal.go b/pkg/sentry/fsimpl/devpts/terminal.go new file mode 100644 index 000000000..b44e673d8 --- /dev/null +++ b/pkg/sentry/fsimpl/devpts/terminal.go @@ -0,0 +1,124 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package devpts + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/usermem" +) + +// LINT.IfChanges + +// Terminal is a pseudoterminal. +// +// +stateify savable +type Terminal struct { + // n is the terminal index. It is immutable. + n uint32 + + // ld is the line discipline of the terminal. It is immutable. + ld *lineDiscipline + + // masterKTTY contains the controlling process of the master end of + // this terminal. This field is immutable. + masterKTTY *kernel.TTY + + // slaveKTTY contains the controlling process of the slave end of this + // terminal. This field is immutable. + slaveKTTY *kernel.TTY +} + +func newTerminal(n uint32) *Terminal { + termios := linux.DefaultSlaveTermios + t := Terminal{ + n: n, + ld: newLineDiscipline(termios), + masterKTTY: &kernel.TTY{Index: n}, + slaveKTTY: &kernel.TTY{Index: n}, + } + return &t +} + +// setControllingTTY makes tm the controlling terminal of the calling thread +// group. +func (tm *Terminal) setControllingTTY(ctx context.Context, io usermem.IO, args arch.SyscallArguments, isMaster bool) error { + task := kernel.TaskFromContext(ctx) + if task == nil { + panic("setControllingTTY must be called from a task context") + } + + return task.ThreadGroup().SetControllingTTY(tm.tty(isMaster), args[2].Int()) +} + +// releaseControllingTTY removes tm as the controlling terminal of the calling +// thread group. +func (tm *Terminal) releaseControllingTTY(ctx context.Context, io usermem.IO, args arch.SyscallArguments, isMaster bool) error { + task := kernel.TaskFromContext(ctx) + if task == nil { + panic("releaseControllingTTY must be called from a task context") + } + + return task.ThreadGroup().ReleaseControllingTTY(tm.tty(isMaster)) +} + +// foregroundProcessGroup gets the process group ID of tm's foreground process. +func (tm *Terminal) foregroundProcessGroup(ctx context.Context, io usermem.IO, args arch.SyscallArguments, isMaster bool) (uintptr, error) { + task := kernel.TaskFromContext(ctx) + if task == nil { + panic("foregroundProcessGroup must be called from a task context") + } + + ret, err := task.ThreadGroup().ForegroundProcessGroup(tm.tty(isMaster)) + if err != nil { + return 0, err + } + + // Write it out to *arg. + _, err = usermem.CopyObjectOut(ctx, io, args[2].Pointer(), int32(ret), usermem.IOOpts{ + AddressSpaceActive: true, + }) + return 0, err +} + +// foregroundProcessGroup sets tm's foreground process. +func (tm *Terminal) setForegroundProcessGroup(ctx context.Context, io usermem.IO, args arch.SyscallArguments, isMaster bool) (uintptr, error) { + task := kernel.TaskFromContext(ctx) + if task == nil { + panic("setForegroundProcessGroup must be called from a task context") + } + + // Read in the process group ID. + var pgid int32 + if _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &pgid, usermem.IOOpts{ + AddressSpaceActive: true, + }); err != nil { + return 0, err + } + + ret, err := task.ThreadGroup().SetForegroundProcessGroup(tm.tty(isMaster), kernel.ProcessGroupID(pgid)) + return uintptr(ret), err +} + +func (tm *Terminal) tty(isMaster bool) *kernel.TTY { + if isMaster { + return tm.masterKTTY + } + return tm.slaveKTTY +} + +// LINT.ThenChange(../../fs/tty/terminal.go) diff --git a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go index 64f1b142c..142ee53b0 100644 --- a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go +++ b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go @@ -163,16 +163,25 @@ func (a *Accessor) CreateDeviceFile(ctx context.Context, pathname string, kind v func (a *Accessor) UserspaceInit(ctx context.Context) error { actx := a.wrapContext(ctx) - // systemd: src/shared/dev-setup.c:dev_setup() + // Initialize symlinks. for _, symlink := range []struct { source string target string }{ - // /proc/kcore is not implemented. + // systemd: src/shared/dev-setup.c:dev_setup() {source: "fd", target: "/proc/self/fd"}, {source: "stdin", target: "/proc/self/fd/0"}, {source: "stdout", target: "/proc/self/fd/1"}, {source: "stderr", target: "/proc/self/fd/2"}, + // /proc/kcore is not implemented. + + // Linux implements /dev/ptmx as a device node, but advises + // container implementations to create /dev/ptmx as a symlink + // to pts/ptmx (Documentation/filesystems/devpts.txt). Systemd + // follows this advice (src/nspawn/nspawn.c:setup_pts()), while + // LXC tries to create a bind mount and falls back to a symlink + // (src/lxc/conf.c:lxc_setup_devpts()). + {source: "ptmx", target: "pts/ptmx"}, } { if err := a.vfsObj.SymlinkAt(actx, a.creds, a.pathOperationAt(symlink.source), symlink.target); err != nil { return fmt.Errorf("failed to create symlink %q => %q: %v", symlink.source, symlink.target, err) diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 3164d022c..1d46dba25 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -391,7 +391,7 @@ func (fs *Filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf // O_NOFOLLOW have no effect here (they're handled by VFS by setting // appropriate bits in rp), but are returned by // FileDescriptionImpl.StatusFlags(). - opts.Flags &= linux.O_ACCMODE | linux.O_CREAT | linux.O_EXCL | linux.O_TRUNC | linux.O_DIRECTORY | linux.O_NOFOLLOW + opts.Flags &= linux.O_ACCMODE | linux.O_CREAT | linux.O_EXCL | linux.O_TRUNC | linux.O_DIRECTORY | linux.O_NOFOLLOW | linux.O_NONBLOCK ats := vfs.AccessTypesForOpenFlags(&opts) // Do not create new file. diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 9f526359e..a946645f6 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -216,6 +216,11 @@ func (a *InodeAttrs) Init(creds *auth.Credentials, ino uint64, mode linux.FileMo atomic.StoreUint32(&a.nlink, nlink) } +// Ino returns the inode id. +func (a *InodeAttrs) Ino() uint64 { + return atomic.LoadUint64(&a.ino) +} + // Mode implements Inode.Mode. func (a *InodeAttrs) Mode() linux.FileMode { return linux.FileMode(atomic.LoadUint32(&a.mode)) @@ -359,8 +364,8 @@ func (o *OrderedChildren) Destroy() { // cache. Populate returns the number of directories inserted, which the caller // may use to update the link count for the parent directory. // -// Precondition: d.Impl() must be a kernfs Dentry. d must represent a directory -// inode. children must not contain any conflicting entries already in o. +// Precondition: d must represent a directory inode. children must not contain +// any conflicting entries already in o. func (o *OrderedChildren) Populate(d *Dentry, children map[string]*Dentry) uint32 { var links uint32 for name, child := range children { -- cgit v1.2.3 From f01f2132d8d3e551579cba9a1b942b4b70d83f21 Mon Sep 17 00:00:00 2001 From: Rahat Mahmood Date: Thu, 23 Apr 2020 18:18:54 -0700 Subject: Enable automated marshalling for mempolicy syscalls. PiperOrigin-RevId: 308170679 --- pkg/abi/linux/mm.go | 17 +++++++++++------ pkg/sentry/kernel/task.go | 2 +- pkg/sentry/kernel/task_sched.go | 4 ++-- pkg/sentry/mm/mm.go | 3 ++- pkg/sentry/mm/syscalls.go | 4 ++-- pkg/sentry/syscalls/linux/sys_mempolicy.go | 18 +++++++++--------- 6 files changed, 27 insertions(+), 21 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/mm.go b/pkg/abi/linux/mm.go index cd043dac3..07cc1895e 100644 --- a/pkg/abi/linux/mm.go +++ b/pkg/abi/linux/mm.go @@ -90,14 +90,19 @@ const ( MS_SYNC = 1 << 2 ) +// NumaPolicy is the NUMA memory policy for a memory range. See numa(7). +// +// +marshal +type NumaPolicy int32 + // Policies for get_mempolicy(2)/set_mempolicy(2). const ( - MPOL_DEFAULT = 0 - MPOL_PREFERRED = 1 - MPOL_BIND = 2 - MPOL_INTERLEAVE = 3 - MPOL_LOCAL = 4 - MPOL_MAX = 5 + MPOL_DEFAULT NumaPolicy = 0 + MPOL_PREFERRED NumaPolicy = 1 + MPOL_BIND NumaPolicy = 2 + MPOL_INTERLEAVE NumaPolicy = 3 + MPOL_LOCAL NumaPolicy = 4 + MPOL_MAX NumaPolicy = 5 ) // Flags for get_mempolicy(2). diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index e5d133d6c..f48247c94 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -484,7 +484,7 @@ type Task struct { // bit. // // numaPolicy and numaNodeMask are protected by mu. - numaPolicy int32 + numaPolicy linux.NumaPolicy numaNodeMask uint64 // netns is the task's network namespace. netns is never nil. diff --git a/pkg/sentry/kernel/task_sched.go b/pkg/sentry/kernel/task_sched.go index 8b148db35..09366b60c 100644 --- a/pkg/sentry/kernel/task_sched.go +++ b/pkg/sentry/kernel/task_sched.go @@ -653,14 +653,14 @@ func (t *Task) SetNiceness(n int) { } // NumaPolicy returns t's current numa policy. -func (t *Task) NumaPolicy() (policy int32, nodeMask uint64) { +func (t *Task) NumaPolicy() (policy linux.NumaPolicy, nodeMask uint64) { t.mu.Lock() defer t.mu.Unlock() return t.numaPolicy, t.numaNodeMask } // SetNumaPolicy sets t's numa policy. -func (t *Task) SetNumaPolicy(policy int32, nodeMask uint64) { +func (t *Task) SetNumaPolicy(policy linux.NumaPolicy, nodeMask uint64) { t.mu.Lock() defer t.mu.Unlock() t.numaPolicy = policy diff --git a/pkg/sentry/mm/mm.go b/pkg/sentry/mm/mm.go index 34d3bde7a..6db7c3d40 100644 --- a/pkg/sentry/mm/mm.go +++ b/pkg/sentry/mm/mm.go @@ -35,6 +35,7 @@ package mm import ( + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fsbridge" @@ -286,7 +287,7 @@ type vma struct { mlockMode memmap.MLockMode // numaPolicy is the NUMA policy for this vma set by mbind(). - numaPolicy int32 + numaPolicy linux.NumaPolicy // numaNodemask is the NUMA nodemask for this vma set by mbind(). numaNodemask uint64 diff --git a/pkg/sentry/mm/syscalls.go b/pkg/sentry/mm/syscalls.go index c5dfa5972..3f496aa9f 100644 --- a/pkg/sentry/mm/syscalls.go +++ b/pkg/sentry/mm/syscalls.go @@ -974,7 +974,7 @@ func (mm *MemoryManager) MLockAll(ctx context.Context, opts MLockAllOpts) error } // NumaPolicy implements the semantics of Linux's get_mempolicy(MPOL_F_ADDR). -func (mm *MemoryManager) NumaPolicy(addr usermem.Addr) (int32, uint64, error) { +func (mm *MemoryManager) NumaPolicy(addr usermem.Addr) (linux.NumaPolicy, uint64, error) { mm.mappingMu.RLock() defer mm.mappingMu.RUnlock() vseg := mm.vmas.FindSegment(addr) @@ -986,7 +986,7 @@ func (mm *MemoryManager) NumaPolicy(addr usermem.Addr) (int32, uint64, error) { } // SetNumaPolicy implements the semantics of Linux's mbind(). -func (mm *MemoryManager) SetNumaPolicy(addr usermem.Addr, length uint64, policy int32, nodemask uint64) error { +func (mm *MemoryManager) SetNumaPolicy(addr usermem.Addr, length uint64, policy linux.NumaPolicy, nodemask uint64) error { if !addr.IsPageAligned() { return syserror.EINVAL } diff --git a/pkg/sentry/syscalls/linux/sys_mempolicy.go b/pkg/sentry/syscalls/linux/sys_mempolicy.go index ac934dc6f..9b4a5c3f1 100644 --- a/pkg/sentry/syscalls/linux/sys_mempolicy.go +++ b/pkg/sentry/syscalls/linux/sys_mempolicy.go @@ -162,10 +162,10 @@ func GetMempolicy(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. if err != nil { return 0, nil, err } - policy = 0 // maxNodes == 1 + policy = linux.MPOL_DEFAULT // maxNodes == 1 } if mode != 0 { - if _, err := t.CopyOut(mode, policy); err != nil { + if _, err := policy.CopyOut(t, mode); err != nil { return 0, nil, err } } @@ -199,10 +199,10 @@ func GetMempolicy(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. if policy&^linux.MPOL_MODE_FLAGS != linux.MPOL_INTERLEAVE { return 0, nil, syserror.EINVAL } - policy = 0 // maxNodes == 1 + policy = linux.MPOL_DEFAULT // maxNodes == 1 } if mode != 0 { - if _, err := t.CopyOut(mode, policy); err != nil { + if _, err := policy.CopyOut(t, mode); err != nil { return 0, nil, err } } @@ -216,7 +216,7 @@ func GetMempolicy(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. // SetMempolicy implements the syscall set_mempolicy(2). func SetMempolicy(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - modeWithFlags := args[0].Int() + modeWithFlags := linux.NumaPolicy(args[0].Int()) nodemask := args[1].Pointer() maxnode := args[2].Uint() @@ -233,7 +233,7 @@ func SetMempolicy(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. func Mbind(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { addr := args[0].Pointer() length := args[1].Uint64() - mode := args[2].Int() + mode := linux.NumaPolicy(args[2].Int()) nodemask := args[3].Pointer() maxnode := args[4].Uint() flags := args[5].Uint() @@ -258,9 +258,9 @@ func Mbind(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return 0, nil, err } -func copyInMempolicyNodemask(t *kernel.Task, modeWithFlags int32, nodemask usermem.Addr, maxnode uint32) (int32, uint64, error) { - flags := modeWithFlags & linux.MPOL_MODE_FLAGS - mode := modeWithFlags &^ linux.MPOL_MODE_FLAGS +func copyInMempolicyNodemask(t *kernel.Task, modeWithFlags linux.NumaPolicy, nodemask usermem.Addr, maxnode uint32) (linux.NumaPolicy, uint64, error) { + flags := linux.NumaPolicy(modeWithFlags & linux.MPOL_MODE_FLAGS) + mode := linux.NumaPolicy(modeWithFlags &^ linux.MPOL_MODE_FLAGS) if flags == linux.MPOL_MODE_FLAGS { // Can't specify both mode flags simultaneously. return 0, 0, syserror.EINVAL -- cgit v1.2.3 From 40a712c57cd78c51c9875ae04b5e795113c75e62 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 24 Apr 2020 08:19:11 -0700 Subject: Refactor syscall.Fstat calls in hostfs. Just call syscall.Fstat directly each time mode/file owner are needed. This feels more natural than using i.getPermissions(). PiperOrigin-RevId: 308257405 --- pkg/sentry/fsimpl/host/host.go | 35 +++++++++++++---------------------- 1 file changed, 13 insertions(+), 22 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index ae94cfa6e..7847e3cc2 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -169,31 +169,22 @@ func fileFlagsFromHostFD(fd int) (int, error) { // CheckPermissions implements kernfs.Inode. func (i *inode) CheckPermissions(ctx context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error { - mode, uid, gid, err := i.getPermissions() - if err != nil { + var s syscall.Stat_t + if err := syscall.Fstat(i.hostFD, &s); err != nil { return err } - return vfs.GenericCheckPermissions(creds, ats, mode, uid, gid) + return vfs.GenericCheckPermissions(creds, ats, linux.FileMode(s.Mode), auth.KUID(s.Uid), auth.KGID(s.Gid)) } // Mode implements kernfs.Inode. func (i *inode) Mode() linux.FileMode { - mode, _, _, err := i.getPermissions() - // Retrieving the mode from the host fd using fstat(2) should not fail. - // If the syscall does not succeed, something is fundamentally wrong. - if err != nil { - panic(fmt.Sprintf("failed to retrieve mode from host fd %d: %v", i.hostFD, err)) - } - return linux.FileMode(mode) -} - -func (i *inode) getPermissions() (linux.FileMode, auth.KUID, auth.KGID, error) { - // Retrieve metadata. var s syscall.Stat_t if err := syscall.Fstat(i.hostFD, &s); err != nil { - return 0, 0, 0, err + // Retrieving the mode from the host fd using fstat(2) should not fail. + // If the syscall does not succeed, something is fundamentally wrong. + panic(fmt.Sprintf("failed to retrieve mode from host fd %d: %v", i.hostFD, err)) } - return linux.FileMode(s.Mode), auth.KUID(s.Uid), auth.KGID(s.Gid), nil + return linux.FileMode(s.Mode) } // Stat implements kernfs.Inode. @@ -326,11 +317,11 @@ func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Cre if m&^(linux.STATX_MODE|linux.STATX_SIZE|linux.STATX_ATIME|linux.STATX_MTIME) != 0 { return syserror.EPERM } - mode, uid, gid, err := i.getPermissions() - if err != nil { + var hostStat syscall.Stat_t + if err := syscall.Fstat(i.hostFD, &hostStat); err != nil { return err } - if err := vfs.CheckSetStat(ctx, creds, &s, mode.Permissions(), uid, gid); err != nil { + if err := vfs.CheckSetStat(ctx, creds, &s, linux.FileMode(hostStat.Mode&linux.PermissionsMask), auth.KUID(hostStat.Uid), auth.KGID(hostStat.Gid)); err != nil { return err } @@ -374,11 +365,11 @@ func (i *inode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptio } func (i *inode) open(d *vfs.Dentry, mnt *vfs.Mount) (*vfs.FileDescription, error) { - mode, _, _, err := i.getPermissions() - if err != nil { + var s syscall.Stat_t + if err := syscall.Fstat(i.hostFD, &s); err != nil { return nil, err } - fileType := mode.FileType() + fileType := s.Mode & linux.FileTypeMask if fileType == syscall.S_IFSOCK { if i.isTTY { return nil, errors.New("cannot use host socket as TTY") -- cgit v1.2.3 From 1b88c63b3e6b330c8399bf92f148cc80374bee18 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 24 Apr 2020 10:02:22 -0700 Subject: Move hostfs mount to Kernel struct. This is needed to set up host fds passed through a Unix socket. Note that the host package depends on kernel, so we cannot set up the hostfs mount directly in Kernel.Init as we do for sockfs and pipefs. Also, adjust sockfs to make its setup look more like hostfs's and pipefs's. PiperOrigin-RevId: 308274053 --- pkg/sentry/fsimpl/host/host.go | 16 +++++++-------- pkg/sentry/fsimpl/sockfs/sockfs.go | 26 ++++++++++------------- pkg/sentry/kernel/kernel.go | 42 ++++++++++++++++++++++++++++---------- runsc/boot/fds.go | 7 +------ runsc/boot/loader.go | 13 ++++++++++++ 5 files changed, 64 insertions(+), 40 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 7847e3cc2..a26b13067 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -42,7 +42,7 @@ type filesystemType struct{} // GetFilesystem implements FilesystemType.GetFilesystem. func (filesystemType) GetFilesystem(context.Context, *vfs.VirtualFilesystem, *auth.Credentials, string, vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { - panic("cannot instaniate a host filesystem") + panic("host.filesystemType.GetFilesystem should never be called") } // Name implements FilesystemType.Name. @@ -55,14 +55,14 @@ type filesystem struct { kernfs.Filesystem } -// NewMount returns a new disconnected mount in vfsObj that may be passed to ImportFD. -func NewMount(vfsObj *vfs.VirtualFilesystem) (*vfs.Mount, error) { +// NewFilesystem sets up and returns a new hostfs filesystem. +// +// Note that there should only ever be one instance of host.filesystem, +// a global mount for host fds. +func NewFilesystem(vfsObj *vfs.VirtualFilesystem) *vfs.Filesystem { fs := &filesystem{} - fs.Init(vfsObj, &filesystemType{}) - vfsfs := fs.VFSFilesystem() - // NewDisconnectedMount will take an additional reference on vfsfs. - defer vfsfs.DecRef() - return vfsObj.NewDisconnectedMount(vfsfs, nil, &vfs.MountOptions{}) + fs.Init(vfsObj, filesystemType{}) + return fs.VFSFilesystem() } // ImportFD sets up and returns a vfs.FileDescription from a donated fd. diff --git a/pkg/sentry/fsimpl/sockfs/sockfs.go b/pkg/sentry/fsimpl/sockfs/sockfs.go index 3f7ad1d65..632cfde88 100644 --- a/pkg/sentry/fsimpl/sockfs/sockfs.go +++ b/pkg/sentry/fsimpl/sockfs/sockfs.go @@ -24,26 +24,12 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) -// NewFilesystem creates a new sockfs filesystem. -// -// Note that there should only ever be one instance of sockfs.Filesystem, -// backing a global socket mount. -func NewFilesystem(vfsObj *vfs.VirtualFilesystem) *vfs.Filesystem { - fs, _, err := filesystemType{}.GetFilesystem(nil, vfsObj, nil, "", vfs.GetFilesystemOptions{}) - if err != nil { - panic("failed to create sockfs filesystem") - } - return fs -} - // filesystemType implements vfs.FilesystemType. type filesystemType struct{} // GetFilesystem implements FilesystemType.GetFilesystem. func (fsType filesystemType) GetFilesystem(_ context.Context, vfsObj *vfs.VirtualFilesystem, _ *auth.Credentials, _ string, _ vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { - fs := &filesystem{} - fs.Init(vfsObj, fsType) - return fs.VFSFilesystem(), nil, nil + panic("sockfs.filesystemType.GetFilesystem should never be called") } // Name implements FilesystemType.Name. @@ -60,6 +46,16 @@ type filesystem struct { kernfs.Filesystem } +// NewFilesystem sets up and returns a new sockfs filesystem. +// +// Note that there should only ever be one instance of sockfs.Filesystem, +// backing a global socket mount. +func NewFilesystem(vfsObj *vfs.VirtualFilesystem) *vfs.Filesystem { + fs := &filesystem{} + fs.Init(vfsObj, filesystemType{}) + return fs.VFSFilesystem() +} + // inode implements kernfs.Inode. // // TODO(gvisor.dev/issue/1476): Add device numbers to this inode (which are diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index fef60e636..c91b9dce2 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -227,11 +227,6 @@ type Kernel struct { // by extMu. nextSocketEntry uint64 - // socketMount is a disconnected vfs.Mount, not included in k.vfs, - // representing a sockfs.filesystem. socketMount is used to back - // VirtualDentries representing anonymous sockets. - socketMount *vfs.Mount - // deviceRegistry is used to save/restore device.SimpleDevices. deviceRegistry struct{} `state:".(*device.Registry)"` @@ -255,10 +250,22 @@ type Kernel struct { // VFS keeps the filesystem state used across the kernel. vfs vfs.VirtualFilesystem + // hostMount is the Mount used for file descriptors that were imported + // from the host. + hostMount *vfs.Mount + // pipeMount is the Mount used for pipes created by the pipe() and pipe2() // syscalls (as opposed to named pipes created by mknod()). pipeMount *vfs.Mount + // socketMount is the Mount used for sockets created by the socket() and + // socketpair() syscalls. There are several cases where a socket dentry will + // not be contained in socketMount: + // 1. Socket files created by mknod() + // 2. Socket fds imported from the host (Kernel.hostMount is used for these) + // 3. Socket files created by binding Unix sockets to a file path + socketMount *vfs.Mount + // If set to true, report address space activation waits as if the task is in // external wait so that the watchdog doesn't report the task stuck. SleepForAddressSpaceActivation bool @@ -377,7 +384,7 @@ func (k *Kernel) Init(args InitKernelArgs) error { defer socketFilesystem.DecRef() socketMount, err := k.vfs.NewDisconnectedMount(socketFilesystem, nil, &vfs.MountOptions{}) if err != nil { - return fmt.Errorf("failed to initialize socket mount: %v", err) + return fmt.Errorf("failed to create sockfs mount: %v", err) } k.socketMount = socketMount } @@ -1526,11 +1533,6 @@ func (k *Kernel) ListSockets() []*SocketEntry { return socks } -// SocketMount returns the global socket mount. -func (k *Kernel) SocketMount() *vfs.Mount { - return k.socketMount -} - // supervisorContext is a privileged context. type supervisorContext struct { context.NoopSleeper @@ -1629,7 +1631,25 @@ func (k *Kernel) VFS() *vfs.VirtualFilesystem { return &k.vfs } +// SetHostMount sets the hostfs mount. +func (k *Kernel) SetHostMount(mnt *vfs.Mount) { + if k.hostMount != nil { + panic("Kernel.hostMount cannot be set more than once") + } + k.hostMount = mnt +} + +// HostMount returns the hostfs mount. +func (k *Kernel) HostMount() *vfs.Mount { + return k.hostMount +} + // PipeMount returns the pipefs mount. func (k *Kernel) PipeMount() *vfs.Mount { return k.pipeMount } + +// SocketMount returns the sockfs mount. +func (k *Kernel) SocketMount() *vfs.Mount { + return k.socketMount +} diff --git a/runsc/boot/fds.go b/runsc/boot/fds.go index 7e49f6f9f..0cbd63857 100644 --- a/runsc/boot/fds.go +++ b/runsc/boot/fds.go @@ -89,14 +89,9 @@ func createFDTableVFS2(ctx context.Context, console bool, stdioFDs []int) (*kern fdTable := k.NewFDTable() defer fdTable.DecRef() - hostMount, err := vfshost.NewMount(k.VFS()) - if err != nil { - return nil, fmt.Errorf("creating host mount: %w", err) - } - for appFD, hostFD := range stdioFDs { // TODO(gvisor.dev/issue/1482): Add TTY support. - appFile, err := vfshost.ImportFD(hostMount, hostFD, false) + appFile, err := vfshost.ImportFD(k.HostMount(), hostFD, false) if err != nil { return nil, err } diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index 096b0e9f0..3f41d8357 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -36,6 +36,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/host" "gvisor.dev/gvisor/pkg/sentry/fs/user" + vfs2host "gvisor.dev/gvisor/pkg/sentry/fsimpl/host" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" @@ -46,6 +47,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/syscalls/linux/vfs2" "gvisor.dev/gvisor/pkg/sentry/time" "gvisor.dev/gvisor/pkg/sentry/usage" + "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sentry/watchdog" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" @@ -329,6 +331,17 @@ func New(args Args) (*Loader, error) { return nil, fmt.Errorf("creating pod mount hints: %v", err) } + if kernel.VFS2Enabled { + // Set up host mount that will be used for imported fds. + hostFilesystem := vfs2host.NewFilesystem(k.VFS()) + defer hostFilesystem.DecRef() + hostMount, err := k.VFS().NewDisconnectedMount(hostFilesystem, nil, &vfs.MountOptions{}) + if err != nil { + return nil, fmt.Errorf("failed to create hostfs mount: %v", err) + } + k.SetHostMount(hostMount) + } + // Make host FDs stable between invocations. Host FDs must map to the exact // same number when the sandbox is restored. Otherwise the wrong FD will be // used. -- cgit v1.2.3 From 2cc0fd42f462f3942230c4b33ca2825e2a28765d Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Fri, 24 Apr 2020 11:43:49 -0700 Subject: Fixes for procfs - Return ENOENT for /proc/[pid]/task if task is zoombied or terminated - Allow directory to be Seek() to the end - Construct synthetic files for /proc/[pid]/ns/* - Changed GenericDirectoryFD.Init to not register with FileDescription, otherwise other implementation cannot change behavior. Updates #1195,1193 PiperOrigin-RevId: 308294649 --- pkg/sentry/fsimpl/devpts/devpts.go | 6 ++- pkg/sentry/fsimpl/kernfs/fd_impl_util.go | 28 +++++++++-- pkg/sentry/fsimpl/kernfs/filesystem.go | 4 +- pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 8 +-- pkg/sentry/fsimpl/kernfs/kernfs.go | 2 +- pkg/sentry/fsimpl/kernfs/kernfs_test.go | 10 ++-- pkg/sentry/fsimpl/kernfs/symlink.go | 2 +- pkg/sentry/fsimpl/proc/subtasks.go | 49 ++++++++++++++++++- pkg/sentry/fsimpl/proc/task.go | 7 ++- pkg/sentry/fsimpl/proc/task_fds.go | 14 ++++-- pkg/sentry/fsimpl/proc/task_files.go | 76 +++++++++++++++++++++++++++-- pkg/sentry/fsimpl/proc/tasks.go | 6 ++- pkg/sentry/fsimpl/proc/tasks_files.go | 4 +- pkg/sentry/fsimpl/sys/sys.go | 4 +- pkg/sentry/vfs/vfs.go | 8 +++ 15 files changed, 192 insertions(+), 36 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/devpts/devpts.go b/pkg/sentry/fsimpl/devpts/devpts.go index 07a69b940..f36bf50fc 100644 --- a/pkg/sentry/fsimpl/devpts/devpts.go +++ b/pkg/sentry/fsimpl/devpts/devpts.go @@ -161,8 +161,10 @@ func (i *rootInode) masterClose(t *Terminal) { // Open implements kernfs.Inode.Open. func (i *rootInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, &opts) + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &opts) + if err != nil { + return nil, err + } return fd.VFSFileDescription(), nil } diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index e8a4670b8..dd5806301 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -15,6 +15,8 @@ package kernfs import ( + "math" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" @@ -43,15 +45,27 @@ type GenericDirectoryFD struct { off int64 } -// Init initializes a GenericDirectoryFD. -func (fd *GenericDirectoryFD) Init(m *vfs.Mount, d *vfs.Dentry, children *OrderedChildren, opts *vfs.OpenOptions) error { +// NewGenericDirectoryFD creates a new GenericDirectoryFD and returns its +// dentry. +func NewGenericDirectoryFD(m *vfs.Mount, d *vfs.Dentry, children *OrderedChildren, opts *vfs.OpenOptions) (*GenericDirectoryFD, error) { + fd := &GenericDirectoryFD{} + if err := fd.Init(children, opts); err != nil { + return nil, err + } + if err := fd.vfsfd.Init(fd, opts.Flags, m, d, &vfs.FileDescriptionOptions{}); err != nil { + return nil, err + } + return fd, nil +} + +// Init initializes a GenericDirectoryFD. Use it when overriding +// GenericDirectoryFD. Caller must call fd.VFSFileDescription.Init() with the +// correct implementation. +func (fd *GenericDirectoryFD) Init(children *OrderedChildren, opts *vfs.OpenOptions) error { if vfs.AccessTypesForOpenFlags(opts)&vfs.MayWrite != 0 { // Can't open directories for writing. return syserror.EISDIR } - if err := fd.vfsfd.Init(fd, opts.Flags, m, d, &vfs.FileDescriptionOptions{}); err != nil { - return err - } fd.children = children return nil } @@ -187,6 +201,10 @@ func (fd *GenericDirectoryFD) Seek(ctx context.Context, offset int64, whence int // Use offset as given. case linux.SEEK_CUR: offset += fd.off + case linux.SEEK_END: + // TODO(gvisor.dev/issue/1193): This can prevent new files from showing up + // if they are added after SEEK_END. + offset = math.MaxInt64 default: return 0, syserror.EINVAL } diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 1d46dba25..3ccd92fc5 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -82,7 +82,7 @@ afterSymlink: } // Resolve any symlink at current path component. if rp.ShouldFollowSymlink() && next.isSymlink() { - targetVD, targetPathname, err := next.inode.Getlink(ctx) + targetVD, targetPathname, err := next.inode.Getlink(ctx, rp.Mount()) if err != nil { return nil, err } @@ -477,7 +477,7 @@ afterTrailingSymlink: } child := childVFSD.Impl().(*Dentry) if rp.ShouldFollowSymlink() && child.isSymlink() { - targetVD, targetPathname, err := child.inode.Getlink(ctx) + targetVD, targetPathname, err := child.inode.Getlink(ctx, rp.Mount()) if err != nil { return nil, err } diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index a946645f6..02f35a675 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -182,7 +182,7 @@ func (InodeNotSymlink) Readlink(context.Context) (string, error) { } // Getlink implements Inode.Getlink. -func (InodeNotSymlink) Getlink(context.Context) (vfs.VirtualDentry, string, error) { +func (InodeNotSymlink) Getlink(context.Context, *vfs.Mount) (vfs.VirtualDentry, string, error) { return vfs.VirtualDentry{}, "", syserror.EINVAL } @@ -568,8 +568,10 @@ func (s *StaticDirectory) Init(creds *auth.Credentials, ino uint64, perm linux.F // Open implements kernfs.Inode. func (s *StaticDirectory) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &s.OrderedChildren, &opts) + fd, err := NewGenericDirectoryFD(rp.Mount(), vfsd, &s.OrderedChildren, &opts) + if err != nil { + return nil, err + } return fd.VFSFileDescription(), nil } diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index f5041824f..95cf6dc24 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -439,5 +439,5 @@ type inodeSymlink interface { // // - If the inode is not a symlink, Getlink returns (zero-value // VirtualDentry, "", EINVAL). - Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) + Getlink(ctx context.Context, mnt *vfs.Mount) (vfs.VirtualDentry, string, error) } diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index 465451f35..0964d5456 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -117,8 +117,8 @@ func (fs *filesystem) newReadonlyDir(creds *auth.Credentials, mode linux.FileMod } func (d *readonlyDir) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &kernfs.GenericDirectoryFD{} - if err := fd.Init(rp.Mount(), vfsd, &d.OrderedChildren, &opts); err != nil { + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &d.OrderedChildren, &opts) + if err != nil { return nil, err } return fd.VFSFileDescription(), nil @@ -147,8 +147,10 @@ func (fs *filesystem) newDir(creds *auth.Credentials, mode linux.FileMode, conte } func (d *dir) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &d.OrderedChildren, &opts) + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &d.OrderedChildren, &opts) + if err != nil { + return nil, err + } return fd.VFSFileDescription(), nil } diff --git a/pkg/sentry/fsimpl/kernfs/symlink.go b/pkg/sentry/fsimpl/kernfs/symlink.go index 018aa503c..0aa6dc979 100644 --- a/pkg/sentry/fsimpl/kernfs/symlink.go +++ b/pkg/sentry/fsimpl/kernfs/symlink.go @@ -56,7 +56,7 @@ func (s *StaticSymlink) Readlink(_ context.Context) (string, error) { } // Getlink implements Inode.Getlink. -func (s *StaticSymlink) Getlink(_ context.Context) (vfs.VirtualDentry, string, error) { +func (s *StaticSymlink) Getlink(context.Context, *vfs.Mount) (vfs.VirtualDentry, string, error) { return vfs.VirtualDentry{}, s.target, nil } diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go index a21313666..28ec2484a 100644 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -88,6 +88,9 @@ func (i *subtasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallb if len(tasks) == 0 { return offset, syserror.ENOENT } + if relOffset >= int64(len(tasks)) { + return offset, nil + } tids := make([]int, 0, len(tasks)) for _, tid := range tasks { @@ -110,10 +113,52 @@ func (i *subtasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallb return offset, nil } +type subtasksFD struct { + kernfs.GenericDirectoryFD + + task *kernel.Task +} + +func (fd *subtasksFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback) error { + if fd.task.ExitState() >= kernel.TaskExitZombie { + return syserror.ENOENT + } + return fd.GenericDirectoryFD.IterDirents(ctx, cb) +} + +// Seek implements vfs.FileDecriptionImpl.Seek. +func (fd *subtasksFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + if fd.task.ExitState() >= kernel.TaskExitZombie { + return 0, syserror.ENOENT + } + return fd.GenericDirectoryFD.Seek(ctx, offset, whence) +} + +// Stat implements vfs.FileDescriptionImpl.Stat. +func (fd *subtasksFD) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { + if fd.task.ExitState() >= kernel.TaskExitZombie { + return linux.Statx{}, syserror.ENOENT + } + return fd.GenericDirectoryFD.Stat(ctx, opts) +} + +// SetStat implements vfs.FileDescriptionImpl.SetStat. +func (fd *subtasksFD) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { + if fd.task.ExitState() >= kernel.TaskExitZombie { + return syserror.ENOENT + } + return fd.GenericDirectoryFD.SetStat(ctx, opts) +} + // Open implements kernfs.Inode. func (i *subtasksInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, &opts) + fd := &subtasksFD{task: i.task} + if err := fd.Init(&i.OrderedChildren, &opts); err != nil { + return nil, err + } + if err := fd.VFSFileDescription().Init(fd, opts.Flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}); err != nil { + return nil, err + } return fd.VFSFileDescription(), nil } diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index 888afc0fd..e2790d35b 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -44,6 +44,7 @@ type taskInode struct { var _ kernfs.Inode = (*taskInode)(nil) func newTaskInode(inoGen InoGenerator, task *kernel.Task, pidns *kernel.PIDNamespace, isThreadGroup bool, cgroupControllers map[string]string) *kernfs.Dentry { + // TODO(gvisor.dev/issue/164): Fail with ESRCH if task exited. contents := map[string]*kernfs.Dentry{ "auxv": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &auxvData{task: task}), "cmdline": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &cmdlineData{task: task, arg: cmdlineDataArg}), @@ -102,8 +103,10 @@ func (i *taskInode) Valid(ctx context.Context) bool { // Open implements kernfs.Inode. func (i *taskInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, &opts) + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &opts) + if err != nil { + return nil, err + } return fd.VFSFileDescription(), nil } diff --git a/pkg/sentry/fsimpl/proc/task_fds.go b/pkg/sentry/fsimpl/proc/task_fds.go index 046265eca..a7622f1b6 100644 --- a/pkg/sentry/fsimpl/proc/task_fds.go +++ b/pkg/sentry/fsimpl/proc/task_fds.go @@ -143,8 +143,10 @@ func (i *fdDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, erro // Open implements kernfs.Inode. func (i *fdDirInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, &opts) + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &opts) + if err != nil { + return nil, err + } return fd.VFSFileDescription(), nil } @@ -207,7 +209,7 @@ func (s *fdSymlink) Readlink(ctx context.Context) (string, error) { return s.task.Kernel().VFS().PathnameWithDeleted(ctx, root, file.VirtualDentry()) } -func (s *fdSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) { +func (s *fdSymlink) Getlink(ctx context.Context, mnt *vfs.Mount) (vfs.VirtualDentry, string, error) { file, _ := getTaskFD(s.task, s.fd) if file == nil { return vfs.VirtualDentry{}, "", syserror.ENOENT @@ -268,8 +270,10 @@ func (i *fdInfoDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, // Open implements kernfs.Inode. func (i *fdInfoDirInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, &opts) + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &opts) + if err != nil { + return nil, err + } return fd.VFSFileDescription(), nil } diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index f3173e197..410cc3552 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -622,7 +622,7 @@ func (s *exeSymlink) Readlink(ctx context.Context) (string, error) { } // Getlink implements kernfs.Inode.Getlink. -func (s *exeSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) { +func (s *exeSymlink) Getlink(ctx context.Context, _ *vfs.Mount) (vfs.VirtualDentry, string, error) { if !kernel.ContextCanTrace(ctx, s.task, false) { return vfs.VirtualDentry{}, "", syserror.EACCES } @@ -754,9 +754,79 @@ func (s *namespaceSymlink) Readlink(ctx context.Context) (string, error) { } // Getlink implements Inode.Getlink. -func (s *namespaceSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) { +func (s *namespaceSymlink) Getlink(ctx context.Context, mnt *vfs.Mount) (vfs.VirtualDentry, string, error) { if err := checkTaskState(s.task); err != nil { return vfs.VirtualDentry{}, "", err } - return s.StaticSymlink.Getlink(ctx) + + // Create a synthetic inode to represent the namespace. + dentry := &kernfs.Dentry{} + dentry.Init(&namespaceInode{}) + vd := vfs.MakeVirtualDentry(mnt, dentry.VFSDentry()) + vd.IncRef() + dentry.DecRef() + return vd, "", nil +} + +// namespaceInode is a synthetic inode created to represent a namespace in +// /proc/[pid]/ns/*. +type namespaceInode struct { + kernfs.InodeAttrs + kernfs.InodeNoopRefCount + kernfs.InodeNotDirectory + kernfs.InodeNotSymlink +} + +var _ kernfs.Inode = (*namespaceInode)(nil) + +// Init initializes a namespace inode. +func (i *namespaceInode) Init(creds *auth.Credentials, ino uint64, perm linux.FileMode) { + if perm&^linux.PermissionsMask != 0 { + panic(fmt.Sprintf("Only permission mask must be set: %x", perm&linux.PermissionsMask)) + } + i.InodeAttrs.Init(creds, ino, linux.ModeRegular|perm) +} + +// Open implements Inode.Open. +func (i *namespaceInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd := &namespaceFD{inode: i} + i.IncRef() + if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}); err != nil { + return nil, err + } + return &fd.vfsfd, nil +} + +// namespace FD is a synthetic file that represents a namespace in +// /proc/[pid]/ns/*. +type namespaceFD struct { + vfs.FileDescriptionDefaultImpl + + vfsfd vfs.FileDescription + inode *namespaceInode +} + +var _ vfs.FileDescriptionImpl = (*namespaceFD)(nil) + +// Stat implements FileDescriptionImpl. +func (fd *namespaceFD) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { + vfs := fd.vfsfd.VirtualDentry().Mount().Filesystem() + return fd.inode.Stat(vfs, opts) +} + +// SetStat implements FileDescriptionImpl. +func (fd *namespaceFD) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { + vfs := fd.vfsfd.VirtualDentry().Mount().Filesystem() + creds := auth.CredentialsFromContext(ctx) + return fd.inode.SetStat(ctx, vfs, creds, opts) +} + +// Release implements FileDescriptionImpl. +func (fd *namespaceFD) Release() { + fd.inode.DecRef() +} + +// OnClose implements FileDescriptionImpl. +func (*namespaceFD) OnClose(context.Context) error { + return nil } diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index 9f2ef8200..26518ed03 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -202,8 +202,10 @@ func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback // Open implements kernfs.Inode. func (i *tasksInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, &opts) + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &opts) + if err != nil { + return nil, err + } return fd.VFSFileDescription(), nil } diff --git a/pkg/sentry/fsimpl/proc/tasks_files.go b/pkg/sentry/fsimpl/proc/tasks_files.go index 4621e2de0..92007df81 100644 --- a/pkg/sentry/fsimpl/proc/tasks_files.go +++ b/pkg/sentry/fsimpl/proc/tasks_files.go @@ -63,7 +63,7 @@ func (s *selfSymlink) Readlink(ctx context.Context) (string, error) { return strconv.FormatUint(uint64(tgid), 10), nil } -func (s *selfSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) { +func (s *selfSymlink) Getlink(ctx context.Context, _ *vfs.Mount) (vfs.VirtualDentry, string, error) { target, err := s.Readlink(ctx) return vfs.VirtualDentry{}, target, err } @@ -106,7 +106,7 @@ func (s *threadSelfSymlink) Readlink(ctx context.Context) (string, error) { return fmt.Sprintf("%d/task/%d", tgid, tid), nil } -func (s *threadSelfSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) { +func (s *threadSelfSymlink) Getlink(ctx context.Context, _ *vfs.Mount) (vfs.VirtualDentry, string, error) { target, err := s.Readlink(ctx) return vfs.VirtualDentry{}, target, err } diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go index 5c617270e..34e8e0cbe 100644 --- a/pkg/sentry/fsimpl/sys/sys.go +++ b/pkg/sentry/fsimpl/sys/sys.go @@ -106,8 +106,8 @@ func (*dir) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.Set // Open implements kernfs.Inode.Open. func (d *dir) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &kernfs.GenericDirectoryFD{} - if err := fd.Init(rp.Mount(), vfsd, &d.OrderedChildren, &opts); err != nil { + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &d.OrderedChildren, &opts) + if err != nil { return nil, err } return fd.VFSFileDescription(), nil diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index cb5bbd781..9015f2cc1 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -797,6 +797,14 @@ type VirtualDentry struct { dentry *Dentry } +// MakeVirtualDentry creates a VirtualDentry. +func MakeVirtualDentry(mount *Mount, dentry *Dentry) VirtualDentry { + return VirtualDentry{ + mount: mount, + dentry: dentry, + } +} + // Ok returns true if vd is not empty. It does not require that a reference is // held. func (vd VirtualDentry) Ok() bool { -- cgit v1.2.3 From 632b104aff3fedf7798447eedc5662c973525c66 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 24 Apr 2020 12:36:14 -0700 Subject: Plumb context.Context into kernfs.Inode.Open(). PiperOrigin-RevId: 308304793 --- pkg/sentry/fsimpl/devpts/devpts.go | 2 +- pkg/sentry/fsimpl/devpts/master.go | 2 +- pkg/sentry/fsimpl/devpts/slave.go | 2 +- pkg/sentry/fsimpl/host/host.go | 10 +++++----- pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go | 2 +- pkg/sentry/fsimpl/kernfs/filesystem.go | 8 ++++---- pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 4 ++-- pkg/sentry/fsimpl/kernfs/kernfs.go | 2 +- pkg/sentry/fsimpl/kernfs/kernfs_test.go | 4 ++-- pkg/sentry/fsimpl/pipefs/pipefs.go | 5 ++--- pkg/sentry/fsimpl/proc/subtasks.go | 2 +- pkg/sentry/fsimpl/proc/task.go | 2 +- pkg/sentry/fsimpl/proc/task_fds.go | 4 ++-- pkg/sentry/fsimpl/proc/task_files.go | 2 +- pkg/sentry/fsimpl/proc/tasks.go | 2 +- pkg/sentry/fsimpl/sockfs/sockfs.go | 2 +- pkg/sentry/fsimpl/sys/sys.go | 2 +- runsc/boot/fds.go | 2 +- 18 files changed, 29 insertions(+), 30 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/devpts/devpts.go b/pkg/sentry/fsimpl/devpts/devpts.go index f36bf50fc..181d765d3 100644 --- a/pkg/sentry/fsimpl/devpts/devpts.go +++ b/pkg/sentry/fsimpl/devpts/devpts.go @@ -160,7 +160,7 @@ func (i *rootInode) masterClose(t *Terminal) { } // Open implements kernfs.Inode.Open. -func (i *rootInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (i *rootInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &opts) if err != nil { return nil, err diff --git a/pkg/sentry/fsimpl/devpts/master.go b/pkg/sentry/fsimpl/devpts/master.go index 60340c28e..04a292927 100644 --- a/pkg/sentry/fsimpl/devpts/master.go +++ b/pkg/sentry/fsimpl/devpts/master.go @@ -46,7 +46,7 @@ type masterInode struct { var _ kernfs.Inode = (*masterInode)(nil) // Open implements kernfs.Inode.Open. -func (mi *masterInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (mi *masterInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { t, err := mi.root.allocateTerminal(rp.Credentials()) if err != nil { return nil, err diff --git a/pkg/sentry/fsimpl/devpts/slave.go b/pkg/sentry/fsimpl/devpts/slave.go index e7e50d51e..0a98dc896 100644 --- a/pkg/sentry/fsimpl/devpts/slave.go +++ b/pkg/sentry/fsimpl/devpts/slave.go @@ -48,7 +48,7 @@ type slaveInode struct { var _ kernfs.Inode = (*slaveInode)(nil) // Open implements kernfs.Inode.Open. -func (si *slaveInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (si *slaveInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { si.IncRef() fd := &slaveFileDescription{ inode: si, diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index a26b13067..1e53b5c1b 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -66,7 +66,7 @@ func NewFilesystem(vfsObj *vfs.VirtualFilesystem) *vfs.Filesystem { } // ImportFD sets up and returns a vfs.FileDescription from a donated fd. -func ImportFD(mnt *vfs.Mount, hostFD int, isTTY bool) (*vfs.FileDescription, error) { +func ImportFD(ctx context.Context, mnt *vfs.Mount, hostFD int, isTTY bool) (*vfs.FileDescription, error) { fs, ok := mnt.Filesystem().Impl().(*kernfs.Filesystem) if !ok { return nil, fmt.Errorf("can't import host FDs into filesystems of type %T", mnt.Filesystem().Impl()) @@ -108,7 +108,7 @@ func ImportFD(mnt *vfs.Mount, hostFD int, isTTY bool) (*vfs.FileDescription, err // i.open will take a reference on d. defer d.DecRef() - return i.open(d.VFSDentry(), mnt) + return i.open(ctx, d.VFSDentry(), mnt) } // inode implements kernfs.Inode. @@ -360,11 +360,11 @@ func (i *inode) Destroy() { } // Open implements kernfs.Inode. -func (i *inode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - return i.open(vfsd, rp.Mount()) +func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + return i.open(ctx, vfsd, rp.Mount()) } -func (i *inode) open(d *vfs.Dentry, mnt *vfs.Mount) (*vfs.FileDescription, error) { +func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount) (*vfs.FileDescription, error) { var s syscall.Stat_t if err := syscall.Fstat(i.hostFD, &s); err != nil { return nil, err diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go index d8bddbafa..c7779fc11 100644 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -53,7 +53,7 @@ func (f *DynamicBytesFile) Init(creds *auth.Credentials, ino uint64, data vfs.Dy } // Open implements Inode.Open. -func (f *DynamicBytesFile) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (f *DynamicBytesFile) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd := &DynamicBytesFD{} if err := fd.Init(rp.Mount(), vfsd, f.data, opts.Flags); err != nil { return nil, err diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 3ccd92fc5..9e8d80414 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -406,7 +406,7 @@ func (fs *Filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf if err := inode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { return nil, err } - return inode.Open(rp, vfsd, opts) + return inode.Open(ctx, rp, vfsd, opts) } // May create new file. @@ -425,7 +425,7 @@ func (fs *Filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf if err := inode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { return nil, err } - return inode.Open(rp, vfsd, opts) + return inode.Open(ctx, rp, vfsd, opts) } afterTrailingSymlink: parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) @@ -466,7 +466,7 @@ afterTrailingSymlink: } child := childVFSD.Impl().(*Dentry) parentVFSD.Impl().(*Dentry).InsertChild(pc, child) - return child.inode.Open(rp, childVFSD, opts) + return child.inode.Open(ctx, rp, childVFSD, opts) } if err != nil { return nil, err @@ -499,7 +499,7 @@ afterTrailingSymlink: if err := child.inode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { return nil, err } - return child.inode.Open(rp, &child.vfsd, opts) + return child.inode.Open(ctx, rp, &child.vfsd, opts) } // ReadlinkAt implements vfs.FilesystemImpl.ReadlinkAt. diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 02f35a675..615592d5f 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -525,7 +525,7 @@ type InodeSymlink struct { } // Open implements Inode.Open. -func (InodeSymlink) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (InodeSymlink) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { return nil, syserror.ELOOP } @@ -567,7 +567,7 @@ func (s *StaticDirectory) Init(creds *auth.Credentials, ino uint64, perm linux.F } // Open implements kernfs.Inode. -func (s *StaticDirectory) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (s *StaticDirectory) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd, err := NewGenericDirectoryFD(rp.Mount(), vfsd, &s.OrderedChildren, &opts) if err != nil { return nil, err diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index 95cf6dc24..732837933 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -308,7 +308,7 @@ type Inode interface { // // Precondition: rp.Done(). vfsd.Impl() must be the kernfs Dentry containing // the inode on which Open() is being called. - Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) + Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) } type inodeRefs interface { diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index 0964d5456..a9f671bc8 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -116,7 +116,7 @@ func (fs *filesystem) newReadonlyDir(creds *auth.Credentials, mode linux.FileMod return &dir.dentry } -func (d *readonlyDir) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (d *readonlyDir) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &d.OrderedChildren, &opts) if err != nil { return nil, err @@ -146,7 +146,7 @@ func (fs *filesystem) newDir(creds *auth.Credentials, mode linux.FileMode, conte return &dir.dentry } -func (d *dir) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (d *dir) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &d.OrderedChildren, &opts) if err != nil { return nil, err diff --git a/pkg/sentry/fsimpl/pipefs/pipefs.go b/pkg/sentry/fsimpl/pipefs/pipefs.go index faf3179bc..d6bd67467 100644 --- a/pkg/sentry/fsimpl/pipefs/pipefs.go +++ b/pkg/sentry/fsimpl/pipefs/pipefs.go @@ -129,9 +129,8 @@ func (i *inode) SetStat(ctx context.Context, vfsfs *vfs.Filesystem, creds *auth. } // Open implements kernfs.Inode.Open. -func (i *inode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - // FIXME(b/38173783): kernfs does not plumb Context here. - return i.pipe.Open(context.Background(), rp.Mount(), vfsd, opts.Flags) +func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + return i.pipe.Open(ctx, rp.Mount(), vfsd, opts.Flags) } // NewConnectedPipeFDs returns a pair of FileDescriptions representing the read diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go index 28ec2484a..a5cfa8333 100644 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -151,7 +151,7 @@ func (fd *subtasksFD) SetStat(ctx context.Context, opts vfs.SetStatOptions) erro } // Open implements kernfs.Inode. -func (i *subtasksInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (i *subtasksInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd := &subtasksFD{task: i.task} if err := fd.Init(&i.OrderedChildren, &opts); err != nil { return nil, err diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index e2790d35b..66419d91b 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -102,7 +102,7 @@ func (i *taskInode) Valid(ctx context.Context) bool { } // Open implements kernfs.Inode. -func (i *taskInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (i *taskInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &opts) if err != nil { return nil, err diff --git a/pkg/sentry/fsimpl/proc/task_fds.go b/pkg/sentry/fsimpl/proc/task_fds.go index a7622f1b6..8ad976073 100644 --- a/pkg/sentry/fsimpl/proc/task_fds.go +++ b/pkg/sentry/fsimpl/proc/task_fds.go @@ -142,7 +142,7 @@ func (i *fdDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, erro } // Open implements kernfs.Inode. -func (i *fdDirInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (i *fdDirInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &opts) if err != nil { return nil, err @@ -269,7 +269,7 @@ func (i *fdInfoDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, } // Open implements kernfs.Inode. -func (i *fdInfoDirInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (i *fdInfoDirInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &opts) if err != nil { return nil, err diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index 410cc3552..515f25327 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -788,7 +788,7 @@ func (i *namespaceInode) Init(creds *auth.Credentials, ino uint64, perm linux.Fi } // Open implements Inode.Open. -func (i *namespaceInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (i *namespaceInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd := &namespaceFD{inode: i} i.IncRef() if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}); err != nil { diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index 26518ed03..5aeda8c9b 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -201,7 +201,7 @@ func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback } // Open implements kernfs.Inode. -func (i *tasksInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (i *tasksInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &opts) if err != nil { return nil, err diff --git a/pkg/sentry/fsimpl/sockfs/sockfs.go b/pkg/sentry/fsimpl/sockfs/sockfs.go index 632cfde88..5ce50625b 100644 --- a/pkg/sentry/fsimpl/sockfs/sockfs.go +++ b/pkg/sentry/fsimpl/sockfs/sockfs.go @@ -69,7 +69,7 @@ type inode struct { } // Open implements kernfs.Inode.Open. -func (i *inode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { return nil, syserror.ENXIO } diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go index 34e8e0cbe..f8d25d35e 100644 --- a/pkg/sentry/fsimpl/sys/sys.go +++ b/pkg/sentry/fsimpl/sys/sys.go @@ -105,7 +105,7 @@ func (*dir) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.Set } // Open implements kernfs.Inode.Open. -func (d *dir) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (d *dir) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &d.OrderedChildren, &opts) if err != nil { return nil, err diff --git a/runsc/boot/fds.go b/runsc/boot/fds.go index 0cbd63857..7e7a31fbd 100644 --- a/runsc/boot/fds.go +++ b/runsc/boot/fds.go @@ -91,7 +91,7 @@ func createFDTableVFS2(ctx context.Context, console bool, stdioFDs []int) (*kern for appFD, hostFD := range stdioFDs { // TODO(gvisor.dev/issue/1482): Add TTY support. - appFile, err := vfshost.ImportFD(k.HostMount(), hostFD, false) + appFile, err := vfshost.ImportFD(ctx, k.HostMount(), hostFD, false) if err != nil { return nil, err } -- cgit v1.2.3 From f13f26d17da56d585fd9857a81175bbd0be8ce60 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 24 Apr 2020 13:45:31 -0700 Subject: Port SCM Rights to VFS2. Fixes #1477. PiperOrigin-RevId: 308317511 --- pkg/sentry/fs/host/control.go | 6 +- pkg/sentry/fsimpl/host/BUILD | 3 + pkg/sentry/fsimpl/host/control.go | 96 ++++++++++++++++++++++ pkg/sentry/socket/control/BUILD | 6 +- pkg/sentry/socket/control/control.go | 24 +++++- pkg/sentry/socket/control/control_vfs2.go | 131 ++++++++++++++++++++++++++++++ pkg/sentry/syscalls/linux/vfs2/socket.go | 2 +- 7 files changed, 260 insertions(+), 8 deletions(-) create mode 100644 pkg/sentry/fsimpl/host/control.go create mode 100644 pkg/sentry/socket/control/control_vfs2.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/host/control.go b/pkg/sentry/fs/host/control.go index 52c0504b6..39299b7e4 100644 --- a/pkg/sentry/fs/host/control.go +++ b/pkg/sentry/fs/host/control.go @@ -23,6 +23,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" ) +// LINT.IfChange + type scmRights struct { fds []int } @@ -32,8 +34,6 @@ func newSCMRights(fds []int) control.SCMRights { } // Files implements control.SCMRights.Files. -// -// TODO(gvisor.dev/issue/2017): Port to VFS2. func (c *scmRights) Files(ctx context.Context, max int) (control.RightsFiles, bool) { n := max var trunc bool @@ -93,3 +93,5 @@ func fdsToFiles(ctx context.Context, fds []int) []*fs.File { } return files } + +// LINT.ThenChange(../../fsimpl/host/control.go) diff --git a/pkg/sentry/fsimpl/host/BUILD b/pkg/sentry/fsimpl/host/BUILD index 44dd9f672..2dcb03a73 100644 --- a/pkg/sentry/fsimpl/host/BUILD +++ b/pkg/sentry/fsimpl/host/BUILD @@ -5,6 +5,7 @@ licenses(["notice"]) go_library( name = "host", srcs = [ + "control.go", "host.go", "ioctl_unsafe.go", "tty.go", @@ -23,6 +24,8 @@ go_library( "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/memmap", + "//pkg/sentry/socket/control", + "//pkg/sentry/socket/unix/transport", "//pkg/sentry/unimpl", "//pkg/sentry/vfs", "//pkg/sync", diff --git a/pkg/sentry/fsimpl/host/control.go b/pkg/sentry/fsimpl/host/control.go new file mode 100644 index 000000000..b9082a20f --- /dev/null +++ b/pkg/sentry/fsimpl/host/control.go @@ -0,0 +1,96 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package host + +import ( + "syscall" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/socket/control" + "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" + "gvisor.dev/gvisor/pkg/sentry/vfs" +) + +type scmRights struct { + fds []int +} + +func newSCMRights(fds []int) control.SCMRightsVFS2 { + return &scmRights{fds} +} + +// Files implements control.SCMRights.Files. +func (c *scmRights) Files(ctx context.Context, max int) (control.RightsFilesVFS2, bool) { + n := max + var trunc bool + if l := len(c.fds); n > l { + n = l + } else if n < l { + trunc = true + } + + rf := control.RightsFilesVFS2(fdsToFiles(ctx, c.fds[:n])) + + // Only consume converted FDs (fdsToFiles may convert fewer than n FDs). + c.fds = c.fds[len(rf):] + return rf, trunc +} + +// Clone implements transport.RightsControlMessage.Clone. +func (c *scmRights) Clone() transport.RightsControlMessage { + // Host rights never need to be cloned. + return nil +} + +// Release implements transport.RightsControlMessage.Release. +func (c *scmRights) Release() { + for _, fd := range c.fds { + syscall.Close(fd) + } + c.fds = nil +} + +// If an error is encountered, only files created before the error will be +// returned. This is what Linux does. +func fdsToFiles(ctx context.Context, fds []int) []*vfs.FileDescription { + files := make([]*vfs.FileDescription, 0, len(fds)) + for _, fd := range fds { + // Get flags. We do it here because they may be modified + // by subsequent functions. + fileFlags, _, errno := syscall.Syscall(syscall.SYS_FCNTL, uintptr(fd), syscall.F_GETFL, 0) + if errno != 0 { + ctx.Warningf("Error retrieving host FD flags: %v", error(errno)) + break + } + + // Create the file backed by hostFD. + file, err := ImportFD(ctx, kernel.KernelFromContext(ctx).HostMount(), fd, false /* isTTY */) + if err != nil { + ctx.Warningf("Error creating file from host FD: %v", err) + break + } + + if err := file.SetStatusFlags(ctx, auth.CredentialsFromContext(ctx), uint32(fileFlags&linux.O_NONBLOCK)); err != nil { + ctx.Warningf("Error setting flags on host FD file: %v", err) + break + } + + files = append(files, file) + } + return files +} diff --git a/pkg/sentry/socket/control/BUILD b/pkg/sentry/socket/control/BUILD index 4d42d29cb..ca16d0381 100644 --- a/pkg/sentry/socket/control/BUILD +++ b/pkg/sentry/socket/control/BUILD @@ -4,7 +4,10 @@ package(licenses = ["notice"]) go_library( name = "control", - srcs = ["control.go"], + srcs = [ + "control.go", + "control_vfs2.go", + ], imports = [ "gvisor.dev/gvisor/pkg/sentry/fs", ], @@ -18,6 +21,7 @@ go_library( "//pkg/sentry/kernel/auth", "//pkg/sentry/socket", "//pkg/sentry/socket/unix/transport", + "//pkg/sentry/vfs", "//pkg/syserror", "//pkg/tcpip", "//pkg/usermem", diff --git a/pkg/sentry/socket/control/control.go b/pkg/sentry/socket/control/control.go index 8834a1e1a..8b439a078 100644 --- a/pkg/sentry/socket/control/control.go +++ b/pkg/sentry/socket/control/control.go @@ -41,6 +41,8 @@ type SCMCredentials interface { Credentials(t *kernel.Task) (kernel.ThreadID, auth.UID, auth.GID) } +// LINT.IfChange + // SCMRights represents a SCM_RIGHTS socket control message. type SCMRights interface { transport.RightsControlMessage @@ -142,6 +144,8 @@ func PackRights(t *kernel.Task, rights SCMRights, cloexec bool, buf []byte, flag return putCmsg(buf, flags, linux.SCM_RIGHTS, align, fds) } +// LINT.ThenChange(./control_vfs2.go) + // scmCredentials represents an SCM_CREDENTIALS socket control message. // // +stateify savable @@ -537,11 +541,19 @@ func Parse(t *kernel.Task, socketOrEndpoint interface{}, buf []byte) (socket.Con } if len(fds) > 0 { - rights, err := NewSCMRights(t, fds) - if err != nil { - return socket.ControlMessages{}, err + if kernel.VFS2Enabled { + rights, err := NewSCMRightsVFS2(t, fds) + if err != nil { + return socket.ControlMessages{}, err + } + cmsgs.Unix.Rights = rights + } else { + rights, err := NewSCMRights(t, fds) + if err != nil { + return socket.ControlMessages{}, err + } + cmsgs.Unix.Rights = rights } - cmsgs.Unix.Rights = rights } return cmsgs, nil @@ -566,6 +578,8 @@ func MakeCreds(t *kernel.Task) SCMCredentials { return &scmCredentials{t, tcred.EffectiveKUID, tcred.EffectiveKGID} } +// LINT.IfChange + // New creates default control messages if needed. func New(t *kernel.Task, socketOrEndpoint interface{}, rights SCMRights) transport.ControlMessages { return transport.ControlMessages{ @@ -573,3 +587,5 @@ func New(t *kernel.Task, socketOrEndpoint interface{}, rights SCMRights) transpo Rights: rights, } } + +// LINT.ThenChange(./control_vfs2.go) diff --git a/pkg/sentry/socket/control/control_vfs2.go b/pkg/sentry/socket/control/control_vfs2.go new file mode 100644 index 000000000..fd08179be --- /dev/null +++ b/pkg/sentry/socket/control/control_vfs2.go @@ -0,0 +1,131 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package control + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// SCMRightsVFS2 represents a SCM_RIGHTS socket control message. +type SCMRightsVFS2 interface { + transport.RightsControlMessage + + // Files returns up to max RightsFiles. + // + // Returned files are consumed and ownership is transferred to the caller. + // Subsequent calls to Files will return the next files. + Files(ctx context.Context, max int) (rf RightsFilesVFS2, truncated bool) +} + +// RightsFiles represents a SCM_RIGHTS socket control message. A reference is +// maintained for each vfs.FileDescription and is release either when an FD is created or +// when the Release method is called. +type RightsFilesVFS2 []*vfs.FileDescription + +// NewSCMRightsVFS2 creates a new SCM_RIGHTS socket control message +// representation using local sentry FDs. +func NewSCMRightsVFS2(t *kernel.Task, fds []int32) (SCMRightsVFS2, error) { + files := make(RightsFilesVFS2, 0, len(fds)) + for _, fd := range fds { + file := t.GetFileVFS2(fd) + if file == nil { + files.Release() + return nil, syserror.EBADF + } + files = append(files, file) + } + return &files, nil +} + +// Files implements SCMRights.Files. +func (fs *RightsFilesVFS2) Files(ctx context.Context, max int) (RightsFilesVFS2, bool) { + n := max + var trunc bool + if l := len(*fs); n > l { + n = l + } else if n < l { + trunc = true + } + rf := (*fs)[:n] + *fs = (*fs)[n:] + return rf, trunc +} + +// Clone implements transport.RightsControlMessage.Clone. +func (fs *RightsFilesVFS2) Clone() transport.RightsControlMessage { + nfs := append(RightsFilesVFS2(nil), *fs...) + for _, nf := range nfs { + nf.IncRef() + } + return &nfs +} + +// Release implements transport.RightsControlMessage.Release. +func (fs *RightsFilesVFS2) Release() { + for _, f := range *fs { + f.DecRef() + } + *fs = nil +} + +// rightsFDsVFS2 gets up to the specified maximum number of FDs. +func rightsFDsVFS2(t *kernel.Task, rights SCMRightsVFS2, cloexec bool, max int) ([]int32, bool) { + files, trunc := rights.Files(t, max) + fds := make([]int32, 0, len(files)) + for i := 0; i < max && len(files) > 0; i++ { + fd, err := t.NewFDFromVFS2(0, files[0], kernel.FDFlags{ + CloseOnExec: cloexec, + }) + files[0].DecRef() + files = files[1:] + if err != nil { + t.Warningf("Error inserting FD: %v", err) + // This is what Linux does. + break + } + + fds = append(fds, int32(fd)) + } + return fds, trunc +} + +// PackRightsVFS2 packs as many FDs as will fit into the unused capacity of buf. +func PackRightsVFS2(t *kernel.Task, rights SCMRightsVFS2, cloexec bool, buf []byte, flags int) ([]byte, int) { + maxFDs := (cap(buf) - len(buf) - linux.SizeOfControlMessageHeader) / 4 + // Linux does not return any FDs if none fit. + if maxFDs <= 0 { + flags |= linux.MSG_CTRUNC + return buf, flags + } + fds, trunc := rightsFDsVFS2(t, rights, cloexec, maxFDs) + if trunc { + flags |= linux.MSG_CTRUNC + } + align := t.Arch().Width() + return putCmsg(buf, flags, linux.SCM_RIGHTS, align, fds) +} + +// NewVFS2 creates default control messages if needed. +func NewVFS2(t *kernel.Task, socketOrEndpoint interface{}, rights SCMRightsVFS2) transport.ControlMessages { + return transport.ControlMessages{ + Credentials: makeCreds(t, socketOrEndpoint), + Rights: rights, + } +} diff --git a/pkg/sentry/syscalls/linux/vfs2/socket.go b/pkg/sentry/syscalls/linux/vfs2/socket.go index b1ede32f0..10b668477 100644 --- a/pkg/sentry/syscalls/linux/vfs2/socket.go +++ b/pkg/sentry/syscalls/linux/vfs2/socket.go @@ -804,7 +804,7 @@ func recvSingleMsg(t *kernel.Task, s socket.SocketVFS2, msgPtr usermem.Addr, fla } if cms.Unix.Rights != nil { - controlData, mflags = control.PackRights(t, cms.Unix.Rights.(control.SCMRights), flags&linux.MSG_CMSG_CLOEXEC != 0, controlData, mflags) + controlData, mflags = control.PackRightsVFS2(t, cms.Unix.Rights.(control.SCMRightsVFS2), flags&linux.MSG_CMSG_CLOEXEC != 0, controlData, mflags) } // Copy the address to the caller. -- cgit v1.2.3 From 15a822a1936e295cb6418df7ddf445d8500dfb2e Mon Sep 17 00:00:00 2001 From: Zach Koopmans Date: Fri, 24 Apr 2020 18:22:21 -0700 Subject: VFS2: Get HelloWorld image tests to pass with VFS2 This change includes: - Modifications to loader_test.go to get TestCreateMountNamespace to pass with VFS2. - Changes necessary to get TestHelloWorld in image tests to pass with VFS2. This means runsc can run the hello-world container with docker on VSF2. Note: Containers that use sockets will not run with these changes. See "//test/image/...". Any tests here with sockets currently fail (which is all of them but HelloWorld). PiperOrigin-RevId: 308363072 --- pkg/sentry/fsimpl/gofer/directory.go | 1 + runsc/boot/BUILD | 2 + runsc/boot/loader.go | 13 +-- runsc/boot/loader_test.go | 152 ++++++++++++++++++++++++----------- runsc/boot/vfs.go | 78 ++++++++++++++---- scripts/docker_tests.sh | 3 + 6 files changed, 183 insertions(+), 66 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/directory.go b/pkg/sentry/fsimpl/gofer/directory.go index c67766ab2..55f9ed911 100644 --- a/pkg/sentry/fsimpl/gofer/directory.go +++ b/pkg/sentry/fsimpl/gofer/directory.go @@ -75,6 +75,7 @@ func (d *dentry) createSyntheticDirectoryLocked(name string, mode linux.FileMode handle: handle{ fd: -1, }, + nlink: uint32(2), } d2.pf.dentry = d2 d2.vfsd.Init(d2) diff --git a/runsc/boot/BUILD b/runsc/boot/BUILD index 69dcc74f2..ed3c8f546 100644 --- a/runsc/boot/BUILD +++ b/runsc/boot/BUILD @@ -119,11 +119,13 @@ go_test( library = ":boot", deps = [ "//pkg/control/server", + "//pkg/fspath", "//pkg/log", "//pkg/p9", "//pkg/sentry/contexttest", "//pkg/sentry/fs", "//pkg/sentry/kernel", + "//pkg/sentry/vfs", "//pkg/sync", "//pkg/unet", "//runsc/fsgofer", diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index 3f41d8357..f6ea4c102 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -625,11 +625,14 @@ func (l *Loader) run() error { // l.stdioFDs are derived from dup() in boot.New() and they are now dup()ed again // either in createFDTable() during initial start or in descriptor.initAfterLoad() - // during restore, we can release l.stdioFDs now. - for _, fd := range l.stdioFDs { - err := syscall.Close(fd) - if err != nil { - return fmt.Errorf("close dup()ed stdioFDs: %v", err) + // during restore, we can release l.stdioFDs now. VFS2 takes ownership of the + // passed FDs, so only close for VFS1. + if !kernel.VFS2Enabled { + for _, fd := range l.stdioFDs { + err := syscall.Close(fd) + if err != nil { + return fmt.Errorf("close dup()ed stdioFDs: %v", err) + } } } diff --git a/runsc/boot/loader_test.go b/runsc/boot/loader_test.go index e7c71734f..55d27a632 100644 --- a/runsc/boot/loader_test.go +++ b/runsc/boot/loader_test.go @@ -26,11 +26,13 @@ import ( specs "github.com/opencontainers/runtime-spec/specs-go" "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/control/server" + "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/unet" "gvisor.dev/gvisor/runsc/fsgofer" @@ -107,14 +109,12 @@ func startGofer(root string) (int, func(), error) { return sandboxEnd, cleanup, nil } -func createLoader(vfsEnabled bool) (*Loader, func(), error) { +func createLoader(vfsEnabled bool, spec *specs.Spec) (*Loader, func(), error) { fd, err := server.CreateSocket(ControlSocketAddr(fmt.Sprintf("%010d", rand.Int())[:10])) if err != nil { return nil, nil, err } conf := testConfig() - spec := testSpec() - conf.VFS2 = vfsEnabled sandEnd, cleanup, err := startGofer(spec.Root.Path) @@ -161,7 +161,7 @@ func TestRunVFS2(t *testing.T) { } func doRun(t *testing.T, vfsEnabled bool) { - l, cleanup, err := createLoader(vfsEnabled) + l, cleanup, err := createLoader(vfsEnabled, testSpec()) if err != nil { t.Fatalf("error creating loader: %v", err) } @@ -210,7 +210,7 @@ func TestStartSignalVFS2(t *testing.T) { } func doStartSignal(t *testing.T, vfsEnabled bool) { - l, cleanup, err := createLoader(vfsEnabled) + l, cleanup, err := createLoader(vfsEnabled, testSpec()) if err != nil { t.Fatalf("error creating loader: %v", err) } @@ -258,18 +258,19 @@ func doStartSignal(t *testing.T, vfsEnabled bool) { } -// Test that MountNamespace can be created with various specs. -func TestCreateMountNamespace(t *testing.T) { - testCases := []struct { - name string - // Spec that will be used to create the mount manager. Note - // that we can't mount procfs without a kernel, so each spec - // MUST contain something other than procfs mounted at /proc. - spec specs.Spec - // Paths that are expected to exist in the resulting fs. - expectedPaths []string - }{ - { +type CreateMountTestcase struct { + name string + // Spec that will be used to create the mount manager. Note + // that we can't mount procfs without a kernel, so each spec + // MUST contain something other than procfs mounted at /proc. + spec specs.Spec + // Paths that are expected to exist in the resulting fs. + expectedPaths []string +} + +func createMountTestcases(vfs2 bool) []*CreateMountTestcase { + testCases := []*CreateMountTestcase{ + &CreateMountTestcase{ // Only proc. name: "only proc mount", spec: specs.Spec{ @@ -311,7 +312,7 @@ func TestCreateMountNamespace(t *testing.T) { // /dev, and /sys. expectedPaths: []string{"/some/very/very/deep/path", "/proc", "/dev", "/sys"}, }, - { + &CreateMountTestcase{ // Mounts are nested inside each other. name: "nested mounts", spec: specs.Spec{ @@ -355,7 +356,7 @@ func TestCreateMountNamespace(t *testing.T) { expectedPaths: []string{"/foo", "/foo/bar", "/foo/bar/baz", "/foo/qux", "/foo/qux-quz", "/foo/some/very/very/deep/path", "/proc", "/dev", "/sys"}, }, - { + &CreateMountTestcase{ name: "mount inside /dev", spec: specs.Spec{ Root: &specs.Root{ @@ -398,40 +399,47 @@ func TestCreateMountNamespace(t *testing.T) { }, expectedPaths: []string{"/proc", "/dev", "/dev/fd-foo", "/dev/foo", "/dev/bar", "/sys"}, }, - { - name: "mounts inside mandatory mounts", - spec: specs.Spec{ - Root: &specs.Root{ - Path: os.TempDir(), - Readonly: true, + } + + vfsCase := &CreateMountTestcase{ + name: "mounts inside mandatory mounts", + spec: specs.Spec{ + Root: &specs.Root{ + Path: os.TempDir(), + Readonly: true, + }, + Mounts: []specs.Mount{ + { + Destination: "/proc", + Type: "tmpfs", }, - Mounts: []specs.Mount{ - { - Destination: "/proc", - Type: "tmpfs", - }, - // We don't include /sys, and /tmp in - // the spec, since they will be added - // automatically. - // - // Instead, add submounts inside these - // directories and make sure they are - // visible under the mandatory mounts. - { - Destination: "/sys/bar", - Type: "tmpfs", - }, - { - Destination: "/tmp/baz", - Type: "tmpfs", - }, + // TODO (gvisor.dev/issue/1487): Re-add this case when sysfs supports + // MkDirAt in VFS2 (and remove the reduntant append). + // { + // Destination: "/sys/bar", + // Type: "tmpfs", + // }, + // + { + Destination: "/tmp/baz", + Type: "tmpfs", }, }, - expectedPaths: []string{"/proc", "/sys", "/sys/bar", "/tmp", "/tmp/baz"}, }, + expectedPaths: []string{"/proc", "/sys" /* "/sys/bar" ,*/, "/tmp", "/tmp/baz"}, } - for _, tc := range testCases { + if !vfs2 { + vfsCase.spec.Mounts = append(vfsCase.spec.Mounts, specs.Mount{Destination: "/sys/bar", Type: "tmpfs"}) + vfsCase.expectedPaths = append(vfsCase.expectedPaths, "/sys/bar") + } + return append(testCases, vfsCase) +} + +// Test that MountNamespace can be created with various specs. +func TestCreateMountNamespace(t *testing.T) { + + for _, tc := range createMountTestcases(false /* vfs2 */) { t.Run(tc.name, func(t *testing.T) { conf := testConfig() ctx := contexttest.Context(t) @@ -466,6 +474,56 @@ func TestCreateMountNamespace(t *testing.T) { } } +// Test that MountNamespace can be created with various specs. +func TestCreateMountNamespaceVFS2(t *testing.T) { + + for _, tc := range createMountTestcases(true /* vfs2 */) { + t.Run(tc.name, func(t *testing.T) { + defer resetSyscallTable() + + spec := testSpec() + spec.Mounts = tc.spec.Mounts + spec.Root = tc.spec.Root + + l, loaderCleanup, err := createLoader(true /* VFS2 Enabled */, spec) + if err != nil { + t.Fatalf("failed to create loader: %v", err) + } + defer l.Destroy() + defer loaderCleanup() + + mntr := newContainerMounter(l.spec, l.goferFDs, l.k, l.mountHints) + if err := mntr.processHints(l.conf); err != nil { + t.Fatalf("failed process hints: %v", err) + } + + ctx := l.rootProcArgs.NewContext(l.k) + mns, err := mntr.setupVFS2(ctx, l.conf, &l.rootProcArgs) + if err != nil { + t.Fatalf("failed to setupVFS2: %v", err) + } + + root := mns.Root() + defer root.DecRef() + for _, p := range tc.expectedPaths { + + target := &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(p), + } + + if d, err := l.k.VFS().GetDentryAt(ctx, l.rootProcArgs.Credentials, target, &vfs.GetDentryOptions{}); err != nil { + t.Errorf("expected path %v to exist with spec %v, but got error %v", p, tc.spec, err) + } else { + d.DecRef() + } + + } + }) + } +} + // TestRestoreEnvironment tests that the correct mounts are collected from the spec and config // in order to build the environment for restoring. func TestRestoreEnvironment(t *testing.T) { diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go index bce3a3593..0b9b0b436 100644 --- a/runsc/boot/vfs.go +++ b/runsc/boot/vfs.go @@ -17,6 +17,7 @@ package boot import ( "fmt" "path" + "sort" "strconv" "strings" @@ -192,14 +193,9 @@ func (c *containerMounter) setupVFS2(ctx context.Context, conf *Config, procArgs return nil, fmt.Errorf("register filesystems: %w", err) } - fd := c.fds.remove() - - opts := strings.Join(p9MountOptionsVFS2(fd, conf.FileAccess), ",") - - log.Infof("Mounting root over 9P, ioFD: %d", fd) - mns, err := c.k.VFS().NewMountNamespace(ctx, creds, "", rootFsName, &vfs.GetFilesystemOptions{Data: opts}) + mns, err := c.createMountNamespaceVFS2(ctx, conf, creds) if err != nil { - return nil, fmt.Errorf("setting up mountnamespace: %w", err) + return nil, fmt.Errorf("creating mount namespace: %w", err) } rootProcArgs.MountNamespaceVFS2 = mns @@ -212,8 +208,23 @@ func (c *containerMounter) setupVFS2(ctx context.Context, conf *Config, procArgs return mns, nil } +func (c *containerMounter) createMountNamespaceVFS2(ctx context.Context, conf *Config, creds *auth.Credentials) (*vfs.MountNamespace, error) { + + fd := c.fds.remove() + opts := strings.Join(p9MountOptionsVFS2(fd, conf.FileAccess), ",") + + log.Infof("Mounting root over 9P, ioFD: %d", fd) + mns, err := c.k.VFS().NewMountNamespace(ctx, creds, "", rootFsName, &vfs.GetFilesystemOptions{Data: opts}) + if err != nil { + return nil, fmt.Errorf("setting up mount namespace: %w", err) + } + return mns, nil +} + func (c *containerMounter) mountSubmountsVFS2(ctx context.Context, conf *Config, mns *vfs.MountNamespace, creds *auth.Credentials) error { + c.prepareMountsVFS2() + for _, submount := range c.mounts { log.Debugf("Mounting %q to %q, type: %s, options: %s", submount.Source, submount.Destination, submount.Type, submount.Options) if err := c.mountSubmountVFS2(ctx, conf, mns, creds, &submount); err != nil { @@ -226,6 +237,11 @@ func (c *containerMounter) mountSubmountsVFS2(ctx context.Context, conf *Config, return c.checkDispenser() } +func (c *containerMounter) prepareMountsVFS2() { + // Sort the mounts so that we don't place children before parents. + sort.Slice(c.mounts, func(i, j int) bool { return len(c.mounts[i].Destination) < len(c.mounts[j].Destination) }) +} + // TODO(gvisor.dev/issue/1487): Implement submount options similar to the VFS1 version. func (c *containerMounter) mountSubmountVFS2(ctx context.Context, conf *Config, mns *vfs.MountNamespace, creds *auth.Credentials, submount *specs.Mount) error { root := mns.Root() @@ -236,11 +252,21 @@ func (c *containerMounter) mountSubmountVFS2(ctx context.Context, conf *Config, Path: fspath.Parse(submount.Destination), } - _, options, useOverlay, err := c.getMountNameAndOptionsVFS2(conf, *submount) + fsName, options, useOverlay, err := c.getMountNameAndOptionsVFS2(conf, *submount) if err != nil { return fmt.Errorf("mountOptions failed: %w", err) } + if fsName == "" { + // Filesystem is not supported (e.g. cgroup), just skip it. + return nil + } + + if err := c.makeSyntheticMount(ctx, submount.Destination, root, creds); err != nil { + return err + } + log.Debugf("directory exists or made directory for submount: %s", submount.Destination) + opts := &vfs.MountOptions{ GetFilesystemOptions: vfs.GetFilesystemOptions{ Data: strings.Join(options, ","), @@ -251,12 +277,6 @@ func (c *containerMounter) mountSubmountVFS2(ctx context.Context, conf *Config, // All writes go to upper, be paranoid and make lower readonly. opts.ReadOnly = useOverlay - if err := c.k.VFS().MkdirAt(ctx, creds, target, &vfs.MkdirOptions{ - ForSyntheticMountpoint: true, - }); err != nil && err != syserror.EEXIST { - // Log a warning, but attempt the mount anyway. - log.Warningf("Failed to create mount point at %q: %v", submount.Destination, err) - } if err := c.k.VFS().MountAt(ctx, creds, "", target, submount.Type, opts); err != nil { return fmt.Errorf("failed to mount %q (type: %s): %w, opts: %v", submount.Destination, submount.Type, err, opts) } @@ -314,3 +334,33 @@ func p9MountOptionsVFS2(fd int, fa FileAccessType) []string { } return opts } + +func (c *containerMounter) makeSyntheticMount(ctx context.Context, currentPath string, root vfs.VirtualDentry, creds *auth.Credentials) error { + + target := &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(currentPath), + } + + _, err := c.k.VFS().StatAt(ctx, creds, target, &vfs.StatOptions{}) + switch { + + case err == syserror.ENOENT: + if err := c.makeSyntheticMount(ctx, path.Dir(currentPath), root, creds); err != nil { + return err + } + + mkdirOpts := &vfs.MkdirOptions{Mode: 0777, ForSyntheticMountpoint: true} + if err := c.k.VFS().MkdirAt(ctx, creds, target, mkdirOpts); err != nil { + return fmt.Errorf("failed to makedir for mount %+v: %w", target, err) + } + return nil + + case err != nil: + return fmt.Errorf("stat failed for mount %+v: %w", target, err) + + default: + return nil + } +} diff --git a/scripts/docker_tests.sh b/scripts/docker_tests.sh index 931ce1aa4..dce0a4085 100755 --- a/scripts/docker_tests.sh +++ b/scripts/docker_tests.sh @@ -20,3 +20,6 @@ make load-all-images install_runsc_for_test docker test_runsc //test/image:image_test //test/e2e:integration_test + +install_runsc_for_test docker --vfs2 +test_runsc //test/image:image_test --test_filter=.*TestHelloWorld -- cgit v1.2.3 From 3c67754663f424f2ebbc0ff2a4c80e30618d5355 Mon Sep 17 00:00:00 2001 From: Rahat Mahmood Date: Sat, 25 Apr 2020 23:54:56 -0700 Subject: Enable automated marshalling for signals and the arch package. PiperOrigin-RevId: 308472331 --- pkg/abi/linux/BUILD | 2 + pkg/abi/linux/ptrace_amd64.go | 52 ++++++++++++++++++++++ pkg/abi/linux/ptrace_arm64.go | 29 ++++++++++++ pkg/sentry/arch/BUILD | 4 +- pkg/sentry/arch/arch_aarch64.go | 26 ++++++----- pkg/sentry/arch/arch_amd64.go | 13 +++--- pkg/sentry/arch/arch_arm64.go | 2 + pkg/sentry/arch/arch_state_aarch64.go | 38 ---------------- pkg/sentry/arch/arch_state_x86.go | 42 ----------------- pkg/sentry/arch/arch_x86.go | 25 ++++++----- pkg/sentry/arch/arch_x86_impl.go | 4 +- pkg/sentry/arch/signal.go | 3 ++ pkg/sentry/arch/signal_act.go | 4 ++ pkg/sentry/arch/signal_stack.go | 3 ++ pkg/sentry/kernel/task_signals.go | 8 ++-- pkg/sentry/platform/kvm/kvm_arm64.go | 5 +-- pkg/sentry/platform/kvm/kvm_test.go | 36 +++++++-------- pkg/sentry/platform/kvm/testutil/BUILD | 1 + pkg/sentry/platform/kvm/testutil/testutil_amd64.go | 17 +++---- pkg/sentry/platform/kvm/testutil/testutil_arm64.go | 13 +++--- pkg/sentry/platform/ptrace/ptrace_amd64.go | 7 ++- pkg/sentry/platform/ptrace/ptrace_arm64.go | 5 +-- pkg/sentry/platform/ptrace/ptrace_unsafe.go | 4 +- pkg/sentry/platform/ptrace/subprocess.go | 8 ++-- pkg/sentry/platform/ptrace/subprocess_amd64.go | 16 +++---- pkg/sentry/platform/ptrace/subprocess_arm64.go | 16 +++---- pkg/sentry/platform/ring0/BUILD | 1 + pkg/sentry/platform/ring0/defs.go | 9 ++-- pkg/sentry/platform/ring0/entry_amd64.go | 6 +-- pkg/sentry/platform/ring0/gen_offsets/BUILD | 1 + pkg/sentry/platform/ring0/offsets_amd64.go | 5 ++- pkg/sentry/platform/ring0/offsets_arm64.go | 5 ++- pkg/sentry/syscalls/linux/sys_signal.go | 10 ++--- 33 files changed, 224 insertions(+), 196 deletions(-) create mode 100644 pkg/abi/linux/ptrace_amd64.go create mode 100644 pkg/abi/linux/ptrace_arm64.go delete mode 100644 pkg/sentry/arch/arch_state_aarch64.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/BUILD b/pkg/abi/linux/BUILD index 59b0e138a..114b516e2 100644 --- a/pkg/abi/linux/BUILD +++ b/pkg/abi/linux/BUILD @@ -44,6 +44,8 @@ go_library( "poll.go", "prctl.go", "ptrace.go", + "ptrace_amd64.go", + "ptrace_arm64.go", "rseq.go", "rusage.go", "sched.go", diff --git a/pkg/abi/linux/ptrace_amd64.go b/pkg/abi/linux/ptrace_amd64.go new file mode 100644 index 000000000..ed3881e27 --- /dev/null +++ b/pkg/abi/linux/ptrace_amd64.go @@ -0,0 +1,52 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build amd64 + +package linux + +// PtraceRegs is the set of CPU registers exposed by ptrace. Source: +// syscall.PtraceRegs. +// +// +marshal +// +stateify savable +type PtraceRegs struct { + R15 uint64 + R14 uint64 + R13 uint64 + R12 uint64 + Rbp uint64 + Rbx uint64 + R11 uint64 + R10 uint64 + R9 uint64 + R8 uint64 + Rax uint64 + Rcx uint64 + Rdx uint64 + Rsi uint64 + Rdi uint64 + Orig_rax uint64 + Rip uint64 + Cs uint64 + Eflags uint64 + Rsp uint64 + Ss uint64 + Fs_base uint64 + Gs_base uint64 + Ds uint64 + Es uint64 + Fs uint64 + Gs uint64 +} diff --git a/pkg/abi/linux/ptrace_arm64.go b/pkg/abi/linux/ptrace_arm64.go new file mode 100644 index 000000000..6147738b3 --- /dev/null +++ b/pkg/abi/linux/ptrace_arm64.go @@ -0,0 +1,29 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package linux + +// PtraceRegs is the set of CPU registers exposed by ptrace. Source: +// syscall.PtraceRegs. +// +// +marshal +// +stateify savable +type PtraceRegs struct { + Regs [31]uint64 + Sp uint64 + Pc uint64 + Pstate uint64 +} diff --git a/pkg/sentry/arch/BUILD b/pkg/sentry/arch/BUILD index e27f21e5e..901e0f320 100644 --- a/pkg/sentry/arch/BUILD +++ b/pkg/sentry/arch/BUILD @@ -11,7 +11,6 @@ go_library( "arch_amd64.go", "arch_amd64.s", "arch_arm64.go", - "arch_state_aarch64.go", "arch_state_x86.go", "arch_x86.go", "arch_x86_impl.go", @@ -26,11 +25,11 @@ go_library( "syscalls_amd64.go", "syscalls_arm64.go", ], + marshal = True, visibility = ["//:sandbox"], deps = [ ":registers_go_proto", "//pkg/abi/linux", - "//pkg/binary", "//pkg/context", "//pkg/cpuid", "//pkg/log", @@ -38,6 +37,7 @@ go_library( "//pkg/sync", "//pkg/syserror", "//pkg/usermem", + "//tools/go_marshal/marshal", ], ) diff --git a/pkg/sentry/arch/arch_aarch64.go b/pkg/sentry/arch/arch_aarch64.go index c29e1b841..529980267 100644 --- a/pkg/sentry/arch/arch_aarch64.go +++ b/pkg/sentry/arch/arch_aarch64.go @@ -17,18 +17,20 @@ package arch import ( + "encoding/binary" "fmt" "io" - "syscall" - "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/log" rpb "gvisor.dev/gvisor/pkg/sentry/arch/registers_go_proto" "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" ) +// Registers represents the CPU registers for this architecture. +type Registers = linux.PtraceRegs + const ( // SyscallWidth is the width of insturctions. SyscallWidth = 4 @@ -90,7 +92,7 @@ func NewFloatingPointData() *FloatingPointData { // file ensures it's only built on aarch64). type State struct { // The system registers. - Regs syscall.PtraceRegs `state:".(syscallPtraceRegs)"` + Regs Registers // Our floating point state. aarch64FPState `state:"wait"` @@ -226,25 +228,27 @@ func (s *State) RegisterMap() (map[string]uintptr, error) { // PtraceGetRegs implements Context.PtraceGetRegs. func (s *State) PtraceGetRegs(dst io.Writer) (int, error) { - return dst.Write(binary.Marshal(nil, usermem.ByteOrder, s.ptraceGetRegs())) + regs := s.ptraceGetRegs() + n, err := regs.WriteTo(dst) + return int(n), err } -func (s *State) ptraceGetRegs() syscall.PtraceRegs { +func (s *State) ptraceGetRegs() Registers { return s.Regs } -var ptraceRegsSize = int(binary.Size(syscall.PtraceRegs{})) +var registersSize = (*Registers)(nil).SizeBytes() // PtraceSetRegs implements Context.PtraceSetRegs. func (s *State) PtraceSetRegs(src io.Reader) (int, error) { - var regs syscall.PtraceRegs - buf := make([]byte, ptraceRegsSize) + var regs Registers + buf := make([]byte, registersSize) if _, err := io.ReadFull(src, buf); err != nil { return 0, err } - binary.Unmarshal(buf, usermem.ByteOrder, ®s) + regs.UnmarshalUnsafe(buf) s.Regs = regs - return ptraceRegsSize, nil + return registersSize, nil } // PtraceGetFPRegs implements Context.PtraceGetFPRegs. diff --git a/pkg/sentry/arch/arch_amd64.go b/pkg/sentry/arch/arch_amd64.go index 85d6acc0f..3b3a0a272 100644 --- a/pkg/sentry/arch/arch_amd64.go +++ b/pkg/sentry/arch/arch_amd64.go @@ -22,7 +22,6 @@ import ( "math/rand" "syscall" - "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/usermem" @@ -301,8 +300,10 @@ func (c *context64) PtracePeekUser(addr uintptr) (interface{}, error) { // PTRACE_PEEKUSER and PTRACE_POKEUSER are only effective on regs and // u_debugreg, returning 0 or silently no-oping for other fields // respectively. - if addr < uintptr(ptraceRegsSize) { - buf := binary.Marshal(nil, usermem.ByteOrder, c.ptraceGetRegs()) + if addr < uintptr(registersSize) { + regs := c.ptraceGetRegs() + buf := make([]byte, regs.SizeBytes()) + regs.MarshalUnsafe(buf) return c.Native(uintptr(usermem.ByteOrder.Uint64(buf[addr:]))), nil } // Note: x86 debug registers are missing. @@ -314,8 +315,10 @@ func (c *context64) PtracePokeUser(addr, data uintptr) error { if addr&7 != 0 || addr >= userStructSize { return syscall.EIO } - if addr < uintptr(ptraceRegsSize) { - buf := binary.Marshal(nil, usermem.ByteOrder, c.ptraceGetRegs()) + if addr < uintptr(registersSize) { + regs := c.ptraceGetRegs() + buf := make([]byte, regs.SizeBytes()) + regs.MarshalUnsafe(buf) usermem.ByteOrder.PutUint64(buf[addr:], uint64(data)) _, err := c.PtraceSetRegs(bytes.NewBuffer(buf)) return err diff --git a/pkg/sentry/arch/arch_arm64.go b/pkg/sentry/arch/arch_arm64.go index db99c5acb..ada7ac7b8 100644 --- a/pkg/sentry/arch/arch_arm64.go +++ b/pkg/sentry/arch/arch_arm64.go @@ -12,6 +12,8 @@ // See the License for the specific language governing permissions and // limitations under the License. +// +build arm64 + package arch import ( diff --git a/pkg/sentry/arch/arch_state_aarch64.go b/pkg/sentry/arch/arch_state_aarch64.go deleted file mode 100644 index 0136a85ad..000000000 --- a/pkg/sentry/arch/arch_state_aarch64.go +++ /dev/null @@ -1,38 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build arm64 - -package arch - -import ( - "syscall" -) - -type syscallPtraceRegs struct { - Regs [31]uint64 - Sp uint64 - Pc uint64 - Pstate uint64 -} - -// saveRegs is invoked by stateify. -func (s *State) saveRegs() syscallPtraceRegs { - return syscallPtraceRegs(s.Regs) -} - -// loadRegs is invoked by stateify. -func (s *State) loadRegs(r syscallPtraceRegs) { - s.Regs = syscall.PtraceRegs(r) -} diff --git a/pkg/sentry/arch/arch_state_x86.go b/pkg/sentry/arch/arch_state_x86.go index aa31169e0..19ce99d25 100644 --- a/pkg/sentry/arch/arch_state_x86.go +++ b/pkg/sentry/arch/arch_state_x86.go @@ -18,7 +18,6 @@ package arch import ( "fmt" - "syscall" "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/usermem" @@ -90,44 +89,3 @@ func (s *State) afterLoadFPState() { // Copy to the new, aligned location. copy(s.x86FPState, old) } - -// +stateify savable -type syscallPtraceRegs struct { - R15 uint64 - R14 uint64 - R13 uint64 - R12 uint64 - Rbp uint64 - Rbx uint64 - R11 uint64 - R10 uint64 - R9 uint64 - R8 uint64 - Rax uint64 - Rcx uint64 - Rdx uint64 - Rsi uint64 - Rdi uint64 - Orig_rax uint64 - Rip uint64 - Cs uint64 - Eflags uint64 - Rsp uint64 - Ss uint64 - Fs_base uint64 - Gs_base uint64 - Ds uint64 - Es uint64 - Fs uint64 - Gs uint64 -} - -// saveRegs is invoked by stateify. -func (s *State) saveRegs() syscallPtraceRegs { - return syscallPtraceRegs(s.Regs) -} - -// loadRegs is invoked by stateify. -func (s *State) loadRegs(r syscallPtraceRegs) { - s.Regs = syscall.PtraceRegs(r) -} diff --git a/pkg/sentry/arch/arch_x86.go b/pkg/sentry/arch/arch_x86.go index 7fc4c0473..dc458b37f 100644 --- a/pkg/sentry/arch/arch_x86.go +++ b/pkg/sentry/arch/arch_x86.go @@ -21,7 +21,7 @@ import ( "io" "syscall" - "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/log" rpb "gvisor.dev/gvisor/pkg/sentry/arch/registers_go_proto" @@ -30,6 +30,9 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) +// Registers represents the CPU registers for this architecture. +type Registers = linux.PtraceRegs + // System-related constants for x86. const ( // SyscallWidth is the width of syscall, sysenter, and int 80 insturctions. @@ -267,10 +270,12 @@ func (s *State) RegisterMap() (map[string]uintptr, error) { // PtraceGetRegs implements Context.PtraceGetRegs. func (s *State) PtraceGetRegs(dst io.Writer) (int, error) { - return dst.Write(binary.Marshal(nil, usermem.ByteOrder, s.ptraceGetRegs())) + regs := s.ptraceGetRegs() + n, err := regs.WriteTo(dst) + return int(n), err } -func (s *State) ptraceGetRegs() syscall.PtraceRegs { +func (s *State) ptraceGetRegs() Registers { regs := s.Regs // These may not be initialized. if regs.Cs == 0 || regs.Ss == 0 || regs.Eflags == 0 { @@ -306,16 +311,16 @@ func (s *State) ptraceGetRegs() syscall.PtraceRegs { return regs } -var ptraceRegsSize = int(binary.Size(syscall.PtraceRegs{})) +var registersSize = (*Registers)(nil).SizeBytes() // PtraceSetRegs implements Context.PtraceSetRegs. func (s *State) PtraceSetRegs(src io.Reader) (int, error) { - var regs syscall.PtraceRegs - buf := make([]byte, ptraceRegsSize) + var regs Registers + buf := make([]byte, registersSize) if _, err := io.ReadFull(src, buf); err != nil { return 0, err } - binary.Unmarshal(buf, usermem.ByteOrder, ®s) + regs.UnmarshalUnsafe(buf) // Truncate segment registers to 16 bits. regs.Cs = uint64(uint16(regs.Cs)) regs.Ds = uint64(uint16(regs.Ds)) @@ -369,7 +374,7 @@ func (s *State) PtraceSetRegs(src io.Reader) (int, error) { } regs.Eflags = (s.Regs.Eflags &^ eflagsPtraceMutable) | (regs.Eflags & eflagsPtraceMutable) s.Regs = regs - return ptraceRegsSize, nil + return registersSize, nil } // isUserSegmentSelector returns true if the given segment selector specifies a @@ -538,7 +543,7 @@ const ( func (s *State) PtraceGetRegSet(regset uintptr, dst io.Writer, maxlen int) (int, error) { switch regset { case _NT_PRSTATUS: - if maxlen < ptraceRegsSize { + if maxlen < registersSize { return 0, syserror.EFAULT } return s.PtraceGetRegs(dst) @@ -558,7 +563,7 @@ func (s *State) PtraceGetRegSet(regset uintptr, dst io.Writer, maxlen int) (int, func (s *State) PtraceSetRegSet(regset uintptr, src io.Reader, maxlen int) (int, error) { switch regset { case _NT_PRSTATUS: - if maxlen < ptraceRegsSize { + if maxlen < registersSize { return 0, syserror.EFAULT } return s.PtraceSetRegs(src) diff --git a/pkg/sentry/arch/arch_x86_impl.go b/pkg/sentry/arch/arch_x86_impl.go index 3edf40764..0c73fcbfb 100644 --- a/pkg/sentry/arch/arch_x86_impl.go +++ b/pkg/sentry/arch/arch_x86_impl.go @@ -17,8 +17,6 @@ package arch import ( - "syscall" - "gvisor.dev/gvisor/pkg/cpuid" ) @@ -28,7 +26,7 @@ import ( // +stateify savable type State struct { // The system registers. - Regs syscall.PtraceRegs `state:".(syscallPtraceRegs)"` + Regs Registers // Our floating point state. x86FPState `state:"wait"` diff --git a/pkg/sentry/arch/signal.go b/pkg/sentry/arch/signal.go index 8b03d0187..c9fb55d00 100644 --- a/pkg/sentry/arch/signal.go +++ b/pkg/sentry/arch/signal.go @@ -22,6 +22,7 @@ import ( // SignalAct represents the action that should be taken when a signal is // delivered, and is equivalent to struct sigaction. // +// +marshal // +stateify savable type SignalAct struct { Handler uint64 @@ -43,6 +44,7 @@ func (s *SignalAct) DeserializeTo(other *SignalAct) { // SignalStack represents information about a user stack, and is equivalent to // stack_t. // +// +marshal // +stateify savable type SignalStack struct { Addr uint64 @@ -64,6 +66,7 @@ func (s *SignalStack) DeserializeTo(other *SignalStack) { // SignalInfo represents information about a signal being delivered, and is // equivalent to struct siginfo in linux kernel(linux/include/uapi/asm-generic/siginfo.h). // +// +marshal // +stateify savable type SignalInfo struct { Signo int32 // Signal number diff --git a/pkg/sentry/arch/signal_act.go b/pkg/sentry/arch/signal_act.go index f9ca2e74e..32173aa20 100644 --- a/pkg/sentry/arch/signal_act.go +++ b/pkg/sentry/arch/signal_act.go @@ -14,6 +14,8 @@ package arch +import "gvisor.dev/gvisor/tools/go_marshal/marshal" + // Special values for SignalAct.Handler. const ( // SignalActDefault is SIG_DFL and specifies that the default behavior for @@ -71,6 +73,8 @@ func (s SignalAct) HasRestorer() bool { // NativeSignalAct is a type that is equivalent to struct sigaction in the // guest architecture. type NativeSignalAct interface { + marshal.Marshallable + // SerializeFrom copies the data in the host SignalAct s into this object. SerializeFrom(s *SignalAct) diff --git a/pkg/sentry/arch/signal_stack.go b/pkg/sentry/arch/signal_stack.go index e58f055c7..0fa738a1d 100644 --- a/pkg/sentry/arch/signal_stack.go +++ b/pkg/sentry/arch/signal_stack.go @@ -18,6 +18,7 @@ package arch import ( "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/tools/go_marshal/marshal" ) const ( @@ -55,6 +56,8 @@ func (s *SignalStack) Contains(sp usermem.Addr) bool { // NativeSignalStack is a type that is equivalent to stack_t in the guest // architecture. type NativeSignalStack interface { + marshal.Marshallable + // SerializeFrom copies the data in the host SignalStack s into this // object. SerializeFrom(s *SignalStack) diff --git a/pkg/sentry/kernel/task_signals.go b/pkg/sentry/kernel/task_signals.go index 7d25e98f7..79766cafe 100644 --- a/pkg/sentry/kernel/task_signals.go +++ b/pkg/sentry/kernel/task_signals.go @@ -716,7 +716,7 @@ func (tg *ThreadGroup) SetSignalAct(sig linux.Signal, actptr *arch.SignalAct) (a func (t *Task) CopyOutSignalAct(addr usermem.Addr, s *arch.SignalAct) error { n := t.Arch().NewSignalAct() n.SerializeFrom(s) - _, err := t.CopyOut(addr, n) + _, err := n.CopyOut(t, addr) return err } @@ -725,7 +725,7 @@ func (t *Task) CopyOutSignalAct(addr usermem.Addr, s *arch.SignalAct) error { func (t *Task) CopyInSignalAct(addr usermem.Addr) (arch.SignalAct, error) { n := t.Arch().NewSignalAct() var s arch.SignalAct - if _, err := t.CopyIn(addr, n); err != nil { + if _, err := n.CopyIn(t, addr); err != nil { return s, err } n.DeserializeTo(&s) @@ -737,7 +737,7 @@ func (t *Task) CopyInSignalAct(addr usermem.Addr) (arch.SignalAct, error) { func (t *Task) CopyOutSignalStack(addr usermem.Addr, s *arch.SignalStack) error { n := t.Arch().NewSignalStack() n.SerializeFrom(s) - _, err := t.CopyOut(addr, n) + _, err := n.CopyOut(t, addr) return err } @@ -746,7 +746,7 @@ func (t *Task) CopyOutSignalStack(addr usermem.Addr, s *arch.SignalStack) error func (t *Task) CopyInSignalStack(addr usermem.Addr) (arch.SignalStack, error) { n := t.Arch().NewSignalStack() var s arch.SignalStack - if _, err := t.CopyIn(addr, n); err != nil { + if _, err := n.CopyIn(t, addr); err != nil { return s, err } n.DeserializeTo(&s) diff --git a/pkg/sentry/platform/kvm/kvm_arm64.go b/pkg/sentry/platform/kvm/kvm_arm64.go index 716198712..29d457a7e 100644 --- a/pkg/sentry/platform/kvm/kvm_arm64.go +++ b/pkg/sentry/platform/kvm/kvm_arm64.go @@ -17,8 +17,7 @@ package kvm import ( - "syscall" - + "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/platform/ring0" ) @@ -37,7 +36,7 @@ type userFpsimdState struct { } type userRegs struct { - Regs syscall.PtraceRegs + Regs arch.Registers sp_el1 uint64 elr_el1 uint64 spsr [KVM_NR_SPSR]uint64 diff --git a/pkg/sentry/platform/kvm/kvm_test.go b/pkg/sentry/platform/kvm/kvm_test.go index c42752d50..6c8f4fa28 100644 --- a/pkg/sentry/platform/kvm/kvm_test.go +++ b/pkg/sentry/platform/kvm/kvm_test.go @@ -117,10 +117,10 @@ func TestKernelFloatingPoint(t *testing.T) { }) } -func applicationTest(t testHarness, useHostMappings bool, target func(), fn func(*vCPU, *syscall.PtraceRegs, *pagetables.PageTables) bool) { +func applicationTest(t testHarness, useHostMappings bool, target func(), fn func(*vCPU, *arch.Registers, *pagetables.PageTables) bool) { // Initialize registers & page tables. var ( - regs syscall.PtraceRegs + regs arch.Registers pt *pagetables.PageTables ) testutil.SetTestTarget(®s, target) @@ -154,7 +154,7 @@ func applicationTest(t testHarness, useHostMappings bool, target func(), fn func } func TestApplicationSyscall(t *testing.T) { - applicationTest(t, true, testutil.SyscallLoop, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { + applicationTest(t, true, testutil.SyscallLoop, func(c *vCPU, regs *arch.Registers, pt *pagetables.PageTables) bool { var si arch.SignalInfo if _, err := c.SwitchToUser(ring0.SwitchOpts{ Registers: regs, @@ -168,7 +168,7 @@ func TestApplicationSyscall(t *testing.T) { } return false }) - applicationTest(t, true, testutil.SyscallLoop, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { + applicationTest(t, true, testutil.SyscallLoop, func(c *vCPU, regs *arch.Registers, pt *pagetables.PageTables) bool { var si arch.SignalInfo if _, err := c.SwitchToUser(ring0.SwitchOpts{ Registers: regs, @@ -184,7 +184,7 @@ func TestApplicationSyscall(t *testing.T) { } func TestApplicationFault(t *testing.T) { - applicationTest(t, true, testutil.Touch, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { + applicationTest(t, true, testutil.Touch, func(c *vCPU, regs *arch.Registers, pt *pagetables.PageTables) bool { testutil.SetTouchTarget(regs, nil) // Cause fault. var si arch.SignalInfo if _, err := c.SwitchToUser(ring0.SwitchOpts{ @@ -199,7 +199,7 @@ func TestApplicationFault(t *testing.T) { } return false }) - applicationTest(t, true, testutil.Touch, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { + applicationTest(t, true, testutil.Touch, func(c *vCPU, regs *arch.Registers, pt *pagetables.PageTables) bool { testutil.SetTouchTarget(regs, nil) // Cause fault. var si arch.SignalInfo if _, err := c.SwitchToUser(ring0.SwitchOpts{ @@ -216,7 +216,7 @@ func TestApplicationFault(t *testing.T) { } func TestRegistersSyscall(t *testing.T) { - applicationTest(t, true, testutil.TwiddleRegsSyscall, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { + applicationTest(t, true, testutil.TwiddleRegsSyscall, func(c *vCPU, regs *arch.Registers, pt *pagetables.PageTables) bool { testutil.SetTestRegs(regs) // Fill values for all registers. for { var si arch.SignalInfo @@ -239,7 +239,7 @@ func TestRegistersSyscall(t *testing.T) { } func TestRegistersFault(t *testing.T) { - applicationTest(t, true, testutil.TwiddleRegsFault, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { + applicationTest(t, true, testutil.TwiddleRegsFault, func(c *vCPU, regs *arch.Registers, pt *pagetables.PageTables) bool { testutil.SetTestRegs(regs) // Fill values for all registers. for { var si arch.SignalInfo @@ -263,7 +263,7 @@ func TestRegistersFault(t *testing.T) { } func TestSegments(t *testing.T) { - applicationTest(t, true, testutil.TwiddleSegments, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { + applicationTest(t, true, testutil.TwiddleSegments, func(c *vCPU, regs *arch.Registers, pt *pagetables.PageTables) bool { testutil.SetTestSegments(regs) for { var si arch.SignalInfo @@ -287,7 +287,7 @@ func TestSegments(t *testing.T) { } func TestBounce(t *testing.T) { - applicationTest(t, true, testutil.SpinLoop, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { + applicationTest(t, true, testutil.SpinLoop, func(c *vCPU, regs *arch.Registers, pt *pagetables.PageTables) bool { go func() { time.Sleep(time.Millisecond) c.BounceToKernel() @@ -302,7 +302,7 @@ func TestBounce(t *testing.T) { } return false }) - applicationTest(t, true, testutil.SpinLoop, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { + applicationTest(t, true, testutil.SpinLoop, func(c *vCPU, regs *arch.Registers, pt *pagetables.PageTables) bool { go func() { time.Sleep(time.Millisecond) c.BounceToKernel() @@ -321,7 +321,7 @@ func TestBounce(t *testing.T) { } func TestBounceStress(t *testing.T) { - applicationTest(t, true, testutil.SpinLoop, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { + applicationTest(t, true, testutil.SpinLoop, func(c *vCPU, regs *arch.Registers, pt *pagetables.PageTables) bool { randomSleep := func() { // O(hundreds of microseconds) is appropriate to ensure // different overlaps and different schedules. @@ -357,7 +357,7 @@ func TestBounceStress(t *testing.T) { func TestInvalidate(t *testing.T) { var data uintptr // Used below. - applicationTest(t, true, testutil.Touch, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { + applicationTest(t, true, testutil.Touch, func(c *vCPU, regs *arch.Registers, pt *pagetables.PageTables) bool { testutil.SetTouchTarget(regs, &data) // Read legitimate value. for { var si arch.SignalInfo @@ -398,7 +398,7 @@ func IsFault(err error, si *arch.SignalInfo) bool { } func TestEmptyAddressSpace(t *testing.T) { - applicationTest(t, false, testutil.SyscallLoop, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { + applicationTest(t, false, testutil.SyscallLoop, func(c *vCPU, regs *arch.Registers, pt *pagetables.PageTables) bool { var si arch.SignalInfo if _, err := c.SwitchToUser(ring0.SwitchOpts{ Registers: regs, @@ -412,7 +412,7 @@ func TestEmptyAddressSpace(t *testing.T) { } return false }) - applicationTest(t, false, testutil.SyscallLoop, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { + applicationTest(t, false, testutil.SyscallLoop, func(c *vCPU, regs *arch.Registers, pt *pagetables.PageTables) bool { var si arch.SignalInfo if _, err := c.SwitchToUser(ring0.SwitchOpts{ Registers: regs, @@ -471,7 +471,7 @@ func BenchmarkApplicationSyscall(b *testing.B) { i int // Iteration includes machine.Get() / machine.Put(). a int // Count for ErrContextInterrupt. ) - applicationTest(b, true, testutil.SyscallLoop, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { + applicationTest(b, true, testutil.SyscallLoop, func(c *vCPU, regs *arch.Registers, pt *pagetables.PageTables) bool { var si arch.SignalInfo if _, err := c.SwitchToUser(ring0.SwitchOpts{ Registers: regs, @@ -493,7 +493,7 @@ func BenchmarkApplicationSyscall(b *testing.B) { func BenchmarkKernelSyscall(b *testing.B) { // Note that the target passed here is irrelevant, we never execute SwitchToUser. - applicationTest(b, true, testutil.Getpid, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { + applicationTest(b, true, testutil.Getpid, func(c *vCPU, regs *arch.Registers, pt *pagetables.PageTables) bool { // iteration does not include machine.Get() / machine.Put(). for i := 0; i < b.N; i++ { testutil.Getpid() @@ -508,7 +508,7 @@ func BenchmarkWorldSwitchToUserRoundtrip(b *testing.B) { i int a int ) - applicationTest(b, true, testutil.SyscallLoop, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { + applicationTest(b, true, testutil.SyscallLoop, func(c *vCPU, regs *arch.Registers, pt *pagetables.PageTables) bool { var si arch.SignalInfo if _, err := c.SwitchToUser(ring0.SwitchOpts{ Registers: regs, diff --git a/pkg/sentry/platform/kvm/testutil/BUILD b/pkg/sentry/platform/kvm/testutil/BUILD index f7605df8a..f7feb8683 100644 --- a/pkg/sentry/platform/kvm/testutil/BUILD +++ b/pkg/sentry/platform/kvm/testutil/BUILD @@ -13,4 +13,5 @@ go_library( "testutil_arm64.s", ], visibility = ["//pkg/sentry/platform/kvm:__pkg__"], + deps = ["//pkg/sentry/arch"], ) diff --git a/pkg/sentry/platform/kvm/testutil/testutil_amd64.go b/pkg/sentry/platform/kvm/testutil/testutil_amd64.go index 4c108abbf..8048eedec 100644 --- a/pkg/sentry/platform/kvm/testutil/testutil_amd64.go +++ b/pkg/sentry/platform/kvm/testutil/testutil_amd64.go @@ -18,19 +18,20 @@ package testutil import ( "reflect" - "syscall" + + "gvisor.dev/gvisor/pkg/sentry/arch" ) // TwiddleSegments reads segments into known registers. func TwiddleSegments() // SetTestTarget sets the rip appropriately. -func SetTestTarget(regs *syscall.PtraceRegs, fn func()) { +func SetTestTarget(regs *arch.Registers, fn func()) { regs.Rip = uint64(reflect.ValueOf(fn).Pointer()) } // SetTouchTarget sets rax appropriately. -func SetTouchTarget(regs *syscall.PtraceRegs, target *uintptr) { +func SetTouchTarget(regs *arch.Registers, target *uintptr) { if target != nil { regs.Rax = uint64(reflect.ValueOf(target).Pointer()) } else { @@ -39,12 +40,12 @@ func SetTouchTarget(regs *syscall.PtraceRegs, target *uintptr) { } // RewindSyscall rewinds a syscall RIP. -func RewindSyscall(regs *syscall.PtraceRegs) { +func RewindSyscall(regs *arch.Registers) { regs.Rip -= 2 } // SetTestRegs initializes registers to known values. -func SetTestRegs(regs *syscall.PtraceRegs) { +func SetTestRegs(regs *arch.Registers) { regs.R15 = 0x15 regs.R14 = 0x14 regs.R13 = 0x13 @@ -64,7 +65,7 @@ func SetTestRegs(regs *syscall.PtraceRegs) { } // CheckTestRegs checks that registers were twiddled per TwiddleRegs. -func CheckTestRegs(regs *syscall.PtraceRegs, full bool) (err error) { +func CheckTestRegs(regs *arch.Registers, full bool) (err error) { if need := ^uint64(0x15); regs.R15 != need { err = addRegisterMismatch(err, "R15", regs.R15, need) } @@ -121,13 +122,13 @@ var fsData uint64 = 0x55 var gsData uint64 = 0x85 // SetTestSegments initializes segments to known values. -func SetTestSegments(regs *syscall.PtraceRegs) { +func SetTestSegments(regs *arch.Registers) { regs.Fs_base = uint64(reflect.ValueOf(&fsData).Pointer()) regs.Gs_base = uint64(reflect.ValueOf(&gsData).Pointer()) } // CheckTestSegments checks that registers were twiddled per TwiddleSegments. -func CheckTestSegments(regs *syscall.PtraceRegs) (err error) { +func CheckTestSegments(regs *arch.Registers) (err error) { if regs.Rax != fsData { err = addRegisterMismatch(err, "Rax", regs.Rax, fsData) } diff --git a/pkg/sentry/platform/kvm/testutil/testutil_arm64.go b/pkg/sentry/platform/kvm/testutil/testutil_arm64.go index 40b2e4acc..ca902c8c1 100644 --- a/pkg/sentry/platform/kvm/testutil/testutil_arm64.go +++ b/pkg/sentry/platform/kvm/testutil/testutil_arm64.go @@ -19,16 +19,17 @@ package testutil import ( "fmt" "reflect" - "syscall" + + "gvisor.dev/gvisor/pkg/sentry/arch" ) // SetTestTarget sets the rip appropriately. -func SetTestTarget(regs *syscall.PtraceRegs, fn func()) { +func SetTestTarget(regs *arch.Registers, fn func()) { regs.Pc = uint64(reflect.ValueOf(fn).Pointer()) } // SetTouchTarget sets rax appropriately. -func SetTouchTarget(regs *syscall.PtraceRegs, target *uintptr) { +func SetTouchTarget(regs *arch.Registers, target *uintptr) { if target != nil { regs.Regs[8] = uint64(reflect.ValueOf(target).Pointer()) } else { @@ -37,19 +38,19 @@ func SetTouchTarget(regs *syscall.PtraceRegs, target *uintptr) { } // RewindSyscall rewinds a syscall RIP. -func RewindSyscall(regs *syscall.PtraceRegs) { +func RewindSyscall(regs *arch.Registers) { regs.Pc -= 4 } // SetTestRegs initializes registers to known values. -func SetTestRegs(regs *syscall.PtraceRegs) { +func SetTestRegs(regs *arch.Registers) { for i := 0; i <= 30; i++ { regs.Regs[i] = uint64(i) + 1 } } // CheckTestRegs checks that registers were twiddled per TwiddleRegs. -func CheckTestRegs(regs *syscall.PtraceRegs, full bool) (err error) { +func CheckTestRegs(regs *arch.Registers, full bool) (err error) { for i := 0; i <= 30; i++ { if need := ^uint64(i + 1); regs.Regs[i] != need { err = addRegisterMismatch(err, fmt.Sprintf("R%d", i), regs.Regs[i], need) diff --git a/pkg/sentry/platform/ptrace/ptrace_amd64.go b/pkg/sentry/platform/ptrace/ptrace_amd64.go index 24fc5dc62..3b9a870a5 100644 --- a/pkg/sentry/platform/ptrace/ptrace_amd64.go +++ b/pkg/sentry/platform/ptrace/ptrace_amd64.go @@ -15,9 +15,8 @@ package ptrace import ( - "syscall" - "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" ) // fpRegSet returns the GETREGSET/SETREGSET register set type to be used. @@ -28,12 +27,12 @@ func fpRegSet(useXsave bool) uintptr { return linux.NT_PRFPREG } -func stackPointer(r *syscall.PtraceRegs) uintptr { +func stackPointer(r *arch.Registers) uintptr { return uintptr(r.Rsp) } // x86 use the fs_base register to store the TLS pointer which can be -// get/set in "func (t *thread) get/setRegs(regs *syscall.PtraceRegs)". +// get/set in "func (t *thread) get/setRegs(regs *arch.Registers)". // So both of the get/setTLS() operations are noop here. // getTLS gets the thread local storage register. diff --git a/pkg/sentry/platform/ptrace/ptrace_arm64.go b/pkg/sentry/platform/ptrace/ptrace_arm64.go index 4db28c534..5c869926a 100644 --- a/pkg/sentry/platform/ptrace/ptrace_arm64.go +++ b/pkg/sentry/platform/ptrace/ptrace_arm64.go @@ -15,9 +15,8 @@ package ptrace import ( - "syscall" - "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" ) // fpRegSet returns the GETREGSET/SETREGSET register set type to be used. @@ -25,6 +24,6 @@ func fpRegSet(_ bool) uintptr { return linux.NT_PRFPREG } -func stackPointer(r *syscall.PtraceRegs) uintptr { +func stackPointer(r *arch.Registers) uintptr { return uintptr(r.Sp) } diff --git a/pkg/sentry/platform/ptrace/ptrace_unsafe.go b/pkg/sentry/platform/ptrace/ptrace_unsafe.go index 6c0ed7b3e..8b72d24e8 100644 --- a/pkg/sentry/platform/ptrace/ptrace_unsafe.go +++ b/pkg/sentry/platform/ptrace/ptrace_unsafe.go @@ -24,7 +24,7 @@ import ( ) // getRegs gets the general purpose register set. -func (t *thread) getRegs(regs *syscall.PtraceRegs) error { +func (t *thread) getRegs(regs *arch.Registers) error { iovec := syscall.Iovec{ Base: (*byte)(unsafe.Pointer(regs)), Len: uint64(unsafe.Sizeof(*regs)), @@ -43,7 +43,7 @@ func (t *thread) getRegs(regs *syscall.PtraceRegs) error { } // setRegs sets the general purpose register set. -func (t *thread) setRegs(regs *syscall.PtraceRegs) error { +func (t *thread) setRegs(regs *arch.Registers) error { iovec := syscall.Iovec{ Base: (*byte)(unsafe.Pointer(regs)), Len: uint64(unsafe.Sizeof(*regs)), diff --git a/pkg/sentry/platform/ptrace/subprocess.go b/pkg/sentry/platform/ptrace/subprocess.go index 773ddb1ed..2389423b0 100644 --- a/pkg/sentry/platform/ptrace/subprocess.go +++ b/pkg/sentry/platform/ptrace/subprocess.go @@ -63,7 +63,7 @@ type thread struct { // initRegs are the initial registers for the first thread. // // These are used for the register set for system calls. - initRegs syscall.PtraceRegs + initRegs arch.Registers } // threadPool is a collection of threads. @@ -317,7 +317,7 @@ const ( ) func (t *thread) dumpAndPanic(message string) { - var regs syscall.PtraceRegs + var regs arch.Registers message += "\n" if err := t.getRegs(®s); err == nil { message += dumpRegs(®s) @@ -423,7 +423,7 @@ func (t *thread) init() { // This is _not_ for use by application system calls, rather it is for use when // a system call must be injected into the remote context (e.g. mmap, munmap). // Note that clones are handled separately. -func (t *thread) syscall(regs *syscall.PtraceRegs) (uintptr, error) { +func (t *thread) syscall(regs *arch.Registers) (uintptr, error) { // Set registers. if err := t.setRegs(regs); err != nil { panic(fmt.Sprintf("ptrace set regs failed: %v", err)) @@ -461,7 +461,7 @@ func (t *thread) syscall(regs *syscall.PtraceRegs) (uintptr, error) { // syscallIgnoreInterrupt ignores interrupts on the system call thread and // restarts the syscall if the kernel indicates that should happen. func (t *thread) syscallIgnoreInterrupt( - initRegs *syscall.PtraceRegs, + initRegs *arch.Registers, sysno uintptr, args ...arch.SyscallArgument) (uintptr, error) { for { diff --git a/pkg/sentry/platform/ptrace/subprocess_amd64.go b/pkg/sentry/platform/ptrace/subprocess_amd64.go index cd74945e7..84b699f0d 100644 --- a/pkg/sentry/platform/ptrace/subprocess_amd64.go +++ b/pkg/sentry/platform/ptrace/subprocess_amd64.go @@ -41,7 +41,7 @@ const ( // resetSysemuRegs sets up emulation registers. // // This should be called prior to calling sysemu. -func (t *thread) resetSysemuRegs(regs *syscall.PtraceRegs) { +func (t *thread) resetSysemuRegs(regs *arch.Registers) { regs.Cs = t.initRegs.Cs regs.Ss = t.initRegs.Ss regs.Ds = t.initRegs.Ds @@ -53,7 +53,7 @@ func (t *thread) resetSysemuRegs(regs *syscall.PtraceRegs) { // createSyscallRegs sets up syscall registers. // // This should be called to generate registers for a system call. -func createSyscallRegs(initRegs *syscall.PtraceRegs, sysno uintptr, args ...arch.SyscallArgument) syscall.PtraceRegs { +func createSyscallRegs(initRegs *arch.Registers, sysno uintptr, args ...arch.SyscallArgument) arch.Registers { // Copy initial registers. regs := *initRegs @@ -82,18 +82,18 @@ func createSyscallRegs(initRegs *syscall.PtraceRegs, sysno uintptr, args ...arch } // isSingleStepping determines if the registers indicate single-stepping. -func isSingleStepping(regs *syscall.PtraceRegs) bool { +func isSingleStepping(regs *arch.Registers) bool { return (regs.Eflags & arch.X86TrapFlag) != 0 } // updateSyscallRegs updates registers after finishing sysemu. -func updateSyscallRegs(regs *syscall.PtraceRegs) { +func updateSyscallRegs(regs *arch.Registers) { // Ptrace puts -ENOSYS in rax on syscall-enter-stop. regs.Rax = regs.Orig_rax } // syscallReturnValue extracts a sensible return from registers. -func syscallReturnValue(regs *syscall.PtraceRegs) (uintptr, error) { +func syscallReturnValue(regs *arch.Registers) (uintptr, error) { rval := int64(regs.Rax) if rval < 0 { return 0, syscall.Errno(-rval) @@ -101,7 +101,7 @@ func syscallReturnValue(regs *syscall.PtraceRegs) (uintptr, error) { return uintptr(rval), nil } -func dumpRegs(regs *syscall.PtraceRegs) string { +func dumpRegs(regs *arch.Registers) string { var m strings.Builder fmt.Fprintf(&m, "Registers:\n") @@ -143,7 +143,7 @@ func (t *thread) adjustInitRegsRip() { } // Pass the expected PPID to the child via R15 when creating stub process. -func initChildProcessPPID(initregs *syscall.PtraceRegs, ppid int32) { +func initChildProcessPPID(initregs *arch.Registers, ppid int32) { initregs.R15 = uint64(ppid) // Rbx has to be set to 1 when creating stub process. initregs.Rbx = 1 @@ -156,7 +156,7 @@ func initChildProcessPPID(initregs *syscall.PtraceRegs, ppid int32) { // // Note that this should only be called after verifying that the signalInfo has // been generated by the kernel. -func patchSignalInfo(regs *syscall.PtraceRegs, signalInfo *arch.SignalInfo) { +func patchSignalInfo(regs *arch.Registers, signalInfo *arch.SignalInfo) { if linux.Signal(signalInfo.Signo) == linux.SIGSYS { signalInfo.Signo = int32(linux.SIGSEGV) diff --git a/pkg/sentry/platform/ptrace/subprocess_arm64.go b/pkg/sentry/platform/ptrace/subprocess_arm64.go index 7f5c393f0..bd618fae8 100644 --- a/pkg/sentry/platform/ptrace/subprocess_arm64.go +++ b/pkg/sentry/platform/ptrace/subprocess_arm64.go @@ -41,13 +41,13 @@ const ( // resetSysemuRegs sets up emulation registers. // // This should be called prior to calling sysemu. -func (t *thread) resetSysemuRegs(regs *syscall.PtraceRegs) { +func (t *thread) resetSysemuRegs(regs *arch.Registers) { } // createSyscallRegs sets up syscall registers. // // This should be called to generate registers for a system call. -func createSyscallRegs(initRegs *syscall.PtraceRegs, sysno uintptr, args ...arch.SyscallArgument) syscall.PtraceRegs { +func createSyscallRegs(initRegs *arch.Registers, sysno uintptr, args ...arch.SyscallArgument) arch.Registers { // Copy initial registers (Pc, Sp, etc.). regs := *initRegs @@ -78,7 +78,7 @@ func createSyscallRegs(initRegs *syscall.PtraceRegs, sysno uintptr, args ...arch } // isSingleStepping determines if the registers indicate single-stepping. -func isSingleStepping(regs *syscall.PtraceRegs) bool { +func isSingleStepping(regs *arch.Registers) bool { // Refer to the ARM SDM D2.12.3: software step state machine // return (regs.Pstate.SS == 1) && (MDSCR_EL1.SS == 1). // @@ -89,13 +89,13 @@ func isSingleStepping(regs *syscall.PtraceRegs) bool { } // updateSyscallRegs updates registers after finishing sysemu. -func updateSyscallRegs(regs *syscall.PtraceRegs) { +func updateSyscallRegs(regs *arch.Registers) { // No special work is necessary. return } // syscallReturnValue extracts a sensible return from registers. -func syscallReturnValue(regs *syscall.PtraceRegs) (uintptr, error) { +func syscallReturnValue(regs *arch.Registers) (uintptr, error) { rval := int64(regs.Regs[0]) if rval < 0 { return 0, syscall.Errno(-rval) @@ -103,7 +103,7 @@ func syscallReturnValue(regs *syscall.PtraceRegs) (uintptr, error) { return uintptr(rval), nil } -func dumpRegs(regs *syscall.PtraceRegs) string { +func dumpRegs(regs *arch.Registers) string { var m strings.Builder fmt.Fprintf(&m, "Registers:\n") @@ -125,7 +125,7 @@ func (t *thread) adjustInitRegsRip() { } // Pass the expected PPID to the child via X7 when creating stub process -func initChildProcessPPID(initregs *syscall.PtraceRegs, ppid int32) { +func initChildProcessPPID(initregs *arch.Registers, ppid int32) { initregs.Regs[7] = uint64(ppid) // R9 has to be set to 1 when creating stub process. initregs.Regs[9] = 1 @@ -138,7 +138,7 @@ func initChildProcessPPID(initregs *syscall.PtraceRegs, ppid int32) { // // Note that this should only be called after verifying that the signalInfo has // been generated by the kernel. -func patchSignalInfo(regs *syscall.PtraceRegs, signalInfo *arch.SignalInfo) { +func patchSignalInfo(regs *arch.Registers, signalInfo *arch.SignalInfo) { if linux.Signal(signalInfo.Signo) == linux.SIGSYS { signalInfo.Signo = int32(linux.SIGSEGV) diff --git a/pkg/sentry/platform/ring0/BUILD b/pkg/sentry/platform/ring0/BUILD index b69520030..679b287c3 100644 --- a/pkg/sentry/platform/ring0/BUILD +++ b/pkg/sentry/platform/ring0/BUILD @@ -79,6 +79,7 @@ go_library( deps = [ "//pkg/cpuid", "//pkg/safecopy", + "//pkg/sentry/arch", "//pkg/sentry/platform/ring0/pagetables", "//pkg/usermem", ], diff --git a/pkg/sentry/platform/ring0/defs.go b/pkg/sentry/platform/ring0/defs.go index 86fd5ed58..e6daf24df 100644 --- a/pkg/sentry/platform/ring0/defs.go +++ b/pkg/sentry/platform/ring0/defs.go @@ -15,8 +15,7 @@ package ring0 import ( - "syscall" - + "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" ) @@ -72,7 +71,7 @@ type CPU struct { // registers is a set of registers; these may be used on kernel system // calls and exceptions via the Registers function. - registers syscall.PtraceRegs + registers arch.Registers // hooks are kernel hooks. hooks Hooks @@ -83,14 +82,14 @@ type CPU struct { // This is explicitly safe to call during KernelException and KernelSyscall. // //go:nosplit -func (c *CPU) Registers() *syscall.PtraceRegs { +func (c *CPU) Registers() *arch.Registers { return &c.registers } // SwitchOpts are passed to the Switch function. type SwitchOpts struct { // Registers are the user register state. - Registers *syscall.PtraceRegs + Registers *arch.Registers // FloatingPointState is a byte pointer where floating point state is // saved and restored. diff --git a/pkg/sentry/platform/ring0/entry_amd64.go b/pkg/sentry/platform/ring0/entry_amd64.go index a5ce67885..7fa43c2f5 100644 --- a/pkg/sentry/platform/ring0/entry_amd64.go +++ b/pkg/sentry/platform/ring0/entry_amd64.go @@ -17,7 +17,7 @@ package ring0 import ( - "syscall" + "gvisor.dev/gvisor/pkg/sentry/arch" ) // This is an assembly function. @@ -41,7 +41,7 @@ func swapgs() // The return code is the vector that interrupted execution. // // See stubs.go for a note regarding the frame size of this function. -func sysret(*CPU, *syscall.PtraceRegs) Vector +func sysret(*CPU, *arch.Registers) Vector // "iret is the cadillac of CPL switching." // @@ -50,7 +50,7 @@ func sysret(*CPU, *syscall.PtraceRegs) Vector // iret is nearly identical to sysret, except an iret is used to fully restore // all user state. This must be called in cases where all registers need to be // restored. -func iret(*CPU, *syscall.PtraceRegs) Vector +func iret(*CPU, *arch.Registers) Vector // exception is the generic exception entry. // diff --git a/pkg/sentry/platform/ring0/gen_offsets/BUILD b/pkg/sentry/platform/ring0/gen_offsets/BUILD index 4cae10459..549f3d228 100644 --- a/pkg/sentry/platform/ring0/gen_offsets/BUILD +++ b/pkg/sentry/platform/ring0/gen_offsets/BUILD @@ -27,6 +27,7 @@ go_binary( visibility = ["//pkg/sentry/platform/ring0:__pkg__"], deps = [ "//pkg/cpuid", + "//pkg/sentry/arch", "//pkg/sentry/platform/ring0/pagetables", "//pkg/usermem", ], diff --git a/pkg/sentry/platform/ring0/offsets_amd64.go b/pkg/sentry/platform/ring0/offsets_amd64.go index 85cc3fdad..b8ab120a0 100644 --- a/pkg/sentry/platform/ring0/offsets_amd64.go +++ b/pkg/sentry/platform/ring0/offsets_amd64.go @@ -20,7 +20,8 @@ import ( "fmt" "io" "reflect" - "syscall" + + "gvisor.dev/gvisor/pkg/sentry/arch" ) // Emit prints architecture-specific offsets. @@ -64,7 +65,7 @@ func Emit(w io.Writer) { fmt.Fprintf(w, "#define SyscallInt80 0x%02x\n", SyscallInt80) fmt.Fprintf(w, "#define Syscall 0x%02x\n", Syscall) - p := &syscall.PtraceRegs{} + p := &arch.Registers{} fmt.Fprintf(w, "\n// Ptrace registers.\n") fmt.Fprintf(w, "#define PTRACE_R15 0x%02x\n", reflect.ValueOf(&p.R15).Pointer()-reflect.ValueOf(p).Pointer()) fmt.Fprintf(w, "#define PTRACE_R14 0x%02x\n", reflect.ValueOf(&p.R14).Pointer()-reflect.ValueOf(p).Pointer()) diff --git a/pkg/sentry/platform/ring0/offsets_arm64.go b/pkg/sentry/platform/ring0/offsets_arm64.go index 057fb5c69..f3de962f0 100644 --- a/pkg/sentry/platform/ring0/offsets_arm64.go +++ b/pkg/sentry/platform/ring0/offsets_arm64.go @@ -20,7 +20,8 @@ import ( "fmt" "io" "reflect" - "syscall" + + "gvisor.dev/gvisor/pkg/sentry/arch" ) // Emit prints architecture-specific offsets. @@ -87,7 +88,7 @@ func Emit(w io.Writer) { fmt.Fprintf(w, "#define Syscall 0x%02x\n", Syscall) fmt.Fprintf(w, "#define VirtualizationException 0x%02x\n", VirtualizationException) - p := &syscall.PtraceRegs{} + p := &arch.Registers{} fmt.Fprintf(w, "\n// Ptrace registers.\n") fmt.Fprintf(w, "#define PTRACE_R0 0x%02x\n", reflect.ValueOf(&p.Regs[0]).Pointer()-reflect.ValueOf(p).Pointer()) fmt.Fprintf(w, "#define PTRACE_R1 0x%02x\n", reflect.ValueOf(&p.Regs[1]).Pointer()-reflect.ValueOf(p).Pointer()) diff --git a/pkg/sentry/syscalls/linux/sys_signal.go b/pkg/sentry/syscalls/linux/sys_signal.go index 7e1747a0c..582d37e03 100644 --- a/pkg/sentry/syscalls/linux/sys_signal.go +++ b/pkg/sentry/syscalls/linux/sys_signal.go @@ -355,7 +355,7 @@ func Pause(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall func RtSigpending(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { addr := args[0].Pointer() pending := t.PendingSignals() - _, err := t.CopyOut(addr, pending) + _, err := pending.CopyOut(t, addr) return 0, nil, err } @@ -392,7 +392,7 @@ func RtSigtimedwait(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne if siginfo != 0 { si.FixSignalCodeForUser() - if _, err := t.CopyOut(siginfo, si); err != nil { + if _, err := si.CopyOut(t, siginfo); err != nil { return 0, nil, err } } @@ -411,7 +411,7 @@ func RtSigqueueinfo(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne // same way), and that the code is in the allowed set. This same logic // appears below in RtSigtgqueueinfo and should be kept in sync. var info arch.SignalInfo - if _, err := t.CopyIn(infoAddr, &info); err != nil { + if _, err := info.CopyIn(t, infoAddr); err != nil { return 0, nil, err } info.Signo = int32(sig) @@ -455,7 +455,7 @@ func RtTgsigqueueinfo(t *kernel.Task, args arch.SyscallArguments) (uintptr, *ker // Copy in the info. See RtSigqueueinfo above. var info arch.SignalInfo - if _, err := t.CopyIn(infoAddr, &info); err != nil { + if _, err := info.CopyIn(t, infoAddr); err != nil { return 0, nil, err } info.Signo = int32(sig) @@ -485,7 +485,7 @@ func RtSigsuspend(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. // Copy in the signal mask. var mask linux.SignalSet - if _, err := t.CopyIn(sigset, &mask); err != nil { + if _, err := mask.CopyIn(t, sigset); err != nil { return 0, nil, err } mask &^= kernel.UnblockableSignals -- cgit v1.2.3 From 292f3f99b73fb901ffdd3ad8ac682718e1e8960a Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Mon, 27 Apr 2020 07:37:45 -0700 Subject: Don't leak vfs.MountNamespace reference if kernel.TaskSet.NewTask fails. PiperOrigin-RevId: 308617610 --- pkg/sentry/kernel/task_start.go | 3 +++ 1 file changed, 3 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/task_start.go b/pkg/sentry/kernel/task_start.go index a5035bb7f..8485fb4b6 100644 --- a/pkg/sentry/kernel/task_start.go +++ b/pkg/sentry/kernel/task_start.go @@ -104,6 +104,9 @@ func (ts *TaskSet) NewTask(cfg *TaskConfig) (*Task, error) { cfg.TaskContext.release() cfg.FSContext.DecRef() cfg.FDTable.DecRef() + if cfg.MountNamespaceVFS2 != nil { + cfg.MountNamespaceVFS2.DecRef() + } return nil, err } return t, nil -- cgit v1.2.3 From 55f0c3316af8ea2a1fcc16511efc580f307623f6 Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Mon, 27 Apr 2020 12:25:10 -0700 Subject: Automated rollback of changelist 308163542 PiperOrigin-RevId: 308674219 --- pkg/sentry/socket/netfilter/tcp_matcher.go | 5 +- pkg/sentry/socket/netfilter/udp_matcher.go | 5 +- pkg/tcpip/buffer/view.go | 55 ++++---------- pkg/tcpip/buffer/view_test.go | 113 ----------------------------- pkg/tcpip/link/loopback/loopback.go | 10 +-- pkg/tcpip/link/rawfile/BUILD | 9 +-- pkg/tcpip/link/rawfile/rawfile_test.go | 46 ------------ pkg/tcpip/link/rawfile/rawfile_unsafe.go | 6 +- pkg/tcpip/link/sharedmem/sharedmem_test.go | 2 +- pkg/tcpip/link/sniffer/sniffer.go | 65 ++++------------- pkg/tcpip/network/arp/arp.go | 5 +- pkg/tcpip/network/ipv4/icmp.go | 20 ++--- pkg/tcpip/network/ipv4/ipv4.go | 12 +-- pkg/tcpip/network/ipv6/icmp.go | 74 +++++++------------ pkg/tcpip/network/ipv6/icmp_test.go | 3 +- pkg/tcpip/network/ipv6/ipv6.go | 6 +- pkg/tcpip/stack/forwarder_test.go | 13 ++-- pkg/tcpip/stack/iptables.go | 22 +----- pkg/tcpip/stack/iptables_targets.go | 23 ++---- pkg/tcpip/stack/nic.go | 34 ++++++--- pkg/tcpip/stack/packet_buffer.go | 8 +- pkg/tcpip/stack/stack_test.go | 10 +-- pkg/tcpip/stack/transport_test.go | 5 +- pkg/tcpip/transport/icmp/endpoint.go | 8 +- pkg/tcpip/transport/tcp/segment.go | 29 +++----- pkg/tcpip/transport/tcp/tcp_test.go | 4 +- pkg/tcpip/transport/udp/endpoint.go | 6 +- pkg/tcpip/transport/udp/protocol.go | 9 +-- 28 files changed, 149 insertions(+), 458 deletions(-) delete mode 100644 pkg/tcpip/link/rawfile/rawfile_test.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/tcp_matcher.go b/pkg/sentry/socket/netfilter/tcp_matcher.go index 55c0f04f3..ff1cfd8f6 100644 --- a/pkg/sentry/socket/netfilter/tcp_matcher.go +++ b/pkg/sentry/socket/netfilter/tcp_matcher.go @@ -121,13 +121,12 @@ func (tm *TCPMatcher) Match(hook stack.Hook, pkt stack.PacketBuffer, interfaceNa tcpHeader = header.TCP(pkt.TransportHeader) } else { // The TCP header hasn't been parsed yet. We have to do it here. - hdr, ok := pkt.Data.PullUp(header.TCPMinimumSize) - if !ok { + if len(pkt.Data.First()) < header.TCPMinimumSize { // There's no valid TCP header here, so we hotdrop the // packet. return false, true } - tcpHeader = header.TCP(hdr) + tcpHeader = header.TCP(pkt.Data.First()) } // Check whether the source and destination ports are within the diff --git a/pkg/sentry/socket/netfilter/udp_matcher.go b/pkg/sentry/socket/netfilter/udp_matcher.go index 04d03d494..3359418c1 100644 --- a/pkg/sentry/socket/netfilter/udp_matcher.go +++ b/pkg/sentry/socket/netfilter/udp_matcher.go @@ -120,13 +120,12 @@ func (um *UDPMatcher) Match(hook stack.Hook, pkt stack.PacketBuffer, interfaceNa udpHeader = header.UDP(pkt.TransportHeader) } else { // The UDP header hasn't been parsed yet. We have to do it here. - hdr, ok := pkt.Data.PullUp(header.UDPMinimumSize) - if !ok { + if len(pkt.Data.First()) < header.UDPMinimumSize { // There's no valid UDP header here, so we hotdrop the // packet. return false, true } - udpHeader = header.UDP(hdr) + udpHeader = header.UDP(pkt.Data.First()) } // Check whether the source and destination ports are within the diff --git a/pkg/tcpip/buffer/view.go b/pkg/tcpip/buffer/view.go index f01217c91..8ec5d5d5c 100644 --- a/pkg/tcpip/buffer/view.go +++ b/pkg/tcpip/buffer/view.go @@ -77,8 +77,7 @@ func NewVectorisedView(size int, views []View) VectorisedView { return VectorisedView{views: views, size: size} } -// TrimFront removes the first "count" bytes of the vectorised view. It panics -// if count > vv.Size(). +// TrimFront removes the first "count" bytes of the vectorised view. func (vv *VectorisedView) TrimFront(count int) { for count > 0 && len(vv.views) > 0 { if count < len(vv.views[0]) { @@ -87,7 +86,7 @@ func (vv *VectorisedView) TrimFront(count int) { return } count -= len(vv.views[0]) - vv.removeFirst() + vv.RemoveFirst() } } @@ -105,7 +104,7 @@ func (vv *VectorisedView) Read(v View) (copied int, err error) { count -= len(vv.views[0]) copy(v[copied:], vv.views[0]) copied += len(vv.views[0]) - vv.removeFirst() + vv.RemoveFirst() } if copied == 0 { return 0, io.EOF @@ -127,7 +126,7 @@ func (vv *VectorisedView) ReadToVV(dstVV *VectorisedView, count int) (copied int count -= len(vv.views[0]) dstVV.AppendView(vv.views[0]) copied += len(vv.views[0]) - vv.removeFirst() + vv.RemoveFirst() } return copied } @@ -163,37 +162,22 @@ func (vv *VectorisedView) Clone(buffer []View) VectorisedView { return VectorisedView{views: append(buffer[:0], vv.views...), size: vv.size} } -// PullUp returns the first "count" bytes of the vectorised view. If those -// bytes aren't already contiguous inside the vectorised view, PullUp will -// reallocate as needed to make them contiguous. PullUp fails and returns false -// when count > vv.Size(). -func (vv *VectorisedView) PullUp(count int) (View, bool) { +// First returns the first view of the vectorised view. +func (vv *VectorisedView) First() View { if len(vv.views) == 0 { - return nil, count == 0 - } - if count <= len(vv.views[0]) { - return vv.views[0][:count], true - } - if count > vv.size { - return nil, false + return nil } + return vv.views[0] +} - newFirst := NewView(count) - i := 0 - for offset := 0; offset < count; i++ { - copy(newFirst[offset:], vv.views[i]) - if count-offset < len(vv.views[i]) { - vv.views[i].TrimFront(count - offset) - break - } - offset += len(vv.views[i]) - vv.views[i] = nil +// RemoveFirst removes the first view of the vectorised view. +func (vv *VectorisedView) RemoveFirst() { + if len(vv.views) == 0 { + return } - // We're guaranteed that i > 0, since count is too large for the first - // view. - vv.views[i-1] = newFirst - vv.views = vv.views[i-1:] - return newFirst, true + vv.size -= len(vv.views[0]) + vv.views[0] = nil + vv.views = vv.views[1:] } // Size returns the size in bytes of the entire content stored in the vectorised view. @@ -241,10 +225,3 @@ func (vv *VectorisedView) Readers() []bytes.Reader { } return readers } - -// removeFirst panics when len(vv.views) < 1. -func (vv *VectorisedView) removeFirst() { - vv.size -= len(vv.views[0]) - vv.views[0] = nil - vv.views = vv.views[1:] -} diff --git a/pkg/tcpip/buffer/view_test.go b/pkg/tcpip/buffer/view_test.go index c56795c7b..106e1994c 100644 --- a/pkg/tcpip/buffer/view_test.go +++ b/pkg/tcpip/buffer/view_test.go @@ -16,7 +16,6 @@ package buffer import ( - "bytes" "reflect" "testing" ) @@ -371,115 +370,3 @@ func TestVVRead(t *testing.T) { }) } } - -var pullUpTestCases = []struct { - comment string - in VectorisedView - count int - want []byte - result VectorisedView - ok bool -}{ - { - comment: "simple case", - in: vv(2, "12"), - count: 1, - want: []byte("1"), - result: vv(2, "12"), - ok: true, - }, - { - comment: "entire View", - in: vv(2, "1", "2"), - count: 1, - want: []byte("1"), - result: vv(2, "1", "2"), - ok: true, - }, - { - comment: "spanning across two Views", - in: vv(3, "1", "23"), - count: 2, - want: []byte("12"), - result: vv(3, "12", "3"), - ok: true, - }, - { - comment: "spanning across all Views", - in: vv(5, "1", "23", "45"), - count: 5, - want: []byte("12345"), - result: vv(5, "12345"), - ok: true, - }, - { - comment: "count = 0", - in: vv(1, "1"), - count: 0, - want: []byte{}, - result: vv(1, "1"), - ok: true, - }, - { - comment: "count = size", - in: vv(1, "1"), - count: 1, - want: []byte("1"), - result: vv(1, "1"), - ok: true, - }, - { - comment: "count too large", - in: vv(3, "1", "23"), - count: 4, - want: nil, - result: vv(3, "1", "23"), - ok: false, - }, - { - comment: "empty vv", - in: vv(0, ""), - count: 1, - want: nil, - result: vv(0, ""), - ok: false, - }, - { - comment: "empty vv, count = 0", - in: vv(0, ""), - count: 0, - want: nil, - result: vv(0, ""), - ok: true, - }, - { - comment: "empty views", - in: vv(3, "", "1", "", "23"), - count: 2, - want: []byte("12"), - result: vv(3, "12", "3"), - ok: true, - }, -} - -func TestPullUp(t *testing.T) { - for _, c := range pullUpTestCases { - got, ok := c.in.PullUp(c.count) - - // Is the return value right? - if ok != c.ok { - t.Errorf("Test %q failed when calling PullUp(%d) on %v. Got an ok of %t. Want %t", - c.comment, c.count, c.in, ok, c.ok) - } - if bytes.Compare(got, View(c.want)) != 0 { - t.Errorf("Test %q failed when calling PullUp(%d) on %v. Got %v. Want %v", - c.comment, c.count, c.in, got, c.want) - } - - // Is the underlying structure right? - if !reflect.DeepEqual(c.in, c.result) { - t.Errorf("Test %q failed when calling PullUp(%d). Got vv with structure %v. Wanted %v", - c.comment, c.count, c.in, c.result) - } - } -} diff --git a/pkg/tcpip/link/loopback/loopback.go b/pkg/tcpip/link/loopback/loopback.go index 073c84ef9..1e2255bfa 100644 --- a/pkg/tcpip/link/loopback/loopback.go +++ b/pkg/tcpip/link/loopback/loopback.go @@ -98,13 +98,13 @@ func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, stack.PacketBufferList // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { - // There should be an ethernet header at the beginning of vv. - hdr, ok := vv.PullUp(header.EthernetMinimumSize) - if !ok { - // Reject the packet if it's shorter than an ethernet header. + // Reject the packet if it's shorter than an ethernet header. + if vv.Size() < header.EthernetMinimumSize { return tcpip.ErrBadAddress } - linkHeader := header.Ethernet(hdr) + + // There should be an ethernet header at the beginning of vv. + linkHeader := header.Ethernet(vv.First()[:header.EthernetMinimumSize]) vv.TrimFront(len(linkHeader)) e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, linkHeader.Type(), stack.PacketBuffer{ Data: vv, diff --git a/pkg/tcpip/link/rawfile/BUILD b/pkg/tcpip/link/rawfile/BUILD index 9cc08d0e2..14b527bc2 100644 --- a/pkg/tcpip/link/rawfile/BUILD +++ b/pkg/tcpip/link/rawfile/BUILD @@ -1,4 +1,4 @@ -load("//tools:defs.bzl", "go_library", "go_test") +load("//tools:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -18,10 +18,3 @@ go_library( "@org_golang_x_sys//unix:go_default_library", ], ) - -go_test( - name = "rawfile_test", - size = "small", - srcs = ["rawfile_test.go"], - library = ":rawfile", -) diff --git a/pkg/tcpip/link/rawfile/rawfile_test.go b/pkg/tcpip/link/rawfile/rawfile_test.go deleted file mode 100644 index 8f14ba761..000000000 --- a/pkg/tcpip/link/rawfile/rawfile_test.go +++ /dev/null @@ -1,46 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build linux - -package rawfile - -import ( - "syscall" - "testing" -) - -func TestNonBlockingWrite3ZeroLength(t *testing.T) { - fd, err := syscall.Open("/dev/null", syscall.O_WRONLY, 0) - if err != nil { - t.Fatalf("failed to open /dev/null: %v", err) - } - defer syscall.Close(fd) - - if err := NonBlockingWrite3(fd, []byte{}, []byte{0}, nil); err != nil { - t.Fatalf("failed to write: %v", err) - } -} - -func TestNonBlockingWrite3Nil(t *testing.T) { - fd, err := syscall.Open("/dev/null", syscall.O_WRONLY, 0) - if err != nil { - t.Fatalf("failed to open /dev/null: %v", err) - } - defer syscall.Close(fd) - - if err := NonBlockingWrite3(fd, nil, []byte{0}, nil); err != nil { - t.Fatalf("failed to write: %v", err) - } -} diff --git a/pkg/tcpip/link/rawfile/rawfile_unsafe.go b/pkg/tcpip/link/rawfile/rawfile_unsafe.go index 92efd0bf8..44e25d475 100644 --- a/pkg/tcpip/link/rawfile/rawfile_unsafe.go +++ b/pkg/tcpip/link/rawfile/rawfile_unsafe.go @@ -76,13 +76,9 @@ func NonBlockingWrite3(fd int, b1, b2, b3 []byte) *tcpip.Error { // We have two buffers. Build the iovec that represents them and issue // a writev syscall. - var base *byte - if len(b1) > 0 { - base = &b1[0] - } iovec := [3]syscall.Iovec{ { - Base: base, + Base: &b1[0], Len: uint64(len(b1)), }, { diff --git a/pkg/tcpip/link/sharedmem/sharedmem_test.go b/pkg/tcpip/link/sharedmem/sharedmem_test.go index 33f640b85..27ea3f531 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem_test.go +++ b/pkg/tcpip/link/sharedmem/sharedmem_test.go @@ -674,7 +674,7 @@ func TestSimpleReceive(t *testing.T) { // Wait for packet to be received, then check it. c.waitForPackets(1, time.After(5*time.Second), "Timeout waiting for packet") c.mu.Lock() - rcvd := []byte(c.packets[0].vv.ToView()) + rcvd := []byte(c.packets[0].vv.First()) c.packets = c.packets[:0] c.mu.Unlock() diff --git a/pkg/tcpip/link/sniffer/sniffer.go b/pkg/tcpip/link/sniffer/sniffer.go index 0799c8f4d..be2537a82 100644 --- a/pkg/tcpip/link/sniffer/sniffer.go +++ b/pkg/tcpip/link/sniffer/sniffer.go @@ -171,7 +171,11 @@ func (e *endpoint) GSOMaxSize() uint32 { func (e *endpoint) dumpPacket(prefix string, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { writer := e.writer if writer == nil && atomic.LoadUint32(&LogPackets) == 1 { - logPacket(prefix, protocol, pkt, gso) + first := pkt.Header.View() + if len(first) == 0 { + first = pkt.Data.First() + } + logPacket(prefix, protocol, first, gso) } if writer != nil && atomic.LoadUint32(&LogPacketsToPCAP) == 1 { totalLength := pkt.Header.UsedLength() + pkt.Data.Size() @@ -234,7 +238,7 @@ func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { // Wait implements stack.LinkEndpoint.Wait. func (e *endpoint) Wait() { e.lower.Wait() } -func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer, gso *stack.GSO) { +func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.View, gso *stack.GSO) { // Figure out the network layer info. var transProto uint8 src := tcpip.Address("unknown") @@ -243,49 +247,28 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.P size := uint16(0) var fragmentOffset uint16 var moreFragments bool - - // Create a clone of pkt, including any headers if present. Avoid allocating - // backing memory for the clone. - views := [8]buffer.View{} - vv := buffer.NewVectorisedView(0, views[:0]) - vv.AppendView(pkt.Header.View()) - vv.Append(pkt.Data) - switch protocol { case header.IPv4ProtocolNumber: - hdr, ok := vv.PullUp(header.IPv4MinimumSize) - if !ok { - return - } - ipv4 := header.IPv4(hdr) + ipv4 := header.IPv4(b) fragmentOffset = ipv4.FragmentOffset() moreFragments = ipv4.Flags()&header.IPv4FlagMoreFragments == header.IPv4FlagMoreFragments src = ipv4.SourceAddress() dst = ipv4.DestinationAddress() transProto = ipv4.Protocol() size = ipv4.TotalLength() - uint16(ipv4.HeaderLength()) - vv.TrimFront(int(ipv4.HeaderLength())) + b = b[ipv4.HeaderLength():] id = int(ipv4.ID()) case header.IPv6ProtocolNumber: - hdr, ok := vv.PullUp(header.IPv6MinimumSize) - if !ok { - return - } - ipv6 := header.IPv6(hdr) + ipv6 := header.IPv6(b) src = ipv6.SourceAddress() dst = ipv6.DestinationAddress() transProto = ipv6.NextHeader() size = ipv6.PayloadLength() - vv.TrimFront(header.IPv6MinimumSize) + b = b[header.IPv6MinimumSize:] case header.ARPProtocolNumber: - hdr, ok := vv.PullUp(header.ARPSize) - if !ok { - return - } - vv.TrimFront(header.ARPSize) - arp := header.ARP(hdr) + arp := header.ARP(b) log.Infof( "%s arp %v (%v) -> %v (%v) valid:%v", prefix, @@ -301,7 +284,7 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.P // We aren't guaranteed to have a transport header - it's possible for // writes via raw endpoints to contain only network headers. - if minSize, ok := transportProtocolMinSizes[tcpip.TransportProtocolNumber(transProto)]; ok && vv.Size() < minSize { + if minSize, ok := transportProtocolMinSizes[tcpip.TransportProtocolNumber(transProto)]; ok && len(b) < minSize { log.Infof("%s %v -> %v transport protocol: %d, but no transport header found (possible raw packet)", prefix, src, dst, transProto) return } @@ -314,11 +297,7 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.P switch tcpip.TransportProtocolNumber(transProto) { case header.ICMPv4ProtocolNumber: transName = "icmp" - hdr, ok := vv.PullUp(header.ICMPv4MinimumSize) - if !ok { - break - } - icmp := header.ICMPv4(hdr) + icmp := header.ICMPv4(b) icmpType := "unknown" if fragmentOffset == 0 { switch icmp.Type() { @@ -351,11 +330,7 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.P case header.ICMPv6ProtocolNumber: transName = "icmp" - hdr, ok := vv.PullUp(header.ICMPv6MinimumSize) - if !ok { - break - } - icmp := header.ICMPv6(hdr) + icmp := header.ICMPv6(b) icmpType := "unknown" switch icmp.Type() { case header.ICMPv6DstUnreachable: @@ -386,11 +361,7 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.P case header.UDPProtocolNumber: transName = "udp" - hdr, ok := vv.PullUp(header.UDPMinimumSize) - if !ok { - break - } - udp := header.UDP(hdr) + udp := header.UDP(b) if fragmentOffset == 0 && len(udp) >= header.UDPMinimumSize { srcPort = udp.SourcePort() dstPort = udp.DestinationPort() @@ -400,11 +371,7 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.P case header.TCPProtocolNumber: transName = "tcp" - hdr, ok := vv.PullUp(header.TCPMinimumSize) - if !ok { - break - } - tcp := header.TCP(hdr) + tcp := header.TCP(b) if fragmentOffset == 0 && len(tcp) >= header.TCPMinimumSize { offset := int(tcp.DataOffset()) if offset < header.TCPMinimumSize { diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index cf73a939e..7acbfa0a8 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -93,10 +93,7 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuf } func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { - v, ok := pkt.Data.PullUp(header.ARPSize) - if !ok { - return - } + v := pkt.Data.First() h := header.ARP(v) if !h.IsValid() { return diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go index 4cbefe5ab..c4bf1ba5c 100644 --- a/pkg/tcpip/network/ipv4/icmp.go +++ b/pkg/tcpip/network/ipv4/icmp.go @@ -25,11 +25,7 @@ import ( // used to find out which transport endpoint must be notified about the ICMP // packet. func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack.PacketBuffer) { - h, ok := pkt.Data.PullUp(header.IPv4MinimumSize) - if !ok { - return - } - hdr := header.IPv4(h) + h := header.IPv4(pkt.Data.First()) // We don't use IsValid() here because ICMP only requires that the IP // header plus 8 bytes of the transport header be included. So it's @@ -38,12 +34,12 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. // // Drop packet if it doesn't have the basic IPv4 header or if the // original source address doesn't match the endpoint's address. - if hdr.SourceAddress() != e.id.LocalAddress { + if len(h) < header.IPv4MinimumSize || h.SourceAddress() != e.id.LocalAddress { return } - hlen := int(hdr.HeaderLength()) - if pkt.Data.Size() < hlen || hdr.FragmentOffset() != 0 { + hlen := int(h.HeaderLength()) + if pkt.Data.Size() < hlen || h.FragmentOffset() != 0 { // We won't be able to handle this if it doesn't contain the // full IPv4 header, or if it's a fragment not at offset 0 // (because it won't have the transport header). @@ -52,15 +48,15 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. // Skip the ip header, then deliver control message. pkt.Data.TrimFront(hlen) - p := hdr.TransportProtocol() - e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, hdr.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) + p := h.TransportProtocol() + e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, h.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) } func (e *endpoint) handleICMP(r *stack.Route, pkt stack.PacketBuffer) { stats := r.Stats() received := stats.ICMP.V4PacketsReceived - v, ok := pkt.Data.PullUp(header.ICMPv4MinimumSize) - if !ok { + v := pkt.Data.First() + if len(v) < header.ICMPv4MinimumSize { received.Invalid.Increment() return } diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index 17202cc7a..104aafbed 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -328,11 +328,7 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuffer) *tcpip.Error { // The packet already has an IP header, but there are a few required // checks. - h, ok := pkt.Data.PullUp(header.IPv4MinimumSize) - if !ok { - return tcpip.ErrInvalidOptionValue - } - ip := header.IPv4(h) + ip := header.IPv4(pkt.Data.First()) if !ip.IsValid(pkt.Data.Size()) { return tcpip.ErrInvalidOptionValue } @@ -382,11 +378,7 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuf // HandlePacket is called by the link layer when new ipv4 packets arrive for // this endpoint. func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { - headerView, ok := pkt.Data.PullUp(header.IPv4MinimumSize) - if !ok { - r.Stats().IP.MalformedPacketsReceived.Increment() - return - } + headerView := pkt.Data.First() h := header.IPv4(headerView) if !h.IsValid(pkt.Data.Size()) { r.Stats().IP.MalformedPacketsReceived.Increment() diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index bdf3a0d25..b68983d10 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -28,11 +28,7 @@ import ( // used to find out which transport endpoint must be notified about the ICMP // packet. func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack.PacketBuffer) { - h, ok := pkt.Data.PullUp(header.IPv6MinimumSize) - if !ok { - return - } - hdr := header.IPv6(h) + h := header.IPv6(pkt.Data.First()) // We don't use IsValid() here because ICMP only requires that up to // 1280 bytes of the original packet be included. So it's likely that it @@ -40,21 +36,17 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. // // Drop packet if it doesn't have the basic IPv6 header or if the // original source address doesn't match the endpoint's address. - if hdr.SourceAddress() != e.id.LocalAddress { + if len(h) < header.IPv6MinimumSize || h.SourceAddress() != e.id.LocalAddress { return } // Skip the IP header, then handle the fragmentation header if there // is one. pkt.Data.TrimFront(header.IPv6MinimumSize) - p := hdr.TransportProtocol() + p := h.TransportProtocol() if p == header.IPv6FragmentHeader { - f, ok := pkt.Data.PullUp(header.IPv6FragmentHeaderSize) - if !ok { - return - } - fragHdr := header.IPv6Fragment(f) - if !fragHdr.IsValid() || fragHdr.FragmentOffset() != 0 { + f := header.IPv6Fragment(pkt.Data.First()) + if !f.IsValid() || f.FragmentOffset() != 0 { // We can't handle fragments that aren't at offset 0 // because they don't have the transport headers. return @@ -63,19 +55,19 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. // Skip fragmentation header and find out the actual protocol // number. pkt.Data.TrimFront(header.IPv6FragmentHeaderSize) - p = fragHdr.TransportProtocol() + p = f.TransportProtocol() } // Deliver the control packet to the transport endpoint. - e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, hdr.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) + e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, h.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) } func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.PacketBuffer, hasFragmentHeader bool) { stats := r.Stats().ICMP sent := stats.V6PacketsSent received := stats.V6PacketsReceived - v, ok := pkt.Data.PullUp(header.ICMPv6HeaderSize) - if !ok { + v := pkt.Data.First() + if len(v) < header.ICMPv6MinimumSize { received.Invalid.Increment() return } @@ -84,9 +76,11 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P // Validate ICMPv6 checksum before processing the packet. // + // Only the first view in vv is accounted for by h. To account for the + // rest of vv, a shallow copy is made and the first view is removed. // This copy is used as extra payload during the checksum calculation. payload := pkt.Data.Clone(nil) - payload.TrimFront(len(h)) + payload.RemoveFirst() if got, want := h.Checksum(), header.ICMPv6Checksum(h, iph.SourceAddress(), iph.DestinationAddress(), payload); got != want { received.Invalid.Increment() return @@ -107,40 +101,34 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P switch h.Type() { case header.ICMPv6PacketTooBig: received.PacketTooBig.Increment() - hdr, ok := pkt.Data.PullUp(header.ICMPv6PacketTooBigMinimumSize) - if !ok { + if len(v) < header.ICMPv6PacketTooBigMinimumSize { received.Invalid.Increment() return } pkt.Data.TrimFront(header.ICMPv6PacketTooBigMinimumSize) - mtu := header.ICMPv6(hdr).MTU() + mtu := h.MTU() e.handleControl(stack.ControlPacketTooBig, calculateMTU(mtu), pkt) case header.ICMPv6DstUnreachable: received.DstUnreachable.Increment() - hdr, ok := pkt.Data.PullUp(header.ICMPv6DstUnreachableMinimumSize) - if !ok { + if len(v) < header.ICMPv6DstUnreachableMinimumSize { received.Invalid.Increment() return } pkt.Data.TrimFront(header.ICMPv6DstUnreachableMinimumSize) - switch header.ICMPv6(hdr).Code() { + switch h.Code() { case header.ICMPv6PortUnreachable: e.handleControl(stack.ControlPortUnreachable, 0, pkt) } case header.ICMPv6NeighborSolicit: received.NeighborSolicit.Increment() - if pkt.Data.Size() < header.ICMPv6NeighborSolicitMinimumSize || !isNDPValid() { + if len(v) < header.ICMPv6NeighborSolicitMinimumSize || !isNDPValid() { received.Invalid.Increment() return } - // The remainder of payload must be only the neighbor solicitation, so - // payload.ToView() always returns the solicitation. Per RFC 6980 section 5, - // NDP messages cannot be fragmented. Also note that in the common case NDP - // datagrams are very small and ToView() will not incur allocations. - ns := header.NDPNeighborSolicit(payload.ToView()) + ns := header.NDPNeighborSolicit(h.NDPPayload()) it, err := ns.Options().Iter(true) if err != nil { // If we have a malformed NDP NS option, drop the packet. @@ -298,16 +286,12 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P case header.ICMPv6NeighborAdvert: received.NeighborAdvert.Increment() - if pkt.Data.Size() < header.ICMPv6NeighborAdvertSize || !isNDPValid() { + if len(v) < header.ICMPv6NeighborAdvertSize || !isNDPValid() { received.Invalid.Increment() return } - // The remainder of payload must be only the neighbor advertisement, so - // payload.ToView() always returns the advertisement. Per RFC 6980 section - // 5, NDP messages cannot be fragmented. Also note that in the common case - // NDP datagrams are very small and ToView() will not incur allocations. - na := header.NDPNeighborAdvert(payload.ToView()) + na := header.NDPNeighborAdvert(h.NDPPayload()) it, err := na.Options().Iter(true) if err != nil { // If we have a malformed NDP NA option, drop the packet. @@ -379,15 +363,14 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P case header.ICMPv6EchoRequest: received.EchoRequest.Increment() - icmpHdr, ok := pkt.Data.PullUp(header.ICMPv6EchoMinimumSize) - if !ok { + if len(v) < header.ICMPv6EchoMinimumSize { received.Invalid.Increment() return } pkt.Data.TrimFront(header.ICMPv6EchoMinimumSize) hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv6EchoMinimumSize) packet := header.ICMPv6(hdr.Prepend(header.ICMPv6EchoMinimumSize)) - copy(packet, icmpHdr) + copy(packet, h) packet.SetType(header.ICMPv6EchoReply) packet.SetChecksum(header.ICMPv6Checksum(packet, r.LocalAddress, r.RemoteAddress, pkt.Data)) if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, stack.PacketBuffer{ @@ -401,7 +384,7 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P case header.ICMPv6EchoReply: received.EchoReply.Increment() - if pkt.Data.Size() < header.ICMPv6EchoMinimumSize { + if len(v) < header.ICMPv6EchoMinimumSize { received.Invalid.Increment() return } @@ -423,9 +406,8 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P case header.ICMPv6RouterAdvert: received.RouterAdvert.Increment() - // Is the NDP payload of sufficient size to hold a Router - // Advertisement? - if pkt.Data.Size()-header.ICMPv6HeaderSize < header.NDPRAMinimumSize || !isNDPValid() { + p := h.NDPPayload() + if len(p) < header.NDPRAMinimumSize || !isNDPValid() { received.Invalid.Increment() return } @@ -443,11 +425,7 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P return } - // The remainder of payload must be only the router advertisement, so - // payload.ToView() always returns the advertisement. Per RFC 6980 section - // 5, NDP messages cannot be fragmented. Also note that in the common case - // NDP datagrams are very small and ToView() will not incur allocations. - ra := header.NDPRouterAdvert(payload.ToView()) + ra := header.NDPRouterAdvert(p) opts := ra.Options() // Are options valid as per the wire format? diff --git a/pkg/tcpip/network/ipv6/icmp_test.go b/pkg/tcpip/network/ipv6/icmp_test.go index d412ff688..bd099a7f8 100644 --- a/pkg/tcpip/network/ipv6/icmp_test.go +++ b/pkg/tcpip/network/ipv6/icmp_test.go @@ -166,8 +166,7 @@ func TestICMPCounts(t *testing.T) { }, { typ: header.ICMPv6NeighborSolicit, - size: header.ICMPv6NeighborSolicitMinimumSize, - }, + size: header.ICMPv6NeighborSolicitMinimumSize}, { typ: header.ICMPv6NeighborAdvert, size: header.ICMPv6NeighborAdvertMinimumSize, diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index 486725131..331b0817b 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -171,11 +171,7 @@ func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuffe // HandlePacket is called by the link layer when new ipv6 packets arrive for // this endpoint. func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { - headerView, ok := pkt.Data.PullUp(header.IPv6MinimumSize) - if !ok { - r.Stats().IP.MalformedPacketsReceived.Increment() - return - } + headerView := pkt.Data.First() h := header.IPv6(headerView) if !h.IsValid(pkt.Data.Size()) { r.Stats().IP.MalformedPacketsReceived.Increment() diff --git a/pkg/tcpip/stack/forwarder_test.go b/pkg/tcpip/stack/forwarder_test.go index c7c663498..e9c652042 100644 --- a/pkg/tcpip/stack/forwarder_test.go +++ b/pkg/tcpip/stack/forwarder_test.go @@ -70,10 +70,7 @@ func (f *fwdTestNetworkEndpoint) ID() *NetworkEndpointID { func (f *fwdTestNetworkEndpoint) HandlePacket(r *Route, pkt PacketBuffer) { // Consume the network header. - b, ok := pkt.Data.PullUp(fwdTestNetHeaderLen) - if !ok { - return - } + b := pkt.Data.First() pkt.Data.TrimFront(fwdTestNetHeaderLen) // Dispatch the packet to the transport protocol. @@ -476,7 +473,7 @@ func TestForwardingWithFakeResolverPartialTimeout(t *testing.T) { t.Fatal("packet not forwarded") } - b := p.Pkt.Data.ToView() + b := p.Pkt.Header.View() if b[0] != 3 { t.Fatalf("got b[0] = %d, want = 3", b[0]) } @@ -520,7 +517,7 @@ func TestForwardingWithFakeResolverTwoPackets(t *testing.T) { t.Fatal("packet not forwarded") } - b := p.Pkt.Data.ToView() + b := p.Pkt.Header.View() if b[0] != 3 { t.Fatalf("got b[0] = %d, want = 3", b[0]) } @@ -567,7 +564,7 @@ func TestForwardingWithFakeResolverManyPackets(t *testing.T) { t.Fatal("packet not forwarded") } - b := p.Pkt.Data.ToView() + b := p.Pkt.Header.View() if b[0] != 3 { t.Fatalf("got b[0] = %d, want = 3", b[0]) } @@ -622,7 +619,7 @@ func TestForwardingWithFakeResolverManyResolutions(t *testing.T) { // The first 5 packets (address 3 to 7) should not be forwarded // because their address resolutions are interrupted. - b := p.Pkt.Data.ToView() + b := p.Pkt.Header.View() if b[0] < 8 { t.Fatalf("got b[0] = %d, want b[0] >= 8", b[0]) } diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go index 6b91159d4..6c0a4b24d 100644 --- a/pkg/tcpip/stack/iptables.go +++ b/pkg/tcpip/stack/iptables.go @@ -212,11 +212,6 @@ func (it *IPTables) Check(hook Hook, pkt PacketBuffer) bool { // CheckPackets runs pkts through the rules for hook and returns a map of packets that // should not go forward. // -// Precondition: pkt is a IPv4 packet of at least length header.IPv4MinimumSize. -// -// TODO(gvisor.dev/issue/170): pk.NetworkHeader will always be set as a -// precondition. -// // NOTE: unlike the Check API the returned map contains packets that should be // dropped. func (it *IPTables) CheckPackets(hook Hook, pkts PacketBufferList) (drop map[*PacketBuffer]struct{}) { @@ -231,9 +226,7 @@ func (it *IPTables) CheckPackets(hook Hook, pkts PacketBufferList) (drop map[*Pa return drop } -// Precondition: pkt is a IPv4 packet of at least length header.IPv4MinimumSize. -// TODO(gvisor.dev/issue/170): pk.NetworkHeader will always be set as a -// precondition. +// Precondition: pkt.NetworkHeader is set. func (it *IPTables) checkChain(hook Hook, pkt PacketBuffer, table Table, ruleIdx int) chainVerdict { // Start from ruleIdx and walk the list of rules until a rule gives us // a verdict. @@ -278,21 +271,14 @@ func (it *IPTables) checkChain(hook Hook, pkt PacketBuffer, table Table, ruleIdx return chainDrop } -// Precondition: pkt is a IPv4 packet of at least length header.IPv4MinimumSize. -// TODO(gvisor.dev/issue/170): pk.NetworkHeader will always be set as a -// precondition. +// Precondition: pk.NetworkHeader is set. func (it *IPTables) checkRule(hook Hook, pkt PacketBuffer, table Table, ruleIdx int) (RuleVerdict, int) { rule := table.Rules[ruleIdx] // If pkt.NetworkHeader hasn't been set yet, it will be contained in - // pkt.Data. + // pkt.Data.First(). if pkt.NetworkHeader == nil { - var ok bool - pkt.NetworkHeader, ok = pkt.Data.PullUp(header.IPv4MinimumSize) - if !ok { - // Precondition has been violated. - panic(fmt.Sprintf("iptables checks require IPv4 headers of at least %d bytes", header.IPv4MinimumSize)) - } + pkt.NetworkHeader = pkt.Data.First() } // Check whether the packet matches the IP header filter. diff --git a/pkg/tcpip/stack/iptables_targets.go b/pkg/tcpip/stack/iptables_targets.go index 8be61f4b1..7b4543caf 100644 --- a/pkg/tcpip/stack/iptables_targets.go +++ b/pkg/tcpip/stack/iptables_targets.go @@ -96,12 +96,9 @@ func (rt RedirectTarget) Action(pkt PacketBuffer) (RuleVerdict, int) { newPkt := pkt.Clone() // Set network header. - headerView, ok := newPkt.Data.PullUp(header.IPv4MinimumSize) - if !ok { - return RuleDrop, 0 - } + headerView := newPkt.Data.First() netHeader := header.IPv4(headerView) - newPkt.NetworkHeader = headerView + newPkt.NetworkHeader = headerView[:header.IPv4MinimumSize] hlen := int(netHeader.HeaderLength()) tlen := int(netHeader.TotalLength()) @@ -120,14 +117,10 @@ func (rt RedirectTarget) Action(pkt PacketBuffer) (RuleVerdict, int) { if newPkt.TransportHeader != nil { udpHeader = header.UDP(newPkt.TransportHeader) } else { - if pkt.Data.Size() < header.UDPMinimumSize { - return RuleDrop, 0 - } - hdr, ok := newPkt.Data.PullUp(header.UDPMinimumSize) - if !ok { + if len(pkt.Data.First()) < header.UDPMinimumSize { return RuleDrop, 0 } - udpHeader = header.UDP(hdr) + udpHeader = header.UDP(newPkt.Data.First()) } udpHeader.SetDestinationPort(rt.MinPort) case header.TCPProtocolNumber: @@ -135,14 +128,10 @@ func (rt RedirectTarget) Action(pkt PacketBuffer) (RuleVerdict, int) { if newPkt.TransportHeader != nil { tcpHeader = header.TCP(newPkt.TransportHeader) } else { - if pkt.Data.Size() < header.TCPMinimumSize { + if len(pkt.Data.First()) < header.TCPMinimumSize { return RuleDrop, 0 } - hdr, ok := newPkt.Data.PullUp(header.TCPMinimumSize) - if !ok { - return RuleDrop, 0 - } - tcpHeader = header.TCP(hdr) + tcpHeader = header.TCP(newPkt.TransportHeader) } // TODO(gvisor.dev/issue/170): Need to recompute checksum // and implement nat connection tracking to support TCP. diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 0c2b1f36a..016dbe15e 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -1203,12 +1203,12 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link n.stack.stats.IP.PacketsReceived.Increment() } - netHeader, ok := pkt.Data.PullUp(netProto.MinimumPacketSize()) - if !ok { + if len(pkt.Data.First()) < netProto.MinimumPacketSize() { n.stack.stats.MalformedRcvdPackets.Increment() return } - src, dst := netProto.ParseAddresses(netHeader) + + src, dst := netProto.ParseAddresses(pkt.Data.First()) if n.stack.handleLocal && !n.isLoopback() && n.getRef(protocol, src) != nil { // The source address is one of our own, so we never should have gotten a @@ -1289,8 +1289,22 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link func (n *NIC) forwardPacket(r *Route, protocol tcpip.NetworkProtocolNumber, pkt PacketBuffer) { // TODO(b/143425874) Decrease the TTL field in forwarded packets. - if linkHeaderLen := int(n.linkEP.MaxHeaderLength()); linkHeaderLen != 0 { - pkt.Header = buffer.NewPrependable(linkHeaderLen) + + firstData := pkt.Data.First() + pkt.Data.RemoveFirst() + + if linkHeaderLen := int(n.linkEP.MaxHeaderLength()); linkHeaderLen == 0 { + pkt.Header = buffer.NewPrependableFromView(firstData) + } else { + firstDataLen := len(firstData) + + // pkt.Header should have enough capacity to hold n.linkEP's headers. + pkt.Header = buffer.NewPrependable(firstDataLen + linkHeaderLen) + + // TODO(b/151227689): avoid copying the packet when forwarding + if n := copy(pkt.Header.Prepend(firstDataLen), firstData); n != firstDataLen { + panic(fmt.Sprintf("copied %d bytes, expected %d", n, firstDataLen)) + } } if err := n.linkEP.WritePacket(r, nil /* gso */, protocol, pkt); err != nil { @@ -1318,13 +1332,12 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN // validly formed. n.stack.demux.deliverRawPacket(r, protocol, pkt) - transHeader, ok := pkt.Data.PullUp(transProto.MinimumPacketSize()) - if !ok { + if len(pkt.Data.First()) < transProto.MinimumPacketSize() { n.stack.stats.MalformedRcvdPackets.Increment() return } - srcPort, dstPort, err := transProto.ParsePorts(transHeader) + srcPort, dstPort, err := transProto.ParsePorts(pkt.Data.First()) if err != nil { n.stack.stats.MalformedRcvdPackets.Increment() return @@ -1362,12 +1375,11 @@ func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcp // ICMPv4 only guarantees that 8 bytes of the transport protocol will // be present in the payload. We know that the ports are within the // first 8 bytes for all known transport protocols. - transHeader, ok := pkt.Data.PullUp(8) - if !ok { + if len(pkt.Data.First()) < 8 { return } - srcPort, dstPort, err := transProto.ParsePorts(transHeader) + srcPort, dstPort, err := transProto.ParsePorts(pkt.Data.First()) if err != nil { return } diff --git a/pkg/tcpip/stack/packet_buffer.go b/pkg/tcpip/stack/packet_buffer.go index 7d36f8e84..dc125f25e 100644 --- a/pkg/tcpip/stack/packet_buffer.go +++ b/pkg/tcpip/stack/packet_buffer.go @@ -37,13 +37,7 @@ type PacketBuffer struct { Data buffer.VectorisedView // Header holds the headers of outbound packets. As a packet is passed - // down the stack, each layer adds to Header. Note that forwarded - // packets don't populate Headers on their way out -- their headers and - // payload are never parsed out and remain in Data. - // - // TODO(gvisor.dev/issue/170): Forwarded packets don't currently - // populate Header, but should. This will be doable once early parsing - // (https://github.com/google/gvisor/pull/1995) is supported. + // down the stack, each layer adds to Header. Header buffer.Prependable // These fields are used by both inbound and outbound packets. They diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index d45d2cc1f..c7634ceb1 100644 --- a/pkg/tcpip/stack/stack_test.go +++ b/pkg/tcpip/stack/stack_test.go @@ -95,18 +95,16 @@ func (f *fakeNetworkEndpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffe f.proto.packetCount[int(f.id.LocalAddress[0])%len(f.proto.packetCount)]++ // Consume the network header. - b, ok := pkt.Data.PullUp(fakeNetHeaderLen) - if !ok { - return - } + b := pkt.Data.First() pkt.Data.TrimFront(fakeNetHeaderLen) // Handle control packets. if b[2] == uint8(fakeControlProtocol) { - nb, ok := pkt.Data.PullUp(fakeNetHeaderLen) - if !ok { + nb := pkt.Data.First() + if len(nb) < fakeNetHeaderLen { return } + pkt.Data.TrimFront(fakeNetHeaderLen) f.dispatcher.DeliverTransportControlPacket(tcpip.Address(nb[1:2]), tcpip.Address(nb[0:1]), fakeNetNumber, tcpip.TransportProtocolNumber(nb[2]), stack.ControlPortUnreachable, 0, pkt) return diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go index a611e44ab..3084e6593 100644 --- a/pkg/tcpip/stack/transport_test.go +++ b/pkg/tcpip/stack/transport_test.go @@ -642,11 +642,10 @@ func TestTransportForwarding(t *testing.T) { t.Fatal("Response packet not forwarded") } - hdrs := p.Pkt.Data.ToView() - if dst := hdrs[0]; dst != 3 { + if dst := p.Pkt.Header.View()[0]; dst != 3 { t.Errorf("Response packet has incorrect destination addresss: got = %d, want = 3", dst) } - if src := hdrs[1]; src != 1 { + if src := p.Pkt.Header.View()[1]; src != 1 { t.Errorf("Response packet has incorrect source addresss: got = %d, want = 3", src) } } diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index b1d820372..feef8dca0 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -747,15 +747,15 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk // Only accept echo replies. switch e.NetProto { case header.IPv4ProtocolNumber: - h, ok := pkt.Data.PullUp(header.ICMPv4MinimumSize) - if !ok || header.ICMPv4(h).Type() != header.ICMPv4EchoReply { + h := header.ICMPv4(pkt.Data.First()) + if h.Type() != header.ICMPv4EchoReply { e.stack.Stats().DroppedPackets.Increment() e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() return } case header.IPv6ProtocolNumber: - h, ok := pkt.Data.PullUp(header.ICMPv6MinimumSize) - if !ok || header.ICMPv6(h).Type() != header.ICMPv6EchoReply { + h := header.ICMPv6(pkt.Data.First()) + if h.Type() != header.ICMPv6EchoReply { e.stack.Stats().DroppedPackets.Increment() e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() return diff --git a/pkg/tcpip/transport/tcp/segment.go b/pkg/tcpip/transport/tcp/segment.go index 7712ce652..40461fd31 100644 --- a/pkg/tcpip/transport/tcp/segment.go +++ b/pkg/tcpip/transport/tcp/segment.go @@ -144,11 +144,7 @@ func (s *segment) logicalLen() seqnum.Size { // TCP checksum and stores the checksum and result of checksum verification in // the csum and csumValid fields of the segment. func (s *segment) parse() bool { - h, ok := s.data.PullUp(header.TCPMinimumSize) - if !ok { - return false - } - hdr := header.TCP(h) + h := header.TCP(s.data.First()) // h is the header followed by the payload. We check that the offset to // the data respects the following constraints: @@ -160,16 +156,12 @@ func (s *segment) parse() bool { // N.B. The segment has already been validated as having at least the // minimum TCP size before reaching here, so it's safe to read the // fields. - offset := int(hdr.DataOffset()) - if offset < header.TCPMinimumSize { - return false - } - hdrWithOpts, ok := s.data.PullUp(offset) - if !ok { + offset := int(h.DataOffset()) + if offset < header.TCPMinimumSize || offset > len(h) { return false } - s.options = []byte(hdrWithOpts[header.TCPMinimumSize:]) + s.options = []byte(h[header.TCPMinimumSize:offset]) s.parsedOptions = header.ParseTCPOptions(s.options) // Query the link capabilities to decide if checksum validation is @@ -181,19 +173,18 @@ func (s *segment) parse() bool { s.data.TrimFront(offset) } if verifyChecksum { - hdr = header.TCP(hdrWithOpts) - s.csum = hdr.Checksum() + s.csum = h.Checksum() xsum := s.route.PseudoHeaderChecksum(ProtocolNumber, uint16(s.data.Size())) - xsum = hdr.CalculateChecksum(xsum) + xsum = h.CalculateChecksum(xsum) s.data.TrimFront(offset) xsum = header.ChecksumVV(s.data, xsum) s.csumValid = xsum == 0xffff } - s.sequenceNumber = seqnum.Value(hdr.SequenceNumber()) - s.ackNumber = seqnum.Value(hdr.AckNumber()) - s.flags = hdr.Flags() - s.window = seqnum.Size(hdr.WindowSize()) + s.sequenceNumber = seqnum.Value(h.SequenceNumber()) + s.ackNumber = seqnum.Value(h.AckNumber()) + s.flags = h.Flags() + s.window = seqnum.Size(h.WindowSize()) return true } diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index 286c66cf5..ab1014c7f 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -3548,7 +3548,7 @@ func TestReceivedInvalidSegmentCountIncrement(t *testing.T) { AckNum: c.IRS.Add(1), RcvWnd: 30000, }) - tcpbuf := vv.ToView()[header.IPv4MinimumSize:] + tcpbuf := vv.First()[header.IPv4MinimumSize:] tcpbuf[header.TCPDataOffset] = ((header.TCPMinimumSize - 1) / 4) << 4 c.SendSegment(vv) @@ -3575,7 +3575,7 @@ func TestReceivedIncorrectChecksumIncrement(t *testing.T) { AckNum: c.IRS.Add(1), RcvWnd: 30000, }) - tcpbuf := vv.ToView()[header.IPv4MinimumSize:] + tcpbuf := vv.First()[header.IPv4MinimumSize:] // Overwrite a byte in the payload which should cause checksum // verification to fail. tcpbuf[(tcpbuf[header.TCPDataOffset]>>4)*4] = 0x4 diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 756ab913a..edb54f0be 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -1250,8 +1250,8 @@ func (e *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { // endpoint. func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) { // Get the header then trim it from the view. - hdr, ok := pkt.Data.PullUp(header.UDPMinimumSize) - if !ok || int(header.UDP(hdr).Length()) > pkt.Data.Size() { + hdr := header.UDP(pkt.Data.First()) + if int(hdr.Length()) > pkt.Data.Size() { // Malformed packet. e.stack.Stats().UDP.MalformedPacketsReceived.Increment() e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() @@ -1286,7 +1286,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk senderAddress: tcpip.FullAddress{ NIC: r.NICID(), Addr: id.RemoteAddress, - Port: header.UDP(hdr).SourcePort(), + Port: hdr.SourcePort(), }, } packet.data = pkt.Data diff --git a/pkg/tcpip/transport/udp/protocol.go b/pkg/tcpip/transport/udp/protocol.go index 52af6de22..6e31a9bac 100644 --- a/pkg/tcpip/transport/udp/protocol.go +++ b/pkg/tcpip/transport/udp/protocol.go @@ -68,13 +68,8 @@ func (*protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) { // that don't match any existing endpoint. func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) bool { // Get the header then trim it from the view. - h, ok := pkt.Data.PullUp(header.UDPMinimumSize) - if !ok { - // Malformed packet. - r.Stack().Stats().UDP.MalformedPacketsReceived.Increment() - return true - } - if int(header.UDP(h).Length()) > pkt.Data.Size() { + hdr := header.UDP(pkt.Data.First()) + if int(hdr.Length()) > pkt.Data.Size() { // Malformed packet. r.Stack().Stats().UDP.MalformedPacketsReceived.Increment() return true -- cgit v1.2.3 From 003e79a6d11b17caed480c1ba556de5cb713abb3 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Mon, 27 Apr 2020 12:25:57 -0700 Subject: Dump stack for stuck start and stuck watchdog The meaning for skipDump was reversed, but not all callers were updated. Change the meaning once again to forceDump, so that the period between stack dump is respected from all callers. PiperOrigin-RevId: 308674373 --- pkg/sentry/watchdog/watchdog.go | 20 +++++++++++--------- 1 file changed, 11 insertions(+), 9 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/watchdog/watchdog.go b/pkg/sentry/watchdog/watchdog.go index fcc46420f..101497ed6 100644 --- a/pkg/sentry/watchdog/watchdog.go +++ b/pkg/sentry/watchdog/watchdog.go @@ -255,7 +255,7 @@ func (w *Watchdog) runTurn() { case <-done: case <-time.After(w.TaskTimeout): // Report if the watchdog is not making progress. - // No one is wathching the watchdog watcher though. + // No one is watching the watchdog watcher though. w.reportStuckWatchdog() <-done } @@ -317,10 +317,8 @@ func (w *Watchdog) report(offenders map[*kernel.Task]*offender, newTaskFound boo buf.WriteString("Search for '(*Task).run(0x..., 0x)' in the stack dump to find the offending goroutine") - // Dump stack only if a new task is detected or if it sometime has - // passed since the last time a stack dump was generated. - showStack := newTaskFound || time.Since(w.lastStackDump) >= stackDumpSameTaskPeriod - w.doAction(w.TaskTimeoutAction, showStack, &buf) + // Force stack dump only if a new task is detected. + w.doAction(w.TaskTimeoutAction, newTaskFound, &buf) } func (w *Watchdog) reportStuckWatchdog() { @@ -329,12 +327,15 @@ func (w *Watchdog) reportStuckWatchdog() { w.doAction(w.TaskTimeoutAction, false, &buf) } -// doAction will take the given action. If the action is LogWarning and -// showStack is false, then the stack printing will be skipped. -func (w *Watchdog) doAction(action Action, showStack bool, msg *bytes.Buffer) { +// doAction will take the given action. If the action is LogWarning, the stack +// is not always dumpped to the log to prevent log flooding. "forceStack" +// guarantees that the stack will be dumped regarless. +func (w *Watchdog) doAction(action Action, forceStack bool, msg *bytes.Buffer) { switch action { case LogWarning: - if !showStack { + // Dump stack only if forced or sometime has passed since the last time a + // stack dump was generated. + if !forceStack && time.Since(w.lastStackDump) < stackDumpSameTaskPeriod { msg.WriteString("\n...[stack dump skipped]...") log.Warningf(msg.String()) return @@ -359,6 +360,7 @@ func (w *Watchdog) doAction(action Action, showStack bool, msg *bytes.Buffer) { case <-time.After(1 * time.Second): } panic(fmt.Sprintf("%s\nStack for running G's are skipped while panicking.", msg.String())) + default: panic(fmt.Sprintf("Unknown watchdog action %v", action)) -- cgit v1.2.3 From 1c2ecbb1a03ffaa3bcdb2ee69c879da5e7076fa5 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Mon, 27 Apr 2020 16:00:39 -0700 Subject: Import host sockets. The FileDescription implementation for hostfs sockets uses the standard Unix socket implementation (unix.SocketVFS2), but is also tied to a hostfs dentry. Updates #1672, #1476 PiperOrigin-RevId: 308716426 --- pkg/sentry/fs/host/socket.go | 4 + pkg/sentry/fs/host/socket_iovec.go | 4 + pkg/sentry/fs/host/socket_unsafe.go | 4 + pkg/sentry/fsimpl/host/BUILD | 10 + pkg/sentry/fsimpl/host/host.go | 57 +++-- pkg/sentry/fsimpl/host/socket.go | 397 ++++++++++++++++++++++++++++++++ pkg/sentry/fsimpl/host/socket_iovec.go | 113 +++++++++ pkg/sentry/fsimpl/host/socket_unsafe.go | 101 ++++++++ pkg/sentry/fsimpl/sockfs/sockfs.go | 20 +- pkg/sentry/socket/unix/BUILD | 1 + pkg/sentry/socket/unix/unix_vfs2.go | 42 ++-- 11 files changed, 698 insertions(+), 55 deletions(-) create mode 100644 pkg/sentry/fsimpl/host/socket.go create mode 100644 pkg/sentry/fsimpl/host/socket_iovec.go create mode 100644 pkg/sentry/fsimpl/host/socket_unsafe.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/host/socket.go b/pkg/sentry/fs/host/socket.go index 06fc2d80a..b6e94583e 100644 --- a/pkg/sentry/fs/host/socket.go +++ b/pkg/sentry/fs/host/socket.go @@ -37,6 +37,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +// LINT.IfChange + // maxSendBufferSize is the maximum host send buffer size allowed for endpoint. // // N.B. 8MB is the default maximum on Linux (2 * sysctl_wmem_max). @@ -388,3 +390,5 @@ func (c *ConnectedEndpoint) Release() { // CloseUnread implements transport.ConnectedEndpoint.CloseUnread. func (c *ConnectedEndpoint) CloseUnread() {} + +// LINT.ThenChange(../../fsimpl/host/socket.go) diff --git a/pkg/sentry/fs/host/socket_iovec.go b/pkg/sentry/fs/host/socket_iovec.go index af6955675..5c18dbd5e 100644 --- a/pkg/sentry/fs/host/socket_iovec.go +++ b/pkg/sentry/fs/host/socket_iovec.go @@ -21,6 +21,8 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) +// LINT.IfChange + // maxIovs is the maximum number of iovecs to pass to the host. var maxIovs = linux.UIO_MAXIOV @@ -111,3 +113,5 @@ func buildIovec(bufs [][]byte, maxlen int64, truncate bool) (length int64, iovec return total, iovecs, nil, err } + +// LINT.ThenChange(../../fsimpl/host/socket_iovec.go) diff --git a/pkg/sentry/fs/host/socket_unsafe.go b/pkg/sentry/fs/host/socket_unsafe.go index f3bbed7ea..5d4f312cf 100644 --- a/pkg/sentry/fs/host/socket_unsafe.go +++ b/pkg/sentry/fs/host/socket_unsafe.go @@ -19,6 +19,8 @@ import ( "unsafe" ) +// LINT.IfChange + // fdReadVec receives from fd to bufs. // // If the total length of bufs is > maxlen, fdReadVec will do a partial read @@ -99,3 +101,5 @@ func fdWriteVec(fd int, bufs [][]byte, maxlen int64, truncate bool) (int64, int6 return int64(n), length, err } + +// LINT.ThenChange(../../fsimpl/host/socket_unsafe.go) diff --git a/pkg/sentry/fsimpl/host/BUILD b/pkg/sentry/fsimpl/host/BUILD index 2dcb03a73..e1c56d89b 100644 --- a/pkg/sentry/fsimpl/host/BUILD +++ b/pkg/sentry/fsimpl/host/BUILD @@ -8,6 +8,9 @@ go_library( "control.go", "host.go", "ioctl_unsafe.go", + "socket.go", + "socket_iovec.go", + "socket_unsafe.go", "tty.go", "util.go", "util_unsafe.go", @@ -16,6 +19,7 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/context", + "//pkg/fdnotifier", "//pkg/log", "//pkg/refs", "//pkg/sentry/arch", @@ -25,12 +29,18 @@ go_library( "//pkg/sentry/kernel/auth", "//pkg/sentry/memmap", "//pkg/sentry/socket/control", + "//pkg/sentry/socket/unix", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/unimpl", + "//pkg/sentry/uniqueid", "//pkg/sentry/vfs", "//pkg/sync", + "//pkg/syserr", "//pkg/syserror", + "//pkg/tcpip", + "//pkg/unet", "//pkg/usermem", + "//pkg/waiter", "@org_golang_x_sys//unix:go_default_library", ], ) diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 1e53b5c1b..05538ea42 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -17,7 +17,6 @@ package host import ( - "errors" "fmt" "math" "syscall" @@ -31,6 +30,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/hostfd" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/memmap" + unixsocket "gvisor.dev/gvisor/pkg/sentry/socket/unix" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" @@ -156,7 +156,7 @@ type inode struct { // Note that these flags may become out of date, since they can be modified // on the host, e.g. with fcntl. -func fileFlagsFromHostFD(fd int) (int, error) { +func fileFlagsFromHostFD(fd int) (uint32, error) { flags, err := unix.FcntlInt(uintptr(fd), syscall.F_GETFL, 0) if err != nil { log.Warningf("Failed to get file flags for donated FD %d: %v", fd, err) @@ -164,7 +164,7 @@ func fileFlagsFromHostFD(fd int) (int, error) { } // TODO(gvisor.dev/issue/1672): implement behavior corresponding to these allowed flags. flags &= syscall.O_ACCMODE | syscall.O_DIRECT | syscall.O_NONBLOCK | syscall.O_DSYNC | syscall.O_SYNC | syscall.O_APPEND - return flags, nil + return uint32(flags), nil } // CheckPermissions implements kernfs.Inode. @@ -361,6 +361,10 @@ func (i *inode) Destroy() { // Open implements kernfs.Inode. func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + // Once created, we cannot re-open a socket fd through /proc/[pid]/fd/. + if i.Mode().FileType() == linux.S_IFSOCK { + return nil, syserror.ENXIO + } return i.open(ctx, vfsd, rp.Mount()) } @@ -370,42 +374,45 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount) (*vfs.F return nil, err } fileType := s.Mode & linux.FileTypeMask + flags, err := fileFlagsFromHostFD(i.hostFD) + if err != nil { + return nil, err + } + if fileType == syscall.S_IFSOCK { if i.isTTY { - return nil, errors.New("cannot use host socket as TTY") + log.Warningf("cannot use host socket fd %d as TTY", i.hostFD) + return nil, syserror.ENOTTY } - // TODO(gvisor.dev/issue/1672): support importing sockets. - return nil, errors.New("importing host sockets not supported") + + ep, err := newEndpoint(ctx, i.hostFD) + if err != nil { + return nil, err + } + // Currently, we only allow Unix sockets to be imported. + return unixsocket.NewFileDescription(ep, ep.Type(), flags, mnt, d) } // TODO(gvisor.dev/issue/1672): Whitelist specific file types here, so that // we don't allow importing arbitrary file types without proper support. - var ( - vfsfd *vfs.FileDescription - fdImpl vfs.FileDescriptionImpl - ) if i.isTTY { fd := &ttyFD{ fileDescription: fileDescription{inode: i}, termios: linux.DefaultSlaveTermios, } - vfsfd = &fd.vfsfd - fdImpl = fd - } else { - // For simplicity, set offset to 0. Technically, we should - // only set to 0 on files that are not seekable (sockets, pipes, etc.), - // and use the offset from the host fd otherwise. - fd := &fileDescription{inode: i} - vfsfd = &fd.vfsfd - fdImpl = fd - } - - flags, err := fileFlagsFromHostFD(i.hostFD) - if err != nil { - return nil, err + vfsfd := &fd.vfsfd + if err := vfsfd.Init(fd, flags, mnt, d, &vfs.FileDescriptionOptions{}); err != nil { + return nil, err + } + return vfsfd, nil } - if err := vfsfd.Init(fdImpl, uint32(flags), mnt, d, &vfs.FileDescriptionOptions{}); err != nil { + // For simplicity, set offset to 0. Technically, we should + // only set to 0 on files that are not seekable (sockets, pipes, etc.), + // and use the offset from the host fd otherwise. + fd := &fileDescription{inode: i} + vfsfd := &fd.vfsfd + if err := vfsfd.Init(fd, flags, mnt, d, &vfs.FileDescriptionOptions{}); err != nil { return nil, err } return vfsfd, nil diff --git a/pkg/sentry/fsimpl/host/socket.go b/pkg/sentry/fsimpl/host/socket.go new file mode 100644 index 000000000..39dd624a0 --- /dev/null +++ b/pkg/sentry/fsimpl/host/socket.go @@ -0,0 +1,397 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package host + +import ( + "fmt" + "syscall" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fdnotifier" + "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/refs" + "gvisor.dev/gvisor/pkg/sentry/socket/control" + "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" + "gvisor.dev/gvisor/pkg/sentry/uniqueid" + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/syserr" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/unet" + "gvisor.dev/gvisor/pkg/waiter" +) + +// Create a new host-backed endpoint from the given fd. +func newEndpoint(ctx context.Context, hostFD int) (transport.Endpoint, error) { + // Set up an external transport.Endpoint using the host fd. + addr := fmt.Sprintf("hostfd:[%d]", hostFD) + var q waiter.Queue + e, err := NewConnectedEndpoint(ctx, hostFD, &q, addr, true /* saveable */) + if err != nil { + return nil, err.ToError() + } + e.Init() + ep := transport.NewExternal(ctx, e.stype, uniqueid.GlobalProviderFromContext(ctx), &q, e, e) + return ep, nil +} + +// maxSendBufferSize is the maximum host send buffer size allowed for endpoint. +// +// N.B. 8MB is the default maximum on Linux (2 * sysctl_wmem_max). +const maxSendBufferSize = 8 << 20 + +// ConnectedEndpoint is an implementation of transport.ConnectedEndpoint and +// transport.Receiver. It is backed by a host fd that was imported at sentry +// startup. This fd is shared with a hostfs inode, which retains ownership of +// it. +// +// ConnectedEndpoint is saveable, since we expect that the host will provide +// the same fd upon restore. +// +// As of this writing, we only allow Unix sockets to be imported. +// +// +stateify savable +type ConnectedEndpoint struct { + // ref keeps track of references to a ConnectedEndpoint. + ref refs.AtomicRefCount + + // mu protects fd below. + mu sync.RWMutex `state:"nosave"` + + // fd is the host fd backing this endpoint. + fd int + + // addr is the address at which this endpoint is bound. + addr string + + queue *waiter.Queue + + // sndbuf is the size of the send buffer. + // + // N.B. When this is smaller than the host size, we present it via + // GetSockOpt and message splitting/rejection in SendMsg, but do not + // prevent lots of small messages from filling the real send buffer + // size on the host. + sndbuf int64 `state:"nosave"` + + // stype is the type of Unix socket. + stype linux.SockType +} + +// init performs initialization required for creating new ConnectedEndpoints and +// for restoring them. +func (c *ConnectedEndpoint) init() *syserr.Error { + family, err := syscall.GetsockoptInt(c.fd, syscall.SOL_SOCKET, syscall.SO_DOMAIN) + if err != nil { + return syserr.FromError(err) + } + + if family != syscall.AF_UNIX { + // We only allow Unix sockets. + return syserr.ErrInvalidEndpointState + } + + stype, err := syscall.GetsockoptInt(c.fd, syscall.SOL_SOCKET, syscall.SO_TYPE) + if err != nil { + return syserr.FromError(err) + } + + if err := syscall.SetNonblock(c.fd, true); err != nil { + return syserr.FromError(err) + } + + sndbuf, err := syscall.GetsockoptInt(c.fd, syscall.SOL_SOCKET, syscall.SO_SNDBUF) + if err != nil { + return syserr.FromError(err) + } + if sndbuf > maxSendBufferSize { + log.Warningf("Socket send buffer too large: %d", sndbuf) + return syserr.ErrInvalidEndpointState + } + + c.stype = linux.SockType(stype) + c.sndbuf = int64(sndbuf) + + return nil +} + +// NewConnectedEndpoint creates a new ConnectedEndpoint backed by a host fd +// imported at sentry startup, +// +// The caller is responsible for calling Init(). Additionaly, Release needs to +// be called twice because ConnectedEndpoint is both a transport.Receiver and +// transport.ConnectedEndpoint. +func NewConnectedEndpoint(ctx context.Context, hostFD int, queue *waiter.Queue, addr string, saveable bool) (*ConnectedEndpoint, *syserr.Error) { + e := ConnectedEndpoint{ + fd: hostFD, + addr: addr, + queue: queue, + } + + if err := e.init(); err != nil { + return nil, err + } + + // AtomicRefCounters start off with a single reference. We need two. + e.ref.IncRef() + e.ref.EnableLeakCheck("host.ConnectedEndpoint") + return &e, nil +} + +// Init will do the initialization required without holding other locks. +func (c *ConnectedEndpoint) Init() { + if err := fdnotifier.AddFD(int32(c.fd), c.queue); err != nil { + panic(err) + } +} + +// Send implements transport.ConnectedEndpoint.Send. +func (c *ConnectedEndpoint) Send(data [][]byte, controlMessages transport.ControlMessages, from tcpip.FullAddress) (int64, bool, *syserr.Error) { + c.mu.RLock() + defer c.mu.RUnlock() + + if !controlMessages.Empty() { + return 0, false, syserr.ErrInvalidEndpointState + } + + // Since stream sockets don't preserve message boundaries, we can write + // only as much of the message as fits in the send buffer. + truncate := c.stype == linux.SOCK_STREAM + + n, totalLen, err := fdWriteVec(c.fd, data, c.sndbuf, truncate) + if n < totalLen && err == nil { + // The host only returns a short write if it would otherwise + // block (and only for stream sockets). + err = syserror.EAGAIN + } + if n > 0 && err != syserror.EAGAIN { + // The caller may need to block to send more data, but + // otherwise there isn't anything that can be done about an + // error with a partial write. + err = nil + } + + // There is no need for the callee to call SendNotify because fdWriteVec + // uses the host's sendmsg(2) and the host kernel's queue. + return n, false, syserr.FromError(err) +} + +// SendNotify implements transport.ConnectedEndpoint.SendNotify. +func (c *ConnectedEndpoint) SendNotify() {} + +// CloseSend implements transport.ConnectedEndpoint.CloseSend. +func (c *ConnectedEndpoint) CloseSend() { + c.mu.Lock() + defer c.mu.Unlock() + + if err := syscall.Shutdown(c.fd, syscall.SHUT_WR); err != nil { + // A well-formed UDS shutdown can't fail. See + // net/unix/af_unix.c:unix_shutdown. + panic(fmt.Sprintf("failed write shutdown on host socket %+v: %v", c, err)) + } +} + +// CloseNotify implements transport.ConnectedEndpoint.CloseNotify. +func (c *ConnectedEndpoint) CloseNotify() {} + +// Writable implements transport.ConnectedEndpoint.Writable. +func (c *ConnectedEndpoint) Writable() bool { + c.mu.RLock() + defer c.mu.RUnlock() + + return fdnotifier.NonBlockingPoll(int32(c.fd), waiter.EventOut)&waiter.EventOut != 0 +} + +// Passcred implements transport.ConnectedEndpoint.Passcred. +func (c *ConnectedEndpoint) Passcred() bool { + // We don't support credential passing for host sockets. + return false +} + +// GetLocalAddress implements transport.ConnectedEndpoint.GetLocalAddress. +func (c *ConnectedEndpoint) GetLocalAddress() (tcpip.FullAddress, *tcpip.Error) { + return tcpip.FullAddress{Addr: tcpip.Address(c.addr)}, nil +} + +// EventUpdate implements transport.ConnectedEndpoint.EventUpdate. +func (c *ConnectedEndpoint) EventUpdate() { + c.mu.RLock() + defer c.mu.RUnlock() + if c.fd != -1 { + fdnotifier.UpdateFD(int32(c.fd)) + } +} + +// Recv implements transport.Receiver.Recv. +func (c *ConnectedEndpoint) Recv(data [][]byte, creds bool, numRights int, peek bool) (int64, int64, transport.ControlMessages, bool, tcpip.FullAddress, bool, *syserr.Error) { + c.mu.RLock() + defer c.mu.RUnlock() + + var cm unet.ControlMessage + if numRights > 0 { + cm.EnableFDs(int(numRights)) + } + + // N.B. Unix sockets don't have a receive buffer, the send buffer + // serves both purposes. + rl, ml, cl, cTrunc, err := fdReadVec(c.fd, data, []byte(cm), peek, c.sndbuf) + if rl > 0 && err != nil { + // We got some data, so all we need to do on error is return + // the data that we got. Short reads are fine, no need to + // block. + err = nil + } + if err != nil { + return 0, 0, transport.ControlMessages{}, false, tcpip.FullAddress{}, false, syserr.FromError(err) + } + + // There is no need for the callee to call RecvNotify because fdReadVec uses + // the host's recvmsg(2) and the host kernel's queue. + + // Trim the control data if we received less than the full amount. + if cl < uint64(len(cm)) { + cm = cm[:cl] + } + + // Avoid extra allocations in the case where there isn't any control data. + if len(cm) == 0 { + return rl, ml, transport.ControlMessages{}, cTrunc, tcpip.FullAddress{Addr: tcpip.Address(c.addr)}, false, nil + } + + fds, err := cm.ExtractFDs() + if err != nil { + return 0, 0, transport.ControlMessages{}, false, tcpip.FullAddress{}, false, syserr.FromError(err) + } + + if len(fds) == 0 { + return rl, ml, transport.ControlMessages{}, cTrunc, tcpip.FullAddress{Addr: tcpip.Address(c.addr)}, false, nil + } + return rl, ml, control.NewVFS2(nil, nil, newSCMRights(fds)), cTrunc, tcpip.FullAddress{Addr: tcpip.Address(c.addr)}, false, nil +} + +// RecvNotify implements transport.Receiver.RecvNotify. +func (c *ConnectedEndpoint) RecvNotify() {} + +// CloseRecv implements transport.Receiver.CloseRecv. +func (c *ConnectedEndpoint) CloseRecv() { + c.mu.Lock() + defer c.mu.Unlock() + + if err := syscall.Shutdown(c.fd, syscall.SHUT_RD); err != nil { + // A well-formed UDS shutdown can't fail. See + // net/unix/af_unix.c:unix_shutdown. + panic(fmt.Sprintf("failed read shutdown on host socket %+v: %v", c, err)) + } +} + +// Readable implements transport.Receiver.Readable. +func (c *ConnectedEndpoint) Readable() bool { + c.mu.RLock() + defer c.mu.RUnlock() + + return fdnotifier.NonBlockingPoll(int32(c.fd), waiter.EventIn)&waiter.EventIn != 0 +} + +// SendQueuedSize implements transport.Receiver.SendQueuedSize. +func (c *ConnectedEndpoint) SendQueuedSize() int64 { + // TODO(gvisor.dev/issue/273): SendQueuedSize isn't supported for host + // sockets because we don't allow the sentry to call ioctl(2). + return -1 +} + +// RecvQueuedSize implements transport.Receiver.RecvQueuedSize. +func (c *ConnectedEndpoint) RecvQueuedSize() int64 { + // TODO(gvisor.dev/issue/273): RecvQueuedSize isn't supported for host + // sockets because we don't allow the sentry to call ioctl(2). + return -1 +} + +// SendMaxQueueSize implements transport.Receiver.SendMaxQueueSize. +func (c *ConnectedEndpoint) SendMaxQueueSize() int64 { + return int64(c.sndbuf) +} + +// RecvMaxQueueSize implements transport.Receiver.RecvMaxQueueSize. +func (c *ConnectedEndpoint) RecvMaxQueueSize() int64 { + // N.B. Unix sockets don't use the receive buffer. We'll claim it is + // the same size as the send buffer. + return int64(c.sndbuf) +} + +func (c *ConnectedEndpoint) destroyLocked() { + fdnotifier.RemoveFD(int32(c.fd)) + c.fd = -1 +} + +// Release implements transport.ConnectedEndpoint.Release and +// transport.Receiver.Release. +func (c *ConnectedEndpoint) Release() { + c.ref.DecRefWithDestructor(func() { + c.mu.Lock() + c.destroyLocked() + c.mu.Unlock() + }) +} + +// CloseUnread implements transport.ConnectedEndpoint.CloseUnread. +func (c *ConnectedEndpoint) CloseUnread() {} + +// SCMConnectedEndpoint represents an endpoint backed by a host fd that was +// passed through a gofer Unix socket. It is almost the same as +// ConnectedEndpoint, with the following differences: +// - SCMConnectedEndpoint is not saveable, because the host cannot guarantee +// the same descriptor number across S/R. +// - SCMConnectedEndpoint holds ownership of its fd and is responsible for +// closing it. +type SCMConnectedEndpoint struct { + ConnectedEndpoint +} + +// Release implements transport.ConnectedEndpoint.Release and +// transport.Receiver.Release. +func (e *SCMConnectedEndpoint) Release() { + e.ref.DecRefWithDestructor(func() { + e.mu.Lock() + if err := syscall.Close(e.fd); err != nil { + log.Warningf("Failed to close host fd %d: %v", err) + } + e.destroyLocked() + e.mu.Unlock() + }) +} + +// NewSCMEndpoint creates a new SCMConnectedEndpoint backed by a host fd that +// was passed through a Unix socket. +// +// The caller is responsible for calling Init(). Additionaly, Release needs to +// be called twice because ConnectedEndpoint is both a transport.Receiver and +// transport.ConnectedEndpoint. +func NewSCMEndpoint(ctx context.Context, hostFD int, queue *waiter.Queue, addr string) (*SCMConnectedEndpoint, *syserr.Error) { + e := SCMConnectedEndpoint{ConnectedEndpoint{ + fd: hostFD, + addr: addr, + queue: queue, + }} + + if err := e.init(); err != nil { + return nil, err + } + + // AtomicRefCounters start off with a single reference. We need two. + e.ref.IncRef() + e.ref.EnableLeakCheck("host.SCMConnectedEndpoint") + return &e, nil +} diff --git a/pkg/sentry/fsimpl/host/socket_iovec.go b/pkg/sentry/fsimpl/host/socket_iovec.go new file mode 100644 index 000000000..584c247d2 --- /dev/null +++ b/pkg/sentry/fsimpl/host/socket_iovec.go @@ -0,0 +1,113 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package host + +import ( + "syscall" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/syserror" +) + +// maxIovs is the maximum number of iovecs to pass to the host. +var maxIovs = linux.UIO_MAXIOV + +// copyToMulti copies as many bytes from src to dst as possible. +func copyToMulti(dst [][]byte, src []byte) { + for _, d := range dst { + done := copy(d, src) + src = src[done:] + if len(src) == 0 { + break + } + } +} + +// copyFromMulti copies as many bytes from src to dst as possible. +func copyFromMulti(dst []byte, src [][]byte) { + for _, s := range src { + done := copy(dst, s) + dst = dst[done:] + if len(dst) == 0 { + break + } + } +} + +// buildIovec builds an iovec slice from the given []byte slice. +// +// If truncate, truncate bufs > maxlen. Otherwise, immediately return an error. +// +// If length < the total length of bufs, err indicates why, even when returning +// a truncated iovec. +// +// If intermediate != nil, iovecs references intermediate rather than bufs and +// the caller must copy to/from bufs as necessary. +func buildIovec(bufs [][]byte, maxlen int64, truncate bool) (length int64, iovecs []syscall.Iovec, intermediate []byte, err error) { + var iovsRequired int + for _, b := range bufs { + length += int64(len(b)) + if len(b) > 0 { + iovsRequired++ + } + } + + stopLen := length + if length > maxlen { + if truncate { + stopLen = maxlen + err = syserror.EAGAIN + } else { + return 0, nil, nil, syserror.EMSGSIZE + } + } + + if iovsRequired > maxIovs { + // The kernel will reject our call if we pass this many iovs. + // Use a single intermediate buffer instead. + b := make([]byte, stopLen) + + return stopLen, []syscall.Iovec{{ + Base: &b[0], + Len: uint64(stopLen), + }}, b, err + } + + var total int64 + iovecs = make([]syscall.Iovec, 0, iovsRequired) + for i := range bufs { + l := len(bufs[i]) + if l == 0 { + continue + } + + stop := int64(l) + if total+stop > stopLen { + stop = stopLen - total + } + + iovecs = append(iovecs, syscall.Iovec{ + Base: &bufs[i][0], + Len: uint64(stop), + }) + + total += stop + if total >= stopLen { + break + } + } + + return total, iovecs, nil, err +} diff --git a/pkg/sentry/fsimpl/host/socket_unsafe.go b/pkg/sentry/fsimpl/host/socket_unsafe.go new file mode 100644 index 000000000..35ded24bc --- /dev/null +++ b/pkg/sentry/fsimpl/host/socket_unsafe.go @@ -0,0 +1,101 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package host + +import ( + "syscall" + "unsafe" +) + +// fdReadVec receives from fd to bufs. +// +// If the total length of bufs is > maxlen, fdReadVec will do a partial read +// and err will indicate why the message was truncated. +func fdReadVec(fd int, bufs [][]byte, control []byte, peek bool, maxlen int64) (readLen int64, msgLen int64, controlLen uint64, controlTrunc bool, err error) { + flags := uintptr(syscall.MSG_DONTWAIT | syscall.MSG_TRUNC) + if peek { + flags |= syscall.MSG_PEEK + } + + // Always truncate the receive buffer. All socket types will truncate + // received messages. + length, iovecs, intermediate, err := buildIovec(bufs, maxlen, true) + if err != nil && len(iovecs) == 0 { + // No partial write to do, return error immediately. + return 0, 0, 0, false, err + } + + var msg syscall.Msghdr + if len(control) != 0 { + msg.Control = &control[0] + msg.Controllen = uint64(len(control)) + } + + if len(iovecs) != 0 { + msg.Iov = &iovecs[0] + msg.Iovlen = uint64(len(iovecs)) + } + + rawN, _, e := syscall.RawSyscall(syscall.SYS_RECVMSG, uintptr(fd), uintptr(unsafe.Pointer(&msg)), flags) + if e != 0 { + // N.B. prioritize the syscall error over the buildIovec error. + return 0, 0, 0, false, e + } + n := int64(rawN) + + // Copy data back to bufs. + if intermediate != nil { + copyToMulti(bufs, intermediate) + } + + controlTrunc = msg.Flags&syscall.MSG_CTRUNC == syscall.MSG_CTRUNC + + if n > length { + return length, n, msg.Controllen, controlTrunc, err + } + + return n, n, msg.Controllen, controlTrunc, err +} + +// fdWriteVec sends from bufs to fd. +// +// If the total length of bufs is > maxlen && truncate, fdWriteVec will do a +// partial write and err will indicate why the message was truncated. +func fdWriteVec(fd int, bufs [][]byte, maxlen int64, truncate bool) (int64, int64, error) { + length, iovecs, intermediate, err := buildIovec(bufs, maxlen, truncate) + if err != nil && len(iovecs) == 0 { + // No partial write to do, return error immediately. + return 0, length, err + } + + // Copy data to intermediate buf. + if intermediate != nil { + copyFromMulti(intermediate, bufs) + } + + var msg syscall.Msghdr + if len(iovecs) > 0 { + msg.Iov = &iovecs[0] + msg.Iovlen = uint64(len(iovecs)) + } + + n, _, e := syscall.RawSyscall(syscall.SYS_SENDMSG, uintptr(fd), uintptr(unsafe.Pointer(&msg)), syscall.MSG_DONTWAIT|syscall.MSG_NOSIGNAL) + if e != 0 { + // N.B. prioritize the syscall error over the buildIovec error. + return 0, length, e + } + + return int64(n), length, err +} diff --git a/pkg/sentry/fsimpl/sockfs/sockfs.go b/pkg/sentry/fsimpl/sockfs/sockfs.go index 5ce50625b..271134af8 100644 --- a/pkg/sentry/fsimpl/sockfs/sockfs.go +++ b/pkg/sentry/fsimpl/sockfs/sockfs.go @@ -73,26 +73,14 @@ func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentr return nil, syserror.ENXIO } -// InitSocket initializes a socket FileDescription, with a corresponding -// Dentry in mnt. -// -// fd should be the FileDescription associated with socketImpl, i.e. its first -// field. mnt should be the global socket mount, Kernel.socketMount. -func InitSocket(socketImpl vfs.FileDescriptionImpl, fd *vfs.FileDescription, mnt *vfs.Mount, creds *auth.Credentials) error { - fsimpl := mnt.Filesystem().Impl() - fs := fsimpl.(*kernfs.Filesystem) - +// NewDentry constructs and returns a sockfs dentry. +func NewDentry(creds *auth.Credentials, ino uint64) *vfs.Dentry { // File mode matches net/socket.c:sock_alloc. filemode := linux.FileMode(linux.S_IFSOCK | 0600) i := &inode{} - i.Init(creds, fs.NextIno(), filemode) + i.Init(creds, ino, filemode) d := &kernfs.Dentry{} d.Init(i) - - opts := &vfs.FileDescriptionOptions{UseDentryMetadata: true} - if err := fd.Init(socketImpl, linux.O_RDWR, mnt, d.VFSDentry(), opts); err != nil { - return err - } - return nil + return d.VFSDentry() } diff --git a/pkg/sentry/socket/unix/BUILD b/pkg/sentry/socket/unix/BUILD index de2cc4bdf..941a91097 100644 --- a/pkg/sentry/socket/unix/BUILD +++ b/pkg/sentry/socket/unix/BUILD @@ -21,6 +21,7 @@ go_library( "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", + "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/fsimpl/sockfs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/time", diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go index 3e54d49c4..315dc274f 100644 --- a/pkg/sentry/socket/unix/unix_vfs2.go +++ b/pkg/sentry/socket/unix/unix_vfs2.go @@ -19,6 +19,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/socket/control" @@ -42,30 +43,44 @@ type SocketVFS2 struct { socketOpsCommon } -// NewVFS2File creates and returns a new vfs.FileDescription for a unix socket. -func NewVFS2File(t *kernel.Task, ep transport.Endpoint, stype linux.SockType) (*vfs.FileDescription, *syserr.Error) { - sock := NewFDImpl(ep, stype) - vfsfd := &sock.vfsfd - if err := sockfs.InitSocket(sock, vfsfd, t.Kernel().SocketMount(), t.Credentials()); err != nil { +// NewSockfsFile creates a new socket file in the global sockfs mount and +// returns a corresponding file description. +func NewSockfsFile(t *kernel.Task, ep transport.Endpoint, stype linux.SockType) (*vfs.FileDescription, *syserr.Error) { + mnt := t.Kernel().SocketMount() + fs := mnt.Filesystem().Impl().(*kernfs.Filesystem) + d := sockfs.NewDentry(t.Credentials(), fs.NextIno()) + + fd, err := NewFileDescription(ep, stype, linux.O_RDWR, mnt, d) + if err != nil { return nil, syserr.FromError(err) } - return vfsfd, nil + return fd, nil } -// NewFDImpl creates and returns a new SocketVFS2. -func NewFDImpl(ep transport.Endpoint, stype linux.SockType) *SocketVFS2 { +// NewFileDescription creates and returns a socket file description +// corresponding to the given mount and dentry. +func NewFileDescription(ep transport.Endpoint, stype linux.SockType, flags uint32, mnt *vfs.Mount, d *vfs.Dentry) (*vfs.FileDescription, error) { // You can create AF_UNIX, SOCK_RAW sockets. They're the same as // SOCK_DGRAM and don't require CAP_NET_RAW. if stype == linux.SOCK_RAW { stype = linux.SOCK_DGRAM } - return &SocketVFS2{ + sock := &SocketVFS2{ socketOpsCommon: socketOpsCommon{ ep: ep, stype: stype, }, } + vfsfd := &sock.vfsfd + if err := vfsfd.Init(sock, flags, mnt, d, &vfs.FileDescriptionOptions{ + DenyPRead: true, + DenyPWrite: true, + UseDentryMetadata: true, + }); err != nil { + return nil, err + } + return vfsfd, nil } // GetSockOpt implements the linux syscall getsockopt(2) for sockets backed by @@ -112,8 +127,7 @@ func (s *SocketVFS2) Accept(t *kernel.Task, peerRequested bool, flags int, block } } - // We expect this to be a FileDescription here. - ns, err := NewVFS2File(t, ep, s.stype) + ns, err := NewSockfsFile(t, ep, s.stype) if err != nil { return 0, nil, 0, err } @@ -307,7 +321,7 @@ func (*providerVFS2) Socket(t *kernel.Task, stype linux.SockType, protocol int) return nil, syserr.ErrInvalidArgument } - f, err := NewVFS2File(t, ep, stype) + f, err := NewSockfsFile(t, ep, stype) if err != nil { ep.Close() return nil, err @@ -331,13 +345,13 @@ func (*providerVFS2) Pair(t *kernel.Task, stype linux.SockType, protocol int) (* // Create the endpoints and sockets. ep1, ep2 := transport.NewPair(t, stype, t.Kernel()) - s1, err := NewVFS2File(t, ep1, stype) + s1, err := NewSockfsFile(t, ep1, stype) if err != nil { ep1.Close() ep2.Close() return nil, nil, err } - s2, err := NewVFS2File(t, ep2, stype) + s2, err := NewSockfsFile(t, ep2, stype) if err != nil { s1.DecRef() ep2.Close() -- cgit v1.2.3 From 44d780aa4bd15cd5fb19151910d1f4e681084333 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Tue, 28 Apr 2020 01:15:38 -0400 Subject: code clean in arch module Signed-off-by: Bin Lu --- pkg/sentry/arch/arch_aarch64.go | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/arch/arch_aarch64.go b/pkg/sentry/arch/arch_aarch64.go index 529980267..343f81f59 100644 --- a/pkg/sentry/arch/arch_aarch64.go +++ b/pkg/sentry/arch/arch_aarch64.go @@ -274,7 +274,7 @@ const ( func (s *State) PtraceGetRegSet(regset uintptr, dst io.Writer, maxlen int) (int, error) { switch regset { case _NT_PRSTATUS: - if maxlen < ptraceRegsSize { + if maxlen < registersSize { return 0, syserror.EFAULT } return s.PtraceGetRegs(dst) @@ -287,7 +287,7 @@ func (s *State) PtraceGetRegSet(regset uintptr, dst io.Writer, maxlen int) (int, func (s *State) PtraceSetRegSet(regset uintptr, src io.Reader, maxlen int) (int, error) { switch regset { case _NT_PRSTATUS: - if maxlen < ptraceRegsSize { + if maxlen < registersSize { return 0, syserror.EFAULT } return s.PtraceSetRegs(src) -- cgit v1.2.3 From f3ca5ca82abe35d13210f8a8035926170bb87436 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 28 Apr 2020 08:32:20 -0700 Subject: Support pipes and sockets in VFS2 gofer fs. Named pipes and sockets can be represented in two ways in gofer fs: 1. As a file on the remote filesystem. In this case, all file operations are passed through 9p. 2. As a synthetic file that is internal to the sandbox. In this case, the dentry stores an endpoint or VFSPipe for sockets and pipes respectively, which replaces interactions with the remote fs through the gofer. In gofer.filesystem.MknodAt, we attempt to call mknod(2) through 9p, and if it fails, fall back to the synthetic version. Updates #1200. PiperOrigin-RevId: 308828161 --- pkg/sentry/fs/gofer/socket.go | 4 + pkg/sentry/fsimpl/ext/filesystem.go | 2 +- pkg/sentry/fsimpl/gofer/BUILD | 8 ++ pkg/sentry/fsimpl/gofer/directory.go | 39 +++++++-- pkg/sentry/fsimpl/gofer/filesystem.go | 128 ++++++++++++++++++++++-------- pkg/sentry/fsimpl/gofer/gofer.go | 32 ++++++-- pkg/sentry/fsimpl/gofer/socket.go | 141 +++++++++++++++++++++++++++++++++ pkg/sentry/fsimpl/kernfs/filesystem.go | 2 +- pkg/sentry/fsimpl/tmpfs/filesystem.go | 2 +- pkg/sentry/socket/unix/unix.go | 2 +- pkg/sentry/vfs/anonfs.go | 2 +- pkg/sentry/vfs/filesystem.go | 2 +- pkg/sentry/vfs/options.go | 18 +++++ pkg/sentry/vfs/vfs.go | 54 ++++++------- 14 files changed, 359 insertions(+), 77 deletions(-) create mode 100644 pkg/sentry/fsimpl/gofer/socket.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/gofer/socket.go b/pkg/sentry/fs/gofer/socket.go index 10ba2f5f0..40f2c1cad 100644 --- a/pkg/sentry/fs/gofer/socket.go +++ b/pkg/sentry/fs/gofer/socket.go @@ -47,6 +47,8 @@ func (i *inodeOperations) BoundEndpoint(inode *fs.Inode, path string) transport. return &endpoint{inode, i.fileState.file.file, path} } +// LINT.IfChange + // endpoint is a Gofer-backed transport.BoundEndpoint. // // An endpoint's lifetime is the time between when InodeOperations.BoundEndpoint() @@ -146,3 +148,5 @@ func (e *endpoint) Release() { func (e *endpoint) Passcred() bool { return false } + +// LINT.ThenChange(../../fsimpl/gofer/socket.go) diff --git a/pkg/sentry/fsimpl/ext/filesystem.go b/pkg/sentry/fsimpl/ext/filesystem.go index 2c22a04af..6c882d0b6 100644 --- a/pkg/sentry/fsimpl/ext/filesystem.go +++ b/pkg/sentry/fsimpl/ext/filesystem.go @@ -485,7 +485,7 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error } // BoundEndpointAt implements FilesystemImpl.BoundEndpointAt. -func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath) (transport.BoundEndpoint, error) { +func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.BoundEndpointOptions) (transport.BoundEndpoint, error) { _, _, err := fs.walk(rp, false) if err != nil { return nil, err diff --git a/pkg/sentry/fsimpl/gofer/BUILD b/pkg/sentry/fsimpl/gofer/BUILD index b9c4beee4..5ce82b793 100644 --- a/pkg/sentry/fsimpl/gofer/BUILD +++ b/pkg/sentry/fsimpl/gofer/BUILD @@ -38,6 +38,7 @@ go_library( "p9file.go", "pagemath.go", "regular_file.go", + "socket.go", "special_file.go", "symlink.go", "time.go", @@ -52,18 +53,25 @@ go_library( "//pkg/p9", "//pkg/safemem", "//pkg/sentry/fs/fsutil", + "//pkg/sentry/fsimpl/host", "//pkg/sentry/hostfd", + "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", + "//pkg/sentry/kernel/pipe", "//pkg/sentry/kernel/time", "//pkg/sentry/memmap", "//pkg/sentry/pgalloc", "//pkg/sentry/platform", + "//pkg/sentry/socket/control", + "//pkg/sentry/socket/unix", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usage", "//pkg/sentry/vfs", + "//pkg/syserr", "//pkg/syserror", "//pkg/unet", "//pkg/usermem", + "//pkg/waiter", ], ) diff --git a/pkg/sentry/fsimpl/gofer/directory.go b/pkg/sentry/fsimpl/gofer/directory.go index 55f9ed911..b98218753 100644 --- a/pkg/sentry/fsimpl/gofer/directory.go +++ b/pkg/sentry/fsimpl/gofer/directory.go @@ -15,6 +15,7 @@ package gofer import ( + "fmt" "sync" "sync/atomic" @@ -22,6 +23,8 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" + "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -59,28 +62,52 @@ func (d *dentry) cacheNegativeLookupLocked(name string) { d.children[name] = nil } -// createSyntheticDirectory creates a synthetic directory with the given name +type createSyntheticOpts struct { + name string + mode linux.FileMode + kuid auth.KUID + kgid auth.KGID + + // The endpoint for a synthetic socket. endpoint should be nil if the file + // being created is not a socket. + endpoint transport.BoundEndpoint + + // pipe should be nil if the file being created is not a pipe. + pipe *pipe.VFSPipe +} + +// createSyntheticChildLocked creates a synthetic file with the given name // in d. // // Preconditions: d.dirMu must be locked. d.isDir(). d does not already contain // a child with the given name. -func (d *dentry) createSyntheticDirectoryLocked(name string, mode linux.FileMode, kuid auth.KUID, kgid auth.KGID) { +func (d *dentry) createSyntheticChildLocked(opts *createSyntheticOpts) { d2 := &dentry{ refs: 1, // held by d fs: d.fs, - mode: uint32(mode) | linux.S_IFDIR, - uid: uint32(kuid), - gid: uint32(kgid), + mode: uint32(opts.mode), + uid: uint32(opts.kuid), + gid: uint32(opts.kgid), blockSize: usermem.PageSize, // arbitrary handle: handle{ fd: -1, }, nlink: uint32(2), } + switch opts.mode.FileType() { + case linux.S_IFDIR: + // Nothing else needs to be done. + case linux.S_IFSOCK: + d2.endpoint = opts.endpoint + case linux.S_IFIFO: + d2.pipe = opts.pipe + default: + panic(fmt.Sprintf("failed to create synthetic file of unrecognized type: %v", opts.mode.FileType())) + } d2.pf.dentry = d2 d2.vfsd.Init(d2) - d.cacheNewChildLocked(d2, name) + d.cacheNewChildLocked(d2, opts.name) d.syntheticChildren++ } diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 98ccb42fd..4a8411371 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -22,9 +22,11 @@ import ( "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // Sync implements vfs.FilesystemImpl.Sync. @@ -652,7 +654,12 @@ func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts v return err } ctx.Infof("Failed to create remote directory %q: %v; falling back to synthetic directory", name, err) - parent.createSyntheticDirectoryLocked(name, opts.Mode, creds.EffectiveKUID, creds.EffectiveKGID) + parent.createSyntheticChildLocked(&createSyntheticOpts{ + name: name, + mode: linux.S_IFDIR | opts.Mode, + kuid: creds.EffectiveKUID, + kgid: creds.EffectiveKGID, + }) } if fs.opts.interop != InteropModeShared { parent.incLinks() @@ -663,7 +670,12 @@ func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts v // Can't create non-synthetic files in synthetic directories. return syserror.EPERM } - parent.createSyntheticDirectoryLocked(name, opts.Mode, creds.EffectiveKUID, creds.EffectiveKGID) + parent.createSyntheticChildLocked(&createSyntheticOpts{ + name: name, + mode: linux.S_IFDIR | opts.Mode, + kuid: creds.EffectiveKUID, + kgid: creds.EffectiveKGID, + }) parent.incLinks() return nil }) @@ -674,6 +686,28 @@ func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v return fs.doCreateAt(ctx, rp, false /* dir */, func(parent *dentry, name string) error { creds := rp.Credentials() _, err := parent.file.mknod(ctx, name, (p9.FileMode)(opts.Mode), opts.DevMajor, opts.DevMinor, (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)) + if err == syserror.EPERM { + switch opts.Mode.FileType() { + case linux.S_IFSOCK: + parent.createSyntheticChildLocked(&createSyntheticOpts{ + name: name, + mode: opts.Mode, + kuid: creds.EffectiveKUID, + kgid: creds.EffectiveKGID, + endpoint: opts.Endpoint, + }) + return nil + case linux.S_IFIFO: + parent.createSyntheticChildLocked(&createSyntheticOpts{ + name: name, + mode: opts.Mode, + kuid: creds.EffectiveKUID, + kgid: creds.EffectiveKGID, + pipe: pipe.NewVFSPipe(true /* isNamed */, pipe.DefaultPipeSize, usermem.PageSize), + }) + return nil + } + } return err }, nil) } @@ -756,20 +790,21 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vf return nil, err } mnt := rp.Mount() - filetype := d.fileType() - switch { - case filetype == linux.S_IFREG && !d.fs.opts.regularFilesUseSpecialFileFD: - if err := d.ensureSharedHandle(ctx, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, opts.Flags&linux.O_TRUNC != 0); err != nil { - return nil, err - } - fd := ®ularFileFD{} - if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{ - AllowDirectIO: true, - }); err != nil { - return nil, err + switch d.fileType() { + case linux.S_IFREG: + if !d.fs.opts.regularFilesUseSpecialFileFD { + if err := d.ensureSharedHandle(ctx, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, opts.Flags&linux.O_TRUNC != 0); err != nil { + return nil, err + } + fd := ®ularFileFD{} + if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{ + AllowDirectIO: true, + }); err != nil { + return nil, err + } + return &fd.vfsfd, nil } - return &fd.vfsfd, nil - case filetype == linux.S_IFDIR: + case linux.S_IFDIR: // Can't open directories with O_CREAT. if opts.Flags&linux.O_CREAT != 0 { return nil, syserror.EISDIR @@ -791,26 +826,40 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vf return nil, err } return &fd.vfsfd, nil - case filetype == linux.S_IFLNK: + case linux.S_IFLNK: // Can't open symlinks without O_PATH (which is unimplemented). return nil, syserror.ELOOP - default: - if opts.Flags&linux.O_DIRECT != 0 { - return nil, syserror.EINVAL - } - h, err := openHandle(ctx, d.file, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, opts.Flags&linux.O_TRUNC != 0) - if err != nil { - return nil, err - } - fd := &specialFileFD{ - handle: h, + case linux.S_IFSOCK: + if d.isSynthetic() { + return nil, syserror.ENXIO } - if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { - h.close(ctx) - return nil, err + case linux.S_IFIFO: + if d.isSynthetic() { + return d.pipe.Open(ctx, mnt, &d.vfsd, opts.Flags) } - return &fd.vfsfd, nil } + return d.openSpecialFileLocked(ctx, mnt, opts) +} + +func (d *dentry) openSpecialFileLocked(ctx context.Context, mnt *vfs.Mount, opts *vfs.OpenOptions) (*vfs.FileDescription, error) { + ats := vfs.AccessTypesForOpenFlags(opts) + // Treat as a special file. This is done for non-synthetic pipes as well as + // regular files when d.fs.opts.regularFilesUseSpecialFileFD is true. + if opts.Flags&linux.O_DIRECT != 0 { + return nil, syserror.EINVAL + } + h, err := openHandle(ctx, d.file, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, opts.Flags&linux.O_TRUNC != 0) + if err != nil { + return nil, err + } + fd := &specialFileFD{ + handle: h, + } + if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { + h.close(ctx) + return nil, err + } + return &fd.vfsfd, nil } // Preconditions: d.fs.renameMu must be locked. d.dirMu must be locked. @@ -1196,15 +1245,28 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error } // BoundEndpointAt implements FilesystemImpl.BoundEndpointAt. -func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath) (transport.BoundEndpoint, error) { +func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.BoundEndpointOptions) (transport.BoundEndpoint, error) { var ds *[]*dentry fs.renameMu.RLock() defer fs.renameMuRUnlockAndCheckCaching(&ds) - _, err := fs.resolveLocked(ctx, rp, &ds) + d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return nil, err } - // TODO(gvisor.dev/issue/1476): Implement BoundEndpointAt. + if err := d.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil { + return nil, err + } + if d.isSocket() { + if !d.isSynthetic() { + d.IncRef() + return &endpoint{ + dentry: d, + file: d.file.file, + path: opts.Addr, + }, nil + } + return d.endpoint, nil + } return nil, syserror.ECONNREFUSED } diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 8b4e91d17..3235816c2 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -46,9 +46,11 @@ import ( "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" + "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/unet" @@ -585,6 +587,14 @@ type dentry struct { // and target are protected by dataMu. haveTarget bool target string + + // If this dentry represents a synthetic socket file, endpoint is the + // transport endpoint bound to this file. + endpoint transport.BoundEndpoint + + // If this dentry represents a synthetic named pipe, pipe is the pipe + // endpoint bound to this file. + pipe *pipe.VFSPipe } // dentryAttrMask returns a p9.AttrMask enabling all attributes used by the @@ -791,10 +801,21 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin setLocalAtime = stat.Mask&linux.STATX_ATIME != 0 setLocalMtime = stat.Mask&linux.STATX_MTIME != 0 stat.Mask &^= linux.STATX_ATIME | linux.STATX_MTIME - if !setLocalMtime && (stat.Mask&linux.STATX_SIZE != 0) { - // Truncate updates mtime. - setLocalMtime = true - stat.Mtime.Nsec = linux.UTIME_NOW + + // Prepare for truncate. + if stat.Mask&linux.STATX_SIZE != 0 { + switch d.mode & linux.S_IFMT { + case linux.S_IFREG: + if !setLocalMtime { + // Truncate updates mtime. + setLocalMtime = true + stat.Mtime.Nsec = linux.UTIME_NOW + } + case linux.S_IFDIR: + return syserror.EISDIR + default: + return syserror.EINVAL + } } } d.metadataMu.Lock() @@ -1049,6 +1070,7 @@ func (d *dentry) destroyLocked() { d.handle.close(ctx) } d.handleMu.Unlock() + if !d.file.isNil() { d.file.close(ctx) d.file = p9file{} @@ -1134,7 +1156,7 @@ func (d *dentry) removexattr(ctx context.Context, creds *auth.Credentials, name return d.file.removeXattr(ctx, name) } -// Preconditions: !d.file.isNil(). d.isRegularFile() || d.isDirectory(). +// Preconditions: !d.isSynthetic(). d.isRegularFile() || d.isDirectory(). func (d *dentry) ensureSharedHandle(ctx context.Context, read, write, trunc bool) error { // O_TRUNC unconditionally requires us to obtain a new handle (opened with // O_TRUNC). diff --git a/pkg/sentry/fsimpl/gofer/socket.go b/pkg/sentry/fsimpl/gofer/socket.go new file mode 100644 index 000000000..73835df91 --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/socket.go @@ -0,0 +1,141 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package gofer + +import ( + "syscall" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/p9" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/host" + "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" + "gvisor.dev/gvisor/pkg/syserr" + "gvisor.dev/gvisor/pkg/waiter" +) + +func (d *dentry) isSocket() bool { + return d.fileType() == linux.S_IFSOCK +} + +// endpoint is a Gofer-backed transport.BoundEndpoint. +// +// An endpoint's lifetime is the time between when filesystem.BoundEndpointAt() +// is called and either BoundEndpoint.BidirectionalConnect or +// BoundEndpoint.UnidirectionalConnect is called. +type endpoint struct { + // dentry is the filesystem dentry which produced this endpoint. + dentry *dentry + + // file is the p9 file that contains a single unopened fid. + file p9.File + + // path is the sentry path where this endpoint is bound. + path string +} + +func sockTypeToP9(t linux.SockType) (p9.ConnectFlags, bool) { + switch t { + case linux.SOCK_STREAM: + return p9.StreamSocket, true + case linux.SOCK_SEQPACKET: + return p9.SeqpacketSocket, true + case linux.SOCK_DGRAM: + return p9.DgramSocket, true + } + return 0, false +} + +// BidirectionalConnect implements ConnectableEndpoint.BidirectionalConnect. +func (e *endpoint) BidirectionalConnect(ctx context.Context, ce transport.ConnectingEndpoint, returnConnect func(transport.Receiver, transport.ConnectedEndpoint)) *syserr.Error { + cf, ok := sockTypeToP9(ce.Type()) + if !ok { + return syserr.ErrConnectionRefused + } + + // No lock ordering required as only the ConnectingEndpoint has a mutex. + ce.Lock() + + // Check connecting state. + if ce.Connected() { + ce.Unlock() + return syserr.ErrAlreadyConnected + } + if ce.Listening() { + ce.Unlock() + return syserr.ErrInvalidEndpointState + } + + c, err := e.newConnectedEndpoint(ctx, cf, ce.WaiterQueue()) + if err != nil { + ce.Unlock() + return err + } + + returnConnect(c, c) + ce.Unlock() + c.Init() + + return nil +} + +// UnidirectionalConnect implements +// transport.BoundEndpoint.UnidirectionalConnect. +func (e *endpoint) UnidirectionalConnect(ctx context.Context) (transport.ConnectedEndpoint, *syserr.Error) { + c, err := e.newConnectedEndpoint(ctx, p9.DgramSocket, &waiter.Queue{}) + if err != nil { + return nil, err + } + c.Init() + + // We don't need the receiver. + c.CloseRecv() + c.Release() + + return c, nil +} + +func (e *endpoint) newConnectedEndpoint(ctx context.Context, flags p9.ConnectFlags, queue *waiter.Queue) (*host.SCMConnectedEndpoint, *syserr.Error) { + hostFile, err := e.file.Connect(flags) + if err != nil { + return nil, syserr.ErrConnectionRefused + } + // Dup the fd so that the new endpoint can manage its lifetime. + hostFD, err := syscall.Dup(hostFile.FD()) + if err != nil { + log.Warningf("Could not dup host socket fd %d: %v", hostFile.FD(), err) + return nil, syserr.FromError(err) + } + // After duplicating, we no longer need hostFile. + hostFile.Close() + + c, serr := host.NewSCMEndpoint(ctx, hostFD, queue, e.path) + if serr != nil { + log.Warningf("Gofer returned invalid host socket for BidirectionalConnect; file %+v flags %+v: %v", e.file, flags, serr) + return nil, serr + } + return c, nil +} + +// Release implements transport.BoundEndpoint.Release. +func (e *endpoint) Release() { + e.dentry.DecRef() +} + +// Passcred implements transport.BoundEndpoint.Passcred. +func (e *endpoint) Passcred() bool { + return false +} diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 9e8d80414..40c1b2104 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -766,7 +766,7 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error } // BoundEndpointAt implements FilesystemImpl.BoundEndpointAt. -func (fs *Filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath) (transport.BoundEndpoint, error) { +func (fs *Filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.BoundEndpointOptions) (transport.BoundEndpoint, error) { fs.mu.RLock() _, _, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 5b62f9ebb..388b98bef 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -697,7 +697,7 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error } // BoundEndpointAt implements FilesystemImpl.BoundEndpointAt. -func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath) (transport.BoundEndpoint, error) { +func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.BoundEndpointOptions) (transport.BoundEndpoint, error) { fs.mu.RLock() defer fs.mu.RUnlock() d, err := resolveLocked(rp) diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go index 7c64f30fa..9aad41c9e 100644 --- a/pkg/sentry/socket/unix/unix.go +++ b/pkg/sentry/socket/unix/unix.go @@ -373,7 +373,7 @@ func extractEndpoint(t *kernel.Task, sockaddr []byte) (transport.BoundEndpoint, Path: p, FollowFinalSymlink: true, } - ep, e := t.Kernel().VFS().BoundEndpointAt(t, t.Credentials(), &pop) + ep, e := t.Kernel().VFS().BoundEndpointAt(t, t.Credentials(), &pop, &vfs.BoundEndpointOptions{path}) root.DecRef() if relPath { start.DecRef() diff --git a/pkg/sentry/vfs/anonfs.go b/pkg/sentry/vfs/anonfs.go index a64d86122..b1a998590 100644 --- a/pkg/sentry/vfs/anonfs.go +++ b/pkg/sentry/vfs/anonfs.go @@ -237,7 +237,7 @@ func (fs *anonFilesystem) UnlinkAt(ctx context.Context, rp *ResolvingPath) error } // BoundEndpointAt implements FilesystemImpl.BoundEndpointAt. -func (fs *anonFilesystem) BoundEndpointAt(ctx context.Context, rp *ResolvingPath) (transport.BoundEndpoint, error) { +func (fs *anonFilesystem) BoundEndpointAt(ctx context.Context, rp *ResolvingPath, opts BoundEndpointOptions) (transport.BoundEndpoint, error) { if !rp.Final() { return nil, syserror.ENOTDIR } diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index 20e5bb072..70385a21f 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -495,7 +495,7 @@ type FilesystemImpl interface { // BoundEndpointAt returns the Unix socket endpoint bound at the path rp. // // - If a non-socket file exists at rp, then BoundEndpointAt returns ECONNREFUSED. - BoundEndpointAt(ctx context.Context, rp *ResolvingPath) (transport.BoundEndpoint, error) + BoundEndpointAt(ctx context.Context, rp *ResolvingPath, opts BoundEndpointOptions) (transport.BoundEndpoint, error) // PrependPath prepends a path from vd to vd.Mount().Root() to b. // diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go index 022bac127..53d364c5c 100644 --- a/pkg/sentry/vfs/options.go +++ b/pkg/sentry/vfs/options.go @@ -151,6 +151,24 @@ type SetStatOptions struct { Stat linux.Statx } +// BoundEndpointOptions contains options to VirtualFilesystem.BoundEndpointAt() +// and FilesystemImpl.BoundEndpointAt(). +type BoundEndpointOptions struct { + // Addr is the path of the file whose socket endpoint is being retrieved. + // It is generally irrelevant: most endpoints are stored at a dentry that + // was created through a bind syscall, so the path can be stored on creation. + // However, if the endpoint was created in FilesystemImpl.BoundEndpointAt(), + // then we may not know what the original bind address was. + // + // For example, if connect(2) is called with address "foo" which corresponds + // a remote named socket in goferfs, we need to generate an endpoint wrapping + // that file. In this case, we can use Addr to set the endpoint address to + // "foo". Note that Addr is only a best-effort attempt--we still do not know + // the exact address that was used on the remote fs to bind the socket (it + // may have been "foo", "./foo", etc.). + Addr string +} + // GetxattrOptions contains options to VirtualFilesystem.GetxattrAt(), // FilesystemImpl.GetxattrAt(), FileDescription.Getxattr(), and // FileDescriptionImpl.Getxattr(). diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 9015f2cc1..8d7f8f8af 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -351,33 +351,6 @@ func (vfs *VirtualFilesystem) MknodAt(ctx context.Context, creds *auth.Credentia } } -// BoundEndpointAt gets the bound endpoint at the given path, if one exists. -func (vfs *VirtualFilesystem) BoundEndpointAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation) (transport.BoundEndpoint, error) { - if !pop.Path.Begin.Ok() { - if pop.Path.Absolute { - return nil, syserror.ECONNREFUSED - } - return nil, syserror.ENOENT - } - rp := vfs.getResolvingPath(creds, pop) - for { - bep, err := rp.mount.fs.impl.BoundEndpointAt(ctx, rp) - if err == nil { - vfs.putResolvingPath(rp) - return bep, nil - } - if checkInvariants { - if rp.canHandleError(err) && rp.Done() { - panic(fmt.Sprintf("%T.BoundEndpointAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) - } - } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) - return nil, err - } - } -} - // OpenAt returns a FileDescription providing access to the file at the given // path. A reference is taken on the returned FileDescription. func (vfs *VirtualFilesystem) OpenAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *OpenOptions) (*FileDescription, error) { @@ -675,6 +648,33 @@ func (vfs *VirtualFilesystem) UnlinkAt(ctx context.Context, creds *auth.Credenti } } +// BoundEndpointAt gets the bound endpoint at the given path, if one exists. +func (vfs *VirtualFilesystem) BoundEndpointAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *BoundEndpointOptions) (transport.BoundEndpoint, error) { + if !pop.Path.Begin.Ok() { + if pop.Path.Absolute { + return nil, syserror.ECONNREFUSED + } + return nil, syserror.ENOENT + } + rp := vfs.getResolvingPath(creds, pop) + for { + bep, err := rp.mount.fs.impl.BoundEndpointAt(ctx, rp, *opts) + if err == nil { + vfs.putResolvingPath(rp) + return bep, nil + } + if checkInvariants { + if rp.canHandleError(err) && rp.Done() { + panic(fmt.Sprintf("%T.BoundEndpointAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) + } + } + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return nil, err + } + } +} + // ListxattrAt returns all extended attribute names for the file at the given // path. func (vfs *VirtualFilesystem) ListxattrAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, size uint64) ([]string, error) { -- cgit v1.2.3 From f93f2fda74f31246e8866783f6c4be2318bdedd6 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 28 Apr 2020 17:42:11 -0700 Subject: Deduplicate unix socket Release() method. PiperOrigin-RevId: 308932254 --- pkg/sentry/socket/unix/unix.go | 2 +- pkg/sentry/socket/unix/unix_vfs2.go | 7 ------- 2 files changed, 1 insertion(+), 8 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go index 9aad41c9e..ddd0eda4b 100644 --- a/pkg/sentry/socket/unix/unix.go +++ b/pkg/sentry/socket/unix/unix.go @@ -103,7 +103,7 @@ func (s *socketOpsCommon) DecRef() { } // Release implemements fs.FileOperations.Release. -func (s *SocketOperations) Release() { +func (s *socketOpsCommon) Release() { // Release only decrements a reference on s because s may be referenced in // the abstract socket namespace. s.DecRef() diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go index 315dc274f..433cde9cb 100644 --- a/pkg/sentry/socket/unix/unix_vfs2.go +++ b/pkg/sentry/socket/unix/unix_vfs2.go @@ -273,13 +273,6 @@ func (s *SocketVFS2) Write(ctx context.Context, src usermem.IOSequence, opts vfs }) } -// Release implements vfs.FileDescriptionImpl. -func (s *SocketVFS2) Release() { - // Release only decrements a reference on s because s may be referenced in - // the abstract socket namespace. - s.DecRef() -} - // Readiness implements waiter.Waitable.Readiness. func (s *SocketVFS2) Readiness(mask waiter.EventMask) waiter.EventMask { return s.socketOpsCommon.Readiness(mask) -- cgit v1.2.3 From ce19497c1c0829af6ba56f0cc68e3a4cb33cf1c8 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 28 Apr 2020 20:11:43 -0700 Subject: Fix Unix socket permissions. Enforce write permission checks in BoundEndpointAt, which corresponds to the permission checks in Linux (net/unix/af_unix.c:unix_find_other). Also, create bound socket files with the correct permissions in VFS2. Fixes #2324. PiperOrigin-RevId: 308949084 --- pkg/sentry/fsimpl/ext/filesystem.go | 5 ++- pkg/sentry/fsimpl/kernfs/filesystem.go | 5 ++- pkg/sentry/fsimpl/tmpfs/filesystem.go | 3 ++ pkg/sentry/socket/unix/unix.go | 5 ++- pkg/sentry/socket/unix/unix_vfs2.go | 12 ++++--- pkg/sentry/vfs/anonfs.go | 3 ++ pkg/sentry/vfs/filesystem.go | 8 ++++- test/syscalls/linux/BUILD | 1 + test/syscalls/linux/socket.cc | 61 ++++++++++++++++++++++++++++++++++ 9 files changed, 94 insertions(+), 9 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/ext/filesystem.go b/pkg/sentry/fsimpl/ext/filesystem.go index 6c882d0b6..77b644275 100644 --- a/pkg/sentry/fsimpl/ext/filesystem.go +++ b/pkg/sentry/fsimpl/ext/filesystem.go @@ -486,10 +486,13 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error // BoundEndpointAt implements FilesystemImpl.BoundEndpointAt. func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.BoundEndpointOptions) (transport.BoundEndpoint, error) { - _, _, err := fs.walk(rp, false) + _, inode, err := fs.walk(rp, false) if err != nil { return nil, err } + if err := inode.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil { + return nil, err + } // TODO(b/134676337): Support sockets. return nil, syserror.ECONNREFUSED diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 40c1b2104..4a12ae245 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -768,12 +768,15 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error // BoundEndpointAt implements FilesystemImpl.BoundEndpointAt. func (fs *Filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.BoundEndpointOptions) (transport.BoundEndpoint, error) { fs.mu.RLock() - _, _, err := fs.walkExistingLocked(ctx, rp) + _, inode, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() fs.processDeferredDecRefs() if err != nil { return nil, err } + if err := inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite); err != nil { + return nil, err + } return nil, syserror.ECONNREFUSED } diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 388b98bef..36ffcb592 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -704,6 +704,9 @@ func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath if err != nil { return nil, err } + if err := d.inode.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil { + return nil, err + } switch impl := d.inode.impl.(type) { case *socketFile: return impl.ep, nil diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go index ddd0eda4b..5b29e9d7f 100644 --- a/pkg/sentry/socket/unix/unix.go +++ b/pkg/sentry/socket/unix/unix.go @@ -323,7 +323,10 @@ func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { // Create the socket. // - // TODO(gvisor.dev/issue/2324): Correctly set file permissions. + // Note that the file permissions here are not set correctly (see + // gvisor.dev/issue/2324). There is no convenient way to get permissions + // on the socket referred to by s, so we will leave this discrepancy + // unresolved until VFS2 replaces this code. childDir, err := d.Bind(t, t.FSContext().RootDirectory(), name, bep, fs.FilePermissions{User: fs.PermMask{Read: true}}) if err != nil { return syserr.ErrPortInUse diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go index 433cde9cb..23db93f33 100644 --- a/pkg/sentry/socket/unix/unix_vfs2.go +++ b/pkg/sentry/socket/unix/unix_vfs2.go @@ -197,11 +197,13 @@ func (s *SocketVFS2) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { Start: start, Path: path, } - err := t.Kernel().VFS().MknodAt(t, t.Credentials(), &pop, &vfs.MknodOptions{ - // TODO(gvisor.dev/issue/2324): The file permissions should be taken - // from s and t.FSContext().Umask() (see net/unix/af_unix.c:unix_bind), - // but VFS1 just always uses 0400. Resolve this inconsistency. - Mode: linux.S_IFSOCK | 0400, + stat, err := s.vfsfd.Stat(t, vfs.StatOptions{Mask: linux.STATX_MODE}) + if err != nil { + return syserr.FromError(err) + } + err = t.Kernel().VFS().MknodAt(t, t.Credentials(), &pop, &vfs.MknodOptions{ + // File permissions correspond to net/unix/af_unix.c:unix_bind. + Mode: linux.FileMode(linux.S_IFSOCK | uint(stat.Mode)&^t.FSContext().Umask()), Endpoint: bep, }) if err == syserror.EEXIST { diff --git a/pkg/sentry/vfs/anonfs.go b/pkg/sentry/vfs/anonfs.go index b1a998590..981bd8caa 100644 --- a/pkg/sentry/vfs/anonfs.go +++ b/pkg/sentry/vfs/anonfs.go @@ -241,6 +241,9 @@ func (fs *anonFilesystem) BoundEndpointAt(ctx context.Context, rp *ResolvingPath if !rp.Final() { return nil, syserror.ENOTDIR } + if err := GenericCheckPermissions(rp.Credentials(), MayWrite, anonFileMode, anonFileUID, anonFileGID); err != nil { + return nil, err + } return nil, syserror.ECONNREFUSED } diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index 70385a21f..1edd584c9 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -494,7 +494,13 @@ type FilesystemImpl interface { // BoundEndpointAt returns the Unix socket endpoint bound at the path rp. // - // - If a non-socket file exists at rp, then BoundEndpointAt returns ECONNREFUSED. + // Errors: + // + // - If the file does not have write permissions, then BoundEndpointAt + // returns EACCES. + // + // - If a non-socket file exists at rp, then BoundEndpointAt returns + // ECONNREFUSED. BoundEndpointAt(ctx context.Context, rp *ResolvingPath, opts BoundEndpointOptions) (transport.BoundEndpoint, error) // PrependPath prepends a path from vd to vd.Mount().Root() to b. diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index d9095c95f..837e56042 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -365,6 +365,7 @@ cc_binary( ":socket_test_util", "//test/util:file_descriptor", gtest, + "//test/util:temp_umask", "//test/util:test_main", "//test/util:test_util", ], diff --git a/test/syscalls/linux/socket.cc b/test/syscalls/linux/socket.cc index 3a07ac8d2..703d594a2 100644 --- a/test/syscalls/linux/socket.cc +++ b/test/syscalls/linux/socket.cc @@ -13,11 +13,14 @@ // limitations under the License. #include +#include +#include #include #include "gtest/gtest.h" #include "test/syscalls/linux/socket_test_util.h" #include "test/util/file_descriptor.h" +#include "test/util/temp_umask.h" #include "test/util/test_util.h" namespace gvisor { @@ -58,11 +61,69 @@ TEST(SocketTest, ProtocolInet) { } } +TEST(SocketTest, UnixSocketFileMode) { + // TODO(gvisor.dev/issue/1624): Re-enable this test once VFS1 is deleted. It + // should pass in VFS2. + SKIP_IF(IsRunningOnGvisor()); + + FileDescriptor bound = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_UNIX, SOCK_STREAM, PF_UNIX)); + + // The permissions of the file created with bind(2) should be defined by the + // permissions of the bound socket and the umask. + mode_t sock_perm = 0765, mask = 0123; + ASSERT_THAT(fchmod(bound.get(), sock_perm), SyscallSucceeds()); + TempUmask m(mask); + + struct sockaddr_un addr = + ASSERT_NO_ERRNO_AND_VALUE(UniqueUnixAddr(/*abstract=*/false, AF_UNIX)); + ASSERT_THAT(bind(bound.get(), reinterpret_cast(&addr), + sizeof(addr)), + SyscallSucceeds()); + + struct stat statbuf = {}; + ASSERT_THAT(stat(addr.sun_path, &statbuf), SyscallSucceeds()); + EXPECT_EQ(statbuf.st_mode, S_IFSOCK | sock_perm & ~mask); +} + +TEST(SocketTest, UnixConnectNeedsWritePerm) { + // TODO(gvisor.dev/issue/1624): Re-enable this test once VFS1 is deleted. It + // should succeed in VFS2. + SKIP_IF(IsRunningOnGvisor()); + + FileDescriptor bound = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_UNIX, SOCK_STREAM, PF_UNIX)); + + struct sockaddr_un addr = + ASSERT_NO_ERRNO_AND_VALUE(UniqueUnixAddr(/*abstract=*/false, AF_UNIX)); + ASSERT_THAT(bind(bound.get(), reinterpret_cast(&addr), + sizeof(addr)), + SyscallSucceeds()); + ASSERT_THAT(listen(bound.get(), 1), SyscallSucceeds()); + + // Connect should fail without write perms. + ASSERT_THAT(chmod(addr.sun_path, 0500), SyscallSucceeds()); + FileDescriptor client = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_UNIX, SOCK_STREAM, PF_UNIX)); + EXPECT_THAT(connect(client.get(), reinterpret_cast(&addr), + sizeof(addr)), + SyscallFailsWithErrno(EACCES)); + + // Connect should succeed with write perms. + ASSERT_THAT(chmod(addr.sun_path, 0200), SyscallSucceeds()); + EXPECT_THAT(connect(client.get(), reinterpret_cast(&addr), + sizeof(addr)), + SyscallSucceeds()); +} + using SocketOpenTest = ::testing::TestWithParam; // UDS cannot be opened. TEST_P(SocketOpenTest, Unix) { // FIXME(b/142001530): Open incorrectly succeeds on gVisor. + // + // TODO(gvisor.dev/issue/1624): Re-enable this test once VFS1 is deleted. It + // should succeed in VFS2. SKIP_IF(IsRunningOnGvisor()); FileDescriptor bound = -- cgit v1.2.3 From a105d185ff9fc24f5bf0c1ca28cbc0f7ec7c4ea5 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Wed, 29 Apr 2020 13:36:29 -0700 Subject: iptables: don't pollute logs The netfilter package uses logs to make debugging the (de)serialization of structs easier. This generates a lot of (usually irrelevant) logs. Logging is now hidden behind a debug flag. PiperOrigin-RevId: 309087115 --- pkg/sentry/socket/netfilter/netfilter.go | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 878f81fd5..72d093aa8 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -53,9 +53,14 @@ type metadata struct { Size uint32 } +// enableLogging controls whether to log the (de)serialization of netfilter +// structs between userspace and netstack. These logs are useful when +// developing iptables, but can pollute sentry logs otherwise. +const enableLogging = false + // nflog logs messages related to the writing and reading of iptables. func nflog(format string, args ...interface{}) { - if log.IsLogging(log.Debug) { + if enableLogging && log.IsLogging(log.Debug) { log.Debugf("netfilter: "+format, args...) } } -- cgit v1.2.3 From ef94401955bbcfb5296daa29ab373423cdd289db Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Wed, 29 Apr 2020 14:34:28 -0700 Subject: Add read/write timeouts for VFS2 socket files. Updates #1476 PiperOrigin-RevId: 309098590 --- pkg/sentry/syscalls/linux/vfs2/read_write.go | 77 +++++++++++++++++++++++++--- 1 file changed, 69 insertions(+), 8 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/vfs2/read_write.go b/pkg/sentry/syscalls/linux/vfs2/read_write.go index 6c6998f45..3a7ef24f5 100644 --- a/pkg/sentry/syscalls/linux/vfs2/read_write.go +++ b/pkg/sentry/syscalls/linux/vfs2/read_write.go @@ -15,9 +15,13 @@ package vfs2 import ( + "time" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" + ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" + "gvisor.dev/gvisor/pkg/sentry/socket" slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" @@ -88,7 +92,12 @@ func Readv(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall func read(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { n, err := file.Read(t, dst, opts) - if err != syserror.ErrWouldBlock || file.StatusFlags()&linux.O_NONBLOCK != 0 { + if err != syserror.ErrWouldBlock { + return n, err + } + + allowBlock, deadline, hasDeadline := blockPolicy(t, file) + if !allowBlock { return n, err } @@ -108,7 +117,12 @@ func read(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, opt if err != syserror.ErrWouldBlock { break } - if err := t.Block(ch); err != nil { + + // Wait for a notification that we should retry. + if err = t.BlockWithDeadline(ch, hasDeadline, deadline); err != nil { + if err == syserror.ETIMEDOUT { + err = syserror.ErrWouldBlock + } break } } @@ -233,7 +247,12 @@ func Preadv2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca func pread(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { n, err := file.PRead(t, dst, offset, opts) - if err != syserror.ErrWouldBlock || file.StatusFlags()&linux.O_NONBLOCK != 0 { + if err != syserror.ErrWouldBlock { + return n, err + } + + allowBlock, deadline, hasDeadline := blockPolicy(t, file) + if !allowBlock { return n, err } @@ -253,7 +272,12 @@ func pread(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, of if err != syserror.ErrWouldBlock { break } - if err := t.Block(ch); err != nil { + + // Wait for a notification that we should retry. + if err = t.BlockWithDeadline(ch, hasDeadline, deadline); err != nil { + if err == syserror.ETIMEDOUT { + err = syserror.ErrWouldBlock + } break } } @@ -320,7 +344,12 @@ func Writev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal func write(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { n, err := file.Write(t, src, opts) - if err != syserror.ErrWouldBlock || file.StatusFlags()&linux.O_NONBLOCK != 0 { + if err != syserror.ErrWouldBlock { + return n, err + } + + allowBlock, deadline, hasDeadline := blockPolicy(t, file) + if !allowBlock { return n, err } @@ -340,7 +369,12 @@ func write(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, op if err != syserror.ErrWouldBlock { break } - if err := t.Block(ch); err != nil { + + // Wait for a notification that we should retry. + if err = t.BlockWithDeadline(ch, hasDeadline, deadline); err != nil { + if err == syserror.ETIMEDOUT { + err = syserror.ErrWouldBlock + } break } } @@ -465,7 +499,12 @@ func Pwritev2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc func pwrite(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { n, err := file.PWrite(t, src, offset, opts) - if err != syserror.ErrWouldBlock || file.StatusFlags()&linux.O_NONBLOCK != 0 { + if err != syserror.ErrWouldBlock { + return n, err + } + + allowBlock, deadline, hasDeadline := blockPolicy(t, file) + if !allowBlock { return n, err } @@ -485,7 +524,12 @@ func pwrite(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, o if err != syserror.ErrWouldBlock { break } - if err := t.Block(ch); err != nil { + + // Wait for a notification that we should retry. + if err = t.BlockWithDeadline(ch, hasDeadline, deadline); err != nil { + if err == syserror.ETIMEDOUT { + err = syserror.ErrWouldBlock + } break } } @@ -494,6 +538,23 @@ func pwrite(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, o return total, err } +func blockPolicy(t *kernel.Task, file *vfs.FileDescription) (allowBlock bool, deadline ktime.Time, hasDeadline bool) { + if file.StatusFlags()&linux.O_NONBLOCK != 0 { + return false, ktime.Time{}, false + } + // Sockets support read/write timeouts. + if s, ok := file.Impl().(socket.SocketVFS2); ok { + dl := s.RecvTimeout() + if dl < 0 { + return false, ktime.Time{}, false + } + if dl > 0 { + return true, t.Kernel().MonotonicClock().Now().Add(time.Duration(dl) * time.Nanosecond), true + } + } + return true, ktime.Time{}, false +} + // Lseek implements Linux syscall lseek(2). func Lseek(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { fd := args[0].Int() -- cgit v1.2.3 From 442fde405d86c555ac09994772c85ca15a3b4fc7 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Thu, 30 Apr 2020 09:46:36 -0700 Subject: Fix proc net bugs in VFS2. The /proc/net/udp header was missing, and /proc/sys/net was set up as /proc/sys/net/net. Discovered while trying to run networking tests for VFS2. PiperOrigin-RevId: 309243758 --- pkg/sentry/fsimpl/proc/task_net.go | 2 ++ pkg/sentry/fsimpl/proc/tasks_sys.go | 4 +--- 2 files changed, 3 insertions(+), 3 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/proc/task_net.go b/pkg/sentry/fsimpl/proc/task_net.go index 6595fcee6..9c329341a 100644 --- a/pkg/sentry/fsimpl/proc/task_net.go +++ b/pkg/sentry/fsimpl/proc/task_net.go @@ -511,6 +511,8 @@ func (d *netUDPData) Generate(ctx context.Context, buf *bytes.Buffer) error { // degrade gracefully and retrieve what we can. t := kernel.TaskFromContext(ctx) + buf.WriteString(" sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode ref pointer drops \n") + for _, se := range d.kernel.ListSockets() { s := se.SockVFS2 if !s.TryIncRef() { diff --git a/pkg/sentry/fsimpl/proc/tasks_sys.go b/pkg/sentry/fsimpl/proc/tasks_sys.go index f08668ca2..0e90e02fe 100644 --- a/pkg/sentry/fsimpl/proc/tasks_sys.go +++ b/pkg/sentry/fsimpl/proc/tasks_sys.go @@ -112,9 +112,7 @@ func newSysNetDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) } } - return kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "net": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, contents), - }) + return kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, contents) } // mmapMinAddrData implements vfs.DynamicBytesSource for -- cgit v1.2.3 From bae30a0c69dbc2bf93db9598e1cf07ef8757a571 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Thu, 30 Apr 2020 14:18:31 -0700 Subject: Implement waiter.Waitable methods on VFS2 host inodes. This fixes bash in Ubuntu. Updates #1672. PiperOrigin-RevId: 309298252 --- pkg/sentry/fsimpl/host/host.go | 53 ++++++++++++++++++++++++++++++++++++++---- 1 file changed, 48 insertions(+), 5 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 05538ea42..33bd205af 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -24,6 +24,7 @@ import ( "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fdnotifier" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" @@ -35,6 +36,7 @@ import ( "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" ) // filesystemType implements vfs.FilesystemType. @@ -88,11 +90,12 @@ func ImportFD(ctx context.Context, mnt *vfs.Mount, hostFD int, isTTY bool) (*vfs seekable := err != syserror.ESPIPE i := &inode{ - hostFD: hostFD, - seekable: seekable, - isTTY: isTTY, - canMap: canMap(uint32(fileType)), - ino: fs.NextIno(), + hostFD: hostFD, + seekable: seekable, + isTTY: isTTY, + canMap: canMap(uint32(fileType)), + wouldBlock: wouldBlock(uint32(fileType)), + ino: fs.NextIno(), // For simplicity, set offset to 0. Technically, we should use the existing // offset on the host if the file is seekable. offset: 0, @@ -103,6 +106,17 @@ func ImportFD(ctx context.Context, mnt *vfs.Mount, hostFD int, isTTY bool) (*vfs panic("files that can return EWOULDBLOCK (sockets, pipes, etc.) cannot be memory mapped") } + // If the hostFD would block, we must set it to non-blocking and handle + // blocking behavior in the sentry. + if i.wouldBlock { + if err := syscall.SetNonblock(i.hostFD, true); err != nil { + return nil, err + } + if err := fdnotifier.AddFD(int32(i.hostFD), &i.queue); err != nil { + return nil, err + } + } + d := &kernfs.Dentry{} d.Init(i) // i.open will take a reference on d. @@ -125,6 +139,12 @@ type inode struct { // This field is initialized at creation time and is immutable. hostFD int + // wouldBlock is true if the host FD would return EWOULDBLOCK for + // operations that would block. + // + // This field is initialized at creation time and is immutable. + wouldBlock bool + // seekable is false if the host fd points to a file representing a stream, // e.g. a socket or a pipe. Such files are not seekable and can return // EWOULDBLOCK for I/O operations. @@ -152,6 +172,9 @@ type inode struct { // offset specifies the current file offset. offset int64 + + // Event queue for blocking operations. + queue waiter.Queue } // Note that these flags may become out of date, since they can be modified @@ -354,6 +377,9 @@ func (i *inode) DecRef() { // Destroy implements kernfs.Inode. func (i *inode) Destroy() { + if i.wouldBlock { + fdnotifier.RemoveFD(int32(i.hostFD)) + } if err := unix.Close(i.hostFD); err != nil { log.Warningf("failed to close host fd %d: %v", i.hostFD, err) } @@ -614,3 +640,20 @@ func (f *fileDescription) ConfigureMMap(_ context.Context, opts *memmap.MMapOpts // TODO(gvisor.dev/issue/1672): Implement ConfigureMMap and Mappable interface. return syserror.ENODEV } + +// EventRegister implements waiter.Waitable.EventRegister. +func (f *fileDescription) EventRegister(e *waiter.Entry, mask waiter.EventMask) { + f.inode.queue.EventRegister(e, mask) + fdnotifier.UpdateFD(int32(f.inode.hostFD)) +} + +// EventUnregister implements waiter.Waitable.EventUnregister. +func (f *fileDescription) EventUnregister(e *waiter.Entry) { + f.inode.queue.EventUnregister(e) + fdnotifier.UpdateFD(int32(f.inode.hostFD)) +} + +// Readiness uses the poll() syscall to check the status of the underlying FD. +func (f *fileDescription) Readiness(mask waiter.EventMask) waiter.EventMask { + return fdnotifier.NonBlockingPoll(int32(f.inode.hostFD), mask) +} -- cgit v1.2.3 From 01beec3bb457a2a3a7313c7fe6dc795817f47746 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Thu, 30 Apr 2020 16:03:04 -0700 Subject: Add gofer.InternalFilesystemOptions.LeakConnection. PiperOrigin-RevId: 309317605 --- pkg/sentry/fsimpl/gofer/gofer.go | 28 +++++++++++++++++++++++++--- 1 file changed, 25 insertions(+), 3 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 3235816c2..2881c7bdd 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -72,7 +72,8 @@ type filesystem struct { mfp pgalloc.MemoryFileProvider // Immutable options. - opts filesystemOptions + opts filesystemOptions + iopts InternalFilesystemOptions // client is the client used by this filesystem. client is immutable. client *p9.Client @@ -209,6 +210,16 @@ const ( InteropModeShared ) +// InternalFilesystemOptions may be passed as +// vfs.GetFilesystemOptions.InternalData to FilesystemType.GetFilesystem. +type InternalFilesystemOptions struct { + // If LeakConnection is true, do not close the connection to the server + // when the Filesystem is released. This is necessary for deployments in + // which servers can handle only a single client and report failure if that + // client disconnects. + LeakConnection bool +} + // Name implements vfs.FilesystemType.Name. func (FilesystemType) Name() string { return Name @@ -347,6 +358,14 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt return nil, nil, syserror.EINVAL } + // Handle internal options. + iopts, ok := opts.InternalData.(InternalFilesystemOptions) + if opts.InternalData != nil && !ok { + ctx.Warningf("gofer.FilesystemType.GetFilesystem: GetFilesystemOptions.InternalData has type %T, wanted gofer.InternalFilesystemOptions", opts.InternalData) + return nil, nil, syserror.EINVAL + } + // If !ok, iopts being the zero value is correct. + // Establish a connection with the server. conn, err := unet.NewSocket(fsopts.fd) if err != nil { @@ -383,6 +402,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt fs := &filesystem{ mfp: mfp, opts: fsopts, + iopts: iopts, uid: creds.EffectiveKUID, gid: creds.EffectiveKGID, client: client, @@ -440,8 +460,10 @@ func (fs *filesystem) Release() { // fs. fs.syncMu.Unlock() - // Close the connection to the server. This implicitly clunks all fids. - fs.client.Close() + if !fs.iopts.LeakConnection { + // Close the connection to the server. This implicitly clunks all fids. + fs.client.Close() + } } // dentry implements vfs.DentryImpl. -- cgit v1.2.3 From 82bae30ceea0984c94af3085866b58ec9e69ea67 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 1 May 2020 12:53:15 -0700 Subject: Port netstack, hostinet, and netlink sockets to VFS2. All three follow the same pattern: 1. Refactor VFS1 sockets into socketOpsCommon, so that most of the methods can be shared with VFS2. 2. Create a FileDescriptionImpl with the corresponding socket operations, rewriting the few that cannot be shared with VFS1. 3. Set up a VFS2 socket provider that creates a socket by setting up a dentry in the global Kernel.socketMount and connecting it with a new FileDescription. This mostly completes the work for porting sockets to VFS2, and many syscall tests can be enabled as a result. There are several networking-related syscall tests that are still not passing: 1. net gofer tests 2. socketpair gofer tests 2. sendfile tests (splice is not implemented in VFS2 yet) Updates #1478, #1484, #1485 PiperOrigin-RevId: 309457331 --- pkg/sentry/fsimpl/gofer/socket.go | 9 +- pkg/sentry/fsimpl/host/host.go | 2 +- pkg/sentry/fsimpl/host/socket.go | 53 +++-- pkg/sentry/fsimpl/sockfs/sockfs.go | 5 + pkg/sentry/socket/hostinet/BUILD | 5 + pkg/sentry/socket/hostinet/socket.go | 105 ++++++--- pkg/sentry/socket/hostinet/socket_unsafe.go | 10 +- pkg/sentry/socket/hostinet/socket_vfs2.go | 183 ++++++++++++++++ pkg/sentry/socket/netlink/BUILD | 5 + pkg/sentry/socket/netlink/provider.go | 5 + pkg/sentry/socket/netlink/provider_vfs2.go | 71 ++++++ pkg/sentry/socket/netlink/socket.go | 72 +++--- pkg/sentry/socket/netlink/socket_vfs2.go | 138 ++++++++++++ pkg/sentry/socket/netstack/BUILD | 5 + pkg/sentry/socket/netstack/netstack.go | 72 +++--- pkg/sentry/socket/netstack/netstack_vfs2.go | 327 ++++++++++++++++++++++++++++ pkg/sentry/socket/netstack/provider.go | 4 + pkg/sentry/socket/netstack/provider_vfs2.go | 141 ++++++++++++ pkg/sentry/socket/unix/unix_vfs2.go | 4 +- 19 files changed, 1090 insertions(+), 126 deletions(-) create mode 100644 pkg/sentry/socket/hostinet/socket_vfs2.go create mode 100644 pkg/sentry/socket/netlink/provider_vfs2.go create mode 100644 pkg/sentry/socket/netlink/socket_vfs2.go create mode 100644 pkg/sentry/socket/netstack/netstack_vfs2.go create mode 100644 pkg/sentry/socket/netstack/provider_vfs2.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/socket.go b/pkg/sentry/fsimpl/gofer/socket.go index 73835df91..d6dbe9092 100644 --- a/pkg/sentry/fsimpl/gofer/socket.go +++ b/pkg/sentry/fsimpl/gofer/socket.go @@ -87,7 +87,9 @@ func (e *endpoint) BidirectionalConnect(ctx context.Context, ce transport.Connec returnConnect(c, c) ce.Unlock() - c.Init() + if err := c.Init(); err != nil { + return syserr.FromError(err) + } return nil } @@ -99,7 +101,10 @@ func (e *endpoint) UnidirectionalConnect(ctx context.Context) (transport.Connect if err != nil { return nil, err } - c.Init() + + if err := c.Init(); err != nil { + return nil, syserr.FromError(err) + } // We don't need the receiver. c.CloseRecv() diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 33bd205af..34fbc69af 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -411,7 +411,7 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount) (*vfs.F return nil, syserror.ENOTTY } - ep, err := newEndpoint(ctx, i.hostFD) + ep, err := newEndpoint(ctx, i.hostFD, &i.queue) if err != nil { return nil, err } diff --git a/pkg/sentry/fsimpl/host/socket.go b/pkg/sentry/fsimpl/host/socket.go index 39dd624a0..38f1fbfba 100644 --- a/pkg/sentry/fsimpl/host/socket.go +++ b/pkg/sentry/fsimpl/host/socket.go @@ -34,17 +34,16 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) -// Create a new host-backed endpoint from the given fd. -func newEndpoint(ctx context.Context, hostFD int) (transport.Endpoint, error) { +// Create a new host-backed endpoint from the given fd and its corresponding +// notification queue. +func newEndpoint(ctx context.Context, hostFD int, queue *waiter.Queue) (transport.Endpoint, error) { // Set up an external transport.Endpoint using the host fd. addr := fmt.Sprintf("hostfd:[%d]", hostFD) - var q waiter.Queue - e, err := NewConnectedEndpoint(ctx, hostFD, &q, addr, true /* saveable */) + e, err := NewConnectedEndpoint(ctx, hostFD, addr, true /* saveable */) if err != nil { return nil, err.ToError() } - e.Init() - ep := transport.NewExternal(ctx, e.stype, uniqueid.GlobalProviderFromContext(ctx), &q, e, e) + ep := transport.NewExternal(ctx, e.stype, uniqueid.GlobalProviderFromContext(ctx), queue, e, e) return ep, nil } @@ -77,8 +76,6 @@ type ConnectedEndpoint struct { // addr is the address at which this endpoint is bound. addr string - queue *waiter.Queue - // sndbuf is the size of the send buffer. // // N.B. When this is smaller than the host size, we present it via @@ -134,11 +131,10 @@ func (c *ConnectedEndpoint) init() *syserr.Error { // The caller is responsible for calling Init(). Additionaly, Release needs to // be called twice because ConnectedEndpoint is both a transport.Receiver and // transport.ConnectedEndpoint. -func NewConnectedEndpoint(ctx context.Context, hostFD int, queue *waiter.Queue, addr string, saveable bool) (*ConnectedEndpoint, *syserr.Error) { +func NewConnectedEndpoint(ctx context.Context, hostFD int, addr string, saveable bool) (*ConnectedEndpoint, *syserr.Error) { e := ConnectedEndpoint{ - fd: hostFD, - addr: addr, - queue: queue, + fd: hostFD, + addr: addr, } if err := e.init(); err != nil { @@ -151,13 +147,6 @@ func NewConnectedEndpoint(ctx context.Context, hostFD int, queue *waiter.Queue, return &e, nil } -// Init will do the initialization required without holding other locks. -func (c *ConnectedEndpoint) Init() { - if err := fdnotifier.AddFD(int32(c.fd), c.queue); err != nil { - panic(err) - } -} - // Send implements transport.ConnectedEndpoint.Send. func (c *ConnectedEndpoint) Send(data [][]byte, controlMessages transport.ControlMessages, from tcpip.FullAddress) (int64, bool, *syserr.Error) { c.mu.RLock() @@ -332,7 +321,6 @@ func (c *ConnectedEndpoint) RecvMaxQueueSize() int64 { } func (c *ConnectedEndpoint) destroyLocked() { - fdnotifier.RemoveFD(int32(c.fd)) c.fd = -1 } @@ -350,14 +338,20 @@ func (c *ConnectedEndpoint) Release() { func (c *ConnectedEndpoint) CloseUnread() {} // SCMConnectedEndpoint represents an endpoint backed by a host fd that was -// passed through a gofer Unix socket. It is almost the same as -// ConnectedEndpoint, with the following differences: +// passed through a gofer Unix socket. It resembles ConnectedEndpoint, with the +// following differences: // - SCMConnectedEndpoint is not saveable, because the host cannot guarantee // the same descriptor number across S/R. -// - SCMConnectedEndpoint holds ownership of its fd and is responsible for -// closing it. +// - SCMConnectedEndpoint holds ownership of its fd and notification queue. type SCMConnectedEndpoint struct { ConnectedEndpoint + + queue *waiter.Queue +} + +// Init will do the initialization required without holding other locks. +func (e *SCMConnectedEndpoint) Init() error { + return fdnotifier.AddFD(int32(e.fd), e.queue) } // Release implements transport.ConnectedEndpoint.Release and @@ -368,6 +362,7 @@ func (e *SCMConnectedEndpoint) Release() { if err := syscall.Close(e.fd); err != nil { log.Warningf("Failed to close host fd %d: %v", err) } + fdnotifier.RemoveFD(int32(e.fd)) e.destroyLocked() e.mu.Unlock() }) @@ -380,11 +375,13 @@ func (e *SCMConnectedEndpoint) Release() { // be called twice because ConnectedEndpoint is both a transport.Receiver and // transport.ConnectedEndpoint. func NewSCMEndpoint(ctx context.Context, hostFD int, queue *waiter.Queue, addr string) (*SCMConnectedEndpoint, *syserr.Error) { - e := SCMConnectedEndpoint{ConnectedEndpoint{ - fd: hostFD, - addr: addr, + e := SCMConnectedEndpoint{ + ConnectedEndpoint: ConnectedEndpoint{ + fd: hostFD, + addr: addr, + }, queue: queue, - }} + } if err := e.init(); err != nil { return nil, err diff --git a/pkg/sentry/fsimpl/sockfs/sockfs.go b/pkg/sentry/fsimpl/sockfs/sockfs.go index 271134af8..dac2389fc 100644 --- a/pkg/sentry/fsimpl/sockfs/sockfs.go +++ b/pkg/sentry/fsimpl/sockfs/sockfs.go @@ -74,6 +74,11 @@ func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentr } // NewDentry constructs and returns a sockfs dentry. +// +// TODO(gvisor.dev/issue/1476): Currently, we are using +// sockfs.filesystem.NextIno() to get inode numbers. We should use +// device-specific numbers, so that we are not using the same generator for +// netstack, unix, etc. func NewDentry(creds *auth.Credentials, ino uint64) *vfs.Dentry { // File mode matches net/socket.c:sock_alloc. filemode := linux.FileMode(linux.S_IFSOCK | 0600) diff --git a/pkg/sentry/socket/hostinet/BUILD b/pkg/sentry/socket/hostinet/BUILD index 023bad156..deedd35f7 100644 --- a/pkg/sentry/socket/hostinet/BUILD +++ b/pkg/sentry/socket/hostinet/BUILD @@ -10,6 +10,7 @@ go_library( "save_restore.go", "socket.go", "socket_unsafe.go", + "socket_vfs2.go", "sockopt_impl.go", "stack.go", ], @@ -25,11 +26,15 @@ go_library( "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", + "//pkg/sentry/fsimpl/kernfs", + "//pkg/sentry/fsimpl/sockfs", + "//pkg/sentry/hostfd", "//pkg/sentry/inet", "//pkg/sentry/kernel", "//pkg/sentry/kernel/time", "//pkg/sentry/socket", "//pkg/sentry/socket/control", + "//pkg/sentry/vfs", "//pkg/syserr", "//pkg/syserror", "//pkg/tcpip/stack", diff --git a/pkg/sentry/socket/hostinet/socket.go b/pkg/sentry/socket/hostinet/socket.go index 22f78d2e2..b49433326 100644 --- a/pkg/sentry/socket/hostinet/socket.go +++ b/pkg/sentry/socket/hostinet/socket.go @@ -25,6 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/fdnotifier" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/safemem" + "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -49,6 +50,8 @@ const ( maxControlLen = 1024 ) +// LINT.IfChange + // socketOperations implements fs.FileOperations and socket.Socket for a socket // implemented using a host socket. type socketOperations struct { @@ -59,23 +62,37 @@ type socketOperations struct { fsutil.FileNoSplice `state:"nosave"` fsutil.FileNoopFlush `state:"nosave"` fsutil.FileUseInodeUnstableAttr `state:"nosave"` + + socketOpsCommon +} + +// socketOpsCommon contains the socket operations common to VFS1 and VFS2. +// +// +stateify savable +type socketOpsCommon struct { socket.SendReceiveTimeout family int // Read-only. stype linux.SockType // Read-only. protocol int // Read-only. - fd int // must be O_NONBLOCK queue waiter.Queue + + // fd is the host socket fd. It must have O_NONBLOCK, so that operations + // will return EWOULDBLOCK instead of blocking on the host. This allows us to + // handle blocking behavior independently in the sentry. + fd int } var _ = socket.Socket(&socketOperations{}) func newSocketFile(ctx context.Context, family int, stype linux.SockType, protocol int, fd int, nonblock bool) (*fs.File, *syserr.Error) { s := &socketOperations{ - family: family, - stype: stype, - protocol: protocol, - fd: fd, + socketOpsCommon: socketOpsCommon{ + family: family, + stype: stype, + protocol: protocol, + fd: fd, + }, } if err := fdnotifier.AddFD(int32(fd), &s.queue); err != nil { return nil, syserr.FromError(err) @@ -86,28 +103,33 @@ func newSocketFile(ctx context.Context, family int, stype linux.SockType, protoc } // Release implements fs.FileOperations.Release. -func (s *socketOperations) Release() { +func (s *socketOpsCommon) Release() { fdnotifier.RemoveFD(int32(s.fd)) syscall.Close(s.fd) } // Readiness implements waiter.Waitable.Readiness. -func (s *socketOperations) Readiness(mask waiter.EventMask) waiter.EventMask { +func (s *socketOpsCommon) Readiness(mask waiter.EventMask) waiter.EventMask { return fdnotifier.NonBlockingPoll(int32(s.fd), mask) } // EventRegister implements waiter.Waitable.EventRegister. -func (s *socketOperations) EventRegister(e *waiter.Entry, mask waiter.EventMask) { +func (s *socketOpsCommon) EventRegister(e *waiter.Entry, mask waiter.EventMask) { s.queue.EventRegister(e, mask) fdnotifier.UpdateFD(int32(s.fd)) } // EventUnregister implements waiter.Waitable.EventUnregister. -func (s *socketOperations) EventUnregister(e *waiter.Entry) { +func (s *socketOpsCommon) EventUnregister(e *waiter.Entry) { s.queue.EventUnregister(e) fdnotifier.UpdateFD(int32(s.fd)) } +// Ioctl implements fs.FileOperations.Ioctl. +func (s *socketOperations) Ioctl(ctx context.Context, _ *fs.File, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { + return ioctl(ctx, s.fd, io, args) +} + // Read implements fs.FileOperations.Read. func (s *socketOperations) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, _ int64) (int64, error) { n, err := dst.CopyOutFrom(ctx, safemem.ReaderFunc(func(dsts safemem.BlockSeq) (uint64, error) { @@ -155,7 +177,7 @@ func (s *socketOperations) Write(ctx context.Context, _ *fs.File, src usermem.IO } // Connect implements socket.Socket.Connect. -func (s *socketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking bool) *syserr.Error { +func (s *socketOpsCommon) Connect(t *kernel.Task, sockaddr []byte, blocking bool) *syserr.Error { if len(sockaddr) > sizeofSockaddr { sockaddr = sockaddr[:sizeofSockaddr] } @@ -195,7 +217,7 @@ func (s *socketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking boo } // Accept implements socket.Socket.Accept. -func (s *socketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (int32, linux.SockAddr, uint32, *syserr.Error) { +func (s *socketOpsCommon) Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (int32, linux.SockAddr, uint32, *syserr.Error) { var peerAddr linux.SockAddr var peerAddrBuf []byte var peerAddrlen uint32 @@ -209,7 +231,7 @@ func (s *socketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, } // Conservatively ignore all flags specified by the application and add - // SOCK_NONBLOCK since socketOperations requires it. + // SOCK_NONBLOCK since socketOpsCommon requires it. fd, syscallErr := accept4(s.fd, peerAddrPtr, peerAddrlenPtr, syscall.SOCK_NONBLOCK|syscall.SOCK_CLOEXEC) if blocking { var ch chan struct{} @@ -235,23 +257,41 @@ func (s *socketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, return 0, peerAddr, peerAddrlen, syserr.FromError(syscallErr) } - f, err := newSocketFile(t, s.family, s.stype, s.protocol, fd, flags&syscall.SOCK_NONBLOCK != 0) - if err != nil { - syscall.Close(fd) - return 0, nil, 0, err - } - defer f.DecRef() + var ( + kfd int32 + kerr error + ) + if kernel.VFS2Enabled { + f, err := newVFS2Socket(t, s.family, s.stype, s.protocol, fd, uint32(flags&syscall.SOCK_NONBLOCK)) + if err != nil { + syscall.Close(fd) + return 0, nil, 0, err + } + defer f.DecRef() - kfd, kerr := t.NewFDFrom(0, f, kernel.FDFlags{ - CloseOnExec: flags&syscall.SOCK_CLOEXEC != 0, - }) - t.Kernel().RecordSocket(f) + kfd, kerr = t.NewFDFromVFS2(0, f, kernel.FDFlags{ + CloseOnExec: flags&syscall.SOCK_CLOEXEC != 0, + }) + t.Kernel().RecordSocketVFS2(f) + } else { + f, err := newSocketFile(t, s.family, s.stype, s.protocol, fd, flags&syscall.SOCK_NONBLOCK != 0) + if err != nil { + syscall.Close(fd) + return 0, nil, 0, err + } + defer f.DecRef() + + kfd, kerr = t.NewFDFrom(0, f, kernel.FDFlags{ + CloseOnExec: flags&syscall.SOCK_CLOEXEC != 0, + }) + t.Kernel().RecordSocket(f) + } return kfd, peerAddr, peerAddrlen, syserr.FromError(kerr) } // Bind implements socket.Socket.Bind. -func (s *socketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { +func (s *socketOpsCommon) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { if len(sockaddr) > sizeofSockaddr { sockaddr = sockaddr[:sizeofSockaddr] } @@ -264,12 +304,12 @@ func (s *socketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { } // Listen implements socket.Socket.Listen. -func (s *socketOperations) Listen(t *kernel.Task, backlog int) *syserr.Error { +func (s *socketOpsCommon) Listen(t *kernel.Task, backlog int) *syserr.Error { return syserr.FromError(syscall.Listen(s.fd, backlog)) } // Shutdown implements socket.Socket.Shutdown. -func (s *socketOperations) Shutdown(t *kernel.Task, how int) *syserr.Error { +func (s *socketOpsCommon) Shutdown(t *kernel.Task, how int) *syserr.Error { switch how { case syscall.SHUT_RD, syscall.SHUT_WR, syscall.SHUT_RDWR: return syserr.FromError(syscall.Shutdown(s.fd, how)) @@ -279,7 +319,7 @@ func (s *socketOperations) Shutdown(t *kernel.Task, how int) *syserr.Error { } // GetSockOpt implements socket.Socket.GetSockOpt. -func (s *socketOperations) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) { +func (s *socketOpsCommon) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) { if outLen < 0 { return nil, syserr.ErrInvalidArgument } @@ -328,7 +368,7 @@ func (s *socketOperations) GetSockOpt(t *kernel.Task, level int, name int, outPt } // SetSockOpt implements socket.Socket.SetSockOpt. -func (s *socketOperations) SetSockOpt(t *kernel.Task, level int, name int, opt []byte) *syserr.Error { +func (s *socketOpsCommon) SetSockOpt(t *kernel.Task, level int, name int, opt []byte) *syserr.Error { // Whitelist options and constrain option length. optlen := setSockOptLen(t, level, name) switch level { @@ -374,7 +414,7 @@ func (s *socketOperations) SetSockOpt(t *kernel.Task, level int, name int, opt [ } // RecvMsg implements socket.Socket.RecvMsg. -func (s *socketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, haveDeadline bool, deadline ktime.Time, senderRequested bool, controlLen uint64) (int, int, linux.SockAddr, uint32, socket.ControlMessages, *syserr.Error) { +func (s *socketOpsCommon) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, haveDeadline bool, deadline ktime.Time, senderRequested bool, controlLen uint64) (int, int, linux.SockAddr, uint32, socket.ControlMessages, *syserr.Error) { // Whitelist flags. // // FIXME(jamieliu): We can't support MSG_ERRQUEUE because it uses ancillary @@ -496,7 +536,7 @@ func (s *socketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags } // SendMsg implements socket.Socket.SendMsg. -func (s *socketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to []byte, flags int, haveDeadline bool, deadline ktime.Time, controlMessages socket.ControlMessages) (int, *syserr.Error) { +func (s *socketOpsCommon) SendMsg(t *kernel.Task, src usermem.IOSequence, to []byte, flags int, haveDeadline bool, deadline ktime.Time, controlMessages socket.ControlMessages) (int, *syserr.Error) { // Whitelist flags. if flags&^(syscall.MSG_DONTWAIT|syscall.MSG_EOR|syscall.MSG_FASTOPEN|syscall.MSG_MORE|syscall.MSG_NOSIGNAL) != 0 { return 0, syserr.ErrInvalidArgument @@ -585,7 +625,7 @@ func translateIOSyscallError(err error) error { } // State implements socket.Socket.State. -func (s *socketOperations) State() uint32 { +func (s *socketOpsCommon) State() uint32 { info := linux.TCPInfo{} buf, err := getsockopt(s.fd, syscall.SOL_TCP, syscall.TCP_INFO, linux.SizeOfTCPInfo) if err != nil { @@ -607,7 +647,7 @@ func (s *socketOperations) State() uint32 { } // Type implements socket.Socket.Type. -func (s *socketOperations) Type() (family int, skType linux.SockType, protocol int) { +func (s *socketOpsCommon) Type() (family int, skType linux.SockType, protocol int) { return s.family, s.stype, s.protocol } @@ -663,8 +703,11 @@ func (p *socketProvider) Pair(t *kernel.Task, stype linux.SockType, protocol int return nil, nil, nil } +// LINT.ThenChange(./socket_vfs2.go) + func init() { for _, family := range []int{syscall.AF_INET, syscall.AF_INET6} { socket.RegisterProvider(family, &socketProvider{family}) + socket.RegisterProviderVFS2(family, &socketProviderVFS2{}) } } diff --git a/pkg/sentry/socket/hostinet/socket_unsafe.go b/pkg/sentry/socket/hostinet/socket_unsafe.go index cd67234d2..3f420c2ec 100644 --- a/pkg/sentry/socket/hostinet/socket_unsafe.go +++ b/pkg/sentry/socket/hostinet/socket_unsafe.go @@ -21,7 +21,6 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/syserr" @@ -54,12 +53,11 @@ func writev(fd int, srcs []syscall.Iovec) (uint64, error) { return uint64(n), nil } -// Ioctl implements fs.FileOperations.Ioctl. -func (s *socketOperations) Ioctl(ctx context.Context, _ *fs.File, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { +func ioctl(ctx context.Context, fd int, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { switch cmd := uintptr(args[1].Int()); cmd { case syscall.TIOCINQ, syscall.TIOCOUTQ: var val int32 - if _, _, errno := syscall.Syscall(syscall.SYS_IOCTL, uintptr(s.fd), cmd, uintptr(unsafe.Pointer(&val))); errno != 0 { + if _, _, errno := syscall.Syscall(syscall.SYS_IOCTL, uintptr(fd), cmd, uintptr(unsafe.Pointer(&val))); errno != 0 { return 0, translateIOSyscallError(errno) } var buf [4]byte @@ -93,7 +91,7 @@ func getsockopt(fd int, level, name int, optlen int) ([]byte, error) { } // GetSockName implements socket.Socket.GetSockName. -func (s *socketOperations) GetSockName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) { +func (s *socketOpsCommon) GetSockName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) { addr := make([]byte, sizeofSockaddr) addrlen := uint32(len(addr)) _, _, errno := syscall.Syscall(syscall.SYS_GETSOCKNAME, uintptr(s.fd), uintptr(unsafe.Pointer(&addr[0])), uintptr(unsafe.Pointer(&addrlen))) @@ -104,7 +102,7 @@ func (s *socketOperations) GetSockName(t *kernel.Task) (linux.SockAddr, uint32, } // GetPeerName implements socket.Socket.GetPeerName. -func (s *socketOperations) GetPeerName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) { +func (s *socketOpsCommon) GetPeerName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) { addr := make([]byte, sizeofSockaddr) addrlen := uint32(len(addr)) _, _, errno := syscall.Syscall(syscall.SYS_GETPEERNAME, uintptr(s.fd), uintptr(unsafe.Pointer(&addr[0])), uintptr(unsafe.Pointer(&addrlen))) diff --git a/pkg/sentry/socket/hostinet/socket_vfs2.go b/pkg/sentry/socket/hostinet/socket_vfs2.go new file mode 100644 index 000000000..b03ca2f26 --- /dev/null +++ b/pkg/sentry/socket/hostinet/socket_vfs2.go @@ -0,0 +1,183 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package hostinet + +import ( + "syscall" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fdnotifier" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" + "gvisor.dev/gvisor/pkg/sentry/hostfd" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/socket" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserr" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +type socketVFS2 struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl + // TODO(gvisor.dev/issue/1484): VFS1 stores internal metadata for hostinet. + // We should perhaps rely on the host, much like in hostfs. + vfs.DentryMetadataFileDescriptionImpl + + socketOpsCommon +} + +var _ = socket.SocketVFS2(&socketVFS2{}) + +func newVFS2Socket(t *kernel.Task, family int, stype linux.SockType, protocol int, fd int, flags uint32) (*vfs.FileDescription, *syserr.Error) { + mnt := t.Kernel().SocketMount() + fs := mnt.Filesystem().Impl().(*kernfs.Filesystem) + d := sockfs.NewDentry(t.Credentials(), fs.NextIno()) + + s := &socketVFS2{ + socketOpsCommon: socketOpsCommon{ + family: family, + stype: stype, + protocol: protocol, + fd: fd, + }, + } + if err := fdnotifier.AddFD(int32(fd), &s.queue); err != nil { + return nil, syserr.FromError(err) + } + vfsfd := &s.vfsfd + if err := vfsfd.Init(s, linux.O_RDWR|(flags&linux.O_NONBLOCK), mnt, d, &vfs.FileDescriptionOptions{ + DenyPRead: true, + DenyPWrite: true, + UseDentryMetadata: true, + }); err != nil { + return nil, syserr.FromError(err) + } + return vfsfd, nil +} + +// Readiness implements waiter.Waitable.Readiness. +func (s *socketVFS2) Readiness(mask waiter.EventMask) waiter.EventMask { + return s.socketOpsCommon.Readiness(mask) +} + +// EventRegister implements waiter.Waitable.EventRegister. +func (s *socketVFS2) EventRegister(e *waiter.Entry, mask waiter.EventMask) { + s.socketOpsCommon.EventRegister(e, mask) +} + +// EventUnregister implements waiter.Waitable.EventUnregister. +func (s *socketVFS2) EventUnregister(e *waiter.Entry) { + s.socketOpsCommon.EventUnregister(e) +} + +// Ioctl implements vfs.FileDescriptionImpl. +func (s *socketVFS2) Ioctl(ctx context.Context, uio usermem.IO, args arch.SyscallArguments) (uintptr, error) { + return ioctl(ctx, s.fd, uio, args) +} + +// PRead implements vfs.FileDescriptionImpl. +func (s *socketVFS2) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + return 0, syserror.ESPIPE +} + +// Read implements vfs.FileDescriptionImpl. +func (s *socketVFS2) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + // All flags other than RWF_NOWAIT should be ignored. + // TODO(gvisor.dev/issue/2601): Support RWF_NOWAIT. + if opts.Flags != 0 { + return 0, syserror.EOPNOTSUPP + } + + reader := hostfd.GetReadWriterAt(int32(s.fd), -1, opts.Flags) + n, err := dst.CopyOutFrom(ctx, reader) + hostfd.PutReadWriterAt(reader) + return int64(n), err +} + +// PWrite implements vfs.FileDescriptionImpl. +func (s *socketVFS2) PWrite(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + return 0, syserror.ESPIPE +} + +// Write implements vfs.FileDescriptionImpl. +func (s *socketVFS2) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + // All flags other than RWF_NOWAIT should be ignored. + // TODO(gvisor.dev/issue/2601): Support RWF_NOWAIT. + if opts.Flags != 0 { + return 0, syserror.EOPNOTSUPP + } + + writer := hostfd.GetReadWriterAt(int32(s.fd), -1, opts.Flags) + n, err := src.CopyInTo(ctx, writer) + hostfd.PutReadWriterAt(writer) + return int64(n), err +} + +type socketProviderVFS2 struct { + family int +} + +// Socket implements socket.ProviderVFS2.Socket. +func (p *socketProviderVFS2) Socket(t *kernel.Task, stypeflags linux.SockType, protocol int) (*vfs.FileDescription, *syserr.Error) { + // Check that we are using the host network stack. + stack := t.NetworkContext() + if stack == nil { + return nil, nil + } + if _, ok := stack.(*Stack); !ok { + return nil, nil + } + + // Only accept TCP and UDP. + stype := stypeflags & linux.SOCK_TYPE_MASK + switch stype { + case syscall.SOCK_STREAM: + switch protocol { + case 0, syscall.IPPROTO_TCP: + // ok + default: + return nil, nil + } + case syscall.SOCK_DGRAM: + switch protocol { + case 0, syscall.IPPROTO_UDP: + // ok + default: + return nil, nil + } + default: + return nil, nil + } + + // Conservatively ignore all flags specified by the application and add + // SOCK_NONBLOCK since socketOperations requires it. Pass a protocol of 0 + // to simplify the syscall filters, since 0 and IPPROTO_* are equivalent. + fd, err := syscall.Socket(p.family, int(stype)|syscall.SOCK_NONBLOCK|syscall.SOCK_CLOEXEC, 0) + if err != nil { + return nil, syserr.FromError(err) + } + return newVFS2Socket(t, p.family, stype, protocol, fd, uint32(stypeflags&syscall.SOCK_NONBLOCK)) +} + +// Pair implements socket.Provider.Pair. +func (p *socketProviderVFS2) Pair(t *kernel.Task, stype linux.SockType, protocol int) (*vfs.FileDescription, *vfs.FileDescription, *syserr.Error) { + // Not supported by AF_INET/AF_INET6. + return nil, nil, nil +} diff --git a/pkg/sentry/socket/netlink/BUILD b/pkg/sentry/socket/netlink/BUILD index 1911cd9b8..09ca00a4a 100644 --- a/pkg/sentry/socket/netlink/BUILD +++ b/pkg/sentry/socket/netlink/BUILD @@ -7,7 +7,9 @@ go_library( srcs = [ "message.go", "provider.go", + "provider_vfs2.go", "socket.go", + "socket_vfs2.go", ], visibility = ["//pkg/sentry:internal"], deps = [ @@ -18,6 +20,8 @@ go_library( "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", + "//pkg/sentry/fsimpl/kernfs", + "//pkg/sentry/fsimpl/sockfs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/time", @@ -25,6 +29,7 @@ go_library( "//pkg/sentry/socket/netlink/port", "//pkg/sentry/socket/unix", "//pkg/sentry/socket/unix/transport", + "//pkg/sentry/vfs", "//pkg/sync", "//pkg/syserr", "//pkg/syserror", diff --git a/pkg/sentry/socket/netlink/provider.go b/pkg/sentry/socket/netlink/provider.go index b0dc70e5c..0d45e5053 100644 --- a/pkg/sentry/socket/netlink/provider.go +++ b/pkg/sentry/socket/netlink/provider.go @@ -67,6 +67,8 @@ func RegisterProvider(protocol int, provider Provider) { protocols[protocol] = provider } +// LINT.IfChange + // socketProvider implements socket.Provider. type socketProvider struct { } @@ -105,7 +107,10 @@ func (*socketProvider) Pair(*kernel.Task, linux.SockType, int) (*fs.File, *fs.Fi return nil, nil, syserr.ErrNotSupported } +// LINT.ThenChange(./provider_vfs2.go) + // init registers the socket provider. func init() { socket.RegisterProvider(linux.AF_NETLINK, &socketProvider{}) + socket.RegisterProviderVFS2(linux.AF_NETLINK, &socketProviderVFS2{}) } diff --git a/pkg/sentry/socket/netlink/provider_vfs2.go b/pkg/sentry/socket/netlink/provider_vfs2.go new file mode 100644 index 000000000..dcd92b5cd --- /dev/null +++ b/pkg/sentry/socket/netlink/provider_vfs2.go @@ -0,0 +1,71 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package netlink + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserr" +) + +// socketProviderVFS2 implements socket.Provider. +type socketProviderVFS2 struct { +} + +// Socket implements socket.Provider.Socket. +func (*socketProviderVFS2) Socket(t *kernel.Task, stype linux.SockType, protocol int) (*vfs.FileDescription, *syserr.Error) { + // Netlink sockets must be specified as datagram or raw, but they + // behave the same regardless of type. + if stype != linux.SOCK_DGRAM && stype != linux.SOCK_RAW { + return nil, syserr.ErrSocketNotSupported + } + + provider, ok := protocols[protocol] + if !ok { + return nil, syserr.ErrProtocolNotSupported + } + + p, err := provider(t) + if err != nil { + return nil, err + } + + s, err := NewVFS2(t, stype, p) + if err != nil { + return nil, err + } + + vfsfd := &s.vfsfd + mnt := t.Kernel().SocketMount() + fs := mnt.Filesystem().Impl().(*kernfs.Filesystem) + d := sockfs.NewDentry(t.Credentials(), fs.NextIno()) + if err := vfsfd.Init(s, linux.O_RDWR, mnt, d, &vfs.FileDescriptionOptions{ + DenyPRead: true, + DenyPWrite: true, + UseDentryMetadata: true, + }); err != nil { + return nil, syserr.FromError(err) + } + return vfsfd, nil +} + +// Pair implements socket.Provider.Pair by returning an error. +func (*socketProviderVFS2) Pair(*kernel.Task, linux.SockType, int) (*vfs.FileDescription, *vfs.FileDescription, *syserr.Error) { + // Netlink sockets never supports creating socket pairs. + return nil, nil, syserr.ErrNotSupported +} diff --git a/pkg/sentry/socket/netlink/socket.go b/pkg/sentry/socket/netlink/socket.go index 2ca02567d..81f34c5a2 100644 --- a/pkg/sentry/socket/netlink/socket.go +++ b/pkg/sentry/socket/netlink/socket.go @@ -58,6 +58,8 @@ var errNoFilter = syserr.New("no filter attached", linux.ENOENT) // netlinkSocketDevice is the netlink socket virtual device. var netlinkSocketDevice = device.NewAnonDevice() +// LINT.IfChange + // Socket is the base socket type for netlink sockets. // // This implementation only supports userspace sending and receiving messages @@ -74,6 +76,14 @@ type Socket struct { fsutil.FileNoSplice `state:"nosave"` fsutil.FileNoopFlush `state:"nosave"` fsutil.FileUseInodeUnstableAttr `state:"nosave"` + + socketOpsCommon +} + +// socketOpsCommon contains the socket operations common to VFS1 and VFS2. +// +// +stateify savable +type socketOpsCommon struct { socket.SendReceiveTimeout // ports provides netlink port allocation. @@ -140,17 +150,19 @@ func NewSocket(t *kernel.Task, skType linux.SockType, protocol Protocol) (*Socke } return &Socket{ - ports: t.Kernel().NetlinkPorts(), - protocol: protocol, - skType: skType, - ep: ep, - connection: connection, - sendBufferSize: defaultSendBufferSize, + socketOpsCommon: socketOpsCommon{ + ports: t.Kernel().NetlinkPorts(), + protocol: protocol, + skType: skType, + ep: ep, + connection: connection, + sendBufferSize: defaultSendBufferSize, + }, }, nil } // Release implements fs.FileOperations.Release. -func (s *Socket) Release() { +func (s *socketOpsCommon) Release() { s.connection.Release() s.ep.Close() @@ -160,7 +172,7 @@ func (s *Socket) Release() { } // Readiness implements waiter.Waitable.Readiness. -func (s *Socket) Readiness(mask waiter.EventMask) waiter.EventMask { +func (s *socketOpsCommon) Readiness(mask waiter.EventMask) waiter.EventMask { // ep holds messages to be read and thus handles EventIn readiness. ready := s.ep.Readiness(mask) @@ -174,18 +186,18 @@ func (s *Socket) Readiness(mask waiter.EventMask) waiter.EventMask { } // EventRegister implements waiter.Waitable.EventRegister. -func (s *Socket) EventRegister(e *waiter.Entry, mask waiter.EventMask) { +func (s *socketOpsCommon) EventRegister(e *waiter.Entry, mask waiter.EventMask) { s.ep.EventRegister(e, mask) // Writable readiness never changes, so no registration is needed. } // EventUnregister implements waiter.Waitable.EventUnregister. -func (s *Socket) EventUnregister(e *waiter.Entry) { +func (s *socketOpsCommon) EventUnregister(e *waiter.Entry) { s.ep.EventUnregister(e) } // Passcred implements transport.Credentialer.Passcred. -func (s *Socket) Passcred() bool { +func (s *socketOpsCommon) Passcred() bool { s.mu.Lock() passcred := s.passcred s.mu.Unlock() @@ -193,7 +205,7 @@ func (s *Socket) Passcred() bool { } // ConnectedPasscred implements transport.Credentialer.ConnectedPasscred. -func (s *Socket) ConnectedPasscred() bool { +func (s *socketOpsCommon) ConnectedPasscred() bool { // This socket is connected to the kernel, which doesn't need creds. // // This is arbitrary, as ConnectedPasscred on this type has no callers. @@ -227,7 +239,7 @@ func ExtractSockAddr(b []byte) (*linux.SockAddrNetlink, *syserr.Error) { // port of 0 defaults to the ThreadGroup ID. // // Preconditions: mu is held. -func (s *Socket) bindPort(t *kernel.Task, port int32) *syserr.Error { +func (s *socketOpsCommon) bindPort(t *kernel.Task, port int32) *syserr.Error { if s.bound { // Re-binding is only allowed if the port doesn't change. if port != s.portID { @@ -251,7 +263,7 @@ func (s *Socket) bindPort(t *kernel.Task, port int32) *syserr.Error { } // Bind implements socket.Socket.Bind. -func (s *Socket) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { +func (s *socketOpsCommon) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { a, err := ExtractSockAddr(sockaddr) if err != nil { return err @@ -269,7 +281,7 @@ func (s *Socket) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { } // Connect implements socket.Socket.Connect. -func (s *Socket) Connect(t *kernel.Task, sockaddr []byte, blocking bool) *syserr.Error { +func (s *socketOpsCommon) Connect(t *kernel.Task, sockaddr []byte, blocking bool) *syserr.Error { a, err := ExtractSockAddr(sockaddr) if err != nil { return err @@ -300,25 +312,25 @@ func (s *Socket) Connect(t *kernel.Task, sockaddr []byte, blocking bool) *syserr } // Accept implements socket.Socket.Accept. -func (s *Socket) Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (int32, linux.SockAddr, uint32, *syserr.Error) { +func (s *socketOpsCommon) Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (int32, linux.SockAddr, uint32, *syserr.Error) { // Netlink sockets never support accept. return 0, nil, 0, syserr.ErrNotSupported } // Listen implements socket.Socket.Listen. -func (s *Socket) Listen(t *kernel.Task, backlog int) *syserr.Error { +func (s *socketOpsCommon) Listen(t *kernel.Task, backlog int) *syserr.Error { // Netlink sockets never support listen. return syserr.ErrNotSupported } // Shutdown implements socket.Socket.Shutdown. -func (s *Socket) Shutdown(t *kernel.Task, how int) *syserr.Error { +func (s *socketOpsCommon) Shutdown(t *kernel.Task, how int) *syserr.Error { // Netlink sockets never support shutdown. return syserr.ErrNotSupported } // GetSockOpt implements socket.Socket.GetSockOpt. -func (s *Socket) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) { +func (s *socketOpsCommon) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) { switch level { case linux.SOL_SOCKET: switch name { @@ -369,7 +381,7 @@ func (s *Socket) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem. } // SetSockOpt implements socket.Socket.SetSockOpt. -func (s *Socket) SetSockOpt(t *kernel.Task, level int, name int, opt []byte) *syserr.Error { +func (s *socketOpsCommon) SetSockOpt(t *kernel.Task, level int, name int, opt []byte) *syserr.Error { switch level { case linux.SOL_SOCKET: switch name { @@ -466,7 +478,7 @@ func (s *Socket) SetSockOpt(t *kernel.Task, level int, name int, opt []byte) *sy } // GetSockName implements socket.Socket.GetSockName. -func (s *Socket) GetSockName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) { +func (s *socketOpsCommon) GetSockName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) { s.mu.Lock() defer s.mu.Unlock() @@ -478,7 +490,7 @@ func (s *Socket) GetSockName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Er } // GetPeerName implements socket.Socket.GetPeerName. -func (s *Socket) GetPeerName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) { +func (s *socketOpsCommon) GetPeerName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) { sa := &linux.SockAddrNetlink{ Family: linux.AF_NETLINK, // TODO(b/68878065): Support non-kernel peers. For now the peer @@ -489,7 +501,7 @@ func (s *Socket) GetPeerName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Er } // RecvMsg implements socket.Socket.RecvMsg. -func (s *Socket) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, haveDeadline bool, deadline ktime.Time, senderRequested bool, controlDataLen uint64) (int, int, linux.SockAddr, uint32, socket.ControlMessages, *syserr.Error) { +func (s *socketOpsCommon) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, haveDeadline bool, deadline ktime.Time, senderRequested bool, controlDataLen uint64) (int, int, linux.SockAddr, uint32, socket.ControlMessages, *syserr.Error) { from := &linux.SockAddrNetlink{ Family: linux.AF_NETLINK, PortID: 0, @@ -590,7 +602,7 @@ func (kernelSCM) Credentials(*kernel.Task) (kernel.ThreadID, auth.UID, auth.GID) var kernelCreds = &kernelSCM{} // sendResponse sends the response messages in ms back to userspace. -func (s *Socket) sendResponse(ctx context.Context, ms *MessageSet) *syserr.Error { +func (s *socketOpsCommon) sendResponse(ctx context.Context, ms *MessageSet) *syserr.Error { // Linux combines multiple netlink messages into a single datagram. bufs := make([][]byte, 0, len(ms.Messages)) for _, m := range ms.Messages { @@ -666,7 +678,7 @@ func dumpAckMesage(hdr linux.NetlinkMessageHeader, ms *MessageSet) { // processMessages handles each message in buf, passing it to the protocol // handler for final handling. -func (s *Socket) processMessages(ctx context.Context, buf []byte) *syserr.Error { +func (s *socketOpsCommon) processMessages(ctx context.Context, buf []byte) *syserr.Error { for len(buf) > 0 { msg, rest, ok := ParseMessage(buf) if !ok { @@ -698,7 +710,7 @@ func (s *Socket) processMessages(ctx context.Context, buf []byte) *syserr.Error } // sendMsg is the core of message send, used for SendMsg and Write. -func (s *Socket) sendMsg(ctx context.Context, src usermem.IOSequence, to []byte, flags int, controlMessages socket.ControlMessages) (int, *syserr.Error) { +func (s *socketOpsCommon) sendMsg(ctx context.Context, src usermem.IOSequence, to []byte, flags int, controlMessages socket.ControlMessages) (int, *syserr.Error) { dstPort := int32(0) if len(to) != 0 { @@ -745,7 +757,7 @@ func (s *Socket) sendMsg(ctx context.Context, src usermem.IOSequence, to []byte, } // SendMsg implements socket.Socket.SendMsg. -func (s *Socket) SendMsg(t *kernel.Task, src usermem.IOSequence, to []byte, flags int, haveDeadline bool, deadline ktime.Time, controlMessages socket.ControlMessages) (int, *syserr.Error) { +func (s *socketOpsCommon) SendMsg(t *kernel.Task, src usermem.IOSequence, to []byte, flags int, haveDeadline bool, deadline ktime.Time, controlMessages socket.ControlMessages) (int, *syserr.Error) { return s.sendMsg(t, src, to, flags, controlMessages) } @@ -756,11 +768,13 @@ func (s *Socket) Write(ctx context.Context, _ *fs.File, src usermem.IOSequence, } // State implements socket.Socket.State. -func (s *Socket) State() uint32 { +func (s *socketOpsCommon) State() uint32 { return s.ep.State() } // Type implements socket.Socket.Type. -func (s *Socket) Type() (family int, skType linux.SockType, protocol int) { +func (s *socketOpsCommon) Type() (family int, skType linux.SockType, protocol int) { return linux.AF_NETLINK, s.skType, s.protocol.Protocol() } + +// LINT.ThenChange(./socket_vfs2.go) diff --git a/pkg/sentry/socket/netlink/socket_vfs2.go b/pkg/sentry/socket/netlink/socket_vfs2.go new file mode 100644 index 000000000..b854bf990 --- /dev/null +++ b/pkg/sentry/socket/netlink/socket_vfs2.go @@ -0,0 +1,138 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package netlink + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/socket" + "gvisor.dev/gvisor/pkg/sentry/socket/unix" + "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserr" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +// SocketVFS2 is the base VFS2 socket type for netlink sockets. +// +// This implementation only supports userspace sending and receiving messages +// to/from the kernel. +// +// SocketVFS2 implements socket.SocketVFS2 and transport.Credentialer. +type SocketVFS2 struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl + vfs.DentryMetadataFileDescriptionImpl + + socketOpsCommon +} + +var _ socket.SocketVFS2 = (*SocketVFS2)(nil) +var _ transport.Credentialer = (*SocketVFS2)(nil) + +// NewVFS2 creates a new SocketVFS2. +func NewVFS2(t *kernel.Task, skType linux.SockType, protocol Protocol) (*SocketVFS2, *syserr.Error) { + // Datagram endpoint used to buffer kernel -> user messages. + ep := transport.NewConnectionless(t) + + // Bind the endpoint for good measure so we can connect to it. The + // bound address will never be exposed. + if err := ep.Bind(tcpip.FullAddress{Addr: "dummy"}, nil); err != nil { + ep.Close() + return nil, err + } + + // Create a connection from which the kernel can write messages. + connection, err := ep.(transport.BoundEndpoint).UnidirectionalConnect(t) + if err != nil { + ep.Close() + return nil, err + } + + return &SocketVFS2{ + socketOpsCommon: socketOpsCommon{ + ports: t.Kernel().NetlinkPorts(), + protocol: protocol, + skType: skType, + ep: ep, + connection: connection, + sendBufferSize: defaultSendBufferSize, + }, + }, nil +} + +// Readiness implements waiter.Waitable.Readiness. +func (s *SocketVFS2) Readiness(mask waiter.EventMask) waiter.EventMask { + return s.socketOpsCommon.Readiness(mask) +} + +// EventRegister implements waiter.Waitable.EventRegister. +func (s *SocketVFS2) EventRegister(e *waiter.Entry, mask waiter.EventMask) { + s.socketOpsCommon.EventRegister(e, mask) +} + +// EventUnregister implements waiter.Waitable.EventUnregister. +func (s *SocketVFS2) EventUnregister(e *waiter.Entry) { + s.socketOpsCommon.EventUnregister(e) +} + +// Ioctl implements vfs.FileDescriptionImpl. +func (*SocketVFS2) Ioctl(context.Context, usermem.IO, arch.SyscallArguments) (uintptr, error) { + // TODO(b/68878065): no ioctls supported. + return 0, syserror.ENOTTY +} + +// PRead implements vfs.FileDescriptionImpl. +func (s *SocketVFS2) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + return 0, syserror.ESPIPE +} + +// Read implements vfs.FileDescriptionImpl. +func (s *SocketVFS2) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + // All flags other than RWF_NOWAIT should be ignored. + // TODO(gvisor.dev/issue/2601): Support RWF_NOWAIT. + if opts.Flags != 0 { + return 0, syserror.EOPNOTSUPP + } + + if dst.NumBytes() == 0 { + return 0, nil + } + return dst.CopyOutFrom(ctx, &unix.EndpointReader{ + Endpoint: s.ep, + }) +} + +// PWrite implements vfs.FileDescriptionImpl. +func (s *SocketVFS2) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + return 0, syserror.ESPIPE +} + +// Write implements vfs.FileDescriptionImpl. +func (s *SocketVFS2) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + // All flags other than RWF_NOWAIT should be ignored. + // TODO(gvisor.dev/issue/2601): Support RWF_NOWAIT. + if opts.Flags != 0 { + return 0, syserror.EOPNOTSUPP + } + + n, err := s.sendMsg(ctx, src, nil, 0, socket.ControlMessages{}) + return int64(n), err.ToError() +} diff --git a/pkg/sentry/socket/netstack/BUILD b/pkg/sentry/socket/netstack/BUILD index cbf46b1e9..ccf9fcf5c 100644 --- a/pkg/sentry/socket/netstack/BUILD +++ b/pkg/sentry/socket/netstack/BUILD @@ -7,7 +7,9 @@ go_library( srcs = [ "device.go", "netstack.go", + "netstack_vfs2.go", "provider.go", + "provider_vfs2.go", "save_restore.go", "stack.go", ], @@ -25,6 +27,8 @@ go_library( "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", + "//pkg/sentry/fsimpl/kernfs", + "//pkg/sentry/fsimpl/sockfs", "//pkg/sentry/inet", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", @@ -32,6 +36,7 @@ go_library( "//pkg/sentry/socket", "//pkg/sentry/socket/netfilter", "//pkg/sentry/unimpl", + "//pkg/sentry/vfs", "//pkg/sync", "//pkg/syserr", "//pkg/syserror", diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index d5879c10f..81053d8ef 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -252,6 +252,8 @@ type commonEndpoint interface { GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) } +// LINT.IfChange + // SocketOperations encapsulates all the state needed to represent a network stack // endpoint in the kernel context. // @@ -263,6 +265,14 @@ type SocketOperations struct { fsutil.FileNoFsync `state:"nosave"` fsutil.FileNoMMap `state:"nosave"` fsutil.FileUseInodeUnstableAttr `state:"nosave"` + + socketOpsCommon +} + +// socketOpsCommon contains the socket operations common to VFS1 and VFS2. +// +// +stateify savable +type socketOpsCommon struct { socket.SendReceiveTimeout *waiter.Queue @@ -314,11 +324,13 @@ func New(t *kernel.Task, family int, skType linux.SockType, protocol int, queue dirent := socket.NewDirent(t, netstackDevice) defer dirent.DecRef() return fs.NewFile(t, dirent, fs.FileFlags{Read: true, Write: true, NonSeekable: true}, &SocketOperations{ - Queue: queue, - family: family, - Endpoint: endpoint, - skType: skType, - protocol: protocol, + socketOpsCommon: socketOpsCommon{ + Queue: queue, + family: family, + Endpoint: endpoint, + skType: skType, + protocol: protocol, + }, }), nil } @@ -417,7 +429,7 @@ func AddressAndFamily(addr []byte) (tcpip.FullAddress, uint16, *syserr.Error) { } } -func (s *SocketOperations) isPacketBased() bool { +func (s *socketOpsCommon) isPacketBased() bool { return s.skType == linux.SOCK_DGRAM || s.skType == linux.SOCK_SEQPACKET || s.skType == linux.SOCK_RDM || s.skType == linux.SOCK_RAW } @@ -425,7 +437,7 @@ func (s *SocketOperations) isPacketBased() bool { // empty. It assumes that the socket is locked. // // Precondition: s.readMu must be held. -func (s *SocketOperations) fetchReadView() *syserr.Error { +func (s *socketOpsCommon) fetchReadView() *syserr.Error { if len(s.readView) > 0 { return nil } @@ -446,7 +458,7 @@ func (s *SocketOperations) fetchReadView() *syserr.Error { } // Release implements fs.FileOperations.Release. -func (s *SocketOperations) Release() { +func (s *socketOpsCommon) Release() { s.Endpoint.Close() } @@ -633,7 +645,7 @@ func (s *SocketOperations) ReadFrom(ctx context.Context, _ *fs.File, r io.Reader } // Readiness returns a mask of ready events for socket s. -func (s *SocketOperations) Readiness(mask waiter.EventMask) waiter.EventMask { +func (s *socketOpsCommon) Readiness(mask waiter.EventMask) waiter.EventMask { r := s.Endpoint.Readiness(mask) // Check our cached value iff the caller asked for readability and the @@ -647,7 +659,7 @@ func (s *SocketOperations) Readiness(mask waiter.EventMask) waiter.EventMask { return r } -func (s *SocketOperations) checkFamily(family uint16, exact bool) *syserr.Error { +func (s *socketOpsCommon) checkFamily(family uint16, exact bool) *syserr.Error { if family == uint16(s.family) { return nil } @@ -670,7 +682,7 @@ func (s *SocketOperations) checkFamily(family uint16, exact bool) *syserr.Error // represented by the empty string. // // TODO(gvisor.dev/issue/1556): remove this function. -func (s *SocketOperations) mapFamily(addr tcpip.FullAddress, family uint16) tcpip.FullAddress { +func (s *socketOpsCommon) mapFamily(addr tcpip.FullAddress, family uint16) tcpip.FullAddress { if len(addr.Addr) == 0 && s.family == linux.AF_INET6 && family == linux.AF_INET { addr.Addr = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\xff\xff\x00\x00\x00\x00" } @@ -679,7 +691,7 @@ func (s *SocketOperations) mapFamily(addr tcpip.FullAddress, family uint16) tcpi // Connect implements the linux syscall connect(2) for sockets backed by // tpcip.Endpoint. -func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking bool) *syserr.Error { +func (s *socketOpsCommon) Connect(t *kernel.Task, sockaddr []byte, blocking bool) *syserr.Error { addr, family, err := AddressAndFamily(sockaddr) if err != nil { return err @@ -725,7 +737,7 @@ func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking boo // Bind implements the linux syscall bind(2) for sockets backed by // tcpip.Endpoint. -func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { +func (s *socketOpsCommon) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { if len(sockaddr) < 2 { return syserr.ErrInvalidArgument } @@ -771,13 +783,13 @@ func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { // Listen implements the linux syscall listen(2) for sockets backed by // tcpip.Endpoint. -func (s *SocketOperations) Listen(t *kernel.Task, backlog int) *syserr.Error { +func (s *socketOpsCommon) Listen(t *kernel.Task, backlog int) *syserr.Error { return syserr.TranslateNetstackError(s.Endpoint.Listen(backlog)) } // blockingAccept implements a blocking version of accept(2), that is, if no // connections are ready to be accept, it will block until one becomes ready. -func (s *SocketOperations) blockingAccept(t *kernel.Task) (tcpip.Endpoint, *waiter.Queue, *syserr.Error) { +func (s *socketOpsCommon) blockingAccept(t *kernel.Task) (tcpip.Endpoint, *waiter.Queue, *syserr.Error) { // Register for notifications. e, ch := waiter.NewChannelEntry(nil) s.EventRegister(&e, waiter.EventIn) @@ -863,7 +875,7 @@ func ConvertShutdown(how int) (tcpip.ShutdownFlags, *syserr.Error) { // Shutdown implements the linux syscall shutdown(2) for sockets backed by // tcpip.Endpoint. -func (s *SocketOperations) Shutdown(t *kernel.Task, how int) *syserr.Error { +func (s *socketOpsCommon) Shutdown(t *kernel.Task, how int) *syserr.Error { f, err := ConvertShutdown(how) if err != nil { return err @@ -2258,7 +2270,7 @@ func ConvertAddress(family int, addr tcpip.FullAddress) (linux.SockAddr, uint32) // GetSockName implements the linux syscall getsockname(2) for sockets backed by // tcpip.Endpoint. -func (s *SocketOperations) GetSockName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) { +func (s *socketOpsCommon) GetSockName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) { addr, err := s.Endpoint.GetLocalAddress() if err != nil { return nil, 0, syserr.TranslateNetstackError(err) @@ -2270,7 +2282,7 @@ func (s *SocketOperations) GetSockName(t *kernel.Task) (linux.SockAddr, uint32, // GetPeerName implements the linux syscall getpeername(2) for sockets backed by // tcpip.Endpoint. -func (s *SocketOperations) GetPeerName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) { +func (s *socketOpsCommon) GetPeerName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) { addr, err := s.Endpoint.GetRemoteAddress() if err != nil { return nil, 0, syserr.TranslateNetstackError(err) @@ -2285,7 +2297,7 @@ func (s *SocketOperations) GetPeerName(t *kernel.Task) (linux.SockAddr, uint32, // caller. // // Precondition: s.readMu must be locked. -func (s *SocketOperations) coalescingRead(ctx context.Context, dst usermem.IOSequence, discard bool) (int, *syserr.Error) { +func (s *socketOpsCommon) coalescingRead(ctx context.Context, dst usermem.IOSequence, discard bool) (int, *syserr.Error) { var err *syserr.Error var copied int @@ -2337,7 +2349,7 @@ func (s *SocketOperations) coalescingRead(ctx context.Context, dst usermem.IOSeq return 0, err } -func (s *SocketOperations) fillCmsgInq(cmsg *socket.ControlMessages) { +func (s *socketOpsCommon) fillCmsgInq(cmsg *socket.ControlMessages) { if !s.sockOptInq { return } @@ -2352,7 +2364,7 @@ func (s *SocketOperations) fillCmsgInq(cmsg *socket.ControlMessages) { // nonBlockingRead issues a non-blocking read. // // TODO(b/78348848): Support timestamps for stream sockets. -func (s *SocketOperations) nonBlockingRead(ctx context.Context, dst usermem.IOSequence, peek, trunc, senderRequested bool) (int, int, linux.SockAddr, uint32, socket.ControlMessages, *syserr.Error) { +func (s *socketOpsCommon) nonBlockingRead(ctx context.Context, dst usermem.IOSequence, peek, trunc, senderRequested bool) (int, int, linux.SockAddr, uint32, socket.ControlMessages, *syserr.Error) { isPacket := s.isPacketBased() // Fast path for regular reads from stream (e.g., TCP) endpoints. Note @@ -2461,7 +2473,7 @@ func (s *SocketOperations) nonBlockingRead(ctx context.Context, dst usermem.IOSe return n, flags, addr, addrLen, cmsg, syserr.FromError(err) } -func (s *SocketOperations) controlMessages() socket.ControlMessages { +func (s *socketOpsCommon) controlMessages() socket.ControlMessages { return socket.ControlMessages{ IP: tcpip.ControlMessages{ HasTimestamp: s.readCM.HasTimestamp && s.sockOptTimestamp, @@ -2480,7 +2492,7 @@ func (s *SocketOperations) controlMessages() socket.ControlMessages { // successfully writing packet data out to userspace. // // Precondition: s.readMu must be locked. -func (s *SocketOperations) updateTimestamp() { +func (s *socketOpsCommon) updateTimestamp() { // Save the SIOCGSTAMP timestamp only if SO_TIMESTAMP is disabled. if !s.sockOptTimestamp { s.timestampValid = true @@ -2490,7 +2502,7 @@ func (s *SocketOperations) updateTimestamp() { // RecvMsg implements the linux syscall recvmsg(2) for sockets backed by // tcpip.Endpoint. -func (s *SocketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, haveDeadline bool, deadline ktime.Time, senderRequested bool, controlDataLen uint64) (n int, msgFlags int, senderAddr linux.SockAddr, senderAddrLen uint32, controlMessages socket.ControlMessages, err *syserr.Error) { +func (s *socketOpsCommon) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, haveDeadline bool, deadline ktime.Time, senderRequested bool, controlDataLen uint64) (n int, msgFlags int, senderAddr linux.SockAddr, senderAddrLen uint32, controlMessages socket.ControlMessages, err *syserr.Error) { trunc := flags&linux.MSG_TRUNC != 0 peek := flags&linux.MSG_PEEK != 0 dontWait := flags&linux.MSG_DONTWAIT != 0 @@ -2558,7 +2570,7 @@ func (s *SocketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags // SendMsg implements the linux syscall sendmsg(2) for sockets backed by // tcpip.Endpoint. -func (s *SocketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to []byte, flags int, haveDeadline bool, deadline ktime.Time, controlMessages socket.ControlMessages) (int, *syserr.Error) { +func (s *socketOpsCommon) SendMsg(t *kernel.Task, src usermem.IOSequence, to []byte, flags int, haveDeadline bool, deadline ktime.Time, controlMessages socket.ControlMessages) (int, *syserr.Error) { // Reject Unix control messages. if !controlMessages.Unix.Empty() { return 0, syserr.ErrInvalidArgument @@ -2634,6 +2646,10 @@ func (s *SocketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to [] // Ioctl implements fs.FileOperations.Ioctl. func (s *SocketOperations) Ioctl(ctx context.Context, _ *fs.File, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { + return s.socketOpsCommon.ioctl(ctx, io, args) +} + +func (s *socketOpsCommon) ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { // SIOCGSTAMP is implemented by netstack rather than all commonEndpoint // sockets. // TODO(b/78348848): Add a commonEndpoint method to support SIOCGSTAMP. @@ -2973,7 +2989,7 @@ func nicStateFlagsToLinux(f stack.NICStateFlags) uint32 { // State implements socket.Socket.State. State translates the internal state // returned by netstack to values defined by Linux. -func (s *SocketOperations) State() uint32 { +func (s *socketOpsCommon) State() uint32 { if s.family != linux.AF_INET && s.family != linux.AF_INET6 { // States not implemented for this socket's family. return 0 @@ -3033,6 +3049,8 @@ func (s *SocketOperations) State() uint32 { } // Type implements socket.Socket.Type. -func (s *SocketOperations) Type() (family int, skType linux.SockType, protocol int) { +func (s *socketOpsCommon) Type() (family int, skType linux.SockType, protocol int) { return s.family, s.skType, s.protocol } + +// LINT.ThenChange(./netstack_vfs2.go) diff --git a/pkg/sentry/socket/netstack/netstack_vfs2.go b/pkg/sentry/socket/netstack/netstack_vfs2.go new file mode 100644 index 000000000..eec71035d --- /dev/null +++ b/pkg/sentry/socket/netstack/netstack_vfs2.go @@ -0,0 +1,327 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package netstack + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" + "gvisor.dev/gvisor/pkg/sentry/inet" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/socket/netfilter" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserr" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +// SocketVFS2 encapsulates all the state needed to represent a network stack +// endpoint in the kernel context. +type SocketVFS2 struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl + vfs.DentryMetadataFileDescriptionImpl + + socketOpsCommon +} + +// NewVFS2 creates a new endpoint socket. +func NewVFS2(t *kernel.Task, family int, skType linux.SockType, protocol int, queue *waiter.Queue, endpoint tcpip.Endpoint) (*vfs.FileDescription, *syserr.Error) { + if skType == linux.SOCK_STREAM { + if err := endpoint.SetSockOptBool(tcpip.DelayOption, true); err != nil { + return nil, syserr.TranslateNetstackError(err) + } + } + + mnt := t.Kernel().SocketMount() + fs := mnt.Filesystem().Impl().(*kernfs.Filesystem) + d := sockfs.NewDentry(t.Credentials(), fs.NextIno()) + + s := &SocketVFS2{ + socketOpsCommon: socketOpsCommon{ + Queue: queue, + family: family, + Endpoint: endpoint, + skType: skType, + protocol: protocol, + }, + } + vfsfd := &s.vfsfd + if err := vfsfd.Init(s, linux.O_RDWR, mnt, d, &vfs.FileDescriptionOptions{ + DenyPRead: true, + DenyPWrite: true, + UseDentryMetadata: true, + }); err != nil { + return nil, syserr.FromError(err) + } + return vfsfd, nil +} + +// Readiness implements waiter.Waitable.Readiness. +func (s *SocketVFS2) Readiness(mask waiter.EventMask) waiter.EventMask { + return s.socketOpsCommon.Readiness(mask) +} + +// EventRegister implements waiter.Waitable.EventRegister. +func (s *SocketVFS2) EventRegister(e *waiter.Entry, mask waiter.EventMask) { + s.socketOpsCommon.EventRegister(e, mask) +} + +// EventUnregister implements waiter.Waitable.EventUnregister. +func (s *SocketVFS2) EventUnregister(e *waiter.Entry) { + s.socketOpsCommon.EventUnregister(e) +} + +// PRead implements vfs.FileDescriptionImpl. +func (s *SocketVFS2) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + return 0, syserror.ESPIPE +} + +// Read implements vfs.FileDescriptionImpl. +func (s *SocketVFS2) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + // All flags other than RWF_NOWAIT should be ignored. + // TODO(gvisor.dev/issue/2601): Support RWF_NOWAIT. + if opts.Flags != 0 { + return 0, syserror.EOPNOTSUPP + } + + if dst.NumBytes() == 0 { + return 0, nil + } + n, _, _, _, _, err := s.nonBlockingRead(ctx, dst, false, false, false) + if err == syserr.ErrWouldBlock { + return int64(n), syserror.ErrWouldBlock + } + if err != nil { + return 0, err.ToError() + } + return int64(n), nil +} + +// PWrite implements vfs.FileDescriptionImpl. +func (s *SocketVFS2) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + return 0, syserror.ESPIPE +} + +// Write implements vfs.FileDescriptionImpl. +func (s *SocketVFS2) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + // All flags other than RWF_NOWAIT should be ignored. + // TODO(gvisor.dev/issue/2601): Support RWF_NOWAIT. + if opts.Flags != 0 { + return 0, syserror.EOPNOTSUPP + } + + f := &ioSequencePayload{ctx: ctx, src: src} + n, resCh, err := s.Endpoint.Write(f, tcpip.WriteOptions{}) + if err == tcpip.ErrWouldBlock { + return 0, syserror.ErrWouldBlock + } + + if resCh != nil { + t := kernel.TaskFromContext(ctx) + if err := t.Block(resCh); err != nil { + return 0, syserr.FromError(err).ToError() + } + + n, _, err = s.Endpoint.Write(f, tcpip.WriteOptions{}) + } + + if err != nil { + return 0, syserr.TranslateNetstackError(err).ToError() + } + + if int64(n) < src.NumBytes() { + return int64(n), syserror.ErrWouldBlock + } + + return int64(n), nil +} + +// Accept implements the linux syscall accept(2) for sockets backed by +// tcpip.Endpoint. +func (s *SocketVFS2) Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (int32, linux.SockAddr, uint32, *syserr.Error) { + // Issue the accept request to get the new endpoint. + ep, wq, terr := s.Endpoint.Accept() + if terr != nil { + if terr != tcpip.ErrWouldBlock || !blocking { + return 0, nil, 0, syserr.TranslateNetstackError(terr) + } + + var err *syserr.Error + ep, wq, err = s.blockingAccept(t) + if err != nil { + return 0, nil, 0, err + } + } + + ns, err := NewVFS2(t, s.family, s.skType, s.protocol, wq, ep) + if err != nil { + return 0, nil, 0, err + } + defer ns.DecRef() + + if err := ns.SetStatusFlags(t, t.Credentials(), uint32(flags&linux.SOCK_NONBLOCK)); err != nil { + return 0, nil, 0, syserr.FromError(err) + } + + var addr linux.SockAddr + var addrLen uint32 + if peerRequested { + // Get address of the peer and write it to peer slice. + var err *syserr.Error + addr, addrLen, err = ns.Impl().(*SocketVFS2).GetPeerName(t) + if err != nil { + return 0, nil, 0, err + } + } + + fd, e := t.NewFDFromVFS2(0, ns, kernel.FDFlags{ + CloseOnExec: flags&linux.SOCK_CLOEXEC != 0, + }) + + t.Kernel().RecordSocketVFS2(ns) + + return fd, addr, addrLen, syserr.FromError(e) +} + +// Ioctl implements vfs.FileDescriptionImpl. +func (s *SocketVFS2) Ioctl(ctx context.Context, uio usermem.IO, args arch.SyscallArguments) (uintptr, error) { + return s.socketOpsCommon.ioctl(ctx, uio, args) +} + +// GetSockOpt implements the linux syscall getsockopt(2) for sockets backed by +// tcpip.Endpoint. +func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) { + // TODO(b/78348848): Unlike other socket options, SO_TIMESTAMP is + // implemented specifically for netstack.SocketVFS2 rather than + // commonEndpoint. commonEndpoint should be extended to support socket + // options where the implementation is not shared, as unix sockets need + // their own support for SO_TIMESTAMP. + if level == linux.SOL_SOCKET && name == linux.SO_TIMESTAMP { + if outLen < sizeOfInt32 { + return nil, syserr.ErrInvalidArgument + } + val := int32(0) + s.readMu.Lock() + defer s.readMu.Unlock() + if s.sockOptTimestamp { + val = 1 + } + return val, nil + } + if level == linux.SOL_TCP && name == linux.TCP_INQ { + if outLen < sizeOfInt32 { + return nil, syserr.ErrInvalidArgument + } + val := int32(0) + s.readMu.Lock() + defer s.readMu.Unlock() + if s.sockOptInq { + val = 1 + } + return val, nil + } + + if s.skType == linux.SOCK_RAW && level == linux.IPPROTO_IP { + switch name { + case linux.IPT_SO_GET_INFO: + if outLen < linux.SizeOfIPTGetinfo { + return nil, syserr.ErrInvalidArgument + } + + stack := inet.StackFromContext(t) + if stack == nil { + return nil, syserr.ErrNoDevice + } + info, err := netfilter.GetInfo(t, stack.(*Stack).Stack, outPtr) + if err != nil { + return nil, err + } + return info, nil + + case linux.IPT_SO_GET_ENTRIES: + if outLen < linux.SizeOfIPTGetEntries { + return nil, syserr.ErrInvalidArgument + } + + stack := inet.StackFromContext(t) + if stack == nil { + return nil, syserr.ErrNoDevice + } + entries, err := netfilter.GetEntries(t, stack.(*Stack).Stack, outPtr, outLen) + if err != nil { + return nil, err + } + return entries, nil + + } + } + + return GetSockOpt(t, s, s.Endpoint, s.family, s.skType, level, name, outLen) +} + +// SetSockOpt implements the linux syscall setsockopt(2) for sockets backed by +// tcpip.Endpoint. +func (s *SocketVFS2) SetSockOpt(t *kernel.Task, level int, name int, optVal []byte) *syserr.Error { + // TODO(b/78348848): Unlike other socket options, SO_TIMESTAMP is + // implemented specifically for netstack.SocketVFS2 rather than + // commonEndpoint. commonEndpoint should be extended to support socket + // options where the implementation is not shared, as unix sockets need + // their own support for SO_TIMESTAMP. + if level == linux.SOL_SOCKET && name == linux.SO_TIMESTAMP { + if len(optVal) < sizeOfInt32 { + return syserr.ErrInvalidArgument + } + s.readMu.Lock() + defer s.readMu.Unlock() + s.sockOptTimestamp = usermem.ByteOrder.Uint32(optVal) != 0 + return nil + } + if level == linux.SOL_TCP && name == linux.TCP_INQ { + if len(optVal) < sizeOfInt32 { + return syserr.ErrInvalidArgument + } + s.readMu.Lock() + defer s.readMu.Unlock() + s.sockOptInq = usermem.ByteOrder.Uint32(optVal) != 0 + return nil + } + + if s.skType == linux.SOCK_RAW && level == linux.IPPROTO_IP { + switch name { + case linux.IPT_SO_SET_REPLACE: + if len(optVal) < linux.SizeOfIPTReplace { + return syserr.ErrInvalidArgument + } + + stack := inet.StackFromContext(t) + if stack == nil { + return syserr.ErrNoDevice + } + // Stack must be a netstack stack. + return netfilter.SetEntries(stack.(*Stack).Stack, optVal) + + case linux.IPT_SO_SET_ADD_COUNTERS: + // TODO(gvisor.dev/issue/170): Counter support. + return nil + } + } + + return SetSockOpt(t, s, s.Endpoint, level, name, optVal) +} diff --git a/pkg/sentry/socket/netstack/provider.go b/pkg/sentry/socket/netstack/provider.go index c3f04b613..ead3b2b79 100644 --- a/pkg/sentry/socket/netstack/provider.go +++ b/pkg/sentry/socket/netstack/provider.go @@ -33,6 +33,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +// LINT.IfChange + // provider is an inet socket provider. type provider struct { family int @@ -167,6 +169,8 @@ func packetSocket(t *kernel.Task, epStack *Stack, stype linux.SockType, protocol return New(t, linux.AF_PACKET, stype, protocol, wq, ep) } +// LINT.ThenChange(./provider_vfs2.go) + // Pair just returns nil sockets (not supported). func (*provider) Pair(*kernel.Task, linux.SockType, int) (*fs.File, *fs.File, *syserr.Error) { return nil, nil, nil diff --git a/pkg/sentry/socket/netstack/provider_vfs2.go b/pkg/sentry/socket/netstack/provider_vfs2.go new file mode 100644 index 000000000..2a01143f6 --- /dev/null +++ b/pkg/sentry/socket/netstack/provider_vfs2.go @@ -0,0 +1,141 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package netstack + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/socket" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserr" + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" + "gvisor.dev/gvisor/pkg/tcpip/network/ipv6" + "gvisor.dev/gvisor/pkg/waiter" +) + +// providerVFS2 is an inet socket provider. +type providerVFS2 struct { + family int + netProto tcpip.NetworkProtocolNumber +} + +// Socket creates a new socket object for the AF_INET, AF_INET6, or AF_PACKET +// family. +func (p *providerVFS2) Socket(t *kernel.Task, stype linux.SockType, protocol int) (*vfs.FileDescription, *syserr.Error) { + // Fail right away if we don't have a stack. + stack := t.NetworkContext() + if stack == nil { + // Don't propagate an error here. Instead, allow the socket + // code to continue searching for another provider. + return nil, nil + } + eps, ok := stack.(*Stack) + if !ok { + return nil, nil + } + + // Packet sockets are handled separately, since they are neither INET + // nor INET6 specific. + if p.family == linux.AF_PACKET { + return packetSocketVFS2(t, eps, stype, protocol) + } + + // Figure out the transport protocol. + transProto, associated, err := getTransportProtocol(t, stype, protocol) + if err != nil { + return nil, err + } + + // Create the endpoint. + var ep tcpip.Endpoint + var e *tcpip.Error + wq := &waiter.Queue{} + if stype == linux.SOCK_RAW { + ep, e = eps.Stack.NewRawEndpoint(transProto, p.netProto, wq, associated) + } else { + ep, e = eps.Stack.NewEndpoint(transProto, p.netProto, wq) + + // Assign task to PacketOwner interface to get the UID and GID for + // iptables owner matching. + if e == nil { + ep.SetOwner(t) + } + } + if e != nil { + return nil, syserr.TranslateNetstackError(e) + } + + return NewVFS2(t, p.family, stype, int(transProto), wq, ep) +} + +func packetSocketVFS2(t *kernel.Task, epStack *Stack, stype linux.SockType, protocol int) (*vfs.FileDescription, *syserr.Error) { + // Packet sockets require CAP_NET_RAW. + creds := auth.CredentialsFromContext(t) + if !creds.HasCapability(linux.CAP_NET_RAW) { + return nil, syserr.ErrNotPermitted + } + + // "cooked" packets don't contain link layer information. + var cooked bool + switch stype { + case linux.SOCK_DGRAM: + cooked = true + case linux.SOCK_RAW: + cooked = false + default: + return nil, syserr.ErrProtocolNotSupported + } + + // protocol is passed in network byte order, but netstack wants it in + // host order. + netProto := tcpip.NetworkProtocolNumber(ntohs(uint16(protocol))) + + wq := &waiter.Queue{} + ep, err := epStack.Stack.NewPacketEndpoint(cooked, netProto, wq) + if err != nil { + return nil, syserr.TranslateNetstackError(err) + } + + return NewVFS2(t, linux.AF_PACKET, stype, protocol, wq, ep) +} + +// Pair just returns nil sockets (not supported). +func (*providerVFS2) Pair(*kernel.Task, linux.SockType, int) (*vfs.FileDescription, *vfs.FileDescription, *syserr.Error) { + return nil, nil, nil +} + +// init registers socket providers for AF_INET, AF_INET6, and AF_PACKET. +func init() { + // Providers backed by netstack. + p := []providerVFS2{ + { + family: linux.AF_INET, + netProto: ipv4.ProtocolNumber, + }, + { + family: linux.AF_INET6, + netProto: ipv6.ProtocolNumber, + }, + { + family: linux.AF_PACKET, + }, + } + + for i := range p { + socket.RegisterProviderVFS2(p[i].family, &p[i]) + } +} diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go index 23db93f33..5edc3cdf4 100644 --- a/pkg/sentry/socket/unix/unix_vfs2.go +++ b/pkg/sentry/socket/unix/unix_vfs2.go @@ -229,7 +229,7 @@ func (s *SocketVFS2) PRead(ctx context.Context, dst usermem.IOSequence, offset i // Read implements vfs.FileDescriptionImpl. func (s *SocketVFS2) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { // All flags other than RWF_NOWAIT should be ignored. - // TODO(gvisor.dev/issue/1476): Support RWF_NOWAIT. + // TODO(gvisor.dev/issue/2601): Support RWF_NOWAIT. if opts.Flags != 0 { return 0, syserror.EOPNOTSUPP } @@ -254,7 +254,7 @@ func (s *SocketVFS2) PWrite(ctx context.Context, src usermem.IOSequence, offset // Write implements vfs.FileDescriptionImpl. func (s *SocketVFS2) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { // All flags other than RWF_NOWAIT should be ignored. - // TODO(gvisor.dev/issue/1476): Support RWF_NOWAIT. + // TODO(gvisor.dev/issue/2601): Support RWF_NOWAIT. if opts.Flags != 0 { return 0, syserror.EOPNOTSUPP } -- cgit v1.2.3 From 5e1e61fbcbe8fa3cc8b104fadb8cdef3ad29c31f Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Fri, 1 May 2020 16:08:26 -0700 Subject: Automated rollback of changelist 308674219 PiperOrigin-RevId: 309491861 --- pkg/sentry/socket/netfilter/tcp_matcher.go | 5 +- pkg/sentry/socket/netfilter/udp_matcher.go | 5 +- pkg/tcpip/buffer/view.go | 55 ++++++++++---- pkg/tcpip/buffer/view_test.go | 113 +++++++++++++++++++++++++++++ pkg/tcpip/link/fdbased/endpoint.go | 3 + pkg/tcpip/link/loopback/loopback.go | 10 +-- pkg/tcpip/link/sharedmem/sharedmem_test.go | 2 +- pkg/tcpip/link/sniffer/sniffer.go | 65 +++++++++++++---- pkg/tcpip/network/arp/arp.go | 5 +- pkg/tcpip/network/ipv4/icmp.go | 20 +++-- pkg/tcpip/network/ipv4/ipv4.go | 12 ++- pkg/tcpip/network/ipv6/icmp.go | 74 ++++++++++++------- pkg/tcpip/network/ipv6/icmp_test.go | 3 +- pkg/tcpip/network/ipv6/ipv6.go | 6 +- pkg/tcpip/stack/forwarder_test.go | 13 ++-- pkg/tcpip/stack/iptables.go | 22 +++++- pkg/tcpip/stack/iptables_targets.go | 23 ++++-- pkg/tcpip/stack/nic.go | 34 +++------ pkg/tcpip/stack/packet_buffer.go | 8 +- pkg/tcpip/stack/stack_test.go | 10 ++- pkg/tcpip/stack/transport_test.go | 5 +- pkg/tcpip/transport/icmp/endpoint.go | 8 +- pkg/tcpip/transport/tcp/segment.go | 29 +++++--- pkg/tcpip/transport/tcp/tcp_test.go | 4 +- pkg/tcpip/transport/udp/endpoint.go | 6 +- pkg/tcpip/transport/udp/protocol.go | 9 ++- 26 files changed, 402 insertions(+), 147 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/tcp_matcher.go b/pkg/sentry/socket/netfilter/tcp_matcher.go index ff1cfd8f6..55c0f04f3 100644 --- a/pkg/sentry/socket/netfilter/tcp_matcher.go +++ b/pkg/sentry/socket/netfilter/tcp_matcher.go @@ -121,12 +121,13 @@ func (tm *TCPMatcher) Match(hook stack.Hook, pkt stack.PacketBuffer, interfaceNa tcpHeader = header.TCP(pkt.TransportHeader) } else { // The TCP header hasn't been parsed yet. We have to do it here. - if len(pkt.Data.First()) < header.TCPMinimumSize { + hdr, ok := pkt.Data.PullUp(header.TCPMinimumSize) + if !ok { // There's no valid TCP header here, so we hotdrop the // packet. return false, true } - tcpHeader = header.TCP(pkt.Data.First()) + tcpHeader = header.TCP(hdr) } // Check whether the source and destination ports are within the diff --git a/pkg/sentry/socket/netfilter/udp_matcher.go b/pkg/sentry/socket/netfilter/udp_matcher.go index 3359418c1..04d03d494 100644 --- a/pkg/sentry/socket/netfilter/udp_matcher.go +++ b/pkg/sentry/socket/netfilter/udp_matcher.go @@ -120,12 +120,13 @@ func (um *UDPMatcher) Match(hook stack.Hook, pkt stack.PacketBuffer, interfaceNa udpHeader = header.UDP(pkt.TransportHeader) } else { // The UDP header hasn't been parsed yet. We have to do it here. - if len(pkt.Data.First()) < header.UDPMinimumSize { + hdr, ok := pkt.Data.PullUp(header.UDPMinimumSize) + if !ok { // There's no valid UDP header here, so we hotdrop the // packet. return false, true } - udpHeader = header.UDP(pkt.Data.First()) + udpHeader = header.UDP(hdr) } // Check whether the source and destination ports are within the diff --git a/pkg/tcpip/buffer/view.go b/pkg/tcpip/buffer/view.go index 8ec5d5d5c..f01217c91 100644 --- a/pkg/tcpip/buffer/view.go +++ b/pkg/tcpip/buffer/view.go @@ -77,7 +77,8 @@ func NewVectorisedView(size int, views []View) VectorisedView { return VectorisedView{views: views, size: size} } -// TrimFront removes the first "count" bytes of the vectorised view. +// TrimFront removes the first "count" bytes of the vectorised view. It panics +// if count > vv.Size(). func (vv *VectorisedView) TrimFront(count int) { for count > 0 && len(vv.views) > 0 { if count < len(vv.views[0]) { @@ -86,7 +87,7 @@ func (vv *VectorisedView) TrimFront(count int) { return } count -= len(vv.views[0]) - vv.RemoveFirst() + vv.removeFirst() } } @@ -104,7 +105,7 @@ func (vv *VectorisedView) Read(v View) (copied int, err error) { count -= len(vv.views[0]) copy(v[copied:], vv.views[0]) copied += len(vv.views[0]) - vv.RemoveFirst() + vv.removeFirst() } if copied == 0 { return 0, io.EOF @@ -126,7 +127,7 @@ func (vv *VectorisedView) ReadToVV(dstVV *VectorisedView, count int) (copied int count -= len(vv.views[0]) dstVV.AppendView(vv.views[0]) copied += len(vv.views[0]) - vv.RemoveFirst() + vv.removeFirst() } return copied } @@ -162,22 +163,37 @@ func (vv *VectorisedView) Clone(buffer []View) VectorisedView { return VectorisedView{views: append(buffer[:0], vv.views...), size: vv.size} } -// First returns the first view of the vectorised view. -func (vv *VectorisedView) First() View { +// PullUp returns the first "count" bytes of the vectorised view. If those +// bytes aren't already contiguous inside the vectorised view, PullUp will +// reallocate as needed to make them contiguous. PullUp fails and returns false +// when count > vv.Size(). +func (vv *VectorisedView) PullUp(count int) (View, bool) { if len(vv.views) == 0 { - return nil + return nil, count == 0 + } + if count <= len(vv.views[0]) { + return vv.views[0][:count], true + } + if count > vv.size { + return nil, false } - return vv.views[0] -} -// RemoveFirst removes the first view of the vectorised view. -func (vv *VectorisedView) RemoveFirst() { - if len(vv.views) == 0 { - return + newFirst := NewView(count) + i := 0 + for offset := 0; offset < count; i++ { + copy(newFirst[offset:], vv.views[i]) + if count-offset < len(vv.views[i]) { + vv.views[i].TrimFront(count - offset) + break + } + offset += len(vv.views[i]) + vv.views[i] = nil } - vv.size -= len(vv.views[0]) - vv.views[0] = nil - vv.views = vv.views[1:] + // We're guaranteed that i > 0, since count is too large for the first + // view. + vv.views[i-1] = newFirst + vv.views = vv.views[i-1:] + return newFirst, true } // Size returns the size in bytes of the entire content stored in the vectorised view. @@ -225,3 +241,10 @@ func (vv *VectorisedView) Readers() []bytes.Reader { } return readers } + +// removeFirst panics when len(vv.views) < 1. +func (vv *VectorisedView) removeFirst() { + vv.size -= len(vv.views[0]) + vv.views[0] = nil + vv.views = vv.views[1:] +} diff --git a/pkg/tcpip/buffer/view_test.go b/pkg/tcpip/buffer/view_test.go index 106e1994c..c56795c7b 100644 --- a/pkg/tcpip/buffer/view_test.go +++ b/pkg/tcpip/buffer/view_test.go @@ -16,6 +16,7 @@ package buffer import ( + "bytes" "reflect" "testing" ) @@ -370,3 +371,115 @@ func TestVVRead(t *testing.T) { }) } } + +var pullUpTestCases = []struct { + comment string + in VectorisedView + count int + want []byte + result VectorisedView + ok bool +}{ + { + comment: "simple case", + in: vv(2, "12"), + count: 1, + want: []byte("1"), + result: vv(2, "12"), + ok: true, + }, + { + comment: "entire View", + in: vv(2, "1", "2"), + count: 1, + want: []byte("1"), + result: vv(2, "1", "2"), + ok: true, + }, + { + comment: "spanning across two Views", + in: vv(3, "1", "23"), + count: 2, + want: []byte("12"), + result: vv(3, "12", "3"), + ok: true, + }, + { + comment: "spanning across all Views", + in: vv(5, "1", "23", "45"), + count: 5, + want: []byte("12345"), + result: vv(5, "12345"), + ok: true, + }, + { + comment: "count = 0", + in: vv(1, "1"), + count: 0, + want: []byte{}, + result: vv(1, "1"), + ok: true, + }, + { + comment: "count = size", + in: vv(1, "1"), + count: 1, + want: []byte("1"), + result: vv(1, "1"), + ok: true, + }, + { + comment: "count too large", + in: vv(3, "1", "23"), + count: 4, + want: nil, + result: vv(3, "1", "23"), + ok: false, + }, + { + comment: "empty vv", + in: vv(0, ""), + count: 1, + want: nil, + result: vv(0, ""), + ok: false, + }, + { + comment: "empty vv, count = 0", + in: vv(0, ""), + count: 0, + want: nil, + result: vv(0, ""), + ok: true, + }, + { + comment: "empty views", + in: vv(3, "", "1", "", "23"), + count: 2, + want: []byte("12"), + result: vv(3, "12", "3"), + ok: true, + }, +} + +func TestPullUp(t *testing.T) { + for _, c := range pullUpTestCases { + got, ok := c.in.PullUp(c.count) + + // Is the return value right? + if ok != c.ok { + t.Errorf("Test %q failed when calling PullUp(%d) on %v. Got an ok of %t. Want %t", + c.comment, c.count, c.in, ok, c.ok) + } + if bytes.Compare(got, View(c.want)) != 0 { + t.Errorf("Test %q failed when calling PullUp(%d) on %v. Got %v. Want %v", + c.comment, c.count, c.in, got, c.want) + } + + // Is the underlying structure right? + if !reflect.DeepEqual(c.in, c.result) { + t.Errorf("Test %q failed when calling PullUp(%d). Got vv with structure %v. Wanted %v", + c.comment, c.count, c.in, c.result) + } + } +} diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go index 53a9712c6..affa1bbdf 100644 --- a/pkg/tcpip/link/fdbased/endpoint.go +++ b/pkg/tcpip/link/fdbased/endpoint.go @@ -436,6 +436,9 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne if pkt.Data.Size() == 0 { return rawfile.NonBlockingWrite(fd, pkt.Header.View()) } + if pkt.Header.UsedLength() == 0 { + return rawfile.NonBlockingWrite(fd, pkt.Data.ToView()) + } return rawfile.NonBlockingWrite3(fd, pkt.Header.View(), pkt.Data.ToView(), nil) } diff --git a/pkg/tcpip/link/loopback/loopback.go b/pkg/tcpip/link/loopback/loopback.go index 1e2255bfa..073c84ef9 100644 --- a/pkg/tcpip/link/loopback/loopback.go +++ b/pkg/tcpip/link/loopback/loopback.go @@ -98,13 +98,13 @@ func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, stack.PacketBufferList // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { - // Reject the packet if it's shorter than an ethernet header. - if vv.Size() < header.EthernetMinimumSize { + // There should be an ethernet header at the beginning of vv. + hdr, ok := vv.PullUp(header.EthernetMinimumSize) + if !ok { + // Reject the packet if it's shorter than an ethernet header. return tcpip.ErrBadAddress } - - // There should be an ethernet header at the beginning of vv. - linkHeader := header.Ethernet(vv.First()[:header.EthernetMinimumSize]) + linkHeader := header.Ethernet(hdr) vv.TrimFront(len(linkHeader)) e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, linkHeader.Type(), stack.PacketBuffer{ Data: vv, diff --git a/pkg/tcpip/link/sharedmem/sharedmem_test.go b/pkg/tcpip/link/sharedmem/sharedmem_test.go index 27ea3f531..33f640b85 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem_test.go +++ b/pkg/tcpip/link/sharedmem/sharedmem_test.go @@ -674,7 +674,7 @@ func TestSimpleReceive(t *testing.T) { // Wait for packet to be received, then check it. c.waitForPackets(1, time.After(5*time.Second), "Timeout waiting for packet") c.mu.Lock() - rcvd := []byte(c.packets[0].vv.First()) + rcvd := []byte(c.packets[0].vv.ToView()) c.packets = c.packets[:0] c.mu.Unlock() diff --git a/pkg/tcpip/link/sniffer/sniffer.go b/pkg/tcpip/link/sniffer/sniffer.go index be2537a82..0799c8f4d 100644 --- a/pkg/tcpip/link/sniffer/sniffer.go +++ b/pkg/tcpip/link/sniffer/sniffer.go @@ -171,11 +171,7 @@ func (e *endpoint) GSOMaxSize() uint32 { func (e *endpoint) dumpPacket(prefix string, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { writer := e.writer if writer == nil && atomic.LoadUint32(&LogPackets) == 1 { - first := pkt.Header.View() - if len(first) == 0 { - first = pkt.Data.First() - } - logPacket(prefix, protocol, first, gso) + logPacket(prefix, protocol, pkt, gso) } if writer != nil && atomic.LoadUint32(&LogPacketsToPCAP) == 1 { totalLength := pkt.Header.UsedLength() + pkt.Data.Size() @@ -238,7 +234,7 @@ func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { // Wait implements stack.LinkEndpoint.Wait. func (e *endpoint) Wait() { e.lower.Wait() } -func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.View, gso *stack.GSO) { +func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer, gso *stack.GSO) { // Figure out the network layer info. var transProto uint8 src := tcpip.Address("unknown") @@ -247,28 +243,49 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie size := uint16(0) var fragmentOffset uint16 var moreFragments bool + + // Create a clone of pkt, including any headers if present. Avoid allocating + // backing memory for the clone. + views := [8]buffer.View{} + vv := buffer.NewVectorisedView(0, views[:0]) + vv.AppendView(pkt.Header.View()) + vv.Append(pkt.Data) + switch protocol { case header.IPv4ProtocolNumber: - ipv4 := header.IPv4(b) + hdr, ok := vv.PullUp(header.IPv4MinimumSize) + if !ok { + return + } + ipv4 := header.IPv4(hdr) fragmentOffset = ipv4.FragmentOffset() moreFragments = ipv4.Flags()&header.IPv4FlagMoreFragments == header.IPv4FlagMoreFragments src = ipv4.SourceAddress() dst = ipv4.DestinationAddress() transProto = ipv4.Protocol() size = ipv4.TotalLength() - uint16(ipv4.HeaderLength()) - b = b[ipv4.HeaderLength():] + vv.TrimFront(int(ipv4.HeaderLength())) id = int(ipv4.ID()) case header.IPv6ProtocolNumber: - ipv6 := header.IPv6(b) + hdr, ok := vv.PullUp(header.IPv6MinimumSize) + if !ok { + return + } + ipv6 := header.IPv6(hdr) src = ipv6.SourceAddress() dst = ipv6.DestinationAddress() transProto = ipv6.NextHeader() size = ipv6.PayloadLength() - b = b[header.IPv6MinimumSize:] + vv.TrimFront(header.IPv6MinimumSize) case header.ARPProtocolNumber: - arp := header.ARP(b) + hdr, ok := vv.PullUp(header.ARPSize) + if !ok { + return + } + vv.TrimFront(header.ARPSize) + arp := header.ARP(hdr) log.Infof( "%s arp %v (%v) -> %v (%v) valid:%v", prefix, @@ -284,7 +301,7 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie // We aren't guaranteed to have a transport header - it's possible for // writes via raw endpoints to contain only network headers. - if minSize, ok := transportProtocolMinSizes[tcpip.TransportProtocolNumber(transProto)]; ok && len(b) < minSize { + if minSize, ok := transportProtocolMinSizes[tcpip.TransportProtocolNumber(transProto)]; ok && vv.Size() < minSize { log.Infof("%s %v -> %v transport protocol: %d, but no transport header found (possible raw packet)", prefix, src, dst, transProto) return } @@ -297,7 +314,11 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie switch tcpip.TransportProtocolNumber(transProto) { case header.ICMPv4ProtocolNumber: transName = "icmp" - icmp := header.ICMPv4(b) + hdr, ok := vv.PullUp(header.ICMPv4MinimumSize) + if !ok { + break + } + icmp := header.ICMPv4(hdr) icmpType := "unknown" if fragmentOffset == 0 { switch icmp.Type() { @@ -330,7 +351,11 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie case header.ICMPv6ProtocolNumber: transName = "icmp" - icmp := header.ICMPv6(b) + hdr, ok := vv.PullUp(header.ICMPv6MinimumSize) + if !ok { + break + } + icmp := header.ICMPv6(hdr) icmpType := "unknown" switch icmp.Type() { case header.ICMPv6DstUnreachable: @@ -361,7 +386,11 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie case header.UDPProtocolNumber: transName = "udp" - udp := header.UDP(b) + hdr, ok := vv.PullUp(header.UDPMinimumSize) + if !ok { + break + } + udp := header.UDP(hdr) if fragmentOffset == 0 && len(udp) >= header.UDPMinimumSize { srcPort = udp.SourcePort() dstPort = udp.DestinationPort() @@ -371,7 +400,11 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, b buffer.Vie case header.TCPProtocolNumber: transName = "tcp" - tcp := header.TCP(b) + hdr, ok := vv.PullUp(header.TCPMinimumSize) + if !ok { + break + } + tcp := header.TCP(hdr) if fragmentOffset == 0 && len(tcp) >= header.TCPMinimumSize { offset := int(tcp.DataOffset()) if offset < header.TCPMinimumSize { diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index 9f47b4ff2..9d0797af7 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -99,7 +99,10 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuf } func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { - v := pkt.Data.First() + v, ok := pkt.Data.PullUp(header.ARPSize) + if !ok { + return + } h := header.ARP(v) if !h.IsValid() { return diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go index c4bf1ba5c..4cbefe5ab 100644 --- a/pkg/tcpip/network/ipv4/icmp.go +++ b/pkg/tcpip/network/ipv4/icmp.go @@ -25,7 +25,11 @@ import ( // used to find out which transport endpoint must be notified about the ICMP // packet. func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack.PacketBuffer) { - h := header.IPv4(pkt.Data.First()) + h, ok := pkt.Data.PullUp(header.IPv4MinimumSize) + if !ok { + return + } + hdr := header.IPv4(h) // We don't use IsValid() here because ICMP only requires that the IP // header plus 8 bytes of the transport header be included. So it's @@ -34,12 +38,12 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. // // Drop packet if it doesn't have the basic IPv4 header or if the // original source address doesn't match the endpoint's address. - if len(h) < header.IPv4MinimumSize || h.SourceAddress() != e.id.LocalAddress { + if hdr.SourceAddress() != e.id.LocalAddress { return } - hlen := int(h.HeaderLength()) - if pkt.Data.Size() < hlen || h.FragmentOffset() != 0 { + hlen := int(hdr.HeaderLength()) + if pkt.Data.Size() < hlen || hdr.FragmentOffset() != 0 { // We won't be able to handle this if it doesn't contain the // full IPv4 header, or if it's a fragment not at offset 0 // (because it won't have the transport header). @@ -48,15 +52,15 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. // Skip the ip header, then deliver control message. pkt.Data.TrimFront(hlen) - p := h.TransportProtocol() - e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, h.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) + p := hdr.TransportProtocol() + e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, hdr.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) } func (e *endpoint) handleICMP(r *stack.Route, pkt stack.PacketBuffer) { stats := r.Stats() received := stats.ICMP.V4PacketsReceived - v := pkt.Data.First() - if len(v) < header.ICMPv4MinimumSize { + v, ok := pkt.Data.PullUp(header.ICMPv4MinimumSize) + if !ok { received.Invalid.Increment() return } diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index a9dec0c0e..1d61fddad 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -333,7 +333,11 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuffer) *tcpip.Error { // The packet already has an IP header, but there are a few required // checks. - ip := header.IPv4(pkt.Data.First()) + h, ok := pkt.Data.PullUp(header.IPv4MinimumSize) + if !ok { + return tcpip.ErrInvalidOptionValue + } + ip := header.IPv4(h) if !ip.IsValid(pkt.Data.Size()) { return tcpip.ErrInvalidOptionValue } @@ -383,7 +387,11 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuf // HandlePacket is called by the link layer when new ipv4 packets arrive for // this endpoint. func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { - headerView := pkt.Data.First() + headerView, ok := pkt.Data.PullUp(header.IPv4MinimumSize) + if !ok { + r.Stats().IP.MalformedPacketsReceived.Increment() + return + } h := header.IPv4(headerView) if !h.IsValid(pkt.Data.Size()) { r.Stats().IP.MalformedPacketsReceived.Increment() diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index b68983d10..bdf3a0d25 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -28,7 +28,11 @@ import ( // used to find out which transport endpoint must be notified about the ICMP // packet. func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack.PacketBuffer) { - h := header.IPv6(pkt.Data.First()) + h, ok := pkt.Data.PullUp(header.IPv6MinimumSize) + if !ok { + return + } + hdr := header.IPv6(h) // We don't use IsValid() here because ICMP only requires that up to // 1280 bytes of the original packet be included. So it's likely that it @@ -36,17 +40,21 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. // // Drop packet if it doesn't have the basic IPv6 header or if the // original source address doesn't match the endpoint's address. - if len(h) < header.IPv6MinimumSize || h.SourceAddress() != e.id.LocalAddress { + if hdr.SourceAddress() != e.id.LocalAddress { return } // Skip the IP header, then handle the fragmentation header if there // is one. pkt.Data.TrimFront(header.IPv6MinimumSize) - p := h.TransportProtocol() + p := hdr.TransportProtocol() if p == header.IPv6FragmentHeader { - f := header.IPv6Fragment(pkt.Data.First()) - if !f.IsValid() || f.FragmentOffset() != 0 { + f, ok := pkt.Data.PullUp(header.IPv6FragmentHeaderSize) + if !ok { + return + } + fragHdr := header.IPv6Fragment(f) + if !fragHdr.IsValid() || fragHdr.FragmentOffset() != 0 { // We can't handle fragments that aren't at offset 0 // because they don't have the transport headers. return @@ -55,19 +63,19 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. // Skip fragmentation header and find out the actual protocol // number. pkt.Data.TrimFront(header.IPv6FragmentHeaderSize) - p = f.TransportProtocol() + p = fragHdr.TransportProtocol() } // Deliver the control packet to the transport endpoint. - e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, h.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) + e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, hdr.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) } func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.PacketBuffer, hasFragmentHeader bool) { stats := r.Stats().ICMP sent := stats.V6PacketsSent received := stats.V6PacketsReceived - v := pkt.Data.First() - if len(v) < header.ICMPv6MinimumSize { + v, ok := pkt.Data.PullUp(header.ICMPv6HeaderSize) + if !ok { received.Invalid.Increment() return } @@ -76,11 +84,9 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P // Validate ICMPv6 checksum before processing the packet. // - // Only the first view in vv is accounted for by h. To account for the - // rest of vv, a shallow copy is made and the first view is removed. // This copy is used as extra payload during the checksum calculation. payload := pkt.Data.Clone(nil) - payload.RemoveFirst() + payload.TrimFront(len(h)) if got, want := h.Checksum(), header.ICMPv6Checksum(h, iph.SourceAddress(), iph.DestinationAddress(), payload); got != want { received.Invalid.Increment() return @@ -101,34 +107,40 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P switch h.Type() { case header.ICMPv6PacketTooBig: received.PacketTooBig.Increment() - if len(v) < header.ICMPv6PacketTooBigMinimumSize { + hdr, ok := pkt.Data.PullUp(header.ICMPv6PacketTooBigMinimumSize) + if !ok { received.Invalid.Increment() return } pkt.Data.TrimFront(header.ICMPv6PacketTooBigMinimumSize) - mtu := h.MTU() + mtu := header.ICMPv6(hdr).MTU() e.handleControl(stack.ControlPacketTooBig, calculateMTU(mtu), pkt) case header.ICMPv6DstUnreachable: received.DstUnreachable.Increment() - if len(v) < header.ICMPv6DstUnreachableMinimumSize { + hdr, ok := pkt.Data.PullUp(header.ICMPv6DstUnreachableMinimumSize) + if !ok { received.Invalid.Increment() return } pkt.Data.TrimFront(header.ICMPv6DstUnreachableMinimumSize) - switch h.Code() { + switch header.ICMPv6(hdr).Code() { case header.ICMPv6PortUnreachable: e.handleControl(stack.ControlPortUnreachable, 0, pkt) } case header.ICMPv6NeighborSolicit: received.NeighborSolicit.Increment() - if len(v) < header.ICMPv6NeighborSolicitMinimumSize || !isNDPValid() { + if pkt.Data.Size() < header.ICMPv6NeighborSolicitMinimumSize || !isNDPValid() { received.Invalid.Increment() return } - ns := header.NDPNeighborSolicit(h.NDPPayload()) + // The remainder of payload must be only the neighbor solicitation, so + // payload.ToView() always returns the solicitation. Per RFC 6980 section 5, + // NDP messages cannot be fragmented. Also note that in the common case NDP + // datagrams are very small and ToView() will not incur allocations. + ns := header.NDPNeighborSolicit(payload.ToView()) it, err := ns.Options().Iter(true) if err != nil { // If we have a malformed NDP NS option, drop the packet. @@ -286,12 +298,16 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P case header.ICMPv6NeighborAdvert: received.NeighborAdvert.Increment() - if len(v) < header.ICMPv6NeighborAdvertSize || !isNDPValid() { + if pkt.Data.Size() < header.ICMPv6NeighborAdvertSize || !isNDPValid() { received.Invalid.Increment() return } - na := header.NDPNeighborAdvert(h.NDPPayload()) + // The remainder of payload must be only the neighbor advertisement, so + // payload.ToView() always returns the advertisement. Per RFC 6980 section + // 5, NDP messages cannot be fragmented. Also note that in the common case + // NDP datagrams are very small and ToView() will not incur allocations. + na := header.NDPNeighborAdvert(payload.ToView()) it, err := na.Options().Iter(true) if err != nil { // If we have a malformed NDP NA option, drop the packet. @@ -363,14 +379,15 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P case header.ICMPv6EchoRequest: received.EchoRequest.Increment() - if len(v) < header.ICMPv6EchoMinimumSize { + icmpHdr, ok := pkt.Data.PullUp(header.ICMPv6EchoMinimumSize) + if !ok { received.Invalid.Increment() return } pkt.Data.TrimFront(header.ICMPv6EchoMinimumSize) hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv6EchoMinimumSize) packet := header.ICMPv6(hdr.Prepend(header.ICMPv6EchoMinimumSize)) - copy(packet, h) + copy(packet, icmpHdr) packet.SetType(header.ICMPv6EchoReply) packet.SetChecksum(header.ICMPv6Checksum(packet, r.LocalAddress, r.RemoteAddress, pkt.Data)) if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, stack.PacketBuffer{ @@ -384,7 +401,7 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P case header.ICMPv6EchoReply: received.EchoReply.Increment() - if len(v) < header.ICMPv6EchoMinimumSize { + if pkt.Data.Size() < header.ICMPv6EchoMinimumSize { received.Invalid.Increment() return } @@ -406,8 +423,9 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P case header.ICMPv6RouterAdvert: received.RouterAdvert.Increment() - p := h.NDPPayload() - if len(p) < header.NDPRAMinimumSize || !isNDPValid() { + // Is the NDP payload of sufficient size to hold a Router + // Advertisement? + if pkt.Data.Size()-header.ICMPv6HeaderSize < header.NDPRAMinimumSize || !isNDPValid() { received.Invalid.Increment() return } @@ -425,7 +443,11 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P return } - ra := header.NDPRouterAdvert(p) + // The remainder of payload must be only the router advertisement, so + // payload.ToView() always returns the advertisement. Per RFC 6980 section + // 5, NDP messages cannot be fragmented. Also note that in the common case + // NDP datagrams are very small and ToView() will not incur allocations. + ra := header.NDPRouterAdvert(payload.ToView()) opts := ra.Options() // Are options valid as per the wire format? diff --git a/pkg/tcpip/network/ipv6/icmp_test.go b/pkg/tcpip/network/ipv6/icmp_test.go index bd099a7f8..d412ff688 100644 --- a/pkg/tcpip/network/ipv6/icmp_test.go +++ b/pkg/tcpip/network/ipv6/icmp_test.go @@ -166,7 +166,8 @@ func TestICMPCounts(t *testing.T) { }, { typ: header.ICMPv6NeighborSolicit, - size: header.ICMPv6NeighborSolicitMinimumSize}, + size: header.ICMPv6NeighborSolicitMinimumSize, + }, { typ: header.ICMPv6NeighborAdvert, size: header.ICMPv6NeighborAdvertMinimumSize, diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index 82928fb66..daf1fcbc6 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -171,7 +171,11 @@ func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuffe // HandlePacket is called by the link layer when new ipv6 packets arrive for // this endpoint. func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { - headerView := pkt.Data.First() + headerView, ok := pkt.Data.PullUp(header.IPv6MinimumSize) + if !ok { + r.Stats().IP.MalformedPacketsReceived.Increment() + return + } h := header.IPv6(headerView) if !h.IsValid(pkt.Data.Size()) { r.Stats().IP.MalformedPacketsReceived.Increment() diff --git a/pkg/tcpip/stack/forwarder_test.go b/pkg/tcpip/stack/forwarder_test.go index 9bc97b84e..8084d50bc 100644 --- a/pkg/tcpip/stack/forwarder_test.go +++ b/pkg/tcpip/stack/forwarder_test.go @@ -70,7 +70,10 @@ func (f *fwdTestNetworkEndpoint) ID() *NetworkEndpointID { func (f *fwdTestNetworkEndpoint) HandlePacket(r *Route, pkt PacketBuffer) { // Consume the network header. - b := pkt.Data.First() + b, ok := pkt.Data.PullUp(fwdTestNetHeaderLen) + if !ok { + return + } pkt.Data.TrimFront(fwdTestNetHeaderLen) // Dispatch the packet to the transport protocol. @@ -477,7 +480,7 @@ func TestForwardingWithFakeResolverPartialTimeout(t *testing.T) { t.Fatal("packet not forwarded") } - b := p.Pkt.Header.View() + b := p.Pkt.Data.ToView() if b[0] != 3 { t.Fatalf("got b[0] = %d, want = 3", b[0]) } @@ -521,7 +524,7 @@ func TestForwardingWithFakeResolverTwoPackets(t *testing.T) { t.Fatal("packet not forwarded") } - b := p.Pkt.Header.View() + b := p.Pkt.Data.ToView() if b[0] != 3 { t.Fatalf("got b[0] = %d, want = 3", b[0]) } @@ -568,7 +571,7 @@ func TestForwardingWithFakeResolverManyPackets(t *testing.T) { t.Fatal("packet not forwarded") } - b := p.Pkt.Header.View() + b := p.Pkt.Data.ToView() if b[0] != 3 { t.Fatalf("got b[0] = %d, want = 3", b[0]) } @@ -623,7 +626,7 @@ func TestForwardingWithFakeResolverManyResolutions(t *testing.T) { // The first 5 packets (address 3 to 7) should not be forwarded // because their address resolutions are interrupted. - b := p.Pkt.Header.View() + b := p.Pkt.Data.ToView() if b[0] < 8 { t.Fatalf("got b[0] = %d, want b[0] >= 8", b[0]) } diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go index 6c0a4b24d..6b91159d4 100644 --- a/pkg/tcpip/stack/iptables.go +++ b/pkg/tcpip/stack/iptables.go @@ -212,6 +212,11 @@ func (it *IPTables) Check(hook Hook, pkt PacketBuffer) bool { // CheckPackets runs pkts through the rules for hook and returns a map of packets that // should not go forward. // +// Precondition: pkt is a IPv4 packet of at least length header.IPv4MinimumSize. +// +// TODO(gvisor.dev/issue/170): pk.NetworkHeader will always be set as a +// precondition. +// // NOTE: unlike the Check API the returned map contains packets that should be // dropped. func (it *IPTables) CheckPackets(hook Hook, pkts PacketBufferList) (drop map[*PacketBuffer]struct{}) { @@ -226,7 +231,9 @@ func (it *IPTables) CheckPackets(hook Hook, pkts PacketBufferList) (drop map[*Pa return drop } -// Precondition: pkt.NetworkHeader is set. +// Precondition: pkt is a IPv4 packet of at least length header.IPv4MinimumSize. +// TODO(gvisor.dev/issue/170): pk.NetworkHeader will always be set as a +// precondition. func (it *IPTables) checkChain(hook Hook, pkt PacketBuffer, table Table, ruleIdx int) chainVerdict { // Start from ruleIdx and walk the list of rules until a rule gives us // a verdict. @@ -271,14 +278,21 @@ func (it *IPTables) checkChain(hook Hook, pkt PacketBuffer, table Table, ruleIdx return chainDrop } -// Precondition: pk.NetworkHeader is set. +// Precondition: pkt is a IPv4 packet of at least length header.IPv4MinimumSize. +// TODO(gvisor.dev/issue/170): pk.NetworkHeader will always be set as a +// precondition. func (it *IPTables) checkRule(hook Hook, pkt PacketBuffer, table Table, ruleIdx int) (RuleVerdict, int) { rule := table.Rules[ruleIdx] // If pkt.NetworkHeader hasn't been set yet, it will be contained in - // pkt.Data.First(). + // pkt.Data. if pkt.NetworkHeader == nil { - pkt.NetworkHeader = pkt.Data.First() + var ok bool + pkt.NetworkHeader, ok = pkt.Data.PullUp(header.IPv4MinimumSize) + if !ok { + // Precondition has been violated. + panic(fmt.Sprintf("iptables checks require IPv4 headers of at least %d bytes", header.IPv4MinimumSize)) + } } // Check whether the packet matches the IP header filter. diff --git a/pkg/tcpip/stack/iptables_targets.go b/pkg/tcpip/stack/iptables_targets.go index 7b4543caf..8be61f4b1 100644 --- a/pkg/tcpip/stack/iptables_targets.go +++ b/pkg/tcpip/stack/iptables_targets.go @@ -96,9 +96,12 @@ func (rt RedirectTarget) Action(pkt PacketBuffer) (RuleVerdict, int) { newPkt := pkt.Clone() // Set network header. - headerView := newPkt.Data.First() + headerView, ok := newPkt.Data.PullUp(header.IPv4MinimumSize) + if !ok { + return RuleDrop, 0 + } netHeader := header.IPv4(headerView) - newPkt.NetworkHeader = headerView[:header.IPv4MinimumSize] + newPkt.NetworkHeader = headerView hlen := int(netHeader.HeaderLength()) tlen := int(netHeader.TotalLength()) @@ -117,10 +120,14 @@ func (rt RedirectTarget) Action(pkt PacketBuffer) (RuleVerdict, int) { if newPkt.TransportHeader != nil { udpHeader = header.UDP(newPkt.TransportHeader) } else { - if len(pkt.Data.First()) < header.UDPMinimumSize { + if pkt.Data.Size() < header.UDPMinimumSize { + return RuleDrop, 0 + } + hdr, ok := newPkt.Data.PullUp(header.UDPMinimumSize) + if !ok { return RuleDrop, 0 } - udpHeader = header.UDP(newPkt.Data.First()) + udpHeader = header.UDP(hdr) } udpHeader.SetDestinationPort(rt.MinPort) case header.TCPProtocolNumber: @@ -128,10 +135,14 @@ func (rt RedirectTarget) Action(pkt PacketBuffer) (RuleVerdict, int) { if newPkt.TransportHeader != nil { tcpHeader = header.TCP(newPkt.TransportHeader) } else { - if len(pkt.Data.First()) < header.TCPMinimumSize { + if pkt.Data.Size() < header.TCPMinimumSize { return RuleDrop, 0 } - tcpHeader = header.TCP(newPkt.TransportHeader) + hdr, ok := newPkt.Data.PullUp(header.TCPMinimumSize) + if !ok { + return RuleDrop, 0 + } + tcpHeader = header.TCP(hdr) } // TODO(gvisor.dev/issue/170): Need to recompute checksum // and implement nat connection tracking to support TCP. diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 440970a21..7b54919bb 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -1212,12 +1212,12 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link n.stack.stats.IP.PacketsReceived.Increment() } - if len(pkt.Data.First()) < netProto.MinimumPacketSize() { + netHeader, ok := pkt.Data.PullUp(netProto.MinimumPacketSize()) + if !ok { n.stack.stats.MalformedRcvdPackets.Increment() return } - - src, dst := netProto.ParseAddresses(pkt.Data.First()) + src, dst := netProto.ParseAddresses(netHeader) if n.stack.handleLocal && !n.isLoopback() && n.getRef(protocol, src) != nil { // The source address is one of our own, so we never should have gotten a @@ -1298,22 +1298,8 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link func (n *NIC) forwardPacket(r *Route, protocol tcpip.NetworkProtocolNumber, pkt PacketBuffer) { // TODO(b/143425874) Decrease the TTL field in forwarded packets. - - firstData := pkt.Data.First() - pkt.Data.RemoveFirst() - - if linkHeaderLen := int(n.linkEP.MaxHeaderLength()); linkHeaderLen == 0 { - pkt.Header = buffer.NewPrependableFromView(firstData) - } else { - firstDataLen := len(firstData) - - // pkt.Header should have enough capacity to hold n.linkEP's headers. - pkt.Header = buffer.NewPrependable(firstDataLen + linkHeaderLen) - - // TODO(b/151227689): avoid copying the packet when forwarding - if n := copy(pkt.Header.Prepend(firstDataLen), firstData); n != firstDataLen { - panic(fmt.Sprintf("copied %d bytes, expected %d", n, firstDataLen)) - } + if linkHeaderLen := int(n.linkEP.MaxHeaderLength()); linkHeaderLen != 0 { + pkt.Header = buffer.NewPrependable(linkHeaderLen) } if err := n.linkEP.WritePacket(r, nil /* gso */, protocol, pkt); err != nil { @@ -1341,12 +1327,13 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN // validly formed. n.stack.demux.deliverRawPacket(r, protocol, pkt) - if len(pkt.Data.First()) < transProto.MinimumPacketSize() { + transHeader, ok := pkt.Data.PullUp(transProto.MinimumPacketSize()) + if !ok { n.stack.stats.MalformedRcvdPackets.Increment() return } - srcPort, dstPort, err := transProto.ParsePorts(pkt.Data.First()) + srcPort, dstPort, err := transProto.ParsePorts(transHeader) if err != nil { n.stack.stats.MalformedRcvdPackets.Increment() return @@ -1384,11 +1371,12 @@ func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcp // ICMPv4 only guarantees that 8 bytes of the transport protocol will // be present in the payload. We know that the ports are within the // first 8 bytes for all known transport protocols. - if len(pkt.Data.First()) < 8 { + transHeader, ok := pkt.Data.PullUp(8) + if !ok { return } - srcPort, dstPort, err := transProto.ParsePorts(pkt.Data.First()) + srcPort, dstPort, err := transProto.ParsePorts(transHeader) if err != nil { return } diff --git a/pkg/tcpip/stack/packet_buffer.go b/pkg/tcpip/stack/packet_buffer.go index 06d312207..9ff80ab24 100644 --- a/pkg/tcpip/stack/packet_buffer.go +++ b/pkg/tcpip/stack/packet_buffer.go @@ -37,7 +37,13 @@ type PacketBuffer struct { Data buffer.VectorisedView // Header holds the headers of outbound packets. As a packet is passed - // down the stack, each layer adds to Header. + // down the stack, each layer adds to Header. Note that forwarded + // packets don't populate Headers on their way out -- their headers and + // payload are never parsed out and remain in Data. + // + // TODO(gvisor.dev/issue/170): Forwarded packets don't currently + // populate Header, but should. This will be doable once early parsing + // (https://github.com/google/gvisor/pull/1995) is supported. Header buffer.Prependable // These fields are used by both inbound and outbound packets. They diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index 3f4e08434..1a2cf007c 100644 --- a/pkg/tcpip/stack/stack_test.go +++ b/pkg/tcpip/stack/stack_test.go @@ -95,16 +95,18 @@ func (f *fakeNetworkEndpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffe f.proto.packetCount[int(f.id.LocalAddress[0])%len(f.proto.packetCount)]++ // Consume the network header. - b := pkt.Data.First() + b, ok := pkt.Data.PullUp(fakeNetHeaderLen) + if !ok { + return + } pkt.Data.TrimFront(fakeNetHeaderLen) // Handle control packets. if b[2] == uint8(fakeControlProtocol) { - nb := pkt.Data.First() - if len(nb) < fakeNetHeaderLen { + nb, ok := pkt.Data.PullUp(fakeNetHeaderLen) + if !ok { return } - pkt.Data.TrimFront(fakeNetHeaderLen) f.dispatcher.DeliverTransportControlPacket(tcpip.Address(nb[1:2]), tcpip.Address(nb[0:1]), fakeNetNumber, tcpip.TransportProtocolNumber(nb[2]), stack.ControlPortUnreachable, 0, pkt) return diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go index 3084e6593..a611e44ab 100644 --- a/pkg/tcpip/stack/transport_test.go +++ b/pkg/tcpip/stack/transport_test.go @@ -642,10 +642,11 @@ func TestTransportForwarding(t *testing.T) { t.Fatal("Response packet not forwarded") } - if dst := p.Pkt.Header.View()[0]; dst != 3 { + hdrs := p.Pkt.Data.ToView() + if dst := hdrs[0]; dst != 3 { t.Errorf("Response packet has incorrect destination addresss: got = %d, want = 3", dst) } - if src := p.Pkt.Header.View()[1]; src != 1 { + if src := hdrs[1]; src != 1 { t.Errorf("Response packet has incorrect source addresss: got = %d, want = 3", src) } } diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index feef8dca0..b1d820372 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -747,15 +747,15 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk // Only accept echo replies. switch e.NetProto { case header.IPv4ProtocolNumber: - h := header.ICMPv4(pkt.Data.First()) - if h.Type() != header.ICMPv4EchoReply { + h, ok := pkt.Data.PullUp(header.ICMPv4MinimumSize) + if !ok || header.ICMPv4(h).Type() != header.ICMPv4EchoReply { e.stack.Stats().DroppedPackets.Increment() e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() return } case header.IPv6ProtocolNumber: - h := header.ICMPv6(pkt.Data.First()) - if h.Type() != header.ICMPv6EchoReply { + h, ok := pkt.Data.PullUp(header.ICMPv6MinimumSize) + if !ok || header.ICMPv6(h).Type() != header.ICMPv6EchoReply { e.stack.Stats().DroppedPackets.Increment() e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() return diff --git a/pkg/tcpip/transport/tcp/segment.go b/pkg/tcpip/transport/tcp/segment.go index 40461fd31..7712ce652 100644 --- a/pkg/tcpip/transport/tcp/segment.go +++ b/pkg/tcpip/transport/tcp/segment.go @@ -144,7 +144,11 @@ func (s *segment) logicalLen() seqnum.Size { // TCP checksum and stores the checksum and result of checksum verification in // the csum and csumValid fields of the segment. func (s *segment) parse() bool { - h := header.TCP(s.data.First()) + h, ok := s.data.PullUp(header.TCPMinimumSize) + if !ok { + return false + } + hdr := header.TCP(h) // h is the header followed by the payload. We check that the offset to // the data respects the following constraints: @@ -156,12 +160,16 @@ func (s *segment) parse() bool { // N.B. The segment has already been validated as having at least the // minimum TCP size before reaching here, so it's safe to read the // fields. - offset := int(h.DataOffset()) - if offset < header.TCPMinimumSize || offset > len(h) { + offset := int(hdr.DataOffset()) + if offset < header.TCPMinimumSize { + return false + } + hdrWithOpts, ok := s.data.PullUp(offset) + if !ok { return false } - s.options = []byte(h[header.TCPMinimumSize:offset]) + s.options = []byte(hdrWithOpts[header.TCPMinimumSize:]) s.parsedOptions = header.ParseTCPOptions(s.options) // Query the link capabilities to decide if checksum validation is @@ -173,18 +181,19 @@ func (s *segment) parse() bool { s.data.TrimFront(offset) } if verifyChecksum { - s.csum = h.Checksum() + hdr = header.TCP(hdrWithOpts) + s.csum = hdr.Checksum() xsum := s.route.PseudoHeaderChecksum(ProtocolNumber, uint16(s.data.Size())) - xsum = h.CalculateChecksum(xsum) + xsum = hdr.CalculateChecksum(xsum) s.data.TrimFront(offset) xsum = header.ChecksumVV(s.data, xsum) s.csumValid = xsum == 0xffff } - s.sequenceNumber = seqnum.Value(h.SequenceNumber()) - s.ackNumber = seqnum.Value(h.AckNumber()) - s.flags = h.Flags() - s.window = seqnum.Size(h.WindowSize()) + s.sequenceNumber = seqnum.Value(hdr.SequenceNumber()) + s.ackNumber = seqnum.Value(hdr.AckNumber()) + s.flags = hdr.Flags() + s.window = seqnum.Size(hdr.WindowSize()) return true } diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index 7e574859b..33e2b9a09 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -3556,7 +3556,7 @@ func TestReceivedInvalidSegmentCountIncrement(t *testing.T) { AckNum: c.IRS.Add(1), RcvWnd: 30000, }) - tcpbuf := vv.First()[header.IPv4MinimumSize:] + tcpbuf := vv.ToView()[header.IPv4MinimumSize:] tcpbuf[header.TCPDataOffset] = ((header.TCPMinimumSize - 1) / 4) << 4 c.SendSegment(vv) @@ -3583,7 +3583,7 @@ func TestReceivedIncorrectChecksumIncrement(t *testing.T) { AckNum: c.IRS.Add(1), RcvWnd: 30000, }) - tcpbuf := vv.First()[header.IPv4MinimumSize:] + tcpbuf := vv.ToView()[header.IPv4MinimumSize:] // Overwrite a byte in the payload which should cause checksum // verification to fail. tcpbuf[(tcpbuf[header.TCPDataOffset]>>4)*4] = 0x4 diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index edb54f0be..756ab913a 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -1250,8 +1250,8 @@ func (e *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { // endpoint. func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) { // Get the header then trim it from the view. - hdr := header.UDP(pkt.Data.First()) - if int(hdr.Length()) > pkt.Data.Size() { + hdr, ok := pkt.Data.PullUp(header.UDPMinimumSize) + if !ok || int(header.UDP(hdr).Length()) > pkt.Data.Size() { // Malformed packet. e.stack.Stats().UDP.MalformedPacketsReceived.Increment() e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() @@ -1286,7 +1286,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk senderAddress: tcpip.FullAddress{ NIC: r.NICID(), Addr: id.RemoteAddress, - Port: hdr.SourcePort(), + Port: header.UDP(hdr).SourcePort(), }, } packet.data = pkt.Data diff --git a/pkg/tcpip/transport/udp/protocol.go b/pkg/tcpip/transport/udp/protocol.go index 6e31a9bac..52af6de22 100644 --- a/pkg/tcpip/transport/udp/protocol.go +++ b/pkg/tcpip/transport/udp/protocol.go @@ -68,8 +68,13 @@ func (*protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) { // that don't match any existing endpoint. func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) bool { // Get the header then trim it from the view. - hdr := header.UDP(pkt.Data.First()) - if int(hdr.Length()) > pkt.Data.Size() { + h, ok := pkt.Data.PullUp(header.UDPMinimumSize) + if !ok { + // Malformed packet. + r.Stack().Stats().UDP.MalformedPacketsReceived.Increment() + return true + } + if int(header.UDP(h).Length()) > pkt.Data.Size() { // Malformed packet. r.Stack().Stats().UDP.MalformedPacketsReceived.Increment() return true -- cgit v1.2.3 From b660f16d18827f0310594c80d9387de11430f15f Mon Sep 17 00:00:00 2001 From: Nayana Bidari Date: Fri, 27 Mar 2020 12:18:45 -0700 Subject: Support for connection tracking of TCP packets. Connection tracking is used to track packets in prerouting and output hooks of iptables. The NAT rules modify the tuples in connections. The connection tracking code modifies the packets by looking at the modified tuples. --- pkg/sentry/socket/netfilter/netfilter.go | 14 +- pkg/sentry/socket/netfilter/targets.go | 3 +- pkg/sentry/socket/netfilter/tcp_matcher.go | 17 +- pkg/sentry/socket/netfilter/udp_matcher.go | 17 +- pkg/tcpip/header/tcp.go | 17 + pkg/tcpip/network/ipv4/ipv4.go | 46 ++- pkg/tcpip/stack/BUILD | 2 + pkg/tcpip/stack/conntrack.go | 480 ++++++++++++++++++++++ pkg/tcpip/stack/iptables.go | 51 ++- pkg/tcpip/stack/iptables_targets.go | 115 +++--- pkg/tcpip/stack/iptables_types.go | 4 +- pkg/tcpip/stack/nic.go | 4 +- pkg/tcpip/stack/packet_buffer.go | 4 + pkg/tcpip/stack/route.go | 13 + pkg/tcpip/stack/stack.go | 19 + pkg/tcpip/transport/tcp/BUILD | 2 +- pkg/tcpip/transport/tcp/rcv.go | 19 +- pkg/tcpip/transport/tcp/rcv_test.go | 6 +- pkg/tcpip/transport/tcpconntrack/BUILD | 1 - pkg/tcpip/transport/tcpconntrack/tcp_conntrack.go | 16 +- test/iptables/filter_output.go | 2 +- test/iptables/iptables_test.go | 66 ++- test/iptables/iptables_util.go | 2 +- test/iptables/nat.go | 103 ++++- 24 files changed, 858 insertions(+), 165 deletions(-) create mode 100644 pkg/tcpip/stack/conntrack.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 72d093aa8..40736fb38 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -251,7 +251,7 @@ func marshalTarget(target stack.Target) []byte { case stack.ReturnTarget: return marshalStandardTarget(stack.RuleReturn) case stack.RedirectTarget: - return marshalRedirectTarget() + return marshalRedirectTarget(tg) case JumpTarget: return marshalJumpTarget(tg) default: @@ -288,7 +288,7 @@ func marshalErrorTarget(errorName string) []byte { return binary.Marshal(ret, usermem.ByteOrder, target) } -func marshalRedirectTarget() []byte { +func marshalRedirectTarget(rt stack.RedirectTarget) []byte { // This is a redirect target named redirect target := linux.XTRedirectTarget{ Target: linux.XTEntryTarget{ @@ -298,6 +298,16 @@ func marshalRedirectTarget() []byte { copy(target.Target.Name[:], redirectTargetName) ret := make([]byte, 0, linux.SizeOfXTRedirectTarget) + target.NfRange.RangeSize = 1 + if rt.RangeProtoSpecified { + target.NfRange.RangeIPV4.Flags |= linux.NF_NAT_RANGE_PROTO_SPECIFIED + } + // Convert port from little endian to big endian. + port := make([]byte, 2) + binary.LittleEndian.PutUint16(port, rt.MinPort) + target.NfRange.RangeIPV4.MinPort = binary.BigEndian.Uint16(port) + binary.LittleEndian.PutUint16(port, rt.MaxPort) + target.NfRange.RangeIPV4.MaxPort = binary.BigEndian.Uint16(port) return binary.Marshal(ret, usermem.ByteOrder, target) } diff --git a/pkg/sentry/socket/netfilter/targets.go b/pkg/sentry/socket/netfilter/targets.go index c948de876..84abe8d29 100644 --- a/pkg/sentry/socket/netfilter/targets.go +++ b/pkg/sentry/socket/netfilter/targets.go @@ -15,6 +15,7 @@ package netfilter import ( + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -29,6 +30,6 @@ type JumpTarget struct { } // Action implements stack.Target.Action. -func (jt JumpTarget) Action(stack.PacketBuffer) (stack.RuleVerdict, int) { +func (jt JumpTarget) Action(*stack.PacketBuffer, *stack.ConnTrackTable, stack.Hook, *stack.GSO, *stack.Route, tcpip.Address) (stack.RuleVerdict, int) { return stack.RuleJump, jt.RuleNum } diff --git a/pkg/sentry/socket/netfilter/tcp_matcher.go b/pkg/sentry/socket/netfilter/tcp_matcher.go index 55c0f04f3..57a1e1c12 100644 --- a/pkg/sentry/socket/netfilter/tcp_matcher.go +++ b/pkg/sentry/socket/netfilter/tcp_matcher.go @@ -120,14 +120,27 @@ func (tm *TCPMatcher) Match(hook stack.Hook, pkt stack.PacketBuffer, interfaceNa if pkt.TransportHeader != nil { tcpHeader = header.TCP(pkt.TransportHeader) } else { + var length int + if hook == stack.Prerouting { + // The network header hasn't been parsed yet. We have to do it here. + hdr, ok := pkt.Data.PullUp(header.IPv4MinimumSize) + if !ok { + // There's no valid TCP header here, so we hotdrop the + // packet. + return false, true + } + h := header.IPv4(hdr) + pkt.NetworkHeader = hdr + length = int(h.HeaderLength()) + } // The TCP header hasn't been parsed yet. We have to do it here. - hdr, ok := pkt.Data.PullUp(header.TCPMinimumSize) + hdr, ok := pkt.Data.PullUp(length + header.TCPMinimumSize) if !ok { // There's no valid TCP header here, so we hotdrop the // packet. return false, true } - tcpHeader = header.TCP(hdr) + tcpHeader = header.TCP(hdr[length:]) } // Check whether the source and destination ports are within the diff --git a/pkg/sentry/socket/netfilter/udp_matcher.go b/pkg/sentry/socket/netfilter/udp_matcher.go index 04d03d494..cfa9e621d 100644 --- a/pkg/sentry/socket/netfilter/udp_matcher.go +++ b/pkg/sentry/socket/netfilter/udp_matcher.go @@ -119,14 +119,27 @@ func (um *UDPMatcher) Match(hook stack.Hook, pkt stack.PacketBuffer, interfaceNa if pkt.TransportHeader != nil { udpHeader = header.UDP(pkt.TransportHeader) } else { + var length int + if hook == stack.Prerouting { + // The network header hasn't been parsed yet. We have to do it here. + hdr, ok := pkt.Data.PullUp(header.IPv4MinimumSize) + if !ok { + // There's no valid UDP header here, so we hotdrop the + // packet. + return false, true + } + h := header.IPv4(hdr) + pkt.NetworkHeader = hdr + length = int(h.HeaderLength()) + } // The UDP header hasn't been parsed yet. We have to do it here. - hdr, ok := pkt.Data.PullUp(header.UDPMinimumSize) + hdr, ok := pkt.Data.PullUp(length + header.UDPMinimumSize) if !ok { // There's no valid UDP header here, so we hotdrop the // packet. return false, true } - udpHeader = header.UDP(hdr) + udpHeader = header.UDP(hdr[length:]) } // Check whether the source and destination ports are within the diff --git a/pkg/tcpip/header/tcp.go b/pkg/tcpip/header/tcp.go index 13480687d..21581257b 100644 --- a/pkg/tcpip/header/tcp.go +++ b/pkg/tcpip/header/tcp.go @@ -594,3 +594,20 @@ func AddTCPOptionPadding(options []byte, offset int) int { } return paddingToAdd } + +// Acceptable checks if a segment that starts at segSeq and has length segLen is +// "acceptable" for arriving in a receive window that starts at rcvNxt and ends +// before rcvAcc, according to the table on page 26 and 69 of RFC 793. +func Acceptable(segSeq seqnum.Value, segLen seqnum.Size, rcvNxt, rcvAcc seqnum.Value) bool { + if rcvNxt == rcvAcc { + return segLen == 0 && segSeq == rcvNxt + } + if segLen == 0 { + // rcvWnd is incremented by 1 because that is Linux's behavior despite the + // RFC. + return segSeq.InRange(rcvNxt, rcvAcc.Add(1)) + } + // Page 70 of RFC 793 allows packets that can be made "acceptable" by trimming + // the payload, so we'll accept any payload that overlaps the receieve window. + return rcvNxt.LessThan(segSeq.Add(segLen)) && segSeq.LessThan(rcvAcc) +} diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index 1d61fddad..9db42b2a4 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -252,11 +252,31 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw // iptables filtering. All packets that reach here are locally // generated. ipt := e.stack.IPTables() - if ok := ipt.Check(stack.Output, pkt); !ok { + if ok := ipt.Check(stack.Output, &pkt, gso, r, ""); !ok { // iptables is telling us to drop the packet. return nil } + if pkt.NatDone { + // If the packet is manipulated as per NAT Ouput rules, handle packet + // based on destination address and do not send the packet to link layer. + netHeader := header.IPv4(pkt.NetworkHeader) + ep, err := e.stack.FindNetworkEndpoint(header.IPv4ProtocolNumber, netHeader.DestinationAddress()) + if err == nil { + src := netHeader.SourceAddress() + dst := netHeader.DestinationAddress() + route := r.ReverseRoute(src, dst) + + views := make([]buffer.View, 1, 1+len(pkt.Data.Views())) + views[0] = pkt.Header.View() + views = append(views, pkt.Data.Views()...) + packet := stack.PacketBuffer{ + Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views)} + ep.HandlePacket(&route, packet) + return nil + } + } + if r.Loop&stack.PacketLoop != 0 { // The inbound path expects the network header to still be in // the PacketBuffer's Data field. @@ -302,8 +322,8 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe // iptables filtering. All packets that reach here are locally // generated. ipt := e.stack.IPTables() - dropped := ipt.CheckPackets(stack.Output, pkts) - if len(dropped) == 0 { + dropped, natPkts := ipt.CheckPackets(stack.Output, pkts, gso, r) + if len(dropped) == 0 && len(natPkts) == 0 { // Fast path: If no packets are to be dropped then we can just invoke the // faster WritePackets API directly. n, err := e.linkEP.WritePackets(r, gso, pkts, ProtocolNumber) @@ -318,6 +338,24 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe if _, ok := dropped[pkt]; ok { continue } + if _, ok := natPkts[pkt]; ok { + netHeader := header.IPv4(pkt.NetworkHeader) + ep, err := e.stack.FindNetworkEndpoint(header.IPv4ProtocolNumber, netHeader.DestinationAddress()) + if err == nil { + src := netHeader.SourceAddress() + dst := netHeader.DestinationAddress() + route := r.ReverseRoute(src, dst) + + views := make([]buffer.View, 1, 1+len(pkt.Data.Views())) + views[0] = pkt.Header.View() + views = append(views, pkt.Data.Views()...) + packet := stack.PacketBuffer{ + Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views)} + ep.HandlePacket(&route, packet) + n++ + continue + } + } if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, *pkt); err != nil { r.Stats().IP.PacketsSent.IncrementBy(uint64(n)) return n, err @@ -407,7 +445,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { // iptables filtering. All packets that reach here are intended for // this machine and will not be forwarded. ipt := e.stack.IPTables() - if ok := ipt.Check(stack.Input, pkt); !ok { + if ok := ipt.Check(stack.Input, &pkt, nil, nil, ""); !ok { // iptables is telling us to drop the packet. return } diff --git a/pkg/tcpip/stack/BUILD b/pkg/tcpip/stack/BUILD index 5e963a4af..f71073207 100644 --- a/pkg/tcpip/stack/BUILD +++ b/pkg/tcpip/stack/BUILD @@ -30,6 +30,7 @@ go_template_instance( go_library( name = "stack", srcs = [ + "conntrack.go", "dhcpv6configurationfromndpra_string.go", "forwarder.go", "icmp_rate_limit.go", @@ -62,6 +63,7 @@ go_library( "//pkg/tcpip/header", "//pkg/tcpip/ports", "//pkg/tcpip/seqnum", + "//pkg/tcpip/transport/tcpconntrack", "//pkg/waiter", "@org_golang_x_time//rate:go_default_library", ], diff --git a/pkg/tcpip/stack/conntrack.go b/pkg/tcpip/stack/conntrack.go new file mode 100644 index 000000000..7d1ede1f2 --- /dev/null +++ b/pkg/tcpip/stack/conntrack.go @@ -0,0 +1,480 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package stack + +import ( + "encoding/binary" + "sync" + "time" + + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/hash/jenkins" + "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/transport/tcpconntrack" +) + +// Connection tracking is used to track and manipulate packets for NAT rules. +// The connection is created for a packet if it does not exist. Every connection +// contains two tuples (original and reply). The tuples are manipulated if there +// is a matching NAT rule. The packet is modified by looking at the tuples in the +// Prerouting and Output hooks. + +// Direction of the tuple. +type ctDirection int + +const ( + dirOriginal ctDirection = iota + dirReply +) + +// Status of connection. +// TODO(gvisor.dev/issue/170): Add other states of connection. +type connStatus int + +const ( + connNew connStatus = iota + connEstablished +) + +// Manipulation type for the connection. +type manipType int + +const ( + manipDstPrerouting manipType = iota + manipDstOutput +) + +// connTrackMutable is the manipulatable part of the tuple. +type connTrackMutable struct { + // addr is source address of the tuple. + addr tcpip.Address + + // port is source port of the tuple. + port uint16 + + // protocol is network layer protocol. + protocol tcpip.NetworkProtocolNumber +} + +// connTrackImmutable is the non-manipulatable part of the tuple. +type connTrackImmutable struct { + // addr is destination address of the tuple. + addr tcpip.Address + + // direction is direction (original or reply) of the tuple. + direction ctDirection + + // port is destination port of the tuple. + port uint16 + + // protocol is transport layer protocol. + protocol tcpip.TransportProtocolNumber +} + +// connTrackTuple represents the tuple which is created from the +// packet. +type connTrackTuple struct { + // dst is non-manipulatable part of the tuple. + dst connTrackImmutable + + // src is manipulatable part of the tuple. + src connTrackMutable +} + +// connTrackTupleHolder is the container of tuple and connection. +type ConnTrackTupleHolder struct { + // conn is pointer to the connection tracking entry. + conn *connTrack + + // tuple is original or reply tuple. + tuple connTrackTuple +} + +// connTrack is the connection. +type connTrack struct { + // originalTupleHolder contains tuple in original direction. + originalTupleHolder ConnTrackTupleHolder + + // replyTupleHolder contains tuple in reply direction. + replyTupleHolder ConnTrackTupleHolder + + // status indicates connection is new or established. + status connStatus + + // timeout indicates the time connection should be active. + timeout time.Duration + + // manip indicates if the packet should be manipulated. + manip manipType + + // tcb is TCB control block. It is used to keep track of states + // of tcp connection. + tcb tcpconntrack.TCB + + // tcbHook indicates if the packet is inbound or outbound to + // update the state of tcb. + tcbHook Hook +} + +// ConnTrackTable contains a map of all existing connections created for +// NAT rules. +type ConnTrackTable struct { + // connMu protects connTrackTable. + connMu sync.RWMutex + + // connTrackTable maintains a map of tuples needed for connection tracking + // for iptables NAT rules. The key for the map is an integer calculated + // using seed, source address, destination address, source port and + // destination port. + CtMap map[uint32]ConnTrackTupleHolder + + // seed is a one-time random value initialized at stack startup + // and is used in calculation of hash key for connection tracking + // table. + Seed uint32 +} + +// parseHeaders sets headers in the packet. +func parseHeaders(pkt *PacketBuffer) { + newPkt := pkt.Clone() + + // Set network header. + hdr, ok := newPkt.Data.PullUp(header.IPv4MinimumSize) + if !ok { + return + } + netHeader := header.IPv4(hdr) + newPkt.NetworkHeader = hdr + length := int(netHeader.HeaderLength()) + + // TODO(gvisor.dev/issue/170): Need to support for other + // protocols as well. + // Set transport header. + switch protocol := netHeader.TransportProtocol(); protocol { + case header.UDPProtocolNumber: + if newPkt.TransportHeader == nil { + h, ok := newPkt.Data.PullUp(length + header.UDPMinimumSize) + if !ok { + return + } + newPkt.TransportHeader = buffer.View(header.UDP(h[length:])) + } + case header.TCPProtocolNumber: + if newPkt.TransportHeader == nil { + h, ok := newPkt.Data.PullUp(length + header.TCPMinimumSize) + if !ok { + return + } + newPkt.TransportHeader = buffer.View(header.TCP(h[length:])) + } + } + pkt.NetworkHeader = newPkt.NetworkHeader + pkt.TransportHeader = newPkt.TransportHeader +} + +// packetToTuple converts packet to a tuple in original direction. +func packetToTuple(pkt PacketBuffer, hook Hook) (connTrackTuple, *tcpip.Error) { + var tuple connTrackTuple + + netHeader := header.IPv4(pkt.NetworkHeader) + // TODO(gvisor.dev/issue/170): Need to support for other + // protocols as well. + if netHeader == nil || netHeader.TransportProtocol() != header.TCPProtocolNumber { + return tuple, tcpip.ErrUnknownProtocol + } + tcpHeader := header.TCP(pkt.TransportHeader) + if tcpHeader == nil { + return tuple, tcpip.ErrUnknownProtocol + } + + tuple.src.addr = netHeader.SourceAddress() + tuple.src.port = tcpHeader.SourcePort() + tuple.src.protocol = header.IPv4ProtocolNumber + + tuple.dst.addr = netHeader.DestinationAddress() + tuple.dst.port = tcpHeader.DestinationPort() + tuple.dst.protocol = netHeader.TransportProtocol() + + return tuple, nil +} + +// getReplyTuple creates reply tuple for the given tuple. +func getReplyTuple(tuple connTrackTuple) connTrackTuple { + var replyTuple connTrackTuple + replyTuple.src.addr = tuple.dst.addr + replyTuple.src.port = tuple.dst.port + replyTuple.src.protocol = tuple.src.protocol + replyTuple.dst.addr = tuple.src.addr + replyTuple.dst.port = tuple.src.port + replyTuple.dst.protocol = tuple.dst.protocol + replyTuple.dst.direction = dirReply + + return replyTuple +} + +// makeNewConn creates new connection. +func makeNewConn(tuple, replyTuple connTrackTuple) connTrack { + var conn connTrack + conn.status = connNew + conn.originalTupleHolder.tuple = tuple + conn.originalTupleHolder.conn = &conn + conn.replyTupleHolder.tuple = replyTuple + conn.replyTupleHolder.conn = &conn + + return conn +} + +// getTupleHash returns hash of the tuple. The fields used for +// generating hash are seed (generated once for stack), source address, +// destination address, source port and destination ports. +func (ct *ConnTrackTable) getTupleHash(tuple connTrackTuple) uint32 { + h := jenkins.Sum32(ct.Seed) + h.Write([]byte(tuple.src.addr)) + h.Write([]byte(tuple.dst.addr)) + portBuf := make([]byte, 2) + binary.LittleEndian.PutUint16(portBuf, tuple.src.port) + h.Write([]byte(portBuf)) + binary.LittleEndian.PutUint16(portBuf, tuple.dst.port) + h.Write([]byte(portBuf)) + + return h.Sum32() +} + +// connTrackForPacket returns connTrack for packet. +// TODO(gvisor.dev/issue/170): Only TCP packets are supported. Need to support other +// transport protocols. +func (ct *ConnTrackTable) connTrackForPacket(pkt *PacketBuffer, hook Hook, createConn bool) (*connTrack, ctDirection) { + if hook == Prerouting { + // Headers will not be set in Prerouting. + // TODO(gvisor.dev/issue/170): Change this after parsing headers + // code is added. + parseHeaders(pkt) + } + + var dir ctDirection + tuple, err := packetToTuple(*pkt, hook) + if err != nil { + return nil, dir + } + + ct.connMu.Lock() + defer ct.connMu.Unlock() + + connTrackTable := ct.CtMap + hash := ct.getTupleHash(tuple) + + var conn *connTrack + switch createConn { + case true: + // If connection does not exist for the hash, create a new + // connection. + replyTuple := getReplyTuple(tuple) + replyHash := ct.getTupleHash(replyTuple) + newConn := makeNewConn(tuple, replyTuple) + conn = &newConn + + // Add tupleHolders to the map. + // TODO(gvisor.dev/issue/170): Need to support collisions using linked list. + ct.CtMap[hash] = conn.originalTupleHolder + ct.CtMap[replyHash] = conn.replyTupleHolder + default: + tupleHolder, ok := connTrackTable[hash] + if !ok { + return nil, dir + } + + // If this is the reply of new connection, set the connection + // status as ESTABLISHED. + conn = tupleHolder.conn + if conn.status == connNew && tupleHolder.tuple.dst.direction == dirReply { + conn.status = connEstablished + } + if tupleHolder.conn == nil { + panic("tupleHolder has null connection tracking entry") + } + + dir = tupleHolder.tuple.dst.direction + } + return conn, dir +} + +// SetNatInfo will manipulate the tuples according to iptables NAT rules. +func (ct *ConnTrackTable) SetNatInfo(pkt *PacketBuffer, rt RedirectTarget, hook Hook) { + // Get the connection. Connection is always created before this + // function is called. + conn, _ := ct.connTrackForPacket(pkt, hook, false) + if conn == nil { + panic("connection should be created to manipulate tuples.") + } + replyTuple := conn.replyTupleHolder.tuple + replyHash := ct.getTupleHash(replyTuple) + + // TODO(gvisor.dev/issue/170): Support only redirect of ports. Need to + // support changing of address for Prerouting. + + // Change the port as per the iptables rule. This tuple will be used + // to manipulate the packet in HandlePacket. + conn.replyTupleHolder.tuple.src.addr = rt.MinIP + conn.replyTupleHolder.tuple.src.port = rt.MinPort + newHash := ct.getTupleHash(conn.replyTupleHolder.tuple) + + // Add the changed tuple to the map. + ct.connMu.Lock() + defer ct.connMu.Unlock() + ct.CtMap[newHash] = conn.replyTupleHolder + if hook == Output { + conn.replyTupleHolder.conn.manip = manipDstOutput + } + + // Delete the old tuple. + delete(ct.CtMap, replyHash) +} + +// handlePacketPrerouting manipulates ports for packets in Prerouting hook. +// TODO(gvisor.dev/issue/170): Change address for Prerouting hook.. +func handlePacketPrerouting(pkt *PacketBuffer, conn *connTrack, dir ctDirection) { + netHeader := header.IPv4(pkt.NetworkHeader) + tcpHeader := header.TCP(pkt.TransportHeader) + + // For prerouting redirection, packets going in the original direction + // have their destinations modified and replies have their sources + // modified. + switch dir { + case dirOriginal: + port := conn.replyTupleHolder.tuple.src.port + tcpHeader.SetDestinationPort(port) + netHeader.SetDestinationAddress(conn.replyTupleHolder.tuple.src.addr) + case dirReply: + port := conn.originalTupleHolder.tuple.dst.port + tcpHeader.SetSourcePort(port) + netHeader.SetSourceAddress(conn.originalTupleHolder.tuple.dst.addr) + } + + netHeader.SetChecksum(0) + netHeader.SetChecksum(^netHeader.CalculateChecksum()) +} + +// handlePacketOutput manipulates ports for packets in Output hook. +func handlePacketOutput(pkt *PacketBuffer, conn *connTrack, gso *GSO, r *Route, dir ctDirection) { + netHeader := header.IPv4(pkt.NetworkHeader) + tcpHeader := header.TCP(pkt.TransportHeader) + + // For output redirection, packets going in the original direction + // have their destinations modified and replies have their sources + // modified. For prerouting redirection, we only reach this point + // when replying, so packet sources are modified. + if conn.manip == manipDstOutput && dir == dirOriginal { + port := conn.replyTupleHolder.tuple.src.port + tcpHeader.SetDestinationPort(port) + netHeader.SetDestinationAddress(conn.replyTupleHolder.tuple.src.addr) + } else { + port := conn.originalTupleHolder.tuple.dst.port + tcpHeader.SetSourcePort(port) + netHeader.SetSourceAddress(conn.originalTupleHolder.tuple.dst.addr) + } + + // Calculate the TCP checksum and set it. + tcpHeader.SetChecksum(0) + hdr := &pkt.Header + length := uint16(pkt.Data.Size()+hdr.UsedLength()) - uint16(netHeader.HeaderLength()) + xsum := r.PseudoHeaderChecksum(header.TCPProtocolNumber, length) + if gso != nil && gso.NeedsCsum { + tcpHeader.SetChecksum(xsum) + } else if r.Capabilities()&CapabilityTXChecksumOffload == 0 { + xsum = header.ChecksumVVWithOffset(pkt.Data, xsum, int(tcpHeader.DataOffset()), pkt.Data.Size()) + tcpHeader.SetChecksum(^tcpHeader.CalculateChecksum(xsum)) + } + + netHeader.SetChecksum(0) + netHeader.SetChecksum(^netHeader.CalculateChecksum()) +} + +// HandlePacket will manipulate the port and address of the packet if the +// connection exists. +func (ct *ConnTrackTable) HandlePacket(pkt *PacketBuffer, hook Hook, gso *GSO, r *Route) { + if pkt.NatDone { + return + } + + if hook != Prerouting && hook != Output { + return + } + + conn, dir := ct.connTrackForPacket(pkt, hook, false) + // Connection or Rule not found for the packet. + if conn == nil { + return + } + + netHeader := header.IPv4(pkt.NetworkHeader) + // TODO(gvisor.dev/issue/170): Need to support for other transport + // protocols as well. + if netHeader == nil || netHeader.TransportProtocol() != header.TCPProtocolNumber { + return + } + + tcpHeader := header.TCP(pkt.TransportHeader) + if tcpHeader == nil { + return + } + + switch hook { + case Prerouting: + handlePacketPrerouting(pkt, conn, dir) + case Output: + handlePacketOutput(pkt, conn, gso, r, dir) + } + pkt.NatDone = true + + // Update the state of tcb. + // TODO(gvisor.dev/issue/170): Add support in tcpcontrack to handle + // other tcp states. + var st tcpconntrack.Result + if conn.tcb.IsEmpty() { + conn.tcb.Init(tcpHeader) + conn.tcbHook = hook + } else { + switch hook { + case conn.tcbHook: + st = conn.tcb.UpdateStateOutbound(tcpHeader) + default: + st = conn.tcb.UpdateStateInbound(tcpHeader) + } + } + + // Delete conntrack if tcp connection is closed. + if st == tcpconntrack.ResultClosedByPeer || st == tcpconntrack.ResultClosedBySelf || st == tcpconntrack.ResultReset { + ct.deleteConnTrack(conn) + } +} + +// deleteConnTrack deletes the connection. +func (ct *ConnTrackTable) deleteConnTrack(conn *connTrack) { + if conn == nil { + return + } + + tuple := conn.originalTupleHolder.tuple + hash := ct.getTupleHash(tuple) + replyTuple := conn.replyTupleHolder.tuple + replyHash := ct.getTupleHash(replyTuple) + + ct.connMu.Lock() + defer ct.connMu.Unlock() + + delete(ct.CtMap, hash) + delete(ct.CtMap, replyHash) +} diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go index 6b91159d4..7c3c47d50 100644 --- a/pkg/tcpip/stack/iptables.go +++ b/pkg/tcpip/stack/iptables.go @@ -17,6 +17,7 @@ package stack import ( "fmt" + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/header" ) @@ -110,6 +111,10 @@ func DefaultTables() IPTables { Prerouting: []string{TablenameMangle, TablenameNat}, Output: []string{TablenameMangle, TablenameNat, TablenameFilter}, }, + connections: ConnTrackTable{ + CtMap: make(map[uint32]ConnTrackTupleHolder), + Seed: generateRandUint32(), + }, } } @@ -173,12 +178,16 @@ const ( // dropped. // // Precondition: pkt.NetworkHeader is set. -func (it *IPTables) Check(hook Hook, pkt PacketBuffer) bool { +func (it *IPTables) Check(hook Hook, pkt *PacketBuffer, gso *GSO, r *Route, address tcpip.Address) bool { + // Packets are manipulated only if connection and matching + // NAT rule exists. + it.connections.HandlePacket(pkt, hook, gso, r) + // Go through each table containing the hook. for _, tablename := range it.Priorities[hook] { table := it.Tables[tablename] ruleIdx := table.BuiltinChains[hook] - switch verdict := it.checkChain(hook, pkt, table, ruleIdx); verdict { + switch verdict := it.checkChain(hook, pkt, table, ruleIdx, gso, r, address); verdict { // If the table returns Accept, move on to the next table. case chainAccept: continue @@ -189,7 +198,7 @@ func (it *IPTables) Check(hook Hook, pkt PacketBuffer) bool { // Any Return from a built-in chain means we have to // call the underflow. underflow := table.Rules[table.Underflows[hook]] - switch v, _ := underflow.Target.Action(pkt); v { + switch v, _ := underflow.Target.Action(pkt, &it.connections, hook, gso, r, address); v { case RuleAccept: continue case RuleDrop: @@ -219,26 +228,34 @@ func (it *IPTables) Check(hook Hook, pkt PacketBuffer) bool { // // NOTE: unlike the Check API the returned map contains packets that should be // dropped. -func (it *IPTables) CheckPackets(hook Hook, pkts PacketBufferList) (drop map[*PacketBuffer]struct{}) { +func (it *IPTables) CheckPackets(hook Hook, pkts PacketBufferList, gso *GSO, r *Route) (drop map[*PacketBuffer]struct{}, natPkts map[*PacketBuffer]struct{}) { for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() { - if ok := it.Check(hook, *pkt); !ok { - if drop == nil { - drop = make(map[*PacketBuffer]struct{}) + if !pkt.NatDone { + if ok := it.Check(hook, pkt, gso, r, ""); !ok { + if drop == nil { + drop = make(map[*PacketBuffer]struct{}) + } + drop[pkt] = struct{}{} + } + if pkt.NatDone { + if natPkts == nil { + natPkts = make(map[*PacketBuffer]struct{}) + } + natPkts[pkt] = struct{}{} } - drop[pkt] = struct{}{} } } - return drop + return drop, natPkts } // Precondition: pkt is a IPv4 packet of at least length header.IPv4MinimumSize. -// TODO(gvisor.dev/issue/170): pk.NetworkHeader will always be set as a +// TODO(gvisor.dev/issue/170): pkt.NetworkHeader will always be set as a // precondition. -func (it *IPTables) checkChain(hook Hook, pkt PacketBuffer, table Table, ruleIdx int) chainVerdict { +func (it *IPTables) checkChain(hook Hook, pkt *PacketBuffer, table Table, ruleIdx int, gso *GSO, r *Route, address tcpip.Address) chainVerdict { // Start from ruleIdx and walk the list of rules until a rule gives us // a verdict. for ruleIdx < len(table.Rules) { - switch verdict, jumpTo := it.checkRule(hook, pkt, table, ruleIdx); verdict { + switch verdict, jumpTo := it.checkRule(hook, pkt, table, ruleIdx, gso, r, address); verdict { case RuleAccept: return chainAccept @@ -255,7 +272,7 @@ func (it *IPTables) checkChain(hook Hook, pkt PacketBuffer, table Table, ruleIdx ruleIdx++ continue } - switch verdict := it.checkChain(hook, pkt, table, jumpTo); verdict { + switch verdict := it.checkChain(hook, pkt, table, jumpTo, gso, r, address); verdict { case chainAccept: return chainAccept case chainDrop: @@ -279,9 +296,9 @@ func (it *IPTables) checkChain(hook Hook, pkt PacketBuffer, table Table, ruleIdx } // Precondition: pkt is a IPv4 packet of at least length header.IPv4MinimumSize. -// TODO(gvisor.dev/issue/170): pk.NetworkHeader will always be set as a +// TODO(gvisor.dev/issue/170): pkt.NetworkHeader will always be set as a // precondition. -func (it *IPTables) checkRule(hook Hook, pkt PacketBuffer, table Table, ruleIdx int) (RuleVerdict, int) { +func (it *IPTables) checkRule(hook Hook, pkt *PacketBuffer, table Table, ruleIdx int, gso *GSO, r *Route, address tcpip.Address) (RuleVerdict, int) { rule := table.Rules[ruleIdx] // If pkt.NetworkHeader hasn't been set yet, it will be contained in @@ -304,7 +321,7 @@ func (it *IPTables) checkRule(hook Hook, pkt PacketBuffer, table Table, ruleIdx // Go through each rule matcher. If they all match, run // the rule target. for _, matcher := range rule.Matchers { - matches, hotdrop := matcher.Match(hook, pkt, "") + matches, hotdrop := matcher.Match(hook, *pkt, "") if hotdrop { return RuleDrop, 0 } @@ -315,7 +332,7 @@ func (it *IPTables) checkRule(hook Hook, pkt PacketBuffer, table Table, ruleIdx } // All the matchers matched, so run the target. - return rule.Target.Action(pkt) + return rule.Target.Action(pkt, &it.connections, hook, gso, r, address) } func filterMatch(filter IPHeaderFilter, hdr header.IPv4) bool { diff --git a/pkg/tcpip/stack/iptables_targets.go b/pkg/tcpip/stack/iptables_targets.go index 8be61f4b1..36cc6275d 100644 --- a/pkg/tcpip/stack/iptables_targets.go +++ b/pkg/tcpip/stack/iptables_targets.go @@ -24,7 +24,7 @@ import ( type AcceptTarget struct{} // Action implements Target.Action. -func (AcceptTarget) Action(packet PacketBuffer) (RuleVerdict, int) { +func (AcceptTarget) Action(*PacketBuffer, *ConnTrackTable, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { return RuleAccept, 0 } @@ -32,7 +32,7 @@ func (AcceptTarget) Action(packet PacketBuffer) (RuleVerdict, int) { type DropTarget struct{} // Action implements Target.Action. -func (DropTarget) Action(packet PacketBuffer) (RuleVerdict, int) { +func (DropTarget) Action(*PacketBuffer, *ConnTrackTable, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { return RuleDrop, 0 } @@ -41,7 +41,7 @@ func (DropTarget) Action(packet PacketBuffer) (RuleVerdict, int) { type ErrorTarget struct{} // Action implements Target.Action. -func (ErrorTarget) Action(packet PacketBuffer) (RuleVerdict, int) { +func (ErrorTarget) Action(*PacketBuffer, *ConnTrackTable, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { log.Debugf("ErrorTarget triggered.") return RuleDrop, 0 } @@ -52,7 +52,7 @@ type UserChainTarget struct { } // Action implements Target.Action. -func (UserChainTarget) Action(PacketBuffer) (RuleVerdict, int) { +func (UserChainTarget) Action(*PacketBuffer, *ConnTrackTable, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { panic("UserChainTarget should never be called.") } @@ -61,7 +61,7 @@ func (UserChainTarget) Action(PacketBuffer) (RuleVerdict, int) { type ReturnTarget struct{} // Action implements Target.Action. -func (ReturnTarget) Action(PacketBuffer) (RuleVerdict, int) { +func (ReturnTarget) Action(*PacketBuffer, *ConnTrackTable, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { return RuleReturn, 0 } @@ -75,16 +75,16 @@ type RedirectTarget struct { // redirect. RangeProtoSpecified bool - // Min address used to redirect. + // MinIP indicates address used to redirect. MinIP tcpip.Address - // Max address used to redirect. + // MaxIP indicates address used to redirect. MaxIP tcpip.Address - // Min port used to redirect. + // MinPort indicates port used to redirect. MinPort uint16 - // Max port used to redirect. + // MaxPort indicates port used to redirect. MaxPort uint16 } @@ -92,61 +92,76 @@ type RedirectTarget struct { // TODO(gvisor.dev/issue/170): Parse headers without copying. The current // implementation only works for PREROUTING and calls pkt.Clone(), neither // of which should be the case. -func (rt RedirectTarget) Action(pkt PacketBuffer) (RuleVerdict, int) { - newPkt := pkt.Clone() +func (rt RedirectTarget) Action(pkt *PacketBuffer, ct *ConnTrackTable, hook Hook, gso *GSO, r *Route, address tcpip.Address) (RuleVerdict, int) { + // Packet is already manipulated. + if pkt.NatDone { + return RuleAccept, 0 + } // Set network header. - headerView, ok := newPkt.Data.PullUp(header.IPv4MinimumSize) - if !ok { - return RuleDrop, 0 + if hook == Prerouting { + parseHeaders(pkt) } - netHeader := header.IPv4(headerView) - newPkt.NetworkHeader = headerView - hlen := int(netHeader.HeaderLength()) - tlen := int(netHeader.TotalLength()) - newPkt.Data.TrimFront(hlen) - newPkt.Data.CapLength(tlen - hlen) + // Drop the packet if network and transport header are not set. + if pkt.NetworkHeader == nil || pkt.TransportHeader == nil { + return RuleDrop, 0 + } - // TODO(gvisor.dev/issue/170): Change destination address to - // loopback or interface address on which the packet was - // received. + // Change the address to localhost (127.0.0.1) in Output and + // to primary address of the incoming interface in Prerouting. + switch hook { + case Output: + rt.MinIP = tcpip.Address([]byte{127, 0, 0, 1}) + rt.MaxIP = tcpip.Address([]byte{127, 0, 0, 1}) + case Prerouting: + rt.MinIP = address + rt.MaxIP = address + default: + panic("redirect target is supported only on output and prerouting hooks") + } // TODO(gvisor.dev/issue/170): Check Flags in RedirectTarget if // we need to change dest address (for OUTPUT chain) or ports. + netHeader := header.IPv4(pkt.NetworkHeader) switch protocol := netHeader.TransportProtocol(); protocol { case header.UDPProtocolNumber: - var udpHeader header.UDP - if newPkt.TransportHeader != nil { - udpHeader = header.UDP(newPkt.TransportHeader) - } else { - if pkt.Data.Size() < header.UDPMinimumSize { - return RuleDrop, 0 - } - hdr, ok := newPkt.Data.PullUp(header.UDPMinimumSize) - if !ok { - return RuleDrop, 0 + udpHeader := header.UDP(pkt.TransportHeader) + udpHeader.SetDestinationPort(rt.MinPort) + + // Calculate UDP checksum and set it. + if hook == Output { + udpHeader.SetChecksum(0) + hdr := &pkt.Header + length := uint16(pkt.Data.Size()+hdr.UsedLength()) - uint16(netHeader.HeaderLength()) + + // Only calculate the checksum if offloading isn't supported. + if r.Capabilities()&CapabilityTXChecksumOffload == 0 { + xsum := r.PseudoHeaderChecksum(protocol, length) + for _, v := range pkt.Data.Views() { + xsum = header.Checksum(v, xsum) + } + udpHeader.SetChecksum(0) + udpHeader.SetChecksum(^udpHeader.CalculateChecksum(xsum)) } - udpHeader = header.UDP(hdr) } - udpHeader.SetDestinationPort(rt.MinPort) + // Change destination address. + netHeader.SetDestinationAddress(rt.MinIP) + netHeader.SetChecksum(0) + netHeader.SetChecksum(^netHeader.CalculateChecksum()) + pkt.NatDone = true case header.TCPProtocolNumber: - var tcpHeader header.TCP - if newPkt.TransportHeader != nil { - tcpHeader = header.TCP(newPkt.TransportHeader) - } else { - if pkt.Data.Size() < header.TCPMinimumSize { - return RuleDrop, 0 - } - hdr, ok := newPkt.Data.PullUp(header.TCPMinimumSize) - if !ok { - return RuleDrop, 0 - } - tcpHeader = header.TCP(hdr) + if ct == nil { + return RuleAccept, 0 + } + + // Set up conection for matching NAT rule. + // Only the first packet of the connection comes here. + // Other packets will be manipulated in connection tracking. + if conn, _ := ct.connTrackForPacket(pkt, hook, true); conn != nil { + ct.SetNatInfo(pkt, rt, hook) + ct.HandlePacket(pkt, hook, gso, r) } - // TODO(gvisor.dev/issue/170): Need to recompute checksum - // and implement nat connection tracking to support TCP. - tcpHeader.SetDestinationPort(rt.MinPort) default: return RuleDrop, 0 } diff --git a/pkg/tcpip/stack/iptables_types.go b/pkg/tcpip/stack/iptables_types.go index 2ffb55f2a..1bb0ba1bd 100644 --- a/pkg/tcpip/stack/iptables_types.go +++ b/pkg/tcpip/stack/iptables_types.go @@ -82,6 +82,8 @@ type IPTables struct { // list is the order in which each table should be visited for that // hook. Priorities map[Hook][]string + + connections ConnTrackTable } // A Table defines a set of chains and hooks into the network stack. It is @@ -176,5 +178,5 @@ type Target interface { // Action takes an action on the packet and returns a verdict on how // traversal should (or should not) continue. If the return value is // Jump, it also returns the index of the rule to jump to. - Action(packet PacketBuffer) (RuleVerdict, int) + Action(packet *PacketBuffer, connections *ConnTrackTable, hook Hook, gso *GSO, r *Route, address tcpip.Address) (RuleVerdict, int) } diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 7b54919bb..8f4c1fe42 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -1230,8 +1230,10 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link // TODO(gvisor.dev/issue/170): Not supporting iptables for IPv6 yet. if protocol == header.IPv4ProtocolNumber { + // iptables filtering. ipt := n.stack.IPTables() - if ok := ipt.Check(Prerouting, pkt); !ok { + address := n.primaryAddress(protocol) + if ok := ipt.Check(Prerouting, &pkt, nil, nil, address.Address); !ok { // iptables is telling us to drop the packet. return } diff --git a/pkg/tcpip/stack/packet_buffer.go b/pkg/tcpip/stack/packet_buffer.go index 9ff80ab24..926df4d7b 100644 --- a/pkg/tcpip/stack/packet_buffer.go +++ b/pkg/tcpip/stack/packet_buffer.go @@ -72,6 +72,10 @@ type PacketBuffer struct { EgressRoute *Route GSOOptions *GSO NetworkProtocolNumber tcpip.NetworkProtocolNumber + + // NatDone indicates if the packet has been manipulated as per NAT + // iptables rule. + NatDone bool } // Clone makes a copy of pk. It clones the Data field, which creates a new diff --git a/pkg/tcpip/stack/route.go b/pkg/tcpip/stack/route.go index 53148dc03..150297ab9 100644 --- a/pkg/tcpip/stack/route.go +++ b/pkg/tcpip/stack/route.go @@ -261,3 +261,16 @@ func (r *Route) MakeLoopedRoute() Route { func (r *Route) Stack() *Stack { return r.ref.stack() } + +// ReverseRoute returns new route with given source and destination address. +func (r *Route) ReverseRoute(src tcpip.Address, dst tcpip.Address) Route { + return Route{ + NetProto: r.NetProto, + LocalAddress: dst, + LocalLinkAddress: r.RemoteLinkAddress, + RemoteAddress: src, + RemoteLinkAddress: r.LocalLinkAddress, + ref: r.ref, + Loop: r.Loop, + } +} diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index 4a2dc3dc6..e33fae4eb 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -1885,3 +1885,22 @@ func generateRandInt64() int64 { } return v } + +// FindNetworkEndpoint returns the network endpoint for the given address. +func (s *Stack) FindNetworkEndpoint(netProto tcpip.NetworkProtocolNumber, address tcpip.Address) (NetworkEndpoint, *tcpip.Error) { + s.mu.Lock() + defer s.mu.Unlock() + + for _, nic := range s.nics { + id := NetworkEndpointID{address} + + if ref, ok := nic.mu.endpoints[id]; ok { + nic.mu.RLock() + defer nic.mu.RUnlock() + + // An endpoint with this id exists, check if it can be used and return it. + return ref.ep, nil + } + } + return nil, tcpip.ErrBadAddress +} diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD index f2aa69069..f38eb6833 100644 --- a/pkg/tcpip/transport/tcp/BUILD +++ b/pkg/tcpip/transport/tcp/BUILD @@ -115,7 +115,7 @@ go_test( size = "small", srcs = ["rcv_test.go"], deps = [ - ":tcp", + "//pkg/tcpip/header", "//pkg/tcpip/seqnum", ], ) diff --git a/pkg/tcpip/transport/tcp/rcv.go b/pkg/tcpip/transport/tcp/rcv.go index a4b73b588..6fe97fefd 100644 --- a/pkg/tcpip/transport/tcp/rcv.go +++ b/pkg/tcpip/transport/tcp/rcv.go @@ -70,24 +70,7 @@ func newReceiver(ep *endpoint, irs seqnum.Value, rcvWnd seqnum.Size, rcvWndScale // acceptable checks if the segment sequence number range is acceptable // according to the table on page 26 of RFC 793. func (r *receiver) acceptable(segSeq seqnum.Value, segLen seqnum.Size) bool { - return Acceptable(segSeq, segLen, r.rcvNxt, r.rcvAcc) -} - -// Acceptable checks if a segment that starts at segSeq and has length segLen is -// "acceptable" for arriving in a receive window that starts at rcvNxt and ends -// before rcvAcc, according to the table on page 26 and 69 of RFC 793. -func Acceptable(segSeq seqnum.Value, segLen seqnum.Size, rcvNxt, rcvAcc seqnum.Value) bool { - if rcvNxt == rcvAcc { - return segLen == 0 && segSeq == rcvNxt - } - if segLen == 0 { - // rcvWnd is incremented by 1 because that is Linux's behavior despite the - // RFC. - return segSeq.InRange(rcvNxt, rcvAcc.Add(1)) - } - // Page 70 of RFC 793 allows packets that can be made "acceptable" by trimming - // the payload, so we'll accept any payload that overlaps the receieve window. - return rcvNxt.LessThan(segSeq.Add(segLen)) && segSeq.LessThan(rcvAcc) + return header.Acceptable(segSeq, segLen, r.rcvNxt, r.rcvAcc) } // getSendParams returns the parameters needed by the sender when building diff --git a/pkg/tcpip/transport/tcp/rcv_test.go b/pkg/tcpip/transport/tcp/rcv_test.go index dc02729ce..c9eeff935 100644 --- a/pkg/tcpip/transport/tcp/rcv_test.go +++ b/pkg/tcpip/transport/tcp/rcv_test.go @@ -17,8 +17,8 @@ package rcv_test import ( "testing" + "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/seqnum" - "gvisor.dev/gvisor/pkg/tcpip/transport/tcp" ) func TestAcceptable(t *testing.T) { @@ -67,8 +67,8 @@ func TestAcceptable(t *testing.T) { {105, 2, 108, 108, false}, {105, 2, 109, 109, false}, } { - if got := tcp.Acceptable(tt.segSeq, tt.segLen, tt.rcvNxt, tt.rcvAcc); got != tt.want { - t.Errorf("tcp.Acceptable(%d, %d, %d, %d) = %t, want %t", tt.segSeq, tt.segLen, tt.rcvNxt, tt.rcvAcc, got, tt.want) + if got := header.Acceptable(tt.segSeq, tt.segLen, tt.rcvNxt, tt.rcvAcc); got != tt.want { + t.Errorf("header.Acceptable(%d, %d, %d, %d) = %t, want %t", tt.segSeq, tt.segLen, tt.rcvNxt, tt.rcvAcc, got, tt.want) } } } diff --git a/pkg/tcpip/transport/tcpconntrack/BUILD b/pkg/tcpip/transport/tcpconntrack/BUILD index 2025ff757..3ad6994a7 100644 --- a/pkg/tcpip/transport/tcpconntrack/BUILD +++ b/pkg/tcpip/transport/tcpconntrack/BUILD @@ -9,7 +9,6 @@ go_library( deps = [ "//pkg/tcpip/header", "//pkg/tcpip/seqnum", - "//pkg/tcpip/transport/tcp", ], ) diff --git a/pkg/tcpip/transport/tcpconntrack/tcp_conntrack.go b/pkg/tcpip/transport/tcpconntrack/tcp_conntrack.go index 30d05200f..12bc1b5b5 100644 --- a/pkg/tcpip/transport/tcpconntrack/tcp_conntrack.go +++ b/pkg/tcpip/transport/tcpconntrack/tcp_conntrack.go @@ -20,7 +20,6 @@ package tcpconntrack import ( "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/seqnum" - "gvisor.dev/gvisor/pkg/tcpip/transport/tcp" ) // Result is returned when the state of a TCB is updated in response to an @@ -312,7 +311,7 @@ type stream struct { // the window is zero, if it's a packet with no payload and sequence number // equal to una. func (s *stream) acceptable(segSeq seqnum.Value, segLen seqnum.Size) bool { - return tcp.Acceptable(segSeq, segLen, s.una, s.end) + return header.Acceptable(segSeq, segLen, s.una, s.end) } // closed determines if the stream has already been closed. This happens when @@ -338,3 +337,16 @@ func logicalLen(tcp header.TCP) seqnum.Size { } return l } + +// IsEmpty returns true if tcb is not initialized. +func (t *TCB) IsEmpty() bool { + if t.inbound != (stream{}) || t.outbound != (stream{}) { + return false + } + + if t.firstFin != nil || t.state != ResultDrop { + return false + } + + return true +} diff --git a/test/iptables/filter_output.go b/test/iptables/filter_output.go index f6d974b85..b1382d353 100644 --- a/test/iptables/filter_output.go +++ b/test/iptables/filter_output.go @@ -42,7 +42,7 @@ func (FilterOutputDropTCPDestPort) Name() string { // ContainerAction implements TestCase.ContainerAction. func (FilterOutputDropTCPDestPort) ContainerAction(ip net.IP) error { - if err := filterTable("-A", "OUTPUT", "-p", "tcp", "-m", "tcp", "--dport", fmt.Sprintf("%d", dropPort), "-j", "DROP"); err != nil { + if err := filterTable("-A", "OUTPUT", "-p", "tcp", "-m", "tcp", "--dport", "1024:65535", "-j", "DROP"); err != nil { return err } diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go index 334d8e676..63a862d35 100644 --- a/test/iptables/iptables_test.go +++ b/test/iptables/iptables_test.go @@ -115,30 +115,6 @@ func TestFilterInputDropOnlyUDP(t *testing.T) { singleTest(t, FilterInputDropOnlyUDP{}) } -func TestNATRedirectUDPPort(t *testing.T) { - // TODO(gvisor.dev/issue/170): Enable when supported. - t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - singleTest(t, NATRedirectUDPPort{}) -} - -func TestNATRedirectTCPPort(t *testing.T) { - // TODO(gvisor.dev/issue/170): Enable when supported. - t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - singleTest(t, NATRedirectTCPPort{}) -} - -func TestNATDropUDP(t *testing.T) { - // TODO(gvisor.dev/issue/170): Enable when supported. - t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - singleTest(t, NATDropUDP{}) -} - -func TestNATAcceptAll(t *testing.T) { - // TODO(gvisor.dev/issue/170): Enable when supported. - t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") - singleTest(t, NATAcceptAll{}) -} - func TestFilterInputDropTCPDestPort(t *testing.T) { singleTest(t, FilterInputDropTCPDestPort{}) } @@ -164,14 +140,10 @@ func TestFilterInputReturnUnderflow(t *testing.T) { } func TestFilterOutputDropTCPDestPort(t *testing.T) { - // TODO(gvisor.dev/issue/170): Enable when supported. - t.Skip("filter OUTPUT isn't supported yet (gvisor.dev/issue/170).") singleTest(t, FilterOutputDropTCPDestPort{}) } func TestFilterOutputDropTCPSrcPort(t *testing.T) { - // TODO(gvisor.dev/issue/170): Enable when supported. - t.Skip("filter OUTPUT isn't supported yet (gvisor.dev/issue/170).") singleTest(t, FilterOutputDropTCPSrcPort{}) } @@ -235,44 +207,54 @@ func TestOutputInvertDestination(t *testing.T) { singleTest(t, FilterOutputInvertDestination{}) } +func TestNATPreRedirectUDPPort(t *testing.T) { + singleTest(t, NATPreRedirectUDPPort{}) +} + +func TestNATPreRedirectTCPPort(t *testing.T) { + singleTest(t, NATPreRedirectTCPPort{}) +} + +func TestNATOutRedirectUDPPort(t *testing.T) { + singleTest(t, NATOutRedirectUDPPort{}) +} + +func TestNATOutRedirectTCPPort(t *testing.T) { + singleTest(t, NATOutRedirectTCPPort{}) +} + +func TestNATDropUDP(t *testing.T) { + singleTest(t, NATDropUDP{}) +} + +func TestNATAcceptAll(t *testing.T) { + singleTest(t, NATAcceptAll{}) +} + func TestNATOutRedirectIP(t *testing.T) { - // TODO(gvisor.dev/issue/170): Enable when supported. - t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") singleTest(t, NATOutRedirectIP{}) } func TestNATOutDontRedirectIP(t *testing.T) { - // TODO(gvisor.dev/issue/170): Enable when supported. - t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") singleTest(t, NATOutDontRedirectIP{}) } func TestNATOutRedirectInvert(t *testing.T) { - // TODO(gvisor.dev/issue/170): Enable when supported. - t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") singleTest(t, NATOutRedirectInvert{}) } func TestNATPreRedirectIP(t *testing.T) { - // TODO(gvisor.dev/issue/170): Enable when supported. - t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") singleTest(t, NATPreRedirectIP{}) } func TestNATPreDontRedirectIP(t *testing.T) { - // TODO(gvisor.dev/issue/170): Enable when supported. - t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") singleTest(t, NATPreDontRedirectIP{}) } func TestNATPreRedirectInvert(t *testing.T) { - // TODO(gvisor.dev/issue/170): Enable when supported. - t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") singleTest(t, NATPreRedirectInvert{}) } func TestNATRedirectRequiresProtocol(t *testing.T) { - // TODO(gvisor.dev/issue/170): Enable when supported. - t.Skip("NAT isn't supported yet (gvisor.dev/issue/170).") singleTest(t, NATRedirectRequiresProtocol{}) } diff --git a/test/iptables/iptables_util.go b/test/iptables/iptables_util.go index 2a00677be..2f988cd18 100644 --- a/test/iptables/iptables_util.go +++ b/test/iptables/iptables_util.go @@ -151,7 +151,7 @@ func connectTCP(ip net.IP, port int, timeout time.Duration) error { return err } if err := testutil.Poll(callback, timeout); err != nil { - return fmt.Errorf("timed out waiting to connect IP, most recent error: %v", err) + return fmt.Errorf("timed out waiting to connect IP on port %v, most recent error: %v", port, err) } return nil diff --git a/test/iptables/nat.go b/test/iptables/nat.go index 40096901c..0a10ce7fe 100644 --- a/test/iptables/nat.go +++ b/test/iptables/nat.go @@ -26,8 +26,10 @@ const ( ) func init() { - RegisterTestCase(NATRedirectUDPPort{}) - RegisterTestCase(NATRedirectTCPPort{}) + RegisterTestCase(NATPreRedirectUDPPort{}) + RegisterTestCase(NATPreRedirectTCPPort{}) + RegisterTestCase(NATOutRedirectUDPPort{}) + RegisterTestCase(NATOutRedirectTCPPort{}) RegisterTestCase(NATDropUDP{}) RegisterTestCase(NATAcceptAll{}) RegisterTestCase(NATPreRedirectIP{}) @@ -39,16 +41,16 @@ func init() { RegisterTestCase(NATRedirectRequiresProtocol{}) } -// NATRedirectUDPPort tests that packets are redirected to different port. -type NATRedirectUDPPort struct{} +// NATPreRedirectUDPPort tests that packets are redirected to different port. +type NATPreRedirectUDPPort struct{} // Name implements TestCase.Name. -func (NATRedirectUDPPort) Name() string { - return "NATRedirectUDPPort" +func (NATPreRedirectUDPPort) Name() string { + return "NATPreRedirectUDPPort" } // ContainerAction implements TestCase.ContainerAction. -func (NATRedirectUDPPort) ContainerAction(ip net.IP) error { +func (NATPreRedirectUDPPort) ContainerAction(ip net.IP) error { if err := natTable("-A", "PREROUTING", "-p", "udp", "-j", "REDIRECT", "--to-ports", fmt.Sprintf("%d", redirectPort)); err != nil { return err } @@ -61,33 +63,53 @@ func (NATRedirectUDPPort) ContainerAction(ip net.IP) error { } // LocalAction implements TestCase.LocalAction. -func (NATRedirectUDPPort) LocalAction(ip net.IP) error { +func (NATPreRedirectUDPPort) LocalAction(ip net.IP) error { return sendUDPLoop(ip, acceptPort, sendloopDuration) } -// NATRedirectTCPPort tests that connections are redirected on specified ports. -type NATRedirectTCPPort struct{} +// NATPreRedirectTCPPort tests that connections are redirected on specified ports. +type NATPreRedirectTCPPort struct{} // Name implements TestCase.Name. -func (NATRedirectTCPPort) Name() string { - return "NATRedirectTCPPort" +func (NATPreRedirectTCPPort) Name() string { + return "NATPreRedirectTCPPort" } // ContainerAction implements TestCase.ContainerAction. -func (NATRedirectTCPPort) ContainerAction(ip net.IP) error { - if err := natTable("-A", "PREROUTING", "-p", "tcp", "-m", "tcp", "--dport", fmt.Sprintf("%d", dropPort), "-j", "REDIRECT", "--to-ports", fmt.Sprintf("%d", redirectPort)); err != nil { +func (NATPreRedirectTCPPort) ContainerAction(ip net.IP) error { + if err := natTable("-A", "PREROUTING", "-p", "tcp", "-m", "tcp", "--dport", fmt.Sprintf("%d", dropPort), "-j", "REDIRECT", "--to-ports", fmt.Sprintf("%d", acceptPort)); err != nil { return err } // Listen for TCP packets on redirect port. - return listenTCP(redirectPort, sendloopDuration) + return listenTCP(acceptPort, sendloopDuration) } // LocalAction implements TestCase.LocalAction. -func (NATRedirectTCPPort) LocalAction(ip net.IP) error { +func (NATPreRedirectTCPPort) LocalAction(ip net.IP) error { return connectTCP(ip, dropPort, sendloopDuration) } +// NATOutRedirectUDPPort tests that packets are redirected to different port. +type NATOutRedirectUDPPort struct{} + +// Name implements TestCase.Name. +func (NATOutRedirectUDPPort) Name() string { + return "NATOutRedirectUDPPort" +} + +// ContainerAction implements TestCase.ContainerAction. +func (NATOutRedirectUDPPort) ContainerAction(ip net.IP) error { + dest := []byte{200, 0, 0, 1} + return loopbackTest(dest, "-A", "OUTPUT", "-p", "udp", "-j", "REDIRECT", "--to-ports", fmt.Sprintf("%d", acceptPort)) +} + +// LocalAction implements TestCase.LocalAction. +func (NATOutRedirectUDPPort) LocalAction(ip net.IP) error { + // No-op. + return nil +} + // NATDropUDP tests that packets are not received in ports other than redirect // port. type NATDropUDP struct{} @@ -329,3 +351,52 @@ func loopbackTest(dest net.IP, args ...string) error { // sendCh will always take the full sendloop time. return <-sendCh } + +// NATOutRedirectTCPPort tests that connections are redirected on specified ports. +type NATOutRedirectTCPPort struct{} + +// Name implements TestCase.Name. +func (NATOutRedirectTCPPort) Name() string { + return "NATOutRedirectTCPPort" +} + +// ContainerAction implements TestCase.ContainerAction. +func (NATOutRedirectTCPPort) ContainerAction(ip net.IP) error { + if err := natTable("-A", "OUTPUT", "-p", "tcp", "-m", "tcp", "--dport", fmt.Sprintf("%d", dropPort), "-j", "REDIRECT", "--to-ports", fmt.Sprintf("%d", acceptPort)); err != nil { + return err + } + + timeout := 20 * time.Second + dest := []byte{127, 0, 0, 1} + localAddr := net.TCPAddr{ + IP: dest, + Port: acceptPort, + } + + // Starts listening on port. + lConn, err := net.ListenTCP("tcp", &localAddr) + if err != nil { + return err + } + defer lConn.Close() + + // Accept connections on port. + lConn.SetDeadline(time.Now().Add(timeout)) + err = connectTCP(ip, dropPort, timeout) + if err != nil { + return err + } + + conn, err := lConn.AcceptTCP() + if err != nil { + return err + } + conn.Close() + + return nil +} + +// LocalAction implements TestCase.LocalAction. +func (NATOutRedirectTCPPort) LocalAction(ip net.IP) error { + return nil +} -- cgit v1.2.3 From 2c986870e35f967c88ebc1b7df7b576aad2c46d4 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Mon, 4 May 2020 10:39:36 -0700 Subject: Fix flaky monotonic time. This change ensures that even platforms with some TSC issues (e.g. KVM), can get reliable monotonic time by applied a lower bound on each read. PiperOrigin-RevId: 309773801 --- pkg/sentry/kernel/timekeeper.go | 19 +++++++++++++++++++ 1 file changed, 19 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/timekeeper.go b/pkg/sentry/kernel/timekeeper.go index dc99301de..da0ea7bb5 100644 --- a/pkg/sentry/kernel/timekeeper.go +++ b/pkg/sentry/kernel/timekeeper.go @@ -16,6 +16,7 @@ package kernel import ( "fmt" + "sync/atomic" "time" "gvisor.dev/gvisor/pkg/log" @@ -48,6 +49,9 @@ type Timekeeper struct { // It is set only once, by SetClocks. monotonicOffset int64 `state:"nosave"` + // monotonicLowerBound is the lowerBound for monotonic time. + monotonicLowerBound int64 `state:"nosave"` + // restored, if non-nil, indicates that this Timekeeper was restored // from a state file. The clocks are not set until restored is closed. restored chan struct{} `state:"nosave"` @@ -271,6 +275,21 @@ func (t *Timekeeper) GetTime(c sentrytime.ClockID) (int64, error) { now, err := t.clocks.GetTime(c) if err == nil && c == sentrytime.Monotonic { now += t.monotonicOffset + for { + // It's possible that the clock is shaky. This may be due to + // platform issues, e.g. the KVM platform relies on the guest + // TSC and host TSC, which may not be perfectly in sync. To + // work around this issue, ensure that the monotonic time is + // always bounded by the last time read. + oldLowerBound := atomic.LoadInt64(&t.monotonicLowerBound) + if now < oldLowerBound { + now = oldLowerBound + break + } + if atomic.CompareAndSwapInt64(&t.monotonicLowerBound, oldLowerBound, now) { + break + } + } } return now, err } -- cgit v1.2.3 From cbc5bef2a66ece1f9e63b213d4dfa616db488df8 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Mon, 4 May 2020 10:58:01 -0700 Subject: Add TTY support on VFS2 to runsc Updates #1623, #1487 PiperOrigin-RevId: 309777922 --- pkg/sentry/control/BUILD | 3 + pkg/sentry/control/proc.go | 98 +++++++----------- pkg/sentry/fdimport/BUILD | 19 ++++ pkg/sentry/fdimport/fdimport.go | 129 ++++++++++++++++++++++++ pkg/sentry/fsimpl/host/host.go | 2 +- pkg/sentry/fsimpl/host/tty.go | 26 ++--- pkg/sentry/vfs/context.go | 24 +++++ pkg/sentry/vfs/epoll.go | 1 + runsc/boot/BUILD | 3 +- runsc/boot/fds.go | 108 -------------------- runsc/boot/loader.go | 204 +++++++++++++++++++++++++------------- runsc/boot/vfs.go | 42 +++----- runsc/container/container_test.go | 107 ++++++++++---------- 13 files changed, 434 insertions(+), 332 deletions(-) create mode 100644 pkg/sentry/fdimport/BUILD create mode 100644 pkg/sentry/fdimport/fdimport.go delete mode 100644 runsc/boot/fds.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/control/BUILD b/pkg/sentry/control/BUILD index d16d78aa5..e74275d2d 100644 --- a/pkg/sentry/control/BUILD +++ b/pkg/sentry/control/BUILD @@ -20,9 +20,11 @@ go_library( "//pkg/fd", "//pkg/fspath", "//pkg/log", + "//pkg/sentry/fdimport", "//pkg/sentry/fs", "//pkg/sentry/fs/host", "//pkg/sentry/fsbridge", + "//pkg/sentry/fsimpl/host", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/time", @@ -36,6 +38,7 @@ go_library( "//pkg/syserror", "//pkg/tcpip/link/sniffer", "//pkg/urpc", + "@org_golang_x_sys//unix:go_default_library", ], ) diff --git a/pkg/sentry/control/proc.go b/pkg/sentry/control/proc.go index b51fb3959..2ed17ee09 100644 --- a/pkg/sentry/control/proc.go +++ b/pkg/sentry/control/proc.go @@ -24,13 +24,16 @@ import ( "text/tabwriter" "time" + "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/sentry/fdimport" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/host" "gvisor.dev/gvisor/pkg/sentry/fsbridge" + hostvfs2 "gvisor.dev/gvisor/pkg/sentry/fsimpl/host" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" @@ -84,15 +87,13 @@ type ExecArgs struct { // the root group if not set explicitly. KGID auth.KGID - // ExtraKGIDs is the list of additional groups to which the user - // belongs. + // ExtraKGIDs is the list of additional groups to which the user belongs. ExtraKGIDs []auth.KGID // Capabilities is the list of capabilities to give to the process. Capabilities *auth.TaskCapabilities - // StdioIsPty indicates that FDs 0, 1, and 2 are connected to a host - // pty FD. + // StdioIsPty indicates that FDs 0, 1, and 2 are connected to a host pty FD. StdioIsPty bool // FilePayload determines the files to give to the new process. @@ -117,7 +118,7 @@ func (args ExecArgs) String() string { // Exec runs a new task. func (proc *Proc) Exec(args *ExecArgs, waitStatus *uint32) error { - newTG, _, _, err := proc.execAsync(args) + newTG, _, _, _, err := proc.execAsync(args) if err != nil { return err } @@ -130,26 +131,18 @@ func (proc *Proc) Exec(args *ExecArgs, waitStatus *uint32) error { // ExecAsync runs a new task, but doesn't wait for it to finish. It is defined // as a function rather than a method to avoid exposing execAsync as an RPC. -func ExecAsync(proc *Proc, args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadID, *host.TTYFileOperations, error) { +func ExecAsync(proc *Proc, args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadID, *host.TTYFileOperations, *hostvfs2.TTYFileDescription, error) { return proc.execAsync(args) } // execAsync runs a new task, but doesn't wait for it to finish. It returns the // newly created thread group and its PID. If the stdio FDs are TTYs, then a // TTYFileOperations that wraps the TTY is also returned. -func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadID, *host.TTYFileOperations, error) { +func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadID, *host.TTYFileOperations, *hostvfs2.TTYFileDescription, error) { // Import file descriptors. fdTable := proc.Kernel.NewFDTable() defer fdTable.DecRef() - // No matter what happens, we should close all files in the FilePayload - // before returning. Any files that are imported will be duped. - defer func() { - for _, f := range args.FilePayload.Files { - f.Close() - } - }() - creds := auth.NewUserCredentials( args.KUID, args.KGID, @@ -202,7 +195,7 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI vfsObj := proc.Kernel.VFS() file, err := ResolveExecutablePath(ctx, vfsObj, initArgs.WorkingDirectory, initArgs.Argv[0], paths) if err != nil { - return nil, 0, nil, fmt.Errorf("error finding executable %q in PATH %v: %v", initArgs.Argv[0], paths, err) + return nil, 0, nil, nil, fmt.Errorf("error finding executable %q in PATH %v: %v", initArgs.Argv[0], paths, err) } initArgs.File = fsbridge.NewVFSFile(file) } else { @@ -214,74 +207,57 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI // initArgs must hold a reference on MountNamespace, which will // be donated to the new process in CreateProcess. - initArgs.MountNamespaceVFS2.IncRef() + initArgs.MountNamespace.IncRef() } f, err := initArgs.MountNamespace.ResolveExecutablePath(ctx, initArgs.WorkingDirectory, initArgs.Argv[0], paths) if err != nil { - return nil, 0, nil, fmt.Errorf("error finding executable %q in PATH %v: %v", initArgs.Argv[0], paths, err) + return nil, 0, nil, nil, fmt.Errorf("error finding executable %q in PATH %v: %v", initArgs.Argv[0], paths, err) } initArgs.Filename = f } } - // TODO(gvisor.dev/issue/1623): Use host FD when supported in VFS2. - var ttyFile *fs.File - for appFD, hostFile := range args.FilePayload.Files { - var appFile *fs.File - - if args.StdioIsPty && appFD < 3 { - // Import the file as a host TTY file. - if ttyFile == nil { - var err error - appFile, err = host.ImportFile(ctx, int(hostFile.Fd()), true /* isTTY */) - if err != nil { - return nil, 0, nil, err - } - defer appFile.DecRef() - - // Remember this in the TTY file, as we will - // use it for the other stdio FDs. - ttyFile = appFile - } else { - // Re-use the existing TTY file, as all three - // stdio FDs must point to the same fs.File in - // order to share TTY state, specifically the - // foreground process group id. - appFile = ttyFile - } - } else { - // Import the file as a regular host file. - var err error - appFile, err = host.ImportFile(ctx, int(hostFile.Fd()), false /* isTTY */) + fds := make([]int, len(args.FilePayload.Files)) + for i, file := range args.FilePayload.Files { + if kernel.VFS2Enabled { + // Need to dup to remove ownership from os.File. + dup, err := unix.Dup(int(file.Fd())) if err != nil { - return nil, 0, nil, err + return nil, 0, nil, nil, fmt.Errorf("duplicating payload files: %w", err) } - defer appFile.DecRef() + fds[i] = dup + } else { + // VFS1 dups the file on import. + fds[i] = int(file.Fd()) } - - // Add the file to the FD map. - if err := fdTable.NewFDAt(ctx, int32(appFD), appFile, kernel.FDFlags{}); err != nil { - return nil, 0, nil, err + } + ttyFile, ttyFileVFS2, err := fdimport.Import(ctx, fdTable, args.StdioIsPty, fds) + if err != nil { + if kernel.VFS2Enabled { + for _, fd := range fds { + unix.Close(fd) + } } + return nil, 0, nil, nil, err } tg, tid, err := proc.Kernel.CreateProcess(initArgs) if err != nil { - return nil, 0, nil, err + return nil, 0, nil, nil, err } - var ttyFileOps *host.TTYFileOperations - if ttyFile != nil { - // Set the foreground process group on the TTY before starting - // the process. - ttyFileOps = ttyFile.FileOperations.(*host.TTYFileOperations) - ttyFileOps.InitForegroundProcessGroup(tg.ProcessGroup()) + // Set the foreground process group on the TTY before starting the process. + switch { + case ttyFile != nil: + ttyFile.InitForegroundProcessGroup(tg.ProcessGroup()) + case ttyFileVFS2 != nil: + ttyFileVFS2.InitForegroundProcessGroup(tg.ProcessGroup()) } // Start the newly created process. proc.Kernel.StartProcess(tg) - return tg, tid, ttyFileOps, nil + return tg, tid, ttyFile, ttyFileVFS2, nil } // PsArgs is the set of arguments to ps. diff --git a/pkg/sentry/fdimport/BUILD b/pkg/sentry/fdimport/BUILD new file mode 100644 index 000000000..5e41ceb4e --- /dev/null +++ b/pkg/sentry/fdimport/BUILD @@ -0,0 +1,19 @@ +load("//tools:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_library( + name = "fdimport", + srcs = [ + "fdimport.go", + ], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/context", + "//pkg/sentry/fs", + "//pkg/sentry/fs/host", + "//pkg/sentry/fsimpl/host", + "//pkg/sentry/kernel", + "//pkg/sentry/vfs", + ], +) diff --git a/pkg/sentry/fdimport/fdimport.go b/pkg/sentry/fdimport/fdimport.go new file mode 100644 index 000000000..a4199f9e9 --- /dev/null +++ b/pkg/sentry/fdimport/fdimport.go @@ -0,0 +1,129 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package fdimport + +import ( + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/fs/host" + hostvfs2 "gvisor.dev/gvisor/pkg/sentry/fsimpl/host" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/vfs" +) + +// Import imports a slice of FDs into the given FDTable. If console is true, +// sets up TTY for the first 3 FDs in the slice representing stdin, stdout, +// stderr. Upon success, Import takes ownership of all FDs. +func Import(ctx context.Context, fdTable *kernel.FDTable, console bool, fds []int) (*host.TTYFileOperations, *hostvfs2.TTYFileDescription, error) { + if kernel.VFS2Enabled { + ttyFile, err := importVFS2(ctx, fdTable, console, fds) + return nil, ttyFile, err + } + ttyFile, err := importFS(ctx, fdTable, console, fds) + return ttyFile, nil, err +} + +func importFS(ctx context.Context, fdTable *kernel.FDTable, console bool, fds []int) (*host.TTYFileOperations, error) { + var ttyFile *fs.File + for appFD, hostFD := range fds { + var appFile *fs.File + + if console && appFD < 3 { + // Import the file as a host TTY file. + if ttyFile == nil { + var err error + appFile, err = host.ImportFile(ctx, hostFD, true /* isTTY */) + if err != nil { + return nil, err + } + defer appFile.DecRef() + + // Remember this in the TTY file, as we will + // use it for the other stdio FDs. + ttyFile = appFile + } else { + // Re-use the existing TTY file, as all three + // stdio FDs must point to the same fs.File in + // order to share TTY state, specifically the + // foreground process group id. + appFile = ttyFile + } + } else { + // Import the file as a regular host file. + var err error + appFile, err = host.ImportFile(ctx, hostFD, false /* isTTY */) + if err != nil { + return nil, err + } + defer appFile.DecRef() + } + + // Add the file to the FD map. + if err := fdTable.NewFDAt(ctx, int32(appFD), appFile, kernel.FDFlags{}); err != nil { + return nil, err + } + } + + if ttyFile == nil { + return nil, nil + } + return ttyFile.FileOperations.(*host.TTYFileOperations), nil +} + +func importVFS2(ctx context.Context, fdTable *kernel.FDTable, console bool, stdioFDs []int) (*hostvfs2.TTYFileDescription, error) { + k := kernel.KernelFromContext(ctx) + + var ttyFile *vfs.FileDescription + for appFD, hostFD := range stdioFDs { + var appFile *vfs.FileDescription + + if console && appFD < 3 { + // Import the file as a host TTY file. + if ttyFile == nil { + var err error + appFile, err = hostvfs2.ImportFD(ctx, k.HostMount(), hostFD, true /* isTTY */) + if err != nil { + return nil, err + } + defer appFile.DecRef() + + // Remember this in the TTY file, as we will use it for the other stdio + // FDs. + ttyFile = appFile + } else { + // Re-use the existing TTY file, as all three stdio FDs must point to + // the same fs.File in order to share TTY state, specifically the + // foreground process group id. + appFile = ttyFile + } + } else { + var err error + appFile, err = hostvfs2.ImportFD(ctx, k.HostMount(), hostFD, false /* isTTY */) + if err != nil { + return nil, err + } + defer appFile.DecRef() + } + + if err := fdTable.NewFDAtVFS2(ctx, int32(appFD), appFile, kernel.FDFlags{}); err != nil { + return nil, err + } + } + + if ttyFile == nil { + return nil, nil + } + return ttyFile.Impl().(*hostvfs2.TTYFileDescription), nil +} diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 34fbc69af..2be498afc 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -422,7 +422,7 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount) (*vfs.F // TODO(gvisor.dev/issue/1672): Whitelist specific file types here, so that // we don't allow importing arbitrary file types without proper support. if i.isTTY { - fd := &ttyFD{ + fd := &TTYFileDescription{ fileDescription: fileDescription{inode: i}, termios: linux.DefaultSlaveTermios, } diff --git a/pkg/sentry/fsimpl/host/tty.go b/pkg/sentry/fsimpl/host/tty.go index 8936afb06..68af6e5af 100644 --- a/pkg/sentry/fsimpl/host/tty.go +++ b/pkg/sentry/fsimpl/host/tty.go @@ -26,15 +26,15 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) -// ttyFD implements vfs.FileDescriptionImpl for a host file descriptor -// that wraps a TTY FD. -type ttyFD struct { +// TTYFileDescription implements vfs.FileDescriptionImpl for a host file +// descriptor that wraps a TTY FD. +type TTYFileDescription struct { fileDescription // mu protects the fields below. mu sync.Mutex `state:"nosave"` - // session is the session attached to this ttyFD. + // session is the session attached to this TTYFileDescription. session *kernel.Session // fgProcessGroup is the foreground process group that is currently @@ -48,7 +48,7 @@ type ttyFD struct { // InitForegroundProcessGroup sets the foreground process group and session for // the TTY. This should only be called once, after the foreground process group // has been created, but before it has started running. -func (t *ttyFD) InitForegroundProcessGroup(pg *kernel.ProcessGroup) { +func (t *TTYFileDescription) InitForegroundProcessGroup(pg *kernel.ProcessGroup) { t.mu.Lock() defer t.mu.Unlock() if t.fgProcessGroup != nil { @@ -59,14 +59,14 @@ func (t *ttyFD) InitForegroundProcessGroup(pg *kernel.ProcessGroup) { } // ForegroundProcessGroup returns the foreground process for the TTY. -func (t *ttyFD) ForegroundProcessGroup() *kernel.ProcessGroup { +func (t *TTYFileDescription) ForegroundProcessGroup() *kernel.ProcessGroup { t.mu.Lock() defer t.mu.Unlock() return t.fgProcessGroup } // Release implements fs.FileOperations.Release. -func (t *ttyFD) Release() { +func (t *TTYFileDescription) Release() { t.mu.Lock() t.fgProcessGroup = nil t.mu.Unlock() @@ -78,7 +78,7 @@ func (t *ttyFD) Release() { // // Reading from a TTY is only allowed for foreground process groups. Background // process groups will either get EIO or a SIGTTIN. -func (t *ttyFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { +func (t *TTYFileDescription) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { t.mu.Lock() defer t.mu.Unlock() @@ -96,7 +96,7 @@ func (t *ttyFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, // // Reading from a TTY is only allowed for foreground process groups. Background // process groups will either get EIO or a SIGTTIN. -func (t *ttyFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { +func (t *TTYFileDescription) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { t.mu.Lock() defer t.mu.Unlock() @@ -111,7 +111,7 @@ func (t *ttyFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadO } // PWrite implements vfs.FileDescriptionImpl. -func (t *ttyFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { +func (t *TTYFileDescription) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { t.mu.Lock() defer t.mu.Unlock() @@ -126,7 +126,7 @@ func (t *ttyFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64 } // Write implements vfs.FileDescriptionImpl. -func (t *ttyFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { +func (t *TTYFileDescription) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { t.mu.Lock() defer t.mu.Unlock() @@ -141,7 +141,7 @@ func (t *ttyFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.Writ } // Ioctl implements vfs.FileDescriptionImpl. -func (t *ttyFD) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { +func (t *TTYFileDescription) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { // Ignore arg[0]. This is the real FD: fd := t.inode.hostFD ioctl := args[1].Uint64() @@ -321,7 +321,7 @@ func (t *ttyFD) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArgum // is a bit convoluted, but documented inline. // // Preconditions: t.mu must be held. -func (t *ttyFD) checkChange(ctx context.Context, sig linux.Signal) error { +func (t *TTYFileDescription) checkChange(ctx context.Context, sig linux.Signal) error { task := kernel.TaskFromContext(ctx) if task == nil { // No task? Linux does not have an analog for this case, but diff --git a/pkg/sentry/vfs/context.go b/pkg/sentry/vfs/context.go index 82781e6d3..c9e724fef 100644 --- a/pkg/sentry/vfs/context.go +++ b/pkg/sentry/vfs/context.go @@ -49,3 +49,27 @@ func RootFromContext(ctx context.Context) VirtualDentry { } return VirtualDentry{} } + +type rootContext struct { + context.Context + root VirtualDentry +} + +// WithRoot returns a copy of ctx with the given root. +func WithRoot(ctx context.Context, root VirtualDentry) context.Context { + return &rootContext{ + Context: ctx, + root: root, + } +} + +// Value implements Context.Value. +func (rc rootContext) Value(key interface{}) interface{} { + switch key { + case CtxRoot: + rc.root.IncRef() + return rc.root + default: + return rc.Context.Value(key) + } +} diff --git a/pkg/sentry/vfs/epoll.go b/pkg/sentry/vfs/epoll.go index 8e0b40841..8297f964b 100644 --- a/pkg/sentry/vfs/epoll.go +++ b/pkg/sentry/vfs/epoll.go @@ -192,6 +192,7 @@ func (ep *EpollInstance) AddInterest(file *FileDescription, num int32, event lin mask: mask, userData: event.Data, } + epi.waiter.Callback = epi ep.interest[key] = epi wmask := waiter.EventMaskFromLinux(mask) file.EventRegister(&epi.waiter, wmask) diff --git a/runsc/boot/BUILD b/runsc/boot/BUILD index abcaf4206..0e71e800b 100644 --- a/runsc/boot/BUILD +++ b/runsc/boot/BUILD @@ -12,7 +12,6 @@ go_library( "controller.go", "debug.go", "events.go", - "fds.go", "fs.go", "limits.go", "loader.go", @@ -43,6 +42,7 @@ go_library( "//pkg/sentry/arch:registers_go_proto", "//pkg/sentry/control", "//pkg/sentry/devices/memdev", + "//pkg/sentry/fdimport", "//pkg/sentry/fs", "//pkg/sentry/fs/dev", "//pkg/sentry/fs/gofer", @@ -53,6 +53,7 @@ go_library( "//pkg/sentry/fs/tmpfs", "//pkg/sentry/fs/tty", "//pkg/sentry/fs/user", + "//pkg/sentry/fsimpl/devpts", "//pkg/sentry/fsimpl/devtmpfs", "//pkg/sentry/fsimpl/gofer", "//pkg/sentry/fsimpl/host", diff --git a/runsc/boot/fds.go b/runsc/boot/fds.go deleted file mode 100644 index 7e7a31fbd..000000000 --- a/runsc/boot/fds.go +++ /dev/null @@ -1,108 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package boot - -import ( - "fmt" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fs/host" - vfshost "gvisor.dev/gvisor/pkg/sentry/fsimpl/host" - "gvisor.dev/gvisor/pkg/sentry/kernel" -) - -// createFDTable creates an FD table that contains stdin, stdout, and stderr. -// If console is true, then ioctl calls will be passed through to the host FD. -// Upon success, createFDMap dups then closes stdioFDs. -func createFDTable(ctx context.Context, console bool, stdioFDs []int) (*kernel.FDTable, error) { - if len(stdioFDs) != 3 { - return nil, fmt.Errorf("stdioFDs should contain exactly 3 FDs (stdin, stdout, and stderr), but %d FDs received", len(stdioFDs)) - } - - if kernel.VFS2Enabled { - return createFDTableVFS2(ctx, console, stdioFDs) - } - - k := kernel.KernelFromContext(ctx) - fdTable := k.NewFDTable() - defer fdTable.DecRef() - - var ttyFile *fs.File - for appFD, hostFD := range stdioFDs { - var appFile *fs.File - - if console && appFD < 3 { - // Import the file as a host TTY file. - if ttyFile == nil { - var err error - appFile, err = host.ImportFile(ctx, hostFD, true /* isTTY */) - if err != nil { - return nil, err - } - defer appFile.DecRef() - - // Remember this in the TTY file, as we will - // use it for the other stdio FDs. - ttyFile = appFile - } else { - // Re-use the existing TTY file, as all three - // stdio FDs must point to the same fs.File in - // order to share TTY state, specifically the - // foreground process group id. - appFile = ttyFile - } - } else { - // Import the file as a regular host file. - var err error - appFile, err = host.ImportFile(ctx, hostFD, false /* isTTY */) - if err != nil { - return nil, err - } - defer appFile.DecRef() - } - - // Add the file to the FD map. - if err := fdTable.NewFDAt(ctx, int32(appFD), appFile, kernel.FDFlags{}); err != nil { - return nil, err - } - } - - fdTable.IncRef() - return fdTable, nil -} - -func createFDTableVFS2(ctx context.Context, console bool, stdioFDs []int) (*kernel.FDTable, error) { - k := kernel.KernelFromContext(ctx) - fdTable := k.NewFDTable() - defer fdTable.DecRef() - - for appFD, hostFD := range stdioFDs { - // TODO(gvisor.dev/issue/1482): Add TTY support. - appFile, err := vfshost.ImportFD(ctx, k.HostMount(), hostFD, false) - if err != nil { - return nil, err - } - - if err := fdTable.NewFDAtVFS2(ctx, int32(appFD), appFile, kernel.FDFlags{}); err != nil { - appFile.DecRef() - return nil, err - } - appFile.DecRef() - } - - fdTable.IncRef() - return fdTable, nil -} diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index f6ea4c102..79ef3a880 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -27,16 +27,18 @@ import ( specs "github.com/opencontainers/runtime-spec/specs-go" "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/memutil" "gvisor.dev/gvisor/pkg/rand" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/control" + "gvisor.dev/gvisor/pkg/sentry/fdimport" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/host" "gvisor.dev/gvisor/pkg/sentry/fs/user" - vfs2host "gvisor.dev/gvisor/pkg/sentry/fsimpl/host" + hostvfs2 "gvisor.dev/gvisor/pkg/sentry/fsimpl/host" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" @@ -143,6 +145,9 @@ type execProcess struct { // tty will be nil if the process is not attached to a terminal. tty *host.TTYFileOperations + // tty will be nil if the process is not attached to a terminal. + ttyVFS2 *hostvfs2.TTYFileDescription + // pidnsPath is the pid namespace path in spec pidnsPath string } @@ -333,7 +338,7 @@ func New(args Args) (*Loader, error) { if kernel.VFS2Enabled { // Set up host mount that will be used for imported fds. - hostFilesystem := vfs2host.NewFilesystem(k.VFS()) + hostFilesystem := hostvfs2.NewFilesystem(k.VFS()) defer hostFilesystem.DecRef() hostMount, err := k.VFS().NewDisconnectedMount(hostFilesystem, nil, &vfs.MountOptions{}) if err != nil { @@ -528,6 +533,8 @@ func (l *Loader) run() error { // If we are restoring, we do not want to create a process. // l.restore is set by the container manager when a restore call is made. + var ttyFile *host.TTYFileOperations + var ttyFileVFS2 *hostvfs2.TTYFileDescription if !l.restore { if l.conf.ProfileEnable { pprof.Initialize() @@ -542,13 +549,14 @@ func (l *Loader) run() error { // Create the FD map, which will set stdin, stdout, and stderr. If console // is true, then ioctl calls will be passed through to the host fd. ctx := l.rootProcArgs.NewContext(l.k) - fdTable, err := createFDTable(ctx, l.console, l.stdioFDs) + var err error + + // CreateProcess takes a reference on FDMap if successful. We won't need + // ours either way. + l.rootProcArgs.FDTable, ttyFile, ttyFileVFS2, err = createFDTable(ctx, l.console, l.stdioFDs) if err != nil { return fmt.Errorf("importing fds: %v", err) } - // CreateProcess takes a reference on FDMap if successful. We won't need - // ours either way. - l.rootProcArgs.FDTable = fdTable // Setup the root container file system. l.startGoferMonitor(l.sandboxID, l.goferFDs) @@ -591,14 +599,16 @@ func (l *Loader) run() error { ep.pidnsPath = ns.Path } if l.console { - ttyFile, _ := l.rootProcArgs.FDTable.Get(0) - defer ttyFile.DecRef() - ep.tty = ttyFile.FileOperations.(*host.TTYFileOperations) - - // Set the foreground process group on the TTY to the global - // init process group, since that is what we are about to - // start running. - ep.tty.InitForegroundProcessGroup(ep.tg.ProcessGroup()) + // Set the foreground process group on the TTY to the global init process + // group, since that is what we are about to start running. + switch { + case ttyFileVFS2 != nil: + ep.ttyVFS2 = ttyFileVFS2 + ttyFileVFS2.InitForegroundProcessGroup(ep.tg.ProcessGroup()) + case ttyFile != nil: + ep.tty = ttyFile + ttyFile.InitForegroundProcessGroup(ep.tg.ProcessGroup()) + } } // Handle signals by forwarding them to the root container process @@ -719,7 +729,7 @@ func (l *Loader) startContainer(spec *specs.Spec, conf *Config, cid string, file // Create the FD map, which will set stdin, stdout, and stderr. ctx := procArgs.NewContext(l.k) - fdTable, err := createFDTable(ctx, false, stdioFDs) + fdTable, _, _, err := createFDTable(ctx, false, stdioFDs) if err != nil { return fmt.Errorf("importing fds: %v", err) } @@ -804,14 +814,14 @@ func (l *Loader) destroyContainer(cid string) error { l.mu.Lock() defer l.mu.Unlock() - _, _, started, err := l.threadGroupFromIDLocked(execID{cid: cid}) + tg, err := l.tryThreadGroupFromIDLocked(execID{cid: cid}) if err != nil { // Container doesn't exist. return err } - // The container exists, has it been started? - if started { + // The container exists, but has it been started? + if tg != nil { if err := l.signalAllProcesses(cid, int32(linux.SIGKILL)); err != nil { return fmt.Errorf("sending SIGKILL to all container processes: %v", err) } @@ -853,48 +863,65 @@ func (l *Loader) executeAsync(args *control.ExecArgs) (kernel.ThreadID, error) { l.mu.Lock() defer l.mu.Unlock() - tg, _, started, err := l.threadGroupFromIDLocked(execID{cid: args.ContainerID}) + tg, err := l.tryThreadGroupFromIDLocked(execID{cid: args.ContainerID}) if err != nil { return 0, err } - if !started { + if tg == nil { return 0, fmt.Errorf("container %q not started", args.ContainerID) } - // TODO(gvisor.dev/issue/1623): Add VFS2 support - // Get the container MountNamespace from the Task. - tg.Leader().WithMuLocked(func(t *kernel.Task) { + if kernel.VFS2Enabled { // task.MountNamespace() does not take a ref, so we must do so ourselves. - args.MountNamespace = t.MountNamespace() - args.MountNamespace.IncRef() - }) - if args.MountNamespace != nil { - defer args.MountNamespace.DecRef() + args.MountNamespaceVFS2 = tg.Leader().MountNamespaceVFS2() + args.MountNamespaceVFS2.IncRef() + } else { + tg.Leader().WithMuLocked(func(t *kernel.Task) { + // task.MountNamespace() does not take a ref, so we must do so ourselves. + args.MountNamespace = t.MountNamespace() + args.MountNamespace.IncRef() + }) } // Add the HOME environment variable if it is not already set. - root := args.MountNamespace.Root() - defer root.DecRef() - ctx := fs.WithRoot(l.k.SupervisorContext(), root) - envv, err := user.MaybeAddExecUserHome(ctx, args.MountNamespace, args.KUID, args.Envv) - if err != nil { - return 0, err + if kernel.VFS2Enabled { + defer args.MountNamespaceVFS2.DecRef() + + root := args.MountNamespaceVFS2.Root() + defer root.DecRef() + ctx := vfs.WithRoot(l.k.SupervisorContext(), root) + envv, err := user.MaybeAddExecUserHomeVFS2(ctx, args.MountNamespaceVFS2, args.KUID, args.Envv) + if err != nil { + return 0, err + } + args.Envv = envv + } else { + defer args.MountNamespace.DecRef() + + root := args.MountNamespace.Root() + defer root.DecRef() + ctx := fs.WithRoot(l.k.SupervisorContext(), root) + envv, err := user.MaybeAddExecUserHome(ctx, args.MountNamespace, args.KUID, args.Envv) + if err != nil { + return 0, err + } + args.Envv = envv } - args.Envv = envv // Start the process. proc := control.Proc{Kernel: l.k} args.PIDNamespace = tg.PIDNamespace() - newTG, tgid, ttyFile, err := control.ExecAsync(&proc, args) + newTG, tgid, ttyFile, ttyFileVFS2, err := control.ExecAsync(&proc, args) if err != nil { return 0, err } eid := execID{cid: args.ContainerID, pid: tgid} l.processes[eid] = &execProcess{ - tg: newTG, - tty: ttyFile, + tg: newTG, + tty: ttyFile, + ttyVFS2: ttyFileVFS2, } log.Debugf("updated processes: %v", l.processes) @@ -905,7 +932,7 @@ func (l *Loader) executeAsync(args *control.ExecArgs) (kernel.ThreadID, error) { func (l *Loader) waitContainer(cid string, waitStatus *uint32) error { // Don't defer unlock, as doing so would make it impossible for // multiple clients to wait on the same container. - tg, _, err := l.threadGroupFromID(execID{cid: cid}) + tg, err := l.threadGroupFromID(execID{cid: cid}) if err != nil { return fmt.Errorf("can't wait for container %q: %v", cid, err) } @@ -924,7 +951,7 @@ func (l *Loader) waitPID(tgid kernel.ThreadID, cid string, waitStatus *uint32) e // Try to find a process that was exec'd eid := execID{cid: cid, pid: tgid} - execTG, _, err := l.threadGroupFromID(eid) + execTG, err := l.threadGroupFromID(eid) if err == nil { ws := l.wait(execTG) *waitStatus = ws @@ -938,7 +965,7 @@ func (l *Loader) waitPID(tgid kernel.ThreadID, cid string, waitStatus *uint32) e // The caller may be waiting on a process not started directly via exec. // In this case, find the process in the container's PID namespace. - initTG, _, err := l.threadGroupFromID(execID{cid: cid}) + initTG, err := l.threadGroupFromID(execID{cid: cid}) if err != nil { return fmt.Errorf("waiting for PID %d: %v", tgid, err) } @@ -1089,8 +1116,7 @@ func (l *Loader) signal(cid string, pid, signo int32, mode SignalDeliveryMode) e return fmt.Errorf("PID (%d) cannot be set when signaling all processes", pid) } // Check that the container has actually started before signaling it. - _, _, err := l.threadGroupFromID(execID{cid: cid}) - if err != nil { + if _, err := l.threadGroupFromID(execID{cid: cid}); err != nil { return err } if err := l.signalAllProcesses(cid, signo); err != nil { @@ -1104,7 +1130,7 @@ func (l *Loader) signal(cid string, pid, signo int32, mode SignalDeliveryMode) e } func (l *Loader) signalProcess(cid string, tgid kernel.ThreadID, signo int32) error { - execTG, _, err := l.threadGroupFromID(execID{cid: cid, pid: tgid}) + execTG, err := l.threadGroupFromID(execID{cid: cid, pid: tgid}) if err == nil { // Send signal directly to the identified process. return l.k.SendExternalSignalThreadGroup(execTG, &arch.SignalInfo{Signo: signo}) @@ -1113,7 +1139,7 @@ func (l *Loader) signalProcess(cid string, tgid kernel.ThreadID, signo int32) er // The caller may be signaling a process not started directly via exec. // In this case, find the process in the container's PID namespace and // signal it. - initTG, _, err := l.threadGroupFromID(execID{cid: cid}) + initTG, err := l.threadGroupFromID(execID{cid: cid}) if err != nil { return fmt.Errorf("no thread group found: %v", err) } @@ -1127,17 +1153,35 @@ func (l *Loader) signalProcess(cid string, tgid kernel.ThreadID, signo int32) er return l.k.SendExternalSignalThreadGroup(tg, &arch.SignalInfo{Signo: signo}) } +// signalForegrondProcessGroup looks up foreground process group from the TTY +// for the given "tgid" inside container "cid", and send the signal to it. func (l *Loader) signalForegrondProcessGroup(cid string, tgid kernel.ThreadID, signo int32) error { - // Lookup foreground process group from the TTY for the given process, - // and send the signal to it. - tg, tty, err := l.threadGroupFromID(execID{cid: cid, pid: tgid}) + l.mu.Lock() + tg, err := l.tryThreadGroupFromIDLocked(execID{cid: cid, pid: tgid}) if err != nil { + l.mu.Unlock() return fmt.Errorf("no thread group found: %v", err) } - if tty == nil { + if tg == nil { + l.mu.Unlock() + return fmt.Errorf("container %q not started", cid) + } + + tty, ttyVFS2, err := l.ttyFromIDLocked(execID{cid: cid, pid: tgid}) + l.mu.Unlock() + if err != nil { + return fmt.Errorf("no thread group found: %v", err) + } + + var pg *kernel.ProcessGroup + switch { + case ttyVFS2 != nil: + pg = ttyVFS2.ForegroundProcessGroup() + case tty != nil: + pg = tty.ForegroundProcessGroup() + default: return fmt.Errorf("no TTY attached") } - pg := tty.ForegroundProcessGroup() if pg == nil { // No foreground process group has been set. Signal the // original thread group. @@ -1168,33 +1212,57 @@ func (l *Loader) signalAllProcesses(cid string, signo int32) error { return l.k.SendContainerSignal(cid, &arch.SignalInfo{Signo: signo}) } -// threadGroupFromID same as threadGroupFromIDLocked except that it acquires -// mutex before calling it. -func (l *Loader) threadGroupFromID(key execID) (*kernel.ThreadGroup, *host.TTYFileOperations, error) { +// threadGroupFromID is similar to tryThreadGroupFromIDLocked except that it +// acquires mutex before calling it and fails in case container hasn't started +// yet. +func (l *Loader) threadGroupFromID(key execID) (*kernel.ThreadGroup, error) { l.mu.Lock() defer l.mu.Unlock() - tg, tty, ok, err := l.threadGroupFromIDLocked(key) + tg, err := l.tryThreadGroupFromIDLocked(key) if err != nil { - return nil, nil, err + return nil, err } - if !ok { - return nil, nil, fmt.Errorf("container %q not started", key.cid) + if tg == nil { + return nil, fmt.Errorf("container %q not started", key.cid) } - return tg, tty, nil + return tg, nil } -// threadGroupFromIDLocked returns the thread group and TTY for the given -// execution ID. TTY may be nil if the process is not attached to a terminal. -// Also returns a boolean indicating whether the container has already started. -// Returns error if execution ID is invalid or if the container cannot be -// found (maybe it has been deleted). Caller must hold 'mu'. -func (l *Loader) threadGroupFromIDLocked(key execID) (*kernel.ThreadGroup, *host.TTYFileOperations, bool, error) { +// tryThreadGroupFromIDLocked returns the thread group for the given execution +// ID. It may return nil in case the container has not started yet. Returns +// error if execution ID is invalid or if the container cannot be found (maybe +// it has been deleted). Caller must hold 'mu'. +func (l *Loader) tryThreadGroupFromIDLocked(key execID) (*kernel.ThreadGroup, error) { ep := l.processes[key] if ep == nil { - return nil, nil, false, fmt.Errorf("container %q not found", key.cid) + return nil, fmt.Errorf("container %q not found", key.cid) } - if ep.tg == nil { - return nil, nil, false, nil + return ep.tg, nil +} + +// ttyFromIDLocked returns the TTY files for the given execution ID. It may +// return nil in case the container has not started yet. Returns error if +// execution ID is invalid or if the container cannot be found (maybe it has +// been deleted). Caller must hold 'mu'. +func (l *Loader) ttyFromIDLocked(key execID) (*host.TTYFileOperations, *hostvfs2.TTYFileDescription, error) { + ep := l.processes[key] + if ep == nil { + return nil, nil, fmt.Errorf("container %q not found", key.cid) + } + return ep.tty, ep.ttyVFS2, nil +} + +func createFDTable(ctx context.Context, console bool, stdioFDs []int) (*kernel.FDTable, *host.TTYFileOperations, *hostvfs2.TTYFileDescription, error) { + if len(stdioFDs) != 3 { + return nil, nil, nil, fmt.Errorf("stdioFDs should contain exactly 3 FDs (stdin, stdout, and stderr), but %d FDs received", len(stdioFDs)) + } + + k := kernel.KernelFromContext(ctx) + fdTable := k.NewFDTable() + ttyFile, ttyFileVFS2, err := fdimport.Import(ctx, fdTable, console, stdioFDs) + if err != nil { + fdTable.DecRef() + return nil, nil, nil, err } - return ep.tg, ep.tty, true, nil + return fdTable, ttyFile, ttyFileVFS2, nil } diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go index 0b9b0b436..448fc4459 100644 --- a/runsc/boot/vfs.go +++ b/runsc/boot/vfs.go @@ -26,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/devices/memdev" "gvisor.dev/gvisor/pkg/sentry/fs" + devpts2 "gvisor.dev/gvisor/pkg/sentry/fsimpl/devpts" devtmpfsimpl "gvisor.dev/gvisor/pkg/sentry/fsimpl/devtmpfs" goferimpl "gvisor.dev/gvisor/pkg/sentry/fsimpl/gofer" procimpl "gvisor.dev/gvisor/pkg/sentry/fsimpl/proc" @@ -41,37 +42,28 @@ import ( ) func registerFilesystems(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials) error { - - vfsObj.MustRegisterFilesystemType(rootFsName, &goferimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserList: true, - }) - - vfsObj.MustRegisterFilesystemType(bind, &goferimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + vfsObj.MustRegisterFilesystemType(devpts2.Name, &devpts2.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserList: true, + // TODO(b/29356795): Users may mount this once the terminals are in a + // usable state. + AllowUserMount: false, }) - - vfsObj.MustRegisterFilesystemType(devpts, &devtmpfsimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + vfsObj.MustRegisterFilesystemType(devtmpfsimpl.Name, &devtmpfsimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, AllowUserList: true, }) - - vfsObj.MustRegisterFilesystemType(devtmpfs, &devtmpfsimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - AllowUserList: true, - }) - vfsObj.MustRegisterFilesystemType(proc, &procimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - AllowUserList: true, + vfsObj.MustRegisterFilesystemType(goferimpl.Name, &goferimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserList: true, }) - vfsObj.MustRegisterFilesystemType(sysfs, &sysimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + vfsObj.MustRegisterFilesystemType(procimpl.Name, &procimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, AllowUserList: true, }) - vfsObj.MustRegisterFilesystemType(tmpfs, &tmpfsimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + vfsObj.MustRegisterFilesystemType(sysimpl.Name, &sysimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, AllowUserList: true, }) - vfsObj.MustRegisterFilesystemType(nonefs, &sysimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + vfsObj.MustRegisterFilesystemType(tmpfsimpl.Name, &tmpfsimpl.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, AllowUserList: true, }) @@ -108,7 +100,6 @@ func setupContainerVFS2(ctx context.Context, conf *Config, mntr *containerMounte } func setExecutablePathVFS2(ctx context.Context, procArgs *kernel.CreateProcessArgs) error { - exe := procArgs.Argv[0] // Absolute paths can be used directly. @@ -120,11 +111,9 @@ func setExecutablePathVFS2(ctx context.Context, procArgs *kernel.CreateProcessAr // Paths with '/' in them should be joined to the working directory, or // to the root if working directory is not set. if strings.IndexByte(exe, '/') > 0 { - if !path.IsAbs(procArgs.WorkingDirectory) { return fmt.Errorf("working directory %q must be absolute", procArgs.WorkingDirectory) } - procArgs.Filename = path.Join(procArgs.WorkingDirectory, exe) return nil } @@ -144,21 +133,17 @@ func setExecutablePathVFS2(ctx context.Context, procArgs *kernel.CreateProcessAr creds := procArgs.Credentials for _, p := range paths { - binPath := path.Join(p, exe) - pop := &vfs.PathOperation{ Root: root, Start: root, Path: fspath.Parse(binPath), FollowFinalSymlink: true, } - opts := &vfs.OpenOptions{ FileExec: true, Flags: linux.O_RDONLY, } - dentry, err := root.Mount().Filesystem().VirtualFilesystem().OpenAt(ctx, creds, pop, opts) if err == syserror.ENOENT || err == syserror.EACCES { // Didn't find it here. @@ -209,7 +194,6 @@ func (c *containerMounter) setupVFS2(ctx context.Context, conf *Config, procArgs } func (c *containerMounter) createMountNamespaceVFS2(ctx context.Context, conf *Config, creds *auth.Credentials) (*vfs.MountNamespace, error) { - fd := c.fds.remove() opts := strings.Join(p9MountOptionsVFS2(fd, conf.FileAccess), ",") @@ -222,7 +206,6 @@ func (c *containerMounter) createMountNamespaceVFS2(ctx context.Context, conf *C } func (c *containerMounter) mountSubmountsVFS2(ctx context.Context, conf *Config, mns *vfs.MountNamespace, creds *auth.Credentials) error { - c.prepareMountsVFS2() for _, submount := range c.mounts { @@ -256,7 +239,6 @@ func (c *containerMounter) mountSubmountVFS2(ctx context.Context, conf *Config, if err != nil { return fmt.Errorf("mountOptions failed: %w", err) } - if fsName == "" { // Filesystem is not supported (e.g. cgroup), just skip it. return nil @@ -277,7 +259,7 @@ func (c *containerMounter) mountSubmountVFS2(ctx context.Context, conf *Config, // All writes go to upper, be paranoid and make lower readonly. opts.ReadOnly = useOverlay - if err := c.k.VFS().MountAt(ctx, creds, "", target, submount.Type, opts); err != nil { + if err := c.k.VFS().MountAt(ctx, creds, "", target, fsName, opts); err != nil { return fmt.Errorf("failed to mount %q (type: %s): %w, opts: %v", submount.Destination, submount.Type, err, opts) } log.Infof("Mounted %q to %q type: %s, internal-options: %q", submount.Source, submount.Destination, submount.Type, opts) diff --git a/runsc/container/container_test.go b/runsc/container/container_test.go index f607fe8af..1bea5db02 100644 --- a/runsc/container/container_test.go +++ b/runsc/container/container_test.go @@ -2189,63 +2189,70 @@ func TestTTYField(t *testing.T) { } for _, test := range testCases { - t.Run(test.name, func(t *testing.T) { - conf := testutil.TestConfig(t) - - // We will run /bin/sleep, possibly with an open TTY. - cmd := []string{"/bin/sleep", "10000"} - if test.useTTY { - // Run inside the "pty-runner". - cmd = append([]string{testApp, "pty-runner"}, cmd...) - } - - spec := testutil.NewSpecWithArgs(cmd...) - _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) - if err != nil { - t.Fatalf("error setting up container: %v", err) - } - defer cleanup() + for _, vfs2 := range []bool{false, true} { + name := test.name + if vfs2 { + name += "-vfs2" + } + t.Run(name, func(t *testing.T) { + conf := testutil.TestConfig(t) + conf.VFS2 = vfs2 + + // We will run /bin/sleep, possibly with an open TTY. + cmd := []string{"/bin/sleep", "10000"} + if test.useTTY { + // Run inside the "pty-runner". + cmd = append([]string{testApp, "pty-runner"}, cmd...) + } - // Create and start the container. - args := Args{ - ID: testutil.RandomContainerID(), - Spec: spec, - BundleDir: bundleDir, - } - c, err := New(conf, args) - if err != nil { - t.Fatalf("error creating container: %v", err) - } - defer c.Destroy() - if err := c.Start(conf); err != nil { - t.Fatalf("error starting container: %v", err) - } + spec := testutil.NewSpecWithArgs(cmd...) + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() - // Wait for sleep to be running, and check the TTY - // field. - var gotTTYField string - cb := func() error { - ps, err := c.Processes() + // Create and start the container. + args := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + c, err := New(conf, args) if err != nil { - err = fmt.Errorf("error getting process data from container: %v", err) - return &backoff.PermanentError{Err: err} + t.Fatalf("error creating container: %v", err) + } + defer c.Destroy() + if err := c.Start(conf); err != nil { + t.Fatalf("error starting container: %v", err) } - for _, p := range ps { - if strings.Contains(p.Cmd, "sleep") { - gotTTYField = p.TTY - return nil + + // Wait for sleep to be running, and check the TTY + // field. + var gotTTYField string + cb := func() error { + ps, err := c.Processes() + if err != nil { + err = fmt.Errorf("error getting process data from container: %v", err) + return &backoff.PermanentError{Err: err} } + for _, p := range ps { + if strings.Contains(p.Cmd, "sleep") { + gotTTYField = p.TTY + return nil + } + } + return fmt.Errorf("sleep not running") + } + if err := testutil.Poll(cb, 30*time.Second); err != nil { + t.Fatalf("error waiting for sleep process: %v", err) } - return fmt.Errorf("sleep not running") - } - if err := testutil.Poll(cb, 30*time.Second); err != nil { - t.Fatalf("error waiting for sleep process: %v", err) - } - if gotTTYField != test.wantTTYField { - t.Errorf("tty field got %q, want %q", gotTTYField, test.wantTTYField) - } - }) + if gotTTYField != test.wantTTYField { + t.Errorf("tty field got %q, want %q", gotTTYField, test.wantTTYField) + } + }) + } } } -- cgit v1.2.3 From 57dbd7f3624528a416664a618ce9edd4c9096d8d Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Mon, 4 May 2020 13:39:15 -0700 Subject: Remove kernfs.Filesystem cast from GenericDirectoryFD This allows for kerfs.Filesystem to be overridden by different implementations. Updates #1672 PiperOrigin-RevId: 309809321 --- pkg/sentry/fsimpl/kernfs/fd_impl_util.go | 30 +++++++++++++++++------------- 1 file changed, 17 insertions(+), 13 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index dd5806301..8284e76a7 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -22,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" ) @@ -36,13 +37,20 @@ import ( // inode. // // Must be initialize with Init before first use. +// +// Lock ordering: mu => children.mu. type GenericDirectoryFD struct { vfs.FileDescriptionDefaultImpl vfs.DirectoryFileDescriptionDefaultImpl vfsfd vfs.FileDescription children *OrderedChildren - off int64 + + // mu protects the fields below. + mu sync.Mutex + + // off is the current directory offset. Protected by "mu". + off int64 } // NewGenericDirectoryFD creates a new GenericDirectoryFD and returns its @@ -115,17 +123,13 @@ func (fd *GenericDirectoryFD) inode() Inode { // IterDirents implements vfs.FileDecriptionImpl.IterDirents. IterDirents holds // o.mu when calling cb. func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback) error { - vfsFS := fd.filesystem() - fs := vfsFS.Impl().(*Filesystem) - vfsd := fd.vfsfd.VirtualDentry().Dentry() - - fs.mu.Lock() - defer fs.mu.Unlock() + fd.mu.Lock() + defer fd.mu.Unlock() opts := vfs.StatOptions{Mask: linux.STATX_INO} // Handle ".". if fd.off == 0 { - stat, err := fd.inode().Stat(vfsFS, opts) + stat, err := fd.inode().Stat(fd.filesystem(), opts) if err != nil { return err } @@ -143,8 +147,9 @@ func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirent // Handle "..". if fd.off == 1 { + vfsd := fd.vfsfd.VirtualDentry().Dentry() parentInode := genericParentOrSelf(vfsd.Impl().(*Dentry)).inode - stat, err := parentInode.Stat(vfsFS, opts) + stat, err := parentInode.Stat(fd.filesystem(), opts) if err != nil { return err } @@ -168,7 +173,7 @@ func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirent childIdx := fd.off - 2 for it := fd.children.nthLocked(childIdx); it != nil; it = it.Next() { inode := it.Dentry.Impl().(*Dentry).inode - stat, err := inode.Stat(vfsFS, opts) + stat, err := inode.Stat(fd.filesystem(), opts) if err != nil { return err } @@ -192,9 +197,8 @@ func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirent // Seek implements vfs.FileDecriptionImpl.Seek. func (fd *GenericDirectoryFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - fs := fd.filesystem().Impl().(*Filesystem) - fs.mu.Lock() - defer fs.mu.Unlock() + fd.mu.Lock() + defer fd.mu.Unlock() switch whence { case linux.SEEK_SET: -- cgit v1.2.3 From da71dc7fddda387232b243c6176de21a1208ad0c Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Mon, 4 May 2020 16:00:22 -0700 Subject: Port eventfd to VFS2. And move sys_timerfd.go to just timerfd.go for consistency. Updates #1475. PiperOrigin-RevId: 309835029 --- pkg/sentry/syscalls/linux/sys_eventfd.go | 17 +- pkg/sentry/syscalls/linux/vfs2/BUILD | 3 +- pkg/sentry/syscalls/linux/vfs2/eventfd.go | 59 +++++ .../syscalls/linux/vfs2/linux64_override_amd64.go | 4 +- pkg/sentry/syscalls/linux/vfs2/sys_timerfd.go | 123 --------- pkg/sentry/syscalls/linux/vfs2/timerfd.go | 123 +++++++++ pkg/sentry/vfs/BUILD | 4 + pkg/sentry/vfs/eventfd.go | 282 +++++++++++++++++++++ pkg/sentry/vfs/eventfd_test.go | 96 +++++++ test/syscalls/linux/eventfd.cc | 16 ++ 10 files changed, 589 insertions(+), 138 deletions(-) create mode 100644 pkg/sentry/syscalls/linux/vfs2/eventfd.go delete mode 100644 pkg/sentry/syscalls/linux/vfs2/sys_timerfd.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/timerfd.go create mode 100644 pkg/sentry/vfs/eventfd.go create mode 100644 pkg/sentry/vfs/eventfd_test.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/sys_eventfd.go b/pkg/sentry/syscalls/linux/sys_eventfd.go index 8a34c4e99..ed3413ca6 100644 --- a/pkg/sentry/syscalls/linux/sys_eventfd.go +++ b/pkg/sentry/syscalls/linux/sys_eventfd.go @@ -15,6 +15,7 @@ package linux import ( + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -22,32 +23,24 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) -const ( - // EFD_SEMAPHORE is a flag used in syscall eventfd(2) and eventfd2(2). Please - // see its man page for more information. - EFD_SEMAPHORE = 1 - EFD_NONBLOCK = 0x800 - EFD_CLOEXEC = 0x80000 -) - // Eventfd2 implements linux syscall eventfd2(2). func Eventfd2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { initVal := args[0].Int() flags := uint(args[1].Uint()) - allOps := uint(EFD_SEMAPHORE | EFD_NONBLOCK | EFD_CLOEXEC) + allOps := uint(linux.EFD_SEMAPHORE | linux.EFD_NONBLOCK | linux.EFD_CLOEXEC) if flags & ^allOps != 0 { return 0, nil, syserror.EINVAL } - event := eventfd.New(t, uint64(initVal), flags&EFD_SEMAPHORE != 0) + event := eventfd.New(t, uint64(initVal), flags&linux.EFD_SEMAPHORE != 0) event.SetFlags(fs.SettableFileFlags{ - NonBlocking: flags&EFD_NONBLOCK != 0, + NonBlocking: flags&linux.EFD_NONBLOCK != 0, }) defer event.DecRef() fd, err := t.NewFDFrom(0, event, kernel.FDFlags{ - CloseOnExec: flags&EFD_CLOEXEC != 0, + CloseOnExec: flags&linux.EFD_CLOEXEC != 0, }) if err != nil { return 0, nil, err diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD index f6fb0f219..ffca627d4 100644 --- a/pkg/sentry/syscalls/linux/vfs2/BUILD +++ b/pkg/sentry/syscalls/linux/vfs2/BUILD @@ -6,6 +6,7 @@ go_library( name = "vfs2", srcs = [ "epoll.go", + "eventfd.go", "execve.go", "fd.go", "filesystem.go", @@ -26,7 +27,7 @@ go_library( "stat_amd64.go", "stat_arm64.go", "sync.go", - "sys_timerfd.go", + "timerfd.go", "xattr.go", ], marshal = True, diff --git a/pkg/sentry/syscalls/linux/vfs2/eventfd.go b/pkg/sentry/syscalls/linux/vfs2/eventfd.go new file mode 100644 index 000000000..bd2194972 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/eventfd.go @@ -0,0 +1,59 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/syserror" +) + +// Eventfd2 implements linux syscall eventfd2(2). +func Eventfd2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + initVal := uint64(args[0].Uint()) + flags := uint(args[1].Uint()) + allOps := uint(linux.EFD_SEMAPHORE | linux.EFD_NONBLOCK | linux.EFD_CLOEXEC) + + if flags & ^allOps != 0 { + return 0, nil, syserror.EINVAL + } + + fileFlags := uint32(linux.O_RDWR) + if flags&linux.EFD_NONBLOCK != 0 { + fileFlags |= linux.O_NONBLOCK + } + semMode := flags&linux.EFD_SEMAPHORE != 0 + eventfd, err := t.Kernel().VFS().NewEventFD(initVal, semMode, fileFlags) + if err != nil { + return 0, nil, err + } + defer eventfd.DecRef() + + fd, err := t.NewFDFromVFS2(0, eventfd, kernel.FDFlags{ + CloseOnExec: flags&linux.EFD_CLOEXEC != 0, + }) + if err != nil { + return 0, nil, err + } + + return uintptr(fd), nil, nil +} + +// Eventfd implements linux syscall eventfd(2). +func Eventfd(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + args[1].Value = 0 + return Eventfd2(t, args) +} diff --git a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go index 74920f785..074f58e5d 100644 --- a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go +++ b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go @@ -141,14 +141,14 @@ func Override(table map[uintptr]kernel.Syscall) { table[281] = syscalls.Supported("epoll_pwait", EpollPwait) delete(table, 282) // signalfd table[283] = syscalls.Supported("timerfd_create", TimerfdCreate) - delete(table, 284) // eventfd + table[284] = syscalls.Supported("eventfd", Eventfd) delete(table, 285) // fallocate table[286] = syscalls.Supported("timerfd_settime", TimerfdSettime) table[287] = syscalls.Supported("timerfd_gettime", TimerfdGettime) // TODO(gvisor.dev/issue/1485): Port all socket variants to VFS2. table[288] = syscalls.PartiallySupported("accept4", Accept4, "In process of porting socket syscalls to VFS2.", nil) delete(table, 289) // signalfd4 - delete(table, 290) // eventfd2 + table[290] = syscalls.Supported("eventfd2", Eventfd2) table[291] = syscalls.Supported("epoll_create1", EpollCreate1) table[292] = syscalls.Supported("dup3", Dup3) table[293] = syscalls.Supported("pipe2", Pipe2) diff --git a/pkg/sentry/syscalls/linux/vfs2/sys_timerfd.go b/pkg/sentry/syscalls/linux/vfs2/sys_timerfd.go deleted file mode 100644 index 7938a5249..000000000 --- a/pkg/sentry/syscalls/linux/vfs2/sys_timerfd.go +++ /dev/null @@ -1,123 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package vfs2 - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/kernel" - ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -// TimerfdCreate implements Linux syscall timerfd_create(2). -func TimerfdCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - clockID := args[0].Int() - flags := args[1].Int() - - if flags&^(linux.TFD_CLOEXEC|linux.TFD_NONBLOCK) != 0 { - return 0, nil, syserror.EINVAL - } - - var fileFlags uint32 - if flags&linux.TFD_NONBLOCK != 0 { - fileFlags = linux.O_NONBLOCK - } - - var clock ktime.Clock - switch clockID { - case linux.CLOCK_REALTIME: - clock = t.Kernel().RealtimeClock() - case linux.CLOCK_MONOTONIC, linux.CLOCK_BOOTTIME: - clock = t.Kernel().MonotonicClock() - default: - return 0, nil, syserror.EINVAL - } - file, err := t.Kernel().VFS().NewTimerFD(clock, fileFlags) - if err != nil { - return 0, nil, err - } - defer file.DecRef() - fd, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{ - CloseOnExec: flags&linux.TFD_CLOEXEC != 0, - }) - if err != nil { - return 0, nil, err - } - return uintptr(fd), nil, nil -} - -// TimerfdSettime implements Linux syscall timerfd_settime(2). -func TimerfdSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := args[0].Int() - flags := args[1].Int() - newValAddr := args[2].Pointer() - oldValAddr := args[3].Pointer() - - if flags&^(linux.TFD_TIMER_ABSTIME) != 0 { - return 0, nil, syserror.EINVAL - } - - file := t.GetFileVFS2(fd) - if file == nil { - return 0, nil, syserror.EBADF - } - defer file.DecRef() - - tfd, ok := file.Impl().(*vfs.TimerFileDescription) - if !ok { - return 0, nil, syserror.EINVAL - } - - var newVal linux.Itimerspec - if _, err := t.CopyIn(newValAddr, &newVal); err != nil { - return 0, nil, err - } - newS, err := ktime.SettingFromItimerspec(newVal, flags&linux.TFD_TIMER_ABSTIME != 0, tfd.Clock()) - if err != nil { - return 0, nil, err - } - tm, oldS := tfd.SetTime(newS) - if oldValAddr != 0 { - oldVal := ktime.ItimerspecFromSetting(tm, oldS) - if _, err := t.CopyOut(oldValAddr, &oldVal); err != nil { - return 0, nil, err - } - } - return 0, nil, nil -} - -// TimerfdGettime implements Linux syscall timerfd_gettime(2). -func TimerfdGettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := args[0].Int() - curValAddr := args[1].Pointer() - - file := t.GetFileVFS2(fd) - if file == nil { - return 0, nil, syserror.EBADF - } - defer file.DecRef() - - tfd, ok := file.Impl().(*vfs.TimerFileDescription) - if !ok { - return 0, nil, syserror.EINVAL - } - - tm, s := tfd.GetTime() - curVal := ktime.ItimerspecFromSetting(tm, s) - _, err := t.CopyOut(curValAddr, &curVal) - return 0, nil, err -} diff --git a/pkg/sentry/syscalls/linux/vfs2/timerfd.go b/pkg/sentry/syscalls/linux/vfs2/timerfd.go new file mode 100644 index 000000000..7938a5249 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/timerfd.go @@ -0,0 +1,123 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// TimerfdCreate implements Linux syscall timerfd_create(2). +func TimerfdCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + clockID := args[0].Int() + flags := args[1].Int() + + if flags&^(linux.TFD_CLOEXEC|linux.TFD_NONBLOCK) != 0 { + return 0, nil, syserror.EINVAL + } + + var fileFlags uint32 + if flags&linux.TFD_NONBLOCK != 0 { + fileFlags = linux.O_NONBLOCK + } + + var clock ktime.Clock + switch clockID { + case linux.CLOCK_REALTIME: + clock = t.Kernel().RealtimeClock() + case linux.CLOCK_MONOTONIC, linux.CLOCK_BOOTTIME: + clock = t.Kernel().MonotonicClock() + default: + return 0, nil, syserror.EINVAL + } + file, err := t.Kernel().VFS().NewTimerFD(clock, fileFlags) + if err != nil { + return 0, nil, err + } + defer file.DecRef() + fd, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{ + CloseOnExec: flags&linux.TFD_CLOEXEC != 0, + }) + if err != nil { + return 0, nil, err + } + return uintptr(fd), nil, nil +} + +// TimerfdSettime implements Linux syscall timerfd_settime(2). +func TimerfdSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + flags := args[1].Int() + newValAddr := args[2].Pointer() + oldValAddr := args[3].Pointer() + + if flags&^(linux.TFD_TIMER_ABSTIME) != 0 { + return 0, nil, syserror.EINVAL + } + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + tfd, ok := file.Impl().(*vfs.TimerFileDescription) + if !ok { + return 0, nil, syserror.EINVAL + } + + var newVal linux.Itimerspec + if _, err := t.CopyIn(newValAddr, &newVal); err != nil { + return 0, nil, err + } + newS, err := ktime.SettingFromItimerspec(newVal, flags&linux.TFD_TIMER_ABSTIME != 0, tfd.Clock()) + if err != nil { + return 0, nil, err + } + tm, oldS := tfd.SetTime(newS) + if oldValAddr != 0 { + oldVal := ktime.ItimerspecFromSetting(tm, oldS) + if _, err := t.CopyOut(oldValAddr, &oldVal); err != nil { + return 0, nil, err + } + } + return 0, nil, nil +} + +// TimerfdGettime implements Linux syscall timerfd_gettime(2). +func TimerfdGettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + curValAddr := args[1].Pointer() + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + tfd, ok := file.Impl().(*vfs.TimerFileDescription) + if !ok { + return 0, nil, syserror.EINVAL + } + + tm, s := tfd.GetTime() + curVal := ktime.ItimerspecFromSetting(tm, s) + _, err := t.CopyOut(curValAddr, &curVal) + return 0, nil, err +} diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index 9aeb83fb0..c62505fe2 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -25,6 +25,7 @@ go_library( "device.go", "epoll.go", "epoll_interest_list.go", + "eventfd.go", "file_description.go", "file_description_impl_util.go", "filesystem.go", @@ -44,6 +45,7 @@ go_library( "//pkg/abi/linux", "//pkg/context", "//pkg/fd", + "//pkg/fdnotifier", "//pkg/fspath", "//pkg/gohacks", "//pkg/log", @@ -68,6 +70,7 @@ go_test( name = "vfs_test", size = "small", srcs = [ + "eventfd_test.go", "file_description_impl_util_test.go", "mount_test.go", ], @@ -79,5 +82,6 @@ go_test( "//pkg/sync", "//pkg/syserror", "//pkg/usermem", + "//pkg/waiter", ], ) diff --git a/pkg/sentry/vfs/eventfd.go b/pkg/sentry/vfs/eventfd.go new file mode 100644 index 000000000..f39dacacf --- /dev/null +++ b/pkg/sentry/vfs/eventfd.go @@ -0,0 +1,282 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs + +import ( + "math" + "sync" + "syscall" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fdnotifier" + "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +// EventFileDescription implements FileDescriptionImpl for file-based event +// notification (eventfd). Eventfds are usually internal to the Sentry but in +// certain situations they may be converted into a host-backed eventfd. +type EventFileDescription struct { + vfsfd FileDescription + FileDescriptionDefaultImpl + DentryMetadataFileDescriptionImpl + + // queue is used to notify interested parties when the event object + // becomes readable or writable. + queue waiter.Queue `state:"zerovalue"` + + // mu protects the fields below. + mu sync.Mutex `state:"nosave"` + + // val is the current value of the event counter. + val uint64 + + // semMode specifies whether the event is in "semaphore" mode. + semMode bool + + // hostfd indicates whether this eventfd is passed through to the host. + hostfd int +} + +var _ FileDescriptionImpl = (*EventFileDescription)(nil) + +// NewEventFD creates a new event fd. +func (vfs *VirtualFilesystem) NewEventFD(initVal uint64, semMode bool, flags uint32) (*FileDescription, error) { + vd := vfs.NewAnonVirtualDentry("[eventfd]") + defer vd.DecRef() + efd := &EventFileDescription{ + val: initVal, + semMode: semMode, + hostfd: -1, + } + if err := efd.vfsfd.Init(efd, flags, vd.Mount(), vd.Dentry(), &FileDescriptionOptions{ + UseDentryMetadata: true, + DenyPRead: true, + DenyPWrite: true, + }); err != nil { + return nil, err + } + return &efd.vfsfd, nil +} + +// HostFD returns the host eventfd associated with this event. +func (efd *EventFileDescription) HostFD() (int, error) { + efd.mu.Lock() + defer efd.mu.Unlock() + if efd.hostfd >= 0 { + return efd.hostfd, nil + } + + flags := linux.EFD_NONBLOCK + if efd.semMode { + flags |= linux.EFD_SEMAPHORE + } + + fd, _, errno := syscall.Syscall(syscall.SYS_EVENTFD2, uintptr(efd.val), uintptr(flags), 0) + if errno != 0 { + return -1, errno + } + + if err := fdnotifier.AddFD(int32(fd), &efd.queue); err != nil { + if closeErr := syscall.Close(int(fd)); closeErr != nil { + log.Warningf("close(%d) eventfd failed: %v", fd, closeErr) + } + return -1, err + } + + efd.hostfd = int(fd) + return efd.hostfd, nil +} + +// Release implements FileDescriptionImpl.Release() +func (efd *EventFileDescription) Release() { + efd.mu.Lock() + defer efd.mu.Unlock() + if efd.hostfd >= 0 { + fdnotifier.RemoveFD(int32(efd.hostfd)) + if closeErr := syscall.Close(int(efd.hostfd)); closeErr != nil { + log.Warningf("close(%d) eventfd failed: %v", efd.hostfd, closeErr) + } + efd.hostfd = -1 + } +} + +// Read implements FileDescriptionImpl.Read. +func (efd *EventFileDescription) Read(ctx context.Context, dst usermem.IOSequence, _ ReadOptions) (int64, error) { + if dst.NumBytes() < 8 { + return 0, syscall.EINVAL + } + if err := efd.read(ctx, dst); err != nil { + return 0, err + } + return 8, nil +} + +// Write implements FileDescriptionImpl.Write. +func (efd *EventFileDescription) Write(ctx context.Context, src usermem.IOSequence, _ WriteOptions) (int64, error) { + if src.NumBytes() < 8 { + return 0, syscall.EINVAL + } + if err := efd.write(ctx, src); err != nil { + return 0, err + } + return 8, nil +} + +// Preconditions: Must be called with efd.mu locked. +func (efd *EventFileDescription) hostReadLocked(ctx context.Context, dst usermem.IOSequence) error { + var buf [8]byte + if _, err := syscall.Read(efd.hostfd, buf[:]); err != nil { + if err == syscall.EWOULDBLOCK { + return syserror.ErrWouldBlock + } + return err + } + _, err := dst.CopyOut(ctx, buf[:]) + return err +} + +func (efd *EventFileDescription) read(ctx context.Context, dst usermem.IOSequence) error { + efd.mu.Lock() + if efd.hostfd >= 0 { + defer efd.mu.Unlock() + return efd.hostReadLocked(ctx, dst) + } + + // We can't complete the read if the value is currently zero. + if efd.val == 0 { + efd.mu.Unlock() + return syserror.ErrWouldBlock + } + + // Update the value based on the mode the event is operating in. + var val uint64 + if efd.semMode { + val = 1 + // Consistent with Linux, this is done even if writing to memory fails. + efd.val-- + } else { + val = efd.val + efd.val = 0 + } + + efd.mu.Unlock() + + // Notify writers. We do this even if we were already writable because + // it is possible that a writer is waiting to write the maximum value + // to the event. + efd.queue.Notify(waiter.EventOut) + + var buf [8]byte + usermem.ByteOrder.PutUint64(buf[:], val) + _, err := dst.CopyOut(ctx, buf[:]) + return err +} + +// Preconditions: Must be called with efd.mu locked. +func (efd *EventFileDescription) hostWriteLocked(val uint64) error { + var buf [8]byte + usermem.ByteOrder.PutUint64(buf[:], val) + _, err := syscall.Write(efd.hostfd, buf[:]) + if err == syscall.EWOULDBLOCK { + return syserror.ErrWouldBlock + } + return err +} + +func (efd *EventFileDescription) write(ctx context.Context, src usermem.IOSequence) error { + var buf [8]byte + if _, err := src.CopyIn(ctx, buf[:]); err != nil { + return err + } + val := usermem.ByteOrder.Uint64(buf[:]) + + return efd.Signal(val) +} + +// Signal is an internal function to signal the event fd. +func (efd *EventFileDescription) Signal(val uint64) error { + if val == math.MaxUint64 { + return syscall.EINVAL + } + + efd.mu.Lock() + + if efd.hostfd >= 0 { + defer efd.mu.Unlock() + return efd.hostWriteLocked(val) + } + + // We only allow writes that won't cause the value to go over the max + // uint64 minus 1. + if val > math.MaxUint64-1-efd.val { + efd.mu.Unlock() + return syserror.ErrWouldBlock + } + + efd.val += val + efd.mu.Unlock() + + // Always trigger a notification. + efd.queue.Notify(waiter.EventIn) + + return nil +} + +// Readiness implements waiter.Waitable.Readiness. +func (efd *EventFileDescription) Readiness(mask waiter.EventMask) waiter.EventMask { + efd.mu.Lock() + defer efd.mu.Unlock() + + if efd.hostfd >= 0 { + return fdnotifier.NonBlockingPoll(int32(efd.hostfd), mask) + } + + ready := waiter.EventMask(0) + if efd.val > 0 { + ready |= waiter.EventIn + } + + if efd.val < math.MaxUint64-1 { + ready |= waiter.EventOut + } + + return mask & ready +} + +// EventRegister implements waiter.Waitable.EventRegister. +func (efd *EventFileDescription) EventRegister(entry *waiter.Entry, mask waiter.EventMask) { + efd.queue.EventRegister(entry, mask) + + efd.mu.Lock() + defer efd.mu.Unlock() + if efd.hostfd >= 0 { + fdnotifier.UpdateFD(int32(efd.hostfd)) + } +} + +// EventUnregister implements waiter.Waitable.EventUnregister. +func (efd *EventFileDescription) EventUnregister(entry *waiter.Entry) { + efd.queue.EventUnregister(entry) + + efd.mu.Lock() + defer efd.mu.Unlock() + if efd.hostfd >= 0 { + fdnotifier.UpdateFD(int32(efd.hostfd)) + } +} diff --git a/pkg/sentry/vfs/eventfd_test.go b/pkg/sentry/vfs/eventfd_test.go new file mode 100644 index 000000000..2dff2d10b --- /dev/null +++ b/pkg/sentry/vfs/eventfd_test.go @@ -0,0 +1,96 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs + +import ( + "testing" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/contexttest" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +func TestEventFD(t *testing.T) { + initVals := []uint64{ + 0, + // Using a non-zero initial value verifies that writing to an + // eventfd signals when the eventfd's counter was already + // non-zero. + 343, + } + + for _, initVal := range initVals { + ctx := contexttest.Context(t) + vfsObj := &VirtualFilesystem{} + if err := vfsObj.Init(); err != nil { + t.Fatalf("VFS init: %v", err) + } + + // Make a new eventfd that is writable. + eventfd, err := vfsObj.NewEventFD(initVal, false, linux.O_RDWR) + if err != nil { + t.Fatalf("NewEventFD failed: %v", err) + } + defer eventfd.DecRef() + + // Register a callback for a write event. + w, ch := waiter.NewChannelEntry(nil) + eventfd.EventRegister(&w, waiter.EventIn) + defer eventfd.EventUnregister(&w) + + data := []byte("00000124") + // Create and submit a write request. + n, err := eventfd.Write(ctx, usermem.BytesIOSequence(data), WriteOptions{}) + if err != nil { + t.Fatal(err) + } + if n != 8 { + t.Errorf("eventfd.write wrote %d bytes, not full int64", n) + } + + // Check if the callback fired due to the write event. + select { + case <-ch: + default: + t.Errorf("Didn't get notified of EventIn after write") + } + } +} + +func TestEventFDStat(t *testing.T) { + ctx := contexttest.Context(t) + vfsObj := &VirtualFilesystem{} + if err := vfsObj.Init(); err != nil { + t.Fatalf("VFS init: %v", err) + } + + // Make a new eventfd that is writable. + eventfd, err := vfsObj.NewEventFD(0, false, linux.O_RDWR) + if err != nil { + t.Fatalf("NewEventFD failed: %v", err) + } + defer eventfd.DecRef() + + statx, err := eventfd.Stat(ctx, StatOptions{ + Mask: linux.STATX_BASIC_STATS, + }) + if err != nil { + t.Fatalf("eventfd.Stat failed: %v", err) + } + if statx.Size != 0 { + t.Errorf("eventfd size should be 0") + } +} diff --git a/test/syscalls/linux/eventfd.cc b/test/syscalls/linux/eventfd.cc index 927001eee..548b05a64 100644 --- a/test/syscalls/linux/eventfd.cc +++ b/test/syscalls/linux/eventfd.cc @@ -100,6 +100,22 @@ TEST(EventfdTest, SmallRead) { ASSERT_THAT(read(efd.get(), &l, 4), SyscallFailsWithErrno(EINVAL)); } +TEST(EventfdTest, PreadIllegalSeek) { + FileDescriptor efd = + ASSERT_NO_ERRNO_AND_VALUE(NewEventFD(0, EFD_NONBLOCK | EFD_SEMAPHORE)); + + uint64_t l = 0; + ASSERT_THAT(pread(efd.get(), &l, 4, 0), SyscallFailsWithErrno(ESPIPE)); +} + +TEST(EventfdTest, PwriteIllegalSeek) { + FileDescriptor efd = + ASSERT_NO_ERRNO_AND_VALUE(NewEventFD(0, EFD_NONBLOCK | EFD_SEMAPHORE)); + + uint64_t l = 0; + ASSERT_THAT(pwrite(efd.get(), &l, 4, 0), SyscallFailsWithErrno(ESPIPE)); +} + TEST(EventfdTest, BigWrite) { FileDescriptor efd = ASSERT_NO_ERRNO_AND_VALUE(NewEventFD(0, EFD_NONBLOCK | EFD_SEMAPHORE)); -- cgit v1.2.3 From 35951c3671f3d429399eb581ad9da3b56e2a5f5a Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Mon, 4 May 2020 18:00:56 -0700 Subject: Translate p9.NoUID/GID to OverflowUID/GID. p9.NoUID/GID (== uint32(-1) == auth.NoID) is not a valid auth.KUID/KGID; in particular, using it for file ownership causes capabilities to be ineffective since file capabilities require that the file's KUID and KGID are mapped into the capability holder's user namespace [1], and auth.NoID is not mapped into any user namespace. Map p9.NoUID/GID to a different, valid KUID/KGID; in the unlikely case that an application actually using the overflow KUID/KGID attempts an operation that is consequently permitted by client permission checks, the remote operation will still fail with EPERM. Since this changes the VFS2 gofer client to no longer ignore the invalid IDs entirely, this CL both permits and requires that we change synthetic mount point creation to use root credentials. [1] See fs.Inode.CheckCapability or vfs.GenericCheckPermissions. PiperOrigin-RevId: 309856455 --- pkg/sentry/fs/gofer/attr.go | 12 ++++++++++-- pkg/sentry/fsimpl/gofer/gofer.go | 26 ++++++++++++++++++++------ 2 files changed, 30 insertions(+), 8 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/gofer/attr.go b/pkg/sentry/fs/gofer/attr.go index 6db4b762d..d481baf77 100644 --- a/pkg/sentry/fs/gofer/attr.go +++ b/pkg/sentry/fs/gofer/attr.go @@ -75,10 +75,18 @@ func owner(mounter fs.FileOwner, valid p9.AttrMask, pattr p9.Attr) fs.FileOwner // task's EUID/EGID. owner := mounter if valid.UID { - owner.UID = auth.KUID(pattr.UID) + if pattr.UID.Ok() { + owner.UID = auth.KUID(pattr.UID) + } else { + owner.UID = auth.KUID(auth.OverflowUID) + } } if valid.GID { - owner.GID = auth.KGID(pattr.GID) + if pattr.GID.Ok() { + owner.GID = auth.KGID(pattr.GID) + } else { + owner.GID = auth.KGID(auth.OverflowGID) + } } return owner } diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 2881c7bdd..1d9caf127 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -663,11 +663,11 @@ func (fs *filesystem) newDentry(ctx context.Context, file p9file, qid p9.QID, ma }, } d.pf.dentry = d - if mask.UID && attr.UID != auth.NoID { - d.uid = uint32(attr.UID) + if mask.UID { + d.uid = dentryUIDFromP9UID(attr.UID) } - if mask.GID && attr.GID != auth.NoID { - d.gid = uint32(attr.GID) + if mask.GID { + d.gid = dentryGIDFromP9GID(attr.GID) } if mask.Size { d.size = attr.Size @@ -718,10 +718,10 @@ func (d *dentry) updateFromP9Attrs(mask p9.AttrMask, attr *p9.Attr) { atomic.StoreUint32(&d.mode, uint32(attr.Mode)) } if mask.UID { - atomic.StoreUint32(&d.uid, uint32(attr.UID)) + atomic.StoreUint32(&d.uid, dentryUIDFromP9UID(attr.UID)) } if mask.GID { - atomic.StoreUint32(&d.gid, uint32(attr.GID)) + atomic.StoreUint32(&d.gid, dentryGIDFromP9GID(attr.GID)) } // There is no P9_GETATTR_* bit for I/O block size. if attr.BlockSize != 0 { @@ -939,6 +939,20 @@ func (d *dentry) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes) return vfs.GenericCheckPermissions(creds, ats, linux.FileMode(atomic.LoadUint32(&d.mode)), auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))) } +func dentryUIDFromP9UID(uid p9.UID) uint32 { + if !uid.Ok() { + return uint32(auth.OverflowUID) + } + return uint32(uid) +} + +func dentryGIDFromP9GID(gid p9.GID) uint32 { + if !gid.Ok() { + return uint32(auth.OverflowGID) + } + return uint32(gid) +} + // IncRef implements vfs.DentryImpl.IncRef. func (d *dentry) IncRef() { // d.refs may be 0 if d.fs.renameMu is locked, which serializes against -- cgit v1.2.3 From b3bd41434c17a95a87d67490f2b9bfd71e1ad705 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Tue, 5 May 2020 09:18:21 -0700 Subject: Return correct name for imported host files Implement PrependPath() in host.filesystem to correctly format name for host files. Updates #1672 PiperOrigin-RevId: 309959135 --- pkg/sentry/fsimpl/devpts/devpts.go | 2 +- pkg/sentry/fsimpl/host/BUILD | 1 + pkg/sentry/fsimpl/host/host.go | 68 ++++++++++++++++++--------------- pkg/sentry/fsimpl/kernfs/kernfs.go | 12 +++--- pkg/sentry/fsimpl/kernfs/kernfs_test.go | 2 +- pkg/sentry/fsimpl/pipefs/pipefs.go | 28 ++++++-------- pkg/sentry/fsimpl/sockfs/sockfs.go | 9 +---- pkg/sentry/fsimpl/sys/sys.go | 2 +- 8 files changed, 60 insertions(+), 64 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/devpts/devpts.go b/pkg/sentry/fsimpl/devpts/devpts.go index 181d765d3..94db8fe5c 100644 --- a/pkg/sentry/fsimpl/devpts/devpts.go +++ b/pkg/sentry/fsimpl/devpts/devpts.go @@ -59,7 +59,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt // master inode. It returns the filesystem and root Dentry. func (fstype FilesystemType) newFilesystem(vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials) (*kernfs.Filesystem, *kernfs.Dentry) { fs := &kernfs.Filesystem{} - fs.Init(vfsObj, fstype) + fs.VFSFilesystem().Init(vfsObj, fstype, fs) // Construct the root directory. This is always inode id 1. root := &rootInode{ diff --git a/pkg/sentry/fsimpl/host/BUILD b/pkg/sentry/fsimpl/host/BUILD index e1c56d89b..39509f703 100644 --- a/pkg/sentry/fsimpl/host/BUILD +++ b/pkg/sentry/fsimpl/host/BUILD @@ -20,6 +20,7 @@ go_library( "//pkg/abi/linux", "//pkg/context", "//pkg/fdnotifier", + "//pkg/fspath", "//pkg/log", "//pkg/refs", "//pkg/sentry/arch", diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 2be498afc..144e04905 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -25,6 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fdnotifier" + "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" @@ -39,37 +40,9 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) -// filesystemType implements vfs.FilesystemType. -type filesystemType struct{} - -// GetFilesystem implements FilesystemType.GetFilesystem. -func (filesystemType) GetFilesystem(context.Context, *vfs.VirtualFilesystem, *auth.Credentials, string, vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { - panic("host.filesystemType.GetFilesystem should never be called") -} - -// Name implements FilesystemType.Name. -func (filesystemType) Name() string { - return "none" -} - -// filesystem implements vfs.FilesystemImpl. -type filesystem struct { - kernfs.Filesystem -} - -// NewFilesystem sets up and returns a new hostfs filesystem. -// -// Note that there should only ever be one instance of host.filesystem, -// a global mount for host fds. -func NewFilesystem(vfsObj *vfs.VirtualFilesystem) *vfs.Filesystem { - fs := &filesystem{} - fs.Init(vfsObj, filesystemType{}) - return fs.VFSFilesystem() -} - // ImportFD sets up and returns a vfs.FileDescription from a donated fd. func ImportFD(ctx context.Context, mnt *vfs.Mount, hostFD int, isTTY bool) (*vfs.FileDescription, error) { - fs, ok := mnt.Filesystem().Impl().(*kernfs.Filesystem) + fs, ok := mnt.Filesystem().Impl().(*filesystem) if !ok { return nil, fmt.Errorf("can't import host FDs into filesystems of type %T", mnt.Filesystem().Impl()) } @@ -119,12 +92,47 @@ func ImportFD(ctx context.Context, mnt *vfs.Mount, hostFD int, isTTY bool) (*vfs d := &kernfs.Dentry{} d.Init(i) + // i.open will take a reference on d. defer d.DecRef() - return i.open(ctx, d.VFSDentry(), mnt) } +// filesystemType implements vfs.FilesystemType. +type filesystemType struct{} + +// GetFilesystem implements FilesystemType.GetFilesystem. +func (filesystemType) GetFilesystem(context.Context, *vfs.VirtualFilesystem, *auth.Credentials, string, vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { + panic("host.filesystemType.GetFilesystem should never be called") +} + +// Name implements FilesystemType.Name. +func (filesystemType) Name() string { + return "none" +} + +// NewFilesystem sets up and returns a new hostfs filesystem. +// +// Note that there should only ever be one instance of host.filesystem, +// a global mount for host fds. +func NewFilesystem(vfsObj *vfs.VirtualFilesystem) *vfs.Filesystem { + fs := &filesystem{} + fs.VFSFilesystem().Init(vfsObj, filesystemType{}, fs) + return fs.VFSFilesystem() +} + +// filesystem implements vfs.FilesystemImpl. +type filesystem struct { + kernfs.Filesystem +} + +func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { + d := vd.Dentry().Impl().(*kernfs.Dentry) + inode := d.Inode().(*inode) + b.PrependComponent(fmt.Sprintf("host:[%d]", inode.ino)) + return vfs.PrependPathSyntheticError{} +} + // inode implements kernfs.Inode. type inode struct { kernfs.InodeNotDirectory diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index 732837933..a83151ad3 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -132,13 +132,6 @@ func (fs *Filesystem) processDeferredDecRefsLocked() { fs.droppedDentriesMu.Unlock() } -// Init initializes a kernfs filesystem. This should be called from during -// vfs.FilesystemType.NewFilesystem for the concrete filesystem embedding -// kernfs. -func (fs *Filesystem) Init(vfsObj *vfs.VirtualFilesystem, fsType vfs.FilesystemType) { - fs.vfsfs.Init(vfsObj, fsType, fs) -} - // VFSFilesystem returns the generic vfs filesystem object. func (fs *Filesystem) VFSFilesystem() *vfs.Filesystem { return &fs.vfsfs @@ -261,6 +254,11 @@ func (d *Dentry) insertChildLocked(name string, child *Dentry) { d.children[name] = child } +// Inode returns the dentry's inode. +func (d *Dentry) Inode() Inode { + return d.inode +} + // The Inode interface maps filesystem-level operations that operate on paths to // equivalent operations on specific filesystem nodes. // diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index a9f671bc8..1c5d3e7e7 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -195,7 +195,7 @@ func (fsType) Name() string { func (fst fsType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opt vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { fs := &filesystem{} - fs.Init(vfsObj, &fst) + fs.VFSFilesystem().Init(vfsObj, &fst, fs) root := fst.rootFn(creds, fs) return fs.VFSFilesystem(), root.VFSDentry(), nil } diff --git a/pkg/sentry/fsimpl/pipefs/pipefs.go b/pkg/sentry/fsimpl/pipefs/pipefs.go index d6bd67467..5375e5e75 100644 --- a/pkg/sentry/fsimpl/pipefs/pipefs.go +++ b/pkg/sentry/fsimpl/pipefs/pipefs.go @@ -40,25 +40,19 @@ func (filesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFile panic("pipefs.filesystemType.GetFilesystem should never be called") } -// filesystem implements vfs.FilesystemImpl. -type filesystem struct { - kernfs.Filesystem - - // TODO(gvisor.dev/issue/1193): - // - // - kernfs does not provide a way to implement statfs, from which we - // should indicate PIPEFS_MAGIC. - // - // - kernfs does not provide a way to override names for - // vfs.FilesystemImpl.PrependPath(); pipefs inodes should use synthetic - // name fmt.Sprintf("pipe:[%d]", inode.ino). -} +// TODO(gvisor.dev/issue/1193): +// +// - kernfs does not provide a way to implement statfs, from which we +// should indicate PIPEFS_MAGIC. +// +// - kernfs does not provide a way to override names for +// vfs.FilesystemImpl.PrependPath(); pipefs inodes should use synthetic +// name fmt.Sprintf("pipe:[%d]", inode.ino). -// NewFilesystem sets up and returns a new vfs.Filesystem implemented by -// pipefs. +// NewFilesystem sets up and returns a new vfs.Filesystem implemented by pipefs. func NewFilesystem(vfsObj *vfs.VirtualFilesystem) *vfs.Filesystem { - fs := &filesystem{} - fs.Init(vfsObj, filesystemType{}) + fs := &kernfs.Filesystem{} + fs.VFSFilesystem().Init(vfsObj, filesystemType{}, fs) return fs.VFSFilesystem() } diff --git a/pkg/sentry/fsimpl/sockfs/sockfs.go b/pkg/sentry/fsimpl/sockfs/sockfs.go index dac2389fc..3f085d3ca 100644 --- a/pkg/sentry/fsimpl/sockfs/sockfs.go +++ b/pkg/sentry/fsimpl/sockfs/sockfs.go @@ -41,18 +41,13 @@ func (filesystemType) Name() string { return "sockfs" } -// filesystem implements vfs.FilesystemImpl. -type filesystem struct { - kernfs.Filesystem -} - // NewFilesystem sets up and returns a new sockfs filesystem. // // Note that there should only ever be one instance of sockfs.Filesystem, // backing a global socket mount. func NewFilesystem(vfsObj *vfs.VirtualFilesystem) *vfs.Filesystem { - fs := &filesystem{} - fs.Init(vfsObj, filesystemType{}) + fs := &kernfs.Filesystem{} + fs.VFSFilesystem().Init(vfsObj, filesystemType{}, fs) return fs.VFSFilesystem() } diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go index f8d25d35e..00f7d6214 100644 --- a/pkg/sentry/fsimpl/sys/sys.go +++ b/pkg/sentry/fsimpl/sys/sys.go @@ -47,7 +47,7 @@ func (FilesystemType) Name() string { // GetFilesystem implements vfs.FilesystemType.GetFilesystem. func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { fs := &filesystem{} - fs.Filesystem.Init(vfsObj, &fsType) + fs.VFSFilesystem().Init(vfsObj, &fsType, fs) k := kernel.KernelFromContext(ctx) maxCPUCores := k.ApplicationCores() defaultSysDirMode := linux.FileMode(0755) -- cgit v1.2.3 From a6dbf9596de58f1a264c236bf5afb8dfcfe78174 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 5 May 2020 10:00:02 -0700 Subject: Update comments for synthetic gofer files in vfs2. PiperOrigin-RevId: 309966538 --- pkg/sentry/fsimpl/gofer/filesystem.go | 2 ++ pkg/sentry/fsimpl/gofer/gofer.go | 6 ++---- 2 files changed, 4 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 4a8411371..4a32821bd 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -686,6 +686,8 @@ func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v return fs.doCreateAt(ctx, rp, false /* dir */, func(parent *dentry, name string) error { creds := rp.Credentials() _, err := parent.file.mknod(ctx, name, (p9.FileMode)(opts.Mode), opts.DevMajor, opts.DevMinor, (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)) + // If the gofer does not allow creating a socket or pipe, create a + // synthetic one, i.e. one that is kept entirely in memory. if err == syserror.EPERM { switch opts.Mode.FileType() { case linux.S_IFSOCK: diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 1d9caf127..9ab8fdc65 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -496,10 +496,8 @@ type dentry struct { // file is the unopened p9.File that backs this dentry. file is immutable. // // If file.isNil(), this dentry represents a synthetic file, i.e. a file - // that does not exist on the remote filesystem. As of this writing, this - // is only possible for a directory created with - // MkdirOptions.ForSyntheticMountpoint == true. - // TODO(gvisor.dev/issue/1476): Support synthetic sockets (and pipes). + // that does not exist on the remote filesystem. As of this writing, the + // only files that can be synthetic are sockets, pipes, and directories. file p9file // If deleted is non-zero, the file represented by this dentry has been -- cgit v1.2.3 From faf89dd31a44b8409b32919d7193834e194ecc56 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 5 May 2020 12:09:39 -0700 Subject: Update vfs2 socket TODOs. Three updates: - Mark all vfs2 socket syscalls as supported. - Use the same dev number and ino number generator for all types of sockets, unlike in VFS1. - Do not use host fd for hostinet metadata. Fixes #1476, #1478, #1484, 1485, #2017. PiperOrigin-RevId: 309994579 --- pkg/sentry/fsimpl/sockfs/sockfs.go | 9 +---- pkg/sentry/socket/hostinet/socket_vfs2.go | 7 ++-- pkg/sentry/socket/netstack/netstack_vfs2.go | 3 ++ pkg/sentry/socket/unix/unix_vfs2.go | 3 ++ .../syscalls/linux/vfs2/linux64_override_amd64.go | 40 ++++++++++------------ 5 files changed, 30 insertions(+), 32 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/sockfs/sockfs.go b/pkg/sentry/fsimpl/sockfs/sockfs.go index 3f085d3ca..239a9f4b4 100644 --- a/pkg/sentry/fsimpl/sockfs/sockfs.go +++ b/pkg/sentry/fsimpl/sockfs/sockfs.go @@ -53,9 +53,7 @@ func NewFilesystem(vfsObj *vfs.VirtualFilesystem) *vfs.Filesystem { // inode implements kernfs.Inode. // -// TODO(gvisor.dev/issue/1476): Add device numbers to this inode (which are -// not included in InodeAttrs) to store the numbers of the appropriate -// socket device. Override InodeAttrs.Stat() accordingly. +// TODO(gvisor.dev/issue/1193): Device numbers. type inode struct { kernfs.InodeNotDirectory kernfs.InodeNotSymlink @@ -69,11 +67,6 @@ func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentr } // NewDentry constructs and returns a sockfs dentry. -// -// TODO(gvisor.dev/issue/1476): Currently, we are using -// sockfs.filesystem.NextIno() to get inode numbers. We should use -// device-specific numbers, so that we are not using the same generator for -// netstack, unix, etc. func NewDentry(creds *auth.Credentials, ino uint64) *vfs.Dentry { // File mode matches net/socket.c:sock_alloc. filemode := linux.FileMode(linux.S_IFSOCK | 0600) diff --git a/pkg/sentry/socket/hostinet/socket_vfs2.go b/pkg/sentry/socket/hostinet/socket_vfs2.go index b03ca2f26..a8278bffc 100644 --- a/pkg/sentry/socket/hostinet/socket_vfs2.go +++ b/pkg/sentry/socket/hostinet/socket_vfs2.go @@ -36,8 +36,11 @@ import ( type socketVFS2 struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl - // TODO(gvisor.dev/issue/1484): VFS1 stores internal metadata for hostinet. - // We should perhaps rely on the host, much like in hostfs. + + // We store metadata for hostinet sockets internally. Technically, we should + // access metadata (e.g. through stat, chmod) on the host for correctness, + // but this is not very useful for inet socket fds, which do not belong to a + // concrete file anyway. vfs.DentryMetadataFileDescriptionImpl socketOpsCommon diff --git a/pkg/sentry/socket/netstack/netstack_vfs2.go b/pkg/sentry/socket/netstack/netstack_vfs2.go index eec71035d..f7d9b2ff4 100644 --- a/pkg/sentry/socket/netstack/netstack_vfs2.go +++ b/pkg/sentry/socket/netstack/netstack_vfs2.go @@ -22,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/netfilter" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserr" @@ -41,6 +42,8 @@ type SocketVFS2 struct { socketOpsCommon } +var _ = socket.SocketVFS2(&SocketVFS2{}) + // NewVFS2 creates a new endpoint socket. func NewVFS2(t *kernel.Task, family int, skType linux.SockType, protocol int, queue *waiter.Queue, endpoint tcpip.Endpoint) (*vfs.FileDescription, *syserr.Error) { if skType == linux.SOCK_STREAM { diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go index 5edc3cdf4..06d838868 100644 --- a/pkg/sentry/socket/unix/unix_vfs2.go +++ b/pkg/sentry/socket/unix/unix_vfs2.go @@ -22,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/control" "gvisor.dev/gvisor/pkg/sentry/socket/netstack" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" @@ -43,6 +44,8 @@ type SocketVFS2 struct { socketOpsCommon } +var _ = socket.SocketVFS2(&SocketVFS2{}) + // NewSockfsFile creates a new socket file in the global sockfs mount and // returns a corresponding file description. func NewSockfsFile(t *kernel.Task, ep transport.Endpoint, stype linux.SockType) (*vfs.FileDescription, *syserr.Error) { diff --git a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go index 074f58e5d..47c5d18e7 100644 --- a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go +++ b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go @@ -44,22 +44,21 @@ func Override(table map[uintptr]kernel.Syscall) { table[32] = syscalls.Supported("dup", Dup) table[33] = syscalls.Supported("dup2", Dup2) delete(table, 40) // sendfile - // TODO(gvisor.dev/issue/1485): Port all socket variants to VFS2. - table[41] = syscalls.PartiallySupported("socket", Socket, "In process of porting socket syscalls to VFS2.", nil) - table[42] = syscalls.PartiallySupported("connect", Connect, "In process of porting socket syscalls to VFS2.", nil) - table[43] = syscalls.PartiallySupported("accept", Accept, "In process of porting socket syscalls to VFS2.", nil) - table[44] = syscalls.PartiallySupported("sendto", SendTo, "In process of porting socket syscalls to VFS2.", nil) - table[45] = syscalls.PartiallySupported("recvfrom", RecvFrom, "In process of porting socket syscalls to VFS2.", nil) - table[46] = syscalls.PartiallySupported("sendmsg", SendMsg, "In process of porting socket syscalls to VFS2.", nil) - table[47] = syscalls.PartiallySupported("recvmsg", RecvMsg, "In process of porting socket syscalls to VFS2.", nil) - table[48] = syscalls.PartiallySupported("shutdown", Shutdown, "In process of porting socket syscalls to VFS2.", nil) - table[49] = syscalls.PartiallySupported("bind", Bind, "In process of porting socket syscalls to VFS2.", nil) - table[50] = syscalls.PartiallySupported("listen", Listen, "In process of porting socket syscalls to VFS2.", nil) - table[51] = syscalls.PartiallySupported("getsockname", GetSockName, "In process of porting socket syscalls to VFS2.", nil) - table[52] = syscalls.PartiallySupported("getpeername", GetPeerName, "In process of porting socket syscalls to VFS2.", nil) - table[53] = syscalls.PartiallySupported("socketpair", SocketPair, "In process of porting socket syscalls to VFS2.", nil) - table[54] = syscalls.PartiallySupported("setsockopt", SetSockOpt, "In process of porting socket syscalls to VFS2.", nil) - table[55] = syscalls.PartiallySupported("getsockopt", GetSockOpt, "In process of porting socket syscalls to VFS2.", nil) + table[41] = syscalls.Supported("socket", Socket) + table[42] = syscalls.Supported("connect", Connect) + table[43] = syscalls.Supported("accept", Accept) + table[44] = syscalls.Supported("sendto", SendTo) + table[45] = syscalls.Supported("recvfrom", RecvFrom) + table[46] = syscalls.Supported("sendmsg", SendMsg) + table[47] = syscalls.Supported("recvmsg", RecvMsg) + table[48] = syscalls.Supported("shutdown", Shutdown) + table[49] = syscalls.Supported("bind", Bind) + table[50] = syscalls.Supported("listen", Listen) + table[51] = syscalls.Supported("getsockname", GetSockName) + table[52] = syscalls.Supported("getpeername", GetPeerName) + table[53] = syscalls.Supported("socketpair", SocketPair) + table[54] = syscalls.Supported("setsockopt", SetSockOpt) + table[55] = syscalls.Supported("getsockopt", GetSockOpt) table[59] = syscalls.Supported("execve", Execve) table[72] = syscalls.Supported("fcntl", Fcntl) delete(table, 73) // flock @@ -145,8 +144,7 @@ func Override(table map[uintptr]kernel.Syscall) { delete(table, 285) // fallocate table[286] = syscalls.Supported("timerfd_settime", TimerfdSettime) table[287] = syscalls.Supported("timerfd_gettime", TimerfdGettime) - // TODO(gvisor.dev/issue/1485): Port all socket variants to VFS2. - table[288] = syscalls.PartiallySupported("accept4", Accept4, "In process of porting socket syscalls to VFS2.", nil) + table[288] = syscalls.Supported("accept4", Accept4) delete(table, 289) // signalfd4 table[290] = syscalls.Supported("eventfd2", Eventfd2) table[291] = syscalls.Supported("epoll_create1", EpollCreate1) @@ -155,11 +153,9 @@ func Override(table map[uintptr]kernel.Syscall) { delete(table, 294) // inotify_init1 table[295] = syscalls.Supported("preadv", Preadv) table[296] = syscalls.Supported("pwritev", Pwritev) - // TODO(gvisor.dev/issue/1485): Port all socket variants to VFS2. - table[299] = syscalls.PartiallySupported("recvmmsg", RecvMMsg, "In process of porting socket syscalls to VFS2.", nil) + table[299] = syscalls.Supported("recvmmsg", RecvMMsg) table[306] = syscalls.Supported("syncfs", Syncfs) - // TODO(gvisor.dev/issue/1485): Port all socket variants to VFS2. - table[307] = syscalls.PartiallySupported("sendmmsg", SendMMsg, "In process of porting socket syscalls to VFS2.", nil) + table[307] = syscalls.Supported("sendmmsg", SendMMsg) table[316] = syscalls.Supported("renameat2", Renameat2) delete(table, 319) // memfd_create table[322] = syscalls.Supported("execveat", Execveat) -- cgit v1.2.3 From 8416da33d25bf28cfcfdcbe658f696d6883eebcd Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Tue, 5 May 2020 17:57:22 -0700 Subject: Internal change. PiperOrigin-RevId: 310057834 --- pkg/sentry/pgalloc/pgalloc.go | 14 ++++++++++++++ 1 file changed, 14 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/pgalloc/pgalloc.go b/pkg/sentry/pgalloc/pgalloc.go index 577e9306a..2b11ea4ae 100644 --- a/pkg/sentry/pgalloc/pgalloc.go +++ b/pkg/sentry/pgalloc/pgalloc.go @@ -180,6 +180,11 @@ type MemoryFileOpts struct { // notifications to determine when eviction is necessary. This option has // no effect unless DelayedEviction is DelayedEvictionEnabled. UseHostMemcgPressure bool + + // If ManualZeroing is true, MemoryFile must not assume that new pages + // obtained from the host are zero-filled, such that MemoryFile must manually + // zero newly-allocated pages. + ManualZeroing bool } // DelayedEvictionType is the type of MemoryFileOpts.DelayedEviction. @@ -432,6 +437,15 @@ func (f *MemoryFile) Allocate(length uint64, kind usage.MemoryKind) (platform.Fi // Mark selected pages as in use. fr := platform.FileRange{start, end} + if f.opts.ManualZeroing { + if err := f.forEachMappingSlice(fr, func(bs []byte) { + for i := range bs { + bs[i] = 0 + } + }); err != nil { + return platform.FileRange{}, err + } + } if !f.usage.Add(fr, usageInfo{ kind: kind, refs: 1, -- cgit v1.2.3 From 591ff0e424e3b30eb143bce06618cb8656784b90 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Wed, 6 May 2020 10:28:50 -0700 Subject: Add maximum memory limit. PiperOrigin-RevId: 310179277 --- pkg/sentry/fs/proc/meminfo.go | 10 +++++++--- pkg/sentry/fsimpl/proc/tasks_files.go | 10 +++++++--- pkg/sentry/syscalls/linux/sys_sysinfo.go | 7 ++++++- pkg/sentry/usage/memory.go | 24 ++++++++++++++++-------- 4 files changed, 36 insertions(+), 15 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/proc/meminfo.go b/pkg/sentry/fs/proc/meminfo.go index 465b47da9..91617267d 100644 --- a/pkg/sentry/fs/proc/meminfo.go +++ b/pkg/sentry/fs/proc/meminfo.go @@ -58,12 +58,16 @@ func (d *meminfoData) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) var buf bytes.Buffer fmt.Fprintf(&buf, "MemTotal: %8d kB\n", totalSize/1024) - memFree := (totalSize - totalUsage) / 1024 + memFree := totalSize - totalUsage + if memFree > totalSize { + // Underflow. + memFree = 0 + } // We use MemFree as MemAvailable because we don't swap. // TODO(rahat): When reclaim is implemented the value of MemAvailable // should change. - fmt.Fprintf(&buf, "MemFree: %8d kB\n", memFree) - fmt.Fprintf(&buf, "MemAvailable: %8d kB\n", memFree) + fmt.Fprintf(&buf, "MemFree: %8d kB\n", memFree/1024) + fmt.Fprintf(&buf, "MemAvailable: %8d kB\n", memFree/1024) fmt.Fprintf(&buf, "Buffers: 0 kB\n") // memory usage by block devices fmt.Fprintf(&buf, "Cached: %8d kB\n", (file+snapshot.Tmpfs)/1024) // Emulate a system with no swap, which disables inactivation of anon pages. diff --git a/pkg/sentry/fsimpl/proc/tasks_files.go b/pkg/sentry/fsimpl/proc/tasks_files.go index 92007df81..e5f13b69e 100644 --- a/pkg/sentry/fsimpl/proc/tasks_files.go +++ b/pkg/sentry/fsimpl/proc/tasks_files.go @@ -272,12 +272,16 @@ func (*meminfoData) Generate(ctx context.Context, buf *bytes.Buffer) error { inactiveFile := file - activeFile fmt.Fprintf(buf, "MemTotal: %8d kB\n", totalSize/1024) - memFree := (totalSize - totalUsage) / 1024 + memFree := totalSize - totalUsage + if memFree > totalSize { + // Underflow. + memFree = 0 + } // We use MemFree as MemAvailable because we don't swap. // TODO(rahat): When reclaim is implemented the value of MemAvailable // should change. - fmt.Fprintf(buf, "MemFree: %8d kB\n", memFree) - fmt.Fprintf(buf, "MemAvailable: %8d kB\n", memFree) + fmt.Fprintf(buf, "MemFree: %8d kB\n", memFree/1024) + fmt.Fprintf(buf, "MemAvailable: %8d kB\n", memFree/1024) fmt.Fprintf(buf, "Buffers: 0 kB\n") // memory usage by block devices fmt.Fprintf(buf, "Cached: %8d kB\n", (file+snapshot.Tmpfs)/1024) // Emulate a system with no swap, which disables inactivation of anon pages. diff --git a/pkg/sentry/syscalls/linux/sys_sysinfo.go b/pkg/sentry/syscalls/linux/sys_sysinfo.go index a65b560c8..297de052a 100644 --- a/pkg/sentry/syscalls/linux/sys_sysinfo.go +++ b/pkg/sentry/syscalls/linux/sys_sysinfo.go @@ -29,13 +29,18 @@ func Sysinfo(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca mf.UpdateUsage() _, totalUsage := usage.MemoryAccounting.Copy() totalSize := usage.TotalMemory(mf.TotalSize(), totalUsage) + memFree := totalSize - totalUsage + if memFree > totalSize { + // Underflow. + memFree = 0 + } // Only a subset of the fields in sysinfo_t make sense to return. si := linux.Sysinfo{ Procs: uint16(len(t.PIDNamespace().Tasks())), Uptime: t.Kernel().MonotonicClock().Now().Seconds(), TotalRAM: totalSize, - FreeRAM: totalSize - totalUsage, + FreeRAM: memFree, Unit: 1, } _, err := t.CopyOut(addr, si) diff --git a/pkg/sentry/usage/memory.go b/pkg/sentry/usage/memory.go index 4320ad17f..ab1d140d2 100644 --- a/pkg/sentry/usage/memory.go +++ b/pkg/sentry/usage/memory.go @@ -252,18 +252,23 @@ func (m *MemoryLocked) Copy() (MemoryStats, uint64) { return ms, m.totalLocked() } -// MinimumTotalMemoryBytes is the minimum reported total system memory. -// -// This can be configured through options provided to the Sentry at start. -// This number is purely synthetic. This is only set before the application -// starts executing, and must not be modified. -var MinimumTotalMemoryBytes uint64 = 2 << 30 // 2 GB +// These options control how much total memory the is reported to the application. +// They may only be set before the application starts executing, and must not +// be modified. +var ( + // MinimumTotalMemoryBytes is the minimum reported total system memory. + MinimumTotalMemoryBytes uint64 = 2 << 30 // 2 GB + + // MaximumTotalMemoryBytes is the maximum reported total system memory. + // The 0 value indicates no maximum. + MaximumTotalMemoryBytes uint64 +) // TotalMemory returns the "total usable memory" available. // // This number doesn't really have a true value so it's based on the following -// inputs and further bounded to be above some minimum guaranteed value (2GB), -// additionally ensuring that total memory reported is always less than used. +// inputs and further bounded to be above the MinumumTotalMemoryBytes and below +// MaximumTotalMemoryBytes. // // memSize should be the platform.Memory size reported by platform.Memory.TotalSize() // used is the total memory reported by MemoryLocked.Total() @@ -279,5 +284,8 @@ func TotalMemory(memSize, used uint64) uint64 { memSize = uint64(1) << (uint(msb) + 1) } } + if MaximumTotalMemoryBytes > 0 && memSize > MaximumTotalMemoryBytes { + memSize = MaximumTotalMemoryBytes + } return memSize } -- cgit v1.2.3 From 279f1eb7abb28966ef633fa61418bffad4a716b0 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Wed, 6 May 2020 14:12:08 -0700 Subject: Fix runsc syscall documentation generation. We can register any number of tables with any number of architectures, and need not limit the definitions to the architecture in question. This allows runsc to generate documentation for all architectures simultaneously. Similarly, this simplifies the VFSv2 patching process. PiperOrigin-RevId: 310224827 --- pkg/sentry/kernel/syscalls.go | 58 +- pkg/sentry/syscalls/linux/BUILD | 2 - pkg/sentry/syscalls/linux/linux64.go | 709 +++++++++++++++++++++ pkg/sentry/syscalls/linux/linux64_amd64.go | 406 ------------ pkg/sentry/syscalls/linux/linux64_arm64.go | 340 ---------- pkg/sentry/syscalls/linux/vfs2/BUILD | 4 +- pkg/sentry/syscalls/linux/vfs2/linux64.go | 16 - .../syscalls/linux/vfs2/linux64_override_amd64.go | 165 ----- .../syscalls/linux/vfs2/linux64_override_arm64.go | 27 - pkg/sentry/syscalls/linux/vfs2/vfs2.go | 172 +++++ runsc/boot/BUILD | 4 - runsc/boot/loader.go | 12 +- runsc/boot/loader_amd64.go | 26 - runsc/boot/loader_arm64.go | 26 - runsc/boot/loader_test.go | 12 - runsc/cmd/syscalls.go | 23 +- 16 files changed, 930 insertions(+), 1072 deletions(-) delete mode 100644 pkg/sentry/syscalls/linux/linux64_amd64.go delete mode 100644 pkg/sentry/syscalls/linux/linux64_arm64.go delete mode 100644 pkg/sentry/syscalls/linux/vfs2/linux64.go delete mode 100644 pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go delete mode 100644 pkg/sentry/syscalls/linux/vfs2/linux64_override_arm64.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/vfs2.go delete mode 100644 runsc/boot/loader_amd64.go delete mode 100644 runsc/boot/loader_arm64.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/syscalls.go b/pkg/sentry/kernel/syscalls.go index 84156d5a1..413111faf 100644 --- a/pkg/sentry/kernel/syscalls.go +++ b/pkg/sentry/kernel/syscalls.go @@ -29,7 +29,7 @@ import ( // // The types below create fast lookup slices for all syscalls. This maximum // serves as a sanity check that we don't allocate huge slices for a very large -// syscall. +// syscall. This is checked during registration. const maxSyscallNum = 2000 // SyscallSupportLevel is a syscall support levels. @@ -266,6 +266,16 @@ type SyscallTable struct { FeatureEnable SyscallFlagsTable } +// MaxSysno returns the largest system call number. +func (s *SyscallTable) MaxSysno() (max uintptr) { + for num := range s.Table { + if num > max { + max = num + } + } + return max +} + // allSyscallTables contains all known tables. var allSyscallTables []*SyscallTable @@ -286,6 +296,20 @@ func LookupSyscallTable(os abi.OS, a arch.Arch) (*SyscallTable, bool) { // RegisterSyscallTable registers a new syscall table for use by a Kernel. func RegisterSyscallTable(s *SyscallTable) { + if max := s.MaxSysno(); max > maxSyscallNum { + panic(fmt.Sprintf("SyscallTable %+v contains too large syscall number %d", s, max)) + } + if _, ok := LookupSyscallTable(s.OS, s.Arch); ok { + panic(fmt.Sprintf("Duplicate SyscallTable registered for OS %v Arch %v", s.OS, s.Arch)) + } + allSyscallTables = append(allSyscallTables, s) + s.Init() +} + +// Init initializes the system call table. +// +// This should normally be called only during registration. +func (s *SyscallTable) Init() { if s.Table == nil { // Ensure non-nil lookup table. s.Table = make(map[uintptr]Syscall) @@ -295,42 +319,16 @@ func RegisterSyscallTable(s *SyscallTable) { s.Emulate = make(map[usermem.Addr]uintptr) } - var max uintptr - for num := range s.Table { - if num > max { - max = num - } - } - - if max > maxSyscallNum { - panic(fmt.Sprintf("SyscallTable %+v contains too large syscall number %d", s, max)) - } - - s.lookup = make([]SyscallFn, max+1) + max := s.MaxSysno() // Checked during RegisterSyscallTable. // Initialize the fast-lookup table. + s.lookup = make([]SyscallFn, max+1) for num, sc := range s.Table { s.lookup[num] = sc.Fn } + // Initialize all features. s.FeatureEnable.init(s.Table, max) - - if _, ok := LookupSyscallTable(s.OS, s.Arch); ok { - panic(fmt.Sprintf("Duplicate SyscallTable registered for OS %v Arch %v", s.OS, s.Arch)) - } - - // Save a reference to this table. - // - // This is required for a Kernel to find the table and for save/restore - // operations below. - allSyscallTables = append(allSyscallTables, s) -} - -// FlushSyscallTablesTestOnly flushes the syscall tables for tests. Used for -// parameterized VFSv2 tests. -// TODO(gvisor.dv/issue/1624): Remove when VFS1 is no longer supported. -func FlushSyscallTablesTestOnly() { - allSyscallTables = nil } // Lookup returns the syscall implementation, if one exists. diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD index 0d24fd3c4..245e8fe1e 100644 --- a/pkg/sentry/syscalls/linux/BUILD +++ b/pkg/sentry/syscalls/linux/BUILD @@ -8,8 +8,6 @@ go_library( "error.go", "flags.go", "linux64.go", - "linux64_amd64.go", - "linux64_arm64.go", "sigset.go", "sys_aio.go", "sys_capability.go", diff --git a/pkg/sentry/syscalls/linux/linux64.go b/pkg/sentry/syscalls/linux/linux64.go index 68589a377..ea4f9b1a7 100644 --- a/pkg/sentry/syscalls/linux/linux64.go +++ b/pkg/sentry/syscalls/linux/linux64.go @@ -15,6 +15,16 @@ // Package linux provides syscall tables for amd64 Linux. package linux +import ( + "gvisor.dev/gvisor/pkg/abi" + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/syscalls" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + const ( // LinuxSysname is the OS name advertised by gVisor. LinuxSysname = "Linux" @@ -25,3 +35,702 @@ const ( // LinuxVersion is the version info advertised by gVisor. LinuxVersion = "#1 SMP Sun Jan 10 15:06:54 PST 2016" ) + +// AMD64 is a table of Linux amd64 syscall API with the corresponding syscall +// numbers from Linux 4.4. +var AMD64 = &kernel.SyscallTable{ + OS: abi.Linux, + Arch: arch.AMD64, + Version: kernel.Version{ + // Version 4.4 is chosen as a stable, longterm version of Linux, which + // guides the interface provided by this syscall table. The build + // version is that for a clean build with default kernel config, at 5 + // minutes after v4.4 was tagged. + Sysname: LinuxSysname, + Release: LinuxRelease, + Version: LinuxVersion, + }, + AuditNumber: linux.AUDIT_ARCH_X86_64, + Table: map[uintptr]kernel.Syscall{ + 0: syscalls.Supported("read", Read), + 1: syscalls.Supported("write", Write), + 2: syscalls.PartiallySupported("open", Open, "Options O_DIRECT, O_NOATIME, O_PATH, O_TMPFILE, O_SYNC are not supported.", nil), + 3: syscalls.Supported("close", Close), + 4: syscalls.Supported("stat", Stat), + 5: syscalls.Supported("fstat", Fstat), + 6: syscalls.Supported("lstat", Lstat), + 7: syscalls.Supported("poll", Poll), + 8: syscalls.Supported("lseek", Lseek), + 9: syscalls.PartiallySupported("mmap", Mmap, "Generally supported with exceptions. Options MAP_FIXED_NOREPLACE, MAP_SHARED_VALIDATE, MAP_SYNC MAP_GROWSDOWN, MAP_HUGETLB are not supported.", nil), + 10: syscalls.Supported("mprotect", Mprotect), + 11: syscalls.Supported("munmap", Munmap), + 12: syscalls.Supported("brk", Brk), + 13: syscalls.Supported("rt_sigaction", RtSigaction), + 14: syscalls.Supported("rt_sigprocmask", RtSigprocmask), + 15: syscalls.Supported("rt_sigreturn", RtSigreturn), + 16: syscalls.PartiallySupported("ioctl", Ioctl, "Only a few ioctls are implemented for backing devices and file systems.", nil), + 17: syscalls.Supported("pread64", Pread64), + 18: syscalls.Supported("pwrite64", Pwrite64), + 19: syscalls.Supported("readv", Readv), + 20: syscalls.Supported("writev", Writev), + 21: syscalls.Supported("access", Access), + 22: syscalls.Supported("pipe", Pipe), + 23: syscalls.Supported("select", Select), + 24: syscalls.Supported("sched_yield", SchedYield), + 25: syscalls.Supported("mremap", Mremap), + 26: syscalls.PartiallySupported("msync", Msync, "Full data flush is not guaranteed at this time.", nil), + 27: syscalls.PartiallySupported("mincore", Mincore, "Stub implementation. The sandbox does not have access to this information. Reports all mapped pages are resident.", nil), + 28: syscalls.PartiallySupported("madvise", Madvise, "Options MADV_DONTNEED, MADV_DONTFORK are supported. Other advice is ignored.", nil), + 29: syscalls.PartiallySupported("shmget", Shmget, "Option SHM_HUGETLB is not supported.", nil), + 30: syscalls.PartiallySupported("shmat", Shmat, "Option SHM_RND is not supported.", nil), + 31: syscalls.PartiallySupported("shmctl", Shmctl, "Options SHM_LOCK, SHM_UNLOCK are not supported.", nil), + 32: syscalls.Supported("dup", Dup), + 33: syscalls.Supported("dup2", Dup2), + 34: syscalls.Supported("pause", Pause), + 35: syscalls.Supported("nanosleep", Nanosleep), + 36: syscalls.Supported("getitimer", Getitimer), + 37: syscalls.Supported("alarm", Alarm), + 38: syscalls.Supported("setitimer", Setitimer), + 39: syscalls.Supported("getpid", Getpid), + 40: syscalls.Supported("sendfile", Sendfile), + 41: syscalls.PartiallySupported("socket", Socket, "Limited support for AF_NETLINK, NETLINK_ROUTE sockets. Limited support for SOCK_RAW.", nil), + 42: syscalls.Supported("connect", Connect), + 43: syscalls.Supported("accept", Accept), + 44: syscalls.Supported("sendto", SendTo), + 45: syscalls.Supported("recvfrom", RecvFrom), + 46: syscalls.Supported("sendmsg", SendMsg), + 47: syscalls.PartiallySupported("recvmsg", RecvMsg, "Not all flags and control messages are supported.", nil), + 48: syscalls.PartiallySupported("shutdown", Shutdown, "Not all flags and control messages are supported.", nil), + 49: syscalls.PartiallySupported("bind", Bind, "Autobind for abstract Unix sockets is not supported.", nil), + 50: syscalls.Supported("listen", Listen), + 51: syscalls.Supported("getsockname", GetSockName), + 52: syscalls.Supported("getpeername", GetPeerName), + 53: syscalls.Supported("socketpair", SocketPair), + 54: syscalls.PartiallySupported("setsockopt", SetSockOpt, "Not all socket options are supported.", nil), + 55: syscalls.PartiallySupported("getsockopt", GetSockOpt, "Not all socket options are supported.", nil), + 56: syscalls.PartiallySupported("clone", Clone, "Mount namespace (CLONE_NEWNS) not supported. Options CLONE_PARENT, CLONE_SYSVSEM not supported.", nil), + 57: syscalls.Supported("fork", Fork), + 58: syscalls.Supported("vfork", Vfork), + 59: syscalls.Supported("execve", Execve), + 60: syscalls.Supported("exit", Exit), + 61: syscalls.Supported("wait4", Wait4), + 62: syscalls.Supported("kill", Kill), + 63: syscalls.Supported("uname", Uname), + 64: syscalls.Supported("semget", Semget), + 65: syscalls.PartiallySupported("semop", Semop, "Option SEM_UNDO not supported.", nil), + 66: syscalls.PartiallySupported("semctl", Semctl, "Options IPC_INFO, SEM_INFO, IPC_STAT, SEM_STAT, SEM_STAT_ANY, GETNCNT, GETZCNT not supported.", nil), + 67: syscalls.Supported("shmdt", Shmdt), + 68: syscalls.ErrorWithEvent("msgget", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) + 69: syscalls.ErrorWithEvent("msgsnd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) + 70: syscalls.ErrorWithEvent("msgrcv", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) + 71: syscalls.ErrorWithEvent("msgctl", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) + 72: syscalls.PartiallySupported("fcntl", Fcntl, "Not all options are supported.", nil), + 73: syscalls.PartiallySupported("flock", Flock, "Locks are held within the sandbox only.", nil), + 74: syscalls.PartiallySupported("fsync", Fsync, "Full data flush is not guaranteed at this time.", nil), + 75: syscalls.PartiallySupported("fdatasync", Fdatasync, "Full data flush is not guaranteed at this time.", nil), + 76: syscalls.Supported("truncate", Truncate), + 77: syscalls.Supported("ftruncate", Ftruncate), + 78: syscalls.Supported("getdents", Getdents), + 79: syscalls.Supported("getcwd", Getcwd), + 80: syscalls.Supported("chdir", Chdir), + 81: syscalls.Supported("fchdir", Fchdir), + 82: syscalls.Supported("rename", Rename), + 83: syscalls.Supported("mkdir", Mkdir), + 84: syscalls.Supported("rmdir", Rmdir), + 85: syscalls.Supported("creat", Creat), + 86: syscalls.Supported("link", Link), + 87: syscalls.Supported("unlink", Unlink), + 88: syscalls.Supported("symlink", Symlink), + 89: syscalls.Supported("readlink", Readlink), + 90: syscalls.Supported("chmod", Chmod), + 91: syscalls.PartiallySupported("fchmod", Fchmod, "Options S_ISUID and S_ISGID not supported.", nil), + 92: syscalls.Supported("chown", Chown), + 93: syscalls.Supported("fchown", Fchown), + 94: syscalls.Supported("lchown", Lchown), + 95: syscalls.Supported("umask", Umask), + 96: syscalls.Supported("gettimeofday", Gettimeofday), + 97: syscalls.Supported("getrlimit", Getrlimit), + 98: syscalls.PartiallySupported("getrusage", Getrusage, "Fields ru_maxrss, ru_minflt, ru_majflt, ru_inblock, ru_oublock are not supported. Fields ru_utime and ru_stime have low precision.", nil), + 99: syscalls.PartiallySupported("sysinfo", Sysinfo, "Fields loads, sharedram, bufferram, totalswap, freeswap, totalhigh, freehigh not supported.", nil), + 100: syscalls.Supported("times", Times), + 101: syscalls.PartiallySupported("ptrace", Ptrace, "Options PTRACE_PEEKSIGINFO, PTRACE_SECCOMP_GET_FILTER not supported.", nil), + 102: syscalls.Supported("getuid", Getuid), + 103: syscalls.PartiallySupported("syslog", Syslog, "Outputs a dummy message for security reasons.", nil), + 104: syscalls.Supported("getgid", Getgid), + 105: syscalls.Supported("setuid", Setuid), + 106: syscalls.Supported("setgid", Setgid), + 107: syscalls.Supported("geteuid", Geteuid), + 108: syscalls.Supported("getegid", Getegid), + 109: syscalls.Supported("setpgid", Setpgid), + 110: syscalls.Supported("getppid", Getppid), + 111: syscalls.Supported("getpgrp", Getpgrp), + 112: syscalls.Supported("setsid", Setsid), + 113: syscalls.Supported("setreuid", Setreuid), + 114: syscalls.Supported("setregid", Setregid), + 115: syscalls.Supported("getgroups", Getgroups), + 116: syscalls.Supported("setgroups", Setgroups), + 117: syscalls.Supported("setresuid", Setresuid), + 118: syscalls.Supported("getresuid", Getresuid), + 119: syscalls.Supported("setresgid", Setresgid), + 120: syscalls.Supported("getresgid", Getresgid), + 121: syscalls.Supported("getpgid", Getpgid), + 122: syscalls.ErrorWithEvent("setfsuid", syserror.ENOSYS, "", []string{"gvisor.dev/issue/260"}), // TODO(b/112851702) + 123: syscalls.ErrorWithEvent("setfsgid", syserror.ENOSYS, "", []string{"gvisor.dev/issue/260"}), // TODO(b/112851702) + 124: syscalls.Supported("getsid", Getsid), + 125: syscalls.Supported("capget", Capget), + 126: syscalls.Supported("capset", Capset), + 127: syscalls.Supported("rt_sigpending", RtSigpending), + 128: syscalls.Supported("rt_sigtimedwait", RtSigtimedwait), + 129: syscalls.Supported("rt_sigqueueinfo", RtSigqueueinfo), + 130: syscalls.Supported("rt_sigsuspend", RtSigsuspend), + 131: syscalls.Supported("sigaltstack", Sigaltstack), + 132: syscalls.Supported("utime", Utime), + 133: syscalls.PartiallySupported("mknod", Mknod, "Device creation is not generally supported. Only regular file and FIFO creation are supported.", nil), + 134: syscalls.Error("uselib", syserror.ENOSYS, "Obsolete", nil), + 135: syscalls.ErrorWithEvent("personality", syserror.EINVAL, "Unable to change personality.", nil), + 136: syscalls.ErrorWithEvent("ustat", syserror.ENOSYS, "Needs filesystem support.", nil), + 137: syscalls.PartiallySupported("statfs", Statfs, "Depends on the backing file system implementation.", nil), + 138: syscalls.PartiallySupported("fstatfs", Fstatfs, "Depends on the backing file system implementation.", nil), + 139: syscalls.ErrorWithEvent("sysfs", syserror.ENOSYS, "", []string{"gvisor.dev/issue/165"}), + 140: syscalls.PartiallySupported("getpriority", Getpriority, "Stub implementation.", nil), + 141: syscalls.PartiallySupported("setpriority", Setpriority, "Stub implementation.", nil), + 142: syscalls.CapError("sched_setparam", linux.CAP_SYS_NICE, "", nil), + 143: syscalls.PartiallySupported("sched_getparam", SchedGetparam, "Stub implementation.", nil), + 144: syscalls.PartiallySupported("sched_setscheduler", SchedSetscheduler, "Stub implementation.", nil), + 145: syscalls.PartiallySupported("sched_getscheduler", SchedGetscheduler, "Stub implementation.", nil), + 146: syscalls.PartiallySupported("sched_get_priority_max", SchedGetPriorityMax, "Stub implementation.", nil), + 147: syscalls.PartiallySupported("sched_get_priority_min", SchedGetPriorityMin, "Stub implementation.", nil), + 148: syscalls.ErrorWithEvent("sched_rr_get_interval", syserror.EPERM, "", nil), + 149: syscalls.PartiallySupported("mlock", Mlock, "Stub implementation. The sandbox lacks appropriate permissions.", nil), + 150: syscalls.PartiallySupported("munlock", Munlock, "Stub implementation. The sandbox lacks appropriate permissions.", nil), + 151: syscalls.PartiallySupported("mlockall", Mlockall, "Stub implementation. The sandbox lacks appropriate permissions.", nil), + 152: syscalls.PartiallySupported("munlockall", Munlockall, "Stub implementation. The sandbox lacks appropriate permissions.", nil), + 153: syscalls.CapError("vhangup", linux.CAP_SYS_TTY_CONFIG, "", nil), + 154: syscalls.Error("modify_ldt", syserror.EPERM, "", nil), + 155: syscalls.Error("pivot_root", syserror.EPERM, "", nil), + 156: syscalls.Error("sysctl", syserror.EPERM, "Deprecated. Use /proc/sys instead.", nil), + 157: syscalls.PartiallySupported("prctl", Prctl, "Not all options are supported.", nil), + 158: syscalls.PartiallySupported("arch_prctl", ArchPrctl, "Options ARCH_GET_GS, ARCH_SET_GS not supported.", nil), + 159: syscalls.CapError("adjtimex", linux.CAP_SYS_TIME, "", nil), + 160: syscalls.PartiallySupported("setrlimit", Setrlimit, "Not all rlimits are enforced.", nil), + 161: syscalls.Supported("chroot", Chroot), + 162: syscalls.PartiallySupported("sync", Sync, "Full data flush is not guaranteed at this time.", nil), + 163: syscalls.CapError("acct", linux.CAP_SYS_PACCT, "", nil), + 164: syscalls.CapError("settimeofday", linux.CAP_SYS_TIME, "", nil), + 165: syscalls.PartiallySupported("mount", Mount, "Not all options or file systems are supported.", nil), + 166: syscalls.PartiallySupported("umount2", Umount2, "Not all options or file systems are supported.", nil), + 167: syscalls.CapError("swapon", linux.CAP_SYS_ADMIN, "", nil), + 168: syscalls.CapError("swapoff", linux.CAP_SYS_ADMIN, "", nil), + 169: syscalls.CapError("reboot", linux.CAP_SYS_BOOT, "", nil), + 170: syscalls.Supported("sethostname", Sethostname), + 171: syscalls.Supported("setdomainname", Setdomainname), + 172: syscalls.CapError("iopl", linux.CAP_SYS_RAWIO, "", nil), + 173: syscalls.CapError("ioperm", linux.CAP_SYS_RAWIO, "", nil), + 174: syscalls.CapError("create_module", linux.CAP_SYS_MODULE, "", nil), + 175: syscalls.CapError("init_module", linux.CAP_SYS_MODULE, "", nil), + 176: syscalls.CapError("delete_module", linux.CAP_SYS_MODULE, "", nil), + 177: syscalls.Error("get_kernel_syms", syserror.ENOSYS, "Not supported in Linux > 2.6.", nil), + 178: syscalls.Error("query_module", syserror.ENOSYS, "Not supported in Linux > 2.6.", nil), + 179: syscalls.CapError("quotactl", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_admin for most operations + 180: syscalls.Error("nfsservctl", syserror.ENOSYS, "Removed after Linux 3.1.", nil), + 181: syscalls.Error("getpmsg", syserror.ENOSYS, "Not implemented in Linux.", nil), + 182: syscalls.Error("putpmsg", syserror.ENOSYS, "Not implemented in Linux.", nil), + 183: syscalls.Error("afs_syscall", syserror.ENOSYS, "Not implemented in Linux.", nil), + 184: syscalls.Error("tuxcall", syserror.ENOSYS, "Not implemented in Linux.", nil), + 185: syscalls.Error("security", syserror.ENOSYS, "Not implemented in Linux.", nil), + 186: syscalls.Supported("gettid", Gettid), + 187: syscalls.Supported("readahead", Readahead), + 188: syscalls.PartiallySupported("setxattr", SetXattr, "Only supported for tmpfs.", nil), + 189: syscalls.PartiallySupported("lsetxattr", LSetXattr, "Only supported for tmpfs.", nil), + 190: syscalls.PartiallySupported("fsetxattr", FSetXattr, "Only supported for tmpfs.", nil), + 191: syscalls.PartiallySupported("getxattr", GetXattr, "Only supported for tmpfs.", nil), + 192: syscalls.PartiallySupported("lgetxattr", LGetXattr, "Only supported for tmpfs.", nil), + 193: syscalls.PartiallySupported("fgetxattr", FGetXattr, "Only supported for tmpfs.", nil), + 194: syscalls.PartiallySupported("listxattr", ListXattr, "Only supported for tmpfs", nil), + 195: syscalls.PartiallySupported("llistxattr", LListXattr, "Only supported for tmpfs", nil), + 196: syscalls.PartiallySupported("flistxattr", FListXattr, "Only supported for tmpfs", nil), + 197: syscalls.PartiallySupported("removexattr", RemoveXattr, "Only supported for tmpfs", nil), + 198: syscalls.PartiallySupported("lremovexattr", LRemoveXattr, "Only supported for tmpfs", nil), + 199: syscalls.PartiallySupported("fremovexattr", FRemoveXattr, "Only supported for tmpfs", nil), + 200: syscalls.Supported("tkill", Tkill), + 201: syscalls.Supported("time", Time), + 202: syscalls.PartiallySupported("futex", Futex, "Robust futexes not supported.", nil), + 203: syscalls.PartiallySupported("sched_setaffinity", SchedSetaffinity, "Stub implementation.", nil), + 204: syscalls.PartiallySupported("sched_getaffinity", SchedGetaffinity, "Stub implementation.", nil), + 205: syscalls.Error("set_thread_area", syserror.ENOSYS, "Expected to return ENOSYS on 64-bit", nil), + 206: syscalls.PartiallySupported("io_setup", IoSetup, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 207: syscalls.PartiallySupported("io_destroy", IoDestroy, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 208: syscalls.PartiallySupported("io_getevents", IoGetevents, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 209: syscalls.PartiallySupported("io_submit", IoSubmit, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 210: syscalls.PartiallySupported("io_cancel", IoCancel, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 211: syscalls.Error("get_thread_area", syserror.ENOSYS, "Expected to return ENOSYS on 64-bit", nil), + 212: syscalls.CapError("lookup_dcookie", linux.CAP_SYS_ADMIN, "", nil), + 213: syscalls.Supported("epoll_create", EpollCreate), + 214: syscalls.ErrorWithEvent("epoll_ctl_old", syserror.ENOSYS, "Deprecated.", nil), + 215: syscalls.ErrorWithEvent("epoll_wait_old", syserror.ENOSYS, "Deprecated.", nil), + 216: syscalls.ErrorWithEvent("remap_file_pages", syserror.ENOSYS, "Deprecated since Linux 3.16.", nil), + 217: syscalls.Supported("getdents64", Getdents64), + 218: syscalls.Supported("set_tid_address", SetTidAddress), + 219: syscalls.Supported("restart_syscall", RestartSyscall), + 220: syscalls.ErrorWithEvent("semtimedop", syserror.ENOSYS, "", []string{"gvisor.dev/issue/137"}), + 221: syscalls.PartiallySupported("fadvise64", Fadvise64, "Not all options are supported.", nil), + 222: syscalls.Supported("timer_create", TimerCreate), + 223: syscalls.Supported("timer_settime", TimerSettime), + 224: syscalls.Supported("timer_gettime", TimerGettime), + 225: syscalls.Supported("timer_getoverrun", TimerGetoverrun), + 226: syscalls.Supported("timer_delete", TimerDelete), + 227: syscalls.Supported("clock_settime", ClockSettime), + 228: syscalls.Supported("clock_gettime", ClockGettime), + 229: syscalls.Supported("clock_getres", ClockGetres), + 230: syscalls.Supported("clock_nanosleep", ClockNanosleep), + 231: syscalls.Supported("exit_group", ExitGroup), + 232: syscalls.Supported("epoll_wait", EpollWait), + 233: syscalls.Supported("epoll_ctl", EpollCtl), + 234: syscalls.Supported("tgkill", Tgkill), + 235: syscalls.Supported("utimes", Utimes), + 236: syscalls.Error("vserver", syserror.ENOSYS, "Not implemented by Linux", nil), + 237: syscalls.PartiallySupported("mbind", Mbind, "Stub implementation. Only a single NUMA node is advertised, and mempolicy is ignored accordingly, but mbind() will succeed and has effects reflected by get_mempolicy.", []string{"gvisor.dev/issue/262"}), + 238: syscalls.PartiallySupported("set_mempolicy", SetMempolicy, "Stub implementation.", nil), + 239: syscalls.PartiallySupported("get_mempolicy", GetMempolicy, "Stub implementation.", nil), + 240: syscalls.ErrorWithEvent("mq_open", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 241: syscalls.ErrorWithEvent("mq_unlink", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 242: syscalls.ErrorWithEvent("mq_timedsend", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 243: syscalls.ErrorWithEvent("mq_timedreceive", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 244: syscalls.ErrorWithEvent("mq_notify", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 245: syscalls.ErrorWithEvent("mq_getsetattr", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 246: syscalls.CapError("kexec_load", linux.CAP_SYS_BOOT, "", nil), + 247: syscalls.Supported("waitid", Waitid), + 248: syscalls.Error("add_key", syserror.EACCES, "Not available to user.", nil), + 249: syscalls.Error("request_key", syserror.EACCES, "Not available to user.", nil), + 250: syscalls.Error("keyctl", syserror.EACCES, "Not available to user.", nil), + 251: syscalls.CapError("ioprio_set", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_nice or cap_sys_admin (depending) + 252: syscalls.CapError("ioprio_get", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_nice or cap_sys_admin (depending) + 253: syscalls.PartiallySupported("inotify_init", InotifyInit, "inotify events are only available inside the sandbox.", nil), + 254: syscalls.PartiallySupported("inotify_add_watch", InotifyAddWatch, "inotify events are only available inside the sandbox.", nil), + 255: syscalls.PartiallySupported("inotify_rm_watch", InotifyRmWatch, "inotify events are only available inside the sandbox.", nil), + 256: syscalls.CapError("migrate_pages", linux.CAP_SYS_NICE, "", nil), + 257: syscalls.Supported("openat", Openat), + 258: syscalls.Supported("mkdirat", Mkdirat), + 259: syscalls.Supported("mknodat", Mknodat), + 260: syscalls.Supported("fchownat", Fchownat), + 261: syscalls.Supported("futimesat", Futimesat), + 262: syscalls.Supported("fstatat", Fstatat), + 263: syscalls.Supported("unlinkat", Unlinkat), + 264: syscalls.Supported("renameat", Renameat), + 265: syscalls.Supported("linkat", Linkat), + 266: syscalls.Supported("symlinkat", Symlinkat), + 267: syscalls.Supported("readlinkat", Readlinkat), + 268: syscalls.Supported("fchmodat", Fchmodat), + 269: syscalls.Supported("faccessat", Faccessat), + 270: syscalls.Supported("pselect", Pselect), + 271: syscalls.Supported("ppoll", Ppoll), + 272: syscalls.PartiallySupported("unshare", Unshare, "Mount, cgroup namespaces not supported. Network namespaces supported but must be empty.", nil), + 273: syscalls.Error("set_robust_list", syserror.ENOSYS, "Obsolete.", nil), + 274: syscalls.Error("get_robust_list", syserror.ENOSYS, "Obsolete.", nil), + 275: syscalls.Supported("splice", Splice), + 276: syscalls.Supported("tee", Tee), + 277: syscalls.PartiallySupported("sync_file_range", SyncFileRange, "Full data flush is not guaranteed at this time.", nil), + 278: syscalls.ErrorWithEvent("vmsplice", syserror.ENOSYS, "", []string{"gvisor.dev/issue/138"}), // TODO(b/29354098) + 279: syscalls.CapError("move_pages", linux.CAP_SYS_NICE, "", nil), // requires cap_sys_nice (mostly) + 280: syscalls.Supported("utimensat", Utimensat), + 281: syscalls.Supported("epoll_pwait", EpollPwait), + 282: syscalls.PartiallySupported("signalfd", Signalfd, "Semantics are slightly different.", []string{"gvisor.dev/issue/139"}), + 283: syscalls.Supported("timerfd_create", TimerfdCreate), + 284: syscalls.Supported("eventfd", Eventfd), + 285: syscalls.PartiallySupported("fallocate", Fallocate, "Not all options are supported.", nil), + 286: syscalls.Supported("timerfd_settime", TimerfdSettime), + 287: syscalls.Supported("timerfd_gettime", TimerfdGettime), + 288: syscalls.Supported("accept4", Accept4), + 289: syscalls.PartiallySupported("signalfd4", Signalfd4, "Semantics are slightly different.", []string{"gvisor.dev/issue/139"}), + 290: syscalls.Supported("eventfd2", Eventfd2), + 291: syscalls.Supported("epoll_create1", EpollCreate1), + 292: syscalls.Supported("dup3", Dup3), + 293: syscalls.Supported("pipe2", Pipe2), + 294: syscalls.Supported("inotify_init1", InotifyInit1), + 295: syscalls.Supported("preadv", Preadv), + 296: syscalls.Supported("pwritev", Pwritev), + 297: syscalls.Supported("rt_tgsigqueueinfo", RtTgsigqueueinfo), + 298: syscalls.ErrorWithEvent("perf_event_open", syserror.ENODEV, "No support for perf counters", nil), + 299: syscalls.PartiallySupported("recvmmsg", RecvMMsg, "Not all flags and control messages are supported.", nil), + 300: syscalls.ErrorWithEvent("fanotify_init", syserror.ENOSYS, "Needs CONFIG_FANOTIFY", nil), + 301: syscalls.ErrorWithEvent("fanotify_mark", syserror.ENOSYS, "Needs CONFIG_FANOTIFY", nil), + 302: syscalls.Supported("prlimit64", Prlimit64), + 303: syscalls.Error("name_to_handle_at", syserror.EOPNOTSUPP, "Not supported by gVisor filesystems", nil), + 304: syscalls.Error("open_by_handle_at", syserror.EOPNOTSUPP, "Not supported by gVisor filesystems", nil), + 305: syscalls.CapError("clock_adjtime", linux.CAP_SYS_TIME, "", nil), + 306: syscalls.PartiallySupported("syncfs", Syncfs, "Depends on backing file system.", nil), + 307: syscalls.PartiallySupported("sendmmsg", SendMMsg, "Not all flags and control messages are supported.", nil), + 308: syscalls.ErrorWithEvent("setns", syserror.EOPNOTSUPP, "Needs filesystem support", []string{"gvisor.dev/issue/140"}), // TODO(b/29354995) + 309: syscalls.Supported("getcpu", Getcpu), + 310: syscalls.ErrorWithEvent("process_vm_readv", syserror.ENOSYS, "", []string{"gvisor.dev/issue/158"}), + 311: syscalls.ErrorWithEvent("process_vm_writev", syserror.ENOSYS, "", []string{"gvisor.dev/issue/158"}), + 312: syscalls.CapError("kcmp", linux.CAP_SYS_PTRACE, "", nil), + 313: syscalls.CapError("finit_module", linux.CAP_SYS_MODULE, "", nil), + 314: syscalls.ErrorWithEvent("sched_setattr", syserror.ENOSYS, "gVisor does not implement a scheduler.", []string{"gvisor.dev/issue/264"}), // TODO(b/118902272) + 315: syscalls.ErrorWithEvent("sched_getattr", syserror.ENOSYS, "gVisor does not implement a scheduler.", []string{"gvisor.dev/issue/264"}), // TODO(b/118902272) + 316: syscalls.ErrorWithEvent("renameat2", syserror.ENOSYS, "", []string{"gvisor.dev/issue/263"}), // TODO(b/118902772) + 317: syscalls.Supported("seccomp", Seccomp), + 318: syscalls.Supported("getrandom", GetRandom), + 319: syscalls.Supported("memfd_create", MemfdCreate), + 320: syscalls.CapError("kexec_file_load", linux.CAP_SYS_BOOT, "", nil), + 321: syscalls.CapError("bpf", linux.CAP_SYS_ADMIN, "", nil), + 322: syscalls.Supported("execveat", Execveat), + 323: syscalls.ErrorWithEvent("userfaultfd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/266"}), // TODO(b/118906345) + 324: syscalls.ErrorWithEvent("membarrier", syserror.ENOSYS, "", []string{"gvisor.dev/issue/267"}), // TODO(gvisor.dev/issue/267) + 325: syscalls.PartiallySupported("mlock2", Mlock2, "Stub implementation. The sandbox lacks appropriate permissions.", nil), + + // Syscalls implemented after 325 are "backports" from versions + // of Linux after 4.4. + 326: syscalls.ErrorWithEvent("copy_file_range", syserror.ENOSYS, "", nil), + 327: syscalls.Supported("preadv2", Preadv2), + 328: syscalls.PartiallySupported("pwritev2", Pwritev2, "Flag RWF_HIPRI is not supported.", nil), + 329: syscalls.ErrorWithEvent("pkey_mprotect", syserror.ENOSYS, "", nil), + 330: syscalls.ErrorWithEvent("pkey_alloc", syserror.ENOSYS, "", nil), + 331: syscalls.ErrorWithEvent("pkey_free", syserror.ENOSYS, "", nil), + 332: syscalls.Supported("statx", Statx), + 333: syscalls.ErrorWithEvent("io_pgetevents", syserror.ENOSYS, "", nil), + 334: syscalls.PartiallySupported("rseq", RSeq, "Not supported on all platforms.", nil), + + // Linux skips ahead to syscall 424 to sync numbers between arches. + 424: syscalls.ErrorWithEvent("pidfd_send_signal", syserror.ENOSYS, "", nil), + 425: syscalls.ErrorWithEvent("io_uring_setup", syserror.ENOSYS, "", nil), + 426: syscalls.ErrorWithEvent("io_uring_enter", syserror.ENOSYS, "", nil), + 427: syscalls.ErrorWithEvent("io_uring_register", syserror.ENOSYS, "", nil), + 428: syscalls.ErrorWithEvent("open_tree", syserror.ENOSYS, "", nil), + 429: syscalls.ErrorWithEvent("move_mount", syserror.ENOSYS, "", nil), + 430: syscalls.ErrorWithEvent("fsopen", syserror.ENOSYS, "", nil), + 431: syscalls.ErrorWithEvent("fsconfig", syserror.ENOSYS, "", nil), + 432: syscalls.ErrorWithEvent("fsmount", syserror.ENOSYS, "", nil), + 433: syscalls.ErrorWithEvent("fspick", syserror.ENOSYS, "", nil), + 434: syscalls.ErrorWithEvent("pidfd_open", syserror.ENOSYS, "", nil), + 435: syscalls.ErrorWithEvent("clone3", syserror.ENOSYS, "", nil), + }, + Emulate: map[usermem.Addr]uintptr{ + 0xffffffffff600000: 96, // vsyscall gettimeofday(2) + 0xffffffffff600400: 201, // vsyscall time(2) + 0xffffffffff600800: 309, // vsyscall getcpu(2) + }, + Missing: func(t *kernel.Task, sysno uintptr, args arch.SyscallArguments) (uintptr, error) { + t.Kernel().EmitUnimplementedEvent(t) + return 0, syserror.ENOSYS + }, +} + +// ARM64 is a table of Linux arm64 syscall API with the corresponding syscall +// numbers from Linux 4.4. +var ARM64 = &kernel.SyscallTable{ + OS: abi.Linux, + Arch: arch.ARM64, + Version: kernel.Version{ + Sysname: LinuxSysname, + Release: LinuxRelease, + Version: LinuxVersion, + }, + AuditNumber: linux.AUDIT_ARCH_AARCH64, + Table: map[uintptr]kernel.Syscall{ + 0: syscalls.PartiallySupported("io_setup", IoSetup, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 1: syscalls.PartiallySupported("io_destroy", IoDestroy, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 2: syscalls.PartiallySupported("io_submit", IoSubmit, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 3: syscalls.PartiallySupported("io_cancel", IoCancel, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 4: syscalls.PartiallySupported("io_getevents", IoGetevents, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 5: syscalls.PartiallySupported("setxattr", SetXattr, "Only supported for tmpfs.", nil), + 6: syscalls.PartiallySupported("lsetxattr", LSetXattr, "Only supported for tmpfs.", nil), + 7: syscalls.PartiallySupported("fsetxattr", FSetXattr, "Only supported for tmpfs.", nil), + 8: syscalls.PartiallySupported("getxattr", GetXattr, "Only supported for tmpfs.", nil), + 9: syscalls.PartiallySupported("lgetxattr", LGetXattr, "Only supported for tmpfs.", nil), + 10: syscalls.PartiallySupported("fgetxattr", FGetXattr, "Only supported for tmpfs.", nil), + 11: syscalls.PartiallySupported("listxattr", ListXattr, "Only supported for tmpfs", nil), + 12: syscalls.PartiallySupported("llistxattr", LListXattr, "Only supported for tmpfs", nil), + 13: syscalls.PartiallySupported("flistxattr", FListXattr, "Only supported for tmpfs", nil), + 14: syscalls.PartiallySupported("removexattr", RemoveXattr, "Only supported for tmpfs", nil), + 15: syscalls.PartiallySupported("lremovexattr", LRemoveXattr, "Only supported for tmpfs", nil), + 16: syscalls.PartiallySupported("fremovexattr", FRemoveXattr, "Only supported for tmpfs", nil), + 17: syscalls.Supported("getcwd", Getcwd), + 18: syscalls.CapError("lookup_dcookie", linux.CAP_SYS_ADMIN, "", nil), + 19: syscalls.Supported("eventfd2", Eventfd2), + 20: syscalls.Supported("epoll_create1", EpollCreate1), + 21: syscalls.Supported("epoll_ctl", EpollCtl), + 22: syscalls.Supported("epoll_pwait", EpollPwait), + 23: syscalls.Supported("dup", Dup), + 24: syscalls.Supported("dup3", Dup3), + 25: syscalls.PartiallySupported("fcntl", Fcntl, "Not all options are supported.", nil), + 26: syscalls.Supported("inotify_init1", InotifyInit1), + 27: syscalls.PartiallySupported("inotify_add_watch", InotifyAddWatch, "inotify events are only available inside the sandbox.", nil), + 28: syscalls.PartiallySupported("inotify_rm_watch", InotifyRmWatch, "inotify events are only available inside the sandbox.", nil), + 29: syscalls.PartiallySupported("ioctl", Ioctl, "Only a few ioctls are implemented for backing devices and file systems.", nil), + 30: syscalls.CapError("ioprio_set", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_nice or cap_sys_admin (depending) + 31: syscalls.CapError("ioprio_get", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_nice or cap_sys_admin (depending) + 32: syscalls.PartiallySupported("flock", Flock, "Locks are held within the sandbox only.", nil), + 33: syscalls.Supported("mknodat", Mknodat), + 34: syscalls.Supported("mkdirat", Mkdirat), + 35: syscalls.Supported("unlinkat", Unlinkat), + 36: syscalls.Supported("symlinkat", Symlinkat), + 37: syscalls.Supported("linkat", Linkat), + 38: syscalls.Supported("renameat", Renameat), + 39: syscalls.PartiallySupported("umount2", Umount2, "Not all options or file systems are supported.", nil), + 40: syscalls.PartiallySupported("mount", Mount, "Not all options or file systems are supported.", nil), + 41: syscalls.Error("pivot_root", syserror.EPERM, "", nil), + 42: syscalls.Error("nfsservctl", syserror.ENOSYS, "Removed after Linux 3.1.", nil), + 43: syscalls.PartiallySupported("statfs", Statfs, "Depends on the backing file system implementation.", nil), + 44: syscalls.PartiallySupported("fstatfs", Fstatfs, "Depends on the backing file system implementation.", nil), + 45: syscalls.Supported("truncate", Truncate), + 46: syscalls.Supported("ftruncate", Ftruncate), + 47: syscalls.PartiallySupported("fallocate", Fallocate, "Not all options are supported.", nil), + 48: syscalls.Supported("faccessat", Faccessat), + 49: syscalls.Supported("chdir", Chdir), + 50: syscalls.Supported("fchdir", Fchdir), + 51: syscalls.Supported("chroot", Chroot), + 52: syscalls.PartiallySupported("fchmod", Fchmod, "Options S_ISUID and S_ISGID not supported.", nil), + 53: syscalls.Supported("fchmodat", Fchmodat), + 54: syscalls.Supported("fchownat", Fchownat), + 55: syscalls.Supported("fchown", Fchown), + 56: syscalls.Supported("openat", Openat), + 57: syscalls.Supported("close", Close), + 58: syscalls.CapError("vhangup", linux.CAP_SYS_TTY_CONFIG, "", nil), + 59: syscalls.Supported("pipe2", Pipe2), + 60: syscalls.CapError("quotactl", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_admin for most operations + 61: syscalls.Supported("getdents64", Getdents64), + 62: syscalls.Supported("lseek", Lseek), + 63: syscalls.Supported("read", Read), + 64: syscalls.Supported("write", Write), + 65: syscalls.Supported("readv", Readv), + 66: syscalls.Supported("writev", Writev), + 67: syscalls.Supported("pread64", Pread64), + 68: syscalls.Supported("pwrite64", Pwrite64), + 69: syscalls.Supported("preadv", Preadv), + 70: syscalls.Supported("pwritev", Pwritev), + 71: syscalls.Supported("sendfile", Sendfile), + 72: syscalls.Supported("pselect", Pselect), + 73: syscalls.Supported("ppoll", Ppoll), + 74: syscalls.PartiallySupported("signalfd4", Signalfd4, "Semantics are slightly different.", []string{"gvisor.dev/issue/139"}), + 75: syscalls.ErrorWithEvent("vmsplice", syserror.ENOSYS, "", []string{"gvisor.dev/issue/138"}), // TODO(b/29354098) + 76: syscalls.PartiallySupported("splice", Splice, "Stub implementation.", []string{"gvisor.dev/issue/138"}), // TODO(b/29354098) + 77: syscalls.Supported("tee", Tee), + 78: syscalls.Supported("readlinkat", Readlinkat), + 79: syscalls.Supported("fstatat", Fstatat), + 80: syscalls.Supported("fstat", Fstat), + 81: syscalls.PartiallySupported("sync", Sync, "Full data flush is not guaranteed at this time.", nil), + 82: syscalls.PartiallySupported("fsync", Fsync, "Full data flush is not guaranteed at this time.", nil), + 83: syscalls.PartiallySupported("fdatasync", Fdatasync, "Full data flush is not guaranteed at this time.", nil), + 84: syscalls.PartiallySupported("sync_file_range", SyncFileRange, "Full data flush is not guaranteed at this time.", nil), + 85: syscalls.Supported("timerfd_create", TimerfdCreate), + 86: syscalls.Supported("timerfd_settime", TimerfdSettime), + 87: syscalls.Supported("timerfd_gettime", TimerfdGettime), + 88: syscalls.Supported("utimensat", Utimensat), + 89: syscalls.CapError("acct", linux.CAP_SYS_PACCT, "", nil), + 90: syscalls.Supported("capget", Capget), + 91: syscalls.Supported("capset", Capset), + 92: syscalls.ErrorWithEvent("personality", syserror.EINVAL, "Unable to change personality.", nil), + 93: syscalls.Supported("exit", Exit), + 94: syscalls.Supported("exit_group", ExitGroup), + 95: syscalls.Supported("waitid", Waitid), + 96: syscalls.Supported("set_tid_address", SetTidAddress), + 97: syscalls.PartiallySupported("unshare", Unshare, "Mount, cgroup namespaces not supported. Network namespaces supported but must be empty.", nil), + 98: syscalls.PartiallySupported("futex", Futex, "Robust futexes not supported.", nil), + 99: syscalls.Error("set_robust_list", syserror.ENOSYS, "Obsolete.", nil), + 100: syscalls.Error("get_robust_list", syserror.ENOSYS, "Obsolete.", nil), + 101: syscalls.Supported("nanosleep", Nanosleep), + 102: syscalls.Supported("getitimer", Getitimer), + 103: syscalls.Supported("setitimer", Setitimer), + 104: syscalls.CapError("kexec_load", linux.CAP_SYS_BOOT, "", nil), + 105: syscalls.CapError("init_module", linux.CAP_SYS_MODULE, "", nil), + 106: syscalls.CapError("delete_module", linux.CAP_SYS_MODULE, "", nil), + 107: syscalls.Supported("timer_create", TimerCreate), + 108: syscalls.Supported("timer_gettime", TimerGettime), + 109: syscalls.Supported("timer_getoverrun", TimerGetoverrun), + 110: syscalls.Supported("timer_settime", TimerSettime), + 111: syscalls.Supported("timer_delete", TimerDelete), + 112: syscalls.Supported("clock_settime", ClockSettime), + 113: syscalls.Supported("clock_gettime", ClockGettime), + 114: syscalls.Supported("clock_getres", ClockGetres), + 115: syscalls.Supported("clock_nanosleep", ClockNanosleep), + 116: syscalls.PartiallySupported("syslog", Syslog, "Outputs a dummy message for security reasons.", nil), + 117: syscalls.PartiallySupported("ptrace", Ptrace, "Options PTRACE_PEEKSIGINFO, PTRACE_SECCOMP_GET_FILTER not supported.", nil), + 118: syscalls.CapError("sched_setparam", linux.CAP_SYS_NICE, "", nil), + 119: syscalls.PartiallySupported("sched_setscheduler", SchedSetscheduler, "Stub implementation.", nil), + 120: syscalls.PartiallySupported("sched_getscheduler", SchedGetscheduler, "Stub implementation.", nil), + 121: syscalls.PartiallySupported("sched_getparam", SchedGetparam, "Stub implementation.", nil), + 122: syscalls.PartiallySupported("sched_setaffinity", SchedSetaffinity, "Stub implementation.", nil), + 123: syscalls.PartiallySupported("sched_getaffinity", SchedGetaffinity, "Stub implementation.", nil), + 124: syscalls.Supported("sched_yield", SchedYield), + 125: syscalls.PartiallySupported("sched_get_priority_max", SchedGetPriorityMax, "Stub implementation.", nil), + 126: syscalls.PartiallySupported("sched_get_priority_min", SchedGetPriorityMin, "Stub implementation.", nil), + 127: syscalls.ErrorWithEvent("sched_rr_get_interval", syserror.EPERM, "", nil), + 128: syscalls.Supported("restart_syscall", RestartSyscall), + 129: syscalls.Supported("kill", Kill), + 130: syscalls.Supported("tkill", Tkill), + 131: syscalls.Supported("tgkill", Tgkill), + 132: syscalls.Supported("sigaltstack", Sigaltstack), + 133: syscalls.Supported("rt_sigsuspend", RtSigsuspend), + 134: syscalls.Supported("rt_sigaction", RtSigaction), + 135: syscalls.Supported("rt_sigprocmask", RtSigprocmask), + 136: syscalls.Supported("rt_sigpending", RtSigpending), + 137: syscalls.Supported("rt_sigtimedwait", RtSigtimedwait), + 138: syscalls.Supported("rt_sigqueueinfo", RtSigqueueinfo), + 139: syscalls.Supported("rt_sigreturn", RtSigreturn), + 140: syscalls.PartiallySupported("setpriority", Setpriority, "Stub implementation.", nil), + 141: syscalls.PartiallySupported("getpriority", Getpriority, "Stub implementation.", nil), + 142: syscalls.CapError("reboot", linux.CAP_SYS_BOOT, "", nil), + 143: syscalls.Supported("setregid", Setregid), + 144: syscalls.Supported("setgid", Setgid), + 145: syscalls.Supported("setreuid", Setreuid), + 146: syscalls.Supported("setuid", Setuid), + 147: syscalls.Supported("setresuid", Setresuid), + 148: syscalls.Supported("getresuid", Getresuid), + 149: syscalls.Supported("setresgid", Setresgid), + 150: syscalls.Supported("getresgid", Getresgid), + 151: syscalls.ErrorWithEvent("setfsuid", syserror.ENOSYS, "", []string{"gvisor.dev/issue/260"}), // TODO(b/112851702) + 152: syscalls.ErrorWithEvent("setfsgid", syserror.ENOSYS, "", []string{"gvisor.dev/issue/260"}), // TODO(b/112851702) + 153: syscalls.Supported("times", Times), + 154: syscalls.Supported("setpgid", Setpgid), + 155: syscalls.Supported("getpgid", Getpgid), + 156: syscalls.Supported("getsid", Getsid), + 157: syscalls.Supported("setsid", Setsid), + 158: syscalls.Supported("getgroups", Getgroups), + 159: syscalls.Supported("setgroups", Setgroups), + 160: syscalls.Supported("uname", Uname), + 161: syscalls.Supported("sethostname", Sethostname), + 162: syscalls.Supported("setdomainname", Setdomainname), + 163: syscalls.Supported("getrlimit", Getrlimit), + 164: syscalls.PartiallySupported("setrlimit", Setrlimit, "Not all rlimits are enforced.", nil), + 165: syscalls.PartiallySupported("getrusage", Getrusage, "Fields ru_maxrss, ru_minflt, ru_majflt, ru_inblock, ru_oublock are not supported. Fields ru_utime and ru_stime have low precision.", nil), + 166: syscalls.Supported("umask", Umask), + 167: syscalls.PartiallySupported("prctl", Prctl, "Not all options are supported.", nil), + 168: syscalls.Supported("getcpu", Getcpu), + 169: syscalls.Supported("gettimeofday", Gettimeofday), + 170: syscalls.CapError("settimeofday", linux.CAP_SYS_TIME, "", nil), + 171: syscalls.CapError("adjtimex", linux.CAP_SYS_TIME, "", nil), + 172: syscalls.Supported("getpid", Getpid), + 173: syscalls.Supported("getppid", Getppid), + 174: syscalls.Supported("getuid", Getuid), + 175: syscalls.Supported("geteuid", Geteuid), + 176: syscalls.Supported("getgid", Getgid), + 177: syscalls.Supported("getegid", Getegid), + 178: syscalls.Supported("gettid", Gettid), + 179: syscalls.PartiallySupported("sysinfo", Sysinfo, "Fields loads, sharedram, bufferram, totalswap, freeswap, totalhigh, freehigh not supported.", nil), + 180: syscalls.ErrorWithEvent("mq_open", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 181: syscalls.ErrorWithEvent("mq_unlink", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 182: syscalls.ErrorWithEvent("mq_timedsend", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 183: syscalls.ErrorWithEvent("mq_timedreceive", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 184: syscalls.ErrorWithEvent("mq_notify", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 185: syscalls.ErrorWithEvent("mq_getsetattr", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 186: syscalls.ErrorWithEvent("msgget", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) + 187: syscalls.ErrorWithEvent("msgctl", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) + 188: syscalls.ErrorWithEvent("msgrcv", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) + 189: syscalls.ErrorWithEvent("msgsnd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) + 190: syscalls.Supported("semget", Semget), + 191: syscalls.PartiallySupported("semctl", Semctl, "Options IPC_INFO, SEM_INFO, IPC_STAT, SEM_STAT, SEM_STAT_ANY, GETNCNT, GETZCNT not supported.", nil), + 192: syscalls.ErrorWithEvent("semtimedop", syserror.ENOSYS, "", []string{"gvisor.dev/issue/137"}), + 193: syscalls.PartiallySupported("semop", Semop, "Option SEM_UNDO not supported.", nil), + 194: syscalls.PartiallySupported("shmget", Shmget, "Option SHM_HUGETLB is not supported.", nil), + 195: syscalls.PartiallySupported("shmctl", Shmctl, "Options SHM_LOCK, SHM_UNLOCK are not supported.", nil), + 196: syscalls.PartiallySupported("shmat", Shmat, "Option SHM_RND is not supported.", nil), + 197: syscalls.Supported("shmdt", Shmdt), + 198: syscalls.PartiallySupported("socket", Socket, "Limited support for AF_NETLINK, NETLINK_ROUTE sockets. Limited support for SOCK_RAW.", nil), + 199: syscalls.Supported("socketpair", SocketPair), + 200: syscalls.PartiallySupported("bind", Bind, "Autobind for abstract Unix sockets is not supported.", nil), + 201: syscalls.Supported("listen", Listen), + 202: syscalls.Supported("accept", Accept), + 203: syscalls.Supported("connect", Connect), + 204: syscalls.Supported("getsockname", GetSockName), + 205: syscalls.Supported("getpeername", GetPeerName), + 206: syscalls.Supported("sendto", SendTo), + 207: syscalls.Supported("recvfrom", RecvFrom), + 208: syscalls.PartiallySupported("setsockopt", SetSockOpt, "Not all socket options are supported.", nil), + 209: syscalls.PartiallySupported("getsockopt", GetSockOpt, "Not all socket options are supported.", nil), + 210: syscalls.PartiallySupported("shutdown", Shutdown, "Not all flags and control messages are supported.", nil), + 211: syscalls.Supported("sendmsg", SendMsg), + 212: syscalls.PartiallySupported("recvmsg", RecvMsg, "Not all flags and control messages are supported.", nil), + 213: syscalls.Supported("readahead", Readahead), + 214: syscalls.Supported("brk", Brk), + 215: syscalls.Supported("munmap", Munmap), + 216: syscalls.Supported("mremap", Mremap), + 217: syscalls.Error("add_key", syserror.EACCES, "Not available to user.", nil), + 218: syscalls.Error("request_key", syserror.EACCES, "Not available to user.", nil), + 219: syscalls.Error("keyctl", syserror.EACCES, "Not available to user.", nil), + 220: syscalls.PartiallySupported("clone", Clone, "Mount namespace (CLONE_NEWNS) not supported. Options CLONE_PARENT, CLONE_SYSVSEM not supported.", nil), + 221: syscalls.Supported("execve", Execve), + 222: syscalls.PartiallySupported("mmap", Mmap, "Generally supported with exceptions. Options MAP_FIXED_NOREPLACE, MAP_SHARED_VALIDATE, MAP_SYNC MAP_GROWSDOWN, MAP_HUGETLB are not supported.", nil), + 223: syscalls.PartiallySupported("fadvise64", Fadvise64, "Not all options are supported.", nil), + 224: syscalls.CapError("swapon", linux.CAP_SYS_ADMIN, "", nil), + 225: syscalls.CapError("swapoff", linux.CAP_SYS_ADMIN, "", nil), + 226: syscalls.Supported("mprotect", Mprotect), + 227: syscalls.PartiallySupported("msync", Msync, "Full data flush is not guaranteed at this time.", nil), + 228: syscalls.PartiallySupported("mlock", Mlock, "Stub implementation. The sandbox lacks appropriate permissions.", nil), + 229: syscalls.PartiallySupported("munlock", Munlock, "Stub implementation. The sandbox lacks appropriate permissions.", nil), + 230: syscalls.PartiallySupported("mlockall", Mlockall, "Stub implementation. The sandbox lacks appropriate permissions.", nil), + 231: syscalls.PartiallySupported("munlockall", Munlockall, "Stub implementation. The sandbox lacks appropriate permissions.", nil), + 232: syscalls.PartiallySupported("mincore", Mincore, "Stub implementation. The sandbox does not have access to this information. Reports all mapped pages are resident.", nil), + 233: syscalls.PartiallySupported("madvise", Madvise, "Options MADV_DONTNEED, MADV_DONTFORK are supported. Other advice is ignored.", nil), + 234: syscalls.ErrorWithEvent("remap_file_pages", syserror.ENOSYS, "Deprecated since Linux 3.16.", nil), + 235: syscalls.PartiallySupported("mbind", Mbind, "Stub implementation. Only a single NUMA node is advertised, and mempolicy is ignored accordingly, but mbind() will succeed and has effects reflected by get_mempolicy.", []string{"gvisor.dev/issue/262"}), + 236: syscalls.PartiallySupported("get_mempolicy", GetMempolicy, "Stub implementation.", nil), + 237: syscalls.PartiallySupported("set_mempolicy", SetMempolicy, "Stub implementation.", nil), + 238: syscalls.CapError("migrate_pages", linux.CAP_SYS_NICE, "", nil), + 239: syscalls.CapError("move_pages", linux.CAP_SYS_NICE, "", nil), // requires cap_sys_nice (mostly) + 240: syscalls.Supported("rt_tgsigqueueinfo", RtTgsigqueueinfo), + 241: syscalls.ErrorWithEvent("perf_event_open", syserror.ENODEV, "No support for perf counters", nil), + 242: syscalls.Supported("accept4", Accept4), + 243: syscalls.PartiallySupported("recvmmsg", RecvMMsg, "Not all flags and control messages are supported.", nil), + 260: syscalls.Supported("wait4", Wait4), + 261: syscalls.Supported("prlimit64", Prlimit64), + 262: syscalls.ErrorWithEvent("fanotify_init", syserror.ENOSYS, "Needs CONFIG_FANOTIFY", nil), + 263: syscalls.ErrorWithEvent("fanotify_mark", syserror.ENOSYS, "Needs CONFIG_FANOTIFY", nil), + 264: syscalls.Error("name_to_handle_at", syserror.EOPNOTSUPP, "Not supported by gVisor filesystems", nil), + 265: syscalls.Error("open_by_handle_at", syserror.EOPNOTSUPP, "Not supported by gVisor filesystems", nil), + 266: syscalls.CapError("clock_adjtime", linux.CAP_SYS_TIME, "", nil), + 267: syscalls.PartiallySupported("syncfs", Syncfs, "Depends on backing file system.", nil), + 268: syscalls.ErrorWithEvent("setns", syserror.EOPNOTSUPP, "Needs filesystem support", []string{"gvisor.dev/issue/140"}), // TODO(b/29354995) + 269: syscalls.PartiallySupported("sendmmsg", SendMMsg, "Not all flags and control messages are supported.", nil), + 270: syscalls.ErrorWithEvent("process_vm_readv", syserror.ENOSYS, "", []string{"gvisor.dev/issue/158"}), + 271: syscalls.ErrorWithEvent("process_vm_writev", syserror.ENOSYS, "", []string{"gvisor.dev/issue/158"}), + 272: syscalls.CapError("kcmp", linux.CAP_SYS_PTRACE, "", nil), + 273: syscalls.CapError("finit_module", linux.CAP_SYS_MODULE, "", nil), + 274: syscalls.ErrorWithEvent("sched_setattr", syserror.ENOSYS, "gVisor does not implement a scheduler.", []string{"gvisor.dev/issue/264"}), // TODO(b/118902272) + 275: syscalls.ErrorWithEvent("sched_getattr", syserror.ENOSYS, "gVisor does not implement a scheduler.", []string{"gvisor.dev/issue/264"}), // TODO(b/118902272) + 276: syscalls.ErrorWithEvent("renameat2", syserror.ENOSYS, "", []string{"gvisor.dev/issue/263"}), // TODO(b/118902772) + 277: syscalls.Supported("seccomp", Seccomp), + 278: syscalls.Supported("getrandom", GetRandom), + 279: syscalls.Supported("memfd_create", MemfdCreate), + 280: syscalls.CapError("bpf", linux.CAP_SYS_ADMIN, "", nil), + 281: syscalls.Supported("execveat", Execveat), + 282: syscalls.ErrorWithEvent("userfaultfd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/266"}), // TODO(b/118906345) + 283: syscalls.ErrorWithEvent("membarrier", syserror.ENOSYS, "", []string{"gvisor.dev/issue/267"}), // TODO(gvisor.dev/issue/267) + 284: syscalls.PartiallySupported("mlock2", Mlock2, "Stub implementation. The sandbox lacks appropriate permissions.", nil), + + // Syscalls after 284 are "backports" from versions of Linux after 4.4. + 285: syscalls.ErrorWithEvent("copy_file_range", syserror.ENOSYS, "", nil), + 286: syscalls.Supported("preadv2", Preadv2), + 287: syscalls.PartiallySupported("pwritev2", Pwritev2, "Flag RWF_HIPRI is not supported.", nil), + 288: syscalls.ErrorWithEvent("pkey_mprotect", syserror.ENOSYS, "", nil), + 289: syscalls.ErrorWithEvent("pkey_alloc", syserror.ENOSYS, "", nil), + 290: syscalls.ErrorWithEvent("pkey_free", syserror.ENOSYS, "", nil), + 291: syscalls.Supported("statx", Statx), + 292: syscalls.ErrorWithEvent("io_pgetevents", syserror.ENOSYS, "", nil), + 293: syscalls.PartiallySupported("rseq", RSeq, "Not supported on all platforms.", nil), + + // Linux skips ahead to syscall 424 to sync numbers between arches. + 424: syscalls.ErrorWithEvent("pidfd_send_signal", syserror.ENOSYS, "", nil), + 425: syscalls.ErrorWithEvent("io_uring_setup", syserror.ENOSYS, "", nil), + 426: syscalls.ErrorWithEvent("io_uring_enter", syserror.ENOSYS, "", nil), + 427: syscalls.ErrorWithEvent("io_uring_register", syserror.ENOSYS, "", nil), + 428: syscalls.ErrorWithEvent("open_tree", syserror.ENOSYS, "", nil), + 429: syscalls.ErrorWithEvent("move_mount", syserror.ENOSYS, "", nil), + 430: syscalls.ErrorWithEvent("fsopen", syserror.ENOSYS, "", nil), + 431: syscalls.ErrorWithEvent("fsconfig", syserror.ENOSYS, "", nil), + 432: syscalls.ErrorWithEvent("fsmount", syserror.ENOSYS, "", nil), + 433: syscalls.ErrorWithEvent("fspick", syserror.ENOSYS, "", nil), + 434: syscalls.ErrorWithEvent("pidfd_open", syserror.ENOSYS, "", nil), + 435: syscalls.ErrorWithEvent("clone3", syserror.ENOSYS, "", nil), + }, + Emulate: map[usermem.Addr]uintptr{}, + Missing: func(t *kernel.Task, sysno uintptr, args arch.SyscallArguments) (uintptr, error) { + t.Kernel().EmitUnimplementedEvent(t) + return 0, syserror.ENOSYS + }, +} + +func init() { + kernel.RegisterSyscallTable(AMD64) + kernel.RegisterSyscallTable(ARM64) +} diff --git a/pkg/sentry/syscalls/linux/linux64_amd64.go b/pkg/sentry/syscalls/linux/linux64_amd64.go deleted file mode 100644 index 79066ad2a..000000000 --- a/pkg/sentry/syscalls/linux/linux64_amd64.go +++ /dev/null @@ -1,406 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package linux - -import ( - "gvisor.dev/gvisor/pkg/abi" - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/syscalls" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -// AMD64 is a table of Linux amd64 syscall API with the corresponding syscall -// numbers from Linux 4.4. -var AMD64 = &kernel.SyscallTable{ - OS: abi.Linux, - Arch: arch.AMD64, - Version: kernel.Version{ - // Version 4.4 is chosen as a stable, longterm version of Linux, which - // guides the interface provided by this syscall table. The build - // version is that for a clean build with default kernel config, at 5 - // minutes after v4.4 was tagged. - Sysname: LinuxSysname, - Release: LinuxRelease, - Version: LinuxVersion, - }, - AuditNumber: linux.AUDIT_ARCH_X86_64, - Table: map[uintptr]kernel.Syscall{ - 0: syscalls.Supported("read", Read), - 1: syscalls.Supported("write", Write), - 2: syscalls.PartiallySupported("open", Open, "Options O_DIRECT, O_NOATIME, O_PATH, O_TMPFILE, O_SYNC are not supported.", nil), - 3: syscalls.Supported("close", Close), - 4: syscalls.Supported("stat", Stat), - 5: syscalls.Supported("fstat", Fstat), - 6: syscalls.Supported("lstat", Lstat), - 7: syscalls.Supported("poll", Poll), - 8: syscalls.Supported("lseek", Lseek), - 9: syscalls.PartiallySupported("mmap", Mmap, "Generally supported with exceptions. Options MAP_FIXED_NOREPLACE, MAP_SHARED_VALIDATE, MAP_SYNC MAP_GROWSDOWN, MAP_HUGETLB are not supported.", nil), - 10: syscalls.Supported("mprotect", Mprotect), - 11: syscalls.Supported("munmap", Munmap), - 12: syscalls.Supported("brk", Brk), - 13: syscalls.Supported("rt_sigaction", RtSigaction), - 14: syscalls.Supported("rt_sigprocmask", RtSigprocmask), - 15: syscalls.Supported("rt_sigreturn", RtSigreturn), - 16: syscalls.PartiallySupported("ioctl", Ioctl, "Only a few ioctls are implemented for backing devices and file systems.", nil), - 17: syscalls.Supported("pread64", Pread64), - 18: syscalls.Supported("pwrite64", Pwrite64), - 19: syscalls.Supported("readv", Readv), - 20: syscalls.Supported("writev", Writev), - 21: syscalls.Supported("access", Access), - 22: syscalls.Supported("pipe", Pipe), - 23: syscalls.Supported("select", Select), - 24: syscalls.Supported("sched_yield", SchedYield), - 25: syscalls.Supported("mremap", Mremap), - 26: syscalls.PartiallySupported("msync", Msync, "Full data flush is not guaranteed at this time.", nil), - 27: syscalls.PartiallySupported("mincore", Mincore, "Stub implementation. The sandbox does not have access to this information. Reports all mapped pages are resident.", nil), - 28: syscalls.PartiallySupported("madvise", Madvise, "Options MADV_DONTNEED, MADV_DONTFORK are supported. Other advice is ignored.", nil), - 29: syscalls.PartiallySupported("shmget", Shmget, "Option SHM_HUGETLB is not supported.", nil), - 30: syscalls.PartiallySupported("shmat", Shmat, "Option SHM_RND is not supported.", nil), - 31: syscalls.PartiallySupported("shmctl", Shmctl, "Options SHM_LOCK, SHM_UNLOCK are not supported.", nil), - 32: syscalls.Supported("dup", Dup), - 33: syscalls.Supported("dup2", Dup2), - 34: syscalls.Supported("pause", Pause), - 35: syscalls.Supported("nanosleep", Nanosleep), - 36: syscalls.Supported("getitimer", Getitimer), - 37: syscalls.Supported("alarm", Alarm), - 38: syscalls.Supported("setitimer", Setitimer), - 39: syscalls.Supported("getpid", Getpid), - 40: syscalls.Supported("sendfile", Sendfile), - 41: syscalls.PartiallySupported("socket", Socket, "Limited support for AF_NETLINK, NETLINK_ROUTE sockets. Limited support for SOCK_RAW.", nil), - 42: syscalls.Supported("connect", Connect), - 43: syscalls.Supported("accept", Accept), - 44: syscalls.Supported("sendto", SendTo), - 45: syscalls.Supported("recvfrom", RecvFrom), - 46: syscalls.Supported("sendmsg", SendMsg), - 47: syscalls.PartiallySupported("recvmsg", RecvMsg, "Not all flags and control messages are supported.", nil), - 48: syscalls.PartiallySupported("shutdown", Shutdown, "Not all flags and control messages are supported.", nil), - 49: syscalls.PartiallySupported("bind", Bind, "Autobind for abstract Unix sockets is not supported.", nil), - 50: syscalls.Supported("listen", Listen), - 51: syscalls.Supported("getsockname", GetSockName), - 52: syscalls.Supported("getpeername", GetPeerName), - 53: syscalls.Supported("socketpair", SocketPair), - 54: syscalls.PartiallySupported("setsockopt", SetSockOpt, "Not all socket options are supported.", nil), - 55: syscalls.PartiallySupported("getsockopt", GetSockOpt, "Not all socket options are supported.", nil), - 56: syscalls.PartiallySupported("clone", Clone, "Mount namespace (CLONE_NEWNS) not supported. Options CLONE_PARENT, CLONE_SYSVSEM not supported.", nil), - 57: syscalls.Supported("fork", Fork), - 58: syscalls.Supported("vfork", Vfork), - 59: syscalls.Supported("execve", Execve), - 60: syscalls.Supported("exit", Exit), - 61: syscalls.Supported("wait4", Wait4), - 62: syscalls.Supported("kill", Kill), - 63: syscalls.Supported("uname", Uname), - 64: syscalls.Supported("semget", Semget), - 65: syscalls.PartiallySupported("semop", Semop, "Option SEM_UNDO not supported.", nil), - 66: syscalls.PartiallySupported("semctl", Semctl, "Options IPC_INFO, SEM_INFO, IPC_STAT, SEM_STAT, SEM_STAT_ANY, GETNCNT, GETZCNT not supported.", nil), - 67: syscalls.Supported("shmdt", Shmdt), - 68: syscalls.ErrorWithEvent("msgget", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) - 69: syscalls.ErrorWithEvent("msgsnd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) - 70: syscalls.ErrorWithEvent("msgrcv", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) - 71: syscalls.ErrorWithEvent("msgctl", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) - 72: syscalls.PartiallySupported("fcntl", Fcntl, "Not all options are supported.", nil), - 73: syscalls.PartiallySupported("flock", Flock, "Locks are held within the sandbox only.", nil), - 74: syscalls.PartiallySupported("fsync", Fsync, "Full data flush is not guaranteed at this time.", nil), - 75: syscalls.PartiallySupported("fdatasync", Fdatasync, "Full data flush is not guaranteed at this time.", nil), - 76: syscalls.Supported("truncate", Truncate), - 77: syscalls.Supported("ftruncate", Ftruncate), - 78: syscalls.Supported("getdents", Getdents), - 79: syscalls.Supported("getcwd", Getcwd), - 80: syscalls.Supported("chdir", Chdir), - 81: syscalls.Supported("fchdir", Fchdir), - 82: syscalls.Supported("rename", Rename), - 83: syscalls.Supported("mkdir", Mkdir), - 84: syscalls.Supported("rmdir", Rmdir), - 85: syscalls.Supported("creat", Creat), - 86: syscalls.Supported("link", Link), - 87: syscalls.Supported("unlink", Unlink), - 88: syscalls.Supported("symlink", Symlink), - 89: syscalls.Supported("readlink", Readlink), - 90: syscalls.Supported("chmod", Chmod), - 91: syscalls.PartiallySupported("fchmod", Fchmod, "Options S_ISUID and S_ISGID not supported.", nil), - 92: syscalls.Supported("chown", Chown), - 93: syscalls.Supported("fchown", Fchown), - 94: syscalls.Supported("lchown", Lchown), - 95: syscalls.Supported("umask", Umask), - 96: syscalls.Supported("gettimeofday", Gettimeofday), - 97: syscalls.Supported("getrlimit", Getrlimit), - 98: syscalls.PartiallySupported("getrusage", Getrusage, "Fields ru_maxrss, ru_minflt, ru_majflt, ru_inblock, ru_oublock are not supported. Fields ru_utime and ru_stime have low precision.", nil), - 99: syscalls.PartiallySupported("sysinfo", Sysinfo, "Fields loads, sharedram, bufferram, totalswap, freeswap, totalhigh, freehigh not supported.", nil), - 100: syscalls.Supported("times", Times), - 101: syscalls.PartiallySupported("ptrace", Ptrace, "Options PTRACE_PEEKSIGINFO, PTRACE_SECCOMP_GET_FILTER not supported.", nil), - 102: syscalls.Supported("getuid", Getuid), - 103: syscalls.PartiallySupported("syslog", Syslog, "Outputs a dummy message for security reasons.", nil), - 104: syscalls.Supported("getgid", Getgid), - 105: syscalls.Supported("setuid", Setuid), - 106: syscalls.Supported("setgid", Setgid), - 107: syscalls.Supported("geteuid", Geteuid), - 108: syscalls.Supported("getegid", Getegid), - 109: syscalls.Supported("setpgid", Setpgid), - 110: syscalls.Supported("getppid", Getppid), - 111: syscalls.Supported("getpgrp", Getpgrp), - 112: syscalls.Supported("setsid", Setsid), - 113: syscalls.Supported("setreuid", Setreuid), - 114: syscalls.Supported("setregid", Setregid), - 115: syscalls.Supported("getgroups", Getgroups), - 116: syscalls.Supported("setgroups", Setgroups), - 117: syscalls.Supported("setresuid", Setresuid), - 118: syscalls.Supported("getresuid", Getresuid), - 119: syscalls.Supported("setresgid", Setresgid), - 120: syscalls.Supported("getresgid", Getresgid), - 121: syscalls.Supported("getpgid", Getpgid), - 122: syscalls.ErrorWithEvent("setfsuid", syserror.ENOSYS, "", []string{"gvisor.dev/issue/260"}), // TODO(b/112851702) - 123: syscalls.ErrorWithEvent("setfsgid", syserror.ENOSYS, "", []string{"gvisor.dev/issue/260"}), // TODO(b/112851702) - 124: syscalls.Supported("getsid", Getsid), - 125: syscalls.Supported("capget", Capget), - 126: syscalls.Supported("capset", Capset), - 127: syscalls.Supported("rt_sigpending", RtSigpending), - 128: syscalls.Supported("rt_sigtimedwait", RtSigtimedwait), - 129: syscalls.Supported("rt_sigqueueinfo", RtSigqueueinfo), - 130: syscalls.Supported("rt_sigsuspend", RtSigsuspend), - 131: syscalls.Supported("sigaltstack", Sigaltstack), - 132: syscalls.Supported("utime", Utime), - 133: syscalls.PartiallySupported("mknod", Mknod, "Device creation is not generally supported. Only regular file and FIFO creation are supported.", nil), - 134: syscalls.Error("uselib", syserror.ENOSYS, "Obsolete", nil), - 135: syscalls.ErrorWithEvent("personality", syserror.EINVAL, "Unable to change personality.", nil), - 136: syscalls.ErrorWithEvent("ustat", syserror.ENOSYS, "Needs filesystem support.", nil), - 137: syscalls.PartiallySupported("statfs", Statfs, "Depends on the backing file system implementation.", nil), - 138: syscalls.PartiallySupported("fstatfs", Fstatfs, "Depends on the backing file system implementation.", nil), - 139: syscalls.ErrorWithEvent("sysfs", syserror.ENOSYS, "", []string{"gvisor.dev/issue/165"}), - 140: syscalls.PartiallySupported("getpriority", Getpriority, "Stub implementation.", nil), - 141: syscalls.PartiallySupported("setpriority", Setpriority, "Stub implementation.", nil), - 142: syscalls.CapError("sched_setparam", linux.CAP_SYS_NICE, "", nil), - 143: syscalls.PartiallySupported("sched_getparam", SchedGetparam, "Stub implementation.", nil), - 144: syscalls.PartiallySupported("sched_setscheduler", SchedSetscheduler, "Stub implementation.", nil), - 145: syscalls.PartiallySupported("sched_getscheduler", SchedGetscheduler, "Stub implementation.", nil), - 146: syscalls.PartiallySupported("sched_get_priority_max", SchedGetPriorityMax, "Stub implementation.", nil), - 147: syscalls.PartiallySupported("sched_get_priority_min", SchedGetPriorityMin, "Stub implementation.", nil), - 148: syscalls.ErrorWithEvent("sched_rr_get_interval", syserror.EPERM, "", nil), - 149: syscalls.PartiallySupported("mlock", Mlock, "Stub implementation. The sandbox lacks appropriate permissions.", nil), - 150: syscalls.PartiallySupported("munlock", Munlock, "Stub implementation. The sandbox lacks appropriate permissions.", nil), - 151: syscalls.PartiallySupported("mlockall", Mlockall, "Stub implementation. The sandbox lacks appropriate permissions.", nil), - 152: syscalls.PartiallySupported("munlockall", Munlockall, "Stub implementation. The sandbox lacks appropriate permissions.", nil), - 153: syscalls.CapError("vhangup", linux.CAP_SYS_TTY_CONFIG, "", nil), - 154: syscalls.Error("modify_ldt", syserror.EPERM, "", nil), - 155: syscalls.Error("pivot_root", syserror.EPERM, "", nil), - 156: syscalls.Error("sysctl", syserror.EPERM, "Deprecated. Use /proc/sys instead.", nil), - 157: syscalls.PartiallySupported("prctl", Prctl, "Not all options are supported.", nil), - 158: syscalls.PartiallySupported("arch_prctl", ArchPrctl, "Options ARCH_GET_GS, ARCH_SET_GS not supported.", nil), - 159: syscalls.CapError("adjtimex", linux.CAP_SYS_TIME, "", nil), - 160: syscalls.PartiallySupported("setrlimit", Setrlimit, "Not all rlimits are enforced.", nil), - 161: syscalls.Supported("chroot", Chroot), - 162: syscalls.PartiallySupported("sync", Sync, "Full data flush is not guaranteed at this time.", nil), - 163: syscalls.CapError("acct", linux.CAP_SYS_PACCT, "", nil), - 164: syscalls.CapError("settimeofday", linux.CAP_SYS_TIME, "", nil), - 165: syscalls.PartiallySupported("mount", Mount, "Not all options or file systems are supported.", nil), - 166: syscalls.PartiallySupported("umount2", Umount2, "Not all options or file systems are supported.", nil), - 167: syscalls.CapError("swapon", linux.CAP_SYS_ADMIN, "", nil), - 168: syscalls.CapError("swapoff", linux.CAP_SYS_ADMIN, "", nil), - 169: syscalls.CapError("reboot", linux.CAP_SYS_BOOT, "", nil), - 170: syscalls.Supported("sethostname", Sethostname), - 171: syscalls.Supported("setdomainname", Setdomainname), - 172: syscalls.CapError("iopl", linux.CAP_SYS_RAWIO, "", nil), - 173: syscalls.CapError("ioperm", linux.CAP_SYS_RAWIO, "", nil), - 174: syscalls.CapError("create_module", linux.CAP_SYS_MODULE, "", nil), - 175: syscalls.CapError("init_module", linux.CAP_SYS_MODULE, "", nil), - 176: syscalls.CapError("delete_module", linux.CAP_SYS_MODULE, "", nil), - 177: syscalls.Error("get_kernel_syms", syserror.ENOSYS, "Not supported in Linux > 2.6.", nil), - 178: syscalls.Error("query_module", syserror.ENOSYS, "Not supported in Linux > 2.6.", nil), - 179: syscalls.CapError("quotactl", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_admin for most operations - 180: syscalls.Error("nfsservctl", syserror.ENOSYS, "Removed after Linux 3.1.", nil), - 181: syscalls.Error("getpmsg", syserror.ENOSYS, "Not implemented in Linux.", nil), - 182: syscalls.Error("putpmsg", syserror.ENOSYS, "Not implemented in Linux.", nil), - 183: syscalls.Error("afs_syscall", syserror.ENOSYS, "Not implemented in Linux.", nil), - 184: syscalls.Error("tuxcall", syserror.ENOSYS, "Not implemented in Linux.", nil), - 185: syscalls.Error("security", syserror.ENOSYS, "Not implemented in Linux.", nil), - 186: syscalls.Supported("gettid", Gettid), - 187: syscalls.Supported("readahead", Readahead), - 188: syscalls.PartiallySupported("setxattr", SetXattr, "Only supported for tmpfs.", nil), - 189: syscalls.PartiallySupported("lsetxattr", LSetXattr, "Only supported for tmpfs.", nil), - 190: syscalls.PartiallySupported("fsetxattr", FSetXattr, "Only supported for tmpfs.", nil), - 191: syscalls.PartiallySupported("getxattr", GetXattr, "Only supported for tmpfs.", nil), - 192: syscalls.PartiallySupported("lgetxattr", LGetXattr, "Only supported for tmpfs.", nil), - 193: syscalls.PartiallySupported("fgetxattr", FGetXattr, "Only supported for tmpfs.", nil), - 194: syscalls.PartiallySupported("listxattr", ListXattr, "Only supported for tmpfs", nil), - 195: syscalls.PartiallySupported("llistxattr", LListXattr, "Only supported for tmpfs", nil), - 196: syscalls.PartiallySupported("flistxattr", FListXattr, "Only supported for tmpfs", nil), - 197: syscalls.PartiallySupported("removexattr", RemoveXattr, "Only supported for tmpfs", nil), - 198: syscalls.PartiallySupported("lremovexattr", LRemoveXattr, "Only supported for tmpfs", nil), - 199: syscalls.PartiallySupported("fremovexattr", FRemoveXattr, "Only supported for tmpfs", nil), - 200: syscalls.Supported("tkill", Tkill), - 201: syscalls.Supported("time", Time), - 202: syscalls.PartiallySupported("futex", Futex, "Robust futexes not supported.", nil), - 203: syscalls.PartiallySupported("sched_setaffinity", SchedSetaffinity, "Stub implementation.", nil), - 204: syscalls.PartiallySupported("sched_getaffinity", SchedGetaffinity, "Stub implementation.", nil), - 205: syscalls.Error("set_thread_area", syserror.ENOSYS, "Expected to return ENOSYS on 64-bit", nil), - 206: syscalls.PartiallySupported("io_setup", IoSetup, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 207: syscalls.PartiallySupported("io_destroy", IoDestroy, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 208: syscalls.PartiallySupported("io_getevents", IoGetevents, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 209: syscalls.PartiallySupported("io_submit", IoSubmit, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 210: syscalls.PartiallySupported("io_cancel", IoCancel, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 211: syscalls.Error("get_thread_area", syserror.ENOSYS, "Expected to return ENOSYS on 64-bit", nil), - 212: syscalls.CapError("lookup_dcookie", linux.CAP_SYS_ADMIN, "", nil), - 213: syscalls.Supported("epoll_create", EpollCreate), - 214: syscalls.ErrorWithEvent("epoll_ctl_old", syserror.ENOSYS, "Deprecated.", nil), - 215: syscalls.ErrorWithEvent("epoll_wait_old", syserror.ENOSYS, "Deprecated.", nil), - 216: syscalls.ErrorWithEvent("remap_file_pages", syserror.ENOSYS, "Deprecated since Linux 3.16.", nil), - 217: syscalls.Supported("getdents64", Getdents64), - 218: syscalls.Supported("set_tid_address", SetTidAddress), - 219: syscalls.Supported("restart_syscall", RestartSyscall), - 220: syscalls.ErrorWithEvent("semtimedop", syserror.ENOSYS, "", []string{"gvisor.dev/issue/137"}), - 221: syscalls.PartiallySupported("fadvise64", Fadvise64, "Not all options are supported.", nil), - 222: syscalls.Supported("timer_create", TimerCreate), - 223: syscalls.Supported("timer_settime", TimerSettime), - 224: syscalls.Supported("timer_gettime", TimerGettime), - 225: syscalls.Supported("timer_getoverrun", TimerGetoverrun), - 226: syscalls.Supported("timer_delete", TimerDelete), - 227: syscalls.Supported("clock_settime", ClockSettime), - 228: syscalls.Supported("clock_gettime", ClockGettime), - 229: syscalls.Supported("clock_getres", ClockGetres), - 230: syscalls.Supported("clock_nanosleep", ClockNanosleep), - 231: syscalls.Supported("exit_group", ExitGroup), - 232: syscalls.Supported("epoll_wait", EpollWait), - 233: syscalls.Supported("epoll_ctl", EpollCtl), - 234: syscalls.Supported("tgkill", Tgkill), - 235: syscalls.Supported("utimes", Utimes), - 236: syscalls.Error("vserver", syserror.ENOSYS, "Not implemented by Linux", nil), - 237: syscalls.PartiallySupported("mbind", Mbind, "Stub implementation. Only a single NUMA node is advertised, and mempolicy is ignored accordingly, but mbind() will succeed and has effects reflected by get_mempolicy.", []string{"gvisor.dev/issue/262"}), - 238: syscalls.PartiallySupported("set_mempolicy", SetMempolicy, "Stub implementation.", nil), - 239: syscalls.PartiallySupported("get_mempolicy", GetMempolicy, "Stub implementation.", nil), - 240: syscalls.ErrorWithEvent("mq_open", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) - 241: syscalls.ErrorWithEvent("mq_unlink", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) - 242: syscalls.ErrorWithEvent("mq_timedsend", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) - 243: syscalls.ErrorWithEvent("mq_timedreceive", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) - 244: syscalls.ErrorWithEvent("mq_notify", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) - 245: syscalls.ErrorWithEvent("mq_getsetattr", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) - 246: syscalls.CapError("kexec_load", linux.CAP_SYS_BOOT, "", nil), - 247: syscalls.Supported("waitid", Waitid), - 248: syscalls.Error("add_key", syserror.EACCES, "Not available to user.", nil), - 249: syscalls.Error("request_key", syserror.EACCES, "Not available to user.", nil), - 250: syscalls.Error("keyctl", syserror.EACCES, "Not available to user.", nil), - 251: syscalls.CapError("ioprio_set", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_nice or cap_sys_admin (depending) - 252: syscalls.CapError("ioprio_get", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_nice or cap_sys_admin (depending) - 253: syscalls.PartiallySupported("inotify_init", InotifyInit, "inotify events are only available inside the sandbox.", nil), - 254: syscalls.PartiallySupported("inotify_add_watch", InotifyAddWatch, "inotify events are only available inside the sandbox.", nil), - 255: syscalls.PartiallySupported("inotify_rm_watch", InotifyRmWatch, "inotify events are only available inside the sandbox.", nil), - 256: syscalls.CapError("migrate_pages", linux.CAP_SYS_NICE, "", nil), - 257: syscalls.Supported("openat", Openat), - 258: syscalls.Supported("mkdirat", Mkdirat), - 259: syscalls.Supported("mknodat", Mknodat), - 260: syscalls.Supported("fchownat", Fchownat), - 261: syscalls.Supported("futimesat", Futimesat), - 262: syscalls.Supported("fstatat", Fstatat), - 263: syscalls.Supported("unlinkat", Unlinkat), - 264: syscalls.Supported("renameat", Renameat), - 265: syscalls.Supported("linkat", Linkat), - 266: syscalls.Supported("symlinkat", Symlinkat), - 267: syscalls.Supported("readlinkat", Readlinkat), - 268: syscalls.Supported("fchmodat", Fchmodat), - 269: syscalls.Supported("faccessat", Faccessat), - 270: syscalls.Supported("pselect", Pselect), - 271: syscalls.Supported("ppoll", Ppoll), - 272: syscalls.PartiallySupported("unshare", Unshare, "Mount, cgroup namespaces not supported. Network namespaces supported but must be empty.", nil), - 273: syscalls.Error("set_robust_list", syserror.ENOSYS, "Obsolete.", nil), - 274: syscalls.Error("get_robust_list", syserror.ENOSYS, "Obsolete.", nil), - 275: syscalls.Supported("splice", Splice), - 276: syscalls.Supported("tee", Tee), - 277: syscalls.PartiallySupported("sync_file_range", SyncFileRange, "Full data flush is not guaranteed at this time.", nil), - 278: syscalls.ErrorWithEvent("vmsplice", syserror.ENOSYS, "", []string{"gvisor.dev/issue/138"}), // TODO(b/29354098) - 279: syscalls.CapError("move_pages", linux.CAP_SYS_NICE, "", nil), // requires cap_sys_nice (mostly) - 280: syscalls.Supported("utimensat", Utimensat), - 281: syscalls.Supported("epoll_pwait", EpollPwait), - 282: syscalls.PartiallySupported("signalfd", Signalfd, "Semantics are slightly different.", []string{"gvisor.dev/issue/139"}), - 283: syscalls.Supported("timerfd_create", TimerfdCreate), - 284: syscalls.Supported("eventfd", Eventfd), - 285: syscalls.PartiallySupported("fallocate", Fallocate, "Not all options are supported.", nil), - 286: syscalls.Supported("timerfd_settime", TimerfdSettime), - 287: syscalls.Supported("timerfd_gettime", TimerfdGettime), - 288: syscalls.Supported("accept4", Accept4), - 289: syscalls.PartiallySupported("signalfd4", Signalfd4, "Semantics are slightly different.", []string{"gvisor.dev/issue/139"}), - 290: syscalls.Supported("eventfd2", Eventfd2), - 291: syscalls.Supported("epoll_create1", EpollCreate1), - 292: syscalls.Supported("dup3", Dup3), - 293: syscalls.Supported("pipe2", Pipe2), - 294: syscalls.Supported("inotify_init1", InotifyInit1), - 295: syscalls.Supported("preadv", Preadv), - 296: syscalls.Supported("pwritev", Pwritev), - 297: syscalls.Supported("rt_tgsigqueueinfo", RtTgsigqueueinfo), - 298: syscalls.ErrorWithEvent("perf_event_open", syserror.ENODEV, "No support for perf counters", nil), - 299: syscalls.PartiallySupported("recvmmsg", RecvMMsg, "Not all flags and control messages are supported.", nil), - 300: syscalls.ErrorWithEvent("fanotify_init", syserror.ENOSYS, "Needs CONFIG_FANOTIFY", nil), - 301: syscalls.ErrorWithEvent("fanotify_mark", syserror.ENOSYS, "Needs CONFIG_FANOTIFY", nil), - 302: syscalls.Supported("prlimit64", Prlimit64), - 303: syscalls.Error("name_to_handle_at", syserror.EOPNOTSUPP, "Not supported by gVisor filesystems", nil), - 304: syscalls.Error("open_by_handle_at", syserror.EOPNOTSUPP, "Not supported by gVisor filesystems", nil), - 305: syscalls.CapError("clock_adjtime", linux.CAP_SYS_TIME, "", nil), - 306: syscalls.PartiallySupported("syncfs", Syncfs, "Depends on backing file system.", nil), - 307: syscalls.PartiallySupported("sendmmsg", SendMMsg, "Not all flags and control messages are supported.", nil), - 308: syscalls.ErrorWithEvent("setns", syserror.EOPNOTSUPP, "Needs filesystem support", []string{"gvisor.dev/issue/140"}), // TODO(b/29354995) - 309: syscalls.Supported("getcpu", Getcpu), - 310: syscalls.ErrorWithEvent("process_vm_readv", syserror.ENOSYS, "", []string{"gvisor.dev/issue/158"}), - 311: syscalls.ErrorWithEvent("process_vm_writev", syserror.ENOSYS, "", []string{"gvisor.dev/issue/158"}), - 312: syscalls.CapError("kcmp", linux.CAP_SYS_PTRACE, "", nil), - 313: syscalls.CapError("finit_module", linux.CAP_SYS_MODULE, "", nil), - 314: syscalls.ErrorWithEvent("sched_setattr", syserror.ENOSYS, "gVisor does not implement a scheduler.", []string{"gvisor.dev/issue/264"}), // TODO(b/118902272) - 315: syscalls.ErrorWithEvent("sched_getattr", syserror.ENOSYS, "gVisor does not implement a scheduler.", []string{"gvisor.dev/issue/264"}), // TODO(b/118902272) - 316: syscalls.ErrorWithEvent("renameat2", syserror.ENOSYS, "", []string{"gvisor.dev/issue/263"}), // TODO(b/118902772) - 317: syscalls.Supported("seccomp", Seccomp), - 318: syscalls.Supported("getrandom", GetRandom), - 319: syscalls.Supported("memfd_create", MemfdCreate), - 320: syscalls.CapError("kexec_file_load", linux.CAP_SYS_BOOT, "", nil), - 321: syscalls.CapError("bpf", linux.CAP_SYS_ADMIN, "", nil), - 322: syscalls.Supported("execveat", Execveat), - 323: syscalls.ErrorWithEvent("userfaultfd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/266"}), // TODO(b/118906345) - 324: syscalls.ErrorWithEvent("membarrier", syserror.ENOSYS, "", []string{"gvisor.dev/issue/267"}), // TODO(gvisor.dev/issue/267) - 325: syscalls.PartiallySupported("mlock2", Mlock2, "Stub implementation. The sandbox lacks appropriate permissions.", nil), - - // Syscalls implemented after 325 are "backports" from versions - // of Linux after 4.4. - 326: syscalls.ErrorWithEvent("copy_file_range", syserror.ENOSYS, "", nil), - 327: syscalls.Supported("preadv2", Preadv2), - 328: syscalls.PartiallySupported("pwritev2", Pwritev2, "Flag RWF_HIPRI is not supported.", nil), - 329: syscalls.ErrorWithEvent("pkey_mprotect", syserror.ENOSYS, "", nil), - 330: syscalls.ErrorWithEvent("pkey_alloc", syserror.ENOSYS, "", nil), - 331: syscalls.ErrorWithEvent("pkey_free", syserror.ENOSYS, "", nil), - 332: syscalls.Supported("statx", Statx), - 333: syscalls.ErrorWithEvent("io_pgetevents", syserror.ENOSYS, "", nil), - 334: syscalls.PartiallySupported("rseq", RSeq, "Not supported on all platforms.", nil), - - // Linux skips ahead to syscall 424 to sync numbers between arches. - 424: syscalls.ErrorWithEvent("pidfd_send_signal", syserror.ENOSYS, "", nil), - 425: syscalls.ErrorWithEvent("io_uring_setup", syserror.ENOSYS, "", nil), - 426: syscalls.ErrorWithEvent("io_uring_enter", syserror.ENOSYS, "", nil), - 427: syscalls.ErrorWithEvent("io_uring_register", syserror.ENOSYS, "", nil), - 428: syscalls.ErrorWithEvent("open_tree", syserror.ENOSYS, "", nil), - 429: syscalls.ErrorWithEvent("move_mount", syserror.ENOSYS, "", nil), - 430: syscalls.ErrorWithEvent("fsopen", syserror.ENOSYS, "", nil), - 431: syscalls.ErrorWithEvent("fsconfig", syserror.ENOSYS, "", nil), - 432: syscalls.ErrorWithEvent("fsmount", syserror.ENOSYS, "", nil), - 433: syscalls.ErrorWithEvent("fspick", syserror.ENOSYS, "", nil), - 434: syscalls.ErrorWithEvent("pidfd_open", syserror.ENOSYS, "", nil), - 435: syscalls.ErrorWithEvent("clone3", syserror.ENOSYS, "", nil), - }, - - Emulate: map[usermem.Addr]uintptr{ - 0xffffffffff600000: 96, // vsyscall gettimeofday(2) - 0xffffffffff600400: 201, // vsyscall time(2) - 0xffffffffff600800: 309, // vsyscall getcpu(2) - }, - Missing: func(t *kernel.Task, sysno uintptr, args arch.SyscallArguments) (uintptr, error) { - t.Kernel().EmitUnimplementedEvent(t) - return 0, syserror.ENOSYS - }, -} diff --git a/pkg/sentry/syscalls/linux/linux64_arm64.go b/pkg/sentry/syscalls/linux/linux64_arm64.go deleted file mode 100644 index 7421619de..000000000 --- a/pkg/sentry/syscalls/linux/linux64_arm64.go +++ /dev/null @@ -1,340 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package linux - -import ( - "gvisor.dev/gvisor/pkg/abi" - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/syscalls" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -// ARM64 is a table of Linux arm64 syscall API with the corresponding syscall -// numbers from Linux 4.4. -var ARM64 = &kernel.SyscallTable{ - OS: abi.Linux, - Arch: arch.ARM64, - Version: kernel.Version{ - Sysname: LinuxSysname, - Release: LinuxRelease, - Version: LinuxVersion, - }, - AuditNumber: linux.AUDIT_ARCH_AARCH64, - Table: map[uintptr]kernel.Syscall{ - 0: syscalls.PartiallySupported("io_setup", IoSetup, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 1: syscalls.PartiallySupported("io_destroy", IoDestroy, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 2: syscalls.PartiallySupported("io_submit", IoSubmit, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 3: syscalls.PartiallySupported("io_cancel", IoCancel, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 4: syscalls.PartiallySupported("io_getevents", IoGetevents, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 5: syscalls.PartiallySupported("setxattr", SetXattr, "Only supported for tmpfs.", nil), - 6: syscalls.PartiallySupported("lsetxattr", LSetXattr, "Only supported for tmpfs.", nil), - 7: syscalls.PartiallySupported("fsetxattr", FSetXattr, "Only supported for tmpfs.", nil), - 8: syscalls.PartiallySupported("getxattr", GetXattr, "Only supported for tmpfs.", nil), - 9: syscalls.PartiallySupported("lgetxattr", LGetXattr, "Only supported for tmpfs.", nil), - 10: syscalls.PartiallySupported("fgetxattr", FGetXattr, "Only supported for tmpfs.", nil), - 11: syscalls.PartiallySupported("listxattr", ListXattr, "Only supported for tmpfs", nil), - 12: syscalls.PartiallySupported("llistxattr", LListXattr, "Only supported for tmpfs", nil), - 13: syscalls.PartiallySupported("flistxattr", FListXattr, "Only supported for tmpfs", nil), - 14: syscalls.PartiallySupported("removexattr", RemoveXattr, "Only supported for tmpfs", nil), - 15: syscalls.PartiallySupported("lremovexattr", LRemoveXattr, "Only supported for tmpfs", nil), - 16: syscalls.PartiallySupported("fremovexattr", FRemoveXattr, "Only supported for tmpfs", nil), - 17: syscalls.Supported("getcwd", Getcwd), - 18: syscalls.CapError("lookup_dcookie", linux.CAP_SYS_ADMIN, "", nil), - 19: syscalls.Supported("eventfd2", Eventfd2), - 20: syscalls.Supported("epoll_create1", EpollCreate1), - 21: syscalls.Supported("epoll_ctl", EpollCtl), - 22: syscalls.Supported("epoll_pwait", EpollPwait), - 23: syscalls.Supported("dup", Dup), - 24: syscalls.Supported("dup3", Dup3), - 25: syscalls.PartiallySupported("fcntl", Fcntl, "Not all options are supported.", nil), - 26: syscalls.Supported("inotify_init1", InotifyInit1), - 27: syscalls.PartiallySupported("inotify_add_watch", InotifyAddWatch, "inotify events are only available inside the sandbox.", nil), - 28: syscalls.PartiallySupported("inotify_rm_watch", InotifyRmWatch, "inotify events are only available inside the sandbox.", nil), - 29: syscalls.PartiallySupported("ioctl", Ioctl, "Only a few ioctls are implemented for backing devices and file systems.", nil), - 30: syscalls.CapError("ioprio_set", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_nice or cap_sys_admin (depending) - 31: syscalls.CapError("ioprio_get", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_nice or cap_sys_admin (depending) - 32: syscalls.PartiallySupported("flock", Flock, "Locks are held within the sandbox only.", nil), - 33: syscalls.Supported("mknodat", Mknodat), - 34: syscalls.Supported("mkdirat", Mkdirat), - 35: syscalls.Supported("unlinkat", Unlinkat), - 36: syscalls.Supported("symlinkat", Symlinkat), - 37: syscalls.Supported("linkat", Linkat), - 38: syscalls.Supported("renameat", Renameat), - 39: syscalls.PartiallySupported("umount2", Umount2, "Not all options or file systems are supported.", nil), - 40: syscalls.PartiallySupported("mount", Mount, "Not all options or file systems are supported.", nil), - 41: syscalls.Error("pivot_root", syserror.EPERM, "", nil), - 42: syscalls.Error("nfsservctl", syserror.ENOSYS, "Removed after Linux 3.1.", nil), - 43: syscalls.PartiallySupported("statfs", Statfs, "Depends on the backing file system implementation.", nil), - 44: syscalls.PartiallySupported("fstatfs", Fstatfs, "Depends on the backing file system implementation.", nil), - 45: syscalls.Supported("truncate", Truncate), - 46: syscalls.Supported("ftruncate", Ftruncate), - 47: syscalls.PartiallySupported("fallocate", Fallocate, "Not all options are supported.", nil), - 48: syscalls.Supported("faccessat", Faccessat), - 49: syscalls.Supported("chdir", Chdir), - 50: syscalls.Supported("fchdir", Fchdir), - 51: syscalls.Supported("chroot", Chroot), - 52: syscalls.PartiallySupported("fchmod", Fchmod, "Options S_ISUID and S_ISGID not supported.", nil), - 53: syscalls.Supported("fchmodat", Fchmodat), - 54: syscalls.Supported("fchownat", Fchownat), - 55: syscalls.Supported("fchown", Fchown), - 56: syscalls.Supported("openat", Openat), - 57: syscalls.Supported("close", Close), - 58: syscalls.CapError("vhangup", linux.CAP_SYS_TTY_CONFIG, "", nil), - 59: syscalls.Supported("pipe2", Pipe2), - 60: syscalls.CapError("quotactl", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_admin for most operations - 61: syscalls.Supported("getdents64", Getdents64), - 62: syscalls.Supported("lseek", Lseek), - 63: syscalls.Supported("read", Read), - 64: syscalls.Supported("write", Write), - 65: syscalls.Supported("readv", Readv), - 66: syscalls.Supported("writev", Writev), - 67: syscalls.Supported("pread64", Pread64), - 68: syscalls.Supported("pwrite64", Pwrite64), - 69: syscalls.Supported("preadv", Preadv), - 70: syscalls.Supported("pwritev", Pwritev), - 71: syscalls.Supported("sendfile", Sendfile), - 72: syscalls.Supported("pselect", Pselect), - 73: syscalls.Supported("ppoll", Ppoll), - 74: syscalls.PartiallySupported("signalfd4", Signalfd4, "Semantics are slightly different.", []string{"gvisor.dev/issue/139"}), - 75: syscalls.ErrorWithEvent("vmsplice", syserror.ENOSYS, "", []string{"gvisor.dev/issue/138"}), // TODO(b/29354098) - 76: syscalls.PartiallySupported("splice", Splice, "Stub implementation.", []string{"gvisor.dev/issue/138"}), // TODO(b/29354098) - 77: syscalls.Supported("tee", Tee), - 78: syscalls.Supported("readlinkat", Readlinkat), - 79: syscalls.Supported("fstatat", Fstatat), - 80: syscalls.Supported("fstat", Fstat), - 81: syscalls.PartiallySupported("sync", Sync, "Full data flush is not guaranteed at this time.", nil), - 82: syscalls.PartiallySupported("fsync", Fsync, "Full data flush is not guaranteed at this time.", nil), - 83: syscalls.PartiallySupported("fdatasync", Fdatasync, "Full data flush is not guaranteed at this time.", nil), - 84: syscalls.PartiallySupported("sync_file_range", SyncFileRange, "Full data flush is not guaranteed at this time.", nil), - 85: syscalls.Supported("timerfd_create", TimerfdCreate), - 86: syscalls.Supported("timerfd_settime", TimerfdSettime), - 87: syscalls.Supported("timerfd_gettime", TimerfdGettime), - 88: syscalls.Supported("utimensat", Utimensat), - 89: syscalls.CapError("acct", linux.CAP_SYS_PACCT, "", nil), - 90: syscalls.Supported("capget", Capget), - 91: syscalls.Supported("capset", Capset), - 92: syscalls.ErrorWithEvent("personality", syserror.EINVAL, "Unable to change personality.", nil), - 93: syscalls.Supported("exit", Exit), - 94: syscalls.Supported("exit_group", ExitGroup), - 95: syscalls.Supported("waitid", Waitid), - 96: syscalls.Supported("set_tid_address", SetTidAddress), - 97: syscalls.PartiallySupported("unshare", Unshare, "Mount, cgroup namespaces not supported. Network namespaces supported but must be empty.", nil), - 98: syscalls.PartiallySupported("futex", Futex, "Robust futexes not supported.", nil), - 99: syscalls.Error("set_robust_list", syserror.ENOSYS, "Obsolete.", nil), - 100: syscalls.Error("get_robust_list", syserror.ENOSYS, "Obsolete.", nil), - 101: syscalls.Supported("nanosleep", Nanosleep), - 102: syscalls.Supported("getitimer", Getitimer), - 103: syscalls.Supported("setitimer", Setitimer), - 104: syscalls.CapError("kexec_load", linux.CAP_SYS_BOOT, "", nil), - 105: syscalls.CapError("init_module", linux.CAP_SYS_MODULE, "", nil), - 106: syscalls.CapError("delete_module", linux.CAP_SYS_MODULE, "", nil), - 107: syscalls.Supported("timer_create", TimerCreate), - 108: syscalls.Supported("timer_gettime", TimerGettime), - 109: syscalls.Supported("timer_getoverrun", TimerGetoverrun), - 110: syscalls.Supported("timer_settime", TimerSettime), - 111: syscalls.Supported("timer_delete", TimerDelete), - 112: syscalls.Supported("clock_settime", ClockSettime), - 113: syscalls.Supported("clock_gettime", ClockGettime), - 114: syscalls.Supported("clock_getres", ClockGetres), - 115: syscalls.Supported("clock_nanosleep", ClockNanosleep), - 116: syscalls.PartiallySupported("syslog", Syslog, "Outputs a dummy message for security reasons.", nil), - 117: syscalls.PartiallySupported("ptrace", Ptrace, "Options PTRACE_PEEKSIGINFO, PTRACE_SECCOMP_GET_FILTER not supported.", nil), - 118: syscalls.CapError("sched_setparam", linux.CAP_SYS_NICE, "", nil), - 119: syscalls.PartiallySupported("sched_setscheduler", SchedSetscheduler, "Stub implementation.", nil), - 120: syscalls.PartiallySupported("sched_getscheduler", SchedGetscheduler, "Stub implementation.", nil), - 121: syscalls.PartiallySupported("sched_getparam", SchedGetparam, "Stub implementation.", nil), - 122: syscalls.PartiallySupported("sched_setaffinity", SchedSetaffinity, "Stub implementation.", nil), - 123: syscalls.PartiallySupported("sched_getaffinity", SchedGetaffinity, "Stub implementation.", nil), - 124: syscalls.Supported("sched_yield", SchedYield), - 125: syscalls.PartiallySupported("sched_get_priority_max", SchedGetPriorityMax, "Stub implementation.", nil), - 126: syscalls.PartiallySupported("sched_get_priority_min", SchedGetPriorityMin, "Stub implementation.", nil), - 127: syscalls.ErrorWithEvent("sched_rr_get_interval", syserror.EPERM, "", nil), - 128: syscalls.Supported("restart_syscall", RestartSyscall), - 129: syscalls.Supported("kill", Kill), - 130: syscalls.Supported("tkill", Tkill), - 131: syscalls.Supported("tgkill", Tgkill), - 132: syscalls.Supported("sigaltstack", Sigaltstack), - 133: syscalls.Supported("rt_sigsuspend", RtSigsuspend), - 134: syscalls.Supported("rt_sigaction", RtSigaction), - 135: syscalls.Supported("rt_sigprocmask", RtSigprocmask), - 136: syscalls.Supported("rt_sigpending", RtSigpending), - 137: syscalls.Supported("rt_sigtimedwait", RtSigtimedwait), - 138: syscalls.Supported("rt_sigqueueinfo", RtSigqueueinfo), - 139: syscalls.Supported("rt_sigreturn", RtSigreturn), - 140: syscalls.PartiallySupported("setpriority", Setpriority, "Stub implementation.", nil), - 141: syscalls.PartiallySupported("getpriority", Getpriority, "Stub implementation.", nil), - 142: syscalls.CapError("reboot", linux.CAP_SYS_BOOT, "", nil), - 143: syscalls.Supported("setregid", Setregid), - 144: syscalls.Supported("setgid", Setgid), - 145: syscalls.Supported("setreuid", Setreuid), - 146: syscalls.Supported("setuid", Setuid), - 147: syscalls.Supported("setresuid", Setresuid), - 148: syscalls.Supported("getresuid", Getresuid), - 149: syscalls.Supported("setresgid", Setresgid), - 150: syscalls.Supported("getresgid", Getresgid), - 151: syscalls.ErrorWithEvent("setfsuid", syserror.ENOSYS, "", []string{"gvisor.dev/issue/260"}), // TODO(b/112851702) - 152: syscalls.ErrorWithEvent("setfsgid", syserror.ENOSYS, "", []string{"gvisor.dev/issue/260"}), // TODO(b/112851702) - 153: syscalls.Supported("times", Times), - 154: syscalls.Supported("setpgid", Setpgid), - 155: syscalls.Supported("getpgid", Getpgid), - 156: syscalls.Supported("getsid", Getsid), - 157: syscalls.Supported("setsid", Setsid), - 158: syscalls.Supported("getgroups", Getgroups), - 159: syscalls.Supported("setgroups", Setgroups), - 160: syscalls.Supported("uname", Uname), - 161: syscalls.Supported("sethostname", Sethostname), - 162: syscalls.Supported("setdomainname", Setdomainname), - 163: syscalls.Supported("getrlimit", Getrlimit), - 164: syscalls.PartiallySupported("setrlimit", Setrlimit, "Not all rlimits are enforced.", nil), - 165: syscalls.PartiallySupported("getrusage", Getrusage, "Fields ru_maxrss, ru_minflt, ru_majflt, ru_inblock, ru_oublock are not supported. Fields ru_utime and ru_stime have low precision.", nil), - 166: syscalls.Supported("umask", Umask), - 167: syscalls.PartiallySupported("prctl", Prctl, "Not all options are supported.", nil), - 168: syscalls.Supported("getcpu", Getcpu), - 169: syscalls.Supported("gettimeofday", Gettimeofday), - 170: syscalls.CapError("settimeofday", linux.CAP_SYS_TIME, "", nil), - 171: syscalls.CapError("adjtimex", linux.CAP_SYS_TIME, "", nil), - 172: syscalls.Supported("getpid", Getpid), - 173: syscalls.Supported("getppid", Getppid), - 174: syscalls.Supported("getuid", Getuid), - 175: syscalls.Supported("geteuid", Geteuid), - 176: syscalls.Supported("getgid", Getgid), - 177: syscalls.Supported("getegid", Getegid), - 178: syscalls.Supported("gettid", Gettid), - 179: syscalls.PartiallySupported("sysinfo", Sysinfo, "Fields loads, sharedram, bufferram, totalswap, freeswap, totalhigh, freehigh not supported.", nil), - 180: syscalls.ErrorWithEvent("mq_open", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) - 181: syscalls.ErrorWithEvent("mq_unlink", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) - 182: syscalls.ErrorWithEvent("mq_timedsend", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) - 183: syscalls.ErrorWithEvent("mq_timedreceive", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) - 184: syscalls.ErrorWithEvent("mq_notify", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) - 185: syscalls.ErrorWithEvent("mq_getsetattr", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) - 186: syscalls.ErrorWithEvent("msgget", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) - 187: syscalls.ErrorWithEvent("msgctl", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) - 188: syscalls.ErrorWithEvent("msgrcv", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) - 189: syscalls.ErrorWithEvent("msgsnd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) - 190: syscalls.Supported("semget", Semget), - 191: syscalls.PartiallySupported("semctl", Semctl, "Options IPC_INFO, SEM_INFO, IPC_STAT, SEM_STAT, SEM_STAT_ANY, GETNCNT, GETZCNT not supported.", nil), - 192: syscalls.ErrorWithEvent("semtimedop", syserror.ENOSYS, "", []string{"gvisor.dev/issue/137"}), - 193: syscalls.PartiallySupported("semop", Semop, "Option SEM_UNDO not supported.", nil), - 194: syscalls.PartiallySupported("shmget", Shmget, "Option SHM_HUGETLB is not supported.", nil), - 195: syscalls.PartiallySupported("shmctl", Shmctl, "Options SHM_LOCK, SHM_UNLOCK are not supported.", nil), - 196: syscalls.PartiallySupported("shmat", Shmat, "Option SHM_RND is not supported.", nil), - 197: syscalls.Supported("shmdt", Shmdt), - 198: syscalls.PartiallySupported("socket", Socket, "Limited support for AF_NETLINK, NETLINK_ROUTE sockets. Limited support for SOCK_RAW.", nil), - 199: syscalls.Supported("socketpair", SocketPair), - 200: syscalls.PartiallySupported("bind", Bind, "Autobind for abstract Unix sockets is not supported.", nil), - 201: syscalls.Supported("listen", Listen), - 202: syscalls.Supported("accept", Accept), - 203: syscalls.Supported("connect", Connect), - 204: syscalls.Supported("getsockname", GetSockName), - 205: syscalls.Supported("getpeername", GetPeerName), - 206: syscalls.Supported("sendto", SendTo), - 207: syscalls.Supported("recvfrom", RecvFrom), - 208: syscalls.PartiallySupported("setsockopt", SetSockOpt, "Not all socket options are supported.", nil), - 209: syscalls.PartiallySupported("getsockopt", GetSockOpt, "Not all socket options are supported.", nil), - 210: syscalls.PartiallySupported("shutdown", Shutdown, "Not all flags and control messages are supported.", nil), - 211: syscalls.Supported("sendmsg", SendMsg), - 212: syscalls.PartiallySupported("recvmsg", RecvMsg, "Not all flags and control messages are supported.", nil), - 213: syscalls.Supported("readahead", Readahead), - 214: syscalls.Supported("brk", Brk), - 215: syscalls.Supported("munmap", Munmap), - 216: syscalls.Supported("mremap", Mremap), - 217: syscalls.Error("add_key", syserror.EACCES, "Not available to user.", nil), - 218: syscalls.Error("request_key", syserror.EACCES, "Not available to user.", nil), - 219: syscalls.Error("keyctl", syserror.EACCES, "Not available to user.", nil), - 220: syscalls.PartiallySupported("clone", Clone, "Mount namespace (CLONE_NEWNS) not supported. Options CLONE_PARENT, CLONE_SYSVSEM not supported.", nil), - 221: syscalls.Supported("execve", Execve), - 222: syscalls.PartiallySupported("mmap", Mmap, "Generally supported with exceptions. Options MAP_FIXED_NOREPLACE, MAP_SHARED_VALIDATE, MAP_SYNC MAP_GROWSDOWN, MAP_HUGETLB are not supported.", nil), - 223: syscalls.PartiallySupported("fadvise64", Fadvise64, "Not all options are supported.", nil), - 224: syscalls.CapError("swapon", linux.CAP_SYS_ADMIN, "", nil), - 225: syscalls.CapError("swapoff", linux.CAP_SYS_ADMIN, "", nil), - 226: syscalls.Supported("mprotect", Mprotect), - 227: syscalls.PartiallySupported("msync", Msync, "Full data flush is not guaranteed at this time.", nil), - 228: syscalls.PartiallySupported("mlock", Mlock, "Stub implementation. The sandbox lacks appropriate permissions.", nil), - 229: syscalls.PartiallySupported("munlock", Munlock, "Stub implementation. The sandbox lacks appropriate permissions.", nil), - 230: syscalls.PartiallySupported("mlockall", Mlockall, "Stub implementation. The sandbox lacks appropriate permissions.", nil), - 231: syscalls.PartiallySupported("munlockall", Munlockall, "Stub implementation. The sandbox lacks appropriate permissions.", nil), - 232: syscalls.PartiallySupported("mincore", Mincore, "Stub implementation. The sandbox does not have access to this information. Reports all mapped pages are resident.", nil), - 233: syscalls.PartiallySupported("madvise", Madvise, "Options MADV_DONTNEED, MADV_DONTFORK are supported. Other advice is ignored.", nil), - 234: syscalls.ErrorWithEvent("remap_file_pages", syserror.ENOSYS, "Deprecated since Linux 3.16.", nil), - 235: syscalls.PartiallySupported("mbind", Mbind, "Stub implementation. Only a single NUMA node is advertised, and mempolicy is ignored accordingly, but mbind() will succeed and has effects reflected by get_mempolicy.", []string{"gvisor.dev/issue/262"}), - 236: syscalls.PartiallySupported("get_mempolicy", GetMempolicy, "Stub implementation.", nil), - 237: syscalls.PartiallySupported("set_mempolicy", SetMempolicy, "Stub implementation.", nil), - 238: syscalls.CapError("migrate_pages", linux.CAP_SYS_NICE, "", nil), - 239: syscalls.CapError("move_pages", linux.CAP_SYS_NICE, "", nil), // requires cap_sys_nice (mostly) - 240: syscalls.Supported("rt_tgsigqueueinfo", RtTgsigqueueinfo), - 241: syscalls.ErrorWithEvent("perf_event_open", syserror.ENODEV, "No support for perf counters", nil), - 242: syscalls.Supported("accept4", Accept4), - 243: syscalls.PartiallySupported("recvmmsg", RecvMMsg, "Not all flags and control messages are supported.", nil), - 260: syscalls.Supported("wait4", Wait4), - 261: syscalls.Supported("prlimit64", Prlimit64), - 262: syscalls.ErrorWithEvent("fanotify_init", syserror.ENOSYS, "Needs CONFIG_FANOTIFY", nil), - 263: syscalls.ErrorWithEvent("fanotify_mark", syserror.ENOSYS, "Needs CONFIG_FANOTIFY", nil), - 264: syscalls.Error("name_to_handle_at", syserror.EOPNOTSUPP, "Not supported by gVisor filesystems", nil), - 265: syscalls.Error("open_by_handle_at", syserror.EOPNOTSUPP, "Not supported by gVisor filesystems", nil), - 266: syscalls.CapError("clock_adjtime", linux.CAP_SYS_TIME, "", nil), - 267: syscalls.PartiallySupported("syncfs", Syncfs, "Depends on backing file system.", nil), - 268: syscalls.ErrorWithEvent("setns", syserror.EOPNOTSUPP, "Needs filesystem support", []string{"gvisor.dev/issue/140"}), // TODO(b/29354995) - 269: syscalls.PartiallySupported("sendmmsg", SendMMsg, "Not all flags and control messages are supported.", nil), - 270: syscalls.ErrorWithEvent("process_vm_readv", syserror.ENOSYS, "", []string{"gvisor.dev/issue/158"}), - 271: syscalls.ErrorWithEvent("process_vm_writev", syserror.ENOSYS, "", []string{"gvisor.dev/issue/158"}), - 272: syscalls.CapError("kcmp", linux.CAP_SYS_PTRACE, "", nil), - 273: syscalls.CapError("finit_module", linux.CAP_SYS_MODULE, "", nil), - 274: syscalls.ErrorWithEvent("sched_setattr", syserror.ENOSYS, "gVisor does not implement a scheduler.", []string{"gvisor.dev/issue/264"}), // TODO(b/118902272) - 275: syscalls.ErrorWithEvent("sched_getattr", syserror.ENOSYS, "gVisor does not implement a scheduler.", []string{"gvisor.dev/issue/264"}), // TODO(b/118902272) - 276: syscalls.ErrorWithEvent("renameat2", syserror.ENOSYS, "", []string{"gvisor.dev/issue/263"}), // TODO(b/118902772) - 277: syscalls.Supported("seccomp", Seccomp), - 278: syscalls.Supported("getrandom", GetRandom), - 279: syscalls.Supported("memfd_create", MemfdCreate), - 280: syscalls.CapError("bpf", linux.CAP_SYS_ADMIN, "", nil), - 281: syscalls.Supported("execveat", Execveat), - 282: syscalls.ErrorWithEvent("userfaultfd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/266"}), // TODO(b/118906345) - 283: syscalls.ErrorWithEvent("membarrier", syserror.ENOSYS, "", []string{"gvisor.dev/issue/267"}), // TODO(gvisor.dev/issue/267) - 284: syscalls.PartiallySupported("mlock2", Mlock2, "Stub implementation. The sandbox lacks appropriate permissions.", nil), - - // Syscalls after 284 are "backports" from versions of Linux after 4.4. - 285: syscalls.ErrorWithEvent("copy_file_range", syserror.ENOSYS, "", nil), - 286: syscalls.Supported("preadv2", Preadv2), - 287: syscalls.PartiallySupported("pwritev2", Pwritev2, "Flag RWF_HIPRI is not supported.", nil), - 288: syscalls.ErrorWithEvent("pkey_mprotect", syserror.ENOSYS, "", nil), - 289: syscalls.ErrorWithEvent("pkey_alloc", syserror.ENOSYS, "", nil), - 290: syscalls.ErrorWithEvent("pkey_free", syserror.ENOSYS, "", nil), - 291: syscalls.Supported("statx", Statx), - 292: syscalls.ErrorWithEvent("io_pgetevents", syserror.ENOSYS, "", nil), - 293: syscalls.PartiallySupported("rseq", RSeq, "Not supported on all platforms.", nil), - - // Linux skips ahead to syscall 424 to sync numbers between arches. - 424: syscalls.ErrorWithEvent("pidfd_send_signal", syserror.ENOSYS, "", nil), - 425: syscalls.ErrorWithEvent("io_uring_setup", syserror.ENOSYS, "", nil), - 426: syscalls.ErrorWithEvent("io_uring_enter", syserror.ENOSYS, "", nil), - 427: syscalls.ErrorWithEvent("io_uring_register", syserror.ENOSYS, "", nil), - 428: syscalls.ErrorWithEvent("open_tree", syserror.ENOSYS, "", nil), - 429: syscalls.ErrorWithEvent("move_mount", syserror.ENOSYS, "", nil), - 430: syscalls.ErrorWithEvent("fsopen", syserror.ENOSYS, "", nil), - 431: syscalls.ErrorWithEvent("fsconfig", syserror.ENOSYS, "", nil), - 432: syscalls.ErrorWithEvent("fsmount", syserror.ENOSYS, "", nil), - 433: syscalls.ErrorWithEvent("fspick", syserror.ENOSYS, "", nil), - 434: syscalls.ErrorWithEvent("pidfd_open", syserror.ENOSYS, "", nil), - 435: syscalls.ErrorWithEvent("clone3", syserror.ENOSYS, "", nil), - }, - Emulate: map[usermem.Addr]uintptr{}, - - Missing: func(t *kernel.Task, sysno uintptr, args arch.SyscallArguments) (uintptr, error) { - t.Kernel().EmitUnimplementedEvent(t) - return 0, syserror.ENOSYS - }, -} diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD index ffca627d4..4c7b8f819 100644 --- a/pkg/sentry/syscalls/linux/vfs2/BUILD +++ b/pkg/sentry/syscalls/linux/vfs2/BUILD @@ -13,9 +13,6 @@ go_library( "fscontext.go", "getdents.go", "ioctl.go", - "linux64.go", - "linux64_override_amd64.go", - "linux64_override_arm64.go", "mmap.go", "path.go", "pipe.go", @@ -28,6 +25,7 @@ go_library( "stat_arm64.go", "sync.go", "timerfd.go", + "vfs2.go", "xattr.go", ], marshal = True, diff --git a/pkg/sentry/syscalls/linux/vfs2/linux64.go b/pkg/sentry/syscalls/linux/vfs2/linux64.go deleted file mode 100644 index 19ee36081..000000000 --- a/pkg/sentry/syscalls/linux/vfs2/linux64.go +++ /dev/null @@ -1,16 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package vfs2 provides syscall implementations that use VFS2. -package vfs2 diff --git a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go deleted file mode 100644 index 47c5d18e7..000000000 --- a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go +++ /dev/null @@ -1,165 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build amd64 - -package vfs2 - -import ( - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/syscalls" -) - -// Override syscall table to add syscalls implementations from this package. -func Override(table map[uintptr]kernel.Syscall) { - table[0] = syscalls.Supported("read", Read) - table[1] = syscalls.Supported("write", Write) - table[2] = syscalls.Supported("open", Open) - table[3] = syscalls.Supported("close", Close) - table[4] = syscalls.Supported("stat", Stat) - table[5] = syscalls.Supported("fstat", Fstat) - table[6] = syscalls.Supported("lstat", Lstat) - table[7] = syscalls.Supported("poll", Poll) - table[8] = syscalls.Supported("lseek", Lseek) - table[9] = syscalls.Supported("mmap", Mmap) - table[16] = syscalls.Supported("ioctl", Ioctl) - table[17] = syscalls.Supported("pread64", Pread64) - table[18] = syscalls.Supported("pwrite64", Pwrite64) - table[19] = syscalls.Supported("readv", Readv) - table[20] = syscalls.Supported("writev", Writev) - table[21] = syscalls.Supported("access", Access) - table[22] = syscalls.Supported("pipe", Pipe) - table[23] = syscalls.Supported("select", Select) - table[32] = syscalls.Supported("dup", Dup) - table[33] = syscalls.Supported("dup2", Dup2) - delete(table, 40) // sendfile - table[41] = syscalls.Supported("socket", Socket) - table[42] = syscalls.Supported("connect", Connect) - table[43] = syscalls.Supported("accept", Accept) - table[44] = syscalls.Supported("sendto", SendTo) - table[45] = syscalls.Supported("recvfrom", RecvFrom) - table[46] = syscalls.Supported("sendmsg", SendMsg) - table[47] = syscalls.Supported("recvmsg", RecvMsg) - table[48] = syscalls.Supported("shutdown", Shutdown) - table[49] = syscalls.Supported("bind", Bind) - table[50] = syscalls.Supported("listen", Listen) - table[51] = syscalls.Supported("getsockname", GetSockName) - table[52] = syscalls.Supported("getpeername", GetPeerName) - table[53] = syscalls.Supported("socketpair", SocketPair) - table[54] = syscalls.Supported("setsockopt", SetSockOpt) - table[55] = syscalls.Supported("getsockopt", GetSockOpt) - table[59] = syscalls.Supported("execve", Execve) - table[72] = syscalls.Supported("fcntl", Fcntl) - delete(table, 73) // flock - table[74] = syscalls.Supported("fsync", Fsync) - table[75] = syscalls.Supported("fdatasync", Fdatasync) - table[76] = syscalls.Supported("truncate", Truncate) - table[77] = syscalls.Supported("ftruncate", Ftruncate) - table[78] = syscalls.Supported("getdents", Getdents) - table[79] = syscalls.Supported("getcwd", Getcwd) - table[80] = syscalls.Supported("chdir", Chdir) - table[81] = syscalls.Supported("fchdir", Fchdir) - table[82] = syscalls.Supported("rename", Rename) - table[83] = syscalls.Supported("mkdir", Mkdir) - table[84] = syscalls.Supported("rmdir", Rmdir) - table[85] = syscalls.Supported("creat", Creat) - table[86] = syscalls.Supported("link", Link) - table[87] = syscalls.Supported("unlink", Unlink) - table[88] = syscalls.Supported("symlink", Symlink) - table[89] = syscalls.Supported("readlink", Readlink) - table[90] = syscalls.Supported("chmod", Chmod) - table[91] = syscalls.Supported("fchmod", Fchmod) - table[92] = syscalls.Supported("chown", Chown) - table[93] = syscalls.Supported("fchown", Fchown) - table[94] = syscalls.Supported("lchown", Lchown) - table[132] = syscalls.Supported("utime", Utime) - table[133] = syscalls.Supported("mknod", Mknod) - table[137] = syscalls.Supported("statfs", Statfs) - table[138] = syscalls.Supported("fstatfs", Fstatfs) - table[161] = syscalls.Supported("chroot", Chroot) - table[162] = syscalls.Supported("sync", Sync) - delete(table, 165) // mount - delete(table, 166) // umount2 - delete(table, 187) // readahead - table[188] = syscalls.Supported("setxattr", Setxattr) - table[189] = syscalls.Supported("lsetxattr", Lsetxattr) - table[190] = syscalls.Supported("fsetxattr", Fsetxattr) - table[191] = syscalls.Supported("getxattr", Getxattr) - table[192] = syscalls.Supported("lgetxattr", Lgetxattr) - table[193] = syscalls.Supported("fgetxattr", Fgetxattr) - table[194] = syscalls.Supported("listxattr", Listxattr) - table[195] = syscalls.Supported("llistxattr", Llistxattr) - table[196] = syscalls.Supported("flistxattr", Flistxattr) - table[197] = syscalls.Supported("removexattr", Removexattr) - table[198] = syscalls.Supported("lremovexattr", Lremovexattr) - table[199] = syscalls.Supported("fremovexattr", Fremovexattr) - delete(table, 206) // io_setup - delete(table, 207) // io_destroy - delete(table, 208) // io_getevents - delete(table, 209) // io_submit - delete(table, 210) // io_cancel - table[213] = syscalls.Supported("epoll_create", EpollCreate) - table[217] = syscalls.Supported("getdents64", Getdents64) - delete(table, 221) // fdavise64 - table[232] = syscalls.Supported("epoll_wait", EpollWait) - table[233] = syscalls.Supported("epoll_ctl", EpollCtl) - table[235] = syscalls.Supported("utimes", Utimes) - delete(table, 253) // inotify_init - delete(table, 254) // inotify_add_watch - delete(table, 255) // inotify_rm_watch - table[257] = syscalls.Supported("openat", Openat) - table[258] = syscalls.Supported("mkdirat", Mkdirat) - table[259] = syscalls.Supported("mknodat", Mknodat) - table[260] = syscalls.Supported("fchownat", Fchownat) - table[261] = syscalls.Supported("futimens", Futimens) - table[262] = syscalls.Supported("newfstatat", Newfstatat) - table[263] = syscalls.Supported("unlinkat", Unlinkat) - table[264] = syscalls.Supported("renameat", Renameat) - table[265] = syscalls.Supported("linkat", Linkat) - table[266] = syscalls.Supported("symlinkat", Symlinkat) - table[267] = syscalls.Supported("readlinkat", Readlinkat) - table[268] = syscalls.Supported("fchmodat", Fchmodat) - table[269] = syscalls.Supported("faccessat", Faccessat) - table[270] = syscalls.Supported("pselect", Pselect) - table[271] = syscalls.Supported("ppoll", Ppoll) - delete(table, 275) // splice - delete(table, 276) // tee - table[277] = syscalls.Supported("sync_file_range", SyncFileRange) - table[280] = syscalls.Supported("utimensat", Utimensat) - table[281] = syscalls.Supported("epoll_pwait", EpollPwait) - delete(table, 282) // signalfd - table[283] = syscalls.Supported("timerfd_create", TimerfdCreate) - table[284] = syscalls.Supported("eventfd", Eventfd) - delete(table, 285) // fallocate - table[286] = syscalls.Supported("timerfd_settime", TimerfdSettime) - table[287] = syscalls.Supported("timerfd_gettime", TimerfdGettime) - table[288] = syscalls.Supported("accept4", Accept4) - delete(table, 289) // signalfd4 - table[290] = syscalls.Supported("eventfd2", Eventfd2) - table[291] = syscalls.Supported("epoll_create1", EpollCreate1) - table[292] = syscalls.Supported("dup3", Dup3) - table[293] = syscalls.Supported("pipe2", Pipe2) - delete(table, 294) // inotify_init1 - table[295] = syscalls.Supported("preadv", Preadv) - table[296] = syscalls.Supported("pwritev", Pwritev) - table[299] = syscalls.Supported("recvmmsg", RecvMMsg) - table[306] = syscalls.Supported("syncfs", Syncfs) - table[307] = syscalls.Supported("sendmmsg", SendMMsg) - table[316] = syscalls.Supported("renameat2", Renameat2) - delete(table, 319) // memfd_create - table[322] = syscalls.Supported("execveat", Execveat) - table[327] = syscalls.Supported("preadv2", Preadv2) - table[328] = syscalls.Supported("pwritev2", Pwritev2) - table[332] = syscalls.Supported("statx", Statx) -} diff --git a/pkg/sentry/syscalls/linux/vfs2/linux64_override_arm64.go b/pkg/sentry/syscalls/linux/vfs2/linux64_override_arm64.go deleted file mode 100644 index a6b367468..000000000 --- a/pkg/sentry/syscalls/linux/vfs2/linux64_override_arm64.go +++ /dev/null @@ -1,27 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build arm64 - -package vfs2 - -import ( - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/syscalls" -) - -// Override syscall table to add syscalls implementations from this package. -func Override(table map[uintptr]kernel.Syscall) { - table[63] = syscalls.Supported("read", Read) -} diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2.go b/pkg/sentry/syscalls/linux/vfs2/vfs2.go new file mode 100644 index 000000000..f1b697844 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/vfs2.go @@ -0,0 +1,172 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package vfs2 provides syscall implementations that use VFS2. +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/sentry/syscalls" + "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" +) + +// Override syscall table to add syscalls implementations from this package. +func Override() { + // Override AMD64. + s := linux.AMD64 + s.Table[0] = syscalls.Supported("read", Read) + s.Table[1] = syscalls.Supported("write", Write) + s.Table[2] = syscalls.Supported("open", Open) + s.Table[3] = syscalls.Supported("close", Close) + s.Table[4] = syscalls.Supported("stat", Stat) + s.Table[5] = syscalls.Supported("fstat", Fstat) + s.Table[6] = syscalls.Supported("lstat", Lstat) + s.Table[7] = syscalls.Supported("poll", Poll) + s.Table[8] = syscalls.Supported("lseek", Lseek) + s.Table[9] = syscalls.Supported("mmap", Mmap) + s.Table[16] = syscalls.Supported("ioctl", Ioctl) + s.Table[17] = syscalls.Supported("pread64", Pread64) + s.Table[18] = syscalls.Supported("pwrite64", Pwrite64) + s.Table[19] = syscalls.Supported("readv", Readv) + s.Table[20] = syscalls.Supported("writev", Writev) + s.Table[21] = syscalls.Supported("access", Access) + s.Table[22] = syscalls.Supported("pipe", Pipe) + s.Table[23] = syscalls.Supported("select", Select) + s.Table[32] = syscalls.Supported("dup", Dup) + s.Table[33] = syscalls.Supported("dup2", Dup2) + delete(s.Table, 40) // sendfile + s.Table[41] = syscalls.Supported("socket", Socket) + s.Table[42] = syscalls.Supported("connect", Connect) + s.Table[43] = syscalls.Supported("accept", Accept) + s.Table[44] = syscalls.Supported("sendto", SendTo) + s.Table[45] = syscalls.Supported("recvfrom", RecvFrom) + s.Table[46] = syscalls.Supported("sendmsg", SendMsg) + s.Table[47] = syscalls.Supported("recvmsg", RecvMsg) + s.Table[48] = syscalls.Supported("shutdown", Shutdown) + s.Table[49] = syscalls.Supported("bind", Bind) + s.Table[50] = syscalls.Supported("listen", Listen) + s.Table[51] = syscalls.Supported("getsockname", GetSockName) + s.Table[52] = syscalls.Supported("getpeername", GetPeerName) + s.Table[53] = syscalls.Supported("socketpair", SocketPair) + s.Table[54] = syscalls.Supported("setsockopt", SetSockOpt) + s.Table[55] = syscalls.Supported("getsockopt", GetSockOpt) + s.Table[59] = syscalls.Supported("execve", Execve) + s.Table[72] = syscalls.Supported("fcntl", Fcntl) + delete(s.Table, 73) // flock + s.Table[74] = syscalls.Supported("fsync", Fsync) + s.Table[75] = syscalls.Supported("fdatasync", Fdatasync) + s.Table[76] = syscalls.Supported("truncate", Truncate) + s.Table[77] = syscalls.Supported("ftruncate", Ftruncate) + s.Table[78] = syscalls.Supported("getdents", Getdents) + s.Table[79] = syscalls.Supported("getcwd", Getcwd) + s.Table[80] = syscalls.Supported("chdir", Chdir) + s.Table[81] = syscalls.Supported("fchdir", Fchdir) + s.Table[82] = syscalls.Supported("rename", Rename) + s.Table[83] = syscalls.Supported("mkdir", Mkdir) + s.Table[84] = syscalls.Supported("rmdir", Rmdir) + s.Table[85] = syscalls.Supported("creat", Creat) + s.Table[86] = syscalls.Supported("link", Link) + s.Table[87] = syscalls.Supported("unlink", Unlink) + s.Table[88] = syscalls.Supported("symlink", Symlink) + s.Table[89] = syscalls.Supported("readlink", Readlink) + s.Table[90] = syscalls.Supported("chmod", Chmod) + s.Table[91] = syscalls.Supported("fchmod", Fchmod) + s.Table[92] = syscalls.Supported("chown", Chown) + s.Table[93] = syscalls.Supported("fchown", Fchown) + s.Table[94] = syscalls.Supported("lchown", Lchown) + s.Table[132] = syscalls.Supported("utime", Utime) + s.Table[133] = syscalls.Supported("mknod", Mknod) + s.Table[137] = syscalls.Supported("statfs", Statfs) + s.Table[138] = syscalls.Supported("fstatfs", Fstatfs) + s.Table[161] = syscalls.Supported("chroot", Chroot) + s.Table[162] = syscalls.Supported("sync", Sync) + delete(s.Table, 165) // mount + delete(s.Table, 166) // umount2 + delete(s.Table, 187) // readahead + s.Table[188] = syscalls.Supported("setxattr", Setxattr) + s.Table[189] = syscalls.Supported("lsetxattr", Lsetxattr) + s.Table[190] = syscalls.Supported("fsetxattr", Fsetxattr) + s.Table[191] = syscalls.Supported("getxattr", Getxattr) + s.Table[192] = syscalls.Supported("lgetxattr", Lgetxattr) + s.Table[193] = syscalls.Supported("fgetxattr", Fgetxattr) + s.Table[194] = syscalls.Supported("listxattr", Listxattr) + s.Table[195] = syscalls.Supported("llistxattr", Llistxattr) + s.Table[196] = syscalls.Supported("flistxattr", Flistxattr) + s.Table[197] = syscalls.Supported("removexattr", Removexattr) + s.Table[198] = syscalls.Supported("lremovexattr", Lremovexattr) + s.Table[199] = syscalls.Supported("fremovexattr", Fremovexattr) + delete(s.Table, 206) // io_setup + delete(s.Table, 207) // io_destroy + delete(s.Table, 208) // io_getevents + delete(s.Table, 209) // io_submit + delete(s.Table, 210) // io_cancel + s.Table[213] = syscalls.Supported("epoll_create", EpollCreate) + s.Table[217] = syscalls.Supported("getdents64", Getdents64) + delete(s.Table, 221) // fdavise64 + s.Table[232] = syscalls.Supported("epoll_wait", EpollWait) + s.Table[233] = syscalls.Supported("epoll_ctl", EpollCtl) + s.Table[235] = syscalls.Supported("utimes", Utimes) + delete(s.Table, 253) // inotify_init + delete(s.Table, 254) // inotify_add_watch + delete(s.Table, 255) // inotify_rm_watch + s.Table[257] = syscalls.Supported("openat", Openat) + s.Table[258] = syscalls.Supported("mkdirat", Mkdirat) + s.Table[259] = syscalls.Supported("mknodat", Mknodat) + s.Table[260] = syscalls.Supported("fchownat", Fchownat) + s.Table[261] = syscalls.Supported("futimens", Futimens) + s.Table[262] = syscalls.Supported("newfstatat", Newfstatat) + s.Table[263] = syscalls.Supported("unlinkat", Unlinkat) + s.Table[264] = syscalls.Supported("renameat", Renameat) + s.Table[265] = syscalls.Supported("linkat", Linkat) + s.Table[266] = syscalls.Supported("symlinkat", Symlinkat) + s.Table[267] = syscalls.Supported("readlinkat", Readlinkat) + s.Table[268] = syscalls.Supported("fchmodat", Fchmodat) + s.Table[269] = syscalls.Supported("faccessat", Faccessat) + s.Table[270] = syscalls.Supported("pselect", Pselect) + s.Table[271] = syscalls.Supported("ppoll", Ppoll) + delete(s.Table, 275) // splice + delete(s.Table, 276) // tee + s.Table[277] = syscalls.Supported("sync_file_range", SyncFileRange) + s.Table[280] = syscalls.Supported("utimensat", Utimensat) + s.Table[281] = syscalls.Supported("epoll_pwait", EpollPwait) + delete(s.Table, 282) // signalfd + s.Table[283] = syscalls.Supported("timerfd_create", TimerfdCreate) + s.Table[284] = syscalls.Supported("eventfd", Eventfd) + delete(s.Table, 285) // fallocate + s.Table[286] = syscalls.Supported("timerfd_settime", TimerfdSettime) + s.Table[287] = syscalls.Supported("timerfd_gettime", TimerfdGettime) + s.Table[288] = syscalls.Supported("accept4", Accept4) + delete(s.Table, 289) // signalfd4 + s.Table[290] = syscalls.Supported("eventfd2", Eventfd2) + s.Table[291] = syscalls.Supported("epoll_create1", EpollCreate1) + s.Table[292] = syscalls.Supported("dup3", Dup3) + s.Table[293] = syscalls.Supported("pipe2", Pipe2) + delete(s.Table, 294) // inotify_init1 + s.Table[295] = syscalls.Supported("preadv", Preadv) + s.Table[296] = syscalls.Supported("pwritev", Pwritev) + s.Table[299] = syscalls.Supported("recvmmsg", RecvMMsg) + s.Table[306] = syscalls.Supported("syncfs", Syncfs) + s.Table[307] = syscalls.Supported("sendmmsg", SendMMsg) + s.Table[316] = syscalls.Supported("renameat2", Renameat2) + delete(s.Table, 319) // memfd_create + s.Table[322] = syscalls.Supported("execveat", Execveat) + s.Table[327] = syscalls.Supported("preadv2", Preadv2) + s.Table[328] = syscalls.Supported("pwritev2", Pwritev2) + s.Table[332] = syscalls.Supported("statx", Statx) + s.Init() + + // Override ARM64. + s = linux.ARM64 + s.Table[63] = syscalls.Supported("read", Read) + s.Init() +} diff --git a/runsc/boot/BUILD b/runsc/boot/BUILD index 0e71e800b..a907c103b 100644 --- a/runsc/boot/BUILD +++ b/runsc/boot/BUILD @@ -15,8 +15,6 @@ go_library( "fs.go", "limits.go", "loader.go", - "loader_amd64.go", - "loader_arm64.go", "network.go", "strace.go", "vfs.go", @@ -77,7 +75,6 @@ go_library( "//pkg/sentry/socket/unix", "//pkg/sentry/state", "//pkg/sentry/strace", - "//pkg/sentry/syscalls/linux", "//pkg/sentry/syscalls/linux/vfs2", "//pkg/sentry/time", "//pkg/sentry/unimpl:unimplemented_syscall_go_proto", @@ -126,7 +123,6 @@ go_test( "//pkg/p9", "//pkg/sentry/contexttest", "//pkg/sentry/fs", - "//pkg/sentry/kernel", "//pkg/sentry/vfs", "//pkg/sync", "//pkg/unet", diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index 79ef3a880..8c8bad11c 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -77,8 +77,6 @@ import ( _ "gvisor.dev/gvisor/pkg/sentry/socket/unix" ) -var syscallTable *kernel.SyscallTable - // Loader keeps state needed to start the kernel and run the container.. type Loader struct { // k is the kernel. @@ -204,14 +202,12 @@ func New(args Args) (*Loader, error) { return nil, fmt.Errorf("setting up memory usage: %v", err) } - // Patch the syscall table. - kernel.VFS2Enabled = args.Conf.VFS2 - if kernel.VFS2Enabled { - vfs2.Override(syscallTable.Table) + // Is this a VFSv2 kernel? + if args.Conf.VFS2 { + kernel.VFS2Enabled = true + vfs2.Override() } - kernel.RegisterSyscallTable(syscallTable) - // Create kernel and platform. p, err := createPlatform(args.Conf, args.Device) if err != nil { diff --git a/runsc/boot/loader_amd64.go b/runsc/boot/loader_amd64.go deleted file mode 100644 index 78df86611..000000000 --- a/runsc/boot/loader_amd64.go +++ /dev/null @@ -1,26 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build amd64 - -package boot - -import ( - "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" -) - -func init() { - // Set the global syscall table. - syscallTable = linux.AMD64 -} diff --git a/runsc/boot/loader_arm64.go b/runsc/boot/loader_arm64.go deleted file mode 100644 index 250785010..000000000 --- a/runsc/boot/loader_arm64.go +++ /dev/null @@ -1,26 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build arm64 - -package boot - -import ( - "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" -) - -func init() { - // Set the global syscall table. - syscallTable = linux.ARM64 -} diff --git a/runsc/boot/loader_test.go b/runsc/boot/loader_test.go index 7a30fea70..e448fd773 100644 --- a/runsc/boot/loader_test.go +++ b/runsc/boot/loader_test.go @@ -31,7 +31,6 @@ import ( "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/unet" @@ -69,11 +68,6 @@ func testSpec() *specs.Spec { } } -func resetSyscallTable() { - kernel.VFS2Enabled = false - kernel.FlushSyscallTablesTestOnly() -} - // startGofer starts a new gofer routine serving 'root' path. It returns the // sandbox side of the connection, and a function that when called will stop the // gofer. @@ -150,13 +144,11 @@ func createLoader(vfsEnabled bool, spec *specs.Spec) (*Loader, func(), error) { // TestRun runs a simple application in a sandbox and checks that it succeeds. func TestRun(t *testing.T) { - defer resetSyscallTable() doRun(t, false) } // TestRunVFS2 runs TestRun in VFSv2. func TestRunVFS2(t *testing.T) { - defer resetSyscallTable() doRun(t, true) } @@ -199,13 +191,11 @@ func doRun(t *testing.T, vfsEnabled bool) { // TestStartSignal tests that the controller Start message will cause // WaitForStartSignal to return. func TestStartSignal(t *testing.T) { - defer resetSyscallTable() doStartSignal(t, false) } // TestStartSignalVFS2 does TestStartSignal with VFS2. func TestStartSignalVFS2(t *testing.T) { - defer resetSyscallTable() doStartSignal(t, true) } @@ -477,8 +467,6 @@ func TestCreateMountNamespace(t *testing.T) { func TestCreateMountNamespaceVFS2(t *testing.T) { for _, tc := range createMountTestcases(true /* vfs2 */) { t.Run(tc.name, func(t *testing.T) { - defer resetSyscallTable() - spec := testSpec() spec.Mounts = tc.spec.Mounts spec.Root = tc.spec.Root diff --git a/runsc/cmd/syscalls.go b/runsc/cmd/syscalls.go index 7072547be..a37d66139 100644 --- a/runsc/cmd/syscalls.go +++ b/runsc/cmd/syscalls.go @@ -32,9 +32,10 @@ import ( // Syscalls implements subcommands.Command for the "syscalls" command. type Syscalls struct { - output string - os string - arch string + format string + os string + arch string + filename string } // CompatibilityInfo is a map of system and architecture to compatibility doc. @@ -95,16 +96,17 @@ func (*Syscalls) Usage() string { // SetFlags implements subcommands.Command.SetFlags. func (s *Syscalls) SetFlags(f *flag.FlagSet) { - f.StringVar(&s.output, "o", "table", "Output format (table, csv, json).") + f.StringVar(&s.format, "format", "table", "Output format (table, csv, json).") f.StringVar(&s.os, "os", osAll, "The OS (e.g. linux)") f.StringVar(&s.arch, "arch", archAll, "The CPU architecture (e.g. amd64).") + f.StringVar(&s.filename, "filename", "", "Output filename (otherwise stdout).") } // Execute implements subcommands.Command.Execute. func (s *Syscalls) Execute(_ context.Context, f *flag.FlagSet, args ...interface{}) subcommands.ExitStatus { - out, ok := outputMap[s.output] + out, ok := outputMap[s.format] if !ok { - Fatalf("Unsupported output format %q", s.output) + Fatalf("Unsupported output format %q", s.format) } // Build map of all supported architectures. @@ -124,7 +126,14 @@ func (s *Syscalls) Execute(_ context.Context, f *flag.FlagSet, args ...interface Fatalf("%v", err) } - if err := out(os.Stdout, info); err != nil { + w := os.Stdout // Default. + if s.filename != "" { + w, err = os.OpenFile(s.filename, os.O_CREATE|os.O_TRUNC|os.O_WRONLY, 0644) + if err != nil { + Fatalf("Error opening %q: %v", s.filename, err) + } + } + if err := out(w, info); err != nil { Fatalf("Error writing output: %v", err) } -- cgit v1.2.3 From 7cd54c1f1437dccdf1840e8f893de06f9d1e70e6 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Wed, 6 May 2020 16:06:49 -0700 Subject: Remove vfs.FileDescriptionOptions.InvalidWrite. Compare: https://elixir.bootlin.com/linux/v5.6/source/fs/timerfd.c#L431 PiperOrigin-RevId: 310246908 --- pkg/sentry/syscalls/linux/vfs2/timerfd.go | 5 ++++- pkg/sentry/vfs/file_description.go | 7 ------- pkg/sentry/vfs/timerfd.go | 1 - 3 files changed, 4 insertions(+), 9 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/vfs2/timerfd.go b/pkg/sentry/syscalls/linux/vfs2/timerfd.go index 7938a5249..839a07db1 100644 --- a/pkg/sentry/syscalls/linux/vfs2/timerfd.go +++ b/pkg/sentry/syscalls/linux/vfs2/timerfd.go @@ -46,7 +46,10 @@ func TimerfdCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel default: return 0, nil, syserror.EINVAL } - file, err := t.Kernel().VFS().NewTimerFD(clock, fileFlags) + // Timerfds aren't writable per se (their implementation of Write just + // returns EINVAL), but they are "opened for writing", which is necessary + // to actually reach said implementation of Write. + file, err := t.Kernel().VFS().NewTimerFD(clock, linux.O_RDWR|fileFlags) if err != nil { return 0, nil, err } diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 418d69b96..cfabd936c 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -91,10 +91,6 @@ type FileDescriptionOptions struct { // ESPIPE. DenyPWrite bool - // if InvalidWrite is true, calls to FileDescription.Write() return - // EINVAL. - InvalidWrite bool - // If UseDentryMetadata is true, calls to FileDescription methods that // interact with file and filesystem metadata (Stat, SetStat, StatFS, // Listxattr, Getxattr, Setxattr, Removexattr) are implemented by calling @@ -570,9 +566,6 @@ func (fd *FileDescription) PWrite(ctx context.Context, src usermem.IOSequence, o // Write is similar to PWrite, but does not specify an offset. func (fd *FileDescription) Write(ctx context.Context, src usermem.IOSequence, opts WriteOptions) (int64, error) { - if fd.opts.InvalidWrite { - return 0, syserror.EINVAL - } if !fd.writable { return 0, syserror.EBADF } diff --git a/pkg/sentry/vfs/timerfd.go b/pkg/sentry/vfs/timerfd.go index 42b880656..cc536ceaf 100644 --- a/pkg/sentry/vfs/timerfd.go +++ b/pkg/sentry/vfs/timerfd.go @@ -53,7 +53,6 @@ func (vfs *VirtualFilesystem) NewTimerFD(clock ktime.Clock, flags uint32) (*File UseDentryMetadata: true, DenyPRead: true, DenyPWrite: true, - InvalidWrite: true, }); err != nil { return nil, err } -- cgit v1.2.3 From e0089a20e4830d6b6659414e393905e0fd6a2e66 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Thu, 7 May 2020 09:52:39 -0700 Subject: Remove outdated TODO for VFS2 AccessAt. Fixes #1965. PiperOrigin-RevId: 310380433 --- pkg/sentry/vfs/anonfs.go | 2 -- 1 file changed, 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/vfs/anonfs.go b/pkg/sentry/vfs/anonfs.go index 981bd8caa..adebaeefb 100644 --- a/pkg/sentry/vfs/anonfs.go +++ b/pkg/sentry/vfs/anonfs.go @@ -91,8 +91,6 @@ func (fs *anonFilesystem) Sync(ctx context.Context) error { } // AccessAt implements vfs.Filesystem.Impl.AccessAt. -// -// TODO(gvisor.dev/issue/1965): Implement access permissions. func (fs *anonFilesystem) AccessAt(ctx context.Context, rp *ResolvingPath, creds *auth.Credentials, ats AccessTypes) error { if !rp.Done() { return syserror.ENOTDIR -- cgit v1.2.3 From 16da7e790f570d9743fb47e50c304e6a24834772 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Thu, 7 May 2020 10:19:11 -0700 Subject: Update privateunixsocket TODOs. Synthetic sockets do not have the race condition issue in VFS2, and we will get rid of privateunixsocket as well. Fixes #1200. PiperOrigin-RevId: 310386474 --- pkg/sentry/fs/gofer/session.go | 6 +++--- runsc/boot/vfs.go | 4 ++-- 2 files changed, 5 insertions(+), 5 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/gofer/session.go b/pkg/sentry/fs/gofer/session.go index f6b3ef178..b5efc86f2 100644 --- a/pkg/sentry/fs/gofer/session.go +++ b/pkg/sentry/fs/gofer/session.go @@ -190,9 +190,9 @@ type session struct { // be socket/pipe files. This allows unix domain sockets and named pipes to // be used with paths that belong to a gofer. // - // TODO(gvisor.dev/issue/1200): there are few possible races with someone - // stat'ing the file and another deleting it concurrently, where the file - // will not be reported as socket file. + // There are a few possible races with someone stat'ing the file and another + // deleting it concurrently, where the file will not be reported as socket + // file. overrides *overrideMaps `state:"wait"` } diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go index d1397ed2c..b94211866 100644 --- a/runsc/boot/vfs.go +++ b/runsc/boot/vfs.go @@ -301,8 +301,8 @@ func (c *containerMounter) getMountNameAndOptionsVFS2(conf *Config, m specs.Moun } // p9MountOptions creates a slice of options for a p9 mount. -// TODO(gvisor.dev/issue/1200): Remove this version in favor of the one in -// fs.go when privateunixsocket lands. +// TODO(gvisor.dev/issue/1624): Remove this version once privateunixsocket is +// deleted, along with the rest of VFS1. func p9MountOptionsVFS2(fd int, fa FileAccessType) []string { opts := []string{ "trans=fd", -- cgit v1.2.3 From 26c60d7d5d31fdf3ad380a0f09b4f33afae3d9d3 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Thu, 7 May 2020 11:40:01 -0700 Subject: Port signalfd to vfs2. PiperOrigin-RevId: 310404113 --- pkg/sentry/fsimpl/signalfd/BUILD | 20 +++++ pkg/sentry/fsimpl/signalfd/signalfd.go | 135 +++++++++++++++++++++++++++++++ pkg/sentry/syscalls/linux/sigset.go | 8 +- pkg/sentry/syscalls/linux/sys_epoll.go | 2 +- pkg/sentry/syscalls/linux/sys_poll.go | 4 +- pkg/sentry/syscalls/linux/sys_signal.go | 6 +- pkg/sentry/syscalls/linux/vfs2/BUILD | 2 + pkg/sentry/syscalls/linux/vfs2/signal.go | 100 +++++++++++++++++++++++ pkg/sentry/syscalls/linux/vfs2/vfs2.go | 4 +- pkg/sentry/vfs/BUILD | 1 + 10 files changed, 272 insertions(+), 10 deletions(-) create mode 100644 pkg/sentry/fsimpl/signalfd/BUILD create mode 100644 pkg/sentry/fsimpl/signalfd/signalfd.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/signal.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/signalfd/BUILD b/pkg/sentry/fsimpl/signalfd/BUILD new file mode 100644 index 000000000..067c1657f --- /dev/null +++ b/pkg/sentry/fsimpl/signalfd/BUILD @@ -0,0 +1,20 @@ +load("//tools:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_library( + name = "signalfd", + srcs = ["signalfd.go"], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/abi/linux", + "//pkg/binary", + "//pkg/context", + "//pkg/sentry/kernel", + "//pkg/sentry/vfs", + "//pkg/sync", + "//pkg/syserror", + "//pkg/usermem", + "//pkg/waiter", + ], +) diff --git a/pkg/sentry/fsimpl/signalfd/signalfd.go b/pkg/sentry/fsimpl/signalfd/signalfd.go new file mode 100644 index 000000000..d29ef3f83 --- /dev/null +++ b/pkg/sentry/fsimpl/signalfd/signalfd.go @@ -0,0 +1,135 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package signalfd + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +// SignalFileDescription implements FileDescriptionImpl for signal fds. +type SignalFileDescription struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl + vfs.DentryMetadataFileDescriptionImpl + + // target is the original signal target task. + // + // The semantics here are a bit broken. Linux will always use current + // for all reads, regardless of where the signalfd originated. We can't + // do exactly that because we need to plumb the context through + // EventRegister in order to support proper blocking behavior. This + // will undoubtedly become very complicated quickly. + target *kernel.Task + + // mu protects mask. + mu sync.Mutex + + // mask is the signal mask. Protected by mu. + mask linux.SignalSet +} + +var _ vfs.FileDescriptionImpl = (*SignalFileDescription)(nil) + +// New creates a new signal fd. +func New(vfsObj *vfs.VirtualFilesystem, target *kernel.Task, mask linux.SignalSet, flags uint32) (*vfs.FileDescription, error) { + vd := vfsObj.NewAnonVirtualDentry("[signalfd]") + defer vd.DecRef() + sfd := &SignalFileDescription{ + target: target, + mask: mask, + } + if err := sfd.vfsfd.Init(sfd, flags, vd.Mount(), vd.Dentry(), &vfs.FileDescriptionOptions{ + UseDentryMetadata: true, + DenyPRead: true, + DenyPWrite: true, + }); err != nil { + return nil, err + } + return &sfd.vfsfd, nil +} + +// Mask returns the signal mask. +func (sfd *SignalFileDescription) Mask() linux.SignalSet { + sfd.mu.Lock() + defer sfd.mu.Unlock() + return sfd.mask +} + +// SetMask sets the signal mask. +func (sfd *SignalFileDescription) SetMask(mask linux.SignalSet) { + sfd.mu.Lock() + defer sfd.mu.Unlock() + sfd.mask = mask +} + +// Read implements FileDescriptionImpl.Read. +func (sfd *SignalFileDescription) Read(ctx context.Context, dst usermem.IOSequence, _ vfs.ReadOptions) (int64, error) { + // Attempt to dequeue relevant signals. + info, err := sfd.target.Sigtimedwait(sfd.Mask(), 0) + if err != nil { + // There must be no signal available. + return 0, syserror.ErrWouldBlock + } + + // Copy out the signal info using the specified format. + var buf [128]byte + binary.Marshal(buf[:0], usermem.ByteOrder, &linux.SignalfdSiginfo{ + Signo: uint32(info.Signo), + Errno: info.Errno, + Code: info.Code, + PID: uint32(info.Pid()), + UID: uint32(info.Uid()), + Status: info.Status(), + Overrun: uint32(info.Overrun()), + Addr: info.Addr(), + }) + n, err := dst.CopyOut(ctx, buf[:]) + return int64(n), err +} + +// Readiness implements waiter.Waitable.Readiness. +func (sfd *SignalFileDescription) Readiness(mask waiter.EventMask) waiter.EventMask { + sfd.mu.Lock() + defer sfd.mu.Unlock() + if mask&waiter.EventIn != 0 && sfd.target.PendingSignals()&sfd.mask != 0 { + return waiter.EventIn // Pending signals. + } + return 0 +} + +// EventRegister implements waiter.Waitable.EventRegister. +func (sfd *SignalFileDescription) EventRegister(entry *waiter.Entry, _ waiter.EventMask) { + sfd.mu.Lock() + defer sfd.mu.Unlock() + // Register for the signal set; ignore the passed events. + sfd.target.SignalRegister(entry, waiter.EventMask(sfd.mask)) +} + +// EventUnregister implements waiter.Waitable.EventUnregister. +func (sfd *SignalFileDescription) EventUnregister(entry *waiter.Entry) { + // Unregister the original entry. + sfd.target.SignalUnregister(entry) +} + +// Release implements FileDescriptionImpl.Release() +func (sfd *SignalFileDescription) Release() {} diff --git a/pkg/sentry/syscalls/linux/sigset.go b/pkg/sentry/syscalls/linux/sigset.go index 2ddb2b146..434559b80 100644 --- a/pkg/sentry/syscalls/linux/sigset.go +++ b/pkg/sentry/syscalls/linux/sigset.go @@ -21,9 +21,13 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) -// copyInSigSet copies in a sigset_t, checks its size, and ensures that KILL and +// CopyInSigSet copies in a sigset_t, checks its size, and ensures that KILL and // STOP are clear. -func copyInSigSet(t *kernel.Task, sigSetAddr usermem.Addr, size uint) (linux.SignalSet, error) { +// +// TODO(gvisor.dev/issue/1624): This is only exported because +// syscalls/vfs2/signal.go depends on it. Once vfs1 is deleted and the vfs2 +// syscalls are moved into this package, then they can be unexported. +func CopyInSigSet(t *kernel.Task, sigSetAddr usermem.Addr, size uint) (linux.SignalSet, error) { if size != linux.SignalSetSize { return 0, syserror.EINVAL } diff --git a/pkg/sentry/syscalls/linux/sys_epoll.go b/pkg/sentry/syscalls/linux/sys_epoll.go index 51bf205cf..7f460d30b 100644 --- a/pkg/sentry/syscalls/linux/sys_epoll.go +++ b/pkg/sentry/syscalls/linux/sys_epoll.go @@ -131,7 +131,7 @@ func EpollPwait(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy maskSize := uint(args[5].Uint()) if maskAddr != 0 { - mask, err := copyInSigSet(t, maskAddr, maskSize) + mask, err := CopyInSigSet(t, maskAddr, maskSize) if err != nil { return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/sys_poll.go b/pkg/sentry/syscalls/linux/sys_poll.go index 4f8762d7d..f0198141c 100644 --- a/pkg/sentry/syscalls/linux/sys_poll.go +++ b/pkg/sentry/syscalls/linux/sys_poll.go @@ -443,7 +443,7 @@ func Ppoll(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } if maskAddr != 0 { - mask, err := copyInSigSet(t, maskAddr, maskSize) + mask, err := CopyInSigSet(t, maskAddr, maskSize) if err != nil { return 0, nil, err } @@ -525,7 +525,7 @@ func Pselect(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca } if maskAddr != 0 { - mask, err := copyInSigSet(t, maskAddr, size) + mask, err := CopyInSigSet(t, maskAddr, size) if err != nil { return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/sys_signal.go b/pkg/sentry/syscalls/linux/sys_signal.go index 582d37e03..d2b0012ae 100644 --- a/pkg/sentry/syscalls/linux/sys_signal.go +++ b/pkg/sentry/syscalls/linux/sys_signal.go @@ -295,7 +295,7 @@ func RtSigprocmask(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel } oldmask := t.SignalMask() if setaddr != 0 { - mask, err := copyInSigSet(t, setaddr, sigsetsize) + mask, err := CopyInSigSet(t, setaddr, sigsetsize) if err != nil { return 0, nil, err } @@ -366,7 +366,7 @@ func RtSigtimedwait(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne timespec := args[2].Pointer() sigsetsize := args[3].SizeT() - mask, err := copyInSigSet(t, sigset, sigsetsize) + mask, err := CopyInSigSet(t, sigset, sigsetsize) if err != nil { return 0, nil, err } @@ -518,7 +518,7 @@ func RestartSyscall(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne // sharedSignalfd is shared between the two calls. func sharedSignalfd(t *kernel.Task, fd int32, sigset usermem.Addr, sigsetsize uint, flags int32) (uintptr, *kernel.SyscallControl, error) { // Copy in the signal mask. - mask, err := copyInSigSet(t, sigset, sigsetsize) + mask, err := CopyInSigSet(t, sigset, sigsetsize) if err != nil { return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD index 4c7b8f819..14838aa2c 100644 --- a/pkg/sentry/syscalls/linux/vfs2/BUILD +++ b/pkg/sentry/syscalls/linux/vfs2/BUILD @@ -19,6 +19,7 @@ go_library( "poll.go", "read_write.go", "setstat.go", + "signal.go", "socket.go", "stat.go", "stat_amd64.go", @@ -39,6 +40,7 @@ go_library( "//pkg/sentry/arch", "//pkg/sentry/fsbridge", "//pkg/sentry/fsimpl/pipefs", + "//pkg/sentry/fsimpl/signalfd", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/pipe", diff --git a/pkg/sentry/syscalls/linux/vfs2/signal.go b/pkg/sentry/syscalls/linux/vfs2/signal.go new file mode 100644 index 000000000..623992f6f --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/signal.go @@ -0,0 +1,100 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/signalfd" + "gvisor.dev/gvisor/pkg/sentry/kernel" + slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +// sharedSignalfd is shared between the two calls. +func sharedSignalfd(t *kernel.Task, fd int32, sigset usermem.Addr, sigsetsize uint, flags int32) (uintptr, *kernel.SyscallControl, error) { + // Copy in the signal mask. + mask, err := slinux.CopyInSigSet(t, sigset, sigsetsize) + if err != nil { + return 0, nil, err + } + + // Always check for valid flags, even if not creating. + if flags&^(linux.SFD_NONBLOCK|linux.SFD_CLOEXEC) != 0 { + return 0, nil, syserror.EINVAL + } + + // Is this a change to an existing signalfd? + // + // The spec indicates that this should adjust the mask. + if fd != -1 { + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Is this a signalfd? + if sfd, ok := file.Impl().(*signalfd.SignalFileDescription); ok { + sfd.SetMask(mask) + return 0, nil, nil + } + + // Not a signalfd. + return 0, nil, syserror.EINVAL + } + + fileFlags := uint32(linux.O_RDWR) + if flags&linux.SFD_NONBLOCK != 0 { + fileFlags |= linux.O_NONBLOCK + } + + // Create a new file. + vfsObj := t.Kernel().VFS() + file, err := signalfd.New(vfsObj, t, mask, fileFlags) + if err != nil { + return 0, nil, err + } + defer file.DecRef() + + // Create a new descriptor. + fd, err = t.NewFDFromVFS2(0, file, kernel.FDFlags{ + CloseOnExec: flags&linux.SFD_CLOEXEC != 0, + }) + if err != nil { + return 0, nil, err + } + + // Done. + return uintptr(fd), nil, nil +} + +// Signalfd implements the linux syscall signalfd(2). +func Signalfd(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + sigset := args[1].Pointer() + sigsetsize := args[2].SizeT() + return sharedSignalfd(t, fd, sigset, sigsetsize, 0) +} + +// Signalfd4 implements the linux syscall signalfd4(2). +func Signalfd4(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + sigset := args[1].Pointer() + sigsetsize := args[2].SizeT() + flags := args[3].Int() + return sharedSignalfd(t, fd, sigset, sigsetsize, flags) +} diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2.go b/pkg/sentry/syscalls/linux/vfs2/vfs2.go index f1b697844..9c04677f1 100644 --- a/pkg/sentry/syscalls/linux/vfs2/vfs2.go +++ b/pkg/sentry/syscalls/linux/vfs2/vfs2.go @@ -139,14 +139,14 @@ func Override() { s.Table[277] = syscalls.Supported("sync_file_range", SyncFileRange) s.Table[280] = syscalls.Supported("utimensat", Utimensat) s.Table[281] = syscalls.Supported("epoll_pwait", EpollPwait) - delete(s.Table, 282) // signalfd + s.Table[282] = syscalls.Supported("signalfd", Signalfd) s.Table[283] = syscalls.Supported("timerfd_create", TimerfdCreate) s.Table[284] = syscalls.Supported("eventfd", Eventfd) delete(s.Table, 285) // fallocate s.Table[286] = syscalls.Supported("timerfd_settime", TimerfdSettime) s.Table[287] = syscalls.Supported("timerfd_gettime", TimerfdGettime) s.Table[288] = syscalls.Supported("accept4", Accept4) - delete(s.Table, 289) // signalfd4 + s.Table[289] = syscalls.Supported("signalfd4", Signalfd4) s.Table[290] = syscalls.Supported("eventfd2", Eventfd2) s.Table[291] = syscalls.Supported("epoll_create1", EpollCreate1) s.Table[292] = syscalls.Supported("dup3", Dup3) diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index c62505fe2..86046dd99 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -43,6 +43,7 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", + "//pkg/binary", "//pkg/context", "//pkg/fd", "//pkg/fdnotifier", -- cgit v1.2.3 From d0b1d0233dc8a8ac837d534cd0664eabb9dd0a71 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Thu, 7 May 2020 12:42:46 -0700 Subject: Move pkg/sentry/vfs/{eventfd,timerfd} to new packages in pkg/sentry/fsimpl. They don't depend on anything in VFS2, so they should be their own packages. PiperOrigin-RevId: 310416807 --- pkg/sentry/fsimpl/eventfd/BUILD | 33 ++++ pkg/sentry/fsimpl/eventfd/eventfd.go | 284 ++++++++++++++++++++++++++++++ pkg/sentry/fsimpl/eventfd/eventfd_test.go | 97 ++++++++++ pkg/sentry/fsimpl/timerfd/BUILD | 17 ++ pkg/sentry/fsimpl/timerfd/timerfd.go | 143 +++++++++++++++ pkg/sentry/kernel/BUILD | 1 + pkg/sentry/kernel/kernel.go | 11 +- pkg/sentry/syscalls/linux/vfs2/BUILD | 2 + pkg/sentry/syscalls/linux/vfs2/eventfd.go | 4 +- pkg/sentry/syscalls/linux/vfs2/timerfd.go | 19 +- pkg/sentry/vfs/BUILD | 4 - pkg/sentry/vfs/eventfd.go | 282 ----------------------------- pkg/sentry/vfs/eventfd_test.go | 96 ---------- pkg/sentry/vfs/timerfd.go | 141 --------------- 14 files changed, 596 insertions(+), 538 deletions(-) create mode 100644 pkg/sentry/fsimpl/eventfd/BUILD create mode 100644 pkg/sentry/fsimpl/eventfd/eventfd.go create mode 100644 pkg/sentry/fsimpl/eventfd/eventfd_test.go create mode 100644 pkg/sentry/fsimpl/timerfd/BUILD create mode 100644 pkg/sentry/fsimpl/timerfd/timerfd.go delete mode 100644 pkg/sentry/vfs/eventfd.go delete mode 100644 pkg/sentry/vfs/eventfd_test.go delete mode 100644 pkg/sentry/vfs/timerfd.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/eventfd/BUILD b/pkg/sentry/fsimpl/eventfd/BUILD new file mode 100644 index 000000000..ea167d38c --- /dev/null +++ b/pkg/sentry/fsimpl/eventfd/BUILD @@ -0,0 +1,33 @@ +load("//tools:defs.bzl", "go_library", "go_test") + +licenses(["notice"]) + +go_library( + name = "eventfd", + srcs = ["eventfd.go"], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/abi/linux", + "//pkg/context", + "//pkg/fdnotifier", + "//pkg/log", + "//pkg/sentry/vfs", + "//pkg/syserror", + "//pkg/usermem", + "//pkg/waiter", + ], +) + +go_test( + name = "eventfd_test", + size = "small", + srcs = ["eventfd_test.go"], + library = ":eventfd", + deps = [ + "//pkg/abi/linux", + "//pkg/sentry/contexttest", + "//pkg/sentry/vfs", + "//pkg/usermem", + "//pkg/waiter", + ], +) diff --git a/pkg/sentry/fsimpl/eventfd/eventfd.go b/pkg/sentry/fsimpl/eventfd/eventfd.go new file mode 100644 index 000000000..c573d7935 --- /dev/null +++ b/pkg/sentry/fsimpl/eventfd/eventfd.go @@ -0,0 +1,284 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package eventfd implements event fds. +package eventfd + +import ( + "math" + "sync" + "syscall" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fdnotifier" + "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +// EventFileDescription implements FileDescriptionImpl for file-based event +// notification (eventfd). Eventfds are usually internal to the Sentry but in +// certain situations they may be converted into a host-backed eventfd. +type EventFileDescription struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl + vfs.DentryMetadataFileDescriptionImpl + + // queue is used to notify interested parties when the event object + // becomes readable or writable. + queue waiter.Queue `state:"zerovalue"` + + // mu protects the fields below. + mu sync.Mutex `state:"nosave"` + + // val is the current value of the event counter. + val uint64 + + // semMode specifies whether the event is in "semaphore" mode. + semMode bool + + // hostfd indicates whether this eventfd is passed through to the host. + hostfd int +} + +var _ vfs.FileDescriptionImpl = (*EventFileDescription)(nil) + +// New creates a new event fd. +func New(vfsObj *vfs.VirtualFilesystem, initVal uint64, semMode bool, flags uint32) (*vfs.FileDescription, error) { + vd := vfsObj.NewAnonVirtualDentry("[eventfd]") + defer vd.DecRef() + efd := &EventFileDescription{ + val: initVal, + semMode: semMode, + hostfd: -1, + } + if err := efd.vfsfd.Init(efd, flags, vd.Mount(), vd.Dentry(), &vfs.FileDescriptionOptions{ + UseDentryMetadata: true, + DenyPRead: true, + DenyPWrite: true, + }); err != nil { + return nil, err + } + return &efd.vfsfd, nil +} + +// HostFD returns the host eventfd associated with this event. +func (efd *EventFileDescription) HostFD() (int, error) { + efd.mu.Lock() + defer efd.mu.Unlock() + if efd.hostfd >= 0 { + return efd.hostfd, nil + } + + flags := linux.EFD_NONBLOCK + if efd.semMode { + flags |= linux.EFD_SEMAPHORE + } + + fd, _, errno := syscall.Syscall(syscall.SYS_EVENTFD2, uintptr(efd.val), uintptr(flags), 0) + if errno != 0 { + return -1, errno + } + + if err := fdnotifier.AddFD(int32(fd), &efd.queue); err != nil { + if closeErr := syscall.Close(int(fd)); closeErr != nil { + log.Warningf("close(%d) eventfd failed: %v", fd, closeErr) + } + return -1, err + } + + efd.hostfd = int(fd) + return efd.hostfd, nil +} + +// Release implements FileDescriptionImpl.Release() +func (efd *EventFileDescription) Release() { + efd.mu.Lock() + defer efd.mu.Unlock() + if efd.hostfd >= 0 { + fdnotifier.RemoveFD(int32(efd.hostfd)) + if closeErr := syscall.Close(int(efd.hostfd)); closeErr != nil { + log.Warningf("close(%d) eventfd failed: %v", efd.hostfd, closeErr) + } + efd.hostfd = -1 + } +} + +// Read implements FileDescriptionImpl.Read. +func (efd *EventFileDescription) Read(ctx context.Context, dst usermem.IOSequence, _ vfs.ReadOptions) (int64, error) { + if dst.NumBytes() < 8 { + return 0, syscall.EINVAL + } + if err := efd.read(ctx, dst); err != nil { + return 0, err + } + return 8, nil +} + +// Write implements FileDescriptionImpl.Write. +func (efd *EventFileDescription) Write(ctx context.Context, src usermem.IOSequence, _ vfs.WriteOptions) (int64, error) { + if src.NumBytes() < 8 { + return 0, syscall.EINVAL + } + if err := efd.write(ctx, src); err != nil { + return 0, err + } + return 8, nil +} + +// Preconditions: Must be called with efd.mu locked. +func (efd *EventFileDescription) hostReadLocked(ctx context.Context, dst usermem.IOSequence) error { + var buf [8]byte + if _, err := syscall.Read(efd.hostfd, buf[:]); err != nil { + if err == syscall.EWOULDBLOCK { + return syserror.ErrWouldBlock + } + return err + } + _, err := dst.CopyOut(ctx, buf[:]) + return err +} + +func (efd *EventFileDescription) read(ctx context.Context, dst usermem.IOSequence) error { + efd.mu.Lock() + if efd.hostfd >= 0 { + defer efd.mu.Unlock() + return efd.hostReadLocked(ctx, dst) + } + + // We can't complete the read if the value is currently zero. + if efd.val == 0 { + efd.mu.Unlock() + return syserror.ErrWouldBlock + } + + // Update the value based on the mode the event is operating in. + var val uint64 + if efd.semMode { + val = 1 + // Consistent with Linux, this is done even if writing to memory fails. + efd.val-- + } else { + val = efd.val + efd.val = 0 + } + + efd.mu.Unlock() + + // Notify writers. We do this even if we were already writable because + // it is possible that a writer is waiting to write the maximum value + // to the event. + efd.queue.Notify(waiter.EventOut) + + var buf [8]byte + usermem.ByteOrder.PutUint64(buf[:], val) + _, err := dst.CopyOut(ctx, buf[:]) + return err +} + +// Preconditions: Must be called with efd.mu locked. +func (efd *EventFileDescription) hostWriteLocked(val uint64) error { + var buf [8]byte + usermem.ByteOrder.PutUint64(buf[:], val) + _, err := syscall.Write(efd.hostfd, buf[:]) + if err == syscall.EWOULDBLOCK { + return syserror.ErrWouldBlock + } + return err +} + +func (efd *EventFileDescription) write(ctx context.Context, src usermem.IOSequence) error { + var buf [8]byte + if _, err := src.CopyIn(ctx, buf[:]); err != nil { + return err + } + val := usermem.ByteOrder.Uint64(buf[:]) + + return efd.Signal(val) +} + +// Signal is an internal function to signal the event fd. +func (efd *EventFileDescription) Signal(val uint64) error { + if val == math.MaxUint64 { + return syscall.EINVAL + } + + efd.mu.Lock() + + if efd.hostfd >= 0 { + defer efd.mu.Unlock() + return efd.hostWriteLocked(val) + } + + // We only allow writes that won't cause the value to go over the max + // uint64 minus 1. + if val > math.MaxUint64-1-efd.val { + efd.mu.Unlock() + return syserror.ErrWouldBlock + } + + efd.val += val + efd.mu.Unlock() + + // Always trigger a notification. + efd.queue.Notify(waiter.EventIn) + + return nil +} + +// Readiness implements waiter.Waitable.Readiness. +func (efd *EventFileDescription) Readiness(mask waiter.EventMask) waiter.EventMask { + efd.mu.Lock() + defer efd.mu.Unlock() + + if efd.hostfd >= 0 { + return fdnotifier.NonBlockingPoll(int32(efd.hostfd), mask) + } + + ready := waiter.EventMask(0) + if efd.val > 0 { + ready |= waiter.EventIn + } + + if efd.val < math.MaxUint64-1 { + ready |= waiter.EventOut + } + + return mask & ready +} + +// EventRegister implements waiter.Waitable.EventRegister. +func (efd *EventFileDescription) EventRegister(entry *waiter.Entry, mask waiter.EventMask) { + efd.queue.EventRegister(entry, mask) + + efd.mu.Lock() + defer efd.mu.Unlock() + if efd.hostfd >= 0 { + fdnotifier.UpdateFD(int32(efd.hostfd)) + } +} + +// EventUnregister implements waiter.Waitable.EventUnregister. +func (efd *EventFileDescription) EventUnregister(entry *waiter.Entry) { + efd.queue.EventUnregister(entry) + + efd.mu.Lock() + defer efd.mu.Unlock() + if efd.hostfd >= 0 { + fdnotifier.UpdateFD(int32(efd.hostfd)) + } +} diff --git a/pkg/sentry/fsimpl/eventfd/eventfd_test.go b/pkg/sentry/fsimpl/eventfd/eventfd_test.go new file mode 100644 index 000000000..20e3adffc --- /dev/null +++ b/pkg/sentry/fsimpl/eventfd/eventfd_test.go @@ -0,0 +1,97 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package eventfd + +import ( + "testing" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/contexttest" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +func TestEventFD(t *testing.T) { + initVals := []uint64{ + 0, + // Using a non-zero initial value verifies that writing to an + // eventfd signals when the eventfd's counter was already + // non-zero. + 343, + } + + for _, initVal := range initVals { + ctx := contexttest.Context(t) + vfsObj := &vfs.VirtualFilesystem{} + if err := vfsObj.Init(); err != nil { + t.Fatalf("VFS init: %v", err) + } + + // Make a new eventfd that is writable. + eventfd, err := New(vfsObj, initVal, false, linux.O_RDWR) + if err != nil { + t.Fatalf("New() failed: %v", err) + } + defer eventfd.DecRef() + + // Register a callback for a write event. + w, ch := waiter.NewChannelEntry(nil) + eventfd.EventRegister(&w, waiter.EventIn) + defer eventfd.EventUnregister(&w) + + data := []byte("00000124") + // Create and submit a write request. + n, err := eventfd.Write(ctx, usermem.BytesIOSequence(data), vfs.WriteOptions{}) + if err != nil { + t.Fatal(err) + } + if n != 8 { + t.Errorf("eventfd.write wrote %d bytes, not full int64", n) + } + + // Check if the callback fired due to the write event. + select { + case <-ch: + default: + t.Errorf("Didn't get notified of EventIn after write") + } + } +} + +func TestEventFDStat(t *testing.T) { + ctx := contexttest.Context(t) + vfsObj := &vfs.VirtualFilesystem{} + if err := vfsObj.Init(); err != nil { + t.Fatalf("VFS init: %v", err) + } + + // Make a new eventfd that is writable. + eventfd, err := New(vfsObj, 0, false, linux.O_RDWR) + if err != nil { + t.Fatalf("New() failed: %v", err) + } + defer eventfd.DecRef() + + statx, err := eventfd.Stat(ctx, vfs.StatOptions{ + Mask: linux.STATX_BASIC_STATS, + }) + if err != nil { + t.Fatalf("eventfd.Stat failed: %v", err) + } + if statx.Size != 0 { + t.Errorf("eventfd size should be 0") + } +} diff --git a/pkg/sentry/fsimpl/timerfd/BUILD b/pkg/sentry/fsimpl/timerfd/BUILD new file mode 100644 index 000000000..fbb02a271 --- /dev/null +++ b/pkg/sentry/fsimpl/timerfd/BUILD @@ -0,0 +1,17 @@ +load("//tools:defs.bzl", "go_library") + +licenses(["notice"]) + +go_library( + name = "timerfd", + srcs = ["timerfd.go"], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/context", + "//pkg/sentry/kernel/time", + "//pkg/sentry/vfs", + "//pkg/syserror", + "//pkg/usermem", + "//pkg/waiter", + ], +) diff --git a/pkg/sentry/fsimpl/timerfd/timerfd.go b/pkg/sentry/fsimpl/timerfd/timerfd.go new file mode 100644 index 000000000..60c92d626 --- /dev/null +++ b/pkg/sentry/fsimpl/timerfd/timerfd.go @@ -0,0 +1,143 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package timerfd implements timer fds. +package timerfd + +import ( + "sync/atomic" + + "gvisor.dev/gvisor/pkg/context" + ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +// TimerFileDescription implements FileDescriptionImpl for timer fds. It also +// implements ktime.TimerListener. +type TimerFileDescription struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl + vfs.DentryMetadataFileDescriptionImpl + + events waiter.Queue + timer *ktime.Timer + + // val is the number of timer expirations since the last successful + // call to PRead, or SetTime. val must be accessed using atomic memory + // operations. + val uint64 +} + +var _ vfs.FileDescriptionImpl = (*TimerFileDescription)(nil) +var _ ktime.TimerListener = (*TimerFileDescription)(nil) + +// New returns a new timer fd. +func New(vfsObj *vfs.VirtualFilesystem, clock ktime.Clock, flags uint32) (*vfs.FileDescription, error) { + vd := vfsObj.NewAnonVirtualDentry("[timerfd]") + defer vd.DecRef() + tfd := &TimerFileDescription{} + tfd.timer = ktime.NewTimer(clock, tfd) + if err := tfd.vfsfd.Init(tfd, flags, vd.Mount(), vd.Dentry(), &vfs.FileDescriptionOptions{ + UseDentryMetadata: true, + DenyPRead: true, + DenyPWrite: true, + }); err != nil { + return nil, err + } + return &tfd.vfsfd, nil +} + +// Read implements FileDescriptionImpl.Read. +func (tfd *TimerFileDescription) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + const sizeofUint64 = 8 + if dst.NumBytes() < sizeofUint64 { + return 0, syserror.EINVAL + } + if val := atomic.SwapUint64(&tfd.val, 0); val != 0 { + var buf [sizeofUint64]byte + usermem.ByteOrder.PutUint64(buf[:], val) + if _, err := dst.CopyOut(ctx, buf[:]); err != nil { + // Linux does not undo consuming the number of + // expirations even if writing to userspace fails. + return 0, err + } + return sizeofUint64, nil + } + return 0, syserror.ErrWouldBlock +} + +// Clock returns the timer fd's Clock. +func (tfd *TimerFileDescription) Clock() ktime.Clock { + return tfd.timer.Clock() +} + +// GetTime returns the associated Timer's setting and the time at which it was +// observed. +func (tfd *TimerFileDescription) GetTime() (ktime.Time, ktime.Setting) { + return tfd.timer.Get() +} + +// SetTime atomically changes the associated Timer's setting, resets the number +// of expirations to 0, and returns the previous setting and the time at which +// it was observed. +func (tfd *TimerFileDescription) SetTime(s ktime.Setting) (ktime.Time, ktime.Setting) { + return tfd.timer.SwapAnd(s, func() { atomic.StoreUint64(&tfd.val, 0) }) +} + +// Readiness implements waiter.Waitable.Readiness. +func (tfd *TimerFileDescription) Readiness(mask waiter.EventMask) waiter.EventMask { + var ready waiter.EventMask + if atomic.LoadUint64(&tfd.val) != 0 { + ready |= waiter.EventIn + } + return ready +} + +// EventRegister implements waiter.Waitable.EventRegister. +func (tfd *TimerFileDescription) EventRegister(e *waiter.Entry, mask waiter.EventMask) { + tfd.events.EventRegister(e, mask) +} + +// EventUnregister implements waiter.Waitable.EventUnregister. +func (tfd *TimerFileDescription) EventUnregister(e *waiter.Entry) { + tfd.events.EventUnregister(e) +} + +// PauseTimer pauses the associated Timer. +func (tfd *TimerFileDescription) PauseTimer() { + tfd.timer.Pause() +} + +// ResumeTimer resumes the associated Timer. +func (tfd *TimerFileDescription) ResumeTimer() { + tfd.timer.Resume() +} + +// Release implements FileDescriptionImpl.Release() +func (tfd *TimerFileDescription) Release() { + tfd.timer.Destroy() +} + +// Notify implements ktime.TimerListener.Notify. +func (tfd *TimerFileDescription) Notify(exp uint64, setting ktime.Setting) (ktime.Setting, bool) { + atomic.AddUint64(&tfd.val, exp) + tfd.events.Notify(waiter.EventIn) + return ktime.Setting{}, false +} + +// Destroy implements ktime.TimerListener.Destroy. +func (tfd *TimerFileDescription) Destroy() {} diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD index e47af66d6..8104f50f3 100644 --- a/pkg/sentry/kernel/BUILD +++ b/pkg/sentry/kernel/BUILD @@ -172,6 +172,7 @@ go_library( "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/fsimpl/pipefs", "//pkg/sentry/fsimpl/sockfs", + "//pkg/sentry/fsimpl/timerfd", "//pkg/sentry/hostcpu", "//pkg/sentry/inet", "//pkg/sentry/kernel/auth", diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index c91b9dce2..271ea5faf 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -48,10 +48,11 @@ import ( "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fs/timerfd" + oldtimerfd "gvisor.dev/gvisor/pkg/sentry/fs/timerfd" "gvisor.dev/gvisor/pkg/sentry/fsbridge" "gvisor.dev/gvisor/pkg/sentry/fsimpl/pipefs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/timerfd" "gvisor.dev/gvisor/pkg/sentry/hostcpu" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" @@ -1068,11 +1069,11 @@ func (k *Kernel) pauseTimeLocked() { if t.fdTable != nil { t.fdTable.forEach(func(_ int32, file *fs.File, fd *vfs.FileDescription, _ FDFlags) { if VFS2Enabled { - if tfd, ok := fd.Impl().(*vfs.TimerFileDescription); ok { + if tfd, ok := fd.Impl().(*timerfd.TimerFileDescription); ok { tfd.PauseTimer() } } else { - if tfd, ok := file.FileOperations.(*timerfd.TimerOperations); ok { + if tfd, ok := file.FileOperations.(*oldtimerfd.TimerOperations); ok { tfd.PauseTimer() } } @@ -1104,11 +1105,11 @@ func (k *Kernel) resumeTimeLocked() { if t.fdTable != nil { t.fdTable.forEach(func(_ int32, file *fs.File, fd *vfs.FileDescription, _ FDFlags) { if VFS2Enabled { - if tfd, ok := fd.Impl().(*vfs.TimerFileDescription); ok { + if tfd, ok := fd.Impl().(*timerfd.TimerFileDescription); ok { tfd.ResumeTimer() } } else { - if tfd, ok := file.FileOperations.(*timerfd.TimerOperations); ok { + if tfd, ok := file.FileOperations.(*oldtimerfd.TimerOperations); ok { tfd.ResumeTimer() } } diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD index 14838aa2c..c32f942fb 100644 --- a/pkg/sentry/syscalls/linux/vfs2/BUILD +++ b/pkg/sentry/syscalls/linux/vfs2/BUILD @@ -39,8 +39,10 @@ go_library( "//pkg/gohacks", "//pkg/sentry/arch", "//pkg/sentry/fsbridge", + "//pkg/sentry/fsimpl/eventfd", "//pkg/sentry/fsimpl/pipefs", "//pkg/sentry/fsimpl/signalfd", + "//pkg/sentry/fsimpl/timerfd", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/pipe", diff --git a/pkg/sentry/syscalls/linux/vfs2/eventfd.go b/pkg/sentry/syscalls/linux/vfs2/eventfd.go index bd2194972..aff1a2070 100644 --- a/pkg/sentry/syscalls/linux/vfs2/eventfd.go +++ b/pkg/sentry/syscalls/linux/vfs2/eventfd.go @@ -17,6 +17,7 @@ package vfs2 import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/eventfd" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/syserror" ) @@ -31,12 +32,13 @@ func Eventfd2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc return 0, nil, syserror.EINVAL } + vfsObj := t.Kernel().VFS() fileFlags := uint32(linux.O_RDWR) if flags&linux.EFD_NONBLOCK != 0 { fileFlags |= linux.O_NONBLOCK } semMode := flags&linux.EFD_SEMAPHORE != 0 - eventfd, err := t.Kernel().VFS().NewEventFD(initVal, semMode, fileFlags) + eventfd, err := eventfd.New(vfsObj, initVal, semMode, fileFlags) if err != nil { return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/vfs2/timerfd.go b/pkg/sentry/syscalls/linux/vfs2/timerfd.go index 839a07db1..5ac79bc09 100644 --- a/pkg/sentry/syscalls/linux/vfs2/timerfd.go +++ b/pkg/sentry/syscalls/linux/vfs2/timerfd.go @@ -17,9 +17,9 @@ package vfs2 import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/timerfd" "gvisor.dev/gvisor/pkg/sentry/kernel" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" ) @@ -32,9 +32,12 @@ func TimerfdCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel return 0, nil, syserror.EINVAL } - var fileFlags uint32 + // Timerfds aren't writable per se (their implementation of Write just + // returns EINVAL), but they are "opened for writing", which is necessary + // to actually reach said implementation of Write. + fileFlags := uint32(linux.O_RDWR) if flags&linux.TFD_NONBLOCK != 0 { - fileFlags = linux.O_NONBLOCK + fileFlags |= linux.O_NONBLOCK } var clock ktime.Clock @@ -46,10 +49,8 @@ func TimerfdCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel default: return 0, nil, syserror.EINVAL } - // Timerfds aren't writable per se (their implementation of Write just - // returns EINVAL), but they are "opened for writing", which is necessary - // to actually reach said implementation of Write. - file, err := t.Kernel().VFS().NewTimerFD(clock, linux.O_RDWR|fileFlags) + vfsObj := t.Kernel().VFS() + file, err := timerfd.New(vfsObj, clock, fileFlags) if err != nil { return 0, nil, err } @@ -80,7 +81,7 @@ func TimerfdSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne } defer file.DecRef() - tfd, ok := file.Impl().(*vfs.TimerFileDescription) + tfd, ok := file.Impl().(*timerfd.TimerFileDescription) if !ok { return 0, nil, syserror.EINVAL } @@ -114,7 +115,7 @@ func TimerfdGettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne } defer file.DecRef() - tfd, ok := file.Impl().(*vfs.TimerFileDescription) + tfd, ok := file.Impl().(*timerfd.TimerFileDescription) if !ok { return 0, nil, syserror.EINVAL } diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index 86046dd99..94d69c1cc 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -25,7 +25,6 @@ go_library( "device.go", "epoll.go", "epoll_interest_list.go", - "eventfd.go", "file_description.go", "file_description_impl_util.go", "filesystem.go", @@ -37,7 +36,6 @@ go_library( "pathname.go", "permissions.go", "resolving_path.go", - "timerfd.go", "vfs.go", ], visibility = ["//pkg/sentry:internal"], @@ -71,7 +69,6 @@ go_test( name = "vfs_test", size = "small", srcs = [ - "eventfd_test.go", "file_description_impl_util_test.go", "mount_test.go", ], @@ -83,6 +80,5 @@ go_test( "//pkg/sync", "//pkg/syserror", "//pkg/usermem", - "//pkg/waiter", ], ) diff --git a/pkg/sentry/vfs/eventfd.go b/pkg/sentry/vfs/eventfd.go deleted file mode 100644 index f39dacacf..000000000 --- a/pkg/sentry/vfs/eventfd.go +++ /dev/null @@ -1,282 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package vfs - -import ( - "math" - "sync" - "syscall" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fdnotifier" - "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" - "gvisor.dev/gvisor/pkg/waiter" -) - -// EventFileDescription implements FileDescriptionImpl for file-based event -// notification (eventfd). Eventfds are usually internal to the Sentry but in -// certain situations they may be converted into a host-backed eventfd. -type EventFileDescription struct { - vfsfd FileDescription - FileDescriptionDefaultImpl - DentryMetadataFileDescriptionImpl - - // queue is used to notify interested parties when the event object - // becomes readable or writable. - queue waiter.Queue `state:"zerovalue"` - - // mu protects the fields below. - mu sync.Mutex `state:"nosave"` - - // val is the current value of the event counter. - val uint64 - - // semMode specifies whether the event is in "semaphore" mode. - semMode bool - - // hostfd indicates whether this eventfd is passed through to the host. - hostfd int -} - -var _ FileDescriptionImpl = (*EventFileDescription)(nil) - -// NewEventFD creates a new event fd. -func (vfs *VirtualFilesystem) NewEventFD(initVal uint64, semMode bool, flags uint32) (*FileDescription, error) { - vd := vfs.NewAnonVirtualDentry("[eventfd]") - defer vd.DecRef() - efd := &EventFileDescription{ - val: initVal, - semMode: semMode, - hostfd: -1, - } - if err := efd.vfsfd.Init(efd, flags, vd.Mount(), vd.Dentry(), &FileDescriptionOptions{ - UseDentryMetadata: true, - DenyPRead: true, - DenyPWrite: true, - }); err != nil { - return nil, err - } - return &efd.vfsfd, nil -} - -// HostFD returns the host eventfd associated with this event. -func (efd *EventFileDescription) HostFD() (int, error) { - efd.mu.Lock() - defer efd.mu.Unlock() - if efd.hostfd >= 0 { - return efd.hostfd, nil - } - - flags := linux.EFD_NONBLOCK - if efd.semMode { - flags |= linux.EFD_SEMAPHORE - } - - fd, _, errno := syscall.Syscall(syscall.SYS_EVENTFD2, uintptr(efd.val), uintptr(flags), 0) - if errno != 0 { - return -1, errno - } - - if err := fdnotifier.AddFD(int32(fd), &efd.queue); err != nil { - if closeErr := syscall.Close(int(fd)); closeErr != nil { - log.Warningf("close(%d) eventfd failed: %v", fd, closeErr) - } - return -1, err - } - - efd.hostfd = int(fd) - return efd.hostfd, nil -} - -// Release implements FileDescriptionImpl.Release() -func (efd *EventFileDescription) Release() { - efd.mu.Lock() - defer efd.mu.Unlock() - if efd.hostfd >= 0 { - fdnotifier.RemoveFD(int32(efd.hostfd)) - if closeErr := syscall.Close(int(efd.hostfd)); closeErr != nil { - log.Warningf("close(%d) eventfd failed: %v", efd.hostfd, closeErr) - } - efd.hostfd = -1 - } -} - -// Read implements FileDescriptionImpl.Read. -func (efd *EventFileDescription) Read(ctx context.Context, dst usermem.IOSequence, _ ReadOptions) (int64, error) { - if dst.NumBytes() < 8 { - return 0, syscall.EINVAL - } - if err := efd.read(ctx, dst); err != nil { - return 0, err - } - return 8, nil -} - -// Write implements FileDescriptionImpl.Write. -func (efd *EventFileDescription) Write(ctx context.Context, src usermem.IOSequence, _ WriteOptions) (int64, error) { - if src.NumBytes() < 8 { - return 0, syscall.EINVAL - } - if err := efd.write(ctx, src); err != nil { - return 0, err - } - return 8, nil -} - -// Preconditions: Must be called with efd.mu locked. -func (efd *EventFileDescription) hostReadLocked(ctx context.Context, dst usermem.IOSequence) error { - var buf [8]byte - if _, err := syscall.Read(efd.hostfd, buf[:]); err != nil { - if err == syscall.EWOULDBLOCK { - return syserror.ErrWouldBlock - } - return err - } - _, err := dst.CopyOut(ctx, buf[:]) - return err -} - -func (efd *EventFileDescription) read(ctx context.Context, dst usermem.IOSequence) error { - efd.mu.Lock() - if efd.hostfd >= 0 { - defer efd.mu.Unlock() - return efd.hostReadLocked(ctx, dst) - } - - // We can't complete the read if the value is currently zero. - if efd.val == 0 { - efd.mu.Unlock() - return syserror.ErrWouldBlock - } - - // Update the value based on the mode the event is operating in. - var val uint64 - if efd.semMode { - val = 1 - // Consistent with Linux, this is done even if writing to memory fails. - efd.val-- - } else { - val = efd.val - efd.val = 0 - } - - efd.mu.Unlock() - - // Notify writers. We do this even if we were already writable because - // it is possible that a writer is waiting to write the maximum value - // to the event. - efd.queue.Notify(waiter.EventOut) - - var buf [8]byte - usermem.ByteOrder.PutUint64(buf[:], val) - _, err := dst.CopyOut(ctx, buf[:]) - return err -} - -// Preconditions: Must be called with efd.mu locked. -func (efd *EventFileDescription) hostWriteLocked(val uint64) error { - var buf [8]byte - usermem.ByteOrder.PutUint64(buf[:], val) - _, err := syscall.Write(efd.hostfd, buf[:]) - if err == syscall.EWOULDBLOCK { - return syserror.ErrWouldBlock - } - return err -} - -func (efd *EventFileDescription) write(ctx context.Context, src usermem.IOSequence) error { - var buf [8]byte - if _, err := src.CopyIn(ctx, buf[:]); err != nil { - return err - } - val := usermem.ByteOrder.Uint64(buf[:]) - - return efd.Signal(val) -} - -// Signal is an internal function to signal the event fd. -func (efd *EventFileDescription) Signal(val uint64) error { - if val == math.MaxUint64 { - return syscall.EINVAL - } - - efd.mu.Lock() - - if efd.hostfd >= 0 { - defer efd.mu.Unlock() - return efd.hostWriteLocked(val) - } - - // We only allow writes that won't cause the value to go over the max - // uint64 minus 1. - if val > math.MaxUint64-1-efd.val { - efd.mu.Unlock() - return syserror.ErrWouldBlock - } - - efd.val += val - efd.mu.Unlock() - - // Always trigger a notification. - efd.queue.Notify(waiter.EventIn) - - return nil -} - -// Readiness implements waiter.Waitable.Readiness. -func (efd *EventFileDescription) Readiness(mask waiter.EventMask) waiter.EventMask { - efd.mu.Lock() - defer efd.mu.Unlock() - - if efd.hostfd >= 0 { - return fdnotifier.NonBlockingPoll(int32(efd.hostfd), mask) - } - - ready := waiter.EventMask(0) - if efd.val > 0 { - ready |= waiter.EventIn - } - - if efd.val < math.MaxUint64-1 { - ready |= waiter.EventOut - } - - return mask & ready -} - -// EventRegister implements waiter.Waitable.EventRegister. -func (efd *EventFileDescription) EventRegister(entry *waiter.Entry, mask waiter.EventMask) { - efd.queue.EventRegister(entry, mask) - - efd.mu.Lock() - defer efd.mu.Unlock() - if efd.hostfd >= 0 { - fdnotifier.UpdateFD(int32(efd.hostfd)) - } -} - -// EventUnregister implements waiter.Waitable.EventUnregister. -func (efd *EventFileDescription) EventUnregister(entry *waiter.Entry) { - efd.queue.EventUnregister(entry) - - efd.mu.Lock() - defer efd.mu.Unlock() - if efd.hostfd >= 0 { - fdnotifier.UpdateFD(int32(efd.hostfd)) - } -} diff --git a/pkg/sentry/vfs/eventfd_test.go b/pkg/sentry/vfs/eventfd_test.go deleted file mode 100644 index 2dff2d10b..000000000 --- a/pkg/sentry/vfs/eventfd_test.go +++ /dev/null @@ -1,96 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package vfs - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/usermem" - "gvisor.dev/gvisor/pkg/waiter" -) - -func TestEventFD(t *testing.T) { - initVals := []uint64{ - 0, - // Using a non-zero initial value verifies that writing to an - // eventfd signals when the eventfd's counter was already - // non-zero. - 343, - } - - for _, initVal := range initVals { - ctx := contexttest.Context(t) - vfsObj := &VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { - t.Fatalf("VFS init: %v", err) - } - - // Make a new eventfd that is writable. - eventfd, err := vfsObj.NewEventFD(initVal, false, linux.O_RDWR) - if err != nil { - t.Fatalf("NewEventFD failed: %v", err) - } - defer eventfd.DecRef() - - // Register a callback for a write event. - w, ch := waiter.NewChannelEntry(nil) - eventfd.EventRegister(&w, waiter.EventIn) - defer eventfd.EventUnregister(&w) - - data := []byte("00000124") - // Create and submit a write request. - n, err := eventfd.Write(ctx, usermem.BytesIOSequence(data), WriteOptions{}) - if err != nil { - t.Fatal(err) - } - if n != 8 { - t.Errorf("eventfd.write wrote %d bytes, not full int64", n) - } - - // Check if the callback fired due to the write event. - select { - case <-ch: - default: - t.Errorf("Didn't get notified of EventIn after write") - } - } -} - -func TestEventFDStat(t *testing.T) { - ctx := contexttest.Context(t) - vfsObj := &VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { - t.Fatalf("VFS init: %v", err) - } - - // Make a new eventfd that is writable. - eventfd, err := vfsObj.NewEventFD(0, false, linux.O_RDWR) - if err != nil { - t.Fatalf("NewEventFD failed: %v", err) - } - defer eventfd.DecRef() - - statx, err := eventfd.Stat(ctx, StatOptions{ - Mask: linux.STATX_BASIC_STATS, - }) - if err != nil { - t.Fatalf("eventfd.Stat failed: %v", err) - } - if statx.Size != 0 { - t.Errorf("eventfd size should be 0") - } -} diff --git a/pkg/sentry/vfs/timerfd.go b/pkg/sentry/vfs/timerfd.go deleted file mode 100644 index cc536ceaf..000000000 --- a/pkg/sentry/vfs/timerfd.go +++ /dev/null @@ -1,141 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package vfs - -import ( - "sync/atomic" - - "gvisor.dev/gvisor/pkg/context" - ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" - "gvisor.dev/gvisor/pkg/waiter" -) - -// TimerFileDescription implements FileDescriptionImpl for timer fds. It also -// implements ktime.TimerListener. -type TimerFileDescription struct { - vfsfd FileDescription - FileDescriptionDefaultImpl - DentryMetadataFileDescriptionImpl - - events waiter.Queue - timer *ktime.Timer - - // val is the number of timer expirations since the last successful - // call to PRead, or SetTime. val must be accessed using atomic memory - // operations. - val uint64 -} - -var _ FileDescriptionImpl = (*TimerFileDescription)(nil) -var _ ktime.TimerListener = (*TimerFileDescription)(nil) - -// NewTimerFD returns a new timer fd. -func (vfs *VirtualFilesystem) NewTimerFD(clock ktime.Clock, flags uint32) (*FileDescription, error) { - vd := vfs.NewAnonVirtualDentry("[timerfd]") - defer vd.DecRef() - tfd := &TimerFileDescription{} - tfd.timer = ktime.NewTimer(clock, tfd) - if err := tfd.vfsfd.Init(tfd, flags, vd.Mount(), vd.Dentry(), &FileDescriptionOptions{ - UseDentryMetadata: true, - DenyPRead: true, - DenyPWrite: true, - }); err != nil { - return nil, err - } - return &tfd.vfsfd, nil -} - -// Read implements FileDescriptionImpl.Read. -func (tfd *TimerFileDescription) Read(ctx context.Context, dst usermem.IOSequence, opts ReadOptions) (int64, error) { - const sizeofUint64 = 8 - if dst.NumBytes() < sizeofUint64 { - return 0, syserror.EINVAL - } - if val := atomic.SwapUint64(&tfd.val, 0); val != 0 { - var buf [sizeofUint64]byte - usermem.ByteOrder.PutUint64(buf[:], val) - if _, err := dst.CopyOut(ctx, buf[:]); err != nil { - // Linux does not undo consuming the number of - // expirations even if writing to userspace fails. - return 0, err - } - return sizeofUint64, nil - } - return 0, syserror.ErrWouldBlock -} - -// Clock returns the timer fd's Clock. -func (tfd *TimerFileDescription) Clock() ktime.Clock { - return tfd.timer.Clock() -} - -// GetTime returns the associated Timer's setting and the time at which it was -// observed. -func (tfd *TimerFileDescription) GetTime() (ktime.Time, ktime.Setting) { - return tfd.timer.Get() -} - -// SetTime atomically changes the associated Timer's setting, resets the number -// of expirations to 0, and returns the previous setting and the time at which -// it was observed. -func (tfd *TimerFileDescription) SetTime(s ktime.Setting) (ktime.Time, ktime.Setting) { - return tfd.timer.SwapAnd(s, func() { atomic.StoreUint64(&tfd.val, 0) }) -} - -// Readiness implements waiter.Waitable.Readiness. -func (tfd *TimerFileDescription) Readiness(mask waiter.EventMask) waiter.EventMask { - var ready waiter.EventMask - if atomic.LoadUint64(&tfd.val) != 0 { - ready |= waiter.EventIn - } - return ready -} - -// EventRegister implements waiter.Waitable.EventRegister. -func (tfd *TimerFileDescription) EventRegister(e *waiter.Entry, mask waiter.EventMask) { - tfd.events.EventRegister(e, mask) -} - -// EventUnregister implements waiter.Waitable.EventUnregister. -func (tfd *TimerFileDescription) EventUnregister(e *waiter.Entry) { - tfd.events.EventUnregister(e) -} - -// PauseTimer pauses the associated Timer. -func (tfd *TimerFileDescription) PauseTimer() { - tfd.timer.Pause() -} - -// ResumeTimer resumes the associated Timer. -func (tfd *TimerFileDescription) ResumeTimer() { - tfd.timer.Resume() -} - -// Release implements FileDescriptionImpl.Release() -func (tfd *TimerFileDescription) Release() { - tfd.timer.Destroy() -} - -// Notify implements ktime.TimerListener.Notify. -func (tfd *TimerFileDescription) Notify(exp uint64, setting ktime.Setting) (ktime.Setting, bool) { - atomic.AddUint64(&tfd.val, exp) - tfd.events.Notify(waiter.EventIn) - return ktime.Setting{}, false -} - -// Destroy implements ktime.TimerListener.Destroy. -func (tfd *TimerFileDescription) Destroy() {} -- cgit v1.2.3 From 9115f26851b6f00ae01e9c130e3b5b342495c9e5 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Thu, 7 May 2020 14:00:36 -0700 Subject: Allocate device numbers for VFS2 filesystems. Updates #1197, #1198, #1672 PiperOrigin-RevId: 310432006 --- pkg/abi/linux/dev.go | 4 + pkg/sentry/fsimpl/devpts/devpts.go | 40 +++++-- pkg/sentry/fsimpl/ext/ext.go | 16 ++- pkg/sentry/fsimpl/ext/filesystem.go | 8 +- pkg/sentry/fsimpl/ext/inode.go | 2 + pkg/sentry/fsimpl/gofer/gofer.go | 19 +++- pkg/sentry/fsimpl/host/host.go | 143 +++++++++++++------------ pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go | 4 +- pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 36 +++++-- pkg/sentry/fsimpl/kernfs/kernfs_test.go | 6 +- pkg/sentry/fsimpl/kernfs/symlink.go | 8 +- pkg/sentry/fsimpl/pipefs/BUILD | 1 + pkg/sentry/fsimpl/pipefs/pipefs.go | 79 +++++++++----- pkg/sentry/fsimpl/proc/filesystem.go | 29 +++-- pkg/sentry/fsimpl/proc/subtasks.go | 12 +-- pkg/sentry/fsimpl/proc/task.go | 64 +++++------ pkg/sentry/fsimpl/proc/task_fds.go | 28 ++--- pkg/sentry/fsimpl/proc/task_files.go | 16 +-- pkg/sentry/fsimpl/proc/task_net.go | 38 +++---- pkg/sentry/fsimpl/proc/tasks.go | 47 ++++---- pkg/sentry/fsimpl/proc/tasks_files.go | 8 +- pkg/sentry/fsimpl/proc/tasks_sys.go | 108 +++++++++---------- pkg/sentry/fsimpl/sockfs/BUILD | 1 + pkg/sentry/fsimpl/sockfs/sockfs.go | 46 ++++++-- pkg/sentry/fsimpl/sys/sys.go | 21 +++- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 39 ++++--- pkg/sentry/kernel/kernel.go | 10 +- pkg/sentry/socket/hostinet/BUILD | 1 - pkg/sentry/socket/hostinet/socket_vfs2.go | 4 +- pkg/sentry/socket/netlink/BUILD | 1 - pkg/sentry/socket/netlink/provider_vfs2.go | 4 +- pkg/sentry/socket/netstack/BUILD | 1 - pkg/sentry/socket/netstack/netstack_vfs2.go | 4 +- pkg/sentry/socket/unix/BUILD | 1 - pkg/sentry/socket/unix/unix_vfs2.go | 4 +- pkg/sentry/vfs/anonfs.go | 2 +- pkg/sentry/vfs/device.go | 2 +- runsc/boot/loader.go | 5 +- 38 files changed, 517 insertions(+), 345 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/dev.go b/pkg/abi/linux/dev.go index 89f9a793f..fa3ae5f18 100644 --- a/pkg/abi/linux/dev.go +++ b/pkg/abi/linux/dev.go @@ -36,6 +36,10 @@ func DecodeDeviceID(rdev uint32) (uint16, uint32) { // // See Documentations/devices.txt and uapi/linux/major.h. const ( + // UNNAMED_MAJOR is the major device number for "unnamed" devices, whose + // minor numbers are dynamically allocated by the kernel. + UNNAMED_MAJOR = 0 + // MEM_MAJOR is the major device number for "memory" character devices. MEM_MAJOR = 1 diff --git a/pkg/sentry/fsimpl/devpts/devpts.go b/pkg/sentry/fsimpl/devpts/devpts.go index 94db8fe5c..c03c65445 100644 --- a/pkg/sentry/fsimpl/devpts/devpts.go +++ b/pkg/sentry/fsimpl/devpts/devpts.go @@ -51,21 +51,37 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt return nil, nil, syserror.EINVAL } - fs, root := fstype.newFilesystem(vfsObj, creds) - return fs.VFSFilesystem(), root.VFSDentry(), nil + fs, root, err := fstype.newFilesystem(vfsObj, creds) + if err != nil { + return nil, nil, err + } + return fs.Filesystem.VFSFilesystem(), root.VFSDentry(), nil +} + +type filesystem struct { + kernfs.Filesystem + + devMinor uint32 } // newFilesystem creates a new devpts filesystem with root directory and ptmx // master inode. It returns the filesystem and root Dentry. -func (fstype FilesystemType) newFilesystem(vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials) (*kernfs.Filesystem, *kernfs.Dentry) { - fs := &kernfs.Filesystem{} - fs.VFSFilesystem().Init(vfsObj, fstype, fs) +func (fstype FilesystemType) newFilesystem(vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials) (*filesystem, *kernfs.Dentry, error) { + devMinor, err := vfsObj.GetAnonBlockDevMinor() + if err != nil { + return nil, nil, err + } + + fs := &filesystem{ + devMinor: devMinor, + } + fs.Filesystem.VFSFilesystem().Init(vfsObj, fstype, fs) // Construct the root directory. This is always inode id 1. root := &rootInode{ slaves: make(map[uint32]*slaveInode), } - root.InodeAttrs.Init(creds, 1, linux.ModeDirectory|0555) + root.InodeAttrs.Init(creds, linux.UNNAMED_MAJOR, devMinor, 1, linux.ModeDirectory|0555) root.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) root.dentry.Init(root) @@ -74,7 +90,7 @@ func (fstype FilesystemType) newFilesystem(vfsObj *vfs.VirtualFilesystem, creds master := &masterInode{ root: root, } - master.InodeAttrs.Init(creds, 2, linux.ModeCharacterDevice|0666) + master.InodeAttrs.Init(creds, linux.UNNAMED_MAJOR, devMinor, 2, linux.ModeCharacterDevice|0666) master.dentry.Init(master) // Add the master as a child of the root. @@ -83,7 +99,13 @@ func (fstype FilesystemType) newFilesystem(vfsObj *vfs.VirtualFilesystem, creds }) root.IncLinks(links) - return fs, &root.dentry + return fs, &root.dentry, nil +} + +// Release implements vfs.FilesystemImpl.Release. +func (fs *filesystem) Release() { + fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) + fs.Filesystem.Release() } // rootInode is the root directory inode for the devpts mounts. @@ -140,7 +162,7 @@ func (i *rootInode) allocateTerminal(creds *auth.Credentials) (*Terminal, error) } // Linux always uses pty index + 3 as the inode id. See // fs/devpts/inode.c:devpts_pty_new(). - slave.InodeAttrs.Init(creds, uint64(idx+3), linux.ModeCharacterDevice|0600) + slave.InodeAttrs.Init(creds, i.InodeAttrs.DevMajor(), i.InodeAttrs.DevMinor(), uint64(idx+3), linux.ModeCharacterDevice|0600) slave.dentry.Init(slave) i.slaves[idx] = slave diff --git a/pkg/sentry/fsimpl/ext/ext.go b/pkg/sentry/fsimpl/ext/ext.go index 7176af6d1..dac6effbf 100644 --- a/pkg/sentry/fsimpl/ext/ext.go +++ b/pkg/sentry/fsimpl/ext/ext.go @@ -105,36 +105,50 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt // EACCESS should be returned according to mount(2). Filesystem independent // flags (like readonly) are currently not available in pkg/sentry/vfs. + devMinor, err := vfsObj.GetAnonBlockDevMinor() + if err != nil { + return nil, nil, err + } + dev, err := getDeviceFd(source, opts) if err != nil { return nil, nil, err } - fs := filesystem{dev: dev, inodeCache: make(map[uint32]*inode)} + fs := filesystem{ + dev: dev, + inodeCache: make(map[uint32]*inode), + devMinor: devMinor, + } fs.vfsfs.Init(vfsObj, &fsType, &fs) fs.sb, err = readSuperBlock(dev) if err != nil { + fs.vfsfs.DecRef() return nil, nil, err } if fs.sb.Magic() != linux.EXT_SUPER_MAGIC { // mount(2) specifies that EINVAL should be returned if the superblock is // invalid. + fs.vfsfs.DecRef() return nil, nil, syserror.EINVAL } // Refuse to mount if the filesystem is incompatible. if !isCompatible(fs.sb) { + fs.vfsfs.DecRef() return nil, nil, syserror.EINVAL } fs.bgs, err = readBlockGroups(dev, fs.sb) if err != nil { + fs.vfsfs.DecRef() return nil, nil, err } rootInode, err := fs.getOrCreateInodeLocked(disklayout.RootDirInode) if err != nil { + fs.vfsfs.DecRef() return nil, nil, err } rootInode.incRef() diff --git a/pkg/sentry/fsimpl/ext/filesystem.go b/pkg/sentry/fsimpl/ext/filesystem.go index 77b644275..557963e03 100644 --- a/pkg/sentry/fsimpl/ext/filesystem.go +++ b/pkg/sentry/fsimpl/ext/filesystem.go @@ -64,6 +64,10 @@ type filesystem struct { // bgs represents all the block group descriptors for the filesystem. // Immutable after initialization. bgs []disklayout.BlockGroup + + // devMinor is this filesystem's device minor number. Immutable after + // initialization. + devMinor uint32 } // Compiles only if filesystem implements vfs.FilesystemImpl. @@ -366,7 +370,9 @@ func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linu } // Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() {} +func (fs *filesystem) Release() { + fs.vfsfs.VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) +} // Sync implements vfs.FilesystemImpl.Sync. func (fs *filesystem) Sync(ctx context.Context) error { diff --git a/pkg/sentry/fsimpl/ext/inode.go b/pkg/sentry/fsimpl/ext/inode.go index a98512350..485f86f4b 100644 --- a/pkg/sentry/fsimpl/ext/inode.go +++ b/pkg/sentry/fsimpl/ext/inode.go @@ -204,6 +204,8 @@ func (in *inode) statTo(stat *linux.Statx) { stat.Atime = in.diskInode.AccessTime().StatxTimestamp() stat.Ctime = in.diskInode.ChangeTime().StatxTimestamp() stat.Mtime = in.diskInode.ModificationTime().StatxTimestamp() + stat.DevMajor = linux.UNNAMED_MAJOR + stat.DevMinor = in.fs.devMinor // TODO(b/134676337): Set stat.Blocks which is the number of 512 byte blocks // (including metadata blocks) required to represent this file. } diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 9ab8fdc65..e68e37ebc 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -81,6 +81,9 @@ type filesystem struct { // clock is a realtime clock used to set timestamps in file operations. clock ktime.Clock + // devMinor is the filesystem's minor device number. devMinor is immutable. + devMinor uint32 + // uid and gid are the effective KUID and KGID of the filesystem's creator, // and are used as the owner and group for files that don't specify one. // uid and gid are immutable. @@ -399,14 +402,21 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt } // Construct the filesystem object. + devMinor, err := vfsObj.GetAnonBlockDevMinor() + if err != nil { + attachFile.close(ctx) + client.Close() + return nil, nil, err + } fs := &filesystem{ mfp: mfp, opts: fsopts, iopts: iopts, - uid: creds.EffectiveKUID, - gid: creds.EffectiveKGID, client: client, clock: ktime.RealtimeClockFromContext(ctx), + devMinor: devMinor, + uid: creds.EffectiveKUID, + gid: creds.EffectiveKGID, syncableDentries: make(map[*dentry]struct{}), specialFileFDs: make(map[*specialFileFD]struct{}), } @@ -464,6 +474,8 @@ func (fs *filesystem) Release() { // Close the connection to the server. This implicitly clunks all fids. fs.client.Close() } + + fs.vfsfs.VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) } // dentry implements vfs.DentryImpl. @@ -796,7 +808,8 @@ func (d *dentry) statTo(stat *linux.Statx) { stat.Btime = statxTimestampFromDentry(atomic.LoadInt64(&d.btime)) stat.Ctime = statxTimestampFromDentry(atomic.LoadInt64(&d.ctime)) stat.Mtime = statxTimestampFromDentry(atomic.LoadInt64(&d.mtime)) - // TODO(gvisor.dev/issue/1198): device number + stat.DevMajor = linux.UNNAMED_MAJOR + stat.DevMinor = d.fs.devMinor } func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *linux.Statx, mnt *vfs.Mount) error { diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 144e04905..55de9c438 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -115,15 +115,28 @@ func (filesystemType) Name() string { // // Note that there should only ever be one instance of host.filesystem, // a global mount for host fds. -func NewFilesystem(vfsObj *vfs.VirtualFilesystem) *vfs.Filesystem { - fs := &filesystem{} +func NewFilesystem(vfsObj *vfs.VirtualFilesystem) (*vfs.Filesystem, error) { + devMinor, err := vfsObj.GetAnonBlockDevMinor() + if err != nil { + return nil, err + } + fs := &filesystem{ + devMinor: devMinor, + } fs.VFSFilesystem().Init(vfsObj, filesystemType{}, fs) - return fs.VFSFilesystem() + return fs.VFSFilesystem(), nil } // filesystem implements vfs.FilesystemImpl. type filesystem struct { kernfs.Filesystem + + devMinor uint32 +} + +func (fs *filesystem) Release() { + fs.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) + fs.Filesystem.Release() } func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { @@ -219,7 +232,7 @@ func (i *inode) Mode() linux.FileMode { } // Stat implements kernfs.Inode. -func (i *inode) Stat(_ *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { +func (i *inode) Stat(vfsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { if opts.Mask&linux.STATX__RESERVED != 0 { return linux.Statx{}, syserror.EINVAL } @@ -227,73 +240,73 @@ func (i *inode) Stat(_ *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, erro return linux.Statx{}, syserror.EINVAL } + fs := vfsfs.Impl().(*filesystem) + // Limit our host call only to known flags. mask := opts.Mask & linux.STATX_ALL var s unix.Statx_t err := unix.Statx(i.hostFD, "", int(unix.AT_EMPTY_PATH|opts.Sync), int(mask), &s) - // Fallback to fstat(2), if statx(2) is not supported on the host. - // - // TODO(b/151263641): Remove fallback. if err == syserror.ENOSYS { - return i.fstat(opts) - } else if err != nil { + // Fallback to fstat(2), if statx(2) is not supported on the host. + // + // TODO(b/151263641): Remove fallback. + return i.fstat(fs) + } + if err != nil { return linux.Statx{}, err } - ls := linux.Statx{Mask: mask} - // Unconditionally fill blksize, attributes, and device numbers, as indicated - // by /include/uapi/linux/stat.h. - // - // RdevMajor/RdevMinor are left as zero, so as not to expose host device - // numbers. - // - // TODO(gvisor.dev/issue/1672): Use kernfs-specific, internally defined - // device numbers. If we use the device number from the host, it may collide - // with another sentry-internal device number. We handle device/inode - // numbers without relying on the host to prevent collisions. - ls.Blksize = s.Blksize - ls.Attributes = s.Attributes - ls.AttributesMask = s.Attributes_mask - - if mask&linux.STATX_TYPE != 0 { + // Unconditionally fill blksize, attributes, and device numbers, as + // indicated by /include/uapi/linux/stat.h. Inode number is always + // available, since we use our own rather than the host's. + ls := linux.Statx{ + Mask: linux.STATX_INO, + Blksize: s.Blksize, + Attributes: s.Attributes, + Ino: i.ino, + AttributesMask: s.Attributes_mask, + DevMajor: linux.UNNAMED_MAJOR, + DevMinor: fs.devMinor, + } + + // Copy other fields that were returned by the host. RdevMajor/RdevMinor + // are never copied (and therefore left as zero), so as not to expose host + // device numbers. + ls.Mask |= s.Mask & linux.STATX_ALL + if s.Mask&linux.STATX_TYPE != 0 { ls.Mode |= s.Mode & linux.S_IFMT } - if mask&linux.STATX_MODE != 0 { + if s.Mask&linux.STATX_MODE != 0 { ls.Mode |= s.Mode &^ linux.S_IFMT } - if mask&linux.STATX_NLINK != 0 { + if s.Mask&linux.STATX_NLINK != 0 { ls.Nlink = s.Nlink } - if mask&linux.STATX_UID != 0 { + if s.Mask&linux.STATX_UID != 0 { ls.UID = s.Uid } - if mask&linux.STATX_GID != 0 { + if s.Mask&linux.STATX_GID != 0 { ls.GID = s.Gid } - if mask&linux.STATX_ATIME != 0 { + if s.Mask&linux.STATX_ATIME != 0 { ls.Atime = unixToLinuxStatxTimestamp(s.Atime) } - if mask&linux.STATX_BTIME != 0 { + if s.Mask&linux.STATX_BTIME != 0 { ls.Btime = unixToLinuxStatxTimestamp(s.Btime) } - if mask&linux.STATX_CTIME != 0 { + if s.Mask&linux.STATX_CTIME != 0 { ls.Ctime = unixToLinuxStatxTimestamp(s.Ctime) } - if mask&linux.STATX_MTIME != 0 { + if s.Mask&linux.STATX_MTIME != 0 { ls.Mtime = unixToLinuxStatxTimestamp(s.Mtime) } - if mask&linux.STATX_SIZE != 0 { + if s.Mask&linux.STATX_SIZE != 0 { ls.Size = s.Size } - if mask&linux.STATX_BLOCKS != 0 { + if s.Mask&linux.STATX_BLOCKS != 0 { ls.Blocks = s.Blocks } - // Use our own internal inode number. - if mask&linux.STATX_INO != 0 { - ls.Ino = i.ino - } - return ls, nil } @@ -305,36 +318,30 @@ func (i *inode) Stat(_ *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, erro // of a mask or sync flags. fstat(2) does not provide any metadata // equivalent to Statx.Attributes, Statx.AttributesMask, or Statx.Btime, so // those fields remain empty. -func (i *inode) fstat(opts vfs.StatOptions) (linux.Statx, error) { +func (i *inode) fstat(fs *filesystem) (linux.Statx, error) { var s unix.Stat_t if err := unix.Fstat(i.hostFD, &s); err != nil { return linux.Statx{}, err } - // Note that rdev numbers are left as 0; do not expose host device numbers. - ls := linux.Statx{ - Mask: linux.STATX_BASIC_STATS, - Blksize: uint32(s.Blksize), - Nlink: uint32(s.Nlink), - UID: s.Uid, - GID: s.Gid, - Mode: uint16(s.Mode), - Size: uint64(s.Size), - Blocks: uint64(s.Blocks), - Atime: timespecToStatxTimestamp(s.Atim), - Ctime: timespecToStatxTimestamp(s.Ctim), - Mtime: timespecToStatxTimestamp(s.Mtim), - } - - // Use our own internal inode number. - // - // TODO(gvisor.dev/issue/1672): Use a kernfs-specific device number as well. - // If we use the device number from the host, it may collide with another - // sentry-internal device number. We handle device/inode numbers without - // relying on the host to prevent collisions. - ls.Ino = i.ino - - return ls, nil + // As with inode.Stat(), we always use internal device and inode numbers, + // and never expose the host's represented device numbers. + return linux.Statx{ + Mask: linux.STATX_BASIC_STATS, + Blksize: uint32(s.Blksize), + Nlink: uint32(s.Nlink), + UID: s.Uid, + GID: s.Gid, + Mode: uint16(s.Mode), + Ino: i.ino, + Size: uint64(s.Size), + Blocks: uint64(s.Blocks), + Atime: timespecToStatxTimestamp(s.Atim), + Ctime: timespecToStatxTimestamp(s.Ctim), + Mtime: timespecToStatxTimestamp(s.Mtim), + DevMajor: linux.UNNAMED_MAJOR, + DevMinor: fs.devMinor, + }, nil } // SetStat implements kernfs.Inode. @@ -453,8 +460,6 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount) (*vfs.F } // fileDescription is embedded by host fd implementations of FileDescriptionImpl. -// -// TODO(gvisor.dev/issue/1672): Implement Waitable interface. type fileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl @@ -471,12 +476,12 @@ type fileDescription struct { // SetStat implements vfs.FileDescriptionImpl. func (f *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { creds := auth.CredentialsFromContext(ctx) - return f.inode.SetStat(ctx, nil, creds, opts) + return f.inode.SetStat(ctx, f.vfsfd.Mount().Filesystem(), creds, opts) } // Stat implements vfs.FileDescriptionImpl. func (f *fileDescription) Stat(_ context.Context, opts vfs.StatOptions) (linux.Statx, error) { - return f.inode.Stat(nil, opts) + return f.inode.Stat(f.vfsfd.Mount().Filesystem(), opts) } // Release implements vfs.FileDescriptionImpl. diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go index c7779fc11..1568a9d49 100644 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -44,11 +44,11 @@ type DynamicBytesFile struct { var _ Inode = (*DynamicBytesFile)(nil) // Init initializes a dynamic bytes file. -func (f *DynamicBytesFile) Init(creds *auth.Credentials, ino uint64, data vfs.DynamicBytesSource, perm linux.FileMode) { +func (f *DynamicBytesFile) Init(creds *auth.Credentials, devMajor, devMinor uint32, ino uint64, data vfs.DynamicBytesSource, perm linux.FileMode) { if perm&^linux.PermissionsMask != 0 { panic(fmt.Sprintf("Only permission mask must be set: %x", perm&linux.PermissionsMask)) } - f.InodeAttrs.Init(creds, ino, linux.ModeRegular|perm) + f.InodeAttrs.Init(creds, devMajor, devMinor, ino, linux.ModeRegular|perm) f.data = data } diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 615592d5f..982daa2e6 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -192,15 +192,17 @@ func (InodeNotSymlink) Getlink(context.Context, *vfs.Mount) (vfs.VirtualDentry, // // Must be initialized by Init prior to first use. type InodeAttrs struct { - ino uint64 - mode uint32 - uid uint32 - gid uint32 - nlink uint32 + devMajor uint32 + devMinor uint32 + ino uint64 + mode uint32 + uid uint32 + gid uint32 + nlink uint32 } // Init initializes this InodeAttrs. -func (a *InodeAttrs) Init(creds *auth.Credentials, ino uint64, mode linux.FileMode) { +func (a *InodeAttrs) Init(creds *auth.Credentials, devMajor, devMinor uint32, ino uint64, mode linux.FileMode) { if mode.FileType() == 0 { panic(fmt.Sprintf("No file type specified in 'mode' for InodeAttrs.Init(): mode=0%o", mode)) } @@ -209,6 +211,8 @@ func (a *InodeAttrs) Init(creds *auth.Credentials, ino uint64, mode linux.FileMo if mode.FileType() == linux.ModeDirectory { nlink = 2 } + a.devMajor = devMajor + a.devMinor = devMinor atomic.StoreUint64(&a.ino, ino) atomic.StoreUint32(&a.mode, uint32(mode)) atomic.StoreUint32(&a.uid, uint32(creds.EffectiveKUID)) @@ -216,6 +220,16 @@ func (a *InodeAttrs) Init(creds *auth.Credentials, ino uint64, mode linux.FileMo atomic.StoreUint32(&a.nlink, nlink) } +// DevMajor returns the device major number. +func (a *InodeAttrs) DevMajor() uint32 { + return a.devMajor +} + +// DevMinor returns the device minor number. +func (a *InodeAttrs) DevMinor() uint32 { + return a.devMinor +} + // Ino returns the inode id. func (a *InodeAttrs) Ino() uint64 { return atomic.LoadUint64(&a.ino) @@ -232,6 +246,8 @@ func (a *InodeAttrs) Mode() linux.FileMode { func (a *InodeAttrs) Stat(*vfs.Filesystem, vfs.StatOptions) (linux.Statx, error) { var stat linux.Statx stat.Mask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_UID | linux.STATX_GID | linux.STATX_INO | linux.STATX_NLINK + stat.DevMajor = a.devMajor + stat.DevMinor = a.devMinor stat.Ino = atomic.LoadUint64(&a.ino) stat.Mode = uint16(a.Mode()) stat.UID = atomic.LoadUint32(&a.uid) @@ -544,9 +560,9 @@ type StaticDirectory struct { var _ Inode = (*StaticDirectory)(nil) // NewStaticDir creates a new static directory and returns its dentry. -func NewStaticDir(creds *auth.Credentials, ino uint64, perm linux.FileMode, children map[string]*Dentry) *Dentry { +func NewStaticDir(creds *auth.Credentials, devMajor, devMinor uint32, ino uint64, perm linux.FileMode, children map[string]*Dentry) *Dentry { inode := &StaticDirectory{} - inode.Init(creds, ino, perm) + inode.Init(creds, devMajor, devMinor, ino, perm) dentry := &Dentry{} dentry.Init(inode) @@ -559,11 +575,11 @@ func NewStaticDir(creds *auth.Credentials, ino uint64, perm linux.FileMode, chil } // Init initializes StaticDirectory. -func (s *StaticDirectory) Init(creds *auth.Credentials, ino uint64, perm linux.FileMode) { +func (s *StaticDirectory) Init(creds *auth.Credentials, devMajor, devMinor uint32, ino uint64, perm linux.FileMode) { if perm&^linux.PermissionsMask != 0 { panic(fmt.Sprintf("Only permission mask must be set: %x", perm&linux.PermissionsMask)) } - s.InodeAttrs.Init(creds, ino, linux.ModeDirectory|perm) + s.InodeAttrs.Init(creds, devMajor, devMinor, ino, linux.ModeDirectory|perm) } // Open implements kernfs.Inode. diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index 1c5d3e7e7..412cf6ac9 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -75,7 +75,7 @@ type file struct { func (fs *filesystem) newFile(creds *auth.Credentials, content string) *kernfs.Dentry { f := &file{} f.content = content - f.DynamicBytesFile.Init(creds, fs.NextIno(), f, 0777) + f.DynamicBytesFile.Init(creds, 0 /* devMajor */, 0 /* devMinor */, fs.NextIno(), f, 0777) d := &kernfs.Dentry{} d.Init(f) @@ -107,7 +107,7 @@ type readonlyDir struct { func (fs *filesystem) newReadonlyDir(creds *auth.Credentials, mode linux.FileMode, contents map[string]*kernfs.Dentry) *kernfs.Dentry { dir := &readonlyDir{} - dir.attrs.Init(creds, fs.NextIno(), linux.ModeDirectory|mode) + dir.attrs.Init(creds, 0 /* devMajor */, 0 /* devMinor */, fs.NextIno(), linux.ModeDirectory|mode) dir.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) dir.dentry.Init(dir) @@ -137,7 +137,7 @@ type dir struct { func (fs *filesystem) newDir(creds *auth.Credentials, mode linux.FileMode, contents map[string]*kernfs.Dentry) *kernfs.Dentry { dir := &dir{} dir.fs = fs - dir.attrs.Init(creds, fs.NextIno(), linux.ModeDirectory|mode) + dir.attrs.Init(creds, 0 /* devMajor */, 0 /* devMinor */, fs.NextIno(), linux.ModeDirectory|mode) dir.OrderedChildren.Init(kernfs.OrderedChildrenOptions{Writable: true}) dir.dentry.Init(dir) diff --git a/pkg/sentry/fsimpl/kernfs/symlink.go b/pkg/sentry/fsimpl/kernfs/symlink.go index 0aa6dc979..2ab3f53fd 100644 --- a/pkg/sentry/fsimpl/kernfs/symlink.go +++ b/pkg/sentry/fsimpl/kernfs/symlink.go @@ -35,9 +35,9 @@ type StaticSymlink struct { var _ Inode = (*StaticSymlink)(nil) // NewStaticSymlink creates a new symlink file pointing to 'target'. -func NewStaticSymlink(creds *auth.Credentials, ino uint64, target string) *Dentry { +func NewStaticSymlink(creds *auth.Credentials, devMajor, devMinor uint32, ino uint64, target string) *Dentry { inode := &StaticSymlink{} - inode.Init(creds, ino, target) + inode.Init(creds, devMajor, devMinor, ino, target) d := &Dentry{} d.Init(inode) @@ -45,9 +45,9 @@ func NewStaticSymlink(creds *auth.Credentials, ino uint64, target string) *Dentr } // Init initializes the instance. -func (s *StaticSymlink) Init(creds *auth.Credentials, ino uint64, target string) { +func (s *StaticSymlink) Init(creds *auth.Credentials, devMajor uint32, devMinor uint32, ino uint64, target string) { s.target = target - s.InodeAttrs.Init(creds, ino, linux.ModeSymlink|0777) + s.InodeAttrs.Init(creds, devMajor, devMinor, ino, linux.ModeSymlink|0777) } // Readlink implements Inode. diff --git a/pkg/sentry/fsimpl/pipefs/BUILD b/pkg/sentry/fsimpl/pipefs/BUILD index 0d411606f..5950a2d59 100644 --- a/pkg/sentry/fsimpl/pipefs/BUILD +++ b/pkg/sentry/fsimpl/pipefs/BUILD @@ -9,6 +9,7 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/context", + "//pkg/fspath", "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/pipe", diff --git a/pkg/sentry/fsimpl/pipefs/pipefs.go b/pkg/sentry/fsimpl/pipefs/pipefs.go index 5375e5e75..cab771211 100644 --- a/pkg/sentry/fsimpl/pipefs/pipefs.go +++ b/pkg/sentry/fsimpl/pipefs/pipefs.go @@ -17,8 +17,11 @@ package pipefs import ( + "fmt" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" @@ -40,20 +43,36 @@ func (filesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFile panic("pipefs.filesystemType.GetFilesystem should never be called") } -// TODO(gvisor.dev/issue/1193): -// -// - kernfs does not provide a way to implement statfs, from which we -// should indicate PIPEFS_MAGIC. -// -// - kernfs does not provide a way to override names for -// vfs.FilesystemImpl.PrependPath(); pipefs inodes should use synthetic -// name fmt.Sprintf("pipe:[%d]", inode.ino). +type filesystem struct { + kernfs.Filesystem + + devMinor uint32 +} // NewFilesystem sets up and returns a new vfs.Filesystem implemented by pipefs. -func NewFilesystem(vfsObj *vfs.VirtualFilesystem) *vfs.Filesystem { - fs := &kernfs.Filesystem{} - fs.VFSFilesystem().Init(vfsObj, filesystemType{}, fs) - return fs.VFSFilesystem() +func NewFilesystem(vfsObj *vfs.VirtualFilesystem) (*vfs.Filesystem, error) { + devMinor, err := vfsObj.GetAnonBlockDevMinor() + if err != nil { + return nil, err + } + fs := &filesystem{ + devMinor: devMinor, + } + fs.Filesystem.VFSFilesystem().Init(vfsObj, filesystemType{}, fs) + return fs.Filesystem.VFSFilesystem(), nil +} + +// Release implements vfs.FilesystemImpl.Release. +func (fs *filesystem) Release() { + fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) + fs.Filesystem.Release() +} + +// PrependPath implements vfs.FilesystemImpl.PrependPath. +func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { + inode := vd.Dentry().Impl().(*kernfs.Dentry).Inode().(*inode) + b.PrependComponent(fmt.Sprintf("pipe:[%d]", inode.ino)) + return vfs.PrependPathSyntheticError{} } // inode implements kernfs.Inode. @@ -71,11 +90,11 @@ type inode struct { ctime ktime.Time } -func newInode(ctx context.Context, fs *kernfs.Filesystem) *inode { +func newInode(ctx context.Context, fs *filesystem) *inode { creds := auth.CredentialsFromContext(ctx) return &inode{ pipe: pipe.NewVFSPipe(false /* isNamed */, pipe.DefaultPipeSize, usermem.PageSize), - ino: fs.NextIno(), + ino: fs.Filesystem.NextIno(), uid: creds.EffectiveKUID, gid: creds.EffectiveKGID, ctime: ktime.NowFromContext(ctx), @@ -98,19 +117,20 @@ func (i *inode) Mode() linux.FileMode { func (i *inode) Stat(vfsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { ts := linux.NsecToStatxTimestamp(i.ctime.Nanoseconds()) return linux.Statx{ - Mask: linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_NLINK | linux.STATX_UID | linux.STATX_GID | linux.STATX_ATIME | linux.STATX_MTIME | linux.STATX_CTIME | linux.STATX_INO | linux.STATX_SIZE | linux.STATX_BLOCKS, - Blksize: usermem.PageSize, - Nlink: 1, - UID: uint32(i.uid), - GID: uint32(i.gid), - Mode: pipeMode, - Ino: i.ino, - Size: 0, - Blocks: 0, - Atime: ts, - Ctime: ts, - Mtime: ts, - // TODO(gvisor.dev/issue/1197): Device number. + Mask: linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_NLINK | linux.STATX_UID | linux.STATX_GID | linux.STATX_ATIME | linux.STATX_MTIME | linux.STATX_CTIME | linux.STATX_INO | linux.STATX_SIZE | linux.STATX_BLOCKS, + Blksize: usermem.PageSize, + Nlink: 1, + UID: uint32(i.uid), + GID: uint32(i.gid), + Mode: pipeMode, + Ino: i.ino, + Size: 0, + Blocks: 0, + Atime: ts, + Ctime: ts, + Mtime: ts, + DevMajor: linux.UNNAMED_MAJOR, + DevMinor: vfsfs.Impl().(*filesystem).devMinor, }, nil } @@ -122,6 +142,9 @@ func (i *inode) SetStat(ctx context.Context, vfsfs *vfs.Filesystem, creds *auth. return syserror.EPERM } +// TODO(gvisor.dev/issue/1193): kernfs does not provide a way to implement +// statfs, from which we should indicate PIPEFS_MAGIC. + // Open implements kernfs.Inode.Open. func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { return i.pipe.Open(ctx, rp.Mount(), vfsd, opts.Flags) @@ -132,7 +155,7 @@ func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentr // // Preconditions: mnt.Filesystem() must have been returned by NewFilesystem(). func NewConnectedPipeFDs(ctx context.Context, mnt *vfs.Mount, flags uint32) (*vfs.FileDescription, *vfs.FileDescription) { - fs := mnt.Filesystem().Impl().(*kernfs.Filesystem) + fs := mnt.Filesystem().Impl().(*filesystem) inode := newInode(ctx, fs) var d kernfs.Dentry d.Init(inode) diff --git a/pkg/sentry/fsimpl/proc/filesystem.go b/pkg/sentry/fsimpl/proc/filesystem.go index 104fc9030..609210253 100644 --- a/pkg/sentry/fsimpl/proc/filesystem.go +++ b/pkg/sentry/fsimpl/proc/filesystem.go @@ -41,6 +41,12 @@ func (FilesystemType) Name() string { return Name } +type filesystem struct { + kernfs.Filesystem + + devMinor uint32 +} + // GetFilesystem implements vfs.FilesystemType.GetFilesystem. func (ft FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { k := kernel.KernelFromContext(ctx) @@ -51,8 +57,13 @@ func (ft FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualF if pidns == nil { return nil, nil, fmt.Errorf("procfs requires a PID namespace") } - - procfs := &kernfs.Filesystem{} + devMinor, err := vfsObj.GetAnonBlockDevMinor() + if err != nil { + return nil, nil, err + } + procfs := &filesystem{ + devMinor: devMinor, + } procfs.VFSFilesystem().Init(vfsObj, &ft, procfs) var cgroups map[string]string @@ -61,21 +72,27 @@ func (ft FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualF cgroups = data.Cgroups } - _, dentry := newTasksInode(procfs, k, pidns, cgroups) + _, dentry := procfs.newTasksInode(k, pidns, cgroups) return procfs.VFSFilesystem(), dentry.VFSDentry(), nil } +// Release implements vfs.FilesystemImpl.Release. +func (fs *filesystem) Release() { + fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) + fs.Filesystem.Release() +} + // dynamicInode is an overfitted interface for common Inodes with // dynamicByteSource types used in procfs. type dynamicInode interface { kernfs.Inode vfs.DynamicBytesSource - Init(creds *auth.Credentials, ino uint64, data vfs.DynamicBytesSource, perm linux.FileMode) + Init(creds *auth.Credentials, devMajor, devMinor uint32, ino uint64, data vfs.DynamicBytesSource, perm linux.FileMode) } -func newDentry(creds *auth.Credentials, ino uint64, perm linux.FileMode, inode dynamicInode) *kernfs.Dentry { - inode.Init(creds, ino, inode, perm) +func (fs *filesystem) newDentry(creds *auth.Credentials, ino uint64, perm linux.FileMode, inode dynamicInode) *kernfs.Dentry { + inode.Init(creds, linux.UNNAMED_MAJOR, fs.devMinor, ino, inode, perm) d := &kernfs.Dentry{} d.Init(inode) diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go index a5cfa8333..36a911db4 100644 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -37,23 +37,23 @@ type subtasksInode struct { kernfs.OrderedChildren kernfs.AlwaysValid + fs *filesystem task *kernel.Task pidns *kernel.PIDNamespace - inoGen InoGenerator cgroupControllers map[string]string } var _ kernfs.Inode = (*subtasksInode)(nil) -func newSubtasks(task *kernel.Task, pidns *kernel.PIDNamespace, inoGen InoGenerator, cgroupControllers map[string]string) *kernfs.Dentry { +func (fs *filesystem) newSubtasks(task *kernel.Task, pidns *kernel.PIDNamespace, cgroupControllers map[string]string) *kernfs.Dentry { subInode := &subtasksInode{ + fs: fs, task: task, pidns: pidns, - inoGen: inoGen, cgroupControllers: cgroupControllers, } // Note: credentials are overridden by taskOwnedInode. - subInode.InodeAttrs.Init(task.Credentials(), inoGen.NextIno(), linux.ModeDirectory|0555) + subInode.InodeAttrs.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0555) subInode.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) inode := &taskOwnedInode{Inode: subInode, owner: task} @@ -78,7 +78,7 @@ func (i *subtasksInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, e return nil, syserror.ENOENT } - subTaskDentry := newTaskInode(i.inoGen, subTask, i.pidns, false, i.cgroupControllers) + subTaskDentry := i.fs.newTaskInode(subTask, i.pidns, false, i.cgroupControllers) return subTaskDentry.VFSDentry(), nil } @@ -102,7 +102,7 @@ func (i *subtasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallb dirent := vfs.Dirent{ Name: strconv.FormatUint(uint64(tid), 10), Type: linux.DT_DIR, - Ino: i.inoGen.NextIno(), + Ino: i.fs.NextIno(), NextOff: offset + 1, } if err := cb.Handle(dirent); err != nil { diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index 66419d91b..482055db1 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -43,45 +43,45 @@ type taskInode struct { var _ kernfs.Inode = (*taskInode)(nil) -func newTaskInode(inoGen InoGenerator, task *kernel.Task, pidns *kernel.PIDNamespace, isThreadGroup bool, cgroupControllers map[string]string) *kernfs.Dentry { +func (fs *filesystem) newTaskInode(task *kernel.Task, pidns *kernel.PIDNamespace, isThreadGroup bool, cgroupControllers map[string]string) *kernfs.Dentry { // TODO(gvisor.dev/issue/164): Fail with ESRCH if task exited. contents := map[string]*kernfs.Dentry{ - "auxv": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &auxvData{task: task}), - "cmdline": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &cmdlineData{task: task, arg: cmdlineDataArg}), - "comm": newComm(task, inoGen.NextIno(), 0444), - "environ": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &cmdlineData{task: task, arg: environDataArg}), - "exe": newExeSymlink(task, inoGen.NextIno()), - "fd": newFDDirInode(task, inoGen), - "fdinfo": newFDInfoDirInode(task, inoGen), - "gid_map": newTaskOwnedFile(task, inoGen.NextIno(), 0644, &idMapData{task: task, gids: true}), - "io": newTaskOwnedFile(task, inoGen.NextIno(), 0400, newIO(task, isThreadGroup)), - "maps": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &mapsData{task: task}), - "mountinfo": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &mountInfoData{task: task}), - "mounts": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &mountsData{task: task}), - "net": newTaskNetDir(task, inoGen), - "ns": newTaskOwnedDir(task, inoGen.NextIno(), 0511, map[string]*kernfs.Dentry{ - "net": newNamespaceSymlink(task, inoGen.NextIno(), "net"), - "pid": newNamespaceSymlink(task, inoGen.NextIno(), "pid"), - "user": newNamespaceSymlink(task, inoGen.NextIno(), "user"), + "auxv": fs.newTaskOwnedFile(task, fs.NextIno(), 0444, &auxvData{task: task}), + "cmdline": fs.newTaskOwnedFile(task, fs.NextIno(), 0444, &cmdlineData{task: task, arg: cmdlineDataArg}), + "comm": fs.newComm(task, fs.NextIno(), 0444), + "environ": fs.newTaskOwnedFile(task, fs.NextIno(), 0444, &cmdlineData{task: task, arg: environDataArg}), + "exe": fs.newExeSymlink(task, fs.NextIno()), + "fd": fs.newFDDirInode(task), + "fdinfo": fs.newFDInfoDirInode(task), + "gid_map": fs.newTaskOwnedFile(task, fs.NextIno(), 0644, &idMapData{task: task, gids: true}), + "io": fs.newTaskOwnedFile(task, fs.NextIno(), 0400, newIO(task, isThreadGroup)), + "maps": fs.newTaskOwnedFile(task, fs.NextIno(), 0444, &mapsData{task: task}), + "mountinfo": fs.newTaskOwnedFile(task, fs.NextIno(), 0444, &mountInfoData{task: task}), + "mounts": fs.newTaskOwnedFile(task, fs.NextIno(), 0444, &mountsData{task: task}), + "net": fs.newTaskNetDir(task), + "ns": fs.newTaskOwnedDir(task, fs.NextIno(), 0511, map[string]*kernfs.Dentry{ + "net": fs.newNamespaceSymlink(task, fs.NextIno(), "net"), + "pid": fs.newNamespaceSymlink(task, fs.NextIno(), "pid"), + "user": fs.newNamespaceSymlink(task, fs.NextIno(), "user"), }), - "oom_score": newTaskOwnedFile(task, inoGen.NextIno(), 0444, newStaticFile("0\n")), - "oom_score_adj": newTaskOwnedFile(task, inoGen.NextIno(), 0644, &oomScoreAdj{task: task}), - "smaps": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &smapsData{task: task}), - "stat": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &taskStatData{task: task, pidns: pidns, tgstats: isThreadGroup}), - "statm": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &statmData{task: task}), - "status": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &statusData{task: task, pidns: pidns}), - "uid_map": newTaskOwnedFile(task, inoGen.NextIno(), 0644, &idMapData{task: task, gids: false}), + "oom_score": fs.newTaskOwnedFile(task, fs.NextIno(), 0444, newStaticFile("0\n")), + "oom_score_adj": fs.newTaskOwnedFile(task, fs.NextIno(), 0644, &oomScoreAdj{task: task}), + "smaps": fs.newTaskOwnedFile(task, fs.NextIno(), 0444, &smapsData{task: task}), + "stat": fs.newTaskOwnedFile(task, fs.NextIno(), 0444, &taskStatData{task: task, pidns: pidns, tgstats: isThreadGroup}), + "statm": fs.newTaskOwnedFile(task, fs.NextIno(), 0444, &statmData{task: task}), + "status": fs.newTaskOwnedFile(task, fs.NextIno(), 0444, &statusData{task: task, pidns: pidns}), + "uid_map": fs.newTaskOwnedFile(task, fs.NextIno(), 0644, &idMapData{task: task, gids: false}), } if isThreadGroup { - contents["task"] = newSubtasks(task, pidns, inoGen, cgroupControllers) + contents["task"] = fs.newSubtasks(task, pidns, cgroupControllers) } if len(cgroupControllers) > 0 { - contents["cgroup"] = newTaskOwnedFile(task, inoGen.NextIno(), 0444, newCgroupData(cgroupControllers)) + contents["cgroup"] = fs.newTaskOwnedFile(task, fs.NextIno(), 0444, newCgroupData(cgroupControllers)) } taskInode := &taskInode{task: task} // Note: credentials are overridden by taskOwnedInode. - taskInode.InodeAttrs.Init(task.Credentials(), inoGen.NextIno(), linux.ModeDirectory|0555) + taskInode.InodeAttrs.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0555) inode := &taskOwnedInode{Inode: taskInode, owner: task} dentry := &kernfs.Dentry{} @@ -126,9 +126,9 @@ type taskOwnedInode struct { var _ kernfs.Inode = (*taskOwnedInode)(nil) -func newTaskOwnedFile(task *kernel.Task, ino uint64, perm linux.FileMode, inode dynamicInode) *kernfs.Dentry { +func (fs *filesystem) newTaskOwnedFile(task *kernel.Task, ino uint64, perm linux.FileMode, inode dynamicInode) *kernfs.Dentry { // Note: credentials are overridden by taskOwnedInode. - inode.Init(task.Credentials(), ino, inode, perm) + inode.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, ino, inode, perm) taskInode := &taskOwnedInode{Inode: inode, owner: task} d := &kernfs.Dentry{} @@ -136,11 +136,11 @@ func newTaskOwnedFile(task *kernel.Task, ino uint64, perm linux.FileMode, inode return d } -func newTaskOwnedDir(task *kernel.Task, ino uint64, perm linux.FileMode, children map[string]*kernfs.Dentry) *kernfs.Dentry { +func (fs *filesystem) newTaskOwnedDir(task *kernel.Task, ino uint64, perm linux.FileMode, children map[string]*kernfs.Dentry) *kernfs.Dentry { dir := &kernfs.StaticDirectory{} // Note: credentials are overridden by taskOwnedInode. - dir.Init(task.Credentials(), ino, perm) + dir.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, ino, perm) inode := &taskOwnedInode{Inode: dir, owner: task} d := &kernfs.Dentry{} diff --git a/pkg/sentry/fsimpl/proc/task_fds.go b/pkg/sentry/fsimpl/proc/task_fds.go index 8ad976073..44ccc9e4a 100644 --- a/pkg/sentry/fsimpl/proc/task_fds.go +++ b/pkg/sentry/fsimpl/proc/task_fds.go @@ -53,8 +53,8 @@ func taskFDExists(t *kernel.Task, fd int32) bool { } type fdDir struct { - inoGen InoGenerator - task *kernel.Task + fs *filesystem + task *kernel.Task // When produceSymlinks is set, dirents produces for the FDs are reported // as symlink. Otherwise, they are reported as regular files. @@ -85,7 +85,7 @@ func (i *fdDir) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, abs dirent := vfs.Dirent{ Name: strconv.FormatUint(uint64(fd), 10), Type: typ, - Ino: i.inoGen.NextIno(), + Ino: i.fs.NextIno(), NextOff: offset + 1, } if err := cb.Handle(dirent); err != nil { @@ -110,15 +110,15 @@ type fdDirInode struct { var _ kernfs.Inode = (*fdDirInode)(nil) -func newFDDirInode(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry { +func (fs *filesystem) newFDDirInode(task *kernel.Task) *kernfs.Dentry { inode := &fdDirInode{ fdDir: fdDir{ - inoGen: inoGen, + fs: fs, task: task, produceSymlink: true, }, } - inode.InodeAttrs.Init(task.Credentials(), inoGen.NextIno(), linux.ModeDirectory|0555) + inode.InodeAttrs.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0555) dentry := &kernfs.Dentry{} dentry.Init(inode) @@ -137,7 +137,7 @@ func (i *fdDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, erro if !taskFDExists(i.task, fd) { return nil, syserror.ENOENT } - taskDentry := newFDSymlink(i.task, fd, i.inoGen.NextIno()) + taskDentry := i.fs.newFDSymlink(i.task, fd, i.fs.NextIno()) return taskDentry.VFSDentry(), nil } @@ -186,12 +186,12 @@ type fdSymlink struct { var _ kernfs.Inode = (*fdSymlink)(nil) -func newFDSymlink(task *kernel.Task, fd int32, ino uint64) *kernfs.Dentry { +func (fs *filesystem) newFDSymlink(task *kernel.Task, fd int32, ino uint64) *kernfs.Dentry { inode := &fdSymlink{ task: task, fd: fd, } - inode.Init(task.Credentials(), ino, linux.ModeSymlink|0777) + inode.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, ino, linux.ModeSymlink|0777) d := &kernfs.Dentry{} d.Init(inode) @@ -234,14 +234,14 @@ type fdInfoDirInode struct { var _ kernfs.Inode = (*fdInfoDirInode)(nil) -func newFDInfoDirInode(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry { +func (fs *filesystem) newFDInfoDirInode(task *kernel.Task) *kernfs.Dentry { inode := &fdInfoDirInode{ fdDir: fdDir{ - inoGen: inoGen, - task: task, + fs: fs, + task: task, }, } - inode.InodeAttrs.Init(task.Credentials(), inoGen.NextIno(), linux.ModeDirectory|0555) + inode.InodeAttrs.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0555) dentry := &kernfs.Dentry{} dentry.Init(inode) @@ -264,7 +264,7 @@ func (i *fdInfoDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, task: i.task, fd: fd, } - dentry := newTaskOwnedFile(i.task, i.inoGen.NextIno(), 0444, data) + dentry := i.fs.newTaskOwnedFile(i.task, i.fs.NextIno(), 0444, data) return dentry.VFSDentry(), nil } diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index 515f25327..2f297e48a 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -241,9 +241,9 @@ type commInode struct { task *kernel.Task } -func newComm(task *kernel.Task, ino uint64, perm linux.FileMode) *kernfs.Dentry { +func (fs *filesystem) newComm(task *kernel.Task, ino uint64, perm linux.FileMode) *kernfs.Dentry { inode := &commInode{task: task} - inode.DynamicBytesFile.Init(task.Credentials(), ino, &commData{task: task}, perm) + inode.DynamicBytesFile.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, ino, &commData{task: task}, perm) d := &kernfs.Dentry{} d.Init(inode) @@ -596,9 +596,9 @@ type exeSymlink struct { var _ kernfs.Inode = (*exeSymlink)(nil) -func newExeSymlink(task *kernel.Task, ino uint64) *kernfs.Dentry { +func (fs *filesystem) newExeSymlink(task *kernel.Task, ino uint64) *kernfs.Dentry { inode := &exeSymlink{task: task} - inode.Init(task.Credentials(), ino, linux.ModeSymlink|0777) + inode.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, ino, linux.ModeSymlink|0777) d := &kernfs.Dentry{} d.Init(inode) @@ -729,7 +729,7 @@ type namespaceSymlink struct { task *kernel.Task } -func newNamespaceSymlink(task *kernel.Task, ino uint64, ns string) *kernfs.Dentry { +func (fs *filesystem) newNamespaceSymlink(task *kernel.Task, ino uint64, ns string) *kernfs.Dentry { // Namespace symlinks should contain the namespace name and the inode number // for the namespace instance, so for example user:[123456]. We currently fake // the inode number by sticking the symlink inode in its place. @@ -737,7 +737,7 @@ func newNamespaceSymlink(task *kernel.Task, ino uint64, ns string) *kernfs.Dentr inode := &namespaceSymlink{task: task} // Note: credentials are overridden by taskOwnedInode. - inode.Init(task.Credentials(), ino, target) + inode.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, ino, target) taskInode := &taskOwnedInode{Inode: inode, owner: task} d := &kernfs.Dentry{} @@ -780,11 +780,11 @@ type namespaceInode struct { var _ kernfs.Inode = (*namespaceInode)(nil) // Init initializes a namespace inode. -func (i *namespaceInode) Init(creds *auth.Credentials, ino uint64, perm linux.FileMode) { +func (i *namespaceInode) Init(creds *auth.Credentials, devMajor, devMinor uint32, ino uint64, perm linux.FileMode) { if perm&^linux.PermissionsMask != 0 { panic(fmt.Sprintf("Only permission mask must be set: %x", perm&linux.PermissionsMask)) } - i.InodeAttrs.Init(creds, ino, linux.ModeRegular|perm) + i.InodeAttrs.Init(creds, devMajor, devMinor, ino, linux.ModeRegular|perm) } // Open implements Inode.Open. diff --git a/pkg/sentry/fsimpl/proc/task_net.go b/pkg/sentry/fsimpl/proc/task_net.go index 9c329341a..6bde27376 100644 --- a/pkg/sentry/fsimpl/proc/task_net.go +++ b/pkg/sentry/fsimpl/proc/task_net.go @@ -37,7 +37,7 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) -func newTaskNetDir(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry { +func (fs *filesystem) newTaskNetDir(task *kernel.Task) *kernfs.Dentry { k := task.Kernel() pidns := task.PIDNamespace() root := auth.NewRootCredentials(pidns.UserNamespace()) @@ -57,37 +57,37 @@ func newTaskNetDir(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry { // TODO(gvisor.dev/issue/1833): Make sure file contents reflect the task // network namespace. contents = map[string]*kernfs.Dentry{ - "dev": newDentry(root, inoGen.NextIno(), 0444, &netDevData{stack: stack}), - "snmp": newDentry(root, inoGen.NextIno(), 0444, &netSnmpData{stack: stack}), + "dev": fs.newDentry(root, fs.NextIno(), 0444, &netDevData{stack: stack}), + "snmp": fs.newDentry(root, fs.NextIno(), 0444, &netSnmpData{stack: stack}), // The following files are simple stubs until they are implemented in // netstack, if the file contains a header the stub is just the header // otherwise it is an empty file. - "arp": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(arp)), - "netlink": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(netlink)), - "netstat": newDentry(root, inoGen.NextIno(), 0444, &netStatData{}), - "packet": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(packet)), - "protocols": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(protocols)), + "arp": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile(arp)), + "netlink": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile(netlink)), + "netstat": fs.newDentry(root, fs.NextIno(), 0444, &netStatData{}), + "packet": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile(packet)), + "protocols": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile(protocols)), // Linux sets psched values to: nsec per usec, psched tick in ns, 1000000, // high res timer ticks per sec (ClockGetres returns 1ns resolution). - "psched": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(psched)), - "ptype": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(ptype)), - "route": newDentry(root, inoGen.NextIno(), 0444, &netRouteData{stack: stack}), - "tcp": newDentry(root, inoGen.NextIno(), 0444, &netTCPData{kernel: k}), - "udp": newDentry(root, inoGen.NextIno(), 0444, &netUDPData{kernel: k}), - "unix": newDentry(root, inoGen.NextIno(), 0444, &netUnixData{kernel: k}), + "psched": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile(psched)), + "ptype": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile(ptype)), + "route": fs.newDentry(root, fs.NextIno(), 0444, &netRouteData{stack: stack}), + "tcp": fs.newDentry(root, fs.NextIno(), 0444, &netTCPData{kernel: k}), + "udp": fs.newDentry(root, fs.NextIno(), 0444, &netUDPData{kernel: k}), + "unix": fs.newDentry(root, fs.NextIno(), 0444, &netUnixData{kernel: k}), } if stack.SupportsIPv6() { - contents["if_inet6"] = newDentry(root, inoGen.NextIno(), 0444, &ifinet6{stack: stack}) - contents["ipv6_route"] = newDentry(root, inoGen.NextIno(), 0444, newStaticFile("")) - contents["tcp6"] = newDentry(root, inoGen.NextIno(), 0444, &netTCP6Data{kernel: k}) - contents["udp6"] = newDentry(root, inoGen.NextIno(), 0444, newStaticFile(upd6)) + contents["if_inet6"] = fs.newDentry(root, fs.NextIno(), 0444, &ifinet6{stack: stack}) + contents["ipv6_route"] = fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("")) + contents["tcp6"] = fs.newDentry(root, fs.NextIno(), 0444, &netTCP6Data{kernel: k}) + contents["udp6"] = fs.newDentry(root, fs.NextIno(), 0444, newStaticFile(upd6)) } } - return newTaskOwnedDir(task, inoGen.NextIno(), 0555, contents) + return fs.newTaskOwnedDir(task, fs.NextIno(), 0555, contents) } // ifinet6 implements vfs.DynamicBytesSource for /proc/net/if_inet6. diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index 5aeda8c9b..b51d43954 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -33,11 +33,6 @@ const ( threadSelfName = "thread-self" ) -// InoGenerator generates unique inode numbers for a given filesystem. -type InoGenerator interface { - NextIno() uint64 -} - // tasksInode represents the inode for /proc/ directory. // // +stateify savable @@ -48,8 +43,8 @@ type tasksInode struct { kernfs.OrderedChildren kernfs.AlwaysValid - inoGen InoGenerator - pidns *kernel.PIDNamespace + fs *filesystem + pidns *kernel.PIDNamespace // '/proc/self' and '/proc/thread-self' have custom directory offsets in // Linux. So handle them outside of OrderedChildren. @@ -64,29 +59,29 @@ type tasksInode struct { var _ kernfs.Inode = (*tasksInode)(nil) -func newTasksInode(inoGen InoGenerator, k *kernel.Kernel, pidns *kernel.PIDNamespace, cgroupControllers map[string]string) (*tasksInode, *kernfs.Dentry) { +func (fs *filesystem) newTasksInode(k *kernel.Kernel, pidns *kernel.PIDNamespace, cgroupControllers map[string]string) (*tasksInode, *kernfs.Dentry) { root := auth.NewRootCredentials(pidns.UserNamespace()) contents := map[string]*kernfs.Dentry{ - "cpuinfo": newDentry(root, inoGen.NextIno(), 0444, newStaticFileSetStat(cpuInfoData(k))), - "filesystems": newDentry(root, inoGen.NextIno(), 0444, &filesystemsData{}), - "loadavg": newDentry(root, inoGen.NextIno(), 0444, &loadavgData{}), - "sys": newSysDir(root, inoGen, k), - "meminfo": newDentry(root, inoGen.NextIno(), 0444, &meminfoData{}), - "mounts": kernfs.NewStaticSymlink(root, inoGen.NextIno(), "self/mounts"), - "net": kernfs.NewStaticSymlink(root, inoGen.NextIno(), "self/net"), - "stat": newDentry(root, inoGen.NextIno(), 0444, &statData{}), - "uptime": newDentry(root, inoGen.NextIno(), 0444, &uptimeData{}), - "version": newDentry(root, inoGen.NextIno(), 0444, &versionData{}), + "cpuinfo": fs.newDentry(root, fs.NextIno(), 0444, newStaticFileSetStat(cpuInfoData(k))), + "filesystems": fs.newDentry(root, fs.NextIno(), 0444, &filesystemsData{}), + "loadavg": fs.newDentry(root, fs.NextIno(), 0444, &loadavgData{}), + "sys": fs.newSysDir(root, k), + "meminfo": fs.newDentry(root, fs.NextIno(), 0444, &meminfoData{}), + "mounts": kernfs.NewStaticSymlink(root, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), "self/mounts"), + "net": kernfs.NewStaticSymlink(root, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), "self/net"), + "stat": fs.newDentry(root, fs.NextIno(), 0444, &statData{}), + "uptime": fs.newDentry(root, fs.NextIno(), 0444, &uptimeData{}), + "version": fs.newDentry(root, fs.NextIno(), 0444, &versionData{}), } inode := &tasksInode{ pidns: pidns, - inoGen: inoGen, - selfSymlink: newSelfSymlink(root, inoGen.NextIno(), pidns).VFSDentry(), - threadSelfSymlink: newThreadSelfSymlink(root, inoGen.NextIno(), pidns).VFSDentry(), + fs: fs, + selfSymlink: fs.newSelfSymlink(root, fs.NextIno(), pidns).VFSDentry(), + threadSelfSymlink: fs.newThreadSelfSymlink(root, fs.NextIno(), pidns).VFSDentry(), cgroupControllers: cgroupControllers, } - inode.InodeAttrs.Init(root, inoGen.NextIno(), linux.ModeDirectory|0555) + inode.InodeAttrs.Init(root, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0555) dentry := &kernfs.Dentry{} dentry.Init(inode) @@ -118,7 +113,7 @@ func (i *tasksInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, erro return nil, syserror.ENOENT } - taskDentry := newTaskInode(i.inoGen, task, i.pidns, true, i.cgroupControllers) + taskDentry := i.fs.newTaskInode(task, i.pidns, true, i.cgroupControllers) return taskDentry.VFSDentry(), nil } @@ -144,7 +139,7 @@ func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback dirent := vfs.Dirent{ Name: selfName, Type: linux.DT_LNK, - Ino: i.inoGen.NextIno(), + Ino: i.fs.NextIno(), NextOff: offset + 1, } if err := cb.Handle(dirent); err != nil { @@ -156,7 +151,7 @@ func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback dirent := vfs.Dirent{ Name: threadSelfName, Type: linux.DT_LNK, - Ino: i.inoGen.NextIno(), + Ino: i.fs.NextIno(), NextOff: offset + 1, } if err := cb.Handle(dirent); err != nil { @@ -189,7 +184,7 @@ func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback dirent := vfs.Dirent{ Name: strconv.FormatUint(uint64(tid), 10), Type: linux.DT_DIR, - Ino: i.inoGen.NextIno(), + Ino: i.fs.NextIno(), NextOff: FIRST_PROCESS_ENTRY + 2 + int64(tid) + 1, } if err := cb.Handle(dirent); err != nil { diff --git a/pkg/sentry/fsimpl/proc/tasks_files.go b/pkg/sentry/fsimpl/proc/tasks_files.go index e5f13b69e..7d8983aa5 100644 --- a/pkg/sentry/fsimpl/proc/tasks_files.go +++ b/pkg/sentry/fsimpl/proc/tasks_files.go @@ -41,9 +41,9 @@ type selfSymlink struct { var _ kernfs.Inode = (*selfSymlink)(nil) -func newSelfSymlink(creds *auth.Credentials, ino uint64, pidns *kernel.PIDNamespace) *kernfs.Dentry { +func (fs *filesystem) newSelfSymlink(creds *auth.Credentials, ino uint64, pidns *kernel.PIDNamespace) *kernfs.Dentry { inode := &selfSymlink{pidns: pidns} - inode.Init(creds, ino, linux.ModeSymlink|0777) + inode.Init(creds, linux.UNNAMED_MAJOR, fs.devMinor, ino, linux.ModeSymlink|0777) d := &kernfs.Dentry{} d.Init(inode) @@ -83,9 +83,9 @@ type threadSelfSymlink struct { var _ kernfs.Inode = (*threadSelfSymlink)(nil) -func newThreadSelfSymlink(creds *auth.Credentials, ino uint64, pidns *kernel.PIDNamespace) *kernfs.Dentry { +func (fs *filesystem) newThreadSelfSymlink(creds *auth.Credentials, ino uint64, pidns *kernel.PIDNamespace) *kernfs.Dentry { inode := &threadSelfSymlink{pidns: pidns} - inode.Init(creds, ino, linux.ModeSymlink|0777) + inode.Init(creds, linux.UNNAMED_MAJOR, fs.devMinor, ino, linux.ModeSymlink|0777) d := &kernfs.Dentry{} d.Init(inode) diff --git a/pkg/sentry/fsimpl/proc/tasks_sys.go b/pkg/sentry/fsimpl/proc/tasks_sys.go index 0e90e02fe..6dac2afa4 100644 --- a/pkg/sentry/fsimpl/proc/tasks_sys.go +++ b/pkg/sentry/fsimpl/proc/tasks_sys.go @@ -30,89 +30,89 @@ import ( ) // newSysDir returns the dentry corresponding to /proc/sys directory. -func newSysDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) *kernfs.Dentry { - return kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "kernel": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "hostname": newDentry(root, inoGen.NextIno(), 0444, &hostnameData{}), - "shmall": newDentry(root, inoGen.NextIno(), 0444, shmData(linux.SHMALL)), - "shmmax": newDentry(root, inoGen.NextIno(), 0444, shmData(linux.SHMMAX)), - "shmmni": newDentry(root, inoGen.NextIno(), 0444, shmData(linux.SHMMNI)), +func (fs *filesystem) newSysDir(root *auth.Credentials, k *kernel.Kernel) *kernfs.Dentry { + return kernfs.NewStaticDir(root, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), 0555, map[string]*kernfs.Dentry{ + "kernel": kernfs.NewStaticDir(root, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), 0555, map[string]*kernfs.Dentry{ + "hostname": fs.newDentry(root, fs.NextIno(), 0444, &hostnameData{}), + "shmall": fs.newDentry(root, fs.NextIno(), 0444, shmData(linux.SHMALL)), + "shmmax": fs.newDentry(root, fs.NextIno(), 0444, shmData(linux.SHMMAX)), + "shmmni": fs.newDentry(root, fs.NextIno(), 0444, shmData(linux.SHMMNI)), }), - "vm": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "mmap_min_addr": newDentry(root, inoGen.NextIno(), 0444, &mmapMinAddrData{k: k}), - "overcommit_memory": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0\n")), + "vm": kernfs.NewStaticDir(root, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), 0555, map[string]*kernfs.Dentry{ + "mmap_min_addr": fs.newDentry(root, fs.NextIno(), 0444, &mmapMinAddrData{k: k}), + "overcommit_memory": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0\n")), }), - "net": newSysNetDir(root, inoGen, k), + "net": fs.newSysNetDir(root, k), }) } // newSysNetDir returns the dentry corresponding to /proc/sys/net directory. -func newSysNetDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) *kernfs.Dentry { +func (fs *filesystem) newSysNetDir(root *auth.Credentials, k *kernel.Kernel) *kernfs.Dentry { var contents map[string]*kernfs.Dentry // TODO(gvisor.dev/issue/1833): Support for using the network stack in the // network namespace of the calling process. if stack := k.RootNetworkNamespace().Stack(); stack != nil { contents = map[string]*kernfs.Dentry{ - "ipv4": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "tcp_sack": newDentry(root, inoGen.NextIno(), 0644, &tcpSackData{stack: stack}), + "ipv4": kernfs.NewStaticDir(root, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), 0555, map[string]*kernfs.Dentry{ + "tcp_sack": fs.newDentry(root, fs.NextIno(), 0644, &tcpSackData{stack: stack}), // The following files are simple stubs until they are implemented in // netstack, most of these files are configuration related. We use the // value closest to the actual netstack behavior or any empty file, all // of these files will have mode 0444 (read-only for all users). - "ip_local_port_range": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("16000 65535")), - "ip_local_reserved_ports": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("")), - "ipfrag_time": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("30")), - "ip_nonlocal_bind": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "ip_no_pmtu_disc": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), + "ip_local_port_range": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("16000 65535")), + "ip_local_reserved_ports": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("")), + "ipfrag_time": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("30")), + "ip_nonlocal_bind": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0")), + "ip_no_pmtu_disc": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("1")), // tcp_allowed_congestion_control tell the user what they are able to // do as an unprivledged process so we leave it empty. - "tcp_allowed_congestion_control": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("")), - "tcp_available_congestion_control": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("reno")), - "tcp_congestion_control": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("reno")), + "tcp_allowed_congestion_control": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("")), + "tcp_available_congestion_control": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("reno")), + "tcp_congestion_control": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("reno")), // Many of the following stub files are features netstack doesn't // support. The unsupported features return "0" to indicate they are // disabled. - "tcp_base_mss": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1280")), - "tcp_dsack": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_early_retrans": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_fack": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_fastopen": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_fastopen_key": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("")), - "tcp_invalid_ratelimit": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_keepalive_intvl": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_keepalive_probes": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_keepalive_time": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("7200")), - "tcp_mtu_probing": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_no_metrics_save": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), - "tcp_probe_interval": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_probe_threshold": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_retries1": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("3")), - "tcp_retries2": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("15")), - "tcp_rfc1337": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), - "tcp_slow_start_after_idle": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), - "tcp_synack_retries": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("5")), - "tcp_syn_retries": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("3")), - "tcp_timestamps": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), + "tcp_base_mss": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("1280")), + "tcp_dsack": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0")), + "tcp_early_retrans": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0")), + "tcp_fack": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0")), + "tcp_fastopen": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0")), + "tcp_fastopen_key": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("")), + "tcp_invalid_ratelimit": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0")), + "tcp_keepalive_intvl": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0")), + "tcp_keepalive_probes": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0")), + "tcp_keepalive_time": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("7200")), + "tcp_mtu_probing": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0")), + "tcp_no_metrics_save": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("1")), + "tcp_probe_interval": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0")), + "tcp_probe_threshold": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0")), + "tcp_retries1": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("3")), + "tcp_retries2": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("15")), + "tcp_rfc1337": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("1")), + "tcp_slow_start_after_idle": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("1")), + "tcp_synack_retries": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("5")), + "tcp_syn_retries": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("3")), + "tcp_timestamps": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("1")), }), - "core": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "default_qdisc": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("pfifo_fast")), - "message_burst": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("10")), - "message_cost": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("5")), - "optmem_max": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "rmem_default": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), - "rmem_max": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), - "somaxconn": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("128")), - "wmem_default": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), - "wmem_max": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), + "core": kernfs.NewStaticDir(root, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), 0555, map[string]*kernfs.Dentry{ + "default_qdisc": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("pfifo_fast")), + "message_burst": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("10")), + "message_cost": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("5")), + "optmem_max": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("0")), + "rmem_default": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("212992")), + "rmem_max": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("212992")), + "somaxconn": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("128")), + "wmem_default": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("212992")), + "wmem_max": fs.newDentry(root, fs.NextIno(), 0444, newStaticFile("212992")), }), } } - return kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, contents) + return kernfs.NewStaticDir(root, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), 0555, contents) } // mmapMinAddrData implements vfs.DynamicBytesSource for diff --git a/pkg/sentry/fsimpl/sockfs/BUILD b/pkg/sentry/fsimpl/sockfs/BUILD index 52084ddb5..9453277b8 100644 --- a/pkg/sentry/fsimpl/sockfs/BUILD +++ b/pkg/sentry/fsimpl/sockfs/BUILD @@ -9,6 +9,7 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/context", + "//pkg/fspath", "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/kernel/auth", "//pkg/sentry/vfs", diff --git a/pkg/sentry/fsimpl/sockfs/sockfs.go b/pkg/sentry/fsimpl/sockfs/sockfs.go index 239a9f4b4..ee0828a15 100644 --- a/pkg/sentry/fsimpl/sockfs/sockfs.go +++ b/pkg/sentry/fsimpl/sockfs/sockfs.go @@ -16,8 +16,11 @@ package sockfs import ( + "fmt" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" @@ -41,19 +44,42 @@ func (filesystemType) Name() string { return "sockfs" } +type filesystem struct { + kernfs.Filesystem + + devMinor uint32 +} + // NewFilesystem sets up and returns a new sockfs filesystem. // // Note that there should only ever be one instance of sockfs.Filesystem, // backing a global socket mount. -func NewFilesystem(vfsObj *vfs.VirtualFilesystem) *vfs.Filesystem { - fs := &kernfs.Filesystem{} - fs.VFSFilesystem().Init(vfsObj, filesystemType{}, fs) - return fs.VFSFilesystem() +func NewFilesystem(vfsObj *vfs.VirtualFilesystem) (*vfs.Filesystem, error) { + devMinor, err := vfsObj.GetAnonBlockDevMinor() + if err != nil { + return nil, err + } + fs := &filesystem{ + devMinor: devMinor, + } + fs.Filesystem.VFSFilesystem().Init(vfsObj, filesystemType{}, fs) + return fs.Filesystem.VFSFilesystem(), nil +} + +// Release implements vfs.FilesystemImpl.Release. +func (fs *filesystem) Release() { + fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) + fs.Filesystem.Release() +} + +// PrependPath implements vfs.FilesystemImpl.PrependPath. +func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { + inode := vd.Dentry().Impl().(*kernfs.Dentry).Inode().(*inode) + b.PrependComponent(fmt.Sprintf("socket:[%d]", inode.InodeAttrs.Ino())) + return vfs.PrependPathSyntheticError{} } // inode implements kernfs.Inode. -// -// TODO(gvisor.dev/issue/1193): Device numbers. type inode struct { kernfs.InodeNotDirectory kernfs.InodeNotSymlink @@ -67,11 +93,15 @@ func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentr } // NewDentry constructs and returns a sockfs dentry. -func NewDentry(creds *auth.Credentials, ino uint64) *vfs.Dentry { +// +// Preconditions: mnt.Filesystem() must have been returned by NewFilesystem(). +func NewDentry(creds *auth.Credentials, mnt *vfs.Mount) *vfs.Dentry { + fs := mnt.Filesystem().Impl().(*filesystem) + // File mode matches net/socket.c:sock_alloc. filemode := linux.FileMode(linux.S_IFSOCK | 0600) i := &inode{} - i.Init(creds, ino, filemode) + i.InodeAttrs.Init(creds, linux.UNNAMED_MAJOR, fs.devMinor, fs.Filesystem.NextIno(), filemode) d := &kernfs.Dentry{} d.Init(i) diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go index 00f7d6214..0af373604 100644 --- a/pkg/sentry/fsimpl/sys/sys.go +++ b/pkg/sentry/fsimpl/sys/sys.go @@ -37,6 +37,8 @@ type FilesystemType struct{} // filesystem implements vfs.FilesystemImpl. type filesystem struct { kernfs.Filesystem + + devMinor uint32 } // Name implements vfs.FilesystemType.Name. @@ -46,7 +48,14 @@ func (FilesystemType) Name() string { // GetFilesystem implements vfs.FilesystemType.GetFilesystem. func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { - fs := &filesystem{} + devMinor, err := vfsObj.GetAnonBlockDevMinor() + if err != nil { + return nil, nil, err + } + + fs := &filesystem{ + devMinor: devMinor, + } fs.VFSFilesystem().Init(vfsObj, &fsType, fs) k := kernel.KernelFromContext(ctx) maxCPUCores := k.ApplicationCores() @@ -77,6 +86,12 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt return fs.VFSFilesystem(), root.VFSDentry(), nil } +// Release implements vfs.FilesystemImpl.Release. +func (fs *filesystem) Release() { + fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) + fs.Filesystem.Release() +} + // dir implements kernfs.Inode. type dir struct { kernfs.InodeAttrs @@ -90,7 +105,7 @@ type dir struct { func (fs *filesystem) newDir(creds *auth.Credentials, mode linux.FileMode, contents map[string]*kernfs.Dentry) *kernfs.Dentry { d := &dir{} - d.InodeAttrs.Init(creds, fs.NextIno(), linux.ModeDirectory|0755) + d.InodeAttrs.Init(creds, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0755) d.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) d.dentry.Init(d) @@ -127,7 +142,7 @@ func (c *cpuFile) Generate(ctx context.Context, buf *bytes.Buffer) error { func (fs *filesystem) newCPUFile(creds *auth.Credentials, maxCores uint, mode linux.FileMode) *kernfs.Dentry { c := &cpuFile{maxCores: maxCores} - c.DynamicBytesFile.Init(creds, fs.NextIno(), c, mode) + c.DynamicBytesFile.Init(creds, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), c, mode) d := &kernfs.Dentry{} d.Init(c) return d diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index efc931468..405928bd0 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -63,6 +63,9 @@ type filesystem struct { // clock is a realtime clock used to set timestamps in file operations. clock time.Clock + // devMinor is the filesystem's minor device number. devMinor is immutable. + devMinor uint32 + // mu serializes changes to the Dentry tree. mu sync.RWMutex @@ -96,11 +99,6 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt if memFileProvider == nil { panic("MemoryFileProviderFromContext returned nil") } - clock := time.RealtimeClockFromContext(ctx) - fs := filesystem{ - memFile: memFileProvider.MemoryFile(), - clock: clock, - } rootFileType := uint16(linux.S_IFDIR) newFSType := vfs.FilesystemType(&fstype) @@ -114,6 +112,16 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt } } + devMinor, err := vfsObj.GetAnonBlockDevMinor() + if err != nil { + return nil, nil, err + } + clock := time.RealtimeClockFromContext(ctx) + fs := filesystem{ + memFile: memFileProvider.MemoryFile(), + clock: clock, + devMinor: devMinor, + } fs.vfsfs.Init(vfsObj, newFSType, &fs) var root *dentry @@ -125,6 +133,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt case linux.S_IFDIR: root = &fs.newDirectory(creds, 01777).dentry default: + fs.vfsfs.DecRef() return nil, nil, fmt.Errorf("invalid tmpfs root file type: %#o", rootFileType) } return &fs.vfsfs, &root.vfsd, nil @@ -132,6 +141,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt // Release implements vfs.FilesystemImpl.Release. func (fs *filesystem) Release() { + fs.vfsfs.VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) } // dentry implements vfs.DentryImpl. @@ -188,8 +198,8 @@ func (d *dentry) DecRef() { // inode represents a filesystem object. type inode struct { - // clock is a realtime clock used to set timestamps in file operations. - clock time.Clock + // fs is the owning filesystem. fs is immutable. + fs *filesystem // refs is a reference count. refs is accessed using atomic memory // operations. @@ -232,7 +242,7 @@ func (i *inode) init(impl interface{}, fs *filesystem, creds *auth.Credentials, if mode.FileType() == 0 { panic("file type is required in FileMode") } - i.clock = fs.clock + i.fs = fs i.refs = 1 i.mode = uint32(mode) i.uid = uint32(creds.EffectiveKUID) @@ -327,7 +337,8 @@ func (i *inode) statTo(stat *linux.Statx) { stat.Atime = linux.NsecToStatxTimestamp(i.atime) stat.Ctime = linux.NsecToStatxTimestamp(i.ctime) stat.Mtime = linux.NsecToStatxTimestamp(i.mtime) - // TODO(gvisor.dev/issue/1197): Device number. + stat.DevMajor = linux.UNNAMED_MAJOR + stat.DevMinor = i.fs.devMinor switch impl := i.impl.(type) { case *regularFile: stat.Mask |= linux.STATX_SIZE | linux.STATX_BLOCKS @@ -401,7 +412,7 @@ func (i *inode) setStat(ctx context.Context, creds *auth.Credentials, stat *linu return syserror.EINVAL } } - now := i.clock.Now().Nanoseconds() + now := i.fs.clock.Now().Nanoseconds() if mask&linux.STATX_ATIME != 0 { if stat.Atime.Nsec == linux.UTIME_NOW { atomic.StoreInt64(&i.atime, now) @@ -518,7 +529,7 @@ func (i *inode) touchAtime(mnt *vfs.Mount) { if err := mnt.CheckBeginWrite(); err != nil { return } - now := i.clock.Now().Nanoseconds() + now := i.fs.clock.Now().Nanoseconds() i.mu.Lock() atomic.StoreInt64(&i.atime, now) i.mu.Unlock() @@ -527,7 +538,7 @@ func (i *inode) touchAtime(mnt *vfs.Mount) { // Preconditions: The caller has called vfs.Mount.CheckBeginWrite(). func (i *inode) touchCtime() { - now := i.clock.Now().Nanoseconds() + now := i.fs.clock.Now().Nanoseconds() i.mu.Lock() atomic.StoreInt64(&i.ctime, now) i.mu.Unlock() @@ -535,7 +546,7 @@ func (i *inode) touchCtime() { // Preconditions: The caller has called vfs.Mount.CheckBeginWrite(). func (i *inode) touchCMtime() { - now := i.clock.Now().Nanoseconds() + now := i.fs.clock.Now().Nanoseconds() i.mu.Lock() atomic.StoreInt64(&i.mtime, now) atomic.StoreInt64(&i.ctime, now) @@ -545,7 +556,7 @@ func (i *inode) touchCMtime() { // Preconditions: The caller has called vfs.Mount.CheckBeginWrite() and holds // inode.mu. func (i *inode) touchCMtimeLocked() { - now := i.clock.Now().Nanoseconds() + now := i.fs.clock.Now().Nanoseconds() atomic.StoreInt64(&i.mtime, now) atomic.StoreInt64(&i.ctime, now) } diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 271ea5faf..3617da8c6 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -373,7 +373,10 @@ func (k *Kernel) Init(args InitKernelArgs) error { return fmt.Errorf("failed to initialize VFS: %v", err) } - pipeFilesystem := pipefs.NewFilesystem(&k.vfs) + pipeFilesystem, err := pipefs.NewFilesystem(&k.vfs) + if err != nil { + return fmt.Errorf("failed to create pipefs filesystem: %v", err) + } defer pipeFilesystem.DecRef() pipeMount, err := k.vfs.NewDisconnectedMount(pipeFilesystem, nil, &vfs.MountOptions{}) if err != nil { @@ -381,7 +384,10 @@ func (k *Kernel) Init(args InitKernelArgs) error { } k.pipeMount = pipeMount - socketFilesystem := sockfs.NewFilesystem(&k.vfs) + socketFilesystem, err := sockfs.NewFilesystem(&k.vfs) + if err != nil { + return fmt.Errorf("failed to create sockfs filesystem: %v", err) + } defer socketFilesystem.DecRef() socketMount, err := k.vfs.NewDisconnectedMount(socketFilesystem, nil, &vfs.MountOptions{}) if err != nil { diff --git a/pkg/sentry/socket/hostinet/BUILD b/pkg/sentry/socket/hostinet/BUILD index deedd35f7..e82d6cd1e 100644 --- a/pkg/sentry/socket/hostinet/BUILD +++ b/pkg/sentry/socket/hostinet/BUILD @@ -26,7 +26,6 @@ go_library( "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", - "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/fsimpl/sockfs", "//pkg/sentry/hostfd", "//pkg/sentry/inet", diff --git a/pkg/sentry/socket/hostinet/socket_vfs2.go b/pkg/sentry/socket/hostinet/socket_vfs2.go index a8278bffc..677743113 100644 --- a/pkg/sentry/socket/hostinet/socket_vfs2.go +++ b/pkg/sentry/socket/hostinet/socket_vfs2.go @@ -21,7 +21,6 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fdnotifier" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" "gvisor.dev/gvisor/pkg/sentry/hostfd" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -50,8 +49,7 @@ var _ = socket.SocketVFS2(&socketVFS2{}) func newVFS2Socket(t *kernel.Task, family int, stype linux.SockType, protocol int, fd int, flags uint32) (*vfs.FileDescription, *syserr.Error) { mnt := t.Kernel().SocketMount() - fs := mnt.Filesystem().Impl().(*kernfs.Filesystem) - d := sockfs.NewDentry(t.Credentials(), fs.NextIno()) + d := sockfs.NewDentry(t.Credentials(), mnt) s := &socketVFS2{ socketOpsCommon: socketOpsCommon{ diff --git a/pkg/sentry/socket/netlink/BUILD b/pkg/sentry/socket/netlink/BUILD index 09ca00a4a..7212d8644 100644 --- a/pkg/sentry/socket/netlink/BUILD +++ b/pkg/sentry/socket/netlink/BUILD @@ -20,7 +20,6 @@ go_library( "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", - "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/fsimpl/sockfs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", diff --git a/pkg/sentry/socket/netlink/provider_vfs2.go b/pkg/sentry/socket/netlink/provider_vfs2.go index dcd92b5cd..bb205be0d 100644 --- a/pkg/sentry/socket/netlink/provider_vfs2.go +++ b/pkg/sentry/socket/netlink/provider_vfs2.go @@ -16,7 +16,6 @@ package netlink import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/vfs" @@ -52,8 +51,7 @@ func (*socketProviderVFS2) Socket(t *kernel.Task, stype linux.SockType, protocol vfsfd := &s.vfsfd mnt := t.Kernel().SocketMount() - fs := mnt.Filesystem().Impl().(*kernfs.Filesystem) - d := sockfs.NewDentry(t.Credentials(), fs.NextIno()) + d := sockfs.NewDentry(t.Credentials(), mnt) if err := vfsfd.Init(s, linux.O_RDWR, mnt, d, &vfs.FileDescriptionOptions{ DenyPRead: true, DenyPWrite: true, diff --git a/pkg/sentry/socket/netstack/BUILD b/pkg/sentry/socket/netstack/BUILD index ccf9fcf5c..6129fb83d 100644 --- a/pkg/sentry/socket/netstack/BUILD +++ b/pkg/sentry/socket/netstack/BUILD @@ -27,7 +27,6 @@ go_library( "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", - "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/fsimpl/sockfs", "//pkg/sentry/inet", "//pkg/sentry/kernel", diff --git a/pkg/sentry/socket/netstack/netstack_vfs2.go b/pkg/sentry/socket/netstack/netstack_vfs2.go index f7d9b2ff4..191970d41 100644 --- a/pkg/sentry/socket/netstack/netstack_vfs2.go +++ b/pkg/sentry/socket/netstack/netstack_vfs2.go @@ -18,7 +18,6 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -53,8 +52,7 @@ func NewVFS2(t *kernel.Task, family int, skType linux.SockType, protocol int, qu } mnt := t.Kernel().SocketMount() - fs := mnt.Filesystem().Impl().(*kernfs.Filesystem) - d := sockfs.NewDentry(t.Credentials(), fs.NextIno()) + d := sockfs.NewDentry(t.Credentials(), mnt) s := &SocketVFS2{ socketOpsCommon: socketOpsCommon{ diff --git a/pkg/sentry/socket/unix/BUILD b/pkg/sentry/socket/unix/BUILD index 941a91097..de2cc4bdf 100644 --- a/pkg/sentry/socket/unix/BUILD +++ b/pkg/sentry/socket/unix/BUILD @@ -21,7 +21,6 @@ go_library( "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", - "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/fsimpl/sockfs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/time", diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go index 06d838868..45e109361 100644 --- a/pkg/sentry/socket/unix/unix_vfs2.go +++ b/pkg/sentry/socket/unix/unix_vfs2.go @@ -19,7 +19,6 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/socket" @@ -50,8 +49,7 @@ var _ = socket.SocketVFS2(&SocketVFS2{}) // returns a corresponding file description. func NewSockfsFile(t *kernel.Task, ep transport.Endpoint, stype linux.SockType) (*vfs.FileDescription, *syserr.Error) { mnt := t.Kernel().SocketMount() - fs := mnt.Filesystem().Impl().(*kernfs.Filesystem) - d := sockfs.NewDentry(t.Credentials(), fs.NextIno()) + d := sockfs.NewDentry(t.Credentials(), mnt) fd, err := NewFileDescription(ep, stype, linux.O_RDWR, mnt, d) if err != nil { diff --git a/pkg/sentry/vfs/anonfs.go b/pkg/sentry/vfs/anonfs.go index adebaeefb..caf770fd5 100644 --- a/pkg/sentry/vfs/anonfs.go +++ b/pkg/sentry/vfs/anonfs.go @@ -202,7 +202,7 @@ func (fs *anonFilesystem) StatAt(ctx context.Context, rp *ResolvingPath, opts St Ino: 1, Size: 0, Blocks: 0, - DevMajor: 0, + DevMajor: linux.UNNAMED_MAJOR, DevMinor: fs.devMinor, }, nil } diff --git a/pkg/sentry/vfs/device.go b/pkg/sentry/vfs/device.go index bda5576fa..1e9dffc8f 100644 --- a/pkg/sentry/vfs/device.go +++ b/pkg/sentry/vfs/device.go @@ -103,7 +103,7 @@ func (vfs *VirtualFilesystem) OpenDeviceSpecialFile(ctx context.Context, mnt *Mo } // GetAnonBlockDevMinor allocates and returns an unused minor device number for -// an "anonymous" block device with major number 0. +// an "anonymous" block device with major number UNNAMED_MAJOR. func (vfs *VirtualFilesystem) GetAnonBlockDevMinor() (uint32, error) { vfs.anonBlockDevMinorMu.Lock() defer vfs.anonBlockDevMinorMu.Unlock() diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index 8c8bad11c..f802bc9fb 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -334,7 +334,10 @@ func New(args Args) (*Loader, error) { if kernel.VFS2Enabled { // Set up host mount that will be used for imported fds. - hostFilesystem := hostvfs2.NewFilesystem(k.VFS()) + hostFilesystem, err := hostvfs2.NewFilesystem(k.VFS()) + if err != nil { + return nil, fmt.Errorf("failed to create hostfs filesystem: %v", err) + } defer hostFilesystem.DecRef() hostMount, err := k.VFS().NewDisconnectedMount(hostFilesystem, nil, &vfs.MountOptions{}) if err != nil { -- cgit v1.2.3 From 7b4a913f36e2fededca9b1d2f73588eca9c4b683 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Thu, 7 May 2020 15:17:09 -0700 Subject: Fix ARM64 build. The common syscall definitions mean that ARM64-exclusive files need stubs in the ARM64 build. PiperOrigin-RevId: 310446698 --- pkg/sentry/syscalls/linux/BUILD | 3 +- pkg/sentry/syscalls/linux/sys_tls.go | 52 ------------------------------ pkg/sentry/syscalls/linux/sys_tls_amd64.go | 52 ++++++++++++++++++++++++++++++ pkg/sentry/syscalls/linux/sys_tls_arm64.go | 28 ++++++++++++++++ pkg/sentry/syscalls/linux/sys_utsname.go | 2 -- 5 files changed, 82 insertions(+), 55 deletions(-) delete mode 100644 pkg/sentry/syscalls/linux/sys_tls.go create mode 100644 pkg/sentry/syscalls/linux/sys_tls_amd64.go create mode 100644 pkg/sentry/syscalls/linux/sys_tls_arm64.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD index 245e8fe1e..217fcfef2 100644 --- a/pkg/sentry/syscalls/linux/BUILD +++ b/pkg/sentry/syscalls/linux/BUILD @@ -49,7 +49,8 @@ go_library( "sys_time.go", "sys_timer.go", "sys_timerfd.go", - "sys_tls.go", + "sys_tls_amd64.go", + "sys_tls_arm64.go", "sys_utsname.go", "sys_write.go", "sys_xattr.go", diff --git a/pkg/sentry/syscalls/linux/sys_tls.go b/pkg/sentry/syscalls/linux/sys_tls.go deleted file mode 100644 index b3eb96a1c..000000000 --- a/pkg/sentry/syscalls/linux/sys_tls.go +++ /dev/null @@ -1,52 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -//+build amd64 - -package linux - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/syserror" -) - -// ArchPrctl implements linux syscall arch_prctl(2). -// It sets architecture-specific process or thread state for t. -func ArchPrctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - switch args[0].Int() { - case linux.ARCH_GET_FS: - addr := args[1].Pointer() - fsbase := t.Arch().TLS() - _, err := t.CopyOut(addr, uint64(fsbase)) - if err != nil { - return 0, nil, err - } - - case linux.ARCH_SET_FS: - fsbase := args[1].Uint64() - if !t.Arch().SetTLS(uintptr(fsbase)) { - return 0, nil, syserror.EPERM - } - - case linux.ARCH_GET_GS, linux.ARCH_SET_GS: - t.Kernel().EmitUnimplementedEvent(t) - fallthrough - default: - return 0, nil, syserror.EINVAL - } - - return 0, nil, nil -} diff --git a/pkg/sentry/syscalls/linux/sys_tls_amd64.go b/pkg/sentry/syscalls/linux/sys_tls_amd64.go new file mode 100644 index 000000000..b3eb96a1c --- /dev/null +++ b/pkg/sentry/syscalls/linux/sys_tls_amd64.go @@ -0,0 +1,52 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +//+build amd64 + +package linux + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/syserror" +) + +// ArchPrctl implements linux syscall arch_prctl(2). +// It sets architecture-specific process or thread state for t. +func ArchPrctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + switch args[0].Int() { + case linux.ARCH_GET_FS: + addr := args[1].Pointer() + fsbase := t.Arch().TLS() + _, err := t.CopyOut(addr, uint64(fsbase)) + if err != nil { + return 0, nil, err + } + + case linux.ARCH_SET_FS: + fsbase := args[1].Uint64() + if !t.Arch().SetTLS(uintptr(fsbase)) { + return 0, nil, syserror.EPERM + } + + case linux.ARCH_GET_GS, linux.ARCH_SET_GS: + t.Kernel().EmitUnimplementedEvent(t) + fallthrough + default: + return 0, nil, syserror.EINVAL + } + + return 0, nil, nil +} diff --git a/pkg/sentry/syscalls/linux/sys_tls_arm64.go b/pkg/sentry/syscalls/linux/sys_tls_arm64.go new file mode 100644 index 000000000..fb08a356e --- /dev/null +++ b/pkg/sentry/syscalls/linux/sys_tls_arm64.go @@ -0,0 +1,28 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +//+build arm64 + +package linux + +import ( + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/syserror" +) + +// ArchPrctl is not defined for ARM64. +func ArchPrctl(*kernel.Task, arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + return 0, nil, syserror.ENOSYS +} diff --git a/pkg/sentry/syscalls/linux/sys_utsname.go b/pkg/sentry/syscalls/linux/sys_utsname.go index a393e28c1..e9d702e8e 100644 --- a/pkg/sentry/syscalls/linux/sys_utsname.go +++ b/pkg/sentry/syscalls/linux/sys_utsname.go @@ -12,8 +12,6 @@ // See the License for the specific language governing permissions and // limitations under the License. -// +build amd64 arm64 - package linux import ( -- cgit v1.2.3 From 21b71395a6aa2eafbc4c59222574d56c2db2e23b Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Fri, 8 May 2020 11:34:15 -0700 Subject: Pass flags to fsimpl/host.inode.open(). This has two effects: It makes flags passed to open("/proc/[pid]/fd/[hostfd]") effective, and it prevents imported pipes/sockets/character devices from being opened with O_NONBLOCK unconditionally (because the underlying host FD was set to non-blocking in ImportFD()). PiperOrigin-RevId: 310596062 --- pkg/sentry/fsimpl/host/host.go | 35 ++++++++++++++--------------------- 1 file changed, 14 insertions(+), 21 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 55de9c438..36bceeaa4 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -53,13 +53,19 @@ func ImportFD(ctx context.Context, mnt *vfs.Mount, hostFD int, isTTY bool) (*vfs return nil, err } + // Get flags for the imported FD. + flags, err := unix.FcntlInt(uintptr(hostFD), syscall.F_GETFL, 0) + if err != nil { + return nil, err + } + fileMode := linux.FileMode(s.Mode) fileType := fileMode.FileType() // Determine if hostFD is seekable. If not, this syscall will return ESPIPE // (see fs/read_write.c:llseek), e.g. for pipes, sockets, and some character // devices. - _, err := unix.Seek(hostFD, 0, linux.SEEK_CUR) + _, err = unix.Seek(hostFD, 0, linux.SEEK_CUR) seekable := err != syserror.ESPIPE i := &inode{ @@ -95,7 +101,7 @@ func ImportFD(ctx context.Context, mnt *vfs.Mount, hostFD int, isTTY bool) (*vfs // i.open will take a reference on d. defer d.DecRef() - return i.open(ctx, d.VFSDentry(), mnt) + return i.open(ctx, d.VFSDentry(), mnt, uint32(flags)) } // filesystemType implements vfs.FilesystemType. @@ -198,19 +204,6 @@ type inode struct { queue waiter.Queue } -// Note that these flags may become out of date, since they can be modified -// on the host, e.g. with fcntl. -func fileFlagsFromHostFD(fd int) (uint32, error) { - flags, err := unix.FcntlInt(uintptr(fd), syscall.F_GETFL, 0) - if err != nil { - log.Warningf("Failed to get file flags for donated FD %d: %v", fd, err) - return 0, err - } - // TODO(gvisor.dev/issue/1672): implement behavior corresponding to these allowed flags. - flags &= syscall.O_ACCMODE | syscall.O_DIRECT | syscall.O_NONBLOCK | syscall.O_DSYNC | syscall.O_SYNC | syscall.O_APPEND - return uint32(flags), nil -} - // CheckPermissions implements kernfs.Inode. func (i *inode) CheckPermissions(ctx context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error { var s syscall.Stat_t @@ -406,19 +399,19 @@ func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentr if i.Mode().FileType() == linux.S_IFSOCK { return nil, syserror.ENXIO } - return i.open(ctx, vfsd, rp.Mount()) + return i.open(ctx, vfsd, rp.Mount(), opts.Flags) } -func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount) (*vfs.FileDescription, error) { +func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount, flags uint32) (*vfs.FileDescription, error) { var s syscall.Stat_t if err := syscall.Fstat(i.hostFD, &s); err != nil { return nil, err } fileType := s.Mode & linux.FileTypeMask - flags, err := fileFlagsFromHostFD(i.hostFD) - if err != nil { - return nil, err - } + + // Constrain flags to a subset we can handle. + // TODO(gvisor.dev/issue/1672): implement behavior corresponding to these allowed flags. + flags &= syscall.O_ACCMODE | syscall.O_DIRECT | syscall.O_NONBLOCK | syscall.O_DSYNC | syscall.O_SYNC | syscall.O_APPEND if fileType == syscall.S_IFSOCK { if i.isTTY { -- cgit v1.2.3 From cfd30665c1d857f20dd05e67c6da6833770e2141 Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Fri, 8 May 2020 15:39:04 -0700 Subject: iptables - filter packets using outgoing interface. Enables commands with -o (--out-interface) for iptables rules. $ iptables -A OUTPUT -o eth0 -j ACCEPT PiperOrigin-RevId: 310642286 --- pkg/abi/linux/netfilter.go | 2 +- pkg/sentry/socket/netfilter/netfilter.go | 40 ++++++-- pkg/tcpip/network/ipv4/ipv4.go | 8 +- pkg/tcpip/stack/iptables.go | 42 ++++++-- pkg/tcpip/stack/iptables_types.go | 13 +++ pkg/tcpip/stack/nic.go | 2 +- pkg/tcpip/stack/stack.go | 16 ++- test/iptables/filter_output.go | 170 +++++++++++++++++++++++++++++++ test/iptables/iptables_test.go | 24 +++++ test/iptables/iptables_util.go | 15 +++ 10 files changed, 309 insertions(+), 23 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/netfilter.go b/pkg/abi/linux/netfilter.go index a8d4f9d69..46d8b0b42 100644 --- a/pkg/abi/linux/netfilter.go +++ b/pkg/abi/linux/netfilter.go @@ -146,7 +146,7 @@ type IPTIP struct { // OutputInterface is the output network interface. OutputInterface [IFNAMSIZ]byte - // InputInterfaceMask is the intput interface mask. + // InputInterfaceMask is the input interface mask. InputInterfaceMask [IFNAMSIZ]byte // OuputInterfaceMask is the output interface mask. diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 40736fb38..41a1ce031 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -17,6 +17,7 @@ package netfilter import ( + "bytes" "errors" "fmt" @@ -204,6 +205,16 @@ func convertNetstackToBinary(tablename string, table stack.Table) (linux.KernelI TargetOffset: linux.SizeOfIPTEntry, }, } + copy(entry.IPTEntry.IP.Dst[:], rule.Filter.Dst) + copy(entry.IPTEntry.IP.DstMask[:], rule.Filter.DstMask) + copy(entry.IPTEntry.IP.OutputInterface[:], rule.Filter.OutputInterface) + copy(entry.IPTEntry.IP.OutputInterfaceMask[:], rule.Filter.OutputInterfaceMask) + if rule.Filter.DstInvert { + entry.IPTEntry.IP.InverseFlags |= linux.IPT_INV_DSTIP + } + if rule.Filter.OutputInterfaceInvert { + entry.IPTEntry.IP.InverseFlags |= linux.IPT_INV_VIA_OUT + } for _, matcher := range rule.Matchers { // Serialize the matcher and add it to the @@ -719,11 +730,27 @@ func filterFromIPTIP(iptip linux.IPTIP) (stack.IPHeaderFilter, error) { if len(iptip.Dst) != header.IPv4AddressSize || len(iptip.DstMask) != header.IPv4AddressSize { return stack.IPHeaderFilter{}, fmt.Errorf("incorrect length of destination (%d) and/or destination mask (%d) fields", len(iptip.Dst), len(iptip.DstMask)) } + + n := bytes.IndexByte([]byte(iptip.OutputInterface[:]), 0) + if n == -1 { + n = len(iptip.OutputInterface) + } + ifname := string(iptip.OutputInterface[:n]) + + n = bytes.IndexByte([]byte(iptip.OutputInterfaceMask[:]), 0) + if n == -1 { + n = len(iptip.OutputInterfaceMask) + } + ifnameMask := string(iptip.OutputInterfaceMask[:n]) + return stack.IPHeaderFilter{ - Protocol: tcpip.TransportProtocolNumber(iptip.Protocol), - Dst: tcpip.Address(iptip.Dst[:]), - DstMask: tcpip.Address(iptip.DstMask[:]), - DstInvert: iptip.InverseFlags&linux.IPT_INV_DSTIP != 0, + Protocol: tcpip.TransportProtocolNumber(iptip.Protocol), + Dst: tcpip.Address(iptip.Dst[:]), + DstMask: tcpip.Address(iptip.DstMask[:]), + DstInvert: iptip.InverseFlags&linux.IPT_INV_DSTIP != 0, + OutputInterface: ifname, + OutputInterfaceMask: ifnameMask, + OutputInterfaceInvert: iptip.InverseFlags&linux.IPT_INV_VIA_OUT != 0, }, nil } @@ -732,16 +759,15 @@ func containsUnsupportedFields(iptip linux.IPTIP) bool { // - Protocol // - Dst and DstMask // - The inverse destination IP check flag + // - OutputInterface, OutputInterfaceMask and its inverse. var emptyInetAddr = linux.InetAddr{} var emptyInterface = [linux.IFNAMSIZ]byte{} // Disable any supported inverse flags. - inverseMask := uint8(linux.IPT_INV_DSTIP) + inverseMask := uint8(linux.IPT_INV_DSTIP) | uint8(linux.IPT_INV_VIA_OUT) return iptip.Src != emptyInetAddr || iptip.SrcMask != emptyInetAddr || iptip.InputInterface != emptyInterface || - iptip.OutputInterface != emptyInterface || iptip.InputInterfaceMask != emptyInterface || - iptip.OutputInterfaceMask != emptyInterface || iptip.Flags != 0 || iptip.InverseFlags&^inverseMask != 0 } diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index 9db42b2a4..64046cbbf 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -249,10 +249,11 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw ip := e.addIPHeader(r, &pkt.Header, pkt.Data.Size(), params) pkt.NetworkHeader = buffer.View(ip) + nicName := e.stack.FindNICNameFromID(e.NICID()) // iptables filtering. All packets that reach here are locally // generated. ipt := e.stack.IPTables() - if ok := ipt.Check(stack.Output, &pkt, gso, r, ""); !ok { + if ok := ipt.Check(stack.Output, &pkt, gso, r, "", nicName); !ok { // iptables is telling us to drop the packet. return nil } @@ -319,10 +320,11 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe pkt = pkt.Next() } + nicName := e.stack.FindNICNameFromID(e.NICID()) // iptables filtering. All packets that reach here are locally // generated. ipt := e.stack.IPTables() - dropped, natPkts := ipt.CheckPackets(stack.Output, pkts, gso, r) + dropped, natPkts := ipt.CheckPackets(stack.Output, pkts, gso, r, nicName) if len(dropped) == 0 && len(natPkts) == 0 { // Fast path: If no packets are to be dropped then we can just invoke the // faster WritePackets API directly. @@ -445,7 +447,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { // iptables filtering. All packets that reach here are intended for // this machine and will not be forwarded. ipt := e.stack.IPTables() - if ok := ipt.Check(stack.Input, &pkt, nil, nil, ""); !ok { + if ok := ipt.Check(stack.Input, &pkt, nil, nil, "", ""); !ok { // iptables is telling us to drop the packet. return } diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go index 7c3c47d50..443423b3c 100644 --- a/pkg/tcpip/stack/iptables.go +++ b/pkg/tcpip/stack/iptables.go @@ -16,6 +16,7 @@ package stack import ( "fmt" + "strings" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/header" @@ -178,7 +179,7 @@ const ( // dropped. // // Precondition: pkt.NetworkHeader is set. -func (it *IPTables) Check(hook Hook, pkt *PacketBuffer, gso *GSO, r *Route, address tcpip.Address) bool { +func (it *IPTables) Check(hook Hook, pkt *PacketBuffer, gso *GSO, r *Route, address tcpip.Address, nicName string) bool { // Packets are manipulated only if connection and matching // NAT rule exists. it.connections.HandlePacket(pkt, hook, gso, r) @@ -187,7 +188,7 @@ func (it *IPTables) Check(hook Hook, pkt *PacketBuffer, gso *GSO, r *Route, addr for _, tablename := range it.Priorities[hook] { table := it.Tables[tablename] ruleIdx := table.BuiltinChains[hook] - switch verdict := it.checkChain(hook, pkt, table, ruleIdx, gso, r, address); verdict { + switch verdict := it.checkChain(hook, pkt, table, ruleIdx, gso, r, address, nicName); verdict { // If the table returns Accept, move on to the next table. case chainAccept: continue @@ -228,10 +229,10 @@ func (it *IPTables) Check(hook Hook, pkt *PacketBuffer, gso *GSO, r *Route, addr // // NOTE: unlike the Check API the returned map contains packets that should be // dropped. -func (it *IPTables) CheckPackets(hook Hook, pkts PacketBufferList, gso *GSO, r *Route) (drop map[*PacketBuffer]struct{}, natPkts map[*PacketBuffer]struct{}) { +func (it *IPTables) CheckPackets(hook Hook, pkts PacketBufferList, gso *GSO, r *Route, nicName string) (drop map[*PacketBuffer]struct{}, natPkts map[*PacketBuffer]struct{}) { for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() { if !pkt.NatDone { - if ok := it.Check(hook, pkt, gso, r, ""); !ok { + if ok := it.Check(hook, pkt, gso, r, "", nicName); !ok { if drop == nil { drop = make(map[*PacketBuffer]struct{}) } @@ -251,11 +252,11 @@ func (it *IPTables) CheckPackets(hook Hook, pkts PacketBufferList, gso *GSO, r * // Precondition: pkt is a IPv4 packet of at least length header.IPv4MinimumSize. // TODO(gvisor.dev/issue/170): pkt.NetworkHeader will always be set as a // precondition. -func (it *IPTables) checkChain(hook Hook, pkt *PacketBuffer, table Table, ruleIdx int, gso *GSO, r *Route, address tcpip.Address) chainVerdict { +func (it *IPTables) checkChain(hook Hook, pkt *PacketBuffer, table Table, ruleIdx int, gso *GSO, r *Route, address tcpip.Address, nicName string) chainVerdict { // Start from ruleIdx and walk the list of rules until a rule gives us // a verdict. for ruleIdx < len(table.Rules) { - switch verdict, jumpTo := it.checkRule(hook, pkt, table, ruleIdx, gso, r, address); verdict { + switch verdict, jumpTo := it.checkRule(hook, pkt, table, ruleIdx, gso, r, address, nicName); verdict { case RuleAccept: return chainAccept @@ -272,7 +273,7 @@ func (it *IPTables) checkChain(hook Hook, pkt *PacketBuffer, table Table, ruleId ruleIdx++ continue } - switch verdict := it.checkChain(hook, pkt, table, jumpTo, gso, r, address); verdict { + switch verdict := it.checkChain(hook, pkt, table, jumpTo, gso, r, address, nicName); verdict { case chainAccept: return chainAccept case chainDrop: @@ -298,7 +299,7 @@ func (it *IPTables) checkChain(hook Hook, pkt *PacketBuffer, table Table, ruleId // Precondition: pkt is a IPv4 packet of at least length header.IPv4MinimumSize. // TODO(gvisor.dev/issue/170): pkt.NetworkHeader will always be set as a // precondition. -func (it *IPTables) checkRule(hook Hook, pkt *PacketBuffer, table Table, ruleIdx int, gso *GSO, r *Route, address tcpip.Address) (RuleVerdict, int) { +func (it *IPTables) checkRule(hook Hook, pkt *PacketBuffer, table Table, ruleIdx int, gso *GSO, r *Route, address tcpip.Address, nicName string) (RuleVerdict, int) { rule := table.Rules[ruleIdx] // If pkt.NetworkHeader hasn't been set yet, it will be contained in @@ -313,7 +314,7 @@ func (it *IPTables) checkRule(hook Hook, pkt *PacketBuffer, table Table, ruleIdx } // Check whether the packet matches the IP header filter. - if !filterMatch(rule.Filter, header.IPv4(pkt.NetworkHeader)) { + if !filterMatch(rule.Filter, header.IPv4(pkt.NetworkHeader), hook, nicName) { // Continue on to the next rule. return RuleJump, ruleIdx + 1 } @@ -335,7 +336,7 @@ func (it *IPTables) checkRule(hook Hook, pkt *PacketBuffer, table Table, ruleIdx return rule.Target.Action(pkt, &it.connections, hook, gso, r, address) } -func filterMatch(filter IPHeaderFilter, hdr header.IPv4) bool { +func filterMatch(filter IPHeaderFilter, hdr header.IPv4, hook Hook, nicName string) bool { // TODO(gvisor.dev/issue/170): Support other fields of the filter. // Check the transport protocol. if filter.Protocol != 0 && filter.Protocol != hdr.TransportProtocol() { @@ -355,5 +356,26 @@ func filterMatch(filter IPHeaderFilter, hdr header.IPv4) bool { return false } + // Check the output interface. + // TODO(gvisor.dev/issue/170): Add the check for FORWARD and POSTROUTING + // hooks after supported. + if hook == Output { + n := len(filter.OutputInterface) + if n == 0 { + return true + } + + // If the interface name ends with '+', any interface which begins + // with the name should be matched. + ifName := filter.OutputInterface + matches = true + if strings.HasSuffix(ifName, "+") { + matches = strings.HasPrefix(nicName, ifName[:n-1]) + } else { + matches = nicName == ifName + } + return filter.OutputInterfaceInvert != matches + } + return true } diff --git a/pkg/tcpip/stack/iptables_types.go b/pkg/tcpip/stack/iptables_types.go index 1bb0ba1bd..fe06007ae 100644 --- a/pkg/tcpip/stack/iptables_types.go +++ b/pkg/tcpip/stack/iptables_types.go @@ -158,6 +158,19 @@ type IPHeaderFilter struct { // true the filter will match packets that fail the destination // comparison. DstInvert bool + + // OutputInterface matches the name of the outgoing interface for the + // packet. + OutputInterface string + + // OutputInterfaceMask masks the characters of the interface name when + // comparing with OutputInterface. + OutputInterfaceMask string + + // OutputInterfaceInvert inverts the meaning of outgoing interface check, + // i.e. when true the filter will match packets that fail the outgoing + // interface comparison. + OutputInterfaceInvert bool } // A Matcher is the interface for matching packets. diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 8f4c1fe42..54103fdb3 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -1233,7 +1233,7 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link // iptables filtering. ipt := n.stack.IPTables() address := n.primaryAddress(protocol) - if ok := ipt.Check(Prerouting, &pkt, nil, nil, address.Address); !ok { + if ok := ipt.Check(Prerouting, &pkt, nil, nil, address.Address, ""); !ok { // iptables is telling us to drop the packet. return } diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index e33fae4eb..b39ffa9fb 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -1898,9 +1898,23 @@ func (s *Stack) FindNetworkEndpoint(netProto tcpip.NetworkProtocolNumber, addres nic.mu.RLock() defer nic.mu.RUnlock() - // An endpoint with this id exists, check if it can be used and return it. + // An endpoint with this id exists, check if it can be + // used and return it. return ref.ep, nil } } return nil, tcpip.ErrBadAddress } + +// FindNICNameFromID returns the name of the nic for the given NICID. +func (s *Stack) FindNICNameFromID(id tcpip.NICID) string { + s.mu.Lock() + defer s.mu.Unlock() + + nic, ok := s.nics[id] + if !ok { + return "" + } + + return nic.Name() +} diff --git a/test/iptables/filter_output.go b/test/iptables/filter_output.go index b1382d353..c145bd1e9 100644 --- a/test/iptables/filter_output.go +++ b/test/iptables/filter_output.go @@ -29,6 +29,12 @@ func init() { RegisterTestCase(FilterOutputAcceptUDPOwner{}) RegisterTestCase(FilterOutputDropUDPOwner{}) RegisterTestCase(FilterOutputOwnerFail{}) + RegisterTestCase(FilterOutputInterfaceAccept{}) + RegisterTestCase(FilterOutputInterfaceDrop{}) + RegisterTestCase(FilterOutputInterface{}) + RegisterTestCase(FilterOutputInterfaceBeginsWith{}) + RegisterTestCase(FilterOutputInterfaceInvertDrop{}) + RegisterTestCase(FilterOutputInterfaceInvertAccept{}) } // FilterOutputDropTCPDestPort tests that connections are not accepted on @@ -286,3 +292,167 @@ func (FilterOutputInvertDestination) ContainerAction(ip net.IP) error { func (FilterOutputInvertDestination) LocalAction(ip net.IP) error { return listenUDP(acceptPort, sendloopDuration) } + +// FilterOutputInterfaceAccept tests that packets are sent via interface +// matching the iptables rule. +type FilterOutputInterfaceAccept struct{} + +// Name implements TestCase.Name. +func (FilterOutputInterfaceAccept) Name() string { + return "FilterOutputInterfaceAccept" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterOutputInterfaceAccept) ContainerAction(ip net.IP) error { + ifname, ok := getInterfaceName() + if !ok { + return fmt.Errorf("no interface is present, except loopback") + } + if err := filterTable("-A", "OUTPUT", "-p", "udp", "-o", ifname, "-j", "ACCEPT"); err != nil { + return err + } + + return sendUDPLoop(ip, acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (FilterOutputInterfaceAccept) LocalAction(ip net.IP) error { + return listenUDP(acceptPort, sendloopDuration) +} + +// FilterOutputInterfaceDrop tests that packets are not sent via interface +// matching the iptables rule. +type FilterOutputInterfaceDrop struct{} + +// Name implements TestCase.Name. +func (FilterOutputInterfaceDrop) Name() string { + return "FilterOutputInterfaceDrop" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterOutputInterfaceDrop) ContainerAction(ip net.IP) error { + ifname, ok := getInterfaceName() + if !ok { + return fmt.Errorf("no interface is present, except loopback") + } + if err := filterTable("-A", "OUTPUT", "-p", "udp", "-o", ifname, "-j", "DROP"); err != nil { + return err + } + + return sendUDPLoop(ip, acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (FilterOutputInterfaceDrop) LocalAction(ip net.IP) error { + if err := listenUDP(acceptPort, sendloopDuration); err == nil { + return fmt.Errorf("packets should not be received on port %v, but are received", acceptPort) + } + + return nil +} + +// FilterOutputInterface tests that packets are sent via interface which is +// not matching the interface name in the iptables rule. +type FilterOutputInterface struct{} + +// Name implements TestCase.Name. +func (FilterOutputInterface) Name() string { + return "FilterOutputInterface" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterOutputInterface) ContainerAction(ip net.IP) error { + if err := filterTable("-A", "OUTPUT", "-p", "udp", "-o", "lo", "-j", "DROP"); err != nil { + return err + } + + return sendUDPLoop(ip, acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (FilterOutputInterface) LocalAction(ip net.IP) error { + return listenUDP(acceptPort, sendloopDuration) +} + +// FilterOutputInterfaceBeginsWith tests that packets are not sent via an +// interface which begins with the given interface name. +type FilterOutputInterfaceBeginsWith struct{} + +// Name implements TestCase.Name. +func (FilterOutputInterfaceBeginsWith) Name() string { + return "FilterOutputInterfaceBeginsWith" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterOutputInterfaceBeginsWith) ContainerAction(ip net.IP) error { + if err := filterTable("-A", "OUTPUT", "-p", "udp", "-o", "e+", "-j", "DROP"); err != nil { + return err + } + + return sendUDPLoop(ip, acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (FilterOutputInterfaceBeginsWith) LocalAction(ip net.IP) error { + if err := listenUDP(acceptPort, sendloopDuration); err == nil { + return fmt.Errorf("packets should not be received on port %v, but are received", acceptPort) + } + + return nil +} + +// FilterOutputInterfaceInvertDrop tests that we selectively do not send +// packets via interface not matching the interface name. +type FilterOutputInterfaceInvertDrop struct{} + +// Name implements TestCase.Name. +func (FilterOutputInterfaceInvertDrop) Name() string { + return "FilterOutputInterfaceInvertDrop" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterOutputInterfaceInvertDrop) ContainerAction(ip net.IP) error { + if err := filterTable("-A", "OUTPUT", "-p", "tcp", "!", "-o", "lo", "-j", "DROP"); err != nil { + return err + } + + // Listen for TCP packets on accept port. + if err := listenTCP(acceptPort, sendloopDuration); err == nil { + return fmt.Errorf("connection on port %d should not be accepted, but got accepted", acceptPort) + } + + return nil +} + +// LocalAction implements TestCase.LocalAction. +func (FilterOutputInterfaceInvertDrop) LocalAction(ip net.IP) error { + if err := connectTCP(ip, acceptPort, sendloopDuration); err == nil { + return fmt.Errorf("connection destined to port %d should not be accepted, but got accepted", acceptPort) + } + + return nil +} + +// FilterOutputInterfaceInvertAccept tests that we can selectively send packets +// not matching the specific outgoing interface. +type FilterOutputInterfaceInvertAccept struct{} + +// Name implements TestCase.Name. +func (FilterOutputInterfaceInvertAccept) Name() string { + return "FilterOutputInterfaceInvertAccept" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterOutputInterfaceInvertAccept) ContainerAction(ip net.IP) error { + if err := filterTable("-A", "OUTPUT", "-p", "tcp", "!", "-o", "lo", "-j", "ACCEPT"); err != nil { + return err + } + + // Listen for TCP packets on accept port. + return listenTCP(acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (FilterOutputInterfaceInvertAccept) LocalAction(ip net.IP) error { + return connectTCP(ip, acceptPort, sendloopDuration) +} diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go index 63a862d35..84eb75a40 100644 --- a/test/iptables/iptables_test.go +++ b/test/iptables/iptables_test.go @@ -167,6 +167,30 @@ func TestFilterOutputOwnerFail(t *testing.T) { singleTest(t, FilterOutputOwnerFail{}) } +func TestFilterOutputInterfaceAccept(t *testing.T) { + singleTest(t, FilterOutputInterfaceAccept{}) +} + +func TestFilterOutputInterfaceDrop(t *testing.T) { + singleTest(t, FilterOutputInterfaceDrop{}) +} + +func TestFilterOutputInterface(t *testing.T) { + singleTest(t, FilterOutputInterface{}) +} + +func TestFilterOutputInterfaceBeginsWith(t *testing.T) { + singleTest(t, FilterOutputInterfaceBeginsWith{}) +} + +func TestFilterOutputInterfaceInvertDrop(t *testing.T) { + singleTest(t, FilterOutputInterfaceInvertDrop{}) +} + +func TestFilterOutputInterfaceInvertAccept(t *testing.T) { + singleTest(t, FilterOutputInterfaceInvertAccept{}) +} + func TestJumpSerialize(t *testing.T) { singleTest(t, FilterInputSerializeJump{}) } diff --git a/test/iptables/iptables_util.go b/test/iptables/iptables_util.go index 2f988cd18..7146edbb9 100644 --- a/test/iptables/iptables_util.go +++ b/test/iptables/iptables_util.go @@ -169,3 +169,18 @@ func localAddrs() ([]string, error) { } return addrStrs, nil } + +// getInterfaceName returns the name of the interface other than loopback. +func getInterfaceName() (string, bool) { + var ifname string + if interfaces, err := net.Interfaces(); err == nil { + for _, intf := range interfaces { + if intf.Name != "lo" { + ifname = intf.Name + break + } + } + } + + return ifname, ifname != "" +} -- cgit v1.2.3 From c52195d25825abc749c5187f4c18834eabe4bfee Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Sun, 10 May 2020 17:51:01 -0700 Subject: Stop avoiding preadv2 and pwritev2, and add them to the filters. Some code paths needed these syscalls anyways, so they should be included in the filters. Given that we depend on these syscalls in some cases, there's no real reason to avoid them any more. PiperOrigin-RevId: 310829126 --- pkg/sentry/hostfd/hostfd_unsafe.go | 46 ++++++++++---------------------------- runsc/boot/filter/config.go | 2 ++ 2 files changed, 14 insertions(+), 34 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/hostfd/hostfd_unsafe.go b/pkg/sentry/hostfd/hostfd_unsafe.go index 5e9e60fc4..cd4dc67fb 100644 --- a/pkg/sentry/hostfd/hostfd_unsafe.go +++ b/pkg/sentry/hostfd/hostfd_unsafe.go @@ -34,24 +34,13 @@ func Preadv2(fd int32, dsts safemem.BlockSeq, offset int64, flags uint32) (uint6 n uintptr e syscall.Errno ) - // Avoid preadv2(2) if possible, since it's relatively new and thus least - // likely to be supported by the host kernel. - if flags == 0 { - if dsts.NumBlocks() == 1 { - // Use read() or pread() to avoid iovec allocation and copying. - dst := dsts.Head() - if offset == -1 { - n, _, e = syscall.Syscall(unix.SYS_READ, uintptr(fd), dst.Addr(), uintptr(dst.Len())) - } else { - n, _, e = syscall.Syscall6(unix.SYS_PREAD64, uintptr(fd), dst.Addr(), uintptr(dst.Len()), uintptr(offset), 0 /* pos_h */, 0 /* unused */) - } + if flags == 0 && dsts.NumBlocks() == 1 { + // Use read() or pread() to avoid iovec allocation and copying. + dst := dsts.Head() + if offset == -1 { + n, _, e = syscall.Syscall(unix.SYS_READ, uintptr(fd), dst.Addr(), uintptr(dst.Len())) } else { - iovs := safemem.IovecsFromBlockSeq(dsts) - if offset == -1 { - n, _, e = syscall.Syscall(unix.SYS_READV, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs))) - } else { - n, _, e = syscall.Syscall6(unix.SYS_PREADV, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs)), uintptr(offset), 0 /* pos_h */, 0 /* unused */) - } + n, _, e = syscall.Syscall6(unix.SYS_PREAD64, uintptr(fd), dst.Addr(), uintptr(dst.Len()), uintptr(offset), 0 /* pos_h */, 0 /* unused */) } } else { iovs := safemem.IovecsFromBlockSeq(dsts) @@ -77,24 +66,13 @@ func Pwritev2(fd int32, srcs safemem.BlockSeq, offset int64, flags uint32) (uint n uintptr e syscall.Errno ) - // Avoid pwritev2(2) if possible, since it's relatively new and thus least - // likely to be supported by the host kernel. - if flags == 0 { - if srcs.NumBlocks() == 1 { - // Use write() or pwrite() to avoid iovec allocation and copying. - src := srcs.Head() - if offset == -1 { - n, _, e = syscall.Syscall(unix.SYS_WRITE, uintptr(fd), src.Addr(), uintptr(src.Len())) - } else { - n, _, e = syscall.Syscall6(unix.SYS_PWRITE64, uintptr(fd), src.Addr(), uintptr(src.Len()), uintptr(offset), 0 /* pos_h */, 0 /* unused */) - } + if flags == 0 && srcs.NumBlocks() == 1 { + // Use write() or pwrite() to avoid iovec allocation and copying. + src := srcs.Head() + if offset == -1 { + n, _, e = syscall.Syscall(unix.SYS_WRITE, uintptr(fd), src.Addr(), uintptr(src.Len())) } else { - iovs := safemem.IovecsFromBlockSeq(srcs) - if offset == -1 { - n, _, e = syscall.Syscall(unix.SYS_WRITEV, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs))) - } else { - n, _, e = syscall.Syscall6(unix.SYS_PWRITEV, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs)), uintptr(offset), 0 /* pos_h */, 0 /* unused */) - } + n, _, e = syscall.Syscall6(unix.SYS_PWRITE64, uintptr(fd), src.Addr(), uintptr(src.Len()), uintptr(offset), 0 /* pos_h */, 0 /* unused */) } } else { iovs := safemem.IovecsFromBlockSeq(srcs) diff --git a/runsc/boot/filter/config.go b/runsc/boot/filter/config.go index 1828d116a..98cdd90dd 100644 --- a/runsc/boot/filter/config.go +++ b/runsc/boot/filter/config.go @@ -230,8 +230,10 @@ var allowedSyscalls = seccomp.SyscallRules{ syscall.SYS_PPOLL: {}, syscall.SYS_PREAD64: {}, syscall.SYS_PREADV: {}, + unix.SYS_PREADV2: {}, syscall.SYS_PWRITE64: {}, syscall.SYS_PWRITEV: {}, + unix.SYS_PWRITEV2: {}, syscall.SYS_READ: {}, syscall.SYS_RECVMSG: []seccomp.Rule{ { -- cgit v1.2.3 From 15de8cc9e0e7789c3d55595171b3272ec726931f Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Mon, 11 May 2020 16:13:14 -0700 Subject: Add fsimpl/gofer.InternalFilesystemOptions.OpenSocketsByConnecting. PiperOrigin-RevId: 311014995 --- pkg/sentry/fsimpl/gofer/filesystem.go | 27 +++++++++++++++++++++-- pkg/sentry/fsimpl/gofer/gofer.go | 4 ++++ pkg/sentry/fsimpl/host/host.go | 41 +++++++++++++++++++++++++++-------- 3 files changed, 61 insertions(+), 11 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 4a32821bd..7ca0cf3f4 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -21,6 +21,8 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/p9" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/host" + "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" @@ -835,6 +837,9 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vf if d.isSynthetic() { return nil, syserror.ENXIO } + if d.fs.iopts.OpenSocketsByConnecting { + return d.connectSocketLocked(ctx, opts) + } case linux.S_IFIFO: if d.isSynthetic() { return d.pipe.Open(ctx, mnt, &d.vfsd, opts.Flags) @@ -843,10 +848,28 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vf return d.openSpecialFileLocked(ctx, mnt, opts) } +func (d *dentry) connectSocketLocked(ctx context.Context, opts *vfs.OpenOptions) (*vfs.FileDescription, error) { + if opts.Flags&linux.O_DIRECT != 0 { + return nil, syserror.EINVAL + } + fdObj, err := d.file.connect(ctx, p9.AnonymousSocket) + if err != nil { + return nil, err + } + fd, err := host.NewFD(ctx, kernel.KernelFromContext(ctx).HostMount(), fdObj.FD(), &host.NewFDOptions{ + HaveFlags: true, + Flags: opts.Flags, + }) + if err != nil { + fdObj.Close() + return nil, err + } + fdObj.Release() + return fd, nil +} + func (d *dentry) openSpecialFileLocked(ctx context.Context, mnt *vfs.Mount, opts *vfs.OpenOptions) (*vfs.FileDescription, error) { ats := vfs.AccessTypesForOpenFlags(opts) - // Treat as a special file. This is done for non-synthetic pipes as well as - // regular files when d.fs.opts.regularFilesUseSpecialFileFD is true. if opts.Flags&linux.O_DIRECT != 0 { return nil, syserror.EINVAL } diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index e68e37ebc..1da8d5d82 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -221,6 +221,10 @@ type InternalFilesystemOptions struct { // which servers can handle only a single client and report failure if that // client disconnects. LeakConnection bool + + // If OpenSocketsByConnecting is true, silently translate attempts to open + // files identifying as sockets to connect RPCs. + OpenSocketsByConnecting bool } // Name implements vfs.FilesystemType.Name. diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 36bceeaa4..8caf55a1b 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -40,8 +40,20 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) -// ImportFD sets up and returns a vfs.FileDescription from a donated fd. -func ImportFD(ctx context.Context, mnt *vfs.Mount, hostFD int, isTTY bool) (*vfs.FileDescription, error) { +// NewFDOptions contains options to NewFD. +type NewFDOptions struct { + // If IsTTY is true, the file descriptor is a TTY. + IsTTY bool + + // If HaveFlags is true, use Flags for the new file description. Otherwise, + // the new file description will inherit flags from hostFD. + HaveFlags bool + Flags uint32 +} + +// NewFD returns a vfs.FileDescription representing the given host file +// descriptor. mnt must be Kernel.HostMount(). +func NewFD(ctx context.Context, mnt *vfs.Mount, hostFD int, opts *NewFDOptions) (*vfs.FileDescription, error) { fs, ok := mnt.Filesystem().Impl().(*filesystem) if !ok { return nil, fmt.Errorf("can't import host FDs into filesystems of type %T", mnt.Filesystem().Impl()) @@ -53,10 +65,14 @@ func ImportFD(ctx context.Context, mnt *vfs.Mount, hostFD int, isTTY bool) (*vfs return nil, err } - // Get flags for the imported FD. - flags, err := unix.FcntlInt(uintptr(hostFD), syscall.F_GETFL, 0) - if err != nil { - return nil, err + flags := opts.Flags + if !opts.HaveFlags { + // Get flags for the imported FD. + flagsInt, err := unix.FcntlInt(uintptr(hostFD), syscall.F_GETFL, 0) + if err != nil { + return nil, err + } + flags = uint32(flagsInt) } fileMode := linux.FileMode(s.Mode) @@ -65,13 +81,13 @@ func ImportFD(ctx context.Context, mnt *vfs.Mount, hostFD int, isTTY bool) (*vfs // Determine if hostFD is seekable. If not, this syscall will return ESPIPE // (see fs/read_write.c:llseek), e.g. for pipes, sockets, and some character // devices. - _, err = unix.Seek(hostFD, 0, linux.SEEK_CUR) + _, err := unix.Seek(hostFD, 0, linux.SEEK_CUR) seekable := err != syserror.ESPIPE i := &inode{ hostFD: hostFD, seekable: seekable, - isTTY: isTTY, + isTTY: opts.IsTTY, canMap: canMap(uint32(fileType)), wouldBlock: wouldBlock(uint32(fileType)), ino: fs.NextIno(), @@ -101,7 +117,14 @@ func ImportFD(ctx context.Context, mnt *vfs.Mount, hostFD int, isTTY bool) (*vfs // i.open will take a reference on d. defer d.DecRef() - return i.open(ctx, d.VFSDentry(), mnt, uint32(flags)) + return i.open(ctx, d.VFSDentry(), mnt, flags) +} + +// ImportFD sets up and returns a vfs.FileDescription from a donated fd. +func ImportFD(ctx context.Context, mnt *vfs.Mount, hostFD int, isTTY bool) (*vfs.FileDescription, error) { + return NewFD(ctx, mnt, hostFD, &NewFDOptions{ + IsTTY: isTTY, + }) } // filesystemType implements vfs.FilesystemType. -- cgit v1.2.3 From 9bd9882b81dfe2eea9af0a12bbaab35ddbd1ab47 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Mon, 11 May 2020 01:52:45 -0400 Subject: Add fpsimd support in sigreturn on Arm64 Signed-off-by: Bin Lu --- pkg/sentry/arch/signal_arm64.go | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/arch/signal_arm64.go b/pkg/sentry/arch/signal_arm64.go index 1cb1adf8c..642c79dda 100644 --- a/pkg/sentry/arch/signal_arm64.go +++ b/pkg/sentry/arch/signal_arm64.go @@ -19,6 +19,7 @@ import ( "syscall" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/usermem" ) @@ -134,6 +135,11 @@ func (c *context64) SignalSetup(st *Stack, act *SignalAct, info *SignalInfo, alt c.Regs.Regs[1] = uint64(infoAddr) c.Regs.Regs[2] = uint64(ucAddr) c.Regs.Regs[30] = uint64(act.Restorer) + + // Save the thread's floating point state. + c.sigFPState = append(c.sigFPState, c.aarch64FPState) + // Signal handler gets a clean floating point state. + c.aarch64FPState = newAarch64FPState() return nil } @@ -155,5 +161,21 @@ func (c *context64) SignalRestore(st *Stack, rt bool) (linux.SignalSet, SignalSt c.Regs.Sp = uc.MContext.Sp c.Regs.Pstate = uc.MContext.Pstate + // Restore floating point state. + l := len(c.sigFPState) + if l > 0 { + c.aarch64FPState = c.sigFPState[l-1] + // NOTE(cl/133042258): State save requires that any slice + // elements from '[len:cap]' to be zero value. + c.sigFPState[l-1] = nil + c.sigFPState = c.sigFPState[0 : l-1] + } else { + // This might happen if sigreturn(2) calls are unbalanced with + // respect to signal handler entries. This is not expected so + // don't bother to do anything fancy with the floating point + // state. + log.Warningf("sigreturn unable to restore application fpstate") + } + return uc.Sigset, uc.Stack, nil } -- cgit v1.2.3 From 87225fad2a468e1516784f13abe8bb946d0172c6 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Mon, 23 Mar 2020 14:33:20 -0700 Subject: iptables: check for truly unconditional rules We weren't properly checking whether the inserted default rule was unconditional. --- pkg/sentry/socket/netfilter/netfilter.go | 22 ++++++++++++++++++++-- 1 file changed, 20 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 41a1ce031..789bb94c8 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -59,6 +59,13 @@ type metadata struct { // developing iptables, but can pollute sentry logs otherwise. const enableLogging = false +// emptyFilter is for comparison with a rule's filters to determine whether it +// is also empty. It is immutable. +var emptyFilter = stack.IPHeaderFilter{ + Dst: "\x00\x00\x00\x00", + DstMask: "\x00\x00\x00\x00", +} + // nflog logs messages related to the writing and reading of iptables. func nflog(format string, args ...interface{}) { if enableLogging && log.IsLogging(log.Debug) { @@ -484,7 +491,7 @@ func SetEntries(stk *stack.Stack, optVal []byte) *syserr.Error { } if offset == replace.Underflow[hook] { if !validUnderflow(table.Rules[ruleIdx]) { - nflog("underflow for hook %d isn't an unconditional ACCEPT or DROP") + nflog("underflow for hook %d isn't an unconditional ACCEPT or DROP", ruleIdx) return syserr.ErrInvalidArgument } table.Underflows[hk] = ruleIdx @@ -547,7 +554,7 @@ func SetEntries(stk *stack.Stack, optVal []byte) *syserr.Error { // make sure all other chains point to ACCEPT rules. for hook, ruleIdx := range table.BuiltinChains { if hook == stack.Forward || hook == stack.Postrouting { - if _, ok := table.Rules[ruleIdx].Target.(stack.AcceptTarget); !ok { + if !isUnconditionalAccept(table.Rules[ruleIdx]) { nflog("hook %d is unsupported.", hook) return syserr.ErrInvalidArgument } @@ -776,6 +783,9 @@ func validUnderflow(rule stack.Rule) bool { if len(rule.Matchers) != 0 { return false } + if rule.Filter != emptyFilter { + return false + } switch rule.Target.(type) { case stack.AcceptTarget, stack.DropTarget: return true @@ -784,6 +794,14 @@ func validUnderflow(rule stack.Rule) bool { } } +func isUnconditionalAccept(rule stack.Rule) bool { + if !validUnderflow(rule) { + return false + } + _, ok := rule.Target.(stack.AcceptTarget) + return ok +} + func hookFromLinux(hook int) stack.Hook { switch hook { case linux.NF_INET_PRE_ROUTING: -- cgit v1.2.3 From 94251aedb4abcf2498a4f6bc71f32d2847ac0f3e Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Mon, 11 May 2020 20:02:09 -0700 Subject: Internal change. PiperOrigin-RevId: 311046755 --- pkg/sentry/fsimpl/gofer/filesystem.go | 35 ++++++++++++++++++--------------- pkg/sentry/fsimpl/gofer/special_file.go | 28 ++++++++++++++++++-------- 2 files changed, 39 insertions(+), 24 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 7ca0cf3f4..7f2181216 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -877,10 +877,15 @@ func (d *dentry) openSpecialFileLocked(ctx context.Context, mnt *vfs.Mount, opts if err != nil { return nil, err } + seekable := d.fileType() == linux.S_IFREG fd := &specialFileFD{ - handle: h, + handle: h, + seekable: seekable, } - if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { + if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{ + DenyPRead: !seekable, + DenyPWrite: !seekable, + }); err != nil { h.close(ctx) return nil, err } @@ -911,7 +916,11 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving } creds := rp.Credentials() name := rp.Component() - fdobj, openFile, createQID, _, err := dirfile.create(ctx, name, (p9.OpenFlags)(opts.Flags), (p9.FileMode)(opts.Mode), (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)) + // Filter file creation flags and O_LARGEFILE out; the create RPC already + // has the semantics of O_CREAT|O_EXCL, while some servers will choke on + // O_LARGEFILE. + createFlags := p9.OpenFlags(opts.Flags &^ (linux.O_CREAT | linux.O_EXCL | linux.O_NOCTTY | linux.O_TRUNC | linux.O_LARGEFILE)) + fdobj, openFile, createQID, _, err := dirfile.create(ctx, name, createFlags, (p9.FileMode)(opts.Mode), (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)) if err != nil { dirfile.close(ctx) return nil, err @@ -919,7 +928,7 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving // Then we need to walk to the file we just created to get a non-open fid // representing it, and to get its metadata. This must use d.file since, as // explained above, dirfile was invalidated by dirfile.Create(). - walkQID, nonOpenFile, attrMask, attr, err := d.file.walkGetAttrOne(ctx, name) + _, nonOpenFile, attrMask, attr, err := d.file.walkGetAttrOne(ctx, name) if err != nil { openFile.close(ctx) if fdobj != nil { @@ -927,17 +936,6 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving } return nil, err } - // Sanity-check that we walked to the file we created. - if createQID.Path != walkQID.Path { - // Probably due to concurrent remote filesystem mutation? - ctx.Warningf("gofer.dentry.createAndOpenChildLocked: created file has QID %v before walk, QID %v after (interop=%v)", createQID, walkQID, d.fs.opts.interop) - nonOpenFile.close(ctx) - openFile.close(ctx) - if fdobj != nil { - fdobj.Close() - } - return nil, syserror.EAGAIN - } // Construct the new dentry. child, err := d.fs.newDentry(ctx, nonOpenFile, createQID, attrMask, &attr) @@ -983,16 +981,21 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving } childVFSFD = &fd.vfsfd } else { + seekable := child.fileType() == linux.S_IFREG fd := &specialFileFD{ handle: handle{ file: openFile, fd: -1, }, + seekable: seekable, } if fdobj != nil { fd.handle.fd = int32(fdobj.Release()) } - if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &child.vfsd, &vfs.FileDescriptionOptions{}); err != nil { + if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &child.vfsd, &vfs.FileDescriptionOptions{ + DenyPRead: !seekable, + DenyPWrite: !seekable, + }); err != nil { fd.handle.close(ctx) return nil, err } diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go index 507e0e276..a464e6a94 100644 --- a/pkg/sentry/fsimpl/gofer/special_file.go +++ b/pkg/sentry/fsimpl/gofer/special_file.go @@ -33,13 +33,14 @@ import ( type specialFileFD struct { fileDescription - // handle is immutable. + // handle is used for file I/O. handle is immutable. handle handle - // off is the file offset. off is protected by mu. (POSIX 2.9.7 only - // requires operations using the file offset to be atomic for regular files - // and symlinks; however, since specialFileFD may be used for regular - // files, we apply this atomicity unconditionally.) + // seekable is true if this file description represents a file for which + // file offset is significant, i.e. a regular file. seekable is immutable. + seekable bool + + // If seekable is true, off is the file offset. off is protected by mu. mu sync.Mutex off int64 } @@ -63,7 +64,7 @@ func (fd *specialFileFD) OnClose(ctx context.Context) error { // PRead implements vfs.FileDescriptionImpl.PRead. func (fd *specialFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - if offset < 0 { + if fd.seekable && offset < 0 { return 0, syserror.EINVAL } if opts.Flags != 0 { @@ -91,6 +92,10 @@ func (fd *specialFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offs // Read implements vfs.FileDescriptionImpl.Read. func (fd *specialFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + if !fd.seekable { + return fd.PRead(ctx, dst, -1, opts) + } + fd.mu.Lock() n, err := fd.PRead(ctx, dst, fd.off, opts) fd.off += n @@ -100,14 +105,14 @@ func (fd *specialFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts // PWrite implements vfs.FileDescriptionImpl.PWrite. func (fd *specialFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - if offset < 0 { + if fd.seekable && offset < 0 { return 0, syserror.EINVAL } if opts.Flags != 0 { return 0, syserror.EOPNOTSUPP } - if fd.dentry().fileType() == linux.S_IFREG { + if fd.seekable { limit, err := vfs.CheckLimit(ctx, offset, src.NumBytes()) if err != nil { return 0, err @@ -130,6 +135,10 @@ func (fd *specialFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off // Write implements vfs.FileDescriptionImpl.Write. func (fd *specialFileFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + if !fd.seekable { + return fd.PWrite(ctx, src, -1, opts) + } + fd.mu.Lock() n, err := fd.PWrite(ctx, src, fd.off, opts) fd.off += n @@ -139,6 +148,9 @@ func (fd *specialFileFD) Write(ctx context.Context, src usermem.IOSequence, opts // Seek implements vfs.FileDescriptionImpl.Seek. func (fd *specialFileFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + if !fd.seekable { + return 0, syserror.ESPIPE + } fd.mu.Lock() defer fd.mu.Unlock() switch whence { -- cgit v1.2.3 From 27b1f19cabe04effbb37fa6a680b65987b379313 Mon Sep 17 00:00:00 2001 From: Nayana Bidari Date: Tue, 12 May 2020 12:14:56 -0700 Subject: iptables: support gid match for owner matching. - Added support for matching gid owner and invert flag for uid and gid. $ iptables -A OUTPUT -p tcp -m owner --gid-owner root -j ACCEPT $ iptables -A OUTPUT -p tcp -m owner ! --uid-owner root -j ACCEPT $ iptables -A OUTPUT -p tcp -m owner ! --gid-owner root -j DROP - Added tests for uid, gid and invert flags. --- pkg/sentry/socket/netfilter/owner_matcher.go | 81 ++++++++----- test/iptables/filter_output.go | 169 +++++++++++++++++++++++++-- test/iptables/iptables_test.go | 20 ++++ 3 files changed, 230 insertions(+), 40 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/owner_matcher.go b/pkg/sentry/socket/netfilter/owner_matcher.go index 5949a7c29..3863293c7 100644 --- a/pkg/sentry/socket/netfilter/owner_matcher.go +++ b/pkg/sentry/socket/netfilter/owner_matcher.go @@ -45,14 +45,18 @@ func (ownerMarshaler) marshal(mr stack.Matcher) []byte { GID: matcher.gid, } - // Support for UID match. - // TODO(gvisor.dev/issue/170): Need to support gid match. + // Support for UID and GID match. if matcher.matchUID { iptOwnerInfo.Match = linux.XT_OWNER_UID - } else if matcher.matchGID { - panic("GID match is not supported.") - } else { - panic("UID match is not set.") + if matcher.invertUID { + iptOwnerInfo.Invert = linux.XT_OWNER_UID + } + } + if matcher.matchGID { + iptOwnerInfo.Match |= linux.XT_OWNER_GID + if matcher.invertGID { + iptOwnerInfo.Invert |= linux.XT_OWNER_GID + } } buf := make([]byte, 0, linux.SizeOfIPTOwnerInfo) @@ -71,31 +75,34 @@ func (ownerMarshaler) unmarshal(buf []byte, filter stack.IPHeaderFilter) (stack. binary.Unmarshal(buf[:linux.SizeOfIPTOwnerInfo], usermem.ByteOrder, &matchData) nflog("parseMatchers: parsed IPTOwnerInfo: %+v", matchData) - if matchData.Invert != 0 { - return nil, fmt.Errorf("invert flag is not supported for owner match") - } - - // Support for UID match. - // TODO(gvisor.dev/issue/170): Need to support gid match. - if matchData.Match&linux.XT_OWNER_UID != linux.XT_OWNER_UID { - return nil, fmt.Errorf("owner match is only supported for uid") - } - - // Check Flags. var owner OwnerMatcher owner.uid = matchData.UID owner.gid = matchData.GID - owner.matchUID = true + + // Check flags. + if matchData.Match&linux.XT_OWNER_UID != 0 { + owner.matchUID = true + if matchData.Invert&linux.XT_OWNER_UID != 0 { + owner.invertUID = true + } + } + if matchData.Match&linux.XT_OWNER_GID != 0 { + owner.matchGID = true + if matchData.Invert&linux.XT_OWNER_GID != 0 { + owner.invertGID = true + } + } return &owner, nil } type OwnerMatcher struct { - uid uint32 - gid uint32 - matchUID bool - matchGID bool - invert uint8 + uid uint32 + gid uint32 + matchUID bool + matchGID bool + invertUID bool + invertGID bool } // Name implements Matcher.Name. @@ -112,16 +119,30 @@ func (om *OwnerMatcher) Match(hook stack.Hook, pkt stack.PacketBuffer, interface } // If the packet owner is not set, drop the packet. - // Support for uid match. - // TODO(gvisor.dev/issue/170): Need to support gid match. - if pkt.Owner == nil || !om.matchUID { + if pkt.Owner == nil { return false, true } - // TODO(gvisor.dev/issue/170): Need to add tests to verify - // drop rule when packet UID does not match owner matcher UID. - if pkt.Owner.UID() != om.uid { - return false, false + var matches bool + // Check for UID match. + if om.matchUID { + if pkt.Owner.UID() == om.uid { + matches = true + } + if matches == om.invertUID { + return false, false + } + } + + // Check for GID match. + if om.matchGID { + matches = false + if pkt.Owner.GID() == om.gid { + matches = true + } + if matches == om.invertGID { + return false, false + } } return true, false diff --git a/test/iptables/filter_output.go b/test/iptables/filter_output.go index c145bd1e9..ba0d6fc29 100644 --- a/test/iptables/filter_output.go +++ b/test/iptables/filter_output.go @@ -29,6 +29,11 @@ func init() { RegisterTestCase(FilterOutputAcceptUDPOwner{}) RegisterTestCase(FilterOutputDropUDPOwner{}) RegisterTestCase(FilterOutputOwnerFail{}) + RegisterTestCase(FilterOutputAcceptGIDOwner{}) + RegisterTestCase(FilterOutputDropGIDOwner{}) + RegisterTestCase(FilterOutputInvertGIDOwner{}) + RegisterTestCase(FilterOutputInvertUIDOwner{}) + RegisterTestCase(FilterOutputInvertUIDAndGIDOwner{}) RegisterTestCase(FilterOutputInterfaceAccept{}) RegisterTestCase(FilterOutputInterfaceDrop{}) RegisterTestCase(FilterOutputInterface{}) @@ -116,20 +121,12 @@ func (FilterOutputAcceptTCPOwner) ContainerAction(ip net.IP) error { } // Listen for TCP packets on accept port. - if err := listenTCP(acceptPort, sendloopDuration); err != nil { - return fmt.Errorf("connection on port %d should be accepted, but got dropped", acceptPort) - } - - return nil + return listenTCP(acceptPort, sendloopDuration) } // LocalAction implements TestCase.LocalAction. func (FilterOutputAcceptTCPOwner) LocalAction(ip net.IP) error { - if err := connectTCP(ip, acceptPort, sendloopDuration); err != nil { - return fmt.Errorf("connection destined to port %d should be accepted, but got dropped", acceptPort) - } - - return nil + return connectTCP(ip, acceptPort, sendloopDuration) } // FilterOutputDropTCPOwner tests that TCP connections from uid owner are dropped. @@ -239,6 +236,158 @@ func (FilterOutputOwnerFail) LocalAction(ip net.IP) error { return nil } +// FilterOutputAcceptGIDOwner tests that TCP connections from gid owner are accepted. +type FilterOutputAcceptGIDOwner struct{} + +// Name implements TestCase.Name. +func (FilterOutputAcceptGIDOwner) Name() string { + return "FilterOutputAcceptGIDOwner" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterOutputAcceptGIDOwner) ContainerAction(ip net.IP) error { + if err := filterTable("-A", "OUTPUT", "-p", "tcp", "-m", "owner", "--gid-owner", "root", "-j", "ACCEPT"); err != nil { + return err + } + + // Listen for TCP packets on accept port. + return listenTCP(acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (FilterOutputAcceptGIDOwner) LocalAction(ip net.IP) error { + return connectTCP(ip, acceptPort, sendloopDuration) +} + +// FilterOutputDropGIDOwner tests that TCP connections from gid owner are dropped. +type FilterOutputDropGIDOwner struct{} + +// Name implements TestCase.Name. +func (FilterOutputDropGIDOwner) Name() string { + return "FilterOutputDropGIDOwner" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterOutputDropGIDOwner) ContainerAction(ip net.IP) error { + if err := filterTable("-A", "OUTPUT", "-p", "tcp", "-m", "owner", "--gid-owner", "root", "-j", "DROP"); err != nil { + return err + } + + // Listen for TCP packets on accept port. + if err := listenTCP(acceptPort, sendloopDuration); err == nil { + return fmt.Errorf("connection on port %d should not be accepted, but got accepted", acceptPort) + } + + return nil +} + +// LocalAction implements TestCase.LocalAction. +func (FilterOutputDropGIDOwner) LocalAction(ip net.IP) error { + if err := connectTCP(ip, acceptPort, sendloopDuration); err == nil { + return fmt.Errorf("connection destined to port %d should not be accepted, but got accepted", acceptPort) + } + + return nil +} + +// FilterOutputInvertGIDOwner tests that TCP connections from gid owner are dropped. +type FilterOutputInvertGIDOwner struct{} + +// Name implements TestCase.Name. +func (FilterOutputInvertGIDOwner) Name() string { + return "FilterOutputInvertGIDOwner" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterOutputInvertGIDOwner) ContainerAction(ip net.IP) error { + rules := [][]string{ + {"-A", "OUTPUT", "-p", "tcp", "-m", "owner", "!", "--gid-owner", "root", "-j", "ACCEPT"}, + {"-A", "OUTPUT", "-p", "tcp", "-j", "DROP"}, + } + if err := filterTableRules(rules); err != nil { + return err + } + + // Listen for TCP packets on accept port. + if err := listenTCP(acceptPort, sendloopDuration); err == nil { + return fmt.Errorf("connection on port %d should not be accepted, but got accepted", acceptPort) + } + + return nil +} + +// LocalAction implements TestCase.LocalAction. +func (FilterOutputInvertGIDOwner) LocalAction(ip net.IP) error { + if err := connectTCP(ip, acceptPort, sendloopDuration); err == nil { + return fmt.Errorf("connection destined to port %d should not be accepted, but got accepted", acceptPort) + } + + return nil +} + +// FilterOutputInvertUIDOwner tests that TCP connections from gid owner are dropped. +type FilterOutputInvertUIDOwner struct{} + +// Name implements TestCase.Name. +func (FilterOutputInvertUIDOwner) Name() string { + return "FilterOutputInvertUIDOwner" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterOutputInvertUIDOwner) ContainerAction(ip net.IP) error { + rules := [][]string{ + {"-A", "OUTPUT", "-p", "tcp", "-m", "owner", "!", "--uid-owner", "root", "-j", "DROP"}, + {"-A", "OUTPUT", "-p", "tcp", "-j", "ACCEPT"}, + } + if err := filterTableRules(rules); err != nil { + return err + } + + // Listen for TCP packets on accept port. + return listenTCP(acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (FilterOutputInvertUIDOwner) LocalAction(ip net.IP) error { + return connectTCP(ip, acceptPort, sendloopDuration) +} + +// FilterOutputInvertUIDAndGIDOwner tests that TCP connections from uid and gid +// owner are dropped. +type FilterOutputInvertUIDAndGIDOwner struct{} + +// Name implements TestCase.Name. +func (FilterOutputInvertUIDAndGIDOwner) Name() string { + return "FilterOutputInvertUIDAndGIDOwner" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterOutputInvertUIDAndGIDOwner) ContainerAction(ip net.IP) error { + rules := [][]string{ + {"-A", "OUTPUT", "-p", "tcp", "-m", "owner", "!", "--uid-owner", "root", "!", "--gid-owner", "root", "-j", "ACCEPT"}, + {"-A", "OUTPUT", "-p", "tcp", "-j", "DROP"}, + } + if err := filterTableRules(rules); err != nil { + return err + } + + // Listen for TCP packets on accept port. + if err := listenTCP(acceptPort, sendloopDuration); err == nil { + return fmt.Errorf("connection on port %d should not be accepted, but got accepted", acceptPort) + } + + return nil +} + +// LocalAction implements TestCase.LocalAction. +func (FilterOutputInvertUIDAndGIDOwner) LocalAction(ip net.IP) error { + if err := connectTCP(ip, acceptPort, sendloopDuration); err == nil { + return fmt.Errorf("connection destined to port %d should not be accepted, but got accepted", acceptPort) + } + + return nil +} + // FilterOutputDestination tests that we can selectively allow packets to // certain destinations. type FilterOutputDestination struct{} diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go index 84eb75a40..4fd2cb46a 100644 --- a/test/iptables/iptables_test.go +++ b/test/iptables/iptables_test.go @@ -167,6 +167,26 @@ func TestFilterOutputOwnerFail(t *testing.T) { singleTest(t, FilterOutputOwnerFail{}) } +func TestFilterOutputAcceptGIDOwner(t *testing.T) { + singleTest(t, FilterOutputAcceptGIDOwner{}) +} + +func TestFilterOutputDropGIDOwner(t *testing.T) { + singleTest(t, FilterOutputDropGIDOwner{}) +} + +func TestFilterOutputInvertGIDOwner(t *testing.T) { + singleTest(t, FilterOutputInvertGIDOwner{}) +} + +func TestFilterOutputInvertUIDOwner(t *testing.T) { + singleTest(t, FilterOutputInvertUIDOwner{}) +} + +func TestFilterOutputInvertUIDAndGIDOwner(t *testing.T) { + singleTest(t, FilterOutputInvertUIDAndGIDOwner{}) +} + func TestFilterOutputInterfaceAccept(t *testing.T) { singleTest(t, FilterOutputInterfaceAccept{}) } -- cgit v1.2.3 From 8dd1d5b75a95100e747b1a88e9e557d5d2c30b64 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Tue, 12 May 2020 12:24:23 -0700 Subject: Don't call kernel.Task.Block() from netstack.SocketOperations.Write(). kernel.Task.Block() requires that the caller is running on the task goroutine. netstack.SocketOperations.Write() uses kernel.TaskFromContext() to call kernel.Task.Block() even if it's not running on the task goroutine. Stop doing that. PiperOrigin-RevId: 311178335 --- pkg/amutex/BUILD | 1 + pkg/amutex/amutex.go | 17 +++++++++++++++++ pkg/sentry/socket/netstack/BUILD | 1 + pkg/sentry/socket/netstack/netstack.go | 13 +++++-------- pkg/sentry/socket/netstack/netstack_vfs2.go | 17 +++-------------- 5 files changed, 27 insertions(+), 22 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/amutex/BUILD b/pkg/amutex/BUILD index 9612f072e..ffc918846 100644 --- a/pkg/amutex/BUILD +++ b/pkg/amutex/BUILD @@ -6,6 +6,7 @@ go_library( name = "amutex", srcs = ["amutex.go"], visibility = ["//:sandbox"], + deps = ["//pkg/syserror"], ) go_test( diff --git a/pkg/amutex/amutex.go b/pkg/amutex/amutex.go index 1c4fd1784..a078a31db 100644 --- a/pkg/amutex/amutex.go +++ b/pkg/amutex/amutex.go @@ -18,6 +18,8 @@ package amutex import ( "sync/atomic" + + "gvisor.dev/gvisor/pkg/syserror" ) // Sleeper must be implemented by users of the abortable mutex to allow for @@ -53,6 +55,21 @@ func (NoopSleeper) SleepFinish(success bool) {} // Interrupted implements Sleeper.Interrupted. func (NoopSleeper) Interrupted() bool { return false } +// Block blocks until either receiving from ch succeeds (in which case it +// returns nil) or sleeper is interrupted (in which case it returns +// syserror.ErrInterrupted). +func Block(sleeper Sleeper, ch <-chan struct{}) error { + cancel := sleeper.SleepStart() + select { + case <-ch: + sleeper.SleepFinish(true) + return nil + case <-cancel: + sleeper.SleepFinish(false) + return syserror.ErrInterrupted + } +} + // AbortableMutex is an abortable mutex. It allows Lock() to be aborted while it // waits to acquire the mutex. type AbortableMutex struct { diff --git a/pkg/sentry/socket/netstack/BUILD b/pkg/sentry/socket/netstack/BUILD index 6129fb83d..333e0042e 100644 --- a/pkg/sentry/socket/netstack/BUILD +++ b/pkg/sentry/socket/netstack/BUILD @@ -18,6 +18,7 @@ go_library( ], deps = [ "//pkg/abi/linux", + "//pkg/amutex", "//pkg/binary", "//pkg/context", "//pkg/log", diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 81053d8ef..9d032f052 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -34,6 +34,7 @@ import ( "time" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/amutex" "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" @@ -553,11 +554,9 @@ func (s *SocketOperations) Write(ctx context.Context, _ *fs.File, src usermem.IO } if resCh != nil { - t := kernel.TaskFromContext(ctx) - if err := t.Block(resCh); err != nil { - return 0, syserr.FromError(err).ToError() + if err := amutex.Block(ctx, resCh); err != nil { + return 0, err } - n, _, err = s.Endpoint.Write(f, tcpip.WriteOptions{}) } @@ -626,11 +625,9 @@ func (s *SocketOperations) ReadFrom(ctx context.Context, _ *fs.File, r io.Reader } if resCh != nil { - t := kernel.TaskFromContext(ctx) - if err := t.Block(resCh); err != nil { - return 0, syserr.FromError(err).ToError() + if err := amutex.Block(ctx, resCh); err != nil { + return 0, err } - n, _, err = s.Endpoint.Write(f, tcpip.WriteOptions{ Atomic: true, // See above. }) diff --git a/pkg/sentry/socket/netstack/netstack_vfs2.go b/pkg/sentry/socket/netstack/netstack_vfs2.go index 191970d41..fcd8013c0 100644 --- a/pkg/sentry/socket/netstack/netstack_vfs2.go +++ b/pkg/sentry/socket/netstack/netstack_vfs2.go @@ -16,6 +16,7 @@ package netstack import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/amutex" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" @@ -89,11 +90,6 @@ func (s *SocketVFS2) EventUnregister(e *waiter.Entry) { s.socketOpsCommon.EventUnregister(e) } -// PRead implements vfs.FileDescriptionImpl. -func (s *SocketVFS2) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - return 0, syserror.ESPIPE -} - // Read implements vfs.FileDescriptionImpl. func (s *SocketVFS2) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { // All flags other than RWF_NOWAIT should be ignored. @@ -115,11 +111,6 @@ func (s *SocketVFS2) Read(ctx context.Context, dst usermem.IOSequence, opts vfs. return int64(n), nil } -// PWrite implements vfs.FileDescriptionImpl. -func (s *SocketVFS2) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - return 0, syserror.ESPIPE -} - // Write implements vfs.FileDescriptionImpl. func (s *SocketVFS2) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { // All flags other than RWF_NOWAIT should be ignored. @@ -135,11 +126,9 @@ func (s *SocketVFS2) Write(ctx context.Context, src usermem.IOSequence, opts vfs } if resCh != nil { - t := kernel.TaskFromContext(ctx) - if err := t.Block(resCh); err != nil { - return 0, syserr.FromError(err).ToError() + if err := amutex.Block(ctx, resCh); err != nil { + return 0, err } - n, _, err = s.Endpoint.Write(f, tcpip.WriteOptions{}) } -- cgit v1.2.3 From 7b691ab73c7a3534e3351a5ca55a82e73ec63f75 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Tue, 12 May 2020 13:41:47 -0700 Subject: Don't allow rename across different gofer or tmpfs mounts. Fixes #2651. PiperOrigin-RevId: 311193661 --- pkg/sentry/fs/gofer/path.go | 24 +++++++++++------------- pkg/sentry/fs/tmpfs/tmpfs.go | 11 +++++------ 2 files changed, 16 insertions(+), 19 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/gofer/path.go b/pkg/sentry/fs/gofer/path.go index a35c3a23d..cf9800100 100644 --- a/pkg/sentry/fs/gofer/path.go +++ b/pkg/sentry/fs/gofer/path.go @@ -16,7 +16,6 @@ package gofer import ( "fmt" - "syscall" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" @@ -68,7 +67,7 @@ func (i *inodeOperations) Lookup(ctx context.Context, dir *fs.Inode, name string // Get a p9.File for name. qids, newFile, mask, p9attr, err := i.fileState.file.walkGetAttr(ctx, []string{name}) if err != nil { - if err == syscall.ENOENT { + if err == syserror.ENOENT { if cp.cacheNegativeDirents() { // Return a negative Dirent. It will stay cached until something // is created over it. @@ -207,7 +206,7 @@ func (i *inodeOperations) CreateHardLink(ctx context.Context, inode *fs.Inode, t targetOpts, ok := target.InodeOperations.(*inodeOperations) if !ok { - return syscall.EXDEV + return syserror.EXDEV } if err := i.fileState.file.link(ctx, &targetOpts.fileState.file, newName); err != nil { @@ -251,7 +250,7 @@ func (i *inodeOperations) Bind(ctx context.Context, dir *fs.Inode, name string, } if i.session().overrides == nil { - return nil, syscall.EOPNOTSUPP + return nil, syserror.EOPNOTSUPP } // Stabilize the override map while creation is in progress. @@ -280,7 +279,7 @@ func (i *inodeOperations) CreateFifo(ctx context.Context, dir *fs.Inode, name st // N.B. FIFOs use major/minor numbers 0. if _, err := i.fileState.file.mknod(ctx, name, mode, 0, 0, p9.UID(owner.UID), p9.GID(owner.GID)); err != nil { - if i.session().overrides == nil || err != syscall.EPERM { + if i.session().overrides == nil || err != syserror.EPERM { return err } // If gofer doesn't support mknod, check if we can create an internal fifo. @@ -427,17 +426,16 @@ func (i *inodeOperations) Rename(ctx context.Context, inode *fs.Inode, oldParent return syserror.ENAMETOOLONG } - // Unwrap the new parent to a *inodeOperations. - newParentInodeOperations, ok := newParent.InodeOperations.(*inodeOperations) - if !ok { - return syscall.EXDEV + // Don't allow renames across different mounts. + if newParent.MountSource != oldParent.MountSource { + return syserror.EXDEV } + // Unwrap the new parent to a *inodeOperations. + newParentInodeOperations := newParent.InodeOperations.(*inodeOperations) + // Unwrap the old parent to a *inodeOperations. - oldParentInodeOperations, ok := oldParent.InodeOperations.(*inodeOperations) - if !ok { - return syscall.EXDEV - } + oldParentInodeOperations := oldParent.InodeOperations.(*inodeOperations) // Do the rename. if err := i.fileState.file.rename(ctx, newParentInodeOperations.fileState.file, newName); err != nil { diff --git a/pkg/sentry/fs/tmpfs/tmpfs.go b/pkg/sentry/fs/tmpfs/tmpfs.go index 3c2b583ae..b095312fe 100644 --- a/pkg/sentry/fs/tmpfs/tmpfs.go +++ b/pkg/sentry/fs/tmpfs/tmpfs.go @@ -39,14 +39,13 @@ var fsInfo = fs.Info{ // rename implements fs.InodeOperations.Rename for tmpfs nodes. func rename(ctx context.Context, oldParent *fs.Inode, oldName string, newParent *fs.Inode, newName string, replacement bool) error { - op, ok := oldParent.InodeOperations.(*Dir) - if !ok { - return syserror.EXDEV - } - np, ok := newParent.InodeOperations.(*Dir) - if !ok { + // Don't allow renames across different mounts. + if newParent.MountSource != oldParent.MountSource { return syserror.EXDEV } + + op := oldParent.InodeOperations.(*Dir) + np := newParent.InodeOperations.(*Dir) return ramfs.Rename(ctx, op.ramfsDir, oldName, np.ramfsDir, newName, replacement) } -- cgit v1.2.3 From a19c8f0b9224dfd6f43dbb2b9d8efa3917d7a68f Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Wed, 13 May 2020 04:23:01 -0400 Subject: adding the methods to get/set TLS for Arm64 kvm platform Signed-off-by: Bin Lu --- pkg/sentry/platform/ring0/lib_arm64.go | 6 ++++++ pkg/sentry/platform/ring0/lib_arm64.s | 10 ++++++++++ 2 files changed, 16 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/ring0/lib_arm64.go b/pkg/sentry/platform/ring0/lib_arm64.go index 444a83913..a6345010d 100644 --- a/pkg/sentry/platform/ring0/lib_arm64.go +++ b/pkg/sentry/platform/ring0/lib_arm64.go @@ -38,6 +38,12 @@ func SaveVRegs(*byte) // LoadVRegs loads V0-V31 registers. func LoadVRegs(*byte) +// GetTLS returns the value of TPIDR_EL0 register. +func GetTLS() (value uint64) + +// SetTLS writes the TPIDR_EL0 value. +func SetTLS(value uint64) + // Init sets function pointers based on architectural features. // // This must be called prior to using ring0. diff --git a/pkg/sentry/platform/ring0/lib_arm64.s b/pkg/sentry/platform/ring0/lib_arm64.s index 0e6a6235b..b63e14b41 100644 --- a/pkg/sentry/platform/ring0/lib_arm64.s +++ b/pkg/sentry/platform/ring0/lib_arm64.s @@ -15,6 +15,16 @@ #include "funcdata.h" #include "textflag.h" +TEXT ·GetTLS(SB),NOSPLIT,$0-8 + MRS TPIDR_EL0, R1 + MOVD R1, ret+0(FP) + RET + +TEXT ·SetTLS(SB),NOSPLIT,$0-8 + MOVD addr+0(FP), R1 + MSR R1, TPIDR_EL0 + RET + TEXT ·CPACREL1(SB),NOSPLIT,$0-8 WORD $0xd5381041 // MRS CPACR_EL1, R1 MOVD R1, ret+0(FP) -- cgit v1.2.3 From 5fa480a877b4aa4633797e2fd8cd0364eebc7366 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Wed, 13 May 2020 04:39:25 -0400 Subject: PROT_NONE should be specially treated in the step of mapPhysical It's a workaround to treat PROT_NONE as RDONLY temporarily. TODO(gvisor.dev/issue/2686): PROT_NONE should be specially treated. Signed-off-by: Bin Lu --- pkg/sentry/platform/kvm/machine_arm64.go | 6 ++++++ 1 file changed, 6 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/kvm/machine_arm64.go b/pkg/sentry/platform/kvm/machine_arm64.go index e42505542..750751aa3 100644 --- a/pkg/sentry/platform/kvm/machine_arm64.go +++ b/pkg/sentry/platform/kvm/machine_arm64.go @@ -60,6 +60,12 @@ func rdonlyRegionsForSetMem() (phyRegions []physicalRegion) { if !vr.accessType.Write && vr.accessType.Read { rdonlyRegions = append(rdonlyRegions, vr.region) } + + // TODO(gvisor.dev/issue/2686): PROT_NONE should be specially treated. + // Workaround: treated as rdonly temporarily. + if !vr.accessType.Write && !vr.accessType.Read && !vr.accessType.Execute { + rdonlyRegions = append(rdonlyRegions, vr.region) + } }) for _, r := range rdonlyRegions { -- cgit v1.2.3 From d84607762889d999f93b89e64e09d2a4dda63bf7 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Wed, 13 May 2020 10:51:50 -0700 Subject: Enable overlayfs_stale_read by default for runsc. Linux 4.18 and later make reads and writes coherent between pre-copy-up and post-copy-up FDs representing the same file on an overlay filesystem. However, memory mappings remain incoherent: - Documentation/filesystems/overlayfs.rst, "Non-standard behavior": "If a file residing on a lower layer is opened for read-only and then memory mapped with MAP_SHARED, then subsequent changes to the file are not reflected in the memory mapping." - fs/overlay/file.c:ovl_mmap() passes through to the underlying FD without any management of coherence in the overlay. - Experimentally on Linux 5.2: ``` $ cat mmap_cat_page.c #include #include #include #include #include #include int main(int argc, char **argv) { if (argc < 2) { errx(1, "syntax: %s [FILE]", argv[0]); } const int fd = open(argv[1], O_RDONLY); if (fd < 0) { err(1, "open(%s)", argv[1]); } const size_t page_size = sysconf(_SC_PAGE_SIZE); void* page = mmap(NULL, page_size, PROT_READ, MAP_SHARED, fd, 0); if (page == MAP_FAILED) { err(1, "mmap"); } for (;;) { write(1, page, strnlen(page, page_size)); if (getc(stdin) == EOF) { break; } } return 0; } $ gcc -O2 -o mmap_cat_page mmap_cat_page.c $ mkdir lowerdir upperdir workdir overlaydir $ echo old > lowerdir/file $ sudo mount -t overlay -o "lowerdir=lowerdir,upperdir=upperdir,workdir=workdir" none overlaydir $ ./mmap_cat_page overlaydir/file old ^Z [1]+ Stopped ./mmap_cat_page overlaydir/file $ echo new > overlaydir/file $ cat overlaydir/file new $ fg ./mmap_cat_page overlaydir/file old ``` Therefore, while the VFS1 gofer client's behavior of reopening read FDs is only necessary pre-4.18, replacing existing memory mappings (in both sentry and application address spaces) with mappings of the new FD is required regardless of kernel version, and this latter behavior is common to both VFS1 and VFS2. Re-document accordingly, and change the runsc flag to enabled by default. New test: - Before this CL: https://source.cloud.google.com/results/invocations/5b222d2c-e918-4bae-afc4-407f5bac509b - After this CL: https://source.cloud.google.com/results/invocations/f28c747e-d89c-4d8c-a461-602b33e71aab PiperOrigin-RevId: 311361267 --- images/hostoverlaytest/Dockerfile | 7 +++ images/hostoverlaytest/test.c | 88 +++++++++++++++++++++++++++++++++++++ images/hostoverlaytest/testfile.txt | 1 + pkg/sentry/fs/gofer/fs.go | 3 +- pkg/sentry/fsimpl/gofer/gofer.go | 9 ++-- runsc/boot/config.go | 6 ++- runsc/container/container_test.go | 37 ---------------- runsc/main.go | 2 +- test/e2e/integration_test.go | 15 +++++++ 9 files changed, 123 insertions(+), 45 deletions(-) create mode 100644 images/hostoverlaytest/Dockerfile create mode 100644 images/hostoverlaytest/test.c create mode 100644 images/hostoverlaytest/testfile.txt (limited to 'pkg/sentry') diff --git a/images/hostoverlaytest/Dockerfile b/images/hostoverlaytest/Dockerfile new file mode 100644 index 000000000..d83439e9c --- /dev/null +++ b/images/hostoverlaytest/Dockerfile @@ -0,0 +1,7 @@ +FROM ubuntu:bionic + +WORKDIR /root +COPY . . + +RUN apt-get update && apt-get install -y gcc +RUN gcc -O2 -o test test.c diff --git a/images/hostoverlaytest/test.c b/images/hostoverlaytest/test.c new file mode 100644 index 000000000..088f90746 --- /dev/null +++ b/images/hostoverlaytest/test.c @@ -0,0 +1,88 @@ +#include +#include +#include +#include +#include +#include + +int main(int argc, char** argv) { + const char kTestFilePath[] = "testfile.txt"; + const char kOldFileData[] = "old data\n"; + const char kNewFileData[] = "new data\n"; + const size_t kPageSize = sysconf(_SC_PAGE_SIZE); + + // Open a file that already exists in a host overlayfs lower layer. + const int fd_rdonly = open(kTestFilePath, O_RDONLY); + if (fd_rdonly < 0) { + err(1, "open(%s, O_RDONLY)", kTestFilePath); + } + + // Check that the file's initial contents are what we expect when read via + // syscall. + char oldbuf[sizeof(kOldFileData)] = {}; + ssize_t n = pread(fd_rdonly, oldbuf, sizeof(oldbuf), 0); + if (n < 0) { + err(1, "initial pread"); + } + if (n != strlen(kOldFileData)) { + errx(1, "short initial pread (%ld/%lu bytes)", n, strlen(kOldFileData)); + } + if (strcmp(oldbuf, kOldFileData) != 0) { + errx(1, "initial pread returned wrong data: %s", oldbuf); + } + + // Check that the file's initial contents are what we expect when read via + // memory mapping. + void* page = mmap(NULL, kPageSize, PROT_READ, MAP_SHARED, fd_rdonly, 0); + if (page == MAP_FAILED) { + err(1, "mmap"); + } + if (strcmp(page, kOldFileData) != 0) { + errx(1, "mapping contains wrong initial data: %s", (const char*)page); + } + + // Open the same file writably, causing host overlayfs to copy it up, and + // replace its contents. + const int fd_rdwr = open(kTestFilePath, O_RDWR); + if (fd_rdwr < 0) { + err(1, "open(%s, O_RDWR)", kTestFilePath); + } + n = write(fd_rdwr, kNewFileData, strlen(kNewFileData)); + if (n < 0) { + err(1, "write"); + } + if (n != strlen(kNewFileData)) { + errx(1, "short write (%ld/%lu bytes)", n, strlen(kNewFileData)); + } + if (ftruncate(fd_rdwr, strlen(kNewFileData)) < 0) { + err(1, "truncate"); + } + + int failed = 0; + + // Check that syscalls on the old FD return updated contents. (Before Linux + // 4.18, this requires that runsc use a post-copy-up FD to service the read.) + char newbuf[sizeof(kNewFileData)] = {}; + n = pread(fd_rdonly, newbuf, sizeof(newbuf), 0); + if (n < 0) { + err(1, "final pread"); + } + if (n != strlen(kNewFileData)) { + warnx("short final pread (%ld/%lu bytes)", n, strlen(kNewFileData)); + failed = 1; + } else if (strcmp(newbuf, kNewFileData) != 0) { + warnx("final pread returned wrong data: %s", newbuf); + failed = 1; + } + + // Check that the memory mapping of the old FD has been updated. (Linux + // overlayfs does not do this, so regardless of kernel version this requires + // that runsc replace existing memory mappings with mappings of a + // post-copy-up FD.) + if (strcmp(page, kNewFileData) != 0) { + warnx("mapping contains wrong final data: %s", (const char*)page); + failed = 1; + } + + return failed; +} diff --git a/images/hostoverlaytest/testfile.txt b/images/hostoverlaytest/testfile.txt new file mode 100644 index 000000000..e4188c841 --- /dev/null +++ b/images/hostoverlaytest/testfile.txt @@ -0,0 +1 @@ +old data diff --git a/pkg/sentry/fs/gofer/fs.go b/pkg/sentry/fs/gofer/fs.go index 9d41fcbdb..8ae2d78d7 100644 --- a/pkg/sentry/fs/gofer/fs.go +++ b/pkg/sentry/fs/gofer/fs.go @@ -60,8 +60,7 @@ const ( limitHostFDTranslationKey = "limit_host_fd_translation" // overlayfsStaleRead if present closes cached readonly file after the first - // write. This is done to workaround a limitation of overlayfs in kernels - // before 4.19 where open FDs are not updated after the file is copied up. + // write. This is done to workaround a limitation of Linux overlayfs. overlayfsStaleRead = "overlayfs_stale_read" ) diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 1da8d5d82..353e2cf5b 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -143,9 +143,12 @@ type filesystemOptions struct { // If overlayfsStaleRead is true, O_RDONLY host FDs provided by the remote // filesystem may not be coherent with writable host FDs opened later, so - // mappings of the former must be replaced by mappings of the latter. This - // is usually only the case when the remote filesystem is an overlayfs - // mount on Linux < 4.19. + // all uses of the former must be replaced by uses of the latter. This is + // usually only the case when the remote filesystem is a Linux overlayfs + // mount. (Prior to Linux 4.18, patch series centered on commit + // d1d04ef8572b "ovl: stack file ops", both I/O and memory mappings were + // incoherent between pre-copy-up and post-copy-up FDs; after that patch + // series, only memory mappings are incoherent.) overlayfsStaleRead bool // If regularFilesUseSpecialFileFD is true, application FDs representing diff --git a/runsc/boot/config.go b/runsc/boot/config.go index 6d6a705f8..bcec7e4db 100644 --- a/runsc/boot/config.go +++ b/runsc/boot/config.go @@ -241,8 +241,10 @@ type Config struct { // ReferenceLeakMode sets reference leak check mode ReferenceLeakMode refs.LeakMode - // OverlayfsStaleRead causes cached FDs to reopen after a file is opened for - // write to workaround overlayfs limitation on kernels before 4.19. + // OverlayfsStaleRead instructs the sandbox to assume that the root mount + // is on a Linux overlayfs mount, which does not necessarily preserve + // coherence between read-only and subsequent writable file descriptors + // representing the "same" file. OverlayfsStaleRead bool // TestOnlyAllowRunAsCurrentUserWithoutChroot should only be used in diff --git a/runsc/container/container_test.go b/runsc/container/container_test.go index acf988aa0..7ba301331 100644 --- a/runsc/container/container_test.go +++ b/runsc/container/container_test.go @@ -2126,43 +2126,6 @@ func TestNetRaw(t *testing.T) { } } -// TestOverlayfsStaleRead most basic test that '--overlayfs-stale-read' works. -func TestOverlayfsStaleRead(t *testing.T) { - conf := testutil.TestConfig(t) - conf.OverlayfsStaleRead = true - - in, err := ioutil.TempFile(testutil.TmpDir(), "stale-read.in") - if err != nil { - t.Fatalf("ioutil.TempFile() failed: %v", err) - } - defer in.Close() - if _, err := in.WriteString("stale data"); err != nil { - t.Fatalf("in.Write() failed: %v", err) - } - - out, err := ioutil.TempFile(testutil.TmpDir(), "stale-read.out") - if err != nil { - t.Fatalf("ioutil.TempFile() failed: %v", err) - } - defer out.Close() - - const want = "foobar" - cmd := fmt.Sprintf("cat %q >&2 && echo %q> %q && cp %q %q", in.Name(), want, in.Name(), in.Name(), out.Name()) - spec := testutil.NewSpecWithArgs("/bin/bash", "-c", cmd) - if err := run(spec, conf); err != nil { - t.Fatalf("Error running container: %v", err) - } - - gotBytes, err := ioutil.ReadAll(out) - if err != nil { - t.Fatalf("out.Read() failed: %v", err) - } - got := strings.TrimSpace(string(gotBytes)) - if want != got { - t.Errorf("Wrong content in out file, got: %q. want: %q", got, want) - } -} - // TestTTYField checks TTY field returned by container.Processes(). func TestTTYField(t *testing.T) { stop := testutil.StartReaper() diff --git a/runsc/main.go b/runsc/main.go index 0625a06e0..920ed84a5 100644 --- a/runsc/main.go +++ b/runsc/main.go @@ -76,7 +76,7 @@ var ( fileAccess = flag.String("file-access", "exclusive", "specifies which filesystem to use for the root mount: exclusive (default), shared. Volume mounts are always shared.") fsGoferHostUDS = flag.Bool("fsgofer-host-uds", false, "allow the gofer to mount Unix Domain Sockets.") overlay = flag.Bool("overlay", false, "wrap filesystem mounts with writable overlay. All modifications are stored in memory inside the sandbox.") - overlayfsStaleRead = flag.Bool("overlayfs-stale-read", false, "reopen cached FDs after a file is opened for write to workaround overlayfs limitation on kernels before 4.19.") + overlayfsStaleRead = flag.Bool("overlayfs-stale-read", true, "assume root mount is an overlay filesystem") watchdogAction = flag.String("watchdog-action", "log", "sets what action the watchdog takes when triggered: log (default), panic.") panicSignal = flag.Int("panic-signal", -1, "register signal handling that panics. Usually set to SIGUSR2(12) to troubleshoot hangs. -1 disables it.") profile = flag.Bool("profile", false, "prepares the sandbox to use Golang profiler. Note that enabling profiler loosens the seccomp protection added to the sandbox (DO NOT USE IN PRODUCTION).") diff --git a/test/e2e/integration_test.go b/test/e2e/integration_test.go index 404e37689..ff856883a 100644 --- a/test/e2e/integration_test.go +++ b/test/e2e/integration_test.go @@ -361,6 +361,21 @@ func TestTmpFile(t *testing.T) { } } +// TestHostOverlayfsCopyUp tests that the --overlayfs-stale-read option causes +// runsc to hide the incoherence of FDs opened before and after overlayfs +// copy-up on the host. +func TestHostOverlayfsCopyUp(t *testing.T) { + d := dockerutil.MakeDocker(t) + defer d.CleanUp() + + if _, err := d.Run(dockerutil.RunOpts{ + Image: "hostoverlaytest", + WorkDir: "/root", + }, "./test"); err != nil { + t.Fatalf("docker run failed: %v", err) + } +} + func TestMain(m *testing.M) { dockerutil.EnsureSupportedDockerVersion() flag.Parse() -- cgit v1.2.3 From db655f020ea556524f0e341e538e81c16d4f95e7 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Wed, 13 May 2020 17:35:04 -0700 Subject: Resolve remaining TODOs for tmpfs. Closes #1197 PiperOrigin-RevId: 311438223 --- pkg/sentry/fsimpl/tmpfs/BUILD | 1 + pkg/sentry/fsimpl/tmpfs/filesystem.go | 25 +++-- pkg/sentry/fsimpl/tmpfs/regular_file_test.go | 136 ----------------------- pkg/sentry/fsimpl/tmpfs/stat_test.go | 2 - pkg/sentry/fsimpl/tmpfs/tmpfs_test.go | 156 +++++++++++++++++++++++++++ test/syscalls/linux/BUILD | 1 + test/syscalls/linux/socket.cc | 9 +- test/syscalls/linux/symlink.cc | 25 +++++ 8 files changed, 208 insertions(+), 147 deletions(-) create mode 100644 pkg/sentry/fsimpl/tmpfs/tmpfs_test.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD index a2d9649e7..9a076ad71 100644 --- a/pkg/sentry/fsimpl/tmpfs/BUILD +++ b/pkg/sentry/fsimpl/tmpfs/BUILD @@ -96,6 +96,7 @@ go_test( "pipe_test.go", "regular_file_test.go", "stat_test.go", + "tmpfs_test.go", ], library = ":tmpfs", deps = [ diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 36ffcb592..e0ad82769 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -16,6 +16,7 @@ package tmpfs import ( "fmt" + "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" @@ -24,6 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // Sync implements vfs.FilesystemImpl.Sync. @@ -76,8 +78,8 @@ afterSymlink: return nil, err } if symlink, ok := child.inode.impl.(*symlink); ok && rp.ShouldFollowSymlink() { - // TODO(gvisor.dev/issue/1197): Symlink traversals updates - // access time. + // Symlink traversal updates access time. + atomic.StoreInt64(&d.inode.atime, d.inode.fs.clock.Now().Nanoseconds()) if err := rp.HandleSymlink(symlink.target); err != nil { return nil, err } @@ -361,8 +363,8 @@ afterTrailingSymlink: } // Do we need to resolve a trailing symlink? if symlink, ok := child.inode.impl.(*symlink); ok && rp.ShouldFollowSymlink() { - // TODO(gvisor.dev/issue/1197): Symlink traversals updates - // access time. + // Symlink traversal updates access time. + atomic.StoreInt64(&child.inode.atime, child.inode.fs.clock.Now().Nanoseconds()) if err := rp.HandleSymlink(symlink.target); err != nil { return nil, err } @@ -636,12 +638,19 @@ func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { fs.mu.RLock() defer fs.mu.RUnlock() - _, err := resolveLocked(rp) - if err != nil { + if _, err := resolveLocked(rp); err != nil { return linux.Statfs{}, err } - // TODO(gvisor.dev/issue/1197): Actually implement statfs. - return linux.Statfs{}, syserror.ENOSYS + statfs := linux.Statfs{ + Type: linux.TMPFS_MAGIC, + BlockSize: usermem.PageSize, + FragmentSize: usermem.PageSize, + NameLength: linux.NAME_MAX, + // TODO(b/29637826): Allow configuring a tmpfs size and enforce it. + Blocks: 0, + BlocksFree: 0, + } + return statfs, nil } // SymlinkAt implements vfs.FilesystemImpl.SymlinkAt. diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go index 0399725cf..f2bc96d51 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go @@ -18,152 +18,16 @@ import ( "bytes" "fmt" "io" - "sync/atomic" "testing" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/fs/lock" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" ) -// nextFileID is used to generate unique file names. -var nextFileID int64 - -// newTmpfsRoot creates a new tmpfs mount, and returns the root. If the error -// is not nil, then cleanup should be called when the root is no longer needed. -func newTmpfsRoot(ctx context.Context) (*vfs.VirtualFilesystem, vfs.VirtualDentry, func(), error) { - creds := auth.CredentialsFromContext(ctx) - - vfsObj := &vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { - return nil, vfs.VirtualDentry{}, nil, fmt.Errorf("VFS init: %v", err) - } - - vfsObj.MustRegisterFilesystemType("tmpfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{}) - if err != nil { - return nil, vfs.VirtualDentry{}, nil, fmt.Errorf("failed to create tmpfs root mount: %v", err) - } - root := mntns.Root() - return vfsObj, root, func() { - root.DecRef() - mntns.DecRef() - }, nil -} - -// newFileFD creates a new file in a new tmpfs mount, and returns the FD. If -// the returned err is not nil, then cleanup should be called when the FD is no -// longer needed. -func newFileFD(ctx context.Context, mode linux.FileMode) (*vfs.FileDescription, func(), error) { - creds := auth.CredentialsFromContext(ctx) - vfsObj, root, cleanup, err := newTmpfsRoot(ctx) - if err != nil { - return nil, nil, err - } - - filename := fmt.Sprintf("tmpfs-test-file-%d", atomic.AddInt64(&nextFileID, 1)) - - // Create the file that will be write/read. - fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(filename), - }, &vfs.OpenOptions{ - Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL, - Mode: linux.ModeRegular | mode, - }) - if err != nil { - cleanup() - return nil, nil, fmt.Errorf("failed to create file %q: %v", filename, err) - } - - return fd, cleanup, nil -} - -// newDirFD is like newFileFD, but for directories. -func newDirFD(ctx context.Context, mode linux.FileMode) (*vfs.FileDescription, func(), error) { - creds := auth.CredentialsFromContext(ctx) - vfsObj, root, cleanup, err := newTmpfsRoot(ctx) - if err != nil { - return nil, nil, err - } - - dirname := fmt.Sprintf("tmpfs-test-dir-%d", atomic.AddInt64(&nextFileID, 1)) - - // Create the dir. - if err := vfsObj.MkdirAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(dirname), - }, &vfs.MkdirOptions{ - Mode: linux.ModeDirectory | mode, - }); err != nil { - cleanup() - return nil, nil, fmt.Errorf("failed to create directory %q: %v", dirname, err) - } - - // Open the dir and return it. - fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(dirname), - }, &vfs.OpenOptions{ - Flags: linux.O_RDONLY | linux.O_DIRECTORY, - }) - if err != nil { - cleanup() - return nil, nil, fmt.Errorf("failed to open directory %q: %v", dirname, err) - } - - return fd, cleanup, nil -} - -// newPipeFD is like newFileFD, but for pipes. -func newPipeFD(ctx context.Context, mode linux.FileMode) (*vfs.FileDescription, func(), error) { - creds := auth.CredentialsFromContext(ctx) - vfsObj, root, cleanup, err := newTmpfsRoot(ctx) - if err != nil { - return nil, nil, err - } - - pipename := fmt.Sprintf("tmpfs-test-pipe-%d", atomic.AddInt64(&nextFileID, 1)) - - // Create the pipe. - if err := vfsObj.MknodAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(pipename), - }, &vfs.MknodOptions{ - Mode: linux.ModeNamedPipe | mode, - }); err != nil { - cleanup() - return nil, nil, fmt.Errorf("failed to create pipe %q: %v", pipename, err) - } - - // Open the pipe and return it. - fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(pipename), - }, &vfs.OpenOptions{ - Flags: linux.O_RDWR, - }) - if err != nil { - cleanup() - return nil, nil, fmt.Errorf("failed to open pipe %q: %v", pipename, err) - } - - return fd, cleanup, nil -} - // Test that we can write some data to a file and read it back.` func TestSimpleWriteRead(t *testing.T) { ctx := contexttest.Context(t) diff --git a/pkg/sentry/fsimpl/tmpfs/stat_test.go b/pkg/sentry/fsimpl/tmpfs/stat_test.go index 60c2c980e..f52755092 100644 --- a/pkg/sentry/fsimpl/tmpfs/stat_test.go +++ b/pkg/sentry/fsimpl/tmpfs/stat_test.go @@ -29,7 +29,6 @@ func TestStatAfterCreate(t *testing.T) { mode := linux.FileMode(0644) // Run with different file types. - // TODO(gvisor.dev/issue/1197): Also test symlinks and sockets. for _, typ := range []string{"file", "dir", "pipe"} { t.Run(fmt.Sprintf("type=%q", typ), func(t *testing.T) { var ( @@ -175,7 +174,6 @@ func TestSetStat(t *testing.T) { mode := linux.FileMode(0644) // Run with different file types. - // TODO(gvisor.dev/issue/1197): Also test symlinks and sockets. for _, typ := range []string{"file", "dir", "pipe"} { t.Run(fmt.Sprintf("type=%q", typ), func(t *testing.T) { var ( diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs_test.go b/pkg/sentry/fsimpl/tmpfs/tmpfs_test.go new file mode 100644 index 000000000..a240fb276 --- /dev/null +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs_test.go @@ -0,0 +1,156 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tmpfs + +import ( + "fmt" + "sync/atomic" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" +) + +// nextFileID is used to generate unique file names. +var nextFileID int64 + +// newTmpfsRoot creates a new tmpfs mount, and returns the root. If the error +// is not nil, then cleanup should be called when the root is no longer needed. +func newTmpfsRoot(ctx context.Context) (*vfs.VirtualFilesystem, vfs.VirtualDentry, func(), error) { + creds := auth.CredentialsFromContext(ctx) + + vfsObj := &vfs.VirtualFilesystem{} + if err := vfsObj.Init(); err != nil { + return nil, vfs.VirtualDentry{}, nil, fmt.Errorf("VFS init: %v", err) + } + + vfsObj.MustRegisterFilesystemType("tmpfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + }) + mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{}) + if err != nil { + return nil, vfs.VirtualDentry{}, nil, fmt.Errorf("failed to create tmpfs root mount: %v", err) + } + root := mntns.Root() + return vfsObj, root, func() { + root.DecRef() + mntns.DecRef() + }, nil +} + +// newFileFD creates a new file in a new tmpfs mount, and returns the FD. If +// the returned err is not nil, then cleanup should be called when the FD is no +// longer needed. +func newFileFD(ctx context.Context, mode linux.FileMode) (*vfs.FileDescription, func(), error) { + creds := auth.CredentialsFromContext(ctx) + vfsObj, root, cleanup, err := newTmpfsRoot(ctx) + if err != nil { + return nil, nil, err + } + + filename := fmt.Sprintf("tmpfs-test-file-%d", atomic.AddInt64(&nextFileID, 1)) + + // Create the file that will be write/read. + fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(filename), + }, &vfs.OpenOptions{ + Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL, + Mode: linux.ModeRegular | mode, + }) + if err != nil { + cleanup() + return nil, nil, fmt.Errorf("failed to create file %q: %v", filename, err) + } + + return fd, cleanup, nil +} + +// newDirFD is like newFileFD, but for directories. +func newDirFD(ctx context.Context, mode linux.FileMode) (*vfs.FileDescription, func(), error) { + creds := auth.CredentialsFromContext(ctx) + vfsObj, root, cleanup, err := newTmpfsRoot(ctx) + if err != nil { + return nil, nil, err + } + + dirname := fmt.Sprintf("tmpfs-test-dir-%d", atomic.AddInt64(&nextFileID, 1)) + + // Create the dir. + if err := vfsObj.MkdirAt(ctx, creds, &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(dirname), + }, &vfs.MkdirOptions{ + Mode: linux.ModeDirectory | mode, + }); err != nil { + cleanup() + return nil, nil, fmt.Errorf("failed to create directory %q: %v", dirname, err) + } + + // Open the dir and return it. + fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(dirname), + }, &vfs.OpenOptions{ + Flags: linux.O_RDONLY | linux.O_DIRECTORY, + }) + if err != nil { + cleanup() + return nil, nil, fmt.Errorf("failed to open directory %q: %v", dirname, err) + } + + return fd, cleanup, nil +} + +// newPipeFD is like newFileFD, but for pipes. +func newPipeFD(ctx context.Context, mode linux.FileMode) (*vfs.FileDescription, func(), error) { + creds := auth.CredentialsFromContext(ctx) + vfsObj, root, cleanup, err := newTmpfsRoot(ctx) + if err != nil { + return nil, nil, err + } + + name := fmt.Sprintf("tmpfs-test-%d", atomic.AddInt64(&nextFileID, 1)) + + if err := vfsObj.MknodAt(ctx, creds, &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(name), + }, &vfs.MknodOptions{ + Mode: linux.ModeNamedPipe | mode, + }); err != nil { + cleanup() + return nil, nil, fmt.Errorf("failed to create pipe %q: %v", name, err) + } + + fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(name), + }, &vfs.OpenOptions{ + Flags: linux.O_RDWR, + }) + if err != nil { + cleanup() + return nil, nil, fmt.Errorf("failed to open pipe %q: %v", name, err) + } + + return fd, cleanup, nil +} diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index adf259bba..e6bc63474 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -3288,6 +3288,7 @@ cc_binary( "//test/util:capability_util", "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/time", gtest, "//test/util:temp_path", "//test/util:test_main", diff --git a/test/syscalls/linux/socket.cc b/test/syscalls/linux/socket.cc index 703d594a2..afa59c1da 100644 --- a/test/syscalls/linux/socket.cc +++ b/test/syscalls/linux/socket.cc @@ -61,7 +61,7 @@ TEST(SocketTest, ProtocolInet) { } } -TEST(SocketTest, UnixSocketFileMode) { +TEST(SocketTest, UnixSocketStat) { // TODO(gvisor.dev/issue/1624): Re-enable this test once VFS1 is deleted. It // should pass in VFS2. SKIP_IF(IsRunningOnGvisor()); @@ -83,7 +83,14 @@ TEST(SocketTest, UnixSocketFileMode) { struct stat statbuf = {}; ASSERT_THAT(stat(addr.sun_path, &statbuf), SyscallSucceeds()); + + // Mode should be S_IFSOCK. EXPECT_EQ(statbuf.st_mode, S_IFSOCK | sock_perm & ~mask); + + // Timestamps should be equal and non-zero. + EXPECT_NE(statbuf.st_atime, 0); + EXPECT_EQ(statbuf.st_atime, statbuf.st_mtime); + EXPECT_EQ(statbuf.st_atime, statbuf.st_ctime); } TEST(SocketTest, UnixConnectNeedsWritePerm) { diff --git a/test/syscalls/linux/symlink.cc b/test/syscalls/linux/symlink.cc index 03ee1250d..a17ff62e9 100644 --- a/test/syscalls/linux/symlink.cc +++ b/test/syscalls/linux/symlink.cc @@ -20,6 +20,7 @@ #include #include "gtest/gtest.h" +#include "absl/time/clock.h" #include "test/util/capability_util.h" #include "test/util/file_descriptor.h" #include "test/util/fs_util.h" @@ -272,6 +273,30 @@ TEST(SymlinkTest, ChmodSymlink) { EXPECT_EQ(FilePermission(newpath), 0777); } +// Test that following a symlink updates the atime on the symlink. +TEST(SymlinkTest, FollowUpdatesATime) { + const auto file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); + const std::string link = NewTempAbsPath(); + EXPECT_THAT(symlink(file.path().c_str(), link.c_str()), SyscallSucceeds()); + + // Lstat the symlink. + struct stat st_before_follow; + ASSERT_THAT(lstat(link.c_str(), &st_before_follow), SyscallSucceeds()); + + // Let the clock advance. + absl::SleepFor(absl::Seconds(1)); + + // Open the file via the symlink. + int fd; + ASSERT_THAT(fd = open(link.c_str(), O_RDWR, 0666), SyscallSucceeds()); + FileDescriptor fd_closer(fd); + + // Lstat the symlink again, and check that atime is updated. + struct stat st_after_follow; + ASSERT_THAT(lstat(link.c_str(), &st_after_follow), SyscallSucceeds()); + EXPECT_LT(st_before_follow.st_atime, st_after_follow.st_atime); +} + class ParamSymlinkTest : public ::testing::TestWithParam {}; // Test that creating an existing symlink with creat will create the target. -- cgit v1.2.3 From 8b8774d7152eb61fc6273bbae679e80c34188517 Mon Sep 17 00:00:00 2001 From: Bhasker Hariharan Date: Wed, 13 May 2020 19:47:42 -0700 Subject: Stub support for TCP_SYNCNT and TCP_WINDOW_CLAMP. This change adds support for TCP_SYNCNT and TCP_WINDOW_CLAMP options in GetSockOpt/SetSockOpt. This change does not really change any behaviour in Netstack and only stores/returns the stored value. Actual honoring of these options will be added as required. Fixes #2626, #2625 PiperOrigin-RevId: 311453777 --- pkg/sentry/socket/netstack/netstack.go | 39 ++++++ pkg/tcpip/tcpip.go | 17 +++ pkg/tcpip/transport/tcp/endpoint.go | 64 ++++++++- pkg/tcpip/transport/tcp/protocol.go | 21 +++ test/syscalls/linux/socket_ip_tcp_generic.cc | 45 +++++++ test/syscalls/linux/tcp_socket.cc | 185 ++++++++++++++++++++++++++- 6 files changed, 367 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 9d032f052..9dea2b5ff 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -1321,6 +1321,29 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa return int32(time.Duration(v) / time.Second), nil + case linux.TCP_SYNCNT: + if outLen < sizeOfInt32 { + return nil, syserr.ErrInvalidArgument + } + + v, err := ep.GetSockOptInt(tcpip.TCPSynCountOption) + if err != nil { + return nil, syserr.TranslateNetstackError(err) + } + + return int32(v), nil + + case linux.TCP_WINDOW_CLAMP: + if outLen < sizeOfInt32 { + return nil, syserr.ErrInvalidArgument + } + + v, err := ep.GetSockOptInt(tcpip.TCPWindowClampOption) + if err != nil { + return nil, syserr.TranslateNetstackError(err) + } + + return int32(v), nil default: emitUnimplementedEventTCP(t, name) } @@ -1790,6 +1813,22 @@ func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) * } return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.TCPDeferAcceptOption(time.Second * time.Duration(v)))) + case linux.TCP_SYNCNT: + if len(optVal) < sizeOfInt32 { + return syserr.ErrInvalidArgument + } + v := usermem.ByteOrder.Uint32(optVal) + + return syserr.TranslateNetstackError(ep.SetSockOptInt(tcpip.TCPSynCountOption, int(v))) + + case linux.TCP_WINDOW_CLAMP: + if len(optVal) < sizeOfInt32 { + return syserr.ErrInvalidArgument + } + v := usermem.ByteOrder.Uint32(optVal) + + return syserr.TranslateNetstackError(ep.SetSockOptInt(tcpip.TCPWindowClampOption, int(v))) + case linux.TCP_REPAIR_OPTIONS: t.Kernel().EmitUnimplementedEvent(t) diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 1ca4088c9..6270146fa 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -622,6 +622,19 @@ const ( // // A zero value indicates the default. TTLOption + + // TCPSynCountOption is used by SetSockOpt/GetSockOpt to specify the number of + // SYN retransmits that TCP should send before aborting the attempt to + // connect. It cannot exceed 255. + // + // NOTE: This option is currently only stubbed out and is no-op. + TCPSynCountOption + + // TCPWindowClampOption is used by SetSockOpt/GetSockOpt to bound the size + // of the advertised window to this value. + // + // NOTE: This option is currently only stubed out and is a no-op + TCPWindowClampOption ) // ErrorOption is used in GetSockOpt to specify that the last error reported by @@ -690,6 +703,10 @@ type TCPMinRTOOption time.Duration // switches to using SYN cookies. type TCPSynRcvdCountThresholdOption uint64 +// TCPSynRetriesOption is used by SetSockOpt/GetSockOpt to specify stack-wide +// default for number of times SYN is retransmitted before aborting a connect. +type TCPSynRetriesOption uint8 + // MulticastInterfaceOption is used by SetSockOpt/GetSockOpt to specify a // default interface for multicast. type MulticastInterfaceOption struct { diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 07d3e64c8..71735029e 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -470,6 +470,17 @@ type endpoint struct { // for this endpoint using the TCP_MAXSEG setsockopt. userMSS uint16 + // maxSynRetries is the maximum number of SYN retransmits that TCP should + // send before aborting the attempt to connect. It cannot exceed 255. + // + // NOTE: This is currently a no-op and does not change the SYN + // retransmissions. + maxSynRetries uint8 + + // windowClamp is used to bound the size of the advertised window to + // this value. + windowClamp uint32 + // The following fields are used to manage the send buffer. When // segments are ready to be sent, they are added to sndQueue and the // protocol goroutine is signaled via sndWaker. @@ -795,8 +806,10 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue interval: 75 * time.Second, count: 9, }, - uniqueID: s.UniqueID(), - txHash: s.Rand().Uint32(), + uniqueID: s.UniqueID(), + txHash: s.Rand().Uint32(), + windowClamp: DefaultReceiveBufferSize, + maxSynRetries: DefaultSynRetries, } var ss SendBufferSizeOption @@ -829,6 +842,11 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue e.tcpLingerTimeout = time.Duration(tcpLT) } + var synRetries tcpip.TCPSynRetriesOption + if err := s.TransportProtocolOption(ProtocolNumber, &synRetries); err == nil { + e.maxSynRetries = uint8(synRetries) + } + if p := s.GetTCPProbe(); p != nil { e.probe = p } @@ -1603,6 +1621,36 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { e.ttl = uint8(v) e.UnlockUser() + case tcpip.TCPSynCountOption: + if v < 1 || v > 255 { + return tcpip.ErrInvalidOptionValue + } + e.LockUser() + e.maxSynRetries = uint8(v) + e.UnlockUser() + + case tcpip.TCPWindowClampOption: + if v == 0 { + e.LockUser() + switch e.EndpointState() { + case StateClose, StateInitial: + e.windowClamp = 0 + e.UnlockUser() + return nil + default: + e.UnlockUser() + return tcpip.ErrInvalidOptionValue + } + } + var rs ReceiveBufferSizeOption + if err := e.stack.TransportProtocolOption(ProtocolNumber, &rs); err == nil { + if v < rs.Min/2 { + v = rs.Min / 2 + } + } + e.LockUser() + e.windowClamp = uint32(v) + e.UnlockUser() } return nil } @@ -1826,6 +1874,18 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { e.UnlockUser() return v, nil + case tcpip.TCPSynCountOption: + e.LockUser() + v := int(e.maxSynRetries) + e.UnlockUser() + return v, nil + + case tcpip.TCPWindowClampOption: + e.LockUser() + v := int(e.windowClamp) + e.UnlockUser() + return v, nil + default: return -1, tcpip.ErrUnknownProtocolOption } diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go index cfd9a4e8e..1e64283b5 100644 --- a/pkg/tcpip/transport/tcp/protocol.go +++ b/pkg/tcpip/transport/tcp/protocol.go @@ -64,6 +64,10 @@ const ( // DefaultTCPTimeWaitTimeout is the amount of time that sockets linger // in TIME_WAIT state before being marked closed. DefaultTCPTimeWaitTimeout = 60 * time.Second + + // DefaultSynRetries is the default value for the number of SYN retransmits + // before a connect is aborted. + DefaultSynRetries = 6 ) // SACKEnabled option can be used to enable SACK support in the TCP @@ -164,6 +168,7 @@ type protocol struct { tcpTimeWaitTimeout time.Duration minRTO time.Duration synRcvdCount synRcvdCounter + synRetries uint8 dispatcher *dispatcher } @@ -346,6 +351,15 @@ func (p *protocol) SetOption(option interface{}) *tcpip.Error { p.mu.Unlock() return nil + case tcpip.TCPSynRetriesOption: + if v < 1 || v > 255 { + return tcpip.ErrInvalidOptionValue + } + p.mu.Lock() + p.synRetries = uint8(v) + p.mu.Unlock() + return nil + default: return tcpip.ErrUnknownProtocolOption } @@ -420,6 +434,12 @@ func (p *protocol) Option(option interface{}) *tcpip.Error { p.mu.RUnlock() return nil + case *tcpip.TCPSynRetriesOption: + p.mu.RLock() + *v = tcpip.TCPSynRetriesOption(p.synRetries) + p.mu.RUnlock() + return nil + default: return tcpip.ErrUnknownProtocolOption } @@ -452,6 +472,7 @@ func NewProtocol() stack.TransportProtocol { tcpTimeWaitTimeout: DefaultTCPTimeWaitTimeout, synRcvdCount: synRcvdCounter{threshold: SynRcvdCountThreshold}, dispatcher: newDispatcher(runtime.GOMAXPROCS(0)), + synRetries: DefaultSynRetries, minRTO: MinRTO, } } diff --git a/test/syscalls/linux/socket_ip_tcp_generic.cc b/test/syscalls/linux/socket_ip_tcp_generic.cc index 27779e47c..fa81845fd 100644 --- a/test/syscalls/linux/socket_ip_tcp_generic.cc +++ b/test/syscalls/linux/socket_ip_tcp_generic.cc @@ -876,6 +876,51 @@ TEST_P(TCPSocketPairTest, SetTCPUserTimeoutAboveZero) { EXPECT_EQ(get, kAbove); } +TEST_P(TCPSocketPairTest, SetTCPWindowClampBelowMinRcvBufConnectedSocket) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + // Discover minimum receive buf by setting a really low value + // for the receive buffer. + constexpr int kZero = 0; + EXPECT_THAT(setsockopt(sockets->first_fd(), SOL_SOCKET, SO_RCVBUF, &kZero, + sizeof(kZero)), + SyscallSucceeds()); + + // Now retrieve the minimum value for SO_RCVBUF as the set above should + // have caused SO_RCVBUF for the socket to be set to the minimum. + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT( + getsockopt(sockets->first_fd(), SOL_SOCKET, SO_RCVBUF, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + int min_so_rcvbuf = get; + + { + // Setting TCP_WINDOW_CLAMP to zero for a connected socket is not permitted. + constexpr int kZero = 0; + EXPECT_THAT(setsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_WINDOW_CLAMP, + &kZero, sizeof(kZero)), + SyscallFailsWithErrno(EINVAL)); + + // Non-zero clamp values below MIN_SO_RCVBUF/2 should result in the clamp + // being set to MIN_SO_RCVBUF/2. + int below_half_min_so_rcvbuf = min_so_rcvbuf / 2 - 1; + EXPECT_THAT( + setsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_WINDOW_CLAMP, + &below_half_min_so_rcvbuf, sizeof(below_half_min_so_rcvbuf)), + SyscallSucceeds()); + + int get = -1; + socklen_t get_len = sizeof(get); + + ASSERT_THAT(getsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_WINDOW_CLAMP, + &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(min_so_rcvbuf / 2, get); + } +} + TEST_P(TCPSocketPairTest, TCPResetDuringClose_NoRandomSave) { DisableSave ds; // Too many syscalls. constexpr int kThreadCount = 1000; diff --git a/test/syscalls/linux/tcp_socket.cc b/test/syscalls/linux/tcp_socket.cc index d9c1ac0e1..a4d2953e1 100644 --- a/test/syscalls/linux/tcp_socket.cc +++ b/test/syscalls/linux/tcp_socket.cc @@ -1313,7 +1313,7 @@ TEST_P(SimpleTcpSocketTest, SetTCPDeferAcceptNeg) { int get = -1; socklen_t get_len = sizeof(get); ASSERT_THAT( - getsockopt(s.get(), IPPROTO_TCP, TCP_USER_TIMEOUT, &get, &get_len), + getsockopt(s.get(), IPPROTO_TCP, TCP_DEFER_ACCEPT, &get, &get_len), SyscallSucceedsWithValue(0)); EXPECT_EQ(get_len, sizeof(get)); EXPECT_EQ(get, 0); @@ -1326,7 +1326,7 @@ TEST_P(SimpleTcpSocketTest, GetTCPDeferAcceptDefault) { int get = -1; socklen_t get_len = sizeof(get); ASSERT_THAT( - getsockopt(s.get(), IPPROTO_TCP, TCP_USER_TIMEOUT, &get, &get_len), + getsockopt(s.get(), IPPROTO_TCP, TCP_DEFER_ACCEPT, &get, &get_len), SyscallSucceedsWithValue(0)); EXPECT_EQ(get_len, sizeof(get)); EXPECT_EQ(get, 0); @@ -1378,6 +1378,187 @@ TEST_P(SimpleTcpSocketTest, TCPConnectSoRcvBufRace) { SyscallSucceedsWithValue(0)); } +TEST_P(SimpleTcpSocketTest, SetTCPSynCntLessThanOne) { + FileDescriptor s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT(getsockopt(s.get(), IPPROTO_TCP, TCP_SYNCNT, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + int default_syn_cnt = get; + + { + // TCP_SYNCNT less than 1 should be rejected with an EINVAL. + constexpr int kZero = 0; + EXPECT_THAT( + setsockopt(s.get(), IPPROTO_TCP, TCP_SYNCNT, &kZero, sizeof(kZero)), + SyscallFailsWithErrno(EINVAL)); + + // TCP_SYNCNT less than 1 should be rejected with an EINVAL. + constexpr int kNeg = -1; + EXPECT_THAT( + setsockopt(s.get(), IPPROTO_TCP, TCP_SYNCNT, &kNeg, sizeof(kNeg)), + SyscallFailsWithErrno(EINVAL)); + + int get = -1; + socklen_t get_len = sizeof(get); + + ASSERT_THAT(getsockopt(s.get(), IPPROTO_TCP, TCP_SYNCNT, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(default_syn_cnt, get); + } +} + +TEST_P(SimpleTcpSocketTest, GetTCPSynCntDefault) { + FileDescriptor s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + + int get = -1; + socklen_t get_len = sizeof(get); + constexpr int kDefaultSynCnt = 6; + + ASSERT_THAT(getsockopt(s.get(), IPPROTO_TCP, TCP_SYNCNT, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kDefaultSynCnt); +} + +TEST_P(SimpleTcpSocketTest, SetTCPSynCntGreaterThanOne) { + FileDescriptor s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + constexpr int kTCPSynCnt = 20; + ASSERT_THAT(setsockopt(s.get(), IPPROTO_TCP, TCP_SYNCNT, &kTCPSynCnt, + sizeof(kTCPSynCnt)), + SyscallSucceeds()); + + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT(getsockopt(s.get(), IPPROTO_TCP, TCP_SYNCNT, &get, &get_len), + SyscallSucceeds()); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kTCPSynCnt); +} + +TEST_P(SimpleTcpSocketTest, SetTCPSynCntAboveMax) { + FileDescriptor s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT(getsockopt(s.get(), IPPROTO_TCP, TCP_SYNCNT, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + int default_syn_cnt = get; + { + constexpr int kTCPSynCnt = 256; + ASSERT_THAT(setsockopt(s.get(), IPPROTO_TCP, TCP_SYNCNT, &kTCPSynCnt, + sizeof(kTCPSynCnt)), + SyscallFailsWithErrno(EINVAL)); + + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT(getsockopt(s.get(), IPPROTO_TCP, TCP_SYNCNT, &get, &get_len), + SyscallSucceeds()); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, default_syn_cnt); + } +} + +TEST_P(SimpleTcpSocketTest, SetTCPWindowClampBelowMinRcvBuf) { + FileDescriptor s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + + // Discover minimum receive buf by setting a really low value + // for the receive buffer. + constexpr int kZero = 0; + EXPECT_THAT(setsockopt(s.get(), SOL_SOCKET, SO_RCVBUF, &kZero, sizeof(kZero)), + SyscallSucceeds()); + + // Now retrieve the minimum value for SO_RCVBUF as the set above should + // have caused SO_RCVBUF for the socket to be set to the minimum. + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT(getsockopt(s.get(), SOL_SOCKET, SO_RCVBUF, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + int min_so_rcvbuf = get; + + { + // TCP_WINDOW_CLAMP less than min_so_rcvbuf/2 should be set to + // min_so_rcvbuf/2. + int below_half_min_rcvbuf = min_so_rcvbuf / 2 - 1; + EXPECT_THAT( + setsockopt(s.get(), IPPROTO_TCP, TCP_WINDOW_CLAMP, + &below_half_min_rcvbuf, sizeof(below_half_min_rcvbuf)), + SyscallSucceeds()); + + int get = -1; + socklen_t get_len = sizeof(get); + + ASSERT_THAT( + getsockopt(s.get(), IPPROTO_TCP, TCP_WINDOW_CLAMP, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(min_so_rcvbuf / 2, get); + } +} + +TEST_P(SimpleTcpSocketTest, SetTCPWindowClampZeroClosedSocket) { + FileDescriptor s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + constexpr int kZero = 0; + ASSERT_THAT( + setsockopt(s.get(), IPPROTO_TCP, TCP_WINDOW_CLAMP, &kZero, sizeof(kZero)), + SyscallSucceeds()); + + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT( + getsockopt(s.get(), IPPROTO_TCP, TCP_WINDOW_CLAMP, &get, &get_len), + SyscallSucceeds()); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, kZero); +} + +TEST_P(SimpleTcpSocketTest, SetTCPWindowClampAboveHalfMinRcvBuf) { + FileDescriptor s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + + // Discover minimum receive buf by setting a really low value + // for the receive buffer. + constexpr int kZero = 0; + EXPECT_THAT(setsockopt(s.get(), SOL_SOCKET, SO_RCVBUF, &kZero, sizeof(kZero)), + SyscallSucceeds()); + + // Now retrieve the minimum value for SO_RCVBUF as the set above should + // have caused SO_RCVBUF for the socket to be set to the minimum. + int get = -1; + socklen_t get_len = sizeof(get); + ASSERT_THAT(getsockopt(s.get(), SOL_SOCKET, SO_RCVBUF, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + int min_so_rcvbuf = get; + + { + int above_half_min_rcv_buf = min_so_rcvbuf / 2 + 1; + EXPECT_THAT( + setsockopt(s.get(), IPPROTO_TCP, TCP_WINDOW_CLAMP, + &above_half_min_rcv_buf, sizeof(above_half_min_rcv_buf)), + SyscallSucceeds()); + + int get = -1; + socklen_t get_len = sizeof(get); + + ASSERT_THAT( + getsockopt(s.get(), IPPROTO_TCP, TCP_WINDOW_CLAMP, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(above_half_min_rcv_buf, get); + } +} + INSTANTIATE_TEST_SUITE_P(AllInetTests, SimpleTcpSocketTest, ::testing::Values(AF_INET, AF_INET6)); -- cgit v1.2.3 From 47dfba76616a69887f0d5a4be6eb82b5dc5d0f52 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Thu, 14 May 2020 09:34:21 -0700 Subject: Port memfd_create to vfs2 and finish implementation of file seals. Closes #2612. PiperOrigin-RevId: 311548074 --- pkg/sentry/fsimpl/tmpfs/BUILD | 2 - pkg/sentry/fsimpl/tmpfs/filesystem.go | 21 +++++++++- pkg/sentry/fsimpl/tmpfs/regular_file.go | 42 +++++++++++++++++++ pkg/sentry/fsimpl/tmpfs/regular_file_test.go | 2 +- pkg/sentry/fsimpl/tmpfs/stat_test.go | 2 +- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 38 ++++++++++++++++- pkg/sentry/kernel/BUILD | 1 + pkg/sentry/kernel/kernel.go | 25 +++++++++++ pkg/sentry/syscalls/linux/vfs2/BUILD | 2 + pkg/sentry/syscalls/linux/vfs2/fd.go | 10 +++++ pkg/sentry/syscalls/linux/vfs2/memfd.go | 63 ++++++++++++++++++++++++++++ pkg/sentry/syscalls/linux/vfs2/vfs2.go | 2 +- 12 files changed, 203 insertions(+), 7 deletions(-) create mode 100644 pkg/sentry/syscalls/linux/vfs2/memfd.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD index 9a076ad71..007be1572 100644 --- a/pkg/sentry/fsimpl/tmpfs/BUILD +++ b/pkg/sentry/fsimpl/tmpfs/BUILD @@ -52,7 +52,6 @@ go_library( "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", "//pkg/sentry/fs/lock", - "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/pipe", "//pkg/sentry/kernel/time", @@ -106,7 +105,6 @@ go_test( "//pkg/sentry/contexttest", "//pkg/sentry/fs/lock", "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/contexttest", "//pkg/sentry/vfs", "//pkg/syserror", "//pkg/usermem", diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index e0ad82769..80fa7b29d 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -772,5 +772,24 @@ func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { fs.mu.RLock() defer fs.mu.RUnlock() - return genericPrependPath(vfsroot, vd.Mount(), vd.Dentry().Impl().(*dentry), b) + mnt := vd.Mount() + d := vd.Dentry().Impl().(*dentry) + for { + if mnt == vfsroot.Mount() && &d.vfsd == vfsroot.Dentry() { + return vfs.PrependPathAtVFSRootError{} + } + if &d.vfsd == mnt.Root() { + return nil + } + if d.parent == nil { + if d.name != "" { + // This must be an anonymous memfd file. + b.PrependComponent("/" + d.name) + return vfs.PrependPathSyntheticError{} + } + return vfs.PrependPathAtNonMountRootError{} + } + b.PrependComponent(d.name) + d = d.parent + } } diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index 57e5e28ec..3f433d666 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -88,6 +88,7 @@ type regularFile struct { func (fs *filesystem) newRegularFile(creds *auth.Credentials, mode linux.FileMode) *inode { file := ®ularFile{ memFile: fs.memFile, + seals: linux.F_SEAL_SEAL, } file.inode.init(file, fs, creds, linux.S_IFREG|mode) file.inode.nlink = 1 // from parent directory @@ -577,3 +578,44 @@ exitLoop: return done, retErr } + +// GetSeals returns the current set of seals on a memfd inode. +func GetSeals(fd *vfs.FileDescription) (uint32, error) { + f, ok := fd.Impl().(*regularFileFD) + if !ok { + return 0, syserror.EINVAL + } + rf := f.inode().impl.(*regularFile) + rf.dataMu.RLock() + defer rf.dataMu.RUnlock() + return rf.seals, nil +} + +// AddSeals adds new file seals to a memfd inode. +func AddSeals(fd *vfs.FileDescription, val uint32) error { + f, ok := fd.Impl().(*regularFileFD) + if !ok { + return syserror.EINVAL + } + rf := f.inode().impl.(*regularFile) + rf.mapsMu.Lock() + defer rf.mapsMu.Unlock() + rf.dataMu.RLock() + defer rf.dataMu.RUnlock() + + if rf.seals&linux.F_SEAL_SEAL != 0 { + // Seal applied which prevents addition of any new seals. + return syserror.EPERM + } + + // F_SEAL_WRITE can only be added if there are no active writable maps. + if rf.seals&linux.F_SEAL_WRITE == 0 && val&linux.F_SEAL_WRITE != 0 { + if rf.writableMappingPages > 0 { + return syserror.EBUSY + } + } + + // Seals can only be added, never removed. + rf.seals |= val + return nil +} diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go index f2bc96d51..64e1c40ad 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go @@ -21,8 +21,8 @@ import ( "testing" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/fs/lock" - "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" diff --git a/pkg/sentry/fsimpl/tmpfs/stat_test.go b/pkg/sentry/fsimpl/tmpfs/stat_test.go index f52755092..f7ee4aab2 100644 --- a/pkg/sentry/fsimpl/tmpfs/stat_test.go +++ b/pkg/sentry/fsimpl/tmpfs/stat_test.go @@ -19,8 +19,8 @@ import ( "testing" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" "gvisor.dev/gvisor/pkg/sentry/vfs" ) diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 405928bd0..1e781aecd 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -94,7 +94,7 @@ type FilesystemOpts struct { } // GetFilesystem implements vfs.FilesystemType.GetFilesystem. -func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { +func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, _ string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { memFileProvider := pgalloc.MemoryFileProviderFromContext(ctx) if memFileProvider == nil { panic("MemoryFileProviderFromContext returned nil") @@ -139,6 +139,11 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt return &fs.vfsfs, &root.vfsd, nil } +// NewFilesystem returns a new tmpfs filesystem. +func NewFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials) (*vfs.Filesystem, *vfs.Dentry, error) { + return FilesystemType{}.GetFilesystem(ctx, vfsObj, creds, "", vfs.GetFilesystemOptions{}) +} + // Release implements vfs.FilesystemImpl.Release. func (fs *filesystem) Release() { fs.vfsfs.VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) @@ -658,3 +663,34 @@ func (fd *fileDescription) Setxattr(ctx context.Context, opts vfs.SetxattrOption func (fd *fileDescription) Removexattr(ctx context.Context, name string) error { return fd.inode().removexattr(auth.CredentialsFromContext(ctx), name) } + +// NewMemfd creates a new tmpfs regular file and file description that can back +// an anonymous fd created by memfd_create. +func NewMemfd(mount *vfs.Mount, creds *auth.Credentials, allowSeals bool, name string) (*vfs.FileDescription, error) { + fs, ok := mount.Filesystem().Impl().(*filesystem) + if !ok { + panic("NewMemfd() called with non-tmpfs mount") + } + + // Per Linux, mm/shmem.c:__shmem_file_setup(), memfd inodes are set up with + // S_IRWXUGO. + mode := linux.FileMode(0777) + inode := fs.newRegularFile(creds, mode) + rf := inode.impl.(*regularFile) + if allowSeals { + rf.seals = 0 + } + + d := fs.newDentry(inode) + defer d.DecRef() + d.name = name + + // Per Linux, mm/shmem.c:__shmem_file_setup(), memfd files are set up with + // FMODE_READ | FMODE_WRITE. + var fd regularFileFD + flags := uint32(linux.O_RDWR) + if err := fd.vfsfd.Init(&fd, flags, mount, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { + return nil, err + } + return &fd.vfsfd, nil +} diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD index 8104f50f3..a28eab8b8 100644 --- a/pkg/sentry/kernel/BUILD +++ b/pkg/sentry/kernel/BUILD @@ -173,6 +173,7 @@ go_library( "//pkg/sentry/fsimpl/pipefs", "//pkg/sentry/fsimpl/sockfs", "//pkg/sentry/fsimpl/timerfd", + "//pkg/sentry/fsimpl/tmpfs", "//pkg/sentry/hostcpu", "//pkg/sentry/inet", "//pkg/sentry/kernel/auth", diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 3617da8c6..5efeb3767 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -53,6 +53,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fsimpl/pipefs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/timerfd" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" "gvisor.dev/gvisor/pkg/sentry/hostcpu" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" @@ -259,6 +260,10 @@ type Kernel struct { // syscalls (as opposed to named pipes created by mknod()). pipeMount *vfs.Mount + // shmMount is the Mount used for anonymous files created by the + // memfd_create() syscalls. It is analagous to Linux's shm_mnt. + shmMount *vfs.Mount + // socketMount is the Mount used for sockets created by the socket() and // socketpair() syscalls. There are several cases where a socket dentry will // not be contained in socketMount: @@ -330,6 +335,9 @@ func (k *Kernel) Init(args InitKernelArgs) error { if args.Timekeeper == nil { return fmt.Errorf("Timekeeper is nil") } + if args.Timekeeper.clocks == nil { + return fmt.Errorf("Must call Timekeeper.SetClocks() before Kernel.Init()") + } if args.RootUserNamespace == nil { return fmt.Errorf("RootUserNamespace is nil") } @@ -384,6 +392,18 @@ func (k *Kernel) Init(args InitKernelArgs) error { } k.pipeMount = pipeMount + tmpfsFilesystem, tmpfsRoot, err := tmpfs.NewFilesystem(k.SupervisorContext(), &k.vfs, auth.NewRootCredentials(k.rootUserNamespace)) + if err != nil { + return fmt.Errorf("failed to create tmpfs filesystem: %v", err) + } + defer tmpfsFilesystem.DecRef() + defer tmpfsRoot.DecRef() + shmMount, err := k.vfs.NewDisconnectedMount(tmpfsFilesystem, tmpfsRoot, &vfs.MountOptions{}) + if err != nil { + return fmt.Errorf("failed to create tmpfs mount: %v", err) + } + k.shmMount = shmMount + socketFilesystem, err := sockfs.NewFilesystem(&k.vfs) if err != nil { return fmt.Errorf("failed to create sockfs filesystem: %v", err) @@ -1656,6 +1676,11 @@ func (k *Kernel) PipeMount() *vfs.Mount { return k.pipeMount } +// ShmMount returns the tmpfs mount. +func (k *Kernel) ShmMount() *vfs.Mount { + return k.shmMount +} + // SocketMount returns the sockfs mount. func (k *Kernel) SocketMount() *vfs.Mount { return k.socketMount diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD index c32f942fb..f882ef840 100644 --- a/pkg/sentry/syscalls/linux/vfs2/BUILD +++ b/pkg/sentry/syscalls/linux/vfs2/BUILD @@ -13,6 +13,7 @@ go_library( "fscontext.go", "getdents.go", "ioctl.go", + "memfd.go", "mmap.go", "path.go", "pipe.go", @@ -43,6 +44,7 @@ go_library( "//pkg/sentry/fsimpl/pipefs", "//pkg/sentry/fsimpl/signalfd", "//pkg/sentry/fsimpl/timerfd", + "//pkg/sentry/fsimpl/tmpfs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/pipe", diff --git a/pkg/sentry/syscalls/linux/vfs2/fd.go b/pkg/sentry/syscalls/linux/vfs2/fd.go index 8181d80f4..ca0f7fd1e 100644 --- a/pkg/sentry/syscalls/linux/vfs2/fd.go +++ b/pkg/sentry/syscalls/linux/vfs2/fd.go @@ -17,6 +17,7 @@ package vfs2 import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" @@ -157,6 +158,15 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return 0, nil, syserror.EBADF } return uintptr(pipefile.PipeSize()), nil, nil + case linux.F_GET_SEALS: + val, err := tmpfs.GetSeals(file) + return uintptr(val), nil, err + case linux.F_ADD_SEALS: + if !file.IsWritable() { + return 0, nil, syserror.EPERM + } + err := tmpfs.AddSeals(file, args[2].Uint()) + return 0, nil, err default: // TODO(gvisor.dev/issue/1623): Everything else is not yet supported. return 0, nil, syserror.EINVAL diff --git a/pkg/sentry/syscalls/linux/vfs2/memfd.go b/pkg/sentry/syscalls/linux/vfs2/memfd.go new file mode 100644 index 000000000..bbe248d17 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/memfd.go @@ -0,0 +1,63 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/syserror" +) + +const ( + memfdPrefix = "memfd:" + memfdMaxNameLen = linux.NAME_MAX - len(memfdPrefix) + memfdAllFlags = uint32(linux.MFD_CLOEXEC | linux.MFD_ALLOW_SEALING) +) + +// MemfdCreate implements the linux syscall memfd_create(2). +func MemfdCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + addr := args[0].Pointer() + flags := args[1].Uint() + + if flags&^memfdAllFlags != 0 { + // Unknown bits in flags. + return 0, nil, syserror.EINVAL + } + + allowSeals := flags&linux.MFD_ALLOW_SEALING != 0 + cloExec := flags&linux.MFD_CLOEXEC != 0 + + name, err := t.CopyInString(addr, memfdMaxNameLen) + if err != nil { + return 0, nil, err + } + + shmMount := t.Kernel().ShmMount() + file, err := tmpfs.NewMemfd(shmMount, t.Credentials(), allowSeals, memfdPrefix+name) + if err != nil { + return 0, nil, err + } + + fd, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{ + CloseOnExec: cloExec, + }) + if err != nil { + return 0, nil, err + } + + return uintptr(fd), nil, nil +} diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2.go b/pkg/sentry/syscalls/linux/vfs2/vfs2.go index 9c04677f1..ec8da7f06 100644 --- a/pkg/sentry/syscalls/linux/vfs2/vfs2.go +++ b/pkg/sentry/syscalls/linux/vfs2/vfs2.go @@ -158,7 +158,7 @@ func Override() { s.Table[306] = syscalls.Supported("syncfs", Syncfs) s.Table[307] = syscalls.Supported("sendmmsg", SendMMsg) s.Table[316] = syscalls.Supported("renameat2", Renameat2) - delete(s.Table, 319) // memfd_create + s.Table[319] = syscalls.Supported("memfd_create", MemfdCreate) s.Table[322] = syscalls.Supported("execveat", Execveat) s.Table[327] = syscalls.Supported("preadv2", Preadv2) s.Table[328] = syscalls.Supported("pwritev2", Pwritev2) -- cgit v1.2.3 From fb7e5f1676ad9e6e7e83312e09fe55f91ade41bc Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Thu, 14 May 2020 20:08:25 -0700 Subject: Make utimes_test pass on VFS2. PiperOrigin-RevId: 311657502 --- pkg/sentry/fsimpl/gofer/gofer.go | 4 +- pkg/sentry/syscalls/linux/vfs2/setstat.go | 123 +++++++++++++++++++----------- pkg/sentry/syscalls/linux/vfs2/vfs2.go | 2 +- test/syscalls/linux/utimes.cc | 33 ++++---- 4 files changed, 101 insertions(+), 61 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 353e2cf5b..ebf063a58 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -869,8 +869,8 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin Size: stat.Mask&linux.STATX_SIZE != 0, ATime: stat.Mask&linux.STATX_ATIME != 0, MTime: stat.Mask&linux.STATX_MTIME != 0, - ATimeNotSystemTime: stat.Atime.Nsec != linux.UTIME_NOW, - MTimeNotSystemTime: stat.Mtime.Nsec != linux.UTIME_NOW, + ATimeNotSystemTime: stat.Mask&linux.STATX_ATIME != 0 && stat.Atime.Nsec != linux.UTIME_NOW, + MTimeNotSystemTime: stat.Mask&linux.STATX_MTIME != 0 && stat.Mtime.Nsec != linux.UTIME_NOW, }, p9.SetAttr{ Permissions: p9.FileMode(stat.Mode), UID: p9.UID(stat.UID), diff --git a/pkg/sentry/syscalls/linux/vfs2/setstat.go b/pkg/sentry/syscalls/linux/vfs2/setstat.go index 4e61f1452..09ecfed26 100644 --- a/pkg/sentry/syscalls/linux/vfs2/setstat.go +++ b/pkg/sentry/syscalls/linux/vfs2/setstat.go @@ -246,73 +246,104 @@ func Utimes(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal return 0, nil, err } - opts := vfs.SetStatOptions{ - Stat: linux.Statx{ - Mask: linux.STATX_ATIME | linux.STATX_MTIME, - }, - } - if timesAddr == 0 { - opts.Stat.Atime.Nsec = linux.UTIME_NOW - opts.Stat.Mtime.Nsec = linux.UTIME_NOW - } else { - var times [2]linux.Timeval - if _, err := t.CopyIn(timesAddr, ×); err != nil { - return 0, nil, err - } - opts.Stat.Atime = linux.StatxTimestamp{ - Sec: times[0].Sec, - Nsec: uint32(times[0].Usec * 1000), - } - opts.Stat.Mtime = linux.StatxTimestamp{ - Sec: times[1].Sec, - Nsec: uint32(times[1].Usec * 1000), - } + var opts vfs.SetStatOptions + if err := populateSetStatOptionsForUtimes(t, timesAddr, &opts); err != nil { + return 0, nil, err } return 0, nil, setstatat(t, linux.AT_FDCWD, path, disallowEmptyPath, followFinalSymlink, &opts) } -// Utimensat implements Linux syscall utimensat(2). -func Utimensat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { +// Futimesat implements Linux syscall futimesat(2). +func Futimesat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { dirfd := args[0].Int() pathAddr := args[1].Pointer() timesAddr := args[2].Pointer() - flags := args[3].Int() - if flags&^linux.AT_SYMLINK_NOFOLLOW != 0 { - return 0, nil, syserror.EINVAL - } - - path, err := copyInPath(t, pathAddr) - if err != nil { - return 0, nil, err + // "If filename is NULL and dfd refers to an open file, then operate on the + // file. Otherwise look up filename, possibly using dfd as a starting + // point." - fs/utimes.c + var path fspath.Path + shouldAllowEmptyPath := allowEmptyPath + if dirfd == linux.AT_FDCWD || pathAddr != 0 { + var err error + path, err = copyInPath(t, pathAddr) + if err != nil { + return 0, nil, err + } + shouldAllowEmptyPath = disallowEmptyPath } var opts vfs.SetStatOptions - if err := populateSetStatOptionsForUtimens(t, timesAddr, &opts); err != nil { + if err := populateSetStatOptionsForUtimes(t, timesAddr, &opts); err != nil { return 0, nil, err } - return 0, nil, setstatat(t, dirfd, path, disallowEmptyPath, followFinalSymlink, &opts) + return 0, nil, setstatat(t, dirfd, path, shouldAllowEmptyPath, followFinalSymlink, &opts) } -// Futimens implements Linux syscall futimens(2). -func Futimens(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := args[0].Int() - timesAddr := args[1].Pointer() - - file := t.GetFileVFS2(fd) - if file == nil { - return 0, nil, syserror.EBADF +func populateSetStatOptionsForUtimes(t *kernel.Task, timesAddr usermem.Addr, opts *vfs.SetStatOptions) error { + if timesAddr == 0 { + opts.Stat.Mask = linux.STATX_ATIME | linux.STATX_MTIME + opts.Stat.Atime.Nsec = linux.UTIME_NOW + opts.Stat.Mtime.Nsec = linux.UTIME_NOW + return nil } - defer file.DecRef() + var times [2]linux.Timeval + if _, err := t.CopyIn(timesAddr, ×); err != nil { + return err + } + if times[0].Usec < 0 || times[0].Usec > 999999 || times[1].Usec < 0 || times[1].Usec > 999999 { + return syserror.EINVAL + } + opts.Stat.Mask = linux.STATX_ATIME | linux.STATX_MTIME + opts.Stat.Atime = linux.StatxTimestamp{ + Sec: times[0].Sec, + Nsec: uint32(times[0].Usec * 1000), + } + opts.Stat.Mtime = linux.StatxTimestamp{ + Sec: times[1].Sec, + Nsec: uint32(times[1].Usec * 1000), + } + return nil +} +// Utimensat implements Linux syscall utimensat(2). +func Utimensat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + dirfd := args[0].Int() + pathAddr := args[1].Pointer() + timesAddr := args[2].Pointer() + flags := args[3].Int() + + // Linux requires that the UTIME_OMIT check occur before checking path or + // flags. var opts vfs.SetStatOptions if err := populateSetStatOptionsForUtimens(t, timesAddr, &opts); err != nil { return 0, nil, err } + if opts.Stat.Mask == 0 { + return 0, nil, nil + } - return 0, nil, file.SetStat(t, opts) + if flags&^linux.AT_SYMLINK_NOFOLLOW != 0 { + return 0, nil, syserror.EINVAL + } + + // "If filename is NULL and dfd refers to an open file, then operate on the + // file. Otherwise look up filename, possibly using dfd as a starting + // point." - fs/utimes.c + var path fspath.Path + shouldAllowEmptyPath := allowEmptyPath + if dirfd == linux.AT_FDCWD || pathAddr != 0 { + var err error + path, err = copyInPath(t, pathAddr) + if err != nil { + return 0, nil, err + } + shouldAllowEmptyPath = disallowEmptyPath + } + + return 0, nil, setstatat(t, dirfd, path, shouldAllowEmptyPath, shouldFollowFinalSymlink(flags&linux.AT_SYMLINK_NOFOLLOW == 0), &opts) } func populateSetStatOptionsForUtimens(t *kernel.Task, timesAddr usermem.Addr, opts *vfs.SetStatOptions) error { @@ -327,6 +358,9 @@ func populateSetStatOptionsForUtimens(t *kernel.Task, timesAddr usermem.Addr, op return err } if times[0].Nsec != linux.UTIME_OMIT { + if times[0].Nsec != linux.UTIME_NOW && (times[0].Nsec < 0 || times[0].Nsec > 999999999) { + return syserror.EINVAL + } opts.Stat.Mask |= linux.STATX_ATIME opts.Stat.Atime = linux.StatxTimestamp{ Sec: times[0].Sec, @@ -334,6 +368,9 @@ func populateSetStatOptionsForUtimens(t *kernel.Task, timesAddr usermem.Addr, op } } if times[1].Nsec != linux.UTIME_OMIT { + if times[1].Nsec != linux.UTIME_NOW && (times[1].Nsec < 0 || times[1].Nsec > 999999999) { + return syserror.EINVAL + } opts.Stat.Mask |= linux.STATX_MTIME opts.Stat.Mtime = linux.StatxTimestamp{ Sec: times[1].Sec, diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2.go b/pkg/sentry/syscalls/linux/vfs2/vfs2.go index ec8da7f06..a332d01bd 100644 --- a/pkg/sentry/syscalls/linux/vfs2/vfs2.go +++ b/pkg/sentry/syscalls/linux/vfs2/vfs2.go @@ -123,7 +123,7 @@ func Override() { s.Table[258] = syscalls.Supported("mkdirat", Mkdirat) s.Table[259] = syscalls.Supported("mknodat", Mknodat) s.Table[260] = syscalls.Supported("fchownat", Fchownat) - s.Table[261] = syscalls.Supported("futimens", Futimens) + s.Table[261] = syscalls.Supported("futimesat", Futimesat) s.Table[262] = syscalls.Supported("newfstatat", Newfstatat) s.Table[263] = syscalls.Supported("unlinkat", Unlinkat) s.Table[264] = syscalls.Supported("renameat", Renameat) diff --git a/test/syscalls/linux/utimes.cc b/test/syscalls/linux/utimes.cc index 22e6d1a85..e647d2896 100644 --- a/test/syscalls/linux/utimes.cc +++ b/test/syscalls/linux/utimes.cc @@ -48,12 +48,15 @@ void TimeBoxed(absl::Time* before, absl::Time* after, // filesystems set it to 1, so we don't do any truncation. struct timespec ts; EXPECT_THAT(clock_gettime(CLOCK_REALTIME_COARSE, &ts), SyscallSucceeds()); - *before = absl::TimeFromTimespec(ts); + // FIXME(b/132819225): gVisor filesystem timestamps inconsistently use the + // internal or host clock, which may diverge slightly. Allow some slack on + // times to account for the difference. + *before = absl::TimeFromTimespec(ts) - absl::Seconds(1); fn(); EXPECT_THAT(clock_gettime(CLOCK_REALTIME_COARSE, &ts), SyscallSucceeds()); - *after = absl::TimeFromTimespec(ts); + *after = absl::TimeFromTimespec(ts) + absl::Seconds(1); if (*after < *before) { // Clock jumped backwards; retry. @@ -68,11 +71,11 @@ void TimeBoxed(absl::Time* before, absl::Time* after, void TestUtimesOnPath(std::string const& path) { struct stat statbuf; - struct timeval times[2] = {{1, 0}, {2, 0}}; + struct timeval times[2] = {{10, 0}, {20, 0}}; EXPECT_THAT(utimes(path.c_str(), times), SyscallSucceeds()); EXPECT_THAT(stat(path.c_str(), &statbuf), SyscallSucceeds()); - EXPECT_EQ(1, statbuf.st_atime); - EXPECT_EQ(2, statbuf.st_mtime); + EXPECT_EQ(10, statbuf.st_atime); + EXPECT_EQ(20, statbuf.st_mtime); absl::Time before; absl::Time after; @@ -103,18 +106,18 @@ TEST(UtimesTest, OnDir) { TEST(UtimesTest, MissingPath) { auto path = NewTempAbsPath(); - struct timeval times[2] = {{1, 0}, {2, 0}}; + struct timeval times[2] = {{10, 0}, {20, 0}}; EXPECT_THAT(utimes(path.c_str(), times), SyscallFailsWithErrno(ENOENT)); } void TestFutimesat(int dirFd, std::string const& path) { struct stat statbuf; - struct timeval times[2] = {{1, 0}, {2, 0}}; + struct timeval times[2] = {{10, 0}, {20, 0}}; EXPECT_THAT(futimesat(dirFd, path.c_str(), times), SyscallSucceeds()); EXPECT_THAT(fstatat(dirFd, path.c_str(), &statbuf, 0), SyscallSucceeds()); - EXPECT_EQ(1, statbuf.st_atime); - EXPECT_EQ(2, statbuf.st_mtime); + EXPECT_EQ(10, statbuf.st_atime); + EXPECT_EQ(20, statbuf.st_mtime); absl::Time before; absl::Time after; @@ -175,11 +178,11 @@ TEST(FutimesatTest, InvalidNsec) { void TestUtimensat(int dirFd, std::string const& path) { struct stat statbuf; - const struct timespec times[2] = {{1, 0}, {2, 0}}; + const struct timespec times[2] = {{10, 0}, {20, 0}}; EXPECT_THAT(utimensat(dirFd, path.c_str(), times, 0), SyscallSucceeds()); EXPECT_THAT(fstatat(dirFd, path.c_str(), &statbuf, 0), SyscallSucceeds()); - EXPECT_EQ(1, statbuf.st_atime); - EXPECT_EQ(2, statbuf.st_mtime); + EXPECT_EQ(10, statbuf.st_atime); + EXPECT_EQ(20, statbuf.st_mtime); // Test setting with UTIME_NOW and UTIME_OMIT. struct stat statbuf2; @@ -301,13 +304,13 @@ TEST(Utimensat, NullPath) { auto f = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); const FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(f.path(), O_RDWR)); struct stat statbuf; - const struct timespec times[2] = {{1, 0}, {2, 0}}; + const struct timespec times[2] = {{10, 0}, {20, 0}}; // Call syscall directly. EXPECT_THAT(syscall(SYS_utimensat, fd.get(), NULL, times, 0), SyscallSucceeds()); EXPECT_THAT(fstatat(0, f.path().c_str(), &statbuf, 0), SyscallSucceeds()); - EXPECT_EQ(1, statbuf.st_atime); - EXPECT_EQ(2, statbuf.st_mtime); + EXPECT_EQ(10, statbuf.st_atime); + EXPECT_EQ(20, statbuf.st_mtime); } } // namespace -- cgit v1.2.3 From 679fd2527bdcaf2ca4dd05dad48a75ffc9400973 Mon Sep 17 00:00:00 2001 From: Bhasker Hariharan Date: Fri, 15 May 2020 15:04:33 -0700 Subject: Remove debug log left behind by mistake. PiperOrigin-RevId: 311808460 --- pkg/sentry/syscalls/linux/sys_splice.go | 2 -- 1 file changed, 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/sys_splice.go b/pkg/sentry/syscalls/linux/sys_splice.go index df0d0f461..39f2b79ec 100644 --- a/pkg/sentry/syscalls/linux/sys_splice.go +++ b/pkg/sentry/syscalls/linux/sys_splice.go @@ -16,7 +16,6 @@ package linux import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -26,7 +25,6 @@ import ( // doSplice implements a blocking splice operation. func doSplice(t *kernel.Task, outFile, inFile *fs.File, opts fs.SpliceOpts, nonBlocking bool) (int64, error) { - log.Infof("NLAC: doSplice opts: %+v", opts) if opts.Length < 0 || opts.SrcStart < 0 || opts.DstStart < 0 || (opts.SrcStart+opts.Length < 0) { return 0, syserror.EINVAL } -- cgit v1.2.3 From 420b791a3d6e0e6e2fc30c6f8be013bce7ca6549 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Fri, 15 May 2020 20:03:54 -0700 Subject: Minor formatting updates for gvisor.dev. * Aggregate architecture Overview in "What is gVisor?" as it makes more sense in one place. * Drop "user-space kernel" and use "application kernel". The term "user-space kernel" is confusing when some platform implementation do not run in user-space (instead running in guest ring zero). * Clear up the relationship between the Platform page in the user guide and the Platform page in the architecture guide, and ensure they are cross-linked. * Restore the call-to-action quick start link in the main page, and drop the GitHub link (which also appears in the top-right). * Improve image formatting by centering all doc and blog images, and move the image captions to the alt text. PiperOrigin-RevId: 311845158 --- README.md | 79 ++--- g3doc/BUILD | 4 + g3doc/Layers.png | Bin 0 -> 11044 bytes g3doc/Layers.svg | 1 + g3doc/Machine-Virtualization.png | Bin 0 -> 13205 bytes g3doc/Machine-Virtualization.svg | 1 + g3doc/README.md | 161 +++++++++- g3doc/Rule-Based-Execution.png | Bin 0 -> 6780 bytes g3doc/Rule-Based-Execution.svg | 1 + g3doc/Sentry-Gofer.png | Bin 0 -> 9064 bytes g3doc/Sentry-Gofer.svg | 1 + g3doc/architecture_guide/BUILD | 30 +- g3doc/architecture_guide/Layers.png | Bin 11044 -> 0 bytes g3doc/architecture_guide/Layers.svg | 1 - .../architecture_guide/Machine-Virtualization.png | Bin 13205 -> 0 bytes .../architecture_guide/Machine-Virtualization.svg | 1 - g3doc/architecture_guide/README.md | 83 ----- g3doc/architecture_guide/Rule-Based-Execution.png | Bin 6780 -> 0 bytes g3doc/architecture_guide/Rule-Based-Execution.svg | 1 - g3doc/architecture_guide/Sentry-Gofer.png | Bin 9064 -> 0 bytes g3doc/architecture_guide/Sentry-Gofer.svg | 1 - g3doc/architecture_guide/performance.md | 35 ++- g3doc/architecture_guide/platforms.md | 109 +++---- g3doc/architecture_guide/platforms.png | Bin 0 -> 21384 bytes g3doc/architecture_guide/platforms.svg | 334 +++++++++++++++++++++ g3doc/architecture_guide/resources.md | 27 +- g3doc/architecture_guide/resources.png | Bin 0 -> 16621 bytes g3doc/architecture_guide/resources.svg | 208 +++++++++++++ g3doc/architecture_guide/security.md | 28 +- g3doc/architecture_guide/security.png | Bin 0 -> 16932 bytes g3doc/architecture_guide/security.svg | 153 ++++++++++ g3doc/user_guide/filesystem.md | 4 +- g3doc/user_guide/platforms.md | 100 +++--- pkg/sentry/arch/syscalls_arm64.go | 2 +- pkg/sentry/kernel/pipe/pipe_util.go | 2 +- pkg/sentry/kernel/task_syscall.go | 4 +- pkg/sentry/socket/netstack/netstack.go | 6 +- pkg/tcpip/stack/stack.go | 4 +- pkg/tcpip/transport/tcp/endpoint.go | 2 +- pkg/tcpip/transport/tcp/tcp_test.go | 2 +- runsc/cmd/help.go | 12 +- website/BUILD | 1 - website/_layouts/docs.html | 2 + website/_sass/front.scss | 4 +- website/_sass/style.scss | 10 + website/blog/2019-11-18-security-basics.md | 28 +- website/blog/2020-04-02-networking-security.md | 8 +- website/index.md | 10 +- 48 files changed, 1054 insertions(+), 406 deletions(-) create mode 100644 g3doc/Layers.png create mode 100644 g3doc/Layers.svg create mode 100644 g3doc/Machine-Virtualization.png create mode 100644 g3doc/Machine-Virtualization.svg create mode 100644 g3doc/Rule-Based-Execution.png create mode 100644 g3doc/Rule-Based-Execution.svg create mode 100644 g3doc/Sentry-Gofer.png create mode 100644 g3doc/Sentry-Gofer.svg delete mode 100644 g3doc/architecture_guide/Layers.png delete mode 100644 g3doc/architecture_guide/Layers.svg delete mode 100644 g3doc/architecture_guide/Machine-Virtualization.png delete mode 100644 g3doc/architecture_guide/Machine-Virtualization.svg delete mode 100644 g3doc/architecture_guide/README.md delete mode 100644 g3doc/architecture_guide/Rule-Based-Execution.png delete mode 100644 g3doc/architecture_guide/Rule-Based-Execution.svg delete mode 100644 g3doc/architecture_guide/Sentry-Gofer.png delete mode 100644 g3doc/architecture_guide/Sentry-Gofer.svg create mode 100644 g3doc/architecture_guide/platforms.png create mode 100644 g3doc/architecture_guide/platforms.svg create mode 100644 g3doc/architecture_guide/resources.png create mode 100644 g3doc/architecture_guide/resources.svg create mode 100644 g3doc/architecture_guide/security.png create mode 100644 g3doc/architecture_guide/security.svg (limited to 'pkg/sentry') diff --git a/README.md b/README.md index de3e06f4e..442f5672a 100644 --- a/README.md +++ b/README.md @@ -5,7 +5,7 @@ ## What is gVisor? -**gVisor** is a user-space kernel, written in Go, that implements a substantial +**gVisor** is a application kernel, written in Go, that implements a substantial portion of the Linux system surface. It includes an [Open Container Initiative (OCI)][oci] runtime called `runsc` that provides an isolation boundary between the application and the host kernel. The `runsc` @@ -15,16 +15,17 @@ containers. ## Why does gVisor exist? Containers are not a [**sandbox**][sandbox]. While containers have -revolutionized how we develop, package, and deploy applications, running -untrusted or potentially malicious code without additional isolation is not a -good idea. The efficiency and performance gains from using a single, shared -kernel also mean that container escape is possible with a single vulnerability. - -gVisor is a user-space kernel for containers. It limits the host kernel surface -accessible to the application while still giving the application access to all -the features it expects. Unlike most kernels, gVisor does not assume or require -a fixed set of physical resources; instead, it leverages existing host kernel -functionality and runs as a normal user-space process. In other words, gVisor +revolutionized how we develop, package, and deploy applications, using them to +run untrusted or potentially malicious code without additional isolation is not +a good idea. While using a single, shared kernel allows for efficiency and +performance gains, it also means that container escape is possible with a single +vulnerability. + +gVisor is an application kernel for containers. It limits the host kernel +surface accessible to the application while still giving the application access +to all the features it expects. Unlike most kernels, gVisor does not assume or +require a fixed set of physical resources; instead, it leverages existing host +kernel functionality and runs as a normal process. In other words, gVisor implements Linux by way of Linux. gVisor should not be confused with technologies and tools to harden containers @@ -39,33 +40,24 @@ be found at [gvisor.dev][gvisor-dev]. ## Installing from source -gVisor currently requires x86\_64 Linux to build, though support for other -architectures may become available in the future. +gVisor builds on x86_64 and ARM64. Other architectures may become available in +the future. + +For the purposes of these instructions, [bazel][bazel] and other build +dependencies are wrapped in a build container. It is possible to use +[bazel][bazel] directly, or type `make help` for standard targets. ### Requirements Make sure the following dependencies are installed: * Linux 4.14.77+ ([older linux][old-linux]) -* [git][git] -* [Bazel][bazel] 1.2+ -* [Python][python] * [Docker version 17.09.0 or greater][docker] -* C++ toolchain supporting C++17 (GCC 7+, Clang 5+) -* Gold linker (e.g. `binutils-gold` package on Ubuntu) ### Building Build and install the `runsc` binary: -``` -bazel build runsc -sudo cp ./bazel-bin/runsc/linux_amd64_pure_stripped/runsc /usr/local/bin -``` - -If you don't want to install bazel on your system, you can build runsc in a -Docker container: - ``` make runsc sudo cp ./bazel-bin/runsc/linux_amd64_pure_stripped/runsc /usr/local/bin @@ -73,41 +65,19 @@ sudo cp ./bazel-bin/runsc/linux_amd64_pure_stripped/runsc /usr/local/bin ### Testing -The test suite can be run with Bazel: - -``` -bazel test //... -``` - -or in a Docker container: +To run standard test suites, you can use: ``` make unit-tests make tests ``` -### Using remote execution - -If you have a [Remote Build Execution][rbe] environment, you can use it to speed -up build and test cycles. - -You must authenticate with the project first: +To run specific tests, you can specify the target: ``` -gcloud auth application-default login --no-launch-browser +make test TARGET="//runsc:version_test" ``` -Then invoke bazel with the following flags: - -``` ---config=remote ---project_id=$PROJECT ---remote_instance_name=projects/$PROJECT/instances/default_instance -``` - -You can also add those flags to your local ~/.bazelrc to avoid needing to -specify them each time on the command line. - ### Using `go get` This project uses [bazel][bazel] to build and manage dependencies. A synthetic @@ -128,7 +98,7 @@ development on this branch is not supported. Development should occur on the ## Community & Governance -The governance model is documented in our [community][community] repository. +See [GOVERNANCE.md](GOVERANCE.md) for project governance information. The [gvisor-users mailing list][gvisor-users-list] and [gvisor-dev mailing list][gvisor-dev-list] are good starting points for @@ -145,12 +115,9 @@ See [Contributing.md](CONTRIBUTING.md). [bazel]: https://bazel.build [community]: https://gvisor.googlesource.com/community [docker]: https://www.docker.com -[git]: https://git-scm.com [gvisor-users-list]: https://groups.google.com/forum/#!forum/gvisor-users +[gvisor-dev]: https://gvisor.dev [gvisor-dev-list]: https://groups.google.com/forum/#!forum/gvisor-dev [oci]: https://www.opencontainers.org [old-linux]: https://gvisor.dev/docs/user_guide/networking/#gso -[python]: https://python.org -[rbe]: https://blog.bazel.build/2018/10/05/remote-build-execution.html [sandbox]: https://en.wikipedia.org/wiki/Sandbox_(computer_security) -[gvisor-dev]: https://gvisor.dev diff --git a/g3doc/BUILD b/g3doc/BUILD index 24177ad06..dbbf96204 100644 --- a/g3doc/BUILD +++ b/g3doc/BUILD @@ -9,6 +9,10 @@ doc( name = "index", src = "README.md", category = "Project", + data = glob([ + "*.png", + "*.svg", + ]), permalink = "/docs/", weight = "0", ) diff --git a/g3doc/Layers.png b/g3doc/Layers.png new file mode 100644 index 000000000..308c6c451 Binary files /dev/null and b/g3doc/Layers.png differ diff --git a/g3doc/Layers.svg b/g3doc/Layers.svg new file mode 100644 index 000000000..0a366f841 --- /dev/null +++ b/g3doc/Layers.svg @@ -0,0 +1 @@ + \ No newline at end of file diff --git a/g3doc/Machine-Virtualization.png b/g3doc/Machine-Virtualization.png new file mode 100644 index 000000000..1ba2ed6b2 Binary files /dev/null and b/g3doc/Machine-Virtualization.png differ diff --git a/g3doc/Machine-Virtualization.svg b/g3doc/Machine-Virtualization.svg new file mode 100644 index 000000000..5352da07b --- /dev/null +++ b/g3doc/Machine-Virtualization.svg @@ -0,0 +1 @@ + \ No newline at end of file diff --git a/g3doc/README.md b/g3doc/README.md index 7999f5d47..304a91493 100644 --- a/g3doc/README.md +++ b/g3doc/README.md @@ -1,6 +1,6 @@ # What is gVisor? -gVisor is a user-space kernel, written in Go, that implements a substantial +gVisor is an application kernel, written in Go, that implements a substantial portion of the [Linux system call interface][linux]. It provides an additional layer of isolation between running applications and the host operating system. @@ -9,19 +9,160 @@ that makes it easy to work with existing container tooling. The `runsc` runtime integrates with Docker and Kubernetes, making it simple to run sandboxed containers. -gVisor takes a distinct approach to container sandboxing and makes a different -set of technical trade-offs compared to existing sandbox technologies, thus -providing new tools and ideas for the container security landscape. - gVisor can be used with Docker, Kubernetes, or directly using `runsc`. Use the links below to see detailed instructions for each of them: -* [Docker](./user_guide/quick_start/docker/): The quickest and easiest way to - get started. -* [Kubernetes](./user_guide/quick_start/kubernetes/): Isolate Pods in your K8s - cluster with gVisor. -* [OCI Quick Start](./user_guide/quick_start/oci/): Expert mode. Customize +* [Docker](./user_guide/quick_start/docker.md): The quickest and easiest way + to get started. +* [Kubernetes](./user_guide/quick_start/kubernetes.md): Isolate Pods in your + K8s cluster with gVisor. +* [OCI Quick Start](./user_guide/quick_start/oci.md): Expert mode. Customize gVisor for your environment. +## What does gVisor do? + +gVisor provides a virtualized environment in order to sandbox containers. The +system interfaces normally implemented by the host kernel are moved into a +distinct, per-sandbox application kernel in order to minimize the risk of an +container escape exploit. gVisor does not introduce large fixed overheads +however, and still retains a process-like model with respect to resource +utilization. + +## How is this different? + +Two other approaches are commonly taken to provide stronger isolation than +native containers. + +**Machine-level virtualization**, such as [KVM][kvm] and [Xen][xen], exposes +virtualized hardware to a guest kernel via a Virtual Machine Monitor (VMM). This +virtualized hardware is generally enlightened (paravirtualized) and additional +mechanisms can be used to improve the visibility between the guest and host +(e.g. balloon drivers, paravirtualized spinlocks). Running containers in +distinct virtual machines can provide great isolation, compatibility and +performance (though nested virtualization may bring challenges in this area), +but for containers it often requires additional proxies and agents, and may +require a larger resource footprint and slower start-up times. + +![Machine-level virtualization](Machine-Virtualization.png "Machine-level virtualization") + +**Rule-based execution**, such as [seccomp][seccomp], [SELinux][selinux] and +[AppArmor][apparmor], allows the specification of a fine-grained security policy +for an application or container. These schemes typically rely on hooks +implemented inside the host kernel to enforce the rules. If the surface can be +made small enough, then this is an excellent way to sandbox applications and +maintain native performance. However, in practice it can be extremely difficult +(if not impossible) to reliably define a policy for arbitrary, previously +unknown applications, making this approach challenging to apply universally. + +![Rule-based execution](Rule-Based-Execution.png "Rule-based execution") + +Rule-based execution is often combined with additional layers for +defense-in-depth. + +**gVisor** provides a third isolation mechanism, distinct from those above. + +gVisor intercepts application system calls and acts as the guest kernel, without +the need for translation through virtualized hardware. gVisor may be thought of +as either a merged guest kernel and VMM, or as seccomp on steroids. This +architecture allows it to provide a flexible resource footprint (i.e. one based +on threads and memory mappings, not fixed guest physical resources) while also +lowering the fixed costs of virtualization. However, this comes at the price of +reduced application compatibility and higher per-system call overhead. + +![gVisor](Layers.png "gVisor") + +On top of this, gVisor employs rule-based execution to provide defense-in-depth +(details below). + +gVisor's approach is similar to [User Mode Linux (UML)][uml], although UML +virtualizes hardware internally and thus provides a fixed resource footprint. + +Each of the above approaches may excel in distinct scenarios. For example, +machine-level virtualization will face challenges achieving high density, while +gVisor may provide poor performance for system call heavy workloads. + +## Why Go? + +gVisor is written in [Go][golang] in order to avoid security pitfalls that can +plague kernels. With Go, there are strong types, built-in bounds checks, no +uninitialized variables, no use-after-free, no stack overflow, and a built-in +race detector. However, the use of Go has its challenges, and the runtime often +introduces performance overhead. + +## What are the different components? + +A gVisor sandbox consists of multiple processes. These processes collectively +comprise an environment in which one or more containers can be run. + +Each sandbox has its own isolated instance of: + +* The **Sentry**, which is a kernel that runs the containers and intercepts + and responds to system calls made by the application. + +Each container running in the sandbox has its own isolated instance of: + +* A **Gofer** which provides file system access to the containers. + +![gVisor architecture diagram](Sentry-Gofer.png "gVisor architecture diagram") + +## What is runsc? + +The entrypoint to running a sandboxed container is the `runsc` executable. +`runsc` implements the [Open Container Initiative (OCI)][oci] runtime +specification, which is used by Docker and Kubernetes. This means that OCI +compatible _filesystem bundles_ can be run by `runsc`. Filesystem bundles are +comprised of a `config.json` file containing container configuration, and a root +filesystem for the container. Please see the [OCI runtime spec][runtime-spec] +for more information on filesystem bundles. `runsc` implements multiple commands +that perform various functions such as starting, stopping, listing, and querying +the status of containers. + +### Sentry + + + +The Sentry is the largest component of gVisor. It can be thought of as a +application kernel. The Sentry implements all the kernel functionality needed by +the application, including: system calls, signal delivery, memory management and +page faulting logic, the threading model, and more. + +When the application makes a system call, the +[Platform](./architecture_guide/platforms.md) redirects the call to the Sentry, +which will do the necessary work to service it. It is important to note that the +Sentry does not pass system calls through to the host kernel. As a userspace +application, the Sentry will make some host system calls to support its +operation, but it does not allow the application to directly control the system +calls it makes. For example, the Sentry is not able to open files directly; file +system operations that extend beyond the sandbox (not internal `/proc` files, +pipes, etc) are sent to the Gofer, described below. + +### Gofer + + + +The Gofer is a standard host process which is started with each container and +communicates with the Sentry via the [9P protocol][9p] over a socket or shared +memory channel. The Sentry process is started in a restricted seccomp container +without access to file system resources. The Gofer mediates all access to the +these resources, providing an additional level of isolation. + +### Application + +The application is a normal Linux binary provided to gVisor in an OCI runtime +bundle. gVisor aims to provide an environment equivalent to Linux v4.4, so +applications should be able to run unmodified. However, gVisor does not +presently implement every system call, `/proc` file, or `/sys` file so some +incompatibilities may occur. See [Commpatibility](./user_guide/compatibility.md) +for more information. + +[9p]: https://en.wikipedia.org/wiki/9P_(protocol) +[apparmor]: https://wiki.ubuntu.com/AppArmor +[golang]: https://golang.org +[kvm]: https://www.linux-kvm.org [linux]: https://en.wikipedia.org/wiki/Linux_kernel_interfaces [oci]: https://www.opencontainers.org +[runtime-spec]: https://github.com/opencontainers/runtime-spec +[seccomp]: https://www.kernel.org/doc/Documentation/prctl/seccomp_filter.txt +[selinux]: https://selinuxproject.org +[uml]: http://user-mode-linux.sourceforge.net/ +[xen]: https://www.xenproject.org diff --git a/g3doc/Rule-Based-Execution.png b/g3doc/Rule-Based-Execution.png new file mode 100644 index 000000000..b42654a90 Binary files /dev/null and b/g3doc/Rule-Based-Execution.png differ diff --git a/g3doc/Rule-Based-Execution.svg b/g3doc/Rule-Based-Execution.svg new file mode 100644 index 000000000..bd6717043 --- /dev/null +++ b/g3doc/Rule-Based-Execution.svg @@ -0,0 +1 @@ + \ No newline at end of file diff --git a/g3doc/Sentry-Gofer.png b/g3doc/Sentry-Gofer.png new file mode 100644 index 000000000..ca2c27ef7 Binary files /dev/null and b/g3doc/Sentry-Gofer.png differ diff --git a/g3doc/Sentry-Gofer.svg b/g3doc/Sentry-Gofer.svg new file mode 100644 index 000000000..5c10750d2 --- /dev/null +++ b/g3doc/Sentry-Gofer.svg @@ -0,0 +1 @@ + \ No newline at end of file diff --git a/g3doc/architecture_guide/BUILD b/g3doc/architecture_guide/BUILD index 72038305b..404f627a4 100644 --- a/g3doc/architecture_guide/BUILD +++ b/g3doc/architecture_guide/BUILD @@ -5,31 +5,13 @@ package( licenses = ["notice"], ) -doc( - name = "index", - src = "README.md", - category = "Architecture Guide", - data = [ - "Layers.png", - "Layers.svg", - "Machine-Virtualization.png", - "Machine-Virtualization.svg", - "Rule-Based-Execution.png", - "Rule-Based-Execution.svg", - "Sentry-Gofer.png", - "Sentry-Gofer.svg", - ], - permalink = "/docs/architecture_guide/", - weight = "0", -) - doc( name = "platforms", src = "platforms.md", category = "Architecture Guide", data = [ - "Sentry-Gofer.png", - "Sentry-Gofer.svg", + "platforms.png", + "platforms.svg", ], permalink = "/docs/architecture_guide/platforms/", weight = "40", @@ -39,6 +21,10 @@ doc( name = "resources", src = "resources.md", category = "Architecture Guide", + data = [ + "resources.png", + "resources.svg", + ], permalink = "/docs/architecture_guide/resources/", weight = "30", ) @@ -48,8 +34,8 @@ doc( src = "security.md", category = "Architecture Guide", data = [ - "Layers.png", - "Layers.svg", + "security.png", + "security.svg", ], permalink = "/docs/architecture_guide/security/", weight = "10", diff --git a/g3doc/architecture_guide/Layers.png b/g3doc/architecture_guide/Layers.png deleted file mode 100644 index 308c6c451..000000000 Binary files a/g3doc/architecture_guide/Layers.png and /dev/null differ diff --git a/g3doc/architecture_guide/Layers.svg b/g3doc/architecture_guide/Layers.svg deleted file mode 100644 index 0a366f841..000000000 --- a/g3doc/architecture_guide/Layers.svg +++ /dev/null @@ -1 +0,0 @@ - \ No newline at end of file diff --git a/g3doc/architecture_guide/Machine-Virtualization.png b/g3doc/architecture_guide/Machine-Virtualization.png deleted file mode 100644 index 1ba2ed6b2..000000000 Binary files a/g3doc/architecture_guide/Machine-Virtualization.png and /dev/null differ diff --git a/g3doc/architecture_guide/Machine-Virtualization.svg b/g3doc/architecture_guide/Machine-Virtualization.svg deleted file mode 100644 index 5352da07b..000000000 --- a/g3doc/architecture_guide/Machine-Virtualization.svg +++ /dev/null @@ -1 +0,0 @@ - \ No newline at end of file diff --git a/g3doc/architecture_guide/README.md b/g3doc/architecture_guide/README.md deleted file mode 100644 index ab9ef7174..000000000 --- a/g3doc/architecture_guide/README.md +++ /dev/null @@ -1,83 +0,0 @@ -# Overview - -gVisor provides a virtualized environment in order to sandbox untrusted -containers. The system interfaces normally implemented by the host kernel are -moved into a distinct, per-sandbox user space kernel in order to minimize the -risk of an exploit. gVisor does not introduce large fixed overheads however, and -still retains a process-like model with respect to resource utilization. - -## How is this different? - -Two other approaches are commonly taken to provide stronger isolation than -native containers. - -**Machine-level virtualization**, such as [KVM][kvm] and [Xen][xen], exposes -virtualized hardware to a guest kernel via a Virtual Machine Monitor (VMM). This -virtualized hardware is generally enlightened (paravirtualized) and additional -mechanisms can be used to improve the visibility between the guest and host -(e.g. balloon drivers, paravirtualized spinlocks). Running containers in -distinct virtual machines can provide great isolation, compatibility and -performance (though nested virtualization may bring challenges in this area), -but for containers it often requires additional proxies and agents, and may -require a larger resource footprint and slower start-up times. - -![Machine-level virtualization](Machine-Virtualization.png "Machine-level virtualization") - -**Rule-based execution**, such as [seccomp][seccomp], [SELinux][selinux] and -[AppArmor][apparmor], allows the specification of a fine-grained security policy -for an application or container. These schemes typically rely on hooks -implemented inside the host kernel to enforce the rules. If the surface can be -made small enough (i.e. a sufficiently complete policy defined), then this is an -excellent way to sandbox applications and maintain native performance. However, -in practice it can be extremely difficult (if not impossible) to reliably define -a policy for arbitrary, previously unknown applications, making this approach -challenging to apply universally. - -![Rule-based execution](Rule-Based-Execution.png "Rule-based execution") - -Rule-based execution is often combined with additional layers for -defense-in-depth. - -**gVisor** provides a third isolation mechanism, distinct from those above. - -gVisor intercepts application system calls and acts as the guest kernel, without -the need for translation through virtualized hardware. gVisor may be thought of -as either a merged guest kernel and VMM, or as seccomp on steroids. This -architecture allows it to provide a flexible resource footprint (i.e. one based -on threads and memory mappings, not fixed guest physical resources) while also -lowering the fixed costs of virtualization. However, this comes at the price of -reduced application compatibility and higher per-system call overhead. - -![gVisor](Layers.png "gVisor") - -On top of this, gVisor employs rule-based execution to provide defense-in-depth -(details below). - -gVisor's approach is similar to [User Mode Linux (UML)][uml], although UML -virtualizes hardware internally and thus provides a fixed resource footprint. - -Each of the above approaches may excel in distinct scenarios. For example, -machine-level virtualization will face challenges achieving high density, while -gVisor may provide poor performance for system call heavy workloads. - -### Why Go? - -gVisor is written in [Go][golang] in order to avoid security pitfalls that can -plague kernels. With Go, there are strong types, built-in bounds checks, no -uninitialized variables, no use-after-free, no stack overflow, and a built-in -race detector. (The use of Go has its challenges too, and isn't free.) - -### What about Gofers? - - - -Gofers mediate file system interactions, and are used to provide additional -isolation. For more details, see the [Platform Guide](./platforms.md). - -[apparmor]: https://wiki.ubuntu.com/AppArmor -[golang]: https://golang.org -[kvm]: https://www.linux-kvm.org -[seccomp]: https://www.kernel.org/doc/Documentation/prctl/seccomp_filter.txt -[selinux]: https://selinuxproject.org -[uml]: http://user-mode-linux.sourceforge.net/ -[xen]: https://www.xenproject.org diff --git a/g3doc/architecture_guide/Rule-Based-Execution.png b/g3doc/architecture_guide/Rule-Based-Execution.png deleted file mode 100644 index b42654a90..000000000 Binary files a/g3doc/architecture_guide/Rule-Based-Execution.png and /dev/null differ diff --git a/g3doc/architecture_guide/Rule-Based-Execution.svg b/g3doc/architecture_guide/Rule-Based-Execution.svg deleted file mode 100644 index bd6717043..000000000 --- a/g3doc/architecture_guide/Rule-Based-Execution.svg +++ /dev/null @@ -1 +0,0 @@ - \ No newline at end of file diff --git a/g3doc/architecture_guide/Sentry-Gofer.png b/g3doc/architecture_guide/Sentry-Gofer.png deleted file mode 100644 index ca2c27ef7..000000000 Binary files a/g3doc/architecture_guide/Sentry-Gofer.png and /dev/null differ diff --git a/g3doc/architecture_guide/Sentry-Gofer.svg b/g3doc/architecture_guide/Sentry-Gofer.svg deleted file mode 100644 index 5c10750d2..000000000 --- a/g3doc/architecture_guide/Sentry-Gofer.svg +++ /dev/null @@ -1 +0,0 @@ - \ No newline at end of file diff --git a/g3doc/architecture_guide/performance.md b/g3doc/architecture_guide/performance.md index 3862d78ee..39dbb0045 100644 --- a/g3doc/architecture_guide/performance.md +++ b/g3doc/architecture_guide/performance.md @@ -13,12 +13,13 @@ forms: additional cycles and memory usage, which may manifest as increased latency, reduced throughput or density, or not at all. In general, these costs come from two different sources. -First, the existence of the [Sentry](../) means that additional memory will be -required, and application system calls must traverse additional layers of -software. The design emphasizes [security](../security/) and therefore we chose -to use a language for the Sentry that provides benefits in this domain but may -not yet offer the raw performance of other choices. Costs imposed by these -design choices are **structural costs**. +First, the existence of the [Sentry](../README.md#sentry) means that additional +memory will be required, and application system calls must traverse additional +layers of software. The design emphasizes +[security](/docs/architecture_guide/security/) and therefore we chose to use a +language for the Sentry that provides benefits in this domain but may not yet +offer the raw performance of other choices. Costs imposed by these design +choices are **structural costs**. Second, as gVisor is an independent implementation of the system call surface, many of the subsystems or specific calls are not as optimized as more mature @@ -50,7 +51,7 @@ Virtual Machines (VMs) with the following specifications: Through this document, `runsc` is used to indicate the runtime provided by gVisor. When relevant, we use the name `runsc-platform` to describe a specific -[platform choice](../platforms/). +[platform choice](/docs/architecture_guide/platforms/). **Except where specified, all tests below are conducted with the `ptrace` platform. The `ptrace` platform works everywhere and does not require hardware @@ -131,11 +132,11 @@ full start-up and run time for the workload, which trains a model. ## System calls Some **structural costs** of gVisor are heavily influenced by the -[platform choice](../platforms/), which implements system call interception. -Today, gVisor supports a variety of platforms. These platforms present distinct -performance, compatibility and security trade-offs. For example, the KVM -platform has low overhead system call interception but runs poorly with nested -virtualization. +[platform choice](/docs/architecture_guide/platforms/), which implements system +call interception. Today, gVisor supports a variety of platforms. These +platforms present distinct performance, compatibility and security trade-offs. +For example, the KVM platform has low overhead system call interception but runs +poorly with nested virtualization. {% include graph.html id="syscall" url="/performance/syscall.csv" title="perf.py syscall --runtime=runc --runtime=runsc-ptrace --runtime=runsc-kvm" y_min="100" @@ -163,7 +164,8 @@ overhead. Some of these costs above are **structural costs**, and `redis` is likely to remain a challenging performance scenario. However, optimizing the -[platform](../platforms/) will also have a dramatic impact. +[platform](/docs/architecture_guide/platforms/) will also have a dramatic +impact. ## Start-up time @@ -184,7 +186,7 @@ similarly loads a number of modules and binds an HTTP server. > Note: most of the time overhead above is associated Docker itself. This is > evident with the empty `runc` benchmark. To avoid these costs with `runsc`, > you may also consider using `runsc do` mode or invoking the -> [OCI runtime](../../user_guide/quick_start/oci/) directly. +> [OCI runtime](../user_guide/quick_start/oci.md) directly. ## Network @@ -222,8 +224,9 @@ In terms of raw disk I/O, gVisor does not introduce significant fundamental overhead. For general file operations, gVisor introduces a small fixed overhead for data that transitions across the sandbox boundary. This manifests as **structural costs** in some cases, since these operations must be routed -through the [Gofer](../) as a result of our [security model](../security/), but -in most cases are dominated by **implementation costs**, due to an internal +through the [Gofer](../README.md#gofer) as a result of our +[Security Model](/docs/architecture_guide/security/), but in most cases are +dominated by **implementation costs**, due to an internal [Virtual File System][vfs] (VFS) implementation that needs improvement. {% include graph.html id="fio-bw" url="/performance/fio.csv" title="perf.py fio diff --git a/g3doc/architecture_guide/platforms.md b/g3doc/architecture_guide/platforms.md index 6e63da8ce..d112c9a28 100644 --- a/g3doc/architecture_guide/platforms.md +++ b/g3doc/architecture_guide/platforms.md @@ -1,86 +1,61 @@ # Platform Guide -A gVisor sandbox consists of multiple processes when running. These processes -collectively comprise a shared environment in which one or more containers can -be run. +[TOC] -Each sandbox has its own isolated instance of: - -* The **Sentry**, A user-space kernel that runs the container and intercepts - and responds to system calls made by the application. - -Each container running in the sandbox has its own isolated instance of: - -* A **Gofer** which provides file system access to the container. - -![gVisor architecture diagram](Sentry-Gofer.png "gVisor architecture diagram") - -## runsc - -The entrypoint to running a sandboxed container is the `runsc` executable. -`runsc` implements the [Open Container Initiative (OCI)][oci] runtime -specification. This means that OCI compatible _filesystem bundles_ can be run by -`runsc`. Filesystem bundles are comprised of a `config.json` file containing -container configuration, and a root filesystem for the container. Please see the -[OCI runtime spec][runtime-spec] for more information on filesystem bundles. -`runsc` implements multiple commands that perform various functions such as -starting, stopping, listing, and querying the status of containers. +gVisor requires a platform to implement interception of syscalls, basic context +switching, and memory mapping functionality. Internally, gVisor uses an +abstraction sensibly called [Platform][platform]. A simplified version of this +interface looks like: -## Sentry +```golang +type Platform interface { + NewAddressSpace() (AddressSpace, error) + NewContext() Context +} -The Sentry is the largest component of gVisor. It can be thought of as a -userspace OS kernel. The Sentry implements all the kernel functionality needed -by the untrusted application. It implements all of the supported system calls, -signal delivery, memory management and page faulting logic, the threading model, -and more. +type Context interface { + Switch(as AddressSpace, ac arch.Context) (..., error) +} -When the untrusted application makes a system call, the currently used platform -redirects the call to the Sentry, which will do the necessary work to service -it. It is important to note that the Sentry will not simply pass through system -calls to the host kernel. As a userspace application, the Sentry will make some -host system calls to support its operation, but it will not allow the -application to directly control the system calls it makes. +type AddressSpace interface { + MapFile(addr usermem.Addr, f File, fr FileRange, at usermem.AccessType, ...) error + Unmap(addr usermem.Addr, length uint64) +} +``` -The Sentry aims to present an equivalent environment to (upstream) Linux v4.4. +There are a number of different ways to implement this interface that come with +various trade-offs, generally around performance and hardware requirements. -File system operations that extend beyond the sandbox (not internal /proc files, -pipes, etc) are sent to the Gofer, described below. +## Implementations -## Platforms +The choice of platform depends on the context in which `runsc` is executing. In +general, virtualized platforms may be limited to platforms that do not require +hardware virtualized support (since the hardware is already in use): -gVisor requires a platform to implement interception of syscalls, basic context -switching, and memory mapping functionality. +![Platforms](platforms.png "Platform examples.") ### ptrace -The ptrace platform uses `PTRACE_SYSEMU` to execute user code without allowing -it to execute host system calls. This platform can run anywhere that ptrace -works (even VMs without nested virtualization). - -### KVM (experimental) +The ptrace platform uses [PTRACE_SYSEMU][ptrace] to execute user code without +allowing it to execute host system calls. This platform can run anywhere that +`ptrace` works (even VMs without nested virtualization), which is ubiquitous. -The KVM platform allows the Sentry to act as both guest OS and VMM, switching -back and forth between the two worlds seamlessly. The KVM platform can run on -bare-metal or in a VM with nested virtualization enabled. While there is no -virtualized hardware layer -- the sandbox retains a process model -- gVisor -leverages virtualization extensions available on modern processors in order to -improve isolation and performance of address space switches. +Unfortunately, the ptrace platform has high context switch overhead, so system +call-heavy applications may pay a [performance penalty](./performance.md). -## Gofer +### KVM -The Gofer is a normal host Linux process. The Gofer is started with each sandbox -and connected to the Sentry. The Sentry process is started in a restricted -seccomp container without access to file system resources. The Gofer provides -the Sentry access to file system resources via the 9P protocol and provides an -additional level of isolation. +The KVM platform uses the kernel's [KVM][kvm] functionality to allow the Sentry +to act as both guest OS and VMM. The KVM platform can run on bare-metal or in a +VM with nested virtualization enabled. While there is no virtualized hardware +layer -- the sandbox retains a process model -- gVisor leverages virtualization +extensions available on modern processors in order to improve isolation and +performance of address space switches. -## Application +## Changing Platforms -The application (aka the untrusted application) is a normal Linux binary -provided to gVisor in an OCI runtime bundle. gVisor aims to provide an -environment equivalent to Linux v4.4, so applications should be able to run -unmodified. However, gVisor does not presently implement every system call, -/proc file, or /sys file so some incompatibilities may occur. +See [Changing Platforms](../user_guide/platforms.md). -[oci]: https://www.opencontainers.org -[runtime-spec]: https://github.com/opencontainers/runtime-spec +[kvm]: https://www.kernel.org/doc/Documentation/virtual/kvm/api.txt +[platform]: https://cs.opensource.google/gvisor/gvisor/+/release-20190304.1:pkg/sentry/platform/platform.go;l=33 +[ptrace]: http://man7.org/linux/man-pages/man2/ptrace.2.html diff --git a/g3doc/architecture_guide/platforms.png b/g3doc/architecture_guide/platforms.png new file mode 100644 index 000000000..005d56feb Binary files /dev/null and b/g3doc/architecture_guide/platforms.png differ diff --git a/g3doc/architecture_guide/platforms.svg b/g3doc/architecture_guide/platforms.svg new file mode 100644 index 000000000..b0bac9ba7 --- /dev/null +++ b/g3doc/architecture_guide/platforms.svg @@ -0,0 +1,334 @@ + + + + + + + + + + image/svg+xml + + + + + + + + + + gVisor + workload + host + + + + gVisor + workload + + KVM + + + gVisor + workload + + ptrace + + ptrace + VM + + guest + + + gVisor + workload + + ptrace + + diff --git a/g3doc/architecture_guide/resources.md b/g3doc/architecture_guide/resources.md index 894f995ae..1dec37bd1 100644 --- a/g3doc/architecture_guide/resources.md +++ b/g3doc/architecture_guide/resources.md @@ -10,9 +10,10 @@ sandbox to be highly dynamic in terms of resource usage: spanning a large number of cores and large amount of memory when busy, and yielding those resources back to the host when not. -Some of the details here may depend on the [platform](../platforms/), but in -general this page describes the resource model used by gVisor. If you're not -familiar with the terms here, uou may want to start with the [Overview](../). +In order words, the shape of the sandbox should closely track the shape of the +sandboxed process: + +![Resource model](resources.png "Workloads of different shapes.") ## Processes @@ -23,9 +24,9 @@ the sandbox (e.g. via a [Docker exec][exec]). ## Networking -Similarly to processes, the sandbox attaches a network endpoint to the system, -but runs it's own network stack. All network resources, other than packets in -flight, exist only inside the sandbox, bound by relevant resource limits. +The sandbox attaches a network endpoint to the system, but runs it's own network +stack. All network resources, other than packets in flight on the host, exist +only inside the sandbox, bound by relevant resource limits. You can interact with network endpoints exposed by the sandbox, just as you would any other container, but network introspection similarly requires entering @@ -33,15 +34,14 @@ the sandbox. ## Files -Files may be backed by different implementations. For host-native files (where a -file descriptor is available), the Gofer may return a file descriptor to the -Sentry via [SCM_RIGHTS][scmrights][^1]. +Files in the sandbox may be backed by different implementations. For host-native +files (where a file descriptor is available), the Gofer may return a file +descriptor to the Sentry via [SCM_RIGHTS][scmrights][^1]. These files may be read from and written to through standard system calls, and also mapped into the associated application's address space. This allows the same host memory to be shared across multiple sandboxes, although this mechanism -does not preclude the use of side-channels (see the -[security model](../security/)). +does not preclude the use of side-channels (see [Security Model](./security.md). Note that some file systems exist only within the context of the sandbox. For example, in many cases a `tmpfs` mount will be available at `/tmp` or @@ -64,8 +64,9 @@ scheduling decisions about all application threads. ## Time Time in the sandbox is provided by the Sentry, through its own [vDSO][vdso] and -timekeeping implementation. This is divorced from the host time, and no state is -shared with the host, although the time will be initialized with the host clock. +time-keeping implementation. This is distinct from the host time, and no state +is shared with the host, although the time will be initialized with the host +clock. The Sentry runs timers to note the passage of time, much like a kernel running on hardware (though the timers are software timers, in this case). These timers diff --git a/g3doc/architecture_guide/resources.png b/g3doc/architecture_guide/resources.png new file mode 100644 index 000000000..f715008ec Binary files /dev/null and b/g3doc/architecture_guide/resources.png differ diff --git a/g3doc/architecture_guide/resources.svg b/g3doc/architecture_guide/resources.svg new file mode 100644 index 000000000..fd7805d90 --- /dev/null +++ b/g3doc/architecture_guide/resources.svg @@ -0,0 +1,208 @@ + + + + + + + + + + image/svg+xml + + + + + + + + + + + + + + gVisor + gVisor + gVisor + workload + workload + workload + host + + diff --git a/g3doc/architecture_guide/security.md b/g3doc/architecture_guide/security.md index f78586291..b99b86332 100644 --- a/g3doc/architecture_guide/security.md +++ b/g3doc/architecture_guide/security.md @@ -86,15 +86,17 @@ a substitute for a secure architecture*. ## Goals: Limiting Exposure -gVisor’s primary design goal is to minimize the System API attack vector while -still providing a process model. There are two primary security principles that -inform this design. First, the application’s direct interactions with the host -System API are intercepted by the Sentry, which implements the System API -instead. Second, the System API accessible to the Sentry itself is minimized to -a safer, restricted set. The first principle minimizes the possibility of direct -exploitation of the host System API by applications, and the second principle -minimizes indirect exploitability, which is the exploitation by an exploited or -buggy Sentry (e.g. chaining an exploit). +![Threat model](security.png "Threat model.") + +gVisor’s primary design goal is to minimize the System API attack vector through +multiple layers of defense, while still providing a process model. There are two +primary security principles that inform this design. First, the application’s +direct interactions with the host System API are intercepted by the Sentry, +which implements the System API instead. Second, the System API accessible to +the Sentry itself is minimized to a safer, restricted set. The first principle +minimizes the possibility of direct exploitation of the host System API by +applications, and the second principle minimizes indirect exploitability, which +is the exploitation by an exploited or buggy Sentry (e.g. chaining an exploit). The first principle is similar to the security basis for a Virtual Machine (VM). With a VM, an application’s interactions with the host are replaced by @@ -210,9 +212,9 @@ crashes are recorded and triaged to similarly identify material issues. ### Is this more or less secure than a Virtual Machine? The security of a VM depends to a large extent on what is exposed from the host -kernel and user space support code. For example, device emulation code in the +kernel and userspace support code. For example, device emulation code in the host kernel (e.g. APIC) or optimizations (e.g. vhost) can be more complex than a -simple system call, and exploits carry the same risks. Similarly, the user space +simple system call, and exploits carry the same risks. Similarly, the userspace support code is frequently unsandboxed, and exploits, while rare, may allow unfettered access to the system. @@ -245,8 +247,8 @@ In gVisor, the platforms that use ptrace operate differently. The stubs that are traced are never allowed to continue execution into the host kernel and complete a call directly. Instead, all system calls are interpreted and handled by the Sentry itself, who reflects resulting register state back into the tracee before -continuing execution in user space. This is very similar to the mechanism used -by User-Mode Linux (UML). +continuing execution in userspace. This is very similar to the mechanism used by +User-Mode Linux (UML). [dirtycow]: https://en.wikipedia.org/wiki/Dirty_COW [clang]: https://en.wikipedia.org/wiki/C_(programming_language) diff --git a/g3doc/architecture_guide/security.png b/g3doc/architecture_guide/security.png new file mode 100644 index 000000000..c29befbf6 Binary files /dev/null and b/g3doc/architecture_guide/security.png differ diff --git a/g3doc/architecture_guide/security.svg b/g3doc/architecture_guide/security.svg new file mode 100644 index 000000000..0575e2dec --- /dev/null +++ b/g3doc/architecture_guide/security.svg @@ -0,0 +1,153 @@ + + + + + + + + + + image/svg+xml + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + diff --git a/g3doc/user_guide/filesystem.md b/g3doc/user_guide/filesystem.md index 6c69f42a1..cd00762dd 100644 --- a/g3doc/user_guide/filesystem.md +++ b/g3doc/user_guide/filesystem.md @@ -4,8 +4,8 @@ gVisor accesses the filesystem through a file proxy, called the Gofer. The gofer runs as a separate process, that is isolated from the sandbox. Gofer instances -communicate with their respective sentry using the 9P protocol. For a more -detailed explanation see [Overview > Gofer](../../architecture_guide/#gofer). +communicate with their respective sentry using the 9P protocol. For another +explanation see [What is gVisor?](../README.md). ## Sandbox overlay diff --git a/g3doc/user_guide/platforms.md b/g3doc/user_guide/platforms.md index eefb6b222..752025881 100644 --- a/g3doc/user_guide/platforms.md +++ b/g3doc/user_guide/platforms.md @@ -1,56 +1,27 @@ -# Platforms (KVM) +# Changing Platforms [TOC] -This document will help you set up your system to use a different gVisor -platform. +This guide described how to change the +[platform](../architecture_guide/platforms.md) used by `runsc`. -## What is a Platform? +## Prerequisites -gVisor requires a *platform* to implement interception of syscalls, basic -context switching, and memory mapping functionality. These are described in more -depth in the [Platform Design](../../architecture_guide/platforms/). +If you intend to run the KVM platform, you will also to have KVM installed on +your system. If you are running a Debian based system like Debian or Ubuntu you +can usually do this by ensuring the module is loaded, and permissions are +appropriately set on the `/dev/kvm` device. -## Selecting a Platform - -The platform is selected by the `--platform` command line flag passed to -`runsc`. By default, the ptrace platform is selected. To select a different -platform, modify your Docker configuration (`/etc/docker/daemon.json`) to pass -this argument: - -```json -{ - "runtimes": { - "runsc": { - "path": "/usr/local/bin/runsc", - "runtimeArgs": [ - "--platform=kvm" - ] - } - } -} -``` - -You must restart the Docker daemon after making changes to this file, typically -this is done via `systemd`: +If you have an Intel CPU: ```bash -sudo systemctl restart docker +sudo modprobe kvm-intel && sudo chmod a+rw /dev/kvm ``` -## Example: Using the KVM Platform - -The KVM platform is currently experimental; however, it provides several -benefits over the default ptrace platform. - -### Prerequisites - -You will also to have KVM installed on your system. If you are running a Debian -based system like Debian or Ubuntu you can usually do this by installing the -`qemu-kvm` package. +If you have an AMD CPU: ```bash -sudo apt-get install qemu-kvm +sudo modprobe kvm-amd && sudo chmod a+rw /dev/kvm ``` If you are using a virtual machine you will need to make sure that nested @@ -68,31 +39,22 @@ cause of security issues (e.g. [CVE-2018-12904](https://nvd.nist.gov/vuln/detail/CVE-2018-12904)). It is not recommended for production.*** -### Configuring Docker - -Per above, you will need to configure Docker to use `runsc` with the KVM -platform. You will remember from the Docker Quick Start that you configured -Docker to use `runsc` as the runtime. Docker allows you to add multiple runtimes -to the Docker configuration. +## Configuring Docker -Add a new entry for the KVM platform entry to your Docker configuration -(`/etc/docker/daemon.json`) in order to provide the `--platform=kvm` runtime -argument. - -In the end, the file should look something like: +The platform is selected by the `--platform` command line flag passed to +`runsc`. By default, the ptrace platform is selected. For example, to select the +KVM platform, modify your Docker configuration (`/etc/docker/daemon.json`) to +pass the `--platform` argument: ```json { "runtimes": { "runsc": { - "path": "/usr/local/bin/runsc" - }, - "runsc-kvm": { "path": "/usr/local/bin/runsc", "runtimeArgs": [ "--platform=kvm" ] - } + } } } ``` @@ -104,13 +66,27 @@ this is done via `systemd`: sudo systemctl restart docker ``` -## Running a container +Note that you may configure multiple runtimes using different platforms. For +example, the following configuration has one configuration for ptrace and one +for the KVM platform: -Now run your container using the `runsc-kvm` runtime. This will run the -container using the KVM platform: - -```bash -docker run --runtime=runsc-kvm --rm hello-world +```json +{ + "runtimes": { + "runsc-ptrace": { + "path": "/usr/local/bin/runsc", + "runtimeArgs": [ + "--platform=ptrace" + ] + }, + "runsc-kvm": { + "path": "/usr/local/bin/runsc", + "runtimeArgs": [ + "--platform=kvm" + ] + } + } +} ``` [nested-azure]: https://docs.microsoft.com/en-us/azure/virtual-machines/windows/nested-virtualization diff --git a/pkg/sentry/arch/syscalls_arm64.go b/pkg/sentry/arch/syscalls_arm64.go index 92d062513..95dfd1e90 100644 --- a/pkg/sentry/arch/syscalls_arm64.go +++ b/pkg/sentry/arch/syscalls_arm64.go @@ -23,7 +23,7 @@ const restartSyscallNr = uintptr(128) // // In linux, at the entry of the syscall handler(el0_svc_common()), value of R0 // is saved to the pt_regs.orig_x0 in kernel code. But currently, the orig_x0 -// was not accessible to the user space application, so we have to do the same +// was not accessible to the userspace application, so we have to do the same // operation in the sentry code to save the R0 value into the App context. func (c *context64) SyscallSaveOrig() { c.OrigR0 = c.Regs.Regs[0] diff --git a/pkg/sentry/kernel/pipe/pipe_util.go b/pkg/sentry/kernel/pipe/pipe_util.go index 5a1d4fd57..aacf28da2 100644 --- a/pkg/sentry/kernel/pipe/pipe_util.go +++ b/pkg/sentry/kernel/pipe/pipe_util.go @@ -144,7 +144,7 @@ func (p *Pipe) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArgume if v > math.MaxInt32 { v = math.MaxInt32 // Silently truncate. } - // Copy result to user-space. + // Copy result to userspace. _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), int32(v), usermem.IOOpts{ AddressSpaceActive: true, }) diff --git a/pkg/sentry/kernel/task_syscall.go b/pkg/sentry/kernel/task_syscall.go index c9db78e06..a5903b0b5 100644 --- a/pkg/sentry/kernel/task_syscall.go +++ b/pkg/sentry/kernel/task_syscall.go @@ -199,10 +199,10 @@ func (t *Task) doSyscall() taskRunState { // // On x86, register rax was shared by syscall number and return // value, and at the entry of the syscall handler, the rax was - // saved to regs.orig_rax which was exposed to user space. + // saved to regs.orig_rax which was exposed to userspace. // But on arm64, syscall number was passed through X8, and the X0 // was shared by the first syscall argument and return value. The - // X0 was saved to regs.orig_x0 which was not exposed to user space. + // X0 was saved to regs.orig_x0 which was not exposed to userspace. // So we have to do the same operation here to save the X0 value // into the task context. t.Arch().SyscallSaveOrig() diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 9dea2b5ff..60df51dae 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -2718,7 +2718,7 @@ func (s *socketOpsCommon) ioctl(ctx context.Context, io usermem.IO, args arch.Sy v = math.MaxInt32 } - // Copy result to user-space. + // Copy result to userspace. _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), int32(v), usermem.IOOpts{ AddressSpaceActive: true, }) @@ -2787,7 +2787,7 @@ func Ioctl(ctx context.Context, ep commonEndpoint, io usermem.IO, args arch.Sysc if v > math.MaxInt32 { v = math.MaxInt32 } - // Copy result to user-space. + // Copy result to userspace. _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), int32(v), usermem.IOOpts{ AddressSpaceActive: true, }) @@ -2803,7 +2803,7 @@ func Ioctl(ctx context.Context, ep commonEndpoint, io usermem.IO, args arch.Sysc v = math.MaxInt32 } - // Copy result to user-space. + // Copy result to userspace. _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), int32(v), usermem.IOOpts{ AddressSpaceActive: true, }) diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index b39ffa9fb..0ab4c3e19 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -235,11 +235,11 @@ type RcvBufAutoTuneParams struct { // was started. MeasureTime time.Time - // CopiedBytes is the number of bytes copied to user space since + // CopiedBytes is the number of bytes copied to userspace since // this measure began. CopiedBytes int - // PrevCopiedBytes is the number of bytes copied to user space in + // PrevCopiedBytes is the number of bytes copied to userspace in // the previous RTT period. PrevCopiedBytes int diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 71735029e..b5ba972f1 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -1097,7 +1097,7 @@ func (e *endpoint) initialReceiveWindow() int { } // ModerateRecvBuf adjusts the receive buffer and the advertised window -// based on the number of bytes copied to user space. +// based on the number of bytes copied to userspace. func (e *endpoint) ModerateRecvBuf(copied int) { e.LockUser() defer e.UnlockUser() diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index 0b4512c65..6ef32a1b3 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -5869,7 +5869,7 @@ func TestReceiveBufferAutoTuning(t *testing.T) { // Invoke the moderation API. This is required for auto-tuning // to happen. This method is normally expected to be invoked // from a higher layer than tcpip.Endpoint. So we simulate - // copying to user-space by invoking it explicitly here. + // copying to userspace by invoking it explicitly here. c.EP.ModerateRecvBuf(totalCopied) // Now send a keep-alive packet to trigger an ACK so that we can diff --git a/runsc/cmd/help.go b/runsc/cmd/help.go index c7d210140..cd85dabbb 100644 --- a/runsc/cmd/help.go +++ b/runsc/cmd/help.go @@ -65,16 +65,10 @@ func (h *Help) Execute(ctx context.Context, f *flag.FlagSet, args ...interface{} switch f.NArg() { case 0: fmt.Fprintf(h.cdr.Output, "Usage: %s \n\n", h.cdr.Name()) - fmt.Fprintf(h.cdr.Output, `runsc is a command line client for running applications packaged in the Open -Container Initiative (OCI) format. Applications run by runsc are run in an -isolated gVisor sandbox that emulates a Linux environment. + fmt.Fprintf(h.cdr.Output, `runsc is the gVisor container runtime. -gVisor is a user-space kernel, written in Go, that implements a substantial -portion of the Linux system call interface. It provides an additional layer -of isolation between running applications and the host operating system. - -Functionality is provided by subcommands. For additonal help on individual -subcommands use "%s %s ". +Functionality is provided by subcommands. For help with a specific subcommand, +use "%s %s ". `, h.cdr.Name(), h.Name()) h.cdr.VisitGroups(func(g *subcommands.CommandGroup) { diff --git a/website/BUILD b/website/BUILD index d6afd5f44..c97b2560b 100644 --- a/website/BUILD +++ b/website/BUILD @@ -138,7 +138,6 @@ docs( "//g3doc:community", "//g3doc:index", "//g3doc:roadmap", - "//g3doc/architecture_guide:index", "//g3doc/architecture_guide:performance", "//g3doc/architecture_guide:platforms", "//g3doc/architecture_guide:resources", diff --git a/website/_layouts/docs.html b/website/_layouts/docs.html index 33ea8e1de..549305089 100644 --- a/website/_layouts/docs.html +++ b/website/_layouts/docs.html @@ -51,7 +51,9 @@ categories: Create issue

{% endif %} +
{{ content }} +
diff --git a/website/_sass/front.scss b/website/_sass/front.scss index 44a7e3473..0e4208f3c 100644 --- a/website/_sass/front.scss +++ b/website/_sass/front.scss @@ -4,12 +4,14 @@ background-repeat: no-repeat; background-size: cover; background-blend-mode: darken; - background-color: rgba(0, 0, 0, 0.1); + background-color: rgba(0, 0, 0, 0.3); p { color: #fff; margin-top: 0; margin-bottom: 0; font-weight: 300; + font-size: 24px; + line-height: 30px; } } diff --git a/website/_sass/style.scss b/website/_sass/style.scss index 520ea469a..4deb945d4 100644 --- a/website/_sass/style.scss +++ b/website/_sass/style.scss @@ -142,3 +142,13 @@ table th { margin-top: 10px; margin-bottom: 20px; } + +.docs-content * img { + display: block; + margin: 20px auto; +} + +.blog-content * img { + display: block; + margin: 20px auto; +} diff --git a/website/blog/2019-11-18-security-basics.md b/website/blog/2019-11-18-security-basics.md index ed6d97ffe..fbdd511dd 100644 --- a/website/blog/2019-11-18-security-basics.md +++ b/website/blog/2019-11-18-security-basics.md @@ -56,15 +56,9 @@ in combination: redundant walls, scattered draw bridges, small bottle-neck entrances, moats, etc. A simplified version of the design is below -([more detailed version](/docs/architecture_guide/))[^2]: +([more detailed version](/docs/))[^2]: --------------------------------------------------------------------------------- - -![Figure 1](/assets/images/2019-11-18-security-basics-figure1.png) - -Figure 1: Simplified design of gVisor. - --------------------------------------------------------------------------------- +![Figure 1](/assets/images/2019-11-18-security-basics-figure1.png "Simplified design of gVisor.") In order to discuss design principles, the following components are important to know: @@ -134,13 +128,7 @@ minimum level of permission is required for it to perform its function. Specifically, the closer you are to the untrusted application, the less privilege you have. --------------------------------------------------------------------------------- - -![Figure 2](/assets/images/2019-11-18-security-basics-figure2.png) - -Figure 2: runsc components and their privileges. - --------------------------------------------------------------------------------- +![Figure 2](/assets/images/2019-11-18-security-basics-figure2.png "runsc components and their privileges.") This is evident in how runsc (the drop in gVisor binary for Docker/Kubernetes) constructs the sandbox. The Sentry has the least privilege possible (it can't @@ -222,15 +210,7 @@ the host Linux syscalls. In other words, with gVisor, applications get the vast majority (and growing) functionality of Linux containers for only 68 possible syscalls to the Host OS. 350 syscalls to 68 is attack surface reduction. --------------------------------------------------------------------------------- - -![Figure 3](/assets/images/2019-11-18-security-basics-figure3.png) - -Figure 3: Reduction of Attack Surface of the Syscall Table. Note that the -Senty's Syscall Emulation Layer keeps the Containerized Process from ever -calling the Host OS. - --------------------------------------------------------------------------------- +![Figure 3](/assets/images/2019-11-18-security-basics-figure3.png "Reduction of Attack Surface of the Syscall Table. Note that the Senty's Syscall Emulation Layer keeps the Containerized Process from ever calling the Host OS.") ## Secure-by-default diff --git a/website/blog/2020-04-02-networking-security.md b/website/blog/2020-04-02-networking-security.md index 78f0a6714..5a5e38fd7 100644 --- a/website/blog/2020-04-02-networking-security.md +++ b/website/blog/2020-04-02-networking-security.md @@ -69,13 +69,7 @@ a similar syscall). Moreover, because packets typically come from off-host (e.g. the internet), the Host OS's packet processing code has received a lot of scrutiny, hopefully resulting in a high degree of hardening. --------------------------------------------------------------------------------- - -![Figure 1](/assets/images/2020-04-02-networking-security-figure1.png) - -Figure 1: Netstack and gVisor - --------------------------------------------------------------------------------- +![Figure 1](/assets/images/2020-04-02-networking-security-figure1.png "Network and gVisor.") ## Writing a network stack diff --git a/website/index.md b/website/index.md index 95d5d16f0..84f877d49 100644 --- a/website/index.md +++ b/website/index.md @@ -3,10 +3,10 @@
-

gVisor is an application kernel and container runtime providing defense-in-depth for containers anywhere.

+

gVisor is an application kernel for containers that provides efficient defense-in-depth anywhere.

+ Quick start  Learn More  - GitHub 

@@ -19,8 +19,8 @@

Container-native Security

-

By providing each container with its own userspace kernel, gVisor limits - the attack surface of the host. This protection does not limit +

By providing each container with its own application kernel, gVisor + limits the attack surface of the host. This protection does not limit functionality: gVisor runs unmodified binaries and integrates with container orchestration systems, such as Docker and Kubernetes, and supports features such as volumes and sidecars.

@@ -43,7 +43,7 @@ The pluggable platform architecture of gVisor allows it to run anywhere, enabling consistent security policies across multiple environments without having to rearchitect your infrastructure.

- Get Started » + Read More »
-- cgit v1.2.3 From a534b2c65236201703096bb010f8d6b742f14c0d Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Fri, 15 May 2020 05:28:46 -0400 Subject: adding the VM-Exit method for Arm64 On amd64, it uses 'HLT' to leave the guest. Unlike amd64, arm64 can only uses mmio_exit/psci to leave the guest. So, I designed the HYPERCALL_VMEXIT to be compatible with amd64/arm64. To keep it simple, I used the address of exception table as the MMIO base address, so that I can trigger a MMIO-EXIT by forcibly writing this space. Then, in host user space, I can calculate this address to find out which hypercall. Signed-off-by: Bin Lu --- pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go | 7 +++++ pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go | 13 ++++++++ pkg/sentry/platform/kvm/bluepill_unsafe.go | 39 ++++++++++++++++-------- pkg/sentry/platform/kvm/kvm_arm64.go | 3 ++ pkg/sentry/platform/kvm/kvm_const.go | 10 ++++++ pkg/sentry/platform/kvm/kvm_const_arm64.go | 7 +++++ pkg/sentry/platform/kvm/machine_arm64_unsafe.go | 4 +++ pkg/sentry/platform/ring0/entry_arm64.s | 14 +++++++-- 8 files changed, 81 insertions(+), 16 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go b/pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go index 99cac665d..0d1e83e6c 100644 --- a/pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go @@ -53,3 +53,10 @@ func dieArchSetup(c *vCPU, context *arch.SignalContext64, guestRegs *userRegs) { context.Rbx = uint64(uintptr(unsafe.Pointer(c))) context.Rip = uint64(dieTrampolineAddr) } + +// getHypercallID returns hypercall ID. +// +//go:nosplit +func getHypercallID(addr uintptr) int { + return _KVM_HYPERCALL_MAX +} diff --git a/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go b/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go index 4ca2b7717..abd36f973 100644 --- a/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go @@ -61,3 +61,16 @@ func dieArchSetup(c *vCPU, context *arch.SignalContext64, guestRegs *userRegs) { func bluepillArchFpContext(context unsafe.Pointer) *arch.FpsimdContext { return &((*arch.SignalContext64)(context).Fpsimd64) } + +// getHypercallID returns hypercall ID. +// +// On Arm64, the MMIO address should be 64-bit aligned. +// +//go:nosplit +func getHypercallID(addr uintptr) int { + if addr < arm64HypercallMMIOBase || addr >= (arm64HypercallMMIOBase+_AARCH64_HYPERCALL_MMIO_SIZE) { + return _KVM_HYPERCALL_MAX + } else { + return int(((addr) - arm64HypercallMMIOBase) >> 3) + } +} diff --git a/pkg/sentry/platform/kvm/bluepill_unsafe.go b/pkg/sentry/platform/kvm/bluepill_unsafe.go index 9add7c944..e79097b5d 100644 --- a/pkg/sentry/platform/kvm/bluepill_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_unsafe.go @@ -58,6 +58,24 @@ func bluepillArchContext(context unsafe.Pointer) *arch.SignalContext64 { return &((*arch.UContext64)(context).MContext) } +// bluepillHandleHlt is reponsible for handling VM-Exit. +// +//go:nosplit +func bluepillGuestExit(c *vCPU, context unsafe.Pointer) { + // Copy out registers. + bluepillArchExit(c, bluepillArchContext(context)) + + // Return to the vCPUReady state; notify any waiters. + user := atomic.LoadUint32(&c.state) & vCPUUser + switch atomic.SwapUint32(&c.state, user) { + case user | vCPUGuest: // Expected case. + case user | vCPUGuest | vCPUWaiter: + c.notify() + default: + throw("invalid state") + } +} + // bluepillHandler is called from the signal stub. // // The world may be stopped while this is executing, and it executes on the @@ -156,25 +174,20 @@ func bluepillHandler(context unsafe.Pointer) { c.die(bluepillArchContext(context), "debug") return case _KVM_EXIT_HLT: - // Copy out registers. - bluepillArchExit(c, bluepillArchContext(context)) - - // Return to the vCPUReady state; notify any waiters. - user := atomic.LoadUint32(&c.state) & vCPUUser - switch atomic.SwapUint32(&c.state, user) { - case user | vCPUGuest: // Expected case. - case user | vCPUGuest | vCPUWaiter: - c.notify() - default: - throw("invalid state") - } + bluepillGuestExit(c, context) return case _KVM_EXIT_MMIO: + physical := uintptr(c.runData.data[0]) + if getHypercallID(physical) == _KVM_HYPERCALL_VMEXIT { + bluepillGuestExit(c, context) + return + } + // Increment the fault count. atomic.AddUint32(&c.faults, 1) // For MMIO, the physical address is the first data item. - physical := uintptr(c.runData.data[0]) + physical = uintptr(c.runData.data[0]) virtual, ok := handleBluepillFault(c.machine, physical, physicalRegions, _KVM_MEM_FLAGS_NONE) if !ok { c.die(bluepillArchContext(context), "invalid physical address") diff --git a/pkg/sentry/platform/kvm/kvm_arm64.go b/pkg/sentry/platform/kvm/kvm_arm64.go index 29d457a7e..3134a076b 100644 --- a/pkg/sentry/platform/kvm/kvm_arm64.go +++ b/pkg/sentry/platform/kvm/kvm_arm64.go @@ -26,6 +26,9 @@ type kvmOneReg struct { addr uint64 } +// arm64HypercallMMIOBase is MMIO base address used to dispatch hypercalls. +var arm64HypercallMMIOBase uintptr + const KVM_NR_SPSR = 5 type userFpsimdState struct { diff --git a/pkg/sentry/platform/kvm/kvm_const.go b/pkg/sentry/platform/kvm/kvm_const.go index 1d5c77ff4..6a7676468 100644 --- a/pkg/sentry/platform/kvm/kvm_const.go +++ b/pkg/sentry/platform/kvm/kvm_const.go @@ -71,3 +71,13 @@ const ( _KVM_MEM_READONLY = uint32(1) << 1 _KVM_MEM_FLAGS_NONE = 0 ) + +// KVM hypercall list. +// Canonical list of hypercalls supported. +const ( + // On amd64, it uses 'HLT' to leave the guest. + // Unlike amd64, arm64 can only uses mmio_exit/psci to leave the guest. + // _KVM_HYPERCALL_VMEXIT is only used on Arm64 for now. + _KVM_HYPERCALL_VMEXIT int = iota + _KVM_HYPERCALL_MAX +) diff --git a/pkg/sentry/platform/kvm/kvm_const_arm64.go b/pkg/sentry/platform/kvm/kvm_const_arm64.go index 531ae8b1e..6f0539c29 100644 --- a/pkg/sentry/platform/kvm/kvm_const_arm64.go +++ b/pkg/sentry/platform/kvm/kvm_const_arm64.go @@ -131,3 +131,10 @@ const ( _ESR_SEGV_PEMERR_L2 = 0xe _ESR_SEGV_PEMERR_L3 = 0xf ) + +// Arm64: MMIO base address used to dispatch hypercalls. +const ( + // on Arm64, the MMIO address must be 64-bit aligned. + // Currently, we only need 1 hypercall: hypercall_vmexit. + _AARCH64_HYPERCALL_MMIO_SIZE = 1 << 3 +) diff --git a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go index 3c02cef7c..48c834499 100644 --- a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go @@ -159,6 +159,10 @@ func (c *vCPU) initArchState() error { return err } + // Use the address of the exception vector table as + // the MMIO address base. + arm64HypercallMMIOBase = toLocation + data = ring0.PsrDefaultSet | ring0.KernelFlagsSet reg.id = _KVM_ARM64_REGS_PSTATE if err := c.setOneRegister(®); err != nil { diff --git a/pkg/sentry/platform/ring0/entry_arm64.s b/pkg/sentry/platform/ring0/entry_arm64.s index db6465663..2bc5f3ecd 100644 --- a/pkg/sentry/platform/ring0/entry_arm64.s +++ b/pkg/sentry/platform/ring0/entry_arm64.s @@ -362,9 +362,17 @@ mmio_exit: MOVD R1, CPU_LAZY_VFP(RSV_REG) VFP_DISABLE - // MMIO_EXIT. - MOVD $0, R9 - MOVD R0, 0xffff000000001000(R9) + // Trigger MMIO_EXIT/_KVM_HYPERCALL_VMEXIT. + // + // To keep it simple, I used the address of exception table as the + // MMIO base address, so that I can trigger a MMIO-EXIT by forcibly writing + // a read-only space. + // Also, the length is engough to match a sufficient number of hypercall ID. + // Then, in host user space, I can calculate this address to find out + // which hypercall. + MRS VBAR_EL1, R9 + MOVD R0, 0x0(R9) + RET // HaltAndResume halts execution and point the pointer to the resume function. -- cgit v1.2.3 From 20e6efd302746554c485028f9fc1f2fbf88b234e Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Mon, 18 May 2020 14:24:47 -0700 Subject: Remove IfChange/ThenChange lint from VFS2 As new functionality is added to VFS2, corresponding files in VFS1 don't need to be changed. PiperOrigin-RevId: 312153799 --- pkg/sentry/fsimpl/devpts/line_discipline.go | 4 ---- pkg/sentry/fsimpl/devpts/master.go | 4 ---- pkg/sentry/fsimpl/devpts/queue.go | 4 ---- pkg/sentry/fsimpl/devpts/slave.go | 4 ---- pkg/sentry/fsimpl/devpts/terminal.go | 4 ---- 5 files changed, 20 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/devpts/line_discipline.go b/pkg/sentry/fsimpl/devpts/line_discipline.go index e201801d6..f7bc325d1 100644 --- a/pkg/sentry/fsimpl/devpts/line_discipline.go +++ b/pkg/sentry/fsimpl/devpts/line_discipline.go @@ -27,8 +27,6 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) -// LINT.IfChange - const ( // canonMaxBytes is the number of bytes that fit into a single line of // terminal input in canonical mode. This corresponds to N_TTY_BUF_SIZE @@ -445,5 +443,3 @@ func (l *lineDiscipline) peek(b []byte) int { } return size } - -// LINT.ThenChange(../../fs/tty/line_discipline.go) diff --git a/pkg/sentry/fsimpl/devpts/master.go b/pkg/sentry/fsimpl/devpts/master.go index 04a292927..7a7ce5d81 100644 --- a/pkg/sentry/fsimpl/devpts/master.go +++ b/pkg/sentry/fsimpl/devpts/master.go @@ -27,8 +27,6 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) -// LINT.IfChange - // masterInode is the inode for the master end of the Terminal. type masterInode struct { kernfs.InodeAttrs @@ -222,5 +220,3 @@ func maybeEmitUnimplementedEvent(ctx context.Context, cmd uint32) { unimpl.EmitUnimplementedEvent(ctx) } } - -// LINT.ThenChange(../../fs/tty/master.go) diff --git a/pkg/sentry/fsimpl/devpts/queue.go b/pkg/sentry/fsimpl/devpts/queue.go index 29a6be858..dffb4232c 100644 --- a/pkg/sentry/fsimpl/devpts/queue.go +++ b/pkg/sentry/fsimpl/devpts/queue.go @@ -25,8 +25,6 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) -// LINT.IfChange - // waitBufMaxBytes is the maximum size of a wait buffer. It is based on // TTYB_DEFAULT_MEM_LIMIT. const waitBufMaxBytes = 131072 @@ -236,5 +234,3 @@ func (q *queue) waitBufAppend(b []byte) { q.waitBuf = append(q.waitBuf, b) q.waitBufLen += uint64(len(b)) } - -// LINT.ThenChange(../../fs/tty/queue.go) diff --git a/pkg/sentry/fsimpl/devpts/slave.go b/pkg/sentry/fsimpl/devpts/slave.go index 0a98dc896..526cd406c 100644 --- a/pkg/sentry/fsimpl/devpts/slave.go +++ b/pkg/sentry/fsimpl/devpts/slave.go @@ -26,8 +26,6 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) -// LINT.IfChange - // slaveInode is the inode for the slave end of the Terminal. type slaveInode struct { kernfs.InodeAttrs @@ -182,5 +180,3 @@ func (sfd *slaveFileDescription) Stat(ctx context.Context, opts vfs.StatOptions) fs := sfd.vfsfd.VirtualDentry().Mount().Filesystem() return sfd.inode.Stat(fs, opts) } - -// LINT.ThenChange(../../fs/tty/slave.go) diff --git a/pkg/sentry/fsimpl/devpts/terminal.go b/pkg/sentry/fsimpl/devpts/terminal.go index b44e673d8..7d2781c54 100644 --- a/pkg/sentry/fsimpl/devpts/terminal.go +++ b/pkg/sentry/fsimpl/devpts/terminal.go @@ -22,8 +22,6 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) -// LINT.IfChanges - // Terminal is a pseudoterminal. // // +stateify savable @@ -120,5 +118,3 @@ func (tm *Terminal) tty(isMaster bool) *kernel.TTY { } return tm.slaveKTTY } - -// LINT.ThenChange(../../fs/tty/terminal.go) -- cgit v1.2.3 From d06de1bede9c40078fe674a6df4849fe3322a861 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 19 May 2020 11:40:15 -0700 Subject: Fix flaky udp tests by polling before reading. On native Linux, calling recv/read right after send/write sometimes returns EWOULDBLOCK, if the data has not made it to the receiving socket (even though the endpoints are on the same host). Poll before reading to avoid this. Making this change also uncovered a hostinet bug (gvisor.dev/issue/2726), which is noted in this CL. PiperOrigin-RevId: 312320587 --- pkg/sentry/socket/hostinet/socket.go | 2 +- test/syscalls/linux/udp_socket_test_cases.cc | 48 +++++++++++++++++++++++++++- 2 files changed, 48 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/hostinet/socket.go b/pkg/sentry/socket/hostinet/socket.go index b49433326..c11e82c10 100644 --- a/pkg/sentry/socket/hostinet/socket.go +++ b/pkg/sentry/socket/hostinet/socket.go @@ -555,7 +555,7 @@ func (s *socketOpsCommon) SendMsg(t *kernel.Task, src usermem.IOSequence, to []b if uint64(src.NumBytes()) != srcs.NumBytes() { return 0, nil } - if srcs.IsEmpty() { + if srcs.IsEmpty() && len(controlBuf) == 0 { return 0, nil } diff --git a/test/syscalls/linux/udp_socket_test_cases.cc b/test/syscalls/linux/udp_socket_test_cases.cc index 740c7986d..4a3ae5f3f 100644 --- a/test/syscalls/linux/udp_socket_test_cases.cc +++ b/test/syscalls/linux/udp_socket_test_cases.cc @@ -17,6 +17,7 @@ #include #include #include +#include #include #include #include @@ -673,6 +674,11 @@ TEST_P(UdpSocketTest, ZerolengthWriteAllowed) { char buf[3]; // Send zero length packet from s_ to t_. ASSERT_THAT(write(s_, buf, 0), SyscallSucceedsWithValue(0)); + + struct pollfd pfd = {t_, POLLIN, 0}; + ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, /*timeout=*/1000), + SyscallSucceedsWithValue(1)); + // Receive the packet. char received[3]; EXPECT_THAT(read(t_, received, sizeof(received)), @@ -698,6 +704,11 @@ TEST_P(UdpSocketTest, ZerolengthWriteAllowedNonBlockRead) { char buf[3]; // Send zero length packet from s_ to t_. ASSERT_THAT(write(s_, buf, 0), SyscallSucceedsWithValue(0)); + + struct pollfd pfd = {t_, POLLIN, 0}; + ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, /*timeout=*/1000), + SyscallSucceedsWithValue(1)); + // Receive the packet. char received[3]; EXPECT_THAT(read(t_, received, sizeof(received)), @@ -859,6 +870,10 @@ TEST_P(UdpSocketTest, ReadShutdownNonblockPendingData) { EXPECT_THAT(shutdown(s_, SHUT_RD), SyscallSucceeds()); + struct pollfd pfd = {s_, POLLIN, 0}; + ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, /*timeout=*/1000), + SyscallSucceedsWithValue(1)); + // We should get the data even though read has been shutdown. EXPECT_THAT(recv(s_, received, 2, 0), SyscallSucceedsWithValue(2)); @@ -1112,6 +1127,10 @@ TEST_P(UdpSocketTest, FIONREADWriteShutdown) { ASSERT_THAT(send(s_, str, sizeof(str), 0), SyscallSucceedsWithValue(sizeof(str))); + struct pollfd pfd = {s_, POLLIN, 0}; + ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, /*timeout=*/1000), + SyscallSucceedsWithValue(1)); + n = -1; EXPECT_THAT(ioctl(s_, FIONREAD, &n), SyscallSucceedsWithValue(0)); EXPECT_EQ(n, sizeof(str)); @@ -1123,7 +1142,7 @@ TEST_P(UdpSocketTest, FIONREADWriteShutdown) { EXPECT_EQ(n, sizeof(str)); } -TEST_P(UdpSocketTest, Fionread) { +TEST_P(UdpSocketTest, FIONREAD) { // Bind s_ to loopback:TestPort. ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); @@ -1138,10 +1157,14 @@ TEST_P(UdpSocketTest, Fionread) { char buf[3 * psize]; RandomizeBuffer(buf, sizeof(buf)); + struct pollfd pfd = {s_, POLLIN, 0}; for (int i = 0; i < 3; ++i) { ASSERT_THAT(sendto(t_, buf + i * psize, psize, 0, addr_[0], addrlen_), SyscallSucceedsWithValue(psize)); + ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, /*timeout=*/1000), + SyscallSucceedsWithValue(1)); + // Check that regardless of how many packets are in the queue, the size // reported is that of a single packet. n = -1; @@ -1165,10 +1188,18 @@ TEST_P(UdpSocketTest, FIONREADZeroLengthPacket) { char buf[3 * psize]; RandomizeBuffer(buf, sizeof(buf)); + struct pollfd pfd = {s_, POLLIN, 0}; for (int i = 0; i < 3; ++i) { ASSERT_THAT(sendto(t_, buf + i * psize, 0, 0, addr_[0], addrlen_), SyscallSucceedsWithValue(0)); + // TODO(gvisor.dev/issue/2726): sending a zero-length message to a hostinet + // socket does not cause a poll event to be triggered. + if (!IsRunningWithHostinet()) { + ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, /*timeout=*/1000), + SyscallSucceedsWithValue(1)); + } + // Check that regardless of how many packets are in the queue, the size // reported is that of a single packet. n = -1; @@ -1235,6 +1266,10 @@ TEST_P(UdpSocketTest, SoTimestamp) { // Send zero length packet from t_ to s_. ASSERT_THAT(RetryEINTR(write)(t_, buf, 0), SyscallSucceedsWithValue(0)); + struct pollfd pfd = {s_, POLLIN, 0}; + ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, /*timeout=*/1000), + SyscallSucceedsWithValue(1)); + char cmsgbuf[CMSG_SPACE(sizeof(struct timeval))]; msghdr msg; memset(&msg, 0, sizeof(msg)); @@ -1278,6 +1313,10 @@ TEST_P(UdpSocketTest, TimestampIoctl) { ASSERT_THAT(RetryEINTR(write)(t_, buf, sizeof(buf)), SyscallSucceedsWithValue(sizeof(buf))); + struct pollfd pfd = {s_, POLLIN, 0}; + ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, /*timeout=*/1000), + SyscallSucceedsWithValue(1)); + // There should be no control messages. char recv_buf[sizeof(buf)]; ASSERT_NO_FATAL_FAILURE(RecvNoCmsg(s_, recv_buf, sizeof(recv_buf))); @@ -1315,6 +1354,10 @@ TEST_P(UdpSocketTest, TimestampIoctlPersistence) { SyscallSucceedsWithValue(sizeof(buf))); ASSERT_THAT(RetryEINTR(write)(t_, buf, 0), SyscallSucceedsWithValue(0)); + struct pollfd pfd = {s_, POLLIN, 0}; + ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, /*timeout=*/1000), + SyscallSucceedsWithValue(1)); + // There should be no control messages. char recv_buf[sizeof(buf)]; ASSERT_NO_FATAL_FAILURE(RecvNoCmsg(s_, recv_buf, sizeof(recv_buf))); @@ -1330,6 +1373,9 @@ TEST_P(UdpSocketTest, TimestampIoctlPersistence) { SyscallSucceeds()); ASSERT_THAT(RetryEINTR(write)(t_, buf, 0), SyscallSucceedsWithValue(0)); + ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, /*timeout=*/1000), + SyscallSucceedsWithValue(1)); + // There should be a message for SO_TIMESTAMP. char cmsgbuf[CMSG_SPACE(sizeof(struct timeval))]; msghdr msg = {}; -- cgit v1.2.3 From 05c89af6edde6158844d4debfe68bc598fec4418 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 19 May 2020 13:45:23 -0700 Subject: Implement mmap for host fs in vfs2. In VFS1, both fs/host and fs/gofer used the same utils for host file mappings. Refactor parts of fsimpl/gofer to create similar utils to share with fsimpl/host (memory accounting code moved to fsutil, page rounding arithmetic moved to usermem). Updates #1476. PiperOrigin-RevId: 312345090 --- pkg/sentry/fs/fsutil/frame_ref_set.go | 40 ++++++++++ pkg/sentry/fsimpl/gofer/BUILD | 1 - pkg/sentry/fsimpl/gofer/gofer.go | 4 +- pkg/sentry/fsimpl/gofer/pagemath.go | 31 -------- pkg/sentry/fsimpl/gofer/regular_file.go | 47 +++--------- pkg/sentry/fsimpl/host/BUILD | 4 + pkg/sentry/fsimpl/host/host.go | 70 ++++++++++++----- pkg/sentry/fsimpl/host/mmap.go | 132 ++++++++++++++++++++++++++++++++ pkg/usermem/addr.go | 17 ++++ 9 files changed, 254 insertions(+), 92 deletions(-) delete mode 100644 pkg/sentry/fsimpl/gofer/pagemath.go create mode 100644 pkg/sentry/fsimpl/host/mmap.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/fsutil/frame_ref_set.go b/pkg/sentry/fs/fsutil/frame_ref_set.go index 6564fd0c6..dd6f5aba6 100644 --- a/pkg/sentry/fs/fsutil/frame_ref_set.go +++ b/pkg/sentry/fs/fsutil/frame_ref_set.go @@ -18,6 +18,7 @@ import ( "math" "gvisor.dev/gvisor/pkg/sentry/platform" + "gvisor.dev/gvisor/pkg/sentry/usage" ) // FrameRefSetFunctions implements segment.Functions for FrameRefSet. @@ -49,3 +50,42 @@ func (FrameRefSetFunctions) Merge(_ platform.FileRange, val1 uint64, _ platform. func (FrameRefSetFunctions) Split(_ platform.FileRange, val uint64, _ uint64) (uint64, uint64) { return val, val } + +// IncRefAndAccount adds a reference on the range fr. All newly inserted segments +// are accounted as host page cache memory mappings. +func (refs *FrameRefSet) IncRefAndAccount(fr platform.FileRange) { + seg, gap := refs.Find(fr.Start) + for { + switch { + case seg.Ok() && seg.Start() < fr.End: + seg = refs.Isolate(seg, fr) + seg.SetValue(seg.Value() + 1) + seg, gap = seg.NextNonEmpty() + case gap.Ok() && gap.Start() < fr.End: + newRange := gap.Range().Intersect(fr) + usage.MemoryAccounting.Inc(newRange.Length(), usage.Mapped) + seg, gap = refs.InsertWithoutMerging(gap, newRange, 1).NextNonEmpty() + default: + refs.MergeAdjacent(fr) + return + } + } +} + +// DecRefAndAccount removes a reference on the range fr and untracks segments +// that are removed from memory accounting. +func (refs *FrameRefSet) DecRefAndAccount(fr platform.FileRange) { + seg := refs.FindSegment(fr.Start) + + for seg.Ok() && seg.Start() < fr.End { + seg = refs.Isolate(seg, fr) + if old := seg.Value(); old == 1 { + usage.MemoryAccounting.Dec(seg.Range().Length(), usage.Mapped) + seg = refs.Remove(seg).NextSegment() + } else { + seg.SetValue(old - 1) + seg = seg.NextSegment() + } + } + refs.MergeAdjacent(fr) +} diff --git a/pkg/sentry/fsimpl/gofer/BUILD b/pkg/sentry/fsimpl/gofer/BUILD index 5ce82b793..67e916525 100644 --- a/pkg/sentry/fsimpl/gofer/BUILD +++ b/pkg/sentry/fsimpl/gofer/BUILD @@ -36,7 +36,6 @@ go_library( "gofer.go", "handle.go", "p9file.go", - "pagemath.go", "regular_file.go", "socket.go", "special_file.go", diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index ebf063a58..6295f6b54 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -928,8 +928,8 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin // so we can't race with Write or another truncate.) d.dataMu.Unlock() if d.size < oldSize { - oldpgend := pageRoundUp(oldSize) - newpgend := pageRoundUp(d.size) + oldpgend, _ := usermem.PageRoundUp(oldSize) + newpgend, _ := usermem.PageRoundUp(d.size) if oldpgend != newpgend { d.mapsMu.Lock() d.mappings.Invalidate(memmap.MappableRange{newpgend, oldpgend}, memmap.InvalidateOpts{ diff --git a/pkg/sentry/fsimpl/gofer/pagemath.go b/pkg/sentry/fsimpl/gofer/pagemath.go deleted file mode 100644 index 847cb0784..000000000 --- a/pkg/sentry/fsimpl/gofer/pagemath.go +++ /dev/null @@ -1,31 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package gofer - -import ( - "gvisor.dev/gvisor/pkg/usermem" -) - -// This are equivalent to usermem.Addr.RoundDown/Up, but without the -// potentially truncating conversion to usermem.Addr. This is necessary because -// there is no way to define generic "PageRoundDown/Up" functions in Go. - -func pageRoundDown(x uint64) uint64 { - return x &^ (usermem.PageSize - 1) -} - -func pageRoundUp(x uint64) uint64 { - return pageRoundDown(x + usermem.PageSize - 1) -} diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go index 857f7c74e..0d10cf7ac 100644 --- a/pkg/sentry/fsimpl/gofer/regular_file.go +++ b/pkg/sentry/fsimpl/gofer/regular_file.go @@ -148,9 +148,9 @@ func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off return 0, err } // Remove touched pages from the cache. - pgstart := pageRoundDown(uint64(offset)) - pgend := pageRoundUp(uint64(offset + src.NumBytes())) - if pgend < pgstart { + pgstart := usermem.PageRoundDown(uint64(offset)) + pgend, ok := usermem.PageRoundUp(uint64(offset + src.NumBytes())) + if !ok { return 0, syserror.EINVAL } mr := memmap.MappableRange{pgstart, pgend} @@ -306,9 +306,10 @@ func (rw *dentryReadWriter) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) if fillCache { // Read into the cache, then re-enter the loop to read from the // cache. + gapEnd, _ := usermem.PageRoundUp(gapMR.End) reqMR := memmap.MappableRange{ - Start: pageRoundDown(gapMR.Start), - End: pageRoundUp(gapMR.End), + Start: usermem.PageRoundDown(gapMR.Start), + End: gapEnd, } optMR := gap.Range() err := rw.d.cache.Fill(rw.ctx, reqMR, maxFillRange(reqMR, optMR), mf, usage.PageCache, rw.d.handle.readToBlocksAt) @@ -671,7 +672,7 @@ func (d *dentry) Translate(ctx context.Context, required, optional memmap.Mappab // Constrain translations to d.size (rounded up) to prevent translation to // pages that may be concurrently truncated. - pgend := pageRoundUp(d.size) + pgend, _ := usermem.PageRoundUp(d.size) var beyondEOF bool if required.End > pgend { if required.Start >= pgend { @@ -818,43 +819,15 @@ type dentryPlatformFile struct { // IncRef implements platform.File.IncRef. func (d *dentryPlatformFile) IncRef(fr platform.FileRange) { d.dataMu.Lock() - seg, gap := d.fdRefs.Find(fr.Start) - for { - switch { - case seg.Ok() && seg.Start() < fr.End: - seg = d.fdRefs.Isolate(seg, fr) - seg.SetValue(seg.Value() + 1) - seg, gap = seg.NextNonEmpty() - case gap.Ok() && gap.Start() < fr.End: - newRange := gap.Range().Intersect(fr) - usage.MemoryAccounting.Inc(newRange.Length(), usage.Mapped) - seg, gap = d.fdRefs.InsertWithoutMerging(gap, newRange, 1).NextNonEmpty() - default: - d.fdRefs.MergeAdjacent(fr) - d.dataMu.Unlock() - return - } - } + d.fdRefs.IncRefAndAccount(fr) + d.dataMu.Unlock() } // DecRef implements platform.File.DecRef. func (d *dentryPlatformFile) DecRef(fr platform.FileRange) { d.dataMu.Lock() - seg := d.fdRefs.FindSegment(fr.Start) - - for seg.Ok() && seg.Start() < fr.End { - seg = d.fdRefs.Isolate(seg, fr) - if old := seg.Value(); old == 1 { - usage.MemoryAccounting.Dec(seg.Range().Length(), usage.Mapped) - seg = d.fdRefs.Remove(seg).NextSegment() - } else { - seg.SetValue(old - 1) - seg = seg.NextSegment() - } - } - d.fdRefs.MergeAdjacent(fr) + d.fdRefs.DecRefAndAccount(fr) d.dataMu.Unlock() - } // MapInternal implements platform.File.MapInternal. diff --git a/pkg/sentry/fsimpl/host/BUILD b/pkg/sentry/fsimpl/host/BUILD index 39509f703..ca0fe6d2b 100644 --- a/pkg/sentry/fsimpl/host/BUILD +++ b/pkg/sentry/fsimpl/host/BUILD @@ -8,6 +8,7 @@ go_library( "control.go", "host.go", "ioctl_unsafe.go", + "mmap.go", "socket.go", "socket_iovec.go", "socket_unsafe.go", @@ -23,12 +24,15 @@ go_library( "//pkg/fspath", "//pkg/log", "//pkg/refs", + "//pkg/safemem", "//pkg/sentry/arch", + "//pkg/sentry/fs/fsutil", "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/hostfd", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/memmap", + "//pkg/sentry/platform", "//pkg/sentry/socket/control", "//pkg/sentry/socket/unix", "//pkg/sentry/socket/unix/transport", diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 8caf55a1b..65981197d 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -86,15 +86,16 @@ func NewFD(ctx context.Context, mnt *vfs.Mount, hostFD int, opts *NewFDOptions) i := &inode{ hostFD: hostFD, - seekable: seekable, + ino: fs.NextIno(), isTTY: opts.IsTTY, - canMap: canMap(uint32(fileType)), wouldBlock: wouldBlock(uint32(fileType)), - ino: fs.NextIno(), + seekable: seekable, // For simplicity, set offset to 0. Technically, we should use the existing // offset on the host if the file is seekable. offset: 0, + canMap: canMap(uint32(fileType)), } + i.pf.inode = i // Non-seekable files can't be memory mapped, assert this. if !i.seekable && i.canMap { @@ -189,11 +190,15 @@ type inode struct { // This field is initialized at creation time and is immutable. hostFD int - // wouldBlock is true if the host FD would return EWOULDBLOCK for - // operations that would block. + // ino is an inode number unique within this filesystem. // // This field is initialized at creation time and is immutable. - wouldBlock bool + ino uint64 + + // isTTY is true if this file represents a TTY. + // + // This field is initialized at creation time and is immutable. + isTTY bool // seekable is false if the host fd points to a file representing a stream, // e.g. a socket or a pipe. Such files are not seekable and can return @@ -202,29 +207,36 @@ type inode struct { // This field is initialized at creation time and is immutable. seekable bool - // isTTY is true if this file represents a TTY. + // offsetMu protects offset. + offsetMu sync.Mutex + + // offset specifies the current file offset. It is only meaningful when + // seekable is true. + offset int64 + + // wouldBlock is true if the host FD would return EWOULDBLOCK for + // operations that would block. // // This field is initialized at creation time and is immutable. - isTTY bool + wouldBlock bool + + // Event queue for blocking operations. + queue waiter.Queue // canMap specifies whether we allow the file to be memory mapped. // // This field is initialized at creation time and is immutable. canMap bool - // ino is an inode number unique within this filesystem. - // - // This field is initialized at creation time and is immutable. - ino uint64 - - // offsetMu protects offset. - offsetMu sync.Mutex + // mapsMu protects mappings. + mapsMu sync.Mutex - // offset specifies the current file offset. - offset int64 + // If canMap is true, mappings tracks mappings of hostFD into + // memmap.MappingSpaces. + mappings memmap.MappingSet - // Event queue for blocking operations. - queue waiter.Queue + // pf implements platform.File for mappings of hostFD. + pf inodePlatformFile } // CheckPermissions implements kernfs.Inode. @@ -388,6 +400,21 @@ func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Cre if err := syscall.Ftruncate(i.hostFD, int64(s.Size)); err != nil { return err } + oldSize := uint64(hostStat.Size) + if s.Size < oldSize { + oldpgend, _ := usermem.PageRoundUp(oldSize) + newpgend, _ := usermem.PageRoundUp(s.Size) + if oldpgend != newpgend { + i.mapsMu.Lock() + i.mappings.Invalidate(memmap.MappableRange{newpgend, oldpgend}, memmap.InvalidateOpts{ + // Compare Linux's mm/truncate.c:truncate_setsize() => + // truncate_pagecache() => + // mm/memory.c:unmap_mapping_range(evencows=1). + InvalidatePrivate: true, + }) + i.mapsMu.Unlock() + } + } } if m&(linux.STATX_ATIME|linux.STATX_MTIME) != 0 { ts := [2]syscall.Timespec{ @@ -666,8 +693,9 @@ func (f *fileDescription) ConfigureMMap(_ context.Context, opts *memmap.MMapOpts if !f.inode.canMap { return syserror.ENODEV } - // TODO(gvisor.dev/issue/1672): Implement ConfigureMMap and Mappable interface. - return syserror.ENODEV + i := f.inode + i.pf.fileMapperInitOnce.Do(i.pf.fileMapper.Init) + return vfs.GenericConfigureMMap(&f.vfsfd, i, opts) } // EventRegister implements waiter.Waitable.EventRegister. diff --git a/pkg/sentry/fsimpl/host/mmap.go b/pkg/sentry/fsimpl/host/mmap.go new file mode 100644 index 000000000..8545a82f0 --- /dev/null +++ b/pkg/sentry/fsimpl/host/mmap.go @@ -0,0 +1,132 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package host + +import ( + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/safemem" + "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" + "gvisor.dev/gvisor/pkg/sentry/memmap" + "gvisor.dev/gvisor/pkg/sentry/platform" + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/usermem" +) + +// inodePlatformFile implements platform.File. It exists solely because inode +// cannot implement both kernfs.Inode.IncRef and platform.File.IncRef. +// +// inodePlatformFile should only be used if inode.canMap is true. +type inodePlatformFile struct { + *inode + + // fdRefsMu protects fdRefs. + fdRefsMu sync.Mutex + + // fdRefs counts references on platform.File offsets. It is used solely for + // memory accounting. + fdRefs fsutil.FrameRefSet + + // fileMapper caches mappings of the host file represented by this inode. + fileMapper fsutil.HostFileMapper + + // fileMapperInitOnce is used to lazily initialize fileMapper. + fileMapperInitOnce sync.Once +} + +// IncRef implements platform.File.IncRef. +// +// Precondition: i.inode.canMap must be true. +func (i *inodePlatformFile) IncRef(fr platform.FileRange) { + i.fdRefsMu.Lock() + i.fdRefs.IncRefAndAccount(fr) + i.fdRefsMu.Unlock() +} + +// DecRef implements platform.File.DecRef. +// +// Precondition: i.inode.canMap must be true. +func (i *inodePlatformFile) DecRef(fr platform.FileRange) { + i.fdRefsMu.Lock() + i.fdRefs.DecRefAndAccount(fr) + i.fdRefsMu.Unlock() +} + +// MapInternal implements platform.File.MapInternal. +// +// Precondition: i.inode.canMap must be true. +func (i *inodePlatformFile) MapInternal(fr platform.FileRange, at usermem.AccessType) (safemem.BlockSeq, error) { + return i.fileMapper.MapInternal(fr, i.hostFD, at.Write) +} + +// FD implements platform.File.FD. +func (i *inodePlatformFile) FD() int { + return i.hostFD +} + +// AddMapping implements memmap.Mappable.AddMapping. +// +// Precondition: i.inode.canMap must be true. +func (i *inode) AddMapping(ctx context.Context, ms memmap.MappingSpace, ar usermem.AddrRange, offset uint64, writable bool) error { + i.mapsMu.Lock() + mapped := i.mappings.AddMapping(ms, ar, offset, writable) + for _, r := range mapped { + i.pf.fileMapper.IncRefOn(r) + } + i.mapsMu.Unlock() + return nil +} + +// RemoveMapping implements memmap.Mappable.RemoveMapping. +// +// Precondition: i.inode.canMap must be true. +func (i *inode) RemoveMapping(ctx context.Context, ms memmap.MappingSpace, ar usermem.AddrRange, offset uint64, writable bool) { + i.mapsMu.Lock() + unmapped := i.mappings.RemoveMapping(ms, ar, offset, writable) + for _, r := range unmapped { + i.pf.fileMapper.DecRefOn(r) + } + i.mapsMu.Unlock() +} + +// CopyMapping implements memmap.Mappable.CopyMapping. +// +// Precondition: i.inode.canMap must be true. +func (i *inode) CopyMapping(ctx context.Context, ms memmap.MappingSpace, srcAR, dstAR usermem.AddrRange, offset uint64, writable bool) error { + return i.AddMapping(ctx, ms, dstAR, offset, writable) +} + +// Translate implements memmap.Mappable.Translate. +// +// Precondition: i.inode.canMap must be true. +func (i *inode) Translate(ctx context.Context, required, optional memmap.MappableRange, at usermem.AccessType) ([]memmap.Translation, error) { + mr := optional + return []memmap.Translation{ + { + Source: mr, + File: &i.pf, + Offset: mr.Start, + Perms: usermem.AnyAccess, + }, + }, nil +} + +// InvalidateUnsavable implements memmap.Mappable.InvalidateUnsavable. +// +// Precondition: i.inode.canMap must be true. +func (i *inode) InvalidateUnsavable(ctx context.Context) error { + // We expect the same host fd across save/restore, so all translations + // should be valid. + return nil +} diff --git a/pkg/usermem/addr.go b/pkg/usermem/addr.go index e79210804..c4100481e 100644 --- a/pkg/usermem/addr.go +++ b/pkg/usermem/addr.go @@ -106,3 +106,20 @@ func (ar AddrRange) IsPageAligned() bool { func (ar AddrRange) String() string { return fmt.Sprintf("[%#x, %#x)", ar.Start, ar.End) } + +// PageRoundDown/Up are equivalent to Addr.RoundDown/Up, but without the +// potentially truncating conversion from uint64 to Addr. This is necessary +// because there is no way to define generic "PageRoundDown/Up" functions in Go. + +// PageRoundDown returns x rounded down to the nearest page boundary. +func PageRoundDown(x uint64) uint64 { + return x &^ (PageSize - 1) +} + +// PageRoundUp returns x rounded up to the nearest page boundary. +// ok is true iff rounding up did not wrap around. +func PageRoundUp(x uint64) (addr uint64, ok bool) { + addr = PageRoundDown(x + PageSize - 1) + ok = addr >= x + return +} -- cgit v1.2.3 From 76369b6480b69bb7bf680db36ce8787fb324911c Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Wed, 20 May 2020 14:49:40 -0700 Subject: Move fsimpl/host file offset from inode to fileDescription. PiperOrigin-RevId: 312559861 --- pkg/sentry/fsimpl/host/host.go | 78 ++++++++++++++++++++---------------------- 1 file changed, 38 insertions(+), 40 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 65981197d..18b127521 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -90,10 +90,7 @@ func NewFD(ctx context.Context, mnt *vfs.Mount, hostFD int, opts *NewFDOptions) isTTY: opts.IsTTY, wouldBlock: wouldBlock(uint32(fileType)), seekable: seekable, - // For simplicity, set offset to 0. Technically, we should use the existing - // offset on the host if the file is seekable. - offset: 0, - canMap: canMap(uint32(fileType)), + canMap: canMap(uint32(fileType)), } i.pf.inode = i @@ -118,6 +115,10 @@ func NewFD(ctx context.Context, mnt *vfs.Mount, hostFD int, opts *NewFDOptions) // i.open will take a reference on d. defer d.DecRef() + + // For simplicity, fileDescription.offset is set to 0. Technically, we + // should only set to 0 on files that are not seekable (sockets, pipes, + // etc.), and use the offset from the host fd otherwise when importing. return i.open(ctx, d.VFSDentry(), mnt, flags) } @@ -207,13 +208,6 @@ type inode struct { // This field is initialized at creation time and is immutable. seekable bool - // offsetMu protects offset. - offsetMu sync.Mutex - - // offset specifies the current file offset. It is only meaningful when - // seekable is true. - offset int64 - // wouldBlock is true if the host FD would return EWOULDBLOCK for // operations that would block. // @@ -491,9 +485,6 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount, flags u return vfsfd, nil } - // For simplicity, set offset to 0. Technically, we should - // only set to 0 on files that are not seekable (sockets, pipes, etc.), - // and use the offset from the host fd otherwise. fd := &fileDescription{inode: i} vfsfd := &fd.vfsfd if err := vfsfd.Init(fd, flags, mnt, d, &vfs.FileDescriptionOptions{}); err != nil { @@ -514,6 +505,13 @@ type fileDescription struct { // // inode is immutable after fileDescription creation. inode *inode + + // offsetMu protects offset. + offsetMu sync.Mutex + + // offset specifies the current file offset. It is only meaningful when + // inode.seekable is true. + offset int64 } // SetStat implements vfs.FileDescriptionImpl. @@ -559,10 +557,10 @@ func (f *fileDescription) Read(ctx context.Context, dst usermem.IOSequence, opts return n, err } // TODO(gvisor.dev/issue/1672): Cache pages, when forced to do so. - i.offsetMu.Lock() - n, err := readFromHostFD(ctx, i.hostFD, dst, i.offset, opts.Flags) - i.offset += n - i.offsetMu.Unlock() + f.offsetMu.Lock() + n, err := readFromHostFD(ctx, i.hostFD, dst, f.offset, opts.Flags) + f.offset += n + f.offsetMu.Unlock() return n, err } @@ -599,10 +597,10 @@ func (f *fileDescription) Write(ctx context.Context, src usermem.IOSequence, opt } // TODO(gvisor.dev/issue/1672): Cache pages, when forced to do so. // TODO(gvisor.dev/issue/1672): Write to end of file and update offset if O_APPEND is set on this file. - i.offsetMu.Lock() - n, err := writeToHostFD(ctx, i.hostFD, src, i.offset, opts.Flags) - i.offset += n - i.offsetMu.Unlock() + f.offsetMu.Lock() + n, err := writeToHostFD(ctx, i.hostFD, src, f.offset, opts.Flags) + f.offset += n + f.offsetMu.Unlock() return n, err } @@ -627,41 +625,41 @@ func (f *fileDescription) Seek(_ context.Context, offset int64, whence int32) (i return 0, syserror.ESPIPE } - i.offsetMu.Lock() - defer i.offsetMu.Unlock() + f.offsetMu.Lock() + defer f.offsetMu.Unlock() switch whence { case linux.SEEK_SET: if offset < 0 { - return i.offset, syserror.EINVAL + return f.offset, syserror.EINVAL } - i.offset = offset + f.offset = offset case linux.SEEK_CUR: - // Check for overflow. Note that underflow cannot occur, since i.offset >= 0. - if offset > math.MaxInt64-i.offset { - return i.offset, syserror.EOVERFLOW + // Check for overflow. Note that underflow cannot occur, since f.offset >= 0. + if offset > math.MaxInt64-f.offset { + return f.offset, syserror.EOVERFLOW } - if i.offset+offset < 0 { - return i.offset, syserror.EINVAL + if f.offset+offset < 0 { + return f.offset, syserror.EINVAL } - i.offset += offset + f.offset += offset case linux.SEEK_END: var s syscall.Stat_t if err := syscall.Fstat(i.hostFD, &s); err != nil { - return i.offset, err + return f.offset, err } size := s.Size // Check for overflow. Note that underflow cannot occur, since size >= 0. if offset > math.MaxInt64-size { - return i.offset, syserror.EOVERFLOW + return f.offset, syserror.EOVERFLOW } if size+offset < 0 { - return i.offset, syserror.EINVAL + return f.offset, syserror.EINVAL } - i.offset = size + offset + f.offset = size + offset case linux.SEEK_DATA, linux.SEEK_HOLE: // Modifying the offset in the host file table should not matter, since @@ -670,16 +668,16 @@ func (f *fileDescription) Seek(_ context.Context, offset int64, whence int32) (i // For reading and writing, we always rely on our internal offset. n, err := unix.Seek(i.hostFD, offset, int(whence)) if err != nil { - return i.offset, err + return f.offset, err } - i.offset = n + f.offset = n default: // Invalid whence. - return i.offset, syserror.EINVAL + return f.offset, syserror.EINVAL } - return i.offset, nil + return f.offset, nil } // Sync implements FileDescriptionImpl. -- cgit v1.2.3 From 059879e14301660c9fce1e5e59bdfaef89fc4aaf Mon Sep 17 00:00:00 2001 From: Reapor-Yurnero Date: Wed, 20 May 2020 22:48:41 -0700 Subject: Implement gap tracking in the segment set. This change was derived from a change by: Reapor-Yurnero And has been modified by: Adin Scannell (The original change author is preserved for the commit.) This change implements gap tracking in the segment set by adding additional information in each node, and using that information to speed up gap finding from a linear scan to a O(log(n)) walk of the tree. This gap tracking is optional, and will default to off except for segment instances that set gapTracking equal to 1 in their const lists. PiperOrigin-RevId: 312621607 --- pkg/segment/BUILD | 2 + pkg/segment/set.go | 400 +++++++++++++++++++++++++++++++++++++- pkg/segment/test/BUILD | 18 ++ pkg/segment/test/segment_test.go | 397 ++++++++++++++++++++++++++++++++----- pkg/segment/test/set_functions.go | 32 ++- pkg/sentry/mm/BUILD | 1 + pkg/sentry/mm/vma.go | 4 +- tools/go_generics/generics.go | 4 +- 8 files changed, 792 insertions(+), 66 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/segment/BUILD b/pkg/segment/BUILD index 1b487b887..f57ccc170 100644 --- a/pkg/segment/BUILD +++ b/pkg/segment/BUILD @@ -21,6 +21,8 @@ go_template( ], opt_consts = [ "minDegree", + # trackGaps must either be 0 or 1. + "trackGaps", ], types = [ "Key", diff --git a/pkg/segment/set.go b/pkg/segment/set.go index 03e4f258f..1a17ad9cb 100644 --- a/pkg/segment/set.go +++ b/pkg/segment/set.go @@ -36,6 +36,34 @@ type Range interface{} // Value is a required type parameter. type Value interface{} +// trackGaps is an optional parameter. +// +// If trackGaps is 1, the Set will track maximum gap size recursively, +// enabling the GapIterator.{Prev,Next}LargeEnoughGap functions. In this +// case, Key must be an unsigned integer. +// +// trackGaps must be 0 or 1. +const trackGaps = 0 + +var _ = uint8(trackGaps << 7) // Will fail if not zero or one. + +// dynamicGap is a type that disappears if trackGaps is 0. +type dynamicGap [trackGaps]Key + +// Get returns the value of the gap. +// +// Precondition: trackGaps must be non-zero. +func (d *dynamicGap) Get() Key { + return d[:][0] +} + +// Set sets the value of the gap. +// +// Precondition: trackGaps must be non-zero. +func (d *dynamicGap) Set(v Key) { + d[:][0] = v +} + // Functions is a required type parameter that must be a struct implementing // the methods defined by Functions. type Functions interface { @@ -327,8 +355,12 @@ func (s *Set) Insert(gap GapIterator, r Range, val Value) Iterator { } if prev.Ok() && prev.End() == r.Start { if mval, ok := (Functions{}).Merge(prev.Range(), prev.Value(), r, val); ok { + shrinkMaxGap := trackGaps != 0 && gap.Range().Length() == gap.node.maxGap.Get() prev.SetEndUnchecked(r.End) prev.SetValue(mval) + if shrinkMaxGap { + gap.node.updateMaxGapLeaf() + } if next.Ok() && next.Start() == r.End { val = mval if mval, ok := (Functions{}).Merge(prev.Range(), val, next.Range(), next.Value()); ok { @@ -342,11 +374,16 @@ func (s *Set) Insert(gap GapIterator, r Range, val Value) Iterator { } if next.Ok() && next.Start() == r.End { if mval, ok := (Functions{}).Merge(r, val, next.Range(), next.Value()); ok { + shrinkMaxGap := trackGaps != 0 && gap.Range().Length() == gap.node.maxGap.Get() next.SetStartUnchecked(r.Start) next.SetValue(mval) + if shrinkMaxGap { + gap.node.updateMaxGapLeaf() + } return next } } + // InsertWithoutMergingUnchecked will maintain maxGap if necessary. return s.InsertWithoutMergingUnchecked(gap, r, val) } @@ -373,11 +410,15 @@ func (s *Set) InsertWithoutMerging(gap GapIterator, r Range, val Value) Iterator // Preconditions: r.Start >= gap.Start(); r.End <= gap.End(). func (s *Set) InsertWithoutMergingUnchecked(gap GapIterator, r Range, val Value) Iterator { gap = gap.node.rebalanceBeforeInsert(gap) + splitMaxGap := trackGaps != 0 && (gap.node.nrSegments == 0 || gap.Range().Length() == gap.node.maxGap.Get()) copy(gap.node.keys[gap.index+1:], gap.node.keys[gap.index:gap.node.nrSegments]) copy(gap.node.values[gap.index+1:], gap.node.values[gap.index:gap.node.nrSegments]) gap.node.keys[gap.index] = r gap.node.values[gap.index] = val gap.node.nrSegments++ + if splitMaxGap { + gap.node.updateMaxGapLeaf() + } return Iterator{gap.node, gap.index} } @@ -399,12 +440,23 @@ func (s *Set) Remove(seg Iterator) GapIterator { // overlap. seg.SetRangeUnchecked(victim.Range()) seg.SetValue(victim.Value()) + // Need to update the nextAdjacentNode's maxGap because the gap in between + // must have been modified by updating seg.Range() to victim.Range(). + // seg.NextSegment() must exist since the last segment can't be in a + // non-leaf node. + nextAdjacentNode := seg.NextSegment().node + if trackGaps != 0 { + nextAdjacentNode.updateMaxGapLeaf() + } return s.Remove(victim).NextGap() } copy(seg.node.keys[seg.index:], seg.node.keys[seg.index+1:seg.node.nrSegments]) copy(seg.node.values[seg.index:], seg.node.values[seg.index+1:seg.node.nrSegments]) Functions{}.ClearValue(&seg.node.values[seg.node.nrSegments-1]) seg.node.nrSegments-- + if trackGaps != 0 { + seg.node.updateMaxGapLeaf() + } return seg.node.rebalanceAfterRemove(GapIterator{seg.node, seg.index}) } @@ -455,6 +507,7 @@ func (s *Set) MergeUnchecked(first, second Iterator) Iterator { // overlaps second. first.SetEndUnchecked(second.End()) first.SetValue(mval) + // Remove will handle the maxGap update if necessary. return s.Remove(second).PrevSegment() } } @@ -631,6 +684,12 @@ type node struct { // than "isLeaf" because false must be the correct value for an empty root. hasChildren bool + // The longest gap within this node. If the node is a leaf, it's simply the + // maximum gap among all the (nrSegments+1) gaps formed by its nrSegments keys + // including the 0th and nrSegments-th gap possibly shared with its upper-level + // nodes; if it's a non-leaf node, it's the max of all children's maxGap. + maxGap dynamicGap + // Nodes store keys and values in separate arrays to maximize locality in // the common case (scanning keys for lookup). keys [maxDegree - 1]Range @@ -676,12 +735,12 @@ func (n *node) nextSibling() *node { // required for insertion, and returns an updated iterator to the position // represented by gap. func (n *node) rebalanceBeforeInsert(gap GapIterator) GapIterator { - if n.parent != nil { - gap = n.parent.rebalanceBeforeInsert(gap) - } if n.nrSegments < maxDegree-1 { return gap } + if n.parent != nil { + gap = n.parent.rebalanceBeforeInsert(gap) + } if n.parent == nil { // n is root. Move all segments before and after n's median segment // into new child nodes adjacent to the median segment, which is now @@ -719,6 +778,13 @@ func (n *node) rebalanceBeforeInsert(gap GapIterator) GapIterator { n.hasChildren = true n.children[0] = left n.children[1] = right + // In this case, n's maxGap won't violated as it's still the root, + // but the left and right children should be updated locally as they + // are newly split from n. + if trackGaps != 0 { + left.updateMaxGapLocal() + right.updateMaxGapLocal() + } if gap.node != n { return gap } @@ -758,6 +824,12 @@ func (n *node) rebalanceBeforeInsert(gap GapIterator) GapIterator { } } n.nrSegments = minDegree - 1 + // MaxGap of n's parent is not violated because the segments within is not changed. + // n and its sibling's maxGap need to be updated locally as they are two new nodes split from old n. + if trackGaps != 0 { + n.updateMaxGapLocal() + sibling.updateMaxGapLocal() + } // gap.node can't be n.parent because gaps are always in leaf nodes. if gap.node != n { return gap @@ -821,6 +893,12 @@ func (n *node) rebalanceAfterRemove(gap GapIterator) GapIterator { } n.nrSegments++ sibling.nrSegments-- + // n's parent's maxGap does not need to be updated as its content is unmodified. + // n and its sibling must be updated with (new) maxGap because of the shift of keys. + if trackGaps != 0 { + n.updateMaxGapLocal() + sibling.updateMaxGapLocal() + } if gap.node == sibling && gap.index == sibling.nrSegments { return GapIterator{n, 0} } @@ -849,6 +927,12 @@ func (n *node) rebalanceAfterRemove(gap GapIterator) GapIterator { } n.nrSegments++ sibling.nrSegments-- + // n's parent's maxGap does not need to be updated as its content is unmodified. + // n and its sibling must be updated with (new) maxGap because of the shift of keys. + if trackGaps != 0 { + n.updateMaxGapLocal() + sibling.updateMaxGapLocal() + } if gap.node == sibling { if gap.index == 0 { return GapIterator{n, n.nrSegments} @@ -886,6 +970,7 @@ func (n *node) rebalanceAfterRemove(gap GapIterator) GapIterator { p.children[0] = nil p.children[1] = nil } + // No need to update maxGap of p as its content is not changed. if gap.node == left { return GapIterator{p, gap.index} } @@ -932,11 +1017,152 @@ func (n *node) rebalanceAfterRemove(gap GapIterator) GapIterator { } p.children[p.nrSegments] = nil p.nrSegments-- + // Update maxGap of left locally, no need to change p and right because + // p's contents is not changed and right is already invalid. + if trackGaps != 0 { + left.updateMaxGapLocal() + } // This process robs p of one segment, so recurse into rebalancing p. n = p } } +// updateMaxGapLeaf updates maxGap bottom-up from the calling leaf until no +// necessary update. +// +// Preconditions: n must be a leaf node, trackGaps must be 1. +func (n *node) updateMaxGapLeaf() { + if n.hasChildren { + panic(fmt.Sprintf("updateMaxGapLeaf should always be called on leaf node: %v", n)) + } + max := n.calculateMaxGapLeaf() + if max == n.maxGap.Get() { + // If new max equals the old maxGap, no update is needed. + return + } + oldMax := n.maxGap.Get() + n.maxGap.Set(max) + if max > oldMax { + // Grow ancestor maxGaps. + for p := n.parent; p != nil; p = p.parent { + if p.maxGap.Get() >= max { + // p and its ancestors already contain an equal or larger gap. + break + } + // Only if new maxGap is larger than parent's + // old maxGap, propagate this update to parent. + p.maxGap.Set(max) + } + return + } + // Shrink ancestor maxGaps. + for p := n.parent; p != nil; p = p.parent { + if p.maxGap.Get() > oldMax { + // p and its ancestors still contain a larger gap. + break + } + // If new max is smaller than the old maxGap, and this gap used + // to be the maxGap of its parent, iterate parent's children + // and calculate parent's new maxGap.(It's probable that parent + // has two children with the old maxGap, but we need to check it anyway.) + parentNewMax := p.calculateMaxGapInternal() + if p.maxGap.Get() == parentNewMax { + // p and its ancestors still contain a gap of at least equal size. + break + } + // If p's new maxGap differs from the old one, propagate this update. + p.maxGap.Set(parentNewMax) + } +} + +// updateMaxGapLocal updates maxGap of the calling node solely with no +// propagation to ancestor nodes. +// +// Precondition: trackGaps must be 1. +func (n *node) updateMaxGapLocal() { + if !n.hasChildren { + // Leaf node iterates its gaps. + n.maxGap.Set(n.calculateMaxGapLeaf()) + } else { + // Non-leaf node iterates its children. + n.maxGap.Set(n.calculateMaxGapInternal()) + } +} + +// calculateMaxGapLeaf iterates the gaps within a leaf node and calculate the +// max. +// +// Preconditions: n must be a leaf node. +func (n *node) calculateMaxGapLeaf() Key { + max := GapIterator{n, 0}.Range().Length() + for i := 1; i <= n.nrSegments; i++ { + if current := (GapIterator{n, i}).Range().Length(); current > max { + max = current + } + } + return max +} + +// calculateMaxGapInternal iterates children's maxGap within an internal node n +// and calculate the max. +// +// Preconditions: n must be a non-leaf node. +func (n *node) calculateMaxGapInternal() Key { + max := n.children[0].maxGap.Get() + for i := 1; i <= n.nrSegments; i++ { + if current := n.children[i].maxGap.Get(); current > max { + max = current + } + } + return max +} + +// searchFirstLargeEnoughGap returns the first gap having at least minSize length +// in the subtree rooted by n. If not found, return a terminal gap iterator. +func (n *node) searchFirstLargeEnoughGap(minSize Key) GapIterator { + if n.maxGap.Get() < minSize { + return GapIterator{} + } + if n.hasChildren { + for i := 0; i <= n.nrSegments; i++ { + if largeEnoughGap := n.children[i].searchFirstLargeEnoughGap(minSize); largeEnoughGap.Ok() { + return largeEnoughGap + } + } + } else { + for i := 0; i <= n.nrSegments; i++ { + currentGap := GapIterator{n, i} + if currentGap.Range().Length() >= minSize { + return currentGap + } + } + } + panic(fmt.Sprintf("invalid maxGap in %v", n)) +} + +// searchLastLargeEnoughGap returns the last gap having at least minSize length +// in the subtree rooted by n. If not found, return a terminal gap iterator. +func (n *node) searchLastLargeEnoughGap(minSize Key) GapIterator { + if n.maxGap.Get() < minSize { + return GapIterator{} + } + if n.hasChildren { + for i := n.nrSegments; i >= 0; i-- { + if largeEnoughGap := n.children[i].searchLastLargeEnoughGap(minSize); largeEnoughGap.Ok() { + return largeEnoughGap + } + } + } else { + for i := n.nrSegments; i >= 0; i-- { + currentGap := GapIterator{n, i} + if currentGap.Range().Length() >= minSize { + return currentGap + } + } + } + panic(fmt.Sprintf("invalid maxGap in %v", n)) +} + // A Iterator is conceptually one of: // // - A pointer to a segment in a set; or @@ -1243,6 +1469,122 @@ func (gap GapIterator) NextGap() GapIterator { return seg.NextGap() } +// NextLargeEnoughGap returns the iterated gap's first next gap with larger +// length than minSize. If not found, return a terminal gap iterator (does NOT +// include this gap itself). +// +// Precondition: trackGaps must be 1. +func (gap GapIterator) NextLargeEnoughGap(minSize Key) GapIterator { + if trackGaps != 1 { + panic("set is not tracking gaps") + } + if gap.node != nil && gap.node.hasChildren && gap.index == gap.node.nrSegments { + // If gap is the trailing gap of an non-leaf node, + // translate it to the equivalent gap on leaf level. + gap.node = gap.NextSegment().node + gap.index = 0 + return gap.nextLargeEnoughGapHelper(minSize) + } + return gap.nextLargeEnoughGapHelper(minSize) +} + +// nextLargeEnoughGapHelper is the helper function used by NextLargeEnoughGap +// to do the real recursions. +// +// Preconditions: gap is NOT the trailing gap of a non-leaf node. +func (gap GapIterator) nextLargeEnoughGapHelper(minSize Key) GapIterator { + // Crawl up the tree if no large enough gap in current node or the + // current gap is the trailing one on leaf level. + for gap.node != nil && + (gap.node.maxGap.Get() < minSize || (!gap.node.hasChildren && gap.index == gap.node.nrSegments)) { + gap.node, gap.index = gap.node.parent, gap.node.parentIndex + } + // If no large enough gap throughout the whole set, return a terminal + // gap iterator. + if gap.node == nil { + return GapIterator{} + } + // Iterate subsequent gaps. + gap.index++ + for gap.index <= gap.node.nrSegments { + if gap.node.hasChildren { + if largeEnoughGap := gap.node.children[gap.index].searchFirstLargeEnoughGap(minSize); largeEnoughGap.Ok() { + return largeEnoughGap + } + } else { + if gap.Range().Length() >= minSize { + return gap + } + } + gap.index++ + } + gap.node, gap.index = gap.node.parent, gap.node.parentIndex + if gap.node != nil && gap.index == gap.node.nrSegments { + // If gap is the trailing gap of a non-leaf node, crawl up to + // parent again and do recursion. + gap.node, gap.index = gap.node.parent, gap.node.parentIndex + } + return gap.nextLargeEnoughGapHelper(minSize) +} + +// PrevLargeEnoughGap returns the iterated gap's first prev gap with larger or +// equal length than minSize. If not found, return a terminal gap iterator +// (does NOT include this gap itself). +// +// Precondition: trackGaps must be 1. +func (gap GapIterator) PrevLargeEnoughGap(minSize Key) GapIterator { + if trackGaps != 1 { + panic("set is not tracking gaps") + } + if gap.node != nil && gap.node.hasChildren && gap.index == 0 { + // If gap is the first gap of an non-leaf node, + // translate it to the equivalent gap on leaf level. + gap.node = gap.PrevSegment().node + gap.index = gap.node.nrSegments + return gap.prevLargeEnoughGapHelper(minSize) + } + return gap.prevLargeEnoughGapHelper(minSize) +} + +// prevLargeEnoughGapHelper is the helper function used by PrevLargeEnoughGap +// to do the real recursions. +// +// Preconditions: gap is NOT the first gap of a non-leaf node. +func (gap GapIterator) prevLargeEnoughGapHelper(minSize Key) GapIterator { + // Crawl up the tree if no large enough gap in current node or the + // current gap is the first one on leaf level. + for gap.node != nil && + (gap.node.maxGap.Get() < minSize || (!gap.node.hasChildren && gap.index == 0)) { + gap.node, gap.index = gap.node.parent, gap.node.parentIndex + } + // If no large enough gap throughout the whole set, return a terminal + // gap iterator. + if gap.node == nil { + return GapIterator{} + } + // Iterate previous gaps. + gap.index-- + for gap.index >= 0 { + if gap.node.hasChildren { + if largeEnoughGap := gap.node.children[gap.index].searchLastLargeEnoughGap(minSize); largeEnoughGap.Ok() { + return largeEnoughGap + } + } else { + if gap.Range().Length() >= minSize { + return gap + } + } + gap.index-- + } + gap.node, gap.index = gap.node.parent, gap.node.parentIndex + if gap.node != nil && gap.index == 0 { + // If gap is the first gap of a non-leaf node, crawl up to + // parent again and do recursion. + gap.node, gap.index = gap.node.parent, gap.node.parentIndex + } + return gap.prevLargeEnoughGapHelper(minSize) +} + // segmentBeforePosition returns the predecessor segment of the position given // by n.children[i], which may or may not contain a child. If no such segment // exists, segmentBeforePosition returns a terminal iterator. @@ -1271,7 +1613,7 @@ func segmentAfterPosition(n *node, i int) Iterator { func zeroValueSlice(slice []Value) { // TODO(jamieliu): check if Go is actually smart enough to optimize a - // ClearValue that assigns nil to a memset here + // ClearValue that assigns nil to a memset here. for i := range slice { Functions{}.ClearValue(&slice[i]) } @@ -1310,7 +1652,15 @@ func (n *node) writeDebugString(buf *bytes.Buffer, prefix string) { child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, i)) } buf.WriteString(prefix) - buf.WriteString(fmt.Sprintf("- % 3d: %v => %v\n", i, n.keys[i], n.values[i])) + if n.hasChildren { + if trackGaps != 0 { + buf.WriteString(fmt.Sprintf("- % 3d: %v => %v, maxGap: %d\n", i, n.keys[i], n.values[i], n.maxGap.Get())) + } else { + buf.WriteString(fmt.Sprintf("- % 3d: %v => %v\n", i, n.keys[i], n.values[i])) + } + } else { + buf.WriteString(fmt.Sprintf("- % 3d: %v => %v\n", i, n.keys[i], n.values[i])) + } } if child := n.children[n.nrSegments]; child != nil { child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, n.nrSegments)) @@ -1362,3 +1712,43 @@ func (s *Set) ImportSortedSlices(sds *SegmentDataSlices) error { } return nil } + +// segmentTestCheck returns an error if s is incorrectly sorted, does not +// contain exactly expectedSegments segments, or contains a segment which +// fails the passed check. +// +// This should be used only for testing, and has been added to this package for +// templating convenience. +func (s *Set) segmentTestCheck(expectedSegments int, segFunc func(int, Range, Value) error) error { + havePrev := false + prev := Key(0) + nrSegments := 0 + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + next := seg.Start() + if havePrev && prev >= next { + return fmt.Errorf("incorrect order: key %d (segment %d) >= key %d (segment %d)", prev, nrSegments-1, next, nrSegments) + } + if segFunc != nil { + if err := segFunc(nrSegments, seg.Range(), seg.Value()); err != nil { + return err + } + } + prev = next + havePrev = true + nrSegments++ + } + if nrSegments != expectedSegments { + return fmt.Errorf("incorrect number of segments: got %d, wanted %d", nrSegments, expectedSegments) + } + return nil +} + +// countSegments counts the number of segments in the set. +// +// Similar to Check, this should only be used for testing. +func (s *Set) countSegments() (segments int) { + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + segments++ + } + return segments +} diff --git a/pkg/segment/test/BUILD b/pkg/segment/test/BUILD index f2d8462d8..131bf09b9 100644 --- a/pkg/segment/test/BUILD +++ b/pkg/segment/test/BUILD @@ -29,10 +29,28 @@ go_template_instance( }, ) +go_template_instance( + name = "gap_set", + out = "gap_set.go", + consts = { + "trackGaps": "1", + }, + package = "segment", + prefix = "gap", + template = "//pkg/segment:generic_set", + types = { + "Key": "int", + "Range": "Range", + "Value": "int", + "Functions": "gapSetFunctions", + }, +) + go_library( name = "segment", testonly = 1, srcs = [ + "gap_set.go", "int_range.go", "int_set.go", "set_functions.go", diff --git a/pkg/segment/test/segment_test.go b/pkg/segment/test/segment_test.go index 97b16c158..85fa19096 100644 --- a/pkg/segment/test/segment_test.go +++ b/pkg/segment/test/segment_test.go @@ -17,6 +17,7 @@ package segment import ( "fmt" "math/rand" + "reflect" "testing" ) @@ -32,61 +33,65 @@ const ( // valueOffset is the difference between the value and start of test // segments. valueOffset = 100000 + + // intervalLength is the interval used by random gap tests. + intervalLength = 10 ) func shuffle(xs []int) { - for i := range xs { - j := rand.Intn(i + 1) - xs[i], xs[j] = xs[j], xs[i] - } + rand.Shuffle(len(xs), func(i, j int) { xs[i], xs[j] = xs[j], xs[i] }) } -func randPermutation(size int) []int { +func randIntervalPermutation(size int) []int { p := make([]int, size) for i := range p { - p[i] = i + p[i] = intervalLength * i } shuffle(p) return p } -// checkSet returns an error if s is incorrectly sorted, does not contain -// exactly expectedSegments segments, or contains a segment for which val != -// key + valueOffset. -func checkSet(s *Set, expectedSegments int) error { - havePrev := false - prev := 0 - nrSegments := 0 - for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { - next := seg.Start() - if havePrev && prev >= next { - return fmt.Errorf("incorrect order: key %d (segment %d) >= key %d (segment %d)", prev, nrSegments-1, next, nrSegments) - } - if got, want := seg.Value(), seg.Start()+valueOffset; got != want { - return fmt.Errorf("segment %d has key %d, value %d (expected %d)", nrSegments, seg.Start(), got, want) - } - prev = next - havePrev = true - nrSegments++ - } - if nrSegments != expectedSegments { - return fmt.Errorf("incorrect number of segments: got %d, wanted %d", nrSegments, expectedSegments) +// validate can be passed to Check. +func validate(nr int, r Range, v int) error { + if got, want := v, r.Start+valueOffset; got != want { + return fmt.Errorf("segment %d has key %d, value %d (expected %d)", nr, r.Start, got, want) } return nil } -// countSegmentsIn returns the number of segments in s. -func countSegmentsIn(s *Set) int { - var count int - for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { - count++ +// checkSetMaxGap returns an error if maxGap inside all nodes of s is not well +// maintained. +func checkSetMaxGap(s *gapSet) error { + n := s.root + return checkNodeMaxGap(&n) +} + +// checkNodeMaxGap returns an error if maxGap inside the subtree rooted by n is +// not well maintained. +func checkNodeMaxGap(n *gapnode) error { + var max int + if !n.hasChildren { + max = n.calculateMaxGapLeaf() + } else { + for i := 0; i <= n.nrSegments; i++ { + child := n.children[i] + if err := checkNodeMaxGap(child); err != nil { + return err + } + if temp := child.maxGap.Get(); i == 0 || temp > max { + max = temp + } + } + } + if max != n.maxGap.Get() { + return fmt.Errorf("maxGap wrong in node\n%vexpected: %d got: %d", n, max, n.maxGap) } - return count + return nil } func TestAddRandom(t *testing.T) { var s Set - order := randPermutation(testSize) + order := rand.Perm(testSize) var nrInsertions int for i, j := range order { if !s.AddWithoutMerging(Range{j, j + 1}, j+valueOffset) { @@ -94,12 +99,12 @@ func TestAddRandom(t *testing.T) { break } nrInsertions++ - if err := checkSet(&s, nrInsertions); err != nil { + if err := s.segmentTestCheck(nrInsertions, validate); err != nil { t.Errorf("Iteration %d: %v", i, err) break } } - if got, want := countSegmentsIn(&s), nrInsertions; got != want { + if got, want := s.countSegments(), nrInsertions; got != want { t.Errorf("Wrong final number of segments: got %d, wanted %d", got, want) } if t.Failed() { @@ -115,7 +120,156 @@ func TestRemoveRandom(t *testing.T) { t.Fatalf("Failed to insert segment %d", i) } } - order := randPermutation(testSize) + order := rand.Perm(testSize) + var nrRemovals int + for i, j := range order { + seg := s.FindSegment(j) + if !seg.Ok() { + t.Errorf("Iteration %d: failed to find segment with key %d", i, j) + break + } + s.Remove(seg) + nrRemovals++ + if err := s.segmentTestCheck(testSize-nrRemovals, validate); err != nil { + t.Errorf("Iteration %d: %v", i, err) + break + } + } + if got, want := s.countSegments(), testSize-nrRemovals; got != want { + t.Errorf("Wrong final number of segments: got %d, wanted %d", got, want) + } + if t.Failed() { + t.Logf("Removal order: %v", order[:nrRemovals]) + t.Logf("Set contents:\n%v", &s) + t.FailNow() + } +} + +func TestMaxGapAddRandom(t *testing.T) { + var s gapSet + order := rand.Perm(testSize) + var nrInsertions int + for i, j := range order { + if !s.AddWithoutMerging(Range{j, j + 1}, j+valueOffset) { + t.Errorf("Iteration %d: failed to insert segment with key %d", i, j) + break + } + nrInsertions++ + if err := s.segmentTestCheck(nrInsertions, validate); err != nil { + t.Errorf("Iteration %d: %v", i, err) + break + } + if err := checkSetMaxGap(&s); err != nil { + t.Errorf("When inserting %d: %v", j, err) + break + } + } + if got, want := s.countSegments(), nrInsertions; got != want { + t.Errorf("Wrong final number of segments: got %d, wanted %d", got, want) + } + if t.Failed() { + t.Logf("Insertion order: %v", order[:nrInsertions]) + t.Logf("Set contents:\n%v", &s) + } +} + +func TestMaxGapAddRandomWithRandomInterval(t *testing.T) { + var s gapSet + order := randIntervalPermutation(testSize) + var nrInsertions int + for i, j := range order { + if !s.AddWithoutMerging(Range{j, j + rand.Intn(intervalLength-1) + 1}, j+valueOffset) { + t.Errorf("Iteration %d: failed to insert segment with key %d", i, j) + break + } + nrInsertions++ + if err := s.segmentTestCheck(nrInsertions, validate); err != nil { + t.Errorf("Iteration %d: %v", i, err) + break + } + if err := checkSetMaxGap(&s); err != nil { + t.Errorf("When inserting %d: %v", j, err) + break + } + } + if got, want := s.countSegments(), nrInsertions; got != want { + t.Errorf("Wrong final number of segments: got %d, wanted %d", got, want) + } + if t.Failed() { + t.Logf("Insertion order: %v", order[:nrInsertions]) + t.Logf("Set contents:\n%v", &s) + } +} + +func TestMaxGapAddRandomWithMerge(t *testing.T) { + var s gapSet + order := randIntervalPermutation(testSize) + nrInsertions := 1 + for i, j := range order { + if !s.Add(Range{j, j + intervalLength}, j+valueOffset) { + t.Errorf("Iteration %d: failed to insert segment with key %d", i, j) + break + } + if err := checkSetMaxGap(&s); err != nil { + t.Errorf("When inserting %d: %v", j, err) + break + } + } + if got, want := s.countSegments(), nrInsertions; got != want { + t.Errorf("Wrong final number of segments: got %d, wanted %d", got, want) + } + if t.Failed() { + t.Logf("Insertion order: %v", order) + t.Logf("Set contents:\n%v", &s) + } +} + +func TestMaxGapRemoveRandom(t *testing.T) { + var s gapSet + for i := 0; i < testSize; i++ { + if !s.AddWithoutMerging(Range{i, i + 1}, i+valueOffset) { + t.Fatalf("Failed to insert segment %d", i) + } + } + order := rand.Perm(testSize) + var nrRemovals int + for i, j := range order { + seg := s.FindSegment(j) + if !seg.Ok() { + t.Errorf("Iteration %d: failed to find segment with key %d", i, j) + break + } + temprange := seg.Range() + s.Remove(seg) + nrRemovals++ + if err := s.segmentTestCheck(testSize-nrRemovals, validate); err != nil { + t.Errorf("Iteration %d: %v", i, err) + break + } + if err := checkSetMaxGap(&s); err != nil { + t.Errorf("When removing %v: %v", temprange, err) + break + } + } + if got, want := s.countSegments(), testSize-nrRemovals; got != want { + t.Errorf("Wrong final number of segments: got %d, wanted %d", got, want) + } + if t.Failed() { + t.Logf("Removal order: %v", order[:nrRemovals]) + t.Logf("Set contents:\n%v", &s) + t.FailNow() + } +} + +func TestMaxGapRemoveHalfRandom(t *testing.T) { + var s gapSet + for i := 0; i < testSize; i++ { + if !s.AddWithoutMerging(Range{intervalLength * i, intervalLength*i + rand.Intn(intervalLength-1) + 1}, intervalLength*i+valueOffset) { + t.Fatalf("Failed to insert segment %d", i) + } + } + order := randIntervalPermutation(testSize) + order = order[:testSize/2] var nrRemovals int for i, j := range order { seg := s.FindSegment(j) @@ -123,14 +277,19 @@ func TestRemoveRandom(t *testing.T) { t.Errorf("Iteration %d: failed to find segment with key %d", i, j) break } + temprange := seg.Range() s.Remove(seg) nrRemovals++ - if err := checkSet(&s, testSize-nrRemovals); err != nil { + if err := s.segmentTestCheck(testSize-nrRemovals, validate); err != nil { t.Errorf("Iteration %d: %v", i, err) break } + if err := checkSetMaxGap(&s); err != nil { + t.Errorf("When removing %v: %v", temprange, err) + break + } } - if got, want := countSegmentsIn(&s), testSize-nrRemovals; got != want { + if got, want := s.countSegments(), testSize-nrRemovals; got != want { t.Errorf("Wrong final number of segments: got %d, wanted %d", got, want) } if t.Failed() { @@ -140,6 +299,148 @@ func TestRemoveRandom(t *testing.T) { } } +func TestMaxGapAddRandomRemoveRandomHalfWithMerge(t *testing.T) { + var s gapSet + order := randIntervalPermutation(testSize * 2) + order = order[:testSize] + for i, j := range order { + if !s.Add(Range{j, j + intervalLength}, j+valueOffset) { + t.Errorf("Iteration %d: failed to insert segment with key %d", i, j) + break + } + if err := checkSetMaxGap(&s); err != nil { + t.Errorf("When inserting %d: %v", j, err) + break + } + } + shuffle(order) + var nrRemovals int + for _, j := range order { + seg := s.FindSegment(j) + if !seg.Ok() { + continue + } + temprange := seg.Range() + s.Remove(seg) + nrRemovals++ + if err := checkSetMaxGap(&s); err != nil { + t.Errorf("When removing %v: %v", temprange, err) + break + } + } + if t.Failed() { + t.Logf("Removal order: %v", order[:nrRemovals]) + t.Logf("Set contents:\n%v", &s) + t.FailNow() + } +} + +func TestNextLargeEnoughGap(t *testing.T) { + var s gapSet + order := randIntervalPermutation(testSize * 2) + order = order[:testSize] + for i, j := range order { + if !s.Add(Range{j, j + rand.Intn(intervalLength-1) + 1}, j+valueOffset) { + t.Errorf("Iteration %d: failed to insert segment with key %d", i, j) + break + } + if err := checkSetMaxGap(&s); err != nil { + t.Errorf("When inserting %d: %v", j, err) + break + } + } + shuffle(order) + order = order[:testSize/2] + for _, j := range order { + seg := s.FindSegment(j) + if !seg.Ok() { + continue + } + temprange := seg.Range() + s.Remove(seg) + if err := checkSetMaxGap(&s); err != nil { + t.Errorf("When removing %v: %v", temprange, err) + break + } + } + minSize := 7 + var gapArr1 []int + for gap := s.LowerBoundGap(0).NextLargeEnoughGap(minSize); gap.Ok(); gap = gap.NextLargeEnoughGap(minSize) { + if gap.Range().Length() < minSize { + t.Errorf("NextLargeEnoughGap wrong, gap %v has length %d, wanted %d", gap.Range(), gap.Range().Length(), minSize) + } else { + gapArr1 = append(gapArr1, gap.Range().Start) + } + } + var gapArr2 []int + for gap := s.LowerBoundGap(0).NextGap(); gap.Ok(); gap = gap.NextGap() { + if gap.Range().Length() >= minSize { + gapArr2 = append(gapArr2, gap.Range().Start) + } + } + + if !reflect.DeepEqual(gapArr2, gapArr1) { + t.Errorf("Search result not correct, got: %v, wanted: %v", gapArr1, gapArr2) + } + if t.Failed() { + t.Logf("Set contents:\n%v", &s) + t.FailNow() + } +} + +func TestPrevLargeEnoughGap(t *testing.T) { + var s gapSet + order := randIntervalPermutation(testSize * 2) + order = order[:testSize] + for i, j := range order { + if !s.Add(Range{j, j + rand.Intn(intervalLength-1) + 1}, j+valueOffset) { + t.Errorf("Iteration %d: failed to insert segment with key %d", i, j) + break + } + if err := checkSetMaxGap(&s); err != nil { + t.Errorf("When inserting %d: %v", j, err) + break + } + } + end := s.LastSegment().End() + shuffle(order) + order = order[:testSize/2] + for _, j := range order { + seg := s.FindSegment(j) + if !seg.Ok() { + continue + } + temprange := seg.Range() + s.Remove(seg) + if err := checkSetMaxGap(&s); err != nil { + t.Errorf("When removing %v: %v", temprange, err) + break + } + } + minSize := 7 + var gapArr1 []int + for gap := s.UpperBoundGap(end + intervalLength).PrevLargeEnoughGap(minSize); gap.Ok(); gap = gap.PrevLargeEnoughGap(minSize) { + if gap.Range().Length() < minSize { + t.Errorf("PrevLargeEnoughGap wrong, gap length %d, wanted %d", gap.Range().Length(), minSize) + } else { + gapArr1 = append(gapArr1, gap.Range().Start) + } + } + var gapArr2 []int + for gap := s.UpperBoundGap(end + intervalLength).PrevGap(); gap.Ok(); gap = gap.PrevGap() { + if gap.Range().Length() >= minSize { + gapArr2 = append(gapArr2, gap.Range().Start) + } + } + if !reflect.DeepEqual(gapArr2, gapArr1) { + t.Errorf("Search result not correct, got: %v, wanted: %v", gapArr1, gapArr2) + } + if t.Failed() { + t.Logf("Set contents:\n%v", &s) + t.FailNow() + } +} + func TestAddSequentialAdjacent(t *testing.T) { var s Set var nrInsertions int @@ -148,12 +449,12 @@ func TestAddSequentialAdjacent(t *testing.T) { t.Fatalf("Failed to insert segment %d", i) } nrInsertions++ - if err := checkSet(&s, nrInsertions); err != nil { + if err := s.segmentTestCheck(nrInsertions, validate); err != nil { t.Errorf("Iteration %d: %v", i, err) break } } - if got, want := countSegmentsIn(&s), nrInsertions; got != want { + if got, want := s.countSegments(), nrInsertions; got != want { t.Errorf("Wrong final number of segments: got %d, wanted %d", got, want) } if t.Failed() { @@ -202,12 +503,12 @@ func TestAddSequentialNonAdjacent(t *testing.T) { t.Fatalf("Failed to insert segment %d", i) } nrInsertions++ - if err := checkSet(&s, nrInsertions); err != nil { + if err := s.segmentTestCheck(nrInsertions, validate); err != nil { t.Errorf("Iteration %d: %v", i, err) break } } - if got, want := countSegmentsIn(&s), nrInsertions; got != want { + if got, want := s.countSegments(), nrInsertions; got != want { t.Errorf("Wrong final number of segments: got %d, wanted %d", got, want) } if t.Failed() { @@ -293,7 +594,7 @@ Tests: var i int for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { if i > len(test.final) { - t.Errorf("%s: Incorrect number of segments: got %d, wanted %d; set contents:\n%v", test.name, countSegmentsIn(&s), len(test.final), &s) + t.Errorf("%s: Incorrect number of segments: got %d, wanted %d; set contents:\n%v", test.name, s.countSegments(), len(test.final), &s) continue Tests } if got, want := seg.Range(), test.final[i]; got != want { @@ -351,7 +652,7 @@ Tests: var i int for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { if i > len(test.final) { - t.Errorf("%s: Incorrect number of segments: got %d, wanted %d; set contents:\n%v", test.name, countSegmentsIn(&s), len(test.final), &s) + t.Errorf("%s: Incorrect number of segments: got %d, wanted %d; set contents:\n%v", test.name, s.countSegments(), len(test.final), &s) continue Tests } if got, want := seg.Range(), test.final[i]; got != want { @@ -378,7 +679,7 @@ func benchmarkAddSequential(b *testing.B, size int) { } func benchmarkAddRandom(b *testing.B, size int) { - order := randPermutation(size) + order := rand.Perm(size) b.ResetTimer() for n := 0; n < b.N; n++ { @@ -416,7 +717,7 @@ func benchmarkFindRandom(b *testing.B, size int) { b.Fatalf("Failed to insert segment %d", i) } } - order := randPermutation(size) + order := rand.Perm(size) b.ResetTimer() for n := 0; n < b.N; n++ { @@ -470,7 +771,7 @@ func benchmarkAddFindRemoveSequential(b *testing.B, size int) { } func benchmarkAddFindRemoveRandom(b *testing.B, size int) { - order := randPermutation(size) + order := rand.Perm(size) b.ResetTimer() for n := 0; n < b.N; n++ { diff --git a/pkg/segment/test/set_functions.go b/pkg/segment/test/set_functions.go index bcddb39bb..7cd895cc7 100644 --- a/pkg/segment/test/set_functions.go +++ b/pkg/segment/test/set_functions.go @@ -14,21 +14,16 @@ package segment -// Basic numeric constants that we define because the math package doesn't. -// TODO(nlacasse): These should be Math.MaxInt64/MinInt64? -const ( - maxInt = int(^uint(0) >> 1) - minInt = -maxInt - 1 -) - type setFunctions struct{} -func (setFunctions) MinKey() int { - return minInt +// MinKey returns the minimum key for the set. +func (s setFunctions) MinKey() int { + return -s.MaxKey() - 1 } +// MaxKey returns the maximum key for the set. func (setFunctions) MaxKey() int { - return maxInt + return int(^uint(0) >> 1) } func (setFunctions) ClearValue(*int) {} @@ -40,3 +35,20 @@ func (setFunctions) Merge(_ Range, val1 int, _ Range, _ int) (int, bool) { func (setFunctions) Split(_ Range, val int, _ int) (int, int) { return val, val } + +type gapSetFunctions struct { + setFunctions +} + +// MinKey is adjusted to make sure no add overflow would happen in test cases. +// e.g. A gap with range {MinInt32, 2} would cause overflow in Range().Length(). +// +// Normally Keys should be unsigned to avoid these issues. +func (s gapSetFunctions) MinKey() int { + return s.setFunctions.MinKey() / 2 +} + +// MaxKey returns the maximum key for the set. +func (s gapSetFunctions) MaxKey() int { + return s.setFunctions.MaxKey() / 2 +} diff --git a/pkg/sentry/mm/BUILD b/pkg/sentry/mm/BUILD index 73591dab7..a036ce53c 100644 --- a/pkg/sentry/mm/BUILD +++ b/pkg/sentry/mm/BUILD @@ -25,6 +25,7 @@ go_template_instance( out = "vma_set.go", consts = { "minDegree": "8", + "trackGaps": "1", }, imports = { "usermem": "gvisor.dev/gvisor/pkg/usermem", diff --git a/pkg/sentry/mm/vma.go b/pkg/sentry/mm/vma.go index 9a14e69e6..16d8207e9 100644 --- a/pkg/sentry/mm/vma.go +++ b/pkg/sentry/mm/vma.go @@ -195,7 +195,7 @@ func (mm *MemoryManager) applicationAddrRange() usermem.AddrRange { // Preconditions: mm.mappingMu must be locked. func (mm *MemoryManager) findLowestAvailableLocked(length, alignment uint64, bounds usermem.AddrRange) (usermem.Addr, error) { - for gap := mm.vmas.LowerBoundGap(bounds.Start); gap.Ok() && gap.Start() < bounds.End; gap = gap.NextGap() { + for gap := mm.vmas.LowerBoundGap(bounds.Start); gap.Ok() && gap.Start() < bounds.End; gap = gap.NextLargeEnoughGap(usermem.Addr(length)) { if gr := gap.availableRange().Intersect(bounds); uint64(gr.Length()) >= length { // Can we shift up to match the alignment? if offset := uint64(gr.Start) % alignment; offset != 0 { @@ -214,7 +214,7 @@ func (mm *MemoryManager) findLowestAvailableLocked(length, alignment uint64, bou // Preconditions: mm.mappingMu must be locked. func (mm *MemoryManager) findHighestAvailableLocked(length, alignment uint64, bounds usermem.AddrRange) (usermem.Addr, error) { - for gap := mm.vmas.UpperBoundGap(bounds.End); gap.Ok() && gap.End() > bounds.Start; gap = gap.PrevGap() { + for gap := mm.vmas.UpperBoundGap(bounds.End); gap.Ok() && gap.End() > bounds.Start; gap = gap.PrevLargeEnoughGap(usermem.Addr(length)) { if gr := gap.availableRange().Intersect(bounds); uint64(gr.Length()) >= length { // Can we shift down to match the alignment? start := gr.End - usermem.Addr(length) diff --git a/tools/go_generics/generics.go b/tools/go_generics/generics.go index e9cc2c753..0860ca9db 100644 --- a/tools/go_generics/generics.go +++ b/tools/go_generics/generics.go @@ -223,7 +223,9 @@ func main() { } else { switch kind { case globals.KindType, globals.KindVar, globals.KindConst, globals.KindFunction: - ident.Name = *prefix + ident.Name + *suffix + if ident.Name != "_" { + ident.Name = *prefix + ident.Name + *suffix + } case globals.KindTag: // Modify the state tag appropriately. if m := stateTagRegexp.FindStringSubmatch(ident.Name); m != nil { -- cgit v1.2.3 From 7938d8734870404c083379d2a73f7cd37fd2eda8 Mon Sep 17 00:00:00 2001 From: Rahat Mahmood Date: Wed, 13 May 2020 19:57:27 -0400 Subject: Write initial design doc for FUSE. --- pkg/sentry/fs/g3doc/.gitignore | 1 + pkg/sentry/fs/g3doc/fuse.md | 258 +++++++++++++++++++++++++++++++++++++++++ 2 files changed, 259 insertions(+) create mode 100644 pkg/sentry/fs/g3doc/.gitignore create mode 100644 pkg/sentry/fs/g3doc/fuse.md (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/g3doc/.gitignore b/pkg/sentry/fs/g3doc/.gitignore new file mode 100644 index 000000000..2d19fc766 --- /dev/null +++ b/pkg/sentry/fs/g3doc/.gitignore @@ -0,0 +1 @@ +*.html diff --git a/pkg/sentry/fs/g3doc/fuse.md b/pkg/sentry/fs/g3doc/fuse.md new file mode 100644 index 000000000..c3988aa43 --- /dev/null +++ b/pkg/sentry/fs/g3doc/fuse.md @@ -0,0 +1,258 @@ +# Foreword + +This document describes an on-going project to support FUSE filesystems within +the sentry. This is intended to become the final documentation for this +subsystem, and is therefore written in the past tense. However FUSE support is +currently incomplete and the document will be updated as things progress. + +# FUSE: Filesystem in Userspace + +The sentry supports dispatching filesystem operations to a FUSE server, +allowing FUSE filesystem to be used with a sandbox. + +## Overview + +FUSE has two main components: + +1. A client kernel driver (canonically `fuse.ko` in Linux), which forwards + filesystem operations (usually initiated by syscalls) to the server. + +2. A server, which is a userspace daemon that implements the actual filesystem. + +The sentry implements the client component, which allows a server daemon +running within the sandbox to implement a filesystem within the sandbox. + +A FUSE filesystem is initialized with `mount(2)`, typically with the help of a +utility like `fusermount(1)`. Various mount options exist for establishing +ownership and access permissions on the filesystem, but the most important mount +option is a file descriptor used to establish communication between the client +and server. + +The FUSE device FD is obtained by opening `/dev/fuse`. During regular operation, +the client and server use the FUSE protocol described in `fuse(4)` to service +filesystem operations. See the "Protocol" section below for more +information about this protocol. The core of the sentry support for FUSE is the +client-side implementation of this protocol. + +## FUSE in the Sentry + +The sentry's FUSE client targets VFS2 and has the following components: + +- An implementation of `/dev/fuse`. + +- A VFS2 filesystem for mapping syscalls to FUSE ops. Since we're targeting + VFS2, one point of contention may be the lack of inodes in VFS2. We can + tentatively implement a kernfs-based filesystem to bridge the gap in APIs. The + kernfs base functionality can serve the role of the Linux inode cache and, the + filesystem can map VFS2 syscalls to kernfs inode operations; see the + `kernfs.Inode` interface. + +The FUSE protocol lends itself well to marshaling with `go_marshal`. The +various request and response packets can be defined in the ABI package and +converted to and from the wire format using `go_marshal`. + +### Design Goals + +- While filesystem performance is always important, the sentry's FUSE support is + primarily concerned with compatibility, with performance as a secondary + concern. + +- Avoiding deadlocks from a hung server daemon. + +- Consider the potential for denial of service from a malicious server + daemon. Protecting itself from userspace is already a design goal for the + sentry, but needs additional consideration for FUSE. Normally, an operating + system doesn't rely on userspace to make progress with filesystem + operations. Since this changes with FUSE, it opens up the possibility of + creating a chain of dependencies controlled by userspace, which could affect + an entire sandbox. For example: a FUSE op can block a syscall, which could be + holding a subsystem lock, which can then block another task goroutine. + +### Milestones + +Below are some broad goals to aim for while implementing FUSE in the sentry. +Many FUSE ops can be grouped into broad categories of functionality, and most +ops can be implemented in parallel. + +#### Minimal client that can mount a trivial FUSE filesystem. + +- Implement `/dev/fuse`. + +- Implement basic FUSE ops like `FUSE_INIT`, `FUSE_DESTROY`. + +#### Read-only mount with basic file operations + +- Implement the majority of file, directory and file descriptor FUSE ops. For + this milestone, we can skip uncommon or complex operations like mmap, mknod, + file locking, poll, and extended attributes. We can stub these out along with + any ops that modify the filesystem. The exact list of required ops are to be + determined, but the goal is to mount a real filesystem as read-only, and be + able to read contents from the filesystem in the sentry. + +#### Full read-write support + +- Implement the remaining FUSE ops and decide if we can omit rarely used + operations like ioctl. + +# Appendix + +## FUSE Protocol + +The FUSE protocol is a request-response protocol. All requests are initiated by +the client. The wire-format for the protocol is raw c structs serialized to +memory. + +All FUSE requests begin with the following request header: + +```c +struct fuse_in_header { + uint32_t len; // Length of the request, including this header. + uint32_t opcode; // Requested operation. + uint64_t unique; // A unique identifier for this request. + uint64_t nodeid; // ID of the filesystem object being operated on. + uint32_t uid; // UID of the requesting process. + uint32_t gid; // GID of the requesting process. + uint32_t pid; // PID of the requesting process. + uint32_t padding; +}; +``` + +The request is then followed by a payload specific to the `opcode`. + +All responses begin with this response header: + +```c +struct fuse_out_header { + uint32_t len; // Length of the response, including this header. + int32_t error; // Status of the request, 0 if success. + uint64_t unique; // The unique identifier from the corresponding request. +}; +``` + +The response payload also depends on the request `opcode`. If `error != 0`, the +response payload must be empty. + +### Operations + +The following is a list of all FUSE operations used in `fuse_in_header.opcode` +as of Linux v4.4, and a brief description of their purpose. These are defined in +`uapi/linux/fuse.h`. Many of these have a corresponding request and response +payload struct; `fuse(4)` has details for some of these. We also note how these +operations map to the sentry virtual filesystem. + +#### FUSE meta-operations + +These operations are specific to FUSE and don't have a corresponding action in a +generic filesystem. + +- `FUSE_INIT`: This operation initializes a new FUSE filesystem, and is the + first message sent by the client after mount. This is used for version and + feature negotiation. This is related to `mount(2)`. +- `FUSE_DESTROY`: Teardown a FUSE filesystem, related to `unmount(2)`. +- `FUSE_INTERRUPT`: Interrupts an in-flight operation, specified by the + `fuse_in_header.unique` value provided in the corresponding request + header. The client can send at most one of these per request, and will enter + an uninterruptible wait for a reply. The server is expected to reply promptly. +- `FUSE_FORGET`: A hint to the server that server should evict the indicate node + from any caches. This is wired up to `(struct super_operations).evict_inode` + in Linux, which is in turned hooked as the inode cache shrinker which is + typically triggered by system memory pressure. +- `FUSE_BATCH_FORGET`: Batch version of `FUSE_FORGET`. + +#### Filesystem Syscalls + +These FUSE ops map directly to an equivalent filesystem syscall, or family of +syscalls. The relevant syscalls have a similar name to the operation, unless +otherwise noted. + +Node creation: + +- `FUSE_MKNOD` +- `FUSE_MKDIR` +- `FUSE_CREATE`: This is equivalent to `open(2)` and `creat(2)`, which + atomically creates and opens a node. + +Node attributes and extended attributes: + +- `FUSE_GETATTR` +- `FUSE_SETATTR` +- `FUSE_SETXATTR` +- `FUSE_GETXATTR` +- `FUSE_LISTXATTR` +- `FUSE_REMOVEXATTR` + +Node link manipulation: + +- `FUSE_READLINK` +- `FUSE_LINK` +- `FUSE_SYMLINK` +- `FUSE_UNLINK` + +Directory operations: + +- `FUSE_RMDIR` +- `FUSE_RENAME` +- `FUSE_RENAME2` +- `FUSE_OPENDIR`: `open(2)` for directories. +- `FUSE_RELEASEDIR`: `close(2)` for directories. +- `FUSE_READDIR` +- `FUSE_READDIRPLUS` +- `FUSE_FSYNCDIR`: `fsync(2)` for directories. +- `FUSE_LOOKUP`: Establishes a unique identifier for a FS node. This is + reminiscent of `VirtualFilesystem.GetDentryAt` in that it resolves a path + component to a node. However the returned identifier is opaque to the + client. The server must remember this mapping, as this is how the client will + reference the node in the future. + +File operations: + +- `FUSE_OPEN`: `open(2)` for files. +- `FUSE_RELEASE`: `close(2)` for files. +- `FUSE_FSYNC` +- `FUSE_FALLOCATE` +- `FUSE_SETUPMAPPING`: Creates a memory map on a file for `mmap(2)`. +- `FUSE_REMOVEMAPPING`: Removes a memory map for `munmap(2)`. + +File locking: + +- `FUSE_GETLK` +- `FUSE_SETLK` +- `FUSE_SETLKW` +- `FUSE_COPY_FILE_RANGE` + +File descriptor operations: + +- `FUSE_IOCTL` +- `FUSE_POLL` +- `FUSE_LSEEK` + +Filesystem operations: + +- `FUSE_STATFS` + +#### Permissions + +- `FUSE_ACCESS` is used to check if a node is accessible, as part of many + syscall implementations. Maps to `vfs.FilesystemImpl.AccessAt` + in the sentry. + +#### I/O Operations + +These ops are used to read and write file pages. They're used to implement both +I/O syscalls like `read(2)`, `write(2)` and `mmap(2)`. + +- `FUSE_READ` +- `FUSE_WRITE` + +#### Miscellaneous + +- `FUSE_FLUSH`: Used by the client to indicate when a file descriptor is + closed. Distinct from `FUSE_FSYNC`, which corresponds to an `fsync(2)` syscall + from the user. Maps to `vfs.FileDescriptorImpl.Release` in the sentry. +- `FUSE_BMAP`: Old address space API for block defrag. Probably not needed. +- `FUSE_NOTIFY_REPLY`: [TODO: what does this do?] + +# References + +- `fuse(4)` manpage. +- Linux kernel FUSE documentation: https://www.kernel.org/doc/html/latest/filesystems/fuse.html -- cgit v1.2.3 From af3121a52383fb60579d769994be5d91bd788015 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Tue, 26 May 2020 21:42:07 -0700 Subject: Implement splice(2) and tee(2) for VFS2. Updates #138 PiperOrigin-RevId: 313326354 --- pkg/buffer/safemem.go | 82 ++++----- pkg/sentry/fsimpl/tmpfs/regular_file.go | 2 +- pkg/sentry/kernel/pipe/BUILD | 2 + pkg/sentry/kernel/pipe/pipe.go | 6 + pkg/sentry/kernel/pipe/pipe_unsafe.go | 35 ++++ pkg/sentry/kernel/pipe/vfs.go | 219 ++++++++++++++++++++++- pkg/sentry/syscalls/linux/vfs2/BUILD | 1 + pkg/sentry/syscalls/linux/vfs2/splice.go | 286 +++++++++++++++++++++++++++++++ pkg/sentry/syscalls/linux/vfs2/vfs2.go | 4 +- pkg/sentry/vfs/file_description.go | 5 + test/syscalls/linux/splice.cc | 49 ++++++ 11 files changed, 648 insertions(+), 43 deletions(-) create mode 100644 pkg/sentry/kernel/pipe/pipe_unsafe.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/splice.go (limited to 'pkg/sentry') diff --git a/pkg/buffer/safemem.go b/pkg/buffer/safemem.go index 0e5b86344..b789e56e9 100644 --- a/pkg/buffer/safemem.go +++ b/pkg/buffer/safemem.go @@ -28,12 +28,11 @@ func (b *buffer) ReadBlock() safemem.Block { return safemem.BlockFromSafeSlice(b.ReadSlice()) } -// WriteFromBlocks implements safemem.Writer.WriteFromBlocks. -// -// This will advance the write index. -func (v *View) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) { - need := int(srcs.NumBytes()) - if need == 0 { +// WriteFromSafememReader writes up to count bytes from r to v and advances the +// write index by the number of bytes written. It calls r.ReadToBlocks() at +// most once. +func (v *View) WriteFromSafememReader(r safemem.Reader, count uint64) (uint64, error) { + if count == 0 { return 0, nil } @@ -50,32 +49,33 @@ func (v *View) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) { } // Does the last block have sufficient capacity alone? - if l := firstBuf.WriteSize(); l >= need { - dst = safemem.BlockSeqOf(firstBuf.WriteBlock()) + if l := uint64(firstBuf.WriteSize()); l >= count { + dst = safemem.BlockSeqOf(firstBuf.WriteBlock().TakeFirst64(count)) } else { // Append blocks until sufficient. - need -= l + count -= l blocks = append(blocks, firstBuf.WriteBlock()) - for need > 0 { + for count > 0 { emptyBuf := bufferPool.Get().(*buffer) v.data.PushBack(emptyBuf) - need -= emptyBuf.WriteSize() - blocks = append(blocks, emptyBuf.WriteBlock()) + block := emptyBuf.WriteBlock().TakeFirst64(count) + count -= uint64(block.Len()) + blocks = append(blocks, block) } dst = safemem.BlockSeqFromSlice(blocks) } - // Perform the copy. - n, err := safemem.CopySeq(dst, srcs) + // Perform I/O. + n, err := r.ReadToBlocks(dst) v.size += int64(n) // Update all indices. - for left := int(n); left > 0; firstBuf = firstBuf.Next() { - if l := firstBuf.WriteSize(); left >= l { + for left := n; left > 0; firstBuf = firstBuf.Next() { + if l := firstBuf.WriteSize(); left >= uint64(l) { firstBuf.WriteMove(l) // Whole block. - left -= l + left -= uint64(l) } else { - firstBuf.WriteMove(left) // Partial block. + firstBuf.WriteMove(int(left)) // Partial block. left = 0 } } @@ -83,14 +83,16 @@ func (v *View) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) { return n, err } -// ReadToBlocks implements safemem.Reader.ReadToBlocks. -// -// This will not advance the read index; the caller should follow -// this call with a call to TrimFront in order to remove the read -// data from the buffer. This is done to support pipe sematics. -func (v *View) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) { - need := int(dsts.NumBytes()) - if need == 0 { +// WriteFromBlocks implements safemem.Writer.WriteFromBlocks. It advances the +// write index by the number of bytes written. +func (v *View) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) { + return v.WriteFromSafememReader(&safemem.BlockSeqReader{srcs}, srcs.NumBytes()) +} + +// ReadToSafememWriter reads up to count bytes from v to w. It does not advance +// the read index. It calls w.WriteFromBlocks() at most once. +func (v *View) ReadToSafememWriter(w safemem.Writer, count uint64) (uint64, error) { + if count == 0 { return 0, nil } @@ -105,25 +107,27 @@ func (v *View) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) { } // Is all the data in a single block? - if l := firstBuf.ReadSize(); l >= need { - src = safemem.BlockSeqOf(firstBuf.ReadBlock()) + if l := uint64(firstBuf.ReadSize()); l >= count { + src = safemem.BlockSeqOf(firstBuf.ReadBlock().TakeFirst64(count)) } else { // Build a list of all the buffers. - need -= l + count -= l blocks = append(blocks, firstBuf.ReadBlock()) - for buf := firstBuf.Next(); buf != nil && need > 0; buf = buf.Next() { - need -= buf.ReadSize() - blocks = append(blocks, buf.ReadBlock()) + for buf := firstBuf.Next(); buf != nil && count > 0; buf = buf.Next() { + block := buf.ReadBlock().TakeFirst64(count) + count -= uint64(block.Len()) + blocks = append(blocks, block) } src = safemem.BlockSeqFromSlice(blocks) } - // Perform the copy. - n, err := safemem.CopySeq(dsts, src) - - // See above: we would normally advance the read index here, but we - // don't do that in order to support pipe semantics. We rely on a - // separate call to TrimFront() in this case. + // Perform I/O. As documented, we don't advance the read index. + return w.WriteFromBlocks(src) +} - return n, err +// ReadToBlocks implements safemem.Reader.ReadToBlocks. It does not advance the +// read index by the number of bytes read, such that it's only safe to call if +// the caller guarantees that ReadToBlocks will only be called once. +func (v *View) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) { + return v.ReadToSafememWriter(&safemem.BlockSeqWriter{dsts}, dsts.NumBytes()) } diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index 3f433d666..fee174375 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -312,7 +312,7 @@ func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off f := fd.inode().impl.(*regularFile) if end := offset + srclen; end < offset { // Overflow. - return 0, syserror.EFBIG + return 0, syserror.EINVAL } var err error diff --git a/pkg/sentry/kernel/pipe/BUILD b/pkg/sentry/kernel/pipe/BUILD index f29dc0472..7bfa9075a 100644 --- a/pkg/sentry/kernel/pipe/BUILD +++ b/pkg/sentry/kernel/pipe/BUILD @@ -8,6 +8,7 @@ go_library( "device.go", "node.go", "pipe.go", + "pipe_unsafe.go", "pipe_util.go", "reader.go", "reader_writer.go", @@ -20,6 +21,7 @@ go_library( "//pkg/amutex", "//pkg/buffer", "//pkg/context", + "//pkg/safemem", "//pkg/sentry/arch", "//pkg/sentry/device", "//pkg/sentry/fs", diff --git a/pkg/sentry/kernel/pipe/pipe.go b/pkg/sentry/kernel/pipe/pipe.go index 62c8691f1..79645d7d2 100644 --- a/pkg/sentry/kernel/pipe/pipe.go +++ b/pkg/sentry/kernel/pipe/pipe.go @@ -207,7 +207,10 @@ func (p *Pipe) read(ctx context.Context, ops readOps) (int64, error) { p.mu.Lock() defer p.mu.Unlock() + return p.readLocked(ctx, ops) +} +func (p *Pipe) readLocked(ctx context.Context, ops readOps) (int64, error) { // Is the pipe empty? if p.view.Size() == 0 { if !p.HasWriters() { @@ -246,7 +249,10 @@ type writeOps struct { func (p *Pipe) write(ctx context.Context, ops writeOps) (int64, error) { p.mu.Lock() defer p.mu.Unlock() + return p.writeLocked(ctx, ops) +} +func (p *Pipe) writeLocked(ctx context.Context, ops writeOps) (int64, error) { // Can't write to a pipe with no readers. if !p.HasReaders() { return 0, syscall.EPIPE diff --git a/pkg/sentry/kernel/pipe/pipe_unsafe.go b/pkg/sentry/kernel/pipe/pipe_unsafe.go new file mode 100644 index 000000000..dd60cba24 --- /dev/null +++ b/pkg/sentry/kernel/pipe/pipe_unsafe.go @@ -0,0 +1,35 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package pipe + +import ( + "unsafe" +) + +// lockTwoPipes locks both x.mu and y.mu in an order that is guaranteed to be +// consistent for both lockTwoPipes(x, y) and lockTwoPipes(y, x), such that +// concurrent calls cannot deadlock. +// +// Preconditions: x != y. +func lockTwoPipes(x, y *Pipe) { + // Lock the two pipes in order of increasing address. + if uintptr(unsafe.Pointer(x)) < uintptr(unsafe.Pointer(y)) { + x.mu.Lock() + y.mu.Lock() + } else { + y.mu.Lock() + x.mu.Lock() + } +} diff --git a/pkg/sentry/kernel/pipe/vfs.go b/pkg/sentry/kernel/pipe/vfs.go index b54f08a30..2602bed72 100644 --- a/pkg/sentry/kernel/pipe/vfs.go +++ b/pkg/sentry/kernel/pipe/vfs.go @@ -16,7 +16,9 @@ package pipe import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/buffer" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" @@ -150,7 +152,9 @@ func (vp *VFSPipe) newFD(mnt *vfs.Mount, vfsd *vfs.Dentry, statusFlags uint32) * return &fd.vfsfd } -// VFSPipeFD implements vfs.FileDescriptionImpl for pipes. +// VFSPipeFD implements vfs.FileDescriptionImpl for pipes. It also implements +// non-atomic usermem.IO methods, allowing it to be passed as usermem.IO to +// other FileDescriptions for splice(2) and tee(2). type VFSPipeFD struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl @@ -229,3 +233,216 @@ func (fd *VFSPipeFD) PipeSize() int64 { func (fd *VFSPipeFD) SetPipeSize(size int64) (int64, error) { return fd.pipe.SetFifoSize(size) } + +// IOSequence returns a useremm.IOSequence that reads up to count bytes from, +// or writes up to count bytes to, fd. +func (fd *VFSPipeFD) IOSequence(count int64) usermem.IOSequence { + return usermem.IOSequence{ + IO: fd, + Addrs: usermem.AddrRangeSeqOf(usermem.AddrRange{0, usermem.Addr(count)}), + } +} + +// CopyIn implements usermem.IO.CopyIn. +func (fd *VFSPipeFD) CopyIn(ctx context.Context, addr usermem.Addr, dst []byte, opts usermem.IOOpts) (int, error) { + origCount := int64(len(dst)) + n, err := fd.pipe.read(ctx, readOps{ + left: func() int64 { + return int64(len(dst)) + }, + limit: func(l int64) { + dst = dst[:l] + }, + read: func(view *buffer.View) (int64, error) { + n, err := view.ReadAt(dst, 0) + view.TrimFront(int64(n)) + return int64(n), err + }, + }) + if n > 0 { + fd.pipe.Notify(waiter.EventOut) + } + if err == nil && n != origCount { + return int(n), syserror.ErrWouldBlock + } + return int(n), err +} + +// CopyOut implements usermem.IO.CopyOut. +func (fd *VFSPipeFD) CopyOut(ctx context.Context, addr usermem.Addr, src []byte, opts usermem.IOOpts) (int, error) { + origCount := int64(len(src)) + n, err := fd.pipe.write(ctx, writeOps{ + left: func() int64 { + return int64(len(src)) + }, + limit: func(l int64) { + src = src[:l] + }, + write: func(view *buffer.View) (int64, error) { + view.Append(src) + return int64(len(src)), nil + }, + }) + if n > 0 { + fd.pipe.Notify(waiter.EventIn) + } + if err == nil && n != origCount { + return int(n), syserror.ErrWouldBlock + } + return int(n), err +} + +// ZeroOut implements usermem.IO.ZeroOut. +func (fd *VFSPipeFD) ZeroOut(ctx context.Context, addr usermem.Addr, toZero int64, opts usermem.IOOpts) (int64, error) { + origCount := toZero + n, err := fd.pipe.write(ctx, writeOps{ + left: func() int64 { + return toZero + }, + limit: func(l int64) { + toZero = l + }, + write: func(view *buffer.View) (int64, error) { + view.Grow(view.Size()+toZero, true /* zero */) + return toZero, nil + }, + }) + if n > 0 { + fd.pipe.Notify(waiter.EventIn) + } + if err == nil && n != origCount { + return n, syserror.ErrWouldBlock + } + return n, err +} + +// CopyInTo implements usermem.IO.CopyInTo. +func (fd *VFSPipeFD) CopyInTo(ctx context.Context, ars usermem.AddrRangeSeq, dst safemem.Writer, opts usermem.IOOpts) (int64, error) { + count := ars.NumBytes() + if count == 0 { + return 0, nil + } + origCount := count + n, err := fd.pipe.read(ctx, readOps{ + left: func() int64 { + return count + }, + limit: func(l int64) { + count = l + }, + read: func(view *buffer.View) (int64, error) { + n, err := view.ReadToSafememWriter(dst, uint64(count)) + view.TrimFront(int64(n)) + return int64(n), err + }, + }) + if n > 0 { + fd.pipe.Notify(waiter.EventOut) + } + if err == nil && n != origCount { + return n, syserror.ErrWouldBlock + } + return n, err +} + +// CopyOutFrom implements usermem.IO.CopyOutFrom. +func (fd *VFSPipeFD) CopyOutFrom(ctx context.Context, ars usermem.AddrRangeSeq, src safemem.Reader, opts usermem.IOOpts) (int64, error) { + count := ars.NumBytes() + if count == 0 { + return 0, nil + } + origCount := count + n, err := fd.pipe.write(ctx, writeOps{ + left: func() int64 { + return count + }, + limit: func(l int64) { + count = l + }, + write: func(view *buffer.View) (int64, error) { + n, err := view.WriteFromSafememReader(src, uint64(count)) + return int64(n), err + }, + }) + if n > 0 { + fd.pipe.Notify(waiter.EventIn) + } + if err == nil && n != origCount { + return n, syserror.ErrWouldBlock + } + return n, err +} + +// SwapUint32 implements usermem.IO.SwapUint32. +func (fd *VFSPipeFD) SwapUint32(ctx context.Context, addr usermem.Addr, new uint32, opts usermem.IOOpts) (uint32, error) { + // How did a pipe get passed as the virtual address space to futex(2)? + panic("VFSPipeFD.SwapUint32 called unexpectedly") +} + +// CompareAndSwapUint32 implements usermem.IO.CompareAndSwapUint32. +func (fd *VFSPipeFD) CompareAndSwapUint32(ctx context.Context, addr usermem.Addr, old, new uint32, opts usermem.IOOpts) (uint32, error) { + panic("VFSPipeFD.CompareAndSwapUint32 called unexpectedly") +} + +// LoadUint32 implements usermem.IO.LoadUint32. +func (fd *VFSPipeFD) LoadUint32(ctx context.Context, addr usermem.Addr, opts usermem.IOOpts) (uint32, error) { + panic("VFSPipeFD.LoadUint32 called unexpectedly") +} + +// Splice reads up to count bytes from src and writes them to dst. It returns +// the number of bytes moved. +// +// Preconditions: count > 0. +func Splice(ctx context.Context, dst, src *VFSPipeFD, count int64) (int64, error) { + return spliceOrTee(ctx, dst, src, count, true /* removeFromSrc */) +} + +// Tee reads up to count bytes from src and writes them to dst, without +// removing the read bytes from src. It returns the number of bytes copied. +// +// Preconditions: count > 0. +func Tee(ctx context.Context, dst, src *VFSPipeFD, count int64) (int64, error) { + return spliceOrTee(ctx, dst, src, count, false /* removeFromSrc */) +} + +// Preconditions: count > 0. +func spliceOrTee(ctx context.Context, dst, src *VFSPipeFD, count int64, removeFromSrc bool) (int64, error) { + if dst.pipe == src.pipe { + return 0, syserror.EINVAL + } + + lockTwoPipes(dst.pipe, src.pipe) + defer dst.pipe.mu.Unlock() + defer src.pipe.mu.Unlock() + + n, err := dst.pipe.writeLocked(ctx, writeOps{ + left: func() int64 { + return count + }, + limit: func(l int64) { + count = l + }, + write: func(dstView *buffer.View) (int64, error) { + return src.pipe.readLocked(ctx, readOps{ + left: func() int64 { + return count + }, + limit: func(l int64) { + count = l + }, + read: func(srcView *buffer.View) (int64, error) { + n, err := srcView.ReadToSafememWriter(dstView, uint64(count)) + if n > 0 && removeFromSrc { + srcView.TrimFront(int64(n)) + } + return int64(n), err + }, + }) + }, + }) + if n > 0 { + dst.pipe.Notify(waiter.EventIn) + src.pipe.Notify(waiter.EventOut) + } + return n, err +} diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD index f882ef840..d56927ff5 100644 --- a/pkg/sentry/syscalls/linux/vfs2/BUILD +++ b/pkg/sentry/syscalls/linux/vfs2/BUILD @@ -22,6 +22,7 @@ go_library( "setstat.go", "signal.go", "socket.go", + "splice.go", "stat.go", "stat_amd64.go", "stat_arm64.go", diff --git a/pkg/sentry/syscalls/linux/vfs2/splice.go b/pkg/sentry/syscalls/linux/vfs2/splice.go new file mode 100644 index 000000000..8f3c22a02 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/splice.go @@ -0,0 +1,286 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/waiter" +) + +// Splice implements Linux syscall splice(2). +func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + inFD := args[0].Int() + inOffsetPtr := args[1].Pointer() + outFD := args[2].Int() + outOffsetPtr := args[3].Pointer() + count := int64(args[4].SizeT()) + flags := args[5].Int() + + if count == 0 { + return 0, nil, nil + } + if count > int64(kernel.MAX_RW_COUNT) { + count = int64(kernel.MAX_RW_COUNT) + } + + // Check for invalid flags. + if flags&^(linux.SPLICE_F_MOVE|linux.SPLICE_F_NONBLOCK|linux.SPLICE_F_MORE|linux.SPLICE_F_GIFT) != 0 { + return 0, nil, syserror.EINVAL + } + + // Get file descriptions. + inFile := t.GetFileVFS2(inFD) + if inFile == nil { + return 0, nil, syserror.EBADF + } + defer inFile.DecRef() + outFile := t.GetFileVFS2(outFD) + if outFile == nil { + return 0, nil, syserror.EBADF + } + defer outFile.DecRef() + + // Check that both files support the required directionality. + if !inFile.IsReadable() || !outFile.IsWritable() { + return 0, nil, syserror.EBADF + } + + // The operation is non-blocking if anything is non-blocking. + // + // N.B. This is a rather simplistic heuristic that avoids some + // poor edge case behavior since the exact semantics here are + // underspecified and vary between versions of Linux itself. + nonBlock := ((inFile.StatusFlags()|outFile.StatusFlags())&linux.O_NONBLOCK != 0) || (flags&linux.SPLICE_F_NONBLOCK != 0) + + // At least one file description must represent a pipe. + inPipeFD, inIsPipe := inFile.Impl().(*pipe.VFSPipeFD) + outPipeFD, outIsPipe := outFile.Impl().(*pipe.VFSPipeFD) + if !inIsPipe && !outIsPipe { + return 0, nil, syserror.EINVAL + } + + // Copy in offsets. + inOffset := int64(-1) + if inOffsetPtr != 0 { + if inIsPipe { + return 0, nil, syserror.ESPIPE + } + if inFile.Options().DenyPRead { + return 0, nil, syserror.EINVAL + } + if _, err := t.CopyIn(inOffsetPtr, &inOffset); err != nil { + return 0, nil, err + } + if inOffset < 0 { + return 0, nil, syserror.EINVAL + } + } + outOffset := int64(-1) + if outOffsetPtr != 0 { + if outIsPipe { + return 0, nil, syserror.ESPIPE + } + if outFile.Options().DenyPWrite { + return 0, nil, syserror.EINVAL + } + if _, err := t.CopyIn(outOffsetPtr, &outOffset); err != nil { + return 0, nil, err + } + if outOffset < 0 { + return 0, nil, syserror.EINVAL + } + } + + // Move data. + var ( + n int64 + err error + inCh chan struct{} + outCh chan struct{} + ) + for { + // If both input and output are pipes, delegate to the pipe + // implementation. Otherwise, exactly one end is a pipe, which we + // ensure is consistently ordered after the non-pipe FD's locks by + // passing the pipe FD as usermem.IO to the non-pipe end. + switch { + case inIsPipe && outIsPipe: + n, err = pipe.Splice(t, outPipeFD, inPipeFD, count) + case inIsPipe: + if outOffset != -1 { + n, err = outFile.PWrite(t, inPipeFD.IOSequence(count), outOffset, vfs.WriteOptions{}) + outOffset += n + } else { + n, err = outFile.Write(t, inPipeFD.IOSequence(count), vfs.WriteOptions{}) + } + case outIsPipe: + if inOffset != -1 { + n, err = inFile.PRead(t, outPipeFD.IOSequence(count), inOffset, vfs.ReadOptions{}) + inOffset += n + } else { + n, err = inFile.Read(t, outPipeFD.IOSequence(count), vfs.ReadOptions{}) + } + } + if n != 0 || err != syserror.ErrWouldBlock || nonBlock { + break + } + + // Note that the blocking behavior here is a bit different than the + // normal pattern. Because we need to have both data to read and data + // to write simultaneously, we actually explicitly block on both of + // these cases in turn before returning to the splice operation. + if inFile.Readiness(eventMaskRead)&eventMaskRead == 0 { + if inCh == nil { + inCh = make(chan struct{}, 1) + inW, _ := waiter.NewChannelEntry(inCh) + inFile.EventRegister(&inW, eventMaskRead) + defer inFile.EventUnregister(&inW) + continue // Need to refresh readiness. + } + if err = t.Block(inCh); err != nil { + break + } + } + if outFile.Readiness(eventMaskWrite)&eventMaskWrite == 0 { + if outCh == nil { + outCh = make(chan struct{}, 1) + outW, _ := waiter.NewChannelEntry(outCh) + outFile.EventRegister(&outW, eventMaskWrite) + defer outFile.EventUnregister(&outW) + continue // Need to refresh readiness. + } + if err = t.Block(outCh); err != nil { + break + } + } + } + + // Copy updated offsets out. + if inOffsetPtr != 0 { + if _, err := t.CopyOut(inOffsetPtr, &inOffset); err != nil { + return 0, nil, err + } + } + if outOffsetPtr != 0 { + if _, err := t.CopyOut(outOffsetPtr, &outOffset); err != nil { + return 0, nil, err + } + } + + if n == 0 { + return 0, nil, err + } + return uintptr(n), nil, nil +} + +// Tee implements Linux syscall tee(2). +func Tee(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + inFD := args[0].Int() + outFD := args[1].Int() + count := int64(args[2].SizeT()) + flags := args[3].Int() + + if count == 0 { + return 0, nil, nil + } + if count > int64(kernel.MAX_RW_COUNT) { + count = int64(kernel.MAX_RW_COUNT) + } + + // Check for invalid flags. + if flags&^(linux.SPLICE_F_MOVE|linux.SPLICE_F_NONBLOCK|linux.SPLICE_F_MORE|linux.SPLICE_F_GIFT) != 0 { + return 0, nil, syserror.EINVAL + } + + // Get file descriptions. + inFile := t.GetFileVFS2(inFD) + if inFile == nil { + return 0, nil, syserror.EBADF + } + defer inFile.DecRef() + outFile := t.GetFileVFS2(outFD) + if outFile == nil { + return 0, nil, syserror.EBADF + } + defer outFile.DecRef() + + // Check that both files support the required directionality. + if !inFile.IsReadable() || !outFile.IsWritable() { + return 0, nil, syserror.EBADF + } + + // The operation is non-blocking if anything is non-blocking. + // + // N.B. This is a rather simplistic heuristic that avoids some + // poor edge case behavior since the exact semantics here are + // underspecified and vary between versions of Linux itself. + nonBlock := ((inFile.StatusFlags()|outFile.StatusFlags())&linux.O_NONBLOCK != 0) || (flags&linux.SPLICE_F_NONBLOCK != 0) + + // Both file descriptions must represent pipes. + inPipeFD, inIsPipe := inFile.Impl().(*pipe.VFSPipeFD) + outPipeFD, outIsPipe := outFile.Impl().(*pipe.VFSPipeFD) + if !inIsPipe || !outIsPipe { + return 0, nil, syserror.EINVAL + } + + // Copy data. + var ( + inCh chan struct{} + outCh chan struct{} + ) + for { + n, err := pipe.Tee(t, outPipeFD, inPipeFD, count) + if n != 0 { + return uintptr(n), nil, nil + } + if err != syserror.ErrWouldBlock || nonBlock { + return 0, nil, err + } + + // Note that the blocking behavior here is a bit different than the + // normal pattern. Because we need to have both data to read and data + // to write simultaneously, we actually explicitly block on both of + // these cases in turn before returning to the tee operation. + if inFile.Readiness(eventMaskRead)&eventMaskRead == 0 { + if inCh == nil { + inCh = make(chan struct{}, 1) + inW, _ := waiter.NewChannelEntry(inCh) + inFile.EventRegister(&inW, eventMaskRead) + defer inFile.EventUnregister(&inW) + continue // Need to refresh readiness. + } + if err := t.Block(inCh); err != nil { + return 0, nil, err + } + } + if outFile.Readiness(eventMaskWrite)&eventMaskWrite == 0 { + if outCh == nil { + outCh = make(chan struct{}, 1) + outW, _ := waiter.NewChannelEntry(outCh) + outFile.EventRegister(&outW, eventMaskWrite) + defer outFile.EventUnregister(&outW) + continue // Need to refresh readiness. + } + if err := t.Block(outCh); err != nil { + return 0, nil, err + } + } + } +} diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2.go b/pkg/sentry/syscalls/linux/vfs2/vfs2.go index a332d01bd..083fdcf82 100644 --- a/pkg/sentry/syscalls/linux/vfs2/vfs2.go +++ b/pkg/sentry/syscalls/linux/vfs2/vfs2.go @@ -134,8 +134,8 @@ func Override() { s.Table[269] = syscalls.Supported("faccessat", Faccessat) s.Table[270] = syscalls.Supported("pselect", Pselect) s.Table[271] = syscalls.Supported("ppoll", Ppoll) - delete(s.Table, 275) // splice - delete(s.Table, 276) // tee + s.Table[275] = syscalls.Supported("splice", Splice) + s.Table[276] = syscalls.Supported("tee", Tee) s.Table[277] = syscalls.Supported("sync_file_range", SyncFileRange) s.Table[280] = syscalls.Supported("utimensat", Utimensat) s.Table[281] = syscalls.Supported("epoll_pwait", EpollPwait) diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index cfabd936c..bb294563d 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -210,6 +210,11 @@ func (fd *FileDescription) VirtualDentry() VirtualDentry { return fd.vd } +// Options returns the options passed to fd.Init(). +func (fd *FileDescription) Options() FileDescriptionOptions { + return fd.opts +} + // StatusFlags returns file description status flags, as for fcntl(F_GETFL). func (fd *FileDescription) StatusFlags() uint32 { return atomic.LoadUint32(&fd.statusFlags) diff --git a/test/syscalls/linux/splice.cc b/test/syscalls/linux/splice.cc index f103e2e56..08fc4b1b7 100644 --- a/test/syscalls/linux/splice.cc +++ b/test/syscalls/linux/splice.cc @@ -430,6 +430,55 @@ TEST(SpliceTest, TwoPipes) { EXPECT_EQ(memcmp(rbuf.data(), buf.data(), kPageSize), 0); } +TEST(SpliceTest, TwoPipesCircular) { + // This test deadlocks the sentry on VFS1 because VFS1 splice ordering is + // based on fs.File.UniqueID, which does not prevent circular ordering between + // e.g. inode-level locks taken by fs.FileOperations. + SKIP_IF(IsRunningWithVFS1()); + + // Create two pipes. + int fds[2]; + ASSERT_THAT(pipe(fds), SyscallSucceeds()); + const FileDescriptor first_rfd(fds[0]); + const FileDescriptor first_wfd(fds[1]); + ASSERT_THAT(pipe(fds), SyscallSucceeds()); + const FileDescriptor second_rfd(fds[0]); + const FileDescriptor second_wfd(fds[1]); + + // On Linux, each pipe is normally limited to + // include/linux/pipe_fs_i.h:PIPE_DEF_BUFFERS buffers worth of data. + constexpr size_t PIPE_DEF_BUFFERS = 16; + + // Write some data to each pipe. Below we splice 1 byte at a time between + // pipes, which very quickly causes each byte to be stored in a separate + // buffer, so we must ensure that the total amount of data in the system is <= + // PIPE_DEF_BUFFERS bytes. + std::vector buf(PIPE_DEF_BUFFERS / 2); + RandomizeBuffer(buf.data(), buf.size()); + ASSERT_THAT(write(first_wfd.get(), buf.data(), buf.size()), + SyscallSucceedsWithValue(buf.size())); + ASSERT_THAT(write(second_wfd.get(), buf.data(), buf.size()), + SyscallSucceedsWithValue(buf.size())); + + // Have another thread splice from the second pipe to the first, while we + // splice from the first to the second. The test passes if this does not + // deadlock. + const int kIterations = 1000; + DisableSave ds; + ScopedThread t([&]() { + for (int i = 0; i < kIterations; i++) { + ASSERT_THAT( + splice(second_rfd.get(), nullptr, first_wfd.get(), nullptr, 1, 0), + SyscallSucceedsWithValue(1)); + } + }); + for (int i = 0; i < kIterations; i++) { + ASSERT_THAT( + splice(first_rfd.get(), nullptr, second_wfd.get(), nullptr, 1, 0), + SyscallSucceedsWithValue(1)); + } +} + TEST(SpliceTest, Blocking) { // Create two new pipes. int first[2], second[2]; -- cgit v1.2.3 From e028714a0dd390b2321c4beeac62c5b2904cd917 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Tue, 26 May 2020 22:42:54 -0700 Subject: Support dfltuid and dfltgid mount options in the VFS2 gofer client. PiperOrigin-RevId: 313332542 --- pkg/sentry/fsimpl/gofer/gofer.go | 48 +++++++++++++++++++++++++++++++--------- 1 file changed, 38 insertions(+), 10 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 6295f6b54..131da332f 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -84,12 +84,6 @@ type filesystem struct { // devMinor is the filesystem's minor device number. devMinor is immutable. devMinor uint32 - // uid and gid are the effective KUID and KGID of the filesystem's creator, - // and are used as the owner and group for files that don't specify one. - // uid and gid are immutable. - uid auth.KUID - gid auth.KGID - // renameMu serves two purposes: // // - It synchronizes path resolution with renaming initiated by this @@ -122,6 +116,8 @@ type filesystemOptions struct { fd int aname string interop InteropMode // derived from the "cache" mount option + dfltuid auth.KUID + dfltgid auth.KGID msize uint32 version string @@ -230,6 +226,15 @@ type InternalFilesystemOptions struct { OpenSocketsByConnecting bool } +// _V9FS_DEFUID and _V9FS_DEFGID (from Linux's fs/9p/v9fs.h) are the default +// UIDs and GIDs used for files that do not provide a specific owner or group +// respectively. +const ( + // uint32(-2) doesn't work in Go. + _V9FS_DEFUID = auth.KUID(4294967294) + _V9FS_DEFGID = auth.KGID(4294967294) +) + // Name implements vfs.FilesystemType.Name. func (FilesystemType) Name() string { return Name @@ -315,6 +320,31 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt } } + // Parse the default UID and GID. + fsopts.dfltuid = _V9FS_DEFUID + if dfltuidstr, ok := mopts["dfltuid"]; ok { + delete(mopts, "dfltuid") + dfltuid, err := strconv.ParseUint(dfltuidstr, 10, 32) + if err != nil { + ctx.Warningf("gofer.FilesystemType.GetFilesystem: invalid default UID: dfltuid=%s", dfltuidstr) + return nil, nil, syserror.EINVAL + } + // In Linux, dfltuid is interpreted as a UID and is converted to a KUID + // in the caller's user namespace, but goferfs isn't + // application-mountable. + fsopts.dfltuid = auth.KUID(dfltuid) + } + fsopts.dfltgid = _V9FS_DEFGID + if dfltgidstr, ok := mopts["dfltgid"]; ok { + delete(mopts, "dfltgid") + dfltgid, err := strconv.ParseUint(dfltgidstr, 10, 32) + if err != nil { + ctx.Warningf("gofer.FilesystemType.GetFilesystem: invalid default UID: dfltgid=%s", dfltgidstr) + return nil, nil, syserror.EINVAL + } + fsopts.dfltgid = auth.KGID(dfltgid) + } + // Parse the 9P message size. fsopts.msize = 1024 * 1024 // 1M, tested to give good enough performance up to 64M if msizestr, ok := mopts["msize"]; ok { @@ -422,8 +452,6 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt client: client, clock: ktime.RealtimeClockFromContext(ctx), devMinor: devMinor, - uid: creds.EffectiveKUID, - gid: creds.EffectiveKGID, syncableDentries: make(map[*dentry]struct{}), specialFileFDs: make(map[*specialFileFD]struct{}), } @@ -672,8 +700,8 @@ func (fs *filesystem) newDentry(ctx context.Context, file p9file, qid p9.QID, ma file: file, ino: qid.Path, mode: uint32(attr.Mode), - uid: uint32(fs.uid), - gid: uint32(fs.gid), + uid: uint32(fs.opts.dfltuid), + gid: uint32(fs.opts.dfltgid), blockSize: usermem.PageSize, handle: handle{ fd: -1, -- cgit v1.2.3 From 17d500f907c7bfc4d8365cceb4b305e9e3e6733c Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Wed, 27 May 2020 10:16:09 -0700 Subject: g3doc/fuse: add more references --- pkg/sentry/fs/g3doc/fuse.md | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/g3doc/fuse.md b/pkg/sentry/fs/g3doc/fuse.md index 635cc009b..cc08c14f5 100644 --- a/pkg/sentry/fs/g3doc/fuse.md +++ b/pkg/sentry/fs/g3doc/fuse.md @@ -99,7 +99,7 @@ ops can be implemented in parallel. ## FUSE Protocol The FUSE protocol is a request-response protocol. All requests are initiated by -the client. The wire-format for the protocol is raw c structs serialized to +the client. The wire-format for the protocol is raw C structs serialized to memory. All FUSE requests begin with the following request header: @@ -255,6 +255,7 @@ I/O syscalls like `read(2)`, `write(2)` and `mmap(2)`. # References -- `fuse(4)` manpage. -- Linux kernel FUSE documentation: - https://www.kernel.org/doc/html/latest/filesystems/fuse.html +- [fuse(4) Linux manual page](https://www.man7.org/linux/man-pages/man4/fuse.4.html) +- [Linux kernel FUSE documentation](https://www.kernel.org/doc/html/latest/filesystems/fuse.html) +- [The reference implementation of the Linux FUSE (Filesystem in Userspace) interface](https://github.com/libfuse/libfuse) +- [The kernel interface of FUSE](https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/include/uapi/linux/fuse.h) -- cgit v1.2.3 From 32021bce9607e09f3d21d3d28047d49340da231e Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Wed, 27 May 2020 18:18:17 -0700 Subject: Correctly update link and ref counts in rmdir. Inotify sends events when a watch target is reaches a link count of 0 (see include/linux/fsnotify.h:fsnotify_inoderemove). Currently, we do not account for both dir/ and dir/.. in unlink, causing syscalls/linux/inotify.cc:WatchTargetDeletionGeneratesEvent to fail because the expected inotify events are not generated. Furthermore, we should DecRef() once the inode reaches zero links; otherwise, we will leak a reference. PiperOrigin-RevId: 313502091 --- pkg/sentry/fsimpl/tmpfs/filesystem.go | 4 +++- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 15 ++++++++------- 2 files changed, 11 insertions(+), 8 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 80fa7b29d..7c04570f1 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -603,8 +603,10 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error return err } parentDir.removeChildLocked(child) - parentDir.inode.decLinksLocked() // from child's ".." + // Remove links for child, child/., and child/.. child.inode.decLinksLocked() + child.inode.decLinksLocked() + parentDir.inode.decLinksLocked() vfsObj.CommitDeleteDentry(&child.vfsd) parentDir.inode.touchCMtime() return nil diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 1e781aecd..b739095b7 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -209,11 +209,9 @@ type inode struct { // refs is a reference count. refs is accessed using atomic memory // operations. // - // A reference is held on all inodes that are reachable in the filesystem - // tree. For non-directories (which may have multiple hard links), this - // means that a reference is dropped when nlink reaches 0. For directories, - // nlink never reaches 0 due to the "." entry; instead, - // filesystem.RmdirAt() drops the reference. + // A reference is held on all inodes as long as they are reachable in the + // filesystem tree, i.e. nlink is nonzero. This reference is dropped when + // nlink reaches 0. refs int64 // xattrs implements extended attributes. @@ -276,14 +274,17 @@ func (i *inode) incLinksLocked() { atomic.AddUint32(&i.nlink, 1) } -// decLinksLocked decrements i's link count. +// decLinksLocked decrements i's link count. If the link count reaches 0, we +// remove a reference on i as well. // // Preconditions: filesystem.mu must be locked for writing. i.nlink != 0. func (i *inode) decLinksLocked() { if i.nlink == 0 { panic("tmpfs.inode.decLinksLocked() called with no existing links") } - atomic.AddUint32(&i.nlink, ^uint32(0)) + if atomic.AddUint32(&i.nlink, ^uint32(0)) == 0 { + i.decRef() + } } func (i *inode) incRef() { -- cgit v1.2.3 From c55b84e16aeb4481106661e3877c50edbf281762 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Thu, 28 May 2020 16:44:15 -0700 Subject: Enable iptables source filtering (-s/--source) --- pkg/sentry/socket/netfilter/netfilter.go | 21 ++++++++--- pkg/tcpip/stack/iptables.go | 47 +----------------------- pkg/tcpip/stack/iptables_types.go | 62 ++++++++++++++++++++++++++++++++ test/iptables/filter_input.go | 60 +++++++++++++++++++++++++++++++ test/iptables/iptables_test.go | 8 +++++ 5 files changed, 147 insertions(+), 51 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 789bb94c8..47ff48c00 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -64,6 +64,8 @@ const enableLogging = false var emptyFilter = stack.IPHeaderFilter{ Dst: "\x00\x00\x00\x00", DstMask: "\x00\x00\x00\x00", + Src: "\x00\x00\x00\x00", + SrcMask: "\x00\x00\x00\x00", } // nflog logs messages related to the writing and reading of iptables. @@ -214,11 +216,16 @@ func convertNetstackToBinary(tablename string, table stack.Table) (linux.KernelI } copy(entry.IPTEntry.IP.Dst[:], rule.Filter.Dst) copy(entry.IPTEntry.IP.DstMask[:], rule.Filter.DstMask) + copy(entry.IPTEntry.IP.Src[:], rule.Filter.Src) + copy(entry.IPTEntry.IP.SrcMask[:], rule.Filter.SrcMask) copy(entry.IPTEntry.IP.OutputInterface[:], rule.Filter.OutputInterface) copy(entry.IPTEntry.IP.OutputInterfaceMask[:], rule.Filter.OutputInterfaceMask) if rule.Filter.DstInvert { entry.IPTEntry.IP.InverseFlags |= linux.IPT_INV_DSTIP } + if rule.Filter.SrcInvert { + entry.IPTEntry.IP.InverseFlags |= linux.IPT_INV_SRCIP + } if rule.Filter.OutputInterfaceInvert { entry.IPTEntry.IP.InverseFlags |= linux.IPT_INV_VIA_OUT } @@ -737,6 +744,9 @@ func filterFromIPTIP(iptip linux.IPTIP) (stack.IPHeaderFilter, error) { if len(iptip.Dst) != header.IPv4AddressSize || len(iptip.DstMask) != header.IPv4AddressSize { return stack.IPHeaderFilter{}, fmt.Errorf("incorrect length of destination (%d) and/or destination mask (%d) fields", len(iptip.Dst), len(iptip.DstMask)) } + if len(iptip.Src) != header.IPv4AddressSize || len(iptip.SrcMask) != header.IPv4AddressSize { + return stack.IPHeaderFilter{}, fmt.Errorf("incorrect length of source (%d) and/or source mask (%d) fields", len(iptip.Src), len(iptip.SrcMask)) + } n := bytes.IndexByte([]byte(iptip.OutputInterface[:]), 0) if n == -1 { @@ -755,6 +765,9 @@ func filterFromIPTIP(iptip linux.IPTIP) (stack.IPHeaderFilter, error) { Dst: tcpip.Address(iptip.Dst[:]), DstMask: tcpip.Address(iptip.DstMask[:]), DstInvert: iptip.InverseFlags&linux.IPT_INV_DSTIP != 0, + Src: tcpip.Address(iptip.Src[:]), + SrcMask: tcpip.Address(iptip.SrcMask[:]), + SrcInvert: iptip.InverseFlags&linux.IPT_INV_SRCIP != 0, OutputInterface: ifname, OutputInterfaceMask: ifnameMask, OutputInterfaceInvert: iptip.InverseFlags&linux.IPT_INV_VIA_OUT != 0, @@ -765,15 +778,13 @@ func containsUnsupportedFields(iptip linux.IPTIP) bool { // The following features are supported: // - Protocol // - Dst and DstMask + // - Src and SrcMask // - The inverse destination IP check flag // - OutputInterface, OutputInterfaceMask and its inverse. - var emptyInetAddr = linux.InetAddr{} var emptyInterface = [linux.IFNAMSIZ]byte{} // Disable any supported inverse flags. - inverseMask := uint8(linux.IPT_INV_DSTIP) | uint8(linux.IPT_INV_VIA_OUT) - return iptip.Src != emptyInetAddr || - iptip.SrcMask != emptyInetAddr || - iptip.InputInterface != emptyInterface || + inverseMask := uint8(linux.IPT_INV_DSTIP) | uint8(linux.IPT_INV_SRCIP) | uint8(linux.IPT_INV_VIA_OUT) + return iptip.InputInterface != emptyInterface || iptip.InputInterfaceMask != emptyInterface || iptip.Flags != 0 || iptip.InverseFlags&^inverseMask != 0 diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go index 443423b3c..709ede3fa 100644 --- a/pkg/tcpip/stack/iptables.go +++ b/pkg/tcpip/stack/iptables.go @@ -16,7 +16,6 @@ package stack import ( "fmt" - "strings" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/header" @@ -314,7 +313,7 @@ func (it *IPTables) checkRule(hook Hook, pkt *PacketBuffer, table Table, ruleIdx } // Check whether the packet matches the IP header filter. - if !filterMatch(rule.Filter, header.IPv4(pkt.NetworkHeader), hook, nicName) { + if !rule.Filter.match(header.IPv4(pkt.NetworkHeader), hook, nicName) { // Continue on to the next rule. return RuleJump, ruleIdx + 1 } @@ -335,47 +334,3 @@ func (it *IPTables) checkRule(hook Hook, pkt *PacketBuffer, table Table, ruleIdx // All the matchers matched, so run the target. return rule.Target.Action(pkt, &it.connections, hook, gso, r, address) } - -func filterMatch(filter IPHeaderFilter, hdr header.IPv4, hook Hook, nicName string) bool { - // TODO(gvisor.dev/issue/170): Support other fields of the filter. - // Check the transport protocol. - if filter.Protocol != 0 && filter.Protocol != hdr.TransportProtocol() { - return false - } - - // Check the destination IP. - dest := hdr.DestinationAddress() - matches := true - for i := range filter.Dst { - if dest[i]&filter.DstMask[i] != filter.Dst[i] { - matches = false - break - } - } - if matches == filter.DstInvert { - return false - } - - // Check the output interface. - // TODO(gvisor.dev/issue/170): Add the check for FORWARD and POSTROUTING - // hooks after supported. - if hook == Output { - n := len(filter.OutputInterface) - if n == 0 { - return true - } - - // If the interface name ends with '+', any interface which begins - // with the name should be matched. - ifName := filter.OutputInterface - matches = true - if strings.HasSuffix(ifName, "+") { - matches = strings.HasPrefix(nicName, ifName[:n-1]) - } else { - matches = nicName == ifName - } - return filter.OutputInterfaceInvert != matches - } - - return true -} diff --git a/pkg/tcpip/stack/iptables_types.go b/pkg/tcpip/stack/iptables_types.go index fe06007ae..a3bd3e700 100644 --- a/pkg/tcpip/stack/iptables_types.go +++ b/pkg/tcpip/stack/iptables_types.go @@ -15,7 +15,10 @@ package stack import ( + "strings" + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/header" ) // A Hook specifies one of the hooks built into the network stack. @@ -159,6 +162,16 @@ type IPHeaderFilter struct { // comparison. DstInvert bool + // Src matches the source IP address. + Src tcpip.Address + + // SrcMask masks bits of the source IP address when comparing with Src. + SrcMask tcpip.Address + + // SrcInvert inverts the meaning of the source IP check, i.e. when true the + // filter will match packets that fail the source comparison. + SrcInvert bool + // OutputInterface matches the name of the outgoing interface for the // packet. OutputInterface string @@ -173,6 +186,55 @@ type IPHeaderFilter struct { OutputInterfaceInvert bool } +// match returns whether hdr matches the filter. +func (fl IPHeaderFilter) match(hdr header.IPv4, hook Hook, nicName string) bool { + // TODO(gvisor.dev/issue/170): Support other fields of the filter. + // Check the transport protocol. + if fl.Protocol != 0 && fl.Protocol != hdr.TransportProtocol() { + return false + } + + // Check the source and destination IPs. + if !filterAddress(hdr.DestinationAddress(), fl.DstMask, fl.Dst, fl.DstInvert) || !filterAddress(hdr.SourceAddress(), fl.SrcMask, fl.Src, fl.SrcInvert) { + return false + } + + // Check the output interface. + // TODO(gvisor.dev/issue/170): Add the check for FORWARD and POSTROUTING + // hooks after supported. + if hook == Output { + n := len(fl.OutputInterface) + if n == 0 { + return true + } + + // If the interface name ends with '+', any interface which begins + // with the name should be matched. + ifName := fl.OutputInterface + matches := true + if strings.HasSuffix(ifName, "+") { + matches = strings.HasPrefix(nicName, ifName[:n-1]) + } else { + matches = nicName == ifName + } + return fl.OutputInterfaceInvert != matches + } + + return true +} + +// filterAddress returns whether addr matches the filter. +func filterAddress(addr, mask, filterAddr tcpip.Address, invert bool) bool { + matches := true + for i := range filterAddr { + if addr[i]&mask[i] != filterAddr[i] { + matches = false + break + } + } + return matches != invert +} + // A Matcher is the interface for matching packets. type Matcher interface { // Name returns the name of the Matcher. diff --git a/test/iptables/filter_input.go b/test/iptables/filter_input.go index 41e0cfa8d..14e385f5a 100644 --- a/test/iptables/filter_input.go +++ b/test/iptables/filter_input.go @@ -49,6 +49,8 @@ func init() { RegisterTestCase(FilterInputJumpTwice{}) RegisterTestCase(FilterInputDestination{}) RegisterTestCase(FilterInputInvertDestination{}) + RegisterTestCase(FilterInputSource{}) + RegisterTestCase(FilterInputInvertSource{}) } // FilterInputDropUDP tests that we can drop UDP traffic. @@ -667,3 +669,61 @@ func (FilterInputInvertDestination) ContainerAction(ip net.IP) error { func (FilterInputInvertDestination) LocalAction(ip net.IP) error { return sendUDPLoop(ip, acceptPort, sendloopDuration) } + +// FilterInputSource verifies that we can filter packets via `-d +// `. +type FilterInputSource struct{} + +// Name implements TestCase.Name. +func (FilterInputSource) Name() string { + return "FilterInputSource" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterInputSource) ContainerAction(ip net.IP) error { + // Make INPUT's default action DROP, then ACCEPT all packets from this + // machine. + rules := [][]string{ + {"-P", "INPUT", "DROP"}, + {"-A", "INPUT", "-s", fmt.Sprintf("%v", ip), "-j", "ACCEPT"}, + } + if err := filterTableRules(rules); err != nil { + return err + } + + return listenUDP(acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (FilterInputSource) LocalAction(ip net.IP) error { + return sendUDPLoop(ip, acceptPort, sendloopDuration) +} + +// FilterInputInvertSource verifies that we can filter packets via `! -d +// `. +type FilterInputInvertSource struct{} + +// Name implements TestCase.Name. +func (FilterInputInvertSource) Name() string { + return "FilterInputInvertSource" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterInputInvertSource) ContainerAction(ip net.IP) error { + // Make INPUT's default action DROP, then ACCEPT all packets not bound + // for 127.0.0.1. + rules := [][]string{ + {"-P", "INPUT", "DROP"}, + {"-A", "INPUT", "!", "-s", localIP, "-j", "ACCEPT"}, + } + if err := filterTableRules(rules); err != nil { + return err + } + + return listenUDP(acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (FilterInputInvertSource) LocalAction(ip net.IP) error { + return sendUDPLoop(ip, acceptPort, sendloopDuration) +} diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go index 4fd2cb46a..172ad9e16 100644 --- a/test/iptables/iptables_test.go +++ b/test/iptables/iptables_test.go @@ -302,3 +302,11 @@ func TestNATPreRedirectInvert(t *testing.T) { func TestNATRedirectRequiresProtocol(t *testing.T) { singleTest(t, NATRedirectRequiresProtocol{}) } + +func TestInputSource(t *testing.T) { + singleTest(t, FilterInputSource{}) +} + +func TestInputInvertSource(t *testing.T) { + singleTest(t, FilterInputInvertSource{}) +} -- cgit v1.2.3 From 50844fee29aed500fff41f417f6183867ca0f0b5 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Thu, 28 May 2020 19:14:26 -0700 Subject: Split VFS.MountAt into VFS.MountDisconnected and VFS.ConnectMountAt. This makes it straightforward to create bind mounts internally in VFS2: Given a bind mount root represented by vfs.VirtualDentry vd: - Create a new mount with VFS.NewDisconnectedMount(vd.Mount().Filesystem(), vd.Dentry()). - Connect the resulting mount in the appropriate namespace with VFS.ConnectMountAt(). Note that the resulting bind mount is non-recursive; recursive bind mounting requires explicitly duplicating all children of the original mount, which is best handled internally by VFS. Updates #179 PiperOrigin-RevId: 313703963 --- pkg/sentry/vfs/mount.go | 56 ++++++++++++++++++++++++++++++++++++------------- pkg/sentry/vfs/vfs.go | 4 ++++ 2 files changed, 45 insertions(+), 15 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index 02850b65c..e4ac6524b 100644 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go @@ -28,9 +28,6 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) -// lastMountID is used to allocate mount ids. Must be accessed atomically. -var lastMountID uint64 - // A Mount is a replacement of a Dentry (Mount.key.point) from one Filesystem // (Mount.key.parent.fs) with a Dentry (Mount.root) from another Filesystem // (Mount.fs), which applies to path resolution in the context of a particular @@ -97,7 +94,7 @@ type Mount struct { func newMount(vfs *VirtualFilesystem, fs *Filesystem, root *Dentry, mntns *MountNamespace, opts *MountOptions) *Mount { mnt := &Mount{ - ID: atomic.AddUint64(&lastMountID, 1), + ID: atomic.AddUint64(&vfs.lastMountID, 1), vfs: vfs, fs: fs, root: root, @@ -111,6 +108,16 @@ func newMount(vfs *VirtualFilesystem, fs *Filesystem, root *Dentry, mntns *Mount return mnt } +// Options returns a copy of the MountOptions currently applicable to mnt. +func (mnt *Mount) Options() MountOptions { + mnt.vfs.mountMu.Lock() + defer mnt.vfs.mountMu.Unlock() + return MountOptions{ + Flags: mnt.flags, + ReadOnly: mnt.readOnly(), + } +} + // A MountNamespace is a collection of Mounts. // // MountNamespaces are reference-counted. Unless otherwise specified, all @@ -148,7 +155,7 @@ type MountNamespace struct { func (vfs *VirtualFilesystem) NewMountNamespace(ctx context.Context, creds *auth.Credentials, source, fsTypeName string, opts *GetFilesystemOptions) (*MountNamespace, error) { rft := vfs.getFilesystemType(fsTypeName) if rft == nil { - ctx.Warningf("Unknown filesystem: %s", fsTypeName) + ctx.Warningf("Unknown filesystem type: %s", fsTypeName) return nil, syserror.ENODEV } fs, root, err := rft.fsType.GetFilesystem(ctx, vfs, creds, source, *opts) @@ -175,26 +182,34 @@ func (vfs *VirtualFilesystem) NewDisconnectedMount(fs *Filesystem, root *Dentry, return newMount(vfs, fs, root, nil /* mntns */, opts), nil } -// MountAt creates and mounts a Filesystem configured by the given arguments. -func (vfs *VirtualFilesystem) MountAt(ctx context.Context, creds *auth.Credentials, source string, target *PathOperation, fsTypeName string, opts *MountOptions) error { +// MountDisconnected creates a Filesystem configured by the given arguments, +// then returns a Mount representing it. The new Mount is not associated with +// any MountNamespace and is not connected to any other Mounts. +func (vfs *VirtualFilesystem) MountDisconnected(ctx context.Context, creds *auth.Credentials, source string, fsTypeName string, opts *MountOptions) (*Mount, error) { rft := vfs.getFilesystemType(fsTypeName) if rft == nil { - return syserror.ENODEV + return nil, syserror.ENODEV } if !opts.InternalMount && !rft.opts.AllowUserMount { - return syserror.ENODEV + return nil, syserror.ENODEV } fs, root, err := rft.fsType.GetFilesystem(ctx, vfs, creds, source, opts.GetFilesystemOptions) if err != nil { - return err + return nil, err } + defer root.DecRef() + defer fs.DecRef() + return vfs.NewDisconnectedMount(fs, root, opts) +} +// ConnectMountAt connects mnt at the path represented by target. +// +// Preconditions: mnt must be disconnected. +func (vfs *VirtualFilesystem) ConnectMountAt(ctx context.Context, creds *auth.Credentials, mnt *Mount, target *PathOperation) error { // We can't hold vfs.mountMu while calling FilesystemImpl methods due to // lock ordering. vd, err := vfs.GetDentryAt(ctx, creds, target, &GetDentryOptions{}) if err != nil { - root.DecRef() - fs.DecRef() return err } vfs.mountMu.Lock() @@ -204,8 +219,6 @@ func (vfs *VirtualFilesystem) MountAt(ctx context.Context, creds *auth.Credentia vd.dentry.mu.Unlock() vfs.mountMu.Unlock() vd.DecRef() - root.DecRef() - fs.DecRef() return syserror.ENOENT } // vd might have been mounted over between vfs.GetDentryAt() and @@ -238,7 +251,6 @@ func (vfs *VirtualFilesystem) MountAt(ctx context.Context, creds *auth.Credentia // point and the mount root are directories, or neither are, and returns // ENOTDIR if this is not the case. mntns := vd.mount.ns - mnt := newMount(vfs, fs, root, mntns, opts) vfs.mounts.seq.BeginWrite() vfs.connectLocked(mnt, vd, mntns) vfs.mounts.seq.EndWrite() @@ -247,6 +259,19 @@ func (vfs *VirtualFilesystem) MountAt(ctx context.Context, creds *auth.Credentia return nil } +// MountAt creates and mounts a Filesystem configured by the given arguments. +func (vfs *VirtualFilesystem) MountAt(ctx context.Context, creds *auth.Credentials, source string, target *PathOperation, fsTypeName string, opts *MountOptions) error { + mnt, err := vfs.MountDisconnected(ctx, creds, source, fsTypeName, opts) + if err != nil { + return err + } + if err := vfs.ConnectMountAt(ctx, creds, mnt, target); err != nil { + mnt.DecRef() + return err + } + return nil +} + // UmountAt removes the Mount at the given path. func (vfs *VirtualFilesystem) UmountAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *UmountOptions) error { if opts.Flags&^(linux.MNT_FORCE|linux.MNT_DETACH) != 0 { @@ -377,6 +402,7 @@ func (vfs *VirtualFilesystem) connectLocked(mnt *Mount, vd VirtualDentry, mntns } vd.mount.children[mnt] = struct{}{} atomic.AddUint32(&vd.dentry.mounts, 1) + mnt.ns = mntns mntns.mountpoints[vd.dentry]++ vfs.mounts.insertSeqed(mnt) vfsmpmounts, ok := vfs.mountpoints[vd.dentry] diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 8d7f8f8af..6d2ba53ea 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -82,6 +82,10 @@ type VirtualFilesystem struct { // mountpoints is analogous to Linux's mountpoint_hashtable. mountpoints map[*Dentry]map[*Mount]struct{} + // lastMountID is the last allocated mount ID. lastMountID is accessed + // using atomic memory operations. + lastMountID uint64 + // anonMount is a Mount, not included in mounts or mountpoints, // representing an anonFilesystem. anonMount is used to back // VirtualDentries returned by VirtualFilesystem.NewAnonVirtualDentry(). -- cgit v1.2.3 From fe464f44b7d3696bafd9a2faf3750e1dc4d56d80 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 29 May 2020 08:07:44 -0700 Subject: Port inotify to vfs2, with support in tmpfs. Support in other filesystem impls is still needed. Unlike in Linux and vfs1, we need to plumb inotify down to each filesystem implementation in order to keep track of links/inode structures properly. IN_EXCL_UNLINK still needs to be implemented, as well as a few inotify hooks that are not present in either vfs1 or vfs2. Those will be addressed in subsequent changes. Updates #1479. PiperOrigin-RevId: 313781995 --- pkg/sentry/fsimpl/ext/dentry.go | 12 + pkg/sentry/fsimpl/gofer/gofer.go | 12 + pkg/sentry/fsimpl/kernfs/kernfs.go | 12 + pkg/sentry/fsimpl/tmpfs/BUILD | 1 + pkg/sentry/fsimpl/tmpfs/directory.go | 1 + pkg/sentry/fsimpl/tmpfs/filesystem.go | 40 +- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 60 ++- pkg/sentry/kernel/fd_table.go | 8 +- pkg/sentry/syscalls/linux/vfs2/BUILD | 1 + pkg/sentry/syscalls/linux/vfs2/inotify.go | 134 ++++++ pkg/sentry/syscalls/linux/vfs2/read_write.go | 36 ++ pkg/sentry/syscalls/linux/vfs2/vfs2.go | 8 +- pkg/sentry/vfs/BUILD | 15 + pkg/sentry/vfs/anonfs.go | 12 + pkg/sentry/vfs/dentry.go | 27 ++ pkg/sentry/vfs/inotify.go | 675 +++++++++++++++++++++++++++ pkg/sentry/vfs/vfs.go | 1 + test/syscalls/linux/BUILD | 5 +- test/syscalls/linux/inotify.cc | 176 ++++++- 19 files changed, 1211 insertions(+), 25 deletions(-) create mode 100644 pkg/sentry/syscalls/linux/vfs2/inotify.go create mode 100644 pkg/sentry/vfs/inotify.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/ext/dentry.go b/pkg/sentry/fsimpl/ext/dentry.go index bfbd7c3d4..4d0deaf03 100644 --- a/pkg/sentry/fsimpl/ext/dentry.go +++ b/pkg/sentry/fsimpl/ext/dentry.go @@ -60,3 +60,15 @@ func (d *dentry) DecRef() { // inode.decRef(). d.inode.decRef() } + +// InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. +// +// TODO(gvisor.dev/issue/1479): Implement inotify. +func (d *dentry) InotifyWithParent(events uint32, cookie uint32) {} + +// Watches implements vfs.DentryImpl.Watches. +// +// TODO(gvisor.dev/issue/1479): Implement inotify. +func (d *dentry) Watches() *vfs.Watches { + return nil +} diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 131da332f..850482a19 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -1039,6 +1039,18 @@ func (d *dentry) decRefLocked() { } } +// InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. +// +// TODO(gvisor.dev/issue/1479): Implement inotify. +func (d *dentry) InotifyWithParent(events uint32, cookie uint32) {} + +// Watches implements vfs.DentryImpl.Watches. +// +// TODO(gvisor.dev/issue/1479): Implement inotify. +func (d *dentry) Watches() *vfs.Watches { + return nil +} + // checkCachingLocked should be called after d's reference count becomes 0 or it // becomes disowned. // diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index a83151ad3..682545994 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -225,6 +225,18 @@ func (d *Dentry) destroy() { } } +// InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. +// +// TODO(gvisor.dev/issue/1479): Implement inotify. +func (d *Dentry) InotifyWithParent(events uint32, cookie uint32) {} + +// Watches implements vfs.DentryImpl.Watches. +// +// TODO(gvisor.dev/issue/1479): Implement inotify. +func (d *Dentry) Watches() *vfs.Watches { + return nil +} + // InsertChild inserts child into the vfs dentry cache with the given name under // this dentry. This does not update the directory inode, so calling this on // it's own isn't sufficient to insert a child into a directory. InsertChild diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD index 007be1572..062321cbc 100644 --- a/pkg/sentry/fsimpl/tmpfs/BUILD +++ b/pkg/sentry/fsimpl/tmpfs/BUILD @@ -59,6 +59,7 @@ go_library( "//pkg/sentry/pgalloc", "//pkg/sentry/platform", "//pkg/sentry/socket/unix/transport", + "//pkg/sentry/uniqueid", "//pkg/sentry/usage", "//pkg/sentry/vfs", "//pkg/sentry/vfs/lock", diff --git a/pkg/sentry/fsimpl/tmpfs/directory.go b/pkg/sentry/fsimpl/tmpfs/directory.go index f2399981b..8bc475f88 100644 --- a/pkg/sentry/fsimpl/tmpfs/directory.go +++ b/pkg/sentry/fsimpl/tmpfs/directory.go @@ -112,6 +112,7 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba dir.iterMu.Lock() defer dir.iterMu.Unlock() + fd.dentry().InotifyWithParent(linux.IN_ACCESS, 0) fd.inode().touchAtime(fd.vfsfd.Mount()) if fd.off == 0 { diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 7c04570f1..b4159f904 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -177,6 +177,12 @@ func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(pa if err := create(parentDir, name); err != nil { return err } + + ev := linux.IN_CREATE + if dir { + ev |= linux.IN_ISDIR + } + parentDir.inode.watches.Notify(name, uint32(ev), 0) parentDir.inode.touchCMtime() return nil } @@ -241,6 +247,7 @@ func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. return syserror.EMLINK } d.inode.incLinksLocked() + d.inode.watches.Notify("", linux.IN_ATTRIB, 0) parentDir.insertChildLocked(fs.newDentry(d.inode), name) return nil }) @@ -354,6 +361,7 @@ afterTrailingSymlink: if err != nil { return nil, err } + parentDir.inode.watches.Notify(name, linux.IN_CREATE, 0) parentDir.inode.touchCMtime() return fd, nil } @@ -559,6 +567,8 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa newParentDir.inode.touchCMtime() } renamed.inode.touchCtime() + + vfs.InotifyRename(ctx, &renamed.inode.watches, &oldParentDir.inode.watches, &newParentDir.inode.watches, oldName, newName, renamed.inode.isDir()) return nil } @@ -603,6 +613,7 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error return err } parentDir.removeChildLocked(child) + parentDir.inode.watches.Notify(name, linux.IN_DELETE|linux.IN_ISDIR, 0) // Remove links for child, child/., and child/.. child.inode.decLinksLocked() child.inode.decLinksLocked() @@ -620,7 +631,14 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts if err != nil { return err } - return d.inode.setStat(ctx, rp.Credentials(), &opts.Stat) + if err := d.inode.setStat(ctx, rp.Credentials(), &opts.Stat); err != nil { + return err + } + + if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { + d.InotifyWithParent(ev, 0) + } + return nil } // StatAt implements vfs.FilesystemImpl.StatAt. @@ -700,6 +718,12 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error if err := vfsObj.PrepareDeleteDentry(mntns, &child.vfsd); err != nil { return err } + + // Generate inotify events. Note that this must take place before the link + // count of the child is decremented, or else the watches may be dropped + // before these events are added. + vfs.InotifyRemoveChild(&child.inode.watches, &parentDir.inode.watches, name) + parentDir.removeChildLocked(child) child.inode.decLinksLocked() vfsObj.CommitDeleteDentry(&child.vfsd) @@ -756,7 +780,12 @@ func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt if err != nil { return err } - return d.inode.setxattr(rp.Credentials(), &opts) + if err := d.inode.setxattr(rp.Credentials(), &opts); err != nil { + return err + } + + d.InotifyWithParent(linux.IN_ATTRIB, 0) + return nil } // RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. @@ -767,7 +796,12 @@ func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, if err != nil { return err } - return d.inode.removexattr(rp.Credentials(), name) + if err := d.inode.removexattr(rp.Credentials(), name); err != nil { + return err + } + + d.InotifyWithParent(linux.IN_ATTRIB, 0) + return nil } // PrependPath implements vfs.FilesystemImpl.PrependPath. diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index b739095b7..1d83b6840 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -201,6 +201,26 @@ func (d *dentry) DecRef() { d.inode.decRef() } +// InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. +func (d *dentry) InotifyWithParent(events uint32, cookie uint32) { + if d.inode.isDir() { + events |= linux.IN_ISDIR + } + + // The ordering below is important, Linux always notifies the parent first. + if d.parent != nil { + // Note that d.parent or d.name may be stale if there is a concurrent + // rename operation. Inotify does not provide consistency guarantees. + d.parent.inode.watches.Notify(d.name, events, cookie) + } + d.inode.watches.Notify("", events, cookie) +} + +// Watches implements vfs.DentryImpl.Watches. +func (d *dentry) Watches() *vfs.Watches { + return &d.inode.watches +} + // inode represents a filesystem object. type inode struct { // fs is the owning filesystem. fs is immutable. @@ -236,6 +256,9 @@ type inode struct { // Advisory file locks, which lock at the inode level. locks lock.FileLocks + // Inotify watches for this inode. + watches vfs.Watches + impl interface{} // immutable } @@ -257,6 +280,7 @@ func (i *inode) init(impl interface{}, fs *filesystem, creds *auth.Credentials, i.ctime = now i.mtime = now // i.nlink initialized by caller + i.watches = vfs.Watches{} i.impl = impl } @@ -307,6 +331,7 @@ func (i *inode) tryIncRef() bool { func (i *inode) decRef() { if refs := atomic.AddInt64(&i.refs, -1); refs == 0 { + i.watches.HandleDeletion() if regFile, ok := i.impl.(*regularFile); ok { // Release memory used by regFile to store data. Since regFile is // no longer usable, we don't need to grab any locks or update any @@ -628,8 +653,12 @@ func (fd *fileDescription) filesystem() *filesystem { return fd.vfsfd.Mount().Filesystem().Impl().(*filesystem) } +func (fd *fileDescription) dentry() *dentry { + return fd.vfsfd.Dentry().Impl().(*dentry) +} + func (fd *fileDescription) inode() *inode { - return fd.vfsfd.Dentry().Impl().(*dentry).inode + return fd.dentry().inode } // Stat implements vfs.FileDescriptionImpl.Stat. @@ -642,7 +671,16 @@ func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linu // SetStat implements vfs.FileDescriptionImpl.SetStat. func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { creds := auth.CredentialsFromContext(ctx) - return fd.inode().setStat(ctx, creds, &opts.Stat) + d := fd.dentry() + if err := d.inode.setStat(ctx, creds, &opts.Stat); err != nil { + return err + } + + // Generate inotify events. + if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { + d.InotifyWithParent(ev, 0) + } + return nil } // Listxattr implements vfs.FileDescriptionImpl.Listxattr. @@ -657,12 +695,26 @@ func (fd *fileDescription) Getxattr(ctx context.Context, opts vfs.GetxattrOption // Setxattr implements vfs.FileDescriptionImpl.Setxattr. func (fd *fileDescription) Setxattr(ctx context.Context, opts vfs.SetxattrOptions) error { - return fd.inode().setxattr(auth.CredentialsFromContext(ctx), &opts) + d := fd.dentry() + if err := d.inode.setxattr(auth.CredentialsFromContext(ctx), &opts); err != nil { + return err + } + + // Generate inotify events. + d.InotifyWithParent(linux.IN_ATTRIB, 0) + return nil } // Removexattr implements vfs.FileDescriptionImpl.Removexattr. func (fd *fileDescription) Removexattr(ctx context.Context, name string) error { - return fd.inode().removexattr(auth.CredentialsFromContext(ctx), name) + d := fd.dentry() + if err := d.inode.removexattr(auth.CredentialsFromContext(ctx), name); err != nil { + return err + } + + // Generate inotify events. + d.InotifyWithParent(linux.IN_ATTRIB, 0) + return nil } // NewMemfd creates a new tmpfs regular file and file description that can back diff --git a/pkg/sentry/kernel/fd_table.go b/pkg/sentry/kernel/fd_table.go index ed40b5303..ef73e1169 100644 --- a/pkg/sentry/kernel/fd_table.go +++ b/pkg/sentry/kernel/fd_table.go @@ -152,7 +152,13 @@ func (f *FDTable) drop(file *fs.File) { // dropVFS2 drops the table reference. func (f *FDTable) dropVFS2(file *vfs.FileDescription) { // TODO(gvisor.dev/issue/1480): Release locks. - // TODO(gvisor.dev/issue/1479): Send inotify events. + + // Generate inotify events. + ev := uint32(linux.IN_CLOSE_NOWRITE) + if file.IsWritable() { + ev = linux.IN_CLOSE_WRITE + } + file.Dentry().InotifyWithParent(ev, 0) // Drop the table reference. file.DecRef() diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD index d56927ff5..9c8b44f64 100644 --- a/pkg/sentry/syscalls/linux/vfs2/BUILD +++ b/pkg/sentry/syscalls/linux/vfs2/BUILD @@ -12,6 +12,7 @@ go_library( "filesystem.go", "fscontext.go", "getdents.go", + "inotify.go", "ioctl.go", "memfd.go", "mmap.go", diff --git a/pkg/sentry/syscalls/linux/vfs2/inotify.go b/pkg/sentry/syscalls/linux/vfs2/inotify.go new file mode 100644 index 000000000..7d50b6a16 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/inotify.go @@ -0,0 +1,134 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +const allFlags = linux.IN_NONBLOCK | linux.IN_CLOEXEC + +// InotifyInit1 implements the inotify_init1() syscalls. +func InotifyInit1(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + flags := args[0].Int() + if flags&^allFlags != 0 { + return 0, nil, syserror.EINVAL + } + + ino, err := vfs.NewInotifyFD(t, t.Kernel().VFS(), uint32(flags)) + if err != nil { + return 0, nil, err + } + defer ino.DecRef() + + fd, err := t.NewFDFromVFS2(0, ino, kernel.FDFlags{ + CloseOnExec: flags&linux.IN_CLOEXEC != 0, + }) + + if err != nil { + return 0, nil, err + } + + return uintptr(fd), nil, nil +} + +// InotifyInit implements the inotify_init() syscalls. +func InotifyInit(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + args[0].Value = 0 + return InotifyInit1(t, args) +} + +// fdToInotify resolves an fd to an inotify object. If successful, the file will +// have an extra ref and the caller is responsible for releasing the ref. +func fdToInotify(t *kernel.Task, fd int32) (*vfs.Inotify, *vfs.FileDescription, error) { + f := t.GetFileVFS2(fd) + if f == nil { + // Invalid fd. + return nil, nil, syserror.EBADF + } + + ino, ok := f.Impl().(*vfs.Inotify) + if !ok { + // Not an inotify fd. + f.DecRef() + return nil, nil, syserror.EINVAL + } + + return ino, f, nil +} + +// InotifyAddWatch implements the inotify_add_watch() syscall. +func InotifyAddWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + addr := args[1].Pointer() + mask := args[2].Uint() + + // "EINVAL: The given event mask contains no valid events." + // -- inotify_add_watch(2) + if validBits := mask & linux.ALL_INOTIFY_BITS; validBits == 0 { + return 0, nil, syserror.EINVAL + } + + // "IN_DONT_FOLLOW: Don't dereference pathname if it is a symbolic link." + // -- inotify(7) + follow := followFinalSymlink + if mask&linux.IN_DONT_FOLLOW == 0 { + follow = nofollowFinalSymlink + } + + ino, f, err := fdToInotify(t, fd) + if err != nil { + return 0, nil, err + } + defer f.DecRef() + + path, err := copyInPath(t, addr) + if err != nil { + return 0, nil, err + } + if mask&linux.IN_ONLYDIR != 0 { + path.Dir = true + } + tpop, err := getTaskPathOperation(t, linux.AT_FDCWD, path, disallowEmptyPath, follow) + if err != nil { + return 0, nil, err + } + defer tpop.Release() + d, err := t.Kernel().VFS().GetDentryAt(t, t.Credentials(), &tpop.pop, &vfs.GetDentryOptions{}) + if err != nil { + return 0, nil, err + } + defer d.DecRef() + + fd = ino.AddWatch(d.Dentry(), mask) + return uintptr(fd), nil, err +} + +// InotifyRmWatch implements the inotify_rm_watch() syscall. +func InotifyRmWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + wd := args[1].Int() + + ino, f, err := fdToInotify(t, fd) + if err != nil { + return 0, nil, err + } + defer f.DecRef() + return 0, nil, ino.RmWatch(wd) +} diff --git a/pkg/sentry/syscalls/linux/vfs2/read_write.go b/pkg/sentry/syscalls/linux/vfs2/read_write.go index 3a7ef24f5..92b5631a3 100644 --- a/pkg/sentry/syscalls/linux/vfs2/read_write.go +++ b/pkg/sentry/syscalls/linux/vfs2/read_write.go @@ -93,11 +93,17 @@ func Readv(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall func read(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { n, err := file.Read(t, dst, opts) if err != syserror.ErrWouldBlock { + if n > 0 { + file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0) + } return n, err } allowBlock, deadline, hasDeadline := blockPolicy(t, file) if !allowBlock { + if n > 0 { + file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0) + } return n, err } @@ -128,6 +134,9 @@ func read(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, opt } file.EventUnregister(&w) + if total > 0 { + file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0) + } return total, err } @@ -248,11 +257,17 @@ func Preadv2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca func pread(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { n, err := file.PRead(t, dst, offset, opts) if err != syserror.ErrWouldBlock { + if n > 0 { + file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0) + } return n, err } allowBlock, deadline, hasDeadline := blockPolicy(t, file) if !allowBlock { + if n > 0 { + file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0) + } return n, err } @@ -283,6 +298,9 @@ func pread(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, of } file.EventUnregister(&w) + if total > 0 { + file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0) + } return total, err } @@ -345,11 +363,17 @@ func Writev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal func write(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { n, err := file.Write(t, src, opts) if err != syserror.ErrWouldBlock { + if n > 0 { + file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0) + } return n, err } allowBlock, deadline, hasDeadline := blockPolicy(t, file) if !allowBlock { + if n > 0 { + file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0) + } return n, err } @@ -380,6 +404,9 @@ func write(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, op } file.EventUnregister(&w) + if total > 0 { + file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0) + } return total, err } @@ -500,11 +527,17 @@ func Pwritev2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc func pwrite(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { n, err := file.PWrite(t, src, offset, opts) if err != syserror.ErrWouldBlock { + if n > 0 { + file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0) + } return n, err } allowBlock, deadline, hasDeadline := blockPolicy(t, file) if !allowBlock { + if n > 0 { + file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0) + } return n, err } @@ -535,6 +568,9 @@ func pwrite(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, o } file.EventUnregister(&w) + if total > 0 { + file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0) + } return total, err } diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2.go b/pkg/sentry/syscalls/linux/vfs2/vfs2.go index 083fdcf82..ef8358b8a 100644 --- a/pkg/sentry/syscalls/linux/vfs2/vfs2.go +++ b/pkg/sentry/syscalls/linux/vfs2/vfs2.go @@ -116,9 +116,9 @@ func Override() { s.Table[232] = syscalls.Supported("epoll_wait", EpollWait) s.Table[233] = syscalls.Supported("epoll_ctl", EpollCtl) s.Table[235] = syscalls.Supported("utimes", Utimes) - delete(s.Table, 253) // inotify_init - delete(s.Table, 254) // inotify_add_watch - delete(s.Table, 255) // inotify_rm_watch + s.Table[253] = syscalls.PartiallySupported("inotify_init", InotifyInit, "inotify events are only available inside the sandbox.", nil) + s.Table[254] = syscalls.PartiallySupported("inotify_add_watch", InotifyAddWatch, "inotify events are only available inside the sandbox.", nil) + s.Table[255] = syscalls.PartiallySupported("inotify_rm_watch", InotifyRmWatch, "inotify events are only available inside the sandbox.", nil) s.Table[257] = syscalls.Supported("openat", Openat) s.Table[258] = syscalls.Supported("mkdirat", Mkdirat) s.Table[259] = syscalls.Supported("mknodat", Mknodat) @@ -151,7 +151,7 @@ func Override() { s.Table[291] = syscalls.Supported("epoll_create1", EpollCreate1) s.Table[292] = syscalls.Supported("dup3", Dup3) s.Table[293] = syscalls.Supported("pipe2", Pipe2) - delete(s.Table, 294) // inotify_init1 + s.Table[294] = syscalls.PartiallySupported("inotify_init1", InotifyInit1, "inotify events are only available inside the sandbox.", nil) s.Table[295] = syscalls.Supported("preadv", Preadv) s.Table[296] = syscalls.Supported("pwritev", Pwritev) s.Table[299] = syscalls.Supported("recvmmsg", RecvMMsg) diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index 94d69c1cc..774cc66cc 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -15,6 +15,18 @@ go_template_instance( }, ) +go_template_instance( + name = "event_list", + out = "event_list.go", + package = "vfs", + prefix = "event", + template = "//pkg/ilist:generic_list", + types = { + "Element": "*Event", + "Linker": "*Event", + }, +) + go_library( name = "vfs", srcs = [ @@ -25,11 +37,13 @@ go_library( "device.go", "epoll.go", "epoll_interest_list.go", + "event_list.go", "file_description.go", "file_description_impl_util.go", "filesystem.go", "filesystem_impl_util.go", "filesystem_type.go", + "inotify.go", "mount.go", "mount_unsafe.go", "options.go", @@ -57,6 +71,7 @@ go_library( "//pkg/sentry/limits", "//pkg/sentry/memmap", "//pkg/sentry/socket/unix/transport", + "//pkg/sentry/uniqueid", "//pkg/sync", "//pkg/syserror", "//pkg/usermem", diff --git a/pkg/sentry/vfs/anonfs.go b/pkg/sentry/vfs/anonfs.go index caf770fd5..55a3d54cc 100644 --- a/pkg/sentry/vfs/anonfs.go +++ b/pkg/sentry/vfs/anonfs.go @@ -297,3 +297,15 @@ func (d *anonDentry) TryIncRef() bool { func (d *anonDentry) DecRef() { // no-op } + +// InotifyWithParent implements DentryImpl.InotifyWithParent. +// +// TODO(gvisor.dev/issue/1479): Implement inotify. +func (d *anonDentry) InotifyWithParent(events uint32, cookie uint32) {} + +// Watches implements DentryImpl.Watches. +// +// TODO(gvisor.dev/issue/1479): Implement inotify. +func (d *anonDentry) Watches() *Watches { + return nil +} diff --git a/pkg/sentry/vfs/dentry.go b/pkg/sentry/vfs/dentry.go index 8624dbd5d..d61b9e09b 100644 --- a/pkg/sentry/vfs/dentry.go +++ b/pkg/sentry/vfs/dentry.go @@ -103,6 +103,22 @@ type DentryImpl interface { // DecRef decrements the Dentry's reference count. DecRef() + + // InotifyWithParent notifies all watches on the targets represented by this + // dentry and its parent. The parent's watches are notified first, followed + // by this dentry's. + // + // InotifyWithParent automatically adds the IN_ISDIR flag for dentries + // representing directories. + // + // Note that the events may not actually propagate up to the user, depending + // on the event masks. + InotifyWithParent(events uint32, cookie uint32) + + // Watches returns the set of inotify watches for the file corresponding to + // the Dentry. Dentries that are hard links to the same underlying file + // share the same watches. + Watches() *Watches } // IncRef increments d's reference count. @@ -133,6 +149,17 @@ func (d *Dentry) isMounted() bool { return atomic.LoadUint32(&d.mounts) != 0 } +// InotifyWithParent notifies all watches on the inodes for this dentry and +// its parent of events. +func (d *Dentry) InotifyWithParent(events uint32, cookie uint32) { + d.impl.InotifyWithParent(events, cookie) +} + +// Watches returns the set of inotify watches associated with d. +func (d *Dentry) Watches() *Watches { + return d.impl.Watches() +} + // The following functions are exported so that filesystem implementations can // use them. The vfs package, and users of VFS, should not call these // functions. diff --git a/pkg/sentry/vfs/inotify.go b/pkg/sentry/vfs/inotify.go new file mode 100644 index 000000000..1d28ccb46 --- /dev/null +++ b/pkg/sentry/vfs/inotify.go @@ -0,0 +1,675 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs + +import ( + "bytes" + "fmt" + "sync/atomic" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/uniqueid" + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +// inotifyEventBaseSize is the base size of linux's struct inotify_event. This +// must be a power 2 for rounding below. +const inotifyEventBaseSize = 16 + +// Inotify represents an inotify instance created by inotify_init(2) or +// inotify_init1(2). Inotify implements FileDescriptionImpl. +// +// Lock ordering: +// Inotify.mu -> Watches.mu -> Inotify.evMu +// +// +stateify savable +type Inotify struct { + vfsfd FileDescription + FileDescriptionDefaultImpl + DentryMetadataFileDescriptionImpl + + // Unique identifier for this inotify instance. We don't just reuse the + // inotify fd because fds can be duped. These should not be exposed to the + // user, since we may aggressively reuse an id on S/R. + id uint64 + + // queue is used to notify interested parties when the inotify instance + // becomes readable or writable. + queue waiter.Queue `state:"nosave"` + + // evMu *only* protects the events list. We need a separate lock while + // queuing events: using mu may violate lock ordering, since at that point + // the calling goroutine may already hold Watches.mu. + evMu sync.Mutex `state:"nosave"` + + // A list of pending events for this inotify instance. Protected by evMu. + events eventList + + // A scratch buffer, used to serialize inotify events. Allocate this + // ahead of time for the sake of performance. Protected by evMu. + scratch []byte + + // mu protects the fields below. + mu sync.Mutex `state:"nosave"` + + // nextWatchMinusOne is used to allocate watch descriptors on this Inotify + // instance. Note that Linux starts numbering watch descriptors from 1. + nextWatchMinusOne int32 + + // Map from watch descriptors to watch objects. + watches map[int32]*Watch +} + +var _ FileDescriptionImpl = (*Inotify)(nil) + +// NewInotifyFD constructs a new Inotify instance. +func NewInotifyFD(ctx context.Context, vfsObj *VirtualFilesystem, flags uint32) (*FileDescription, error) { + // O_CLOEXEC affects file descriptors, so it must be handled outside of vfs. + flags &^= linux.O_CLOEXEC + if flags&^linux.O_NONBLOCK != 0 { + return nil, syserror.EINVAL + } + + id := uniqueid.GlobalFromContext(ctx) + vd := vfsObj.NewAnonVirtualDentry(fmt.Sprintf("[inotifyfd:%d]", id)) + defer vd.DecRef() + fd := &Inotify{ + id: id, + scratch: make([]byte, inotifyEventBaseSize), + watches: make(map[int32]*Watch), + } + if err := fd.vfsfd.Init(fd, flags, vd.Mount(), vd.Dentry(), &FileDescriptionOptions{ + UseDentryMetadata: true, + DenyPRead: true, + DenyPWrite: true, + }); err != nil { + return nil, err + } + return &fd.vfsfd, nil +} + +// Release implements FileDescriptionImpl.Release. Release removes all +// watches and frees all resources for an inotify instance. +func (i *Inotify) Release() { + // We need to hold i.mu to avoid a race with concurrent calls to + // Inotify.handleDeletion from Watches. There's no risk of Watches + // accessing this Inotify after the destructor ends, because we remove all + // references to it below. + i.mu.Lock() + defer i.mu.Unlock() + for _, w := range i.watches { + // Remove references to the watch from the watches set on the target. We + // don't need to worry about the references from i.watches, since this + // file description is about to be destroyed. + w.set.Remove(i.id) + } +} + +// EventRegister implements waiter.Waitable. +func (i *Inotify) EventRegister(e *waiter.Entry, mask waiter.EventMask) { + i.queue.EventRegister(e, mask) +} + +// EventUnregister implements waiter.Waitable. +func (i *Inotify) EventUnregister(e *waiter.Entry) { + i.queue.EventUnregister(e) +} + +// Readiness implements waiter.Waitable.Readiness. +// +// Readiness indicates whether there are pending events for an inotify instance. +func (i *Inotify) Readiness(mask waiter.EventMask) waiter.EventMask { + ready := waiter.EventMask(0) + + i.evMu.Lock() + defer i.evMu.Unlock() + + if !i.events.Empty() { + ready |= waiter.EventIn + } + + return mask & ready +} + +// PRead implements FileDescriptionImpl. +func (*Inotify) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts ReadOptions) (int64, error) { + return 0, syserror.ESPIPE +} + +// PWrite implements FileDescriptionImpl. +func (*Inotify) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts WriteOptions) (int64, error) { + return 0, syserror.ESPIPE +} + +// Write implements FileDescriptionImpl.Write. +func (*Inotify) Write(ctx context.Context, src usermem.IOSequence, opts WriteOptions) (int64, error) { + return 0, syserror.EBADF +} + +// Read implements FileDescriptionImpl.Read. +func (i *Inotify) Read(ctx context.Context, dst usermem.IOSequence, opts ReadOptions) (int64, error) { + if dst.NumBytes() < inotifyEventBaseSize { + return 0, syserror.EINVAL + } + + i.evMu.Lock() + defer i.evMu.Unlock() + + if i.events.Empty() { + // Nothing to read yet, tell caller to block. + return 0, syserror.ErrWouldBlock + } + + var writeLen int64 + for it := i.events.Front(); it != nil; { + // Advance `it` before the element is removed from the list, or else + // it.Next() will always be nil. + event := it + it = it.Next() + + // Does the buffer have enough remaining space to hold the event we're + // about to write out? + if dst.NumBytes() < int64(event.sizeOf()) { + if writeLen > 0 { + // Buffer wasn't big enough for all pending events, but we did + // write some events out. + return writeLen, nil + } + return 0, syserror.EINVAL + } + + // Linux always dequeues an available event as long as there's enough + // buffer space to copy it out, even if the copy below fails. Emulate + // this behaviour. + i.events.Remove(event) + + // Buffer has enough space, copy event to the read buffer. + n, err := event.CopyTo(ctx, i.scratch, dst) + if err != nil { + return 0, err + } + + writeLen += n + dst = dst.DropFirst64(n) + } + return writeLen, nil +} + +// Ioctl implements fs.FileOperations.Ioctl. +func (i *Inotify) Ioctl(ctx context.Context, uio usermem.IO, args arch.SyscallArguments) (uintptr, error) { + switch args[1].Int() { + case linux.FIONREAD: + i.evMu.Lock() + defer i.evMu.Unlock() + var n uint32 + for e := i.events.Front(); e != nil; e = e.Next() { + n += uint32(e.sizeOf()) + } + var buf [4]byte + usermem.ByteOrder.PutUint32(buf[:], n) + _, err := uio.CopyOut(ctx, args[2].Pointer(), buf[:], usermem.IOOpts{}) + return 0, err + + default: + return 0, syserror.ENOTTY + } +} + +func (i *Inotify) queueEvent(ev *Event) { + i.evMu.Lock() + + // Check if we should coalesce the event we're about to queue with the last + // one currently in the queue. Events are coalesced if they are identical. + if last := i.events.Back(); last != nil { + if ev.equals(last) { + // "Coalesce" the two events by simply not queuing the new one. We + // don't need to raise a waiter.EventIn notification because no new + // data is available for reading. + i.evMu.Unlock() + return + } + } + + i.events.PushBack(ev) + + // Release mutex before notifying waiters because we don't control what they + // can do. + i.evMu.Unlock() + + i.queue.Notify(waiter.EventIn) +} + +// newWatchLocked creates and adds a new watch to target. +// +// Precondition: i.mu must be locked. +func (i *Inotify) newWatchLocked(target *Dentry, mask uint32) *Watch { + targetWatches := target.Watches() + w := &Watch{ + owner: i, + wd: i.nextWatchIDLocked(), + set: targetWatches, + mask: mask, + } + + // Hold the watch in this inotify instance as well as the watch set on the + // target. + i.watches[w.wd] = w + targetWatches.Add(w) + return w +} + +// newWatchIDLocked allocates and returns a new watch descriptor. +// +// Precondition: i.mu must be locked. +func (i *Inotify) nextWatchIDLocked() int32 { + i.nextWatchMinusOne++ + return i.nextWatchMinusOne +} + +// handleDeletion handles the deletion of the target of watch w. It removes w +// from i.watches and a watch removal event is generated. +func (i *Inotify) handleDeletion(w *Watch) { + i.mu.Lock() + _, found := i.watches[w.wd] + delete(i.watches, w.wd) + i.mu.Unlock() + + if found { + i.queueEvent(newEvent(w.wd, "", linux.IN_IGNORED, 0)) + } +} + +// AddWatch constructs a new inotify watch and adds it to the target. It +// returns the watch descriptor returned by inotify_add_watch(2). +func (i *Inotify) AddWatch(target *Dentry, mask uint32) int32 { + // Note: Locking this inotify instance protects the result returned by + // Lookup() below. With the lock held, we know for sure the lookup result + // won't become stale because it's impossible for *this* instance to + // add/remove watches on target. + i.mu.Lock() + defer i.mu.Unlock() + + // Does the target already have a watch from this inotify instance? + if existing := target.Watches().Lookup(i.id); existing != nil { + newmask := mask + if mask&linux.IN_MASK_ADD != 0 { + // "Add (OR) events to watch mask for this pathname if it already + // exists (instead of replacing mask)." -- inotify(7) + newmask |= atomic.LoadUint32(&existing.mask) + } + atomic.StoreUint32(&existing.mask, newmask) + return existing.wd + } + + // No existing watch, create a new watch. + w := i.newWatchLocked(target, mask) + return w.wd +} + +// RmWatch looks up an inotify watch for the given 'wd' and configures the +// target to stop sending events to this inotify instance. +func (i *Inotify) RmWatch(wd int32) error { + i.mu.Lock() + + // Find the watch we were asked to removed. + w, ok := i.watches[wd] + if !ok { + i.mu.Unlock() + return syserror.EINVAL + } + + // Remove the watch from this instance. + delete(i.watches, wd) + + // Remove the watch from the watch target. + w.set.Remove(w.OwnerID()) + i.mu.Unlock() + + // Generate the event for the removal. + i.queueEvent(newEvent(wd, "", linux.IN_IGNORED, 0)) + + return nil +} + +// Watches is the collection of all inotify watches on a single file. +// +// +stateify savable +type Watches struct { + // mu protects the fields below. + mu sync.RWMutex `state:"nosave"` + + // ws is the map of active watches in this collection, keyed by the inotify + // instance id of the owner. + ws map[uint64]*Watch +} + +// Lookup returns the watch owned by an inotify instance with the given id. +// Returns nil if no such watch exists. +// +// Precondition: the inotify instance with the given id must be locked to +// prevent the returned watch from being concurrently modified or replaced in +// Inotify.watches. +func (w *Watches) Lookup(id uint64) *Watch { + w.mu.Lock() + defer w.mu.Unlock() + return w.ws[id] +} + +// Add adds watch into this set of watches. +// +// Precondition: the inotify instance with the given id must be locked. +func (w *Watches) Add(watch *Watch) { + w.mu.Lock() + defer w.mu.Unlock() + + owner := watch.OwnerID() + // Sanity check, we should never have two watches for one owner on the + // same target. + if _, exists := w.ws[owner]; exists { + panic(fmt.Sprintf("Watch collision with ID %+v", owner)) + } + if w.ws == nil { + w.ws = make(map[uint64]*Watch) + } + w.ws[owner] = watch +} + +// Remove removes a watch with the given id from this set of watches and +// releases it. The caller is responsible for generating any watch removal +// event, as appropriate. The provided id must match an existing watch in this +// collection. +// +// Precondition: the inotify instance with the given id must be locked. +func (w *Watches) Remove(id uint64) { + w.mu.Lock() + defer w.mu.Unlock() + + if w.ws == nil { + // This watch set is being destroyed. The thread executing the + // destructor is already in the process of deleting all our watches. We + // got here with no references on the target because we raced with the + // destructor notifying all the watch owners of destruction. See the + // comment in Watches.HandleDeletion for why this race exists. + return + } + + if _, ok := w.ws[id]; !ok { + // While there's technically no problem with silently ignoring a missing + // watch, this is almost certainly a bug. + panic(fmt.Sprintf("Attempt to remove a watch, but no watch found with provided id %+v.", id)) + } + delete(w.ws, id) +} + +// Notify queues a new event with all watches in this set. +func (w *Watches) Notify(name string, events, cookie uint32) { + // N.B. We don't defer the unlocks because Notify is in the hot path of + // all IO operations, and the defer costs too much for small IO + // operations. + w.mu.RLock() + for _, watch := range w.ws { + // TODO(gvisor.dev/issue/1479): Skip for IN_EXCL_UNLINK cases. + watch.Notify(name, events, cookie) + } + w.mu.RUnlock() +} + +// HandleDeletion is called when the watch target is destroyed to emit +// the appropriate events. +func (w *Watches) HandleDeletion() { + w.Notify("", linux.IN_DELETE_SELF, 0) + + // TODO(gvisor.dev/issue/1479): This doesn't work because maps are not copied + // by value. Ideally, we wouldn't have this circular locking so we can just + // notify of IN_DELETE_SELF in the same loop below. + // + // We can't hold w.mu while calling watch.handleDeletion to preserve lock + // ordering w.r.t to the owner inotify instances. Instead, atomically move + // the watches map into a local variable so we can iterate over it safely. + // + // Because of this however, it is possible for the watches' owners to reach + // this inode while the inode has no refs. This is still safe because the + // owners can only reach the inode until this function finishes calling + // watch.handleDeletion below and the inode is guaranteed to exist in the + // meantime. But we still have to be very careful not to rely on inode state + // that may have been already destroyed. + var ws map[uint64]*Watch + w.mu.Lock() + ws = w.ws + w.ws = nil + w.mu.Unlock() + + for _, watch := range ws { + // TODO(gvisor.dev/issue/1479): consider refactoring this. + watch.handleDeletion() + } +} + +// Watch represent a particular inotify watch created by inotify_add_watch. +// +// +stateify savable +type Watch struct { + // Inotify instance which owns this watch. + owner *Inotify + + // Descriptor for this watch. This is unique across an inotify instance. + wd int32 + + // set is the watch set containing this watch. It belongs to the target file + // of this watch. + set *Watches + + // Events being monitored via this watch. Must be accessed with atomic + // memory operations. + mask uint32 +} + +// OwnerID returns the id of the inotify instance that owns this watch. +func (w *Watch) OwnerID() uint64 { + return w.owner.id +} + +// ExcludeUnlinkedChildren indicates whether the watched object should continue +// to be notified of events of its children after they have been unlinked, e.g. +// for an open file descriptor. +// +// TODO(gvisor.dev/issue/1479): Implement IN_EXCL_UNLINK. +// We can do this by keeping track of the set of unlinked children in Watches +// to skip notification. +func (w *Watch) ExcludeUnlinkedChildren() bool { + return atomic.LoadUint32(&w.mask)&linux.IN_EXCL_UNLINK != 0 +} + +// Notify queues a new event on this watch. +func (w *Watch) Notify(name string, events uint32, cookie uint32) { + mask := atomic.LoadUint32(&w.mask) + if mask&events == 0 { + // We weren't watching for this event. + return + } + + // Event mask should include bits matched from the watch plus all control + // event bits. + unmaskableBits := ^uint32(0) &^ linux.IN_ALL_EVENTS + effectiveMask := unmaskableBits | mask + matchedEvents := effectiveMask & events + w.owner.queueEvent(newEvent(w.wd, name, matchedEvents, cookie)) +} + +// handleDeletion handles the deletion of w's target. +func (w *Watch) handleDeletion() { + w.owner.handleDeletion(w) +} + +// Event represents a struct inotify_event from linux. +// +// +stateify savable +type Event struct { + eventEntry + + wd int32 + mask uint32 + cookie uint32 + + // len is computed based on the name field is set automatically by + // Event.setName. It should be 0 when no name is set; otherwise it is the + // length of the name slice. + len uint32 + + // The name field has special padding requirements and should only be set by + // calling Event.setName. + name []byte +} + +func newEvent(wd int32, name string, events, cookie uint32) *Event { + e := &Event{ + wd: wd, + mask: events, + cookie: cookie, + } + if name != "" { + e.setName(name) + } + return e +} + +// paddedBytes converts a go string to a null-terminated c-string, padded with +// null bytes to a total size of 'l'. 'l' must be large enough for all the bytes +// in the 's' plus at least one null byte. +func paddedBytes(s string, l uint32) []byte { + if l < uint32(len(s)+1) { + panic("Converting string to byte array results in truncation, this can lead to buffer-overflow due to the missing null-byte!") + } + b := make([]byte, l) + copy(b, s) + + // b was zero-value initialized during make(), so the rest of the slice is + // already filled with null bytes. + + return b +} + +// setName sets the optional name for this event. +func (e *Event) setName(name string) { + // We need to pad the name such that the entire event length ends up a + // multiple of inotifyEventBaseSize. + unpaddedLen := len(name) + 1 + // Round up to nearest multiple of inotifyEventBaseSize. + e.len = uint32((unpaddedLen + inotifyEventBaseSize - 1) & ^(inotifyEventBaseSize - 1)) + // Make sure we haven't overflowed and wrapped around when rounding. + if unpaddedLen > int(e.len) { + panic("Overflow when rounding inotify event size, the 'name' field was too big.") + } + e.name = paddedBytes(name, e.len) +} + +func (e *Event) sizeOf() int { + s := inotifyEventBaseSize + int(e.len) + if s < inotifyEventBaseSize { + panic("overflow") + } + return s +} + +// CopyTo serializes this event to dst. buf is used as a scratch buffer to +// construct the output. We use a buffer allocated ahead of time for +// performance. buf must be at least inotifyEventBaseSize bytes. +func (e *Event) CopyTo(ctx context.Context, buf []byte, dst usermem.IOSequence) (int64, error) { + usermem.ByteOrder.PutUint32(buf[0:], uint32(e.wd)) + usermem.ByteOrder.PutUint32(buf[4:], e.mask) + usermem.ByteOrder.PutUint32(buf[8:], e.cookie) + usermem.ByteOrder.PutUint32(buf[12:], e.len) + + writeLen := 0 + + n, err := dst.CopyOut(ctx, buf) + if err != nil { + return 0, err + } + writeLen += n + dst = dst.DropFirst(n) + + if e.len > 0 { + n, err = dst.CopyOut(ctx, e.name) + if err != nil { + return 0, err + } + writeLen += n + } + + // Santiy check. + if writeLen != e.sizeOf() { + panic(fmt.Sprintf("Serialized unexpected amount of data for an event, expected %d, wrote %d.", e.sizeOf(), writeLen)) + } + + return int64(writeLen), nil +} + +func (e *Event) equals(other *Event) bool { + return e.wd == other.wd && + e.mask == other.mask && + e.cookie == other.cookie && + e.len == other.len && + bytes.Equal(e.name, other.name) +} + +// InotifyEventFromStatMask generates the appropriate events for an operation +// that set the stats specified in mask. +func InotifyEventFromStatMask(mask uint32) uint32 { + var ev uint32 + if mask&(linux.STATX_UID|linux.STATX_GID|linux.STATX_MODE) != 0 { + ev |= linux.IN_ATTRIB + } + if mask&linux.STATX_SIZE != 0 { + ev |= linux.IN_MODIFY + } + + if (mask & (linux.STATX_ATIME | linux.STATX_MTIME)) == (linux.STATX_ATIME | linux.STATX_MTIME) { + // Both times indicates a utime(s) call. + ev |= linux.IN_ATTRIB + } else if mask&linux.STATX_ATIME != 0 { + ev |= linux.IN_ACCESS + } else if mask&linux.STATX_MTIME != 0 { + mask |= linux.IN_MODIFY + } + return ev +} + +// InotifyRemoveChild sends the appriopriate notifications to the watch sets of +// the child being removed and its parent. +func InotifyRemoveChild(self, parent *Watches, name string) { + self.Notify("", linux.IN_ATTRIB, 0) + parent.Notify(name, linux.IN_DELETE, 0) + // TODO(gvisor.dev/issue/1479): implement IN_EXCL_UNLINK. +} + +// InotifyRename sends the appriopriate notifications to the watch sets of the +// file being renamed and its old/new parents. +func InotifyRename(ctx context.Context, renamed, oldParent, newParent *Watches, oldName, newName string, isDir bool) { + var dirEv uint32 + if isDir { + dirEv = linux.IN_ISDIR + } + cookie := uniqueid.InotifyCookie(ctx) + oldParent.Notify(oldName, dirEv|linux.IN_MOVED_FROM, cookie) + newParent.Notify(newName, dirEv|linux.IN_MOVED_TO, cookie) + // Somewhat surprisingly, self move events do not have a cookie. + renamed.Notify("", linux.IN_MOVE_SELF, 0) +} diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 6d2ba53ea..be6f21dba 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -422,6 +422,7 @@ func (vfs *VirtualFilesystem) OpenAt(ctx context.Context, creds *auth.Credential } } + fd.Dentry().InotifyWithParent(linux.IN_OPEN, 0) return fd, nil } if !rp.handleError(err) { diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 5acdb8438..f4b5de18d 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -951,6 +951,7 @@ cc_binary( "//test/util:epoll_util", "//test/util:file_descriptor", "//test/util:fs_util", + "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", @@ -1382,7 +1383,7 @@ cc_binary( srcs = ["partial_bad_buffer.cc"], linkstatic = 1, deps = [ - "//test/syscalls/linux:socket_test_util", + ":socket_test_util", "//test/util:file_descriptor", "//test/util:fs_util", "@com_google_absl//absl/time", @@ -3461,7 +3462,7 @@ cc_binary( deps = [ ":socket_test_util", gtest, - "//test/syscalls/linux:socket_netlink_route_util", + ":socket_netlink_route_util", "//test/util:capability_util", "//test/util:file_descriptor", "//test/util:fs_util", diff --git a/test/syscalls/linux/inotify.cc b/test/syscalls/linux/inotify.cc index 0e13ad190..e4565467b 100644 --- a/test/syscalls/linux/inotify.cc +++ b/test/syscalls/linux/inotify.cc @@ -33,6 +33,7 @@ #include "test/util/epoll_util.h" #include "test/util/file_descriptor.h" #include "test/util/fs_util.h" +#include "test/util/posix_error.h" #include "test/util/temp_path.h" #include "test/util/test_util.h" #include "test/util/thread_util.h" @@ -335,6 +336,11 @@ TEST(Inotify, InotifyFdNotWritable) { EXPECT_THAT(write(fd.get(), "x", 1), SyscallFailsWithErrno(EBADF)); } +TEST(Inotify, InitFlags) { + EXPECT_THAT(inotify_init1(IN_NONBLOCK | IN_CLOEXEC), SyscallSucceeds()); + EXPECT_THAT(inotify_init1(12345), SyscallFailsWithErrno(EINVAL)); +} + TEST(Inotify, NonBlockingReadReturnsEagain) { const FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); @@ -395,7 +401,7 @@ TEST(Inotify, CanDeleteFileAfterRemovingWatch) { file1.reset(); } -TEST(Inotify, CanRemoveWatchAfterDeletingFile) { +TEST(Inotify, RemoveWatchAfterDeletingFileFails) { const TempPath root = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); TempPath file1 = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(root.path())); @@ -491,17 +497,23 @@ TEST(Inotify, DeletingChildGeneratesEvents) { Event(IN_DELETE, root_wd, Basename(file1_path))})); } +// Creating a file in "parent/child" should generate events for child, but not +// parent. TEST(Inotify, CreatingFileGeneratesEvents) { - const TempPath root = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const TempPath parent = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const TempPath child = + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDirIn(parent.path())); const FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(fd.get(), parent.path(), IN_ALL_EVENTS)); const int wd = ASSERT_NO_ERRNO_AND_VALUE( - InotifyAddWatch(fd.get(), root.path(), IN_ALL_EVENTS)); + InotifyAddWatch(fd.get(), child.path(), IN_ALL_EVENTS)); // Create a new file in the directory. const TempPath file1 = - ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(root.path())); + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(child.path())); const std::vector events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(fd.get())); @@ -554,6 +566,47 @@ TEST(Inotify, WritingFileGeneratesModifyEvent) { ASSERT_THAT(events, Are({Event(IN_MODIFY, wd, Basename(file1.path()))})); } +TEST(Inotify, SizeZeroReadWriteGeneratesNothing) { + const TempPath root = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + const TempPath file1 = + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(root.path())); + + const FileDescriptor file1_fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(file1.path(), O_RDWR)); + ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(fd.get(), root.path(), IN_ALL_EVENTS)); + + // Read from the empty file. + int val; + ASSERT_THAT(read(file1_fd.get(), &val, sizeof(val)), + SyscallSucceedsWithValue(0)); + + // Write zero bytes. + ASSERT_THAT(write(file1_fd.get(), "", 0), SyscallSucceedsWithValue(0)); + + const std::vector events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(fd.get())); + ASSERT_THAT(events, Are({})); +} + +TEST(Inotify, FailedFileCreationGeneratesNoEvents) { + const TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(fd.get(), dir.path(), IN_ALL_EVENTS)); + + const char* p = dir.path().c_str(); + ASSERT_THAT(mkdir(p, 0777), SyscallFails()); + ASSERT_THAT(mknod(p, S_IFIFO, 0777), SyscallFails()); + ASSERT_THAT(symlink(p, p), SyscallFails()); + ASSERT_THAT(link(p, p), SyscallFails()); + std::vector events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(fd.get())); + ASSERT_THAT(events, Are({})); +} + TEST(Inotify, WatchSetAfterOpenReportsCloseFdEvent) { const TempPath root = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); const FileDescriptor fd = @@ -602,7 +655,7 @@ TEST(Inotify, ChildrenDeletionInWatchedDirGeneratesEvent) { Event(IN_DELETE | IN_ISDIR, wd, Basename(dir1_path))})); } -TEST(Inotify, WatchTargetDeletionGeneratesEvent) { +TEST(Inotify, RmdirOnWatchedTargetGeneratesEvent) { const TempPath root = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); const FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); @@ -1228,7 +1281,7 @@ TEST(Inotify, LinkGeneratesAttribAndCreateEvents) { InotifyAddWatch(fd.get(), file1.path(), IN_ALL_EVENTS)); const int rc = link(file1.path().c_str(), link1.path().c_str()); - // link(2) is only supported on tmpfs in the sandbox. + // NOTE(b/34861058): link(2) is only supported on tmpfs in the sandbox. SKIP_IF(IsRunningOnGvisor() && rc != 0 && (errno == EPERM || errno == ENOENT)); ASSERT_THAT(rc, SyscallSucceeds()); @@ -1322,21 +1375,27 @@ TEST(Inotify, HardlinksReuseSameWatch) { Event(IN_DELETE, root_wd, Basename(file1_path))})); } +// Calling mkdir within "parent/child" should generate an event for child, but +// not parent. TEST(Inotify, MkdirGeneratesCreateEventWithDirFlag) { - const TempPath root = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const TempPath parent = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const TempPath child = + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDirIn(parent.path())); const FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); - const int root_wd = ASSERT_NO_ERRNO_AND_VALUE( - InotifyAddWatch(fd.get(), root.path(), IN_ALL_EVENTS)); + ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(fd.get(), parent.path(), IN_ALL_EVENTS)); + const int child_wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(fd.get(), child.path(), IN_ALL_EVENTS)); - const TempPath dir1(NewTempAbsPathInDir(root.path())); + const TempPath dir1(NewTempAbsPathInDir(child.path())); ASSERT_THAT(mkdir(dir1.path().c_str(), 0777), SyscallSucceeds()); const std::vector events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(fd.get())); ASSERT_THAT( events, - Are({Event(IN_CREATE | IN_ISDIR, root_wd, Basename(dir1.path()))})); + Are({Event(IN_CREATE | IN_ISDIR, child_wd, Basename(dir1.path()))})); } TEST(Inotify, MultipleInotifyInstancesAndWatchesAllGetEvents) { @@ -1597,6 +1656,8 @@ TEST(Inotify, EpollNoDeadlock) { } TEST(Inotify, SpliceEvent) { + // TODO(gvisor.dev/issue/138): Implement splice in VFS2. + SKIP_IF(IsRunningOnGvisor() && !IsRunningWithVFS1()); int pipes[2]; ASSERT_THAT(pipe2(pipes, O_NONBLOCK), SyscallSucceeds()); @@ -1624,6 +1685,99 @@ TEST(Inotify, SpliceEvent) { ASSERT_THAT(events, Are({Event(IN_ACCESS, watcher)})); } +// Watches on a parent should not be triggered by actions on a hard link to one +// of its children that has a different parent. +TEST(Inotify, LinkOnOtherParent) { + const TempPath dir1 = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const TempPath dir2 = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const TempPath file = + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(dir1.path())); + std::string link_path = NewTempAbsPathInDir(dir2.path()); + + const int rc = link(file.path().c_str(), link_path.c_str()); + // NOTE(b/34861058): link(2) is only supported on tmpfs in the sandbox. + SKIP_IF(IsRunningOnGvisor() && rc != 0 && + (errno == EPERM || errno == ENOENT)); + ASSERT_THAT(rc, SyscallSucceeds()); + + const FileDescriptor inotify_fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(inotify_fd.get(), dir1.path(), IN_ALL_EVENTS)); + + // Perform various actions on the link outside of dir1, which should trigger + // no inotify events. + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(link_path.c_str(), O_RDWR)); + int val = 0; + ASSERT_THAT(write(fd.get(), &val, sizeof(val)), SyscallSucceeds()); + ASSERT_THAT(read(fd.get(), &val, sizeof(val)), SyscallSucceeds()); + ASSERT_THAT(ftruncate(fd.get(), 12345), SyscallSucceeds()); + ASSERT_THAT(unlink(link_path.c_str()), SyscallSucceeds()); + const std::vector events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({})); +} + +TEST(Inotify, Exec) { + const TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const TempPath bin = ASSERT_NO_ERRNO_AND_VALUE( + TempPath::CreateSymlinkTo(dir.path(), "/bin/true")); + + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + const int wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(fd.get(), bin.path(), IN_ALL_EVENTS)); + + // Perform exec. + ScopedThread t([&bin]() { + ASSERT_THAT(execl(bin.path().c_str(), bin.path().c_str(), (char*)nullptr), + SyscallSucceeds()); + }); + t.Join(); + + std::vector events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(fd.get())); + EXPECT_THAT(events, Are({Event(IN_OPEN, wd), Event(IN_ACCESS, wd)})); +} + +// Watches without IN_EXCL_UNLINK, should continue to emit events for file +// descriptors after their corresponding files have been unlinked. +// +// We need to disable S/R because there are filesystems where we cannot re-open +// fds to an unlinked file across S/R, e.g. gofer-backed filesytems. +TEST(Inotify, IncludeUnlinkedFile_NoRandomSave) { + const DisableSave ds; + + const TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const TempPath file = ASSERT_NO_ERRNO_AND_VALUE( + TempPath::CreateFileWith(dir.path(), "123", TempPath::kDefaultFileMode)); + + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(file.path(), O_RDWR)); + + const FileDescriptor inotify_fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + const int dir_wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(inotify_fd.get(), dir.path(), IN_ALL_EVENTS)); + const int file_wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(inotify_fd.get(), file.path(), IN_ALL_EVENTS)); + + ASSERT_THAT(unlink(file.path().c_str()), SyscallSucceeds()); + int val = 0; + ASSERT_THAT(read(fd.get(), &val, sizeof(val)), SyscallSucceeds()); + ASSERT_THAT(write(fd.get(), &val, sizeof(val)), SyscallSucceeds()); + const std::vector events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({ + Event(IN_ATTRIB, file_wd), + Event(IN_DELETE, dir_wd, Basename(file.path())), + Event(IN_ACCESS, dir_wd, Basename(file.path())), + Event(IN_ACCESS, file_wd), + Event(IN_MODIFY, dir_wd, Basename(file.path())), + Event(IN_MODIFY, file_wd), + })); +} + } // namespace } // namespace testing } // namespace gvisor -- cgit v1.2.3 From 9ada8c972e43e0c3e144b432fe57d95f823a1847 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 29 May 2020 11:27:00 -0700 Subject: Fix the smallest of typos. PiperOrigin-RevId: 313817646 --- pkg/sentry/fsimpl/kernfs/kernfs.go | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index 682545994..3f1220791 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -239,7 +239,7 @@ func (d *Dentry) Watches() *vfs.Watches { // InsertChild inserts child into the vfs dentry cache with the given name under // this dentry. This does not update the directory inode, so calling this on -// it's own isn't sufficient to insert a child into a directory. InsertChild +// its own isn't sufficient to insert a child into a directory. InsertChild // updates the link count on d if required. // // Precondition: d must represent a directory inode. -- cgit v1.2.3 From ccf69bdd7e05a4e5f404fbef89a7f49f218645e2 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 29 May 2020 12:27:15 -0700 Subject: Implement IN_EXCL_UNLINK inotify option in vfs2. Limited to tmpfs. Inotify support in other filesystem implementations to follow. Updates #1479 PiperOrigin-RevId: 313828648 --- pkg/sentry/fsimpl/ext/dentry.go | 2 +- pkg/sentry/fsimpl/gofer/gofer.go | 2 +- pkg/sentry/fsimpl/kernfs/kernfs.go | 2 +- pkg/sentry/fsimpl/tmpfs/directory.go | 3 +- pkg/sentry/fsimpl/tmpfs/filesystem.go | 14 +-- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 18 +-- pkg/sentry/kernel/fd_table.go | 2 +- pkg/sentry/syscalls/linux/vfs2/read_write.go | 24 ++-- pkg/sentry/vfs/anonfs.go | 2 +- pkg/sentry/vfs/dentry.go | 6 +- pkg/sentry/vfs/inotify.go | 38 +++++-- pkg/sentry/vfs/vfs.go | 2 +- test/syscalls/linux/inotify.cc | 164 +++++++++++++++++++++++++++ 13 files changed, 235 insertions(+), 44 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/ext/dentry.go b/pkg/sentry/fsimpl/ext/dentry.go index 4d0deaf03..6bd1a9fc6 100644 --- a/pkg/sentry/fsimpl/ext/dentry.go +++ b/pkg/sentry/fsimpl/ext/dentry.go @@ -64,7 +64,7 @@ func (d *dentry) DecRef() { // InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. // // TODO(gvisor.dev/issue/1479): Implement inotify. -func (d *dentry) InotifyWithParent(events uint32, cookie uint32) {} +func (d *dentry) InotifyWithParent(events uint32, cookie uint32, et vfs.EventType) {} // Watches implements vfs.DentryImpl.Watches. // diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 850482a19..3f3bd56f0 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -1042,7 +1042,7 @@ func (d *dentry) decRefLocked() { // InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. // // TODO(gvisor.dev/issue/1479): Implement inotify. -func (d *dentry) InotifyWithParent(events uint32, cookie uint32) {} +func (d *dentry) InotifyWithParent(events uint32, cookie uint32, et vfs.EventType) {} // Watches implements vfs.DentryImpl.Watches. // diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index 3f1220791..bbee8ccda 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -228,7 +228,7 @@ func (d *Dentry) destroy() { // InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. // // TODO(gvisor.dev/issue/1479): Implement inotify. -func (d *Dentry) InotifyWithParent(events uint32, cookie uint32) {} +func (d *Dentry) InotifyWithParent(events uint32, cookie uint32, et vfs.EventType) {} // Watches implements vfs.DentryImpl.Watches. // diff --git a/pkg/sentry/fsimpl/tmpfs/directory.go b/pkg/sentry/fsimpl/tmpfs/directory.go index 8bc475f88..70387cb9c 100644 --- a/pkg/sentry/fsimpl/tmpfs/directory.go +++ b/pkg/sentry/fsimpl/tmpfs/directory.go @@ -79,6 +79,7 @@ func (dir *directory) removeChildLocked(child *dentry) { dir.iterMu.Lock() dir.childList.Remove(child) dir.iterMu.Unlock() + child.unlinked = true } type directoryFD struct { @@ -112,7 +113,7 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba dir.iterMu.Lock() defer dir.iterMu.Unlock() - fd.dentry().InotifyWithParent(linux.IN_ACCESS, 0) + fd.dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) fd.inode().touchAtime(fd.vfsfd.Mount()) if fd.off == 0 { diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index b4159f904..183eb975c 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -182,7 +182,7 @@ func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(pa if dir { ev |= linux.IN_ISDIR } - parentDir.inode.watches.Notify(name, uint32(ev), 0) + parentDir.inode.watches.Notify(name, uint32(ev), 0, vfs.InodeEvent) parentDir.inode.touchCMtime() return nil } @@ -247,7 +247,7 @@ func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. return syserror.EMLINK } d.inode.incLinksLocked() - d.inode.watches.Notify("", linux.IN_ATTRIB, 0) + d.inode.watches.Notify("", linux.IN_ATTRIB, 0, vfs.InodeEvent) parentDir.insertChildLocked(fs.newDentry(d.inode), name) return nil }) @@ -361,7 +361,7 @@ afterTrailingSymlink: if err != nil { return nil, err } - parentDir.inode.watches.Notify(name, linux.IN_CREATE, 0) + parentDir.inode.watches.Notify(name, linux.IN_CREATE, 0, vfs.PathEvent) parentDir.inode.touchCMtime() return fd, nil } @@ -613,7 +613,7 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error return err } parentDir.removeChildLocked(child) - parentDir.inode.watches.Notify(name, linux.IN_DELETE|linux.IN_ISDIR, 0) + parentDir.inode.watches.Notify(name, linux.IN_DELETE|linux.IN_ISDIR, 0, vfs.InodeEvent) // Remove links for child, child/., and child/.. child.inode.decLinksLocked() child.inode.decLinksLocked() @@ -636,7 +636,7 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts } if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { - d.InotifyWithParent(ev, 0) + d.InotifyWithParent(ev, 0, vfs.InodeEvent) } return nil } @@ -784,7 +784,7 @@ func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt return err } - d.InotifyWithParent(linux.IN_ATTRIB, 0) + d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil } @@ -800,7 +800,7 @@ func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, return err } - d.InotifyWithParent(linux.IN_ATTRIB, 0) + d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil } diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 1d83b6840..f0e098702 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -163,6 +163,11 @@ type dentry struct { // filesystem.mu. name string + // unlinked indicates whether this dentry has been unlinked from its parent. + // It is only set to true on an unlink operation, and never set from true to + // false. unlinked is protected by filesystem.mu. + unlinked bool + // dentryEntry (ugh) links dentries into their parent directory.childList. dentryEntry @@ -202,7 +207,7 @@ func (d *dentry) DecRef() { } // InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. -func (d *dentry) InotifyWithParent(events uint32, cookie uint32) { +func (d *dentry) InotifyWithParent(events uint32, cookie uint32, et vfs.EventType) { if d.inode.isDir() { events |= linux.IN_ISDIR } @@ -211,9 +216,9 @@ func (d *dentry) InotifyWithParent(events uint32, cookie uint32) { if d.parent != nil { // Note that d.parent or d.name may be stale if there is a concurrent // rename operation. Inotify does not provide consistency guarantees. - d.parent.inode.watches.Notify(d.name, events, cookie) + d.parent.inode.watches.NotifyWithExclusions(d.name, events, cookie, et, d.unlinked) } - d.inode.watches.Notify("", events, cookie) + d.inode.watches.Notify("", events, cookie, et) } // Watches implements vfs.DentryImpl.Watches. @@ -676,9 +681,8 @@ func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) return err } - // Generate inotify events. if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { - d.InotifyWithParent(ev, 0) + d.InotifyWithParent(ev, 0, vfs.InodeEvent) } return nil } @@ -701,7 +705,7 @@ func (fd *fileDescription) Setxattr(ctx context.Context, opts vfs.SetxattrOption } // Generate inotify events. - d.InotifyWithParent(linux.IN_ATTRIB, 0) + d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil } @@ -713,7 +717,7 @@ func (fd *fileDescription) Removexattr(ctx context.Context, name string) error { } // Generate inotify events. - d.InotifyWithParent(linux.IN_ATTRIB, 0) + d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil } diff --git a/pkg/sentry/kernel/fd_table.go b/pkg/sentry/kernel/fd_table.go index ef73e1169..dbfcef0fa 100644 --- a/pkg/sentry/kernel/fd_table.go +++ b/pkg/sentry/kernel/fd_table.go @@ -158,7 +158,7 @@ func (f *FDTable) dropVFS2(file *vfs.FileDescription) { if file.IsWritable() { ev = linux.IN_CLOSE_WRITE } - file.Dentry().InotifyWithParent(ev, 0) + file.Dentry().InotifyWithParent(ev, 0, vfs.PathEvent) // Drop the table reference. file.DecRef() diff --git a/pkg/sentry/syscalls/linux/vfs2/read_write.go b/pkg/sentry/syscalls/linux/vfs2/read_write.go index 92b5631a3..7f9debd4a 100644 --- a/pkg/sentry/syscalls/linux/vfs2/read_write.go +++ b/pkg/sentry/syscalls/linux/vfs2/read_write.go @@ -94,7 +94,7 @@ func read(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, opt n, err := file.Read(t, dst, opts) if err != syserror.ErrWouldBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0) + file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) } return n, err } @@ -102,7 +102,7 @@ func read(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, opt allowBlock, deadline, hasDeadline := blockPolicy(t, file) if !allowBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0) + file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) } return n, err } @@ -135,7 +135,7 @@ func read(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, opt file.EventUnregister(&w) if total > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0) + file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) } return total, err } @@ -258,7 +258,7 @@ func pread(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, of n, err := file.PRead(t, dst, offset, opts) if err != syserror.ErrWouldBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0) + file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) } return n, err } @@ -266,7 +266,7 @@ func pread(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, of allowBlock, deadline, hasDeadline := blockPolicy(t, file) if !allowBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0) + file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) } return n, err } @@ -299,7 +299,7 @@ func pread(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, of file.EventUnregister(&w) if total > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0) + file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) } return total, err } @@ -364,7 +364,7 @@ func write(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, op n, err := file.Write(t, src, opts) if err != syserror.ErrWouldBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0) + file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) } return n, err } @@ -372,7 +372,7 @@ func write(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, op allowBlock, deadline, hasDeadline := blockPolicy(t, file) if !allowBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0) + file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) } return n, err } @@ -405,7 +405,7 @@ func write(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, op file.EventUnregister(&w) if total > 0 { - file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0) + file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) } return total, err } @@ -528,7 +528,7 @@ func pwrite(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, o n, err := file.PWrite(t, src, offset, opts) if err != syserror.ErrWouldBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0) + file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) } return n, err } @@ -536,7 +536,7 @@ func pwrite(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, o allowBlock, deadline, hasDeadline := blockPolicy(t, file) if !allowBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0) + file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) } return n, err } @@ -569,7 +569,7 @@ func pwrite(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, o file.EventUnregister(&w) if total > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0) + file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) } return total, err } diff --git a/pkg/sentry/vfs/anonfs.go b/pkg/sentry/vfs/anonfs.go index 55a3d54cc..b7c6b60b8 100644 --- a/pkg/sentry/vfs/anonfs.go +++ b/pkg/sentry/vfs/anonfs.go @@ -301,7 +301,7 @@ func (d *anonDentry) DecRef() { // InotifyWithParent implements DentryImpl.InotifyWithParent. // // TODO(gvisor.dev/issue/1479): Implement inotify. -func (d *anonDentry) InotifyWithParent(events uint32, cookie uint32) {} +func (d *anonDentry) InotifyWithParent(events uint32, cookie uint32, et EventType) {} // Watches implements DentryImpl.Watches. // diff --git a/pkg/sentry/vfs/dentry.go b/pkg/sentry/vfs/dentry.go index d61b9e09b..24af13eb1 100644 --- a/pkg/sentry/vfs/dentry.go +++ b/pkg/sentry/vfs/dentry.go @@ -113,7 +113,7 @@ type DentryImpl interface { // // Note that the events may not actually propagate up to the user, depending // on the event masks. - InotifyWithParent(events uint32, cookie uint32) + InotifyWithParent(events uint32, cookie uint32, et EventType) // Watches returns the set of inotify watches for the file corresponding to // the Dentry. Dentries that are hard links to the same underlying file @@ -151,8 +151,8 @@ func (d *Dentry) isMounted() bool { // InotifyWithParent notifies all watches on the inodes for this dentry and // its parent of events. -func (d *Dentry) InotifyWithParent(events uint32, cookie uint32) { - d.impl.InotifyWithParent(events, cookie) +func (d *Dentry) InotifyWithParent(events uint32, cookie uint32, et EventType) { + d.impl.InotifyWithParent(events, cookie, et) } // Watches returns the set of inotify watches associated with d. diff --git a/pkg/sentry/vfs/inotify.go b/pkg/sentry/vfs/inotify.go index 1d28ccb46..05a3051a4 100644 --- a/pkg/sentry/vfs/inotify.go +++ b/pkg/sentry/vfs/inotify.go @@ -33,6 +33,19 @@ import ( // must be a power 2 for rounding below. const inotifyEventBaseSize = 16 +// EventType defines different kinds of inotfiy events. +// +// The way events are labelled appears somewhat arbitrary, but they must match +// Linux so that IN_EXCL_UNLINK behaves as it does in Linux. +type EventType uint8 + +// PathEvent and InodeEvent correspond to FSNOTIFY_EVENT_PATH and +// FSNOTIFY_EVENT_INODE in Linux. +const ( + PathEvent EventType = iota + InodeEvent EventType = iota +) + // Inotify represents an inotify instance created by inotify_init(2) or // inotify_init1(2). Inotify implements FileDescriptionImpl. // @@ -419,13 +432,22 @@ func (w *Watches) Remove(id uint64) { } // Notify queues a new event with all watches in this set. -func (w *Watches) Notify(name string, events, cookie uint32) { +func (w *Watches) Notify(name string, events, cookie uint32, et EventType) { + w.NotifyWithExclusions(name, events, cookie, et, false) +} + +// NotifyWithExclusions queues a new event with watches in this set. Watches +// with IN_EXCL_UNLINK are skipped if the event is coming from a child that +// has been unlinked. +func (w *Watches) NotifyWithExclusions(name string, events, cookie uint32, et EventType, unlinked bool) { // N.B. We don't defer the unlocks because Notify is in the hot path of // all IO operations, and the defer costs too much for small IO // operations. w.mu.RLock() for _, watch := range w.ws { - // TODO(gvisor.dev/issue/1479): Skip for IN_EXCL_UNLINK cases. + if unlinked && watch.ExcludeUnlinkedChildren() && et == PathEvent { + continue + } watch.Notify(name, events, cookie) } w.mu.RUnlock() @@ -434,7 +456,7 @@ func (w *Watches) Notify(name string, events, cookie uint32) { // HandleDeletion is called when the watch target is destroyed to emit // the appropriate events. func (w *Watches) HandleDeletion() { - w.Notify("", linux.IN_DELETE_SELF, 0) + w.Notify("", linux.IN_DELETE_SELF, 0, InodeEvent) // TODO(gvisor.dev/issue/1479): This doesn't work because maps are not copied // by value. Ideally, we wouldn't have this circular locking so we can just @@ -655,8 +677,8 @@ func InotifyEventFromStatMask(mask uint32) uint32 { // InotifyRemoveChild sends the appriopriate notifications to the watch sets of // the child being removed and its parent. func InotifyRemoveChild(self, parent *Watches, name string) { - self.Notify("", linux.IN_ATTRIB, 0) - parent.Notify(name, linux.IN_DELETE, 0) + self.Notify("", linux.IN_ATTRIB, 0, InodeEvent) + parent.Notify(name, linux.IN_DELETE, 0, InodeEvent) // TODO(gvisor.dev/issue/1479): implement IN_EXCL_UNLINK. } @@ -668,8 +690,8 @@ func InotifyRename(ctx context.Context, renamed, oldParent, newParent *Watches, dirEv = linux.IN_ISDIR } cookie := uniqueid.InotifyCookie(ctx) - oldParent.Notify(oldName, dirEv|linux.IN_MOVED_FROM, cookie) - newParent.Notify(newName, dirEv|linux.IN_MOVED_TO, cookie) + oldParent.Notify(oldName, dirEv|linux.IN_MOVED_FROM, cookie, InodeEvent) + newParent.Notify(newName, dirEv|linux.IN_MOVED_TO, cookie, InodeEvent) // Somewhat surprisingly, self move events do not have a cookie. - renamed.Notify("", linux.IN_MOVE_SELF, 0) + renamed.Notify("", linux.IN_MOVE_SELF, 0, InodeEvent) } diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index be6f21dba..52643a7c5 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -422,7 +422,7 @@ func (vfs *VirtualFilesystem) OpenAt(ctx context.Context, creds *auth.Credential } } - fd.Dentry().InotifyWithParent(linux.IN_OPEN, 0) + fd.Dentry().InotifyWithParent(linux.IN_OPEN, 0, PathEvent) return fd, nil } if !rp.handleError(err) { diff --git a/test/syscalls/linux/inotify.cc b/test/syscalls/linux/inotify.cc index e4565467b..2306d9cab 100644 --- a/test/syscalls/linux/inotify.cc +++ b/test/syscalls/linux/inotify.cc @@ -1778,6 +1778,170 @@ TEST(Inotify, IncludeUnlinkedFile_NoRandomSave) { })); } +// Watches created with IN_EXCL_UNLINK will stop emitting events on fds for +// children that have already been unlinked. +// +// We need to disable S/R because there are filesystems where we cannot re-open +// fds to an unlinked file across S/R, e.g. gofer-backed filesytems. +TEST(Inotify, ExcludeUnlink_NoRandomSave) { + const DisableSave ds; + // TODO(gvisor.dev/issue/1624): This test fails on VFS1. + SKIP_IF(IsRunningWithVFS1()); + + const TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const TempPath file = + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(dir.path())); + + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(file.path(), O_RDWR)); + + const FileDescriptor inotify_fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + const int wd = ASSERT_NO_ERRNO_AND_VALUE(InotifyAddWatch( + inotify_fd.get(), dir.path(), IN_ALL_EVENTS | IN_EXCL_UNLINK)); + + // Unlink the child, which should cause further operations on the open file + // descriptor to be ignored. + ASSERT_THAT(unlink(file.path().c_str()), SyscallSucceeds()); + int val = 0; + ASSERT_THAT(write(fd.get(), &val, sizeof(val)), SyscallSucceeds()); + ASSERT_THAT(read(fd.get(), &val, sizeof(val)), SyscallSucceeds()); + const std::vector events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({Event(IN_DELETE, wd, Basename(file.path()))})); +} + +// We need to disable S/R because there are filesystems where we cannot re-open +// fds to an unlinked file across S/R, e.g. gofer-backed filesytems. +TEST(Inotify, ExcludeUnlinkDirectory_NoRandomSave) { + const DisableSave ds; + + const TempPath parent = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + TempPath dir = + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDirIn(parent.path())); + std::string dirPath = dir.path(); + const FileDescriptor inotify_fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(dirPath.c_str(), O_RDONLY | O_DIRECTORY)); + const int wd = ASSERT_NO_ERRNO_AND_VALUE(InotifyAddWatch( + inotify_fd.get(), parent.path(), IN_ALL_EVENTS | IN_EXCL_UNLINK)); + + // Unlink the dir, and then close the open fd. + ASSERT_THAT(rmdir(dirPath.c_str()), SyscallSucceeds()); + dir.reset(); + + const std::vector events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + // No close event should appear. + ASSERT_THAT(events, + Are({Event(IN_DELETE | IN_ISDIR, wd, Basename(dirPath))})); +} + +// If "dir/child" and "dir/child2" are links to the same file, and "dir/child" +// is unlinked, a watch on "dir" with IN_EXCL_UNLINK will exclude future events +// for fds on "dir/child" but not "dir/child2". +// +// We need to disable S/R because there are filesystems where we cannot re-open +// fds to an unlinked file across S/R, e.g. gofer-backed filesytems. +TEST(Inotify, ExcludeUnlinkMultipleChildren_NoRandomSave) { + const DisableSave ds; + // TODO(gvisor.dev/issue/1624): This test fails on VFS1. + SKIP_IF(IsRunningWithVFS1()); + + const TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const TempPath file = + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(dir.path())); + std::string path1 = file.path(); + std::string path2 = NewTempAbsPathInDir(dir.path()); + + const int rc = link(path1.c_str(), path2.c_str()); + // NOTE(b/34861058): link(2) is only supported on tmpfs in the sandbox. + SKIP_IF(IsRunningOnGvisor() && rc != 0 && + (errno == EPERM || errno == ENOENT)); + ASSERT_THAT(rc, SyscallSucceeds()); + const FileDescriptor fd1 = + ASSERT_NO_ERRNO_AND_VALUE(Open(path1.c_str(), O_RDWR)); + const FileDescriptor fd2 = + ASSERT_NO_ERRNO_AND_VALUE(Open(path2.c_str(), O_RDWR)); + + const FileDescriptor inotify_fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + const int wd = ASSERT_NO_ERRNO_AND_VALUE(InotifyAddWatch( + inotify_fd.get(), dir.path(), IN_ALL_EVENTS | IN_EXCL_UNLINK)); + + // After unlinking path1, only events on the fd for path2 should be generated. + ASSERT_THAT(unlink(path1.c_str()), SyscallSucceeds()); + ASSERT_THAT(write(fd1.get(), "x", 1), SyscallSucceeds()); + ASSERT_THAT(write(fd2.get(), "x", 1), SyscallSucceeds()); + + const std::vector events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({ + Event(IN_DELETE, wd, Basename(path1)), + Event(IN_MODIFY, wd, Basename(path2)), + })); +} + +// On native Linux, actions of data type FSNOTIFY_EVENT_INODE are not affected +// by IN_EXCL_UNLINK (see +// fs/notify/inotify/inotify_fsnotify.c:inotify_handle_event). Inode-level +// events include changes to metadata and extended attributes. +// +// We need to disable S/R because there are filesystems where we cannot re-open +// fds to an unlinked file across S/R, e.g. gofer-backed filesytems. +TEST(Inotify, ExcludeUnlinkInodeEvents_NoRandomSave) { + const DisableSave ds; + + const TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const TempPath file = + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(dir.path())); + + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(file.path().c_str(), O_RDWR)); + const FileDescriptor inotify_fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + const int wd = ASSERT_NO_ERRNO_AND_VALUE(InotifyAddWatch( + inotify_fd.get(), dir.path(), IN_ALL_EVENTS | IN_EXCL_UNLINK)); + + // NOTE(b/157163751): Create another link before unlinking. This is needed for + // the gofer filesystem in gVisor, where open fds will not work once the link + // count hits zero. In VFS2, we end up skipping the gofer test anyway, because + // hard links are not supported for gofer fs. + if (IsRunningOnGvisor()) { + std::string link_path = NewTempAbsPath(); + const int rc = link(file.path().c_str(), link_path.c_str()); + // NOTE(b/34861058): link(2) is only supported on tmpfs in the sandbox. + SKIP_IF(rc != 0 && (errno == EPERM || errno == ENOENT)); + ASSERT_THAT(rc, SyscallSucceeds()); + } + + // Even after unlinking, inode-level operations will trigger events regardless + // of IN_EXCL_UNLINK. + ASSERT_THAT(unlink(file.path().c_str()), SyscallSucceeds()); + + // Perform various actions on fd. + ASSERT_THAT(ftruncate(fd.get(), 12345), SyscallSucceeds()); + std::vector events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({ + Event(IN_DELETE, wd, Basename(file.path())), + Event(IN_MODIFY, wd, Basename(file.path())), + })); + + struct timeval times[2] = {{1, 0}, {2, 0}}; + ASSERT_THAT(futimes(fd.get(), times), SyscallSucceeds()); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({Event(IN_ATTRIB, wd, Basename(file.path()))})); + + // S/R is disabled on this entire test due to behavior with unlink; it must + // also be disabled after this point because of fchmod. + ASSERT_THAT(fchmod(fd.get(), 0777), SyscallSucceeds()); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({Event(IN_ATTRIB, wd, Basename(file.path()))})); +} + } // namespace } // namespace testing } // namespace gvisor -- cgit v1.2.3 From 65569cfca08a99e3700108cec64f3aa443c357b0 Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Fri, 29 May 2020 15:40:06 -0700 Subject: Update Go version build tags None of the dependencies have changed in 1.15. It may be possible to simplify some of the wrappers in rawfile following 1.13, but that can come in a later change. PiperOrigin-RevId: 313863264 --- pkg/procid/procid_amd64.s | 2 +- pkg/procid/procid_arm64.s | 2 +- pkg/sentry/platform/kvm/bluepill_unsafe.go | 2 +- pkg/sentry/platform/kvm/machine_unsafe.go | 2 +- pkg/sentry/platform/ptrace/subprocess_unsafe.go | 2 +- pkg/sentry/vfs/mount_unsafe.go | 2 +- pkg/sleep/sleep_unsafe.go | 2 +- pkg/sync/memmove_unsafe.go | 2 +- pkg/sync/mutex_unsafe.go | 2 +- pkg/sync/rwmutex_unsafe.go | 2 +- pkg/syncevent/waiter_unsafe.go | 2 +- pkg/tcpip/link/rawfile/blockingpoll_yield_unsafe.go | 2 +- pkg/tcpip/time_unsafe.go | 2 +- 13 files changed, 13 insertions(+), 13 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/procid/procid_amd64.s b/pkg/procid/procid_amd64.s index 38cea9be3..7c622e5d7 100644 --- a/pkg/procid/procid_amd64.s +++ b/pkg/procid/procid_amd64.s @@ -14,7 +14,7 @@ // +build amd64 // +build go1.8 -// +build !go1.15 +// +build !go1.16 #include "textflag.h" diff --git a/pkg/procid/procid_arm64.s b/pkg/procid/procid_arm64.s index 4f4b70fef..48ebb5fd1 100644 --- a/pkg/procid/procid_arm64.s +++ b/pkg/procid/procid_arm64.s @@ -14,7 +14,7 @@ // +build arm64 // +build go1.8 -// +build !go1.15 +// +build !go1.16 #include "textflag.h" diff --git a/pkg/sentry/platform/kvm/bluepill_unsafe.go b/pkg/sentry/platform/kvm/bluepill_unsafe.go index 9add7c944..2407014e9 100644 --- a/pkg/sentry/platform/kvm/bluepill_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_unsafe.go @@ -13,7 +13,7 @@ // limitations under the License. // +build go1.12 -// +build !go1.15 +// +build !go1.16 // Check go:linkname function signatures when updating Go version. diff --git a/pkg/sentry/platform/kvm/machine_unsafe.go b/pkg/sentry/platform/kvm/machine_unsafe.go index f04be2ab5..de7df4f80 100644 --- a/pkg/sentry/platform/kvm/machine_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_unsafe.go @@ -13,7 +13,7 @@ // limitations under the License. // +build go1.12 -// +build !go1.15 +// +build !go1.16 // Check go:linkname function signatures when updating Go version. diff --git a/pkg/sentry/platform/ptrace/subprocess_unsafe.go b/pkg/sentry/platform/ptrace/subprocess_unsafe.go index 2ae6b9f9d..0bee995e4 100644 --- a/pkg/sentry/platform/ptrace/subprocess_unsafe.go +++ b/pkg/sentry/platform/ptrace/subprocess_unsafe.go @@ -13,7 +13,7 @@ // limitations under the License. // +build go1.12 -// +build !go1.15 +// +build !go1.16 // Check go:linkname function signatures when updating Go version. diff --git a/pkg/sentry/vfs/mount_unsafe.go b/pkg/sentry/vfs/mount_unsafe.go index bc7581698..70f850ca4 100644 --- a/pkg/sentry/vfs/mount_unsafe.go +++ b/pkg/sentry/vfs/mount_unsafe.go @@ -13,7 +13,7 @@ // limitations under the License. // +build go1.12 -// +build !go1.15 +// +build !go1.16 // Check go:linkname function signatures when updating Go version. diff --git a/pkg/sleep/sleep_unsafe.go b/pkg/sleep/sleep_unsafe.go index 65bfcf778..f68c12620 100644 --- a/pkg/sleep/sleep_unsafe.go +++ b/pkg/sleep/sleep_unsafe.go @@ -13,7 +13,7 @@ // limitations under the License. // +build go1.11 -// +build !go1.15 +// +build !go1.16 // Check go:linkname function signatures when updating Go version. diff --git a/pkg/sync/memmove_unsafe.go b/pkg/sync/memmove_unsafe.go index ad4a3a37e..1d7780695 100644 --- a/pkg/sync/memmove_unsafe.go +++ b/pkg/sync/memmove_unsafe.go @@ -4,7 +4,7 @@ // license that can be found in the LICENSE file. // +build go1.12 -// +build !go1.15 +// +build !go1.16 // Check go:linkname function signatures when updating Go version. diff --git a/pkg/sync/mutex_unsafe.go b/pkg/sync/mutex_unsafe.go index 3dd15578b..dc034d561 100644 --- a/pkg/sync/mutex_unsafe.go +++ b/pkg/sync/mutex_unsafe.go @@ -4,7 +4,7 @@ // license that can be found in the LICENSE file. // +build go1.13 -// +build !go1.15 +// +build !go1.16 // When updating the build constraint (above), check that syncMutex matches the // standard library sync.Mutex definition. diff --git a/pkg/sync/rwmutex_unsafe.go b/pkg/sync/rwmutex_unsafe.go index ea6cdc447..995c0346e 100644 --- a/pkg/sync/rwmutex_unsafe.go +++ b/pkg/sync/rwmutex_unsafe.go @@ -4,7 +4,7 @@ // license that can be found in the LICENSE file. // +build go1.13 -// +build !go1.15 +// +build !go1.16 // Check go:linkname function signatures when updating Go version. diff --git a/pkg/syncevent/waiter_unsafe.go b/pkg/syncevent/waiter_unsafe.go index 112e0e604..ad271e1a0 100644 --- a/pkg/syncevent/waiter_unsafe.go +++ b/pkg/syncevent/waiter_unsafe.go @@ -13,7 +13,7 @@ // limitations under the License. // +build go1.11 -// +build !go1.15 +// +build !go1.16 // Check go:linkname function signatures when updating Go version. diff --git a/pkg/tcpip/link/rawfile/blockingpoll_yield_unsafe.go b/pkg/tcpip/link/rawfile/blockingpoll_yield_unsafe.go index 0b5a6cf49..99313ee25 100644 --- a/pkg/tcpip/link/rawfile/blockingpoll_yield_unsafe.go +++ b/pkg/tcpip/link/rawfile/blockingpoll_yield_unsafe.go @@ -14,7 +14,7 @@ // +build linux,amd64 linux,arm64 // +build go1.12 -// +build !go1.15 +// +build !go1.16 // Check go:linkname function signatures when updating Go version. diff --git a/pkg/tcpip/time_unsafe.go b/pkg/tcpip/time_unsafe.go index 2f98a996f..7f172f978 100644 --- a/pkg/tcpip/time_unsafe.go +++ b/pkg/tcpip/time_unsafe.go @@ -13,7 +13,7 @@ // limitations under the License. // +build go1.9 -// +build !go1.15 +// +build !go1.16 // Check go:linkname function signatures when updating Go version. -- cgit v1.2.3 From 93edb36cbb653f8415d5c813e40ef21fed123417 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Fri, 29 May 2020 16:33:50 -0700 Subject: Refactor the ResolveExecutablePath logic. PiperOrigin-RevId: 313871804 --- pkg/sentry/control/BUILD | 1 + pkg/sentry/control/proc.go | 68 ++++-------------- pkg/sentry/fs/mounts.go | 72 ------------------- pkg/sentry/fs/user/BUILD | 7 +- pkg/sentry/fs/user/path.go | 169 +++++++++++++++++++++++++++++++++++++++++++++ pkg/sentry/fs/user/user.go | 2 + runsc/boot/fs.go | 14 ++-- runsc/boot/vfs.go | 76 +++----------------- 8 files changed, 205 insertions(+), 204 deletions(-) create mode 100644 pkg/sentry/fs/user/path.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/control/BUILD b/pkg/sentry/control/BUILD index e74275d2d..0c9a62f0d 100644 --- a/pkg/sentry/control/BUILD +++ b/pkg/sentry/control/BUILD @@ -23,6 +23,7 @@ go_library( "//pkg/sentry/fdimport", "//pkg/sentry/fs", "//pkg/sentry/fs/host", + "//pkg/sentry/fs/user", "//pkg/sentry/fsbridge", "//pkg/sentry/fsimpl/host", "//pkg/sentry/kernel", diff --git a/pkg/sentry/control/proc.go b/pkg/sentry/control/proc.go index 2ed17ee09..8767430b7 100644 --- a/pkg/sentry/control/proc.go +++ b/pkg/sentry/control/proc.go @@ -18,7 +18,6 @@ import ( "bytes" "encoding/json" "fmt" - "path" "sort" "strings" "text/tabwriter" @@ -28,10 +27,10 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/fdimport" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/host" + "gvisor.dev/gvisor/pkg/sentry/fs/user" "gvisor.dev/gvisor/pkg/sentry/fsbridge" hostvfs2 "gvisor.dev/gvisor/pkg/sentry/fsimpl/host" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -190,17 +189,12 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI // transferred to the new process. initArgs.MountNamespaceVFS2 = proc.Kernel.GlobalInit().Leader().MountNamespaceVFS2() } - - paths := fs.GetPath(initArgs.Envv) - vfsObj := proc.Kernel.VFS() - file, err := ResolveExecutablePath(ctx, vfsObj, initArgs.WorkingDirectory, initArgs.Argv[0], paths) + file, err := getExecutableFD(ctx, creds, proc.Kernel.VFS(), initArgs.MountNamespaceVFS2, initArgs.Envv, initArgs.WorkingDirectory, initArgs.Argv[0]) if err != nil { - return nil, 0, nil, nil, fmt.Errorf("error finding executable %q in PATH %v: %v", initArgs.Argv[0], paths, err) + return nil, 0, nil, nil, fmt.Errorf("error finding executable %q in environment %v: %v", initArgs.Argv[0], initArgs.Envv, err) } initArgs.File = fsbridge.NewVFSFile(file) } else { - // Get the full path to the filename from the PATH env variable. - paths := fs.GetPath(initArgs.Envv) if initArgs.MountNamespace == nil { // Set initArgs so that 'ctx' returns the namespace. initArgs.MountNamespace = proc.Kernel.GlobalInit().Leader().MountNamespace() @@ -209,9 +203,9 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI // be donated to the new process in CreateProcess. initArgs.MountNamespace.IncRef() } - f, err := initArgs.MountNamespace.ResolveExecutablePath(ctx, initArgs.WorkingDirectory, initArgs.Argv[0], paths) + f, err := user.ResolveExecutablePath(ctx, creds, initArgs.MountNamespace, initArgs.Envv, initArgs.WorkingDirectory, initArgs.Argv[0]) if err != nil { - return nil, 0, nil, nil, fmt.Errorf("error finding executable %q in PATH %v: %v", initArgs.Argv[0], paths, err) + return nil, 0, nil, nil, fmt.Errorf("error finding executable %q in PATH %v: %v", initArgs.Argv[0], initArgs.Envv, err) } initArgs.Filename = f } @@ -429,53 +423,17 @@ func ttyName(tty *kernel.TTY) string { return fmt.Sprintf("pts/%d", tty.Index) } -// ResolveExecutablePath resolves the given executable name given a set of -// paths that might contain it. -func ResolveExecutablePath(ctx context.Context, vfsObj *vfs.VirtualFilesystem, wd, name string, paths []string) (*vfs.FileDescription, error) { - root := vfs.RootFromContext(ctx) - defer root.DecRef() - creds := auth.CredentialsFromContext(ctx) - - // Absolute paths can be used directly. - if path.IsAbs(name) { - return openExecutable(ctx, vfsObj, creds, root, name) - } - - // Paths with '/' in them should be joined to the working directory, or - // to the root if working directory is not set. - if strings.IndexByte(name, '/') > 0 { - if len(wd) == 0 { - wd = "/" - } - if !path.IsAbs(wd) { - return nil, fmt.Errorf("working directory %q must be absolute", wd) - } - return openExecutable(ctx, vfsObj, creds, root, path.Join(wd, name)) +// getExecutableFD resolves the given executable name and returns a +// vfs.FileDescription for the executable file. +func getExecutableFD(ctx context.Context, creds *auth.Credentials, vfsObj *vfs.VirtualFilesystem, mns *vfs.MountNamespace, envv []string, wd, name string) (*vfs.FileDescription, error) { + path, err := user.ResolveExecutablePathVFS2(ctx, creds, mns, envv, wd, name) + if err != nil { + return nil, err } - // Otherwise, we must lookup the name in the paths, starting from the - // calling context's root directory. - for _, p := range paths { - if !path.IsAbs(p) { - // Relative paths aren't safe, no one should be using them. - log.Warningf("Skipping relative path %q in $PATH", p) - continue - } - - binPath := path.Join(p, name) - f, err := openExecutable(ctx, vfsObj, creds, root, binPath) - if err != nil { - return nil, err - } - if f == nil { - continue // Not found/no access. - } - return f, nil - } - return nil, syserror.ENOENT -} + root := vfs.RootFromContext(ctx) + defer root.DecRef() -func openExecutable(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, root vfs.VirtualDentry, path string) (*vfs.FileDescription, error) { pop := vfs.PathOperation{ Root: root, Start: root, // binPath is absolute, Start can be anything. diff --git a/pkg/sentry/fs/mounts.go b/pkg/sentry/fs/mounts.go index b414ddaee..3f2bd0e87 100644 --- a/pkg/sentry/fs/mounts.go +++ b/pkg/sentry/fs/mounts.go @@ -17,13 +17,9 @@ package fs import ( "fmt" "math" - "path" - "strings" "syscall" - "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sync" @@ -625,71 +621,3 @@ func (mns *MountNamespace) SyncAll(ctx context.Context) { defer mns.mu.Unlock() mns.root.SyncAll(ctx) } - -// ResolveExecutablePath resolves the given executable name given a set of -// paths that might contain it. -func (mns *MountNamespace) ResolveExecutablePath(ctx context.Context, wd, name string, paths []string) (string, error) { - // Absolute paths can be used directly. - if path.IsAbs(name) { - return name, nil - } - - // Paths with '/' in them should be joined to the working directory, or - // to the root if working directory is not set. - if strings.IndexByte(name, '/') > 0 { - if wd == "" { - wd = "/" - } - if !path.IsAbs(wd) { - return "", fmt.Errorf("working directory %q must be absolute", wd) - } - return path.Join(wd, name), nil - } - - // Otherwise, We must lookup the name in the paths, starting from the - // calling context's root directory. - root := RootFromContext(ctx) - if root == nil { - // Caller has no root. Don't bother traversing anything. - return "", syserror.ENOENT - } - defer root.DecRef() - for _, p := range paths { - binPath := path.Join(p, name) - traversals := uint(linux.MaxSymlinkTraversals) - d, err := mns.FindInode(ctx, root, nil, binPath, &traversals) - if err == syserror.ENOENT || err == syserror.EACCES { - // Didn't find it here. - continue - } - if err != nil { - return "", err - } - defer d.DecRef() - - // Check that it is a regular file. - if !IsRegular(d.Inode.StableAttr) { - continue - } - - // Check whether we can read and execute the found file. - if err := d.Inode.CheckPermission(ctx, PermMask{Read: true, Execute: true}); err != nil { - log.Infof("Found executable at %q, but user cannot execute it: %v", binPath, err) - continue - } - return path.Join("/", p, name), nil - } - return "", syserror.ENOENT -} - -// GetPath returns the PATH as a slice of strings given the environment -// variables. -func GetPath(env []string) []string { - const prefix = "PATH=" - for _, e := range env { - if strings.HasPrefix(e, prefix) { - return strings.Split(strings.TrimPrefix(e, prefix), ":") - } - } - return nil -} diff --git a/pkg/sentry/fs/user/BUILD b/pkg/sentry/fs/user/BUILD index f37f979f1..bd5dac373 100644 --- a/pkg/sentry/fs/user/BUILD +++ b/pkg/sentry/fs/user/BUILD @@ -4,15 +4,20 @@ package(licenses = ["notice"]) go_library( name = "user", - srcs = ["user.go"], + srcs = [ + "path.go", + "user.go", + ], visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", "//pkg/context", "//pkg/fspath", + "//pkg/log", "//pkg/sentry/fs", "//pkg/sentry/kernel/auth", "//pkg/sentry/vfs", + "//pkg/syserror", "//pkg/usermem", ], ) diff --git a/pkg/sentry/fs/user/path.go b/pkg/sentry/fs/user/path.go new file mode 100644 index 000000000..fbd4547a7 --- /dev/null +++ b/pkg/sentry/fs/user/path.go @@ -0,0 +1,169 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package user + +import ( + "fmt" + "path" + "strings" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// ResolveExecutablePath resolves the given executable name given the working +// dir and environment. +func ResolveExecutablePath(ctx context.Context, creds *auth.Credentials, mns *fs.MountNamespace, envv []string, wd, name string) (string, error) { + // Absolute paths can be used directly. + if path.IsAbs(name) { + return name, nil + } + + // Paths with '/' in them should be joined to the working directory, or + // to the root if working directory is not set. + if strings.IndexByte(name, '/') > 0 { + if wd == "" { + wd = "/" + } + if !path.IsAbs(wd) { + return "", fmt.Errorf("working directory %q must be absolute", wd) + } + return path.Join(wd, name), nil + } + + // Otherwise, We must lookup the name in the paths, starting from the + // calling context's root directory. + paths := getPath(envv) + + root := fs.RootFromContext(ctx) + if root == nil { + // Caller has no root. Don't bother traversing anything. + return "", syserror.ENOENT + } + defer root.DecRef() + for _, p := range paths { + if !path.IsAbs(p) { + // Relative paths aren't safe, no one should be using them. + log.Warningf("Skipping relative path %q in $PATH", p) + continue + } + + binPath := path.Join(p, name) + traversals := uint(linux.MaxSymlinkTraversals) + d, err := mns.FindInode(ctx, root, nil, binPath, &traversals) + if err == syserror.ENOENT || err == syserror.EACCES { + // Didn't find it here. + continue + } + if err != nil { + return "", err + } + defer d.DecRef() + + // Check that it is a regular file. + if !fs.IsRegular(d.Inode.StableAttr) { + continue + } + + // Check whether we can read and execute the found file. + if err := d.Inode.CheckPermission(ctx, fs.PermMask{Read: true, Execute: true}); err != nil { + log.Infof("Found executable at %q, but user cannot execute it: %v", binPath, err) + continue + } + return path.Join("/", p, name), nil + } + + // Couldn't find it. + return "", syserror.ENOENT +} + +// ResolveExecutablePathVFS2 resolves the given executable name given the +// working dir and environment. +func ResolveExecutablePathVFS2(ctx context.Context, creds *auth.Credentials, mns *vfs.MountNamespace, envv []string, wd, name string) (string, error) { + // Absolute paths can be used directly. + if path.IsAbs(name) { + return name, nil + } + + // Paths with '/' in them should be joined to the working directory, or + // to the root if working directory is not set. + if strings.IndexByte(name, '/') > 0 { + if wd == "" { + wd = "/" + } + if !path.IsAbs(wd) { + return "", fmt.Errorf("working directory %q must be absolute", wd) + } + return path.Join(wd, name), nil + } + + // Otherwise, We must lookup the name in the paths, starting from the + // calling context's root directory. + paths := getPath(envv) + + root := mns.Root() + defer root.DecRef() + for _, p := range paths { + if !path.IsAbs(p) { + // Relative paths aren't safe, no one should be using them. + log.Warningf("Skipping relative path %q in $PATH", p) + continue + } + + binPath := path.Join(p, name) + pop := &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(binPath), + FollowFinalSymlink: true, + } + opts := &vfs.OpenOptions{ + FileExec: true, + Flags: linux.O_RDONLY, + } + dentry, err := root.Mount().Filesystem().VirtualFilesystem().OpenAt(ctx, creds, pop, opts) + if err == syserror.ENOENT || err == syserror.EACCES { + // Didn't find it here. + continue + } + if err != nil { + return "", err + } + dentry.DecRef() + + return binPath, nil + } + + // Couldn't find it. + return "", syserror.ENOENT +} + +// getPath returns the PATH as a slice of strings given the environment +// variables. +func getPath(env []string) []string { + const prefix = "PATH=" + for _, e := range env { + if strings.HasPrefix(e, prefix) { + return strings.Split(strings.TrimPrefix(e, prefix), ":") + } + } + return nil +} diff --git a/pkg/sentry/fs/user/user.go b/pkg/sentry/fs/user/user.go index fe7f67c00..f4d525523 100644 --- a/pkg/sentry/fs/user/user.go +++ b/pkg/sentry/fs/user/user.go @@ -12,6 +12,8 @@ // See the License for the specific language governing permissions and // limitations under the License. +// Package user contains methods for resolving filesystem paths based on the +// user and their environment. package user import ( diff --git a/runsc/boot/fs.go b/runsc/boot/fs.go index 9cd4f97da..52f8344ca 100644 --- a/runsc/boot/fs.go +++ b/runsc/boot/fs.go @@ -37,6 +37,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/gofer" "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" + "gvisor.dev/gvisor/pkg/sentry/fs/user" "gvisor.dev/gvisor/pkg/sentry/fsimpl/devpts" "gvisor.dev/gvisor/pkg/sentry/fsimpl/devtmpfs" gofervfs2 "gvisor.dev/gvisor/pkg/sentry/fsimpl/gofer" @@ -291,17 +292,10 @@ func setupContainerFS(ctx context.Context, conf *Config, mntr *containerMounter, // Set namespace here so that it can be found in ctx. procArgs.MountNamespace = mns - return setExecutablePath(ctx, procArgs) -} - -// setExecutablePath sets the procArgs.Filename by searching the PATH for an -// executable matching the procArgs.Argv[0]. -func setExecutablePath(ctx context.Context, procArgs *kernel.CreateProcessArgs) error { - paths := fs.GetPath(procArgs.Envv) - exe := procArgs.Argv[0] - f, err := procArgs.MountNamespace.ResolveExecutablePath(ctx, procArgs.WorkingDirectory, exe, paths) + // Resolve the executable path from working dir and environment. + f, err := user.ResolveExecutablePath(ctx, procArgs.Credentials, procArgs.MountNamespace, procArgs.Envv, procArgs.WorkingDirectory, procArgs.Argv[0]) if err != nil { - return fmt.Errorf("searching for executable %q, cwd: %q, $PATH=%q: %v", exe, procArgs.WorkingDirectory, strings.Join(paths, ":"), err) + return fmt.Errorf("searching for executable %q, cwd: %q, envv: %q: %v", procArgs.Argv[0], procArgs.WorkingDirectory, procArgs.Envv, err) } procArgs.Filename = f return nil diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go index fefeeff58..f48e6b0f1 100644 --- a/runsc/boot/vfs.go +++ b/runsc/boot/vfs.go @@ -22,22 +22,21 @@ import ( specs "github.com/opencontainers/runtime-spec/specs-go" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/devices/memdev" - "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/fs/user" "gvisor.dev/gvisor/pkg/sentry/fsimpl/devpts" "gvisor.dev/gvisor/pkg/sentry/fsimpl/devtmpfs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/gofer" "gvisor.dev/gvisor/pkg/sentry/fsimpl/proc" "gvisor.dev/gvisor/pkg/sentry/fsimpl/sys" "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" - "gvisor.dev/gvisor/pkg/syserror" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" ) func registerFilesystems(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials) error { @@ -95,69 +94,14 @@ func setupContainerVFS2(ctx context.Context, conf *Config, mntr *containerMounte return fmt.Errorf("failed to setupFS: %w", err) } procArgs.MountNamespaceVFS2 = mns - return setExecutablePathVFS2(ctx, procArgs) -} - -func setExecutablePathVFS2(ctx context.Context, procArgs *kernel.CreateProcessArgs) error { - exe := procArgs.Argv[0] - // Absolute paths can be used directly. - if path.IsAbs(exe) { - procArgs.Filename = exe - return nil - } - - // Paths with '/' in them should be joined to the working directory, or - // to the root if working directory is not set. - if strings.IndexByte(exe, '/') > 0 { - if !path.IsAbs(procArgs.WorkingDirectory) { - return fmt.Errorf("working directory %q must be absolute", procArgs.WorkingDirectory) - } - procArgs.Filename = path.Join(procArgs.WorkingDirectory, exe) - return nil - } - - // Paths with a '/' are relative to the CWD. - if strings.IndexByte(exe, '/') > 0 { - procArgs.Filename = path.Join(procArgs.WorkingDirectory, exe) - return nil - } - - // Otherwise, We must lookup the name in the paths, starting from the - // root directory. - root := procArgs.MountNamespaceVFS2.Root() - defer root.DecRef() - - paths := fs.GetPath(procArgs.Envv) - creds := procArgs.Credentials - - for _, p := range paths { - binPath := path.Join(p, exe) - pop := &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(binPath), - FollowFinalSymlink: true, - } - opts := &vfs.OpenOptions{ - FileExec: true, - Flags: linux.O_RDONLY, - } - dentry, err := root.Mount().Filesystem().VirtualFilesystem().OpenAt(ctx, creds, pop, opts) - if err == syserror.ENOENT || err == syserror.EACCES { - // Didn't find it here. - continue - } - if err != nil { - return err - } - dentry.DecRef() - - procArgs.Filename = binPath - return nil + // Resolve the executable path from working dir and environment. + f, err := user.ResolveExecutablePathVFS2(ctx, procArgs.Credentials, procArgs.MountNamespaceVFS2, procArgs.Envv, procArgs.WorkingDirectory, procArgs.Argv[0]) + if err != nil { + return fmt.Errorf("searching for executable %q, cwd: %q, envv: %q: %v", procArgs.Argv[0], procArgs.WorkingDirectory, procArgs.Envv, err) } - - return fmt.Errorf("executable %q not found in $PATH=%q", exe, strings.Join(paths, ":")) + procArgs.Filename = f + return nil } func (c *containerMounter) setupVFS2(ctx context.Context, conf *Config, procArgs *kernel.CreateProcessArgs) (*vfs.MountNamespace, error) { -- cgit v1.2.3 From 3a987160aa09f814a8459ed3f6192ce741b701a3 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Mon, 1 Jun 2020 15:31:59 -0700 Subject: Handle gofer blocking opens of host named pipes in VFS2. Using tee instead of read to detect when a O_RDONLY|O_NONBLOCK pipe FD has a writer circumvents the problem of what to do with the byte read from the pipe, avoiding much of the complexity of the fdpipe package. PiperOrigin-RevId: 314216146 --- pkg/sentry/fsimpl/gofer/BUILD | 3 + pkg/sentry/fsimpl/gofer/filesystem.go | 56 ++++++++++------- pkg/sentry/fsimpl/gofer/host_named_pipe.go | 97 ++++++++++++++++++++++++++++++ pkg/sentry/fsimpl/gofer/special_file.go | 77 ++++++++++++++++++++++-- runsc/boot/filter/config.go | 8 +++ 5 files changed, 214 insertions(+), 27 deletions(-) create mode 100644 pkg/sentry/fsimpl/gofer/host_named_pipe.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/BUILD b/pkg/sentry/fsimpl/gofer/BUILD index 67e916525..f5f35a3bc 100644 --- a/pkg/sentry/fsimpl/gofer/BUILD +++ b/pkg/sentry/fsimpl/gofer/BUILD @@ -35,6 +35,7 @@ go_library( "fstree.go", "gofer.go", "handle.go", + "host_named_pipe.go", "p9file.go", "regular_file.go", "socket.go", @@ -47,6 +48,7 @@ go_library( "//pkg/abi/linux", "//pkg/context", "//pkg/fd", + "//pkg/fdnotifier", "//pkg/fspath", "//pkg/log", "//pkg/p9", @@ -71,6 +73,7 @@ go_library( "//pkg/unet", "//pkg/usermem", "//pkg/waiter", + "@org_golang_x_sys//unix:go_default_library", ], ) diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 7f2181216..0000ca6ba 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -873,19 +873,37 @@ func (d *dentry) openSpecialFileLocked(ctx context.Context, mnt *vfs.Mount, opts if opts.Flags&linux.O_DIRECT != 0 { return nil, syserror.EINVAL } - h, err := openHandle(ctx, d.file, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, opts.Flags&linux.O_TRUNC != 0) + // We assume that the server silently inserts O_NONBLOCK in the open flags + // for all named pipes (because all existing gofers do this). + // + // NOTE(b/133875563): This makes named pipe opens racy, because the + // mechanisms for translating nonblocking to blocking opens can only detect + // the instantaneous presence of a peer holding the other end of the pipe + // open, not whether the pipe was *previously* opened by a peer that has + // since closed its end. + isBlockingOpenOfNamedPipe := d.fileType() == linux.S_IFIFO && opts.Flags&linux.O_NONBLOCK == 0 +retry: + h, err := openHandle(ctx, d.file, ats.MayRead(), ats.MayWrite(), opts.Flags&linux.O_TRUNC != 0) if err != nil { + if isBlockingOpenOfNamedPipe && ats == vfs.MayWrite && err == syserror.ENXIO { + // An attempt to open a named pipe with O_WRONLY|O_NONBLOCK fails + // with ENXIO if opening the same named pipe with O_WRONLY would + // block because there are no readers of the pipe. + if err := sleepBetweenNamedPipeOpenChecks(ctx); err != nil { + return nil, err + } + goto retry + } return nil, err } - seekable := d.fileType() == linux.S_IFREG - fd := &specialFileFD{ - handle: h, - seekable: seekable, + if isBlockingOpenOfNamedPipe && ats == vfs.MayRead && h.fd >= 0 { + if err := blockUntilNonblockingPipeHasWriter(ctx, h.fd); err != nil { + h.close(ctx) + return nil, err + } } - if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{ - DenyPRead: !seekable, - DenyPWrite: !seekable, - }); err != nil { + fd, err := newSpecialFileFD(h, mnt, d, opts.Flags) + if err != nil { h.close(ctx) return nil, err } @@ -981,22 +999,16 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving } childVFSFD = &fd.vfsfd } else { - seekable := child.fileType() == linux.S_IFREG - fd := &specialFileFD{ - handle: handle{ - file: openFile, - fd: -1, - }, - seekable: seekable, + h := handle{ + file: openFile, + fd: -1, } if fdobj != nil { - fd.handle.fd = int32(fdobj.Release()) + h.fd = int32(fdobj.Release()) } - if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &child.vfsd, &vfs.FileDescriptionOptions{ - DenyPRead: !seekable, - DenyPWrite: !seekable, - }); err != nil { - fd.handle.close(ctx) + fd, err := newSpecialFileFD(h, mnt, child, opts.Flags) + if err != nil { + h.close(ctx) return nil, err } childVFSFD = &fd.vfsfd diff --git a/pkg/sentry/fsimpl/gofer/host_named_pipe.go b/pkg/sentry/fsimpl/gofer/host_named_pipe.go new file mode 100644 index 000000000..7294de7d6 --- /dev/null +++ b/pkg/sentry/fsimpl/gofer/host_named_pipe.go @@ -0,0 +1,97 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package gofer + +import ( + "fmt" + "sync" + "time" + + "golang.org/x/sys/unix" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/syserror" +) + +// Global pipe used by blockUntilNonblockingPipeHasWriter since we can't create +// pipes after sentry initialization due to syscall filters. +var ( + tempPipeMu sync.Mutex + tempPipeReadFD int + tempPipeWriteFD int + tempPipeBuf [1]byte +) + +func init() { + var pipeFDs [2]int + if err := unix.Pipe(pipeFDs[:]); err != nil { + panic(fmt.Sprintf("failed to create pipe for gofer.blockUntilNonblockingPipeHasWriter: %v", err)) + } + tempPipeReadFD = pipeFDs[0] + tempPipeWriteFD = pipeFDs[1] +} + +func blockUntilNonblockingPipeHasWriter(ctx context.Context, fd int32) error { + for { + ok, err := nonblockingPipeHasWriter(fd) + if err != nil { + return err + } + if ok { + return nil + } + if err := sleepBetweenNamedPipeOpenChecks(ctx); err != nil { + return err + } + } +} + +func nonblockingPipeHasWriter(fd int32) (bool, error) { + tempPipeMu.Lock() + defer tempPipeMu.Unlock() + // Copy 1 byte from fd into the temporary pipe. + n, err := unix.Tee(int(fd), tempPipeWriteFD, 1, unix.SPLICE_F_NONBLOCK) + if err == syserror.EAGAIN { + // The pipe represented by fd is empty, but has a writer. + return true, nil + } + if err != nil { + return false, err + } + if n == 0 { + // The pipe represented by fd is empty and has no writer. + return false, nil + } + // The pipe represented by fd is non-empty, so it either has, or has + // previously had, a writer. Remove the byte copied to the temporary pipe + // before returning. + if n, err := unix.Read(tempPipeReadFD, tempPipeBuf[:]); err != nil || n != 1 { + panic(fmt.Sprintf("failed to drain pipe for gofer.blockUntilNonblockingPipeHasWriter: got (%d, %v), wanted (1, nil)", n, err)) + } + return true, nil +} + +func sleepBetweenNamedPipeOpenChecks(ctx context.Context) error { + t := time.NewTimer(100 * time.Millisecond) + defer t.Stop() + cancel := ctx.SleepStart() + select { + case <-t.C: + ctx.SleepFinish(true) + return nil + case <-cancel: + ctx.SleepFinish(false) + return syserror.ErrInterrupted + } +} diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go index a464e6a94..ff6126b87 100644 --- a/pkg/sentry/fsimpl/gofer/special_file.go +++ b/pkg/sentry/fsimpl/gofer/special_file.go @@ -19,17 +19,18 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fdnotifier" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" ) -// specialFileFD implements vfs.FileDescriptionImpl for files other than -// regular files, directories, and symlinks: pipes, sockets, etc. It is also -// used for regular files when filesystemOptions.specialRegularFiles is in -// effect. specialFileFD differs from regularFileFD by using per-FD handles -// instead of shared per-dentry handles, and never buffering I/O. +// specialFileFD implements vfs.FileDescriptionImpl for pipes, sockets, device +// special files, and (when filesystemOptions.specialRegularFiles is in effect) +// regular files. specialFileFD differs from regularFileFD by using per-FD +// handles instead of shared per-dentry handles, and never buffering I/O. type specialFileFD struct { fileDescription @@ -40,13 +41,47 @@ type specialFileFD struct { // file offset is significant, i.e. a regular file. seekable is immutable. seekable bool + // mayBlock is true if this file description represents a file for which + // queue may send I/O readiness events. mayBlock is immutable. + mayBlock bool + queue waiter.Queue + // If seekable is true, off is the file offset. off is protected by mu. mu sync.Mutex off int64 } +func newSpecialFileFD(h handle, mnt *vfs.Mount, d *dentry, flags uint32) (*specialFileFD, error) { + ftype := d.fileType() + seekable := ftype == linux.S_IFREG + mayBlock := ftype == linux.S_IFIFO || ftype == linux.S_IFSOCK + fd := &specialFileFD{ + handle: h, + seekable: seekable, + mayBlock: mayBlock, + } + if mayBlock && h.fd >= 0 { + if err := fdnotifier.AddFD(h.fd, &fd.queue); err != nil { + return nil, err + } + } + if err := fd.vfsfd.Init(fd, flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{ + DenyPRead: !seekable, + DenyPWrite: !seekable, + }); err != nil { + if mayBlock && h.fd >= 0 { + fdnotifier.RemoveFD(h.fd) + } + return nil, err + } + return fd, nil +} + // Release implements vfs.FileDescriptionImpl.Release. func (fd *specialFileFD) Release() { + if fd.mayBlock && fd.handle.fd >= 0 { + fdnotifier.RemoveFD(fd.handle.fd) + } fd.handle.close(context.Background()) fs := fd.vfsfd.Mount().Filesystem().Impl().(*filesystem) fs.syncMu.Lock() @@ -62,6 +97,32 @@ func (fd *specialFileFD) OnClose(ctx context.Context) error { return fd.handle.file.flush(ctx) } +// Readiness implements waiter.Waitable.Readiness. +func (fd *specialFileFD) Readiness(mask waiter.EventMask) waiter.EventMask { + if fd.mayBlock { + return fdnotifier.NonBlockingPoll(fd.handle.fd, mask) + } + return fd.fileDescription.Readiness(mask) +} + +// EventRegister implements waiter.Waitable.EventRegister. +func (fd *specialFileFD) EventRegister(e *waiter.Entry, mask waiter.EventMask) { + if fd.mayBlock { + fd.queue.EventRegister(e, mask) + return + } + fd.fileDescription.EventRegister(e, mask) +} + +// EventUnregister implements waiter.Waitable.EventUnregister. +func (fd *specialFileFD) EventUnregister(e *waiter.Entry) { + if fd.mayBlock { + fd.queue.EventUnregister(e) + return + } + fd.fileDescription.EventUnregister(e) +} + // PRead implements vfs.FileDescriptionImpl.PRead. func (fd *specialFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { if fd.seekable && offset < 0 { @@ -81,6 +142,9 @@ func (fd *specialFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offs } buf := make([]byte, dst.NumBytes()) n, err := fd.handle.readToBlocksAt(ctx, safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf)), uint64(offset)) + if err == syserror.EAGAIN { + err = syserror.ErrWouldBlock + } if n == 0 { return 0, err } @@ -130,6 +194,9 @@ func (fd *specialFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off return 0, err } n, err := fd.handle.writeFromBlocksAt(ctx, safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf)), uint64(offset)) + if err == syserror.EAGAIN { + err = syserror.ErrWouldBlock + } return int64(n), err } diff --git a/runsc/boot/filter/config.go b/runsc/boot/filter/config.go index 98cdd90dd..60e33425f 100644 --- a/runsc/boot/filter/config.go +++ b/runsc/boot/filter/config.go @@ -288,6 +288,14 @@ var allowedSyscalls = seccomp.SyscallRules{ syscall.SYS_SIGALTSTACK: {}, unix.SYS_STATX: {}, syscall.SYS_SYNC_FILE_RANGE: {}, + syscall.SYS_TEE: []seccomp.Rule{ + { + seccomp.AllowAny{}, + seccomp.AllowAny{}, + seccomp.AllowValue(1), /* len */ + seccomp.AllowValue(unix.SPLICE_F_NONBLOCK), /* flags */ + }, + }, syscall.SYS_TGKILL: []seccomp.Rule{ { seccomp.AllowValue(uint64(os.Getpid())), -- cgit v1.2.3 From 49a9b78f74fca28cc9312dfb29ccbe70e3b5fcc3 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Mon, 1 Jun 2020 18:11:35 -0700 Subject: Fix VFS2 gofer open(O_CREAT) reference leak. gofer.filesystem.createAndOpenChildLocked() doesn't need to take a reference on the new dentry since vfs.FileDescription.Init() will do so. PiperOrigin-RevId: 314242127 --- pkg/sentry/fsimpl/gofer/filesystem.go | 9 +++------ 1 file changed, 3 insertions(+), 6 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 0000ca6ba..36e0e1856 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -760,7 +760,7 @@ afterTrailingSymlink: parent.dirMu.Unlock() return nil, syserror.EPERM } - fd, err := parent.createAndOpenChildLocked(ctx, rp, &opts) + fd, err := parent.createAndOpenChildLocked(ctx, rp, &opts, &ds) parent.dirMu.Unlock() return fd, err } @@ -912,7 +912,7 @@ retry: // Preconditions: d.fs.renameMu must be locked. d.dirMu must be locked. // !d.isSynthetic(). -func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.OpenOptions) (*vfs.FileDescription, error) { +func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.OpenOptions, ds **[]*dentry) (*vfs.FileDescription, error) { if err := d.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil { return nil, err } @@ -965,6 +965,7 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving } return nil, err } + *ds = appendDentry(*ds, child) // Incorporate the fid that was opened by lcreate. useRegularFileFD := child.fileType() == linux.S_IFREG && !d.fs.opts.regularFilesUseSpecialFileFD if useRegularFileFD { @@ -977,10 +978,6 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving child.handleWritable = vfs.MayWriteFileWithOpenFlags(opts.Flags) child.handleMu.Unlock() } - // Take a reference on the new dentry to be held by the new file - // description. (This reference also means that the new dentry is not - // eligible for caching yet, so we don't need to append to a dentry slice.) - child.refs = 1 // Insert the dentry into the tree. d.cacheNewChildLocked(child, name) if d.cachedMetadataAuthoritative() { -- cgit v1.2.3 From 050d8e6e331e01d732471e4641dc51346e7a7d3b Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Mon, 1 Jun 2020 18:30:09 -0700 Subject: Add inotify events for extended attributes and splice. Splice, setxattr and removexattr should generate events. Note that VFS2 already generates events for extended attributes. Updates #1479. PiperOrigin-RevId: 314244261 --- pkg/sentry/syscalls/linux/sys_splice.go | 6 ++ pkg/sentry/syscalls/linux/sys_xattr.go | 12 +++- pkg/sentry/syscalls/linux/vfs2/splice.go | 5 ++ test/syscalls/linux/inotify.cc | 94 +++++++++++++++++++++++++++++++- 4 files changed, 112 insertions(+), 5 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/sys_splice.go b/pkg/sentry/syscalls/linux/sys_splice.go index 39f2b79ec..77c78889d 100644 --- a/pkg/sentry/syscalls/linux/sys_splice.go +++ b/pkg/sentry/syscalls/linux/sys_splice.go @@ -80,6 +80,12 @@ func doSplice(t *kernel.Task, outFile, inFile *fs.File, opts fs.SpliceOpts, nonB } } + if total > 0 { + // On Linux, inotify behavior is not very consistent with splice(2). We try + // our best to emulate Linux for very basic calls to splice, where for some + // reason, events are generated for output files, but not input files. + outFile.Dirent.InotifyEvent(linux.IN_MODIFY, 0) + } return total, err } diff --git a/pkg/sentry/syscalls/linux/sys_xattr.go b/pkg/sentry/syscalls/linux/sys_xattr.go index 2de5e3422..c24946160 100644 --- a/pkg/sentry/syscalls/linux/sys_xattr.go +++ b/pkg/sentry/syscalls/linux/sys_xattr.go @@ -207,7 +207,11 @@ func setXattr(t *kernel.Task, d *fs.Dirent, nameAddr, valueAddr usermem.Addr, si return syserror.EOPNOTSUPP } - return d.Inode.SetXattr(t, d, name, value, flags) + if err := d.Inode.SetXattr(t, d, name, value, flags); err != nil { + return err + } + d.InotifyEvent(linux.IN_ATTRIB, 0) + return nil } func copyInXattrName(t *kernel.Task, nameAddr usermem.Addr) (string, error) { @@ -418,7 +422,11 @@ func removeXattr(t *kernel.Task, d *fs.Dirent, nameAddr usermem.Addr) error { return syserror.EOPNOTSUPP } - return d.Inode.RemoveXattr(t, d, name) + if err := d.Inode.RemoveXattr(t, d, name); err != nil { + return err + } + d.InotifyEvent(linux.IN_ATTRIB, 0) + return nil } // LINT.ThenChange(vfs2/xattr.go) diff --git a/pkg/sentry/syscalls/linux/vfs2/splice.go b/pkg/sentry/syscalls/linux/vfs2/splice.go index 8f3c22a02..945a364a7 100644 --- a/pkg/sentry/syscalls/linux/vfs2/splice.go +++ b/pkg/sentry/syscalls/linux/vfs2/splice.go @@ -187,6 +187,11 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if n == 0 { return 0, nil, err } + + // On Linux, inotify behavior is not very consistent with splice(2). We try + // our best to emulate Linux for very basic calls to splice, where for some + // reason, events are generated for output files, but not input files. + outFile.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) return uintptr(n), nil, nil } diff --git a/test/syscalls/linux/inotify.cc b/test/syscalls/linux/inotify.cc index ac5ef2e51..1d1a7171d 100644 --- a/test/syscalls/linux/inotify.cc +++ b/test/syscalls/linux/inotify.cc @@ -19,6 +19,7 @@ #include #include #include +#include #include #include @@ -1655,9 +1656,44 @@ TEST(Inotify, EpollNoDeadlock) { } } -TEST(Inotify, SpliceEvent) { - // TODO(gvisor.dev/issue/138): Implement splice in VFS2. - SKIP_IF(IsRunningOnGvisor() && !IsRunningWithVFS1()); +// On Linux, inotify behavior is not very consistent with splice(2). We try our +// best to emulate Linux for very basic calls to splice. +TEST(Inotify, SpliceOnWatchTarget) { + int pipes[2]; + ASSERT_THAT(pipe2(pipes, O_NONBLOCK), SyscallSucceeds()); + + const TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const FileDescriptor inotify_fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileWith( + dir.path(), "some content", TempPath::kDefaultFileMode)); + + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(file.path(), O_RDWR)); + const int dir_wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(inotify_fd.get(), dir.path(), IN_ALL_EVENTS)); + const int file_wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(inotify_fd.get(), file.path(), IN_ALL_EVENTS)); + + EXPECT_THAT(splice(fd.get(), nullptr, pipes[1], nullptr, 1, /*flags=*/0), + SyscallSucceedsWithValue(1)); + + // Surprisingly, events are not generated in Linux if we read from a file. + std::vector events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + ASSERT_THAT(events, Are({})); + + EXPECT_THAT(splice(pipes[0], nullptr, fd.get(), nullptr, 1, /*flags=*/0), + SyscallSucceedsWithValue(1)); + + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + ASSERT_THAT(events, Are({ + Event(IN_MODIFY, dir_wd, Basename(file.path())), + Event(IN_MODIFY, file_wd), + })); +} + +TEST(Inotify, SpliceOnInotifyFD) { int pipes[2]; ASSERT_THAT(pipe2(pipes, O_NONBLOCK), SyscallSucceeds()); @@ -1719,6 +1755,58 @@ TEST(Inotify, LinkOnOtherParent) { EXPECT_THAT(events, Are({})); } +TEST(Inotify, Xattr) { + // TODO(gvisor.dev/issue/1636): Support extended attributes in runsc gofer. + SKIP_IF(IsRunningOnGvisor()); + + const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); + const std::string path = file.path(); + const FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(path, O_RDWR)); + const FileDescriptor inotify_fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + const int wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(inotify_fd.get(), path, IN_ALL_EVENTS)); + + const char* cpath = path.c_str(); + const char* name = "user.test"; + int val = 123; + ASSERT_THAT(setxattr(cpath, name, &val, sizeof(val), /*flags=*/0), + SyscallSucceeds()); + std::vector events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({Event(IN_ATTRIB, wd)})); + + ASSERT_THAT(getxattr(cpath, name, &val, sizeof(val)), SyscallSucceeds()); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({})); + + char list[100]; + ASSERT_THAT(listxattr(cpath, list, sizeof(list)), SyscallSucceeds()); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({})); + + ASSERT_THAT(removexattr(cpath, name), SyscallSucceeds()); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({Event(IN_ATTRIB, wd)})); + + ASSERT_THAT(fsetxattr(fd.get(), name, &val, sizeof(val), /*flags=*/0), + SyscallSucceeds()); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({Event(IN_ATTRIB, wd)})); + + ASSERT_THAT(fgetxattr(fd.get(), name, &val, sizeof(val)), SyscallSucceeds()); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({})); + + ASSERT_THAT(flistxattr(fd.get(), list, sizeof(list)), SyscallSucceeds()); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({})); + + ASSERT_THAT(fremovexattr(fd.get(), name), SyscallSucceeds()); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({Event(IN_ATTRIB, wd)})); +} + TEST(Inotify, Exec) { const TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); const TempPath bin = ASSERT_NO_ERRNO_AND_VALUE( -- cgit v1.2.3 From 5173c960212a2d78d69e94f028629cbcac8c04f9 Mon Sep 17 00:00:00 2001 From: Ridwan Sharif Date: Tue, 2 Jun 2020 22:08:23 +0000 Subject: Add some detail to milestone #1 This change adds more information about what needs to be done to implement `/dev/fuse` --- pkg/sentry/fs/g3doc/fuse.md | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/g3doc/fuse.md b/pkg/sentry/fs/g3doc/fuse.md index b43c082a7..2ca84dd74 100644 --- a/pkg/sentry/fs/g3doc/fuse.md +++ b/pkg/sentry/fs/g3doc/fuse.md @@ -76,7 +76,8 @@ ops can be implemented in parallel. #### Minimal client that can mount a trivial FUSE filesystem. -- Implement `/dev/fuse`. +- Implement `/dev/fuse` - a character device used to establish an FD for + communication between the sentry and the server daemon. - Implement basic FUSE ops like `FUSE_INIT`, `FUSE_DESTROY`. -- cgit v1.2.3 From 43279c6734109eec2725513e3ada93342f4de8d2 Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Tue, 2 Jun 2020 15:23:36 -0700 Subject: Fix typos PiperOrigin-RevId: 314415253 --- pkg/sentry/watchdog/watchdog.go | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/watchdog/watchdog.go b/pkg/sentry/watchdog/watchdog.go index 101497ed6..26ce1077f 100644 --- a/pkg/sentry/watchdog/watchdog.go +++ b/pkg/sentry/watchdog/watchdog.go @@ -328,8 +328,8 @@ func (w *Watchdog) reportStuckWatchdog() { } // doAction will take the given action. If the action is LogWarning, the stack -// is not always dumpped to the log to prevent log flooding. "forceStack" -// guarantees that the stack will be dumped regarless. +// is not always dumped to the log to prevent log flooding. "forceStack" +// guarantees that the stack will be dumped regardless. func (w *Watchdog) doAction(action Action, forceStack bool, msg *bytes.Buffer) { switch action { case LogWarning: -- cgit v1.2.3 From e6334e81ca8d951e56f03d8ea0629e3c85556cf1 Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Tue, 2 Jun 2020 19:17:53 -0700 Subject: Check that two sockets with different types can't be connected to each other PiperOrigin-RevId: 314450191 --- pkg/sentry/socket/unix/transport/connectioned.go | 2 +- pkg/sentry/socket/unix/unix.go | 13 +++++++- test/syscalls/linux/accept_bind.cc | 42 ++++++++++++++++++++++++ 3 files changed, 55 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/unix/transport/connectioned.go b/pkg/sentry/socket/unix/transport/connectioned.go index ce5b94ee7..09c6d3b27 100644 --- a/pkg/sentry/socket/unix/transport/connectioned.go +++ b/pkg/sentry/socket/unix/transport/connectioned.go @@ -252,7 +252,7 @@ func (e *connectionedEndpoint) Close() { // BidirectionalConnect implements BoundEndpoint.BidirectionalConnect. func (e *connectionedEndpoint) BidirectionalConnect(ctx context.Context, ce ConnectingEndpoint, returnConnect func(Receiver, ConnectedEndpoint)) *syserr.Error { if ce.Type() != e.stype { - return syserr.ErrConnectionRefused + return syserr.ErrWrongProtocolForSocket } // Check if ce is e to avoid a deadlock. diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go index 5b29e9d7f..c4c9db81b 100644 --- a/pkg/sentry/socket/unix/unix.go +++ b/pkg/sentry/socket/unix/unix.go @@ -417,7 +417,18 @@ func (s *socketOpsCommon) Connect(t *kernel.Task, sockaddr []byte, blocking bool defer ep.Release() // Connect the server endpoint. - return s.ep.Connect(t, ep) + err = s.ep.Connect(t, ep) + + if err == syserr.ErrWrongProtocolForSocket { + // Linux for abstract sockets returns ErrConnectionRefused + // instead of ErrWrongProtocolForSocket. + path, _ := extractPath(sockaddr) + if len(path) > 0 && path[0] == 0 { + err = syserr.ErrConnectionRefused + } + } + + return err } // Write implements fs.FileOperations.Write. diff --git a/test/syscalls/linux/accept_bind.cc b/test/syscalls/linux/accept_bind.cc index e08c578f0..f65a14fb8 100644 --- a/test/syscalls/linux/accept_bind.cc +++ b/test/syscalls/linux/accept_bind.cc @@ -13,6 +13,7 @@ // limitations under the License. #include +#include #include #include @@ -141,6 +142,47 @@ TEST_P(AllSocketPairTest, Connect) { SyscallSucceeds()); } +TEST_P(AllSocketPairTest, ConnectWithWrongType) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + int type; + socklen_t typelen = sizeof(type); + EXPECT_THAT( + getsockopt(sockets->first_fd(), SOL_SOCKET, SO_TYPE, &type, &typelen), + SyscallSucceeds()); + switch (type) { + case SOCK_STREAM: + type = SOCK_SEQPACKET; + break; + case SOCK_SEQPACKET: + type = SOCK_STREAM; + break; + } + + const FileDescriptor another_socket = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_UNIX, type, 0)); + + ASSERT_THAT(bind(sockets->first_fd(), sockets->first_addr(), + sockets->first_addr_size()), + SyscallSucceeds()); + + ASSERT_THAT(listen(sockets->first_fd(), 5), SyscallSucceeds()); + + if (sockets->first_addr()->sa_data[0] != 0) { + ASSERT_THAT(connect(another_socket.get(), sockets->first_addr(), + sockets->first_addr_size()), + SyscallFailsWithErrno(EPROTOTYPE)); + } else { + ASSERT_THAT(connect(another_socket.get(), sockets->first_addr(), + sockets->first_addr_size()), + SyscallFailsWithErrno(ECONNREFUSED)); + } + + ASSERT_THAT(connect(sockets->second_fd(), sockets->first_addr(), + sockets->first_addr_size()), + SyscallSucceeds()); +} + TEST_P(AllSocketPairTest, ConnectNonListening) { auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); -- cgit v1.2.3 From c8e79683891db9e780449112f78ee4004a2de833 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Wed, 3 Jun 2020 11:01:34 -0700 Subject: Take Mount reference in VFS.connectLocked. Updates #179 PiperOrigin-RevId: 314563830 --- pkg/sentry/vfs/README.md | 4 ++-- pkg/sentry/vfs/mount.go | 16 ++++++++++++++-- 2 files changed, 16 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/vfs/README.md b/pkg/sentry/vfs/README.md index 9aa133bcb..66f3105bd 100644 --- a/pkg/sentry/vfs/README.md +++ b/pkg/sentry/vfs/README.md @@ -39,8 +39,8 @@ Mount references are held by: - Mount: Each referenced Mount holds a reference on its parent, which is the mount containing its mount point. -- VirtualFilesystem: A reference is held on each Mount that has not been - umounted. +- VirtualFilesystem: A reference is held on each Mount that has been connected + to a mount point, but not yet umounted. MountNamespace and FileDescription references are held by users of VFS. The expectation is that each `kernel.Task` holds a reference on its corresponding diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index e4ac6524b..3adb7c97d 100644 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go @@ -265,8 +265,8 @@ func (vfs *VirtualFilesystem) MountAt(ctx context.Context, creds *auth.Credentia if err != nil { return err } + defer mnt.DecRef() if err := vfs.ConnectMountAt(ctx, creds, mnt, target); err != nil { - mnt.DecRef() return err } return nil @@ -394,8 +394,15 @@ func (vfs *VirtualFilesystem) umountRecursiveLocked(mnt *Mount, opts *umountRecu // references held by vd. // // Preconditions: vfs.mountMu must be locked. vfs.mounts.seq must be in a -// writer critical section. d.mu must be locked. mnt.parent() == nil. +// writer critical section. d.mu must be locked. mnt.parent() == nil, i.e. mnt +// must not already be connected. func (vfs *VirtualFilesystem) connectLocked(mnt *Mount, vd VirtualDentry, mntns *MountNamespace) { + if checkInvariants { + if mnt.parent() != nil { + panic("VFS.connectLocked called on connected mount") + } + } + mnt.IncRef() // dropped by callers of umountRecursiveLocked mnt.storeKey(vd) if vd.mount.children == nil { vd.mount.children = make(map[*Mount]struct{}) @@ -420,6 +427,11 @@ func (vfs *VirtualFilesystem) connectLocked(mnt *Mount, vd VirtualDentry, mntns // writer critical section. mnt.parent() != nil. func (vfs *VirtualFilesystem) disconnectLocked(mnt *Mount) VirtualDentry { vd := mnt.loadKey() + if checkInvariants { + if vd.mount != nil { + panic("VFS.disconnectLocked called on disconnected mount") + } + } mnt.storeKey(VirtualDentry{}) delete(vd.mount.children, mnt) atomic.AddUint32(&vd.dentry.mounts, math.MaxUint32) // -1 -- cgit v1.2.3 From b2e2a081a8a180764677111ae3c0b6179be81d31 Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Wed, 3 Jun 2020 11:33:39 -0700 Subject: Add metric for startup watchdog timeout PiperOrigin-RevId: 314570894 --- pkg/sentry/watchdog/watchdog.go | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/watchdog/watchdog.go b/pkg/sentry/watchdog/watchdog.go index 26ce1077f..e2894f9f5 100644 --- a/pkg/sentry/watchdog/watchdog.go +++ b/pkg/sentry/watchdog/watchdog.go @@ -77,7 +77,10 @@ var DefaultOpts = Opts{ // trigger it. const descheduleThreshold = 1 * time.Second -var stuckTasks = metric.MustCreateNewUint64Metric("/watchdog/stuck_tasks_detected", true /* sync */, "Cumulative count of stuck tasks detected") +var ( + stuckStartup = metric.MustCreateNewUint64Metric("/watchdog/stuck_startup_detected", true /* sync */, "Incremented once on startup watchdog timeout") + stuckTasks = metric.MustCreateNewUint64Metric("/watchdog/stuck_tasks_detected", true /* sync */, "Cumulative count of stuck tasks detected") +) // Amount of time to wait before dumping the stack to the log again when the same task(s) remains stuck. var stackDumpSameTaskPeriod = time.Minute @@ -220,6 +223,9 @@ func (w *Watchdog) waitForStart() { // We are fine. return } + + stuckStartup.Increment() + var buf bytes.Buffer buf.WriteString(fmt.Sprintf("Watchdog.Start() not called within %s", w.StartupTimeout)) w.doAction(w.StartupTimeoutAction, false, &buf) -- cgit v1.2.3 From 7da69fe9719b26f5d15e80b2764459c4cac80b0e Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Wed, 3 Jun 2020 12:33:51 -0700 Subject: Fix data race on f.offset. We must hold f.mu to write f.offset. PiperOrigin-RevId: 314582968 --- pkg/sentry/fs/file.go | 9 ++++----- pkg/sentry/fs/fs.go | 3 ++- 2 files changed, 6 insertions(+), 6 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/file.go b/pkg/sentry/fs/file.go index 846252c89..2a278fbe3 100644 --- a/pkg/sentry/fs/file.go +++ b/pkg/sentry/fs/file.go @@ -310,7 +310,6 @@ func (f *File) Writev(ctx context.Context, src usermem.IOSequence) (int64, error if !f.mu.Lock(ctx) { return 0, syserror.ErrInterrupted } - unlockAppendMu := f.Dirent.Inode.lockAppendMu(f.Flags().Append) // Handle append mode. if f.Flags().Append { @@ -355,7 +354,6 @@ func (f *File) Pwritev(ctx context.Context, src usermem.IOSequence, offset int64 // offset." unlockAppendMu := f.Dirent.Inode.lockAppendMu(f.Flags().Append) defer unlockAppendMu() - if f.Flags().Append { if err := f.offsetForAppend(ctx, &offset); err != nil { return 0, err @@ -374,9 +372,10 @@ func (f *File) Pwritev(ctx context.Context, src usermem.IOSequence, offset int64 return f.FileOperations.Write(ctx, f, src, offset) } -// offsetForAppend sets the given offset to the end of the file. +// offsetForAppend atomically sets the given offset to the end of the file. // -// Precondition: the file.Dirent.Inode.appendMu mutex should be held for writing. +// Precondition: the file.Dirent.Inode.appendMu mutex should be held for +// writing. func (f *File) offsetForAppend(ctx context.Context, offset *int64) error { uattr, err := f.Dirent.Inode.UnstableAttr(ctx) if err != nil { @@ -386,7 +385,7 @@ func (f *File) offsetForAppend(ctx context.Context, offset *int64) error { } // Update the offset. - *offset = uattr.Size + atomic.StoreInt64(offset, uattr.Size) return nil } diff --git a/pkg/sentry/fs/fs.go b/pkg/sentry/fs/fs.go index bdba6efe5..d2dbff268 100644 --- a/pkg/sentry/fs/fs.go +++ b/pkg/sentry/fs/fs.go @@ -42,9 +42,10 @@ // Dirent.dirMu // Dirent.mu // DirentCache.mu -// Locks in InodeOperations implementations or overlayEntry // Inode.Watches.mu (see `Inotify` for other lock ordering) // MountSource.mu +// Inode.appendMu +// Locks in InodeOperations implementations or overlayEntry // // If multiple Dirent or MountSource locks must be taken, locks in the parent must be // taken before locks in their children. -- cgit v1.2.3 From d3a8bffe04595910714ec67231585bc33dab2b5b Mon Sep 17 00:00:00 2001 From: Ting-Yu Wang Date: Wed, 3 Jun 2020 14:57:57 -0700 Subject: Pass PacketBuffer as pointer. Historically we've been passing PacketBuffer by shallow copying through out the stack. Right now, this is only correct as the caller would not use PacketBuffer after passing into the next layer in netstack. With new buffer management effort in gVisor/netstack, PacketBuffer will own a Buffer (to be added). Internally, both PacketBuffer and Buffer may have pointers and shallow copying shouldn't be used. Updates #2404. PiperOrigin-RevId: 314610879 --- pkg/sentry/socket/netfilter/owner_matcher.go | 2 +- pkg/sentry/socket/netfilter/tcp_matcher.go | 2 +- pkg/sentry/socket/netfilter/udp_matcher.go | 2 +- pkg/tcpip/link/channel/channel.go | 8 ++--- pkg/tcpip/link/fdbased/endpoint.go | 4 +-- pkg/tcpip/link/fdbased/endpoint_test.go | 10 +++---- pkg/tcpip/link/fdbased/mmap.go | 2 +- pkg/tcpip/link/fdbased/packet_dispatchers.go | 4 +-- pkg/tcpip/link/loopback/loopback.go | 6 ++-- pkg/tcpip/link/muxed/injectable.go | 4 +-- pkg/tcpip/link/muxed/injectable_test.go | 4 +-- pkg/tcpip/link/qdisc/fifo/endpoint.go | 6 ++-- pkg/tcpip/link/sharedmem/sharedmem.go | 4 +-- pkg/tcpip/link/sharedmem/sharedmem_test.go | 26 ++++++++--------- pkg/tcpip/link/sniffer/sniffer.go | 8 ++--- pkg/tcpip/link/tun/device.go | 2 +- pkg/tcpip/link/waitable/waitable.go | 4 +-- pkg/tcpip/link/waitable/waitable_test.go | 16 +++++----- pkg/tcpip/network/arp/arp.go | 10 +++---- pkg/tcpip/network/arp/arp_test.go | 2 +- pkg/tcpip/network/ip_test.go | 30 ++++++++++++------- pkg/tcpip/network/ipv4/icmp.go | 12 +++++--- pkg/tcpip/network/ipv4/ipv4.go | 34 +++++++++++----------- pkg/tcpip/network/ipv4/ipv4_test.go | 28 +++++++++++------- pkg/tcpip/network/ipv6/icmp.go | 10 +++---- pkg/tcpip/network/ipv6/icmp_test.go | 14 ++++----- pkg/tcpip/network/ipv6/ipv6.go | 8 ++--- pkg/tcpip/network/ipv6/ipv6_test.go | 8 ++--- pkg/tcpip/network/ipv6/ndp_test.go | 10 +++---- pkg/tcpip/stack/conntrack.go | 4 +-- pkg/tcpip/stack/forwarder.go | 4 +-- pkg/tcpip/stack/forwarder_test.go | 34 +++++++++++----------- pkg/tcpip/stack/iptables.go | 2 +- pkg/tcpip/stack/iptables_types.go | 2 +- pkg/tcpip/stack/ndp.go | 4 +-- pkg/tcpip/stack/ndp_test.go | 14 ++++----- pkg/tcpip/stack/nic.go | 12 ++++---- pkg/tcpip/stack/nic_test.go | 2 +- pkg/tcpip/stack/packet_buffer.go | 33 +++++++++++++++++++-- pkg/tcpip/stack/registration.go | 26 ++++++++--------- pkg/tcpip/stack/route.go | 4 +-- pkg/tcpip/stack/stack.go | 4 +-- pkg/tcpip/stack/stack_test.go | 26 ++++++++--------- pkg/tcpip/stack/transport_demuxer.go | 14 ++++----- pkg/tcpip/stack/transport_demuxer_test.go | 4 +-- pkg/tcpip/stack/transport_test.go | 22 +++++++------- pkg/tcpip/transport/icmp/endpoint.go | 8 ++--- pkg/tcpip/transport/icmp/protocol.go | 2 +- pkg/tcpip/transport/packet/endpoint.go | 2 +- pkg/tcpip/transport/raw/endpoint.go | 6 ++-- pkg/tcpip/transport/tcp/connect.go | 4 +-- pkg/tcpip/transport/tcp/dispatcher.go | 2 +- pkg/tcpip/transport/tcp/endpoint.go | 4 +-- pkg/tcpip/transport/tcp/forwarder.go | 2 +- pkg/tcpip/transport/tcp/protocol.go | 4 +-- pkg/tcpip/transport/tcp/segment.go | 2 +- pkg/tcpip/transport/tcp/testing/context/context.go | 10 +++---- pkg/tcpip/transport/udp/endpoint.go | 10 +++++-- pkg/tcpip/transport/udp/forwarder.go | 4 +-- pkg/tcpip/transport/udp/protocol.go | 6 ++-- pkg/tcpip/transport/udp/udp_test.go | 4 +-- 61 files changed, 306 insertions(+), 255 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/owner_matcher.go b/pkg/sentry/socket/netfilter/owner_matcher.go index 3863293c7..1b4e0ad79 100644 --- a/pkg/sentry/socket/netfilter/owner_matcher.go +++ b/pkg/sentry/socket/netfilter/owner_matcher.go @@ -111,7 +111,7 @@ func (*OwnerMatcher) Name() string { } // Match implements Matcher.Match. -func (om *OwnerMatcher) Match(hook stack.Hook, pkt stack.PacketBuffer, interfaceName string) (bool, bool) { +func (om *OwnerMatcher) Match(hook stack.Hook, pkt *stack.PacketBuffer, interfaceName string) (bool, bool) { // Support only for OUTPUT chain. // TODO(gvisor.dev/issue/170): Need to support for POSTROUTING chain also. if hook != stack.Output { diff --git a/pkg/sentry/socket/netfilter/tcp_matcher.go b/pkg/sentry/socket/netfilter/tcp_matcher.go index 57a1e1c12..ebabdf334 100644 --- a/pkg/sentry/socket/netfilter/tcp_matcher.go +++ b/pkg/sentry/socket/netfilter/tcp_matcher.go @@ -96,7 +96,7 @@ func (*TCPMatcher) Name() string { } // Match implements Matcher.Match. -func (tm *TCPMatcher) Match(hook stack.Hook, pkt stack.PacketBuffer, interfaceName string) (bool, bool) { +func (tm *TCPMatcher) Match(hook stack.Hook, pkt *stack.PacketBuffer, interfaceName string) (bool, bool) { netHeader := header.IPv4(pkt.NetworkHeader) if netHeader.TransportProtocol() != header.TCPProtocolNumber { diff --git a/pkg/sentry/socket/netfilter/udp_matcher.go b/pkg/sentry/socket/netfilter/udp_matcher.go index cfa9e621d..98b9943f8 100644 --- a/pkg/sentry/socket/netfilter/udp_matcher.go +++ b/pkg/sentry/socket/netfilter/udp_matcher.go @@ -93,7 +93,7 @@ func (*UDPMatcher) Name() string { } // Match implements Matcher.Match. -func (um *UDPMatcher) Match(hook stack.Hook, pkt stack.PacketBuffer, interfaceName string) (bool, bool) { +func (um *UDPMatcher) Match(hook stack.Hook, pkt *stack.PacketBuffer, interfaceName string) (bool, bool) { netHeader := header.IPv4(pkt.NetworkHeader) // TODO(gvisor.dev/issue/170): Proto checks should ultimately be moved diff --git a/pkg/tcpip/link/channel/channel.go b/pkg/tcpip/link/channel/channel.go index 5eb78b398..20b183da0 100644 --- a/pkg/tcpip/link/channel/channel.go +++ b/pkg/tcpip/link/channel/channel.go @@ -181,12 +181,12 @@ func (e *Endpoint) NumQueued() int { } // InjectInbound injects an inbound packet. -func (e *Endpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { +func (e *Endpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { e.InjectLinkAddr(protocol, "", pkt) } // InjectLinkAddr injects an inbound packet with a remote link address. -func (e *Endpoint) InjectLinkAddr(protocol tcpip.NetworkProtocolNumber, remote tcpip.LinkAddress, pkt stack.PacketBuffer) { +func (e *Endpoint) InjectLinkAddr(protocol tcpip.NetworkProtocolNumber, remote tcpip.LinkAddress, pkt *stack.PacketBuffer) { e.dispatcher.DeliverNetworkPacket(remote, "" /* local */, protocol, pkt) } @@ -229,13 +229,13 @@ func (e *Endpoint) LinkAddress() tcpip.LinkAddress { } // WritePacket stores outbound packets into the channel. -func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { +func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { // Clone r then release its resource so we only get the relevant fields from // stack.Route without holding a reference to a NIC's endpoint. route := r.Clone() route.Release() p := PacketInfo{ - Pkt: &pkt, + Pkt: pkt, Proto: protocol, GSO: gso, Route: route, diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go index 5ee508d48..f34082e1a 100644 --- a/pkg/tcpip/link/fdbased/endpoint.go +++ b/pkg/tcpip/link/fdbased/endpoint.go @@ -387,7 +387,7 @@ const ( // WritePacket writes outbound packets to the file descriptor. If it is not // currently writable, the packet is dropped. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { if e.hdrSize > 0 { // Add ethernet header if needed. eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize)) @@ -641,7 +641,7 @@ func (e *InjectableEndpoint) Attach(dispatcher stack.NetworkDispatcher) { } // InjectInbound injects an inbound packet. -func (e *InjectableEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { +func (e *InjectableEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { e.dispatcher.DeliverNetworkPacket("" /* remote */, "" /* local */, protocol, pkt) } diff --git a/pkg/tcpip/link/fdbased/endpoint_test.go b/pkg/tcpip/link/fdbased/endpoint_test.go index 6f41a71a8..eaee7e5d7 100644 --- a/pkg/tcpip/link/fdbased/endpoint_test.go +++ b/pkg/tcpip/link/fdbased/endpoint_test.go @@ -45,7 +45,7 @@ const ( type packetInfo struct { raddr tcpip.LinkAddress proto tcpip.NetworkProtocolNumber - contents stack.PacketBuffer + contents *stack.PacketBuffer } type context struct { @@ -103,7 +103,7 @@ func (c *context) cleanup() { } } -func (c *context) DeliverNetworkPacket(remote tcpip.LinkAddress, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { +func (c *context) DeliverNetworkPacket(remote tcpip.LinkAddress, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { c.ch <- packetInfo{remote, protocol, pkt} } @@ -179,7 +179,7 @@ func testWritePacket(t *testing.T, plen int, eth bool, gsoMaxSize uint32, hash u L3HdrLen: header.IPv4MaximumHeaderSize, } } - if err := c.ep.WritePacket(r, gso, proto, stack.PacketBuffer{ + if err := c.ep.WritePacket(r, gso, proto, &stack.PacketBuffer{ Header: hdr, Data: payload.ToVectorisedView(), Hash: hash, @@ -295,7 +295,7 @@ func TestPreserveSrcAddress(t *testing.T) { // WritePacket panics given a prependable with anything less than // the minimum size of the ethernet header. hdr := buffer.NewPrependable(header.EthernetMinimumSize) - if err := c.ep.WritePacket(r, nil /* gso */, proto, stack.PacketBuffer{ + if err := c.ep.WritePacket(r, nil /* gso */, proto, &stack.PacketBuffer{ Header: hdr, Data: buffer.VectorisedView{}, }); err != nil { @@ -358,7 +358,7 @@ func TestDeliverPacket(t *testing.T) { want := packetInfo{ raddr: raddr, proto: proto, - contents: stack.PacketBuffer{ + contents: &stack.PacketBuffer{ Data: buffer.View(b).ToVectorisedView(), LinkHeader: buffer.View(hdr), }, diff --git a/pkg/tcpip/link/fdbased/mmap.go b/pkg/tcpip/link/fdbased/mmap.go index ca4229ed6..2dfd29aa9 100644 --- a/pkg/tcpip/link/fdbased/mmap.go +++ b/pkg/tcpip/link/fdbased/mmap.go @@ -191,7 +191,7 @@ func (d *packetMMapDispatcher) dispatch() (bool, *tcpip.Error) { } pkt = pkt[d.e.hdrSize:] - d.e.dispatcher.DeliverNetworkPacket(remote, local, p, stack.PacketBuffer{ + d.e.dispatcher.DeliverNetworkPacket(remote, local, p, &stack.PacketBuffer{ Data: buffer.View(pkt).ToVectorisedView(), LinkHeader: buffer.View(eth), }) diff --git a/pkg/tcpip/link/fdbased/packet_dispatchers.go b/pkg/tcpip/link/fdbased/packet_dispatchers.go index 26c96a655..f04738cfb 100644 --- a/pkg/tcpip/link/fdbased/packet_dispatchers.go +++ b/pkg/tcpip/link/fdbased/packet_dispatchers.go @@ -139,7 +139,7 @@ func (d *readVDispatcher) dispatch() (bool, *tcpip.Error) { } used := d.capViews(n, BufConfig) - pkt := stack.PacketBuffer{ + pkt := &stack.PacketBuffer{ Data: buffer.NewVectorisedView(n, append([]buffer.View(nil), d.views[:used]...)), LinkHeader: buffer.View(eth), } @@ -296,7 +296,7 @@ func (d *recvMMsgDispatcher) dispatch() (bool, *tcpip.Error) { } used := d.capViews(k, int(n), BufConfig) - pkt := stack.PacketBuffer{ + pkt := &stack.PacketBuffer{ Data: buffer.NewVectorisedView(int(n), append([]buffer.View(nil), d.views[k][:used]...)), LinkHeader: buffer.View(eth), } diff --git a/pkg/tcpip/link/loopback/loopback.go b/pkg/tcpip/link/loopback/loopback.go index 20d9e95f6..568c6874f 100644 --- a/pkg/tcpip/link/loopback/loopback.go +++ b/pkg/tcpip/link/loopback/loopback.go @@ -76,7 +76,7 @@ func (*endpoint) Wait() {} // WritePacket implements stack.LinkEndpoint.WritePacket. It delivers outbound // packets to the network-layer dispatcher. -func (e *endpoint) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { +func (e *endpoint) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { views := make([]buffer.View, 1, 1+len(pkt.Data.Views())) views[0] = pkt.Header.View() views = append(views, pkt.Data.Views()...) @@ -84,7 +84,7 @@ func (e *endpoint) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.Netw // Because we're immediately turning around and writing the packet back // to the rx path, we intentionally don't preserve the remote and local // link addresses from the stack.Route we're passed. - e.dispatcher.DeliverNetworkPacket("" /* remote */, "" /* local */, protocol, stack.PacketBuffer{ + e.dispatcher.DeliverNetworkPacket("" /* remote */, "" /* local */, protocol, &stack.PacketBuffer{ Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views), }) @@ -106,7 +106,7 @@ func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { } linkHeader := header.Ethernet(hdr) vv.TrimFront(len(linkHeader)) - e.dispatcher.DeliverNetworkPacket("" /* remote */, "" /* local */, linkHeader.Type(), stack.PacketBuffer{ + e.dispatcher.DeliverNetworkPacket("" /* remote */, "" /* local */, linkHeader.Type(), &stack.PacketBuffer{ Data: vv, LinkHeader: buffer.View(linkHeader), }) diff --git a/pkg/tcpip/link/muxed/injectable.go b/pkg/tcpip/link/muxed/injectable.go index f0769830a..c69d6b7e9 100644 --- a/pkg/tcpip/link/muxed/injectable.go +++ b/pkg/tcpip/link/muxed/injectable.go @@ -80,7 +80,7 @@ func (m *InjectableEndpoint) IsAttached() bool { } // InjectInbound implements stack.InjectableLinkEndpoint. -func (m *InjectableEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { +func (m *InjectableEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { m.dispatcher.DeliverNetworkPacket("" /* remote */, "" /* local */, protocol, pkt) } @@ -98,7 +98,7 @@ func (m *InjectableEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts s // WritePacket writes outbound packets to the appropriate LinkInjectableEndpoint // based on the RemoteAddress. HandleLocal only works if r.RemoteAddress has a // route registered in this endpoint. -func (m *InjectableEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { +func (m *InjectableEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { if endpoint, ok := m.routes[r.RemoteAddress]; ok { return endpoint.WritePacket(r, gso, protocol, pkt) } diff --git a/pkg/tcpip/link/muxed/injectable_test.go b/pkg/tcpip/link/muxed/injectable_test.go index 87c734c1f..0744f66d6 100644 --- a/pkg/tcpip/link/muxed/injectable_test.go +++ b/pkg/tcpip/link/muxed/injectable_test.go @@ -50,7 +50,7 @@ func TestInjectableEndpointDispatch(t *testing.T) { hdr.Prepend(1)[0] = 0xFA packetRoute := stack.Route{RemoteAddress: dstIP} - endpoint.WritePacket(&packetRoute, nil /* gso */, ipv4.ProtocolNumber, stack.PacketBuffer{ + endpoint.WritePacket(&packetRoute, nil /* gso */, ipv4.ProtocolNumber, &stack.PacketBuffer{ Header: hdr, Data: buffer.NewViewFromBytes([]byte{0xFB}).ToVectorisedView(), }) @@ -70,7 +70,7 @@ func TestInjectableEndpointDispatchHdrOnly(t *testing.T) { hdr := buffer.NewPrependable(1) hdr.Prepend(1)[0] = 0xFA packetRoute := stack.Route{RemoteAddress: dstIP} - endpoint.WritePacket(&packetRoute, nil /* gso */, ipv4.ProtocolNumber, stack.PacketBuffer{ + endpoint.WritePacket(&packetRoute, nil /* gso */, ipv4.ProtocolNumber, &stack.PacketBuffer{ Header: hdr, Data: buffer.NewView(0).ToVectorisedView(), }) diff --git a/pkg/tcpip/link/qdisc/fifo/endpoint.go b/pkg/tcpip/link/qdisc/fifo/endpoint.go index ec5c5048a..b5dfb7850 100644 --- a/pkg/tcpip/link/qdisc/fifo/endpoint.go +++ b/pkg/tcpip/link/qdisc/fifo/endpoint.go @@ -102,7 +102,7 @@ func (q *queueDispatcher) dispatchLoop() { } // DeliverNetworkPacket implements stack.NetworkDispatcher.DeliverNetworkPacket. -func (e *endpoint) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { +func (e *endpoint) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { e.dispatcher.DeliverNetworkPacket(remote, local, protocol, pkt) } @@ -146,7 +146,7 @@ func (e *endpoint) GSOMaxSize() uint32 { } // WritePacket implements stack.LinkEndpoint.WritePacket. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { // WritePacket caller's do not set the following fields in PacketBuffer // so we populate them here. newRoute := r.Clone() @@ -154,7 +154,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne pkt.GSOOptions = gso pkt.NetworkProtocolNumber = protocol d := e.dispatchers[int(pkt.Hash)%len(e.dispatchers)] - if !d.q.enqueue(&pkt) { + if !d.q.enqueue(pkt) { return tcpip.ErrNoBufferSpace } d.newPacketWaker.Assert() diff --git a/pkg/tcpip/link/sharedmem/sharedmem.go b/pkg/tcpip/link/sharedmem/sharedmem.go index f5dec0a7f..0374a2441 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem.go +++ b/pkg/tcpip/link/sharedmem/sharedmem.go @@ -185,7 +185,7 @@ func (e *endpoint) LinkAddress() tcpip.LinkAddress { // WritePacket writes outbound packets to the file descriptor. If it is not // currently writable, the packet is dropped. -func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { +func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { // Add the ethernet header here. eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize)) pkt.LinkHeader = buffer.View(eth) @@ -275,7 +275,7 @@ func (e *endpoint) dispatchLoop(d stack.NetworkDispatcher) { // Send packet up the stack. eth := header.Ethernet(b[:header.EthernetMinimumSize]) - d.DeliverNetworkPacket(eth.SourceAddress(), eth.DestinationAddress(), eth.Type(), stack.PacketBuffer{ + d.DeliverNetworkPacket(eth.SourceAddress(), eth.DestinationAddress(), eth.Type(), &stack.PacketBuffer{ Data: buffer.View(b[header.EthernetMinimumSize:]).ToVectorisedView(), LinkHeader: buffer.View(eth), }) diff --git a/pkg/tcpip/link/sharedmem/sharedmem_test.go b/pkg/tcpip/link/sharedmem/sharedmem_test.go index f3fc62607..28a2e88ba 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem_test.go +++ b/pkg/tcpip/link/sharedmem/sharedmem_test.go @@ -131,7 +131,7 @@ func newTestContext(t *testing.T, mtu, bufferSize uint32, addr tcpip.LinkAddress return c } -func (c *testContext) DeliverNetworkPacket(remoteLinkAddr, localLinkAddr tcpip.LinkAddress, proto tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { +func (c *testContext) DeliverNetworkPacket(remoteLinkAddr, localLinkAddr tcpip.LinkAddress, proto tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { c.mu.Lock() c.packets = append(c.packets, packetInfo{ addr: remoteLinkAddr, @@ -273,7 +273,7 @@ func TestSimpleSend(t *testing.T) { randomFill(buf) proto := tcpip.NetworkProtocolNumber(rand.Intn(0x10000)) - if err := c.ep.WritePacket(&r, nil /* gso */, proto, stack.PacketBuffer{ + if err := c.ep.WritePacket(&r, nil /* gso */, proto, &stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != nil { @@ -345,7 +345,7 @@ func TestPreserveSrcAddressInSend(t *testing.T) { hdr := buffer.NewPrependable(header.EthernetMinimumSize) proto := tcpip.NetworkProtocolNumber(rand.Intn(0x10000)) - if err := c.ep.WritePacket(&r, nil /* gso */, proto, stack.PacketBuffer{ + if err := c.ep.WritePacket(&r, nil /* gso */, proto, &stack.PacketBuffer{ Header: hdr, }); err != nil { t.Fatalf("WritePacket failed: %v", err) @@ -401,7 +401,7 @@ func TestFillTxQueue(t *testing.T) { for i := queuePipeSize / 40; i > 0; i-- { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, &stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != nil { @@ -419,7 +419,7 @@ func TestFillTxQueue(t *testing.T) { // Next attempt to write must fail. hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ + if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, &stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != want { @@ -447,7 +447,7 @@ func TestFillTxQueueAfterBadCompletion(t *testing.T) { // Send two packets so that the id slice has at least two slots. for i := 2; i > 0; i-- { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, &stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != nil { @@ -470,7 +470,7 @@ func TestFillTxQueueAfterBadCompletion(t *testing.T) { ids := make(map[uint64]struct{}) for i := queuePipeSize / 40; i > 0; i-- { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, &stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != nil { @@ -488,7 +488,7 @@ func TestFillTxQueueAfterBadCompletion(t *testing.T) { // Next attempt to write must fail. hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ + if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, &stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != want { @@ -514,7 +514,7 @@ func TestFillTxMemory(t *testing.T) { ids := make(map[uint64]struct{}) for i := queueDataSize / bufferSize; i > 0; i-- { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, &stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != nil { @@ -533,7 +533,7 @@ func TestFillTxMemory(t *testing.T) { // Next attempt to write must fail. hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ + err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, &stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }) @@ -561,7 +561,7 @@ func TestFillTxMemoryWithMultiBuffer(t *testing.T) { // until there is only one buffer left. for i := queueDataSize/bufferSize - 1; i > 0; i-- { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, &stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != nil { @@ -577,7 +577,7 @@ func TestFillTxMemoryWithMultiBuffer(t *testing.T) { { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) uu := buffer.NewView(bufferSize).ToVectorisedView() - if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ + if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, &stack.PacketBuffer{ Header: hdr, Data: uu, }); err != want { @@ -588,7 +588,7 @@ func TestFillTxMemoryWithMultiBuffer(t *testing.T) { // Attempt to write the one-buffer packet again. It must succeed. { hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, stack.PacketBuffer{ + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, &stack.PacketBuffer{ Header: hdr, Data: buf.ToVectorisedView(), }); err != nil { diff --git a/pkg/tcpip/link/sniffer/sniffer.go b/pkg/tcpip/link/sniffer/sniffer.go index b060d4627..ae3186314 100644 --- a/pkg/tcpip/link/sniffer/sniffer.go +++ b/pkg/tcpip/link/sniffer/sniffer.go @@ -120,8 +120,8 @@ func NewWithWriter(lower stack.LinkEndpoint, writer io.Writer, snapLen uint32) ( // DeliverNetworkPacket implements the stack.NetworkDispatcher interface. It is // called by the link-layer endpoint being wrapped when a packet arrives, and // logs the packet before forwarding to the actual dispatcher. -func (e *endpoint) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { - e.dumpPacket("recv", nil, protocol, &pkt) +func (e *endpoint) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + e.dumpPacket("recv", nil, protocol, pkt) e.dispatcher.DeliverNetworkPacket(remote, local, protocol, pkt) } @@ -208,8 +208,8 @@ func (e *endpoint) dumpPacket(prefix string, gso *stack.GSO, protocol tcpip.Netw // WritePacket implements the stack.LinkEndpoint interface. It is called by // higher-level protocols to write packets; it just logs the packet and // forwards the request to the lower endpoint. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { - e.dumpPacket("send", gso, protocol, &pkt) +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { + e.dumpPacket("send", gso, protocol, pkt) return e.lower.WritePacket(r, gso, protocol, pkt) } diff --git a/pkg/tcpip/link/tun/device.go b/pkg/tcpip/link/tun/device.go index 617446ea2..6bc9033d0 100644 --- a/pkg/tcpip/link/tun/device.go +++ b/pkg/tcpip/link/tun/device.go @@ -213,7 +213,7 @@ func (d *Device) Write(data []byte) (int64, error) { remote = tcpip.LinkAddress(zeroMAC[:]) } - pkt := stack.PacketBuffer{ + pkt := &stack.PacketBuffer{ Data: buffer.View(data).ToVectorisedView(), } if ethHdr != nil { diff --git a/pkg/tcpip/link/waitable/waitable.go b/pkg/tcpip/link/waitable/waitable.go index f5a05929f..949b3f2b2 100644 --- a/pkg/tcpip/link/waitable/waitable.go +++ b/pkg/tcpip/link/waitable/waitable.go @@ -50,7 +50,7 @@ func New(lower stack.LinkEndpoint) *Endpoint { // It is called by the link-layer endpoint being wrapped when a packet arrives, // and only forwards to the actual dispatcher if Wait or WaitDispatch haven't // been called. -func (e *Endpoint) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { +func (e *Endpoint) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { if !e.dispatchGate.Enter() { return } @@ -99,7 +99,7 @@ func (e *Endpoint) LinkAddress() tcpip.LinkAddress { // WritePacket implements stack.LinkEndpoint.WritePacket. It is called by // higher-level protocols to write packets. It only forwards packets to the // lower endpoint if Wait or WaitWrite haven't been called. -func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { +func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { if !e.writeGate.Enter() { return nil } diff --git a/pkg/tcpip/link/waitable/waitable_test.go b/pkg/tcpip/link/waitable/waitable_test.go index 0a9b99f18..63bf40562 100644 --- a/pkg/tcpip/link/waitable/waitable_test.go +++ b/pkg/tcpip/link/waitable/waitable_test.go @@ -35,7 +35,7 @@ type countedEndpoint struct { dispatcher stack.NetworkDispatcher } -func (e *countedEndpoint) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { +func (e *countedEndpoint) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { e.dispatchCount++ } @@ -65,7 +65,7 @@ func (e *countedEndpoint) LinkAddress() tcpip.LinkAddress { return e.linkAddr } -func (e *countedEndpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { +func (e *countedEndpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { e.writeCount++ return nil } @@ -89,21 +89,21 @@ func TestWaitWrite(t *testing.T) { wep := New(ep) // Write and check that it goes through. - wep.WritePacket(nil, nil /* gso */, 0, stack.PacketBuffer{}) + wep.WritePacket(nil, nil /* gso */, 0, &stack.PacketBuffer{}) if want := 1; ep.writeCount != want { t.Fatalf("Unexpected writeCount: got=%v, want=%v", ep.writeCount, want) } // Wait on dispatches, then try to write. It must go through. wep.WaitDispatch() - wep.WritePacket(nil, nil /* gso */, 0, stack.PacketBuffer{}) + wep.WritePacket(nil, nil /* gso */, 0, &stack.PacketBuffer{}) if want := 2; ep.writeCount != want { t.Fatalf("Unexpected writeCount: got=%v, want=%v", ep.writeCount, want) } // Wait on writes, then try to write. It must not go through. wep.WaitWrite() - wep.WritePacket(nil, nil /* gso */, 0, stack.PacketBuffer{}) + wep.WritePacket(nil, nil /* gso */, 0, &stack.PacketBuffer{}) if want := 2; ep.writeCount != want { t.Fatalf("Unexpected writeCount: got=%v, want=%v", ep.writeCount, want) } @@ -120,21 +120,21 @@ func TestWaitDispatch(t *testing.T) { } // Dispatch and check that it goes through. - ep.dispatcher.DeliverNetworkPacket("", "", 0, stack.PacketBuffer{}) + ep.dispatcher.DeliverNetworkPacket("", "", 0, &stack.PacketBuffer{}) if want := 1; ep.dispatchCount != want { t.Fatalf("Unexpected dispatchCount: got=%v, want=%v", ep.dispatchCount, want) } // Wait on writes, then try to dispatch. It must go through. wep.WaitWrite() - ep.dispatcher.DeliverNetworkPacket("", "", 0, stack.PacketBuffer{}) + ep.dispatcher.DeliverNetworkPacket("", "", 0, &stack.PacketBuffer{}) if want := 2; ep.dispatchCount != want { t.Fatalf("Unexpected dispatchCount: got=%v, want=%v", ep.dispatchCount, want) } // Wait on dispatches, then try to dispatch. It must not go through. wep.WaitDispatch() - ep.dispatcher.DeliverNetworkPacket("", "", 0, stack.PacketBuffer{}) + ep.dispatcher.DeliverNetworkPacket("", "", 0, &stack.PacketBuffer{}) if want := 2; ep.dispatchCount != want { t.Fatalf("Unexpected dispatchCount: got=%v, want=%v", ep.dispatchCount, want) } diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index 9d0797af7..ea1acba83 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -80,7 +80,7 @@ func (e *endpoint) MaxHeaderLength() uint16 { func (e *endpoint) Close() {} -func (e *endpoint) WritePacket(*stack.Route, *stack.GSO, stack.NetworkHeaderParams, stack.PacketBuffer) *tcpip.Error { +func (e *endpoint) WritePacket(*stack.Route, *stack.GSO, stack.NetworkHeaderParams, *stack.PacketBuffer) *tcpip.Error { return tcpip.ErrNotSupported } @@ -94,11 +94,11 @@ func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, stack.PacketBufferList return 0, tcpip.ErrNotSupported } -func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuffer) *tcpip.Error { +func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBuffer) *tcpip.Error { return tcpip.ErrNotSupported } -func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { +func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { v, ok := pkt.Data.PullUp(header.ARPSize) if !ok { return @@ -122,7 +122,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { copy(packet.ProtocolAddressSender(), h.ProtocolAddressTarget()) copy(packet.HardwareAddressTarget(), h.HardwareAddressSender()) copy(packet.ProtocolAddressTarget(), h.ProtocolAddressSender()) - e.linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, stack.PacketBuffer{ + e.linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, &stack.PacketBuffer{ Header: hdr, }) fallthrough // also fill the cache from requests @@ -177,7 +177,7 @@ func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, linkEP stack. copy(h.ProtocolAddressSender(), localAddr) copy(h.ProtocolAddressTarget(), addr) - return linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, stack.PacketBuffer{ + return linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, &stack.PacketBuffer{ Header: hdr, }) } diff --git a/pkg/tcpip/network/arp/arp_test.go b/pkg/tcpip/network/arp/arp_test.go index 1646d9cde..66e67429c 100644 --- a/pkg/tcpip/network/arp/arp_test.go +++ b/pkg/tcpip/network/arp/arp_test.go @@ -103,7 +103,7 @@ func TestDirectRequest(t *testing.T) { inject := func(addr tcpip.Address) { copy(h.ProtocolAddressTarget(), addr) - c.linkEP.InjectInbound(arp.ProtocolNumber, stack.PacketBuffer{ + c.linkEP.InjectInbound(arp.ProtocolNumber, &stack.PacketBuffer{ Data: v.ToVectorisedView(), }) } diff --git a/pkg/tcpip/network/ip_test.go b/pkg/tcpip/network/ip_test.go index 4c20301c6..d9b62f2db 100644 --- a/pkg/tcpip/network/ip_test.go +++ b/pkg/tcpip/network/ip_test.go @@ -96,7 +96,7 @@ func (t *testObject) checkValues(protocol tcpip.TransportProtocolNumber, vv buff // DeliverTransportPacket is called by network endpoints after parsing incoming // packets. This is used by the test object to verify that the results of the // parsing are expected. -func (t *testObject) DeliverTransportPacket(r *stack.Route, protocol tcpip.TransportProtocolNumber, pkt stack.PacketBuffer) { +func (t *testObject) DeliverTransportPacket(r *stack.Route, protocol tcpip.TransportProtocolNumber, pkt *stack.PacketBuffer) { t.checkValues(protocol, pkt.Data, r.RemoteAddress, r.LocalAddress) t.dataCalls++ } @@ -104,7 +104,7 @@ func (t *testObject) DeliverTransportPacket(r *stack.Route, protocol tcpip.Trans // DeliverTransportControlPacket is called by network endpoints after parsing // incoming control (ICMP) packets. This is used by the test object to verify // that the results of the parsing are expected. -func (t *testObject) DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ stack.ControlType, extra uint32, pkt stack.PacketBuffer) { +func (t *testObject) DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ stack.ControlType, extra uint32, pkt *stack.PacketBuffer) { t.checkValues(trans, pkt.Data, remote, local) if typ != t.typ { t.t.Errorf("typ = %v, want %v", typ, t.typ) @@ -150,7 +150,7 @@ func (*testObject) Wait() {} // WritePacket is called by network endpoints after producing a packet and // writing it to the link endpoint. This is used by the test object to verify // that the produced packet is as expected. -func (t *testObject) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { +func (t *testObject) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { var prot tcpip.TransportProtocolNumber var srcAddr tcpip.Address var dstAddr tcpip.Address @@ -246,7 +246,11 @@ func TestIPv4Send(t *testing.T) { if err != nil { t.Fatalf("could not find route: %v", err) } - if err := ep.WritePacket(&r, nil /* gso */, stack.NetworkHeaderParams{Protocol: 123, TTL: 123, TOS: stack.DefaultTOS}, stack.PacketBuffer{ + if err := ep.WritePacket(&r, nil /* gso */, stack.NetworkHeaderParams{ + Protocol: 123, + TTL: 123, + TOS: stack.DefaultTOS, + }, &stack.PacketBuffer{ Header: hdr, Data: payload.ToVectorisedView(), }); err != nil { @@ -289,7 +293,7 @@ func TestIPv4Receive(t *testing.T) { if err != nil { t.Fatalf("could not find route: %v", err) } - ep.HandlePacket(&r, stack.PacketBuffer{ + ep.HandlePacket(&r, &stack.PacketBuffer{ Data: view.ToVectorisedView(), }) if o.dataCalls != 1 { @@ -379,7 +383,7 @@ func TestIPv4ReceiveControl(t *testing.T) { o.extra = c.expectedExtra vv := view[:len(view)-c.trunc].ToVectorisedView() - ep.HandlePacket(&r, stack.PacketBuffer{ + ep.HandlePacket(&r, &stack.PacketBuffer{ Data: vv, }) if want := c.expectedCount; o.controlCalls != want { @@ -444,7 +448,7 @@ func TestIPv4FragmentationReceive(t *testing.T) { } // Send first segment. - ep.HandlePacket(&r, stack.PacketBuffer{ + ep.HandlePacket(&r, &stack.PacketBuffer{ Data: frag1.ToVectorisedView(), }) if o.dataCalls != 0 { @@ -452,7 +456,7 @@ func TestIPv4FragmentationReceive(t *testing.T) { } // Send second segment. - ep.HandlePacket(&r, stack.PacketBuffer{ + ep.HandlePacket(&r, &stack.PacketBuffer{ Data: frag2.ToVectorisedView(), }) if o.dataCalls != 1 { @@ -487,7 +491,11 @@ func TestIPv6Send(t *testing.T) { if err != nil { t.Fatalf("could not find route: %v", err) } - if err := ep.WritePacket(&r, nil /* gso */, stack.NetworkHeaderParams{Protocol: 123, TTL: 123, TOS: stack.DefaultTOS}, stack.PacketBuffer{ + if err := ep.WritePacket(&r, nil /* gso */, stack.NetworkHeaderParams{ + Protocol: 123, + TTL: 123, + TOS: stack.DefaultTOS, + }, &stack.PacketBuffer{ Header: hdr, Data: payload.ToVectorisedView(), }); err != nil { @@ -530,7 +538,7 @@ func TestIPv6Receive(t *testing.T) { t.Fatalf("could not find route: %v", err) } - ep.HandlePacket(&r, stack.PacketBuffer{ + ep.HandlePacket(&r, &stack.PacketBuffer{ Data: view.ToVectorisedView(), }) if o.dataCalls != 1 { @@ -644,7 +652,7 @@ func TestIPv6ReceiveControl(t *testing.T) { // Set ICMPv6 checksum. icmp.SetChecksum(header.ICMPv6Checksum(icmp, outerSrcAddr, localIpv6Addr, buffer.VectorisedView{})) - ep.HandlePacket(&r, stack.PacketBuffer{ + ep.HandlePacket(&r, &stack.PacketBuffer{ Data: view[:len(view)-c.trunc].ToVectorisedView(), }) if want := c.expectedCount; o.controlCalls != want { diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go index 4cbefe5ab..d1c3ae835 100644 --- a/pkg/tcpip/network/ipv4/icmp.go +++ b/pkg/tcpip/network/ipv4/icmp.go @@ -24,7 +24,7 @@ import ( // the original packet that caused the ICMP one to be sent. This information is // used to find out which transport endpoint must be notified about the ICMP // packet. -func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack.PacketBuffer) { +func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt *stack.PacketBuffer) { h, ok := pkt.Data.PullUp(header.IPv4MinimumSize) if !ok { return @@ -56,7 +56,7 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, hdr.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) } -func (e *endpoint) handleICMP(r *stack.Route, pkt stack.PacketBuffer) { +func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer) { stats := r.Stats() received := stats.ICMP.V4PacketsReceived v, ok := pkt.Data.PullUp(header.ICMPv4MinimumSize) @@ -88,7 +88,7 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt stack.PacketBuffer) { // It's possible that a raw socket expects to receive this. h.SetChecksum(wantChecksum) - e.dispatcher.DeliverTransportPacket(r, header.ICMPv4ProtocolNumber, stack.PacketBuffer{ + e.dispatcher.DeliverTransportPacket(r, header.ICMPv4ProtocolNumber, &stack.PacketBuffer{ Data: pkt.Data.Clone(nil), NetworkHeader: append(buffer.View(nil), pkt.NetworkHeader...), }) @@ -102,7 +102,11 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt stack.PacketBuffer) { pkt.SetChecksum(0) pkt.SetChecksum(^header.Checksum(pkt, header.ChecksumVV(vv, 0))) sent := stats.ICMP.V4PacketsSent - if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, stack.PacketBuffer{ + if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{ + Protocol: header.ICMPv4ProtocolNumber, + TTL: r.DefaultTTL(), + TOS: stack.DefaultTOS, + }, &stack.PacketBuffer{ Header: hdr, Data: vv, TransportHeader: buffer.View(pkt), diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index 64046cbbf..9cd7592f4 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -129,7 +129,7 @@ func (e *endpoint) NetworkProtocolNumber() tcpip.NetworkProtocolNumber { // packet's stated length matches the length of the header+payload. mtu // includes the IP header and options. This does not support the DontFragment // IP flag. -func (e *endpoint) writePacketFragments(r *stack.Route, gso *stack.GSO, mtu int, pkt stack.PacketBuffer) *tcpip.Error { +func (e *endpoint) writePacketFragments(r *stack.Route, gso *stack.GSO, mtu int, pkt *stack.PacketBuffer) *tcpip.Error { // This packet is too big, it needs to be fragmented. ip := header.IPv4(pkt.Header.View()) flags := ip.Flags() @@ -169,7 +169,7 @@ func (e *endpoint) writePacketFragments(r *stack.Route, gso *stack.GSO, mtu int, if i > 0 { newPayload := pkt.Data.Clone(nil) newPayload.CapLength(innerMTU) - if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, stack.PacketBuffer{ + if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, &stack.PacketBuffer{ Header: pkt.Header, Data: newPayload, NetworkHeader: buffer.View(h), @@ -188,7 +188,7 @@ func (e *endpoint) writePacketFragments(r *stack.Route, gso *stack.GSO, mtu int, newPayload := pkt.Data.Clone(nil) newPayloadLength := outerMTU - pkt.Header.UsedLength() newPayload.CapLength(newPayloadLength) - if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, stack.PacketBuffer{ + if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, &stack.PacketBuffer{ Header: pkt.Header, Data: newPayload, NetworkHeader: buffer.View(h), @@ -202,7 +202,7 @@ func (e *endpoint) writePacketFragments(r *stack.Route, gso *stack.GSO, mtu int, startOfHdr := pkt.Header startOfHdr.TrimBack(pkt.Header.UsedLength() - outerMTU) emptyVV := buffer.NewVectorisedView(0, []buffer.View{}) - if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, stack.PacketBuffer{ + if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, &stack.PacketBuffer{ Header: startOfHdr, Data: emptyVV, NetworkHeader: buffer.View(h), @@ -245,7 +245,7 @@ func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadS } // WritePacket writes a packet to the given destination address and protocol. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, pkt stack.PacketBuffer) *tcpip.Error { +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, pkt *stack.PacketBuffer) *tcpip.Error { ip := e.addIPHeader(r, &pkt.Header, pkt.Data.Size(), params) pkt.NetworkHeader = buffer.View(ip) @@ -253,7 +253,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw // iptables filtering. All packets that reach here are locally // generated. ipt := e.stack.IPTables() - if ok := ipt.Check(stack.Output, &pkt, gso, r, "", nicName); !ok { + if ok := ipt.Check(stack.Output, pkt, gso, r, "", nicName); !ok { // iptables is telling us to drop the packet. return nil } @@ -271,9 +271,9 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw views := make([]buffer.View, 1, 1+len(pkt.Data.Views())) views[0] = pkt.Header.View() views = append(views, pkt.Data.Views()...) - packet := stack.PacketBuffer{ - Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views)} - ep.HandlePacket(&route, packet) + ep.HandlePacket(&route, &stack.PacketBuffer{ + Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views), + }) return nil } } @@ -286,7 +286,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw views = append(views, pkt.Data.Views()...) loopedR := r.MakeLoopedRoute() - e.HandlePacket(&loopedR, stack.PacketBuffer{ + e.HandlePacket(&loopedR, &stack.PacketBuffer{ Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views), }) @@ -351,14 +351,14 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe views := make([]buffer.View, 1, 1+len(pkt.Data.Views())) views[0] = pkt.Header.View() views = append(views, pkt.Data.Views()...) - packet := stack.PacketBuffer{ - Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views)} - ep.HandlePacket(&route, packet) + ep.HandlePacket(&route, &stack.PacketBuffer{ + Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views), + }) n++ continue } } - if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, *pkt); err != nil { + if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, pkt); err != nil { r.Stats().IP.PacketsSent.IncrementBy(uint64(n)) return n, err } @@ -370,7 +370,7 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe // WriteHeaderIncludedPacket writes a packet already containing a network // header through the given route. -func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuffer) *tcpip.Error { +func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBuffer) *tcpip.Error { // The packet already has an IP header, but there are a few required // checks. h, ok := pkt.Data.PullUp(header.IPv4MinimumSize) @@ -426,7 +426,7 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuf // HandlePacket is called by the link layer when new ipv4 packets arrive for // this endpoint. -func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { +func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { headerView, ok := pkt.Data.PullUp(header.IPv4MinimumSize) if !ok { r.Stats().IP.MalformedPacketsReceived.Increment() @@ -447,7 +447,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { // iptables filtering. All packets that reach here are intended for // this machine and will not be forwarded. ipt := e.stack.IPTables() - if ok := ipt.Check(stack.Input, &pkt, nil, nil, "", ""); !ok { + if ok := ipt.Check(stack.Input, pkt, nil, nil, "", ""); !ok { // iptables is telling us to drop the packet. return } diff --git a/pkg/tcpip/network/ipv4/ipv4_test.go b/pkg/tcpip/network/ipv4/ipv4_test.go index 36035c820..c208ebd99 100644 --- a/pkg/tcpip/network/ipv4/ipv4_test.go +++ b/pkg/tcpip/network/ipv4/ipv4_test.go @@ -114,7 +114,7 @@ func makeHdrAndPayload(hdrLength int, extraLength int, viewSizes []int) (buffer. // comparePayloads compared the contents of all the packets against the contents // of the source packet. -func compareFragments(t *testing.T, packets []stack.PacketBuffer, sourcePacketInfo stack.PacketBuffer, mtu uint32) { +func compareFragments(t *testing.T, packets []*stack.PacketBuffer, sourcePacketInfo *stack.PacketBuffer, mtu uint32) { t.Helper() // Make a complete array of the sourcePacketInfo packet. source := header.IPv4(packets[0].Header.View()[:header.IPv4MinimumSize]) @@ -174,7 +174,7 @@ func compareFragments(t *testing.T, packets []stack.PacketBuffer, sourcePacketIn type errorChannel struct { *channel.Endpoint - Ch chan stack.PacketBuffer + Ch chan *stack.PacketBuffer packetCollectorErrors []*tcpip.Error } @@ -184,7 +184,7 @@ type errorChannel struct { func newErrorChannel(size int, mtu uint32, linkAddr tcpip.LinkAddress, packetCollectorErrors []*tcpip.Error) *errorChannel { return &errorChannel{ Endpoint: channel.New(size, mtu, linkAddr), - Ch: make(chan stack.PacketBuffer, size), + Ch: make(chan *stack.PacketBuffer, size), packetCollectorErrors: packetCollectorErrors, } } @@ -203,7 +203,7 @@ func (e *errorChannel) Drain() int { } // WritePacket stores outbound packets into the channel. -func (e *errorChannel) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) *tcpip.Error { +func (e *errorChannel) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { select { case e.Ch <- pkt: default: @@ -282,13 +282,17 @@ func TestFragmentation(t *testing.T) { for _, ft := range fragTests { t.Run(ft.description, func(t *testing.T) { hdr, payload := makeHdrAndPayload(ft.hdrLength, ft.extraLength, ft.payloadViewsSizes) - source := stack.PacketBuffer{ + source := &stack.PacketBuffer{ Header: hdr, // Save the source payload because WritePacket will modify it. Data: payload.Clone(nil), } c := buildContext(t, nil, ft.mtu) - err := c.Route.WritePacket(ft.gso, stack.NetworkHeaderParams{Protocol: tcp.ProtocolNumber, TTL: 42, TOS: stack.DefaultTOS}, stack.PacketBuffer{ + err := c.Route.WritePacket(ft.gso, stack.NetworkHeaderParams{ + Protocol: tcp.ProtocolNumber, + TTL: 42, + TOS: stack.DefaultTOS, + }, &stack.PacketBuffer{ Header: hdr, Data: payload, }) @@ -296,7 +300,7 @@ func TestFragmentation(t *testing.T) { t.Errorf("err got %v, want %v", err, nil) } - var results []stack.PacketBuffer + var results []*stack.PacketBuffer L: for { select { @@ -338,7 +342,11 @@ func TestFragmentationErrors(t *testing.T) { t.Run(ft.description, func(t *testing.T) { hdr, payload := makeHdrAndPayload(ft.hdrLength, header.IPv4MinimumSize, ft.payloadViewsSizes) c := buildContext(t, ft.packetCollectorErrors, ft.mtu) - err := c.Route.WritePacket(&stack.GSO{}, stack.NetworkHeaderParams{Protocol: tcp.ProtocolNumber, TTL: 42, TOS: stack.DefaultTOS}, stack.PacketBuffer{ + err := c.Route.WritePacket(&stack.GSO{}, stack.NetworkHeaderParams{ + Protocol: tcp.ProtocolNumber, + TTL: 42, + TOS: stack.DefaultTOS, + }, &stack.PacketBuffer{ Header: hdr, Data: payload, }) @@ -460,7 +468,7 @@ func TestInvalidFragments(t *testing.T) { s.CreateNIC(nicID, sniffer.New(ep)) for _, pkt := range tc.packets { - ep.InjectLinkAddr(header.IPv4ProtocolNumber, remoteLinkAddr, stack.PacketBuffer{ + ep.InjectLinkAddr(header.IPv4ProtocolNumber, remoteLinkAddr, &stack.PacketBuffer{ Data: buffer.NewVectorisedView(len(pkt), []buffer.View{pkt}), }) } @@ -698,7 +706,7 @@ func TestReceiveFragments(t *testing.T) { vv := hdr.View().ToVectorisedView() vv.AppendView(frag.payload) - e.InjectInbound(header.IPv4ProtocolNumber, stack.PacketBuffer{ + e.InjectInbound(header.IPv4ProtocolNumber, &stack.PacketBuffer{ Data: vv, }) } diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index bdf3a0d25..b62fb1de6 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -27,7 +27,7 @@ import ( // the original packet that caused the ICMP one to be sent. This information is // used to find out which transport endpoint must be notified about the ICMP // packet. -func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack.PacketBuffer) { +func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt *stack.PacketBuffer) { h, ok := pkt.Data.PullUp(header.IPv6MinimumSize) if !ok { return @@ -70,7 +70,7 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt stack. e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, hdr.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) } -func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.PacketBuffer, hasFragmentHeader bool) { +func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt *stack.PacketBuffer, hasFragmentHeader bool) { stats := r.Stats().ICMP sent := stats.V6PacketsSent received := stats.V6PacketsReceived @@ -288,7 +288,7 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P // // The IP Hop Limit field has a value of 255, i.e., the packet // could not possibly have been forwarded by a router. - if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, TOS: stack.DefaultTOS}, stack.PacketBuffer{ + if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, TOS: stack.DefaultTOS}, &stack.PacketBuffer{ Header: hdr, }); err != nil { sent.Dropped.Increment() @@ -390,7 +390,7 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P copy(packet, icmpHdr) packet.SetType(header.ICMPv6EchoReply) packet.SetChecksum(header.ICMPv6Checksum(packet, r.LocalAddress, r.RemoteAddress, pkt.Data)) - if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, stack.PacketBuffer{ + if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, &stack.PacketBuffer{ Header: hdr, Data: pkt.Data, }); err != nil { @@ -532,7 +532,7 @@ func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, linkEP stack. }) // TODO(stijlist): count this in ICMP stats. - return linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, stack.PacketBuffer{ + return linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, &stack.PacketBuffer{ Header: hdr, }) } diff --git a/pkg/tcpip/network/ipv6/icmp_test.go b/pkg/tcpip/network/ipv6/icmp_test.go index d412ff688..a720f626f 100644 --- a/pkg/tcpip/network/ipv6/icmp_test.go +++ b/pkg/tcpip/network/ipv6/icmp_test.go @@ -57,7 +57,7 @@ func (*stubLinkEndpoint) LinkAddress() tcpip.LinkAddress { return "" } -func (*stubLinkEndpoint) WritePacket(*stack.Route, *stack.GSO, tcpip.NetworkProtocolNumber, stack.PacketBuffer) *tcpip.Error { +func (*stubLinkEndpoint) WritePacket(*stack.Route, *stack.GSO, tcpip.NetworkProtocolNumber, *stack.PacketBuffer) *tcpip.Error { return nil } @@ -67,7 +67,7 @@ type stubDispatcher struct { stack.TransportDispatcher } -func (*stubDispatcher) DeliverTransportPacket(*stack.Route, tcpip.TransportProtocolNumber, stack.PacketBuffer) { +func (*stubDispatcher) DeliverTransportPacket(*stack.Route, tcpip.TransportProtocolNumber, *stack.PacketBuffer) { } type stubLinkAddressCache struct { @@ -189,7 +189,7 @@ func TestICMPCounts(t *testing.T) { SrcAddr: r.LocalAddress, DstAddr: r.RemoteAddress, }) - ep.HandlePacket(&r, stack.PacketBuffer{ + ep.HandlePacket(&r, &stack.PacketBuffer{ Data: hdr.View().ToVectorisedView(), }) } @@ -328,7 +328,7 @@ func routeICMPv6Packet(t *testing.T, args routeArgs, fn func(*testing.T, header. views := []buffer.View{pi.Pkt.Header.View(), pi.Pkt.Data.ToView()} size := pi.Pkt.Header.UsedLength() + pi.Pkt.Data.Size() vv := buffer.NewVectorisedView(size, views) - args.dst.InjectLinkAddr(pi.Proto, args.dst.LinkAddress(), stack.PacketBuffer{ + args.dst.InjectLinkAddr(pi.Proto, args.dst.LinkAddress(), &stack.PacketBuffer{ Data: vv, }) } @@ -563,7 +563,7 @@ func TestICMPChecksumValidationSimple(t *testing.T) { SrcAddr: lladdr1, DstAddr: lladdr0, }) - e.InjectInbound(ProtocolNumber, stack.PacketBuffer{ + e.InjectInbound(ProtocolNumber, &stack.PacketBuffer{ Data: hdr.View().ToVectorisedView(), }) } @@ -740,7 +740,7 @@ func TestICMPChecksumValidationWithPayload(t *testing.T) { SrcAddr: lladdr1, DstAddr: lladdr0, }) - e.InjectInbound(ProtocolNumber, stack.PacketBuffer{ + e.InjectInbound(ProtocolNumber, &stack.PacketBuffer{ Data: hdr.View().ToVectorisedView(), }) } @@ -918,7 +918,7 @@ func TestICMPChecksumValidationWithPayloadMultipleViews(t *testing.T) { SrcAddr: lladdr1, DstAddr: lladdr0, }) - e.InjectInbound(ProtocolNumber, stack.PacketBuffer{ + e.InjectInbound(ProtocolNumber, &stack.PacketBuffer{ Data: buffer.NewVectorisedView(header.IPv6MinimumSize+size+payloadSize, []buffer.View{hdr.View(), payload}), }) } diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index daf1fcbc6..0d94ad122 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -116,7 +116,7 @@ func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadS } // WritePacket writes a packet to the given destination address and protocol. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, pkt stack.PacketBuffer) *tcpip.Error { +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, pkt *stack.PacketBuffer) *tcpip.Error { ip := e.addIPHeader(r, &pkt.Header, pkt.Data.Size(), params) pkt.NetworkHeader = buffer.View(ip) @@ -128,7 +128,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw views = append(views, pkt.Data.Views()...) loopedR := r.MakeLoopedRoute() - e.HandlePacket(&loopedR, stack.PacketBuffer{ + e.HandlePacket(&loopedR, &stack.PacketBuffer{ Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views), }) @@ -163,14 +163,14 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe // WriteHeaderIncludedPacker implements stack.NetworkEndpoint. It is not yet // supported by IPv6. -func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuffer) *tcpip.Error { +func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBuffer) *tcpip.Error { // TODO(b/146666412): Support IPv6 header-included packets. return tcpip.ErrNotSupported } // HandlePacket is called by the link layer when new ipv6 packets arrive for // this endpoint. -func (e *endpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { +func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { headerView, ok := pkt.Data.PullUp(header.IPv6MinimumSize) if !ok { r.Stats().IP.MalformedPacketsReceived.Increment() diff --git a/pkg/tcpip/network/ipv6/ipv6_test.go b/pkg/tcpip/network/ipv6/ipv6_test.go index 841a0cb7a..213ff64f2 100644 --- a/pkg/tcpip/network/ipv6/ipv6_test.go +++ b/pkg/tcpip/network/ipv6/ipv6_test.go @@ -65,7 +65,7 @@ func testReceiveICMP(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst DstAddr: dst, }) - e.InjectInbound(ProtocolNumber, stack.PacketBuffer{ + e.InjectInbound(ProtocolNumber, &stack.PacketBuffer{ Data: hdr.View().ToVectorisedView(), }) @@ -123,7 +123,7 @@ func testReceiveUDP(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst DstAddr: dst, }) - e.InjectInbound(ProtocolNumber, stack.PacketBuffer{ + e.InjectInbound(ProtocolNumber, &stack.PacketBuffer{ Data: hdr.View().ToVectorisedView(), }) @@ -637,7 +637,7 @@ func TestReceiveIPv6ExtHdrs(t *testing.T) { DstAddr: addr2, }) - e.InjectInbound(ProtocolNumber, stack.PacketBuffer{ + e.InjectInbound(ProtocolNumber, &stack.PacketBuffer{ Data: hdr.View().ToVectorisedView(), }) @@ -1238,7 +1238,7 @@ func TestReceiveIPv6Fragments(t *testing.T) { vv := hdr.View().ToVectorisedView() vv.Append(f.data) - e.InjectInbound(ProtocolNumber, stack.PacketBuffer{ + e.InjectInbound(ProtocolNumber, &stack.PacketBuffer{ Data: vv, }) } diff --git a/pkg/tcpip/network/ipv6/ndp_test.go b/pkg/tcpip/network/ipv6/ndp_test.go index 12b70f7e9..3c141b91b 100644 --- a/pkg/tcpip/network/ipv6/ndp_test.go +++ b/pkg/tcpip/network/ipv6/ndp_test.go @@ -136,7 +136,7 @@ func TestNeighorSolicitationWithSourceLinkLayerOption(t *testing.T) { t.Fatalf("got invalid = %d, want = 0", got) } - e.InjectInbound(ProtocolNumber, stack.PacketBuffer{ + e.InjectInbound(ProtocolNumber, &stack.PacketBuffer{ Data: hdr.View().ToVectorisedView(), }) @@ -380,7 +380,7 @@ func TestNeighorSolicitationResponse(t *testing.T) { t.Fatalf("got invalid = %d, want = 0", got) } - e.InjectLinkAddr(ProtocolNumber, test.nsSrcLinkAddr, stack.PacketBuffer{ + e.InjectLinkAddr(ProtocolNumber, test.nsSrcLinkAddr, &stack.PacketBuffer{ Data: hdr.View().ToVectorisedView(), }) @@ -497,7 +497,7 @@ func TestNeighorAdvertisementWithTargetLinkLayerOption(t *testing.T) { t.Fatalf("got invalid = %d, want = 0", got) } - e.InjectInbound(ProtocolNumber, stack.PacketBuffer{ + e.InjectInbound(ProtocolNumber, &stack.PacketBuffer{ Data: hdr.View().ToVectorisedView(), }) @@ -568,7 +568,7 @@ func TestNDPValidation(t *testing.T) { SrcAddr: r.LocalAddress, DstAddr: r.RemoteAddress, }) - ep.HandlePacket(r, stack.PacketBuffer{ + ep.HandlePacket(r, &stack.PacketBuffer{ Data: hdr.View().ToVectorisedView(), }) } @@ -884,7 +884,7 @@ func TestRouterAdvertValidation(t *testing.T) { t.Fatalf("got rxRA = %d, want = 0", got) } - e.InjectInbound(header.IPv6ProtocolNumber, stack.PacketBuffer{ + e.InjectInbound(header.IPv6ProtocolNumber, &stack.PacketBuffer{ Data: hdr.View().ToVectorisedView(), }) diff --git a/pkg/tcpip/stack/conntrack.go b/pkg/tcpip/stack/conntrack.go index 7d1ede1f2..d4053be08 100644 --- a/pkg/tcpip/stack/conntrack.go +++ b/pkg/tcpip/stack/conntrack.go @@ -186,7 +186,7 @@ func parseHeaders(pkt *PacketBuffer) { } // packetToTuple converts packet to a tuple in original direction. -func packetToTuple(pkt PacketBuffer, hook Hook) (connTrackTuple, *tcpip.Error) { +func packetToTuple(pkt *PacketBuffer, hook Hook) (connTrackTuple, *tcpip.Error) { var tuple connTrackTuple netHeader := header.IPv4(pkt.NetworkHeader) @@ -265,7 +265,7 @@ func (ct *ConnTrackTable) connTrackForPacket(pkt *PacketBuffer, hook Hook, creat } var dir ctDirection - tuple, err := packetToTuple(*pkt, hook) + tuple, err := packetToTuple(pkt, hook) if err != nil { return nil, dir } diff --git a/pkg/tcpip/stack/forwarder.go b/pkg/tcpip/stack/forwarder.go index 6b64cd37f..3eff141e6 100644 --- a/pkg/tcpip/stack/forwarder.go +++ b/pkg/tcpip/stack/forwarder.go @@ -32,7 +32,7 @@ type pendingPacket struct { nic *NIC route *Route proto tcpip.NetworkProtocolNumber - pkt PacketBuffer + pkt *PacketBuffer } type forwardQueue struct { @@ -50,7 +50,7 @@ func newForwardQueue() *forwardQueue { return &forwardQueue{packets: make(map[<-chan struct{}][]*pendingPacket)} } -func (f *forwardQueue) enqueue(ch <-chan struct{}, n *NIC, r *Route, protocol tcpip.NetworkProtocolNumber, pkt PacketBuffer) { +func (f *forwardQueue) enqueue(ch <-chan struct{}, n *NIC, r *Route, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { shouldWait := false f.Lock() diff --git a/pkg/tcpip/stack/forwarder_test.go b/pkg/tcpip/stack/forwarder_test.go index 344d60baa..63537aaad 100644 --- a/pkg/tcpip/stack/forwarder_test.go +++ b/pkg/tcpip/stack/forwarder_test.go @@ -68,7 +68,7 @@ func (f *fwdTestNetworkEndpoint) ID() *NetworkEndpointID { return &f.id } -func (f *fwdTestNetworkEndpoint) HandlePacket(r *Route, pkt PacketBuffer) { +func (f *fwdTestNetworkEndpoint) HandlePacket(r *Route, pkt *PacketBuffer) { // Consume the network header. b, ok := pkt.Data.PullUp(fwdTestNetHeaderLen) if !ok { @@ -96,7 +96,7 @@ func (f *fwdTestNetworkEndpoint) NetworkProtocolNumber() tcpip.NetworkProtocolNu return f.proto.Number() } -func (f *fwdTestNetworkEndpoint) WritePacket(r *Route, gso *GSO, params NetworkHeaderParams, pkt PacketBuffer) *tcpip.Error { +func (f *fwdTestNetworkEndpoint) WritePacket(r *Route, gso *GSO, params NetworkHeaderParams, pkt *PacketBuffer) *tcpip.Error { // Add the protocol's header to the packet and send it to the link // endpoint. b := pkt.Header.Prepend(fwdTestNetHeaderLen) @@ -112,7 +112,7 @@ func (f *fwdTestNetworkEndpoint) WritePackets(r *Route, gso *GSO, pkts PacketBuf panic("not implemented") } -func (*fwdTestNetworkEndpoint) WriteHeaderIncludedPacket(r *Route, pkt PacketBuffer) *tcpip.Error { +func (*fwdTestNetworkEndpoint) WriteHeaderIncludedPacket(r *Route, pkt *PacketBuffer) *tcpip.Error { return tcpip.ErrNotSupported } @@ -190,7 +190,7 @@ func (f *fwdTestNetworkProtocol) LinkAddressProtocol() tcpip.NetworkProtocolNumb type fwdTestPacketInfo struct { RemoteLinkAddress tcpip.LinkAddress LocalLinkAddress tcpip.LinkAddress - Pkt PacketBuffer + Pkt *PacketBuffer } type fwdTestLinkEndpoint struct { @@ -203,12 +203,12 @@ type fwdTestLinkEndpoint struct { } // InjectInbound injects an inbound packet. -func (e *fwdTestLinkEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt PacketBuffer) { +func (e *fwdTestLinkEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { e.InjectLinkAddr(protocol, "", pkt) } // InjectLinkAddr injects an inbound packet with a remote link address. -func (e *fwdTestLinkEndpoint) InjectLinkAddr(protocol tcpip.NetworkProtocolNumber, remote tcpip.LinkAddress, pkt PacketBuffer) { +func (e *fwdTestLinkEndpoint) InjectLinkAddr(protocol tcpip.NetworkProtocolNumber, remote tcpip.LinkAddress, pkt *PacketBuffer) { e.dispatcher.DeliverNetworkPacket(remote, "" /* local */, protocol, pkt) } @@ -251,7 +251,7 @@ func (e *fwdTestLinkEndpoint) LinkAddress() tcpip.LinkAddress { return e.linkAddr } -func (e fwdTestLinkEndpoint) WritePacket(r *Route, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkt PacketBuffer) *tcpip.Error { +func (e fwdTestLinkEndpoint) WritePacket(r *Route, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) *tcpip.Error { p := fwdTestPacketInfo{ RemoteLinkAddress: r.RemoteLinkAddress, LocalLinkAddress: r.LocalLinkAddress, @@ -270,7 +270,7 @@ func (e fwdTestLinkEndpoint) WritePacket(r *Route, gso *GSO, protocol tcpip.Netw func (e *fwdTestLinkEndpoint) WritePackets(r *Route, gso *GSO, pkts PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { n := 0 for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() { - e.WritePacket(r, gso, protocol, *pkt) + e.WritePacket(r, gso, protocol, pkt) n++ } @@ -280,7 +280,7 @@ func (e *fwdTestLinkEndpoint) WritePackets(r *Route, gso *GSO, pkts PacketBuffer // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. func (e *fwdTestLinkEndpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { p := fwdTestPacketInfo{ - Pkt: PacketBuffer{Data: vv}, + Pkt: &PacketBuffer{Data: vv}, } select { @@ -362,7 +362,7 @@ func TestForwardingWithStaticResolver(t *testing.T) { // forwarded to NIC 2. buf := buffer.NewView(30) buf[0] = 3 - ep1.InjectInbound(fwdTestNetNumber, PacketBuffer{ + ep1.InjectInbound(fwdTestNetNumber, &PacketBuffer{ Data: buf.ToVectorisedView(), }) @@ -399,7 +399,7 @@ func TestForwardingWithFakeResolver(t *testing.T) { // forwarded to NIC 2. buf := buffer.NewView(30) buf[0] = 3 - ep1.InjectInbound(fwdTestNetNumber, PacketBuffer{ + ep1.InjectInbound(fwdTestNetNumber, &PacketBuffer{ Data: buf.ToVectorisedView(), }) @@ -430,7 +430,7 @@ func TestForwardingWithNoResolver(t *testing.T) { // forwarded to NIC 2. buf := buffer.NewView(30) buf[0] = 3 - ep1.InjectInbound(fwdTestNetNumber, PacketBuffer{ + ep1.InjectInbound(fwdTestNetNumber, &PacketBuffer{ Data: buf.ToVectorisedView(), }) @@ -460,7 +460,7 @@ func TestForwardingWithFakeResolverPartialTimeout(t *testing.T) { // not be forwarded. buf := buffer.NewView(30) buf[0] = 4 - ep1.InjectInbound(fwdTestNetNumber, PacketBuffer{ + ep1.InjectInbound(fwdTestNetNumber, &PacketBuffer{ Data: buf.ToVectorisedView(), }) @@ -468,7 +468,7 @@ func TestForwardingWithFakeResolverPartialTimeout(t *testing.T) { // forwarded to NIC 2. buf = buffer.NewView(30) buf[0] = 3 - ep1.InjectInbound(fwdTestNetNumber, PacketBuffer{ + ep1.InjectInbound(fwdTestNetNumber, &PacketBuffer{ Data: buf.ToVectorisedView(), }) @@ -510,7 +510,7 @@ func TestForwardingWithFakeResolverTwoPackets(t *testing.T) { for i := 0; i < 2; i++ { buf := buffer.NewView(30) buf[0] = 3 - ep1.InjectInbound(fwdTestNetNumber, PacketBuffer{ + ep1.InjectInbound(fwdTestNetNumber, &PacketBuffer{ Data: buf.ToVectorisedView(), }) } @@ -557,7 +557,7 @@ func TestForwardingWithFakeResolverManyPackets(t *testing.T) { buf[0] = 3 // Set the packet sequence number. binary.BigEndian.PutUint16(buf[fwdTestNetHeaderLen:], uint16(i)) - ep1.InjectInbound(fwdTestNetNumber, PacketBuffer{ + ep1.InjectInbound(fwdTestNetNumber, &PacketBuffer{ Data: buf.ToVectorisedView(), }) } @@ -610,7 +610,7 @@ func TestForwardingWithFakeResolverManyResolutions(t *testing.T) { // maxPendingResolutions + 7). buf := buffer.NewView(30) buf[0] = byte(3 + i) - ep1.InjectInbound(fwdTestNetNumber, PacketBuffer{ + ep1.InjectInbound(fwdTestNetNumber, &PacketBuffer{ Data: buf.ToVectorisedView(), }) } diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go index 709ede3fa..d989dbe91 100644 --- a/pkg/tcpip/stack/iptables.go +++ b/pkg/tcpip/stack/iptables.go @@ -321,7 +321,7 @@ func (it *IPTables) checkRule(hook Hook, pkt *PacketBuffer, table Table, ruleIdx // Go through each rule matcher. If they all match, run // the rule target. for _, matcher := range rule.Matchers { - matches, hotdrop := matcher.Match(hook, *pkt, "") + matches, hotdrop := matcher.Match(hook, pkt, "") if hotdrop { return RuleDrop, 0 } diff --git a/pkg/tcpip/stack/iptables_types.go b/pkg/tcpip/stack/iptables_types.go index a3bd3e700..af72b9c46 100644 --- a/pkg/tcpip/stack/iptables_types.go +++ b/pkg/tcpip/stack/iptables_types.go @@ -245,7 +245,7 @@ type Matcher interface { // used for suspicious packets. // // Precondition: packet.NetworkHeader is set. - Match(hook Hook, packet PacketBuffer, interfaceName string) (matches bool, hotdrop bool) + Match(hook Hook, packet *PacketBuffer, interfaceName string) (matches bool, hotdrop bool) } // A Target is the interface for taking an action for a packet. diff --git a/pkg/tcpip/stack/ndp.go b/pkg/tcpip/stack/ndp.go index 526c7d6ff..ae7a8f740 100644 --- a/pkg/tcpip/stack/ndp.go +++ b/pkg/tcpip/stack/ndp.go @@ -750,7 +750,7 @@ func (ndp *ndpState) sendDADPacket(addr tcpip.Address) *tcpip.Error { Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, TOS: DefaultTOS, - }, PacketBuffer{Header: hdr}, + }, &PacketBuffer{Header: hdr}, ); err != nil { sent.Dropped.Increment() return err @@ -1881,7 +1881,7 @@ func (ndp *ndpState) startSolicitingRouters() { Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, TOS: DefaultTOS, - }, PacketBuffer{Header: hdr}, + }, &PacketBuffer{Header: hdr}, ); err != nil { sent.Dropped.Increment() log.Printf("startSolicitingRouters: error writing NDP router solicit message on NIC(%d); err = %s", ndp.nic.ID(), err) diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go index b3d174cdd..58f1ebf60 100644 --- a/pkg/tcpip/stack/ndp_test.go +++ b/pkg/tcpip/stack/ndp_test.go @@ -613,7 +613,7 @@ func TestDADFail(t *testing.T) { // Receive a packet to simulate multiple nodes owning or // attempting to own the same address. hdr := test.makeBuf(addr1) - e.InjectInbound(header.IPv6ProtocolNumber, stack.PacketBuffer{ + e.InjectInbound(header.IPv6ProtocolNumber, &stack.PacketBuffer{ Data: hdr.View().ToVectorisedView(), }) @@ -935,7 +935,7 @@ func TestSetNDPConfigurations(t *testing.T) { // raBufWithOptsAndDHCPv6 returns a valid NDP Router Advertisement with options // and DHCPv6 configurations specified. -func raBufWithOptsAndDHCPv6(ip tcpip.Address, rl uint16, managedAddress, otherConfigurations bool, optSer header.NDPOptionsSerializer) stack.PacketBuffer { +func raBufWithOptsAndDHCPv6(ip tcpip.Address, rl uint16, managedAddress, otherConfigurations bool, optSer header.NDPOptionsSerializer) *stack.PacketBuffer { icmpSize := header.ICMPv6HeaderSize + header.NDPRAMinimumSize + int(optSer.Length()) hdr := buffer.NewPrependable(header.IPv6MinimumSize + icmpSize) pkt := header.ICMPv6(hdr.Prepend(icmpSize)) @@ -970,14 +970,14 @@ func raBufWithOptsAndDHCPv6(ip tcpip.Address, rl uint16, managedAddress, otherCo DstAddr: header.IPv6AllNodesMulticastAddress, }) - return stack.PacketBuffer{Data: hdr.View().ToVectorisedView()} + return &stack.PacketBuffer{Data: hdr.View().ToVectorisedView()} } // raBufWithOpts returns a valid NDP Router Advertisement with options. // // Note, raBufWithOpts does not populate any of the RA fields other than the // Router Lifetime. -func raBufWithOpts(ip tcpip.Address, rl uint16, optSer header.NDPOptionsSerializer) stack.PacketBuffer { +func raBufWithOpts(ip tcpip.Address, rl uint16, optSer header.NDPOptionsSerializer) *stack.PacketBuffer { return raBufWithOptsAndDHCPv6(ip, rl, false, false, optSer) } @@ -986,7 +986,7 @@ func raBufWithOpts(ip tcpip.Address, rl uint16, optSer header.NDPOptionsSerializ // // Note, raBufWithDHCPv6 does not populate any of the RA fields other than the // DHCPv6 related ones. -func raBufWithDHCPv6(ip tcpip.Address, managedAddresses, otherConfiguratiosns bool) stack.PacketBuffer { +func raBufWithDHCPv6(ip tcpip.Address, managedAddresses, otherConfiguratiosns bool) *stack.PacketBuffer { return raBufWithOptsAndDHCPv6(ip, 0, managedAddresses, otherConfiguratiosns, header.NDPOptionsSerializer{}) } @@ -994,7 +994,7 @@ func raBufWithDHCPv6(ip tcpip.Address, managedAddresses, otherConfiguratiosns bo // // Note, raBuf does not populate any of the RA fields other than the // Router Lifetime. -func raBuf(ip tcpip.Address, rl uint16) stack.PacketBuffer { +func raBuf(ip tcpip.Address, rl uint16) *stack.PacketBuffer { return raBufWithOpts(ip, rl, header.NDPOptionsSerializer{}) } @@ -1003,7 +1003,7 @@ func raBuf(ip tcpip.Address, rl uint16) stack.PacketBuffer { // // Note, raBufWithPI does not populate any of the RA fields other than the // Router Lifetime. -func raBufWithPI(ip tcpip.Address, rl uint16, prefix tcpip.AddressWithPrefix, onLink, auto bool, vl, pl uint32) stack.PacketBuffer { +func raBufWithPI(ip tcpip.Address, rl uint16, prefix tcpip.AddressWithPrefix, onLink, auto bool, vl, pl uint32) *stack.PacketBuffer { flags := uint8(0) if onLink { // The OnLink flag is the 7th bit in the flags byte. diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 05646e5e2..ec8e3cb85 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -1153,7 +1153,7 @@ func (n *NIC) isInGroup(addr tcpip.Address) bool { return joins != 0 } -func handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, localLinkAddr, remotelinkAddr tcpip.LinkAddress, ref *referencedNetworkEndpoint, pkt PacketBuffer) { +func handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, localLinkAddr, remotelinkAddr tcpip.LinkAddress, ref *referencedNetworkEndpoint, pkt *PacketBuffer) { r := makeRoute(protocol, dst, src, localLinkAddr, ref, false /* handleLocal */, false /* multicastLoop */) r.RemoteLinkAddress = remotelinkAddr @@ -1167,7 +1167,7 @@ func handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, // Note that the ownership of the slice backing vv is retained by the caller. // This rule applies only to the slice itself, not to the items of the slice; // the ownership of the items is not retained by the caller. -func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt PacketBuffer) { +func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { n.mu.RLock() enabled := n.mu.enabled // If the NIC is not yet enabled, don't receive any packets. @@ -1233,7 +1233,7 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp // iptables filtering. ipt := n.stack.IPTables() address := n.primaryAddress(protocol) - if ok := ipt.Check(Prerouting, &pkt, nil, nil, address.Address, ""); !ok { + if ok := ipt.Check(Prerouting, pkt, nil, nil, address.Address, ""); !ok { // iptables is telling us to drop the packet. return } @@ -1298,7 +1298,7 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp } } -func (n *NIC) forwardPacket(r *Route, protocol tcpip.NetworkProtocolNumber, pkt PacketBuffer) { +func (n *NIC) forwardPacket(r *Route, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { // TODO(b/143425874) Decrease the TTL field in forwarded packets. if linkHeaderLen := int(n.linkEP.MaxHeaderLength()); linkHeaderLen != 0 { pkt.Header = buffer.NewPrependable(linkHeaderLen) @@ -1318,7 +1318,7 @@ func (n *NIC) forwardPacket(r *Route, protocol tcpip.NetworkProtocolNumber, pkt // DeliverTransportPacket delivers the packets to the appropriate transport // protocol endpoint. -func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt PacketBuffer) { +func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt *PacketBuffer) { state, ok := n.stack.transportProtocols[protocol] if !ok { n.stack.stats.UnknownProtocolRcvdPackets.Increment() @@ -1365,7 +1365,7 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN // DeliverTransportControlPacket delivers control packets to the appropriate // transport protocol endpoint. -func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, pkt PacketBuffer) { +func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, pkt *PacketBuffer) { state, ok := n.stack.transportProtocols[trans] if !ok { return diff --git a/pkg/tcpip/stack/nic_test.go b/pkg/tcpip/stack/nic_test.go index b01b3f476..fea46158c 100644 --- a/pkg/tcpip/stack/nic_test.go +++ b/pkg/tcpip/stack/nic_test.go @@ -44,7 +44,7 @@ func TestDisabledRxStatsWhenNICDisabled(t *testing.T) { t.FailNow() } - nic.DeliverNetworkPacket("", "", 0, PacketBuffer{Data: buffer.View([]byte{1, 2, 3, 4}).ToVectorisedView()}) + nic.DeliverNetworkPacket("", "", 0, &PacketBuffer{Data: buffer.View([]byte{1, 2, 3, 4}).ToVectorisedView()}) if got := nic.stats.DisabledRx.Packets.Value(); got != 1 { t.Errorf("got DisabledRx.Packets = %d, want = 1", got) diff --git a/pkg/tcpip/stack/packet_buffer.go b/pkg/tcpip/stack/packet_buffer.go index 926df4d7b..1b5da6017 100644 --- a/pkg/tcpip/stack/packet_buffer.go +++ b/pkg/tcpip/stack/packet_buffer.go @@ -24,6 +24,8 @@ import ( // multiple endpoints. Clone() should be called in such cases so that // modifications to the Data field do not affect other copies. type PacketBuffer struct { + _ noCopy + // PacketBufferEntry is used to build an intrusive list of // PacketBuffers. PacketBufferEntry @@ -82,7 +84,32 @@ type PacketBuffer struct { // VectorisedView but does not deep copy the underlying bytes. // // Clone also does not deep copy any of its other fields. -func (pk PacketBuffer) Clone() PacketBuffer { - pk.Data = pk.Data.Clone(nil) - return pk +// +// FIXME(b/153685824): Data gets copied but not other header references. +func (pk *PacketBuffer) Clone() *PacketBuffer { + return &PacketBuffer{ + PacketBufferEntry: pk.PacketBufferEntry, + Data: pk.Data.Clone(nil), + Header: pk.Header, + LinkHeader: pk.LinkHeader, + NetworkHeader: pk.NetworkHeader, + TransportHeader: pk.TransportHeader, + Hash: pk.Hash, + Owner: pk.Owner, + EgressRoute: pk.EgressRoute, + GSOOptions: pk.GSOOptions, + NetworkProtocolNumber: pk.NetworkProtocolNumber, + NatDone: pk.NatDone, + } } + +// noCopy may be embedded into structs which must not be copied +// after the first use. +// +// See https://golang.org/issues/8005#issuecomment-190753527 +// for details. +type noCopy struct{} + +// Lock is a no-op used by -copylocks checker from `go vet`. +func (*noCopy) Lock() {} +func (*noCopy) Unlock() {} diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go index db89234e8..94f177841 100644 --- a/pkg/tcpip/stack/registration.go +++ b/pkg/tcpip/stack/registration.go @@ -67,12 +67,12 @@ type TransportEndpoint interface { // this transport endpoint. It sets pkt.TransportHeader. // // HandlePacket takes ownership of pkt. - HandlePacket(r *Route, id TransportEndpointID, pkt PacketBuffer) + HandlePacket(r *Route, id TransportEndpointID, pkt *PacketBuffer) // HandleControlPacket is called by the stack when new control (e.g. // ICMP) packets arrive to this transport endpoint. // HandleControlPacket takes ownership of pkt. - HandleControlPacket(id TransportEndpointID, typ ControlType, extra uint32, pkt PacketBuffer) + HandleControlPacket(id TransportEndpointID, typ ControlType, extra uint32, pkt *PacketBuffer) // Abort initiates an expedited endpoint teardown. It puts the endpoint // in a closed state and frees all resources associated with it. This @@ -100,7 +100,7 @@ type RawTransportEndpoint interface { // layer up. // // HandlePacket takes ownership of pkt. - HandlePacket(r *Route, pkt PacketBuffer) + HandlePacket(r *Route, pkt *PacketBuffer) } // PacketEndpoint is the interface that needs to be implemented by packet @@ -118,7 +118,7 @@ type PacketEndpoint interface { // should construct its own ethernet header for applications. // // HandlePacket takes ownership of pkt. - HandlePacket(nicID tcpip.NICID, addr tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, pkt PacketBuffer) + HandlePacket(nicID tcpip.NICID, addr tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, pkt *PacketBuffer) } // TransportProtocol is the interface that needs to be implemented by transport @@ -150,7 +150,7 @@ type TransportProtocol interface { // stats purposes only). // // HandleUnknownDestinationPacket takes ownership of pkt. - HandleUnknownDestinationPacket(r *Route, id TransportEndpointID, pkt PacketBuffer) bool + HandleUnknownDestinationPacket(r *Route, id TransportEndpointID, pkt *PacketBuffer) bool // SetOption allows enabling/disabling protocol specific features. // SetOption returns an error if the option is not supported or the @@ -180,7 +180,7 @@ type TransportDispatcher interface { // pkt.NetworkHeader must be set before calling DeliverTransportPacket. // // DeliverTransportPacket takes ownership of pkt. - DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt PacketBuffer) + DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt *PacketBuffer) // DeliverTransportControlPacket delivers control packets to the // appropriate transport protocol endpoint. @@ -189,7 +189,7 @@ type TransportDispatcher interface { // DeliverTransportControlPacket. // // DeliverTransportControlPacket takes ownership of pkt. - DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, pkt PacketBuffer) + DeliverTransportControlPacket(local, remote tcpip.Address, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, pkt *PacketBuffer) } // PacketLooping specifies where an outbound packet should be sent. @@ -242,7 +242,7 @@ type NetworkEndpoint interface { // WritePacket writes a packet to the given destination address and // protocol. It takes ownership of pkt. pkt.TransportHeader must have already // been set. - WritePacket(r *Route, gso *GSO, params NetworkHeaderParams, pkt PacketBuffer) *tcpip.Error + WritePacket(r *Route, gso *GSO, params NetworkHeaderParams, pkt *PacketBuffer) *tcpip.Error // WritePackets writes packets to the given destination address and // protocol. pkts must not be zero length. It takes ownership of pkts and @@ -251,7 +251,7 @@ type NetworkEndpoint interface { // WriteHeaderIncludedPacket writes a packet that includes a network // header to the given destination address. It takes ownership of pkt. - WriteHeaderIncludedPacket(r *Route, pkt PacketBuffer) *tcpip.Error + WriteHeaderIncludedPacket(r *Route, pkt *PacketBuffer) *tcpip.Error // ID returns the network protocol endpoint ID. ID() *NetworkEndpointID @@ -266,7 +266,7 @@ type NetworkEndpoint interface { // this network endpoint. It sets pkt.NetworkHeader. // // HandlePacket takes ownership of pkt. - HandlePacket(r *Route, pkt PacketBuffer) + HandlePacket(r *Route, pkt *PacketBuffer) // Close is called when the endpoint is reomved from a stack. Close() @@ -327,7 +327,7 @@ type NetworkDispatcher interface { // packets sent via loopback), and won't have the field set. // // DeliverNetworkPacket takes ownership of pkt. - DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt PacketBuffer) + DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) } // LinkEndpointCapabilities is the type associated with the capabilities @@ -389,7 +389,7 @@ type LinkEndpoint interface { // To participate in transparent bridging, a LinkEndpoint implementation // should call eth.Encode with header.EthernetFields.SrcAddr set to // r.LocalLinkAddress if it is provided. - WritePacket(r *Route, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkt PacketBuffer) *tcpip.Error + WritePacket(r *Route, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) *tcpip.Error // WritePackets writes packets with the given protocol through the // given route. pkts must not be zero length. It takes ownership of pkts and @@ -431,7 +431,7 @@ type InjectableLinkEndpoint interface { LinkEndpoint // InjectInbound injects an inbound packet. - InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt PacketBuffer) + InjectInbound(protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) // InjectOutbound writes a fully formed outbound packet directly to the // link. diff --git a/pkg/tcpip/stack/route.go b/pkg/tcpip/stack/route.go index 3d0e5cc6e..f5b6ca0b9 100644 --- a/pkg/tcpip/stack/route.go +++ b/pkg/tcpip/stack/route.go @@ -153,7 +153,7 @@ func (r *Route) IsResolutionRequired() bool { } // WritePacket writes the packet through the given route. -func (r *Route) WritePacket(gso *GSO, params NetworkHeaderParams, pkt PacketBuffer) *tcpip.Error { +func (r *Route) WritePacket(gso *GSO, params NetworkHeaderParams, pkt *PacketBuffer) *tcpip.Error { if !r.ref.isValidForOutgoing() { return tcpip.ErrInvalidEndpointState } @@ -199,7 +199,7 @@ func (r *Route) WritePackets(gso *GSO, pkts PacketBufferList, params NetworkHead // WriteHeaderIncludedPacket writes a packet already containing a network // header through the given route. -func (r *Route) WriteHeaderIncludedPacket(pkt PacketBuffer) *tcpip.Error { +func (r *Route) WriteHeaderIncludedPacket(pkt *PacketBuffer) *tcpip.Error { if !r.ref.isValidForOutgoing() { return tcpip.ErrInvalidEndpointState } diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index 0ab4c3e19..8af06cb9a 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -52,7 +52,7 @@ const ( type transportProtocolState struct { proto TransportProtocol - defaultHandler func(r *Route, id TransportEndpointID, pkt PacketBuffer) bool + defaultHandler func(r *Route, id TransportEndpointID, pkt *PacketBuffer) bool } // TCPProbeFunc is the expected function type for a TCP probe function to be @@ -778,7 +778,7 @@ func (s *Stack) TransportProtocolOption(transport tcpip.TransportProtocolNumber, // // It must be called only during initialization of the stack. Changing it as the // stack is operating is not supported. -func (s *Stack) SetTransportProtocolHandler(p tcpip.TransportProtocolNumber, h func(*Route, TransportEndpointID, PacketBuffer) bool) { +func (s *Stack) SetTransportProtocolHandler(p tcpip.TransportProtocolNumber, h func(*Route, TransportEndpointID, *PacketBuffer) bool) { state := s.transportProtocols[p] if state != nil { state.defaultHandler = h diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index 1a2cf007c..f6ddc3ced 100644 --- a/pkg/tcpip/stack/stack_test.go +++ b/pkg/tcpip/stack/stack_test.go @@ -90,7 +90,7 @@ func (f *fakeNetworkEndpoint) ID() *stack.NetworkEndpointID { return &f.id } -func (f *fakeNetworkEndpoint) HandlePacket(r *stack.Route, pkt stack.PacketBuffer) { +func (f *fakeNetworkEndpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { // Increment the received packet count in the protocol descriptor. f.proto.packetCount[int(f.id.LocalAddress[0])%len(f.proto.packetCount)]++ @@ -132,7 +132,7 @@ func (f *fakeNetworkEndpoint) NetworkProtocolNumber() tcpip.NetworkProtocolNumbe return f.proto.Number() } -func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, pkt stack.PacketBuffer) *tcpip.Error { +func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, pkt *stack.PacketBuffer) *tcpip.Error { // Increment the sent packet count in the protocol descriptor. f.proto.sendPacketCount[int(r.RemoteAddress[0])%len(f.proto.sendPacketCount)]++ @@ -147,7 +147,7 @@ func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, params views := make([]buffer.View, 1, 1+len(pkt.Data.Views())) views[0] = pkt.Header.View() views = append(views, pkt.Data.Views()...) - f.HandlePacket(r, stack.PacketBuffer{ + f.HandlePacket(r, &stack.PacketBuffer{ Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views), }) } @@ -163,7 +163,7 @@ func (f *fakeNetworkEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts panic("not implemented") } -func (*fakeNetworkEndpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt stack.PacketBuffer) *tcpip.Error { +func (*fakeNetworkEndpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBuffer) *tcpip.Error { return tcpip.ErrNotSupported } @@ -293,7 +293,7 @@ func TestNetworkReceive(t *testing.T) { // Make sure packet with wrong address is not delivered. buf[0] = 3 - ep.InjectInbound(fakeNetNumber, stack.PacketBuffer{ + ep.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if fakeNet.packetCount[1] != 0 { @@ -305,7 +305,7 @@ func TestNetworkReceive(t *testing.T) { // Make sure packet is delivered to first endpoint. buf[0] = 1 - ep.InjectInbound(fakeNetNumber, stack.PacketBuffer{ + ep.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if fakeNet.packetCount[1] != 1 { @@ -317,7 +317,7 @@ func TestNetworkReceive(t *testing.T) { // Make sure packet is delivered to second endpoint. buf[0] = 2 - ep.InjectInbound(fakeNetNumber, stack.PacketBuffer{ + ep.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if fakeNet.packetCount[1] != 1 { @@ -328,7 +328,7 @@ func TestNetworkReceive(t *testing.T) { } // Make sure packet is not delivered if protocol number is wrong. - ep.InjectInbound(fakeNetNumber-1, stack.PacketBuffer{ + ep.InjectInbound(fakeNetNumber-1, &stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if fakeNet.packetCount[1] != 1 { @@ -340,7 +340,7 @@ func TestNetworkReceive(t *testing.T) { // Make sure packet that is too small is dropped. buf.CapLength(2) - ep.InjectInbound(fakeNetNumber, stack.PacketBuffer{ + ep.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if fakeNet.packetCount[1] != 1 { @@ -362,7 +362,7 @@ func sendTo(s *stack.Stack, addr tcpip.Address, payload buffer.View) *tcpip.Erro func send(r stack.Route, payload buffer.View) *tcpip.Error { hdr := buffer.NewPrependable(int(r.MaxHeaderLength())) - return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}, stack.PacketBuffer{ + return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}, &stack.PacketBuffer{ Header: hdr, Data: payload.ToVectorisedView(), }) @@ -420,7 +420,7 @@ func testFailingRecv(t *testing.T, fakeNet *fakeNetworkProtocol, localAddrByte b func testRecvInternal(t *testing.T, fakeNet *fakeNetworkProtocol, localAddrByte byte, ep *channel.Endpoint, buf buffer.View, want int) { t.Helper() - ep.InjectInbound(fakeNetNumber, stack.PacketBuffer{ + ep.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if got := fakeNet.PacketCount(localAddrByte); got != want { @@ -2263,7 +2263,7 @@ func TestNICStats(t *testing.T) { // Send a packet to address 1. buf := buffer.NewView(30) - ep1.InjectInbound(fakeNetNumber, stack.PacketBuffer{ + ep1.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if got, want := s.NICInfo()[1].Stats.Rx.Packets.Value(), uint64(1); got != want { @@ -2345,7 +2345,7 @@ func TestNICForwarding(t *testing.T) { // Send a packet to dstAddr. buf := buffer.NewView(30) buf[0] = dstAddr[0] - ep1.InjectInbound(fakeNetNumber, stack.PacketBuffer{ + ep1.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) diff --git a/pkg/tcpip/stack/transport_demuxer.go b/pkg/tcpip/stack/transport_demuxer.go index 9a33ed375..e09866405 100644 --- a/pkg/tcpip/stack/transport_demuxer.go +++ b/pkg/tcpip/stack/transport_demuxer.go @@ -152,7 +152,7 @@ func (epsByNIC *endpointsByNIC) transportEndpoints() []TransportEndpoint { // HandlePacket is called by the stack when new packets arrive to this transport // endpoint. -func (epsByNIC *endpointsByNIC) handlePacket(r *Route, id TransportEndpointID, pkt PacketBuffer) { +func (epsByNIC *endpointsByNIC) handlePacket(r *Route, id TransportEndpointID, pkt *PacketBuffer) { epsByNIC.mu.RLock() mpep, ok := epsByNIC.endpoints[r.ref.nic.ID()] @@ -183,7 +183,7 @@ func (epsByNIC *endpointsByNIC) handlePacket(r *Route, id TransportEndpointID, p } // HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket. -func (epsByNIC *endpointsByNIC) handleControlPacket(n *NIC, id TransportEndpointID, typ ControlType, extra uint32, pkt PacketBuffer) { +func (epsByNIC *endpointsByNIC) handleControlPacket(n *NIC, id TransportEndpointID, typ ControlType, extra uint32, pkt *PacketBuffer) { epsByNIC.mu.RLock() defer epsByNIC.mu.RUnlock() @@ -251,7 +251,7 @@ type transportDemuxer struct { // the dispatcher to delivery packets to the QueuePacket method instead of // calling HandlePacket directly on the endpoint. type queuedTransportProtocol interface { - QueuePacket(r *Route, ep TransportEndpoint, id TransportEndpointID, pkt PacketBuffer) + QueuePacket(r *Route, ep TransportEndpoint, id TransportEndpointID, pkt *PacketBuffer) } func newTransportDemuxer(stack *Stack) *transportDemuxer { @@ -379,7 +379,7 @@ func selectEndpoint(id TransportEndpointID, mpep *multiPortEndpoint, seed uint32 return mpep.endpoints[idx] } -func (ep *multiPortEndpoint) handlePacketAll(r *Route, id TransportEndpointID, pkt PacketBuffer) { +func (ep *multiPortEndpoint) handlePacketAll(r *Route, id TransportEndpointID, pkt *PacketBuffer) { ep.mu.RLock() queuedProtocol, mustQueue := ep.demux.queuedProtocols[protocolIDs{ep.netProto, ep.transProto}] // HandlePacket takes ownership of pkt, so each endpoint needs @@ -470,7 +470,7 @@ func (d *transportDemuxer) unregisterEndpoint(netProtos []tcpip.NetworkProtocolN // deliverPacket attempts to find one or more matching transport endpoints, and // then, if matches are found, delivers the packet to them. Returns true if // the packet no longer needs to be handled. -func (d *transportDemuxer) deliverPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt PacketBuffer, id TransportEndpointID) bool { +func (d *transportDemuxer) deliverPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt *PacketBuffer, id TransportEndpointID) bool { eps, ok := d.protocol[protocolIDs{r.NetProto, protocol}] if !ok { return false @@ -520,7 +520,7 @@ func (d *transportDemuxer) deliverPacket(r *Route, protocol tcpip.TransportProto // deliverRawPacket attempts to deliver the given packet and returns whether it // was delivered successfully. -func (d *transportDemuxer) deliverRawPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt PacketBuffer) bool { +func (d *transportDemuxer) deliverRawPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt *PacketBuffer) bool { eps, ok := d.protocol[protocolIDs{r.NetProto, protocol}] if !ok { return false @@ -544,7 +544,7 @@ func (d *transportDemuxer) deliverRawPacket(r *Route, protocol tcpip.TransportPr // deliverControlPacket attempts to deliver the given control packet. Returns // true if it found an endpoint, false otherwise. -func (d *transportDemuxer) deliverControlPacket(n *NIC, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, pkt PacketBuffer, id TransportEndpointID) bool { +func (d *transportDemuxer) deliverControlPacket(n *NIC, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, pkt *PacketBuffer, id TransportEndpointID) bool { eps, ok := d.protocol[protocolIDs{net, trans}] if !ok { return false diff --git a/pkg/tcpip/stack/transport_demuxer_test.go b/pkg/tcpip/stack/transport_demuxer_test.go index 2474a7db3..67d778137 100644 --- a/pkg/tcpip/stack/transport_demuxer_test.go +++ b/pkg/tcpip/stack/transport_demuxer_test.go @@ -127,7 +127,7 @@ func (c *testContext) sendV4Packet(payload []byte, h *headers, linkEpID tcpip.NI u.SetChecksum(^u.CalculateChecksum(xsum)) // Inject packet. - c.linkEps[linkEpID].InjectInbound(ipv4.ProtocolNumber, stack.PacketBuffer{ + c.linkEps[linkEpID].InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ Data: buf.ToVectorisedView(), NetworkHeader: buffer.View(ip), TransportHeader: buffer.View(u), @@ -165,7 +165,7 @@ func (c *testContext) sendV6Packet(payload []byte, h *headers, linkEpID tcpip.NI u.SetChecksum(^u.CalculateChecksum(xsum)) // Inject packet. - c.linkEps[linkEpID].InjectInbound(ipv6.ProtocolNumber, stack.PacketBuffer{ + c.linkEps[linkEpID].InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ Data: buf.ToVectorisedView(), NetworkHeader: buffer.View(ip), TransportHeader: buffer.View(u), diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go index a611e44ab..cb350ead3 100644 --- a/pkg/tcpip/stack/transport_test.go +++ b/pkg/tcpip/stack/transport_test.go @@ -88,7 +88,7 @@ func (f *fakeTransportEndpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions if err != nil { return 0, nil, err } - if err := f.route.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}, stack.PacketBuffer{ + if err := f.route.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}, &stack.PacketBuffer{ Header: hdr, Data: buffer.View(v).ToVectorisedView(), }); err != nil { @@ -215,7 +215,7 @@ func (*fakeTransportEndpoint) GetRemoteAddress() (tcpip.FullAddress, *tcpip.Erro return tcpip.FullAddress{}, nil } -func (f *fakeTransportEndpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, _ stack.PacketBuffer) { +func (f *fakeTransportEndpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, _ *stack.PacketBuffer) { // Increment the number of received packets. f.proto.packetCount++ if f.acceptQueue != nil { @@ -232,7 +232,7 @@ func (f *fakeTransportEndpoint) HandlePacket(r *stack.Route, id stack.TransportE } } -func (f *fakeTransportEndpoint) HandleControlPacket(stack.TransportEndpointID, stack.ControlType, uint32, stack.PacketBuffer) { +func (f *fakeTransportEndpoint) HandleControlPacket(stack.TransportEndpointID, stack.ControlType, uint32, *stack.PacketBuffer) { // Increment the number of received control packets. f.proto.controlCount++ } @@ -289,7 +289,7 @@ func (*fakeTransportProtocol) ParsePorts(buffer.View) (src, dst uint16, err *tcp return 0, 0, nil } -func (*fakeTransportProtocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, stack.PacketBuffer) bool { +func (*fakeTransportProtocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, *stack.PacketBuffer) bool { return true } @@ -369,7 +369,7 @@ func TestTransportReceive(t *testing.T) { // Make sure packet with wrong protocol is not delivered. buf[0] = 1 buf[2] = 0 - linkEP.InjectInbound(fakeNetNumber, stack.PacketBuffer{ + linkEP.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if fakeTrans.packetCount != 0 { @@ -380,7 +380,7 @@ func TestTransportReceive(t *testing.T) { buf[0] = 1 buf[1] = 3 buf[2] = byte(fakeTransNumber) - linkEP.InjectInbound(fakeNetNumber, stack.PacketBuffer{ + linkEP.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if fakeTrans.packetCount != 0 { @@ -391,7 +391,7 @@ func TestTransportReceive(t *testing.T) { buf[0] = 1 buf[1] = 2 buf[2] = byte(fakeTransNumber) - linkEP.InjectInbound(fakeNetNumber, stack.PacketBuffer{ + linkEP.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if fakeTrans.packetCount != 1 { @@ -446,7 +446,7 @@ func TestTransportControlReceive(t *testing.T) { buf[fakeNetHeaderLen+0] = 0 buf[fakeNetHeaderLen+1] = 1 buf[fakeNetHeaderLen+2] = 0 - linkEP.InjectInbound(fakeNetNumber, stack.PacketBuffer{ + linkEP.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if fakeTrans.controlCount != 0 { @@ -457,7 +457,7 @@ func TestTransportControlReceive(t *testing.T) { buf[fakeNetHeaderLen+0] = 3 buf[fakeNetHeaderLen+1] = 1 buf[fakeNetHeaderLen+2] = byte(fakeTransNumber) - linkEP.InjectInbound(fakeNetNumber, stack.PacketBuffer{ + linkEP.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if fakeTrans.controlCount != 0 { @@ -468,7 +468,7 @@ func TestTransportControlReceive(t *testing.T) { buf[fakeNetHeaderLen+0] = 2 buf[fakeNetHeaderLen+1] = 1 buf[fakeNetHeaderLen+2] = byte(fakeTransNumber) - linkEP.InjectInbound(fakeNetNumber, stack.PacketBuffer{ + linkEP.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) if fakeTrans.controlCount != 1 { @@ -623,7 +623,7 @@ func TestTransportForwarding(t *testing.T) { req[0] = 1 req[1] = 3 req[2] = byte(fakeTransNumber) - ep2.InjectInbound(fakeNetNumber, stack.PacketBuffer{ + ep2.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ Data: req.ToVectorisedView(), }) diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index b1d820372..29ff68df3 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -450,7 +450,7 @@ func send4(r *stack.Route, ident uint16, data buffer.View, ttl uint8, owner tcpi if ttl == 0 { ttl = r.DefaultTTL() } - return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}, stack.PacketBuffer{ + return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}, &stack.PacketBuffer{ Header: hdr, Data: data.ToVectorisedView(), TransportHeader: buffer.View(icmpv4), @@ -481,7 +481,7 @@ func send6(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Err if ttl == 0 { ttl = r.DefaultTTL() } - return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}, stack.PacketBuffer{ + return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}, &stack.PacketBuffer{ Header: hdr, Data: dataVV, TransportHeader: buffer.View(icmpv6), @@ -743,7 +743,7 @@ func (e *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { // HandlePacket is called by the stack when new packets arrive to this transport // endpoint. -func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) { +func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt *stack.PacketBuffer) { // Only accept echo replies. switch e.NetProto { case header.IPv4ProtocolNumber: @@ -805,7 +805,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk } // HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket. -func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.ControlType, extra uint32, pkt stack.PacketBuffer) { +func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.ControlType, extra uint32, pkt *stack.PacketBuffer) { } // State implements tcpip.Endpoint.State. The ICMP endpoint currently doesn't diff --git a/pkg/tcpip/transport/icmp/protocol.go b/pkg/tcpip/transport/icmp/protocol.go index 3c47692b2..2ec6749c7 100644 --- a/pkg/tcpip/transport/icmp/protocol.go +++ b/pkg/tcpip/transport/icmp/protocol.go @@ -104,7 +104,7 @@ func (p *protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) // HandleUnknownDestinationPacket handles packets targeted at this protocol but // that don't match any existing endpoint. -func (*protocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, stack.PacketBuffer) bool { +func (*protocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, *stack.PacketBuffer) bool { return true } diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go index 23158173d..bab2d63ae 100644 --- a/pkg/tcpip/transport/packet/endpoint.go +++ b/pkg/tcpip/transport/packet/endpoint.go @@ -298,7 +298,7 @@ func (ep *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { } // HandlePacket implements stack.PacketEndpoint.HandlePacket. -func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, pkt stack.PacketBuffer) { +func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { ep.rcvMu.Lock() // Drop the packet if our buffer is currently full. diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index eee754a5a..25a17940d 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -348,7 +348,7 @@ func (e *endpoint) finishWrite(payloadBytes []byte, route *stack.Route) (int64, switch e.NetProto { case header.IPv4ProtocolNumber: if !e.associated { - if err := route.WriteHeaderIncludedPacket(stack.PacketBuffer{ + if err := route.WriteHeaderIncludedPacket(&stack.PacketBuffer{ Data: buffer.View(payloadBytes).ToVectorisedView(), }); err != nil { return 0, nil, err @@ -357,7 +357,7 @@ func (e *endpoint) finishWrite(payloadBytes []byte, route *stack.Route) (int64, } hdr := buffer.NewPrependable(len(payloadBytes) + int(route.MaxHeaderLength())) - if err := route.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: e.TransProto, TTL: route.DefaultTTL(), TOS: stack.DefaultTOS}, stack.PacketBuffer{ + if err := route.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: e.TransProto, TTL: route.DefaultTTL(), TOS: stack.DefaultTOS}, &stack.PacketBuffer{ Header: hdr, Data: buffer.View(payloadBytes).ToVectorisedView(), Owner: e.owner, @@ -584,7 +584,7 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { } // HandlePacket implements stack.RawTransportEndpoint.HandlePacket. -func (e *endpoint) HandlePacket(route *stack.Route, pkt stack.PacketBuffer) { +func (e *endpoint) HandlePacket(route *stack.Route, pkt *stack.PacketBuffer) { e.rcvMu.Lock() // Drop the packet if our buffer is currently full. diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index e4a06c9e1..7da93dcc4 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -833,13 +833,13 @@ func sendTCP(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso *stac return sendTCPBatch(r, tf, data, gso, owner) } - pkt := stack.PacketBuffer{ + pkt := &stack.PacketBuffer{ Header: buffer.NewPrependable(header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen), Data: data, Hash: tf.txHash, Owner: owner, } - buildTCPHdr(r, tf, &pkt, gso) + buildTCPHdr(r, tf, pkt, gso) if tf.ttl == 0 { tf.ttl = r.DefaultTTL() diff --git a/pkg/tcpip/transport/tcp/dispatcher.go b/pkg/tcpip/transport/tcp/dispatcher.go index 6062ca916..047704c80 100644 --- a/pkg/tcpip/transport/tcp/dispatcher.go +++ b/pkg/tcpip/transport/tcp/dispatcher.go @@ -186,7 +186,7 @@ func (d *dispatcher) wait() { } } -func (d *dispatcher) queuePacket(r *stack.Route, stackEP stack.TransportEndpoint, id stack.TransportEndpointID, pkt stack.PacketBuffer) { +func (d *dispatcher) queuePacket(r *stack.Route, stackEP stack.TransportEndpoint, id stack.TransportEndpointID, pkt *stack.PacketBuffer) { ep := stackEP.(*endpoint) s := newSegment(r, id, pkt) if !s.parse() { diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index b5ba972f1..d048ef90c 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -2462,7 +2462,7 @@ func (e *endpoint) GetRemoteAddress() (tcpip.FullAddress, *tcpip.Error) { }, nil } -func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) { +func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt *stack.PacketBuffer) { // TCP HandlePacket is not required anymore as inbound packets first // land at the Dispatcher which then can either delivery using the // worker go routine or directly do the invoke the tcp processing inline @@ -2481,7 +2481,7 @@ func (e *endpoint) enqueueSegment(s *segment) bool { } // HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket. -func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.ControlType, extra uint32, pkt stack.PacketBuffer) { +func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.ControlType, extra uint32, pkt *stack.PacketBuffer) { switch typ { case stack.ControlPacketTooBig: e.sndBufMu.Lock() diff --git a/pkg/tcpip/transport/tcp/forwarder.go b/pkg/tcpip/transport/tcp/forwarder.go index 704d01c64..070b634b4 100644 --- a/pkg/tcpip/transport/tcp/forwarder.go +++ b/pkg/tcpip/transport/tcp/forwarder.go @@ -61,7 +61,7 @@ func NewForwarder(s *stack.Stack, rcvWnd, maxInFlight int, handler func(*Forward // // This function is expected to be passed as an argument to the // stack.SetTransportProtocolHandler function. -func (f *Forwarder) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) bool { +func (f *Forwarder) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt *stack.PacketBuffer) bool { s := newSegment(r, id, pkt) defer s.decRef() diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go index 2a2a7ddeb..c827d0277 100644 --- a/pkg/tcpip/transport/tcp/protocol.go +++ b/pkg/tcpip/transport/tcp/protocol.go @@ -206,7 +206,7 @@ func (*protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) { // to a specific processing queue. Each queue is serviced by its own processor // goroutine which is responsible for dequeuing and doing full TCP dispatch of // the packet. -func (p *protocol) QueuePacket(r *stack.Route, ep stack.TransportEndpoint, id stack.TransportEndpointID, pkt stack.PacketBuffer) { +func (p *protocol) QueuePacket(r *stack.Route, ep stack.TransportEndpoint, id stack.TransportEndpointID, pkt *stack.PacketBuffer) { p.dispatcher.queuePacket(r, ep, id, pkt) } @@ -217,7 +217,7 @@ func (p *protocol) QueuePacket(r *stack.Route, ep stack.TransportEndpoint, id st // a reset is sent in response to any incoming segment except another reset. In // particular, SYNs addressed to a non-existent connection are rejected by this // means." -func (*protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) bool { +func (*protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, pkt *stack.PacketBuffer) bool { s := newSegment(r, id, pkt) defer s.decRef() diff --git a/pkg/tcpip/transport/tcp/segment.go b/pkg/tcpip/transport/tcp/segment.go index 074edded6..0c099e2fd 100644 --- a/pkg/tcpip/transport/tcp/segment.go +++ b/pkg/tcpip/transport/tcp/segment.go @@ -60,7 +60,7 @@ type segment struct { xmitCount uint32 } -func newSegment(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) *segment { +func newSegment(r *stack.Route, id stack.TransportEndpointID, pkt *stack.PacketBuffer) *segment { s := &segment{ refCnt: 1, id: id, diff --git a/pkg/tcpip/transport/tcp/testing/context/context.go b/pkg/tcpip/transport/tcp/testing/context/context.go index 7b1d72cf4..9721f6caf 100644 --- a/pkg/tcpip/transport/tcp/testing/context/context.go +++ b/pkg/tcpip/transport/tcp/testing/context/context.go @@ -316,7 +316,7 @@ func (c *Context) SendICMPPacket(typ header.ICMPv4Type, code uint8, p1, p2 []byt copy(icmp[header.ICMPv4PayloadOffset:], p2) // Inject packet. - c.linkEP.InjectInbound(ipv4.ProtocolNumber, stack.PacketBuffer{ + c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) } @@ -372,7 +372,7 @@ func (c *Context) BuildSegmentWithAddrs(payload []byte, h *Headers, src, dst tcp // SendSegment sends a TCP segment that has already been built and written to a // buffer.VectorisedView. func (c *Context) SendSegment(s buffer.VectorisedView) { - c.linkEP.InjectInbound(ipv4.ProtocolNumber, stack.PacketBuffer{ + c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ Data: s, }) } @@ -380,7 +380,7 @@ func (c *Context) SendSegment(s buffer.VectorisedView) { // SendPacket builds and sends a TCP segment(with the provided payload & TCP // headers) in an IPv4 packet via the link layer endpoint. func (c *Context) SendPacket(payload []byte, h *Headers) { - c.linkEP.InjectInbound(ipv4.ProtocolNumber, stack.PacketBuffer{ + c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ Data: c.BuildSegment(payload, h), }) } @@ -389,7 +389,7 @@ func (c *Context) SendPacket(payload []byte, h *Headers) { // & TCPheaders) in an IPv4 packet via the link layer endpoint using the // provided source and destination IPv4 addresses. func (c *Context) SendPacketWithAddrs(payload []byte, h *Headers, src, dst tcpip.Address) { - c.linkEP.InjectInbound(ipv4.ProtocolNumber, stack.PacketBuffer{ + c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ Data: c.BuildSegmentWithAddrs(payload, h, src, dst), }) } @@ -564,7 +564,7 @@ func (c *Context) SendV6PacketWithAddrs(payload []byte, h *Headers, src, dst tcp t.SetChecksum(^t.CalculateChecksum(xsum)) // Inject packet. - c.linkEP.InjectInbound(ipv6.ProtocolNumber, stack.PacketBuffer{ + c.linkEP.InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) } diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 647b2067a..79faa7869 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -921,7 +921,11 @@ func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort u if useDefaultTTL { ttl = r.DefaultTTL() } - if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}, stack.PacketBuffer{ + if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{ + Protocol: ProtocolNumber, + TTL: ttl, + TOS: tos, + }, &stack.PacketBuffer{ Header: hdr, Data: data, TransportHeader: buffer.View(udp), @@ -1269,7 +1273,7 @@ func (e *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { // HandlePacket is called by the stack when new packets arrive to this transport // endpoint. -func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) { +func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt *stack.PacketBuffer) { // Get the header then trim it from the view. hdr, ok := pkt.Data.PullUp(header.UDPMinimumSize) if !ok || int(header.UDP(hdr).Length()) > pkt.Data.Size() { @@ -1336,7 +1340,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk } // HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket. -func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.ControlType, extra uint32, pkt stack.PacketBuffer) { +func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.ControlType, extra uint32, pkt *stack.PacketBuffer) { if typ == stack.ControlPortUnreachable { e.mu.RLock() defer e.mu.RUnlock() diff --git a/pkg/tcpip/transport/udp/forwarder.go b/pkg/tcpip/transport/udp/forwarder.go index a674ceb68..7abfa0ed2 100644 --- a/pkg/tcpip/transport/udp/forwarder.go +++ b/pkg/tcpip/transport/udp/forwarder.go @@ -43,7 +43,7 @@ func NewForwarder(s *stack.Stack, handler func(*ForwarderRequest)) *Forwarder { // // This function is expected to be passed as an argument to the // stack.SetTransportProtocolHandler function. -func (f *Forwarder) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) bool { +func (f *Forwarder) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt *stack.PacketBuffer) bool { f.handler(&ForwarderRequest{ stack: f.stack, route: r, @@ -61,7 +61,7 @@ type ForwarderRequest struct { stack *stack.Stack route *stack.Route id stack.TransportEndpointID - pkt stack.PacketBuffer + pkt *stack.PacketBuffer } // ID returns the 4-tuple (src address, src port, dst address, dst port) that diff --git a/pkg/tcpip/transport/udp/protocol.go b/pkg/tcpip/transport/udp/protocol.go index 52af6de22..e320c5758 100644 --- a/pkg/tcpip/transport/udp/protocol.go +++ b/pkg/tcpip/transport/udp/protocol.go @@ -66,7 +66,7 @@ func (*protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) { // HandleUnknownDestinationPacket handles packets targeted at this protocol but // that don't match any existing endpoint. -func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, pkt stack.PacketBuffer) bool { +func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, pkt *stack.PacketBuffer) bool { // Get the header then trim it from the view. h, ok := pkt.Data.PullUp(header.UDPMinimumSize) if !ok { @@ -140,7 +140,7 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans pkt.SetType(header.ICMPv4DstUnreachable) pkt.SetCode(header.ICMPv4PortUnreachable) pkt.SetChecksum(header.ICMPv4Checksum(pkt, payload)) - r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, stack.PacketBuffer{ + r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, &stack.PacketBuffer{ Header: hdr, Data: payload, }) @@ -177,7 +177,7 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans pkt.SetType(header.ICMPv6DstUnreachable) pkt.SetCode(header.ICMPv6PortUnreachable) pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, payload)) - r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, stack.PacketBuffer{ + r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, &stack.PacketBuffer{ Header: hdr, Data: payload, }) diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index 8acaa607a..e8ade882b 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -440,7 +440,7 @@ func (c *testContext) injectV6Packet(payload []byte, h *header4Tuple, valid bool u.SetChecksum(^u.CalculateChecksum(xsum)) // Inject packet. - c.linkEP.InjectInbound(ipv6.ProtocolNumber, stack.PacketBuffer{ + c.linkEP.InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ Data: buf.ToVectorisedView(), NetworkHeader: buffer.View(ip), TransportHeader: buffer.View(u), @@ -487,7 +487,7 @@ func (c *testContext) injectV4Packet(payload []byte, h *header4Tuple, valid bool // Inject packet. - c.linkEP.InjectInbound(ipv4.ProtocolNumber, stack.PacketBuffer{ + c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ Data: buf.ToVectorisedView(), NetworkHeader: buffer.View(ip), TransportHeader: buffer.View(u), -- cgit v1.2.3 From 6acd185bd3c04d6d6f57ccc9335513abaadf0651 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Wed, 27 May 2020 01:00:30 -0400 Subject: avoid runtime fails with missing stack maps in race mode on Arm64 In race mode, when calling the go function in asm code, there will be an missing stack maps issue. The root cause is: The function of 'muldiv64' has a non-empty frame, so it needs stack maps for locals, for which the macro NO_LOCAL_POINTERS will do. Also, the macro GO_ARGS can covers arguments. Signed-off-by: Bin Lu --- pkg/sentry/time/muldiv_arm64.s | 3 +++ pkg/sentry/time/parameters_test.go | 15 +++++++++++++++ 2 files changed, 18 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/time/muldiv_arm64.s b/pkg/sentry/time/muldiv_arm64.s index 5ad57a8a3..8afc62d53 100644 --- a/pkg/sentry/time/muldiv_arm64.s +++ b/pkg/sentry/time/muldiv_arm64.s @@ -12,12 +12,15 @@ // See the License for the specific language governing permissions and // limitations under the License. +#include "funcdata.h" #include "textflag.h" // Documentation is available in parameters.go. // // func muldiv64(value, multiplier, divisor uint64) (uint64, bool) TEXT ·muldiv64(SB),NOSPLIT,$40-33 + GO_ARGS + NO_LOCAL_POINTERS MOVD value+0(FP), R0 MOVD multiplier+8(FP), R1 MOVD divisor+16(FP), R2 diff --git a/pkg/sentry/time/parameters_test.go b/pkg/sentry/time/parameters_test.go index e1b9084ac..0ce1257f6 100644 --- a/pkg/sentry/time/parameters_test.go +++ b/pkg/sentry/time/parameters_test.go @@ -484,3 +484,18 @@ func TestMulDivOverflow(t *testing.T) { }) } } + +func BenchmarkMuldiv64(b *testing.B) { + var v uint64 = math.MaxUint64 + for i := uint64(1); i <= 1000000; i++ { + mult := uint64(1000000000) + div := i * mult + res, ok := muldiv64(v, mult, div) + if !ok { + b.Errorf("Result of %v * %v / %v ok got false want true", v, mult, div) + } + if want := v / i; res != want { + b.Errorf("Bad result of %v * %v / %v: got %v, want %v", v, mult, div, res, want) + } + } +} -- cgit v1.2.3 From 41da7a568b1e4f46b3bc09724996556fb18b4d16 Mon Sep 17 00:00:00 2001 From: Ting-Yu Wang Date: Thu, 4 Jun 2020 15:38:33 -0700 Subject: Fix copylocks error about copying IPTables. IPTables.connections contains a sync.RWMutex. Copying it will trigger copylocks analysis. Tested by manually enabling nogo tests. sync.RWMutex is added to IPTables for the additional race condition discovered. PiperOrigin-RevId: 314817019 --- pkg/sentry/socket/netfilter/netfilter.go | 36 ++++++++++++--------------- pkg/sentry/socket/netstack/stack.go | 9 +++---- pkg/tcpip/stack/iptables.go | 42 +++++++++++++++++++++++++++----- pkg/tcpip/stack/iptables_types.go | 15 ++++++++---- pkg/tcpip/stack/stack.go | 23 ++++------------- pkg/tcpip/transport/icmp/endpoint.go | 5 ---- pkg/tcpip/transport/packet/endpoint.go | 5 ---- pkg/tcpip/transport/raw/endpoint.go | 5 ---- pkg/tcpip/transport/tcp/endpoint.go | 5 ---- pkg/tcpip/transport/udp/endpoint.go | 5 ---- runsc/boot/loader.go | 2 +- 11 files changed, 71 insertions(+), 81 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 47ff48c00..66015e2bc 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -144,31 +144,27 @@ func GetEntries(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr, outLen } func findTable(stk *stack.Stack, tablename linux.TableName) (stack.Table, error) { - ipt := stk.IPTables() - table, ok := ipt.Tables[tablename.String()] + table, ok := stk.IPTables().GetTable(tablename.String()) if !ok { return stack.Table{}, fmt.Errorf("couldn't find table %q", tablename) } return table, nil } -// FillDefaultIPTables sets stack's IPTables to the default tables and -// populates them with metadata. -func FillDefaultIPTables(stk *stack.Stack) { - ipt := stack.DefaultTables() - - // In order to fill in the metadata, we have to translate ipt from its - // netstack format to Linux's giant-binary-blob format. - for name, table := range ipt.Tables { - _, metadata, err := convertNetstackToBinary(name, table) - if err != nil { - panic(fmt.Errorf("Unable to set default IP tables: %v", err)) +// FillIPTablesMetadata populates stack's IPTables with metadata. +func FillIPTablesMetadata(stk *stack.Stack) { + stk.IPTables().ModifyTables(func(tables map[string]stack.Table) { + // In order to fill in the metadata, we have to translate ipt from its + // netstack format to Linux's giant-binary-blob format. + for name, table := range tables { + _, metadata, err := convertNetstackToBinary(name, table) + if err != nil { + panic(fmt.Errorf("Unable to set default IP tables: %v", err)) + } + table.SetMetadata(metadata) + tables[name] = table } - table.SetMetadata(metadata) - ipt.Tables[name] = table - } - - stk.SetIPTables(ipt) + }) } // convertNetstackToBinary converts the iptables as stored in netstack to the @@ -573,15 +569,13 @@ func SetEntries(stk *stack.Stack, optVal []byte) *syserr.Error { // - There are no chains without an unconditional final rule. // - There are no chains without an unconditional underflow rule. - ipt := stk.IPTables() table.SetMetadata(metadata{ HookEntry: replace.HookEntry, Underflow: replace.Underflow, NumEntries: replace.NumEntries, Size: replace.Size, }) - ipt.Tables[replace.Name.String()] = table - stk.SetIPTables(ipt) + stk.IPTables().ReplaceTable(replace.Name.String(), table) return nil } diff --git a/pkg/sentry/socket/netstack/stack.go b/pkg/sentry/socket/netstack/stack.go index f5fa18136..9b44c2b89 100644 --- a/pkg/sentry/socket/netstack/stack.go +++ b/pkg/sentry/socket/netstack/stack.go @@ -362,14 +362,13 @@ func (s *Stack) RouteTable() []inet.Route { } // IPTables returns the stack's iptables. -func (s *Stack) IPTables() (stack.IPTables, error) { +func (s *Stack) IPTables() (*stack.IPTables, error) { return s.Stack.IPTables(), nil } -// FillDefaultIPTables sets the stack's iptables to the default tables, which -// allow and do not modify all traffic. -func (s *Stack) FillDefaultIPTables() { - netfilter.FillDefaultIPTables(s.Stack) +// FillIPTablesMetadata populates stack's IPTables with metadata. +func (s *Stack) FillIPTablesMetadata() { + netfilter.FillIPTablesMetadata(s.Stack) } // Resume implements inet.Stack.Resume. diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go index d989dbe91..4e9b404c8 100644 --- a/pkg/tcpip/stack/iptables.go +++ b/pkg/tcpip/stack/iptables.go @@ -43,11 +43,11 @@ const HookUnset = -1 // DefaultTables returns a default set of tables. Each chain is set to accept // all packets. -func DefaultTables() IPTables { +func DefaultTables() *IPTables { // TODO(gvisor.dev/issue/170): We may be able to swap out some strings for // iotas. - return IPTables{ - Tables: map[string]Table{ + return &IPTables{ + tables: map[string]Table{ TablenameNat: Table{ Rules: []Rule{ Rule{Target: AcceptTarget{}}, @@ -106,7 +106,7 @@ func DefaultTables() IPTables { UserChains: map[string]int{}, }, }, - Priorities: map[Hook][]string{ + priorities: map[Hook][]string{ Input: []string{TablenameNat, TablenameFilter}, Prerouting: []string{TablenameMangle, TablenameNat}, Output: []string{TablenameMangle, TablenameNat, TablenameFilter}, @@ -158,6 +158,36 @@ func EmptyNatTable() Table { } } +// GetTable returns table by name. +func (it *IPTables) GetTable(name string) (Table, bool) { + it.mu.RLock() + defer it.mu.RUnlock() + t, ok := it.tables[name] + return t, ok +} + +// ReplaceTable replaces or inserts table by name. +func (it *IPTables) ReplaceTable(name string, table Table) { + it.mu.Lock() + defer it.mu.Unlock() + it.tables[name] = table +} + +// ModifyTables acquires write-lock and calls fn with internal name-to-table +// map. This function can be used to update multiple tables atomically. +func (it *IPTables) ModifyTables(fn func(map[string]Table)) { + it.mu.Lock() + defer it.mu.Unlock() + fn(it.tables) +} + +// GetPriorities returns slice of priorities associated with hook. +func (it *IPTables) GetPriorities(hook Hook) []string { + it.mu.RLock() + defer it.mu.RUnlock() + return it.priorities[hook] +} + // A chainVerdict is what a table decides should be done with a packet. type chainVerdict int @@ -184,8 +214,8 @@ func (it *IPTables) Check(hook Hook, pkt *PacketBuffer, gso *GSO, r *Route, addr it.connections.HandlePacket(pkt, hook, gso, r) // Go through each table containing the hook. - for _, tablename := range it.Priorities[hook] { - table := it.Tables[tablename] + for _, tablename := range it.GetPriorities(hook) { + table, _ := it.GetTable(tablename) ruleIdx := table.BuiltinChains[hook] switch verdict := it.checkChain(hook, pkt, table, ruleIdx, gso, r, address, nicName); verdict { // If the table returns Accept, move on to the next table. diff --git a/pkg/tcpip/stack/iptables_types.go b/pkg/tcpip/stack/iptables_types.go index af72b9c46..4a6a5c6f1 100644 --- a/pkg/tcpip/stack/iptables_types.go +++ b/pkg/tcpip/stack/iptables_types.go @@ -16,6 +16,7 @@ package stack import ( "strings" + "sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/header" @@ -78,13 +79,17 @@ const ( // IPTables holds all the tables for a netstack. type IPTables struct { - // Tables maps table names to tables. User tables have arbitrary names. - Tables map[string]Table + // mu protects tables and priorities. + mu sync.RWMutex - // Priorities maps each hook to a list of table names. The order of the + // tables maps table names to tables. User tables have arbitrary names. mu + // needs to be locked for accessing. + tables map[string]Table + + // priorities maps each hook to a list of table names. The order of the // list is the order in which each table should be visited for that - // hook. - Priorities map[Hook][]string + // hook. mu needs to be locked for accessing. + priorities map[Hook][]string connections ConnTrackTable } diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index 8af06cb9a..294ce8775 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -424,12 +424,8 @@ type Stack struct { // handleLocal allows non-loopback interfaces to loop packets. handleLocal bool - // tablesMu protects iptables. - tablesMu sync.RWMutex - - // tables are the iptables packet filtering and manipulation rules. The are - // protected by tablesMu.` - tables IPTables + // tables are the iptables packet filtering and manipulation rules. + tables *IPTables // resumableEndpoints is a list of endpoints that need to be resumed if the // stack is being restored. @@ -676,6 +672,7 @@ func New(opts Options) *Stack { clock: clock, stats: opts.Stats.FillIn(), handleLocal: opts.HandleLocal, + tables: DefaultTables(), icmpRateLimiter: NewICMPRateLimiter(), seed: generateRandUint32(), ndpConfigs: opts.NDPConfigs, @@ -1741,18 +1738,8 @@ func (s *Stack) IsInGroup(nicID tcpip.NICID, multicastAddr tcpip.Address) (bool, } // IPTables returns the stack's iptables. -func (s *Stack) IPTables() IPTables { - s.tablesMu.RLock() - t := s.tables - s.tablesMu.RUnlock() - return t -} - -// SetIPTables sets the stack's iptables. -func (s *Stack) SetIPTables(ipt IPTables) { - s.tablesMu.Lock() - s.tables = ipt - s.tablesMu.Unlock() +func (s *Stack) IPTables() *IPTables { + return s.tables } // ICMPLimit returns the maximum number of ICMP messages that can be sent diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index 29ff68df3..3bc72bc19 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -140,11 +140,6 @@ func (e *endpoint) SetOwner(owner tcpip.PacketOwner) { e.owner = owner } -// IPTables implements tcpip.Endpoint.IPTables. -func (e *endpoint) IPTables() (stack.IPTables, error) { - return e.stack.IPTables(), nil -} - // Read reads data from the endpoint. This method does not block if // there is no data pending. func (e *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) { diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go index bab2d63ae..baf08eda6 100644 --- a/pkg/tcpip/transport/packet/endpoint.go +++ b/pkg/tcpip/transport/packet/endpoint.go @@ -132,11 +132,6 @@ func (ep *endpoint) Close() { // ModerateRecvBuf implements tcpip.Endpoint.ModerateRecvBuf. func (ep *endpoint) ModerateRecvBuf(copied int) {} -// IPTables implements tcpip.Endpoint.IPTables. -func (ep *endpoint) IPTables() (stack.IPTables, error) { - return ep.stack.IPTables(), nil -} - // Read implements tcpip.Endpoint.Read. func (ep *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) { ep.rcvMu.Lock() diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index 25a17940d..21c34fac2 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -166,11 +166,6 @@ func (e *endpoint) SetOwner(owner tcpip.PacketOwner) { e.owner = owner } -// IPTables implements tcpip.Endpoint.IPTables. -func (e *endpoint) IPTables() (stack.IPTables, error) { - return e.stack.IPTables(), nil -} - // Read implements tcpip.Endpoint.Read. func (e *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) { if !e.associated { diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index d048ef90c..edca98160 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -1172,11 +1172,6 @@ func (e *endpoint) SetOwner(owner tcpip.PacketOwner) { e.owner = owner } -// IPTables implements tcpip.Endpoint.IPTables. -func (e *endpoint) IPTables() (stack.IPTables, error) { - return e.stack.IPTables(), nil -} - // Read reads data from the endpoint. func (e *endpoint) Read(*tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) { e.LockUser() diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 79faa7869..663af8fec 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -247,11 +247,6 @@ func (e *endpoint) Close() { // ModerateRecvBuf implements tcpip.Endpoint.ModerateRecvBuf. func (e *endpoint) ModerateRecvBuf(copied int) {} -// IPTables implements tcpip.Endpoint.IPTables. -func (e *endpoint) IPTables() (stack.IPTables, error) { - return e.stack.IPTables(), nil -} - // Read reads data from the endpoint. This method does not block if // there is no data pending. func (e *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) { diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index f802bc9fb..002479612 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -1056,7 +1056,7 @@ func newEmptySandboxNetworkStack(clock tcpip.Clock, uniqueID stack.UniqueID) (in return nil, fmt.Errorf("SetTransportProtocolOption failed: %v", err) } - s.FillDefaultIPTables() + s.FillIPTablesMetadata() return &s, nil } -- cgit v1.2.3 From 526df4f52a07a02687dd43ceb752621a41883f95 Mon Sep 17 00:00:00 2001 From: Bhasker Hariharan Date: Fri, 5 Jun 2020 13:41:19 -0700 Subject: Fix error code returned due to Port exhaustion. For TCP sockets gVisor incorrectly returns EAGAIN when no ephemeral ports are available to bind during a connect. Linux returns EADDRNOTAVAIL. This change fixes gVisor to return the correct code and adds a test for the same. This change also fixes a minor bug for ping sockets where connect() would fail with EINVAL unless the socket was bound first. Also added tests for testing UDP Port exhaustion and Ping socket port exhaustion. PiperOrigin-RevId: 314988525 --- pkg/sentry/socket/netstack/BUILD | 1 + pkg/sentry/socket/netstack/netstack.go | 9 ++ pkg/tcpip/transport/icmp/endpoint.go | 1 + test/syscalls/BUILD | 16 ++ test/syscalls/linux/BUILD | 35 +++++ test/syscalls/linux/ping_socket.cc | 91 +++++++++++ test/syscalls/linux/poll.cc | 6 +- .../linux/socket_inet_loopback_nogotsan.cc | 171 +++++++++++++++++++++ test/syscalls/linux/socket_ipv4_udp_unbound.cc | 33 ++++ 9 files changed, 360 insertions(+), 3 deletions(-) create mode 100644 test/syscalls/linux/ping_socket.cc create mode 100644 test/syscalls/linux/socket_inet_loopback_nogotsan.cc (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/BUILD b/pkg/sentry/socket/netstack/BUILD index 333e0042e..8f0f5466e 100644 --- a/pkg/sentry/socket/netstack/BUILD +++ b/pkg/sentry/socket/netstack/BUILD @@ -50,5 +50,6 @@ go_library( "//pkg/tcpip/transport/udp", "//pkg/usermem", "//pkg/waiter", + "@org_golang_x_sys//unix:go_default_library", ], ) diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 60df51dae..e1e0c5931 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -33,6 +33,7 @@ import ( "syscall" "time" + "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/amutex" "gvisor.dev/gvisor/pkg/binary" @@ -719,6 +720,14 @@ func (s *socketOpsCommon) Connect(t *kernel.Task, sockaddr []byte, blocking bool defer s.EventUnregister(&e) if err := s.Endpoint.Connect(addr); err != tcpip.ErrConnectStarted && err != tcpip.ErrAlreadyConnecting { + if (s.family == unix.AF_INET || s.family == unix.AF_INET6) && s.skType == linux.SOCK_STREAM { + // TCP unlike UDP returns EADDRNOTAVAIL when it can't + // find an available local ephemeral port. + if err == tcpip.ErrNoPortAvailable { + return syserr.ErrAddressNotAvailable + } + } + return syserr.TranslateNetstackError(err) } diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index 3bc72bc19..57e0a069b 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -506,6 +506,7 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { nicID := addr.NIC localPort := uint16(0) switch e.state { + case stateInitial: case stateBound, stateConnected: localPort = e.ID.LocalPort if e.BindNICID == 0 { diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 3406a2de8..d68afbe44 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -400,6 +400,14 @@ syscall_test( vfs2 = "True", ) +syscall_test( + size = "medium", + # Takes too long under gotsan to run. + tags = ["nogotsan"], + test = "//test/syscalls/linux:ping_socket_test", + vfs2 = "True", +) + syscall_test( size = "large", add_overlay = True, @@ -697,6 +705,14 @@ syscall_test( test = "//test/syscalls/linux:socket_inet_loopback_test", ) +syscall_test( + size = "large", + shard_count = 50, + # Takes too long for TSAN. Creates a lot of TCP sockets. + tags = ["nogotsan"], + test = "//test/syscalls/linux:socket_inet_loopback_nogotsan_test", +) + syscall_test( size = "large", shard_count = 50, diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index f4b5de18d..ae2aa44dc 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -1411,6 +1411,21 @@ cc_binary( ], ) +cc_binary( + name = "ping_socket_test", + testonly = 1, + srcs = ["ping_socket.cc"], + linkstatic = 1, + deps = [ + ":socket_test_util", + "//test/util:file_descriptor", + gtest, + "//test/util:save_util", + "//test/util:test_main", + "//test/util:test_util", + ], +) + cc_binary( name = "pipe_test", testonly = 1, @@ -2780,6 +2795,26 @@ cc_binary( ], ) +cc_binary( + name = "socket_inet_loopback_nogotsan_test", + testonly = 1, + srcs = ["socket_inet_loopback_nogotsan.cc"], + linkstatic = 1, + deps = [ + ":ip_socket_test_util", + ":socket_test_util", + "//test/util:file_descriptor", + "@com_google_absl//absl/memory", + "@com_google_absl//absl/strings", + gtest, + "//test/util:posix_error", + "//test/util:save_util", + "//test/util:test_main", + "//test/util:test_util", + "//test/util:thread_util", + ], +) + cc_binary( name = "socket_netlink_test", testonly = 1, diff --git a/test/syscalls/linux/ping_socket.cc b/test/syscalls/linux/ping_socket.cc new file mode 100644 index 000000000..a9bfdb37b --- /dev/null +++ b/test/syscalls/linux/ping_socket.cc @@ -0,0 +1,91 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include +#include +#include +#include +#include +#include + +#include + +#include "gtest/gtest.h" +#include "test/syscalls/linux/socket_test_util.h" +#include "test/util/file_descriptor.h" +#include "test/util/save_util.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { +namespace { + +class PingSocket : public ::testing::Test { + protected: + // Creates a socket to be used in tests. + void SetUp() override; + + // Closes the socket created by SetUp(). + void TearDown() override; + + // The loopback address. + struct sockaddr_in addr_; +}; + +void PingSocket::SetUp() { + // On some hosts ping sockets are restricted to specific groups using the + // sysctl "ping_group_range". + int s = socket(AF_INET, SOCK_RAW, IPPROTO_ICMP); + if (s < 0 && errno == EPERM) { + GTEST_SKIP(); + } + close(s); + + addr_ = {}; + // Just a random port as the destination port number is irrelevant for ping + // sockets. + addr_.sin_port = 12345; + addr_.sin_addr.s_addr = htonl(INADDR_LOOPBACK); + addr_.sin_family = AF_INET; +} + +void PingSocket::TearDown() {} + +// Test ICMP port exhaustion returns EAGAIN. +// +// We disable both random/cooperative S/R for this test as it makes way too many +// syscalls. +TEST_F(PingSocket, ICMPPortExhaustion_NoRandomSave) { + DisableSave ds; + std::vector sockets; + constexpr int kSockets = 65536; + addr_.sin_port = 0; + for (int i = 0; i < kSockets; i++) { + auto s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_DGRAM, IPPROTO_ICMP)); + int ret = connect(s.get(), reinterpret_cast(&addr_), + sizeof(addr_)); + if (ret == 0) { + sockets.push_back(std::move(s)); + continue; + } + ASSERT_THAT(ret, SyscallFailsWithErrno(EAGAIN)); + break; + } +} + +} // namespace + +} // namespace testing +} // namespace gvisor diff --git a/test/syscalls/linux/poll.cc b/test/syscalls/linux/poll.cc index 1e35a4a8b..7a316427d 100644 --- a/test/syscalls/linux/poll.cc +++ b/test/syscalls/linux/poll.cc @@ -259,9 +259,9 @@ TEST_F(PollTest, Nfds) { TEST_PCHECK(getrlimit(RLIMIT_NOFILE, &rlim) == 0); // gVisor caps the number of FDs that epoll can use beyond RLIMIT_NOFILE. - constexpr rlim_t gVisorMax = 1048576; - if (rlim.rlim_cur > gVisorMax) { - rlim.rlim_cur = gVisorMax; + constexpr rlim_t maxFD = 4096; + if (rlim.rlim_cur > maxFD) { + rlim.rlim_cur = maxFD; TEST_PCHECK(setrlimit(RLIMIT_NOFILE, &rlim) == 0); } diff --git a/test/syscalls/linux/socket_inet_loopback_nogotsan.cc b/test/syscalls/linux/socket_inet_loopback_nogotsan.cc new file mode 100644 index 000000000..2324c7f6a --- /dev/null +++ b/test/syscalls/linux/socket_inet_loopback_nogotsan.cc @@ -0,0 +1,171 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include +#include +#include +#include + +#include +#include +#include +#include +#include + +#include "gmock/gmock.h" +#include "gtest/gtest.h" +#include "absl/strings/str_cat.h" +#include "test/syscalls/linux/ip_socket_test_util.h" +#include "test/syscalls/linux/socket_test_util.h" +#include "test/util/file_descriptor.h" +#include "test/util/posix_error.h" +#include "test/util/save_util.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +namespace { + +using ::testing::Gt; + +PosixErrorOr AddrPort(int family, sockaddr_storage const& addr) { + switch (family) { + case AF_INET: + return static_cast( + reinterpret_cast(&addr)->sin_port); + case AF_INET6: + return static_cast( + reinterpret_cast(&addr)->sin6_port); + default: + return PosixError(EINVAL, + absl::StrCat("unknown socket family: ", family)); + } +} + +PosixError SetAddrPort(int family, sockaddr_storage* addr, uint16_t port) { + switch (family) { + case AF_INET: + reinterpret_cast(addr)->sin_port = port; + return NoError(); + case AF_INET6: + reinterpret_cast(addr)->sin6_port = port; + return NoError(); + default: + return PosixError(EINVAL, + absl::StrCat("unknown socket family: ", family)); + } +} + +struct TestParam { + TestAddress listener; + TestAddress connector; +}; + +std::string DescribeTestParam(::testing::TestParamInfo const& info) { + return absl::StrCat("Listen", info.param.listener.description, "_Connect", + info.param.connector.description); +} + +using SocketInetLoopbackTest = ::testing::TestWithParam; + +// This test verifies that connect returns EADDRNOTAVAIL if all local ephemeral +// ports are already in use for a given destination ip/port. +// We disable S/R because this test creates a large number of sockets. +TEST_P(SocketInetLoopbackTest, TestTCPPortExhaustion_NoRandomSave) { + auto const& param = GetParam(); + TestAddress const& listener = param.listener; + TestAddress const& connector = param.connector; + + constexpr int kBacklog = 10; + constexpr int kClients = 65536; + + // Create the listening socket. + auto listen_fd = ASSERT_NO_ERRNO_AND_VALUE( + Socket(listener.family(), SOCK_STREAM, IPPROTO_TCP)); + sockaddr_storage listen_addr = listener.addr; + ASSERT_THAT(bind(listen_fd.get(), reinterpret_cast(&listen_addr), + listener.addr_len), + SyscallSucceeds()); + ASSERT_THAT(listen(listen_fd.get(), kBacklog), SyscallSucceeds()); + + // Get the port bound by the listening socket. + socklen_t addrlen = listener.addr_len; + ASSERT_THAT(getsockname(listen_fd.get(), + reinterpret_cast(&listen_addr), &addrlen), + SyscallSucceeds()); + uint16_t const port = + ASSERT_NO_ERRNO_AND_VALUE(AddrPort(listener.family(), listen_addr)); + + // Disable cooperative S/R as we are making too many syscalls. + DisableSave ds; + + // Now we keep opening connections till we run out of local ephemeral ports. + // and assert the error we get back. + sockaddr_storage conn_addr = connector.addr; + ASSERT_NO_ERRNO(SetAddrPort(connector.family(), &conn_addr, port)); + std::vector clients; + std::vector servers; + + for (int i = 0; i < kClients; i++) { + FileDescriptor client = ASSERT_NO_ERRNO_AND_VALUE( + Socket(connector.family(), SOCK_STREAM, IPPROTO_TCP)); + int ret = connect(client.get(), reinterpret_cast(&conn_addr), + connector.addr_len); + if (ret == 0) { + clients.push_back(std::move(client)); + FileDescriptor server = + ASSERT_NO_ERRNO_AND_VALUE(Accept(listen_fd.get(), nullptr, nullptr)); + servers.push_back(std::move(server)); + continue; + } + ASSERT_THAT(ret, SyscallFailsWithErrno(EADDRNOTAVAIL)); + break; + } +} + +INSTANTIATE_TEST_SUITE_P( + All, SocketInetLoopbackTest, + ::testing::Values( + // Listeners bound to IPv4 addresses refuse connections using IPv6 + // addresses. + TestParam{V4Any(), V4Any()}, TestParam{V4Any(), V4Loopback()}, + TestParam{V4Any(), V4MappedAny()}, + TestParam{V4Any(), V4MappedLoopback()}, + TestParam{V4Loopback(), V4Any()}, TestParam{V4Loopback(), V4Loopback()}, + TestParam{V4Loopback(), V4MappedLoopback()}, + TestParam{V4MappedAny(), V4Any()}, + TestParam{V4MappedAny(), V4Loopback()}, + TestParam{V4MappedAny(), V4MappedAny()}, + TestParam{V4MappedAny(), V4MappedLoopback()}, + TestParam{V4MappedLoopback(), V4Any()}, + TestParam{V4MappedLoopback(), V4Loopback()}, + TestParam{V4MappedLoopback(), V4MappedLoopback()}, + + // Listeners bound to IN6ADDR_ANY accept all connections. + TestParam{V6Any(), V4Any()}, TestParam{V6Any(), V4Loopback()}, + TestParam{V6Any(), V4MappedAny()}, + TestParam{V6Any(), V4MappedLoopback()}, TestParam{V6Any(), V6Any()}, + TestParam{V6Any(), V6Loopback()}, + + // Listeners bound to IN6ADDR_LOOPBACK refuse connections using IPv4 + // addresses. + TestParam{V6Loopback(), V6Any()}, + TestParam{V6Loopback(), V6Loopback()}), + DescribeTestParam); + +} // namespace + +} // namespace testing +} // namespace gvisor diff --git a/test/syscalls/linux/socket_ipv4_udp_unbound.cc b/test/syscalls/linux/socket_ipv4_udp_unbound.cc index bc4b07a62..1294d9050 100644 --- a/test/syscalls/linux/socket_ipv4_udp_unbound.cc +++ b/test/syscalls/linux/socket_ipv4_udp_unbound.cc @@ -2129,6 +2129,39 @@ TEST_P(IPv4UDPUnboundSocketTest, ReuseAddrReusePortDistribution) { SyscallSucceedsWithValue(kMessageSize)); } +// Check that connect returns EADDRNOTAVAIL when out of local ephemeral ports. +// We disable S/R because this test creates a large number of sockets. +TEST_P(IPv4UDPUnboundSocketTest, UDPConnectPortExhaustion_NoRandomSave) { + auto receiver1 = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + constexpr int kClients = 65536; + // Bind the first socket to the loopback and take note of the selected port. + auto addr = V4Loopback(); + ASSERT_THAT(bind(receiver1->get(), reinterpret_cast(&addr.addr), + addr.addr_len), + SyscallSucceeds()); + socklen_t addr_len = addr.addr_len; + ASSERT_THAT(getsockname(receiver1->get(), + reinterpret_cast(&addr.addr), &addr_len), + SyscallSucceeds()); + EXPECT_EQ(addr_len, addr.addr_len); + + // Disable cooperative S/R as we are making too many syscalls. + DisableSave ds; + std::vector> sockets; + for (int i = 0; i < kClients; i++) { + auto s = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + + int ret = connect(s->get(), reinterpret_cast(&addr.addr), + addr.addr_len); + if (ret == 0) { + sockets.push_back(std::move(s)); + continue; + } + ASSERT_THAT(ret, SyscallFailsWithErrno(EAGAIN)); + break; + } +} + // Test that socket will receive packet info control message. TEST_P(IPv4UDPUnboundSocketTest, SetAndReceiveIPPKTINFO) { // TODO(gvisor.dev/issue/1202): ioctl() is not supported by hostinet. -- cgit v1.2.3 From 8c1f5b5cd8b634a5e7255944f42e82c5c9de3149 Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Fri, 5 Jun 2020 14:43:56 -0700 Subject: Unshare files on exec The current task can share its fdtable with a few other tasks, but after exec, this should be a completely separate process. PiperOrigin-RevId: 314999565 --- pkg/sentry/kernel/task_exec.go | 4 ++++ test/syscalls/linux/exec.cc | 27 +++++++++++++++++++++++++++ 2 files changed, 31 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/task_exec.go b/pkg/sentry/kernel/task_exec.go index 00c425cca..9b69f3cbe 100644 --- a/pkg/sentry/kernel/task_exec.go +++ b/pkg/sentry/kernel/task_exec.go @@ -198,6 +198,10 @@ func (r *runSyscallAfterExecStop) execute(t *Task) taskRunState { t.tg.oldRSeqCritical.Store(&OldRSeqCriticalRegion{}) t.tg.pidns.owner.mu.Unlock() + oldFDTable := t.fdTable + t.fdTable = t.fdTable.Fork() + oldFDTable.DecRef() + // Remove FDs with the CloseOnExec flag set. t.fdTable.RemoveIf(func(_ *fs.File, _ *vfs.FileDescription, flags FDFlags) bool { return flags.CloseOnExec diff --git a/test/syscalls/linux/exec.cc b/test/syscalls/linux/exec.cc index 12c9b05ca..e09afafe9 100644 --- a/test/syscalls/linux/exec.cc +++ b/test/syscalls/linux/exec.cc @@ -673,6 +673,33 @@ TEST(ExecveatTest, SymlinkNoFollowWithRelativePath) { EXPECT_EQ(execve_errno, ELOOP); } +TEST(ExecveatTest, UnshareFiles) { + TempPath tempFile = ASSERT_NO_ERRNO_AND_VALUE( + TempPath::CreateFileWith(GetAbsoluteTestTmpdir(), "bar", 0755)); + const FileDescriptor fd_closed_on_exec = + ASSERT_NO_ERRNO_AND_VALUE(Open(tempFile.path(), O_RDONLY | O_CLOEXEC)); + + pid_t child; + EXPECT_THAT(child = syscall(__NR_clone, SIGCHLD | CLONE_VFORK | CLONE_FILES, + 0, 0, 0, 0), + SyscallSucceeds()); + if (child == 0) { + ExecveArray argv = {"test"}; + ExecveArray envp; + ASSERT_THAT( + execve(RunfilePath(kBasicWorkload).c_str(), argv.get(), envp.get()), + SyscallSucceeds()); + _exit(1); + } + + int status; + ASSERT_THAT(RetryEINTR(waitpid)(child, &status, 0), SyscallSucceeds()); + EXPECT_EQ(status, 0); + + struct stat st; + EXPECT_THAT(fstat(fd_closed_on_exec.get(), &st), SyscallSucceeds()); +} + TEST(ExecveatTest, SymlinkNoFollowWithAbsolutePath) { std::string parent_dir = "/tmp"; TempPath link = ASSERT_NO_ERRNO_AND_VALUE( -- cgit v1.2.3 From 9aaca5a6da39c6154ce08f4386e702ee4a18d03a Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Fri, 5 Jun 2020 15:38:38 -0700 Subject: Use top-down allocation for pgalloc. This change has multiple small components. First, the chunk size is bumped to 1GB in order to avoid creating excessive VMAs in the Sentry, which can lead to VMA exhaustion (and hitting limits). Second, gap-tracking is added to the usage set in order to efficiently scan for available regions. Third, reclaim is moved to a simple segment set. This is done to allow the order of reclaim to align with the Allocate order (which becomes much more complex when trying to track a "max page" as opposed to "min page", so we just track explicit segments instead, which should make reclaim scanning faster anyways). Finally, the findAvailable function attempts to scan from the top-down, in order to maximize opportunities for VMA merging in applications (hopefully preventing the same VMA exhaustion that can affect the Sentry). PiperOrigin-RevId: 315009249 --- pkg/sentry/pgalloc/BUILD | 22 ++++ pkg/sentry/pgalloc/pgalloc.go | 215 ++++++++++++++++++++----------------- pkg/sentry/pgalloc/pgalloc_test.go | 198 +++++++++++++++++++++++----------- test/e2e/integration_test.go | 4 +- 4 files changed, 273 insertions(+), 166 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/pgalloc/BUILD b/pkg/sentry/pgalloc/BUILD index 1eeb9f317..a9836ba71 100644 --- a/pkg/sentry/pgalloc/BUILD +++ b/pkg/sentry/pgalloc/BUILD @@ -33,6 +33,7 @@ go_template_instance( out = "usage_set.go", consts = { "minDegree": "10", + "trackGaps": "1", }, imports = { "platform": "gvisor.dev/gvisor/pkg/sentry/platform", @@ -48,6 +49,26 @@ go_template_instance( }, ) +go_template_instance( + name = "reclaim_set", + out = "reclaim_set.go", + consts = { + "minDegree": "10", + }, + imports = { + "platform": "gvisor.dev/gvisor/pkg/sentry/platform", + }, + package = "pgalloc", + prefix = "reclaim", + template = "//pkg/segment:generic_set", + types = { + "Key": "uint64", + "Range": "platform.FileRange", + "Value": "reclaimSetValue", + "Functions": "reclaimSetFunctions", + }, +) + go_library( name = "pgalloc", srcs = [ @@ -56,6 +77,7 @@ go_library( "evictable_range_set.go", "pgalloc.go", "pgalloc_unsafe.go", + "reclaim_set.go", "save_restore.go", "usage_set.go", ], diff --git a/pkg/sentry/pgalloc/pgalloc.go b/pkg/sentry/pgalloc/pgalloc.go index 2b11ea4ae..c8d9facc2 100644 --- a/pkg/sentry/pgalloc/pgalloc.go +++ b/pkg/sentry/pgalloc/pgalloc.go @@ -108,12 +108,6 @@ type MemoryFile struct { usageSwapped uint64 usageLast time.Time - // minUnallocatedPage is the minimum page that may be unallocated. - // i.e., there are no unallocated pages below minUnallocatedPage. - // - // minUnallocatedPage is protected by mu. - minUnallocatedPage uint64 - // fileSize is the size of the backing memory file in bytes. fileSize is // always a power-of-two multiple of chunkSize. // @@ -146,11 +140,9 @@ type MemoryFile struct { // is protected by mu. reclaimable bool - // minReclaimablePage is the minimum page that may be reclaimable. - // i.e., all reclaimable pages are >= minReclaimablePage. - // - // minReclaimablePage is protected by mu. - minReclaimablePage uint64 + // relcaim is the collection of regions for reclaim. relcaim is protected + // by mu. + reclaim reclaimSet // reclaimCond is signaled (with mu locked) when reclaimable or destroyed // transitions from false to true. @@ -273,12 +265,10 @@ type evictableMemoryUserInfo struct { } const ( - chunkShift = 24 - chunkSize = 1 << chunkShift // 16 MB + chunkShift = 30 + chunkSize = 1 << chunkShift // 1 GB chunkMask = chunkSize - 1 - initialSize = chunkSize - // maxPage is the highest 64-bit page. maxPage = math.MaxUint64 &^ (usermem.PageSize - 1) ) @@ -302,19 +292,12 @@ func NewMemoryFile(file *os.File, opts MemoryFileOpts) (*MemoryFile, error) { if err := file.Truncate(0); err != nil { return nil, err } - if err := file.Truncate(initialSize); err != nil { - return nil, err - } f := &MemoryFile{ - opts: opts, - fileSize: initialSize, - file: file, - // No pages are reclaimable. DecRef will always be able to - // decrease minReclaimablePage from this point. - minReclaimablePage: maxPage, - evictable: make(map[EvictableMemoryUser]*evictableMemoryUserInfo), + opts: opts, + file: file, + evictable: make(map[EvictableMemoryUser]*evictableMemoryUserInfo), } - f.mappings.Store(make([]uintptr, initialSize/chunkSize)) + f.mappings.Store(make([]uintptr, 0)) f.reclaimCond.L = &f.mu if f.opts.DelayedEviction == DelayedEvictionEnabled && f.opts.UseHostMemcgPressure { @@ -404,39 +387,28 @@ func (f *MemoryFile) Allocate(length uint64, kind usage.MemoryKind) (platform.Fi alignment = usermem.HugePageSize } - start, minUnallocatedPage := findUnallocatedRange(&f.usage, f.minUnallocatedPage, length, alignment) - end := start + length - // File offsets are int64s. Since length must be strictly positive, end - // cannot legitimately be 0. - if end < start || int64(end) <= 0 { + // Find a range in the underlying file. + fr, ok := findAvailableRange(&f.usage, f.fileSize, length, alignment) + if !ok { return platform.FileRange{}, syserror.ENOMEM } - // Expand the file if needed. Double the file size on each expansion; - // uncommitted pages have effectively no cost. - fileSize := f.fileSize - for int64(end) > fileSize { - if fileSize >= 2*fileSize { - // fileSize overflow. - return platform.FileRange{}, syserror.ENOMEM - } - fileSize *= 2 - } - if fileSize > f.fileSize { - if err := f.file.Truncate(fileSize); err != nil { + // Expand the file if needed. Note that findAvailableRange will + // appropriately double the fileSize when required. + if int64(fr.End) > f.fileSize { + if err := f.file.Truncate(int64(fr.End)); err != nil { return platform.FileRange{}, err } - f.fileSize = fileSize + f.fileSize = int64(fr.End) f.mappingsMu.Lock() oldMappings := f.mappings.Load().([]uintptr) - newMappings := make([]uintptr, fileSize>>chunkShift) + newMappings := make([]uintptr, f.fileSize>>chunkShift) copy(newMappings, oldMappings) f.mappings.Store(newMappings) f.mappingsMu.Unlock() } // Mark selected pages as in use. - fr := platform.FileRange{start, end} if f.opts.ManualZeroing { if err := f.forEachMappingSlice(fr, func(bs []byte) { for i := range bs { @@ -453,49 +425,71 @@ func (f *MemoryFile) Allocate(length uint64, kind usage.MemoryKind) (platform.Fi panic(fmt.Sprintf("allocating %v: failed to insert into usage set:\n%v", fr, &f.usage)) } - if minUnallocatedPage < start { - f.minUnallocatedPage = minUnallocatedPage - } else { - // start was the first unallocated page. The next must be - // somewhere beyond end. - f.minUnallocatedPage = end - } - return fr, nil } -// findUnallocatedRange returns the first unallocated page in usage of the -// specified length and alignment beginning at page start and the first single -// unallocated page. -func findUnallocatedRange(usage *usageSet, start, length, alignment uint64) (uint64, uint64) { - // Only searched until the first page is found. - firstPage := start - foundFirstPage := false - alignMask := alignment - 1 - for seg := usage.LowerBoundSegment(start); seg.Ok(); seg = seg.NextSegment() { - r := seg.Range() +// findAvailableRange returns an available range in the usageSet. +// +// Note that scanning for available slots takes place from end first backwards, +// then forwards. This heuristic has important consequence for how sequential +// mappings can be merged in the host VMAs, given that addresses for both +// application and sentry mappings are allocated top-down (from higher to +// lower addresses). The file is also grown expoentially in order to create +// space for mappings to be allocated downwards. +// +// Precondition: alignment must be a power of 2. +func findAvailableRange(usage *usageSet, fileSize int64, length, alignment uint64) (platform.FileRange, bool) { + alignmentMask := alignment - 1 + for gap := usage.UpperBoundGap(uint64(fileSize)); gap.Ok(); gap = gap.PrevLargeEnoughGap(length) { + // Start searching only at end of file. + end := gap.End() + if end > uint64(fileSize) { + end = uint64(fileSize) + } - if !foundFirstPage && r.Start > firstPage { - foundFirstPage = true + // Start at the top and align downwards. + start := end - length + if start > end { + break // Underflow. } + start &^= alignmentMask - if start >= r.End { - // start was rounded up to an alignment boundary from the end - // of a previous segment and is now beyond r.End. + // Is the gap still sufficient? + if start < gap.Start() { continue } - // This segment represents allocated or reclaimable pages; only the - // range from start to the segment's beginning is allocatable, and the - // next allocatable range begins after the segment. - if r.Start > start && r.Start-start >= length { - break + + // Allocate in the given gap. + return platform.FileRange{start, start + length}, true + } + + // Check that it's possible to fit this allocation at the end of a file of any size. + min := usage.LastGap().Start() + min = (min + alignmentMask) &^ alignmentMask + if min+length < min { + // Overflow. + return platform.FileRange{}, false + } + + // Determine the minimum file size required to fit this allocation at its end. + for { + if fileSize >= 2*fileSize { + // Is this because it's initially empty? + if fileSize == 0 { + fileSize += chunkSize + } else { + // fileSize overflow. + return platform.FileRange{}, false + } + } else { + // Double the current fileSize. + fileSize *= 2 } - start = (r.End + alignMask) &^ alignMask - if !foundFirstPage { - firstPage = r.End + start := (uint64(fileSize) - length) &^ alignmentMask + if start >= min { + return platform.FileRange{start, start + length}, true } } - return start, firstPage } // AllocateAndFill allocates memory of the given kind and fills it by calling @@ -616,6 +610,7 @@ func (f *MemoryFile) DecRef(fr platform.FileRange) { } val.refs-- if val.refs == 0 { + f.reclaim.Add(seg.Range(), reclaimSetValue{}) freed = true // Reclassify memory as System, until it's freed by the reclaim // goroutine. @@ -628,10 +623,6 @@ func (f *MemoryFile) DecRef(fr platform.FileRange) { f.usage.MergeAdjacent(fr) if freed { - if fr.Start < f.minReclaimablePage { - // We've freed at least one lower page. - f.minReclaimablePage = fr.Start - } f.reclaimable = true f.reclaimCond.Signal() } @@ -1030,6 +1021,7 @@ func (f *MemoryFile) String() string { // for allocation. func (f *MemoryFile) runReclaim() { for { + // N.B. We must call f.markReclaimed on the returned FrameRange. fr, ok := f.findReclaimable() if !ok { break @@ -1085,6 +1077,10 @@ func (f *MemoryFile) runReclaim() { } } +// findReclaimable finds memory that has been marked for reclaim. +// +// Note that there returned range will be removed from tracking. It +// must be reclaimed (removed from f.usage) at this point. func (f *MemoryFile) findReclaimable() (platform.FileRange, bool) { f.mu.Lock() defer f.mu.Unlock() @@ -1103,18 +1099,15 @@ func (f *MemoryFile) findReclaimable() (platform.FileRange, bool) { } f.reclaimCond.Wait() } - // Allocate returns the first usable range in offset order and is - // currently a linear scan, so reclaiming from the beginning of the - // file minimizes the expected latency of Allocate. - for seg := f.usage.LowerBoundSegment(f.minReclaimablePage); seg.Ok(); seg = seg.NextSegment() { - if seg.ValuePtr().refs == 0 { - f.minReclaimablePage = seg.End() - return seg.Range(), true - } + // Allocate works from the back of the file inwards, so reclaim + // preserves this order to minimize the cost of the search. + if seg := f.reclaim.LastSegment(); seg.Ok() { + fr := seg.Range() + f.reclaim.Remove(seg) + return fr, true } - // No pages are reclaimable. + // Nothing is reclaimable. f.reclaimable = false - f.minReclaimablePage = maxPage } } @@ -1122,8 +1115,8 @@ func (f *MemoryFile) markReclaimed(fr platform.FileRange) { f.mu.Lock() defer f.mu.Unlock() seg := f.usage.FindSegment(fr.Start) - // All of fr should be mapped to a single uncommitted reclaimable segment - // accounted to System. + // All of fr should be mapped to a single uncommitted reclaimable + // segment accounted to System. if !seg.Ok() { panic(fmt.Sprintf("reclaimed pages %v include unreferenced pages:\n%v", fr, &f.usage)) } @@ -1137,14 +1130,10 @@ func (f *MemoryFile) markReclaimed(fr platform.FileRange) { }); got != want { panic(fmt.Sprintf("reclaimed pages %v in segment %v has incorrect state %v, wanted %v:\n%v", fr, seg.Range(), got, want, &f.usage)) } - // Deallocate reclaimed pages. Even though all of seg is reclaimable, the - // caller of markReclaimed may not have decommitted it, so we can only mark - // fr as reclaimed. + // Deallocate reclaimed pages. Even though all of seg is reclaimable, + // the caller of markReclaimed may not have decommitted it, so we can + // only mark fr as reclaimed. f.usage.Remove(f.usage.Isolate(seg, fr)) - if fr.Start < f.minUnallocatedPage { - // We've deallocated at least one lower page. - f.minUnallocatedPage = fr.Start - } } // StartEvictions requests that f evict all evictable allocations. It does not @@ -1255,3 +1244,27 @@ func (evictableRangeSetFunctions) Merge(_ EvictableRange, _ evictableRangeSetVal func (evictableRangeSetFunctions) Split(_ EvictableRange, _ evictableRangeSetValue, _ uint64) (evictableRangeSetValue, evictableRangeSetValue) { return evictableRangeSetValue{}, evictableRangeSetValue{} } + +// reclaimSetValue is the value type of reclaimSet. +type reclaimSetValue struct{} + +type reclaimSetFunctions struct{} + +func (reclaimSetFunctions) MinKey() uint64 { + return 0 +} + +func (reclaimSetFunctions) MaxKey() uint64 { + return math.MaxUint64 +} + +func (reclaimSetFunctions) ClearValue(val *reclaimSetValue) { +} + +func (reclaimSetFunctions) Merge(_ platform.FileRange, _ reclaimSetValue, _ platform.FileRange, _ reclaimSetValue) (reclaimSetValue, bool) { + return reclaimSetValue{}, true +} + +func (reclaimSetFunctions) Split(_ platform.FileRange, _ reclaimSetValue, _ uint64) (reclaimSetValue, reclaimSetValue) { + return reclaimSetValue{}, reclaimSetValue{} +} diff --git a/pkg/sentry/pgalloc/pgalloc_test.go b/pkg/sentry/pgalloc/pgalloc_test.go index 293f22c6b..b5b68eb52 100644 --- a/pkg/sentry/pgalloc/pgalloc_test.go +++ b/pkg/sentry/pgalloc/pgalloc_test.go @@ -23,39 +23,49 @@ import ( const ( page = usermem.PageSize hugepage = usermem.HugePageSize + topPage = (1 << 63) - page ) func TestFindUnallocatedRange(t *testing.T) { for _, test := range []struct { - desc string - usage *usageSegmentDataSlices - start uint64 - length uint64 - alignment uint64 - unallocated uint64 - minUnallocated uint64 + desc string + usage *usageSegmentDataSlices + fileSize int64 + length uint64 + alignment uint64 + start uint64 + expectFail bool }{ { - desc: "Initial allocation succeeds", - usage: &usageSegmentDataSlices{}, - start: 0, - length: page, - alignment: page, - unallocated: 0, - minUnallocated: 0, + desc: "Initial allocation succeeds", + usage: &usageSegmentDataSlices{}, + length: page, + alignment: page, + start: chunkSize - page, // Grows by chunkSize, allocate down. }, { - desc: "Allocation begins at start of file", + desc: "Allocation finds empty space at start of file", usage: &usageSegmentDataSlices{ Start: []uint64{page}, End: []uint64{2 * page}, Values: []usageInfo{{refs: 1}}, }, - start: 0, - length: page, - alignment: page, - unallocated: 0, - minUnallocated: 0, + fileSize: 2 * page, + length: page, + alignment: page, + start: 0, + }, + { + desc: "Allocation finds empty space at end of file", + usage: &usageSegmentDataSlices{ + Start: []uint64{0}, + End: []uint64{page}, + Values: []usageInfo{{refs: 1}}, + }, + fileSize: 2 * page, + length: page, + alignment: page, + start: page, }, { desc: "In-use frames are not allocatable", @@ -64,11 +74,10 @@ func TestFindUnallocatedRange(t *testing.T) { End: []uint64{page, 2 * page}, Values: []usageInfo{{refs: 1}, {refs: 2}}, }, - start: 0, - length: page, - alignment: page, - unallocated: 2 * page, - minUnallocated: 2 * page, + fileSize: 2 * page, + length: page, + alignment: page, + start: 3 * page, // Double fileSize, allocate top-down. }, { desc: "Reclaimable frames are not allocatable", @@ -77,11 +86,10 @@ func TestFindUnallocatedRange(t *testing.T) { End: []uint64{page, 2 * page, 3 * page}, Values: []usageInfo{{refs: 1}, {refs: 0}, {refs: 1}}, }, - start: 0, - length: page, - alignment: page, - unallocated: 3 * page, - minUnallocated: 3 * page, + fileSize: 3 * page, + length: page, + alignment: page, + start: 5 * page, // Double fileSize, grow down. }, { desc: "Gaps between in-use frames are allocatable", @@ -90,11 +98,10 @@ func TestFindUnallocatedRange(t *testing.T) { End: []uint64{page, 3 * page}, Values: []usageInfo{{refs: 1}, {refs: 1}}, }, - start: 0, - length: page, - alignment: page, - unallocated: page, - minUnallocated: page, + fileSize: 3 * page, + length: page, + alignment: page, + start: page, }, { desc: "Inadequately-sized gaps are rejected", @@ -103,14 +110,13 @@ func TestFindUnallocatedRange(t *testing.T) { End: []uint64{page, 3 * page}, Values: []usageInfo{{refs: 1}, {refs: 1}}, }, - start: 0, - length: 2 * page, - alignment: page, - unallocated: 3 * page, - minUnallocated: page, + fileSize: 3 * page, + length: 2 * page, + alignment: page, + start: 4 * page, // Double fileSize, grow down. }, { - desc: "Hugepage alignment is honored", + desc: "Alignment is honored at end of file", usage: &usageSegmentDataSlices{ Start: []uint64{0, hugepage + page}, // Hugepage-sized gap here that shouldn't be allocated from @@ -118,37 +124,95 @@ func TestFindUnallocatedRange(t *testing.T) { End: []uint64{page, hugepage + 2*page}, Values: []usageInfo{{refs: 1}, {refs: 1}}, }, - start: 0, - length: hugepage, - alignment: hugepage, - unallocated: 2 * hugepage, - minUnallocated: page, + fileSize: hugepage + 2*page, + length: hugepage, + alignment: hugepage, + start: 3 * hugepage, // Double fileSize until alignment is satisfied, grow down. + }, + { + desc: "Alignment is honored before end of file", + usage: &usageSegmentDataSlices{ + Start: []uint64{0, 2*hugepage + page}, + // Page will need to be shifted down from top. + End: []uint64{page, 2*hugepage + 2*page}, + Values: []usageInfo{{refs: 1}, {refs: 1}}, + }, + fileSize: 2*hugepage + 2*page, + length: hugepage, + alignment: hugepage, + start: hugepage, }, { - desc: "Pages before start ignored", + desc: "Allocations are compact if possible", usage: &usageSegmentDataSlices{ Start: []uint64{page, 3 * page}, End: []uint64{2 * page, 4 * page}, Values: []usageInfo{{refs: 1}, {refs: 2}}, }, - start: page, - length: page, - alignment: page, - unallocated: 2 * page, - minUnallocated: 2 * page, + fileSize: 4 * page, + length: page, + alignment: page, + start: 2 * page, + }, + { + desc: "Top-down allocation within one gap", + usage: &usageSegmentDataSlices{ + Start: []uint64{page, 4 * page, 7 * page}, + End: []uint64{2 * page, 5 * page, 8 * page}, + Values: []usageInfo{{refs: 1}, {refs: 2}, {refs: 1}}, + }, + fileSize: 8 * page, + length: page, + alignment: page, + start: 6 * page, + }, + { + desc: "Top-down allocation between multiple gaps", + usage: &usageSegmentDataSlices{ + Start: []uint64{page, 3 * page, 5 * page}, + End: []uint64{2 * page, 4 * page, 6 * page}, + Values: []usageInfo{{refs: 1}, {refs: 2}, {refs: 1}}, + }, + fileSize: 6 * page, + length: page, + alignment: page, + start: 4 * page, }, { - desc: "start may be in the middle of segment", + desc: "Top-down allocation with large top gap", usage: &usageSegmentDataSlices{ - Start: []uint64{0, 3 * page}, + Start: []uint64{page, 3 * page}, End: []uint64{2 * page, 4 * page}, Values: []usageInfo{{refs: 1}, {refs: 2}}, }, - start: page, - length: page, - alignment: page, - unallocated: 2 * page, - minUnallocated: 2 * page, + fileSize: 8 * page, + length: page, + alignment: page, + start: 7 * page, + }, + { + desc: "Gaps found with possible overflow", + usage: &usageSegmentDataSlices{ + Start: []uint64{page, topPage - page}, + End: []uint64{2 * page, topPage}, + Values: []usageInfo{{refs: 1}, {refs: 1}}, + }, + fileSize: topPage, + length: page, + alignment: page, + start: topPage - 2*page, + }, + { + desc: "Overflow detected", + usage: &usageSegmentDataSlices{ + Start: []uint64{page}, + End: []uint64{topPage}, + Values: []usageInfo{{refs: 1}}, + }, + fileSize: topPage, + length: 2 * page, + alignment: page, + expectFail: true, }, } { t.Run(test.desc, func(t *testing.T) { @@ -156,12 +220,18 @@ func TestFindUnallocatedRange(t *testing.T) { if err := usage.ImportSortedSlices(test.usage); err != nil { t.Fatalf("Failed to initialize usage from %v: %v", test.usage, err) } - unallocated, minUnallocated := findUnallocatedRange(&usage, test.start, test.length, test.alignment) - if unallocated != test.unallocated { - t.Errorf("findUnallocatedRange(%v, %x, %x, %x): got unallocated %x, wanted %x", test.usage, test.start, test.length, test.alignment, unallocated, test.unallocated) + fr, ok := findAvailableRange(&usage, test.fileSize, test.length, test.alignment) + if !test.expectFail && !ok { + t.Fatalf("findAvailableRange(%v, %x, %x, %x): got %x, false wanted %x, true", test.usage, test.fileSize, test.length, test.alignment, fr.Start, test.start) + } + if test.expectFail && ok { + t.Fatalf("findAvailableRange(%v, %x, %x, %x): got %x, true wanted %x, false", test.usage, test.fileSize, test.length, test.alignment, fr.Start, test.start) + } + if ok && fr.Start != test.start { + t.Errorf("findAvailableRange(%v, %x, %x, %x): got start=%x, wanted %x", test.usage, test.fileSize, test.length, test.alignment, fr.Start, test.start) } - if minUnallocated != test.minUnallocated { - t.Errorf("findUnallocatedRange(%v, %x, %x, %x): got minUnallocated %x, wanted %x", test.usage, test.start, test.length, test.alignment, minUnallocated, test.minUnallocated) + if ok && fr.End != test.start+test.length { + t.Errorf("findAvailableRange(%v, %x, %x, %x): got end=%x, wanted %x", test.usage, test.fileSize, test.length, test.alignment, fr.End, test.start+test.length) } }) } diff --git a/test/e2e/integration_test.go b/test/e2e/integration_test.go index 9cbb2ed5b..91c956e10 100644 --- a/test/e2e/integration_test.go +++ b/test/e2e/integration_test.go @@ -239,7 +239,9 @@ func TestMemLimit(t *testing.T) { d := dockerutil.MakeDocker(t) defer d.CleanUp() - allocMemory := 500 * 1024 + // N.B. Because the size of the memory file may grow in large chunks, + // there is a minimum threshold of 1GB for the MemTotal figure. + allocMemory := 1024 * 1024 out, err := d.Run(dockerutil.RunOpts{ Image: "basic/alpine", Memory: allocMemory, // In kB. -- cgit v1.2.3 From 527d08f6afdea1e142c76b8abb2266525a98c2ea Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Fri, 5 Jun 2020 17:24:12 -0700 Subject: Add +checkescape annotations to kvm/ring0. This analysis also catches a potential bug, which is a split on mapPhysical. This would have led to potential guest-exit during Mapping (although this would have been handled by the now-unecessary retryInGuest loop). PiperOrigin-RevId: 315025106 --- pkg/sentry/platform/kvm/BUILD | 2 +- pkg/sentry/platform/kvm/address_space.go | 73 +++++++++------ pkg/sentry/platform/kvm/allocator.go | 76 ---------------- pkg/sentry/platform/kvm/bluepill_allocator.go | 100 +++++++++++++++++++++ pkg/sentry/platform/kvm/bluepill_amd64.go | 12 ++- pkg/sentry/platform/kvm/bluepill_arm64.go | 4 + pkg/sentry/platform/kvm/bluepill_unsafe.go | 9 +- pkg/sentry/platform/kvm/machine.go | 52 ++++++----- pkg/sentry/platform/kvm/machine_amd64.go | 34 ++----- pkg/sentry/platform/kvm/machine_amd64_unsafe.go | 2 +- pkg/sentry/platform/kvm/machine_unsafe.go | 2 +- pkg/sentry/platform/ring0/kernel.go | 24 ++++- pkg/sentry/platform/ring0/kernel_amd64.go | 12 +-- pkg/sentry/platform/ring0/pagetables/allocator.go | 11 ++- pkg/sentry/platform/ring0/pagetables/pagetables.go | 8 ++ 15 files changed, 244 insertions(+), 177 deletions(-) delete mode 100644 pkg/sentry/platform/kvm/allocator.go create mode 100644 pkg/sentry/platform/kvm/bluepill_allocator.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/kvm/BUILD b/pkg/sentry/platform/kvm/BUILD index 159f7eafd..4792454c4 100644 --- a/pkg/sentry/platform/kvm/BUILD +++ b/pkg/sentry/platform/kvm/BUILD @@ -6,8 +6,8 @@ go_library( name = "kvm", srcs = [ "address_space.go", - "allocator.go", "bluepill.go", + "bluepill_allocator.go", "bluepill_amd64.go", "bluepill_amd64.s", "bluepill_amd64_unsafe.go", diff --git a/pkg/sentry/platform/kvm/address_space.go b/pkg/sentry/platform/kvm/address_space.go index be213bfe8..faf1d5e1c 100644 --- a/pkg/sentry/platform/kvm/address_space.go +++ b/pkg/sentry/platform/kvm/address_space.go @@ -26,16 +26,15 @@ import ( // dirtySet tracks vCPUs for invalidation. type dirtySet struct { - vCPUs []uint64 + vCPUMasks []uint64 } // forEach iterates over all CPUs in the dirty set. +// +//go:nosplit func (ds *dirtySet) forEach(m *machine, fn func(c *vCPU)) { - m.mu.RLock() - defer m.mu.RUnlock() - - for index := range ds.vCPUs { - mask := atomic.SwapUint64(&ds.vCPUs[index], 0) + for index := range ds.vCPUMasks { + mask := atomic.SwapUint64(&ds.vCPUMasks[index], 0) if mask != 0 { for bit := 0; bit < 64; bit++ { if mask&(1< 0 { physical, length, ok := translateToPhysical(m.addr) if !ok { @@ -133,18 +137,10 @@ func (as *addressSpace) mapHost(addr usermem.Addr, m hostMapEntry, at usermem.Ac // important; if the pagetable mappings were installed before // ensuring the physical pages were available, then some other // thread could theoretically access them. - // - // Due to the way KVM's shadow paging implementation works, - // modifications to the page tables while in host mode may not - // be trapped, leading to the shadow pages being out of sync. - // Therefore, we need to ensure that we are in guest mode for - // page table modifications. See the call to bluepill, below. - as.machine.retryInGuest(func() { - inv = as.pageTables.Map(addr, length, pagetables.MapOpts{ - AccessType: at, - User: true, - }, physical) || inv - }) + inv = as.pageTables.Map(addr, length, pagetables.MapOpts{ + AccessType: at, + User: true, + }, physical) || inv m.addr += length m.length -= length addr += usermem.Addr(length) @@ -176,6 +172,10 @@ func (as *addressSpace) MapFile(addr usermem.Addr, f platform.File, fr platform. return err } + // See block in mapLocked. + as.pageTables.Allocator.(*allocator).cpu = as.machine.Get() + defer as.machine.Put(as.pageTables.Allocator.(*allocator).cpu) + // Map the mappings in the sentry's address space (guest physical memory) // into the application's address space (guest virtual memory). inv := false @@ -190,7 +190,12 @@ func (as *addressSpace) MapFile(addr usermem.Addr, f platform.File, fr platform. _ = s[i] // Touch to commit. } } - prev := as.mapHost(addr, hostMapEntry{ + + // See bluepill_allocator.go. + bluepill(as.pageTables.Allocator.(*allocator).cpu) + + // Perform the mapping. + prev := as.mapLocked(addr, hostMapEntry{ addr: b.Addr(), length: uintptr(b.Len()), }, at) @@ -204,17 +209,27 @@ func (as *addressSpace) MapFile(addr usermem.Addr, f platform.File, fr platform. return nil } +// unmapLocked is an escape-checked wrapped around Unmap. +// +// +checkescape:hard,stack +// +//go:nosplit +func (as *addressSpace) unmapLocked(addr usermem.Addr, length uint64) bool { + return as.pageTables.Unmap(addr, uintptr(length)) +} + // Unmap unmaps the given range by calling pagetables.PageTables.Unmap. func (as *addressSpace) Unmap(addr usermem.Addr, length uint64) { as.mu.Lock() defer as.mu.Unlock() - // See above re: retryInGuest. - var prev bool - as.machine.retryInGuest(func() { - prev = as.pageTables.Unmap(addr, uintptr(length)) || prev - }) - if prev { + // See above & bluepill_allocator.go. + as.pageTables.Allocator.(*allocator).cpu = as.machine.Get() + defer as.machine.Put(as.pageTables.Allocator.(*allocator).cpu) + bluepill(as.pageTables.Allocator.(*allocator).cpu) + + if prev := as.unmapLocked(addr, length); prev { + // Invalidate all active vCPUs. as.invalidate() // Recycle any freed intermediate pages. @@ -227,7 +242,7 @@ func (as *addressSpace) Release() { as.Unmap(0, ^uint64(0)) // Free all pages from the allocator. - as.pageTables.Allocator.(allocator).base.Drain() + as.pageTables.Allocator.(*allocator).base.Drain() // Drop all cached machine references. as.machine.dropPageTables(as.pageTables) diff --git a/pkg/sentry/platform/kvm/allocator.go b/pkg/sentry/platform/kvm/allocator.go deleted file mode 100644 index 3f35414bb..000000000 --- a/pkg/sentry/platform/kvm/allocator.go +++ /dev/null @@ -1,76 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package kvm - -import ( - "fmt" - - "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" -) - -type allocator struct { - base *pagetables.RuntimeAllocator -} - -// newAllocator is used to define the allocator. -func newAllocator() allocator { - return allocator{ - base: pagetables.NewRuntimeAllocator(), - } -} - -// NewPTEs implements pagetables.Allocator.NewPTEs. -// -//go:nosplit -func (a allocator) NewPTEs() *pagetables.PTEs { - return a.base.NewPTEs() -} - -// PhysicalFor returns the physical address for a set of PTEs. -// -//go:nosplit -func (a allocator) PhysicalFor(ptes *pagetables.PTEs) uintptr { - virtual := a.base.PhysicalFor(ptes) - physical, _, ok := translateToPhysical(virtual) - if !ok { - panic(fmt.Sprintf("PhysicalFor failed for %p", ptes)) - } - return physical -} - -// LookupPTEs implements pagetables.Allocator.LookupPTEs. -// -//go:nosplit -func (a allocator) LookupPTEs(physical uintptr) *pagetables.PTEs { - virtualStart, physicalStart, _, ok := calculateBluepillFault(physical, physicalRegions) - if !ok { - panic(fmt.Sprintf("LookupPTEs failed for 0x%x", physical)) - } - return a.base.LookupPTEs(virtualStart + (physical - physicalStart)) -} - -// FreePTEs implements pagetables.Allocator.FreePTEs. -// -//go:nosplit -func (a allocator) FreePTEs(ptes *pagetables.PTEs) { - a.base.FreePTEs(ptes) -} - -// Recycle implements pagetables.Allocator.Recycle. -// -//go:nosplit -func (a allocator) Recycle() { - a.base.Recycle() -} diff --git a/pkg/sentry/platform/kvm/bluepill_allocator.go b/pkg/sentry/platform/kvm/bluepill_allocator.go new file mode 100644 index 000000000..9485e1301 --- /dev/null +++ b/pkg/sentry/platform/kvm/bluepill_allocator.go @@ -0,0 +1,100 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package kvm + +import ( + "fmt" + + "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" +) + +type allocator struct { + base pagetables.RuntimeAllocator + + // cpu must be set prior to any pagetable operation. + // + // Due to the way KVM's shadow paging implementation works, + // modifications to the page tables while in host mode may not be + // trapped, leading to the shadow pages being out of sync. Therefore, + // we need to ensure that we are in guest mode for page table + // modifications. See the call to bluepill, below. + cpu *vCPU +} + +// newAllocator is used to define the allocator. +func newAllocator() *allocator { + a := new(allocator) + a.base.Init() + return a +} + +// NewPTEs implements pagetables.Allocator.NewPTEs. +// +// +checkescape:all +// +//go:nosplit +func (a *allocator) NewPTEs() *pagetables.PTEs { + ptes := a.base.NewPTEs() // escapes: bluepill below. + if a.cpu != nil { + bluepill(a.cpu) + } + return ptes +} + +// PhysicalFor returns the physical address for a set of PTEs. +// +// +checkescape:all +// +//go:nosplit +func (a *allocator) PhysicalFor(ptes *pagetables.PTEs) uintptr { + virtual := a.base.PhysicalFor(ptes) + physical, _, ok := translateToPhysical(virtual) + if !ok { + panic(fmt.Sprintf("PhysicalFor failed for %p", ptes)) // escapes: panic. + } + return physical +} + +// LookupPTEs implements pagetables.Allocator.LookupPTEs. +// +// +checkescape:all +// +//go:nosplit +func (a *allocator) LookupPTEs(physical uintptr) *pagetables.PTEs { + virtualStart, physicalStart, _, ok := calculateBluepillFault(physical, physicalRegions) + if !ok { + panic(fmt.Sprintf("LookupPTEs failed for 0x%x", physical)) // escapes: panic. + } + return a.base.LookupPTEs(virtualStart + (physical - physicalStart)) +} + +// FreePTEs implements pagetables.Allocator.FreePTEs. +// +// +checkescape:all +// +//go:nosplit +func (a *allocator) FreePTEs(ptes *pagetables.PTEs) { + a.base.FreePTEs(ptes) // escapes: bluepill below. + if a.cpu != nil { + bluepill(a.cpu) + } +} + +// Recycle implements pagetables.Allocator.Recycle. +// +//go:nosplit +func (a *allocator) Recycle() { + a.base.Recycle() +} diff --git a/pkg/sentry/platform/kvm/bluepill_amd64.go b/pkg/sentry/platform/kvm/bluepill_amd64.go index 133c2203d..ddc1554d5 100644 --- a/pkg/sentry/platform/kvm/bluepill_amd64.go +++ b/pkg/sentry/platform/kvm/bluepill_amd64.go @@ -63,6 +63,8 @@ func bluepillArchEnter(context *arch.SignalContext64) *vCPU { // KernelSyscall handles kernel syscalls. // +// +checkescape:all +// //go:nosplit func (c *vCPU) KernelSyscall() { regs := c.Registers() @@ -72,13 +74,15 @@ func (c *vCPU) KernelSyscall() { // We only trigger a bluepill entry in the bluepill function, and can // therefore be guaranteed that there is no floating point state to be // loaded on resuming from halt. We only worry about saving on exit. - ring0.SaveFloatingPoint((*byte)(c.floatingPointState)) + ring0.SaveFloatingPoint((*byte)(c.floatingPointState)) // escapes: no. ring0.Halt() - ring0.WriteFS(uintptr(regs.Fs_base)) // Reload host segment. + ring0.WriteFS(uintptr(regs.Fs_base)) // escapes: no, reload host segment. } // KernelException handles kernel exceptions. // +// +checkescape:all +// //go:nosplit func (c *vCPU) KernelException(vector ring0.Vector) { regs := c.Registers() @@ -89,9 +93,9 @@ func (c *vCPU) KernelException(vector ring0.Vector) { regs.Rip = 0 } // See above. - ring0.SaveFloatingPoint((*byte)(c.floatingPointState)) + ring0.SaveFloatingPoint((*byte)(c.floatingPointState)) // escapes: no. ring0.Halt() - ring0.WriteFS(uintptr(regs.Fs_base)) // Reload host segment. + ring0.WriteFS(uintptr(regs.Fs_base)) // escapes: no; reload host segment. } // bluepillArchExit is called during bluepillEnter. diff --git a/pkg/sentry/platform/kvm/bluepill_arm64.go b/pkg/sentry/platform/kvm/bluepill_arm64.go index c215d443c..83643c602 100644 --- a/pkg/sentry/platform/kvm/bluepill_arm64.go +++ b/pkg/sentry/platform/kvm/bluepill_arm64.go @@ -66,6 +66,8 @@ func bluepillArchExit(c *vCPU, context *arch.SignalContext64) { // KernelSyscall handles kernel syscalls. // +// +checkescape:all +// //go:nosplit func (c *vCPU) KernelSyscall() { regs := c.Registers() @@ -88,6 +90,8 @@ func (c *vCPU) KernelSyscall() { // KernelException handles kernel exceptions. // +// +checkescape:all +// //go:nosplit func (c *vCPU) KernelException(vector ring0.Vector) { regs := c.Registers() diff --git a/pkg/sentry/platform/kvm/bluepill_unsafe.go b/pkg/sentry/platform/kvm/bluepill_unsafe.go index 2407014e9..c025aa0bb 100644 --- a/pkg/sentry/platform/kvm/bluepill_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_unsafe.go @@ -64,6 +64,8 @@ func bluepillArchContext(context unsafe.Pointer) *arch.SignalContext64 { // signal stack. It should only execute raw system calls and functions that are // explicitly marked go:nosplit. // +// +checkescape:all +// //go:nosplit func bluepillHandler(context unsafe.Pointer) { // Sanitize the registers; interrupts must always be disabled. @@ -82,7 +84,8 @@ func bluepillHandler(context unsafe.Pointer) { } for { - switch _, _, errno := syscall.RawSyscall(syscall.SYS_IOCTL, uintptr(c.fd), _KVM_RUN, 0); errno { + _, _, errno := syscall.RawSyscall(syscall.SYS_IOCTL, uintptr(c.fd), _KVM_RUN, 0) // escapes: no. + switch errno { case 0: // Expected case. case syscall.EINTR: // First, we process whatever pending signal @@ -90,7 +93,7 @@ func bluepillHandler(context unsafe.Pointer) { // currently, all signals are masked and the signal // must have been delivered directly to this thread. timeout := syscall.Timespec{} - sig, _, errno := syscall.RawSyscall6( + sig, _, errno := syscall.RawSyscall6( // escapes: no. syscall.SYS_RT_SIGTIMEDWAIT, uintptr(unsafe.Pointer(&bounceSignalMask)), 0, // siginfo. @@ -125,7 +128,7 @@ func bluepillHandler(context unsafe.Pointer) { // MMIO exit we receive EFAULT from the run ioctl. We // always inject an NMI here since we may be in kernel // mode and have interrupts disabled. - if _, _, errno := syscall.RawSyscall( + if _, _, errno := syscall.RawSyscall( // escapes: no. syscall.SYS_IOCTL, uintptr(c.fd), _KVM_NMI, 0); errno != 0 { diff --git a/pkg/sentry/platform/kvm/machine.go b/pkg/sentry/platform/kvm/machine.go index f1afc74dc..6c54712d1 100644 --- a/pkg/sentry/platform/kvm/machine.go +++ b/pkg/sentry/platform/kvm/machine.go @@ -52,16 +52,19 @@ type machine struct { // available is notified when vCPUs are available. available sync.Cond - // vCPUs are the machine vCPUs. + // vCPUsByTID are the machine vCPUs. // // These are populated dynamically. - vCPUs map[uint64]*vCPU + vCPUsByTID map[uint64]*vCPU // vCPUsByID are the machine vCPUs, can be indexed by the vCPU's ID. - vCPUsByID map[int]*vCPU + vCPUsByID []*vCPU // maxVCPUs is the maximum number of vCPUs supported by the machine. maxVCPUs int + + // nextID is the next vCPU ID. + nextID uint32 } const ( @@ -137,9 +140,8 @@ type dieState struct { // // Precondition: mu must be held. func (m *machine) newVCPU() *vCPU { - id := len(m.vCPUs) - // Create the vCPU. + id := int(atomic.AddUint32(&m.nextID, 1) - 1) fd, _, errno := syscall.RawSyscall(syscall.SYS_IOCTL, uintptr(m.fd), _KVM_CREATE_VCPU, uintptr(id)) if errno != 0 { panic(fmt.Sprintf("error creating new vCPU: %v", errno)) @@ -176,11 +178,7 @@ func (m *machine) newVCPU() *vCPU { // newMachine returns a new VM context. func newMachine(vm int) (*machine, error) { // Create the machine. - m := &machine{ - fd: vm, - vCPUs: make(map[uint64]*vCPU), - vCPUsByID: make(map[int]*vCPU), - } + m := &machine{fd: vm} m.available.L = &m.mu m.kernel.Init(ring0.KernelOpts{ PageTables: pagetables.New(newAllocator()), @@ -194,6 +192,10 @@ func newMachine(vm int) (*machine, error) { } log.Debugf("The maximum number of vCPUs is %d.", m.maxVCPUs) + // Create the vCPUs map/slices. + m.vCPUsByTID = make(map[uint64]*vCPU) + m.vCPUsByID = make([]*vCPU, m.maxVCPUs) + // Apply the physical mappings. Note that these mappings may point to // guest physical addresses that are not actually available. These // physical pages are mapped on demand, see kernel_unsafe.go. @@ -274,6 +276,8 @@ func newMachine(vm int) (*machine, error) { // not available. This attempts to be efficient for calls in the hot path. // // This panics on error. +// +//go:nosplit func (m *machine) mapPhysical(physical, length uintptr, phyRegions []physicalRegion, flags uint32) { for end := physical + length; physical < end; { _, physicalStart, length, ok := calculateBluepillFault(physical, phyRegions) @@ -304,7 +308,11 @@ func (m *machine) Destroy() { runtime.SetFinalizer(m, nil) // Destroy vCPUs. - for _, c := range m.vCPUs { + for _, c := range m.vCPUsByID { + if c == nil { + continue + } + // Ensure the vCPU is not still running in guest mode. This is // possible iff teardown has been done by other threads, and // somehow a single thread has not executed any system calls. @@ -337,7 +345,7 @@ func (m *machine) Get() *vCPU { tid := procid.Current() // Check for an exact match. - if c := m.vCPUs[tid]; c != nil { + if c := m.vCPUsByTID[tid]; c != nil { c.lock() m.mu.RUnlock() return c @@ -356,7 +364,7 @@ func (m *machine) Get() *vCPU { tid = procid.Current() // Recheck for an exact match. - if c := m.vCPUs[tid]; c != nil { + if c := m.vCPUsByTID[tid]; c != nil { c.lock() m.mu.Unlock() return c @@ -364,10 +372,10 @@ func (m *machine) Get() *vCPU { for { // Scan for an available vCPU. - for origTID, c := range m.vCPUs { + for origTID, c := range m.vCPUsByTID { if atomic.CompareAndSwapUint32(&c.state, vCPUReady, vCPUUser) { - delete(m.vCPUs, origTID) - m.vCPUs[tid] = c + delete(m.vCPUsByTID, origTID) + m.vCPUsByTID[tid] = c m.mu.Unlock() c.loadSegments(tid) return c @@ -375,17 +383,17 @@ func (m *machine) Get() *vCPU { } // Create a new vCPU (maybe). - if len(m.vCPUs) < m.maxVCPUs { + if int(m.nextID) < m.maxVCPUs { c := m.newVCPU() c.lock() - m.vCPUs[tid] = c + m.vCPUsByTID[tid] = c m.mu.Unlock() c.loadSegments(tid) return c } // Scan for something not in user mode. - for origTID, c := range m.vCPUs { + for origTID, c := range m.vCPUsByTID { if !atomic.CompareAndSwapUint32(&c.state, vCPUGuest, vCPUGuest|vCPUWaiter) { continue } @@ -403,8 +411,8 @@ func (m *machine) Get() *vCPU { } // Steal the vCPU. - delete(m.vCPUs, origTID) - m.vCPUs[tid] = c + delete(m.vCPUsByTID, origTID) + m.vCPUsByTID[tid] = c m.mu.Unlock() c.loadSegments(tid) return c @@ -431,7 +439,7 @@ func (m *machine) Put(c *vCPU) { // newDirtySet returns a new dirty set. func (m *machine) newDirtySet() *dirtySet { return &dirtySet{ - vCPUs: make([]uint64, (m.maxVCPUs+63)/64, (m.maxVCPUs+63)/64), + vCPUMasks: make([]uint64, (m.maxVCPUs+63)/64, (m.maxVCPUs+63)/64), } } diff --git a/pkg/sentry/platform/kvm/machine_amd64.go b/pkg/sentry/platform/kvm/machine_amd64.go index 923ce3909..acc823ba6 100644 --- a/pkg/sentry/platform/kvm/machine_amd64.go +++ b/pkg/sentry/platform/kvm/machine_amd64.go @@ -51,9 +51,10 @@ func (m *machine) initArchState() error { recover() debug.SetPanicOnFault(old) }() - m.retryInGuest(func() { - ring0.SetCPUIDFaulting(true) - }) + c := m.Get() + defer m.Put(c) + bluepill(c) + ring0.SetCPUIDFaulting(true) return nil } @@ -89,8 +90,8 @@ func (m *machine) dropPageTables(pt *pagetables.PageTables) { defer m.mu.Unlock() // Clear from all PCIDs. - for _, c := range m.vCPUs { - if c.PCIDs != nil { + for _, c := range m.vCPUsByID { + if c != nil && c.PCIDs != nil { c.PCIDs.Drop(pt) } } @@ -335,29 +336,6 @@ func (c *vCPU) SwitchToUser(switchOpts ring0.SwitchOpts, info *arch.SignalInfo) } } -// retryInGuest runs the given function in guest mode. -// -// If the function does not complete in guest mode (due to execution of a -// system call due to a GC stall, for example), then it will be retried. The -// given function must be idempotent as a result of the retry mechanism. -func (m *machine) retryInGuest(fn func()) { - c := m.Get() - defer m.Put(c) - for { - c.ClearErrorCode() // See below. - bluepill(c) // Force guest mode. - fn() // Execute the given function. - _, user := c.ErrorCode() - if user { - // If user is set, then we haven't bailed back to host - // mode via a kernel exception or system call. We - // consider the full function to have executed in guest - // mode and we can return. - break - } - } -} - // On x86 platform, the flags for "setMemoryRegion" can always be set as 0. // There is no need to return read-only physicalRegions. func rdonlyRegionsForSetMem() (phyRegions []physicalRegion) { diff --git a/pkg/sentry/platform/kvm/machine_amd64_unsafe.go b/pkg/sentry/platform/kvm/machine_amd64_unsafe.go index 7156c245f..290f035dd 100644 --- a/pkg/sentry/platform/kvm/machine_amd64_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_amd64_unsafe.go @@ -154,7 +154,7 @@ func (c *vCPU) setUserRegisters(uregs *userRegs) error { // //go:nosplit func (c *vCPU) getUserRegisters(uregs *userRegs) syscall.Errno { - if _, _, errno := syscall.RawSyscall( + if _, _, errno := syscall.RawSyscall( // escapes: no. syscall.SYS_IOCTL, uintptr(c.fd), _KVM_GET_REGS, diff --git a/pkg/sentry/platform/kvm/machine_unsafe.go b/pkg/sentry/platform/kvm/machine_unsafe.go index de7df4f80..9f86f6a7a 100644 --- a/pkg/sentry/platform/kvm/machine_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_unsafe.go @@ -115,7 +115,7 @@ func (a *atomicAddressSpace) get() *addressSpace { // //go:nosplit func (c *vCPU) notify() { - _, _, errno := syscall.RawSyscall6( + _, _, errno := syscall.RawSyscall6( // escapes: no. syscall.SYS_FUTEX, uintptr(unsafe.Pointer(&c.state)), linux.FUTEX_WAKE|linux.FUTEX_PRIVATE_FLAG, diff --git a/pkg/sentry/platform/ring0/kernel.go b/pkg/sentry/platform/ring0/kernel.go index 900c0bba7..021693791 100644 --- a/pkg/sentry/platform/ring0/kernel.go +++ b/pkg/sentry/platform/ring0/kernel.go @@ -31,23 +31,39 @@ type defaultHooks struct{} // KernelSyscall implements Hooks.KernelSyscall. // +// +checkescape:all +// //go:nosplit -func (defaultHooks) KernelSyscall() { Halt() } +func (defaultHooks) KernelSyscall() { + Halt() +} // KernelException implements Hooks.KernelException. // +// +checkescape:all +// //go:nosplit -func (defaultHooks) KernelException(Vector) { Halt() } +func (defaultHooks) KernelException(Vector) { + Halt() +} // kernelSyscall is a trampoline. // +// +checkescape:hard,stack +// //go:nosplit -func kernelSyscall(c *CPU) { c.hooks.KernelSyscall() } +func kernelSyscall(c *CPU) { + c.hooks.KernelSyscall() +} // kernelException is a trampoline. // +// +checkescape:hard,stack +// //go:nosplit -func kernelException(c *CPU, vector Vector) { c.hooks.KernelException(vector) } +func kernelException(c *CPU, vector Vector) { + c.hooks.KernelException(vector) +} // Init initializes a new CPU. // diff --git a/pkg/sentry/platform/ring0/kernel_amd64.go b/pkg/sentry/platform/ring0/kernel_amd64.go index 0feff8778..d37981dbf 100644 --- a/pkg/sentry/platform/ring0/kernel_amd64.go +++ b/pkg/sentry/platform/ring0/kernel_amd64.go @@ -178,6 +178,8 @@ func IsCanonical(addr uint64) bool { // // Precondition: the Rip, Rsp, Fs and Gs registers must be canonical. // +// +checkescape:all +// //go:nosplit func (c *CPU) SwitchToUser(switchOpts SwitchOpts) (vector Vector) { userCR3 := switchOpts.PageTables.CR3(!switchOpts.Flush, switchOpts.UserPCID) @@ -192,9 +194,9 @@ func (c *CPU) SwitchToUser(switchOpts SwitchOpts) (vector Vector) { // Perform the switch. swapgs() // GS will be swapped on return. - WriteFS(uintptr(regs.Fs_base)) // Set application FS. - WriteGS(uintptr(regs.Gs_base)) // Set application GS. - LoadFloatingPoint(switchOpts.FloatingPointState) // Copy in floating point. + WriteFS(uintptr(regs.Fs_base)) // escapes: no. Set application FS. + WriteGS(uintptr(regs.Gs_base)) // escapes: no. Set application GS. + LoadFloatingPoint(switchOpts.FloatingPointState) // escapes: no. Copy in floating point. jumpToKernel() // Switch to upper half. writeCR3(uintptr(userCR3)) // Change to user address space. if switchOpts.FullRestore { @@ -204,8 +206,8 @@ func (c *CPU) SwitchToUser(switchOpts SwitchOpts) (vector Vector) { } writeCR3(uintptr(kernelCR3)) // Return to kernel address space. jumpToUser() // Return to lower half. - SaveFloatingPoint(switchOpts.FloatingPointState) // Copy out floating point. - WriteFS(uintptr(c.registers.Fs_base)) // Restore kernel FS. + SaveFloatingPoint(switchOpts.FloatingPointState) // escapes: no. Copy out floating point. + WriteFS(uintptr(c.registers.Fs_base)) // escapes: no. Restore kernel FS. return } diff --git a/pkg/sentry/platform/ring0/pagetables/allocator.go b/pkg/sentry/platform/ring0/pagetables/allocator.go index 23fd5c352..8d75b7599 100644 --- a/pkg/sentry/platform/ring0/pagetables/allocator.go +++ b/pkg/sentry/platform/ring0/pagetables/allocator.go @@ -53,9 +53,14 @@ type RuntimeAllocator struct { // NewRuntimeAllocator returns an allocator that uses runtime allocation. func NewRuntimeAllocator() *RuntimeAllocator { - return &RuntimeAllocator{ - used: make(map[*PTEs]struct{}), - } + r := new(RuntimeAllocator) + r.Init() + return r +} + +// Init initializes a RuntimeAllocator. +func (r *RuntimeAllocator) Init() { + r.used = make(map[*PTEs]struct{}) } // Recycle returns freed pages to the pool. diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables.go b/pkg/sentry/platform/ring0/pagetables/pagetables.go index 87e88e97d..7f18ac296 100644 --- a/pkg/sentry/platform/ring0/pagetables/pagetables.go +++ b/pkg/sentry/platform/ring0/pagetables/pagetables.go @@ -86,6 +86,8 @@ func (*mapVisitor) requiresSplit() bool { return true } // // Precondition: addr & length must be page-aligned, their sum must not overflow. // +// +checkescape:hard,stack +// //go:nosplit func (p *PageTables) Map(addr usermem.Addr, length uintptr, opts MapOpts, physical uintptr) bool { if !opts.AccessType.Any() { @@ -128,6 +130,8 @@ func (v *unmapVisitor) visit(start uintptr, pte *PTE, align uintptr) { // // Precondition: addr & length must be page-aligned. // +// +checkescape:hard,stack +// //go:nosplit func (p *PageTables) Unmap(addr usermem.Addr, length uintptr) bool { w := unmapWalker{ @@ -162,6 +166,8 @@ func (v *emptyVisitor) visit(start uintptr, pte *PTE, align uintptr) { // // Precondition: addr & length must be page-aligned. // +// +checkescape:hard,stack +// //go:nosplit func (p *PageTables) IsEmpty(addr usermem.Addr, length uintptr) bool { w := emptyWalker{ @@ -197,6 +203,8 @@ func (*lookupVisitor) requiresSplit() bool { return false } // Lookup returns the physical address for the given virtual address. // +// +checkescape:hard,stack +// //go:nosplit func (p *PageTables) Lookup(addr usermem.Addr) (physical uintptr, opts MapOpts) { mask := uintptr(usermem.PageSize - 1) -- cgit v1.2.3 From 21b6bc7280f68f43360a008ffd02a4f461ec9fc8 Mon Sep 17 00:00:00 2001 From: Rahat Mahmood Date: Fri, 5 Jun 2020 19:10:28 -0700 Subject: Implement mount(2) and umount2(2) for VFS2. This is mostly syscall plumbing, VFS2 already implements the internals of mounts. In addition to the syscall defintions, the following mount-related mechanisms are updated: - Implement MS_NOATIME for VFS2, but only for tmpfs and goferfs. The other VFS2 filesystems don't implement node-level timestamps yet. - Implement the 'mode', 'uid' and 'gid' mount options for VFS2's tmpfs. - Plumb mount namespace ownership, which is necessary for checking appropriate capabilities during mount(2). Updates #1035 PiperOrigin-RevId: 315035352 --- pkg/sentry/fsimpl/gofer/time.go | 3 + pkg/sentry/fsimpl/tmpfs/tmpfs.go | 41 +++++++- pkg/sentry/kernel/auth/credentials.go | 28 +++++ pkg/sentry/syscalls/linux/vfs2/BUILD | 1 + pkg/sentry/syscalls/linux/vfs2/mount.go | 145 ++++++++++++++++++++++++++ pkg/sentry/syscalls/linux/vfs2/vfs2.go | 4 +- pkg/sentry/vfs/genericfstree/genericfstree.go | 3 +- pkg/sentry/vfs/mount.go | 34 +++--- pkg/sentry/vfs/options.go | 4 + pkg/sentry/vfs/vfs.go | 2 +- runsc/boot/vfs.go | 2 +- 11 files changed, 247 insertions(+), 20 deletions(-) create mode 100644 pkg/sentry/syscalls/linux/vfs2/mount.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/time.go b/pkg/sentry/fsimpl/gofer/time.go index 2608e7e1d..1d5aa82dc 100644 --- a/pkg/sentry/fsimpl/gofer/time.go +++ b/pkg/sentry/fsimpl/gofer/time.go @@ -38,6 +38,9 @@ func statxTimestampFromDentry(ns int64) linux.StatxTimestamp { // Preconditions: fs.interop != InteropModeShared. func (d *dentry) touchAtime(mnt *vfs.Mount) { + if mnt.Flags.NoATime { + return + } if err := mnt.CheckBeginWrite(); err != nil { return } diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index f0e098702..3777ebdf2 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -30,6 +30,7 @@ package tmpfs import ( "fmt" "math" + "strconv" "strings" "sync/atomic" @@ -124,14 +125,45 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt } fs.vfsfs.Init(vfsObj, newFSType, &fs) + mopts := vfs.GenericParseMountOptions(opts.Data) + + defaultMode := linux.FileMode(0777) + if modeStr, ok := mopts["mode"]; ok { + mode, err := strconv.ParseUint(modeStr, 8, 32) + if err != nil { + return nil, nil, fmt.Errorf("Mount option \"mode='%v'\" not parsable: %v", modeStr, err) + } + defaultMode = linux.FileMode(mode) + } + + defaultOwnerCreds := creds.Fork() + if uidStr, ok := mopts["uid"]; ok { + uid, err := strconv.ParseInt(uidStr, 10, 32) + if err != nil { + return nil, nil, fmt.Errorf("Mount option \"uid='%v'\" not parsable: %v", uidStr, err) + } + if err := defaultOwnerCreds.SetUID(auth.UID(uid)); err != nil { + return nil, nil, fmt.Errorf("Error using mount option \"uid='%v'\": %v", uidStr, err) + } + } + if gidStr, ok := mopts["gid"]; ok { + gid, err := strconv.ParseInt(gidStr, 10, 32) + if err != nil { + return nil, nil, fmt.Errorf("Mount option \"gid='%v'\" not parsable: %v", gidStr, err) + } + if err := defaultOwnerCreds.SetGID(auth.GID(gid)); err != nil { + return nil, nil, fmt.Errorf("Error using mount option \"gid='%v'\": %v", gidStr, err) + } + } + var root *dentry switch rootFileType { case linux.S_IFREG: - root = fs.newDentry(fs.newRegularFile(creds, 0777)) + root = fs.newDentry(fs.newRegularFile(defaultOwnerCreds, defaultMode)) case linux.S_IFLNK: - root = fs.newDentry(fs.newSymlink(creds, tmpfsOpts.RootSymlinkTarget)) + root = fs.newDentry(fs.newSymlink(defaultOwnerCreds, tmpfsOpts.RootSymlinkTarget)) case linux.S_IFDIR: - root = &fs.newDirectory(creds, 01777).dentry + root = &fs.newDirectory(defaultOwnerCreds, defaultMode).dentry default: fs.vfsfs.DecRef() return nil, nil, fmt.Errorf("invalid tmpfs root file type: %#o", rootFileType) @@ -562,6 +594,9 @@ func (i *inode) isDir() bool { } func (i *inode) touchAtime(mnt *vfs.Mount) { + if mnt.Flags.NoATime { + return + } if err := mnt.CheckBeginWrite(); err != nil { return } diff --git a/pkg/sentry/kernel/auth/credentials.go b/pkg/sentry/kernel/auth/credentials.go index e057d2c6d..6862f2ef5 100644 --- a/pkg/sentry/kernel/auth/credentials.go +++ b/pkg/sentry/kernel/auth/credentials.go @@ -232,3 +232,31 @@ func (c *Credentials) UseGID(gid GID) (KGID, error) { } return NoID, syserror.EPERM } + +// SetUID translates the provided uid to the root user namespace and updates c's +// uids to it. This performs no permissions or capabilities checks, the caller +// is responsible for ensuring the calling context is permitted to modify c. +func (c *Credentials) SetUID(uid UID) error { + kuid := c.UserNamespace.MapToKUID(uid) + if !kuid.Ok() { + return syserror.EINVAL + } + c.RealKUID = kuid + c.EffectiveKUID = kuid + c.SavedKUID = kuid + return nil +} + +// SetGID translates the provided gid to the root user namespace and updates c's +// gids to it. This performs no permissions or capabilities checks, the caller +// is responsible for ensuring the calling context is permitted to modify c. +func (c *Credentials) SetGID(gid GID) error { + kgid := c.UserNamespace.MapToKGID(gid) + if !kgid.Ok() { + return syserror.EINVAL + } + c.RealKGID = kgid + c.EffectiveKGID = kgid + c.SavedKGID = kgid + return nil +} diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD index 9c8b44f64..c0d005247 100644 --- a/pkg/sentry/syscalls/linux/vfs2/BUILD +++ b/pkg/sentry/syscalls/linux/vfs2/BUILD @@ -16,6 +16,7 @@ go_library( "ioctl.go", "memfd.go", "mmap.go", + "mount.go", "path.go", "pipe.go", "poll.go", diff --git a/pkg/sentry/syscalls/linux/vfs2/mount.go b/pkg/sentry/syscalls/linux/vfs2/mount.go new file mode 100644 index 000000000..adeaa39cc --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/mount.go @@ -0,0 +1,145 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +// Mount implements Linux syscall mount(2). +func Mount(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + sourceAddr := args[0].Pointer() + targetAddr := args[1].Pointer() + typeAddr := args[2].Pointer() + flags := args[3].Uint64() + dataAddr := args[4].Pointer() + + // For null-terminated strings related to mount(2), Linux copies in at most + // a page worth of data. See fs/namespace.c:copy_mount_string(). + fsType, err := t.CopyInString(typeAddr, usermem.PageSize) + if err != nil { + return 0, nil, err + } + source, err := t.CopyInString(sourceAddr, usermem.PageSize) + if err != nil { + return 0, nil, err + } + + targetPath, err := copyInPath(t, targetAddr) + if err != nil { + return 0, nil, err + } + + data := "" + if dataAddr != 0 { + // In Linux, a full page is always copied in regardless of null + // character placement, and the address is passed to each file system. + // Most file systems always treat this data as a string, though, and so + // do all of the ones we implement. + data, err = t.CopyInString(dataAddr, usermem.PageSize) + if err != nil { + return 0, nil, err + } + } + + // Ignore magic value that was required before Linux 2.4. + if flags&linux.MS_MGC_MSK == linux.MS_MGC_VAL { + flags = flags &^ linux.MS_MGC_MSK + } + + // Must have CAP_SYS_ADMIN in the current mount namespace's associated user + // namespace. + creds := t.Credentials() + if !creds.HasCapabilityIn(linux.CAP_SYS_ADMIN, t.MountNamespaceVFS2().Owner) { + return 0, nil, syserror.EPERM + } + + const unsupportedOps = linux.MS_REMOUNT | linux.MS_BIND | + linux.MS_SHARED | linux.MS_PRIVATE | linux.MS_SLAVE | + linux.MS_UNBINDABLE | linux.MS_MOVE + + // Silently allow MS_NOSUID, since we don't implement set-id bits + // anyway. + const unsupportedFlags = linux.MS_NODEV | + linux.MS_NODIRATIME | linux.MS_STRICTATIME + + // Linux just allows passing any flags to mount(2) - it won't fail when + // unknown or unsupported flags are passed. Since we don't implement + // everything, we fail explicitly on flags that are unimplemented. + if flags&(unsupportedOps|unsupportedFlags) != 0 { + return 0, nil, syserror.EINVAL + } + + var opts vfs.MountOptions + if flags&linux.MS_NOATIME == linux.MS_NOATIME { + opts.Flags.NoATime = true + } + if flags&linux.MS_NOEXEC == linux.MS_NOEXEC { + opts.Flags.NoExec = true + } + if flags&linux.MS_RDONLY == linux.MS_RDONLY { + opts.ReadOnly = true + } + opts.GetFilesystemOptions.Data = data + + target, err := getTaskPathOperation(t, linux.AT_FDCWD, targetPath, disallowEmptyPath, nofollowFinalSymlink) + if err != nil { + return 0, nil, err + } + defer target.Release() + + return 0, nil, t.Kernel().VFS().MountAt(t, creds, source, &target.pop, fsType, &opts) +} + +// Umount2 implements Linux syscall umount2(2). +func Umount2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + addr := args[0].Pointer() + flags := args[1].Int() + + // Must have CAP_SYS_ADMIN in the mount namespace's associated user + // namespace. + // + // Currently, this is always the init task's user namespace. + creds := t.Credentials() + if !creds.HasCapabilityIn(linux.CAP_SYS_ADMIN, t.MountNamespaceVFS2().Owner) { + return 0, nil, syserror.EPERM + } + + const unsupported = linux.MNT_FORCE | linux.MNT_EXPIRE + if flags&unsupported != 0 { + return 0, nil, syserror.EINVAL + } + + path, err := copyInPath(t, addr) + if err != nil { + return 0, nil, err + } + tpop, err := getTaskPathOperation(t, linux.AT_FDCWD, path, disallowEmptyPath, nofollowFinalSymlink) + if err != nil { + return 0, nil, err + } + defer tpop.Release() + + opts := vfs.UmountOptions{ + Flags: uint32(flags), + } + + return 0, nil, t.Kernel().VFS().UmountAt(t, creds, &tpop.pop, &opts) +} diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2.go b/pkg/sentry/syscalls/linux/vfs2/vfs2.go index ef8358b8a..7b6e7571a 100644 --- a/pkg/sentry/syscalls/linux/vfs2/vfs2.go +++ b/pkg/sentry/syscalls/linux/vfs2/vfs2.go @@ -90,8 +90,8 @@ func Override() { s.Table[138] = syscalls.Supported("fstatfs", Fstatfs) s.Table[161] = syscalls.Supported("chroot", Chroot) s.Table[162] = syscalls.Supported("sync", Sync) - delete(s.Table, 165) // mount - delete(s.Table, 166) // umount2 + s.Table[165] = syscalls.Supported("mount", Mount) + s.Table[166] = syscalls.Supported("umount2", Umount2) delete(s.Table, 187) // readahead s.Table[188] = syscalls.Supported("setxattr", Setxattr) s.Table[189] = syscalls.Supported("lsetxattr", Lsetxattr) diff --git a/pkg/sentry/vfs/genericfstree/genericfstree.go b/pkg/sentry/vfs/genericfstree/genericfstree.go index 286510195..8882fa84a 100644 --- a/pkg/sentry/vfs/genericfstree/genericfstree.go +++ b/pkg/sentry/vfs/genericfstree/genericfstree.go @@ -43,7 +43,7 @@ type Dentry struct { // IsAncestorDentry returns true if d is an ancestor of d2; that is, d is // either d2's parent or an ancestor of d2's parent. func IsAncestorDentry(d, d2 *Dentry) bool { - for { + for d2 != nil { // Stop at root, where d2.parent == nil. if d2.parent == d { return true } @@ -52,6 +52,7 @@ func IsAncestorDentry(d, d2 *Dentry) bool { } d2 = d2.parent } + return false } // ParentOrSelf returns d.parent. If d.parent is nil, ParentOrSelf returns d. diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index 3adb7c97d..32f901bd8 100644 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go @@ -55,6 +55,10 @@ type Mount struct { // ID is the immutable mount ID. ID uint64 + // Flags contains settings as specified for mount(2), e.g. MS_NOEXEC, except + // for MS_RDONLY which is tracked in "writers". Immutable. + Flags MountFlags + // key is protected by VirtualFilesystem.mountMu and // VirtualFilesystem.mounts.seq, and may be nil. References are held on // key.parent and key.point if they are not nil. @@ -81,10 +85,6 @@ type Mount struct { // umounted is true. umounted is protected by VirtualFilesystem.mountMu. umounted bool - // flags contains settings as specified for mount(2), e.g. MS_NOEXEC, except - // for MS_RDONLY which is tracked in "writers". - flags MountFlags - // The lower 63 bits of writers is the number of calls to // Mount.CheckBeginWrite() that have not yet been paired with a call to // Mount.EndWrite(). The MSB of writers is set if MS_RDONLY is in effect. @@ -95,10 +95,10 @@ type Mount struct { func newMount(vfs *VirtualFilesystem, fs *Filesystem, root *Dentry, mntns *MountNamespace, opts *MountOptions) *Mount { mnt := &Mount{ ID: atomic.AddUint64(&vfs.lastMountID, 1), + Flags: opts.Flags, vfs: vfs, fs: fs, root: root, - flags: opts.Flags, ns: mntns, refs: 1, } @@ -113,13 +113,12 @@ func (mnt *Mount) Options() MountOptions { mnt.vfs.mountMu.Lock() defer mnt.vfs.mountMu.Unlock() return MountOptions{ - Flags: mnt.flags, + Flags: mnt.Flags, ReadOnly: mnt.readOnly(), } } -// A MountNamespace is a collection of Mounts. -// +// A MountNamespace is a collection of Mounts.// // MountNamespaces are reference-counted. Unless otherwise specified, all // MountNamespace methods require that a reference is held. // @@ -127,6 +126,9 @@ func (mnt *Mount) Options() MountOptions { // // +stateify savable type MountNamespace struct { + // Owner is the usernamespace that owns this mount namespace. + Owner *auth.UserNamespace + // root is the MountNamespace's root mount. root is immutable. root *Mount @@ -163,6 +165,7 @@ func (vfs *VirtualFilesystem) NewMountNamespace(ctx context.Context, creds *auth return nil, err } mntns := &MountNamespace{ + Owner: creds.UserNamespace, refs: 1, mountpoints: make(map[*Dentry]uint32), } @@ -279,6 +282,9 @@ func (vfs *VirtualFilesystem) UmountAt(ctx context.Context, creds *auth.Credenti } // MNT_FORCE is currently unimplemented except for the permission check. + // Force unmounting specifically requires CAP_SYS_ADMIN in the root user + // namespace, and not in the owner user namespace for the target mount. See + // fs/namespace.c:SYSCALL_DEFINE2(umount, ...) if opts.Flags&linux.MNT_FORCE != 0 && creds.HasCapabilityIn(linux.CAP_SYS_ADMIN, creds.UserNamespace.Root()) { return syserror.EPERM } @@ -753,7 +759,10 @@ func (vfs *VirtualFilesystem) GenerateProcMounts(ctx context.Context, taskRootDi if mnt.readOnly() { opts = "ro" } - if mnt.flags.NoExec { + if mnt.Flags.NoATime { + opts = ",noatime" + } + if mnt.Flags.NoExec { opts += ",noexec" } @@ -838,11 +847,12 @@ func (vfs *VirtualFilesystem) GenerateProcMountInfo(ctx context.Context, taskRoo if mnt.readOnly() { opts = "ro" } - if mnt.flags.NoExec { + if mnt.Flags.NoATime { + opts = ",noatime" + } + if mnt.Flags.NoExec { opts += ",noexec" } - // TODO(gvisor.dev/issue/1193): Add "noatime" if MS_NOATIME is - // set. fmt.Fprintf(buf, "%s ", opts) // (7) Optional fields: zero or more fields of the form "tag[:value]". diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go index 53d364c5c..f223aeda8 100644 --- a/pkg/sentry/vfs/options.go +++ b/pkg/sentry/vfs/options.go @@ -75,6 +75,10 @@ type MknodOptions struct { type MountFlags struct { // NoExec is equivalent to MS_NOEXEC. NoExec bool + + // NoATime is equivalent to MS_NOATIME and indicates that the + // filesystem should not update access time in-place. + NoATime bool } // MountOptions contains options to VirtualFilesystem.MountAt(). diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 52643a7c5..9acca8bc7 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -405,7 +405,7 @@ func (vfs *VirtualFilesystem) OpenAt(ctx context.Context, creds *auth.Credential vfs.putResolvingPath(rp) if opts.FileExec { - if fd.Mount().flags.NoExec { + if fd.Mount().Flags.NoExec { fd.DecRef() return nil, syserror.EACCES } diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go index 6c84f0794..7ed6801b4 100644 --- a/runsc/boot/vfs.go +++ b/runsc/boot/vfs.go @@ -272,7 +272,7 @@ func (c *containerMounter) getMountNameAndOptionsVFS2(conf *Config, m *mountAndF case "ro": opts.ReadOnly = true case "noatime": - // TODO(gvisor.dev/issue/1193): Implement MS_NOATIME. + opts.Flags.NoATime = true case "noexec": opts.Flags.NoExec = true default: -- cgit v1.2.3 From 32b823fcdb00a7d6eb5ddcd378f19a659edc3da3 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Sun, 7 Jun 2020 13:37:25 -0700 Subject: netstack: parse incoming packet headers up-front Netstack has traditionally parsed headers on-demand as a packet moves up the stack. This is conceptually simple and convenient, but incompatible with iptables, where headers can be inspected and mangled before even a routing decision is made. This changes header parsing to happen early in the incoming packet path, as soon as the NIC gets the packet from a link endpoint. Even if an invalid packet is found (e.g. a TCP header of insufficient length), the packet is passed up the stack for proper stats bookkeeping. PiperOrigin-RevId: 315179302 --- pkg/sentry/socket/netfilter/tcp_matcher.go | 34 +------ pkg/sentry/socket/netfilter/udp_matcher.go | 34 +------ pkg/tcpip/header/ipv4.go | 5 + pkg/tcpip/header/ipv6_extension_headers.go | 7 ++ pkg/tcpip/network/arp/arp.go | 17 +++- pkg/tcpip/network/fragmentation/fragmentation.go | 4 +- pkg/tcpip/network/ip_test.go | 48 +++++---- pkg/tcpip/network/ipv4/icmp.go | 3 + pkg/tcpip/network/ipv4/ipv4.go | 74 +++++++------- pkg/tcpip/network/ipv4/ipv4_test.go | 12 +++ pkg/tcpip/network/ipv6/icmp.go | 7 +- pkg/tcpip/network/ipv6/icmp_test.go | 43 ++++----- pkg/tcpip/network/ipv6/ipv6.go | 118 +++++++++++++++++++---- pkg/tcpip/network/ipv6/ndp_test.go | 33 ++++--- pkg/tcpip/stack/conntrack.go | 46 --------- pkg/tcpip/stack/forwarder_test.go | 78 ++++++++------- pkg/tcpip/stack/iptables_targets.go | 5 - pkg/tcpip/stack/nic.go | 51 ++++++++-- pkg/tcpip/stack/registration.go | 13 +++ pkg/tcpip/stack/stack_test.go | 70 ++++++++------ pkg/tcpip/stack/transport_test.go | 19 +++- pkg/tcpip/transport/icmp/protocol.go | 10 ++ pkg/tcpip/transport/raw/endpoint.go | 5 +- pkg/tcpip/transport/tcp/protocol.go | 21 ++++ pkg/tcpip/transport/tcp/segment.go | 35 +++---- pkg/tcpip/transport/udp/endpoint.go | 6 +- pkg/tcpip/transport/udp/protocol.go | 45 +++++---- pkg/tcpip/transport/udp/udp_test.go | 60 ++++++++++-- 28 files changed, 539 insertions(+), 364 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/tcp_matcher.go b/pkg/sentry/socket/netfilter/tcp_matcher.go index ebabdf334..4f98ee2d5 100644 --- a/pkg/sentry/socket/netfilter/tcp_matcher.go +++ b/pkg/sentry/socket/netfilter/tcp_matcher.go @@ -111,36 +111,10 @@ func (tm *TCPMatcher) Match(hook stack.Hook, pkt *stack.PacketBuffer, interfaceN return false, false } - // Now we need the transport header. However, this may not have been set - // yet. - // TODO(gvisor.dev/issue/170): Parsing the transport header should - // ultimately be moved into the stack.Check codepath as matchers are - // added. - var tcpHeader header.TCP - if pkt.TransportHeader != nil { - tcpHeader = header.TCP(pkt.TransportHeader) - } else { - var length int - if hook == stack.Prerouting { - // The network header hasn't been parsed yet. We have to do it here. - hdr, ok := pkt.Data.PullUp(header.IPv4MinimumSize) - if !ok { - // There's no valid TCP header here, so we hotdrop the - // packet. - return false, true - } - h := header.IPv4(hdr) - pkt.NetworkHeader = hdr - length = int(h.HeaderLength()) - } - // The TCP header hasn't been parsed yet. We have to do it here. - hdr, ok := pkt.Data.PullUp(length + header.TCPMinimumSize) - if !ok { - // There's no valid TCP header here, so we hotdrop the - // packet. - return false, true - } - tcpHeader = header.TCP(hdr[length:]) + tcpHeader := header.TCP(pkt.TransportHeader) + if len(tcpHeader) < header.TCPMinimumSize { + // There's no valid TCP header here, so we drop the packet immediately. + return false, true } // Check whether the source and destination ports are within the diff --git a/pkg/sentry/socket/netfilter/udp_matcher.go b/pkg/sentry/socket/netfilter/udp_matcher.go index 98b9943f8..3f20fc891 100644 --- a/pkg/sentry/socket/netfilter/udp_matcher.go +++ b/pkg/sentry/socket/netfilter/udp_matcher.go @@ -110,36 +110,10 @@ func (um *UDPMatcher) Match(hook stack.Hook, pkt *stack.PacketBuffer, interfaceN return false, false } - // Now we need the transport header. However, this may not have been set - // yet. - // TODO(gvisor.dev/issue/170): Parsing the transport header should - // ultimately be moved into the stack.Check codepath as matchers are - // added. - var udpHeader header.UDP - if pkt.TransportHeader != nil { - udpHeader = header.UDP(pkt.TransportHeader) - } else { - var length int - if hook == stack.Prerouting { - // The network header hasn't been parsed yet. We have to do it here. - hdr, ok := pkt.Data.PullUp(header.IPv4MinimumSize) - if !ok { - // There's no valid UDP header here, so we hotdrop the - // packet. - return false, true - } - h := header.IPv4(hdr) - pkt.NetworkHeader = hdr - length = int(h.HeaderLength()) - } - // The UDP header hasn't been parsed yet. We have to do it here. - hdr, ok := pkt.Data.PullUp(length + header.UDPMinimumSize) - if !ok { - // There's no valid UDP header here, so we hotdrop the - // packet. - return false, true - } - udpHeader = header.UDP(hdr[length:]) + udpHeader := header.UDP(pkt.TransportHeader) + if len(udpHeader) < header.UDPMinimumSize { + // There's no valid UDP header here, so we drop the packet immediately. + return false, true } // Check whether the source and destination ports are within the diff --git a/pkg/tcpip/header/ipv4.go b/pkg/tcpip/header/ipv4.go index 76839eb92..62ac932bb 100644 --- a/pkg/tcpip/header/ipv4.go +++ b/pkg/tcpip/header/ipv4.go @@ -159,6 +159,11 @@ func (b IPv4) Flags() uint8 { return uint8(binary.BigEndian.Uint16(b[flagsFO:]) >> 13) } +// More returns whether the more fragments flag is set. +func (b IPv4) More() bool { + return b.Flags()&IPv4FlagMoreFragments != 0 +} + // TTL returns the "TTL" field of the ipv4 header. func (b IPv4) TTL() uint8 { return b[ttl] diff --git a/pkg/tcpip/header/ipv6_extension_headers.go b/pkg/tcpip/header/ipv6_extension_headers.go index 2c4591409..3499d8399 100644 --- a/pkg/tcpip/header/ipv6_extension_headers.go +++ b/pkg/tcpip/header/ipv6_extension_headers.go @@ -354,6 +354,13 @@ func (b IPv6FragmentExtHdr) ID() uint32 { return binary.BigEndian.Uint32(b[ipv6FragmentExtHdrIdentificationOffset:]) } +// IsAtomic returns whether the fragment header indicates an atomic fragment. An +// atomic fragment is a fragment that contains all the data required to +// reassemble a full packet. +func (b IPv6FragmentExtHdr) IsAtomic() bool { + return !b.More() && b.FragmentOffset() == 0 +} + // IPv6PayloadIterator is an iterator over the contents of an IPv6 payload. // // The IPv6 payload may contain IPv6 extension headers before any upper layer diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index ea1acba83..7f27a840d 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -99,11 +99,7 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBu } func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { - v, ok := pkt.Data.PullUp(header.ARPSize) - if !ok { - return - } - h := header.ARP(v) + h := header.ARP(pkt.NetworkHeader) if !h.IsValid() { return } @@ -209,6 +205,17 @@ func (*protocol) Close() {} // Wait implements stack.TransportProtocol.Wait. func (*protocol) Wait() {} +// Parse implements stack.NetworkProtocol.Parse. +func (*protocol) Parse(pkt *stack.PacketBuffer) (proto tcpip.TransportProtocolNumber, hasTransportHdr bool, ok bool) { + hdr, ok := pkt.Data.PullUp(header.ARPSize) + if !ok { + return 0, false, false + } + pkt.NetworkHeader = hdr + pkt.Data.TrimFront(header.ARPSize) + return 0, false, true +} + var broadcastMAC = tcpip.LinkAddress([]byte{0xff, 0xff, 0xff, 0xff, 0xff, 0xff}) // NewProtocol returns an ARP network protocol. diff --git a/pkg/tcpip/network/fragmentation/fragmentation.go b/pkg/tcpip/network/fragmentation/fragmentation.go index f42abc4bb..2982450f8 100644 --- a/pkg/tcpip/network/fragmentation/fragmentation.go +++ b/pkg/tcpip/network/fragmentation/fragmentation.go @@ -81,8 +81,8 @@ func NewFragmentation(highMemoryLimit, lowMemoryLimit int, reassemblingTimeout t } } -// Process processes an incoming fragment belonging to an ID -// and returns a complete packet when all the packets belonging to that ID have been received. +// Process processes an incoming fragment belonging to an ID and returns a +// complete packet when all the packets belonging to that ID have been received. func (f *Fragmentation) Process(id uint32, first, last uint16, more bool, vv buffer.VectorisedView) (buffer.VectorisedView, bool, error) { f.mu.Lock() r, ok := f.reassemblers[id] diff --git a/pkg/tcpip/network/ip_test.go b/pkg/tcpip/network/ip_test.go index d9b62f2db..7c8fb3e0a 100644 --- a/pkg/tcpip/network/ip_test.go +++ b/pkg/tcpip/network/ip_test.go @@ -293,9 +293,9 @@ func TestIPv4Receive(t *testing.T) { if err != nil { t.Fatalf("could not find route: %v", err) } - ep.HandlePacket(&r, &stack.PacketBuffer{ - Data: view.ToVectorisedView(), - }) + pkt := stack.PacketBuffer{Data: view.ToVectorisedView()} + proto.Parse(&pkt) + ep.HandlePacket(&r, &pkt) if o.dataCalls != 1 { t.Fatalf("Bad number of data calls: got %x, want 1", o.dataCalls) } @@ -382,10 +382,7 @@ func TestIPv4ReceiveControl(t *testing.T) { o.typ = c.expectedTyp o.extra = c.expectedExtra - vv := view[:len(view)-c.trunc].ToVectorisedView() - ep.HandlePacket(&r, &stack.PacketBuffer{ - Data: vv, - }) + ep.HandlePacket(&r, truncatedPacket(view, c.trunc, header.IPv4MinimumSize)) if want := c.expectedCount; o.controlCalls != want { t.Fatalf("Bad number of control calls for %q case: got %v, want %v", c.name, o.controlCalls, want) } @@ -448,17 +445,17 @@ func TestIPv4FragmentationReceive(t *testing.T) { } // Send first segment. - ep.HandlePacket(&r, &stack.PacketBuffer{ - Data: frag1.ToVectorisedView(), - }) + pkt := stack.PacketBuffer{Data: frag1.ToVectorisedView()} + proto.Parse(&pkt) + ep.HandlePacket(&r, &pkt) if o.dataCalls != 0 { t.Fatalf("Bad number of data calls: got %x, want 0", o.dataCalls) } // Send second segment. - ep.HandlePacket(&r, &stack.PacketBuffer{ - Data: frag2.ToVectorisedView(), - }) + pkt = stack.PacketBuffer{Data: frag2.ToVectorisedView()} + proto.Parse(&pkt) + ep.HandlePacket(&r, &pkt) if o.dataCalls != 1 { t.Fatalf("Bad number of data calls: got %x, want 1", o.dataCalls) } @@ -538,9 +535,9 @@ func TestIPv6Receive(t *testing.T) { t.Fatalf("could not find route: %v", err) } - ep.HandlePacket(&r, &stack.PacketBuffer{ - Data: view.ToVectorisedView(), - }) + pkt := stack.PacketBuffer{Data: view.ToVectorisedView()} + proto.Parse(&pkt) + ep.HandlePacket(&r, &pkt) if o.dataCalls != 1 { t.Fatalf("Bad number of data calls: got %x, want 1", o.dataCalls) } @@ -652,12 +649,25 @@ func TestIPv6ReceiveControl(t *testing.T) { // Set ICMPv6 checksum. icmp.SetChecksum(header.ICMPv6Checksum(icmp, outerSrcAddr, localIpv6Addr, buffer.VectorisedView{})) - ep.HandlePacket(&r, &stack.PacketBuffer{ - Data: view[:len(view)-c.trunc].ToVectorisedView(), - }) + ep.HandlePacket(&r, truncatedPacket(view, c.trunc, header.IPv6MinimumSize)) if want := c.expectedCount; o.controlCalls != want { t.Fatalf("Bad number of control calls for %q case: got %v, want %v", c.name, o.controlCalls, want) } }) } } + +// truncatedPacket returns a PacketBuffer based on a truncated view. If view, +// after truncation, is large enough to hold a network header, it makes part of +// view the packet's NetworkHeader and the rest its Data. Otherwise all of view +// becomes Data. +func truncatedPacket(view buffer.View, trunc, netHdrLen int) *stack.PacketBuffer { + v := view[:len(view)-trunc] + if len(v) < netHdrLen { + return &stack.PacketBuffer{Data: v.ToVectorisedView()} + } + return &stack.PacketBuffer{ + NetworkHeader: v[:netHdrLen], + Data: v[netHdrLen:].ToVectorisedView(), + } +} diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go index d1c3ae835..1b67aa066 100644 --- a/pkg/tcpip/network/ipv4/icmp.go +++ b/pkg/tcpip/network/ipv4/icmp.go @@ -59,6 +59,9 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt *stack func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer) { stats := r.Stats() received := stats.ICMP.V4PacketsReceived + // TODO(gvisor.dev/issue/170): ICMP packets don't have their + // TransportHeader fields set. See icmp/protocol.go:protocol.Parse for a + // full explanation. v, ok := pkt.Data.PullUp(header.ICMPv4MinimumSize) if !ok { received.Invalid.Increment() diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index 959f7e007..7e9f16c90 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -21,6 +21,7 @@ package ipv4 import ( + "fmt" "sync/atomic" "gvisor.dev/gvisor/pkg/tcpip" @@ -268,14 +269,14 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw ep, err := e.stack.FindNetworkEndpoint(header.IPv4ProtocolNumber, netHeader.DestinationAddress()) if err == nil { route := r.ReverseRoute(netHeader.SourceAddress(), netHeader.DestinationAddress()) - handleLoopback(&route, pkt, ep) + ep.HandlePacket(&route, pkt) return nil } } if r.Loop&stack.PacketLoop != 0 { loopedR := r.MakeLoopedRoute() - handleLoopback(&loopedR, pkt, e) + e.HandlePacket(&loopedR, pkt) loopedR.Release() } if r.Loop&stack.PacketOut == 0 { @@ -291,17 +292,6 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw return nil } -func handleLoopback(route *stack.Route, pkt *stack.PacketBuffer, ep stack.NetworkEndpoint) { - // The inbound path expects the network header to still be in - // the PacketBuffer's Data field. - views := make([]buffer.View, 1, 1+len(pkt.Data.Views())) - views[0] = pkt.Header.View() - views = append(views, pkt.Data.Views()...) - ep.HandlePacket(route, &stack.PacketBuffer{ - Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views), - }) -} - // WritePackets implements stack.NetworkEndpoint.WritePackets. func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.PacketBufferList, params stack.NetworkHeaderParams) (int, *tcpip.Error) { if r.Loop&stack.PacketLoop != 0 { @@ -339,12 +329,11 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe } if _, ok := natPkts[pkt]; ok { netHeader := header.IPv4(pkt.NetworkHeader) - ep, err := e.stack.FindNetworkEndpoint(header.IPv4ProtocolNumber, netHeader.DestinationAddress()) - if err == nil { + if ep, err := e.stack.FindNetworkEndpoint(header.IPv4ProtocolNumber, netHeader.DestinationAddress()); err == nil { src := netHeader.SourceAddress() dst := netHeader.DestinationAddress() route := r.ReverseRoute(src, dst) - handleLoopback(&route, pkt, ep) + ep.HandlePacket(&route, pkt) n++ continue } @@ -418,22 +407,11 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBu // HandlePacket is called by the link layer when new ipv4 packets arrive for // this endpoint. func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { - headerView, ok := pkt.Data.PullUp(header.IPv4MinimumSize) - if !ok { + h := header.IPv4(pkt.NetworkHeader) + if !h.IsValid(pkt.Data.Size() + len(pkt.NetworkHeader) + len(pkt.TransportHeader)) { r.Stats().IP.MalformedPacketsReceived.Increment() return } - h := header.IPv4(headerView) - if !h.IsValid(pkt.Data.Size()) { - r.Stats().IP.MalformedPacketsReceived.Increment() - return - } - pkt.NetworkHeader = headerView[:h.HeaderLength()] - - hlen := int(h.HeaderLength()) - tlen := int(h.TotalLength()) - pkt.Data.TrimFront(hlen) - pkt.Data.CapLength(tlen - hlen) // iptables filtering. All packets that reach here are intended for // this machine and will not be forwarded. @@ -443,9 +421,8 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { return } - more := (h.Flags() & header.IPv4FlagMoreFragments) != 0 - if more || h.FragmentOffset() != 0 { - if pkt.Data.Size() == 0 { + if h.More() || h.FragmentOffset() != 0 { + if pkt.Data.Size()+len(pkt.TransportHeader) == 0 { // Drop the packet as it's marked as a fragment but has // no payload. r.Stats().IP.MalformedPacketsReceived.Increment() @@ -464,7 +441,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { } var ready bool var err error - pkt.Data, ready, err = e.fragmentation.Process(hash.IPv4FragmentHash(h), h.FragmentOffset(), last, more, pkt.Data) + pkt.Data, ready, err = e.fragmentation.Process(hash.IPv4FragmentHash(h), h.FragmentOffset(), last, h.More(), pkt.Data) if err != nil { r.Stats().IP.MalformedPacketsReceived.Increment() r.Stats().IP.MalformedFragmentsReceived.Increment() @@ -476,7 +453,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { } p := h.TransportProtocol() if p == header.ICMPv4ProtocolNumber { - headerView.CapLength(hlen) + pkt.NetworkHeader.CapLength(int(h.HeaderLength())) e.handleICMP(r, pkt) return } @@ -556,6 +533,35 @@ func (*protocol) Close() {} // Wait implements stack.TransportProtocol.Wait. func (*protocol) Wait() {} +// Parse implements stack.TransportProtocol.Parse. +func (*protocol) Parse(pkt *stack.PacketBuffer) (proto tcpip.TransportProtocolNumber, hasTransportHdr bool, ok bool) { + hdr, ok := pkt.Data.PullUp(header.IPv4MinimumSize) + if !ok { + return 0, false, false + } + ipHdr := header.IPv4(hdr) + + // If there are options, pull those into hdr as well. + if headerLen := int(ipHdr.HeaderLength()); headerLen > header.IPv4MinimumSize && headerLen <= pkt.Data.Size() { + hdr, ok = pkt.Data.PullUp(headerLen) + if !ok { + panic(fmt.Sprintf("There are only %d bytes in pkt.Data, but there should be at least %d", pkt.Data.Size(), headerLen)) + } + ipHdr = header.IPv4(hdr) + } + + // If this is a fragment, don't bother parsing the transport header. + parseTransportHeader := true + if ipHdr.More() || ipHdr.FragmentOffset() != 0 { + parseTransportHeader = false + } + + pkt.NetworkHeader = hdr + pkt.Data.TrimFront(len(hdr)) + pkt.Data.CapLength(int(ipHdr.TotalLength()) - len(hdr)) + return ipHdr.TransportProtocol(), parseTransportHeader, true +} + // calculateMTU calculates the network-layer payload MTU based on the link-layer // payload mtu. func calculateMTU(mtu uint32) uint32 { diff --git a/pkg/tcpip/network/ipv4/ipv4_test.go b/pkg/tcpip/network/ipv4/ipv4_test.go index c208ebd99..11e579c4b 100644 --- a/pkg/tcpip/network/ipv4/ipv4_test.go +++ b/pkg/tcpip/network/ipv4/ipv4_test.go @@ -652,6 +652,18 @@ func TestReceiveFragments(t *testing.T) { }, expectedPayloads: [][]byte{udpPayload1, udpPayload2}, }, + { + name: "Fragment without followup", + fragments: []fragmentData{ + { + id: 1, + flags: header.IPv4FlagMoreFragments, + fragmentOffset: 0, + payload: ipv4Payload1[:64], + }, + }, + expectedPayloads: nil, + }, } for _, test := range tests { diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index b62fb1de6..2ff7eedf4 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -70,17 +70,20 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt *stack e.dispatcher.DeliverTransportControlPacket(e.id.LocalAddress, hdr.DestinationAddress(), ProtocolNumber, p, typ, extra, pkt) } -func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt *stack.PacketBuffer, hasFragmentHeader bool) { +func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragmentHeader bool) { stats := r.Stats().ICMP sent := stats.V6PacketsSent received := stats.V6PacketsReceived + // TODO(gvisor.dev/issue/170): ICMP packets don't have their + // TransportHeader fields set. See icmp/protocol.go:protocol.Parse for a + // full explanation. v, ok := pkt.Data.PullUp(header.ICMPv6HeaderSize) if !ok { received.Invalid.Increment() return } h := header.ICMPv6(v) - iph := header.IPv6(netHeader) + iph := header.IPv6(pkt.NetworkHeader) // Validate ICMPv6 checksum before processing the packet. // diff --git a/pkg/tcpip/network/ipv6/icmp_test.go b/pkg/tcpip/network/ipv6/icmp_test.go index a720f626f..52a01b44e 100644 --- a/pkg/tcpip/network/ipv6/icmp_test.go +++ b/pkg/tcpip/network/ipv6/icmp_test.go @@ -179,36 +179,32 @@ func TestICMPCounts(t *testing.T) { }, } - handleIPv6Payload := func(hdr buffer.Prependable) { - payloadLength := hdr.UsedLength() - ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) + handleIPv6Payload := func(icmp header.ICMPv6) { + ip := header.IPv6(buffer.NewView(header.IPv6MinimumSize)) ip.Encode(&header.IPv6Fields{ - PayloadLength: uint16(payloadLength), + PayloadLength: uint16(len(icmp)), NextHeader: uint8(header.ICMPv6ProtocolNumber), HopLimit: header.NDPHopLimit, SrcAddr: r.LocalAddress, DstAddr: r.RemoteAddress, }) ep.HandlePacket(&r, &stack.PacketBuffer{ - Data: hdr.View().ToVectorisedView(), + NetworkHeader: buffer.View(ip), + Data: buffer.View(icmp).ToVectorisedView(), }) } for _, typ := range types { - extraDataLen := len(typ.extraData) - hdr := buffer.NewPrependable(header.IPv6MinimumSize + typ.size + extraDataLen) - extraData := buffer.View(hdr.Prepend(extraDataLen)) - copy(extraData, typ.extraData) - pkt := header.ICMPv6(hdr.Prepend(typ.size)) - pkt.SetType(typ.typ) - pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, extraData.ToVectorisedView())) - - handleIPv6Payload(hdr) + icmp := header.ICMPv6(buffer.NewView(typ.size + len(typ.extraData))) + copy(icmp[typ.size:], typ.extraData) + icmp.SetType(typ.typ) + icmp.SetChecksum(header.ICMPv6Checksum(icmp[:typ.size], r.LocalAddress, r.RemoteAddress, buffer.View(typ.extraData).ToVectorisedView())) + handleIPv6Payload(icmp) } // Construct an empty ICMP packet so that // Stats().ICMP.ICMPv6ReceivedPacketStats.Invalid is incremented. - handleIPv6Payload(buffer.NewPrependable(header.IPv6MinimumSize)) + handleIPv6Payload(header.ICMPv6(buffer.NewView(header.IPv6MinimumSize))) icmpv6Stats := s.Stats().ICMP.V6PacketsReceived visitStats(reflect.ValueOf(&icmpv6Stats).Elem(), func(name string, s *tcpip.StatCounter) { @@ -546,25 +542,22 @@ func TestICMPChecksumValidationSimple(t *testing.T) { } handleIPv6Payload := func(checksum bool) { - extraDataLen := len(typ.extraData) - hdr := buffer.NewPrependable(header.IPv6MinimumSize + typ.size + extraDataLen) - extraData := buffer.View(hdr.Prepend(extraDataLen)) - copy(extraData, typ.extraData) - pkt := header.ICMPv6(hdr.Prepend(typ.size)) - pkt.SetType(typ.typ) + icmp := header.ICMPv6(buffer.NewView(typ.size + len(typ.extraData))) + copy(icmp[typ.size:], typ.extraData) + icmp.SetType(typ.typ) if checksum { - pkt.SetChecksum(header.ICMPv6Checksum(pkt, lladdr1, lladdr0, extraData.ToVectorisedView())) + icmp.SetChecksum(header.ICMPv6Checksum(icmp, lladdr1, lladdr0, buffer.View{}.ToVectorisedView())) } - ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) + ip := header.IPv6(buffer.NewView(header.IPv6MinimumSize)) ip.Encode(&header.IPv6Fields{ - PayloadLength: uint16(typ.size + extraDataLen), + PayloadLength: uint16(len(icmp)), NextHeader: uint8(header.ICMPv6ProtocolNumber), HopLimit: header.NDPHopLimit, SrcAddr: lladdr1, DstAddr: lladdr0, }) e.InjectInbound(ProtocolNumber, &stack.PacketBuffer{ - Data: hdr.View().ToVectorisedView(), + Data: buffer.NewVectorisedView(len(ip)+len(icmp), []buffer.View{buffer.View(ip), buffer.View(icmp)}), }) } diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index 0d94ad122..95fbcf2d1 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -171,22 +171,20 @@ func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBuff // HandlePacket is called by the link layer when new ipv6 packets arrive for // this endpoint. func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { - headerView, ok := pkt.Data.PullUp(header.IPv6MinimumSize) - if !ok { + h := header.IPv6(pkt.NetworkHeader) + if !h.IsValid(pkt.Data.Size() + len(pkt.NetworkHeader) + len(pkt.TransportHeader)) { r.Stats().IP.MalformedPacketsReceived.Increment() return } - h := header.IPv6(headerView) - if !h.IsValid(pkt.Data.Size()) { - r.Stats().IP.MalformedPacketsReceived.Increment() - return - } - - pkt.NetworkHeader = headerView[:header.IPv6MinimumSize] - pkt.Data.TrimFront(header.IPv6MinimumSize) - pkt.Data.CapLength(int(h.PayloadLength())) - it := header.MakeIPv6PayloadIterator(header.IPv6ExtensionHeaderIdentifier(h.NextHeader()), pkt.Data) + // vv consists of: + // - Any IPv6 header bytes after the first 40 (i.e. extensions). + // - The transport header, if present. + // - Any other payload data. + vv := pkt.NetworkHeader[header.IPv6MinimumSize:].ToVectorisedView() + vv.AppendView(pkt.TransportHeader) + vv.Append(pkt.Data) + it := header.MakeIPv6PayloadIterator(header.IPv6ExtensionHeaderIdentifier(h.NextHeader()), vv) hasFragmentHeader := false for firstHeader := true; ; firstHeader = false { @@ -262,9 +260,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { case header.IPv6FragmentExtHdr: hasFragmentHeader = true - fragmentOffset := extHdr.FragmentOffset() - more := extHdr.More() - if !more && fragmentOffset == 0 { + if extHdr.IsAtomic() { // This fragment extension header indicates that this packet is an // atomic fragment. An atomic fragment is a fragment that contains // all the data required to reassemble a full packet. As per RFC 6946, @@ -277,9 +273,9 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { // Don't consume the iterator if we have the first fragment because we // will use it to validate that the first fragment holds the upper layer // header. - rawPayload := it.AsRawHeader(fragmentOffset != 0 /* consume */) + rawPayload := it.AsRawHeader(extHdr.FragmentOffset() != 0 /* consume */) - if fragmentOffset == 0 { + if extHdr.FragmentOffset() == 0 { // Check that the iterator ends with a raw payload as the first fragment // should include all headers up to and including any upper layer // headers, as per RFC 8200 section 4.5; only upper layer data @@ -332,7 +328,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { } // The packet is a fragment, let's try to reassemble it. - start := fragmentOffset * header.IPv6FragmentExtHdrFragmentOffsetBytesPerUnit + start := extHdr.FragmentOffset() * header.IPv6FragmentExtHdrFragmentOffsetBytesPerUnit last := start + uint16(fragmentPayloadLen) - 1 // Drop the packet if the fragmentOffset is incorrect. i.e the @@ -345,7 +341,9 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { } var ready bool - pkt.Data, ready, err = e.fragmentation.Process(hash.IPv6FragmentHash(h, extHdr.ID()), start, last, more, rawPayload.Buf) + // Note that pkt doesn't have its transport header set after reassembly, + // and won't until DeliverNetworkPacket sets it. + pkt.Data, ready, err = e.fragmentation.Process(hash.IPv6FragmentHash(h, extHdr.ID()), start, last, extHdr.More(), rawPayload.Buf) if err != nil { r.Stats().IP.MalformedPacketsReceived.Increment() r.Stats().IP.MalformedFragmentsReceived.Increment() @@ -394,10 +392,17 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { case header.IPv6RawPayloadHeader: // If the last header in the payload isn't a known IPv6 extension header, // handle it as if it is transport layer data. + + // For unfragmented packets, extHdr still contains the transport header. + // Get rid of it. + // + // For reassembled fragments, pkt.TransportHeader is unset, so this is a + // no-op and pkt.Data begins with the transport header. + extHdr.Buf.TrimFront(len(pkt.TransportHeader)) pkt.Data = extHdr.Buf if p := tcpip.TransportProtocolNumber(extHdr.Identifier); p == header.ICMPv6ProtocolNumber { - e.handleICMP(r, headerView, pkt, hasFragmentHeader) + e.handleICMP(r, pkt, hasFragmentHeader) } else { r.Stats().IP.PacketsDelivered.Increment() // TODO(b/152019344): Send an ICMPv6 Parameter Problem, Code 1 error @@ -505,6 +510,79 @@ func (*protocol) Close() {} // Wait implements stack.TransportProtocol.Wait. func (*protocol) Wait() {} +// Parse implements stack.TransportProtocol.Parse. +func (*protocol) Parse(pkt *stack.PacketBuffer) (proto tcpip.TransportProtocolNumber, hasTransportHdr bool, ok bool) { + hdr, ok := pkt.Data.PullUp(header.IPv6MinimumSize) + if !ok { + return 0, false, false + } + ipHdr := header.IPv6(hdr) + + // dataClone consists of: + // - Any IPv6 header bytes after the first 40 (i.e. extensions). + // - The transport header, if present. + // - Any other payload data. + views := [8]buffer.View{} + dataClone := pkt.Data.Clone(views[:]) + dataClone.TrimFront(header.IPv6MinimumSize) + it := header.MakeIPv6PayloadIterator(header.IPv6ExtensionHeaderIdentifier(ipHdr.NextHeader()), dataClone) + + // Iterate over the IPv6 extensions to find their length. + // + // Parsing occurs again in HandlePacket because we don't track the + // extensions in PacketBuffer. Unfortunately, that means HandlePacket + // has to do the parsing work again. + var nextHdr tcpip.TransportProtocolNumber + foundNext := true + extensionsSize := 0 +traverseExtensions: + for extHdr, done, err := it.Next(); ; extHdr, done, err = it.Next() { + if err != nil { + break + } + // If we exhaust the extension list, the entire packet is the IPv6 header + // and (possibly) extensions. + if done { + extensionsSize = dataClone.Size() + foundNext = false + break + } + + switch extHdr := extHdr.(type) { + case header.IPv6FragmentExtHdr: + // If this is an atomic fragment, we don't have to treat it specially. + if !extHdr.More() && extHdr.FragmentOffset() == 0 { + continue + } + // This is a non-atomic fragment and has to be re-assembled before we can + // examine the payload for a transport header. + foundNext = false + + case header.IPv6RawPayloadHeader: + // We've found the payload after any extensions. + extensionsSize = dataClone.Size() - extHdr.Buf.Size() + nextHdr = tcpip.TransportProtocolNumber(extHdr.Identifier) + break traverseExtensions + + default: + // Any other extension is a no-op, keep looping until we find the payload. + } + } + + // Put the IPv6 header with extensions in pkt.NetworkHeader. + hdr, ok = pkt.Data.PullUp(header.IPv6MinimumSize + extensionsSize) + if !ok { + panic(fmt.Sprintf("pkt.Data should have at least %d bytes, but only has %d.", header.IPv6MinimumSize+extensionsSize, pkt.Data.Size())) + } + ipHdr = header.IPv6(hdr) + + pkt.NetworkHeader = hdr + pkt.Data.TrimFront(len(hdr)) + pkt.Data.CapLength(int(ipHdr.PayloadLength())) + + return nextHdr, foundNext, true +} + // calculateMTU calculates the network-layer payload MTU based on the link-layer // payload mtu. func calculateMTU(mtu uint32) uint32 { diff --git a/pkg/tcpip/network/ipv6/ndp_test.go b/pkg/tcpip/network/ipv6/ndp_test.go index 3c141b91b..64239ce9a 100644 --- a/pkg/tcpip/network/ipv6/ndp_test.go +++ b/pkg/tcpip/network/ipv6/ndp_test.go @@ -551,25 +551,29 @@ func TestNDPValidation(t *testing.T) { return s, ep, r } - handleIPv6Payload := func(hdr buffer.Prependable, hopLimit uint8, atomicFragment bool, ep stack.NetworkEndpoint, r *stack.Route) { + handleIPv6Payload := func(payload buffer.View, hopLimit uint8, atomicFragment bool, ep stack.NetworkEndpoint, r *stack.Route) { nextHdr := uint8(header.ICMPv6ProtocolNumber) + var extensions buffer.View if atomicFragment { - bytes := hdr.Prepend(header.IPv6FragmentExtHdrLength) - bytes[0] = nextHdr + extensions = buffer.NewView(header.IPv6FragmentExtHdrLength) + extensions[0] = nextHdr nextHdr = uint8(header.IPv6FragmentExtHdrIdentifier) } - payloadLength := hdr.UsedLength() - ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) + ip := header.IPv6(buffer.NewView(header.IPv6MinimumSize + len(extensions))) ip.Encode(&header.IPv6Fields{ - PayloadLength: uint16(payloadLength), + PayloadLength: uint16(len(payload) + len(extensions)), NextHeader: nextHdr, HopLimit: hopLimit, SrcAddr: r.LocalAddress, DstAddr: r.RemoteAddress, }) + if n := copy(ip[header.IPv6MinimumSize:], extensions); n != len(extensions) { + t.Fatalf("expected to write %d bytes of extensions, but wrote %d", len(extensions), n) + } ep.HandlePacket(r, &stack.PacketBuffer{ - Data: hdr.View().ToVectorisedView(), + NetworkHeader: buffer.View(ip), + Data: payload.ToVectorisedView(), }) } @@ -676,14 +680,11 @@ func TestNDPValidation(t *testing.T) { invalid := stats.Invalid typStat := typ.statCounter(stats) - extraDataLen := len(typ.extraData) - hdr := buffer.NewPrependable(header.IPv6MinimumSize + typ.size + extraDataLen + header.IPv6FragmentExtHdrLength) - extraData := buffer.View(hdr.Prepend(extraDataLen)) - copy(extraData, typ.extraData) - pkt := header.ICMPv6(hdr.Prepend(typ.size)) - pkt.SetType(typ.typ) - pkt.SetCode(test.code) - pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, extraData.ToVectorisedView())) + icmp := header.ICMPv6(buffer.NewView(typ.size + len(typ.extraData))) + copy(icmp[typ.size:], typ.extraData) + icmp.SetType(typ.typ) + icmp.SetCode(test.code) + icmp.SetChecksum(header.ICMPv6Checksum(icmp[:typ.size], r.LocalAddress, r.RemoteAddress, buffer.View(typ.extraData).ToVectorisedView())) // Rx count of the NDP message should initially be 0. if got := typStat.Value(); got != 0 { @@ -699,7 +700,7 @@ func TestNDPValidation(t *testing.T) { t.FailNow() } - handleIPv6Payload(hdr, test.hopLimit, test.atomicFragment, ep, &r) + handleIPv6Payload(buffer.View(icmp), test.hopLimit, test.atomicFragment, ep, &r) // Rx count of the NDP packet should have increased. if got := typStat.Value(); got != 1 { diff --git a/pkg/tcpip/stack/conntrack.go b/pkg/tcpip/stack/conntrack.go index d4053be08..05bf62788 100644 --- a/pkg/tcpip/stack/conntrack.go +++ b/pkg/tcpip/stack/conntrack.go @@ -20,7 +20,6 @@ import ( "time" "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/hash/jenkins" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/transport/tcpconntrack" @@ -147,44 +146,6 @@ type ConnTrackTable struct { Seed uint32 } -// parseHeaders sets headers in the packet. -func parseHeaders(pkt *PacketBuffer) { - newPkt := pkt.Clone() - - // Set network header. - hdr, ok := newPkt.Data.PullUp(header.IPv4MinimumSize) - if !ok { - return - } - netHeader := header.IPv4(hdr) - newPkt.NetworkHeader = hdr - length := int(netHeader.HeaderLength()) - - // TODO(gvisor.dev/issue/170): Need to support for other - // protocols as well. - // Set transport header. - switch protocol := netHeader.TransportProtocol(); protocol { - case header.UDPProtocolNumber: - if newPkt.TransportHeader == nil { - h, ok := newPkt.Data.PullUp(length + header.UDPMinimumSize) - if !ok { - return - } - newPkt.TransportHeader = buffer.View(header.UDP(h[length:])) - } - case header.TCPProtocolNumber: - if newPkt.TransportHeader == nil { - h, ok := newPkt.Data.PullUp(length + header.TCPMinimumSize) - if !ok { - return - } - newPkt.TransportHeader = buffer.View(header.TCP(h[length:])) - } - } - pkt.NetworkHeader = newPkt.NetworkHeader - pkt.TransportHeader = newPkt.TransportHeader -} - // packetToTuple converts packet to a tuple in original direction. func packetToTuple(pkt *PacketBuffer, hook Hook) (connTrackTuple, *tcpip.Error) { var tuple connTrackTuple @@ -257,13 +218,6 @@ func (ct *ConnTrackTable) getTupleHash(tuple connTrackTuple) uint32 { // TODO(gvisor.dev/issue/170): Only TCP packets are supported. Need to support other // transport protocols. func (ct *ConnTrackTable) connTrackForPacket(pkt *PacketBuffer, hook Hook, createConn bool) (*connTrack, ctDirection) { - if hook == Prerouting { - // Headers will not be set in Prerouting. - // TODO(gvisor.dev/issue/170): Change this after parsing headers - // code is added. - parseHeaders(pkt) - } - var dir ctDirection tuple, err := packetToTuple(pkt, hook) if err != nil { diff --git a/pkg/tcpip/stack/forwarder_test.go b/pkg/tcpip/stack/forwarder_test.go index 63537aaad..a6546cef0 100644 --- a/pkg/tcpip/stack/forwarder_test.go +++ b/pkg/tcpip/stack/forwarder_test.go @@ -33,6 +33,10 @@ const ( // except where another value is explicitly used. It is chosen to match // the MTU of loopback interfaces on linux systems. fwdTestNetDefaultMTU = 65536 + + dstAddrOffset = 0 + srcAddrOffset = 1 + protocolNumberOffset = 2 ) // fwdTestNetworkEndpoint is a network-layer protocol endpoint. @@ -69,15 +73,8 @@ func (f *fwdTestNetworkEndpoint) ID() *NetworkEndpointID { } func (f *fwdTestNetworkEndpoint) HandlePacket(r *Route, pkt *PacketBuffer) { - // Consume the network header. - b, ok := pkt.Data.PullUp(fwdTestNetHeaderLen) - if !ok { - return - } - pkt.Data.TrimFront(fwdTestNetHeaderLen) - // Dispatch the packet to the transport protocol. - f.dispatcher.DeliverTransportPacket(r, tcpip.TransportProtocolNumber(b[2]), pkt) + f.dispatcher.DeliverTransportPacket(r, tcpip.TransportProtocolNumber(pkt.NetworkHeader[protocolNumberOffset]), pkt) } func (f *fwdTestNetworkEndpoint) MaxHeaderLength() uint16 { @@ -100,9 +97,9 @@ func (f *fwdTestNetworkEndpoint) WritePacket(r *Route, gso *GSO, params NetworkH // Add the protocol's header to the packet and send it to the link // endpoint. b := pkt.Header.Prepend(fwdTestNetHeaderLen) - b[0] = r.RemoteAddress[0] - b[1] = f.id.LocalAddress[0] - b[2] = byte(params.Protocol) + b[dstAddrOffset] = r.RemoteAddress[0] + b[srcAddrOffset] = f.id.LocalAddress[0] + b[protocolNumberOffset] = byte(params.Protocol) return f.ep.WritePacket(r, gso, fwdTestNetNumber, pkt) } @@ -140,7 +137,17 @@ func (f *fwdTestNetworkProtocol) DefaultPrefixLen() int { } func (*fwdTestNetworkProtocol) ParseAddresses(v buffer.View) (src, dst tcpip.Address) { - return tcpip.Address(v[1:2]), tcpip.Address(v[0:1]) + return tcpip.Address(v[srcAddrOffset : srcAddrOffset+1]), tcpip.Address(v[dstAddrOffset : dstAddrOffset+1]) +} + +func (*fwdTestNetworkProtocol) Parse(pkt *PacketBuffer) (tcpip.TransportProtocolNumber, bool, bool) { + netHeader, ok := pkt.Data.PullUp(fwdTestNetHeaderLen) + if !ok { + return 0, false, false + } + pkt.NetworkHeader = netHeader + pkt.Data.TrimFront(fwdTestNetHeaderLen) + return tcpip.TransportProtocolNumber(pkt.NetworkHeader[protocolNumberOffset]), true, true } func (f *fwdTestNetworkProtocol) NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache LinkAddressCache, dispatcher TransportDispatcher, ep LinkEndpoint, _ *Stack) (NetworkEndpoint, *tcpip.Error) { @@ -361,7 +368,7 @@ func TestForwardingWithStaticResolver(t *testing.T) { // Inject an inbound packet to address 3 on NIC 1, and see if it is // forwarded to NIC 2. buf := buffer.NewView(30) - buf[0] = 3 + buf[dstAddrOffset] = 3 ep1.InjectInbound(fwdTestNetNumber, &PacketBuffer{ Data: buf.ToVectorisedView(), }) @@ -398,7 +405,7 @@ func TestForwardingWithFakeResolver(t *testing.T) { // Inject an inbound packet to address 3 on NIC 1, and see if it is // forwarded to NIC 2. buf := buffer.NewView(30) - buf[0] = 3 + buf[dstAddrOffset] = 3 ep1.InjectInbound(fwdTestNetNumber, &PacketBuffer{ Data: buf.ToVectorisedView(), }) @@ -429,7 +436,7 @@ func TestForwardingWithNoResolver(t *testing.T) { // inject an inbound packet to address 3 on NIC 1, and see if it is // forwarded to NIC 2. buf := buffer.NewView(30) - buf[0] = 3 + buf[dstAddrOffset] = 3 ep1.InjectInbound(fwdTestNetNumber, &PacketBuffer{ Data: buf.ToVectorisedView(), }) @@ -459,7 +466,7 @@ func TestForwardingWithFakeResolverPartialTimeout(t *testing.T) { // Inject an inbound packet to address 4 on NIC 1. This packet should // not be forwarded. buf := buffer.NewView(30) - buf[0] = 4 + buf[dstAddrOffset] = 4 ep1.InjectInbound(fwdTestNetNumber, &PacketBuffer{ Data: buf.ToVectorisedView(), }) @@ -467,7 +474,7 @@ func TestForwardingWithFakeResolverPartialTimeout(t *testing.T) { // Inject an inbound packet to address 3 on NIC 1, and see if it is // forwarded to NIC 2. buf = buffer.NewView(30) - buf[0] = 3 + buf[dstAddrOffset] = 3 ep1.InjectInbound(fwdTestNetNumber, &PacketBuffer{ Data: buf.ToVectorisedView(), }) @@ -480,9 +487,8 @@ func TestForwardingWithFakeResolverPartialTimeout(t *testing.T) { t.Fatal("packet not forwarded") } - b := p.Pkt.Data.ToView() - if b[0] != 3 { - t.Fatalf("got b[0] = %d, want = 3", b[0]) + if p.Pkt.NetworkHeader[dstAddrOffset] != 3 { + t.Fatalf("got p.Pkt.NetworkHeader[dstAddrOffset] = %d, want = 3", p.Pkt.NetworkHeader[dstAddrOffset]) } // Test that the address resolution happened correctly. @@ -509,7 +515,7 @@ func TestForwardingWithFakeResolverTwoPackets(t *testing.T) { // Inject two inbound packets to address 3 on NIC 1. for i := 0; i < 2; i++ { buf := buffer.NewView(30) - buf[0] = 3 + buf[dstAddrOffset] = 3 ep1.InjectInbound(fwdTestNetNumber, &PacketBuffer{ Data: buf.ToVectorisedView(), }) @@ -524,9 +530,8 @@ func TestForwardingWithFakeResolverTwoPackets(t *testing.T) { t.Fatal("packet not forwarded") } - b := p.Pkt.Data.ToView() - if b[0] != 3 { - t.Fatalf("got b[0] = %d, want = 3", b[0]) + if p.Pkt.NetworkHeader[dstAddrOffset] != 3 { + t.Fatalf("got p.Pkt.NetworkHeader[dstAddrOffset] = %d, want = 3", p.Pkt.NetworkHeader[dstAddrOffset]) } // Test that the address resolution happened correctly. @@ -554,7 +559,7 @@ func TestForwardingWithFakeResolverManyPackets(t *testing.T) { for i := 0; i < maxPendingPacketsPerResolution+5; i++ { // Inject inbound 'maxPendingPacketsPerResolution + 5' packets on NIC 1. buf := buffer.NewView(30) - buf[0] = 3 + buf[dstAddrOffset] = 3 // Set the packet sequence number. binary.BigEndian.PutUint16(buf[fwdTestNetHeaderLen:], uint16(i)) ep1.InjectInbound(fwdTestNetNumber, &PacketBuffer{ @@ -571,14 +576,18 @@ func TestForwardingWithFakeResolverManyPackets(t *testing.T) { t.Fatal("packet not forwarded") } - b := p.Pkt.Data.ToView() - if b[0] != 3 { - t.Fatalf("got b[0] = %d, want = 3", b[0]) + if b := p.Pkt.Header.View(); b[dstAddrOffset] != 3 { + t.Fatalf("got b[dstAddrOffset] = %d, want = 3", b[dstAddrOffset]) + } + seqNumBuf, ok := p.Pkt.Data.PullUp(2) // The sequence number is a uint16 (2 bytes). + if !ok { + t.Fatalf("p.Pkt.Data is too short to hold a sequence number: %d", p.Pkt.Data.Size()) } - // The first 5 packets should not be forwarded so the the - // sequemnce number should start with 5. + + // The first 5 packets should not be forwarded so the sequence number should + // start with 5. want := uint16(i + 5) - if n := binary.BigEndian.Uint16(b[fwdTestNetHeaderLen:]); n != want { + if n := binary.BigEndian.Uint16(seqNumBuf); n != want { t.Fatalf("got the packet #%d, want = #%d", n, want) } @@ -609,7 +618,7 @@ func TestForwardingWithFakeResolverManyResolutions(t *testing.T) { // Each packet has a different destination address (3 to // maxPendingResolutions + 7). buf := buffer.NewView(30) - buf[0] = byte(3 + i) + buf[dstAddrOffset] = byte(3 + i) ep1.InjectInbound(fwdTestNetNumber, &PacketBuffer{ Data: buf.ToVectorisedView(), }) @@ -626,9 +635,8 @@ func TestForwardingWithFakeResolverManyResolutions(t *testing.T) { // The first 5 packets (address 3 to 7) should not be forwarded // because their address resolutions are interrupted. - b := p.Pkt.Data.ToView() - if b[0] < 8 { - t.Fatalf("got b[0] = %d, want b[0] >= 8", b[0]) + if p.Pkt.NetworkHeader[dstAddrOffset] < 8 { + t.Fatalf("got p.Pkt.NetworkHeader[dstAddrOffset] = %d, want p.Pkt.NetworkHeader[dstAddrOffset] >= 8", p.Pkt.NetworkHeader[dstAddrOffset]) } // Test that the address resolution happened correctly. diff --git a/pkg/tcpip/stack/iptables_targets.go b/pkg/tcpip/stack/iptables_targets.go index 36cc6275d..92e31643e 100644 --- a/pkg/tcpip/stack/iptables_targets.go +++ b/pkg/tcpip/stack/iptables_targets.go @@ -98,11 +98,6 @@ func (rt RedirectTarget) Action(pkt *PacketBuffer, ct *ConnTrackTable, hook Hook return RuleAccept, 0 } - // Set network header. - if hook == Prerouting { - parseHeaders(pkt) - } - // Drop the packet if network and transport header are not set. if pkt.NetworkHeader == nil || pkt.TransportHeader == nil { return RuleDrop, 0 diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 6664aea06..d756ae6f5 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -1212,12 +1212,21 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp n.stack.stats.IP.PacketsReceived.Increment() } - netHeader, ok := pkt.Data.PullUp(netProto.MinimumPacketSize()) + // Parse headers. + transProtoNum, hasTransportHdr, ok := netProto.Parse(pkt) if !ok { + // The packet is too small to contain a network header. n.stack.stats.MalformedRcvdPackets.Increment() return } - src, dst := netProto.ParseAddresses(netHeader) + if hasTransportHdr { + // Parse the transport header if present. + if state, ok := n.stack.transportProtocols[transProtoNum]; ok { + state.proto.Parse(pkt) + } + } + + src, dst := netProto.ParseAddresses(pkt.NetworkHeader) if n.stack.handleLocal && !n.isLoopback() && n.getRef(protocol, src) != nil { // The source address is one of our own, so we never should have gotten a @@ -1301,8 +1310,18 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp func (n *NIC) forwardPacket(r *Route, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { // TODO(b/143425874) Decrease the TTL field in forwarded packets. - if linkHeaderLen := int(n.linkEP.MaxHeaderLength()); linkHeaderLen != 0 { - pkt.Header = buffer.NewPrependable(linkHeaderLen) + // TODO(b/151227689): Avoid copying the packet when forwarding. We can do this + // by having lower layers explicity write each header instead of just + // pkt.Header. + + // pkt may have set its NetworkHeader and TransportHeader. If we're + // forwarding, we'll have to copy them into pkt.Header. + pkt.Header = buffer.NewPrependable(int(n.linkEP.MaxHeaderLength()) + len(pkt.NetworkHeader) + len(pkt.TransportHeader)) + if n := copy(pkt.Header.Prepend(len(pkt.TransportHeader)), pkt.TransportHeader); n != len(pkt.TransportHeader) { + panic(fmt.Sprintf("copied %d bytes, expected %d", n, len(pkt.TransportHeader))) + } + if n := copy(pkt.Header.Prepend(len(pkt.NetworkHeader)), pkt.NetworkHeader); n != len(pkt.NetworkHeader) { + panic(fmt.Sprintf("copied %d bytes, expected %d", n, len(pkt.NetworkHeader))) } // WritePacket takes ownership of pkt, calculate numBytes first. @@ -1333,13 +1352,31 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN // validly formed. n.stack.demux.deliverRawPacket(r, protocol, pkt) - transHeader, ok := pkt.Data.PullUp(transProto.MinimumPacketSize()) - if !ok { + // TransportHeader is nil only when pkt is an ICMP packet or was reassembled + // from fragments. + if pkt.TransportHeader == nil { + // TODO(gvisor.dev/issue/170): ICMP packets don't have their + // TransportHeader fields set. See icmp/protocol.go:protocol.Parse for a + // full explanation. + if protocol == header.ICMPv4ProtocolNumber || protocol == header.ICMPv6ProtocolNumber { + transHeader, ok := pkt.Data.PullUp(transProto.MinimumPacketSize()) + if !ok { + n.stack.stats.MalformedRcvdPackets.Increment() + return + } + pkt.TransportHeader = transHeader + } else { + // This is either a bad packet or was re-assembled from fragments. + transProto.Parse(pkt) + } + } + + if len(pkt.TransportHeader) < transProto.MinimumPacketSize() { n.stack.stats.MalformedRcvdPackets.Increment() return } - srcPort, dstPort, err := transProto.ParsePorts(transHeader) + srcPort, dstPort, err := transProto.ParsePorts(pkt.TransportHeader) if err != nil { n.stack.stats.MalformedRcvdPackets.Increment() return diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go index 94f177841..5cbc946b6 100644 --- a/pkg/tcpip/stack/registration.go +++ b/pkg/tcpip/stack/registration.go @@ -168,6 +168,11 @@ type TransportProtocol interface { // Wait waits for any worker goroutines owned by the protocol to stop. Wait() + + // Parse sets pkt.TransportHeader and trims pkt.Data appropriately. It does + // neither and returns false if pkt.Data is too small, i.e. pkt.Data.Size() < + // MinimumPacketSize() + Parse(pkt *PacketBuffer) (ok bool) } // TransportDispatcher contains the methods used by the network stack to deliver @@ -313,6 +318,14 @@ type NetworkProtocol interface { // Wait waits for any worker goroutines owned by the protocol to stop. Wait() + + // Parse sets pkt.NetworkHeader and trims pkt.Data appropriately. It + // returns: + // - The encapsulated protocol, if present. + // - Whether there is an encapsulated transport protocol payload (e.g. ARP + // does not encapsulate anything). + // - Whether pkt.Data was large enough to parse and set pkt.NetworkHeader. + Parse(pkt *PacketBuffer) (proto tcpip.TransportProtocolNumber, hasTransportHdr bool, ok bool) } // NetworkDispatcher contains the methods used by the network stack to deliver diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index f6ddc3ced..ffef9bc2c 100644 --- a/pkg/tcpip/stack/stack_test.go +++ b/pkg/tcpip/stack/stack_test.go @@ -52,6 +52,10 @@ const ( // where another value is explicitly used. It is chosen to match the MTU // of loopback interfaces on linux systems. defaultMTU = 65536 + + dstAddrOffset = 0 + srcAddrOffset = 1 + protocolNumberOffset = 2 ) // fakeNetworkEndpoint is a network-layer protocol endpoint. It counts sent and @@ -94,26 +98,24 @@ func (f *fakeNetworkEndpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuff // Increment the received packet count in the protocol descriptor. f.proto.packetCount[int(f.id.LocalAddress[0])%len(f.proto.packetCount)]++ - // Consume the network header. - b, ok := pkt.Data.PullUp(fakeNetHeaderLen) - if !ok { - return - } - pkt.Data.TrimFront(fakeNetHeaderLen) - // Handle control packets. - if b[2] == uint8(fakeControlProtocol) { + if pkt.NetworkHeader[protocolNumberOffset] == uint8(fakeControlProtocol) { nb, ok := pkt.Data.PullUp(fakeNetHeaderLen) if !ok { return } pkt.Data.TrimFront(fakeNetHeaderLen) - f.dispatcher.DeliverTransportControlPacket(tcpip.Address(nb[1:2]), tcpip.Address(nb[0:1]), fakeNetNumber, tcpip.TransportProtocolNumber(nb[2]), stack.ControlPortUnreachable, 0, pkt) + f.dispatcher.DeliverTransportControlPacket( + tcpip.Address(nb[srcAddrOffset:srcAddrOffset+1]), + tcpip.Address(nb[dstAddrOffset:dstAddrOffset+1]), + fakeNetNumber, + tcpip.TransportProtocolNumber(nb[protocolNumberOffset]), + stack.ControlPortUnreachable, 0, pkt) return } // Dispatch the packet to the transport protocol. - f.dispatcher.DeliverTransportPacket(r, tcpip.TransportProtocolNumber(b[2]), pkt) + f.dispatcher.DeliverTransportPacket(r, tcpip.TransportProtocolNumber(pkt.NetworkHeader[protocolNumberOffset]), pkt) } func (f *fakeNetworkEndpoint) MaxHeaderLength() uint16 { @@ -138,18 +140,13 @@ func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, params // Add the protocol's header to the packet and send it to the link // endpoint. - b := pkt.Header.Prepend(fakeNetHeaderLen) - b[0] = r.RemoteAddress[0] - b[1] = f.id.LocalAddress[0] - b[2] = byte(params.Protocol) + pkt.NetworkHeader = pkt.Header.Prepend(fakeNetHeaderLen) + pkt.NetworkHeader[dstAddrOffset] = r.RemoteAddress[0] + pkt.NetworkHeader[srcAddrOffset] = f.id.LocalAddress[0] + pkt.NetworkHeader[protocolNumberOffset] = byte(params.Protocol) if r.Loop&stack.PacketLoop != 0 { - views := make([]buffer.View, 1, 1+len(pkt.Data.Views())) - views[0] = pkt.Header.View() - views = append(views, pkt.Data.Views()...) - f.HandlePacket(r, &stack.PacketBuffer{ - Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views), - }) + f.HandlePacket(r, pkt) } if r.Loop&stack.PacketOut == 0 { return nil @@ -205,7 +202,7 @@ func (f *fakeNetworkProtocol) PacketCount(intfAddr byte) int { } func (*fakeNetworkProtocol) ParseAddresses(v buffer.View) (src, dst tcpip.Address) { - return tcpip.Address(v[1:2]), tcpip.Address(v[0:1]) + return tcpip.Address(v[srcAddrOffset : srcAddrOffset+1]), tcpip.Address(v[dstAddrOffset : dstAddrOffset+1]) } func (f *fakeNetworkProtocol) NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, ep stack.LinkEndpoint, _ *stack.Stack) (stack.NetworkEndpoint, *tcpip.Error) { @@ -247,6 +244,17 @@ func (*fakeNetworkProtocol) Close() {} // Wait implements TransportProtocol.Wait. func (*fakeNetworkProtocol) Wait() {} +// Parse implements TransportProtocol.Parse. +func (*fakeNetworkProtocol) Parse(pkt *stack.PacketBuffer) (tcpip.TransportProtocolNumber, bool, bool) { + hdr, ok := pkt.Data.PullUp(fakeNetHeaderLen) + if !ok { + return 0, false, false + } + pkt.NetworkHeader = hdr + pkt.Data.TrimFront(fakeNetHeaderLen) + return tcpip.TransportProtocolNumber(hdr[protocolNumberOffset]), true, true +} + func fakeNetFactory() stack.NetworkProtocol { return &fakeNetworkProtocol{} } @@ -292,7 +300,7 @@ func TestNetworkReceive(t *testing.T) { buf := buffer.NewView(30) // Make sure packet with wrong address is not delivered. - buf[0] = 3 + buf[dstAddrOffset] = 3 ep.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) @@ -304,7 +312,7 @@ func TestNetworkReceive(t *testing.T) { } // Make sure packet is delivered to first endpoint. - buf[0] = 1 + buf[dstAddrOffset] = 1 ep.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) @@ -316,7 +324,7 @@ func TestNetworkReceive(t *testing.T) { } // Make sure packet is delivered to second endpoint. - buf[0] = 2 + buf[dstAddrOffset] = 2 ep.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) @@ -982,7 +990,7 @@ func TestAddressRemoval(t *testing.T) { buf := buffer.NewView(30) // Send and receive packets, and verify they are received. - buf[0] = localAddrByte + buf[dstAddrOffset] = localAddrByte testRecv(t, fakeNet, localAddrByte, ep, buf) testSendTo(t, s, remoteAddr, ep, nil) @@ -1032,7 +1040,7 @@ func TestAddressRemovalWithRouteHeld(t *testing.T) { } // Send and receive packets, and verify they are received. - buf[0] = localAddrByte + buf[dstAddrOffset] = localAddrByte testRecv(t, fakeNet, localAddrByte, ep, buf) testSend(t, r, ep, nil) testSendTo(t, s, remoteAddr, ep, nil) @@ -1114,7 +1122,7 @@ func TestEndpointExpiration(t *testing.T) { fakeNet := s.NetworkProtocolInstance(fakeNetNumber).(*fakeNetworkProtocol) buf := buffer.NewView(30) - buf[0] = localAddrByte + buf[dstAddrOffset] = localAddrByte if promiscuous { if err := s.SetPromiscuousMode(nicID, true); err != nil { @@ -1277,7 +1285,7 @@ func TestPromiscuousMode(t *testing.T) { // Write a packet, and check that it doesn't get delivered as we don't // have a matching endpoint. const localAddrByte byte = 0x01 - buf[0] = localAddrByte + buf[dstAddrOffset] = localAddrByte testFailingRecv(t, fakeNet, localAddrByte, ep, buf) // Set promiscuous mode, then check that packet is delivered. @@ -1658,7 +1666,7 @@ func TestAddressRangeAcceptsMatchingPacket(t *testing.T) { buf := buffer.NewView(30) const localAddrByte byte = 0x01 - buf[0] = localAddrByte + buf[dstAddrOffset] = localAddrByte subnet, err := tcpip.NewSubnet(tcpip.Address("\x00"), tcpip.AddressMask("\xF0")) if err != nil { t.Fatal("NewSubnet failed:", err) @@ -1766,7 +1774,7 @@ func TestAddressRangeRejectsNonmatchingPacket(t *testing.T) { buf := buffer.NewView(30) const localAddrByte byte = 0x01 - buf[0] = localAddrByte + buf[dstAddrOffset] = localAddrByte subnet, err := tcpip.NewSubnet(tcpip.Address("\x10"), tcpip.AddressMask("\xF0")) if err != nil { t.Fatal("NewSubnet failed:", err) @@ -2344,7 +2352,7 @@ func TestNICForwarding(t *testing.T) { // Send a packet to dstAddr. buf := buffer.NewView(30) - buf[0] = dstAddr[0] + buf[dstAddrOffset] = dstAddr[0] ep1.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ Data: buf.ToVectorisedView(), }) diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go index cb350ead3..ad61c09d6 100644 --- a/pkg/tcpip/stack/transport_test.go +++ b/pkg/tcpip/stack/transport_test.go @@ -83,7 +83,8 @@ func (f *fakeTransportEndpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions return 0, nil, tcpip.ErrNoRoute } - hdr := buffer.NewPrependable(int(f.route.MaxHeaderLength())) + hdr := buffer.NewPrependable(int(f.route.MaxHeaderLength()) + fakeTransHeaderLen) + hdr.Prepend(fakeTransHeaderLen) v, err := p.FullPayload() if err != nil { return 0, nil, err @@ -324,6 +325,17 @@ func (*fakeTransportProtocol) Close() {} // Wait implements TransportProtocol.Wait. func (*fakeTransportProtocol) Wait() {} +// Parse implements TransportProtocol.Parse. +func (*fakeTransportProtocol) Parse(pkt *stack.PacketBuffer) bool { + hdr, ok := pkt.Data.PullUp(fakeTransHeaderLen) + if !ok { + return false + } + pkt.TransportHeader = hdr + pkt.Data.TrimFront(fakeTransHeaderLen) + return true +} + func fakeTransFactory() stack.TransportProtocol { return &fakeTransportProtocol{} } @@ -642,11 +654,10 @@ func TestTransportForwarding(t *testing.T) { t.Fatal("Response packet not forwarded") } - hdrs := p.Pkt.Data.ToView() - if dst := hdrs[0]; dst != 3 { + if dst := p.Pkt.NetworkHeader[0]; dst != 3 { t.Errorf("Response packet has incorrect destination addresss: got = %d, want = 3", dst) } - if src := hdrs[1]; src != 1 { + if src := p.Pkt.NetworkHeader[1]; src != 1 { t.Errorf("Response packet has incorrect source addresss: got = %d, want = 3", src) } } diff --git a/pkg/tcpip/transport/icmp/protocol.go b/pkg/tcpip/transport/icmp/protocol.go index 2ec6749c7..74ef6541e 100644 --- a/pkg/tcpip/transport/icmp/protocol.go +++ b/pkg/tcpip/transport/icmp/protocol.go @@ -124,6 +124,16 @@ func (*protocol) Close() {} // Wait implements stack.TransportProtocol.Wait. func (*protocol) Wait() {} +// Parse implements stack.TransportProtocol.Parse. +func (*protocol) Parse(pkt *stack.PacketBuffer) bool { + // TODO(gvisor.dev/issue/170): Implement parsing of ICMP. + // + // Right now, the Parse() method is tied to enabled protocols passed into + // stack.New. This works for UDP and TCP, but we handle ICMP traffic even + // when netstack users don't pass ICMP as a supported protocol. + return false +} + // NewProtocol4 returns an ICMPv4 transport protocol. func NewProtocol4() stack.TransportProtocol { return &protocol{ProtocolNumber4} diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index 21c34fac2..a406d815e 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -627,8 +627,9 @@ func (e *endpoint) HandlePacket(route *stack.Route, pkt *stack.PacketBuffer) { }, } - networkHeader := append(buffer.View(nil), pkt.NetworkHeader...) - combinedVV := networkHeader.ToVectorisedView() + headers := append(buffer.View(nil), pkt.NetworkHeader...) + headers = append(headers, pkt.TransportHeader...) + combinedVV := headers.ToVectorisedView() combinedVV.Append(pkt.Data) packet.data = combinedVV packet.timestampNS = e.stack.NowNanoseconds() diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go index c827d0277..73b8a6782 100644 --- a/pkg/tcpip/transport/tcp/protocol.go +++ b/pkg/tcpip/transport/tcp/protocol.go @@ -21,6 +21,7 @@ package tcp import ( + "fmt" "runtime" "strings" "time" @@ -490,6 +491,26 @@ func (p *protocol) SynRcvdCounter() *synRcvdCounter { return &p.synRcvdCount } +// Parse implements stack.TransportProtocol.Parse. +func (*protocol) Parse(pkt *stack.PacketBuffer) bool { + hdr, ok := pkt.Data.PullUp(header.TCPMinimumSize) + if !ok { + return false + } + + // If the header has options, pull those up as well. + if offset := int(header.TCP(hdr).DataOffset()); offset > header.TCPMinimumSize && offset <= pkt.Data.Size() { + hdr, ok = pkt.Data.PullUp(offset) + if !ok { + panic(fmt.Sprintf("There should be at least %d bytes in pkt.Data.", offset)) + } + } + + pkt.TransportHeader = hdr + pkt.Data.TrimFront(len(hdr)) + return true +} + // NewProtocol returns a TCP transport protocol. func NewProtocol() stack.TransportProtocol { return &protocol{ diff --git a/pkg/tcpip/transport/tcp/segment.go b/pkg/tcpip/transport/tcp/segment.go index 0c099e2fd..0280892a8 100644 --- a/pkg/tcpip/transport/tcp/segment.go +++ b/pkg/tcpip/transport/tcp/segment.go @@ -35,6 +35,7 @@ type segment struct { id stack.TransportEndpointID `state:"manual"` route stack.Route `state:"manual"` data buffer.VectorisedView `state:".(buffer.VectorisedView)"` + hdr header.TCP // views is used as buffer for data when its length is large // enough to store a VectorisedView. views [8]buffer.View `state:"nosave"` @@ -67,6 +68,7 @@ func newSegment(r *stack.Route, id stack.TransportEndpointID, pkt *stack.PacketB route: r.Clone(), } s.data = pkt.Data.Clone(s.views[:]) + s.hdr = header.TCP(pkt.TransportHeader) s.rcvdTime = time.Now() return s } @@ -146,12 +148,6 @@ func (s *segment) logicalLen() seqnum.Size { // TCP checksum and stores the checksum and result of checksum verification in // the csum and csumValid fields of the segment. func (s *segment) parse() bool { - h, ok := s.data.PullUp(header.TCPMinimumSize) - if !ok { - return false - } - hdr := header.TCP(h) - // h is the header followed by the payload. We check that the offset to // the data respects the following constraints: // 1. That it's at least the minimum header size; if we don't do this @@ -162,16 +158,12 @@ func (s *segment) parse() bool { // N.B. The segment has already been validated as having at least the // minimum TCP size before reaching here, so it's safe to read the // fields. - offset := int(hdr.DataOffset()) - if offset < header.TCPMinimumSize { - return false - } - hdrWithOpts, ok := s.data.PullUp(offset) - if !ok { + offset := int(s.hdr.DataOffset()) + if offset < header.TCPMinimumSize || offset > len(s.hdr) { return false } - s.options = []byte(hdrWithOpts[header.TCPMinimumSize:]) + s.options = []byte(s.hdr[header.TCPMinimumSize:]) s.parsedOptions = header.ParseTCPOptions(s.options) // Query the link capabilities to decide if checksum validation is @@ -180,22 +172,19 @@ func (s *segment) parse() bool { if s.route.Capabilities()&stack.CapabilityRXChecksumOffload != 0 { s.csumValid = true verifyChecksum = false - s.data.TrimFront(offset) } if verifyChecksum { - hdr = header.TCP(hdrWithOpts) - s.csum = hdr.Checksum() - xsum := s.route.PseudoHeaderChecksum(ProtocolNumber, uint16(s.data.Size())) - xsum = hdr.CalculateChecksum(xsum) - s.data.TrimFront(offset) + s.csum = s.hdr.Checksum() + xsum := s.route.PseudoHeaderChecksum(ProtocolNumber, uint16(s.data.Size()+len(s.hdr))) + xsum = s.hdr.CalculateChecksum(xsum) xsum = header.ChecksumVV(s.data, xsum) s.csumValid = xsum == 0xffff } - s.sequenceNumber = seqnum.Value(hdr.SequenceNumber()) - s.ackNumber = seqnum.Value(hdr.AckNumber()) - s.flags = hdr.Flags() - s.window = seqnum.Size(hdr.WindowSize()) + s.sequenceNumber = seqnum.Value(s.hdr.SequenceNumber()) + s.ackNumber = seqnum.Value(s.hdr.AckNumber()) + s.flags = s.hdr.Flags() + s.window = seqnum.Size(s.hdr.WindowSize()) return true } diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 663af8fec..8c7895713 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -1270,16 +1270,14 @@ func (e *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { // endpoint. func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt *stack.PacketBuffer) { // Get the header then trim it from the view. - hdr, ok := pkt.Data.PullUp(header.UDPMinimumSize) - if !ok || int(header.UDP(hdr).Length()) > pkt.Data.Size() { + hdr := header.UDP(pkt.TransportHeader) + if int(hdr.Length()) > pkt.Data.Size()+header.UDPMinimumSize { // Malformed packet. e.stack.Stats().UDP.MalformedPacketsReceived.Increment() e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() return } - pkt.Data.TrimFront(header.UDPMinimumSize) - e.rcvMu.Lock() e.stack.Stats().UDP.PacketsReceived.Increment() e.stats.PacketsReceived.Increment() diff --git a/pkg/tcpip/transport/udp/protocol.go b/pkg/tcpip/transport/udp/protocol.go index e320c5758..4218e7d03 100644 --- a/pkg/tcpip/transport/udp/protocol.go +++ b/pkg/tcpip/transport/udp/protocol.go @@ -67,14 +67,8 @@ func (*protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) { // HandleUnknownDestinationPacket handles packets targeted at this protocol but // that don't match any existing endpoint. func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, pkt *stack.PacketBuffer) bool { - // Get the header then trim it from the view. - h, ok := pkt.Data.PullUp(header.UDPMinimumSize) - if !ok { - // Malformed packet. - r.Stack().Stats().UDP.MalformedPacketsReceived.Increment() - return true - } - if int(header.UDP(h).Length()) > pkt.Data.Size() { + hdr := header.UDP(pkt.TransportHeader) + if int(hdr.Length()) > pkt.Data.Size()+header.UDPMinimumSize { // Malformed packet. r.Stack().Stats().UDP.MalformedPacketsReceived.Increment() return true @@ -121,7 +115,7 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans } headerLen := int(r.MaxHeaderLength()) + header.ICMPv4MinimumSize available := int(mtu) - headerLen - payloadLen := len(pkt.NetworkHeader) + pkt.Data.Size() + payloadLen := len(pkt.NetworkHeader) + len(pkt.TransportHeader) + pkt.Data.Size() if payloadLen > available { payloadLen = available } @@ -130,9 +124,10 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans // For example, a raw or packet socket may use what UDP // considers an unreachable destination. Thus we deep copy pkt // to prevent multiple ownership and SR errors. - newNetHeader := append(buffer.View(nil), pkt.NetworkHeader...) - payload := newNetHeader.ToVectorisedView() - payload.Append(pkt.Data.ToView().ToVectorisedView()) + newHeader := append(buffer.View(nil), pkt.NetworkHeader...) + newHeader = append(newHeader, pkt.TransportHeader...) + payload := newHeader.ToVectorisedView() + payload.AppendView(pkt.Data.ToView()) payload.CapLength(payloadLen) hdr := buffer.NewPrependable(headerLen) @@ -141,8 +136,9 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans pkt.SetCode(header.ICMPv4PortUnreachable) pkt.SetChecksum(header.ICMPv4Checksum(pkt, payload)) r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, &stack.PacketBuffer{ - Header: hdr, - Data: payload, + Header: hdr, + TransportHeader: buffer.View(pkt), + Data: payload, }) case header.IPv6AddressSize: @@ -164,11 +160,11 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans } headerLen := int(r.MaxHeaderLength()) + header.ICMPv6DstUnreachableMinimumSize available := int(mtu) - headerLen - payloadLen := len(pkt.NetworkHeader) + pkt.Data.Size() + payloadLen := len(pkt.NetworkHeader) + len(pkt.TransportHeader) + pkt.Data.Size() if payloadLen > available { payloadLen = available } - payload := buffer.NewVectorisedView(len(pkt.NetworkHeader), []buffer.View{pkt.NetworkHeader}) + payload := buffer.NewVectorisedView(len(pkt.NetworkHeader)+len(pkt.TransportHeader), []buffer.View{pkt.NetworkHeader, pkt.TransportHeader}) payload.Append(pkt.Data) payload.CapLength(payloadLen) @@ -178,8 +174,9 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans pkt.SetCode(header.ICMPv6PortUnreachable) pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, payload)) r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, &stack.PacketBuffer{ - Header: hdr, - Data: payload, + Header: hdr, + TransportHeader: buffer.View(pkt), + Data: payload, }) } return true @@ -201,6 +198,18 @@ func (*protocol) Close() {} // Wait implements stack.TransportProtocol.Wait. func (*protocol) Wait() {} +// Parse implements stack.TransportProtocol.Parse. +func (*protocol) Parse(pkt *stack.PacketBuffer) bool { + h, ok := pkt.Data.PullUp(header.UDPMinimumSize) + if !ok { + // Packet is too small + return false + } + pkt.TransportHeader = h + pkt.Data.TrimFront(header.UDPMinimumSize) + return true +} + // NewProtocol returns a UDP transport protocol. func NewProtocol() stack.TransportProtocol { return &protocol{} diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index e8ade882b..313a3f117 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -441,9 +441,7 @@ func (c *testContext) injectV6Packet(payload []byte, h *header4Tuple, valid bool // Inject packet. c.linkEP.InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ - Data: buf.ToVectorisedView(), - NetworkHeader: buffer.View(ip), - TransportHeader: buffer.View(u), + Data: buf.ToVectorisedView(), }) } @@ -488,9 +486,7 @@ func (c *testContext) injectV4Packet(payload []byte, h *header4Tuple, valid bool // Inject packet. c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ - Data: buf.ToVectorisedView(), - NetworkHeader: buffer.View(ip), - TransportHeader: buffer.View(u), + Data: buf.ToVectorisedView(), }) } @@ -1720,6 +1716,58 @@ func TestIncrementMalformedPacketsReceived(t *testing.T) { } } +// TestShortHeader verifies that when a packet with a too-short UDP header is +// received, the malformed received global stat gets incremented. +func TestShortHeader(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpoint(ipv6.ProtocolNumber) + // Bind to wildcard. + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + c.t.Fatalf("Bind failed: %s", err) + } + + c.t.Helper() + h := unicastV6.header4Tuple(incoming) + + // Allocate a buffer for an IPv6 and too-short UDP header. + const udpSize = header.UDPMinimumSize - 1 + buf := buffer.NewView(header.IPv6MinimumSize + udpSize) + // Initialize the IP header. + ip := header.IPv6(buf) + ip.Encode(&header.IPv6Fields{ + TrafficClass: testTOS, + PayloadLength: uint16(udpSize), + NextHeader: uint8(udp.ProtocolNumber), + HopLimit: 65, + SrcAddr: h.srcAddr.Addr, + DstAddr: h.dstAddr.Addr, + }) + + // Initialize the UDP header. + udpHdr := header.UDP(buffer.NewView(header.UDPMinimumSize)) + udpHdr.Encode(&header.UDPFields{ + SrcPort: h.srcAddr.Port, + DstPort: h.dstAddr.Port, + Length: header.UDPMinimumSize, + }) + // Calculate the UDP pseudo-header checksum. + xsum := header.PseudoHeaderChecksum(udp.ProtocolNumber, h.srcAddr.Addr, h.dstAddr.Addr, uint16(len(udpHdr))) + udpHdr.SetChecksum(^udpHdr.CalculateChecksum(xsum)) + // Copy all but the last byte of the UDP header into the packet. + copy(buf[header.IPv6MinimumSize:], udpHdr) + + // Inject packet. + c.linkEP.InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) + + if got, want := c.s.Stats().MalformedRcvdPackets.Value(), uint64(1); got != want { + t.Errorf("got c.s.Stats().MalformedRcvdPackets.Value() = %d, want = %d", got, want) + } +} + // TestShutdownRead verifies endpoint read shutdown and error // stats increment on packet receive. func TestShutdownRead(t *testing.T) { -- cgit v1.2.3 From dc029b4b96e92719b2850e9d5556f68837737997 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Mon, 8 Jun 2020 13:27:06 -0700 Subject: Implement VFS2 tmpfs mount options. As in VFS1, the mode, uid, and gid options are supported. Updates #1197 PiperOrigin-RevId: 315340510 --- pkg/sentry/fsimpl/tmpfs/device_file.go | 4 +- pkg/sentry/fsimpl/tmpfs/directory.go | 4 +- pkg/sentry/fsimpl/tmpfs/filesystem.go | 20 ++++--- pkg/sentry/fsimpl/tmpfs/named_pipe.go | 4 +- pkg/sentry/fsimpl/tmpfs/regular_file.go | 4 +- pkg/sentry/fsimpl/tmpfs/socket_file.go | 4 +- pkg/sentry/fsimpl/tmpfs/symlink.go | 4 +- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 94 ++++++++++++++++++++------------- 8 files changed, 81 insertions(+), 57 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/tmpfs/device_file.go b/pkg/sentry/fsimpl/tmpfs/device_file.go index 83bf885ee..ac54d420d 100644 --- a/pkg/sentry/fsimpl/tmpfs/device_file.go +++ b/pkg/sentry/fsimpl/tmpfs/device_file.go @@ -29,7 +29,7 @@ type deviceFile struct { minor uint32 } -func (fs *filesystem) newDeviceFile(creds *auth.Credentials, mode linux.FileMode, kind vfs.DeviceKind, major, minor uint32) *inode { +func (fs *filesystem) newDeviceFile(kuid auth.KUID, kgid auth.KGID, mode linux.FileMode, kind vfs.DeviceKind, major, minor uint32) *inode { file := &deviceFile{ kind: kind, major: major, @@ -43,7 +43,7 @@ func (fs *filesystem) newDeviceFile(creds *auth.Credentials, mode linux.FileMode default: panic(fmt.Sprintf("invalid DeviceKind: %v", kind)) } - file.inode.init(file, fs, creds, mode) + file.inode.init(file, fs, kuid, kgid, mode) file.inode.nlink = 1 // from parent directory return &file.inode } diff --git a/pkg/sentry/fsimpl/tmpfs/directory.go b/pkg/sentry/fsimpl/tmpfs/directory.go index 70387cb9c..913b8a6c5 100644 --- a/pkg/sentry/fsimpl/tmpfs/directory.go +++ b/pkg/sentry/fsimpl/tmpfs/directory.go @@ -48,9 +48,9 @@ type directory struct { childList dentryList } -func (fs *filesystem) newDirectory(creds *auth.Credentials, mode linux.FileMode) *directory { +func (fs *filesystem) newDirectory(kuid auth.KUID, kgid auth.KGID, mode linux.FileMode) *directory { dir := &directory{} - dir.inode.init(dir, fs, creds, linux.S_IFDIR|mode) + dir.inode.init(dir, fs, kuid, kgid, linux.S_IFDIR|mode) dir.inode.nlink = 2 // from "." and parent directory or ".." for root dir.dentry.inode = &dir.inode dir.dentry.vfsd.Init(&dir.dentry) diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 183eb975c..e801680e8 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -256,11 +256,12 @@ func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. // MkdirAt implements vfs.FilesystemImpl.MkdirAt. func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error { return fs.doCreateAt(rp, true /* dir */, func(parentDir *directory, name string) error { + creds := rp.Credentials() if parentDir.inode.nlink == maxLinks { return syserror.EMLINK } parentDir.inode.incLinksLocked() // from child's ".." - childDir := fs.newDirectory(rp.Credentials(), opts.Mode) + childDir := fs.newDirectory(creds.EffectiveKUID, creds.EffectiveKGID, opts.Mode) parentDir.insertChildLocked(&childDir.dentry, name) return nil }) @@ -269,18 +270,19 @@ func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts v // MknodAt implements vfs.FilesystemImpl.MknodAt. func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MknodOptions) error { return fs.doCreateAt(rp, false /* dir */, func(parentDir *directory, name string) error { + creds := rp.Credentials() var childInode *inode switch opts.Mode.FileType() { case 0, linux.S_IFREG: - childInode = fs.newRegularFile(rp.Credentials(), opts.Mode) + childInode = fs.newRegularFile(creds.EffectiveKUID, creds.EffectiveKGID, opts.Mode) case linux.S_IFIFO: - childInode = fs.newNamedPipe(rp.Credentials(), opts.Mode) + childInode = fs.newNamedPipe(creds.EffectiveKUID, creds.EffectiveKGID, opts.Mode) case linux.S_IFBLK: - childInode = fs.newDeviceFile(rp.Credentials(), opts.Mode, vfs.BlockDevice, opts.DevMajor, opts.DevMinor) + childInode = fs.newDeviceFile(creds.EffectiveKUID, creds.EffectiveKGID, opts.Mode, vfs.BlockDevice, opts.DevMajor, opts.DevMinor) case linux.S_IFCHR: - childInode = fs.newDeviceFile(rp.Credentials(), opts.Mode, vfs.CharDevice, opts.DevMajor, opts.DevMinor) + childInode = fs.newDeviceFile(creds.EffectiveKUID, creds.EffectiveKGID, opts.Mode, vfs.CharDevice, opts.DevMajor, opts.DevMinor) case linux.S_IFSOCK: - childInode = fs.newSocketFile(rp.Credentials(), opts.Mode, opts.Endpoint) + childInode = fs.newSocketFile(creds.EffectiveKUID, creds.EffectiveKGID, opts.Mode, opts.Endpoint) default: return syserror.EINVAL } @@ -355,7 +357,8 @@ afterTrailingSymlink: } defer rp.Mount().EndWrite() // Create and open the child. - child := fs.newDentry(fs.newRegularFile(rp.Credentials(), opts.Mode)) + creds := rp.Credentials() + child := fs.newDentry(fs.newRegularFile(creds.EffectiveKUID, creds.EffectiveKGID, opts.Mode)) parentDir.insertChildLocked(child, name) fd, err := child.open(ctx, rp, &opts, true) if err != nil { @@ -676,7 +679,8 @@ func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linu // SymlinkAt implements vfs.FilesystemImpl.SymlinkAt. func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, target string) error { return fs.doCreateAt(rp, false /* dir */, func(parentDir *directory, name string) error { - child := fs.newDentry(fs.newSymlink(rp.Credentials(), target)) + creds := rp.Credentials() + child := fs.newDentry(fs.newSymlink(creds.EffectiveKUID, creds.EffectiveKGID, 0777, target)) parentDir.insertChildLocked(child, name) return nil }) diff --git a/pkg/sentry/fsimpl/tmpfs/named_pipe.go b/pkg/sentry/fsimpl/tmpfs/named_pipe.go index 8d77b3fa8..739350cf0 100644 --- a/pkg/sentry/fsimpl/tmpfs/named_pipe.go +++ b/pkg/sentry/fsimpl/tmpfs/named_pipe.go @@ -30,9 +30,9 @@ type namedPipe struct { // Preconditions: // * fs.mu must be locked. // * rp.Mount().CheckBeginWrite() has been called successfully. -func (fs *filesystem) newNamedPipe(creds *auth.Credentials, mode linux.FileMode) *inode { +func (fs *filesystem) newNamedPipe(kuid auth.KUID, kgid auth.KGID, mode linux.FileMode) *inode { file := &namedPipe{pipe: pipe.NewVFSPipe(true /* isNamed */, pipe.DefaultPipeSize, usermem.PageSize)} - file.inode.init(file, fs, creds, linux.S_IFIFO|mode) + file.inode.init(file, fs, kuid, kgid, linux.S_IFIFO|mode) file.inode.nlink = 1 // Only the parent has a link. return &file.inode } diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index fee174375..4f2ae04d2 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -85,12 +85,12 @@ type regularFile struct { size uint64 } -func (fs *filesystem) newRegularFile(creds *auth.Credentials, mode linux.FileMode) *inode { +func (fs *filesystem) newRegularFile(kuid auth.KUID, kgid auth.KGID, mode linux.FileMode) *inode { file := ®ularFile{ memFile: fs.memFile, seals: linux.F_SEAL_SEAL, } - file.inode.init(file, fs, creds, linux.S_IFREG|mode) + file.inode.init(file, fs, kuid, kgid, linux.S_IFREG|mode) file.inode.nlink = 1 // from parent directory return &file.inode } diff --git a/pkg/sentry/fsimpl/tmpfs/socket_file.go b/pkg/sentry/fsimpl/tmpfs/socket_file.go index 25c2321af..3ed650474 100644 --- a/pkg/sentry/fsimpl/tmpfs/socket_file.go +++ b/pkg/sentry/fsimpl/tmpfs/socket_file.go @@ -26,9 +26,9 @@ type socketFile struct { ep transport.BoundEndpoint } -func (fs *filesystem) newSocketFile(creds *auth.Credentials, mode linux.FileMode, ep transport.BoundEndpoint) *inode { +func (fs *filesystem) newSocketFile(kuid auth.KUID, kgid auth.KGID, mode linux.FileMode, ep transport.BoundEndpoint) *inode { file := &socketFile{ep: ep} - file.inode.init(file, fs, creds, mode) + file.inode.init(file, fs, kuid, kgid, mode) file.inode.nlink = 1 // from parent directory return &file.inode } diff --git a/pkg/sentry/fsimpl/tmpfs/symlink.go b/pkg/sentry/fsimpl/tmpfs/symlink.go index 47e075ed4..b0de5fabe 100644 --- a/pkg/sentry/fsimpl/tmpfs/symlink.go +++ b/pkg/sentry/fsimpl/tmpfs/symlink.go @@ -24,11 +24,11 @@ type symlink struct { target string // immutable } -func (fs *filesystem) newSymlink(creds *auth.Credentials, target string) *inode { +func (fs *filesystem) newSymlink(kuid auth.KUID, kgid auth.KGID, mode linux.FileMode, target string) *inode { link := &symlink{ target: target, } - link.inode.init(link, fs, creds, linux.S_IFLNK|0777) + link.inode.init(link, fs, kuid, kgid, linux.S_IFLNK|mode) link.inode.nlink = 1 // from parent directory return &link.inode } diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 3777ebdf2..7ce1b86c7 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -113,57 +113,78 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt } } - devMinor, err := vfsObj.GetAnonBlockDevMinor() - if err != nil { - return nil, nil, err - } - clock := time.RealtimeClockFromContext(ctx) - fs := filesystem{ - memFile: memFileProvider.MemoryFile(), - clock: clock, - devMinor: devMinor, - } - fs.vfsfs.Init(vfsObj, newFSType, &fs) - mopts := vfs.GenericParseMountOptions(opts.Data) - - defaultMode := linux.FileMode(0777) - if modeStr, ok := mopts["mode"]; ok { + rootMode := linux.FileMode(0777) + if rootFileType == linux.S_IFDIR { + rootMode = 01777 + } + modeStr, ok := mopts["mode"] + if ok { + delete(mopts, "mode") mode, err := strconv.ParseUint(modeStr, 8, 32) if err != nil { - return nil, nil, fmt.Errorf("Mount option \"mode='%v'\" not parsable: %v", modeStr, err) + ctx.Warningf("tmpfs.FilesystemType.GetFilesystem: invalid mode: %q", modeStr) + return nil, nil, syserror.EINVAL } - defaultMode = linux.FileMode(mode) + rootMode = linux.FileMode(mode & 07777) } - - defaultOwnerCreds := creds.Fork() - if uidStr, ok := mopts["uid"]; ok { - uid, err := strconv.ParseInt(uidStr, 10, 32) + rootKUID := creds.EffectiveKUID + uidStr, ok := mopts["uid"] + if ok { + delete(mopts, "uid") + uid, err := strconv.ParseUint(uidStr, 10, 32) if err != nil { - return nil, nil, fmt.Errorf("Mount option \"uid='%v'\" not parsable: %v", uidStr, err) + ctx.Warningf("tmpfs.FilesystemType.GetFilesystem: invalid uid: %q", uidStr) + return nil, nil, syserror.EINVAL } - if err := defaultOwnerCreds.SetUID(auth.UID(uid)); err != nil { - return nil, nil, fmt.Errorf("Error using mount option \"uid='%v'\": %v", uidStr, err) + kuid := creds.UserNamespace.MapToKUID(auth.UID(uid)) + if !kuid.Ok() { + ctx.Warningf("tmpfs.FilesystemType.GetFilesystem: unmapped uid: %d", uid) + return nil, nil, syserror.EINVAL } + rootKUID = kuid } - if gidStr, ok := mopts["gid"]; ok { - gid, err := strconv.ParseInt(gidStr, 10, 32) + rootKGID := creds.EffectiveKGID + gidStr, ok := mopts["gid"] + if ok { + delete(mopts, "gid") + gid, err := strconv.ParseUint(gidStr, 10, 32) if err != nil { - return nil, nil, fmt.Errorf("Mount option \"gid='%v'\" not parsable: %v", gidStr, err) + ctx.Warningf("tmpfs.FilesystemType.GetFilesystem: invalid gid: %q", gidStr) + return nil, nil, syserror.EINVAL } - if err := defaultOwnerCreds.SetGID(auth.GID(gid)); err != nil { - return nil, nil, fmt.Errorf("Error using mount option \"gid='%v'\": %v", gidStr, err) + kgid := creds.UserNamespace.MapToKGID(auth.GID(gid)) + if !kgid.Ok() { + ctx.Warningf("tmpfs.FilesystemType.GetFilesystem: unmapped gid: %d", gid) + return nil, nil, syserror.EINVAL } + rootKGID = kgid + } + if len(mopts) != 0 { + ctx.Warningf("tmpfs.FilesystemType.GetFilesystem: unknown options: %v", mopts) + return nil, nil, syserror.EINVAL + } + + devMinor, err := vfsObj.GetAnonBlockDevMinor() + if err != nil { + return nil, nil, err } + clock := time.RealtimeClockFromContext(ctx) + fs := filesystem{ + memFile: memFileProvider.MemoryFile(), + clock: clock, + devMinor: devMinor, + } + fs.vfsfs.Init(vfsObj, newFSType, &fs) var root *dentry switch rootFileType { case linux.S_IFREG: - root = fs.newDentry(fs.newRegularFile(defaultOwnerCreds, defaultMode)) + root = fs.newDentry(fs.newRegularFile(rootKUID, rootKGID, rootMode)) case linux.S_IFLNK: - root = fs.newDentry(fs.newSymlink(defaultOwnerCreds, tmpfsOpts.RootSymlinkTarget)) + root = fs.newDentry(fs.newSymlink(rootKUID, rootKGID, rootMode, tmpfsOpts.RootSymlinkTarget)) case linux.S_IFDIR: - root = &fs.newDirectory(defaultOwnerCreds, defaultMode).dentry + root = &fs.newDirectory(rootKUID, rootKGID, rootMode).dentry default: fs.vfsfs.DecRef() return nil, nil, fmt.Errorf("invalid tmpfs root file type: %#o", rootFileType) @@ -301,15 +322,15 @@ type inode struct { const maxLinks = math.MaxUint32 -func (i *inode) init(impl interface{}, fs *filesystem, creds *auth.Credentials, mode linux.FileMode) { +func (i *inode) init(impl interface{}, fs *filesystem, kuid auth.KUID, kgid auth.KGID, mode linux.FileMode) { if mode.FileType() == 0 { panic("file type is required in FileMode") } i.fs = fs i.refs = 1 i.mode = uint32(mode) - i.uid = uint32(creds.EffectiveKUID) - i.gid = uint32(creds.EffectiveKGID) + i.uid = uint32(kuid) + i.gid = uint32(kgid) i.ino = atomic.AddUint64(&fs.nextInoMinusOne, 1) // Tmpfs creation sets atime, ctime, and mtime to current time. now := fs.clock.Now().Nanoseconds() @@ -766,8 +787,7 @@ func NewMemfd(mount *vfs.Mount, creds *auth.Credentials, allowSeals bool, name s // Per Linux, mm/shmem.c:__shmem_file_setup(), memfd inodes are set up with // S_IRWXUGO. - mode := linux.FileMode(0777) - inode := fs.newRegularFile(creds, mode) + inode := fs.newRegularFile(creds.EffectiveKUID, creds.EffectiveKGID, 0777) rf := inode.impl.(*regularFile) if allowSeals { rf.seals = 0 -- cgit v1.2.3 From 4e96b94915633cc06bf04bd680f4eeba6a764dc9 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Mon, 8 Jun 2020 23:06:50 -0700 Subject: Combine executable lookup code Run vs. exec, VFS1 vs. VFS2 were executable lookup were slightly different from each other. Combine them all into the same logic. PiperOrigin-RevId: 315426443 --- pkg/sentry/control/BUILD | 4 - pkg/sentry/control/proc.go | 84 +++------- pkg/sentry/fs/user/BUILD | 1 + pkg/sentry/fs/user/path.go | 57 +++---- runsc/boot/fs.go | 6 +- runsc/boot/vfs.go | 6 +- runsc/container/container_test.go | 272 +++++++++++++++++++++++--------- runsc/container/multi_container_test.go | 45 +++--- 8 files changed, 279 insertions(+), 196 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/control/BUILD b/pkg/sentry/control/BUILD index 0c9a62f0d..2c5d14be5 100644 --- a/pkg/sentry/control/BUILD +++ b/pkg/sentry/control/BUILD @@ -16,15 +16,12 @@ go_library( ], deps = [ "//pkg/abi/linux", - "//pkg/context", "//pkg/fd", - "//pkg/fspath", "//pkg/log", "//pkg/sentry/fdimport", "//pkg/sentry/fs", "//pkg/sentry/fs/host", "//pkg/sentry/fs/user", - "//pkg/sentry/fsbridge", "//pkg/sentry/fsimpl/host", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", @@ -36,7 +33,6 @@ go_library( "//pkg/sentry/vfs", "//pkg/sentry/watchdog", "//pkg/sync", - "//pkg/syserror", "//pkg/tcpip/link/sniffer", "//pkg/urpc", "@org_golang_x_sys//unix:go_default_library", diff --git a/pkg/sentry/control/proc.go b/pkg/sentry/control/proc.go index 8767430b7..1bae7cfaf 100644 --- a/pkg/sentry/control/proc.go +++ b/pkg/sentry/control/proc.go @@ -25,13 +25,10 @@ import ( "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/fdimport" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/host" "gvisor.dev/gvisor/pkg/sentry/fs/user" - "gvisor.dev/gvisor/pkg/sentry/fsbridge" hostvfs2 "gvisor.dev/gvisor/pkg/sentry/fsimpl/host" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" @@ -39,7 +36,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/urpc" ) @@ -107,6 +103,9 @@ type ExecArgs struct { // String prints the arguments as a string. func (args ExecArgs) String() string { + if len(args.Argv) == 0 { + return args.Filename + } a := make([]string, len(args.Argv)) copy(a, args.Argv) if args.Filename != "" { @@ -179,37 +178,30 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI } ctx := initArgs.NewContext(proc.Kernel) - if initArgs.Filename == "" { - if kernel.VFS2Enabled { - // Get the full path to the filename from the PATH env variable. - if initArgs.MountNamespaceVFS2 == nil { - // Set initArgs so that 'ctx' returns the namespace. - // - // MountNamespaceVFS2 adds a reference to the namespace, which is - // transferred to the new process. - initArgs.MountNamespaceVFS2 = proc.Kernel.GlobalInit().Leader().MountNamespaceVFS2() - } - file, err := getExecutableFD(ctx, creds, proc.Kernel.VFS(), initArgs.MountNamespaceVFS2, initArgs.Envv, initArgs.WorkingDirectory, initArgs.Argv[0]) - if err != nil { - return nil, 0, nil, nil, fmt.Errorf("error finding executable %q in environment %v: %v", initArgs.Argv[0], initArgs.Envv, err) - } - initArgs.File = fsbridge.NewVFSFile(file) - } else { - if initArgs.MountNamespace == nil { - // Set initArgs so that 'ctx' returns the namespace. - initArgs.MountNamespace = proc.Kernel.GlobalInit().Leader().MountNamespace() + if kernel.VFS2Enabled { + // Get the full path to the filename from the PATH env variable. + if initArgs.MountNamespaceVFS2 == nil { + // Set initArgs so that 'ctx' returns the namespace. + // + // MountNamespaceVFS2 adds a reference to the namespace, which is + // transferred to the new process. + initArgs.MountNamespaceVFS2 = proc.Kernel.GlobalInit().Leader().MountNamespaceVFS2() + } + } else { + if initArgs.MountNamespace == nil { + // Set initArgs so that 'ctx' returns the namespace. + initArgs.MountNamespace = proc.Kernel.GlobalInit().Leader().MountNamespace() - // initArgs must hold a reference on MountNamespace, which will - // be donated to the new process in CreateProcess. - initArgs.MountNamespace.IncRef() - } - f, err := user.ResolveExecutablePath(ctx, creds, initArgs.MountNamespace, initArgs.Envv, initArgs.WorkingDirectory, initArgs.Argv[0]) - if err != nil { - return nil, 0, nil, nil, fmt.Errorf("error finding executable %q in PATH %v: %v", initArgs.Argv[0], initArgs.Envv, err) - } - initArgs.Filename = f + // initArgs must hold a reference on MountNamespace, which will + // be donated to the new process in CreateProcess. + initArgs.MountNamespace.IncRef() } } + resolved, err := user.ResolveExecutablePath(ctx, &initArgs) + if err != nil { + return nil, 0, nil, nil, err + } + initArgs.Filename = resolved fds := make([]int, len(args.FilePayload.Files)) for i, file := range args.FilePayload.Files { @@ -422,31 +414,3 @@ func ttyName(tty *kernel.TTY) string { } return fmt.Sprintf("pts/%d", tty.Index) } - -// getExecutableFD resolves the given executable name and returns a -// vfs.FileDescription for the executable file. -func getExecutableFD(ctx context.Context, creds *auth.Credentials, vfsObj *vfs.VirtualFilesystem, mns *vfs.MountNamespace, envv []string, wd, name string) (*vfs.FileDescription, error) { - path, err := user.ResolveExecutablePathVFS2(ctx, creds, mns, envv, wd, name) - if err != nil { - return nil, err - } - - root := vfs.RootFromContext(ctx) - defer root.DecRef() - - pop := vfs.PathOperation{ - Root: root, - Start: root, // binPath is absolute, Start can be anything. - Path: fspath.Parse(path), - FollowFinalSymlink: true, - } - opts := &vfs.OpenOptions{ - Flags: linux.O_RDONLY, - FileExec: true, - } - f, err := vfsObj.OpenAt(ctx, creds, &pop, opts) - if err == syserror.ENOENT || err == syserror.EACCES { - return nil, nil - } - return f, err -} diff --git a/pkg/sentry/fs/user/BUILD b/pkg/sentry/fs/user/BUILD index bd5dac373..66e949c95 100644 --- a/pkg/sentry/fs/user/BUILD +++ b/pkg/sentry/fs/user/BUILD @@ -15,6 +15,7 @@ go_library( "//pkg/fspath", "//pkg/log", "//pkg/sentry/fs", + "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/vfs", "//pkg/syserror", diff --git a/pkg/sentry/fs/user/path.go b/pkg/sentry/fs/user/path.go index fbd4547a7..397e96045 100644 --- a/pkg/sentry/fs/user/path.go +++ b/pkg/sentry/fs/user/path.go @@ -24,6 +24,7 @@ import ( "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" @@ -31,7 +32,15 @@ import ( // ResolveExecutablePath resolves the given executable name given the working // dir and environment. -func ResolveExecutablePath(ctx context.Context, creds *auth.Credentials, mns *fs.MountNamespace, envv []string, wd, name string) (string, error) { +func ResolveExecutablePath(ctx context.Context, args *kernel.CreateProcessArgs) (string, error) { + name := args.Filename + if len(name) == 0 { + if len(args.Argv) == 0 { + return "", fmt.Errorf("no filename or command provided") + } + name = args.Argv[0] + } + // Absolute paths can be used directly. if path.IsAbs(name) { return name, nil @@ -40,6 +49,7 @@ func ResolveExecutablePath(ctx context.Context, creds *auth.Credentials, mns *fs // Paths with '/' in them should be joined to the working directory, or // to the root if working directory is not set. if strings.IndexByte(name, '/') > 0 { + wd := args.WorkingDirectory if wd == "" { wd = "/" } @@ -49,10 +59,24 @@ func ResolveExecutablePath(ctx context.Context, creds *auth.Credentials, mns *fs return path.Join(wd, name), nil } - // Otherwise, We must lookup the name in the paths, starting from the - // calling context's root directory. - paths := getPath(envv) + // Otherwise, We must lookup the name in the paths. + paths := getPath(args.Envv) + if kernel.VFS2Enabled { + f, err := resolveVFS2(ctx, args.Credentials, args.MountNamespaceVFS2, paths, name) + if err != nil { + return "", fmt.Errorf("error finding executable %q in PATH %v: %v", name, paths, err) + } + return f, nil + } + + f, err := resolve(ctx, args.MountNamespace, paths, name) + if err != nil { + return "", fmt.Errorf("error finding executable %q in PATH %v: %v", name, paths, err) + } + return f, nil +} +func resolve(ctx context.Context, mns *fs.MountNamespace, paths []string, name string) (string, error) { root := fs.RootFromContext(ctx) if root == nil { // Caller has no root. Don't bother traversing anything. @@ -95,30 +119,7 @@ func ResolveExecutablePath(ctx context.Context, creds *auth.Credentials, mns *fs return "", syserror.ENOENT } -// ResolveExecutablePathVFS2 resolves the given executable name given the -// working dir and environment. -func ResolveExecutablePathVFS2(ctx context.Context, creds *auth.Credentials, mns *vfs.MountNamespace, envv []string, wd, name string) (string, error) { - // Absolute paths can be used directly. - if path.IsAbs(name) { - return name, nil - } - - // Paths with '/' in them should be joined to the working directory, or - // to the root if working directory is not set. - if strings.IndexByte(name, '/') > 0 { - if wd == "" { - wd = "/" - } - if !path.IsAbs(wd) { - return "", fmt.Errorf("working directory %q must be absolute", wd) - } - return path.Join(wd, name), nil - } - - // Otherwise, We must lookup the name in the paths, starting from the - // calling context's root directory. - paths := getPath(envv) - +func resolveVFS2(ctx context.Context, creds *auth.Credentials, mns *vfs.MountNamespace, paths []string, name string) (string, error) { root := mns.Root() defer root.DecRef() for _, p := range paths { diff --git a/runsc/boot/fs.go b/runsc/boot/fs.go index b98a1eb50..e83584b82 100644 --- a/runsc/boot/fs.go +++ b/runsc/boot/fs.go @@ -293,11 +293,11 @@ func setupContainerFS(ctx context.Context, conf *Config, mntr *containerMounter, procArgs.MountNamespace = mns // Resolve the executable path from working dir and environment. - f, err := user.ResolveExecutablePath(ctx, procArgs.Credentials, procArgs.MountNamespace, procArgs.Envv, procArgs.WorkingDirectory, procArgs.Argv[0]) + resolved, err := user.ResolveExecutablePath(ctx, procArgs) if err != nil { - return fmt.Errorf("searching for executable %q, cwd: %q, envv: %q: %v", procArgs.Argv[0], procArgs.WorkingDirectory, procArgs.Envv, err) + return err } - procArgs.Filename = f + procArgs.Filename = resolved return nil } diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go index 7ed6801b4..8eeb43e79 100644 --- a/runsc/boot/vfs.go +++ b/runsc/boot/vfs.go @@ -96,11 +96,11 @@ func setupContainerVFS2(ctx context.Context, conf *Config, mntr *containerMounte procArgs.MountNamespaceVFS2 = mns // Resolve the executable path from working dir and environment. - f, err := user.ResolveExecutablePathVFS2(ctx, procArgs.Credentials, procArgs.MountNamespaceVFS2, procArgs.Envv, procArgs.WorkingDirectory, procArgs.Argv[0]) + resolved, err := user.ResolveExecutablePath(ctx, procArgs) if err != nil { - return fmt.Errorf("searching for executable %q, cwd: %q, envv: %q: %v", procArgs.Argv[0], procArgs.WorkingDirectory, procArgs.Envv, err) + return err } - procArgs.Filename = f + procArgs.Filename = resolved return nil } diff --git a/runsc/container/container_test.go b/runsc/container/container_test.go index e7715b6f7..cd76645bd 100644 --- a/runsc/container/container_test.go +++ b/runsc/container/container_test.go @@ -20,6 +20,7 @@ import ( "fmt" "io" "io/ioutil" + "math" "os" "path" "path/filepath" @@ -53,9 +54,8 @@ func waitForProcessList(cont *Container, want []*control.Process) error { err = fmt.Errorf("error getting process data from container: %v", err) return &backoff.PermanentError{Err: err} } - if r, err := procListsEqual(got, want); !r { - return fmt.Errorf("container got process list: %s, want: %s: error: %v", - procListToString(got), procListToString(want), err) + if !procListsEqual(got, want) { + return fmt.Errorf("container got process list: %s, want: %s", procListToString(got), procListToString(want)) } return nil } @@ -92,36 +92,72 @@ func blockUntilWaitable(pid int) error { return err } -// procListsEqual is used to check whether 2 Process lists are equal for all -// implemented fields. -func procListsEqual(got, want []*control.Process) (bool, error) { - if len(got) != len(want) { - return false, nil - } - for i := range got { - pd1 := got[i] - pd2 := want[i] - // Zero out timing dependant fields. - pd1.Time = "" - pd1.STime = "" - pd1.C = 0 - // Ignore TTY field too, since it's not relevant in the cases - // where we use this method. Tests that care about the TTY - // field should check for it themselves. - pd1.TTY = "" - pd1Json, err := control.ProcessListToJSON([]*control.Process{pd1}) - if err != nil { - return false, err +// procListsEqual is used to check whether 2 Process lists are equal. Fields +// set to -1 in wants are ignored. Timestamp and threads fields are always +// ignored. +func procListsEqual(gots, wants []*control.Process) bool { + if len(gots) != len(wants) { + return false + } + for i := range gots { + got := gots[i] + want := wants[i] + + if want.UID != math.MaxUint32 && want.UID != got.UID { + return false } - pd2Json, err := control.ProcessListToJSON([]*control.Process{pd2}) - if err != nil { - return false, err + if want.PID != -1 && want.PID != got.PID { + return false } - if pd1Json != pd2Json { - return false, nil + if want.PPID != -1 && want.PPID != got.PPID { + return false + } + if len(want.TTY) != 0 && want.TTY != got.TTY { + return false + } + if len(want.Cmd) != 0 && want.Cmd != got.Cmd { + return false } } - return true, nil + return true +} + +type processBuilder struct { + process control.Process +} + +func newProcessBuilder() *processBuilder { + return &processBuilder{ + process: control.Process{ + UID: math.MaxUint32, + PID: -1, + PPID: -1, + }, + } +} + +func (p *processBuilder) Cmd(cmd string) *processBuilder { + p.process.Cmd = cmd + return p +} + +func (p *processBuilder) PID(pid kernel.ThreadID) *processBuilder { + p.process.PID = pid + return p +} + +func (p *processBuilder) PPID(ppid kernel.ThreadID) *processBuilder { + p.process.PPID = ppid + return p +} + +func (p *processBuilder) UID(uid auth.KUID) *processBuilder { + p.process.UID = uid + return p +} + +func (p *processBuilder) Process() *control.Process { + return &p.process } func procListToString(pl []*control.Process) string { @@ -323,14 +359,7 @@ func TestLifecycle(t *testing.T) { // expectedPL lists the expected process state of the container. expectedPL := []*control.Process{ - { - UID: 0, - PID: 1, - PPID: 0, - C: 0, - Cmd: "sleep", - Threads: []kernel.ThreadID{1}, - }, + newProcessBuilder().Cmd("sleep").Process(), } // Create the container. args := Args{ @@ -608,10 +637,14 @@ func doAppExitStatus(t *testing.T, vfs2 bool) { // TestExec verifies that a container can exec a new program. func TestExec(t *testing.T) { - for name, conf := range configsWithVFS2(t, overlay) { + for name, conf := range configsWithVFS2(t, all...) { t.Run(name, func(t *testing.T) { - const uid = 343 - spec := testutil.NewSpecWithArgs("sleep", "100") + dir, err := ioutil.TempDir(testutil.TmpDir(), "exec-test") + if err != nil { + t.Fatalf("error creating temporary directory: %v", err) + } + cmd := fmt.Sprintf("ln -s /bin/true %q/symlink && sleep 100", dir) + spec := testutil.NewSpecWithArgs("sh", "-c", cmd) _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) if err != nil { @@ -634,29 +667,127 @@ func TestExec(t *testing.T) { t.Fatalf("error starting container: %v", err) } - // expectedPL lists the expected process state of the container. + // Wait until sleep is running to ensure the symlink was created. expectedPL := []*control.Process{ + newProcessBuilder().Cmd("sh").Process(), + newProcessBuilder().Cmd("sleep").Process(), + } + if err := waitForProcessList(cont, expectedPL); err != nil { + t.Fatalf("waitForProcessList: %v", err) + } + + for _, tc := range []struct { + name string + args control.ExecArgs + }{ + { + name: "complete", + args: control.ExecArgs{ + Filename: "/bin/true", + Argv: []string{"/bin/true"}, + }, + }, + { + name: "filename", + args: control.ExecArgs{ + Filename: "/bin/true", + }, + }, + { + name: "argv", + args: control.ExecArgs{ + Argv: []string{"/bin/true"}, + }, + }, + { + name: "filename resolution", + args: control.ExecArgs{ + Filename: "true", + Envv: []string{"PATH=/bin"}, + }, + }, + { + name: "argv resolution", + args: control.ExecArgs{ + Argv: []string{"true"}, + Envv: []string{"PATH=/bin"}, + }, + }, { - UID: 0, - PID: 1, - PPID: 0, - C: 0, - Cmd: "sleep", - Threads: []kernel.ThreadID{1}, + name: "argv symlink", + args: control.ExecArgs{ + Argv: []string{filepath.Join(dir, "symlink")}, + }, }, { - UID: uid, - PID: 2, - PPID: 0, - C: 0, - Cmd: "sleep", - Threads: []kernel.ThreadID{2}, + name: "working dir", + args: control.ExecArgs{ + Argv: []string{"/bin/sh", "-c", `if [[ "${PWD}" != "/tmp" ]]; then exit 1; fi`}, + WorkingDirectory: "/tmp", + }, }, + { + name: "user", + args: control.ExecArgs{ + Argv: []string{"/bin/sh", "-c", `if [[ "$(id -u)" != "343" ]]; then exit 1; fi`}, + KUID: 343, + }, + }, + { + name: "group", + args: control.ExecArgs{ + Argv: []string{"/bin/sh", "-c", `if [[ "$(id -g)" != "343" ]]; then exit 1; fi`}, + KGID: 343, + }, + }, + { + name: "env", + args: control.ExecArgs{ + Argv: []string{"/bin/sh", "-c", `if [[ "${FOO}" != "123" ]]; then exit 1; fi`}, + Envv: []string{"FOO=123"}, + }, + }, + } { + t.Run(tc.name, func(t *testing.T) { + // t.Parallel() + if ws, err := cont.executeSync(&tc.args); err != nil { + t.Fatalf("executeAsync(%+v): %v", tc.args, err) + } else if ws != 0 { + t.Fatalf("executeAsync(%+v) failed with exit: %v", tc.args, ws) + } + }) } + }) + } +} - // Verify that "sleep 100" is running. - if err := waitForProcessList(cont, expectedPL[:1]); err != nil { - t.Error(err) +// TestExecProcList verifies that a container can exec a new program and it +// shows correcly in the process list. +func TestExecProcList(t *testing.T) { + for name, conf := range configsWithVFS2(t, all...) { + t.Run(name, func(t *testing.T) { + const uid = 343 + spec := testutil.NewSpecWithArgs("sleep", "100") + + _, bundleDir, cleanup, err := testutil.SetupContainer(spec, conf) + if err != nil { + t.Fatalf("error setting up container: %v", err) + } + defer cleanup() + + // Create and start the container. + args := Args{ + ID: testutil.RandomContainerID(), + Spec: spec, + BundleDir: bundleDir, + } + cont, err := New(conf, args) + if err != nil { + t.Fatalf("error creating container: %v", err) + } + defer cont.Destroy() + if err := cont.Start(conf); err != nil { + t.Fatalf("error starting container: %v", err) } execArgs := &control.ExecArgs{ @@ -666,9 +797,8 @@ func TestExec(t *testing.T) { KUID: uid, } - // Verify that "sleep 100" and "sleep 5" are running - // after exec. First, start running exec (whick - // blocks). + // Verify that "sleep 100" and "sleep 5" are running after exec. First, + // start running exec (which blocks). ch := make(chan error) go func() { exitStatus, err := cont.executeSync(execArgs) @@ -681,6 +811,11 @@ func TestExec(t *testing.T) { } }() + // expectedPL lists the expected process state of the container. + expectedPL := []*control.Process{ + newProcessBuilder().PID(1).PPID(0).Cmd("sleep").UID(0).Process(), + newProcessBuilder().PID(2).PPID(0).Cmd("sleep").UID(uid).Process(), + } if err := waitForProcessList(cont, expectedPL); err != nil { t.Fatalf("error waiting for processes: %v", err) } @@ -1242,24 +1377,9 @@ func TestCapabilities(t *testing.T) { // expectedPL lists the expected process state of the container. expectedPL := []*control.Process{ - { - UID: 0, - PID: 1, - PPID: 0, - C: 0, - Cmd: "sleep", - Threads: []kernel.ThreadID{1}, - }, - { - UID: uid, - PID: 2, - PPID: 0, - C: 0, - Cmd: "exe", - Threads: []kernel.ThreadID{2}, - }, + newProcessBuilder().Cmd("sleep").Process(), } - if err := waitForProcessList(cont, expectedPL[:1]); err != nil { + if err := waitForProcessList(cont, expectedPL); err != nil { t.Fatalf("Failed to wait for sleep to start, err: %v", err) } diff --git a/runsc/container/multi_container_test.go b/runsc/container/multi_container_test.go index 207206dd2..c2b54696c 100644 --- a/runsc/container/multi_container_test.go +++ b/runsc/container/multi_container_test.go @@ -149,13 +149,13 @@ func TestMultiContainerSanity(t *testing.T) { // Check via ps that multiple processes are running. expectedPL := []*control.Process{ - {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, + newProcessBuilder().PID(1).PPID(0).Cmd("sleep").Process(), } if err := waitForProcessList(containers[0], expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) } expectedPL = []*control.Process{ - {PID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{2}}, + newProcessBuilder().PID(2).PPID(0).Cmd("sleep").Process(), } if err := waitForProcessList(containers[1], expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) @@ -195,13 +195,13 @@ func TestMultiPIDNS(t *testing.T) { // Check via ps that multiple processes are running. expectedPL := []*control.Process{ - {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, + newProcessBuilder().PID(1).Cmd("sleep").Process(), } if err := waitForProcessList(containers[0], expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) } expectedPL = []*control.Process{ - {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, + newProcessBuilder().PID(1).Cmd("sleep").Process(), } if err := waitForProcessList(containers[1], expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) @@ -257,7 +257,7 @@ func TestMultiPIDNSPath(t *testing.T) { // Check via ps that multiple processes are running. expectedPL := []*control.Process{ - {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, + newProcessBuilder().PID(1).PPID(0).Cmd("sleep").Process(), } if err := waitForProcessList(containers[0], expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) @@ -267,7 +267,7 @@ func TestMultiPIDNSPath(t *testing.T) { } expectedPL = []*control.Process{ - {PID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{2}}, + newProcessBuilder().PID(2).PPID(0).Cmd("sleep").Process(), } if err := waitForProcessList(containers[1], expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) @@ -300,7 +300,7 @@ func TestMultiContainerWait(t *testing.T) { // Check via ps that multiple processes are running. expectedPL := []*control.Process{ - {PID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{2}}, + newProcessBuilder().PID(2).PPID(0).Cmd("sleep").Process(), } if err := waitForProcessList(containers[1], expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) @@ -345,7 +345,7 @@ func TestMultiContainerWait(t *testing.T) { // After Wait returns, ensure that the root container is running and // the child has finished. expectedPL = []*control.Process{ - {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, + newProcessBuilder().Cmd("sleep").Process(), } if err := waitForProcessList(containers[0], expectedPL); err != nil { t.Errorf("failed to wait for %q to start: %v", strings.Join(containers[0].Spec.Process.Args, " "), err) @@ -377,7 +377,7 @@ func TestExecWait(t *testing.T) { // Check via ps that process is running. expectedPL := []*control.Process{ - {PID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{2}}, + newProcessBuilder().Cmd("sleep").Process(), } if err := waitForProcessList(containers[1], expectedPL); err != nil { t.Fatalf("failed to wait for sleep to start: %v", err) @@ -412,7 +412,7 @@ func TestExecWait(t *testing.T) { // Wait for the exec'd process to exit. expectedPL = []*control.Process{ - {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, + newProcessBuilder().PID(1).Cmd("sleep").Process(), } if err := waitForProcessList(containers[0], expectedPL); err != nil { t.Fatalf("failed to wait for second container to stop: %v", err) @@ -498,9 +498,8 @@ func TestMultiContainerSignal(t *testing.T) { // Check via ps that container 1 process is running. expectedPL := []*control.Process{ - {PID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{2}}, + newProcessBuilder().Cmd("sleep").Process(), } - if err := waitForProcessList(containers[1], expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) } @@ -512,7 +511,7 @@ func TestMultiContainerSignal(t *testing.T) { // Make sure process 1 is still running. expectedPL = []*control.Process{ - {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, + newProcessBuilder().PID(1).Cmd("sleep").Process(), } if err := waitForProcessList(containers[0], expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) @@ -626,8 +625,10 @@ func TestMultiContainerDestroy(t *testing.T) { if err != nil { t.Fatalf("error getting process data from sandbox: %v", err) } - expectedPL := []*control.Process{{PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}} - if r, err := procListsEqual(pss, expectedPL); !r { + expectedPL := []*control.Process{ + newProcessBuilder().PID(1).Cmd("sleep").Process(), + } + if !procListsEqual(pss, expectedPL) { t.Errorf("container got process list: %s, want: %s: error: %v", procListToString(pss), procListToString(expectedPL), err) } @@ -664,7 +665,7 @@ func TestMultiContainerProcesses(t *testing.T) { // Check root's container process list doesn't include other containers. expectedPL0 := []*control.Process{ - {PID: 1, Cmd: "sleep", Threads: []kernel.ThreadID{1}}, + newProcessBuilder().PID(1).Cmd("sleep").Process(), } if err := waitForProcessList(containers[0], expectedPL0); err != nil { t.Errorf("failed to wait for process to start: %v", err) @@ -672,8 +673,8 @@ func TestMultiContainerProcesses(t *testing.T) { // Same for the other container. expectedPL1 := []*control.Process{ - {PID: 2, Cmd: "sh", Threads: []kernel.ThreadID{2}}, - {PID: 3, PPID: 2, Cmd: "sleep", Threads: []kernel.ThreadID{3}}, + newProcessBuilder().PID(2).Cmd("sh").Process(), + newProcessBuilder().PID(3).PPID(2).Cmd("sleep").Process(), } if err := waitForProcessList(containers[1], expectedPL1); err != nil { t.Errorf("failed to wait for process to start: %v", err) @@ -687,7 +688,7 @@ func TestMultiContainerProcesses(t *testing.T) { if _, err := containers[1].Execute(args); err != nil { t.Fatalf("error exec'ing: %v", err) } - expectedPL1 = append(expectedPL1, &control.Process{PID: 4, Cmd: "sleep", Threads: []kernel.ThreadID{4}}) + expectedPL1 = append(expectedPL1, newProcessBuilder().PID(4).Cmd("sleep").Process()) if err := waitForProcessList(containers[1], expectedPL1); err != nil { t.Errorf("failed to wait for process to start: %v", err) } @@ -1505,7 +1506,7 @@ func TestMultiContainerGoferKilled(t *testing.T) { // Ensure container is running c := containers[2] expectedPL := []*control.Process{ - {PID: 3, Cmd: "sleep", Threads: []kernel.ThreadID{3}}, + newProcessBuilder().PID(3).Cmd("sleep").Process(), } if err := waitForProcessList(c, expectedPL); err != nil { t.Errorf("failed to wait for sleep to start: %v", err) @@ -1533,7 +1534,7 @@ func TestMultiContainerGoferKilled(t *testing.T) { continue // container[2] has been killed. } pl := []*control.Process{ - {PID: kernel.ThreadID(i + 1), Cmd: "sleep", Threads: []kernel.ThreadID{kernel.ThreadID(i + 1)}}, + newProcessBuilder().PID(kernel.ThreadID(i + 1)).Cmd("sleep").Process(), } if err := waitForProcessList(c, pl); err != nil { t.Errorf("Container %q was affected by another container: %v", c.ID, err) @@ -1553,7 +1554,7 @@ func TestMultiContainerGoferKilled(t *testing.T) { // Wait until sandbox stops. waitForProcessList will loop until sandbox exits // and RPC errors out. impossiblePL := []*control.Process{ - {PID: 100, Cmd: "non-existent-process", Threads: []kernel.ThreadID{100}}, + newProcessBuilder().Cmd("non-existent-process").Process(), } if err := waitForProcessList(c, impossiblePL); err == nil { t.Fatalf("Sandbox was not killed after gofer death") -- cgit v1.2.3 From 5ad1025d2dde4af9cad2e2d46e4b1c5a26f55409 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Mon, 18 May 2020 03:20:18 -0400 Subject: initialize an empty fp state area for sentry on Arm64 We need to initialize an empty fp state area for the sentry. Signed-off-by: Bin Lu --- pkg/sentry/arch/arch_aarch64.go | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/arch/arch_aarch64.go b/pkg/sentry/arch/arch_aarch64.go index 343f81f59..daba8b172 100644 --- a/pkg/sentry/arch/arch_aarch64.go +++ b/pkg/sentry/arch/arch_aarch64.go @@ -17,7 +17,6 @@ package arch import ( - "encoding/binary" "fmt" "io" @@ -49,9 +48,14 @@ const ARMTrapFlag = uint64(1) << 21 type aarch64FPState []byte // initAarch64FPState sets up initial state. +// +// Related code in Linux kernel: fpsimd_flush_thread(). +// FPCR = FPCR_RM_RN (0x0 << 22). +// +// Currently, aarch64FPState is only a space of 0x210 length for fpstate. +// The fp head is useless in sentry/ptrace/kvm. +// func initAarch64FPState(data aarch64FPState) { - binary.LittleEndian.PutUint32(data, fpsimdMagic) - binary.LittleEndian.PutUint32(data[4:], fpsimdContextSize) } func newAarch64FPStateSlice() []byte { -- cgit v1.2.3 From b792cc1ca9c89cafdf8dafc23ef8956eb4f72b46 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Tue, 9 Jun 2020 05:28:09 -0400 Subject: minor change in kvm module for Arm64 Signed-off-by: Bin Lu --- pkg/sentry/platform/kvm/machine_arm64.go | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/kvm/machine_arm64.go b/pkg/sentry/platform/kvm/machine_arm64.go index 750751aa3..f3bf973de 100644 --- a/pkg/sentry/platform/kvm/machine_arm64.go +++ b/pkg/sentry/platform/kvm/machine_arm64.go @@ -106,7 +106,7 @@ func (m *machine) dropPageTables(pt *pagetables.PageTables) { defer m.mu.Unlock() // Clear from all PCIDs. - for _, c := range m.vCPUs { + for _, c := range m.vCPUsByID { if c.PCIDs != nil { c.PCIDs.Drop(pt) } -- cgit v1.2.3 From ecff24930cb2dd5b0910da859d6e712f2f1d32c4 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Tue, 9 Jun 2020 11:14:24 -0700 Subject: Ensure pgalloc.MemoryFile.fileSize is always chunk-aligned. findAvailableLocked() may return a non-aligned FileRange.End after expansion since it may round FileRange.Start down to a hugepage boundary. PiperOrigin-RevId: 315520321 --- pkg/sentry/pgalloc/pgalloc.go | 11 ++++++----- 1 file changed, 6 insertions(+), 5 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/pgalloc/pgalloc.go b/pkg/sentry/pgalloc/pgalloc.go index c8d9facc2..46f19d218 100644 --- a/pkg/sentry/pgalloc/pgalloc.go +++ b/pkg/sentry/pgalloc/pgalloc.go @@ -393,16 +393,17 @@ func (f *MemoryFile) Allocate(length uint64, kind usage.MemoryKind) (platform.Fi return platform.FileRange{}, syserror.ENOMEM } - // Expand the file if needed. Note that findAvailableRange will - // appropriately double the fileSize when required. + // Expand the file if needed. if int64(fr.End) > f.fileSize { - if err := f.file.Truncate(int64(fr.End)); err != nil { + // Round the new file size up to be chunk-aligned. + newFileSize := (int64(fr.End) + chunkMask) &^ chunkMask + if err := f.file.Truncate(newFileSize); err != nil { return platform.FileRange{}, err } - f.fileSize = int64(fr.End) + f.fileSize = newFileSize f.mappingsMu.Lock() oldMappings := f.mappings.Load().([]uintptr) - newMappings := make([]uintptr, f.fileSize>>chunkShift) + newMappings := make([]uintptr, newFileSize>>chunkShift) copy(newMappings, oldMappings) f.mappings.Store(newMappings) f.mappingsMu.Unlock() -- cgit v1.2.3 From 6722b1e56fa63f3409f222a63241705aa3f3ace4 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Tue, 9 Jun 2020 16:34:42 -0700 Subject: Don't WriteOut to readonly mounts When the file closes, it attempts to write dirty cached attributes to the file. This should not be done when the mount is readonly. PiperOrigin-RevId: 315585058 --- pkg/sentry/fs/gofer/inode.go | 2 +- pkg/sentry/fs/host/inode.go | 3 +++ 2 files changed, 4 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/gofer/inode.go b/pkg/sentry/fs/gofer/inode.go index a016c896e..51d7368a1 100644 --- a/pkg/sentry/fs/gofer/inode.go +++ b/pkg/sentry/fs/gofer/inode.go @@ -640,7 +640,7 @@ func (i *inodeOperations) Allocate(ctx context.Context, inode *fs.Inode, offset, // WriteOut implements fs.InodeOperations.WriteOut. func (i *inodeOperations) WriteOut(ctx context.Context, inode *fs.Inode) error { - if !i.session().cachePolicy.cacheUAttrs(inode) { + if inode.MountSource.Flags.ReadOnly || !i.session().cachePolicy.cacheUAttrs(inode) { return nil } diff --git a/pkg/sentry/fs/host/inode.go b/pkg/sentry/fs/host/inode.go index 62f1246aa..fbfba1b58 100644 --- a/pkg/sentry/fs/host/inode.go +++ b/pkg/sentry/fs/host/inode.go @@ -368,6 +368,9 @@ func (i *inodeOperations) Allocate(ctx context.Context, inode *fs.Inode, offset, // WriteOut implements fs.InodeOperations.WriteOut. func (i *inodeOperations) WriteOut(ctx context.Context, inode *fs.Inode) error { + if inode.MountSource.Flags.ReadOnly { + return nil + } // Have we been using host kernel metadata caches? if !inode.MountSource.Flags.ForcePageCache || !canMap(inode) { // Then the metadata is already up to date on the host. -- cgit v1.2.3 From 67565078bbcdd8f797206d996605df8f6658d00a Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Tue, 9 Jun 2020 18:44:57 -0700 Subject: Implement flock(2) in VFS2 LockFD is the generic implementation that can be embedded in FileDescriptionImpl implementations. Unique lock ID is maintained in vfs.FileDescription and is created on demand. Updates #1480 PiperOrigin-RevId: 315604825 --- pkg/sentry/devices/memdev/full.go | 1 + pkg/sentry/devices/memdev/null.go | 1 + pkg/sentry/devices/memdev/random.go | 1 + pkg/sentry/devices/memdev/zero.go | 1 + pkg/sentry/fs/file.go | 2 +- pkg/sentry/fs/lock/lock.go | 41 +++----- pkg/sentry/fs/lock/lock_set_functions.go | 8 +- pkg/sentry/fs/lock/lock_test.go | 111 +++++++++++----------- pkg/sentry/fsimpl/devpts/BUILD | 1 + pkg/sentry/fsimpl/devpts/devpts.go | 5 +- pkg/sentry/fsimpl/devpts/master.go | 5 + pkg/sentry/fsimpl/devpts/slave.go | 5 + pkg/sentry/fsimpl/eventfd/eventfd.go | 1 + pkg/sentry/fsimpl/ext/BUILD | 1 + pkg/sentry/fsimpl/ext/file_description.go | 1 + pkg/sentry/fsimpl/ext/inode.go | 6 ++ pkg/sentry/fsimpl/ext/regular_file.go | 1 + pkg/sentry/fsimpl/gofer/BUILD | 2 + pkg/sentry/fsimpl/gofer/filesystem.go | 9 +- pkg/sentry/fsimpl/gofer/gofer.go | 23 +++++ pkg/sentry/fsimpl/gofer/special_file.go | 4 +- pkg/sentry/fsimpl/host/BUILD | 1 + pkg/sentry/fsimpl/host/host.go | 8 +- pkg/sentry/fsimpl/kernfs/BUILD | 2 + pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go | 10 +- pkg/sentry/fsimpl/kernfs/fd_impl_util.go | 9 +- pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 5 +- pkg/sentry/fsimpl/kernfs/kernfs_test.go | 13 ++- pkg/sentry/fsimpl/pipefs/BUILD | 1 + pkg/sentry/fsimpl/pipefs/pipefs.go | 6 +- pkg/sentry/fsimpl/proc/BUILD | 1 + pkg/sentry/fsimpl/proc/subtasks.go | 5 +- pkg/sentry/fsimpl/proc/task.go | 5 +- pkg/sentry/fsimpl/proc/task_fds.go | 7 +- pkg/sentry/fsimpl/proc/task_files.go | 10 +- pkg/sentry/fsimpl/proc/tasks.go | 5 +- pkg/sentry/fsimpl/signalfd/signalfd.go | 1 + pkg/sentry/fsimpl/sys/BUILD | 1 + pkg/sentry/fsimpl/sys/sys.go | 7 +- pkg/sentry/fsimpl/timerfd/timerfd.go | 1 + pkg/sentry/fsimpl/tmpfs/filesystem.go | 6 +- pkg/sentry/fsimpl/tmpfs/regular_file.go | 23 ----- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 41 +------- pkg/sentry/kernel/fd_table.go | 15 +-- pkg/sentry/kernel/kernel.go | 5 - pkg/sentry/kernel/pipe/BUILD | 1 + pkg/sentry/kernel/pipe/vfs.go | 13 ++- pkg/sentry/socket/hostinet/BUILD | 1 + pkg/sentry/socket/hostinet/socket_vfs2.go | 3 + pkg/sentry/socket/netlink/BUILD | 1 + pkg/sentry/socket/netlink/socket_vfs2.go | 8 +- pkg/sentry/socket/netstack/BUILD | 1 + pkg/sentry/socket/netstack/netstack_vfs2.go | 3 + pkg/sentry/socket/unix/BUILD | 1 + pkg/sentry/socket/unix/unix_vfs2.go | 7 +- pkg/sentry/syscalls/linux/sys_file.go | 39 ++------ pkg/sentry/syscalls/linux/vfs2/BUILD | 2 + pkg/sentry/syscalls/linux/vfs2/lock.go | 64 +++++++++++++ pkg/sentry/syscalls/linux/vfs2/vfs2.go | 2 +- pkg/sentry/vfs/BUILD | 1 + pkg/sentry/vfs/epoll.go | 1 + pkg/sentry/vfs/file_description.go | 25 ++++- pkg/sentry/vfs/file_description_impl_util.go | 80 ++++++++++++---- pkg/sentry/vfs/file_description_impl_util_test.go | 1 + pkg/sentry/vfs/inotify.go | 1 + test/syscalls/linux/BUILD | 3 + test/syscalls/linux/flock.cc | 75 ++++++++++++--- 67 files changed, 470 insertions(+), 281 deletions(-) create mode 100644 pkg/sentry/syscalls/linux/vfs2/lock.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/devices/memdev/full.go b/pkg/sentry/devices/memdev/full.go index c7e197691..af66fe4dc 100644 --- a/pkg/sentry/devices/memdev/full.go +++ b/pkg/sentry/devices/memdev/full.go @@ -42,6 +42,7 @@ type fullFD struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl vfs.DentryMetadataFileDescriptionImpl + vfs.NoLockFD } // Release implements vfs.FileDescriptionImpl.Release. diff --git a/pkg/sentry/devices/memdev/null.go b/pkg/sentry/devices/memdev/null.go index 33d060d02..92d3d71be 100644 --- a/pkg/sentry/devices/memdev/null.go +++ b/pkg/sentry/devices/memdev/null.go @@ -43,6 +43,7 @@ type nullFD struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl vfs.DentryMetadataFileDescriptionImpl + vfs.NoLockFD } // Release implements vfs.FileDescriptionImpl.Release. diff --git a/pkg/sentry/devices/memdev/random.go b/pkg/sentry/devices/memdev/random.go index acfa23149..6b81da5ef 100644 --- a/pkg/sentry/devices/memdev/random.go +++ b/pkg/sentry/devices/memdev/random.go @@ -48,6 +48,7 @@ type randomFD struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl vfs.DentryMetadataFileDescriptionImpl + vfs.NoLockFD // off is the "file offset". off is accessed using atomic memory // operations. diff --git a/pkg/sentry/devices/memdev/zero.go b/pkg/sentry/devices/memdev/zero.go index 3b1372b9e..c6f15054d 100644 --- a/pkg/sentry/devices/memdev/zero.go +++ b/pkg/sentry/devices/memdev/zero.go @@ -44,6 +44,7 @@ type zeroFD struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl vfs.DentryMetadataFileDescriptionImpl + vfs.NoLockFD } // Release implements vfs.FileDescriptionImpl.Release. diff --git a/pkg/sentry/fs/file.go b/pkg/sentry/fs/file.go index 2a278fbe3..ca41520b4 100644 --- a/pkg/sentry/fs/file.go +++ b/pkg/sentry/fs/file.go @@ -146,7 +146,7 @@ func (f *File) DecRef() { f.DecRefWithDestructor(func() { // Drop BSD style locks. lockRng := lock.LockRange{Start: 0, End: lock.LockEOF} - f.Dirent.Inode.LockCtx.BSD.UnlockRegion(lock.UniqueID(f.UniqueID), lockRng) + f.Dirent.Inode.LockCtx.BSD.UnlockRegion(f, lockRng) // Release resources held by the FileOperations. f.FileOperations.Release() diff --git a/pkg/sentry/fs/lock/lock.go b/pkg/sentry/fs/lock/lock.go index 926538d90..8a5d9c7eb 100644 --- a/pkg/sentry/fs/lock/lock.go +++ b/pkg/sentry/fs/lock/lock.go @@ -62,7 +62,7 @@ import ( type LockType int // UniqueID is a unique identifier of the holder of a regional file lock. -type UniqueID uint64 +type UniqueID interface{} const ( // ReadLock describes a POSIX regional file lock to be taken @@ -98,12 +98,7 @@ type Lock struct { // If len(Readers) > 0 then HasWriter must be false. Readers map[UniqueID]bool - // HasWriter indicates that this is a write lock held by a single - // UniqueID. - HasWriter bool - - // Writer is only valid if HasWriter is true. It identifies a - // single write lock holder. + // Writer holds the writer unique ID. It's nil if there are no writers. Writer UniqueID } @@ -186,7 +181,6 @@ func makeLock(uid UniqueID, t LockType) Lock { case ReadLock: value.Readers[uid] = true case WriteLock: - value.HasWriter = true value.Writer = uid default: panic(fmt.Sprintf("makeLock: invalid lock type %d", t)) @@ -196,10 +190,7 @@ func makeLock(uid UniqueID, t LockType) Lock { // isHeld returns true if uid is a holder of Lock. func (l Lock) isHeld(uid UniqueID) bool { - if l.HasWriter && l.Writer == uid { - return true - } - return l.Readers[uid] + return l.Writer == uid || l.Readers[uid] } // lock sets uid as a holder of a typed lock on Lock. @@ -214,20 +205,20 @@ func (l *Lock) lock(uid UniqueID, t LockType) { } // We cannot downgrade a write lock to a read lock unless the // uid is the same. - if l.HasWriter { + if l.Writer != nil { if l.Writer != uid { panic(fmt.Sprintf("lock: cannot downgrade write lock to read lock for uid %d, writer is %d", uid, l.Writer)) } // Ensure that there is only one reader if upgrading. l.Readers = make(map[UniqueID]bool) // Ensure that there is no longer a writer. - l.HasWriter = false + l.Writer = nil } l.Readers[uid] = true return case WriteLock: // If we are already the writer, then this is a no-op. - if l.HasWriter && l.Writer == uid { + if l.Writer == uid { return } // We can only upgrade a read lock to a write lock if there @@ -243,7 +234,6 @@ func (l *Lock) lock(uid UniqueID, t LockType) { } // Ensure that there is only a writer. l.Readers = make(map[UniqueID]bool) - l.HasWriter = true l.Writer = uid default: panic(fmt.Sprintf("lock: invalid lock type %d", t)) @@ -277,9 +267,8 @@ func (l LockSet) canLock(uid UniqueID, t LockType, r LockRange) bool { switch t { case ReadLock: return l.lockable(r, func(value Lock) bool { - // If there is no writer, there's no problem adding - // another reader. - if !value.HasWriter { + // If there is no writer, there's no problem adding another reader. + if value.Writer == nil { return true } // If there is a writer, then it must be the same uid @@ -289,10 +278,9 @@ func (l LockSet) canLock(uid UniqueID, t LockType, r LockRange) bool { case WriteLock: return l.lockable(r, func(value Lock) bool { // If there are only readers. - if !value.HasWriter { - // Then this uid can only take a write lock if - // this is a private upgrade, meaning that the - // only reader is uid. + if value.Writer == nil { + // Then this uid can only take a write lock if this is a private + // upgrade, meaning that the only reader is uid. return len(value.Readers) == 1 && value.Readers[uid] } // If the uid is already a writer on this region, then @@ -304,7 +292,8 @@ func (l LockSet) canLock(uid UniqueID, t LockType, r LockRange) bool { } } -// lock returns true if uid took a lock of type t on the entire range of LockRange. +// lock returns true if uid took a lock of type t on the entire range of +// LockRange. // // Preconditions: r.Start <= r.End (will panic otherwise). func (l *LockSet) lock(uid UniqueID, t LockType, r LockRange) bool { @@ -339,7 +328,7 @@ func (l *LockSet) lock(uid UniqueID, t LockType, r LockRange) bool { seg, _ = l.SplitUnchecked(seg, r.End) } - // Set the lock on the segment. This is guaranteed to + // Set the lock on the segment. This is guaranteed to // always be safe, given canLock above. value := seg.ValuePtr() value.lock(uid, t) @@ -386,7 +375,7 @@ func (l *LockSet) unlock(uid UniqueID, r LockRange) { value := seg.Value() var remove bool - if value.HasWriter && value.Writer == uid { + if value.Writer == uid { // If we are unlocking a writer, then since there can // only ever be one writer and no readers, then this // lock should always be removed from the set. diff --git a/pkg/sentry/fs/lock/lock_set_functions.go b/pkg/sentry/fs/lock/lock_set_functions.go index 8a3ace0c1..50a16e662 100644 --- a/pkg/sentry/fs/lock/lock_set_functions.go +++ b/pkg/sentry/fs/lock/lock_set_functions.go @@ -44,14 +44,9 @@ func (lockSetFunctions) Merge(r1 LockRange, val1 Lock, r2 LockRange, val2 Lock) return Lock{}, false } } - if val1.HasWriter != val2.HasWriter { + if val1.Writer != val2.Writer { return Lock{}, false } - if val1.HasWriter { - if val1.Writer != val2.Writer { - return Lock{}, false - } - } return val1, true } @@ -62,7 +57,6 @@ func (lockSetFunctions) Split(r LockRange, val Lock, split uint64) (Lock, Lock) for k, v := range val.Readers { val0.Readers[k] = v } - val0.HasWriter = val.HasWriter val0.Writer = val.Writer return val, val0 diff --git a/pkg/sentry/fs/lock/lock_test.go b/pkg/sentry/fs/lock/lock_test.go index ba002aeb7..fad90984b 100644 --- a/pkg/sentry/fs/lock/lock_test.go +++ b/pkg/sentry/fs/lock/lock_test.go @@ -42,9 +42,6 @@ func equals(e0, e1 []entry) bool { if !reflect.DeepEqual(e0[i].LockRange, e1[i].LockRange) { return false } - if e0[i].Lock.HasWriter != e1[i].Lock.HasWriter { - return false - } if e0[i].Lock.Writer != e1[i].Lock.Writer { return false } @@ -105,7 +102,7 @@ func TestCanLock(t *testing.T) { LockRange: LockRange{2048, 3072}, }, { - Lock: Lock{HasWriter: true, Writer: 1}, + Lock: Lock{Writer: 1}, LockRange: LockRange{3072, 4096}, }, }) @@ -241,7 +238,7 @@ func TestSetLock(t *testing.T) { // 0 max uint64 after: []entry{ { - Lock: Lock{HasWriter: true, Writer: 0}, + Lock: Lock{Writer: 0}, LockRange: LockRange{0, LockEOF}, }, }, @@ -254,7 +251,7 @@ func TestSetLock(t *testing.T) { // 0 max uint64 before: []entry{ { - Lock: Lock{HasWriter: true, Writer: 0}, + Lock: Lock{Writer: 0}, LockRange: LockRange{0, LockEOF}, }, }, @@ -273,7 +270,7 @@ func TestSetLock(t *testing.T) { LockRange: LockRange{0, 4096}, }, { - Lock: Lock{HasWriter: true, Writer: 0}, + Lock: Lock{Writer: 0}, LockRange: LockRange{4096, LockEOF}, }, }, @@ -301,7 +298,7 @@ func TestSetLock(t *testing.T) { // 0 4096 max uint64 after: []entry{ { - Lock: Lock{HasWriter: true, Writer: 0}, + Lock: Lock{Writer: 0}, LockRange: LockRange{0, 4096}, }, { @@ -318,7 +315,7 @@ func TestSetLock(t *testing.T) { // 0 max uint64 before: []entry{ { - Lock: Lock{HasWriter: true, Writer: 0}, + Lock: Lock{Writer: 0}, LockRange: LockRange{0, LockEOF}, }, }, @@ -550,7 +547,7 @@ func TestSetLock(t *testing.T) { LockRange: LockRange{0, 1024}, }, { - Lock: Lock{HasWriter: true, Writer: 0}, + Lock: Lock{Writer: 0}, LockRange: LockRange{1024, 4096}, }, { @@ -594,7 +591,7 @@ func TestSetLock(t *testing.T) { LockRange: LockRange{0, 1024}, }, { - Lock: Lock{HasWriter: true, Writer: 0}, + Lock: Lock{Writer: 0}, LockRange: LockRange{1024, 3072}, }, { @@ -633,7 +630,7 @@ func TestSetLock(t *testing.T) { // 0 1024 2048 4096 max uint64 before: []entry{ { - Lock: Lock{HasWriter: true, Writer: 0}, + Lock: Lock{Writer: 0}, LockRange: LockRange{0, 1024}, }, { @@ -663,11 +660,11 @@ func TestSetLock(t *testing.T) { // 0 1024 max uint64 after: []entry{ { - Lock: Lock{HasWriter: true, Writer: 0}, + Lock: Lock{Writer: 0}, LockRange: LockRange{0, 1024}, }, { - Lock: Lock{HasWriter: true, Writer: 0}, + Lock: Lock{Writer: 0}, LockRange: LockRange{1024, LockEOF}, }, }, @@ -675,28 +672,30 @@ func TestSetLock(t *testing.T) { } for _, test := range tests { - l := fill(test.before) + t.Run(test.name, func(t *testing.T) { + l := fill(test.before) - r := LockRange{Start: test.start, End: test.end} - success := l.lock(test.uid, test.lockType, r) - var got []entry - for seg := l.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { - got = append(got, entry{ - Lock: seg.Value(), - LockRange: seg.Range(), - }) - } + r := LockRange{Start: test.start, End: test.end} + success := l.lock(test.uid, test.lockType, r) + var got []entry + for seg := l.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + got = append(got, entry{ + Lock: seg.Value(), + LockRange: seg.Range(), + }) + } - if success != test.success { - t.Errorf("%s: setlock(%v, %+v, %d, %d) got success %v, want %v", test.name, test.before, r, test.uid, test.lockType, success, test.success) - continue - } + if success != test.success { + t.Errorf("setlock(%v, %+v, %d, %d) got success %v, want %v", test.before, r, test.uid, test.lockType, success, test.success) + return + } - if success { - if !equals(got, test.after) { - t.Errorf("%s: got set %+v, want %+v", test.name, got, test.after) + if success { + if !equals(got, test.after) { + t.Errorf("got set %+v, want %+v", got, test.after) + } } - } + }) } } @@ -782,7 +781,7 @@ func TestUnlock(t *testing.T) { // 0 max uint64 before: []entry{ { - Lock: Lock{HasWriter: true, Writer: 0}, + Lock: Lock{Writer: 0}, LockRange: LockRange{0, LockEOF}, }, }, @@ -824,7 +823,7 @@ func TestUnlock(t *testing.T) { // 0 max uint64 before: []entry{ { - Lock: Lock{HasWriter: true, Writer: 0}, + Lock: Lock{Writer: 0}, LockRange: LockRange{0, LockEOF}, }, }, @@ -837,7 +836,7 @@ func TestUnlock(t *testing.T) { // 0 4096 max uint64 after: []entry{ { - Lock: Lock{HasWriter: true, Writer: 0}, + Lock: Lock{Writer: 0}, LockRange: LockRange{4096, LockEOF}, }, }, @@ -876,7 +875,7 @@ func TestUnlock(t *testing.T) { // 0 max uint64 before: []entry{ { - Lock: Lock{HasWriter: true, Writer: 0}, + Lock: Lock{Writer: 0}, LockRange: LockRange{0, LockEOF}, }, }, @@ -889,7 +888,7 @@ func TestUnlock(t *testing.T) { // 0 4096 after: []entry{ { - Lock: Lock{HasWriter: true, Writer: 0}, + Lock: Lock{Writer: 0}, LockRange: LockRange{0, 4096}, }, }, @@ -906,7 +905,7 @@ func TestUnlock(t *testing.T) { LockRange: LockRange{0, 1024}, }, { - Lock: Lock{HasWriter: true, Writer: 0}, + Lock: Lock{Writer: 0}, LockRange: LockRange{1024, 4096}, }, { @@ -974,7 +973,7 @@ func TestUnlock(t *testing.T) { // 0 1024 4096 max uint64 before: []entry{ { - Lock: Lock{HasWriter: true, Writer: 0}, + Lock: Lock{Writer: 0}, LockRange: LockRange{0, 1024}, }, { @@ -991,7 +990,7 @@ func TestUnlock(t *testing.T) { // 0 8 4096 max uint64 after: []entry{ { - Lock: Lock{HasWriter: true, Writer: 0}, + Lock: Lock{Writer: 0}, LockRange: LockRange{0, 8}, }, { @@ -1008,7 +1007,7 @@ func TestUnlock(t *testing.T) { // 0 1024 4096 max uint64 before: []entry{ { - Lock: Lock{HasWriter: true, Writer: 0}, + Lock: Lock{Writer: 0}, LockRange: LockRange{0, 1024}, }, { @@ -1025,7 +1024,7 @@ func TestUnlock(t *testing.T) { // 0 1024 4096 8192 max uint64 after: []entry{ { - Lock: Lock{HasWriter: true, Writer: 0}, + Lock: Lock{Writer: 0}, LockRange: LockRange{0, 1024}, }, { @@ -1041,19 +1040,21 @@ func TestUnlock(t *testing.T) { } for _, test := range tests { - l := fill(test.before) + t.Run(test.name, func(t *testing.T) { + l := fill(test.before) - r := LockRange{Start: test.start, End: test.end} - l.unlock(test.uid, r) - var got []entry - for seg := l.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { - got = append(got, entry{ - Lock: seg.Value(), - LockRange: seg.Range(), - }) - } - if !equals(got, test.after) { - t.Errorf("%s: got set %+v, want %+v", test.name, got, test.after) - } + r := LockRange{Start: test.start, End: test.end} + l.unlock(test.uid, r) + var got []entry + for seg := l.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + got = append(got, entry{ + Lock: seg.Value(), + LockRange: seg.Range(), + }) + } + if !equals(got, test.after) { + t.Errorf("got set %+v, want %+v", got, test.after) + } + }) } } diff --git a/pkg/sentry/fsimpl/devpts/BUILD b/pkg/sentry/fsimpl/devpts/BUILD index 585764223..cf440dce8 100644 --- a/pkg/sentry/fsimpl/devpts/BUILD +++ b/pkg/sentry/fsimpl/devpts/BUILD @@ -23,6 +23,7 @@ go_library( "//pkg/sentry/kernel/auth", "//pkg/sentry/unimpl", "//pkg/sentry/vfs", + "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserror", "//pkg/usermem", diff --git a/pkg/sentry/fsimpl/devpts/devpts.go b/pkg/sentry/fsimpl/devpts/devpts.go index c03c65445..9b0e0cca2 100644 --- a/pkg/sentry/fsimpl/devpts/devpts.go +++ b/pkg/sentry/fsimpl/devpts/devpts.go @@ -28,6 +28,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" ) @@ -116,6 +117,8 @@ type rootInode struct { kernfs.InodeNotSymlink kernfs.OrderedChildren + locks lock.FileLocks + // Keep a reference to this inode's dentry. dentry kernfs.Dentry @@ -183,7 +186,7 @@ func (i *rootInode) masterClose(t *Terminal) { // Open implements kernfs.Inode.Open. func (i *rootInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &opts) + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &i.locks, &opts) if err != nil { return nil, err } diff --git a/pkg/sentry/fsimpl/devpts/master.go b/pkg/sentry/fsimpl/devpts/master.go index 7a7ce5d81..1d22adbe3 100644 --- a/pkg/sentry/fsimpl/devpts/master.go +++ b/pkg/sentry/fsimpl/devpts/master.go @@ -22,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/unimpl" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" @@ -34,6 +35,8 @@ type masterInode struct { kernfs.InodeNotDirectory kernfs.InodeNotSymlink + locks lock.FileLocks + // Keep a reference to this inode's dentry. dentry kernfs.Dentry @@ -55,6 +58,7 @@ func (mi *masterInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vf inode: mi, t: t, } + fd.LockFD.Init(&mi.locks) if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}); err != nil { mi.DecRef() return nil, err @@ -85,6 +89,7 @@ func (mi *masterInode) SetStat(ctx context.Context, vfsfs *vfs.Filesystem, creds type masterFileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl + vfs.LockFD inode *masterInode t *Terminal diff --git a/pkg/sentry/fsimpl/devpts/slave.go b/pkg/sentry/fsimpl/devpts/slave.go index 526cd406c..7fe475080 100644 --- a/pkg/sentry/fsimpl/devpts/slave.go +++ b/pkg/sentry/fsimpl/devpts/slave.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" @@ -33,6 +34,8 @@ type slaveInode struct { kernfs.InodeNotDirectory kernfs.InodeNotSymlink + locks lock.FileLocks + // Keep a reference to this inode's dentry. dentry kernfs.Dentry @@ -51,6 +54,7 @@ func (si *slaveInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs fd := &slaveFileDescription{ inode: si, } + fd.LockFD.Init(&si.locks) if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}); err != nil { si.DecRef() return nil, err @@ -91,6 +95,7 @@ func (si *slaveInode) SetStat(ctx context.Context, vfsfs *vfs.Filesystem, creds type slaveFileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl + vfs.LockFD inode *slaveInode } diff --git a/pkg/sentry/fsimpl/eventfd/eventfd.go b/pkg/sentry/fsimpl/eventfd/eventfd.go index c573d7935..d12d78b84 100644 --- a/pkg/sentry/fsimpl/eventfd/eventfd.go +++ b/pkg/sentry/fsimpl/eventfd/eventfd.go @@ -37,6 +37,7 @@ type EventFileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl vfs.DentryMetadataFileDescriptionImpl + vfs.NoLockFD // queue is used to notify interested parties when the event object // becomes readable or writable. diff --git a/pkg/sentry/fsimpl/ext/BUILD b/pkg/sentry/fsimpl/ext/BUILD index ff861d0fe..973fa0def 100644 --- a/pkg/sentry/fsimpl/ext/BUILD +++ b/pkg/sentry/fsimpl/ext/BUILD @@ -60,6 +60,7 @@ go_library( "//pkg/sentry/socket/unix/transport", "//pkg/sentry/syscalls/linux", "//pkg/sentry/vfs", + "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserror", "//pkg/usermem", diff --git a/pkg/sentry/fsimpl/ext/file_description.go b/pkg/sentry/fsimpl/ext/file_description.go index 92f7da40d..90b086468 100644 --- a/pkg/sentry/fsimpl/ext/file_description.go +++ b/pkg/sentry/fsimpl/ext/file_description.go @@ -26,6 +26,7 @@ import ( type fileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl + vfs.LockFD } func (fd *fileDescription) filesystem() *filesystem { diff --git a/pkg/sentry/fsimpl/ext/inode.go b/pkg/sentry/fsimpl/ext/inode.go index 485f86f4b..e4b434b13 100644 --- a/pkg/sentry/fsimpl/ext/inode.go +++ b/pkg/sentry/fsimpl/ext/inode.go @@ -22,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" ) @@ -54,6 +55,8 @@ type inode struct { // diskInode gives us access to the inode struct on disk. Immutable. diskInode disklayout.Inode + locks lock.FileLocks + // This is immutable. The first field of the implementations must have inode // as the first field to ensure temporality. impl interface{} @@ -157,6 +160,7 @@ func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts *vfs.OpenOpt switch in.impl.(type) { case *regularFile: var fd regularFileFD + fd.LockFD.Init(&in.locks) if err := fd.vfsfd.Init(&fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{}); err != nil { return nil, err } @@ -168,6 +172,7 @@ func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts *vfs.OpenOpt return nil, syserror.EISDIR } var fd directoryFD + fd.LockFD.Init(&in.locks) if err := fd.vfsfd.Init(&fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{}); err != nil { return nil, err } @@ -178,6 +183,7 @@ func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts *vfs.OpenOpt return nil, syserror.ELOOP } var fd symlinkFD + fd.LockFD.Init(&in.locks) fd.vfsfd.Init(&fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{}) return &fd.vfsfd, nil default: diff --git a/pkg/sentry/fsimpl/ext/regular_file.go b/pkg/sentry/fsimpl/ext/regular_file.go index 30135ddb0..f7015c44f 100644 --- a/pkg/sentry/fsimpl/ext/regular_file.go +++ b/pkg/sentry/fsimpl/ext/regular_file.go @@ -77,6 +77,7 @@ func (in *inode) isRegular() bool { // vfs.FileDescriptionImpl. type regularFileFD struct { fileDescription + vfs.LockFD // off is the file offset. off is accessed using atomic memory operations. off int64 diff --git a/pkg/sentry/fsimpl/gofer/BUILD b/pkg/sentry/fsimpl/gofer/BUILD index f5f35a3bc..5cdeeaeb5 100644 --- a/pkg/sentry/fsimpl/gofer/BUILD +++ b/pkg/sentry/fsimpl/gofer/BUILD @@ -54,6 +54,7 @@ go_library( "//pkg/p9", "//pkg/safemem", "//pkg/sentry/fs/fsutil", + "//pkg/sentry/fs/lock", "//pkg/sentry/fsimpl/host", "//pkg/sentry/hostfd", "//pkg/sentry/kernel", @@ -68,6 +69,7 @@ go_library( "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usage", "//pkg/sentry/vfs", + "//pkg/sentry/vfs/lock", "//pkg/syserr", "//pkg/syserror", "//pkg/unet", diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 36e0e1856..40933b74b 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -801,6 +801,7 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vf return nil, err } fd := ®ularFileFD{} + fd.LockFD.Init(&d.locks) if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{ AllowDirectIO: true, }); err != nil { @@ -826,6 +827,7 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vf } } fd := &directoryFD{} + fd.LockFD.Init(&d.locks) if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { return nil, err } @@ -842,7 +844,7 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vf } case linux.S_IFIFO: if d.isSynthetic() { - return d.pipe.Open(ctx, mnt, &d.vfsd, opts.Flags) + return d.pipe.Open(ctx, mnt, &d.vfsd, opts.Flags, &d.locks) } } return d.openSpecialFileLocked(ctx, mnt, opts) @@ -902,7 +904,7 @@ retry: return nil, err } } - fd, err := newSpecialFileFD(h, mnt, d, opts.Flags) + fd, err := newSpecialFileFD(h, mnt, d, &d.locks, opts.Flags) if err != nil { h.close(ctx) return nil, err @@ -989,6 +991,7 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving var childVFSFD *vfs.FileDescription if useRegularFileFD { fd := ®ularFileFD{} + fd.LockFD.Init(&child.locks) if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &child.vfsd, &vfs.FileDescriptionOptions{ AllowDirectIO: true, }); err != nil { @@ -1003,7 +1006,7 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving if fdobj != nil { h.fd = int32(fdobj.Release()) } - fd, err := newSpecialFileFD(h, mnt, child, opts.Flags) + fd, err := newSpecialFileFD(h, mnt, child, &d.locks, opts.Flags) if err != nil { h.close(ctx) return nil, err diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 3f3bd56f0..0d88a328e 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -45,6 +45,7 @@ import ( "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" @@ -52,6 +53,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/unet" "gvisor.dev/gvisor/pkg/usermem" @@ -662,6 +664,8 @@ type dentry struct { // If this dentry represents a synthetic named pipe, pipe is the pipe // endpoint bound to this file. pipe *pipe.VFSPipe + + locks lock.FileLocks } // dentryAttrMask returns a p9.AttrMask enabling all attributes used by the @@ -1366,6 +1370,9 @@ func (d *dentry) decLinks() { type fileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl + vfs.LockFD + + lockLogging sync.Once } func (fd *fileDescription) filesystem() *filesystem { @@ -1416,3 +1423,19 @@ func (fd *fileDescription) Setxattr(ctx context.Context, opts vfs.SetxattrOption func (fd *fileDescription) Removexattr(ctx context.Context, name string) error { return fd.dentry().removexattr(ctx, auth.CredentialsFromContext(ctx), name) } + +// LockBSD implements vfs.FileDescriptionImpl.LockBSD. +func (fd *fileDescription) LockBSD(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, block fslock.Blocker) error { + fd.lockLogging.Do(func() { + log.Infof("File lock using gofer file handled internally.") + }) + return fd.LockFD.LockBSD(ctx, uid, t, block) +} + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (fd *fileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, rng fslock.LockRange, block fslock.Blocker) error { + fd.lockLogging.Do(func() { + log.Infof("Range lock using gofer file handled internally.") + }) + return fd.LockFD.LockPOSIX(ctx, uid, t, rng, block) +} diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go index ff6126b87..289efdd25 100644 --- a/pkg/sentry/fsimpl/gofer/special_file.go +++ b/pkg/sentry/fsimpl/gofer/special_file.go @@ -22,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/fdnotifier" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" @@ -51,7 +52,7 @@ type specialFileFD struct { off int64 } -func newSpecialFileFD(h handle, mnt *vfs.Mount, d *dentry, flags uint32) (*specialFileFD, error) { +func newSpecialFileFD(h handle, mnt *vfs.Mount, d *dentry, locks *lock.FileLocks, flags uint32) (*specialFileFD, error) { ftype := d.fileType() seekable := ftype == linux.S_IFREG mayBlock := ftype == linux.S_IFIFO || ftype == linux.S_IFSOCK @@ -60,6 +61,7 @@ func newSpecialFileFD(h handle, mnt *vfs.Mount, d *dentry, flags uint32) (*speci seekable: seekable, mayBlock: mayBlock, } + fd.LockFD.Init(locks) if mayBlock && h.fd >= 0 { if err := fdnotifier.AddFD(h.fd, &fd.queue); err != nil { return nil, err diff --git a/pkg/sentry/fsimpl/host/BUILD b/pkg/sentry/fsimpl/host/BUILD index ca0fe6d2b..54f16ad63 100644 --- a/pkg/sentry/fsimpl/host/BUILD +++ b/pkg/sentry/fsimpl/host/BUILD @@ -39,6 +39,7 @@ go_library( "//pkg/sentry/unimpl", "//pkg/sentry/uniqueid", "//pkg/sentry/vfs", + "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserr", "//pkg/syserror", diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 18b127521..5ec5100b8 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -34,6 +34,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/memmap" unixsocket "gvisor.dev/gvisor/pkg/sentry/socket/unix" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -182,6 +183,8 @@ type inode struct { kernfs.InodeNotDirectory kernfs.InodeNotSymlink + locks lock.FileLocks + // When the reference count reaches zero, the host fd is closed. refs.AtomicRefCount @@ -468,7 +471,7 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount, flags u return nil, err } // Currently, we only allow Unix sockets to be imported. - return unixsocket.NewFileDescription(ep, ep.Type(), flags, mnt, d) + return unixsocket.NewFileDescription(ep, ep.Type(), flags, mnt, d, &i.locks) } // TODO(gvisor.dev/issue/1672): Whitelist specific file types here, so that @@ -478,6 +481,7 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount, flags u fileDescription: fileDescription{inode: i}, termios: linux.DefaultSlaveTermios, } + fd.LockFD.Init(&i.locks) vfsfd := &fd.vfsfd if err := vfsfd.Init(fd, flags, mnt, d, &vfs.FileDescriptionOptions{}); err != nil { return nil, err @@ -486,6 +490,7 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount, flags u } fd := &fileDescription{inode: i} + fd.LockFD.Init(&i.locks) vfsfd := &fd.vfsfd if err := vfsfd.Init(fd, flags, mnt, d, &vfs.FileDescriptionOptions{}); err != nil { return nil, err @@ -497,6 +502,7 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount, flags u type fileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl + vfs.LockFD // inode is vfsfd.Dentry().Impl().(*kernfs.Dentry).Inode().(*inode), but // cached to reduce indirections and casting. fileDescription does not hold diff --git a/pkg/sentry/fsimpl/kernfs/BUILD b/pkg/sentry/fsimpl/kernfs/BUILD index ef34cb28a..0299dbde9 100644 --- a/pkg/sentry/fsimpl/kernfs/BUILD +++ b/pkg/sentry/fsimpl/kernfs/BUILD @@ -49,6 +49,7 @@ go_library( "//pkg/sentry/memmap", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/vfs", + "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserror", "//pkg/usermem", @@ -67,6 +68,7 @@ go_test( "//pkg/sentry/fsimpl/testutil", "//pkg/sentry/kernel/auth", "//pkg/sentry/vfs", + "//pkg/sentry/vfs/lock", "//pkg/syserror", "//pkg/usermem", "@com_github_google_go-cmp//cmp:go_default_library", diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go index 1568a9d49..6418de0a3 100644 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" ) @@ -38,7 +39,8 @@ type DynamicBytesFile struct { InodeNotDirectory InodeNotSymlink - data vfs.DynamicBytesSource + locks lock.FileLocks + data vfs.DynamicBytesSource } var _ Inode = (*DynamicBytesFile)(nil) @@ -55,7 +57,7 @@ func (f *DynamicBytesFile) Init(creds *auth.Credentials, devMajor, devMinor uint // Open implements Inode.Open. func (f *DynamicBytesFile) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd := &DynamicBytesFD{} - if err := fd.Init(rp.Mount(), vfsd, f.data, opts.Flags); err != nil { + if err := fd.Init(rp.Mount(), vfsd, f.data, &f.locks, opts.Flags); err != nil { return nil, err } return &fd.vfsfd, nil @@ -77,13 +79,15 @@ func (*DynamicBytesFile) SetStat(context.Context, *vfs.Filesystem, *auth.Credent type DynamicBytesFD struct { vfs.FileDescriptionDefaultImpl vfs.DynamicBytesFileDescriptionImpl + vfs.LockFD vfsfd vfs.FileDescription inode Inode } // Init initializes a DynamicBytesFD. -func (fd *DynamicBytesFD) Init(m *vfs.Mount, d *vfs.Dentry, data vfs.DynamicBytesSource, flags uint32) error { +func (fd *DynamicBytesFD) Init(m *vfs.Mount, d *vfs.Dentry, data vfs.DynamicBytesSource, locks *lock.FileLocks, flags uint32) error { + fd.LockFD.Init(locks) if err := fd.vfsfd.Init(fd, flags, m, d, &vfs.FileDescriptionOptions{}); err != nil { return err } diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index 8284e76a7..33a5968ca 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -22,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -42,6 +43,7 @@ import ( type GenericDirectoryFD struct { vfs.FileDescriptionDefaultImpl vfs.DirectoryFileDescriptionDefaultImpl + vfs.LockFD vfsfd vfs.FileDescription children *OrderedChildren @@ -55,9 +57,9 @@ type GenericDirectoryFD struct { // NewGenericDirectoryFD creates a new GenericDirectoryFD and returns its // dentry. -func NewGenericDirectoryFD(m *vfs.Mount, d *vfs.Dentry, children *OrderedChildren, opts *vfs.OpenOptions) (*GenericDirectoryFD, error) { +func NewGenericDirectoryFD(m *vfs.Mount, d *vfs.Dentry, children *OrderedChildren, locks *lock.FileLocks, opts *vfs.OpenOptions) (*GenericDirectoryFD, error) { fd := &GenericDirectoryFD{} - if err := fd.Init(children, opts); err != nil { + if err := fd.Init(children, locks, opts); err != nil { return nil, err } if err := fd.vfsfd.Init(fd, opts.Flags, m, d, &vfs.FileDescriptionOptions{}); err != nil { @@ -69,11 +71,12 @@ func NewGenericDirectoryFD(m *vfs.Mount, d *vfs.Dentry, children *OrderedChildre // Init initializes a GenericDirectoryFD. Use it when overriding // GenericDirectoryFD. Caller must call fd.VFSFileDescription.Init() with the // correct implementation. -func (fd *GenericDirectoryFD) Init(children *OrderedChildren, opts *vfs.OpenOptions) error { +func (fd *GenericDirectoryFD) Init(children *OrderedChildren, locks *lock.FileLocks, opts *vfs.OpenOptions) error { if vfs.AccessTypesForOpenFlags(opts)&vfs.MayWrite != 0 { // Can't open directories for writing. return syserror.EISDIR } + fd.LockFD.Init(locks) fd.children = children return nil } diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 982daa2e6..0e4927215 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -23,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) @@ -555,6 +556,8 @@ type StaticDirectory struct { InodeAttrs InodeNoDynamicLookup OrderedChildren + + locks lock.FileLocks } var _ Inode = (*StaticDirectory)(nil) @@ -584,7 +587,7 @@ func (s *StaticDirectory) Init(creds *auth.Credentials, devMajor, devMinor uint3 // Open implements kernfs.Inode. func (s *StaticDirectory) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := NewGenericDirectoryFD(rp.Mount(), vfsd, &s.OrderedChildren, &opts) + fd, err := NewGenericDirectoryFD(rp.Mount(), vfsd, &s.OrderedChildren, &s.locks, &opts) if err != nil { return nil, err } diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index 412cf6ac9..6749facf7 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -27,6 +27,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fsimpl/testutil" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" ) @@ -100,8 +101,10 @@ type readonlyDir struct { kernfs.InodeNotSymlink kernfs.InodeNoDynamicLookup kernfs.InodeDirectoryNoNewChildren - kernfs.OrderedChildren + + locks lock.FileLocks + dentry kernfs.Dentry } @@ -117,7 +120,7 @@ func (fs *filesystem) newReadonlyDir(creds *auth.Credentials, mode linux.FileMod } func (d *readonlyDir) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &d.OrderedChildren, &opts) + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &d.OrderedChildren, &d.locks, &opts) if err != nil { return nil, err } @@ -128,10 +131,12 @@ type dir struct { attrs kernfs.InodeNotSymlink kernfs.InodeNoDynamicLookup + kernfs.OrderedChildren + + locks lock.FileLocks fs *filesystem dentry kernfs.Dentry - kernfs.OrderedChildren } func (fs *filesystem) newDir(creds *auth.Credentials, mode linux.FileMode, contents map[string]*kernfs.Dentry) *kernfs.Dentry { @@ -147,7 +152,7 @@ func (fs *filesystem) newDir(creds *auth.Credentials, mode linux.FileMode, conte } func (d *dir) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &d.OrderedChildren, &opts) + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &d.OrderedChildren, &d.locks, &opts) if err != nil { return nil, err } diff --git a/pkg/sentry/fsimpl/pipefs/BUILD b/pkg/sentry/fsimpl/pipefs/BUILD index 5950a2d59..c618dbe6c 100644 --- a/pkg/sentry/fsimpl/pipefs/BUILD +++ b/pkg/sentry/fsimpl/pipefs/BUILD @@ -15,6 +15,7 @@ go_library( "//pkg/sentry/kernel/pipe", "//pkg/sentry/kernel/time", "//pkg/sentry/vfs", + "//pkg/sentry/vfs/lock", "//pkg/syserror", "//pkg/usermem", ], diff --git a/pkg/sentry/fsimpl/pipefs/pipefs.go b/pkg/sentry/fsimpl/pipefs/pipefs.go index cab771211..e4dabaa33 100644 --- a/pkg/sentry/fsimpl/pipefs/pipefs.go +++ b/pkg/sentry/fsimpl/pipefs/pipefs.go @@ -27,6 +27,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" ) @@ -81,7 +82,8 @@ type inode struct { kernfs.InodeNotSymlink kernfs.InodeNoopRefCount - pipe *pipe.VFSPipe + locks lock.FileLocks + pipe *pipe.VFSPipe ino uint64 uid auth.KUID @@ -147,7 +149,7 @@ func (i *inode) SetStat(ctx context.Context, vfsfs *vfs.Filesystem, creds *auth. // Open implements kernfs.Inode.Open. func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - return i.pipe.Open(ctx, rp.Mount(), vfsd, opts.Flags) + return i.pipe.Open(ctx, rp.Mount(), vfsd, opts.Flags, &i.locks) } // NewConnectedPipeFDs returns a pair of FileDescriptions representing the read diff --git a/pkg/sentry/fsimpl/proc/BUILD b/pkg/sentry/fsimpl/proc/BUILD index 17c1342b5..351ba4ee9 100644 --- a/pkg/sentry/fsimpl/proc/BUILD +++ b/pkg/sentry/fsimpl/proc/BUILD @@ -35,6 +35,7 @@ go_library( "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usage", "//pkg/sentry/vfs", + "//pkg/sentry/vfs/lock", "//pkg/syserror", "//pkg/tcpip/header", "//pkg/usermem", diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go index 36a911db4..e2cdb7ee9 100644 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -24,6 +24,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" ) @@ -37,6 +38,8 @@ type subtasksInode struct { kernfs.OrderedChildren kernfs.AlwaysValid + locks lock.FileLocks + fs *filesystem task *kernel.Task pidns *kernel.PIDNamespace @@ -153,7 +156,7 @@ func (fd *subtasksFD) SetStat(ctx context.Context, opts vfs.SetStatOptions) erro // Open implements kernfs.Inode. func (i *subtasksInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd := &subtasksFD{task: i.task} - if err := fd.Init(&i.OrderedChildren, &opts); err != nil { + if err := fd.Init(&i.OrderedChildren, &i.locks, &opts); err != nil { return nil, err } if err := fd.VFSFileDescription().Init(fd, opts.Flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}); err != nil { diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index 482055db1..44078a765 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -25,6 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/mm" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" ) @@ -38,6 +39,8 @@ type taskInode struct { kernfs.InodeAttrs kernfs.OrderedChildren + locks lock.FileLocks + task *kernel.Task } @@ -103,7 +106,7 @@ func (i *taskInode) Valid(ctx context.Context) bool { // Open implements kernfs.Inode. func (i *taskInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &opts) + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &i.locks, &opts) if err != nil { return nil, err } diff --git a/pkg/sentry/fsimpl/proc/task_fds.go b/pkg/sentry/fsimpl/proc/task_fds.go index 44ccc9e4a..ef6c1d04f 100644 --- a/pkg/sentry/fsimpl/proc/task_fds.go +++ b/pkg/sentry/fsimpl/proc/task_fds.go @@ -27,6 +27,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" ) @@ -53,6 +54,8 @@ func taskFDExists(t *kernel.Task, fd int32) bool { } type fdDir struct { + locks lock.FileLocks + fs *filesystem task *kernel.Task @@ -143,7 +146,7 @@ func (i *fdDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, erro // Open implements kernfs.Inode. func (i *fdDirInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &opts) + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &i.locks, &opts) if err != nil { return nil, err } @@ -270,7 +273,7 @@ func (i *fdInfoDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, // Open implements kernfs.Inode. func (i *fdInfoDirInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &opts) + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &i.locks, &opts) if err != nil { return nil, err } diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index 2f297e48a..e5eaa91cd 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -30,6 +30,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/mm" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" ) @@ -775,6 +776,8 @@ type namespaceInode struct { kernfs.InodeNoopRefCount kernfs.InodeNotDirectory kernfs.InodeNotSymlink + + locks lock.FileLocks } var _ kernfs.Inode = (*namespaceInode)(nil) @@ -791,6 +794,7 @@ func (i *namespaceInode) Init(creds *auth.Credentials, devMajor, devMinor uint32 func (i *namespaceInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd := &namespaceFD{inode: i} i.IncRef() + fd.LockFD.Init(&i.locks) if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}); err != nil { return nil, err } @@ -801,6 +805,7 @@ func (i *namespaceInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd * // /proc/[pid]/ns/*. type namespaceFD struct { vfs.FileDescriptionDefaultImpl + vfs.LockFD vfsfd vfs.FileDescription inode *namespaceInode @@ -825,8 +830,3 @@ func (fd *namespaceFD) SetStat(ctx context.Context, opts vfs.SetStatOptions) err func (fd *namespaceFD) Release() { fd.inode.DecRef() } - -// OnClose implements FileDescriptionImpl. -func (*namespaceFD) OnClose(context.Context) error { - return nil -} diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index b51d43954..58c8b9d05 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -25,6 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" ) @@ -43,6 +44,8 @@ type tasksInode struct { kernfs.OrderedChildren kernfs.AlwaysValid + locks lock.FileLocks + fs *filesystem pidns *kernel.PIDNamespace @@ -197,7 +200,7 @@ func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback // Open implements kernfs.Inode. func (i *tasksInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &opts) + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &i.locks, &opts) if err != nil { return nil, err } diff --git a/pkg/sentry/fsimpl/signalfd/signalfd.go b/pkg/sentry/fsimpl/signalfd/signalfd.go index d29ef3f83..242ba9b5d 100644 --- a/pkg/sentry/fsimpl/signalfd/signalfd.go +++ b/pkg/sentry/fsimpl/signalfd/signalfd.go @@ -31,6 +31,7 @@ type SignalFileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl vfs.DentryMetadataFileDescriptionImpl + vfs.NoLockFD // target is the original signal target task. // diff --git a/pkg/sentry/fsimpl/sys/BUILD b/pkg/sentry/fsimpl/sys/BUILD index a741e2bb6..237f17def 100644 --- a/pkg/sentry/fsimpl/sys/BUILD +++ b/pkg/sentry/fsimpl/sys/BUILD @@ -15,6 +15,7 @@ go_library( "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/vfs", + "//pkg/sentry/vfs/lock", "//pkg/syserror", ], ) diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go index 0af373604..b84463d3a 100644 --- a/pkg/sentry/fsimpl/sys/sys.go +++ b/pkg/sentry/fsimpl/sys/sys.go @@ -25,6 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" ) @@ -98,8 +99,10 @@ type dir struct { kernfs.InodeNoDynamicLookup kernfs.InodeNotSymlink kernfs.InodeDirectoryNoNewChildren - kernfs.OrderedChildren + + locks lock.FileLocks + dentry kernfs.Dentry } @@ -121,7 +124,7 @@ func (*dir) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.Set // Open implements kernfs.Inode.Open. func (d *dir) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &d.OrderedChildren, &opts) + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &d.OrderedChildren, &d.locks, &opts) if err != nil { return nil, err } diff --git a/pkg/sentry/fsimpl/timerfd/timerfd.go b/pkg/sentry/fsimpl/timerfd/timerfd.go index 60c92d626..2dc90d484 100644 --- a/pkg/sentry/fsimpl/timerfd/timerfd.go +++ b/pkg/sentry/fsimpl/timerfd/timerfd.go @@ -32,6 +32,7 @@ type TimerFileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl vfs.DentryMetadataFileDescriptionImpl + vfs.NoLockFD events waiter.Queue timer *ktime.Timer diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index e801680e8..72399b321 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -399,6 +399,7 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.Open switch impl := d.inode.impl.(type) { case *regularFile: var fd regularFileFD + fd.LockFD.Init(&d.inode.locks) if err := fd.vfsfd.Init(&fd, opts.Flags, rp.Mount(), &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { return nil, err } @@ -414,15 +415,16 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.Open return nil, syserror.EISDIR } var fd directoryFD + fd.LockFD.Init(&d.inode.locks) if err := fd.vfsfd.Init(&fd, opts.Flags, rp.Mount(), &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { return nil, err } return &fd.vfsfd, nil case *symlink: - // Can't open symlinks without O_PATH (which is unimplemented). + // TODO(gvisor.dev/issue/2782): Can't open symlinks without O_PATH. return nil, syserror.ELOOP case *namedPipe: - return impl.pipe.Open(ctx, rp.Mount(), &d.vfsd, opts.Flags) + return impl.pipe.Open(ctx, rp.Mount(), &d.vfsd, opts.Flags, &d.inode.locks) case *deviceFile: return rp.VirtualFilesystem().OpenDeviceSpecialFile(ctx, rp.Mount(), &d.vfsd, impl.kind, impl.major, impl.minor, opts) case *socketFile: diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index 4f2ae04d2..77447b32c 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -25,7 +25,6 @@ import ( "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" @@ -366,28 +365,6 @@ func (fd *regularFileFD) Sync(ctx context.Context) error { return nil } -// LockBSD implements vfs.FileDescriptionImpl.LockBSD. -func (fd *regularFileFD) LockBSD(ctx context.Context, uid lock.UniqueID, t lock.LockType, block lock.Blocker) error { - return fd.inode().lockBSD(uid, t, block) -} - -// UnlockBSD implements vfs.FileDescriptionImpl.UnlockBSD. -func (fd *regularFileFD) UnlockBSD(ctx context.Context, uid lock.UniqueID) error { - fd.inode().unlockBSD(uid) - return nil -} - -// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. -func (fd *regularFileFD) LockPOSIX(ctx context.Context, uid lock.UniqueID, t lock.LockType, rng lock.LockRange, block lock.Blocker) error { - return fd.inode().lockPOSIX(uid, t, rng, block) -} - -// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. -func (fd *regularFileFD) UnlockPOSIX(ctx context.Context, uid lock.UniqueID, rng lock.LockRange) error { - fd.inode().unlockPOSIX(uid, rng) - return nil -} - // ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. func (fd *regularFileFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { file := fd.inode().impl.(*regularFile) diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 7ce1b86c7..71a7522af 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -36,7 +36,6 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" - fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/pgalloc" @@ -311,7 +310,6 @@ type inode struct { ctime int64 // nanoseconds mtime int64 // nanoseconds - // Advisory file locks, which lock at the inode level. locks lock.FileLocks // Inotify watches for this inode. @@ -539,44 +537,6 @@ func (i *inode) setStat(ctx context.Context, creds *auth.Credentials, stat *linu return nil } -// TODO(gvisor.dev/issue/1480): support file locking for file types other than regular. -func (i *inode) lockBSD(uid fslock.UniqueID, t fslock.LockType, block fslock.Blocker) error { - switch i.impl.(type) { - case *regularFile: - return i.locks.LockBSD(uid, t, block) - } - return syserror.EBADF -} - -// TODO(gvisor.dev/issue/1480): support file locking for file types other than regular. -func (i *inode) unlockBSD(uid fslock.UniqueID) error { - switch i.impl.(type) { - case *regularFile: - i.locks.UnlockBSD(uid) - return nil - } - return syserror.EBADF -} - -// TODO(gvisor.dev/issue/1480): support file locking for file types other than regular. -func (i *inode) lockPOSIX(uid fslock.UniqueID, t fslock.LockType, rng fslock.LockRange, block fslock.Blocker) error { - switch i.impl.(type) { - case *regularFile: - return i.locks.LockPOSIX(uid, t, rng, block) - } - return syserror.EBADF -} - -// TODO(gvisor.dev/issue/1480): support file locking for file types other than regular. -func (i *inode) unlockPOSIX(uid fslock.UniqueID, rng fslock.LockRange) error { - switch i.impl.(type) { - case *regularFile: - i.locks.UnlockPOSIX(uid, rng) - return nil - } - return syserror.EBADF -} - // allocatedBlocksForSize returns the number of 512B blocks needed to // accommodate the given size in bytes, as appropriate for struct // stat::st_blocks and struct statx::stx_blocks. (Note that this 512B block @@ -708,6 +668,7 @@ func (i *inode) userXattrSupported() bool { type fileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl + vfs.LockFD } func (fd *fileDescription) filesystem() *filesystem { diff --git a/pkg/sentry/kernel/fd_table.go b/pkg/sentry/kernel/fd_table.go index dbfcef0fa..b35afafe3 100644 --- a/pkg/sentry/kernel/fd_table.go +++ b/pkg/sentry/kernel/fd_table.go @@ -80,9 +80,6 @@ type FDTable struct { refs.AtomicRefCount k *Kernel - // uid is a unique identifier. - uid uint64 - // mu protects below. mu sync.Mutex `state:"nosave"` @@ -130,7 +127,7 @@ func (f *FDTable) loadDescriptorTable(m map[int32]descriptor) { // drop drops the table reference. func (f *FDTable) drop(file *fs.File) { // Release locks. - file.Dirent.Inode.LockCtx.Posix.UnlockRegion(lock.UniqueID(f.uid), lock.LockRange{0, lock.LockEOF}) + file.Dirent.Inode.LockCtx.Posix.UnlockRegion(f, lock.LockRange{0, lock.LockEOF}) // Send inotify events. d := file.Dirent @@ -164,17 +161,9 @@ func (f *FDTable) dropVFS2(file *vfs.FileDescription) { file.DecRef() } -// ID returns a unique identifier for this FDTable. -func (f *FDTable) ID() uint64 { - return f.uid -} - // NewFDTable allocates a new FDTable that may be used by tasks in k. func (k *Kernel) NewFDTable() *FDTable { - f := &FDTable{ - k: k, - uid: atomic.AddUint64(&k.fdMapUids, 1), - } + f := &FDTable{k: k} f.init() return f } diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 5efeb3767..bcbeb6a39 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -194,11 +194,6 @@ type Kernel struct { // cpuClockTickerSetting is protected by runningTasksMu. cpuClockTickerSetting ktime.Setting - // fdMapUids is an ever-increasing counter for generating FDTable uids. - // - // fdMapUids is mutable, and is accessed using atomic memory operations. - fdMapUids uint64 - // uniqueID is used to generate unique identifiers. // // uniqueID is mutable, and is accessed using atomic memory operations. diff --git a/pkg/sentry/kernel/pipe/BUILD b/pkg/sentry/kernel/pipe/BUILD index 7bfa9075a..0db546b98 100644 --- a/pkg/sentry/kernel/pipe/BUILD +++ b/pkg/sentry/kernel/pipe/BUILD @@ -27,6 +27,7 @@ go_library( "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", "//pkg/sentry/vfs", + "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserror", "//pkg/usermem", diff --git a/pkg/sentry/kernel/pipe/vfs.go b/pkg/sentry/kernel/pipe/vfs.go index 2602bed72..c0e9ee1f4 100644 --- a/pkg/sentry/kernel/pipe/vfs.go +++ b/pkg/sentry/kernel/pipe/vfs.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -61,11 +62,13 @@ func NewVFSPipe(isNamed bool, sizeBytes, atomicIOBytes int64) *VFSPipe { // // Preconditions: statusFlags should not contain an open access mode. func (vp *VFSPipe) ReaderWriterPair(mnt *vfs.Mount, vfsd *vfs.Dentry, statusFlags uint32) (*vfs.FileDescription, *vfs.FileDescription) { - return vp.newFD(mnt, vfsd, linux.O_RDONLY|statusFlags), vp.newFD(mnt, vfsd, linux.O_WRONLY|statusFlags) + // Connected pipes share the same locks. + locks := &lock.FileLocks{} + return vp.newFD(mnt, vfsd, linux.O_RDONLY|statusFlags, locks), vp.newFD(mnt, vfsd, linux.O_WRONLY|statusFlags, locks) } // Open opens the pipe represented by vp. -func (vp *VFSPipe) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, statusFlags uint32) (*vfs.FileDescription, error) { +func (vp *VFSPipe) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, statusFlags uint32, locks *lock.FileLocks) (*vfs.FileDescription, error) { vp.mu.Lock() defer vp.mu.Unlock() @@ -75,7 +78,7 @@ func (vp *VFSPipe) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, s return nil, syserror.EINVAL } - fd := vp.newFD(mnt, vfsd, statusFlags) + fd := vp.newFD(mnt, vfsd, statusFlags, locks) // Named pipes have special blocking semantics during open: // @@ -127,10 +130,11 @@ func (vp *VFSPipe) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, s } // Preconditions: vp.mu must be held. -func (vp *VFSPipe) newFD(mnt *vfs.Mount, vfsd *vfs.Dentry, statusFlags uint32) *vfs.FileDescription { +func (vp *VFSPipe) newFD(mnt *vfs.Mount, vfsd *vfs.Dentry, statusFlags uint32, locks *lock.FileLocks) *vfs.FileDescription { fd := &VFSPipeFD{ pipe: &vp.pipe, } + fd.LockFD.Init(locks) fd.vfsfd.Init(fd, statusFlags, mnt, vfsd, &vfs.FileDescriptionOptions{ DenyPRead: true, DenyPWrite: true, @@ -159,6 +163,7 @@ type VFSPipeFD struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl vfs.DentryMetadataFileDescriptionImpl + vfs.LockFD pipe *Pipe } diff --git a/pkg/sentry/socket/hostinet/BUILD b/pkg/sentry/socket/hostinet/BUILD index e82d6cd1e..60c9896fc 100644 --- a/pkg/sentry/socket/hostinet/BUILD +++ b/pkg/sentry/socket/hostinet/BUILD @@ -34,6 +34,7 @@ go_library( "//pkg/sentry/socket", "//pkg/sentry/socket/control", "//pkg/sentry/vfs", + "//pkg/sentry/vfs/lock", "//pkg/syserr", "//pkg/syserror", "//pkg/tcpip/stack", diff --git a/pkg/sentry/socket/hostinet/socket_vfs2.go b/pkg/sentry/socket/hostinet/socket_vfs2.go index 677743113..027add1fd 100644 --- a/pkg/sentry/socket/hostinet/socket_vfs2.go +++ b/pkg/sentry/socket/hostinet/socket_vfs2.go @@ -26,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -35,6 +36,7 @@ import ( type socketVFS2 struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl + vfs.LockFD // We store metadata for hostinet sockets internally. Technically, we should // access metadata (e.g. through stat, chmod) on the host for correctness, @@ -59,6 +61,7 @@ func newVFS2Socket(t *kernel.Task, family int, stype linux.SockType, protocol in fd: fd, }, } + s.LockFD.Init(&lock.FileLocks{}) if err := fdnotifier.AddFD(int32(fd), &s.queue); err != nil { return nil, syserr.FromError(err) } diff --git a/pkg/sentry/socket/netlink/BUILD b/pkg/sentry/socket/netlink/BUILD index 7212d8644..420e573c9 100644 --- a/pkg/sentry/socket/netlink/BUILD +++ b/pkg/sentry/socket/netlink/BUILD @@ -29,6 +29,7 @@ go_library( "//pkg/sentry/socket/unix", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/vfs", + "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserr", "//pkg/syserror", diff --git a/pkg/sentry/socket/netlink/socket_vfs2.go b/pkg/sentry/socket/netlink/socket_vfs2.go index b854bf990..8bfee5193 100644 --- a/pkg/sentry/socket/netlink/socket_vfs2.go +++ b/pkg/sentry/socket/netlink/socket_vfs2.go @@ -23,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/socket/unix" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip" @@ -40,6 +41,7 @@ type SocketVFS2 struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl vfs.DentryMetadataFileDescriptionImpl + vfs.LockFD socketOpsCommon } @@ -66,7 +68,7 @@ func NewVFS2(t *kernel.Task, skType linux.SockType, protocol Protocol) (*SocketV return nil, err } - return &SocketVFS2{ + fd := &SocketVFS2{ socketOpsCommon: socketOpsCommon{ ports: t.Kernel().NetlinkPorts(), protocol: protocol, @@ -75,7 +77,9 @@ func NewVFS2(t *kernel.Task, skType linux.SockType, protocol Protocol) (*SocketV connection: connection, sendBufferSize: defaultSendBufferSize, }, - }, nil + } + fd.LockFD.Init(&lock.FileLocks{}) + return fd, nil } // Readiness implements waiter.Waitable.Readiness. diff --git a/pkg/sentry/socket/netstack/BUILD b/pkg/sentry/socket/netstack/BUILD index 8f0f5466e..0f592ecc3 100644 --- a/pkg/sentry/socket/netstack/BUILD +++ b/pkg/sentry/socket/netstack/BUILD @@ -37,6 +37,7 @@ go_library( "//pkg/sentry/socket/netfilter", "//pkg/sentry/unimpl", "//pkg/sentry/vfs", + "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserr", "//pkg/syserror", diff --git a/pkg/sentry/socket/netstack/netstack_vfs2.go b/pkg/sentry/socket/netstack/netstack_vfs2.go index fcd8013c0..1412a4810 100644 --- a/pkg/sentry/socket/netstack/netstack_vfs2.go +++ b/pkg/sentry/socket/netstack/netstack_vfs2.go @@ -25,6 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/netfilter" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip" @@ -38,6 +39,7 @@ type SocketVFS2 struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl vfs.DentryMetadataFileDescriptionImpl + vfs.LockFD socketOpsCommon } @@ -64,6 +66,7 @@ func NewVFS2(t *kernel.Task, family int, skType linux.SockType, protocol int, qu protocol: protocol, }, } + s.LockFD.Init(&lock.FileLocks{}) vfsfd := &s.vfsfd if err := vfsfd.Init(s, linux.O_RDWR, mnt, d, &vfs.FileDescriptionOptions{ DenyPRead: true, diff --git a/pkg/sentry/socket/unix/BUILD b/pkg/sentry/socket/unix/BUILD index de2cc4bdf..7d4cc80fe 100644 --- a/pkg/sentry/socket/unix/BUILD +++ b/pkg/sentry/socket/unix/BUILD @@ -29,6 +29,7 @@ go_library( "//pkg/sentry/socket/netstack", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/vfs", + "//pkg/sentry/vfs/lock", "//pkg/syserr", "//pkg/syserror", "//pkg/tcpip", diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go index 45e109361..8c32371a2 100644 --- a/pkg/sentry/socket/unix/unix_vfs2.go +++ b/pkg/sentry/socket/unix/unix_vfs2.go @@ -26,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/socket/netstack" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip" @@ -39,6 +40,7 @@ type SocketVFS2 struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl vfs.DentryMetadataFileDescriptionImpl + vfs.LockFD socketOpsCommon } @@ -51,7 +53,7 @@ func NewSockfsFile(t *kernel.Task, ep transport.Endpoint, stype linux.SockType) mnt := t.Kernel().SocketMount() d := sockfs.NewDentry(t.Credentials(), mnt) - fd, err := NewFileDescription(ep, stype, linux.O_RDWR, mnt, d) + fd, err := NewFileDescription(ep, stype, linux.O_RDWR, mnt, d, &lock.FileLocks{}) if err != nil { return nil, syserr.FromError(err) } @@ -60,7 +62,7 @@ func NewSockfsFile(t *kernel.Task, ep transport.Endpoint, stype linux.SockType) // NewFileDescription creates and returns a socket file description // corresponding to the given mount and dentry. -func NewFileDescription(ep transport.Endpoint, stype linux.SockType, flags uint32, mnt *vfs.Mount, d *vfs.Dentry) (*vfs.FileDescription, error) { +func NewFileDescription(ep transport.Endpoint, stype linux.SockType, flags uint32, mnt *vfs.Mount, d *vfs.Dentry, locks *lock.FileLocks) (*vfs.FileDescription, error) { // You can create AF_UNIX, SOCK_RAW sockets. They're the same as // SOCK_DGRAM and don't require CAP_NET_RAW. if stype == linux.SOCK_RAW { @@ -73,6 +75,7 @@ func NewFileDescription(ep transport.Endpoint, stype linux.SockType, flags uint3 stype: stype, }, } + sock.LockFD.Init(locks) vfsfd := &sock.vfsfd if err := vfsfd.Init(sock, flags, mnt, d, &vfs.FileDescriptionOptions{ DenyPRead: true, diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go index 35a98212a..8347617bd 100644 --- a/pkg/sentry/syscalls/linux/sys_file.go +++ b/pkg/sentry/syscalls/linux/sys_file.go @@ -998,9 +998,6 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return 0, nil, err } - // The lock uid is that of the Task's FDTable. - lockUniqueID := lock.UniqueID(t.FDTable().ID()) - // These locks don't block; execute the non-blocking operation using the inode's lock // context directly. switch flock.Type { @@ -1010,12 +1007,12 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } if cmd == linux.F_SETLK { // Non-blocking lock, provide a nil lock.Blocker. - if !file.Dirent.Inode.LockCtx.Posix.LockRegion(lockUniqueID, lock.ReadLock, rng, nil) { + if !file.Dirent.Inode.LockCtx.Posix.LockRegion(t.FDTable(), lock.ReadLock, rng, nil) { return 0, nil, syserror.EAGAIN } } else { // Blocking lock, pass in the task to satisfy the lock.Blocker interface. - if !file.Dirent.Inode.LockCtx.Posix.LockRegion(lockUniqueID, lock.ReadLock, rng, t) { + if !file.Dirent.Inode.LockCtx.Posix.LockRegion(t.FDTable(), lock.ReadLock, rng, t) { return 0, nil, syserror.EINTR } } @@ -1026,18 +1023,18 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } if cmd == linux.F_SETLK { // Non-blocking lock, provide a nil lock.Blocker. - if !file.Dirent.Inode.LockCtx.Posix.LockRegion(lockUniqueID, lock.WriteLock, rng, nil) { + if !file.Dirent.Inode.LockCtx.Posix.LockRegion(t.FDTable(), lock.WriteLock, rng, nil) { return 0, nil, syserror.EAGAIN } } else { // Blocking lock, pass in the task to satisfy the lock.Blocker interface. - if !file.Dirent.Inode.LockCtx.Posix.LockRegion(lockUniqueID, lock.WriteLock, rng, t) { + if !file.Dirent.Inode.LockCtx.Posix.LockRegion(t.FDTable(), lock.WriteLock, rng, t) { return 0, nil, syserror.EINTR } } return 0, nil, nil case linux.F_UNLCK: - file.Dirent.Inode.LockCtx.Posix.UnlockRegion(lockUniqueID, rng) + file.Dirent.Inode.LockCtx.Posix.UnlockRegion(t.FDTable(), rng) return 0, nil, nil default: return 0, nil, syserror.EINVAL @@ -2157,22 +2154,6 @@ func Flock(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall nonblocking := operation&linux.LOCK_NB != 0 operation &^= linux.LOCK_NB - // flock(2): - // Locks created by flock() are associated with an open file table entry. This means that - // duplicate file descriptors (created by, for example, fork(2) or dup(2)) refer to the - // same lock, and this lock may be modified or released using any of these descriptors. Furthermore, - // the lock is released either by an explicit LOCK_UN operation on any of these duplicate - // descriptors, or when all such descriptors have been closed. - // - // If a process uses open(2) (or similar) to obtain more than one descriptor for the same file, - // these descriptors are treated independently by flock(). An attempt to lock the file using - // one of these file descriptors may be denied by a lock that the calling process has already placed via - // another descriptor. - // - // We use the File UniqueID as the lock UniqueID because it needs to reference the same lock across dup(2) - // and fork(2). - lockUniqueID := lock.UniqueID(file.UniqueID) - // A BSD style lock spans the entire file. rng := lock.LockRange{ Start: 0, @@ -2183,29 +2164,29 @@ func Flock(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall case linux.LOCK_EX: if nonblocking { // Since we're nonblocking we pass a nil lock.Blocker implementation. - if !file.Dirent.Inode.LockCtx.BSD.LockRegion(lockUniqueID, lock.WriteLock, rng, nil) { + if !file.Dirent.Inode.LockCtx.BSD.LockRegion(file, lock.WriteLock, rng, nil) { return 0, nil, syserror.EWOULDBLOCK } } else { // Because we're blocking we will pass the task to satisfy the lock.Blocker interface. - if !file.Dirent.Inode.LockCtx.BSD.LockRegion(lockUniqueID, lock.WriteLock, rng, t) { + if !file.Dirent.Inode.LockCtx.BSD.LockRegion(file, lock.WriteLock, rng, t) { return 0, nil, syserror.EINTR } } case linux.LOCK_SH: if nonblocking { // Since we're nonblocking we pass a nil lock.Blocker implementation. - if !file.Dirent.Inode.LockCtx.BSD.LockRegion(lockUniqueID, lock.ReadLock, rng, nil) { + if !file.Dirent.Inode.LockCtx.BSD.LockRegion(file, lock.ReadLock, rng, nil) { return 0, nil, syserror.EWOULDBLOCK } } else { // Because we're blocking we will pass the task to satisfy the lock.Blocker interface. - if !file.Dirent.Inode.LockCtx.BSD.LockRegion(lockUniqueID, lock.ReadLock, rng, t) { + if !file.Dirent.Inode.LockCtx.BSD.LockRegion(file, lock.ReadLock, rng, t) { return 0, nil, syserror.EINTR } } case linux.LOCK_UN: - file.Dirent.Inode.LockCtx.BSD.UnlockRegion(lockUniqueID, rng) + file.Dirent.Inode.LockCtx.BSD.UnlockRegion(file, rng) default: // flock(2): EINVAL operation is invalid. return 0, nil, syserror.EINVAL diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD index c0d005247..9f93f4354 100644 --- a/pkg/sentry/syscalls/linux/vfs2/BUILD +++ b/pkg/sentry/syscalls/linux/vfs2/BUILD @@ -14,6 +14,7 @@ go_library( "getdents.go", "inotify.go", "ioctl.go", + "lock.go", "memfd.go", "mmap.go", "mount.go", @@ -42,6 +43,7 @@ go_library( "//pkg/fspath", "//pkg/gohacks", "//pkg/sentry/arch", + "//pkg/sentry/fs/lock", "//pkg/sentry/fsbridge", "//pkg/sentry/fsimpl/eventfd", "//pkg/sentry/fsimpl/pipefs", diff --git a/pkg/sentry/syscalls/linux/vfs2/lock.go b/pkg/sentry/syscalls/linux/vfs2/lock.go new file mode 100644 index 000000000..bf19028c4 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/lock.go @@ -0,0 +1,64 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fs/lock" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/syserror" +) + +// Flock implements linux syscall flock(2). +func Flock(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + operation := args[1].Int() + + file := t.GetFileVFS2(fd) + if file == nil { + // flock(2): EBADF fd is not an open file descriptor. + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + nonblocking := operation&linux.LOCK_NB != 0 + operation &^= linux.LOCK_NB + + var blocker lock.Blocker + if !nonblocking { + blocker = t + } + + switch operation { + case linux.LOCK_EX: + if err := file.LockBSD(t, lock.WriteLock, blocker); err != nil { + return 0, nil, err + } + case linux.LOCK_SH: + if err := file.LockBSD(t, lock.ReadLock, blocker); err != nil { + return 0, nil, err + } + case linux.LOCK_UN: + if err := file.UnlockBSD(t); err != nil { + return 0, nil, err + } + default: + // flock(2): EINVAL operation is invalid. + return 0, nil, syserror.EINVAL + } + + return 0, nil, nil +} diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2.go b/pkg/sentry/syscalls/linux/vfs2/vfs2.go index 7b6e7571a..954c82f97 100644 --- a/pkg/sentry/syscalls/linux/vfs2/vfs2.go +++ b/pkg/sentry/syscalls/linux/vfs2/vfs2.go @@ -62,7 +62,7 @@ func Override() { s.Table[55] = syscalls.Supported("getsockopt", GetSockOpt) s.Table[59] = syscalls.Supported("execve", Execve) s.Table[72] = syscalls.Supported("fcntl", Fcntl) - delete(s.Table, 73) // flock + s.Table[73] = syscalls.Supported("fcntl", Flock) s.Table[74] = syscalls.Supported("fsync", Fsync) s.Table[75] = syscalls.Supported("fdatasync", Fdatasync) s.Table[76] = syscalls.Supported("truncate", Truncate) diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index 774cc66cc..16d9f3a28 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -72,6 +72,7 @@ go_library( "//pkg/sentry/memmap", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/uniqueid", + "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserror", "//pkg/usermem", diff --git a/pkg/sentry/vfs/epoll.go b/pkg/sentry/vfs/epoll.go index 8297f964b..599c3131c 100644 --- a/pkg/sentry/vfs/epoll.go +++ b/pkg/sentry/vfs/epoll.go @@ -31,6 +31,7 @@ type EpollInstance struct { vfsfd FileDescription FileDescriptionDefaultImpl DentryMetadataFileDescriptionImpl + NoLockFD // q holds waiters on this EpollInstance. q waiter.Queue diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index bb294563d..97b9b18d7 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -73,6 +73,8 @@ type FileDescription struct { // writable is analogous to Linux's FMODE_WRITE. writable bool + usedLockBSD uint32 + // impl is the FileDescriptionImpl associated with this Filesystem. impl is // immutable. This should be the last field in FileDescription. impl FileDescriptionImpl @@ -175,6 +177,12 @@ func (fd *FileDescription) DecRef() { } ep.interestMu.Unlock() } + + // If BSD locks were used, release any lock that it may have acquired. + if atomic.LoadUint32(&fd.usedLockBSD) != 0 { + fd.impl.UnlockBSD(context.Background(), fd) + } + // Release implementation resources. fd.impl.Release() if fd.writable { @@ -420,13 +428,9 @@ type FileDescriptionImpl interface { Removexattr(ctx context.Context, name string) error // LockBSD tries to acquire a BSD-style advisory file lock. - // - // TODO(gvisor.dev/issue/1480): BSD-style file locking LockBSD(ctx context.Context, uid lock.UniqueID, t lock.LockType, block lock.Blocker) error - // LockBSD releases a BSD-style advisory file lock. - // - // TODO(gvisor.dev/issue/1480): BSD-style file locking + // UnlockBSD releases a BSD-style advisory file lock. UnlockBSD(ctx context.Context, uid lock.UniqueID) error // LockPOSIX tries to acquire a POSIX-style advisory file lock. @@ -736,3 +740,14 @@ func (fd *FileDescription) InodeID() uint64 { func (fd *FileDescription) Msync(ctx context.Context, mr memmap.MappableRange) error { return fd.Sync(ctx) } + +// LockBSD tries to acquire a BSD-style advisory file lock. +func (fd *FileDescription) LockBSD(ctx context.Context, lockType lock.LockType, blocker lock.Blocker) error { + atomic.StoreUint32(&fd.usedLockBSD, 1) + return fd.impl.LockBSD(ctx, fd, lockType, blocker) +} + +// UnlockBSD releases a BSD-style advisory file lock. +func (fd *FileDescription) UnlockBSD(ctx context.Context) error { + return fd.impl.UnlockBSD(ctx, fd) +} diff --git a/pkg/sentry/vfs/file_description_impl_util.go b/pkg/sentry/vfs/file_description_impl_util.go index f4c111926..af7213dfd 100644 --- a/pkg/sentry/vfs/file_description_impl_util.go +++ b/pkg/sentry/vfs/file_description_impl_util.go @@ -21,8 +21,9 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/fs/lock" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/memmap" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -153,26 +154,6 @@ func (FileDescriptionDefaultImpl) Removexattr(ctx context.Context, name string) return syserror.ENOTSUP } -// LockBSD implements FileDescriptionImpl.LockBSD. -func (FileDescriptionDefaultImpl) LockBSD(ctx context.Context, uid lock.UniqueID, t lock.LockType, block lock.Blocker) error { - return syserror.EBADF -} - -// UnlockBSD implements FileDescriptionImpl.UnlockBSD. -func (FileDescriptionDefaultImpl) UnlockBSD(ctx context.Context, uid lock.UniqueID) error { - return syserror.EBADF -} - -// LockPOSIX implements FileDescriptionImpl.LockPOSIX. -func (FileDescriptionDefaultImpl) LockPOSIX(ctx context.Context, uid lock.UniqueID, t lock.LockType, rng lock.LockRange, block lock.Blocker) error { - return syserror.EBADF -} - -// UnlockPOSIX implements FileDescriptionImpl.UnlockPOSIX. -func (FileDescriptionDefaultImpl) UnlockPOSIX(ctx context.Context, uid lock.UniqueID, rng lock.LockRange) error { - return syserror.EBADF -} - // DirectoryFileDescriptionDefaultImpl may be embedded by implementations of // FileDescriptionImpl that always represent directories to obtain // implementations of non-directory I/O methods that return EISDIR. @@ -384,3 +365,60 @@ func GenericConfigureMMap(fd *FileDescription, m memmap.Mappable, opts *memmap.M fd.IncRef() return nil } + +// LockFD may be used by most implementations of FileDescriptionImpl.Lock* +// functions. Caller must call Init(). +type LockFD struct { + locks *lock.FileLocks +} + +// Init initializes fd with FileLocks to use. +func (fd *LockFD) Init(locks *lock.FileLocks) { + fd.locks = locks +} + +// LockBSD implements vfs.FileDescriptionImpl.LockBSD. +func (fd *LockFD) LockBSD(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, block fslock.Blocker) error { + return fd.locks.LockBSD(uid, t, block) +} + +// UnlockBSD implements vfs.FileDescriptionImpl.UnlockBSD. +func (fd *LockFD) UnlockBSD(ctx context.Context, uid fslock.UniqueID) error { + fd.locks.UnlockBSD(uid) + return nil +} + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (fd *LockFD) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, rng fslock.LockRange, block fslock.Blocker) error { + return fd.locks.LockPOSIX(uid, t, rng, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (fd *LockFD) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, rng fslock.LockRange) error { + fd.locks.UnlockPOSIX(uid, rng) + return nil +} + +// NoLockFD implements Lock*/Unlock* portion of FileDescriptionImpl interface +// returning ENOLCK. +type NoLockFD struct{} + +// LockBSD implements vfs.FileDescriptionImpl.LockBSD. +func (NoLockFD) LockBSD(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, block fslock.Blocker) error { + return syserror.ENOLCK +} + +// UnlockBSD implements vfs.FileDescriptionImpl.UnlockBSD. +func (NoLockFD) UnlockBSD(ctx context.Context, uid fslock.UniqueID) error { + return syserror.ENOLCK +} + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (NoLockFD) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, rng fslock.LockRange, block fslock.Blocker) error { + return syserror.ENOLCK +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (NoLockFD) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, rng fslock.LockRange) error { + return syserror.ENOLCK +} diff --git a/pkg/sentry/vfs/file_description_impl_util_test.go b/pkg/sentry/vfs/file_description_impl_util_test.go index 3a75d4d62..5061f6ac9 100644 --- a/pkg/sentry/vfs/file_description_impl_util_test.go +++ b/pkg/sentry/vfs/file_description_impl_util_test.go @@ -33,6 +33,7 @@ import ( type fileDescription struct { vfsfd FileDescription FileDescriptionDefaultImpl + NoLockFD } // genCount contains the number of times its DynamicBytesSource.Generate() diff --git a/pkg/sentry/vfs/inotify.go b/pkg/sentry/vfs/inotify.go index 05a3051a4..7fa7d2d0c 100644 --- a/pkg/sentry/vfs/inotify.go +++ b/pkg/sentry/vfs/inotify.go @@ -57,6 +57,7 @@ type Inotify struct { vfsfd FileDescription FileDescriptionDefaultImpl DentryMetadataFileDescriptionImpl + NoLockFD // Unique identifier for this inotify instance. We don't just reuse the // inotify fd because fds can be duped. These should not be exposed to the diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index ae2aa44dc..4a1486e14 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -802,10 +802,13 @@ cc_binary( ], linkstatic = 1, deps = [ + ":socket_test_util", "//test/util:file_descriptor", "@com_google_absl//absl/strings", "@com_google_absl//absl/time", gtest, + "//test/util:epoll_util", + "//test/util:eventfd_util", "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_main", diff --git a/test/syscalls/linux/flock.cc b/test/syscalls/linux/flock.cc index 3ecb8db8e..638a93979 100644 --- a/test/syscalls/linux/flock.cc +++ b/test/syscalls/linux/flock.cc @@ -21,6 +21,7 @@ #include "absl/time/clock.h" #include "absl/time/time.h" #include "test/syscalls/linux/file_base.h" +#include "test/syscalls/linux/socket_test_util.h" #include "test/util/file_descriptor.h" #include "test/util/temp_path.h" #include "test/util/test_util.h" @@ -34,11 +35,6 @@ namespace { class FlockTest : public FileTest {}; -TEST_F(FlockTest, BadFD) { - // EBADF: fd is not an open file descriptor. - ASSERT_THAT(flock(-1, 0), SyscallFailsWithErrno(EBADF)); -} - TEST_F(FlockTest, InvalidOpCombinations) { // The operation cannot be both exclusive and shared. EXPECT_THAT(flock(test_file_fd_.get(), LOCK_EX | LOCK_SH | LOCK_NB), @@ -57,15 +53,6 @@ TEST_F(FlockTest, NoOperationSpecified) { SyscallFailsWithErrno(EINVAL)); } -TEST(FlockTestNoFixture, FlockSupportsPipes) { - int fds[2]; - ASSERT_THAT(pipe(fds), SyscallSucceeds()); - - EXPECT_THAT(flock(fds[0], LOCK_EX | LOCK_NB), SyscallSucceeds()); - EXPECT_THAT(close(fds[0]), SyscallSucceeds()); - EXPECT_THAT(close(fds[1]), SyscallSucceeds()); -} - TEST_F(FlockTest, TestSimpleExLock) { // Test that we can obtain an exclusive lock (no other holders) // and that we can unlock it. @@ -583,6 +570,66 @@ TEST_F(FlockTest, BlockingLockFirstExclusiveSecondExclusive_NoRandomSave) { EXPECT_THAT(flock(test_file_fd_.get(), LOCK_UN), SyscallSucceeds()); } +TEST(FlockTestNoFixture, BadFD) { + // EBADF: fd is not an open file descriptor. + ASSERT_THAT(flock(-1, 0), SyscallFailsWithErrno(EBADF)); +} + +TEST(FlockTestNoFixture, FlockDir) { + auto dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + auto fd = ASSERT_NO_ERRNO_AND_VALUE(Open(dir.path(), O_RDONLY, 0000)); + EXPECT_THAT(flock(fd.get(), LOCK_EX | LOCK_NB), SyscallSucceeds()); +} + +TEST(FlockTestNoFixture, FlockSymlink) { + // TODO(gvisor.dev/issue/2782): Replace with IsRunningWithVFS1() when O_PATH + // is supported. + SKIP_IF(IsRunningOnGvisor()); + + auto file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); + auto symlink = ASSERT_NO_ERRNO_AND_VALUE( + TempPath::CreateSymlinkTo(GetAbsoluteTestTmpdir(), file.path())); + + auto fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(symlink.path(), O_RDONLY | O_PATH, 0000)); + EXPECT_THAT(flock(fd.get(), LOCK_EX | LOCK_NB), SyscallFailsWithErrno(EBADF)); +} + +TEST(FlockTestNoFixture, FlockProc) { + auto fd = + ASSERT_NO_ERRNO_AND_VALUE(Open("/proc/self/status", O_RDONLY, 0000)); + EXPECT_THAT(flock(fd.get(), LOCK_EX | LOCK_NB), SyscallSucceeds()); +} + +TEST(FlockTestNoFixture, FlockPipe) { + int fds[2]; + ASSERT_THAT(pipe(fds), SyscallSucceeds()); + + EXPECT_THAT(flock(fds[0], LOCK_EX | LOCK_NB), SyscallSucceeds()); + // Check that the pipe was locked above. + EXPECT_THAT(flock(fds[1], LOCK_EX | LOCK_NB), SyscallFailsWithErrno(EAGAIN)); + + EXPECT_THAT(flock(fds[0], LOCK_UN), SyscallSucceeds()); + EXPECT_THAT(flock(fds[1], LOCK_EX | LOCK_NB), SyscallSucceeds()); + + EXPECT_THAT(close(fds[0]), SyscallSucceeds()); + EXPECT_THAT(close(fds[1]), SyscallSucceeds()); +} + +TEST(FlockTestNoFixture, FlockSocket) { + int sock = socket(AF_UNIX, SOCK_STREAM, 0); + ASSERT_THAT(sock, SyscallSucceeds()); + + struct sockaddr_un addr = + ASSERT_NO_ERRNO_AND_VALUE(UniqueUnixAddr(true /* abstract */, AF_UNIX)); + ASSERT_THAT( + bind(sock, reinterpret_cast(&addr), sizeof(addr)), + SyscallSucceeds()); + + EXPECT_THAT(flock(sock, LOCK_EX | LOCK_NB), SyscallSucceeds()); + EXPECT_THAT(close(sock), SyscallSucceeds()); +} + } // namespace } // namespace testing -- cgit v1.2.3 From f1f85f475d5f98ad9e9d574a11997c0829c3b189 Mon Sep 17 00:00:00 2001 From: Gaurav Singh Date: Fri, 22 May 2020 23:45:21 -0400 Subject: sentry: use defer wg.Done() unconditionally Signed-off-by: Gaurav Singh --- pkg/sentry/kernel/timekeeper.go | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/timekeeper.go b/pkg/sentry/kernel/timekeeper.go index da0ea7bb5..0adf25691 100644 --- a/pkg/sentry/kernel/timekeeper.go +++ b/pkg/sentry/kernel/timekeeper.go @@ -186,6 +186,7 @@ func (t *Timekeeper) startUpdater() { timer := time.NewTicker(sentrytime.ApproxUpdateInterval) t.wg.Add(1) go func() { // S/R-SAFE: stopped during save. + defer t.wg.Done() for { // Start with an update immediately, so the clocks are // ready ASAP. @@ -220,7 +221,6 @@ func (t *Timekeeper) startUpdater() { select { case <-timer.C: case <-t.stop: - t.wg.Done() return } } -- cgit v1.2.3 From 203dc121f6fb3122e3bc0f234f4b4e4b276fb3b0 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Wed, 10 Jun 2020 06:50:10 -0700 Subject: Redirect TODOs to more specific issues Closes #1623 PiperOrigin-RevId: 315681993 --- pkg/sentry/loader/vdso.go | 2 +- pkg/sentry/syscalls/linux/vfs2/fd.go | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/loader/vdso.go b/pkg/sentry/loader/vdso.go index 00977fc08..165869028 100644 --- a/pkg/sentry/loader/vdso.go +++ b/pkg/sentry/loader/vdso.go @@ -75,7 +75,7 @@ var _ fs.FileOperations = (*byteReader)(nil) // newByteReaderFile creates a fake file to read data from. // -// TODO(gvisor.dev/issue/1623): Convert to VFS2. +// TODO(gvisor.dev/issue/2921): Convert to VFS2. func newByteReaderFile(ctx context.Context, data []byte) *fs.File { // Create a fake inode. inode := fs.NewInode( diff --git a/pkg/sentry/syscalls/linux/vfs2/fd.go b/pkg/sentry/syscalls/linux/vfs2/fd.go index ca0f7fd1e..6006758a5 100644 --- a/pkg/sentry/syscalls/linux/vfs2/fd.go +++ b/pkg/sentry/syscalls/linux/vfs2/fd.go @@ -168,7 +168,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall err := tmpfs.AddSeals(file, args[2].Uint()) return 0, nil, err default: - // TODO(gvisor.dev/issue/1623): Everything else is not yet supported. + // TODO(gvisor.dev/issue/2920): Everything else is not yet supported. return 0, nil, syserror.EINVAL } } -- cgit v1.2.3 From a5a4f804879f8d1b5e6de6005aef6d3e14e7dca2 Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Wed, 10 Jun 2020 13:25:17 -0700 Subject: socket/unix: handle sendto address argument for connected sockets In case of SOCK_SEQPACKET, it has to be ignored. In case of SOCK_STREAM, EISCONN or EOPNOTSUPP has to be returned. PiperOrigin-RevId: 315755972 --- pkg/sentry/socket/unix/unix.go | 26 ++++++++++++++++++-------- test/syscalls/linux/socket_unix_seqpacket.cc | 18 ++++++++++++++++++ test/syscalls/linux/socket_unix_stream.cc | 14 ++++++++++++++ 3 files changed, 50 insertions(+), 8 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go index c4c9db81b..4bb2b6ff4 100644 --- a/pkg/sentry/socket/unix/unix.go +++ b/pkg/sentry/socket/unix/unix.go @@ -459,15 +459,25 @@ func (s *socketOpsCommon) SendMsg(t *kernel.Task, src usermem.IOSequence, to []b To: nil, } if len(to) > 0 { - ep, err := extractEndpoint(t, to) - if err != nil { - return 0, err - } - defer ep.Release() - w.To = ep + switch s.stype { + case linux.SOCK_SEQPACKET: + to = nil + case linux.SOCK_STREAM: + if s.State() == linux.SS_CONNECTED { + return 0, syserr.ErrAlreadyConnected + } + return 0, syserr.ErrNotSupported + default: + ep, err := extractEndpoint(t, to) + if err != nil { + return 0, err + } + defer ep.Release() + w.To = ep - if ep.Passcred() && w.Control.Credentials == nil { - w.Control.Credentials = control.MakeCreds(t) + if ep.Passcred() && w.Control.Credentials == nil { + w.Control.Credentials = control.MakeCreds(t) + } } } diff --git a/test/syscalls/linux/socket_unix_seqpacket.cc b/test/syscalls/linux/socket_unix_seqpacket.cc index 84d3a569e..6d03df4d9 100644 --- a/test/syscalls/linux/socket_unix_seqpacket.cc +++ b/test/syscalls/linux/socket_unix_seqpacket.cc @@ -43,6 +43,24 @@ TEST_P(SeqpacketUnixSocketPairTest, ReadOneSideClosed) { SyscallSucceedsWithValue(0)); } +TEST_P(SeqpacketUnixSocketPairTest, Sendto) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + struct sockaddr_un addr = {}; + addr.sun_family = AF_UNIX; + constexpr char kPath[] = "\0nonexistent"; + memcpy(addr.sun_path, kPath, sizeof(kPath)); + + constexpr char kStr[] = "abc"; + ASSERT_THAT(sendto(sockets->second_fd(), kStr, 3, 0, (struct sockaddr*)&addr, + sizeof(addr)), + SyscallSucceedsWithValue(3)); + + char data[10] = {}; + ASSERT_THAT(read(sockets->first_fd(), data, sizeof(data)), + SyscallSucceedsWithValue(3)); +} + } // namespace } // namespace testing diff --git a/test/syscalls/linux/socket_unix_stream.cc b/test/syscalls/linux/socket_unix_stream.cc index 563467365..99e77b89e 100644 --- a/test/syscalls/linux/socket_unix_stream.cc +++ b/test/syscalls/linux/socket_unix_stream.cc @@ -89,6 +89,20 @@ TEST_P(StreamUnixSocketPairTest, ReadOneSideClosedWithUnreadData) { SyscallFailsWithErrno(ECONNRESET)); } +TEST_P(StreamUnixSocketPairTest, Sendto) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + struct sockaddr_un addr = {}; + addr.sun_family = AF_UNIX; + constexpr char kPath[] = "\0nonexistent"; + memcpy(addr.sun_path, kPath, sizeof(kPath)); + + constexpr char kStr[] = "abc"; + ASSERT_THAT(sendto(sockets->second_fd(), kStr, 3, 0, (struct sockaddr*)&addr, + sizeof(addr)), + SyscallFailsWithErrno(EISCONN)); +} + INSTANTIATE_TEST_SUITE_P( AllUnixDomainSockets, StreamUnixSocketPairTest, ::testing::ValuesIn(IncludeReversals(VecCat( -- cgit v1.2.3 From 4b9652d63b319414e764696f1b77ee39cd36d96d Mon Sep 17 00:00:00 2001 From: Nayana Bidari Date: Wed, 10 Jun 2020 13:36:02 -0700 Subject: {S,G}etsockopt for TCP_KEEPCNT option. TCP_KEEPCNT is used to set the maximum keepalive probes to be sent before dropping the connection. WANT_LGTM=jchacon PiperOrigin-RevId: 315758094 --- pkg/abi/linux/tcp.go | 1 + pkg/sentry/socket/netstack/netstack.go | 35 ++++++++----- test/syscalls/linux/socket_ip_tcp_generic.cc | 73 ++++++++++++++++++++++++++++ 3 files changed, 98 insertions(+), 11 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/tcp.go b/pkg/abi/linux/tcp.go index 174d470e2..2a8d4708b 100644 --- a/pkg/abi/linux/tcp.go +++ b/pkg/abi/linux/tcp.go @@ -57,4 +57,5 @@ const ( const ( MAX_TCP_KEEPIDLE = 32767 MAX_TCP_KEEPINTVL = 32767 + MAX_TCP_KEEPCNT = 127 ) diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index e1e0c5931..738277391 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -1246,6 +1246,18 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa return int32(time.Duration(v) / time.Second), nil + case linux.TCP_KEEPCNT: + if outLen < sizeOfInt32 { + return nil, syserr.ErrInvalidArgument + } + + v, err := ep.GetSockOptInt(tcpip.KeepaliveCountOption) + if err != nil { + return nil, syserr.TranslateNetstackError(err) + } + + return int32(v), nil + case linux.TCP_USER_TIMEOUT: if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument @@ -1786,6 +1798,17 @@ func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) * } return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.KeepaliveIntervalOption(time.Second * time.Duration(v)))) + case linux.TCP_KEEPCNT: + if len(optVal) < sizeOfInt32 { + return syserr.ErrInvalidArgument + } + + v := usermem.ByteOrder.Uint32(optVal) + if v < 1 || v > linux.MAX_TCP_KEEPCNT { + return syserr.ErrInvalidArgument + } + return syserr.TranslateNetstackError(ep.SetSockOptInt(tcpip.KeepaliveCountOption, int(v))) + case linux.TCP_USER_TIMEOUT: if len(optVal) < sizeOfInt32 { return syserr.ErrInvalidArgument @@ -2115,30 +2138,20 @@ func emitUnimplementedEventTCP(t *kernel.Task, name int) { switch name { case linux.TCP_CONGESTION, linux.TCP_CORK, - linux.TCP_DEFER_ACCEPT, linux.TCP_FASTOPEN, linux.TCP_FASTOPEN_CONNECT, linux.TCP_FASTOPEN_KEY, linux.TCP_FASTOPEN_NO_COOKIE, - linux.TCP_KEEPCNT, - linux.TCP_KEEPIDLE, - linux.TCP_KEEPINTVL, - linux.TCP_LINGER2, - linux.TCP_MAXSEG, linux.TCP_QUEUE_SEQ, - linux.TCP_QUICKACK, linux.TCP_REPAIR, linux.TCP_REPAIR_QUEUE, linux.TCP_REPAIR_WINDOW, linux.TCP_SAVED_SYN, linux.TCP_SAVE_SYN, - linux.TCP_SYNCNT, linux.TCP_THIN_DUPACK, linux.TCP_THIN_LINEAR_TIMEOUTS, linux.TCP_TIMESTAMP, - linux.TCP_ULP, - linux.TCP_USER_TIMEOUT, - linux.TCP_WINDOW_CLAMP: + linux.TCP_ULP: t.Kernel().EmitUnimplementedEvent(t) } diff --git a/test/syscalls/linux/socket_ip_tcp_generic.cc b/test/syscalls/linux/socket_ip_tcp_generic.cc index fa81845fd..15adc8d0e 100644 --- a/test/syscalls/linux/socket_ip_tcp_generic.cc +++ b/test/syscalls/linux/socket_ip_tcp_generic.cc @@ -524,6 +524,7 @@ TEST_P(TCPSocketPairTest, SetTCPKeepintvlZero) { // Copied from include/net/tcp.h. constexpr int MAX_TCP_KEEPIDLE = 32767; constexpr int MAX_TCP_KEEPINTVL = 32767; +constexpr int MAX_TCP_KEEPCNT = 127; TEST_P(TCPSocketPairTest, SetTCPKeepidleAboveMax) { auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); @@ -575,6 +576,78 @@ TEST_P(TCPSocketPairTest, SetTCPKeepintvlToMax) { EXPECT_EQ(get, MAX_TCP_KEEPINTVL); } +TEST_P(TCPSocketPairTest, TCPKeepcountDefault) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + int get = -1; + socklen_t get_len = sizeof(get); + EXPECT_THAT( + getsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_KEEPCNT, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, 9); // 9 keepalive probes. +} + +TEST_P(TCPSocketPairTest, SetTCPKeepcountZero) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + constexpr int kZero = 0; + EXPECT_THAT(setsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_KEEPCNT, &kZero, + sizeof(kZero)), + SyscallFailsWithErrno(EINVAL)); +} + +TEST_P(TCPSocketPairTest, SetTCPKeepcountAboveMax) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + constexpr int kAboveMax = MAX_TCP_KEEPCNT + 1; + EXPECT_THAT(setsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_KEEPCNT, + &kAboveMax, sizeof(kAboveMax)), + SyscallFailsWithErrno(EINVAL)); +} + +TEST_P(TCPSocketPairTest, SetTCPKeepcountToMax) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + EXPECT_THAT(setsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_KEEPCNT, + &MAX_TCP_KEEPCNT, sizeof(MAX_TCP_KEEPCNT)), + SyscallSucceedsWithValue(0)); + + int get = -1; + socklen_t get_len = sizeof(get); + EXPECT_THAT( + getsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_KEEPCNT, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, MAX_TCP_KEEPCNT); +} + +TEST_P(TCPSocketPairTest, SetTCPKeepcountToOne) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + int keepaliveCount = 1; + EXPECT_THAT(setsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_KEEPCNT, + &keepaliveCount, sizeof(keepaliveCount)), + SyscallSucceedsWithValue(0)); + + int get = -1; + socklen_t get_len = sizeof(get); + EXPECT_THAT( + getsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_KEEPCNT, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, keepaliveCount); +} + +TEST_P(TCPSocketPairTest, SetTCPKeepcountToNegative) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + int keepaliveCount = -5; + EXPECT_THAT(setsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_KEEPCNT, + &keepaliveCount, sizeof(keepaliveCount)), + SyscallFailsWithErrno(EINVAL)); +} + TEST_P(TCPSocketPairTest, SetOOBInline) { auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); -- cgit v1.2.3 From cbfb537eef7c531b596c3aecd947ee6920bea2d7 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Tue, 2 Jun 2020 22:55:31 -0400 Subject: Deleting the maxSendBufferSize from fs/host When I do high-performance networking, the value of wmem_max is often set very high, specially for 10/25/50 Gigabit NIC. I think maybe this restriction is not suitable. Signed-off-by: Bin Lu --- pkg/sentry/fs/host/socket.go | 10 ---------- 1 file changed, 10 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/host/socket.go b/pkg/sentry/fs/host/socket.go index b6e94583e..cfb089e43 100644 --- a/pkg/sentry/fs/host/socket.go +++ b/pkg/sentry/fs/host/socket.go @@ -22,7 +22,6 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fd" "gvisor.dev/gvisor/pkg/fdnotifier" - "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/socket/control" @@ -39,11 +38,6 @@ import ( // LINT.IfChange -// maxSendBufferSize is the maximum host send buffer size allowed for endpoint. -// -// N.B. 8MB is the default maximum on Linux (2 * sysctl_wmem_max). -const maxSendBufferSize = 8 << 20 - // ConnectedEndpoint is a host FD backed implementation of // transport.ConnectedEndpoint and transport.Receiver. // @@ -103,10 +97,6 @@ func (c *ConnectedEndpoint) init() *syserr.Error { if err != nil { return syserr.FromError(err) } - if sndbuf > maxSendBufferSize { - log.Warningf("Socket send buffer too large: %d", sndbuf) - return syserr.ErrInvalidEndpointState - } c.stype = linux.SockType(stype) c.sndbuf = int64(sndbuf) -- cgit v1.2.3 From a87c74bc548b1eebc9a118fcc192d906b9fb2e39 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Wed, 10 Jun 2020 23:21:35 -0700 Subject: Remove duplicate colon from warning log. doAction()->log.TracebackAll() will append a colon. PiperOrigin-RevId: 315842611 --- pkg/sentry/watchdog/watchdog.go | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/watchdog/watchdog.go b/pkg/sentry/watchdog/watchdog.go index e2894f9f5..748273366 100644 --- a/pkg/sentry/watchdog/watchdog.go +++ b/pkg/sentry/watchdog/watchdog.go @@ -329,7 +329,7 @@ func (w *Watchdog) report(offenders map[*kernel.Task]*offender, newTaskFound boo func (w *Watchdog) reportStuckWatchdog() { var buf bytes.Buffer - buf.WriteString("Watchdog goroutine is stuck:") + buf.WriteString("Watchdog goroutine is stuck") w.doAction(w.TaskTimeoutAction, false, &buf) } -- cgit v1.2.3 From d58d57606a460d49b8870d2c48cb75f662f65fda Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Thu, 11 Jun 2020 14:55:04 -0700 Subject: Don't copy structs with sync.Mutex during initialization During inititalization inode struct was copied around, but it isn't great pratice to copy it around since it contains ref count and sync.Mutex. Updates #1480 PiperOrigin-RevId: 315983788 --- pkg/sentry/fsimpl/ext/block_map_file.go | 9 +++++---- pkg/sentry/fsimpl/ext/block_map_test.go | 29 +++++++++++++---------------- pkg/sentry/fsimpl/ext/directory.go | 11 +++++------ pkg/sentry/fsimpl/ext/extent_file.go | 5 +++-- pkg/sentry/fsimpl/ext/extent_test.go | 22 ++++++++++------------ pkg/sentry/fsimpl/ext/inode.go | 23 +++++++++++++++++++---- pkg/sentry/fsimpl/ext/regular_file.go | 17 ++++------------- pkg/sentry/fsimpl/ext/symlink.go | 13 ++++++------- 8 files changed, 65 insertions(+), 64 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/ext/block_map_file.go b/pkg/sentry/fsimpl/ext/block_map_file.go index a2d8c3ad6..8bb104ff0 100644 --- a/pkg/sentry/fsimpl/ext/block_map_file.go +++ b/pkg/sentry/fsimpl/ext/block_map_file.go @@ -58,15 +58,16 @@ var _ io.ReaderAt = (*blockMapFile)(nil) // newBlockMapFile is the blockMapFile constructor. It initializes the file to // physical blocks map with (at most) the first 12 (direct) blocks. -func newBlockMapFile(regFile regularFile) (*blockMapFile, error) { - file := &blockMapFile{regFile: regFile} +func newBlockMapFile(args inodeArgs) (*blockMapFile, error) { + file := &blockMapFile{} file.regFile.impl = file + file.regFile.inode.init(args, &file.regFile) for i := uint(0); i < 4; i++ { - file.coverage[i] = getCoverage(regFile.inode.blkSize, i) + file.coverage[i] = getCoverage(file.regFile.inode.blkSize, i) } - blkMap := regFile.inode.diskInode.Data() + blkMap := file.regFile.inode.diskInode.Data() binary.Unmarshal(blkMap[:numDirectBlks*4], binary.LittleEndian, &file.directBlks) binary.Unmarshal(blkMap[numDirectBlks*4:(numDirectBlks+1)*4], binary.LittleEndian, &file.indirectBlk) binary.Unmarshal(blkMap[(numDirectBlks+1)*4:(numDirectBlks+2)*4], binary.LittleEndian, &file.doubleIndirectBlk) diff --git a/pkg/sentry/fsimpl/ext/block_map_test.go b/pkg/sentry/fsimpl/ext/block_map_test.go index 181727ef7..6fa84e7aa 100644 --- a/pkg/sentry/fsimpl/ext/block_map_test.go +++ b/pkg/sentry/fsimpl/ext/block_map_test.go @@ -85,20 +85,6 @@ func (n *blkNumGen) next() uint32 { // the inode covers and that is written to disk. func blockMapSetUp(t *testing.T) (*blockMapFile, []byte) { mockDisk := make([]byte, mockBMDiskSize) - regFile := regularFile{ - inode: inode{ - fs: &filesystem{ - dev: bytes.NewReader(mockDisk), - }, - diskInode: &disklayout.InodeNew{ - InodeOld: disklayout.InodeOld{ - SizeLo: getMockBMFileFize(), - }, - }, - blkSize: uint64(mockBMBlkSize), - }, - } - var fileData []byte blkNums := newBlkNumGen() var data []byte @@ -125,9 +111,20 @@ func blockMapSetUp(t *testing.T) (*blockMapFile, []byte) { data = binary.Marshal(data, binary.LittleEndian, triplyIndirectBlk) fileData = append(fileData, writeFileDataToBlock(mockDisk, triplyIndirectBlk, 3, blkNums)...) - copy(regFile.inode.diskInode.Data(), data) + args := inodeArgs{ + fs: &filesystem{ + dev: bytes.NewReader(mockDisk), + }, + diskInode: &disklayout.InodeNew{ + InodeOld: disklayout.InodeOld{ + SizeLo: getMockBMFileFize(), + }, + }, + blkSize: uint64(mockBMBlkSize), + } + copy(args.diskInode.Data(), data) - mockFile, err := newBlockMapFile(regFile) + mockFile, err := newBlockMapFile(args) if err != nil { t.Fatalf("newBlockMapFile failed: %v", err) } diff --git a/pkg/sentry/fsimpl/ext/directory.go b/pkg/sentry/fsimpl/ext/directory.go index 12b875c8f..43be6928a 100644 --- a/pkg/sentry/fsimpl/ext/directory.go +++ b/pkg/sentry/fsimpl/ext/directory.go @@ -54,16 +54,15 @@ type directory struct { } // newDirectory is the directory constructor. -func newDirectory(inode inode, newDirent bool) (*directory, error) { +func newDirectory(args inodeArgs, newDirent bool) (*directory, error) { file := &directory{ - inode: inode, childCache: make(map[string]*dentry), childMap: make(map[string]*dirent), } - file.inode.impl = file + file.inode.init(args, file) // Initialize childList by reading dirents from the underlying file. - if inode.diskInode.Flags().Index { + if args.diskInode.Flags().Index { // TODO(b/134676337): Support hash tree directories. Currently only the '.' // and '..' entries are read in. @@ -74,7 +73,7 @@ func newDirectory(inode inode, newDirent bool) (*directory, error) { // The dirents are organized in a linear array in the file data. // Extract the file data and decode the dirents. - regFile, err := newRegularFile(inode) + regFile, err := newRegularFile(args) if err != nil { return nil, err } @@ -82,7 +81,7 @@ func newDirectory(inode inode, newDirent bool) (*directory, error) { // buf is used as scratch space for reading in dirents from disk and // unmarshalling them into dirent structs. buf := make([]byte, disklayout.DirentSize) - size := inode.diskInode.Size() + size := args.diskInode.Size() for off, inc := uint64(0), uint64(0); off < size; off += inc { toRead := size - off if toRead > disklayout.DirentSize { diff --git a/pkg/sentry/fsimpl/ext/extent_file.go b/pkg/sentry/fsimpl/ext/extent_file.go index 11dcc0346..c36225a7c 100644 --- a/pkg/sentry/fsimpl/ext/extent_file.go +++ b/pkg/sentry/fsimpl/ext/extent_file.go @@ -38,9 +38,10 @@ var _ io.ReaderAt = (*extentFile)(nil) // newExtentFile is the extent file constructor. It reads the entire extent // tree into memory. // TODO(b/134676337): Build extent tree on demand to reduce memory usage. -func newExtentFile(regFile regularFile) (*extentFile, error) { - file := &extentFile{regFile: regFile} +func newExtentFile(args inodeArgs) (*extentFile, error) { + file := &extentFile{} file.regFile.impl = file + file.regFile.inode.init(args, &file.regFile) err := file.buildExtTree() if err != nil { return nil, err diff --git a/pkg/sentry/fsimpl/ext/extent_test.go b/pkg/sentry/fsimpl/ext/extent_test.go index a2382daa3..cd10d46ee 100644 --- a/pkg/sentry/fsimpl/ext/extent_test.go +++ b/pkg/sentry/fsimpl/ext/extent_test.go @@ -177,21 +177,19 @@ func extentTreeSetUp(t *testing.T, root *disklayout.ExtentNode) (*extentFile, [] t.Helper() mockDisk := make([]byte, mockExtentBlkSize*10) - mockExtentFile := &extentFile{ - regFile: regularFile{ - inode: inode{ - fs: &filesystem{ - dev: bytes.NewReader(mockDisk), - }, - diskInode: &disklayout.InodeNew{ - InodeOld: disklayout.InodeOld{ - SizeLo: uint32(mockExtentBlkSize) * getNumPhyBlks(root), - }, - }, - blkSize: mockExtentBlkSize, + mockExtentFile := &extentFile{} + args := inodeArgs{ + fs: &filesystem{ + dev: bytes.NewReader(mockDisk), + }, + diskInode: &disklayout.InodeNew{ + InodeOld: disklayout.InodeOld{ + SizeLo: uint32(mockExtentBlkSize) * getNumPhyBlks(root), }, }, + blkSize: mockExtentBlkSize, } + mockExtentFile.regFile.inode.init(args, &mockExtentFile.regFile) fileData := writeTree(&mockExtentFile.regFile.inode, mockDisk, node0, mockExtentBlkSize) diff --git a/pkg/sentry/fsimpl/ext/inode.go b/pkg/sentry/fsimpl/ext/inode.go index e4b434b13..5caaf14ed 100644 --- a/pkg/sentry/fsimpl/ext/inode.go +++ b/pkg/sentry/fsimpl/ext/inode.go @@ -118,7 +118,7 @@ func newInode(fs *filesystem, inodeNum uint32) (*inode, error) { } // Build the inode based on its type. - inode := inode{ + args := inodeArgs{ fs: fs, inodeNum: inodeNum, blkSize: blkSize, @@ -127,19 +127,19 @@ func newInode(fs *filesystem, inodeNum uint32) (*inode, error) { switch diskInode.Mode().FileType() { case linux.ModeSymlink: - f, err := newSymlink(inode) + f, err := newSymlink(args) if err != nil { return nil, err } return &f.inode, nil case linux.ModeRegular: - f, err := newRegularFile(inode) + f, err := newRegularFile(args) if err != nil { return nil, err } return &f.inode, nil case linux.ModeDirectory: - f, err := newDirectory(inode, fs.sb.IncompatibleFeatures().DirentFileType) + f, err := newDirectory(args, fs.sb.IncompatibleFeatures().DirentFileType) if err != nil { return nil, err } @@ -150,6 +150,21 @@ func newInode(fs *filesystem, inodeNum uint32) (*inode, error) { } } +type inodeArgs struct { + fs *filesystem + inodeNum uint32 + blkSize uint64 + diskInode disklayout.Inode +} + +func (in *inode) init(args inodeArgs, impl interface{}) { + in.fs = args.fs + in.inodeNum = args.inodeNum + in.blkSize = args.blkSize + in.diskInode = args.diskInode + in.impl = impl +} + // open creates and returns a file description for the dentry passed in. func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts *vfs.OpenOptions) (*vfs.FileDescription, error) { ats := vfs.AccessTypesForOpenFlags(opts) diff --git a/pkg/sentry/fsimpl/ext/regular_file.go b/pkg/sentry/fsimpl/ext/regular_file.go index f7015c44f..152036b2e 100644 --- a/pkg/sentry/fsimpl/ext/regular_file.go +++ b/pkg/sentry/fsimpl/ext/regular_file.go @@ -43,28 +43,19 @@ type regularFile struct { // newRegularFile is the regularFile constructor. It figures out what kind of // file this is and initializes the fileReader. -func newRegularFile(inode inode) (*regularFile, error) { - regFile := regularFile{ - inode: inode, - } - - inodeFlags := inode.diskInode.Flags() - - if inodeFlags.Extents { - file, err := newExtentFile(regFile) +func newRegularFile(args inodeArgs) (*regularFile, error) { + if args.diskInode.Flags().Extents { + file, err := newExtentFile(args) if err != nil { return nil, err } - - file.regFile.inode.impl = &file.regFile return &file.regFile, nil } - file, err := newBlockMapFile(regFile) + file, err := newBlockMapFile(args) if err != nil { return nil, err } - file.regFile.inode.impl = &file.regFile return &file.regFile, nil } diff --git a/pkg/sentry/fsimpl/ext/symlink.go b/pkg/sentry/fsimpl/ext/symlink.go index 1447a4dc1..acb28d85b 100644 --- a/pkg/sentry/fsimpl/ext/symlink.go +++ b/pkg/sentry/fsimpl/ext/symlink.go @@ -30,18 +30,17 @@ type symlink struct { // newSymlink is the symlink constructor. It reads out the symlink target from // the inode (however it might have been stored). -func newSymlink(inode inode) (*symlink, error) { - var file *symlink +func newSymlink(args inodeArgs) (*symlink, error) { var link []byte // If the symlink target is lesser than 60 bytes, its stores in inode.Data(). // Otherwise either extents or block maps will be used to store the link. - size := inode.diskInode.Size() + size := args.diskInode.Size() if size < 60 { - link = inode.diskInode.Data()[:size] + link = args.diskInode.Data()[:size] } else { // Create a regular file out of this inode and read out the target. - regFile, err := newRegularFile(inode) + regFile, err := newRegularFile(args) if err != nil { return nil, err } @@ -52,8 +51,8 @@ func newSymlink(inode inode) (*symlink, error) { } } - file = &symlink{inode: inode, target: string(link)} - file.inode.impl = file + file := &symlink{target: string(link)} + file.inode.init(args, file) return file, nil } -- cgit v1.2.3 From 77c206e3719962d27cc60781ca59712b040587ab Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Thu, 11 Jun 2020 18:33:35 -0700 Subject: Add //pkg/sentry/fsimpl/overlay. Major differences from existing overlay filesystems: - Linux allows lower layers in an overlay to require revalidation, but not the upper layer. VFS1 allows the upper layer in an overlay to require revalidation, but not the lower layer. VFS2 does not allow any layers to require revalidation. (Now that vfs.MkdirOptions.ForSyntheticMountpoint exists, no uses of overlay in VFS1 are believed to require upper layer revalidation; in particular, the requirement that the upper layer support the creation of "trusted." extended attributes for whiteouts effectively required the upper filesystem to be tmpfs in most cases.) - Like VFS1, but unlike Linux, VFS2 overlay does not attempt to make mutations of the upper layer atomic using a working directory and features like RENAME_WHITEOUT. (This may change in the future, since not having a working directory makes error recovery for some operations, e.g. rmdir, particularly painful.) - Like Linux, but unlike VFS1, VFS2 represents whiteouts using character devices with rdev == 0; the equivalent of the whiteout attribute on directories is xattr trusted.overlay.opaque = "y"; and there is no equivalent to the whiteout attribute on non-directories since non-directories are never merged with lower layers. - Device and inode numbers work as follows: - In Linux, modulo the xino feature and a special case for when all layers are the same filesystem: - Directories use the overlay filesystem's device number and an ephemeral inode number assigned by the overlay. - Non-directories that have been copied up use the device and inode number assigned by the upper filesystem. - Non-directories that have not been copied up use a per-(overlay, layer)-pair device number and the inode number assigned by the lower filesystem. - In VFS1, device and inode numbers always come from the lower layer unless "whited out"; this has the adverse effect of requiring interaction with the lower filesystem even for non-directory files that exist on the upper layer. - In VFS2, device and inode numbers are assigned as in Linux, except that xino and the samefs special case are not supported. - Like Linux, but unlike VFS1, VFS2 does not attempt to maintain memory mapping coherence across copy-up. (This may have to change in the future, as users may be dependent on this property.) - Like Linux, but unlike VFS1, VFS2 uses the overlayfs mounter's credentials when interacting with the overlay's layers, rather than the caller's. - Like Linux, but unlike VFS1, VFS2 permits multiple lower layers in an overlay. - Like Linux, but unlike VFS1, VFS2's overlay filesystem is application-mountable. Updates #1199 PiperOrigin-RevId: 316019067 --- pkg/sentry/fsimpl/gofer/filesystem.go | 12 +- pkg/sentry/fsimpl/kernfs/filesystem.go | 10 +- pkg/sentry/fsimpl/overlay/BUILD | 41 + pkg/sentry/fsimpl/overlay/copy_up.go | 262 ++++++ pkg/sentry/fsimpl/overlay/directory.go | 265 ++++++ pkg/sentry/fsimpl/overlay/filesystem.go | 1364 ++++++++++++++++++++++++++++ pkg/sentry/fsimpl/overlay/non_directory.go | 266 ++++++ pkg/sentry/fsimpl/overlay/overlay.go | 612 +++++++++++++ pkg/sentry/vfs/file_description.go | 17 +- pkg/syserror/syserror.go | 1 + runsc/boot/BUILD | 1 + runsc/boot/vfs.go | 5 + 12 files changed, 2843 insertions(+), 13 deletions(-) create mode 100644 pkg/sentry/fsimpl/overlay/BUILD create mode 100644 pkg/sentry/fsimpl/overlay/copy_up.go create mode 100644 pkg/sentry/fsimpl/overlay/directory.go create mode 100644 pkg/sentry/fsimpl/overlay/filesystem.go create mode 100644 pkg/sentry/fsimpl/overlay/non_directory.go create mode 100644 pkg/sentry/fsimpl/overlay/overlay.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 40933b74b..3c467e313 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -118,7 +118,7 @@ func putDentrySlice(ds *[]*dentry) { // must be up to date. // // Postconditions: The returned dentry's cached metadata is up to date. -func (fs *filesystem) stepLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry, ds **[]*dentry) (*dentry, error) { +func (fs *filesystem) stepLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry, mayFollowSymlinks bool, ds **[]*dentry) (*dentry, error) { if !d.isDir() { return nil, syserror.ENOTDIR } @@ -168,7 +168,7 @@ afterSymlink: if err := rp.CheckMount(&child.vfsd); err != nil { return nil, err } - if child.isSymlink() && rp.ShouldFollowSymlink() { + if child.isSymlink() && mayFollowSymlinks && rp.ShouldFollowSymlink() { target, err := child.readlink(ctx, rp.Mount()) if err != nil { return nil, err @@ -275,7 +275,7 @@ func (fs *filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.Vir func (fs *filesystem) walkParentDirLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry, ds **[]*dentry) (*dentry, error) { for !rp.Final() { d.dirMu.Lock() - next, err := fs.stepLocked(ctx, rp, d, ds) + next, err := fs.stepLocked(ctx, rp, d, true /* mayFollowSymlinks */, ds) d.dirMu.Unlock() if err != nil { return nil, err @@ -301,7 +301,7 @@ func (fs *filesystem) resolveLocked(ctx context.Context, rp *vfs.ResolvingPath, } for !rp.Done() { d.dirMu.Lock() - next, err := fs.stepLocked(ctx, rp, d, ds) + next, err := fs.stepLocked(ctx, rp, d, true /* mayFollowSymlinks */, ds) d.dirMu.Unlock() if err != nil { return nil, err @@ -754,7 +754,7 @@ afterTrailingSymlink: } // Determine whether or not we need to create a file. parent.dirMu.Lock() - child, err := fs.stepLocked(ctx, rp, parent, &ds) + child, err := fs.stepLocked(ctx, rp, parent, false /* mayFollowSymlinks */, &ds) if err == syserror.ENOENT && mayCreate { if parent.isSynthetic() { parent.dirMu.Unlock() @@ -939,7 +939,7 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving // Filter file creation flags and O_LARGEFILE out; the create RPC already // has the semantics of O_CREAT|O_EXCL, while some servers will choke on // O_LARGEFILE. - createFlags := p9.OpenFlags(opts.Flags &^ (linux.O_CREAT | linux.O_EXCL | linux.O_NOCTTY | linux.O_TRUNC | linux.O_LARGEFILE)) + createFlags := p9.OpenFlags(opts.Flags &^ (vfs.FileCreationFlags | linux.O_LARGEFILE)) fdobj, openFile, createQID, _, err := dirfile.create(ctx, name, createFlags, (p9.FileMode)(opts.Mode), (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)) if err != nil { dirfile.close(ctx) diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 4a12ae245..8939871c1 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -35,7 +35,7 @@ import ( // Preconditions: Filesystem.mu must be locked for at least reading. !rp.Done(). // // Postcondition: Caller must call fs.processDeferredDecRefs*. -func (fs *Filesystem) stepExistingLocked(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry) (*vfs.Dentry, error) { +func (fs *Filesystem) stepExistingLocked(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, mayFollowSymlinks bool) (*vfs.Dentry, error) { d := vfsd.Impl().(*Dentry) if !d.isDir() { return nil, syserror.ENOTDIR @@ -81,7 +81,7 @@ afterSymlink: return nil, err } // Resolve any symlink at current path component. - if rp.ShouldFollowSymlink() && next.isSymlink() { + if mayFollowSymlinks && rp.ShouldFollowSymlink() && next.isSymlink() { targetVD, targetPathname, err := next.inode.Getlink(ctx, rp.Mount()) if err != nil { return nil, err @@ -152,7 +152,7 @@ func (fs *Filesystem) walkExistingLocked(ctx context.Context, rp *vfs.ResolvingP vfsd := rp.Start() for !rp.Done() { var err error - vfsd, err = fs.stepExistingLocked(ctx, rp, vfsd) + vfsd, err = fs.stepExistingLocked(ctx, rp, vfsd, true /* mayFollowSymlinks */) if err != nil { return nil, nil, err } @@ -178,7 +178,7 @@ func (fs *Filesystem) walkParentDirLocked(ctx context.Context, rp *vfs.Resolving vfsd := rp.Start() for !rp.Final() { var err error - vfsd, err = fs.stepExistingLocked(ctx, rp, vfsd) + vfsd, err = fs.stepExistingLocked(ctx, rp, vfsd, true /* mayFollowSymlinks */) if err != nil { return nil, nil, err } @@ -449,7 +449,7 @@ afterTrailingSymlink: return nil, syserror.ENAMETOOLONG } // Determine whether or not we need to create a file. - childVFSD, err := fs.stepExistingLocked(ctx, rp, parentVFSD) + childVFSD, err := fs.stepExistingLocked(ctx, rp, parentVFSD, false /* mayFollowSymlinks */) if err == syserror.ENOENT { // Already checked for searchability above; now check for writability. if err := parentInode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite); err != nil { diff --git a/pkg/sentry/fsimpl/overlay/BUILD b/pkg/sentry/fsimpl/overlay/BUILD new file mode 100644 index 000000000..f9413bbdd --- /dev/null +++ b/pkg/sentry/fsimpl/overlay/BUILD @@ -0,0 +1,41 @@ +load("//tools:defs.bzl", "go_library") +load("//tools/go_generics:defs.bzl", "go_template_instance") + +licenses(["notice"]) + +go_template_instance( + name = "fstree", + out = "fstree.go", + package = "overlay", + prefix = "generic", + template = "//pkg/sentry/vfs/genericfstree:generic_fstree", + types = { + "Dentry": "dentry", + }, +) + +go_library( + name = "overlay", + srcs = [ + "copy_up.go", + "directory.go", + "filesystem.go", + "fstree.go", + "non_directory.go", + "overlay.go", + ], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/abi/linux", + "//pkg/context", + "//pkg/fspath", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/memmap", + "//pkg/sentry/socket/unix/transport", + "//pkg/sentry/vfs", + "//pkg/sentry/vfs/lock", + "//pkg/sync", + "//pkg/syserror", + "//pkg/usermem", + ], +) diff --git a/pkg/sentry/fsimpl/overlay/copy_up.go b/pkg/sentry/fsimpl/overlay/copy_up.go new file mode 100644 index 000000000..8f8dcfafe --- /dev/null +++ b/pkg/sentry/fsimpl/overlay/copy_up.go @@ -0,0 +1,262 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package overlay + +import ( + "fmt" + "io" + "sync/atomic" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +func (d *dentry) isCopiedUp() bool { + return atomic.LoadUint32(&d.copiedUp) != 0 +} + +// copyUpLocked ensures that d exists on the upper layer, i.e. d.upperVD.Ok(). +// +// Preconditions: filesystem.renameMu must be locked. +func (d *dentry) copyUpLocked(ctx context.Context) error { + // Fast path. + if d.isCopiedUp() { + return nil + } + + ftype := atomic.LoadUint32(&d.mode) & linux.S_IFMT + switch ftype { + case linux.S_IFREG, linux.S_IFDIR, linux.S_IFLNK, linux.S_IFBLK, linux.S_IFCHR: + // Can be copied-up. + default: + // Can't be copied-up. + return syserror.EPERM + } + + // Ensure that our parent directory is copied-up. + if d.parent == nil { + // d is a filesystem root with no upper layer. + return syserror.EROFS + } + if err := d.parent.copyUpLocked(ctx); err != nil { + return err + } + + d.copyMu.Lock() + defer d.copyMu.Unlock() + if d.upperVD.Ok() { + // Raced with another call to d.copyUpLocked(). + return nil + } + if d.vfsd.IsDead() { + // Raced with deletion of d. + return syserror.ENOENT + } + + // Perform copy-up. + vfsObj := d.fs.vfsfs.VirtualFilesystem() + newpop := vfs.PathOperation{ + Root: d.parent.upperVD, + Start: d.parent.upperVD, + Path: fspath.Parse(d.name), + } + cleanupUndoCopyUp := func() { + var err error + if ftype == linux.S_IFDIR { + err = vfsObj.RmdirAt(ctx, d.fs.creds, &newpop) + } else { + err = vfsObj.UnlinkAt(ctx, d.fs.creds, &newpop) + } + if err != nil { + ctx.Warningf("Unrecoverable overlayfs inconsistency: failed to delete upper layer file after copy-up error: %v", err) + } + } + switch ftype { + case linux.S_IFREG: + oldFD, err := vfsObj.OpenAt(ctx, d.fs.creds, &vfs.PathOperation{ + Root: d.lowerVDs[0], + Start: d.lowerVDs[0], + }, &vfs.OpenOptions{ + Flags: linux.O_RDONLY, + }) + if err != nil { + return err + } + defer oldFD.DecRef() + newFD, err := vfsObj.OpenAt(ctx, d.fs.creds, &newpop, &vfs.OpenOptions{ + Flags: linux.O_WRONLY | linux.O_CREAT | linux.O_EXCL, + Mode: linux.FileMode(d.mode &^ linux.S_IFMT), + }) + if err != nil { + return err + } + defer newFD.DecRef() + bufIOSeq := usermem.BytesIOSequence(make([]byte, 32*1024)) // arbitrary buffer size + for { + readN, readErr := oldFD.Read(ctx, bufIOSeq, vfs.ReadOptions{}) + if readErr != nil && readErr != io.EOF { + cleanupUndoCopyUp() + return readErr + } + total := int64(0) + for total < readN { + writeN, writeErr := newFD.Write(ctx, bufIOSeq.DropFirst64(total), vfs.WriteOptions{}) + total += writeN + if writeErr != nil { + cleanupUndoCopyUp() + return writeErr + } + } + if readErr == io.EOF { + break + } + } + if err := newFD.SetStat(ctx, vfs.SetStatOptions{ + Stat: linux.Statx{ + Mask: linux.STATX_UID | linux.STATX_GID, + UID: d.uid, + GID: d.gid, + }, + }); err != nil { + cleanupUndoCopyUp() + return err + } + d.upperVD = newFD.VirtualDentry() + d.upperVD.IncRef() + + case linux.S_IFDIR: + if err := vfsObj.MkdirAt(ctx, d.fs.creds, &newpop, &vfs.MkdirOptions{ + Mode: linux.FileMode(d.mode &^ linux.S_IFMT), + }); err != nil { + return err + } + if err := vfsObj.SetStatAt(ctx, d.fs.creds, &newpop, &vfs.SetStatOptions{ + Stat: linux.Statx{ + Mask: linux.STATX_UID | linux.STATX_GID, + UID: d.uid, + GID: d.gid, + }, + }); err != nil { + cleanupUndoCopyUp() + return err + } + upperVD, err := vfsObj.GetDentryAt(ctx, d.fs.creds, &newpop, &vfs.GetDentryOptions{}) + if err != nil { + cleanupUndoCopyUp() + return err + } + d.upperVD = upperVD + + case linux.S_IFLNK: + target, err := vfsObj.ReadlinkAt(ctx, d.fs.creds, &vfs.PathOperation{ + Root: d.lowerVDs[0], + Start: d.lowerVDs[0], + }) + if err != nil { + return err + } + if err := vfsObj.SymlinkAt(ctx, d.fs.creds, &newpop, target); err != nil { + return err + } + if err := vfsObj.SetStatAt(ctx, d.fs.creds, &newpop, &vfs.SetStatOptions{ + Stat: linux.Statx{ + Mask: linux.STATX_MODE | linux.STATX_UID | linux.STATX_GID, + Mode: uint16(d.mode), + UID: d.uid, + GID: d.gid, + }, + }); err != nil { + cleanupUndoCopyUp() + return err + } + upperVD, err := vfsObj.GetDentryAt(ctx, d.fs.creds, &newpop, &vfs.GetDentryOptions{}) + if err != nil { + cleanupUndoCopyUp() + return err + } + d.upperVD = upperVD + + case linux.S_IFBLK, linux.S_IFCHR: + lowerStat, err := vfsObj.StatAt(ctx, d.fs.creds, &vfs.PathOperation{ + Root: d.lowerVDs[0], + Start: d.lowerVDs[0], + }, &vfs.StatOptions{}) + if err != nil { + return err + } + if err := vfsObj.MknodAt(ctx, d.fs.creds, &newpop, &vfs.MknodOptions{ + Mode: linux.FileMode(d.mode), + DevMajor: lowerStat.RdevMajor, + DevMinor: lowerStat.RdevMinor, + }); err != nil { + return err + } + if err := vfsObj.SetStatAt(ctx, d.fs.creds, &newpop, &vfs.SetStatOptions{ + Stat: linux.Statx{ + Mask: linux.STATX_UID | linux.STATX_GID, + UID: d.uid, + GID: d.gid, + }, + }); err != nil { + cleanupUndoCopyUp() + return err + } + upperVD, err := vfsObj.GetDentryAt(ctx, d.fs.creds, &newpop, &vfs.GetDentryOptions{}) + if err != nil { + cleanupUndoCopyUp() + return err + } + d.upperVD = upperVD + + default: + // Should have rejected this at the beginning of this function? + panic(fmt.Sprintf("unexpected file type %o", ftype)) + } + + // TODO(gvisor.dev/issue/1199): copy up xattrs + + // Update the dentry's device and inode numbers (except for directories, + // for which these remain overlay-assigned). + if ftype != linux.S_IFDIR { + upperStat, err := vfsObj.StatAt(ctx, d.fs.creds, &vfs.PathOperation{ + Root: d.upperVD, + Start: d.upperVD, + }, &vfs.StatOptions{ + Mask: linux.STATX_INO, + }) + if err != nil { + d.upperVD.DecRef() + d.upperVD = vfs.VirtualDentry{} + cleanupUndoCopyUp() + return err + } + if upperStat.Mask&linux.STATX_INO == 0 { + d.upperVD.DecRef() + d.upperVD = vfs.VirtualDentry{} + cleanupUndoCopyUp() + return syserror.EREMOTE + } + atomic.StoreUint32(&d.devMajor, upperStat.DevMajor) + atomic.StoreUint32(&d.devMinor, upperStat.DevMinor) + atomic.StoreUint64(&d.ino, upperStat.Ino) + } + + atomic.StoreUint32(&d.copiedUp, 1) + return nil +} diff --git a/pkg/sentry/fsimpl/overlay/directory.go b/pkg/sentry/fsimpl/overlay/directory.go new file mode 100644 index 000000000..6f47167d3 --- /dev/null +++ b/pkg/sentry/fsimpl/overlay/directory.go @@ -0,0 +1,265 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package overlay + +import ( + "sync/atomic" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/syserror" +) + +func (d *dentry) isDir() bool { + return atomic.LoadUint32(&d.mode)&linux.S_IFMT == linux.S_IFDIR +} + +// Preconditions: d.dirMu must be locked. d.isDir(). +func (d *dentry) collectWhiteoutsForRmdirLocked(ctx context.Context) (map[string]bool, error) { + vfsObj := d.fs.vfsfs.VirtualFilesystem() + var readdirErr error + whiteouts := make(map[string]bool) + var maybeWhiteouts []string + d.iterLayers(func(layerVD vfs.VirtualDentry, isUpper bool) bool { + layerFD, err := vfsObj.OpenAt(ctx, d.fs.creds, &vfs.PathOperation{ + Root: layerVD, + Start: layerVD, + }, &vfs.OpenOptions{ + Flags: linux.O_RDONLY | linux.O_DIRECTORY, + }) + if err != nil { + readdirErr = err + return false + } + defer layerFD.DecRef() + + // Reuse slice allocated for maybeWhiteouts from a previous layer to + // reduce allocations. + maybeWhiteouts = maybeWhiteouts[:0] + if err := layerFD.IterDirents(ctx, vfs.IterDirentsCallbackFunc(func(dirent vfs.Dirent) error { + if dirent.Name == "." || dirent.Name == ".." { + return nil + } + if _, ok := whiteouts[dirent.Name]; ok { + // This file has been whited-out in a previous layer. + return nil + } + if dirent.Type == linux.DT_CHR { + // We have to determine if this is a whiteout, which doesn't + // count against the directory's emptiness. However, we can't + // do so while holding locks held by layerFD.IterDirents(). + maybeWhiteouts = append(maybeWhiteouts, dirent.Name) + return nil + } + // Non-whiteout file in the directory prevents rmdir. + return syserror.ENOTEMPTY + })); err != nil { + readdirErr = err + return false + } + + for _, maybeWhiteoutName := range maybeWhiteouts { + stat, err := vfsObj.StatAt(ctx, d.fs.creds, &vfs.PathOperation{ + Root: layerVD, + Start: layerVD, + Path: fspath.Parse(maybeWhiteoutName), + }, &vfs.StatOptions{}) + if err != nil { + readdirErr = err + return false + } + if stat.RdevMajor != 0 || stat.RdevMinor != 0 { + // This file is a real character device, not a whiteout. + readdirErr = syserror.ENOTEMPTY + return false + } + whiteouts[maybeWhiteoutName] = isUpper + } + // Continue iteration since we haven't found any non-whiteout files in + // this directory yet. + return true + }) + return whiteouts, readdirErr +} + +type directoryFD struct { + fileDescription + vfs.DirectoryFileDescriptionDefaultImpl + vfs.DentryMetadataFileDescriptionImpl + + mu sync.Mutex + off int64 + dirents []vfs.Dirent +} + +// Release implements vfs.FileDescriptionImpl.Release. +func (fd *directoryFD) Release() { +} + +// IterDirents implements vfs.FileDescriptionImpl.IterDirents. +func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback) error { + fd.mu.Lock() + defer fd.mu.Unlock() + + d := fd.dentry() + if fd.dirents == nil { + ds, err := d.getDirents(ctx) + if err != nil { + return err + } + fd.dirents = ds + } + + for fd.off < int64(len(fd.dirents)) { + if err := cb.Handle(fd.dirents[fd.off]); err != nil { + return err + } + fd.off++ + } + return nil +} + +// Preconditions: d.isDir(). +func (d *dentry) getDirents(ctx context.Context) ([]vfs.Dirent, error) { + d.fs.renameMu.RLock() + defer d.fs.renameMu.RUnlock() + d.dirMu.Lock() + defer d.dirMu.Unlock() + + if d.dirents != nil { + return d.dirents, nil + } + + parent := genericParentOrSelf(d) + dirents := []vfs.Dirent{ + { + Name: ".", + Type: linux.DT_DIR, + Ino: d.ino, + NextOff: 1, + }, + { + Name: "..", + Type: uint8(atomic.LoadUint32(&parent.mode) >> 12), + Ino: parent.ino, + NextOff: 2, + }, + } + + // Merge dirents from all layers comprising this directory. + vfsObj := d.fs.vfsfs.VirtualFilesystem() + var readdirErr error + prevDirents := make(map[string]struct{}) + var maybeWhiteouts []vfs.Dirent + d.iterLayers(func(layerVD vfs.VirtualDentry, isUpper bool) bool { + layerFD, err := vfsObj.OpenAt(ctx, d.fs.creds, &vfs.PathOperation{ + Root: layerVD, + Start: layerVD, + }, &vfs.OpenOptions{ + Flags: linux.O_RDONLY | linux.O_DIRECTORY, + }) + if err != nil { + readdirErr = err + return false + } + defer layerFD.DecRef() + + // Reuse slice allocated for maybeWhiteouts from a previous layer to + // reduce allocations. + maybeWhiteouts = maybeWhiteouts[:0] + if err := layerFD.IterDirents(ctx, vfs.IterDirentsCallbackFunc(func(dirent vfs.Dirent) error { + if dirent.Name == "." || dirent.Name == ".." { + return nil + } + if _, ok := prevDirents[dirent.Name]; ok { + // This file is hidden by, or merged with, another file with + // the same name in a previous layer. + return nil + } + prevDirents[dirent.Name] = struct{}{} + if dirent.Type == linux.DT_CHR { + // We can't determine if this file is a whiteout while holding + // locks held by layerFD.IterDirents(). + maybeWhiteouts = append(maybeWhiteouts, dirent) + return nil + } + dirent.NextOff = int64(len(dirents) + 1) + dirents = append(dirents, dirent) + return nil + })); err != nil { + readdirErr = err + return false + } + + for _, dirent := range maybeWhiteouts { + stat, err := vfsObj.StatAt(ctx, d.fs.creds, &vfs.PathOperation{ + Root: layerVD, + Start: layerVD, + Path: fspath.Parse(dirent.Name), + }, &vfs.StatOptions{}) + if err != nil { + readdirErr = err + return false + } + if stat.RdevMajor == 0 && stat.RdevMinor == 0 { + // This file is a whiteout; don't emit a dirent for it. + continue + } + dirent.NextOff = int64(len(dirents) + 1) + dirents = append(dirents, dirent) + } + return true + }) + if readdirErr != nil { + return nil, readdirErr + } + + // Cache dirents for future directoryFDs. + d.dirents = dirents + return dirents, nil +} + +// Seek implements vfs.FileDescriptionImpl.Seek. +func (fd *directoryFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + fd.mu.Lock() + defer fd.mu.Unlock() + + switch whence { + case linux.SEEK_SET: + if offset < 0 { + return 0, syserror.EINVAL + } + if offset == 0 { + // Ensure that the next call to fd.IterDirents() calls + // fd.dentry().getDirents(). + fd.dirents = nil + } + fd.off = offset + return fd.off, nil + case linux.SEEK_CUR: + offset += fd.off + if offset < 0 { + return 0, syserror.EINVAL + } + // Don't clear fd.dirents in this case, even if offset == 0. + fd.off = offset + return fd.off, nil + default: + return 0, syserror.EINVAL + } +} diff --git a/pkg/sentry/fsimpl/overlay/filesystem.go b/pkg/sentry/fsimpl/overlay/filesystem.go new file mode 100644 index 000000000..ff82e1f20 --- /dev/null +++ b/pkg/sentry/fsimpl/overlay/filesystem.go @@ -0,0 +1,1364 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package overlay + +import ( + "sync/atomic" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/syserror" +) + +// _OVL_XATTR_OPAQUE is an extended attribute key whose value is set to "y" for +// opaque directories. +// Linux: fs/overlayfs/overlayfs.h:OVL_XATTR_OPAQUE +const _OVL_XATTR_OPAQUE = "trusted.overlay.opaque" + +func isWhiteout(stat *linux.Statx) bool { + return stat.Mode&linux.S_IFMT == linux.S_IFCHR && stat.RdevMajor == 0 && stat.RdevMinor == 0 +} + +// Sync implements vfs.FilesystemImpl.Sync. +func (fs *filesystem) Sync(ctx context.Context) error { + if fs.opts.UpperRoot.Ok() { + return fs.opts.UpperRoot.Mount().Filesystem().Impl().Sync(ctx) + } + return nil +} + +var dentrySlicePool = sync.Pool{ + New: func() interface{} { + ds := make([]*dentry, 0, 4) // arbitrary non-zero initial capacity + return &ds + }, +} + +func appendDentry(ds *[]*dentry, d *dentry) *[]*dentry { + if ds == nil { + ds = dentrySlicePool.Get().(*[]*dentry) + } + *ds = append(*ds, d) + return ds +} + +// Preconditions: ds != nil. +func putDentrySlice(ds *[]*dentry) { + // Allow dentries to be GC'd. + for i := range *ds { + (*ds)[i] = nil + } + *ds = (*ds)[:0] + dentrySlicePool.Put(ds) +} + +// renameMuRUnlockAndCheckDrop calls fs.renameMu.RUnlock(), then calls +// dentry.checkDropLocked on all dentries in *ds with fs.renameMu locked for +// writing. +// +// ds is a pointer-to-pointer since defer evaluates its arguments immediately, +// but dentry slices are allocated lazily, and it's much easier to say "defer +// fs.renameMuRUnlockAndCheckDrop(&ds)" than "defer func() { +// fs.renameMuRUnlockAndCheckDrop(ds) }()" to work around this. +func (fs *filesystem) renameMuRUnlockAndCheckDrop(ds **[]*dentry) { + fs.renameMu.RUnlock() + if *ds == nil { + return + } + if len(**ds) != 0 { + fs.renameMu.Lock() + for _, d := range **ds { + d.checkDropLocked() + } + fs.renameMu.Unlock() + } + putDentrySlice(*ds) +} + +func (fs *filesystem) renameMuUnlockAndCheckDrop(ds **[]*dentry) { + if *ds == nil { + fs.renameMu.Unlock() + return + } + for _, d := range **ds { + d.checkDropLocked() + } + fs.renameMu.Unlock() + putDentrySlice(*ds) +} + +// stepLocked resolves rp.Component() to an existing file, starting from the +// given directory. +// +// Dentries which may have a reference count of zero, and which therefore +// should be dropped once traversal is complete, are appended to ds. +// +// Preconditions: fs.renameMu must be locked. d.dirMu must be locked. +// !rp.Done(). +func (fs *filesystem) stepLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry, mayFollowSymlinks bool, ds **[]*dentry) (*dentry, error) { + if !d.isDir() { + return nil, syserror.ENOTDIR + } + if err := d.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { + return nil, err + } +afterSymlink: + name := rp.Component() + if name == "." { + rp.Advance() + return d, nil + } + if name == ".." { + if isRoot, err := rp.CheckRoot(&d.vfsd); err != nil { + return nil, err + } else if isRoot || d.parent == nil { + rp.Advance() + return d, nil + } + if err := rp.CheckMount(&d.parent.vfsd); err != nil { + return nil, err + } + rp.Advance() + return d.parent, nil + } + child, err := fs.getChildLocked(ctx, d, name, ds) + if err != nil { + return nil, err + } + if err := rp.CheckMount(&child.vfsd); err != nil { + return nil, err + } + if child.isSymlink() && mayFollowSymlinks && rp.ShouldFollowSymlink() { + target, err := child.readlink(ctx) + if err != nil { + return nil, err + } + if err := rp.HandleSymlink(target); err != nil { + return nil, err + } + goto afterSymlink // don't check the current directory again + } + rp.Advance() + return child, nil +} + +// Preconditions: fs.renameMu must be locked. d.dirMu must be locked. +func (fs *filesystem) getChildLocked(ctx context.Context, parent *dentry, name string, ds **[]*dentry) (*dentry, error) { + if child, ok := parent.children[name]; ok { + return child, nil + } + child, err := fs.lookupLocked(ctx, parent, name) + if err != nil { + return nil, err + } + if parent.children == nil { + parent.children = make(map[string]*dentry) + } + parent.children[name] = child + // child's refcount is initially 0, so it may be dropped after traversal. + *ds = appendDentry(*ds, child) + return child, nil +} + +// Preconditions: fs.renameMu must be locked. parent.dirMu must be locked. +func (fs *filesystem) lookupLocked(ctx context.Context, parent *dentry, name string) (*dentry, error) { + childPath := fspath.Parse(name) + child := fs.newDentry() + existsOnAnyLayer := false + var lookupErr error + + vfsObj := fs.vfsfs.VirtualFilesystem() + parent.iterLayers(func(parentVD vfs.VirtualDentry, isUpper bool) bool { + childVD, err := vfsObj.GetDentryAt(ctx, fs.creds, &vfs.PathOperation{ + Root: parentVD, + Start: parentVD, + Path: childPath, + }, &vfs.GetDentryOptions{}) + if err == syserror.ENOENT || err == syserror.ENAMETOOLONG { + // The file doesn't exist on this layer. Proceed to the next one. + return true + } + if err != nil { + lookupErr = err + return false + } + + mask := uint32(linux.STATX_TYPE) + if !existsOnAnyLayer { + // Mode, UID, GID, and (for non-directories) inode number come from + // the topmost layer on which the file exists. + mask |= linux.STATX_MODE | linux.STATX_UID | linux.STATX_GID | linux.STATX_INO + } + stat, err := vfsObj.StatAt(ctx, fs.creds, &vfs.PathOperation{ + Root: childVD, + Start: childVD, + }, &vfs.StatOptions{ + Mask: mask, + }) + if err != nil { + lookupErr = err + return false + } + if stat.Mask&mask != mask { + lookupErr = syserror.EREMOTE + return false + } + + if isWhiteout(&stat) { + // This is a whiteout, so it "doesn't exist" on this layer, and + // layers below this one are ignored. + return false + } + isDir := stat.Mode&linux.S_IFMT == linux.S_IFDIR + if existsOnAnyLayer && !isDir { + // Directories are not merged with non-directory files from lower + // layers; instead, layers including and below the first + // non-directory file are ignored. (This file must be a directory + // on previous layers, since lower layers aren't searched for + // non-directory files.) + return false + } + + // Update child to include this layer. + if isUpper { + child.upperVD = childVD + child.copiedUp = 1 + } else { + child.lowerVDs = append(child.lowerVDs, childVD) + } + if !existsOnAnyLayer { + existsOnAnyLayer = true + child.mode = uint32(stat.Mode) + child.uid = stat.UID + child.gid = stat.GID + child.devMajor = stat.DevMajor + child.devMinor = stat.DevMinor + child.ino = stat.Ino + } + + // For non-directory files, only the topmost layer that contains a file + // matters. + if !isDir { + return false + } + + // Directories are merged with directories from lower layers if they + // are not explicitly opaque. + opaqueVal, err := vfsObj.GetxattrAt(ctx, fs.creds, &vfs.PathOperation{ + Root: childVD, + Start: childVD, + }, &vfs.GetxattrOptions{ + Name: _OVL_XATTR_OPAQUE, + Size: 1, + }) + return !(err == nil && opaqueVal == "y") + }) + + if lookupErr != nil { + child.destroyLocked() + return nil, lookupErr + } + if !existsOnAnyLayer { + child.destroyLocked() + return nil, syserror.ENOENT + } + + // Device and inode numbers were copied from the topmost layer above; + // override them if necessary. + if child.isDir() { + child.devMajor = linux.UNNAMED_MAJOR + child.devMinor = fs.dirDevMinor + child.ino = fs.newDirIno() + } else if !child.upperVD.Ok() { + child.devMajor = linux.UNNAMED_MAJOR + child.devMinor = fs.lowerDevMinors[child.lowerVDs[0].Mount().Filesystem()] + } + + parent.IncRef() + child.parent = parent + child.name = name + return child, nil +} + +// lookupLayerLocked is similar to lookupLocked, but only returns information +// about the file rather than a dentry. +// +// Preconditions: fs.renameMu must be locked. parent.dirMu must be locked. +func (fs *filesystem) lookupLayerLocked(ctx context.Context, parent *dentry, name string) (lookupLayer, error) { + childPath := fspath.Parse(name) + lookupLayer := lookupLayerNone + var lookupErr error + + parent.iterLayers(func(parentVD vfs.VirtualDentry, isUpper bool) bool { + stat, err := fs.vfsfs.VirtualFilesystem().StatAt(ctx, fs.creds, &vfs.PathOperation{ + Root: parentVD, + Start: parentVD, + Path: childPath, + }, &vfs.StatOptions{ + Mask: linux.STATX_TYPE, + }) + if err == syserror.ENOENT || err == syserror.ENAMETOOLONG { + // The file doesn't exist on this layer. Proceed to the next + // one. + return true + } + if err != nil { + lookupErr = err + return false + } + if stat.Mask&linux.STATX_TYPE == 0 { + // Linux's overlayfs tends to return EREMOTE in cases where a file + // is unusable for reasons that are not better captured by another + // errno. + lookupErr = syserror.EREMOTE + return false + } + if isWhiteout(&stat) { + // This is a whiteout, so it "doesn't exist" on this layer, and + // layers below this one are ignored. + if isUpper { + lookupLayer = lookupLayerUpperWhiteout + } + return false + } + // The file exists; we can stop searching. + if isUpper { + lookupLayer = lookupLayerUpper + } else { + lookupLayer = lookupLayerLower + } + return false + }) + + return lookupLayer, lookupErr +} + +type lookupLayer int + +const ( + // lookupLayerNone indicates that no file exists at the given path on the + // upper layer, and is either whited out or does not exist on lower layers. + // Therefore, the file does not exist in the overlay filesystem, and file + // creation may proceed normally (if an upper layer exists). + lookupLayerNone lookupLayer = iota + + // lookupLayerLower indicates that no file exists at the given path on the + // upper layer, but exists on a lower layer. Therefore, the file exists in + // the overlay filesystem, but must be copied-up before mutation. + lookupLayerLower + + // lookupLayerUpper indicates that a non-whiteout file exists at the given + // path on the upper layer. Therefore, the file exists in the overlay + // filesystem, and is already copied-up. + lookupLayerUpper + + // lookupLayerUpperWhiteout indicates that a whiteout exists at the given + // path on the upper layer. Therefore, the file does not exist in the + // overlay filesystem, and file creation must remove the whiteout before + // proceeding. + lookupLayerUpperWhiteout +) + +func (ll lookupLayer) existsInOverlay() bool { + return ll == lookupLayerLower || ll == lookupLayerUpper +} + +// walkParentDirLocked resolves all but the last path component of rp to an +// existing directory, starting from the given directory (which is usually +// rp.Start().Impl().(*dentry)). It does not check that the returned directory +// is searchable by the provider of rp. +// +// Preconditions: fs.renameMu must be locked. !rp.Done(). +func (fs *filesystem) walkParentDirLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry, ds **[]*dentry) (*dentry, error) { + for !rp.Final() { + d.dirMu.Lock() + next, err := fs.stepLocked(ctx, rp, d, true /* mayFollowSymlinks */, ds) + d.dirMu.Unlock() + if err != nil { + return nil, err + } + d = next + } + if !d.isDir() { + return nil, syserror.ENOTDIR + } + return d, nil +} + +// resolveLocked resolves rp to an existing file. +// +// Preconditions: fs.renameMu must be locked. +func (fs *filesystem) resolveLocked(ctx context.Context, rp *vfs.ResolvingPath, ds **[]*dentry) (*dentry, error) { + d := rp.Start().Impl().(*dentry) + for !rp.Done() { + d.dirMu.Lock() + next, err := fs.stepLocked(ctx, rp, d, true /* mayFollowSymlinks */, ds) + d.dirMu.Unlock() + if err != nil { + return nil, err + } + d = next + } + if rp.MustBeDir() && !d.isDir() { + return nil, syserror.ENOTDIR + } + return d, nil +} + +// doCreateAt checks that creating a file at rp is permitted, then invokes +// create to do so. +// +// Preconditions: !rp.Done(). For the final path component in rp, +// !rp.ShouldFollowSymlink(). +func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir bool, create func(parent *dentry, name string, haveUpperWhiteout bool) error) error { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckDrop(&ds) + start := rp.Start().Impl().(*dentry) + parent, err := fs.walkParentDirLocked(ctx, rp, start, &ds) + if err != nil { + return err + } + if err := parent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { + return err + } + name := rp.Component() + if name == "." || name == ".." { + return syserror.EEXIST + } + if !dir && rp.MustBeDir() { + return syserror.ENOENT + } + if parent.vfsd.IsDead() { + return syserror.ENOENT + } + mnt := rp.Mount() + if err := mnt.CheckBeginWrite(); err != nil { + return err + } + defer mnt.EndWrite() + parent.dirMu.Lock() + defer parent.dirMu.Unlock() + + // Determine if a file already exists at name. + if _, ok := parent.children[name]; ok { + return syserror.EEXIST + } + childLayer, err := fs.lookupLayerLocked(ctx, parent, name) + if err != nil { + return err + } + if childLayer.existsInOverlay() { + return syserror.EEXIST + } + + // Ensure that the parent directory is copied-up so that we can create the + // new file in the upper layer. + if err := parent.copyUpLocked(ctx); err != nil { + return err + } + + // Finally create the new file. + if err := create(parent, name, childLayer == lookupLayerUpperWhiteout); err != nil { + return err + } + parent.dirents = nil + return nil +} + +// Preconditions: pop's parent directory has been copied up. +func (fs *filesystem) createWhiteout(ctx context.Context, vfsObj *vfs.VirtualFilesystem, pop *vfs.PathOperation) error { + return vfsObj.MknodAt(ctx, fs.creds, pop, &vfs.MknodOptions{ + Mode: linux.S_IFCHR, // permissions == include/linux/fs.h:WHITEOUT_MODE == 0 + // DevMajor == DevMinor == 0, from include/linux/fs.h:WHITEOUT_DEV + }) +} + +func (fs *filesystem) cleanupRecreateWhiteout(ctx context.Context, vfsObj *vfs.VirtualFilesystem, pop *vfs.PathOperation) { + if err := fs.createWhiteout(ctx, vfsObj, pop); err != nil { + ctx.Warningf("Unrecoverable overlayfs inconsistency: failed to recreate whiteout after failed file creation: %v", err) + } +} + +// AccessAt implements vfs.Filesystem.Impl.AccessAt. +func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckDrop(&ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return err + } + return d.checkPermissions(creds, ats) +} + +// BoundEndpointAt implements vfs.FilesystemImpl.BoundEndpointAt. +func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.BoundEndpointOptions) (transport.BoundEndpoint, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckDrop(&ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return nil, err + } + if err := d.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil { + return nil, err + } + layerVD := d.topLayer() + return fs.vfsfs.VirtualFilesystem().BoundEndpointAt(ctx, fs.creds, &vfs.PathOperation{ + Root: layerVD, + Start: layerVD, + }, &opts) +} + +// GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. +func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckDrop(&ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return nil, err + } + if opts.CheckSearchable { + if !d.isDir() { + return nil, syserror.ENOTDIR + } + if err := d.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { + return nil, err + } + } + d.IncRef() + return &d.vfsd, nil +} + +// GetParentDentryAt implements vfs.FilesystemImpl.GetParentDentryAt. +func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckDrop(&ds) + start := rp.Start().Impl().(*dentry) + d, err := fs.walkParentDirLocked(ctx, rp, start, &ds) + if err != nil { + return nil, err + } + d.IncRef() + return &d.vfsd, nil +} + +// LinkAt implements vfs.FilesystemImpl.LinkAt. +func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry) error { + return fs.doCreateAt(ctx, rp, false /* dir */, func(parent *dentry, childName string, haveUpperWhiteout bool) error { + if rp.Mount() != vd.Mount() { + return syserror.EXDEV + } + old := vd.Dentry().Impl().(*dentry) + if old.isDir() { + return syserror.EPERM + } + if err := old.copyUpLocked(ctx); err != nil { + return err + } + vfsObj := fs.vfsfs.VirtualFilesystem() + newpop := vfs.PathOperation{ + Root: parent.upperVD, + Start: parent.upperVD, + Path: fspath.Parse(childName), + } + if haveUpperWhiteout { + if err := vfsObj.UnlinkAt(ctx, fs.creds, &newpop); err != nil { + return err + } + } + if err := vfsObj.LinkAt(ctx, fs.creds, &vfs.PathOperation{ + Root: old.upperVD, + Start: old.upperVD, + }, &newpop); err != nil { + if haveUpperWhiteout { + fs.cleanupRecreateWhiteout(ctx, vfsObj, &newpop) + } + return err + } + creds := rp.Credentials() + if err := vfsObj.SetStatAt(ctx, fs.creds, &newpop, &vfs.SetStatOptions{ + Stat: linux.Statx{ + Mask: linux.STATX_UID | linux.STATX_GID, + UID: uint32(creds.EffectiveKUID), + GID: uint32(creds.EffectiveKGID), + }, + }); err != nil { + if cleanupErr := vfsObj.UnlinkAt(ctx, fs.creds, &newpop); cleanupErr != nil { + ctx.Warningf("Unrecoverable overlayfs inconsistency: failed to delete upper layer file after LinkAt metadata update failure: %v", cleanupErr) + } else if haveUpperWhiteout { + fs.cleanupRecreateWhiteout(ctx, vfsObj, &newpop) + } + return err + } + return nil + }) +} + +// MkdirAt implements vfs.FilesystemImpl.MkdirAt. +func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error { + return fs.doCreateAt(ctx, rp, true /* dir */, func(parent *dentry, childName string, haveUpperWhiteout bool) error { + vfsObj := fs.vfsfs.VirtualFilesystem() + pop := vfs.PathOperation{ + Root: parent.upperVD, + Start: parent.upperVD, + Path: fspath.Parse(childName), + } + if haveUpperWhiteout { + if err := vfsObj.UnlinkAt(ctx, fs.creds, &pop); err != nil { + return err + } + } + if err := vfsObj.MkdirAt(ctx, fs.creds, &pop, &opts); err != nil { + if haveUpperWhiteout { + fs.cleanupRecreateWhiteout(ctx, vfsObj, &pop) + } + return err + } + creds := rp.Credentials() + if err := vfsObj.SetStatAt(ctx, fs.creds, &pop, &vfs.SetStatOptions{ + Stat: linux.Statx{ + Mask: linux.STATX_UID | linux.STATX_GID, + UID: uint32(creds.EffectiveKUID), + GID: uint32(creds.EffectiveKGID), + }, + }); err != nil { + if cleanupErr := vfsObj.RmdirAt(ctx, fs.creds, &pop); cleanupErr != nil { + ctx.Warningf("Unrecoverable overlayfs inconsistency: failed to delete upper layer directory after MkdirAt metadata update failure: %v", cleanupErr) + } else if haveUpperWhiteout { + fs.cleanupRecreateWhiteout(ctx, vfsObj, &pop) + } + return err + } + if haveUpperWhiteout { + // There may be directories on lower layers (previously hidden by + // the whiteout) that the new directory should not be merged with. + // Mark it opaque to prevent merging. + if err := vfsObj.SetxattrAt(ctx, fs.creds, &pop, &vfs.SetxattrOptions{ + Name: _OVL_XATTR_OPAQUE, + Value: "y", + }); err != nil { + if cleanupErr := vfsObj.RmdirAt(ctx, fs.creds, &pop); cleanupErr != nil { + ctx.Warningf("Unrecoverable overlayfs inconsistency: failed to delete upper layer directory after MkdirAt set-opaque failure: %v", cleanupErr) + } else { + fs.cleanupRecreateWhiteout(ctx, vfsObj, &pop) + } + return err + } + } + return nil + }) +} + +// MknodAt implements vfs.FilesystemImpl.MknodAt. +func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MknodOptions) error { + return fs.doCreateAt(ctx, rp, false /* dir */, func(parent *dentry, childName string, haveUpperWhiteout bool) error { + // Disallow attempts to create whiteouts. + if opts.Mode&linux.S_IFMT == linux.S_IFCHR && opts.DevMajor == 0 && opts.DevMinor == 0 { + return syserror.EPERM + } + vfsObj := fs.vfsfs.VirtualFilesystem() + pop := vfs.PathOperation{ + Root: parent.upperVD, + Start: parent.upperVD, + Path: fspath.Parse(childName), + } + if haveUpperWhiteout { + if err := vfsObj.UnlinkAt(ctx, fs.creds, &pop); err != nil { + return err + } + } + if err := vfsObj.MknodAt(ctx, fs.creds, &pop, &opts); err != nil { + if haveUpperWhiteout { + fs.cleanupRecreateWhiteout(ctx, vfsObj, &pop) + } + return err + } + creds := rp.Credentials() + if err := vfsObj.SetStatAt(ctx, fs.creds, &pop, &vfs.SetStatOptions{ + Stat: linux.Statx{ + Mask: linux.STATX_UID | linux.STATX_GID, + UID: uint32(creds.EffectiveKUID), + GID: uint32(creds.EffectiveKGID), + }, + }); err != nil { + if cleanupErr := vfsObj.UnlinkAt(ctx, fs.creds, &pop); cleanupErr != nil { + ctx.Warningf("Unrecoverable overlayfs inconsistency: failed to delete upper layer file after MknodAt metadata update failure: %v", cleanupErr) + } else if haveUpperWhiteout { + fs.cleanupRecreateWhiteout(ctx, vfsObj, &pop) + } + return err + } + return nil + }) +} + +// OpenAt implements vfs.FilesystemImpl.OpenAt. +func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + mayCreate := opts.Flags&linux.O_CREAT != 0 + mustCreate := opts.Flags&(linux.O_CREAT|linux.O_EXCL) == (linux.O_CREAT | linux.O_EXCL) + + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckDrop(&ds) + + start := rp.Start().Impl().(*dentry) + if rp.Done() { + if mustCreate { + return nil, syserror.EEXIST + } + return start.openLocked(ctx, rp, &opts) + } + +afterTrailingSymlink: + parent, err := fs.walkParentDirLocked(ctx, rp, start, &ds) + if err != nil { + return nil, err + } + // Check for search permission in the parent directory. + if err := parent.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { + return nil, err + } + // Determine whether or not we need to create a file. + parent.dirMu.Lock() + child, err := fs.stepLocked(ctx, rp, parent, false /* mayFollowSymlinks */, &ds) + if err == syserror.ENOENT && mayCreate { + fd, err := fs.createAndOpenLocked(ctx, rp, parent, &opts, &ds) + parent.dirMu.Unlock() + return fd, err + } + if err != nil { + parent.dirMu.Unlock() + return nil, err + } + // Open existing child or follow symlink. + parent.dirMu.Unlock() + if mustCreate { + return nil, syserror.EEXIST + } + if child.isSymlink() && rp.ShouldFollowSymlink() { + target, err := child.readlink(ctx) + if err != nil { + return nil, err + } + if err := rp.HandleSymlink(target); err != nil { + return nil, err + } + start = parent + goto afterTrailingSymlink + } + return child.openLocked(ctx, rp, &opts) +} + +// Preconditions: fs.renameMu must be locked. +func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.OpenOptions) (*vfs.FileDescription, error) { + ats := vfs.AccessTypesForOpenFlags(opts) + if err := d.checkPermissions(rp.Credentials(), ats); err != nil { + return nil, err + } + if ats.MayWrite() { + if err := d.copyUpLocked(ctx); err != nil { + return nil, err + } + } + mnt := rp.Mount() + + // Directory FDs open FDs from each layer when directory entries are read, + // so they don't require opening an FD from d.topLayer() up front. + ftype := atomic.LoadUint32(&d.mode) & linux.S_IFMT + if ftype == linux.S_IFDIR { + // Can't open directories with O_CREAT. + if opts.Flags&linux.O_CREAT != 0 { + return nil, syserror.EISDIR + } + // Can't open directories writably. + if ats&vfs.MayWrite != 0 { + return nil, syserror.EISDIR + } + if opts.Flags&linux.O_DIRECT != 0 { + return nil, syserror.EINVAL + } + fd := &directoryFD{} + fd.LockFD.Init(&d.locks) + if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{ + UseDentryMetadata: true, + }); err != nil { + return nil, err + } + return &fd.vfsfd, nil + } + + layerVD, isUpper := d.topLayerInfo() + layerFD, err := rp.VirtualFilesystem().OpenAt(ctx, d.fs.creds, &vfs.PathOperation{ + Root: layerVD, + Start: layerVD, + }, opts) + if err != nil { + return nil, err + } + layerFlags := layerFD.StatusFlags() + fd := &nonDirectoryFD{ + copiedUp: isUpper, + cachedFD: layerFD, + cachedFlags: layerFlags, + } + fd.LockFD.Init(&d.locks) + layerFDOpts := layerFD.Options() + if err := fd.vfsfd.Init(fd, layerFlags, mnt, &d.vfsd, &layerFDOpts); err != nil { + layerFD.DecRef() + return nil, err + } + return &fd.vfsfd, nil +} + +// Preconditions: parent.dirMu must be locked. parent does not already contain +// a child named rp.Component(). +func (fs *filesystem) createAndOpenLocked(ctx context.Context, rp *vfs.ResolvingPath, parent *dentry, opts *vfs.OpenOptions, ds **[]*dentry) (*vfs.FileDescription, error) { + creds := rp.Credentials() + if err := parent.checkPermissions(creds, vfs.MayWrite); err != nil { + return nil, err + } + if parent.vfsd.IsDead() { + return nil, syserror.ENOENT + } + mnt := rp.Mount() + if err := mnt.CheckBeginWrite(); err != nil { + return nil, err + } + defer mnt.EndWrite() + + if err := parent.copyUpLocked(ctx); err != nil { + return nil, err + } + + vfsObj := fs.vfsfs.VirtualFilesystem() + childName := rp.Component() + pop := vfs.PathOperation{ + Root: parent.upperVD, + Start: parent.upperVD, + Path: fspath.Parse(childName), + } + // We don't know if a whiteout exists on the upper layer; speculatively + // unlink it. + // + // TODO(gvisor.dev/issue/1199): Modify OpenAt => stepLocked so that we do + // know whether a whiteout exists. + var haveUpperWhiteout bool + switch err := vfsObj.UnlinkAt(ctx, fs.creds, &pop); err { + case nil: + haveUpperWhiteout = true + case syserror.ENOENT: + haveUpperWhiteout = false + default: + return nil, err + } + // Create the file on the upper layer, and get an FD representing it. + upperFD, err := vfsObj.OpenAt(ctx, fs.creds, &pop, &vfs.OpenOptions{ + Flags: opts.Flags&^vfs.FileCreationFlags | linux.O_CREAT | linux.O_EXCL, + Mode: opts.Mode, + }) + if err != nil { + if haveUpperWhiteout { + fs.cleanupRecreateWhiteout(ctx, vfsObj, &pop) + } + return nil, err + } + // Change the file's owner to the caller. We can't use upperFD.SetStat() + // because it will pick up creds from ctx. + if err := vfsObj.SetStatAt(ctx, fs.creds, &pop, &vfs.SetStatOptions{ + Stat: linux.Statx{ + Mask: linux.STATX_UID | linux.STATX_GID, + UID: uint32(creds.EffectiveKUID), + GID: uint32(creds.EffectiveKGID), + }, + }); err != nil { + if cleanupErr := vfsObj.UnlinkAt(ctx, fs.creds, &pop); cleanupErr != nil { + ctx.Warningf("Unrecoverable overlayfs inconsistency: failed to delete upper layer file after OpenAt(O_CREAT) metadata update failure: %v", cleanupErr) + } else if haveUpperWhiteout { + fs.cleanupRecreateWhiteout(ctx, vfsObj, &pop) + } + return nil, err + } + // Re-lookup to get a dentry representing the new file, which is needed for + // the returned FD. + child, err := fs.getChildLocked(ctx, parent, childName, ds) + if err != nil { + if cleanupErr := vfsObj.UnlinkAt(ctx, fs.creds, &pop); cleanupErr != nil { + ctx.Warningf("Unrecoverable overlayfs inconsistency: failed to delete upper layer file after OpenAt(O_CREAT) dentry lookup failure: %v", cleanupErr) + } else if haveUpperWhiteout { + fs.cleanupRecreateWhiteout(ctx, vfsObj, &pop) + } + return nil, err + } + // Finally construct the overlay FD. + upperFlags := upperFD.StatusFlags() + fd := &nonDirectoryFD{ + copiedUp: true, + cachedFD: upperFD, + cachedFlags: upperFlags, + } + fd.LockFD.Init(&child.locks) + upperFDOpts := upperFD.Options() + if err := fd.vfsfd.Init(fd, upperFlags, mnt, &child.vfsd, &upperFDOpts); err != nil { + upperFD.DecRef() + // Don't bother with cleanup; the file was created successfully, we + // just can't open it anymore for some reason. + return nil, err + } + return &fd.vfsfd, nil +} + +// ReadlinkAt implements vfs.FilesystemImpl.ReadlinkAt. +func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckDrop(&ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return "", err + } + layerVD := d.topLayer() + return fs.vfsfs.VirtualFilesystem().ReadlinkAt(ctx, d.fs.creds, &vfs.PathOperation{ + Root: layerVD, + Start: layerVD, + }) +} + +// RenameAt implements vfs.FilesystemImpl.RenameAt. +func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldParentVD vfs.VirtualDentry, oldName string, opts vfs.RenameOptions) error { + if opts.Flags != 0 { + return syserror.EINVAL + } + + var ds *[]*dentry + fs.renameMu.Lock() + defer fs.renameMuUnlockAndCheckDrop(&ds) + newParent, err := fs.walkParentDirLocked(ctx, rp, rp.Start().Impl().(*dentry), &ds) + if err != nil { + return err + } + newName := rp.Component() + if newName == "." || newName == ".." { + return syserror.EBUSY + } + mnt := rp.Mount() + if mnt != oldParentVD.Mount() { + return syserror.EXDEV + } + if err := mnt.CheckBeginWrite(); err != nil { + return err + } + defer mnt.EndWrite() + + // FIXME(gvisor.dev/issue/1199): Actually implement rename. + _ = newParent + return syserror.EXDEV +} + +// RmdirAt implements vfs.FilesystemImpl.RmdirAt. +func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckDrop(&ds) + start := rp.Start().Impl().(*dentry) + parent, err := fs.walkParentDirLocked(ctx, rp, start, &ds) + if err != nil { + return err + } + if err := parent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { + return err + } + if err := rp.Mount().CheckBeginWrite(); err != nil { + return err + } + defer rp.Mount().EndWrite() + name := rp.Component() + if name == "." { + return syserror.EINVAL + } + if name == ".." { + return syserror.ENOTEMPTY + } + vfsObj := rp.VirtualFilesystem() + mntns := vfs.MountNamespaceFromContext(ctx) + defer mntns.DecRef() + parent.dirMu.Lock() + defer parent.dirMu.Unlock() + + // Ensure that parent is copied-up before potentially holding child.copyMu + // below. + if err := parent.copyUpLocked(ctx); err != nil { + return err + } + + // Unlike UnlinkAt, we need a dentry representing the child directory being + // removed in order to verify that it's empty. + child, err := fs.getChildLocked(ctx, parent, name, &ds) + if err != nil { + return err + } + if !child.isDir() { + return syserror.ENOTDIR + } + child.dirMu.Lock() + defer child.dirMu.Unlock() + whiteouts, err := child.collectWhiteoutsForRmdirLocked(ctx) + if err != nil { + return err + } + child.copyMu.RLock() + defer child.copyMu.RUnlock() + if err := vfsObj.PrepareDeleteDentry(mntns, &child.vfsd); err != nil { + return err + } + + pop := vfs.PathOperation{ + Root: parent.upperVD, + Start: parent.upperVD, + Path: fspath.Parse(name), + } + if child.upperVD.Ok() { + cleanupRecreateWhiteouts := func() { + if !child.upperVD.Ok() { + return + } + for whiteoutName, whiteoutUpper := range whiteouts { + if !whiteoutUpper { + continue + } + if err := fs.createWhiteout(ctx, vfsObj, &vfs.PathOperation{ + Root: child.upperVD, + Start: child.upperVD, + Path: fspath.Parse(whiteoutName), + }); err != nil && err != syserror.EEXIST { + ctx.Warningf("Unrecoverable overlayfs inconsistency: failed to recreate deleted whiteout after RmdirAt failure: %v", err) + } + } + } + // Remove existing whiteouts on the upper layer. + for whiteoutName, whiteoutUpper := range whiteouts { + if !whiteoutUpper { + continue + } + if err := vfsObj.UnlinkAt(ctx, fs.creds, &vfs.PathOperation{ + Root: child.upperVD, + Start: child.upperVD, + Path: fspath.Parse(whiteoutName), + }); err != nil { + cleanupRecreateWhiteouts() + vfsObj.AbortDeleteDentry(&child.vfsd) + return err + } + } + // Remove the existing directory on the upper layer. + if err := vfsObj.RmdirAt(ctx, fs.creds, &pop); err != nil { + cleanupRecreateWhiteouts() + vfsObj.AbortDeleteDentry(&child.vfsd) + return err + } + } + if err := fs.createWhiteout(ctx, vfsObj, &pop); err != nil { + // Don't attempt to recover from this: the original directory is + // already gone, so any dentries representing it are invalid, and + // creating a new directory won't undo that. + ctx.Warningf("Unrecoverable overlayfs inconsistency: failed to create whiteout during RmdirAt: %v", err) + vfsObj.AbortDeleteDentry(&child.vfsd) + return err + } + + vfsObj.CommitDeleteDentry(&child.vfsd) + delete(parent.children, name) + ds = appendDentry(ds, child) + parent.dirents = nil + return nil +} + +// SetStatAt implements vfs.FilesystemImpl.SetStatAt. +func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckDrop(&ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return err + } + + mode := linux.FileMode(atomic.LoadUint32(&d.mode)) + if err := vfs.CheckSetStat(ctx, rp.Credentials(), &opts.Stat, mode, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))); err != nil { + return err + } + mnt := rp.Mount() + if err := mnt.CheckBeginWrite(); err != nil { + return err + } + defer mnt.EndWrite() + if err := d.copyUpLocked(ctx); err != nil { + return err + } + // Changes to d's attributes are serialized by d.copyMu. + d.copyMu.Lock() + defer d.copyMu.Unlock() + if err := d.fs.vfsfs.VirtualFilesystem().SetStatAt(ctx, d.fs.creds, &vfs.PathOperation{ + Root: d.upperVD, + Start: d.upperVD, + }, &opts); err != nil { + return err + } + d.updateAfterSetStatLocked(&opts) + return nil +} + +// StatAt implements vfs.FilesystemImpl.StatAt. +func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckDrop(&ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return linux.Statx{}, err + } + + var stat linux.Statx + if layerMask := opts.Mask &^ statInternalMask; layerMask != 0 { + layerVD := d.topLayer() + stat, err = fs.vfsfs.VirtualFilesystem().StatAt(ctx, fs.creds, &vfs.PathOperation{ + Root: layerVD, + Start: layerVD, + }, &vfs.StatOptions{ + Mask: layerMask, + Sync: opts.Sync, + }) + if err != nil { + return linux.Statx{}, err + } + } + d.statInternalTo(ctx, &opts, &stat) + return stat, nil +} + +// StatFSAt implements vfs.FilesystemImpl.StatFSAt. +func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckDrop(&ds) + _, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return linux.Statfs{}, err + } + return fs.statFS(ctx) +} + +// SymlinkAt implements vfs.FilesystemImpl.SymlinkAt. +func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, target string) error { + return fs.doCreateAt(ctx, rp, false /* dir */, func(parent *dentry, childName string, haveUpperWhiteout bool) error { + vfsObj := fs.vfsfs.VirtualFilesystem() + pop := vfs.PathOperation{ + Root: parent.upperVD, + Start: parent.upperVD, + Path: fspath.Parse(childName), + } + if haveUpperWhiteout { + if err := vfsObj.UnlinkAt(ctx, fs.creds, &pop); err != nil { + return err + } + } + if err := vfsObj.SymlinkAt(ctx, fs.creds, &pop, target); err != nil { + if haveUpperWhiteout { + fs.cleanupRecreateWhiteout(ctx, vfsObj, &pop) + } + return err + } + creds := rp.Credentials() + if err := vfsObj.SetStatAt(ctx, fs.creds, &pop, &vfs.SetStatOptions{ + Stat: linux.Statx{ + Mask: linux.STATX_UID | linux.STATX_GID, + UID: uint32(creds.EffectiveKUID), + GID: uint32(creds.EffectiveKGID), + }, + }); err != nil { + if cleanupErr := vfsObj.UnlinkAt(ctx, fs.creds, &pop); cleanupErr != nil { + ctx.Warningf("Unrecoverable overlayfs inconsistency: failed to delete upper layer file after SymlinkAt metadata update failure: %v", cleanupErr) + } else if haveUpperWhiteout { + fs.cleanupRecreateWhiteout(ctx, vfsObj, &pop) + } + return err + } + return nil + }) +} + +// UnlinkAt implements vfs.FilesystemImpl.UnlinkAt. +func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckDrop(&ds) + start := rp.Start().Impl().(*dentry) + parent, err := fs.walkParentDirLocked(ctx, rp, start, &ds) + if err != nil { + return err + } + if err := parent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { + return err + } + if err := rp.Mount().CheckBeginWrite(); err != nil { + return err + } + defer rp.Mount().EndWrite() + name := rp.Component() + if name == "." || name == ".." { + return syserror.EISDIR + } + if rp.MustBeDir() { + return syserror.ENOTDIR + } + vfsObj := rp.VirtualFilesystem() + mntns := vfs.MountNamespaceFromContext(ctx) + defer mntns.DecRef() + parent.dirMu.Lock() + defer parent.dirMu.Unlock() + + // Ensure that parent is copied-up before potentially holding child.copyMu + // below. + if err := parent.copyUpLocked(ctx); err != nil { + return err + } + + child := parent.children[name] + var childLayer lookupLayer + if child != nil { + if child.isDir() { + return syserror.EISDIR + } + if err := vfsObj.PrepareDeleteDentry(mntns, &child.vfsd); err != nil { + return err + } + // Hold child.copyMu to prevent it from being copied-up during + // deletion. + child.copyMu.RLock() + defer child.copyMu.RUnlock() + if child.upperVD.Ok() { + childLayer = lookupLayerUpper + } else { + childLayer = lookupLayerLower + } + } else { + // Determine if the file being unlinked actually exists. Holding + // parent.dirMu prevents a dentry from being instantiated for the file, + // which in turn prevents it from being copied-up, so this result is + // stable. + childLayer, err = fs.lookupLayerLocked(ctx, parent, name) + if err != nil { + return err + } + if !childLayer.existsInOverlay() { + return syserror.ENOENT + } + } + + pop := vfs.PathOperation{ + Root: parent.upperVD, + Start: parent.upperVD, + Path: fspath.Parse(name), + } + if childLayer == lookupLayerUpper { + // Remove the existing file on the upper layer. + if err := vfsObj.UnlinkAt(ctx, fs.creds, &pop); err != nil { + if child != nil { + vfsObj.AbortDeleteDentry(&child.vfsd) + } + return err + } + } + if err := fs.createWhiteout(ctx, vfsObj, &pop); err != nil { + ctx.Warningf("Unrecoverable overlayfs inconsistency: failed to create whiteout during UnlinkAt: %v", err) + if child != nil { + vfsObj.AbortDeleteDentry(&child.vfsd) + } + return err + } + + if child != nil { + vfsObj.CommitDeleteDentry(&child.vfsd) + delete(parent.children, name) + ds = appendDentry(ds, child) + } + parent.dirents = nil + return nil +} + +// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. +func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, size uint64) ([]string, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckDrop(&ds) + _, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return nil, err + } + // TODO(gvisor.dev/issue/1199): Linux overlayfs actually allows listxattr, + // but not any other xattr syscalls. For now we just reject all of them. + return nil, syserror.ENOTSUP +} + +// GetxattrAt implements vfs.FilesystemImpl.GetxattrAt. +func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetxattrOptions) (string, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckDrop(&ds) + _, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return "", err + } + return "", syserror.ENOTSUP +} + +// SetxattrAt implements vfs.FilesystemImpl.SetxattrAt. +func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckDrop(&ds) + _, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return err + } + return syserror.ENOTSUP +} + +// RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. +func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckDrop(&ds) + _, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return err + } + return syserror.ENOTSUP +} + +// PrependPath implements vfs.FilesystemImpl.PrependPath. +func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { + fs.renameMu.RLock() + defer fs.renameMu.RUnlock() + return genericPrependPath(vfsroot, vd.Mount(), vd.Dentry().Impl().(*dentry), b) +} diff --git a/pkg/sentry/fsimpl/overlay/non_directory.go b/pkg/sentry/fsimpl/overlay/non_directory.go new file mode 100644 index 000000000..a3c1f7a8d --- /dev/null +++ b/pkg/sentry/fsimpl/overlay/non_directory.go @@ -0,0 +1,266 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package overlay + +import ( + "sync/atomic" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/memmap" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/usermem" +) + +func (d *dentry) isSymlink() bool { + return atomic.LoadUint32(&d.mode)&linux.S_IFMT == linux.S_IFLNK +} + +func (d *dentry) readlink(ctx context.Context) (string, error) { + layerVD := d.topLayer() + return d.fs.vfsfs.VirtualFilesystem().ReadlinkAt(ctx, d.fs.creds, &vfs.PathOperation{ + Root: layerVD, + Start: layerVD, + }) +} + +type nonDirectoryFD struct { + fileDescription + + // If copiedUp is false, cachedFD represents + // fileDescription.dentry().lowerVDs[0]; otherwise, cachedFD represents + // fileDescription.dentry().upperVD. cachedFlags is the last known value of + // cachedFD.StatusFlags(). copiedUp, cachedFD, and cachedFlags are + // protected by mu. + mu sync.Mutex + copiedUp bool + cachedFD *vfs.FileDescription + cachedFlags uint32 +} + +func (fd *nonDirectoryFD) getCurrentFD(ctx context.Context) (*vfs.FileDescription, error) { + fd.mu.Lock() + defer fd.mu.Unlock() + wrappedFD, err := fd.currentFDLocked(ctx) + if err != nil { + return nil, err + } + wrappedFD.IncRef() + return wrappedFD, nil +} + +func (fd *nonDirectoryFD) currentFDLocked(ctx context.Context) (*vfs.FileDescription, error) { + d := fd.dentry() + statusFlags := fd.vfsfd.StatusFlags() + if !fd.copiedUp && d.isCopiedUp() { + // Switch to the copied-up file. + upperVD := d.topLayer() + upperFD, err := fd.filesystem().vfsfs.VirtualFilesystem().OpenAt(ctx, d.fs.creds, &vfs.PathOperation{ + Root: upperVD, + Start: upperVD, + }, &vfs.OpenOptions{ + Flags: statusFlags, + }) + if err != nil { + return nil, err + } + oldOff, oldOffErr := fd.cachedFD.Seek(ctx, 0, linux.SEEK_CUR) + if oldOffErr == nil { + if _, err := upperFD.Seek(ctx, oldOff, linux.SEEK_SET); err != nil { + upperFD.DecRef() + return nil, err + } + } + fd.cachedFD.DecRef() + fd.copiedUp = true + fd.cachedFD = upperFD + fd.cachedFlags = statusFlags + } else if fd.cachedFlags != statusFlags { + if err := fd.cachedFD.SetStatusFlags(ctx, d.fs.creds, statusFlags); err != nil { + return nil, err + } + fd.cachedFlags = statusFlags + } + return fd.cachedFD, nil +} + +// Release implements vfs.FileDescriptionImpl.Release. +func (fd *nonDirectoryFD) Release() { + fd.cachedFD.DecRef() + fd.cachedFD = nil +} + +// OnClose implements vfs.FileDescriptionImpl.OnClose. +func (fd *nonDirectoryFD) OnClose(ctx context.Context) error { + // Linux doesn't define ovl_file_operations.flush at all (i.e. its + // equivalent to OnClose is a no-op). We pass through to + // fd.cachedFD.OnClose() without upgrading if fd.dentry() has been + // copied-up, since OnClose is mostly used to define post-close writeback, + // and if fd.cachedFD hasn't been updated then it can't have been used to + // mutate fd.dentry() anyway. + fd.mu.Lock() + if statusFlags := fd.vfsfd.StatusFlags(); fd.cachedFlags != statusFlags { + if err := fd.cachedFD.SetStatusFlags(ctx, fd.filesystem().creds, statusFlags); err != nil { + fd.mu.Unlock() + return err + } + fd.cachedFlags = statusFlags + } + wrappedFD := fd.cachedFD + defer wrappedFD.IncRef() + fd.mu.Unlock() + return wrappedFD.OnClose(ctx) +} + +// Stat implements vfs.FileDescriptionImpl.Stat. +func (fd *nonDirectoryFD) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { + var stat linux.Statx + if layerMask := opts.Mask &^ statInternalMask; layerMask != 0 { + wrappedFD, err := fd.getCurrentFD(ctx) + if err != nil { + return linux.Statx{}, err + } + stat, err = wrappedFD.Stat(ctx, vfs.StatOptions{ + Mask: layerMask, + Sync: opts.Sync, + }) + wrappedFD.DecRef() + if err != nil { + return linux.Statx{}, err + } + } + fd.dentry().statInternalTo(ctx, &opts, &stat) + return stat, nil +} + +// SetStat implements vfs.FileDescriptionImpl.SetStat. +func (fd *nonDirectoryFD) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { + d := fd.dentry() + mode := linux.FileMode(atomic.LoadUint32(&d.mode)) + if err := vfs.CheckSetStat(ctx, auth.CredentialsFromContext(ctx), &opts.Stat, mode, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))); err != nil { + return err + } + mnt := fd.vfsfd.Mount() + if err := mnt.CheckBeginWrite(); err != nil { + return err + } + defer mnt.EndWrite() + if err := d.copyUpLocked(ctx); err != nil { + return err + } + // Changes to d's attributes are serialized by d.copyMu. + d.copyMu.Lock() + defer d.copyMu.Unlock() + wrappedFD, err := fd.currentFDLocked(ctx) + if err != nil { + return err + } + if err := wrappedFD.SetStat(ctx, opts); err != nil { + return err + } + d.updateAfterSetStatLocked(&opts) + return nil +} + +// StatFS implements vfs.FileDesciptionImpl.StatFS. +func (fd *nonDirectoryFD) StatFS(ctx context.Context) (linux.Statfs, error) { + return fd.filesystem().statFS(ctx) +} + +// PRead implements vfs.FileDescriptionImpl.PRead. +func (fd *nonDirectoryFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + wrappedFD, err := fd.getCurrentFD(ctx) + if err != nil { + return 0, err + } + defer wrappedFD.DecRef() + return wrappedFD.PRead(ctx, dst, offset, opts) +} + +// Read implements vfs.FileDescriptionImpl.Read. +func (fd *nonDirectoryFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + // Hold fd.mu during the read to serialize the file offset. + fd.mu.Lock() + defer fd.mu.Unlock() + wrappedFD, err := fd.currentFDLocked(ctx) + if err != nil { + return 0, err + } + return wrappedFD.Read(ctx, dst, opts) +} + +// PWrite implements vfs.FileDescriptionImpl.PWrite. +func (fd *nonDirectoryFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + wrappedFD, err := fd.getCurrentFD(ctx) + if err != nil { + return 0, err + } + defer wrappedFD.DecRef() + return wrappedFD.PWrite(ctx, src, offset, opts) +} + +// Write implements vfs.FileDescriptionImpl.Write. +func (fd *nonDirectoryFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + // Hold fd.mu during the write to serialize the file offset. + fd.mu.Lock() + defer fd.mu.Unlock() + wrappedFD, err := fd.currentFDLocked(ctx) + if err != nil { + return 0, err + } + return wrappedFD.Write(ctx, src, opts) +} + +// Seek implements vfs.FileDescriptionImpl.Seek. +func (fd *nonDirectoryFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + // Hold fd.mu during the seek to serialize the file offset. + fd.mu.Lock() + defer fd.mu.Unlock() + wrappedFD, err := fd.currentFDLocked(ctx) + if err != nil { + return 0, err + } + return wrappedFD.Seek(ctx, offset, whence) +} + +// Sync implements vfs.FileDescriptionImpl.Sync. +func (fd *nonDirectoryFD) Sync(ctx context.Context) error { + fd.mu.Lock() + if !fd.dentry().isCopiedUp() { + fd.mu.Unlock() + return nil + } + wrappedFD, err := fd.currentFDLocked(ctx) + if err != nil { + fd.mu.Unlock() + return err + } + wrappedFD.IncRef() + defer wrappedFD.DecRef() + fd.mu.Unlock() + return wrappedFD.Sync(ctx) +} + +// ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. +func (fd *nonDirectoryFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { + wrappedFD, err := fd.getCurrentFD(ctx) + if err != nil { + return err + } + defer wrappedFD.DecRef() + return wrappedFD.ConfigureMMap(ctx, opts) +} diff --git a/pkg/sentry/fsimpl/overlay/overlay.go b/pkg/sentry/fsimpl/overlay/overlay.go new file mode 100644 index 000000000..e660d0e2c --- /dev/null +++ b/pkg/sentry/fsimpl/overlay/overlay.go @@ -0,0 +1,612 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package overlay provides an overlay filesystem implementation, which +// synthesizes a filesystem by composing one or more immutable filesystems +// ("lower layers") with an optional mutable filesystem ("upper layer"). +// +// Lock order: +// +// directoryFD.mu / nonDirectoryFD.mu +// filesystem.renameMu +// dentry.dirMu +// dentry.copyMu +// +// Locking dentry.dirMu in multiple dentries requires that parent dentries are +// locked before child dentries, and that filesystem.renameMu is locked to +// stabilize this relationship. +package overlay + +import ( + "strings" + "sync/atomic" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sentry/vfs/lock" + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/syserror" +) + +// Name is the default filesystem name. +const Name = "overlay" + +// FilesystemType implements vfs.FilesystemType. +type FilesystemType struct{} + +// Name implements vfs.FilesystemType.Name. +func (FilesystemType) Name() string { + return Name +} + +// FilesystemOptions may be passed as vfs.GetFilesystemOptions.InternalData to +// FilesystemType.GetFilesystem. +type FilesystemOptions struct { + // Callers passing FilesystemOptions to + // overlay.FilesystemType.GetFilesystem() are responsible for ensuring that + // the vfs.Mounts comprising the layers of the overlay filesystem do not + // contain submounts. + + // If UpperRoot.Ok(), it is the root of the writable upper layer of the + // overlay. + UpperRoot vfs.VirtualDentry + + // LowerRoots contains the roots of the immutable lower layers of the + // overlay. LowerRoots is immutable. + LowerRoots []vfs.VirtualDentry +} + +// filesystem implements vfs.FilesystemImpl. +type filesystem struct { + vfsfs vfs.Filesystem + + // Immutable options. + opts FilesystemOptions + + // creds is a copy of the filesystem's creator's credentials, which are + // used for accesses to the filesystem's layers. creds is immutable. + creds *auth.Credentials + + // dirDevMinor is the device minor number used for directories. dirDevMinor + // is immutable. + dirDevMinor uint32 + + // lowerDevMinors maps lower layer filesystems to device minor numbers + // assigned to non-directory files originating from that filesystem. + // lowerDevMinors is immutable. + lowerDevMinors map[*vfs.Filesystem]uint32 + + // renameMu synchronizes renaming with non-renaming operations in order to + // ensure consistent lock ordering between dentry.dirMu in different + // dentries. + renameMu sync.RWMutex + + // lastDirIno is the last inode number assigned to a directory. lastDirIno + // is accessed using atomic memory operations. + lastDirIno uint64 +} + +// GetFilesystem implements vfs.FilesystemType.GetFilesystem. +func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { + mopts := vfs.GenericParseMountOptions(opts.Data) + fsoptsRaw := opts.InternalData + fsopts, haveFSOpts := fsoptsRaw.(FilesystemOptions) + if fsoptsRaw != nil && !haveFSOpts { + ctx.Warningf("overlay.FilesystemType.GetFilesystem: GetFilesystemOptions.InternalData has type %T, wanted overlay.FilesystemOptions or nil", fsoptsRaw) + return nil, nil, syserror.EINVAL + } + if haveFSOpts { + if len(fsopts.LowerRoots) == 0 { + ctx.Warningf("overlay.FilesystemType.GetFilesystem: LowerRoots must be non-empty") + return nil, nil, syserror.EINVAL + } + if len(fsopts.LowerRoots) < 2 && !fsopts.UpperRoot.Ok() { + ctx.Warningf("overlay.FilesystemType.GetFilesystem: at least two LowerRoots are required when UpperRoot is unspecified") + return nil, nil, syserror.EINVAL + } + // We don't enforce a maximum number of lower layers when not + // configured by applications; the sandbox owner can have an overlay + // filesystem with any number of lower layers. + } else { + vfsroot := vfs.RootFromContext(ctx) + defer vfsroot.DecRef() + upperPathname, ok := mopts["upperdir"] + if ok { + delete(mopts, "upperdir") + // Linux overlayfs also requires a workdir when upperdir is + // specified; we don't, so silently ignore this option. + delete(mopts, "workdir") + upperPath := fspath.Parse(upperPathname) + if !upperPath.Absolute { + ctx.Warningf("overlay.FilesystemType.GetFilesystem: upperdir %q must be absolute", upperPathname) + return nil, nil, syserror.EINVAL + } + upperRoot, err := vfsObj.GetDentryAt(ctx, creds, &vfs.PathOperation{ + Root: vfsroot, + Start: vfsroot, + Path: upperPath, + FollowFinalSymlink: true, + }, &vfs.GetDentryOptions{ + CheckSearchable: true, + }) + if err != nil { + ctx.Warningf("overlay.FilesystemType.GetFilesystem: failed to resolve upperdir %q: %v", upperPathname, err) + return nil, nil, err + } + defer upperRoot.DecRef() + privateUpperRoot, err := clonePrivateMount(vfsObj, upperRoot, false /* forceReadOnly */) + if err != nil { + ctx.Warningf("overlay.FilesystemType.GetFilesystem: failed to make private bind mount of upperdir %q: %v", upperPathname, err) + return nil, nil, err + } + defer privateUpperRoot.DecRef() + fsopts.UpperRoot = privateUpperRoot + } + lowerPathnamesStr, ok := mopts["lowerdir"] + if !ok { + ctx.Warningf("overlay.FilesystemType.GetFilesystem: missing required option lowerdir") + return nil, nil, syserror.EINVAL + } + delete(mopts, "lowerdir") + lowerPathnames := strings.Split(lowerPathnamesStr, ":") + const maxLowerLayers = 500 // Linux: fs/overlay/super.c:OVL_MAX_STACK + if len(lowerPathnames) < 2 && !fsopts.UpperRoot.Ok() { + ctx.Warningf("overlay.FilesystemType.GetFilesystem: at least two lowerdirs are required when upperdir is unspecified") + return nil, nil, syserror.EINVAL + } + if len(lowerPathnames) > maxLowerLayers { + ctx.Warningf("overlay.FilesystemType.GetFilesystem: %d lowerdirs specified, maximum %d", len(lowerPathnames), maxLowerLayers) + return nil, nil, syserror.EINVAL + } + for _, lowerPathname := range lowerPathnames { + lowerPath := fspath.Parse(lowerPathname) + if !lowerPath.Absolute { + ctx.Warningf("overlay.FilesystemType.GetFilesystem: lowerdir %q must be absolute", lowerPathname) + return nil, nil, syserror.EINVAL + } + lowerRoot, err := vfsObj.GetDentryAt(ctx, creds, &vfs.PathOperation{ + Root: vfsroot, + Start: vfsroot, + Path: lowerPath, + FollowFinalSymlink: true, + }, &vfs.GetDentryOptions{ + CheckSearchable: true, + }) + if err != nil { + ctx.Warningf("overlay.FilesystemType.GetFilesystem: failed to resolve lowerdir %q: %v", lowerPathname, err) + return nil, nil, err + } + defer lowerRoot.DecRef() + privateLowerRoot, err := clonePrivateMount(vfsObj, lowerRoot, true /* forceReadOnly */) + if err != nil { + ctx.Warningf("overlay.FilesystemType.GetFilesystem: failed to make private bind mount of lowerdir %q: %v", lowerPathname, err) + return nil, nil, err + } + defer privateLowerRoot.DecRef() + fsopts.LowerRoots = append(fsopts.LowerRoots, privateLowerRoot) + } + } + if len(mopts) != 0 { + ctx.Warningf("overlay.FilesystemType.GetFilesystem: unused options: %v", mopts) + return nil, nil, syserror.EINVAL + } + + // Allocate device numbers. + dirDevMinor, err := vfsObj.GetAnonBlockDevMinor() + if err != nil { + return nil, nil, err + } + lowerDevMinors := make(map[*vfs.Filesystem]uint32) + for _, lowerRoot := range fsopts.LowerRoots { + lowerFS := lowerRoot.Mount().Filesystem() + if _, ok := lowerDevMinors[lowerFS]; !ok { + devMinor, err := vfsObj.GetAnonBlockDevMinor() + if err != nil { + vfsObj.PutAnonBlockDevMinor(dirDevMinor) + for _, lowerDevMinor := range lowerDevMinors { + vfsObj.PutAnonBlockDevMinor(lowerDevMinor) + } + return nil, nil, err + } + lowerDevMinors[lowerFS] = devMinor + } + } + + // Take extra references held by the filesystem. + if fsopts.UpperRoot.Ok() { + fsopts.UpperRoot.IncRef() + } + for _, lowerRoot := range fsopts.LowerRoots { + lowerRoot.IncRef() + } + + fs := &filesystem{ + opts: fsopts, + creds: creds.Fork(), + dirDevMinor: dirDevMinor, + lowerDevMinors: lowerDevMinors, + } + fs.vfsfs.Init(vfsObj, &fstype, fs) + + // Construct the root dentry. + root := fs.newDentry() + root.refs = 1 + if fs.opts.UpperRoot.Ok() { + fs.opts.UpperRoot.IncRef() + root.copiedUp = 1 + root.upperVD = fs.opts.UpperRoot + } + for _, lowerRoot := range fs.opts.LowerRoots { + lowerRoot.IncRef() + root.lowerVDs = append(root.lowerVDs, lowerRoot) + } + rootTopVD := root.topLayer() + // Get metadata from the topmost layer. See fs.lookupLocked(). + const rootStatMask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_UID | linux.STATX_GID | linux.STATX_INO + rootStat, err := vfsObj.StatAt(ctx, creds, &vfs.PathOperation{ + Root: rootTopVD, + Start: rootTopVD, + }, &vfs.StatOptions{ + Mask: rootStatMask, + }) + if err != nil { + root.destroyLocked() + fs.vfsfs.DecRef() + return nil, nil, err + } + if rootStat.Mask&rootStatMask != rootStatMask { + root.destroyLocked() + fs.vfsfs.DecRef() + return nil, nil, syserror.EREMOTE + } + if isWhiteout(&rootStat) { + ctx.Warningf("overlay.FilesystemType.GetFilesystem: filesystem root is a whiteout") + root.destroyLocked() + fs.vfsfs.DecRef() + return nil, nil, syserror.EINVAL + } + root.mode = uint32(rootStat.Mode) + root.uid = rootStat.UID + root.gid = rootStat.GID + if rootStat.Mode&linux.S_IFMT == linux.S_IFDIR { + root.devMajor = linux.UNNAMED_MAJOR + root.devMinor = fs.dirDevMinor + root.ino = fs.newDirIno() + } else if !root.upperVD.Ok() { + root.devMajor = linux.UNNAMED_MAJOR + root.devMinor = fs.lowerDevMinors[root.lowerVDs[0].Mount().Filesystem()] + root.ino = rootStat.Ino + } else { + root.devMajor = rootStat.DevMajor + root.devMinor = rootStat.DevMinor + root.ino = rootStat.Ino + } + + return &fs.vfsfs, &root.vfsd, nil +} + +// clonePrivateMount creates a non-recursive bind mount rooted at vd, not +// associated with any MountNamespace, and returns the root of the new mount. +// (This is required to ensure that each layer of an overlay comprises only a +// single mount, and therefore can't cross into e.g. the overlay filesystem +// itself, risking lock recursion.) A reference is held on the returned +// VirtualDentry. +func clonePrivateMount(vfsObj *vfs.VirtualFilesystem, vd vfs.VirtualDentry, forceReadOnly bool) (vfs.VirtualDentry, error) { + oldmnt := vd.Mount() + opts := oldmnt.Options() + if forceReadOnly { + opts.ReadOnly = true + } + newmnt, err := vfsObj.NewDisconnectedMount(oldmnt.Filesystem(), vd.Dentry(), &opts) + if err != nil { + return vfs.VirtualDentry{}, err + } + return vfs.MakeVirtualDentry(newmnt, vd.Dentry()), nil +} + +// Release implements vfs.FilesystemImpl.Release. +func (fs *filesystem) Release() { + vfsObj := fs.vfsfs.VirtualFilesystem() + vfsObj.PutAnonBlockDevMinor(fs.dirDevMinor) + for _, lowerDevMinor := range fs.lowerDevMinors { + vfsObj.PutAnonBlockDevMinor(lowerDevMinor) + } + if fs.opts.UpperRoot.Ok() { + fs.opts.UpperRoot.DecRef() + } + for _, lowerRoot := range fs.opts.LowerRoots { + lowerRoot.DecRef() + } +} + +func (fs *filesystem) statFS(ctx context.Context) (linux.Statfs, error) { + // Always statfs the root of the topmost layer. Compare Linux's + // fs/overlayfs/super.c:ovl_statfs(). + var rootVD vfs.VirtualDentry + if fs.opts.UpperRoot.Ok() { + rootVD = fs.opts.UpperRoot + } else { + rootVD = fs.opts.LowerRoots[0] + } + fsstat, err := fs.vfsfs.VirtualFilesystem().StatFSAt(ctx, fs.creds, &vfs.PathOperation{ + Root: rootVD, + Start: rootVD, + }) + if err != nil { + return linux.Statfs{}, err + } + fsstat.Type = linux.OVERLAYFS_SUPER_MAGIC + return fsstat, nil +} + +func (fs *filesystem) newDirIno() uint64 { + return atomic.AddUint64(&fs.lastDirIno, 1) +} + +// dentry implements vfs.DentryImpl. +type dentry struct { + vfsd vfs.Dentry + + refs int64 + + // fs is the owning filesystem. fs is immutable. + fs *filesystem + + // mode, uid, and gid are the file mode, owner, and group of the file in + // the topmost layer (and therefore the overlay file as well), and are used + // for permission checks on this dentry. These fields are protected by + // copyMu and accessed using atomic memory operations. + mode uint32 + uid uint32 + gid uint32 + + // copiedUp is 1 if this dentry has been copied-up (i.e. upperVD.Ok()) and + // 0 otherwise. copiedUp is accessed using atomic memory operations. + copiedUp uint32 + + // parent is the dentry corresponding to this dentry's parent directory. + // name is this dentry's name in parent. If this dentry is a filesystem + // root, parent is nil and name is the empty string. parent and name are + // protected by fs.renameMu. + parent *dentry + name string + + // If this dentry represents a directory, children maps the names of + // children for which dentries have been instantiated to those dentries, + // and dirents (if not nil) is a cache of dirents as returned by + // directoryFDs representing this directory. children is protected by + // dirMu. + dirMu sync.Mutex + children map[string]*dentry + dirents []vfs.Dirent + + // upperVD and lowerVDs are the files from the overlay filesystem's layers + // that comprise the file on the overlay filesystem. + // + // If !upperVD.Ok(), it can transition to a valid vfs.VirtualDentry (i.e. + // be copied up) with copyMu locked for writing; otherwise, it is + // immutable. lowerVDs is always immutable. + copyMu sync.RWMutex + upperVD vfs.VirtualDentry + lowerVDs []vfs.VirtualDentry + + // inlineLowerVDs backs lowerVDs in the common case where len(lowerVDs) <= + // len(inlineLowerVDs). + inlineLowerVDs [1]vfs.VirtualDentry + + // devMajor, devMinor, and ino are the device major/minor and inode numbers + // used by this dentry. These fields are protected by copyMu and accessed + // using atomic memory operations. + devMajor uint32 + devMinor uint32 + ino uint64 + + locks lock.FileLocks +} + +// newDentry creates a new dentry. The dentry initially has no references; it +// is the caller's responsibility to set the dentry's reference count and/or +// call dentry.destroy() as appropriate. The dentry is initially invalid in +// that it contains no layers; the caller is responsible for setting them. +func (fs *filesystem) newDentry() *dentry { + d := &dentry{ + fs: fs, + } + d.lowerVDs = d.inlineLowerVDs[:0] + d.vfsd.Init(d) + return d +} + +// IncRef implements vfs.DentryImpl.IncRef. +func (d *dentry) IncRef() { + // d.refs may be 0 if d.fs.renameMu is locked, which serializes against + // d.checkDropLocked(). + atomic.AddInt64(&d.refs, 1) +} + +// TryIncRef implements vfs.DentryImpl.TryIncRef. +func (d *dentry) TryIncRef() bool { + for { + refs := atomic.LoadInt64(&d.refs) + if refs <= 0 { + return false + } + if atomic.CompareAndSwapInt64(&d.refs, refs, refs+1) { + return true + } + } +} + +// DecRef implements vfs.DentryImpl.DecRef. +func (d *dentry) DecRef() { + if refs := atomic.AddInt64(&d.refs, -1); refs == 0 { + d.fs.renameMu.Lock() + d.checkDropLocked() + d.fs.renameMu.Unlock() + } else if refs < 0 { + panic("overlay.dentry.DecRef() called without holding a reference") + } +} + +// checkDropLocked should be called after d's reference count becomes 0 or it +// becomes deleted. +// +// Preconditions: d.fs.renameMu must be locked for writing. +func (d *dentry) checkDropLocked() { + // Dentries with a positive reference count must be retained. (The only way + // to obtain a reference on a dentry with zero references is via path + // resolution, which requires renameMu, so if d.refs is zero then it will + // remain zero while we hold renameMu for writing.) Dentries with a + // negative reference count have already been destroyed. + if atomic.LoadInt64(&d.refs) != 0 { + return + } + // Refs is still zero; destroy it. + d.destroyLocked() + return +} + +// destroyLocked destroys the dentry. +// +// Preconditions: d.fs.renameMu must be locked for writing. d.refs == 0. +func (d *dentry) destroyLocked() { + switch atomic.LoadInt64(&d.refs) { + case 0: + // Mark the dentry destroyed. + atomic.StoreInt64(&d.refs, -1) + case -1: + panic("overlay.dentry.destroyLocked() called on already destroyed dentry") + default: + panic("overlay.dentry.destroyLocked() called with references on the dentry") + } + + if d.upperVD.Ok() { + d.upperVD.DecRef() + } + for _, lowerVD := range d.lowerVDs { + lowerVD.DecRef() + } + + if d.parent != nil { + d.parent.dirMu.Lock() + if !d.vfsd.IsDead() { + delete(d.parent.children, d.name) + } + d.parent.dirMu.Unlock() + // Drop the reference held by d on its parent without recursively + // locking d.fs.renameMu. + if refs := atomic.AddInt64(&d.parent.refs, -1); refs == 0 { + d.parent.checkDropLocked() + } else if refs < 0 { + panic("overlay.dentry.DecRef() called without holding a reference") + } + } +} + +// InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. +func (d *dentry) InotifyWithParent(events uint32, cookie uint32, et vfs.EventType) { + // TODO(gvisor.dev/issue/1479): Implement inotify. +} + +// Watches implements vfs.DentryImpl.Watches. +func (d *dentry) Watches() *vfs.Watches { + // TODO(gvisor.dev/issue/1479): Implement inotify. + return nil +} + +// iterLayers invokes yield on each layer comprising d, from top to bottom. If +// any call to yield returns false, iterLayer stops iteration. +func (d *dentry) iterLayers(yield func(vd vfs.VirtualDentry, isUpper bool) bool) { + if d.isCopiedUp() { + if !yield(d.upperVD, true) { + return + } + } + for _, lowerVD := range d.lowerVDs { + if !yield(lowerVD, false) { + return + } + } +} + +func (d *dentry) topLayerInfo() (vd vfs.VirtualDentry, isUpper bool) { + if d.isCopiedUp() { + return d.upperVD, true + } + return d.lowerVDs[0], false +} + +func (d *dentry) topLayer() vfs.VirtualDentry { + vd, _ := d.topLayerInfo() + return vd +} + +func (d *dentry) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes) error { + return vfs.GenericCheckPermissions(creds, ats, linux.FileMode(atomic.LoadUint32(&d.mode)), auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))) +} + +// statInternalMask is the set of stat fields that is set by +// dentry.statInternalTo(). +const statInternalMask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_UID | linux.STATX_GID | linux.STATX_INO + +// statInternalTo writes fields to stat that are stored in d, and therefore do +// not requiring invoking StatAt on the overlay's layers. +func (d *dentry) statInternalTo(ctx context.Context, opts *vfs.StatOptions, stat *linux.Statx) { + stat.Mask |= statInternalMask + if d.isDir() { + // Linux sets nlink to 1 for merged directories + // (fs/overlayfs/inode.c:ovl_getattr()); we set it to 2 because this is + // correct more often ("." and the directory's entry in its parent), + // and some of our tests expect this. + stat.Nlink = 2 + } + stat.UID = atomic.LoadUint32(&d.uid) + stat.GID = atomic.LoadUint32(&d.gid) + stat.Mode = uint16(atomic.LoadUint32(&d.mode)) + stat.Ino = atomic.LoadUint64(&d.ino) + stat.DevMajor = atomic.LoadUint32(&d.devMajor) + stat.DevMinor = atomic.LoadUint32(&d.devMinor) +} + +// Preconditions: d.copyMu must be locked for writing. +func (d *dentry) updateAfterSetStatLocked(opts *vfs.SetStatOptions) { + if opts.Stat.Mask&linux.STATX_MODE != 0 { + atomic.StoreUint32(&d.mode, (d.mode&linux.S_IFMT)|uint32(opts.Stat.Mode&^linux.S_IFMT)) + } + if opts.Stat.Mask&linux.STATX_UID != 0 { + atomic.StoreUint32(&d.uid, opts.Stat.UID) + } + if opts.Stat.Mask&linux.STATX_GID != 0 { + atomic.StoreUint32(&d.gid, opts.Stat.GID) + } +} + +// fileDescription is embedded by overlay implementations of +// vfs.FileDescriptionImpl. +type fileDescription struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl + vfs.LockFD +} + +func (fd *fileDescription) filesystem() *filesystem { + return fd.vfsfd.Mount().Filesystem().Impl().(*filesystem) +} + +func (fd *fileDescription) dentry() *dentry { + return fd.vfsfd.Dentry().Impl().(*dentry) +} diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 97b9b18d7..13c48824e 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -108,6 +108,10 @@ type FileDescriptionOptions struct { UseDentryMetadata bool } +// FileCreationFlags are the set of flags passed to FileDescription.Init() but +// omitted from FileDescription.StatusFlags(). +const FileCreationFlags = linux.O_CREAT | linux.O_EXCL | linux.O_NOCTTY | linux.O_TRUNC + // Init must be called before first use of fd. If it succeeds, it takes // references on mnt and d. flags is the initial file description flags, which // is usually the full set of flags passed to open(2). @@ -122,8 +126,8 @@ func (fd *FileDescription) Init(impl FileDescriptionImpl, flags uint32, mnt *Mou fd.refs = 1 // Remove "file creation flags" to mirror the behavior from file.f_flags in - // fs/open.c:do_dentry_open - fd.statusFlags = flags &^ (linux.O_CREAT | linux.O_EXCL | linux.O_NOCTTY | linux.O_TRUNC) + // fs/open.c:do_dentry_open. + fd.statusFlags = flags &^ FileCreationFlags fd.vd = VirtualDentry{ mount: mnt, dentry: d, @@ -471,6 +475,15 @@ type IterDirentsCallback interface { Handle(dirent Dirent) error } +// IterDirentsCallbackFunc implements IterDirentsCallback for a function with +// the semantics of IterDirentsCallback.Handle. +type IterDirentsCallbackFunc func(dirent Dirent) error + +// Handle implements IterDirentsCallback.Handle. +func (f IterDirentsCallbackFunc) Handle(dirent Dirent) error { + return f(dirent) +} + // OnClose is called when a file descriptor representing the FileDescription is // closed. Returning a non-nil error should not prevent the file descriptor // from being closed. diff --git a/pkg/syserror/syserror.go b/pkg/syserror/syserror.go index f86db0999..c73072c42 100644 --- a/pkg/syserror/syserror.go +++ b/pkg/syserror/syserror.go @@ -72,6 +72,7 @@ var ( EPERM = error(syscall.EPERM) EPIPE = error(syscall.EPIPE) ERANGE = error(syscall.ERANGE) + EREMOTE = error(syscall.EREMOTE) EROFS = error(syscall.EROFS) ESPIPE = error(syscall.ESPIPE) ESRCH = error(syscall.ESRCH) diff --git a/runsc/boot/BUILD b/runsc/boot/BUILD index a907c103b..2b1e6b13e 100644 --- a/runsc/boot/BUILD +++ b/runsc/boot/BUILD @@ -55,6 +55,7 @@ go_library( "//pkg/sentry/fsimpl/devtmpfs", "//pkg/sentry/fsimpl/gofer", "//pkg/sentry/fsimpl/host", + "//pkg/sentry/fsimpl/overlay", "//pkg/sentry/fsimpl/proc", "//pkg/sentry/fsimpl/sys", "//pkg/sentry/fsimpl/tmpfs", diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go index 8eeb43e79..d1653b279 100644 --- a/runsc/boot/vfs.go +++ b/runsc/boot/vfs.go @@ -30,6 +30,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fsimpl/devpts" "gvisor.dev/gvisor/pkg/sentry/fsimpl/devtmpfs" "gvisor.dev/gvisor/pkg/sentry/fsimpl/gofer" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/overlay" "gvisor.dev/gvisor/pkg/sentry/fsimpl/proc" "gvisor.dev/gvisor/pkg/sentry/fsimpl/sys" "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" @@ -53,6 +54,10 @@ func registerFilesystems(ctx context.Context, vfsObj *vfs.VirtualFilesystem, cre vfsObj.MustRegisterFilesystemType(gofer.Name, &gofer.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserList: true, }) + vfsObj.MustRegisterFilesystemType(overlay.Name, &overlay.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + AllowUserList: true, + }) vfsObj.MustRegisterFilesystemType(proc.Name, &proc.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, AllowUserList: true, -- cgit v1.2.3 From b504a0b82b81330067e2dd64d5fc626104e3b4ed Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Fri, 12 Jun 2020 03:38:44 -0400 Subject: vfs-arm64-1:add vfs2-syscall support to Arm64 Add vfs2-syscall support to Arm64 Signed-off-by: Bin Lu --- pkg/sentry/syscalls/linux/vfs2/vfs2.go | 102 ++++++++++++++++++++++++++++++++- 1 file changed, 101 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2.go b/pkg/sentry/syscalls/linux/vfs2/vfs2.go index 954c82f97..428a95fbc 100644 --- a/pkg/sentry/syscalls/linux/vfs2/vfs2.go +++ b/pkg/sentry/syscalls/linux/vfs2/vfs2.go @@ -62,7 +62,7 @@ func Override() { s.Table[55] = syscalls.Supported("getsockopt", GetSockOpt) s.Table[59] = syscalls.Supported("execve", Execve) s.Table[72] = syscalls.Supported("fcntl", Fcntl) - s.Table[73] = syscalls.Supported("fcntl", Flock) + s.Table[73] = syscalls.Supported("flock", Flock) s.Table[74] = syscalls.Supported("fsync", Fsync) s.Table[75] = syscalls.Supported("fdatasync", Fdatasync) s.Table[76] = syscalls.Supported("truncate", Truncate) @@ -167,6 +167,106 @@ func Override() { // Override ARM64. s = linux.ARM64 + s.Table[5] = syscalls.Supported("setxattr", Setxattr) + s.Table[6] = syscalls.Supported("lsetxattr", Lsetxattr) + s.Table[7] = syscalls.Supported("fsetxattr", Fsetxattr) + s.Table[8] = syscalls.Supported("getxattr", Getxattr) + s.Table[9] = syscalls.Supported("lgetxattr", Lgetxattr) + s.Table[10] = syscalls.Supported("fgetxattr", Fgetxattr) + s.Table[11] = syscalls.Supported("listxattr", Listxattr) + s.Table[12] = syscalls.Supported("llistxattr", Llistxattr) + s.Table[13] = syscalls.Supported("flistxattr", Flistxattr) + s.Table[14] = syscalls.Supported("removexattr", Removexattr) + s.Table[15] = syscalls.Supported("lremovexattr", Lremovexattr) + s.Table[16] = syscalls.Supported("fremovexattr", Fremovexattr) + s.Table[17] = syscalls.Supported("getcwd", Getcwd) + s.Table[19] = syscalls.Supported("eventfd2", Eventfd2) + s.Table[20] = syscalls.Supported("epoll_create1", EpollCreate1) + s.Table[21] = syscalls.Supported("epoll_ctl", EpollCtl) + s.Table[22] = syscalls.Supported("epoll_pwait", EpollPwait) + s.Table[23] = syscalls.Supported("dup", Dup) + s.Table[24] = syscalls.Supported("dup3", Dup3) + s.Table[25] = syscalls.Supported("fcntl", Fcntl) + s.Table[26] = syscalls.PartiallySupported("inotify_init1", InotifyInit1, "inotify events are only available inside the sandbox.", nil) + s.Table[27] = syscalls.PartiallySupported("inotify_add_watch", InotifyAddWatch, "inotify events are only available inside the sandbox.", nil) + s.Table[28] = syscalls.PartiallySupported("inotify_rm_watch", InotifyRmWatch, "inotify events are only available inside the sandbox.", nil) + s.Table[29] = syscalls.Supported("ioctl", Ioctl) + s.Table[32] = syscalls.Supported("flock", Flock) + s.Table[33] = syscalls.Supported("mknodat", Mknodat) + s.Table[34] = syscalls.Supported("mkdirat", Mkdirat) + s.Table[35] = syscalls.Supported("unlinkat", Unlinkat) + s.Table[36] = syscalls.Supported("symlinkat", Symlinkat) + s.Table[37] = syscalls.Supported("linkat", Linkat) + s.Table[38] = syscalls.Supported("renameat", Renameat) + s.Table[39] = syscalls.Supported("umount2", Umount2) + s.Table[40] = syscalls.Supported("mount", Mount) + s.Table[43] = syscalls.Supported("statfs", Statfs) + s.Table[44] = syscalls.Supported("fstatfs", Fstatfs) + s.Table[45] = syscalls.Supported("truncate", Truncate) + s.Table[46] = syscalls.Supported("ftruncate", Ftruncate) + s.Table[48] = syscalls.Supported("faccessat", Faccessat) + s.Table[49] = syscalls.Supported("chdir", Chdir) + s.Table[50] = syscalls.Supported("fchdir", Fchdir) + s.Table[51] = syscalls.Supported("chroot", Chroot) + s.Table[52] = syscalls.Supported("fchmod", Fchmod) + s.Table[53] = syscalls.Supported("fchmodat", Fchmodat) + s.Table[54] = syscalls.Supported("fchownat", Fchownat) + s.Table[55] = syscalls.Supported("fchown", Fchown) + s.Table[56] = syscalls.Supported("openat", Openat) + s.Table[57] = syscalls.Supported("close", Close) + s.Table[59] = syscalls.Supported("pipe2", Pipe2) + s.Table[61] = syscalls.Supported("getdents64", Getdents64) + s.Table[62] = syscalls.Supported("lseek", Lseek) s.Table[63] = syscalls.Supported("read", Read) + s.Table[64] = syscalls.Supported("write", Write) + s.Table[65] = syscalls.Supported("readv", Readv) + s.Table[66] = syscalls.Supported("writev", Writev) + s.Table[67] = syscalls.Supported("pread64", Pread64) + s.Table[68] = syscalls.Supported("pwrite64", Pwrite64) + s.Table[69] = syscalls.Supported("preadv", Preadv) + s.Table[70] = syscalls.Supported("pwritev", Pwritev) + s.Table[72] = syscalls.Supported("pselect", Pselect) + s.Table[73] = syscalls.Supported("ppoll", Ppoll) + s.Table[74] = syscalls.Supported("signalfd4", Signalfd4) + s.Table[76] = syscalls.Supported("splice", Splice) + s.Table[77] = syscalls.Supported("tee", Tee) + s.Table[78] = syscalls.Supported("readlinkat", Readlinkat) + s.Table[80] = syscalls.Supported("fstat", Fstat) + s.Table[81] = syscalls.Supported("sync", Sync) + s.Table[82] = syscalls.Supported("fsync", Fsync) + s.Table[83] = syscalls.Supported("fdatasync", Fdatasync) + s.Table[84] = syscalls.Supported("sync_file_range", SyncFileRange) + s.Table[85] = syscalls.Supported("timerfd_create", TimerfdCreate) + s.Table[86] = syscalls.Supported("timerfd_settime", TimerfdSettime) + s.Table[87] = syscalls.Supported("timerfd_gettime", TimerfdGettime) + s.Table[88] = syscalls.Supported("utimensat", Utimensat) + s.Table[198] = syscalls.Supported("socket", Socket) + s.Table[199] = syscalls.Supported("socketpair", SocketPair) + s.Table[200] = syscalls.Supported("bind", Bind) + s.Table[201] = syscalls.Supported("listen", Listen) + s.Table[202] = syscalls.Supported("accept", Accept) + s.Table[203] = syscalls.Supported("connect", Connect) + s.Table[204] = syscalls.Supported("getsockname", GetSockName) + s.Table[205] = syscalls.Supported("getpeername", GetPeerName) + s.Table[206] = syscalls.Supported("sendto", SendTo) + s.Table[207] = syscalls.Supported("recvfrom", RecvFrom) + s.Table[208] = syscalls.Supported("setsockopt", SetSockOpt) + s.Table[209] = syscalls.Supported("getsockopt", GetSockOpt) + s.Table[210] = syscalls.Supported("shutdown", Shutdown) + s.Table[211] = syscalls.Supported("sendmsg", SendMsg) + s.Table[212] = syscalls.Supported("recvmsg", RecvMsg) + s.Table[221] = syscalls.Supported("execve", Execve) + s.Table[222] = syscalls.Supported("mmap", Mmap) + s.Table[242] = syscalls.Supported("accept4", Accept4) + s.Table[243] = syscalls.Supported("recvmmsg", RecvMMsg) + s.Table[267] = syscalls.Supported("syncfs", Syncfs) + s.Table[269] = syscalls.Supported("sendmmsg", SendMMsg) + s.Table[276] = syscalls.Supported("renameat2", Renameat2) + s.Table[279] = syscalls.Supported("memfd_create", MemfdCreate) + s.Table[281] = syscalls.Supported("execveat", Execveat) + s.Table[286] = syscalls.Supported("preadv2", Preadv2) + s.Table[287] = syscalls.Supported("pwritev2", Pwritev2) + s.Table[291] = syscalls.Supported("statx", Statx) + s.Init() } -- cgit v1.2.3 From 6ec9d60403fdf7a33072eaa023e62bfd56ed9f5c Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Fri, 12 Jun 2020 11:56:43 -0700 Subject: vfs2: implement fcntl(fd, F_SETFL, flags) PiperOrigin-RevId: 316148074 --- pkg/sentry/kernel/fd_table.go | 23 +++++++++++++++++++++++ pkg/sentry/syscalls/linux/sys_file.go | 4 ++-- pkg/sentry/syscalls/linux/vfs2/fd.go | 4 ++-- test/syscalls/linux/fcntl.cc | 9 +++++++++ 4 files changed, 36 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/fd_table.go b/pkg/sentry/kernel/fd_table.go index b35afafe3..48911240f 100644 --- a/pkg/sentry/kernel/fd_table.go +++ b/pkg/sentry/kernel/fd_table.go @@ -458,6 +458,29 @@ func (f *FDTable) SetFlags(fd int32, flags FDFlags) error { return nil } +// SetFlagsVFS2 sets the flags for the given file descriptor. +// +// True is returned iff flags were changed. +func (f *FDTable) SetFlagsVFS2(fd int32, flags FDFlags) error { + if fd < 0 { + // Don't accept negative FDs. + return syscall.EBADF + } + + f.mu.Lock() + defer f.mu.Unlock() + + file, _, _ := f.getVFS2(fd) + if file == nil { + // No file found. + return syscall.EBADF + } + + // Update the flags. + f.setVFS2(fd, file, flags) + return nil +} + // Get returns a reference to the file and the flags for the FD or nil if no // file is defined for the given fd. // diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go index 8347617bd..696e1c8d3 100644 --- a/pkg/sentry/syscalls/linux/sys_file.go +++ b/pkg/sentry/syscalls/linux/sys_file.go @@ -935,10 +935,10 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return uintptr(flags.ToLinuxFDFlags()), nil, nil case linux.F_SETFD: flags := args[2].Uint() - t.FDTable().SetFlags(fd, kernel.FDFlags{ + err := t.FDTable().SetFlags(fd, kernel.FDFlags{ CloseOnExec: flags&linux.FD_CLOEXEC != 0, }) - return 0, nil, nil + return 0, nil, err case linux.F_GETFL: return uintptr(file.Flags().ToLinux()), nil, nil case linux.F_SETFL: diff --git a/pkg/sentry/syscalls/linux/vfs2/fd.go b/pkg/sentry/syscalls/linux/vfs2/fd.go index 6006758a5..f9ccb303c 100644 --- a/pkg/sentry/syscalls/linux/vfs2/fd.go +++ b/pkg/sentry/syscalls/linux/vfs2/fd.go @@ -134,10 +134,10 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return uintptr(flags.ToLinuxFDFlags()), nil, nil case linux.F_SETFD: flags := args[2].Uint() - t.FDTable().SetFlags(fd, kernel.FDFlags{ + err := t.FDTable().SetFlagsVFS2(fd, kernel.FDFlags{ CloseOnExec: flags&linux.FD_CLOEXEC != 0, }) - return 0, nil, nil + return 0, nil, err case linux.F_GETFL: return uintptr(file.StatusFlags()), nil, nil case linux.F_SETFL: diff --git a/test/syscalls/linux/fcntl.cc b/test/syscalls/linux/fcntl.cc index c7cc5816e..35e8a4ff3 100644 --- a/test/syscalls/linux/fcntl.cc +++ b/test/syscalls/linux/fcntl.cc @@ -115,6 +115,15 @@ PosixErrorOr SubprocessLock(std::string const& path, bool for_write, return std::move(cleanup); } +TEST(FcntlTest, SetCloExecBadFD) { + // Open an eventfd file descriptor with FD_CLOEXEC descriptor flag not set. + FileDescriptor f = ASSERT_NO_ERRNO_AND_VALUE(NewEventFD(0, 0)); + auto fd = f.get(); + f.reset(); + ASSERT_THAT(fcntl(fd, F_GETFD), SyscallFailsWithErrno(EBADF)); + ASSERT_THAT(fcntl(fd, F_SETFD, FD_CLOEXEC), SyscallFailsWithErrno(EBADF)); +} + TEST(FcntlTest, SetCloExec) { // Open an eventfd file descriptor with FD_CLOEXEC descriptor flag not set. FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NewEventFD(0, 0)); -- cgit v1.2.3 From 3b0b1f104d963a1d11973c444934e6744ab7e79b Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Tue, 16 Jun 2020 00:14:07 -0700 Subject: Miscellaneous VFS2 fixes. PiperOrigin-RevId: 316627764 --- pkg/sentry/fsimpl/gofer/gofer.go | 26 ++++++++++++++++---------- pkg/sentry/kernel/kernel.go | 2 +- pkg/sentry/syscalls/linux/vfs2/ioctl.go | 29 +++++++++++++++++++++++++++++ 3 files changed, 46 insertions(+), 11 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 0d88a328e..ac051b3a7 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -835,6 +835,14 @@ func (d *dentry) statTo(stat *linux.Statx) { stat.Mask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_NLINK | linux.STATX_UID | linux.STATX_GID | linux.STATX_ATIME | linux.STATX_MTIME | linux.STATX_CTIME | linux.STATX_INO | linux.STATX_SIZE | linux.STATX_BLOCKS | linux.STATX_BTIME stat.Blksize = atomic.LoadUint32(&d.blockSize) stat.Nlink = atomic.LoadUint32(&d.nlink) + if stat.Nlink == 0 { + // The remote filesystem doesn't support link count; just make + // something up. This is consistent with Linux, where + // fs/inode.c:inode_init_always() initializes link count to 1, and + // fs/9p/vfs_inode_dotl.c:v9fs_stat2inode_dotl() doesn't touch it if + // it's not provided by the remote filesystem. + stat.Nlink = 1 + } stat.UID = atomic.LoadUint32(&d.uid) stat.GID = atomic.LoadUint32(&d.gid) stat.Mode = uint16(atomic.LoadUint32(&d.mode)) @@ -1346,23 +1354,21 @@ func (d *dentry) ensureSharedHandle(ctx context.Context, read, write, trunc bool } // incLinks increments link count. -// -// Preconditions: d.nlink != 0 && d.nlink < math.MaxUint32. func (d *dentry) incLinks() { - v := atomic.AddUint32(&d.nlink, 1) - if v < 2 { - panic(fmt.Sprintf("dentry.nlink is invalid (was 0 or overflowed): %d", v)) + if atomic.LoadUint32(&d.nlink) == 0 { + // The remote filesystem doesn't support link count. + return } + atomic.AddUint32(&d.nlink, 1) } // decLinks decrements link count. -// -// Preconditions: d.nlink > 1. func (d *dentry) decLinks() { - v := atomic.AddUint32(&d.nlink, ^uint32(0)) - if v == 0 { - panic(fmt.Sprintf("dentry.nlink must be greater than 0: %d", v)) + if atomic.LoadUint32(&d.nlink) == 0 { + // The remote filesystem doesn't support link count. + return } + atomic.AddUint32(&d.nlink, ^uint32(0)) } // fileDescription is embedded by gofer implementations of diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index bcbeb6a39..52491da7a 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -892,7 +892,7 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, if mntnsVFS2 == nil { // MountNamespaceVFS2 adds a reference to the namespace, which is // transferred to the new process. - mntnsVFS2 = k.GlobalInit().Leader().MountNamespaceVFS2() + mntnsVFS2 = k.globalInit.Leader().MountNamespaceVFS2() } // Get the root directory from the MountNamespace. root := args.MountNamespaceVFS2.Root() diff --git a/pkg/sentry/syscalls/linux/vfs2/ioctl.go b/pkg/sentry/syscalls/linux/vfs2/ioctl.go index 5a2418da9..0399c0db4 100644 --- a/pkg/sentry/syscalls/linux/vfs2/ioctl.go +++ b/pkg/sentry/syscalls/linux/vfs2/ioctl.go @@ -15,6 +15,7 @@ package vfs2 import ( + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/syserror" @@ -30,6 +31,34 @@ func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } defer file.DecRef() + // Handle ioctls that apply to all FDs. + switch args[1].Int() { + case linux.FIONCLEX: + t.FDTable().SetFlagsVFS2(fd, kernel.FDFlags{ + CloseOnExec: false, + }) + return 0, nil, nil + + case linux.FIOCLEX: + t.FDTable().SetFlagsVFS2(fd, kernel.FDFlags{ + CloseOnExec: true, + }) + return 0, nil, nil + + case linux.FIONBIO: + var set int32 + if _, err := t.CopyIn(args[2].Pointer(), &set); err != nil { + return 0, nil, err + } + flags := file.StatusFlags() + if set != 0 { + flags |= linux.O_NONBLOCK + } else { + flags &^= linux.O_NONBLOCK + } + return 0, nil, file.SetStatusFlags(t, t.Credentials(), flags) + } + ret, err := file.Ioctl(t, t.MemoryManager(), args) return ret, nil, err } -- cgit v1.2.3 From 810748f5c9c72f713d81d14bcc89a8eb4ca49eb6 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Tue, 16 Jun 2020 08:47:04 -0700 Subject: Port aio to VFS2. In order to make sure all aio goroutines have stopped during S/R, a new WaitGroup was added to TaskSet, analagous to runningGoroutines. This WaitGroup is incremented with each aio goroutine, and waited on during kernel.Pause. The old VFS1 aio code was changed to use this new WaitGroup, rather than fs.Async. The only uses of fs.Async are now inode and mount Release operations, which do not call fs.Async recursively. This fixes a lock-ordering violation that can cause deadlocks. Updates #1035. PiperOrigin-RevId: 316689380 --- pkg/abi/linux/aio.go | 60 ++++++++- pkg/sentry/kernel/BUILD | 1 + pkg/sentry/kernel/aio.go | 81 +++++++++++++ pkg/sentry/kernel/context.go | 53 -------- pkg/sentry/kernel/kernel.go | 10 +- pkg/sentry/kernel/threads.go | 7 ++ pkg/sentry/syscalls/linux/sys_aio.go | 169 +++++++++----------------- pkg/sentry/syscalls/linux/vfs2/BUILD | 3 + pkg/sentry/syscalls/linux/vfs2/aio.go | 216 +++++++++++++++++++++++++++++++++ pkg/sentry/syscalls/linux/vfs2/vfs2.go | 6 +- 10 files changed, 431 insertions(+), 175 deletions(-) create mode 100644 pkg/sentry/kernel/aio.go create mode 100644 pkg/sentry/syscalls/linux/vfs2/aio.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/aio.go b/pkg/abi/linux/aio.go index 3c6e0079d..86ee3f8b5 100644 --- a/pkg/abi/linux/aio.go +++ b/pkg/abi/linux/aio.go @@ -14,7 +14,63 @@ package linux +import "encoding/binary" + +// AIORingSize is sizeof(struct aio_ring). +const AIORingSize = 32 + +// I/O commands. const ( - // AIORingSize is sizeof(struct aio_ring). - AIORingSize = 32 + IOCB_CMD_PREAD = 0 + IOCB_CMD_PWRITE = 1 + IOCB_CMD_FSYNC = 2 + IOCB_CMD_FDSYNC = 3 + // 4 was the experimental IOCB_CMD_PREADX. + IOCB_CMD_POLL = 5 + IOCB_CMD_NOOP = 6 + IOCB_CMD_PREADV = 7 + IOCB_CMD_PWRITEV = 8 ) + +// I/O flags. +const ( + IOCB_FLAG_RESFD = 1 + IOCB_FLAG_IOPRIO = 2 +) + +// IOCallback describes an I/O request. +// +// The priority field is currently ignored in the implementation below. Also +// note that the IOCB_FLAG_RESFD feature is not supported. +type IOCallback struct { + Data uint64 + Key uint32 + _ uint32 + + OpCode uint16 + ReqPrio int16 + FD int32 + + Buf uint64 + Bytes uint64 + Offset int64 + + Reserved2 uint64 + Flags uint32 + + // eventfd to signal if IOCB_FLAG_RESFD is set in flags. + ResFD int32 +} + +// IOEvent describes an I/O result. +// +// +stateify savable +type IOEvent struct { + Data uint64 + Obj uint64 + Result int64 + Result2 int64 +} + +// IOEventSize is the size of an ioEvent encoded. +var IOEventSize = binary.Size(IOEvent{}) diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD index a28eab8b8..1510a7c26 100644 --- a/pkg/sentry/kernel/BUILD +++ b/pkg/sentry/kernel/BUILD @@ -85,6 +85,7 @@ go_library( name = "kernel", srcs = [ "abstract_socket_namespace.go", + "aio.go", "context.go", "fd_table.go", "fd_table_unsafe.go", diff --git a/pkg/sentry/kernel/aio.go b/pkg/sentry/kernel/aio.go new file mode 100644 index 000000000..0ac78c0b8 --- /dev/null +++ b/pkg/sentry/kernel/aio.go @@ -0,0 +1,81 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package kernel + +import ( + "time" + + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/log" +) + +// AIOCallback is an function that does asynchronous I/O on behalf of a task. +type AIOCallback func(context.Context) + +// QueueAIO queues an AIOCallback which will be run asynchronously. +func (t *Task) QueueAIO(cb AIOCallback) { + ctx := taskAsyncContext{t: t} + wg := &t.TaskSet().aioGoroutines + wg.Add(1) + go func() { + cb(ctx) + wg.Done() + }() +} + +type taskAsyncContext struct { + context.NoopSleeper + t *Task +} + +// Debugf implements log.Logger.Debugf. +func (ctx taskAsyncContext) Debugf(format string, v ...interface{}) { + ctx.t.Debugf(format, v...) +} + +// Infof implements log.Logger.Infof. +func (ctx taskAsyncContext) Infof(format string, v ...interface{}) { + ctx.t.Infof(format, v...) +} + +// Warningf implements log.Logger.Warningf. +func (ctx taskAsyncContext) Warningf(format string, v ...interface{}) { + ctx.t.Warningf(format, v...) +} + +// IsLogging implements log.Logger.IsLogging. +func (ctx taskAsyncContext) IsLogging(level log.Level) bool { + return ctx.t.IsLogging(level) +} + +// Deadline implements context.Context.Deadline. +func (ctx taskAsyncContext) Deadline() (time.Time, bool) { + return ctx.t.Deadline() +} + +// Done implements context.Context.Done. +func (ctx taskAsyncContext) Done() <-chan struct{} { + return ctx.t.Done() +} + +// Err implements context.Context.Err. +func (ctx taskAsyncContext) Err() error { + return ctx.t.Err() +} + +// Value implements context.Context.Value. +func (ctx taskAsyncContext) Value(key interface{}) interface{} { + return ctx.t.Value(key) +} diff --git a/pkg/sentry/kernel/context.go b/pkg/sentry/kernel/context.go index 0c40bf315..dd5f0f5fa 100644 --- a/pkg/sentry/kernel/context.go +++ b/pkg/sentry/kernel/context.go @@ -18,7 +18,6 @@ import ( "time" "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/log" ) // contextID is the kernel package's type for context.Context.Value keys. @@ -113,55 +112,3 @@ func (*Task) Done() <-chan struct{} { func (*Task) Err() error { return nil } - -// AsyncContext returns a context.Context that may be used by goroutines that -// do work on behalf of t and therefore share its contextual values, but are -// not t's task goroutine (e.g. asynchronous I/O). -func (t *Task) AsyncContext() context.Context { - return taskAsyncContext{t: t} -} - -type taskAsyncContext struct { - context.NoopSleeper - t *Task -} - -// Debugf implements log.Logger.Debugf. -func (ctx taskAsyncContext) Debugf(format string, v ...interface{}) { - ctx.t.Debugf(format, v...) -} - -// Infof implements log.Logger.Infof. -func (ctx taskAsyncContext) Infof(format string, v ...interface{}) { - ctx.t.Infof(format, v...) -} - -// Warningf implements log.Logger.Warningf. -func (ctx taskAsyncContext) Warningf(format string, v ...interface{}) { - ctx.t.Warningf(format, v...) -} - -// IsLogging implements log.Logger.IsLogging. -func (ctx taskAsyncContext) IsLogging(level log.Level) bool { - return ctx.t.IsLogging(level) -} - -// Deadline implements context.Context.Deadline. -func (ctx taskAsyncContext) Deadline() (time.Time, bool) { - return ctx.t.Deadline() -} - -// Done implements context.Context.Done. -func (ctx taskAsyncContext) Done() <-chan struct{} { - return ctx.t.Done() -} - -// Err implements context.Context.Err. -func (ctx taskAsyncContext) Err() error { - return ctx.t.Err() -} - -// Value implements context.Context.Value. -func (ctx taskAsyncContext) Value(key interface{}) interface{} { - return ctx.t.Value(key) -} diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 52491da7a..554a42e05 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -452,9 +452,7 @@ func (k *Kernel) SaveTo(w io.Writer) error { return err } - // Ensure that all pending asynchronous work is complete: - // - inode and mount release - // - asynchronuous IO + // Ensure that all inode and mount release operations have completed. fs.AsyncBarrier() // Once all fs work has completed (flushed references have all been released), @@ -1249,13 +1247,15 @@ func (k *Kernel) Kill(es ExitStatus) { } // Pause requests that all tasks in k temporarily stop executing, and blocks -// until all tasks in k have stopped. Multiple calls to Pause nest and require -// an equal number of calls to Unpause to resume execution. +// until all tasks and asynchronous I/O operations in k have stopped. Multiple +// calls to Pause nest and require an equal number of calls to Unpause to +// resume execution. func (k *Kernel) Pause() { k.extMu.Lock() k.tasks.BeginExternalStop() k.extMu.Unlock() k.tasks.runningGoroutines.Wait() + k.tasks.aioGoroutines.Wait() } // Unpause ends the effect of a previous call to Pause. If Unpause is called diff --git a/pkg/sentry/kernel/threads.go b/pkg/sentry/kernel/threads.go index bf2dabb6e..872e1a82d 100644 --- a/pkg/sentry/kernel/threads.go +++ b/pkg/sentry/kernel/threads.go @@ -87,6 +87,13 @@ type TaskSet struct { // at time of save (but note that this is not necessarily the same thing as // sync.WaitGroup's zero value). runningGoroutines sync.WaitGroup `state:"nosave"` + + // aioGoroutines is the number of goroutines running async I/O + // callbacks. + // + // aioGoroutines is not saved but is required to be zero at the time of + // save. + aioGoroutines sync.WaitGroup `state:"nosave"` } // newTaskSet returns a new, empty TaskSet. diff --git a/pkg/sentry/syscalls/linux/sys_aio.go b/pkg/sentry/syscalls/linux/sys_aio.go index d781d6a04..ba2557c52 100644 --- a/pkg/sentry/syscalls/linux/sys_aio.go +++ b/pkg/sentry/syscalls/linux/sys_aio.go @@ -15,8 +15,8 @@ package linux import ( - "encoding/binary" - + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -27,59 +27,6 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) -// I/O commands. -const ( - _IOCB_CMD_PREAD = 0 - _IOCB_CMD_PWRITE = 1 - _IOCB_CMD_FSYNC = 2 - _IOCB_CMD_FDSYNC = 3 - _IOCB_CMD_NOOP = 6 - _IOCB_CMD_PREADV = 7 - _IOCB_CMD_PWRITEV = 8 -) - -// I/O flags. -const ( - _IOCB_FLAG_RESFD = 1 -) - -// ioCallback describes an I/O request. -// -// The priority field is currently ignored in the implementation below. Also -// note that the IOCB_FLAG_RESFD feature is not supported. -type ioCallback struct { - Data uint64 - Key uint32 - Reserved1 uint32 - - OpCode uint16 - ReqPrio int16 - FD int32 - - Buf uint64 - Bytes uint64 - Offset int64 - - Reserved2 uint64 - Flags uint32 - - // eventfd to signal if IOCB_FLAG_RESFD is set in flags. - ResFD int32 -} - -// ioEvent describes an I/O result. -// -// +stateify savable -type ioEvent struct { - Data uint64 - Obj uint64 - Result int64 - Result2 int64 -} - -// ioEventSize is the size of an ioEvent encoded. -var ioEventSize = binary.Size(ioEvent{}) - // IoSetup implements linux syscall io_setup(2). func IoSetup(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { nrEvents := args[0].Int() @@ -192,7 +139,7 @@ func IoGetevents(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S } } - ev := v.(*ioEvent) + ev := v.(*linux.IOEvent) // Copy out the result. if _, err := t.CopyOut(eventsAddr, ev); err != nil { @@ -204,7 +151,7 @@ func IoGetevents(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S } // Keep rolling. - eventsAddr += usermem.Addr(ioEventSize) + eventsAddr += usermem.Addr(linux.IOEventSize) } // Everything finished. @@ -231,7 +178,7 @@ func waitForRequest(ctx *mm.AIOContext, t *kernel.Task, haveDeadline bool, deadl } // memoryFor returns appropriate memory for the given callback. -func memoryFor(t *kernel.Task, cb *ioCallback) (usermem.IOSequence, error) { +func memoryFor(t *kernel.Task, cb *linux.IOCallback) (usermem.IOSequence, error) { bytes := int(cb.Bytes) if bytes < 0 { // Linux also requires that this field fit in ssize_t. @@ -242,17 +189,17 @@ func memoryFor(t *kernel.Task, cb *ioCallback) (usermem.IOSequence, error) { // we have no guarantee that t's AddressSpace will be active during the // I/O. switch cb.OpCode { - case _IOCB_CMD_PREAD, _IOCB_CMD_PWRITE: + case linux.IOCB_CMD_PREAD, linux.IOCB_CMD_PWRITE: return t.SingleIOSequence(usermem.Addr(cb.Buf), bytes, usermem.IOOpts{ AddressSpaceActive: false, }) - case _IOCB_CMD_PREADV, _IOCB_CMD_PWRITEV: + case linux.IOCB_CMD_PREADV, linux.IOCB_CMD_PWRITEV: return t.IovecsIOSequence(usermem.Addr(cb.Buf), bytes, usermem.IOOpts{ AddressSpaceActive: false, }) - case _IOCB_CMD_FSYNC, _IOCB_CMD_FDSYNC, _IOCB_CMD_NOOP: + case linux.IOCB_CMD_FSYNC, linux.IOCB_CMD_FDSYNC, linux.IOCB_CMD_NOOP: return usermem.IOSequence{}, nil default: @@ -261,54 +208,62 @@ func memoryFor(t *kernel.Task, cb *ioCallback) (usermem.IOSequence, error) { } } -func performCallback(t *kernel.Task, file *fs.File, cbAddr usermem.Addr, cb *ioCallback, ioseq usermem.IOSequence, ctx *mm.AIOContext, eventFile *fs.File) { - if ctx.Dead() { - ctx.CancelPendingRequest() - return - } - ev := &ioEvent{ - Data: cb.Data, - Obj: uint64(cbAddr), - } +// IoCancel implements linux syscall io_cancel(2). +// +// It is not presently supported (ENOSYS indicates no support on this +// architecture). +func IoCancel(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + return 0, nil, syserror.ENOSYS +} - // Construct a context.Context that will not be interrupted if t is - // interrupted. - c := t.AsyncContext() +// LINT.IfChange - var err error - switch cb.OpCode { - case _IOCB_CMD_PREAD, _IOCB_CMD_PREADV: - ev.Result, err = file.Preadv(c, ioseq, cb.Offset) - case _IOCB_CMD_PWRITE, _IOCB_CMD_PWRITEV: - ev.Result, err = file.Pwritev(c, ioseq, cb.Offset) - case _IOCB_CMD_FSYNC: - err = file.Fsync(c, 0, fs.FileMaxOffset, fs.SyncAll) - case _IOCB_CMD_FDSYNC: - err = file.Fsync(c, 0, fs.FileMaxOffset, fs.SyncData) - } +func getAIOCallback(t *kernel.Task, file *fs.File, cbAddr usermem.Addr, cb *linux.IOCallback, ioseq usermem.IOSequence, actx *mm.AIOContext, eventFile *fs.File) kernel.AIOCallback { + return func(ctx context.Context) { + if actx.Dead() { + actx.CancelPendingRequest() + return + } + ev := &linux.IOEvent{ + Data: cb.Data, + Obj: uint64(cbAddr), + } - // Update the result. - if err != nil { - err = handleIOError(t, ev.Result != 0 /* partial */, err, nil /* never interrupted */, "aio", file) - ev.Result = -int64(kernel.ExtractErrno(err, 0)) - } + var err error + switch cb.OpCode { + case linux.IOCB_CMD_PREAD, linux.IOCB_CMD_PREADV: + ev.Result, err = file.Preadv(ctx, ioseq, cb.Offset) + case linux.IOCB_CMD_PWRITE, linux.IOCB_CMD_PWRITEV: + ev.Result, err = file.Pwritev(ctx, ioseq, cb.Offset) + case linux.IOCB_CMD_FSYNC: + err = file.Fsync(ctx, 0, fs.FileMaxOffset, fs.SyncAll) + case linux.IOCB_CMD_FDSYNC: + err = file.Fsync(ctx, 0, fs.FileMaxOffset, fs.SyncData) + } + + // Update the result. + if err != nil { + err = handleIOError(t, ev.Result != 0 /* partial */, err, nil /* never interrupted */, "aio", file) + ev.Result = -int64(kernel.ExtractErrno(err, 0)) + } - file.DecRef() + file.DecRef() - // Queue the result for delivery. - ctx.FinishRequest(ev) + // Queue the result for delivery. + actx.FinishRequest(ev) - // Notify the event file if one was specified. This needs to happen - // *after* queueing the result to avoid racing with the thread we may - // wake up. - if eventFile != nil { - eventFile.FileOperations.(*eventfd.EventOperations).Signal(1) - eventFile.DecRef() + // Notify the event file if one was specified. This needs to happen + // *after* queueing the result to avoid racing with the thread we may + // wake up. + if eventFile != nil { + eventFile.FileOperations.(*eventfd.EventOperations).Signal(1) + eventFile.DecRef() + } } } // submitCallback processes a single callback. -func submitCallback(t *kernel.Task, id uint64, cb *ioCallback, cbAddr usermem.Addr) error { +func submitCallback(t *kernel.Task, id uint64, cb *linux.IOCallback, cbAddr usermem.Addr) error { file := t.GetFile(cb.FD) if file == nil { // File not found. @@ -318,7 +273,7 @@ func submitCallback(t *kernel.Task, id uint64, cb *ioCallback, cbAddr usermem.Ad // Was there an eventFD? Extract it. var eventFile *fs.File - if cb.Flags&_IOCB_FLAG_RESFD != 0 { + if cb.Flags&linux.IOCB_FLAG_RESFD != 0 { eventFile = t.GetFile(cb.ResFD) if eventFile == nil { // Bad FD. @@ -340,7 +295,7 @@ func submitCallback(t *kernel.Task, id uint64, cb *ioCallback, cbAddr usermem.Ad // Check offset for reads/writes. switch cb.OpCode { - case _IOCB_CMD_PREAD, _IOCB_CMD_PREADV, _IOCB_CMD_PWRITE, _IOCB_CMD_PWRITEV: + case linux.IOCB_CMD_PREAD, linux.IOCB_CMD_PREADV, linux.IOCB_CMD_PWRITE, linux.IOCB_CMD_PWRITEV: if cb.Offset < 0 { return syserror.EINVAL } @@ -366,7 +321,7 @@ func submitCallback(t *kernel.Task, id uint64, cb *ioCallback, cbAddr usermem.Ad // Perform the request asynchronously. file.IncRef() - fs.Async(func() { performCallback(t, file, cbAddr, cb, ioseq, ctx, eventFile) }) + t.QueueAIO(getAIOCallback(t, file, cbAddr, cb, ioseq, ctx, eventFile)) // All set. return nil @@ -395,7 +350,7 @@ func IoSubmit(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc } // Copy in this callback. - var cb ioCallback + var cb linux.IOCallback cbAddr := usermem.Addr(t.Arch().Value(cbAddrNative)) if _, err := t.CopyIn(cbAddr, &cb); err != nil { @@ -424,10 +379,4 @@ func IoSubmit(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc return uintptr(nrEvents), nil, nil } -// IoCancel implements linux syscall io_cancel(2). -// -// It is not presently supported (ENOSYS indicates no support on this -// architecture). -func IoCancel(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - return 0, nil, syserror.ENOSYS -} +// LINT.ThenChange(vfs2/aio.go) diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD index 9f93f4354..c301a0991 100644 --- a/pkg/sentry/syscalls/linux/vfs2/BUILD +++ b/pkg/sentry/syscalls/linux/vfs2/BUILD @@ -5,6 +5,7 @@ package(licenses = ["notice"]) go_library( name = "vfs2", srcs = [ + "aio.go", "epoll.go", "eventfd.go", "execve.go", @@ -40,6 +41,7 @@ go_library( "//pkg/abi/linux", "//pkg/binary", "//pkg/bits", + "//pkg/context", "//pkg/fspath", "//pkg/gohacks", "//pkg/sentry/arch", @@ -57,6 +59,7 @@ go_library( "//pkg/sentry/limits", "//pkg/sentry/loader", "//pkg/sentry/memmap", + "//pkg/sentry/mm", "//pkg/sentry/socket", "//pkg/sentry/socket/control", "//pkg/sentry/socket/unix/transport", diff --git a/pkg/sentry/syscalls/linux/vfs2/aio.go b/pkg/sentry/syscalls/linux/vfs2/aio.go new file mode 100644 index 000000000..e5cdefc50 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/aio.go @@ -0,0 +1,216 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/eventfd" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/mm" + slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +// IoSubmit implements linux syscall io_submit(2). +func IoSubmit(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + id := args[0].Uint64() + nrEvents := args[1].Int() + addr := args[2].Pointer() + + if nrEvents < 0 { + return 0, nil, syserror.EINVAL + } + + for i := int32(0); i < nrEvents; i++ { + // Copy in the address. + cbAddrNative := t.Arch().Native(0) + if _, err := t.CopyIn(addr, cbAddrNative); err != nil { + if i > 0 { + // Some successful. + return uintptr(i), nil, nil + } + // Nothing done. + return 0, nil, err + } + + // Copy in this callback. + var cb linux.IOCallback + cbAddr := usermem.Addr(t.Arch().Value(cbAddrNative)) + if _, err := t.CopyIn(cbAddr, &cb); err != nil { + if i > 0 { + // Some have been successful. + return uintptr(i), nil, nil + } + // Nothing done. + return 0, nil, err + } + + // Process this callback. + if err := submitCallback(t, id, &cb, cbAddr); err != nil { + if i > 0 { + // Partial success. + return uintptr(i), nil, nil + } + // Nothing done. + return 0, nil, err + } + + // Advance to the next one. + addr += usermem.Addr(t.Arch().Width()) + } + + return uintptr(nrEvents), nil, nil +} + +// submitCallback processes a single callback. +func submitCallback(t *kernel.Task, id uint64, cb *linux.IOCallback, cbAddr usermem.Addr) error { + if cb.Reserved2 != 0 { + return syserror.EINVAL + } + + fd := t.GetFileVFS2(cb.FD) + if fd == nil { + return syserror.EBADF + } + defer fd.DecRef() + + // Was there an eventFD? Extract it. + var eventFD *vfs.FileDescription + if cb.Flags&linux.IOCB_FLAG_RESFD != 0 { + eventFD = t.GetFileVFS2(cb.ResFD) + if eventFD == nil { + return syserror.EBADF + } + defer eventFD.DecRef() + + // Check that it is an eventfd. + if _, ok := eventFD.Impl().(*eventfd.EventFileDescription); !ok { + return syserror.EINVAL + } + } + + ioseq, err := memoryFor(t, cb) + if err != nil { + return err + } + + // Check offset for reads/writes. + switch cb.OpCode { + case linux.IOCB_CMD_PREAD, linux.IOCB_CMD_PREADV, linux.IOCB_CMD_PWRITE, linux.IOCB_CMD_PWRITEV: + if cb.Offset < 0 { + return syserror.EINVAL + } + } + + // Prepare the request. + aioCtx, ok := t.MemoryManager().LookupAIOContext(t, id) + if !ok { + return syserror.EINVAL + } + if ready := aioCtx.Prepare(); !ready { + // Context is busy. + return syserror.EAGAIN + } + + if eventFD != nil { + // The request is set. Make sure there's a ref on the file. + // + // This is necessary when the callback executes on completion, + // which is also what will release this reference. + eventFD.IncRef() + } + + // Perform the request asynchronously. + fd.IncRef() + t.QueueAIO(getAIOCallback(t, fd, eventFD, cbAddr, cb, ioseq, aioCtx)) + return nil +} + +func getAIOCallback(t *kernel.Task, fd, eventFD *vfs.FileDescription, cbAddr usermem.Addr, cb *linux.IOCallback, ioseq usermem.IOSequence, aioCtx *mm.AIOContext) kernel.AIOCallback { + return func(ctx context.Context) { + if aioCtx.Dead() { + aioCtx.CancelPendingRequest() + return + } + ev := &linux.IOEvent{ + Data: cb.Data, + Obj: uint64(cbAddr), + } + + var err error + switch cb.OpCode { + case linux.IOCB_CMD_PREAD, linux.IOCB_CMD_PREADV: + ev.Result, err = fd.PRead(ctx, ioseq, cb.Offset, vfs.ReadOptions{}) + case linux.IOCB_CMD_PWRITE, linux.IOCB_CMD_PWRITEV: + ev.Result, err = fd.PWrite(ctx, ioseq, cb.Offset, vfs.WriteOptions{}) + case linux.IOCB_CMD_FSYNC, linux.IOCB_CMD_FDSYNC: + err = fd.Sync(ctx) + } + + // Update the result. + if err != nil { + err = slinux.HandleIOErrorVFS2(t, ev.Result != 0 /* partial */, err, nil /* never interrupted */, "aio", fd) + ev.Result = -int64(kernel.ExtractErrno(err, 0)) + } + + fd.DecRef() + + // Queue the result for delivery. + aioCtx.FinishRequest(ev) + + // Notify the event file if one was specified. This needs to happen + // *after* queueing the result to avoid racing with the thread we may + // wake up. + if eventFD != nil { + eventFD.Impl().(*eventfd.EventFileDescription).Signal(1) + eventFD.DecRef() + } + } +} + +// memoryFor returns appropriate memory for the given callback. +func memoryFor(t *kernel.Task, cb *linux.IOCallback) (usermem.IOSequence, error) { + bytes := int(cb.Bytes) + if bytes < 0 { + // Linux also requires that this field fit in ssize_t. + return usermem.IOSequence{}, syserror.EINVAL + } + + // Since this I/O will be asynchronous with respect to t's task goroutine, + // we have no guarantee that t's AddressSpace will be active during the + // I/O. + switch cb.OpCode { + case linux.IOCB_CMD_PREAD, linux.IOCB_CMD_PWRITE: + return t.SingleIOSequence(usermem.Addr(cb.Buf), bytes, usermem.IOOpts{ + AddressSpaceActive: false, + }) + + case linux.IOCB_CMD_PREADV, linux.IOCB_CMD_PWRITEV: + return t.IovecsIOSequence(usermem.Addr(cb.Buf), bytes, usermem.IOOpts{ + AddressSpaceActive: false, + }) + + case linux.IOCB_CMD_FSYNC, linux.IOCB_CMD_FDSYNC, linux.IOCB_CMD_NOOP: + return usermem.IOSequence{}, nil + + default: + // Not a supported command. + return usermem.IOSequence{}, syserror.EINVAL + } +} diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2.go b/pkg/sentry/syscalls/linux/vfs2/vfs2.go index 954c82f97..caa6a98ff 100644 --- a/pkg/sentry/syscalls/linux/vfs2/vfs2.go +++ b/pkg/sentry/syscalls/linux/vfs2/vfs2.go @@ -105,11 +105,7 @@ func Override() { s.Table[197] = syscalls.Supported("removexattr", Removexattr) s.Table[198] = syscalls.Supported("lremovexattr", Lremovexattr) s.Table[199] = syscalls.Supported("fremovexattr", Fremovexattr) - delete(s.Table, 206) // io_setup - delete(s.Table, 207) // io_destroy - delete(s.Table, 208) // io_getevents - delete(s.Table, 209) // io_submit - delete(s.Table, 210) // io_cancel + s.Table[209] = syscalls.PartiallySupported("io_submit", IoSubmit, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}) s.Table[213] = syscalls.Supported("epoll_create", EpollCreate) s.Table[217] = syscalls.Supported("getdents64", Getdents64) delete(s.Table, 221) // fdavise64 -- cgit v1.2.3 From e61acfb5eaec4450587116573f708284a0fe7849 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Tue, 16 Jun 2020 16:14:26 -0700 Subject: Correctly handle multiple resizings in pgalloc.findAvailableRange(). PiperOrigin-RevId: 316778032 --- pkg/sentry/pgalloc/pgalloc.go | 60 +++++++++++++++++++++----------------- pkg/sentry/pgalloc/pgalloc_test.go | 8 +++++ 2 files changed, 42 insertions(+), 26 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/pgalloc/pgalloc.go b/pkg/sentry/pgalloc/pgalloc.go index 46f19d218..afab97c0a 100644 --- a/pkg/sentry/pgalloc/pgalloc.go +++ b/pkg/sentry/pgalloc/pgalloc.go @@ -441,53 +441,61 @@ func (f *MemoryFile) Allocate(length uint64, kind usage.MemoryKind) (platform.Fi // Precondition: alignment must be a power of 2. func findAvailableRange(usage *usageSet, fileSize int64, length, alignment uint64) (platform.FileRange, bool) { alignmentMask := alignment - 1 - for gap := usage.UpperBoundGap(uint64(fileSize)); gap.Ok(); gap = gap.PrevLargeEnoughGap(length) { - // Start searching only at end of file. + + // Search for space in existing gaps, starting at the current end of the + // file and working backward. + lastGap := usage.LastGap() + gap := lastGap + for { end := gap.End() if end > uint64(fileSize) { end = uint64(fileSize) } - // Start at the top and align downwards. - start := end - length - if start > end { - break // Underflow. + // Try to allocate from the end of this gap, with the start of the + // allocated range aligned down to alignment. + unalignedStart := end - length + if unalignedStart > end { + // Negative overflow: this and all preceding gaps are too small to + // accommodate length. + break } - start &^= alignmentMask - - // Is the gap still sufficient? - if start < gap.Start() { - continue + if start := unalignedStart &^ alignmentMask; start >= gap.Start() { + return platform.FileRange{start, start + length}, true } - // Allocate in the given gap. - return platform.FileRange{start, start + length}, true + gap = gap.PrevLargeEnoughGap(length) + if !gap.Ok() { + break + } } // Check that it's possible to fit this allocation at the end of a file of any size. - min := usage.LastGap().Start() + min := lastGap.Start() min = (min + alignmentMask) &^ alignmentMask if min+length < min { - // Overflow. + // Overflow: allocation would exceed the range of uint64. return platform.FileRange{}, false } // Determine the minimum file size required to fit this allocation at its end. for { - if fileSize >= 2*fileSize { - // Is this because it's initially empty? - if fileSize == 0 { - fileSize += chunkSize - } else { - // fileSize overflow. + newFileSize := 2 * fileSize + if newFileSize <= fileSize { + if fileSize != 0 { + // Overflow: allocation would exceed the range of int64. return platform.FileRange{}, false } - } else { - // Double the current fileSize. - fileSize *= 2 + newFileSize = chunkSize + } + fileSize = newFileSize + + unalignedStart := uint64(fileSize) - length + if unalignedStart > uint64(fileSize) { + // Negative overflow: fileSize is still inadequate. + continue } - start := (uint64(fileSize) - length) &^ alignmentMask - if start >= min { + if start := unalignedStart &^ alignmentMask; start >= min { return platform.FileRange{start, start + length}, true } } diff --git a/pkg/sentry/pgalloc/pgalloc_test.go b/pkg/sentry/pgalloc/pgalloc_test.go index b5b68eb52..405db141f 100644 --- a/pkg/sentry/pgalloc/pgalloc_test.go +++ b/pkg/sentry/pgalloc/pgalloc_test.go @@ -142,6 +142,14 @@ func TestFindUnallocatedRange(t *testing.T) { alignment: hugepage, start: hugepage, }, + { + desc: "Allocation doubles file size more than once if necessary", + usage: &usageSegmentDataSlices{}, + fileSize: page, + length: 4 * page, + alignment: page, + start: 0, + }, { desc: "Allocations are compact if possible", usage: &usageSegmentDataSlices{ -- cgit v1.2.3 From 4232be8480625de6832713aa2e15dc5378896f58 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Thu, 26 Mar 2020 23:15:42 -0400 Subject: support sError injection in kvm module on Arm64 There are 3 types of asynchronous exceptions on Arm64: sError, IRQ, FIQ. In this case, we use the sError injection method in bluepillHandler to force the guest to quit. So that the test case of "TestBounce" can be passed on Arm64. Signed-off-by: Bin Lu --- pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go | 25 ++++++++++++++++++++++++ pkg/sentry/platform/kvm/bluepill_arm64.go | 11 +++++++++++ pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go | 21 ++++++++++++++++++++ pkg/sentry/platform/kvm/bluepill_unsafe.go | 20 +++++-------------- pkg/sentry/platform/kvm/kvm_arm64.go | 12 ++++++++++++ pkg/sentry/platform/kvm/kvm_const.go | 8 ++++++-- pkg/sentry/platform/kvm/machine_arm64_unsafe.go | 4 ++-- 7 files changed, 82 insertions(+), 19 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go b/pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go index 0d1e83e6c..03a98512e 100644 --- a/pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go @@ -17,6 +17,7 @@ package kvm import ( + "syscall" "unsafe" "gvisor.dev/gvisor/pkg/sentry/arch" @@ -60,3 +61,27 @@ func dieArchSetup(c *vCPU, context *arch.SignalContext64, guestRegs *userRegs) { func getHypercallID(addr uintptr) int { return _KVM_HYPERCALL_MAX } + +// bluepillStopGuest is reponsible for injecting interrupt. +// +//go:nosplit +func bluepillStopGuest(c *vCPU) { + // Interrupt: we must have requested an interrupt + // window; set the interrupt line. + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(c.fd), + _KVM_INTERRUPT, + uintptr(unsafe.Pointer(&bounce))); errno != 0 { + throw("interrupt injection failed") + } + // Clear previous injection request. + c.runData.requestInterruptWindow = 0 +} + +// bluepillReadyStopGuest checks whether the current vCPU is ready for interrupt injection. +// +//go:nosplit +func bluepillReadyStopGuest(c *vCPU) bool { + return c.runData.readyForInterruptInjection != 0 +} diff --git a/pkg/sentry/platform/kvm/bluepill_arm64.go b/pkg/sentry/platform/kvm/bluepill_arm64.go index 83643c602..dba563160 100644 --- a/pkg/sentry/platform/kvm/bluepill_arm64.go +++ b/pkg/sentry/platform/kvm/bluepill_arm64.go @@ -26,6 +26,17 @@ import ( var ( // The action for bluepillSignal is changed by sigaction(). bluepillSignal = syscall.SIGILL + + // vcpuSErr is the event of system error. + vcpuSErr = kvmVcpuEvents{ + exception: exception{ + sErrPending: 1, + sErrHasEsr: 0, + pad: [6]uint8{0, 0, 0, 0, 0, 0}, + sErrEsr: 1, + }, + rsvd: [12]uint32{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}, + } ) // bluepillArchEnter is called during bluepillEnter. diff --git a/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go b/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go index abd36f973..8b64f3a1e 100644 --- a/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go @@ -17,6 +17,7 @@ package kvm import ( + "syscall" "unsafe" "gvisor.dev/gvisor/pkg/sentry/arch" @@ -74,3 +75,23 @@ func getHypercallID(addr uintptr) int { return int(((addr) - arm64HypercallMMIOBase) >> 3) } } + +// bluepillStopGuest is reponsible for injecting sError. +// +//go:nosplit +func bluepillStopGuest(c *vCPU) { + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(c.fd), + _KVM_SET_VCPU_EVENTS, + uintptr(unsafe.Pointer(&vcpuSErr))); errno != 0 { + throw("sErr injection failed") + } +} + +// bluepillReadyStopGuest checks whether the current vCPU is ready for sError injection. +// +//go:nosplit +func bluepillReadyStopGuest(c *vCPU) bool { + return true +} diff --git a/pkg/sentry/platform/kvm/bluepill_unsafe.go b/pkg/sentry/platform/kvm/bluepill_unsafe.go index a5b9be36d..bf357de1a 100644 --- a/pkg/sentry/platform/kvm/bluepill_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_unsafe.go @@ -133,12 +133,12 @@ func bluepillHandler(context unsafe.Pointer) { // PIC, we can't inject an interrupt while they are // masked. We need to request a window if it's not // ready. - if c.runData.readyForInterruptInjection == 0 { - c.runData.requestInterruptWindow = 1 - continue // Rerun vCPU. - } else { + if bluepillReadyStopGuest(c) { // Force injection below; the vCPU is ready. c.runData.exitReason = _KVM_EXIT_IRQ_WINDOW_OPEN + } else { + c.runData.requestInterruptWindow = 1 + continue // Rerun vCPU. } case syscall.EFAULT: // If a fault is not serviceable due to the host @@ -217,17 +217,7 @@ func bluepillHandler(context unsafe.Pointer) { } } case _KVM_EXIT_IRQ_WINDOW_OPEN: - // Interrupt: we must have requested an interrupt - // window; set the interrupt line. - if _, _, errno := syscall.RawSyscall( - syscall.SYS_IOCTL, - uintptr(c.fd), - _KVM_INTERRUPT, - uintptr(unsafe.Pointer(&bounce))); errno != 0 { - throw("interrupt injection failed") - } - // Clear previous injection request. - c.runData.requestInterruptWindow = 0 + bluepillStopGuest(c) case _KVM_EXIT_SHUTDOWN: c.die(bluepillArchContext(context), "shutdown") return diff --git a/pkg/sentry/platform/kvm/kvm_arm64.go b/pkg/sentry/platform/kvm/kvm_arm64.go index 3134a076b..0b06a923a 100644 --- a/pkg/sentry/platform/kvm/kvm_arm64.go +++ b/pkg/sentry/platform/kvm/kvm_arm64.go @@ -46,6 +46,18 @@ type userRegs struct { fpRegs userFpsimdState } +type exception struct { + sErrPending uint8 + sErrHasEsr uint8 + pad [6]uint8 + sErrEsr uint64 +} + +type kvmVcpuEvents struct { + exception + rsvd [12]uint32 +} + // updateGlobalOnce does global initialization. It has to be called only once. func updateGlobalOnce(fd int) error { physicalInit() diff --git a/pkg/sentry/platform/kvm/kvm_const.go b/pkg/sentry/platform/kvm/kvm_const.go index 6a7676468..3bf918446 100644 --- a/pkg/sentry/platform/kvm/kvm_const.go +++ b/pkg/sentry/platform/kvm/kvm_const.go @@ -35,6 +35,8 @@ const ( _KVM_GET_SUPPORTED_CPUID = 0xc008ae05 _KVM_SET_CPUID2 = 0x4008ae90 _KVM_SET_SIGNAL_MASK = 0x4004ae8b + _KVM_GET_VCPU_EVENTS = 0x8040ae9f + _KVM_SET_VCPU_EVENTS = 0x4040aea0 ) // KVM exit reasons. @@ -54,8 +56,10 @@ const ( // KVM capability options. const ( - _KVM_CAP_MAX_VCPUS = 0x42 - _KVM_CAP_ARM_VM_IPA_SIZE = 0xa5 + _KVM_CAP_MAX_VCPUS = 0x42 + _KVM_CAP_ARM_VM_IPA_SIZE = 0xa5 + _KVM_CAP_VCPU_EVENTS = 0x29 + _KVM_CAP_ARM_INJECT_SERROR_ESR = 0x9e ) // KVM limits. diff --git a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go index 48c834499..8bed34922 100644 --- a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go @@ -273,8 +273,8 @@ func (c *vCPU) SwitchToUser(switchOpts ring0.SwitchOpts, info *arch.SignalInfo) case ring0.PageFault: return c.fault(int32(syscall.SIGSEGV), info) - case 0xaa: - return usermem.NoAccess, nil + case ring0.Vector(bounce): // ring0.VirtualizationException + return usermem.NoAccess, platform.ErrContextInterrupt default: return usermem.NoAccess, platform.ErrContextSignal } -- cgit v1.2.3 From 96519e2c9d3fa1f15537c4dfc081a19d8d1ce1a2 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Wed, 17 Jun 2020 10:02:41 -0700 Subject: Implement POSIX locks - Change FileDescriptionImpl Lock/UnlockPOSIX signature to take {start,length,whence}, so the correct offset can be calculated in the implementations. - Create PosixLocker interface to make it possible to share the same locking code from different implementations. Closes #1480 PiperOrigin-RevId: 316910286 --- pkg/sentry/fsimpl/devpts/BUILD | 2 +- pkg/sentry/fsimpl/devpts/devpts.go | 3 +- pkg/sentry/fsimpl/devpts/master.go | 14 +++- pkg/sentry/fsimpl/devpts/slave.go | 14 +++- pkg/sentry/fsimpl/ext/BUILD | 2 +- pkg/sentry/fsimpl/ext/directory.go | 11 +++ pkg/sentry/fsimpl/ext/inode.go | 3 +- pkg/sentry/fsimpl/ext/regular_file.go | 11 +++ pkg/sentry/fsimpl/ext/symlink.go | 1 + pkg/sentry/fsimpl/gofer/BUILD | 1 - pkg/sentry/fsimpl/gofer/gofer.go | 12 ++- pkg/sentry/fsimpl/gofer/special_file.go | 3 +- pkg/sentry/fsimpl/host/BUILD | 2 +- pkg/sentry/fsimpl/host/host.go | 14 +++- pkg/sentry/fsimpl/host/tty.go | 11 +++ pkg/sentry/fsimpl/kernfs/BUILD | 3 +- pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go | 16 +++- pkg/sentry/fsimpl/kernfs/fd_impl_util.go | 16 +++- pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 3 +- pkg/sentry/fsimpl/kernfs/kernfs_test.go | 5 +- pkg/sentry/fsimpl/overlay/BUILD | 2 +- pkg/sentry/fsimpl/overlay/overlay.go | 14 +++- pkg/sentry/fsimpl/pipefs/BUILD | 1 - pkg/sentry/fsimpl/pipefs/pipefs.go | 3 +- pkg/sentry/fsimpl/proc/BUILD | 2 +- pkg/sentry/fsimpl/proc/subtasks.go | 3 +- pkg/sentry/fsimpl/proc/task.go | 3 +- pkg/sentry/fsimpl/proc/task_fds.go | 3 +- pkg/sentry/fsimpl/proc/task_files.go | 14 +++- pkg/sentry/fsimpl/proc/tasks.go | 3 +- pkg/sentry/fsimpl/sys/BUILD | 1 - pkg/sentry/fsimpl/sys/sys.go | 3 +- pkg/sentry/fsimpl/tmpfs/BUILD | 1 - pkg/sentry/fsimpl/tmpfs/regular_file_test.go | 33 +++----- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 15 +++- pkg/sentry/kernel/fd_table.go | 10 ++- pkg/sentry/kernel/pipe/BUILD | 2 +- pkg/sentry/kernel/pipe/vfs.go | 18 +++- pkg/sentry/socket/hostinet/BUILD | 2 +- pkg/sentry/socket/hostinet/socket_vfs2.go | 14 +++- pkg/sentry/socket/netlink/BUILD | 2 +- pkg/sentry/socket/netlink/socket_vfs2.go | 14 +++- pkg/sentry/socket/netstack/BUILD | 2 +- pkg/sentry/socket/netstack/netstack_vfs2.go | 14 +++- pkg/sentry/socket/unix/BUILD | 2 +- pkg/sentry/socket/unix/unix_vfs2.go | 16 +++- pkg/sentry/syscalls/linux/vfs2/fd.go | 38 +++++++++ pkg/sentry/vfs/BUILD | 2 +- pkg/sentry/vfs/file_description.go | 18 ++-- pkg/sentry/vfs/file_description_impl_util.go | 25 ++---- pkg/sentry/vfs/lock.go | 109 +++++++++++++++++++++++++ pkg/sentry/vfs/lock/BUILD | 13 --- pkg/sentry/vfs/lock/lock.go | 72 ---------------- test/syscalls/linux/BUILD | 1 + test/syscalls/linux/fcntl.cc | 96 +++++++++++++++------- 55 files changed, 484 insertions(+), 234 deletions(-) create mode 100644 pkg/sentry/vfs/lock.go delete mode 100644 pkg/sentry/vfs/lock/BUILD delete mode 100644 pkg/sentry/vfs/lock/lock.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/devpts/BUILD b/pkg/sentry/fsimpl/devpts/BUILD index cf440dce8..93512c9b6 100644 --- a/pkg/sentry/fsimpl/devpts/BUILD +++ b/pkg/sentry/fsimpl/devpts/BUILD @@ -18,12 +18,12 @@ go_library( "//pkg/context", "//pkg/safemem", "//pkg/sentry/arch", + "//pkg/sentry/fs/lock", "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/unimpl", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserror", "//pkg/usermem", diff --git a/pkg/sentry/fsimpl/devpts/devpts.go b/pkg/sentry/fsimpl/devpts/devpts.go index 9b0e0cca2..e6fda2b4f 100644 --- a/pkg/sentry/fsimpl/devpts/devpts.go +++ b/pkg/sentry/fsimpl/devpts/devpts.go @@ -28,7 +28,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" ) @@ -117,7 +116,7 @@ type rootInode struct { kernfs.InodeNotSymlink kernfs.OrderedChildren - locks lock.FileLocks + locks vfs.FileLocks // Keep a reference to this inode's dentry. dentry kernfs.Dentry diff --git a/pkg/sentry/fsimpl/devpts/master.go b/pkg/sentry/fsimpl/devpts/master.go index 1d22adbe3..69879498a 100644 --- a/pkg/sentry/fsimpl/devpts/master.go +++ b/pkg/sentry/fsimpl/devpts/master.go @@ -18,11 +18,11 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/unimpl" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" @@ -35,7 +35,7 @@ type masterInode struct { kernfs.InodeNotDirectory kernfs.InodeNotSymlink - locks lock.FileLocks + locks vfs.FileLocks // Keep a reference to this inode's dentry. dentry kernfs.Dentry @@ -189,6 +189,16 @@ func (mfd *masterFileDescription) Stat(ctx context.Context, opts vfs.StatOptions return mfd.inode.Stat(fs, opts) } +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (mfd *masterFileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return mfd.Locks().LockPOSIX(ctx, &mfd.vfsfd, uid, t, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (mfd *masterFileDescription) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return mfd.Locks().UnlockPOSIX(ctx, &mfd.vfsfd, uid, start, length, whence) +} + // maybeEmitUnimplementedEvent emits unimplemented event if cmd is valid. func maybeEmitUnimplementedEvent(ctx context.Context, cmd uint32) { switch cmd { diff --git a/pkg/sentry/fsimpl/devpts/slave.go b/pkg/sentry/fsimpl/devpts/slave.go index 7fe475080..cf1a0f0ac 100644 --- a/pkg/sentry/fsimpl/devpts/slave.go +++ b/pkg/sentry/fsimpl/devpts/slave.go @@ -18,10 +18,10 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" @@ -34,7 +34,7 @@ type slaveInode struct { kernfs.InodeNotDirectory kernfs.InodeNotSymlink - locks lock.FileLocks + locks vfs.FileLocks // Keep a reference to this inode's dentry. dentry kernfs.Dentry @@ -185,3 +185,13 @@ func (sfd *slaveFileDescription) Stat(ctx context.Context, opts vfs.StatOptions) fs := sfd.vfsfd.VirtualDentry().Mount().Filesystem() return sfd.inode.Stat(fs, opts) } + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (sfd *slaveFileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return sfd.Locks().LockPOSIX(ctx, &sfd.vfsfd, uid, t, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (sfd *slaveFileDescription) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return sfd.Locks().UnlockPOSIX(ctx, &sfd.vfsfd, uid, start, length, whence) +} diff --git a/pkg/sentry/fsimpl/ext/BUILD b/pkg/sentry/fsimpl/ext/BUILD index 973fa0def..ef24f8159 100644 --- a/pkg/sentry/fsimpl/ext/BUILD +++ b/pkg/sentry/fsimpl/ext/BUILD @@ -54,13 +54,13 @@ go_library( "//pkg/safemem", "//pkg/sentry/arch", "//pkg/sentry/fs", + "//pkg/sentry/fs/lock", "//pkg/sentry/fsimpl/ext/disklayout", "//pkg/sentry/kernel/auth", "//pkg/sentry/memmap", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/syscalls/linux", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserror", "//pkg/usermem", diff --git a/pkg/sentry/fsimpl/ext/directory.go b/pkg/sentry/fsimpl/ext/directory.go index 43be6928a..357512c7e 100644 --- a/pkg/sentry/fsimpl/ext/directory.go +++ b/pkg/sentry/fsimpl/ext/directory.go @@ -20,6 +20,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/fs" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" @@ -305,3 +306,13 @@ func (fd *directoryFD) Seek(ctx context.Context, offset int64, whence int32) (in fd.off = offset return offset, nil } + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (fd *directoryFD) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return fd.Locks().LockPOSIX(ctx, &fd.vfsfd, uid, t, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (fd *directoryFD) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return fd.Locks().UnlockPOSIX(ctx, &fd.vfsfd, uid, start, length, whence) +} diff --git a/pkg/sentry/fsimpl/ext/inode.go b/pkg/sentry/fsimpl/ext/inode.go index 5caaf14ed..30636cf66 100644 --- a/pkg/sentry/fsimpl/ext/inode.go +++ b/pkg/sentry/fsimpl/ext/inode.go @@ -22,7 +22,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" ) @@ -55,7 +54,7 @@ type inode struct { // diskInode gives us access to the inode struct on disk. Immutable. diskInode disklayout.Inode - locks lock.FileLocks + locks vfs.FileLocks // This is immutable. The first field of the implementations must have inode // as the first field to ensure temporality. diff --git a/pkg/sentry/fsimpl/ext/regular_file.go b/pkg/sentry/fsimpl/ext/regular_file.go index 152036b2e..66d14bb95 100644 --- a/pkg/sentry/fsimpl/ext/regular_file.go +++ b/pkg/sentry/fsimpl/ext/regular_file.go @@ -20,6 +20,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/safemem" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" @@ -149,3 +150,13 @@ func (fd *regularFileFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpt // TODO(b/134676337): Implement mmap(2). return syserror.ENODEV } + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (fd *regularFileFD) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return fd.Locks().LockPOSIX(ctx, &fd.vfsfd, uid, t, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (fd *regularFileFD) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return fd.Locks().UnlockPOSIX(ctx, &fd.vfsfd, uid, start, length, whence) +} diff --git a/pkg/sentry/fsimpl/ext/symlink.go b/pkg/sentry/fsimpl/ext/symlink.go index acb28d85b..62efd4095 100644 --- a/pkg/sentry/fsimpl/ext/symlink.go +++ b/pkg/sentry/fsimpl/ext/symlink.go @@ -66,6 +66,7 @@ func (in *inode) isSymlink() bool { // O_PATH. For this reason most of the functions return EBADF. type symlinkFD struct { fileDescription + vfs.NoLockFD } // Compiles only if symlinkFD implements vfs.FileDescriptionImpl. diff --git a/pkg/sentry/fsimpl/gofer/BUILD b/pkg/sentry/fsimpl/gofer/BUILD index 5cdeeaeb5..4a800dcf9 100644 --- a/pkg/sentry/fsimpl/gofer/BUILD +++ b/pkg/sentry/fsimpl/gofer/BUILD @@ -69,7 +69,6 @@ go_library( "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usage", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/syserr", "//pkg/syserror", "//pkg/unet", diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index ac051b3a7..d8ae475ed 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -53,7 +53,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/unet" "gvisor.dev/gvisor/pkg/usermem" @@ -665,7 +664,7 @@ type dentry struct { // endpoint bound to this file. pipe *pipe.VFSPipe - locks lock.FileLocks + locks vfs.FileLocks } // dentryAttrMask returns a p9.AttrMask enabling all attributes used by the @@ -1439,9 +1438,14 @@ func (fd *fileDescription) LockBSD(ctx context.Context, uid fslock.UniqueID, t f } // LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. -func (fd *fileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, rng fslock.LockRange, block fslock.Blocker) error { +func (fd *fileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { fd.lockLogging.Do(func() { log.Infof("Range lock using gofer file handled internally.") }) - return fd.LockFD.LockPOSIX(ctx, uid, t, rng, block) + return fd.Locks().LockPOSIX(ctx, &fd.vfsfd, uid, t, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (fd *fileDescription) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return fd.Locks().UnlockPOSIX(ctx, &fd.vfsfd, uid, start, length, whence) } diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go index 289efdd25..e6e29b329 100644 --- a/pkg/sentry/fsimpl/gofer/special_file.go +++ b/pkg/sentry/fsimpl/gofer/special_file.go @@ -22,7 +22,6 @@ import ( "gvisor.dev/gvisor/pkg/fdnotifier" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" @@ -52,7 +51,7 @@ type specialFileFD struct { off int64 } -func newSpecialFileFD(h handle, mnt *vfs.Mount, d *dentry, locks *lock.FileLocks, flags uint32) (*specialFileFD, error) { +func newSpecialFileFD(h handle, mnt *vfs.Mount, d *dentry, locks *vfs.FileLocks, flags uint32) (*specialFileFD, error) { ftype := d.fileType() seekable := ftype == linux.S_IFREG mayBlock := ftype == linux.S_IFIFO || ftype == linux.S_IFSOCK diff --git a/pkg/sentry/fsimpl/host/BUILD b/pkg/sentry/fsimpl/host/BUILD index 54f16ad63..44a09d87a 100644 --- a/pkg/sentry/fsimpl/host/BUILD +++ b/pkg/sentry/fsimpl/host/BUILD @@ -27,6 +27,7 @@ go_library( "//pkg/safemem", "//pkg/sentry/arch", "//pkg/sentry/fs/fsutil", + "//pkg/sentry/fs/lock", "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/hostfd", "//pkg/sentry/kernel", @@ -39,7 +40,6 @@ go_library( "//pkg/sentry/unimpl", "//pkg/sentry/uniqueid", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserr", "//pkg/syserror", diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 5ec5100b8..7906242c9 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -28,13 +28,13 @@ import ( "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/refs" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/hostfd" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/memmap" unixsocket "gvisor.dev/gvisor/pkg/sentry/socket/unix" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -183,7 +183,7 @@ type inode struct { kernfs.InodeNotDirectory kernfs.InodeNotSymlink - locks lock.FileLocks + locks vfs.FileLocks // When the reference count reaches zero, the host fd is closed. refs.AtomicRefCount @@ -718,3 +718,13 @@ func (f *fileDescription) EventUnregister(e *waiter.Entry) { func (f *fileDescription) Readiness(mask waiter.EventMask) waiter.EventMask { return fdnotifier.NonBlockingPoll(int32(f.inode.hostFD), mask) } + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (f *fileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return f.Locks().LockPOSIX(ctx, &f.vfsfd, uid, t, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (f *fileDescription) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return f.Locks().UnlockPOSIX(ctx, &f.vfsfd, uid, start, length, whence) +} diff --git a/pkg/sentry/fsimpl/host/tty.go b/pkg/sentry/fsimpl/host/tty.go index 68af6e5af..0fbc543b1 100644 --- a/pkg/sentry/fsimpl/host/tty.go +++ b/pkg/sentry/fsimpl/host/tty.go @@ -18,6 +18,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/unimpl" "gvisor.dev/gvisor/pkg/sentry/vfs" @@ -377,3 +378,13 @@ func (t *TTYFileDescription) checkChange(ctx context.Context, sig linux.Signal) _ = pg.SendSignal(kernel.SignalInfoPriv(sig)) return kernel.ERESTARTSYS } + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (t *TTYFileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, typ fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return t.Locks().LockPOSIX(ctx, &t.vfsfd, uid, typ, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (t *TTYFileDescription) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return t.Locks().UnlockPOSIX(ctx, &t.vfsfd, uid, start, length, whence) +} diff --git a/pkg/sentry/fsimpl/kernfs/BUILD b/pkg/sentry/fsimpl/kernfs/BUILD index 0299dbde9..179df6c1e 100644 --- a/pkg/sentry/fsimpl/kernfs/BUILD +++ b/pkg/sentry/fsimpl/kernfs/BUILD @@ -45,11 +45,11 @@ go_library( "//pkg/fspath", "//pkg/log", "//pkg/refs", + "//pkg/sentry/fs/lock", "//pkg/sentry/kernel/auth", "//pkg/sentry/memmap", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserror", "//pkg/usermem", @@ -68,7 +68,6 @@ go_test( "//pkg/sentry/fsimpl/testutil", "//pkg/sentry/kernel/auth", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/syserror", "//pkg/usermem", "@com_github_google_go-cmp//cmp:go_default_library", diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go index 6418de0a3..c1215b70a 100644 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -19,9 +19,9 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" ) @@ -39,7 +39,7 @@ type DynamicBytesFile struct { InodeNotDirectory InodeNotSymlink - locks lock.FileLocks + locks vfs.FileLocks data vfs.DynamicBytesSource } @@ -86,7 +86,7 @@ type DynamicBytesFD struct { } // Init initializes a DynamicBytesFD. -func (fd *DynamicBytesFD) Init(m *vfs.Mount, d *vfs.Dentry, data vfs.DynamicBytesSource, locks *lock.FileLocks, flags uint32) error { +func (fd *DynamicBytesFD) Init(m *vfs.Mount, d *vfs.Dentry, data vfs.DynamicBytesSource, locks *vfs.FileLocks, flags uint32) error { fd.LockFD.Init(locks) if err := fd.vfsfd.Init(fd, flags, m, d, &vfs.FileDescriptionOptions{}); err != nil { return err @@ -135,3 +135,13 @@ func (fd *DynamicBytesFD) SetStat(context.Context, vfs.SetStatOptions) error { // DynamicBytesFiles are immutable. return syserror.EPERM } + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (fd *DynamicBytesFD) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return fd.Locks().LockPOSIX(ctx, &fd.vfsfd, uid, t, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (fd *DynamicBytesFD) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return fd.Locks().UnlockPOSIX(ctx, &fd.vfsfd, uid, start, length, whence) +} diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index 33a5968ca..5f7853a2a 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -19,10 +19,10 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -57,7 +57,7 @@ type GenericDirectoryFD struct { // NewGenericDirectoryFD creates a new GenericDirectoryFD and returns its // dentry. -func NewGenericDirectoryFD(m *vfs.Mount, d *vfs.Dentry, children *OrderedChildren, locks *lock.FileLocks, opts *vfs.OpenOptions) (*GenericDirectoryFD, error) { +func NewGenericDirectoryFD(m *vfs.Mount, d *vfs.Dentry, children *OrderedChildren, locks *vfs.FileLocks, opts *vfs.OpenOptions) (*GenericDirectoryFD, error) { fd := &GenericDirectoryFD{} if err := fd.Init(children, locks, opts); err != nil { return nil, err @@ -71,7 +71,7 @@ func NewGenericDirectoryFD(m *vfs.Mount, d *vfs.Dentry, children *OrderedChildre // Init initializes a GenericDirectoryFD. Use it when overriding // GenericDirectoryFD. Caller must call fd.VFSFileDescription.Init() with the // correct implementation. -func (fd *GenericDirectoryFD) Init(children *OrderedChildren, locks *lock.FileLocks, opts *vfs.OpenOptions) error { +func (fd *GenericDirectoryFD) Init(children *OrderedChildren, locks *vfs.FileLocks, opts *vfs.OpenOptions) error { if vfs.AccessTypesForOpenFlags(opts)&vfs.MayWrite != 0 { // Can't open directories for writing. return syserror.EISDIR @@ -235,3 +235,13 @@ func (fd *GenericDirectoryFD) SetStat(ctx context.Context, opts vfs.SetStatOptio inode := fd.vfsfd.VirtualDentry().Dentry().Impl().(*Dentry).inode return inode.SetStat(ctx, fd.filesystem(), creds, opts) } + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (fd *GenericDirectoryFD) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return fd.Locks().LockPOSIX(ctx, &fd.vfsfd, uid, t, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (fd *GenericDirectoryFD) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return fd.Locks().UnlockPOSIX(ctx, &fd.vfsfd, uid, start, length, whence) +} diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 0e4927215..650bd7b88 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -23,7 +23,6 @@ import ( "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) @@ -557,7 +556,7 @@ type StaticDirectory struct { InodeNoDynamicLookup OrderedChildren - locks lock.FileLocks + locks vfs.FileLocks } var _ Inode = (*StaticDirectory)(nil) diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index 6749facf7..dc407eb1d 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -27,7 +27,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fsimpl/testutil" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" ) @@ -103,7 +102,7 @@ type readonlyDir struct { kernfs.InodeDirectoryNoNewChildren kernfs.OrderedChildren - locks lock.FileLocks + locks vfs.FileLocks dentry kernfs.Dentry } @@ -133,7 +132,7 @@ type dir struct { kernfs.InodeNoDynamicLookup kernfs.OrderedChildren - locks lock.FileLocks + locks vfs.FileLocks fs *filesystem dentry kernfs.Dentry diff --git a/pkg/sentry/fsimpl/overlay/BUILD b/pkg/sentry/fsimpl/overlay/BUILD index f9413bbdd..8cf5b35d3 100644 --- a/pkg/sentry/fsimpl/overlay/BUILD +++ b/pkg/sentry/fsimpl/overlay/BUILD @@ -29,11 +29,11 @@ go_library( "//pkg/abi/linux", "//pkg/context", "//pkg/fspath", + "//pkg/sentry/fs/lock", "//pkg/sentry/kernel/auth", "//pkg/sentry/memmap", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserror", "//pkg/usermem", diff --git a/pkg/sentry/fsimpl/overlay/overlay.go b/pkg/sentry/fsimpl/overlay/overlay.go index e660d0e2c..e11a3ff19 100644 --- a/pkg/sentry/fsimpl/overlay/overlay.go +++ b/pkg/sentry/fsimpl/overlay/overlay.go @@ -35,9 +35,9 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) @@ -415,7 +415,7 @@ type dentry struct { devMinor uint32 ino uint64 - locks lock.FileLocks + locks vfs.FileLocks } // newDentry creates a new dentry. The dentry initially has no references; it @@ -610,3 +610,13 @@ func (fd *fileDescription) filesystem() *filesystem { func (fd *fileDescription) dentry() *dentry { return fd.vfsfd.Dentry().Impl().(*dentry) } + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (fd *fileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return fd.Locks().LockPOSIX(ctx, &fd.vfsfd, uid, t, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (fd *fileDescription) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return fd.Locks().UnlockPOSIX(ctx, &fd.vfsfd, uid, start, length, whence) +} diff --git a/pkg/sentry/fsimpl/pipefs/BUILD b/pkg/sentry/fsimpl/pipefs/BUILD index c618dbe6c..5950a2d59 100644 --- a/pkg/sentry/fsimpl/pipefs/BUILD +++ b/pkg/sentry/fsimpl/pipefs/BUILD @@ -15,7 +15,6 @@ go_library( "//pkg/sentry/kernel/pipe", "//pkg/sentry/kernel/time", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/syserror", "//pkg/usermem", ], diff --git a/pkg/sentry/fsimpl/pipefs/pipefs.go b/pkg/sentry/fsimpl/pipefs/pipefs.go index e4dabaa33..dd7eaf4a8 100644 --- a/pkg/sentry/fsimpl/pipefs/pipefs.go +++ b/pkg/sentry/fsimpl/pipefs/pipefs.go @@ -27,7 +27,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" ) @@ -82,7 +81,7 @@ type inode struct { kernfs.InodeNotSymlink kernfs.InodeNoopRefCount - locks lock.FileLocks + locks vfs.FileLocks pipe *pipe.VFSPipe ino uint64 diff --git a/pkg/sentry/fsimpl/proc/BUILD b/pkg/sentry/fsimpl/proc/BUILD index 351ba4ee9..6014138ff 100644 --- a/pkg/sentry/fsimpl/proc/BUILD +++ b/pkg/sentry/fsimpl/proc/BUILD @@ -22,6 +22,7 @@ go_library( "//pkg/log", "//pkg/refs", "//pkg/safemem", + "//pkg/sentry/fs/lock", "//pkg/sentry/fsbridge", "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/inet", @@ -35,7 +36,6 @@ go_library( "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usage", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/syserror", "//pkg/tcpip/header", "//pkg/usermem", diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go index e2cdb7ee9..36a89540c 100644 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -24,7 +24,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" ) @@ -38,7 +37,7 @@ type subtasksInode struct { kernfs.OrderedChildren kernfs.AlwaysValid - locks lock.FileLocks + locks vfs.FileLocks fs *filesystem task *kernel.Task diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index 44078a765..8bb2b0ce1 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -25,7 +25,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/mm" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" ) @@ -39,7 +38,7 @@ type taskInode struct { kernfs.InodeAttrs kernfs.OrderedChildren - locks lock.FileLocks + locks vfs.FileLocks task *kernel.Task } diff --git a/pkg/sentry/fsimpl/proc/task_fds.go b/pkg/sentry/fsimpl/proc/task_fds.go index ef6c1d04f..7debdb07a 100644 --- a/pkg/sentry/fsimpl/proc/task_fds.go +++ b/pkg/sentry/fsimpl/proc/task_fds.go @@ -27,7 +27,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" ) @@ -54,7 +53,7 @@ func taskFDExists(t *kernel.Task, fd int32) bool { } type fdDir struct { - locks lock.FileLocks + locks vfs.FileLocks fs *filesystem task *kernel.Task diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index e5eaa91cd..ba4405026 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -22,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/safemem" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/fsbridge" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -30,7 +31,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/mm" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" ) @@ -777,7 +777,7 @@ type namespaceInode struct { kernfs.InodeNotDirectory kernfs.InodeNotSymlink - locks lock.FileLocks + locks vfs.FileLocks } var _ kernfs.Inode = (*namespaceInode)(nil) @@ -830,3 +830,13 @@ func (fd *namespaceFD) SetStat(ctx context.Context, opts vfs.SetStatOptions) err func (fd *namespaceFD) Release() { fd.inode.DecRef() } + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (fd *namespaceFD) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return fd.Locks().LockPOSIX(ctx, &fd.vfsfd, uid, t, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (fd *namespaceFD) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return fd.Locks().UnlockPOSIX(ctx, &fd.vfsfd, uid, start, length, whence) +} diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index 58c8b9d05..2f214d0c2 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -25,7 +25,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" ) @@ -44,7 +43,7 @@ type tasksInode struct { kernfs.OrderedChildren kernfs.AlwaysValid - locks lock.FileLocks + locks vfs.FileLocks fs *filesystem pidns *kernel.PIDNamespace diff --git a/pkg/sentry/fsimpl/sys/BUILD b/pkg/sentry/fsimpl/sys/BUILD index 237f17def..a741e2bb6 100644 --- a/pkg/sentry/fsimpl/sys/BUILD +++ b/pkg/sentry/fsimpl/sys/BUILD @@ -15,7 +15,6 @@ go_library( "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/syserror", ], ) diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go index b84463d3a..fe02f7ee9 100644 --- a/pkg/sentry/fsimpl/sys/sys.go +++ b/pkg/sentry/fsimpl/sys/sys.go @@ -25,7 +25,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserror" ) @@ -101,7 +100,7 @@ type dir struct { kernfs.InodeDirectoryNoNewChildren kernfs.OrderedChildren - locks lock.FileLocks + locks vfs.FileLocks dentry kernfs.Dentry } diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD index 062321cbc..e73732a6b 100644 --- a/pkg/sentry/fsimpl/tmpfs/BUILD +++ b/pkg/sentry/fsimpl/tmpfs/BUILD @@ -62,7 +62,6 @@ go_library( "//pkg/sentry/uniqueid", "//pkg/sentry/usage", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/sentry/vfs/memxattr", "//pkg/sync", "//pkg/syserror", diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go index 64e1c40ad..146c7fdfe 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go @@ -138,48 +138,37 @@ func TestLocks(t *testing.T) { } defer cleanup() - var ( - uid1 lock.UniqueID - uid2 lock.UniqueID - // Non-blocking. - block lock.Blocker - ) - - uid1 = 123 - uid2 = 456 - - if err := fd.Impl().LockBSD(ctx, uid1, lock.ReadLock, block); err != nil { + uid1 := 123 + uid2 := 456 + if err := fd.Impl().LockBSD(ctx, uid1, lock.ReadLock, nil); err != nil { t.Fatalf("fd.Impl().LockBSD failed: err = %v", err) } - if err := fd.Impl().LockBSD(ctx, uid2, lock.ReadLock, block); err != nil { + if err := fd.Impl().LockBSD(ctx, uid2, lock.ReadLock, nil); err != nil { t.Fatalf("fd.Impl().LockBSD failed: err = %v", err) } - if got, want := fd.Impl().LockBSD(ctx, uid2, lock.WriteLock, block), syserror.ErrWouldBlock; got != want { + if got, want := fd.Impl().LockBSD(ctx, uid2, lock.WriteLock, nil), syserror.ErrWouldBlock; got != want { t.Fatalf("fd.Impl().LockBSD failed: got = %v, want = %v", got, want) } if err := fd.Impl().UnlockBSD(ctx, uid1); err != nil { t.Fatalf("fd.Impl().UnlockBSD failed: err = %v", err) } - if err := fd.Impl().LockBSD(ctx, uid2, lock.WriteLock, block); err != nil { + if err := fd.Impl().LockBSD(ctx, uid2, lock.WriteLock, nil); err != nil { t.Fatalf("fd.Impl().LockBSD failed: err = %v", err) } - rng1 := lock.LockRange{0, 1} - rng2 := lock.LockRange{1, 2} - - if err := fd.Impl().LockPOSIX(ctx, uid1, lock.ReadLock, rng1, block); err != nil { + if err := fd.Impl().LockPOSIX(ctx, uid1, lock.ReadLock, 0, 1, linux.SEEK_SET, nil); err != nil { t.Fatalf("fd.Impl().LockPOSIX failed: err = %v", err) } - if err := fd.Impl().LockPOSIX(ctx, uid2, lock.ReadLock, rng2, block); err != nil { + if err := fd.Impl().LockPOSIX(ctx, uid2, lock.ReadLock, 1, 2, linux.SEEK_SET, nil); err != nil { t.Fatalf("fd.Impl().LockPOSIX failed: err = %v", err) } - if err := fd.Impl().LockPOSIX(ctx, uid1, lock.WriteLock, rng1, block); err != nil { + if err := fd.Impl().LockPOSIX(ctx, uid1, lock.WriteLock, 0, 1, linux.SEEK_SET, nil); err != nil { t.Fatalf("fd.Impl().LockPOSIX failed: err = %v", err) } - if got, want := fd.Impl().LockPOSIX(ctx, uid2, lock.ReadLock, rng1, block), syserror.ErrWouldBlock; got != want { + if got, want := fd.Impl().LockPOSIX(ctx, uid2, lock.ReadLock, 0, 1, linux.SEEK_SET, nil), syserror.ErrWouldBlock; got != want { t.Fatalf("fd.Impl().LockPOSIX failed: got = %v, want = %v", got, want) } - if err := fd.Impl().UnlockPOSIX(ctx, uid1, rng1); err != nil { + if err := fd.Impl().UnlockPOSIX(ctx, uid1, 0, 1, linux.SEEK_SET); err != nil { t.Fatalf("fd.Impl().UnlockPOSIX failed: err = %v", err) } } diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 71a7522af..d0a3e1a5c 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -36,11 +36,11 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/sentry/vfs/memxattr" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" @@ -310,7 +310,7 @@ type inode struct { ctime int64 // nanoseconds mtime int64 // nanoseconds - locks lock.FileLocks + locks vfs.FileLocks // Inotify watches for this inode. watches vfs.Watches @@ -761,9 +761,20 @@ func NewMemfd(mount *vfs.Mount, creds *auth.Credentials, allowSeals bool, name s // Per Linux, mm/shmem.c:__shmem_file_setup(), memfd files are set up with // FMODE_READ | FMODE_WRITE. var fd regularFileFD + fd.Init(&inode.locks) flags := uint32(linux.O_RDWR) if err := fd.vfsfd.Init(&fd, flags, mount, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { return nil, err } return &fd.vfsfd, nil } + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (fd *fileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return fd.Locks().LockPOSIX(ctx, &fd.vfsfd, uid, t, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (fd *fileDescription) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return fd.Locks().UnlockPOSIX(ctx, &fd.vfsfd, uid, start, length, whence) +} diff --git a/pkg/sentry/kernel/fd_table.go b/pkg/sentry/kernel/fd_table.go index 48911240f..4b7d234a4 100644 --- a/pkg/sentry/kernel/fd_table.go +++ b/pkg/sentry/kernel/fd_table.go @@ -29,6 +29,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/syserror" ) // FDFlags define flags for an individual descriptor. @@ -148,7 +149,12 @@ func (f *FDTable) drop(file *fs.File) { // dropVFS2 drops the table reference. func (f *FDTable) dropVFS2(file *vfs.FileDescription) { - // TODO(gvisor.dev/issue/1480): Release locks. + // Release any POSIX lock possibly held by the FDTable. Range {0, 0} means the + // entire file. + err := file.UnlockPOSIX(context.Background(), f, 0, 0, linux.SEEK_SET) + if err != nil && err != syserror.ENOLCK { + panic(fmt.Sprintf("UnlockPOSIX failed: %v", err)) + } // Generate inotify events. ev := uint32(linux.IN_CLOSE_NOWRITE) @@ -157,7 +163,7 @@ func (f *FDTable) dropVFS2(file *vfs.FileDescription) { } file.Dentry().InotifyWithParent(ev, 0, vfs.PathEvent) - // Drop the table reference. + // Drop the table's reference. file.DecRef() } diff --git a/pkg/sentry/kernel/pipe/BUILD b/pkg/sentry/kernel/pipe/BUILD index 0db546b98..449643118 100644 --- a/pkg/sentry/kernel/pipe/BUILD +++ b/pkg/sentry/kernel/pipe/BUILD @@ -26,8 +26,8 @@ go_library( "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", + "//pkg/sentry/fs/lock", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserror", "//pkg/usermem", diff --git a/pkg/sentry/kernel/pipe/vfs.go b/pkg/sentry/kernel/pipe/vfs.go index c0e9ee1f4..a4519363f 100644 --- a/pkg/sentry/kernel/pipe/vfs.go +++ b/pkg/sentry/kernel/pipe/vfs.go @@ -20,8 +20,8 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/arch" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -63,12 +63,12 @@ func NewVFSPipe(isNamed bool, sizeBytes, atomicIOBytes int64) *VFSPipe { // Preconditions: statusFlags should not contain an open access mode. func (vp *VFSPipe) ReaderWriterPair(mnt *vfs.Mount, vfsd *vfs.Dentry, statusFlags uint32) (*vfs.FileDescription, *vfs.FileDescription) { // Connected pipes share the same locks. - locks := &lock.FileLocks{} + locks := &vfs.FileLocks{} return vp.newFD(mnt, vfsd, linux.O_RDONLY|statusFlags, locks), vp.newFD(mnt, vfsd, linux.O_WRONLY|statusFlags, locks) } // Open opens the pipe represented by vp. -func (vp *VFSPipe) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, statusFlags uint32, locks *lock.FileLocks) (*vfs.FileDescription, error) { +func (vp *VFSPipe) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, statusFlags uint32, locks *vfs.FileLocks) (*vfs.FileDescription, error) { vp.mu.Lock() defer vp.mu.Unlock() @@ -130,7 +130,7 @@ func (vp *VFSPipe) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, s } // Preconditions: vp.mu must be held. -func (vp *VFSPipe) newFD(mnt *vfs.Mount, vfsd *vfs.Dentry, statusFlags uint32, locks *lock.FileLocks) *vfs.FileDescription { +func (vp *VFSPipe) newFD(mnt *vfs.Mount, vfsd *vfs.Dentry, statusFlags uint32, locks *vfs.FileLocks) *vfs.FileDescription { fd := &VFSPipeFD{ pipe: &vp.pipe, } @@ -451,3 +451,13 @@ func spliceOrTee(ctx context.Context, dst, src *VFSPipeFD, count int64, removeFr } return n, err } + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (fd *VFSPipeFD) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return fd.Locks().LockPOSIX(ctx, &fd.vfsfd, uid, t, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (fd *VFSPipeFD) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return fd.Locks().UnlockPOSIX(ctx, &fd.vfsfd, uid, start, length, whence) +} diff --git a/pkg/sentry/socket/hostinet/BUILD b/pkg/sentry/socket/hostinet/BUILD index 60c9896fc..ff81ea6e6 100644 --- a/pkg/sentry/socket/hostinet/BUILD +++ b/pkg/sentry/socket/hostinet/BUILD @@ -26,6 +26,7 @@ go_library( "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", + "//pkg/sentry/fs/lock", "//pkg/sentry/fsimpl/sockfs", "//pkg/sentry/hostfd", "//pkg/sentry/inet", @@ -34,7 +35,6 @@ go_library( "//pkg/sentry/socket", "//pkg/sentry/socket/control", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/syserr", "//pkg/syserror", "//pkg/tcpip/stack", diff --git a/pkg/sentry/socket/hostinet/socket_vfs2.go b/pkg/sentry/socket/hostinet/socket_vfs2.go index 027add1fd..ad5f64799 100644 --- a/pkg/sentry/socket/hostinet/socket_vfs2.go +++ b/pkg/sentry/socket/hostinet/socket_vfs2.go @@ -21,12 +21,12 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fdnotifier" "gvisor.dev/gvisor/pkg/sentry/arch" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" "gvisor.dev/gvisor/pkg/sentry/hostfd" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -61,7 +61,7 @@ func newVFS2Socket(t *kernel.Task, family int, stype linux.SockType, protocol in fd: fd, }, } - s.LockFD.Init(&lock.FileLocks{}) + s.LockFD.Init(&vfs.FileLocks{}) if err := fdnotifier.AddFD(int32(fd), &s.queue); err != nil { return nil, syserr.FromError(err) } @@ -134,6 +134,16 @@ func (s *socketVFS2) Write(ctx context.Context, src usermem.IOSequence, opts vfs return int64(n), err } +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (s *socketVFS2) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return s.Locks().LockPOSIX(ctx, &s.vfsfd, uid, t, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (s *socketVFS2) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return s.Locks().UnlockPOSIX(ctx, &s.vfsfd, uid, start, length, whence) +} + type socketProviderVFS2 struct { family int } diff --git a/pkg/sentry/socket/netlink/BUILD b/pkg/sentry/socket/netlink/BUILD index 420e573c9..d5ca3ac56 100644 --- a/pkg/sentry/socket/netlink/BUILD +++ b/pkg/sentry/socket/netlink/BUILD @@ -20,6 +20,7 @@ go_library( "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", + "//pkg/sentry/fs/lock", "//pkg/sentry/fsimpl/sockfs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", @@ -29,7 +30,6 @@ go_library( "//pkg/sentry/socket/unix", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserr", "//pkg/syserror", diff --git a/pkg/sentry/socket/netlink/socket_vfs2.go b/pkg/sentry/socket/netlink/socket_vfs2.go index 8bfee5193..dbcd8b49a 100644 --- a/pkg/sentry/socket/netlink/socket_vfs2.go +++ b/pkg/sentry/socket/netlink/socket_vfs2.go @@ -18,12 +18,12 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/unix" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip" @@ -78,7 +78,7 @@ func NewVFS2(t *kernel.Task, skType linux.SockType, protocol Protocol) (*SocketV sendBufferSize: defaultSendBufferSize, }, } - fd.LockFD.Init(&lock.FileLocks{}) + fd.LockFD.Init(&vfs.FileLocks{}) return fd, nil } @@ -140,3 +140,13 @@ func (s *SocketVFS2) Write(ctx context.Context, src usermem.IOSequence, opts vfs n, err := s.sendMsg(ctx, src, nil, 0, socket.ControlMessages{}) return int64(n), err.ToError() } + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (s *SocketVFS2) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return s.Locks().LockPOSIX(ctx, &s.vfsfd, uid, t, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (s *SocketVFS2) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return s.Locks().UnlockPOSIX(ctx, &s.vfsfd, uid, start, length, whence) +} diff --git a/pkg/sentry/socket/netstack/BUILD b/pkg/sentry/socket/netstack/BUILD index 0f592ecc3..ea6ebd0e2 100644 --- a/pkg/sentry/socket/netstack/BUILD +++ b/pkg/sentry/socket/netstack/BUILD @@ -28,6 +28,7 @@ go_library( "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", + "//pkg/sentry/fs/lock", "//pkg/sentry/fsimpl/sockfs", "//pkg/sentry/inet", "//pkg/sentry/kernel", @@ -37,7 +38,6 @@ go_library( "//pkg/sentry/socket/netfilter", "//pkg/sentry/unimpl", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserr", "//pkg/syserror", diff --git a/pkg/sentry/socket/netstack/netstack_vfs2.go b/pkg/sentry/socket/netstack/netstack_vfs2.go index 1412a4810..d65a89316 100644 --- a/pkg/sentry/socket/netstack/netstack_vfs2.go +++ b/pkg/sentry/socket/netstack/netstack_vfs2.go @@ -19,13 +19,13 @@ import ( "gvisor.dev/gvisor/pkg/amutex" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/netfilter" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip" @@ -66,7 +66,7 @@ func NewVFS2(t *kernel.Task, family int, skType linux.SockType, protocol int, qu protocol: protocol, }, } - s.LockFD.Init(&lock.FileLocks{}) + s.LockFD.Init(&vfs.FileLocks{}) vfsfd := &s.vfsfd if err := vfsfd.Init(s, linux.O_RDWR, mnt, d, &vfs.FileDescriptionOptions{ DenyPRead: true, @@ -318,3 +318,13 @@ func (s *SocketVFS2) SetSockOpt(t *kernel.Task, level int, name int, optVal []by return SetSockOpt(t, s, s.Endpoint, level, name, optVal) } + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (s *SocketVFS2) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return s.Locks().LockPOSIX(ctx, &s.vfsfd, uid, t, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (s *SocketVFS2) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return s.Locks().UnlockPOSIX(ctx, &s.vfsfd, uid, start, length, whence) +} diff --git a/pkg/sentry/socket/unix/BUILD b/pkg/sentry/socket/unix/BUILD index 7d4cc80fe..cca5e70f1 100644 --- a/pkg/sentry/socket/unix/BUILD +++ b/pkg/sentry/socket/unix/BUILD @@ -21,6 +21,7 @@ go_library( "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", + "//pkg/sentry/fs/lock", "//pkg/sentry/fsimpl/sockfs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/time", @@ -29,7 +30,6 @@ go_library( "//pkg/sentry/socket/netstack", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", "//pkg/syserr", "//pkg/syserror", "//pkg/tcpip", diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go index 8c32371a2..ff2149250 100644 --- a/pkg/sentry/socket/unix/unix_vfs2.go +++ b/pkg/sentry/socket/unix/unix_vfs2.go @@ -19,6 +19,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/arch" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/socket" @@ -26,7 +27,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/socket/netstack" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip" @@ -53,7 +53,7 @@ func NewSockfsFile(t *kernel.Task, ep transport.Endpoint, stype linux.SockType) mnt := t.Kernel().SocketMount() d := sockfs.NewDentry(t.Credentials(), mnt) - fd, err := NewFileDescription(ep, stype, linux.O_RDWR, mnt, d, &lock.FileLocks{}) + fd, err := NewFileDescription(ep, stype, linux.O_RDWR, mnt, d, &vfs.FileLocks{}) if err != nil { return nil, syserr.FromError(err) } @@ -62,7 +62,7 @@ func NewSockfsFile(t *kernel.Task, ep transport.Endpoint, stype linux.SockType) // NewFileDescription creates and returns a socket file description // corresponding to the given mount and dentry. -func NewFileDescription(ep transport.Endpoint, stype linux.SockType, flags uint32, mnt *vfs.Mount, d *vfs.Dentry, locks *lock.FileLocks) (*vfs.FileDescription, error) { +func NewFileDescription(ep transport.Endpoint, stype linux.SockType, flags uint32, mnt *vfs.Mount, d *vfs.Dentry, locks *vfs.FileLocks) (*vfs.FileDescription, error) { // You can create AF_UNIX, SOCK_RAW sockets. They're the same as // SOCK_DGRAM and don't require CAP_NET_RAW. if stype == linux.SOCK_RAW { @@ -300,6 +300,16 @@ func (s *SocketVFS2) SetSockOpt(t *kernel.Task, level int, name int, optVal []by return netstack.SetSockOpt(t, s, s.ep, level, name, optVal) } +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (s *SocketVFS2) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return s.Locks().LockPOSIX(ctx, &s.vfsfd, uid, t, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (s *SocketVFS2) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return s.Locks().UnlockPOSIX(ctx, &s.vfsfd, uid, start, length, whence) +} + // providerVFS2 is a unix domain socket provider for VFS2. type providerVFS2 struct{} diff --git a/pkg/sentry/syscalls/linux/vfs2/fd.go b/pkg/sentry/syscalls/linux/vfs2/fd.go index f9ccb303c..f5eaa076b 100644 --- a/pkg/sentry/syscalls/linux/vfs2/fd.go +++ b/pkg/sentry/syscalls/linux/vfs2/fd.go @@ -17,10 +17,12 @@ package vfs2 import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" + "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" ) @@ -167,8 +169,44 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } err := tmpfs.AddSeals(file, args[2].Uint()) return 0, nil, err + case linux.F_SETLK, linux.F_SETLKW: + return 0, nil, posixLock(t, args, file, cmd) default: // TODO(gvisor.dev/issue/2920): Everything else is not yet supported. return 0, nil, syserror.EINVAL } } + +func posixLock(t *kernel.Task, args arch.SyscallArguments, file *vfs.FileDescription, cmd int32) error { + // Copy in the lock request. + flockAddr := args[2].Pointer() + var flock linux.Flock + if _, err := t.CopyIn(flockAddr, &flock); err != nil { + return err + } + + var blocker lock.Blocker + if cmd == linux.F_SETLKW { + blocker = t + } + + switch flock.Type { + case linux.F_RDLCK: + if !file.IsReadable() { + return syserror.EBADF + } + return file.LockPOSIX(t, t.FDTable(), lock.ReadLock, uint64(flock.Start), uint64(flock.Len), flock.Whence, blocker) + + case linux.F_WRLCK: + if !file.IsWritable() { + return syserror.EBADF + } + return file.LockPOSIX(t, t.FDTable(), lock.WriteLock, uint64(flock.Start), uint64(flock.Len), flock.Whence, blocker) + + case linux.F_UNLCK: + return file.UnlockPOSIX(t, t.FDTable(), uint64(flock.Start), uint64(flock.Len), flock.Whence) + + default: + return syserror.EINVAL + } +} diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index 16d9f3a28..642769e7c 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -44,6 +44,7 @@ go_library( "filesystem_impl_util.go", "filesystem_type.go", "inotify.go", + "lock.go", "mount.go", "mount_unsafe.go", "options.go", @@ -72,7 +73,6 @@ go_library( "//pkg/sentry/memmap", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/uniqueid", - "//pkg/sentry/vfs/lock", "//pkg/sync", "//pkg/syserror", "//pkg/usermem", diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 13c48824e..e0538ea53 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -438,14 +438,10 @@ type FileDescriptionImpl interface { UnlockBSD(ctx context.Context, uid lock.UniqueID) error // LockPOSIX tries to acquire a POSIX-style advisory file lock. - // - // TODO(gvisor.dev/issue/1480): POSIX-style file locking - LockPOSIX(ctx context.Context, uid lock.UniqueID, t lock.LockType, rng lock.LockRange, block lock.Blocker) error + LockPOSIX(ctx context.Context, uid lock.UniqueID, t lock.LockType, start, length uint64, whence int16, block lock.Blocker) error // UnlockPOSIX releases a POSIX-style advisory file lock. - // - // TODO(gvisor.dev/issue/1480): POSIX-style file locking - UnlockPOSIX(ctx context.Context, uid lock.UniqueID, rng lock.LockRange) error + UnlockPOSIX(ctx context.Context, uid lock.UniqueID, start, length uint64, whence int16) error } // Dirent holds the information contained in struct linux_dirent64. @@ -764,3 +760,13 @@ func (fd *FileDescription) LockBSD(ctx context.Context, lockType lock.LockType, func (fd *FileDescription) UnlockBSD(ctx context.Context) error { return fd.impl.UnlockBSD(ctx, fd) } + +// LockPOSIX locks a POSIX-style file range lock. +func (fd *FileDescription) LockPOSIX(ctx context.Context, uid lock.UniqueID, t lock.LockType, start, end uint64, whence int16, block lock.Blocker) error { + return fd.impl.LockPOSIX(ctx, uid, t, start, end, whence, block) +} + +// UnlockPOSIX unlocks a POSIX-style file range lock. +func (fd *FileDescription) UnlockPOSIX(ctx context.Context, uid lock.UniqueID, start, end uint64, whence int16) error { + return fd.impl.UnlockPOSIX(ctx, uid, start, end, whence) +} diff --git a/pkg/sentry/vfs/file_description_impl_util.go b/pkg/sentry/vfs/file_description_impl_util.go index af7213dfd..1e66997ce 100644 --- a/pkg/sentry/vfs/file_description_impl_util.go +++ b/pkg/sentry/vfs/file_description_impl_util.go @@ -23,7 +23,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -369,14 +368,19 @@ func GenericConfigureMMap(fd *FileDescription, m memmap.Mappable, opts *memmap.M // LockFD may be used by most implementations of FileDescriptionImpl.Lock* // functions. Caller must call Init(). type LockFD struct { - locks *lock.FileLocks + locks *FileLocks } // Init initializes fd with FileLocks to use. -func (fd *LockFD) Init(locks *lock.FileLocks) { +func (fd *LockFD) Init(locks *FileLocks) { fd.locks = locks } +// Locks returns the locks associated with this file. +func (fd *LockFD) Locks() *FileLocks { + return fd.locks +} + // LockBSD implements vfs.FileDescriptionImpl.LockBSD. func (fd *LockFD) LockBSD(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, block fslock.Blocker) error { return fd.locks.LockBSD(uid, t, block) @@ -388,17 +392,6 @@ func (fd *LockFD) UnlockBSD(ctx context.Context, uid fslock.UniqueID) error { return nil } -// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. -func (fd *LockFD) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, rng fslock.LockRange, block fslock.Blocker) error { - return fd.locks.LockPOSIX(uid, t, rng, block) -} - -// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. -func (fd *LockFD) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, rng fslock.LockRange) error { - fd.locks.UnlockPOSIX(uid, rng) - return nil -} - // NoLockFD implements Lock*/Unlock* portion of FileDescriptionImpl interface // returning ENOLCK. type NoLockFD struct{} @@ -414,11 +407,11 @@ func (NoLockFD) UnlockBSD(ctx context.Context, uid fslock.UniqueID) error { } // LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. -func (NoLockFD) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, rng fslock.LockRange, block fslock.Blocker) error { +func (NoLockFD) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { return syserror.ENOLCK } // UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. -func (NoLockFD) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, rng fslock.LockRange) error { +func (NoLockFD) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { return syserror.ENOLCK } diff --git a/pkg/sentry/vfs/lock.go b/pkg/sentry/vfs/lock.go new file mode 100644 index 000000000..6c7583a81 --- /dev/null +++ b/pkg/sentry/vfs/lock.go @@ -0,0 +1,109 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package lock provides POSIX and BSD style file locking for VFS2 file +// implementations. +// +// The actual implementations can be found in the lock package under +// sentry/fs/lock. +package vfs + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" + "gvisor.dev/gvisor/pkg/syserror" +) + +// FileLocks supports POSIX and BSD style locks, which correspond to fcntl(2) +// and flock(2) respectively in Linux. It can be embedded into various file +// implementations for VFS2 that support locking. +// +// Note that in Linux these two types of locks are _not_ cooperative, because +// race and deadlock conditions make merging them prohibitive. We do the same +// and keep them oblivious to each other. +type FileLocks struct { + // bsd is a set of BSD-style advisory file wide locks, see flock(2). + bsd fslock.Locks + + // posix is a set of POSIX-style regional advisory locks, see fcntl(2). + posix fslock.Locks +} + +// LockBSD tries to acquire a BSD-style lock on the entire file. +func (fl *FileLocks) LockBSD(uid fslock.UniqueID, t fslock.LockType, block fslock.Blocker) error { + if fl.bsd.LockRegion(uid, t, fslock.LockRange{0, fslock.LockEOF}, block) { + return nil + } + return syserror.ErrWouldBlock +} + +// UnlockBSD releases a BSD-style lock on the entire file. +// +// This operation is always successful, even if there did not exist a lock on +// the requested region held by uid in the first place. +func (fl *FileLocks) UnlockBSD(uid fslock.UniqueID) { + fl.bsd.UnlockRegion(uid, fslock.LockRange{0, fslock.LockEOF}) +} + +// LockPOSIX tries to acquire a POSIX-style lock on a file region. +func (fl *FileLocks) LockPOSIX(ctx context.Context, fd *FileDescription, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + rng, err := computeRange(ctx, fd, start, length, whence) + if err != nil { + return err + } + if fl.posix.LockRegion(uid, t, rng, block) { + return nil + } + return syserror.ErrWouldBlock +} + +// UnlockPOSIX releases a POSIX-style lock on a file region. +// +// This operation is always successful, even if there did not exist a lock on +// the requested region held by uid in the first place. +func (fl *FileLocks) UnlockPOSIX(ctx context.Context, fd *FileDescription, uid fslock.UniqueID, start, length uint64, whence int16) error { + rng, err := computeRange(ctx, fd, start, length, whence) + if err != nil { + return err + } + fl.posix.UnlockRegion(uid, rng) + return nil +} + +func computeRange(ctx context.Context, fd *FileDescription, start uint64, length uint64, whence int16) (fslock.LockRange, error) { + var off int64 + switch whence { + case linux.SEEK_SET: + off = 0 + case linux.SEEK_CUR: + // Note that Linux does not hold any mutexes while retrieving the file + // offset, see fs/locks.c:flock_to_posix_lock and fs/locks.c:fcntl_setlk. + curOff, err := fd.Seek(ctx, 0, linux.SEEK_CUR) + if err != nil { + return fslock.LockRange{}, err + } + off = curOff + case linux.SEEK_END: + stat, err := fd.Stat(ctx, StatOptions{Mask: linux.STATX_SIZE}) + if err != nil { + return fslock.LockRange{}, err + } + off = int64(stat.Size) + default: + return fslock.LockRange{}, syserror.EINVAL + } + + return fslock.ComputeRange(int64(start), int64(length), off) +} diff --git a/pkg/sentry/vfs/lock/BUILD b/pkg/sentry/vfs/lock/BUILD deleted file mode 100644 index d9ab063b7..000000000 --- a/pkg/sentry/vfs/lock/BUILD +++ /dev/null @@ -1,13 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "lock", - srcs = ["lock.go"], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/sentry/fs/lock", - "//pkg/syserror", - ], -) diff --git a/pkg/sentry/vfs/lock/lock.go b/pkg/sentry/vfs/lock/lock.go deleted file mode 100644 index 724dfe743..000000000 --- a/pkg/sentry/vfs/lock/lock.go +++ /dev/null @@ -1,72 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package lock provides POSIX and BSD style file locking for VFS2 file -// implementations. -// -// The actual implementations can be found in the lock package under -// sentry/fs/lock. -package lock - -import ( - fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" - "gvisor.dev/gvisor/pkg/syserror" -) - -// FileLocks supports POSIX and BSD style locks, which correspond to fcntl(2) -// and flock(2) respectively in Linux. It can be embedded into various file -// implementations for VFS2 that support locking. -// -// Note that in Linux these two types of locks are _not_ cooperative, because -// race and deadlock conditions make merging them prohibitive. We do the same -// and keep them oblivious to each other. -type FileLocks struct { - // bsd is a set of BSD-style advisory file wide locks, see flock(2). - bsd fslock.Locks - - // posix is a set of POSIX-style regional advisory locks, see fcntl(2). - posix fslock.Locks -} - -// LockBSD tries to acquire a BSD-style lock on the entire file. -func (fl *FileLocks) LockBSD(uid fslock.UniqueID, t fslock.LockType, block fslock.Blocker) error { - if fl.bsd.LockRegion(uid, t, fslock.LockRange{0, fslock.LockEOF}, block) { - return nil - } - return syserror.ErrWouldBlock -} - -// UnlockBSD releases a BSD-style lock on the entire file. -// -// This operation is always successful, even if there did not exist a lock on -// the requested region held by uid in the first place. -func (fl *FileLocks) UnlockBSD(uid fslock.UniqueID) { - fl.bsd.UnlockRegion(uid, fslock.LockRange{0, fslock.LockEOF}) -} - -// LockPOSIX tries to acquire a POSIX-style lock on a file region. -func (fl *FileLocks) LockPOSIX(uid fslock.UniqueID, t fslock.LockType, rng fslock.LockRange, block fslock.Blocker) error { - if fl.posix.LockRegion(uid, t, rng, block) { - return nil - } - return syserror.ErrWouldBlock -} - -// UnlockPOSIX releases a POSIX-style lock on a file region. -// -// This operation is always successful, even if there did not exist a lock on -// the requested region held by uid in the first place. -func (fl *FileLocks) UnlockPOSIX(uid fslock.UniqueID, rng fslock.LockRange) { - fl.posix.UnlockRegion(uid, rng) -} diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 96044928e..078e4a284 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -791,6 +791,7 @@ cc_binary( deps = [ ":socket_test_util", "//test/util:cleanup", + "//test/util:epoll_util", "//test/util:eventfd_util", "//test/util:fs_util", "@com_google_absl//absl/base:core_headers", diff --git a/test/syscalls/linux/fcntl.cc b/test/syscalls/linux/fcntl.cc index 35e8a4ff3..25bef2522 100644 --- a/test/syscalls/linux/fcntl.cc +++ b/test/syscalls/linux/fcntl.cc @@ -191,45 +191,85 @@ TEST(FcntlTest, SetFlags) { EXPECT_EQ(rflags, expected); } -TEST_F(FcntlLockTest, SetLockBadFd) { +void TestLock(int fd, short lock_type = F_RDLCK) { // NOLINT, type in flock struct flock fl; - fl.l_type = F_WRLCK; + fl.l_type = lock_type; fl.l_whence = SEEK_SET; fl.l_start = 0; - // len 0 has a special meaning: lock all bytes despite how - // large the file grows. + // len 0 locks all bytes despite how large the file grows. fl.l_len = 0; - EXPECT_THAT(fcntl(-1, F_SETLK, &fl), SyscallFailsWithErrno(EBADF)); + EXPECT_THAT(fcntl(fd, F_SETLK, &fl), SyscallSucceeds()); } -TEST_F(FcntlLockTest, SetLockPipe) { - int fds[2]; - ASSERT_THAT(pipe(fds), SyscallSucceeds()); - +void TestLockBadFD(int fd, + short lock_type = F_RDLCK) { // NOLINT, type in flock struct flock fl; - fl.l_type = F_WRLCK; + fl.l_type = lock_type; fl.l_whence = SEEK_SET; fl.l_start = 0; - // Same as SetLockBadFd, but doesn't matter, we expect this to fail. + // len 0 locks all bytes despite how large the file grows. fl.l_len = 0; - EXPECT_THAT(fcntl(fds[0], F_SETLK, &fl), SyscallFailsWithErrno(EBADF)); - EXPECT_THAT(close(fds[0]), SyscallSucceeds()); - EXPECT_THAT(close(fds[1]), SyscallSucceeds()); + EXPECT_THAT(fcntl(fd, F_SETLK, &fl), SyscallFailsWithErrno(EBADF)); } +TEST_F(FcntlLockTest, SetLockBadFd) { TestLockBadFD(-1); } + TEST_F(FcntlLockTest, SetLockDir) { auto dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); - FileDescriptor fd = - ASSERT_NO_ERRNO_AND_VALUE(Open(dir.path(), O_RDONLY, 0666)); + auto fd = ASSERT_NO_ERRNO_AND_VALUE(Open(dir.path(), O_RDONLY, 0000)); + TestLock(fd.get()); +} - struct flock fl; - fl.l_type = F_RDLCK; - fl.l_whence = SEEK_SET; - fl.l_start = 0; - // Same as SetLockBadFd. - fl.l_len = 0; +TEST_F(FcntlLockTest, SetLockSymlink) { + // TODO(gvisor.dev/issue/2782): Replace with IsRunningWithVFS1() when O_PATH + // is supported. + SKIP_IF(IsRunningOnGvisor()); - EXPECT_THAT(fcntl(fd.get(), F_SETLK, &fl), SyscallSucceeds()); + auto file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); + auto symlink = ASSERT_NO_ERRNO_AND_VALUE( + TempPath::CreateSymlinkTo(GetAbsoluteTestTmpdir(), file.path())); + + auto fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(symlink.path(), O_RDONLY | O_PATH, 0000)); + TestLockBadFD(fd.get()); +} + +TEST_F(FcntlLockTest, SetLockProc) { + auto fd = + ASSERT_NO_ERRNO_AND_VALUE(Open("/proc/self/status", O_RDONLY, 0000)); + TestLock(fd.get()); +} + +TEST_F(FcntlLockTest, SetLockPipe) { + SKIP_IF(IsRunningWithVFS1()); + + int fds[2]; + ASSERT_THAT(pipe(fds), SyscallSucceeds()); + + TestLock(fds[0]); + TestLockBadFD(fds[0], F_WRLCK); + + TestLock(fds[1], F_WRLCK); + TestLockBadFD(fds[1]); + + EXPECT_THAT(close(fds[0]), SyscallSucceeds()); + EXPECT_THAT(close(fds[1]), SyscallSucceeds()); +} + +TEST_F(FcntlLockTest, SetLockSocket) { + SKIP_IF(IsRunningWithVFS1()); + + int sock = socket(AF_UNIX, SOCK_STREAM, 0); + ASSERT_THAT(sock, SyscallSucceeds()); + + struct sockaddr_un addr = + ASSERT_NO_ERRNO_AND_VALUE(UniqueUnixAddr(true /* abstract */, AF_UNIX)); + ASSERT_THAT( + bind(sock, reinterpret_cast(&addr), sizeof(addr)), + SyscallSucceeds()); + + TestLock(sock); + EXPECT_THAT(close(sock), SyscallSucceeds()); } TEST_F(FcntlLockTest, SetLockBadOpenFlagsWrite) { @@ -241,8 +281,7 @@ TEST_F(FcntlLockTest, SetLockBadOpenFlagsWrite) { fl0.l_type = F_WRLCK; fl0.l_whence = SEEK_SET; fl0.l_start = 0; - // Same as SetLockBadFd. - fl0.l_len = 0; + fl0.l_len = 0; // Lock all file // Expect that setting a write lock using a read only file descriptor // won't work. @@ -704,7 +743,7 @@ TEST_F(FcntlLockTest, SetWriteLockThenBlockingWriteLock) { << "Exited with code: " << status; } -// This test will veirfy that blocking works as expected when another process +// This test will verify that blocking works as expected when another process // holds a read lock when obtaining a write lock. This test will hold the lock // for some amount of time and then wait for the second process to send over the // socket_fd the amount of time it was blocked for before the lock succeeded. @@ -1109,8 +1148,7 @@ int main(int argc, char** argv) { fl.l_start = absl::GetFlag(FLAGS_child_setlock_start); fl.l_len = absl::GetFlag(FLAGS_child_setlock_len); - // Test the fcntl, no need to log, the error is unambiguously - // from fcntl at this point. + // Test the fcntl. int err = 0; int ret = 0; @@ -1123,6 +1161,8 @@ int main(int argc, char** argv) { if (ret == -1 && errno != 0) { err = errno; + std::cerr << "CHILD lock " << setlock_on << " failed " << err + << std::endl; } // If there is a socket fd let's send back the time in microseconds it took -- cgit v1.2.3 From e5d97cbcc1e64185b8fab1cf563c8754edd2e52e Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Wed, 17 Jun 2020 11:32:23 -0700 Subject: Refactor host.canMap. Simplify the canMap check. We do not have plans to allow mmap for anything beyond regular files, so we can just inline canMap() as a simple file mode check. Updates #1672. PiperOrigin-RevId: 316929654 --- pkg/sentry/fsimpl/host/host.go | 4 +++- pkg/sentry/fsimpl/host/util.go | 10 ---------- 2 files changed, 3 insertions(+), 11 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 7906242c9..2b084860e 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -91,7 +91,9 @@ func NewFD(ctx context.Context, mnt *vfs.Mount, hostFD int, opts *NewFDOptions) isTTY: opts.IsTTY, wouldBlock: wouldBlock(uint32(fileType)), seekable: seekable, - canMap: canMap(uint32(fileType)), + // NOTE(b/38213152): Technically, some obscure char devices can be memory + // mapped, but we only allow regular files. + canMap: fileType == linux.S_IFREG, } i.pf.inode = i diff --git a/pkg/sentry/fsimpl/host/util.go b/pkg/sentry/fsimpl/host/util.go index 2bc757b1a..412bdb2eb 100644 --- a/pkg/sentry/fsimpl/host/util.go +++ b/pkg/sentry/fsimpl/host/util.go @@ -49,16 +49,6 @@ func wouldBlock(fileType uint32) bool { return fileType == syscall.S_IFIFO || fileType == syscall.S_IFCHR || fileType == syscall.S_IFSOCK } -// canMap returns true if a file with fileType is allowed to be memory mapped. -// This is ported over from VFS1, but it's probably not the best way for us -// to check if a file can be memory mapped. -func canMap(fileType uint32) bool { - // TODO(gvisor.dev/issue/1672): Also allow "special files" to be mapped (see fs/host:canMap()). - // - // TODO(b/38213152): Some obscure character devices can be mapped. - return fileType == syscall.S_IFREG -} - // isBlockError checks if an error is EAGAIN or EWOULDBLOCK. // If so, they can be transformed into syserror.ErrWouldBlock. func isBlockError(err error) bool { -- cgit v1.2.3 From 6d806ee7198422973a2e4efa9b539de7792b933f Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Wed, 17 Jun 2020 12:32:59 -0700 Subject: Remove various uses of 'blacklist' Updates #2972 PiperOrigin-RevId: 316942245 --- pkg/sentry/fs/host/tty.go | 6 +++--- pkg/sentry/fsimpl/host/tty.go | 6 +++--- test/runner/runner.go | 6 +++--- tools/nogo/matchers.go | 2 +- 4 files changed, 10 insertions(+), 10 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/host/tty.go b/pkg/sentry/fs/host/tty.go index cb91355ab..82a02fcb2 100644 --- a/pkg/sentry/fs/host/tty.go +++ b/pkg/sentry/fs/host/tty.go @@ -308,9 +308,9 @@ func (t *TTYFileOperations) checkChange(ctx context.Context, sig linux.Signal) e task := kernel.TaskFromContext(ctx) if task == nil { // No task? Linux does not have an analog for this case, but - // tty_check_change is more of a blacklist of cases than a - // whitelist, and is surprisingly permissive. Allowing the - // change seems most appropriate. + // tty_check_change only blocks specific cases and is + // surprisingly permissive. Allowing the change seems + // appropriate. return nil } diff --git a/pkg/sentry/fsimpl/host/tty.go b/pkg/sentry/fsimpl/host/tty.go index 0fbc543b1..4ee9270cc 100644 --- a/pkg/sentry/fsimpl/host/tty.go +++ b/pkg/sentry/fsimpl/host/tty.go @@ -326,9 +326,9 @@ func (t *TTYFileDescription) checkChange(ctx context.Context, sig linux.Signal) task := kernel.TaskFromContext(ctx) if task == nil { // No task? Linux does not have an analog for this case, but - // tty_check_change is more of a blacklist of cases than a - // whitelist, and is surprisingly permissive. Allowing the - // change seems most appropriate. + // tty_check_change only blocks specific cases and is + // surprisingly permissive. Allowing the change seems + // appropriate. return nil } diff --git a/test/runner/runner.go b/test/runner/runner.go index f8baf61b0..5456e46a6 100644 --- a/test/runner/runner.go +++ b/test/runner/runner.go @@ -391,12 +391,12 @@ func runTestCaseRunsc(testBin string, tc gtest.TestCase, t *testing.T) { } } -// filterEnv returns an environment with the blacklisted variables removed. -func filterEnv(env, blacklist []string) []string { +// filterEnv returns an environment with the excluded variables removed. +func filterEnv(env, exclude []string) []string { var out []string for _, kv := range env { ok := true - for _, k := range blacklist { + for _, k := range exclude { if strings.HasPrefix(kv, k+"=") { ok = false break diff --git a/tools/nogo/matchers.go b/tools/nogo/matchers.go index bc5772303..32f099925 100644 --- a/tools/nogo/matchers.go +++ b/tools/nogo/matchers.go @@ -89,7 +89,7 @@ func (r resultExcluded) ShouldReport(d analysis.Diagnostic, _ *token.FileSet) bo return false } } - return true // Not blacklisted. + return true // Not excluded. } // andMatcher is a composite matcher. -- cgit v1.2.3 From a5f4deeca76b83745c0bae5c2b300faf5776b640 Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Wed, 17 Jun 2020 15:08:55 -0700 Subject: Internal change. PiperOrigin-RevId: 316973783 --- pkg/sentry/control/logging.go | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/control/logging.go b/pkg/sentry/control/logging.go index 811f24324..8a500a515 100644 --- a/pkg/sentry/control/logging.go +++ b/pkg/sentry/control/logging.go @@ -70,8 +70,8 @@ type LoggingArgs struct { type Logging struct{} // Change will change the log level and strace arguments. Although -// this functions signature requires an error it never acctually -// return san error. It's required by the URPC interface. +// this functions signature requires an error it never actually +// returns an error. It's required by the URPC interface. // Additionally, it may look odd that this is the only method // attached to an empty struct but this is also part of how // URPC dispatches. -- cgit v1.2.3 From 6e0c170522279ca72119b17c41e2e1dc93c49d6a Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Wed, 17 Jun 2020 21:21:08 -0700 Subject: Implement Sync() to directories Updates #1035, #1199 PiperOrigin-RevId: 317028108 --- pkg/sentry/fsimpl/gofer/directory.go | 5 +++++ pkg/sentry/fsimpl/host/host.go | 3 ++- pkg/sentry/fsimpl/overlay/directory.go | 22 ++++++++++++++++++++++ pkg/sentry/fsimpl/tmpfs/regular_file.go | 5 ----- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 6 ++++++ 5 files changed, 35 insertions(+), 6 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/directory.go b/pkg/sentry/fsimpl/gofer/directory.go index b98218753..480510b2a 100644 --- a/pkg/sentry/fsimpl/gofer/directory.go +++ b/pkg/sentry/fsimpl/gofer/directory.go @@ -299,3 +299,8 @@ func (fd *directoryFD) Seek(ctx context.Context, offset int64, whence int32) (in return 0, syserror.EINVAL } } + +// Sync implements vfs.FileDescriptionImpl.Sync. +func (fd *directoryFD) Sync(ctx context.Context) error { + return fd.dentry().handle.sync(ctx) +} diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 2b084860e..a3a312edb 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -690,7 +690,8 @@ func (f *fileDescription) Seek(_ context.Context, offset int64, whence int32) (i // Sync implements FileDescriptionImpl. func (f *fileDescription) Sync(context.Context) error { - // TODO(gvisor.dev/issue/1672): Currently we do not support the SyncData optimization, so we always sync everything. + // TODO(gvisor.dev/issue/1672): Currently we do not support the SyncData + // optimization, so we always sync everything. return unix.Fsync(f.inode.hostFD) } diff --git a/pkg/sentry/fsimpl/overlay/directory.go b/pkg/sentry/fsimpl/overlay/directory.go index 6f47167d3..f5c2462a5 100644 --- a/pkg/sentry/fsimpl/overlay/directory.go +++ b/pkg/sentry/fsimpl/overlay/directory.go @@ -263,3 +263,25 @@ func (fd *directoryFD) Seek(ctx context.Context, offset int64, whence int32) (in return 0, syserror.EINVAL } } + +// Sync implements vfs.FileDescriptionImpl.Sync. Forwards sync to the upper +// layer, if there is one. The lower layer doesn't need to sync because it +// never changes. +func (fd *directoryFD) Sync(ctx context.Context) error { + d := fd.dentry() + if !d.isCopiedUp() { + return nil + } + vfsObj := d.fs.vfsfs.VirtualFilesystem() + pop := vfs.PathOperation{ + Root: d.upperVD, + Start: d.upperVD, + } + upperFD, err := vfsObj.OpenAt(ctx, d.fs.creds, &pop, &vfs.OpenOptions{Flags: linux.O_RDONLY | linux.O_DIRECTORY}) + if err != nil { + return err + } + err = upperFD.Sync(ctx) + upperFD.DecRef() + return err +} diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index 77447b32c..bfd9c5995 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -360,11 +360,6 @@ func (fd *regularFileFD) Seek(ctx context.Context, offset int64, whence int32) ( return offset, nil } -// Sync implements vfs.FileDescriptionImpl.Sync. -func (fd *regularFileFD) Sync(ctx context.Context) error { - return nil -} - // ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. func (fd *regularFileFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { file := fd.inode().impl.(*regularFile) diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index d0a3e1a5c..a94333ee0 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -778,3 +778,9 @@ func (fd *fileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t func (fd *fileDescription) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { return fd.Locks().UnlockPOSIX(ctx, &fd.vfsfd, uid, start, length, whence) } + +// Sync implements vfs.FileDescriptionImpl.Sync. It does nothing because all +// filesystem state is in-memory. +func (*fileDescription) Sync(context.Context) error { + return nil +} -- cgit v1.2.3 From 07ff909e76d8233827e705476ec116fc2cecec2f Mon Sep 17 00:00:00 2001 From: Bhasker Hariharan Date: Thu, 18 Jun 2020 06:05:47 -0700 Subject: Support setsockopt SO_SNDBUF/SO_RCVBUF for raw/udp sockets. Updates #173,#6 Fixes #2888 PiperOrigin-RevId: 317087652 --- benchmarks/tcp/tcp_proxy.go | 8 +- pkg/sentry/socket/netstack/stack.go | 12 +- pkg/tcpip/tcpip.go | 26 ++ pkg/tcpip/transport/raw/endpoint.go | 72 +++- pkg/tcpip/transport/tcp/connect.go | 2 +- pkg/tcpip/transport/tcp/endpoint.go | 16 +- pkg/tcpip/transport/tcp/endpoint_state.go | 2 +- pkg/tcpip/transport/tcp/protocol.go | 59 ++-- pkg/tcpip/transport/tcp/tcp_sack_test.go | 4 +- pkg/tcpip/transport/tcp/tcp_test.go | 26 +- pkg/tcpip/transport/tcp/testing/context/context.go | 10 +- pkg/tcpip/transport/udp/endpoint.go | 57 +++- pkg/tcpip/transport/udp/protocol.go | 70 +++- runsc/boot/loader.go | 6 +- test/syscalls/linux/raw_socket_ipv4.cc | 379 +++++++++++++++++++++ test/syscalls/linux/socket_ipv4_udp_unbound.cc | 216 ++++++++++++ test/syscalls/linux/udp_socket_test_cases.cc | 115 +++++++ 17 files changed, 984 insertions(+), 96 deletions(-) (limited to 'pkg/sentry') diff --git a/benchmarks/tcp/tcp_proxy.go b/benchmarks/tcp/tcp_proxy.go index f5aa0b515..b3a4dbea3 100644 --- a/benchmarks/tcp/tcp_proxy.go +++ b/benchmarks/tcp/tcp_proxy.go @@ -228,19 +228,19 @@ func newNetstackImpl(mode string) (impl, error) { }) // Set protocol options. - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SACKEnabled(*sack)); err != nil { - return nil, fmt.Errorf("SetTransportProtocolOption for SACKEnabled failed: %v", err) + if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackSACKEnabled(*sack)); err != nil { + return nil, fmt.Errorf("SetTransportProtocolOption for SACKEnabled failed: %s", err) } // Enable Receive Buffer Auto-Tuning. if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.ModerateReceiveBufferOption(*moderateRecvBuf)); err != nil { - return nil, fmt.Errorf("SetTransportProtocolOption failed: %v", err) + return nil, fmt.Errorf("SetTransportProtocolOption failed: %s", err) } // Set Congestion Control to cubic if requested. if *cubic { if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.CongestionControlOption("cubic")); err != nil { - return nil, fmt.Errorf("SetTransportProtocolOption for CongestionControlOption(cubic) failed: %v", err) + return nil, fmt.Errorf("SetTransportProtocolOption for CongestionControlOption(cubic) failed: %s", err) } } diff --git a/pkg/sentry/socket/netstack/stack.go b/pkg/sentry/socket/netstack/stack.go index 9b44c2b89..f97f9b6f3 100644 --- a/pkg/sentry/socket/netstack/stack.go +++ b/pkg/sentry/socket/netstack/stack.go @@ -144,7 +144,7 @@ func (s *Stack) AddInterfaceAddr(idx int32, addr inet.InterfaceAddr) error { // TCPReceiveBufferSize implements inet.Stack.TCPReceiveBufferSize. func (s *Stack) TCPReceiveBufferSize() (inet.TCPBufferSize, error) { - var rs tcp.ReceiveBufferSizeOption + var rs tcpip.StackReceiveBufferSizeOption err := s.Stack.TransportProtocolOption(tcp.ProtocolNumber, &rs) return inet.TCPBufferSize{ Min: rs.Min, @@ -155,7 +155,7 @@ func (s *Stack) TCPReceiveBufferSize() (inet.TCPBufferSize, error) { // SetTCPReceiveBufferSize implements inet.Stack.SetTCPReceiveBufferSize. func (s *Stack) SetTCPReceiveBufferSize(size inet.TCPBufferSize) error { - rs := tcp.ReceiveBufferSizeOption{ + rs := tcpip.StackReceiveBufferSizeOption{ Min: size.Min, Default: size.Default, Max: size.Max, @@ -165,7 +165,7 @@ func (s *Stack) SetTCPReceiveBufferSize(size inet.TCPBufferSize) error { // TCPSendBufferSize implements inet.Stack.TCPSendBufferSize. func (s *Stack) TCPSendBufferSize() (inet.TCPBufferSize, error) { - var ss tcp.SendBufferSizeOption + var ss tcpip.StackSendBufferSizeOption err := s.Stack.TransportProtocolOption(tcp.ProtocolNumber, &ss) return inet.TCPBufferSize{ Min: ss.Min, @@ -176,7 +176,7 @@ func (s *Stack) TCPSendBufferSize() (inet.TCPBufferSize, error) { // SetTCPSendBufferSize implements inet.Stack.SetTCPSendBufferSize. func (s *Stack) SetTCPSendBufferSize(size inet.TCPBufferSize) error { - ss := tcp.SendBufferSizeOption{ + ss := tcpip.StackSendBufferSizeOption{ Min: size.Min, Default: size.Default, Max: size.Max, @@ -186,14 +186,14 @@ func (s *Stack) SetTCPSendBufferSize(size inet.TCPBufferSize) error { // TCPSACKEnabled implements inet.Stack.TCPSACKEnabled. func (s *Stack) TCPSACKEnabled() (bool, error) { - var sack tcp.SACKEnabled + var sack tcpip.StackSACKEnabled err := s.Stack.TransportProtocolOption(tcp.ProtocolNumber, &sack) return bool(sack), syserr.TranslateNetstackError(err).ToError() } // SetTCPSACKEnabled implements inet.Stack.SetTCPSACKEnabled. func (s *Stack) SetTCPSACKEnabled(enabled bool) error { - return syserr.TranslateNetstackError(s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SACKEnabled(enabled))).ToError() + return syserr.TranslateNetstackError(s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackSACKEnabled(enabled))).ToError() } // Statistics implements inet.Stack.Statistics. diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index b7b227328..3ad130b23 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -813,6 +813,32 @@ type OutOfBandInlineOption int // a default TTL. type DefaultTTLOption uint8 +// StackSACKEnabled is used by stack.(*Stack).TransportProtocolOption to +// enable/disable SACK support in TCP. See: https://tools.ietf.org/html/rfc2018. +type StackSACKEnabled bool + +// StackDelayEnabled is used by stack.(Stack*).TransportProtocolOption to +// enable/disable Nagle's algorithm in TCP. +type StackDelayEnabled bool + +// StackSendBufferSizeOption is used by stack.(Stack*).TransportProtocolOption +// to get/set the default, min and max send buffer sizes. +type StackSendBufferSizeOption struct { + Min int + Default int + Max int +} + +// StackReceiveBufferSizeOption is used by +// stack.(Stack*).TransportProtocolOption to get/set the default, min and max +// receive buffer sizes. +type StackReceiveBufferSizeOption struct { + Min int + Default int + Max int +} + +// // IPPacketInfo is the message struture for IP_PKTINFO. // // +stateify savable diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index a406d815e..6a7977259 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -26,6 +26,8 @@ package raw import ( + "fmt" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" @@ -66,16 +68,17 @@ type endpoint struct { // protected by rcvMu. rcvMu sync.Mutex `state:"nosave"` rcvList rawPacketList - rcvBufSizeMax int `state:".(int)"` rcvBufSize int + rcvBufSizeMax int `state:".(int)"` rcvClosed bool // The following fields are protected by mu. - mu sync.RWMutex `state:"nosave"` - sndBufSize int - closed bool - connected bool - bound bool + mu sync.RWMutex `state:"nosave"` + sndBufSize int + sndBufSizeMax int + closed bool + connected bool + bound bool // route is the route to a remote network endpoint. It is set via // Connect(), and is valid only when conneted is true. route stack.Route `state:"manual"` @@ -103,10 +106,21 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProt }, waiterQueue: waiterQueue, rcvBufSizeMax: 32 * 1024, - sndBufSize: 32 * 1024, + sndBufSizeMax: 32 * 1024, associated: associated, } + // Override with stack defaults. + var ss tcpip.StackSendBufferSizeOption + if err := s.TransportProtocolOption(transProto, &ss); err == nil { + e.sndBufSizeMax = ss.Default + } + + var rs tcpip.StackReceiveBufferSizeOption + if err := s.TransportProtocolOption(transProto, &rs); err == nil { + e.rcvBufSizeMax = rs.Default + } + // Unassociated endpoints are write-only and users call Write() with IP // headers included. Because they're write-only, We don't need to // register with the stack. @@ -523,7 +537,46 @@ func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { // SetSockOptInt implements tcpip.Endpoint.SetSockOptInt. func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { - return tcpip.ErrUnknownProtocolOption + switch opt { + case tcpip.SendBufferSizeOption: + // Make sure the send buffer size is within the min and max + // allowed. + var ss tcpip.StackSendBufferSizeOption + if err := e.stack.TransportProtocolOption(e.TransProto, &ss); err != nil { + panic(fmt.Sprintf("s.TransportProtocolOption(%d, %+v) = %s", e.TransProto, ss, err)) + } + if v > ss.Max { + v = ss.Max + } + if v < ss.Min { + v = ss.Min + } + e.mu.Lock() + e.sndBufSizeMax = v + e.mu.Unlock() + return nil + + case tcpip.ReceiveBufferSizeOption: + // Make sure the receive buffer size is within the min and max + // allowed. + var rs tcpip.StackReceiveBufferSizeOption + if err := e.stack.TransportProtocolOption(e.TransProto, &rs); err != nil { + panic(fmt.Sprintf("s.TransportProtocolOption(%d, %+v) = %s", e.TransProto, rs, err)) + } + if v > rs.Max { + v = rs.Max + } + if v < rs.Min { + v = rs.Min + } + e.rcvMu.Lock() + e.rcvBufSizeMax = v + e.rcvMu.Unlock() + return nil + + default: + return tcpip.ErrUnknownProtocolOption + } } // GetSockOpt implements tcpip.Endpoint.GetSockOpt. @@ -563,7 +616,7 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { case tcpip.SendBufferSizeOption: e.mu.Lock() - v := e.sndBufSize + v := e.sndBufSizeMax e.mu.Unlock() return v, nil @@ -636,7 +689,6 @@ func (e *endpoint) HandlePacket(route *stack.Route, pkt *stack.PacketBuffer) { e.rcvList.PushBack(packet) e.rcvBufSize += packet.data.Size() - e.rcvMu.Unlock() e.stats.PacketsReceived.Increment() // Notify waiters that there's data to be read. diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 9d4dce826..377643b82 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -521,7 +521,7 @@ func (h *handshake) execute() *tcpip.Error { s.AddWaker(&h.ep.newSegmentWaker, wakerForNewSegment) defer s.Done() - var sackEnabled SACKEnabled + var sackEnabled tcpip.StackSACKEnabled if err := h.ep.stack.TransportProtocolOption(ProtocolNumber, &sackEnabled); err != nil { // If stack returned an error when checking for SACKEnabled // status then just default to switching off SACK negotiation. diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index f225b00e7..10df2bcd5 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -851,12 +851,12 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue maxSynRetries: DefaultSynRetries, } - var ss SendBufferSizeOption + var ss tcpip.StackSendBufferSizeOption if err := s.TransportProtocolOption(ProtocolNumber, &ss); err == nil { e.sndBufSize = ss.Default } - var rs ReceiveBufferSizeOption + var rs tcpip.StackReceiveBufferSizeOption if err := s.TransportProtocolOption(ProtocolNumber, &rs); err == nil { e.rcvBufSize = rs.Default } @@ -871,7 +871,7 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue e.rcvAutoParams.disabled = !bool(mrb) } - var de DelayEnabled + var de tcpip.StackDelayEnabled if err := s.TransportProtocolOption(ProtocolNumber, &de); err == nil && de { e.SetSockOptBool(tcpip.DelayOption, true) } @@ -1588,7 +1588,7 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { case tcpip.ReceiveBufferSizeOption: // Make sure the receive buffer size is within the min and max // allowed. - var rs ReceiveBufferSizeOption + var rs tcpip.StackReceiveBufferSizeOption if err := e.stack.TransportProtocolOption(ProtocolNumber, &rs); err == nil { if v < rs.Min { v = rs.Min @@ -1638,7 +1638,7 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { case tcpip.SendBufferSizeOption: // Make sure the send buffer size is within the min and max // allowed. - var ss SendBufferSizeOption + var ss tcpip.StackSendBufferSizeOption if err := e.stack.TransportProtocolOption(ProtocolNumber, &ss); err == nil { if v < ss.Min { v = ss.Min @@ -1678,7 +1678,7 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { return tcpip.ErrInvalidOptionValue } } - var rs ReceiveBufferSizeOption + var rs tcpip.StackReceiveBufferSizeOption if err := e.stack.TransportProtocolOption(ProtocolNumber, &rs); err == nil { if v < rs.Min/2 { v = rs.Min / 2 @@ -2609,7 +2609,7 @@ func (e *endpoint) receiveBufferSize() int { } func (e *endpoint) maxReceiveBufferSize() int { - var rs ReceiveBufferSizeOption + var rs tcpip.StackReceiveBufferSizeOption if err := e.stack.TransportProtocolOption(ProtocolNumber, &rs); err != nil { // As a fallback return the hardcoded max buffer size. return MaxBufferSize @@ -2690,7 +2690,7 @@ func timeStampOffset() uint32 { // if the SYN options indicate that the SACK option was negotiated and the TCP // stack is configured to enable TCP SACK option. func (e *endpoint) maybeEnableSACKPermitted(synOpts *header.TCPSynOptions) { - var v SACKEnabled + var v tcpip.StackSACKEnabled if err := e.stack.TransportProtocolOption(ProtocolNumber, &v); err != nil { // Stack doesn't support SACK. So just return. return diff --git a/pkg/tcpip/transport/tcp/endpoint_state.go b/pkg/tcpip/transport/tcp/endpoint_state.go index cbb779666..0bebec2d1 100644 --- a/pkg/tcpip/transport/tcp/endpoint_state.go +++ b/pkg/tcpip/transport/tcp/endpoint_state.go @@ -186,7 +186,7 @@ func (e *endpoint) Resume(s *stack.Stack) { epState := e.origEndpointState switch epState { case StateInitial, StateBound, StateListen, StateConnecting, StateEstablished: - var ss SendBufferSizeOption + var ss tcpip.StackSendBufferSizeOption if err := e.stack.TransportProtocolOption(ProtocolNumber, &ss); err == nil { if e.sndBufSize < ss.Min || e.sndBufSize > ss.Max { panic(fmt.Sprintf("endpoint.sndBufSize %d is outside the min and max allowed [%d, %d]", e.sndBufSize, ss.Min, ss.Max)) diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go index 73b8a6782..3cff55afa 100644 --- a/pkg/tcpip/transport/tcp/protocol.go +++ b/pkg/tcpip/transport/tcp/protocol.go @@ -71,29 +71,6 @@ const ( DefaultSynRetries = 6 ) -// SACKEnabled option can be used to enable SACK support in the TCP -// protocol. See: https://tools.ietf.org/html/rfc2018. -type SACKEnabled bool - -// DelayEnabled option can be used to enable Nagle's algorithm in the TCP protocol. -type DelayEnabled bool - -// SendBufferSizeOption allows the default, min and max send buffer sizes for -// TCP endpoints to be queried or configured. -type SendBufferSizeOption struct { - Min int - Default int - Max int -} - -// ReceiveBufferSizeOption allows the default, min and max receive buffer size -// for TCP endpoints to be queried or configured. -type ReceiveBufferSizeOption struct { - Min int - Default int - Max int -} - const ( ccReno = "reno" ccCubic = "cubic" @@ -160,8 +137,8 @@ type protocol struct { mu sync.RWMutex sackEnabled bool delayEnabled bool - sendBufferSize SendBufferSizeOption - recvBufferSize ReceiveBufferSizeOption + sendBufferSize tcpip.StackSendBufferSizeOption + recvBufferSize tcpip.StackReceiveBufferSizeOption congestionControl string availableCongestionControl []string moderateReceiveBuffer bool @@ -272,19 +249,19 @@ func replyWithReset(s *segment, tos, ttl uint8) { // SetOption implements stack.TransportProtocol.SetOption. func (p *protocol) SetOption(option interface{}) *tcpip.Error { switch v := option.(type) { - case SACKEnabled: + case tcpip.StackSACKEnabled: p.mu.Lock() p.sackEnabled = bool(v) p.mu.Unlock() return nil - case DelayEnabled: + case tcpip.StackDelayEnabled: p.mu.Lock() p.delayEnabled = bool(v) p.mu.Unlock() return nil - case SendBufferSizeOption: + case tcpip.StackSendBufferSizeOption: if v.Min <= 0 || v.Default < v.Min || v.Default > v.Max { return tcpip.ErrInvalidOptionValue } @@ -293,7 +270,7 @@ func (p *protocol) SetOption(option interface{}) *tcpip.Error { p.mu.Unlock() return nil - case ReceiveBufferSizeOption: + case tcpip.StackReceiveBufferSizeOption: if v.Min <= 0 || v.Default < v.Min || v.Default > v.Max { return tcpip.ErrInvalidOptionValue } @@ -386,25 +363,25 @@ func (p *protocol) SetOption(option interface{}) *tcpip.Error { // Option implements stack.TransportProtocol.Option. func (p *protocol) Option(option interface{}) *tcpip.Error { switch v := option.(type) { - case *SACKEnabled: + case *tcpip.StackSACKEnabled: p.mu.RLock() - *v = SACKEnabled(p.sackEnabled) + *v = tcpip.StackSACKEnabled(p.sackEnabled) p.mu.RUnlock() return nil - case *DelayEnabled: + case *tcpip.StackDelayEnabled: p.mu.RLock() - *v = DelayEnabled(p.delayEnabled) + *v = tcpip.StackDelayEnabled(p.delayEnabled) p.mu.RUnlock() return nil - case *SendBufferSizeOption: + case *tcpip.StackSendBufferSizeOption: p.mu.RLock() *v = p.sendBufferSize p.mu.RUnlock() return nil - case *ReceiveBufferSizeOption: + case *tcpip.StackReceiveBufferSizeOption: p.mu.RLock() *v = p.recvBufferSize p.mu.RUnlock() @@ -514,8 +491,16 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) bool { // NewProtocol returns a TCP transport protocol. func NewProtocol() stack.TransportProtocol { return &protocol{ - sendBufferSize: SendBufferSizeOption{MinBufferSize, DefaultSendBufferSize, MaxBufferSize}, - recvBufferSize: ReceiveBufferSizeOption{MinBufferSize, DefaultReceiveBufferSize, MaxBufferSize}, + sendBufferSize: tcpip.StackSendBufferSizeOption{ + Min: MinBufferSize, + Default: DefaultSendBufferSize, + Max: MaxBufferSize, + }, + recvBufferSize: tcpip.StackReceiveBufferSizeOption{ + Min: MinBufferSize, + Default: DefaultReceiveBufferSize, + Max: MaxBufferSize, + }, congestionControl: ccReno, availableCongestionControl: []string{ccReno, ccCubic}, tcpLingerTimeout: DefaultTCPLingerTimeout, diff --git a/pkg/tcpip/transport/tcp/tcp_sack_test.go b/pkg/tcpip/transport/tcp/tcp_sack_test.go index fcc165f17..812e503bc 100644 --- a/pkg/tcpip/transport/tcp/tcp_sack_test.go +++ b/pkg/tcpip/transport/tcp/tcp_sack_test.go @@ -46,8 +46,8 @@ func createConnectedWithSACKAndTS(c *context.Context) *context.RawEndpoint { func setStackSACKPermitted(t *testing.T, c *context.Context, enable bool) { t.Helper() - if err := c.Stack().SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SACKEnabled(enable)); err != nil { - t.Fatalf("c.s.SetTransportProtocolOption(tcp.ProtocolNumber, SACKEnabled(%v) = %v", enable, err) + if err := c.Stack().SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackSACKEnabled(enable)); err != nil { + t.Fatalf("c.s.SetTransportProtocolOption(tcp.ProtocolNumber, StackSACKEnabled(%t) = %s", enable, err) } } diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index 0668cedc9..aca6a7951 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -3987,7 +3987,10 @@ func TestDefaultBufferSizes(t *testing.T) { checkRecvBufferSize(t, ep, tcp.DefaultReceiveBufferSize) // Change the default send buffer size. - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SendBufferSizeOption{1, tcp.DefaultSendBufferSize * 2, tcp.DefaultSendBufferSize * 20}); err != nil { + if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackSendBufferSizeOption{ + Min: 1, + Default: tcp.DefaultSendBufferSize * 2, + Max: tcp.DefaultSendBufferSize * 20}); err != nil { t.Fatalf("SetTransportProtocolOption failed: %s", err) } @@ -4001,8 +4004,11 @@ func TestDefaultBufferSizes(t *testing.T) { checkRecvBufferSize(t, ep, tcp.DefaultReceiveBufferSize) // Change the default receive buffer size. - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.ReceiveBufferSizeOption{1, tcp.DefaultReceiveBufferSize * 3, tcp.DefaultReceiveBufferSize * 30}); err != nil { - t.Fatalf("SetTransportProtocolOption failed: %s", err) + if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackReceiveBufferSizeOption{ + Min: 1, + Default: tcp.DefaultReceiveBufferSize * 3, + Max: tcp.DefaultReceiveBufferSize * 30}); err != nil { + t.Fatalf("SetTransportProtocolOption failed: %v", err) } ep.Close() @@ -4029,11 +4035,11 @@ func TestMinMaxBufferSizes(t *testing.T) { defer ep.Close() // Change the min/max values for send/receive - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.ReceiveBufferSizeOption{200, tcp.DefaultReceiveBufferSize * 2, tcp.DefaultReceiveBufferSize * 20}); err != nil { + if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackReceiveBufferSizeOption{Min: 200, Default: tcp.DefaultReceiveBufferSize * 2, Max: tcp.DefaultReceiveBufferSize * 20}); err != nil { t.Fatalf("SetTransportProtocolOption failed: %s", err) } - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SendBufferSizeOption{300, tcp.DefaultSendBufferSize * 3, tcp.DefaultSendBufferSize * 30}); err != nil { + if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackSendBufferSizeOption{Min: 300, Default: tcp.DefaultSendBufferSize * 3, Max: tcp.DefaultSendBufferSize * 30}); err != nil { t.Fatalf("SetTransportProtocolOption failed: %s", err) } @@ -5672,7 +5678,7 @@ func TestReceiveBufferAutoTuningApplicationLimited(t *testing.T) { // the segment queue holding unprocessed packets is limited to 500. const receiveBufferSize = 80 << 10 // 80KB. const maxReceiveBufferSize = receiveBufferSize * 10 - if err := stk.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.ReceiveBufferSizeOption{1, receiveBufferSize, maxReceiveBufferSize}); err != nil { + if err := stk.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackReceiveBufferSizeOption{Min: 1, Default: receiveBufferSize, Max: maxReceiveBufferSize}); err != nil { t.Fatalf("SetTransportProtocolOption failed: %s", err) } @@ -5793,7 +5799,7 @@ func TestReceiveBufferAutoTuning(t *testing.T) { // the segment queue holding unprocessed packets is limited to 300. const receiveBufferSize = 80 << 10 // 80KB. const maxReceiveBufferSize = receiveBufferSize * 10 - if err := stk.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.ReceiveBufferSizeOption{1, receiveBufferSize, maxReceiveBufferSize}); err != nil { + if err := stk.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackReceiveBufferSizeOption{Min: 1, Default: receiveBufferSize, Max: maxReceiveBufferSize}); err != nil { t.Fatalf("SetTransportProtocolOption failed: %s", err) } @@ -5935,7 +5941,7 @@ func TestDelayEnabled(t *testing.T) { checkDelayOption(t, c, false, false) // Delay is disabled by default. for _, v := range []struct { - delayEnabled tcp.DelayEnabled + delayEnabled tcpip.StackDelayEnabled wantDelayOption bool }{ {delayEnabled: false, wantDelayOption: false}, @@ -5950,10 +5956,10 @@ func TestDelayEnabled(t *testing.T) { } } -func checkDelayOption(t *testing.T, c *context.Context, wantDelayEnabled tcp.DelayEnabled, wantDelayOption bool) { +func checkDelayOption(t *testing.T, c *context.Context, wantDelayEnabled tcpip.StackDelayEnabled, wantDelayOption bool) { t.Helper() - var gotDelayEnabled tcp.DelayEnabled + var gotDelayEnabled tcpip.StackDelayEnabled if err := c.Stack().TransportProtocolOption(tcp.ProtocolNumber, &gotDelayEnabled); err != nil { t.Fatalf("TransportProtocolOption(tcp, &gotDelayEnabled) failed: %s", err) } diff --git a/pkg/tcpip/transport/tcp/testing/context/context.go b/pkg/tcpip/transport/tcp/testing/context/context.go index 9721f6caf..9e262c272 100644 --- a/pkg/tcpip/transport/tcp/testing/context/context.go +++ b/pkg/tcpip/transport/tcp/testing/context/context.go @@ -144,12 +144,12 @@ func New(t *testing.T, mtu uint32) *Context { }) // Allow minimum send/receive buffer sizes to be 1 during tests. - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SendBufferSizeOption{1, tcp.DefaultSendBufferSize, 10 * tcp.DefaultSendBufferSize}); err != nil { - t.Fatalf("SetTransportProtocolOption failed: %v", err) + if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackSendBufferSizeOption{Min: 1, Default: tcp.DefaultSendBufferSize, Max: 10 * tcp.DefaultSendBufferSize}); err != nil { + t.Fatalf("SetTransportProtocolOption failed: %s", err) } - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.ReceiveBufferSizeOption{1, tcp.DefaultReceiveBufferSize, 10 * tcp.DefaultReceiveBufferSize}); err != nil { - t.Fatalf("SetTransportProtocolOption failed: %v", err) + if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackReceiveBufferSizeOption{Min: 1, Default: tcp.DefaultReceiveBufferSize, Max: 10 * tcp.DefaultReceiveBufferSize}); err != nil { + t.Fatalf("SetTransportProtocolOption failed: %s", err) } // Increase minimum RTO in tests to avoid test flakes due to early @@ -1091,7 +1091,7 @@ func (c *Context) PassiveConnectWithOptions(maxPayload, wndScale int, synOptions // SACKEnabled returns true if the TCP Protocol option SACKEnabled is set to true // for the Stack in the context. func (c *Context) SACKEnabled() bool { - var v tcp.SACKEnabled + var v tcpip.StackSACKEnabled if err := c.Stack().TransportProtocolOption(tcp.ProtocolNumber, &v); err != nil { // Stack doesn't support SACK. So just return. return false diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index df5efbf6a..f51988047 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -15,6 +15,8 @@ package udp import ( + "fmt" + "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" @@ -94,6 +96,7 @@ type endpoint struct { // The following fields are protected by the mu mutex. mu sync.RWMutex `state:"nosave"` sndBufSize int + sndBufSizeMax int state EndpointState route stack.Route `state:"manual"` dstPort uint16 @@ -159,7 +162,7 @@ type multicastMembership struct { } func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) *endpoint { - return &endpoint{ + e := &endpoint{ stack: s, TransportEndpointInfo: stack.TransportEndpointInfo{ NetProto: netProto, @@ -181,10 +184,23 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue multicastTTL: 1, multicastLoop: true, rcvBufSizeMax: 32 * 1024, - sndBufSize: 32 * 1024, + sndBufSizeMax: 32 * 1024, state: StateInitial, uniqueID: s.UniqueID(), } + + // Override with stack defaults. + var ss tcpip.StackSendBufferSizeOption + if err := s.TransportProtocolOption(ProtocolNumber, &ss); err == nil { + e.sndBufSizeMax = ss.Default + } + + var rs tcpip.StackReceiveBufferSizeOption + if err := s.TransportProtocolOption(ProtocolNumber, &rs); err == nil { + e.rcvBufSizeMax = rs.Default + } + + return e } // UniqueID implements stack.TransportEndpoint.UniqueID. @@ -611,8 +627,43 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { e.mu.Unlock() case tcpip.ReceiveBufferSizeOption: + // Make sure the receive buffer size is within the min and max + // allowed. + var rs tcpip.StackReceiveBufferSizeOption + if err := e.stack.TransportProtocolOption(ProtocolNumber, &rs); err != nil { + panic(fmt.Sprintf("e.stack.TransportProtocolOption(%d, %+v) = %s", ProtocolNumber, rs, err)) + } + + if v < rs.Min { + v = rs.Min + } + if v > rs.Max { + v = rs.Max + } + + e.mu.Lock() + e.rcvBufSizeMax = v + e.mu.Unlock() + return nil case tcpip.SendBufferSizeOption: + // Make sure the send buffer size is within the min and max + // allowed. + var ss tcpip.StackSendBufferSizeOption + if err := e.stack.TransportProtocolOption(ProtocolNumber, &ss); err != nil { + panic(fmt.Sprintf("e.stack.TransportProtocolOption(%d, %+v) = %s", ProtocolNumber, ss, err)) + } + + if v < ss.Min { + v = ss.Min + } + if v > ss.Max { + v = ss.Max + } + e.mu.Lock() + e.sndBufSizeMax = v + e.mu.Unlock() + return nil } return nil @@ -861,7 +912,7 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { case tcpip.SendBufferSizeOption: e.mu.Lock() - v := e.sndBufSize + v := e.sndBufSizeMax e.mu.Unlock() return v, nil diff --git a/pkg/tcpip/transport/udp/protocol.go b/pkg/tcpip/transport/udp/protocol.go index 4218e7d03..fc93f93c0 100644 --- a/pkg/tcpip/transport/udp/protocol.go +++ b/pkg/tcpip/transport/udp/protocol.go @@ -21,6 +21,7 @@ package udp import ( + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" @@ -32,9 +33,27 @@ import ( const ( // ProtocolNumber is the udp protocol number. ProtocolNumber = header.UDPProtocolNumber + + // MinBufferSize is the smallest size of a receive or send buffer. + MinBufferSize = 4 << 10 // 4KiB bytes. + + // DefaultSendBufferSize is the default size of the send buffer for + // an endpoint. + DefaultSendBufferSize = 32 << 10 // 32KiB + + // DefaultReceiveBufferSize is the default size of the receive buffer + // for an endpoint. + DefaultReceiveBufferSize = 32 << 10 // 32KiB + + // MaxBufferSize is the largest size a receive/send buffer can grow to. + MaxBufferSize = 4 << 20 // 4MiB ) -type protocol struct{} +type protocol struct { + mu sync.RWMutex + sendBufferSize tcpip.StackSendBufferSizeOption + recvBufferSize tcpip.StackReceiveBufferSizeOption +} // Number returns the udp protocol number. func (*protocol) Number() tcpip.TransportProtocolNumber { @@ -183,13 +202,49 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans } // SetOption implements stack.TransportProtocol.SetOption. -func (*protocol) SetOption(option interface{}) *tcpip.Error { - return tcpip.ErrUnknownProtocolOption +func (p *protocol) SetOption(option interface{}) *tcpip.Error { + switch v := option.(type) { + case tcpip.StackSendBufferSizeOption: + if v.Min <= 0 || v.Default < v.Min || v.Default > v.Max { + return tcpip.ErrInvalidOptionValue + } + p.mu.Lock() + p.sendBufferSize = v + p.mu.Unlock() + return nil + + case tcpip.StackReceiveBufferSizeOption: + if v.Min <= 0 || v.Default < v.Min || v.Default > v.Max { + return tcpip.ErrInvalidOptionValue + } + p.mu.Lock() + p.recvBufferSize = v + p.mu.Unlock() + return nil + + default: + return tcpip.ErrUnknownProtocolOption + } } // Option implements stack.TransportProtocol.Option. -func (*protocol) Option(option interface{}) *tcpip.Error { - return tcpip.ErrUnknownProtocolOption +func (p *protocol) Option(option interface{}) *tcpip.Error { + switch v := option.(type) { + case *tcpip.StackSendBufferSizeOption: + p.mu.RLock() + *v = p.sendBufferSize + p.mu.RUnlock() + return nil + + case *tcpip.StackReceiveBufferSizeOption: + p.mu.RLock() + *v = p.recvBufferSize + p.mu.RUnlock() + return nil + + default: + return tcpip.ErrUnknownProtocolOption + } } // Close implements stack.TransportProtocol.Close. @@ -212,5 +267,8 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) bool { // NewProtocol returns a UDP transport protocol. func NewProtocol() stack.TransportProtocol { - return &protocol{} + return &protocol{ + sendBufferSize: tcpip.StackSendBufferSizeOption{Min: MinBufferSize, Default: DefaultSendBufferSize, Max: MaxBufferSize}, + recvBufferSize: tcpip.StackReceiveBufferSizeOption{Min: MinBufferSize, Default: DefaultReceiveBufferSize, Max: MaxBufferSize}, + } } diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index b05a8bd45..c6efcdc83 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -1058,8 +1058,8 @@ func newEmptySandboxNetworkStack(clock tcpip.Clock, uniqueID stack.UniqueID) (in })} // Enable SACK Recovery. - if err := s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SACKEnabled(true)); err != nil { - return nil, fmt.Errorf("failed to enable SACK: %v", err) + if err := s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackSACKEnabled(true)); err != nil { + return nil, fmt.Errorf("failed to enable SACK: %s", err) } // Set default TTLs as required by socket/netstack. @@ -1068,7 +1068,7 @@ func newEmptySandboxNetworkStack(clock tcpip.Clock, uniqueID stack.UniqueID) (in // Enable Receive Buffer Auto-Tuning. if err := s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.ModerateReceiveBufferOption(true)); err != nil { - return nil, fmt.Errorf("SetTransportProtocolOption failed: %v", err) + return nil, fmt.Errorf("SetTransportProtocolOption failed: %s", err) } s.FillIPTablesMetadata() diff --git a/test/syscalls/linux/raw_socket_ipv4.cc b/test/syscalls/linux/raw_socket_ipv4.cc index cde2f07c9..0116c3e94 100644 --- a/test/syscalls/linux/raw_socket_ipv4.cc +++ b/test/syscalls/linux/raw_socket_ipv4.cc @@ -357,10 +357,389 @@ TEST_P(RawSocketTest, BindConnectSendAndReceive) { EXPECT_EQ(memcmp(recv_buf + sizeof(struct iphdr), kBuf, sizeof(kBuf)), 0); } +// Check that setting SO_RCVBUF below min is clamped to the minimum +// receive buffer size. +TEST_P(RawSocketTest, SetSocketRecvBufBelowMin) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + // Discover minimum receive buf size by trying to set it to zero. + // See: + // https://github.com/torvalds/linux/blob/a5dc8300df75e8b8384b4c82225f1e4a0b4d9b55/net/core/sock.c#L820 + constexpr int kRcvBufSz = 0; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, sizeof(kRcvBufSz)), + SyscallSucceeds()); + + int min = 0; + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &min, &min_len), + SyscallSucceeds()); + + // Linux doubles the value so let's use a value that when doubled will still + // be smaller than min. + int below_min = min / 2 - 1; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &below_min, sizeof(below_min)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &val, &val_len), + SyscallSucceeds()); + + ASSERT_EQ(min, val); +} + +// Check that setting SO_RCVBUF above max is clamped to the maximum +// receive buffer size. +TEST_P(RawSocketTest, SetSocketRecvBufAboveMax) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + // Discover max buf size by trying to set the largest possible buffer size. + constexpr int kRcvBufSz = 0xffffffff; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, sizeof(kRcvBufSz)), + SyscallSucceeds()); + + int max = 0; + socklen_t max_len = sizeof(max); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &max, &max_len), + SyscallSucceeds()); + + int above_max = max + 1; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &above_max, sizeof(above_max)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &val, &val_len), + SyscallSucceeds()); + ASSERT_EQ(max, val); +} + +// Check that setting SO_RCVBUF min <= kRcvBufSz <= max is honored. +TEST_P(RawSocketTest, SetSocketRecvBuf) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + int max = 0; + int min = 0; + { + // Discover max buf size by trying to set a really large buffer size. + constexpr int kRcvBufSz = 0xffffffff; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, sizeof(kRcvBufSz)), + SyscallSucceeds()); + + max = 0; + socklen_t max_len = sizeof(max); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &max, &max_len), + SyscallSucceeds()); + } + + { + // Discover minimum buffer size by trying to set a zero size receive buffer + // size. + // See: + // https://github.com/torvalds/linux/blob/a5dc8300df75e8b8384b4c82225f1e4a0b4d9b55/net/core/sock.c#L820 + constexpr int kRcvBufSz = 0; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, sizeof(kRcvBufSz)), + SyscallSucceeds()); + + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &min, &min_len), + SyscallSucceeds()); + } + + int quarter_sz = min + (max - min) / 4; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &quarter_sz, sizeof(quarter_sz)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &val, &val_len), + SyscallSucceeds()); + + // Linux doubles the value set by SO_SNDBUF/SO_RCVBUF. + // TODO(gvisor.dev/issue/2926): Remove when Netstack matches linux behavior. + if (!IsRunningOnGvisor()) { + quarter_sz *= 2; + } + ASSERT_EQ(quarter_sz, val); +} + +// Check that setting SO_SNDBUF below min is clamped to the minimum +// receive buffer size. +TEST_P(RawSocketTest, SetSocketSendBufBelowMin) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + // Discover minimum buffer size by trying to set it to zero. + constexpr int kSndBufSz = 0; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &kSndBufSz, sizeof(kSndBufSz)), + SyscallSucceeds()); + + int min = 0; + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &min, &min_len), + SyscallSucceeds()); + + // Linux doubles the value so let's use a value that when doubled will still + // be smaller than min. + int below_min = min / 2 - 1; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &below_min, sizeof(below_min)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &val, &val_len), + SyscallSucceeds()); + + ASSERT_EQ(min, val); +} + +// Check that setting SO_SNDBUF above max is clamped to the maximum +// send buffer size. +TEST_P(RawSocketTest, SetSocketSendBufAboveMax) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + // Discover maximum buffer size by trying to set it to a large value. + constexpr int kSndBufSz = 0xffffffff; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &kSndBufSz, sizeof(kSndBufSz)), + SyscallSucceeds()); + + int max = 0; + socklen_t max_len = sizeof(max); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &max, &max_len), + SyscallSucceeds()); + + int above_max = max + 1; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &above_max, sizeof(above_max)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &val, &val_len), + SyscallSucceeds()); + ASSERT_EQ(max, val); +} + +// Check that setting SO_SNDBUF min <= kSndBufSz <= max is honored. +TEST_P(RawSocketTest, SetSocketSendBuf) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + int max = 0; + int min = 0; + { + // Discover maximum buffer size by trying to set it to a large value. + constexpr int kSndBufSz = 0xffffffff; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &kSndBufSz, sizeof(kSndBufSz)), + SyscallSucceeds()); + + max = 0; + socklen_t max_len = sizeof(max); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &max, &max_len), + SyscallSucceeds()); + } + + { + // Discover minimum buffer size by trying to set it to zero. + constexpr int kSndBufSz = 0; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &kSndBufSz, sizeof(kSndBufSz)), + SyscallSucceeds()); + + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &min, &min_len), + SyscallSucceeds()); + } + + int quarter_sz = min + (max - min) / 4; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &quarter_sz, sizeof(quarter_sz)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &val, &val_len), + SyscallSucceeds()); + + // Linux doubles the value set by SO_SNDBUF/SO_RCVBUF. + // TODO(gvisor.dev/issue/2926): Remove the gvisor special casing when Netstack + // matches linux behavior. + if (!IsRunningOnGvisor()) { + quarter_sz *= 2; + } + + ASSERT_EQ(quarter_sz, val); +} + void RawSocketTest::SendBuf(const char* buf, int buf_len) { ASSERT_NO_FATAL_FAILURE(SendBufTo(s_, addr_, buf, buf_len)); } +// Test that receive buffer limits are not enforced when the recv buffer is +// empty. +TEST_P(RawSocketTest, RecvBufLimitsEmptyRecvBuffer) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + ASSERT_THAT( + bind(s_, reinterpret_cast(&addr_), sizeof(addr_)), + SyscallSucceeds()); + ASSERT_THAT( + connect(s_, reinterpret_cast(&addr_), sizeof(addr_)), + SyscallSucceeds()); + + int min = 0; + { + // Discover minimum buffer size by trying to set it to zero. + constexpr int kRcvBufSz = 0; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, sizeof(kRcvBufSz)), + SyscallSucceeds()); + + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &min, &min_len), + SyscallSucceeds()); + } + + { + // Send data of size min and verify that it's received. + std::vector buf(min); + RandomizeBuffer(buf.data(), buf.size()); + ASSERT_NO_FATAL_FAILURE(SendBuf(buf.data(), buf.size())); + + // Receive the packet and make sure it's identical. + std::vector recv_buf(buf.size() + sizeof(struct iphdr)); + ASSERT_NO_FATAL_FAILURE(ReceiveBuf(recv_buf.data(), recv_buf.size())); + EXPECT_EQ( + memcmp(recv_buf.data() + sizeof(struct iphdr), buf.data(), buf.size()), + 0); + } + + { + // Send data of size min + 1 and verify that its received. Both linux and + // Netstack accept a dgram that exceeds rcvBuf limits if the receive buffer + // is currently empty. + std::vector buf(min + 1); + RandomizeBuffer(buf.data(), buf.size()); + ASSERT_NO_FATAL_FAILURE(SendBuf(buf.data(), buf.size())); + // Receive the packet and make sure it's identical. + std::vector recv_buf(buf.size() + sizeof(struct iphdr)); + ASSERT_NO_FATAL_FAILURE(ReceiveBuf(recv_buf.data(), recv_buf.size())); + EXPECT_EQ( + memcmp(recv_buf.data() + sizeof(struct iphdr), buf.data(), buf.size()), + 0); + } +} + +TEST_P(RawSocketTest, RecvBufLimits) { + // TCP stack generates RSTs for unknown endpoints and it complicates the test + // as we have to deal with the RST packets as well. For testing the raw socket + // endpoints buffer limit enforcement we can just test for UDP. + // + // We don't use SKIP_IF here because root_test_runner explicitly fails if a + // test is skipped. + if (Protocol() == IPPROTO_TCP) { + return; + } + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + ASSERT_THAT( + bind(s_, reinterpret_cast(&addr_), sizeof(addr_)), + SyscallSucceeds()); + ASSERT_THAT( + connect(s_, reinterpret_cast(&addr_), sizeof(addr_)), + SyscallSucceeds()); + + int min = 0; + { + // Discover minimum buffer size by trying to set it to zero. + constexpr int kRcvBufSz = 0; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, sizeof(kRcvBufSz)), + SyscallSucceeds()); + + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &min, &min_len), + SyscallSucceeds()); + } + + // Now set the limit to min * 2. + int new_rcv_buf_sz = min * 4; + if (!IsRunningOnGvisor()) { + // Linux doubles the value specified so just set to min. + new_rcv_buf_sz = min * 2; + } + + ASSERT_THAT(setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &new_rcv_buf_sz, + sizeof(new_rcv_buf_sz)), + SyscallSucceeds()); + int rcv_buf_sz = 0; + { + socklen_t rcv_buf_len = sizeof(rcv_buf_sz); + ASSERT_THAT( + getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &rcv_buf_sz, &rcv_buf_len), + SyscallSucceeds()); + } + + // Set a receive timeout so that we don't block forever on reads if the test + // fails. + struct timeval tv { + .tv_sec = 1, .tv_usec = 0, + }; + ASSERT_THAT(setsockopt(s_, SOL_SOCKET, SO_RCVTIMEO, &tv, sizeof(tv)), + SyscallSucceeds()); + + { + std::vector buf(min); + RandomizeBuffer(buf.data(), buf.size()); + + ASSERT_NO_FATAL_FAILURE(SendBuf(buf.data(), buf.size())); + ASSERT_NO_FATAL_FAILURE(SendBuf(buf.data(), buf.size())); + ASSERT_NO_FATAL_FAILURE(SendBuf(buf.data(), buf.size())); + ASSERT_NO_FATAL_FAILURE(SendBuf(buf.data(), buf.size())); + int sent = 4; + if (IsRunningOnGvisor()) { + // Linux seems to drop the 4th packet even though technically it should + // fit in the receive buffer. + ASSERT_NO_FATAL_FAILURE(SendBuf(buf.data(), buf.size())); + sent++; + } + + // Verify that the expected number of packets are available to be read. + for (int i = 0; i < sent - 1; i++) { + // Receive the packet and make sure it's identical. + std::vector recv_buf(buf.size() + sizeof(struct iphdr)); + ASSERT_NO_FATAL_FAILURE(ReceiveBuf(recv_buf.data(), recv_buf.size())); + EXPECT_EQ(memcmp(recv_buf.data() + sizeof(struct iphdr), buf.data(), + buf.size()), + 0); + } + + // Assert that the last packet is dropped because the receive buffer should + // be full after the first four packets. + std::vector recv_buf(buf.size() + sizeof(struct iphdr)); + struct iovec iov = {}; + iov.iov_base = static_cast(const_cast(recv_buf.data())); + iov.iov_len = buf.size(); + struct msghdr msg = {}; + msg.msg_iov = &iov; + msg.msg_iovlen = 1; + msg.msg_control = NULL; + msg.msg_controllen = 0; + msg.msg_flags = 0; + ASSERT_THAT(RetryEINTR(recvmsg)(s_, &msg, MSG_DONTWAIT), + SyscallFailsWithErrno(EAGAIN)); + } +} + void RawSocketTest::SendBufTo(int sock, const struct sockaddr_in& addr, const char* buf, int buf_len) { // It's safe to use const_cast here because sendmsg won't modify the iovec or diff --git a/test/syscalls/linux/socket_ipv4_udp_unbound.cc b/test/syscalls/linux/socket_ipv4_udp_unbound.cc index 4db9553e1..de0f5f01b 100644 --- a/test/syscalls/linux/socket_ipv4_udp_unbound.cc +++ b/test/syscalls/linux/socket_ipv4_udp_unbound.cc @@ -18,6 +18,7 @@ #include #include #include +#include #include #include @@ -2236,5 +2237,220 @@ TEST_P(IPv4UDPUnboundSocketTest, SetAndReceiveIPPKTINFO) { EXPECT_EQ(received_pktinfo.ipi_spec_dst.s_addr, htonl(INADDR_LOOPBACK)); EXPECT_EQ(received_pktinfo.ipi_addr.s_addr, htonl(INADDR_LOOPBACK)); } + +// Check that setting SO_RCVBUF below min is clamped to the minimum +// receive buffer size. +TEST_P(IPv4UDPUnboundSocketTest, SetSocketRecvBufBelowMin) { + auto s = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + + // Discover minimum buffer size by setting it to zero. + constexpr int kRcvBufSz = 0; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, + sizeof(kRcvBufSz)), + SyscallSucceeds()); + + int min = 0; + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &min, &min_len), + SyscallSucceeds()); + + // Linux doubles the value so let's use a value that when doubled will still + // be smaller than min. + int below_min = min / 2 - 1; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &below_min, + sizeof(below_min)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &val, &val_len), + SyscallSucceeds()); + + ASSERT_EQ(min, val); +} + +// Check that setting SO_RCVBUF above max is clamped to the maximum +// receive buffer size. +TEST_P(IPv4UDPUnboundSocketTest, SetSocketRecvBufAboveMax) { + auto s = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + + // Discover maxmimum buffer size by setting to a really large value. + constexpr int kRcvBufSz = 0xffffffff; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, + sizeof(kRcvBufSz)), + SyscallSucceeds()); + + int max = 0; + socklen_t max_len = sizeof(max); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &max, &max_len), + SyscallSucceeds()); + + int above_max = max + 1; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &above_max, + sizeof(above_max)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &val, &val_len), + SyscallSucceeds()); + ASSERT_EQ(max, val); +} + +// Check that setting SO_RCVBUF min <= rcvBufSz <= max is honored. +TEST_P(IPv4UDPUnboundSocketTest, SetSocketRecvBuf) { + auto s = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + + int max = 0; + int min = 0; + { + // Discover maxmimum buffer size by setting to a really large value. + constexpr int kRcvBufSz = 0xffffffff; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, + sizeof(kRcvBufSz)), + SyscallSucceeds()); + + max = 0; + socklen_t max_len = sizeof(max); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &max, &max_len), + SyscallSucceeds()); + } + + { + // Discover minimum buffer size by setting it to zero. + constexpr int kRcvBufSz = 0; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, + sizeof(kRcvBufSz)), + SyscallSucceeds()); + + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &min, &min_len), + SyscallSucceeds()); + } + + int quarter_sz = min + (max - min) / 4; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &quarter_sz, + sizeof(quarter_sz)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &val, &val_len), + SyscallSucceeds()); + + // Linux doubles the value set by SO_SNDBUF/SO_RCVBUF. + if (!IsRunningOnGvisor()) { + quarter_sz *= 2; + } + ASSERT_EQ(quarter_sz, val); +} + +// Check that setting SO_SNDBUF below min is clamped to the minimum +// send buffer size. +TEST_P(IPv4UDPUnboundSocketTest, SetSocketSendBufBelowMin) { + auto s = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + + // Discover minimum buffer size by setting it to zero. + constexpr int kSndBufSz = 0; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &kSndBufSz, + sizeof(kSndBufSz)), + SyscallSucceeds()); + + int min = 0; + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &min, &min_len), + SyscallSucceeds()); + + // Linux doubles the value so let's use a value that when doubled will still + // be smaller than min. + int below_min = min / 2 - 1; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &below_min, + sizeof(below_min)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &val, &val_len), + SyscallSucceeds()); + + ASSERT_EQ(min, val); +} + +// Check that setting SO_SNDBUF above max is clamped to the maximum +// send buffer size. +TEST_P(IPv4UDPUnboundSocketTest, SetSocketSendBufAboveMax) { + auto s = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + + // Discover maxmimum buffer size by setting to a really large value. + constexpr int kSndBufSz = 0xffffffff; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &kSndBufSz, + sizeof(kSndBufSz)), + SyscallSucceeds()); + + int max = 0; + socklen_t max_len = sizeof(max); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &max, &max_len), + SyscallSucceeds()); + + int above_max = max + 1; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &above_max, + sizeof(above_max)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &val, &val_len), + SyscallSucceeds()); + ASSERT_EQ(max, val); +} + +// Check that setting SO_SNDBUF min <= kSndBufSz <= max is honored. +TEST_P(IPv4UDPUnboundSocketTest, SetSocketSendBuf) { + auto s = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + + int max = 0; + int min = 0; + { + // Discover maxmimum buffer size by setting to a really large value. + constexpr int kSndBufSz = 0xffffffff; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &kSndBufSz, + sizeof(kSndBufSz)), + SyscallSucceeds()); + + max = 0; + socklen_t max_len = sizeof(max); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &max, &max_len), + SyscallSucceeds()); + } + + { + // Discover minimum buffer size by setting it to zero. + constexpr int kSndBufSz = 0; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &kSndBufSz, + sizeof(kSndBufSz)), + SyscallSucceeds()); + + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &min, &min_len), + SyscallSucceeds()); + } + + int quarter_sz = min + (max - min) / 4; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &quarter_sz, + sizeof(quarter_sz)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &val, &val_len), + SyscallSucceeds()); + + // Linux doubles the value set by SO_SNDBUF/SO_RCVBUF. + if (!IsRunningOnGvisor()) { + quarter_sz *= 2; + } + + ASSERT_EQ(quarter_sz, val); +} } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/udp_socket_test_cases.cc b/test/syscalls/linux/udp_socket_test_cases.cc index 42521efef..cc1db3de8 100644 --- a/test/syscalls/linux/udp_socket_test_cases.cc +++ b/test/syscalls/linux/udp_socket_test_cases.cc @@ -1543,5 +1543,120 @@ TEST_P(UdpSocketTest, SendAndReceiveTOS) { memcpy(&received_tos, CMSG_DATA(cmsg), sizeof(received_tos)); EXPECT_EQ(received_tos, sent_tos); } + +TEST_P(UdpSocketTest, RecvBufLimitsEmptyRcvBuf) { + // Discover minimum buffer size by setting it to zero. + constexpr int kRcvBufSz = 0; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, sizeof(kRcvBufSz)), + SyscallSucceeds()); + + int min = 0; + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &min, &min_len), + SyscallSucceeds()); + + // Bind s_ to loopback. + ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); + + { + // Send data of size min and verify that it's received. + std::vector buf(min); + RandomizeBuffer(buf.data(), buf.size()); + ASSERT_THAT(sendto(t_, buf.data(), buf.size(), 0, addr_[0], addrlen_), + SyscallSucceedsWithValue(buf.size())); + std::vector received(buf.size()); + EXPECT_THAT(recv(s_, received.data(), received.size(), MSG_DONTWAIT), + SyscallSucceedsWithValue(received.size())); + } + + { + // Send data of size min + 1 and verify that its received. Both linux and + // Netstack accept a dgram that exceeds rcvBuf limits if the receive buffer + // is currently empty. + std::vector buf(min + 1); + RandomizeBuffer(buf.data(), buf.size()); + ASSERT_THAT(sendto(t_, buf.data(), buf.size(), 0, addr_[0], addrlen_), + SyscallSucceedsWithValue(buf.size())); + + std::vector received(buf.size()); + EXPECT_THAT(recv(s_, received.data(), received.size(), MSG_DONTWAIT), + SyscallSucceedsWithValue(received.size())); + } +} + +// Test that receive buffer limits are enforced. +TEST_P(UdpSocketTest, RecvBufLimits) { + // Bind s_ to loopback. + ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); + + int min = 0; + { + // Discover minimum buffer size by trying to set it to zero. + constexpr int kRcvBufSz = 0; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, sizeof(kRcvBufSz)), + SyscallSucceeds()); + + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &min, &min_len), + SyscallSucceeds()); + } + + // Now set the limit to min * 4. + int new_rcv_buf_sz = min * 4; + if (!IsRunningOnGvisor() || IsRunningWithHostinet()) { + // Linux doubles the value specified so just set to min * 2. + new_rcv_buf_sz = min * 2; + } + + ASSERT_THAT(setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &new_rcv_buf_sz, + sizeof(new_rcv_buf_sz)), + SyscallSucceeds()); + int rcv_buf_sz = 0; + { + socklen_t rcv_buf_len = sizeof(rcv_buf_sz); + ASSERT_THAT( + getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &rcv_buf_sz, &rcv_buf_len), + SyscallSucceeds()); + } + + { + std::vector buf(min); + RandomizeBuffer(buf.data(), buf.size()); + + ASSERT_THAT(sendto(t_, buf.data(), buf.size(), 0, addr_[0], addrlen_), + SyscallSucceedsWithValue(buf.size())); + ASSERT_THAT(sendto(t_, buf.data(), buf.size(), 0, addr_[0], addrlen_), + SyscallSucceedsWithValue(buf.size())); + ASSERT_THAT(sendto(t_, buf.data(), buf.size(), 0, addr_[0], addrlen_), + SyscallSucceedsWithValue(buf.size())); + ASSERT_THAT(sendto(t_, buf.data(), buf.size(), 0, addr_[0], addrlen_), + SyscallSucceedsWithValue(buf.size())); + int sent = 4; + if (IsRunningOnGvisor() && !IsRunningWithHostinet()) { + // Linux seems to drop the 4th packet even though technically it should + // fit in the receive buffer. + ASSERT_THAT(sendto(t_, buf.data(), buf.size(), 0, addr_[0], addrlen_), + SyscallSucceedsWithValue(buf.size())); + sent++; + } + + for (int i = 0; i < sent - 1; i++) { + // Receive the data. + std::vector received(buf.size()); + EXPECT_THAT(recv(s_, received.data(), received.size(), MSG_DONTWAIT), + SyscallSucceedsWithValue(received.size())); + EXPECT_EQ(memcmp(buf.data(), received.data(), buf.size()), 0); + } + + // The last receive should fail with EAGAIN as the last packet should have + // been dropped due to lack of space in the receive buffer. + std::vector received(buf.size()); + EXPECT_THAT(recv(s_, received.data(), received.size(), MSG_DONTWAIT), + SyscallFailsWithErrno(EAGAIN)); + } +} + } // namespace testing } // namespace gvisor -- cgit v1.2.3 From 3970c127434817304f67a2ad192cbe8094ad3353 Mon Sep 17 00:00:00 2001 From: Michael Pratt Date: Thu, 18 Jun 2020 09:02:14 -0700 Subject: Remove various uses of 'whitelist' Updates #2972 PiperOrigin-RevId: 317113059 --- pkg/bpf/interpreter_test.go | 2 +- pkg/cpuid/cpuid_arm64.go | 5 +++-- pkg/cpuid/cpuid_x86.go | 7 +++---- pkg/seccomp/seccomp_rules.go | 4 ++-- pkg/sentry/fs/filesystems.go | 14 -------------- pkg/sentry/fsimpl/host/host.go | 5 +++-- pkg/sentry/socket/hostinet/socket.go | 8 ++++---- pkg/sentry/vfs/README.md | 2 -- test/syscalls/linux/xattr.cc | 5 +++-- tools/nogo/matchers.go | 27 ++++++++++++++++----------- 10 files changed, 35 insertions(+), 44 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/bpf/interpreter_test.go b/pkg/bpf/interpreter_test.go index 547921d0a..c85d786b9 100644 --- a/pkg/bpf/interpreter_test.go +++ b/pkg/bpf/interpreter_test.go @@ -767,7 +767,7 @@ func TestSimpleFilter(t *testing.T) { expectedRet: 0, }, { - desc: "Whitelisted syscall is allowed", + desc: "Allowed syscall is indeed allowed", seccompData: seccompData{nr: 231 /* __NR_exit_group */, arch: 0xc000003e}, expectedRet: 0x7fff0000, }, diff --git a/pkg/cpuid/cpuid_arm64.go b/pkg/cpuid/cpuid_arm64.go index 08381c1c0..ac7bb6774 100644 --- a/pkg/cpuid/cpuid_arm64.go +++ b/pkg/cpuid/cpuid_arm64.go @@ -312,8 +312,9 @@ func HostFeatureSet() *FeatureSet { } } -// Reads bogomips from host /proc/cpuinfo. Must run before whitelisting. -// This value is used to create the fake /proc/cpuinfo from a FeatureSet. +// Reads bogomips from host /proc/cpuinfo. Must run before syscall filter +// installation. This value is used to create the fake /proc/cpuinfo from a +// FeatureSet. func initCPUInfo() { cpuinfob, err := ioutil.ReadFile("/proc/cpuinfo") if err != nil { diff --git a/pkg/cpuid/cpuid_x86.go b/pkg/cpuid/cpuid_x86.go index 562f8f405..17a89c00d 100644 --- a/pkg/cpuid/cpuid_x86.go +++ b/pkg/cpuid/cpuid_x86.go @@ -1057,9 +1057,9 @@ func HostFeatureSet() *FeatureSet { } } -// Reads max cpu frequency from host /proc/cpuinfo. Must run before -// whitelisting. This value is used to create the fake /proc/cpuinfo from a -// FeatureSet. +// Reads max cpu frequency from host /proc/cpuinfo. Must run before syscall +// filter installation. This value is used to create the fake /proc/cpuinfo +// from a FeatureSet. func initCPUFreq() { cpuinfob, err := ioutil.ReadFile("/proc/cpuinfo") if err != nil { @@ -1106,7 +1106,6 @@ func initFeaturesFromString() { } func init() { - // initCpuFreq must be run before whitelists are enabled. initCPUFreq() initFeaturesFromString() } diff --git a/pkg/seccomp/seccomp_rules.go b/pkg/seccomp/seccomp_rules.go index 06308cd29..a52dc1b4e 100644 --- a/pkg/seccomp/seccomp_rules.go +++ b/pkg/seccomp/seccomp_rules.go @@ -56,7 +56,7 @@ func (a AllowValue) String() (s string) { return fmt.Sprintf("%#x ", uintptr(a)) } -// Rule stores the whitelist of syscall arguments. +// Rule stores the allowed syscall arguments. // // For example: // rule := Rule { @@ -82,7 +82,7 @@ func (r Rule) String() (s string) { return } -// SyscallRules stores a map of OR'ed whitelist rules indexed by the syscall number. +// SyscallRules stores a map of OR'ed argument rules indexed by the syscall number. // If the 'Rules' is empty, we treat it as any argument is allowed. // // For example: diff --git a/pkg/sentry/fs/filesystems.go b/pkg/sentry/fs/filesystems.go index 084da2a8d..d41f30bbb 100644 --- a/pkg/sentry/fs/filesystems.go +++ b/pkg/sentry/fs/filesystems.go @@ -87,20 +87,6 @@ func RegisterFilesystem(f Filesystem) { filesystems.registered[f.Name()] = f } -// UnregisterFilesystem removes a file system from the global set. To keep the -// file system set compatible with save/restore, UnregisterFilesystem must be -// called before save/restore methods. -// -// For instance, packages may unregister their file system after it is mounted. -// This makes sense for pseudo file systems that should not be visible or -// mountable. See whitelistfs in fs/host/fs.go for one example. -func UnregisterFilesystem(name string) { - filesystems.mu.Lock() - defer filesystems.mu.Unlock() - - delete(filesystems.registered, name) -} - // FindFilesystem returns a Filesystem registered at name or (nil, false) if name // is not a file system type that can be found in /proc/filesystems. func FindFilesystem(name string) (Filesystem, bool) { diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index a3a312edb..43a173bc9 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -476,8 +476,9 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount, flags u return unixsocket.NewFileDescription(ep, ep.Type(), flags, mnt, d, &i.locks) } - // TODO(gvisor.dev/issue/1672): Whitelist specific file types here, so that - // we don't allow importing arbitrary file types without proper support. + // TODO(gvisor.dev/issue/1672): Allow only specific file types here, so + // that we don't allow importing arbitrary file types without proper + // support. if i.isTTY { fd := &TTYFileDescription{ fileDescription: fileDescription{inode: i}, diff --git a/pkg/sentry/socket/hostinet/socket.go b/pkg/sentry/socket/hostinet/socket.go index c11e82c10..a92aed2c9 100644 --- a/pkg/sentry/socket/hostinet/socket.go +++ b/pkg/sentry/socket/hostinet/socket.go @@ -324,7 +324,7 @@ func (s *socketOpsCommon) GetSockOpt(t *kernel.Task, level int, name int, outPtr return nil, syserr.ErrInvalidArgument } - // Whitelist options and constrain option length. + // Only allow known and safe options. optlen := getSockOptLen(t, level, name) switch level { case linux.SOL_IP: @@ -369,7 +369,7 @@ func (s *socketOpsCommon) GetSockOpt(t *kernel.Task, level int, name int, outPtr // SetSockOpt implements socket.Socket.SetSockOpt. func (s *socketOpsCommon) SetSockOpt(t *kernel.Task, level int, name int, opt []byte) *syserr.Error { - // Whitelist options and constrain option length. + // Only allow known and safe options. optlen := setSockOptLen(t, level, name) switch level { case linux.SOL_IP: @@ -415,7 +415,7 @@ func (s *socketOpsCommon) SetSockOpt(t *kernel.Task, level int, name int, opt [] // RecvMsg implements socket.Socket.RecvMsg. func (s *socketOpsCommon) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, haveDeadline bool, deadline ktime.Time, senderRequested bool, controlLen uint64) (int, int, linux.SockAddr, uint32, socket.ControlMessages, *syserr.Error) { - // Whitelist flags. + // Only allow known and safe flags. // // FIXME(jamieliu): We can't support MSG_ERRQUEUE because it uses ancillary // messages that gvisor/pkg/tcpip/transport/unix doesn't understand. Kill the @@ -537,7 +537,7 @@ func (s *socketOpsCommon) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags // SendMsg implements socket.Socket.SendMsg. func (s *socketOpsCommon) SendMsg(t *kernel.Task, src usermem.IOSequence, to []byte, flags int, haveDeadline bool, deadline ktime.Time, controlMessages socket.ControlMessages) (int, *syserr.Error) { - // Whitelist flags. + // Only allow known and safe flags. if flags&^(syscall.MSG_DONTWAIT|syscall.MSG_EOR|syscall.MSG_FASTOPEN|syscall.MSG_MORE|syscall.MSG_NOSIGNAL) != 0 { return 0, syserr.ErrInvalidArgument } diff --git a/pkg/sentry/vfs/README.md b/pkg/sentry/vfs/README.md index 66f3105bd..4b9faf2ea 100644 --- a/pkg/sentry/vfs/README.md +++ b/pkg/sentry/vfs/README.md @@ -169,8 +169,6 @@ This construction, which is essentially a type-safe analogue to Linux's - binder, which is similarly far too incomplete to use. - - whitelistfs, which we are already actively attempting to remove. - - Save/restore. For instance, it is unclear if the current implementation of the `state` package supports the inheritance pattern described above. diff --git a/test/syscalls/linux/xattr.cc b/test/syscalls/linux/xattr.cc index 3231732ec..cbcf08451 100644 --- a/test/syscalls/linux/xattr.cc +++ b/test/syscalls/linux/xattr.cc @@ -73,9 +73,10 @@ TEST_F(XattrTest, XattrLargeName) { std::string name = "user."; name += std::string(XATTR_NAME_MAX - name.length(), 'a'); - // An xattr should be whitelisted before it can be accessed--do not allow - // arbitrary xattrs to be read/written in gVisor. if (!IsRunningOnGvisor()) { + // In gVisor, access to xattrs is controlled with an explicit list of + // allowed names. This name isn't going to be configured to allow access, so + // don't test it. EXPECT_THAT(setxattr(path, name.c_str(), nullptr, 0, /*flags=*/0), SyscallSucceeds()); EXPECT_THAT(getxattr(path, name.c_str(), nullptr, 0), diff --git a/tools/nogo/matchers.go b/tools/nogo/matchers.go index 32f099925..57a250501 100644 --- a/tools/nogo/matchers.go +++ b/tools/nogo/matchers.go @@ -27,10 +27,15 @@ type matcher interface { ShouldReport(d analysis.Diagnostic, fs *token.FileSet) bool } -// pathRegexps excludes explicit paths. +// pathRegexps filters explicit paths. type pathRegexps struct { - expr []*regexp.Regexp - whitelist bool + expr []*regexp.Regexp + + // include, if true, indicates that paths matching any regexp in expr + // match. + // + // If false, paths matching no regexps in expr match. + include bool } // buildRegexps builds a list of regular expressions. @@ -49,33 +54,33 @@ func (p *pathRegexps) ShouldReport(d analysis.Diagnostic, fs *token.FileSet) boo fullPos := fs.Position(d.Pos).String() for _, path := range p.expr { if path.MatchString(fullPos) { - return p.whitelist + return p.include } } - return !p.whitelist + return !p.include } // internalExcluded excludes specific internal paths. func internalExcluded(paths ...string) *pathRegexps { return &pathRegexps{ - expr: buildRegexps(internalPrefix, paths...), - whitelist: false, + expr: buildRegexps(internalPrefix, paths...), + include: false, } } // excludedExcluded excludes specific external paths. func externalExcluded(paths ...string) *pathRegexps { return &pathRegexps{ - expr: buildRegexps(externalPrefix, paths...), - whitelist: false, + expr: buildRegexps(externalPrefix, paths...), + include: false, } } // internalMatches returns a path matcher for internal packages. func internalMatches() *pathRegexps { return &pathRegexps{ - expr: buildRegexps(internalPrefix, ".*"), - whitelist: true, + expr: buildRegexps(internalPrefix, ".*"), + include: true, } } -- cgit v1.2.3 From f97122f44c5c4218bb2239686cd323358c661d60 Mon Sep 17 00:00:00 2001 From: Ting-Yu Wang Date: Thu, 18 Jun 2020 14:28:06 -0700 Subject: Acquire lock when accessing MultiDevice's cache in String(). PiperOrigin-RevId: 317180925 --- pkg/sentry/device/device.go | 3 +++ 1 file changed, 3 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/device/device.go b/pkg/sentry/device/device.go index 69e71e322..f45b2bd2b 100644 --- a/pkg/sentry/device/device.go +++ b/pkg/sentry/device/device.go @@ -188,6 +188,9 @@ type MultiDevice struct { // String stringifies MultiDevice. func (m *MultiDevice) String() string { + m.mu.Lock() + defer m.mu.Unlock() + buf := bytes.NewBuffer(nil) buf.WriteString("cache{") for k, v := range m.cache { -- cgit v1.2.3 From 28b8a5cc3ac538333756084da28d7f13f13b5c87 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Thu, 18 Jun 2020 17:00:47 -0700 Subject: iptables: remove metadata struct Metadata was useful for debugging and safety, but enough tests exist that we should see failures when (de)serialization is broken. It made stack initialization more cumbersome and it's also getting in the way of ip6tables. PiperOrigin-RevId: 317210653 --- pkg/sentry/socket/netfilter/netfilter.go | 103 ++++++------------------------- pkg/sentry/socket/netstack/stack.go | 6 -- pkg/tcpip/stack/iptables.go | 8 --- pkg/tcpip/stack/iptables_types.go | 16 +---- runsc/boot/loader.go | 2 - 5 files changed, 21 insertions(+), 114 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 66015e2bc..f7abe77d3 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -41,19 +41,6 @@ const errorTargetName = "ERROR" // change the destination port/destination IP for packets. const redirectTargetName = "REDIRECT" -// Metadata is used to verify that we are correctly serializing and -// deserializing iptables into structs consumable by the iptables tool. We save -// a metadata struct when the tables are written, and when they are read out we -// verify that certain fields are the same. -// -// metadata is used by this serialization/deserializing code, not netstack. -type metadata struct { - HookEntry [linux.NF_INET_NUMHOOKS]uint32 - Underflow [linux.NF_INET_NUMHOOKS]uint32 - NumEntries uint32 - Size uint32 -} - // enableLogging controls whether to log the (de)serialization of netfilter // structs between userspace and netstack. These logs are useful when // developing iptables, but can pollute sentry logs otherwise. @@ -83,29 +70,13 @@ func GetInfo(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr) (linux.IPT return linux.IPTGetinfo{}, syserr.FromError(err) } - // Find the appropriate table. - table, err := findTable(stack, info.Name) + _, info, err := convertNetstackToBinary(stack, info.Name) if err != nil { - nflog("%v", err) + nflog("couldn't convert iptables: %v", err) return linux.IPTGetinfo{}, syserr.ErrInvalidArgument } - // Get the hooks that apply to this table. - info.ValidHooks = table.ValidHooks() - - // Grab the metadata struct, which is used to store information (e.g. - // the number of entries) that applies to the user's encoding of - // iptables, but not netstack's. - metadata := table.Metadata().(metadata) - - // Set values from metadata. - info.HookEntry = metadata.HookEntry - info.Underflow = metadata.Underflow - info.NumEntries = metadata.NumEntries - info.Size = metadata.Size - nflog("returning info: %+v", info) - return info, nil } @@ -118,23 +89,13 @@ func GetEntries(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr, outLen return linux.KernelIPTGetEntries{}, syserr.FromError(err) } - // Find the appropriate table. - table, err := findTable(stack, userEntries.Name) - if err != nil { - nflog("%v", err) - return linux.KernelIPTGetEntries{}, syserr.ErrInvalidArgument - } - // Convert netstack's iptables rules to something that the iptables // tool can understand. - entries, meta, err := convertNetstackToBinary(userEntries.Name.String(), table) + entries, _, err := convertNetstackToBinary(stack, userEntries.Name) if err != nil { nflog("couldn't read entries: %v", err) return linux.KernelIPTGetEntries{}, syserr.ErrInvalidArgument } - if meta != table.Metadata().(metadata) { - panic(fmt.Sprintf("Table %q metadata changed between writing and reading. Was saved as %+v, but is now %+v", userEntries.Name.String(), table.Metadata().(metadata), meta)) - } if binary.Size(entries) > uintptr(outLen) { nflog("insufficient GetEntries output size: %d", uintptr(outLen)) return linux.KernelIPTGetEntries{}, syserr.ErrInvalidArgument @@ -143,44 +104,26 @@ func GetEntries(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr, outLen return entries, nil } -func findTable(stk *stack.Stack, tablename linux.TableName) (stack.Table, error) { - table, ok := stk.IPTables().GetTable(tablename.String()) - if !ok { - return stack.Table{}, fmt.Errorf("couldn't find table %q", tablename) - } - return table, nil -} - -// FillIPTablesMetadata populates stack's IPTables with metadata. -func FillIPTablesMetadata(stk *stack.Stack) { - stk.IPTables().ModifyTables(func(tables map[string]stack.Table) { - // In order to fill in the metadata, we have to translate ipt from its - // netstack format to Linux's giant-binary-blob format. - for name, table := range tables { - _, metadata, err := convertNetstackToBinary(name, table) - if err != nil { - panic(fmt.Errorf("Unable to set default IP tables: %v", err)) - } - table.SetMetadata(metadata) - tables[name] = table - } - }) -} - // convertNetstackToBinary converts the iptables as stored in netstack to the // format expected by the iptables tool. Linux stores each table as a binary // blob that can only be traversed by parsing a bit, reading some offsets, // jumping to those offsets, parsing again, etc. -func convertNetstackToBinary(tablename string, table stack.Table) (linux.KernelIPTGetEntries, metadata, error) { - // Return values. +func convertNetstackToBinary(stack *stack.Stack, tablename linux.TableName) (linux.KernelIPTGetEntries, linux.IPTGetinfo, error) { + table, ok := stack.IPTables().GetTable(tablename.String()) + if !ok { + return linux.KernelIPTGetEntries{}, linux.IPTGetinfo{}, fmt.Errorf("couldn't find table %q", tablename) + } + var entries linux.KernelIPTGetEntries - var meta metadata + var info linux.IPTGetinfo + info.ValidHooks = table.ValidHooks() // The table name has to fit in the struct. if linux.XT_TABLE_MAXNAMELEN < len(tablename) { - return linux.KernelIPTGetEntries{}, metadata{}, fmt.Errorf("table name %q too long.", tablename) + return linux.KernelIPTGetEntries{}, linux.IPTGetinfo{}, fmt.Errorf("table name %q too long", tablename) } - copy(entries.Name[:], tablename) + copy(info.Name[:], tablename[:]) + copy(entries.Name[:], tablename[:]) for ruleIdx, rule := range table.Rules { nflog("convert to binary: current offset: %d", entries.Size) @@ -189,14 +132,14 @@ func convertNetstackToBinary(tablename string, table stack.Table) (linux.KernelI for hook, hookRuleIdx := range table.BuiltinChains { if hookRuleIdx == ruleIdx { nflog("convert to binary: found hook %d at offset %d", hook, entries.Size) - meta.HookEntry[hook] = entries.Size + info.HookEntry[hook] = entries.Size } } // Is this a chain underflow point? for underflow, underflowRuleIdx := range table.Underflows { if underflowRuleIdx == ruleIdx { nflog("convert to binary: found underflow %d at offset %d", underflow, entries.Size) - meta.Underflow[underflow] = entries.Size + info.Underflow[underflow] = entries.Size } } @@ -251,12 +194,12 @@ func convertNetstackToBinary(tablename string, table stack.Table) (linux.KernelI entries.Size += uint32(entry.NextOffset) entries.Entrytable = append(entries.Entrytable, entry) - meta.NumEntries++ + info.NumEntries++ } - nflog("convert to binary: finished with an marshalled size of %d", meta.Size) - meta.Size = entries.Size - return entries, meta, nil + nflog("convert to binary: finished with an marshalled size of %d", info.Size) + info.Size = entries.Size + return entries, info, nil } func marshalTarget(target stack.Target) []byte { @@ -569,12 +512,6 @@ func SetEntries(stk *stack.Stack, optVal []byte) *syserr.Error { // - There are no chains without an unconditional final rule. // - There are no chains without an unconditional underflow rule. - table.SetMetadata(metadata{ - HookEntry: replace.HookEntry, - Underflow: replace.Underflow, - NumEntries: replace.NumEntries, - Size: replace.Size, - }) stk.IPTables().ReplaceTable(replace.Name.String(), table) return nil diff --git a/pkg/sentry/socket/netstack/stack.go b/pkg/sentry/socket/netstack/stack.go index f97f9b6f3..ee11742a6 100644 --- a/pkg/sentry/socket/netstack/stack.go +++ b/pkg/sentry/socket/netstack/stack.go @@ -18,7 +18,6 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/inet" - "gvisor.dev/gvisor/pkg/sentry/socket/netfilter" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip" @@ -366,11 +365,6 @@ func (s *Stack) IPTables() (*stack.IPTables, error) { return s.Stack.IPTables(), nil } -// FillIPTablesMetadata populates stack's IPTables with metadata. -func (s *Stack) FillIPTablesMetadata() { - netfilter.FillIPTablesMetadata(s.Stack) -} - // Resume implements inet.Stack.Resume. func (s *Stack) Resume() { s.Stack.Resume() diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go index 4e9b404c8..dc2b77c9d 100644 --- a/pkg/tcpip/stack/iptables.go +++ b/pkg/tcpip/stack/iptables.go @@ -173,14 +173,6 @@ func (it *IPTables) ReplaceTable(name string, table Table) { it.tables[name] = table } -// ModifyTables acquires write-lock and calls fn with internal name-to-table -// map. This function can be used to update multiple tables atomically. -func (it *IPTables) ModifyTables(fn func(map[string]Table)) { - it.mu.Lock() - defer it.mu.Unlock() - fn(it.tables) -} - // GetPriorities returns slice of priorities associated with hook. func (it *IPTables) GetPriorities(hook Hook) []string { it.mu.RLock() diff --git a/pkg/tcpip/stack/iptables_types.go b/pkg/tcpip/stack/iptables_types.go index 4a6a5c6f1..72f1dd329 100644 --- a/pkg/tcpip/stack/iptables_types.go +++ b/pkg/tcpip/stack/iptables_types.go @@ -95,7 +95,7 @@ type IPTables struct { } // A Table defines a set of chains and hooks into the network stack. It is -// really just a list of rules with some metadata for entrypoints and such. +// really just a list of rules. type Table struct { // Rules holds the rules that make up the table. Rules []Rule @@ -110,10 +110,6 @@ type Table struct { // UserChains holds user-defined chains for the keyed by name. Users // can give their chains arbitrary names. UserChains map[string]int - - // Metadata holds information about the Table that is useful to users - // of IPTables, but not to the netstack IPTables code itself. - metadata interface{} } // ValidHooks returns a bitmap of the builtin hooks for the given table. @@ -125,16 +121,6 @@ func (table *Table) ValidHooks() uint32 { return hooks } -// Metadata returns the metadata object stored in table. -func (table *Table) Metadata() interface{} { - return table.metadata -} - -// SetMetadata sets the metadata object stored in table. -func (table *Table) SetMetadata(metadata interface{}) { - table.metadata = metadata -} - // A Rule is a packet processing rule. It consists of two pieces. First it // contains zero or more matchers, each of which is a specification of which // packets this rule applies to. If there are no matchers in the rule, it diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index c6efcdc83..081db39c1 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -1071,8 +1071,6 @@ func newEmptySandboxNetworkStack(clock tcpip.Clock, uniqueID stack.UniqueID) (in return nil, fmt.Errorf("SetTransportProtocolOption failed: %s", err) } - s.FillIPTablesMetadata() - return &s, nil } -- cgit v1.2.3 From 70c45e09cfd15abeebd640a9a2ddd214078db651 Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Thu, 18 Jun 2020 20:26:29 -0700 Subject: socket/unix: (*connectionedEndpoint).State() has to take the endpoint lock It accesses e.receiver which is protected by the endpoint lock. WARNING: DATA RACE Write at 0x00c0006aa2b8 by goroutine 189: pkg/sentry/socket/unix/transport.(*connectionedEndpoint).Connect.func1() pkg/sentry/socket/unix/transport/connectioned.go:359 +0x50 pkg/sentry/socket/unix/transport.(*connectionedEndpoint).BidirectionalConnect() pkg/sentry/socket/unix/transport/connectioned.go:327 +0xa3c pkg/sentry/socket/unix/transport.(*connectionedEndpoint).Connect() pkg/sentry/socket/unix/transport/connectioned.go:363 +0xca pkg/sentry/socket/unix.(*socketOpsCommon).Connect() pkg/sentry/socket/unix/unix.go:420 +0x13a pkg/sentry/socket/unix.(*SocketOperations).Connect() :1 +0x78 pkg/sentry/syscalls/linux.Connect() pkg/sentry/syscalls/linux/sys_socket.go:286 +0x251 Previous read at 0x00c0006aa2b8 by goroutine 270: pkg/sentry/socket/unix/transport.(*baseEndpoint).Connected() pkg/sentry/socket/unix/transport/unix.go:789 +0x42 pkg/sentry/socket/unix/transport.(*connectionedEndpoint).State() pkg/sentry/socket/unix/transport/connectioned.go:479 +0x2f pkg/sentry/socket/unix.(*socketOpsCommon).State() pkg/sentry/socket/unix/unix.go:714 +0xc3e pkg/sentry/socket/unix.(*socketOpsCommon).SendMsg() pkg/sentry/socket/unix/unix.go:466 +0xc44 pkg/sentry/socket/unix.(*SocketOperations).SendMsg() :1 +0x173 pkg/sentry/syscalls/linux.sendTo() pkg/sentry/syscalls/linux/sys_socket.go:1121 +0x4c5 pkg/sentry/syscalls/linux.SendTo() pkg/sentry/syscalls/linux/sys_socket.go:1134 +0x87 Reported-by: syzbot+c2be37eedc672ed59a86@syzkaller.appspotmail.com PiperOrigin-RevId: 317236996 --- pkg/sentry/socket/unix/transport/connectioned.go | 3 +++ 1 file changed, 3 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/unix/transport/connectioned.go b/pkg/sentry/socket/unix/transport/connectioned.go index 09c6d3b27..a1e49cc57 100644 --- a/pkg/sentry/socket/unix/transport/connectioned.go +++ b/pkg/sentry/socket/unix/transport/connectioned.go @@ -476,6 +476,9 @@ func (e *connectionedEndpoint) Readiness(mask waiter.EventMask) waiter.EventMask // State implements socket.Socket.State. func (e *connectionedEndpoint) State() uint32 { + e.Lock() + defer e.Unlock() + if e.Connected() { return linux.SS_CONNECTED } -- cgit v1.2.3 From 408f3d2cd64cae6b2f76a940c76236e9841c095f Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Thu, 18 Jun 2020 22:00:56 -0700 Subject: Fix vfs2 tmpfs link permission checks. Updates #2923. PiperOrigin-RevId: 317246916 --- pkg/sentry/fsimpl/tmpfs/filesystem.go | 16 ++++++++++------ pkg/sentry/vfs/permissions.go | 31 +++++++++++++++++++++++++++++++ test/syscalls/BUILD | 1 + test/syscalls/linux/link.cc | 15 +++++++++++++-- test/util/test_util.h | 1 + 5 files changed, 56 insertions(+), 8 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 72399b321..ac359cf7b 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -237,18 +237,22 @@ func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. return syserror.EXDEV } d := vd.Dentry().Impl().(*dentry) - if d.inode.isDir() { + i := d.inode + if i.isDir() { return syserror.EPERM } - if d.inode.nlink == 0 { + if err := vfs.MayLink(auth.CredentialsFromContext(ctx), linux.FileMode(atomic.LoadUint32(&i.mode)), auth.KUID(atomic.LoadUint32(&i.uid)), auth.KGID(atomic.LoadUint32(&i.gid))); err != nil { + return err + } + if i.nlink == 0 { return syserror.ENOENT } - if d.inode.nlink == maxLinks { + if i.nlink == maxLinks { return syserror.EMLINK } - d.inode.incLinksLocked() - d.inode.watches.Notify("", linux.IN_ATTRIB, 0, vfs.InodeEvent) - parentDir.insertChildLocked(fs.newDentry(d.inode), name) + i.incLinksLocked() + i.watches.Notify("", linux.IN_ATTRIB, 0, vfs.InodeEvent) + parentDir.insertChildLocked(fs.newDentry(i), name) return nil }) } diff --git a/pkg/sentry/vfs/permissions.go b/pkg/sentry/vfs/permissions.go index f9647f90e..afe2be8d7 100644 --- a/pkg/sentry/vfs/permissions.go +++ b/pkg/sentry/vfs/permissions.go @@ -94,6 +94,37 @@ func GenericCheckPermissions(creds *auth.Credentials, ats AccessTypes, mode linu return syserror.EACCES } +// MayLink determines whether creating a hard link to a file with the given +// mode, kuid, and kgid is permitted. +// +// This corresponds to Linux's fs/namei.c:may_linkat. +func MayLink(creds *auth.Credentials, mode linux.FileMode, kuid auth.KUID, kgid auth.KGID) error { + // Source inode owner can hardlink all they like; otherwise, it must be a + // safe source. + if CanActAsOwner(creds, kuid) { + return nil + } + + // Only regular files can be hard linked. + if mode.FileType() != linux.S_IFREG { + return syserror.EPERM + } + + // Setuid files should not get pinned to the filesystem. + if mode&linux.S_ISUID != 0 { + return syserror.EPERM + } + + // Executable setgid files should not get pinned to the filesystem, but we + // don't support S_IXGRP anyway. + + // Hardlinking to unreadable or unwritable sources is dangerous. + if err := GenericCheckPermissions(creds, MayRead|MayWrite, mode, kuid, kgid); err != nil { + return syserror.EPERM + } + return nil +} + // AccessTypesForOpenFlags returns the access types required to open a file // with the given OpenOptions.Flags. Note that this is NOT the same thing as // the set of accesses permitted for the opened file: diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 1638a11c7..65a6a7f37 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -305,6 +305,7 @@ syscall_test( add_overlay = True, test = "//test/syscalls/linux:link_test", use_tmpfs = True, # gofer needs CAP_DAC_READ_SEARCH to use AT_EMPTY_PATH with linkat(2) + vfs2 = "True", ) syscall_test( diff --git a/test/syscalls/linux/link.cc b/test/syscalls/linux/link.cc index e74fa2ed5..544681168 100644 --- a/test/syscalls/linux/link.cc +++ b/test/syscalls/linux/link.cc @@ -79,8 +79,13 @@ TEST(LinkTest, PermissionDenied) { // Make the file "unsafe" to link by making it only readable, but not // writable. - const auto oldfile = + const auto unwriteable_file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileMode(0400)); + const std::string special_path = NewTempAbsPath(); + ASSERT_THAT(mkfifo(special_path.c_str(), 0666), SyscallSucceeds()); + const auto setuid_file = + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileMode(0666 | S_ISUID)); + const std::string newname = NewTempAbsPath(); // Do setuid in a separate thread so that after finishing this test, the @@ -97,8 +102,14 @@ TEST(LinkTest, PermissionDenied) { EXPECT_THAT(syscall(SYS_setuid, absl::GetFlag(FLAGS_scratch_uid)), SyscallSucceeds()); - EXPECT_THAT(link(oldfile.path().c_str(), newname.c_str()), + EXPECT_THAT(link(unwriteable_file.path().c_str(), newname.c_str()), + SyscallFailsWithErrno(EPERM)); + EXPECT_THAT(link(special_path.c_str(), newname.c_str()), SyscallFailsWithErrno(EPERM)); + if (!IsRunningWithVFS1()) { + EXPECT_THAT(link(setuid_file.path().c_str(), newname.c_str()), + SyscallFailsWithErrno(EPERM)); + } }); } diff --git a/test/util/test_util.h b/test/util/test_util.h index 8e3245b27..e635827e6 100644 --- a/test/util/test_util.h +++ b/test/util/test_util.h @@ -220,6 +220,7 @@ constexpr char kKVM[] = "kvm"; bool IsRunningOnGvisor(); const std::string GvisorPlatform(); bool IsRunningWithHostinet(); +// TODO(gvisor.dev/issue/1624): Delete once VFS1 is gone. bool IsRunningWithVFS1(); #ifdef __linux__ -- cgit v1.2.3 From 46957ed24f21396683cf9aff13fa0cd3086ea466 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 19 Jun 2020 05:55:35 -0700 Subject: Fix synthetic file bugs in gofer fs. Always check if a synthetic file already exists at a location before creating a file there, and do not try to delete synthetic gofer files from the remote fs. This fixes runsc_ptrace socket tests that create/unlink synthetic, named socket files. Updates #2923. PiperOrigin-RevId: 317293648 --- pkg/sentry/fsimpl/gofer/filesystem.go | 19 +++++++++++-------- test/syscalls/BUILD | 15 +++++++++++++++ 2 files changed, 26 insertions(+), 8 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 3c467e313..21eb976cb 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -374,13 +374,16 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir return nil } if fs.opts.interop == InteropModeShared { - // The existence of a dentry at name would be inconclusive because the - // file it represents may have been deleted from the remote filesystem, - // so we would need to make an RPC to revalidate the dentry. Just - // attempt the file creation RPC instead. If a file does exist, the RPC - // will fail with EEXIST like we would have. If the RPC succeeds, and a - // stale dentry exists, the dentry will fail revalidation next time - // it's used. + if child := parent.children[name]; child != nil && child.isSynthetic() { + return syserror.EEXIST + } + // The existence of a non-synthetic dentry at name would be inconclusive + // because the file it represents may have been deleted from the remote + // filesystem, so we would need to make an RPC to revalidate the dentry. + // Just attempt the file creation RPC instead. If a file does exist, the + // RPC will fail with EEXIST like we would have. If the RPC succeeds, and a + // stale dentry exists, the dentry will fail revalidation next time it's + // used. return createInRemoteDir(parent, name) } if child := parent.children[name]; child != nil { @@ -518,7 +521,7 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b if child == nil { return syserror.ENOENT } - } else { + } else if child == nil || !child.isSynthetic() { err = parent.file.unlinkAt(ctx, name, flags) if err != nil { if child != nil { diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 65a6a7f37..7b0248a39 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -57,6 +57,7 @@ syscall_test( size = "large", add_overlay = True, test = "//test/syscalls/linux:bind_test", + vfs2 = "True", ) syscall_test( @@ -700,6 +701,7 @@ syscall_test( size = "medium", add_overlay = True, test = "//test/syscalls/linux:socket_filesystem_non_blocking_test", + vfs2 = "True", ) syscall_test( @@ -707,12 +709,14 @@ syscall_test( add_overlay = True, shard_count = 50, test = "//test/syscalls/linux:socket_filesystem_test", + vfs2 = "True", ) syscall_test( size = "large", shard_count = 50, test = "//test/syscalls/linux:socket_inet_loopback_test", + vfs2 = "True", ) syscall_test( @@ -721,6 +725,7 @@ syscall_test( # Takes too long for TSAN. Creates a lot of TCP sockets. tags = ["nogotsan"], test = "//test/syscalls/linux:socket_inet_loopback_nogotsan_test", + vfs2 = "True", ) syscall_test( @@ -796,6 +801,7 @@ syscall_test( syscall_test( test = "//test/syscalls/linux:socket_blocking_local_test", + vfs2 = "True", ) syscall_test( @@ -805,6 +811,7 @@ syscall_test( syscall_test( test = "//test/syscalls/linux:socket_non_stream_blocking_local_test", + vfs2 = "True", ) syscall_test( @@ -815,6 +822,7 @@ syscall_test( syscall_test( size = "large", test = "//test/syscalls/linux:socket_stream_blocking_local_test", + vfs2 = "True", ) syscall_test( @@ -826,11 +834,13 @@ syscall_test( syscall_test( size = "medium", test = "//test/syscalls/linux:socket_stream_local_test", + vfs2 = "True", ) syscall_test( size = "medium", test = "//test/syscalls/linux:socket_stream_nonblock_local_test", + vfs2 = "True", ) syscall_test( @@ -838,6 +848,7 @@ syscall_test( size = "enormous", shard_count = 5, test = "//test/syscalls/linux:socket_unix_dgram_local_test", + vfs2 = "True", ) syscall_test( @@ -859,11 +870,13 @@ syscall_test( size = "enormous", shard_count = 5, test = "//test/syscalls/linux:socket_unix_seqpacket_local_test", + vfs2 = "True", ) syscall_test( size = "medium", test = "//test/syscalls/linux:socket_unix_stream_test", + vfs2 = "True", ) syscall_test( @@ -881,6 +894,7 @@ syscall_test( syscall_test( size = "medium", test = "//test/syscalls/linux:socket_unix_unbound_filesystem_test", + vfs2 = "True", ) syscall_test( @@ -894,6 +908,7 @@ syscall_test( size = "large", shard_count = 50, test = "//test/syscalls/linux:socket_unix_unbound_stream_test", + vfs2 = "True", ) syscall_test( -- cgit v1.2.3 From 6b69b955d7613ff391984661a7269eabc86020e3 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 19 Jun 2020 06:37:40 -0700 Subject: Support all seek options in gofer specialFileFD.Seek. Updates #2923. PiperOrigin-RevId: 317298186 --- pkg/sentry/fsimpl/gofer/regular_file.go | 16 ++++++++++++---- pkg/sentry/fsimpl/gofer/special_file.go | 19 +++++-------------- 2 files changed, 17 insertions(+), 18 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go index 0d10cf7ac..f7a15a2ed 100644 --- a/pkg/sentry/fsimpl/gofer/regular_file.go +++ b/pkg/sentry/fsimpl/gofer/regular_file.go @@ -489,15 +489,24 @@ func (d *dentry) writeback(ctx context.Context, offset, size int64) error { func (fd *regularFileFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { fd.mu.Lock() defer fd.mu.Unlock() + newOffset, err := regularFileSeekLocked(ctx, fd.dentry(), fd.off, offset, whence) + if err != nil { + return 0, err + } + fd.off = newOffset + return newOffset, nil +} + +// Calculate the new offset for a seek operation on a regular file. +func regularFileSeekLocked(ctx context.Context, d *dentry, fdOffset, offset int64, whence int32) (int64, error) { switch whence { case linux.SEEK_SET: // Use offset as specified. case linux.SEEK_CUR: - offset += fd.off + offset += fdOffset case linux.SEEK_END, linux.SEEK_DATA, linux.SEEK_HOLE: // Ensure file size is up to date. - d := fd.dentry() - if fd.filesystem().opts.interop == InteropModeShared { + if !d.cachedMetadataAuthoritative() { if err := d.updateFromGetattr(ctx); err != nil { return 0, err } @@ -525,7 +534,6 @@ func (fd *regularFileFD) Seek(ctx context.Context, offset int64, whence int32) ( if offset < 0 { return 0, syserror.EINVAL } - fd.off = offset return offset, nil } diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go index e6e29b329..a92008208 100644 --- a/pkg/sentry/fsimpl/gofer/special_file.go +++ b/pkg/sentry/fsimpl/gofer/special_file.go @@ -221,21 +221,12 @@ func (fd *specialFileFD) Seek(ctx context.Context, offset int64, whence int32) ( } fd.mu.Lock() defer fd.mu.Unlock() - switch whence { - case linux.SEEK_SET: - // Use offset as given. - case linux.SEEK_CUR: - offset += fd.off - default: - // SEEK_END, SEEK_DATA, and SEEK_HOLE aren't supported since it's not - // clear that file size is even meaningful for these files. - return 0, syserror.EINVAL - } - if offset < 0 { - return 0, syserror.EINVAL + newOffset, err := regularFileSeekLocked(ctx, fd.dentry(), fd.off, offset, whence) + if err != nil { + return 0, err } - fd.off = offset - return offset, nil + fd.off = newOffset + return newOffset, nil } // Sync implements vfs.FileDescriptionImpl.Sync. -- cgit v1.2.3 From f40d023ad6f8c19898ca105842a88961b3c2994c Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 19 Jun 2020 08:44:26 -0700 Subject: Don't adjust parent link count if we replace a child dir with another. Updates #2923. PiperOrigin-RevId: 317314460 --- pkg/sentry/fsimpl/gofer/filesystem.go | 3 ++- test/syscalls/BUILD | 1 + 2 files changed, 3 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 21eb976cb..5501781ac 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -1196,7 +1196,8 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa if newParent.cachedMetadataAuthoritative() { newParent.dirents = nil newParent.touchCMtime() - if renamed.isDir() { + if renamed.isDir() && (replaced == nil || !replaced.isDir()) { + // Increase the link count if we did not replace another directory. newParent.incLinks() } } diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 7b0248a39..131f09ab4 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -572,6 +572,7 @@ syscall_test( size = "medium", add_overlay = True, test = "//test/syscalls/linux:rename_test", + vfs2 = "True", ) syscall_test( -- cgit v1.2.3 From a609fff9d1516a095341c2016ec36f952550a46f Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 19 Jun 2020 10:18:35 -0700 Subject: Fix vfs2 handling of preadv2/pwritev2 flags. Check for unsupported flags, and silently support RWF_HIPRI by doing nothing. From pkg/abi/linux/file.go: "gVisor does not implement the RWF_HIPRI feature, but the flag is accepted as a valid flag argument for preadv2/pwritev2." Updates #2923. PiperOrigin-RevId: 317330631 --- pkg/sentry/fsimpl/gofer/regular_file.go | 9 +++++++-- pkg/sentry/fsimpl/gofer/special_file.go | 8 ++++++-- pkg/sentry/fsimpl/tmpfs/regular_file.go | 12 ++++++++++++ test/syscalls/BUILD | 1 + 4 files changed, 26 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go index f7a15a2ed..404a452c4 100644 --- a/pkg/sentry/fsimpl/gofer/regular_file.go +++ b/pkg/sentry/fsimpl/gofer/regular_file.go @@ -72,7 +72,9 @@ func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offs if offset < 0 { return 0, syserror.EINVAL } - if opts.Flags != 0 { + + // Check that flags are supported. Silently ignore RWF_HIPRI. + if opts.Flags&^linux.RWF_HIPRI != 0 { return 0, syserror.EOPNOTSUPP } @@ -123,9 +125,12 @@ func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off if offset < 0 { return 0, syserror.EINVAL } - if opts.Flags != 0 { + + // Check that flags are supported. Silently ignore RWF_HIPRI. + if opts.Flags&^linux.RWF_HIPRI != 0 { return 0, syserror.EOPNOTSUPP } + limit, err := vfs.CheckLimit(ctx, offset, src.NumBytes()) if err != nil { return 0, err diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go index a92008208..a016cbae1 100644 --- a/pkg/sentry/fsimpl/gofer/special_file.go +++ b/pkg/sentry/fsimpl/gofer/special_file.go @@ -129,7 +129,9 @@ func (fd *specialFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offs if fd.seekable && offset < 0 { return 0, syserror.EINVAL } - if opts.Flags != 0 { + + // Check that flags are supported. Silently ignore RWF_HIPRI. + if opts.Flags&^linux.RWF_HIPRI != 0 { return 0, syserror.EOPNOTSUPP } @@ -173,7 +175,9 @@ func (fd *specialFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off if fd.seekable && offset < 0 { return 0, syserror.EINVAL } - if opts.Flags != 0 { + + // Check that flags are supported. Silently ignore RWF_HIPRI. + if opts.Flags&^linux.RWF_HIPRI != 0 { return 0, syserror.EOPNOTSUPP } diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index bfd9c5995..b805aadd0 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -279,6 +279,12 @@ func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offs if offset < 0 { return 0, syserror.EINVAL } + + // Check that flags are supported. Silently ignore RWF_HIPRI. + if opts.Flags&^linux.RWF_HIPRI != 0 { + return 0, syserror.EOPNOTSUPP + } + if dst.NumBytes() == 0 { return 0, nil } @@ -304,6 +310,12 @@ func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off if offset < 0 { return 0, syserror.EINVAL } + + // Check that flags are supported. Silently ignore RWF_HIPRI. + if opts.Flags&^linux.RWF_HIPRI != 0 { + return 0, syserror.EOPNOTSUPP + } + srclen := src.NumBytes() if srclen == 0 { return 0, nil diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 131f09ab4..9555dc308 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -463,6 +463,7 @@ syscall_test( syscall_test( add_overlay = True, test = "//test/syscalls/linux:preadv2_test", + vfs2 = "True", ) syscall_test( -- cgit v1.2.3 From d962f9f3842c5c352bc61411cf27e38ba2219317 Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Fri, 19 Jun 2020 11:41:37 -0700 Subject: Implement UDP cheksum verification. Test: - TestIncrementChecksumErrors Fixes #2943 PiperOrigin-RevId: 317348158 --- pkg/sentry/socket/netstack/netstack.go | 1 + pkg/sentry/socket/netstack/stack.go | 2 +- pkg/tcpip/tcpip.go | 6 + pkg/tcpip/transport/udp/endpoint.go | 18 ++ pkg/tcpip/transport/udp/udp_test.go | 320 +++++++++++++++++++++++++-------- 5 files changed, 276 insertions(+), 71 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 738277391..c0b63a803 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -191,6 +191,7 @@ var Metrics = tcpip.Stats{ MalformedPacketsReceived: mustCreateMetric("/netstack/udp/malformed_packets_received", "Number of incoming UDP datagrams dropped due to the UDP header being in a malformed state."), PacketsSent: mustCreateMetric("/netstack/udp/packets_sent", "Number of UDP datagrams sent."), PacketSendErrors: mustCreateMetric("/netstack/udp/packet_send_errors", "Number of UDP datagrams failed to be sent."), + ChecksumErrors: mustCreateMetric("/netstack/udp/checksum_errors", "Number of UDP datagrams dropped due to bad checksums."), }, } diff --git a/pkg/sentry/socket/netstack/stack.go b/pkg/sentry/socket/netstack/stack.go index ee11742a6..d2fb655ea 100644 --- a/pkg/sentry/socket/netstack/stack.go +++ b/pkg/sentry/socket/netstack/stack.go @@ -313,7 +313,7 @@ func (s *Stack) Statistics(stat interface{}, arg string) error { udp.PacketsSent.Value(), // OutDatagrams. udp.ReceiveBufferErrors.Value(), // RcvbufErrors. 0, // Udp/SndbufErrors. - 0, // Udp/InCsumErrors. + udp.ChecksumErrors.Value(), // Udp/InCsumErrors. 0, // Udp/IgnoredMulti. } default: diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 3ad130b23..956232a44 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -1224,6 +1224,9 @@ type UDPStats struct { // PacketSendErrors is the number of datagrams failed to be sent. PacketSendErrors *StatCounter + + // ChecksumErrors is the number of datagrams dropped due to bad checksums. + ChecksumErrors *StatCounter } // Stats holds statistics about the networking stack. @@ -1267,6 +1270,9 @@ type ReceiveErrors struct { // ClosedReceiver is the number of received packets dropped because // of receiving endpoint state being closed. ClosedReceiver StatCounter + + // ChecksumErrors is the number of packets dropped due to bad checksums. + ChecksumErrors StatCounter } // SendErrors collects packet send errors within the transport layer for diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index f51988047..40d66ef09 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -1350,6 +1350,24 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk return } + // Verify checksum unless RX checksum offload is enabled. + // On IPv4, UDP checksum is optional, and a zero value means + // the transmitter omitted the checksum generation (RFC768). + // On IPv6, UDP checksum is not optional (RFC2460 Section 8.1). + if r.Capabilities()&stack.CapabilityRXChecksumOffload == 0 && + (hdr.Checksum() != 0 || r.NetProto == header.IPv6ProtocolNumber) { + xsum := r.PseudoHeaderChecksum(ProtocolNumber, hdr.Length()) + for _, v := range pkt.Data.Views() { + xsum = header.Checksum(v, xsum) + } + if hdr.CalculateChecksum(xsum) != 0xffff { + // Checksum Error. + e.stack.Stats().UDP.ChecksumErrors.Increment() + e.stats.ReceiveErrors.ChecksumErrors.Increment() + return + } + } + e.rcvMu.Lock() e.stack.Stats().UDP.PacketsReceived.Increment() e.stats.PacketsReceived.Increment() diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index 313a3f117..ff9f60cf9 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -292,15 +292,15 @@ func newDualTestContextWithOptions(t *testing.T, mtu uint32, options stack.Optio wep = sniffer.New(ep) } if err := s.CreateNIC(1, wep); err != nil { - t.Fatalf("CreateNIC failed: %v", err) + t.Fatalf("CreateNIC failed: %s", err) } if err := s.AddAddress(1, ipv4.ProtocolNumber, stackAddr); err != nil { - t.Fatalf("AddAddress failed: %v", err) + t.Fatalf("AddAddress failed: %s", err) } if err := s.AddAddress(1, ipv6.ProtocolNumber, stackV6Addr); err != nil { - t.Fatalf("AddAddress failed: %v", err) + t.Fatalf("AddAddress failed: %s", err) } s.SetRouteTable([]tcpip.Route{ @@ -391,17 +391,21 @@ func (c *testContext) injectPacket(flow testFlow, payload []byte) { h := flow.header4Tuple(incoming) if flow.isV4() { - c.injectV4Packet(payload, &h, true /* valid */) + buf := c.buildV4Packet(payload, &h) + c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) } else { - c.injectV6Packet(payload, &h, true /* valid */) + buf := c.buildV6Packet(payload, &h) + c.linkEP.InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) } } -// injectV6Packet creates a V6 test packet with the given payload and header -// values, and injects it into the link endpoint. valid indicates if the -// caller intends to inject a packet with a valid or an invalid UDP header. -// We can invalidate the header by corrupting the UDP payload length. -func (c *testContext) injectV6Packet(payload []byte, h *header4Tuple, valid bool) { +// buildV6Packet creates a V6 test packet with the given payload and header +// values in a buffer. +func (c *testContext) buildV6Packet(payload []byte, h *header4Tuple) buffer.View { // Allocate a buffer for data and headers. buf := buffer.NewView(header.UDPMinimumSize + header.IPv6MinimumSize + len(payload)) payloadStart := len(buf) - len(payload) @@ -420,16 +424,10 @@ func (c *testContext) injectV6Packet(payload []byte, h *header4Tuple, valid bool // Initialize the UDP header. u := header.UDP(buf[header.IPv6MinimumSize:]) - l := uint16(header.UDPMinimumSize + len(payload)) - if !valid { - // Change the UDP payload length to corrupt the header - // as requested by the caller. - l++ - } u.Encode(&header.UDPFields{ SrcPort: h.srcAddr.Port, DstPort: h.dstAddr.Port, - Length: l, + Length: uint16(header.UDPMinimumSize + len(payload)), }) // Calculate the UDP pseudo-header checksum. @@ -439,17 +437,12 @@ func (c *testContext) injectV6Packet(payload []byte, h *header4Tuple, valid bool xsum = header.Checksum(payload, xsum) u.SetChecksum(^u.CalculateChecksum(xsum)) - // Inject packet. - c.linkEP.InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ - Data: buf.ToVectorisedView(), - }) + return buf } -// injectV4Packet creates a V4 test packet with the given payload and header -// values, and injects it into the link endpoint. valid indicates if the -// caller intends to inject a packet with a valid or an invalid UDP header. -// We can invalidate the header by corrupting the UDP payload length. -func (c *testContext) injectV4Packet(payload []byte, h *header4Tuple, valid bool) { +// buildV4Packet creates a V4 test packet with the given payload and header +// values in a buffer. +func (c *testContext) buildV4Packet(payload []byte, h *header4Tuple) buffer.View { // Allocate a buffer for data and headers. buf := buffer.NewView(header.UDPMinimumSize + header.IPv4MinimumSize + len(payload)) payloadStart := len(buf) - len(payload) @@ -483,11 +476,7 @@ func (c *testContext) injectV4Packet(payload []byte, h *header4Tuple, valid bool xsum = header.Checksum(payload, xsum) u.SetChecksum(^u.CalculateChecksum(xsum)) - // Inject packet. - - c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ - Data: buf.ToVectorisedView(), - }) + return buf } func newPayload() []byte { @@ -509,7 +498,7 @@ func TestBindToDeviceOption(t *testing.T) { ep, err := s.NewEndpoint(udp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) if err != nil { - t.Fatalf("NewEndpoint failed; %v", err) + t.Fatalf("NewEndpoint failed; %s", err) } defer ep.Close() @@ -643,7 +632,7 @@ func TestBindEphemeralPort(t *testing.T) { c.createEndpoint(ipv6.ProtocolNumber) if err := c.ep.Bind(tcpip.FullAddress{}); err != nil { - t.Fatalf("ep.Bind(...) failed: %v", err) + t.Fatalf("ep.Bind(...) failed: %s", err) } } @@ -654,19 +643,19 @@ func TestBindReservedPort(t *testing.T) { c.createEndpoint(ipv6.ProtocolNumber) if err := c.ep.Connect(tcpip.FullAddress{Addr: testV6Addr, Port: testPort}); err != nil { - c.t.Fatalf("Connect failed: %v", err) + c.t.Fatalf("Connect failed: %s", err) } addr, err := c.ep.GetLocalAddress() if err != nil { - t.Fatalf("GetLocalAddress failed: %v", err) + t.Fatalf("GetLocalAddress failed: %s", err) } // We can't bind the address reserved by the connected endpoint above. { ep, err := c.s.NewEndpoint(udp.ProtocolNumber, ipv6.ProtocolNumber, &c.wq) if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) + t.Fatalf("NewEndpoint failed: %s", err) } defer ep.Close() if got, want := ep.Bind(addr), tcpip.ErrPortInUse; got != want { @@ -677,7 +666,7 @@ func TestBindReservedPort(t *testing.T) { func() { ep, err := c.s.NewEndpoint(udp.ProtocolNumber, ipv4.ProtocolNumber, &c.wq) if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) + t.Fatalf("NewEndpoint failed: %s", err) } defer ep.Close() // We can't bind ipv4-any on the port reserved by the connected endpoint @@ -687,7 +676,7 @@ func TestBindReservedPort(t *testing.T) { } // We can bind an ipv4 address on this port, though. if err := ep.Bind(tcpip.FullAddress{Addr: stackAddr, Port: addr.Port}); err != nil { - t.Fatalf("ep.Bind(...) failed: %v", err) + t.Fatalf("ep.Bind(...) failed: %s", err) } }() @@ -697,11 +686,11 @@ func TestBindReservedPort(t *testing.T) { func() { ep, err := c.s.NewEndpoint(udp.ProtocolNumber, ipv4.ProtocolNumber, &c.wq) if err != nil { - t.Fatalf("NewEndpoint failed: %v", err) + t.Fatalf("NewEndpoint failed: %s", err) } defer ep.Close() if err := ep.Bind(tcpip.FullAddress{Port: addr.Port}); err != nil { - t.Fatalf("ep.Bind(...) failed: %v", err) + t.Fatalf("ep.Bind(...) failed: %s", err) } }() } @@ -714,7 +703,7 @@ func TestV4ReadOnV6(t *testing.T) { // Bind to wildcard. if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) + c.t.Fatalf("Bind failed: %s", err) } // Test acceptance. @@ -729,7 +718,7 @@ func TestV4ReadOnBoundToV4MappedWildcard(t *testing.T) { // Bind to v4 mapped wildcard. if err := c.ep.Bind(tcpip.FullAddress{Addr: v4MappedWildcardAddr, Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) + c.t.Fatalf("Bind failed: %s", err) } // Test acceptance. @@ -744,7 +733,7 @@ func TestV4ReadOnBoundToV4Mapped(t *testing.T) { // Bind to local address. if err := c.ep.Bind(tcpip.FullAddress{Addr: stackV4MappedAddr, Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) + c.t.Fatalf("Bind failed: %s", err) } // Test acceptance. @@ -759,7 +748,7 @@ func TestV6ReadOnV6(t *testing.T) { // Bind to wildcard. if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) + c.t.Fatalf("Bind failed: %s", err) } // Test acceptance. @@ -796,7 +785,10 @@ func TestV4ReadSelfSource(t *testing.T) { h := unicastV4.header4Tuple(incoming) h.srcAddr = h.dstAddr - c.injectV4Packet(payload, &h, true /* valid */) + buf := c.buildV4Packet(payload, &h) + c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) if got := c.s.Stats().IP.InvalidSourceAddressesReceived.Value(); got != tt.wantInvalidSource { t.Errorf("c.s.Stats().IP.InvalidSourceAddressesReceived got %d, want %d", got, tt.wantInvalidSource) @@ -817,7 +809,7 @@ func TestV4ReadOnV4(t *testing.T) { // Bind to wildcard. if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) + c.t.Fatalf("Bind failed: %s", err) } // Test acceptance. @@ -955,7 +947,7 @@ func testWriteInternal(c *testContext, flow testFlow, setDest bool, checkers ... payload := buffer.View(newPayload()) n, _, err := c.ep.Write(tcpip.SlicePayload(payload), writeOpts) if err != nil { - c.t.Fatalf("Write failed: %v", err) + c.t.Fatalf("Write failed: %s", err) } if n != int64(len(payload)) { c.t.Fatalf("Bad number of bytes written: got %v, want %v", n, len(payload)) @@ -1005,7 +997,7 @@ func TestDualWriteBoundToWildcard(t *testing.T) { // Bind to wildcard. if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) + c.t.Fatalf("Bind failed: %s", err) } p := testDualWrite(c) @@ -1022,7 +1014,7 @@ func TestDualWriteConnectedToV6(t *testing.T) { // Connect to v6 address. if err := c.ep.Connect(tcpip.FullAddress{Addr: testV6Addr, Port: testPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) + c.t.Fatalf("Bind failed: %s", err) } testWrite(c, unicastV6) @@ -1043,7 +1035,7 @@ func TestDualWriteConnectedToV4Mapped(t *testing.T) { // Connect to v4 mapped address. if err := c.ep.Connect(tcpip.FullAddress{Addr: testV4MappedAddr, Port: testPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) + c.t.Fatalf("Bind failed: %s", err) } testWrite(c, unicastV4in6) @@ -1070,7 +1062,7 @@ func TestV6WriteOnBoundToV4Mapped(t *testing.T) { // Bind to v4 mapped address. if err := c.ep.Bind(tcpip.FullAddress{Addr: stackV4MappedAddr, Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) + c.t.Fatalf("Bind failed: %s", err) } // Write to v6 address. @@ -1085,7 +1077,7 @@ func TestV6WriteOnConnected(t *testing.T) { // Connect to v6 address. if err := c.ep.Connect(tcpip.FullAddress{Addr: testV6Addr, Port: testPort}); err != nil { - c.t.Fatalf("Connect failed: %v", err) + c.t.Fatalf("Connect failed: %s", err) } testWriteWithoutDestination(c, unicastV6) @@ -1099,7 +1091,7 @@ func TestV4WriteOnConnected(t *testing.T) { // Connect to v4 mapped address. if err := c.ep.Connect(tcpip.FullAddress{Addr: testV4MappedAddr, Port: testPort}); err != nil { - c.t.Fatalf("Connect failed: %v", err) + c.t.Fatalf("Connect failed: %s", err) } testWriteWithoutDestination(c, unicastV4) @@ -1234,7 +1226,7 @@ func TestReadIncrementsPacketsReceived(t *testing.T) { // Bind to wildcard. if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) + c.t.Fatalf("Bind failed: %s", err) } testRead(c, unicastV4) @@ -1506,12 +1498,12 @@ func TestMulticastInterfaceOption(t *testing.T) { Port: stackPort, } if err := c.ep.Connect(addr); err != nil { - c.t.Fatalf("Connect failed: %v", err) + c.t.Fatalf("Connect failed: %s", err) } } if err := c.ep.SetSockOpt(ifoptSet); err != nil { - c.t.Fatalf("SetSockOpt failed: %v", err) + c.t.Fatalf("SetSockOpt failed: %s", err) } // Verify multicast interface addr and NIC were set correctly. @@ -1519,7 +1511,7 @@ func TestMulticastInterfaceOption(t *testing.T) { ifoptWant := tcpip.MulticastInterfaceOption{NIC: 1, InterfaceAddr: ifoptSet.InterfaceAddr} var ifoptGot tcpip.MulticastInterfaceOption if err := c.ep.GetSockOpt(&ifoptGot); err != nil { - c.t.Fatalf("GetSockOpt failed: %v", err) + c.t.Fatalf("GetSockOpt failed: %s", err) } if ifoptGot != ifoptWant { c.t.Errorf("got GetSockOpt() = %#v, want = %#v", ifoptGot, ifoptWant) @@ -1691,7 +1683,7 @@ func TestV6UnknownDestination(t *testing.T) { } // TestIncrementMalformedPacketsReceived verifies if the malformed received -// global and endpoint stats get incremented. +// global and endpoint stats are incremented. func TestIncrementMalformedPacketsReceived(t *testing.T) { c := newDualTestContext(t, defaultMTU) defer c.cleanup() @@ -1699,20 +1691,25 @@ func TestIncrementMalformedPacketsReceived(t *testing.T) { c.createEndpoint(ipv6.ProtocolNumber) // Bind to wildcard. if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) + c.t.Fatalf("Bind failed: %s", err) } payload := newPayload() - c.t.Helper() h := unicastV6.header4Tuple(incoming) - c.injectV6Packet(payload, &h, false /* !valid */) + buf := c.buildV6Packet(payload, &h) + // Invalidate the packet length field in the UDP header by adding one. + u := header.UDP(buf[header.IPv6MinimumSize:]) + u.SetLength(u.Length() + 1) + c.linkEP.InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) - var want uint64 = 1 + const want = 1 if got := c.s.Stats().UDP.MalformedPacketsReceived.Value(); got != want { - t.Errorf("got stats.UDP.MalformedPacketsReceived.Value() = %v, want = %v", got, want) + t.Errorf("got stats.UDP.MalformedPacketsReceived.Value() = %d, want = %d", got, want) } if got := c.ep.Stats().(*tcpip.TransportEndpointStats).ReceiveErrors.MalformedPacketsReceived.Value(); got != want { - t.Errorf("got EP Stats.ReceiveErrors.MalformedPacketsReceived stats = %v, want = %v", got, want) + t.Errorf("got EP Stats.ReceiveErrors.MalformedPacketsReceived stats = %d, want = %d", got, want) } } @@ -1728,7 +1725,6 @@ func TestShortHeader(t *testing.T) { c.t.Fatalf("Bind failed: %s", err) } - c.t.Helper() h := unicastV6.header4Tuple(incoming) // Allocate a buffer for an IPv6 and too-short UDP header. @@ -1768,6 +1764,190 @@ func TestShortHeader(t *testing.T) { } } +// TestIncrementChecksumErrorsV4 verifies if a checksum error is detected, +// global and endpoint stats are incremented. +func TestIncrementChecksumErrorsV4(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpoint(ipv4.ProtocolNumber) + // Bind to wildcard. + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + c.t.Fatalf("Bind failed: %s", err) + } + + payload := newPayload() + h := unicastV4.header4Tuple(incoming) + buf := c.buildV4Packet(payload, &h) + // Invalidate the checksum field in the UDP header by adding one. + u := header.UDP(buf[header.IPv4MinimumSize:]) + u.SetChecksum(u.Checksum() + 1) + c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) + + const want = 1 + if got := c.s.Stats().UDP.ChecksumErrors.Value(); got != want { + t.Errorf("got stats.UDP.ChecksumErrors.Value() = %d, want = %d", got, want) + } + if got := c.ep.Stats().(*tcpip.TransportEndpointStats).ReceiveErrors.ChecksumErrors.Value(); got != want { + t.Errorf("got EP Stats.ReceiveErrors.ChecksumErrors stats = %d, want = %d", got, want) + } +} + +// TestIncrementChecksumErrorsV6 verifies if a checksum error is detected, +// global and endpoint stats are incremented. +func TestIncrementChecksumErrorsV6(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpoint(ipv6.ProtocolNumber) + // Bind to wildcard. + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + c.t.Fatalf("Bind failed: %s", err) + } + + payload := newPayload() + h := unicastV6.header4Tuple(incoming) + buf := c.buildV6Packet(payload, &h) + // Invalidate the checksum field in the UDP header by adding one. + u := header.UDP(buf[header.IPv6MinimumSize:]) + u.SetChecksum(u.Checksum() + 1) + c.linkEP.InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) + + const want = 1 + if got := c.s.Stats().UDP.ChecksumErrors.Value(); got != want { + t.Errorf("got stats.UDP.ChecksumErrors.Value() = %d, want = %d", got, want) + } + if got := c.ep.Stats().(*tcpip.TransportEndpointStats).ReceiveErrors.ChecksumErrors.Value(); got != want { + t.Errorf("got EP Stats.ReceiveErrors.ChecksumErrors stats = %d, want = %d", got, want) + } +} + +// TestPayloadModifiedV4 verifies if a checksum error is detected, +// global and endpoint stats are incremented. +func TestPayloadModifiedV4(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpoint(ipv4.ProtocolNumber) + // Bind to wildcard. + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + c.t.Fatalf("Bind failed: %s", err) + } + + payload := newPayload() + h := unicastV4.header4Tuple(incoming) + buf := c.buildV4Packet(payload, &h) + // Modify the payload so that the checksum value in the UDP header will be incorrect. + buf[len(buf)-1]++ + c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) + + const want = 1 + if got := c.s.Stats().UDP.ChecksumErrors.Value(); got != want { + t.Errorf("got stats.UDP.ChecksumErrors.Value() = %d, want = %d", got, want) + } + if got := c.ep.Stats().(*tcpip.TransportEndpointStats).ReceiveErrors.ChecksumErrors.Value(); got != want { + t.Errorf("got EP Stats.ReceiveErrors.ChecksumErrors stats = %d, want = %d", got, want) + } +} + +// TestPayloadModifiedV6 verifies if a checksum error is detected, +// global and endpoint stats are incremented. +func TestPayloadModifiedV6(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpoint(ipv6.ProtocolNumber) + // Bind to wildcard. + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + c.t.Fatalf("Bind failed: %s", err) + } + + payload := newPayload() + h := unicastV6.header4Tuple(incoming) + buf := c.buildV6Packet(payload, &h) + // Modify the payload so that the checksum value in the UDP header will be incorrect. + buf[len(buf)-1]++ + c.linkEP.InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) + + const want = 1 + if got := c.s.Stats().UDP.ChecksumErrors.Value(); got != want { + t.Errorf("got stats.UDP.ChecksumErrors.Value() = %d, want = %d", got, want) + } + if got := c.ep.Stats().(*tcpip.TransportEndpointStats).ReceiveErrors.ChecksumErrors.Value(); got != want { + t.Errorf("got EP Stats.ReceiveErrors.ChecksumErrors stats = %d, want = %d", got, want) + } +} + +// TestChecksumZeroV4 verifies if the checksum value is zero, global and +// endpoint states are *not* incremented (UDP checksum is optional on IPv4). +func TestChecksumZeroV4(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpoint(ipv4.ProtocolNumber) + // Bind to wildcard. + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + c.t.Fatalf("Bind failed: %s", err) + } + + payload := newPayload() + h := unicastV4.header4Tuple(incoming) + buf := c.buildV4Packet(payload, &h) + // Set the checksum field in the UDP header to zero. + u := header.UDP(buf[header.IPv4MinimumSize:]) + u.SetChecksum(0) + c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) + + const want = 0 + if got := c.s.Stats().UDP.ChecksumErrors.Value(); got != want { + t.Errorf("got stats.UDP.ChecksumErrors.Value() = %d, want = %d", got, want) + } + if got := c.ep.Stats().(*tcpip.TransportEndpointStats).ReceiveErrors.ChecksumErrors.Value(); got != want { + t.Errorf("got EP Stats.ReceiveErrors.ChecksumErrors stats = %d, want = %d", got, want) + } +} + +// TestChecksumZeroV6 verifies if the checksum value is zero, global and +// endpoint states are incremented (UDP checksum is *not* optional on IPv6). +func TestChecksumZeroV6(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpoint(ipv6.ProtocolNumber) + // Bind to wildcard. + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + c.t.Fatalf("Bind failed: %s", err) + } + + payload := newPayload() + h := unicastV6.header4Tuple(incoming) + buf := c.buildV6Packet(payload, &h) + // Set the checksum field in the UDP header to zero. + u := header.UDP(buf[header.IPv6MinimumSize:]) + u.SetChecksum(0) + c.linkEP.InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) + + const want = 1 + if got := c.s.Stats().UDP.ChecksumErrors.Value(); got != want { + t.Errorf("got stats.UDP.ChecksumErrors.Value() = %d, want = %d", got, want) + } + if got := c.ep.Stats().(*tcpip.TransportEndpointStats).ReceiveErrors.ChecksumErrors.Value(); got != want { + t.Errorf("got EP Stats.ReceiveErrors.ChecksumErrors stats = %d, want = %d", got, want) + } +} + // TestShutdownRead verifies endpoint read shutdown and error // stats increment on packet receive. func TestShutdownRead(t *testing.T) { @@ -1778,15 +1958,15 @@ func TestShutdownRead(t *testing.T) { // Bind to wildcard. if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { - c.t.Fatalf("Bind failed: %v", err) + c.t.Fatalf("Bind failed: %s", err) } if err := c.ep.Connect(tcpip.FullAddress{Addr: testV6Addr, Port: testPort}); err != nil { - c.t.Fatalf("Connect failed: %v", err) + c.t.Fatalf("Connect failed: %s", err) } if err := c.ep.Shutdown(tcpip.ShutdownRead); err != nil { - t.Fatalf("Shutdown failed: %v", err) + t.Fatalf("Shutdown failed: %s", err) } testFailingRead(c, unicastV6, true /* expectReadError */) @@ -1809,11 +1989,11 @@ func TestShutdownWrite(t *testing.T) { c.createEndpoint(ipv6.ProtocolNumber) if err := c.ep.Connect(tcpip.FullAddress{Addr: testV6Addr, Port: testPort}); err != nil { - c.t.Fatalf("Connect failed: %v", err) + c.t.Fatalf("Connect failed: %s", err) } if err := c.ep.Shutdown(tcpip.ShutdownWrite); err != nil { - t.Fatalf("Shutdown failed: %v", err) + t.Fatalf("Shutdown failed: %s", err) } testFailingWrite(c, unicastV6, tcpip.ErrClosedForSend) -- cgit v1.2.3 From 7db196c4dbb33562dbfaa28a9476c590f356b2b5 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 19 Jun 2020 11:48:24 -0700 Subject: Port fadvise64 to vfs2. Like vfs1, we have a trivial implementation that ignores all valid advice. Updates #2923. PiperOrigin-RevId: 317349505 --- pkg/abi/linux/BUILD | 1 + pkg/abi/linux/fadvise.go | 24 +++++++++++++++++++++ pkg/sentry/syscalls/linux/sys_file.go | 25 ++++++---------------- pkg/sentry/syscalls/linux/vfs2/fd.go | 38 ++++++++++++++++++++++++++++++++++ pkg/sentry/syscalls/linux/vfs2/vfs2.go | 2 +- test/syscalls/BUILD | 1 + 6 files changed, 71 insertions(+), 20 deletions(-) create mode 100644 pkg/abi/linux/fadvise.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/BUILD b/pkg/abi/linux/BUILD index 114b516e2..2b789c4ec 100644 --- a/pkg/abi/linux/BUILD +++ b/pkg/abi/linux/BUILD @@ -23,6 +23,7 @@ go_library( "errors.go", "eventfd.go", "exec.go", + "fadvise.go", "fcntl.go", "file.go", "file_amd64.go", diff --git a/pkg/abi/linux/fadvise.go b/pkg/abi/linux/fadvise.go new file mode 100644 index 000000000..b06ff9964 --- /dev/null +++ b/pkg/abi/linux/fadvise.go @@ -0,0 +1,24 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package linux + +const ( + POSIX_FADV_NORMAL = 0 + POSIX_FADV_RANDOM = 1 + POSIX_FADV_SEQUENTIAL = 2 + POSIX_FADV_WILLNEED = 3 + POSIX_FADV_DONTNEED = 4 + POSIX_FADV_NOREUSE = 5 +) diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go index 696e1c8d3..35eba20c5 100644 --- a/pkg/sentry/syscalls/linux/sys_file.go +++ b/pkg/sentry/syscalls/linux/sys_file.go @@ -1111,17 +1111,6 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } } -// LINT.ThenChange(vfs2/fd.go) - -const ( - _FADV_NORMAL = 0 - _FADV_RANDOM = 1 - _FADV_SEQUENTIAL = 2 - _FADV_WILLNEED = 3 - _FADV_DONTNEED = 4 - _FADV_NOREUSE = 5 -) - // Fadvise64 implements linux syscall fadvise64(2). // This implementation currently ignores the provided advice. func Fadvise64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { @@ -1146,12 +1135,12 @@ func Fadvise64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys } switch advice { - case _FADV_NORMAL: - case _FADV_RANDOM: - case _FADV_SEQUENTIAL: - case _FADV_WILLNEED: - case _FADV_DONTNEED: - case _FADV_NOREUSE: + case linux.POSIX_FADV_NORMAL: + case linux.POSIX_FADV_RANDOM: + case linux.POSIX_FADV_SEQUENTIAL: + case linux.POSIX_FADV_WILLNEED: + case linux.POSIX_FADV_DONTNEED: + case linux.POSIX_FADV_NOREUSE: default: return 0, nil, syserror.EINVAL } @@ -1160,8 +1149,6 @@ func Fadvise64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys return 0, nil, nil } -// LINT.IfChange - func mkdirAt(t *kernel.Task, dirFD int32, addr usermem.Addr, mode linux.FileMode) error { path, _, err := copyInPath(t, addr, false /* allowEmpty */) if err != nil { diff --git a/pkg/sentry/syscalls/linux/vfs2/fd.go b/pkg/sentry/syscalls/linux/vfs2/fd.go index f5eaa076b..e68b20bed 100644 --- a/pkg/sentry/syscalls/linux/vfs2/fd.go +++ b/pkg/sentry/syscalls/linux/vfs2/fd.go @@ -210,3 +210,41 @@ func posixLock(t *kernel.Task, args arch.SyscallArguments, file *vfs.FileDescrip return syserror.EINVAL } } + +// Fadvise64 implements fadvise64(2). +// This implementation currently ignores the provided advice. +func Fadvise64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + length := args[2].Int64() + advice := args[3].Int() + + // Note: offset is allowed to be negative. + if length < 0 { + return 0, nil, syserror.EINVAL + } + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // If the FD refers to a pipe or FIFO, return error. + if _, isPipe := file.Impl().(*pipe.VFSPipeFD); isPipe { + return 0, nil, syserror.ESPIPE + } + + switch advice { + case linux.POSIX_FADV_NORMAL: + case linux.POSIX_FADV_RANDOM: + case linux.POSIX_FADV_SEQUENTIAL: + case linux.POSIX_FADV_WILLNEED: + case linux.POSIX_FADV_DONTNEED: + case linux.POSIX_FADV_NOREUSE: + default: + return 0, nil, syserror.EINVAL + } + + // Sure, whatever. + return 0, nil, nil +} diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2.go b/pkg/sentry/syscalls/linux/vfs2/vfs2.go index caa6a98ff..b463edf2a 100644 --- a/pkg/sentry/syscalls/linux/vfs2/vfs2.go +++ b/pkg/sentry/syscalls/linux/vfs2/vfs2.go @@ -108,7 +108,7 @@ func Override() { s.Table[209] = syscalls.PartiallySupported("io_submit", IoSubmit, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}) s.Table[213] = syscalls.Supported("epoll_create", EpollCreate) s.Table[217] = syscalls.Supported("getdents64", Getdents64) - delete(s.Table, 221) // fdavise64 + s.Table[221] = syscalls.PartiallySupported("fadvise64", Fadvise64, "The syscall is 'supported', but ignores all provided advice.", nil) s.Table[232] = syscalls.Supported("epoll_wait", EpollWait) s.Table[233] = syscalls.Supported("epoll_ctl", EpollCtl) s.Table[235] = syscalls.Supported("utimes", Utimes) diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 9555dc308..3479358b6 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -191,6 +191,7 @@ syscall_test( syscall_test( add_overlay = True, test = "//test/syscalls/linux:fadvise64_test", + vfs2 = "True", ) syscall_test( -- cgit v1.2.3 From 8655fb72482e179923987759f378543b2f489f08 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 19 Jun 2020 13:35:02 -0700 Subject: Fix vfs2 proc/self/fd dirent iteration. Make proc/self/fd iteration work properly. Also, the comment on kernfs.Inode.IterDirents did not accurately reflect how parameters should be used/were used in kernfs.Inode impls other than fdDir. Updates #2923. PiperOrigin-RevId: 317370325 --- pkg/sentry/fsimpl/kernfs/kernfs.go | 8 ++++---- pkg/sentry/fsimpl/proc/task_fds.go | 3 +-- 2 files changed, 5 insertions(+), 6 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index bbee8ccda..07a9dc830 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -425,10 +425,10 @@ type inodeDynamicLookup interface { // IterDirents is used to iterate over dynamically created entries. It invokes // cb on each entry in the directory represented by the FileDescription. // 'offset' is the offset for the entire IterDirents call, which may include - // results from the caller. 'relOffset' is the offset inside the entries - // returned by this IterDirents invocation. In other words, - // 'offset+relOffset+1' is the value that should be set in vfs.Dirent.NextOff, - // while 'relOffset' is the place where iteration should start from. + // results from the caller (e.g. "." and ".."). 'relOffset' is the offset + // inside the entries returned by this IterDirents invocation. In other words, + // 'offset' should be used to calculate each vfs.Dirent.NextOff as well as + // the return value, while 'relOffset' is the place to start iteration. IterDirents(ctx context.Context, callback vfs.IterDirentsCallback, offset, relOffset int64) (newOffset int64, err error) } diff --git a/pkg/sentry/fsimpl/proc/task_fds.go b/pkg/sentry/fsimpl/proc/task_fds.go index 7debdb07a..fea29e5f0 100644 --- a/pkg/sentry/fsimpl/proc/task_fds.go +++ b/pkg/sentry/fsimpl/proc/task_fds.go @@ -64,7 +64,7 @@ type fdDir struct { } // IterDirents implements kernfs.inodeDynamicLookup. -func (i *fdDir) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, absOffset, relOffset int64) (int64, error) { +func (i *fdDir) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, offset, relOffset int64) (int64, error) { var fds []int32 i.task.WithMuLocked(func(t *kernel.Task) { if fdTable := t.FDTable(); fdTable != nil { @@ -72,7 +72,6 @@ func (i *fdDir) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, abs } }) - offset := absOffset + relOffset typ := uint8(linux.DT_REG) if i.produceSymlink { typ = linux.DT_LNK -- cgit v1.2.3 From ad9f4691741cfada0ae09f73053d6195d43465ae Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 19 Jun 2020 14:39:31 -0700 Subject: Fix bugs in vfs2 to make symlink tests pass. - Return ENOENT if target path is empty. - Make sure open(2) with O_CREAT|O_EXCL returns EEXIST when necessary. - Correctly update atime in tmpfs using touchATime(). Updates #2923. PiperOrigin-RevId: 317382655 --- pkg/sentry/fsimpl/tmpfs/filesystem.go | 7 +++++-- pkg/sentry/syscalls/linux/vfs2/filesystem.go | 3 +++ 2 files changed, 8 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index ac359cf7b..637d84e04 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -79,7 +79,7 @@ afterSymlink: } if symlink, ok := child.inode.impl.(*symlink); ok && rp.ShouldFollowSymlink() { // Symlink traversal updates access time. - atomic.StoreInt64(&d.inode.atime, d.inode.fs.clock.Now().Nanoseconds()) + child.inode.touchAtime(rp.Mount()) if err := rp.HandleSymlink(symlink.target); err != nil { return nil, err } @@ -372,6 +372,9 @@ afterTrailingSymlink: parentDir.inode.touchCMtime() return fd, nil } + if mustCreate { + return nil, syserror.EEXIST + } // Is the file mounted over? if err := rp.CheckMount(&child.vfsd); err != nil { return nil, err @@ -379,7 +382,7 @@ afterTrailingSymlink: // Do we need to resolve a trailing symlink? if symlink, ok := child.inode.impl.(*symlink); ok && rp.ShouldFollowSymlink() { // Symlink traversal updates access time. - atomic.StoreInt64(&child.inode.atime, child.inode.fs.clock.Now().Nanoseconds()) + child.inode.touchAtime(rp.Mount()) if err := rp.HandleSymlink(symlink.target); err != nil { return nil, err } diff --git a/pkg/sentry/syscalls/linux/vfs2/filesystem.go b/pkg/sentry/syscalls/linux/vfs2/filesystem.go index 46d3e189c..5dac77e4d 100644 --- a/pkg/sentry/syscalls/linux/vfs2/filesystem.go +++ b/pkg/sentry/syscalls/linux/vfs2/filesystem.go @@ -313,6 +313,9 @@ func symlinkat(t *kernel.Task, targetAddr usermem.Addr, newdirfd int32, linkpath if err != nil { return err } + if len(target) == 0 { + return syserror.ENOENT + } linkpath, err := copyInPath(t, linkpathAddr) if err != nil { return err -- cgit v1.2.3 From 00928d142dd580c44a392e8e51246b543dc4f957 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Sun, 21 Jun 2020 21:46:57 -0700 Subject: Fix vfs2 extended attributes. Correct behavior when given zero size arguments and trying to set user.* xattrs on files other than regular files or directories. Updates #2923. PiperOrigin-RevId: 317590409 --- pkg/sentry/fsimpl/gofer/gofer.go | 18 +++++++++++++++++- 1 file changed, 17 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index d8ae475ed..43c8153a4 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -1206,7 +1206,7 @@ func (d *dentry) setDeleted() { // We only support xattrs prefixed with "user." (see b/148380782). Currently, // there is no need to expose any other xattrs through a gofer. func (d *dentry) listxattr(ctx context.Context, creds *auth.Credentials, size uint64) ([]string, error) { - if d.file.isNil() { + if d.file.isNil() || !d.userXattrSupported() { return nil, nil } xattrMap, err := d.file.listXattr(ctx, size) @@ -1232,6 +1232,9 @@ func (d *dentry) getxattr(ctx context.Context, creds *auth.Credentials, opts *vf if !strings.HasPrefix(opts.Name, linux.XATTR_USER_PREFIX) { return "", syserror.EOPNOTSUPP } + if !d.userXattrSupported() { + return "", syserror.ENODATA + } return d.file.getXattr(ctx, opts.Name, opts.Size) } @@ -1245,6 +1248,9 @@ func (d *dentry) setxattr(ctx context.Context, creds *auth.Credentials, opts *vf if !strings.HasPrefix(opts.Name, linux.XATTR_USER_PREFIX) { return syserror.EOPNOTSUPP } + if !d.userXattrSupported() { + return syserror.EPERM + } return d.file.setXattr(ctx, opts.Name, opts.Value, opts.Flags) } @@ -1258,9 +1264,19 @@ func (d *dentry) removexattr(ctx context.Context, creds *auth.Credentials, name if !strings.HasPrefix(name, linux.XATTR_USER_PREFIX) { return syserror.EOPNOTSUPP } + if !d.userXattrSupported() { + return syserror.EPERM + } return d.file.removeXattr(ctx, name) } +// Extended attributes in the user.* namespace are only supported for regular +// files and directories. +func (d *dentry) userXattrSupported() bool { + filetype := linux.S_IFMT & atomic.LoadUint32(&d.mode) + return filetype == linux.S_IFREG || filetype == linux.S_IFDIR +} + // Preconditions: !d.isSynthetic(). d.isRegularFile() || d.isDirectory(). func (d *dentry) ensureSharedHandle(ctx context.Context, read, write, trunc bool) error { // O_TRUNC unconditionally requires us to obtain a new handle (opened with -- cgit v1.2.3 From 4573e7d863d59d59c6a4f72f396f72b0f6458cb2 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Mon, 22 Jun 2020 11:38:25 -0700 Subject: Check for invalid trailing / when traversing path in gofer OpenAt. Updates #2923. PiperOrigin-RevId: 317700049 --- pkg/sentry/fsimpl/gofer/filesystem.go | 8 +++++--- test/syscalls/BUILD | 1 + test/syscalls/linux/open.cc | 6 ++++++ 3 files changed, 12 insertions(+), 3 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 5501781ac..f065c4bad 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -767,15 +767,17 @@ afterTrailingSymlink: parent.dirMu.Unlock() return fd, err } + parent.dirMu.Unlock() if err != nil { - parent.dirMu.Unlock() return nil, err } - // Open existing child or follow symlink. - parent.dirMu.Unlock() if mustCreate { return nil, syserror.EEXIST } + if !child.isDir() && rp.MustBeDir() { + return nil, syserror.ENOTDIR + } + // Open existing child or follow symlink. if child.isSymlink() && rp.ShouldFollowSymlink() { target, err := child.readlink(ctx, rp.Mount()) if err != nil { diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 46a98a4cd..f94c383ae 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -181,6 +181,7 @@ syscall_test( size = "medium", add_overlay = True, test = "//test/syscalls/linux:exec_binary_test", + vfs2 = "True", ) syscall_test( diff --git a/test/syscalls/linux/open.cc b/test/syscalls/linux/open.cc index 670c0284b..bb7d108e8 100644 --- a/test/syscalls/linux/open.cc +++ b/test/syscalls/linux/open.cc @@ -439,6 +439,12 @@ TEST_F(OpenTest, CanTruncateWithStrangePermissions) { EXPECT_THAT(close(fd), SyscallSucceeds()); } +TEST_F(OpenTest, OpenNonDirectoryWithTrailingSlash) { + const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); + const std::string bad_path = file.path() + "/"; + EXPECT_THAT(open(bad_path.c_str(), O_RDONLY), SyscallFailsWithErrno(ENOTDIR)); +} + } // namespace } // namespace testing -- cgit v1.2.3 From 38d7b2fe5630a8f3169cfef7703921c4bc4056c2 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Mon, 22 Jun 2020 21:29:31 -0700 Subject: Only allow regular files, sockets, pipes, and char devices to be imported. PiperOrigin-RevId: 317796028 --- pkg/sentry/fsimpl/host/host.go | 35 +++++++++++++++++++---------------- 1 file changed, 19 insertions(+), 16 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 43a173bc9..a65543028 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -462,7 +462,8 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount, flags u // TODO(gvisor.dev/issue/1672): implement behavior corresponding to these allowed flags. flags &= syscall.O_ACCMODE | syscall.O_DIRECT | syscall.O_NONBLOCK | syscall.O_DSYNC | syscall.O_SYNC | syscall.O_APPEND - if fileType == syscall.S_IFSOCK { + switch fileType { + case syscall.S_IFSOCK: if i.isTTY { log.Warningf("cannot use host socket fd %d as TTY", i.hostFD) return nil, syserror.ENOTTY @@ -474,31 +475,33 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount, flags u } // Currently, we only allow Unix sockets to be imported. return unixsocket.NewFileDescription(ep, ep.Type(), flags, mnt, d, &i.locks) - } - // TODO(gvisor.dev/issue/1672): Allow only specific file types here, so - // that we don't allow importing arbitrary file types without proper - // support. - if i.isTTY { - fd := &TTYFileDescription{ - fileDescription: fileDescription{inode: i}, - termios: linux.DefaultSlaveTermios, + case syscall.S_IFREG, syscall.S_IFIFO, syscall.S_IFCHR: + if i.isTTY { + fd := &TTYFileDescription{ + fileDescription: fileDescription{inode: i}, + termios: linux.DefaultSlaveTermios, + } + fd.LockFD.Init(&i.locks) + vfsfd := &fd.vfsfd + if err := vfsfd.Init(fd, flags, mnt, d, &vfs.FileDescriptionOptions{}); err != nil { + return nil, err + } + return vfsfd, nil } + + fd := &fileDescription{inode: i} fd.LockFD.Init(&i.locks) vfsfd := &fd.vfsfd if err := vfsfd.Init(fd, flags, mnt, d, &vfs.FileDescriptionOptions{}); err != nil { return nil, err } return vfsfd, nil - } - fd := &fileDescription{inode: i} - fd.LockFD.Init(&i.locks) - vfsfd := &fd.vfsfd - if err := vfsfd.Init(fd, flags, mnt, d, &vfs.FileDescriptionOptions{}); err != nil { - return nil, err + default: + log.Warningf("cannot import host fd %d with file type %o", i.hostFD, fileType) + return nil, syserror.EPERM } - return vfsfd, nil } // fileDescription is embedded by host fd implementations of FileDescriptionImpl. -- cgit v1.2.3 From edea9a8d02d9877a05bb14046a26ebd9633f34f6 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Tue, 23 Jun 2020 14:51:43 -0700 Subject: Port readahead to VFS2. It preserves the same functionality (almost none) as in VFS1. Updates #2923 #1035 PiperOrigin-RevId: 317943522 --- pkg/sentry/syscalls/linux/vfs2/read_write.go | 33 ++++++++++++++++++++++++++++ pkg/sentry/syscalls/linux/vfs2/vfs2.go | 2 +- test/syscalls/BUILD | 1 + 3 files changed, 35 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/vfs2/read_write.go b/pkg/sentry/syscalls/linux/vfs2/read_write.go index 7f9debd4a..cd25597a7 100644 --- a/pkg/sentry/syscalls/linux/vfs2/read_write.go +++ b/pkg/sentry/syscalls/linux/vfs2/read_write.go @@ -606,3 +606,36 @@ func Lseek(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall newoff, err := file.Seek(t, offset, whence) return uintptr(newoff), nil, err } + +// Readahead implements readahead(2). +func Readahead(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + offset := args[1].Int64() + size := args[2].SizeT() + + file := t.GetFileVFS2(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Check that the file is readable. + if !file.IsReadable() { + return 0, nil, syserror.EBADF + } + + // Check that the size is valid. + if int(size) < 0 { + return 0, nil, syserror.EINVAL + } + + // Check that the offset is legitimate and does not overflow. + if offset < 0 || offset+int64(size) < 0 { + return 0, nil, syserror.EINVAL + } + + // Return EINVAL; if the underlying file type does not support readahead, + // then Linux will return EINVAL to indicate as much. In the future, we + // may extend this function to actually support readahead hints. + return 0, nil, syserror.EINVAL +} diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2.go b/pkg/sentry/syscalls/linux/vfs2/vfs2.go index b463edf2a..9e60c4a1c 100644 --- a/pkg/sentry/syscalls/linux/vfs2/vfs2.go +++ b/pkg/sentry/syscalls/linux/vfs2/vfs2.go @@ -92,7 +92,7 @@ func Override() { s.Table[162] = syscalls.Supported("sync", Sync) s.Table[165] = syscalls.Supported("mount", Mount) s.Table[166] = syscalls.Supported("umount2", Umount2) - delete(s.Table, 187) // readahead + s.Table[187] = syscalls.Supported("readahead", Readahead) s.Table[188] = syscalls.Supported("setxattr", Setxattr) s.Table[189] = syscalls.Supported("lsetxattr", Lsetxattr) s.Table[190] = syscalls.Supported("fsetxattr", Fsetxattr) diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index f94c383ae..8ce5d24f4 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -557,6 +557,7 @@ syscall_test( syscall_test( add_overlay = True, test = "//test/syscalls/linux:readahead_test", + vfs2 = "True", ) syscall_test( -- cgit v1.2.3 From 0c628c3152a727fff287a98897d83ee45ad990e5 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 23 Jun 2020 16:11:31 -0700 Subject: Support inotify in vfs2 gofer fs. MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Because there is no inode structure stored in the sandbox, inotify watches must be held on the dentry. This would be an issue in the presence of hard links, where multiple dentries would need to share the same set of watches, but in VFS2, we do not support the internal creation of hard links on gofer fs. As a result, we make the assumption that every dentry corresponds to a unique inode. Furthermore, dentries can be cached and then evicted, even if the underlying file has not be deleted. We must prevent this from occurring if there are any watches that would be lost. Note that if the dentry was deleted or invalidated (d.vfsd.IsDead()), we should still destroy it along with its watches. Additionally, when a dentry’s last watch is removed, we cache it if it also has zero references. This way, the dentry can eventually be evicted from memory if it is no longer needed. This is accomplished with a new dentry method, OnZeroWatches(), which is called by Inotify.RmWatch and Inotify.Release. Note that it must be called after all inotify locks are released to avoid violating lock order. Stress tests are added to make sure that inotify operations don't deadlock with gofer.OnZeroWatches. Updates #1479. PiperOrigin-RevId: 317958034 --- pkg/sentry/fsimpl/ext/dentry.go | 11 +- pkg/sentry/fsimpl/gofer/directory.go | 1 + pkg/sentry/fsimpl/gofer/filesystem.go | 59 ++++++- pkg/sentry/fsimpl/gofer/gofer.go | 73 +++++++-- pkg/sentry/fsimpl/kernfs/kernfs.go | 7 +- pkg/sentry/fsimpl/overlay/overlay.go | 5 + pkg/sentry/fsimpl/tmpfs/directory.go | 4 +- pkg/sentry/fsimpl/tmpfs/filesystem.go | 12 +- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 19 ++- pkg/sentry/syscalls/linux/vfs2/inotify.go | 7 +- pkg/sentry/vfs/anonfs.go | 7 +- pkg/sentry/vfs/dentry.go | 32 +++- pkg/sentry/vfs/inotify.go | 108 +++++++++---- pkg/sentry/vfs/vfs.go | 3 + test/syscalls/linux/BUILD | 1 + test/syscalls/linux/inotify.cc | 251 +++++++++++++++++++++++++++++- 16 files changed, 523 insertions(+), 77 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/ext/dentry.go b/pkg/sentry/fsimpl/ext/dentry.go index 6bd1a9fc6..55902322a 100644 --- a/pkg/sentry/fsimpl/ext/dentry.go +++ b/pkg/sentry/fsimpl/ext/dentry.go @@ -63,12 +63,17 @@ func (d *dentry) DecRef() { // InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. // -// TODO(gvisor.dev/issue/1479): Implement inotify. -func (d *dentry) InotifyWithParent(events uint32, cookie uint32, et vfs.EventType) {} +// TODO(b/134676337): Implement inotify. +func (d *dentry) InotifyWithParent(events, cookie uint32, et vfs.EventType) {} // Watches implements vfs.DentryImpl.Watches. // -// TODO(gvisor.dev/issue/1479): Implement inotify. +// TODO(b/134676337): Implement inotify. func (d *dentry) Watches() *vfs.Watches { return nil } + +// OnZeroWatches implements vfs.Dentry.OnZeroWatches. +// +// TODO(b/134676337): Implement inotify. +func (d *dentry) OnZeroWatches() {} diff --git a/pkg/sentry/fsimpl/gofer/directory.go b/pkg/sentry/fsimpl/gofer/directory.go index 480510b2a..5d83fe363 100644 --- a/pkg/sentry/fsimpl/gofer/directory.go +++ b/pkg/sentry/fsimpl/gofer/directory.go @@ -138,6 +138,7 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba fd.dirents = ds } + d.InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) if d.cachedMetadataAuthoritative() { d.touchAtime(fd.vfsfd.Mount()) } diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index f065c4bad..0321fd384 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -371,6 +371,11 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir } parent.touchCMtime() parent.dirents = nil + ev := linux.IN_CREATE + if dir { + ev |= linux.IN_ISDIR + } + parent.watches.Notify(name, uint32(ev), 0, vfs.InodeEvent) return nil } if fs.opts.interop == InteropModeShared { @@ -400,6 +405,11 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir } parent.touchCMtime() parent.dirents = nil + ev := linux.IN_CREATE + if dir { + ev |= linux.IN_ISDIR + } + parent.watches.Notify(name, uint32(ev), 0, vfs.InodeEvent) return nil } @@ -530,6 +540,18 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b return err } } + + // Generate inotify events for rmdir or unlink. + if dir { + parent.watches.Notify(name, linux.IN_DELETE|linux.IN_ISDIR, 0, vfs.InodeEvent) + } else { + var cw *vfs.Watches + if child != nil { + cw = &child.watches + } + vfs.InotifyRemoveChild(cw, &parent.watches, name) + } + if child != nil { vfsObj.CommitDeleteDentry(&child.vfsd) child.setDeleted() @@ -1018,6 +1040,7 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving } childVFSFD = &fd.vfsfd } + d.watches.Notify(name, linux.IN_CREATE, 0, vfs.PathEvent) return childVFSFD, nil } @@ -1203,6 +1226,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa newParent.incLinks() } } + vfs.InotifyRename(ctx, &renamed.watches, &oldParent.watches, &newParent.watches, oldName, newName, renamed.isDir()) return nil } @@ -1215,12 +1239,21 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { + fs.renameMuRUnlockAndCheckCaching(&ds) + return err + } + if err := d.setStat(ctx, rp.Credentials(), &opts.Stat, rp.Mount()); err != nil { + fs.renameMuRUnlockAndCheckCaching(&ds) return err } - return d.setStat(ctx, rp.Credentials(), &opts.Stat, rp.Mount()) + fs.renameMuRUnlockAndCheckCaching(&ds) + + if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { + d.InotifyWithParent(ev, 0, vfs.InodeEvent) + } + return nil } // StatAt implements vfs.FilesystemImpl.StatAt. @@ -1344,24 +1377,38 @@ func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { + fs.renameMuRUnlockAndCheckCaching(&ds) return err } - return d.setxattr(ctx, rp.Credentials(), &opts) + if err := d.setxattr(ctx, rp.Credentials(), &opts); err != nil { + fs.renameMuRUnlockAndCheckCaching(&ds) + return err + } + fs.renameMuRUnlockAndCheckCaching(&ds) + + d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) + return nil } // RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { + fs.renameMuRUnlockAndCheckCaching(&ds) + return err + } + if err := d.removexattr(ctx, rp.Credentials(), name); err != nil { + fs.renameMuRUnlockAndCheckCaching(&ds) return err } - return d.removexattr(ctx, rp.Credentials(), name) + fs.renameMuRUnlockAndCheckCaching(&ds) + + d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) + return nil } // PrependPath implements vfs.FilesystemImpl.PrependPath. diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 43c8153a4..c6c284ff3 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -665,6 +665,9 @@ type dentry struct { pipe *pipe.VFSPipe locks vfs.FileLocks + + // Inotify watches for this dentry. + watches vfs.Watches } // dentryAttrMask returns a p9.AttrMask enabling all attributes used by the @@ -947,6 +950,8 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin } else { atomic.StoreInt64(&d.atime, dentryTimestampFromStatx(stat.Atime)) } + // Restore mask bits that we cleared earlier. + stat.Mask |= linux.STATX_ATIME } if setLocalMtime { if stat.Mtime.Nsec == linux.UTIME_NOW { @@ -954,6 +959,8 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin } else { atomic.StoreInt64(&d.mtime, dentryTimestampFromStatx(stat.Mtime)) } + // Restore mask bits that we cleared earlier. + stat.Mask |= linux.STATX_MTIME } atomic.StoreInt64(&d.ctime, now) if stat.Mask&linux.STATX_SIZE != 0 { @@ -1051,15 +1058,34 @@ func (d *dentry) decRefLocked() { } // InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. -// -// TODO(gvisor.dev/issue/1479): Implement inotify. -func (d *dentry) InotifyWithParent(events uint32, cookie uint32, et vfs.EventType) {} +func (d *dentry) InotifyWithParent(events, cookie uint32, et vfs.EventType) { + if d.isDir() { + events |= linux.IN_ISDIR + } + + d.fs.renameMu.RLock() + // The ordering below is important, Linux always notifies the parent first. + if d.parent != nil { + d.parent.watches.NotifyWithExclusions(d.name, events, cookie, et, d.isDeleted()) + } + d.watches.Notify("", events, cookie, et) + d.fs.renameMu.RUnlock() +} // Watches implements vfs.DentryImpl.Watches. -// -// TODO(gvisor.dev/issue/1479): Implement inotify. func (d *dentry) Watches() *vfs.Watches { - return nil + return &d.watches +} + +// OnZeroWatches implements vfs.DentryImpl.OnZeroWatches. +// +// If no watches are left on this dentry and it has no references, cache it. +func (d *dentry) OnZeroWatches() { + if atomic.LoadInt64(&d.refs) == 0 { + d.fs.renameMu.Lock() + d.checkCachingLocked() + d.fs.renameMu.Unlock() + } } // checkCachingLocked should be called after d's reference count becomes 0 or it @@ -1093,6 +1119,9 @@ func (d *dentry) checkCachingLocked() { // Deleted and invalidated dentries with zero references are no longer // reachable by path resolution and should be dropped immediately. if d.vfsd.IsDead() { + if d.isDeleted() { + d.watches.HandleDeletion() + } if d.cached { d.fs.cachedDentries.Remove(d) d.fs.cachedDentriesLen-- @@ -1101,6 +1130,14 @@ func (d *dentry) checkCachingLocked() { d.destroyLocked() return } + // If d still has inotify watches and it is not deleted or invalidated, we + // cannot cache it and allow it to be evicted. Otherwise, we will lose its + // watches, even if a new dentry is created for the same file in the future. + // Note that the size of d.watches cannot concurrently transition from zero + // to non-zero, because adding a watch requires holding a reference on d. + if d.watches.Size() > 0 { + return + } // If d is already cached, just move it to the front of the LRU. if d.cached { d.fs.cachedDentries.Remove(d) @@ -1277,7 +1314,7 @@ func (d *dentry) userXattrSupported() bool { return filetype == linux.S_IFREG || filetype == linux.S_IFDIR } -// Preconditions: !d.isSynthetic(). d.isRegularFile() || d.isDirectory(). +// Preconditions: !d.isSynthetic(). d.isRegularFile() || d.isDir(). func (d *dentry) ensureSharedHandle(ctx context.Context, read, write, trunc bool) error { // O_TRUNC unconditionally requires us to obtain a new handle (opened with // O_TRUNC). @@ -1422,7 +1459,13 @@ func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linu // SetStat implements vfs.FileDescriptionImpl.SetStat. func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { - return fd.dentry().setStat(ctx, auth.CredentialsFromContext(ctx), &opts.Stat, fd.vfsfd.Mount()) + if err := fd.dentry().setStat(ctx, auth.CredentialsFromContext(ctx), &opts.Stat, fd.vfsfd.Mount()); err != nil { + return err + } + if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { + fd.dentry().InotifyWithParent(ev, 0, vfs.InodeEvent) + } + return nil } // Listxattr implements vfs.FileDescriptionImpl.Listxattr. @@ -1437,12 +1480,22 @@ func (fd *fileDescription) Getxattr(ctx context.Context, opts vfs.GetxattrOption // Setxattr implements vfs.FileDescriptionImpl.Setxattr. func (fd *fileDescription) Setxattr(ctx context.Context, opts vfs.SetxattrOptions) error { - return fd.dentry().setxattr(ctx, auth.CredentialsFromContext(ctx), &opts) + d := fd.dentry() + if err := d.setxattr(ctx, auth.CredentialsFromContext(ctx), &opts); err != nil { + return err + } + d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) + return nil } // Removexattr implements vfs.FileDescriptionImpl.Removexattr. func (fd *fileDescription) Removexattr(ctx context.Context, name string) error { - return fd.dentry().removexattr(ctx, auth.CredentialsFromContext(ctx), name) + d := fd.dentry() + if err := d.removexattr(ctx, auth.CredentialsFromContext(ctx), name); err != nil { + return err + } + d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) + return nil } // LockBSD implements vfs.FileDescriptionImpl.LockBSD. diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index 07a9dc830..55349f2a3 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -228,7 +228,7 @@ func (d *Dentry) destroy() { // InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. // // TODO(gvisor.dev/issue/1479): Implement inotify. -func (d *Dentry) InotifyWithParent(events uint32, cookie uint32, et vfs.EventType) {} +func (d *Dentry) InotifyWithParent(events, cookie uint32, et vfs.EventType) {} // Watches implements vfs.DentryImpl.Watches. // @@ -237,6 +237,11 @@ func (d *Dentry) Watches() *vfs.Watches { return nil } +// OnZeroWatches implements vfs.Dentry.OnZeroWatches. +// +// TODO(gvisor.dev/issue/1479): Implement inotify. +func (d *Dentry) OnZeroWatches() {} + // InsertChild inserts child into the vfs dentry cache with the given name under // this dentry. This does not update the directory inode, so calling this on // its own isn't sufficient to insert a child into a directory. InsertChild diff --git a/pkg/sentry/fsimpl/overlay/overlay.go b/pkg/sentry/fsimpl/overlay/overlay.go index e11a3ff19..e720d4825 100644 --- a/pkg/sentry/fsimpl/overlay/overlay.go +++ b/pkg/sentry/fsimpl/overlay/overlay.go @@ -528,6 +528,11 @@ func (d *dentry) Watches() *vfs.Watches { return nil } +// OnZeroWatches implements vfs.DentryImpl.OnZeroWatches. +// +// TODO(gvisor.dev/issue/1479): Implement inotify. +func (d *dentry) OnZeroWatches() {} + // iterLayers invokes yield on each layer comprising d, from top to bottom. If // any call to yield returns false, iterLayer stops iteration. func (d *dentry) iterLayers(yield func(vd vfs.VirtualDentry, isUpper bool) bool) { diff --git a/pkg/sentry/fsimpl/tmpfs/directory.go b/pkg/sentry/fsimpl/tmpfs/directory.go index 913b8a6c5..b172abc15 100644 --- a/pkg/sentry/fsimpl/tmpfs/directory.go +++ b/pkg/sentry/fsimpl/tmpfs/directory.go @@ -79,7 +79,6 @@ func (dir *directory) removeChildLocked(child *dentry) { dir.iterMu.Lock() dir.childList.Remove(child) dir.iterMu.Unlock() - child.unlinked = true } type directoryFD struct { @@ -107,13 +106,14 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba fs := fd.filesystem() dir := fd.inode().impl.(*directory) + defer fd.dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + // fs.mu is required to read d.parent and dentry.name. fs.mu.RLock() defer fs.mu.RUnlock() dir.iterMu.Lock() defer dir.iterMu.Unlock() - fd.dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) fd.inode().touchAtime(fd.vfsfd.Mount()) if fd.off == 0 { diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 637d84e04..85d8e37b2 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -638,14 +638,16 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error // SetStatAt implements vfs.FilesystemImpl.SetStatAt. func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { fs.mu.RLock() - defer fs.mu.RUnlock() d, err := resolveLocked(rp) if err != nil { + fs.mu.RUnlock() return err } if err := d.inode.setStat(ctx, rp.Credentials(), &opts.Stat); err != nil { + fs.mu.RUnlock() return err } + fs.mu.RUnlock() if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { d.InotifyWithParent(ev, 0, vfs.InodeEvent) @@ -788,14 +790,16 @@ func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt // SetxattrAt implements vfs.FilesystemImpl.SetxattrAt. func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { fs.mu.RLock() - defer fs.mu.RUnlock() d, err := resolveLocked(rp) if err != nil { + fs.mu.RUnlock() return err } if err := d.inode.setxattr(rp.Credentials(), &opts); err != nil { + fs.mu.RUnlock() return err } + fs.mu.RUnlock() d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil @@ -804,14 +808,16 @@ func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt // RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { fs.mu.RLock() - defer fs.mu.RUnlock() d, err := resolveLocked(rp) if err != nil { + fs.mu.RUnlock() return err } if err := d.inode.removexattr(rp.Credentials(), name); err != nil { + fs.mu.RUnlock() return err } + fs.mu.RUnlock() d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index a94333ee0..a85bfc968 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -215,11 +215,6 @@ type dentry struct { // filesystem.mu. name string - // unlinked indicates whether this dentry has been unlinked from its parent. - // It is only set to true on an unlink operation, and never set from true to - // false. unlinked is protected by filesystem.mu. - unlinked bool - // dentryEntry (ugh) links dentries into their parent directory.childList. dentryEntry @@ -259,18 +254,20 @@ func (d *dentry) DecRef() { } // InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. -func (d *dentry) InotifyWithParent(events uint32, cookie uint32, et vfs.EventType) { +func (d *dentry) InotifyWithParent(events, cookie uint32, et vfs.EventType) { if d.inode.isDir() { events |= linux.IN_ISDIR } + d.inode.fs.mu.RLock() // The ordering below is important, Linux always notifies the parent first. if d.parent != nil { - // Note that d.parent or d.name may be stale if there is a concurrent - // rename operation. Inotify does not provide consistency guarantees. - d.parent.inode.watches.NotifyWithExclusions(d.name, events, cookie, et, d.unlinked) + // tmpfs never calls VFS.InvalidateDentry(), so d.vfsd.IsDead() indicates + // that d was deleted. + d.parent.inode.watches.NotifyWithExclusions(d.name, events, cookie, et, d.vfsd.IsDead()) } d.inode.watches.Notify("", events, cookie, et) + d.inode.fs.mu.RUnlock() } // Watches implements vfs.DentryImpl.Watches. @@ -278,6 +275,9 @@ func (d *dentry) Watches() *vfs.Watches { return &d.inode.watches } +// OnZeroWatches implements vfs.Dentry.OnZeroWatches. +func (d *dentry) OnZeroWatches() {} + // inode represents a filesystem object. type inode struct { // fs is the owning filesystem. fs is immutable. @@ -336,7 +336,6 @@ func (i *inode) init(impl interface{}, fs *filesystem, kuid auth.KUID, kgid auth i.ctime = now i.mtime = now // i.nlink initialized by caller - i.watches = vfs.Watches{} i.impl = impl } diff --git a/pkg/sentry/syscalls/linux/vfs2/inotify.go b/pkg/sentry/syscalls/linux/vfs2/inotify.go index 7d50b6a16..74e5287b7 100644 --- a/pkg/sentry/syscalls/linux/vfs2/inotify.go +++ b/pkg/sentry/syscalls/linux/vfs2/inotify.go @@ -116,8 +116,11 @@ func InotifyAddWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kern } defer d.DecRef() - fd = ino.AddWatch(d.Dentry(), mask) - return uintptr(fd), nil, err + fd, err = ino.AddWatch(d.Dentry(), mask) + if err != nil { + return 0, nil, err + } + return uintptr(fd), nil, nil } // InotifyRmWatch implements the inotify_rm_watch() syscall. diff --git a/pkg/sentry/vfs/anonfs.go b/pkg/sentry/vfs/anonfs.go index b7c6b60b8..746af03ec 100644 --- a/pkg/sentry/vfs/anonfs.go +++ b/pkg/sentry/vfs/anonfs.go @@ -301,7 +301,7 @@ func (d *anonDentry) DecRef() { // InotifyWithParent implements DentryImpl.InotifyWithParent. // // TODO(gvisor.dev/issue/1479): Implement inotify. -func (d *anonDentry) InotifyWithParent(events uint32, cookie uint32, et EventType) {} +func (d *anonDentry) InotifyWithParent(events, cookie uint32, et EventType) {} // Watches implements DentryImpl.Watches. // @@ -309,3 +309,8 @@ func (d *anonDentry) InotifyWithParent(events uint32, cookie uint32, et EventTyp func (d *anonDentry) Watches() *Watches { return nil } + +// OnZeroWatches implements Dentry.OnZeroWatches. +// +// TODO(gvisor.dev/issue/1479): Implement inotify. +func (d *anonDentry) OnZeroWatches() {} diff --git a/pkg/sentry/vfs/dentry.go b/pkg/sentry/vfs/dentry.go index 24af13eb1..cea3e6955 100644 --- a/pkg/sentry/vfs/dentry.go +++ b/pkg/sentry/vfs/dentry.go @@ -113,12 +113,29 @@ type DentryImpl interface { // // Note that the events may not actually propagate up to the user, depending // on the event masks. - InotifyWithParent(events uint32, cookie uint32, et EventType) + InotifyWithParent(events, cookie uint32, et EventType) // Watches returns the set of inotify watches for the file corresponding to // the Dentry. Dentries that are hard links to the same underlying file // share the same watches. + // + // Watches may return nil if the dentry belongs to a FilesystemImpl that + // does not support inotify. If an implementation returns a non-nil watch + // set, it must always return a non-nil watch set. Likewise, if an + // implementation returns a nil watch set, it must always return a nil watch + // set. + // + // The caller does not need to hold a reference on the dentry. Watches() *Watches + + // OnZeroWatches is called whenever the number of watches on a dentry drops + // to zero. This is needed by some FilesystemImpls (e.g. gofer) to manage + // dentry lifetime. + // + // The caller does not need to hold a reference on the dentry. OnZeroWatches + // may acquire inotify locks, so to prevent deadlock, no inotify locks should + // be held by the caller. + OnZeroWatches() } // IncRef increments d's reference count. @@ -149,17 +166,26 @@ func (d *Dentry) isMounted() bool { return atomic.LoadUint32(&d.mounts) != 0 } -// InotifyWithParent notifies all watches on the inodes for this dentry and +// InotifyWithParent notifies all watches on the targets represented by d and // its parent of events. -func (d *Dentry) InotifyWithParent(events uint32, cookie uint32, et EventType) { +func (d *Dentry) InotifyWithParent(events, cookie uint32, et EventType) { d.impl.InotifyWithParent(events, cookie, et) } // Watches returns the set of inotify watches associated with d. +// +// Watches will return nil if d belongs to a FilesystemImpl that does not +// support inotify. func (d *Dentry) Watches() *Watches { return d.impl.Watches() } +// OnZeroWatches performs cleanup tasks whenever the number of watches on a +// dentry drops to zero. +func (d *Dentry) OnZeroWatches() { + d.impl.OnZeroWatches() +} + // The following functions are exported so that filesystem implementations can // use them. The vfs package, and users of VFS, should not call these // functions. diff --git a/pkg/sentry/vfs/inotify.go b/pkg/sentry/vfs/inotify.go index 7fa7d2d0c..36e7a3767 100644 --- a/pkg/sentry/vfs/inotify.go +++ b/pkg/sentry/vfs/inotify.go @@ -49,9 +49,6 @@ const ( // Inotify represents an inotify instance created by inotify_init(2) or // inotify_init1(2). Inotify implements FileDescriptionImpl. // -// Lock ordering: -// Inotify.mu -> Watches.mu -> Inotify.evMu -// // +stateify savable type Inotify struct { vfsfd FileDescription @@ -122,17 +119,31 @@ func NewInotifyFD(ctx context.Context, vfsObj *VirtualFilesystem, flags uint32) // Release implements FileDescriptionImpl.Release. Release removes all // watches and frees all resources for an inotify instance. func (i *Inotify) Release() { + var ds []*Dentry // We need to hold i.mu to avoid a race with concurrent calls to // Inotify.handleDeletion from Watches. There's no risk of Watches // accessing this Inotify after the destructor ends, because we remove all // references to it below. i.mu.Lock() - defer i.mu.Unlock() for _, w := range i.watches { // Remove references to the watch from the watches set on the target. We // don't need to worry about the references from i.watches, since this // file description is about to be destroyed. - w.set.Remove(i.id) + d := w.target + ws := d.Watches() + // Watchable dentries should never return a nil watch set. + if ws == nil { + panic("Cannot remove watch from an unwatchable dentry") + } + ws.Remove(i.id) + if ws.Size() == 0 { + ds = append(ds, d) + } + } + i.mu.Unlock() + + for _, d := range ds { + d.OnZeroWatches() } } @@ -272,20 +283,19 @@ func (i *Inotify) queueEvent(ev *Event) { // newWatchLocked creates and adds a new watch to target. // -// Precondition: i.mu must be locked. -func (i *Inotify) newWatchLocked(target *Dentry, mask uint32) *Watch { - targetWatches := target.Watches() +// Precondition: i.mu must be locked. ws must be the watch set for target d. +func (i *Inotify) newWatchLocked(d *Dentry, ws *Watches, mask uint32) *Watch { w := &Watch{ - owner: i, - wd: i.nextWatchIDLocked(), - set: targetWatches, - mask: mask, + owner: i, + wd: i.nextWatchIDLocked(), + target: d, + mask: mask, } // Hold the watch in this inotify instance as well as the watch set on the // target. i.watches[w.wd] = w - targetWatches.Add(w) + ws.Add(w) return w } @@ -312,7 +322,9 @@ func (i *Inotify) handleDeletion(w *Watch) { // AddWatch constructs a new inotify watch and adds it to the target. It // returns the watch descriptor returned by inotify_add_watch(2). -func (i *Inotify) AddWatch(target *Dentry, mask uint32) int32 { +// +// The caller must hold a reference on target. +func (i *Inotify) AddWatch(target *Dentry, mask uint32) (int32, error) { // Note: Locking this inotify instance protects the result returned by // Lookup() below. With the lock held, we know for sure the lookup result // won't become stale because it's impossible for *this* instance to @@ -320,8 +332,14 @@ func (i *Inotify) AddWatch(target *Dentry, mask uint32) int32 { i.mu.Lock() defer i.mu.Unlock() + ws := target.Watches() + if ws == nil { + // While Linux supports inotify watches on all filesystem types, watches on + // filesystems like kernfs are not generally useful, so we do not. + return 0, syserror.EPERM + } // Does the target already have a watch from this inotify instance? - if existing := target.Watches().Lookup(i.id); existing != nil { + if existing := ws.Lookup(i.id); existing != nil { newmask := mask if mask&linux.IN_MASK_ADD != 0 { // "Add (OR) events to watch mask for this pathname if it already @@ -329,12 +347,12 @@ func (i *Inotify) AddWatch(target *Dentry, mask uint32) int32 { newmask |= atomic.LoadUint32(&existing.mask) } atomic.StoreUint32(&existing.mask, newmask) - return existing.wd + return existing.wd, nil } // No existing watch, create a new watch. - w := i.newWatchLocked(target, mask) - return w.wd + w := i.newWatchLocked(target, ws, mask) + return w.wd, nil } // RmWatch looks up an inotify watch for the given 'wd' and configures the @@ -353,9 +371,19 @@ func (i *Inotify) RmWatch(wd int32) error { delete(i.watches, wd) // Remove the watch from the watch target. - w.set.Remove(w.OwnerID()) + ws := w.target.Watches() + // AddWatch ensures that w.target has a non-nil watch set. + if ws == nil { + panic("Watched dentry cannot have nil watch set") + } + ws.Remove(w.OwnerID()) + remaining := ws.Size() i.mu.Unlock() + if remaining == 0 { + w.target.OnZeroWatches() + } + // Generate the event for the removal. i.queueEvent(newEvent(wd, "", linux.IN_IGNORED, 0)) @@ -374,6 +402,13 @@ type Watches struct { ws map[uint64]*Watch } +// Size returns the number of watches held by w. +func (w *Watches) Size() int { + w.mu.Lock() + defer w.mu.Unlock() + return len(w.ws) +} + // Lookup returns the watch owned by an inotify instance with the given id. // Returns nil if no such watch exists. // @@ -459,10 +494,6 @@ func (w *Watches) NotifyWithExclusions(name string, events, cookie uint32, et Ev func (w *Watches) HandleDeletion() { w.Notify("", linux.IN_DELETE_SELF, 0, InodeEvent) - // TODO(gvisor.dev/issue/1479): This doesn't work because maps are not copied - // by value. Ideally, we wouldn't have this circular locking so we can just - // notify of IN_DELETE_SELF in the same loop below. - // // We can't hold w.mu while calling watch.handleDeletion to preserve lock // ordering w.r.t to the owner inotify instances. Instead, atomically move // the watches map into a local variable so we can iterate over it safely. @@ -495,9 +526,8 @@ type Watch struct { // Descriptor for this watch. This is unique across an inotify instance. wd int32 - // set is the watch set containing this watch. It belongs to the target file - // of this watch. - set *Watches + // target is a dentry representing the watch target. Its watch set contains this watch. + target *Dentry // Events being monitored via this watch. Must be accessed with atomic // memory operations. @@ -606,7 +636,7 @@ func (e *Event) setName(name string) { func (e *Event) sizeOf() int { s := inotifyEventBaseSize + int(e.len) if s < inotifyEventBaseSize { - panic("overflow") + panic("Overflowed event size") } return s } @@ -676,11 +706,15 @@ func InotifyEventFromStatMask(mask uint32) uint32 { } // InotifyRemoveChild sends the appriopriate notifications to the watch sets of -// the child being removed and its parent. +// the child being removed and its parent. Note that unlike most pairs of +// parent/child notifications, the child is notified first in this case. func InotifyRemoveChild(self, parent *Watches, name string) { - self.Notify("", linux.IN_ATTRIB, 0, InodeEvent) - parent.Notify(name, linux.IN_DELETE, 0, InodeEvent) - // TODO(gvisor.dev/issue/1479): implement IN_EXCL_UNLINK. + if self != nil { + self.Notify("", linux.IN_ATTRIB, 0, InodeEvent) + } + if parent != nil { + parent.Notify(name, linux.IN_DELETE, 0, InodeEvent) + } } // InotifyRename sends the appriopriate notifications to the watch sets of the @@ -691,8 +725,14 @@ func InotifyRename(ctx context.Context, renamed, oldParent, newParent *Watches, dirEv = linux.IN_ISDIR } cookie := uniqueid.InotifyCookie(ctx) - oldParent.Notify(oldName, dirEv|linux.IN_MOVED_FROM, cookie, InodeEvent) - newParent.Notify(newName, dirEv|linux.IN_MOVED_TO, cookie, InodeEvent) + if oldParent != nil { + oldParent.Notify(oldName, dirEv|linux.IN_MOVED_FROM, cookie, InodeEvent) + } + if newParent != nil { + newParent.Notify(newName, dirEv|linux.IN_MOVED_TO, cookie, InodeEvent) + } // Somewhat surprisingly, self move events do not have a cookie. - renamed.Notify("", linux.IN_MOVE_SELF, 0, InodeEvent) + if renamed != nil { + renamed.Notify("", linux.IN_MOVE_SELF, 0, InodeEvent) + } } diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 9acca8bc7..58c7ad778 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -24,6 +24,9 @@ // Locks acquired by FilesystemImpls between Prepare{Delete,Rename}Dentry and Commit{Delete,Rename*}Dentry // VirtualFilesystem.filesystemsMu // EpollInstance.mu +// Inotify.mu +// Watches.mu +// Inotify.evMu // VirtualFilesystem.fsTypesMu // // Locking Dentry.mu in multiple Dentries requires holding diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 078e4a284..7282d675e 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -974,6 +974,7 @@ cc_binary( "//test/util:thread_util", "@com_google_absl//absl/strings", "@com_google_absl//absl/strings:str_format", + "@com_google_absl//absl/synchronization", "@com_google_absl//absl/time", ], ) diff --git a/test/syscalls/linux/inotify.cc b/test/syscalls/linux/inotify.cc index 1d1a7171d..13e096e9c 100644 --- a/test/syscalls/linux/inotify.cc +++ b/test/syscalls/linux/inotify.cc @@ -29,6 +29,7 @@ #include "absl/strings/str_cat.h" #include "absl/strings/str_format.h" #include "absl/strings/str_join.h" +#include "absl/synchronization/mutex.h" #include "absl/time/clock.h" #include "absl/time/time.h" #include "test/util/epoll_util.h" @@ -1305,7 +1306,7 @@ TEST(Inotify, UtimesGeneratesAttribEvent) { const int wd = ASSERT_NO_ERRNO_AND_VALUE( InotifyAddWatch(fd.get(), root.path(), IN_ALL_EVENTS)); - struct timeval times[2] = {{1, 0}, {2, 0}}; + const struct timeval times[2] = {{1, 0}, {2, 0}}; EXPECT_THAT(futimes(file1_fd.get(), times), SyscallSucceeds()); const std::vector events = @@ -2018,7 +2019,7 @@ TEST(Inotify, ExcludeUnlinkInodeEvents_NoRandomSave) { Event(IN_MODIFY, wd, Basename(file.path())), })); - struct timeval times[2] = {{1, 0}, {2, 0}}; + const struct timeval times[2] = {{1, 0}, {2, 0}}; ASSERT_THAT(futimes(fd.get(), times), SyscallSucceeds()); events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); EXPECT_THAT(events, Are({Event(IN_ATTRIB, wd, Basename(file.path()))})); @@ -2030,6 +2031,252 @@ TEST(Inotify, ExcludeUnlinkInodeEvents_NoRandomSave) { EXPECT_THAT(events, Are({Event(IN_ATTRIB, wd, Basename(file.path()))})); } +// This test helps verify that the lock order of filesystem and inotify locks +// is respected when inotify instances and watch targets are concurrently being +// destroyed. +TEST(InotifyTest, InotifyAndTargetDestructionDoNotDeadlock_NoRandomSave) { + const DisableSave ds; // Too many syscalls. + + // A file descriptor protected by a mutex. This ensures that while a + // descriptor is in use, it cannot be closed and reused for a different file + // description. + struct atomic_fd { + int fd; + absl::Mutex mu; + }; + + // Set up initial inotify instances. + constexpr int num_fds = 3; + std::vector fds(num_fds); + for (int i = 0; i < num_fds; i++) { + int fd; + ASSERT_THAT(fd = inotify_init1(IN_NONBLOCK), SyscallSucceeds()); + fds[i].fd = fd; + } + + // Set up initial watch targets. + std::vector paths; + for (int i = 0; i < 3; i++) { + paths.push_back(NewTempAbsPath()); + ASSERT_THAT(mknod(paths[i].c_str(), S_IFREG | 0600, 0), SyscallSucceeds()); + } + + constexpr absl::Duration runtime = absl::Seconds(4); + + // Constantly replace each inotify instance with a new one. + auto replace_fds = [&] { + for (auto start = absl::Now(); absl::Now() - start < runtime;) { + for (auto& afd : fds) { + int new_fd; + ASSERT_THAT(new_fd = inotify_init1(IN_NONBLOCK), SyscallSucceeds()); + absl::MutexLock l(&afd.mu); + ASSERT_THAT(close(afd.fd), SyscallSucceeds()); + afd.fd = new_fd; + for (auto& p : paths) { + // inotify_add_watch may fail if the file at p was deleted. + ASSERT_THAT(inotify_add_watch(afd.fd, p.c_str(), IN_ALL_EVENTS), + AnyOf(SyscallSucceeds(), SyscallFailsWithErrno(ENOENT))); + } + } + sched_yield(); + } + }; + + std::list ts; + for (int i = 0; i < 3; i++) { + ts.emplace_back(replace_fds); + } + + // Constantly replace each watch target with a new one. + for (auto start = absl::Now(); absl::Now() - start < runtime;) { + for (auto& p : paths) { + ASSERT_THAT(unlink(p.c_str()), SyscallSucceeds()); + ASSERT_THAT(mknod(p.c_str(), S_IFREG | 0600, 0), SyscallSucceeds()); + } + sched_yield(); + } +} + +// This test helps verify that the lock order of filesystem and inotify locks +// is respected when adding/removing watches occurs concurrently with the +// removal of their targets. +TEST(InotifyTest, AddRemoveUnlinkDoNotDeadlock_NoRandomSave) { + const DisableSave ds; // Too many syscalls. + + // Set up inotify instances. + constexpr int num_fds = 3; + std::vector fds(num_fds); + for (int i = 0; i < num_fds; i++) { + ASSERT_THAT(fds[i] = inotify_init1(IN_NONBLOCK), SyscallSucceeds()); + } + + // Set up initial watch targets. + std::vector paths; + for (int i = 0; i < 3; i++) { + paths.push_back(NewTempAbsPath()); + ASSERT_THAT(mknod(paths[i].c_str(), S_IFREG | 0600, 0), SyscallSucceeds()); + } + + constexpr absl::Duration runtime = absl::Seconds(1); + + // Constantly add/remove watches for each inotify instance/watch target pair. + auto add_remove_watches = [&] { + for (auto start = absl::Now(); absl::Now() - start < runtime;) { + for (int fd : fds) { + for (auto& p : paths) { + // Do not assert on inotify_add_watch and inotify_rm_watch. They may + // fail if the file at p was deleted. inotify_add_watch may also fail + // if another thread beat us to adding a watch. + const int wd = inotify_add_watch(fd, p.c_str(), IN_ALL_EVENTS); + if (wd > 0) { + inotify_rm_watch(fd, wd); + } + } + } + sched_yield(); + } + }; + + std::list ts; + for (int i = 0; i < 15; i++) { + ts.emplace_back(add_remove_watches); + } + + // Constantly replace each watch target with a new one. + for (auto start = absl::Now(); absl::Now() - start < runtime;) { + for (auto& p : paths) { + ASSERT_THAT(unlink(p.c_str()), SyscallSucceeds()); + ASSERT_THAT(mknod(p.c_str(), S_IFREG | 0600, 0), SyscallSucceeds()); + } + sched_yield(); + } +} + +// This test helps verify that the lock order of filesystem and inotify locks +// is respected when many inotify events and filesystem operations occur +// simultaneously. +TEST(InotifyTest, NotifyNoDeadlock_NoRandomSave) { + const DisableSave ds; // Too many syscalls. + + const TempPath parent = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const std::string dir = parent.path(); + + // mu protects file, which will change on rename. + absl::Mutex mu; + std::string file = NewTempAbsPathInDir(dir); + ASSERT_THAT(mknod(file.c_str(), 0644 | S_IFREG, 0), SyscallSucceeds()); + + const absl::Duration runtime = absl::Milliseconds(300); + + // Add/remove watches on dir and file. + ScopedThread add_remove_watches([&] { + const FileDescriptor ifd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + int dir_wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(ifd.get(), dir, IN_ALL_EVENTS)); + int file_wd; + { + absl::ReaderMutexLock l(&mu); + file_wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(ifd.get(), file, IN_ALL_EVENTS)); + } + for (auto start = absl::Now(); absl::Now() - start < runtime;) { + ASSERT_THAT(inotify_rm_watch(ifd.get(), file_wd), SyscallSucceeds()); + ASSERT_THAT(inotify_rm_watch(ifd.get(), dir_wd), SyscallSucceeds()); + dir_wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(ifd.get(), dir, IN_ALL_EVENTS)); + { + absl::ReaderMutexLock l(&mu); + file_wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(ifd.get(), file, IN_ALL_EVENTS)); + } + sched_yield(); + } + }); + + // Modify attributes on dir and file. + ScopedThread stats([&] { + int fd, dir_fd; + { + absl::ReaderMutexLock l(&mu); + ASSERT_THAT(fd = open(file.c_str(), O_RDONLY), SyscallSucceeds()); + } + ASSERT_THAT(dir_fd = open(dir.c_str(), O_RDONLY | O_DIRECTORY), + SyscallSucceeds()); + const struct timeval times[2] = {{1, 0}, {2, 0}}; + + for (auto start = absl::Now(); absl::Now() - start < runtime;) { + { + absl::ReaderMutexLock l(&mu); + EXPECT_THAT(utimes(file.c_str(), times), SyscallSucceeds()); + } + EXPECT_THAT(futimes(fd, times), SyscallSucceeds()); + EXPECT_THAT(utimes(dir.c_str(), times), SyscallSucceeds()); + EXPECT_THAT(futimes(dir_fd, times), SyscallSucceeds()); + sched_yield(); + } + }); + + // Modify extended attributes on dir and file. + ScopedThread xattrs([&] { + // TODO(gvisor.dev/issue/1636): Support extended attributes in runsc gofer. + if (!IsRunningOnGvisor()) { + int fd; + { + absl::ReaderMutexLock l(&mu); + ASSERT_THAT(fd = open(file.c_str(), O_RDONLY), SyscallSucceeds()); + } + + const char* name = "user.test"; + int val = 123; + for (auto start = absl::Now(); absl::Now() - start < runtime;) { + { + absl::ReaderMutexLock l(&mu); + ASSERT_THAT( + setxattr(file.c_str(), name, &val, sizeof(val), /*flags=*/0), + SyscallSucceeds()); + ASSERT_THAT(removexattr(file.c_str(), name), SyscallSucceeds()); + } + + ASSERT_THAT(fsetxattr(fd, name, &val, sizeof(val), /*flags=*/0), + SyscallSucceeds()); + ASSERT_THAT(fremovexattr(fd, name), SyscallSucceeds()); + sched_yield(); + } + } + }); + + // Read and write file's contents. Read and write dir's entries. + ScopedThread read_write([&] { + int fd; + { + absl::ReaderMutexLock l(&mu); + ASSERT_THAT(fd = open(file.c_str(), O_RDWR), SyscallSucceeds()); + } + for (auto start = absl::Now(); absl::Now() - start < runtime;) { + int val = 123; + ASSERT_THAT(write(fd, &val, sizeof(val)), SyscallSucceeds()); + ASSERT_THAT(read(fd, &val, sizeof(val)), SyscallSucceeds()); + TempPath new_file = + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(dir)); + ASSERT_NO_ERRNO(ListDir(dir, false)); + new_file.reset(); + sched_yield(); + } + }); + + // Rename file. + for (auto start = absl::Now(); absl::Now() - start < runtime;) { + const std::string new_path = NewTempAbsPathInDir(dir); + { + absl::WriterMutexLock l(&mu); + ASSERT_THAT(rename(file.c_str(), new_path.c_str()), SyscallSucceeds()); + file = new_path; + } + sched_yield(); + } +} + } // namespace } // namespace testing } // namespace gvisor -- cgit v1.2.3 From 65a587dedf1a30b3614a66532d2b448026b9c540 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 23 Jun 2020 18:31:53 -0700 Subject: Complete inotify IN_EXCL_UNLINK implementation in VFS2. Events were only skipped on parent directories after their children were unlinked; events on the unlinked file itself need to be skipped as well. As a result, all Watches.Notify() calls need to know whether the dentry where the call came from was unlinked. Updates #1479. PiperOrigin-RevId: 317979476 --- pkg/sentry/fsimpl/gofer/filesystem.go | 8 ++-- pkg/sentry/fsimpl/gofer/gofer.go | 4 +- pkg/sentry/fsimpl/tmpfs/filesystem.go | 8 ++-- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 10 +++-- pkg/sentry/vfs/inotify.go | 39 +++++++--------- test/syscalls/linux/inotify.cc | 84 ++++++++++++++++++++++++++--------- 6 files changed, 95 insertions(+), 58 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 0321fd384..d253c996c 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -375,7 +375,7 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir if dir { ev |= linux.IN_ISDIR } - parent.watches.Notify(name, uint32(ev), 0, vfs.InodeEvent) + parent.watches.Notify(name, uint32(ev), 0, vfs.InodeEvent, false /* unlinked */) return nil } if fs.opts.interop == InteropModeShared { @@ -409,7 +409,7 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir if dir { ev |= linux.IN_ISDIR } - parent.watches.Notify(name, uint32(ev), 0, vfs.InodeEvent) + parent.watches.Notify(name, uint32(ev), 0, vfs.InodeEvent, false /* unlinked */) return nil } @@ -543,7 +543,7 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b // Generate inotify events for rmdir or unlink. if dir { - parent.watches.Notify(name, linux.IN_DELETE|linux.IN_ISDIR, 0, vfs.InodeEvent) + parent.watches.Notify(name, linux.IN_DELETE|linux.IN_ISDIR, 0, vfs.InodeEvent, true /* unlinked */) } else { var cw *vfs.Watches if child != nil { @@ -1040,7 +1040,7 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving } childVFSFD = &fd.vfsfd } - d.watches.Notify(name, linux.IN_CREATE, 0, vfs.PathEvent) + d.watches.Notify(name, linux.IN_CREATE, 0, vfs.PathEvent, false /* unlinked */) return childVFSFD, nil } diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index c6c284ff3..b0b6d8c64 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -1066,9 +1066,9 @@ func (d *dentry) InotifyWithParent(events, cookie uint32, et vfs.EventType) { d.fs.renameMu.RLock() // The ordering below is important, Linux always notifies the parent first. if d.parent != nil { - d.parent.watches.NotifyWithExclusions(d.name, events, cookie, et, d.isDeleted()) + d.parent.watches.Notify(d.name, events, cookie, et, d.isDeleted()) } - d.watches.Notify("", events, cookie, et) + d.watches.Notify("", events, cookie, et, d.isDeleted()) d.fs.renameMu.RUnlock() } diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 85d8e37b2..f766b7ce2 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -182,7 +182,7 @@ func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(pa if dir { ev |= linux.IN_ISDIR } - parentDir.inode.watches.Notify(name, uint32(ev), 0, vfs.InodeEvent) + parentDir.inode.watches.Notify(name, uint32(ev), 0, vfs.InodeEvent, false /* unlinked */) parentDir.inode.touchCMtime() return nil } @@ -251,7 +251,7 @@ func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. return syserror.EMLINK } i.incLinksLocked() - i.watches.Notify("", linux.IN_ATTRIB, 0, vfs.InodeEvent) + i.watches.Notify("", linux.IN_ATTRIB, 0, vfs.InodeEvent, false /* unlinked */) parentDir.insertChildLocked(fs.newDentry(i), name) return nil }) @@ -368,7 +368,7 @@ afterTrailingSymlink: if err != nil { return nil, err } - parentDir.inode.watches.Notify(name, linux.IN_CREATE, 0, vfs.PathEvent) + parentDir.inode.watches.Notify(name, linux.IN_CREATE, 0, vfs.PathEvent, false /* unlinked */) parentDir.inode.touchCMtime() return fd, nil } @@ -625,7 +625,7 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error return err } parentDir.removeChildLocked(child) - parentDir.inode.watches.Notify(name, linux.IN_DELETE|linux.IN_ISDIR, 0, vfs.InodeEvent) + parentDir.inode.watches.Notify(name, linux.IN_DELETE|linux.IN_ISDIR, 0, vfs.InodeEvent, true /* unlinked */) // Remove links for child, child/., and child/.. child.inode.decLinksLocked() child.inode.decLinksLocked() diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index a85bfc968..d7f4f0779 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -259,14 +259,16 @@ func (d *dentry) InotifyWithParent(events, cookie uint32, et vfs.EventType) { events |= linux.IN_ISDIR } + // tmpfs never calls VFS.InvalidateDentry(), so d.vfsd.IsDead() indicates + // that d was deleted. + deleted := d.vfsd.IsDead() + d.inode.fs.mu.RLock() // The ordering below is important, Linux always notifies the parent first. if d.parent != nil { - // tmpfs never calls VFS.InvalidateDentry(), so d.vfsd.IsDead() indicates - // that d was deleted. - d.parent.inode.watches.NotifyWithExclusions(d.name, events, cookie, et, d.vfsd.IsDead()) + d.parent.inode.watches.Notify(d.name, events, cookie, et, deleted) } - d.inode.watches.Notify("", events, cookie, et) + d.inode.watches.Notify("", events, cookie, et, deleted) d.inode.fs.mu.RUnlock() } diff --git a/pkg/sentry/vfs/inotify.go b/pkg/sentry/vfs/inotify.go index 36e7a3767..6043a0566 100644 --- a/pkg/sentry/vfs/inotify.go +++ b/pkg/sentry/vfs/inotify.go @@ -467,21 +467,16 @@ func (w *Watches) Remove(id uint64) { delete(w.ws, id) } -// Notify queues a new event with all watches in this set. -func (w *Watches) Notify(name string, events, cookie uint32, et EventType) { - w.NotifyWithExclusions(name, events, cookie, et, false) -} - -// NotifyWithExclusions queues a new event with watches in this set. Watches -// with IN_EXCL_UNLINK are skipped if the event is coming from a child that -// has been unlinked. -func (w *Watches) NotifyWithExclusions(name string, events, cookie uint32, et EventType, unlinked bool) { +// Notify queues a new event with watches in this set. Watches with +// IN_EXCL_UNLINK are skipped if the event is coming from a child that has been +// unlinked. +func (w *Watches) Notify(name string, events, cookie uint32, et EventType, unlinked bool) { // N.B. We don't defer the unlocks because Notify is in the hot path of // all IO operations, and the defer costs too much for small IO // operations. w.mu.RLock() for _, watch := range w.ws { - if unlinked && watch.ExcludeUnlinkedChildren() && et == PathEvent { + if unlinked && watch.ExcludeUnlinked() && et == PathEvent { continue } watch.Notify(name, events, cookie) @@ -492,7 +487,7 @@ func (w *Watches) NotifyWithExclusions(name string, events, cookie uint32, et Ev // HandleDeletion is called when the watch target is destroyed to emit // the appropriate events. func (w *Watches) HandleDeletion() { - w.Notify("", linux.IN_DELETE_SELF, 0, InodeEvent) + w.Notify("", linux.IN_DELETE_SELF, 0, InodeEvent, true /* unlinked */) // We can't hold w.mu while calling watch.handleDeletion to preserve lock // ordering w.r.t to the owner inotify instances. Instead, atomically move @@ -539,14 +534,12 @@ func (w *Watch) OwnerID() uint64 { return w.owner.id } -// ExcludeUnlinkedChildren indicates whether the watched object should continue -// to be notified of events of its children after they have been unlinked, e.g. -// for an open file descriptor. +// ExcludeUnlinked indicates whether the watched object should continue to be +// notified of events originating from a path that has been unlinked. // -// TODO(gvisor.dev/issue/1479): Implement IN_EXCL_UNLINK. -// We can do this by keeping track of the set of unlinked children in Watches -// to skip notification. -func (w *Watch) ExcludeUnlinkedChildren() bool { +// For example, if "foo/bar" is opened and then unlinked, operations on the +// open fd may be ignored by watches on "foo" and "foo/bar" with IN_EXCL_UNLINK. +func (w *Watch) ExcludeUnlinked() bool { return atomic.LoadUint32(&w.mask)&linux.IN_EXCL_UNLINK != 0 } @@ -710,10 +703,10 @@ func InotifyEventFromStatMask(mask uint32) uint32 { // parent/child notifications, the child is notified first in this case. func InotifyRemoveChild(self, parent *Watches, name string) { if self != nil { - self.Notify("", linux.IN_ATTRIB, 0, InodeEvent) + self.Notify("", linux.IN_ATTRIB, 0, InodeEvent, true /* unlinked */) } if parent != nil { - parent.Notify(name, linux.IN_DELETE, 0, InodeEvent) + parent.Notify(name, linux.IN_DELETE, 0, InodeEvent, true /* unlinked */) } } @@ -726,13 +719,13 @@ func InotifyRename(ctx context.Context, renamed, oldParent, newParent *Watches, } cookie := uniqueid.InotifyCookie(ctx) if oldParent != nil { - oldParent.Notify(oldName, dirEv|linux.IN_MOVED_FROM, cookie, InodeEvent) + oldParent.Notify(oldName, dirEv|linux.IN_MOVED_FROM, cookie, InodeEvent, false /* unlinked */) } if newParent != nil { - newParent.Notify(newName, dirEv|linux.IN_MOVED_TO, cookie, InodeEvent) + newParent.Notify(newName, dirEv|linux.IN_MOVED_TO, cookie, InodeEvent, false /* unlinked */) } // Somewhat surprisingly, self move events do not have a cookie. if renamed != nil { - renamed.Notify("", linux.IN_MOVE_SELF, 0, InodeEvent) + renamed.Notify("", linux.IN_MOVE_SELF, 0, InodeEvent, false /* unlinked */) } } diff --git a/test/syscalls/linux/inotify.cc b/test/syscalls/linux/inotify.cc index 13e096e9c..731c7046c 100644 --- a/test/syscalls/linux/inotify.cc +++ b/test/syscalls/linux/inotify.cc @@ -1840,9 +1840,7 @@ TEST(Inotify, IncludeUnlinkedFile_NoRandomSave) { const TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); const TempPath file = ASSERT_NO_ERRNO_AND_VALUE( TempPath::CreateFileWith(dir.path(), "123", TempPath::kDefaultFileMode)); - - const FileDescriptor fd = - ASSERT_NO_ERRNO_AND_VALUE(Open(file.path(), O_RDWR)); + FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(file.path(), O_RDWR)); const FileDescriptor inotify_fd = ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); @@ -1855,7 +1853,7 @@ TEST(Inotify, IncludeUnlinkedFile_NoRandomSave) { int val = 0; ASSERT_THAT(read(fd.get(), &val, sizeof(val)), SyscallSucceeds()); ASSERT_THAT(write(fd.get(), &val, sizeof(val)), SyscallSucceeds()); - const std::vector events = + std::vector events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); EXPECT_THAT(events, Are({ Event(IN_ATTRIB, file_wd), @@ -1865,6 +1863,15 @@ TEST(Inotify, IncludeUnlinkedFile_NoRandomSave) { Event(IN_MODIFY, dir_wd, Basename(file.path())), Event(IN_MODIFY, file_wd), })); + + fd.reset(); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({ + Event(IN_CLOSE_WRITE, dir_wd, Basename(file.path())), + Event(IN_CLOSE_WRITE, file_wd), + Event(IN_DELETE_SELF, file_wd), + Event(IN_IGNORED, file_wd), + })); } // Watches created with IN_EXCL_UNLINK will stop emitting events on fds for @@ -1881,13 +1888,14 @@ TEST(Inotify, ExcludeUnlink_NoRandomSave) { const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(dir.path())); - const FileDescriptor fd = - ASSERT_NO_ERRNO_AND_VALUE(Open(file.path(), O_RDWR)); + FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(file.path(), O_RDWR)); const FileDescriptor inotify_fd = ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); - const int wd = ASSERT_NO_ERRNO_AND_VALUE(InotifyAddWatch( + const int dir_wd = ASSERT_NO_ERRNO_AND_VALUE(InotifyAddWatch( inotify_fd.get(), dir.path(), IN_ALL_EVENTS | IN_EXCL_UNLINK)); + const int file_wd = ASSERT_NO_ERRNO_AND_VALUE(InotifyAddWatch( + inotify_fd.get(), file.path(), IN_ALL_EVENTS | IN_EXCL_UNLINK)); // Unlink the child, which should cause further operations on the open file // descriptor to be ignored. @@ -1895,14 +1903,28 @@ TEST(Inotify, ExcludeUnlink_NoRandomSave) { int val = 0; ASSERT_THAT(write(fd.get(), &val, sizeof(val)), SyscallSucceeds()); ASSERT_THAT(read(fd.get(), &val, sizeof(val)), SyscallSucceeds()); - const std::vector events = + std::vector events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); - EXPECT_THAT(events, Are({Event(IN_DELETE, wd, Basename(file.path()))})); + EXPECT_THAT(events, Are({ + Event(IN_ATTRIB, file_wd), + Event(IN_DELETE, dir_wd, Basename(file.path())), + })); + + fd.reset(); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + ASSERT_THAT(events, Are({ + Event(IN_DELETE_SELF, file_wd), + Event(IN_IGNORED, file_wd), + })); } // We need to disable S/R because there are filesystems where we cannot re-open // fds to an unlinked file across S/R, e.g. gofer-backed filesytems. TEST(Inotify, ExcludeUnlinkDirectory_NoRandomSave) { + // TODO(gvisor.dev/issue/1624): This test fails on VFS1. Remove once VFS1 is + // deleted. + SKIP_IF(IsRunningWithVFS1()); + const DisableSave ds; const TempPath parent = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); @@ -1912,20 +1934,29 @@ TEST(Inotify, ExcludeUnlinkDirectory_NoRandomSave) { const FileDescriptor inotify_fd = ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); - const FileDescriptor fd = + FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(dirPath.c_str(), O_RDONLY | O_DIRECTORY)); - const int wd = ASSERT_NO_ERRNO_AND_VALUE(InotifyAddWatch( + const int parent_wd = ASSERT_NO_ERRNO_AND_VALUE(InotifyAddWatch( inotify_fd.get(), parent.path(), IN_ALL_EVENTS | IN_EXCL_UNLINK)); + const int self_wd = ASSERT_NO_ERRNO_AND_VALUE(InotifyAddWatch( + inotify_fd.get(), dir.path(), IN_ALL_EVENTS | IN_EXCL_UNLINK)); // Unlink the dir, and then close the open fd. ASSERT_THAT(rmdir(dirPath.c_str()), SyscallSucceeds()); dir.reset(); - const std::vector events = + std::vector events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); // No close event should appear. ASSERT_THAT(events, - Are({Event(IN_DELETE | IN_ISDIR, wd, Basename(dirPath))})); + Are({Event(IN_DELETE | IN_ISDIR, parent_wd, Basename(dirPath))})); + + fd.reset(); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + ASSERT_THAT(events, Are({ + Event(IN_DELETE_SELF, self_wd), + Event(IN_IGNORED, self_wd), + })); } // If "dir/child" and "dir/child2" are links to the same file, and "dir/child" @@ -1989,10 +2020,6 @@ TEST(Inotify, ExcludeUnlinkInodeEvents_NoRandomSave) { const FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(file.path().c_str(), O_RDWR)); - const FileDescriptor inotify_fd = - ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); - const int wd = ASSERT_NO_ERRNO_AND_VALUE(InotifyAddWatch( - inotify_fd.get(), dir.path(), IN_ALL_EVENTS | IN_EXCL_UNLINK)); // NOTE(b/157163751): Create another link before unlinking. This is needed for // the gofer filesystem in gVisor, where open fds will not work once the link @@ -2006,6 +2033,13 @@ TEST(Inotify, ExcludeUnlinkInodeEvents_NoRandomSave) { ASSERT_THAT(rc, SyscallSucceeds()); } + const FileDescriptor inotify_fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + const int dir_wd = ASSERT_NO_ERRNO_AND_VALUE(InotifyAddWatch( + inotify_fd.get(), dir.path(), IN_ALL_EVENTS | IN_EXCL_UNLINK)); + const int file_wd = ASSERT_NO_ERRNO_AND_VALUE(InotifyAddWatch( + inotify_fd.get(), file.path(), IN_ALL_EVENTS | IN_EXCL_UNLINK)); + // Even after unlinking, inode-level operations will trigger events regardless // of IN_EXCL_UNLINK. ASSERT_THAT(unlink(file.path().c_str()), SyscallSucceeds()); @@ -2015,20 +2049,28 @@ TEST(Inotify, ExcludeUnlinkInodeEvents_NoRandomSave) { std::vector events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); EXPECT_THAT(events, Are({ - Event(IN_DELETE, wd, Basename(file.path())), - Event(IN_MODIFY, wd, Basename(file.path())), + Event(IN_ATTRIB, file_wd), + Event(IN_DELETE, dir_wd, Basename(file.path())), + Event(IN_MODIFY, dir_wd, Basename(file.path())), + Event(IN_MODIFY, file_wd), })); const struct timeval times[2] = {{1, 0}, {2, 0}}; ASSERT_THAT(futimes(fd.get(), times), SyscallSucceeds()); events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); - EXPECT_THAT(events, Are({Event(IN_ATTRIB, wd, Basename(file.path()))})); + EXPECT_THAT(events, Are({ + Event(IN_ATTRIB, dir_wd, Basename(file.path())), + Event(IN_ATTRIB, file_wd), + })); // S/R is disabled on this entire test due to behavior with unlink; it must // also be disabled after this point because of fchmod. ASSERT_THAT(fchmod(fd.get(), 0777), SyscallSucceeds()); events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); - EXPECT_THAT(events, Are({Event(IN_ATTRIB, wd, Basename(file.path()))})); + EXPECT_THAT(events, Are({ + Event(IN_ATTRIB, dir_wd, Basename(file.path())), + Event(IN_ATTRIB, file_wd), + })); } // This test helps verify that the lock order of filesystem and inotify locks -- cgit v1.2.3 From 0f328beb0d830a6e3b88b9f2cc1c9cb48f55c752 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Tue, 23 Jun 2020 18:47:22 -0700 Subject: Port /dev/tty device to VFS2. Support is limited to the functionality that exists in VFS1. Updates #2923 #1035 PiperOrigin-RevId: 317981417 --- pkg/sentry/devices/ttydev/BUILD | 16 +++++++ pkg/sentry/devices/ttydev/ttydev.go | 91 +++++++++++++++++++++++++++++++++++++ runsc/boot/BUILD | 1 + runsc/boot/vfs.go | 7 +++ test/syscalls/BUILD | 1 + 5 files changed, 116 insertions(+) create mode 100644 pkg/sentry/devices/ttydev/BUILD create mode 100644 pkg/sentry/devices/ttydev/ttydev.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/devices/ttydev/BUILD b/pkg/sentry/devices/ttydev/BUILD new file mode 100644 index 000000000..12e49b58a --- /dev/null +++ b/pkg/sentry/devices/ttydev/BUILD @@ -0,0 +1,16 @@ +load("//tools:defs.bzl", "go_library") + +licenses(["notice"]) + +go_library( + name = "ttydev", + srcs = ["ttydev.go"], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/abi/linux", + "//pkg/context", + "//pkg/sentry/fsimpl/devtmpfs", + "//pkg/sentry/vfs", + "//pkg/usermem", + ], +) diff --git a/pkg/sentry/devices/ttydev/ttydev.go b/pkg/sentry/devices/ttydev/ttydev.go new file mode 100644 index 000000000..fbb7fd92c --- /dev/null +++ b/pkg/sentry/devices/ttydev/ttydev.go @@ -0,0 +1,91 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package ttydev implements devices for /dev/tty and (eventually) +// /dev/console. +// +// TODO(b/159623826): Support /dev/console. +package ttydev + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/devtmpfs" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/usermem" +) + +const ( + // See drivers/tty/tty_io.c:tty_init(). + ttyDevMinor = 0 + consoleDevMinor = 1 +) + +// ttyDevice implements vfs.Device for /dev/tty. +type ttyDevice struct{} + +// Open implements vfs.Device.Open. +func (ttyDevice) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd := &ttyFD{} + if err := fd.vfsfd.Init(fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{ + UseDentryMetadata: true, + }); err != nil { + return nil, err + } + return &fd.vfsfd, nil +} + +// ttyFD implements vfs.FileDescriptionImpl for /dev/tty. +type ttyFD struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl + vfs.DentryMetadataFileDescriptionImpl + vfs.NoLockFD +} + +// Release implements vfs.FileDescriptionImpl.Release. +func (fd *ttyFD) Release() {} + +// PRead implements vfs.FileDescriptionImpl.PRead. +func (fd *ttyFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + return 0, nil +} + +// Read implements vfs.FileDescriptionImpl.Read. +func (fd *ttyFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + return 0, nil +} + +// PWrite implements vfs.FileDescriptionImpl.PWrite. +func (fd *ttyFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + return src.NumBytes(), nil +} + +// Write implements vfs.FileDescriptionImpl.Write. +func (fd *ttyFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + return src.NumBytes(), nil +} + +// Register registers all devices implemented by this package in vfsObj. +func Register(vfsObj *vfs.VirtualFilesystem) error { + return vfsObj.RegisterDevice(vfs.CharDevice, linux.TTYAUX_MAJOR, ttyDevMinor, ttyDevice{}, &vfs.RegisterDeviceOptions{ + GroupName: "tty", + }) +} + +// CreateDevtmpfsFiles creates device special files in dev representing all +// devices implemented by this package. +func CreateDevtmpfsFiles(ctx context.Context, dev *devtmpfs.Accessor) error { + return dev.CreateDeviceFile(ctx, "tty", vfs.CharDevice, linux.TTYAUX_MAJOR, ttyDevMinor, 0666 /* mode */) +} diff --git a/runsc/boot/BUILD b/runsc/boot/BUILD index 2b1e6b13e..b701c65d2 100644 --- a/runsc/boot/BUILD +++ b/runsc/boot/BUILD @@ -40,6 +40,7 @@ go_library( "//pkg/sentry/arch:registers_go_proto", "//pkg/sentry/control", "//pkg/sentry/devices/memdev", + "//pkg/sentry/devices/ttydev", "//pkg/sentry/fdimport", "//pkg/sentry/fs", "//pkg/sentry/fs/dev", diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go index d1653b279..2fdddc719 100644 --- a/runsc/boot/vfs.go +++ b/runsc/boot/vfs.go @@ -26,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/devices/memdev" + "gvisor.dev/gvisor/pkg/sentry/devices/ttydev" "gvisor.dev/gvisor/pkg/sentry/fs/user" "gvisor.dev/gvisor/pkg/sentry/fsimpl/devpts" "gvisor.dev/gvisor/pkg/sentry/fsimpl/devtmpfs" @@ -75,6 +76,9 @@ func registerFilesystems(ctx context.Context, vfsObj *vfs.VirtualFilesystem, cre if err := memdev.Register(vfsObj); err != nil { return fmt.Errorf("registering memdev: %w", err) } + if err := ttydev.Register(vfsObj); err != nil { + return fmt.Errorf("registering ttydev: %w", err) + } a, err := devtmpfs.NewAccessor(ctx, vfsObj, creds, devtmpfs.Name) if err != nil { return fmt.Errorf("creating devtmpfs accessor: %w", err) @@ -87,6 +91,9 @@ func registerFilesystems(ctx context.Context, vfsObj *vfs.VirtualFilesystem, cre if err := memdev.CreateDevtmpfsFiles(ctx, a); err != nil { return fmt.Errorf("creating devtmpfs files: %w", err) } + if err := ttydev.CreateDevtmpfsFiles(ctx, a); err != nil { + return fmt.Errorf("creating devtmpfs files: %w", err) + } return nil } diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 8ce5d24f4..315af9efe 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -147,6 +147,7 @@ syscall_test( syscall_test( test = "//test/syscalls/linux:dev_test", + vfs2 = "True", ) syscall_test( -- cgit v1.2.3 From 2189e0a660e8355167bee4939fffeb57f0312b5d Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 23 Jun 2020 19:22:45 -0700 Subject: Clean up hostfs TODOs. This CL does a handful of things: - Support O_DSYNC, O_SYNC - Support O_APPEND and document an unavoidable race condition - Ignore O_DIRECT; we probably don't want to allow applications to set O_DIRECT on the host fd itself. - Leave a TODO for supporting O_NONBLOCK, which is a simple fix once RWF_NOWAIT is supported. - Get rid of caching TODO; force_page_cache is not configurable for host fs in vfs1 or vfs2 after whitelist fs was removed. - For the remaining TODOs, link to more specific bugs. Fixes #1672. PiperOrigin-RevId: 317985269 --- pkg/sentry/fsimpl/host/host.go | 47 ++++++++++++++++++++++++++++-------------- 1 file changed, 32 insertions(+), 15 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index a65543028..d5e73ddae 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -459,8 +459,9 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount, flags u fileType := s.Mode & linux.FileTypeMask // Constrain flags to a subset we can handle. - // TODO(gvisor.dev/issue/1672): implement behavior corresponding to these allowed flags. - flags &= syscall.O_ACCMODE | syscall.O_DIRECT | syscall.O_NONBLOCK | syscall.O_DSYNC | syscall.O_SYNC | syscall.O_APPEND + // + // TODO(gvisor.dev/issue/2601): Support O_NONBLOCK by adding RWF_NOWAIT to pread/pwrite calls. + flags &= syscall.O_ACCMODE | syscall.O_NONBLOCK | syscall.O_DSYNC | syscall.O_SYNC | syscall.O_APPEND switch fileType { case syscall.S_IFSOCK: @@ -568,7 +569,7 @@ func (f *fileDescription) Read(ctx context.Context, dst usermem.IOSequence, opts } return n, err } - // TODO(gvisor.dev/issue/1672): Cache pages, when forced to do so. + f.offsetMu.Lock() n, err := readFromHostFD(ctx, i.hostFD, dst, f.offset, opts.Flags) f.offset += n @@ -577,7 +578,7 @@ func (f *fileDescription) Read(ctx context.Context, dst usermem.IOSequence, opts } func readFromHostFD(ctx context.Context, hostFD int, dst usermem.IOSequence, offset int64, flags uint32) (int64, error) { - // TODO(gvisor.dev/issue/1672): Support select preadv2 flags. + // TODO(gvisor.dev/issue/2601): Support select preadv2 flags. if flags != 0 { return 0, syserror.EOPNOTSUPP } @@ -589,41 +590,58 @@ func readFromHostFD(ctx context.Context, hostFD int, dst usermem.IOSequence, off // PWrite implements FileDescriptionImpl. func (f *fileDescription) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - i := f.inode - if !i.seekable { + if !f.inode.seekable { return 0, syserror.ESPIPE } - return writeToHostFD(ctx, i.hostFD, src, offset, opts.Flags) + return f.writeToHostFD(ctx, src, offset, opts.Flags) } // Write implements FileDescriptionImpl. func (f *fileDescription) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { i := f.inode if !i.seekable { - n, err := writeToHostFD(ctx, i.hostFD, src, -1, opts.Flags) + n, err := f.writeToHostFD(ctx, src, -1, opts.Flags) if isBlockError(err) { err = syserror.ErrWouldBlock } return n, err } - // TODO(gvisor.dev/issue/1672): Cache pages, when forced to do so. - // TODO(gvisor.dev/issue/1672): Write to end of file and update offset if O_APPEND is set on this file. + f.offsetMu.Lock() - n, err := writeToHostFD(ctx, i.hostFD, src, f.offset, opts.Flags) + // NOTE(gvisor.dev/issue/2983): O_APPEND may cause memory corruption if + // another process modifies the host file between retrieving the file size + // and writing to the host fd. This is an unavoidable race condition because + // we cannot enforce synchronization on the host. + if f.vfsfd.StatusFlags()&linux.O_APPEND != 0 { + var s syscall.Stat_t + if err := syscall.Fstat(i.hostFD, &s); err != nil { + f.offsetMu.Unlock() + return 0, err + } + f.offset = s.Size + } + n, err := f.writeToHostFD(ctx, src, f.offset, opts.Flags) f.offset += n f.offsetMu.Unlock() return n, err } -func writeToHostFD(ctx context.Context, hostFD int, src usermem.IOSequence, offset int64, flags uint32) (int64, error) { - // TODO(gvisor.dev/issue/1672): Support select pwritev2 flags. +func (f *fileDescription) writeToHostFD(ctx context.Context, src usermem.IOSequence, offset int64, flags uint32) (int64, error) { + hostFD := f.inode.hostFD + // TODO(gvisor.dev/issue/2601): Support select pwritev2 flags. if flags != 0 { return 0, syserror.EOPNOTSUPP } writer := hostfd.GetReadWriterAt(int32(hostFD), offset, flags) n, err := src.CopyInTo(ctx, writer) hostfd.PutReadWriterAt(writer) + // NOTE(gvisor.dev/issue/2979): We always sync everything, even for O_DSYNC. + if n > 0 && f.vfsfd.StatusFlags()&(linux.O_DSYNC|linux.O_SYNC) != 0 { + if syncErr := unix.Fsync(hostFD); syncErr != nil { + return int64(n), syncErr + } + } return int64(n), err } @@ -694,8 +712,7 @@ func (f *fileDescription) Seek(_ context.Context, offset int64, whence int32) (i // Sync implements FileDescriptionImpl. func (f *fileDescription) Sync(context.Context) error { - // TODO(gvisor.dev/issue/1672): Currently we do not support the SyncData - // optimization, so we always sync everything. + // TODO(gvisor.dev/issue/1897): Currently, we always sync everything. return unix.Fsync(f.inode.hostFD) } -- cgit v1.2.3 From 399c52888db609296fd1341ed0daa994ad2d02b0 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 23 Jun 2020 20:04:15 -0700 Subject: Resolve remaining inotify TODOs. Also refactor HandleDeletion(). Updates #1479. PiperOrigin-RevId: 317989000 --- pkg/sentry/fsimpl/kernfs/kernfs.go | 8 +++----- pkg/sentry/vfs/anonfs.go | 8 +++----- pkg/sentry/vfs/filesystem.go | 2 -- pkg/sentry/vfs/inotify.go | 42 ++++++++++++++++++-------------------- 4 files changed, 26 insertions(+), 34 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index 55349f2a3..596de1edf 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -227,19 +227,17 @@ func (d *Dentry) destroy() { // InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. // -// TODO(gvisor.dev/issue/1479): Implement inotify. +// Although Linux technically supports inotify on pseudo filesystems (inotify +// is implemented at the vfs layer), it is not particularly useful. It is left +// unimplemented until someone actually needs it. func (d *Dentry) InotifyWithParent(events, cookie uint32, et vfs.EventType) {} // Watches implements vfs.DentryImpl.Watches. -// -// TODO(gvisor.dev/issue/1479): Implement inotify. func (d *Dentry) Watches() *vfs.Watches { return nil } // OnZeroWatches implements vfs.Dentry.OnZeroWatches. -// -// TODO(gvisor.dev/issue/1479): Implement inotify. func (d *Dentry) OnZeroWatches() {} // InsertChild inserts child into the vfs dentry cache with the given name under diff --git a/pkg/sentry/vfs/anonfs.go b/pkg/sentry/vfs/anonfs.go index 746af03ec..641e3e502 100644 --- a/pkg/sentry/vfs/anonfs.go +++ b/pkg/sentry/vfs/anonfs.go @@ -300,17 +300,15 @@ func (d *anonDentry) DecRef() { // InotifyWithParent implements DentryImpl.InotifyWithParent. // -// TODO(gvisor.dev/issue/1479): Implement inotify. +// Although Linux technically supports inotify on pseudo filesystems (inotify +// is implemented at the vfs layer), it is not particularly useful. It is left +// unimplemented until someone actually needs it. func (d *anonDentry) InotifyWithParent(events, cookie uint32, et EventType) {} // Watches implements DentryImpl.Watches. -// -// TODO(gvisor.dev/issue/1479): Implement inotify. func (d *anonDentry) Watches() *Watches { return nil } // OnZeroWatches implements Dentry.OnZeroWatches. -// -// TODO(gvisor.dev/issue/1479): Implement inotify. func (d *anonDentry) OnZeroWatches() {} diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index 1edd584c9..6bb9ca180 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -524,8 +524,6 @@ type FilesystemImpl interface { // // Preconditions: vd.Mount().Filesystem().Impl() == this FilesystemImpl. PrependPath(ctx context.Context, vfsroot, vd VirtualDentry, b *fspath.Builder) error - - // TODO(gvisor.dev/issue/1479): inotify_add_watch() } // PrependPathAtVFSRootError is returned by implementations of diff --git a/pkg/sentry/vfs/inotify.go b/pkg/sentry/vfs/inotify.go index 6043a0566..35960394c 100644 --- a/pkg/sentry/vfs/inotify.go +++ b/pkg/sentry/vfs/inotify.go @@ -120,6 +120,7 @@ func NewInotifyFD(ctx context.Context, vfsObj *VirtualFilesystem, flags uint32) // watches and frees all resources for an inotify instance. func (i *Inotify) Release() { var ds []*Dentry + // We need to hold i.mu to avoid a race with concurrent calls to // Inotify.handleDeletion from Watches. There's no risk of Watches // accessing this Inotify after the destructor ends, because we remove all @@ -307,19 +308,6 @@ func (i *Inotify) nextWatchIDLocked() int32 { return i.nextWatchMinusOne } -// handleDeletion handles the deletion of the target of watch w. It removes w -// from i.watches and a watch removal event is generated. -func (i *Inotify) handleDeletion(w *Watch) { - i.mu.Lock() - _, found := i.watches[w.wd] - delete(i.watches, w.wd) - i.mu.Unlock() - - if found { - i.queueEvent(newEvent(w.wd, "", linux.IN_IGNORED, 0)) - } -} - // AddWatch constructs a new inotify watch and adds it to the target. It // returns the watch descriptor returned by inotify_add_watch(2). // @@ -484,8 +472,9 @@ func (w *Watches) Notify(name string, events, cookie uint32, et EventType, unlin w.mu.RUnlock() } -// HandleDeletion is called when the watch target is destroyed to emit -// the appropriate events. +// HandleDeletion is called when the watch target is destroyed. Clear the +// watch set, detach watches from the inotify instances they belong to, and +// generate the appropriate events. func (w *Watches) HandleDeletion() { w.Notify("", linux.IN_DELETE_SELF, 0, InodeEvent, true /* unlinked */) @@ -505,9 +494,23 @@ func (w *Watches) HandleDeletion() { w.ws = nil w.mu.Unlock() + // Remove each watch from its owner's watch set, and generate a corresponding + // watch removal event. for _, watch := range ws { - // TODO(gvisor.dev/issue/1479): consider refactoring this. - watch.handleDeletion() + i := watch.owner + i.mu.Lock() + _, found := i.watches[watch.wd] + delete(i.watches, watch.wd) + + // Release mutex before notifying waiters because we don't control what + // they can do. + i.mu.Unlock() + + // If watch was not found, it was removed from the inotify instance before + // we could get to it, in which case we should not generate an event. + if found { + i.queueEvent(newEvent(watch.wd, "", linux.IN_IGNORED, 0)) + } } } @@ -559,11 +562,6 @@ func (w *Watch) Notify(name string, events uint32, cookie uint32) { w.owner.queueEvent(newEvent(w.wd, name, matchedEvents, cookie)) } -// handleDeletion handles the deletion of w's target. -func (w *Watch) handleDeletion() { - w.owner.handleDeletion(w) -} - // Event represents a struct inotify_event from linux. // // +stateify savable -- cgit v1.2.3 From 364ac92baf83f2352f78b718090472639bd92a76 Mon Sep 17 00:00:00 2001 From: Adin Scannell Date: Tue, 23 Jun 2020 23:32:23 -0700 Subject: Support for saving pointers to fields in the state package. Previously, it was not possible to encode/decode an object graph which contained a pointer to a field within another type. This was because the encoder was previously unable to disambiguate a pointer to an object and a pointer within the object. This CL remedies this by constructing an address map tracking the full memory range object occupy. The encoded Refvalue message has been extended to allow references to children objects within another object. Because the encoding process may learn about object structure over time, we cannot encode any objects under the entire graph has been generated. This CL also updates the state package to use standard interfaces intead of reflection-based dispatch in order to improve performance overall. This includes a custom wire protocol to significantly reduce the number of allocations and take advantage of structure packing. As part of these changes, there are a small number of minor changes in other places of the code base: * The lists used during encoding are changed to use intrusive lists with the objectEncodeState directly, which required that the ilist Len() method is updated to work properly with the ElementMapper mechanism. * A bug is fixed in the list code wherein Remove() called on an element that is already removed can corrupt the list (removing the element if there's only a single element). Now the behavior is correct. * Standard error wrapping is introduced. * Compressio was updated to implement the new wire.Reader and wire.Writer inteface methods directly. The lack of a ReadByte and WriteByte caused issues not due to interface dispatch, but because underlying slices for a Read or Write call through an interface would always escape to the heap! * Statify has been updated to support the new APIs. See README.md for a description of how the new mechanism works. PiperOrigin-RevId: 318010298 --- pkg/compressio/compressio.go | 54 +- pkg/gohacks/BUILD | 1 + pkg/ilist/list.go | 6 +- pkg/sentry/kernel/BUILD | 1 + pkg/sentry/kernel/kernel.go | 22 +- pkg/sentry/pgalloc/BUILD | 1 + pkg/sentry/pgalloc/save_restore.go | 13 +- pkg/state/BUILD | 68 ++- pkg/state/README.md | 158 ++++++ pkg/state/decode.go | 918 ++++++++++++++++++-------------- pkg/state/decode_unsafe.go | 27 + pkg/state/encode.go | 1025 ++++++++++++++++++++++++------------ pkg/state/encode_unsafe.go | 48 -- pkg/state/map.go | 232 -------- pkg/state/object.proto | 140 ----- pkg/state/pretty/BUILD | 13 + pkg/state/pretty/pretty.go | 273 ++++++++++ pkg/state/printer.go | 251 --------- pkg/state/state.go | 360 ++++++------- pkg/state/state_norace.go | 19 + pkg/state/state_race.go | 19 + pkg/state/state_test.go | 721 ------------------------- pkg/state/statefile/BUILD | 1 + pkg/state/statefile/statefile.go | 15 +- pkg/state/stats.go | 117 ++-- pkg/state/tests/BUILD | 43 ++ pkg/state/tests/array.go | 35 ++ pkg/state/tests/array_test.go | 134 +++++ pkg/state/tests/bench.go | 24 + pkg/state/tests/bench_test.go | 153 ++++++ pkg/state/tests/bool_test.go | 31 ++ pkg/state/tests/float_test.go | 118 +++++ pkg/state/tests/integer.go | 163 ++++++ pkg/state/tests/integer_test.go | 94 ++++ pkg/state/tests/load.go | 61 +++ pkg/state/tests/load_test.go | 70 +++ pkg/state/tests/map.go | 28 + pkg/state/tests/map_test.go | 90 ++++ pkg/state/tests/register.go | 21 + pkg/state/tests/register_test.go | 167 ++++++ pkg/state/tests/string_test.go | 34 ++ pkg/state/tests/struct.go | 65 +++ pkg/state/tests/struct_test.go | 89 ++++ pkg/state/tests/tests.go | 215 ++++++++ pkg/state/types.go | 361 +++++++++++++ pkg/state/wire/BUILD | 12 + pkg/state/wire/wire.go | 970 ++++++++++++++++++++++++++++++++++ runsc/cmd/BUILD | 2 +- runsc/cmd/statefile.go | 12 +- tools/checkescape/checkescape.go | 4 +- tools/go_stateify/main.go | 182 ++++--- 51 files changed, 5171 insertions(+), 2510 deletions(-) create mode 100644 pkg/state/README.md create mode 100644 pkg/state/decode_unsafe.go delete mode 100644 pkg/state/map.go delete mode 100644 pkg/state/object.proto create mode 100644 pkg/state/pretty/BUILD create mode 100644 pkg/state/pretty/pretty.go delete mode 100644 pkg/state/printer.go create mode 100644 pkg/state/state_norace.go create mode 100644 pkg/state/state_race.go delete mode 100644 pkg/state/state_test.go create mode 100644 pkg/state/tests/BUILD create mode 100644 pkg/state/tests/array.go create mode 100644 pkg/state/tests/array_test.go create mode 100644 pkg/state/tests/bench.go create mode 100644 pkg/state/tests/bench_test.go create mode 100644 pkg/state/tests/bool_test.go create mode 100644 pkg/state/tests/float_test.go create mode 100644 pkg/state/tests/integer.go create mode 100644 pkg/state/tests/integer_test.go create mode 100644 pkg/state/tests/load.go create mode 100644 pkg/state/tests/load_test.go create mode 100644 pkg/state/tests/map.go create mode 100644 pkg/state/tests/map_test.go create mode 100644 pkg/state/tests/register.go create mode 100644 pkg/state/tests/register_test.go create mode 100644 pkg/state/tests/string_test.go create mode 100644 pkg/state/tests/struct.go create mode 100644 pkg/state/tests/struct_test.go create mode 100644 pkg/state/tests/tests.go create mode 100644 pkg/state/types.go create mode 100644 pkg/state/wire/BUILD create mode 100644 pkg/state/wire/wire.go (limited to 'pkg/sentry') diff --git a/pkg/compressio/compressio.go b/pkg/compressio/compressio.go index 5f52cbe74..b094c5662 100644 --- a/pkg/compressio/compressio.go +++ b/pkg/compressio/compressio.go @@ -346,20 +346,22 @@ func (p *pool) schedule(c *chunk, callback func(*chunk) error) error { } } -// reader chunks reads and decompresses. -type reader struct { +// Reader is a compressed reader. +type Reader struct { pool // in is the source. in io.Reader } +var _ io.Reader = (*Reader)(nil) + // NewReader returns a new compressed reader. If key is non-nil, the data stream // is assumed to contain expected hash values, which will be compared against // hash values computed from the compressed bytes. See package comments for // details. -func NewReader(in io.Reader, key []byte) (io.Reader, error) { - r := &reader{ +func NewReader(in io.Reader, key []byte) (*Reader, error) { + r := &Reader{ in: in, } @@ -394,8 +396,19 @@ var errNewBuffer = errors.New("buffer ready") // ErrHashMismatch is returned if the hash does not match. var ErrHashMismatch = errors.New("hash mismatch") +// ReadByte implements wire.Reader.ReadByte. +func (r *Reader) ReadByte() (byte, error) { + var p [1]byte + n, err := r.Read(p[:]) + if n != 1 { + return p[0], err + } + // Suppress EOF. + return p[0], nil +} + // Read implements io.Reader.Read. -func (r *reader) Read(p []byte) (int, error) { +func (r *Reader) Read(p []byte) (int, error) { r.mu.Lock() defer r.mu.Unlock() @@ -551,8 +564,8 @@ func (r *reader) Read(p []byte) (int, error) { return done, nil } -// writer chunks and schedules writes. -type writer struct { +// Writer is a compressed writer. +type Writer struct { pool // out is the underlying writer. @@ -562,6 +575,8 @@ type writer struct { closed bool } +var _ io.Writer = (*Writer)(nil) + // NewWriter returns a new compressed writer. If key is non-nil, hash values are // generated and written out for compressed bytes. See package comments for // details. @@ -569,8 +584,8 @@ type writer struct { // The recommended chunkSize is on the order of 1M. Extra memory may be // buffered (in the form of read-ahead, or buffered writes), and is limited to // O(chunkSize * [1+GOMAXPROCS]). -func NewWriter(out io.Writer, key []byte, chunkSize uint32, level int) (io.WriteCloser, error) { - w := &writer{ +func NewWriter(out io.Writer, key []byte, chunkSize uint32, level int) (*Writer, error) { + w := &Writer{ pool: pool{ chunkSize: chunkSize, buf: bufPool.Get().(*bytes.Buffer), @@ -597,7 +612,7 @@ func NewWriter(out io.Writer, key []byte, chunkSize uint32, level int) (io.Write } // flush writes a single buffer. -func (w *writer) flush(c *chunk) error { +func (w *Writer) flush(c *chunk) error { // Prefix each chunk with a length; this allows the reader to safely // limit reads while buffering. l := uint32(c.compressed.Len()) @@ -624,8 +639,23 @@ func (w *writer) flush(c *chunk) error { return nil } +// WriteByte implements wire.Writer.WriteByte. +// +// Note that this implementation is necessary on the object itself, as an +// interface-based dispatch cannot tell whether the array backing the slice +// escapes, therefore the all bytes written will generate an escape. +func (w *Writer) WriteByte(b byte) error { + var p [1]byte + p[0] = b + n, err := w.Write(p[:]) + if n != 1 { + return err + } + return nil +} + // Write implements io.Writer.Write. -func (w *writer) Write(p []byte) (int, error) { +func (w *Writer) Write(p []byte) (int, error) { w.mu.Lock() defer w.mu.Unlock() @@ -710,7 +740,7 @@ func (w *writer) Write(p []byte) (int, error) { } // Close implements io.Closer.Close. -func (w *writer) Close() error { +func (w *Writer) Close() error { w.mu.Lock() defer w.mu.Unlock() diff --git a/pkg/gohacks/BUILD b/pkg/gohacks/BUILD index 798a65eca..35683fe98 100644 --- a/pkg/gohacks/BUILD +++ b/pkg/gohacks/BUILD @@ -7,5 +7,6 @@ go_library( srcs = [ "gohacks_unsafe.go", ], + stateify = False, visibility = ["//:sandbox"], ) diff --git a/pkg/ilist/list.go b/pkg/ilist/list.go index 0d07da3b1..f4a4c33d3 100644 --- a/pkg/ilist/list.go +++ b/pkg/ilist/list.go @@ -90,7 +90,7 @@ func (l *List) Back() Element { // // NOTE: This is an O(n) operation. func (l *List) Len() (count int) { - for e := l.Front(); e != nil; e = e.Next() { + for e := l.Front(); e != nil; e = (ElementMapper{}.linkerFor(e)).Next() { count++ } return count @@ -182,13 +182,13 @@ func (l *List) Remove(e Element) { if prev != nil { ElementMapper{}.linkerFor(prev).SetNext(next) - } else { + } else if l.head == e { l.head = next } if next != nil { ElementMapper{}.linkerFor(next).SetPrev(prev) - } else { + } else if l.tail == e { l.tail = prev } diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD index 1510a7c26..25fe1921b 100644 --- a/pkg/sentry/kernel/BUILD +++ b/pkg/sentry/kernel/BUILD @@ -200,6 +200,7 @@ go_library( "//pkg/sentry/vfs", "//pkg/state", "//pkg/state/statefile", + "//pkg/state/wire", "//pkg/sync", "//pkg/syserr", "//pkg/syserror", diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 554a42e05..2177b785a 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -34,7 +34,6 @@ package kernel import ( "errors" "fmt" - "io" "path/filepath" "sync/atomic" "time" @@ -73,6 +72,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/uniqueid" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/state" + "gvisor.dev/gvisor/pkg/state/wire" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" ) @@ -417,7 +417,7 @@ func (k *Kernel) Init(args InitKernelArgs) error { // SaveTo saves the state of k to w. // // Preconditions: The kernel must be paused throughout the call to SaveTo. -func (k *Kernel) SaveTo(w io.Writer) error { +func (k *Kernel) SaveTo(w wire.Writer) error { saveStart := time.Now() ctx := k.SupervisorContext() @@ -473,18 +473,18 @@ func (k *Kernel) SaveTo(w io.Writer) error { // // N.B. This will also be saved along with the full kernel save below. cpuidStart := time.Now() - if err := state.Save(k.SupervisorContext(), w, k.FeatureSet(), nil); err != nil { + if _, err := state.Save(k.SupervisorContext(), w, k.FeatureSet()); err != nil { return err } log.Infof("CPUID save took [%s].", time.Since(cpuidStart)) // Save the kernel state. kernelStart := time.Now() - var stats state.Stats - if err := state.Save(k.SupervisorContext(), w, k, &stats); err != nil { + stats, err := state.Save(k.SupervisorContext(), w, k) + if err != nil { return err } - log.Infof("Kernel save stats: %s", &stats) + log.Infof("Kernel save stats: %s", stats.String()) log.Infof("Kernel save took [%s].", time.Since(kernelStart)) // Save the memory file's state. @@ -629,7 +629,7 @@ func (ts *TaskSet) unregisterEpollWaiters() { } // LoadFrom returns a new Kernel loaded from args. -func (k *Kernel) LoadFrom(r io.Reader, net inet.Stack, clocks sentrytime.Clocks) error { +func (k *Kernel) LoadFrom(r wire.Reader, net inet.Stack, clocks sentrytime.Clocks) error { loadStart := time.Now() initAppCores := k.applicationCores @@ -640,7 +640,7 @@ func (k *Kernel) LoadFrom(r io.Reader, net inet.Stack, clocks sentrytime.Clocks) // don't need to explicitly install it in the Kernel. cpuidStart := time.Now() var features cpuid.FeatureSet - if err := state.Load(k.SupervisorContext(), r, &features, nil); err != nil { + if _, err := state.Load(k.SupervisorContext(), r, &features); err != nil { return err } log.Infof("CPUID load took [%s].", time.Since(cpuidStart)) @@ -655,11 +655,11 @@ func (k *Kernel) LoadFrom(r io.Reader, net inet.Stack, clocks sentrytime.Clocks) // Load the kernel state. kernelStart := time.Now() - var stats state.Stats - if err := state.Load(k.SupervisorContext(), r, k, &stats); err != nil { + stats, err := state.Load(k.SupervisorContext(), r, k) + if err != nil { return err } - log.Infof("Kernel load stats: %s", &stats) + log.Infof("Kernel load stats: %s", stats.String()) log.Infof("Kernel load took [%s].", time.Since(kernelStart)) // rootNetworkNamespace should be populated after loading the state file. diff --git a/pkg/sentry/pgalloc/BUILD b/pkg/sentry/pgalloc/BUILD index a9836ba71..e1fcb175f 100644 --- a/pkg/sentry/pgalloc/BUILD +++ b/pkg/sentry/pgalloc/BUILD @@ -92,6 +92,7 @@ go_library( "//pkg/sentry/platform", "//pkg/sentry/usage", "//pkg/state", + "//pkg/state/wire", "//pkg/sync", "//pkg/syserror", "//pkg/usermem", diff --git a/pkg/sentry/pgalloc/save_restore.go b/pkg/sentry/pgalloc/save_restore.go index f8385c146..78317fa35 100644 --- a/pkg/sentry/pgalloc/save_restore.go +++ b/pkg/sentry/pgalloc/save_restore.go @@ -26,11 +26,12 @@ import ( "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/state" + "gvisor.dev/gvisor/pkg/state/wire" "gvisor.dev/gvisor/pkg/usermem" ) // SaveTo writes f's state to the given stream. -func (f *MemoryFile) SaveTo(ctx context.Context, w io.Writer) error { +func (f *MemoryFile) SaveTo(ctx context.Context, w wire.Writer) error { // Wait for reclaim. f.mu.Lock() defer f.mu.Unlock() @@ -79,10 +80,10 @@ func (f *MemoryFile) SaveTo(ctx context.Context, w io.Writer) error { } // Save metadata. - if err := state.Save(ctx, w, &f.fileSize, nil); err != nil { + if _, err := state.Save(ctx, w, &f.fileSize); err != nil { return err } - if err := state.Save(ctx, w, &f.usage, nil); err != nil { + if _, err := state.Save(ctx, w, &f.usage); err != nil { return err } @@ -115,9 +116,9 @@ func (f *MemoryFile) SaveTo(ctx context.Context, w io.Writer) error { } // LoadFrom loads MemoryFile state from the given stream. -func (f *MemoryFile) LoadFrom(ctx context.Context, r io.Reader) error { +func (f *MemoryFile) LoadFrom(ctx context.Context, r wire.Reader) error { // Load metadata. - if err := state.Load(ctx, r, &f.fileSize, nil); err != nil { + if _, err := state.Load(ctx, r, &f.fileSize); err != nil { return err } if err := f.file.Truncate(f.fileSize); err != nil { @@ -125,7 +126,7 @@ func (f *MemoryFile) LoadFrom(ctx context.Context, r io.Reader) error { } newMappings := make([]uintptr, f.fileSize>>chunkShift) f.mappings.Store(newMappings) - if err := state.Load(ctx, r, &f.usage, nil); err != nil { + if _, err := state.Load(ctx, r, &f.usage); err != nil { return err } diff --git a/pkg/state/BUILD b/pkg/state/BUILD index 2b1350135..089b3bbef 100644 --- a/pkg/state/BUILD +++ b/pkg/state/BUILD @@ -1,8 +1,46 @@ -load("//tools:defs.bzl", "go_library", "go_test", "proto_library") +load("//tools:defs.bzl", "go_library") load("//tools/go_generics:defs.bzl", "go_template_instance") package(licenses = ["notice"]) +go_template_instance( + name = "pending_list", + out = "pending_list.go", + package = "state", + prefix = "pending", + template = "//pkg/ilist:generic_list", + types = { + "Element": "*objectEncodeState", + "ElementMapper": "pendingMapper", + "Linker": "*pendingEntry", + }, +) + +go_template_instance( + name = "deferred_list", + out = "deferred_list.go", + package = "state", + prefix = "deferred", + template = "//pkg/ilist:generic_list", + types = { + "Element": "*objectEncodeState", + "ElementMapper": "deferredMapper", + "Linker": "*deferredEntry", + }, +) + +go_template_instance( + name = "complete_list", + out = "complete_list.go", + package = "state", + prefix = "complete", + template = "//pkg/ilist:generic_list", + types = { + "Element": "*objectDecodeState", + "Linker": "*objectDecodeState", + }, +) + go_template_instance( name = "addr_range", out = "addr_range.go", @@ -29,7 +67,7 @@ go_template_instance( types = { "Key": "uintptr", "Range": "addrRange", - "Value": "reflect.Value", + "Value": "*objectEncodeState", "Functions": "addrSetFunctions", }, ) @@ -39,32 +77,24 @@ go_library( srcs = [ "addr_range.go", "addr_set.go", + "complete_list.go", "decode.go", + "decode_unsafe.go", + "deferred_list.go", "encode.go", "encode_unsafe.go", - "map.go", - "printer.go", + "pending_list.go", "state.go", + "state_norace.go", + "state_race.go", "stats.go", + "types.go", ], marshal = False, stateify = False, visibility = ["//:sandbox"], deps = [ - ":object_go_proto", - "@com_github_golang_protobuf//proto:go_default_library", + "//pkg/log", + "//pkg/state/wire", ], ) - -proto_library( - name = "object", - srcs = ["object.proto"], - visibility = ["//:sandbox"], -) - -go_test( - name = "state_test", - timeout = "long", - srcs = ["state_test.go"], - library = ":state", -) diff --git a/pkg/state/README.md b/pkg/state/README.md new file mode 100644 index 000000000..1aa401193 --- /dev/null +++ b/pkg/state/README.md @@ -0,0 +1,158 @@ +# State Encoding and Decoding + +The state package implements the encoding and decoding of data structures for +`go_stateify`. This package is designed for use cases other than the standard +encoding packages, e.g. `gob` and `json`. Principally: + +* This package operates on complex object graphs and accurately serializes and + restores all relationships. That is, you can have things like: intrusive + pointers, cycles, and pointer chains of arbitrary depths. These are not + handled appropriately by existing encoders. This is not an implementation + flaw: the formats themselves are not capable of representing these graphs, + as they can only generate directed trees. + +* This package allows installing order-dependent load callbacks and then + resolves that graph at load time, with cycle detection. Similarly, there is + no analogous feature possible in the standard encoders. + +* This package handles the resolution of interfaces, based on a registered + type name. For interface objects type information is saved in the serialized + format. This is generally true for `gob` as well, but it works differently. + +Here's an overview of how encoding and decoding works. + +## Encoding + +Encoding produces a `statefile`, which contains a list of chunks of the form +`(header, payload)`. The payload can either be some raw data, or a series of +encoded wire objects representing some object graph. All encoded objects are +defined in the `wire` subpackage. + +Encoding of an object graph begins with `encodeState.Save`. + +### 1. Memory Map & Encoding + +To discover relationships between potentially interdependent data structures +(for example, a struct may contain pointers to members of other data +structures), the encoder first walks the object graph and constructs a memory +map of the objects in the input graph. As this walk progresses, objects are +queued in the `pending` list and items are placed on the `deferred` list as they +are discovered. No single object will be encoded multiple times, but the +discovered relationships between objects may change as more parts of the overall +object graph are discovered. + +The encoder starts at the root object and recursively visits all reachable +objects, recording the address ranges containing the underlying data for each +object. This is stored as a segment set (`addrSet`), mapping address ranges to +the of the object occupying the range; see `encodeState.values`. Note that there +is special handling for zero-sized types and map objects during this process. + +Additionally, the encoder assigns each object a unique identifier which is used +to indicate relationships between objects in the statefile; see `objectID` in +`encode.go`. + +### 2. Type Serialization + +The enoder will subsequently serialize all information about discovered types, +including field names. These are used during decoding to reconcile these types +with other internally registered types. + +### 3. Object Serialization + +With a full address map, and all objects correctly encoded, all object encodings +are serialized. The assigned `objectID`s aren't explicitly encoded in the +statefile. The order of object messages in the stream determine their IDs. + +### Example + +Given the following data structure definitions: + +```go +type system struct { + o *outer + i *inner +} + +type outer struct { + a int64 + cn *container +} + +type container struct { + n uint64 + elem *inner +} + +type inner struct { + c container + x, y uint64 +} +``` + +Initialized like this: + +```go +o := outer{ + a: 10, + cn: nil, +} +i := inner{ + x: 20, + y: 30, + c: container{}, +} +s := system{ + o: &o, + i: &i, +} + +o.cn = &i.c +o.cn.elem = &i + +``` + +Encoding will produce an object stream like this: + +``` +g0r1 = struct{ + i: g0r3, + o: g0r2, +} +g0r2 = struct{ + a: 10, + cn: g0r3.c, +} +g0r3 = struct{ + c: struct{ + elem: g0r3, + n: 0u, + }, + x: 20u, + y: 30u, +} +``` + +Note how `g0r3.c` is correctly encoded as the underlying `container` object for +`inner.c`, and how the pointer from `outer.cn` points to it, despite `system.i` +being discovered after the pointer to it in `system.o.cn`. Also note that +decoding isn't strictly reliant on the order of encoded object stream, as long +as the relationship between objects are correctly encoded. + +## Decoding + +Decoding reads the statefile and reconstructs the object graph. Decoding begins +in `decodeState.Load`. Decoding is performed in a single pass over the object +stream in the statefile, and a subsequent pass over all deserialized objects is +done to fire off all loading callbacks in the correctly defined order. Note that +introducing cycles is possible here, but these are detected and an error will be +returned. + +Decoding is relatively straight forward. For most primitive values, the decoder +constructs an appropriate object and fills it with the values encoded in the +statefile. Pointers need special handling, as they must point to a value +allocated elsewhere. When values are constructed, the decoder indexes them by +their `objectID`s in `decodeState.objectsByID`. The target of pointers are +resolved by searching for the target in this index by their `objectID`; see +`decodeState.register`. For pointers to values inside another value (fields in a +pointer, elements of an array), the decoder uses the accessor path to walk to +the appropriate location; see `walkChild`. diff --git a/pkg/state/decode.go b/pkg/state/decode.go index 590c241a3..c9971cdf6 100644 --- a/pkg/state/decode.go +++ b/pkg/state/decode.go @@ -17,28 +17,49 @@ package state import ( "bytes" "context" - "encoding/binary" - "errors" "fmt" - "io" + "math" "reflect" - "sort" - "github.com/golang/protobuf/proto" - pb "gvisor.dev/gvisor/pkg/state/object_go_proto" + "gvisor.dev/gvisor/pkg/state/wire" ) -// objectState represents an object that may be in the process of being +// internalCallback is a interface called on object completion. +// +// There are two implementations: objectDecodeState & userCallback. +type internalCallback interface { + // source returns the dependent object. May be nil. + source() *objectDecodeState + + // callbackRun executes the callback. + callbackRun() +} + +// userCallback is an implementation of internalCallback. +type userCallback func() + +// source implements internalCallback.source. +func (userCallback) source() *objectDecodeState { + return nil +} + +// callbackRun implements internalCallback.callbackRun. +func (uc userCallback) callbackRun() { + uc() +} + +// objectDecodeState represents an object that may be in the process of being // decoded. Specifically, it represents either a decoded object, or an an // interest in a future object that will be decoded. When that interest is // registered (via register), the storage for the object will be created, but // it will not be decoded until the object is encountered in the stream. -type objectState struct { +type objectDecodeState struct { // id is the id for this object. - // - // If this field is zero, then this is an anonymous (unregistered, - // non-reference primitive) object. This is immutable. - id uint64 + id objectID + + // typ is the id for this typeID. This may be zero if this is not a + // type-registered structure. + typ typeID // obj is the object. This may or may not be valid yet, depending on // whether complete returns true. However, regardless of whether the @@ -57,69 +78,52 @@ type objectState struct { // blockedBy is the number of dependencies this object has. blockedBy int - // blocking is a list of the objects blocked by this one. - blocking []*objectState + // callbacksInline is inline storage for callbacks. + callbacksInline [2]internalCallback // callbacks is a set of callbacks to execute on load. - callbacks []func() - - // path is the decoding path to the object. - path recoverable -} - -// complete indicates the object is complete. -func (os *objectState) complete() bool { - return os.blockedBy == 0 && len(os.callbacks) == 0 -} - -// checkComplete checks for completion. If the object is complete, pending -// callbacks will be executed and checkComplete will be called on downstream -// objects (those depending on this one). -func (os *objectState) checkComplete(stats *Stats) { - if os.blockedBy > 0 { - return - } - stats.Start(os.obj) + callbacks []internalCallback - // Fire all callbacks. - for _, fn := range os.callbacks { - fn() - } - os.callbacks = nil - - // Clear all blocked objects. - for _, other := range os.blocking { - other.blockedBy-- - other.checkComplete(stats) - } - os.blocking = nil - stats.Done() + completeEntry } -// waitFor queues a dependency on the given object. -func (os *objectState) waitFor(other *objectState, callback func()) { - os.blockedBy++ - other.blocking = append(other.blocking, os) - if callback != nil { - other.callbacks = append(other.callbacks, callback) +// addCallback adds a callback to the objectDecodeState. +func (ods *objectDecodeState) addCallback(ic internalCallback) { + if ods.callbacks == nil { + ods.callbacks = ods.callbacksInline[:0] } + ods.callbacks = append(ods.callbacks, ic) } // findCycleFor returns when the given object is found in the blocking set. -func (os *objectState) findCycleFor(target *objectState) []*objectState { - for _, other := range os.blocking { - if other == target { - return []*objectState{target} +func (ods *objectDecodeState) findCycleFor(target *objectDecodeState) []*objectDecodeState { + for _, ic := range ods.callbacks { + other := ic.source() + if other != nil && other == target { + return []*objectDecodeState{target} } else if childList := other.findCycleFor(target); childList != nil { return append(childList, other) } } - return nil + + // This should not occur. + Failf("no deadlock found?") + panic("unreachable") } // findCycle finds a dependency cycle. -func (os *objectState) findCycle() []*objectState { - return append(os.findCycleFor(os), os) +func (ods *objectDecodeState) findCycle() []*objectDecodeState { + return append(ods.findCycleFor(ods), ods) +} + +// source implements internalCallback.source. +func (ods *objectDecodeState) source() *objectDecodeState { + return ods +} + +// callbackRun implements internalCallback.callbackRun. +func (ods *objectDecodeState) callbackRun() { + ods.blockedBy-- } // decodeState is a graph of objects in the process of being decoded. @@ -137,30 +141,66 @@ type decodeState struct { // ctx is the decode context. ctx context.Context + // r is the input stream. + r wire.Reader + + // types is the type database. + types typeDecodeDatabase + // objectByID is the set of objects in progress. - objectsByID map[uint64]*objectState + objectsByID []*objectDecodeState // deferred are objects that have been read, by no interest has been // registered yet. These will be decoded once interest in registered. - deferred map[uint64]*pb.Object + deferred map[objectID]wire.Object - // outstanding is the number of outstanding objects. - outstanding uint32 + // pending is the set of objects that are not yet complete. + pending completeList - // r is the input stream. - r io.Reader - - // stats is the passed stats object. - stats *Stats - - // recoverable is the panic recover facility. - recoverable + // stats tracks time data. + stats Stats } // lookup looks up an object in decodeState or returns nil if no such object // has been previously registered. -func (ds *decodeState) lookup(id uint64) *objectState { - return ds.objectsByID[id] +func (ds *decodeState) lookup(id objectID) *objectDecodeState { + if len(ds.objectsByID) < int(id) { + return nil + } + return ds.objectsByID[id-1] +} + +// checkComplete checks for completion. +func (ds *decodeState) checkComplete(ods *objectDecodeState) bool { + // Still blocked? + if ods.blockedBy > 0 { + return false + } + + // Track stats if relevant. + if ods.callbacks != nil && ods.typ != 0 { + ds.stats.start(ods.typ) + defer ds.stats.done() + } + + // Fire all callbacks. + for _, ic := range ods.callbacks { + ic.callbackRun() + } + + // Mark completed. + cbs := ods.callbacks + ods.callbacks = nil + ds.pending.Remove(ods) + + // Recursively check others. + for _, ic := range cbs { + if other := ic.source(); other != nil && other.blockedBy == 0 { + ds.checkComplete(other) + } + } + + return true // All set. } // wait registers a dependency on an object. @@ -168,11 +208,8 @@ func (ds *decodeState) lookup(id uint64) *objectState { // As a special case, we always allow _useable_ references back to the first // decoding object because it may have fields that are already decoded. We also // allow trivial self reference, since they can be handled internally. -func (ds *decodeState) wait(waiter *objectState, id uint64, callback func()) { +func (ds *decodeState) wait(waiter *objectDecodeState, id objectID, callback func()) { switch id { - case 0: - // Nil pointer; nothing to wait for. - fallthrough case waiter.id: // Trivial self reference. fallthrough @@ -184,107 +221,188 @@ func (ds *decodeState) wait(waiter *objectState, id uint64, callback func()) { return } + // Mark as blocked. + waiter.blockedBy++ + // No nil can be returned here. - waiter.waitFor(ds.lookup(id), callback) + other := ds.lookup(id) + if callback != nil { + // Add the additional user callback. + other.addCallback(userCallback(callback)) + } + + // Mark waiter as unblocked. + other.addCallback(waiter) } // waitObject notes a blocking relationship. -func (ds *decodeState) waitObject(os *objectState, p *pb.Object, callback func()) { - if rv, ok := p.Value.(*pb.Object_RefValue); ok { +func (ds *decodeState) waitObject(ods *objectDecodeState, encoded wire.Object, callback func()) { + if rv, ok := encoded.(*wire.Ref); ok && rv.Root != 0 { // Refs can encode pointers and maps. - ds.wait(os, rv.RefValue, callback) - } else if sv, ok := p.Value.(*pb.Object_SliceValue); ok { + ds.wait(ods, objectID(rv.Root), callback) + } else if sv, ok := encoded.(*wire.Slice); ok && sv.Ref.Root != 0 { // See decodeObject; we need to wait for the array (if non-nil). - ds.wait(os, sv.SliceValue.RefValue, callback) - } else if iv, ok := p.Value.(*pb.Object_InterfaceValue); ok { + ds.wait(ods, objectID(sv.Ref.Root), callback) + } else if iv, ok := encoded.(*wire.Interface); ok { // It's an interface (wait recurisvely). - ds.waitObject(os, iv.InterfaceValue.Value, callback) + ds.waitObject(ods, iv.Value, callback) } else if callback != nil { // Nothing to wait for: execute the callback immediately. callback() } } +// walkChild returns a child object from obj, given an accessor path. This is +// the decode-side equivalent to traverse in encode.go. +// +// For the purposes of this function, a child object is either a field within a +// struct or an array element, with one such indirection per element in +// path. The returned value may be an unexported field, so it may not be +// directly assignable. See unsafePointerTo. +func walkChild(path []wire.Dot, obj reflect.Value) reflect.Value { + // See wire.Ref.Dots. The path here is specified in reverse order. + for i := len(path) - 1; i >= 0; i-- { + switch pc := path[i].(type) { + case *wire.FieldName: // Must be a pointer. + if obj.Kind() != reflect.Struct { + Failf("next component in child path is a field name, but the current object is not a struct. Path: %v, current obj: %#v", path, obj) + } + obj = obj.FieldByName(string(*pc)) + case wire.Index: // Embedded. + if obj.Kind() != reflect.Array { + Failf("next component in child path is an array index, but the current object is not an array. Path: %v, current obj: %#v", path, obj) + } + obj = obj.Index(int(pc)) + default: + panic("unreachable: switch should be exhaustive") + } + } + return obj +} + // register registers a decode with a type. // // This type is only used to instantiate a new object if it has not been -// registered previously. -func (ds *decodeState) register(id uint64, typ reflect.Type) *objectState { - os, ok := ds.objectsByID[id] - if ok { - return os +// registered previously. This depends on the type provided if none is +// available in the object itself. +func (ds *decodeState) register(r *wire.Ref, typ reflect.Type) reflect.Value { + // Grow the objectsByID slice. + id := objectID(r.Root) + if len(ds.objectsByID) < int(id) { + ds.objectsByID = append(ds.objectsByID, make([]*objectDecodeState, int(id)-len(ds.objectsByID))...) + } + + // Does this object already exist? + ods := ds.objectsByID[id-1] + if ods != nil { + return walkChild(r.Dots, ods.obj) + } + + // Create the object. + if len(r.Dots) != 0 { + typ = ds.findType(r.Type) } + v := reflect.New(typ) + ods = &objectDecodeState{ + id: id, + obj: v.Elem(), + } + ds.objectsByID[id-1] = ods + ds.pending.PushBack(ods) - // Record in the object index. - if typ.Kind() == reflect.Map { - os = &objectState{id: id, obj: reflect.MakeMap(typ), path: ds.recoverable.copy()} - } else { - os = &objectState{id: id, obj: reflect.New(typ).Elem(), path: ds.recoverable.copy()} + // Process any deferred objects & callbacks. + if encoded, ok := ds.deferred[id]; ok { + delete(ds.deferred, id) + ds.decodeObject(ods, ods.obj, encoded) } - ds.objectsByID[id] = os - if o, ok := ds.deferred[id]; ok { - // There is a deferred object. - delete(ds.deferred, id) // Free memory. - ds.decodeObject(os, os.obj, o, "", nil) - } else { - // There is no deferred object. - ds.outstanding++ + return walkChild(r.Dots, ods.obj) +} + +// objectDecoder is for decoding structs. +type objectDecoder struct { + // ds is decodeState. + ds *decodeState + + // ods is current object being decoded. + ods *objectDecodeState + + // reconciledTypeEntry is the reconciled type information. + rte *reconciledTypeEntry + + // encoded is the encoded object state. + encoded *wire.Struct +} + +// load is helper for the public methods on Source. +func (od *objectDecoder) load(slot int, objPtr reflect.Value, wait bool, fn func()) { + // Note that we have reconciled the type and may remap the fields here + // to match what's expected by the decoder. The "slot" parameter here + // is in terms of the local type, where the fields in the encoded + // object are in terms of the wire object's type, which might be in a + // different order (but will have the same fields). + v := *od.encoded.Field(od.rte.FieldOrder[slot]) + od.ds.decodeObject(od.ods, objPtr.Elem(), v) + if wait { + // Mark this individual object a blocker. + od.ds.waitObject(od.ods, v, fn) } +} - return os +// aterLoad implements Source.AfterLoad. +func (od *objectDecoder) afterLoad(fn func()) { + // Queue the local callback; this will execute when all of the above + // data dependencies have been cleared. + od.ods.addCallback(userCallback(fn)) } // decodeStruct decodes a struct value. -func (ds *decodeState) decodeStruct(os *objectState, obj reflect.Value, s *pb.Struct) { - // Set the fields. - m := Map{newInternalMap(nil, ds, os)} - defer internalMapPool.Put(m.internalMap) - for _, field := range s.Fields { - m.data = append(m.data, entry{ - name: field.Name, - object: field.Value, - }) - } - - // Sort the fields for efficient searching. - // - // Technically, these should already appear in sorted order in the - // state ordering, so this cost is effectively a single scan to ensure - // that the order is correct. - if len(m.data) > 1 { - sort.Slice(m.data, func(i, j int) bool { - return m.data[i].name < m.data[j].name - }) - } - - // Invoke the load; this will recursively decode other objects. - fns, ok := registeredTypes.lookupFns(obj.Addr().Type()) - if ok { - // Invoke the loader. - fns.invokeLoad(obj.Addr(), m) - } else if obj.NumField() == 0 { - // Allow anonymous empty structs. - return - } else { +func (ds *decodeState) decodeStruct(ods *objectDecodeState, obj reflect.Value, encoded *wire.Struct) { + if encoded.TypeID == 0 { + // Allow anonymous empty structs, but only if the encoded + // object also has no fields. + if encoded.Fields() == 0 && obj.NumField() == 0 { + return + } + // Propagate an error. - panic(fmt.Errorf("unregistered type %s", obj.Type())) + Failf("empty struct on wire %#v has field mismatch with type %q", encoded, obj.Type().Name()) + } + + // Lookup the object type. + rte := ds.types.Lookup(typeID(encoded.TypeID), obj.Type()) + ods.typ = typeID(encoded.TypeID) + + // Invoke the loader. + od := objectDecoder{ + ds: ds, + ods: ods, + rte: rte, + encoded: encoded, + } + ds.stats.start(ods.typ) + defer ds.stats.done() + if sl, ok := obj.Addr().Interface().(SaverLoader); ok { + // Note: may be a registered empty struct which does not + // implement the saver/loader interfaces. + sl.StateLoad(Source{internal: od}) } } // decodeMap decodes a map value. -func (ds *decodeState) decodeMap(os *objectState, obj reflect.Value, m *pb.Map) { +func (ds *decodeState) decodeMap(ods *objectDecodeState, obj reflect.Value, encoded *wire.Map) { if obj.IsNil() { + // See pointerTo. obj.Set(reflect.MakeMap(obj.Type())) } - for i := 0; i < len(m.Keys); i++ { + for i := 0; i < len(encoded.Keys); i++ { // Decode the objects. kv := reflect.New(obj.Type().Key()).Elem() vv := reflect.New(obj.Type().Elem()).Elem() - ds.decodeObject(os, kv, m.Keys[i], ".(key %d)", i) - ds.decodeObject(os, vv, m.Values[i], "[%#v]", kv.Interface()) - ds.waitObject(os, m.Keys[i], nil) - ds.waitObject(os, m.Values[i], nil) + ds.decodeObject(ods, kv, encoded.Keys[i]) + ds.decodeObject(ods, vv, encoded.Values[i]) + ds.waitObject(ods, encoded.Keys[i], nil) + ds.waitObject(ods, encoded.Values[i], nil) // Set in the map. obj.SetMapIndex(kv, vv) @@ -292,271 +410,294 @@ func (ds *decodeState) decodeMap(os *objectState, obj reflect.Value, m *pb.Map) } // decodeArray decodes an array value. -func (ds *decodeState) decodeArray(os *objectState, obj reflect.Value, a *pb.Array) { - if len(a.Contents) != obj.Len() { - panic(fmt.Errorf("mismatching array length expect=%d, actual=%d", obj.Len(), len(a.Contents))) +func (ds *decodeState) decodeArray(ods *objectDecodeState, obj reflect.Value, encoded *wire.Array) { + if len(encoded.Contents) != obj.Len() { + Failf("mismatching array length expect=%d, actual=%d", obj.Len(), len(encoded.Contents)) } // Decode the contents into the array. - for i := 0; i < len(a.Contents); i++ { - ds.decodeObject(os, obj.Index(i), a.Contents[i], "[%d]", i) - ds.waitObject(os, a.Contents[i], nil) + for i := 0; i < len(encoded.Contents); i++ { + ds.decodeObject(ods, obj.Index(i), encoded.Contents[i]) + ds.waitObject(ods, encoded.Contents[i], nil) } } -// decodeInterface decodes an interface value. -func (ds *decodeState) decodeInterface(os *objectState, obj reflect.Value, i *pb.Interface) { - // Is this a nil value? - if i.Type == "" { - return // Just leave obj alone. +// findType finds the type for the given wire.TypeSpecs. +func (ds *decodeState) findType(t wire.TypeSpec) reflect.Type { + switch x := t.(type) { + case wire.TypeID: + typ := ds.types.LookupType(typeID(x)) + rte := ds.types.Lookup(typeID(x), typ) + return rte.LocalType + case *wire.TypeSpecPointer: + return reflect.PtrTo(ds.findType(x.Type)) + case *wire.TypeSpecArray: + return reflect.ArrayOf(int(x.Count), ds.findType(x.Type)) + case *wire.TypeSpecSlice: + return reflect.SliceOf(ds.findType(x.Type)) + case *wire.TypeSpecMap: + return reflect.MapOf(ds.findType(x.Key), ds.findType(x.Value)) + default: + // Should not happen. + Failf("unknown type %#v", t) } + panic("unreachable") +} - // Get the dispatchable type. This may not be used if the given - // reference has already been resolved, but if not we need to know the - // type to create. - t, ok := registeredTypes.lookupType(i.Type) - if !ok { - panic(fmt.Errorf("no valid type for %q", i.Type)) +// decodeInterface decodes an interface value. +func (ds *decodeState) decodeInterface(ods *objectDecodeState, obj reflect.Value, encoded *wire.Interface) { + if _, ok := encoded.Type.(wire.TypeSpecNil); ok { + // Special case; the nil object. Just decode directly, which + // will read nil from the wire (if encoded correctly). + ds.decodeObject(ods, obj, encoded.Value) + return } - if obj.Kind() != reflect.Map { - // Set the obj to be the given typed value; this actually sets - // obj to be a non-zero value -- namely, it inserts type - // information. There's no need to do this for maps. - obj.Set(reflect.Zero(t)) + // We now need to resolve the actual type. + typ := ds.findType(encoded.Type) + + // We need to imbue type information here, then we can proceed to + // decode normally. In order to avoid issues with setting value-types, + // we create a new non-interface version of this object. We will then + // set the interface object to be equal to whatever we decode. + origObj := obj + obj = reflect.New(typ).Elem() + defer origObj.Set(obj) + + // With the object now having sufficient type information to actually + // have Set called on it, we can proceed to decode the value. + ds.decodeObject(ods, obj, encoded.Value) +} + +// isFloatEq determines if x and y represent the same value. +func isFloatEq(x float64, y float64) bool { + switch { + case math.IsNaN(x): + return math.IsNaN(y) + case math.IsInf(x, 1): + return math.IsInf(y, 1) + case math.IsInf(x, -1): + return math.IsInf(y, -1) + default: + return x == y } +} - // Decode the dereferenced element; there is no need to wait here, as - // the interface object shares the current object state. - ds.decodeObject(os, obj, i.Value, ".(%s)", i.Type) +// isComplexEq determines if x and y represent the same value. +func isComplexEq(x complex128, y complex128) bool { + return isFloatEq(real(x), real(y)) && isFloatEq(imag(x), imag(y)) } // decodeObject decodes a object value. -func (ds *decodeState) decodeObject(os *objectState, obj reflect.Value, object *pb.Object, format string, param interface{}) { - ds.push(false, format, param) - ds.stats.Add(obj) - ds.stats.Start(obj) - - switch x := object.GetValue().(type) { - case *pb.Object_BoolValue: - obj.SetBool(x.BoolValue) - case *pb.Object_StringValue: - obj.SetString(string(x.StringValue)) - case *pb.Object_Int64Value: - obj.SetInt(x.Int64Value) - if obj.Int() != x.Int64Value { - panic(fmt.Errorf("signed integer truncated in %v for %s", object, obj.Type())) +func (ds *decodeState) decodeObject(ods *objectDecodeState, obj reflect.Value, encoded wire.Object) { + switch x := encoded.(type) { + case wire.Nil: // Fast path: first. + // We leave obj alone here. That's because if obj represents an + // interface, it may have been imbued with type information in + // decodeInterface, and we don't want to destroy that. + case *wire.Ref: + // Nil pointers may be encoded in a "forceValue" context. For + // those we just leave it alone as the value will already be + // correct (nil). + if id := objectID(x.Root); id == 0 { + return } - case *pb.Object_Uint64Value: - obj.SetUint(x.Uint64Value) - if obj.Uint() != x.Uint64Value { - panic(fmt.Errorf("unsigned integer truncated in %v for %s", object, obj.Type())) - } - case *pb.Object_DoubleValue: - obj.SetFloat(x.DoubleValue) - if obj.Float() != x.DoubleValue { - panic(fmt.Errorf("float truncated in %v for %s", object, obj.Type())) - } - case *pb.Object_RefValue: - // Resolve the pointer itself, even though the object may not - // be decoded yet. You need to use wait() in order to ensure - // that is the case. See wait above, and Map.Barrier. - if id := x.RefValue; id != 0 { - // Decoding the interface should have imparted type - // information, so from this point it's safe to resolve - // and use this dynamic information for actually - // creating the object in register. - // - // (For non-interfaces this is a no-op). - dyntyp := reflect.TypeOf(obj.Interface()) - if dyntyp.Kind() == reflect.Map { - // Remove the map object count here to avoid - // double counting, as this object will be - // counted again when it gets processed later. - // We do not add a reference count as the - // reference is artificial. - ds.stats.Remove(obj) - obj.Set(ds.register(id, dyntyp).obj) - } else if dyntyp.Kind() == reflect.Ptr { - ds.push(true /* dereference */, "", nil) - obj.Set(ds.register(id, dyntyp.Elem()).obj.Addr()) - ds.pop() - } else { - obj.Set(ds.register(id, dyntyp.Elem()).obj.Addr()) + + // Note that if this is a map type, we go through a level of + // indirection to allow for map aliasing. + if obj.Kind() == reflect.Map { + v := ds.register(x, obj.Type()) + if v.IsNil() { + // Note that we don't want to clobber the map + // if has already been decoded by decodeMap. We + // just make it so that we have a consistent + // reference when that eventually does happen. + v.Set(reflect.MakeMap(v.Type())) } - } else { - // We leave obj alone here. That's because if obj - // represents an interface, it may have been embued - // with type information in decodeInterface, and we - // don't want to destroy that information. + obj.Set(v) + return } - case *pb.Object_SliceValue: - // It's okay to slice the array here, since the contents will - // still be provided later on. These semantics are a bit - // strange but they are handled in the Map.Barrier properly. - // - // The special semantics of zero ref apply here too. - if id := x.SliceValue.RefValue; id != 0 && x.SliceValue.Capacity > 0 { - v := reflect.ArrayOf(int(x.SliceValue.Capacity), obj.Type().Elem()) - obj.Set(ds.register(id, v).obj.Slice3(0, int(x.SliceValue.Length), int(x.SliceValue.Capacity))) + + // Normal assignment: authoritative only if no dots. + v := ds.register(x, obj.Type().Elem()) + if v.IsValid() { + obj.Set(unsafePointerTo(v)) } - case *pb.Object_ArrayValue: - ds.decodeArray(os, obj, x.ArrayValue) - case *pb.Object_StructValue: - ds.decodeStruct(os, obj, x.StructValue) - case *pb.Object_MapValue: - ds.decodeMap(os, obj, x.MapValue) - case *pb.Object_InterfaceValue: - ds.decodeInterface(os, obj, x.InterfaceValue) - case *pb.Object_ByteArrayValue: - copyArray(obj, reflect.ValueOf(x.ByteArrayValue)) - case *pb.Object_Uint16ArrayValue: - // 16-bit slices are serialized as 32-bit slices. - // See object.proto for details. - s := x.Uint16ArrayValue.Values - t := obj.Slice(0, obj.Len()).Interface().([]uint16) - if len(t) != len(s) { - panic(fmt.Errorf("mismatching array length expect=%d, actual=%d", len(t), len(s))) + case wire.Bool: + obj.SetBool(bool(x)) + case wire.Int: + obj.SetInt(int64(x)) + if obj.Int() != int64(x) { + Failf("signed integer truncated from %v to %v", int64(x), obj.Int()) } - for i := range s { - t[i] = uint16(s[i]) + case wire.Uint: + obj.SetUint(uint64(x)) + if obj.Uint() != uint64(x) { + Failf("unsigned integer truncated from %v to %v", uint64(x), obj.Uint()) } - case *pb.Object_Uint32ArrayValue: - copyArray(obj, reflect.ValueOf(x.Uint32ArrayValue.Values)) - case *pb.Object_Uint64ArrayValue: - copyArray(obj, reflect.ValueOf(x.Uint64ArrayValue.Values)) - case *pb.Object_UintptrArrayValue: - copyArray(obj, castSlice(reflect.ValueOf(x.UintptrArrayValue.Values), reflect.TypeOf(uintptr(0)))) - case *pb.Object_Int8ArrayValue: - copyArray(obj, castSlice(reflect.ValueOf(x.Int8ArrayValue.Values), reflect.TypeOf(int8(0)))) - case *pb.Object_Int16ArrayValue: - // 16-bit slices are serialized as 32-bit slices. - // See object.proto for details. - s := x.Int16ArrayValue.Values - t := obj.Slice(0, obj.Len()).Interface().([]int16) - if len(t) != len(s) { - panic(fmt.Errorf("mismatching array length expect=%d, actual=%d", len(t), len(s))) + case wire.Float32: + obj.SetFloat(float64(x)) + case wire.Float64: + obj.SetFloat(float64(x)) + if !isFloatEq(obj.Float(), float64(x)) { + Failf("floating point number truncated from %v to %v", float64(x), obj.Float()) } - for i := range s { - t[i] = int16(s[i]) + case *wire.Complex64: + obj.SetComplex(complex128(*x)) + case *wire.Complex128: + obj.SetComplex(complex128(*x)) + if !isComplexEq(obj.Complex(), complex128(*x)) { + Failf("complex number truncated from %v to %v", complex128(*x), obj.Complex()) } - case *pb.Object_Int32ArrayValue: - copyArray(obj, reflect.ValueOf(x.Int32ArrayValue.Values)) - case *pb.Object_Int64ArrayValue: - copyArray(obj, reflect.ValueOf(x.Int64ArrayValue.Values)) - case *pb.Object_BoolArrayValue: - copyArray(obj, reflect.ValueOf(x.BoolArrayValue.Values)) - case *pb.Object_Float64ArrayValue: - copyArray(obj, reflect.ValueOf(x.Float64ArrayValue.Values)) - case *pb.Object_Float32ArrayValue: - copyArray(obj, reflect.ValueOf(x.Float32ArrayValue.Values)) + case *wire.String: + obj.SetString(string(*x)) + case *wire.Slice: + // See *wire.Ref above; same applies. + if id := objectID(x.Ref.Root); id == 0 { + return + } + // Note that it's fine to slice the array here and assume that + // contents will still be filled in later on. + typ := reflect.ArrayOf(int(x.Capacity), obj.Type().Elem()) // The object type. + v := ds.register(&x.Ref, typ) + obj.Set(v.Slice3(0, int(x.Length), int(x.Capacity))) + case *wire.Array: + ds.decodeArray(ods, obj, x) + case *wire.Struct: + ds.decodeStruct(ods, obj, x) + case *wire.Map: + ds.decodeMap(ods, obj, x) + case *wire.Interface: + ds.decodeInterface(ods, obj, x) default: // Shoud not happen, not propagated as an error. - panic(fmt.Sprintf("unknown object %v for %s", object, obj.Type())) - } - - ds.stats.Done() - ds.pop() -} - -func copyArray(dest reflect.Value, src reflect.Value) { - if dest.Len() != src.Len() { - panic(fmt.Errorf("mismatching array length expect=%d, actual=%d", dest.Len(), src.Len())) + Failf("unknown object %#v for %q", encoded, obj.Type().Name()) } - reflect.Copy(dest, castSlice(src, dest.Type().Elem())) } -// Deserialize deserializes the object state. +// Load deserializes the object graph rooted at obj. // // This function may panic and should be run in safely(). -func (ds *decodeState) Deserialize(obj reflect.Value) { - ds.objectsByID[1] = &objectState{id: 1, obj: obj, path: ds.recoverable.copy()} - ds.outstanding = 1 // The root object. +func (ds *decodeState) Load(obj reflect.Value) { + ds.stats.init() + defer ds.stats.fini(func(id typeID) string { + return ds.types.LookupName(id) + }) + + // Create the root object. + ds.objectsByID = append(ds.objectsByID, &objectDecodeState{ + id: 1, + obj: obj, + }) + + // Read the number of objects. + lastID, object, err := ReadHeader(ds.r) + if err != nil { + Failf("header error: %w", err) + } + if !object { + Failf("object missing") + } + + // Decode all objects. + var ( + encoded wire.Object + ods *objectDecodeState + id = objectID(1) + tid = typeID(1) + ) + if err := safely(func() { + // Decode all objects in the stream. + // + // Note that the structure of this decoding loop should match + // the raw decoding loop in printer.go. + for id <= objectID(lastID) { + // Unmarshal the object. + encoded = wire.Load(ds.r) + + // Is this a type object? Handle inline. + if wt, ok := encoded.(*wire.Type); ok { + ds.types.Register(wt) + tid++ + encoded = nil + continue + } - // Decode all objects in the stream. - // - // See above, we never process objects while we have no outstanding - // interests (other than the very first object). - for id := uint64(1); ds.outstanding > 0; id++ { - os := ds.lookup(id) - ds.stats.Start(os.obj) - - o, err := ds.readObject() - if err != nil { - panic(err) - } + // Actually resolve the object. + ods = ds.lookup(id) + if ods != nil { + // Decode the object. + ds.decodeObject(ods, ods.obj, encoded) + } else { + // If an object hasn't had interest registered + // previously or isn't yet valid, we deferred + // decoding until interest is registered. + ds.deferred[id] = encoded + } - if os != nil { - // Decode the object. - ds.from = &os.path - ds.decodeObject(os, os.obj, o, "", nil) - ds.outstanding-- + // For error handling. + ods = nil + encoded = nil + id++ + } + }); err != nil { + // Include as much information as we can, taking into account + // the possible state transitions above. + if ods != nil { + Failf("error decoding object ID %d (%T) from %#v: %w", id, ods.obj.Interface(), encoded, err) + } else if encoded != nil { + Failf("lookup error decoding object ID %d from %#v: %w", id, encoded, err) } else { - // If an object hasn't had interest registered - // previously, we deferred decoding until interest is - // registered. - ds.deferred[id] = o + Failf("general decoding error: %w", err) } - - ds.stats.Done() - } - - // Check the zero-length header at the end. - length, object, err := ReadHeader(ds.r) - if err != nil { - panic(err) - } - if length != 0 { - panic(fmt.Sprintf("expected zero-length terminal, got %d", length)) - } - if object { - panic("expected non-object terminal") } // Check if we have any deferred objects. - if count := len(ds.deferred); count > 0 { - // Shoud not happen, not propagated as an error. - panic(fmt.Sprintf("still have %d deferred objects", count)) - } - - // Scan and fire all callbacks. - for _, os := range ds.objectsByID { - os.checkComplete(ds.stats) + for id, encoded := range ds.deferred { + // Shoud never happen, the graph was bogus. + Failf("still have deferred objects: one is ID %d, %#v", id, encoded) } - // Check if we have any remaining dependency cycles. - for _, os := range ds.objectsByID { - if !os.complete() { - // This must be the result of a dependency cycle. - cycle := os.findCycle() - var buf bytes.Buffer - buf.WriteString("dependency cycle: {") - for i, cycleOS := range cycle { - if i > 0 { - buf.WriteString(" => ") + // Scan and fire all callbacks. We iterate over the list of incomplete + // objects until all have been finished. We stop iterating if no + // objects become complete (there is a dependency cycle). + // + // Note that we iterate backwards here, because there will be a strong + // tendendcy for blocking relationships to go from earlier objects to + // later (deeper) objects in the graph. This will reduce the number of + // iterations required to finish all objects. + if err := safely(func() { + for ds.pending.Back() != nil { + thisCycle := false + for ods = ds.pending.Back(); ods != nil; { + if ds.checkComplete(ods) { + thisCycle = true + break } - buf.WriteString(fmt.Sprintf("%s", cycleOS.obj.Type())) + ods = ods.Prev() + } + if !thisCycle { + break } - buf.WriteString("}") - // Panic as an error; propagate to the caller. - panic(errors.New(string(buf.Bytes()))) } - } -} - -type byteReader struct { - io.Reader -} - -// ReadByte implements io.ByteReader. -func (br byteReader) ReadByte() (byte, error) { - var b [1]byte - n, err := br.Reader.Read(b[:]) - if n > 0 { - return b[0], nil - } else if err != nil { - return 0, err - } else { - return 0, io.ErrUnexpectedEOF + }); err != nil { + Failf("error executing callbacks for %#v: %w", ods.obj.Interface(), err) + } + + // Check if we have any remaining dependency cycles. If there are any + // objects left in the pending list, then it must be due to a cycle. + if ods := ds.pending.Front(); ods != nil { + // This must be the result of a dependency cycle. + cycle := ods.findCycle() + var buf bytes.Buffer + buf.WriteString("dependency cycle: {") + for i, cycleOS := range cycle { + if i > 0 { + buf.WriteString(" => ") + } + fmt.Fprintf(&buf, "%q", cycleOS.obj.Type()) + } + buf.WriteString("}") + Failf("incomplete graph: %s", string(buf.Bytes())) } } @@ -565,45 +706,20 @@ func (br byteReader) ReadByte() (byte, error) { // Each object written to the statefile is prefixed with a header. See // WriteHeader for more information; these functions are exported to allow // non-state writes to the file to play nice with debugging tools. -func ReadHeader(r io.Reader) (length uint64, object bool, err error) { +func ReadHeader(r wire.Reader) (length uint64, object bool, err error) { // Read the header. - length, err = binary.ReadUvarint(byteReader{r}) + err = safely(func() { + length = wire.LoadUint(r) + }) if err != nil { - return + // On the header, pass raw I/O errors. + if sErr, ok := err.(*ErrState); ok { + return 0, false, sErr.Unwrap() + } } // Decode whether the object is valid. - object = length&0x1 != 0 - length = length >> 1 + object = length&objectFlag != 0 + length &^= objectFlag return } - -// readObject reads an object from the stream. -func (ds *decodeState) readObject() (*pb.Object, error) { - // Read the header. - length, object, err := ReadHeader(ds.r) - if err != nil { - return nil, err - } - if !object { - return nil, fmt.Errorf("invalid object header") - } - - // Read the object. - buf := make([]byte, length) - for done := 0; done < len(buf); { - n, err := ds.r.Read(buf[done:]) - done += n - if n == 0 && err != nil { - return nil, err - } - } - - // Unmarshal. - obj := new(pb.Object) - if err := proto.Unmarshal(buf, obj); err != nil { - return nil, err - } - - return obj, nil -} diff --git a/pkg/state/decode_unsafe.go b/pkg/state/decode_unsafe.go new file mode 100644 index 000000000..d048f61a1 --- /dev/null +++ b/pkg/state/decode_unsafe.go @@ -0,0 +1,27 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package state + +import ( + "reflect" + "unsafe" +) + +// unsafePointerTo is logically equivalent to reflect.Value.Addr, but works on +// values representing unexported fields. This bypasses visibility, but not +// type safety. +func unsafePointerTo(obj reflect.Value) reflect.Value { + return reflect.NewAt(obj.Type(), unsafe.Pointer(obj.UnsafeAddr())) +} diff --git a/pkg/state/encode.go b/pkg/state/encode.go index c5118d3a9..92fcad4e9 100644 --- a/pkg/state/encode.go +++ b/pkg/state/encode.go @@ -15,437 +15,797 @@ package state import ( - "container/list" "context" - "encoding/binary" - "fmt" - "io" "reflect" - "sort" - "github.com/golang/protobuf/proto" - pb "gvisor.dev/gvisor/pkg/state/object_go_proto" + "gvisor.dev/gvisor/pkg/state/wire" ) -// queuedObject is an object queued for encoding. -type queuedObject struct { - id uint64 - obj reflect.Value - path recoverable +// objectEncodeState the type and identity of an object occupying a memory +// address range. This is the value type for addrSet, and the intrusive entry +// for the pending and deferred lists. +type objectEncodeState struct { + // id is the assigned ID for this object. + id objectID + + // obj is the object value. Note that this may be replaced if we + // encounter an object that contains this object. When this happens (in + // resolve), we will update existing references approprately, below, + // and defer a re-encoding of the object. + obj reflect.Value + + // encoded is the encoded value of this object. Note that this may not + // be up to date if this object is still in the deferred list. + encoded wire.Object + + // how indicates whether this object should be encoded as a value. This + // is used only for deferred encoding. + how encodeStrategy + + // refs are the list of reference objects used by other objects + // referring to this object. When the object is updated, these + // references may be updated directly and automatically. + refs []*wire.Ref + + pendingEntry + deferredEntry } // encodeState is state used for encoding. // -// The encoding process is a breadth-first traversal of the object graph. The -// inherent races and dependencies are much simpler than the decode case. +// The encoding process constructs a representation of the in-memory graph of +// objects before a single object is serialized. This is done to ensure that +// all references can be fully disambiguated. See resolve for more details. type encodeState struct { // ctx is the encode context. ctx context.Context - // lastID is the last object ID. - // - // See idsByObject for context. Because of the special zero encoding - // used for reference values, the first ID must be 1. - lastID uint64 + // w is the output stream. + w wire.Writer - // idsByObject is a set of objects, indexed via: - // - // reflect.ValueOf(x).UnsafeAddr - // - // This provides IDs for objects. - idsByObject map[uintptr]uint64 + // types is the type database. + types typeEncodeDatabase + + // lastID is the last allocated object ID. + lastID objectID - // values stores values that span the addresses. + // values tracks the address ranges occupied by objects, along with the + // types of these objects. This is used to locate pointer targets, + // including pointers to fields within another type. // - // addrSet is a a generated type which efficiently stores ranges of - // addresses. When encoding pointers, these ranges are filled in and - // used to check for overlapping or conflicting pointers. This would - // indicate a pointer to an field, or a non-type safe value, neither of - // which are currently decodable. + // Multiple objects may overlap in memory iff the larger object fully + // contains the smaller one, and the type of the smaller object matches + // a field or array element's type at the appropriate offset. An + // arbitrary number of objects may be nested in this manner. // - // See the usage of values below for more context. + // Note that this does not track zero-sized objects, those are tracked + // by zeroValues below. values addrSet - // w is the output stream. - w io.Writer + // zeroValues tracks zero-sized objects. + zeroValues map[reflect.Type]*objectEncodeState - // pending is the list of objects to be serialized. - // - // This is a set of queuedObjects. - pending list.List + // deferred is the list of objects to be encoded. + deferred deferredList - // done is the a list of finished objects. - // - // This is kept to prevent garbage collection and address reuse. - done list.List + // pendingTypes is the list of types to be serialized. Serialization + // will occur when all objects have been encoded, but before pending is + // serialized. + pendingTypes []wire.Type - // stats is the passed stats object. - stats *Stats + // pending is the list of objects to be serialized. Serialization does + // not actually occur until the full object graph is computed. + pending pendingList - // recoverable is the panic recover facility. - recoverable + // stats tracks time data. + stats Stats } -// register looks up an ID, registering if necessary. +// isSameSizeParent returns true if child is a field value or element within +// parent. Only a struct or array can have a child value. +// +// isSameSizeParent deals with objects like this: +// +// struct child { +// // fields.. +// } // -// If the object was not previously registered, it is enqueued to be serialized. -// See the documentation for idsByObject for more information. -func (es *encodeState) register(obj reflect.Value) uint64 { - // It is not legal to call register for any non-pointer objects (see - // below), so we panic with a recoverable error if this is a mismatch. - if obj.Kind() != reflect.Ptr && obj.Kind() != reflect.Map { - panic(fmt.Errorf("non-pointer %#v registered", obj.Interface())) +// struct parent { +// c child +// } +// +// var p parent +// record(&p.c) +// +// Here, &p and &p.c occupy the exact same address range. +// +// Or like this: +// +// struct child { +// // fields +// } +// +// var arr [1]parent +// record(&arr[0]) +// +// Similarly, &arr[0] and &arr[0].c have the exact same address range. +// +// Precondition: parent and child must occupy the same memory. +func isSameSizeParent(parent reflect.Value, childType reflect.Type) bool { + switch parent.Kind() { + case reflect.Struct: + for i := 0; i < parent.NumField(); i++ { + field := parent.Field(i) + if field.Type() == childType { + return true + } + // Recurse through any intermediate types. + if isSameSizeParent(field, childType) { + return true + } + // Does it make sense to keep going if the first field + // doesn't match? Yes, because there might be an + // arbitrary number of zero-sized fields before we get + // a match, and childType itself can be zero-sized. + } + return false + case reflect.Array: + // The only case where an array with more than one elements can + // return true is if childType is zero-sized. In such cases, + // it's ambiguous which element contains the match since a + // zero-sized child object fully fits in any of the zero-sized + // elements in an array... However since all elements are of + // the same type, we only need to check one element. + // + // For non-zero-sized childTypes, parent.Len() must be 1, but a + // combination of the precondition and an implicit comparison + // between the array element size and childType ensures this. + return parent.Len() > 0 && isSameSizeParent(parent.Index(0), childType) + default: + return false } +} - addr := obj.Pointer() - if obj.Kind() == reflect.Ptr && obj.Elem().Type().Size() == 0 { - // For zero-sized objects, we always provide a unique ID. - // That's because the runtime internally multiplexes pointers - // to the same address. We can't be certain what the intent is - // with pointers to zero-sized objects, so we just give them - // all unique identities. - } else if id, ok := es.idsByObject[addr]; ok { - // Already registered. - return id - } - - // Ensure that the first ID given out is one. See note on lastID. The - // ID zero is used to indicate nil values. +// nextID returns the next valid ID. +func (es *encodeState) nextID() objectID { es.lastID++ - id := es.lastID - es.idsByObject[addr] = id - if obj.Kind() == reflect.Ptr { - // Dereference and treat as a pointer. - es.pending.PushBack(queuedObject{id: id, obj: obj.Elem(), path: es.recoverable.copy()}) - - // Register this object at all addresses. - typ := obj.Elem().Type() - if size := typ.Size(); size > 0 { - r := addrRange{addr, addr + size} - if !es.values.IsEmptyRange(r) { - old := es.values.LowerBoundSegment(addr).Value().Interface().(recoverable) - panic(fmt.Errorf("overlapping objects: [new object] %#v [existing object path] %s", obj.Interface(), old.path())) + return objectID(es.lastID) +} + +// dummyAddr points to the dummy zero-sized address. +var dummyAddr = reflect.ValueOf(new(struct{})).Pointer() + +// resolve records the address range occupied by an object. +func (es *encodeState) resolve(obj reflect.Value, ref *wire.Ref) { + addr := obj.Pointer() + + // Is this a map pointer? Just record the single address. It is not + // possible to take any pointers into the map internals. + if obj.Kind() == reflect.Map { + if addr == 0 { + // Just leave the nil reference alone. This is fine, we + // may need to encode as a reference in this way. We + // return nil for our objectEncodeState so that anyone + // depending on this value knows there's nothing there. + return + } + if seg, _ := es.values.Find(addr); seg.Ok() { + // Ensure the map types match. + existing := seg.Value() + if existing.obj.Type() != obj.Type() { + Failf("overlapping map objects at 0x%x: [new object] %#v [existing object type] %s", addr, obj, existing.obj) } - es.values.Add(r, reflect.ValueOf(es.recoverable.copy())) + + // No sense recording refs, maps may not be replaced by + // covering objects, they are maximal. + ref.Root = wire.Uint(existing.id) + return } + + // Record the map. + oes := &objectEncodeState{ + id: es.nextID(), + obj: obj, + how: encodeMapAsValue, + } + es.values.Add(addrRange{addr, addr + 1}, oes) + es.pending.PushBack(oes) + es.deferred.PushBack(oes) + + // See above: no ref recording. + ref.Root = wire.Uint(oes.id) + return + } + + // If not a map, then the object must be a pointer. + if obj.Kind() != reflect.Ptr { + Failf("attempt to record non-map and non-pointer object %#v", obj) + } + + obj = obj.Elem() // Value from here. + + // Is this a zero-sized type? + typ := obj.Type() + size := typ.Size() + if size == 0 { + if addr == dummyAddr { + // Zero-sized objects point to a dummy byte within the + // runtime. There's no sense recording this in the + // address map. We add this to the dedicated + // zeroValues. + // + // Note that zero-sized objects must be *true* + // zero-sized objects. They cannot be part of some + // larger object. In that case, they are assigned a + // 1-byte address at the end of the object. + oes, ok := es.zeroValues[typ] + if !ok { + oes = &objectEncodeState{ + id: es.nextID(), + obj: obj, + } + es.zeroValues[typ] = oes + es.pending.PushBack(oes) + es.deferred.PushBack(oes) + } + + // There's also no sense tracking back references. We + // know that this is a true zero-sized object, and not + // part of a larger container, so it will not change. + ref.Root = wire.Uint(oes.id) + return + } + size = 1 // See above. + } + + // Calculate the container. + end := addr + size + r := addrRange{addr, end} + if seg, _ := es.values.Find(addr); seg.Ok() { + existing := seg.Value() + switch { + case seg.Start() == addr && seg.End() == end && obj.Type() == existing.obj.Type(): + // The object is a perfect match. Happy path. Avoid the + // traversal and just return directly. We don't need to + // encode the type information or any dots here. + ref.Root = wire.Uint(existing.id) + existing.refs = append(existing.refs, ref) + return + + case (seg.Start() < addr && seg.End() >= end) || (seg.Start() <= addr && seg.End() > end): + // The previously registered object is larger than + // this, no need to update. But we expect some + // traversal below. + + case seg.Start() == addr && seg.End() == end: + if !isSameSizeParent(obj, existing.obj.Type()) { + break // Needs traversal. + } + fallthrough // Needs update. + + case (seg.Start() > addr && seg.End() <= end) || (seg.Start() >= addr && seg.End() < end): + // Update the object and redo the encoding. + old := existing.obj + existing.obj = obj + es.deferred.Remove(existing) + es.deferred.PushBack(existing) + + // The previously registered object is superseded by + // this new object. We are guaranteed to not have any + // mergeable neighbours in this segment set. + if !raceEnabled { + seg.SetRangeUnchecked(r) + } else { + // Add extra paranoid. This will be statically + // removed at compile time unless a race build. + es.values.Remove(seg) + es.values.Add(r, existing) + seg = es.values.LowerBoundSegment(addr) + } + + // Compute the traversal required & update references. + dots := traverse(obj.Type(), old.Type(), addr, seg.Start()) + wt := es.findType(obj.Type()) + for _, ref := range existing.refs { + ref.Dots = append(ref.Dots, dots...) + ref.Type = wt + } + default: + // There is a non-sensical overlap. + Failf("overlapping objects: [new object] %#v [existing object] %#v", obj, existing.obj) + } + + // Compute the new reference, record and return it. + ref.Root = wire.Uint(existing.id) + ref.Dots = traverse(existing.obj.Type(), obj.Type(), seg.Start(), addr) + ref.Type = es.findType(obj.Type()) + existing.refs = append(existing.refs, ref) + return + } + + // The only remaining case is a pointer value that doesn't overlap with + // any registered addresses. Create a new entry for it, and start + // tracking the first reference we just created. + oes := &objectEncodeState{ + id: es.nextID(), + obj: obj, + } + if !raceEnabled { + es.values.AddWithoutMerging(r, oes) } else { - // Push back the map itself; when maps are encoded from the - // top-level, forceMap will be equal to true. - es.pending.PushBack(queuedObject{id: id, obj: obj, path: es.recoverable.copy()}) + // Merges should never happen. This is just enabled extra + // sanity checks because the Merge function below will panic. + es.values.Add(r, oes) + } + es.pending.PushBack(oes) + es.deferred.PushBack(oes) + ref.Root = wire.Uint(oes.id) + oes.refs = append(oes.refs, ref) +} + +// traverse searches for a target object within a root object, where the target +// object is a struct field or array element within root, with potentially +// multiple intervening types. traverse returns the set of field or element +// traversals required to reach the target. +// +// Note that for efficiency, traverse returns the dots in the reverse order. +// That is, the first traversal required will be the last element of the list. +// +// Precondition: The target object must lie completely within the range defined +// by [rootAddr, rootAddr + sizeof(rootType)]. +func traverse(rootType, targetType reflect.Type, rootAddr, targetAddr uintptr) []wire.Dot { + // Recursion base case: the types actually match. + if targetType == rootType && targetAddr == rootAddr { + return nil } - return id + switch rootType.Kind() { + case reflect.Struct: + offset := targetAddr - rootAddr + for i := rootType.NumField(); i > 0; i-- { + field := rootType.Field(i - 1) + // The first field from the end with an offset that is + // smaller than or equal to our address offset is where + // the target is located. Traverse from there. + if field.Offset <= offset { + dots := traverse(field.Type, targetType, rootAddr+field.Offset, targetAddr) + fieldName := wire.FieldName(field.Name) + return append(dots, &fieldName) + } + } + // Should never happen; the target should be reachable. + Failf("no field in root type %v contains target type %v", rootType, targetType) + + case reflect.Array: + // Since arrays have homogenous types, all elements have the + // same size and we can compute where the target lives. This + // does not matter for the purpose of typing, but matters for + // the purpose of computing the address of the given index. + elemSize := int(rootType.Elem().Size()) + n := int(targetAddr-rootAddr) / elemSize // Relies on integer division rounding down. + if rootType.Len() < n { + Failf("traversal target of type %v @%x is beyond the end of the array type %v @%x with %v elements", + targetType, targetAddr, rootType, rootAddr, rootType.Len()) + } + dots := traverse(rootType.Elem(), targetType, rootAddr+uintptr(n*elemSize), targetAddr) + return append(dots, wire.Index(n)) + + default: + // For any other type, there's no possibility of aliasing so if + // the types didn't match earlier then we have an addresss + // collision which shouldn't be possible at this point. + Failf("traverse failed for root type %v and target type %v", rootType, targetType) + } + panic("unreachable") } // encodeMap encodes a map. -func (es *encodeState) encodeMap(obj reflect.Value) *pb.Map { - var ( - keys []*pb.Object - values []*pb.Object - ) +func (es *encodeState) encodeMap(obj reflect.Value, dest *wire.Object) { + if obj.IsNil() { + // Because there is a difference between a nil map and an empty + // map, we need to not decode in the case of a truly nil map. + *dest = wire.Nil{} + return + } + l := obj.Len() + m := &wire.Map{ + Keys: make([]wire.Object, l), + Values: make([]wire.Object, l), + } + *dest = m for i, k := range obj.MapKeys() { v := obj.MapIndex(k) - kp := es.encodeObject(k, false, ".(key %d)", i) - vp := es.encodeObject(v, false, "[%#v]", k.Interface()) - keys = append(keys, kp) - values = append(values, vp) + // Map keys must be encoded using the full value because the + // type will be omitted after the first key. + es.encodeObject(k, encodeAsValue, &m.Keys[i]) + es.encodeObject(v, encodeAsValue, &m.Values[i]) } - return &pb.Map{Keys: keys, Values: values} +} + +// objectEncoder is for encoding structs. +type objectEncoder struct { + // es is encodeState. + es *encodeState + + // encoded is the encoded struct. + encoded *wire.Struct +} + +// save is called by the public methods on Sink. +func (oe *objectEncoder) save(slot int, obj reflect.Value) { + fieldValue := oe.encoded.Field(slot) + oe.es.encodeObject(obj, encodeDefault, fieldValue) } // encodeStruct encodes a composite object. -func (es *encodeState) encodeStruct(obj reflect.Value) *pb.Struct { - // Invoke the save. - m := Map{newInternalMap(es, nil, nil)} - defer internalMapPool.Put(m.internalMap) +func (es *encodeState) encodeStruct(obj reflect.Value, dest *wire.Object) { + // Ensure that the obj is addressable. There are two cases when it is + // not. First, is when this is dispatched via SaveValue. Second, when + // this is a map key as a struct. Either way, we need to make a copy to + // obtain an addressable value. if !obj.CanAddr() { - // Force it to a * type of the above; this involves a copy. localObj := reflect.New(obj.Type()) localObj.Elem().Set(obj) obj = localObj.Elem() } - fns, ok := registeredTypes.lookupFns(obj.Addr().Type()) - if ok { - // Invoke the provided saver. - fns.invokeSave(obj.Addr(), m) - } else if obj.NumField() == 0 { - // Allow unregistered anonymous, empty structs. - return &pb.Struct{} - } else { - // Propagate an error. - panic(fmt.Errorf("unregistered type %T", obj.Interface())) - } - - // Sort the underlying slice, and check for duplicates. This is done - // once instead of on each add, because performing this sort once is - // far more efficient. - if len(m.data) > 1 { - sort.Slice(m.data, func(i, j int) bool { - return m.data[i].name < m.data[j].name - }) - for i := range m.data { - if i > 0 && m.data[i-1].name == m.data[i].name { - panic(fmt.Errorf("duplicate name %s", m.data[i].name)) - } + + // Prepare the value. + s := &wire.Struct{} + *dest = s + + // Look the type up in the database. + te, ok := es.types.Lookup(obj.Type()) + if te == nil { + if obj.NumField() == 0 { + // Allow unregistered anonymous, empty structs. This + // will just return success without ever invoking the + // passed function. This uses the immutable EmptyStruct + // variable to prevent an allocation in this case. + // + // Note that this mechanism does *not* work for + // interfaces in general. So you can't dispatch + // non-registered empty structs via interfaces because + // then they can't be restored. + s.Alloc(0) + return } + // We need a SaverLoader for struct types. + Failf("struct %T does not implement SaverLoader", obj.Interface()) } - - // Encode the resulting fields. - fields := make([]*pb.Field, 0, len(m.data)) - for _, e := range m.data { - fields = append(fields, &pb.Field{ - Name: e.name, - Value: e.object, - }) + if !ok { + // Queue the type to be serialized. + es.pendingTypes = append(es.pendingTypes, te.Type) } - // Return the encoded object. - return &pb.Struct{Fields: fields} + // Invoke the provided saver. + s.TypeID = wire.TypeID(te.ID) + s.Alloc(len(te.Fields)) + oe := objectEncoder{ + es: es, + encoded: s, + } + es.stats.start(te.ID) + defer es.stats.done() + if sl, ok := obj.Addr().Interface().(SaverLoader); ok { + // Note: may be a registered empty struct which does not + // implement the saver/loader interfaces. + sl.StateSave(Sink{internal: oe}) + } } // encodeArray encodes an array. -func (es *encodeState) encodeArray(obj reflect.Value) *pb.Array { - var ( - contents []*pb.Object - ) - for i := 0; i < obj.Len(); i++ { - entry := es.encodeObject(obj.Index(i), false, "[%d]", i) - contents = append(contents, entry) - } - return &pb.Array{Contents: contents} +func (es *encodeState) encodeArray(obj reflect.Value, dest *wire.Object) { + l := obj.Len() + a := &wire.Array{ + Contents: make([]wire.Object, l), + } + *dest = a + for i := 0; i < l; i++ { + // We need to encode the full value because arrays are encoded + // using the type information from only the first element. + es.encodeObject(obj.Index(i), encodeAsValue, &a.Contents[i]) + } +} + +// findType recursively finds type information. +func (es *encodeState) findType(typ reflect.Type) wire.TypeSpec { + // First: check if this is a proper type. It's possible for pointers, + // slices, arrays, maps, etc to all have some different type. + te, ok := es.types.Lookup(typ) + if te != nil { + if !ok { + // See encodeStruct. + es.pendingTypes = append(es.pendingTypes, te.Type) + } + return wire.TypeID(te.ID) + } + + switch typ.Kind() { + case reflect.Ptr: + return &wire.TypeSpecPointer{ + Type: es.findType(typ.Elem()), + } + case reflect.Slice: + return &wire.TypeSpecSlice{ + Type: es.findType(typ.Elem()), + } + case reflect.Array: + return &wire.TypeSpecArray{ + Count: wire.Uint(typ.Len()), + Type: es.findType(typ.Elem()), + } + case reflect.Map: + return &wire.TypeSpecMap{ + Key: es.findType(typ.Key()), + Value: es.findType(typ.Elem()), + } + default: + // After potentially chasing many pointers, the + // ultimate type of the object is not known. + Failf("type %q is not known", typ) + } + panic("unreachable") } // encodeInterface encodes an interface. -// -// Precondition: the value is not nil. -func (es *encodeState) encodeInterface(obj reflect.Value) *pb.Interface { - // Check for the nil interface. - obj = reflect.ValueOf(obj.Interface()) +func (es *encodeState) encodeInterface(obj reflect.Value, dest *wire.Object) { + // Dereference the object. + obj = obj.Elem() if !obj.IsValid() { - return &pb.Interface{ - Type: "", // left alone in decode. - Value: &pb.Object{Value: &pb.Object_RefValue{0}}, + // Special case: the nil object. + *dest = &wire.Interface{ + Type: wire.TypeSpecNil{}, + Value: wire.Nil{}, } + return } - // We have an interface value here. How do we save that? We - // resolve the underlying type and save it as a dispatchable. - typName, ok := registeredTypes.lookupName(obj.Type()) - if !ok { - panic(fmt.Errorf("type %s is not registered", obj.Type())) + + // Encode underlying object. + i := &wire.Interface{ + Type: es.findType(obj.Type()), } + *dest = i + es.encodeObject(obj, encodeAsValue, &i.Value) +} - // Encode the object again. - return &pb.Interface{ - Type: typName, - Value: es.encodeObject(obj, false, ".(%s)", typName), +// isPrimitive returns true if this is a primitive object, or a composite +// object composed entirely of primitives. +func isPrimitiveZero(typ reflect.Type) bool { + switch typ.Kind() { + case reflect.Ptr: + // Pointers are always treated as primitive types because we + // won't encode directly from here. Returning true here won't + // prevent the object from being encoded correctly. + return true + case reflect.Bool: + return true + case reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64: + return true + case reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr: + return true + case reflect.Float32, reflect.Float64: + return true + case reflect.Complex64, reflect.Complex128: + return true + case reflect.String: + return true + case reflect.Slice: + // The slice itself a primitive, but not necessarily the array + // that points to. This is similar to a pointer. + return true + case reflect.Array: + // We cannot treat an array as a primitive, because it may be + // composed of structures or other things with side-effects. + return isPrimitiveZero(typ.Elem()) + case reflect.Interface: + // Since we now that this type is the zero type, the interface + // value must be zero. Therefore this is primitive. + return true + case reflect.Struct: + return false + case reflect.Map: + // The isPrimitiveZero function is called only on zero-types to + // see if it's safe to serialize. Since a zero map has no + // elements, it is safe to treat as a primitive. + return true + default: + Failf("unknown type %q", typ.Name()) } + panic("unreachable") } -// encodeObject encodes an object. -// -// If mapAsValue is true, then a map will be encoded directly. -func (es *encodeState) encodeObject(obj reflect.Value, mapAsValue bool, format string, param interface{}) (object *pb.Object) { - es.push(false, format, param) - es.stats.Add(obj) - es.stats.Start(obj) +// encodeStrategy is the strategy used for encodeObject. +type encodeStrategy int +const ( + // encodeDefault means types are encoded normally as references. + encodeDefault encodeStrategy = iota + + // encodeAsValue means that types will never take short-circuited and + // will always be encoded as a normal value. + encodeAsValue + + // encodeMapAsValue means that even maps will be fully encoded. + encodeMapAsValue +) + +// encodeObject encodes an object. +func (es *encodeState) encodeObject(obj reflect.Value, how encodeStrategy, dest *wire.Object) { + if how == encodeDefault && isPrimitiveZero(obj.Type()) && obj.IsZero() { + *dest = wire.Nil{} + return + } switch obj.Kind() { + case reflect.Ptr: // Fast path: first. + r := new(wire.Ref) + *dest = r + if obj.IsNil() { + // May be in an array or elsewhere such that a value is + // required. So we encode as a reference to the zero + // object, which does not exist. Note that this has to + // be handled correctly in the decode path as well. + return + } + es.resolve(obj, r) case reflect.Bool: - object = &pb.Object{Value: &pb.Object_BoolValue{obj.Bool()}} + *dest = wire.Bool(obj.Bool()) case reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64: - object = &pb.Object{Value: &pb.Object_Int64Value{obj.Int()}} + *dest = wire.Int(obj.Int()) case reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr: - object = &pb.Object{Value: &pb.Object_Uint64Value{obj.Uint()}} - case reflect.Float32, reflect.Float64: - object = &pb.Object{Value: &pb.Object_DoubleValue{obj.Float()}} + *dest = wire.Uint(obj.Uint()) + case reflect.Float32: + *dest = wire.Float32(obj.Float()) + case reflect.Float64: + *dest = wire.Float64(obj.Float()) + case reflect.Complex64: + c := wire.Complex64(obj.Complex()) + *dest = &c // Needs alloc. + case reflect.Complex128: + c := wire.Complex128(obj.Complex()) + *dest = &c // Needs alloc. + case reflect.String: + s := wire.String(obj.String()) + *dest = &s // Needs alloc. case reflect.Array: - switch obj.Type().Elem().Kind() { - case reflect.Uint8: - object = &pb.Object{Value: &pb.Object_ByteArrayValue{pbSlice(obj).Interface().([]byte)}} - case reflect.Uint16: - // 16-bit slices are serialized as 32-bit slices. - // See object.proto for details. - s := pbSlice(obj).Interface().([]uint16) - t := make([]uint32, len(s)) - for i := range s { - t[i] = uint32(s[i]) - } - object = &pb.Object{Value: &pb.Object_Uint16ArrayValue{&pb.Uint16S{Values: t}}} - case reflect.Uint32: - object = &pb.Object{Value: &pb.Object_Uint32ArrayValue{&pb.Uint32S{Values: pbSlice(obj).Interface().([]uint32)}}} - case reflect.Uint64: - object = &pb.Object{Value: &pb.Object_Uint64ArrayValue{&pb.Uint64S{Values: pbSlice(obj).Interface().([]uint64)}}} - case reflect.Uintptr: - object = &pb.Object{Value: &pb.Object_UintptrArrayValue{&pb.Uintptrs{Values: pbSlice(obj).Interface().([]uint64)}}} - case reflect.Int8: - object = &pb.Object{Value: &pb.Object_Int8ArrayValue{&pb.Int8S{Values: pbSlice(obj).Interface().([]byte)}}} - case reflect.Int16: - // 16-bit slices are serialized as 32-bit slices. - // See object.proto for details. - s := pbSlice(obj).Interface().([]int16) - t := make([]int32, len(s)) - for i := range s { - t[i] = int32(s[i]) - } - object = &pb.Object{Value: &pb.Object_Int16ArrayValue{&pb.Int16S{Values: t}}} - case reflect.Int32: - object = &pb.Object{Value: &pb.Object_Int32ArrayValue{&pb.Int32S{Values: pbSlice(obj).Interface().([]int32)}}} - case reflect.Int64: - object = &pb.Object{Value: &pb.Object_Int64ArrayValue{&pb.Int64S{Values: pbSlice(obj).Interface().([]int64)}}} - case reflect.Bool: - object = &pb.Object{Value: &pb.Object_BoolArrayValue{&pb.Bools{Values: pbSlice(obj).Interface().([]bool)}}} - case reflect.Float32: - object = &pb.Object{Value: &pb.Object_Float32ArrayValue{&pb.Float32S{Values: pbSlice(obj).Interface().([]float32)}}} - case reflect.Float64: - object = &pb.Object{Value: &pb.Object_Float64ArrayValue{&pb.Float64S{Values: pbSlice(obj).Interface().([]float64)}}} - default: - object = &pb.Object{Value: &pb.Object_ArrayValue{es.encodeArray(obj)}} - } + es.encodeArray(obj, dest) case reflect.Slice: - if obj.IsNil() || obj.Cap() == 0 { - // Handled specially in decode; store as nil value. - object = &pb.Object{Value: &pb.Object_RefValue{0}} - } else { - // Serialize a slice as the array plus length and capacity. - object = &pb.Object{Value: &pb.Object_SliceValue{&pb.Slice{ - Capacity: uint32(obj.Cap()), - Length: uint32(obj.Len()), - RefValue: es.register(arrayFromSlice(obj)), - }}} + s := &wire.Slice{ + Capacity: wire.Uint(obj.Cap()), + Length: wire.Uint(obj.Len()), } - case reflect.String: - object = &pb.Object{Value: &pb.Object_StringValue{[]byte(obj.String())}} - case reflect.Ptr: + *dest = s + // Note that we do need to provide a wire.Slice type here as + // how is not encodeDefault. If this were the case, then it + // would have been caught by the IsZero check above and we + // would have just used wire.Nil{}. if obj.IsNil() { - // Handled specially in decode; store as a nil value. - object = &pb.Object{Value: &pb.Object_RefValue{0}} - } else { - es.push(true /* dereference */, "", nil) - object = &pb.Object{Value: &pb.Object_RefValue{es.register(obj)}} - es.pop() + return } + // Slices need pointer resolution. + es.resolve(arrayFromSlice(obj), &s.Ref) case reflect.Interface: - // We don't check for IsNil here, as we want to encode type - // information. The case of the empty interface (no type, no - // value) is handled by encodeInteface. - object = &pb.Object{Value: &pb.Object_InterfaceValue{es.encodeInterface(obj)}} + es.encodeInterface(obj, dest) case reflect.Struct: - object = &pb.Object{Value: &pb.Object_StructValue{es.encodeStruct(obj)}} + es.encodeStruct(obj, dest) case reflect.Map: - if obj.IsNil() { - // Handled specially in decode; store as a nil value. - object = &pb.Object{Value: &pb.Object_RefValue{0}} - } else if mapAsValue { - // Encode the map directly. - object = &pb.Object{Value: &pb.Object_MapValue{es.encodeMap(obj)}} - } else { - // Encode a reference to the map. - // - // Remove the map object count here to avoid double - // counting, as this object will be counted again when - // it gets processed later. We do not add a reference - // count as the reference is artificial. - es.stats.Remove(obj) - object = &pb.Object{Value: &pb.Object_RefValue{es.register(obj)}} + if how == encodeMapAsValue { + es.encodeMap(obj, dest) + return } + r := new(wire.Ref) + *dest = r + es.resolve(obj, r) default: - panic(fmt.Errorf("unknown primitive %#v", obj.Interface())) + Failf("unknown object %#v", obj.Interface()) + panic("unreachable") } - - es.stats.Done() - es.pop() - return } -// Serialize serializes the object state. -// -// This function may panic and should be run in safely(). -func (es *encodeState) Serialize(obj reflect.Value) { - es.register(obj.Addr()) - - // Pop off the list until we're done. - for es.pending.Len() > 0 { - e := es.pending.Front() - - // Extract the queued object. - qo := e.Value.(queuedObject) - es.stats.Start(qo.obj) +// Save serializes the object graph rooted at obj. +func (es *encodeState) Save(obj reflect.Value) { + es.stats.init() + defer es.stats.fini(func(id typeID) string { + return es.pendingTypes[id-1].Name + }) + + // Resolve the first object, which should queue a pile of additional + // objects on the pending list. All queued objects should be fully + // resolved, and we should be able to serialize after this call. + var root wire.Ref + es.resolve(obj.Addr(), &root) + + // Encode the graph. + var oes *objectEncodeState + if err := safely(func() { + for oes = es.deferred.Front(); oes != nil; oes = es.deferred.Front() { + // Remove and encode the object. Note that as a result + // of this encoding, the object may be enqueued on the + // deferred list yet again. That's expected, and why it + // is removed first. + es.deferred.Remove(oes) + es.encodeObject(oes.obj, oes.how, &oes.encoded) + } + }); err != nil { + // Include the object in the error message. + Failf("encoding error at object %#v: %w", oes.obj.Interface(), err) + } - es.pending.Remove(e) + // Check that items are pending. + if es.pending.Front() == nil { + Failf("pending is empty?") + } - es.from = &qo.path - o := es.encodeObject(qo.obj, true, "", nil) + // Write the header with the number of objects. Note that there is no + // way that es.lastID could conflict with objectID, which would + // indicate that an impossibly large encoding. + if err := WriteHeader(es.w, uint64(es.lastID), true); err != nil { + Failf("error writing header: %w", err) + } - // Emit to our output stream. - if err := es.writeObject(qo.id, o); err != nil { - panic(err) + // Serialize all pending types and pending objects. Note that we don't + // bother removing from this list as we walk it because that just + // wastes time. It will not change after this point. + var id objectID + if err := safely(func() { + for _, wt := range es.pendingTypes { + // Encode the type. + wire.Save(es.w, &wt) } + for oes = es.pending.Front(); oes != nil; oes = oes.pendingEntry.Next() { + id++ // First object is 1. + if oes.id != id { + Failf("expected id %d, got %d", id, oes.id) + } - // Mark as done. - es.done.PushBack(e) - es.stats.Done() + // Marshall the object. + wire.Save(es.w, oes.encoded) + } + }); err != nil { + // Include the object and the error. + Failf("error serializing object %#v: %w", oes.encoded, err) } - // Write a zero-length terminal at the end; this is a sanity check - // applied at decode time as well (see decode.go). - if err := WriteHeader(es.w, 0, false); err != nil { - panic(err) + // Check what we wrote. + if id != es.lastID { + Failf("expected %d objects, wrote %d", es.lastID, id) } } +// objectFlag indicates that the length is a # of objects, rather than a raw +// byte length. When this is set on a length header in the stream, it may be +// decoded appropriately. +const objectFlag uint64 = 1 << 63 + // WriteHeader writes a header. // // Each object written to the statefile should be prefixed with a header. In // order to generate statefiles that play nicely with debugging tools, raw // writes should be prefixed with a header with object set to false and the // appropriate length. This will allow tools to skip these regions. -func WriteHeader(w io.Writer, length uint64, object bool) error { - // The lowest-order bit encodes whether this is a valid object. This is - // a purely internal convention, but allows the object flag to be - // returned from ReadHeader. - length = length << 1 +func WriteHeader(w wire.Writer, length uint64, object bool) error { + // Sanity check the length. + if length&objectFlag != 0 { + Failf("impossibly huge length: %d", length) + } if object { - length |= 0x1 + length |= objectFlag } // Write a header. - var hdr [32]byte - encodedLen := binary.PutUvarint(hdr[:], length) - for done := 0; done < encodedLen; { - n, err := w.Write(hdr[done:encodedLen]) - done += n - if n == 0 && err != nil { - return err - } - } - - return nil + return safely(func() { + wire.SaveUint(w, length) + }) } -// writeObject writes an object to the stream. -func (es *encodeState) writeObject(id uint64, obj *pb.Object) error { - // Marshal the proto. - buf, err := proto.Marshal(obj) - if err != nil { - return err - } +// pendingMapper is for the pending list. +type pendingMapper struct{} - // Write the object header. - if err := WriteHeader(es.w, uint64(len(buf)), true); err != nil { - return err - } +func (pendingMapper) linkerFor(oes *objectEncodeState) *pendingEntry { return &oes.pendingEntry } - // Write the object. - for done := 0; done < len(buf); { - n, err := es.w.Write(buf[done:]) - done += n - if n == 0 && err != nil { - return err - } - } +// deferredMapper is for the deferred list. +type deferredMapper struct{} - return nil -} +func (deferredMapper) linkerFor(oes *objectEncodeState) *deferredEntry { return &oes.deferredEntry } // addrSetFunctions is used by addrSet. type addrSetFunctions struct{} @@ -458,13 +818,24 @@ func (addrSetFunctions) MaxKey() uintptr { return ^uintptr(0) } -func (addrSetFunctions) ClearValue(val *reflect.Value) { +func (addrSetFunctions) ClearValue(val **objectEncodeState) { + *val = nil } -func (addrSetFunctions) Merge(_ addrRange, val1 reflect.Value, _ addrRange, val2 reflect.Value) (reflect.Value, bool) { - return val1, val1 == val2 +func (addrSetFunctions) Merge(r1 addrRange, val1 *objectEncodeState, r2 addrRange, val2 *objectEncodeState) (*objectEncodeState, bool) { + if val1.obj == val2.obj { + // This, should never happen. It would indicate that the same + // object exists in two non-contiguous address ranges. Note + // that this assertion can only be triggered if the race + // detector is enabled. + Failf("unexpected merge in addrSet @ %v and %v: %#v and %#v", r1, r2, val1.obj, val2.obj) + } + // Reject the merge. + return val1, false } -func (addrSetFunctions) Split(_ addrRange, val reflect.Value, _ uintptr) (reflect.Value, reflect.Value) { - return val, val +func (addrSetFunctions) Split(r addrRange, val *objectEncodeState, _ uintptr) (*objectEncodeState, *objectEncodeState) { + // A split should never happen: we don't remove ranges. + Failf("unexpected split in addrSet @ %v: %#v", r, val.obj) + panic("unreachable") } diff --git a/pkg/state/encode_unsafe.go b/pkg/state/encode_unsafe.go index 457e6dbb7..e0dad83b4 100644 --- a/pkg/state/encode_unsafe.go +++ b/pkg/state/encode_unsafe.go @@ -31,51 +31,3 @@ func arrayFromSlice(obj reflect.Value) reflect.Value { reflect.ArrayOf(obj.Cap(), obj.Type().Elem()), unsafe.Pointer(obj.Pointer())) } - -// pbSlice returns a protobuf-supported slice of the array and erase the -// original element type (which could be a defined type or non-supported type). -func pbSlice(obj reflect.Value) reflect.Value { - var typ reflect.Type - switch obj.Type().Elem().Kind() { - case reflect.Uint8: - typ = reflect.TypeOf(byte(0)) - case reflect.Uint16: - typ = reflect.TypeOf(uint16(0)) - case reflect.Uint32: - typ = reflect.TypeOf(uint32(0)) - case reflect.Uint64: - typ = reflect.TypeOf(uint64(0)) - case reflect.Uintptr: - typ = reflect.TypeOf(uint64(0)) - case reflect.Int8: - typ = reflect.TypeOf(byte(0)) - case reflect.Int16: - typ = reflect.TypeOf(int16(0)) - case reflect.Int32: - typ = reflect.TypeOf(int32(0)) - case reflect.Int64: - typ = reflect.TypeOf(int64(0)) - case reflect.Bool: - typ = reflect.TypeOf(bool(false)) - case reflect.Float32: - typ = reflect.TypeOf(float32(0)) - case reflect.Float64: - typ = reflect.TypeOf(float64(0)) - default: - panic("slice element is not of basic value type") - } - return reflect.NewAt( - reflect.ArrayOf(obj.Len(), typ), - unsafe.Pointer(obj.Slice(0, obj.Len()).Pointer()), - ).Elem().Slice(0, obj.Len()) -} - -func castSlice(obj reflect.Value, elemTyp reflect.Type) reflect.Value { - if obj.Type().Elem().Size() != elemTyp.Size() { - panic("cannot cast slice into other element type of different size") - } - return reflect.NewAt( - reflect.ArrayOf(obj.Len(), elemTyp), - unsafe.Pointer(obj.Slice(0, obj.Len()).Pointer()), - ).Elem() -} diff --git a/pkg/state/map.go b/pkg/state/map.go deleted file mode 100644 index 4f3ebb0da..000000000 --- a/pkg/state/map.go +++ /dev/null @@ -1,232 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package state - -import ( - "context" - "fmt" - "reflect" - "sort" - "sync" - - pb "gvisor.dev/gvisor/pkg/state/object_go_proto" -) - -// entry is a single map entry. -type entry struct { - name string - object *pb.Object -} - -// internalMap is the internal Map state. -// -// These are recycled via a pool to avoid churn. -type internalMap struct { - // es is encodeState. - es *encodeState - - // ds is decodeState. - ds *decodeState - - // os is current object being decoded. - // - // This will always be nil during encode. - os *objectState - - // data stores the encoded values. - data []entry -} - -var internalMapPool = sync.Pool{ - New: func() interface{} { - return new(internalMap) - }, -} - -// newInternalMap returns a cached map. -func newInternalMap(es *encodeState, ds *decodeState, os *objectState) *internalMap { - m := internalMapPool.Get().(*internalMap) - m.es = es - m.ds = ds - m.os = os - if m.data != nil { - m.data = m.data[:0] - } - return m -} - -// Map is a generic state container. -// -// This is the object passed to Save and Load in order to store their state. -// -// Detailed documentation is available in individual methods. -type Map struct { - *internalMap -} - -// Save adds the given object to the map. -// -// You should pass always pointers to the object you are saving. For example: -// -// type X struct { -// A int -// B *int -// } -// -// func (x *X) Save(m Map) { -// m.Save("A", &x.A) -// m.Save("B", &x.B) -// } -// -// func (x *X) Load(m Map) { -// m.Load("A", &x.A) -// m.Load("B", &x.B) -// } -func (m Map) Save(name string, objPtr interface{}) { - m.save(name, reflect.ValueOf(objPtr).Elem(), ".%s") -} - -// SaveValue adds the given object value to the map. -// -// This should be used for values where pointers are not available, or casts -// are required during Save/Load. -// -// For example, if we want to cast external package type P.Foo to int64: -// -// type X struct { -// A P.Foo -// } -// -// func (x *X) Save(m Map) { -// m.SaveValue("A", int64(x.A)) -// } -// -// func (x *X) Load(m Map) { -// m.LoadValue("A", new(int64), func(x interface{}) { -// x.A = P.Foo(x.(int64)) -// }) -// } -func (m Map) SaveValue(name string, obj interface{}) { - m.save(name, reflect.ValueOf(obj), ".(value %s)") -} - -// save is helper for the above. It takes the name of value to save the field -// to, the field object (obj), and a format string that specifies how the -// field's saving logic is dispatched from the struct (normal, value, etc.). The -// format string should expect one string parameter, which is the name of the -// field. -func (m Map) save(name string, obj reflect.Value, format string) { - if m.es == nil { - // Not currently encoding. - m.Failf("no encode state for %q", name) - } - - // Attempt the encode. - // - // These are sorted at the end, after all objects are added and will be - // sorted and checked for duplicates (see encodeStruct). - m.data = append(m.data, entry{ - name: name, - object: m.es.encodeObject(obj, false, format, name), - }) -} - -// Load loads the given object from the map. -// -// See Save for an example. -func (m Map) Load(name string, objPtr interface{}) { - m.load(name, reflect.ValueOf(objPtr), false, nil, ".%s") -} - -// LoadWait loads the given objects from the map, and marks it as requiring all -// AfterLoad executions to complete prior to running this object's AfterLoad. -// -// See Save for an example. -func (m Map) LoadWait(name string, objPtr interface{}) { - m.load(name, reflect.ValueOf(objPtr), true, nil, ".(wait %s)") -} - -// LoadValue loads the given object value from the map. -// -// See SaveValue for an example. -func (m Map) LoadValue(name string, objPtr interface{}, fn func(interface{})) { - o := reflect.ValueOf(objPtr) - m.load(name, o, true, func() { fn(o.Elem().Interface()) }, ".(value %s)") -} - -// load is helper for the above. It takes the name of value to load the field -// from, the target field pointer (objPtr), whether load completion of the -// struct depends on the field's load completion (wait), the load completion -// logic (fn), and a format string that specifies how the field's loading logic -// is dispatched from the struct (normal, wait, value, etc.). The format string -// should expect one string parameter, which is the name of the field. -func (m Map) load(name string, objPtr reflect.Value, wait bool, fn func(), format string) { - if m.ds == nil { - // Not currently decoding. - m.Failf("no decode state for %q", name) - } - - // Find the object. - // - // These are sorted up front (and should appear in the state file - // sorted as well), so we can do a binary search here to ensure that - // large structs don't behave badly. - i := sort.Search(len(m.data), func(i int) bool { - return m.data[i].name >= name - }) - if i >= len(m.data) || m.data[i].name != name { - // There is no data for this name? - m.Failf("no data found for %q", name) - } - - // Perform the decode. - m.ds.decodeObject(m.os, objPtr.Elem(), m.data[i].object, format, name) - if wait { - // Mark this individual object a blocker. - m.ds.waitObject(m.os, m.data[i].object, fn) - } -} - -// Failf fails the save or restore with the provided message. Processing will -// stop after calling Failf, as the state package uses a panic & recover -// mechanism for state errors. You should defer any cleanup required. -func (m Map) Failf(format string, args ...interface{}) { - panic(fmt.Errorf(format, args...)) -} - -// AfterLoad schedules a function execution when all objects have been allocated -// and their automated loading and customized load logic have been executed. fn -// will not be executed until all of current object's dependencies' AfterLoad() -// logic, if exist, have been executed. -func (m Map) AfterLoad(fn func()) { - if m.ds == nil { - // Not currently decoding. - m.Failf("not decoding") - } - - // Queue the local callback; this will execute when all of the above - // data dependencies have been cleared. - m.os.callbacks = append(m.os.callbacks, fn) -} - -// Context returns the current context object. -func (m Map) Context() context.Context { - if m.es != nil { - return m.es.ctx - } else if m.ds != nil { - return m.ds.ctx - } - return context.Background() // No context. -} diff --git a/pkg/state/object.proto b/pkg/state/object.proto deleted file mode 100644 index 5ebcfb151..000000000 --- a/pkg/state/object.proto +++ /dev/null @@ -1,140 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -syntax = "proto3"; - -package gvisor.state.statefile; - -// Slice is a slice value. -message Slice { - uint32 length = 1; - uint32 capacity = 2; - uint64 ref_value = 3; -} - -// Array is an array value. -message Array { - repeated Object contents = 1; -} - -// Map is a map value. -message Map { - repeated Object keys = 1; - repeated Object values = 2; -} - -// Interface is an interface value. -message Interface { - string type = 1; - Object value = 2; -} - -// Struct is a basic composite value. -message Struct { - repeated Field fields = 1; -} - -// Field encodes a single field. -message Field { - string name = 1; - Object value = 2; -} - -// Uint16s encodes an uint16 array. To be used inside oneof structure. -message Uint16s { - // There is no 16-bit type in protobuf so we use variable length 32-bit here. - repeated uint32 values = 1; -} - -// Uint32s encodes an uint32 array. To be used inside oneof structure. -message Uint32s { - repeated fixed32 values = 1; -} - -// Uint64s encodes an uint64 array. To be used inside oneof structure. -message Uint64s { - repeated fixed64 values = 1; -} - -// Uintptrs encodes an uintptr array. To be used inside oneof structure. -message Uintptrs { - repeated fixed64 values = 1; -} - -// Int8s encodes an int8 array. To be used inside oneof structure. -message Int8s { - bytes values = 1; -} - -// Int16s encodes an int16 array. To be used inside oneof structure. -message Int16s { - // There is no 16-bit type in protobuf so we use variable length 32-bit here. - repeated int32 values = 1; -} - -// Int32s encodes an int32 array. To be used inside oneof structure. -message Int32s { - repeated sfixed32 values = 1; -} - -// Int64s encodes an int64 array. To be used inside oneof structure. -message Int64s { - repeated sfixed64 values = 1; -} - -// Bools encodes a boolean array. To be used inside oneof structure. -message Bools { - repeated bool values = 1; -} - -// Float64s encodes a float64 array. To be used inside oneof structure. -message Float64s { - repeated double values = 1; -} - -// Float32s encodes a float32 array. To be used inside oneof structure. -message Float32s { - repeated float values = 1; -} - -// Object are primitive encodings. -// -// Note that ref_value references an Object.id, below. -message Object { - oneof value { - bool bool_value = 1; - bytes string_value = 2; - int64 int64_value = 3; - uint64 uint64_value = 4; - double double_value = 5; - uint64 ref_value = 6; - Slice slice_value = 7; - Array array_value = 8; - Interface interface_value = 9; - Struct struct_value = 10; - Map map_value = 11; - bytes byte_array_value = 12; - Uint16s uint16_array_value = 13; - Uint32s uint32_array_value = 14; - Uint64s uint64_array_value = 15; - Uintptrs uintptr_array_value = 16; - Int8s int8_array_value = 17; - Int16s int16_array_value = 18; - Int32s int32_array_value = 19; - Int64s int64_array_value = 20; - Bools bool_array_value = 21; - Float64s float64_array_value = 22; - Float32s float32_array_value = 23; - } -} diff --git a/pkg/state/pretty/BUILD b/pkg/state/pretty/BUILD new file mode 100644 index 000000000..d053802f7 --- /dev/null +++ b/pkg/state/pretty/BUILD @@ -0,0 +1,13 @@ +load("//tools:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_library( + name = "pretty", + srcs = ["pretty.go"], + visibility = ["//:sandbox"], + deps = [ + "//pkg/state", + "//pkg/state/wire", + ], +) diff --git a/pkg/state/pretty/pretty.go b/pkg/state/pretty/pretty.go new file mode 100644 index 000000000..cf37aaa49 --- /dev/null +++ b/pkg/state/pretty/pretty.go @@ -0,0 +1,273 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package pretty is a pretty-printer for state streams. +package pretty + +import ( + "fmt" + "io" + "io/ioutil" + "reflect" + "strings" + + "gvisor.dev/gvisor/pkg/state" + "gvisor.dev/gvisor/pkg/state/wire" +) + +func formatRef(x *wire.Ref, graph uint64, html bool) string { + baseRef := fmt.Sprintf("g%dr%d", graph, x.Root) + fullRef := baseRef + if len(x.Dots) > 0 { + // See wire.Ref; Type valid if Dots non-zero. + typ, _ := formatType(x.Type, graph, html) + var buf strings.Builder + buf.WriteString("(*") + buf.WriteString(typ) + buf.WriteString(")(") + buf.WriteString(baseRef) + for _, component := range x.Dots { + switch v := component.(type) { + case *wire.FieldName: + buf.WriteString(".") + buf.WriteString(string(*v)) + case wire.Index: + buf.WriteString(fmt.Sprintf("[%d]", v)) + default: + panic(fmt.Sprintf("unreachable: switch should be exhaustive, unhandled case %v", reflect.TypeOf(component))) + } + } + buf.WriteString(")") + fullRef = buf.String() + } + if html { + return fmt.Sprintf("%s", baseRef, fullRef) + } + return fullRef +} + +func formatType(t wire.TypeSpec, graph uint64, html bool) (string, bool) { + switch x := t.(type) { + case wire.TypeID: + base := fmt.Sprintf("g%dt%d", graph, x) + if html { + return fmt.Sprintf("%s", base, base), true + } + return fmt.Sprintf("%s", base), true + case wire.TypeSpecNil: + return "", false // Only nil type. + case *wire.TypeSpecPointer: + element, _ := formatType(x.Type, graph, html) + return fmt.Sprintf("(*%s)", element), true + case *wire.TypeSpecArray: + element, _ := formatType(x.Type, graph, html) + return fmt.Sprintf("[%d](%s)", x.Count, element), true + case *wire.TypeSpecSlice: + element, _ := formatType(x.Type, graph, html) + return fmt.Sprintf("([]%s)", element), true + case *wire.TypeSpecMap: + key, _ := formatType(x.Key, graph, html) + value, _ := formatType(x.Value, graph, html) + return fmt.Sprintf("(map[%s]%s)", key, value), true + default: + panic(fmt.Sprintf("unreachable: unknown type %T", t)) + } +} + +// format formats a single object, for pretty-printing. It also returns whether +// the value is a non-zero value. +func format(graph uint64, depth int, encoded wire.Object, html bool) (string, bool) { + switch x := encoded.(type) { + case wire.Nil: + return "nil", false + case *wire.String: + return fmt.Sprintf("%q", *x), *x != "" + case *wire.Complex64: + return fmt.Sprintf("%f+%fi", real(*x), imag(*x)), *x != 0.0 + case *wire.Complex128: + return fmt.Sprintf("%f+%fi", real(*x), imag(*x)), *x != 0.0 + case *wire.Ref: + return formatRef(x, graph, html), x.Root != 0 + case *wire.Type: + tabs := "\n" + strings.Repeat("\t", depth) + items := make([]string, 0, len(x.Fields)+2) + items = append(items, fmt.Sprintf("type %s {", x.Name)) + for i := 0; i < len(x.Fields); i++ { + items = append(items, fmt.Sprintf("\t%d: %s,", i, x.Fields[i])) + } + items = append(items, "}") + return strings.Join(items, tabs), true // No zero value. + case *wire.Slice: + return fmt.Sprintf("%s{len:%d,cap:%d}", formatRef(&x.Ref, graph, html), x.Length, x.Capacity), x.Capacity != 0 + case *wire.Array: + if len(x.Contents) == 0 { + return "[]", false + } + items := make([]string, 0, len(x.Contents)+2) + zeros := make([]string, 0) // used to eliminate zero entries. + items = append(items, "[") + tabs := "\n" + strings.Repeat("\t", depth) + for i := 0; i < len(x.Contents); i++ { + item, ok := format(graph, depth+1, x.Contents[i], html) + if !ok { + zeros = append(zeros, fmt.Sprintf("\t%s,", item)) + continue + } + if len(zeros) > 0 { + items = append(items, zeros...) + zeros = nil + } + items = append(items, fmt.Sprintf("\t%s,", item)) + } + if len(zeros) > 0 { + items = append(items, fmt.Sprintf("\t... (%d zeros),", len(zeros))) + } + items = append(items, "]") + return strings.Join(items, tabs), len(zeros) < len(x.Contents) + case *wire.Struct: + typ, _ := formatType(x.TypeID, graph, html) + if x.Fields() == 0 { + return fmt.Sprintf("struct[%s]{}", typ), false + } + items := make([]string, 0, 2) + items = append(items, fmt.Sprintf("struct[%s]{", typ)) + tabs := "\n" + strings.Repeat("\t", depth) + allZero := true + for i := 0; i < x.Fields(); i++ { + element, ok := format(graph, depth+1, *x.Field(i), html) + allZero = allZero && !ok + items = append(items, fmt.Sprintf("\t%d: %s,", i, element)) + i++ + } + items = append(items, "}") + return strings.Join(items, tabs), !allZero + case *wire.Map: + if len(x.Keys) == 0 { + return "map{}", false + } + items := make([]string, 0, len(x.Keys)+2) + items = append(items, "map{") + tabs := "\n" + strings.Repeat("\t", depth) + for i := 0; i < len(x.Keys); i++ { + key, _ := format(graph, depth+1, x.Keys[i], html) + value, _ := format(graph, depth+1, x.Values[i], html) + items = append(items, fmt.Sprintf("\t%s: %s,", key, value)) + } + items = append(items, "}") + return strings.Join(items, tabs), true + case *wire.Interface: + typ, typOk := formatType(x.Type, graph, html) + element, elementOk := format(graph, depth+1, x.Value, html) + return fmt.Sprintf("interface[%s]{%s}", typ, element), typOk || elementOk + default: + // Must be a primitive; use reflection. + return fmt.Sprintf("%v", encoded), true + } +} + +// printStream is the basic print implementation. +func printStream(w io.Writer, r wire.Reader, html bool) (err error) { + // current graph ID. + var graph uint64 + + if html { + fmt.Fprintf(w, "
")
+		defer fmt.Fprintf(w, "
") + } + + defer func() { + if r := recover(); r != nil { + if rErr, ok := r.(error); ok { + err = rErr // Override return. + return + } + panic(r) // Propagate. + } + }() + + for { + // Find the first object to begin generation. + length, object, err := state.ReadHeader(r) + if err == io.EOF { + // Nothing else to do. + break + } else if err != nil { + return err + } + if !object { + graph++ // Increment the graph. + if length > 0 { + fmt.Fprintf(w, "(%d bytes non-object data)\n", length) + io.Copy(ioutil.Discard, &io.LimitedReader{ + R: r, + N: int64(length), + }) + } + continue + } + + // Read & unmarshal the object. + // + // Note that this loop must match the general structure of the + // loop in decode.go. But we don't register type information, + // etc. and just print the raw structures. + var ( + oid uint64 = 1 + tid uint64 = 1 + ) + for oid <= length { + // Unmarshal the object. + encoded := wire.Load(r) + + // Is this a type? + if _, ok := encoded.(*wire.Type); ok { + str, _ := format(graph, 0, encoded, html) + tag := fmt.Sprintf("g%dt%d", graph, tid) + if html { + // See below. + tag = fmt.Sprintf("%s⚓", tag, tag, tag) + } + if _, err := fmt.Fprintf(w, "%s = %s\n", tag, str); err != nil { + return err + } + tid++ + continue + } + + // Format the node. + str, _ := format(graph, 0, encoded, html) + tag := fmt.Sprintf("g%dr%d", graph, oid) + if html { + // Create a little tag with an anchor next to it for linking. + tag = fmt.Sprintf("%s⚓", tag, tag, tag) + } + if _, err := fmt.Fprintf(w, "%s = %s\n", tag, str); err != nil { + return err + } + oid++ + } + } + + return nil +} + +// PrintText reads the stream from r and prints text to w. +func PrintText(w io.Writer, r wire.Reader) error { + return printStream(w, r, false /* html */) +} + +// PrintHTML reads the stream from r and prints html to w. +func PrintHTML(w io.Writer, r wire.Reader) error { + return printStream(w, r, true /* html */) +} diff --git a/pkg/state/printer.go b/pkg/state/printer.go deleted file mode 100644 index 3ce18242f..000000000 --- a/pkg/state/printer.go +++ /dev/null @@ -1,251 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package state - -import ( - "fmt" - "io" - "io/ioutil" - "reflect" - "strings" - - "github.com/golang/protobuf/proto" - pb "gvisor.dev/gvisor/pkg/state/object_go_proto" -) - -// format formats a single object, for pretty-printing. It also returns whether -// the value is a non-zero value. -func format(graph uint64, depth int, object *pb.Object, html bool) (string, bool) { - switch x := object.GetValue().(type) { - case *pb.Object_BoolValue: - return fmt.Sprintf("%t", x.BoolValue), x.BoolValue != false - case *pb.Object_StringValue: - return fmt.Sprintf("\"%s\"", string(x.StringValue)), len(x.StringValue) != 0 - case *pb.Object_Int64Value: - return fmt.Sprintf("%d", x.Int64Value), x.Int64Value != 0 - case *pb.Object_Uint64Value: - return fmt.Sprintf("%du", x.Uint64Value), x.Uint64Value != 0 - case *pb.Object_DoubleValue: - return fmt.Sprintf("%f", x.DoubleValue), x.DoubleValue != 0.0 - case *pb.Object_RefValue: - if x.RefValue == 0 { - return "nil", false - } - ref := fmt.Sprintf("g%dr%d", graph, x.RefValue) - if html { - ref = fmt.Sprintf("%s", ref, ref) - } - return ref, true - case *pb.Object_SliceValue: - if x.SliceValue.RefValue == 0 { - return "nil", false - } - ref := fmt.Sprintf("g%dr%d", graph, x.SliceValue.RefValue) - if html { - ref = fmt.Sprintf("%s", ref, ref) - } - return fmt.Sprintf("%s[:%d:%d]", ref, x.SliceValue.Length, x.SliceValue.Capacity), true - case *pb.Object_ArrayValue: - if len(x.ArrayValue.Contents) == 0 { - return "[]", false - } - items := make([]string, 0, len(x.ArrayValue.Contents)+2) - zeros := make([]string, 0) // used to eliminate zero entries. - items = append(items, "[") - tabs := "\n" + strings.Repeat("\t", depth) - for i := 0; i < len(x.ArrayValue.Contents); i++ { - item, ok := format(graph, depth+1, x.ArrayValue.Contents[i], html) - if ok { - if len(zeros) > 0 { - items = append(items, zeros...) - zeros = nil - } - items = append(items, fmt.Sprintf("\t%s,", item)) - } else { - zeros = append(zeros, fmt.Sprintf("\t%s,", item)) - } - } - if len(zeros) > 0 { - items = append(items, fmt.Sprintf("\t... (%d zeros),", len(zeros))) - } - items = append(items, "]") - return strings.Join(items, tabs), len(zeros) < len(x.ArrayValue.Contents) - case *pb.Object_StructValue: - if len(x.StructValue.Fields) == 0 { - return "struct{}", false - } - items := make([]string, 0, len(x.StructValue.Fields)+2) - items = append(items, "struct{") - tabs := "\n" + strings.Repeat("\t", depth) - allZero := true - for _, field := range x.StructValue.Fields { - element, ok := format(graph, depth+1, field.Value, html) - allZero = allZero && !ok - items = append(items, fmt.Sprintf("\t%s: %s,", field.Name, element)) - } - items = append(items, "}") - return strings.Join(items, tabs), !allZero - case *pb.Object_MapValue: - if len(x.MapValue.Keys) == 0 { - return "map{}", false - } - items := make([]string, 0, len(x.MapValue.Keys)+2) - items = append(items, "map{") - tabs := "\n" + strings.Repeat("\t", depth) - for i := 0; i < len(x.MapValue.Keys); i++ { - key, _ := format(graph, depth+1, x.MapValue.Keys[i], html) - value, _ := format(graph, depth+1, x.MapValue.Values[i], html) - items = append(items, fmt.Sprintf("\t%s: %s,", key, value)) - } - items = append(items, "}") - return strings.Join(items, tabs), true - case *pb.Object_InterfaceValue: - if x.InterfaceValue.Type == "" { - return "interface(nil){}", false - } - element, _ := format(graph, depth+1, x.InterfaceValue.Value, html) - return fmt.Sprintf("interface(\"%s\"){%s}", x.InterfaceValue.Type, element), true - case *pb.Object_ByteArrayValue: - return printArray(reflect.ValueOf(x.ByteArrayValue)) - case *pb.Object_Uint16ArrayValue: - return printArray(reflect.ValueOf(x.Uint16ArrayValue.Values)) - case *pb.Object_Uint32ArrayValue: - return printArray(reflect.ValueOf(x.Uint32ArrayValue.Values)) - case *pb.Object_Uint64ArrayValue: - return printArray(reflect.ValueOf(x.Uint64ArrayValue.Values)) - case *pb.Object_UintptrArrayValue: - return printArray(castSlice(reflect.ValueOf(x.UintptrArrayValue.Values), reflect.TypeOf(uintptr(0)))) - case *pb.Object_Int8ArrayValue: - return printArray(castSlice(reflect.ValueOf(x.Int8ArrayValue.Values), reflect.TypeOf(int8(0)))) - case *pb.Object_Int16ArrayValue: - return printArray(reflect.ValueOf(x.Int16ArrayValue.Values)) - case *pb.Object_Int32ArrayValue: - return printArray(reflect.ValueOf(x.Int32ArrayValue.Values)) - case *pb.Object_Int64ArrayValue: - return printArray(reflect.ValueOf(x.Int64ArrayValue.Values)) - case *pb.Object_BoolArrayValue: - return printArray(reflect.ValueOf(x.BoolArrayValue.Values)) - case *pb.Object_Float64ArrayValue: - return printArray(reflect.ValueOf(x.Float64ArrayValue.Values)) - case *pb.Object_Float32ArrayValue: - return printArray(reflect.ValueOf(x.Float32ArrayValue.Values)) - } - - // Should not happen, but tolerate. - return fmt.Sprintf("(unknown proto type: %T)", object.GetValue()), true -} - -// PrettyPrint reads the state stream from r, and pretty prints to w. -func PrettyPrint(w io.Writer, r io.Reader, html bool) error { - var ( - // current graph ID. - graph uint64 - - // current object ID. - id uint64 - ) - - if html { - fmt.Fprintf(w, "
")
-		defer fmt.Fprintf(w, "
") - } - - for { - // Find the first object to begin generation. - length, object, err := ReadHeader(r) - if err == io.EOF { - // Nothing else to do. - break - } else if err != nil { - return err - } - if !object { - // Increment the graph number & reset the ID. - graph++ - id = 0 - if length > 0 { - fmt.Fprintf(w, "(%d bytes non-object data)\n", length) - io.Copy(ioutil.Discard, &io.LimitedReader{ - R: r, - N: int64(length), - }) - } - continue - } - - // Read & unmarshal the object. - buf := make([]byte, length) - for done := 0; done < len(buf); { - n, err := r.Read(buf[done:]) - done += n - if n == 0 && err != nil { - return err - } - } - obj := new(pb.Object) - if err := proto.Unmarshal(buf, obj); err != nil { - return err - } - - id++ // First object must be one. - str, _ := format(graph, 0, obj, html) - tag := fmt.Sprintf("g%dr%d", graph, id) - if html { - tag = fmt.Sprintf("%s", tag, tag) - } - if _, err := fmt.Fprintf(w, "%s = %s\n", tag, str); err != nil { - return err - } - } - - return nil -} - -func printArray(s reflect.Value) (string, bool) { - zero := reflect.Zero(s.Type().Elem()).Interface() - z := "0" - switch s.Type().Elem().Kind() { - case reflect.Bool: - z = "false" - case reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64: - case reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr: - case reflect.Float32, reflect.Float64: - default: - return fmt.Sprintf("unexpected non-primitive type array: %#v", s.Interface()), true - } - - zeros := 0 - items := make([]string, 0, s.Len()) - for i := 0; i <= s.Len(); i++ { - if i < s.Len() && reflect.DeepEqual(s.Index(i).Interface(), zero) { - zeros++ - continue - } - if zeros > 0 { - if zeros <= 4 { - for ; zeros > 0; zeros-- { - items = append(items, z) - } - } else { - items = append(items, fmt.Sprintf("(%d %ss)", zeros, z)) - zeros = 0 - } - } - if i < s.Len() { - items = append(items, fmt.Sprintf("%v", s.Index(i).Interface())) - } - } - return "[" + strings.Join(items, ",") + "]", zeros < s.Len() -} diff --git a/pkg/state/state.go b/pkg/state/state.go index 03ae2dbb0..acb629969 100644 --- a/pkg/state/state.go +++ b/pkg/state/state.go @@ -31,210 +31,226 @@ // Uint64 default // Float32 default // Float64 default -// Complex64 custom -// Complex128 custom +// Complex64 default +// Complex128 default // Array default // Chan custom // Func custom -// Interface custom -// Map default (*) +// Interface default +// Map default // Ptr default // Slice default // String default -// Struct custom +// Struct custom (*) Unless zero-sized. // UnsafePointer custom // -// (*) Maps are treated as value types by this package, even if they are -// pointers internally. If you want to save two independent references -// to the same map value, you must explicitly use a pointer to a map. +// See README.md for an overview of how encoding and decoding works. package state import ( "context" "fmt" - "io" "reflect" "runtime" - pb "gvisor.dev/gvisor/pkg/state/object_go_proto" + "gvisor.dev/gvisor/pkg/state/wire" ) +// objectID is a unique identifier assigned to each object to be serialized. +// Each instance of an object is considered separately, i.e. if there are two +// objects of the same type in the object graph being serialized, they'll be +// assigned unique objectIDs. +type objectID uint32 + +// typeID is the identifier for a type. Types are serialized and tracked +// alongside objects in order to avoid the overhead of encoding field names in +// all objects. +type typeID uint32 + // ErrState is returned when an error is encountered during encode/decode. type ErrState struct { // err is the underlying error. err error - // path is the visit path from root to the current object. - path string - // trace is the stack trace. trace string } // Error returns a sensible description of the state error. func (e *ErrState) Error() string { - return fmt.Sprintf("%v:\nstate path: %s\n%s", e.err, e.path, e.trace) + return fmt.Sprintf("%v:\n%s", e.err, e.trace) } -// UnwrapErrState returns the underlying error in ErrState. -// -// If err is not *ErrState, err is returned directly. -func UnwrapErrState(err error) error { - if e, ok := err.(*ErrState); ok { - return e.err - } - return err +// Unwrap implements standard unwrapping. +func (e *ErrState) Unwrap() error { + return e.err } // Save saves the given object state. -func Save(ctx context.Context, w io.Writer, rootPtr interface{}, stats *Stats) error { +func Save(ctx context.Context, w wire.Writer, rootPtr interface{}) (Stats, error) { // Create the encoding state. - es := &encodeState{ - ctx: ctx, - idsByObject: make(map[uintptr]uint64), - w: w, - stats: stats, + es := encodeState{ + ctx: ctx, + w: w, + types: makeTypeEncodeDatabase(), + zeroValues: make(map[reflect.Type]*objectEncodeState), } // Perform the encoding. - return es.safely(func() { - es.Serialize(reflect.ValueOf(rootPtr).Elem()) + err := safely(func() { + es.Save(reflect.ValueOf(rootPtr).Elem()) }) + return es.stats, err } // Load loads a checkpoint. -func Load(ctx context.Context, r io.Reader, rootPtr interface{}, stats *Stats) error { +func Load(ctx context.Context, r wire.Reader, rootPtr interface{}) (Stats, error) { // Create the decoding state. - ds := &decodeState{ - ctx: ctx, - objectsByID: make(map[uint64]*objectState), - deferred: make(map[uint64]*pb.Object), - r: r, - stats: stats, + ds := decodeState{ + ctx: ctx, + r: r, + types: makeTypeDecodeDatabase(), + deferred: make(map[objectID]wire.Object), } // Attempt our decode. - return ds.safely(func() { - ds.Deserialize(reflect.ValueOf(rootPtr).Elem()) + err := safely(func() { + ds.Load(reflect.ValueOf(rootPtr).Elem()) }) + return ds.stats, err } -// Fns are the state dispatch functions. -type Fns struct { - // Save is a function like Save(concreteType, Map). - Save interface{} - - // Load is a function like Load(concreteType, Map). - Load interface{} +// Sink is used for Type.StateSave. +type Sink struct { + internal objectEncoder } -// Save executes the save function. -func (fns *Fns) invokeSave(obj reflect.Value, m Map) { - reflect.ValueOf(fns.Save).Call([]reflect.Value{obj, reflect.ValueOf(m)}) +// Save adds the given object to the map. +// +// You should pass always pointers to the object you are saving. For example: +// +// type X struct { +// A int +// B *int +// } +// +// func (x *X) StateTypeInfo(m Sink) state.TypeInfo { +// return state.TypeInfo{ +// Name: "pkg.X", +// Fields: []string{ +// "A", +// "B", +// }, +// } +// } +// +// func (x *X) StateSave(m Sink) { +// m.Save(0, &x.A) // Field is A. +// m.Save(1, &x.B) // Field is B. +// } +// +// func (x *X) StateLoad(m Source) { +// m.Load(0, &x.A) // Field is A. +// m.Load(1, &x.B) // Field is B. +// } +func (s Sink) Save(slot int, objPtr interface{}) { + s.internal.save(slot, reflect.ValueOf(objPtr).Elem()) } -// Load executes the load function. -func (fns *Fns) invokeLoad(obj reflect.Value, m Map) { - reflect.ValueOf(fns.Load).Call([]reflect.Value{obj, reflect.ValueOf(m)}) +// SaveValue adds the given object value to the map. +// +// This should be used for values where pointers are not available, or casts +// are required during Save/Load. +// +// For example, if we want to cast external package type P.Foo to int64: +// +// func (x *X) StateSave(m Sink) { +// m.SaveValue(0, "A", int64(x.A)) +// } +// +// func (x *X) StateLoad(m Source) { +// m.LoadValue(0, new(int64), func(x interface{}) { +// x.A = P.Foo(x.(int64)) +// }) +// } +func (s Sink) SaveValue(slot int, obj interface{}) { + s.internal.save(slot, reflect.ValueOf(obj)) } -// validateStateFn ensures types are correct. -func validateStateFn(fn interface{}, typ reflect.Type) bool { - fnTyp := reflect.TypeOf(fn) - if fnTyp.Kind() != reflect.Func { - return false - } - if fnTyp.NumIn() != 2 { - return false - } - if fnTyp.NumOut() != 0 { - return false - } - if fnTyp.In(0) != typ { - return false - } - if fnTyp.In(1) != reflect.TypeOf(Map{}) { - return false - } - return true +// Context returns the context object provided at save time. +func (s Sink) Context() context.Context { + return s.internal.es.ctx } -// Validate validates all state functions. -func (fns *Fns) Validate(typ reflect.Type) bool { - return validateStateFn(fns.Save, typ) && validateStateFn(fns.Load, typ) +// Type is an interface that must be implemented by Struct objects. This allows +// these objects to be serialized while minimizing runtime reflection required. +// +// All these methods can be automatically generated by the go_statify tool. +type Type interface { + // StateTypeName returns the type's name. + // + // This is used for matching type information during encoding and + // decoding, as well as dynamic interface dispatch. This should be + // globally unique. + StateTypeName() string + + // StateFields returns information about the type. + // + // Fields is the set of fields for the object. Calls to Sink.Save and + // Source.Load must be made in-order with respect to these fields. + // + // This will be called at most once per serialization. + StateFields() []string } -type typeDatabase struct { - // nameToType is a forward lookup table. - nameToType map[string]reflect.Type - - // typeToName is the reverse lookup table. - typeToName map[reflect.Type]string +// SaverLoader must be implemented by struct types. +type SaverLoader interface { + // StateSave saves the state of the object to the given Map. + StateSave(Sink) - // typeToFns is the function lookup table. - typeToFns map[reflect.Type]Fns + // StateLoad loads the state of the object. + StateLoad(Source) } -// registeredTypes is a database used for SaveInterface and LoadInterface. -var registeredTypes = typeDatabase{ - nameToType: make(map[string]reflect.Type), - typeToName: make(map[reflect.Type]string), - typeToFns: make(map[reflect.Type]Fns), +// Source is used for Type.StateLoad. +type Source struct { + internal objectDecoder } -// register registers a type under the given name. This will generally be -// called via init() methods, and therefore uses panic to propagate errors. -func (t *typeDatabase) register(name string, typ reflect.Type, fns Fns) { - // We can't allow name collisions. - if ot, ok := t.nameToType[name]; ok { - panic(fmt.Sprintf("type %q can't use name %q, already in use by type %q", typ.Name(), name, ot.Name())) - } - - // Or multiple registrations. - if on, ok := t.typeToName[typ]; ok { - panic(fmt.Sprintf("type %q can't be registered as %q, already registered as %q", typ.Name(), name, on)) - } - - t.nameToType[name] = typ - t.typeToName[typ] = name - t.typeToFns[typ] = fns +// Load loads the given object passed as a pointer.. +// +// See Sink.Save for an example. +func (s Source) Load(slot int, objPtr interface{}) { + s.internal.load(slot, reflect.ValueOf(objPtr), false, nil) } -// lookupType finds a type given a name. -func (t *typeDatabase) lookupType(name string) (reflect.Type, bool) { - typ, ok := t.nameToType[name] - return typ, ok +// LoadWait loads the given objects from the map, and marks it as requiring all +// AfterLoad executions to complete prior to running this object's AfterLoad. +// +// See Sink.Save for an example. +func (s Source) LoadWait(slot int, objPtr interface{}) { + s.internal.load(slot, reflect.ValueOf(objPtr), true, nil) } -// lookupName finds a name given a type. -func (t *typeDatabase) lookupName(typ reflect.Type) (string, bool) { - name, ok := t.typeToName[typ] - return name, ok +// LoadValue loads the given object value from the map. +// +// See Sink.SaveValue for an example. +func (s Source) LoadValue(slot int, objPtr interface{}, fn func(interface{})) { + o := reflect.ValueOf(objPtr) + s.internal.load(slot, o, true, func() { fn(o.Elem().Interface()) }) } -// lookupFns finds functions given a type. -func (t *typeDatabase) lookupFns(typ reflect.Type) (Fns, bool) { - fns, ok := t.typeToFns[typ] - return fns, ok +// AfterLoad schedules a function execution when all objects have been +// allocated and their automated loading and customized load logic have been +// executed. fn will not be executed until all of current object's +// dependencies' AfterLoad() logic, if exist, have been executed. +func (s Source) AfterLoad(fn func()) { + s.internal.afterLoad(fn) } -// Register must be called for any interface implementation types that -// implements Loader. -// -// Register should be called either immediately after startup or via init() -// methods. Double registration of either names or types will result in a panic. -// -// No synchronization is provided; this should only be called in init. -// -// Example usage: -// -// state.Register("Foo", (*Foo)(nil), state.Fns{ -// Save: (*Foo).Save, -// Load: (*Foo).Load, -// }) -// -func Register(name string, instance interface{}, fns Fns) { - registeredTypes.register(name, reflect.TypeOf(instance), fns) +// Context returns the context object provided at load time. +func (s Source) Context() context.Context { + return s.internal.ds.ctx } // IsZeroValue checks if the given value is the zero value. @@ -244,72 +260,14 @@ func IsZeroValue(val interface{}) bool { return val == nil || reflect.ValueOf(val).Elem().IsZero() } -// step captures one encoding / decoding step. On each step, there is up to one -// choice made, which is captured by non-nil param. We intentionally do not -// eagerly create the final path string, as that will only be needed upon panic. -type step struct { - // dereference indicate if the current object is obtained by - // dereferencing a pointer. - dereference bool - - // format is the formatting string that takes param below, if - // non-nil. For example, in array indexing case, we have "[%d]". - format string - - // param stores the choice made at the current encoding / decoding step. - // For eaxmple, in array indexing case, param stores the index. When no - // choice is made, e.g. dereference, param should be nil. - param interface{} -} - -// recoverable is the state encoding / decoding panic recovery facility. It is -// also used to store encoding / decoding steps as well as the reference to the -// original queued object from which the current object is dispatched. The -// complete encoding / decoding path is synthesised from the steps in all queued -// objects leading to the current object. -type recoverable struct { - from *recoverable - steps []step +// Failf is a wrapper around panic that should be used to generate errors that +// can be caught during saving and loading. +func Failf(fmtStr string, v ...interface{}) { + panic(fmt.Errorf(fmtStr, v...)) } -// push enters a new context level. -func (sr *recoverable) push(dereference bool, format string, param interface{}) { - sr.steps = append(sr.steps, step{dereference, format, param}) -} - -// pop exits the current context level. -func (sr *recoverable) pop() { - if len(sr.steps) <= 1 { - return - } - sr.steps = sr.steps[:len(sr.steps)-1] -} - -// path returns the complete encoding / decoding path from root. This is only -// called upon panic. -func (sr *recoverable) path() string { - if sr.from == nil { - return "root" - } - p := sr.from.path() - for _, s := range sr.steps { - if s.dereference { - p = fmt.Sprintf("*(%s)", p) - } - if s.param == nil { - p += s.format - } else { - p += fmt.Sprintf(s.format, s.param) - } - } - return p -} - -func (sr *recoverable) copy() recoverable { - return recoverable{from: sr.from, steps: append([]step(nil), sr.steps...)} -} - -// safely executes the given function, catching a panic and unpacking as an error. +// safely executes the given function, catching a panic and unpacking as an +// error. // // The error flow through the state package uses panic and recover. There are // two important reasons for this: @@ -323,9 +281,15 @@ func (sr *recoverable) copy() recoverable { // method doesn't add a lot of value. If there are specific error conditions // that you'd like to handle, you should add appropriate functionality to // objects themselves prior to calling Save() and Load(). -func (sr *recoverable) safely(fn func()) (err error) { +func safely(fn func()) (err error) { defer func() { if r := recover(); r != nil { + if es, ok := r.(*ErrState); ok { + err = es // Propagate. + return + } + + // Build a new state error. es := new(ErrState) if e, ok := r.(error); ok { es.err = e @@ -333,8 +297,6 @@ func (sr *recoverable) safely(fn func()) (err error) { es.err = fmt.Errorf("%v", r) } - es.path = sr.path() - // Make a stack. We don't know how big it will be ahead // of time, but want to make sure we get the whole // thing. So we just do a stupid brute force approach. diff --git a/pkg/state/state_norace.go b/pkg/state/state_norace.go new file mode 100644 index 000000000..4281aed6d --- /dev/null +++ b/pkg/state/state_norace.go @@ -0,0 +1,19 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build !race + +package state + +var raceEnabled = false diff --git a/pkg/state/state_race.go b/pkg/state/state_race.go new file mode 100644 index 000000000..8232981ce --- /dev/null +++ b/pkg/state/state_race.go @@ -0,0 +1,19 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build race + +package state + +var raceEnabled = true diff --git a/pkg/state/state_test.go b/pkg/state/state_test.go deleted file mode 100644 index d7221e9e8..000000000 --- a/pkg/state/state_test.go +++ /dev/null @@ -1,721 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package state - -import ( - "bytes" - "context" - "io/ioutil" - "math" - "reflect" - "testing" -) - -// TestCase is used to define a single success/failure testcase of -// serialization of a set of objects. -type TestCase struct { - // Name is the name of the test case. - Name string - - // Objects is the list of values to serialize. - Objects []interface{} - - // Fail is whether the test case is supposed to fail or not. - Fail bool -} - -// runTest runs all testcases. -func runTest(t *testing.T, tests []TestCase) { - for _, test := range tests { - t.Logf("TEST %s:", test.Name) - for i, root := range test.Objects { - t.Logf(" case#%d: %#v", i, root) - - // Save the passed object. - saveBuffer := &bytes.Buffer{} - saveObjectPtr := reflect.New(reflect.TypeOf(root)) - saveObjectPtr.Elem().Set(reflect.ValueOf(root)) - if err := Save(context.Background(), saveBuffer, saveObjectPtr.Interface(), nil); err != nil && !test.Fail { - t.Errorf(" FAIL: Save failed unexpectedly: %v", err) - continue - } else if err != nil { - t.Logf(" PASS: Save failed as expected: %v", err) - continue - } - - // Load a new copy of the object. - loadObjectPtr := reflect.New(reflect.TypeOf(root)) - if err := Load(context.Background(), bytes.NewReader(saveBuffer.Bytes()), loadObjectPtr.Interface(), nil); err != nil && !test.Fail { - t.Errorf(" FAIL: Load failed unexpectedly: %v", err) - continue - } else if err != nil { - t.Logf(" PASS: Load failed as expected: %v", err) - continue - } - - // Compare the values. - loadedValue := loadObjectPtr.Elem().Interface() - if eq := reflect.DeepEqual(root, loadedValue); !eq && !test.Fail { - t.Errorf(" FAIL: Objects differs; got %#v", loadedValue) - continue - } else if !eq { - t.Logf(" PASS: Object different as expected.") - continue - } - - // Everything went okay. Is that good? - if test.Fail { - t.Errorf(" FAIL: Unexpected success.") - } else { - t.Logf(" PASS: Success.") - } - } - } -} - -// dumbStruct is a struct which does not implement the loader/saver interface. -// We expect that serialization of this struct will fail. -type dumbStruct struct { - A int - B int -} - -// smartStruct is a struct which does implement the loader/saver interface. -// We expect that serialization of this struct will succeed. -type smartStruct struct { - A int - B int -} - -func (s *smartStruct) save(m Map) { - m.Save("A", &s.A) - m.Save("B", &s.B) -} - -func (s *smartStruct) load(m Map) { - m.Load("A", &s.A) - m.Load("B", &s.B) -} - -// valueLoadStruct uses a value load. -type valueLoadStruct struct { - v int -} - -func (v *valueLoadStruct) save(m Map) { - m.SaveValue("v", v.v) -} - -func (v *valueLoadStruct) load(m Map) { - m.LoadValue("v", new(int), func(value interface{}) { - v.v = value.(int) - }) -} - -// afterLoadStruct has an AfterLoad function. -type afterLoadStruct struct { - v int -} - -func (a *afterLoadStruct) save(m Map) { -} - -func (a *afterLoadStruct) load(m Map) { - m.AfterLoad(func() { - a.v++ - }) -} - -// genericContainer is a generic dispatcher. -type genericContainer struct { - v interface{} -} - -func (g *genericContainer) save(m Map) { - m.Save("v", &g.v) -} - -func (g *genericContainer) load(m Map) { - m.Load("v", &g.v) -} - -// sliceContainer is a generic slice. -type sliceContainer struct { - v []interface{} -} - -func (s *sliceContainer) save(m Map) { - m.Save("v", &s.v) -} - -func (s *sliceContainer) load(m Map) { - m.Load("v", &s.v) -} - -// mapContainer is a generic map. -type mapContainer struct { - v map[int]interface{} -} - -func (mc *mapContainer) save(m Map) { - m.Save("v", &mc.v) -} - -func (mc *mapContainer) load(m Map) { - // Some of the test cases below assume legacy behavior wherein maps - // will automatically inherit dependencies. - m.LoadWait("v", &mc.v) -} - -// dumbMap is a map which does not implement the loader/saver interface. -// Serialization of this map will default to the standard encode/decode logic. -type dumbMap map[string]int - -// pointerStruct contains various pointers, shared and non-shared, and pointers -// to pointers. We expect that serialization will respect the structure. -type pointerStruct struct { - A *int - B *int - C *int - D *int - - AA **int - BB **int -} - -func (p *pointerStruct) save(m Map) { - m.Save("A", &p.A) - m.Save("B", &p.B) - m.Save("C", &p.C) - m.Save("D", &p.D) - m.Save("AA", &p.AA) - m.Save("BB", &p.BB) -} - -func (p *pointerStruct) load(m Map) { - m.Load("A", &p.A) - m.Load("B", &p.B) - m.Load("C", &p.C) - m.Load("D", &p.D) - m.Load("AA", &p.AA) - m.Load("BB", &p.BB) -} - -// testInterface is a trivial interface example. -type testInterface interface { - Foo() -} - -// testImpl is a trivial implementation of testInterface. -type testImpl struct { -} - -// Foo satisfies testInterface. -func (t *testImpl) Foo() { -} - -// testImpl is trivially serializable. -func (t *testImpl) save(m Map) { -} - -// testImpl is trivially serializable. -func (t *testImpl) load(m Map) { -} - -// testI demonstrates interface dispatching. -type testI struct { - I testInterface -} - -func (t *testI) save(m Map) { - m.Save("I", &t.I) -} - -func (t *testI) load(m Map) { - m.Load("I", &t.I) -} - -// cycleStruct is used to implement basic cycles. -type cycleStruct struct { - c *cycleStruct -} - -func (c *cycleStruct) save(m Map) { - m.Save("c", &c.c) -} - -func (c *cycleStruct) load(m Map) { - m.Load("c", &c.c) -} - -// badCycleStruct actually has deadlocking dependencies. -// -// This should pass if b.b = {nil|b} and fail otherwise. -type badCycleStruct struct { - b *badCycleStruct -} - -func (b *badCycleStruct) save(m Map) { - m.Save("b", &b.b) -} - -func (b *badCycleStruct) load(m Map) { - m.LoadWait("b", &b.b) - m.AfterLoad(func() { - // This is not executable, since AfterLoad requires that the - // object and all dependencies are complete. This should cause - // a deadlock error during load. - }) -} - -// emptyStructPointer points to an empty struct. -type emptyStructPointer struct { - nothing *struct{} -} - -func (e *emptyStructPointer) save(m Map) { - m.Save("nothing", &e.nothing) -} - -func (e *emptyStructPointer) load(m Map) { - m.Load("nothing", &e.nothing) -} - -// truncateInteger truncates an integer. -type truncateInteger struct { - v int64 - v2 int32 -} - -func (t *truncateInteger) save(m Map) { - t.v2 = int32(t.v) - m.Save("v", &t.v) -} - -func (t *truncateInteger) load(m Map) { - m.Load("v", &t.v2) - t.v = int64(t.v2) -} - -// truncateUnsignedInteger truncates an unsigned integer. -type truncateUnsignedInteger struct { - v uint64 - v2 uint32 -} - -func (t *truncateUnsignedInteger) save(m Map) { - t.v2 = uint32(t.v) - m.Save("v", &t.v) -} - -func (t *truncateUnsignedInteger) load(m Map) { - m.Load("v", &t.v2) - t.v = uint64(t.v2) -} - -// truncateFloat truncates a floating point number. -type truncateFloat struct { - v float64 - v2 float32 -} - -func (t *truncateFloat) save(m Map) { - t.v2 = float32(t.v) - m.Save("v", &t.v) -} - -func (t *truncateFloat) load(m Map) { - m.Load("v", &t.v2) - t.v = float64(t.v2) -} - -func TestTypes(t *testing.T) { - // x and y are basic integers, while xp points to x. - x := 1 - y := 2 - xp := &x - - // cs is a single object cycle. - cs := cycleStruct{nil} - cs.c = &cs - - // cs1 and cs2 are in a two object cycle. - cs1 := cycleStruct{nil} - cs2 := cycleStruct{nil} - cs1.c = &cs2 - cs2.c = &cs1 - - // bs is a single object cycle. - bs := badCycleStruct{nil} - bs.b = &bs - - // bs2 and bs2 are in a deadlocking cycle. - bs1 := badCycleStruct{nil} - bs2 := badCycleStruct{nil} - bs1.b = &bs2 - bs2.b = &bs1 - - // regular nils. - var ( - nilmap dumbMap - nilslice []byte - ) - - // embed points to embedded fields. - embed1 := pointerStruct{} - embed1.AA = &embed1.A - embed2 := pointerStruct{} - embed2.BB = &embed2.B - - // es1 contains two structs pointing to the same empty struct. - es := emptyStructPointer{new(struct{})} - es1 := []emptyStructPointer{es, es} - - tests := []TestCase{ - { - Name: "bool", - Objects: []interface{}{ - true, - false, - }, - }, - { - Name: "integers", - Objects: []interface{}{ - int(0), - int(1), - int(-1), - int8(0), - int8(1), - int8(-1), - int16(0), - int16(1), - int16(-1), - int32(0), - int32(1), - int32(-1), - int64(0), - int64(1), - int64(-1), - }, - }, - { - Name: "unsigned integers", - Objects: []interface{}{ - uint(0), - uint(1), - uint8(0), - uint8(1), - uint16(0), - uint16(1), - uint32(1), - uint64(0), - uint64(1), - }, - }, - { - Name: "strings", - Objects: []interface{}{ - "", - "foo", - "bar", - "\xa0", - }, - }, - { - Name: "slices", - Objects: []interface{}{ - []int{-1, 0, 1}, - []*int{&x, &x, &x}, - []int{1, 2, 3}[0:1], - []int{1, 2, 3}[1:2], - make([]byte, 32), - make([]byte, 32)[:16], - make([]byte, 32)[:16:20], - nilslice, - }, - }, - { - Name: "arrays", - Objects: []interface{}{ - &[1048576]bool{false, true, false, true}, - &[1048576]uint8{0, 1, 2, 3}, - &[1048576]byte{0, 1, 2, 3}, - &[1048576]uint16{0, 1, 2, 3}, - &[1048576]uint{0, 1, 2, 3}, - &[1048576]uint32{0, 1, 2, 3}, - &[1048576]uint64{0, 1, 2, 3}, - &[1048576]uintptr{0, 1, 2, 3}, - &[1048576]int8{0, -1, -2, -3}, - &[1048576]int16{0, -1, -2, -3}, - &[1048576]int32{0, -1, -2, -3}, - &[1048576]int64{0, -1, -2, -3}, - &[1048576]float32{0, 1.1, 2.2, 3.3}, - &[1048576]float64{0, 1.1, 2.2, 3.3}, - }, - }, - { - Name: "pointers", - Objects: []interface{}{ - &pointerStruct{A: &x, B: &x, C: &y, D: &y, AA: &xp, BB: &xp}, - &pointerStruct{}, - }, - }, - { - Name: "empty struct", - Objects: []interface{}{ - struct{}{}, - }, - }, - { - Name: "unenlightened structs", - Objects: []interface{}{ - &dumbStruct{A: 1, B: 2}, - }, - Fail: true, - }, - { - Name: "enlightened structs", - Objects: []interface{}{ - &smartStruct{A: 1, B: 2}, - }, - }, - { - Name: "load-hooks", - Objects: []interface{}{ - &afterLoadStruct{v: 1}, - &valueLoadStruct{v: 1}, - &genericContainer{v: &afterLoadStruct{v: 1}}, - &genericContainer{v: &valueLoadStruct{v: 1}}, - &sliceContainer{v: []interface{}{&afterLoadStruct{v: 1}}}, - &sliceContainer{v: []interface{}{&valueLoadStruct{v: 1}}}, - &mapContainer{v: map[int]interface{}{0: &afterLoadStruct{v: 1}}}, - &mapContainer{v: map[int]interface{}{0: &valueLoadStruct{v: 1}}}, - }, - }, - { - Name: "maps", - Objects: []interface{}{ - dumbMap{"a": -1, "b": 0, "c": 1}, - map[smartStruct]int{{}: 0, {A: 1}: 1}, - nilmap, - &mapContainer{v: map[int]interface{}{0: &smartStruct{A: 1}}}, - }, - }, - { - Name: "interfaces", - Objects: []interface{}{ - &testI{&testImpl{}}, - &testI{nil}, - &testI{(*testImpl)(nil)}, - }, - }, - { - Name: "unregistered-interfaces", - Objects: []interface{}{ - &genericContainer{v: afterLoadStruct{v: 1}}, - &genericContainer{v: valueLoadStruct{v: 1}}, - &sliceContainer{v: []interface{}{afterLoadStruct{v: 1}}}, - &sliceContainer{v: []interface{}{valueLoadStruct{v: 1}}}, - &mapContainer{v: map[int]interface{}{0: afterLoadStruct{v: 1}}}, - &mapContainer{v: map[int]interface{}{0: valueLoadStruct{v: 1}}}, - }, - Fail: true, - }, - { - Name: "cycles", - Objects: []interface{}{ - &cs, - &cs1, - &cycleStruct{&cs1}, - &cycleStruct{&cs}, - &badCycleStruct{nil}, - &bs, - }, - }, - { - Name: "deadlock", - Objects: []interface{}{ - &bs1, - }, - Fail: true, - }, - { - Name: "embed", - Objects: []interface{}{ - &embed1, - &embed2, - }, - Fail: true, - }, - { - Name: "empty structs", - Objects: []interface{}{ - new(struct{}), - es, - es1, - }, - }, - { - Name: "truncated okay", - Objects: []interface{}{ - &truncateInteger{v: 1}, - &truncateUnsignedInteger{v: 1}, - &truncateFloat{v: 1.0}, - }, - }, - { - Name: "truncated bad", - Objects: []interface{}{ - &truncateInteger{v: math.MaxInt32 + 1}, - &truncateUnsignedInteger{v: math.MaxUint32 + 1}, - &truncateFloat{v: math.MaxFloat32 * 2}, - }, - Fail: true, - }, - } - - runTest(t, tests) -} - -// benchStruct is used for benchmarking. -type benchStruct struct { - b *benchStruct - - // Dummy data is included to ensure that these objects are large. - // This is to detect possible regression when registering objects. - _ [4096]byte -} - -func (b *benchStruct) save(m Map) { - m.Save("b", &b.b) -} - -func (b *benchStruct) load(m Map) { - m.LoadWait("b", &b.b) - m.AfterLoad(b.afterLoad) -} - -func (b *benchStruct) afterLoad() { - // Do nothing, just force scheduling. -} - -// buildObject builds a benchmark object. -func buildObject(n int) (b *benchStruct) { - for i := 0; i < n; i++ { - b = &benchStruct{b: b} - } - return -} - -func BenchmarkEncoding(b *testing.B) { - b.StopTimer() - bs := buildObject(b.N) - var stats Stats - b.StartTimer() - if err := Save(context.Background(), ioutil.Discard, bs, &stats); err != nil { - b.Errorf("save failed: %v", err) - } - b.StopTimer() - if b.N > 1000 { - b.Logf("breakdown (n=%d): %s", b.N, &stats) - } -} - -func BenchmarkDecoding(b *testing.B) { - b.StopTimer() - bs := buildObject(b.N) - var newBS benchStruct - buf := &bytes.Buffer{} - if err := Save(context.Background(), buf, bs, nil); err != nil { - b.Errorf("save failed: %v", err) - } - var stats Stats - b.StartTimer() - if err := Load(context.Background(), buf, &newBS, &stats); err != nil { - b.Errorf("load failed: %v", err) - } - b.StopTimer() - if b.N > 1000 { - b.Logf("breakdown (n=%d): %s", b.N, &stats) - } -} - -func init() { - Register("stateTest.smartStruct", (*smartStruct)(nil), Fns{ - Save: (*smartStruct).save, - Load: (*smartStruct).load, - }) - Register("stateTest.afterLoadStruct", (*afterLoadStruct)(nil), Fns{ - Save: (*afterLoadStruct).save, - Load: (*afterLoadStruct).load, - }) - Register("stateTest.valueLoadStruct", (*valueLoadStruct)(nil), Fns{ - Save: (*valueLoadStruct).save, - Load: (*valueLoadStruct).load, - }) - Register("stateTest.genericContainer", (*genericContainer)(nil), Fns{ - Save: (*genericContainer).save, - Load: (*genericContainer).load, - }) - Register("stateTest.sliceContainer", (*sliceContainer)(nil), Fns{ - Save: (*sliceContainer).save, - Load: (*sliceContainer).load, - }) - Register("stateTest.mapContainer", (*mapContainer)(nil), Fns{ - Save: (*mapContainer).save, - Load: (*mapContainer).load, - }) - Register("stateTest.pointerStruct", (*pointerStruct)(nil), Fns{ - Save: (*pointerStruct).save, - Load: (*pointerStruct).load, - }) - Register("stateTest.testImpl", (*testImpl)(nil), Fns{ - Save: (*testImpl).save, - Load: (*testImpl).load, - }) - Register("stateTest.testI", (*testI)(nil), Fns{ - Save: (*testI).save, - Load: (*testI).load, - }) - Register("stateTest.cycleStruct", (*cycleStruct)(nil), Fns{ - Save: (*cycleStruct).save, - Load: (*cycleStruct).load, - }) - Register("stateTest.badCycleStruct", (*badCycleStruct)(nil), Fns{ - Save: (*badCycleStruct).save, - Load: (*badCycleStruct).load, - }) - Register("stateTest.emptyStructPointer", (*emptyStructPointer)(nil), Fns{ - Save: (*emptyStructPointer).save, - Load: (*emptyStructPointer).load, - }) - Register("stateTest.truncateInteger", (*truncateInteger)(nil), Fns{ - Save: (*truncateInteger).save, - Load: (*truncateInteger).load, - }) - Register("stateTest.truncateUnsignedInteger", (*truncateUnsignedInteger)(nil), Fns{ - Save: (*truncateUnsignedInteger).save, - Load: (*truncateUnsignedInteger).load, - }) - Register("stateTest.truncateFloat", (*truncateFloat)(nil), Fns{ - Save: (*truncateFloat).save, - Load: (*truncateFloat).load, - }) - Register("stateTest.benchStruct", (*benchStruct)(nil), Fns{ - Save: (*benchStruct).save, - Load: (*benchStruct).load, - }) -} diff --git a/pkg/state/statefile/BUILD b/pkg/state/statefile/BUILD index e7581c09b..d6c89c7e9 100644 --- a/pkg/state/statefile/BUILD +++ b/pkg/state/statefile/BUILD @@ -9,6 +9,7 @@ go_library( deps = [ "//pkg/binary", "//pkg/compressio", + "//pkg/state/wire", ], ) diff --git a/pkg/state/statefile/statefile.go b/pkg/state/statefile/statefile.go index c0f4c4954..bdfb800fb 100644 --- a/pkg/state/statefile/statefile.go +++ b/pkg/state/statefile/statefile.go @@ -57,6 +57,7 @@ import ( "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/compressio" + "gvisor.dev/gvisor/pkg/state/wire" ) // keySize is the AES-256 key length. @@ -83,10 +84,16 @@ var ErrInvalidMetadataLength = fmt.Errorf("metadata length invalid, maximum size // ErrMetadataInvalid is returned if passed metadata is invalid. var ErrMetadataInvalid = fmt.Errorf("metadata invalid, can't start with _") +// WriteCloser is an io.Closer and wire.Writer. +type WriteCloser interface { + wire.Writer + io.Closer +} + // NewWriter returns a state data writer for a statefile. // // Note that the returned WriteCloser must be closed. -func NewWriter(w io.Writer, key []byte, metadata map[string]string) (io.WriteCloser, error) { +func NewWriter(w io.Writer, key []byte, metadata map[string]string) (WriteCloser, error) { if metadata == nil { metadata = make(map[string]string) } @@ -215,7 +222,7 @@ func metadata(r io.Reader, h hash.Hash) (map[string]string, error) { } // NewReader returns a reader for a statefile. -func NewReader(r io.Reader, key []byte) (io.Reader, map[string]string, error) { +func NewReader(r io.Reader, key []byte) (wire.Reader, map[string]string, error) { // Read the metadata with the hash. h := hmac.New(sha256.New, key) metadata, err := metadata(r, h) @@ -224,9 +231,9 @@ func NewReader(r io.Reader, key []byte) (io.Reader, map[string]string, error) { } // Wrap in compression. - rc, err := compressio.NewReader(r, key) + cr, err := compressio.NewReader(r, key) if err != nil { return nil, nil, err } - return rc, metadata, nil + return cr, metadata, nil } diff --git a/pkg/state/stats.go b/pkg/state/stats.go index eb51cda47..eaec664a1 100644 --- a/pkg/state/stats.go +++ b/pkg/state/stats.go @@ -17,7 +17,6 @@ package state import ( "bytes" "fmt" - "reflect" "sort" "time" ) @@ -35,92 +34,81 @@ type statEntry struct { // All exported receivers accept nil. type Stats struct { // byType contains a breakdown of time spent by type. - byType map[reflect.Type]*statEntry + // + // This is indexed *directly* by typeID, including zero. + byType []statEntry // stack contains objects in progress. - stack []reflect.Type + stack []typeID + + // names contains type names. + // + // This is also indexed *directly* by typeID, including zero, which we + // hard-code as "state.default". This is only resolved by calling fini + // on the stats object. + names []string // last is the last start time. last time.Time } -// sample adds the samples to the given object. -func (s *Stats) sample(typ reflect.Type) { - now := time.Now() - s.byType[typ].total += now.Sub(s.last) - s.last = now +// init initializes statistics. +func (s *Stats) init() { + s.last = time.Now() + s.stack = append(s.stack, 0) } -// Add adds a sample count. -func (s *Stats) Add(obj reflect.Value) { - if s == nil { - return - } - if s.byType == nil { - s.byType = make(map[reflect.Type]*statEntry) - } - typ := obj.Type() - entry, ok := s.byType[typ] - if !ok { - entry = new(statEntry) - s.byType[typ] = entry +// fini finalizes statistics. +func (s *Stats) fini(resolve func(id typeID) string) { + s.done() + + // Resolve all type names. + s.names = make([]string, len(s.byType)) + s.names[0] = "state.default" // See above. + for id := typeID(1); int(id) < len(s.names); id++ { + s.names[id] = resolve(id) } - entry.count++ } -// Remove removes a sample count. It should only be called after a previous -// Add(). -func (s *Stats) Remove(obj reflect.Value) { - if s == nil { - return +// sample adds the samples to the given object. +func (s *Stats) sample(id typeID) { + now := time.Now() + if len(s.byType) <= int(id) { + // Allocate all the missing entries in one fell swoop. + s.byType = append(s.byType, make([]statEntry, 1+int(id)-len(s.byType))...) } - typ := obj.Type() - entry := s.byType[typ] - entry.count-- + s.byType[id].total += now.Sub(s.last) + s.last = now } -// Start starts a sample. -func (s *Stats) Start(obj reflect.Value) { - if s == nil { - return - } - if len(s.stack) > 0 { - last := s.stack[len(s.stack)-1] - s.sample(last) - } else { - // First time sample. - s.last = time.Now() - } - s.stack = append(s.stack, obj.Type()) +// start starts a sample. +func (s *Stats) start(id typeID) { + last := s.stack[len(s.stack)-1] + s.sample(last) + s.stack = append(s.stack, id) } -// Done finishes the current sample. -func (s *Stats) Done() { - if s == nil { - return - } +// done finishes the current sample. +func (s *Stats) done() { last := s.stack[len(s.stack)-1] s.sample(last) + s.byType[last].count++ s.stack = s.stack[:len(s.stack)-1] } type sliceEntry struct { - typ reflect.Type + name string entry *statEntry } // String returns a table representation of the stats. func (s *Stats) String() string { - if s == nil || len(s.byType) == 0 { - return "(no data)" - } - // Build a list of stat entries. ss := make([]sliceEntry, 0, len(s.byType)) - for typ, entry := range s.byType { + for id := 0; id < len(s.names); id++ { ss = append(ss, sliceEntry{ - typ: typ, - entry: entry, + name: s.names[id], + entry: &s.byType[id], }) } @@ -136,17 +124,22 @@ func (s *Stats) String() string { total time.Duration ) buf.WriteString("\n") - buf.WriteString(fmt.Sprintf("%12s | %8s | %8s | %s\n", "total", "count", "per", "type")) - buf.WriteString("-------------+----------+----------+-------------\n") + buf.WriteString(fmt.Sprintf("% 16s | % 8s | % 16s | %s\n", "total", "count", "per", "type")) + buf.WriteString("-----------------+----------+------------------+----------------\n") for _, se := range ss { + if se.entry.count == 0 { + // Since we store all types linearly, we are not + // guaranteed that any entry actually has time. + continue + } count += se.entry.count total += se.entry.total per := se.entry.total / time.Duration(se.entry.count) - buf.WriteString(fmt.Sprintf("%12s | %8d | %8s | %s\n", - se.entry.total, se.entry.count, per, se.typ.String())) + buf.WriteString(fmt.Sprintf("% 16s | %8d | % 16s | %s\n", + se.entry.total, se.entry.count, per, se.name)) } - buf.WriteString("-------------+----------+----------+-------------\n") - buf.WriteString(fmt.Sprintf("%12s | %8d | %8s | [all]", + buf.WriteString("-----------------+----------+------------------+----------------\n") + buf.WriteString(fmt.Sprintf("% 16s | % 8d | % 16s | [all]", total, count, total/time.Duration(count))) return string(buf.Bytes()) } diff --git a/pkg/state/tests/BUILD b/pkg/state/tests/BUILD new file mode 100644 index 000000000..9297cafbe --- /dev/null +++ b/pkg/state/tests/BUILD @@ -0,0 +1,43 @@ +load("//tools:defs.bzl", "go_library", "go_test") + +package(licenses = ["notice"]) + +go_library( + name = "tests", + srcs = [ + "array.go", + "bench.go", + "integer.go", + "load.go", + "map.go", + "register.go", + "struct.go", + "tests.go", + ], + deps = [ + "//pkg/state", + "//pkg/state/pretty", + ], +) + +go_test( + name = "tests_test", + size = "small", + srcs = [ + "array_test.go", + "bench_test.go", + "bool_test.go", + "float_test.go", + "integer_test.go", + "load_test.go", + "map_test.go", + "register_test.go", + "string_test.go", + "struct_test.go", + ], + library = ":tests", + deps = [ + "//pkg/state", + "//pkg/state/wire", + ], +) diff --git a/pkg/state/tests/array.go b/pkg/state/tests/array.go new file mode 100644 index 000000000..0972a80e7 --- /dev/null +++ b/pkg/state/tests/array.go @@ -0,0 +1,35 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tests + +// +stateify savable +type arrayContainer struct { + v [1]interface{} +} + +// +stateify savable +type arrayPtrContainer struct { + v *[1]interface{} +} + +// +stateify savable +type sliceContainer struct { + v []interface{} +} + +// +stateify savable +type slicePtrContainer struct { + v *[]interface{} +} diff --git a/pkg/state/tests/array_test.go b/pkg/state/tests/array_test.go new file mode 100644 index 000000000..a347b2947 --- /dev/null +++ b/pkg/state/tests/array_test.go @@ -0,0 +1,134 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tests + +import ( + "reflect" + "testing" +) + +var allArrayPrimitives = []interface{}{ + [1]bool{}, + [1]bool{true}, + [2]bool{false, true}, + [1]int{}, + [1]int{1}, + [2]int{0, 1}, + [1]int8{}, + [1]int8{1}, + [2]int8{0, 1}, + [1]int16{}, + [1]int16{1}, + [2]int16{0, 1}, + [1]int32{}, + [1]int32{1}, + [2]int32{0, 1}, + [1]int64{}, + [1]int64{1}, + [2]int64{0, 1}, + [1]uint{}, + [1]uint{1}, + [2]uint{0, 1}, + [1]uintptr{}, + [1]uintptr{1}, + [2]uintptr{0, 1}, + [1]uint8{}, + [1]uint8{1}, + [2]uint8{0, 1}, + [1]uint16{}, + [1]uint16{1}, + [2]uint16{0, 1}, + [1]uint32{}, + [1]uint32{1}, + [2]uint32{0, 1}, + [1]uint64{}, + [1]uint64{1}, + [2]uint64{0, 1}, + [1]string{}, + [1]string{""}, + [1]string{nonEmptyString}, + [2]string{"", nonEmptyString}, +} + +func TestArrayPrimitives(t *testing.T) { + runTestCases(t, false, "plain", flatten(allArrayPrimitives)) + runTestCases(t, false, "pointers", pointersTo(flatten(allArrayPrimitives))) + runTestCases(t, false, "interfaces", interfacesTo(flatten(allArrayPrimitives))) + runTestCases(t, false, "interfacesToPointers", interfacesTo(pointersTo(flatten(allArrayPrimitives)))) +} + +func TestSlices(t *testing.T) { + var allSlices = flatten( + filter(allArrayPrimitives, func(o interface{}) (interface{}, bool) { + v := reflect.New(reflect.TypeOf(o)).Elem() + v.Set(reflect.ValueOf(o)) + return v.Slice(0, v.Len()).Interface(), true + }), + filter(allArrayPrimitives, func(o interface{}) (interface{}, bool) { + v := reflect.New(reflect.TypeOf(o)).Elem() + v.Set(reflect.ValueOf(o)) + if v.Len() == 0 { + // Return the pure "nil" value for the slice. + return reflect.New(v.Slice(0, 0).Type()).Elem().Interface(), true + } + return v.Slice(1, v.Len()).Interface(), true + }), + filter(allArrayPrimitives, func(o interface{}) (interface{}, bool) { + v := reflect.New(reflect.TypeOf(o)).Elem() + v.Set(reflect.ValueOf(o)) + if v.Len() == 0 { + // Return the zero-valued slice. + return reflect.MakeSlice(v.Slice(0, 0).Type(), 0, 0).Interface(), true + } + return v.Slice(0, v.Len()-1).Interface(), true + }), + ) + runTestCases(t, false, "plain", allSlices) + runTestCases(t, false, "pointers", pointersTo(allSlices)) + runTestCases(t, false, "interfaces", interfacesTo(allSlices)) + runTestCases(t, false, "interfacesToPointers", interfacesTo(pointersTo(allSlices))) +} + +func TestArrayContainers(t *testing.T) { + var ( + emptyArray [1]interface{} + fullArray [1]interface{} + ) + fullArray[0] = &emptyArray + runTestCases(t, false, "", []interface{}{ + arrayContainer{v: emptyArray}, + arrayContainer{v: fullArray}, + arrayPtrContainer{v: nil}, + arrayPtrContainer{v: &emptyArray}, + arrayPtrContainer{v: &fullArray}, + }) +} + +func TestSliceContainers(t *testing.T) { + var ( + nilSlice []interface{} + emptySlice = make([]interface{}, 0) + fullSlice = []interface{}{nil} + ) + runTestCases(t, false, "", []interface{}{ + sliceContainer{v: nilSlice}, + sliceContainer{v: emptySlice}, + sliceContainer{v: fullSlice}, + slicePtrContainer{v: nil}, + slicePtrContainer{v: &nilSlice}, + slicePtrContainer{v: &emptySlice}, + slicePtrContainer{v: &fullSlice}, + }) +} diff --git a/pkg/state/tests/bench.go b/pkg/state/tests/bench.go new file mode 100644 index 000000000..40869cdfb --- /dev/null +++ b/pkg/state/tests/bench.go @@ -0,0 +1,24 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tests + +// +stateify savable +type benchStruct struct { + B *benchStruct // Must be exported for gob. +} + +func (b *benchStruct) afterLoad() { + // Do nothing, just force scheduling. +} diff --git a/pkg/state/tests/bench_test.go b/pkg/state/tests/bench_test.go new file mode 100644 index 000000000..7e102c907 --- /dev/null +++ b/pkg/state/tests/bench_test.go @@ -0,0 +1,153 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tests + +import ( + "bytes" + "context" + "encoding/gob" + "fmt" + "testing" + + "gvisor.dev/gvisor/pkg/state" + "gvisor.dev/gvisor/pkg/state/wire" +) + +// buildPtrObject builds a benchmark object. +func buildPtrObject(n int) interface{} { + b := new(benchStruct) + for i := 0; i < n; i++ { + b = &benchStruct{B: b} + } + return b +} + +// buildMapObject builds a benchmark object. +func buildMapObject(n int) interface{} { + b := new(benchStruct) + m := make(map[int]*benchStruct) + for i := 0; i < n; i++ { + m[i] = b + } + return &m +} + +// buildSliceObject builds a benchmark object. +func buildSliceObject(n int) interface{} { + b := new(benchStruct) + s := make([]*benchStruct, 0, n) + for i := 0; i < n; i++ { + s = append(s, b) + } + return &s +} + +var allObjects = map[string]struct { + New func(int) interface{} +}{ + "ptr": { + New: buildPtrObject, + }, + "map": { + New: buildMapObject, + }, + "slice": { + New: buildSliceObject, + }, +} + +func buildObjects(n int, fn func(int) interface{}) (iters int, v interface{}) { + // maxSize is the maximum size of an individual object below. For an N + // larger than this, we start to return multiple objects. + const maxSize = 1024 + if n <= maxSize { + return 1, fn(n) + } + iters = (n + maxSize - 1) / maxSize + return iters, fn(maxSize) +} + +// gobSave is a version of save using gob (no stats available). +func gobSave(_ context.Context, w wire.Writer, v interface{}) (_ state.Stats, err error) { + enc := gob.NewEncoder(w) + err = enc.Encode(v) + return +} + +// gobLoad is a version of load using gob (no stats available). +func gobLoad(_ context.Context, r wire.Reader, v interface{}) (_ state.Stats, err error) { + dec := gob.NewDecoder(r) + err = dec.Decode(v) + return +} + +var allAlgos = map[string]struct { + Save func(context.Context, wire.Writer, interface{}) (state.Stats, error) + Load func(context.Context, wire.Reader, interface{}) (state.Stats, error) + MaxPtr int +}{ + "state": { + Save: state.Save, + Load: state.Load, + }, + "gob": { + Save: gobSave, + Load: gobLoad, + }, +} + +func BenchmarkEncoding(b *testing.B) { + for objName, objInfo := range allObjects { + for algoName, algoInfo := range allAlgos { + b.Run(fmt.Sprintf("%s/%s", objName, algoName), func(b *testing.B) { + b.StopTimer() + n, v := buildObjects(b.N, objInfo.New) + b.ReportAllocs() + b.StartTimer() + for i := 0; i < n; i++ { + if _, err := algoInfo.Save(context.Background(), discard{}, v); err != nil { + b.Errorf("save failed: %v", err) + } + } + b.StopTimer() + }) + } + } +} + +func BenchmarkDecoding(b *testing.B) { + for objName, objInfo := range allObjects { + for algoName, algoInfo := range allAlgos { + b.Run(fmt.Sprintf("%s/%s", objName, algoName), func(b *testing.B) { + b.StopTimer() + n, v := buildObjects(b.N, objInfo.New) + buf := new(bytes.Buffer) + if _, err := algoInfo.Save(context.Background(), buf, v); err != nil { + b.Errorf("save failed: %v", err) + } + b.ReportAllocs() + b.StartTimer() + var r bytes.Reader + for i := 0; i < n; i++ { + r.Reset(buf.Bytes()) + if _, err := algoInfo.Load(context.Background(), &r, v); err != nil { + b.Errorf("load failed: %v", err) + } + } + b.StopTimer() + }) + } + } +} diff --git a/pkg/state/tests/bool_test.go b/pkg/state/tests/bool_test.go new file mode 100644 index 000000000..e17cfacf9 --- /dev/null +++ b/pkg/state/tests/bool_test.go @@ -0,0 +1,31 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tests + +import ( + "testing" +) + +var allBools = []bool{ + true, + false, +} + +func TestBool(t *testing.T) { + runTestCases(t, false, "plain", flatten(allBools)) + runTestCases(t, false, "pointers", pointersTo(flatten(allBools))) + runTestCases(t, false, "interfaces", interfacesTo(flatten(allBools))) + runTestCases(t, false, "interfacesToPointers", interfacesTo(pointersTo(flatten(allBools)))) +} diff --git a/pkg/state/tests/float_test.go b/pkg/state/tests/float_test.go new file mode 100644 index 000000000..3e89edd9c --- /dev/null +++ b/pkg/state/tests/float_test.go @@ -0,0 +1,118 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tests + +import ( + "math" + "testing" +) + +var safeFloat32s = []float32{ + float32(0.0), + float32(1.0), + float32(-1.0), + float32(math.Inf(1)), + float32(math.Inf(-1)), +} + +var allFloat32s = append(safeFloat32s, float32(math.NaN())) + +var safeFloat64s = []float64{ + float64(0.0), + float64(1.0), + float64(-1.0), + math.Inf(1), + math.Inf(-1), +} + +var allFloat64s = append(safeFloat64s, math.NaN()) + +func TestFloat(t *testing.T) { + runTestCases(t, false, "plain", flatten( + allFloat32s, + allFloat64s, + )) + // See checkEqual for why NaNs are missing. + runTestCases(t, false, "pointers", pointersTo(flatten( + safeFloat32s, + safeFloat64s, + ))) + runTestCases(t, false, "interfaces", interfacesTo(flatten( + safeFloat32s, + safeFloat64s, + ))) + runTestCases(t, false, "interfacesToPointers", interfacesTo(pointersTo(flatten( + safeFloat32s, + safeFloat64s, + )))) +} + +const onlyDouble float64 = 1.0000000000000002 + +func TestFloatTruncation(t *testing.T) { + runTestCases(t, true, "pass", []interface{}{ + truncatingFloat32{save: onlyDouble}, + }) + runTestCases(t, false, "fail", []interface{}{ + truncatingFloat32{save: 1.0}, + }) +} + +var safeComplex64s = combine(safeFloat32s, safeFloat32s, func(i, j interface{}) interface{} { + return complex(i.(float32), j.(float32)) +}) + +var allComplex64s = combine(allFloat32s, allFloat32s, func(i, j interface{}) interface{} { + return complex(i.(float32), j.(float32)) +}) + +var safeComplex128s = combine(safeFloat64s, safeFloat64s, func(i, j interface{}) interface{} { + return complex(i.(float64), j.(float64)) +}) + +var allComplex128s = combine(allFloat64s, allFloat64s, func(i, j interface{}) interface{} { + return complex(i.(float64), j.(float64)) +}) + +func TestComplex(t *testing.T) { + runTestCases(t, false, "plain", flatten( + allComplex64s, + allComplex128s, + )) + // See TestFloat; same issue. + runTestCases(t, false, "pointers", pointersTo(flatten( + safeComplex64s, + safeComplex128s, + ))) + runTestCases(t, false, "interfacse", interfacesTo(flatten( + safeComplex64s, + safeComplex128s, + ))) + runTestCases(t, false, "interfacesTo", interfacesTo(pointersTo(flatten( + safeComplex64s, + safeComplex128s, + )))) +} + +func TestComplexTruncation(t *testing.T) { + runTestCases(t, true, "pass", []interface{}{ + truncatingComplex64{save: complex(onlyDouble, onlyDouble)}, + truncatingComplex64{save: complex(1.0, onlyDouble)}, + truncatingComplex64{save: complex(onlyDouble, 1.0)}, + }) + runTestCases(t, false, "fail", []interface{}{ + truncatingComplex64{save: complex(1.0, 1.0)}, + }) +} diff --git a/pkg/state/tests/integer.go b/pkg/state/tests/integer.go new file mode 100644 index 000000000..ca403eed1 --- /dev/null +++ b/pkg/state/tests/integer.go @@ -0,0 +1,163 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tests + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +// +stateify type +type truncatingUint8 struct { + save uint64 + load uint8 `state:"nosave"` +} + +func (t *truncatingUint8) StateSave(m state.Sink) { + m.Save(0, &t.save) +} + +func (t *truncatingUint8) StateLoad(m state.Source) { + m.Load(0, &t.load) + t.save = uint64(t.load) + t.load = 0 +} + +var _ state.SaverLoader = (*truncatingUint8)(nil) + +// +stateify type +type truncatingUint16 struct { + save uint64 + load uint16 `state:"nosave"` +} + +func (t *truncatingUint16) StateSave(m state.Sink) { + m.Save(0, &t.save) +} + +func (t *truncatingUint16) StateLoad(m state.Source) { + m.Load(0, &t.load) + t.save = uint64(t.load) + t.load = 0 +} + +var _ state.SaverLoader = (*truncatingUint16)(nil) + +// +stateify type +type truncatingUint32 struct { + save uint64 + load uint32 `state:"nosave"` +} + +func (t *truncatingUint32) StateSave(m state.Sink) { + m.Save(0, &t.save) +} + +func (t *truncatingUint32) StateLoad(m state.Source) { + m.Load(0, &t.load) + t.save = uint64(t.load) + t.load = 0 +} + +var _ state.SaverLoader = (*truncatingUint32)(nil) + +// +stateify type +type truncatingInt8 struct { + save int64 + load int8 `state:"nosave"` +} + +func (t *truncatingInt8) StateSave(m state.Sink) { + m.Save(0, &t.save) +} + +func (t *truncatingInt8) StateLoad(m state.Source) { + m.Load(0, &t.load) + t.save = int64(t.load) + t.load = 0 +} + +var _ state.SaverLoader = (*truncatingInt8)(nil) + +// +stateify type +type truncatingInt16 struct { + save int64 + load int16 `state:"nosave"` +} + +func (t *truncatingInt16) StateSave(m state.Sink) { + m.Save(0, &t.save) +} + +func (t *truncatingInt16) StateLoad(m state.Source) { + m.Load(0, &t.load) + t.save = int64(t.load) + t.load = 0 +} + +var _ state.SaverLoader = (*truncatingInt16)(nil) + +// +stateify type +type truncatingInt32 struct { + save int64 + load int32 `state:"nosave"` +} + +func (t *truncatingInt32) StateSave(m state.Sink) { + m.Save(0, &t.save) +} + +func (t *truncatingInt32) StateLoad(m state.Source) { + m.Load(0, &t.load) + t.save = int64(t.load) + t.load = 0 +} + +var _ state.SaverLoader = (*truncatingInt32)(nil) + +// +stateify type +type truncatingFloat32 struct { + save float64 + load float32 `state:"nosave"` +} + +func (t *truncatingFloat32) StateSave(m state.Sink) { + m.Save(0, &t.save) +} + +func (t *truncatingFloat32) StateLoad(m state.Source) { + m.Load(0, &t.load) + t.save = float64(t.load) + t.load = 0 +} + +var _ state.SaverLoader = (*truncatingFloat32)(nil) + +// +stateify type +type truncatingComplex64 struct { + save complex128 + load complex64 `state:"nosave"` +} + +func (t *truncatingComplex64) StateSave(m state.Sink) { + m.Save(0, &t.save) +} + +func (t *truncatingComplex64) StateLoad(m state.Source) { + m.Load(0, &t.load) + t.save = complex128(t.load) + t.load = 0 +} + +var _ state.SaverLoader = (*truncatingComplex64)(nil) diff --git a/pkg/state/tests/integer_test.go b/pkg/state/tests/integer_test.go new file mode 100644 index 000000000..d3931c952 --- /dev/null +++ b/pkg/state/tests/integer_test.go @@ -0,0 +1,94 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tests + +import ( + "math" + "testing" +) + +var ( + allIntTs = []int{-1, 0, 1} + allInt8s = []int8{math.MinInt8, -1, 0, 1, math.MaxInt8} + allInt16s = []int16{math.MinInt16, -1, 0, 1, math.MaxInt16} + allInt32s = []int32{math.MinInt32, -1, 0, 1, math.MaxInt32} + allInt64s = []int64{math.MinInt64, -1, 0, 1, math.MaxInt64} + allUintTs = []uint{0, 1} + allUintptrs = []uintptr{0, 1, ^uintptr(0)} + allUint8s = []uint8{0, 1, math.MaxUint8} + allUint16s = []uint16{0, 1, math.MaxUint16} + allUint32s = []uint32{0, 1, math.MaxUint32} + allUint64s = []uint64{0, 1, math.MaxUint64} +) + +var allInts = flatten( + allIntTs, + allInt8s, + allInt16s, + allInt32s, + allInt64s, +) + +var allUints = flatten( + allUintTs, + allUintptrs, + allUint8s, + allUint16s, + allUint32s, + allUint64s, +) + +func TestInt(t *testing.T) { + runTestCases(t, false, "plain", allInts) + runTestCases(t, false, "pointers", pointersTo(allInts)) + runTestCases(t, false, "interfaces", interfacesTo(allInts)) + runTestCases(t, false, "interfacesTo", interfacesTo(pointersTo(allInts))) +} + +func TestIntTruncation(t *testing.T) { + runTestCases(t, true, "pass", []interface{}{ + truncatingInt8{save: math.MinInt8 - 1}, + truncatingInt16{save: math.MinInt16 - 1}, + truncatingInt32{save: math.MinInt32 - 1}, + truncatingInt8{save: math.MaxInt8 + 1}, + truncatingInt16{save: math.MaxInt16 + 1}, + truncatingInt32{save: math.MaxInt32 + 1}, + }) + runTestCases(t, false, "fail", []interface{}{ + truncatingInt8{save: 1}, + truncatingInt16{save: 1}, + truncatingInt32{save: 1}, + }) +} + +func TestUint(t *testing.T) { + runTestCases(t, false, "plain", allUints) + runTestCases(t, false, "pointers", pointersTo(allUints)) + runTestCases(t, false, "interfaces", interfacesTo(allUints)) + runTestCases(t, false, "interfacesTo", interfacesTo(pointersTo(allUints))) +} + +func TestUintTruncation(t *testing.T) { + runTestCases(t, true, "pass", []interface{}{ + truncatingUint8{save: math.MaxUint8 + 1}, + truncatingUint16{save: math.MaxUint16 + 1}, + truncatingUint32{save: math.MaxUint32 + 1}, + }) + runTestCases(t, false, "fail", []interface{}{ + truncatingUint8{save: 1}, + truncatingUint16{save: 1}, + truncatingUint32{save: 1}, + }) +} diff --git a/pkg/state/tests/load.go b/pkg/state/tests/load.go new file mode 100644 index 000000000..a8350c0f3 --- /dev/null +++ b/pkg/state/tests/load.go @@ -0,0 +1,61 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tests + +// +stateify savable +type genericContainer struct { + v interface{} +} + +// +stateify savable +type afterLoadStruct struct { + v int `state:"nosave"` +} + +func (a *afterLoadStruct) afterLoad() { + a.v++ +} + +// +stateify savable +type valueLoadStruct struct { + v int `state:".(int64)"` +} + +func (v *valueLoadStruct) saveV() int64 { + return int64(v.v) // Save as int64. +} + +func (v *valueLoadStruct) loadV(value int64) { + v.v = int(value) // Load as int. +} + +// +stateify savable +type cycleStruct struct { + c *cycleStruct +} + +// +stateify savable +type badCycleStruct struct { + b *badCycleStruct `state:"wait"` +} + +func (b *badCycleStruct) afterLoad() { + if b.b != b { + // This is not executable, since AfterLoad requires that the + // object and all dependencies are complete. This should cause + // a deadlock error during load. + panic("badCycleStruct.afterLoad called") + } +} diff --git a/pkg/state/tests/load_test.go b/pkg/state/tests/load_test.go new file mode 100644 index 000000000..1e9794296 --- /dev/null +++ b/pkg/state/tests/load_test.go @@ -0,0 +1,70 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tests + +import ( + "testing" +) + +func TestLoadHooks(t *testing.T) { + runTestCases(t, false, "load-hooks", []interface{}{ + &afterLoadStruct{v: 1}, + &valueLoadStruct{v: 1}, + &genericContainer{v: &afterLoadStruct{v: 1}}, + &genericContainer{v: &valueLoadStruct{v: 1}}, + &sliceContainer{v: []interface{}{&afterLoadStruct{v: 1}}}, + &sliceContainer{v: []interface{}{&valueLoadStruct{v: 1}}}, + &mapContainer{v: map[int]interface{}{0: &afterLoadStruct{v: 1}}}, + &mapContainer{v: map[int]interface{}{0: &valueLoadStruct{v: 1}}}, + }) +} + +func TestCycles(t *testing.T) { + // cs is a single object cycle. + cs := cycleStruct{nil} + cs.c = &cs + + // cs1 and cs2 are in a two object cycle. + cs1 := cycleStruct{nil} + cs2 := cycleStruct{nil} + cs1.c = &cs2 + cs2.c = &cs1 + + runTestCases(t, false, "cycles", []interface{}{ + cs, + cs1, + }) +} + +func TestDeadlock(t *testing.T) { + // bs is a single object cycle. This does not cause deadlock because an + // object cannot wait for itself. + bs := badCycleStruct{nil} + bs.b = &bs + + runTestCases(t, false, "self", []interface{}{ + &bs, + }) + + // bs2 and bs2 are in a deadlocking cycle. + bs1 := badCycleStruct{nil} + bs2 := badCycleStruct{nil} + bs1.b = &bs2 + bs2.b = &bs1 + + runTestCases(t, true, "deadlock", []interface{}{ + &bs1, + }) +} diff --git a/pkg/state/tests/map.go b/pkg/state/tests/map.go new file mode 100644 index 000000000..db4e548f1 --- /dev/null +++ b/pkg/state/tests/map.go @@ -0,0 +1,28 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tests + +// +stateify savable +type mapContainer struct { + v map[int]interface{} +} + +// +stateify savable +type mapPtrContainer struct { + v *map[int]interface{} +} + +// +stateify savable +type registeredMapStruct struct{} diff --git a/pkg/state/tests/map_test.go b/pkg/state/tests/map_test.go new file mode 100644 index 000000000..92bf0fc01 --- /dev/null +++ b/pkg/state/tests/map_test.go @@ -0,0 +1,90 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tests + +import ( + "reflect" + "testing" +) + +var allMapPrimitives = []interface{}{ + bool(true), + int(1), + int8(1), + int16(1), + int32(1), + int64(1), + uint(1), + uintptr(1), + uint8(1), + uint16(1), + uint32(1), + uint64(1), + string(""), + registeredMapStruct{}, +} + +var allMapKeys = flatten(allMapPrimitives, pointersTo(allMapPrimitives)) + +var allMapValues = flatten(allMapPrimitives, pointersTo(allMapPrimitives), interfacesTo(allMapPrimitives)) + +var emptyMaps = combine(allMapKeys, allMapValues, func(v1, v2 interface{}) interface{} { + m := reflect.MakeMap(reflect.MapOf(reflect.TypeOf(v1), reflect.TypeOf(v2))) + return m.Interface() +}) + +var fullMaps = combine(allMapKeys, allMapValues, func(v1, v2 interface{}) interface{} { + m := reflect.MakeMap(reflect.MapOf(reflect.TypeOf(v1), reflect.TypeOf(v2))) + m.SetMapIndex(reflect.Zero(reflect.TypeOf(v1)), reflect.Zero(reflect.TypeOf(v2))) + return m.Interface() +}) + +func TestMapAliasing(t *testing.T) { + v := make(map[int]int) + ptrToV := &v + aliases := []map[int]int{v, v} + runTestCases(t, false, "", []interface{}{ptrToV, aliases}) +} + +func TestMapsEmpty(t *testing.T) { + runTestCases(t, false, "plain", emptyMaps) + runTestCases(t, false, "pointers", pointersTo(emptyMaps)) + runTestCases(t, false, "interfaces", interfacesTo(emptyMaps)) + runTestCases(t, false, "interfacesToPointers", interfacesTo(pointersTo(emptyMaps))) +} + +func TestMapsFull(t *testing.T) { + runTestCases(t, false, "plain", fullMaps) + runTestCases(t, false, "pointers", pointersTo(fullMaps)) + runTestCases(t, false, "interfaces", interfacesTo(fullMaps)) + runTestCases(t, false, "interfacesToPointer", interfacesTo(pointersTo(fullMaps))) +} + +func TestMapContainers(t *testing.T) { + var ( + nilMap map[int]interface{} + emptyMap = make(map[int]interface{}) + fullMap = map[int]interface{}{0: nil} + ) + runTestCases(t, false, "", []interface{}{ + mapContainer{v: nilMap}, + mapContainer{v: emptyMap}, + mapContainer{v: fullMap}, + mapPtrContainer{v: nil}, + mapPtrContainer{v: &nilMap}, + mapPtrContainer{v: &emptyMap}, + mapPtrContainer{v: &fullMap}, + }) +} diff --git a/pkg/state/tests/register.go b/pkg/state/tests/register.go new file mode 100644 index 000000000..074d86315 --- /dev/null +++ b/pkg/state/tests/register.go @@ -0,0 +1,21 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tests + +// +stateify savable +type alreadyRegisteredStruct struct{} + +// +stateify savable +type alreadyRegisteredOther int diff --git a/pkg/state/tests/register_test.go b/pkg/state/tests/register_test.go new file mode 100644 index 000000000..c829753cc --- /dev/null +++ b/pkg/state/tests/register_test.go @@ -0,0 +1,167 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tests + +import ( + "testing" + + "gvisor.dev/gvisor/pkg/state" +) + +// faker calls itself whatever is in the name field. +type faker struct { + Name string + Fields []string +} + +func (f *faker) StateTypeName() string { + return f.Name +} + +func (f *faker) StateFields() []string { + return f.Fields +} + +// fakerWithSaverLoader has all it needs. +type fakerWithSaverLoader struct { + faker +} + +func (f *fakerWithSaverLoader) StateSave(m state.Sink) {} + +func (f *fakerWithSaverLoader) StateLoad(m state.Source) {} + +// fakerOther calls itself .. uh, itself? +type fakerOther string + +func (f *fakerOther) StateTypeName() string { + return string(*f) +} + +func (f *fakerOther) StateFields() []string { + return nil +} + +func newFakerOther(name string) *fakerOther { + f := fakerOther(name) + return &f +} + +// fakerOtherBadFields returns non-nil fields. +type fakerOtherBadFields string + +func (f *fakerOtherBadFields) StateTypeName() string { + return string(*f) +} + +func (f *fakerOtherBadFields) StateFields() []string { + return []string{string(*f)} +} + +func newFakerOtherBadFields(name string) *fakerOtherBadFields { + f := fakerOtherBadFields(name) + return &f +} + +// fakerOtherSaverLoader implements SaverLoader methods. +type fakerOtherSaverLoader string + +func (f *fakerOtherSaverLoader) StateTypeName() string { + return string(*f) +} + +func (f *fakerOtherSaverLoader) StateFields() []string { + return nil +} + +func (f *fakerOtherSaverLoader) StateSave(m state.Sink) {} + +func (f *fakerOtherSaverLoader) StateLoad(m state.Source) {} + +func newFakerOtherSaverLoader(name string) *fakerOtherSaverLoader { + f := fakerOtherSaverLoader(name) + return &f +} + +func TestRegisterPrimitives(t *testing.T) { + for _, typeName := range []string{ + "int", + "int8", + "int16", + "int32", + "int64", + "uint", + "uintptr", + "uint8", + "uint16", + "uint32", + "uint64", + "float32", + "float64", + "complex64", + "complex128", + "string", + } { + t.Run("struct/"+typeName, func(t *testing.T) { + defer func() { + if r := recover(); r == nil { + t.Errorf("Registering type %q did not panic", typeName) + } + }() + state.Register(&faker{ + Name: typeName, + }) + }) + t.Run("other/"+typeName, func(t *testing.T) { + defer func() { + if r := recover(); r == nil { + t.Errorf("Registering type %q did not panic", typeName) + } + }() + state.Register(newFakerOther(typeName)) + }) + } +} + +func TestRegisterBad(t *testing.T) { + const ( + goodName = "foo" + firstField = "a" + secondField = "b" + ) + for name, object := range map[string]state.Type{ + "non-struct-with-fields": newFakerOtherBadFields(goodName), + "non-struct-with-saverloader": newFakerOtherSaverLoader(goodName), + "struct-without-saverloader": &faker{Name: goodName}, + "non-struct-duplicate-with-struct": newFakerOther((new(alreadyRegisteredStruct)).StateTypeName()), + "non-struct-duplicate-with-non-struct": newFakerOther((new(alreadyRegisteredOther)).StateTypeName()), + "struct-duplicate-with-struct": &fakerWithSaverLoader{faker{Name: (new(alreadyRegisteredStruct)).StateTypeName()}}, + "struct-duplicate-with-non-struct": &fakerWithSaverLoader{faker{Name: (new(alreadyRegisteredOther)).StateTypeName()}}, + "struct-with-empty-field": &fakerWithSaverLoader{faker{Name: goodName, Fields: []string{""}}}, + "struct-with-empty-field-and-non-empty": &fakerWithSaverLoader{faker{Name: goodName, Fields: []string{firstField, ""}}}, + "struct-with-duplicate-field": &fakerWithSaverLoader{faker{Name: goodName, Fields: []string{firstField, firstField}}}, + "struct-with-duplicate-field-and-non-dup": &fakerWithSaverLoader{faker{Name: goodName, Fields: []string{firstField, secondField, firstField}}}, + } { + t.Run(name, func(t *testing.T) { + defer func() { + if r := recover(); r == nil { + t.Errorf("Registering object %#v did not panic", object) + } + }() + state.Register(object) + }) + + } +} diff --git a/pkg/state/tests/string_test.go b/pkg/state/tests/string_test.go new file mode 100644 index 000000000..44f5a562c --- /dev/null +++ b/pkg/state/tests/string_test.go @@ -0,0 +1,34 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tests + +import ( + "testing" +) + +const nonEmptyString = "hello world" + +var allStrings = []string{ + "", + nonEmptyString, + "\\0", +} + +func TestString(t *testing.T) { + runTestCases(t, false, "plain", flatten(allStrings)) + runTestCases(t, false, "pointers", pointersTo(flatten(allStrings))) + runTestCases(t, false, "interfaces", interfacesTo(flatten(allStrings))) + runTestCases(t, false, "interfacesToPointers", interfacesTo(pointersTo(flatten(allStrings)))) +} diff --git a/pkg/state/tests/struct.go b/pkg/state/tests/struct.go new file mode 100644 index 000000000..bd2c2b399 --- /dev/null +++ b/pkg/state/tests/struct.go @@ -0,0 +1,65 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tests + +type unregisteredEmptyStruct struct{} + +// typeOnlyEmptyStruct just implements the state.Type interface. +type typeOnlyEmptyStruct struct{} + +func (*typeOnlyEmptyStruct) StateTypeName() string { return "registeredEmptyStruct" } + +func (*typeOnlyEmptyStruct) StateFields() []string { return nil } + +// +stateify savable +type savableEmptyStruct struct{} + +// +stateify savable +type emptyStructPointer struct { + nothing *struct{} +} + +// +stateify savable +type outerSame struct { + inner inner +} + +// +stateify savable +type outerFieldFirst struct { + inner inner + v int64 +} + +// +stateify savable +type outerFieldSecond struct { + v int64 + inner inner +} + +// +stateify savable +type outerArray struct { + inner [2]inner +} + +// +stateify savable +type inner struct { + v int64 +} + +// +stateify savable +type system struct { + v1 interface{} + v2 interface{} +} diff --git a/pkg/state/tests/struct_test.go b/pkg/state/tests/struct_test.go new file mode 100644 index 000000000..de9d17aa7 --- /dev/null +++ b/pkg/state/tests/struct_test.go @@ -0,0 +1,89 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tests + +import ( + "testing" + + "gvisor.dev/gvisor/pkg/state" +) + +func TestEmptyStruct(t *testing.T) { + runTestCases(t, false, "plain", []interface{}{ + unregisteredEmptyStruct{}, + typeOnlyEmptyStruct{}, + savableEmptyStruct{}, + }) + runTestCases(t, false, "pointers", pointersTo([]interface{}{ + unregisteredEmptyStruct{}, + typeOnlyEmptyStruct{}, + savableEmptyStruct{}, + })) + runTestCases(t, false, "interfaces-pass", interfacesTo([]interface{}{ + // Only registered types can be dispatched via interfaces. All + // other types should fail, even if it is the empty struct. + savableEmptyStruct{}, + })) + runTestCases(t, true, "interfaces-fail", interfacesTo([]interface{}{ + unregisteredEmptyStruct{}, + typeOnlyEmptyStruct{}, + })) + runTestCases(t, false, "interfacesToPointers-pass", interfacesTo(pointersTo([]interface{}{ + savableEmptyStruct{}, + }))) + runTestCases(t, true, "interfacesToPointers-fail", interfacesTo(pointersTo([]interface{}{ + unregisteredEmptyStruct{}, + typeOnlyEmptyStruct{}, + }))) + + // Ensuring empty struct aliasing works. + es := emptyStructPointer{new(struct{})} + runTestCases(t, false, "empty-struct-pointers", []interface{}{ + emptyStructPointer{}, + es, + []emptyStructPointer{es, es}, // Same pointer. + }) +} + +func TestRegisterTypeOnlyStruct(t *testing.T) { + defer func() { + if r := recover(); r == nil { + t.Errorf("Register did not panic") + } + }() + state.Register((*typeOnlyEmptyStruct)(nil)) +} + +func TestEmbeddedPointers(t *testing.T) { + var ( + ofs outerSame + of1 outerFieldFirst + of2 outerFieldSecond + oa outerArray + ) + + runTestCases(t, false, "embedded-pointers", []interface{}{ + system{&ofs, &ofs.inner}, + system{&ofs.inner, &ofs}, + system{&of1, &of1.inner}, + system{&of1.inner, &of1}, + system{&of2, &of2.inner}, + system{&of2.inner, &of2}, + system{&oa, &oa.inner[0]}, + system{&oa, &oa.inner[1]}, + system{&oa.inner[0], &oa}, + system{&oa.inner[1], &oa}, + }) +} diff --git a/pkg/state/tests/tests.go b/pkg/state/tests/tests.go new file mode 100644 index 000000000..435a0e9db --- /dev/null +++ b/pkg/state/tests/tests.go @@ -0,0 +1,215 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package tests tests the state packages. +package tests + +import ( + "bytes" + "context" + "fmt" + "math" + "reflect" + "testing" + + "gvisor.dev/gvisor/pkg/state" + "gvisor.dev/gvisor/pkg/state/pretty" +) + +// discard is an implementation of wire.Writer. +type discard struct{} + +// Write implements wire.Writer.Write. +func (discard) Write(p []byte) (int, error) { return len(p), nil } + +// WriteByte implements wire.Writer.WriteByte. +func (discard) WriteByte(byte) error { return nil } + +// checkEqual checks if two objects are equal. +// +// N.B. This only handles one level of dereferences for NaN. Otherwise we +// would need to fork the entire implementation of reflect.DeepEqual. +func checkEqual(root, loadedValue interface{}) bool { + if reflect.DeepEqual(root, loadedValue) { + return true + } + + // NaN is not equal to itself. We handle the case of raw floating point + // primitives here, but don't handle this case nested. + rf32, ok1 := root.(float32) + lf32, ok2 := loadedValue.(float32) + if ok1 && ok2 && math.IsNaN(float64(rf32)) && math.IsNaN(float64(lf32)) { + return true + } + rf64, ok1 := root.(float64) + lf64, ok2 := loadedValue.(float64) + if ok1 && ok2 && math.IsNaN(rf64) && math.IsNaN(lf64) { + return true + } + + // Same real for complex numbers. + rc64, ok1 := root.(complex64) + lc64, ok2 := root.(complex64) + if ok1 && ok2 { + return checkEqual(real(rc64), real(lc64)) && checkEqual(imag(rc64), imag(lc64)) + } + rc128, ok1 := root.(complex128) + lc128, ok2 := root.(complex128) + if ok1 && ok2 { + return checkEqual(real(rc128), real(lc128)) && checkEqual(imag(rc128), imag(lc128)) + } + + return false +} + +// runTestCases runs a test for each object in objects. +func runTestCases(t *testing.T, shouldFail bool, prefix string, objects []interface{}) { + t.Helper() + for i, root := range objects { + t.Run(fmt.Sprintf("%s%d", prefix, i), func(t *testing.T) { + t.Logf("Original object:\n%#v", root) + + // Save the passed object. + saveBuffer := &bytes.Buffer{} + saveObjectPtr := reflect.New(reflect.TypeOf(root)) + saveObjectPtr.Elem().Set(reflect.ValueOf(root)) + saveStats, err := state.Save(context.Background(), saveBuffer, saveObjectPtr.Interface()) + if err != nil { + if shouldFail { + return + } + t.Fatalf("Save failed unexpectedly: %v", err) + } + + // Dump the serialized proto to aid with debugging. + var ppBuf bytes.Buffer + t.Logf("Raw state:\n%v", saveBuffer.Bytes()) + if err := pretty.PrintText(&ppBuf, bytes.NewReader(saveBuffer.Bytes())); err != nil { + // We don't count this as a test failure if we + // have shouldFail set, but we will count as a + // failure if we were not expecting to fail. + if !shouldFail { + t.Errorf("PrettyPrint(html=false) failed unexpected: %v", err) + } + } + if err := pretty.PrintHTML(discard{}, bytes.NewReader(saveBuffer.Bytes())); err != nil { + // See above. + if !shouldFail { + t.Errorf("PrettyPrint(html=true) failed unexpected: %v", err) + } + } + t.Logf("Encoded state:\n%s", ppBuf.String()) + t.Logf("Save stats:\n%s", saveStats.String()) + + // Load a new copy of the object. + loadObjectPtr := reflect.New(reflect.TypeOf(root)) + loadStats, err := state.Load(context.Background(), bytes.NewReader(saveBuffer.Bytes()), loadObjectPtr.Interface()) + if err != nil { + if shouldFail { + return + } + t.Fatalf("Load failed unexpectedly: %v", err) + } + + // Compare the values. + loadedValue := loadObjectPtr.Elem().Interface() + if !checkEqual(root, loadedValue) { + if shouldFail { + return + } + t.Fatalf("Objects differ:\n\toriginal: %#v\n\tloaded: %#v\n", root, loadedValue) + } + + // Everything went okay. Is that good? + if shouldFail { + t.Fatalf("This test was expected to fail, but didn't.") + } + t.Logf("Load stats:\n%s", loadStats.String()) + + // Truncate half the bytes in the byte stream, + // and ensure that we can't restore. Then + // truncate only the final byte and ensure that + // we can't restore. + l := saveBuffer.Len() + halfReader := bytes.NewReader(saveBuffer.Bytes()[:l/2]) + if _, err := state.Load(context.Background(), halfReader, loadObjectPtr.Interface()); err == nil { + t.Errorf("Load with half bytes succeeded unexpectedly.") + } + missingByteReader := bytes.NewReader(saveBuffer.Bytes()[:l-1]) + if _, err := state.Load(context.Background(), missingByteReader, loadObjectPtr.Interface()); err == nil { + t.Errorf("Load with missing byte succeeded unexpectedly.") + } + }) + } +} + +// convert converts the slice to an []interface{}. +func convert(v interface{}) (r []interface{}) { + s := reflect.ValueOf(v) // Must be slice. + for i := 0; i < s.Len(); i++ { + r = append(r, s.Index(i).Interface()) + } + return r +} + +// flatten flattens multiple slices. +func flatten(vs ...interface{}) (r []interface{}) { + for _, v := range vs { + r = append(r, convert(v)...) + } + return r +} + +// filter maps from one slice to another. +func filter(vs interface{}, fn func(interface{}) (interface{}, bool)) (r []interface{}) { + s := reflect.ValueOf(vs) + for i := 0; i < s.Len(); i++ { + v, ok := fn(s.Index(i).Interface()) + if ok { + r = append(r, v) + } + } + return r +} + +// combine combines objects in two slices as specified. +func combine(v1, v2 interface{}, fn func(_, _ interface{}) interface{}) (r []interface{}) { + s1 := reflect.ValueOf(v1) + s2 := reflect.ValueOf(v2) + for i := 0; i < s1.Len(); i++ { + for j := 0; j < s2.Len(); j++ { + // Combine using the given function. + r = append(r, fn(s1.Index(i).Interface(), s2.Index(j).Interface())) + } + } + return r +} + +// pointersTo is a filter function that returns pointers. +func pointersTo(vs interface{}) []interface{} { + return filter(vs, func(o interface{}) (interface{}, bool) { + v := reflect.New(reflect.TypeOf(o)) + v.Elem().Set(reflect.ValueOf(o)) + return v.Interface(), true + }) +} + +// interfacesTo is a filter function that returns interface objects. +func interfacesTo(vs interface{}) []interface{} { + return filter(vs, func(o interface{}) (interface{}, bool) { + var v [1]interface{} + v[0] = o + return v, true + }) +} diff --git a/pkg/state/types.go b/pkg/state/types.go new file mode 100644 index 000000000..215ef80f8 --- /dev/null +++ b/pkg/state/types.go @@ -0,0 +1,361 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package state + +import ( + "reflect" + "sort" + + "gvisor.dev/gvisor/pkg/state/wire" +) + +// assertValidType asserts that the type is valid. +func assertValidType(name string, fields []string) { + if name == "" { + Failf("type has empty name") + } + fieldsCopy := make([]string, len(fields)) + for i := 0; i < len(fields); i++ { + if fields[i] == "" { + Failf("field has empty name for type %q", name) + } + fieldsCopy[i] = fields[i] + } + sort.Slice(fieldsCopy, func(i, j int) bool { + return fieldsCopy[i] < fieldsCopy[j] + }) + for i := range fieldsCopy { + if i > 0 && fieldsCopy[i-1] == fieldsCopy[i] { + Failf("duplicate field %q for type %s", fieldsCopy[i], name) + } + } +} + +// typeEntry is an entry in the typeDatabase. +type typeEntry struct { + ID typeID + wire.Type +} + +// reconciledTypeEntry is a reconciled entry in the typeDatabase. +type reconciledTypeEntry struct { + wire.Type + LocalType reflect.Type + FieldOrder []int +} + +// typeEncodeDatabase is an internal TypeInfo database for encoding. +type typeEncodeDatabase struct { + // byType maps by type to the typeEntry. + byType map[reflect.Type]*typeEntry + + // lastID is the last used ID. + lastID typeID +} + +// makeTypeEncodeDatabase makes a typeDatabase. +func makeTypeEncodeDatabase() typeEncodeDatabase { + return typeEncodeDatabase{ + byType: make(map[reflect.Type]*typeEntry), + } +} + +// typeDecodeDatabase is an internal TypeInfo database for decoding. +type typeDecodeDatabase struct { + // byID maps by ID to type. + byID []*reconciledTypeEntry + + // pending are entries that are pending validation by Lookup. These + // will be reconciled with actual objects. Note that these will also be + // used to lookup types by name, since they may not be reconciled and + // there's little value to deleting from this map. + pending []*wire.Type +} + +// makeTypeDecodeDatabase makes a typeDatabase. +func makeTypeDecodeDatabase() typeDecodeDatabase { + return typeDecodeDatabase{} +} + +// lookupNameFields extracts the name and fields from an object. +func lookupNameFields(typ reflect.Type) (string, []string, bool) { + v := reflect.Zero(reflect.PtrTo(typ)).Interface() + t, ok := v.(Type) + if !ok { + // Is this a primitive? + if typ.Kind() == reflect.Interface { + return interfaceType, nil, true + } + name := typ.Name() + if _, ok := primitiveTypeDatabase[name]; !ok { + // This is not a known type, and not a primitive. The + // encoder may proceed for anonymous empty structs, or + // it may deference the type pointer and try again. + return "", nil, false + } + return name, nil, true + } + // Extract the name from the object. + name := t.StateTypeName() + fields := t.StateFields() + assertValidType(name, fields) + return name, fields, true +} + +// Lookup looks up or registers the given object. +// +// The bool indicates whether this is an existing entry: false means the entry +// did not exist, and true means the entry did exist. If this bool is false and +// the returned typeEntry are nil, then the obj did not implement the Type +// interface. +func (tdb *typeEncodeDatabase) Lookup(typ reflect.Type) (*typeEntry, bool) { + te, ok := tdb.byType[typ] + if !ok { + // Lookup the type information. + name, fields, ok := lookupNameFields(typ) + if !ok { + // Empty structs may still be encoded, so let the + // caller decide what to do from here. + return nil, false + } + + // Register the new type. + tdb.lastID++ + te = &typeEntry{ + ID: tdb.lastID, + Type: wire.Type{ + Name: name, + Fields: fields, + }, + } + + // All done. + tdb.byType[typ] = te + return te, false + } + return te, true +} + +// Register adds a typeID entry. +func (tbd *typeDecodeDatabase) Register(typ *wire.Type) { + assertValidType(typ.Name, typ.Fields) + tbd.pending = append(tbd.pending, typ) +} + +// LookupName looks up the type name by ID. +func (tbd *typeDecodeDatabase) LookupName(id typeID) string { + if len(tbd.pending) < int(id) { + // This is likely an encoder error? + Failf("type ID %d not available", id) + } + return tbd.pending[id-1].Name +} + +// LookupType looks up the type by ID. +func (tbd *typeDecodeDatabase) LookupType(id typeID) reflect.Type { + name := tbd.LookupName(id) + typ, ok := globalTypeDatabase[name] + if !ok { + // If not available, see if it's primitive. + typ, ok = primitiveTypeDatabase[name] + if !ok && name == interfaceType { + // Matches the built-in interface type. + var i interface{} + return reflect.TypeOf(&i).Elem() + } + if !ok { + // The type is perhaps not registered? + Failf("type name %q is not available", name) + } + return typ // Primitive type. + } + return typ // Registered type. +} + +// singleFieldOrder defines the field order for a single field. +var singleFieldOrder = []int{0} + +// Lookup looks up or registers the given object. +// +// First, the typeID is searched to see if this has already been appropriately +// reconciled. If no, then a reconcilation will take place that may result in a +// field ordering. If a nil reconciledTypeEntry is returned from this method, +// then the object does not support the Type interface. +// +// This method never returns nil. +func (tbd *typeDecodeDatabase) Lookup(id typeID, typ reflect.Type) *reconciledTypeEntry { + if len(tbd.byID) > int(id) && tbd.byID[id-1] != nil { + // Already reconciled. + return tbd.byID[id-1] + } + // The ID has not been reconciled yet. That's fine. We need to make + // sure it aligns with the current provided object. + if len(tbd.pending) < int(id) { + // This id was never registered. Probably an encoder error? + Failf("typeDatabase does not contain id %d", id) + } + // Extract the pending info. + pending := tbd.pending[id-1] + // Grow the byID list. + if len(tbd.byID) < int(id) { + tbd.byID = append(tbd.byID, make([]*reconciledTypeEntry, int(id)-len(tbd.byID))...) + } + // Reconcile the type. + name, fields, ok := lookupNameFields(typ) + if !ok { + // Empty structs are decoded only when the type is nil. Since + // this isn't the case, we fail here. + Failf("unsupported type %q during decode; can't reconcile", pending.Name) + } + if name != pending.Name { + // Are these the same type? Print a helpful message as this may + // actually happen in practice if types change. + Failf("typeDatabase contains conflicting definitions for id %d: %s->%v (current) and %s->%v (existing)", + id, name, fields, pending.Name, pending.Fields) + } + rte := &reconciledTypeEntry{ + Type: wire.Type{ + Name: name, + Fields: fields, + }, + LocalType: typ, + } + // If there are zero or one fields, then we skip allocating the field + // slice. There is special handling for decoding in this case. If the + // field name does not match, it will be caught in the general purpose + // code below. + if len(fields) != len(pending.Fields) { + Failf("type %q contains different fields: %v (decode) and %v (encode)", + name, fields, pending.Fields) + } + if len(fields) == 0 { + tbd.byID[id-1] = rte // Save. + return rte + } + if len(fields) == 1 && fields[0] == pending.Fields[0] { + tbd.byID[id-1] = rte // Save. + rte.FieldOrder = singleFieldOrder + return rte + } + // For each field in the current object's information, match it to a + // field in the destination object. We know from the assertion above + // and the insertion on insertion to pending that neither field + // contains any duplicates. + fieldOrder := make([]int, len(fields)) + for i, name := range fields { + fieldOrder[i] = -1 // Sentinel. + // Is it an exact match? + if pending.Fields[i] == name { + fieldOrder[i] = i + continue + } + // Find the matching field. + for j, otherName := range pending.Fields { + if name == otherName { + fieldOrder[i] = j + break + } + } + if fieldOrder[i] == -1 { + // The type name matches but we are lacking some common fields. + Failf("type %q has mismatched fields: %v (decode) and %v (encode)", + name, fields, pending.Fields) + } + } + // The type has been reeconciled. + rte.FieldOrder = fieldOrder + tbd.byID[id-1] = rte + return rte +} + +// interfaceType defines all interfaces. +const interfaceType = "interface" + +// primitiveTypeDatabase is a set of fixed types. +var primitiveTypeDatabase = func() map[string]reflect.Type { + r := make(map[string]reflect.Type) + for _, t := range []reflect.Type{ + reflect.TypeOf(false), + reflect.TypeOf(int(0)), + reflect.TypeOf(int8(0)), + reflect.TypeOf(int16(0)), + reflect.TypeOf(int32(0)), + reflect.TypeOf(int64(0)), + reflect.TypeOf(uint(0)), + reflect.TypeOf(uintptr(0)), + reflect.TypeOf(uint8(0)), + reflect.TypeOf(uint16(0)), + reflect.TypeOf(uint32(0)), + reflect.TypeOf(uint64(0)), + reflect.TypeOf(""), + reflect.TypeOf(float32(0.0)), + reflect.TypeOf(float64(0.0)), + reflect.TypeOf(complex64(0.0)), + reflect.TypeOf(complex128(0.0)), + } { + r[t.Name()] = t + } + return r +}() + +// globalTypeDatabase is used for dispatching interfaces on decode. +var globalTypeDatabase = map[string]reflect.Type{} + +// Register registers a type. +// +// This must be called on init and only done once. +func Register(t Type) { + name := t.StateTypeName() + fields := t.StateFields() + assertValidType(name, fields) + // Register must always be called on pointers. + typ := reflect.TypeOf(t) + if typ.Kind() != reflect.Ptr { + Failf("Register must be called on pointers") + } + typ = typ.Elem() + if typ.Kind() == reflect.Struct { + // All registered structs must implement SaverLoader. We allow + // the registration is non-struct types with just the Type + // interface, but we need to call StateSave/StateLoad methods + // on aggregate types. + if _, ok := t.(SaverLoader); !ok { + Failf("struct %T does not implement SaverLoader", t) + } + } else { + // Non-structs must not have any fields. We don't support + // calling StateSave/StateLoad methods on any non-struct types. + // If custom behavior is required, these types should be + // wrapped in a structure of some kind. + if len(fields) != 0 { + Failf("non-struct %T has non-zero fields %v", t, fields) + } + // We don't allow non-structs to implement StateSave/StateLoad + // methods, because they won't be called and it's confusing. + if _, ok := t.(SaverLoader); ok { + Failf("non-struct %T implements SaverLoader", t) + } + } + if _, ok := primitiveTypeDatabase[name]; ok { + Failf("conflicting primitiveTypeDatabase entry for %T: used by primitive", t) + } + if _, ok := globalTypeDatabase[name]; ok { + Failf("conflicting globalTypeDatabase entries for %T: name conflict", t) + } + if name == interfaceType { + Failf("conflicting name for %T: matches interfaceType", t) + } + globalTypeDatabase[name] = typ +} diff --git a/pkg/state/wire/BUILD b/pkg/state/wire/BUILD new file mode 100644 index 000000000..311b93dcb --- /dev/null +++ b/pkg/state/wire/BUILD @@ -0,0 +1,12 @@ +load("//tools:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_library( + name = "wire", + srcs = ["wire.go"], + marshal = False, + stateify = False, + visibility = ["//:sandbox"], + deps = ["//pkg/gohacks"], +) diff --git a/pkg/state/wire/wire.go b/pkg/state/wire/wire.go new file mode 100644 index 000000000..93dee6740 --- /dev/null +++ b/pkg/state/wire/wire.go @@ -0,0 +1,970 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package wire contains a few basic types that can be composed to serialize +// graph information for the state package. This package defines the wire +// protocol. +// +// Note that these types are careful about how they implement the relevant +// interfaces (either value receiver or pointer receiver), so that native-sized +// types, such as integers and simple pointers, can fit inside the interface +// object. +// +// This package also uses panic as control flow, so called should be careful to +// wrap calls in appropriate handlers. +// +// Testing for this package is driven by the state test package. +package wire + +import ( + "fmt" + "io" + "math" + + "gvisor.dev/gvisor/pkg/gohacks" +) + +// Reader is the required reader interface. +type Reader interface { + io.Reader + ReadByte() (byte, error) +} + +// Writer is the required writer interface. +type Writer interface { + io.Writer + WriteByte(byte) error +} + +// readFull is a utility. The equivalent is not needed for Write, but the API +// contract dictates that it must always complete all bytes given or return an +// error. +func readFull(r io.Reader, p []byte) { + for done := 0; done < len(p); { + n, err := r.Read(p[done:]) + done += n + if n == 0 && err != nil { + panic(err) + } + } +} + +// Object is a generic object. +type Object interface { + // save saves the given object. + // + // Panic is used for error control flow. + save(Writer) + + // load loads a new object of the given type. + // + // Panic is used for error control flow. + load(Reader) Object +} + +// Bool is a boolean. +type Bool bool + +// loadBool loads an object of type Bool. +func loadBool(r Reader) Bool { + b := loadUint(r) + return Bool(b == 1) +} + +// save implements Object.save. +func (b Bool) save(w Writer) { + var v Uint + if b { + v = 1 + } else { + v = 0 + } + v.save(w) +} + +// load implements Object.load. +func (Bool) load(r Reader) Object { return loadBool(r) } + +// Int is a signed integer. +// +// This uses varint encoding. +type Int int64 + +// loadInt loads an object of type Int. +func loadInt(r Reader) Int { + u := loadUint(r) + x := Int(u >> 1) + if u&1 != 0 { + x = ^x + } + return x +} + +// save implements Object.save. +func (i Int) save(w Writer) { + u := Uint(i) << 1 + if i < 0 { + u = ^u + } + u.save(w) +} + +// load implements Object.load. +func (Int) load(r Reader) Object { return loadInt(r) } + +// Uint is an unsigned integer. +type Uint uint64 + +// loadUint loads an object of type Uint. +func loadUint(r Reader) Uint { + var ( + u Uint + s uint + ) + for i := 0; i <= 9; i++ { + b, err := r.ReadByte() + if err != nil { + panic(err) + } + if b < 0x80 { + if i == 9 && b > 1 { + panic("overflow") + } + u |= Uint(b) << s + return u + } + u |= Uint(b&0x7f) << s + s += 7 + } + panic("unreachable") +} + +// save implements Object.save. +func (u Uint) save(w Writer) { + for u >= 0x80 { + if err := w.WriteByte(byte(u) | 0x80); err != nil { + panic(err) + } + u >>= 7 + } + if err := w.WriteByte(byte(u)); err != nil { + panic(err) + } +} + +// load implements Object.load. +func (Uint) load(r Reader) Object { return loadUint(r) } + +// Float32 is a 32-bit floating point number. +type Float32 float32 + +// loadFloat32 loads an object of type Float32. +func loadFloat32(r Reader) Float32 { + n := loadUint(r) + return Float32(math.Float32frombits(uint32(n))) +} + +// save implements Object.save. +func (f Float32) save(w Writer) { + n := Uint(math.Float32bits(float32(f))) + n.save(w) +} + +// load implements Object.load. +func (Float32) load(r Reader) Object { return loadFloat32(r) } + +// Float64 is a 64-bit floating point number. +type Float64 float64 + +// loadFloat64 loads an object of type Float64. +func loadFloat64(r Reader) Float64 { + n := loadUint(r) + return Float64(math.Float64frombits(uint64(n))) +} + +// save implements Object.save. +func (f Float64) save(w Writer) { + n := Uint(math.Float64bits(float64(f))) + n.save(w) +} + +// load implements Object.load. +func (Float64) load(r Reader) Object { return loadFloat64(r) } + +// Complex64 is a 64-bit complex number. +type Complex64 complex128 + +// loadComplex64 loads an object of type Complex64. +func loadComplex64(r Reader) Complex64 { + re := loadFloat32(r) + im := loadFloat32(r) + return Complex64(complex(float32(re), float32(im))) +} + +// save implements Object.save. +func (c *Complex64) save(w Writer) { + re := Float32(real(*c)) + im := Float32(imag(*c)) + re.save(w) + im.save(w) +} + +// load implements Object.load. +func (*Complex64) load(r Reader) Object { + c := loadComplex64(r) + return &c +} + +// Complex128 is a 128-bit complex number. +type Complex128 complex128 + +// loadComplex128 loads an object of type Complex128. +func loadComplex128(r Reader) Complex128 { + re := loadFloat64(r) + im := loadFloat64(r) + return Complex128(complex(float64(re), float64(im))) +} + +// save implements Object.save. +func (c *Complex128) save(w Writer) { + re := Float64(real(*c)) + im := Float64(imag(*c)) + re.save(w) + im.save(w) +} + +// load implements Object.load. +func (*Complex128) load(r Reader) Object { + c := loadComplex128(r) + return &c +} + +// String is a string. +type String string + +// loadString loads an object of type String. +func loadString(r Reader) String { + l := loadUint(r) + p := make([]byte, l) + readFull(r, p) + return String(gohacks.StringFromImmutableBytes(p)) +} + +// save implements Object.save. +func (s *String) save(w Writer) { + l := Uint(len(*s)) + l.save(w) + p := gohacks.ImmutableBytesFromString(string(*s)) + _, err := w.Write(p) // Must write all bytes. + if err != nil { + panic(err) + } +} + +// load implements Object.load. +func (*String) load(r Reader) Object { + s := loadString(r) + return &s +} + +// Dot is a kind of reference: one of Index and FieldName. +type Dot interface { + isDot() +} + +// Index is a reference resolution. +type Index uint32 + +func (Index) isDot() {} + +// FieldName is a reference resolution. +type FieldName string + +func (*FieldName) isDot() {} + +// Ref is a reference to an object. +type Ref struct { + // Root is the root object. + Root Uint + + // Dots is the set of traversals required from the Root object above. + // Note that this will be stored in reverse order for efficiency. + Dots []Dot + + // Type is the base type for the root object. This is non-nil iff Dots + // is non-zero length (that is, this is a complex reference). This is + // not *strictly* necessary, but can be used to simplify decoding. + Type TypeSpec +} + +// loadRef loads an object of type Ref (abstract). +func loadRef(r Reader) Ref { + ref := Ref{ + Root: loadUint(r), + } + l := loadUint(r) + ref.Dots = make([]Dot, l) + for i := 0; i < int(l); i++ { + // Disambiguate between an Index (non-negative) and a field + // name (negative). This does some space and avoids a dedicate + // loadDot function. See Ref.save for the other side. + d := loadInt(r) + if d >= 0 { + ref.Dots[i] = Index(d) + continue + } + p := make([]byte, -d) + readFull(r, p) + fieldName := FieldName(gohacks.StringFromImmutableBytes(p)) + ref.Dots[i] = &fieldName + } + if l != 0 { + // Only if dots is non-zero. + ref.Type = loadTypeSpec(r) + } + return ref +} + +// save implements Object.save. +func (r *Ref) save(w Writer) { + r.Root.save(w) + l := Uint(len(r.Dots)) + l.save(w) + for _, d := range r.Dots { + // See LoadRef. We use non-negative numbers to encode Index + // objects and negative numbers to encode field lengths. + switch x := d.(type) { + case Index: + i := Int(x) + i.save(w) + case *FieldName: + d := Int(-len(*x)) + d.save(w) + p := gohacks.ImmutableBytesFromString(string(*x)) + if _, err := w.Write(p); err != nil { + panic(err) + } + default: + panic("unknown dot implementation") + } + } + if l != 0 { + // See above. + saveTypeSpec(w, r.Type) + } +} + +// load implements Object.load. +func (*Ref) load(r Reader) Object { + ref := loadRef(r) + return &ref +} + +// Nil is a primitive zero value of any type. +type Nil struct{} + +// loadNil loads an object of type Nil. +func loadNil(r Reader) Nil { + return Nil{} +} + +// save implements Object.save. +func (Nil) save(w Writer) {} + +// load implements Object.load. +func (Nil) load(r Reader) Object { return loadNil(r) } + +// Slice is a slice value. +type Slice struct { + Length Uint + Capacity Uint + Ref Ref +} + +// loadSlice loads an object of type Slice. +func loadSlice(r Reader) Slice { + return Slice{ + Length: loadUint(r), + Capacity: loadUint(r), + Ref: loadRef(r), + } +} + +// save implements Object.save. +func (s *Slice) save(w Writer) { + s.Length.save(w) + s.Capacity.save(w) + s.Ref.save(w) +} + +// load implements Object.load. +func (*Slice) load(r Reader) Object { + s := loadSlice(r) + return &s +} + +// Array is an array value. +type Array struct { + Contents []Object +} + +// loadArray loads an object of type Array. +func loadArray(r Reader) Array { + l := loadUint(r) + if l == 0 { + // Note that there isn't a single object available to encode + // the type of, so we need this additional branch. + return Array{} + } + // All the objects here have the same type, so use dynamic dispatch + // only once. All other objects will automatically take the same type + // as the first object. + contents := make([]Object, l) + v := Load(r) + contents[0] = v + for i := 1; i < int(l); i++ { + contents[i] = v.load(r) + } + return Array{ + Contents: contents, + } +} + +// save implements Object.save. +func (a *Array) save(w Writer) { + l := Uint(len(a.Contents)) + l.save(w) + if l == 0 { + // See LoadArray. + return + } + // See above. + Save(w, a.Contents[0]) + for i := 1; i < int(l); i++ { + a.Contents[i].save(w) + } +} + +// load implements Object.load. +func (*Array) load(r Reader) Object { + a := loadArray(r) + return &a +} + +// Map is a map value. +type Map struct { + Keys []Object + Values []Object +} + +// loadMap loads an object of type Map. +func loadMap(r Reader) Map { + l := loadUint(r) + if l == 0 { + // See LoadArray. + return Map{} + } + // See type dispatch notes in Array. + keys := make([]Object, l) + values := make([]Object, l) + k := Load(r) + v := Load(r) + keys[0] = k + values[0] = v + for i := 1; i < int(l); i++ { + keys[i] = k.load(r) + values[i] = v.load(r) + } + return Map{ + Keys: keys, + Values: values, + } +} + +// save implements Object.save. +func (m *Map) save(w Writer) { + l := Uint(len(m.Keys)) + if int(l) != len(m.Values) { + panic(fmt.Sprintf("mismatched keys (%d) Aand values (%d)", len(m.Keys), len(m.Values))) + } + l.save(w) + if l == 0 { + // See LoadArray. + return + } + // See above. + Save(w, m.Keys[0]) + Save(w, m.Values[0]) + for i := 1; i < int(l); i++ { + m.Keys[i].save(w) + m.Values[i].save(w) + } +} + +// load implements Object.load. +func (*Map) load(r Reader) Object { + m := loadMap(r) + return &m +} + +// TypeSpec is a type dereference. +type TypeSpec interface { + isTypeSpec() +} + +// TypeID is a concrete type ID. +type TypeID Uint + +func (TypeID) isTypeSpec() {} + +// TypeSpecPointer is a pointer type. +type TypeSpecPointer struct { + Type TypeSpec +} + +func (*TypeSpecPointer) isTypeSpec() {} + +// TypeSpecArray is an array type. +type TypeSpecArray struct { + Count Uint + Type TypeSpec +} + +func (*TypeSpecArray) isTypeSpec() {} + +// TypeSpecSlice is a slice type. +type TypeSpecSlice struct { + Type TypeSpec +} + +func (*TypeSpecSlice) isTypeSpec() {} + +// TypeSpecMap is a map type. +type TypeSpecMap struct { + Key TypeSpec + Value TypeSpec +} + +func (*TypeSpecMap) isTypeSpec() {} + +// TypeSpecNil is an empty type. +type TypeSpecNil struct{} + +func (TypeSpecNil) isTypeSpec() {} + +// TypeSpec types. +// +// These use a distinct encoding on the wire, as they are used only in the +// interface object. They are decoded through the dedicated loadTypeSpec and +// saveTypeSpec functions. +const ( + typeSpecTypeID Uint = iota + typeSpecPointer + typeSpecArray + typeSpecSlice + typeSpecMap + typeSpecNil +) + +// loadTypeSpec loads TypeSpec values. +func loadTypeSpec(r Reader) TypeSpec { + switch hdr := loadUint(r); hdr { + case typeSpecTypeID: + return TypeID(loadUint(r)) + case typeSpecPointer: + return &TypeSpecPointer{ + Type: loadTypeSpec(r), + } + case typeSpecArray: + return &TypeSpecArray{ + Count: loadUint(r), + Type: loadTypeSpec(r), + } + case typeSpecSlice: + return &TypeSpecSlice{ + Type: loadTypeSpec(r), + } + case typeSpecMap: + return &TypeSpecMap{ + Key: loadTypeSpec(r), + Value: loadTypeSpec(r), + } + case typeSpecNil: + return TypeSpecNil{} + default: + // This is not a valid stream? + panic(fmt.Errorf("unknown header: %d", hdr)) + } +} + +// saveTypeSpec saves TypeSpec values. +func saveTypeSpec(w Writer, t TypeSpec) { + switch x := t.(type) { + case TypeID: + typeSpecTypeID.save(w) + Uint(x).save(w) + case *TypeSpecPointer: + typeSpecPointer.save(w) + saveTypeSpec(w, x.Type) + case *TypeSpecArray: + typeSpecArray.save(w) + x.Count.save(w) + saveTypeSpec(w, x.Type) + case *TypeSpecSlice: + typeSpecSlice.save(w) + saveTypeSpec(w, x.Type) + case *TypeSpecMap: + typeSpecMap.save(w) + saveTypeSpec(w, x.Key) + saveTypeSpec(w, x.Value) + case TypeSpecNil: + typeSpecNil.save(w) + default: + // This should not happen? + panic(fmt.Errorf("unknown type %T", t)) + } +} + +// Interface is an interface value. +type Interface struct { + Type TypeSpec + Value Object +} + +// loadInterface loads an object of type Interface. +func loadInterface(r Reader) Interface { + return Interface{ + Type: loadTypeSpec(r), + Value: Load(r), + } +} + +// save implements Object.save. +func (i *Interface) save(w Writer) { + saveTypeSpec(w, i.Type) + Save(w, i.Value) +} + +// load implements Object.load. +func (*Interface) load(r Reader) Object { + i := loadInterface(r) + return &i +} + +// Type is type information. +type Type struct { + Name string + Fields []string +} + +// loadType loads an object of type Type. +func loadType(r Reader) Type { + name := string(loadString(r)) + l := loadUint(r) + fields := make([]string, l) + for i := 0; i < int(l); i++ { + fields[i] = string(loadString(r)) + } + return Type{ + Name: name, + Fields: fields, + } +} + +// save implements Object.save. +func (t *Type) save(w Writer) { + s := String(t.Name) + s.save(w) + l := Uint(len(t.Fields)) + l.save(w) + for i := 0; i < int(l); i++ { + s := String(t.Fields[i]) + s.save(w) + } +} + +// load implements Object.load. +func (*Type) load(r Reader) Object { + t := loadType(r) + return &t +} + +// multipleObjects is a special type for serializing multiple objects. +type multipleObjects []Object + +// loadMultipleObjects loads a series of objects. +func loadMultipleObjects(r Reader) multipleObjects { + l := loadUint(r) + m := make(multipleObjects, l) + for i := 0; i < int(l); i++ { + m[i] = Load(r) + } + return m +} + +// save implements Object.save. +func (m *multipleObjects) save(w Writer) { + l := Uint(len(*m)) + l.save(w) + for i := 0; i < int(l); i++ { + Save(w, (*m)[i]) + } +} + +// load implements Object.load. +func (*multipleObjects) load(r Reader) Object { + m := loadMultipleObjects(r) + return &m +} + +// noObjects represents no objects. +type noObjects struct{} + +// loadNoObjects loads a sentinel. +func loadNoObjects(r Reader) noObjects { return noObjects{} } + +// save implements Object.save. +func (noObjects) save(w Writer) {} + +// load implements Object.load. +func (noObjects) load(r Reader) Object { return loadNoObjects(r) } + +// Struct is a basic composite value. +type Struct struct { + TypeID TypeID + fields Object // Optionally noObjects or *multipleObjects. +} + +// Field returns a pointer to the given field slot. +// +// This must be called after Alloc. +func (s *Struct) Field(i int) *Object { + if fields, ok := s.fields.(*multipleObjects); ok { + return &((*fields)[i]) + } + if _, ok := s.fields.(noObjects); ok { + // Alloc may be optionally called; can't call twice. + panic("Field called inappropriately, wrong Alloc?") + } + return &s.fields +} + +// Alloc allocates the given number of fields. +// +// This must be called before Add and Save. +// +// Precondition: slots must be positive. +func (s *Struct) Alloc(slots int) { + switch { + case slots == 0: + s.fields = noObjects{} + case slots == 1: + // Leave it alone. + case slots > 1: + fields := make(multipleObjects, slots) + s.fields = &fields + default: + // Violates precondition. + panic(fmt.Sprintf("Alloc called with negative slots %d?", slots)) + } +} + +// Fields returns the number of fields. +func (s *Struct) Fields() int { + switch x := s.fields.(type) { + case *multipleObjects: + return len(*x) + case noObjects: + return 0 + default: + return 1 + } +} + +// loadStruct loads an object of type Struct. +func loadStruct(r Reader) Struct { + return Struct{ + TypeID: TypeID(loadUint(r)), + fields: Load(r), + } +} + +// save implements Object.save. +// +// Precondition: Alloc must have been called, and the fields all filled in +// appropriately. See Alloc and Add for more details. +func (s *Struct) save(w Writer) { + Uint(s.TypeID).save(w) + Save(w, s.fields) +} + +// load implements Object.load. +func (*Struct) load(r Reader) Object { + s := loadStruct(r) + return &s +} + +// Object types. +// +// N.B. Be careful about changing the order or introducing new elements in the +// middle here. This is part of the wire format and shouldn't change. +const ( + typeBool Uint = iota + typeInt + typeUint + typeFloat32 + typeFloat64 + typeNil + typeRef + typeString + typeSlice + typeArray + typeMap + typeStruct + typeNoObjects + typeMultipleObjects + typeInterface + typeComplex64 + typeComplex128 + typeType +) + +// Save saves the given object. +// +// +checkescape all +// +// N.B. This function will panic on error. +func Save(w Writer, obj Object) { + switch x := obj.(type) { + case Bool: + typeBool.save(w) + x.save(w) + case Int: + typeInt.save(w) + x.save(w) + case Uint: + typeUint.save(w) + x.save(w) + case Float32: + typeFloat32.save(w) + x.save(w) + case Float64: + typeFloat64.save(w) + x.save(w) + case Nil: + typeNil.save(w) + x.save(w) + case *Ref: + typeRef.save(w) + x.save(w) + case *String: + typeString.save(w) + x.save(w) + case *Slice: + typeSlice.save(w) + x.save(w) + case *Array: + typeArray.save(w) + x.save(w) + case *Map: + typeMap.save(w) + x.save(w) + case *Struct: + typeStruct.save(w) + x.save(w) + case noObjects: + typeNoObjects.save(w) + x.save(w) + case *multipleObjects: + typeMultipleObjects.save(w) + x.save(w) + case *Interface: + typeInterface.save(w) + x.save(w) + case *Type: + typeType.save(w) + x.save(w) + case *Complex64: + typeComplex64.save(w) + x.save(w) + case *Complex128: + typeComplex128.save(w) + x.save(w) + default: + panic(fmt.Errorf("unknown type: %#v", obj)) + } +} + +// Load loads a new object. +// +// +checkescape all +// +// N.B. This function will panic on error. +func Load(r Reader) Object { + switch hdr := loadUint(r); hdr { + case typeBool: + return loadBool(r) + case typeInt: + return loadInt(r) + case typeUint: + return loadUint(r) + case typeFloat32: + return loadFloat32(r) + case typeFloat64: + return loadFloat64(r) + case typeNil: + return loadNil(r) + case typeRef: + return ((*Ref)(nil)).load(r) // Escapes. + case typeString: + return ((*String)(nil)).load(r) // Escapes. + case typeSlice: + return ((*Slice)(nil)).load(r) // Escapes. + case typeArray: + return ((*Array)(nil)).load(r) // Escapes. + case typeMap: + return ((*Map)(nil)).load(r) // Escapes. + case typeStruct: + return ((*Struct)(nil)).load(r) // Escapes. + case typeNoObjects: // Special for struct. + return loadNoObjects(r) + case typeMultipleObjects: // Special for struct. + return ((*multipleObjects)(nil)).load(r) // Escapes. + case typeInterface: + return ((*Interface)(nil)).load(r) // Escapes. + case typeComplex64: + return ((*Complex64)(nil)).load(r) // Escapes. + case typeComplex128: + return ((*Complex128)(nil)).load(r) // Escapes. + case typeType: + return ((*Type)(nil)).load(r) // Escapes. + default: + // This is not a valid stream? + panic(fmt.Errorf("unknown header: %d", hdr)) + } +} + +// LoadUint loads a single unsigned integer. +// +// N.B. This function will panic on error. +func LoadUint(r Reader) uint64 { + return uint64(loadUint(r)) +} + +// SaveUint saves a single unsigned integer. +// +// N.B. This function will panic on error. +func SaveUint(w Writer, v uint64) { + Uint(v).save(w) +} diff --git a/runsc/cmd/BUILD b/runsc/cmd/BUILD index af3538ef0..dae9b3b3e 100644 --- a/runsc/cmd/BUILD +++ b/runsc/cmd/BUILD @@ -45,7 +45,7 @@ go_library( "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/platform", - "//pkg/state", + "//pkg/state/pretty", "//pkg/state/statefile", "//pkg/sync", "//pkg/unet", diff --git a/runsc/cmd/statefile.go b/runsc/cmd/statefile.go index e6f1907da..daed9e728 100644 --- a/runsc/cmd/statefile.go +++ b/runsc/cmd/statefile.go @@ -20,7 +20,7 @@ import ( "os" "github.com/google/subcommands" - "gvisor.dev/gvisor/pkg/state" + "gvisor.dev/gvisor/pkg/state/pretty" "gvisor.dev/gvisor/pkg/state/statefile" "gvisor.dev/gvisor/runsc/flag" ) @@ -105,8 +105,14 @@ func (s *Statefile) Execute(_ context.Context, f *flag.FlagSet, args ...interfac if err != nil { Fatalf("error parsing statefile: %v", err) } - if err := state.PrettyPrint(output, rc, s.html); err != nil { - Fatalf("error printing state: %v", err) + if s.html { + if err := pretty.PrintHTML(output, rc); err != nil { + Fatalf("error printing state: %v", err) + } + } else { + if err := pretty.PrintText(output, rc); err != nil { + Fatalf("error printing state: %v", err) + } } return subcommands.ExitSuccess } diff --git a/tools/checkescape/checkescape.go b/tools/checkescape/checkescape.go index 571e9a6e6..f8def4823 100644 --- a/tools/checkescape/checkescape.go +++ b/tools/checkescape/checkescape.go @@ -88,7 +88,7 @@ const ( testMagic = "// +mustescape:" // exempt is the exemption annotation. - exempt = "// escapes:" + exempt = "// escapes" ) // escapingBuiltins are builtins known to escape. @@ -546,7 +546,7 @@ func run(pass *analysis.Pass) (interface{}, error) { for _, cg := range f.Comments { for _, c := range cg.List { p := pass.Fset.Position(c.Slash) - if strings.HasPrefix(c.Text, exempt) { + if strings.HasPrefix(strings.ToLower(c.Text), exempt) { exemptions[LinePosition{ Filename: filepath.Base(p.Filename), Line: p.Line, diff --git a/tools/go_stateify/main.go b/tools/go_stateify/main.go index 309ee9c21..4f6ed208a 100644 --- a/tools/go_stateify/main.go +++ b/tools/go_stateify/main.go @@ -103,7 +103,7 @@ type scanFunctions struct { // skipped if nil. // // Fields tagged nosave are skipped. -func scanFields(ss *ast.StructType, fn scanFunctions) { +func scanFields(ss *ast.StructType, prefix string, fn scanFunctions) { if ss.Fields.List == nil { // No fields. return @@ -127,7 +127,16 @@ func scanFields(ss *ast.StructType, fn scanFunctions) { continue } - switch tag := extractStateTag(field.Tag); tag { + // Is this a anonymous struct? If yes, then continue the + // recursion with the given prefix. We don't pay attention to + // any tags on the top-level struct field. + tag := extractStateTag(field.Tag) + if anon, ok := field.Type.(*ast.StructType); ok && tag == "" { + scanFields(anon, name+".", fn) + continue + } + + switch tag { case "zerovalue": if fn.zerovalue != nil { fn.zerovalue(name) @@ -201,28 +210,12 @@ func main() { // initCalls is dumped at the end. var initCalls []string - // Declare our emission closures. + // Common closures. emitRegister := func(name string) { - initCalls = append(initCalls, fmt.Sprintf("%sRegister(\"%s.%s\", (*%s)(nil), state.Fns{Save: (*%s).save, Load: (*%s).load})", statePrefix, *fullPkg, name, name, name, name)) + initCalls = append(initCalls, fmt.Sprintf("%sRegister((*%s)(nil))", statePrefix, name)) } emitZeroCheck := func(name string) { - fmt.Fprintf(outputFile, " if !%sIsZeroValue(&x.%s) { m.Failf(\"%s is %%#v, expected zero\", &x.%s) }\n", statePrefix, name, name, name) - } - emitLoadValue := func(name, typName string) { - fmt.Fprintf(outputFile, " m.LoadValue(\"%s\", new(%s), func(y interface{}) { x.load%s(y.(%s)) })\n", name, typName, camelCased(name), typName) - } - emitLoad := func(name string) { - fmt.Fprintf(outputFile, " m.Load(\"%s\", &x.%s)\n", name, name) - } - emitLoadWait := func(name string) { - fmt.Fprintf(outputFile, " m.LoadWait(\"%s\", &x.%s)\n", name, name) - } - emitSaveValue := func(name, typName string) { - fmt.Fprintf(outputFile, " var %s %s = x.save%s()\n", name, typName, camelCased(name)) - fmt.Fprintf(outputFile, " m.SaveValue(\"%s\", %s)\n", name, name) - } - emitSave := func(name string) { - fmt.Fprintf(outputFile, " m.Save(\"%s\", &x.%s)\n", name, name) + fmt.Fprintf(outputFile, " if !%sIsZeroValue(&x.%s) { %sFailf(\"%s is %%#v, expected zero\", &x.%s) }\n", statePrefix, name, statePrefix, name, name) } // Automated warning. @@ -329,87 +322,140 @@ func main() { continue } - // Only generate code for types marked - // "// +stateify savable" in one of the proceeding - // comment lines. + // Only generate code for types marked "// +stateify + // savable" in one of the proceeding comment lines. If + // the line is marked "// +stateify type" then only + // generate type information and register the type. if d.Doc == nil { continue } - savable := false + var ( + generateTypeInfo = false + generateSaverLoader = false + ) for _, l := range d.Doc.List { if l.Text == "// +stateify savable" { - savable = true + generateTypeInfo = true + generateSaverLoader = true break } + if l.Text == "// +stateify type" { + generateTypeInfo = true + } } - if !savable { + if !generateTypeInfo && !generateSaverLoader { continue } for _, gs := range d.Specs { ts := gs.(*ast.TypeSpec) - switch ts.Type.(type) { - case *ast.InterfaceType, *ast.ChanType, *ast.FuncType, *ast.ParenExpr, *ast.StarExpr: - // Don't register. - break + switch x := ts.Type.(type) { case *ast.StructType: maybeEmitImports() - ss := ts.Type.(*ast.StructType) + // Record the slot for each field. + fieldCount := 0 + fields := make(map[string]int) + emitField := func(name string) { + fmt.Fprintf(outputFile, " \"%s\",\n", name) + fields[name] = fieldCount + fieldCount++ + } + emitFieldValue := func(name string, _ string) { + emitField(name) + } + emitLoadValue := func(name, typName string) { + fmt.Fprintf(outputFile, " m.LoadValue(%d, new(%s), func(y interface{}) { x.load%s(y.(%s)) })\n", fields[name], typName, camelCased(name), typName) + } + emitLoad := func(name string) { + fmt.Fprintf(outputFile, " m.Load(%d, &x.%s)\n", fields[name], name) + } + emitLoadWait := func(name string) { + fmt.Fprintf(outputFile, " m.LoadWait(%d, &x.%s)\n", fields[name], name) + } + emitSaveValue := func(name, typName string) { + fmt.Fprintf(outputFile, " var %s %s = x.save%s()\n", name, typName, camelCased(name)) + fmt.Fprintf(outputFile, " m.SaveValue(%d, %s)\n", fields[name], name) + } + emitSave := func(name string) { + fmt.Fprintf(outputFile, " m.Save(%d, &x.%s)\n", fields[name], name) + } + + // Generate the type name method. + fmt.Fprintf(outputFile, "func (x *%s) StateTypeName() string {\n", ts.Name.Name) + fmt.Fprintf(outputFile, " return \"%s.%s\"\n", *fullPkg, ts.Name.Name) + fmt.Fprintf(outputFile, "}\n\n") + + // Generate the fields method. + fmt.Fprintf(outputFile, "func (x *%s) StateFields() []string {\n", ts.Name.Name) + fmt.Fprintf(outputFile, " return []string{\n") + scanFields(x, "", scanFunctions{ + normal: emitField, + wait: emitField, + value: emitFieldValue, + }) + fmt.Fprintf(outputFile, " }\n") + fmt.Fprintf(outputFile, "}\n\n") - // Define beforeSave if a definition was not found. This - // prevents the code from compiling if a custom beforeSave - // was defined in a file not provided to this binary and - // prevents inherited methods from being called multiple times - // by overriding them. - if _, ok := simpleMethods[method{ts.Name.Name, "beforeSave"}]; !ok { - fmt.Fprintf(outputFile, "func (x *%s) beforeSave() {}\n", ts.Name.Name) + // Define beforeSave if a definition was not found. This prevents + // the code from compiling if a custom beforeSave was defined in a + // file not provided to this binary and prevents inherited methods + // from being called multiple times by overriding them. + if _, ok := simpleMethods[method{ts.Name.Name, "beforeSave"}]; !ok && generateSaverLoader { + fmt.Fprintf(outputFile, "func (x *%s) beforeSave() {}\n\n", ts.Name.Name) } // Generate the save method. - fmt.Fprintf(outputFile, "func (x *%s) save(m %sMap) {\n", ts.Name.Name, statePrefix) - fmt.Fprintf(outputFile, " x.beforeSave()\n") - scanFields(ss, scanFunctions{zerovalue: emitZeroCheck}) - scanFields(ss, scanFunctions{value: emitSaveValue}) - scanFields(ss, scanFunctions{normal: emitSave, wait: emitSave}) - fmt.Fprintf(outputFile, "}\n\n") + // + // N.B. For historical reasons, we perform the value saves first, + // and perform the value loads last. There should be no dependency + // on this specific behavior, but the ability to specify slots + // allows a manual implementation to be order-dependent. + if generateSaverLoader { + fmt.Fprintf(outputFile, "func (x *%s) StateSave(m %sSink) {\n", ts.Name.Name, statePrefix) + fmt.Fprintf(outputFile, " x.beforeSave()\n") + scanFields(x, "", scanFunctions{zerovalue: emitZeroCheck}) + scanFields(x, "", scanFunctions{value: emitSaveValue}) + scanFields(x, "", scanFunctions{normal: emitSave, wait: emitSave}) + fmt.Fprintf(outputFile, "}\n\n") + } - // Define afterLoad if a definition was not found. We do this - // for the same reason that we do it for beforeSave. + // Define afterLoad if a definition was not found. We do this for + // the same reason that we do it for beforeSave. _, hasAfterLoad := simpleMethods[method{ts.Name.Name, "afterLoad"}] - if !hasAfterLoad { - fmt.Fprintf(outputFile, "func (x *%s) afterLoad() {}\n", ts.Name.Name) + if !hasAfterLoad && generateSaverLoader { + fmt.Fprintf(outputFile, "func (x *%s) afterLoad() {}\n\n", ts.Name.Name) } // Generate the load method. // - // Note that the manual loads always follow the - // automated loads. - fmt.Fprintf(outputFile, "func (x *%s) load(m %sMap) {\n", ts.Name.Name, statePrefix) - scanFields(ss, scanFunctions{normal: emitLoad, wait: emitLoadWait}) - scanFields(ss, scanFunctions{value: emitLoadValue}) - if hasAfterLoad { - // The call to afterLoad is made conditionally, because when - // AfterLoad is called, the object encodes a dependency on - // referred objects (i.e. fields). This means that afterLoad - // will not be called until the other afterLoads are called. - fmt.Fprintf(outputFile, " m.AfterLoad(x.afterLoad)\n") + // N.B. See the comment above for the save method. + if generateSaverLoader { + fmt.Fprintf(outputFile, "func (x *%s) StateLoad(m %sSource) {\n", ts.Name.Name, statePrefix) + scanFields(x, "", scanFunctions{normal: emitLoad, wait: emitLoadWait}) + scanFields(x, "", scanFunctions{value: emitLoadValue}) + if hasAfterLoad { + // The call to afterLoad is made conditionally, because when + // AfterLoad is called, the object encodes a dependency on + // referred objects (i.e. fields). This means that afterLoad + // will not be called until the other afterLoads are called. + fmt.Fprintf(outputFile, " m.AfterLoad(x.afterLoad)\n") + } + fmt.Fprintf(outputFile, "}\n\n") } - fmt.Fprintf(outputFile, "}\n\n") // Add to our registration. emitRegister(ts.Name.Name) + case *ast.Ident, *ast.SelectorExpr, *ast.ArrayType: maybeEmitImports() - _, val := resolveTypeName(ts.Name.Name, ts.Type) - - // Dispatch directly. - fmt.Fprintf(outputFile, "func (x *%s) save(m %sMap) {\n", ts.Name.Name, statePrefix) - fmt.Fprintf(outputFile, " m.SaveValue(\"\", (%s)(*x))\n", val) + // Generate the info methods. + fmt.Fprintf(outputFile, "func (x *%s) StateTypeName() string {\n", ts.Name.Name) + fmt.Fprintf(outputFile, " return \"%s.%s\"\n", *fullPkg, ts.Name.Name) fmt.Fprintf(outputFile, "}\n\n") - fmt.Fprintf(outputFile, "func (x *%s) load(m %sMap) {\n", ts.Name.Name, statePrefix) - fmt.Fprintf(outputFile, " m.LoadValue(\"\", new(%s), func(y interface{}) { *x = (%s)(y.(%s)) })\n", val, ts.Name.Name, val) + fmt.Fprintf(outputFile, "func (x *%s) StateFields() []string {\n", ts.Name.Name) + fmt.Fprintf(outputFile, " return nil\n") fmt.Fprintf(outputFile, "}\n\n") // See above. -- cgit v1.2.3 From b070e218c6fe61c6ef98e0a3af5ad58d7e627632 Mon Sep 17 00:00:00 2001 From: Bhasker Hariharan Date: Wed, 24 Jun 2020 10:21:44 -0700 Subject: Add support for Stack level options. Linux controls socket send/receive buffers using a few sysctl variables - net.core.rmem_default - net.core.rmem_max - net.core.wmem_max - net.core.wmem_default - net.ipv4.tcp_rmem - net.ipv4.tcp_wmem The first 4 control the default socket buffer sizes for all sockets raw/packet/tcp/udp and also the maximum permitted socket buffer that can be specified in setsockopt(SOL_SOCKET, SO_(RCV|SND)BUF,...). The last two control the TCP auto-tuning limits and override the default specified in rmem_default/wmem_default as well as the max limits. Netstack today only implements tcp_rmem/tcp_wmem and incorrectly uses it to limit the maximum size in setsockopt() as well as uses it for raw/udp sockets. This changelist introduces the other 4 and updates the udp/raw sockets to use the newly introduced variables. The values for min/max match the current tcp_rmem/wmem values and the default value buffers for UDP/RAW sockets is updated to match the linux value of 212KiB up from the really low current value of 32 KiB. Updates #3043 Fixes #3043 PiperOrigin-RevId: 318089805 --- benchmarks/tcp/tcp_proxy.go | 2 +- pkg/sentry/socket/netstack/stack.go | 12 +-- pkg/tcpip/stack/BUILD | 1 + pkg/tcpip/stack/stack.go | 18 ++++ pkg/tcpip/stack/stack_options.go | 106 +++++++++++++++++++++ pkg/tcpip/stack/stack_test.go | 80 ++++++++++++++++ pkg/tcpip/tcpip.go | 27 +----- pkg/tcpip/transport/raw/endpoint.go | 20 ++-- pkg/tcpip/transport/tcp/connect.go | 2 +- pkg/tcpip/transport/tcp/endpoint.go | 16 ++-- pkg/tcpip/transport/tcp/endpoint_state.go | 10 +- pkg/tcpip/transport/tcp/protocol.go | 53 ++++++++--- pkg/tcpip/transport/tcp/tcp_sack_test.go | 4 +- pkg/tcpip/transport/tcp/tcp_test.go | 18 ++-- pkg/tcpip/transport/tcp/testing/context/context.go | 6 +- pkg/tcpip/transport/udp/endpoint.go | 20 ++-- pkg/tcpip/transport/udp/protocol.go | 49 +--------- runsc/boot/loader.go | 2 +- 18 files changed, 306 insertions(+), 140 deletions(-) create mode 100644 pkg/tcpip/stack/stack_options.go (limited to 'pkg/sentry') diff --git a/benchmarks/tcp/tcp_proxy.go b/benchmarks/tcp/tcp_proxy.go index b3a4dbea3..4b7ca7a14 100644 --- a/benchmarks/tcp/tcp_proxy.go +++ b/benchmarks/tcp/tcp_proxy.go @@ -228,7 +228,7 @@ func newNetstackImpl(mode string) (impl, error) { }) // Set protocol options. - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackSACKEnabled(*sack)); err != nil { + if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SACKEnabled(*sack)); err != nil { return nil, fmt.Errorf("SetTransportProtocolOption for SACKEnabled failed: %s", err) } diff --git a/pkg/sentry/socket/netstack/stack.go b/pkg/sentry/socket/netstack/stack.go index d2fb655ea..548442b96 100644 --- a/pkg/sentry/socket/netstack/stack.go +++ b/pkg/sentry/socket/netstack/stack.go @@ -143,7 +143,7 @@ func (s *Stack) AddInterfaceAddr(idx int32, addr inet.InterfaceAddr) error { // TCPReceiveBufferSize implements inet.Stack.TCPReceiveBufferSize. func (s *Stack) TCPReceiveBufferSize() (inet.TCPBufferSize, error) { - var rs tcpip.StackReceiveBufferSizeOption + var rs tcp.ReceiveBufferSizeOption err := s.Stack.TransportProtocolOption(tcp.ProtocolNumber, &rs) return inet.TCPBufferSize{ Min: rs.Min, @@ -154,7 +154,7 @@ func (s *Stack) TCPReceiveBufferSize() (inet.TCPBufferSize, error) { // SetTCPReceiveBufferSize implements inet.Stack.SetTCPReceiveBufferSize. func (s *Stack) SetTCPReceiveBufferSize(size inet.TCPBufferSize) error { - rs := tcpip.StackReceiveBufferSizeOption{ + rs := tcp.ReceiveBufferSizeOption{ Min: size.Min, Default: size.Default, Max: size.Max, @@ -164,7 +164,7 @@ func (s *Stack) SetTCPReceiveBufferSize(size inet.TCPBufferSize) error { // TCPSendBufferSize implements inet.Stack.TCPSendBufferSize. func (s *Stack) TCPSendBufferSize() (inet.TCPBufferSize, error) { - var ss tcpip.StackSendBufferSizeOption + var ss tcp.SendBufferSizeOption err := s.Stack.TransportProtocolOption(tcp.ProtocolNumber, &ss) return inet.TCPBufferSize{ Min: ss.Min, @@ -175,7 +175,7 @@ func (s *Stack) TCPSendBufferSize() (inet.TCPBufferSize, error) { // SetTCPSendBufferSize implements inet.Stack.SetTCPSendBufferSize. func (s *Stack) SetTCPSendBufferSize(size inet.TCPBufferSize) error { - ss := tcpip.StackSendBufferSizeOption{ + ss := tcp.SendBufferSizeOption{ Min: size.Min, Default: size.Default, Max: size.Max, @@ -185,14 +185,14 @@ func (s *Stack) SetTCPSendBufferSize(size inet.TCPBufferSize) error { // TCPSACKEnabled implements inet.Stack.TCPSACKEnabled. func (s *Stack) TCPSACKEnabled() (bool, error) { - var sack tcpip.StackSACKEnabled + var sack tcp.SACKEnabled err := s.Stack.TransportProtocolOption(tcp.ProtocolNumber, &sack) return bool(sack), syserr.TranslateNetstackError(err).ToError() } // SetTCPSACKEnabled implements inet.Stack.SetTCPSACKEnabled. func (s *Stack) SetTCPSACKEnabled(enabled bool) error { - return syserr.TranslateNetstackError(s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackSACKEnabled(enabled))).ToError() + return syserr.TranslateNetstackError(s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SACKEnabled(enabled))).ToError() } // Statistics implements inet.Stack.Statistics. diff --git a/pkg/tcpip/stack/BUILD b/pkg/tcpip/stack/BUILD index 24f52b735..794ddb5c8 100644 --- a/pkg/tcpip/stack/BUILD +++ b/pkg/tcpip/stack/BUILD @@ -48,6 +48,7 @@ go_library( "route.go", "stack.go", "stack_global_state.go", + "stack_options.go", "transport_demuxer.go", ], visibility = ["//visibility:public"], diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index e92ec0c24..cdcfb8321 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -471,6 +471,14 @@ type Stack struct { // randomGenerator is an injectable pseudo random generator that can be // used when a random number is required. randomGenerator *mathrand.Rand + + // sendBufferSize holds the min/default/max send buffer sizes for + // endpoints other than TCP. + sendBufferSize SendBufferSizeOption + + // receiveBufferSize holds the min/default/max receive buffer sizes for + // endpoints other than TCP. + receiveBufferSize ReceiveBufferSizeOption } // UniqueID is an abstract generator of unique identifiers. @@ -683,6 +691,16 @@ func New(opts Options) *Stack { tempIIDSeed: opts.TempIIDSeed, forwarder: newForwardQueue(), randomGenerator: mathrand.New(randSrc), + sendBufferSize: SendBufferSizeOption{ + Min: MinBufferSize, + Default: DefaultBufferSize, + Max: DefaultMaxBufferSize, + }, + receiveBufferSize: ReceiveBufferSizeOption{ + Min: MinBufferSize, + Default: DefaultBufferSize, + Max: DefaultMaxBufferSize, + }, } // Add specified network protocols. diff --git a/pkg/tcpip/stack/stack_options.go b/pkg/tcpip/stack/stack_options.go new file mode 100644 index 000000000..0b093e6c5 --- /dev/null +++ b/pkg/tcpip/stack/stack_options.go @@ -0,0 +1,106 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package stack + +import "gvisor.dev/gvisor/pkg/tcpip" + +const ( + // MinBufferSize is the smallest size of a receive or send buffer. + MinBufferSize = 4 << 10 // 4 KiB + + // DefaultBufferSize is the default size of the send/recv buffer for a + // transport endpoint. + DefaultBufferSize = 212 << 10 // 212 KiB + + // DefaultMaxBufferSize is the default maximum permitted size of a + // send/receive buffer. + DefaultMaxBufferSize = 4 << 20 // 4 MiB +) + +// SendBufferSizeOption is used by stack.(Stack*).Option/SetOption to +// get/set the default, min and max send buffer sizes. +type SendBufferSizeOption struct { + Min int + Default int + Max int +} + +// ReceiveBufferSizeOption is used by stack.(Stack*).Option/SetOption to +// get/set the default, min and max receive buffer sizes. +type ReceiveBufferSizeOption struct { + Min int + Default int + Max int +} + +// SetOption allows setting stack wide options. +func (s *Stack) SetOption(option interface{}) *tcpip.Error { + switch v := option.(type) { + case SendBufferSizeOption: + // Make sure we don't allow lowering the buffer below minimum + // required for stack to work. + if v.Min < MinBufferSize { + return tcpip.ErrInvalidOptionValue + } + + if v.Default < v.Min || v.Default > v.Max { + return tcpip.ErrInvalidOptionValue + } + + s.mu.Lock() + s.sendBufferSize = v + s.mu.Unlock() + return nil + + case ReceiveBufferSizeOption: + // Make sure we don't allow lowering the buffer below minimum + // required for stack to work. + if v.Min < MinBufferSize { + return tcpip.ErrInvalidOptionValue + } + + if v.Default < v.Min || v.Default > v.Max { + return tcpip.ErrInvalidOptionValue + } + + s.mu.Lock() + s.receiveBufferSize = v + s.mu.Unlock() + return nil + + default: + return tcpip.ErrUnknownProtocolOption + } +} + +// Option allows retrieving stack wide options. +func (s *Stack) Option(option interface{}) *tcpip.Error { + switch v := option.(type) { + case *SendBufferSizeOption: + s.mu.RLock() + *v = s.sendBufferSize + s.mu.RUnlock() + return nil + + case *ReceiveBufferSizeOption: + s.mu.RLock() + *v = s.receiveBufferSize + s.mu.RUnlock() + return nil + + default: + return tcpip.ErrUnknownProtocolOption + } +} diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index 5aacbf53e..7657a4101 100644 --- a/pkg/tcpip/stack/stack_test.go +++ b/pkg/tcpip/stack/stack_test.go @@ -3338,3 +3338,83 @@ func TestDoDADWhenNICEnabled(t *testing.T) { t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (%s, nil), want = (%s, nil)", nicID, header.IPv6ProtocolNumber, got, addr.AddressWithPrefix) } } + +func TestStackReceiveBufferSizeOption(t *testing.T) { + const sMin = stack.MinBufferSize + testCases := []struct { + name string + rs stack.ReceiveBufferSizeOption + err *tcpip.Error + }{ + // Invalid configurations. + {"min_below_zero", stack.ReceiveBufferSizeOption{Min: -1, Default: sMin, Max: sMin}, tcpip.ErrInvalidOptionValue}, + {"min_zero", stack.ReceiveBufferSizeOption{Min: 0, Default: sMin, Max: sMin}, tcpip.ErrInvalidOptionValue}, + {"default_below_min", stack.ReceiveBufferSizeOption{Min: sMin, Default: sMin - 1, Max: sMin - 1}, tcpip.ErrInvalidOptionValue}, + {"default_above_max", stack.ReceiveBufferSizeOption{Min: sMin, Default: sMin + 1, Max: sMin}, tcpip.ErrInvalidOptionValue}, + {"max_below_min", stack.ReceiveBufferSizeOption{Min: sMin, Default: sMin + 1, Max: sMin - 1}, tcpip.ErrInvalidOptionValue}, + + // Valid Configurations + {"in_ascending_order", stack.ReceiveBufferSizeOption{Min: sMin, Default: sMin + 1, Max: sMin + 2}, nil}, + {"all_equal", stack.ReceiveBufferSizeOption{Min: sMin, Default: sMin, Max: sMin}, nil}, + {"min_default_equal", stack.ReceiveBufferSizeOption{Min: sMin, Default: sMin, Max: sMin + 1}, nil}, + {"default_max_equal", stack.ReceiveBufferSizeOption{Min: sMin, Default: sMin + 1, Max: sMin + 1}, nil}, + } + for _, tc := range testCases { + t.Run(tc.name, func(t *testing.T) { + s := stack.New(stack.Options{}) + defer s.Close() + if err := s.SetOption(tc.rs); err != tc.err { + t.Fatalf("s.SetOption(%#v) = %v, want: %v", tc.rs, err, tc.err) + } + var rs stack.ReceiveBufferSizeOption + if tc.err == nil { + if err := s.Option(&rs); err != nil { + t.Fatalf("s.Option(%#v) = %v, want: nil", rs, err) + } + if got, want := rs, tc.rs; got != want { + t.Fatalf("s.Option(..) returned unexpected value got: %#v, want: %#v", got, want) + } + } + }) + } +} + +func TestStackSendBufferSizeOption(t *testing.T) { + const sMin = stack.MinBufferSize + testCases := []struct { + name string + ss stack.SendBufferSizeOption + err *tcpip.Error + }{ + // Invalid configurations. + {"min_below_zero", stack.SendBufferSizeOption{Min: -1, Default: sMin, Max: sMin}, tcpip.ErrInvalidOptionValue}, + {"min_zero", stack.SendBufferSizeOption{Min: 0, Default: sMin, Max: sMin}, tcpip.ErrInvalidOptionValue}, + {"default_below_min", stack.SendBufferSizeOption{Min: 0, Default: sMin - 1, Max: sMin - 1}, tcpip.ErrInvalidOptionValue}, + {"default_above_max", stack.SendBufferSizeOption{Min: 0, Default: sMin + 1, Max: sMin}, tcpip.ErrInvalidOptionValue}, + {"max_below_min", stack.SendBufferSizeOption{Min: sMin, Default: sMin + 1, Max: sMin - 1}, tcpip.ErrInvalidOptionValue}, + + // Valid Configurations + {"in_ascending_order", stack.SendBufferSizeOption{Min: sMin, Default: sMin + 1, Max: sMin + 2}, nil}, + {"all_equal", stack.SendBufferSizeOption{Min: sMin, Default: sMin, Max: sMin}, nil}, + {"min_default_equal", stack.SendBufferSizeOption{Min: sMin, Default: sMin, Max: sMin + 1}, nil}, + {"default_max_equal", stack.SendBufferSizeOption{Min: sMin, Default: sMin + 1, Max: sMin + 1}, nil}, + } + for _, tc := range testCases { + t.Run(tc.name, func(t *testing.T) { + s := stack.New(stack.Options{}) + defer s.Close() + if err := s.SetOption(tc.ss); err != tc.err { + t.Fatalf("s.SetOption(%+v) = %v, want: %v", tc.ss, err, tc.err) + } + var ss stack.SendBufferSizeOption + if tc.err == nil { + if err := s.Option(&ss); err != nil { + t.Fatalf("s.Option(%+v) = %v, want: nil", ss, err) + } + if got, want := ss, tc.ss; got != want { + t.Fatalf("s.Option(..) returned unexpected value got: %#v, want: %#v", got, want) + } + } + }) + } +} diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 956232a44..4d45dcc42 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -813,33 +813,8 @@ type OutOfBandInlineOption int // a default TTL. type DefaultTTLOption uint8 -// StackSACKEnabled is used by stack.(*Stack).TransportProtocolOption to -// enable/disable SACK support in TCP. See: https://tools.ietf.org/html/rfc2018. -type StackSACKEnabled bool - -// StackDelayEnabled is used by stack.(Stack*).TransportProtocolOption to -// enable/disable Nagle's algorithm in TCP. -type StackDelayEnabled bool - -// StackSendBufferSizeOption is used by stack.(Stack*).TransportProtocolOption -// to get/set the default, min and max send buffer sizes. -type StackSendBufferSizeOption struct { - Min int - Default int - Max int -} - -// StackReceiveBufferSizeOption is used by -// stack.(Stack*).TransportProtocolOption to get/set the default, min and max -// receive buffer sizes. -type StackReceiveBufferSizeOption struct { - Min int - Default int - Max int -} - // -// IPPacketInfo is the message struture for IP_PKTINFO. +// IPPacketInfo is the message structure for IP_PKTINFO. // // +stateify savable type IPPacketInfo struct { diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index 6a7977259..dd514d397 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -111,13 +111,13 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProt } // Override with stack defaults. - var ss tcpip.StackSendBufferSizeOption - if err := s.TransportProtocolOption(transProto, &ss); err == nil { + var ss stack.SendBufferSizeOption + if err := s.Option(&ss); err == nil { e.sndBufSizeMax = ss.Default } - var rs tcpip.StackReceiveBufferSizeOption - if err := s.TransportProtocolOption(transProto, &rs); err == nil { + var rs stack.ReceiveBufferSizeOption + if err := s.Option(&rs); err == nil { e.rcvBufSizeMax = rs.Default } @@ -541,9 +541,9 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { case tcpip.SendBufferSizeOption: // Make sure the send buffer size is within the min and max // allowed. - var ss tcpip.StackSendBufferSizeOption - if err := e.stack.TransportProtocolOption(e.TransProto, &ss); err != nil { - panic(fmt.Sprintf("s.TransportProtocolOption(%d, %+v) = %s", e.TransProto, ss, err)) + var ss stack.SendBufferSizeOption + if err := e.stack.Option(&ss); err != nil { + panic(fmt.Sprintf("s.Option(%#v) = %s", ss, err)) } if v > ss.Max { v = ss.Max @@ -559,9 +559,9 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { case tcpip.ReceiveBufferSizeOption: // Make sure the receive buffer size is within the min and max // allowed. - var rs tcpip.StackReceiveBufferSizeOption - if err := e.stack.TransportProtocolOption(e.TransProto, &rs); err != nil { - panic(fmt.Sprintf("s.TransportProtocolOption(%d, %+v) = %s", e.TransProto, rs, err)) + var rs stack.ReceiveBufferSizeOption + if err := e.stack.Option(&rs); err != nil { + panic(fmt.Sprintf("s.Option(%#v) = %s", rs, err)) } if v > rs.Max { v = rs.Max diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 377643b82..9d4dce826 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -521,7 +521,7 @@ func (h *handshake) execute() *tcpip.Error { s.AddWaker(&h.ep.newSegmentWaker, wakerForNewSegment) defer s.Done() - var sackEnabled tcpip.StackSACKEnabled + var sackEnabled SACKEnabled if err := h.ep.stack.TransportProtocolOption(ProtocolNumber, &sackEnabled); err != nil { // If stack returned an error when checking for SACKEnabled // status then just default to switching off SACK negotiation. diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 1e4c2f507..99a691815 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -847,12 +847,12 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue maxSynRetries: DefaultSynRetries, } - var ss tcpip.StackSendBufferSizeOption + var ss SendBufferSizeOption if err := s.TransportProtocolOption(ProtocolNumber, &ss); err == nil { e.sndBufSize = ss.Default } - var rs tcpip.StackReceiveBufferSizeOption + var rs ReceiveBufferSizeOption if err := s.TransportProtocolOption(ProtocolNumber, &rs); err == nil { e.rcvBufSize = rs.Default } @@ -867,7 +867,7 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue e.rcvAutoParams.disabled = !bool(mrb) } - var de tcpip.StackDelayEnabled + var de DelayEnabled if err := s.TransportProtocolOption(ProtocolNumber, &de); err == nil && de { e.SetSockOptBool(tcpip.DelayOption, true) } @@ -1584,7 +1584,7 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { case tcpip.ReceiveBufferSizeOption: // Make sure the receive buffer size is within the min and max // allowed. - var rs tcpip.StackReceiveBufferSizeOption + var rs ReceiveBufferSizeOption if err := e.stack.TransportProtocolOption(ProtocolNumber, &rs); err == nil { if v < rs.Min { v = rs.Min @@ -1634,7 +1634,7 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { case tcpip.SendBufferSizeOption: // Make sure the send buffer size is within the min and max // allowed. - var ss tcpip.StackSendBufferSizeOption + var ss SendBufferSizeOption if err := e.stack.TransportProtocolOption(ProtocolNumber, &ss); err == nil { if v < ss.Min { v = ss.Min @@ -1674,7 +1674,7 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { return tcpip.ErrInvalidOptionValue } } - var rs tcpip.StackReceiveBufferSizeOption + var rs ReceiveBufferSizeOption if err := e.stack.TransportProtocolOption(ProtocolNumber, &rs); err == nil { if v < rs.Min/2 { v = rs.Min / 2 @@ -2595,7 +2595,7 @@ func (e *endpoint) receiveBufferSize() int { } func (e *endpoint) maxReceiveBufferSize() int { - var rs tcpip.StackReceiveBufferSizeOption + var rs ReceiveBufferSizeOption if err := e.stack.TransportProtocolOption(ProtocolNumber, &rs); err != nil { // As a fallback return the hardcoded max buffer size. return MaxBufferSize @@ -2676,7 +2676,7 @@ func timeStampOffset() uint32 { // if the SYN options indicate that the SACK option was negotiated and the TCP // stack is configured to enable TCP SACK option. func (e *endpoint) maybeEnableSACKPermitted(synOpts *header.TCPSynOptions) { - var v tcpip.StackSACKEnabled + var v SACKEnabled if err := e.stack.TransportProtocolOption(ProtocolNumber, &v); err != nil { // Stack doesn't support SACK. So just return. return diff --git a/pkg/tcpip/transport/tcp/endpoint_state.go b/pkg/tcpip/transport/tcp/endpoint_state.go index 8258c0ecc..abf1ac5c9 100644 --- a/pkg/tcpip/transport/tcp/endpoint_state.go +++ b/pkg/tcpip/transport/tcp/endpoint_state.go @@ -182,13 +182,17 @@ func (e *endpoint) Resume(s *stack.Stack) { epState := e.origEndpointState switch epState { case StateInitial, StateBound, StateListen, StateConnecting, StateEstablished: - var ss tcpip.StackSendBufferSizeOption + var ss SendBufferSizeOption if err := e.stack.TransportProtocolOption(ProtocolNumber, &ss); err == nil { if e.sndBufSize < ss.Min || e.sndBufSize > ss.Max { panic(fmt.Sprintf("endpoint.sndBufSize %d is outside the min and max allowed [%d, %d]", e.sndBufSize, ss.Min, ss.Max)) } - if e.rcvBufSize < ss.Min || e.rcvBufSize > ss.Max { - panic(fmt.Sprintf("endpoint.rcvBufSize %d is outside the min and max allowed [%d, %d]", e.rcvBufSize, ss.Min, ss.Max)) + } + + var rs ReceiveBufferSizeOption + if err := e.stack.TransportProtocolOption(ProtocolNumber, &rs); err == nil { + if e.rcvBufSize < rs.Min || e.rcvBufSize > rs.Max { + panic(fmt.Sprintf("endpoint.rcvBufSize %d is outside the min and max allowed [%d, %d]", e.rcvBufSize, rs.Min, rs.Max)) } } } diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go index 3cff55afa..f2ae6ce50 100644 --- a/pkg/tcpip/transport/tcp/protocol.go +++ b/pkg/tcpip/transport/tcp/protocol.go @@ -76,6 +76,31 @@ const ( ccCubic = "cubic" ) +// SACKEnabled is used by stack.(*Stack).TransportProtocolOption to +// enable/disable SACK support in TCP. See: https://tools.ietf.org/html/rfc2018. +type SACKEnabled bool + +// DelayEnabled is used by stack.(Stack*).TransportProtocolOption to +// enable/disable Nagle's algorithm in TCP. +type DelayEnabled bool + +// SendBufferSizeOption is used by stack.(Stack*).TransportProtocolOption +// to get/set the default, min and max TCP send buffer sizes. +type SendBufferSizeOption struct { + Min int + Default int + Max int +} + +// ReceiveBufferSizeOption is used by +// stack.(Stack*).TransportProtocolOption to get/set the default, min and max +// TCP receive buffer sizes. +type ReceiveBufferSizeOption struct { + Min int + Default int + Max int +} + // syncRcvdCounter tracks the number of endpoints in the SYN-RCVD state. The // value is protected by a mutex so that we can increment only when it's // guaranteed not to go above a threshold. @@ -137,8 +162,8 @@ type protocol struct { mu sync.RWMutex sackEnabled bool delayEnabled bool - sendBufferSize tcpip.StackSendBufferSizeOption - recvBufferSize tcpip.StackReceiveBufferSizeOption + sendBufferSize SendBufferSizeOption + recvBufferSize ReceiveBufferSizeOption congestionControl string availableCongestionControl []string moderateReceiveBuffer bool @@ -249,19 +274,19 @@ func replyWithReset(s *segment, tos, ttl uint8) { // SetOption implements stack.TransportProtocol.SetOption. func (p *protocol) SetOption(option interface{}) *tcpip.Error { switch v := option.(type) { - case tcpip.StackSACKEnabled: + case SACKEnabled: p.mu.Lock() p.sackEnabled = bool(v) p.mu.Unlock() return nil - case tcpip.StackDelayEnabled: + case DelayEnabled: p.mu.Lock() p.delayEnabled = bool(v) p.mu.Unlock() return nil - case tcpip.StackSendBufferSizeOption: + case SendBufferSizeOption: if v.Min <= 0 || v.Default < v.Min || v.Default > v.Max { return tcpip.ErrInvalidOptionValue } @@ -270,7 +295,7 @@ func (p *protocol) SetOption(option interface{}) *tcpip.Error { p.mu.Unlock() return nil - case tcpip.StackReceiveBufferSizeOption: + case ReceiveBufferSizeOption: if v.Min <= 0 || v.Default < v.Min || v.Default > v.Max { return tcpip.ErrInvalidOptionValue } @@ -363,25 +388,25 @@ func (p *protocol) SetOption(option interface{}) *tcpip.Error { // Option implements stack.TransportProtocol.Option. func (p *protocol) Option(option interface{}) *tcpip.Error { switch v := option.(type) { - case *tcpip.StackSACKEnabled: + case *SACKEnabled: p.mu.RLock() - *v = tcpip.StackSACKEnabled(p.sackEnabled) + *v = SACKEnabled(p.sackEnabled) p.mu.RUnlock() return nil - case *tcpip.StackDelayEnabled: + case *DelayEnabled: p.mu.RLock() - *v = tcpip.StackDelayEnabled(p.delayEnabled) + *v = DelayEnabled(p.delayEnabled) p.mu.RUnlock() return nil - case *tcpip.StackSendBufferSizeOption: + case *SendBufferSizeOption: p.mu.RLock() *v = p.sendBufferSize p.mu.RUnlock() return nil - case *tcpip.StackReceiveBufferSizeOption: + case *ReceiveBufferSizeOption: p.mu.RLock() *v = p.recvBufferSize p.mu.RUnlock() @@ -491,12 +516,12 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) bool { // NewProtocol returns a TCP transport protocol. func NewProtocol() stack.TransportProtocol { return &protocol{ - sendBufferSize: tcpip.StackSendBufferSizeOption{ + sendBufferSize: SendBufferSizeOption{ Min: MinBufferSize, Default: DefaultSendBufferSize, Max: MaxBufferSize, }, - recvBufferSize: tcpip.StackReceiveBufferSizeOption{ + recvBufferSize: ReceiveBufferSizeOption{ Min: MinBufferSize, Default: DefaultReceiveBufferSize, Max: MaxBufferSize, diff --git a/pkg/tcpip/transport/tcp/tcp_sack_test.go b/pkg/tcpip/transport/tcp/tcp_sack_test.go index 812e503bc..99521f0c1 100644 --- a/pkg/tcpip/transport/tcp/tcp_sack_test.go +++ b/pkg/tcpip/transport/tcp/tcp_sack_test.go @@ -46,8 +46,8 @@ func createConnectedWithSACKAndTS(c *context.Context) *context.RawEndpoint { func setStackSACKPermitted(t *testing.T, c *context.Context, enable bool) { t.Helper() - if err := c.Stack().SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackSACKEnabled(enable)); err != nil { - t.Fatalf("c.s.SetTransportProtocolOption(tcp.ProtocolNumber, StackSACKEnabled(%t) = %s", enable, err) + if err := c.Stack().SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SACKEnabled(enable)); err != nil { + t.Fatalf("c.s.SetTransportProtocolOption(tcp.ProtocolNumber, SACKEnabled(%t) = %s", enable, err) } } diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index 2632a3c67..169adb16b 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -4005,7 +4005,7 @@ func TestDefaultBufferSizes(t *testing.T) { checkRecvBufferSize(t, ep, tcp.DefaultReceiveBufferSize) // Change the default send buffer size. - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackSendBufferSizeOption{ + if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SendBufferSizeOption{ Min: 1, Default: tcp.DefaultSendBufferSize * 2, Max: tcp.DefaultSendBufferSize * 20}); err != nil { @@ -4022,7 +4022,7 @@ func TestDefaultBufferSizes(t *testing.T) { checkRecvBufferSize(t, ep, tcp.DefaultReceiveBufferSize) // Change the default receive buffer size. - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackReceiveBufferSizeOption{ + if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.ReceiveBufferSizeOption{ Min: 1, Default: tcp.DefaultReceiveBufferSize * 3, Max: tcp.DefaultReceiveBufferSize * 30}); err != nil { @@ -4053,11 +4053,11 @@ func TestMinMaxBufferSizes(t *testing.T) { defer ep.Close() // Change the min/max values for send/receive - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackReceiveBufferSizeOption{Min: 200, Default: tcp.DefaultReceiveBufferSize * 2, Max: tcp.DefaultReceiveBufferSize * 20}); err != nil { + if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.ReceiveBufferSizeOption{Min: 200, Default: tcp.DefaultReceiveBufferSize * 2, Max: tcp.DefaultReceiveBufferSize * 20}); err != nil { t.Fatalf("SetTransportProtocolOption failed: %s", err) } - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackSendBufferSizeOption{Min: 300, Default: tcp.DefaultSendBufferSize * 3, Max: tcp.DefaultSendBufferSize * 30}); err != nil { + if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SendBufferSizeOption{Min: 300, Default: tcp.DefaultSendBufferSize * 3, Max: tcp.DefaultSendBufferSize * 30}); err != nil { t.Fatalf("SetTransportProtocolOption failed: %s", err) } @@ -5696,7 +5696,7 @@ func TestReceiveBufferAutoTuningApplicationLimited(t *testing.T) { // the segment queue holding unprocessed packets is limited to 500. const receiveBufferSize = 80 << 10 // 80KB. const maxReceiveBufferSize = receiveBufferSize * 10 - if err := stk.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackReceiveBufferSizeOption{Min: 1, Default: receiveBufferSize, Max: maxReceiveBufferSize}); err != nil { + if err := stk.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.ReceiveBufferSizeOption{Min: 1, Default: receiveBufferSize, Max: maxReceiveBufferSize}); err != nil { t.Fatalf("SetTransportProtocolOption failed: %s", err) } @@ -5817,7 +5817,7 @@ func TestReceiveBufferAutoTuning(t *testing.T) { // the segment queue holding unprocessed packets is limited to 300. const receiveBufferSize = 80 << 10 // 80KB. const maxReceiveBufferSize = receiveBufferSize * 10 - if err := stk.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackReceiveBufferSizeOption{Min: 1, Default: receiveBufferSize, Max: maxReceiveBufferSize}); err != nil { + if err := stk.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.ReceiveBufferSizeOption{Min: 1, Default: receiveBufferSize, Max: maxReceiveBufferSize}); err != nil { t.Fatalf("SetTransportProtocolOption failed: %s", err) } @@ -5959,7 +5959,7 @@ func TestDelayEnabled(t *testing.T) { checkDelayOption(t, c, false, false) // Delay is disabled by default. for _, v := range []struct { - delayEnabled tcpip.StackDelayEnabled + delayEnabled tcp.DelayEnabled wantDelayOption bool }{ {delayEnabled: false, wantDelayOption: false}, @@ -5974,10 +5974,10 @@ func TestDelayEnabled(t *testing.T) { } } -func checkDelayOption(t *testing.T, c *context.Context, wantDelayEnabled tcpip.StackDelayEnabled, wantDelayOption bool) { +func checkDelayOption(t *testing.T, c *context.Context, wantDelayEnabled tcp.DelayEnabled, wantDelayOption bool) { t.Helper() - var gotDelayEnabled tcpip.StackDelayEnabled + var gotDelayEnabled tcp.DelayEnabled if err := c.Stack().TransportProtocolOption(tcp.ProtocolNumber, &gotDelayEnabled); err != nil { t.Fatalf("TransportProtocolOption(tcp, &gotDelayEnabled) failed: %s", err) } diff --git a/pkg/tcpip/transport/tcp/testing/context/context.go b/pkg/tcpip/transport/tcp/testing/context/context.go index 9e262c272..06fde2a79 100644 --- a/pkg/tcpip/transport/tcp/testing/context/context.go +++ b/pkg/tcpip/transport/tcp/testing/context/context.go @@ -144,11 +144,11 @@ func New(t *testing.T, mtu uint32) *Context { }) // Allow minimum send/receive buffer sizes to be 1 during tests. - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackSendBufferSizeOption{Min: 1, Default: tcp.DefaultSendBufferSize, Max: 10 * tcp.DefaultSendBufferSize}); err != nil { + if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SendBufferSizeOption{Min: 1, Default: tcp.DefaultSendBufferSize, Max: 10 * tcp.DefaultSendBufferSize}); err != nil { t.Fatalf("SetTransportProtocolOption failed: %s", err) } - if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackReceiveBufferSizeOption{Min: 1, Default: tcp.DefaultReceiveBufferSize, Max: 10 * tcp.DefaultReceiveBufferSize}); err != nil { + if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.ReceiveBufferSizeOption{Min: 1, Default: tcp.DefaultReceiveBufferSize, Max: 10 * tcp.DefaultReceiveBufferSize}); err != nil { t.Fatalf("SetTransportProtocolOption failed: %s", err) } @@ -1091,7 +1091,7 @@ func (c *Context) PassiveConnectWithOptions(maxPayload, wndScale int, synOptions // SACKEnabled returns true if the TCP Protocol option SACKEnabled is set to true // for the Stack in the context. func (c *Context) SACKEnabled() bool { - var v tcpip.StackSACKEnabled + var v tcp.SACKEnabled if err := c.Stack().TransportProtocolOption(tcp.ProtocolNumber, &v); err != nil { // Stack doesn't support SACK. So just return. return false diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 6ea212093..8bdc1ee1f 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -190,13 +190,13 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue } // Override with stack defaults. - var ss tcpip.StackSendBufferSizeOption - if err := s.TransportProtocolOption(ProtocolNumber, &ss); err == nil { + var ss stack.SendBufferSizeOption + if err := s.Option(&ss); err == nil { e.sndBufSizeMax = ss.Default } - var rs tcpip.StackReceiveBufferSizeOption - if err := s.TransportProtocolOption(ProtocolNumber, &rs); err == nil { + var rs stack.ReceiveBufferSizeOption + if err := s.Option(&rs); err == nil { e.rcvBufSizeMax = rs.Default } @@ -629,9 +629,9 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { case tcpip.ReceiveBufferSizeOption: // Make sure the receive buffer size is within the min and max // allowed. - var rs tcpip.StackReceiveBufferSizeOption - if err := e.stack.TransportProtocolOption(ProtocolNumber, &rs); err != nil { - panic(fmt.Sprintf("e.stack.TransportProtocolOption(%d, %+v) = %s", ProtocolNumber, rs, err)) + var rs stack.ReceiveBufferSizeOption + if err := e.stack.Option(&rs); err != nil { + panic(fmt.Sprintf("e.stack.Option(%#v) = %s", rs, err)) } if v < rs.Min { @@ -648,9 +648,9 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { case tcpip.SendBufferSizeOption: // Make sure the send buffer size is within the min and max // allowed. - var ss tcpip.StackSendBufferSizeOption - if err := e.stack.TransportProtocolOption(ProtocolNumber, &ss); err != nil { - panic(fmt.Sprintf("e.stack.TransportProtocolOption(%d, %+v) = %s", ProtocolNumber, ss, err)) + var ss stack.SendBufferSizeOption + if err := e.stack.Option(&ss); err != nil { + panic(fmt.Sprintf("e.stack.Option(%#v) = %s", ss, err)) } if v < ss.Min { diff --git a/pkg/tcpip/transport/udp/protocol.go b/pkg/tcpip/transport/udp/protocol.go index fc93f93c0..0e7464e3a 100644 --- a/pkg/tcpip/transport/udp/protocol.go +++ b/pkg/tcpip/transport/udp/protocol.go @@ -21,7 +21,6 @@ package udp import ( - "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" @@ -50,9 +49,6 @@ const ( ) type protocol struct { - mu sync.RWMutex - sendBufferSize tcpip.StackSendBufferSizeOption - recvBufferSize tcpip.StackReceiveBufferSizeOption } // Number returns the udp protocol number. @@ -203,48 +199,12 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans // SetOption implements stack.TransportProtocol.SetOption. func (p *protocol) SetOption(option interface{}) *tcpip.Error { - switch v := option.(type) { - case tcpip.StackSendBufferSizeOption: - if v.Min <= 0 || v.Default < v.Min || v.Default > v.Max { - return tcpip.ErrInvalidOptionValue - } - p.mu.Lock() - p.sendBufferSize = v - p.mu.Unlock() - return nil - - case tcpip.StackReceiveBufferSizeOption: - if v.Min <= 0 || v.Default < v.Min || v.Default > v.Max { - return tcpip.ErrInvalidOptionValue - } - p.mu.Lock() - p.recvBufferSize = v - p.mu.Unlock() - return nil - - default: - return tcpip.ErrUnknownProtocolOption - } + return tcpip.ErrUnknownProtocolOption } // Option implements stack.TransportProtocol.Option. func (p *protocol) Option(option interface{}) *tcpip.Error { - switch v := option.(type) { - case *tcpip.StackSendBufferSizeOption: - p.mu.RLock() - *v = p.sendBufferSize - p.mu.RUnlock() - return nil - - case *tcpip.StackReceiveBufferSizeOption: - p.mu.RLock() - *v = p.recvBufferSize - p.mu.RUnlock() - return nil - - default: - return tcpip.ErrUnknownProtocolOption - } + return tcpip.ErrUnknownProtocolOption } // Close implements stack.TransportProtocol.Close. @@ -267,8 +227,5 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) bool { // NewProtocol returns a UDP transport protocol. func NewProtocol() stack.TransportProtocol { - return &protocol{ - sendBufferSize: tcpip.StackSendBufferSizeOption{Min: MinBufferSize, Default: DefaultSendBufferSize, Max: MaxBufferSize}, - recvBufferSize: tcpip.StackReceiveBufferSizeOption{Min: MinBufferSize, Default: DefaultReceiveBufferSize, Max: MaxBufferSize}, - } + return &protocol{} } diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index 081db39c1..b5df1deb9 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -1058,7 +1058,7 @@ func newEmptySandboxNetworkStack(clock tcpip.Clock, uniqueID stack.UniqueID) (in })} // Enable SACK Recovery. - if err := s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.StackSACKEnabled(true)); err != nil { + if err := s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SACKEnabled(true)); err != nil { return nil, fmt.Errorf("failed to enable SACK: %s", err) } -- cgit v1.2.3 From 10930b0f8c1ff2ac83c7a30cc1f78112a35e3183 Mon Sep 17 00:00:00 2001 From: Tamir Duberstein Date: Wed, 24 Jun 2020 13:54:58 -0700 Subject: Remove waiter.Entry.Context This field is redundant since state can be stored in the callback. PiperOrigin-RevId: 318134855 --- pkg/sentry/kernel/epoll/epoll.go | 13 +++++++------ pkg/sentry/kernel/epoll/epoll_state.go | 3 +-- pkg/waiter/waiter.go | 18 ++++++------------ 3 files changed, 14 insertions(+), 20 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/epoll/epoll.go b/pkg/sentry/kernel/epoll/epoll.go index 3d78cd48f..679ab495d 100644 --- a/pkg/sentry/kernel/epoll/epoll.go +++ b/pkg/sentry/kernel/epoll/epoll.go @@ -271,11 +271,13 @@ func (e *EventPoll) ReadEvents(max int) []linux.EpollEvent { // readyCallback is called when one of the files we're polling becomes ready. It // moves said file to the readyList if it's currently in the waiting list. -type readyCallback struct{} +type readyCallback struct { + context *pollEntry +} // Callback implements waiter.EntryCallback.Callback. -func (*readyCallback) Callback(w *waiter.Entry) { - entry := w.Context.(*pollEntry) +func (r *readyCallback) Callback(*waiter.Entry) { + entry := r.context e := entry.epoll e.listsMu.Lock() @@ -310,7 +312,7 @@ func (e *EventPoll) initEntryReadiness(entry *pollEntry) { // Check if the file happens to already be in a ready state. ready := f.Readiness(entry.mask) & entry.mask if ready != 0 { - (*readyCallback).Callback(nil, &entry.waiter) + (&readyCallback{context: entry}).Callback(&entry.waiter) } } @@ -380,10 +382,9 @@ func (e *EventPoll) AddEntry(id FileIdentifier, flags EntryFlags, mask waiter.Ev userData: data, epoll: e, flags: flags, - waiter: waiter.Entry{Callback: &readyCallback{}}, mask: mask, } - entry.waiter.Context = entry + entry.waiter.Callback = &readyCallback{context: entry} e.files[id] = entry entry.file = refs.NewWeakRef(id.File, entry) diff --git a/pkg/sentry/kernel/epoll/epoll_state.go b/pkg/sentry/kernel/epoll/epoll_state.go index 8e9f200d0..02f9aabfa 100644 --- a/pkg/sentry/kernel/epoll/epoll_state.go +++ b/pkg/sentry/kernel/epoll/epoll_state.go @@ -21,8 +21,7 @@ import ( // afterLoad is invoked by stateify. func (p *pollEntry) afterLoad() { - p.waiter = waiter.Entry{Callback: &readyCallback{}} - p.waiter.Context = p + p.waiter.Callback = &readyCallback{context: p} p.file = refs.NewWeakRef(p.id.File, p) p.id.File.EventRegister(&p.waiter, p.mask) } diff --git a/pkg/waiter/waiter.go b/pkg/waiter/waiter.go index 707eb085b..67a950444 100644 --- a/pkg/waiter/waiter.go +++ b/pkg/waiter/waiter.go @@ -128,13 +128,6 @@ type EntryCallback interface { // // +stateify savable type Entry struct { - // Context stores any state the waiter may wish to store in the entry - // itself, which may be used at wake up time. - // - // Note that use of this field is optional and state may alternatively be - // stored in the callback itself. - Context interface{} - Callback EntryCallback // The following fields are protected by the queue lock. @@ -142,13 +135,14 @@ type Entry struct { waiterEntry } -type channelCallback struct{} +type channelCallback struct { + ch chan struct{} +} // Callback implements EntryCallback.Callback. -func (*channelCallback) Callback(e *Entry) { - ch := e.Context.(chan struct{}) +func (c *channelCallback) Callback(*Entry) { select { - case ch <- struct{}{}: + case c.ch <- struct{}{}: default: } } @@ -164,7 +158,7 @@ func NewChannelEntry(c chan struct{}) (Entry, chan struct{}) { c = make(chan struct{}, 1) } - return Entry{Context: c, Callback: &channelCallback{}}, c + return Entry{Callback: &channelCallback{ch: c}}, c } // Queue represents the wait queue where waiters can be added and -- cgit v1.2.3 From 58880bf551f4fdaeecf0a355816f1af353ef81b6 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Wed, 24 Jun 2020 16:21:53 -0700 Subject: Port /dev/net/tun device to VFS2. Updates #2912 #1035 PiperOrigin-RevId: 318162565 --- pkg/sentry/devices/tundev/BUILD | 23 +++++ pkg/sentry/devices/tundev/tundev.go | 178 +++++++++++++++++++++++++++++++++ pkg/sentry/fsimpl/devtmpfs/devtmpfs.go | 16 ++- runsc/boot/BUILD | 1 + runsc/boot/vfs.go | 11 +- test/syscalls/BUILD | 1 + 6 files changed, 227 insertions(+), 3 deletions(-) create mode 100644 pkg/sentry/devices/tundev/BUILD create mode 100644 pkg/sentry/devices/tundev/tundev.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/devices/tundev/BUILD b/pkg/sentry/devices/tundev/BUILD new file mode 100644 index 000000000..71c59287c --- /dev/null +++ b/pkg/sentry/devices/tundev/BUILD @@ -0,0 +1,23 @@ +load("//tools:defs.bzl", "go_library") + +licenses(["notice"]) + +go_library( + name = "tundev", + srcs = ["tundev.go"], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/abi/linux", + "//pkg/context", + "//pkg/sentry/arch", + "//pkg/sentry/fsimpl/devtmpfs", + "//pkg/sentry/inet", + "//pkg/sentry/kernel", + "//pkg/sentry/socket/netstack", + "//pkg/sentry/vfs", + "//pkg/syserror", + "//pkg/tcpip/link/tun", + "//pkg/usermem", + "//pkg/waiter", + ], +) diff --git a/pkg/sentry/devices/tundev/tundev.go b/pkg/sentry/devices/tundev/tundev.go new file mode 100644 index 000000000..dfbd069af --- /dev/null +++ b/pkg/sentry/devices/tundev/tundev.go @@ -0,0 +1,178 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package tundev implements the /dev/net/tun device. +package tundev + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/devtmpfs" + "gvisor.dev/gvisor/pkg/sentry/inet" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/socket/netstack" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/tcpip/link/tun" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" +) + +const ( + netTunDevMajor = 10 + netTunDevMinor = 200 +) + +// tunDevice implements vfs.Device for /dev/net/tun. +type tunDevice struct{} + +// Open implements vfs.Device.Open. +func (tunDevice) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd := &tunFD{} + if err := fd.vfsfd.Init(fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{ + UseDentryMetadata: true, + }); err != nil { + return nil, err + } + return &fd.vfsfd, nil +} + +// tunFD implements vfs.FileDescriptionImpl for /dev/net/tun. +type tunFD struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl + vfs.DentryMetadataFileDescriptionImpl + vfs.NoLockFD + + device tun.Device +} + +// Ioctl implements vfs.FileDescriptionImpl.Ioctl. +func (fd *tunFD) Ioctl(ctx context.Context, uio usermem.IO, args arch.SyscallArguments) (uintptr, error) { + request := args[1].Uint() + data := args[2].Pointer() + + switch request { + case linux.TUNSETIFF: + t := kernel.TaskFromContext(ctx) + if t == nil { + panic("Ioctl should be called from a task context") + } + if !t.HasCapability(linux.CAP_NET_ADMIN) { + return 0, syserror.EPERM + } + stack, ok := t.NetworkContext().(*netstack.Stack) + if !ok { + return 0, syserror.EINVAL + } + + var req linux.IFReq + if _, err := usermem.CopyObjectIn(ctx, uio, data, &req, usermem.IOOpts{ + AddressSpaceActive: true, + }); err != nil { + return 0, err + } + flags := usermem.ByteOrder.Uint16(req.Data[:]) + return 0, fd.device.SetIff(stack.Stack, req.Name(), flags) + + case linux.TUNGETIFF: + var req linux.IFReq + + copy(req.IFName[:], fd.device.Name()) + + // Linux adds IFF_NOFILTER (the same value as IFF_NO_PI unfortunately) when + // there is no sk_filter. See __tun_chr_ioctl() in net/drivers/tun.c. + flags := fd.device.Flags() | linux.IFF_NOFILTER + usermem.ByteOrder.PutUint16(req.Data[:], flags) + + _, err := usermem.CopyObjectOut(ctx, uio, data, &req, usermem.IOOpts{ + AddressSpaceActive: true, + }) + return 0, err + + default: + return 0, syserror.ENOTTY + } +} + +// Release implements vfs.FileDescriptionImpl.Release. +func (fd *tunFD) Release() { + fd.device.Release() +} + +// PRead implements vfs.FileDescriptionImpl.PRead. +func (fd *tunFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + return fd.Read(ctx, dst, opts) +} + +// Read implements vfs.FileDescriptionImpl.Read. +func (fd *tunFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + data, err := fd.device.Read() + if err != nil { + return 0, err + } + n, err := dst.CopyOut(ctx, data) + if n > 0 && n < len(data) { + // Not an error for partial copying. Packet truncated. + err = nil + } + return int64(n), err +} + +// PWrite implements vfs.FileDescriptionImpl.PWrite. +func (fd *tunFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + return fd.Write(ctx, src, opts) +} + +// Write implements vfs.FileDescriptionImpl.Write. +func (fd *tunFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + data := make([]byte, src.NumBytes()) + if _, err := src.CopyIn(ctx, data); err != nil { + return 0, err + } + return fd.device.Write(data) +} + +// Readiness implements watier.Waitable.Readiness. +func (fd *tunFD) Readiness(mask waiter.EventMask) waiter.EventMask { + return fd.device.Readiness(mask) +} + +// EventRegister implements watier.Waitable.EventRegister. +func (fd *tunFD) EventRegister(e *waiter.Entry, mask waiter.EventMask) { + fd.device.EventRegister(e, mask) +} + +// EventUnregister implements watier.Waitable.EventUnregister. +func (fd *tunFD) EventUnregister(e *waiter.Entry) { + fd.device.EventUnregister(e) +} + +// isNetTunSupported returns whether /dev/net/tun device is supported for s. +func isNetTunSupported(s inet.Stack) bool { + _, ok := s.(*netstack.Stack) + return ok +} + +// Register registers all devices implemented by this package in vfsObj. +func Register(vfsObj *vfs.VirtualFilesystem) error { + return vfsObj.RegisterDevice(vfs.CharDevice, netTunDevMajor, netTunDevMinor, tunDevice{}, &vfs.RegisterDeviceOptions{}) +} + +// CreateDevtmpfsFiles creates device special files in dev representing all +// devices implemented by this package. +func CreateDevtmpfsFiles(ctx context.Context, dev *devtmpfs.Accessor) error { + return dev.CreateDeviceFile(ctx, "net/tun", vfs.CharDevice, netTunDevMajor, netTunDevMinor, 0666 /* mode */) +} diff --git a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go index 142ee53b0..d0e06cdc0 100644 --- a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go +++ b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go @@ -136,6 +136,8 @@ func (a *Accessor) pathOperationAt(pathname string) *vfs.PathOperation { // CreateDeviceFile creates a device special file at the given pathname in the // devtmpfs instance accessed by the Accessor. func (a *Accessor) CreateDeviceFile(ctx context.Context, pathname string, kind vfs.DeviceKind, major, minor uint32, perms uint16) error { + actx := a.wrapContext(ctx) + mode := (linux.FileMode)(perms) switch kind { case vfs.BlockDevice: @@ -145,12 +147,24 @@ func (a *Accessor) CreateDeviceFile(ctx context.Context, pathname string, kind v default: panic(fmt.Sprintf("invalid vfs.DeviceKind: %v", kind)) } + + // Create any parent directories. See + // devtmpfs.c:handle_create()=>path_create(). + for it := fspath.Parse(pathname).Begin; it.NextOk(); it = it.Next() { + pop := a.pathOperationAt(it.String()) + if err := a.vfsObj.MkdirAt(actx, a.creds, pop, &vfs.MkdirOptions{ + Mode: 0755, + }); err != nil { + return fmt.Errorf("failed to create directory %q: %v", it.String(), err) + } + } + // NOTE: Linux's devtmpfs refuses to automatically delete files it didn't // create, which it recognizes by storing a pointer to the kdevtmpfs struct // thread in struct inode::i_private. Accessor doesn't yet support deletion // of files at all, and probably won't as long as we don't need to support // kernel modules, so this is moot for now. - return a.vfsObj.MknodAt(a.wrapContext(ctx), a.creds, a.pathOperationAt(pathname), &vfs.MknodOptions{ + return a.vfsObj.MknodAt(actx, a.creds, a.pathOperationAt(pathname), &vfs.MknodOptions{ Mode: mode, DevMajor: major, DevMinor: minor, diff --git a/runsc/boot/BUILD b/runsc/boot/BUILD index b701c65d2..45d7f7d09 100644 --- a/runsc/boot/BUILD +++ b/runsc/boot/BUILD @@ -41,6 +41,7 @@ go_library( "//pkg/sentry/control", "//pkg/sentry/devices/memdev", "//pkg/sentry/devices/ttydev", + "//pkg/sentry/devices/tundev", "//pkg/sentry/fdimport", "//pkg/sentry/fs", "//pkg/sentry/fs/dev", diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go index 2fdddc719..9ba5e9cd9 100644 --- a/runsc/boot/vfs.go +++ b/runsc/boot/vfs.go @@ -27,6 +27,7 @@ import ( "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/devices/memdev" "gvisor.dev/gvisor/pkg/sentry/devices/ttydev" + "gvisor.dev/gvisor/pkg/sentry/devices/tundev" "gvisor.dev/gvisor/pkg/sentry/fs/user" "gvisor.dev/gvisor/pkg/sentry/fsimpl/devpts" "gvisor.dev/gvisor/pkg/sentry/fsimpl/devtmpfs" @@ -79,6 +80,9 @@ func registerFilesystems(ctx context.Context, vfsObj *vfs.VirtualFilesystem, cre if err := ttydev.Register(vfsObj); err != nil { return fmt.Errorf("registering ttydev: %w", err) } + if err := tundev.Register(vfsObj); err != nil { + return fmt.Errorf("registering tundev: %v", err) + } a, err := devtmpfs.NewAccessor(ctx, vfsObj, creds, devtmpfs.Name) if err != nil { return fmt.Errorf("creating devtmpfs accessor: %w", err) @@ -89,10 +93,13 @@ func registerFilesystems(ctx context.Context, vfsObj *vfs.VirtualFilesystem, cre return fmt.Errorf("initializing userspace: %w", err) } if err := memdev.CreateDevtmpfsFiles(ctx, a); err != nil { - return fmt.Errorf("creating devtmpfs files: %w", err) + return fmt.Errorf("creating memdev devtmpfs files: %w", err) } if err := ttydev.CreateDevtmpfsFiles(ctx, a); err != nil { - return fmt.Errorf("creating devtmpfs files: %w", err) + return fmt.Errorf("creating ttydev devtmpfs files: %w", err) + } + if err := tundev.CreateDevtmpfsFiles(ctx, a); err != nil { + return fmt.Errorf("creating tundev devtmpfs files: %v", err) } return nil } diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 315af9efe..39c1f79ad 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -1013,6 +1013,7 @@ syscall_test( syscall_test( test = "//test/syscalls/linux:tuntap_test", + vfs2 = "True", ) syscall_test( -- cgit v1.2.3 From b5e814445a4db5df7f4f58027422a5dba97ea766 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Wed, 24 Jun 2020 19:20:58 -0700 Subject: Fix procfs bugs in vfs2. - Support writing on proc/[pid]/{uid,gid}map - Return EIO for writing to static files. Updates #2923. PiperOrigin-RevId: 318188503 --- pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go | 4 +- pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 2 + pkg/sentry/fsimpl/proc/task_files.go | 60 ++++++++++++++++++++++- pkg/sentry/vfs/file_description_impl_util.go | 2 +- pkg/sentry/vfs/file_description_impl_util_test.go | 8 +-- test/syscalls/BUILD | 2 + test/syscalls/linux/proc.cc | 49 +++++++++++++++++- 7 files changed, 117 insertions(+), 10 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go index c1215b70a..6886b0876 100644 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -101,12 +101,12 @@ func (fd *DynamicBytesFD) Seek(ctx context.Context, offset int64, whence int32) return fd.DynamicBytesFileDescriptionImpl.Seek(ctx, offset, whence) } -// Read implmenets vfs.FileDescriptionImpl.Read. +// Read implements vfs.FileDescriptionImpl.Read. func (fd *DynamicBytesFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { return fd.DynamicBytesFileDescriptionImpl.Read(ctx, dst, opts) } -// PRead implmenets vfs.FileDescriptionImpl.PRead. +// PRead implements vfs.FileDescriptionImpl.PRead. func (fd *DynamicBytesFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { return fd.DynamicBytesFileDescriptionImpl.PRead(ctx, dst, offset, opts) } diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 650bd7b88..53aec4918 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -293,6 +293,8 @@ func (a *InodeAttrs) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *aut // inode numbers are immutable after node creation. // TODO(gvisor.dev/issue/1193): Implement other stat fields like timestamps. + // Also, STATX_SIZE will need some special handling, because read-only static + // files should return EIO for truncate operations. return nil } diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index ba4405026..286c23f01 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -35,6 +35,10 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) +// "There is an (arbitrary) limit on the number of lines in the file. As at +// Linux 3.18, the limit is five lines." - user_namespaces(7) +const maxIDMapLines = 5 + // mm gets the kernel task's MemoryManager. No additional reference is taken on // mm here. This is safe because MemoryManager.destroy is required to leave the // MemoryManager in a state where it's still usable as a DynamicBytesSource. @@ -283,7 +287,8 @@ func (d *commData) Generate(ctx context.Context, buf *bytes.Buffer) error { return nil } -// idMapData implements vfs.DynamicBytesSource for /proc/[pid]/{gid_map|uid_map}. +// idMapData implements vfs.WritableDynamicBytesSource for +// /proc/[pid]/{gid_map|uid_map}. // // +stateify savable type idMapData struct { @@ -309,6 +314,59 @@ func (d *idMapData) Generate(ctx context.Context, buf *bytes.Buffer) error { return nil } +func (d *idMapData) Write(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) { + // "In addition, the number of bytes written to the file must be less than + // the system page size, and the write must be performed at the start of + // the file ..." - user_namespaces(7) + srclen := src.NumBytes() + if srclen >= usermem.PageSize || offset != 0 { + return 0, syserror.EINVAL + } + b := make([]byte, srclen) + if _, err := src.CopyIn(ctx, b); err != nil { + return 0, err + } + + // Truncate from the first NULL byte. + var nul int64 + nul = int64(bytes.IndexByte(b, 0)) + if nul == -1 { + nul = srclen + } + b = b[:nul] + // Remove the last \n. + if nul >= 1 && b[nul-1] == '\n' { + b = b[:nul-1] + } + lines := bytes.SplitN(b, []byte("\n"), maxIDMapLines+1) + if len(lines) > maxIDMapLines { + return 0, syserror.EINVAL + } + + entries := make([]auth.IDMapEntry, len(lines)) + for i, l := range lines { + var e auth.IDMapEntry + _, err := fmt.Sscan(string(l), &e.FirstID, &e.FirstParentID, &e.Length) + if err != nil { + return 0, syserror.EINVAL + } + entries[i] = e + } + var err error + if d.gids { + err = d.task.UserNamespace().SetGIDMap(ctx, entries) + } else { + err = d.task.UserNamespace().SetUIDMap(ctx, entries) + } + if err != nil { + return 0, err + } + + // On success, Linux's kernel/user_namespace.c:map_write() always returns + // count, even if fewer bytes were used. + return int64(srclen), nil +} + // mapsData implements vfs.DynamicBytesSource for /proc/[pid]/maps. // // +stateify savable diff --git a/pkg/sentry/vfs/file_description_impl_util.go b/pkg/sentry/vfs/file_description_impl_util.go index 1e66997ce..3fec0d6d6 100644 --- a/pkg/sentry/vfs/file_description_impl_util.go +++ b/pkg/sentry/vfs/file_description_impl_util.go @@ -327,7 +327,7 @@ func (fd *DynamicBytesFileDescriptionImpl) pwriteLocked(ctx context.Context, src writable, ok := fd.data.(WritableDynamicBytesSource) if !ok { - return 0, syserror.EINVAL + return 0, syserror.EIO } n, err := writable.Write(ctx, src, offset) if err != nil { diff --git a/pkg/sentry/vfs/file_description_impl_util_test.go b/pkg/sentry/vfs/file_description_impl_util_test.go index 5061f6ac9..3b7e1c273 100644 --- a/pkg/sentry/vfs/file_description_impl_util_test.go +++ b/pkg/sentry/vfs/file_description_impl_util_test.go @@ -155,11 +155,11 @@ func TestGenCountFD(t *testing.T) { } // Write and PWrite fails. - if _, err := fd.Write(ctx, ioseq, WriteOptions{}); err != syserror.EINVAL { - t.Errorf("Write: got err %v, wanted %v", err, syserror.EINVAL) + if _, err := fd.Write(ctx, ioseq, WriteOptions{}); err != syserror.EIO { + t.Errorf("Write: got err %v, wanted %v", err, syserror.EIO) } - if _, err := fd.PWrite(ctx, ioseq, 0, WriteOptions{}); err != syserror.EINVAL { - t.Errorf("Write: got err %v, wanted %v", err, syserror.EINVAL) + if _, err := fd.PWrite(ctx, ioseq, 0, WriteOptions{}); err != syserror.EIO { + t.Errorf("Write: got err %v, wanted %v", err, syserror.EIO) } } diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 39c1f79ad..23123006f 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -479,6 +479,7 @@ syscall_test( syscall_test( size = "medium", test = "//test/syscalls/linux:proc_test", + vfs2 = "True", ) syscall_test( @@ -498,6 +499,7 @@ syscall_test( syscall_test( test = "//test/syscalls/linux:proc_pid_uid_gid_map_test", + vfs2 = "True", ) syscall_test( diff --git a/test/syscalls/linux/proc.cc b/test/syscalls/linux/proc.cc index e77e9820e..d6b875dbf 100644 --- a/test/syscalls/linux/proc.cc +++ b/test/syscalls/linux/proc.cc @@ -754,8 +754,53 @@ TEST(ProcCpuinfo, RequiredFieldsArePresent) { } } -TEST(ProcCpuinfo, DeniesWrite) { - EXPECT_THAT(open("/proc/cpuinfo", O_WRONLY), SyscallFailsWithErrno(EACCES)); +TEST(ProcCpuinfo, DeniesWriteNonRoot) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_FOWNER))); + + // Do setuid in a separate thread so that after finishing this test, the + // process can still open files the test harness created before starting this + // test. Otherwise, the files are created by root (UID before the test), but + // cannot be opened by the `uid` set below after the test. After calling + // setuid(non-zero-UID), there is no way to get root privileges back. + ScopedThread([&] { + // Use syscall instead of glibc setuid wrapper because we want this setuid + // call to only apply to this task. POSIX threads, however, require that all + // threads have the same UIDs, so using the setuid wrapper sets all threads' + // real UID. + // Also drops capabilities. + constexpr int kNobody = 65534; + EXPECT_THAT(syscall(SYS_setuid, kNobody), SyscallSucceeds()); + EXPECT_THAT(open("/proc/cpuinfo", O_WRONLY), SyscallFailsWithErrno(EACCES)); + // TODO(gvisor.dev/issue/1193): Properly support setting size attributes in + // kernfs. + if (!IsRunningOnGvisor() || IsRunningWithVFS1()) { + EXPECT_THAT(truncate("/proc/cpuinfo", 123), + SyscallFailsWithErrno(EACCES)); + } + }); +} + +// With root privileges, it is possible to open /proc/cpuinfo with write mode, +// but all write operations will return EIO. +TEST(ProcCpuinfo, DeniesWriteRoot) { + // VFS1 does not behave differently for root/non-root. + SKIP_IF(IsRunningWithVFS1()); + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_FOWNER))); + + int fd; + EXPECT_THAT(fd = open("/proc/cpuinfo", O_WRONLY), SyscallSucceeds()); + if (fd > 0) { + EXPECT_THAT(write(fd, "x", 1), SyscallFailsWithErrno(EIO)); + EXPECT_THAT(pwrite(fd, "x", 1, 123), SyscallFailsWithErrno(EIO)); + } + // TODO(gvisor.dev/issue/1193): Properly support setting size attributes in + // kernfs. + if (!IsRunningOnGvisor() || IsRunningWithVFS1()) { + if (fd > 0) { + EXPECT_THAT(ftruncate(fd, 123), SyscallFailsWithErrno(EIO)); + } + EXPECT_THAT(truncate("/proc/cpuinfo", 123), SyscallFailsWithErrno(EIO)); + } } // Sanity checks that uptime is present. -- cgit v1.2.3 From bd5f0e2dc42866f28437f07b1a24e19d1748c3ea Mon Sep 17 00:00:00 2001 From: Ridwan Sharif Date: Tue, 9 Jun 2020 12:35:39 -0400 Subject: Add FUSE character device This change adds a FUSE character device backed by devtmpfs. This device will be used to establish a connection between the FUSE server daemon and fusefs. The FileDescriptionImpl methods will be implemented as we flesh out fusefs some more. The tests assert that the device can be opened and used. --- pkg/abi/linux/dev.go | 4 ++ pkg/sentry/devices/miscdev/BUILD | 20 +++++++++ pkg/sentry/devices/miscdev/fuse.go | 78 +++++++++++++++++++++++++++++++++++ pkg/sentry/devices/miscdev/miscdev.go | 54 ++++++++++++++++++++++++ test/syscalls/linux/dev.cc | 20 +++++++++ 5 files changed, 176 insertions(+) create mode 100644 pkg/sentry/devices/miscdev/BUILD create mode 100644 pkg/sentry/devices/miscdev/fuse.go create mode 100644 pkg/sentry/devices/miscdev/miscdev.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/dev.go b/pkg/abi/linux/dev.go index fa3ae5f18..192e2093b 100644 --- a/pkg/abi/linux/dev.go +++ b/pkg/abi/linux/dev.go @@ -46,6 +46,10 @@ const ( // TTYAUX_MAJOR is the major device number for alternate TTY devices. TTYAUX_MAJOR = 5 + // MISC_MAJOR is the major device number for non-serial mice, misc feature + // devices. + MISC_MAJOR = 10 + // UNIX98_PTY_MASTER_MAJOR is the initial major device number for // Unix98 PTY masters. UNIX98_PTY_MASTER_MAJOR = 128 diff --git a/pkg/sentry/devices/miscdev/BUILD b/pkg/sentry/devices/miscdev/BUILD new file mode 100644 index 000000000..aaa76c5d2 --- /dev/null +++ b/pkg/sentry/devices/miscdev/BUILD @@ -0,0 +1,20 @@ +load("//tools:defs.bzl", "go_library") + +licenses(["notice"]) + +go_library( + name = "miscdev", + srcs = [ + "fuse.go", + "miscdev.go", + ], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/abi/linux", + "//pkg/context", + "//pkg/sentry/fsimpl/devtmpfs", + "//pkg/sentry/vfs", + "//pkg/syserror", + "//pkg/usermem", + ], +) diff --git a/pkg/sentry/devices/miscdev/fuse.go b/pkg/sentry/devices/miscdev/fuse.go new file mode 100644 index 000000000..d0a963191 --- /dev/null +++ b/pkg/sentry/devices/miscdev/fuse.go @@ -0,0 +1,78 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package miscdev + +import ( + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +const fuseDevMinor = 229 + +// fuseDevice implements vfs.Device for /dev/fuse. +type fuseDevice struct{} + +// Open implements vfs.Device.Open. +func (fuseDevice) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + var fd FUSEDeviceFile + if err := fd.vfsfd.Init(&fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{ + UseDentryMetadata: true, + }); err != nil { + return nil, err + } + return &fd.vfsfd, nil +} + +// FUSEDeviceFile implements vfs.FileDescriptionImpl for /dev/fuse. +type FUSEDeviceFile struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl + vfs.DentryMetadataFileDescriptionImpl + vfs.NoLockFD + + // TODO(gvisor.dev/issue/2987): Add all the data structures needed to enqueue + // and deque requests, control synchronization and establish communication + // between the FUSE kernel module and the /dev/fuse character device. +} + +// Release implements vfs.FileDescriptionImpl.Release. +func (fd *FUSEDeviceFile) Release() {} + +// PRead implements vfs.FileDescriptionImpl.PRead. +func (fd *FUSEDeviceFile) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + return 0, syserror.ENOSYS +} + +// Read implements vfs.FileDescriptionImpl.Read. +func (fd *FUSEDeviceFile) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + return 0, syserror.ENOSYS +} + +// PWrite implements vfs.FileDescriptionImpl.PWrite. +func (fd *FUSEDeviceFile) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + return 0, syserror.ENOSYS +} + +// Write implements vfs.FileDescriptionImpl.Write. +func (fd *FUSEDeviceFile) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + return 0, syserror.ENOSYS +} + +// Seek implements vfs.FileDescriptionImpl.Seek. +func (fd *FUSEDeviceFile) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + return 0, syserror.ENOSYS +} diff --git a/pkg/sentry/devices/miscdev/miscdev.go b/pkg/sentry/devices/miscdev/miscdev.go new file mode 100644 index 000000000..500d92ed9 --- /dev/null +++ b/pkg/sentry/devices/miscdev/miscdev.go @@ -0,0 +1,54 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package miscdev implements "misc" character devices, as implemented in Linux +// by drivers/char/misc.c and fs/fuse/dev.c. +package miscdev + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/devtmpfs" + "gvisor.dev/gvisor/pkg/sentry/vfs" +) + +// miscDevMajor is the major device number for devices defined in this package. +const miscDevMajor = linux.MISC_MAJOR + +// Register registers all devices implemented by this package in vfsObj. +func Register(vfsObj *vfs.VirtualFilesystem) error { + for minor, dev := range map[uint32]vfs.Device{ + fuseDevMinor: fuseDevice{}, + } { + if err := vfsObj.RegisterDevice(vfs.CharDevice, miscDevMajor, minor, dev, &vfs.RegisterDeviceOptions{ + GroupName: "misc", + }); err != nil { + return err + } + } + return nil +} + +// CreateDevtmpfsFiles creates device special files in dev representing all +// devices implemented by this package. +func CreateDevtmpfsFiles(ctx context.Context, dev *devtmpfs.Accessor) error { + for minor, name := range map[uint32]string{ + fuseDevMinor: "fuse", + } { + if err := dev.CreateDeviceFile(ctx, name, vfs.CharDevice, miscDevMajor, minor, 0666 /* mode */); err != nil { + return err + } + } + return nil +} diff --git a/test/syscalls/linux/dev.cc b/test/syscalls/linux/dev.cc index 4dd302eed..82b115981 100644 --- a/test/syscalls/linux/dev.cc +++ b/test/syscalls/linux/dev.cc @@ -146,7 +146,27 @@ TEST(DevTest, WriteDevFull) { EXPECT_THAT(WriteFd(fd.get(), "a", 1), SyscallFailsWithErrno(ENOSPC)); } +TEST(DevTest, ReadDevFuse) { + SKIP_IF(IsRunningWithVFS1()); + + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open("/dev/fuse", O_RDONLY)); + std::vector buf(1); + EXPECT_THAT(ReadFd(fd.get(), buf.data(), sizeof(buf)), SyscallFailsWithErrno(ENOSYS)); +} + +TEST(DevTest, WriteDevFuse) { + SKIP_IF(IsRunningWithVFS1()); + + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open("/dev/fuse", O_WRONLY)); + const char* testStr = "test"; + EXPECT_THAT(WriteFd(fd.get(), testStr, sizeof(testStr)), SyscallFailsWithErrno(ENOSYS)); +} + TEST(DevTest, TTYExists) { + SKIP_IF(!IsRunningWithVFS1()); + struct stat statbuf = {}; ASSERT_THAT(stat("/dev/tty", &statbuf), SyscallSucceeds()); // Check that it's a character device with rw-rw-rw- permissions. -- cgit v1.2.3 From a63db7d90303280de9431f369e5a9c8db351a9e8 Mon Sep 17 00:00:00 2001 From: Ridwan Sharif Date: Wed, 17 Jun 2020 16:23:27 -0400 Subject: Moved FUSE device under the fuse directory --- pkg/sentry/devices/miscdev/BUILD | 20 ------- pkg/sentry/devices/miscdev/fuse.go | 78 -------------------------- pkg/sentry/devices/miscdev/miscdev.go | 54 ------------------ pkg/sentry/fsimpl/fuse/BUILD | 19 +++++++ pkg/sentry/fsimpl/fuse/dev.go | 100 ++++++++++++++++++++++++++++++++++ runsc/boot/BUILD | 1 + runsc/boot/vfs.go | 7 +++ test/syscalls/linux/dev.cc | 3 +- 8 files changed, 129 insertions(+), 153 deletions(-) delete mode 100644 pkg/sentry/devices/miscdev/BUILD delete mode 100644 pkg/sentry/devices/miscdev/fuse.go delete mode 100644 pkg/sentry/devices/miscdev/miscdev.go create mode 100644 pkg/sentry/fsimpl/fuse/BUILD create mode 100644 pkg/sentry/fsimpl/fuse/dev.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/devices/miscdev/BUILD b/pkg/sentry/devices/miscdev/BUILD deleted file mode 100644 index aaa76c5d2..000000000 --- a/pkg/sentry/devices/miscdev/BUILD +++ /dev/null @@ -1,20 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -licenses(["notice"]) - -go_library( - name = "miscdev", - srcs = [ - "fuse.go", - "miscdev.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/sentry/fsimpl/devtmpfs", - "//pkg/sentry/vfs", - "//pkg/syserror", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/devices/miscdev/fuse.go b/pkg/sentry/devices/miscdev/fuse.go deleted file mode 100644 index d0a963191..000000000 --- a/pkg/sentry/devices/miscdev/fuse.go +++ /dev/null @@ -1,78 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package miscdev - -import ( - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -const fuseDevMinor = 229 - -// fuseDevice implements vfs.Device for /dev/fuse. -type fuseDevice struct{} - -// Open implements vfs.Device.Open. -func (fuseDevice) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - var fd FUSEDeviceFile - if err := fd.vfsfd.Init(&fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{ - UseDentryMetadata: true, - }); err != nil { - return nil, err - } - return &fd.vfsfd, nil -} - -// FUSEDeviceFile implements vfs.FileDescriptionImpl for /dev/fuse. -type FUSEDeviceFile struct { - vfsfd vfs.FileDescription - vfs.FileDescriptionDefaultImpl - vfs.DentryMetadataFileDescriptionImpl - vfs.NoLockFD - - // TODO(gvisor.dev/issue/2987): Add all the data structures needed to enqueue - // and deque requests, control synchronization and establish communication - // between the FUSE kernel module and the /dev/fuse character device. -} - -// Release implements vfs.FileDescriptionImpl.Release. -func (fd *FUSEDeviceFile) Release() {} - -// PRead implements vfs.FileDescriptionImpl.PRead. -func (fd *FUSEDeviceFile) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - return 0, syserror.ENOSYS -} - -// Read implements vfs.FileDescriptionImpl.Read. -func (fd *FUSEDeviceFile) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - return 0, syserror.ENOSYS -} - -// PWrite implements vfs.FileDescriptionImpl.PWrite. -func (fd *FUSEDeviceFile) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - return 0, syserror.ENOSYS -} - -// Write implements vfs.FileDescriptionImpl.Write. -func (fd *FUSEDeviceFile) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { - return 0, syserror.ENOSYS -} - -// Seek implements vfs.FileDescriptionImpl.Seek. -func (fd *FUSEDeviceFile) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - return 0, syserror.ENOSYS -} diff --git a/pkg/sentry/devices/miscdev/miscdev.go b/pkg/sentry/devices/miscdev/miscdev.go deleted file mode 100644 index 500d92ed9..000000000 --- a/pkg/sentry/devices/miscdev/miscdev.go +++ /dev/null @@ -1,54 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package miscdev implements "misc" character devices, as implemented in Linux -// by drivers/char/misc.c and fs/fuse/dev.c. -package miscdev - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/devtmpfs" - "gvisor.dev/gvisor/pkg/sentry/vfs" -) - -// miscDevMajor is the major device number for devices defined in this package. -const miscDevMajor = linux.MISC_MAJOR - -// Register registers all devices implemented by this package in vfsObj. -func Register(vfsObj *vfs.VirtualFilesystem) error { - for minor, dev := range map[uint32]vfs.Device{ - fuseDevMinor: fuseDevice{}, - } { - if err := vfsObj.RegisterDevice(vfs.CharDevice, miscDevMajor, minor, dev, &vfs.RegisterDeviceOptions{ - GroupName: "misc", - }); err != nil { - return err - } - } - return nil -} - -// CreateDevtmpfsFiles creates device special files in dev representing all -// devices implemented by this package. -func CreateDevtmpfsFiles(ctx context.Context, dev *devtmpfs.Accessor) error { - for minor, name := range map[uint32]string{ - fuseDevMinor: "fuse", - } { - if err := dev.CreateDeviceFile(ctx, name, vfs.CharDevice, miscDevMajor, minor, 0666 /* mode */); err != nil { - return err - } - } - return nil -} diff --git a/pkg/sentry/fsimpl/fuse/BUILD b/pkg/sentry/fsimpl/fuse/BUILD new file mode 100644 index 000000000..41567967d --- /dev/null +++ b/pkg/sentry/fsimpl/fuse/BUILD @@ -0,0 +1,19 @@ +load("//tools:defs.bzl", "go_library") + +licenses(["notice"]) + +go_library( + name = "fuse", + srcs = [ + "dev.go", + ], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/abi/linux", + "//pkg/context", + "//pkg/sentry/fsimpl/devtmpfs", + "//pkg/sentry/vfs", + "//pkg/syserror", + "//pkg/usermem", + ], +) diff --git a/pkg/sentry/fsimpl/fuse/dev.go b/pkg/sentry/fsimpl/fuse/dev.go new file mode 100644 index 000000000..f6a67d005 --- /dev/null +++ b/pkg/sentry/fsimpl/fuse/dev.go @@ -0,0 +1,100 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package fuse + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/devtmpfs" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +const fuseDevMinor = 229 + +// fuseDevice implements vfs.Device for /dev/fuse. +type fuseDevice struct{} + +// Open implements vfs.Device.Open. +func (fuseDevice) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + var fd DeviceFD + if err := fd.vfsfd.Init(&fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{ + UseDentryMetadata: true, + }); err != nil { + return nil, err + } + return &fd.vfsfd, nil +} + +// DeviceFD implements vfs.FileDescriptionImpl for /dev/fuse. +type DeviceFD struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl + vfs.DentryMetadataFileDescriptionImpl + vfs.NoLockFD + + // TODO(gvisor.dev/issue/2987): Add all the data structures needed to enqueue + // and deque requests, control synchronization and establish communication + // between the FUSE kernel module and the /dev/fuse character device. +} + +// Release implements vfs.FileDescriptionImpl.Release. +func (fd *DeviceFD) Release() {} + +// PRead implements vfs.FileDescriptionImpl.PRead. +func (fd *DeviceFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + return 0, syserror.ENOSYS +} + +// Read implements vfs.FileDescriptionImpl.Read. +func (fd *DeviceFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + return 0, syserror.ENOSYS +} + +// PWrite implements vfs.FileDescriptionImpl.PWrite. +func (fd *DeviceFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + return 0, syserror.ENOSYS +} + +// Write implements vfs.FileDescriptionImpl.Write. +func (fd *DeviceFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + return 0, syserror.ENOSYS +} + +// Seek implements vfs.FileDescriptionImpl.Seek. +func (fd *DeviceFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + return 0, syserror.ENOSYS +} + +// Register registers the FUSE device with vfsObj. +func Register(vfsObj *vfs.VirtualFilesystem) error { + if err := vfsObj.RegisterDevice(vfs.CharDevice, linux.MISC_MAJOR, fuseDevMinor, fuseDevice{}, &vfs.RegisterDeviceOptions{ + GroupName: "misc", + }); err != nil { + return err + } + + return nil +} + +// CreateDevtmpfsFile creates a device special file in devtmpfs. +func CreateDevtmpfsFile(ctx context.Context, dev *devtmpfs.Accessor) error { + if err := dev.CreateDeviceFile(ctx, "fuse", vfs.CharDevice, linux.MISC_MAJOR, fuseDevMinor, 0666 /* mode */); err != nil { + return err + } + + return nil +} diff --git a/runsc/boot/BUILD b/runsc/boot/BUILD index 45d7f7d09..aad2a41de 100644 --- a/runsc/boot/BUILD +++ b/runsc/boot/BUILD @@ -55,6 +55,7 @@ go_library( "//pkg/sentry/fs/user", "//pkg/sentry/fsimpl/devpts", "//pkg/sentry/fsimpl/devtmpfs", + "//pkg/sentry/fsimpl/fuse", "//pkg/sentry/fsimpl/gofer", "//pkg/sentry/fsimpl/host", "//pkg/sentry/fsimpl/overlay", diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go index 9ba5e9cd9..0f5ad1e05 100644 --- a/runsc/boot/vfs.go +++ b/runsc/boot/vfs.go @@ -31,6 +31,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs/user" "gvisor.dev/gvisor/pkg/sentry/fsimpl/devpts" "gvisor.dev/gvisor/pkg/sentry/fsimpl/devtmpfs" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/fuse" "gvisor.dev/gvisor/pkg/sentry/fsimpl/gofer" "gvisor.dev/gvisor/pkg/sentry/fsimpl/overlay" "gvisor.dev/gvisor/pkg/sentry/fsimpl/proc" @@ -79,6 +80,9 @@ func registerFilesystems(ctx context.Context, vfsObj *vfs.VirtualFilesystem, cre } if err := ttydev.Register(vfsObj); err != nil { return fmt.Errorf("registering ttydev: %w", err) + + if err := fuse.Register(vfsObj); err != nil { + return fmt.Errorf("registering /dev/fuse: %w", err) } if err := tundev.Register(vfsObj); err != nil { return fmt.Errorf("registering tundev: %v", err) @@ -101,6 +105,9 @@ func registerFilesystems(ctx context.Context, vfsObj *vfs.VirtualFilesystem, cre if err := tundev.CreateDevtmpfsFiles(ctx, a); err != nil { return fmt.Errorf("creating tundev devtmpfs files: %v", err) } + if err := fuse.CreateDevtmpfsFile(ctx, a); err != nil { + return fmt.Errorf("creating devtmpfs fuse device file: %w", err) + } return nil } diff --git a/test/syscalls/linux/dev.cc b/test/syscalls/linux/dev.cc index 82b115981..6be173c14 100644 --- a/test/syscalls/linux/dev.cc +++ b/test/syscalls/linux/dev.cc @@ -165,7 +165,8 @@ TEST(DevTest, WriteDevFuse) { } TEST(DevTest, TTYExists) { - SKIP_IF(!IsRunningWithVFS1()); + // Run test if running on VFS1 or on Linux. + SKIP_IF(!IsRunningWithVFS1() && IsRunningOnGvisor()); struct stat statbuf = {}; ASSERT_THAT(stat("/dev/tty", &statbuf), SyscallSucceeds()); -- cgit v1.2.3 From 4069461877d843654d18db74a5962b332f1226aa Mon Sep 17 00:00:00 2001 From: Tamir Duberstein Date: Thu, 25 Jun 2020 14:17:00 -0700 Subject: Avoid an allocation in epoll PiperOrigin-RevId: 318346153 --- pkg/sentry/kernel/epoll/epoll.go | 32 ++++++++++++++------------------ pkg/sentry/kernel/epoll/epoll_state.go | 2 +- 2 files changed, 15 insertions(+), 19 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/epoll/epoll.go b/pkg/sentry/kernel/epoll/epoll.go index 679ab495d..4c0f1e41f 100644 --- a/pkg/sentry/kernel/epoll/epoll.go +++ b/pkg/sentry/kernel/epoll/epoll.go @@ -107,7 +107,7 @@ type EventPoll struct { // different lock to avoid circular lock acquisition order involving // the wait queue mutexes and mu. The full order is mu, observed file // wait queue mutex, then listsMu; this allows listsMu to be acquired - // when readyCallback is called. + // when (*pollEntry).Callback is called. // // An entry is always in one of the following lists: // readyList -- when there's a chance that it's ready to have @@ -116,7 +116,7 @@ type EventPoll struct { // readEvents() functions always call the entry's file // Readiness() function to confirm it's ready. // waitingList -- when there's no chance that the entry is ready, - // so it's waiting for the readyCallback to be called + // so it's waiting for the (*pollEntry).Callback to be called // on it before it gets moved to the readyList. // disabledList -- when the entry is disabled. This happens when // a one-shot entry gets delivered via readEvents(). @@ -269,23 +269,19 @@ func (e *EventPoll) ReadEvents(max int) []linux.EpollEvent { return ret } -// readyCallback is called when one of the files we're polling becomes ready. It -// moves said file to the readyList if it's currently in the waiting list. -type readyCallback struct { - context *pollEntry -} - // Callback implements waiter.EntryCallback.Callback. -func (r *readyCallback) Callback(*waiter.Entry) { - entry := r.context - e := entry.epoll +// +// Callback is called when one of the files we're polling becomes ready. It +// moves said file to the readyList if it's currently in the waiting list. +func (p *pollEntry) Callback(*waiter.Entry) { + e := p.epoll e.listsMu.Lock() - if entry.curList == &e.waitingList { - e.waitingList.Remove(entry) - e.readyList.PushBack(entry) - entry.curList = &e.readyList + if p.curList == &e.waitingList { + e.waitingList.Remove(p) + e.readyList.PushBack(p) + p.curList = &e.readyList e.listsMu.Unlock() e.Notify(waiter.EventIn) @@ -312,7 +308,7 @@ func (e *EventPoll) initEntryReadiness(entry *pollEntry) { // Check if the file happens to already be in a ready state. ready := f.Readiness(entry.mask) & entry.mask if ready != 0 { - (&readyCallback{context: entry}).Callback(&entry.waiter) + entry.Callback(&entry.waiter) } } @@ -384,7 +380,7 @@ func (e *EventPoll) AddEntry(id FileIdentifier, flags EntryFlags, mask waiter.Ev flags: flags, mask: mask, } - entry.waiter.Callback = &readyCallback{context: entry} + entry.waiter.Callback = entry e.files[id] = entry entry.file = refs.NewWeakRef(id.File, entry) @@ -407,7 +403,7 @@ func (e *EventPoll) UpdateEntry(id FileIdentifier, flags EntryFlags, mask waiter } // Unregister the old mask and remove entry from the list it's in, so - // readyCallback is guaranteed to not be called on this entry anymore. + // (*pollEntry).Callback is guaranteed to not be called on this entry anymore. entry.id.File.EventUnregister(&entry.waiter) // Remove entry from whatever list it's in. This ensure that no other diff --git a/pkg/sentry/kernel/epoll/epoll_state.go b/pkg/sentry/kernel/epoll/epoll_state.go index 02f9aabfa..7c61e0258 100644 --- a/pkg/sentry/kernel/epoll/epoll_state.go +++ b/pkg/sentry/kernel/epoll/epoll_state.go @@ -21,7 +21,7 @@ import ( // afterLoad is invoked by stateify. func (p *pollEntry) afterLoad() { - p.waiter.Callback = &readyCallback{context: p} + p.waiter.Callback = p p.file = refs.NewWeakRef(p.id.File, p) p.id.File.EventRegister(&p.waiter, p.mask) } -- cgit v1.2.3 From 7fb6cc286fffab2f408a5dbc228e0db706104682 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Thu, 25 Jun 2020 21:20:29 -0700 Subject: conntrack refactor, no behavior changes - Split connTrackForPacket into 2 functions instead of switching on flag - Replace hash with struct keys. - Remove prefixes where possible - Remove unused connStatus, timeout - Flatten ConnTrack struct a bit - some intermediate structs had no meaning outside of the context of their parent. - Protect conn.tcb with a mutex - Remove redundant error checking (e.g. when is pkt.NetworkHeader valid) - Clarify that HandlePacket and CreateConnFor are the expected entrypoints for ConnTrack PiperOrigin-RevId: 318407168 --- pkg/sentry/socket/netfilter/targets.go | 2 +- pkg/tcpip/stack/conntrack.go | 405 ++++++++++++--------------------- pkg/tcpip/stack/iptables.go | 7 +- pkg/tcpip/stack/iptables_targets.go | 23 +- pkg/tcpip/stack/iptables_types.go | 4 +- 5 files changed, 168 insertions(+), 273 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/targets.go b/pkg/sentry/socket/netfilter/targets.go index 84abe8d29..b91ba3ab3 100644 --- a/pkg/sentry/socket/netfilter/targets.go +++ b/pkg/sentry/socket/netfilter/targets.go @@ -30,6 +30,6 @@ type JumpTarget struct { } // Action implements stack.Target.Action. -func (jt JumpTarget) Action(*stack.PacketBuffer, *stack.ConnTrackTable, stack.Hook, *stack.GSO, *stack.Route, tcpip.Address) (stack.RuleVerdict, int) { +func (jt JumpTarget) Action(*stack.PacketBuffer, *stack.ConnTrack, stack.Hook, *stack.GSO, *stack.Route, tcpip.Address) (stack.RuleVerdict, int) { return stack.RuleJump, jt.RuleNum } diff --git a/pkg/tcpip/stack/conntrack.go b/pkg/tcpip/stack/conntrack.go index 05bf62788..af9c325ca 100644 --- a/pkg/tcpip/stack/conntrack.go +++ b/pkg/tcpip/stack/conntrack.go @@ -15,39 +15,29 @@ package stack import ( - "encoding/binary" "sync" - "time" "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/hash/jenkins" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/transport/tcpconntrack" ) // Connection tracking is used to track and manipulate packets for NAT rules. -// The connection is created for a packet if it does not exist. Every connection -// contains two tuples (original and reply). The tuples are manipulated if there -// is a matching NAT rule. The packet is modified by looking at the tuples in the -// Prerouting and Output hooks. +// The connection is created for a packet if it does not exist. Every +// connection contains two tuples (original and reply). The tuples are +// manipulated if there is a matching NAT rule. The packet is modified by +// looking at the tuples in the Prerouting and Output hooks. +// +// Currently, only TCP tracking is supported. // Direction of the tuple. -type ctDirection int +type direction int const ( - dirOriginal ctDirection = iota + dirOriginal direction = iota dirReply ) -// Status of connection. -// TODO(gvisor.dev/issue/170): Add other states of connection. -type connStatus int - -const ( - connNew connStatus = iota - connEstablished -) - // Manipulation type for the connection. type manipType int @@ -56,250 +46,171 @@ const ( manipDstOutput ) -// connTrackMutable is the manipulatable part of the tuple. -type connTrackMutable struct { - // addr is source address of the tuple. - addr tcpip.Address +// tuple holds a connection's identifying and manipulating data in one +// direction. It is immutable. +type tuple struct { + tupleID - // port is source port of the tuple. - port uint16 + // conn is the connection tracking entry this tuple belongs to. + conn *conn - // protocol is network layer protocol. - protocol tcpip.NetworkProtocolNumber + // direction is the direction of the tuple. + direction direction } -// connTrackImmutable is the non-manipulatable part of the tuple. -type connTrackImmutable struct { - // addr is destination address of the tuple. - addr tcpip.Address - - // direction is direction (original or reply) of the tuple. - direction ctDirection - - // port is destination port of the tuple. - port uint16 - - // protocol is transport layer protocol. - protocol tcpip.TransportProtocolNumber +// tupleID uniquely identifies a connection in one direction. It currently +// contains enough information to distinguish between any TCP or UDP +// connection, and will need to be extended to support other protocols. +type tupleID struct { + srcAddr tcpip.Address + srcPort uint16 + dstAddr tcpip.Address + dstPort uint16 + transProto tcpip.TransportProtocolNumber + netProto tcpip.NetworkProtocolNumber } -// connTrackTuple represents the tuple which is created from the -// packet. -type connTrackTuple struct { - // dst is non-manipulatable part of the tuple. - dst connTrackImmutable - - // src is manipulatable part of the tuple. - src connTrackMutable +// reply creates the reply tupleID. +func (ti tupleID) reply() tupleID { + return tupleID{ + srcAddr: ti.dstAddr, + srcPort: ti.dstPort, + dstAddr: ti.srcAddr, + dstPort: ti.srcPort, + transProto: ti.transProto, + netProto: ti.netProto, + } } -// connTrackTupleHolder is the container of tuple and connection. -type ConnTrackTupleHolder struct { - // conn is pointer to the connection tracking entry. - conn *connTrack - - // tuple is original or reply tuple. - tuple connTrackTuple -} +// conn is a tracked connection. +type conn struct { + // original is the tuple in original direction. It is immutable. + original tuple -// connTrack is the connection. -type connTrack struct { - // originalTupleHolder contains tuple in original direction. - originalTupleHolder ConnTrackTupleHolder + // reply is the tuple in reply direction. It is immutable. + reply tuple - // replyTupleHolder contains tuple in reply direction. - replyTupleHolder ConnTrackTupleHolder + // manip indicates if the packet should be manipulated. It is immutable. + manip manipType - // status indicates connection is new or established. - status connStatus + // tcbHook indicates if the packet is inbound or outbound to + // update the state of tcb. It is immutable. + tcbHook Hook - // timeout indicates the time connection should be active. - timeout time.Duration - - // manip indicates if the packet should be manipulated. - manip manipType + // mu protects tcb. + mu sync.Mutex // tcb is TCB control block. It is used to keep track of states - // of tcp connection. + // of tcp connection and is protected by mu. tcb tcpconntrack.TCB - - // tcbHook indicates if the packet is inbound or outbound to - // update the state of tcb. - tcbHook Hook } -// ConnTrackTable contains a map of all existing connections created for -// NAT rules. -type ConnTrackTable struct { - // connMu protects connTrackTable. - connMu sync.RWMutex - - // connTrackTable maintains a map of tuples needed for connection tracking - // for iptables NAT rules. The key for the map is an integer calculated - // using seed, source address, destination address, source port and - // destination port. - CtMap map[uint32]ConnTrackTupleHolder - - // seed is a one-time random value initialized at stack startup - // and is used in calculation of hash key for connection tracking - // table. - Seed uint32 -} +// ConnTrack tracks all connections created for NAT rules. Most users are +// expected to only call handlePacket and createConnFor. +type ConnTrack struct { + // mu protects conns. + mu sync.RWMutex -// packetToTuple converts packet to a tuple in original direction. -func packetToTuple(pkt *PacketBuffer, hook Hook) (connTrackTuple, *tcpip.Error) { - var tuple connTrackTuple + // conns maintains a map of tuples needed for connection tracking for + // iptables NAT rules. It is protected by mu. + conns map[tupleID]tuple +} - netHeader := header.IPv4(pkt.NetworkHeader) +// packetToTupleID converts packet to a tuple ID. It fails when pkt lacks a valid +// TCP header. +func packetToTupleID(pkt *PacketBuffer) (tupleID, *tcpip.Error) { // TODO(gvisor.dev/issue/170): Need to support for other // protocols as well. + netHeader := header.IPv4(pkt.NetworkHeader) if netHeader == nil || netHeader.TransportProtocol() != header.TCPProtocolNumber { - return tuple, tcpip.ErrUnknownProtocol + return tupleID{}, tcpip.ErrUnknownProtocol } tcpHeader := header.TCP(pkt.TransportHeader) if tcpHeader == nil { - return tuple, tcpip.ErrUnknownProtocol + return tupleID{}, tcpip.ErrUnknownProtocol } - tuple.src.addr = netHeader.SourceAddress() - tuple.src.port = tcpHeader.SourcePort() - tuple.src.protocol = header.IPv4ProtocolNumber - - tuple.dst.addr = netHeader.DestinationAddress() - tuple.dst.port = tcpHeader.DestinationPort() - tuple.dst.protocol = netHeader.TransportProtocol() - - return tuple, nil -} - -// getReplyTuple creates reply tuple for the given tuple. -func getReplyTuple(tuple connTrackTuple) connTrackTuple { - var replyTuple connTrackTuple - replyTuple.src.addr = tuple.dst.addr - replyTuple.src.port = tuple.dst.port - replyTuple.src.protocol = tuple.src.protocol - replyTuple.dst.addr = tuple.src.addr - replyTuple.dst.port = tuple.src.port - replyTuple.dst.protocol = tuple.dst.protocol - replyTuple.dst.direction = dirReply - - return replyTuple + return tupleID{ + srcAddr: netHeader.SourceAddress(), + srcPort: tcpHeader.SourcePort(), + dstAddr: netHeader.DestinationAddress(), + dstPort: tcpHeader.DestinationPort(), + transProto: netHeader.TransportProtocol(), + netProto: header.IPv4ProtocolNumber, + }, nil } -// makeNewConn creates new connection. -func makeNewConn(tuple, replyTuple connTrackTuple) connTrack { - var conn connTrack - conn.status = connNew - conn.originalTupleHolder.tuple = tuple - conn.originalTupleHolder.conn = &conn - conn.replyTupleHolder.tuple = replyTuple - conn.replyTupleHolder.conn = &conn - - return conn -} - -// getTupleHash returns hash of the tuple. The fields used for -// generating hash are seed (generated once for stack), source address, -// destination address, source port and destination ports. -func (ct *ConnTrackTable) getTupleHash(tuple connTrackTuple) uint32 { - h := jenkins.Sum32(ct.Seed) - h.Write([]byte(tuple.src.addr)) - h.Write([]byte(tuple.dst.addr)) - portBuf := make([]byte, 2) - binary.LittleEndian.PutUint16(portBuf, tuple.src.port) - h.Write([]byte(portBuf)) - binary.LittleEndian.PutUint16(portBuf, tuple.dst.port) - h.Write([]byte(portBuf)) - - return h.Sum32() +// newConn creates new connection. +func newConn(orig, reply tupleID, manip manipType, hook Hook) *conn { + conn := conn{ + manip: manip, + tcbHook: hook, + } + conn.original = tuple{conn: &conn, tupleID: orig} + conn.reply = tuple{conn: &conn, tupleID: reply, direction: dirReply} + return &conn } -// connTrackForPacket returns connTrack for packet. -// TODO(gvisor.dev/issue/170): Only TCP packets are supported. Need to support other -// transport protocols. -func (ct *ConnTrackTable) connTrackForPacket(pkt *PacketBuffer, hook Hook, createConn bool) (*connTrack, ctDirection) { - var dir ctDirection - tuple, err := packetToTuple(pkt, hook) +// connFor gets the conn for pkt if it exists, or returns nil +// if it does not. It returns an error when pkt does not contain a valid TCP +// header. +// TODO(gvisor.dev/issue/170): Only TCP packets are supported. Need to support +// other transport protocols. +func (ct *ConnTrack) connFor(pkt *PacketBuffer) (*conn, direction) { + tid, err := packetToTupleID(pkt) if err != nil { - return nil, dir + return nil, dirOriginal } - ct.connMu.Lock() - defer ct.connMu.Unlock() - - connTrackTable := ct.CtMap - hash := ct.getTupleHash(tuple) - - var conn *connTrack - switch createConn { - case true: - // If connection does not exist for the hash, create a new - // connection. - replyTuple := getReplyTuple(tuple) - replyHash := ct.getTupleHash(replyTuple) - newConn := makeNewConn(tuple, replyTuple) - conn = &newConn - - // Add tupleHolders to the map. - // TODO(gvisor.dev/issue/170): Need to support collisions using linked list. - ct.CtMap[hash] = conn.originalTupleHolder - ct.CtMap[replyHash] = conn.replyTupleHolder - default: - tupleHolder, ok := connTrackTable[hash] - if !ok { - return nil, dir - } - - // If this is the reply of new connection, set the connection - // status as ESTABLISHED. - conn = tupleHolder.conn - if conn.status == connNew && tupleHolder.tuple.dst.direction == dirReply { - conn.status = connEstablished - } - if tupleHolder.conn == nil { - panic("tupleHolder has null connection tracking entry") - } + ct.mu.Lock() + defer ct.mu.Unlock() - dir = tupleHolder.tuple.dst.direction + tuple, ok := ct.conns[tid] + if !ok { + return nil, dirOriginal } - return conn, dir + return tuple.conn, tuple.direction } -// SetNatInfo will manipulate the tuples according to iptables NAT rules. -func (ct *ConnTrackTable) SetNatInfo(pkt *PacketBuffer, rt RedirectTarget, hook Hook) { - // Get the connection. Connection is always created before this - // function is called. - conn, _ := ct.connTrackForPacket(pkt, hook, false) - if conn == nil { - panic("connection should be created to manipulate tuples.") +// createConnFor creates a new conn for pkt. +func (ct *ConnTrack) createConnFor(pkt *PacketBuffer, hook Hook, rt RedirectTarget) *conn { + tid, err := packetToTupleID(pkt) + if err != nil { + return nil + } + if hook != Prerouting && hook != Output { + return nil } - replyTuple := conn.replyTupleHolder.tuple - replyHash := ct.getTupleHash(replyTuple) - - // TODO(gvisor.dev/issue/170): Support only redirect of ports. Need to - // support changing of address for Prerouting. - // Change the port as per the iptables rule. This tuple will be used - // to manipulate the packet in HandlePacket. - conn.replyTupleHolder.tuple.src.addr = rt.MinIP - conn.replyTupleHolder.tuple.src.port = rt.MinPort - newHash := ct.getTupleHash(conn.replyTupleHolder.tuple) + // Create a new connection and change the port as per the iptables + // rule. This tuple will be used to manipulate the packet in + // handlePacket. + replyTID := tid.reply() + replyTID.srcAddr = rt.MinIP + replyTID.srcPort = rt.MinPort + var manip manipType + switch hook { + case Prerouting: + manip = manipDstPrerouting + case Output: + manip = manipDstOutput + } + conn := newConn(tid, replyTID, manip, hook) // Add the changed tuple to the map. - ct.connMu.Lock() - defer ct.connMu.Unlock() - ct.CtMap[newHash] = conn.replyTupleHolder - if hook == Output { - conn.replyTupleHolder.conn.manip = manipDstOutput - } + // TODO(gvisor.dev/issue/170): Need to support collisions using linked + // list. + ct.mu.Lock() + defer ct.mu.Unlock() + ct.conns[tid] = conn.original + ct.conns[replyTID] = conn.reply - // Delete the old tuple. - delete(ct.CtMap, replyHash) + return conn } // handlePacketPrerouting manipulates ports for packets in Prerouting hook. -// TODO(gvisor.dev/issue/170): Change address for Prerouting hook.. -func handlePacketPrerouting(pkt *PacketBuffer, conn *connTrack, dir ctDirection) { +// TODO(gvisor.dev/issue/170): Change address for Prerouting hook. +func handlePacketPrerouting(pkt *PacketBuffer, conn *conn, dir direction) { netHeader := header.IPv4(pkt.NetworkHeader) tcpHeader := header.TCP(pkt.TransportHeader) @@ -308,13 +219,13 @@ func handlePacketPrerouting(pkt *PacketBuffer, conn *connTrack, dir ctDirection) // modified. switch dir { case dirOriginal: - port := conn.replyTupleHolder.tuple.src.port + port := conn.reply.srcPort tcpHeader.SetDestinationPort(port) - netHeader.SetDestinationAddress(conn.replyTupleHolder.tuple.src.addr) + netHeader.SetDestinationAddress(conn.reply.srcAddr) case dirReply: - port := conn.originalTupleHolder.tuple.dst.port + port := conn.original.dstPort tcpHeader.SetSourcePort(port) - netHeader.SetSourceAddress(conn.originalTupleHolder.tuple.dst.addr) + netHeader.SetSourceAddress(conn.original.dstAddr) } netHeader.SetChecksum(0) @@ -322,7 +233,7 @@ func handlePacketPrerouting(pkt *PacketBuffer, conn *connTrack, dir ctDirection) } // handlePacketOutput manipulates ports for packets in Output hook. -func handlePacketOutput(pkt *PacketBuffer, conn *connTrack, gso *GSO, r *Route, dir ctDirection) { +func handlePacketOutput(pkt *PacketBuffer, conn *conn, gso *GSO, r *Route, dir direction) { netHeader := header.IPv4(pkt.NetworkHeader) tcpHeader := header.TCP(pkt.TransportHeader) @@ -331,13 +242,13 @@ func handlePacketOutput(pkt *PacketBuffer, conn *connTrack, gso *GSO, r *Route, // modified. For prerouting redirection, we only reach this point // when replying, so packet sources are modified. if conn.manip == manipDstOutput && dir == dirOriginal { - port := conn.replyTupleHolder.tuple.src.port + port := conn.reply.srcPort tcpHeader.SetDestinationPort(port) - netHeader.SetDestinationAddress(conn.replyTupleHolder.tuple.src.addr) + netHeader.SetDestinationAddress(conn.reply.srcAddr) } else { - port := conn.originalTupleHolder.tuple.dst.port + port := conn.original.dstPort tcpHeader.SetSourcePort(port) - netHeader.SetSourceAddress(conn.originalTupleHolder.tuple.dst.addr) + netHeader.SetSourceAddress(conn.original.dstAddr) } // Calculate the TCP checksum and set it. @@ -356,9 +267,9 @@ func handlePacketOutput(pkt *PacketBuffer, conn *connTrack, gso *GSO, r *Route, netHeader.SetChecksum(^netHeader.CalculateChecksum()) } -// HandlePacket will manipulate the port and address of the packet if the +// handlePacket will manipulate the port and address of the packet if the // connection exists. -func (ct *ConnTrackTable) HandlePacket(pkt *PacketBuffer, hook Hook, gso *GSO, r *Route) { +func (ct *ConnTrack) handlePacket(pkt *PacketBuffer, hook Hook, gso *GSO, r *Route) { if pkt.NatDone { return } @@ -367,21 +278,9 @@ func (ct *ConnTrackTable) HandlePacket(pkt *PacketBuffer, hook Hook, gso *GSO, r return } - conn, dir := ct.connTrackForPacket(pkt, hook, false) - // Connection or Rule not found for the packet. + conn, dir := ct.connFor(pkt) if conn == nil { - return - } - - netHeader := header.IPv4(pkt.NetworkHeader) - // TODO(gvisor.dev/issue/170): Need to support for other transport - // protocols as well. - if netHeader == nil || netHeader.TransportProtocol() != header.TCPProtocolNumber { - return - } - - tcpHeader := header.TCP(pkt.TransportHeader) - if tcpHeader == nil { + // Connection not found for the packet or the packet is invalid. return } @@ -396,7 +295,10 @@ func (ct *ConnTrackTable) HandlePacket(pkt *PacketBuffer, hook Hook, gso *GSO, r // Update the state of tcb. // TODO(gvisor.dev/issue/170): Add support in tcpcontrack to handle // other tcp states. + conn.mu.Lock() + defer conn.mu.Unlock() var st tcpconntrack.Result + tcpHeader := header.TCP(pkt.TransportHeader) if conn.tcb.IsEmpty() { conn.tcb.Init(tcpHeader) conn.tcbHook = hook @@ -409,26 +311,21 @@ func (ct *ConnTrackTable) HandlePacket(pkt *PacketBuffer, hook Hook, gso *GSO, r } } - // Delete conntrack if tcp connection is closed. + // Delete conn if tcp connection is closed. if st == tcpconntrack.ResultClosedByPeer || st == tcpconntrack.ResultClosedBySelf || st == tcpconntrack.ResultReset { - ct.deleteConnTrack(conn) + ct.deleteConn(conn) } } -// deleteConnTrack deletes the connection. -func (ct *ConnTrackTable) deleteConnTrack(conn *connTrack) { +// deleteConn deletes the connection. +func (ct *ConnTrack) deleteConn(conn *conn) { if conn == nil { return } - tuple := conn.originalTupleHolder.tuple - hash := ct.getTupleHash(tuple) - replyTuple := conn.replyTupleHolder.tuple - replyHash := ct.getTupleHash(replyTuple) - - ct.connMu.Lock() - defer ct.connMu.Unlock() + ct.mu.Lock() + defer ct.mu.Unlock() - delete(ct.CtMap, hash) - delete(ct.CtMap, replyHash) + delete(ct.conns, conn.original.tupleID) + delete(ct.conns, conn.reply.tupleID) } diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go index 62d4eb1b6..974d77c36 100644 --- a/pkg/tcpip/stack/iptables.go +++ b/pkg/tcpip/stack/iptables.go @@ -111,9 +111,8 @@ func DefaultTables() *IPTables { Prerouting: []string{TablenameMangle, TablenameNat}, Output: []string{TablenameMangle, TablenameNat, TablenameFilter}, }, - connections: ConnTrackTable{ - CtMap: make(map[uint32]ConnTrackTupleHolder), - Seed: generateRandUint32(), + connections: ConnTrack{ + conns: make(map[tupleID]tuple), }, } } @@ -213,7 +212,7 @@ func (it *IPTables) Check(hook Hook, pkt *PacketBuffer, gso *GSO, r *Route, addr // Packets are manipulated only if connection and matching // NAT rule exists. - it.connections.HandlePacket(pkt, hook, gso, r) + it.connections.handlePacket(pkt, hook, gso, r) // Go through each table containing the hook. for _, tablename := range it.GetPriorities(hook) { diff --git a/pkg/tcpip/stack/iptables_targets.go b/pkg/tcpip/stack/iptables_targets.go index 92e31643e..d43f60c67 100644 --- a/pkg/tcpip/stack/iptables_targets.go +++ b/pkg/tcpip/stack/iptables_targets.go @@ -24,7 +24,7 @@ import ( type AcceptTarget struct{} // Action implements Target.Action. -func (AcceptTarget) Action(*PacketBuffer, *ConnTrackTable, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { +func (AcceptTarget) Action(*PacketBuffer, *ConnTrack, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { return RuleAccept, 0 } @@ -32,7 +32,7 @@ func (AcceptTarget) Action(*PacketBuffer, *ConnTrackTable, Hook, *GSO, *Route, t type DropTarget struct{} // Action implements Target.Action. -func (DropTarget) Action(*PacketBuffer, *ConnTrackTable, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { +func (DropTarget) Action(*PacketBuffer, *ConnTrack, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { return RuleDrop, 0 } @@ -41,7 +41,7 @@ func (DropTarget) Action(*PacketBuffer, *ConnTrackTable, Hook, *GSO, *Route, tcp type ErrorTarget struct{} // Action implements Target.Action. -func (ErrorTarget) Action(*PacketBuffer, *ConnTrackTable, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { +func (ErrorTarget) Action(*PacketBuffer, *ConnTrack, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { log.Debugf("ErrorTarget triggered.") return RuleDrop, 0 } @@ -52,7 +52,7 @@ type UserChainTarget struct { } // Action implements Target.Action. -func (UserChainTarget) Action(*PacketBuffer, *ConnTrackTable, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { +func (UserChainTarget) Action(*PacketBuffer, *ConnTrack, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { panic("UserChainTarget should never be called.") } @@ -61,7 +61,7 @@ func (UserChainTarget) Action(*PacketBuffer, *ConnTrackTable, Hook, *GSO, *Route type ReturnTarget struct{} // Action implements Target.Action. -func (ReturnTarget) Action(*PacketBuffer, *ConnTrackTable, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { +func (ReturnTarget) Action(*PacketBuffer, *ConnTrack, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { return RuleReturn, 0 } @@ -92,7 +92,7 @@ type RedirectTarget struct { // TODO(gvisor.dev/issue/170): Parse headers without copying. The current // implementation only works for PREROUTING and calls pkt.Clone(), neither // of which should be the case. -func (rt RedirectTarget) Action(pkt *PacketBuffer, ct *ConnTrackTable, hook Hook, gso *GSO, r *Route, address tcpip.Address) (RuleVerdict, int) { +func (rt RedirectTarget) Action(pkt *PacketBuffer, ct *ConnTrack, hook Hook, gso *GSO, r *Route, address tcpip.Address) (RuleVerdict, int) { // Packet is already manipulated. if pkt.NatDone { return RuleAccept, 0 @@ -150,12 +150,11 @@ func (rt RedirectTarget) Action(pkt *PacketBuffer, ct *ConnTrackTable, hook Hook return RuleAccept, 0 } - // Set up conection for matching NAT rule. - // Only the first packet of the connection comes here. - // Other packets will be manipulated in connection tracking. - if conn, _ := ct.connTrackForPacket(pkt, hook, true); conn != nil { - ct.SetNatInfo(pkt, rt, hook) - ct.HandlePacket(pkt, hook, gso, r) + // Set up conection for matching NAT rule. Only the first + // packet of the connection comes here. Other packets will be + // manipulated in connection tracking. + if conn := ct.createConnFor(pkt, hook, rt); conn != nil { + ct.handlePacket(pkt, hook, gso, r) } default: return RuleDrop, 0 diff --git a/pkg/tcpip/stack/iptables_types.go b/pkg/tcpip/stack/iptables_types.go index 7026990c4..c528ec381 100644 --- a/pkg/tcpip/stack/iptables_types.go +++ b/pkg/tcpip/stack/iptables_types.go @@ -96,7 +96,7 @@ type IPTables struct { // don't utilize iptables. modified bool - connections ConnTrackTable + connections ConnTrack } // A Table defines a set of chains and hooks into the network stack. It is @@ -249,5 +249,5 @@ type Target interface { // Action takes an action on the packet and returns a verdict on how // traversal should (or should not) continue. If the return value is // Jump, it also returns the index of the rule to jump to. - Action(packet *PacketBuffer, connections *ConnTrackTable, hook Hook, gso *GSO, r *Route, address tcpip.Address) (RuleVerdict, int) + Action(packet *PacketBuffer, connections *ConnTrack, hook Hook, gso *GSO, r *Route, address tcpip.Address) (RuleVerdict, int) } -- cgit v1.2.3 From 54a31e219ca9d6086a367213a92d2a72ce3af07b Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 26 Jun 2020 13:46:01 -0700 Subject: Support inotify IN_ONESHOT. Also, while we're here, make sure that gofer inotify events are generated when files are created in remote revalidating mode. Updates #1479. PiperOrigin-RevId: 318536354 --- pkg/sentry/fsimpl/gofer/filesystem.go | 10 +++- pkg/sentry/syscalls/linux/vfs2/inotify.go | 2 +- pkg/sentry/vfs/inotify.go | 84 +++++++++++++++++++++++-------- test/syscalls/linux/inotify.cc | 48 ++++++++++++++++-- 4 files changed, 115 insertions(+), 29 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index d253c996c..73bac738d 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -389,7 +389,15 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir // RPC will fail with EEXIST like we would have. If the RPC succeeds, and a // stale dentry exists, the dentry will fail revalidation next time it's // used. - return createInRemoteDir(parent, name) + if err := createInRemoteDir(parent, name); err != nil { + return err + } + ev := linux.IN_CREATE + if dir { + ev |= linux.IN_ISDIR + } + parent.watches.Notify(name, uint32(ev), 0, vfs.InodeEvent, false /* unlinked */) + return nil } if child := parent.children[name]; child != nil { return syserror.EEXIST diff --git a/pkg/sentry/syscalls/linux/vfs2/inotify.go b/pkg/sentry/syscalls/linux/vfs2/inotify.go index 74e5287b7..5d98134a5 100644 --- a/pkg/sentry/syscalls/linux/vfs2/inotify.go +++ b/pkg/sentry/syscalls/linux/vfs2/inotify.go @@ -81,7 +81,7 @@ func InotifyAddWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kern // "EINVAL: The given event mask contains no valid events." // -- inotify_add_watch(2) - if validBits := mask & linux.ALL_INOTIFY_BITS; validBits == 0 { + if mask&linux.ALL_INOTIFY_BITS == 0 { return 0, nil, syserror.EINVAL } diff --git a/pkg/sentry/vfs/inotify.go b/pkg/sentry/vfs/inotify.go index 35960394c..509034531 100644 --- a/pkg/sentry/vfs/inotify.go +++ b/pkg/sentry/vfs/inotify.go @@ -447,29 +447,51 @@ func (w *Watches) Remove(id uint64) { return } - if _, ok := w.ws[id]; !ok { - // While there's technically no problem with silently ignoring a missing - // watch, this is almost certainly a bug. - panic(fmt.Sprintf("Attempt to remove a watch, but no watch found with provided id %+v.", id)) + // It is possible for w.Remove() to be called for the same watch multiple + // times. See the treatment of one-shot watches in Watches.Notify(). + if _, ok := w.ws[id]; ok { + delete(w.ws, id) } - delete(w.ws, id) } // Notify queues a new event with watches in this set. Watches with // IN_EXCL_UNLINK are skipped if the event is coming from a child that has been // unlinked. func (w *Watches) Notify(name string, events, cookie uint32, et EventType, unlinked bool) { - // N.B. We don't defer the unlocks because Notify is in the hot path of - // all IO operations, and the defer costs too much for small IO - // operations. + var hasExpired bool w.mu.RLock() for _, watch := range w.ws { if unlinked && watch.ExcludeUnlinked() && et == PathEvent { continue } - watch.Notify(name, events, cookie) + if watch.Notify(name, events, cookie) { + hasExpired = true + } + } + w.mu.RUnlock() + + if hasExpired { + w.cleanupExpiredWatches() + } +} + +// This function is relatively expensive and should only be called where there +// are expired watches. +func (w *Watches) cleanupExpiredWatches() { + // Because of lock ordering, we cannot acquire Inotify.mu for each watch + // owner while holding w.mu. As a result, store expired watches locally + // before removing. + var toRemove []*Watch + w.mu.RLock() + for _, watch := range w.ws { + if atomic.LoadInt32(&watch.expired) == 1 { + toRemove = append(toRemove, watch) + } } w.mu.RUnlock() + for _, watch := range toRemove { + watch.owner.RmWatch(watch.wd) + } } // HandleDeletion is called when the watch target is destroyed. Clear the @@ -478,16 +500,10 @@ func (w *Watches) Notify(name string, events, cookie uint32, et EventType, unlin func (w *Watches) HandleDeletion() { w.Notify("", linux.IN_DELETE_SELF, 0, InodeEvent, true /* unlinked */) - // We can't hold w.mu while calling watch.handleDeletion to preserve lock - // ordering w.r.t to the owner inotify instances. Instead, atomically move - // the watches map into a local variable so we can iterate over it safely. - // - // Because of this however, it is possible for the watches' owners to reach - // this inode while the inode has no refs. This is still safe because the - // owners can only reach the inode until this function finishes calling - // watch.handleDeletion below and the inode is guaranteed to exist in the - // meantime. But we still have to be very careful not to rely on inode state - // that may have been already destroyed. + // As in Watches.Notify, we can't hold w.mu while acquiring Inotify.mu for + // the owner of each watch being deleted. Instead, atomically store the + // watches map in a local variable and set it to nil so we can iterate over + // it with the assurance that there will be no concurrent accesses. var ws map[uint64]*Watch w.mu.Lock() ws = w.ws @@ -519,17 +535,28 @@ func (w *Watches) HandleDeletion() { // +stateify savable type Watch struct { // Inotify instance which owns this watch. + // + // This field is immutable after creation. owner *Inotify // Descriptor for this watch. This is unique across an inotify instance. + // + // This field is immutable after creation. wd int32 // target is a dentry representing the watch target. Its watch set contains this watch. + // + // This field is immutable after creation. target *Dentry // Events being monitored via this watch. Must be accessed with atomic // memory operations. mask uint32 + + // expired is set to 1 to indicate that this watch is a one-shot that has + // already sent a notification and therefore can be removed. Must be accessed + // with atomic memory operations. + expired int32 } // OwnerID returns the id of the inotify instance that owns this watch. @@ -546,12 +573,20 @@ func (w *Watch) ExcludeUnlinked() bool { return atomic.LoadUint32(&w.mask)&linux.IN_EXCL_UNLINK != 0 } -// Notify queues a new event on this watch. -func (w *Watch) Notify(name string, events uint32, cookie uint32) { +// Notify queues a new event on this watch. Returns true if this is a one-shot +// watch that should be deleted, after this event was successfully queued. +func (w *Watch) Notify(name string, events uint32, cookie uint32) bool { + if atomic.LoadInt32(&w.expired) == 1 { + // This is a one-shot watch that is already in the process of being + // removed. This may happen if a second event reaches the watch target + // before this watch has been removed. + return false + } + mask := atomic.LoadUint32(&w.mask) if mask&events == 0 { // We weren't watching for this event. - return + return false } // Event mask should include bits matched from the watch plus all control @@ -560,6 +595,11 @@ func (w *Watch) Notify(name string, events uint32, cookie uint32) { effectiveMask := unmaskableBits | mask matchedEvents := effectiveMask & events w.owner.queueEvent(newEvent(w.wd, name, matchedEvents, cookie)) + if mask&linux.IN_ONESHOT != 0 { + atomic.StoreInt32(&w.expired, 1) + return true + } + return false } // Event represents a struct inotify_event from linux. diff --git a/test/syscalls/linux/inotify.cc b/test/syscalls/linux/inotify.cc index 731c7046c..bdb645c35 100644 --- a/test/syscalls/linux/inotify.cc +++ b/test/syscalls/linux/inotify.cc @@ -1485,20 +1485,26 @@ TEST(Inotify, DuplicateWatchReturnsSameWatchDescriptor) { TEST(Inotify, UnmatchedEventsAreDiscarded) { const TempPath root = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); - const TempPath file1 = + TempPath file1 = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(root.path())); const FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); - ASSERT_NO_ERRNO_AND_VALUE(InotifyAddWatch(fd.get(), file1.path(), IN_ACCESS)); + const int wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(fd.get(), file1.path(), IN_ACCESS)); - const FileDescriptor file1_fd = + FileDescriptor file1_fd = ASSERT_NO_ERRNO_AND_VALUE(Open(file1.path(), O_WRONLY)); - const std::vector events = - ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(fd.get())); + std::vector events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(fd.get())); // We only asked for access events, the open event should be discarded. ASSERT_THAT(events, Are({})); + + // IN_IGNORED events are always generated, regardless of the mask. + file1_fd.reset(); + file1.reset(); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(fd.get())); + ASSERT_THAT(events, Are({Event(IN_IGNORED, wd)})); } TEST(Inotify, AddWatchWithInvalidEventMaskFails) { @@ -2073,6 +2079,38 @@ TEST(Inotify, ExcludeUnlinkInodeEvents_NoRandomSave) { })); } +TEST(Inotify, OneShot) { + // TODO(gvisor.dev/issue/1624): IN_ONESHOT not supported in VFS1. + SKIP_IF(IsRunningWithVFS1()); + + const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); + const FileDescriptor inotify_fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + + const int wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(inotify_fd.get(), file.path(), IN_MODIFY | IN_ONESHOT)); + + // Open an fd, write to it, and then close it. + FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(file.path(), O_WRONLY)); + ASSERT_THAT(write(fd.get(), "x", 1), SyscallSucceedsWithValue(1)); + fd.reset(); + + // We should get a single event followed by IN_IGNORED indicating removal + // of the one-shot watch. Prior activity (i.e. open) that is not in the mask + // should not trigger removal, and activity after removal (i.e. close) should + // not generate events. + std::vector events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({ + Event(IN_MODIFY, wd), + Event(IN_IGNORED, wd), + })); + + // The watch should already have been removed. + EXPECT_THAT(inotify_rm_watch(inotify_fd.get(), wd), + SyscallFailsWithErrno(EINVAL)); +} + // This test helps verify that the lock order of filesystem and inotify locks // is respected when inotify instances and watch targets are concurrently being // destroyed. -- cgit v1.2.3 From 9cfc15497581824f1c6ba2b9f9ee653d0be0bc5a Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Fri, 26 Jun 2020 16:23:15 -0700 Subject: Require CAP_SYS_ADMIN in the root user namespace for TTY theft PiperOrigin-RevId: 318563543 --- pkg/sentry/kernel/thread_group.go | 3 ++- test/syscalls/linux/pty_root.cc | 22 ++++++++++++++++------ 2 files changed, 18 insertions(+), 7 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/thread_group.go b/pkg/sentry/kernel/thread_group.go index 52849f5b3..4dfd2c990 100644 --- a/pkg/sentry/kernel/thread_group.go +++ b/pkg/sentry/kernel/thread_group.go @@ -366,7 +366,8 @@ func (tg *ThreadGroup) SetControllingTTY(tty *TTY, arg int32) error { // terminal is stolen, and all processes that had it as controlling // terminal lose it." - tty_ioctl(4) if tty.tg != nil && tg.processGroup.session != tty.tg.processGroup.session { - if !auth.CredentialsFromContext(tg.leader).HasCapability(linux.CAP_SYS_ADMIN) || arg != 1 { + // Stealing requires CAP_SYS_ADMIN in the root user namespace. + if creds := auth.CredentialsFromContext(tg.leader); !creds.HasCapabilityIn(linux.CAP_SYS_ADMIN, creds.UserNamespace.Root()) || arg != 1 { return syserror.EPERM } // Steal the TTY away. Unlike TIOCNOTTY, don't send signals. diff --git a/test/syscalls/linux/pty_root.cc b/test/syscalls/linux/pty_root.cc index 14a4af980..1d7dbefdb 100644 --- a/test/syscalls/linux/pty_root.cc +++ b/test/syscalls/linux/pty_root.cc @@ -25,16 +25,26 @@ namespace gvisor { namespace testing { -// These tests should be run as root. namespace { +// StealTTY tests whether privileged processes can steal controlling terminals. +// If the stealing process has CAP_SYS_ADMIN in the root user namespace, the +// test ensures that stealing works. If it has non-root CAP_SYS_ADMIN, it +// ensures stealing fails. TEST(JobControlRootTest, StealTTY) { SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SYS_ADMIN))); - // Make this a session leader, which also drops the controlling terminal. - // In the gVisor test environment, this test will be run as the session - // leader already (as the sentry init process). + bool true_root = true; if (!IsRunningOnGvisor()) { + // If running in Linux, we may only have CAP_SYS_ADMIN in a non-root user + // namespace (i.e. we are not truly root). We use init_module as a proxy for + // whether we are true root, as it returns EPERM immediately. + ASSERT_THAT(syscall(SYS_init_module, nullptr, 0, nullptr), SyscallFails()); + true_root = errno != EPERM; + + // Make this a session leader, which also drops the controlling terminal. + // In the gVisor test environment, this test will be run as the session + // leader already (as the sentry init process). ASSERT_THAT(setsid(), SyscallSucceeds()); } @@ -53,8 +63,8 @@ TEST(JobControlRootTest, StealTTY) { ASSERT_THAT(setsid(), SyscallSucceeds()); // We shouldn't be able to steal the terminal with the wrong arg value. TEST_PCHECK(ioctl(slave.get(), TIOCSCTTY, 0)); - // We should be able to steal it here. - TEST_PCHECK(!ioctl(slave.get(), TIOCSCTTY, 1)); + // We should be able to steal it if we are true root. + TEST_PCHECK(true_root == !ioctl(slave.get(), TIOCSCTTY, 1)); _exit(0); } -- cgit v1.2.3 From 8dbeac53ce1b3c1cf4a5f2f0ccdd7196f4656fd8 Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Fri, 26 Jun 2020 17:49:32 -0700 Subject: Implement SO_NO_CHECK socket option. SO_NO_CHECK is used to skip the UDP checksum generation on a TX socket (UDP checksum is optional on IPv4). Test: - TestNoChecksum - SoNoCheckOffByDefault (UdpSocketTest) - SoNoCheck (UdpSocketTest) Fixes #3055 PiperOrigin-RevId: 318575215 --- pkg/sentry/socket/netstack/netstack.go | 19 +++++ pkg/sentry/socket/socket.go | 1 - pkg/tcpip/checker/checker.go | 16 +++++ pkg/tcpip/tcpip.go | 100 +++++++++++++++------------ pkg/tcpip/transport/udp/endpoint.go | 24 +++++-- pkg/tcpip/transport/udp/udp_test.go | 24 +++++++ test/syscalls/linux/udp_socket_test_cases.cc | 38 ++++++++++ 7 files changed, 173 insertions(+), 49 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index c0b63a803..e7d2c83d7 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -1169,6 +1169,17 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam return int32(v), nil + case linux.SO_NO_CHECK: + if outLen < sizeOfInt32 { + return nil, syserr.ErrInvalidArgument + } + + v, err := ep.GetSockOptBool(tcpip.NoChecksumOption) + if err != nil { + return nil, syserr.TranslateNetstackError(err) + } + return boolToInt32(v), nil + default: socket.GetSockOptEmitUnimplementedEvent(t, name) } @@ -1720,6 +1731,14 @@ func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, nam return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.OutOfBandInlineOption(v))) + case linux.SO_NO_CHECK: + if len(optVal) < sizeOfInt32 { + return syserr.ErrInvalidArgument + } + + v := usermem.ByteOrder.Uint32(optVal) + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.NoChecksumOption, v != 0)) + case linux.SO_LINGER: if len(optVal) < linux.SizeOfLinger { return syserr.ErrInvalidArgument diff --git a/pkg/sentry/socket/socket.go b/pkg/sentry/socket/socket.go index 6580bd6e9..fcd7f9d7f 100644 --- a/pkg/sentry/socket/socket.go +++ b/pkg/sentry/socket/socket.go @@ -407,7 +407,6 @@ func emitUnimplementedEvent(t *kernel.Task, name int) { linux.SO_MARK, linux.SO_MAX_PACING_RATE, linux.SO_NOFCS, - linux.SO_NO_CHECK, linux.SO_OOBINLINE, linux.SO_PASSCRED, linux.SO_PASSSEC, diff --git a/pkg/tcpip/checker/checker.go b/pkg/tcpip/checker/checker.go index c1745ba6a..ee264b726 100644 --- a/pkg/tcpip/checker/checker.go +++ b/pkg/tcpip/checker/checker.go @@ -320,6 +320,22 @@ func DstPort(port uint16) TransportChecker { } } +// NoChecksum creates a checker that checks if the checksum is zero. +func NoChecksum(noChecksum bool) TransportChecker { + return func(t *testing.T, h header.Transport) { + t.Helper() + + udp, ok := h.(header.UDP) + if !ok { + return + } + + if b := udp.Checksum() == 0; b != noChecksum { + t.Errorf("bad checksum state, got %t, want %t", b, noChecksum) + } + } +} + // SeqNum creates a checker that checks the sequence number. func SeqNum(seq uint32) TransportChecker { return func(t *testing.T, h header.Transport) { diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 4d45dcc42..2be1c107a 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -585,59 +585,68 @@ type WriteOptions struct { type SockOptBool int const ( - // BroadcastOption is used by SetSockOpt/GetSockOpt to specify whether - // datagram sockets are allowed to send packets to a broadcast address. + // BroadcastOption is used by SetSockOptBool/GetSockOptBool to specify + // whether datagram sockets are allowed to send packets to a broadcast + // address. BroadcastOption SockOptBool = iota - // CorkOption is used by SetSockOpt/GetSockOpt to specify if data should be - // held until segments are full by the TCP transport protocol. + // CorkOption is used by SetSockOptBool/GetSockOptBool to specify if + // data should be held until segments are full by the TCP transport + // protocol. CorkOption - // DelayOption is used by SetSockOpt/GetSockOpt to specify if data - // should be sent out immediately by the transport protocol. For TCP, - // it determines if the Nagle algorithm is on or off. + // DelayOption is used by SetSockOptBool/GetSockOptBool to specify if + // data should be sent out immediately by the transport protocol. For + // TCP, it determines if the Nagle algorithm is on or off. DelayOption - // KeepaliveEnabledOption is used by SetSockOpt/GetSockOpt to specify whether - // TCP keepalive is enabled for this socket. + // KeepaliveEnabledOption is used by SetSockOptBool/GetSockOptBool to + // specify whether TCP keepalive is enabled for this socket. KeepaliveEnabledOption - // MulticastLoopOption is used by SetSockOpt/GetSockOpt to specify whether - // multicast packets sent over a non-loopback interface will be looped back. + // MulticastLoopOption is used by SetSockOptBool/GetSockOptBool to + // specify whether multicast packets sent over a non-loopback interface + // will be looped back. MulticastLoopOption - // PasscredOption is used by SetSockOpt/GetSockOpt to specify whether - // SCM_CREDENTIALS socket control messages are enabled. + // NoChecksumOption is used by SetSockOptBool/GetSockOptBool to specify + // whether UDP checksum is disabled for this socket. + NoChecksumOption + + // PasscredOption is used by SetSockOptBool/GetSockOptBool to specify + // whether SCM_CREDENTIALS socket control messages are enabled. // // Only supported on Unix sockets. PasscredOption - // QuickAckOption is stubbed out in SetSockOpt/GetSockOpt. + // QuickAckOption is stubbed out in SetSockOptBool/GetSockOptBool. QuickAckOption - // ReceiveTClassOption is used by SetSockOpt/GetSockOpt to specify if the - // IPV6_TCLASS ancillary message is passed with incoming packets. + // ReceiveTClassOption is used by SetSockOptBool/GetSockOptBool to + // specify if the IPV6_TCLASS ancillary message is passed with incoming + // packets. ReceiveTClassOption - // ReceiveTOSOption is used by SetSockOpt/GetSockOpt to specify if the TOS - // ancillary message is passed with incoming packets. + // ReceiveTOSOption is used by SetSockOptBool/GetSockOptBool to specify + // if the TOS ancillary message is passed with incoming packets. ReceiveTOSOption - // ReceiveIPPacketInfoOption is used by {G,S}etSockOptBool to specify - // if more inforamtion is provided with incoming packets such - // as interface index and address. + // ReceiveIPPacketInfoOption is used by SetSockOptBool/GetSockOptBool to + // specify if more inforamtion is provided with incoming packets such as + // interface index and address. ReceiveIPPacketInfoOption - // ReuseAddressOption is used by SetSockOpt/GetSockOpt to specify whether Bind() - // should allow reuse of local address. + // ReuseAddressOption is used by SetSockOptBool/GetSockOptBool to + // specify whether Bind() should allow reuse of local address. ReuseAddressOption - // ReusePortOption is used by SetSockOpt/GetSockOpt to permit multiple sockets - // to be bound to an identical socket address. + // ReusePortOption is used by SetSockOptBool/GetSockOptBool to permit + // multiple sockets to be bound to an identical socket address. ReusePortOption - // V6OnlyOption is used by {G,S}etSockOptBool to specify whether an IPv6 - // socket is to be restricted to sending and receiving IPv6 packets only. + // V6OnlyOption is used by SetSockOptBool/GetSockOptBool to specify + // whether an IPv6 socket is to be restricted to sending and receiving + // IPv6 packets only. V6OnlyOption ) @@ -645,25 +654,27 @@ const ( type SockOptInt int const ( - // KeepaliveCountOption is used by SetSockOpt/GetSockOpt to specify the number - // of un-ACKed TCP keepalives that will be sent before the connection is - // closed. + // KeepaliveCountOption is used by SetSockOptInt/GetSockOptInt to + // specify the number of un-ACKed TCP keepalives that will be sent + // before the connection is closed. KeepaliveCountOption SockOptInt = iota - // IPv4TOSOption is used by SetSockOpt/GetSockOpt to specify TOS + // IPv4TOSOption is used by SetSockOptInt/GetSockOptInt to specify TOS // for all subsequent outgoing IPv4 packets from the endpoint. IPv4TOSOption - // IPv6TrafficClassOption is used by SetSockOpt/GetSockOpt to specify TOS - // for all subsequent outgoing IPv6 packets from the endpoint. + // IPv6TrafficClassOption is used by SetSockOptInt/GetSockOptInt to + // specify TOS for all subsequent outgoing IPv6 packets from the + // endpoint. IPv6TrafficClassOption - // MaxSegOption is used by SetSockOpt/GetSockOpt to set/get the current - // Maximum Segment Size(MSS) value as specified using the TCP_MAXSEG option. + // MaxSegOption is used by SetSockOptInt/GetSockOptInt to set/get the + // current Maximum Segment Size(MSS) value as specified using the + // TCP_MAXSEG option. MaxSegOption - // MulticastTTLOption is used by SetSockOpt/GetSockOpt to control the default - // TTL value for multicast messages. The default is 1. + // MulticastTTLOption is used by SetSockOptInt/GetSockOptInt to control + // the default TTL value for multicast messages. The default is 1. MulticastTTLOption // ReceiveQueueSizeOption is used in GetSockOptInt to specify that the @@ -682,21 +693,22 @@ const ( // number of unread bytes in the output buffer should be returned. SendQueueSizeOption - // TTLOption is used by SetSockOpt/GetSockOpt to control the default TTL/hop - // limit value for unicast messages. The default is protocol specific. + // TTLOption is used by SetSockOptInt/GetSockOptInt to control the + // default TTL/hop limit value for unicast messages. The default is + // protocol specific. // // A zero value indicates the default. TTLOption - // TCPSynCountOption is used by SetSockOpt/GetSockOpt to specify the number of - // SYN retransmits that TCP should send before aborting the attempt to - // connect. It cannot exceed 255. + // TCPSynCountOption is used by SetSockOptInt/GetSockOptInt to specify + // the number of SYN retransmits that TCP should send before aborting + // the attempt to connect. It cannot exceed 255. // // NOTE: This option is currently only stubbed out and is no-op. TCPSynCountOption - // TCPWindowClampOption is used by SetSockOpt/GetSockOpt to bound the size - // of the advertised window to this value. + // TCPWindowClampOption is used by SetSockOptInt/GetSockOptInt to bound + // the size of the advertised window to this value. // // NOTE: This option is currently only stubed out and is a no-op TCPWindowClampOption diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 8bdc1ee1f..cae29fbff 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -109,6 +109,7 @@ type endpoint struct { portFlags ports.Flags bindToDevice tcpip.NICID broadcast bool + noChecksum bool lastErrorMu sync.Mutex `state:"nosave"` lastError *tcpip.Error `state:".(string)"` @@ -529,7 +530,7 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c useDefaultTTL = false } - if err := sendUDP(route, buffer.View(v).ToVectorisedView(), e.ID.LocalPort, dstPort, ttl, useDefaultTTL, e.sendTOS, e.owner); err != nil { + if err := sendUDP(route, buffer.View(v).ToVectorisedView(), e.ID.LocalPort, dstPort, ttl, useDefaultTTL, e.sendTOS, e.owner, e.noChecksum); err != nil { return 0, nil, err } return int64(len(v)), nil, nil @@ -553,6 +554,11 @@ func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { e.multicastLoop = v e.mu.Unlock() + case tcpip.NoChecksumOption: + e.mu.Lock() + e.noChecksum = v + e.mu.Unlock() + case tcpip.ReceiveTOSOption: e.mu.Lock() e.receiveTOS = v @@ -825,6 +831,12 @@ func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { e.mu.RUnlock() return v, nil + case tcpip.NoChecksumOption: + e.mu.RLock() + v := e.noChecksum + e.mu.RUnlock() + return v, nil + case tcpip.ReceiveTOSOption: e.mu.RLock() v := e.receiveTOS @@ -959,7 +971,7 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { // sendUDP sends a UDP segment via the provided network endpoint and under the // provided identity. -func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort uint16, ttl uint8, useDefaultTTL bool, tos uint8, owner tcpip.PacketOwner) *tcpip.Error { +func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort uint16, ttl uint8, useDefaultTTL bool, tos uint8, owner tcpip.PacketOwner, noChecksum bool) *tcpip.Error { // Allocate a buffer for the UDP header. hdr := buffer.NewPrependable(header.UDPMinimumSize + int(r.MaxHeaderLength())) @@ -973,8 +985,12 @@ func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort u Length: length, }) - // Only calculate the checksum if offloading isn't supported. - if r.Capabilities()&stack.CapabilityTXChecksumOffload == 0 { + // Set the checksum field unless TX checksum offload is enabled. + // On IPv4, UDP checksum is optional, and a zero value indicates the + // transmitter skipped the checksum generation (RFC768). + // On IPv6, UDP checksum is not optional (RFC2460 Section 8.1). + if r.Capabilities()&stack.CapabilityTXChecksumOffload == 0 && + (!noChecksum || r.NetProto == header.IPv6ProtocolNumber) { xsum := r.PseudoHeaderChecksum(ProtocolNumber, length) for _, v := range data.Views() { xsum = header.Checksum(v, xsum) diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index ff9f60cf9..db59eb5a0 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -1251,6 +1251,30 @@ func TestWriteIncrementsPacketsSent(t *testing.T) { } } +func TestNoChecksum(t *testing.T) { + for _, flow := range []testFlow{unicastV4, unicastV6} { + t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpointForFlow(flow) + + // Disable the checksum generation. + if err := c.ep.SetSockOptBool(tcpip.NoChecksumOption, true); err != nil { + t.Fatalf("SetSockOptBool failed: %s", err) + } + // This option is effective on IPv4 only. + testWrite(c, flow, checker.UDP(checker.NoChecksum(flow.isV4()))) + + // Enable the checksum generation. + if err := c.ep.SetSockOptBool(tcpip.NoChecksumOption, false); err != nil { + t.Fatalf("SetSockOptBool failed: %s", err) + } + testWrite(c, flow, checker.UDP(checker.NoChecksum(false))) + }) + } +} + func TestTTL(t *testing.T) { for _, flow := range []testFlow{unicastV4, unicastV4in6, unicastV6, unicastV6Only, multicastV4, multicastV4in6, multicastV6, broadcast, broadcastIn6} { t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { diff --git a/test/syscalls/linux/udp_socket_test_cases.cc b/test/syscalls/linux/udp_socket_test_cases.cc index cc1db3de8..1d13432ca 100644 --- a/test/syscalls/linux/udp_socket_test_cases.cc +++ b/test/syscalls/linux/udp_socket_test_cases.cc @@ -1239,6 +1239,44 @@ TEST_P(UdpSocketTest, FIONREADZeroLengthWriteShutdown) { EXPECT_EQ(n, 0); } +TEST_P(UdpSocketTest, SoNoCheckOffByDefault) { + // TODO(gvisor.dev/issue/1202): SO_NO_CHECK socket option not supported by + // hostinet. + SKIP_IF(IsRunningWithHostinet()); + + int v = -1; + socklen_t optlen = sizeof(v); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_NO_CHECK, &v, &optlen), + SyscallSucceeds()); + ASSERT_EQ(v, kSockOptOff); + ASSERT_EQ(optlen, sizeof(v)); +} + +TEST_P(UdpSocketTest, SoNoCheck) { + // TODO(gvisor.dev/issue/1202): SO_NO_CHECK socket option not supported by + // hostinet. + SKIP_IF(IsRunningWithHostinet()); + + int v = kSockOptOn; + socklen_t optlen = sizeof(v); + ASSERT_THAT(setsockopt(s_, SOL_SOCKET, SO_NO_CHECK, &v, optlen), + SyscallSucceeds()); + v = -1; + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_NO_CHECK, &v, &optlen), + SyscallSucceeds()); + ASSERT_EQ(v, kSockOptOn); + ASSERT_EQ(optlen, sizeof(v)); + + v = kSockOptOff; + ASSERT_THAT(setsockopt(s_, SOL_SOCKET, SO_NO_CHECK, &v, optlen), + SyscallSucceeds()); + v = -1; + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_NO_CHECK, &v, &optlen), + SyscallSucceeds()); + ASSERT_EQ(v, kSockOptOff); + ASSERT_EQ(optlen, sizeof(v)); +} + TEST_P(UdpSocketTest, SoTimestampOffByDefault) { // TODO(gvisor.dev/issue/1202): SO_TIMESTAMP socket option not supported by // hostinet. -- cgit v1.2.3 From 691c04278ee6cf579e2b2dafb28e39861ce21bb9 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Sat, 27 Jun 2020 09:17:52 -0700 Subject: Add documentation for vfs2 inotify. Updates #1479. PiperOrigin-RevId: 318631247 --- pkg/sentry/vfs/g3doc/inotify.md | 210 ++++++++++++++++++++++++++++++++++++++++ 1 file changed, 210 insertions(+) create mode 100644 pkg/sentry/vfs/g3doc/inotify.md (limited to 'pkg/sentry') diff --git a/pkg/sentry/vfs/g3doc/inotify.md b/pkg/sentry/vfs/g3doc/inotify.md new file mode 100644 index 000000000..e7da49faa --- /dev/null +++ b/pkg/sentry/vfs/g3doc/inotify.md @@ -0,0 +1,210 @@ +# Inotify + +Inotify is a mechanism for monitoring filesystem events in Linux--see +inotify(7). An inotify instance can be used to monitor files and directories for +modifications, creation/deletion, etc. The inotify API consists of system calls +that create inotify instances (inotify_init/inotify_init1) and add/remove +watches on files to an instance (inotify_add_watch/inotify_rm_watch). Events are +generated from various places in the sentry, including the syscall layer, the +vfs layer, the process fd table, and within each filesystem implementation. This +document outlines the implementation details of inotify in VFS2. + +## Inotify Objects + +Inotify data structures are implemented in the vfs package. + +### vfs.Inotify + +Inotify instances are represented by vfs.Inotify objects, which implement +vfs.FileDescriptionImpl. As in Linux, inotify fds are backed by a +pseudo-filesystem (anonfs). Each inotify instance receives events from a set of +vfs.Watch objects, which can be modified with inotify_add_watch(2) and +inotify_rm_watch(2). An application can retrieve events by reading the inotify +fd. + +### vfs.Watches + +The set of all watches held on a single file (i.e., the watch target) is stored +in vfs.Watches. Each watch will belong to a different inotify instance (an +instance can only have one watch on any watch target). The watches are stored in +a map indexed by their vfs.Inotify owner’s id. Hard links and file descriptions +to a single file will all share the same vfs.Watches. Activity on the target +causes its vfs.Watches to generate notifications on its watches’ inotify +instances. + +### vfs.Watch + +A single watch, owned by one inotify instance and applied to one watch target. +Both the vfs.Inotify owner and vfs.Watches on the target will hold a vfs.Watch, +which leads to some complicated locking behavior (see Lock Ordering). Whenever a +watch is notified of an event on its target, it will queue events to its inotify +instance for delivery to the user. + +### vfs.Event + +vfs.Event is a simple struct encapsulating all the fields for an inotify event. +It is generated by vfs.Watches and forwarded to the watches' owners. It is +serialized to the user during read(2) syscalls on the associated fs.Inotify's +fd. + +## Lock Ordering + +There are three locks related to the inotify implementation: + +Inotify.mu: the inotify instance lock. Inotify.evMu: the inotify event queue +lock. Watches.mu: the watch set lock, used to protect the collection of watches +on a target. + +The correct lock ordering for inotify code is: + +Inotify.mu -> Watches.mu -> Inotify.evMu. + +Note that we use a distinct lock to protect the inotify event queue. If we +simply used Inotify.mu, we could simultaneously have locks being acquired in the +order of Inotify.mu -> Watches.mu and Watches.mu -> Inotify.mu, which would +cause deadlocks. For instance, adding a watch to an inotify instance would +require locking Inotify.mu, and then adding the same watch to the target would +cause Watches.mu to be held. At the same time, generating an event on the target +would require Watches.mu to be held before iterating through each watch, and +then notifying the owner of each watch would cause Inotify.mu to be held. + +See the vfs package comment to understand how inotify locks fit into the overall +ordering of filesystem locks. + +## Watch Targets in Different Filesystem Implementations + +In Linux, watches reside on inodes at the virtual filesystem layer. As a result, +all hard links and file descriptions on a single file will all share the same +watch set. In VFS2, there is no common inode structure across filesystem types +(some may not even have inodes), so we have to plumb inotify support through +each specific filesystem implementation. Some of the technical considerations +are outlined below. + +### Tmpfs + +For filesystems with inodes, like tmpfs, the design is quite similar to that of +Linux, where watches reside on the inode. + +### Pseudo-filesystems + +Technically, because inotify is implemented at the vfs layer in Linux, +pseudo-filesystems on top of kernfs support inotify passively. However, watches +can only track explicit filesystem operations like read/write, open/close, +mknod, etc., so watches on a target like /proc/self/fd will not generate events +every time a new fd is added or removed. As of this writing, we leave inotify +unimplemented in kernfs and anonfs; it does not seem particularly useful. + +### Gofer Filesystem (fsimpl/gofer) + +The gofer filesystem has several traits that make it difficult to support +inotify: + +* **There are no inodes.** A file is represented as a dentry that holds an + unopened p9 file (and possibly an open FID), through which the Sentry + interacts with the gofer. + * *Solution:* Because there is no inode structure stored in the sandbox, + inotify watches must be held on the dentry. This would be an issue in + the presence of hard links, where multiple dentries would need to share + the same set of watches, but in VFS2, we do not support the internal + creation of hard links on gofer fs. As a result, we make the assumption + that every dentry corresponds to a unique inode. However, the next point + raises an issue with this assumption: +* **The Sentry cannot always be aware of hard links on the remote + filesystem.** There is no way for us to confirm whether two files on the + remote filesystem are actually links to the same inode. QIDs and inodes are + not always 1:1. The assumption that dentries and inodes are 1:1 is + inevitably broken if there are remote hard links that we cannot detect. + * *Solution:* this is an issue with gofer fs in general, not only inotify, + and we will have to live with it. +* **Dentries can be cached, and then evicted.** Dentry lifetime does not + correspond to file lifetime. Because gofer fs is not entirely in-memory, the + absence of a dentry does not mean that the corresponding file does not + exist, nor does a dentry reaching zero references mean that the + corresponding file no longer exists. When a dentry reaches zero references, + it will be cached, in case the file at that path is needed again in the + future. However, the dentry may be evicted from the cache, which will cause + a new dentry to be created next time the same file path is used. The + existing watches will be lost. + * *Solution:* When a dentry reaches zero references, do not cache it if it + has any watches, so we can avoid eviction/destruction. Note that if the + dentry was deleted or invalidated (d.vfsd.IsDead()), we should still + destroy it along with its watches. Additionally, when a dentry’s last + watch is removed, we cache it if it also has zero references. This way, + the dentry can eventually be evicted from memory if it is no longer + needed. +* **Dentries can be invalidated.** Another issue with dentry lifetime is that + the remote file at the file path represented may change from underneath the + dentry. In this case, the next time that the dentry is used, it will be + invalidated and a new dentry will replace it. In this case, it is not clear + what should be done with the watches on the old dentry. + * *Solution:* Silently destroy the watches when invalidation occurs. We + have no way of knowing exactly what happened, when it happens. Inotify + instances on NFS files in Linux probably behave in a similar fashion, + since inotify is implemented at the vfs layer and is not aware of the + complexities of remote file systems. + * An alternative would be to issue some kind of event upon invalidation, + e.g. a delete event, but this has several issues: + * We cannot discern whether the remote file was invalidated because it was + moved, deleted, etc. This information is crucial, because these cases + should result in different events. Furthermore, the watches should only + be destroyed if the file has been deleted. + * Moreover, the mechanism for detecting whether the underlying file has + changed is to check whether a new QID is given by the gofer. This may + result in false positives, e.g. suppose that the server closed and + re-opened the same file, which may result in a new QID. + * Finally, the time of the event may be completely different from the time + of the file modification, since a dentry is not immediately notified + when the underlying file has changed. It would be quite unexpected to + receive the notification when invalidation was triggered, i.e. the next + time the file was accessed within the sandbox, because then the + read/write/etc. operation on the file would not result in the expected + event. + * Another point in favor of the first solution: inotify in Linux can + already be lossy on local filesystems (one of the sacrifices made so + that filesystem performance isn’t killed), and it is lossy on NFS for + similar reasons to gofer fs. Therefore, it is better for inotify to be + silent than to emit incorrect notifications. +* **There may be external users of the remote filesystem.** We can only track + operations performed on the file within the sandbox. This is sufficient + under InteropModeExclusive, but whenever there are external users, the set + of actions we are aware of is incomplete. + * *Solution:* We could either return an error or just issue a warning when + inotify is used without InteropModeExclusive. Although faulty, VFS1 + allows it when the filesystem is shared, and Linux does the same for + remote filesystems (as mentioned above, inotify sits at the vfs level). + +## Dentry Interface + +For events that must be generated above the vfs layer, we provide the following +DentryImpl methods to allow interactions with targets on any FilesystemImpl: + +* **InotifyWithParent()** generates events on the dentry’s watches as well as + its parent’s. +* **Watches()** retrieves the watch set of the target represented by the + dentry. This is used to access and modify watches on a target. +* **OnZeroWatches()** performs cleanup tasks after the last watch is removed + from a dentry. This is needed by gofer fs, which must allow a watched dentry + to be cached once it has no more watches. Most implementations can just do + nothing. Note that OnZeroWatches() must be called after all inotify locks + are released to preserve lock ordering, since it may acquire + FilesystemImpl-specific locks. + +## IN_EXCL_UNLINK + +There are several options that can be set for a watch, specified as part of the +mask in inotify_add_watch(2). In particular, IN_EXCL_UNLINK requires some +additional support in each filesystem. + +A watch with IN_EXCL_UNLINK will not generate events for its target if it +corresponds to a path that was unlinked. For instance, if an fd is opened on +“foo/bar” and “foo/bar” is subsequently unlinked, any reads/writes/etc. on the +fd will be ignored by watches on “foo” or “foo/bar” with IN_EXCL_UNLINK. This +requires each DentryImpl to keep track of whether it has been unlinked, in order +to determine whether events should be sent to watches with IN_EXCL_UNLINK. + +## IN_ONESHOT + +One-shot watches expire after generating a single event. When an event occurs, +all one-shot watches on the target that successfully generated an event are +removed. Lock ordering can cause the management of one-shot watches to be quite +expensive; see Watches.Notify() for more information. -- cgit v1.2.3 From 02d552d07c4415978d2ce418fb16baf238d0ff78 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Sat, 27 Jun 2020 14:38:20 -0700 Subject: Support sticky bit in vfs2. Updates #2923. PiperOrigin-RevId: 318648128 --- pkg/sentry/fsimpl/gofer/filesystem.go | 65 ++++++++++++++++++++++++----- pkg/sentry/fsimpl/gofer/gofer.go | 4 ++ pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 4 ++ pkg/sentry/fsimpl/tmpfs/directory.go | 4 ++ pkg/sentry/fsimpl/tmpfs/filesystem.go | 9 ++++ pkg/sentry/vfs/permissions.go | 14 +++++++ test/syscalls/BUILD | 1 + test/syscalls/linux/sticky.cc | 57 +++++++++++++++++-------- 8 files changed, 131 insertions(+), 27 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 73bac738d..51082359d 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -16,6 +16,7 @@ package gofer import ( "sync" + "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" @@ -464,21 +465,61 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b defer mntns.DecRef() parent.dirMu.Lock() defer parent.dirMu.Unlock() + child, ok := parent.children[name] if ok && child == nil { return syserror.ENOENT } - // We only need a dentry representing the file at name if it can be a mount - // point. If child is nil, then it can't be a mount point. If child is - // non-nil but stale, the actual file can't be a mount point either; we - // detect this case by just speculatively calling PrepareDeleteDentry and - // only revalidating the dentry if that fails (indicating that the existing - // dentry is a mount point). + + sticky := atomic.LoadUint32(&parent.mode)&linux.ModeSticky != 0 + if sticky { + if !ok { + // If the sticky bit is set, we need to retrieve the child to determine + // whether removing it is allowed. + child, err = fs.stepLocked(ctx, rp, parent, false /* mayFollowSymlinks */, &ds) + if err != nil { + return err + } + } else if child != nil && !child.cachedMetadataAuthoritative() { + // Make sure the dentry representing the file at name is up to date + // before examining its metadata. + child, err = fs.revalidateChildLocked(ctx, vfsObj, parent, name, child, &ds) + if err != nil { + return err + } + } + if err := parent.mayDelete(rp.Credentials(), child); err != nil { + return err + } + } + + // If a child dentry exists, prepare to delete it. This should fail if it is + // a mount point. We detect mount points by speculatively calling + // PrepareDeleteDentry, which fails if child is a mount point. However, we + // may need to revalidate the file in this case to make sure that it has not + // been deleted or replaced on the remote fs, in which case the mount point + // will have disappeared. If calling PrepareDeleteDentry fails again on the + // up-to-date dentry, we can be sure that it is a mount point. + // + // Also note that if child is nil, then it can't be a mount point. if child != nil { + // Hold child.dirMu so we can check child.children and + // child.syntheticChildren. We don't access these fields until a bit later, + // but locking child.dirMu after calling vfs.PrepareDeleteDentry() would + // create an inconsistent lock ordering between dentry.dirMu and + // vfs.Dentry.mu (in the VFS lock order, it would make dentry.dirMu both "a + // FilesystemImpl lock" and "a lock acquired by a FilesystemImpl between + // PrepareDeleteDentry and CommitDeleteDentry). To avoid this, lock + // child.dirMu before calling PrepareDeleteDentry. child.dirMu.Lock() defer child.dirMu.Unlock() if err := vfsObj.PrepareDeleteDentry(mntns, &child.vfsd); err != nil { - if parent.cachedMetadataAuthoritative() { + // We can skip revalidation in several cases: + // - We are not in InteropModeShared + // - The parent directory is synthetic, in which case the child must also + // be synthetic + // - We already updated the child during the sticky bit check above + if parent.cachedMetadataAuthoritative() || sticky { return err } child, err = fs.revalidateChildLocked(ctx, vfsObj, parent, name, child, &ds) @@ -1100,7 +1141,8 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa return err } } - if err := oldParent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { + creds := rp.Credentials() + if err := oldParent.checkPermissions(creds, vfs.MayWrite|vfs.MayExec); err != nil { return err } vfsObj := rp.VirtualFilesystem() @@ -1115,12 +1157,15 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa if renamed == nil { return syserror.ENOENT } + if err := oldParent.mayDelete(creds, renamed); err != nil { + return err + } if renamed.isDir() { if renamed == newParent || genericIsAncestorDentry(renamed, newParent) { return syserror.EINVAL } if oldParent != newParent { - if err := renamed.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil { + if err := renamed.checkPermissions(creds, vfs.MayWrite); err != nil { return err } } @@ -1131,7 +1176,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa } if oldParent != newParent { - if err := newParent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { + if err := newParent.checkPermissions(creds, vfs.MayWrite|vfs.MayExec); err != nil { return err } newParent.dirMu.Lock() diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index b0b6d8c64..71c8d3ae1 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -1003,6 +1003,10 @@ func (d *dentry) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes) return vfs.GenericCheckPermissions(creds, ats, linux.FileMode(atomic.LoadUint32(&d.mode)), auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))) } +func (d *dentry) mayDelete(creds *auth.Credentials, child *dentry) error { + return vfs.CheckDeleteSticky(creds, linux.FileMode(atomic.LoadUint32(&d.mode)), auth.KUID(atomic.LoadUint32(&child.uid))) +} + func dentryUIDFromP9UID(uid p9.UID) uint32 { if !uid.Ok() { return uint32(auth.OverflowUID) diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 53aec4918..4cb885d87 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -471,6 +471,8 @@ func (o *OrderedChildren) Unlink(ctx context.Context, name string, child *vfs.De if err := o.checkExistingLocked(name, child); err != nil { return err } + + // TODO(gvisor.dev/issue/3027): Check sticky bit before removing. o.removeLocked(name) return nil } @@ -518,6 +520,8 @@ func (o *OrderedChildren) Rename(ctx context.Context, oldname, newname string, c if err := o.checkExistingLocked(oldname, child); err != nil { return nil, err } + + // TODO(gvisor.dev/issue/3027): Check sticky bit before removing. replaced := dst.replaceChildLocked(newname, child) return replaced, nil } diff --git a/pkg/sentry/fsimpl/tmpfs/directory.go b/pkg/sentry/fsimpl/tmpfs/directory.go index b172abc15..0a1ad4765 100644 --- a/pkg/sentry/fsimpl/tmpfs/directory.go +++ b/pkg/sentry/fsimpl/tmpfs/directory.go @@ -81,6 +81,10 @@ func (dir *directory) removeChildLocked(child *dentry) { dir.iterMu.Unlock() } +func (dir *directory) mayDelete(creds *auth.Credentials, child *dentry) error { + return vfs.CheckDeleteSticky(creds, linux.FileMode(atomic.LoadUint32(&dir.inode.mode)), auth.KUID(atomic.LoadUint32(&child.inode.uid))) +} + type directoryFD struct { fileDescription vfs.DirectoryFileDescriptionDefaultImpl diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index f766b7ce2..71ac7b8e6 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -492,6 +492,9 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa if !ok { return syserror.ENOENT } + if err := oldParentDir.mayDelete(rp.Credentials(), renamed); err != nil { + return err + } // Note that we don't need to call rp.CheckMount(), since if renamed is a // mount point then we want to rename the mount point, not anything in the // mounted filesystem. @@ -606,6 +609,9 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error if !ok { return syserror.ENOENT } + if err := parentDir.mayDelete(rp.Credentials(), child); err != nil { + return err + } childDir, ok := child.inode.impl.(*directory) if !ok { return syserror.ENOTDIR @@ -716,6 +722,9 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error if !ok { return syserror.ENOENT } + if err := parentDir.mayDelete(rp.Credentials(), child); err != nil { + return err + } if child.inode.isDir() { return syserror.EISDIR } diff --git a/pkg/sentry/vfs/permissions.go b/pkg/sentry/vfs/permissions.go index afe2be8d7..9cb050597 100644 --- a/pkg/sentry/vfs/permissions.go +++ b/pkg/sentry/vfs/permissions.go @@ -230,6 +230,20 @@ func CheckSetStat(ctx context.Context, creds *auth.Credentials, stat *linux.Stat return nil } +// CheckDeleteSticky checks whether the sticky bit is set on a directory with +// the given file mode, and if so, checks whether creds has permission to +// remove a file owned by childKUID from a directory with the given mode. +// CheckDeleteSticky is consistent with fs/linux.h:check_sticky(). +func CheckDeleteSticky(creds *auth.Credentials, parentMode linux.FileMode, childKUID auth.KUID) error { + if parentMode&linux.ModeSticky == 0 { + return nil + } + if CanActAsOwner(creds, childKUID) { + return nil + } + return syserror.EPERM +} + // CanActAsOwner returns true if creds can act as the owner of a file with the // given owning UID, consistent with Linux's // fs/inode.c:inode_owner_or_capable(). diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index c4fff0ac8..36c178e4a 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -942,6 +942,7 @@ syscall_test( syscall_test( add_overlay = True, test = "//test/syscalls/linux:sticky_test", + vfs2 = "True", ) syscall_test( diff --git a/test/syscalls/linux/sticky.cc b/test/syscalls/linux/sticky.cc index 92eec0449..39f4fb801 100644 --- a/test/syscalls/linux/sticky.cc +++ b/test/syscalls/linux/sticky.cc @@ -40,11 +40,14 @@ namespace { TEST(StickyTest, StickyBitPermDenied) { SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SETUID))); - auto dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); - EXPECT_THAT(chmod(dir.path().c_str(), 0777 | S_ISVTX), SyscallSucceeds()); - const FileDescriptor dirfd = - ASSERT_NO_ERRNO_AND_VALUE(Open(dir.path(), O_DIRECTORY)); - ASSERT_THAT(mkdirat(dirfd.get(), "NewDir", 0755), SyscallSucceeds()); + const TempPath parent = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + EXPECT_THAT(chmod(parent.path().c_str(), 0777 | S_ISVTX), SyscallSucceeds()); + const TempPath file = ASSERT_NO_ERRNO_AND_VALUE( + TempPath::CreateFileWith(parent.path(), "some content", 0755)); + const TempPath dir = + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDirWith(parent.path(), 0755)); + const TempPath link = ASSERT_NO_ERRNO_AND_VALUE( + TempPath::CreateSymlinkTo(parent.path(), file.path())); // Drop privileges and change IDs only in child thread, or else this parent // thread won't be able to open some log files after the test ends. @@ -62,18 +65,26 @@ TEST(StickyTest, StickyBitPermDenied) { syscall(SYS_setresuid, -1, absl::GetFlag(FLAGS_scratch_uid), -1), SyscallSucceeds()); - EXPECT_THAT(unlinkat(dirfd.get(), "NewDir", AT_REMOVEDIR), + std::string new_path = NewTempAbsPath(); + EXPECT_THAT(rename(file.path().c_str(), new_path.c_str()), SyscallFailsWithErrno(EPERM)); + EXPECT_THAT(unlink(file.path().c_str()), SyscallFailsWithErrno(EPERM)); + EXPECT_THAT(rmdir(dir.path().c_str()), SyscallFailsWithErrno(EPERM)); + EXPECT_THAT(unlink(link.path().c_str()), SyscallFailsWithErrno(EPERM)); }); } TEST(StickyTest, StickyBitSameUID) { SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SETUID))); - auto dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); - EXPECT_THAT(chmod(dir.path().c_str(), 0777 | S_ISVTX), SyscallSucceeds()); - std::string path = JoinPath(dir.path(), "NewDir"); - ASSERT_THAT(mkdir(path.c_str(), 0755), SyscallSucceeds()); + const TempPath parent = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + EXPECT_THAT(chmod(parent.path().c_str(), 0777 | S_ISVTX), SyscallSucceeds()); + const TempPath file = ASSERT_NO_ERRNO_AND_VALUE( + TempPath::CreateFileWith(parent.path(), "some content", 0755)); + const TempPath dir = + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDirWith(parent.path(), 0755)); + const TempPath link = ASSERT_NO_ERRNO_AND_VALUE( + TempPath::CreateSymlinkTo(parent.path(), file.path())); // Drop privileges and change IDs only in child thread, or else this parent // thread won't be able to open some log files after the test ends. @@ -89,18 +100,26 @@ TEST(StickyTest, StickyBitSameUID) { SyscallSucceeds()); // We still have the same EUID. - EXPECT_THAT(rmdir(path.c_str()), SyscallSucceeds()); + std::string new_path = NewTempAbsPath(); + EXPECT_THAT(rename(file.path().c_str(), new_path.c_str()), + SyscallSucceeds()); + EXPECT_THAT(unlink(new_path.c_str()), SyscallSucceeds()); + EXPECT_THAT(rmdir(dir.path().c_str()), SyscallSucceeds()); + EXPECT_THAT(unlink(link.path().c_str()), SyscallSucceeds()); }); } TEST(StickyTest, StickyBitCapFOWNER) { SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SETUID))); - auto dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); - EXPECT_THAT(chmod(dir.path().c_str(), 0777 | S_ISVTX), SyscallSucceeds()); - const FileDescriptor dirfd = - ASSERT_NO_ERRNO_AND_VALUE(Open(dir.path(), O_DIRECTORY)); - ASSERT_THAT(mkdirat(dirfd.get(), "NewDir", 0755), SyscallSucceeds()); + const TempPath parent = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + EXPECT_THAT(chmod(parent.path().c_str(), 0777 | S_ISVTX), SyscallSucceeds()); + const TempPath file = ASSERT_NO_ERRNO_AND_VALUE( + TempPath::CreateFileWith(parent.path(), "some content", 0755)); + const TempPath dir = + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDirWith(parent.path(), 0755)); + const TempPath link = ASSERT_NO_ERRNO_AND_VALUE( + TempPath::CreateSymlinkTo(parent.path(), file.path())); // Drop privileges and change IDs only in child thread, or else this parent // thread won't be able to open some log files after the test ends. @@ -117,8 +136,12 @@ TEST(StickyTest, StickyBitCapFOWNER) { SyscallSucceeds()); EXPECT_NO_ERRNO(SetCapability(CAP_FOWNER, true)); - EXPECT_THAT(unlinkat(dirfd.get(), "NewDir", AT_REMOVEDIR), + std::string new_path = NewTempAbsPath(); + EXPECT_THAT(rename(file.path().c_str(), new_path.c_str()), SyscallSucceeds()); + EXPECT_THAT(unlink(new_path.c_str()), SyscallSucceeds()); + EXPECT_THAT(rmdir(dir.path().c_str()), SyscallSucceeds()); + EXPECT_THAT(unlink(link.path().c_str()), SyscallSucceeds()); }); } } // namespace -- cgit v1.2.3 From e8f1a5c1f652ba7abb8c4bd842d6afdcab03865a Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Sat, 27 Jun 2020 21:32:16 -0700 Subject: Port GETOWN, SETOWN fcntls to vfs2. Also make some fixes to vfs1's F_SETOWN. The fcntl test now entirely passes on vfs2. Fixes #2920. PiperOrigin-RevId: 318669529 --- pkg/abi/linux/fcntl.go | 2 +- pkg/sentry/kernel/fasync/BUILD | 1 + pkg/sentry/kernel/fasync/fasync.go | 8 +- pkg/sentry/syscalls/linux/sys_file.go | 15 ++-- pkg/sentry/syscalls/linux/vfs2/BUILD | 1 + pkg/sentry/syscalls/linux/vfs2/fd.go | 93 ++++++++++++++++++++++ pkg/sentry/vfs/file_description.go | 74 ++++++++++++++++-- test/syscalls/linux/BUILD | 1 + test/syscalls/linux/fcntl.cc | 140 ++++++++++++++++++++++++++++++---- 9 files changed, 308 insertions(+), 27 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/fcntl.go b/pkg/abi/linux/fcntl.go index 6663a199c..9242e80a5 100644 --- a/pkg/abi/linux/fcntl.go +++ b/pkg/abi/linux/fcntl.go @@ -55,7 +55,7 @@ type Flock struct { _ [4]byte } -// Flags for F_SETOWN_EX and F_GETOWN_EX. +// Owner types for F_SETOWN_EX and F_GETOWN_EX. const ( F_OWNER_TID = 0 F_OWNER_PID = 1 diff --git a/pkg/sentry/kernel/fasync/BUILD b/pkg/sentry/kernel/fasync/BUILD index b9126e946..2b3955598 100644 --- a/pkg/sentry/kernel/fasync/BUILD +++ b/pkg/sentry/kernel/fasync/BUILD @@ -11,6 +11,7 @@ go_library( "//pkg/sentry/fs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", + "//pkg/sentry/vfs", "//pkg/sync", "//pkg/waiter", ], diff --git a/pkg/sentry/kernel/fasync/fasync.go b/pkg/sentry/kernel/fasync/fasync.go index d32c3e90a..323f1dfa5 100644 --- a/pkg/sentry/kernel/fasync/fasync.go +++ b/pkg/sentry/kernel/fasync/fasync.go @@ -20,15 +20,21 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/waiter" ) -// New creates a new FileAsync. +// New creates a new fs.FileAsync. func New() fs.FileAsync { return &FileAsync{} } +// NewVFS2 creates a new vfs.FileAsync. +func NewVFS2() vfs.FileAsync { + return &FileAsync{} +} + // FileAsync sends signals when the registered file is ready for IO. // // +stateify savable diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go index 35eba20c5..2797c6a72 100644 --- a/pkg/sentry/syscalls/linux/sys_file.go +++ b/pkg/sentry/syscalls/linux/sys_file.go @@ -900,14 +900,20 @@ func fGetOwn(t *kernel.Task, file *fs.File) int32 { // // If who is positive, it represents a PID. If negative, it represents a PGID. // If the PID or PGID is invalid, the owner is silently unset. -func fSetOwn(t *kernel.Task, file *fs.File, who int32) { +func fSetOwn(t *kernel.Task, file *fs.File, who int32) error { a := file.Async(fasync.New).(*fasync.FileAsync) if who < 0 { + // Check for overflow before flipping the sign. + if who-1 > who { + return syserror.EINVAL + } pg := t.PIDNamespace().ProcessGroupWithID(kernel.ProcessGroupID(-who)) a.SetOwnerProcessGroup(t, pg) + } else { + tg := t.PIDNamespace().ThreadGroupWithID(kernel.ThreadID(who)) + a.SetOwnerThreadGroup(t, tg) } - tg := t.PIDNamespace().ThreadGroupWithID(kernel.ThreadID(who)) - a.SetOwnerThreadGroup(t, tg) + return nil } // Fcntl implements linux syscall fcntl(2). @@ -1042,8 +1048,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall case linux.F_GETOWN: return uintptr(fGetOwn(t, file)), nil, nil case linux.F_SETOWN: - fSetOwn(t, file, args[2].Int()) - return 0, nil, nil + return 0, nil, fSetOwn(t, file, args[2].Int()) case linux.F_GETOWN_EX: addr := args[2].Pointer() owner := fGetOwnEx(t, file) diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD index c301a0991..0c740335b 100644 --- a/pkg/sentry/syscalls/linux/vfs2/BUILD +++ b/pkg/sentry/syscalls/linux/vfs2/BUILD @@ -54,6 +54,7 @@ go_library( "//pkg/sentry/fsimpl/tmpfs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", + "//pkg/sentry/kernel/fasync", "//pkg/sentry/kernel/pipe", "//pkg/sentry/kernel/time", "//pkg/sentry/limits", diff --git a/pkg/sentry/syscalls/linux/vfs2/fd.go b/pkg/sentry/syscalls/linux/vfs2/fd.go index e68b20bed..7e4c6a56e 100644 --- a/pkg/sentry/syscalls/linux/vfs2/fd.go +++ b/pkg/sentry/syscalls/linux/vfs2/fd.go @@ -20,6 +20,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/fasync" "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" "gvisor.dev/gvisor/pkg/sentry/vfs" @@ -154,6 +155,47 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return 0, nil, err } return uintptr(n), nil, nil + case linux.F_GETOWN: + a := file.AsyncHandler() + if a == nil { + return 0, nil, nil + } + owner := getAsyncOwner(t, a.(*fasync.FileAsync)) + if owner.Type == linux.F_OWNER_PGRP { + return uintptr(-owner.PID), nil, nil + } + return uintptr(owner.PID), nil, nil + case linux.F_SETOWN: + who := args[2].Int() + ownerType := int32(linux.F_OWNER_PID) + if who < 0 { + // Check for overflow before flipping the sign. + if who-1 > who { + return 0, nil, syserror.EINVAL + } + ownerType = linux.F_OWNER_PGRP + who = -who + } + a := file.SetAsyncHandler(fasync.NewVFS2).(*fasync.FileAsync) + return 0, nil, setAsyncOwner(t, a, ownerType, who) + case linux.F_GETOWN_EX: + a := file.AsyncHandler() + if a == nil { + return 0, nil, nil + } + addr := args[2].Pointer() + owner := getAsyncOwner(t, a.(*fasync.FileAsync)) + _, err := t.CopyOut(addr, &owner) + return 0, nil, err + case linux.F_SETOWN_EX: + addr := args[2].Pointer() + var owner linux.FOwnerEx + n, err := t.CopyIn(addr, &owner) + if err != nil { + return 0, nil, err + } + a := file.SetAsyncHandler(fasync.NewVFS2).(*fasync.FileAsync) + return uintptr(n), nil, setAsyncOwner(t, a, owner.Type, owner.PID) case linux.F_GETPIPE_SZ: pipefile, ok := file.Impl().(*pipe.VFSPipeFD) if !ok { @@ -177,6 +219,57 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } } +func getAsyncOwner(t *kernel.Task, a *fasync.FileAsync) linux.FOwnerEx { + ot, otg, opg := a.Owner() + switch { + case ot != nil: + return linux.FOwnerEx{ + Type: linux.F_OWNER_TID, + PID: int32(t.PIDNamespace().IDOfTask(ot)), + } + case otg != nil: + return linux.FOwnerEx{ + Type: linux.F_OWNER_PID, + PID: int32(t.PIDNamespace().IDOfThreadGroup(otg)), + } + case opg != nil: + return linux.FOwnerEx{ + Type: linux.F_OWNER_PGRP, + PID: int32(t.PIDNamespace().IDOfProcessGroup(opg)), + } + default: + return linux.FOwnerEx{} + } +} + +func setAsyncOwner(t *kernel.Task, a *fasync.FileAsync, ownerType, pid int32) error { + switch ownerType { + case linux.F_OWNER_TID: + task := t.PIDNamespace().TaskWithID(kernel.ThreadID(pid)) + if task == nil { + return syserror.ESRCH + } + a.SetOwnerTask(t, task) + return nil + case linux.F_OWNER_PID: + tg := t.PIDNamespace().ThreadGroupWithID(kernel.ThreadID(pid)) + if tg == nil { + return syserror.ESRCH + } + a.SetOwnerThreadGroup(t, tg) + return nil + case linux.F_OWNER_PGRP: + pg := t.PIDNamespace().ProcessGroupWithID(kernel.ProcessGroupID(pid)) + if pg == nil { + return syserror.ESRCH + } + a.SetOwnerProcessGroup(t, pg) + return nil + default: + return syserror.EINVAL + } +} + func posixLock(t *kernel.Task, args arch.SyscallArguments, file *vfs.FileDescription, cmd int32) error { // Copy in the lock request. flockAddr := args[2].Pointer() diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index e0538ea53..cd1db14ac 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -42,11 +42,20 @@ type FileDescription struct { // operations. refs int64 + // flagsMu protects statusFlags and asyncHandler below. + flagsMu sync.Mutex + // statusFlags contains status flags, "initialized by open(2) and possibly - // modified by fcntl()" - fcntl(2). statusFlags is accessed using atomic - // memory operations. + // modified by fcntl()" - fcntl(2). statusFlags can be read using atomic + // memory operations when it does not need to be synchronized with an + // access to asyncHandler. statusFlags uint32 + // asyncHandler handles O_ASYNC signal generation. It is set with the + // F_SETOWN or F_SETOWN_EX fcntls. For asyncHandler to be used, O_ASYNC must + // also be set by fcntl(2). + asyncHandler FileAsync + // epolls is the set of epollInterests registered for this FileDescription. // epolls is protected by epollMu. epollMu sync.Mutex @@ -193,6 +202,13 @@ func (fd *FileDescription) DecRef() { fd.vd.mount.EndWrite() } fd.vd.DecRef() + fd.flagsMu.Lock() + // TODO(gvisor.dev/issue/1663): We may need to unregister during save, as we do in VFS1. + if fd.statusFlags&linux.O_ASYNC != 0 && fd.asyncHandler != nil { + fd.asyncHandler.Unregister(fd) + } + fd.asyncHandler = nil + fd.flagsMu.Unlock() } else if refs < 0 { panic("FileDescription.DecRef() called without holding a reference") } @@ -276,7 +292,18 @@ func (fd *FileDescription) SetStatusFlags(ctx context.Context, creds *auth.Crede } // TODO(jamieliu): FileDescriptionImpl.SetOAsync()? const settableFlags = linux.O_APPEND | linux.O_ASYNC | linux.O_DIRECT | linux.O_NOATIME | linux.O_NONBLOCK - atomic.StoreUint32(&fd.statusFlags, (oldFlags&^settableFlags)|(flags&settableFlags)) + fd.flagsMu.Lock() + if fd.asyncHandler != nil { + // Use fd.statusFlags instead of oldFlags, which may have become outdated, + // to avoid double registering/unregistering. + if fd.statusFlags&linux.O_ASYNC == 0 && flags&linux.O_ASYNC != 0 { + fd.asyncHandler.Register(fd) + } else if fd.statusFlags&linux.O_ASYNC != 0 && flags&linux.O_ASYNC == 0 { + fd.asyncHandler.Unregister(fd) + } + } + fd.statusFlags = (oldFlags &^ settableFlags) | (flags & settableFlags) + fd.flagsMu.Unlock() return nil } @@ -533,17 +560,23 @@ func (fd *FileDescription) StatFS(ctx context.Context) (linux.Statfs, error) { return fd.impl.StatFS(ctx) } -// Readiness returns fd's I/O readiness. +// Readiness implements waiter.Waitable.Readiness. +// +// It returns fd's I/O readiness. func (fd *FileDescription) Readiness(mask waiter.EventMask) waiter.EventMask { return fd.impl.Readiness(mask) } -// EventRegister registers e for I/O readiness events in mask. +// EventRegister implements waiter.Waitable.EventRegister. +// +// It registers e for I/O readiness events in mask. func (fd *FileDescription) EventRegister(e *waiter.Entry, mask waiter.EventMask) { fd.impl.EventRegister(e, mask) } -// EventUnregister unregisters e for I/O readiness events. +// EventUnregister implements waiter.Waitable.EventUnregister. +// +// It unregisters e for I/O readiness events. func (fd *FileDescription) EventUnregister(e *waiter.Entry) { fd.impl.EventUnregister(e) } @@ -770,3 +803,32 @@ func (fd *FileDescription) LockPOSIX(ctx context.Context, uid lock.UniqueID, t l func (fd *FileDescription) UnlockPOSIX(ctx context.Context, uid lock.UniqueID, start, end uint64, whence int16) error { return fd.impl.UnlockPOSIX(ctx, uid, start, end, whence) } + +// A FileAsync sends signals to its owner when w is ready for IO. This is only +// implemented by pkg/sentry/fasync:FileAsync, but we unfortunately need this +// interface to avoid circular dependencies. +type FileAsync interface { + Register(w waiter.Waitable) + Unregister(w waiter.Waitable) +} + +// AsyncHandler returns the FileAsync for fd. +func (fd *FileDescription) AsyncHandler() FileAsync { + fd.flagsMu.Lock() + defer fd.flagsMu.Unlock() + return fd.asyncHandler +} + +// SetAsyncHandler sets fd.asyncHandler if it has not been set before and +// returns it. +func (fd *FileDescription) SetAsyncHandler(newHandler func() FileAsync) FileAsync { + fd.flagsMu.Lock() + defer fd.flagsMu.Unlock() + if fd.asyncHandler == nil { + fd.asyncHandler = newHandler() + if fd.statusFlags&linux.O_ASYNC != 0 { + fd.asyncHandler.Register(fd) + } + } + return fd.asyncHandler +} diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 270b9e4c4..e141a86bb 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -805,6 +805,7 @@ cc_binary( "//test/util:save_util", "//test/util:temp_path", "//test/util:test_util", + "//test/util:thread_util", "//test/util:timer_util", ], ) diff --git a/test/syscalls/linux/fcntl.cc b/test/syscalls/linux/fcntl.cc index 25bef2522..9130618fa 100644 --- a/test/syscalls/linux/fcntl.cc +++ b/test/syscalls/linux/fcntl.cc @@ -37,6 +37,7 @@ #include "test/util/save_util.h" #include "test/util/temp_path.h" #include "test/util/test_util.h" +#include "test/util/thread_util.h" #include "test/util/timer_util.h" ABSL_FLAG(std::string, child_setlock_on, "", @@ -953,15 +954,18 @@ TEST(FcntlTest, DupAfterO_ASYNC) { EXPECT_EQ(after & O_ASYNC, O_ASYNC); } -TEST(FcntlTest, GetOwn) { +TEST(FcntlTest, GetOwnNone) { FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); - EXPECT_EQ(syscall(__NR_fcntl, s.get(), F_GETOWN), 0); + // Use the raw syscall because the glibc wrapper may convert F_{GET,SET}OWN + // into F_{GET,SET}OWN_EX. + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN), + SyscallSucceedsWithValue(0)); MaybeSave(); } -TEST(FcntlTest, GetOwnEx) { +TEST(FcntlTest, GetOwnExNone) { FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); @@ -970,6 +974,70 @@ TEST(FcntlTest, GetOwnEx) { SyscallSucceedsWithValue(0)); } +TEST(FcntlTest, SetOwnInvalidPid) { + SKIP_IF(IsRunningWithVFS1()); + + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, 12345678), + SyscallFailsWithErrno(ESRCH)); +} + +TEST(FcntlTest, SetOwnInvalidPgrp) { + SKIP_IF(IsRunningWithVFS1()); + + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, -12345678), + SyscallFailsWithErrno(ESRCH)); +} + +TEST(FcntlTest, SetOwnPid) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + pid_t pid; + EXPECT_THAT(pid = getpid(), SyscallSucceeds()); + + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, pid), SyscallSucceeds()); + + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN), + SyscallSucceedsWithValue(pid)); + MaybeSave(); +} + +TEST(FcntlTest, SetOwnPgrp) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + pid_t pgid; + EXPECT_THAT(pgid = getpgrp(), SyscallSucceeds()); + + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, -pgid), SyscallSucceeds()); + + // Verify with F_GETOWN_EX; using F_GETOWN on Linux may incorrectly treat the + // negative return value as an error, converting the return value to -1 and + // setting errno accordingly. + f_owner_ex got_owner = {}; + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN_EX, &got_owner), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(got_owner.type, F_OWNER_PGRP); + EXPECT_EQ(got_owner.pid, pgid); + MaybeSave(); +} + +// F_SETOWN flips the sign of negative values, an operation that is guarded +// against overflow. +TEST(FcntlTest, SetOwnOverflow) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, INT_MIN), + SyscallFailsWithErrno(EINVAL)); +} + TEST(FcntlTest, SetOwnExInvalidType) { FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); @@ -1027,7 +1095,8 @@ TEST(FcntlTest, SetOwnExTid) { ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), SyscallSucceeds()); - EXPECT_EQ(syscall(__NR_fcntl, s.get(), F_GETOWN), owner.pid); + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN), + SyscallSucceedsWithValue(owner.pid)); MaybeSave(); } @@ -1042,7 +1111,8 @@ TEST(FcntlTest, SetOwnExPid) { ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), SyscallSucceeds()); - EXPECT_EQ(syscall(__NR_fcntl, s.get(), F_GETOWN), owner.pid); + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN), + SyscallSucceedsWithValue(owner.pid)); MaybeSave(); } @@ -1050,18 +1120,21 @@ TEST(FcntlTest, SetOwnExPgrp) { FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); - f_owner_ex owner = {}; - owner.type = F_OWNER_PGRP; - EXPECT_THAT(owner.pid = getpgrp(), SyscallSucceeds()); + f_owner_ex set_owner = {}; + set_owner.type = F_OWNER_PGRP; + EXPECT_THAT(set_owner.pid = getpgrp(), SyscallSucceeds()); - ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &set_owner), SyscallSucceeds()); - // NOTE(igudger): I don't understand why, but this is flaky on Linux. - // GetOwnExPgrp (below) does not have this issue. - SKIP_IF(!IsRunningOnGvisor()); - - EXPECT_EQ(syscall(__NR_fcntl, s.get(), F_GETOWN), -owner.pid); + // Verify with F_GETOWN_EX; using F_GETOWN on Linux may incorrectly treat the + // negative return value as an error, converting the return value to -1 and + // setting errno accordingly. + f_owner_ex got_owner = {}; + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN_EX, &got_owner), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(got_owner.type, set_owner.type); + EXPECT_EQ(got_owner.pid, set_owner.pid); MaybeSave(); } @@ -1119,6 +1192,45 @@ TEST(FcntlTest, GetOwnExPgrp) { EXPECT_EQ(got_owner.pid, set_owner.pid); } +// Make sure that making multiple concurrent changes to async signal generation +// does not cause any race issues. +TEST(FcntlTest, SetFlSetOwnDoNotRace) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + pid_t pid; + EXPECT_THAT(pid = getpid(), SyscallSucceeds()); + + constexpr absl::Duration runtime = absl::Milliseconds(300); + auto setAsync = [&s, &runtime] { + for (auto start = absl::Now(); absl::Now() - start < runtime;) { + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETFL, O_ASYNC), + SyscallSucceeds()); + sched_yield(); + } + }; + auto resetAsync = [&s, &runtime] { + for (auto start = absl::Now(); absl::Now() - start < runtime;) { + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETFL, 0), SyscallSucceeds()); + sched_yield(); + } + }; + auto setOwn = [&s, &pid, &runtime] { + for (auto start = absl::Now(); absl::Now() - start < runtime;) { + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, pid), + SyscallSucceeds()); + sched_yield(); + } + }; + + std::list threads; + for (int i = 0; i < 10; i++) { + threads.emplace_back(setAsync); + threads.emplace_back(resetAsync); + threads.emplace_back(setOwn); + } +} + } // namespace } // namespace testing -- cgit v1.2.3 From c4bdd0118f5dc9090979697e31a18731968b4066 Mon Sep 17 00:00:00 2001 From: Bhasker Hariharan Date: Tue, 30 Jun 2020 19:00:13 -0700 Subject: Add missing newline in /sys/devices/systen/cpu/onine PiperOrigin-RevId: 319143410 --- pkg/sentry/fsimpl/sys/sys.go | 2 +- pkg/sentry/fsimpl/sys/sys_test.go | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go index fe02f7ee9..01ce30a4d 100644 --- a/pkg/sentry/fsimpl/sys/sys.go +++ b/pkg/sentry/fsimpl/sys/sys.go @@ -138,7 +138,7 @@ type cpuFile struct { // Generate implements vfs.DynamicBytesSource.Generate. func (c *cpuFile) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "0-%d", c.maxCores-1) + fmt.Fprintf(buf, "0-%d\n", c.maxCores-1) return nil } diff --git a/pkg/sentry/fsimpl/sys/sys_test.go b/pkg/sentry/fsimpl/sys/sys_test.go index 4b3602d47..242d5fd12 100644 --- a/pkg/sentry/fsimpl/sys/sys_test.go +++ b/pkg/sentry/fsimpl/sys/sys_test.go @@ -51,7 +51,7 @@ func TestReadCPUFile(t *testing.T) { k := kernel.KernelFromContext(s.Ctx) maxCPUCores := k.ApplicationCores() - expected := fmt.Sprintf("0-%d", maxCPUCores-1) + expected := fmt.Sprintf("0-%d\n", maxCPUCores-1) for _, fname := range []string{"online", "possible", "present"} { pop := s.PathOpAtRoot(fmt.Sprintf("devices/system/cpu/%s", fname)) -- cgit v1.2.3 From 20d571b0c181023cc02521ad746a2b6d91e6794d Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 30 Jun 2020 20:47:02 -0700 Subject: Allow O_DIRECT on vfs2 tmpfs files. Updates #2923. PiperOrigin-RevId: 319153792 --- pkg/sentry/fsimpl/tmpfs/filesystem.go | 4 ++-- pkg/sentry/vfs/file_description.go | 3 +-- 2 files changed, 3 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 71ac7b8e6..ce0dbb154 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -407,7 +407,7 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.Open case *regularFile: var fd regularFileFD fd.LockFD.Init(&d.inode.locks) - if err := fd.vfsfd.Init(&fd, opts.Flags, rp.Mount(), &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { + if err := fd.vfsfd.Init(&fd, opts.Flags, rp.Mount(), &d.vfsd, &vfs.FileDescriptionOptions{AllowDirectIO: true}); err != nil { return nil, err } if opts.Flags&linux.O_TRUNC != 0 { @@ -423,7 +423,7 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.Open } var fd directoryFD fd.LockFD.Init(&d.inode.locks) - if err := fd.vfsfd.Init(&fd, opts.Flags, rp.Mount(), &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { + if err := fd.vfsfd.Init(&fd, opts.Flags, rp.Mount(), &d.vfsd, &vfs.FileDescriptionOptions{AllowDirectIO: true}); err != nil { return nil, err } return &fd.vfsfd, nil diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index cd1db14ac..eb5dfd7e2 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -91,8 +91,7 @@ type FileDescription struct { // FileDescriptionOptions contains options to FileDescription.Init(). type FileDescriptionOptions struct { - // If AllowDirectIO is true, allow O_DIRECT to be set on the file. This is - // usually only the case if O_DIRECT would actually have an effect. + // If AllowDirectIO is true, allow O_DIRECT to be set on the file. AllowDirectIO bool // If DenyPRead is true, calls to FileDescription.PRead() return ESPIPE. -- cgit v1.2.3 From 43f5dd95a1c58a6e3260c31093bfc3f97885f4b0 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 30 Jun 2020 20:59:32 -0700 Subject: Fix index calculation for /proc/[pid]/cmdline. We were truncating buf using a index relative to the middle of the slice (i.e. where envv begins), but we need to calculate the index relative to the entire slice. Updates #2923. PiperOrigin-RevId: 319154950 --- pkg/sentry/fsimpl/proc/task_files.go | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index 286c23f01..9af43b859 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -231,8 +231,9 @@ func (d *cmdlineData) Generate(ctx context.Context, buf *bytes.Buffer) error { // Linux will return envp up to and including the first NULL character, // so find it. - if end := bytes.IndexByte(buf.Bytes()[ar.Length():], 0); end != -1 { - buf.Truncate(end) + envStart := int(ar.Length()) + if nullIdx := bytes.IndexByte(buf.Bytes()[envStart:], 0); nullIdx != -1 { + buf.Truncate(envStart + nullIdx) } } @@ -300,7 +301,7 @@ type idMapData struct { var _ dynamicInode = (*idMapData)(nil) -// Generate implements vfs.DynamicBytesSource.Generate. +// Generate implements vfs.WritableDynamicBytesSource.Generate. func (d *idMapData) Generate(ctx context.Context, buf *bytes.Buffer) error { var entries []auth.IDMapEntry if d.gids { @@ -314,6 +315,7 @@ func (d *idMapData) Generate(ctx context.Context, buf *bytes.Buffer) error { return nil } +// Write implements vfs.WritableDynamicBytesSource.Write. func (d *idMapData) Write(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) { // "In addition, the number of bytes written to the file must be less than // the system page size, and the write must be performed at the start of -- cgit v1.2.3 From cda2979b63fad37a33706f8aa430664a9c4d0b3b Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Wed, 1 Jul 2020 08:40:31 -0700 Subject: Complete async signal delivery support in vfs2. - Support FIOASYNC, FIO{SET,GET}OWN, SIOC{G,S}PGRP (refactor getting/setting owner in the process). - Unset signal recipient when setting owner with pid == 0 and valid owner type. Updates #2923. PiperOrigin-RevId: 319231420 --- pkg/sentry/kernel/fasync/fasync.go | 10 ++++++ pkg/sentry/syscalls/linux/vfs2/fd.go | 54 +++++++++++++++++++------------ pkg/sentry/syscalls/linux/vfs2/ioctl.go | 43 +++++++++++++++++++++++++ test/syscalls/BUILD | 1 + test/syscalls/linux/fcntl.cc | 57 +++++++++++++++++++++++++++++++++ 5 files changed, 144 insertions(+), 21 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/fasync/fasync.go b/pkg/sentry/kernel/fasync/fasync.go index 323f1dfa5..153d2cd9b 100644 --- a/pkg/sentry/kernel/fasync/fasync.go +++ b/pkg/sentry/kernel/fasync/fasync.go @@ -176,3 +176,13 @@ func (a *FileAsync) SetOwnerProcessGroup(requester *kernel.Task, recipient *kern a.recipientTG = nil a.recipientPG = recipient } + +// ClearOwner unsets the current signal recipient. +func (a *FileAsync) ClearOwner() { + a.mu.Lock() + defer a.mu.Unlock() + a.requester = nil + a.recipientT = nil + a.recipientTG = nil + a.recipientPG = nil +} diff --git a/pkg/sentry/syscalls/linux/vfs2/fd.go b/pkg/sentry/syscalls/linux/vfs2/fd.go index 7e4c6a56e..517394ba9 100644 --- a/pkg/sentry/syscalls/linux/vfs2/fd.go +++ b/pkg/sentry/syscalls/linux/vfs2/fd.go @@ -156,11 +156,10 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } return uintptr(n), nil, nil case linux.F_GETOWN: - a := file.AsyncHandler() - if a == nil { + owner, hasOwner := getAsyncOwner(t, file) + if !hasOwner { return 0, nil, nil } - owner := getAsyncOwner(t, a.(*fasync.FileAsync)) if owner.Type == linux.F_OWNER_PGRP { return uintptr(-owner.PID), nil, nil } @@ -176,26 +175,21 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall ownerType = linux.F_OWNER_PGRP who = -who } - a := file.SetAsyncHandler(fasync.NewVFS2).(*fasync.FileAsync) - return 0, nil, setAsyncOwner(t, a, ownerType, who) + return 0, nil, setAsyncOwner(t, file, ownerType, who) case linux.F_GETOWN_EX: - a := file.AsyncHandler() - if a == nil { + owner, hasOwner := getAsyncOwner(t, file) + if !hasOwner { return 0, nil, nil } - addr := args[2].Pointer() - owner := getAsyncOwner(t, a.(*fasync.FileAsync)) - _, err := t.CopyOut(addr, &owner) + _, err := t.CopyOut(args[2].Pointer(), &owner) return 0, nil, err case linux.F_SETOWN_EX: - addr := args[2].Pointer() var owner linux.FOwnerEx - n, err := t.CopyIn(addr, &owner) + n, err := t.CopyIn(args[2].Pointer(), &owner) if err != nil { return 0, nil, err } - a := file.SetAsyncHandler(fasync.NewVFS2).(*fasync.FileAsync) - return uintptr(n), nil, setAsyncOwner(t, a, owner.Type, owner.PID) + return uintptr(n), nil, setAsyncOwner(t, file, owner.Type, owner.PID) case linux.F_GETPIPE_SZ: pipefile, ok := file.Impl().(*pipe.VFSPipeFD) if !ok { @@ -219,30 +213,48 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } } -func getAsyncOwner(t *kernel.Task, a *fasync.FileAsync) linux.FOwnerEx { - ot, otg, opg := a.Owner() +func getAsyncOwner(t *kernel.Task, fd *vfs.FileDescription) (ownerEx linux.FOwnerEx, hasOwner bool) { + a := fd.AsyncHandler() + if a == nil { + return linux.FOwnerEx{}, false + } + + ot, otg, opg := a.(*fasync.FileAsync).Owner() switch { case ot != nil: return linux.FOwnerEx{ Type: linux.F_OWNER_TID, PID: int32(t.PIDNamespace().IDOfTask(ot)), - } + }, true case otg != nil: return linux.FOwnerEx{ Type: linux.F_OWNER_PID, PID: int32(t.PIDNamespace().IDOfThreadGroup(otg)), - } + }, true case opg != nil: return linux.FOwnerEx{ Type: linux.F_OWNER_PGRP, PID: int32(t.PIDNamespace().IDOfProcessGroup(opg)), - } + }, true default: - return linux.FOwnerEx{} + return linux.FOwnerEx{}, true } } -func setAsyncOwner(t *kernel.Task, a *fasync.FileAsync, ownerType, pid int32) error { +func setAsyncOwner(t *kernel.Task, fd *vfs.FileDescription, ownerType, pid int32) error { + switch ownerType { + case linux.F_OWNER_TID, linux.F_OWNER_PID, linux.F_OWNER_PGRP: + // Acceptable type. + default: + return syserror.EINVAL + } + + a := fd.SetAsyncHandler(fasync.NewVFS2).(*fasync.FileAsync) + if pid == 0 { + a.ClearOwner() + return nil + } + switch ownerType { case linux.F_OWNER_TID: task := t.PIDNamespace().TaskWithID(kernel.ThreadID(pid)) diff --git a/pkg/sentry/syscalls/linux/vfs2/ioctl.go b/pkg/sentry/syscalls/linux/vfs2/ioctl.go index 0399c0db4..fd6ab94b2 100644 --- a/pkg/sentry/syscalls/linux/vfs2/ioctl.go +++ b/pkg/sentry/syscalls/linux/vfs2/ioctl.go @@ -57,6 +57,49 @@ func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall flags &^= linux.O_NONBLOCK } return 0, nil, file.SetStatusFlags(t, t.Credentials(), flags) + + case linux.FIOASYNC: + var set int32 + if _, err := t.CopyIn(args[2].Pointer(), &set); err != nil { + return 0, nil, err + } + flags := file.StatusFlags() + if set != 0 { + flags |= linux.O_ASYNC + } else { + flags &^= linux.O_ASYNC + } + file.SetStatusFlags(t, t.Credentials(), flags) + return 0, nil, nil + + case linux.FIOGETOWN, linux.SIOCGPGRP: + var who int32 + owner, hasOwner := getAsyncOwner(t, file) + if hasOwner { + if owner.Type == linux.F_OWNER_PGRP { + who = -owner.PID + } else { + who = owner.PID + } + } + _, err := t.CopyOut(args[2].Pointer(), &who) + return 0, nil, err + + case linux.FIOSETOWN, linux.SIOCSPGRP: + var who int32 + if _, err := t.CopyIn(args[2].Pointer(), &who); err != nil { + return 0, nil, err + } + ownerType := int32(linux.F_OWNER_PID) + if who < 0 { + // Check for overflow before flipping the sign. + if who-1 > who { + return 0, nil, syserror.EINVAL + } + ownerType = linux.F_OWNER_PGRP + who = -who + } + return 0, nil, setAsyncOwner(t, file, ownerType, who) } ret, err := file.Ioctl(t, t.MemoryManager(), args) diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 36c178e4a..88ed36b69 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -288,6 +288,7 @@ syscall_test( size = "medium", add_overlay = True, test = "//test/syscalls/linux:ioctl_test", + vfs2 = "True", ) syscall_test( diff --git a/test/syscalls/linux/fcntl.cc b/test/syscalls/linux/fcntl.cc index 4b9b4db99..5467fa2c8 100644 --- a/test/syscalls/linux/fcntl.cc +++ b/test/syscalls/linux/fcntl.cc @@ -1031,6 +1031,30 @@ TEST(FcntlTest, SetOwnPgrp) { MaybeSave(); } +TEST(FcntlTest, SetOwnUnset) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + // Set and unset pid. + pid_t pid; + EXPECT_THAT(pid = getpid(), SyscallSucceeds()); + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, pid), SyscallSucceeds()); + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, 0), SyscallSucceeds()); + + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN), + SyscallSucceedsWithValue(0)); + + // Set and unset pgid. + pid_t pgid; + EXPECT_THAT(pgid = getpgrp(), SyscallSucceeds()); + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, -pgid), SyscallSucceeds()); + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, 0), SyscallSucceeds()); + + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN), + SyscallSucceedsWithValue(0)); + MaybeSave(); +} + // F_SETOWN flips the sign of negative values, an operation that is guarded // against overflow. TEST(FcntlTest, SetOwnOverflow) { @@ -1141,6 +1165,39 @@ TEST(FcntlTest, SetOwnExPgrp) { MaybeSave(); } +TEST(FcntlTest, SetOwnExUnset) { + SKIP_IF(IsRunningWithVFS1()); + + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + // Set and unset pid. + f_owner_ex owner = {}; + owner.type = F_OWNER_PID; + EXPECT_THAT(owner.pid = getpid(), SyscallSucceeds()); + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), + SyscallSucceeds()); + owner.pid = 0; + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), + SyscallSucceeds()); + + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN), + SyscallSucceedsWithValue(0)); + + // Set and unset pgid. + owner.type = F_OWNER_PGRP; + EXPECT_THAT(owner.pid = getpgrp(), SyscallSucceeds()); + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), + SyscallSucceeds()); + owner.pid = 0; + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), + SyscallSucceeds()); + + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN), + SyscallSucceedsWithValue(0)); + MaybeSave(); +} + TEST(FcntlTest, GetOwnExTid) { FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); -- cgit v1.2.3 From 6a90c88b97481a6d81b05f09d5c8ed7158225dd5 Mon Sep 17 00:00:00 2001 From: Zach Koopmans Date: Wed, 1 Jul 2020 13:13:04 -0700 Subject: Port fallocate to VFS2. PiperOrigin-RevId: 319283715 --- pkg/p9/p9.go | 13 ++++++++ pkg/sentry/fsimpl/gofer/regular_file.go | 29 ++++++++++++++++ pkg/sentry/fsimpl/host/host.go | 10 ++++++ pkg/sentry/fsimpl/kernfs/fd_impl_util.go | 5 +++ pkg/sentry/fsimpl/tmpfs/regular_file.go | 15 +++++++++ pkg/sentry/kernel/pipe/vfs.go | 5 +++ pkg/sentry/socket/hostinet/socket_vfs2.go | 7 +++- pkg/sentry/syscalls/linux/vfs2/filesystem.go | 50 ++++++++++++++++++++++++++++ pkg/sentry/syscalls/linux/vfs2/vfs2.go | 2 +- pkg/sentry/vfs/file_description.go | 4 +++ pkg/sentry/vfs/file_description_impl_util.go | 11 ++++++ pkg/sentry/vfs/inotify.go | 5 +++ test/syscalls/BUILD | 1 + test/syscalls/linux/BUILD | 5 +++ test/syscalls/linux/fallocate.cc | 45 +++++++++++++++++++++++++ 15 files changed, 205 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/p9/p9.go b/pkg/p9/p9.go index 28d851ff5..122c457d2 100644 --- a/pkg/p9/p9.go +++ b/pkg/p9/p9.go @@ -1091,6 +1091,19 @@ type AllocateMode struct { Unshare bool } +// ToAllocateMode returns an AllocateMode from a fallocate(2) mode. +func ToAllocateMode(mode uint64) AllocateMode { + return AllocateMode{ + KeepSize: mode&unix.FALLOC_FL_KEEP_SIZE != 0, + PunchHole: mode&unix.FALLOC_FL_PUNCH_HOLE != 0, + NoHideStale: mode&unix.FALLOC_FL_NO_HIDE_STALE != 0, + CollapseRange: mode&unix.FALLOC_FL_COLLAPSE_RANGE != 0, + ZeroRange: mode&unix.FALLOC_FL_ZERO_RANGE != 0, + InsertRange: mode&unix.FALLOC_FL_INSERT_RANGE != 0, + Unshare: mode&unix.FALLOC_FL_UNSHARE_RANGE != 0, + } +} + // ToLinux converts to a value compatible with fallocate(2)'s mode. func (a *AllocateMode) ToLinux() uint32 { rv := uint32(0) diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go index 404a452c4..faba73af2 100644 --- a/pkg/sentry/fsimpl/gofer/regular_file.go +++ b/pkg/sentry/fsimpl/gofer/regular_file.go @@ -24,6 +24,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/memmap" @@ -67,6 +68,34 @@ func (fd *regularFileFD) OnClose(ctx context.Context) error { return d.handle.file.flush(ctx) } +// Allocate implements vfs.FileDescriptionImpl.Allocate. +func (fd *regularFileFD) Allocate(ctx context.Context, mode, offset, length uint64) error { + + d := fd.dentry() + d.metadataMu.Lock() + defer d.metadataMu.Unlock() + + size := offset + length + + // Allocating a smaller size is a noop. + if size <= d.size { + return nil + } + + d.handleMu.Lock() + defer d.handleMu.Unlock() + + err := d.handle.file.allocate(ctx, p9.ToAllocateMode(mode), offset, length) + if err != nil { + return err + } + d.size = size + if !d.cachedMetadataAuthoritative() { + d.touchCMtimeLocked() + } + return nil +} + // PRead implements vfs.FileDescriptionImpl.PRead. func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { if offset < 0 { diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index d5e73ddae..007b3332e 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -543,6 +543,16 @@ func (f *fileDescription) Release() { // noop } +// Allocate implements vfs.FileDescriptionImpl. +func (f *fileDescription) Allocate(ctx context.Context, mode, offset, length uint64) error { + if !f.inode.seekable { + return syserror.ESPIPE + } + + // TODO(gvisor.dev/issue/2923): Implement Allocate for non-pipe hostfds. + return syserror.EOPNOTSUPP +} + // PRead implements FileDescriptionImpl. func (f *fileDescription) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { i := f.inode diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index 5f7853a2a..ca8b8c63b 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -236,6 +236,11 @@ func (fd *GenericDirectoryFD) SetStat(ctx context.Context, opts vfs.SetStatOptio return inode.SetStat(ctx, fd.filesystem(), creds, opts) } +// Allocate implements vfs.FileDescriptionImpl.Allocate. +func (fd *GenericDirectoryFD) Allocate(ctx context.Context, mode, offset, length uint64) error { + return fd.DirectoryFileDescriptionDefaultImpl.Allocate(ctx, mode, offset, length) +} + // LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. func (fd *GenericDirectoryFD) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { return fd.Locks().LockPOSIX(ctx, &fd.vfsfd, uid, t, start, length, whence, block) diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index b805aadd0..6691add96 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -274,6 +274,21 @@ func (fd *regularFileFD) Release() { // noop } +// Allocate implements vfs.FileDescriptionImpl.Allocate. +func (fd *regularFileFD) Allocate(ctx context.Context, mode, offset, length uint64) error { + f := fd.inode().impl.(*regularFile) + + f.inode.mu.Lock() + defer f.inode.mu.Unlock() + oldSize := f.size + size := offset + length + if oldSize >= size { + return nil + } + _, err := f.truncateLocked(size) + return err +} + // PRead implements vfs.FileDescriptionImpl.PRead. func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { if offset < 0 { diff --git a/pkg/sentry/kernel/pipe/vfs.go b/pkg/sentry/kernel/pipe/vfs.go index a4519363f..45d4c5fc1 100644 --- a/pkg/sentry/kernel/pipe/vfs.go +++ b/pkg/sentry/kernel/pipe/vfs.go @@ -200,6 +200,11 @@ func (fd *VFSPipeFD) Readiness(mask waiter.EventMask) waiter.EventMask { } } +// Allocate implements vfs.FileDescriptionImpl.Allocate. +func (fd *VFSPipeFD) Allocate(ctx context.Context, mode, offset, length uint64) error { + return syserror.ESPIPE +} + // EventRegister implements waiter.Waitable.EventRegister. func (fd *VFSPipeFD) EventRegister(e *waiter.Entry, mask waiter.EventMask) { fd.pipe.EventRegister(e, mask) diff --git a/pkg/sentry/socket/hostinet/socket_vfs2.go b/pkg/sentry/socket/hostinet/socket_vfs2.go index ad5f64799..8f192c62f 100644 --- a/pkg/sentry/socket/hostinet/socket_vfs2.go +++ b/pkg/sentry/socket/hostinet/socket_vfs2.go @@ -96,7 +96,12 @@ func (s *socketVFS2) Ioctl(ctx context.Context, uio usermem.IO, args arch.Syscal return ioctl(ctx, s.fd, uio, args) } -// PRead implements vfs.FileDescriptionImpl. +// Allocate implements vfs.FileDescriptionImpl.Allocate. +func (s *socketVFS2) Allocate(ctx context.Context, mode, offset, length uint64) error { + return syserror.ENODEV +} + +// PRead implements vfs.FileDescriptionImpl.PRead. func (s *socketVFS2) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { return 0, syserror.ESPIPE } diff --git a/pkg/sentry/syscalls/linux/vfs2/filesystem.go b/pkg/sentry/syscalls/linux/vfs2/filesystem.go index 5dac77e4d..b12b5967b 100644 --- a/pkg/sentry/syscalls/linux/vfs2/filesystem.go +++ b/pkg/sentry/syscalls/linux/vfs2/filesystem.go @@ -18,6 +18,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -239,6 +240,55 @@ func renameat(t *kernel.Task, olddirfd int32, oldpathAddr usermem.Addr, newdirfd }) } +// Fallocate implements linux system call fallocate(2). +func Fallocate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + mode := args[1].Uint64() + offset := args[2].Int64() + length := args[3].Int64() + + file := t.GetFileVFS2(fd) + + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + if !file.IsWritable() { + return 0, nil, syserror.EBADF + } + + if mode != 0 { + return 0, nil, syserror.ENOTSUP + } + + if offset < 0 || length <= 0 { + return 0, nil, syserror.EINVAL + } + + size := offset + length + + if size < 0 { + return 0, nil, syserror.EFBIG + } + + limit := limits.FromContext(t).Get(limits.FileSize).Cur + + if uint64(size) >= limit { + t.SendSignal(&arch.SignalInfo{ + Signo: int32(linux.SIGXFSZ), + Code: arch.SignalInfoUser, + }) + return 0, nil, syserror.EFBIG + } + + return 0, nil, file.Impl().Allocate(t, mode, uint64(offset), uint64(length)) + + // File length modified, generate notification. + // TODO(gvisor.dev/issue/1479): Reenable when Inotify is ported. + // file.Dirent.InotifyEvent(linux.IN_MODIFY, 0) +} + // Rmdir implements Linux syscall rmdir(2). func Rmdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { pathAddr := args[0].Pointer() diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2.go b/pkg/sentry/syscalls/linux/vfs2/vfs2.go index 9e60c4a1c..8f497ecc7 100644 --- a/pkg/sentry/syscalls/linux/vfs2/vfs2.go +++ b/pkg/sentry/syscalls/linux/vfs2/vfs2.go @@ -138,7 +138,7 @@ func Override() { s.Table[282] = syscalls.Supported("signalfd", Signalfd) s.Table[283] = syscalls.Supported("timerfd_create", TimerfdCreate) s.Table[284] = syscalls.Supported("eventfd", Eventfd) - delete(s.Table, 285) // fallocate + s.Table[285] = syscalls.PartiallySupported("fallocate", Fallocate, "Not all options are supported.", nil) s.Table[286] = syscalls.Supported("timerfd_settime", TimerfdSettime) s.Table[287] = syscalls.Supported("timerfd_gettime", TimerfdGettime) s.Table[288] = syscalls.Supported("accept4", Accept4) diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index eb5dfd7e2..0c42574db 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -354,6 +354,10 @@ type FileDescriptionImpl interface { // represented by the FileDescription. StatFS(ctx context.Context) (linux.Statfs, error) + // Allocate grows file represented by FileDescription to offset + length bytes. + // Only mode == 0 is supported currently. + Allocate(ctx context.Context, mode, offset, length uint64) error + // waiter.Waitable methods may be used to poll for I/O events. waiter.Waitable diff --git a/pkg/sentry/vfs/file_description_impl_util.go b/pkg/sentry/vfs/file_description_impl_util.go index 3fec0d6d6..6b8b4ad49 100644 --- a/pkg/sentry/vfs/file_description_impl_util.go +++ b/pkg/sentry/vfs/file_description_impl_util.go @@ -56,6 +56,12 @@ func (FileDescriptionDefaultImpl) StatFS(ctx context.Context) (linux.Statfs, err return linux.Statfs{}, syserror.ENOSYS } +// Allocate implements FileDescriptionImpl.Allocate analogously to +// fallocate called on regular file, directory or FIFO in Linux. +func (FileDescriptionDefaultImpl) Allocate(ctx context.Context, mode, offset, length uint64) error { + return syserror.ENODEV +} + // Readiness implements waiter.Waitable.Readiness analogously to // file_operations::poll == NULL in Linux. func (FileDescriptionDefaultImpl) Readiness(mask waiter.EventMask) waiter.EventMask { @@ -158,6 +164,11 @@ func (FileDescriptionDefaultImpl) Removexattr(ctx context.Context, name string) // implementations of non-directory I/O methods that return EISDIR. type DirectoryFileDescriptionDefaultImpl struct{} +// Allocate implements DirectoryFileDescriptionDefaultImpl.Allocate. +func (DirectoryFileDescriptionDefaultImpl) Allocate(ctx context.Context, mode, offset, length uint64) error { + return syserror.EISDIR +} + // PRead implements FileDescriptionImpl.PRead. func (DirectoryFileDescriptionDefaultImpl) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts ReadOptions) (int64, error) { return 0, syserror.EISDIR diff --git a/pkg/sentry/vfs/inotify.go b/pkg/sentry/vfs/inotify.go index 509034531..c2e21ac5f 100644 --- a/pkg/sentry/vfs/inotify.go +++ b/pkg/sentry/vfs/inotify.go @@ -148,6 +148,11 @@ func (i *Inotify) Release() { } } +// Allocate implements FileDescription.Allocate. +func (i *Inotify) Allocate(ctx context.Context, mode, offset, length uint64) error { + panic("Allocate should not be called on read-only inotify fds") +} + // EventRegister implements waiter.Waitable. func (i *Inotify) EventRegister(e *waiter.Entry, mask waiter.EventMask) { i.queue.EventRegister(e, mask) diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 88ed36b69..67645cc83 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -199,6 +199,7 @@ syscall_test( syscall_test( add_overlay = True, test = "//test/syscalls/linux:fallocate_test", + vfs2 = "True", ) syscall_test( diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 8c7d54b21..9e097c888 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -748,9 +748,14 @@ cc_binary( linkstatic = 1, deps = [ ":file_base", + ":socket_test_util", "//test/util:cleanup", + "//test/util:eventfd_util", "//test/util:file_descriptor", + "@com_google_absl//absl/strings", + "@com_google_absl//absl/time", gtest, + "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", diff --git a/test/syscalls/linux/fallocate.cc b/test/syscalls/linux/fallocate.cc index 7819f4ac3..cabc2b751 100644 --- a/test/syscalls/linux/fallocate.cc +++ b/test/syscalls/linux/fallocate.cc @@ -15,16 +15,27 @@ #include #include #include +#include #include +#include +#include #include +#include #include #include #include +#include + #include "gtest/gtest.h" +#include "absl/strings/str_cat.h" +#include "absl/time/time.h" #include "test/syscalls/linux/file_base.h" +#include "test/syscalls/linux/socket_test_util.h" #include "test/util/cleanup.h" +#include "test/util/eventfd_util.h" #include "test/util/file_descriptor.h" +#include "test/util/posix_error.h" #include "test/util/temp_path.h" #include "test/util/test_util.h" @@ -70,6 +81,12 @@ TEST_F(AllocateTest, Fallocate) { ASSERT_THAT(fallocate(test_file_fd_.get(), 0, 39, 1), SyscallSucceeds()); ASSERT_THAT(fstat(test_file_fd_.get(), &buf), SyscallSucceeds()); EXPECT_EQ(buf.st_size, 40); + + // Given length 0 should fail with EINVAL. + ASSERT_THAT(fallocate(test_file_fd_.get(), 0, 50, 0), + SyscallFailsWithErrno(EINVAL)); + ASSERT_THAT(fstat(test_file_fd_.get(), &buf), SyscallSucceeds()); + EXPECT_EQ(buf.st_size, 40); } TEST_F(AllocateTest, FallocateInvalid) { @@ -136,6 +153,34 @@ TEST_F(AllocateTest, FallocateRlimit) { ASSERT_THAT(sigprocmask(SIG_UNBLOCK, &new_mask, nullptr), SyscallSucceeds()); } +TEST_F(AllocateTest, FallocateOtherFDs) { + int fd; + ASSERT_THAT(fd = timerfd_create(CLOCK_MONOTONIC, 0), SyscallSucceeds()); + auto timer_fd = FileDescriptor(fd); + EXPECT_THAT(fallocate(timer_fd.get(), 0, 0, 10), + SyscallFailsWithErrno(ENODEV)); + + sigset_t mask; + sigemptyset(&mask); + ASSERT_THAT(fd = signalfd(-1, &mask, 0), SyscallSucceeds()); + auto sfd = FileDescriptor(fd); + EXPECT_THAT(fallocate(sfd.get(), 0, 0, 10), SyscallFailsWithErrno(ENODEV)); + + auto efd = + ASSERT_NO_ERRNO_AND_VALUE(NewEventFD(0, EFD_NONBLOCK | EFD_SEMAPHORE)); + EXPECT_THAT(fallocate(efd.get(), 0, 0, 10), SyscallFailsWithErrno(ENODEV)); + + auto sockfd = ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_DGRAM, 0)); + EXPECT_THAT(fallocate(sockfd.get(), 0, 0, 10), SyscallFailsWithErrno(ENODEV)); + + int socks[2]; + ASSERT_THAT(socketpair(AF_UNIX, SOCK_STREAM, PF_UNIX, socks), + SyscallSucceeds()); + auto sock0 = FileDescriptor(socks[0]); + auto sock1 = FileDescriptor(socks[1]); + EXPECT_THAT(fallocate(sock0.get(), 0, 0, 10), SyscallFailsWithErrno(ENODEV)); +} + } // namespace } // namespace testing } // namespace gvisor -- cgit v1.2.3 From 65d99855583a21b6ea511ea74aa52318d0a1e5b2 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Wed, 1 Jul 2020 17:09:26 -0700 Subject: Port vfs1 implementation of sync_file_range to vfs2. Currently, we always perform a full-file sync which could be extremely expensive for some applications. Although vfs1 did not fully support sync_file_range, there were some optimizations that allowed us skip some unnecessary write-outs. Updates #2923, #1897. PiperOrigin-RevId: 319324213 --- pkg/sentry/fsimpl/gofer/special_file.go | 5 +--- pkg/sentry/syscalls/linux/vfs2/sync.go | 42 +++++++++++++++++++++++++++------ test/syscalls/BUILD | 1 + 3 files changed, 37 insertions(+), 11 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go index a016cbae1..2b381af05 100644 --- a/pkg/sentry/fsimpl/gofer/special_file.go +++ b/pkg/sentry/fsimpl/gofer/special_file.go @@ -235,8 +235,5 @@ func (fd *specialFileFD) Seek(ctx context.Context, offset int64, whence int32) ( // Sync implements vfs.FileDescriptionImpl.Sync. func (fd *specialFileFD) Sync(ctx context.Context) error { - if !fd.vfsfd.IsWritable() { - return nil - } - return fd.handle.sync(ctx) + return fd.dentry().syncSharedHandle(ctx) } diff --git a/pkg/sentry/syscalls/linux/vfs2/sync.go b/pkg/sentry/syscalls/linux/vfs2/sync.go index 365250b0b..0d0ebf46a 100644 --- a/pkg/sentry/syscalls/linux/vfs2/sync.go +++ b/pkg/sentry/syscalls/linux/vfs2/sync.go @@ -65,10 +65,8 @@ func SyncFileRange(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel nbytes := args[2].Int64() flags := args[3].Uint() - if offset < 0 { - return 0, nil, syserror.EINVAL - } - if nbytes < 0 { + // Check for negative values and overflow. + if offset < 0 || offset+nbytes < 0 { return 0, nil, syserror.EINVAL } if flags&^(linux.SYNC_FILE_RANGE_WAIT_BEFORE|linux.SYNC_FILE_RANGE_WRITE|linux.SYNC_FILE_RANGE_WAIT_AFTER) != 0 { @@ -81,7 +79,37 @@ func SyncFileRange(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel } defer file.DecRef() - // TODO(gvisor.dev/issue/1897): Avoid writeback of data ranges outside of - // [offset, offset+nbytes). - return 0, nil, file.Sync(t) + // TODO(gvisor.dev/issue/1897): Currently, the only file syncing we support + // is a full-file sync, i.e. fsync(2). As a result, there are severe + // limitations on how much we support sync_file_range: + // - In Linux, sync_file_range(2) doesn't write out the file's metadata, even + // if the file size is changed. We do. + // - We always sync the entire file instead of [offset, offset+nbytes). + // - We do not support the use of WAIT_BEFORE without WAIT_AFTER. For + // correctness, we would have to perform a write-out every time WAIT_BEFORE + // was used, but this would be much more expensive than expected if there + // were no write-out operations in progress. + // - Whenever WAIT_AFTER is used, we sync the file. + // - Ignore WRITE. If this flag is used with WAIT_AFTER, then the file will + // be synced anyway. If this flag is used without WAIT_AFTER, then it is + // safe (and less expensive) to do nothing, because the syscall will not + // wait for the write-out to complete--we only need to make sure that the + // next time WAIT_BEFORE or WAIT_AFTER are used, the write-out completes. + // - According to fs/sync.c, WAIT_BEFORE|WAIT_AFTER "will detect any I/O + // errors or ENOSPC conditions and will return those to the caller, after + // clearing the EIO and ENOSPC flags in the address_space." We don't do + // this. + + if flags&linux.SYNC_FILE_RANGE_WAIT_BEFORE != 0 && + flags&linux.SYNC_FILE_RANGE_WAIT_AFTER == 0 { + t.Kernel().EmitUnimplementedEvent(t) + return 0, nil, syserror.ENOSYS + } + + if flags&linux.SYNC_FILE_RANGE_WAIT_AFTER != 0 { + if err := file.Sync(t); err != nil { + return 0, nil, syserror.ConvertIntr(err, kernel.ERESTARTSYS) + } + } + return 0, nil, nil } diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 67645cc83..790190273 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -962,6 +962,7 @@ syscall_test( syscall_test( add_overlay = True, test = "//test/syscalls/linux:sync_file_range_test", + vfs2 = "True", ) syscall_test( -- cgit v1.2.3 From 3b26d2121e3afbbc005b5e919000d55a235b4bbe Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Wed, 1 Jul 2020 17:39:07 -0700 Subject: Remove maxSendBufferSize from vfs2. Complements cl/315991648. PiperOrigin-RevId: 319327853 --- pkg/sentry/fsimpl/host/socket.go | 9 --------- 1 file changed, 9 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/host/socket.go b/pkg/sentry/fsimpl/host/socket.go index 38f1fbfba..fd16bd92d 100644 --- a/pkg/sentry/fsimpl/host/socket.go +++ b/pkg/sentry/fsimpl/host/socket.go @@ -47,11 +47,6 @@ func newEndpoint(ctx context.Context, hostFD int, queue *waiter.Queue) (transpor return ep, nil } -// maxSendBufferSize is the maximum host send buffer size allowed for endpoint. -// -// N.B. 8MB is the default maximum on Linux (2 * sysctl_wmem_max). -const maxSendBufferSize = 8 << 20 - // ConnectedEndpoint is an implementation of transport.ConnectedEndpoint and // transport.Receiver. It is backed by a host fd that was imported at sentry // startup. This fd is shared with a hostfs inode, which retains ownership of @@ -114,10 +109,6 @@ func (c *ConnectedEndpoint) init() *syserr.Error { if err != nil { return syserr.FromError(err) } - if sndbuf > maxSendBufferSize { - log.Warningf("Socket send buffer too large: %d", sndbuf) - return syserr.ErrInvalidEndpointState - } c.stype = linux.SockType(stype) c.sndbuf = int64(sndbuf) -- cgit v1.2.3 From 52b44719d6e14ec299d0d953b4dc07a712b897fa Mon Sep 17 00:00:00 2001 From: Ayush Ranjan Date: Wed, 1 Jul 2020 19:47:51 -0700 Subject: [vfs2][gofer] Update file size to 0 on O_TRUNC Some Open:TruncateXxx syscall tests were failing because the file size was not being updated when the file was opened with O_TRUNC. Fixes Truncate tests in test/syscalls:open_test_runsc_ptrace_vfs2. Updates #2923 PiperOrigin-RevId: 319340127 --- pkg/sentry/fsimpl/gofer/filesystem.go | 35 ++++++++++++++++-- pkg/sentry/fsimpl/gofer/gofer.go | 69 ++++++++++++++++++----------------- pkg/sentry/fsimpl/gofer/time.go | 12 +++--- pkg/sentry/fsimpl/tmpfs/filesystem.go | 2 +- 4 files changed, 76 insertions(+), 42 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 51082359d..7bcc99b29 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -869,11 +869,22 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vf if err := d.checkPermissions(rp.Credentials(), ats); err != nil { return nil, err } + + trunc := opts.Flags&linux.O_TRUNC != 0 && d.fileType() == linux.S_IFREG + if trunc { + // Lock metadataMu *while* we open a regular file with O_TRUNC because + // open(2) will change the file size on server. + d.metadataMu.Lock() + defer d.metadataMu.Unlock() + } + + var vfd *vfs.FileDescription + var err error mnt := rp.Mount() switch d.fileType() { case linux.S_IFREG: if !d.fs.opts.regularFilesUseSpecialFileFD { - if err := d.ensureSharedHandle(ctx, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, opts.Flags&linux.O_TRUNC != 0); err != nil { + if err := d.ensureSharedHandle(ctx, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, trunc); err != nil { return nil, err } fd := ®ularFileFD{} @@ -883,7 +894,7 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vf }); err != nil { return nil, err } - return &fd.vfsfd, nil + vfd = &fd.vfsfd } case linux.S_IFDIR: // Can't open directories with O_CREAT. @@ -923,7 +934,25 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vf return d.pipe.Open(ctx, mnt, &d.vfsd, opts.Flags, &d.locks) } } - return d.openSpecialFileLocked(ctx, mnt, opts) + + if vfd == nil { + if vfd, err = d.openSpecialFileLocked(ctx, mnt, opts); err != nil { + return nil, err + } + } + + if trunc { + // If no errors occured so far then update file size in memory. This + // step is required even if !d.cachedMetadataAuthoritative() because + // d.mappings has to be updated. + // d.metadataMu has already been acquired if trunc == true. + d.updateFileSizeLocked(0) + + if d.cachedMetadataAuthoritative() { + d.touchCMtimeLocked() + } + } + return vfd, err } func (d *dentry) connectSocketLocked(ctx context.Context, opts *vfs.OpenOptions) (*vfs.FileDescription, error) { diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 71c8d3ae1..709b5e496 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -794,9 +794,7 @@ func (d *dentry) updateFromP9Attrs(mask p9.AttrMask, attr *p9.Attr) { atomic.StoreUint32(&d.nlink, uint32(attr.NLink)) } if mask.Size { - d.dataMu.Lock() - atomic.StoreUint64(&d.size, attr.Size) - d.dataMu.Unlock() + d.updateFileSizeLocked(attr.Size) } d.metadataMu.Unlock() } @@ -964,39 +962,44 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin } atomic.StoreInt64(&d.ctime, now) if stat.Mask&linux.STATX_SIZE != 0 { + d.updateFileSizeLocked(stat.Size) + } + return nil +} + +// Preconditions: d.metadataMu must be locked. +func (d *dentry) updateFileSizeLocked(newSize uint64) { + d.dataMu.Lock() + oldSize := d.size + d.size = newSize + // d.dataMu must be unlocked to lock d.mapsMu and invalidate mappings + // below. This allows concurrent calls to Read/Translate/etc. These + // functions synchronize with truncation by refusing to use cache + // contents beyond the new d.size. (We are still holding d.metadataMu, + // so we can't race with Write or another truncate.) + d.dataMu.Unlock() + if d.size < oldSize { + oldpgend, _ := usermem.PageRoundUp(oldSize) + newpgend, _ := usermem.PageRoundUp(d.size) + if oldpgend != newpgend { + d.mapsMu.Lock() + d.mappings.Invalidate(memmap.MappableRange{newpgend, oldpgend}, memmap.InvalidateOpts{ + // Compare Linux's mm/truncate.c:truncate_setsize() => + // truncate_pagecache() => + // mm/memory.c:unmap_mapping_range(evencows=1). + InvalidatePrivate: true, + }) + d.mapsMu.Unlock() + } + // We are now guaranteed that there are no translations of + // truncated pages, and can remove them from the cache. Since + // truncated pages have been removed from the remote file, they + // should be dropped without being written back. d.dataMu.Lock() - oldSize := d.size - d.size = stat.Size - // d.dataMu must be unlocked to lock d.mapsMu and invalidate mappings - // below. This allows concurrent calls to Read/Translate/etc. These - // functions synchronize with truncation by refusing to use cache - // contents beyond the new d.size. (We are still holding d.metadataMu, - // so we can't race with Write or another truncate.) + d.cache.Truncate(d.size, d.fs.mfp.MemoryFile()) + d.dirty.KeepClean(memmap.MappableRange{d.size, oldpgend}) d.dataMu.Unlock() - if d.size < oldSize { - oldpgend, _ := usermem.PageRoundUp(oldSize) - newpgend, _ := usermem.PageRoundUp(d.size) - if oldpgend != newpgend { - d.mapsMu.Lock() - d.mappings.Invalidate(memmap.MappableRange{newpgend, oldpgend}, memmap.InvalidateOpts{ - // Compare Linux's mm/truncate.c:truncate_setsize() => - // truncate_pagecache() => - // mm/memory.c:unmap_mapping_range(evencows=1). - InvalidatePrivate: true, - }) - d.mapsMu.Unlock() - } - // We are now guaranteed that there are no translations of - // truncated pages, and can remove them from the cache. Since - // truncated pages have been removed from the remote file, they - // should be dropped without being written back. - d.dataMu.Lock() - d.cache.Truncate(d.size, d.fs.mfp.MemoryFile()) - d.dirty.KeepClean(memmap.MappableRange{d.size, oldpgend}) - d.dataMu.Unlock() - } } - return nil } func (d *dentry) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes) error { diff --git a/pkg/sentry/fsimpl/gofer/time.go b/pkg/sentry/fsimpl/gofer/time.go index 1d5aa82dc..0eef4e16e 100644 --- a/pkg/sentry/fsimpl/gofer/time.go +++ b/pkg/sentry/fsimpl/gofer/time.go @@ -36,7 +36,7 @@ func statxTimestampFromDentry(ns int64) linux.StatxTimestamp { } } -// Preconditions: fs.interop != InteropModeShared. +// Preconditions: d.cachedMetadataAuthoritative() == true. func (d *dentry) touchAtime(mnt *vfs.Mount) { if mnt.Flags.NoATime { return @@ -51,8 +51,8 @@ func (d *dentry) touchAtime(mnt *vfs.Mount) { mnt.EndWrite() } -// Preconditions: fs.interop != InteropModeShared. The caller has successfully -// called vfs.Mount.CheckBeginWrite(). +// Preconditions: d.cachedMetadataAuthoritative() == true. The caller has +// successfully called vfs.Mount.CheckBeginWrite(). func (d *dentry) touchCtime() { now := d.fs.clock.Now().Nanoseconds() d.metadataMu.Lock() @@ -60,8 +60,8 @@ func (d *dentry) touchCtime() { d.metadataMu.Unlock() } -// Preconditions: fs.interop != InteropModeShared. The caller has successfully -// called vfs.Mount.CheckBeginWrite(). +// Preconditions: d.cachedMetadataAuthoritative() == true. The caller has +// successfully called vfs.Mount.CheckBeginWrite(). func (d *dentry) touchCMtime() { now := d.fs.clock.Now().Nanoseconds() d.metadataMu.Lock() @@ -70,6 +70,8 @@ func (d *dentry) touchCMtime() { d.metadataMu.Unlock() } +// Preconditions: d.cachedMetadataAuthoritative() == true. The caller has +// locked d.metadataMu. func (d *dentry) touchCMtimeLocked() { now := d.fs.clock.Now().Nanoseconds() atomic.StoreInt64(&d.mtime, now) diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index ce0dbb154..ed40f6b52 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -410,7 +410,7 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.Open if err := fd.vfsfd.Init(&fd, opts.Flags, rp.Mount(), &d.vfsd, &vfs.FileDescriptionOptions{AllowDirectIO: true}); err != nil { return nil, err } - if opts.Flags&linux.O_TRUNC != 0 { + if !afterCreate && opts.Flags&linux.O_TRUNC != 0 { if _, err := impl.truncate(0); err != nil { return nil, err } -- cgit v1.2.3 From 514955c1a8f3927c928a57935d511ffbbf9c6f01 Mon Sep 17 00:00:00 2001 From: Ayush Ranjan Date: Wed, 1 Jul 2020 21:04:43 -0700 Subject: [vfs2][gofer] Fix mmap syscall test. We were not invalidating mappings when the file size changed in shared mode. Enabled the syscall test for vfs2. Updates #2923 PiperOrigin-RevId: 319346569 --- pkg/sentry/fsimpl/gofer/gofer.go | 9 ++++++--- test/syscalls/BUILD | 1 + 2 files changed, 7 insertions(+), 3 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 709b5e496..8e74e60a5 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -900,6 +900,12 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin } d.metadataMu.Lock() defer d.metadataMu.Unlock() + if stat.Mask&linux.STATX_SIZE != 0 { + // The size needs to be changed even when + // !d.cachedMetadataAuthoritative() because d.mappings has to be + // updated. + d.updateFileSizeLocked(stat.Size) + } if !d.isSynthetic() { if stat.Mask != 0 { if err := d.file.setAttr(ctx, p9.SetAttrMask{ @@ -961,9 +967,6 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin stat.Mask |= linux.STATX_MTIME } atomic.StoreInt64(&d.ctime, now) - if stat.Mask&linux.STATX_SIZE != 0 { - d.updateFileSizeLocked(stat.Size) - } return nil } diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 790190273..6fe397af9 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -357,6 +357,7 @@ syscall_test( size = "medium", shard_count = 5, test = "//test/syscalls/linux:mmap_test", + vfs2 = "True", ) syscall_test( -- cgit v1.2.3 From 6c099d830091b3153e0dc39cf500dba441f45707 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Wed, 1 Jul 2020 22:03:14 -0700 Subject: Update preadv2/pwritev2 flag handling in vfs2. We do not support RWF_SYNC/RWF_DSYNC and probably shouldn't silently accept them, since the user may incorrectly believe that we are synchronizing I/O. Remove the pwritev2 test verifying that we support these flags. gvisor.dev/issue/2601 is the tracking bug for deciding which RWF_.* flags we need and supporting them. Updates #2923, #2601. PiperOrigin-RevId: 319351286 --- pkg/abi/linux/file.go | 5 +-- pkg/sentry/fsimpl/gofer/regular_file.go | 8 +++-- pkg/sentry/fsimpl/gofer/special_file.go | 8 +++-- pkg/sentry/fsimpl/host/host.go | 4 ++- pkg/sentry/fsimpl/tmpfs/regular_file.go | 14 +++++--- test/syscalls/BUILD | 1 + test/syscalls/linux/pwritev2.cc | 59 ++++++--------------------------- 7 files changed, 39 insertions(+), 60 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/file.go b/pkg/abi/linux/file.go index 055ac1d7c..e11ca2d62 100644 --- a/pkg/abi/linux/file.go +++ b/pkg/abi/linux/file.go @@ -191,8 +191,9 @@ var DirentType = abi.ValueSet{ // Values for preadv2/pwritev2. const ( - // Note: gVisor does not implement the RWF_HIPRI feature, but the flag is - // accepted as a valid flag argument for preadv2/pwritev2. + // NOTE(b/120162627): gVisor does not implement the RWF_HIPRI feature, but + // the flag is accepted as a valid flag argument for preadv2/pwritev2 and + // silently ignored. RWF_HIPRI = 0x00000001 RWF_DSYNC = 0x00000002 RWF_SYNC = 0x00000004 diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go index faba73af2..3d2d3530a 100644 --- a/pkg/sentry/fsimpl/gofer/regular_file.go +++ b/pkg/sentry/fsimpl/gofer/regular_file.go @@ -102,7 +102,9 @@ func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offs return 0, syserror.EINVAL } - // Check that flags are supported. Silently ignore RWF_HIPRI. + // Check that flags are supported. + // + // TODO(gvisor.dev/issue/2601): Support select preadv2 flags. if opts.Flags&^linux.RWF_HIPRI != 0 { return 0, syserror.EOPNOTSUPP } @@ -155,7 +157,9 @@ func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off return 0, syserror.EINVAL } - // Check that flags are supported. Silently ignore RWF_HIPRI. + // Check that flags are supported. + // + // TODO(gvisor.dev/issue/2601): Support select pwritev2 flags. if opts.Flags&^linux.RWF_HIPRI != 0 { return 0, syserror.EOPNOTSUPP } diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go index 2b381af05..3c4e7e2e4 100644 --- a/pkg/sentry/fsimpl/gofer/special_file.go +++ b/pkg/sentry/fsimpl/gofer/special_file.go @@ -130,7 +130,9 @@ func (fd *specialFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offs return 0, syserror.EINVAL } - // Check that flags are supported. Silently ignore RWF_HIPRI. + // Check that flags are supported. + // + // TODO(gvisor.dev/issue/2601): Support select preadv2 flags. if opts.Flags&^linux.RWF_HIPRI != 0 { return 0, syserror.EOPNOTSUPP } @@ -176,7 +178,9 @@ func (fd *specialFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off return 0, syserror.EINVAL } - // Check that flags are supported. Silently ignore RWF_HIPRI. + // Check that flags are supported. + // + // TODO(gvisor.dev/issue/2601): Support select pwritev2 flags. if opts.Flags&^linux.RWF_HIPRI != 0 { return 0, syserror.EOPNOTSUPP } diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 007b3332e..1cd2982cb 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -588,8 +588,10 @@ func (f *fileDescription) Read(ctx context.Context, dst usermem.IOSequence, opts } func readFromHostFD(ctx context.Context, hostFD int, dst usermem.IOSequence, offset int64, flags uint32) (int64, error) { + // Check that flags are supported. + // // TODO(gvisor.dev/issue/2601): Support select preadv2 flags. - if flags != 0 { + if flags&^linux.RWF_HIPRI != 0 { return 0, syserror.EOPNOTSUPP } reader := hostfd.GetReadWriterAt(int32(hostFD), offset, flags) diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index 6691add96..1cdb46e6f 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -295,8 +295,11 @@ func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offs return 0, syserror.EINVAL } - // Check that flags are supported. Silently ignore RWF_HIPRI. - if opts.Flags&^linux.RWF_HIPRI != 0 { + // Check that flags are supported. RWF_DSYNC/RWF_SYNC can be ignored since + // all state is in-memory. + // + // TODO(gvisor.dev/issue/2601): Support select preadv2 flags. + if opts.Flags&^(linux.RWF_HIPRI|linux.RWF_DSYNC|linux.RWF_SYNC) != 0 { return 0, syserror.EOPNOTSUPP } @@ -326,8 +329,11 @@ func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off return 0, syserror.EINVAL } - // Check that flags are supported. Silently ignore RWF_HIPRI. - if opts.Flags&^linux.RWF_HIPRI != 0 { + // Check that flags are supported. RWF_DSYNC/RWF_SYNC can be ignored since + // all state is in-memory. + // + // TODO(gvisor.dev/issue/2601): Support select preadv2 flags. + if opts.Flags&^(linux.RWF_HIPRI|linux.RWF_DSYNC|linux.RWF_SYNC) != 0 { return 0, syserror.EOPNOTSUPP } diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 6fe397af9..7c4cd8192 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -531,6 +531,7 @@ syscall_test( syscall_test( add_overlay = True, test = "//test/syscalls/linux:pwritev2_test", + vfs2 = "True", ) syscall_test( diff --git a/test/syscalls/linux/pwritev2.cc b/test/syscalls/linux/pwritev2.cc index 3fe5a600f..63b686c62 100644 --- a/test/syscalls/linux/pwritev2.cc +++ b/test/syscalls/linux/pwritev2.cc @@ -69,7 +69,7 @@ ssize_t pwritev2(unsigned long fd, const struct iovec* iov, } // This test is the base case where we call pwritev (no offset, no flags). -TEST(Writev2Test, TestBaseCall) { +TEST(Writev2Test, BaseCall) { SKIP_IF(pwritev2(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS); const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileWith( @@ -97,7 +97,7 @@ TEST(Writev2Test, TestBaseCall) { } // This test is where we call pwritev2 with a positive offset and no flags. -TEST(Pwritev2Test, TestValidPositiveOffset) { +TEST(Pwritev2Test, ValidPositiveOffset) { SKIP_IF(pwritev2(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS); std::string prefix(kBufSize, '0'); @@ -129,7 +129,7 @@ TEST(Pwritev2Test, TestValidPositiveOffset) { // This test is the base case where we call writev by using -1 as the offset. // The write should use the file offset, so the test increments the file offset // prior to call pwritev2. -TEST(Pwritev2Test, TestNegativeOneOffset) { +TEST(Pwritev2Test, NegativeOneOffset) { SKIP_IF(pwritev2(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS); const std::string prefix = "00"; @@ -164,7 +164,7 @@ TEST(Pwritev2Test, TestNegativeOneOffset) { // pwritev2 requires if the RWF_HIPRI flag is passed, the fd must be opened with // O_DIRECT. This test implements a correct call with the RWF_HIPRI flag. -TEST(Pwritev2Test, TestCallWithRWF_HIPRI) { +TEST(Pwritev2Test, CallWithRWF_HIPRI) { SKIP_IF(pwritev2(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS); const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileWith( @@ -189,47 +189,8 @@ TEST(Pwritev2Test, TestCallWithRWF_HIPRI) { EXPECT_EQ(buf, content); } -// This test checks that pwritev2 can be called with valid flags -TEST(Pwritev2Test, TestCallWithValidFlags) { - SKIP_IF(pwritev2(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS); - - const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileWith( - GetAbsoluteTestTmpdir(), "", TempPath::kDefaultFileMode)); - const FileDescriptor fd = - ASSERT_NO_ERRNO_AND_VALUE(Open(file.path(), O_RDWR)); - - std::vector content(kBufSize, '0'); - struct iovec iov; - iov.iov_base = content.data(); - iov.iov_len = content.size(); - - EXPECT_THAT(pwritev2(fd.get(), &iov, /*iovcnt=*/1, - /*offset=*/0, /*flags=*/RWF_DSYNC), - SyscallSucceedsWithValue(kBufSize)); - - std::vector buf(content.size()); - EXPECT_THAT(read(fd.get(), buf.data(), buf.size()), - SyscallSucceedsWithValue(buf.size())); - - EXPECT_EQ(buf, content); - - SetContent(content); - - EXPECT_THAT(pwritev2(fd.get(), &iov, /*iovcnt=*/1, - /*offset=*/0, /*flags=*/0x4), - SyscallSucceedsWithValue(kBufSize)); - - ASSERT_THAT(lseek(fd.get(), 0, SEEK_CUR), - SyscallSucceedsWithValue(content.size())); - - EXPECT_THAT(pread(fd.get(), buf.data(), buf.size(), /*offset=*/0), - SyscallSucceedsWithValue(buf.size())); - - EXPECT_EQ(buf, content); -} - // This test calls pwritev2 with a bad file descriptor. -TEST(Writev2Test, TestBadFile) { +TEST(Writev2Test, BadFile) { SKIP_IF(pwritev2(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS); ASSERT_THAT(pwritev2(/*fd=*/-1, /*iov=*/nullptr, /*iovcnt=*/0, /*offset=*/0, /*flags=*/0), @@ -237,7 +198,7 @@ TEST(Writev2Test, TestBadFile) { } // This test calls pwrite2 with an invalid offset. -TEST(Pwritev2Test, TestInvalidOffset) { +TEST(Pwritev2Test, InvalidOffset) { SKIP_IF(pwritev2(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS); const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileWith( @@ -255,7 +216,7 @@ TEST(Pwritev2Test, TestInvalidOffset) { SyscallFailsWithErrno(EINVAL)); } -TEST(Pwritev2Test, TestUnseekableFileValid) { +TEST(Pwritev2Test, UnseekableFileValid) { SKIP_IF(pwritev2(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS); int pipe_fds[2]; @@ -285,7 +246,7 @@ TEST(Pwritev2Test, TestUnseekableFileValid) { // Calling pwritev2 with a non-negative offset calls pwritev. Calling pwritev // with an unseekable file is not allowed. A pipe is used for an unseekable // file. -TEST(Pwritev2Test, TestUnseekableFileInValid) { +TEST(Pwritev2Test, UnseekableFileInvalid) { SKIP_IF(pwritev2(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS); int pipe_fds[2]; @@ -304,7 +265,7 @@ TEST(Pwritev2Test, TestUnseekableFileInValid) { EXPECT_THAT(close(pipe_fds[1]), SyscallSucceeds()); } -TEST(Pwritev2Test, TestReadOnlyFile) { +TEST(Pwritev2Test, ReadOnlyFile) { SKIP_IF(pwritev2(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS); const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileWith( @@ -323,7 +284,7 @@ TEST(Pwritev2Test, TestReadOnlyFile) { } // This test calls pwritev2 with an invalid flag. -TEST(Pwritev2Test, TestInvalidFlag) { +TEST(Pwritev2Test, InvalidFlag) { SKIP_IF(pwritev2(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS); const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileWith( -- cgit v1.2.3 From 49f38dee10527d67048599740e208b1d11eda3bd Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Fri, 3 Jul 2020 00:06:38 -0400 Subject: allow guest user applications read ctr_el0 on Arm64 At present, when doing syscall_kvm test, we need to enable the function of ESR_ELx_SYS64_ISS_SYS_CTR_READ to successfully pass the test. I set SCTLR_EL1.UCT==1, so that the related cases can passed. Signed-off-by: Bin Lu --- pkg/sentry/platform/kvm/machine_arm64_unsafe.go | 13 ------------- pkg/sentry/platform/ring0/entry_arm64.s | 13 +++++++++++++ 2 files changed, 13 insertions(+), 13 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go index 8bed34922..3de309c1a 100644 --- a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go @@ -78,19 +78,6 @@ func (c *vCPU) initArchState() error { return err } - // sctlr_el1 - regGet.id = _KVM_ARM64_REGS_SCTLR_EL1 - if err := c.getOneRegister(®Get); err != nil { - return err - } - - dataGet |= (_SCTLR_M | _SCTLR_C | _SCTLR_I) - data = dataGet - reg.id = _KVM_ARM64_REGS_SCTLR_EL1 - if err := c.setOneRegister(®); err != nil { - return err - } - // tcr_el1 data = _TCR_TXSZ_VA48 | _TCR_CACHE_FLAGS | _TCR_SHARED | _TCR_TG_FLAGS | _TCR_ASID16 | _TCR_IPS_40BITS reg.id = _KVM_ARM64_REGS_TCR_EL1 diff --git a/pkg/sentry/platform/ring0/entry_arm64.s b/pkg/sentry/platform/ring0/entry_arm64.s index 2bc5f3ecd..6ed73699b 100644 --- a/pkg/sentry/platform/ring0/entry_arm64.s +++ b/pkg/sentry/platform/ring0/entry_arm64.s @@ -40,6 +40,14 @@ #define FPEN_ENABLE (FPEN_NOTRAP << FPEN_SHIFT) +// sctlr_el1: system control register el1. +#define SCTLR_M 1 << 0 +#define SCTLR_C 1 << 2 +#define SCTLR_I 1 << 12 +#define SCTLR_UCT 1 << 15 + +#define SCTLR_EL1_DEFAULT (SCTLR_M | SCTLR_C | SCTLR_I | SCTLR_UCT) + // Saves a register set. // // This is a macro because it may need to executed in contents where a stack is @@ -496,6 +504,11 @@ TEXT ·kernelExitToEl1(SB),NOSPLIT,$0 // Start is the CPU entrypoint. TEXT ·Start(SB),NOSPLIT,$0 IRQ_DISABLE + + // Init. + MOVD $SCTLR_EL1_DEFAULT, R1 + MSR R1, SCTLR_EL1 + MOVD R8, RSV_REG ORR $0xffff000000000000, RSV_REG, RSV_REG WORD $0xd518d092 //MSR R18, TPIDR_EL1 -- cgit v1.2.3 From 057a2666fa50ac0d0f3b0472a6e1f3d909cc7a12 Mon Sep 17 00:00:00 2001 From: Jianfeng Tan Date: Mon, 6 Jul 2020 20:53:51 +0800 Subject: hostinet: fix not specifying family field Creating sockets by hostinet with VFS2 fails due to triggerring a seccomp violation. In essence, we fails to pass down the field of family. We fix this by passing down this field, family. Fixes #3141 Signed-off-by: Jianfeng Tan --- pkg/sentry/socket/hostinet/socket.go | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/hostinet/socket.go b/pkg/sentry/socket/hostinet/socket.go index a92aed2c9..fda19e7bb 100644 --- a/pkg/sentry/socket/hostinet/socket.go +++ b/pkg/sentry/socket/hostinet/socket.go @@ -708,6 +708,6 @@ func (p *socketProvider) Pair(t *kernel.Task, stype linux.SockType, protocol int func init() { for _, family := range []int{syscall.AF_INET, syscall.AF_INET6} { socket.RegisterProvider(family, &socketProvider{family}) - socket.RegisterProviderVFS2(family, &socketProviderVFS2{}) + socket.RegisterProviderVFS2(family, &socketProviderVFS2{family}) } } -- cgit v1.2.3 From bd43368f491a02b050cd9f87f69185dc74386f2b Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Mon, 6 Jul 2020 15:38:51 -0700 Subject: Add inode number to synthetic dentries Reserve the MSB from ino for synthetic dentries to prevent conflict with regular dentries. Log warning in case MSB is set for regular dentries. Updates #1487 PiperOrigin-RevId: 319869858 --- pkg/sentry/fsimpl/gofer/directory.go | 9 +++++---- pkg/sentry/fsimpl/gofer/filesystem.go | 9 ++++++--- pkg/sentry/fsimpl/gofer/gofer.go | 34 +++++++++++++++++++++++++++------- 3 files changed, 38 insertions(+), 14 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/directory.go b/pkg/sentry/fsimpl/gofer/directory.go index 5d83fe363..8c7c8e1b3 100644 --- a/pkg/sentry/fsimpl/gofer/directory.go +++ b/pkg/sentry/fsimpl/gofer/directory.go @@ -85,6 +85,7 @@ func (d *dentry) createSyntheticChildLocked(opts *createSyntheticOpts) { d2 := &dentry{ refs: 1, // held by d fs: d.fs, + ino: d.fs.nextSyntheticIno(), mode: uint32(opts.mode), uid: uint32(opts.kuid), gid: uint32(opts.kgid), @@ -184,13 +185,13 @@ func (d *dentry) getDirents(ctx context.Context) ([]vfs.Dirent, error) { { Name: ".", Type: linux.DT_DIR, - Ino: d.ino, + Ino: uint64(d.ino), NextOff: 1, }, { Name: "..", Type: uint8(atomic.LoadUint32(&parent.mode) >> 12), - Ino: parent.ino, + Ino: uint64(parent.ino), NextOff: 2, }, } @@ -226,7 +227,7 @@ func (d *dentry) getDirents(ctx context.Context) ([]vfs.Dirent, error) { } dirent := vfs.Dirent{ Name: p9d.Name, - Ino: p9d.QID.Path, + Ino: uint64(inoFromPath(p9d.QID.Path)), NextOff: int64(len(dirents) + 1), } // p9 does not expose 9P2000.U's DMDEVICE, DMNAMEDPIPE, or @@ -259,7 +260,7 @@ func (d *dentry) getDirents(ctx context.Context) ([]vfs.Dirent, error) { dirents = append(dirents, vfs.Dirent{ Name: child.name, Type: uint8(atomic.LoadUint32(&child.mode) >> 12), - Ino: child.ino, + Ino: uint64(child.ino), NextOff: int64(len(dirents) + 1), }) } diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 7bcc99b29..cd5f5049e 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -214,9 +214,8 @@ func (fs *filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.Vir return nil, err } if child != nil { - if !file.isNil() && qid.Path == child.ino { - // The file at this path hasn't changed. Just update cached - // metadata. + if !file.isNil() && inoFromPath(qid.Path) == child.ino { + // The file at this path hasn't changed. Just update cached metadata. file.close(ctx) child.updateFromP9Attrs(attrMask, &attr) return child, nil @@ -1499,3 +1498,7 @@ func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDe defer fs.renameMu.RUnlock() return genericPrependPath(vfsroot, vd.Mount(), vd.Dentry().Impl().(*dentry), b) } + +func (fs *filesystem) nextSyntheticIno() inodeNumber { + return inodeNumber(atomic.AddUint64(&fs.syntheticSeq, 1) | syntheticInoMask) +} diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 8e74e60a5..2b83094cd 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -110,6 +110,26 @@ type filesystem struct { syncMu sync.Mutex syncableDentries map[*dentry]struct{} specialFileFDs map[*specialFileFD]struct{} + + // syntheticSeq stores a counter to used to generate unique inodeNumber for + // synthetic dentries. + syntheticSeq uint64 +} + +// inodeNumber represents inode number reported in Dirent.Ino. For regular +// dentries, it comes from QID.Path from the 9P server. Synthetic dentries +// have have their inodeNumber generated sequentially, with the MSB reserved to +// prevent conflicts with regular dentries. +type inodeNumber uint64 + +// Reserve MSB for synthetic mounts. +const syntheticInoMask = uint64(1) << 63 + +func inoFromPath(path uint64) inodeNumber { + if path&syntheticInoMask != 0 { + log.Warningf("Dropping MSB from ino, collision is possible. Original: %d, new: %d", path, path&^syntheticInoMask) + } + return inodeNumber(path &^ syntheticInoMask) } type filesystemOptions struct { @@ -585,11 +605,11 @@ type dentry struct { // Cached metadata; protected by metadataMu and accessed using atomic // memory operations unless otherwise specified. metadataMu sync.Mutex - ino uint64 // immutable - mode uint32 // type is immutable, perms are mutable - uid uint32 // auth.KUID, but stored as raw uint32 for sync/atomic - gid uint32 // auth.KGID, but ... - blockSize uint32 // 0 if unknown + ino inodeNumber // immutable + mode uint32 // type is immutable, perms are mutable + uid uint32 // auth.KUID, but stored as raw uint32 for sync/atomic + gid uint32 // auth.KGID, but ... + blockSize uint32 // 0 if unknown // Timestamps, all nsecs from the Unix epoch. atime int64 mtime int64 @@ -704,7 +724,7 @@ func (fs *filesystem) newDentry(ctx context.Context, file p9file, qid p9.QID, ma d := &dentry{ fs: fs, file: file, - ino: qid.Path, + ino: inoFromPath(qid.Path), mode: uint32(attr.Mode), uid: uint32(fs.opts.dfltuid), gid: uint32(fs.opts.dfltgid), @@ -846,7 +866,7 @@ func (d *dentry) statTo(stat *linux.Statx) { stat.UID = atomic.LoadUint32(&d.uid) stat.GID = atomic.LoadUint32(&d.gid) stat.Mode = uint16(atomic.LoadUint32(&d.mode)) - stat.Ino = d.ino + stat.Ino = uint64(d.ino) stat.Size = atomic.LoadUint64(&d.size) // This is consistent with regularFileFD.Seek(), which treats regular files // as having no holes. -- cgit v1.2.3 From 937912a4847cb1fd2acb79c0dd8fbc4b27695156 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Mon, 6 Jul 2020 17:24:25 -0700 Subject: Ensure sync is called for readonly file Calling sync on a readonly file flushes metadata that may have been modified, like last access time. Updates #1198 PiperOrigin-RevId: 319888290 --- pkg/sentry/fsimpl/gofer/handle.go | 5 +++++ pkg/sentry/fsimpl/gofer/regular_file.go | 25 ++++++++++++------------- 2 files changed, 17 insertions(+), 13 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/handle.go b/pkg/sentry/fsimpl/gofer/handle.go index 724a3f1f7..8792ca4f2 100644 --- a/pkg/sentry/fsimpl/gofer/handle.go +++ b/pkg/sentry/fsimpl/gofer/handle.go @@ -126,11 +126,16 @@ func (h *handle) writeFromBlocksAt(ctx context.Context, srcs safemem.BlockSeq, o } func (h *handle) sync(ctx context.Context) error { + // Handle most common case first. if h.fd >= 0 { ctx.UninterruptibleSleepStart(false) err := syscall.Fsync(int(h.fd)) ctx.UninterruptibleSleepFinish(false) return err } + if h.file.isNil() { + // File hasn't been touched, there is nothing to sync. + return nil + } return h.file.fsync(ctx) } diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go index 3d2d3530a..a2f02d9c7 100644 --- a/pkg/sentry/fsimpl/gofer/regular_file.go +++ b/pkg/sentry/fsimpl/gofer/regular_file.go @@ -582,20 +582,19 @@ func (fd *regularFileFD) Sync(ctx context.Context) error { func (d *dentry) syncSharedHandle(ctx context.Context) error { d.handleMu.RLock() - if !d.handleWritable { - d.handleMu.RUnlock() - return nil - } - d.dataMu.Lock() - // Write dirty cached data to the remote file. - err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, d.fs.mfp.MemoryFile(), d.handle.writeFromBlocksAt) - d.dataMu.Unlock() - if err == nil { - // Sync the remote file. - err = d.handle.sync(ctx) + defer d.handleMu.RUnlock() + + if d.handleWritable { + d.dataMu.Lock() + // Write dirty cached data to the remote file. + err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, d.fs.mfp.MemoryFile(), d.handle.writeFromBlocksAt) + d.dataMu.Unlock() + if err != nil { + return err + } } - d.handleMu.RUnlock() - return err + // Sync the remote file. + return d.handle.sync(ctx) } // ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. -- cgit v1.2.3 From e227450dc1ec1ef07b9a6a5983ce2f2e0dec22b1 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Mon, 6 Jul 2020 22:54:01 -0700 Subject: Call fdnotifier.UpdateFD() from fsimpl/gofer.specialFileFD. The fdnotifier package provides an API to a thread that continually epolls arbitrary host FDs. The set of events polled for each host FD is (intended to be) all events for which a waiter.Entry has expressed interest, as returned by waiter.Queue.Events() for the waiter.Queue registered to the given host FD. When the set of events changes (due to a change in the set of registered waiter.Entries), the mutator must call fdnotifier.UpdateFD() to recalculate the new event set and propagate it to the epoll FD. PiperOrigin-RevId: 319924719 --- pkg/sentry/fsimpl/gofer/special_file.go | 30 ++++++++++++++++-------------- 1 file changed, 16 insertions(+), 14 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go index 3c4e7e2e4..c1e6b13e5 100644 --- a/pkg/sentry/fsimpl/gofer/special_file.go +++ b/pkg/sentry/fsimpl/gofer/special_file.go @@ -41,10 +41,10 @@ type specialFileFD struct { // file offset is significant, i.e. a regular file. seekable is immutable. seekable bool - // mayBlock is true if this file description represents a file for which - // queue may send I/O readiness events. mayBlock is immutable. - mayBlock bool - queue waiter.Queue + // haveQueue is true if this file description represents a file for which + // queue may send I/O readiness events. haveQueue is immutable. + haveQueue bool + queue waiter.Queue // If seekable is true, off is the file offset. off is protected by mu. mu sync.Mutex @@ -54,14 +54,14 @@ type specialFileFD struct { func newSpecialFileFD(h handle, mnt *vfs.Mount, d *dentry, locks *vfs.FileLocks, flags uint32) (*specialFileFD, error) { ftype := d.fileType() seekable := ftype == linux.S_IFREG - mayBlock := ftype == linux.S_IFIFO || ftype == linux.S_IFSOCK + haveQueue := (ftype == linux.S_IFIFO || ftype == linux.S_IFSOCK) && h.fd >= 0 fd := &specialFileFD{ - handle: h, - seekable: seekable, - mayBlock: mayBlock, + handle: h, + seekable: seekable, + haveQueue: haveQueue, } fd.LockFD.Init(locks) - if mayBlock && h.fd >= 0 { + if haveQueue { if err := fdnotifier.AddFD(h.fd, &fd.queue); err != nil { return nil, err } @@ -70,7 +70,7 @@ func newSpecialFileFD(h handle, mnt *vfs.Mount, d *dentry, locks *vfs.FileLocks, DenyPRead: !seekable, DenyPWrite: !seekable, }); err != nil { - if mayBlock && h.fd >= 0 { + if haveQueue { fdnotifier.RemoveFD(h.fd) } return nil, err @@ -80,7 +80,7 @@ func newSpecialFileFD(h handle, mnt *vfs.Mount, d *dentry, locks *vfs.FileLocks, // Release implements vfs.FileDescriptionImpl.Release. func (fd *specialFileFD) Release() { - if fd.mayBlock && fd.handle.fd >= 0 { + if fd.haveQueue { fdnotifier.RemoveFD(fd.handle.fd) } fd.handle.close(context.Background()) @@ -100,7 +100,7 @@ func (fd *specialFileFD) OnClose(ctx context.Context) error { // Readiness implements waiter.Waitable.Readiness. func (fd *specialFileFD) Readiness(mask waiter.EventMask) waiter.EventMask { - if fd.mayBlock { + if fd.haveQueue { return fdnotifier.NonBlockingPoll(fd.handle.fd, mask) } return fd.fileDescription.Readiness(mask) @@ -108,8 +108,9 @@ func (fd *specialFileFD) Readiness(mask waiter.EventMask) waiter.EventMask { // EventRegister implements waiter.Waitable.EventRegister. func (fd *specialFileFD) EventRegister(e *waiter.Entry, mask waiter.EventMask) { - if fd.mayBlock { + if fd.haveQueue { fd.queue.EventRegister(e, mask) + fdnotifier.UpdateFD(fd.handle.fd) return } fd.fileDescription.EventRegister(e, mask) @@ -117,8 +118,9 @@ func (fd *specialFileFD) EventRegister(e *waiter.Entry, mask waiter.EventMask) { // EventUnregister implements waiter.Waitable.EventUnregister. func (fd *specialFileFD) EventUnregister(e *waiter.Entry) { - if fd.mayBlock { + if fd.haveQueue { fd.queue.EventUnregister(e) + fdnotifier.UpdateFD(fd.handle.fd) return } fd.fileDescription.EventUnregister(e) -- cgit v1.2.3 From 10930189c3cd938e7526c55188ba2d814a7b8a43 Mon Sep 17 00:00:00 2001 From: Ayush Ranjan Date: Tue, 7 Jul 2020 15:33:32 -0700 Subject: Fix mknod and inotify syscall test This change fixes a few things: - creating sockets using mknod(2) is supported via vfs2 - fsgofer can create regular files via mknod(2) - mode = 0 for mknod(2) will be interpreted as regular file in vfs2 as well Updates #2923 PiperOrigin-RevId: 320074267 --- pkg/sentry/fsimpl/tmpfs/filesystem.go | 2 +- pkg/sentry/syscalls/linux/vfs2/filesystem.go | 13 ++++-- runsc/fsgofer/filter/config.go | 1 + runsc/fsgofer/fsgofer.go | 60 ++++++++++++++++++++++------ test/syscalls/BUILD | 2 + test/syscalls/linux/mknod.cc | 26 +++++++++++- 6 files changed, 84 insertions(+), 20 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index ed40f6b52..a0f20c2d4 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -277,7 +277,7 @@ func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v creds := rp.Credentials() var childInode *inode switch opts.Mode.FileType() { - case 0, linux.S_IFREG: + case linux.S_IFREG: childInode = fs.newRegularFile(creds.EffectiveKUID, creds.EffectiveKGID, opts.Mode) case linux.S_IFIFO: childInode = fs.newNamedPipe(creds.EffectiveKUID, creds.EffectiveKGID, opts.Mode) diff --git a/pkg/sentry/syscalls/linux/vfs2/filesystem.go b/pkg/sentry/syscalls/linux/vfs2/filesystem.go index b12b5967b..6b14c2bef 100644 --- a/pkg/sentry/syscalls/linux/vfs2/filesystem.go +++ b/pkg/sentry/syscalls/linux/vfs2/filesystem.go @@ -107,7 +107,7 @@ func Mknod(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall addr := args[0].Pointer() mode := args[1].ModeT() dev := args[2].Uint() - return 0, nil, mknodat(t, linux.AT_FDCWD, addr, mode, dev) + return 0, nil, mknodat(t, linux.AT_FDCWD, addr, linux.FileMode(mode), dev) } // Mknodat implements Linux syscall mknodat(2). @@ -116,10 +116,10 @@ func Mknodat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca addr := args[1].Pointer() mode := args[2].ModeT() dev := args[3].Uint() - return 0, nil, mknodat(t, dirfd, addr, mode, dev) + return 0, nil, mknodat(t, dirfd, addr, linux.FileMode(mode), dev) } -func mknodat(t *kernel.Task, dirfd int32, addr usermem.Addr, mode uint, dev uint32) error { +func mknodat(t *kernel.Task, dirfd int32, addr usermem.Addr, mode linux.FileMode, dev uint32) error { path, err := copyInPath(t, addr) if err != nil { return err @@ -129,9 +129,14 @@ func mknodat(t *kernel.Task, dirfd int32, addr usermem.Addr, mode uint, dev uint return err } defer tpop.Release() + + // "Zero file type is equivalent to type S_IFREG." - mknod(2) + if mode.FileType() == 0 { + mode |= linux.ModeRegular + } major, minor := linux.DecodeDeviceID(dev) return t.Kernel().VFS().MknodAt(t, t.Credentials(), &tpop.pop, &vfs.MknodOptions{ - Mode: linux.FileMode(mode &^ t.FSContext().Umask()), + Mode: mode &^ linux.FileMode(t.FSContext().Umask()), DevMajor: uint32(major), DevMinor: minor, }) diff --git a/runsc/fsgofer/filter/config.go b/runsc/fsgofer/filter/config.go index 1dce36965..88814b83c 100644 --- a/runsc/fsgofer/filter/config.go +++ b/runsc/fsgofer/filter/config.go @@ -128,6 +128,7 @@ var allowedSyscalls = seccomp.SyscallRules{ syscall.SYS_MADVISE: {}, unix.SYS_MEMFD_CREATE: {}, /// Used by flipcall.PacketWindowAllocator.Init(). syscall.SYS_MKDIRAT: {}, + syscall.SYS_MKNODAT: {}, // Used by the Go runtime as a temporarily workaround for a Linux // 5.2-5.4 bug. // diff --git a/runsc/fsgofer/fsgofer.go b/runsc/fsgofer/fsgofer.go index 74977c313..b7521bda7 100644 --- a/runsc/fsgofer/fsgofer.go +++ b/runsc/fsgofer/fsgofer.go @@ -139,7 +139,7 @@ func (a *attachPoint) Attach() (p9.File, error) { return nil, fmt.Errorf("unable to open %q: %v", a.prefix, err) } - stat, err := stat(f.FD()) + stat, err := fstat(f.FD()) if err != nil { return nil, fmt.Errorf("unable to stat %q: %v", a.prefix, err) } @@ -352,7 +352,7 @@ func newFDMaybe(file *fd.FD) *fd.FD { return dup } -func stat(fd int) (syscall.Stat_t, error) { +func fstat(fd int) (syscall.Stat_t, error) { var stat syscall.Stat_t if err := syscall.Fstat(fd, &stat); err != nil { return syscall.Stat_t{}, err @@ -360,6 +360,14 @@ func stat(fd int) (syscall.Stat_t, error) { return stat, nil } +func stat(path string) (syscall.Stat_t, error) { + var stat syscall.Stat_t + if err := syscall.Stat(path, &stat); err != nil { + return syscall.Stat_t{}, err + } + return stat, nil +} + func fchown(fd int, uid p9.UID, gid p9.GID) error { return syscall.Fchownat(fd, "", int(uid), int(gid), linux.AT_EMPTY_PATH|unix.AT_SYMLINK_NOFOLLOW) } @@ -388,7 +396,7 @@ func (l *localFile) Open(flags p9.OpenFlags) (*fd.FD, p9.QID, uint32, error) { } } - stat, err := stat(newFile.FD()) + stat, err := fstat(newFile.FD()) if err != nil { if newFile != l.file { newFile.Close() @@ -449,7 +457,7 @@ func (l *localFile) Create(name string, mode p9.OpenFlags, perm p9.FileMode, uid if err := fchown(child.FD(), uid, gid); err != nil { return nil, nil, p9.QID{}, 0, extractErrno(err) } - stat, err := stat(child.FD()) + stat, err := fstat(child.FD()) if err != nil { return nil, nil, p9.QID{}, 0, extractErrno(err) } @@ -497,7 +505,7 @@ func (l *localFile) Mkdir(name string, perm p9.FileMode, uid p9.UID, gid p9.GID) if err := fchown(f.FD(), uid, gid); err != nil { return p9.QID{}, extractErrno(err) } - stat, err := stat(f.FD()) + stat, err := fstat(f.FD()) if err != nil { return p9.QID{}, extractErrno(err) } @@ -517,7 +525,7 @@ func (l *localFile) Walk(names []string) ([]p9.QID, p9.File, error) { return nil, nil, extractErrno(err) } - stat, err := stat(newFile.FD()) + stat, err := fstat(newFile.FD()) if err != nil { newFile.Close() return nil, nil, extractErrno(err) @@ -542,7 +550,7 @@ func (l *localFile) Walk(names []string) ([]p9.QID, p9.File, error) { if err != nil { return nil, nil, extractErrno(err) } - stat, err := stat(f.FD()) + stat, err := fstat(f.FD()) if err != nil { f.Close() return nil, nil, extractErrno(err) @@ -592,7 +600,7 @@ func (l *localFile) FSync() error { // GetAttr implements p9.File. func (l *localFile) GetAttr(_ p9.AttrMask) (p9.QID, p9.AttrMask, p9.Attr, error) { - stat, err := stat(l.file.FD()) + stat, err := fstat(l.file.FD()) if err != nil { return p9.QID{}, p9.AttrMask{}, p9.Attr{}, extractErrno(err) } @@ -880,7 +888,7 @@ func (l *localFile) Symlink(target, newName string, uid p9.UID, gid p9.GID) (p9. if err := fchown(f.FD(), uid, gid); err != nil { return p9.QID{}, extractErrno(err) } - stat, err := stat(f.FD()) + stat, err := fstat(f.FD()) if err != nil { return p9.QID{}, extractErrno(err) } @@ -907,13 +915,39 @@ func (l *localFile) Link(target p9.File, newName string) error { } // Mknod implements p9.File. -// -// Not implemented. -func (*localFile) Mknod(_ string, _ p9.FileMode, _ uint32, _ uint32, _ p9.UID, _ p9.GID) (p9.QID, error) { +func (l *localFile) Mknod(name string, mode p9.FileMode, _ uint32, _ uint32, uid p9.UID, gid p9.GID) (p9.QID, error) { + conf := l.attachPoint.conf + if conf.ROMount { + if conf.PanicOnWrite { + panic("attempt to write to RO mount") + } + return p9.QID{}, syscall.EROFS + } + + hostPath := path.Join(l.hostPath, name) + + // Return EEXIST if the file already exists. + if _, err := stat(hostPath); err == nil { + return p9.QID{}, syscall.EEXIST + } + // From mknod(2) man page: // "EPERM: [...] if the filesystem containing pathname does not support // the type of node requested." - return p9.QID{}, syscall.EPERM + if mode.FileType() != p9.ModeRegular { + return p9.QID{}, syscall.EPERM + } + + // Allow Mknod to create regular files. + if err := syscall.Mknod(hostPath, uint32(mode), 0); err != nil { + return p9.QID{}, err + } + + stat, err := stat(hostPath) + if err != nil { + return p9.QID{}, extractErrno(err) + } + return l.attachPoint.makeQID(stat), nil } // UnlinkAt implements p9.File. diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 7c4cd8192..28ef55945 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -283,6 +283,7 @@ syscall_test( size = "medium", add_overlay = False, # TODO(gvisor.dev/issue/317): enable when fixed. test = "//test/syscalls/linux:inotify_test", + vfs2 = "True", ) syscall_test( @@ -351,6 +352,7 @@ syscall_test( syscall_test( add_overlay = True, test = "//test/syscalls/linux:mknod_test", + vfs2 = "True", ) syscall_test( diff --git a/test/syscalls/linux/mknod.cc b/test/syscalls/linux/mknod.cc index 4c45766c7..05dfb375a 100644 --- a/test/syscalls/linux/mknod.cc +++ b/test/syscalls/linux/mknod.cc @@ -15,6 +15,7 @@ #include #include #include +#include #include #include @@ -39,7 +40,28 @@ TEST(MknodTest, RegularFile) { EXPECT_THAT(mknod(node1.c_str(), 0, 0), SyscallSucceeds()); } -TEST(MknodTest, MknodAtRegularFile) { +TEST(MknodTest, RegularFilePermissions) { + const std::string node = NewTempAbsPath(); + mode_t newUmask = 0077; + umask(newUmask); + + // Attempt to open file with mode 0777. Not specifying file type should create + // a regualar file. + mode_t perms = S_IRWXU | S_IRWXG | S_IRWXO; + EXPECT_THAT(mknod(node.c_str(), perms, 0), SyscallSucceeds()); + + // In the absence of a default ACL, the permissions of the created node are + // (mode & ~umask). -- mknod(2) + mode_t wantPerms = perms & ~newUmask; + struct stat st; + ASSERT_THAT(stat(node.c_str(), &st), SyscallSucceeds()); + ASSERT_EQ(st.st_mode & 0777, wantPerms); + + // "Zero file type is equivalent to type S_IFREG." - mknod(2) + ASSERT_EQ(st.st_mode & S_IFMT, S_IFREG); +} + +TEST(MknodTest, MknodAtFIFO) { const TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); const std::string fifo_relpath = NewTempRelPath(); const std::string fifo = JoinPath(dir.path(), fifo_relpath); @@ -72,7 +94,7 @@ TEST(MknodTest, MknodOnExistingPathFails) { TEST(MknodTest, UnimplementedTypesReturnError) { const std::string path = NewTempAbsPath(); - if (IsRunningOnGvisor()) { + if (IsRunningWithVFS1()) { ASSERT_THAT(mknod(path.c_str(), S_IFSOCK, 0), SyscallFailsWithErrno(EOPNOTSUPP)); } -- cgit v1.2.3 From efa2615eb008a642dc542176759dc560c5f48a2d Mon Sep 17 00:00:00 2001 From: Ayush Ranjan Date: Tue, 7 Jul 2020 21:35:47 -0700 Subject: [vfs2] Remove VFS1 usage in VDSO. Removed VDSO dependency on VFS1. Resolves #2921 PiperOrigin-RevId: 320122176 --- pkg/sentry/fsimpl/testutil/kernel.go | 2 +- pkg/sentry/loader/BUILD | 4 --- pkg/sentry/loader/elf.go | 15 +++++++-- pkg/sentry/loader/loader.go | 17 ---------- pkg/sentry/loader/vdso.go | 61 +++--------------------------------- runsc/boot/loader.go | 4 +-- 6 files changed, 19 insertions(+), 84 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/testutil/kernel.go b/pkg/sentry/fsimpl/testutil/kernel.go index c16a36cdb..ecc81e6bf 100644 --- a/pkg/sentry/fsimpl/testutil/kernel.go +++ b/pkg/sentry/fsimpl/testutil/kernel.go @@ -73,7 +73,7 @@ func Boot() (*kernel.Kernel, error) { k.SetMemoryFile(mf) // Pass k as the platform since it is savable, unlike the actual platform. - vdso, err := loader.PrepareVDSO(nil, k) + vdso, err := loader.PrepareVDSO(k) if err != nil { return nil, fmt.Errorf("creating vdso: %v", err) } diff --git a/pkg/sentry/loader/BUILD b/pkg/sentry/loader/BUILD index c6aa65f28..34bdb0b69 100644 --- a/pkg/sentry/loader/BUILD +++ b/pkg/sentry/loader/BUILD @@ -30,9 +30,6 @@ go_library( "//pkg/rand", "//pkg/safemem", "//pkg/sentry/arch", - "//pkg/sentry/fs", - "//pkg/sentry/fs/anon", - "//pkg/sentry/fs/fsutil", "//pkg/sentry/fsbridge", "//pkg/sentry/kernel/auth", "//pkg/sentry/limits", @@ -45,6 +42,5 @@ go_library( "//pkg/syserr", "//pkg/syserror", "//pkg/usermem", - "//pkg/waiter", ], ) diff --git a/pkg/sentry/loader/elf.go b/pkg/sentry/loader/elf.go index 616fafa2c..ddeaff3db 100644 --- a/pkg/sentry/loader/elf.go +++ b/pkg/sentry/loader/elf.go @@ -90,14 +90,23 @@ type elfInfo struct { sharedObject bool } +// fullReader interface extracts the ReadFull method from fsbridge.File so that +// client code does not need to define an entire fsbridge.File when only read +// functionality is needed. +// +// TODO(gvisor.dev/issue/1035): Once VFS2 ships, rewrite this to wrap +// vfs.FileDescription's PRead/Read instead. +type fullReader interface { + // ReadFull is the same as fsbridge.File.ReadFull. + ReadFull(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) +} + // parseHeader parse the ELF header, verifying that this is a supported ELF // file and returning the ELF program headers. // // This is similar to elf.NewFile, except that it is more strict about what it // accepts from the ELF, and it doesn't parse unnecessary parts of the file. -// -// ctx may be nil if f does not need it. -func parseHeader(ctx context.Context, f fsbridge.File) (elfInfo, error) { +func parseHeader(ctx context.Context, f fullReader) (elfInfo, error) { // Check ident first; it will tell us the endianness of the rest of the // structs. var ident [elf.EI_NIDENT]byte diff --git a/pkg/sentry/loader/loader.go b/pkg/sentry/loader/loader.go index 88449fe95..e65ec4771 100644 --- a/pkg/sentry/loader/loader.go +++ b/pkg/sentry/loader/loader.go @@ -27,7 +27,6 @@ import ( "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/rand" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fsbridge" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/mm" @@ -80,22 +79,6 @@ type LoadArgs struct { Features *cpuid.FeatureSet } -// readFull behaves like io.ReadFull for an *fs.File. -func readFull(ctx context.Context, f *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { - var total int64 - for dst.NumBytes() > 0 { - n, err := f.Preadv(ctx, dst, offset+total) - total += n - if err == io.EOF && total != 0 { - return total, io.ErrUnexpectedEOF - } else if err != nil { - return total, err - } - dst = dst.DropFirst64(n) - } - return total, nil -} - // openPath opens args.Filename and checks that it is valid for loading. // // openPath returns an *fs.Dirent and *fs.File for args.Filename, which is not diff --git a/pkg/sentry/loader/vdso.go b/pkg/sentry/loader/vdso.go index 165869028..05a294fe6 100644 --- a/pkg/sentry/loader/vdso.go +++ b/pkg/sentry/loader/vdso.go @@ -26,10 +26,6 @@ import ( "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fs/anon" - "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/fsbridge" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/mm" "gvisor.dev/gvisor/pkg/sentry/pgalloc" @@ -37,7 +33,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" - "gvisor.dev/gvisor/pkg/waiter" ) const vdsoPrelink = 0xffffffffff700000 @@ -55,52 +50,11 @@ func (f *fileContext) Value(key interface{}) interface{} { } } -// byteReader implements fs.FileOperations for reading from a []byte source. -type byteReader struct { - fsutil.FileNoFsync `state:"nosave"` - fsutil.FileNoIoctl `state:"nosave"` - fsutil.FileNoMMap `state:"nosave"` - fsutil.FileNoSplice `state:"nosave"` - fsutil.FileNoopFlush `state:"nosave"` - fsutil.FileNoopRelease `state:"nosave"` - fsutil.FileNotDirReaddir `state:"nosave"` - fsutil.FilePipeSeek `state:"nosave"` - fsutil.FileUseInodeUnstableAttr `state:"nosave"` - waiter.AlwaysReady `state:"nosave"` - +type byteFullReader struct { data []byte } -var _ fs.FileOperations = (*byteReader)(nil) - -// newByteReaderFile creates a fake file to read data from. -// -// TODO(gvisor.dev/issue/2921): Convert to VFS2. -func newByteReaderFile(ctx context.Context, data []byte) *fs.File { - // Create a fake inode. - inode := fs.NewInode( - ctx, - &fsutil.SimpleFileInode{}, - fs.NewPseudoMountSource(ctx), - fs.StableAttr{ - Type: fs.Anonymous, - DeviceID: anon.PseudoDevice.DeviceID(), - InodeID: anon.PseudoDevice.NextIno(), - BlockSize: usermem.PageSize, - }) - - // Use the fake inode to create a fake dirent. - dirent := fs.NewTransientDirent(inode) - defer dirent.DecRef() - - // Use the fake dirent to make a fake file. - flags := fs.FileFlags{Read: true, Pread: true} - return fs.NewFile(&fileContext{Context: context.Background()}, dirent, flags, &byteReader{ - data: data, - }) -} - -func (b *byteReader) Read(ctx context.Context, file *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { +func (b *byteFullReader) ReadFull(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { if offset < 0 { return 0, syserror.EINVAL } @@ -111,10 +65,6 @@ func (b *byteReader) Read(ctx context.Context, file *fs.File, dst usermem.IOSequ return int64(n), err } -func (b *byteReader) Write(ctx context.Context, file *fs.File, src usermem.IOSequence, offset int64) (int64, error) { - panic("Write not supported") -} - // validateVDSO checks that the VDSO can be loaded by loadVDSO. // // VDSOs are special (see below). Since we are going to map the VDSO directly @@ -130,7 +80,7 @@ func (b *byteReader) Write(ctx context.Context, file *fs.File, src usermem.IOSeq // * PT_LOAD segments don't extend beyond the end of the file. // // ctx may be nil if f does not need it. -func validateVDSO(ctx context.Context, f fsbridge.File, size uint64) (elfInfo, error) { +func validateVDSO(ctx context.Context, f fullReader, size uint64) (elfInfo, error) { info, err := parseHeader(ctx, f) if err != nil { log.Infof("Unable to parse VDSO header: %v", err) @@ -248,13 +198,12 @@ func getSymbolValueFromVDSO(symbol string) (uint64, error) { // PrepareVDSO validates the system VDSO and returns a VDSO, containing the // param page for updating by the kernel. -func PrepareVDSO(ctx context.Context, mfp pgalloc.MemoryFileProvider) (*VDSO, error) { - vdsoFile := fsbridge.NewFSFile(newByteReaderFile(ctx, vdsoBin)) +func PrepareVDSO(mfp pgalloc.MemoryFileProvider) (*VDSO, error) { + vdsoFile := &byteFullReader{data: vdsoBin} // First make sure the VDSO is valid. vdsoFile does not use ctx, so a // nil context can be passed. info, err := validateVDSO(nil, vdsoFile, uint64(len(vdsoBin))) - vdsoFile.DecRef() if err != nil { return nil, err } diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index b5df1deb9..0940d2a6d 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -227,9 +227,7 @@ func New(args Args) (*Loader, error) { // Create VDSO. // // Pass k as the platform since it is savable, unlike the actual platform. - // - // FIXME(b/109889800): Use non-nil context. - vdso, err := loader.PrepareVDSO(nil, k) + vdso, err := loader.PrepareVDSO(k) if err != nil { return nil, fmt.Errorf("creating vdso: %v", err) } -- cgit v1.2.3 From c4815af9475cc4680c6d598d9c930de892c98aae Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Wed, 8 Jul 2020 17:10:35 -0700 Subject: Add shared mount hints to VFS2 Container restart test is disabled for VFS2 for now. Updates #1487 PiperOrigin-RevId: 320296401 --- pkg/sentry/fsimpl/testutil/kernel.go | 6 +- pkg/sentry/loader/loader.go | 4 +- pkg/sentry/vfs/vfs.go | 3 + runsc/boot/fs.go | 9 +- runsc/boot/loader.go | 8 +- runsc/boot/loader_test.go | 2 +- runsc/boot/vfs.go | 116 ++++++++++++++++---- runsc/container/multi_container_test.go | 184 +++++++++++++++----------------- 8 files changed, 208 insertions(+), 124 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/testutil/kernel.go b/pkg/sentry/fsimpl/testutil/kernel.go index ecc81e6bf..e743e8114 100644 --- a/pkg/sentry/fsimpl/testutil/kernel.go +++ b/pkg/sentry/fsimpl/testutil/kernel.go @@ -62,6 +62,7 @@ func Boot() (*kernel.Kernel, error) { return nil, fmt.Errorf("creating platform: %v", err) } + kernel.VFS2Enabled = true k := &kernel.Kernel{ Platform: plat, } @@ -103,11 +104,6 @@ func Boot() (*kernel.Kernel, error) { return nil, fmt.Errorf("initializing kernel: %v", err) } - kernel.VFS2Enabled = true - - if err := k.VFS().Init(); err != nil { - return nil, fmt.Errorf("VFS init: %v", err) - } k.VFS().MustRegisterFilesystemType(tmpfs.Name, &tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, AllowUserList: true, diff --git a/pkg/sentry/loader/loader.go b/pkg/sentry/loader/loader.go index e65ec4771..986c7fb4d 100644 --- a/pkg/sentry/loader/loader.go +++ b/pkg/sentry/loader/loader.go @@ -221,14 +221,14 @@ func Load(ctx context.Context, args LoadArgs, extraAuxv []arch.AuxEntry, vdso *V // Load the executable itself. loaded, ac, file, newArgv, err := loadExecutable(ctx, args) if err != nil { - return 0, nil, "", syserr.NewDynamic(fmt.Sprintf("Failed to load %s: %v", args.Filename, err), syserr.FromError(err).ToLinux()) + return 0, nil, "", syserr.NewDynamic(fmt.Sprintf("failed to load %s: %v", args.Filename, err), syserr.FromError(err).ToLinux()) } defer file.DecRef() // Load the VDSO. vdsoAddr, err := loadVDSO(ctx, args.MemoryManager, vdso, loaded) if err != nil { - return 0, nil, "", syserr.NewDynamic(fmt.Sprintf("Error loading VDSO: %v", err), syserr.FromError(err).ToLinux()) + return 0, nil, "", syserr.NewDynamic(fmt.Sprintf("error loading VDSO: %v", err), syserr.FromError(err).ToLinux()) } // Setup the heap. brk starts at the next page after the end of the diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 58c7ad778..522e27475 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -123,6 +123,9 @@ type VirtualFilesystem struct { // Init initializes a new VirtualFilesystem with no mounts or FilesystemTypes. func (vfs *VirtualFilesystem) Init() error { + if vfs.mountpoints != nil { + panic("VFS already initialized") + } vfs.mountpoints = make(map[*Dentry]map[*Mount]struct{}) vfs.devices = make(map[devTuple]*registeredDevice) vfs.anonBlockDevMinorNext = 1 diff --git a/runsc/boot/fs.go b/runsc/boot/fs.go index e83584b82..59639ba19 100644 --- a/runsc/boot/fs.go +++ b/runsc/boot/fs.go @@ -29,6 +29,7 @@ import ( _ "gvisor.dev/gvisor/pkg/sentry/fs/sys" _ "gvisor.dev/gvisor/pkg/sentry/fs/tmpfs" _ "gvisor.dev/gvisor/pkg/sentry/fs/tty" + "gvisor.dev/gvisor/pkg/sentry/vfs" specs "github.com/opencontainers/runtime-spec/specs-go" "gvisor.dev/gvisor/pkg/abi/linux" @@ -390,6 +391,10 @@ type mountHint struct { // root is the inode where the volume is mounted. For mounts with 'pod' share // the volume is mounted once and then bind mounted inside the containers. root *fs.Inode + + // vfsMount is the master mount for the volume. For mounts with 'pod' share + // the master volume is bind mounted inside the containers. + vfsMount *vfs.Mount } func (m *mountHint) setField(key, val string) error { @@ -571,9 +576,9 @@ func newContainerMounter(spec *specs.Spec, goferFDs []int, k *kernel.Kernel, hin // processHints processes annotations that container hints about how volumes // should be mounted (e.g. a volume shared between containers). It must be // called for the root container only. -func (c *containerMounter) processHints(conf *Config) error { +func (c *containerMounter) processHints(conf *Config, creds *auth.Credentials) error { if conf.VFS2 { - return nil + return c.processHintsVFS2(conf, creds) } ctx := c.k.SupervisorContext() for _, hint := range c.hints.mounts { diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index 0940d2a6d..0c0423ab2 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -298,6 +298,12 @@ func New(args Args) (*Loader, error) { return nil, fmt.Errorf("initializing kernel: %v", err) } + if kernel.VFS2Enabled { + if err := registerFilesystems(k); err != nil { + return nil, fmt.Errorf("registering filesystems: %w", err) + } + } + if err := adjustDirentCache(k); err != nil { return nil, err } @@ -559,7 +565,7 @@ func (l *Loader) run() error { l.startGoferMonitor(l.sandboxID, l.goferFDs) mntr := newContainerMounter(l.spec, l.goferFDs, l.k, l.mountHints) - if err := mntr.processHints(l.conf); err != nil { + if err := mntr.processHints(l.conf, l.rootProcArgs.Credentials); err != nil { return err } if err := setupContainerFS(ctx, l.conf, mntr, &l.rootProcArgs); err != nil { diff --git a/runsc/boot/loader_test.go b/runsc/boot/loader_test.go index e448fd773..b723e4335 100644 --- a/runsc/boot/loader_test.go +++ b/runsc/boot/loader_test.go @@ -480,7 +480,7 @@ func TestCreateMountNamespaceVFS2(t *testing.T) { defer loaderCleanup() mntr := newContainerMounter(l.spec, l.goferFDs, l.k, l.mountHints) - if err := mntr.processHints(l.conf); err != nil { + if err := mntr.processHints(l.conf, l.rootProcArgs.Credentials); err != nil { t.Fatalf("failed process hints: %v", err) } diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go index b68117867..6ee6fae04 100644 --- a/runsc/boot/vfs.go +++ b/runsc/boot/vfs.go @@ -43,7 +43,11 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) -func registerFilesystems(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials) error { +func registerFilesystems(k *kernel.Kernel) error { + ctx := k.SupervisorContext() + creds := auth.NewRootCredentials(k.RootUserNamespace()) + vfsObj := k.VFS() + vfsObj.MustRegisterFilesystemType(devpts.Name, &devpts.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserList: true, // TODO(b/29356795): Users may mount this once the terminals are in a @@ -113,9 +117,6 @@ func registerFilesystems(ctx context.Context, vfsObj *vfs.VirtualFilesystem, cre } func setupContainerVFS2(ctx context.Context, conf *Config, mntr *containerMounter, procArgs *kernel.CreateProcessArgs) error { - if err := mntr.k.VFS().Init(); err != nil { - return fmt.Errorf("failed to initialize VFS: %w", err) - } mns, err := mntr.setupVFS2(ctx, conf, procArgs) if err != nil { return fmt.Errorf("failed to setupFS: %w", err) @@ -144,10 +145,6 @@ func (c *containerMounter) setupVFS2(ctx context.Context, conf *Config, procArgs rootProcArgs.MaxSymlinkTraversals = linux.MaxSymlinkTraversals rootCtx := procArgs.NewContext(c.k) - if err := registerFilesystems(rootCtx, c.k.VFS(), rootCreds); err != nil { - return nil, fmt.Errorf("register filesystems: %w", err) - } - mns, err := c.createMountNamespaceVFS2(rootCtx, conf, rootCreds) if err != nil { return nil, fmt.Errorf("creating mount namespace: %w", err) @@ -182,8 +179,14 @@ func (c *containerMounter) mountSubmountsVFS2(ctx context.Context, conf *Config, for i := range mounts { submount := &mounts[i] log.Debugf("Mounting %q to %q, type: %s, options: %s", submount.Source, submount.Destination, submount.Type, submount.Options) - if err := c.mountSubmountVFS2(ctx, conf, mns, creds, submount); err != nil { - return err + if hint := c.hints.findMount(submount.Mount); hint != nil && hint.isSupported() { + if err := c.mountSharedSubmountVFS2(ctx, conf, mns, creds, submount.Mount, hint); err != nil { + return fmt.Errorf("mount shared mount %q to %q: %v", hint.name, submount.Destination, err) + } + } else { + if err := c.mountSubmountVFS2(ctx, conf, mns, creds, submount); err != nil { + return fmt.Errorf("mount submount %q: %w", submount.Destination, err) + } } } @@ -257,20 +260,18 @@ func (c *containerMounter) mountSubmountVFS2(ctx context.Context, conf *Config, // getMountNameAndOptionsVFS2 retrieves the fsName, opts, and useOverlay values // used for mounts. func (c *containerMounter) getMountNameAndOptionsVFS2(conf *Config, m *mountAndFD) (string, *vfs.MountOptions, error) { - var ( - fsName string - data []string - ) + fsName := m.Type + var data []string // Find filesystem name and FS specific data field. switch m.Type { case devpts.Name, devtmpfs.Name, proc.Name, sys.Name: - fsName = m.Type + // Nothing to do. + case nonefs: fsName = sys.Name - case tmpfs.Name: - fsName = m.Type + case tmpfs.Name: var err error data, err = parseAndFilterOptions(m.Options, tmpfsAllowedData...) if err != nil { @@ -279,10 +280,16 @@ func (c *containerMounter) getMountNameAndOptionsVFS2(conf *Config, m *mountAndF case bind: fsName = gofer.Name + if m.fd == 0 { + // Check that an FD was provided to fails fast. Technically FD=0 is valid, + // but unlikely to be correct in this context. + return "", nil, fmt.Errorf("9P mount requires a connection FD") + } data = p9MountData(m.fd, c.getMountAccessType(m.Mount), true /* vfs2 */) default: log.Warningf("ignoring unknown filesystem type %q", m.Type) + return "", nil, nil } opts := &vfs.MountOptions{ @@ -322,7 +329,7 @@ func (c *containerMounter) makeSyntheticMount(ctx context.Context, currentPath s } _, err := c.k.VFS().StatAt(ctx, creds, target, &vfs.StatOptions{}) if err == nil { - // Mount point exists, nothing else to do. + log.Debugf("Mount point %q already exists", currentPath) return nil } if err != syserror.ENOENT { @@ -400,3 +407,76 @@ func (c *containerMounter) mountTmpVFS2(ctx context.Context, conf *Config, creds return fmt.Errorf(`stating "/tmp" inside container: %w`, err) } } + +// processHintsVFS2 processes annotations that container hints about how volumes +// should be mounted (e.g. a volume shared between containers). It must be +// called for the root container only. +func (c *containerMounter) processHintsVFS2(conf *Config, creds *auth.Credentials) error { + ctx := c.k.SupervisorContext() + for _, hint := range c.hints.mounts { + // TODO(b/142076984): Only support tmpfs for now. Bind mounts require a + // common gofer to mount all shared volumes. + if hint.mount.Type != tmpfs.Name { + continue + } + + log.Infof("Mounting master of shared mount %q from %q type %q", hint.name, hint.mount.Source, hint.mount.Type) + mnt, err := c.mountSharedMasterVFS2(ctx, conf, hint, creds) + if err != nil { + return fmt.Errorf("mounting shared master %q: %v", hint.name, err) + } + hint.vfsMount = mnt + } + return nil +} + +// mountSharedMasterVFS2 mounts the master of a volume that is shared among +// containers in a pod. +func (c *containerMounter) mountSharedMasterVFS2(ctx context.Context, conf *Config, hint *mountHint, creds *auth.Credentials) (*vfs.Mount, error) { + // Map mount type to filesystem name, and parse out the options that we are + // capable of dealing with. + mntFD := &mountAndFD{Mount: hint.mount} + fsName, opts, err := c.getMountNameAndOptionsVFS2(conf, mntFD) + if err != nil { + return nil, err + } + if len(fsName) == 0 { + return nil, fmt.Errorf("mount type not supported %q", hint.mount.Type) + } + return c.k.VFS().MountDisconnected(ctx, creds, "", fsName, opts) +} + +// mountSharedSubmount binds mount to a previously mounted volume that is shared +// among containers in the same pod. +func (c *containerMounter) mountSharedSubmountVFS2(ctx context.Context, conf *Config, mns *vfs.MountNamespace, creds *auth.Credentials, mount specs.Mount, source *mountHint) error { + if err := source.checkCompatible(mount); err != nil { + return err + } + + _, opts, err := c.getMountNameAndOptionsVFS2(conf, &mountAndFD{Mount: mount}) + if err != nil { + return err + } + newMnt, err := c.k.VFS().NewDisconnectedMount(source.vfsMount.Filesystem(), source.vfsMount.Root(), opts) + if err != nil { + return err + } + defer newMnt.DecRef() + + root := mns.Root() + defer root.DecRef() + if err := c.makeSyntheticMount(ctx, mount.Destination, root, creds); err != nil { + return err + } + + target := &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(mount.Destination), + } + if err := c.k.VFS().ConnectMountAt(ctx, creds, newMnt, target); err != nil { + return err + } + log.Infof("Mounted %q type shared bind to %q", mount.Destination, source.name) + return nil +} diff --git a/runsc/container/multi_container_test.go b/runsc/container/multi_container_test.go index a27a01942..e189648f4 100644 --- a/runsc/container/multi_container_test.go +++ b/runsc/container/multi_container_test.go @@ -100,19 +100,20 @@ type execDesc struct { c *Container cmd []string want int - desc string + name string } -func execMany(execs []execDesc) error { +func execMany(t *testing.T, execs []execDesc) { for _, exec := range execs { - args := &control.ExecArgs{Argv: exec.cmd} - if ws, err := exec.c.executeSync(args); err != nil { - return fmt.Errorf("error executing %+v: %v", args, err) - } else if ws.ExitStatus() != exec.want { - return fmt.Errorf("%q: exec %q got exit status: %d, want: %d", exec.desc, exec.cmd, ws.ExitStatus(), exec.want) - } + t.Run(exec.name, func(t *testing.T) { + args := &control.ExecArgs{Argv: exec.cmd} + if ws, err := exec.c.executeSync(args); err != nil { + t.Errorf("error executing %+v: %v", args, err) + } else if ws.ExitStatus() != exec.want { + t.Errorf("%q: exec %q got exit status: %d, want: %d", exec.name, exec.cmd, ws.ExitStatus(), exec.want) + } + }) } - return nil } func createSharedMount(mount specs.Mount, name string, pod ...*specs.Spec) { @@ -1072,7 +1073,7 @@ func TestMultiContainerContainerDestroyStress(t *testing.T) { // Test that pod shared mounts are properly mounted in 2 containers and that // changes from one container is reflected in the other. func TestMultiContainerSharedMount(t *testing.T) { - for name, conf := range configs(t, all...) { + for name, conf := range configsWithVFS2(t, all...) { t.Run(name, func(t *testing.T) { rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { @@ -1110,84 +1111,82 @@ func TestMultiContainerSharedMount(t *testing.T) { { c: containers[0], cmd: []string{"/usr/bin/test", "-d", mnt0.Destination}, - desc: "directory is mounted in container0", + name: "directory is mounted in container0", }, { c: containers[1], cmd: []string{"/usr/bin/test", "-d", mnt1.Destination}, - desc: "directory is mounted in container1", + name: "directory is mounted in container1", }, { c: containers[0], - cmd: []string{"/usr/bin/touch", file0}, - desc: "create file in container0", + cmd: []string{"/bin/touch", file0}, + name: "create file in container0", }, { c: containers[0], cmd: []string{"/usr/bin/test", "-f", file0}, - desc: "file appears in container0", + name: "file appears in container0", }, { c: containers[1], cmd: []string{"/usr/bin/test", "-f", file1}, - desc: "file appears in container1", + name: "file appears in container1", }, { c: containers[1], cmd: []string{"/bin/rm", file1}, - desc: "file removed from container1", + name: "remove file from container1", }, { c: containers[0], cmd: []string{"/usr/bin/test", "!", "-f", file0}, - desc: "file removed from container0", + name: "file removed from container0", }, { c: containers[1], cmd: []string{"/usr/bin/test", "!", "-f", file1}, - desc: "file removed from container1", + name: "file removed from container1", }, { c: containers[1], cmd: []string{"/bin/mkdir", file1}, - desc: "create directory in container1", + name: "create directory in container1", }, { c: containers[0], cmd: []string{"/usr/bin/test", "-d", file0}, - desc: "dir appears in container0", + name: "dir appears in container0", }, { c: containers[1], cmd: []string{"/usr/bin/test", "-d", file1}, - desc: "dir appears in container1", + name: "dir appears in container1", }, { c: containers[0], cmd: []string{"/bin/rmdir", file0}, - desc: "create directory in container0", + name: "remove directory from container0", }, { c: containers[0], cmd: []string{"/usr/bin/test", "!", "-d", file0}, - desc: "dir removed from container0", + name: "dir removed from container0", }, { c: containers[1], cmd: []string{"/usr/bin/test", "!", "-d", file1}, - desc: "dir removed from container1", + name: "dir removed from container1", }, } - if err := execMany(execs); err != nil { - t.Fatal(err.Error()) - } + execMany(t, execs) }) } } // Test that pod mounts are mounted as readonly when requested. func TestMultiContainerSharedMountReadonly(t *testing.T) { - for name, conf := range configs(t, all...) { + for name, conf := range configsWithVFS2(t, all...) { t.Run(name, func(t *testing.T) { rootDir, cleanup, err := testutil.SetupRootDir() if err != nil { @@ -1225,35 +1224,34 @@ func TestMultiContainerSharedMountReadonly(t *testing.T) { { c: containers[0], cmd: []string{"/usr/bin/test", "-d", mnt0.Destination}, - desc: "directory is mounted in container0", + name: "directory is mounted in container0", }, { c: containers[1], cmd: []string{"/usr/bin/test", "-d", mnt1.Destination}, - desc: "directory is mounted in container1", + name: "directory is mounted in container1", }, { c: containers[0], - cmd: []string{"/usr/bin/touch", file0}, + cmd: []string{"/bin/touch", file0}, want: 1, - desc: "fails to write to container0", + name: "fails to write to container0", }, { c: containers[1], - cmd: []string{"/usr/bin/touch", file1}, + cmd: []string{"/bin/touch", file1}, want: 1, - desc: "fails to write to container1", + name: "fails to write to container1", }, } - if err := execMany(execs); err != nil { - t.Fatal(err.Error()) - } + execMany(t, execs) }) } } // Test that shared pod mounts continue to work after container is restarted. func TestMultiContainerSharedMountRestart(t *testing.T) { + //TODO(gvisor.dev/issue/1487): This is failing with VFS2. for name, conf := range configs(t, all...) { t.Run(name, func(t *testing.T) { rootDir, cleanup, err := testutil.SetupRootDir() @@ -1291,23 +1289,21 @@ func TestMultiContainerSharedMountRestart(t *testing.T) { execs := []execDesc{ { c: containers[0], - cmd: []string{"/usr/bin/touch", file0}, - desc: "create file in container0", + cmd: []string{"/bin/touch", file0}, + name: "create file in container0", }, { c: containers[0], cmd: []string{"/usr/bin/test", "-f", file0}, - desc: "file appears in container0", + name: "file appears in container0", }, { c: containers[1], cmd: []string{"/usr/bin/test", "-f", file1}, - desc: "file appears in container1", + name: "file appears in container1", }, } - if err := execMany(execs); err != nil { - t.Fatal(err.Error()) - } + execMany(t, execs) containers[1].Destroy() @@ -1334,32 +1330,30 @@ func TestMultiContainerSharedMountRestart(t *testing.T) { { c: containers[0], cmd: []string{"/usr/bin/test", "-f", file0}, - desc: "file is still in container0", + name: "file is still in container0", }, { c: containers[1], cmd: []string{"/usr/bin/test", "-f", file1}, - desc: "file is still in container1", + name: "file is still in container1", }, { c: containers[1], cmd: []string{"/bin/rm", file1}, - desc: "file removed from container1", + name: "file removed from container1", }, { c: containers[0], cmd: []string{"/usr/bin/test", "!", "-f", file0}, - desc: "file removed from container0", + name: "file removed from container0", }, { c: containers[1], cmd: []string{"/usr/bin/test", "!", "-f", file1}, - desc: "file removed from container1", + name: "file removed from container1", }, } - if err := execMany(execs); err != nil { - t.Fatal(err.Error()) - } + execMany(t, execs) }) } } @@ -1367,53 +1361,53 @@ func TestMultiContainerSharedMountRestart(t *testing.T) { // Test that unsupported pod mounts options are ignored when matching master and // slave mounts. func TestMultiContainerSharedMountUnsupportedOptions(t *testing.T) { - rootDir, cleanup, err := testutil.SetupRootDir() - if err != nil { - t.Fatalf("error creating root dir: %v", err) - } - defer cleanup() - - conf := testutil.TestConfig(t) - conf.RootDir = rootDir + for name, conf := range configsWithVFS2(t, all...) { + t.Run(name, func(t *testing.T) { + rootDir, cleanup, err := testutil.SetupRootDir() + if err != nil { + t.Fatalf("error creating root dir: %v", err) + } + defer cleanup() + conf.RootDir = rootDir - // Setup the containers. - sleep := []string{"/bin/sleep", "100"} - podSpec, ids := createSpecs(sleep, sleep) - mnt0 := specs.Mount{ - Destination: "/mydir/test", - Source: "/some/dir", - Type: "tmpfs", - Options: []string{"rw", "rbind", "relatime"}, - } - podSpec[0].Mounts = append(podSpec[0].Mounts, mnt0) + // Setup the containers. + sleep := []string{"/bin/sleep", "100"} + podSpec, ids := createSpecs(sleep, sleep) + mnt0 := specs.Mount{ + Destination: "/mydir/test", + Source: "/some/dir", + Type: "tmpfs", + Options: []string{"rw", "rbind", "relatime"}, + } + podSpec[0].Mounts = append(podSpec[0].Mounts, mnt0) - mnt1 := mnt0 - mnt1.Destination = "/mydir2/test2" - mnt1.Options = []string{"rw", "nosuid"} - podSpec[1].Mounts = append(podSpec[1].Mounts, mnt1) + mnt1 := mnt0 + mnt1.Destination = "/mydir2/test2" + mnt1.Options = []string{"rw", "nosuid"} + podSpec[1].Mounts = append(podSpec[1].Mounts, mnt1) - createSharedMount(mnt0, "test-mount", podSpec...) + createSharedMount(mnt0, "test-mount", podSpec...) - containers, cleanup, err := startContainers(conf, podSpec, ids) - if err != nil { - t.Fatalf("error starting containers: %v", err) - } - defer cleanup() + containers, cleanup, err := startContainers(conf, podSpec, ids) + if err != nil { + t.Fatalf("error starting containers: %v", err) + } + defer cleanup() - execs := []execDesc{ - { - c: containers[0], - cmd: []string{"/usr/bin/test", "-d", mnt0.Destination}, - desc: "directory is mounted in container0", - }, - { - c: containers[1], - cmd: []string{"/usr/bin/test", "-d", mnt1.Destination}, - desc: "directory is mounted in container1", - }, - } - if err := execMany(execs); err != nil { - t.Fatal(err.Error()) + execs := []execDesc{ + { + c: containers[0], + cmd: []string{"/usr/bin/test", "-d", mnt0.Destination}, + name: "directory is mounted in container0", + }, + { + c: containers[1], + cmd: []string{"/usr/bin/test", "-d", mnt1.Destination}, + name: "directory is mounted in container1", + }, + } + execMany(t, execs) + }) } } -- cgit v1.2.3 From abffebde7be2dcdb4564e45f845d7c150ced0ccb Mon Sep 17 00:00:00 2001 From: Ridwan Sharif Date: Tue, 7 Jul 2020 21:48:25 -0400 Subject: Gate FUSE behind a runsc flag This change gates all FUSE commands (by gating /dev/fuse) behind a runsc flag. In order to use FUSE commands, use the --fuse flag with the --vfs2 flag. Check if FUSE is enabled by running dmesg in the sandbox. --- pkg/sentry/fsimpl/fuse/BUILD | 1 + pkg/sentry/fsimpl/fuse/dev.go | 5 +++++ pkg/sentry/kernel/kernel.go | 4 ++++ pkg/sentry/kernel/syslog.go | 9 +++++++++ runsc/boot/config.go | 7 +++++++ runsc/boot/loader.go | 4 ++++ runsc/boot/vfs.go | 14 ++++++++++---- runsc/main.go | 2 ++ test/runner/defs.bzl | 20 +++++++++++++++++++- test/runner/runner.go | 10 ++++++++++ test/syscalls/BUILD | 1 + test/syscalls/linux/dev.cc | 2 +- test/util/test_util.cc | 6 ++++++ test/util/test_util.h | 1 + 14 files changed, 80 insertions(+), 6 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/fuse/BUILD b/pkg/sentry/fsimpl/fuse/BUILD index 41567967d..3e00c2abb 100644 --- a/pkg/sentry/fsimpl/fuse/BUILD +++ b/pkg/sentry/fsimpl/fuse/BUILD @@ -12,6 +12,7 @@ go_library( "//pkg/abi/linux", "//pkg/context", "//pkg/sentry/fsimpl/devtmpfs", + "//pkg/sentry/kernel", "//pkg/sentry/vfs", "//pkg/syserror", "//pkg/usermem", diff --git a/pkg/sentry/fsimpl/fuse/dev.go b/pkg/sentry/fsimpl/fuse/dev.go index f6a67d005..dc33268af 100644 --- a/pkg/sentry/fsimpl/fuse/dev.go +++ b/pkg/sentry/fsimpl/fuse/dev.go @@ -18,6 +18,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fsimpl/devtmpfs" + "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -30,6 +31,10 @@ type fuseDevice struct{} // Open implements vfs.Device.Open. func (fuseDevice) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + if !kernel.FUSEEnabled { + return nil, syserror.ENOENT + } + var fd DeviceFD if err := fd.vfsfd.Init(&fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{ UseDentryMetadata: true, diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 2177b785a..240cd6fe0 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -81,6 +81,10 @@ import ( // easy access everywhere. To be removed once VFS2 becomes the default. var VFS2Enabled = false +// FUSEEnabled is set to true when FUSE is enabled. Added as a global for allow +// easy access everywhere. To be removed once FUSE is completed. +var FUSEEnabled = false + // Kernel represents an emulated Linux kernel. It must be initialized by calling // Init() or LoadFrom(). // diff --git a/pkg/sentry/kernel/syslog.go b/pkg/sentry/kernel/syslog.go index 4607cde2f..a83ce219c 100644 --- a/pkg/sentry/kernel/syslog.go +++ b/pkg/sentry/kernel/syslog.go @@ -98,6 +98,15 @@ func (s *syslog) Log() []byte { s.msg = append(s.msg, []byte(fmt.Sprintf(format, time, selectMessage()))...) } + if VFS2Enabled { + time += rand.Float64() / 2 + s.msg = append(s.msg, []byte(fmt.Sprintf(format, time, "Setting up VFS2..."))...) + if FUSEEnabled { + time += rand.Float64() / 2 + s.msg = append(s.msg, []byte(fmt.Sprintf(format, time, "Setting up FUSE..."))...) + } + } + time += rand.Float64() / 2 s.msg = append(s.msg, []byte(fmt.Sprintf(format, time, "Ready!"))...) diff --git a/runsc/boot/config.go b/runsc/boot/config.go index bb01b8fb5..80da8b3e6 100644 --- a/runsc/boot/config.go +++ b/runsc/boot/config.go @@ -274,6 +274,9 @@ type Config struct { // Enables VFS2 (not plumbled through yet). VFS2 bool + + // Enables FUSE usage (not plumbled through yet). + FUSE bool } // ToFlags returns a slice of flags that correspond to the given Config. @@ -325,5 +328,9 @@ func (c *Config) ToFlags() []string { f = append(f, "--vfs2=true") } + if c.FUSE { + f = append(f, "--fuse=true") + } + return f } diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index 0c0423ab2..93ac7ec41 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -205,6 +205,10 @@ func New(args Args) (*Loader, error) { // Is this a VFSv2 kernel? if args.Conf.VFS2 { kernel.VFS2Enabled = true + if args.Conf.FUSE { + kernel.FUSEEnabled = true + } + vfs2.Override() } diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go index 6ee6fae04..56f4ba15d 100644 --- a/runsc/boot/vfs.go +++ b/runsc/boot/vfs.go @@ -86,9 +86,12 @@ func registerFilesystems(k *kernel.Kernel) error { return fmt.Errorf("registering ttydev: %w", err) } - if err := fuse.Register(vfsObj); err != nil { - return fmt.Errorf("registering fusedev: %w", err) + if kernel.FUSEEnabled { + if err := fuse.Register(vfsObj); err != nil { + return fmt.Errorf("registering fusedev: %w", err) + } } + if err := tundev.Register(vfsObj); err != nil { return fmt.Errorf("registering tundev: %v", err) } @@ -110,8 +113,11 @@ func registerFilesystems(k *kernel.Kernel) error { if err := tundev.CreateDevtmpfsFiles(ctx, a); err != nil { return fmt.Errorf("creating tundev devtmpfs files: %v", err) } - if err := fuse.CreateDevtmpfsFile(ctx, a); err != nil { - return fmt.Errorf("creating fusedev devtmpfs files: %w", err) + + if kernel.FUSEEnabled { + if err := fuse.CreateDevtmpfsFile(ctx, a); err != nil { + return fmt.Errorf("creating fusedev devtmpfs files: %w", err) + } } return nil } diff --git a/runsc/main.go b/runsc/main.go index c9f47c579..69cb505fa 100644 --- a/runsc/main.go +++ b/runsc/main.go @@ -88,6 +88,7 @@ var ( referenceLeakMode = flag.String("ref-leak-mode", "disabled", "sets reference leak check mode: disabled (default), log-names, log-traces.") cpuNumFromQuota = flag.Bool("cpu-num-from-quota", false, "set cpu number to cpu quota (least integer greater or equal to quota value, but not less than 2)") vfs2Enabled = flag.Bool("vfs2", false, "TEST ONLY; use while VFSv2 is landing. This uses the new experimental VFS layer.") + fuseEnabled = flag.Bool("fuse", false, "TEST ONLY; use while FUSE in VFSv2 is landing. This allows the use of the new experimental FUSE filesystem.") // Test flags, not to be used outside tests, ever. testOnlyAllowRunAsCurrentUserWithoutChroot = flag.Bool("TESTONLY-unsafe-nonroot", false, "TEST ONLY; do not ever use! This skips many security measures that isolate the host from the sandbox.") @@ -242,6 +243,7 @@ func main() { OverlayfsStaleRead: *overlayfsStaleRead, CPUNumFromQuota: *cpuNumFromQuota, VFS2: *vfs2Enabled, + FUSE: *fuseEnabled, QDisc: queueingDiscipline, TestOnlyAllowRunAsCurrentUserWithoutChroot: *testOnlyAllowRunAsCurrentUserWithoutChroot, TestOnlyTestNameEnv: *testOnlyTestNameEnv, diff --git a/test/runner/defs.bzl b/test/runner/defs.bzl index 921e499be..600cb5192 100644 --- a/test/runner/defs.bzl +++ b/test/runner/defs.bzl @@ -61,7 +61,8 @@ def _syscall_test( file_access = "exclusive", overlay = False, add_uds_tree = False, - vfs2 = False): + vfs2 = False, + fuse = False): # Prepend "runsc" to non-native platform names. full_platform = platform if platform == "native" else "runsc_" + platform @@ -73,6 +74,8 @@ def _syscall_test( name += "_overlay" if vfs2: name += "_vfs2" + if fuse: + name += "_fuse" if network != "none": name += "_" + network + "net" @@ -107,6 +110,7 @@ def _syscall_test( "--overlay=" + str(overlay), "--add-uds-tree=" + str(add_uds_tree), "--vfs2=" + str(vfs2), + "--fuse=" + str(fuse), ] # Call the rule above. @@ -129,6 +133,7 @@ def syscall_test( add_uds_tree = False, add_hostinet = False, vfs2 = False, + fuse = False, tags = None): """syscall_test is a macro that will create targets for all platforms. @@ -188,6 +193,19 @@ def syscall_test( vfs2 = True, ) + if vfs2 and fuse: + _syscall_test( + test = test, + shard_count = shard_count, + size = size, + platform = default_platform, + use_tmpfs = use_tmpfs, + add_uds_tree = add_uds_tree, + tags = platforms[default_platform] + vfs2_tags, + vfs2 = True, + fuse = True, + ) + # TODO(gvisor.dev/issue/1487): Enable VFS2 overlay tests. if add_overlay: _syscall_test( diff --git a/test/runner/runner.go b/test/runner/runner.go index 5456e46a6..2296f3a46 100644 --- a/test/runner/runner.go +++ b/test/runner/runner.go @@ -47,6 +47,7 @@ var ( fileAccess = flag.String("file-access", "exclusive", "mounts root in exclusive or shared mode") overlay = flag.Bool("overlay", false, "wrap filesystem mounts with writable tmpfs overlay") vfs2 = flag.Bool("vfs2", false, "enable VFS2") + fuse = flag.Bool("fuse", false, "enable FUSE") parallel = flag.Bool("parallel", false, "run tests in parallel") runscPath = flag.String("runsc", "", "path to runsc binary") @@ -149,6 +150,9 @@ func runRunsc(tc gtest.TestCase, spec *specs.Spec) error { } if *vfs2 { args = append(args, "-vfs2") + if *fuse { + args = append(args, "-fuse") + } } if *debug { args = append(args, "-debug", "-log-packets=true") @@ -358,6 +362,12 @@ func runTestCaseRunsc(testBin string, tc gtest.TestCase, t *testing.T) { vfsVar := "GVISOR_VFS" if *vfs2 { env = append(env, vfsVar+"=VFS2") + fuseVar := "FUSE_ENABLED" + if *fuse { + env = append(env, fuseVar+"=TRUE") + } else { + env = append(env, fuseVar+"=FALSE") + } } else { env = append(env, vfsVar+"=VFS1") } diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 28ef55945..c06a75ada 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -146,6 +146,7 @@ syscall_test( ) syscall_test( + fuse = "True", test = "//test/syscalls/linux:dev_test", vfs2 = "True", ) diff --git a/test/syscalls/linux/dev.cc b/test/syscalls/linux/dev.cc index 3c88c4cbd..6fa16208e 100644 --- a/test/syscalls/linux/dev.cc +++ b/test/syscalls/linux/dev.cc @@ -156,7 +156,7 @@ TEST(DevTest, TTYExists) { TEST(DevTest, OpenDevFuse) { // Note(gvisor.dev/issue/3076) This won't work in the sentry until the new // device registration is complete. - SKIP_IF(IsRunningWithVFS1() || IsRunningOnGvisor()); + SKIP_IF(IsRunningWithVFS1() || IsRunningOnGvisor() || !IsFUSEEnabled()); ASSERT_NO_ERRNO_AND_VALUE(Open("/dev/fuse", O_RDONLY)); } diff --git a/test/util/test_util.cc b/test/util/test_util.cc index 8a037f45f..d0c1d6426 100644 --- a/test/util/test_util.cc +++ b/test/util/test_util.cc @@ -42,6 +42,7 @@ namespace testing { constexpr char kGvisorNetwork[] = "GVISOR_NETWORK"; constexpr char kGvisorVfs[] = "GVISOR_VFS"; +constexpr char kFuseEnabled[] = "FUSE_ENABLED"; bool IsRunningOnGvisor() { return GvisorPlatform() != Platform::kNative; } @@ -68,6 +69,11 @@ bool IsRunningWithVFS1() { return strcmp(env, "VFS1") == 0; } +bool IsFUSEEnabled() { + const char* env = getenv(kFuseEnabled); + return env && strcmp(env, "TRUE") == 0; +} + // Inline cpuid instruction. Preserve %ebx/%rbx register. In PIC compilations // %ebx contains the address of the global offset table. %rbx is occasionally // used to address stack variables in presence of dynamic allocas. diff --git a/test/util/test_util.h b/test/util/test_util.h index 109078fc7..89ac575bd 100644 --- a/test/util/test_util.h +++ b/test/util/test_util.h @@ -225,6 +225,7 @@ const std::string GvisorPlatform(); bool IsRunningWithHostinet(); // TODO(gvisor.dev/issue/1624): Delete once VFS1 is gone. bool IsRunningWithVFS1(); +bool IsFUSEEnabled(); #ifdef __linux__ void SetupGvisorDeathTest(); -- cgit v1.2.3 From 5946f111827fa4e342a2e6e9c043c198d2e5cb03 Mon Sep 17 00:00:00 2001 From: Bhasker Hariharan Date: Thu, 9 Jul 2020 16:24:43 -0700 Subject: Add support for IP_HDRINCL IP option for raw sockets. Updates #2746 Fixes #3158 PiperOrigin-RevId: 320497190 --- pkg/sentry/socket/netstack/netstack.go | 11 ++++- pkg/tcpip/tcpip.go | 5 +++ pkg/tcpip/transport/raw/endpoint.go | 34 +++++++++++---- test/syscalls/linux/raw_socket_hdrincl.cc | 70 ++++++++++++++++++++++++++++++- 4 files changed, 110 insertions(+), 10 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index e7d2c83d7..3b248a953 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -2112,13 +2112,22 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s } return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.ReceiveIPPacketInfoOption, v != 0)) + case linux.IP_HDRINCL: + if len(optVal) == 0 { + return nil + } + v, err := parseIntOrChar(optVal) + if err != nil { + return err + } + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.IPHdrIncludedOption, v != 0)) + case linux.IP_ADD_SOURCE_MEMBERSHIP, linux.IP_BIND_ADDRESS_NO_PORT, linux.IP_BLOCK_SOURCE, linux.IP_CHECKSUM, linux.IP_DROP_SOURCE_MEMBERSHIP, linux.IP_FREEBIND, - linux.IP_HDRINCL, linux.IP_IPSEC_POLICY, linux.IP_MINTTL, linux.IP_MSFILTER, diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 2f9872dc6..25534a10d 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -648,6 +648,11 @@ const ( // whether an IPv6 socket is to be restricted to sending and receiving // IPv6 packets only. V6OnlyOption + + // IPHdrIncludedOption is used by SetSockOpt to indicate for a raw + // endpoint that all packets being written have an IP header and the + // endpoint should not attach an IP header. + IPHdrIncludedOption ) // SockOptInt represents socket options which values have the int type. diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index 766c7648e..5b6e7d102 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -63,6 +63,7 @@ type endpoint struct { stack *stack.Stack `state:"manual"` waiterQueue *waiter.Queue associated bool + hdrIncluded bool // The following fields are used to manage the receive queue and are // protected by rcvMu. @@ -108,6 +109,7 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProt rcvBufSizeMax: 32 * 1024, sndBufSizeMax: 32 * 1024, associated: associated, + hdrIncluded: !associated, } // Override with stack defaults. @@ -182,10 +184,6 @@ func (e *endpoint) SetOwner(owner tcpip.PacketOwner) { // Read implements tcpip.Endpoint.Read. func (e *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) { - if !e.associated { - return buffer.View{}, tcpip.ControlMessages{}, tcpip.ErrInvalidOptionValue - } - e.rcvMu.Lock() // If there's no data to read, return that read would block or that the @@ -263,7 +261,7 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c // If this is an unassociated socket and callee provided a nonzero // destination address, route using that address. - if !e.associated { + if e.hdrIncluded { ip := header.IPv4(payloadBytes) if !ip.IsValid(len(payloadBytes)) { e.mu.RUnlock() @@ -353,7 +351,7 @@ func (e *endpoint) finishWrite(payloadBytes []byte, route *stack.Route) (int64, } } - if !e.associated { + if e.hdrIncluded { if err := route.WriteHeaderIncludedPacket(&stack.PacketBuffer{ Data: buffer.View(payloadBytes).ToVectorisedView(), }); err != nil { @@ -513,6 +511,13 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { // SetSockOptBool implements tcpip.Endpoint.SetSockOptBool. func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { + switch opt { + case tcpip.IPHdrIncludedOption: + e.mu.Lock() + e.hdrIncluded = v + e.mu.Unlock() + return nil + } return tcpip.ErrUnknownProtocolOption } @@ -577,6 +582,12 @@ func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { case tcpip.KeepaliveEnabledOption: return false, nil + case tcpip.IPHdrIncludedOption: + e.mu.Lock() + v := e.hdrIncluded + e.mu.Unlock() + return v, nil + default: return false, tcpip.ErrUnknownProtocolOption } @@ -616,8 +627,15 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { func (e *endpoint) HandlePacket(route *stack.Route, pkt *stack.PacketBuffer) { e.rcvMu.Lock() - // Drop the packet if our buffer is currently full. - if e.rcvClosed { + // Drop the packet if our buffer is currently full or if this is an unassociated + // endpoint (i.e endpoint created w/ IPPROTO_RAW). Such endpoints are send only + // See: https://man7.org/linux/man-pages/man7/raw.7.html + // + // An IPPROTO_RAW socket is send only. If you really want to receive + // all IP packets, use a packet(7) socket with the ETH_P_IP protocol. + // Note that packet sockets don't reassemble IP fragments, unlike raw + // sockets. + if e.rcvClosed || !e.associated { e.rcvMu.Unlock() e.stack.Stats().DroppedPackets.Increment() e.stats.ReceiveErrors.ClosedReceiver.Increment() diff --git a/test/syscalls/linux/raw_socket_hdrincl.cc b/test/syscalls/linux/raw_socket_hdrincl.cc index 16cfc1d75..5bb14d57c 100644 --- a/test/syscalls/linux/raw_socket_hdrincl.cc +++ b/test/syscalls/linux/raw_socket_hdrincl.cc @@ -167,7 +167,7 @@ TEST_F(RawHDRINCL, NotReadable) { // nothing to be read. char buf[117]; ASSERT_THAT(RetryEINTR(recv)(socket_, buf, sizeof(buf), MSG_DONTWAIT), - SyscallFailsWithErrno(EINVAL)); + SyscallFailsWithErrno(EAGAIN)); } // Test that we can connect() to a valid IP (loopback). @@ -332,6 +332,74 @@ TEST_F(RawHDRINCL, SendAndReceiveDifferentAddress) { EXPECT_EQ(absl::gbswap_32(recv_iphdr.daddr), INADDR_LOOPBACK); } +// Send and receive a packet w/ the IP_HDRINCL option set. +TEST_F(RawHDRINCL, SendAndReceiveIPHdrIncl) { + int port = 40000; + if (!IsRunningOnGvisor()) { + port = static_cast(ASSERT_NO_ERRNO_AND_VALUE( + PortAvailable(0, AddressFamily::kIpv4, SocketType::kUdp, false))); + } + + FileDescriptor recv_sock = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_RAW, IPPROTO_UDP)); + + FileDescriptor send_sock = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_RAW, IPPROTO_UDP)); + + // Enable IP_HDRINCL option so that we can build and send w/ an IP + // header. + constexpr int kSockOptOn = 1; + ASSERT_THAT(setsockopt(send_sock.get(), SOL_IP, IP_HDRINCL, &kSockOptOn, + sizeof(kSockOptOn)), + SyscallSucceeds()); + // This is not strictly required but we do it to make sure that setting + // IP_HDRINCL on a non IPPROTO_RAW socket does not prevent it from receiving + // packets. + ASSERT_THAT(setsockopt(recv_sock.get(), SOL_IP, IP_HDRINCL, &kSockOptOn, + sizeof(kSockOptOn)), + SyscallSucceeds()); + + // Construct a packet with an IP header, UDP header, and payload. + constexpr char kPayload[] = "toto"; + char packet[sizeof(struct iphdr) + sizeof(struct udphdr) + sizeof(kPayload)]; + ASSERT_TRUE( + FillPacket(packet, sizeof(packet), port, kPayload, sizeof(kPayload))); + + socklen_t addrlen = sizeof(addr_); + ASSERT_NO_FATAL_FAILURE(sendto(send_sock.get(), &packet, sizeof(packet), 0, + reinterpret_cast(&addr_), + addrlen)); + + // Receive the payload. + char recv_buf[sizeof(packet)]; + struct sockaddr_in src; + socklen_t src_size = sizeof(src); + ASSERT_THAT(recvfrom(recv_sock.get(), recv_buf, sizeof(recv_buf), 0, + reinterpret_cast(&src), &src_size), + SyscallSucceedsWithValue(sizeof(packet))); + EXPECT_EQ( + memcmp(kPayload, recv_buf + sizeof(struct iphdr) + sizeof(struct udphdr), + sizeof(kPayload)), + 0); + // The network stack should have set the source address. + EXPECT_EQ(src.sin_family, AF_INET); + EXPECT_EQ(absl::gbswap_32(src.sin_addr.s_addr), INADDR_LOOPBACK); + struct iphdr iphdr = {}; + memcpy(&iphdr, recv_buf, sizeof(iphdr)); + EXPECT_NE(iphdr.id, 0); + + // Also verify that the packet we just sent was not delivered to the + // IPPROTO_RAW socket. + { + char recv_buf[sizeof(packet)]; + struct sockaddr_in src; + socklen_t src_size = sizeof(src); + ASSERT_THAT(recvfrom(socket_, recv_buf, sizeof(recv_buf), MSG_DONTWAIT, + reinterpret_cast(&src), &src_size), + SyscallFailsWithErrno(EAGAIN)); + } +} + } // namespace } // namespace testing -- cgit v1.2.3 From 5df3a8fedef7e54550d4c6b7172e25216600ee9f Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Thu, 9 Jul 2020 22:33:53 -0700 Subject: Discard multicast UDP source address. RFC-1122 (and others) specify that UDP should not receive datagrams that have a source address that is a multicast address. Packets should never be received FROM a multicast address. See also, RFC 768: 'User Datagram Protocol' J. Postel, ISI, 28 August 1980 A UDP datagram received with an invalid IP source address (e.g., a broadcast or multicast address) must be discarded by UDP or by the IP layer (see rfc 1122 Section 3.2.1.3). This CL does not address TCP or broadcast which is more complicated. Also adds a test for both ipv6 and ipv4 UDP. Fixes #3154 PiperOrigin-RevId: 320547674 --- pkg/sentry/socket/netstack/netstack.go | 1 + pkg/tcpip/tcpip.go | 5 +- pkg/tcpip/transport/udp/endpoint.go | 11 +++- pkg/tcpip/transport/udp/udp_test.go | 104 ++++++++++++++++++++++++++++----- 4 files changed, 103 insertions(+), 18 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 3b248a953..5a3cedd7c 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -192,6 +192,7 @@ var Metrics = tcpip.Stats{ PacketsSent: mustCreateMetric("/netstack/udp/packets_sent", "Number of UDP datagrams sent."), PacketSendErrors: mustCreateMetric("/netstack/udp/packet_send_errors", "Number of UDP datagrams failed to be sent."), ChecksumErrors: mustCreateMetric("/netstack/udp/checksum_errors", "Number of UDP datagrams dropped due to bad checksums."), + InvalidSourceAddress: mustCreateMetric("/netstack/udp/invalid_source", "Number of UDP datagrams dropped due to invalid source address."), }, } diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 25534a10d..cf7291d09 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -782,7 +782,7 @@ type CongestionControlOption string // control algorithms. type AvailableCongestionControlOption string -// buffer moderation. +// ModerateReceiveBufferOption is used by buffer moderation. type ModerateReceiveBufferOption bool // TCPLingerTimeoutOption is used by SetSockOpt/GetSockOpt to set/get the @@ -1244,6 +1244,9 @@ type UDPStats struct { // ChecksumErrors is the number of datagrams dropped due to bad checksums. ChecksumErrors *StatCounter + + // InvalidSourceAddress is the number of invalid sourced datagrams dropped. + InvalidSourceAddress *StatCounter } // Stats holds statistics about the networking stack. diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 0584ec8dc..4e9e114a9 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -1377,6 +1377,15 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk return } + // Never receive from a multicast address. + if header.IsV4MulticastAddress(id.RemoteAddress) || + header.IsV6MulticastAddress(id.RemoteAddress) { + e.stack.Stats().UDP.InvalidSourceAddress.Increment() + e.stack.Stats().IP.InvalidSourceAddressesReceived.Increment() + e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() + return + } + // Verify checksum unless RX checksum offload is enabled. // On IPv4, UDP checksum is optional, and a zero value means // the transmitter omitted the checksum generation (RFC768). @@ -1395,10 +1404,10 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk } } - e.rcvMu.Lock() e.stack.Stats().UDP.PacketsReceived.Increment() e.stats.PacketsReceived.Increment() + e.rcvMu.Lock() // Drop the packet if our buffer is currently full. if !e.rcvReady || e.rcvClosed { e.rcvMu.Unlock() diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index 91ba031fa..90781cf49 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -83,16 +83,18 @@ type header4Tuple struct { type testFlow int const ( - unicastV4 testFlow = iota // V4 unicast on a V4 socket - unicastV4in6 // V4-mapped unicast on a V6-dual socket - unicastV6 // V6 unicast on a V6 socket - unicastV6Only // V6 unicast on a V6-only socket - multicastV4 // V4 multicast on a V4 socket - multicastV4in6 // V4-mapped multicast on a V6-dual socket - multicastV6 // V6 multicast on a V6 socket - multicastV6Only // V6 multicast on a V6-only socket - broadcast // V4 broadcast on a V4 socket - broadcastIn6 // V4-mapped broadcast on a V6-dual socket + unicastV4 testFlow = iota // V4 unicast on a V4 socket + unicastV4in6 // V4-mapped unicast on a V6-dual socket + unicastV6 // V6 unicast on a V6 socket + unicastV6Only // V6 unicast on a V6-only socket + multicastV4 // V4 multicast on a V4 socket + multicastV4in6 // V4-mapped multicast on a V6-dual socket + multicastV6 // V6 multicast on a V6 socket + multicastV6Only // V6 multicast on a V6-only socket + broadcast // V4 broadcast on a V4 socket + broadcastIn6 // V4-mapped broadcast on a V6-dual socket + reverseMulticast4 // V4 multicast src. Must fail. + reverseMulticast6 // V6 multicast src. Must fail. ) func (flow testFlow) String() string { @@ -117,6 +119,10 @@ func (flow testFlow) String() string { return "broadcast" case broadcastIn6: return "broadcastIn6" + case reverseMulticast4: + return "reverseMulticast4" + case reverseMulticast6: + return "reverseMulticast6" default: return "unknown" } @@ -168,6 +174,9 @@ func (flow testFlow) header4Tuple(d packetDirection) header4Tuple { h.dstAddr.Addr = multicastV6Addr } } + if flow.isReverseMulticast() { + h.srcAddr.Addr = flow.getMcastAddr() + } return h } @@ -199,9 +208,9 @@ func (flow testFlow) netProto() tcpip.NetworkProtocolNumber { // endpoint for this flow. func (flow testFlow) sockProto() tcpip.NetworkProtocolNumber { switch flow { - case unicastV4in6, unicastV6, unicastV6Only, multicastV4in6, multicastV6, multicastV6Only, broadcastIn6: + case unicastV4in6, unicastV6, unicastV6Only, multicastV4in6, multicastV6, multicastV6Only, broadcastIn6, reverseMulticast6: return ipv6.ProtocolNumber - case unicastV4, multicastV4, broadcast: + case unicastV4, multicastV4, broadcast, reverseMulticast4: return ipv4.ProtocolNumber default: panic(fmt.Sprintf("invalid testFlow given: %d", flow)) @@ -224,7 +233,7 @@ func (flow testFlow) isV6Only() bool { switch flow { case unicastV6Only, multicastV6Only: return true - case unicastV4, unicastV4in6, unicastV6, multicastV4, multicastV4in6, multicastV6, broadcast, broadcastIn6: + case unicastV4, unicastV4in6, unicastV6, multicastV4, multicastV4in6, multicastV6, broadcast, broadcastIn6, reverseMulticast4, reverseMulticast6: return false default: panic(fmt.Sprintf("invalid testFlow given: %d", flow)) @@ -235,7 +244,7 @@ func (flow testFlow) isMulticast() bool { switch flow { case multicastV4, multicastV4in6, multicastV6, multicastV6Only: return true - case unicastV4, unicastV4in6, unicastV6, unicastV6Only, broadcast, broadcastIn6: + case unicastV4, unicastV4in6, unicastV6, unicastV6Only, broadcast, broadcastIn6, reverseMulticast4, reverseMulticast6: return false default: panic(fmt.Sprintf("invalid testFlow given: %d", flow)) @@ -246,7 +255,7 @@ func (flow testFlow) isBroadcast() bool { switch flow { case broadcast, broadcastIn6: return true - case unicastV4, unicastV4in6, unicastV6, unicastV6Only, multicastV4, multicastV4in6, multicastV6, multicastV6Only: + case unicastV4, unicastV4in6, unicastV6, unicastV6Only, multicastV4, multicastV4in6, multicastV6, multicastV6Only, reverseMulticast4, reverseMulticast6: return false default: panic(fmt.Sprintf("invalid testFlow given: %d", flow)) @@ -257,13 +266,22 @@ func (flow testFlow) isMapped() bool { switch flow { case unicastV4in6, multicastV4in6, broadcastIn6: return true - case unicastV4, unicastV6, unicastV6Only, multicastV4, multicastV6, multicastV6Only, broadcast: + case unicastV4, unicastV6, unicastV6Only, multicastV4, multicastV6, multicastV6Only, broadcast, reverseMulticast4, reverseMulticast6: return false default: panic(fmt.Sprintf("invalid testFlow given: %d", flow)) } } +func (flow testFlow) isReverseMulticast() bool { + switch flow { + case reverseMulticast4, reverseMulticast6: + return true + default: + return false + } +} + type testContext struct { t *testing.T linkEP *channel.Endpoint @@ -872,6 +890,60 @@ func TestV4ReadOnBoundToBroadcast(t *testing.T) { } } +// TestReadFromMulticast checks that an endpoint will NOT receive a packet +// that was sent with multicast SOURCE address. +func TestReadFromMulticast(t *testing.T) { + for _, flow := range []testFlow{reverseMulticast4, reverseMulticast6} { + t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpointForFlow(flow) + + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + t.Fatalf("Bind failed: %s", err) + } + testFailingRead(c, flow, false /* expectReadError */) + }) + } +} + +// TestReadFromMulticaststats checks that a discarded packet +// that that was sent with multicast SOURCE address increments +// the correct counters and that a regular packet does not. +func TestReadFromMulticastStats(t *testing.T) { + t.Helper() + for _, flow := range []testFlow{reverseMulticast4, reverseMulticast6, unicastV4} { + t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpointForFlow(flow) + + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + t.Fatalf("Bind failed: %s", err) + } + + payload := newPayload() + c.injectPacket(flow, payload) + + var want uint64 = 0 + if flow.isReverseMulticast() { + want = 1 + } + if got := c.s.Stats().IP.InvalidSourceAddressesReceived.Value(); got != want { + t.Errorf("got stats.IP.InvalidSourceAddressesReceived.Value() = %d, want = %d", got, want) + } + if got := c.s.Stats().UDP.InvalidSourceAddress.Value(); got != want { + t.Errorf("got stats.UDP.InvalidSourceAddress.Value() = %d, want = %d", got, want) + } + if got := c.ep.Stats().(*tcpip.TransportEndpointStats).ReceiveErrors.MalformedPacketsReceived.Value(); got != want { + t.Errorf("got EP Stats.ReceiveErrors.MalformedPacketsReceived stats = %d, want = %d", got, want) + } + }) + } +} + // TestV4ReadBroadcastOnBoundToWildcard checks that an endpoint can bind to ANY // and receive broadcast and unicast data. func TestV4ReadBroadcastOnBoundToWildcard(t *testing.T) { -- cgit v1.2.3 From feb867bb83d647df45ec0d046db5d25494a162bf Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Fri, 10 Jul 2020 03:09:54 -0400 Subject: Split the kvm ut test cases to correspond to different platforms Split the kvm ut test cases to pass unit-tests on Arm64. I will add the tls and full-context test cases for Arm64 later. Signed-off-by: Bin Lu --- pkg/sentry/platform/kvm/BUILD | 1 + pkg/sentry/platform/kvm/kvm_amd64_test.go | 51 +++++++++++++++++++++++ pkg/sentry/platform/kvm/kvm_test.go | 24 ----------- pkg/sentry/platform/kvm/testutil/testutil_arm64.s | 6 +++ 4 files changed, 58 insertions(+), 24 deletions(-) create mode 100644 pkg/sentry/platform/kvm/kvm_amd64_test.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/kvm/BUILD b/pkg/sentry/platform/kvm/BUILD index 4792454c4..10a10bfe2 100644 --- a/pkg/sentry/platform/kvm/BUILD +++ b/pkg/sentry/platform/kvm/BUILD @@ -60,6 +60,7 @@ go_library( go_test( name = "kvm_test", srcs = [ + "kvm_amd64_test.go", "kvm_test.go", "virtual_map_test.go", ], diff --git a/pkg/sentry/platform/kvm/kvm_amd64_test.go b/pkg/sentry/platform/kvm/kvm_amd64_test.go new file mode 100644 index 000000000..c0b4fd374 --- /dev/null +++ b/pkg/sentry/platform/kvm/kvm_amd64_test.go @@ -0,0 +1,51 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build amd64 + +package kvm + +import ( + "testing" + + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/platform" + "gvisor.dev/gvisor/pkg/sentry/platform/kvm/testutil" + "gvisor.dev/gvisor/pkg/sentry/platform/ring0" + "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" +) + +func TestSegments(t *testing.T) { + applicationTest(t, true, testutil.TwiddleSegments, func(c *vCPU, regs *arch.Registers, pt *pagetables.PageTables) bool { + testutil.SetTestSegments(regs) + for { + var si arch.SignalInfo + if _, err := c.SwitchToUser(ring0.SwitchOpts{ + Registers: regs, + FloatingPointState: dummyFPState, + PageTables: pt, + FullRestore: true, + }, &si); err == platform.ErrContextInterrupt { + continue // Retry. + } else if err != nil { + t.Errorf("application segment check with full restore got unexpected error: %v", err) + } + if err := testutil.CheckTestSegments(regs); err != nil { + t.Errorf("application segment check with full restore failed: %v", err) + } + break // Done. + } + return false + }) +} diff --git a/pkg/sentry/platform/kvm/kvm_test.go b/pkg/sentry/platform/kvm/kvm_test.go index 6c8f4fa28..45b3180f1 100644 --- a/pkg/sentry/platform/kvm/kvm_test.go +++ b/pkg/sentry/platform/kvm/kvm_test.go @@ -262,30 +262,6 @@ func TestRegistersFault(t *testing.T) { }) } -func TestSegments(t *testing.T) { - applicationTest(t, true, testutil.TwiddleSegments, func(c *vCPU, regs *arch.Registers, pt *pagetables.PageTables) bool { - testutil.SetTestSegments(regs) - for { - var si arch.SignalInfo - if _, err := c.SwitchToUser(ring0.SwitchOpts{ - Registers: regs, - FloatingPointState: dummyFPState, - PageTables: pt, - FullRestore: true, - }, &si); err == platform.ErrContextInterrupt { - continue // Retry. - } else if err != nil { - t.Errorf("application segment check with full restore got unexpected error: %v", err) - } - if err := testutil.CheckTestSegments(regs); err != nil { - t.Errorf("application segment check with full restore failed: %v", err) - } - break // Done. - } - return false - }) -} - func TestBounce(t *testing.T) { applicationTest(t, true, testutil.SpinLoop, func(c *vCPU, regs *arch.Registers, pt *pagetables.PageTables) bool { go func() { diff --git a/pkg/sentry/platform/kvm/testutil/testutil_arm64.s b/pkg/sentry/platform/kvm/testutil/testutil_arm64.s index 0bebee852..07658144e 100644 --- a/pkg/sentry/platform/kvm/testutil/testutil_arm64.s +++ b/pkg/sentry/platform/kvm/testutil/testutil_arm64.s @@ -104,3 +104,9 @@ TEXT ·TwiddleRegsSyscall(SB),NOSPLIT,$0 TWIDDLE_REGS() SVC RET // never reached + +TEXT ·TwiddleRegsFault(SB),NOSPLIT,$0 + TWIDDLE_REGS() + // Branch to Register branches unconditionally to an address in . + JMP (R4) // <=> br x4, must fault + RET // never reached -- cgit v1.2.3 From 216dcebc066c82907b0de790a77a3deb6a734805 Mon Sep 17 00:00:00 2001 From: Bhasker Hariharan Date: Sat, 11 Jul 2020 06:21:34 -0700 Subject: Stub out SO_DETACH_FILTER. Updates #2746 PiperOrigin-RevId: 320757963 --- pkg/sentry/socket/netstack/netstack.go | 5 +++ pkg/tcpip/tcpip.go | 5 ++- pkg/tcpip/transport/icmp/endpoint.go | 4 ++ pkg/tcpip/transport/packet/endpoint.go | 8 +++- pkg/tcpip/transport/raw/endpoint.go | 8 +++- pkg/tcpip/transport/tcp/endpoint.go | 3 ++ pkg/tcpip/transport/udp/endpoint.go | 3 ++ test/syscalls/linux/BUILD | 3 ++ test/syscalls/linux/packet_socket_raw.cc | 34 ++++++++++++++++ test/syscalls/linux/raw_socket.cc | 37 +++++++++++++++-- test/syscalls/linux/tcp_socket.cc | 60 ++++++++++++++++++++++++++++ test/syscalls/linux/udp_socket_test_cases.cc | 54 +++++++++++++++++++++++++ 12 files changed, 217 insertions(+), 7 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 5a3cedd7c..78a842973 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -1754,6 +1754,11 @@ func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, nam return nil + case linux.SO_DETACH_FILTER: + // optval is ignored. + var v tcpip.SocketDetachFilterOption + return syserr.TranslateNetstackError(ep.SetSockOpt(v)) + default: socket.SetSockOptEmitUnimplementedEvent(t, name) } diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index cf7291d09..71bcee785 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -855,7 +855,10 @@ type OutOfBandInlineOption int // a default TTL. type DefaultTTLOption uint8 -// +// SocketDetachFilterOption is used by SetSockOpt to detach a previously attached +// classic BPF filter on a given endpoint. +type SocketDetachFilterOption int + // IPPacketInfo is the message structure for IP_PKTINFO. // // +stateify savable diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index 62d1acad4..678f4e016 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -344,6 +344,10 @@ func (e *endpoint) Peek([][]byte) (int64, tcpip.ControlMessages, *tcpip.Error) { // SetSockOpt sets a socket option. func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { + switch opt.(type) { + case tcpip.SocketDetachFilterOption: + return nil + } return nil } diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go index a8f8454dd..57b7f5c19 100644 --- a/pkg/tcpip/transport/packet/endpoint.go +++ b/pkg/tcpip/transport/packet/endpoint.go @@ -278,7 +278,13 @@ func (ep *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { // used with SetSockOpt, and this function always returns // tcpip.ErrNotSupported. func (ep *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { - return tcpip.ErrUnknownProtocolOption + switch opt.(type) { + case tcpip.SocketDetachFilterOption: + return nil + + default: + return tcpip.ErrUnknownProtocolOption + } } // SetSockOptBool implements tcpip.Endpoint.SetSockOptBool. diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index 5b6e7d102..c2e9fd29f 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -506,7 +506,13 @@ func (e *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { // SetSockOpt implements tcpip.Endpoint.SetSockOpt. func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { - return tcpip.ErrUnknownProtocolOption + switch opt.(type) { + case tcpip.SocketDetachFilterOption: + return nil + + default: + return tcpip.ErrUnknownProtocolOption + } } // SetSockOptBool implements tcpip.Endpoint.SetSockOptBool. diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index caac6ef57..83dc10ed0 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -1792,6 +1792,9 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.deferAccept = time.Duration(v) e.UnlockUser() + case tcpip.SocketDetachFilterOption: + return nil + default: return nil } diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 4e9e114a9..a14643ae8 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -816,6 +816,9 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.mu.Lock() e.bindToDevice = id e.mu.Unlock() + + case tcpip.SocketDetachFilterOption: + return nil } return nil } diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 9e097c888..662d780d8 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -1330,6 +1330,7 @@ cc_binary( name = "packet_socket_raw_test", testonly = 1, srcs = ["packet_socket_raw.cc"], + defines = select_system(), linkstatic = 1, deps = [ ":socket_test_util", @@ -1809,6 +1810,7 @@ cc_binary( name = "raw_socket_test", testonly = 1, srcs = ["raw_socket.cc"], + defines = select_system(), linkstatic = 1, deps = [ ":socket_test_util", @@ -3407,6 +3409,7 @@ cc_binary( name = "tcp_socket_test", testonly = 1, srcs = ["tcp_socket.cc"], + defines = select_system(), linkstatic = 1, deps = [ ":socket_test_util", diff --git a/test/syscalls/linux/packet_socket_raw.cc b/test/syscalls/linux/packet_socket_raw.cc index 4093ac813..6a963b12c 100644 --- a/test/syscalls/linux/packet_socket_raw.cc +++ b/test/syscalls/linux/packet_socket_raw.cc @@ -14,6 +14,9 @@ #include #include +#ifndef __fuchsia__ +#include +#endif // __fuchsia__ #include #include #include @@ -556,6 +559,37 @@ TEST_P(RawPacketTest, SetSocketSendBuf) { ASSERT_EQ(quarter_sz, val); } +#ifndef __fuchsia__ + +TEST_P(RawPacketTest, SetSocketDetachFilterNoInstalledFilter) { + // TODO(gvisor.dev/2746): Support SO_ATTACH_FILTER/SO_DETACH_FILTER. + // + // gVisor returns no error on SO_DETACH_FILTER even if there is no filter + // attached unlike linux which does return ENOENT in such cases. This is + // because gVisor doesn't support SO_ATTACH_FILTER and just silently returns + // success. + if (IsRunningOnGvisor()) { + constexpr int val = 0; + ASSERT_THAT(setsockopt(s_, SOL_SOCKET, SO_DETACH_FILTER, &val, sizeof(val)), + SyscallSucceeds()); + return; + } + constexpr int val = 0; + ASSERT_THAT(setsockopt(s_, SOL_SOCKET, SO_DETACH_FILTER, &val, sizeof(val)), + SyscallFailsWithErrno(ENOENT)); +} + +TEST_P(RawPacketTest, GetSocketDetachFilter) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_DETACH_FILTER, &val, &val_len), + SyscallFailsWithErrno(ENOPROTOOPT)); +} + +#endif // __fuchsia__ + INSTANTIATE_TEST_SUITE_P(AllInetTests, RawPacketTest, ::testing::Values(ETH_P_IP, ETH_P_ALL)); diff --git a/test/syscalls/linux/raw_socket.cc b/test/syscalls/linux/raw_socket.cc index 05c4ed03f..ce54dc064 100644 --- a/test/syscalls/linux/raw_socket.cc +++ b/test/syscalls/linux/raw_socket.cc @@ -13,6 +13,9 @@ // limitations under the License. #include +#ifndef __fuchsia__ +#include +#endif // __fuchsia__ #include #include #include @@ -21,6 +24,7 @@ #include #include #include + #include #include "gtest/gtest.h" @@ -790,10 +794,30 @@ void RawSocketTest::ReceiveBufFrom(int sock, char* recv_buf, ASSERT_NO_FATAL_FAILURE(RecvNoCmsg(sock, recv_buf, recv_buf_len)); } -INSTANTIATE_TEST_SUITE_P(AllInetTests, RawSocketTest, - ::testing::Combine( - ::testing::Values(IPPROTO_TCP, IPPROTO_UDP), - ::testing::Values(AF_INET, AF_INET6))); +#ifndef __fuchsia__ + +TEST_P(RawSocketTest, SetSocketDetachFilterNoInstalledFilter) { + // TODO(gvisor.dev/2746): Support SO_ATTACH_FILTER/SO_DETACH_FILTER. + if (IsRunningOnGvisor()) { + constexpr int val = 0; + ASSERT_THAT(setsockopt(s_, SOL_SOCKET, SO_DETACH_FILTER, &val, sizeof(val)), + SyscallSucceeds()); + return; + } + + constexpr int val = 0; + ASSERT_THAT(setsockopt(s_, SOL_SOCKET, SO_DETACH_FILTER, &val, sizeof(val)), + SyscallFailsWithErrno(ENOENT)); +} + +TEST_P(RawSocketTest, GetSocketDetachFilter) { + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_DETACH_FILTER, &val, &val_len), + SyscallFailsWithErrno(ENOPROTOOPT)); +} + +#endif // __fuchsia__ // AF_INET6+SOCK_RAW+IPPROTO_RAW sockets can be created, but not written to. TEST(RawSocketTest, IPv6ProtoRaw) { @@ -813,6 +837,11 @@ TEST(RawSocketTest, IPv6ProtoRaw) { SyscallFailsWithErrno(EINVAL)); } +INSTANTIATE_TEST_SUITE_P( + AllInetTests, RawSocketTest, + ::testing::Combine(::testing::Values(IPPROTO_TCP, IPPROTO_UDP), + ::testing::Values(AF_INET, AF_INET6))); + } // namespace } // namespace testing diff --git a/test/syscalls/linux/tcp_socket.cc b/test/syscalls/linux/tcp_socket.cc index a4d2953e1..0cea7d11f 100644 --- a/test/syscalls/linux/tcp_socket.cc +++ b/test/syscalls/linux/tcp_socket.cc @@ -13,6 +13,9 @@ // limitations under the License. #include +#ifndef __fuchsia__ +#include +#endif // __fuchsia__ #include #include #include @@ -1559,6 +1562,63 @@ TEST_P(SimpleTcpSocketTest, SetTCPWindowClampAboveHalfMinRcvBuf) { } } +#ifndef __fuchsia__ + +// TODO(gvisor.dev/2746): Support SO_ATTACH_FILTER/SO_DETACH_FILTER. +// gVisor currently silently ignores attaching a filter. +TEST_P(SimpleTcpSocketTest, SetSocketAttachDetachFilter) { + FileDescriptor s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + // Program generated using sudo tcpdump -i lo tcp and port 1234 -dd + struct sock_filter code[] = { + {0x28, 0, 0, 0x0000000c}, {0x15, 0, 6, 0x000086dd}, + {0x30, 0, 0, 0x00000014}, {0x15, 0, 15, 0x00000006}, + {0x28, 0, 0, 0x00000036}, {0x15, 12, 0, 0x000004d2}, + {0x28, 0, 0, 0x00000038}, {0x15, 10, 11, 0x000004d2}, + {0x15, 0, 10, 0x00000800}, {0x30, 0, 0, 0x00000017}, + {0x15, 0, 8, 0x00000006}, {0x28, 0, 0, 0x00000014}, + {0x45, 6, 0, 0x00001fff}, {0xb1, 0, 0, 0x0000000e}, + {0x48, 0, 0, 0x0000000e}, {0x15, 2, 0, 0x000004d2}, + {0x48, 0, 0, 0x00000010}, {0x15, 0, 1, 0x000004d2}, + {0x6, 0, 0, 0x00040000}, {0x6, 0, 0, 0x00000000}, + }; + struct sock_fprog bpf = { + .len = ABSL_ARRAYSIZE(code), + .filter = code, + }; + ASSERT_THAT( + setsockopt(s.get(), SOL_SOCKET, SO_ATTACH_FILTER, &bpf, sizeof(bpf)), + SyscallSucceeds()); + + constexpr int val = 0; + ASSERT_THAT( + setsockopt(s.get(), SOL_SOCKET, SO_DETACH_FILTER, &val, sizeof(val)), + SyscallSucceeds()); +} + +TEST_P(SimpleTcpSocketTest, SetSocketDetachFilterNoInstalledFilter) { + // TODO(gvisor.dev/2746): Support SO_ATTACH_FILTER/SO_DETACH_FILTER. + SKIP_IF(IsRunningOnGvisor()); + FileDescriptor s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + constexpr int val = 0; + ASSERT_THAT( + setsockopt(s.get(), SOL_SOCKET, SO_DETACH_FILTER, &val, sizeof(val)), + SyscallFailsWithErrno(ENOENT)); +} + +TEST_P(SimpleTcpSocketTest, GetSocketDetachFilter) { + FileDescriptor s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s.get(), SOL_SOCKET, SO_DETACH_FILTER, &val, &val_len), + SyscallFailsWithErrno(ENOPROTOOPT)); +} + +#endif // __fuchsia__ + INSTANTIATE_TEST_SUITE_P(AllInetTests, SimpleTcpSocketTest, ::testing::Values(AF_INET, AF_INET6)); diff --git a/test/syscalls/linux/udp_socket_test_cases.cc b/test/syscalls/linux/udp_socket_test_cases.cc index 9cc6be4fb..60c48ed6e 100644 --- a/test/syscalls/linux/udp_socket_test_cases.cc +++ b/test/syscalls/linux/udp_socket_test_cases.cc @@ -16,6 +16,9 @@ #include #include +#ifndef __fuchsia__ +#include +#endif // __fuchsia__ #include #include #include @@ -1723,5 +1726,56 @@ TEST_P(UdpSocketTest, RecvBufLimits) { } } +#ifndef __fuchsia__ + +// TODO(gvisor.dev/2746): Support SO_ATTACH_FILTER/SO_DETACH_FILTER. +// gVisor currently silently ignores attaching a filter. +TEST_P(UdpSocketTest, SetSocketDetachFilter) { + // Program generated using sudo tcpdump -i lo udp and port 1234 -dd + struct sock_filter code[] = { + {0x28, 0, 0, 0x0000000c}, {0x15, 0, 6, 0x000086dd}, + {0x30, 0, 0, 0x00000014}, {0x15, 0, 15, 0x00000011}, + {0x28, 0, 0, 0x00000036}, {0x15, 12, 0, 0x000004d2}, + {0x28, 0, 0, 0x00000038}, {0x15, 10, 11, 0x000004d2}, + {0x15, 0, 10, 0x00000800}, {0x30, 0, 0, 0x00000017}, + {0x15, 0, 8, 0x00000011}, {0x28, 0, 0, 0x00000014}, + {0x45, 6, 0, 0x00001fff}, {0xb1, 0, 0, 0x0000000e}, + {0x48, 0, 0, 0x0000000e}, {0x15, 2, 0, 0x000004d2}, + {0x48, 0, 0, 0x00000010}, {0x15, 0, 1, 0x000004d2}, + {0x6, 0, 0, 0x00040000}, {0x6, 0, 0, 0x00000000}, + }; + struct sock_fprog bpf = { + .len = ABSL_ARRAYSIZE(code), + .filter = code, + }; + ASSERT_THAT( + setsockopt(sock_.get(), SOL_SOCKET, SO_ATTACH_FILTER, &bpf, sizeof(bpf)), + SyscallSucceeds()); + + constexpr int val = 0; + ASSERT_THAT( + setsockopt(sock_.get(), SOL_SOCKET, SO_DETACH_FILTER, &val, sizeof(val)), + SyscallSucceeds()); +} + +TEST_P(UdpSocketTest, SetSocketDetachFilterNoInstalledFilter) { + // TODO(gvisor.dev/2746): Support SO_ATTACH_FILTER/SO_DETACH_FILTER. + SKIP_IF(IsRunningOnGvisor()); + constexpr int val = 0; + ASSERT_THAT( + setsockopt(sock_.get(), SOL_SOCKET, SO_DETACH_FILTER, &val, sizeof(val)), + SyscallFailsWithErrno(ENOENT)); +} + +TEST_P(UdpSocketTest, GetSocketDetachFilter) { + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT( + getsockopt(sock_.get(), SOL_SOCKET, SO_DETACH_FILTER, &val, &val_len), + SyscallFailsWithErrno(ENOPROTOOPT)); +} + +#endif // __fuchsia__ + } // namespace testing } // namespace gvisor -- cgit v1.2.3 From 6994f4d5912ce9dc9233aebe9902892824904b71 Mon Sep 17 00:00:00 2001 From: Ayush Ranjan Date: Mon, 13 Jul 2020 13:33:09 -0700 Subject: [vfs2] Make gofer metadata atomics consistent For accessing metadata fields: - If metadataMu is locked, we can access without atomics - If metadataMu is unlocked, we should use atomics For mutating metadata fields: - Always lock metadataMu and use atomics. There were some instances of inconsistencies which have been fixed. PiperOrigin-RevId: 321022895 --- pkg/sentry/fsimpl/gofer/gofer.go | 24 +++++++++++++++--------- pkg/sentry/fsimpl/gofer/regular_file.go | 4 +++- 2 files changed, 18 insertions(+), 10 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 2b83094cd..b74d489a0 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -602,8 +602,14 @@ type dentry struct { // returned by the server. dirents is protected by dirMu. dirents []vfs.Dirent - // Cached metadata; protected by metadataMu and accessed using atomic - // memory operations unless otherwise specified. + // Cached metadata; protected by metadataMu. + // To access: + // - In situations where consistency is not required (like stat), these + // can be accessed using atomic operations only (without locking). + // - Lock metadataMu and can access without atomic operations. + // To mutate: + // - Lock metadataMu and use atomic operations to update because we might + // have atomic readers that don't hold the lock. metadataMu sync.Mutex ino inodeNumber // immutable mode uint32 // type is immutable, perms are mutable @@ -616,7 +622,7 @@ type dentry struct { ctime int64 btime int64 // File size, protected by both metadataMu and dataMu (i.e. both must be - // locked to mutate it). + // locked to mutate it; locking either is sufficient to access it). size uint64 // nlink counts the number of hard links to this dentry. It's updated and @@ -904,14 +910,14 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin // Prepare for truncate. if stat.Mask&linux.STATX_SIZE != 0 { - switch d.mode & linux.S_IFMT { - case linux.S_IFREG: + switch mode.FileType() { + case linux.ModeRegular: if !setLocalMtime { // Truncate updates mtime. setLocalMtime = true stat.Mtime.Nsec = linux.UTIME_NOW } - case linux.S_IFDIR: + case linux.ModeDirectory: return syserror.EISDIR default: return syserror.EINVAL @@ -994,7 +1000,7 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin func (d *dentry) updateFileSizeLocked(newSize uint64) { d.dataMu.Lock() oldSize := d.size - d.size = newSize + atomic.StoreUint64(&d.size, newSize) // d.dataMu must be unlocked to lock d.mapsMu and invalidate mappings // below. This allows concurrent calls to Read/Translate/etc. These // functions synchronize with truncation by refusing to use cache @@ -1340,8 +1346,8 @@ func (d *dentry) removexattr(ctx context.Context, creds *auth.Credentials, name // Extended attributes in the user.* namespace are only supported for regular // files and directories. func (d *dentry) userXattrSupported() bool { - filetype := linux.S_IFMT & atomic.LoadUint32(&d.mode) - return filetype == linux.S_IFREG || filetype == linux.S_IFDIR + filetype := linux.FileMode(atomic.LoadUint32(&d.mode)).FileType() + return filetype == linux.ModeRegular || filetype == linux.ModeDirectory } // Preconditions: !d.isSynthetic(). d.isRegularFile() || d.isDir(). diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go index a2f02d9c7..f10350c97 100644 --- a/pkg/sentry/fsimpl/gofer/regular_file.go +++ b/pkg/sentry/fsimpl/gofer/regular_file.go @@ -89,7 +89,9 @@ func (fd *regularFileFD) Allocate(ctx context.Context, mode, offset, length uint if err != nil { return err } - d.size = size + d.dataMu.Lock() + atomic.StoreUint64(&d.size, size) + d.dataMu.Unlock() if !d.cachedMetadataAuthoritative() { d.touchCMtimeLocked() } -- cgit v1.2.3 From cf0826653c436021a1b808d1f0aea082c9a0fc1b Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Mon, 13 Jul 2020 14:00:21 -0700 Subject: Clean up inotify comments. PiperOrigin-RevId: 321028238 --- pkg/sentry/vfs/inotify.go | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/vfs/inotify.go b/pkg/sentry/vfs/inotify.go index c2e21ac5f..167b731ac 100644 --- a/pkg/sentry/vfs/inotify.go +++ b/pkg/sentry/vfs/inotify.go @@ -179,12 +179,12 @@ func (i *Inotify) Readiness(mask waiter.EventMask) waiter.EventMask { return mask & ready } -// PRead implements FileDescriptionImpl. +// PRead implements FileDescriptionImpl.PRead. func (*Inotify) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts ReadOptions) (int64, error) { return 0, syserror.ESPIPE } -// PWrite implements FileDescriptionImpl. +// PWrite implements FileDescriptionImpl.PWrite. func (*Inotify) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts WriteOptions) (int64, error) { return 0, syserror.ESPIPE } @@ -243,7 +243,7 @@ func (i *Inotify) Read(ctx context.Context, dst usermem.IOSequence, opts ReadOpt return writeLen, nil } -// Ioctl implements fs.FileOperations.Ioctl. +// Ioctl implements FileDescriptionImpl.Ioctl. func (i *Inotify) Ioctl(ctx context.Context, uio usermem.IO, args arch.SyscallArguments) (uintptr, error) { switch args[1].Int() { case linux.FIONREAD: -- cgit v1.2.3 From 59a5479409094b141a60cfcc65f0a53d7871e2e1 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Mon, 13 Jul 2020 15:41:07 -0700 Subject: Disable debug time adjustment logging When --debug is enabled, the following log messages are printed every second filling up the log: D0430 18:04:42.823775 129561 parameters.go:238] Clock(Monotonic): error: 46 ns, adjusted frequency from 3591713733 Hz to 3591714196 Hz D0430 18:04:42.823870 129561 parameters.go:238] Clock(Realtime): error: 36 ns, adjusted frequency from 3591714003 Hz to 3591714169 Hz D0430 18:04:42.823892 129561 timekeeper.go:209] Updating VDSO parameters: {monotonicReady:1 monotonicBaseCycles:15758797714254696 monotonicBaseRef:29000233837 monotonicFrequency:3591714196 realtimeReady:1 realtimeBaseCycles:15758797714610880 realtimeBaseRef:1588269882823867374 realtimeFrequency:3591714169} Info and warning messages for larger changes are kept the same. PiperOrigin-RevId: 321048523 --- pkg/sentry/kernel/timekeeper.go | 3 --- pkg/sentry/time/parameters.go | 12 ++++++++---- 2 files changed, 8 insertions(+), 7 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/timekeeper.go b/pkg/sentry/kernel/timekeeper.go index 0adf25691..5f3908d8b 100644 --- a/pkg/sentry/kernel/timekeeper.go +++ b/pkg/sentry/kernel/timekeeper.go @@ -210,9 +210,6 @@ func (t *Timekeeper) startUpdater() { p.realtimeBaseRef = int64(realtimeParams.BaseRef) p.realtimeFrequency = realtimeParams.Frequency } - - log.Debugf("Updating VDSO parameters: %+v", p) - return p }); err != nil { log.Warningf("Unable to update VDSO parameter page: %v", err) diff --git a/pkg/sentry/time/parameters.go b/pkg/sentry/time/parameters.go index 65868cb26..cd1b95117 100644 --- a/pkg/sentry/time/parameters.go +++ b/pkg/sentry/time/parameters.go @@ -228,11 +228,15 @@ func errorAdjust(prevParams Parameters, newParams Parameters, now TSCValue) (Par // // The log level is determined by the error severity. func logErrorAdjustment(clock ClockID, errorNS ReferenceNS, orig, adjusted Parameters) { - fn := log.Debugf - if int64(errorNS.Magnitude()) > time.Millisecond.Nanoseconds() { + magNS := int64(errorNS.Magnitude()) + if magNS <= 10*time.Microsecond.Nanoseconds() { + // Don't log small errors. + return + } + fn := log.Infof + if magNS > time.Millisecond.Nanoseconds() { + // Upgrade large errors to warning. fn = log.Warningf - } else if int64(errorNS.Magnitude()) > 10*time.Microsecond.Nanoseconds() { - fn = log.Infof } fn("Clock(%v): error: %v ns, adjusted frequency from %v Hz to %v Hz", clock, errorNS, orig.Frequency, adjusted.Frequency) -- cgit v1.2.3 From 8494a0325d2ca7ba7bc9c17e55546f1665500660 Mon Sep 17 00:00:00 2001 From: Craig Chi Date: Tue, 14 Jul 2020 15:42:47 -0700 Subject: Include context in kernfs.Inode.Stat method To implement stat(2) in FUSE, we have to embed credentials and pid in request header. The information should be extracted from the context passed to VFS layer. Therefore `Stat()` signature in `kernfs.Inode` interface should include context as first argument. Some other fs implementations need to be modified as well, such as devpts, host, pipefs, and proc. Fixes #3235 --- pkg/sentry/fsimpl/devpts/master.go | 6 +++--- pkg/sentry/fsimpl/devpts/slave.go | 6 +++--- pkg/sentry/fsimpl/host/host.go | 6 +++--- pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go | 2 +- pkg/sentry/fsimpl/kernfs/fd_impl_util.go | 8 ++++---- pkg/sentry/fsimpl/kernfs/filesystem.go | 2 +- pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 2 +- pkg/sentry/fsimpl/kernfs/kernfs.go | 2 +- pkg/sentry/fsimpl/pipefs/pipefs.go | 2 +- pkg/sentry/fsimpl/proc/subtasks.go | 4 ++-- pkg/sentry/fsimpl/proc/task.go | 4 ++-- pkg/sentry/fsimpl/proc/task_files.go | 2 +- pkg/sentry/fsimpl/proc/tasks.go | 4 ++-- 13 files changed, 25 insertions(+), 25 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/devpts/master.go b/pkg/sentry/fsimpl/devpts/master.go index 69879498a..1081fff52 100644 --- a/pkg/sentry/fsimpl/devpts/master.go +++ b/pkg/sentry/fsimpl/devpts/master.go @@ -67,8 +67,8 @@ func (mi *masterInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vf } // Stat implements kernfs.Inode.Stat. -func (mi *masterInode) Stat(vfsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { - statx, err := mi.InodeAttrs.Stat(vfsfs, opts) +func (mi *masterInode) Stat(ctx context.Context, vfsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { + statx, err := mi.InodeAttrs.Stat(ctx, vfsfs, opts) if err != nil { return linux.Statx{}, err } @@ -186,7 +186,7 @@ func (mfd *masterFileDescription) SetStat(ctx context.Context, opts vfs.SetStatO // Stat implements vfs.FileDescriptionImpl.Stat. func (mfd *masterFileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { fs := mfd.vfsfd.VirtualDentry().Mount().Filesystem() - return mfd.inode.Stat(fs, opts) + return mfd.inode.Stat(ctx, fs, opts) } // LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. diff --git a/pkg/sentry/fsimpl/devpts/slave.go b/pkg/sentry/fsimpl/devpts/slave.go index cf1a0f0ac..2018b978a 100644 --- a/pkg/sentry/fsimpl/devpts/slave.go +++ b/pkg/sentry/fsimpl/devpts/slave.go @@ -73,8 +73,8 @@ func (si *slaveInode) Valid(context.Context) bool { } // Stat implements kernfs.Inode.Stat. -func (si *slaveInode) Stat(vfsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { - statx, err := si.InodeAttrs.Stat(vfsfs, opts) +func (si *slaveInode) Stat(ctx context.Context, vfsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { + statx, err := si.InodeAttrs.Stat(ctx, vfsfs, opts) if err != nil { return linux.Statx{}, err } @@ -183,7 +183,7 @@ func (sfd *slaveFileDescription) SetStat(ctx context.Context, opts vfs.SetStatOp // Stat implements vfs.FileDescriptionImpl.Stat. func (sfd *slaveFileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { fs := sfd.vfsfd.VirtualDentry().Mount().Filesystem() - return sfd.inode.Stat(fs, opts) + return sfd.inode.Stat(ctx, fs, opts) } // LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 1cd2982cb..1a88cb657 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -259,7 +259,7 @@ func (i *inode) Mode() linux.FileMode { } // Stat implements kernfs.Inode. -func (i *inode) Stat(vfsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { +func (i *inode) Stat(ctx context.Context, vfsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { if opts.Mask&linux.STATX__RESERVED != 0 { return linux.Statx{}, syserror.EINVAL } @@ -534,8 +534,8 @@ func (f *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) } // Stat implements vfs.FileDescriptionImpl. -func (f *fileDescription) Stat(_ context.Context, opts vfs.StatOptions) (linux.Statx, error) { - return f.inode.Stat(f.vfsfd.Mount().Filesystem(), opts) +func (f *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { + return f.inode.Stat(ctx, f.vfsfd.Mount().Filesystem(), opts) } // Release implements vfs.FileDescriptionImpl. diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go index 6886b0876..c6c4472e7 100644 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -127,7 +127,7 @@ func (fd *DynamicBytesFD) Release() {} // Stat implements vfs.FileDescriptionImpl.Stat. func (fd *DynamicBytesFD) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { fs := fd.vfsfd.VirtualDentry().Mount().Filesystem() - return fd.inode.Stat(fs, opts) + return fd.inode.Stat(ctx, fs, opts) } // SetStat implements vfs.FileDescriptionImpl.SetStat. diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index ca8b8c63b..3f0aea73a 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -132,7 +132,7 @@ func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirent opts := vfs.StatOptions{Mask: linux.STATX_INO} // Handle ".". if fd.off == 0 { - stat, err := fd.inode().Stat(fd.filesystem(), opts) + stat, err := fd.inode().Stat(ctx, fd.filesystem(), opts) if err != nil { return err } @@ -152,7 +152,7 @@ func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirent if fd.off == 1 { vfsd := fd.vfsfd.VirtualDentry().Dentry() parentInode := genericParentOrSelf(vfsd.Impl().(*Dentry)).inode - stat, err := parentInode.Stat(fd.filesystem(), opts) + stat, err := parentInode.Stat(ctx, fd.filesystem(), opts) if err != nil { return err } @@ -176,7 +176,7 @@ func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirent childIdx := fd.off - 2 for it := fd.children.nthLocked(childIdx); it != nil; it = it.Next() { inode := it.Dentry.Impl().(*Dentry).inode - stat, err := inode.Stat(fd.filesystem(), opts) + stat, err := inode.Stat(ctx, fd.filesystem(), opts) if err != nil { return err } @@ -226,7 +226,7 @@ func (fd *GenericDirectoryFD) Seek(ctx context.Context, offset int64, whence int func (fd *GenericDirectoryFD) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { fs := fd.filesystem() inode := fd.inode() - return inode.Stat(fs, opts) + return inode.Stat(ctx, fs, opts) } // SetStat implements vfs.FileDescriptionImpl.SetStat. diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 8939871c1..61a36cff9 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -684,7 +684,7 @@ func (fs *Filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf if err != nil { return linux.Statx{}, err } - return inode.Stat(fs.VFSFilesystem(), opts) + return inode.Stat(ctx, fs.VFSFilesystem(), opts) } // StatFSAt implements vfs.FilesystemImpl.StatFSAt. diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 4cb885d87..2ab3f1761 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -243,7 +243,7 @@ func (a *InodeAttrs) Mode() linux.FileMode { // Stat partially implements Inode.Stat. Note that this function doesn't provide // all the stat fields, and the embedder should consider extending the result // with filesystem-specific fields. -func (a *InodeAttrs) Stat(*vfs.Filesystem, vfs.StatOptions) (linux.Statx, error) { +func (a *InodeAttrs) Stat(context.Context, *vfs.Filesystem, vfs.StatOptions) (linux.Statx, error) { var stat linux.Statx stat.Mask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_UID | linux.STATX_GID | linux.STATX_INO | linux.STATX_NLINK stat.DevMajor = a.devMajor diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index 596de1edf..46f207664 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -346,7 +346,7 @@ type inodeMetadata interface { // Stat returns the metadata for this inode. This corresponds to // vfs.FilesystemImpl.StatAt. - Stat(fs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) + Stat(ctx context.Context, fs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) // SetStat updates the metadata for this inode. This corresponds to // vfs.FilesystemImpl.SetStatAt. Implementations are responsible for checking diff --git a/pkg/sentry/fsimpl/pipefs/pipefs.go b/pkg/sentry/fsimpl/pipefs/pipefs.go index dd7eaf4a8..811f80a5f 100644 --- a/pkg/sentry/fsimpl/pipefs/pipefs.go +++ b/pkg/sentry/fsimpl/pipefs/pipefs.go @@ -115,7 +115,7 @@ func (i *inode) Mode() linux.FileMode { } // Stat implements kernfs.Inode.Stat. -func (i *inode) Stat(vfsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { +func (i *inode) Stat(_ context.Context, vfsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { ts := linux.NsecToStatxTimestamp(i.ctime.Nanoseconds()) return linux.Statx{ Mask: linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_NLINK | linux.STATX_UID | linux.STATX_GID | linux.STATX_ATIME | linux.STATX_MTIME | linux.STATX_CTIME | linux.STATX_INO | linux.STATX_SIZE | linux.STATX_BLOCKS, diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go index 36a89540c..dad4db1a7 100644 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -165,8 +165,8 @@ func (i *subtasksInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *v } // Stat implements kernfs.Inode. -func (i *subtasksInode) Stat(vsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { - stat, err := i.InodeAttrs.Stat(vsfs, opts) +func (i *subtasksInode) Stat(ctx context.Context, vsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { + stat, err := i.InodeAttrs.Stat(ctx, vsfs, opts) if err != nil { return linux.Statx{}, err } diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index 8bb2b0ce1..a5c7aa470 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -156,8 +156,8 @@ func (fs *filesystem) newTaskOwnedDir(task *kernel.Task, ino uint64, perm linux. } // Stat implements kernfs.Inode. -func (i *taskOwnedInode) Stat(fs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { - stat, err := i.Inode.Stat(fs, opts) +func (i *taskOwnedInode) Stat(ctx context.Context, fs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { + stat, err := i.Inode.Stat(ctx, fs, opts) if err != nil { return linux.Statx{}, err } diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index 9af43b859..859b7d727 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -876,7 +876,7 @@ var _ vfs.FileDescriptionImpl = (*namespaceFD)(nil) // Stat implements FileDescriptionImpl. func (fd *namespaceFD) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { vfs := fd.vfsfd.VirtualDentry().Mount().Filesystem() - return fd.inode.Stat(vfs, opts) + return fd.inode.Stat(ctx, vfs, opts) } // SetStat implements FileDescriptionImpl. diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index 2f214d0c2..6d2b90a8b 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -206,8 +206,8 @@ func (i *tasksInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs. return fd.VFSFileDescription(), nil } -func (i *tasksInode) Stat(vsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { - stat, err := i.InodeAttrs.Stat(vsfs, opts) +func (i *tasksInode) Stat(ctx context.Context, vsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { + stat, err := i.InodeAttrs.Stat(ctx, vsfs, opts) if err != nil { return linux.Statx{}, err } -- cgit v1.2.3 From 1b9965e06a966977a99569484da139d64d1db95e Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 14 Jul 2020 17:29:53 -0700 Subject: Update special file option name in comment. PiperOrigin-RevId: 321269281 --- pkg/sentry/fsimpl/gofer/special_file.go | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go index c1e6b13e5..a7b53b2d2 100644 --- a/pkg/sentry/fsimpl/gofer/special_file.go +++ b/pkg/sentry/fsimpl/gofer/special_file.go @@ -28,9 +28,9 @@ import ( ) // specialFileFD implements vfs.FileDescriptionImpl for pipes, sockets, device -// special files, and (when filesystemOptions.specialRegularFiles is in effect) -// regular files. specialFileFD differs from regularFileFD by using per-FD -// handles instead of shared per-dentry handles, and never buffering I/O. +// special files, and (when filesystemOptions.regularFilesUseSpecialFileFD is +// in effect) regular files. specialFileFD differs from regularFileFD by using +// per-FD handles instead of shared per-dentry handles, and never buffering I/O. type specialFileFD struct { fileDescription -- cgit v1.2.3 From 505bebae43183a68ee8873764817d282a6c49ec6 Mon Sep 17 00:00:00 2001 From: Tiwei Bie Date: Wed, 15 Jul 2020 22:02:26 +0800 Subject: hostinet: fix fd leak in fdnotifier for VFS2 When we failed to create the new socket after adding the fd to fdnotifier, we should remove the fd from fdnotifier, because we are going to close the fd directly. Fixes: #3241 Signed-off-by: Tiwei Bie --- pkg/sentry/socket/hostinet/socket_vfs2.go | 1 + 1 file changed, 1 insertion(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/hostinet/socket_vfs2.go b/pkg/sentry/socket/hostinet/socket_vfs2.go index 8f192c62f..8a1d52ebf 100644 --- a/pkg/sentry/socket/hostinet/socket_vfs2.go +++ b/pkg/sentry/socket/hostinet/socket_vfs2.go @@ -71,6 +71,7 @@ func newVFS2Socket(t *kernel.Task, family int, stype linux.SockType, protocol in DenyPWrite: true, UseDentryMetadata: true, }); err != nil { + fdnotifier.RemoveFD(int32(s.fd)) return nil, syserr.FromError(err) } return vfsfd, nil -- cgit v1.2.3 From fef90c61c6186c113cfdb0bbcf53f4ca70f9741a Mon Sep 17 00:00:00 2001 From: Bhasker Hariharan Date: Wed, 15 Jul 2020 14:13:42 -0700 Subject: Fix minor bugs in a couple of interface IOCTLs. gVisor incorrectly returns the wrong ARP type for SIOGIFHWADDR. This breaks tcpdump as it tries to interpret the packets incorrectly. Similarly, SIOCETHTOOL is used by tcpdump to query interface properties which fails with an EINVAL since we don't implement it. For now change it to return EOPNOTSUPP to indicate that we don't support the query rather than return EINVAL. NOTE: ARPHRD types for link endpoints are distinct from NIC capabilities and NIC flags. In Linux all 3 exist eg. ARPHRD types are stored in dev->type field while NIC capabilities are more like the device features which can be queried using SIOCETHTOOL but not modified and NIC Flags are fields that can be modified from user space. eg. NIC status (UP/DOWN/MULTICAST/BROADCAST) etc. Updates #2746 PiperOrigin-RevId: 321436525 --- pkg/abi/linux/ioctl.go | 27 +++++++++-- pkg/abi/linux/netlink_route.go | 2 + pkg/sentry/socket/netstack/netstack.go | 74 ++++++++++++++++++------------ pkg/sentry/socket/netstack/stack.go | 22 ++++++--- pkg/tcpip/header/arp.go | 77 +++++++++++++++++++++----------- pkg/tcpip/link/channel/BUILD | 1 + pkg/tcpip/link/channel/channel.go | 6 +++ pkg/tcpip/link/fdbased/endpoint.go | 8 ++++ pkg/tcpip/link/loopback/loopback.go | 5 +++ pkg/tcpip/link/muxed/BUILD | 1 + pkg/tcpip/link/muxed/injectable.go | 6 +++ pkg/tcpip/link/nested/BUILD | 1 + pkg/tcpip/link/nested/nested.go | 6 +++ pkg/tcpip/link/qdisc/fifo/BUILD | 1 + pkg/tcpip/link/qdisc/fifo/endpoint.go | 6 +++ pkg/tcpip/link/sharedmem/sharedmem.go | 5 +++ pkg/tcpip/link/tun/device.go | 10 +++++ pkg/tcpip/link/waitable/BUILD | 2 + pkg/tcpip/link/waitable/waitable.go | 6 +++ pkg/tcpip/link/waitable/waitable_test.go | 6 +++ pkg/tcpip/network/ip_test.go | 9 +++- pkg/tcpip/stack/forwarder_test.go | 6 +++ pkg/tcpip/stack/nic_test.go | 5 +++ pkg/tcpip/stack/registration.go | 7 +++ pkg/tcpip/stack/stack.go | 6 +++ test/syscalls/linux/socket_netdevice.cc | 23 ++++++++++ 26 files changed, 263 insertions(+), 65 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/ioctl.go b/pkg/abi/linux/ioctl.go index 2062e6a4b..2c5e56ae5 100644 --- a/pkg/abi/linux/ioctl.go +++ b/pkg/abi/linux/ioctl.go @@ -67,10 +67,29 @@ const ( // ioctl(2) requests provided by uapi/linux/sockios.h const ( - SIOCGIFMEM = 0x891f - SIOCGIFPFLAGS = 0x8935 - SIOCGMIIPHY = 0x8947 - SIOCGMIIREG = 0x8948 + SIOCGIFNAME = 0x8910 + SIOCGIFCONF = 0x8912 + SIOCGIFFLAGS = 0x8913 + SIOCGIFADDR = 0x8915 + SIOCGIFDSTADDR = 0x8917 + SIOCGIFBRDADDR = 0x8919 + SIOCGIFNETMASK = 0x891b + SIOCGIFMETRIC = 0x891d + SIOCGIFMTU = 0x8921 + SIOCGIFMEM = 0x891f + SIOCGIFHWADDR = 0x8927 + SIOCGIFINDEX = 0x8933 + SIOCGIFPFLAGS = 0x8935 + SIOCGIFTXQLEN = 0x8942 + SIOCETHTOOL = 0x8946 + SIOCGMIIPHY = 0x8947 + SIOCGMIIREG = 0x8948 + SIOCGIFMAP = 0x8970 +) + +// ioctl(2) requests provided by uapi/asm-generic/sockios.h +const ( + SIOCGSTAMP = 0x8906 ) // ioctl(2) directions. Used to calculate requests number. diff --git a/pkg/abi/linux/netlink_route.go b/pkg/abi/linux/netlink_route.go index 40bec566c..ceda0a8d3 100644 --- a/pkg/abi/linux/netlink_route.go +++ b/pkg/abi/linux/netlink_route.go @@ -187,6 +187,8 @@ const ( // Device types, from uapi/linux/if_arp.h. const ( + ARPHRD_NONE = 65534 + ARPHRD_ETHER = 1 ARPHRD_LOOPBACK = 772 ) diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 78a842973..0b1be1bd2 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -2747,7 +2747,7 @@ func (s *socketOpsCommon) ioctl(ctx context.Context, io usermem.IO, args arch.Sy // sockets. // TODO(b/78348848): Add a commonEndpoint method to support SIOCGSTAMP. switch args[1].Int() { - case syscall.SIOCGSTAMP: + case linux.SIOCGSTAMP: s.readMu.Lock() defer s.readMu.Unlock() if !s.timestampValid { @@ -2788,18 +2788,19 @@ func (s *socketOpsCommon) ioctl(ctx context.Context, io usermem.IO, args arch.Sy // Ioctl performs a socket ioctl. func Ioctl(ctx context.Context, ep commonEndpoint, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { switch arg := int(args[1].Int()); arg { - case syscall.SIOCGIFFLAGS, - syscall.SIOCGIFADDR, - syscall.SIOCGIFBRDADDR, - syscall.SIOCGIFDSTADDR, - syscall.SIOCGIFHWADDR, - syscall.SIOCGIFINDEX, - syscall.SIOCGIFMAP, - syscall.SIOCGIFMETRIC, - syscall.SIOCGIFMTU, - syscall.SIOCGIFNAME, - syscall.SIOCGIFNETMASK, - syscall.SIOCGIFTXQLEN: + case linux.SIOCGIFFLAGS, + linux.SIOCGIFADDR, + linux.SIOCGIFBRDADDR, + linux.SIOCGIFDSTADDR, + linux.SIOCGIFHWADDR, + linux.SIOCGIFINDEX, + linux.SIOCGIFMAP, + linux.SIOCGIFMETRIC, + linux.SIOCGIFMTU, + linux.SIOCGIFNAME, + linux.SIOCGIFNETMASK, + linux.SIOCGIFTXQLEN, + linux.SIOCETHTOOL: var ifr linux.IFReq if _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &ifr, usermem.IOOpts{ @@ -2815,7 +2816,7 @@ func Ioctl(ctx context.Context, ep commonEndpoint, io usermem.IO, args arch.Sysc }) return 0, err - case syscall.SIOCGIFCONF: + case linux.SIOCGIFCONF: // Return a list of interface addresses or the buffer size // necessary to hold the list. var ifc linux.IFConf @@ -2889,7 +2890,7 @@ func interfaceIoctl(ctx context.Context, io usermem.IO, arg int, ifr *linux.IFRe // SIOCGIFNAME uses ifr.ifr_ifindex rather than ifr.ifr_name to // identify a device. - if arg == syscall.SIOCGIFNAME { + if arg == linux.SIOCGIFNAME { // Gets the name of the interface given the interface index // stored in ifr_ifindex. index = int32(usermem.ByteOrder.Uint32(ifr.Data[:4])) @@ -2912,21 +2913,28 @@ func interfaceIoctl(ctx context.Context, io usermem.IO, arg int, ifr *linux.IFRe } switch arg { - case syscall.SIOCGIFINDEX: + case linux.SIOCGIFINDEX: // Copy out the index to the data. usermem.ByteOrder.PutUint32(ifr.Data[:], uint32(index)) - case syscall.SIOCGIFHWADDR: + case linux.SIOCGIFHWADDR: // Copy the hardware address out. - ifr.Data[0] = 6 // IEEE802.2 arp type. - ifr.Data[1] = 0 + // + // Refer: https://linux.die.net/man/7/netdevice + // SIOCGIFHWADDR, SIOCSIFHWADDR + // + // Get or set the hardware address of a device using + // ifr_hwaddr. The hardware address is specified in a struct + // sockaddr. sa_family contains the ARPHRD_* device type, + // sa_data the L2 hardware address starting from byte 0. Setting + // the hardware address is a privileged operation. + usermem.ByteOrder.PutUint16(ifr.Data[:], iface.DeviceType) n := copy(ifr.Data[2:], iface.Addr) for i := 2 + n; i < len(ifr.Data); i++ { ifr.Data[i] = 0 // Clear padding. } - usermem.ByteOrder.PutUint16(ifr.Data[:2], uint16(n)) - case syscall.SIOCGIFFLAGS: + case linux.SIOCGIFFLAGS: f, err := interfaceStatusFlags(stack, iface.Name) if err != nil { return err @@ -2935,7 +2943,7 @@ func interfaceIoctl(ctx context.Context, io usermem.IO, arg int, ifr *linux.IFRe // matches Linux behavior. usermem.ByteOrder.PutUint16(ifr.Data[:2], uint16(f)) - case syscall.SIOCGIFADDR: + case linux.SIOCGIFADDR: // Copy the IPv4 address out. for _, addr := range stack.InterfaceAddrs()[index] { // This ioctl is only compatible with AF_INET addresses. @@ -2946,32 +2954,32 @@ func interfaceIoctl(ctx context.Context, io usermem.IO, arg int, ifr *linux.IFRe break } - case syscall.SIOCGIFMETRIC: + case linux.SIOCGIFMETRIC: // Gets the metric of the device. As per netdevice(7), this // always just sets ifr_metric to 0. usermem.ByteOrder.PutUint32(ifr.Data[:4], 0) - case syscall.SIOCGIFMTU: + case linux.SIOCGIFMTU: // Gets the MTU of the device. usermem.ByteOrder.PutUint32(ifr.Data[:4], iface.MTU) - case syscall.SIOCGIFMAP: + case linux.SIOCGIFMAP: // Gets the hardware parameters of the device. // TODO(gvisor.dev/issue/505): Implement. - case syscall.SIOCGIFTXQLEN: + case linux.SIOCGIFTXQLEN: // Gets the transmit queue length of the device. // TODO(gvisor.dev/issue/505): Implement. - case syscall.SIOCGIFDSTADDR: + case linux.SIOCGIFDSTADDR: // Gets the destination address of a point-to-point device. // TODO(gvisor.dev/issue/505): Implement. - case syscall.SIOCGIFBRDADDR: + case linux.SIOCGIFBRDADDR: // Gets the broadcast address of a device. // TODO(gvisor.dev/issue/505): Implement. - case syscall.SIOCGIFNETMASK: + case linux.SIOCGIFNETMASK: // Gets the network mask of a device. for _, addr := range stack.InterfaceAddrs()[index] { // This ioctl is only compatible with AF_INET addresses. @@ -2988,6 +2996,14 @@ func interfaceIoctl(ctx context.Context, io usermem.IO, arg int, ifr *linux.IFRe break } + case linux.SIOCETHTOOL: + // Stubbed out for now, Ideally we should implement the required + // sub-commands for ETHTOOL + // + // See: + // https://github.com/torvalds/linux/blob/aa0c9086b40c17a7ad94425b3b70dd1fdd7497bf/net/core/dev_ioctl.c + return syserr.ErrEndpointOperation + default: // Not a valid call. return syserr.ErrInvalidArgument diff --git a/pkg/sentry/socket/netstack/stack.go b/pkg/sentry/socket/netstack/stack.go index 548442b96..67737ae87 100644 --- a/pkg/sentry/socket/netstack/stack.go +++ b/pkg/sentry/socket/netstack/stack.go @@ -15,6 +15,8 @@ package netstack import ( + "fmt" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/inet" @@ -40,19 +42,29 @@ func (s *Stack) SupportsIPv6() bool { return s.Stack.CheckNetworkProtocol(ipv6.ProtocolNumber) } +// Converts Netstack's ARPHardwareType to equivalent linux constants. +func toLinuxARPHardwareType(t header.ARPHardwareType) uint16 { + switch t { + case header.ARPHardwareNone: + return linux.ARPHRD_NONE + case header.ARPHardwareLoopback: + return linux.ARPHRD_LOOPBACK + case header.ARPHardwareEther: + return linux.ARPHRD_ETHER + default: + panic(fmt.Sprintf("unknown ARPHRD type: %d", t)) + } +} + // Interfaces implements inet.Stack.Interfaces. func (s *Stack) Interfaces() map[int32]inet.Interface { is := make(map[int32]inet.Interface) for id, ni := range s.Stack.NICInfo() { - var devType uint16 - if ni.Flags.Loopback { - devType = linux.ARPHRD_LOOPBACK - } is[int32(id)] = inet.Interface{ Name: ni.Name, Addr: []byte(ni.LinkAddress), Flags: uint32(nicStateFlagsToLinux(ni.Flags)), - DeviceType: devType, + DeviceType: toLinuxARPHardwareType(ni.ARPHardwareType), MTU: ni.MTU, } } diff --git a/pkg/tcpip/header/arp.go b/pkg/tcpip/header/arp.go index 718a4720a..83189676e 100644 --- a/pkg/tcpip/header/arp.go +++ b/pkg/tcpip/header/arp.go @@ -14,14 +14,33 @@ package header -import "gvisor.dev/gvisor/pkg/tcpip" +import ( + "encoding/binary" + + "gvisor.dev/gvisor/pkg/tcpip" +) const ( // ARPProtocolNumber is the ARP network protocol number. ARPProtocolNumber tcpip.NetworkProtocolNumber = 0x0806 // ARPSize is the size of an IPv4-over-Ethernet ARP packet. - ARPSize = 2 + 2 + 1 + 1 + 2 + 2*6 + 2*4 + ARPSize = 28 +) + +// ARPHardwareType is the hardware type for LinkEndpoint in an ARP header. +type ARPHardwareType uint16 + +// Typical ARP HardwareType values. Some of the constants have to be specific +// values as they are egressed on the wire in the HTYPE field of an ARP header. +const ( + ARPHardwareNone ARPHardwareType = 0 + // ARPHardwareEther specifically is the HTYPE for Ethernet as specified + // in the IANA list here: + // + // https://www.iana.org/assignments/arp-parameters/arp-parameters.xhtml#arp-parameters-2 + ARPHardwareEther ARPHardwareType = 1 + ARPHardwareLoopback ARPHardwareType = 2 ) // ARPOp is an ARP opcode. @@ -36,54 +55,64 @@ const ( // ARP is an ARP packet stored in a byte array as described in RFC 826. type ARP []byte -func (a ARP) hardwareAddressSpace() uint16 { return uint16(a[0])<<8 | uint16(a[1]) } -func (a ARP) protocolAddressSpace() uint16 { return uint16(a[2])<<8 | uint16(a[3]) } -func (a ARP) hardwareAddressSize() int { return int(a[4]) } -func (a ARP) protocolAddressSize() int { return int(a[5]) } +const ( + hTypeOffset = 0 + protocolOffset = 2 + haAddressSizeOffset = 4 + protoAddressSizeOffset = 5 + opCodeOffset = 6 + senderHAAddressOffset = 8 + senderProtocolAddressOffset = senderHAAddressOffset + EthernetAddressSize + targetHAAddressOffset = senderProtocolAddressOffset + IPv4AddressSize + targetProtocolAddressOffset = targetHAAddressOffset + EthernetAddressSize +) + +func (a ARP) hardwareAddressType() ARPHardwareType { + return ARPHardwareType(binary.BigEndian.Uint16(a[hTypeOffset:])) +} + +func (a ARP) protocolAddressSpace() uint16 { return binary.BigEndian.Uint16(a[protocolOffset:]) } +func (a ARP) hardwareAddressSize() int { return int(a[haAddressSizeOffset]) } +func (a ARP) protocolAddressSize() int { return int(a[protoAddressSizeOffset]) } // Op is the ARP opcode. -func (a ARP) Op() ARPOp { return ARPOp(a[6])<<8 | ARPOp(a[7]) } +func (a ARP) Op() ARPOp { return ARPOp(binary.BigEndian.Uint16(a[opCodeOffset:])) } // SetOp sets the ARP opcode. func (a ARP) SetOp(op ARPOp) { - a[6] = uint8(op >> 8) - a[7] = uint8(op) + binary.BigEndian.PutUint16(a[opCodeOffset:], uint16(op)) } // SetIPv4OverEthernet configures the ARP packet for IPv4-over-Ethernet. func (a ARP) SetIPv4OverEthernet() { - a[0], a[1] = 0, 1 // htypeEthernet - a[2], a[3] = 0x08, 0x00 // IPv4ProtocolNumber - a[4] = 6 // macSize - a[5] = uint8(IPv4AddressSize) + binary.BigEndian.PutUint16(a[hTypeOffset:], uint16(ARPHardwareEther)) + binary.BigEndian.PutUint16(a[protocolOffset:], uint16(IPv4ProtocolNumber)) + a[haAddressSizeOffset] = EthernetAddressSize + a[protoAddressSizeOffset] = uint8(IPv4AddressSize) } // HardwareAddressSender is the link address of the sender. // It is a view on to the ARP packet so it can be used to set the value. func (a ARP) HardwareAddressSender() []byte { - const s = 8 - return a[s : s+6] + return a[senderHAAddressOffset : senderHAAddressOffset+EthernetAddressSize] } // ProtocolAddressSender is the protocol address of the sender. // It is a view on to the ARP packet so it can be used to set the value. func (a ARP) ProtocolAddressSender() []byte { - const s = 8 + 6 - return a[s : s+4] + return a[senderProtocolAddressOffset : senderProtocolAddressOffset+IPv4AddressSize] } // HardwareAddressTarget is the link address of the target. // It is a view on to the ARP packet so it can be used to set the value. func (a ARP) HardwareAddressTarget() []byte { - const s = 8 + 6 + 4 - return a[s : s+6] + return a[targetHAAddressOffset : targetHAAddressOffset+EthernetAddressSize] } // ProtocolAddressTarget is the protocol address of the target. // It is a view on to the ARP packet so it can be used to set the value. func (a ARP) ProtocolAddressTarget() []byte { - const s = 8 + 6 + 4 + 6 - return a[s : s+4] + return a[targetProtocolAddressOffset : targetProtocolAddressOffset+IPv4AddressSize] } // IsValid reports whether this is an ARP packet for IPv4 over Ethernet. @@ -91,10 +120,8 @@ func (a ARP) IsValid() bool { if len(a) < ARPSize { return false } - const htypeEthernet = 1 - const macSize = 6 - return a.hardwareAddressSpace() == htypeEthernet && + return a.hardwareAddressType() == ARPHardwareEther && a.protocolAddressSpace() == uint16(IPv4ProtocolNumber) && - a.hardwareAddressSize() == macSize && + a.hardwareAddressSize() == EthernetAddressSize && a.protocolAddressSize() == IPv4AddressSize } diff --git a/pkg/tcpip/link/channel/BUILD b/pkg/tcpip/link/channel/BUILD index b8b93e78e..39ca774ef 100644 --- a/pkg/tcpip/link/channel/BUILD +++ b/pkg/tcpip/link/channel/BUILD @@ -10,6 +10,7 @@ go_library( "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", + "//pkg/tcpip/header", "//pkg/tcpip/stack", ], ) diff --git a/pkg/tcpip/link/channel/channel.go b/pkg/tcpip/link/channel/channel.go index 20b183da0..a2bb773d4 100644 --- a/pkg/tcpip/link/channel/channel.go +++ b/pkg/tcpip/link/channel/channel.go @@ -23,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -296,3 +297,8 @@ func (e *Endpoint) AddNotify(notify Notification) *NotificationHandle { func (e *Endpoint) RemoveNotify(handle *NotificationHandle) { e.q.RemoveNotify(handle) } + +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType. +func (*Endpoint) ARPHardwareType() header.ARPHardwareType { + return header.ARPHardwareNone +} diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go index f34082e1a..32abe2a13 100644 --- a/pkg/tcpip/link/fdbased/endpoint.go +++ b/pkg/tcpip/link/fdbased/endpoint.go @@ -626,6 +626,14 @@ func (e *endpoint) GSOMaxSize() uint32 { return e.gsoMaxSize } +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType. +func (e *endpoint) ARPHardwareType() header.ARPHardwareType { + if e.hdrSize > 0 { + return header.ARPHardwareEther + } + return header.ARPHardwareNone +} + // InjectableEndpoint is an injectable fd-based endpoint. The endpoint writes // to the FD, but does not read from it. All reads come from injected packets. type InjectableEndpoint struct { diff --git a/pkg/tcpip/link/loopback/loopback.go b/pkg/tcpip/link/loopback/loopback.go index 568c6874f..3b17d8c28 100644 --- a/pkg/tcpip/link/loopback/loopback.go +++ b/pkg/tcpip/link/loopback/loopback.go @@ -113,3 +113,8 @@ func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { return nil } + +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType. +func (*endpoint) ARPHardwareType() header.ARPHardwareType { + return header.ARPHardwareLoopback +} diff --git a/pkg/tcpip/link/muxed/BUILD b/pkg/tcpip/link/muxed/BUILD index 82b441b79..e7493e5c5 100644 --- a/pkg/tcpip/link/muxed/BUILD +++ b/pkg/tcpip/link/muxed/BUILD @@ -9,6 +9,7 @@ go_library( deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", + "//pkg/tcpip/header", "//pkg/tcpip/stack", ], ) diff --git a/pkg/tcpip/link/muxed/injectable.go b/pkg/tcpip/link/muxed/injectable.go index c69d6b7e9..c305d9e86 100644 --- a/pkg/tcpip/link/muxed/injectable.go +++ b/pkg/tcpip/link/muxed/injectable.go @@ -18,6 +18,7 @@ package muxed import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -129,6 +130,11 @@ func (m *InjectableEndpoint) Wait() { } } +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType. +func (*InjectableEndpoint) ARPHardwareType() header.ARPHardwareType { + panic("unsupported operation") +} + // NewInjectableEndpoint creates a new multi-endpoint injectable endpoint. func NewInjectableEndpoint(routes map[tcpip.Address]stack.InjectableLinkEndpoint) *InjectableEndpoint { return &InjectableEndpoint{ diff --git a/pkg/tcpip/link/nested/BUILD b/pkg/tcpip/link/nested/BUILD index bdd5276ad..2cdb23475 100644 --- a/pkg/tcpip/link/nested/BUILD +++ b/pkg/tcpip/link/nested/BUILD @@ -12,6 +12,7 @@ go_library( "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", + "//pkg/tcpip/header", "//pkg/tcpip/stack", ], ) diff --git a/pkg/tcpip/link/nested/nested.go b/pkg/tcpip/link/nested/nested.go index 2998f9c4f..328bd048e 100644 --- a/pkg/tcpip/link/nested/nested.go +++ b/pkg/tcpip/link/nested/nested.go @@ -20,6 +20,7 @@ import ( "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -129,3 +130,8 @@ func (e *Endpoint) GSOMaxSize() uint32 { } return 0 } + +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType +func (e *Endpoint) ARPHardwareType() header.ARPHardwareType { + return e.child.ARPHardwareType() +} diff --git a/pkg/tcpip/link/qdisc/fifo/BUILD b/pkg/tcpip/link/qdisc/fifo/BUILD index 054c213bc..1d0079bd6 100644 --- a/pkg/tcpip/link/qdisc/fifo/BUILD +++ b/pkg/tcpip/link/qdisc/fifo/BUILD @@ -14,6 +14,7 @@ go_library( "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", + "//pkg/tcpip/header", "//pkg/tcpip/stack", ], ) diff --git a/pkg/tcpip/link/qdisc/fifo/endpoint.go b/pkg/tcpip/link/qdisc/fifo/endpoint.go index b5dfb7850..c84fe1bb9 100644 --- a/pkg/tcpip/link/qdisc/fifo/endpoint.go +++ b/pkg/tcpip/link/qdisc/fifo/endpoint.go @@ -22,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -207,3 +208,8 @@ func (e *endpoint) Wait() { e.wg.Wait() } + +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType +func (e *endpoint) ARPHardwareType() header.ARPHardwareType { + return e.lower.ARPHardwareType() +} diff --git a/pkg/tcpip/link/sharedmem/sharedmem.go b/pkg/tcpip/link/sharedmem/sharedmem.go index 0374a2441..a36862c67 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem.go +++ b/pkg/tcpip/link/sharedmem/sharedmem.go @@ -287,3 +287,8 @@ func (e *endpoint) dispatchLoop(d stack.NetworkDispatcher) { e.completed.Done() } + +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType +func (*endpoint) ARPHardwareType() header.ARPHardwareType { + return header.ARPHardwareEther +} diff --git a/pkg/tcpip/link/tun/device.go b/pkg/tcpip/link/tun/device.go index 6bc9033d0..47446efec 100644 --- a/pkg/tcpip/link/tun/device.go +++ b/pkg/tcpip/link/tun/device.go @@ -139,6 +139,7 @@ func attachOrCreateNIC(s *stack.Stack, name, prefix string, linkCaps stack.LinkE stack: s, nicID: id, name: name, + isTap: prefix == "tap", } endpoint.Endpoint.LinkEPCapabilities = linkCaps if endpoint.name == "" { @@ -348,6 +349,7 @@ type tunEndpoint struct { stack *stack.Stack nicID tcpip.NICID name string + isTap bool } // DecRef decrements refcount of e, removes NIC if refcount goes to 0. @@ -356,3 +358,11 @@ func (e *tunEndpoint) DecRef() { e.stack.RemoveNIC(e.nicID) }) } + +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType. +func (e *tunEndpoint) ARPHardwareType() header.ARPHardwareType { + if e.isTap { + return header.ARPHardwareEther + } + return header.ARPHardwareNone +} diff --git a/pkg/tcpip/link/waitable/BUILD b/pkg/tcpip/link/waitable/BUILD index 0956d2c65..ee84c3d96 100644 --- a/pkg/tcpip/link/waitable/BUILD +++ b/pkg/tcpip/link/waitable/BUILD @@ -12,6 +12,7 @@ go_library( "//pkg/gate", "//pkg/tcpip", "//pkg/tcpip/buffer", + "//pkg/tcpip/header", "//pkg/tcpip/stack", ], ) @@ -25,6 +26,7 @@ go_test( deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", + "//pkg/tcpip/header", "//pkg/tcpip/stack", ], ) diff --git a/pkg/tcpip/link/waitable/waitable.go b/pkg/tcpip/link/waitable/waitable.go index 949b3f2b2..24a8dc2eb 100644 --- a/pkg/tcpip/link/waitable/waitable.go +++ b/pkg/tcpip/link/waitable/waitable.go @@ -25,6 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/gate" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -147,3 +148,8 @@ func (e *Endpoint) WaitDispatch() { // Wait implements stack.LinkEndpoint.Wait. func (e *Endpoint) Wait() {} + +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType. +func (e *Endpoint) ARPHardwareType() header.ARPHardwareType { + return e.lower.ARPHardwareType() +} diff --git a/pkg/tcpip/link/waitable/waitable_test.go b/pkg/tcpip/link/waitable/waitable_test.go index 63bf40562..ffb2354be 100644 --- a/pkg/tcpip/link/waitable/waitable_test.go +++ b/pkg/tcpip/link/waitable/waitable_test.go @@ -19,6 +19,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -81,6 +82,11 @@ func (e *countedEndpoint) WriteRawPacket(buffer.VectorisedView) *tcpip.Error { return nil } +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType. +func (*countedEndpoint) ARPHardwareType() header.ARPHardwareType { + panic("unimplemented") +} + // Wait implements stack.LinkEndpoint.Wait. func (*countedEndpoint) Wait() {} diff --git a/pkg/tcpip/network/ip_test.go b/pkg/tcpip/network/ip_test.go index 7c8fb3e0a..a5b780ca2 100644 --- a/pkg/tcpip/network/ip_test.go +++ b/pkg/tcpip/network/ip_test.go @@ -172,14 +172,19 @@ func (t *testObject) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.Ne } // WritePackets implements stack.LinkEndpoint.WritePackets. -func (t *testObject) WritePackets(_ *stack.Route, _ *stack.GSO, pkt stack.PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { +func (*testObject) WritePackets(_ *stack.Route, _ *stack.GSO, pkt stack.PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { panic("not implemented") } -func (t *testObject) WriteRawPacket(_ buffer.VectorisedView) *tcpip.Error { +func (*testObject) WriteRawPacket(_ buffer.VectorisedView) *tcpip.Error { return tcpip.ErrNotSupported } +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType. +func (*testObject) ARPHardwareType() header.ARPHardwareType { + panic("not implemented") +} + func buildIPv4Route(local, remote tcpip.Address) (stack.Route, *tcpip.Error) { s := stack.New(stack.Options{ NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, diff --git a/pkg/tcpip/stack/forwarder_test.go b/pkg/tcpip/stack/forwarder_test.go index a6546cef0..eefb4b07f 100644 --- a/pkg/tcpip/stack/forwarder_test.go +++ b/pkg/tcpip/stack/forwarder_test.go @@ -22,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" ) const ( @@ -301,6 +302,11 @@ func (e *fwdTestLinkEndpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Er // Wait implements stack.LinkEndpoint.Wait. func (*fwdTestLinkEndpoint) Wait() {} +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType. +func (*fwdTestLinkEndpoint) ARPHardwareType() header.ARPHardwareType { + panic("not implemented") +} + func fwdTestNetFactory(t *testing.T, proto *fwdTestNetworkProtocol) (ep1, ep2 *fwdTestLinkEndpoint) { // Create a stack with the network protocol and two NICs. s := New(Options{ diff --git a/pkg/tcpip/stack/nic_test.go b/pkg/tcpip/stack/nic_test.go index 31f865260..3bc9fd831 100644 --- a/pkg/tcpip/stack/nic_test.go +++ b/pkg/tcpip/stack/nic_test.go @@ -84,6 +84,11 @@ func (e *testLinkEndpoint) WriteRawPacket(buffer.VectorisedView) *tcpip.Error { return tcpip.ErrNotSupported } +// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType. +func (*testLinkEndpoint) ARPHardwareType() header.ARPHardwareType { + panic("not implemented") +} + var _ NetworkEndpoint = (*testIPv6Endpoint)(nil) // An IPv6 NetworkEndpoint that throws away outgoing packets. diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go index 5cbc946b6..f260eeb7f 100644 --- a/pkg/tcpip/stack/registration.go +++ b/pkg/tcpip/stack/registration.go @@ -18,6 +18,7 @@ import ( "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/waiter" ) @@ -436,6 +437,12 @@ type LinkEndpoint interface { // Wait will not block if the endpoint hasn't started any goroutines // yet, even if it might later. Wait() + + // ARPHardwareType returns the ARPHRD_TYPE of the link endpoint. + // + // See: + // https://github.com/torvalds/linux/blob/aa0c9086b40c17a7ad94425b3b70dd1fdd7497bf/include/uapi/linux/if_arp.h#L30 + ARPHardwareType() header.ARPHardwareType } // InjectableLinkEndpoint is a LinkEndpoint where inbound packets are diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index 0aa815447..2b7ece851 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -1095,6 +1095,11 @@ type NICInfo struct { // Context is user-supplied data optionally supplied in CreateNICWithOptions. // See type NICOptions for more details. Context NICContext + + // ARPHardwareType holds the ARP Hardware type of the NIC. This is the + // value sent in haType field of an ARP Request sent by this NIC and the + // value expected in the haType field of an ARP response. + ARPHardwareType header.ARPHardwareType } // HasNIC returns true if the NICID is defined in the stack. @@ -1126,6 +1131,7 @@ func (s *Stack) NICInfo() map[tcpip.NICID]NICInfo { MTU: nic.linkEP.MTU(), Stats: nic.stats, Context: nic.context, + ARPHardwareType: nic.linkEP.ARPHardwareType(), } } return nics diff --git a/test/syscalls/linux/socket_netdevice.cc b/test/syscalls/linux/socket_netdevice.cc index 15d4b85a7..5f8d7f981 100644 --- a/test/syscalls/linux/socket_netdevice.cc +++ b/test/syscalls/linux/socket_netdevice.cc @@ -12,6 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. +#include #include #include #include @@ -49,6 +50,7 @@ TEST(NetdeviceTest, Loopback) { // Check that the loopback is zero hardware address. ASSERT_THAT(ioctl(sock.get(), SIOCGIFHWADDR, &ifr), SyscallSucceeds()); + EXPECT_EQ(ifr.ifr_hwaddr.sa_family, ARPHRD_LOOPBACK); EXPECT_EQ(ifr.ifr_hwaddr.sa_data[0], 0); EXPECT_EQ(ifr.ifr_hwaddr.sa_data[1], 0); EXPECT_EQ(ifr.ifr_hwaddr.sa_data[2], 0); @@ -178,6 +180,27 @@ TEST(NetdeviceTest, InterfaceMTU) { EXPECT_GT(ifr.ifr_mtu, 0); } +TEST(NetdeviceTest, EthtoolGetTSInfo) { + FileDescriptor sock = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_DGRAM, 0)); + + struct ethtool_ts_info tsi = {}; + tsi.cmd = ETHTOOL_GET_TS_INFO; // Get NIC's Timestamping capabilities. + + // Prepare the request. + struct ifreq ifr = {}; + snprintf(ifr.ifr_name, IFNAMSIZ, "lo"); + ifr.ifr_data = (void*)&tsi; + + // Check that SIOCGIFMTU returns a nonzero MTU. + if (IsRunningOnGvisor()) { + ASSERT_THAT(ioctl(sock.get(), SIOCETHTOOL, &ifr), + SyscallFailsWithErrno(EOPNOTSUPP)); + return; + } + ASSERT_THAT(ioctl(sock.get(), SIOCETHTOOL, &ifr), SyscallSucceeds()); +} + } // namespace } // namespace testing -- cgit v1.2.3 From db653bb34baeba24b8b658eb0985e4c5185344cb Mon Sep 17 00:00:00 2001 From: Ting-Yu Wang Date: Wed, 15 Jul 2020 15:04:10 -0700 Subject: fdbased: Vectorized write for packet; relax writev syscall filter. Now it calls pkt.Data.ToView() when writing the packet. This may require copying when the packet is large, which puts the worse case in an even worse situation. This sent out in a separate preparation change as it requires syscall filter changes. This change will be followed by the change for the adoption of the new PacketHeader API. PiperOrigin-RevId: 321447003 --- pkg/iovec/BUILD | 18 +++++ pkg/iovec/iovec.go | 75 +++++++++++++++++++ pkg/iovec/iovec_test.go | 121 +++++++++++++++++++++++++++++++ pkg/sentry/fs/host/BUILD | 1 + pkg/sentry/fs/host/socket_iovec.go | 7 +- pkg/sentry/fsimpl/host/BUILD | 1 + pkg/sentry/fsimpl/host/socket_iovec.go | 7 +- pkg/tcpip/link/fdbased/BUILD | 1 + pkg/tcpip/link/fdbased/endpoint.go | 69 ++++++------------ pkg/tcpip/link/rawfile/rawfile_unsafe.go | 32 +------- runsc/boot/filter/config.go | 11 +-- 11 files changed, 248 insertions(+), 95 deletions(-) create mode 100644 pkg/iovec/BUILD create mode 100644 pkg/iovec/iovec.go create mode 100644 pkg/iovec/iovec_test.go (limited to 'pkg/sentry') diff --git a/pkg/iovec/BUILD b/pkg/iovec/BUILD new file mode 100644 index 000000000..eda82cfc1 --- /dev/null +++ b/pkg/iovec/BUILD @@ -0,0 +1,18 @@ +load("//tools:defs.bzl", "go_library", "go_test") + +package(licenses = ["notice"]) + +go_library( + name = "iovec", + srcs = ["iovec.go"], + visibility = ["//:sandbox"], + deps = ["//pkg/abi/linux"], +) + +go_test( + name = "iovec_test", + size = "small", + srcs = ["iovec_test.go"], + library = ":iovec", + deps = ["@org_golang_x_sys//unix:go_default_library"], +) diff --git a/pkg/iovec/iovec.go b/pkg/iovec/iovec.go new file mode 100644 index 000000000..dd70fe80f --- /dev/null +++ b/pkg/iovec/iovec.go @@ -0,0 +1,75 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build linux + +// Package iovec provides helpers to interact with vectorized I/O on host +// system. +package iovec + +import ( + "syscall" + + "gvisor.dev/gvisor/pkg/abi/linux" +) + +// MaxIovs is the maximum number of iovecs host platform can accept. +var MaxIovs = linux.UIO_MAXIOV + +// Builder is a builder for slice of syscall.Iovec. +type Builder struct { + iovec []syscall.Iovec + storage [8]syscall.Iovec + + // overflow tracks the last buffer when iovec length is at MaxIovs. + overflow []byte +} + +// Add adds buf to b preparing to be written. Zero-length buf won't be added. +func (b *Builder) Add(buf []byte) { + if len(buf) == 0 { + return + } + if b.iovec == nil { + b.iovec = b.storage[:0] + } + if len(b.iovec) >= MaxIovs { + b.addByAppend(buf) + return + } + b.iovec = append(b.iovec, syscall.Iovec{ + Base: &buf[0], + Len: uint64(len(buf)), + }) + // Keep the last buf if iovec is at max capacity. We will need to append to it + // for later bufs. + if len(b.iovec) == MaxIovs { + n := len(buf) + b.overflow = buf[:n:n] + } +} + +func (b *Builder) addByAppend(buf []byte) { + b.overflow = append(b.overflow, buf...) + b.iovec[len(b.iovec)-1] = syscall.Iovec{ + Base: &b.overflow[0], + Len: uint64(len(b.overflow)), + } +} + +// Build returns the final Iovec slice. The length of returned iovec will not +// excceed MaxIovs. +func (b *Builder) Build() []syscall.Iovec { + return b.iovec +} diff --git a/pkg/iovec/iovec_test.go b/pkg/iovec/iovec_test.go new file mode 100644 index 000000000..a3900c299 --- /dev/null +++ b/pkg/iovec/iovec_test.go @@ -0,0 +1,121 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build linux + +package iovec + +import ( + "bytes" + "fmt" + "syscall" + "testing" + "unsafe" + + "golang.org/x/sys/unix" +) + +func TestBuilderEmpty(t *testing.T) { + var builder Builder + iovecs := builder.Build() + if got, want := len(iovecs), 0; got != want { + t.Errorf("len(iovecs) = %d, want %d", got, want) + } +} + +func TestBuilderBuild(t *testing.T) { + a := []byte{1, 2} + b := []byte{3, 4, 5} + + var builder Builder + builder.Add(a) + builder.Add(b) + builder.Add(nil) // Nil slice won't be added. + builder.Add([]byte{}) // Empty slice won't be added. + iovecs := builder.Build() + + if got, want := len(iovecs), 2; got != want { + t.Fatalf("len(iovecs) = %d, want %d", got, want) + } + for i, data := range [][]byte{a, b} { + if got, want := *iovecs[i].Base, data[0]; got != want { + t.Fatalf("*iovecs[%d].Base = %d, want %d", i, got, want) + } + if got, want := iovecs[i].Len, uint64(len(data)); got != want { + t.Fatalf("iovecs[%d].Len = %d, want %d", i, got, want) + } + } +} + +func TestBuilderBuildMaxIov(t *testing.T) { + for _, test := range []struct { + numIov int + }{ + { + numIov: MaxIovs - 1, + }, + { + numIov: MaxIovs, + }, + { + numIov: MaxIovs + 1, + }, + { + numIov: MaxIovs + 10, + }, + } { + name := fmt.Sprintf("numIov=%v", test.numIov) + t.Run(name, func(t *testing.T) { + var data []byte + var builder Builder + for i := 0; i < test.numIov; i++ { + buf := []byte{byte(i)} + builder.Add(buf) + data = append(data, buf...) + } + iovec := builder.Build() + + // Check the expected length of iovec. + wantNum := test.numIov + if wantNum > MaxIovs { + wantNum = MaxIovs + } + if got, want := len(iovec), wantNum; got != want { + t.Errorf("len(iovec) = %d, want %d", got, want) + } + + // Test a real read-write. + var fds [2]int + if err := unix.Pipe(fds[:]); err != nil { + t.Fatalf("Pipe: %v", err) + } + defer syscall.Close(fds[0]) + defer syscall.Close(fds[1]) + + wrote, _, e := syscall.RawSyscall(syscall.SYS_WRITEV, uintptr(fds[1]), uintptr(unsafe.Pointer(&iovec[0])), uintptr(len(iovec))) + if int(wrote) != len(data) || e != 0 { + t.Fatalf("writev: %v, %v; want %v, 0", wrote, e, len(data)) + } + + got := make([]byte, len(data)) + if n, err := syscall.Read(fds[0], got); n != len(got) || err != nil { + t.Fatalf("read: %v, %v; want %v, nil", n, err, len(got)) + } + + if !bytes.Equal(got, data) { + t.Errorf("read: got data %v, want %v", got, data) + } + }) + } +} diff --git a/pkg/sentry/fs/host/BUILD b/pkg/sentry/fs/host/BUILD index aabce6cc9..d41d23a43 100644 --- a/pkg/sentry/fs/host/BUILD +++ b/pkg/sentry/fs/host/BUILD @@ -30,6 +30,7 @@ go_library( "//pkg/context", "//pkg/fd", "//pkg/fdnotifier", + "//pkg/iovec", "//pkg/log", "//pkg/refs", "//pkg/safemem", diff --git a/pkg/sentry/fs/host/socket_iovec.go b/pkg/sentry/fs/host/socket_iovec.go index 5c18dbd5e..905afb50d 100644 --- a/pkg/sentry/fs/host/socket_iovec.go +++ b/pkg/sentry/fs/host/socket_iovec.go @@ -17,15 +17,12 @@ package host import ( "syscall" - "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/iovec" "gvisor.dev/gvisor/pkg/syserror" ) // LINT.IfChange -// maxIovs is the maximum number of iovecs to pass to the host. -var maxIovs = linux.UIO_MAXIOV - // copyToMulti copies as many bytes from src to dst as possible. func copyToMulti(dst [][]byte, src []byte) { for _, d := range dst { @@ -76,7 +73,7 @@ func buildIovec(bufs [][]byte, maxlen int64, truncate bool) (length int64, iovec } } - if iovsRequired > maxIovs { + if iovsRequired > iovec.MaxIovs { // The kernel will reject our call if we pass this many iovs. // Use a single intermediate buffer instead. b := make([]byte, stopLen) diff --git a/pkg/sentry/fsimpl/host/BUILD b/pkg/sentry/fsimpl/host/BUILD index 44a09d87a..e86fbe2d5 100644 --- a/pkg/sentry/fsimpl/host/BUILD +++ b/pkg/sentry/fsimpl/host/BUILD @@ -22,6 +22,7 @@ go_library( "//pkg/context", "//pkg/fdnotifier", "//pkg/fspath", + "//pkg/iovec", "//pkg/log", "//pkg/refs", "//pkg/safemem", diff --git a/pkg/sentry/fsimpl/host/socket_iovec.go b/pkg/sentry/fsimpl/host/socket_iovec.go index 584c247d2..fc0d5fd38 100644 --- a/pkg/sentry/fsimpl/host/socket_iovec.go +++ b/pkg/sentry/fsimpl/host/socket_iovec.go @@ -17,13 +17,10 @@ package host import ( "syscall" - "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/iovec" "gvisor.dev/gvisor/pkg/syserror" ) -// maxIovs is the maximum number of iovecs to pass to the host. -var maxIovs = linux.UIO_MAXIOV - // copyToMulti copies as many bytes from src to dst as possible. func copyToMulti(dst [][]byte, src []byte) { for _, d := range dst { @@ -74,7 +71,7 @@ func buildIovec(bufs [][]byte, maxlen int64, truncate bool) (length int64, iovec } } - if iovsRequired > maxIovs { + if iovsRequired > iovec.MaxIovs { // The kernel will reject our call if we pass this many iovs. // Use a single intermediate buffer instead. b := make([]byte, stopLen) diff --git a/pkg/tcpip/link/fdbased/BUILD b/pkg/tcpip/link/fdbased/BUILD index aa6db9aea..507b44abc 100644 --- a/pkg/tcpip/link/fdbased/BUILD +++ b/pkg/tcpip/link/fdbased/BUILD @@ -15,6 +15,7 @@ go_library( visibility = ["//visibility:public"], deps = [ "//pkg/binary", + "//pkg/iovec", "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go index 32abe2a13..6aa1badc7 100644 --- a/pkg/tcpip/link/fdbased/endpoint.go +++ b/pkg/tcpip/link/fdbased/endpoint.go @@ -45,6 +45,7 @@ import ( "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/iovec" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" @@ -406,6 +407,8 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne eth.Encode(ethHdr) } + var builder iovec.Builder + fd := e.fds[pkt.Hash%uint32(len(e.fds))] if e.Capabilities()&stack.CapabilityHardwareGSO != 0 { vnetHdr := virtioNetHdr{} @@ -430,29 +433,25 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne } vnetHdrBuf := binary.Marshal(make([]byte, 0, virtioNetHdrSize), binary.LittleEndian, vnetHdr) - return rawfile.NonBlockingWrite3(fd, vnetHdrBuf, pkt.Header.View(), pkt.Data.ToView()) + builder.Add(vnetHdrBuf) } - if pkt.Data.Size() == 0 { - return rawfile.NonBlockingWrite(fd, pkt.Header.View()) - } - if pkt.Header.UsedLength() == 0 { - return rawfile.NonBlockingWrite(fd, pkt.Data.ToView()) + builder.Add(pkt.Header.View()) + for _, v := range pkt.Data.Views() { + builder.Add(v) } - return rawfile.NonBlockingWrite3(fd, pkt.Header.View(), pkt.Data.ToView(), nil) + return rawfile.NonBlockingWriteIovec(fd, builder.Build()) } func (e *endpoint) sendBatch(batchFD int, batch []*stack.PacketBuffer) (int, *tcpip.Error) { // Send a batch of packets through batchFD. mmsgHdrs := make([]rawfile.MMsgHdr, 0, len(batch)) for _, pkt := range batch { - var ethHdrBuf []byte - iovLen := 0 + var eth header.Ethernet if e.hdrSize > 0 { // Add ethernet header if needed. - ethHdrBuf = make([]byte, header.EthernetMinimumSize) - eth := header.Ethernet(ethHdrBuf) + eth = make(header.Ethernet, header.EthernetMinimumSize) ethHdr := &header.EthernetFields{ DstAddr: pkt.EgressRoute.RemoteLinkAddress, Type: pkt.NetworkProtocolNumber, @@ -465,12 +464,11 @@ func (e *endpoint) sendBatch(batchFD int, batch []*stack.PacketBuffer) (int, *tc ethHdr.SrcAddr = e.addr } eth.Encode(ethHdr) - iovLen++ } - vnetHdr := virtioNetHdr{} var vnetHdrBuf []byte if e.Capabilities()&stack.CapabilityHardwareGSO != 0 { + vnetHdr := virtioNetHdr{} if pkt.GSOOptions != nil { vnetHdr.hdrLen = uint16(pkt.Header.UsedLength()) if pkt.GSOOptions.NeedsCsum { @@ -491,45 +489,20 @@ func (e *endpoint) sendBatch(batchFD int, batch []*stack.PacketBuffer) (int, *tc } } vnetHdrBuf = binary.Marshal(make([]byte, 0, virtioNetHdrSize), binary.LittleEndian, vnetHdr) - iovLen++ } - iovecs := make([]syscall.Iovec, iovLen+1+len(pkt.Data.Views())) + var builder iovec.Builder + builder.Add(vnetHdrBuf) + builder.Add(eth) + builder.Add(pkt.Header.View()) + for _, v := range pkt.Data.Views() { + builder.Add(v) + } + iovecs := builder.Build() + var mmsgHdr rawfile.MMsgHdr mmsgHdr.Msg.Iov = &iovecs[0] - iovecIdx := 0 - if vnetHdrBuf != nil { - v := &iovecs[iovecIdx] - v.Base = &vnetHdrBuf[0] - v.Len = uint64(len(vnetHdrBuf)) - iovecIdx++ - } - if ethHdrBuf != nil { - v := &iovecs[iovecIdx] - v.Base = ðHdrBuf[0] - v.Len = uint64(len(ethHdrBuf)) - iovecIdx++ - } - pktSize := uint64(0) - // Encode L3 Header - v := &iovecs[iovecIdx] - hdr := &pkt.Header - hdrView := hdr.View() - v.Base = &hdrView[0] - v.Len = uint64(len(hdrView)) - pktSize += v.Len - iovecIdx++ - - // Now encode the Transport Payload. - pktViews := pkt.Data.Views() - for i := range pktViews { - vec := &iovecs[iovecIdx] - iovecIdx++ - vec.Base = &pktViews[i][0] - vec.Len = uint64(len(pktViews[i])) - pktSize += vec.Len - } - mmsgHdr.Msg.Iovlen = uint64(iovecIdx) + mmsgHdr.Msg.Iovlen = uint64(len(iovecs)) mmsgHdrs = append(mmsgHdrs, mmsgHdr) } diff --git a/pkg/tcpip/link/rawfile/rawfile_unsafe.go b/pkg/tcpip/link/rawfile/rawfile_unsafe.go index 69de6eb3e..f4c32c2da 100644 --- a/pkg/tcpip/link/rawfile/rawfile_unsafe.go +++ b/pkg/tcpip/link/rawfile/rawfile_unsafe.go @@ -66,38 +66,14 @@ func NonBlockingWrite(fd int, buf []byte) *tcpip.Error { return nil } -// NonBlockingWrite3 writes up to three byte slices to a file descriptor in a -// single syscall. It fails if partial data is written. -func NonBlockingWrite3(fd int, b1, b2, b3 []byte) *tcpip.Error { - // If there is no second and third buffer, issue a regular write. - if len(b2) == 0 && len(b3) == 0 { - return NonBlockingWrite(fd, b1) - } - - // Build the iovec that represents them and issue a writev syscall. - iovec := [3]syscall.Iovec{ - { - Base: &b1[0], - Len: uint64(len(b1)), - }, - { - Base: &b2[0], - Len: uint64(len(b2)), - }, - } - iovecLen := uintptr(2) - - if len(b3) > 0 { - iovecLen++ - iovec[2].Base = &b3[0] - iovec[2].Len = uint64(len(b3)) - } - +// NonBlockingWriteIovec writes iovec to a file descriptor in a single syscall. +// It fails if partial data is written. +func NonBlockingWriteIovec(fd int, iovec []syscall.Iovec) *tcpip.Error { + iovecLen := uintptr(len(iovec)) _, _, e := syscall.RawSyscall(syscall.SYS_WRITEV, uintptr(fd), uintptr(unsafe.Pointer(&iovec[0])), iovecLen) if e != 0 { return TranslateErrno(e) } - return nil } diff --git a/runsc/boot/filter/config.go b/runsc/boot/filter/config.go index 60e33425f..149eb0b1b 100644 --- a/runsc/boot/filter/config.go +++ b/runsc/boot/filter/config.go @@ -310,19 +310,12 @@ var allowedSyscalls = seccomp.SyscallRules{ }, }, syscall.SYS_WRITE: {}, - // The only user in rawfile.NonBlockingWrite3 always passes iovcnt with - // values 2 or 3. Three iovec-s are passed, when the PACKET_VNET_HDR - // option is enabled for a packet socket. + // For rawfile.NonBlockingWriteIovec. syscall.SYS_WRITEV: []seccomp.Rule{ { seccomp.AllowAny{}, seccomp.AllowAny{}, - seccomp.AllowValue(2), - }, - { - seccomp.AllowAny{}, - seccomp.AllowAny{}, - seccomp.AllowValue(3), + seccomp.GreaterThan(0), }, }, } -- cgit v1.2.3 From dcf6ddc2772b8fcf824f1f48e0281e1cc80b93ea Mon Sep 17 00:00:00 2001 From: Bhasker Hariharan Date: Thu, 16 Jul 2020 18:38:28 -0700 Subject: Add support to return protocol in recvmsg for AF_PACKET. Updates #173 PiperOrigin-RevId: 321690756 --- pkg/sentry/socket/netstack/netstack.go | 24 +++++++++++++++++++++--- pkg/tcpip/tcpip.go | 19 +++++++++++++++++++ pkg/tcpip/transport/packet/endpoint.go | 18 ++++++++++++++++-- test/syscalls/linux/packet_socket.cc | 1 + test/syscalls/linux/packet_socket_raw.cc | 1 + 5 files changed, 58 insertions(+), 5 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 0b1be1bd2..49a04e613 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -297,8 +297,9 @@ type socketOpsCommon struct { readView buffer.View // readCM holds control message information for the last packet read // from Endpoint. - readCM tcpip.ControlMessages - sender tcpip.FullAddress + readCM tcpip.ControlMessages + sender tcpip.FullAddress + linkPacketInfo tcpip.LinkPacketInfo // sockOptTimestamp corresponds to SO_TIMESTAMP. When true, timestamps // of returned messages can be returned via control messages. When @@ -447,8 +448,21 @@ func (s *socketOpsCommon) fetchReadView() *syserr.Error { } s.readView = nil s.sender = tcpip.FullAddress{} + s.linkPacketInfo = tcpip.LinkPacketInfo{} - v, cms, err := s.Endpoint.Read(&s.sender) + var v buffer.View + var cms tcpip.ControlMessages + var err *tcpip.Error + + switch e := s.Endpoint.(type) { + // The ordering of these interfaces matters. The most specific + // interfaces must be specified before the more generic Endpoint + // interface. + case tcpip.PacketEndpoint: + v, cms, err = e.ReadPacket(&s.sender, &s.linkPacketInfo) + case tcpip.Endpoint: + v, cms, err = e.Read(&s.sender) + } if err != nil { atomic.StoreUint32(&s.readViewHasData, 0) return syserr.TranslateNetstackError(err) @@ -2509,6 +2523,10 @@ func (s *socketOpsCommon) nonBlockingRead(ctx context.Context, dst usermem.IOSeq var addrLen uint32 if isPacket && senderRequested { addr, addrLen = ConvertAddress(s.family, s.sender) + switch v := addr.(type) { + case *linux.SockAddrLink: + v.Protocol = htons(uint16(s.linkPacketInfo.Protocol)) + } } if peek { diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 71bcee785..48ad56d4d 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -549,6 +549,25 @@ type Endpoint interface { SetOwner(owner PacketOwner) } +// LinkPacketInfo holds Link layer information for a received packet. +// +// +stateify savable +type LinkPacketInfo struct { + // Protocol is the NetworkProtocolNumber for the packet. + Protocol NetworkProtocolNumber +} + +// PacketEndpoint are additional methods that are only implemented by Packet +// endpoints. +type PacketEndpoint interface { + // ReadPacket reads a datagram/packet from the endpoint and optionally + // returns the sender and additional LinkPacketInfo. + // + // This method does not block if there is no data pending. It will also + // either return an error or data, never both. + ReadPacket(*FullAddress, *LinkPacketInfo) (buffer.View, ControlMessages, *Error) +} + // EndpointInfo is the interface implemented by each endpoint info struct. type EndpointInfo interface { // IsEndpointInfo is an empty method to implement the tcpip.EndpointInfo diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go index 92b487381..7b2083a09 100644 --- a/pkg/tcpip/transport/packet/endpoint.go +++ b/pkg/tcpip/transport/packet/endpoint.go @@ -45,6 +45,9 @@ type packet struct { timestampNS int64 // senderAddr is the network address of the sender. senderAddr tcpip.FullAddress + // packetInfo holds additional information like the protocol + // of the packet etc. + packetInfo tcpip.LinkPacketInfo } // endpoint is the packet socket implementation of tcpip.Endpoint. It is legal @@ -151,8 +154,8 @@ func (ep *endpoint) Close() { // ModerateRecvBuf implements tcpip.Endpoint.ModerateRecvBuf. func (ep *endpoint) ModerateRecvBuf(copied int) {} -// Read implements tcpip.Endpoint.Read. -func (ep *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) { +// Read implements tcpip.PacketEndpoint.ReadPacket. +func (ep *endpoint) ReadPacket(addr *tcpip.FullAddress, info *tcpip.LinkPacketInfo) (buffer.View, tcpip.ControlMessages, *tcpip.Error) { ep.rcvMu.Lock() // If there's no data to read, return that read would block or that the @@ -177,9 +180,18 @@ func (ep *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMes *addr = packet.senderAddr } + if info != nil { + *info = packet.packetInfo + } + return packet.data.ToView(), tcpip.ControlMessages{HasTimestamp: true, Timestamp: packet.timestampNS}, nil } +// Read implements tcpip.Endpoint.Read. +func (ep *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) { + return ep.ReadPacket(addr, nil) +} + func (ep *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { // TODO(b/129292371): Implement. return 0, nil, tcpip.ErrInvalidOptionValue @@ -428,12 +440,14 @@ func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress, NIC: nicID, Addr: tcpip.Address(hdr.SourceAddress()), } + packet.packetInfo.Protocol = netProto } else { // Guess the would-be ethernet header. packet.senderAddr = tcpip.FullAddress{ NIC: nicID, Addr: tcpip.Address(localAddr), } + packet.packetInfo.Protocol = netProto } if ep.cooked { diff --git a/test/syscalls/linux/packet_socket.cc b/test/syscalls/linux/packet_socket.cc index ce63adb23..e94ddcb77 100644 --- a/test/syscalls/linux/packet_socket.cc +++ b/test/syscalls/linux/packet_socket.cc @@ -193,6 +193,7 @@ void ReceiveMessage(int sock, int ifindex) { EXPECT_EQ(src.sll_family, AF_PACKET); EXPECT_EQ(src.sll_ifindex, ifindex); EXPECT_EQ(src.sll_halen, ETH_ALEN); + EXPECT_EQ(ntohs(src.sll_protocol), ETH_P_IP); // This came from the loopback device, so the address is all 0s. for (int i = 0; i < src.sll_halen; i++) { EXPECT_EQ(src.sll_addr[i], 0); diff --git a/test/syscalls/linux/packet_socket_raw.cc b/test/syscalls/linux/packet_socket_raw.cc index e44062475..2fca9fe4d 100644 --- a/test/syscalls/linux/packet_socket_raw.cc +++ b/test/syscalls/linux/packet_socket_raw.cc @@ -200,6 +200,7 @@ TEST_P(RawPacketTest, Receive) { EXPECT_EQ(src.sll_family, AF_PACKET); EXPECT_EQ(src.sll_ifindex, GetLoopbackIndex()); EXPECT_EQ(src.sll_halen, ETH_ALEN); + EXPECT_EQ(ntohs(src.sll_protocol), ETH_P_IP); // This came from the loopback device, so the address is all 0s. for (int i = 0; i < src.sll_halen; i++) { EXPECT_EQ(src.sll_addr[i], 0); -- cgit v1.2.3 From 8c9156fed69ef791e6f7fd7f3a611b74333e3e18 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Mon, 20 Jul 2020 01:06:46 -0400 Subject: add asid support to Arm64 Support the operation of asid, so that I can optimize tlb performance by combining with nG. Signed-off-by: Bin Lu --- pkg/sentry/platform/kvm/kvm_arm64_unsafe.go | 4 +++- pkg/sentry/platform/kvm/machine_arm64_unsafe.go | 16 ++++++++++++++++ 2 files changed, 19 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/kvm/kvm_arm64_unsafe.go b/pkg/sentry/platform/kvm/kvm_arm64_unsafe.go index 6531bae1d..48ccf8474 100644 --- a/pkg/sentry/platform/kvm/kvm_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/kvm_arm64_unsafe.go @@ -22,7 +22,8 @@ import ( ) var ( - runDataSize int + runDataSize int + hasGuestPCID bool ) func updateSystemValues(fd int) error { @@ -33,6 +34,7 @@ func updateSystemValues(fd int) error { } // Save the data. runDataSize = int(sz) + hasGuestPCID = true // Success. return nil diff --git a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go index 3de309c1a..ff8c068c0 100644 --- a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go @@ -26,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/ring0" + "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" "gvisor.dev/gvisor/pkg/usermem" ) @@ -156,6 +157,14 @@ func (c *vCPU) initArchState() error { return err } + // Initialize the PCID database. + if hasGuestPCID { + // Note that NewPCIDs may return a nil table here, in which + // case we simply don't use PCID support (see below). In + // practice, this should not happen, however. + c.PCIDs = pagetables.NewPCIDs(fixedKernelPCID+1, poolPCIDs) + } + c.floatingPointState = arch.NewFloatingPointData() return nil } @@ -234,6 +243,13 @@ func (c *vCPU) SwitchToUser(switchOpts ring0.SwitchOpts, info *arch.SignalInfo) return nonCanonical(regs.Sp, int32(syscall.SIGBUS), info) } + // Assign PCIDs. + if c.PCIDs != nil { + var requireFlushPCID bool // Force a flush? + switchOpts.UserASID, requireFlushPCID = c.PCIDs.Assign(switchOpts.PageTables) + switchOpts.Flush = switchOpts.Flush || requireFlushPCID + } + var vector ring0.Vector ttbr0App := switchOpts.PageTables.TTBR0_EL1(false, 0) c.SetTtbr0App(uintptr(ttbr0App)) -- cgit v1.2.3 From 71bf90c55bd888f9b9c493533ca5e4b2b4b3d21d Mon Sep 17 00:00:00 2001 From: Bhasker Hariharan Date: Wed, 22 Jul 2020 15:12:56 -0700 Subject: Support for receiving outbound packets in AF_PACKET. Updates #173 PiperOrigin-RevId: 322665518 --- pkg/abi/linux/socket.go | 9 +++ pkg/sentry/socket/netstack/netstack.go | 19 +++++ pkg/tcpip/link/channel/channel.go | 4 + pkg/tcpip/link/fdbased/endpoint.go | 36 ++++----- pkg/tcpip/link/fdbased/endpoint_test.go | 8 ++ pkg/tcpip/link/loopback/loopback.go | 3 + pkg/tcpip/link/muxed/injectable.go | 4 + pkg/tcpip/link/nested/nested.go | 15 ++++ pkg/tcpip/link/nested/nested_test.go | 4 + pkg/tcpip/link/packetsocket/BUILD | 14 ++++ pkg/tcpip/link/packetsocket/endpoint.go | 50 +++++++++++++ pkg/tcpip/link/qdisc/fifo/endpoint.go | 12 +++ pkg/tcpip/link/sharedmem/sharedmem.go | 21 ++++-- pkg/tcpip/link/sharedmem/sharedmem_test.go | 4 + pkg/tcpip/link/sniffer/sniffer.go | 5 ++ pkg/tcpip/link/tun/device.go | 43 +++++++---- pkg/tcpip/link/waitable/waitable.go | 14 ++++ pkg/tcpip/link/waitable/waitable_test.go | 9 +++ pkg/tcpip/network/ip_test.go | 5 ++ pkg/tcpip/stack/forwarder_test.go | 5 ++ pkg/tcpip/stack/nic.go | 30 ++++++-- pkg/tcpip/stack/nic_test.go | 5 ++ pkg/tcpip/stack/packet_buffer.go | 4 + pkg/tcpip/stack/registration.go | 16 +++- pkg/tcpip/tcpip.go | 25 +++++++ pkg/tcpip/transport/packet/endpoint.go | 52 +++++++++---- runsc/boot/BUILD | 1 + runsc/boot/network.go | 4 + test/syscalls/linux/packet_socket.cc | 116 +++++++++++++++++++++++++++++ 29 files changed, 471 insertions(+), 66 deletions(-) create mode 100644 pkg/tcpip/link/packetsocket/BUILD create mode 100644 pkg/tcpip/link/packetsocket/endpoint.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/socket.go b/pkg/abi/linux/socket.go index 4a14ef691..95337c168 100644 --- a/pkg/abi/linux/socket.go +++ b/pkg/abi/linux/socket.go @@ -134,6 +134,15 @@ const ( SHUT_RDWR = 2 ) +// Packet types from +const ( + PACKET_HOST = 0 // To us + PACKET_BROADCAST = 1 // To all + PACKET_MULTICAST = 2 // To group + PACKET_OTHERHOST = 3 // To someone else + PACKET_OUTGOING = 4 // Outgoing of any type +) + // Socket options from socket.h. const ( SO_DEBUG = 1 diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 49a04e613..964ec8414 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -26,6 +26,7 @@ package netstack import ( "bytes" + "fmt" "io" "math" "reflect" @@ -2468,6 +2469,23 @@ func (s *socketOpsCommon) fillCmsgInq(cmsg *socket.ControlMessages) { cmsg.IP.Inq = int32(len(s.readView) + rcvBufUsed) } +func toLinuxPacketType(pktType tcpip.PacketType) uint8 { + switch pktType { + case tcpip.PacketHost: + return linux.PACKET_HOST + case tcpip.PacketOtherHost: + return linux.PACKET_OTHERHOST + case tcpip.PacketOutgoing: + return linux.PACKET_OUTGOING + case tcpip.PacketBroadcast: + return linux.PACKET_BROADCAST + case tcpip.PacketMulticast: + return linux.PACKET_MULTICAST + default: + panic(fmt.Sprintf("unknown packet type: %d", pktType)) + } +} + // nonBlockingRead issues a non-blocking read. // // TODO(b/78348848): Support timestamps for stream sockets. @@ -2526,6 +2544,7 @@ func (s *socketOpsCommon) nonBlockingRead(ctx context.Context, dst usermem.IOSeq switch v := addr.(type) { case *linux.SockAddrLink: v.Protocol = htons(uint16(s.linkPacketInfo.Protocol)) + v.PacketType = toLinuxPacketType(s.linkPacketInfo.PktType) } } diff --git a/pkg/tcpip/link/channel/channel.go b/pkg/tcpip/link/channel/channel.go index a2bb773d4..e12a5929b 100644 --- a/pkg/tcpip/link/channel/channel.go +++ b/pkg/tcpip/link/channel/channel.go @@ -302,3 +302,7 @@ func (e *Endpoint) RemoveNotify(handle *NotificationHandle) { func (*Endpoint) ARPHardwareType() header.ARPHardwareType { return header.ARPHardwareNone } + +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *Endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { +} diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go index 6aa1badc7..c18bb91fb 100644 --- a/pkg/tcpip/link/fdbased/endpoint.go +++ b/pkg/tcpip/link/fdbased/endpoint.go @@ -386,26 +386,33 @@ const ( _VIRTIO_NET_HDR_GSO_TCPV6 = 4 ) -// WritePacket writes outbound packets to the file descriptor. If it is not -// currently writable, the packet is dropped. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { if e.hdrSize > 0 { // Add ethernet header if needed. eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize)) pkt.LinkHeader = buffer.View(eth) ethHdr := &header.EthernetFields{ - DstAddr: r.RemoteLinkAddress, + DstAddr: remote, Type: protocol, } // Preserve the src address if it's set in the route. - if r.LocalLinkAddress != "" { - ethHdr.SrcAddr = r.LocalLinkAddress + if local != "" { + ethHdr.SrcAddr = local } else { ethHdr.SrcAddr = e.addr } eth.Encode(ethHdr) } +} + +// WritePacket writes outbound packets to the file descriptor. If it is not +// currently writable, the packet is dropped. +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { + if e.hdrSize > 0 { + e.AddHeader(r.LocalLinkAddress, r.RemoteLinkAddress, protocol, pkt) + } var builder iovec.Builder @@ -448,22 +455,8 @@ func (e *endpoint) sendBatch(batchFD int, batch []*stack.PacketBuffer) (int, *tc // Send a batch of packets through batchFD. mmsgHdrs := make([]rawfile.MMsgHdr, 0, len(batch)) for _, pkt := range batch { - var eth header.Ethernet if e.hdrSize > 0 { - // Add ethernet header if needed. - eth = make(header.Ethernet, header.EthernetMinimumSize) - ethHdr := &header.EthernetFields{ - DstAddr: pkt.EgressRoute.RemoteLinkAddress, - Type: pkt.NetworkProtocolNumber, - } - - // Preserve the src address if it's set in the route. - if pkt.EgressRoute.LocalLinkAddress != "" { - ethHdr.SrcAddr = pkt.EgressRoute.LocalLinkAddress - } else { - ethHdr.SrcAddr = e.addr - } - eth.Encode(ethHdr) + e.AddHeader(pkt.EgressRoute.LocalLinkAddress, pkt.EgressRoute.RemoteLinkAddress, pkt.NetworkProtocolNumber, pkt) } var vnetHdrBuf []byte @@ -493,7 +486,6 @@ func (e *endpoint) sendBatch(batchFD int, batch []*stack.PacketBuffer) (int, *tc var builder iovec.Builder builder.Add(vnetHdrBuf) - builder.Add(eth) builder.Add(pkt.Header.View()) for _, v := range pkt.Data.Views() { builder.Add(v) diff --git a/pkg/tcpip/link/fdbased/endpoint_test.go b/pkg/tcpip/link/fdbased/endpoint_test.go index 4bad930c7..7b995b85a 100644 --- a/pkg/tcpip/link/fdbased/endpoint_test.go +++ b/pkg/tcpip/link/fdbased/endpoint_test.go @@ -107,6 +107,10 @@ func (c *context) DeliverNetworkPacket(remote tcpip.LinkAddress, local tcpip.Lin c.ch <- packetInfo{remote, protocol, pkt} } +func (c *context) DeliverOutboundPacket(remote tcpip.LinkAddress, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + panic("unimplemented") +} + func TestNoEthernetProperties(t *testing.T) { c := newContext(t, &Options{MTU: mtu}) defer c.cleanup() @@ -510,6 +514,10 @@ func (d *fakeNetworkDispatcher) DeliverNetworkPacket(remote, local tcpip.LinkAdd d.pkts = append(d.pkts, pkt) } +func (d *fakeNetworkDispatcher) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + panic("unimplemented") +} + func TestDispatchPacketFormat(t *testing.T) { for _, test := range []struct { name string diff --git a/pkg/tcpip/link/loopback/loopback.go b/pkg/tcpip/link/loopback/loopback.go index 3b17d8c28..781cdd317 100644 --- a/pkg/tcpip/link/loopback/loopback.go +++ b/pkg/tcpip/link/loopback/loopback.go @@ -118,3 +118,6 @@ func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { func (*endpoint) ARPHardwareType() header.ARPHardwareType { return header.ARPHardwareLoopback } + +func (e *endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { +} diff --git a/pkg/tcpip/link/muxed/injectable.go b/pkg/tcpip/link/muxed/injectable.go index c305d9e86..56a611825 100644 --- a/pkg/tcpip/link/muxed/injectable.go +++ b/pkg/tcpip/link/muxed/injectable.go @@ -135,6 +135,10 @@ func (*InjectableEndpoint) ARPHardwareType() header.ARPHardwareType { panic("unsupported operation") } +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (*InjectableEndpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { +} + // NewInjectableEndpoint creates a new multi-endpoint injectable endpoint. func NewInjectableEndpoint(routes map[tcpip.Address]stack.InjectableLinkEndpoint) *InjectableEndpoint { return &InjectableEndpoint{ diff --git a/pkg/tcpip/link/nested/nested.go b/pkg/tcpip/link/nested/nested.go index 328bd048e..d40de54df 100644 --- a/pkg/tcpip/link/nested/nested.go +++ b/pkg/tcpip/link/nested/nested.go @@ -61,6 +61,16 @@ func (e *Endpoint) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protoco } } +// DeliverOutboundPacket implements stack.NetworkDispatcher.DeliverOutboundPacket. +func (e *Endpoint) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + e.mu.RLock() + d := e.dispatcher + e.mu.RUnlock() + if d != nil { + d.DeliverOutboundPacket(remote, local, protocol, pkt) + } +} + // Attach implements stack.LinkEndpoint. func (e *Endpoint) Attach(dispatcher stack.NetworkDispatcher) { e.mu.Lock() @@ -135,3 +145,8 @@ func (e *Endpoint) GSOMaxSize() uint32 { func (e *Endpoint) ARPHardwareType() header.ARPHardwareType { return e.child.ARPHardwareType() } + +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *Endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + e.child.AddHeader(local, remote, protocol, pkt) +} diff --git a/pkg/tcpip/link/nested/nested_test.go b/pkg/tcpip/link/nested/nested_test.go index c1a219f02..7d9249c1c 100644 --- a/pkg/tcpip/link/nested/nested_test.go +++ b/pkg/tcpip/link/nested/nested_test.go @@ -55,6 +55,10 @@ func (d *counterDispatcher) DeliverNetworkPacket(tcpip.LinkAddress, tcpip.LinkAd d.count++ } +func (d *counterDispatcher) DeliverOutboundPacket(tcpip.LinkAddress, tcpip.LinkAddress, tcpip.NetworkProtocolNumber, *stack.PacketBuffer) { + panic("unimplemented") +} + func TestNestedLinkEndpoint(t *testing.T) { const emptyAddress = tcpip.LinkAddress("") diff --git a/pkg/tcpip/link/packetsocket/BUILD b/pkg/tcpip/link/packetsocket/BUILD new file mode 100644 index 000000000..6fff160ce --- /dev/null +++ b/pkg/tcpip/link/packetsocket/BUILD @@ -0,0 +1,14 @@ +load("//tools:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_library( + name = "packetsocket", + srcs = ["endpoint.go"], + visibility = ["//visibility:public"], + deps = [ + "//pkg/tcpip", + "//pkg/tcpip/link/nested", + "//pkg/tcpip/stack", + ], +) diff --git a/pkg/tcpip/link/packetsocket/endpoint.go b/pkg/tcpip/link/packetsocket/endpoint.go new file mode 100644 index 000000000..3922c2a04 --- /dev/null +++ b/pkg/tcpip/link/packetsocket/endpoint.go @@ -0,0 +1,50 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package packetsocket provides a link layer endpoint that provides the ability +// to loop outbound packets to any AF_PACKET sockets that may be interested in +// the outgoing packet. +package packetsocket + +import ( + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/link/nested" + "gvisor.dev/gvisor/pkg/tcpip/stack" +) + +type endpoint struct { + nested.Endpoint +} + +// New creates a new packetsocket LinkEndpoint. +func New(lower stack.LinkEndpoint) stack.LinkEndpoint { + e := &endpoint{} + e.Endpoint.Init(lower, e) + return e +} + +// WritePacket implements stack.LinkEndpoint.WritePacket. +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { + e.Endpoint.DeliverOutboundPacket(r.RemoteLinkAddress, r.LocalLinkAddress, protocol, pkt) + return e.Endpoint.WritePacket(r, gso, protocol, pkt) +} + +// WritePackets implements stack.LinkEndpoint.WritePackets. +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.PacketBufferList, proto tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() { + e.Endpoint.DeliverOutboundPacket(pkt.EgressRoute.RemoteLinkAddress, pkt.EgressRoute.LocalLinkAddress, pkt.NetworkProtocolNumber, pkt) + } + + return e.Endpoint.WritePackets(r, gso, pkts, proto) +} diff --git a/pkg/tcpip/link/qdisc/fifo/endpoint.go b/pkg/tcpip/link/qdisc/fifo/endpoint.go index c84fe1bb9..467083239 100644 --- a/pkg/tcpip/link/qdisc/fifo/endpoint.go +++ b/pkg/tcpip/link/qdisc/fifo/endpoint.go @@ -107,6 +107,11 @@ func (e *endpoint) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protoco e.dispatcher.DeliverNetworkPacket(remote, local, protocol, pkt) } +// DeliverOutboundPacket implements stack.NetworkDispatcher.DeliverOutboundPacket. +func (e *endpoint) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + e.dispatcher.DeliverOutboundPacket(remote, local, protocol, pkt) +} + // Attach implements stack.LinkEndpoint.Attach. func (e *endpoint) Attach(dispatcher stack.NetworkDispatcher) { e.dispatcher = dispatcher @@ -194,6 +199,8 @@ func (e *endpoint) WritePackets(_ *stack.Route, _ *stack.GSO, pkts stack.PacketB // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { + // TODO(gvisor.dev/issue/3267/): Queue these packets as well once + // WriteRawPacket takes PacketBuffer instead of VectorisedView. return e.lower.WriteRawPacket(vv) } @@ -213,3 +220,8 @@ func (e *endpoint) Wait() { func (e *endpoint) ARPHardwareType() header.ARPHardwareType { return e.lower.ARPHardwareType() } + +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + e.lower.AddHeader(local, remote, protocol, pkt) +} diff --git a/pkg/tcpip/link/sharedmem/sharedmem.go b/pkg/tcpip/link/sharedmem/sharedmem.go index a36862c67..507c76b76 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem.go +++ b/pkg/tcpip/link/sharedmem/sharedmem.go @@ -183,22 +183,29 @@ func (e *endpoint) LinkAddress() tcpip.LinkAddress { return e.addr } -// WritePacket writes outbound packets to the file descriptor. If it is not -// currently writable, the packet is dropped. -func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { - // Add the ethernet header here. +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + // Add ethernet header if needed. eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize)) pkt.LinkHeader = buffer.View(eth) ethHdr := &header.EthernetFields{ - DstAddr: r.RemoteLinkAddress, + DstAddr: remote, Type: protocol, } - if r.LocalLinkAddress != "" { - ethHdr.SrcAddr = r.LocalLinkAddress + + // Preserve the src address if it's set in the route. + if local != "" { + ethHdr.SrcAddr = local } else { ethHdr.SrcAddr = e.addr } eth.Encode(ethHdr) +} + +// WritePacket writes outbound packets to the file descriptor. If it is not +// currently writable, the packet is dropped. +func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { + e.AddHeader(r.LocalLinkAddress, r.RemoteLinkAddress, protocol, pkt) v := pkt.Data.ToView() // Transmit the packet. diff --git a/pkg/tcpip/link/sharedmem/sharedmem_test.go b/pkg/tcpip/link/sharedmem/sharedmem_test.go index 28a2e88ba..8f3cd9449 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem_test.go +++ b/pkg/tcpip/link/sharedmem/sharedmem_test.go @@ -143,6 +143,10 @@ func (c *testContext) DeliverNetworkPacket(remoteLinkAddr, localLinkAddr tcpip.L c.packetCh <- struct{}{} } +func (c *testContext) DeliverOutboundPacket(remoteLinkAddr, localLinkAddr tcpip.LinkAddress, proto tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + panic("unimplemented") +} + func (c *testContext) cleanup() { c.ep.Close() closeFDs(&c.txCfg) diff --git a/pkg/tcpip/link/sniffer/sniffer.go b/pkg/tcpip/link/sniffer/sniffer.go index d9cd4e83a..509076643 100644 --- a/pkg/tcpip/link/sniffer/sniffer.go +++ b/pkg/tcpip/link/sniffer/sniffer.go @@ -123,6 +123,11 @@ func (e *endpoint) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protoco e.Endpoint.DeliverNetworkPacket(remote, local, protocol, pkt) } +// DeliverOutboundPacket implements stack.NetworkDispatcher.DeliverOutboundPacket. +func (e *endpoint) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + e.Endpoint.DeliverOutboundPacket(remote, local, protocol, pkt) +} + func (e *endpoint) dumpPacket(prefix string, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { writer := e.writer if writer == nil && atomic.LoadUint32(&LogPackets) == 1 { diff --git a/pkg/tcpip/link/tun/device.go b/pkg/tcpip/link/tun/device.go index 47446efec..04ae58e59 100644 --- a/pkg/tcpip/link/tun/device.go +++ b/pkg/tcpip/link/tun/device.go @@ -272,21 +272,9 @@ func (d *Device) encodePkt(info *channel.PacketInfo) (buffer.View, bool) { if d.hasFlags(linux.IFF_TAP) { // Add ethernet header if not provided. if info.Pkt.LinkHeader == nil { - hdr := &header.EthernetFields{ - SrcAddr: info.Route.LocalLinkAddress, - DstAddr: info.Route.RemoteLinkAddress, - Type: info.Proto, - } - if hdr.SrcAddr == "" { - hdr.SrcAddr = d.endpoint.LinkAddress() - } - - eth := make(header.Ethernet, header.EthernetMinimumSize) - eth.Encode(hdr) - vv.AppendView(buffer.View(eth)) - } else { - vv.AppendView(info.Pkt.LinkHeader) + d.endpoint.AddHeader(info.Route.LocalLinkAddress, info.Route.RemoteLinkAddress, info.Proto, info.Pkt) } + vv.AppendView(info.Pkt.LinkHeader) } // Append upper headers. @@ -366,3 +354,30 @@ func (e *tunEndpoint) ARPHardwareType() header.ARPHardwareType { } return header.ARPHardwareNone } + +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *tunEndpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + if !e.isTap { + return + } + eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize)) + pkt.LinkHeader = buffer.View(eth) + hdr := &header.EthernetFields{ + SrcAddr: local, + DstAddr: remote, + Type: protocol, + } + if hdr.SrcAddr == "" { + hdr.SrcAddr = e.LinkAddress() + } + + eth.Encode(hdr) +} + +// MaxHeaderLength returns the maximum size of the link layer header. +func (e *tunEndpoint) MaxHeaderLength() uint16 { + if e.isTap { + return header.EthernetMinimumSize + } + return 0 +} diff --git a/pkg/tcpip/link/waitable/waitable.go b/pkg/tcpip/link/waitable/waitable.go index 24a8dc2eb..b152a0f26 100644 --- a/pkg/tcpip/link/waitable/waitable.go +++ b/pkg/tcpip/link/waitable/waitable.go @@ -60,6 +60,15 @@ func (e *Endpoint) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protoco e.dispatchGate.Leave() } +// DeliverOutboundPacket implements stack.NetworkDispatcher.DeliverOutboundPacket. +func (e *Endpoint) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + if !e.dispatchGate.Enter() { + return + } + e.dispatcher.DeliverOutboundPacket(remote, local, protocol, pkt) + e.dispatchGate.Leave() +} + // Attach implements stack.LinkEndpoint.Attach. It saves the dispatcher and // registers with the lower endpoint as its dispatcher so that "e" is called // for inbound packets. @@ -153,3 +162,8 @@ func (e *Endpoint) Wait() {} func (e *Endpoint) ARPHardwareType() header.ARPHardwareType { return e.lower.ARPHardwareType() } + +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *Endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + e.lower.AddHeader(local, remote, protocol, pkt) +} diff --git a/pkg/tcpip/link/waitable/waitable_test.go b/pkg/tcpip/link/waitable/waitable_test.go index ffb2354be..c448a888f 100644 --- a/pkg/tcpip/link/waitable/waitable_test.go +++ b/pkg/tcpip/link/waitable/waitable_test.go @@ -40,6 +40,10 @@ func (e *countedEndpoint) DeliverNetworkPacket(remote, local tcpip.LinkAddress, e.dispatchCount++ } +func (e *countedEndpoint) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + panic("unimplemented") +} + func (e *countedEndpoint) Attach(dispatcher stack.NetworkDispatcher) { e.attachCount++ e.dispatcher = dispatcher @@ -90,6 +94,11 @@ func (*countedEndpoint) ARPHardwareType() header.ARPHardwareType { // Wait implements stack.LinkEndpoint.Wait. func (*countedEndpoint) Wait() {} +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *countedEndpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + panic("unimplemented") +} + func TestWaitWrite(t *testing.T) { ep := &countedEndpoint{} wep := New(ep) diff --git a/pkg/tcpip/network/ip_test.go b/pkg/tcpip/network/ip_test.go index a5b780ca2..615bae648 100644 --- a/pkg/tcpip/network/ip_test.go +++ b/pkg/tcpip/network/ip_test.go @@ -185,6 +185,11 @@ func (*testObject) ARPHardwareType() header.ARPHardwareType { panic("not implemented") } +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (*testObject) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { + panic("not implemented") +} + func buildIPv4Route(local, remote tcpip.Address) (stack.Route, *tcpip.Error) { s := stack.New(stack.Options{ NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, diff --git a/pkg/tcpip/stack/forwarder_test.go b/pkg/tcpip/stack/forwarder_test.go index eefb4b07f..bca1d940b 100644 --- a/pkg/tcpip/stack/forwarder_test.go +++ b/pkg/tcpip/stack/forwarder_test.go @@ -307,6 +307,11 @@ func (*fwdTestLinkEndpoint) ARPHardwareType() header.ARPHardwareType { panic("not implemented") } +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *fwdTestLinkEndpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { + panic("not implemented") +} + func fwdTestNetFactory(t *testing.T, proto *fwdTestNetworkProtocol) (ep1, ep2 *fwdTestLinkEndpoint) { // Create a stack with the network protocol and two NICs. s := New(Options{ diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 7b80534e6..fea0ce7e8 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -1200,15 +1200,13 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp // Are any packet sockets listening for this network protocol? packetEPs := n.mu.packetEPs[protocol] - // Check whether there are packet sockets listening for every protocol. - // If we received a packet with protocol EthernetProtocolAll, then the - // previous for loop will have handled it. - if protocol != header.EthernetProtocolAll { - packetEPs = append(packetEPs, n.mu.packetEPs[header.EthernetProtocolAll]...) - } + // Add any other packet sockets that maybe listening for all protocols. + packetEPs = append(packetEPs, n.mu.packetEPs[header.EthernetProtocolAll]...) n.mu.RUnlock() for _, ep := range packetEPs { - ep.HandlePacket(n.id, local, protocol, pkt.Clone()) + p := pkt.Clone() + p.PktType = tcpip.PacketHost + ep.HandlePacket(n.id, local, protocol, p) } if netProto.Number() == header.IPv4ProtocolNumber || netProto.Number() == header.IPv6ProtocolNumber { @@ -1311,6 +1309,24 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp } } +// DeliverOutboundPacket implements NetworkDispatcher.DeliverOutboundPacket. +func (n *NIC) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { + n.mu.RLock() + // We do not deliver to protocol specific packet endpoints as on Linux + // only ETH_P_ALL endpoints get outbound packets. + // Add any other packet sockets that maybe listening for all protocols. + packetEPs := n.mu.packetEPs[header.EthernetProtocolAll] + n.mu.RUnlock() + for _, ep := range packetEPs { + p := pkt.Clone() + p.PktType = tcpip.PacketOutgoing + // Add the link layer header as outgoing packets are intercepted + // before the link layer header is created. + n.linkEP.AddHeader(local, remote, protocol, p) + ep.HandlePacket(n.id, local, protocol, p) + } +} + func (n *NIC) forwardPacket(r *Route, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { // TODO(b/143425874) Decrease the TTL field in forwarded packets. // TODO(b/151227689): Avoid copying the packet when forwarding. We can do this diff --git a/pkg/tcpip/stack/nic_test.go b/pkg/tcpip/stack/nic_test.go index 3bc9fd831..c477e31d8 100644 --- a/pkg/tcpip/stack/nic_test.go +++ b/pkg/tcpip/stack/nic_test.go @@ -89,6 +89,11 @@ func (*testLinkEndpoint) ARPHardwareType() header.ARPHardwareType { panic("not implemented") } +// AddHeader implements stack.LinkEndpoint.AddHeader. +func (e *testLinkEndpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { + panic("not implemented") +} + var _ NetworkEndpoint = (*testIPv6Endpoint)(nil) // An IPv6 NetworkEndpoint that throws away outgoing packets. diff --git a/pkg/tcpip/stack/packet_buffer.go b/pkg/tcpip/stack/packet_buffer.go index e3556d5d2..5d6865e35 100644 --- a/pkg/tcpip/stack/packet_buffer.go +++ b/pkg/tcpip/stack/packet_buffer.go @@ -79,6 +79,10 @@ type PacketBuffer struct { // NatDone indicates if the packet has been manipulated as per NAT // iptables rule. NatDone bool + + // PktType indicates the SockAddrLink.PacketType of the packet as defined in + // https://www.man7.org/linux/man-pages/man7/packet.7.html. + PktType tcpip.PacketType } // Clone makes a copy of pk. It clones the Data field, which creates a new diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go index f260eeb7f..cd4b7a449 100644 --- a/pkg/tcpip/stack/registration.go +++ b/pkg/tcpip/stack/registration.go @@ -330,8 +330,7 @@ type NetworkProtocol interface { } // NetworkDispatcher contains the methods used by the network stack to deliver -// packets to the appropriate network endpoint after it has been handled by -// the data link layer. +// inbound/outbound packets to the appropriate network/packet(if any) endpoints. type NetworkDispatcher interface { // DeliverNetworkPacket finds the appropriate network protocol endpoint // and hands the packet over for further processing. @@ -342,6 +341,16 @@ type NetworkDispatcher interface { // // DeliverNetworkPacket takes ownership of pkt. DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) + + // DeliverOutboundPacket is called by link layer when a packet is being + // sent out. + // + // pkt.LinkHeader may or may not be set before calling + // DeliverOutboundPacket. Some packets do not have link headers (e.g. + // packets sent via loopback), and won't have the field set. + // + // DeliverOutboundPacket takes ownership of pkt. + DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) } // LinkEndpointCapabilities is the type associated with the capabilities @@ -443,6 +452,9 @@ type LinkEndpoint interface { // See: // https://github.com/torvalds/linux/blob/aa0c9086b40c17a7ad94425b3b70dd1fdd7497bf/include/uapi/linux/if_arp.h#L30 ARPHardwareType() header.ARPHardwareType + + // AddHeader adds a link layer header to pkt if required. + AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) } // InjectableLinkEndpoint is a LinkEndpoint where inbound packets are diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 48ad56d4d..ff14a3b3c 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -316,6 +316,28 @@ const ( ShutdownWrite ) +// PacketType is used to indicate the destination of the packet. +type PacketType uint8 + +const ( + // PacketHost indicates a packet addressed to the local host. + PacketHost PacketType = iota + + // PacketOtherHost indicates an outgoing packet addressed to + // another host caught by a NIC in promiscuous mode. + PacketOtherHost + + // PacketOutgoing for a packet originating from the local host + // that is looped back to a packet socket. + PacketOutgoing + + // PacketBroadcast indicates a link layer broadcast packet. + PacketBroadcast + + // PacketMulticast indicates a link layer multicast packet. + PacketMulticast +) + // FullAddress represents a full transport node address, as required by the // Connect() and Bind() methods. // @@ -555,6 +577,9 @@ type Endpoint interface { type LinkPacketInfo struct { // Protocol is the NetworkProtocolNumber for the packet. Protocol NetworkProtocolNumber + + // PktType is used to indicate the destination of the packet. + PktType PacketType } // PacketEndpoint are additional methods that are only implemented by Packet diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go index 7b2083a09..8f167391f 100644 --- a/pkg/tcpip/transport/packet/endpoint.go +++ b/pkg/tcpip/transport/packet/endpoint.go @@ -441,6 +441,7 @@ func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress, Addr: tcpip.Address(hdr.SourceAddress()), } packet.packetInfo.Protocol = netProto + packet.packetInfo.PktType = pkt.PktType } else { // Guess the would-be ethernet header. packet.senderAddr = tcpip.FullAddress{ @@ -448,30 +449,53 @@ func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress, Addr: tcpip.Address(localAddr), } packet.packetInfo.Protocol = netProto + packet.packetInfo.PktType = pkt.PktType } if ep.cooked { // Cooked packets can simply be queued. - packet.data = pkt.Data + switch pkt.PktType { + case tcpip.PacketHost: + packet.data = pkt.Data + case tcpip.PacketOutgoing: + // Strip Link Header from the Header. + pkt.Header = buffer.NewPrependableFromView(pkt.Header.View()[len(pkt.LinkHeader):]) + combinedVV := pkt.Header.View().ToVectorisedView() + combinedVV.Append(pkt.Data) + packet.data = combinedVV + default: + panic(fmt.Sprintf("unexpected PktType in pkt: %+v", pkt)) + } + } else { // Raw packets need their ethernet headers prepended before // queueing. var linkHeader buffer.View - if len(pkt.LinkHeader) == 0 { - // We weren't provided with an actual ethernet header, - // so fake one. - ethFields := header.EthernetFields{ - SrcAddr: tcpip.LinkAddress([]byte{0x00, 0x00, 0x00, 0x00, 0x00, 0x00}), - DstAddr: localAddr, - Type: netProto, + var combinedVV buffer.VectorisedView + if pkt.PktType != tcpip.PacketOutgoing { + if len(pkt.LinkHeader) == 0 { + // We weren't provided with an actual ethernet header, + // so fake one. + ethFields := header.EthernetFields{ + SrcAddr: tcpip.LinkAddress([]byte{0x00, 0x00, 0x00, 0x00, 0x00, 0x00}), + DstAddr: localAddr, + Type: netProto, + } + fakeHeader := make(header.Ethernet, header.EthernetMinimumSize) + fakeHeader.Encode(ðFields) + linkHeader = buffer.View(fakeHeader) + } else { + linkHeader = append(buffer.View(nil), pkt.LinkHeader...) } - fakeHeader := make(header.Ethernet, header.EthernetMinimumSize) - fakeHeader.Encode(ðFields) - linkHeader = buffer.View(fakeHeader) - } else { - linkHeader = append(buffer.View(nil), pkt.LinkHeader...) + combinedVV = linkHeader.ToVectorisedView() + } + if pkt.PktType == tcpip.PacketOutgoing { + // For outgoing packets the Link, Network and Transport + // headers are in the pkt.Header fields normally unless + // a Raw socket is in use in which case pkt.Header could + // be nil. + combinedVV.AppendView(pkt.Header.View()) } - combinedVV := linkHeader.ToVectorisedView() combinedVV.Append(pkt.Data) packet.data = combinedVV } diff --git a/runsc/boot/BUILD b/runsc/boot/BUILD index 55d45aaa6..9f52438c2 100644 --- a/runsc/boot/BUILD +++ b/runsc/boot/BUILD @@ -90,6 +90,7 @@ go_library( "//pkg/tcpip", "//pkg/tcpip/link/fdbased", "//pkg/tcpip/link/loopback", + "//pkg/tcpip/link/packetsocket", "//pkg/tcpip/link/qdisc/fifo", "//pkg/tcpip/link/sniffer", "//pkg/tcpip/network/arp", diff --git a/runsc/boot/network.go b/runsc/boot/network.go index 14d2f56a5..4e1fa7665 100644 --- a/runsc/boot/network.go +++ b/runsc/boot/network.go @@ -25,6 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/link/fdbased" "gvisor.dev/gvisor/pkg/tcpip/link/loopback" + "gvisor.dev/gvisor/pkg/tcpip/link/packetsocket" "gvisor.dev/gvisor/pkg/tcpip/link/qdisc/fifo" "gvisor.dev/gvisor/pkg/tcpip/link/sniffer" "gvisor.dev/gvisor/pkg/tcpip/network/arp" @@ -252,6 +253,9 @@ func (n *Network) CreateLinksAndRoutes(args *CreateLinksAndRoutesArgs, _ *struct linkEP = fifo.New(linkEP, runtime.GOMAXPROCS(0), 1000) } + // Enable support for AF_PACKET sockets to receive outgoing packets. + linkEP = packetsocket.New(linkEP) + log.Infof("Enabling interface %q with id %d on addresses %+v (%v) w/ %d channels", link.Name, nicID, link.Addresses, mac, link.NumChannels) if err := n.createNICWithAddrs(nicID, link.Name, linkEP, link.Addresses); err != nil { return err diff --git a/test/syscalls/linux/packet_socket.cc b/test/syscalls/linux/packet_socket.cc index e94ddcb77..40aa9326d 100644 --- a/test/syscalls/linux/packet_socket.cc +++ b/test/syscalls/linux/packet_socket.cc @@ -417,6 +417,122 @@ TEST_P(CookedPacketTest, BindDrop) { EXPECT_THAT(RetryEINTR(poll)(&pfd, 1, 1000), SyscallSucceedsWithValue(0)); } +// Verify that we receive outbound packets. This test requires at least one +// non loopback interface so that we can actually capture an outgoing packet. +TEST_P(CookedPacketTest, ReceiveOutbound) { + // Only ETH_P_ALL sockets can receive outbound packets on linux. + SKIP_IF(GetParam() != ETH_P_ALL); + + // Let's use a simple IP payload: a UDP datagram. + FileDescriptor udp_sock = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_DGRAM, 0)); + + struct ifaddrs* if_addr_list = nullptr; + auto cleanup = Cleanup([&if_addr_list]() { freeifaddrs(if_addr_list); }); + + ASSERT_THAT(getifaddrs(&if_addr_list), SyscallSucceeds()); + + // Get interface other than loopback. + struct ifreq ifr = {}; + for (struct ifaddrs* i = if_addr_list; i; i = i->ifa_next) { + if (strcmp(i->ifa_name, "lo") != 0) { + strncpy(ifr.ifr_name, i->ifa_name, sizeof(ifr.ifr_name)); + break; + } + } + + // Skip if no interface is available other than loopback. + if (strlen(ifr.ifr_name) == 0) { + GTEST_SKIP(); + } + + // Get interface index and name. + EXPECT_THAT(ioctl(socket_, SIOCGIFINDEX, &ifr), SyscallSucceeds()); + EXPECT_NE(ifr.ifr_ifindex, 0); + int ifindex = ifr.ifr_ifindex; + + constexpr int kMACSize = 6; + char hwaddr[kMACSize]; + // Get interface address. + ASSERT_THAT(ioctl(socket_, SIOCGIFHWADDR, &ifr), SyscallSucceeds()); + ASSERT_THAT(ifr.ifr_hwaddr.sa_family, + AnyOf(Eq(ARPHRD_NONE), Eq(ARPHRD_ETHER))); + memcpy(hwaddr, ifr.ifr_hwaddr.sa_data, kMACSize); + + // Just send it to the google dns server 8.8.8.8. It's UDP we don't care + // if it actually gets to the DNS Server we just want to see that we receive + // it on our AF_PACKET socket. + // + // NOTE: We just want to pick an IP that is non-local to avoid having to + // handle ARP as this should cause the UDP packet to be sent to the default + // gateway configured for the system under test. Otherwise the only packet we + // will see is the ARP query unless we picked an IP which will actually + // resolve. The test is a bit brittle but this was the best compromise for + // now. + struct sockaddr_in dest = {}; + ASSERT_EQ(inet_pton(AF_INET, "8.8.8.8", &dest.sin_addr.s_addr), 1); + dest.sin_family = AF_INET; + dest.sin_port = kPort; + EXPECT_THAT(sendto(udp_sock.get(), kMessage, sizeof(kMessage), 0, + reinterpret_cast(&dest), sizeof(dest)), + SyscallSucceedsWithValue(sizeof(kMessage))); + + // Wait and make sure the socket receives the data. + struct pollfd pfd = {}; + pfd.fd = socket_; + pfd.events = POLLIN; + EXPECT_THAT(RetryEINTR(poll)(&pfd, 1, 1000), SyscallSucceedsWithValue(1)); + + // Now read and check that the packet is the one we just sent. + // Read and verify the data. + constexpr size_t packet_size = + sizeof(struct iphdr) + sizeof(struct udphdr) + sizeof(kMessage); + char buf[64]; + struct sockaddr_ll src = {}; + socklen_t src_len = sizeof(src); + ASSERT_THAT(recvfrom(socket_, buf, sizeof(buf), 0, + reinterpret_cast(&src), &src_len), + SyscallSucceedsWithValue(packet_size)); + + // sockaddr_ll ends with an 8 byte physical address field, but ethernet + // addresses only use 6 bytes. Linux used to return sizeof(sockaddr_ll)-2 + // here, but since commit b2cf86e1563e33a14a1c69b3e508d15dc12f804c returns + // sizeof(sockaddr_ll). + ASSERT_THAT(src_len, AnyOf(Eq(sizeof(src)), Eq(sizeof(src) - 2))); + + // Verify the source address. + EXPECT_EQ(src.sll_family, AF_PACKET); + EXPECT_EQ(src.sll_ifindex, ifindex); + EXPECT_EQ(src.sll_halen, ETH_ALEN); + EXPECT_EQ(ntohs(src.sll_protocol), ETH_P_IP); + EXPECT_EQ(src.sll_pkttype, PACKET_OUTGOING); + // Verify the link address of the interface matches that of the non + // non loopback interface address we stored above. + for (int i = 0; i < src.sll_halen; i++) { + EXPECT_EQ(src.sll_addr[i], hwaddr[i]); + } + + // Verify the IP header. + struct iphdr ip = {}; + memcpy(&ip, buf, sizeof(ip)); + EXPECT_EQ(ip.ihl, 5); + EXPECT_EQ(ip.version, 4); + EXPECT_EQ(ip.tot_len, htons(packet_size)); + EXPECT_EQ(ip.protocol, IPPROTO_UDP); + EXPECT_EQ(ip.daddr, dest.sin_addr.s_addr); + EXPECT_NE(ip.saddr, htonl(INADDR_LOOPBACK)); + + // Verify the UDP header. + struct udphdr udp = {}; + memcpy(&udp, buf + sizeof(iphdr), sizeof(udp)); + EXPECT_EQ(udp.dest, kPort); + EXPECT_EQ(udp.len, htons(sizeof(udphdr) + sizeof(kMessage))); + + // Verify the payload. + char* payload = reinterpret_cast(buf + sizeof(iphdr) + sizeof(udphdr)); + EXPECT_EQ(strncmp(payload, kMessage, sizeof(kMessage)), 0); +} + // Bind with invalid address. TEST_P(CookedPacketTest, BindFail) { // Null address. -- cgit v1.2.3 From 39525d64cbd5901ea80651e323c8554af132f4dd Mon Sep 17 00:00:00 2001 From: Ayush Ranjan Date: Wed, 22 Jul 2020 15:40:44 -0700 Subject: Add O_APPEND support in vfs2 gofer. Helps in fixing open syscall tests: AppendConcurrentWrite and AppendOnly. We also now update the file size for seekable special files (regular files) which we were not doing earlier. Updates #2923 PiperOrigin-RevId: 322670843 --- pkg/sentry/fsimpl/gofer/regular_file.go | 41 +++++++++++++++++++----- pkg/sentry/fsimpl/gofer/special_file.go | 56 +++++++++++++++++++++++++++------ test/syscalls/linux/open.cc | 4 +-- 3 files changed, 83 insertions(+), 18 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go index f10350c97..02317a133 100644 --- a/pkg/sentry/fsimpl/gofer/regular_file.go +++ b/pkg/sentry/fsimpl/gofer/regular_file.go @@ -155,26 +155,53 @@ func (fd *regularFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts // PWrite implements vfs.FileDescriptionImpl.PWrite. func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + n, _, err := fd.pwrite(ctx, src, offset, opts) + return n, err +} + +// pwrite returns the number of bytes written, final offset, error. The final +// offset should be ignored by PWrite. +func (fd *regularFileFD) pwrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (written, finalOff int64, err error) { if offset < 0 { - return 0, syserror.EINVAL + return 0, offset, syserror.EINVAL } // Check that flags are supported. // // TODO(gvisor.dev/issue/2601): Support select pwritev2 flags. if opts.Flags&^linux.RWF_HIPRI != 0 { - return 0, syserror.EOPNOTSUPP + return 0, offset, syserror.EOPNOTSUPP + } + + d := fd.dentry() + // If the fd was opened with O_APPEND, make sure the file size is updated. + // There is a possible race here if size is modified externally after + // metadata cache is updated. + if fd.vfsfd.StatusFlags()&linux.O_APPEND != 0 && !d.cachedMetadataAuthoritative() { + if err := d.updateFromGetattr(ctx); err != nil { + return 0, offset, err + } } + d.metadataMu.Lock() + defer d.metadataMu.Unlock() + // Set offset to file size if the fd was opened with O_APPEND. + if fd.vfsfd.StatusFlags()&linux.O_APPEND != 0 { + // Holding d.metadataMu is sufficient for reading d.size. + offset = int64(d.size) + } limit, err := vfs.CheckLimit(ctx, offset, src.NumBytes()) if err != nil { - return 0, err + return 0, offset, err } src = src.TakeFirst64(limit) + n, err := fd.pwriteLocked(ctx, src, offset, opts) + return n, offset + n, err +} +// Preconditions: fd.dentry().metatdataMu must be locked. +func (fd *regularFileFD) pwriteLocked(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { d := fd.dentry() - d.metadataMu.Lock() - defer d.metadataMu.Unlock() if d.fs.opts.interop != InteropModeShared { // Compare Linux's mm/filemap.c:__generic_file_write_iter() => // file_update_time(). This is d.touchCMtime(), but without locking @@ -237,8 +264,8 @@ func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off // Write implements vfs.FileDescriptionImpl.Write. func (fd *regularFileFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { fd.mu.Lock() - n, err := fd.PWrite(ctx, src, fd.off, opts) - fd.off += n + n, off, err := fd.pwrite(ctx, src, fd.off, opts) + fd.off = off fd.mu.Unlock() return n, err } diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go index a7b53b2d2..811528982 100644 --- a/pkg/sentry/fsimpl/gofer/special_file.go +++ b/pkg/sentry/fsimpl/gofer/special_file.go @@ -16,6 +16,7 @@ package gofer import ( "sync" + "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" @@ -144,7 +145,7 @@ func (fd *specialFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offs // mmap due to lock ordering; MM locks precede dentry.dataMu. That doesn't // hold here since specialFileFD doesn't client-cache data. Just buffer the // read instead. - if d := fd.dentry(); d.fs.opts.interop != InteropModeShared { + if d := fd.dentry(); d.cachedMetadataAuthoritative() { d.touchAtime(fd.vfsfd.Mount()) } buf := make([]byte, dst.NumBytes()) @@ -176,39 +177,76 @@ func (fd *specialFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts // PWrite implements vfs.FileDescriptionImpl.PWrite. func (fd *specialFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + n, _, err := fd.pwrite(ctx, src, offset, opts) + return n, err +} + +// pwrite returns the number of bytes written, final offset, error. The final +// offset should be ignored by PWrite. +func (fd *specialFileFD) pwrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (written, finalOff int64, err error) { if fd.seekable && offset < 0 { - return 0, syserror.EINVAL + return 0, offset, syserror.EINVAL } // Check that flags are supported. // // TODO(gvisor.dev/issue/2601): Support select pwritev2 flags. if opts.Flags&^linux.RWF_HIPRI != 0 { - return 0, syserror.EOPNOTSUPP + return 0, offset, syserror.EOPNOTSUPP + } + + d := fd.dentry() + // If the regular file fd was opened with O_APPEND, make sure the file size + // is updated. There is a possible race here if size is modified externally + // after metadata cache is updated. + if fd.seekable && fd.vfsfd.StatusFlags()&linux.O_APPEND != 0 && !d.cachedMetadataAuthoritative() { + if err := d.updateFromGetattr(ctx); err != nil { + return 0, offset, err + } } if fd.seekable { + // We need to hold the metadataMu *while* writing to a regular file. + d.metadataMu.Lock() + defer d.metadataMu.Unlock() + + // Set offset to file size if the regular file was opened with O_APPEND. + if fd.vfsfd.StatusFlags()&linux.O_APPEND != 0 { + // Holding d.metadataMu is sufficient for reading d.size. + offset = int64(d.size) + } limit, err := vfs.CheckLimit(ctx, offset, src.NumBytes()) if err != nil { - return 0, err + return 0, offset, err } src = src.TakeFirst64(limit) } // Do a buffered write. See rationale in PRead. - if d := fd.dentry(); d.fs.opts.interop != InteropModeShared { + if d.cachedMetadataAuthoritative() { d.touchCMtime() } buf := make([]byte, src.NumBytes()) // Don't do partial writes if we get a partial read from src. if _, err := src.CopyIn(ctx, buf); err != nil { - return 0, err + return 0, offset, err } n, err := fd.handle.writeFromBlocksAt(ctx, safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf)), uint64(offset)) if err == syserror.EAGAIN { err = syserror.ErrWouldBlock } - return int64(n), err + finalOff = offset + // Update file size for regular files. + if fd.seekable { + finalOff += int64(n) + // d.metadataMu is already locked at this point. + if uint64(finalOff) > d.size { + d.dataMu.Lock() + defer d.dataMu.Unlock() + atomic.StoreUint64(&d.size, uint64(finalOff)) + } + } + return int64(n), finalOff, err } // Write implements vfs.FileDescriptionImpl.Write. @@ -218,8 +256,8 @@ func (fd *specialFileFD) Write(ctx context.Context, src usermem.IOSequence, opts } fd.mu.Lock() - n, err := fd.PWrite(ctx, src, fd.off, opts) - fd.off += n + n, off, err := fd.pwrite(ctx, src, fd.off, opts) + fd.off = off fd.mu.Unlock() return n, err } diff --git a/test/syscalls/linux/open.cc b/test/syscalls/linux/open.cc index bb7d108e8..bf350946b 100644 --- a/test/syscalls/linux/open.cc +++ b/test/syscalls/linux/open.cc @@ -235,7 +235,7 @@ TEST_F(OpenTest, AppendOnly) { ASSERT_NO_ERRNO_AND_VALUE(Open(test_file_name_, O_RDWR | O_APPEND)); EXPECT_THAT(lseek(fd2.get(), 0, SEEK_CUR), SyscallSucceedsWithValue(0)); - // Then try to write to the first file and make sure the bytes are appended. + // Then try to write to the first fd and make sure the bytes are appended. EXPECT_THAT(WriteFd(fd1.get(), buf.data(), buf.size()), SyscallSucceedsWithValue(buf.size())); @@ -247,7 +247,7 @@ TEST_F(OpenTest, AppendOnly) { EXPECT_THAT(lseek(fd1.get(), 0, SEEK_CUR), SyscallSucceedsWithValue(kBufSize * 2)); - // Then try to write to the second file and make sure the bytes are appended. + // Then try to write to the second fd and make sure the bytes are appended. EXPECT_THAT(WriteFd(fd2.get(), buf.data(), buf.size()), SyscallSucceedsWithValue(buf.size())); -- cgit v1.2.3 From 9654bf04acad7300e0bf964183f7860bd553f563 Mon Sep 17 00:00:00 2001 From: Ayush Ranjan Date: Wed, 22 Jul 2020 15:44:08 -0700 Subject: [vfs2][tmpfs] Implement O_APPEND Updates #2923 PiperOrigin-RevId: 322671489 --- pkg/sentry/fsimpl/tmpfs/regular_file.go | 35 ++++++++++++++++++++++----------- 1 file changed, 23 insertions(+), 12 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index 1cdb46e6f..abbaa5d60 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -325,8 +325,15 @@ func (fd *regularFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts // PWrite implements vfs.FileDescriptionImpl.PWrite. func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + n, _, err := fd.pwrite(ctx, src, offset, opts) + return n, err +} + +// pwrite returns the number of bytes written, final offset and error. The +// final offset should be ignored by PWrite. +func (fd *regularFileFD) pwrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (written, finalOff int64, err error) { if offset < 0 { - return 0, syserror.EINVAL + return 0, offset, syserror.EINVAL } // Check that flags are supported. RWF_DSYNC/RWF_SYNC can be ignored since @@ -334,40 +341,44 @@ func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off // // TODO(gvisor.dev/issue/2601): Support select preadv2 flags. if opts.Flags&^(linux.RWF_HIPRI|linux.RWF_DSYNC|linux.RWF_SYNC) != 0 { - return 0, syserror.EOPNOTSUPP + return 0, offset, syserror.EOPNOTSUPP } srclen := src.NumBytes() if srclen == 0 { - return 0, nil + return 0, offset, nil } f := fd.inode().impl.(*regularFile) + f.inode.mu.Lock() + defer f.inode.mu.Unlock() + // If the file is opened with O_APPEND, update offset to file size. + if fd.vfsfd.StatusFlags()&linux.O_APPEND != 0 { + // Locking f.inode.mu is sufficient for reading f.size. + offset = int64(f.size) + } if end := offset + srclen; end < offset { // Overflow. - return 0, syserror.EINVAL + return 0, offset, syserror.EINVAL } - var err error srclen, err = vfs.CheckLimit(ctx, offset, srclen) if err != nil { - return 0, err + return 0, offset, err } src = src.TakeFirst64(srclen) - f.inode.mu.Lock() rw := getRegularFileReadWriter(f, offset) n, err := src.CopyInTo(ctx, rw) - fd.inode().touchCMtimeLocked() - f.inode.mu.Unlock() + f.inode.touchCMtimeLocked() putRegularFileReadWriter(rw) - return n, err + return n, n + offset, err } // Write implements vfs.FileDescriptionImpl.Write. func (fd *regularFileFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { fd.offMu.Lock() - n, err := fd.PWrite(ctx, src, fd.off, opts) - fd.off += n + n, off, err := fd.pwrite(ctx, src, fd.off, opts) + fd.off = off fd.offMu.Unlock() return n, err } -- cgit v1.2.3 From bd98f820141208d9f19b0e12dee93f6f6de3ac97 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Wed, 22 Jul 2020 16:22:06 -0700 Subject: iptables: replace maps with arrays For iptables users, Check() is a hot path called for every packet one or more times. Let's avoid a bunch of map lookups. PiperOrigin-RevId: 322678699 --- pkg/sentry/socket/netfilter/netfilter.go | 21 ++-- pkg/tcpip/stack/iptables.go | 187 ++++++++++++++++++------------- pkg/tcpip/stack/iptables_types.go | 22 ++-- 3 files changed, 127 insertions(+), 103 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index f7abe77d3..1243143ea 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -342,10 +342,10 @@ func SetEntries(stk *stack.Stack, optVal []byte) *syserr.Error { // TODO(gvisor.dev/issue/170): Support other tables. var table stack.Table switch replace.Name.String() { - case stack.TablenameFilter: + case stack.FilterTable: table = stack.EmptyFilterTable() - case stack.TablenameNat: - table = stack.EmptyNatTable() + case stack.NATTable: + table = stack.EmptyNATTable() default: nflog("we don't yet support writing to the %q table (gvisor.dev/issue/170)", replace.Name.String()) return syserr.ErrInvalidArgument @@ -431,6 +431,8 @@ func SetEntries(stk *stack.Stack, optVal []byte) *syserr.Error { for hook, _ := range replace.HookEntry { if table.ValidHooks()&(1< Date: Thu, 23 Jul 2020 09:43:05 -0700 Subject: [vfs2][gofer] Fix update attributes race condition. We were getting the file attributes before locking the metadataMu which was causing stale updates to the file attributes. Fixes OpenTest_AppendConcurrentWrite. Updates #2923 PiperOrigin-RevId: 322804438 --- pkg/sentry/fsimpl/gofer/filesystem.go | 17 +++++++++++++---- pkg/sentry/fsimpl/gofer/gofer.go | 11 +++++++---- 2 files changed, 20 insertions(+), 8 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index cd5f5049e..491495b16 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -150,11 +150,9 @@ afterSymlink: return nil, err } if d != d.parent && !d.cachedMetadataAuthoritative() { - _, attrMask, attr, err := d.parent.file.getAttr(ctx, dentryAttrMask()) - if err != nil { + if err := d.parent.updateFromGetattr(ctx); err != nil { return nil, err } - d.parent.updateFromP9Attrs(attrMask, &attr) } rp.Advance() return d.parent, nil @@ -209,17 +207,28 @@ func (fs *filesystem) getChildLocked(ctx context.Context, vfsObj *vfs.VirtualFil // Preconditions: As for getChildLocked. !parent.isSynthetic(). func (fs *filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.VirtualFilesystem, parent *dentry, name string, child *dentry, ds **[]*dentry) (*dentry, error) { + if child != nil { + // Need to lock child.metadataMu because we might be updating child + // metadata. We need to hold the lock *before* getting metadata from the + // server and release it after updating local metadata. + child.metadataMu.Lock() + } qid, file, attrMask, attr, err := parent.file.walkGetAttrOne(ctx, name) if err != nil && err != syserror.ENOENT { + if child != nil { + child.metadataMu.Unlock() + } return nil, err } if child != nil { if !file.isNil() && inoFromPath(qid.Path) == child.ino { // The file at this path hasn't changed. Just update cached metadata. file.close(ctx) - child.updateFromP9Attrs(attrMask, &attr) + child.updateFromP9AttrsLocked(attrMask, &attr) + child.metadataMu.Unlock() return child, nil } + child.metadataMu.Unlock() if file.isNil() && child.isSynthetic() { // We have a synthetic file, and no remote file has arisen to // replace it. diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index b74d489a0..ba7f650b3 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -785,8 +785,8 @@ func (d *dentry) cachedMetadataAuthoritative() bool { // updateFromP9Attrs is called to update d's metadata after an update from the // remote filesystem. -func (d *dentry) updateFromP9Attrs(mask p9.AttrMask, attr *p9.Attr) { - d.metadataMu.Lock() +// Precondition: d.metadataMu must be locked. +func (d *dentry) updateFromP9AttrsLocked(mask p9.AttrMask, attr *p9.Attr) { if mask.Mode { if got, want := uint32(attr.Mode.FileType()), d.fileType(); got != want { d.metadataMu.Unlock() @@ -822,7 +822,6 @@ func (d *dentry) updateFromP9Attrs(mask p9.AttrMask, attr *p9.Attr) { if mask.Size { d.updateFileSizeLocked(attr.Size) } - d.metadataMu.Unlock() } // Preconditions: !d.isSynthetic() @@ -834,6 +833,10 @@ func (d *dentry) updateFromGetattr(ctx context.Context) error { file p9file handleMuRLocked bool ) + // d.metadataMu must be locked *before* we getAttr so that we do not end up + // updating stale attributes in d.updateFromP9AttrsLocked(). + d.metadataMu.Lock() + defer d.metadataMu.Unlock() d.handleMu.RLock() if !d.handle.file.isNil() { file = d.handle.file @@ -849,7 +852,7 @@ func (d *dentry) updateFromGetattr(ctx context.Context) error { if err != nil { return err } - d.updateFromP9Attrs(attrMask, &attr) + d.updateFromP9AttrsLocked(attrMask, &attr) return nil } -- cgit v1.2.3 From b396d3882ccccc82e1a3c799c7a291d47c43c2c8 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Thu, 23 Jul 2020 10:34:46 -0700 Subject: Port sendfile to vfs2. And do some refactoring of the wait logic in sendfile/splice/tee. Updates #1035 #2923 PiperOrigin-RevId: 322815521 --- pkg/sentry/syscalls/linux/vfs2/splice.go | 321 +++++++++++++++++++++++++------ pkg/sentry/syscalls/linux/vfs2/vfs2.go | 2 +- test/syscalls/BUILD | 2 + 3 files changed, 261 insertions(+), 64 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/vfs2/splice.go b/pkg/sentry/syscalls/linux/vfs2/splice.go index 945a364a7..63ab11f8c 100644 --- a/pkg/sentry/syscalls/linux/vfs2/splice.go +++ b/pkg/sentry/syscalls/linux/vfs2/splice.go @@ -15,12 +15,15 @@ package vfs2 import ( + "io" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) @@ -110,16 +113,20 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal // Move data. var ( - n int64 - err error - inCh chan struct{} - outCh chan struct{} + n int64 + err error ) + dw := dualWaiter{ + inFile: inFile, + outFile: outFile, + } + defer dw.destroy() for { // If both input and output are pipes, delegate to the pipe - // implementation. Otherwise, exactly one end is a pipe, which we - // ensure is consistently ordered after the non-pipe FD's locks by - // passing the pipe FD as usermem.IO to the non-pipe end. + // implementation. Otherwise, exactly one end is a pipe, which + // we ensure is consistently ordered after the non-pipe FD's + // locks by passing the pipe FD as usermem.IO to the non-pipe + // end. switch { case inIsPipe && outIsPipe: n, err = pipe.Splice(t, outPipeFD, inPipeFD, count) @@ -137,38 +144,15 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal } else { n, err = inFile.Read(t, outPipeFD.IOSequence(count), vfs.ReadOptions{}) } + default: + panic("not possible") } + if n != 0 || err != syserror.ErrWouldBlock || nonBlock { break } - - // Note that the blocking behavior here is a bit different than the - // normal pattern. Because we need to have both data to read and data - // to write simultaneously, we actually explicitly block on both of - // these cases in turn before returning to the splice operation. - if inFile.Readiness(eventMaskRead)&eventMaskRead == 0 { - if inCh == nil { - inCh = make(chan struct{}, 1) - inW, _ := waiter.NewChannelEntry(inCh) - inFile.EventRegister(&inW, eventMaskRead) - defer inFile.EventUnregister(&inW) - continue // Need to refresh readiness. - } - if err = t.Block(inCh); err != nil { - break - } - } - if outFile.Readiness(eventMaskWrite)&eventMaskWrite == 0 { - if outCh == nil { - outCh = make(chan struct{}, 1) - outW, _ := waiter.NewChannelEntry(outCh) - outFile.EventRegister(&outW, eventMaskWrite) - defer outFile.EventUnregister(&outW) - continue // Need to refresh readiness. - } - if err = t.Block(outCh); err != nil { - break - } + if err = dw.waitForBoth(t); err != nil { + break } } @@ -247,45 +231,256 @@ func Tee(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo // Copy data. var ( - inCh chan struct{} - outCh chan struct{} + n int64 + err error ) + dw := dualWaiter{ + inFile: inFile, + outFile: outFile, + } + defer dw.destroy() for { - n, err := pipe.Tee(t, outPipeFD, inPipeFD, count) - if n != 0 { - return uintptr(n), nil, nil + n, err = pipe.Tee(t, outPipeFD, inPipeFD, count) + if n != 0 || err != syserror.ErrWouldBlock || nonBlock { + break + } + if err = dw.waitForBoth(t); err != nil { + break + } + } + if n == 0 { + return 0, nil, err + } + outFile.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) + return uintptr(n), nil, nil +} + +// Sendfile implements linux system call sendfile(2). +func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + outFD := args[0].Int() + inFD := args[1].Int() + offsetAddr := args[2].Pointer() + count := int64(args[3].SizeT()) + + inFile := t.GetFileVFS2(inFD) + if inFile == nil { + return 0, nil, syserror.EBADF + } + defer inFile.DecRef() + if !inFile.IsReadable() { + return 0, nil, syserror.EBADF + } + + outFile := t.GetFileVFS2(outFD) + if outFile == nil { + return 0, nil, syserror.EBADF + } + defer outFile.DecRef() + if !outFile.IsWritable() { + return 0, nil, syserror.EBADF + } + + // Verify that the outFile Append flag is not set. + if outFile.StatusFlags()&linux.O_APPEND != 0 { + return 0, nil, syserror.EINVAL + } + + // Verify that inFile is a regular file or block device. This is a + // requirement; the same check appears in Linux + // (fs/splice.c:splice_direct_to_actor). + if stat, err := inFile.Stat(t, vfs.StatOptions{Mask: linux.STATX_TYPE}); err != nil { + return 0, nil, err + } else if stat.Mask&linux.STATX_TYPE == 0 || + (stat.Mode&linux.S_IFMT != linux.S_IFREG && stat.Mode&linux.S_IFMT != linux.S_IFBLK) { + return 0, nil, syserror.EINVAL + } + + // Copy offset if it exists. + offset := int64(-1) + if offsetAddr != 0 { + if inFile.Options().DenyPRead { + return 0, nil, syserror.ESPIPE } - if err != syserror.ErrWouldBlock || nonBlock { + if _, err := t.CopyIn(offsetAddr, &offset); err != nil { return 0, nil, err } + if offset < 0 { + return 0, nil, syserror.EINVAL + } + if offset+count < 0 { + return 0, nil, syserror.EINVAL + } + } + + // Validate count. This must come after offset checks. + if count < 0 { + return 0, nil, syserror.EINVAL + } + if count == 0 { + return 0, nil, nil + } + if count > int64(kernel.MAX_RW_COUNT) { + count = int64(kernel.MAX_RW_COUNT) + } - // Note that the blocking behavior here is a bit different than the - // normal pattern. Because we need to have both data to read and data - // to write simultaneously, we actually explicitly block on both of - // these cases in turn before returning to the tee operation. - if inFile.Readiness(eventMaskRead)&eventMaskRead == 0 { - if inCh == nil { - inCh = make(chan struct{}, 1) - inW, _ := waiter.NewChannelEntry(inCh) - inFile.EventRegister(&inW, eventMaskRead) - defer inFile.EventUnregister(&inW) - continue // Need to refresh readiness. + // Copy data. + var ( + n int64 + err error + ) + dw := dualWaiter{ + inFile: inFile, + outFile: outFile, + } + defer dw.destroy() + outPipeFD, outIsPipe := outFile.Impl().(*pipe.VFSPipeFD) + // Reading from input file should never block, since it is regular or + // block device. We only need to check if writing to the output file + // can block. + nonBlock := outFile.StatusFlags()&linux.O_NONBLOCK != 0 + if outIsPipe { + for n < count { + var spliceN int64 + if offset != -1 { + spliceN, err = inFile.PRead(t, outPipeFD.IOSequence(count), offset, vfs.ReadOptions{}) + offset += spliceN + } else { + spliceN, err = inFile.Read(t, outPipeFD.IOSequence(count), vfs.ReadOptions{}) } - if err := t.Block(inCh); err != nil { - return 0, nil, err + n += spliceN + if err == syserror.ErrWouldBlock && !nonBlock { + err = dw.waitForBoth(t) + } + if err != nil { + break } } - if outFile.Readiness(eventMaskWrite)&eventMaskWrite == 0 { - if outCh == nil { - outCh = make(chan struct{}, 1) - outW, _ := waiter.NewChannelEntry(outCh) - outFile.EventRegister(&outW, eventMaskWrite) - defer outFile.EventUnregister(&outW) - continue // Need to refresh readiness. + } else { + // Read inFile to buffer, then write the contents to outFile. + buf := make([]byte, count) + for n < count { + var readN int64 + if offset != -1 { + readN, err = inFile.PRead(t, usermem.BytesIOSequence(buf), offset, vfs.ReadOptions{}) + offset += readN + } else { + readN, err = inFile.Read(t, usermem.BytesIOSequence(buf), vfs.ReadOptions{}) + } + if readN == 0 && err == io.EOF { + // We reached the end of the file. Eat the + // error and exit the loop. + err = nil + break } - if err := t.Block(outCh); err != nil { - return 0, nil, err + n += readN + if err != nil { + break + } + + // Write all of the bytes that we read. This may need + // multiple write calls to complete. + wbuf := buf[:n] + for len(wbuf) > 0 { + var writeN int64 + writeN, err = outFile.Write(t, usermem.BytesIOSequence(wbuf), vfs.WriteOptions{}) + wbuf = wbuf[writeN:] + if err == syserror.ErrWouldBlock && !nonBlock { + err = dw.waitForOut(t) + } + if err != nil { + // We didn't complete the write. Only + // report the bytes that were actually + // written, and rewind the offset. + notWritten := int64(len(wbuf)) + n -= notWritten + if offset != -1 { + offset -= notWritten + } + break + } + } + if err == syserror.ErrWouldBlock && !nonBlock { + err = dw.waitForBoth(t) } + if err != nil { + break + } + } + } + + if offsetAddr != 0 { + // Copy out the new offset. + if _, err := t.CopyOut(offsetAddr, offset); err != nil { + return 0, nil, err + } + } + + if n == 0 { + return 0, nil, err + } + + inFile.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + outFile.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) + return uintptr(n), nil, nil +} + +// dualWaiter is used to wait on one or both vfs.FileDescriptions. It is not +// thread-safe, and does not take a reference on the vfs.FileDescriptions. +// +// Users must call destroy() when finished. +type dualWaiter struct { + inFile *vfs.FileDescription + outFile *vfs.FileDescription + + inW waiter.Entry + inCh chan struct{} + outW waiter.Entry + outCh chan struct{} +} + +// waitForBoth waits for both dw.inFile and dw.outFile to be ready. +func (dw *dualWaiter) waitForBoth(t *kernel.Task) error { + if dw.inFile.Readiness(eventMaskRead)&eventMaskRead == 0 { + if dw.inCh == nil { + dw.inW, dw.inCh = waiter.NewChannelEntry(nil) + dw.inFile.EventRegister(&dw.inW, eventMaskRead) + // We might be ready now. Try again before blocking. + return nil + } + if err := t.Block(dw.inCh); err != nil { + return err + } + } + return dw.waitForOut(t) +} + +// waitForOut waits for dw.outfile to be read. +func (dw *dualWaiter) waitForOut(t *kernel.Task) error { + if dw.outFile.Readiness(eventMaskWrite)&eventMaskWrite == 0 { + if dw.outCh == nil { + dw.outW, dw.outCh = waiter.NewChannelEntry(nil) + dw.outFile.EventRegister(&dw.outW, eventMaskWrite) + // We might be ready now. Try again before blocking. + return nil } + if err := t.Block(dw.outCh); err != nil { + return err + } + } + return nil +} + +// destroy cleans up resources help by dw. No more calls to wait* can occur +// after destroy is called. +func (dw *dualWaiter) destroy() { + if dw.inCh != nil { + dw.inFile.EventUnregister(&dw.inW) + dw.inCh = nil + } + if dw.outCh != nil { + dw.outFile.EventUnregister(&dw.outW) + dw.outCh = nil } + dw.inFile = nil + dw.outFile = nil } diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2.go b/pkg/sentry/syscalls/linux/vfs2/vfs2.go index 8f497ecc7..1b2cfad7d 100644 --- a/pkg/sentry/syscalls/linux/vfs2/vfs2.go +++ b/pkg/sentry/syscalls/linux/vfs2/vfs2.go @@ -44,7 +44,7 @@ func Override() { s.Table[23] = syscalls.Supported("select", Select) s.Table[32] = syscalls.Supported("dup", Dup) s.Table[33] = syscalls.Supported("dup2", Dup2) - delete(s.Table, 40) // sendfile + s.Table[40] = syscalls.Supported("sendfile", Sendfile) s.Table[41] = syscalls.Supported("socket", Socket) s.Table[42] = syscalls.Supported("connect", Connect) s.Table[43] = syscalls.Supported("accept", Accept) diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index c06a75ada..15e5fd223 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -640,11 +640,13 @@ syscall_test( syscall_test( add_overlay = True, test = "//test/syscalls/linux:sendfile_socket_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:sendfile_test", + vfs2 = "True", ) syscall_test( -- cgit v1.2.3 From 6f7f73996791bbab6b63c248000df0d3ce652f2b Mon Sep 17 00:00:00 2001 From: Ayush Ranjan Date: Thu, 23 Jul 2020 11:43:40 -0700 Subject: Marshallable socket opitons. Socket option values are now required to implement marshal.Marshallable. Co-authored-by: Rahat Mahmood PiperOrigin-RevId: 322831612 --- pkg/abi/linux/BUILD | 3 + pkg/abi/linux/netfilter.go | 146 ++++++++++++++++++-- pkg/abi/linux/socket.go | 8 ++ pkg/sentry/socket/BUILD | 1 + pkg/sentry/socket/hostinet/BUILD | 2 + pkg/sentry/socket/hostinet/socket.go | 7 +- pkg/sentry/socket/netfilter/netfilter.go | 28 ++-- pkg/sentry/socket/netlink/BUILD | 2 + pkg/sentry/socket/netlink/socket.go | 14 +- pkg/sentry/socket/netstack/BUILD | 2 + pkg/sentry/socket/netstack/netstack.go | 205 ++++++++++++++++++---------- pkg/sentry/socket/netstack/netstack_vfs2.go | 16 ++- pkg/sentry/socket/socket.go | 3 +- pkg/sentry/socket/unix/BUILD | 1 + pkg/sentry/socket/unix/unix.go | 3 +- pkg/sentry/socket/unix/unix_vfs2.go | 3 +- pkg/sentry/syscalls/linux/BUILD | 2 + pkg/sentry/syscalls/linux/sys_socket.go | 17 ++- pkg/sentry/syscalls/linux/vfs2/BUILD | 2 + pkg/sentry/syscalls/linux/vfs2/socket.go | 17 ++- tools/go_marshal/README.md | 8 +- tools/go_marshal/primitive/primitive.go | 72 ++++++++++ 22 files changed, 427 insertions(+), 135 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/BUILD b/pkg/abi/linux/BUILD index 2b789c4ec..a4bb62013 100644 --- a/pkg/abi/linux/BUILD +++ b/pkg/abi/linux/BUILD @@ -72,6 +72,9 @@ go_library( "//pkg/abi", "//pkg/binary", "//pkg/bits", + "//pkg/usermem", + "//tools/go_marshal/marshal", + "//tools/go_marshal/primitive", ], ) diff --git a/pkg/abi/linux/netfilter.go b/pkg/abi/linux/netfilter.go index 46d8b0b42..a91f9f018 100644 --- a/pkg/abi/linux/netfilter.go +++ b/pkg/abi/linux/netfilter.go @@ -14,6 +14,14 @@ package linux +import ( + "io" + + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/tools/go_marshal/marshal" + "gvisor.dev/gvisor/tools/go_marshal/primitive" +) + // This file contains structures required to support netfilter, specifically // the iptables tool. @@ -76,6 +84,8 @@ const ( // IPTEntry is an iptable rule. It corresponds to struct ipt_entry in // include/uapi/linux/netfilter_ipv4/ip_tables.h. +// +// +marshal type IPTEntry struct { // IP is used to filter packets based on the IP header. IP IPTIP @@ -112,21 +122,41 @@ type IPTEntry struct { // SizeOfIPTEntry is the size of an IPTEntry. const SizeOfIPTEntry = 112 -// KernelIPTEntry is identical to IPTEntry, but includes the Elems field. This -// struct marshaled via the binary package to write an IPTEntry to userspace. +// KernelIPTEntry is identical to IPTEntry, but includes the Elems field. +// KernelIPTEntry itself is not Marshallable but it implements some methods of +// marshal.Marshallable that help in other implementations of Marshallable. type KernelIPTEntry struct { - IPTEntry + Entry IPTEntry // Elems holds the data for all this rule's matches followed by the // target. It is variable length -- users have to iterate over any // matches and use TargetOffset and NextOffset to make sense of the // data. - Elems []byte + Elems primitive.ByteSlice +} + +// SizeBytes implements marshal.Marshallable.SizeBytes. +func (ke *KernelIPTEntry) SizeBytes() int { + return ke.Entry.SizeBytes() + ke.Elems.SizeBytes() +} + +// MarshalBytes implements marshal.Marshallable.MarshalBytes. +func (ke *KernelIPTEntry) MarshalBytes(dst []byte) { + ke.Entry.MarshalBytes(dst) + ke.Elems.MarshalBytes(dst[ke.Entry.SizeBytes():]) +} + +// UnmarshalBytes implements marshal.Marshallable.UnmarshalBytes. +func (ke *KernelIPTEntry) UnmarshalBytes(src []byte) { + ke.Entry.UnmarshalBytes(src) + ke.Elems.UnmarshalBytes(src[ke.Entry.SizeBytes():]) } // IPTIP contains information for matching a packet's IP header. // It corresponds to struct ipt_ip in // include/uapi/linux/netfilter_ipv4/ip_tables.h. +// +// +marshal type IPTIP struct { // Src is the source IP address. Src InetAddr @@ -189,6 +219,8 @@ const SizeOfIPTIP = 84 // XTCounters holds packet and byte counts for a rule. It corresponds to struct // xt_counters in include/uapi/linux/netfilter/x_tables.h. +// +// +marshal type XTCounters struct { // Pcnt is the packet count. Pcnt uint64 @@ -321,6 +353,8 @@ const SizeOfXTRedirectTarget = 56 // IPTGetinfo is the argument for the IPT_SO_GET_INFO sockopt. It corresponds // to struct ipt_getinfo in include/uapi/linux/netfilter_ipv4/ip_tables.h. +// +// +marshal type IPTGetinfo struct { Name TableName ValidHooks uint32 @@ -336,6 +370,8 @@ const SizeOfIPTGetinfo = 84 // IPTGetEntries is the argument for the IPT_SO_GET_ENTRIES sockopt. It // corresponds to struct ipt_get_entries in // include/uapi/linux/netfilter_ipv4/ip_tables.h. +// +// +marshal type IPTGetEntries struct { Name TableName Size uint32 @@ -350,13 +386,103 @@ type IPTGetEntries struct { const SizeOfIPTGetEntries = 40 // KernelIPTGetEntries is identical to IPTGetEntries, but includes the -// Entrytable field. This struct marshaled via the binary package to write an -// KernelIPTGetEntries to userspace. +// Entrytable field. This has been manually made marshal.Marshallable since it +// is dynamically sized. type KernelIPTGetEntries struct { IPTGetEntries Entrytable []KernelIPTEntry } +// SizeBytes implements marshal.Marshallable.SizeBytes. +func (ke *KernelIPTGetEntries) SizeBytes() int { + res := ke.IPTGetEntries.SizeBytes() + for _, entry := range ke.Entrytable { + res += entry.SizeBytes() + } + return res +} + +// MarshalBytes implements marshal.Marshallable.MarshalBytes. +func (ke *KernelIPTGetEntries) MarshalBytes(dst []byte) { + ke.IPTGetEntries.MarshalBytes(dst) + marshalledUntil := ke.IPTGetEntries.SizeBytes() + for i := 0; i < len(ke.Entrytable); i++ { + ke.Entrytable[i].MarshalBytes(dst[marshalledUntil:]) + marshalledUntil += ke.Entrytable[i].SizeBytes() + } +} + +// UnmarshalBytes implements marshal.Marshallable.UnmarshalBytes. +func (ke *KernelIPTGetEntries) UnmarshalBytes(src []byte) { + ke.IPTGetEntries.UnmarshalBytes(src) + unmarshalledUntil := ke.IPTGetEntries.SizeBytes() + for i := 0; i < len(ke.Entrytable); i++ { + ke.Entrytable[i].UnmarshalBytes(src[unmarshalledUntil:]) + unmarshalledUntil += ke.Entrytable[i].SizeBytes() + } +} + +// Packed implements marshal.Marshallable.Packed. +func (ke *KernelIPTGetEntries) Packed() bool { + // KernelIPTGetEntries isn't packed because the ke.Entrytable contains an + // indirection to the actual data we want to marshal (the slice data + // pointer), and the memory for KernelIPTGetEntries contains the slice + // header which we don't want to marshal. + return false +} + +// MarshalUnsafe implements marshal.Marshallable.MarshalUnsafe. +func (ke *KernelIPTGetEntries) MarshalUnsafe(dst []byte) { + // Fall back to safe Marshal because the type in not packed. + ke.MarshalBytes(dst) +} + +// UnmarshalUnsafe implements marshal.Marshallable.UnmarshalUnsafe. +func (ke *KernelIPTGetEntries) UnmarshalUnsafe(src []byte) { + // Fall back to safe Unmarshal because the type in not packed. + ke.UnmarshalBytes(src) +} + +// CopyIn implements marshal.Marshallable.CopyIn. +func (ke *KernelIPTGetEntries) CopyIn(task marshal.Task, addr usermem.Addr) (int, error) { + buf := task.CopyScratchBuffer(ke.SizeBytes()) // escapes: okay. + length, err := task.CopyInBytes(addr, buf) // escapes: okay. + // Unmarshal unconditionally. If we had a short copy-in, this results in a + // partially unmarshalled struct. + ke.UnmarshalBytes(buf) // escapes: fallback. + return length, err +} + +// CopyOut implements marshal.Marshallable.CopyOut. +func (ke *KernelIPTGetEntries) CopyOut(task marshal.Task, addr usermem.Addr) (int, error) { + // Type KernelIPTGetEntries doesn't have a packed layout in memory, fall + // back to MarshalBytes. + return task.CopyOutBytes(addr, ke.marshalAll(task)) +} + +// CopyOutN implements marshal.Marshallable.CopyOutN. +func (ke *KernelIPTGetEntries) CopyOutN(task marshal.Task, addr usermem.Addr, limit int) (int, error) { + // Type KernelIPTGetEntries doesn't have a packed layout in memory, fall + // back to MarshalBytes. + return task.CopyOutBytes(addr, ke.marshalAll(task)[:limit]) +} + +func (ke *KernelIPTGetEntries) marshalAll(task marshal.Task) []byte { + buf := task.CopyScratchBuffer(ke.SizeBytes()) + ke.MarshalBytes(buf) + return buf +} + +// WriteTo implements io.WriterTo.WriteTo. +func (ke *KernelIPTGetEntries) WriteTo(w io.Writer) (int64, error) { + buf := make([]byte, ke.SizeBytes()) + ke.MarshalBytes(buf) + length, err := w.Write(buf) + return int64(length), err +} + +var _ marshal.Marshallable = (*KernelIPTGetEntries)(nil) + // IPTReplace is the argument for the IPT_SO_SET_REPLACE sockopt. It // corresponds to struct ipt_replace in // include/uapi/linux/netfilter_ipv4/ip_tables.h. @@ -374,12 +500,6 @@ type IPTReplace struct { // Entries [0]IPTEntry } -// KernelIPTReplace is identical to IPTReplace, but includes the Entries field. -type KernelIPTReplace struct { - IPTReplace - Entries [0]IPTEntry -} - // SizeOfIPTReplace is the size of an IPTReplace. const SizeOfIPTReplace = 96 @@ -392,6 +512,8 @@ func (en ExtensionName) String() string { } // TableName holds the name of a netfilter table. +// +// +marshal type TableName [XT_TABLE_MAXNAMELEN]byte // String implements fmt.Stringer. diff --git a/pkg/abi/linux/socket.go b/pkg/abi/linux/socket.go index 95337c168..c24a8216e 100644 --- a/pkg/abi/linux/socket.go +++ b/pkg/abi/linux/socket.go @@ -234,6 +234,8 @@ const ( const SockAddrMax = 128 // InetAddr is struct in_addr, from uapi/linux/in.h. +// +// +marshal type InetAddr [4]byte // SockAddrInet is struct sockaddr_in, from uapi/linux/in.h. @@ -303,6 +305,8 @@ func (s *SockAddrUnix) implementsSockAddr() {} func (s *SockAddrNetlink) implementsSockAddr() {} // Linger is struct linger, from include/linux/socket.h. +// +// +marshal type Linger struct { OnOff int32 Linger int32 @@ -317,6 +321,8 @@ const SizeOfLinger = 8 // the end of this struct or within existing unusued space, so its size grows // over time. The current iteration is based on linux v4.17. New versions are // always backwards compatible. +// +// +marshal type TCPInfo struct { State uint8 CaState uint8 @@ -414,6 +420,8 @@ var SizeOfControlMessageHeader = int(binary.Size(ControlMessageHeader{})) // A ControlMessageCredentials is an SCM_CREDENTIALS socket control message. // // ControlMessageCredentials represents struct ucred from linux/socket.h. +// +// +marshal type ControlMessageCredentials struct { PID int32 UID uint32 diff --git a/pkg/sentry/socket/BUILD b/pkg/sentry/socket/BUILD index c40c6d673..c0fd3425b 100644 --- a/pkg/sentry/socket/BUILD +++ b/pkg/sentry/socket/BUILD @@ -20,5 +20,6 @@ go_library( "//pkg/syserr", "//pkg/tcpip", "//pkg/usermem", + "//tools/go_marshal/marshal", ], ) diff --git a/pkg/sentry/socket/hostinet/BUILD b/pkg/sentry/socket/hostinet/BUILD index ff81ea6e6..e76e498de 100644 --- a/pkg/sentry/socket/hostinet/BUILD +++ b/pkg/sentry/socket/hostinet/BUILD @@ -40,6 +40,8 @@ go_library( "//pkg/tcpip/stack", "//pkg/usermem", "//pkg/waiter", + "//tools/go_marshal/marshal", + "//tools/go_marshal/primitive", "@org_golang_x_sys//unix:go_default_library", ], ) diff --git a/pkg/sentry/socket/hostinet/socket.go b/pkg/sentry/socket/hostinet/socket.go index a92aed2c9..ec5506efc 100644 --- a/pkg/sentry/socket/hostinet/socket.go +++ b/pkg/sentry/socket/hostinet/socket.go @@ -36,6 +36,8 @@ import ( "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" + "gvisor.dev/gvisor/tools/go_marshal/marshal" + "gvisor.dev/gvisor/tools/go_marshal/primitive" ) const ( @@ -319,7 +321,7 @@ func (s *socketOpsCommon) Shutdown(t *kernel.Task, how int) *syserr.Error { } // GetSockOpt implements socket.Socket.GetSockOpt. -func (s *socketOpsCommon) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) { +func (s *socketOpsCommon) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (marshal.Marshallable, *syserr.Error) { if outLen < 0 { return nil, syserr.ErrInvalidArgument } @@ -364,7 +366,8 @@ func (s *socketOpsCommon) GetSockOpt(t *kernel.Task, level int, name int, outPtr if err != nil { return nil, syserr.FromError(err) } - return opt, nil + optP := primitive.ByteSlice(opt) + return &optP, nil } // SetSockOpt implements socket.Socket.SetSockOpt. diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 1243143ea..d9394055d 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -145,7 +145,7 @@ func convertNetstackToBinary(stack *stack.Stack, tablename linux.TableName) (lin // Each rule corresponds to an entry. entry := linux.KernelIPTEntry{ - IPTEntry: linux.IPTEntry{ + Entry: linux.IPTEntry{ IP: linux.IPTIP{ Protocol: uint16(rule.Filter.Protocol), }, @@ -153,20 +153,20 @@ func convertNetstackToBinary(stack *stack.Stack, tablename linux.TableName) (lin TargetOffset: linux.SizeOfIPTEntry, }, } - copy(entry.IPTEntry.IP.Dst[:], rule.Filter.Dst) - copy(entry.IPTEntry.IP.DstMask[:], rule.Filter.DstMask) - copy(entry.IPTEntry.IP.Src[:], rule.Filter.Src) - copy(entry.IPTEntry.IP.SrcMask[:], rule.Filter.SrcMask) - copy(entry.IPTEntry.IP.OutputInterface[:], rule.Filter.OutputInterface) - copy(entry.IPTEntry.IP.OutputInterfaceMask[:], rule.Filter.OutputInterfaceMask) + copy(entry.Entry.IP.Dst[:], rule.Filter.Dst) + copy(entry.Entry.IP.DstMask[:], rule.Filter.DstMask) + copy(entry.Entry.IP.Src[:], rule.Filter.Src) + copy(entry.Entry.IP.SrcMask[:], rule.Filter.SrcMask) + copy(entry.Entry.IP.OutputInterface[:], rule.Filter.OutputInterface) + copy(entry.Entry.IP.OutputInterfaceMask[:], rule.Filter.OutputInterfaceMask) if rule.Filter.DstInvert { - entry.IPTEntry.IP.InverseFlags |= linux.IPT_INV_DSTIP + entry.Entry.IP.InverseFlags |= linux.IPT_INV_DSTIP } if rule.Filter.SrcInvert { - entry.IPTEntry.IP.InverseFlags |= linux.IPT_INV_SRCIP + entry.Entry.IP.InverseFlags |= linux.IPT_INV_SRCIP } if rule.Filter.OutputInterfaceInvert { - entry.IPTEntry.IP.InverseFlags |= linux.IPT_INV_VIA_OUT + entry.Entry.IP.InverseFlags |= linux.IPT_INV_VIA_OUT } for _, matcher := range rule.Matchers { @@ -178,8 +178,8 @@ func convertNetstackToBinary(stack *stack.Stack, tablename linux.TableName) (lin panic(fmt.Sprintf("matcher %T is not 64-bit aligned", matcher)) } entry.Elems = append(entry.Elems, serialized...) - entry.NextOffset += uint16(len(serialized)) - entry.TargetOffset += uint16(len(serialized)) + entry.Entry.NextOffset += uint16(len(serialized)) + entry.Entry.TargetOffset += uint16(len(serialized)) } // Serialize and append the target. @@ -188,11 +188,11 @@ func convertNetstackToBinary(stack *stack.Stack, tablename linux.TableName) (lin panic(fmt.Sprintf("target %T is not 64-bit aligned", rule.Target)) } entry.Elems = append(entry.Elems, serialized...) - entry.NextOffset += uint16(len(serialized)) + entry.Entry.NextOffset += uint16(len(serialized)) nflog("convert to binary: adding entry: %+v", entry) - entries.Size += uint32(entry.NextOffset) + entries.Size += uint32(entry.Entry.NextOffset) entries.Entrytable = append(entries.Entrytable, entry) info.NumEntries++ } diff --git a/pkg/sentry/socket/netlink/BUILD b/pkg/sentry/socket/netlink/BUILD index d5ca3ac56..0546801bf 100644 --- a/pkg/sentry/socket/netlink/BUILD +++ b/pkg/sentry/socket/netlink/BUILD @@ -36,6 +36,8 @@ go_library( "//pkg/tcpip", "//pkg/usermem", "//pkg/waiter", + "//tools/go_marshal/marshal", + "//tools/go_marshal/primitive", ], ) diff --git a/pkg/sentry/socket/netlink/socket.go b/pkg/sentry/socket/netlink/socket.go index 81f34c5a2..98ca7add0 100644 --- a/pkg/sentry/socket/netlink/socket.go +++ b/pkg/sentry/socket/netlink/socket.go @@ -38,6 +38,8 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" + "gvisor.dev/gvisor/tools/go_marshal/marshal" + "gvisor.dev/gvisor/tools/go_marshal/primitive" ) const sizeOfInt32 int = 4 @@ -330,7 +332,7 @@ func (s *socketOpsCommon) Shutdown(t *kernel.Task, how int) *syserr.Error { } // GetSockOpt implements socket.Socket.GetSockOpt. -func (s *socketOpsCommon) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) { +func (s *socketOpsCommon) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (marshal.Marshallable, *syserr.Error) { switch level { case linux.SOL_SOCKET: switch name { @@ -340,24 +342,26 @@ func (s *socketOpsCommon) GetSockOpt(t *kernel.Task, level int, name int, outPtr } s.mu.Lock() defer s.mu.Unlock() - return int32(s.sendBufferSize), nil + sendBufferSizeP := primitive.Int32(s.sendBufferSize) + return &sendBufferSizeP, nil case linux.SO_RCVBUF: if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } // We don't have limit on receiving size. - return int32(math.MaxInt32), nil + recvBufferSizeP := primitive.Int32(math.MaxInt32) + return &recvBufferSizeP, nil case linux.SO_PASSCRED: if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - var passcred int32 + var passcred primitive.Int32 if s.Passcred() { passcred = 1 } - return passcred, nil + return &passcred, nil default: socket.GetSockOptEmitUnimplementedEvent(t, name) diff --git a/pkg/sentry/socket/netstack/BUILD b/pkg/sentry/socket/netstack/BUILD index ea6ebd0e2..1fb777a6c 100644 --- a/pkg/sentry/socket/netstack/BUILD +++ b/pkg/sentry/socket/netstack/BUILD @@ -51,6 +51,8 @@ go_library( "//pkg/tcpip/transport/udp", "//pkg/usermem", "//pkg/waiter", + "//tools/go_marshal/marshal", + "//tools/go_marshal/primitive", "@org_golang_x_sys//unix:go_default_library", ], ) diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 964ec8414..9856ab8c5 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -62,6 +62,8 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/transport/udp" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" + "gvisor.dev/gvisor/tools/go_marshal/marshal" + "gvisor.dev/gvisor/tools/go_marshal/primitive" ) func mustCreateMetric(name, description string) *tcpip.StatCounter { @@ -910,7 +912,7 @@ func (s *socketOpsCommon) Shutdown(t *kernel.Task, how int) *syserr.Error { // GetSockOpt implements the linux syscall getsockopt(2) for sockets backed by // tcpip.Endpoint. -func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) { +func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (marshal.Marshallable, *syserr.Error) { // TODO(b/78348848): Unlike other socket options, SO_TIMESTAMP is // implemented specifically for netstack.SocketOperations rather than // commonEndpoint. commonEndpoint should be extended to support socket @@ -920,25 +922,25 @@ func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr us if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - val := int32(0) + val := primitive.Int32(0) s.readMu.Lock() defer s.readMu.Unlock() if s.sockOptTimestamp { val = 1 } - return val, nil + return &val, nil } if level == linux.SOL_TCP && name == linux.TCP_INQ { if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - val := int32(0) + val := primitive.Int32(0) s.readMu.Lock() defer s.readMu.Unlock() if s.sockOptInq { val = 1 } - return val, nil + return &val, nil } if s.skType == linux.SOCK_RAW && level == linux.IPPROTO_IP { @@ -956,7 +958,7 @@ func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr us if err != nil { return nil, err } - return info, nil + return &info, nil case linux.IPT_SO_GET_ENTRIES: if outLen < linux.SizeOfIPTGetEntries { @@ -971,7 +973,7 @@ func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr us if err != nil { return nil, err } - return entries, nil + return &entries, nil } } @@ -981,7 +983,7 @@ func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr us // GetSockOpt can be used to implement the linux syscall getsockopt(2) for // sockets backed by a commonEndpoint. -func GetSockOpt(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, family int, skType linux.SockType, level, name, outLen int) (interface{}, *syserr.Error) { +func GetSockOpt(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, family int, skType linux.SockType, level, name, outLen int) (marshal.Marshallable, *syserr.Error) { switch level { case linux.SOL_SOCKET: return getSockOptSocket(t, s, ep, family, skType, name, outLen) @@ -1014,7 +1016,7 @@ func boolToInt32(v bool) int32 { } // getSockOptSocket implements GetSockOpt when level is SOL_SOCKET. -func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, family int, skType linux.SockType, name, outLen int) (interface{}, *syserr.Error) { +func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, family int, skType linux.SockType, name, outLen int) (marshal.Marshallable, *syserr.Error) { // TODO(b/124056281): Stop rejecting short optLen values in getsockopt. switch name { case linux.SO_ERROR: @@ -1025,9 +1027,12 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam // Get the last error and convert it. err := ep.GetSockOpt(tcpip.ErrorOption{}) if err == nil { - return int32(0), nil + optP := primitive.Int32(0) + return &optP, nil } - return int32(syserr.TranslateNetstackError(err).ToLinux().Number()), nil + + optP := primitive.Int32(syserr.TranslateNetstackError(err).ToLinux().Number()) + return &optP, nil case linux.SO_PEERCRED: if family != linux.AF_UNIX || outLen < syscall.SizeofUcred { @@ -1035,11 +1040,12 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam } tcred := t.Credentials() - return syscall.Ucred{ - Pid: int32(t.ThreadGroup().ID()), - Uid: uint32(tcred.EffectiveKUID.In(tcred.UserNamespace).OrOverflow()), - Gid: uint32(tcred.EffectiveKGID.In(tcred.UserNamespace).OrOverflow()), - }, nil + creds := linux.ControlMessageCredentials{ + PID: int32(t.ThreadGroup().ID()), + UID: uint32(tcred.EffectiveKUID.In(tcred.UserNamespace).OrOverflow()), + GID: uint32(tcred.EffectiveKGID.In(tcred.UserNamespace).OrOverflow()), + } + return &creds, nil case linux.SO_PASSCRED: if outLen < sizeOfInt32 { @@ -1050,7 +1056,9 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil case linux.SO_SNDBUF: if outLen < sizeOfInt32 { @@ -1066,7 +1074,8 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam size = math.MaxInt32 } - return int32(size), nil + sizeP := primitive.Int32(size) + return &sizeP, nil case linux.SO_RCVBUF: if outLen < sizeOfInt32 { @@ -1082,7 +1091,8 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam size = math.MaxInt32 } - return int32(size), nil + sizeP := primitive.Int32(size) + return &sizeP, nil case linux.SO_REUSEADDR: if outLen < sizeOfInt32 { @@ -1093,7 +1103,8 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil case linux.SO_REUSEPORT: if outLen < sizeOfInt32 { @@ -1104,7 +1115,9 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil case linux.SO_BINDTODEVICE: var v tcpip.BindToDeviceOption @@ -1112,7 +1125,8 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam return nil, syserr.TranslateNetstackError(err) } if v == 0 { - return []byte{}, nil + var b primitive.ByteSlice + return &b, nil } if outLen < linux.IFNAMSIZ { return nil, syserr.ErrInvalidArgument @@ -1127,7 +1141,9 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam // interface was removed. return nil, syserr.ErrUnknownDevice } - return append([]byte(nic.Name), 0), nil + + name := primitive.ByteSlice(append([]byte(nic.Name), 0)) + return &name, nil case linux.SO_BROADCAST: if outLen < sizeOfInt32 { @@ -1138,7 +1154,9 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil case linux.SO_KEEPALIVE: if outLen < sizeOfInt32 { @@ -1149,13 +1167,17 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil case linux.SO_LINGER: if outLen < linux.SizeOfLinger { return nil, syserr.ErrInvalidArgument } - return linux.Linger{}, nil + + linger := linux.Linger{} + return &linger, nil case linux.SO_SNDTIMEO: // TODO(igudger): Linux allows shorter lengths for partial results. @@ -1163,7 +1185,8 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam return nil, syserr.ErrInvalidArgument } - return linux.NsecToTimeval(s.SendTimeout()), nil + sendTimeout := linux.NsecToTimeval(s.SendTimeout()) + return &sendTimeout, nil case linux.SO_RCVTIMEO: // TODO(igudger): Linux allows shorter lengths for partial results. @@ -1171,7 +1194,8 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam return nil, syserr.ErrInvalidArgument } - return linux.NsecToTimeval(s.RecvTimeout()), nil + recvTimeout := linux.NsecToTimeval(s.RecvTimeout()) + return &recvTimeout, nil case linux.SO_OOBINLINE: if outLen < sizeOfInt32 { @@ -1183,7 +1207,8 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam return nil, syserr.TranslateNetstackError(err) } - return int32(v), nil + vP := primitive.Int32(v) + return &vP, nil case linux.SO_NO_CHECK: if outLen < sizeOfInt32 { @@ -1194,7 +1219,8 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil default: socket.GetSockOptEmitUnimplementedEvent(t, name) @@ -1203,7 +1229,7 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam } // getSockOptTCP implements GetSockOpt when level is SOL_TCP. -func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interface{}, *syserr.Error) { +func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (marshal.Marshallable, *syserr.Error) { switch name { case linux.TCP_NODELAY: if outLen < sizeOfInt32 { @@ -1214,7 +1240,9 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(!v), nil + + vP := primitive.Int32(boolToInt32(!v)) + return &vP, nil case linux.TCP_CORK: if outLen < sizeOfInt32 { @@ -1225,7 +1253,9 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil case linux.TCP_QUICKACK: if outLen < sizeOfInt32 { @@ -1236,7 +1266,9 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil case linux.TCP_MAXSEG: if outLen < sizeOfInt32 { @@ -1247,8 +1279,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa if err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(v), nil + vP := primitive.Int32(v) + return &vP, nil case linux.TCP_KEEPIDLE: if outLen < sizeOfInt32 { @@ -1259,8 +1291,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa if err := ep.GetSockOpt(&v); err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(time.Duration(v) / time.Second), nil + keepAliveIdle := primitive.Int32(time.Duration(v) / time.Second) + return &keepAliveIdle, nil case linux.TCP_KEEPINTVL: if outLen < sizeOfInt32 { @@ -1271,8 +1303,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa if err := ep.GetSockOpt(&v); err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(time.Duration(v) / time.Second), nil + keepAliveInterval := primitive.Int32(time.Duration(v) / time.Second) + return &keepAliveInterval, nil case linux.TCP_KEEPCNT: if outLen < sizeOfInt32 { @@ -1283,8 +1315,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa if err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(v), nil + vP := primitive.Int32(v) + return &vP, nil case linux.TCP_USER_TIMEOUT: if outLen < sizeOfInt32 { @@ -1295,8 +1327,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa if err := ep.GetSockOpt(&v); err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(time.Duration(v) / time.Millisecond), nil + tcpUserTimeout := primitive.Int32(time.Duration(v) / time.Millisecond) + return &tcpUserTimeout, nil case linux.TCP_INFO: var v tcpip.TCPInfoOption @@ -1309,12 +1341,13 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa info := linux.TCPInfo{} // Linux truncates the output binary to outLen. - ib := binary.Marshal(nil, usermem.ByteOrder, &info) - if len(ib) > outLen { - ib = ib[:outLen] + buf := t.CopyScratchBuffer(info.SizeBytes()) + info.MarshalUnsafe(buf) + if len(buf) > outLen { + buf = buf[:outLen] } - - return ib, nil + bufP := primitive.ByteSlice(buf) + return &bufP, nil case linux.TCP_CC_INFO, linux.TCP_NOTSENT_LOWAT, @@ -1344,7 +1377,9 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa } b := make([]byte, toCopy) copy(b, v) - return b, nil + + bP := primitive.ByteSlice(b) + return &bP, nil case linux.TCP_LINGER2: if outLen < sizeOfInt32 { @@ -1356,7 +1391,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa return nil, syserr.TranslateNetstackError(err) } - return int32(time.Duration(v) / time.Second), nil + lingerTimeout := primitive.Int32(time.Duration(v) / time.Second) + return &lingerTimeout, nil case linux.TCP_DEFER_ACCEPT: if outLen < sizeOfInt32 { @@ -1368,7 +1404,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa return nil, syserr.TranslateNetstackError(err) } - return int32(time.Duration(v) / time.Second), nil + tcpDeferAccept := primitive.Int32(time.Duration(v) / time.Second) + return &tcpDeferAccept, nil case linux.TCP_SYNCNT: if outLen < sizeOfInt32 { @@ -1379,8 +1416,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa if err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(v), nil + vP := primitive.Int32(v) + return &vP, nil case linux.TCP_WINDOW_CLAMP: if outLen < sizeOfInt32 { @@ -1391,8 +1428,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa if err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(v), nil + vP := primitive.Int32(v) + return &vP, nil default: emitUnimplementedEventTCP(t, name) } @@ -1400,7 +1437,7 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa } // getSockOptIPv6 implements GetSockOpt when level is SOL_IPV6. -func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interface{}, *syserr.Error) { +func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (marshal.Marshallable, *syserr.Error) { switch name { case linux.IPV6_V6ONLY: if outLen < sizeOfInt32 { @@ -1411,7 +1448,9 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil case linux.IPV6_PATHMTU: t.Kernel().EmitUnimplementedEvent(t) @@ -1419,21 +1458,24 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf case linux.IPV6_TCLASS: // Length handling for parity with Linux. if outLen == 0 { - return make([]byte, 0), nil + var b primitive.ByteSlice + return &b, nil } v, err := ep.GetSockOptInt(tcpip.IPv6TrafficClassOption) if err != nil { return nil, syserr.TranslateNetstackError(err) } - uintv := uint32(v) + uintv := primitive.Uint32(v) // Linux truncates the output binary to outLen. - ib := binary.Marshal(nil, usermem.ByteOrder, &uintv) + ib := t.CopyScratchBuffer(uintv.SizeBytes()) + uintv.MarshalUnsafe(ib) // Handle cases where outLen is lesser than sizeOfInt32. if len(ib) > outLen { ib = ib[:outLen] } - return ib, nil + ibP := primitive.ByteSlice(ib) + return &ibP, nil case linux.IPV6_RECVTCLASS: if outLen < sizeOfInt32 { @@ -1444,7 +1486,9 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil default: emitUnimplementedEventIPv6(t, name) @@ -1453,7 +1497,7 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf } // getSockOptIP implements GetSockOpt when level is SOL_IP. -func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family int) (interface{}, *syserr.Error) { +func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family int) (marshal.Marshallable, *syserr.Error) { switch name { case linux.IP_TTL: if outLen < sizeOfInt32 { @@ -1466,11 +1510,12 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in } // Fill in the default value, if needed. - if v == 0 { - v = DefaultTTL + vP := primitive.Int32(v) + if vP == 0 { + vP = DefaultTTL } - return int32(v), nil + return &vP, nil case linux.IP_MULTICAST_TTL: if outLen < sizeOfInt32 { @@ -1482,7 +1527,8 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in return nil, syserr.TranslateNetstackError(err) } - return int32(v), nil + vP := primitive.Int32(v) + return &vP, nil case linux.IP_MULTICAST_IF: if outLen < len(linux.InetAddr{}) { @@ -1496,7 +1542,7 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in a, _ := ConvertAddress(linux.AF_INET, tcpip.FullAddress{Addr: v.InterfaceAddr}) - return a.(*linux.SockAddrInet).Addr, nil + return &a.(*linux.SockAddrInet).Addr, nil case linux.IP_MULTICAST_LOOP: if outLen < sizeOfInt32 { @@ -1507,21 +1553,26 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil case linux.IP_TOS: // Length handling for parity with Linux. if outLen == 0 { - return []byte(nil), nil + var b primitive.ByteSlice + return &b, nil } v, err := ep.GetSockOptInt(tcpip.IPv4TOSOption) if err != nil { return nil, syserr.TranslateNetstackError(err) } if outLen < sizeOfInt32 { - return uint8(v), nil + vP := primitive.Uint8(v) + return &vP, nil } - return int32(v), nil + vP := primitive.Int32(v) + return &vP, nil case linux.IP_RECVTOS: if outLen < sizeOfInt32 { @@ -1532,7 +1583,9 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil case linux.IP_PKTINFO: if outLen < sizeOfInt32 { @@ -1543,7 +1596,9 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil default: emitUnimplementedEventIP(t, name) diff --git a/pkg/sentry/socket/netstack/netstack_vfs2.go b/pkg/sentry/socket/netstack/netstack_vfs2.go index d65a89316..a9025b0ec 100644 --- a/pkg/sentry/socket/netstack/netstack_vfs2.go +++ b/pkg/sentry/socket/netstack/netstack_vfs2.go @@ -31,6 +31,8 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" + "gvisor.dev/gvisor/tools/go_marshal/marshal" + "gvisor.dev/gvisor/tools/go_marshal/primitive" ) // SocketVFS2 encapsulates all the state needed to represent a network stack @@ -200,7 +202,7 @@ func (s *SocketVFS2) Ioctl(ctx context.Context, uio usermem.IO, args arch.Syscal // GetSockOpt implements the linux syscall getsockopt(2) for sockets backed by // tcpip.Endpoint. -func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) { +func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (marshal.Marshallable, *syserr.Error) { // TODO(b/78348848): Unlike other socket options, SO_TIMESTAMP is // implemented specifically for netstack.SocketVFS2 rather than // commonEndpoint. commonEndpoint should be extended to support socket @@ -210,25 +212,25 @@ func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem. if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - val := int32(0) + val := primitive.Int32(0) s.readMu.Lock() defer s.readMu.Unlock() if s.sockOptTimestamp { val = 1 } - return val, nil + return &val, nil } if level == linux.SOL_TCP && name == linux.TCP_INQ { if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - val := int32(0) + val := primitive.Int32(0) s.readMu.Lock() defer s.readMu.Unlock() if s.sockOptInq { val = 1 } - return val, nil + return &val, nil } if s.skType == linux.SOCK_RAW && level == linux.IPPROTO_IP { @@ -246,7 +248,7 @@ func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem. if err != nil { return nil, err } - return info, nil + return &info, nil case linux.IPT_SO_GET_ENTRIES: if outLen < linux.SizeOfIPTGetEntries { @@ -261,7 +263,7 @@ func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem. if err != nil { return nil, err } - return entries, nil + return &entries, nil } } diff --git a/pkg/sentry/socket/socket.go b/pkg/sentry/socket/socket.go index fcd7f9d7f..d112757fb 100644 --- a/pkg/sentry/socket/socket.go +++ b/pkg/sentry/socket/socket.go @@ -35,6 +35,7 @@ import ( "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/tools/go_marshal/marshal" ) // ControlMessages represents the union of unix control messages and tcpip @@ -86,7 +87,7 @@ type SocketOps interface { Shutdown(t *kernel.Task, how int) *syserr.Error // GetSockOpt implements the getsockopt(2) linux syscall. - GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) + GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (marshal.Marshallable, *syserr.Error) // SetSockOpt implements the setsockopt(2) linux syscall. SetSockOpt(t *kernel.Task, level int, name int, opt []byte) *syserr.Error diff --git a/pkg/sentry/socket/unix/BUILD b/pkg/sentry/socket/unix/BUILD index cca5e70f1..061a689a9 100644 --- a/pkg/sentry/socket/unix/BUILD +++ b/pkg/sentry/socket/unix/BUILD @@ -35,5 +35,6 @@ go_library( "//pkg/tcpip", "//pkg/usermem", "//pkg/waiter", + "//tools/go_marshal/marshal", ], ) diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go index 4bb2b6ff4..0482d33cf 100644 --- a/pkg/sentry/socket/unix/unix.go +++ b/pkg/sentry/socket/unix/unix.go @@ -40,6 +40,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" + "gvisor.dev/gvisor/tools/go_marshal/marshal" ) // SocketOperations is a Unix socket. It is similar to a netstack socket, @@ -184,7 +185,7 @@ func (s *SocketOperations) Ioctl(ctx context.Context, _ *fs.File, io usermem.IO, // GetSockOpt implements the linux syscall getsockopt(2) for sockets backed by // a transport.Endpoint. -func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) { +func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (marshal.Marshallable, *syserr.Error) { return netstack.GetSockOpt(t, s, s.ep, linux.AF_UNIX, s.ep.Type(), level, name, outLen) } diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go index ff2149250..05c16fcfe 100644 --- a/pkg/sentry/socket/unix/unix_vfs2.go +++ b/pkg/sentry/socket/unix/unix_vfs2.go @@ -32,6 +32,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" + "gvisor.dev/gvisor/tools/go_marshal/marshal" ) // SocketVFS2 implements socket.SocketVFS2 (and by extension, @@ -89,7 +90,7 @@ func NewFileDescription(ep transport.Endpoint, stype linux.SockType, flags uint3 // GetSockOpt implements the linux syscall getsockopt(2) for sockets backed by // a transport.Endpoint. -func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) { +func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (marshal.Marshallable, *syserr.Error) { return netstack.GetSockOpt(t, s, s.ep, linux.AF_UNIX, s.ep.Type(), level, name, outLen) } diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD index 217fcfef2..4a9b04fd0 100644 --- a/pkg/sentry/syscalls/linux/BUILD +++ b/pkg/sentry/syscalls/linux/BUILD @@ -99,5 +99,7 @@ go_library( "//pkg/syserror", "//pkg/usermem", "//pkg/waiter", + "//tools/go_marshal/marshal", + "//tools/go_marshal/primitive", ], ) diff --git a/pkg/sentry/syscalls/linux/sys_socket.go b/pkg/sentry/syscalls/linux/sys_socket.go index 0760af77b..414fce8e3 100644 --- a/pkg/sentry/syscalls/linux/sys_socket.go +++ b/pkg/sentry/syscalls/linux/sys_socket.go @@ -29,6 +29,8 @@ import ( "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/tools/go_marshal/marshal" + "gvisor.dev/gvisor/tools/go_marshal/primitive" ) // LINT.IfChange @@ -474,7 +476,7 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy } if v != nil { - if _, err := t.CopyOut(optValAddr, v); err != nil { + if _, err := v.CopyOut(t, optValAddr); err != nil { return 0, nil, err } } @@ -484,7 +486,7 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy // getSockOpt tries to handle common socket options, or dispatches to a specific // socket implementation. -func getSockOpt(t *kernel.Task, s socket.Socket, level, name int, optValAddr usermem.Addr, len int) (interface{}, *syserr.Error) { +func getSockOpt(t *kernel.Task, s socket.Socket, level, name int, optValAddr usermem.Addr, len int) (marshal.Marshallable, *syserr.Error) { if level == linux.SOL_SOCKET { switch name { case linux.SO_TYPE, linux.SO_DOMAIN, linux.SO_PROTOCOL: @@ -496,13 +498,16 @@ func getSockOpt(t *kernel.Task, s socket.Socket, level, name int, optValAddr use switch name { case linux.SO_TYPE: _, skType, _ := s.Type() - return int32(skType), nil + v := primitive.Int32(skType) + return &v, nil case linux.SO_DOMAIN: family, _, _ := s.Type() - return int32(family), nil + v := primitive.Int32(family) + return &v, nil case linux.SO_PROTOCOL: _, _, protocol := s.Type() - return int32(protocol), nil + v := primitive.Int32(protocol) + return &v, nil } } @@ -539,7 +544,7 @@ func SetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy return 0, nil, syserror.EINVAL } buf := t.CopyScratchBuffer(int(optLen)) - if _, err := t.CopyIn(optValAddr, &buf); err != nil { + if _, err := t.CopyInBytes(optValAddr, buf); err != nil { return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD index 0c740335b..64696b438 100644 --- a/pkg/sentry/syscalls/linux/vfs2/BUILD +++ b/pkg/sentry/syscalls/linux/vfs2/BUILD @@ -72,5 +72,7 @@ go_library( "//pkg/syserror", "//pkg/usermem", "//pkg/waiter", + "//tools/go_marshal/marshal", + "//tools/go_marshal/primitive", ], ) diff --git a/pkg/sentry/syscalls/linux/vfs2/socket.go b/pkg/sentry/syscalls/linux/vfs2/socket.go index 10b668477..8096a8f9c 100644 --- a/pkg/sentry/syscalls/linux/vfs2/socket.go +++ b/pkg/sentry/syscalls/linux/vfs2/socket.go @@ -30,6 +30,8 @@ import ( "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/tools/go_marshal/marshal" + "gvisor.dev/gvisor/tools/go_marshal/primitive" ) // minListenBacklog is the minimum reasonable backlog for listening sockets. @@ -477,7 +479,7 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy } if v != nil { - if _, err := t.CopyOut(optValAddr, v); err != nil { + if _, err := v.CopyOut(t, optValAddr); err != nil { return 0, nil, err } } @@ -487,7 +489,7 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy // getSockOpt tries to handle common socket options, or dispatches to a specific // socket implementation. -func getSockOpt(t *kernel.Task, s socket.SocketVFS2, level, name int, optValAddr usermem.Addr, len int) (interface{}, *syserr.Error) { +func getSockOpt(t *kernel.Task, s socket.SocketVFS2, level, name int, optValAddr usermem.Addr, len int) (marshal.Marshallable, *syserr.Error) { if level == linux.SOL_SOCKET { switch name { case linux.SO_TYPE, linux.SO_DOMAIN, linux.SO_PROTOCOL: @@ -499,13 +501,16 @@ func getSockOpt(t *kernel.Task, s socket.SocketVFS2, level, name int, optValAddr switch name { case linux.SO_TYPE: _, skType, _ := s.Type() - return int32(skType), nil + v := primitive.Int32(skType) + return &v, nil case linux.SO_DOMAIN: family, _, _ := s.Type() - return int32(family), nil + v := primitive.Int32(family) + return &v, nil case linux.SO_PROTOCOL: _, _, protocol := s.Type() - return int32(protocol), nil + v := primitive.Int32(protocol) + return &v, nil } } @@ -542,7 +547,7 @@ func SetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy return 0, nil, syserror.EINVAL } buf := t.CopyScratchBuffer(int(optLen)) - if _, err := t.CopyIn(optValAddr, &buf); err != nil { + if _, err := t.CopyInBytes(optValAddr, buf); err != nil { return 0, nil, err } diff --git a/tools/go_marshal/README.md b/tools/go_marshal/README.md index 4886efddf..68d759083 100644 --- a/tools/go_marshal/README.md +++ b/tools/go_marshal/README.md @@ -9,11 +9,9 @@ automatically generating code to marshal go data structures to memory. `binary.Marshal` by moving the go runtime reflection necessary to marshal a struct to compile-time. -`go_marshal` automatically generates implementations for `abi.Marshallable` and -`safemem.{Reader,Writer}`. Call-sites for serialization (typically syscall -implementations) can directly invoke `safemem.Reader.ReadToBlocks` and -`safemem.Writer.WriteFromBlocks`. Data structures that require custom -serialization will have manual implementations for these interfaces. +`go_marshal` automatically generates implementations for `marshal.Marshallable` +and `safemem.{Reader,Writer}`. Data structures that require custom serialization +will have manual implementations for these interfaces. Data structures can be flagged for code generation by adding a struct-level comment `// +marshal`. diff --git a/tools/go_marshal/primitive/primitive.go b/tools/go_marshal/primitive/primitive.go index ebcf130ae..d93edda8b 100644 --- a/tools/go_marshal/primitive/primitive.go +++ b/tools/go_marshal/primitive/primitive.go @@ -17,10 +17,22 @@ package primitive import ( + "io" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/tools/go_marshal/marshal" ) +// Int8 is a marshal.Marshallable implementation for int8. +// +// +marshal slice:Int8Slice:inner +type Int8 int8 + +// Uint8 is a marshal.Marshallable implementation for uint8. +// +// +marshal slice:Uint8Slice:inner +type Uint8 uint8 + // Int16 is a marshal.Marshallable implementation for int16. // // +marshal slice:Int16Slice:inner @@ -51,6 +63,66 @@ type Int64 int64 // +marshal slice:Uint64Slice:inner type Uint64 uint64 +// ByteSlice is a marshal.Marshallable implementation for []byte. +// This is a convenience wrapper around a dynamically sized type, and can't be +// embedded in other marshallable types because it breaks assumptions made by +// go-marshal internals. It violates the "no dynamically-sized types" +// constraint of the go-marshal library. +type ByteSlice []byte + +// SizeBytes implements marshal.Marshallable.SizeBytes. +func (b *ByteSlice) SizeBytes() int { + return len(*b) +} + +// MarshalBytes implements marshal.Marshallable.MarshalBytes. +func (b *ByteSlice) MarshalBytes(dst []byte) { + copy(dst, *b) +} + +// UnmarshalBytes implements marshal.Marshallable.UnmarshalBytes. +func (b *ByteSlice) UnmarshalBytes(src []byte) { + copy(*b, src) +} + +// Packed implements marshal.Marshallable.Packed. +func (b *ByteSlice) Packed() bool { + return false +} + +// MarshalUnsafe implements marshal.Marshallable.MarshalUnsafe. +func (b *ByteSlice) MarshalUnsafe(dst []byte) { + b.MarshalBytes(dst) +} + +// UnmarshalUnsafe implements marshal.Marshallable.UnmarshalUnsafe. +func (b *ByteSlice) UnmarshalUnsafe(src []byte) { + b.UnmarshalBytes(src) +} + +// CopyIn implements marshal.Marshallable.CopyIn. +func (b *ByteSlice) CopyIn(task marshal.Task, addr usermem.Addr) (int, error) { + return task.CopyInBytes(addr, *b) +} + +// CopyOut implements marshal.Marshallable.CopyOut. +func (b *ByteSlice) CopyOut(task marshal.Task, addr usermem.Addr) (int, error) { + return task.CopyOutBytes(addr, *b) +} + +// CopyOutN implements marshal.Marshallable.CopyOutN. +func (b *ByteSlice) CopyOutN(task marshal.Task, addr usermem.Addr, limit int) (int, error) { + return task.CopyOutBytes(addr, (*b)[:limit]) +} + +// WriteTo implements io.WriterTo.WriteTo. +func (b *ByteSlice) WriteTo(w io.Writer) (int64, error) { + n, err := w.Write(*b) + return int64(n), err +} + +var _ marshal.Marshallable = (*ByteSlice)(nil) + // Below, we define some convenience functions for marshalling primitive types // using the newtypes above, without requiring superfluous casts. -- cgit v1.2.3 From 3e0e3b9b11fee58835a0a492d66e72b354459e27 Mon Sep 17 00:00:00 2001 From: Ridwan Sharif Date: Tue, 9 Jun 2020 12:35:39 -0400 Subject: Added stub FUSE filesystem Allow FUSE filesystems to be mounted using libfuse. The appropriate flags and mount options are parsed and understood by fusefs. --- pkg/sentry/fsimpl/fuse/BUILD | 5 + pkg/sentry/fsimpl/fuse/dev.go | 28 +++++ pkg/sentry/fsimpl/fuse/fusefs.go | 201 ++++++++++++++++++++++++++++++++ pkg/sentry/syscalls/linux/vfs2/mount.go | 9 +- pkg/sentry/vfs/options.go | 11 ++ runsc/boot/vfs.go | 5 + test/syscalls/linux/dev.cc | 12 ++ test/syscalls/linux/mount.cc | 28 +++++ 8 files changed, 297 insertions(+), 2 deletions(-) create mode 100644 pkg/sentry/fsimpl/fuse/fusefs.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/fuse/BUILD b/pkg/sentry/fsimpl/fuse/BUILD index 3e00c2abb..1fb9c438b 100644 --- a/pkg/sentry/fsimpl/fuse/BUILD +++ b/pkg/sentry/fsimpl/fuse/BUILD @@ -6,15 +6,20 @@ go_library( name = "fuse", srcs = [ "dev.go", + "fusefs.go", ], visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", "//pkg/context", + "//pkg/log", "//pkg/sentry/fsimpl/devtmpfs", + "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/kernel", + "//pkg/sentry/kernel/auth", "//pkg/sentry/vfs", "//pkg/syserror", "//pkg/usermem", + "@org_golang_x_sys//unix:go_default_library", ], ) diff --git a/pkg/sentry/fsimpl/fuse/dev.go b/pkg/sentry/fsimpl/fuse/dev.go index dc33268af..c9e12a94f 100644 --- a/pkg/sentry/fsimpl/fuse/dev.go +++ b/pkg/sentry/fsimpl/fuse/dev.go @@ -51,6 +51,9 @@ type DeviceFD struct { vfs.DentryMetadataFileDescriptionImpl vfs.NoLockFD + // mounted specifies whether a FUSE filesystem was mounted using the DeviceFD. + mounted bool + // TODO(gvisor.dev/issue/2987): Add all the data structures needed to enqueue // and deque requests, control synchronization and establish communication // between the FUSE kernel module and the /dev/fuse character device. @@ -61,26 +64,51 @@ func (fd *DeviceFD) Release() {} // PRead implements vfs.FileDescriptionImpl.PRead. func (fd *DeviceFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + // Operations on /dev/fuse don't make sense until a FUSE filesystem is mounted. + if !fd.mounted { + return 0, syserror.EPERM + } + return 0, syserror.ENOSYS } // Read implements vfs.FileDescriptionImpl.Read. func (fd *DeviceFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + // Operations on /dev/fuse don't make sense until a FUSE filesystem is mounted. + if !fd.mounted { + return 0, syserror.EPERM + } + return 0, syserror.ENOSYS } // PWrite implements vfs.FileDescriptionImpl.PWrite. func (fd *DeviceFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + // Operations on /dev/fuse don't make sense until a FUSE filesystem is mounted. + if !fd.mounted { + return 0, syserror.EPERM + } + return 0, syserror.ENOSYS } // Write implements vfs.FileDescriptionImpl.Write. func (fd *DeviceFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + // Operations on /dev/fuse don't make sense until a FUSE filesystem is mounted. + if !fd.mounted { + return 0, syserror.EPERM + } + return 0, syserror.ENOSYS } // Seek implements vfs.FileDescriptionImpl.Seek. func (fd *DeviceFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + // Operations on /dev/fuse don't make sense until a FUSE filesystem is mounted. + if !fd.mounted { + return 0, syserror.EPERM + } + return 0, syserror.ENOSYS } diff --git a/pkg/sentry/fsimpl/fuse/fusefs.go b/pkg/sentry/fsimpl/fuse/fusefs.go new file mode 100644 index 000000000..6d40e2d87 --- /dev/null +++ b/pkg/sentry/fsimpl/fuse/fusefs.go @@ -0,0 +1,201 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package fuse implements fusefs. +package fuse + +import ( + "strconv" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// Name is the default filesystem name. +const Name = "fuse" + +// FilesystemType implements vfs.FilesystemType. +type FilesystemType struct{} + +type filesystemOptions struct { + // userID specifies the numeric uid of the mount owner. + // This option should not be specified by the filesystem owner. + // It is set by libfuse (or, if libfuse is not used, must be set + // by the filesystem itself). For more information, see man page + // for fuse(8) + userID uint32 + + // groupID specifies the numeric gid of the mount owner. + // This option should not be specified by the filesystem owner. + // It is set by libfuse (or, if libfuse is not used, must be set + // by the filesystem itself). For more information, see man page + // for fuse(8) + groupID uint32 + + // rootMode specifies the the file mode of the filesystem's root. + rootMode linux.FileMode +} + +// filesystem implements vfs.FilesystemImpl. +type filesystem struct { + kernfs.Filesystem + devMinor uint32 + + // fuseFD is the FD returned when opening /dev/fuse. It is used for communication + // between the FUSE server daemon and the sentry fusefs. + fuseFD *DeviceFD + + // opts is the options the fusefs is initialized with. + opts filesystemOptions +} + +// Name implements vfs.FilesystemType.Name. +func (FilesystemType) Name() string { + return Name +} + +// GetFilesystem implements vfs.FilesystemType.GetFilesystem. +func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { + devMinor, err := vfsObj.GetAnonBlockDevMinor() + if err != nil { + return nil, nil, err + } + + var fsopts filesystemOptions + mopts := vfs.GenericParseMountOptions(opts.Data) + deviceDescriptorStr, ok := mopts["fd"] + if !ok { + log.Warningf("%s.GetFilesystem: communication file descriptor N (obtained by opening /dev/fuse) must be specified as 'fd=N'", fsType.Name()) + return nil, nil, syserror.EINVAL + } + delete(mopts, "fd") + + deviceDescriptor, err := strconv.ParseInt(deviceDescriptorStr, 10 /* base */, 32 /* bitSize */) + if err != nil { + return nil, nil, err + } + + kernelTask := kernel.TaskFromContext(ctx) + if kernelTask == nil { + log.Warningf("%s.GetFilesystem: couldn't get kernel task from context", fsType.Name()) + return nil, nil, syserror.EINVAL + } + fuseFd := kernelTask.GetFileVFS2(int32(deviceDescriptor)) + + // Parse and set all the other supported FUSE mount options. + // TODO: Expand the supported mount options. + if userIDStr, ok := mopts["user_id"]; ok { + delete(mopts, "user_id") + userID, err := strconv.ParseUint(userIDStr, 10, 32) + if err != nil { + log.Warningf("%s.GetFilesystem: invalid user_id: user_id=%s", fsType.Name(), userIDStr) + return nil, nil, syserror.EINVAL + } + fsopts.userID = uint32(userID) + } + + if groupIDStr, ok := mopts["group_id"]; ok { + delete(mopts, "group_id") + groupID, err := strconv.ParseUint(groupIDStr, 10, 32) + if err != nil { + log.Warningf("%s.GetFilesystem: invalid group_id: group_id=%s", fsType.Name(), groupIDStr) + return nil, nil, syserror.EINVAL + } + fsopts.groupID = uint32(groupID) + } + + rootMode := linux.FileMode(0777) + modeStr, ok := mopts["rootmode"] + if ok { + delete(mopts, "rootmode") + mode, err := strconv.ParseUint(modeStr, 8, 32) + if err != nil { + log.Warningf("%s.GetFilesystem: invalid mode: %q", fsType.Name(), modeStr) + return nil, nil, syserror.EINVAL + } + rootMode = linux.FileMode(mode & 07777) + } + fsopts.rootMode = rootMode + + // Check for unparsed options. + if len(mopts) != 0 { + log.Warningf("%s.GetFilesystem: unknown options: %v", fsType.Name(), mopts) + return nil, nil, syserror.EINVAL + } + + // Mark the device as ready so it can be used. /dev/fuse can only be used if the FD was used to + // mount a FUSE filesystem. + fuseFD := fuseFd.Impl().(*DeviceFD) + fuseFD.mounted = true + + fs := &filesystem{ + devMinor: devMinor, + fuseFD: fuseFD, + opts: fsopts, + } + + fs.VFSFilesystem().Init(vfsObj, &fsType, fs) + + // TODO: dispatch a FUSE_INIT request to the FUSE daemon server before + // returning. Mount will not block on this dispatched request. + + // root is the fusefs root directory. + defaultFusefsDirMode := linux.FileMode(0755) + root := fs.newInode(creds, defaultFusefsDirMode) + + return fs.VFSFilesystem(), root.VFSDentry(), nil +} + +// Release implements vfs.FilesystemImpl.Release. +func (fs *filesystem) Release() { + fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) + fs.Filesystem.Release() +} + +// Inode implements kernfs.Inode. +type Inode struct { + kernfs.InodeAttrs + kernfs.InodeNoDynamicLookup + kernfs.InodeNotSymlink + kernfs.InodeDirectoryNoNewChildren + kernfs.OrderedChildren + + locks vfs.FileLocks + + dentry kernfs.Dentry +} + +func (fs *filesystem) newInode(creds *auth.Credentials, mode linux.FileMode) *kernfs.Dentry { + i := &Inode{} + i.InodeAttrs.Init(creds, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0755) + i.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) + i.dentry.Init(i) + + return &i.dentry +} + +// Open implements kernfs.Inode.Open. +func (i *Inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &i.locks, &opts) + if err != nil { + return nil, err + } + return fd.VFSFileDescription(), nil +} diff --git a/pkg/sentry/syscalls/linux/vfs2/mount.go b/pkg/sentry/syscalls/linux/vfs2/mount.go index adeaa39cc..ea337de7c 100644 --- a/pkg/sentry/syscalls/linux/vfs2/mount.go +++ b/pkg/sentry/syscalls/linux/vfs2/mount.go @@ -77,8 +77,7 @@ func Mount(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // Silently allow MS_NOSUID, since we don't implement set-id bits // anyway. - const unsupportedFlags = linux.MS_NODEV | - linux.MS_NODIRATIME | linux.MS_STRICTATIME + const unsupportedFlags = linux.MS_NODIRATIME | linux.MS_STRICTATIME // Linux just allows passing any flags to mount(2) - it won't fail when // unknown or unsupported flags are passed. Since we don't implement @@ -94,6 +93,12 @@ func Mount(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if flags&linux.MS_NOEXEC == linux.MS_NOEXEC { opts.Flags.NoExec = true } + if flags&linux.MS_NODEV == linux.MS_NODEV { + opts.Flags.NoDev = true + } + if flags&linux.MS_NOSUID == linux.MS_NOSUID { + opts.Flags.NoSUID = true + } if flags&linux.MS_RDONLY == linux.MS_RDONLY { opts.ReadOnly = true } diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go index f223aeda8..d37208a1f 100644 --- a/pkg/sentry/vfs/options.go +++ b/pkg/sentry/vfs/options.go @@ -79,6 +79,17 @@ type MountFlags struct { // NoATime is equivalent to MS_NOATIME and indicates that the // filesystem should not update access time in-place. NoATime bool + + // NoDev is equivalent to MS_NODEV and indicates that the + // filesystem should not allow access to devices (special files). + // TODO(gVisor.dev/issue/3186): respect this flag in non FUSE + // filesystems. + NoDev bool + + // NoSUID is equivalent to MS_NOSUID and indicates that the + // filesystem should not honor set-user-ID and set-group-ID bits or + // file capabilities when executing programs. + NoSUID bool } // MountOptions contains options to VirtualFilesystem.MountAt(). diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go index 56f4ba15d..9a1ed8e9e 100644 --- a/runsc/boot/vfs.go +++ b/runsc/boot/vfs.go @@ -77,6 +77,10 @@ func registerFilesystems(k *kernel.Kernel) error { AllowUserMount: true, AllowUserList: true, }) + vfsObj.MustRegisterFilesystemType(fuse.Name, &fuse.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + AllowUserList: true, + }) // Setup files in devtmpfs. if err := memdev.Register(vfsObj); err != nil { @@ -119,6 +123,7 @@ func registerFilesystems(k *kernel.Kernel) error { return fmt.Errorf("creating fusedev devtmpfs files: %w", err) } } + return nil } diff --git a/test/syscalls/linux/dev.cc b/test/syscalls/linux/dev.cc index 6fa16208e..69e8b76d3 100644 --- a/test/syscalls/linux/dev.cc +++ b/test/syscalls/linux/dev.cc @@ -161,6 +161,18 @@ TEST(DevTest, OpenDevFuse) { ASSERT_NO_ERRNO_AND_VALUE(Open("/dev/fuse", O_RDONLY)); } +TEST(DevTest, ReadDevFuseWithoutMount) { + // Note(gvisor.dev/issue/3076) This won't work in the sentry until the new + // device registration is complete. + SKIP_IF(IsRunningWithVFS1() || IsRunningOnGvisor()); + + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open("/dev/fuse", O_RDONLY)); + + std::vector buf(1); + EXPECT_THAT(ReadFd(fd.get(), buf.data(), sizeof(buf)), SyscallFailsWithErrno(EPERM)); +} + } // namespace } // namespace testing diff --git a/test/syscalls/linux/mount.cc b/test/syscalls/linux/mount.cc index a3e9745cf..7664fa73d 100644 --- a/test/syscalls/linux/mount.cc +++ b/test/syscalls/linux/mount.cc @@ -321,6 +321,34 @@ TEST(MountTest, RenameRemoveMountPoint) { ASSERT_THAT(rmdir(dir.path().c_str()), SyscallFailsWithErrno(EBUSY)); } +TEST(MountTest, MountFuseFilesystemNoDevice) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SYS_ADMIN))); + + // Note(gvisor.dev/issue/3076) This won't work in the sentry until the new + // device registration is complete. + SKIP_IF(IsRunningWithVFS1() || IsRunningOnGvisor()); + + auto const dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + EXPECT_THAT(mount("", dir.path().c_str(), "fuse", 0, ""), + SyscallFailsWithErrno(EINVAL)); +} + +TEST(MountTest, MountFuseFilesystem) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SYS_ADMIN))); + + // Note(gvisor.dev/issue/3076) This won't work in the sentry until the new + // device registration is complete. + SKIP_IF(IsRunningWithVFS1() || IsRunningOnGvisor()); + + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open("/dev/fuse", O_WRONLY)); + std::string mopts = "fd=" + std::to_string(fd.get()); + + auto const dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + auto const mount = + ASSERT_NO_ERRNO_AND_VALUE(Mount("", dir.path(), "fuse", 0, mopts, 0)); +} + } // namespace } // namespace testing -- cgit v1.2.3 From 2f78c487f17e12dfee08214311c500073cb03fde Mon Sep 17 00:00:00 2001 From: Ridwan Sharif Date: Mon, 20 Jul 2020 16:24:27 -0400 Subject: Use mode supplied by the mount options --- pkg/sentry/fsimpl/fuse/BUILD | 1 - pkg/sentry/fsimpl/fuse/fusefs.go | 7 +++---- 2 files changed, 3 insertions(+), 5 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/fuse/BUILD b/pkg/sentry/fsimpl/fuse/BUILD index 1fb9c438b..737007748 100644 --- a/pkg/sentry/fsimpl/fuse/BUILD +++ b/pkg/sentry/fsimpl/fuse/BUILD @@ -20,6 +20,5 @@ go_library( "//pkg/sentry/vfs", "//pkg/syserror", "//pkg/usermem", - "@org_golang_x_sys//unix:go_default_library", ], ) diff --git a/pkg/sentry/fsimpl/fuse/fusefs.go b/pkg/sentry/fsimpl/fuse/fusefs.go index 6d40e2d87..f7775fb9b 100644 --- a/pkg/sentry/fsimpl/fuse/fusefs.go +++ b/pkg/sentry/fsimpl/fuse/fusefs.go @@ -1,4 +1,4 @@ -// Copyright 2019 The gVisor Authors. +// Copyright 2020 The gVisor Authors. // // Licensed under the Apache License, Version 2.0 (the "License"); // you may not use this file except in compliance with the License. @@ -130,7 +130,7 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt log.Warningf("%s.GetFilesystem: invalid mode: %q", fsType.Name(), modeStr) return nil, nil, syserror.EINVAL } - rootMode = linux.FileMode(mode & 07777) + rootMode = linux.FileMode(mode) } fsopts.rootMode = rootMode @@ -157,8 +157,7 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt // returning. Mount will not block on this dispatched request. // root is the fusefs root directory. - defaultFusefsDirMode := linux.FileMode(0755) - root := fs.newInode(creds, defaultFusefsDirMode) + root := fs.newInode(creds, fsopts.rootMode) return fs.VFSFilesystem(), root.VFSDentry(), nil } -- cgit v1.2.3 From 4eb3c8c7e59b1f3dd0c0bfc253cd5c8c6067d05d Mon Sep 17 00:00:00 2001 From: lubinszARM <34124929+lubinszARM@users.noreply.github.com> Date: Thu, 23 Jul 2020 16:06:46 -0700 Subject: kvm-tls-2:add the preservation of user-TLS in the Arm64 kvm platform This patch load/save TLS for the container application. Related issue: full context-switch supporting for Arm64 #1238 COPYBARA_INTEGRATE_REVIEW=https://github.com/google/gvisor/pull/2761 from lubinszARM:pr_tls_2 cb5dbca1c9c3f378002406da7a58887f9b5032b3 PiperOrigin-RevId: 322887044 --- pkg/sentry/arch/arch_aarch64.go | 23 +++++++++++++---------- pkg/sentry/arch/arch_amd64.go | 4 ++-- pkg/sentry/arch/arch_arm64.go | 4 ++-- pkg/sentry/arch/arch_x86.go | 16 ++++++++++------ pkg/sentry/platform/ring0/kernel_arm64.go | 6 ++++++ 5 files changed, 33 insertions(+), 20 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/arch/arch_aarch64.go b/pkg/sentry/arch/arch_aarch64.go index daba8b172..fd95eb2d2 100644 --- a/pkg/sentry/arch/arch_aarch64.go +++ b/pkg/sentry/arch/arch_aarch64.go @@ -28,7 +28,14 @@ import ( ) // Registers represents the CPU registers for this architecture. -type Registers = linux.PtraceRegs +// +// +stateify savable +type Registers struct { + linux.PtraceRegs + + // TPIDR_EL0 is the EL0 Read/Write Software Thread ID Register. + TPIDR_EL0 uint64 +} const ( // SyscallWidth is the width of insturctions. @@ -101,9 +108,6 @@ type State struct { // Our floating point state. aarch64FPState `state:"wait"` - // TLS pointer - TPValue uint64 - // FeatureSet is a pointer to the currently active feature set. FeatureSet *cpuid.FeatureSet @@ -157,7 +161,6 @@ func (s *State) Fork() State { return State{ Regs: s.Regs, aarch64FPState: s.aarch64FPState.fork(), - TPValue: s.TPValue, FeatureSet: s.FeatureSet, OrigR0: s.OrigR0, } @@ -241,18 +244,18 @@ func (s *State) ptraceGetRegs() Registers { return s.Regs } -var registersSize = (*Registers)(nil).SizeBytes() +var ptraceRegistersSize = (*linux.PtraceRegs)(nil).SizeBytes() // PtraceSetRegs implements Context.PtraceSetRegs. func (s *State) PtraceSetRegs(src io.Reader) (int, error) { var regs Registers - buf := make([]byte, registersSize) + buf := make([]byte, ptraceRegistersSize) if _, err := io.ReadFull(src, buf); err != nil { return 0, err } regs.UnmarshalUnsafe(buf) s.Regs = regs - return registersSize, nil + return ptraceRegistersSize, nil } // PtraceGetFPRegs implements Context.PtraceGetFPRegs. @@ -278,7 +281,7 @@ const ( func (s *State) PtraceGetRegSet(regset uintptr, dst io.Writer, maxlen int) (int, error) { switch regset { case _NT_PRSTATUS: - if maxlen < registersSize { + if maxlen < ptraceRegistersSize { return 0, syserror.EFAULT } return s.PtraceGetRegs(dst) @@ -291,7 +294,7 @@ func (s *State) PtraceGetRegSet(regset uintptr, dst io.Writer, maxlen int) (int, func (s *State) PtraceSetRegSet(regset uintptr, src io.Reader, maxlen int) (int, error) { switch regset { case _NT_PRSTATUS: - if maxlen < registersSize { + if maxlen < ptraceRegistersSize { return 0, syserror.EFAULT } return s.PtraceSetRegs(src) diff --git a/pkg/sentry/arch/arch_amd64.go b/pkg/sentry/arch/arch_amd64.go index 3b3a0a272..1c3e3c14c 100644 --- a/pkg/sentry/arch/arch_amd64.go +++ b/pkg/sentry/arch/arch_amd64.go @@ -300,7 +300,7 @@ func (c *context64) PtracePeekUser(addr uintptr) (interface{}, error) { // PTRACE_PEEKUSER and PTRACE_POKEUSER are only effective on regs and // u_debugreg, returning 0 or silently no-oping for other fields // respectively. - if addr < uintptr(registersSize) { + if addr < uintptr(ptraceRegistersSize) { regs := c.ptraceGetRegs() buf := make([]byte, regs.SizeBytes()) regs.MarshalUnsafe(buf) @@ -315,7 +315,7 @@ func (c *context64) PtracePokeUser(addr, data uintptr) error { if addr&7 != 0 || addr >= userStructSize { return syscall.EIO } - if addr < uintptr(registersSize) { + if addr < uintptr(ptraceRegistersSize) { regs := c.ptraceGetRegs() buf := make([]byte, regs.SizeBytes()) regs.MarshalUnsafe(buf) diff --git a/pkg/sentry/arch/arch_arm64.go b/pkg/sentry/arch/arch_arm64.go index ada7ac7b8..cabbf60e0 100644 --- a/pkg/sentry/arch/arch_arm64.go +++ b/pkg/sentry/arch/arch_arm64.go @@ -142,7 +142,7 @@ func (c *context64) SetStack(value uintptr) { // TLS returns the current TLS pointer. func (c *context64) TLS() uintptr { - return uintptr(c.TPValue) + return uintptr(c.Regs.TPIDR_EL0) } // SetTLS sets the current TLS pointer. Returns false if value is invalid. @@ -151,7 +151,7 @@ func (c *context64) SetTLS(value uintptr) bool { return false } - c.TPValue = uint64(value) + c.Regs.TPIDR_EL0 = uint64(value) return true } diff --git a/pkg/sentry/arch/arch_x86.go b/pkg/sentry/arch/arch_x86.go index dc458b37f..b9405b320 100644 --- a/pkg/sentry/arch/arch_x86.go +++ b/pkg/sentry/arch/arch_x86.go @@ -31,7 +31,11 @@ import ( ) // Registers represents the CPU registers for this architecture. -type Registers = linux.PtraceRegs +// +// +stateify savable +type Registers struct { + linux.PtraceRegs +} // System-related constants for x86. const ( @@ -311,12 +315,12 @@ func (s *State) ptraceGetRegs() Registers { return regs } -var registersSize = (*Registers)(nil).SizeBytes() +var ptraceRegistersSize = (*linux.PtraceRegs)(nil).SizeBytes() // PtraceSetRegs implements Context.PtraceSetRegs. func (s *State) PtraceSetRegs(src io.Reader) (int, error) { var regs Registers - buf := make([]byte, registersSize) + buf := make([]byte, ptraceRegistersSize) if _, err := io.ReadFull(src, buf); err != nil { return 0, err } @@ -374,7 +378,7 @@ func (s *State) PtraceSetRegs(src io.Reader) (int, error) { } regs.Eflags = (s.Regs.Eflags &^ eflagsPtraceMutable) | (regs.Eflags & eflagsPtraceMutable) s.Regs = regs - return registersSize, nil + return ptraceRegistersSize, nil } // isUserSegmentSelector returns true if the given segment selector specifies a @@ -543,7 +547,7 @@ const ( func (s *State) PtraceGetRegSet(regset uintptr, dst io.Writer, maxlen int) (int, error) { switch regset { case _NT_PRSTATUS: - if maxlen < registersSize { + if maxlen < ptraceRegistersSize { return 0, syserror.EFAULT } return s.PtraceGetRegs(dst) @@ -563,7 +567,7 @@ func (s *State) PtraceGetRegSet(regset uintptr, dst io.Writer, maxlen int) (int, func (s *State) PtraceSetRegSet(regset uintptr, src io.Reader, maxlen int) (int, error) { switch regset { case _NT_PRSTATUS: - if maxlen < registersSize { + if maxlen < ptraceRegistersSize { return 0, syserror.EFAULT } return s.PtraceSetRegs(src) diff --git a/pkg/sentry/platform/ring0/kernel_arm64.go b/pkg/sentry/platform/ring0/kernel_arm64.go index ccacaea6b..fca3a5478 100644 --- a/pkg/sentry/platform/ring0/kernel_arm64.go +++ b/pkg/sentry/platform/ring0/kernel_arm64.go @@ -58,7 +58,13 @@ func (c *CPU) SwitchToUser(switchOpts SwitchOpts) (vector Vector) { regs.Pstate &= ^uint64(UserFlagsClear) regs.Pstate |= UserFlagsSet + + SetTLS(regs.TPIDR_EL0) + kernelExitToEl0() + + regs.TPIDR_EL0 = GetTLS() + vector = c.vecCode // Perform the switch. -- cgit v1.2.3 From 8fed97794edcbaa7069dbd39604030e4fbb6891c Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Thu, 23 Jul 2020 16:22:41 -0700 Subject: Add task work mechanism. Like task_work in Linux, this allows us to register callbacks to be executed before returning to userspace. This is needed for kcov support, which requires coverage information to be up-to-date whenever we are in user mode. We will provide coverage data through the kcov interface to enable coverage-directed fuzzing in syzkaller. One difference from Linux is that task work cannot queue work before the transition to userspace that it precedes; queued work will be picked up before the next transition. PiperOrigin-RevId: 322889984 --- pkg/sentry/kernel/BUILD | 1 + pkg/sentry/kernel/task.go | 15 +++++++++++++++ pkg/sentry/kernel/task_run.go | 17 ++++++++++++++++- pkg/sentry/kernel/task_work.go | 38 ++++++++++++++++++++++++++++++++++++++ 4 files changed, 70 insertions(+), 1 deletion(-) create mode 100644 pkg/sentry/kernel/task_work.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD index 25fe1921b..f6886a758 100644 --- a/pkg/sentry/kernel/BUILD +++ b/pkg/sentry/kernel/BUILD @@ -132,6 +132,7 @@ go_library( "task_stop.go", "task_syscall.go", "task_usermem.go", + "task_work.go", "thread_group.go", "threads.go", "timekeeper.go", diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index f48247c94..b3d655b6e 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -68,6 +68,21 @@ type Task struct { // runState is exclusive to the task goroutine. runState taskRunState + // taskWorkCount represents the current size of the task work queue. It is + // used to avoid acquiring taskWorkMu when the queue is empty. + // + // Must accessed with atomic memory operations. + taskWorkCount int32 + + // taskWorkMu protects taskWork. + taskWorkMu sync.Mutex `state:"nosave"` + + // taskWork is a queue of work to be executed before resuming user execution. + // It is similar to the task_work mechanism in Linux. + // + // taskWork is exclusive to the task goroutine. + taskWork []TaskWorker + // haveSyscallReturn is true if tc.Arch().Return() represents a value // returned by a syscall (or set by ptrace after a syscall). // diff --git a/pkg/sentry/kernel/task_run.go b/pkg/sentry/kernel/task_run.go index d654dd997..7d4f44caf 100644 --- a/pkg/sentry/kernel/task_run.go +++ b/pkg/sentry/kernel/task_run.go @@ -167,7 +167,22 @@ func (app *runApp) execute(t *Task) taskRunState { return (*runInterrupt)(nil) } - // We're about to switch to the application again. If there's still a + // Execute any task work callbacks before returning to user space. + if atomic.LoadInt32(&t.taskWorkCount) > 0 { + t.taskWorkMu.Lock() + queue := t.taskWork + t.taskWork = nil + atomic.StoreInt32(&t.taskWorkCount, 0) + t.taskWorkMu.Unlock() + + // Do not hold taskWorkMu while executing task work, which may register + // more work. + for _, work := range queue { + work.TaskWork(t) + } + } + + // We're about to switch to the application again. If there's still an // unhandled SyscallRestartErrno that wasn't translated to an EINTR, // restart the syscall that was interrupted. If there's a saved signal // mask, restore it. (Note that restoring the saved signal mask may unblock diff --git a/pkg/sentry/kernel/task_work.go b/pkg/sentry/kernel/task_work.go new file mode 100644 index 000000000..dda5a433a --- /dev/null +++ b/pkg/sentry/kernel/task_work.go @@ -0,0 +1,38 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package kernel + +import "sync/atomic" + +// TaskWorker is a deferred task. +// +// This must be savable. +type TaskWorker interface { + // TaskWork will be executed prior to returning to user space. Note that + // TaskWork may call RegisterWork again, but this will not be executed until + // the next return to user space, unlike in Linux. This effectively allows + // registration of indefinite user return hooks, but not by default. + TaskWork(t *Task) +} + +// RegisterWork can be used to register additional task work that will be +// performed prior to returning to user space. See TaskWorker.TaskWork for +// semantics regarding registration. +func (t *Task) RegisterWork(work TaskWorker) { + t.taskWorkMu.Lock() + defer t.taskWorkMu.Unlock() + atomic.AddInt32(&t.taskWorkCount, 1) + t.taskWork = append(t.taskWork, work) +} -- cgit v1.2.3 From 4ec351633206fdbd191bc3aef29a007925a731cc Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Thu, 23 Jul 2020 17:40:46 -0700 Subject: Implement get/set_robust_list. PiperOrigin-RevId: 322904430 --- pkg/abi/linux/futex.go | 18 +++++ pkg/sentry/kernel/futex/futex.go | 8 +-- pkg/sentry/kernel/task.go | 4 ++ pkg/sentry/kernel/task_exec.go | 3 + pkg/sentry/kernel/task_exit.go | 3 + pkg/sentry/kernel/task_futex.go | 125 +++++++++++++++++++++++++++++++++ pkg/sentry/syscalls/linux/linux64.go | 4 +- pkg/sentry/syscalls/linux/sys_futex.go | 48 ++++++++++++- test/syscalls/linux/futex.cc | 92 ++++++++++++++++++++++++ 9 files changed, 298 insertions(+), 7 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/futex.go b/pkg/abi/linux/futex.go index 08bfde3b5..8138088a6 100644 --- a/pkg/abi/linux/futex.go +++ b/pkg/abi/linux/futex.go @@ -60,3 +60,21 @@ const ( FUTEX_WAITERS = 0x80000000 FUTEX_OWNER_DIED = 0x40000000 ) + +// FUTEX_BITSET_MATCH_ANY has all bits set. +const FUTEX_BITSET_MATCH_ANY = 0xffffffff + +// ROBUST_LIST_LIMIT protects against a deliberately circular list. +const ROBUST_LIST_LIMIT = 2048 + +// RobustListHead corresponds to Linux's struct robust_list_head. +// +// +marshal +type RobustListHead struct { + List uint64 + FutexOffset uint64 + ListOpPending uint64 +} + +// SizeOfRobustListHead is the size of a RobustListHead struct. +var SizeOfRobustListHead = (*RobustListHead)(nil).SizeBytes() diff --git a/pkg/sentry/kernel/futex/futex.go b/pkg/sentry/kernel/futex/futex.go index 732e66da4..bcc1b29a8 100644 --- a/pkg/sentry/kernel/futex/futex.go +++ b/pkg/sentry/kernel/futex/futex.go @@ -717,10 +717,10 @@ func (m *Manager) lockPILocked(w *Waiter, t Target, addr usermem.Addr, tid uint3 } } -// UnlockPI unlock the futex following the Priority-inheritance futex -// rules. The address provided must contain the caller's TID. If there are -// waiters, TID of the next waiter (FIFO) is set to the given address, and the -// waiter woken up. If there are no waiters, 0 is set to the address. +// UnlockPI unlocks the futex following the Priority-inheritance futex rules. +// The address provided must contain the caller's TID. If there are waiters, +// TID of the next waiter (FIFO) is set to the given address, and the waiter +// woken up. If there are no waiters, 0 is set to the address. func (m *Manager) UnlockPI(t Target, addr usermem.Addr, tid uint32, private bool) error { k, err := getKey(t, addr, private) if err != nil { diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index b3d655b6e..c4db05bd8 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -565,6 +565,10 @@ type Task struct { // futexWaiter is exclusive to the task goroutine. futexWaiter *futex.Waiter `state:"nosave"` + // robustList is a pointer to the head of the tasks's robust futex + // list. + robustList usermem.Addr + // startTime is the real time at which the task started. It is set when // a Task is created or invokes execve(2). // diff --git a/pkg/sentry/kernel/task_exec.go b/pkg/sentry/kernel/task_exec.go index 9b69f3cbe..7803b98d0 100644 --- a/pkg/sentry/kernel/task_exec.go +++ b/pkg/sentry/kernel/task_exec.go @@ -207,6 +207,9 @@ func (r *runSyscallAfterExecStop) execute(t *Task) taskRunState { return flags.CloseOnExec }) + // Handle the robust futex list. + t.exitRobustList() + // NOTE(b/30815691): We currently do not implement privileged // executables (set-user/group-ID bits and file capabilities). This // allows us to unconditionally enable user dumpability on the new mm. diff --git a/pkg/sentry/kernel/task_exit.go b/pkg/sentry/kernel/task_exit.go index c4ade6e8e..231ac548a 100644 --- a/pkg/sentry/kernel/task_exit.go +++ b/pkg/sentry/kernel/task_exit.go @@ -253,6 +253,9 @@ func (*runExitMain) execute(t *Task) taskRunState { } } + // Handle the robust futex list. + t.exitRobustList() + // Deactivate the address space and update max RSS before releasing the // task's MM. t.Deactivate() diff --git a/pkg/sentry/kernel/task_futex.go b/pkg/sentry/kernel/task_futex.go index a53e77c9f..4b535c949 100644 --- a/pkg/sentry/kernel/task_futex.go +++ b/pkg/sentry/kernel/task_futex.go @@ -15,6 +15,7 @@ package kernel import ( + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/kernel/futex" "gvisor.dev/gvisor/pkg/usermem" ) @@ -52,3 +53,127 @@ func (t *Task) LoadUint32(addr usermem.Addr) (uint32, error) { func (t *Task) GetSharedKey(addr usermem.Addr) (futex.Key, error) { return t.MemoryManager().GetSharedFutexKey(t, addr) } + +// GetRobustList sets the robust futex list for the task. +func (t *Task) GetRobustList() usermem.Addr { + t.mu.Lock() + addr := t.robustList + t.mu.Unlock() + return addr +} + +// SetRobustList sets the robust futex list for the task. +func (t *Task) SetRobustList(addr usermem.Addr) { + t.mu.Lock() + t.robustList = addr + t.mu.Unlock() +} + +// exitRobustList walks the robust futex list, marking locks dead and notifying +// wakers. It corresponds to Linux's exit_robust_list(). Following Linux, +// errors are silently ignored. +func (t *Task) exitRobustList() { + t.mu.Lock() + addr := t.robustList + t.robustList = 0 + t.mu.Unlock() + + if addr == 0 { + return + } + + var rl linux.RobustListHead + if _, err := rl.CopyIn(t, usermem.Addr(addr)); err != nil { + return + } + + next := rl.List + done := 0 + var pendingLockAddr usermem.Addr + if rl.ListOpPending != 0 { + pendingLockAddr = usermem.Addr(rl.ListOpPending + rl.FutexOffset) + } + + // Wake up normal elements. + for usermem.Addr(next) != addr { + // We traverse to the next element of the list before we + // actually wake anything. This prevents the race where waking + // this futex causes a modification of the list. + thisLockAddr := usermem.Addr(next + rl.FutexOffset) + + // Try to decode the next element in the list before waking the + // current futex. But don't check the error until after we've + // woken the current futex. Linux does it in this order too + _, nextErr := t.CopyIn(usermem.Addr(next), &next) + + // Wakeup the current futex if it's not pending. + if thisLockAddr != pendingLockAddr { + t.wakeRobustListOne(thisLockAddr) + } + + // If there was an error copying the next futex, we must bail. + if nextErr != nil { + break + } + + // This is a user structure, so it could be a massive list, or + // even contain a loop if they are trying to mess with us. We + // cap traversal to prevent that. + done++ + if done >= linux.ROBUST_LIST_LIMIT { + break + } + } + + // Is there a pending entry to wake? + if pendingLockAddr != 0 { + t.wakeRobustListOne(pendingLockAddr) + } +} + +// wakeRobustListOne wakes a single futex from the robust list. +func (t *Task) wakeRobustListOne(addr usermem.Addr) { + // Bit 0 in address signals PI futex. + pi := addr&1 == 1 + addr = addr &^ 1 + + // Load the futex. + f, err := t.LoadUint32(addr) + if err != nil { + // Can't read this single value? Ignore the problem. + // We can wake the other futexes in the list. + return + } + + tid := uint32(t.ThreadID()) + for { + // Is this held by someone else? + if f&linux.FUTEX_TID_MASK != tid { + return + } + + // This thread is dying and it's holding this futex. We need to + // set the owner died bit and wake up any waiters. + newF := (f & linux.FUTEX_WAITERS) | linux.FUTEX_OWNER_DIED + if curF, err := t.CompareAndSwapUint32(addr, f, newF); err != nil { + return + } else if curF != f { + // Futex changed out from under us. Try again... + f = curF + continue + } + + // Wake waiters if there are any. + if f&linux.FUTEX_WAITERS != 0 { + private := f&linux.FUTEX_PRIVATE_FLAG != 0 + if pi { + t.Futex().UnlockPI(t, addr, tid, private) + return + } + t.Futex().Wake(t, addr, private, linux.FUTEX_BITSET_MATCH_ANY, 1) + } + + // Done. + return + } +} diff --git a/pkg/sentry/syscalls/linux/linux64.go b/pkg/sentry/syscalls/linux/linux64.go index ea4f9b1a7..80c65164a 100644 --- a/pkg/sentry/syscalls/linux/linux64.go +++ b/pkg/sentry/syscalls/linux/linux64.go @@ -325,8 +325,8 @@ var AMD64 = &kernel.SyscallTable{ 270: syscalls.Supported("pselect", Pselect), 271: syscalls.Supported("ppoll", Ppoll), 272: syscalls.PartiallySupported("unshare", Unshare, "Mount, cgroup namespaces not supported. Network namespaces supported but must be empty.", nil), - 273: syscalls.Error("set_robust_list", syserror.ENOSYS, "Obsolete.", nil), - 274: syscalls.Error("get_robust_list", syserror.ENOSYS, "Obsolete.", nil), + 273: syscalls.Supported("set_robust_list", SetRobustList), + 274: syscalls.Supported("get_robust_list", GetRobustList), 275: syscalls.Supported("splice", Splice), 276: syscalls.Supported("tee", Tee), 277: syscalls.PartiallySupported("sync_file_range", SyncFileRange, "Full data flush is not guaranteed at this time.", nil), diff --git a/pkg/sentry/syscalls/linux/sys_futex.go b/pkg/sentry/syscalls/linux/sys_futex.go index b68261f72..f04d78856 100644 --- a/pkg/sentry/syscalls/linux/sys_futex.go +++ b/pkg/sentry/syscalls/linux/sys_futex.go @@ -198,7 +198,7 @@ func Futex(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall switch cmd { case linux.FUTEX_WAIT: // WAIT uses a relative timeout. - mask = ^uint32(0) + mask = linux.FUTEX_BITSET_MATCH_ANY var timeoutDur time.Duration if !forever { timeoutDur = time.Duration(timespec.ToNsecCapped()) * time.Nanosecond @@ -286,3 +286,49 @@ func Futex(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return 0, nil, syserror.ENOSYS } } + +// SetRobustList implements linux syscall set_robust_list(2). +func SetRobustList(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + // Despite the syscall using the name 'pid' for this variable, it is + // very much a tid. + head := args[0].Pointer() + length := args[1].SizeT() + + if length != uint(linux.SizeOfRobustListHead) { + return 0, nil, syserror.EINVAL + } + t.SetRobustList(head) + return 0, nil, nil +} + +// GetRobustList implements linux syscall get_robust_list(2). +func GetRobustList(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + // Despite the syscall using the name 'pid' for this variable, it is + // very much a tid. + tid := args[0].Int() + head := args[1].Pointer() + size := args[2].Pointer() + + if tid < 0 { + return 0, nil, syserror.EINVAL + } + + ot := t + if tid != 0 { + if ot = t.PIDNamespace().TaskWithID(kernel.ThreadID(tid)); ot == nil { + return 0, nil, syserror.ESRCH + } + } + + // Copy out head pointer. + if _, err := t.CopyOut(head, uint64(ot.GetRobustList())); err != nil { + return 0, nil, err + } + + // Copy out size, which is a constant. + if _, err := t.CopyOut(size, uint64(linux.SizeOfRobustListHead)); err != nil { + return 0, nil, err + } + + return 0, nil, nil +} diff --git a/test/syscalls/linux/futex.cc b/test/syscalls/linux/futex.cc index 40c80a6e1..90b1f0508 100644 --- a/test/syscalls/linux/futex.cc +++ b/test/syscalls/linux/futex.cc @@ -18,6 +18,7 @@ #include #include #include +#include #include #include @@ -737,6 +738,97 @@ TEST_P(PrivateAndSharedFutexTest, PITryLockConcurrency_NoRandomSave) { } } +int get_robust_list(int pid, struct robust_list_head** head_ptr, + size_t* len_ptr) { + return syscall(__NR_get_robust_list, pid, head_ptr, len_ptr); +} + +int set_robust_list(struct robust_list_head* head, size_t len) { + return syscall(__NR_set_robust_list, head, len); +} + +TEST(RobustFutexTest, BasicSetGet) { + struct robust_list_head hd = {}; + struct robust_list_head* hd_ptr = &hd; + + // Set! + EXPECT_THAT(set_robust_list(hd_ptr, sizeof(hd)), SyscallSucceedsWithValue(0)); + + // Get! + struct robust_list_head* new_hd_ptr = hd_ptr; + size_t len; + EXPECT_THAT(get_robust_list(0, &new_hd_ptr, &len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(new_hd_ptr, hd_ptr); + EXPECT_EQ(len, sizeof(hd)); +} + +TEST(RobustFutexTest, GetFromOtherTid) { + // Get the current tid and list head. + pid_t tid = gettid(); + struct robust_list_head* hd_ptr = {}; + size_t len; + EXPECT_THAT(get_robust_list(0, &hd_ptr, &len), SyscallSucceedsWithValue(0)); + + // Create a new thread. + ScopedThread t([&] { + // Current tid list head should be different from parent tid. + struct robust_list_head* got_hd_ptr = {}; + EXPECT_THAT(get_robust_list(0, &got_hd_ptr, &len), + SyscallSucceedsWithValue(0)); + EXPECT_NE(hd_ptr, got_hd_ptr); + + // Get the parent list head by passing its tid. + EXPECT_THAT(get_robust_list(tid, &got_hd_ptr, &len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(hd_ptr, got_hd_ptr); + }); + + // Wait for thread. + t.Join(); +} + +TEST(RobustFutexTest, InvalidSize) { + struct robust_list_head* hd = {}; + EXPECT_THAT(set_robust_list(hd, sizeof(*hd) + 1), + SyscallFailsWithErrno(EINVAL)); +} + +TEST(RobustFutexTest, PthreadMutexAttr) { + constexpr int kNumMutexes = 3; + + // Create a bunch of robust mutexes. + pthread_mutexattr_t attrs[kNumMutexes]; + pthread_mutex_t mtxs[kNumMutexes]; + for (int i = 0; i < kNumMutexes; i++) { + TEST_PCHECK(pthread_mutexattr_init(&attrs[i]) == 0); + TEST_PCHECK(pthread_mutexattr_setrobust(&attrs[i], PTHREAD_MUTEX_ROBUST) == + 0); + TEST_PCHECK(pthread_mutex_init(&mtxs[i], &attrs[i]) == 0); + } + + // Start thread to lock the mutexes and then exit. + ScopedThread t([&] { + for (int i = 0; i < kNumMutexes; i++) { + TEST_PCHECK(pthread_mutex_lock(&mtxs[i]) == 0); + } + pthread_exit(NULL); + }); + + // Wait for thread. + t.Join(); + + // Now try to take the mutexes. + for (int i = 0; i < kNumMutexes; i++) { + // Should get EOWNERDEAD. + EXPECT_EQ(pthread_mutex_lock(&mtxs[i]), EOWNERDEAD); + // Make the mutex consistent. + EXPECT_EQ(pthread_mutex_consistent(&mtxs[i]), 0); + // Unlock. + EXPECT_EQ(pthread_mutex_unlock(&mtxs[i]), 0); + } +} + } // namespace } // namespace testing } // namespace gvisor -- cgit v1.2.3 From 82a5cada5944390e738a8b7235fb861965ca40f7 Mon Sep 17 00:00:00 2001 From: Sam Balana Date: Thu, 23 Jul 2020 17:59:12 -0700 Subject: Add AfterFunc to tcpip.Clock Changes the API of tcpip.Clock to also provide a method for scheduling and rescheduling work after a specified duration. This change also implements the AfterFunc method for existing implementations of tcpip.Clock. This is the groundwork required to mock time within tests. All references to CancellableTimer has been replaced with the tcpip.Job interface, allowing for custom implementations of scheduling work. This is a BREAKING CHANGE for clients that implement their own tcpip.Clock or use tcpip.CancellableTimer. Migration plan: 1. Add AfterFunc(d, f) to tcpip.Clock 2. Replace references of tcpip.CancellableTimer with tcpip.Job 3. Replace calls to tcpip.CancellableTimer#StopLocked with tcpip.Job#Cancel 4. Replace calls to tcpip.CancellableTimer#Reset with tcpip.Job#Schedule 5. Replace calls to tcpip.NewCancellableTimer with tcpip.NewJob. PiperOrigin-RevId: 322906897 --- pkg/sentry/kernel/kernel.go | 5 ++ pkg/sentry/kernel/time/BUILD | 1 + pkg/sentry/kernel/time/tcpip.go | 131 +++++++++++++++++++++++++++++ pkg/tcpip/stack/ndp.go | 140 +++++++++++++++---------------- pkg/tcpip/stack/ndp_test.go | 28 +++---- pkg/tcpip/stack/stack.go | 12 ++- pkg/tcpip/tcpip.go | 27 +++++- pkg/tcpip/time_unsafe.go | 30 ++++++- pkg/tcpip/timer.go | 147 ++++++++++++++++++++------------- pkg/tcpip/timer_test.go | 91 ++++++++++---------- pkg/tcpip/transport/icmp/endpoint.go | 2 +- pkg/tcpip/transport/packet/endpoint.go | 2 +- pkg/tcpip/transport/raw/endpoint.go | 2 +- pkg/tcpip/transport/udp/endpoint.go | 2 +- 14 files changed, 429 insertions(+), 191 deletions(-) create mode 100644 pkg/sentry/kernel/time/tcpip.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 240cd6fe0..15dae0f5b 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -1469,6 +1469,11 @@ func (k *Kernel) NowMonotonic() int64 { return now } +// AfterFunc implements tcpip.Clock.AfterFunc. +func (k *Kernel) AfterFunc(d time.Duration, f func()) tcpip.Timer { + return ktime.TcpipAfterFunc(k.realtimeClock, d, f) +} + // SetMemoryFile sets Kernel.mf. SetMemoryFile must be called before Init or // LoadFrom. func (k *Kernel) SetMemoryFile(mf *pgalloc.MemoryFile) { diff --git a/pkg/sentry/kernel/time/BUILD b/pkg/sentry/kernel/time/BUILD index 7ba7dc50c..2817aa3ba 100644 --- a/pkg/sentry/kernel/time/BUILD +++ b/pkg/sentry/kernel/time/BUILD @@ -6,6 +6,7 @@ go_library( name = "time", srcs = [ "context.go", + "tcpip.go", "time.go", ], visibility = ["//pkg/sentry:internal"], diff --git a/pkg/sentry/kernel/time/tcpip.go b/pkg/sentry/kernel/time/tcpip.go new file mode 100644 index 000000000..c4474c0cf --- /dev/null +++ b/pkg/sentry/kernel/time/tcpip.go @@ -0,0 +1,131 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package time + +import ( + "sync" + "time" +) + +// TcpipAfterFunc waits for duration to elapse according to clock then runs fn. +// The timer is started immediately and will fire exactly once. +func TcpipAfterFunc(clock Clock, duration time.Duration, fn func()) *TcpipTimer { + timer := &TcpipTimer{ + clock: clock, + } + timer.notifier = functionNotifier{ + fn: func() { + // tcpip.Timer.Stop() explicitly states that the function is called in a + // separate goroutine that Stop() does not synchronize with. + // Timer.Destroy() synchronizes with calls to TimerListener.Notify(). + // This is semantically meaningful because, in the former case, it's + // legal to call tcpip.Timer.Stop() while holding locks that may also be + // taken by the function, but this isn't so in the latter case. Most + // immediately, Timer calls TimerListener.Notify() while holding + // Timer.mu. A deadlock occurs without spawning a goroutine: + // T1: (Timer expires) + // => Timer.Tick() <- Timer.mu.Lock() called + // => TimerListener.Notify() + // => Timer.Stop() + // => Timer.Destroy() <- Timer.mu.Lock() called, deadlock! + // + // Spawning a goroutine avoids the deadlock: + // T1: (Timer expires) + // => Timer.Tick() <- Timer.mu.Lock() called + // => TimerListener.Notify() <- Launches T2 + // T2: + // => Timer.Stop() + // => Timer.Destroy() <- Timer.mu.Lock() called, blocks + // T1: + // => (returns) <- Timer.mu.Unlock() called + // T2: + // => (continues) <- No deadlock! + go func() { + timer.Stop() + fn() + }() + }, + } + timer.Reset(duration) + return timer +} + +// TcpipTimer is a resettable timer with variable duration expirations. +// Implements tcpip.Timer, which does not define a Destroy method; instead, all +// resources are released after timer expiration and calls to Timer.Stop. +// +// Must be created by AfterFunc. +type TcpipTimer struct { + // clock is the time source. clock is immutable. + clock Clock + + // notifier is called when the Timer expires. notifier is immutable. + notifier functionNotifier + + // mu protects t. + mu sync.Mutex + + // t stores the latest running Timer. This is replaced whenever Reset is + // called since Timer cannot be restarted once it has been Destroyed by Stop. + // + // This field is nil iff Stop has been called. + t *Timer +} + +// Stop implements tcpip.Timer.Stop. +func (r *TcpipTimer) Stop() bool { + r.mu.Lock() + defer r.mu.Unlock() + + if r.t == nil { + return false + } + _, lastSetting := r.t.Swap(Setting{}) + r.t.Destroy() + r.t = nil + return lastSetting.Enabled +} + +// Reset implements tcpip.Timer.Reset. +func (r *TcpipTimer) Reset(d time.Duration) { + r.mu.Lock() + defer r.mu.Unlock() + + if r.t == nil { + r.t = NewTimer(r.clock, &r.notifier) + } + + r.t.Swap(Setting{ + Enabled: true, + Period: 0, + Next: r.clock.Now().Add(d), + }) +} + +// functionNotifier is a TimerListener that runs a function. +// +// functionNotifier cannot be saved or loaded. +type functionNotifier struct { + fn func() +} + +// Notify implements ktime.TimerListener.Notify. +func (f *functionNotifier) Notify(uint64, Setting) (Setting, bool) { + f.fn() + return Setting{}, false +} + +// Destroy implements ktime.TimerListener.Destroy. +func (f *functionNotifier) Destroy() {} diff --git a/pkg/tcpip/stack/ndp.go b/pkg/tcpip/stack/ndp.go index e28c23d66..9dce11a97 100644 --- a/pkg/tcpip/stack/ndp.go +++ b/pkg/tcpip/stack/ndp.go @@ -469,7 +469,7 @@ type ndpState struct { rtrSolicit struct { // The timer used to send the next router solicitation message. - timer *time.Timer + timer tcpip.Timer // Used to let the Router Solicitation timer know that it has been stopped. // @@ -503,7 +503,7 @@ type ndpState struct { // to the DAD goroutine that DAD should stop. type dadState struct { // The DAD timer to send the next NS message, or resolve the address. - timer *time.Timer + timer tcpip.Timer // Used to let the DAD timer know that it has been stopped. // @@ -515,38 +515,38 @@ type dadState struct { // defaultRouterState holds data associated with a default router discovered by // a Router Advertisement (RA). type defaultRouterState struct { - // Timer to invalidate the default router. + // Job to invalidate the default router. // // Must not be nil. - invalidationTimer *tcpip.CancellableTimer + invalidationJob *tcpip.Job } // onLinkPrefixState holds data associated with an on-link prefix discovered by // a Router Advertisement's Prefix Information option (PI) when the NDP // configurations was configured to do so. type onLinkPrefixState struct { - // Timer to invalidate the on-link prefix. + // Job to invalidate the on-link prefix. // // Must not be nil. - invalidationTimer *tcpip.CancellableTimer + invalidationJob *tcpip.Job } // tempSLAACAddrState holds state associated with a temporary SLAAC address. type tempSLAACAddrState struct { - // Timer to deprecate the temporary SLAAC address. + // Job to deprecate the temporary SLAAC address. // // Must not be nil. - deprecationTimer *tcpip.CancellableTimer + deprecationJob *tcpip.Job - // Timer to invalidate the temporary SLAAC address. + // Job to invalidate the temporary SLAAC address. // // Must not be nil. - invalidationTimer *tcpip.CancellableTimer + invalidationJob *tcpip.Job - // Timer to regenerate the temporary SLAAC address. + // Job to regenerate the temporary SLAAC address. // // Must not be nil. - regenTimer *tcpip.CancellableTimer + regenJob *tcpip.Job createdAt time.Time @@ -561,15 +561,15 @@ type tempSLAACAddrState struct { // slaacPrefixState holds state associated with a SLAAC prefix. type slaacPrefixState struct { - // Timer to deprecate the prefix. + // Job to deprecate the prefix. // // Must not be nil. - deprecationTimer *tcpip.CancellableTimer + deprecationJob *tcpip.Job - // Timer to invalidate the prefix. + // Job to invalidate the prefix. // // Must not be nil. - invalidationTimer *tcpip.CancellableTimer + invalidationJob *tcpip.Job // Nonzero only when the address is not valid forever. validUntil time.Time @@ -651,12 +651,12 @@ func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, ref *ref } var done bool - var timer *time.Timer + var timer tcpip.Timer // We initially start a timer to fire immediately because some of the DAD work // cannot be done while holding the NIC's lock. This is effectively the same // as starting a goroutine but we use a timer that fires immediately so we can // reset it for the next DAD iteration. - timer = time.AfterFunc(0, func() { + timer = ndp.nic.stack.Clock().AfterFunc(0, func() { ndp.nic.mu.Lock() defer ndp.nic.mu.Unlock() @@ -871,9 +871,9 @@ func (ndp *ndpState) handleRA(ip tcpip.Address, ra header.NDPRouterAdvert) { case ok && rl != 0: // This is an already discovered default router. Update - // the invalidation timer. - rtr.invalidationTimer.StopLocked() - rtr.invalidationTimer.Reset(rl) + // the invalidation job. + rtr.invalidationJob.Cancel() + rtr.invalidationJob.Schedule(rl) ndp.defaultRouters[ip] = rtr case ok && rl == 0: @@ -950,7 +950,7 @@ func (ndp *ndpState) invalidateDefaultRouter(ip tcpip.Address) { return } - rtr.invalidationTimer.StopLocked() + rtr.invalidationJob.Cancel() delete(ndp.defaultRouters, ip) // Let the integrator know a discovered default router is invalidated. @@ -979,12 +979,12 @@ func (ndp *ndpState) rememberDefaultRouter(ip tcpip.Address, rl time.Duration) { } state := defaultRouterState{ - invalidationTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { ndp.invalidateDefaultRouter(ip) }), } - state.invalidationTimer.Reset(rl) + state.invalidationJob.Schedule(rl) ndp.defaultRouters[ip] = state } @@ -1009,13 +1009,13 @@ func (ndp *ndpState) rememberOnLinkPrefix(prefix tcpip.Subnet, l time.Duration) } state := onLinkPrefixState{ - invalidationTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { ndp.invalidateOnLinkPrefix(prefix) }), } if l < header.NDPInfiniteLifetime { - state.invalidationTimer.Reset(l) + state.invalidationJob.Schedule(l) } ndp.onLinkPrefixes[prefix] = state @@ -1033,7 +1033,7 @@ func (ndp *ndpState) invalidateOnLinkPrefix(prefix tcpip.Subnet) { return } - s.invalidationTimer.StopLocked() + s.invalidationJob.Cancel() delete(ndp.onLinkPrefixes, prefix) // Let the integrator know a discovered on-link prefix is invalidated. @@ -1082,14 +1082,14 @@ func (ndp *ndpState) handleOnLinkPrefixInformation(pi header.NDPPrefixInformatio // This is an already discovered on-link prefix with a // new non-zero valid lifetime. // - // Update the invalidation timer. + // Update the invalidation job. - prefixState.invalidationTimer.StopLocked() + prefixState.invalidationJob.Cancel() if vl < header.NDPInfiniteLifetime { - // Prefix is valid for a finite lifetime, reset the timer to expire after + // Prefix is valid for a finite lifetime, schedule the job to execute after // the new valid lifetime. - prefixState.invalidationTimer.Reset(vl) + prefixState.invalidationJob.Schedule(vl) } ndp.onLinkPrefixes[prefix] = prefixState @@ -1154,7 +1154,7 @@ func (ndp *ndpState) doSLAAC(prefix tcpip.Subnet, pl, vl time.Duration) { } state := slaacPrefixState{ - deprecationTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + deprecationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { state, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for the deprecated SLAAC prefix %s", prefix)) @@ -1162,7 +1162,7 @@ func (ndp *ndpState) doSLAAC(prefix tcpip.Subnet, pl, vl time.Duration) { ndp.deprecateSLAACAddress(state.stableAddr.ref) }), - invalidationTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { state, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for the invalidated SLAAC prefix %s", prefix)) @@ -1184,19 +1184,19 @@ func (ndp *ndpState) doSLAAC(prefix tcpip.Subnet, pl, vl time.Duration) { if !ndp.generateSLAACAddr(prefix, &state) { // We were unable to generate an address for the prefix, we do not nothing - // further as there is no reason to maintain state or timers for a prefix we + // further as there is no reason to maintain state or jobs for a prefix we // do not have an address for. return } - // Setup the initial timers to deprecate and invalidate prefix. + // Setup the initial jobs to deprecate and invalidate prefix. if pl < header.NDPInfiniteLifetime && pl != 0 { - state.deprecationTimer.Reset(pl) + state.deprecationJob.Schedule(pl) } if vl < header.NDPInfiniteLifetime { - state.invalidationTimer.Reset(vl) + state.invalidationJob.Schedule(vl) state.validUntil = now.Add(vl) } @@ -1428,7 +1428,7 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla } state := tempSLAACAddrState{ - deprecationTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + deprecationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { prefixState, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for %s to deprecate temporary address %s", prefix, generatedAddr)) @@ -1441,7 +1441,7 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla ndp.deprecateSLAACAddress(tempAddrState.ref) }), - invalidationTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { prefixState, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for %s to invalidate temporary address %s", prefix, generatedAddr)) @@ -1454,7 +1454,7 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla ndp.invalidateTempSLAACAddr(prefixState.tempAddrs, generatedAddr.Address, tempAddrState) }), - regenTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + regenJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { prefixState, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for %s to regenerate temporary address after %s", prefix, generatedAddr)) @@ -1481,9 +1481,9 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla ref: ref, } - state.deprecationTimer.Reset(pl) - state.invalidationTimer.Reset(vl) - state.regenTimer.Reset(pl - ndp.configs.RegenAdvanceDuration) + state.deprecationJob.Schedule(pl) + state.invalidationJob.Schedule(vl) + state.regenJob.Schedule(pl - ndp.configs.RegenAdvanceDuration) prefixState.generationAttempts++ prefixState.tempAddrs[generatedAddr.Address] = state @@ -1518,16 +1518,16 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat prefixState.stableAddr.ref.deprecated = false } - // If prefix was preferred for some finite lifetime before, stop the - // deprecation timer so it can be reset. - prefixState.deprecationTimer.StopLocked() + // If prefix was preferred for some finite lifetime before, cancel the + // deprecation job so it can be reset. + prefixState.deprecationJob.Cancel() now := time.Now() - // Reset the deprecation timer if prefix has a finite preferred lifetime. + // Schedule the deprecation job if prefix has a finite preferred lifetime. if pl < header.NDPInfiniteLifetime { if !deprecated { - prefixState.deprecationTimer.Reset(pl) + prefixState.deprecationJob.Schedule(pl) } prefixState.preferredUntil = now.Add(pl) } else { @@ -1546,9 +1546,9 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat // 3) Otherwise, reset the valid lifetime of the prefix to 2 hours. if vl >= header.NDPInfiniteLifetime { - // Handle the infinite valid lifetime separately as we do not keep a timer - // in this case. - prefixState.invalidationTimer.StopLocked() + // Handle the infinite valid lifetime separately as we do not schedule a + // job in this case. + prefixState.invalidationJob.Cancel() prefixState.validUntil = time.Time{} } else { var effectiveVl time.Duration @@ -1569,8 +1569,8 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat } if effectiveVl != 0 { - prefixState.invalidationTimer.StopLocked() - prefixState.invalidationTimer.Reset(effectiveVl) + prefixState.invalidationJob.Cancel() + prefixState.invalidationJob.Schedule(effectiveVl) prefixState.validUntil = now.Add(effectiveVl) } } @@ -1582,7 +1582,7 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat } // Note, we do not need to update the entries in the temporary address map - // after updating the timers because the timers are held as pointers. + // after updating the jobs because the jobs are held as pointers. var regenForAddr tcpip.Address allAddressesRegenerated := true for tempAddr, tempAddrState := range prefixState.tempAddrs { @@ -1596,14 +1596,14 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat } // If the address is no longer valid, invalidate it immediately. Otherwise, - // reset the invalidation timer. + // reset the invalidation job. newValidLifetime := validUntil.Sub(now) if newValidLifetime <= 0 { ndp.invalidateTempSLAACAddr(prefixState.tempAddrs, tempAddr, tempAddrState) continue } - tempAddrState.invalidationTimer.StopLocked() - tempAddrState.invalidationTimer.Reset(newValidLifetime) + tempAddrState.invalidationJob.Cancel() + tempAddrState.invalidationJob.Schedule(newValidLifetime) // As per RFC 4941 section 3.3 step 4, the preferred lifetime of a temporary // address is the lower of the preferred lifetime of the stable address or @@ -1616,17 +1616,17 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat } // If the address is no longer preferred, deprecate it immediately. - // Otherwise, reset the deprecation timer. + // Otherwise, schedule the deprecation job again. newPreferredLifetime := preferredUntil.Sub(now) - tempAddrState.deprecationTimer.StopLocked() + tempAddrState.deprecationJob.Cancel() if newPreferredLifetime <= 0 { ndp.deprecateSLAACAddress(tempAddrState.ref) } else { tempAddrState.ref.deprecated = false - tempAddrState.deprecationTimer.Reset(newPreferredLifetime) + tempAddrState.deprecationJob.Schedule(newPreferredLifetime) } - tempAddrState.regenTimer.StopLocked() + tempAddrState.regenJob.Cancel() if tempAddrState.regenerated { } else { allAddressesRegenerated = false @@ -1637,7 +1637,7 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat // immediately after we finish iterating over the temporary addresses. regenForAddr = tempAddr } else { - tempAddrState.regenTimer.Reset(newPreferredLifetime - ndp.configs.RegenAdvanceDuration) + tempAddrState.regenJob.Schedule(newPreferredLifetime - ndp.configs.RegenAdvanceDuration) } } } @@ -1717,7 +1717,7 @@ func (ndp *ndpState) cleanupSLAACAddrResourcesAndNotify(addr tcpip.AddressWithPr ndp.cleanupSLAACPrefixResources(prefix, state) } -// cleanupSLAACPrefixResources cleansup a SLAAC prefix's timers and entry. +// cleanupSLAACPrefixResources cleans up a SLAAC prefix's jobs and entry. // // Panics if the SLAAC prefix is not known. // @@ -1729,8 +1729,8 @@ func (ndp *ndpState) cleanupSLAACPrefixResources(prefix tcpip.Subnet, state slaa } state.stableAddr.ref = nil - state.deprecationTimer.StopLocked() - state.invalidationTimer.StopLocked() + state.deprecationJob.Cancel() + state.invalidationJob.Cancel() delete(ndp.slaacPrefixes, prefix) } @@ -1775,13 +1775,13 @@ func (ndp *ndpState) cleanupTempSLAACAddrResourcesAndNotify(addr tcpip.AddressWi } // cleanupTempSLAACAddrResourcesAndNotify cleans up a temporary SLAAC address's -// timers and entry. +// jobs and entry. // // The NIC that ndp belongs to MUST be locked. func (ndp *ndpState) cleanupTempSLAACAddrResources(tempAddrs map[tcpip.Address]tempSLAACAddrState, tempAddr tcpip.Address, tempAddrState tempSLAACAddrState) { - tempAddrState.deprecationTimer.StopLocked() - tempAddrState.invalidationTimer.StopLocked() - tempAddrState.regenTimer.StopLocked() + tempAddrState.deprecationJob.Cancel() + tempAddrState.invalidationJob.Cancel() + tempAddrState.regenJob.Cancel() delete(tempAddrs, tempAddr) } @@ -1860,7 +1860,7 @@ func (ndp *ndpState) startSolicitingRouters() { var done bool ndp.rtrSolicit.done = &done - ndp.rtrSolicit.timer = time.AfterFunc(delay, func() { + ndp.rtrSolicit.timer = ndp.nic.stack.Clock().AfterFunc(delay, func() { ndp.nic.mu.Lock() if done { // If we reach this point, it means that the RS timer fired after another diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go index 6f86abc98..644ba7c33 100644 --- a/pkg/tcpip/stack/ndp_test.go +++ b/pkg/tcpip/stack/ndp_test.go @@ -1254,7 +1254,7 @@ func TestRouterDiscovery(t *testing.T) { default: } - // Wait for lladdr2's router invalidation timer to fire. The lifetime + // Wait for lladdr2's router invalidation job to execute. The lifetime // of the router should have been updated to the most recent (smaller) // lifetime. // @@ -1271,7 +1271,7 @@ func TestRouterDiscovery(t *testing.T) { e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr2, 0)) expectRouterEvent(llAddr2, false) - // Wait for lladdr3's router invalidation timer to fire. The lifetime + // Wait for lladdr3's router invalidation job to execute. The lifetime // of the router should have been updated to the most recent (smaller) // lifetime. // @@ -1502,7 +1502,7 @@ func TestPrefixDiscovery(t *testing.T) { default: } - // Wait for prefix2's most recent invalidation timer plus some buffer to + // Wait for prefix2's most recent invalidation job plus some buffer to // expire. select { case e := <-ndpDisp.prefixC: @@ -2395,7 +2395,7 @@ func TestAutoGenTempAddrRegen(t *testing.T) { for _, addr := range tempAddrs { // Wait for a deprecation then invalidation event, or just an invalidation // event. We need to cover both cases but cannot deterministically hit both - // cases because the deprecation and invalidation timers could fire in any + // cases because the deprecation and invalidation jobs could execute in any // order. select { case e := <-ndpDisp.autoGenAddrC: @@ -2432,9 +2432,9 @@ func TestAutoGenTempAddrRegen(t *testing.T) { } } -// TestAutoGenTempAddrRegenTimerUpdates tests that a temporary address's -// regeneration timer gets updated when refreshing the address's lifetimes. -func TestAutoGenTempAddrRegenTimerUpdates(t *testing.T) { +// TestAutoGenTempAddrRegenJobUpdates tests that a temporary address's +// regeneration job gets updated when refreshing the address's lifetimes. +func TestAutoGenTempAddrRegenJobUpdates(t *testing.T) { const ( nicID = 1 regenAfter = 2 * time.Second @@ -2533,7 +2533,7 @@ func TestAutoGenTempAddrRegenTimerUpdates(t *testing.T) { // // A new temporary address should immediately be generated since the // regeneration time has already passed since the last address was generated - // - this regeneration does not depend on a timer. + // - this regeneration does not depend on a job. e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, 100, 100)) expectAutoGenAddrEvent(tempAddr2, newAddr) @@ -2559,11 +2559,11 @@ func TestAutoGenTempAddrRegenTimerUpdates(t *testing.T) { } // Set the maximum lifetimes for temporary addresses such that on the next - // RA, the regeneration timer gets reset. + // RA, the regeneration job gets scheduled again. // // The maximum lifetime is the sum of the minimum lifetimes for temporary // addresses + the time that has already passed since the last address was - // generated so that the regeneration timer is needed to generate the next + // generated so that the regeneration job is needed to generate the next // address. newLifetimes := newMinVLDuration + regenAfter + defaultAsyncNegativeEventTimeout ndpConfigs.MaxTempAddrValidLifetime = newLifetimes @@ -2993,9 +2993,9 @@ func TestAutoGenAddrDeprecateFromPI(t *testing.T) { expectPrimaryAddr(addr2) } -// TestAutoGenAddrTimerDeprecation tests that an address is properly deprecated +// TestAutoGenAddrJobDeprecation tests that an address is properly deprecated // when its preferred lifetime expires. -func TestAutoGenAddrTimerDeprecation(t *testing.T) { +func TestAutoGenAddrJobDeprecation(t *testing.T) { const nicID = 1 const newMinVL = 2 newMinVLDuration := newMinVL * time.Second @@ -3513,8 +3513,8 @@ func TestAutoGenAddrRemoval(t *testing.T) { } expectAutoGenAddrEvent(addr, invalidatedAddr) - // Wait for the original valid lifetime to make sure the original timer - // got stopped/cleaned up. + // Wait for the original valid lifetime to make sure the original job got + // cancelled/cleaned up. select { case <-ndpDisp.autoGenAddrC: t.Fatal("unexpectedly received an auto gen addr event") diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index 2b7ece851..a6faa22c2 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -728,6 +728,11 @@ func New(opts Options) *Stack { return s } +// newJob returns a tcpip.Job using the Stack clock. +func (s *Stack) newJob(l sync.Locker, f func()) *tcpip.Job { + return tcpip.NewJob(s.clock, l, f) +} + // UniqueID returns a unique identifier. func (s *Stack) UniqueID() uint64 { return s.uniqueIDGenerator.UniqueID() @@ -801,9 +806,10 @@ func (s *Stack) SetTransportProtocolHandler(p tcpip.TransportProtocolNumber, h f } } -// NowNanoseconds implements tcpip.Clock.NowNanoseconds. -func (s *Stack) NowNanoseconds() int64 { - return s.clock.NowNanoseconds() +// Clock returns the Stack's clock for retrieving the current time and +// scheduling work. +func (s *Stack) Clock() tcpip.Clock { + return s.clock } // Stats returns a mutable copy of the current stats. diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index ff14a3b3c..21aafb0a2 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -192,7 +192,7 @@ func (e ErrSaveRejection) Error() string { return "save rejected due to unsupported networking state: " + e.Err.Error() } -// A Clock provides the current time. +// A Clock provides the current time and schedules work for execution. // // Times returned by a Clock should always be used for application-visible // time. Only monotonic times should be used for netstack internal timekeeping. @@ -203,6 +203,31 @@ type Clock interface { // NowMonotonic returns a monotonic time value. NowMonotonic() int64 + + // AfterFunc waits for the duration to elapse and then calls f in its own + // goroutine. It returns a Timer that can be used to cancel the call using + // its Stop method. + AfterFunc(d time.Duration, f func()) Timer +} + +// Timer represents a single event. A Timer must be created with +// Clock.AfterFunc. +type Timer interface { + // Stop prevents the Timer from firing. It returns true if the call stops the + // timer, false if the timer has already expired or been stopped. + // + // If Stop returns false, then the timer has already expired and the function + // f of Clock.AfterFunc(d, f) has been started in its own goroutine; Stop + // does not wait for f to complete before returning. If the caller needs to + // know whether f is completed, it must coordinate with f explicitly. + Stop() bool + + // Reset changes the timer to expire after duration d. + // + // Reset should be invoked only on stopped or expired timers. If the timer is + // known to have expired, Reset can be used directly. Otherwise, the caller + // must coordinate with the function f of Clock.AfterFunc(d, f). + Reset(d time.Duration) } // Address is a byte slice cast as a string that represents the address of a diff --git a/pkg/tcpip/time_unsafe.go b/pkg/tcpip/time_unsafe.go index 7f172f978..f32d58091 100644 --- a/pkg/tcpip/time_unsafe.go +++ b/pkg/tcpip/time_unsafe.go @@ -20,7 +20,7 @@ package tcpip import ( - _ "time" // Used with go:linkname. + "time" // Used with go:linkname. _ "unsafe" // Required for go:linkname. ) @@ -45,3 +45,31 @@ func (*StdClock) NowMonotonic() int64 { _, _, mono := now() return mono } + +// AfterFunc implements Clock.AfterFunc. +func (*StdClock) AfterFunc(d time.Duration, f func()) Timer { + return &stdTimer{ + t: time.AfterFunc(d, f), + } +} + +type stdTimer struct { + t *time.Timer +} + +var _ Timer = (*stdTimer)(nil) + +// Stop implements Timer.Stop. +func (st *stdTimer) Stop() bool { + return st.t.Stop() +} + +// Reset implements Timer.Reset. +func (st *stdTimer) Reset(d time.Duration) { + st.t.Reset(d) +} + +// NewStdTimer returns a Timer implemented with the time package. +func NewStdTimer(t *time.Timer) Timer { + return &stdTimer{t: t} +} diff --git a/pkg/tcpip/timer.go b/pkg/tcpip/timer.go index 5554c573f..f1dd7c310 100644 --- a/pkg/tcpip/timer.go +++ b/pkg/tcpip/timer.go @@ -20,50 +20,49 @@ import ( "gvisor.dev/gvisor/pkg/sync" ) -// cancellableTimerInstance is a specific instance of CancellableTimer. +// jobInstance is a specific instance of Job. // -// Different instances are created each time CancellableTimer is Reset so each -// timer has its own earlyReturn signal. This is to address a bug when a -// CancellableTimer is stopped and reset in quick succession resulting in a -// timer instance's earlyReturn signal being affected or seen by another timer -// instance. +// Different instances are created each time Job is scheduled so each timer has +// its own earlyReturn signal. This is to address a bug when a Job is stopped +// and reset in quick succession resulting in a timer instance's earlyReturn +// signal being affected or seen by another timer instance. // // Consider the following sceneario where timer instances share a common // earlyReturn signal (T1 creates, stops and resets a Cancellable timer under a // lock L; T2, T3, T4 and T5 are goroutines that handle the first (A), second // (B), third (C), and fourth (D) instance of the timer firing, respectively): // T1: Obtain L -// T1: Create a new CancellableTimer w/ lock L (create instance A) +// T1: Create a new Job w/ lock L (create instance A) // T2: instance A fires, blocked trying to obtain L. // T1: Attempt to stop instance A (set earlyReturn = true) -// T1: Reset timer (create instance B) +// T1: Schedule timer (create instance B) // T3: instance B fires, blocked trying to obtain L. // T1: Attempt to stop instance B (set earlyReturn = true) -// T1: Reset timer (create instance C) +// T1: Schedule timer (create instance C) // T4: instance C fires, blocked trying to obtain L. // T1: Attempt to stop instance C (set earlyReturn = true) -// T1: Reset timer (create instance D) +// T1: Schedule timer (create instance D) // T5: instance D fires, blocked trying to obtain L. // T1: Release L // -// Now that T1 has released L, any of the 4 timer instances can take L and check -// earlyReturn. If the timers simply check earlyReturn and then do nothing -// further, then instance D will never early return even though it was not -// requested to stop. If the timers reset earlyReturn before early returning, -// then all but one of the timers will do work when only one was expected to. -// If CancellableTimer resets earlyReturn when resetting, then all the timers +// Now that T1 has released L, any of the 4 timer instances can take L and +// check earlyReturn. If the timers simply check earlyReturn and then do +// nothing further, then instance D will never early return even though it was +// not requested to stop. If the timers reset earlyReturn before early +// returning, then all but one of the timers will do work when only one was +// expected to. If Job resets earlyReturn when resetting, then all the timers // will fire (again, when only one was expected to). // // To address the above concerns the simplest solution was to give each timer // its own earlyReturn signal. -type cancellableTimerInstance struct { - timer *time.Timer +type jobInstance struct { + timer Timer // Used to inform the timer to early return when it gets stopped while the // lock the timer tries to obtain when fired is held (T1 is a goroutine that // tries to cancel the timer and T2 is the goroutine that handles the timer // firing): - // T1: Obtain the lock, then call StopLocked() + // T1: Obtain the lock, then call Cancel() // T2: timer fires, and gets blocked on obtaining the lock // T1: Releases lock // T2: Obtains lock does unintended work @@ -74,29 +73,33 @@ type cancellableTimerInstance struct { earlyReturn *bool } -// stop stops the timer instance t from firing if it hasn't fired already. If it +// stop stops the job instance j from firing if it hasn't fired already. If it // has fired and is blocked at obtaining the lock, earlyReturn will be set to // true so that it will early return when it obtains the lock. -func (t *cancellableTimerInstance) stop() { - if t.timer != nil { - t.timer.Stop() - *t.earlyReturn = true +func (j *jobInstance) stop() { + if j.timer != nil { + j.timer.Stop() + *j.earlyReturn = true } } -// CancellableTimer is a timer that does some work and can be safely cancelled -// when it fires at the same time some "related work" is being done. +// Job represents some work that can be scheduled for execution. The work can +// be safely cancelled when it fires at the same time some "related work" is +// being done. // // The term "related work" is defined as some work that needs to be done while // holding some lock that the timer must also hold while doing some work. // -// Note, it is not safe to copy a CancellableTimer as its timer instance creates -// a closure over the address of the CancellableTimer. -type CancellableTimer struct { +// Note, it is not safe to copy a Job as its timer instance creates +// a closure over the address of the Job. +type Job struct { _ sync.NoCopy + // The clock used to schedule the backing timer + clock Clock + // The active instance of a cancellable timer. - instance cancellableTimerInstance + instance jobInstance // locker is the lock taken by the timer immediately after it fires and must // be held when attempting to stop the timer. @@ -113,59 +116,91 @@ type CancellableTimer struct { fn func() } -// StopLocked prevents the Timer from firing if it has not fired already. +// Cancel prevents the Job from executing if it has not executed already. // -// If the timer is blocked on obtaining the t.locker lock when StopLocked is -// called, it will early return instead of calling t.fn. +// Cancel requires appropriate locking to be in place for any resources managed +// by the Job. If the Job is blocked on obtaining the lock when Cancel is +// called, it will early return. // // Note, t will be modified. // -// t.locker MUST be locked. -func (t *CancellableTimer) StopLocked() { - t.instance.stop() +// j.locker MUST be locked. +func (j *Job) Cancel() { + j.instance.stop() // Nothing to do with the stopped instance anymore. - t.instance = cancellableTimerInstance{} + j.instance = jobInstance{} } -// Reset changes the timer to expire after duration d. +// Schedule schedules the Job for execution after duration d. This can be +// called on cancelled or completed Jobs to schedule them again. // -// Note, t will be modified. +// Schedule should be invoked only on unscheduled, cancelled, or completed +// Jobs. To be safe, callers should always call Cancel before calling Schedule. // -// Reset should only be called on stopped or expired timers. To be safe, callers -// should always call StopLocked before calling Reset. -func (t *CancellableTimer) Reset(d time.Duration) { +// Note, j will be modified. +func (j *Job) Schedule(d time.Duration) { // Create a new instance. earlyReturn := false // Capture the locker so that updating the timer does not cause a data race // when a timer fires and tries to obtain the lock (read the timer's locker). - locker := t.locker - t.instance = cancellableTimerInstance{ - timer: time.AfterFunc(d, func() { + locker := j.locker + j.instance = jobInstance{ + timer: j.clock.AfterFunc(d, func() { locker.Lock() defer locker.Unlock() if earlyReturn { // If we reach this point, it means that the timer fired while another - // goroutine called StopLocked while it had the lock. Simply return - // here and do nothing further. + // goroutine called Cancel while it had the lock. Simply return here + // and do nothing further. earlyReturn = false return } - t.fn() + j.fn() }), earlyReturn: &earlyReturn, } } -// NewCancellableTimer returns an unscheduled CancellableTimer with the given -// locker and fn. -// -// fn MUST NOT attempt to lock locker. -// -// Callers must call Reset to schedule the timer to fire. -func NewCancellableTimer(locker sync.Locker, fn func()) *CancellableTimer { - return &CancellableTimer{locker: locker, fn: fn} +// NewJob returns a new Job that can be used to schedule f to run in its own +// gorountine. l will be locked before calling f then unlocked after f returns. +// +// var clock tcpip.StdClock +// var mu sync.Mutex +// message := "foo" +// job := tcpip.NewJob(&clock, &mu, func() { +// fmt.Println(message) +// }) +// job.Schedule(time.Second) +// +// mu.Lock() +// message = "bar" +// mu.Unlock() +// +// // Output: bar +// +// f MUST NOT attempt to lock l. +// +// l MUST be locked prior to calling the returned job's Cancel(). +// +// var clock tcpip.StdClock +// var mu sync.Mutex +// message := "foo" +// job := tcpip.NewJob(&clock, &mu, func() { +// fmt.Println(message) +// }) +// job.Schedule(time.Second) +// +// mu.Lock() +// job.Cancel() +// mu.Unlock() +func NewJob(c Clock, l sync.Locker, f func()) *Job { + return &Job{ + clock: c, + locker: l, + fn: f, + } } diff --git a/pkg/tcpip/timer_test.go b/pkg/tcpip/timer_test.go index b4940e397..a82384c49 100644 --- a/pkg/tcpip/timer_test.go +++ b/pkg/tcpip/timer_test.go @@ -28,8 +28,8 @@ const ( longDuration = 1 * time.Second ) -func TestCancellableTimerReassignment(t *testing.T) { - var timer tcpip.CancellableTimer +func TestJobReschedule(t *testing.T) { + var clock tcpip.StdClock var wg sync.WaitGroup var lock sync.Mutex @@ -43,26 +43,27 @@ func TestCancellableTimerReassignment(t *testing.T) { // that has an active timer (even if it has been stopped as a stopped // timer may be blocked on a lock before it can check if it has been // stopped while another goroutine holds the same lock). - timer = *tcpip.NewCancellableTimer(&lock, func() { + job := tcpip.NewJob(&clock, &lock, func() { wg.Done() }) - timer.Reset(shortDuration) + job.Schedule(shortDuration) lock.Unlock() }() } wg.Wait() } -func TestCancellableTimerFire(t *testing.T) { +func TestJobExecution(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) - timer := tcpip.NewCancellableTimer(&lock, func() { + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) - timer.Reset(shortDuration) + job.Schedule(shortDuration) // Wait for timer to fire. select { @@ -82,17 +83,18 @@ func TestCancellableTimerFire(t *testing.T) { func TestCancellableTimerResetFromLongDuration(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) - timer := tcpip.NewCancellableTimer(&lock, func() { ch <- struct{}{} }) - timer.Reset(middleDuration) + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) + job.Schedule(middleDuration) lock.Lock() - timer.StopLocked() + job.Cancel() lock.Unlock() - timer.Reset(shortDuration) + job.Schedule(shortDuration) // Wait for timer to fire. select { @@ -109,16 +111,17 @@ func TestCancellableTimerResetFromLongDuration(t *testing.T) { } } -func TestCancellableTimerResetFromShortDuration(t *testing.T) { +func TestJobRescheduleFromShortDuration(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) lock.Lock() - timer := tcpip.NewCancellableTimer(&lock, func() { ch <- struct{}{} }) - timer.Reset(shortDuration) - timer.StopLocked() + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) + job.Schedule(shortDuration) + job.Cancel() lock.Unlock() // Wait for timer to fire if it wasn't correctly stopped. @@ -128,7 +131,7 @@ func TestCancellableTimerResetFromShortDuration(t *testing.T) { case <-time.After(middleDuration): } - timer.Reset(shortDuration) + job.Schedule(shortDuration) // Wait for timer to fire. select { @@ -145,17 +148,18 @@ func TestCancellableTimerResetFromShortDuration(t *testing.T) { } } -func TestCancellableTimerImmediatelyStop(t *testing.T) { +func TestJobImmediatelyCancel(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) for i := 0; i < 1000; i++ { lock.Lock() - timer := tcpip.NewCancellableTimer(&lock, func() { ch <- struct{}{} }) - timer.Reset(shortDuration) - timer.StopLocked() + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) + job.Schedule(shortDuration) + job.Cancel() lock.Unlock() } @@ -167,25 +171,26 @@ func TestCancellableTimerImmediatelyStop(t *testing.T) { } } -func TestCancellableTimerStoppedResetWithoutLock(t *testing.T) { +func TestJobCancelledRescheduleWithoutLock(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) lock.Lock() - timer := tcpip.NewCancellableTimer(&lock, func() { ch <- struct{}{} }) - timer.Reset(shortDuration) - timer.StopLocked() + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) + job.Schedule(shortDuration) + job.Cancel() lock.Unlock() for i := 0; i < 10; i++ { - timer.Reset(middleDuration) + job.Schedule(middleDuration) lock.Lock() // Sleep until the timer fires and gets blocked trying to take the lock. time.Sleep(middleDuration * 2) - timer.StopLocked() + job.Cancel() lock.Unlock() } @@ -201,17 +206,18 @@ func TestCancellableTimerStoppedResetWithoutLock(t *testing.T) { func TestManyCancellableTimerResetAfterBlockedOnLock(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) lock.Lock() - timer := tcpip.NewCancellableTimer(&lock, func() { ch <- struct{}{} }) - timer.Reset(shortDuration) + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) + job.Schedule(shortDuration) for i := 0; i < 10; i++ { // Sleep until the timer fires and gets blocked trying to take the lock. time.Sleep(middleDuration) - timer.StopLocked() - timer.Reset(shortDuration) + job.Cancel() + job.Schedule(shortDuration) } lock.Unlock() @@ -230,18 +236,19 @@ func TestManyCancellableTimerResetAfterBlockedOnLock(t *testing.T) { } } -func TestManyCancellableTimerResetUnderLock(t *testing.T) { +func TestManyJobReschedulesUnderLock(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) lock.Lock() - timer := tcpip.NewCancellableTimer(&lock, func() { ch <- struct{}{} }) - timer.Reset(shortDuration) + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) + job.Schedule(shortDuration) for i := 0; i < 10; i++ { - timer.StopLocked() - timer.Reset(shortDuration) + job.Cancel() + job.Schedule(shortDuration) } lock.Unlock() diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index 678f4e016..4612be4e7 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -797,7 +797,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk e.rcvList.PushBack(packet) e.rcvBufSize += packet.data.Size() - packet.timestamp = e.stack.NowNanoseconds() + packet.timestamp = e.stack.Clock().NowNanoseconds() e.rcvMu.Unlock() e.stats.PacketsReceived.Increment() diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go index 8f167391f..0e46e6355 100644 --- a/pkg/tcpip/transport/packet/endpoint.go +++ b/pkg/tcpip/transport/packet/endpoint.go @@ -499,7 +499,7 @@ func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress, combinedVV.Append(pkt.Data) packet.data = combinedVV } - packet.timestampNS = ep.stack.NowNanoseconds() + packet.timestampNS = ep.stack.Clock().NowNanoseconds() ep.rcvList.PushBack(&packet) ep.rcvBufSize += packet.data.Size() diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index aefe0e2b2..f85a68554 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -700,7 +700,7 @@ func (e *endpoint) HandlePacket(route *stack.Route, pkt *stack.PacketBuffer) { } combinedVV.Append(pkt.Data) packet.data = combinedVV - packet.timestampNS = e.stack.NowNanoseconds() + packet.timestampNS = e.stack.Clock().NowNanoseconds() e.rcvList.PushBack(packet) e.rcvBufSize += packet.data.Size() diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index a14643ae8..6e692da07 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -1451,7 +1451,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk packet.tos, _ = header.IPv6(pkt.NetworkHeader).TOS() } - packet.timestamp = e.stack.NowNanoseconds() + packet.timestamp = e.stack.Clock().NowNanoseconds() e.rcvMu.Unlock() -- cgit v1.2.3 From bac4ebaabfac95f7b467b9c777a890fcf31a42ae Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Thu, 23 Jul 2020 18:43:20 -0700 Subject: FileDescription is hard to spell. Fix typos. PiperOrigin-RevId: 322913282 --- pkg/sentry/fsimpl/devpts/slave.go | 2 +- pkg/sentry/fsimpl/kernfs/fd_impl_util.go | 6 +++--- pkg/sentry/fsimpl/overlay/non_directory.go | 2 +- pkg/sentry/fsimpl/proc/subtasks.go | 2 +- 4 files changed, 6 insertions(+), 6 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/devpts/slave.go b/pkg/sentry/fsimpl/devpts/slave.go index 2018b978a..a91cae3ef 100644 --- a/pkg/sentry/fsimpl/devpts/slave.go +++ b/pkg/sentry/fsimpl/devpts/slave.go @@ -132,7 +132,7 @@ func (sfd *slaveFileDescription) Write(ctx context.Context, src usermem.IOSequen return sfd.inode.t.ld.outputQueueWrite(ctx, src) } -// Ioctl implements vfs.FileDescripionImpl.Ioctl. +// Ioctl implements vfs.FileDescriptionImpl.Ioctl. func (sfd *slaveFileDescription) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { switch cmd := args[1].Uint(); cmd { case linux.FIONREAD: // linux.FIONREAD == linux.TIOCINQ diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index 3f0aea73a..1d37ccb98 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -112,7 +112,7 @@ func (fd *GenericDirectoryFD) PWrite(ctx context.Context, src usermem.IOSequence return fd.DirectoryFileDescriptionDefaultImpl.PWrite(ctx, src, offset, opts) } -// Release implements vfs.FileDecriptionImpl.Release. +// Release implements vfs.FileDescriptionImpl.Release. func (fd *GenericDirectoryFD) Release() {} func (fd *GenericDirectoryFD) filesystem() *vfs.Filesystem { @@ -123,7 +123,7 @@ func (fd *GenericDirectoryFD) inode() Inode { return fd.vfsfd.VirtualDentry().Dentry().Impl().(*Dentry).inode } -// IterDirents implements vfs.FileDecriptionImpl.IterDirents. IterDirents holds +// IterDirents implements vfs.FileDescriptionImpl.IterDirents. IterDirents holds // o.mu when calling cb. func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback) error { fd.mu.Lock() @@ -198,7 +198,7 @@ func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirent return err } -// Seek implements vfs.FileDecriptionImpl.Seek. +// Seek implements vfs.FileDescriptionImpl.Seek. func (fd *GenericDirectoryFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { fd.mu.Lock() defer fd.mu.Unlock() diff --git a/pkg/sentry/fsimpl/overlay/non_directory.go b/pkg/sentry/fsimpl/overlay/non_directory.go index a3c1f7a8d..6c8445ea1 100644 --- a/pkg/sentry/fsimpl/overlay/non_directory.go +++ b/pkg/sentry/fsimpl/overlay/non_directory.go @@ -176,7 +176,7 @@ func (fd *nonDirectoryFD) SetStat(ctx context.Context, opts vfs.SetStatOptions) return nil } -// StatFS implements vfs.FileDesciptionImpl.StatFS. +// StatFS implements vfs.FileDescriptionImpl.StatFS. func (fd *nonDirectoryFD) StatFS(ctx context.Context) (linux.Statfs, error) { return fd.filesystem().statFS(ctx) } diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go index dad4db1a7..79c2725f3 100644 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -128,7 +128,7 @@ func (fd *subtasksFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallbac return fd.GenericDirectoryFD.IterDirents(ctx, cb) } -// Seek implements vfs.FileDecriptionImpl.Seek. +// Seek implements vfs.FileDescriptionImpl.Seek. func (fd *subtasksFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { if fd.task.ExitState() >= kernel.TaskExitZombie { return 0, syserror.ENOENT -- cgit v1.2.3 From d9a3f5d0c7d675b3cb4519eccca341bac33456af Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Thu, 23 Jul 2020 18:46:10 -0700 Subject: Add permission checks to vfs2 truncate. - Check write permission on truncate(2). Unlike ftruncate(2), truncate(2) fails if the user does not have write permissions on the file. - For gofers under InteropModeShared, check file type before making a truncate request. We should fail early and avoid making an rpc when possible. Furthermore, depending on the remote host's failure may give us unexpected behavior--if the host converts the truncate request to an ftruncate syscall on an open fd, we will get EINVAL instead of EISDIR. Updates #2923. PiperOrigin-RevId: 322913569 --- pkg/sentry/fsimpl/gofer/filesystem.go | 2 +- pkg/sentry/fsimpl/gofer/gofer.go | 18 +++++++++++++++--- pkg/sentry/fsimpl/host/host.go | 7 +++++-- pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 2 +- pkg/sentry/fsimpl/overlay/filesystem.go | 2 +- pkg/sentry/fsimpl/overlay/non_directory.go | 2 +- pkg/sentry/fsimpl/tmpfs/filesystem.go | 2 +- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 7 ++++--- pkg/sentry/syscalls/linux/vfs2/setstat.go | 5 +++++ pkg/sentry/vfs/options.go | 6 ++++++ pkg/sentry/vfs/permissions.go | 8 +++++++- 11 files changed, 47 insertions(+), 14 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 491495b16..00e3c99cd 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -1334,7 +1334,7 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts fs.renameMuRUnlockAndCheckCaching(&ds) return err } - if err := d.setStat(ctx, rp.Credentials(), &opts.Stat, rp.Mount()); err != nil { + if err := d.setStat(ctx, rp.Credentials(), &opts, rp.Mount()); err != nil { fs.renameMuRUnlockAndCheckCaching(&ds) return err } diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index ba7f650b3..e20de84b5 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -888,7 +888,8 @@ func (d *dentry) statTo(stat *linux.Statx) { stat.DevMinor = d.fs.devMinor } -func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *linux.Statx, mnt *vfs.Mount) error { +func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, opts *vfs.SetStatOptions, mnt *vfs.Mount) error { + stat := &opts.Stat if stat.Mask == 0 { return nil } @@ -896,7 +897,7 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin return syserror.EPERM } mode := linux.FileMode(atomic.LoadUint32(&d.mode)) - if err := vfs.CheckSetStat(ctx, creds, stat, mode, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))); err != nil { + if err := vfs.CheckSetStat(ctx, creds, opts, mode, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))); err != nil { return err } if err := mnt.CheckBeginWrite(); err != nil { @@ -937,6 +938,17 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin } if !d.isSynthetic() { if stat.Mask != 0 { + if stat.Mask&linux.STATX_SIZE != 0 { + // Check whether to allow a truncate request to be made. + switch d.mode & linux.S_IFMT { + case linux.S_IFREG: + // Allow. + case linux.S_IFDIR: + return syserror.EISDIR + default: + return syserror.EINVAL + } + } if err := d.file.setAttr(ctx, p9.SetAttrMask{ Permissions: stat.Mask&linux.STATX_MODE != 0, UID: stat.Mask&linux.STATX_UID != 0, @@ -1498,7 +1510,7 @@ func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linu // SetStat implements vfs.FileDescriptionImpl.SetStat. func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { - if err := fd.dentry().setStat(ctx, auth.CredentialsFromContext(ctx), &opts.Stat, fd.vfsfd.Mount()); err != nil { + if err := fd.dentry().setStat(ctx, auth.CredentialsFromContext(ctx), &opts, fd.vfsfd.Mount()); err != nil { return err } if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 1a88cb657..c894f2ca0 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -373,7 +373,7 @@ func (i *inode) fstat(fs *filesystem) (linux.Statx, error) { // SetStat implements kernfs.Inode. func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { - s := opts.Stat + s := &opts.Stat m := s.Mask if m == 0 { @@ -386,7 +386,7 @@ func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Cre if err := syscall.Fstat(i.hostFD, &hostStat); err != nil { return err } - if err := vfs.CheckSetStat(ctx, creds, &s, linux.FileMode(hostStat.Mode&linux.PermissionsMask), auth.KUID(hostStat.Uid), auth.KGID(hostStat.Gid)); err != nil { + if err := vfs.CheckSetStat(ctx, creds, &opts, linux.FileMode(hostStat.Mode), auth.KUID(hostStat.Uid), auth.KGID(hostStat.Gid)); err != nil { return err } @@ -396,6 +396,9 @@ func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Cre } } if m&linux.STATX_SIZE != 0 { + if hostStat.Mode&linux.S_IFMT != linux.S_IFREG { + return syserror.EINVAL + } if err := syscall.Ftruncate(i.hostFD, int64(s.Size)); err != nil { return err } diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 2ab3f1761..579e627f0 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -267,7 +267,7 @@ func (a *InodeAttrs) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *aut if opts.Stat.Mask&^(linux.STATX_MODE|linux.STATX_UID|linux.STATX_GID) != 0 { return syserror.EPERM } - if err := vfs.CheckSetStat(ctx, creds, &opts.Stat, a.Mode(), auth.KUID(atomic.LoadUint32(&a.uid)), auth.KGID(atomic.LoadUint32(&a.gid))); err != nil { + if err := vfs.CheckSetStat(ctx, creds, &opts, a.Mode(), auth.KUID(atomic.LoadUint32(&a.uid)), auth.KGID(atomic.LoadUint32(&a.gid))); err != nil { return err } diff --git a/pkg/sentry/fsimpl/overlay/filesystem.go b/pkg/sentry/fsimpl/overlay/filesystem.go index ff82e1f20..6b705e955 100644 --- a/pkg/sentry/fsimpl/overlay/filesystem.go +++ b/pkg/sentry/fsimpl/overlay/filesystem.go @@ -1104,7 +1104,7 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts } mode := linux.FileMode(atomic.LoadUint32(&d.mode)) - if err := vfs.CheckSetStat(ctx, rp.Credentials(), &opts.Stat, mode, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))); err != nil { + if err := vfs.CheckSetStat(ctx, rp.Credentials(), &opts, mode, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))); err != nil { return err } mnt := rp.Mount() diff --git a/pkg/sentry/fsimpl/overlay/non_directory.go b/pkg/sentry/fsimpl/overlay/non_directory.go index 6c8445ea1..c0749e711 100644 --- a/pkg/sentry/fsimpl/overlay/non_directory.go +++ b/pkg/sentry/fsimpl/overlay/non_directory.go @@ -151,7 +151,7 @@ func (fd *nonDirectoryFD) Stat(ctx context.Context, opts vfs.StatOptions) (linux func (fd *nonDirectoryFD) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { d := fd.dentry() mode := linux.FileMode(atomic.LoadUint32(&d.mode)) - if err := vfs.CheckSetStat(ctx, auth.CredentialsFromContext(ctx), &opts.Stat, mode, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))); err != nil { + if err := vfs.CheckSetStat(ctx, auth.CredentialsFromContext(ctx), &opts, mode, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))); err != nil { return err } mnt := fd.vfsfd.Mount() diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index a0f20c2d4..ef210a69b 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -649,7 +649,7 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts fs.mu.RUnlock() return err } - if err := d.inode.setStat(ctx, rp.Credentials(), &opts.Stat); err != nil { + if err := d.inode.setStat(ctx, rp.Credentials(), &opts); err != nil { fs.mu.RUnlock() return err } diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index d7f4f0779..2545d88e9 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -452,7 +452,8 @@ func (i *inode) statTo(stat *linux.Statx) { } } -func (i *inode) setStat(ctx context.Context, creds *auth.Credentials, stat *linux.Statx) error { +func (i *inode) setStat(ctx context.Context, creds *auth.Credentials, opts *vfs.SetStatOptions) error { + stat := &opts.Stat if stat.Mask == 0 { return nil } @@ -460,7 +461,7 @@ func (i *inode) setStat(ctx context.Context, creds *auth.Credentials, stat *linu return syserror.EPERM } mode := linux.FileMode(atomic.LoadUint32(&i.mode)) - if err := vfs.CheckSetStat(ctx, creds, stat, mode, auth.KUID(atomic.LoadUint32(&i.uid)), auth.KGID(atomic.LoadUint32(&i.gid))); err != nil { + if err := vfs.CheckSetStat(ctx, creds, opts, mode, auth.KUID(atomic.LoadUint32(&i.uid)), auth.KGID(atomic.LoadUint32(&i.gid))); err != nil { return err } i.mu.Lock() @@ -695,7 +696,7 @@ func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linu func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { creds := auth.CredentialsFromContext(ctx) d := fd.dentry() - if err := d.inode.setStat(ctx, creds, &opts.Stat); err != nil { + if err := d.inode.setStat(ctx, creds, &opts); err != nil { return err } diff --git a/pkg/sentry/syscalls/linux/vfs2/setstat.go b/pkg/sentry/syscalls/linux/vfs2/setstat.go index 09ecfed26..6daedd173 100644 --- a/pkg/sentry/syscalls/linux/vfs2/setstat.go +++ b/pkg/sentry/syscalls/linux/vfs2/setstat.go @@ -178,6 +178,7 @@ func Truncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc Mask: linux.STATX_SIZE, Size: uint64(length), }, + NeedWritePerm: true, }) return 0, nil, handleSetSizeError(t, err) } @@ -197,6 +198,10 @@ func Ftruncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys } defer file.DecRef() + if !file.IsWritable() { + return 0, nil, syserror.EINVAL + } + err := file.SetStat(t, vfs.SetStatOptions{ Stat: linux.Statx{ Mask: linux.STATX_SIZE, diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go index d37208a1f..dfc8573fd 100644 --- a/pkg/sentry/vfs/options.go +++ b/pkg/sentry/vfs/options.go @@ -164,6 +164,12 @@ type SetStatOptions struct { // == UTIME_OMIT (VFS users must unset the corresponding bit in Stat.Mask // instead). Stat linux.Statx + + // NeedWritePerm indicates that write permission on the file is needed for + // this operation. This is needed for truncate(2) (note that ftruncate(2) + // does not require the same check--instead, it checks that the fd is + // writable). + NeedWritePerm bool } // BoundEndpointOptions contains options to VirtualFilesystem.BoundEndpointAt() diff --git a/pkg/sentry/vfs/permissions.go b/pkg/sentry/vfs/permissions.go index 9cb050597..33389c1df 100644 --- a/pkg/sentry/vfs/permissions.go +++ b/pkg/sentry/vfs/permissions.go @@ -183,7 +183,8 @@ func MayWriteFileWithOpenFlags(flags uint32) bool { // CheckSetStat checks that creds has permission to change the metadata of a // file with the given permissions, UID, and GID as specified by stat, subject // to the rules of Linux's fs/attr.c:setattr_prepare(). -func CheckSetStat(ctx context.Context, creds *auth.Credentials, stat *linux.Statx, mode linux.FileMode, kuid auth.KUID, kgid auth.KGID) error { +func CheckSetStat(ctx context.Context, creds *auth.Credentials, opts *SetStatOptions, mode linux.FileMode, kuid auth.KUID, kgid auth.KGID) error { + stat := &opts.Stat if stat.Mask&linux.STATX_SIZE != 0 { limit, err := CheckLimit(ctx, 0, int64(stat.Size)) if err != nil { @@ -215,6 +216,11 @@ func CheckSetStat(ctx context.Context, creds *auth.Credentials, stat *linux.Stat return syserror.EPERM } } + if opts.NeedWritePerm && !creds.HasCapability(linux.CAP_DAC_OVERRIDE) { + if err := GenericCheckPermissions(creds, MayWrite, mode, kuid, kgid); err != nil { + return err + } + } if stat.Mask&(linux.STATX_ATIME|linux.STATX_MTIME|linux.STATX_CTIME) != 0 { if !CanActAsOwner(creds, kuid) { if (stat.Mask&linux.STATX_ATIME != 0 && stat.Atime.Nsec != linux.UTIME_NOW) || -- cgit v1.2.3 From e2c70ee9814f0f76ab5c30478748e4c697e91f33 Mon Sep 17 00:00:00 2001 From: Ayush Ranjan Date: Fri, 24 Jul 2020 01:24:16 -0700 Subject: Enable automated marshalling for netstack. PiperOrigin-RevId: 322954792 --- pkg/abi/linux/netdevice.go | 4 +++ pkg/sentry/fs/file_operations.go | 1 + pkg/sentry/socket/netfilter/netfilter.go | 4 +-- pkg/sentry/socket/netstack/netstack.go | 54 ++++++++++++++------------------ 4 files changed, 31 insertions(+), 32 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/netdevice.go b/pkg/abi/linux/netdevice.go index 7866352b4..0faf015c7 100644 --- a/pkg/abi/linux/netdevice.go +++ b/pkg/abi/linux/netdevice.go @@ -22,6 +22,8 @@ const ( ) // IFReq is an interface request. +// +// +marshal type IFReq struct { // IFName is an encoded name, normally null-terminated. This should be // accessed via the Name and SetName functions. @@ -79,6 +81,8 @@ type IFMap struct { // IFConf is used to return a list of interfaces and their addresses. See // netdevice(7) and struct ifconf for more detail on its use. +// +// +marshal type IFConf struct { Len int32 _ [4]byte // Pad to sizeof(struct ifconf). diff --git a/pkg/sentry/fs/file_operations.go b/pkg/sentry/fs/file_operations.go index beba0f771..f5537411e 100644 --- a/pkg/sentry/fs/file_operations.go +++ b/pkg/sentry/fs/file_operations.go @@ -160,6 +160,7 @@ type FileOperations interface { // refer. // // Preconditions: The AddressSpace (if any) that io refers to is activated. + // Must only be called from a task goroutine. Ioctl(ctx context.Context, file *File, io usermem.IO, args arch.SyscallArguments) (uintptr, error) } diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index d9394055d..a9f0604ae 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -66,7 +66,7 @@ func nflog(format string, args ...interface{}) { func GetInfo(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr) (linux.IPTGetinfo, *syserr.Error) { // Read in the struct and table name. var info linux.IPTGetinfo - if _, err := t.CopyIn(outPtr, &info); err != nil { + if _, err := info.CopyIn(t, outPtr); err != nil { return linux.IPTGetinfo{}, syserr.FromError(err) } @@ -84,7 +84,7 @@ func GetInfo(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr) (linux.IPT func GetEntries(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr, outLen int) (linux.KernelIPTGetEntries, *syserr.Error) { // Read in the struct and table name. var userEntries linux.IPTGetEntries - if _, err := t.CopyIn(outPtr, &userEntries); err != nil { + if _, err := userEntries.CopyIn(t, outPtr); err != nil { nflog("couldn't copy in entries %q", userEntries.Name) return linux.KernelIPTGetEntries{}, syserr.FromError(err) } diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 9856ab8c5..44b3fff46 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -2835,6 +2835,11 @@ func (s *SocketOperations) Ioctl(ctx context.Context, _ *fs.File, io usermem.IO, } func (s *socketOpsCommon) ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { + t := kernel.TaskFromContext(ctx) + if t == nil { + panic("ioctl(2) may only be called from a task goroutine") + } + // SIOCGSTAMP is implemented by netstack rather than all commonEndpoint // sockets. // TODO(b/78348848): Add a commonEndpoint method to support SIOCGSTAMP. @@ -2847,9 +2852,7 @@ func (s *socketOpsCommon) ioctl(ctx context.Context, io usermem.IO, args arch.Sy } tv := linux.NsecToTimeval(s.timestampNS) - _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), &tv, usermem.IOOpts{ - AddressSpaceActive: true, - }) + _, err := tv.CopyOut(t, args[2].Pointer()) return 0, err case linux.TIOCINQ: @@ -2868,9 +2871,8 @@ func (s *socketOpsCommon) ioctl(ctx context.Context, io usermem.IO, args arch.Sy } // Copy result to userspace. - _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), int32(v), usermem.IOOpts{ - AddressSpaceActive: true, - }) + vP := primitive.Int32(v) + _, err := vP.CopyOut(t, args[2].Pointer()) return 0, err } @@ -2879,6 +2881,11 @@ func (s *socketOpsCommon) ioctl(ctx context.Context, io usermem.IO, args arch.Sy // Ioctl performs a socket ioctl. func Ioctl(ctx context.Context, ep commonEndpoint, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { + t := kernel.TaskFromContext(ctx) + if t == nil { + panic("ioctl(2) may only be called from a task goroutine") + } + switch arg := int(args[1].Int()); arg { case linux.SIOCGIFFLAGS, linux.SIOCGIFADDR, @@ -2895,37 +2902,28 @@ func Ioctl(ctx context.Context, ep commonEndpoint, io usermem.IO, args arch.Sysc linux.SIOCETHTOOL: var ifr linux.IFReq - if _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &ifr, usermem.IOOpts{ - AddressSpaceActive: true, - }); err != nil { + if _, err := ifr.CopyIn(t, args[2].Pointer()); err != nil { return 0, err } if err := interfaceIoctl(ctx, io, arg, &ifr); err != nil { return 0, err.ToError() } - _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), &ifr, usermem.IOOpts{ - AddressSpaceActive: true, - }) + _, err := ifr.CopyOut(t, args[2].Pointer()) return 0, err case linux.SIOCGIFCONF: // Return a list of interface addresses or the buffer size // necessary to hold the list. var ifc linux.IFConf - if _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &ifc, usermem.IOOpts{ - AddressSpaceActive: true, - }); err != nil { + if _, err := ifc.CopyIn(t, args[2].Pointer()); err != nil { return 0, err } - if err := ifconfIoctl(ctx, io, &ifc); err != nil { + if err := ifconfIoctl(ctx, t, io, &ifc); err != nil { return 0, err } - _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), ifc, usermem.IOOpts{ - AddressSpaceActive: true, - }) - + _, err := ifc.CopyOut(t, args[2].Pointer()) return 0, err case linux.TIOCINQ: @@ -2938,9 +2936,8 @@ func Ioctl(ctx context.Context, ep commonEndpoint, io usermem.IO, args arch.Sysc v = math.MaxInt32 } // Copy result to userspace. - _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), int32(v), usermem.IOOpts{ - AddressSpaceActive: true, - }) + vP := primitive.Int32(v) + _, err := vP.CopyOut(t, args[2].Pointer()) return 0, err case linux.TIOCOUTQ: @@ -2954,9 +2951,8 @@ func Ioctl(ctx context.Context, ep commonEndpoint, io usermem.IO, args arch.Sysc } // Copy result to userspace. - _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), int32(v), usermem.IOOpts{ - AddressSpaceActive: true, - }) + vP := primitive.Int32(v) + _, err := vP.CopyOut(t, args[2].Pointer()) return 0, err case linux.SIOCGIFMEM, linux.SIOCGIFPFLAGS, linux.SIOCGMIIPHY, linux.SIOCGMIIREG: @@ -3105,7 +3101,7 @@ func interfaceIoctl(ctx context.Context, io usermem.IO, arg int, ifr *linux.IFRe } // ifconfIoctl populates a struct ifconf for the SIOCGIFCONF ioctl. -func ifconfIoctl(ctx context.Context, io usermem.IO, ifc *linux.IFConf) error { +func ifconfIoctl(ctx context.Context, t *kernel.Task, io usermem.IO, ifc *linux.IFConf) error { // If Ptr is NULL, return the necessary buffer size via Len. // Otherwise, write up to Len bytes starting at Ptr containing ifreq // structs. @@ -3142,9 +3138,7 @@ func ifconfIoctl(ctx context.Context, io usermem.IO, ifc *linux.IFConf) error { // Copy the ifr to userspace. dst := uintptr(ifc.Ptr) + uintptr(ifc.Len) ifc.Len += int32(linux.SizeOfIFReq) - if _, err := usermem.CopyObjectOut(ctx, io, usermem.Addr(dst), ifr, usermem.IOOpts{ - AddressSpaceActive: true, - }); err != nil { + if _, err := ifr.CopyOut(t, usermem.Addr(dst)); err != nil { return err } } -- cgit v1.2.3 From ced5863c4938ab2219331dc88bb3a301c1815918 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Fri, 10 Jul 2020 03:21:50 -0400 Subject: allow guest user applications read CNTVCT_EL0/CNTFRQ_EL0 At present, when doing syscall_kvm test, we need to enable the function of ESR_ELx_SYS64_ISS_SYS_CNTVCT/ESR_ELx_SYS64_ISS_SYS_CNTFRQ to successfully pass the test. I set CNTKCTL_EL1.EL0VCTEN==1/CNTKCTL_EL1.EL0PCTEN==1, so that the related cases can passed. Signed-off-by: Bin Lu --- pkg/sentry/platform/ring0/entry_arm64.s | 9 +++++++++ 1 file changed, 9 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/ring0/entry_arm64.s b/pkg/sentry/platform/ring0/entry_arm64.s index 6ed73699b..9fd02d628 100644 --- a/pkg/sentry/platform/ring0/entry_arm64.s +++ b/pkg/sentry/platform/ring0/entry_arm64.s @@ -48,6 +48,12 @@ #define SCTLR_EL1_DEFAULT (SCTLR_M | SCTLR_C | SCTLR_I | SCTLR_UCT) +// cntkctl_el1: counter-timer kernel control register el1. +#define CNTKCTL_EL0PCTEN 1 << 0 +#define CNTKCTL_EL0VCTEN 1 << 1 + +#define CNTKCTL_EL1_DEFAULT (CNTKCTL_EL0PCTEN | CNTKCTL_EL0VCTEN) + // Saves a register set. // // This is a macro because it may need to executed in contents where a stack is @@ -509,6 +515,9 @@ TEXT ·Start(SB),NOSPLIT,$0 MOVD $SCTLR_EL1_DEFAULT, R1 MSR R1, SCTLR_EL1 + MOVD $CNTKCTL_EL1_DEFAULT, R1 + MSR R1, CNTKCTL_EL1 + MOVD R8, RSV_REG ORR $0xffff000000000000, RSV_REG, RSV_REG WORD $0xd518d092 //MSR R18, TPIDR_EL1 -- cgit v1.2.3 From 14087485bdf1f697ab7b5ece168408de2aab4624 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Thu, 18 Jun 2020 23:20:23 -0400 Subject: updated the functions to distinguish IA/DA for Arm64 We need to correctly distinguish instruction_abort/data_abort for mem_abort@Arm64. So, EC/WNR/FSC in esr_el1 should be checked. Signed-off-by: Bin Lu --- pkg/sentry/platform/kvm/kvm_const_arm64.go | 11 +++++ pkg/sentry/platform/kvm/machine_arm64.go | 72 +++++++++++++----------------- 2 files changed, 41 insertions(+), 42 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/kvm/kvm_const_arm64.go b/pkg/sentry/platform/kvm/kvm_const_arm64.go index 6f0539c29..fdc599477 100644 --- a/pkg/sentry/platform/kvm/kvm_const_arm64.go +++ b/pkg/sentry/platform/kvm/kvm_const_arm64.go @@ -116,6 +116,17 @@ const ( // Arm64: Exception Syndrome Register EL1. const ( + _ESR_ELx_EC_SHIFT = 26 + _ESR_ELx_EC_MASK = 0x3F << _ESR_ELx_EC_SHIFT + + _ESR_ELx_EC_IMP_DEF = 0x1f + _ESR_ELx_EC_IABT_LOW = 0x20 + _ESR_ELx_EC_IABT_CUR = 0x21 + _ESR_ELx_EC_PC_ALIGN = 0x22 + + _ESR_ELx_CM = 1 << 8 + _ESR_ELx_WNR = 1 << 6 + _ESR_ELx_FSC = 0x3F _ESR_SEGV_MAPERR_L0 = 0x4 diff --git a/pkg/sentry/platform/kvm/machine_arm64.go b/pkg/sentry/platform/kvm/machine_arm64.go index f3bf973de..9db171af9 100644 --- a/pkg/sentry/platform/kvm/machine_arm64.go +++ b/pkg/sentry/platform/kvm/machine_arm64.go @@ -125,71 +125,59 @@ func nonCanonical(addr uint64, signal int32, info *arch.SignalInfo) (usermem.Acc return usermem.NoAccess, platform.ErrContextSignal } +// isInstructionAbort returns true if it is an instruction abort. +// +//go:nosplit +func isInstructionAbort(code uint64) bool { + value := (code & _ESR_ELx_EC_MASK) >> _ESR_ELx_EC_SHIFT + return value == _ESR_ELx_EC_IABT_LOW +} + +// isWriteFault returns whether it is a write fault. +// +//go:nosplit +func isWriteFault(code uint64) bool { + if isInstructionAbort(code) { + return false + } + + return (code & _ESR_ELx_WNR) != 0 +} + // fault generates an appropriate fault return. // //go:nosplit func (c *vCPU) fault(signal int32, info *arch.SignalInfo) (usermem.AccessType, error) { + bluepill(c) // Probably no-op, but may not be. faultAddr := c.GetFaultAddr() code, user := c.ErrorCode() + if !user { + // The last fault serviced by this CPU was not a user + // fault, so we can't reliably trust the faultAddr or + // the code provided here. We need to re-execute. + return usermem.NoAccess, platform.ErrContextInterrupt + } + // Reset the pointed SignalInfo. *info = arch.SignalInfo{Signo: signal} info.SetAddr(uint64(faultAddr)) - read := true - write := false - execute := true - ret := code & _ESR_ELx_FSC switch ret { case _ESR_SEGV_MAPERR_L0, _ESR_SEGV_MAPERR_L1, _ESR_SEGV_MAPERR_L2, _ESR_SEGV_MAPERR_L3: info.Code = 1 //SEGV_MAPERR - read = false - write = true - execute = false case _ESR_SEGV_ACCERR_L1, _ESR_SEGV_ACCERR_L2, _ESR_SEGV_ACCERR_L3, _ESR_SEGV_PEMERR_L1, _ESR_SEGV_PEMERR_L2, _ESR_SEGV_PEMERR_L3: info.Code = 2 // SEGV_ACCERR. - read = true - write = false - execute = false default: info.Code = 2 } - if !user { - read = true - write = false - execute = true - - } accessType := usermem.AccessType{ - Read: read, - Write: write, - Execute: execute, + Read: !isWriteFault(uint64(code)), + Write: isWriteFault(uint64(code)), + Execute: isInstructionAbort(uint64(code)), } return accessType, platform.ErrContextSignal } - -// retryInGuest runs the given function in guest mode. -// -// If the function does not complete in guest mode (due to execution of a -// system call due to a GC stall, for example), then it will be retried. The -// given function must be idempotent as a result of the retry mechanism. -func (m *machine) retryInGuest(fn func()) { - c := m.Get() - defer m.Put(c) - for { - c.ClearErrorCode() // See below. - bluepill(c) // Force guest mode. - fn() // Execute the given function. - _, user := c.ErrorCode() - if user { - // If user is set, then we haven't bailed back to host - // mode via a kernel exception or system call. We - // consider the full function to have executed in guest - // mode and we can return. - break - } - } -} -- cgit v1.2.3 From f347a578b79c96c13ed492b2cf9aec1cb3e60f3f Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Mon, 27 Jul 2020 11:57:11 -0700 Subject: Move platform.File in memmap The subsequent systrap changes will need to import memmap from the platform package. PiperOrigin-RevId: 323409486 --- pkg/sentry/fs/fsutil/BUILD | 7 +--- pkg/sentry/fs/fsutil/dirty_set.go | 7 ++-- pkg/sentry/fs/fsutil/file_range_set.go | 15 ++++---- pkg/sentry/fs/fsutil/frame_ref_set.go | 10 ++--- pkg/sentry/fs/fsutil/host_file_mapper.go | 5 +-- pkg/sentry/fs/fsutil/host_mappable.go | 19 +++++---- pkg/sentry/fs/fsutil/inode_cached.go | 25 ++++++------ pkg/sentry/fsimpl/gofer/regular_file.go | 27 +++++++------ pkg/sentry/fsimpl/host/BUILD | 1 - pkg/sentry/fsimpl/host/mmap.go | 21 +++++----- pkg/sentry/kernel/shm/BUILD | 1 - pkg/sentry/kernel/shm/shm.go | 3 +- pkg/sentry/kernel/timekeeper.go | 4 +- pkg/sentry/kernel/vdso.go | 6 +-- pkg/sentry/memmap/BUILD | 14 ++++++- pkg/sentry/memmap/memmap.go | 60 +++++++++++++++++++++++++---- pkg/sentry/mm/BUILD | 4 +- pkg/sentry/mm/aio_context.go | 3 +- pkg/sentry/mm/mm.go | 10 ++--- pkg/sentry/mm/pma.go | 25 ++++++------ pkg/sentry/mm/special_mappable.go | 7 ++-- pkg/sentry/pgalloc/BUILD | 10 ++--- pkg/sentry/pgalloc/pgalloc.go | 66 ++++++++++++++++---------------- pkg/sentry/platform/BUILD | 20 +--------- pkg/sentry/platform/kvm/BUILD | 1 + pkg/sentry/platform/kvm/address_space.go | 3 +- pkg/sentry/platform/platform.go | 50 +----------------------- pkg/sentry/platform/ptrace/BUILD | 1 + pkg/sentry/platform/ptrace/subprocess.go | 3 +- 29 files changed, 205 insertions(+), 223 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/fsutil/BUILD b/pkg/sentry/fs/fsutil/BUILD index 789369220..5fb419bcd 100644 --- a/pkg/sentry/fs/fsutil/BUILD +++ b/pkg/sentry/fs/fsutil/BUILD @@ -8,7 +8,6 @@ go_template_instance( out = "dirty_set_impl.go", imports = { "memmap": "gvisor.dev/gvisor/pkg/sentry/memmap", - "platform": "gvisor.dev/gvisor/pkg/sentry/platform", }, package = "fsutil", prefix = "Dirty", @@ -25,14 +24,14 @@ go_template_instance( name = "frame_ref_set_impl", out = "frame_ref_set_impl.go", imports = { - "platform": "gvisor.dev/gvisor/pkg/sentry/platform", + "memmap": "gvisor.dev/gvisor/pkg/sentry/memmap", }, package = "fsutil", prefix = "FrameRef", template = "//pkg/segment:generic_set", types = { "Key": "uint64", - "Range": "platform.FileRange", + "Range": "memmap.FileRange", "Value": "uint64", "Functions": "FrameRefSetFunctions", }, @@ -43,7 +42,6 @@ go_template_instance( out = "file_range_set_impl.go", imports = { "memmap": "gvisor.dev/gvisor/pkg/sentry/memmap", - "platform": "gvisor.dev/gvisor/pkg/sentry/platform", }, package = "fsutil", prefix = "FileRange", @@ -86,7 +84,6 @@ go_library( "//pkg/sentry/kernel/time", "//pkg/sentry/memmap", "//pkg/sentry/pgalloc", - "//pkg/sentry/platform", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usage", "//pkg/state", diff --git a/pkg/sentry/fs/fsutil/dirty_set.go b/pkg/sentry/fs/fsutil/dirty_set.go index c6cd45087..2c9446c1d 100644 --- a/pkg/sentry/fs/fsutil/dirty_set.go +++ b/pkg/sentry/fs/fsutil/dirty_set.go @@ -20,7 +20,6 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/usermem" ) @@ -159,7 +158,7 @@ func (ds *DirtySet) AllowClean(mr memmap.MappableRange) { // repeatedly until all bytes have been written. max is the true size of the // cached object; offsets beyond max will not be passed to writeAt, even if // they are marked dirty. -func SyncDirty(ctx context.Context, mr memmap.MappableRange, cache *FileRangeSet, dirty *DirtySet, max uint64, mem platform.File, writeAt func(ctx context.Context, srcs safemem.BlockSeq, offset uint64) (uint64, error)) error { +func SyncDirty(ctx context.Context, mr memmap.MappableRange, cache *FileRangeSet, dirty *DirtySet, max uint64, mem memmap.File, writeAt func(ctx context.Context, srcs safemem.BlockSeq, offset uint64) (uint64, error)) error { var changedDirty bool defer func() { if changedDirty { @@ -194,7 +193,7 @@ func SyncDirty(ctx context.Context, mr memmap.MappableRange, cache *FileRangeSet // successful partial write, SyncDirtyAll will call it repeatedly until all // bytes have been written. max is the true size of the cached object; offsets // beyond max will not be passed to writeAt, even if they are marked dirty. -func SyncDirtyAll(ctx context.Context, cache *FileRangeSet, dirty *DirtySet, max uint64, mem platform.File, writeAt func(ctx context.Context, srcs safemem.BlockSeq, offset uint64) (uint64, error)) error { +func SyncDirtyAll(ctx context.Context, cache *FileRangeSet, dirty *DirtySet, max uint64, mem memmap.File, writeAt func(ctx context.Context, srcs safemem.BlockSeq, offset uint64) (uint64, error)) error { dseg := dirty.FirstSegment() for dseg.Ok() { if err := syncDirtyRange(ctx, dseg.Range(), cache, max, mem, writeAt); err != nil { @@ -210,7 +209,7 @@ func SyncDirtyAll(ctx context.Context, cache *FileRangeSet, dirty *DirtySet, max } // Preconditions: mr must be page-aligned. -func syncDirtyRange(ctx context.Context, mr memmap.MappableRange, cache *FileRangeSet, max uint64, mem platform.File, writeAt func(ctx context.Context, srcs safemem.BlockSeq, offset uint64) (uint64, error)) error { +func syncDirtyRange(ctx context.Context, mr memmap.MappableRange, cache *FileRangeSet, max uint64, mem memmap.File, writeAt func(ctx context.Context, srcs safemem.BlockSeq, offset uint64) (uint64, error)) error { for cseg := cache.LowerBoundSegment(mr.Start); cseg.Ok() && cseg.Start() < mr.End; cseg = cseg.NextSegment() { wbr := cseg.Range().Intersect(mr) if max < wbr.Start { diff --git a/pkg/sentry/fs/fsutil/file_range_set.go b/pkg/sentry/fs/fsutil/file_range_set.go index 5643cdac9..bbafebf03 100644 --- a/pkg/sentry/fs/fsutil/file_range_set.go +++ b/pkg/sentry/fs/fsutil/file_range_set.go @@ -23,13 +23,12 @@ import ( "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/usermem" ) // FileRangeSet maps offsets into a memmap.Mappable to offsets into a -// platform.File. It is used to implement Mappables that store data in +// memmap.File. It is used to implement Mappables that store data in // sparsely-allocated memory. // // type FileRangeSet @@ -65,20 +64,20 @@ func (FileRangeSetFunctions) Split(mr memmap.MappableRange, frstart uint64, spli } // FileRange returns the FileRange mapped by seg. -func (seg FileRangeIterator) FileRange() platform.FileRange { +func (seg FileRangeIterator) FileRange() memmap.FileRange { return seg.FileRangeOf(seg.Range()) } // FileRangeOf returns the FileRange mapped by mr. // // Preconditions: seg.Range().IsSupersetOf(mr). mr.Length() != 0. -func (seg FileRangeIterator) FileRangeOf(mr memmap.MappableRange) platform.FileRange { +func (seg FileRangeIterator) FileRangeOf(mr memmap.MappableRange) memmap.FileRange { frstart := seg.Value() + (mr.Start - seg.Start()) - return platform.FileRange{frstart, frstart + mr.Length()} + return memmap.FileRange{frstart, frstart + mr.Length()} } // Fill attempts to ensure that all memmap.Mappable offsets in required are -// mapped to a platform.File offset, by allocating from mf with the given +// mapped to a memmap.File offset, by allocating from mf with the given // memory usage kind and invoking readAt to store data into memory. (If readAt // returns a successful partial read, Fill will call it repeatedly until all // bytes have been read.) EOF is handled consistently with the requirements of @@ -141,7 +140,7 @@ func (frs *FileRangeSet) Fill(ctx context.Context, required, optional memmap.Map } // Drop removes segments for memmap.Mappable offsets in mr, freeing the -// corresponding platform.FileRanges. +// corresponding memmap.FileRanges. // // Preconditions: mr must be page-aligned. func (frs *FileRangeSet) Drop(mr memmap.MappableRange, mf *pgalloc.MemoryFile) { @@ -154,7 +153,7 @@ func (frs *FileRangeSet) Drop(mr memmap.MappableRange, mf *pgalloc.MemoryFile) { } // DropAll removes all segments in mr, freeing the corresponding -// platform.FileRanges. +// memmap.FileRanges. func (frs *FileRangeSet) DropAll(mf *pgalloc.MemoryFile) { for seg := frs.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { mf.DecRef(seg.FileRange()) diff --git a/pkg/sentry/fs/fsutil/frame_ref_set.go b/pkg/sentry/fs/fsutil/frame_ref_set.go index dd6f5aba6..a808894df 100644 --- a/pkg/sentry/fs/fsutil/frame_ref_set.go +++ b/pkg/sentry/fs/fsutil/frame_ref_set.go @@ -17,7 +17,7 @@ package fsutil import ( "math" - "gvisor.dev/gvisor/pkg/sentry/platform" + "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/usage" ) @@ -39,7 +39,7 @@ func (FrameRefSetFunctions) ClearValue(val *uint64) { } // Merge implements segment.Functions.Merge. -func (FrameRefSetFunctions) Merge(_ platform.FileRange, val1 uint64, _ platform.FileRange, val2 uint64) (uint64, bool) { +func (FrameRefSetFunctions) Merge(_ memmap.FileRange, val1 uint64, _ memmap.FileRange, val2 uint64) (uint64, bool) { if val1 != val2 { return 0, false } @@ -47,13 +47,13 @@ func (FrameRefSetFunctions) Merge(_ platform.FileRange, val1 uint64, _ platform. } // Split implements segment.Functions.Split. -func (FrameRefSetFunctions) Split(_ platform.FileRange, val uint64, _ uint64) (uint64, uint64) { +func (FrameRefSetFunctions) Split(_ memmap.FileRange, val uint64, _ uint64) (uint64, uint64) { return val, val } // IncRefAndAccount adds a reference on the range fr. All newly inserted segments // are accounted as host page cache memory mappings. -func (refs *FrameRefSet) IncRefAndAccount(fr platform.FileRange) { +func (refs *FrameRefSet) IncRefAndAccount(fr memmap.FileRange) { seg, gap := refs.Find(fr.Start) for { switch { @@ -74,7 +74,7 @@ func (refs *FrameRefSet) IncRefAndAccount(fr platform.FileRange) { // DecRefAndAccount removes a reference on the range fr and untracks segments // that are removed from memory accounting. -func (refs *FrameRefSet) DecRefAndAccount(fr platform.FileRange) { +func (refs *FrameRefSet) DecRefAndAccount(fr memmap.FileRange) { seg := refs.FindSegment(fr.Start) for seg.Ok() && seg.Start() < fr.End { diff --git a/pkg/sentry/fs/fsutil/host_file_mapper.go b/pkg/sentry/fs/fsutil/host_file_mapper.go index e82afd112..ef0113b52 100644 --- a/pkg/sentry/fs/fsutil/host_file_mapper.go +++ b/pkg/sentry/fs/fsutil/host_file_mapper.go @@ -21,7 +21,6 @@ import ( "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/usermem" ) @@ -126,7 +125,7 @@ func (f *HostFileMapper) DecRefOn(mr memmap.MappableRange) { // offsets in fr or until the next call to UnmapAll. // // Preconditions: The caller must hold a reference on all offsets in fr. -func (f *HostFileMapper) MapInternal(fr platform.FileRange, fd int, write bool) (safemem.BlockSeq, error) { +func (f *HostFileMapper) MapInternal(fr memmap.FileRange, fd int, write bool) (safemem.BlockSeq, error) { chunks := ((fr.End + chunkMask) >> chunkShift) - (fr.Start >> chunkShift) f.mapsMu.Lock() defer f.mapsMu.Unlock() @@ -146,7 +145,7 @@ func (f *HostFileMapper) MapInternal(fr platform.FileRange, fd int, write bool) } // Preconditions: f.mapsMu must be locked. -func (f *HostFileMapper) forEachMappingBlockLocked(fr platform.FileRange, fd int, write bool, fn func(safemem.Block)) error { +func (f *HostFileMapper) forEachMappingBlockLocked(fr memmap.FileRange, fd int, write bool, fn func(safemem.Block)) error { prot := syscall.PROT_READ if write { prot |= syscall.PROT_WRITE diff --git a/pkg/sentry/fs/fsutil/host_mappable.go b/pkg/sentry/fs/fsutil/host_mappable.go index 78fec553e..c15d8a946 100644 --- a/pkg/sentry/fs/fsutil/host_mappable.go +++ b/pkg/sentry/fs/fsutil/host_mappable.go @@ -21,18 +21,17 @@ import ( "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/usermem" ) -// HostMappable implements memmap.Mappable and platform.File over a +// HostMappable implements memmap.Mappable and memmap.File over a // CachedFileObject. // // Lock order (compare the lock order model in mm/mm.go): // truncateMu ("fs locks") // mu ("memmap.Mappable locks not taken by Translate") -// ("platform.File locks") +// ("memmap.File locks") // backingFile ("CachedFileObject locks") // // +stateify savable @@ -124,24 +123,24 @@ func (h *HostMappable) NotifyChangeFD() error { return nil } -// MapInternal implements platform.File.MapInternal. -func (h *HostMappable) MapInternal(fr platform.FileRange, at usermem.AccessType) (safemem.BlockSeq, error) { +// MapInternal implements memmap.File.MapInternal. +func (h *HostMappable) MapInternal(fr memmap.FileRange, at usermem.AccessType) (safemem.BlockSeq, error) { return h.hostFileMapper.MapInternal(fr, h.backingFile.FD(), at.Write) } -// FD implements platform.File.FD. +// FD implements memmap.File.FD. func (h *HostMappable) FD() int { return h.backingFile.FD() } -// IncRef implements platform.File.IncRef. -func (h *HostMappable) IncRef(fr platform.FileRange) { +// IncRef implements memmap.File.IncRef. +func (h *HostMappable) IncRef(fr memmap.FileRange) { mr := memmap.MappableRange{Start: fr.Start, End: fr.End} h.hostFileMapper.IncRefOn(mr) } -// DecRef implements platform.File.DecRef. -func (h *HostMappable) DecRef(fr platform.FileRange) { +// DecRef implements memmap.File.DecRef. +func (h *HostMappable) DecRef(fr memmap.FileRange) { mr := memmap.MappableRange{Start: fr.Start, End: fr.End} h.hostFileMapper.DecRefOn(mr) } diff --git a/pkg/sentry/fs/fsutil/inode_cached.go b/pkg/sentry/fs/fsutil/inode_cached.go index 800c8b4e1..fe8b0b6ac 100644 --- a/pkg/sentry/fs/fsutil/inode_cached.go +++ b/pkg/sentry/fs/fsutil/inode_cached.go @@ -26,7 +26,6 @@ import ( ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/usermem" @@ -934,7 +933,7 @@ func maxFillRange(required, optional memmap.MappableRange) memmap.MappableRange // InvalidateUnsavable implements memmap.Mappable.InvalidateUnsavable. func (c *CachingInodeOperations) InvalidateUnsavable(ctx context.Context) error { - // Whether we have a host fd (and consequently what platform.File is + // Whether we have a host fd (and consequently what memmap.File is // mapped) can change across save/restore, so invalidate all translations // unconditionally. c.mapsMu.Lock() @@ -999,10 +998,10 @@ func (c *CachingInodeOperations) Evict(ctx context.Context, er pgalloc.Evictable } } -// IncRef implements platform.File.IncRef. This is used when we directly map an -// underlying host fd and CachingInodeOperations is used as the platform.File +// IncRef implements memmap.File.IncRef. This is used when we directly map an +// underlying host fd and CachingInodeOperations is used as the memmap.File // during translation. -func (c *CachingInodeOperations) IncRef(fr platform.FileRange) { +func (c *CachingInodeOperations) IncRef(fr memmap.FileRange) { // Hot path. Avoid defers. c.dataMu.Lock() seg, gap := c.refs.Find(fr.Start) @@ -1024,10 +1023,10 @@ func (c *CachingInodeOperations) IncRef(fr platform.FileRange) { } } -// DecRef implements platform.File.DecRef. This is used when we directly map an -// underlying host fd and CachingInodeOperations is used as the platform.File +// DecRef implements memmap.File.DecRef. This is used when we directly map an +// underlying host fd and CachingInodeOperations is used as the memmap.File // during translation. -func (c *CachingInodeOperations) DecRef(fr platform.FileRange) { +func (c *CachingInodeOperations) DecRef(fr memmap.FileRange) { // Hot path. Avoid defers. c.dataMu.Lock() seg := c.refs.FindSegment(fr.Start) @@ -1046,15 +1045,15 @@ func (c *CachingInodeOperations) DecRef(fr platform.FileRange) { c.dataMu.Unlock() } -// MapInternal implements platform.File.MapInternal. This is used when we +// MapInternal implements memmap.File.MapInternal. This is used when we // directly map an underlying host fd and CachingInodeOperations is used as the -// platform.File during translation. -func (c *CachingInodeOperations) MapInternal(fr platform.FileRange, at usermem.AccessType) (safemem.BlockSeq, error) { +// memmap.File during translation. +func (c *CachingInodeOperations) MapInternal(fr memmap.FileRange, at usermem.AccessType) (safemem.BlockSeq, error) { return c.hostFileMapper.MapInternal(fr, c.backingFile.FD(), at.Write) } -// FD implements platform.File.FD. This is used when we directly map an -// underlying host fd and CachingInodeOperations is used as the platform.File +// FD implements memmap.File.FD. This is used when we directly map an +// underlying host fd and CachingInodeOperations is used as the memmap.File // during translation. func (c *CachingInodeOperations) FD() int { return c.backingFile.FD() diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go index 02317a133..09f142cfc 100644 --- a/pkg/sentry/fsimpl/gofer/regular_file.go +++ b/pkg/sentry/fsimpl/gofer/regular_file.go @@ -29,7 +29,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" @@ -221,12 +220,12 @@ func (fd *regularFileFD) pwriteLocked(ctx context.Context, src usermem.IOSequenc return 0, syserror.EINVAL } mr := memmap.MappableRange{pgstart, pgend} - var freed []platform.FileRange + var freed []memmap.FileRange d.dataMu.Lock() cseg := d.cache.LowerBoundSegment(mr.Start) for cseg.Ok() && cseg.Start() < mr.End { cseg = d.cache.Isolate(cseg, mr) - freed = append(freed, platform.FileRange{cseg.Value(), cseg.Value() + cseg.Range().Length()}) + freed = append(freed, memmap.FileRange{cseg.Value(), cseg.Value() + cseg.Range().Length()}) cseg = d.cache.Remove(cseg).NextSegment() } d.dataMu.Unlock() @@ -821,7 +820,7 @@ func maxFillRange(required, optional memmap.MappableRange) memmap.MappableRange // InvalidateUnsavable implements memmap.Mappable.InvalidateUnsavable. func (d *dentry) InvalidateUnsavable(ctx context.Context) error { - // Whether we have a host fd (and consequently what platform.File is + // Whether we have a host fd (and consequently what memmap.File is // mapped) can change across save/restore, so invalidate all translations // unconditionally. d.mapsMu.Lock() @@ -869,8 +868,8 @@ func (d *dentry) Evict(ctx context.Context, er pgalloc.EvictableRange) { } } -// dentryPlatformFile implements platform.File. It exists solely because dentry -// cannot implement both vfs.DentryImpl.IncRef and platform.File.IncRef. +// dentryPlatformFile implements memmap.File. It exists solely because dentry +// cannot implement both vfs.DentryImpl.IncRef and memmap.File.IncRef. // // dentryPlatformFile is only used when a host FD representing the remote file // is available (i.e. dentry.handle.fd >= 0), and that FD is used for @@ -878,7 +877,7 @@ func (d *dentry) Evict(ctx context.Context, er pgalloc.EvictableRange) { type dentryPlatformFile struct { *dentry - // fdRefs counts references on platform.File offsets. fdRefs is protected + // fdRefs counts references on memmap.File offsets. fdRefs is protected // by dentry.dataMu. fdRefs fsutil.FrameRefSet @@ -890,29 +889,29 @@ type dentryPlatformFile struct { hostFileMapperInitOnce sync.Once } -// IncRef implements platform.File.IncRef. -func (d *dentryPlatformFile) IncRef(fr platform.FileRange) { +// IncRef implements memmap.File.IncRef. +func (d *dentryPlatformFile) IncRef(fr memmap.FileRange) { d.dataMu.Lock() d.fdRefs.IncRefAndAccount(fr) d.dataMu.Unlock() } -// DecRef implements platform.File.DecRef. -func (d *dentryPlatformFile) DecRef(fr platform.FileRange) { +// DecRef implements memmap.File.DecRef. +func (d *dentryPlatformFile) DecRef(fr memmap.FileRange) { d.dataMu.Lock() d.fdRefs.DecRefAndAccount(fr) d.dataMu.Unlock() } -// MapInternal implements platform.File.MapInternal. -func (d *dentryPlatformFile) MapInternal(fr platform.FileRange, at usermem.AccessType) (safemem.BlockSeq, error) { +// MapInternal implements memmap.File.MapInternal. +func (d *dentryPlatformFile) MapInternal(fr memmap.FileRange, at usermem.AccessType) (safemem.BlockSeq, error) { d.handleMu.RLock() bs, err := d.hostFileMapper.MapInternal(fr, int(d.handle.fd), at.Write) d.handleMu.RUnlock() return bs, err } -// FD implements platform.File.FD. +// FD implements memmap.File.FD. func (d *dentryPlatformFile) FD() int { d.handleMu.RLock() fd := d.handle.fd diff --git a/pkg/sentry/fsimpl/host/BUILD b/pkg/sentry/fsimpl/host/BUILD index e86fbe2d5..bd701bbc7 100644 --- a/pkg/sentry/fsimpl/host/BUILD +++ b/pkg/sentry/fsimpl/host/BUILD @@ -34,7 +34,6 @@ go_library( "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/memmap", - "//pkg/sentry/platform", "//pkg/sentry/socket/control", "//pkg/sentry/socket/unix", "//pkg/sentry/socket/unix/transport", diff --git a/pkg/sentry/fsimpl/host/mmap.go b/pkg/sentry/fsimpl/host/mmap.go index 8545a82f0..65d3af38c 100644 --- a/pkg/sentry/fsimpl/host/mmap.go +++ b/pkg/sentry/fsimpl/host/mmap.go @@ -19,13 +19,12 @@ import ( "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/usermem" ) -// inodePlatformFile implements platform.File. It exists solely because inode -// cannot implement both kernfs.Inode.IncRef and platform.File.IncRef. +// inodePlatformFile implements memmap.File. It exists solely because inode +// cannot implement both kernfs.Inode.IncRef and memmap.File.IncRef. // // inodePlatformFile should only be used if inode.canMap is true. type inodePlatformFile struct { @@ -34,7 +33,7 @@ type inodePlatformFile struct { // fdRefsMu protects fdRefs. fdRefsMu sync.Mutex - // fdRefs counts references on platform.File offsets. It is used solely for + // fdRefs counts references on memmap.File offsets. It is used solely for // memory accounting. fdRefs fsutil.FrameRefSet @@ -45,32 +44,32 @@ type inodePlatformFile struct { fileMapperInitOnce sync.Once } -// IncRef implements platform.File.IncRef. +// IncRef implements memmap.File.IncRef. // // Precondition: i.inode.canMap must be true. -func (i *inodePlatformFile) IncRef(fr platform.FileRange) { +func (i *inodePlatformFile) IncRef(fr memmap.FileRange) { i.fdRefsMu.Lock() i.fdRefs.IncRefAndAccount(fr) i.fdRefsMu.Unlock() } -// DecRef implements platform.File.DecRef. +// DecRef implements memmap.File.DecRef. // // Precondition: i.inode.canMap must be true. -func (i *inodePlatformFile) DecRef(fr platform.FileRange) { +func (i *inodePlatformFile) DecRef(fr memmap.FileRange) { i.fdRefsMu.Lock() i.fdRefs.DecRefAndAccount(fr) i.fdRefsMu.Unlock() } -// MapInternal implements platform.File.MapInternal. +// MapInternal implements memmap.File.MapInternal. // // Precondition: i.inode.canMap must be true. -func (i *inodePlatformFile) MapInternal(fr platform.FileRange, at usermem.AccessType) (safemem.BlockSeq, error) { +func (i *inodePlatformFile) MapInternal(fr memmap.FileRange, at usermem.AccessType) (safemem.BlockSeq, error) { return i.fileMapper.MapInternal(fr, i.hostFD, at.Write) } -// FD implements platform.File.FD. +// FD implements memmap.File.FD. func (i *inodePlatformFile) FD() int { return i.hostFD } diff --git a/pkg/sentry/kernel/shm/BUILD b/pkg/sentry/kernel/shm/BUILD index bfd779837..c211fc8d0 100644 --- a/pkg/sentry/kernel/shm/BUILD +++ b/pkg/sentry/kernel/shm/BUILD @@ -20,7 +20,6 @@ go_library( "//pkg/sentry/kernel/time", "//pkg/sentry/memmap", "//pkg/sentry/pgalloc", - "//pkg/sentry/platform", "//pkg/sentry/usage", "//pkg/sync", "//pkg/syserror", diff --git a/pkg/sentry/kernel/shm/shm.go b/pkg/sentry/kernel/shm/shm.go index f66cfcc7f..55b4c2cdb 100644 --- a/pkg/sentry/kernel/shm/shm.go +++ b/pkg/sentry/kernel/shm/shm.go @@ -45,7 +45,6 @@ import ( ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" @@ -370,7 +369,7 @@ type Shm struct { // fr is the offset into mfp.MemoryFile() that backs this contents of this // segment. Immutable. - fr platform.FileRange + fr memmap.FileRange // mu protects all fields below. mu sync.Mutex `state:"nosave"` diff --git a/pkg/sentry/kernel/timekeeper.go b/pkg/sentry/kernel/timekeeper.go index 5f3908d8b..7c4fefb16 100644 --- a/pkg/sentry/kernel/timekeeper.go +++ b/pkg/sentry/kernel/timekeeper.go @@ -21,8 +21,8 @@ import ( "gvisor.dev/gvisor/pkg/log" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" + "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/platform" sentrytime "gvisor.dev/gvisor/pkg/sentry/time" "gvisor.dev/gvisor/pkg/sync" ) @@ -90,7 +90,7 @@ type Timekeeper struct { // NewTimekeeper does not take ownership of paramPage. // // SetClocks must be called on the returned Timekeeper before it is usable. -func NewTimekeeper(mfp pgalloc.MemoryFileProvider, paramPage platform.FileRange) (*Timekeeper, error) { +func NewTimekeeper(mfp pgalloc.MemoryFileProvider, paramPage memmap.FileRange) (*Timekeeper, error) { return &Timekeeper{ params: NewVDSOParamPage(mfp, paramPage), }, nil diff --git a/pkg/sentry/kernel/vdso.go b/pkg/sentry/kernel/vdso.go index f1b3c212c..290c32466 100644 --- a/pkg/sentry/kernel/vdso.go +++ b/pkg/sentry/kernel/vdso.go @@ -19,8 +19,8 @@ import ( "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/safemem" + "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/usermem" ) @@ -58,7 +58,7 @@ type vdsoParams struct { type VDSOParamPage struct { // The parameter page is fr, allocated from mfp.MemoryFile(). mfp pgalloc.MemoryFileProvider - fr platform.FileRange + fr memmap.FileRange // seq is the current sequence count written to the page. // @@ -81,7 +81,7 @@ type VDSOParamPage struct { // * VDSOParamPage must be the only writer to fr. // // * mfp.MemoryFile().MapInternal(fr) must return a single safemem.Block. -func NewVDSOParamPage(mfp pgalloc.MemoryFileProvider, fr platform.FileRange) *VDSOParamPage { +func NewVDSOParamPage(mfp pgalloc.MemoryFileProvider, fr memmap.FileRange) *VDSOParamPage { return &VDSOParamPage{mfp: mfp, fr: fr} } diff --git a/pkg/sentry/memmap/BUILD b/pkg/sentry/memmap/BUILD index a98b66de1..2c95669cd 100644 --- a/pkg/sentry/memmap/BUILD +++ b/pkg/sentry/memmap/BUILD @@ -28,9 +28,21 @@ go_template_instance( }, ) +go_template_instance( + name = "file_range", + out = "file_range.go", + package = "memmap", + prefix = "File", + template = "//pkg/segment:generic_range", + types = { + "T": "uint64", + }, +) + go_library( name = "memmap", srcs = [ + "file_range.go", "mappable_range.go", "mapping_set.go", "mapping_set_impl.go", @@ -40,7 +52,7 @@ go_library( deps = [ "//pkg/context", "//pkg/log", - "//pkg/sentry/platform", + "//pkg/safemem", "//pkg/syserror", "//pkg/usermem", ], diff --git a/pkg/sentry/memmap/memmap.go b/pkg/sentry/memmap/memmap.go index c6db9fc8f..c188f6c29 100644 --- a/pkg/sentry/memmap/memmap.go +++ b/pkg/sentry/memmap/memmap.go @@ -19,12 +19,12 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/platform" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/usermem" ) // Mappable represents a memory-mappable object, a mutable mapping from uint64 -// offsets to (platform.File, uint64 File offset) pairs. +// offsets to (File, uint64 File offset) pairs. // // See mm/mm.go for Mappable's place in the lock order. // @@ -74,7 +74,7 @@ type Mappable interface { // Translations are valid until invalidated by a callback to // MappingSpace.Invalidate or until the caller removes its mapping of the // translated range. Mappable implementations must ensure that at least one - // reference is held on all pages in a platform.File that may be the result + // reference is held on all pages in a File that may be the result // of a valid Translation. // // Preconditions: required.Length() > 0. optional.IsSupersetOf(required). @@ -100,7 +100,7 @@ type Translation struct { Source MappableRange // File is the mapped file. - File platform.File + File File // Offset is the offset into File at which this Translation begins. Offset uint64 @@ -110,9 +110,9 @@ type Translation struct { Perms usermem.AccessType } -// FileRange returns the platform.FileRange represented by t. -func (t Translation) FileRange() platform.FileRange { - return platform.FileRange{t.Offset, t.Offset + t.Source.Length()} +// FileRange returns the FileRange represented by t. +func (t Translation) FileRange() FileRange { + return FileRange{t.Offset, t.Offset + t.Source.Length()} } // CheckTranslateResult returns an error if (ts, terr) does not satisfy all @@ -361,3 +361,49 @@ type MMapOpts struct { // TODO(jamieliu): Replace entirely with MappingIdentity? Hint string } + +// File represents a host file that may be mapped into an platform.AddressSpace. +type File interface { + // All pages in a File are reference-counted. + + // IncRef increments the reference count on all pages in fr. + // + // Preconditions: fr.Start and fr.End must be page-aligned. fr.Length() > + // 0. At least one reference must be held on all pages in fr. (The File + // interface does not provide a way to acquire an initial reference; + // implementors may define mechanisms for doing so.) + IncRef(fr FileRange) + + // DecRef decrements the reference count on all pages in fr. + // + // Preconditions: fr.Start and fr.End must be page-aligned. fr.Length() > + // 0. At least one reference must be held on all pages in fr. + DecRef(fr FileRange) + + // MapInternal returns a mapping of the given file offsets in the invoking + // process' address space for reading and writing. + // + // Note that fr.Start and fr.End need not be page-aligned. + // + // Preconditions: fr.Length() > 0. At least one reference must be held on + // all pages in fr. + // + // Postconditions: The returned mapping is valid as long as at least one + // reference is held on the mapped pages. + MapInternal(fr FileRange, at usermem.AccessType) (safemem.BlockSeq, error) + + // FD returns the file descriptor represented by the File. + // + // The only permitted operation on the returned file descriptor is to map + // pages from it consistent with the requirements of AddressSpace.MapFile. + FD() int +} + +// FileRange represents a range of uint64 offsets into a File. +// +// type FileRange + +// String implements fmt.Stringer.String. +func (fr FileRange) String() string { + return fmt.Sprintf("[%#x, %#x)", fr.Start, fr.End) +} diff --git a/pkg/sentry/mm/BUILD b/pkg/sentry/mm/BUILD index a036ce53c..f9d0837a1 100644 --- a/pkg/sentry/mm/BUILD +++ b/pkg/sentry/mm/BUILD @@ -7,14 +7,14 @@ go_template_instance( name = "file_refcount_set", out = "file_refcount_set.go", imports = { - "platform": "gvisor.dev/gvisor/pkg/sentry/platform", + "memmap": "gvisor.dev/gvisor/pkg/sentry/memmap", }, package = "mm", prefix = "fileRefcount", template = "//pkg/segment:generic_set", types = { "Key": "uint64", - "Range": "platform.FileRange", + "Range": "memmap.FileRange", "Value": "int32", "Functions": "fileRefcountSetFunctions", }, diff --git a/pkg/sentry/mm/aio_context.go b/pkg/sentry/mm/aio_context.go index 379148903..1999ec706 100644 --- a/pkg/sentry/mm/aio_context.go +++ b/pkg/sentry/mm/aio_context.go @@ -20,7 +20,6 @@ import ( "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" @@ -243,7 +242,7 @@ type aioMappable struct { refs.AtomicRefCount mfp pgalloc.MemoryFileProvider - fr platform.FileRange + fr memmap.FileRange } var aioRingBufferSize = uint64(usermem.Addr(linux.AIORingSize).MustRoundUp()) diff --git a/pkg/sentry/mm/mm.go b/pkg/sentry/mm/mm.go index 6db7c3d40..3e85964e4 100644 --- a/pkg/sentry/mm/mm.go +++ b/pkg/sentry/mm/mm.go @@ -25,7 +25,7 @@ // Locks taken by memmap.Mappable.Translate // mm.privateRefs.mu // platform.AddressSpace locks -// platform.File locks +// memmap.File locks // mm.aioManager.mu // mm.AIOContext.mu // @@ -396,7 +396,7 @@ type pma struct { // file is the file mapped by this pma. Only pmas for which file == // MemoryManager.mfp.MemoryFile() may be saved. pmas hold a reference to // the corresponding file range while they exist. - file platform.File `state:"nosave"` + file memmap.File `state:"nosave"` // off is the offset into file at which this pma begins. // @@ -436,7 +436,7 @@ type pma struct { private bool // If internalMappings is not empty, it is the cached return value of - // file.MapInternal for the platform.FileRange mapped by this pma. + // file.MapInternal for the memmap.FileRange mapped by this pma. internalMappings safemem.BlockSeq `state:"nosave"` } @@ -469,10 +469,10 @@ func (fileRefcountSetFunctions) MaxKey() uint64 { func (fileRefcountSetFunctions) ClearValue(_ *int32) { } -func (fileRefcountSetFunctions) Merge(_ platform.FileRange, rc1 int32, _ platform.FileRange, rc2 int32) (int32, bool) { +func (fileRefcountSetFunctions) Merge(_ memmap.FileRange, rc1 int32, _ memmap.FileRange, rc2 int32) (int32, bool) { return rc1, rc1 == rc2 } -func (fileRefcountSetFunctions) Split(_ platform.FileRange, rc int32, _ uint64) (int32, int32) { +func (fileRefcountSetFunctions) Split(_ memmap.FileRange, rc int32, _ uint64) (int32, int32) { return rc, rc } diff --git a/pkg/sentry/mm/pma.go b/pkg/sentry/mm/pma.go index 62e4c20af..930ec895f 100644 --- a/pkg/sentry/mm/pma.go +++ b/pkg/sentry/mm/pma.go @@ -21,7 +21,6 @@ import ( "gvisor.dev/gvisor/pkg/safecopy" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -604,7 +603,7 @@ func (mm *MemoryManager) invalidateLocked(ar usermem.AddrRange, invalidatePrivat } } -// Pin returns the platform.File ranges currently mapped by addresses in ar in +// Pin returns the memmap.File ranges currently mapped by addresses in ar in // mm, acquiring a reference on the returned ranges which the caller must // release by calling Unpin. If not all addresses are mapped, Pin returns a // non-nil error. Note that Pin may return both a non-empty slice of @@ -674,15 +673,15 @@ type PinnedRange struct { Source usermem.AddrRange // File is the mapped file. - File platform.File + File memmap.File // Offset is the offset into File at which this PinnedRange begins. Offset uint64 } -// FileRange returns the platform.File offsets mapped by pr. -func (pr PinnedRange) FileRange() platform.FileRange { - return platform.FileRange{pr.Offset, pr.Offset + uint64(pr.Source.Length())} +// FileRange returns the memmap.File offsets mapped by pr. +func (pr PinnedRange) FileRange() memmap.FileRange { + return memmap.FileRange{pr.Offset, pr.Offset + uint64(pr.Source.Length())} } // Unpin releases the reference held by prs. @@ -857,7 +856,7 @@ func (mm *MemoryManager) vecInternalMappingsLocked(ars usermem.AddrRangeSeq) saf } // incPrivateRef acquires a reference on private pages in fr. -func (mm *MemoryManager) incPrivateRef(fr platform.FileRange) { +func (mm *MemoryManager) incPrivateRef(fr memmap.FileRange) { mm.privateRefs.mu.Lock() defer mm.privateRefs.mu.Unlock() refSet := &mm.privateRefs.refs @@ -878,8 +877,8 @@ func (mm *MemoryManager) incPrivateRef(fr platform.FileRange) { } // decPrivateRef releases a reference on private pages in fr. -func (mm *MemoryManager) decPrivateRef(fr platform.FileRange) { - var freed []platform.FileRange +func (mm *MemoryManager) decPrivateRef(fr memmap.FileRange) { + var freed []memmap.FileRange mm.privateRefs.mu.Lock() refSet := &mm.privateRefs.refs @@ -951,7 +950,7 @@ func (pmaSetFunctions) Merge(ar1 usermem.AddrRange, pma1 pma, ar2 usermem.AddrRa // Discard internal mappings instead of trying to merge them, since merging // them requires an allocation and getting them again from the - // platform.File might not. + // memmap.File might not. pma1.internalMappings = safemem.BlockSeq{} return pma1, true } @@ -1012,12 +1011,12 @@ func (pseg pmaIterator) getInternalMappingsLocked() error { return nil } -func (pseg pmaIterator) fileRange() platform.FileRange { +func (pseg pmaIterator) fileRange() memmap.FileRange { return pseg.fileRangeOf(pseg.Range()) } // Preconditions: pseg.Range().IsSupersetOf(ar). ar.Length != 0. -func (pseg pmaIterator) fileRangeOf(ar usermem.AddrRange) platform.FileRange { +func (pseg pmaIterator) fileRangeOf(ar usermem.AddrRange) memmap.FileRange { if checkInvariants { if !pseg.Ok() { panic("terminal pma iterator") @@ -1032,5 +1031,5 @@ func (pseg pmaIterator) fileRangeOf(ar usermem.AddrRange) platform.FileRange { pma := pseg.ValuePtr() pstart := pseg.Start() - return platform.FileRange{pma.off + uint64(ar.Start-pstart), pma.off + uint64(ar.End-pstart)} + return memmap.FileRange{pma.off + uint64(ar.Start-pstart), pma.off + uint64(ar.End-pstart)} } diff --git a/pkg/sentry/mm/special_mappable.go b/pkg/sentry/mm/special_mappable.go index 9ad52082d..0e142fb11 100644 --- a/pkg/sentry/mm/special_mappable.go +++ b/pkg/sentry/mm/special_mappable.go @@ -19,7 +19,6 @@ import ( "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -35,7 +34,7 @@ type SpecialMappable struct { refs.AtomicRefCount mfp pgalloc.MemoryFileProvider - fr platform.FileRange + fr memmap.FileRange name string } @@ -44,7 +43,7 @@ type SpecialMappable struct { // SpecialMappable will use the given name in /proc/[pid]/maps. // // Preconditions: fr.Length() != 0. -func NewSpecialMappable(name string, mfp pgalloc.MemoryFileProvider, fr platform.FileRange) *SpecialMappable { +func NewSpecialMappable(name string, mfp pgalloc.MemoryFileProvider, fr memmap.FileRange) *SpecialMappable { m := SpecialMappable{mfp: mfp, fr: fr, name: name} m.EnableLeakCheck("mm.SpecialMappable") return &m @@ -126,7 +125,7 @@ func (m *SpecialMappable) MemoryFileProvider() pgalloc.MemoryFileProvider { // FileRange returns the offsets into MemoryFileProvider().MemoryFile() that // store the SpecialMappable's contents. -func (m *SpecialMappable) FileRange() platform.FileRange { +func (m *SpecialMappable) FileRange() memmap.FileRange { return m.fr } diff --git a/pkg/sentry/pgalloc/BUILD b/pkg/sentry/pgalloc/BUILD index e1fcb175f..7a3311a70 100644 --- a/pkg/sentry/pgalloc/BUILD +++ b/pkg/sentry/pgalloc/BUILD @@ -36,14 +36,14 @@ go_template_instance( "trackGaps": "1", }, imports = { - "platform": "gvisor.dev/gvisor/pkg/sentry/platform", + "memmap": "gvisor.dev/gvisor/pkg/sentry/memmap", }, package = "pgalloc", prefix = "usage", template = "//pkg/segment:generic_set", types = { "Key": "uint64", - "Range": "platform.FileRange", + "Range": "memmap.FileRange", "Value": "usageInfo", "Functions": "usageSetFunctions", }, @@ -56,14 +56,14 @@ go_template_instance( "minDegree": "10", }, imports = { - "platform": "gvisor.dev/gvisor/pkg/sentry/platform", + "memmap": "gvisor.dev/gvisor/pkg/sentry/memmap", }, package = "pgalloc", prefix = "reclaim", template = "//pkg/segment:generic_set", types = { "Key": "uint64", - "Range": "platform.FileRange", + "Range": "memmap.FileRange", "Value": "reclaimSetValue", "Functions": "reclaimSetFunctions", }, @@ -89,7 +89,7 @@ go_library( "//pkg/safemem", "//pkg/sentry/arch", "//pkg/sentry/hostmm", - "//pkg/sentry/platform", + "//pkg/sentry/memmap", "//pkg/sentry/usage", "//pkg/state", "//pkg/state/wire", diff --git a/pkg/sentry/pgalloc/pgalloc.go b/pkg/sentry/pgalloc/pgalloc.go index afab97c0a..3243d7214 100644 --- a/pkg/sentry/pgalloc/pgalloc.go +++ b/pkg/sentry/pgalloc/pgalloc.go @@ -33,14 +33,14 @@ import ( "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/hostmm" - "gvisor.dev/gvisor/pkg/sentry/platform" + "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" ) -// MemoryFile is a platform.File whose pages may be allocated to arbitrary +// MemoryFile is a memmap.File whose pages may be allocated to arbitrary // users. type MemoryFile struct { // opts holds options passed to NewMemoryFile. opts is immutable. @@ -372,7 +372,7 @@ func (f *MemoryFile) Destroy() { // to Allocate. // // Preconditions: length must be page-aligned and non-zero. -func (f *MemoryFile) Allocate(length uint64, kind usage.MemoryKind) (platform.FileRange, error) { +func (f *MemoryFile) Allocate(length uint64, kind usage.MemoryKind) (memmap.FileRange, error) { if length == 0 || length%usermem.PageSize != 0 { panic(fmt.Sprintf("invalid allocation length: %#x", length)) } @@ -390,7 +390,7 @@ func (f *MemoryFile) Allocate(length uint64, kind usage.MemoryKind) (platform.Fi // Find a range in the underlying file. fr, ok := findAvailableRange(&f.usage, f.fileSize, length, alignment) if !ok { - return platform.FileRange{}, syserror.ENOMEM + return memmap.FileRange{}, syserror.ENOMEM } // Expand the file if needed. @@ -398,7 +398,7 @@ func (f *MemoryFile) Allocate(length uint64, kind usage.MemoryKind) (platform.Fi // Round the new file size up to be chunk-aligned. newFileSize := (int64(fr.End) + chunkMask) &^ chunkMask if err := f.file.Truncate(newFileSize); err != nil { - return platform.FileRange{}, err + return memmap.FileRange{}, err } f.fileSize = newFileSize f.mappingsMu.Lock() @@ -416,7 +416,7 @@ func (f *MemoryFile) Allocate(length uint64, kind usage.MemoryKind) (platform.Fi bs[i] = 0 } }); err != nil { - return platform.FileRange{}, err + return memmap.FileRange{}, err } } if !f.usage.Add(fr, usageInfo{ @@ -439,7 +439,7 @@ func (f *MemoryFile) Allocate(length uint64, kind usage.MemoryKind) (platform.Fi // space for mappings to be allocated downwards. // // Precondition: alignment must be a power of 2. -func findAvailableRange(usage *usageSet, fileSize int64, length, alignment uint64) (platform.FileRange, bool) { +func findAvailableRange(usage *usageSet, fileSize int64, length, alignment uint64) (memmap.FileRange, bool) { alignmentMask := alignment - 1 // Search for space in existing gaps, starting at the current end of the @@ -461,7 +461,7 @@ func findAvailableRange(usage *usageSet, fileSize int64, length, alignment uint6 break } if start := unalignedStart &^ alignmentMask; start >= gap.Start() { - return platform.FileRange{start, start + length}, true + return memmap.FileRange{start, start + length}, true } gap = gap.PrevLargeEnoughGap(length) @@ -475,7 +475,7 @@ func findAvailableRange(usage *usageSet, fileSize int64, length, alignment uint6 min = (min + alignmentMask) &^ alignmentMask if min+length < min { // Overflow: allocation would exceed the range of uint64. - return platform.FileRange{}, false + return memmap.FileRange{}, false } // Determine the minimum file size required to fit this allocation at its end. @@ -484,7 +484,7 @@ func findAvailableRange(usage *usageSet, fileSize int64, length, alignment uint6 if newFileSize <= fileSize { if fileSize != 0 { // Overflow: allocation would exceed the range of int64. - return platform.FileRange{}, false + return memmap.FileRange{}, false } newFileSize = chunkSize } @@ -496,7 +496,7 @@ func findAvailableRange(usage *usageSet, fileSize int64, length, alignment uint6 continue } if start := unalignedStart &^ alignmentMask; start >= min { - return platform.FileRange{start, start + length}, true + return memmap.FileRange{start, start + length}, true } } } @@ -508,22 +508,22 @@ func findAvailableRange(usage *usageSet, fileSize int64, length, alignment uint6 // by r.ReadToBlocks(), it returns that error. // // Preconditions: length > 0. length must be page-aligned. -func (f *MemoryFile) AllocateAndFill(length uint64, kind usage.MemoryKind, r safemem.Reader) (platform.FileRange, error) { +func (f *MemoryFile) AllocateAndFill(length uint64, kind usage.MemoryKind, r safemem.Reader) (memmap.FileRange, error) { fr, err := f.Allocate(length, kind) if err != nil { - return platform.FileRange{}, err + return memmap.FileRange{}, err } dsts, err := f.MapInternal(fr, usermem.Write) if err != nil { f.DecRef(fr) - return platform.FileRange{}, err + return memmap.FileRange{}, err } n, err := safemem.ReadFullToBlocks(r, dsts) un := uint64(usermem.Addr(n).RoundDown()) if un < length { // Free unused memory and update fr to contain only the memory that is // still allocated. - f.DecRef(platform.FileRange{fr.Start + un, fr.End}) + f.DecRef(memmap.FileRange{fr.Start + un, fr.End}) fr.End = fr.Start + un } return fr, err @@ -540,7 +540,7 @@ const ( // will read zeroes. // // Preconditions: fr.Length() > 0. -func (f *MemoryFile) Decommit(fr platform.FileRange) error { +func (f *MemoryFile) Decommit(fr memmap.FileRange) error { if !fr.WellFormed() || fr.Length() == 0 || fr.Start%usermem.PageSize != 0 || fr.End%usermem.PageSize != 0 { panic(fmt.Sprintf("invalid range: %v", fr)) } @@ -560,7 +560,7 @@ func (f *MemoryFile) Decommit(fr platform.FileRange) error { return nil } -func (f *MemoryFile) markDecommitted(fr platform.FileRange) { +func (f *MemoryFile) markDecommitted(fr memmap.FileRange) { f.mu.Lock() defer f.mu.Unlock() // Since we're changing the knownCommitted attribute, we need to merge @@ -581,8 +581,8 @@ func (f *MemoryFile) markDecommitted(fr platform.FileRange) { f.usage.MergeRange(fr) } -// IncRef implements platform.File.IncRef. -func (f *MemoryFile) IncRef(fr platform.FileRange) { +// IncRef implements memmap.File.IncRef. +func (f *MemoryFile) IncRef(fr memmap.FileRange) { if !fr.WellFormed() || fr.Length() == 0 || fr.Start%usermem.PageSize != 0 || fr.End%usermem.PageSize != 0 { panic(fmt.Sprintf("invalid range: %v", fr)) } @@ -600,8 +600,8 @@ func (f *MemoryFile) IncRef(fr platform.FileRange) { f.usage.MergeAdjacent(fr) } -// DecRef implements platform.File.DecRef. -func (f *MemoryFile) DecRef(fr platform.FileRange) { +// DecRef implements memmap.File.DecRef. +func (f *MemoryFile) DecRef(fr memmap.FileRange) { if !fr.WellFormed() || fr.Length() == 0 || fr.Start%usermem.PageSize != 0 || fr.End%usermem.PageSize != 0 { panic(fmt.Sprintf("invalid range: %v", fr)) } @@ -637,8 +637,8 @@ func (f *MemoryFile) DecRef(fr platform.FileRange) { } } -// MapInternal implements platform.File.MapInternal. -func (f *MemoryFile) MapInternal(fr platform.FileRange, at usermem.AccessType) (safemem.BlockSeq, error) { +// MapInternal implements memmap.File.MapInternal. +func (f *MemoryFile) MapInternal(fr memmap.FileRange, at usermem.AccessType) (safemem.BlockSeq, error) { if !fr.WellFormed() || fr.Length() == 0 { panic(fmt.Sprintf("invalid range: %v", fr)) } @@ -664,7 +664,7 @@ func (f *MemoryFile) MapInternal(fr platform.FileRange, at usermem.AccessType) ( // forEachMappingSlice invokes fn on a sequence of byte slices that // collectively map all bytes in fr. -func (f *MemoryFile) forEachMappingSlice(fr platform.FileRange, fn func([]byte)) error { +func (f *MemoryFile) forEachMappingSlice(fr memmap.FileRange, fn func([]byte)) error { mappings := f.mappings.Load().([]uintptr) for chunkStart := fr.Start &^ chunkMask; chunkStart < fr.End; chunkStart += chunkSize { chunk := int(chunkStart >> chunkShift) @@ -944,7 +944,7 @@ func (f *MemoryFile) updateUsageLocked(currentUsage uint64, checkCommitted func( continue case !populated && populatedRun: // Finish the run by changing this segment. - runRange := platform.FileRange{ + runRange := memmap.FileRange{ Start: r.Start + uint64(populatedRunStart*usermem.PageSize), End: r.Start + uint64(i*usermem.PageSize), } @@ -1009,7 +1009,7 @@ func (f *MemoryFile) File() *os.File { return f.file } -// FD implements platform.File.FD. +// FD implements memmap.File.FD. func (f *MemoryFile) FD() int { return int(f.file.Fd()) } @@ -1090,13 +1090,13 @@ func (f *MemoryFile) runReclaim() { // // Note that there returned range will be removed from tracking. It // must be reclaimed (removed from f.usage) at this point. -func (f *MemoryFile) findReclaimable() (platform.FileRange, bool) { +func (f *MemoryFile) findReclaimable() (memmap.FileRange, bool) { f.mu.Lock() defer f.mu.Unlock() for { for { if f.destroyed { - return platform.FileRange{}, false + return memmap.FileRange{}, false } if f.reclaimable { break @@ -1120,7 +1120,7 @@ func (f *MemoryFile) findReclaimable() (platform.FileRange, bool) { } } -func (f *MemoryFile) markReclaimed(fr platform.FileRange) { +func (f *MemoryFile) markReclaimed(fr memmap.FileRange) { f.mu.Lock() defer f.mu.Unlock() seg := f.usage.FindSegment(fr.Start) @@ -1222,11 +1222,11 @@ func (usageSetFunctions) MaxKey() uint64 { func (usageSetFunctions) ClearValue(val *usageInfo) { } -func (usageSetFunctions) Merge(_ platform.FileRange, val1 usageInfo, _ platform.FileRange, val2 usageInfo) (usageInfo, bool) { +func (usageSetFunctions) Merge(_ memmap.FileRange, val1 usageInfo, _ memmap.FileRange, val2 usageInfo) (usageInfo, bool) { return val1, val1 == val2 } -func (usageSetFunctions) Split(_ platform.FileRange, val usageInfo, _ uint64) (usageInfo, usageInfo) { +func (usageSetFunctions) Split(_ memmap.FileRange, val usageInfo, _ uint64) (usageInfo, usageInfo) { return val, val } @@ -1270,10 +1270,10 @@ func (reclaimSetFunctions) MaxKey() uint64 { func (reclaimSetFunctions) ClearValue(val *reclaimSetValue) { } -func (reclaimSetFunctions) Merge(_ platform.FileRange, _ reclaimSetValue, _ platform.FileRange, _ reclaimSetValue) (reclaimSetValue, bool) { +func (reclaimSetFunctions) Merge(_ memmap.FileRange, _ reclaimSetValue, _ memmap.FileRange, _ reclaimSetValue) (reclaimSetValue, bool) { return reclaimSetValue{}, true } -func (reclaimSetFunctions) Split(_ platform.FileRange, _ reclaimSetValue, _ uint64) (reclaimSetValue, reclaimSetValue) { +func (reclaimSetFunctions) Split(_ memmap.FileRange, _ reclaimSetValue, _ uint64) (reclaimSetValue, reclaimSetValue) { return reclaimSetValue{}, reclaimSetValue{} } diff --git a/pkg/sentry/platform/BUILD b/pkg/sentry/platform/BUILD index 453241eca..209b28053 100644 --- a/pkg/sentry/platform/BUILD +++ b/pkg/sentry/platform/BUILD @@ -1,39 +1,21 @@ load("//tools:defs.bzl", "go_library") -load("//tools/go_generics:defs.bzl", "go_template_instance") package(licenses = ["notice"]) -go_template_instance( - name = "file_range", - out = "file_range.go", - package = "platform", - prefix = "File", - template = "//pkg/segment:generic_range", - types = { - "T": "uint64", - }, -) - go_library( name = "platform", srcs = [ "context.go", - "file_range.go", "mmap_min_addr.go", "platform.go", ], visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", - "//pkg/atomicbitops", "//pkg/context", - "//pkg/log", - "//pkg/safecopy", - "//pkg/safemem", "//pkg/seccomp", "//pkg/sentry/arch", - "//pkg/sentry/usage", - "//pkg/syserror", + "//pkg/sentry/memmap", "//pkg/usermem", ], ) diff --git a/pkg/sentry/platform/kvm/BUILD b/pkg/sentry/platform/kvm/BUILD index 10a10bfe2..b5d27a72a 100644 --- a/pkg/sentry/platform/kvm/BUILD +++ b/pkg/sentry/platform/kvm/BUILD @@ -47,6 +47,7 @@ go_library( "//pkg/safecopy", "//pkg/seccomp", "//pkg/sentry/arch", + "//pkg/sentry/memmap", "//pkg/sentry/platform", "//pkg/sentry/platform/interrupt", "//pkg/sentry/platform/ring0", diff --git a/pkg/sentry/platform/kvm/address_space.go b/pkg/sentry/platform/kvm/address_space.go index faf1d5e1c..98a3e539d 100644 --- a/pkg/sentry/platform/kvm/address_space.go +++ b/pkg/sentry/platform/kvm/address_space.go @@ -18,6 +18,7 @@ import ( "sync/atomic" "gvisor.dev/gvisor/pkg/atomicbitops" + "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" "gvisor.dev/gvisor/pkg/sync" @@ -150,7 +151,7 @@ func (as *addressSpace) mapLocked(addr usermem.Addr, m hostMapEntry, at usermem. } // MapFile implements platform.AddressSpace.MapFile. -func (as *addressSpace) MapFile(addr usermem.Addr, f platform.File, fr platform.FileRange, at usermem.AccessType, precommit bool) error { +func (as *addressSpace) MapFile(addr usermem.Addr, f memmap.File, fr memmap.FileRange, at usermem.AccessType, precommit bool) error { as.mu.Lock() defer as.mu.Unlock() diff --git a/pkg/sentry/platform/platform.go b/pkg/sentry/platform/platform.go index 171513f3f..4b13eec30 100644 --- a/pkg/sentry/platform/platform.go +++ b/pkg/sentry/platform/platform.go @@ -22,9 +22,9 @@ import ( "os" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/seccomp" "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/usermem" ) @@ -207,7 +207,7 @@ type AddressSpace interface { // Preconditions: addr and fr must be page-aligned. fr.Length() > 0. // at.Any() == true. At least one reference must be held on all pages in // fr, and must continue to be held as long as pages are mapped. - MapFile(addr usermem.Addr, f File, fr FileRange, at usermem.AccessType, precommit bool) error + MapFile(addr usermem.Addr, f memmap.File, fr memmap.FileRange, at usermem.AccessType, precommit bool) error // Unmap unmaps the given range. // @@ -310,52 +310,6 @@ func (f SegmentationFault) Error() string { return fmt.Sprintf("segmentation fault at %#x", f.Addr) } -// File represents a host file that may be mapped into an AddressSpace. -type File interface { - // All pages in a File are reference-counted. - - // IncRef increments the reference count on all pages in fr. - // - // Preconditions: fr.Start and fr.End must be page-aligned. fr.Length() > - // 0. At least one reference must be held on all pages in fr. (The File - // interface does not provide a way to acquire an initial reference; - // implementors may define mechanisms for doing so.) - IncRef(fr FileRange) - - // DecRef decrements the reference count on all pages in fr. - // - // Preconditions: fr.Start and fr.End must be page-aligned. fr.Length() > - // 0. At least one reference must be held on all pages in fr. - DecRef(fr FileRange) - - // MapInternal returns a mapping of the given file offsets in the invoking - // process' address space for reading and writing. - // - // Note that fr.Start and fr.End need not be page-aligned. - // - // Preconditions: fr.Length() > 0. At least one reference must be held on - // all pages in fr. - // - // Postconditions: The returned mapping is valid as long as at least one - // reference is held on the mapped pages. - MapInternal(fr FileRange, at usermem.AccessType) (safemem.BlockSeq, error) - - // FD returns the file descriptor represented by the File. - // - // The only permitted operation on the returned file descriptor is to map - // pages from it consistent with the requirements of AddressSpace.MapFile. - FD() int -} - -// FileRange represents a range of uint64 offsets into a File. -// -// type FileRange - -// String implements fmt.Stringer.String. -func (fr FileRange) String() string { - return fmt.Sprintf("[%#x, %#x)", fr.Start, fr.End) -} - // Requirements is used to specify platform specific requirements. type Requirements struct { // RequiresCurrentPIDNS indicates that the sandbox has to be started in the diff --git a/pkg/sentry/platform/ptrace/BUILD b/pkg/sentry/platform/ptrace/BUILD index 30402c2df..29fd23cc3 100644 --- a/pkg/sentry/platform/ptrace/BUILD +++ b/pkg/sentry/platform/ptrace/BUILD @@ -30,6 +30,7 @@ go_library( "//pkg/seccomp", "//pkg/sentry/arch", "//pkg/sentry/hostcpu", + "//pkg/sentry/memmap", "//pkg/sentry/platform", "//pkg/sentry/platform/interrupt", "//pkg/sync", diff --git a/pkg/sentry/platform/ptrace/subprocess.go b/pkg/sentry/platform/ptrace/subprocess.go index 2389423b0..c990f3454 100644 --- a/pkg/sentry/platform/ptrace/subprocess.go +++ b/pkg/sentry/platform/ptrace/subprocess.go @@ -24,6 +24,7 @@ import ( "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/procid" "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/usermem" @@ -616,7 +617,7 @@ func (s *subprocess) syscall(sysno uintptr, args ...arch.SyscallArgument) (uintp } // MapFile implements platform.AddressSpace.MapFile. -func (s *subprocess) MapFile(addr usermem.Addr, f platform.File, fr platform.FileRange, at usermem.AccessType, precommit bool) error { +func (s *subprocess) MapFile(addr usermem.Addr, f memmap.File, fr memmap.FileRange, at usermem.AccessType, precommit bool) error { var flags int if precommit { flags |= syscall.MAP_POPULATE -- cgit v1.2.3 From 112eb0c5b9e6d45b58470fb5536a9fd91fb4222b Mon Sep 17 00:00:00 2001 From: Ridwan Sharif Date: Mon, 27 Jul 2020 13:25:21 -0700 Subject: Add device implementation for /dev/fuse This PR adds the following: - [x] Marshall-able structs for fuse headers - [x] Data structures needed in /dev/fuse to communicate with the daemon server - [x] Implementation of the device interface - [x] Go unit tests This change adds the `/dev/fuse` implementation. `Connection` controls the communication between the server and the sentry. The FUSE server uses the `FileDescription` interface to interact with the Sentry. The Sentry implmenetation of fusefs, uses `Connection` and the Connection interface to interact with the Server. All communication messages are in the form of `go_marshal` backed structs defined in the ABI package. This change also adds some go unit tests that test (pretty basically) the interfaces and should be used as an example of an end to end FUSE operation. COPYBARA_INTEGRATE_REVIEW=https://github.com/google/gvisor/pull/3083 from ridwanmsharif:ridwanmsharif/fuse-device-impl 69aa2ce970004938fe9f918168dfe57636ab856e PiperOrigin-RevId: 323428180 --- pkg/abi/linux/BUILD | 1 + pkg/abi/linux/fuse.go | 143 ++++++++++++ pkg/sentry/fsimpl/fuse/BUILD | 41 +++- pkg/sentry/fsimpl/fuse/connection.go | 255 +++++++++++++++++++++ pkg/sentry/fsimpl/fuse/dev.go | 289 +++++++++++++++++++++-- pkg/sentry/fsimpl/fuse/dev_test.go | 429 +++++++++++++++++++++++++++++++++++ pkg/sentry/fsimpl/fuse/fusefs.go | 52 +++-- pkg/sentry/fsimpl/fuse/register.go | 42 ++++ 8 files changed, 1223 insertions(+), 29 deletions(-) create mode 100644 pkg/abi/linux/fuse.go create mode 100644 pkg/sentry/fsimpl/fuse/connection.go create mode 100644 pkg/sentry/fsimpl/fuse/dev_test.go create mode 100644 pkg/sentry/fsimpl/fuse/register.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/BUILD b/pkg/abi/linux/BUILD index a4bb62013..05ca5342f 100644 --- a/pkg/abi/linux/BUILD +++ b/pkg/abi/linux/BUILD @@ -29,6 +29,7 @@ go_library( "file_amd64.go", "file_arm64.go", "fs.go", + "fuse.go", "futex.go", "inotify.go", "ioctl.go", diff --git a/pkg/abi/linux/fuse.go b/pkg/abi/linux/fuse.go new file mode 100644 index 000000000..d3ebbccc4 --- /dev/null +++ b/pkg/abi/linux/fuse.go @@ -0,0 +1,143 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package linux + +// +marshal +type FUSEOpcode uint32 + +// +marshal +type FUSEOpID uint64 + +// Opcodes for FUSE operations. Analogous to the opcodes in include/linux/fuse.h. +const ( + FUSE_LOOKUP FUSEOpcode = 1 + FUSE_FORGET = 2 /* no reply */ + FUSE_GETATTR = 3 + FUSE_SETATTR = 4 + FUSE_READLINK = 5 + FUSE_SYMLINK = 6 + _ + FUSE_MKNOD = 8 + FUSE_MKDIR = 9 + FUSE_UNLINK = 10 + FUSE_RMDIR = 11 + FUSE_RENAME = 12 + FUSE_LINK = 13 + FUSE_OPEN = 14 + FUSE_READ = 15 + FUSE_WRITE = 16 + FUSE_STATFS = 17 + FUSE_RELEASE = 18 + _ + FUSE_FSYNC = 20 + FUSE_SETXATTR = 21 + FUSE_GETXATTR = 22 + FUSE_LISTXATTR = 23 + FUSE_REMOVEXATTR = 24 + FUSE_FLUSH = 25 + FUSE_INIT = 26 + FUSE_OPENDIR = 27 + FUSE_READDIR = 28 + FUSE_RELEASEDIR = 29 + FUSE_FSYNCDIR = 30 + FUSE_GETLK = 31 + FUSE_SETLK = 32 + FUSE_SETLKW = 33 + FUSE_ACCESS = 34 + FUSE_CREATE = 35 + FUSE_INTERRUPT = 36 + FUSE_BMAP = 37 + FUSE_DESTROY = 38 + FUSE_IOCTL = 39 + FUSE_POLL = 40 + FUSE_NOTIFY_REPLY = 41 + FUSE_BATCH_FORGET = 42 +) + +const ( + // FUSE_MIN_READ_BUFFER is the minimum size the read can be for any FUSE filesystem. + // This is the minimum size Linux supports. See linux.fuse.h. + FUSE_MIN_READ_BUFFER uint32 = 8192 +) + +// FUSEHeaderIn is the header read by the daemon with each request. +// +// +marshal +type FUSEHeaderIn struct { + // Len specifies the total length of the data, including this header. + Len uint32 + + // Opcode specifies the kind of operation of the request. + Opcode FUSEOpcode + + // Unique specifies the unique identifier for this request. + Unique FUSEOpID + + // NodeID is the ID of the filesystem object being operated on. + NodeID uint64 + + // UID is the UID of the requesting process. + UID uint32 + + // GID is the GID of the requesting process. + GID uint32 + + // PID is the PID of the requesting process. + PID uint32 + + _ uint32 +} + +// FUSEHeaderOut is the header written by the daemon when it processes +// a request and wants to send a reply (almost all operations require a +// reply; if they do not, this will be explicitly documented). +// +// +marshal +type FUSEHeaderOut struct { + // Len specifies the total length of the data, including this header. + Len uint32 + + // Error specifies the error that occurred (0 if none). + Error int32 + + // Unique specifies the unique identifier of the corresponding request. + Unique FUSEOpID +} + +// FUSEWriteIn is the header written by a daemon when it makes a +// write request to the FUSE filesystem. +// +// +marshal +type FUSEWriteIn struct { + // Fh specifies the file handle that is being written to. + Fh uint64 + + // Offset is the offset of the write. + Offset uint64 + + // Size is the size of data being written. + Size uint32 + + // WriteFlags is the flags used during the write. + WriteFlags uint32 + + // LockOwner is the ID of the lock owner. + LockOwner uint64 + + // Flags is the flags for the request. + Flags uint32 + + _ uint32 +} diff --git a/pkg/sentry/fsimpl/fuse/BUILD b/pkg/sentry/fsimpl/fuse/BUILD index 737007748..67649e811 100644 --- a/pkg/sentry/fsimpl/fuse/BUILD +++ b/pkg/sentry/fsimpl/fuse/BUILD @@ -1,12 +1,28 @@ -load("//tools:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") +load("//tools/go_generics:defs.bzl", "go_template_instance") licenses(["notice"]) +go_template_instance( + name = "request_list", + out = "request_list.go", + package = "fuse", + prefix = "request", + template = "//pkg/ilist:generic_list", + types = { + "Element": "*Request", + "Linker": "*Request", + }, +) + go_library( name = "fuse", srcs = [ + "connection.go", "dev.go", "fusefs.go", + "register.go", + "request_list.go", ], visibility = ["//pkg/sentry:internal"], deps = [ @@ -18,7 +34,30 @@ go_library( "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/vfs", + "//pkg/sync", + "//pkg/syserror", + "//pkg/usermem", + "//pkg/waiter", + "//tools/go_marshal/marshal", + "@org_golang_x_sys//unix:go_default_library", + ], +) + +go_test( + name = "dev_test", + size = "small", + srcs = ["dev_test.go"], + library = ":fuse", + deps = [ + "//pkg/abi/linux", + "//pkg/sentry/fsimpl/testutil", + "//pkg/sentry/fsimpl/tmpfs", + "//pkg/sentry/kernel", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/vfs", "//pkg/syserror", "//pkg/usermem", + "//pkg/waiter", + "//tools/go_marshal/marshal", ], ) diff --git a/pkg/sentry/fsimpl/fuse/connection.go b/pkg/sentry/fsimpl/fuse/connection.go new file mode 100644 index 000000000..f330da0bd --- /dev/null +++ b/pkg/sentry/fsimpl/fuse/connection.go @@ -0,0 +1,255 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package fuse + +import ( + "errors" + "fmt" + "syscall" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/waiter" + "gvisor.dev/gvisor/tools/go_marshal/marshal" +) + +// MaxActiveRequestsDefault is the default setting controlling the upper bound +// on the number of active requests at any given time. +const MaxActiveRequestsDefault = 10000 + +var ( + // Ordinary requests have even IDs, while interrupts IDs are odd. + InitReqBit uint64 = 1 + ReqIDStep uint64 = 2 +) + +// Request represents a FUSE operation request that hasn't been sent to the +// server yet. +// +// +stateify savable +type Request struct { + requestEntry + + id linux.FUSEOpID + hdr *linux.FUSEHeaderIn + data []byte +} + +// Response represents an actual response from the server, including the +// response payload. +// +// +stateify savable +type Response struct { + opcode linux.FUSEOpcode + hdr linux.FUSEHeaderOut + data []byte +} + +// Connection is the struct by which the sentry communicates with the FUSE server daemon. +type Connection struct { + fd *DeviceFD + + // MaxWrite is the daemon's maximum size of a write buffer. + // This is negotiated during FUSE_INIT. + MaxWrite uint32 +} + +// NewFUSEConnection creates a FUSE connection to fd +func NewFUSEConnection(_ context.Context, fd *vfs.FileDescription, maxInFlightRequests uint64) (*Connection, error) { + // Mark the device as ready so it can be used. /dev/fuse can only be used if the FD was used to + // mount a FUSE filesystem. + fuseFD := fd.Impl().(*DeviceFD) + fuseFD.mounted = true + + // Create the writeBuf for the header to be stored in. + hdrLen := uint32((*linux.FUSEHeaderOut)(nil).SizeBytes()) + fuseFD.writeBuf = make([]byte, hdrLen) + fuseFD.completions = make(map[linux.FUSEOpID]*futureResponse) + fuseFD.fullQueueCh = make(chan struct{}, maxInFlightRequests) + fuseFD.writeCursor = 0 + + return &Connection{ + fd: fuseFD, + }, nil +} + +// NewRequest creates a new request that can be sent to the FUSE server. +func (conn *Connection) NewRequest(creds *auth.Credentials, pid uint32, ino uint64, opcode linux.FUSEOpcode, payload marshal.Marshallable) (*Request, error) { + conn.fd.mu.Lock() + defer conn.fd.mu.Unlock() + conn.fd.nextOpID += linux.FUSEOpID(ReqIDStep) + + hdrLen := (*linux.FUSEHeaderIn)(nil).SizeBytes() + hdr := linux.FUSEHeaderIn{ + Len: uint32(hdrLen + payload.SizeBytes()), + Opcode: opcode, + Unique: conn.fd.nextOpID, + NodeID: ino, + UID: uint32(creds.EffectiveKUID), + GID: uint32(creds.EffectiveKGID), + PID: pid, + } + + buf := make([]byte, hdr.Len) + hdr.MarshalUnsafe(buf[:hdrLen]) + payload.MarshalUnsafe(buf[hdrLen:]) + + return &Request{ + id: hdr.Unique, + hdr: &hdr, + data: buf, + }, nil +} + +// Call makes a request to the server and blocks the invoking task until a +// server responds with a response. +// NOTE: If no task is provided then the Call will simply enqueue the request +// and return a nil response. No blocking will happen in this case. Instead, +// this is used to signify that the processing of this request will happen by +// the kernel.Task that writes the response. See FUSE_INIT for such an +// invocation. +func (conn *Connection) Call(t *kernel.Task, r *Request) (*Response, error) { + fut, err := conn.callFuture(t, r) + if err != nil { + return nil, err + } + + return fut.resolve(t) +} + +// Error returns the error of the FUSE call. +func (r *Response) Error() error { + errno := r.hdr.Error + if errno >= 0 { + return nil + } + + sysErrNo := syscall.Errno(-errno) + return error(sysErrNo) +} + +// UnmarshalPayload unmarshals the response data into m. +func (r *Response) UnmarshalPayload(m marshal.Marshallable) error { + hdrLen := r.hdr.SizeBytes() + haveDataLen := r.hdr.Len - uint32(hdrLen) + wantDataLen := uint32(m.SizeBytes()) + + if haveDataLen < wantDataLen { + return fmt.Errorf("payload too small. Minimum data lenth required: %d, but got data length %d", wantDataLen, haveDataLen) + } + + m.UnmarshalUnsafe(r.data[hdrLen:]) + return nil +} + +// callFuture makes a request to the server and returns a future response. +// Call resolve() when the response needs to be fulfilled. +func (conn *Connection) callFuture(t *kernel.Task, r *Request) (*futureResponse, error) { + conn.fd.mu.Lock() + defer conn.fd.mu.Unlock() + + // Is the queue full? + // + // We must busy wait here until the request can be queued. We don't + // block on the fd.fullQueueCh with a lock - so after being signalled, + // before we acquire the lock, it is possible that a barging task enters + // and queues a request. As a result, upon acquiring the lock we must + // again check if the room is available. + // + // This can potentially starve a request forever but this can only happen + // if there are always too many ongoing requests all the time. The + // supported maxActiveRequests setting should be really high to avoid this. + for conn.fd.numActiveRequests == conn.fd.fs.opts.maxActiveRequests { + if t == nil { + // Since there is no task that is waiting. We must error out. + return nil, errors.New("FUSE request queue full") + } + + log.Infof("Blocking request %v from being queued. Too many active requests: %v", + r.id, conn.fd.numActiveRequests) + conn.fd.mu.Unlock() + err := t.Block(conn.fd.fullQueueCh) + conn.fd.mu.Lock() + if err != nil { + return nil, err + } + } + + return conn.callFutureLocked(t, r) +} + +// callFutureLocked makes a request to the server and returns a future response. +func (conn *Connection) callFutureLocked(t *kernel.Task, r *Request) (*futureResponse, error) { + conn.fd.queue.PushBack(r) + conn.fd.numActiveRequests += 1 + fut := newFutureResponse(r.hdr.Opcode) + conn.fd.completions[r.id] = fut + + // Signal the readers that there is something to read. + conn.fd.waitQueue.Notify(waiter.EventIn) + + return fut, nil +} + +// futureResponse represents an in-flight request, that may or may not have +// completed yet. Convert it to a resolved Response by calling Resolve, but note +// that this may block. +// +// +stateify savable +type futureResponse struct { + opcode linux.FUSEOpcode + ch chan struct{} + hdr *linux.FUSEHeaderOut + data []byte +} + +// newFutureResponse creates a future response to a FUSE request. +func newFutureResponse(opcode linux.FUSEOpcode) *futureResponse { + return &futureResponse{ + opcode: opcode, + ch: make(chan struct{}), + } +} + +// resolve blocks the task until the server responds to its corresponding request, +// then returns a resolved response. +func (f *futureResponse) resolve(t *kernel.Task) (*Response, error) { + // If there is no Task associated with this request - then we don't try to resolve + // the response. Instead, the task writing the response (proxy to the server) will + // process the response on our behalf. + if t == nil { + log.Infof("fuse.Response.resolve: Not waiting on a response from server.") + return nil, nil + } + + if err := t.Block(f.ch); err != nil { + return nil, err + } + + return f.getResponse(), nil +} + +// getResponse creates a Response from the data the futureResponse has. +func (f *futureResponse) getResponse() *Response { + return &Response{ + opcode: f.opcode, + hdr: *f.hdr, + data: f.data, + } +} diff --git a/pkg/sentry/fsimpl/fuse/dev.go b/pkg/sentry/fsimpl/fuse/dev.go index c9e12a94f..f3443ac71 100644 --- a/pkg/sentry/fsimpl/fuse/dev.go +++ b/pkg/sentry/fsimpl/fuse/dev.go @@ -15,13 +15,17 @@ package fuse import ( + "syscall" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/devtmpfs" + "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" ) const fuseDevMinor = 229 @@ -54,9 +58,43 @@ type DeviceFD struct { // mounted specifies whether a FUSE filesystem was mounted using the DeviceFD. mounted bool - // TODO(gvisor.dev/issue/2987): Add all the data structures needed to enqueue - // and deque requests, control synchronization and establish communication - // between the FUSE kernel module and the /dev/fuse character device. + // nextOpID is used to create new requests. + nextOpID linux.FUSEOpID + + // queue is the list of requests that need to be processed by the FUSE server. + queue requestList + + // numActiveRequests is the number of requests made by the Sentry that has + // yet to be responded to. + numActiveRequests uint64 + + // completions is used to map a request to its response. A Writer will use this + // to notify the caller of a completed response. + completions map[linux.FUSEOpID]*futureResponse + + writeCursor uint32 + + // writeBuf is the memory buffer used to copy in the FUSE out header from + // userspace. + writeBuf []byte + + // writeCursorFR current FR being copied from server. + writeCursorFR *futureResponse + + // mu protects all the queues, maps, buffers and cursors and nextOpID. + mu sync.Mutex + + // waitQueue is used to notify interested parties when the device becomes + // readable or writable. + waitQueue waiter.Queue + + // fullQueueCh is a channel used to synchronize the readers with the writers. + // Writers (inbound requests to the filesystem) block if there are too many + // unprocessed in-flight requests. + fullQueueCh chan struct{} + + // fs is the FUSE filesystem that this FD is being used for. + fs *filesystem } // Release implements vfs.FileDescriptionImpl.Release. @@ -79,7 +117,75 @@ func (fd *DeviceFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.R return 0, syserror.EPERM } - return 0, syserror.ENOSYS + // We require that any Read done on this filesystem have a sane minimum + // read buffer. It must have the capacity for the fixed parts of any request + // header (Linux uses the request header and the FUSEWriteIn header for this + // calculation) + the negotiated MaxWrite room for the data. + minBuffSize := linux.FUSE_MIN_READ_BUFFER + inHdrLen := uint32((*linux.FUSEHeaderIn)(nil).SizeBytes()) + writeHdrLen := uint32((*linux.FUSEWriteIn)(nil).SizeBytes()) + negotiatedMinBuffSize := inHdrLen + writeHdrLen + fd.fs.conn.MaxWrite + if minBuffSize < negotiatedMinBuffSize { + minBuffSize = negotiatedMinBuffSize + } + + // If the read buffer is too small, error out. + if dst.NumBytes() < int64(minBuffSize) { + return 0, syserror.EINVAL + } + + fd.mu.Lock() + defer fd.mu.Unlock() + return fd.readLocked(ctx, dst, opts) +} + +// readLocked implements the reading of the fuse device while locked with DeviceFD.mu. +func (fd *DeviceFD) readLocked(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + if fd.queue.Empty() { + return 0, syserror.ErrWouldBlock + } + + var readCursor uint32 + var bytesRead int64 + for { + req := fd.queue.Front() + if dst.NumBytes() < int64(req.hdr.Len) { + // The request is too large. Cannot process it. All requests must be smaller than the + // negotiated size as specified by Connection.MaxWrite set as part of the FUSE_INIT + // handshake. + errno := -int32(syscall.EIO) + if req.hdr.Opcode == linux.FUSE_SETXATTR { + errno = -int32(syscall.E2BIG) + } + + // Return the error to the calling task. + if err := fd.sendError(ctx, errno, req); err != nil { + return 0, err + } + + // We're done with this request. + fd.queue.Remove(req) + + // Restart the read as this request was invalid. + log.Warningf("fuse.DeviceFD.Read: request found was too large. Restarting read.") + return fd.readLocked(ctx, dst, opts) + } + + n, err := dst.CopyOut(ctx, req.data[readCursor:]) + if err != nil { + return 0, err + } + readCursor += uint32(n) + bytesRead += int64(n) + + if readCursor >= req.hdr.Len { + // Fully done with this req, remove it from the queue. + fd.queue.Remove(req) + break + } + } + + return bytesRead, nil } // PWrite implements vfs.FileDescriptionImpl.PWrite. @@ -94,12 +200,128 @@ func (fd *DeviceFD) PWrite(ctx context.Context, src usermem.IOSequence, offset i // Write implements vfs.FileDescriptionImpl.Write. func (fd *DeviceFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + fd.mu.Lock() + defer fd.mu.Unlock() + return fd.writeLocked(ctx, src, opts) +} + +// writeLocked implements writing to the fuse device while locked with DeviceFD.mu. +func (fd *DeviceFD) writeLocked(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { // Operations on /dev/fuse don't make sense until a FUSE filesystem is mounted. if !fd.mounted { return 0, syserror.EPERM } - return 0, syserror.ENOSYS + var cn, n int64 + hdrLen := uint32((*linux.FUSEHeaderOut)(nil).SizeBytes()) + + for src.NumBytes() > 0 { + if fd.writeCursorFR != nil { + // Already have common header, and we're now copying the payload. + wantBytes := fd.writeCursorFR.hdr.Len + + // Note that the FR data doesn't have the header. Copy it over if its necessary. + if fd.writeCursorFR.data == nil { + fd.writeCursorFR.data = make([]byte, wantBytes) + } + + bytesCopied, err := src.CopyIn(ctx, fd.writeCursorFR.data[fd.writeCursor:wantBytes]) + if err != nil { + return 0, err + } + src = src.DropFirst(bytesCopied) + + cn = int64(bytesCopied) + n += cn + fd.writeCursor += uint32(cn) + if fd.writeCursor == wantBytes { + // Done reading this full response. Clean up and unblock the + // initiator. + break + } + + // Check if we have more data in src. + continue + } + + // Assert that the header isn't read into the writeBuf yet. + if fd.writeCursor >= hdrLen { + return 0, syserror.EINVAL + } + + // We don't have the full common response header yet. + wantBytes := hdrLen - fd.writeCursor + bytesCopied, err := src.CopyIn(ctx, fd.writeBuf[fd.writeCursor:wantBytes]) + if err != nil { + return 0, err + } + src = src.DropFirst(bytesCopied) + + cn = int64(bytesCopied) + n += cn + fd.writeCursor += uint32(cn) + if fd.writeCursor == hdrLen { + // Have full header in the writeBuf. Use it to fetch the actual futureResponse + // from the device's completions map. + var hdr linux.FUSEHeaderOut + hdr.UnmarshalBytes(fd.writeBuf) + + // We have the header now and so the writeBuf has served its purpose. + // We could reset it manually here but instead of doing that, at the + // end of the write, the writeCursor will be set to 0 thereby allowing + // the next request to overwrite whats in the buffer, + + fut, ok := fd.completions[hdr.Unique] + if !ok { + // Server sent us a response for a request we never sent? + return 0, syserror.EINVAL + } + + delete(fd.completions, hdr.Unique) + + // Copy over the header into the future response. The rest of the payload + // will be copied over to the FR's data in the next iteration. + fut.hdr = &hdr + fd.writeCursorFR = fut + + // Next iteration will now try read the complete request, if src has + // any data remaining. Otherwise we're done. + } + } + + if fd.writeCursorFR != nil { + if err := fd.sendResponse(ctx, fd.writeCursorFR); err != nil { + return 0, err + } + + // Ready the device for the next request. + fd.writeCursorFR = nil + fd.writeCursor = 0 + } + + return n, nil +} + +// Readiness implements vfs.FileDescriptionImpl.Readiness. +func (fd *DeviceFD) Readiness(mask waiter.EventMask) waiter.EventMask { + var ready waiter.EventMask + ready |= waiter.EventOut // FD is always writable + if !fd.queue.Empty() { + // Have reqs available, FD is readable. + ready |= waiter.EventIn + } + + return ready & mask +} + +// EventRegister implements waiter.Waitable.EventRegister. +func (fd *DeviceFD) EventRegister(e *waiter.Entry, mask waiter.EventMask) { + fd.waitQueue.EventRegister(e, mask) +} + +// EventUnregister implements waiter.Waitable.EventUnregister. +func (fd *DeviceFD) EventUnregister(e *waiter.Entry) { + fd.waitQueue.EventUnregister(e) } // Seek implements vfs.FileDescriptionImpl.Seek. @@ -112,22 +334,61 @@ func (fd *DeviceFD) Seek(ctx context.Context, offset int64, whence int32) (int64 return 0, syserror.ENOSYS } -// Register registers the FUSE device with vfsObj. -func Register(vfsObj *vfs.VirtualFilesystem) error { - if err := vfsObj.RegisterDevice(vfs.CharDevice, linux.MISC_MAJOR, fuseDevMinor, fuseDevice{}, &vfs.RegisterDeviceOptions{ - GroupName: "misc", - }); err != nil { +// sendResponse sends a response to the waiting task (if any). +func (fd *DeviceFD) sendResponse(ctx context.Context, fut *futureResponse) error { + // See if the running task need to perform some action before returning. + // Since we just finished writing the future, we can be sure that + // getResponse generates a populated response. + if err := fd.noReceiverAction(ctx, fut.getResponse()); err != nil { return err } + // Signal that the queue is no longer full. + select { + case fd.fullQueueCh <- struct{}{}: + default: + } + fd.numActiveRequests -= 1 + + // Signal the task waiting on a response. + close(fut.ch) return nil } -// CreateDevtmpfsFile creates a device special file in devtmpfs. -func CreateDevtmpfsFile(ctx context.Context, dev *devtmpfs.Accessor) error { - if err := dev.CreateDeviceFile(ctx, "fuse", vfs.CharDevice, linux.MISC_MAJOR, fuseDevMinor, 0666 /* mode */); err != nil { +// sendError sends an error response to the waiting task (if any). +func (fd *DeviceFD) sendError(ctx context.Context, errno int32, req *Request) error { + // Return the error to the calling task. + outHdrLen := uint32((*linux.FUSEHeaderOut)(nil).SizeBytes()) + respHdr := linux.FUSEHeaderOut{ + Len: outHdrLen, + Error: errno, + Unique: req.hdr.Unique, + } + + fut, ok := fd.completions[respHdr.Unique] + if !ok { + // Server sent us a response for a request we never sent? + return syserror.EINVAL + } + delete(fd.completions, respHdr.Unique) + + fut.hdr = &respHdr + if err := fd.sendResponse(ctx, fut); err != nil { return err } return nil } + +// noReceiverAction has the calling kernel.Task do some action if its known that no +// receiver is going to be waiting on the future channel. This is to be used by: +// FUSE_INIT. +func (fd *DeviceFD) noReceiverAction(ctx context.Context, r *Response) error { + if r.opcode == linux.FUSE_INIT { + // TODO: process init response here. + // Maybe get the creds from the context? + // creds := auth.CredentialsFromContext(ctx) + } + + return nil +} diff --git a/pkg/sentry/fsimpl/fuse/dev_test.go b/pkg/sentry/fsimpl/fuse/dev_test.go new file mode 100644 index 000000000..fcd77832a --- /dev/null +++ b/pkg/sentry/fsimpl/fuse/dev_test.go @@ -0,0 +1,429 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package fuse + +import ( + "fmt" + "io" + "math/rand" + "testing" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/testutil" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/waiter" + "gvisor.dev/gvisor/tools/go_marshal/marshal" +) + +// echoTestOpcode is the Opcode used during testing. The server used in tests +// will simply echo the payload back with the appropriate headers. +const echoTestOpcode linux.FUSEOpcode = 1000 + +type testPayload struct { + data uint32 +} + +// TestFUSECommunication tests that the communication layer between the Sentry and the +// FUSE server daemon works as expected. +func TestFUSECommunication(t *testing.T) { + s := setup(t) + defer s.Destroy() + + k := kernel.KernelFromContext(s.Ctx) + creds := auth.CredentialsFromContext(s.Ctx) + + // Create test cases with different number of concurrent clients and servers. + testCases := []struct { + Name string + NumClients int + NumServers int + MaxActiveRequests uint64 + }{ + { + Name: "SingleClientSingleServer", + NumClients: 1, + NumServers: 1, + MaxActiveRequests: MaxActiveRequestsDefault, + }, + { + Name: "SingleClientMultipleServers", + NumClients: 1, + NumServers: 10, + MaxActiveRequests: MaxActiveRequestsDefault, + }, + { + Name: "MultipleClientsSingleServer", + NumClients: 10, + NumServers: 1, + MaxActiveRequests: MaxActiveRequestsDefault, + }, + { + Name: "MultipleClientsMultipleServers", + NumClients: 10, + NumServers: 10, + MaxActiveRequests: MaxActiveRequestsDefault, + }, + { + Name: "RequestCapacityFull", + NumClients: 10, + NumServers: 1, + MaxActiveRequests: 1, + }, + { + Name: "RequestCapacityContinuouslyFull", + NumClients: 100, + NumServers: 2, + MaxActiveRequests: 2, + }, + } + + for _, testCase := range testCases { + t.Run(testCase.Name, func(t *testing.T) { + conn, fd, err := newTestConnection(s, k, testCase.MaxActiveRequests) + if err != nil { + t.Fatalf("newTestConnection: %v", err) + } + + clientsDone := make([]chan struct{}, testCase.NumClients) + serversDone := make([]chan struct{}, testCase.NumServers) + serversKill := make([]chan struct{}, testCase.NumServers) + + // FUSE clients. + for i := 0; i < testCase.NumClients; i++ { + clientsDone[i] = make(chan struct{}) + go func(i int) { + fuseClientRun(t, s, k, conn, creds, uint32(i), uint64(i), clientsDone[i]) + }(i) + } + + // FUSE servers. + for j := 0; j < testCase.NumServers; j++ { + serversDone[j] = make(chan struct{}) + serversKill[j] = make(chan struct{}, 1) // The kill command shouldn't block. + go func(j int) { + fuseServerRun(t, s, k, fd, serversDone[j], serversKill[j]) + }(j) + } + + // Tear down. + // + // Make sure all the clients are done. + for i := 0; i < testCase.NumClients; i++ { + <-clientsDone[i] + } + + // Kill any server that is potentially waiting. + for j := 0; j < testCase.NumServers; j++ { + serversKill[j] <- struct{}{} + } + + // Make sure all the servers are done. + for j := 0; j < testCase.NumServers; j++ { + <-serversDone[j] + } + }) + } +} + +// CallTest makes a request to the server and blocks the invoking +// goroutine until a server responds with a response. Doesn't block +// a kernel.Task. Analogous to Connection.Call but used for testing. +func CallTest(conn *Connection, t *kernel.Task, r *Request, i uint32) (*Response, error) { + conn.fd.mu.Lock() + + // Wait until we're certain that a new request can be processed. + for conn.fd.numActiveRequests == conn.fd.fs.opts.maxActiveRequests { + conn.fd.mu.Unlock() + select { + case <-conn.fd.fullQueueCh: + } + conn.fd.mu.Lock() + } + + fut, err := conn.callFutureLocked(t, r) // No task given. + conn.fd.mu.Unlock() + + if err != nil { + return nil, err + } + + // Resolve the response. + // + // Block without a task. + select { + case <-fut.ch: + } + + // A response is ready. Resolve and return it. + return fut.getResponse(), nil +} + +// ReadTest is analogous to vfs.FileDescription.Read and reads from the FUSE +// device. However, it does so by - not blocking the task that is calling - and +// instead just waits on a channel. The behaviour is essentially the same as +// DeviceFD.Read except it guarantees that the task is not blocked. +func ReadTest(serverTask *kernel.Task, fd *vfs.FileDescription, inIOseq usermem.IOSequence, killServer chan struct{}) (int64, bool, error) { + var err error + var n, total int64 + + dev := fd.Impl().(*DeviceFD) + + // Register for notifications. + w, ch := waiter.NewChannelEntry(nil) + dev.EventRegister(&w, waiter.EventIn) + for { + // Issue the request and break out if it completes with anything other than + // "would block". + n, err = dev.Read(serverTask, inIOseq, vfs.ReadOptions{}) + total += n + if err != syserror.ErrWouldBlock { + break + } + + // Wait for a notification that we should retry. + // Emulate the blocking for when no requests are available + select { + case <-ch: + case <-killServer: + // Server killed by the main program. + return 0, true, nil + } + } + + dev.EventUnregister(&w) + return total, false, err +} + +// fuseClientRun emulates all the actions of a normal FUSE request. It creates +// a header, a payload, calls the server, waits for the response, and processes +// the response. +func fuseClientRun(t *testing.T, s *testutil.System, k *kernel.Kernel, conn *Connection, creds *auth.Credentials, pid uint32, inode uint64, clientDone chan struct{}) { + defer func() { clientDone <- struct{}{} }() + + tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) + clientTask, err := testutil.CreateTask(s.Ctx, fmt.Sprintf("fuse-client-%v", pid), tc, s.MntNs, s.Root, s.Root) + if err != nil { + t.Fatal(err) + } + testObj := &testPayload{ + data: rand.Uint32(), + } + + req, err := conn.NewRequest(creds, pid, inode, echoTestOpcode, testObj) + if err != nil { + t.Fatalf("NewRequest creation failed: %v", err) + } + + // Queue up a request. + // Analogous to Call except it doesn't block on the task. + resp, err := CallTest(conn, clientTask, req, pid) + if err != nil { + t.Fatalf("CallTaskNonBlock failed: %v", err) + } + + if err = resp.Error(); err != nil { + t.Fatalf("Server responded with an error: %v", err) + } + + var respTestPayload testPayload + if err := resp.UnmarshalPayload(&respTestPayload); err != nil { + t.Fatalf("Unmarshalling payload error: %v", err) + } + + if resp.hdr.Unique != req.hdr.Unique { + t.Fatalf("got response for another request. Expected response for req %v but got response for req %v", + req.hdr.Unique, resp.hdr.Unique) + } + + if respTestPayload.data != testObj.data { + t.Fatalf("read incorrect data. Data expected: %v, but got %v", testObj.data, respTestPayload.data) + } + +} + +// fuseServerRun creates a task and emulates all the actions of a simple FUSE server +// that simply reads a request and echos the same struct back as a response using the +// appropriate headers. +func fuseServerRun(t *testing.T, s *testutil.System, k *kernel.Kernel, fd *vfs.FileDescription, serverDone, killServer chan struct{}) { + defer func() { serverDone <- struct{}{} }() + + // Create the tasks that the server will be using. + tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) + var readPayload testPayload + + serverTask, err := testutil.CreateTask(s.Ctx, "fuse-server", tc, s.MntNs, s.Root, s.Root) + if err != nil { + t.Fatal(err) + } + + // Read the request. + for { + inHdrLen := uint32((*linux.FUSEHeaderIn)(nil).SizeBytes()) + payloadLen := uint32(readPayload.SizeBytes()) + + // The raed buffer must meet some certain size criteria. + buffSize := inHdrLen + payloadLen + if buffSize < linux.FUSE_MIN_READ_BUFFER { + buffSize = linux.FUSE_MIN_READ_BUFFER + } + inBuf := make([]byte, buffSize) + inIOseq := usermem.BytesIOSequence(inBuf) + + n, serverKilled, err := ReadTest(serverTask, fd, inIOseq, killServer) + if err != nil { + t.Fatalf("Read failed :%v", err) + } + + // Server should shut down. No new requests are going to be made. + if serverKilled { + break + } + + if n <= 0 { + t.Fatalf("Read read no bytes") + } + + var readFUSEHeaderIn linux.FUSEHeaderIn + readFUSEHeaderIn.UnmarshalUnsafe(inBuf[:inHdrLen]) + readPayload.UnmarshalUnsafe(inBuf[inHdrLen : inHdrLen+payloadLen]) + + if readFUSEHeaderIn.Opcode != echoTestOpcode { + t.Fatalf("read incorrect data. Header: %v, Payload: %v", readFUSEHeaderIn, readPayload) + } + + // Write the response. + outHdrLen := uint32((*linux.FUSEHeaderOut)(nil).SizeBytes()) + outBuf := make([]byte, outHdrLen+payloadLen) + outHeader := linux.FUSEHeaderOut{ + Len: outHdrLen + payloadLen, + Error: 0, + Unique: readFUSEHeaderIn.Unique, + } + + // Echo the payload back. + outHeader.MarshalUnsafe(outBuf[:outHdrLen]) + readPayload.MarshalUnsafe(outBuf[outHdrLen:]) + outIOseq := usermem.BytesIOSequence(outBuf) + + n, err = fd.Write(s.Ctx, outIOseq, vfs.WriteOptions{}) + if err != nil { + t.Fatalf("Write failed :%v", err) + } + } +} + +func setup(t *testing.T) *testutil.System { + k, err := testutil.Boot() + if err != nil { + t.Fatalf("Error creating kernel: %v", err) + } + + ctx := k.SupervisorContext() + creds := auth.CredentialsFromContext(ctx) + + k.VFS().MustRegisterFilesystemType(Name, &FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserList: true, + AllowUserMount: true, + }) + + mntns, err := k.VFS().NewMountNamespace(ctx, creds, "", tmpfs.Name, &vfs.GetFilesystemOptions{}) + if err != nil { + t.Fatalf("NewMountNamespace(): %v", err) + } + + return testutil.NewSystem(ctx, t, k.VFS(), mntns) +} + +// newTestConnection creates a fuse connection that the sentry can communicate with +// and the FD for the server to communicate with. +func newTestConnection(system *testutil.System, k *kernel.Kernel, maxActiveRequests uint64) (*Connection, *vfs.FileDescription, error) { + vfsObj := &vfs.VirtualFilesystem{} + fuseDev := &DeviceFD{} + + if err := vfsObj.Init(); err != nil { + return nil, nil, err + } + + vd := vfsObj.NewAnonVirtualDentry("genCountFD") + defer vd.DecRef() + if err := fuseDev.vfsfd.Init(fuseDev, linux.O_RDWR|linux.O_CREAT, vd.Mount(), vd.Dentry(), &vfs.FileDescriptionOptions{}); err != nil { + return nil, nil, err + } + + fsopts := filesystemOptions{ + maxActiveRequests: maxActiveRequests, + } + fs, err := NewFUSEFilesystem(system.Ctx, 0, &fsopts, &fuseDev.vfsfd) + if err != nil { + return nil, nil, err + } + + return fs.conn, &fuseDev.vfsfd, nil +} + +// SizeBytes implements marshal.Marshallable.SizeBytes. +func (t *testPayload) SizeBytes() int { + return 4 +} + +// MarshalBytes implements marshal.Marshallable.MarshalBytes. +func (t *testPayload) MarshalBytes(dst []byte) { + usermem.ByteOrder.PutUint32(dst[:4], t.data) +} + +// UnmarshalBytes implements marshal.Marshallable.UnmarshalBytes. +func (t *testPayload) UnmarshalBytes(src []byte) { + *t = testPayload{data: usermem.ByteOrder.Uint32(src[:4])} +} + +// Packed implements marshal.Marshallable.Packed. +func (t *testPayload) Packed() bool { + return true +} + +// MarshalUnsafe implements marshal.Marshallable.MarshalUnsafe. +func (t *testPayload) MarshalUnsafe(dst []byte) { + t.MarshalBytes(dst) +} + +// UnmarshalUnsafe implements marshal.Marshallable.UnmarshalUnsafe. +func (t *testPayload) UnmarshalUnsafe(src []byte) { + t.UnmarshalBytes(src) +} + +// CopyOutN implements marshal.Marshallable.CopyOutN. +func (t *testPayload) CopyOutN(task marshal.Task, addr usermem.Addr, limit int) (int, error) { + panic("not implemented") +} + +// CopyOut implements marshal.Marshallable.CopyOut. +func (t *testPayload) CopyOut(task marshal.Task, addr usermem.Addr) (int, error) { + panic("not implemented") +} + +// CopyIn implements marshal.Marshallable.CopyIn. +func (t *testPayload) CopyIn(task marshal.Task, addr usermem.Addr) (int, error) { + panic("not implemented") +} + +// WriteTo implements io.WriterTo.WriteTo. +func (t *testPayload) WriteTo(w io.Writer) (int64, error) { + panic("not implemented") +} diff --git a/pkg/sentry/fsimpl/fuse/fusefs.go b/pkg/sentry/fsimpl/fuse/fusefs.go index f7775fb9b..911b6f7cb 100644 --- a/pkg/sentry/fsimpl/fuse/fusefs.go +++ b/pkg/sentry/fsimpl/fuse/fusefs.go @@ -51,6 +51,11 @@ type filesystemOptions struct { // rootMode specifies the the file mode of the filesystem's root. rootMode linux.FileMode + + // maxActiveRequests specifies the maximum number of active requests that can + // exist at any time. Any further requests will block when trying to + // Call the server. + maxActiveRequests uint64 } // filesystem implements vfs.FilesystemImpl. @@ -58,12 +63,12 @@ type filesystem struct { kernfs.Filesystem devMinor uint32 - // fuseFD is the FD returned when opening /dev/fuse. It is used for communication - // between the FUSE server daemon and the sentry fusefs. - fuseFD *DeviceFD + // conn is used for communication between the FUSE server + // daemon and the sentry fusefs. + conn *Connection // opts is the options the fusefs is initialized with. - opts filesystemOptions + opts *filesystemOptions } // Name implements vfs.FilesystemType.Name. @@ -100,7 +105,7 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt fuseFd := kernelTask.GetFileVFS2(int32(deviceDescriptor)) // Parse and set all the other supported FUSE mount options. - // TODO: Expand the supported mount options. + // TODO(gVisor.dev/issue/3229): Expand the supported mount options. if userIDStr, ok := mopts["user_id"]; ok { delete(mopts, "user_id") userID, err := strconv.ParseUint(userIDStr, 10, 32) @@ -134,21 +139,20 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt } fsopts.rootMode = rootMode + // Set the maxInFlightRequests option. + fsopts.maxActiveRequests = MaxActiveRequestsDefault + // Check for unparsed options. if len(mopts) != 0 { log.Warningf("%s.GetFilesystem: unknown options: %v", fsType.Name(), mopts) return nil, nil, syserror.EINVAL } - // Mark the device as ready so it can be used. /dev/fuse can only be used if the FD was used to - // mount a FUSE filesystem. - fuseFD := fuseFd.Impl().(*DeviceFD) - fuseFD.mounted = true - - fs := &filesystem{ - devMinor: devMinor, - fuseFD: fuseFD, - opts: fsopts, + // Create a new FUSE filesystem. + fs, err := NewFUSEFilesystem(ctx, devMinor, &fsopts, fuseFd) + if err != nil { + log.Warningf("%s.NewFUSEFilesystem: failed with error: %v", fsType.Name(), err) + return nil, nil, err } fs.VFSFilesystem().Init(vfsObj, &fsType, fs) @@ -162,6 +166,26 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt return fs.VFSFilesystem(), root.VFSDentry(), nil } +// NewFUSEFilesystem creates a new FUSE filesystem. +func NewFUSEFilesystem(ctx context.Context, devMinor uint32, opts *filesystemOptions, device *vfs.FileDescription) (*filesystem, error) { + fs := &filesystem{ + devMinor: devMinor, + opts: opts, + } + + conn, err := NewFUSEConnection(ctx, device, opts.maxActiveRequests) + if err != nil { + log.Warningf("fuse.NewFUSEFilesystem: NewFUSEConnection failed with error: %v", err) + return nil, syserror.EINVAL + } + + fs.conn = conn + fuseFD := device.Impl().(*DeviceFD) + fuseFD.fs = fs + + return fs, nil +} + // Release implements vfs.FilesystemImpl.Release. func (fs *filesystem) Release() { fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) diff --git a/pkg/sentry/fsimpl/fuse/register.go b/pkg/sentry/fsimpl/fuse/register.go new file mode 100644 index 000000000..b5b581152 --- /dev/null +++ b/pkg/sentry/fsimpl/fuse/register.go @@ -0,0 +1,42 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package fuse + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/devtmpfs" + "gvisor.dev/gvisor/pkg/sentry/vfs" +) + +// Register registers the FUSE device with vfsObj. +func Register(vfsObj *vfs.VirtualFilesystem) error { + if err := vfsObj.RegisterDevice(vfs.CharDevice, linux.MISC_MAJOR, fuseDevMinor, fuseDevice{}, &vfs.RegisterDeviceOptions{ + GroupName: "misc", + }); err != nil { + return err + } + + return nil +} + +// CreateDevtmpfsFile creates a device special file in devtmpfs. +func CreateDevtmpfsFile(ctx context.Context, dev *devtmpfs.Accessor) error { + if err := dev.CreateDeviceFile(ctx, "fuse", vfs.CharDevice, linux.MISC_MAJOR, fuseDevMinor, 0666 /* mode */); err != nil { + return err + } + + return nil +} -- cgit v1.2.3 From 18c246359663ae7e424f94225f11f87cc63eccd9 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Mon, 27 Jul 2020 19:25:31 -0700 Subject: Fix strace for epoll event arrays. PiperOrigin-RevId: 323491461 --- pkg/sentry/strace/epoll.go | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/strace/epoll.go b/pkg/sentry/strace/epoll.go index a6e48b836..5d51a7792 100644 --- a/pkg/sentry/strace/epoll.go +++ b/pkg/sentry/strace/epoll.go @@ -50,10 +50,10 @@ func epollEvents(t *kernel.Task, eventsAddr usermem.Addr, numEvents, maxBytes ui sb.WriteString("...") break } - if _, ok := addr.AddLength(uint64(linux.SizeOfEpollEvent)); !ok { - fmt.Fprintf(&sb, "{error reading event at %#x: EFAULT}", addr) - continue - } + // Allowing addr to overflow is consistent with Linux, and harmless; if + // this isn't the last iteration of the loop, the next call to CopyIn + // will just fail with EFAULT. + addr, _ = addr.AddLength(uint64(linux.SizeOfEpollEvent)) } sb.WriteString("}") return sb.String() @@ -75,7 +75,7 @@ var epollEventEvents = abi.FlagSet{ {Flag: linux.EPOLLPRI, Name: "EPOLLPRI"}, {Flag: linux.EPOLLOUT, Name: "EPOLLOUT"}, {Flag: linux.EPOLLERR, Name: "EPOLLERR"}, - {Flag: linux.EPOLLHUP, Name: "EPULLHUP"}, + {Flag: linux.EPOLLHUP, Name: "EPOLLHUP"}, {Flag: linux.EPOLLRDNORM, Name: "EPOLLRDNORM"}, {Flag: linux.EPOLLRDBAND, Name: "EPOLLRDBAND"}, {Flag: linux.EPOLLWRNORM, Name: "EPOLLWRNORM"}, -- cgit v1.2.3 From f82dd8ddb477b8923d1db12654a62d55d613019c Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Tue, 28 Jul 2020 21:22:52 -0700 Subject: Redirect TODO to GitHub issues PiperOrigin-RevId: 323715260 --- pkg/sentry/socket/netstack/netstack.go | 4 ++-- pkg/tcpip/stack/nic.go | 2 +- pkg/tcpip/stack/stack_test.go | 6 +++--- pkg/tcpip/transport/packet/endpoint.go | 4 ++-- test/syscalls/linux/packet_socket.cc | 4 ++-- test/syscalls/linux/packet_socket_raw.cc | 4 ++-- 6 files changed, 12 insertions(+), 12 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 44b3fff46..f86e6cd7a 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -423,7 +423,7 @@ func AddressAndFamily(addr []byte) (tcpip.FullAddress, uint16, *syserr.Error) { return tcpip.FullAddress{}, family, syserr.ErrInvalidArgument } - // TODO(b/129292371): Return protocol too. + // TODO(gvisor.dev/issue/173): Return protocol too. return tcpip.FullAddress{ NIC: tcpip.NICID(a.InterfaceIndex), Addr: tcpip.Address(a.HardwareAddr[:header.EthernetAddressSize]), @@ -2418,7 +2418,7 @@ func ConvertAddress(family int, addr tcpip.FullAddress) (linux.SockAddr, uint32) return &out, uint32(sockAddrInet6Size) case linux.AF_PACKET: - // TODO(b/129292371): Return protocol too. + // TODO(gvisor.dev/issue/173): Return protocol too. var out linux.SockAddrLink out.Family = linux.AF_PACKET out.InterfaceIndex = int32(addr.NIC) diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index fea0ce7e8..9256d4d43 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -181,7 +181,7 @@ func (n *NIC) disableLocked() *tcpip.Error { return nil } - // TODO(b/147015577): Should Routes that are currently bound to n be + // TODO(gvisor.dev/issue/1491): Should Routes that are currently bound to n be // invalidated? Currently, Routes will continue to work when a NIC is enabled // again, and applications may not know that the underlying NIC was ever // disabled. diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index 7657a4101..101ca2206 100644 --- a/pkg/tcpip/stack/stack_test.go +++ b/pkg/tcpip/stack/stack_test.go @@ -867,9 +867,9 @@ func TestRouteWithDownNIC(t *testing.T) { // Writes with Routes that use NIC1 after being brought up should // succeed. // - // TODO(b/147015577): Should we instead completely invalidate all - // Routes that were bound to a NIC that was brought down at some - // point? + // TODO(gvisor.dev/issue/1491): Should we instead completely + // invalidate all Routes that were bound to a NIC that was brought + // down at some point? if err := upFn(s, nicID1); err != nil { t.Fatalf("test.upFn(_, %d): %s", nicID1, err) } diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go index 0e46e6355..df478115d 100644 --- a/pkg/tcpip/transport/packet/endpoint.go +++ b/pkg/tcpip/transport/packet/endpoint.go @@ -193,7 +193,7 @@ func (ep *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMes } func (ep *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { - // TODO(b/129292371): Implement. + // TODO(gvisor.dev/issue/173): Implement. return 0, nil, tcpip.ErrInvalidOptionValue } @@ -432,7 +432,7 @@ func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress, // Push new packet into receive list and increment the buffer size. var packet packet - // TODO(b/129292371): Return network protocol. + // TODO(gvisor.dev/issue/173): Return network protocol. if len(pkt.LinkHeader) > 0 { // Get info directly from the ethernet header. hdr := header.Ethernet(pkt.LinkHeader) diff --git a/test/syscalls/linux/packet_socket.cc b/test/syscalls/linux/packet_socket.cc index 40aa9326d..861617ff7 100644 --- a/test/syscalls/linux/packet_socket.cc +++ b/test/syscalls/linux/packet_socket.cc @@ -188,7 +188,7 @@ void ReceiveMessage(int sock, int ifindex) { // sizeof(sockaddr_ll). ASSERT_THAT(src_len, AnyOf(Eq(sizeof(src)), Eq(sizeof(src) - 2))); - // TODO(b/129292371): Verify protocol once we return it. + // TODO(gvisor.dev/issue/173): Verify protocol once we return it. // Verify the source address. EXPECT_EQ(src.sll_family, AF_PACKET); EXPECT_EQ(src.sll_ifindex, ifindex); @@ -234,7 +234,7 @@ TEST_P(CookedPacketTest, Receive) { // Send via a packet socket. TEST_P(CookedPacketTest, Send) { - // TODO(b/129292371): Remove once we support packet socket writing. + // TODO(gvisor.dev/issue/173): Remove once we support packet socket writing. SKIP_IF(IsRunningOnGvisor()); // Let's send a UDP packet and receive it using a regular UDP socket. diff --git a/test/syscalls/linux/packet_socket_raw.cc b/test/syscalls/linux/packet_socket_raw.cc index 2fca9fe4d..a11a03415 100644 --- a/test/syscalls/linux/packet_socket_raw.cc +++ b/test/syscalls/linux/packet_socket_raw.cc @@ -195,7 +195,7 @@ TEST_P(RawPacketTest, Receive) { // sizeof(sockaddr_ll). ASSERT_THAT(src_len, AnyOf(Eq(sizeof(src)), Eq(sizeof(src) - 2))); - // TODO(b/129292371): Verify protocol once we return it. + // TODO(gvisor.dev/issue/173): Verify protocol once we return it. // Verify the source address. EXPECT_EQ(src.sll_family, AF_PACKET); EXPECT_EQ(src.sll_ifindex, GetLoopbackIndex()); @@ -240,7 +240,7 @@ TEST_P(RawPacketTest, Receive) { // Send via a packet socket. TEST_P(RawPacketTest, Send) { - // TODO(b/129292371): Remove once we support packet socket writing. + // TODO(gvisor.dev/issue/173): Remove once we support packet socket writing. SKIP_IF(IsRunningOnGvisor()); // Let's send a UDP packet and receive it using a regular UDP socket. -- cgit v1.2.3 From 267f48ebe2d25a69177a887b86ac50f3e9e59ed8 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Mon, 20 Jul 2020 03:50:59 -0400 Subject: load/store user fpsimd on Arm64 full context switch: add fpsimd load/store support to container application. Signed-off-by: Bin Lu --- pkg/sentry/platform/ring0/kernel_arm64.go | 2 + pkg/sentry/platform/ring0/lib_arm64.go | 6 +++ pkg/sentry/platform/ring0/lib_arm64.s | 86 +++++++++++++++++++++++++++++++ 3 files changed, 94 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/ring0/kernel_arm64.go b/pkg/sentry/platform/ring0/kernel_arm64.go index fca3a5478..d483ff03c 100644 --- a/pkg/sentry/platform/ring0/kernel_arm64.go +++ b/pkg/sentry/platform/ring0/kernel_arm64.go @@ -59,11 +59,13 @@ func (c *CPU) SwitchToUser(switchOpts SwitchOpts) (vector Vector) { regs.Pstate &= ^uint64(UserFlagsClear) regs.Pstate |= UserFlagsSet + LoadFloatingPoint(switchOpts.FloatingPointState) SetTLS(regs.TPIDR_EL0) kernelExitToEl0() regs.TPIDR_EL0 = GetTLS() + SaveFloatingPoint(switchOpts.FloatingPointState) vector = c.vecCode diff --git a/pkg/sentry/platform/ring0/lib_arm64.go b/pkg/sentry/platform/ring0/lib_arm64.go index a6345010d..00e52c8af 100644 --- a/pkg/sentry/platform/ring0/lib_arm64.go +++ b/pkg/sentry/platform/ring0/lib_arm64.go @@ -38,6 +38,12 @@ func SaveVRegs(*byte) // LoadVRegs loads V0-V31 registers. func LoadVRegs(*byte) +// LoadFloatingPoint loads floating point state. +func LoadFloatingPoint(*byte) + +// SaveFloatingPoint saves floating point state. +func SaveFloatingPoint(*byte) + // GetTLS returns the value of TPIDR_EL0 register. func GetTLS() (value uint64) diff --git a/pkg/sentry/platform/ring0/lib_arm64.s b/pkg/sentry/platform/ring0/lib_arm64.s index b63e14b41..86bfbe46f 100644 --- a/pkg/sentry/platform/ring0/lib_arm64.s +++ b/pkg/sentry/platform/ring0/lib_arm64.s @@ -129,3 +129,89 @@ TEXT ·LoadVRegs(SB),NOSPLIT,$0-8 ISB $15 RET + +TEXT ·LoadFloatingPoint(SB),NOSPLIT,$0-8 + MOVD addr+0(FP), R0 + + MOVD 0(R0), R1 + MOVD R1, FPSR + MOVD 8(R0), R1 + MOVD R1, NZCV + + FMOVD 16*1(R0), F0 + FMOVD 16*2(R0), F1 + FMOVD 16*3(R0), F2 + FMOVD 16*4(R0), F3 + FMOVD 16*5(R0), F4 + FMOVD 16*6(R0), F5 + FMOVD 16*7(R0), F6 + FMOVD 16*8(R0), F7 + FMOVD 16*9(R0), F8 + FMOVD 16*10(R0), F9 + FMOVD 16*11(R0), F10 + FMOVD 16*12(R0), F11 + FMOVD 16*13(R0), F12 + FMOVD 16*14(R0), F13 + FMOVD 16*15(R0), F14 + FMOVD 16*16(R0), F15 + FMOVD 16*17(R0), F16 + FMOVD 16*18(R0), F17 + FMOVD 16*19(R0), F18 + FMOVD 16*20(R0), F19 + FMOVD 16*21(R0), F20 + FMOVD 16*22(R0), F21 + FMOVD 16*23(R0), F22 + FMOVD 16*24(R0), F23 + FMOVD 16*25(R0), F24 + FMOVD 16*26(R0), F25 + FMOVD 16*27(R0), F26 + FMOVD 16*28(R0), F27 + FMOVD 16*29(R0), F28 + FMOVD 16*30(R0), F29 + FMOVD 16*31(R0), F30 + FMOVD 16*32(R0), F31 + + RET + +TEXT ·SaveFloatingPoint(SB),NOSPLIT,$0-8 + MOVD addr+0(FP), R0 + + MOVD FPSR, R1 + MOVD R1, 0(R0) + MOVD NZCV, R1 + MOVD R1, 8(R0) + + FMOVD F0, 16*1(R0) + FMOVD F1, 16*2(R0) + FMOVD F2, 16*3(R0) + FMOVD F3, 16*4(R0) + FMOVD F4, 16*5(R0) + FMOVD F5, 16*6(R0) + FMOVD F6, 16*7(R0) + FMOVD F7, 16*8(R0) + FMOVD F8, 16*9(R0) + FMOVD F9, 16*10(R0) + FMOVD F10, 16*11(R0) + FMOVD F11, 16*12(R0) + FMOVD F12, 16*13(R0) + FMOVD F13, 16*14(R0) + FMOVD F14, 16*15(R0) + FMOVD F15, 16*16(R0) + FMOVD F16, 16*17(R0) + FMOVD F17, 16*18(R0) + FMOVD F18, 16*19(R0) + FMOVD F19, 16*20(R0) + FMOVD F20, 16*21(R0) + FMOVD F21, 16*22(R0) + FMOVD F22, 16*23(R0) + FMOVD F23, 16*24(R0) + FMOVD F24, 16*25(R0) + FMOVD F25, 16*26(R0) + FMOVD F26, 16*27(R0) + FMOVD F27, 16*28(R0) + FMOVD F28, 16*29(R0) + FMOVD F29, 16*30(R0) + FMOVD F30, 16*31(R0) + FMOVD F31, 16*32(R0) + + RET -- cgit v1.2.3 From 4cd4759238655e35c8dc63723f4e55014b5ea9ea Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Wed, 29 Jul 2020 14:56:14 -0700 Subject: Force registration for EPOLLHUP, not EPOLLRDHUP, in vfs2's epoll. Compare Linux's fs/eventpoll.c:do_epoll_ctl(). I don't know where EPOLLRDHUP came from. PiperOrigin-RevId: 323874419 --- pkg/sentry/vfs/epoll.go | 4 ++-- test/syscalls/linux/epoll.cc | 22 ++++++++++++++++++++++ 2 files changed, 24 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/vfs/epoll.go b/pkg/sentry/vfs/epoll.go index 599c3131c..5b009b928 100644 --- a/pkg/sentry/vfs/epoll.go +++ b/pkg/sentry/vfs/epoll.go @@ -186,7 +186,7 @@ func (ep *EpollInstance) AddInterest(file *FileDescription, num int32, event lin } // Register interest in file. - mask := event.Events | linux.EPOLLERR | linux.EPOLLRDHUP + mask := event.Events | linux.EPOLLERR | linux.EPOLLHUP epi := &epollInterest{ epoll: ep, key: key, @@ -257,7 +257,7 @@ func (ep *EpollInstance) ModifyInterest(file *FileDescription, num int32, event } // Update epi for the next call to ep.ReadEvents(). - mask := event.Events | linux.EPOLLERR | linux.EPOLLRDHUP + mask := event.Events | linux.EPOLLERR | linux.EPOLLHUP ep.mu.Lock() epi.mask = mask epi.userData = event.Data diff --git a/test/syscalls/linux/epoll.cc b/test/syscalls/linux/epoll.cc index f57d38dc7..2101e5c9f 100644 --- a/test/syscalls/linux/epoll.cc +++ b/test/syscalls/linux/epoll.cc @@ -422,6 +422,28 @@ TEST(EpollTest, CloseFile) { SyscallSucceedsWithValue(0)); } +TEST(EpollTest, PipeReaderHupAfterWriterClosed) { + auto epollfd = ASSERT_NO_ERRNO_AND_VALUE(NewEpollFD()); + int pipefds[2]; + ASSERT_THAT(pipe(pipefds), SyscallSucceeds()); + FileDescriptor rfd(pipefds[0]); + FileDescriptor wfd(pipefds[1]); + + ASSERT_NO_ERRNO(RegisterEpollFD(epollfd.get(), rfd.get(), 0, kMagicConstant)); + struct epoll_event result[kFDsPerEpoll]; + // Initially, rfd should not generate any events of interest. + ASSERT_THAT(epoll_wait(epollfd.get(), result, kFDsPerEpoll, 0), + SyscallSucceedsWithValue(0)); + // Close the write end of the pipe. + wfd.reset(); + // rfd should now generate EPOLLHUP, which EPOLL_CTL_ADD unconditionally adds + // to the set of events of interest. + ASSERT_THAT(epoll_wait(epollfd.get(), result, kFDsPerEpoll, 0), + SyscallSucceedsWithValue(1)); + EXPECT_EQ(result[0].events, EPOLLHUP); + EXPECT_EQ(result[0].data.u64, kMagicConstant); +} + } // namespace } // namespace testing -- cgit v1.2.3 From 2e19a8b951e9402b28b4e601e65c51e69c815db1 Mon Sep 17 00:00:00 2001 From: Jinmou Li Date: Sat, 27 Jun 2020 02:14:56 +0000 Subject: Add FUSE_INIT This change allows the sentry to send FUSE_INIT request and process the reply. It adds the corresponding structs, employs the fuse device to send and read the message, and stores the results of negotiation in corresponding places (inside connection struct). It adds a CallAsync() function to the FUSE connection interface: - like Call(), but it's for requests that do not expect immediate response (init, release, interrupt etc.) - will block if the connection hasn't initialized, which is the same for Call() --- pkg/abi/linux/fuse.go | 105 ++++++++++++++++ pkg/sentry/fsimpl/fuse/BUILD | 4 +- pkg/sentry/fsimpl/fuse/connection.go | 234 +++++++++++++++++++++++++++++++---- pkg/sentry/fsimpl/fuse/dev.go | 13 +- pkg/sentry/fsimpl/fuse/dev_test.go | 19 ++- pkg/sentry/fsimpl/fuse/fusefs.go | 22 ++-- pkg/sentry/fsimpl/fuse/init.go | 166 +++++++++++++++++++++++++ pkg/syserror/syserror.go | 1 + 8 files changed, 512 insertions(+), 52 deletions(-) create mode 100644 pkg/sentry/fsimpl/fuse/init.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/fuse.go b/pkg/abi/linux/fuse.go index d3ebbccc4..5c6ffe4a3 100644 --- a/pkg/abi/linux/fuse.go +++ b/pkg/abi/linux/fuse.go @@ -141,3 +141,108 @@ type FUSEWriteIn struct { _ uint32 } + +// FUSE_INIT flags, consistent with the ones in include/uapi/linux/fuse.h. +const ( + FUSE_ASYNC_READ = 1 << 0 + FUSE_POSIX_LOCKS = 1 << 1 + FUSE_FILE_OPS = 1 << 2 + FUSE_ATOMIC_O_TRUNC = 1 << 3 + FUSE_EXPORT_SUPPORT = 1 << 4 + FUSE_BIG_WRITES = 1 << 5 + FUSE_DONT_MASK = 1 << 6 + FUSE_SPLICE_WRITE = 1 << 7 + FUSE_SPLICE_MOVE = 1 << 8 + FUSE_SPLICE_READ = 1 << 9 + FUSE_FLOCK_LOCKS = 1 << 10 + FUSE_HAS_IOCTL_DIR = 1 << 11 + FUSE_AUTO_INVAL_DATA = 1 << 12 + FUSE_DO_READDIRPLUS = 1 << 13 + FUSE_READDIRPLUS_AUTO = 1 << 14 + FUSE_ASYNC_DIO = 1 << 15 + FUSE_WRITEBACK_CACHE = 1 << 16 + FUSE_NO_OPEN_SUPPORT = 1 << 17 + FUSE_PARALLEL_DIROPS = 1 << 18 + FUSE_HANDLE_KILLPRIV = 1 << 19 + FUSE_POSIX_ACL = 1 << 20 + FUSE_ABORT_ERROR = 1 << 21 + FUSE_MAX_PAGES = 1 << 22 + FUSE_CACHE_SYMLINKS = 1 << 23 + FUSE_NO_OPENDIR_SUPPORT = 1 << 24 + FUSE_EXPLICIT_INVAL_DATA = 1 << 25 + FUSE_MAP_ALIGNMENT = 1 << 26 +) + +// currently supported FUSE protocol version numbers. +const ( + FUSE_KERNEL_VERSION = 7 + FUSE_KERNEL_MINOR_VERSION = 31 +) + +// FUSEInitIn is the request sent by the kernel to the daemon, +// to negotiate the version and flags. +// +// +marshal +type FUSEInitIn struct { + // Major version supported by kernel. + Major uint32 + + // Minor version supported by the kernel. + Minor uint32 + + // MaxReadahead is the maximum number of bytes to read-ahead + // decided by the kernel. + MaxReadahead uint32 + + // Flags of this init request. + Flags uint32 +} + +// FUSEInitOut is the reply sent by the daemon to the kernel +// for FUSEInitIn. +// +// +marshal +type FUSEInitOut struct { + // Major version supported by daemon. + Major uint32 + + // Minor version supported by daemon. + Minor uint32 + + // MaxReadahead is the maximum number of bytes to read-ahead. + // Decided by the daemon, after receiving the value from kernel. + MaxReadahead uint32 + + // Flags of this init reply. + Flags uint32 + + // MaxBackground is the maximum number of pending background requests + // that the daemon wants. + MaxBackground uint16 + + // CongestionThreshold is the daemon-decided threshold for + // the number of the pending background requests. + CongestionThreshold uint16 + + // MaxWrite is the daemon's maximum size of a write buffer. + // Kernel adjusts it to the minimum (fuse/init.go:fuseMinMaxWrite). + // if the value from daemon is too small. + MaxWrite uint32 + + // TimeGran is the daemon's time granularity for mtime and ctime metadata. + // The unit is nanosecond. + // Value should be power of 10. + // 1 indicates full nanosecond granularity support. + TimeGran uint32 + + // MaxPages is the daemon's maximum number of pages for one write operation. + // Kernel adjusts it to the maximum (fuse/init.go:FUSE_MAX_MAX_PAGES). + // if the value from daemon is too large. + MaxPages uint16 + + // MapAlignment is an unknown field and not used by this package at this moment. + // Use as a placeholder to be consistent with the FUSE protocol. + MapAlignment uint16 + + _ [8]uint32 +} diff --git a/pkg/sentry/fsimpl/fuse/BUILD b/pkg/sentry/fsimpl/fuse/BUILD index 67649e811..999111deb 100644 --- a/pkg/sentry/fsimpl/fuse/BUILD +++ b/pkg/sentry/fsimpl/fuse/BUILD @@ -21,6 +21,7 @@ go_library( "connection.go", "dev.go", "fusefs.go", + "init.go", "register.go", "request_list.go", ], @@ -44,14 +45,13 @@ go_library( ) go_test( - name = "dev_test", + name = "fuse_test", size = "small", srcs = ["dev_test.go"], library = ":fuse", deps = [ "//pkg/abi/linux", "//pkg/sentry/fsimpl/testutil", - "//pkg/sentry/fsimpl/tmpfs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/vfs", diff --git a/pkg/sentry/fsimpl/fuse/connection.go b/pkg/sentry/fsimpl/fuse/connection.go index f330da0bd..6df2728ab 100644 --- a/pkg/sentry/fsimpl/fuse/connection.go +++ b/pkg/sentry/fsimpl/fuse/connection.go @@ -17,6 +17,8 @@ package fuse import ( "errors" "fmt" + "sync" + "sync/atomic" "syscall" "gvisor.dev/gvisor/pkg/abi/linux" @@ -25,18 +27,29 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/waiter" "gvisor.dev/gvisor/tools/go_marshal/marshal" ) -// MaxActiveRequestsDefault is the default setting controlling the upper bound +// maxActiveRequestsDefault is the default setting controlling the upper bound // on the number of active requests at any given time. -const MaxActiveRequestsDefault = 10000 +const maxActiveRequestsDefault = 10000 -var ( - // Ordinary requests have even IDs, while interrupts IDs are odd. - InitReqBit uint64 = 1 - ReqIDStep uint64 = 2 +// Ordinary requests have even IDs, while interrupts IDs are odd. +// Used to increment the unique ID for each FUSE request. +var reqIDStep uint64 = 2 + +const ( + // fuseDefaultMaxBackground is the default value for MaxBackground. + fuseDefaultMaxBackground = 12 + + // fuseDefaultCongestionThreshold is the default value for CongestionThreshold, + // and is 75% of the default maximum of MaxGround. + fuseDefaultCongestionThreshold = (fuseDefaultMaxBackground * 3 / 4) + + // fuseDefaultMaxPagesPerReq is the default value for MaxPagesPerReq. + fuseDefaultMaxPagesPerReq = 32 ) // Request represents a FUSE operation request that hasn't been sent to the @@ -61,17 +74,125 @@ type Response struct { data []byte } -// Connection is the struct by which the sentry communicates with the FUSE server daemon. -type Connection struct { +// connection is the struct by which the sentry communicates with the FUSE server daemon. +type connection struct { fd *DeviceFD - // MaxWrite is the daemon's maximum size of a write buffer. - // This is negotiated during FUSE_INIT. - MaxWrite uint32 + // The following FUSE_INIT flags are currently unsupported by this implementation: + // - FUSE_ATOMIC_O_TRUNC: requires open(..., O_TRUNC) + // - FUSE_EXPORT_SUPPORT + // - FUSE_HANDLE_KILLPRIV + // - FUSE_POSIX_LOCKS: requires POSIX locks + // - FUSE_FLOCK_LOCKS: requires POSIX locks + // - FUSE_AUTO_INVAL_DATA: requires page caching eviction + // - FUSE_EXPLICIT_INVAL_DATA: requires page caching eviction + // - FUSE_DO_READDIRPLUS/FUSE_READDIRPLUS_AUTO: requires FUSE_READDIRPLUS implementation + // - FUSE_ASYNC_DIO + // - FUSE_POSIX_ACL: affects defaultPermissions, posixACL, xattr handler + + // initialized after receiving FUSE_INIT reply. + // Until it's set, suspend sending FUSE requests. + // Use SetInitialized() and IsInitialized() for atomic access. + initialized int32 + + // initializedChan is used to block requests before initialization. + initializedChan chan struct{} + + // blocked when there are too many outstading backgrounds requests (NumBackground == MaxBackground). + // TODO(gvisor.dev/issue/3185): update the numBackground accordingly; use a channel to block. + blocked bool + + // connected (connection established) when a new FUSE file system is created. + // Set to false when: + // umount, + // connection abort, + // device release. + connected bool + + // aborted via sysfs. + // TODO(gvisor.dev/issue/3185): abort all queued requests. + aborted bool + + // connInitError if FUSE_INIT encountered error (major version mismatch). + // Only set in INIT. + connInitError bool + + // connInitSuccess if FUSE_INIT is successful. + // Only set in INIT. + // Used for destory. + connInitSuccess bool + + // TODO(gvisor.dev/issue/3185): All the queue logic are working in progress. + + // NumberBackground is the number of requests in the background. + numBackground uint16 + + // congestionThreshold for NumBackground. + // Negotiated in FUSE_INIT. + congestionThreshold uint16 + + // maxBackground is the maximum number of NumBackground. + // Block connection when it is reached. + // Negotiated in FUSE_INIT. + maxBackground uint16 + + // numActiveBackground is the number of requests in background and has being marked as active. + numActiveBackground uint16 + + // numWating is the number of requests waiting for completion. + numWaiting uint32 + + // TODO(gvisor.dev/issue/3185): BgQueue + // some queue for background queued requests. + + // bgLock protects: + // MaxBackground, CongestionThreshold, NumBackground, + // NumActiveBackground, BgQueue, Blocked. + bgLock sync.Mutex + + // maxRead is the maximum size of a read buffer in in bytes. + maxRead uint32 + + // maxWrite is the maximum size of a write buffer in bytes. + // Negotiated in FUSE_INIT. + maxWrite uint32 + + // maxPages is the maximum number of pages for a single request to use. + // Negotiated in FUSE_INIT. + maxPages uint16 + + // minor version of the FUSE protocol. + // Negotiated and only set in INIT. + minor uint32 + + // asyncRead if read pages asynchronously. + // Negotiated and only set in INIT. + asyncRead bool + + // abortErr is true if kernel need to return an unique read error after abort. + // Negotiated and only set in INIT. + abortErr bool + + // writebackCache is true for write-back cache policy, + // false for write-through policy. + // Negotiated and only set in INIT. + writebackCache bool + + // cacheSymlinks if filesystem needs to cache READLINK responses in page cache. + // Negotiated and only set in INIT. + cacheSymlinks bool + + // bigWrites if doing multi-page cached writes. + // Negotiated and only set in INIT. + bigWrites bool + + // dontMask if filestestem does not apply umask to creation modes. + // Negotiated in INIT. + dontMask bool } -// NewFUSEConnection creates a FUSE connection to fd -func NewFUSEConnection(_ context.Context, fd *vfs.FileDescription, maxInFlightRequests uint64) (*Connection, error) { +// newFUSEConnection creates a FUSE connection to fd. +func newFUSEConnection(_ context.Context, fd *vfs.FileDescription, maxInFlightRequests uint64) (*connection, error) { // Mark the device as ready so it can be used. /dev/fuse can only be used if the FD was used to // mount a FUSE filesystem. fuseFD := fd.Impl().(*DeviceFD) @@ -84,16 +205,41 @@ func NewFUSEConnection(_ context.Context, fd *vfs.FileDescription, maxInFlightRe fuseFD.fullQueueCh = make(chan struct{}, maxInFlightRequests) fuseFD.writeCursor = 0 - return &Connection{ - fd: fuseFD, + return &connection{ + fd: fuseFD, + maxBackground: fuseDefaultMaxBackground, + congestionThreshold: fuseDefaultCongestionThreshold, + maxPages: fuseDefaultMaxPagesPerReq, + initializedChan: make(chan struct{}), + connected: true, }, nil } +// SetInitialized atomically sets the connection as initialized. +func (conn *connection) SetInitialized() { + // Unblock the requests sent before INIT. + close(conn.initializedChan) + + // Close the channel first to avoid the non-atomic situation + // where conn.initialized is true but there are + // tasks being blocked on the channel. + // And it prevents the newer tasks from gaining + // unnecessary higher chance to be issued before the blocked one. + + atomic.StoreInt32(&(conn.initialized), int32(1)) +} + +// IsInitialized atomically check if the connection is initialized. +// pairs with SetInitialized(). +func (conn *connection) Initialized() bool { + return atomic.LoadInt32(&(conn.initialized)) != 0 +} + // NewRequest creates a new request that can be sent to the FUSE server. -func (conn *Connection) NewRequest(creds *auth.Credentials, pid uint32, ino uint64, opcode linux.FUSEOpcode, payload marshal.Marshallable) (*Request, error) { +func (conn *connection) NewRequest(creds *auth.Credentials, pid uint32, ino uint64, opcode linux.FUSEOpcode, payload marshal.Marshallable) (*Request, error) { conn.fd.mu.Lock() defer conn.fd.mu.Unlock() - conn.fd.nextOpID += linux.FUSEOpID(ReqIDStep) + conn.fd.nextOpID += linux.FUSEOpID(reqIDStep) hdrLen := (*linux.FUSEHeaderIn)(nil).SizeBytes() hdr := linux.FUSEHeaderIn{ @@ -118,13 +264,49 @@ func (conn *Connection) NewRequest(creds *auth.Credentials, pid uint32, ino uint } // Call makes a request to the server and blocks the invoking task until a -// server responds with a response. -// NOTE: If no task is provided then the Call will simply enqueue the request -// and return a nil response. No blocking will happen in this case. Instead, -// this is used to signify that the processing of this request will happen by -// the kernel.Task that writes the response. See FUSE_INIT for such an -// invocation. -func (conn *Connection) Call(t *kernel.Task, r *Request) (*Response, error) { +// server responds with a response. Task should never be nil. +// Requests will not be sent before the connection is initialized. +// For async tasks, use CallAsync(). +func (conn *connection) Call(t *kernel.Task, r *Request) (*Response, error) { + // Block requests sent before connection is initalized. + if !conn.Initialized() { + if err := t.Block(conn.initializedChan); err != nil { + return nil, err + } + } + + return conn.call(t, r) +} + +// CallAsync makes an async (aka background) request. +// Those requests either do not expect a response (e.g. release) or +// the response should be handled by others (e.g. init). +// Return immediately unless the connection is blocked (before initialization). +// Async call example: init, release, forget, aio, interrupt. +// When the Request is FUSE_INIT, it will not be blocked before initialization. +func (conn *connection) CallAsync(t *kernel.Task, r *Request) error { + // Block requests sent before connection is initalized. + if !conn.Initialized() && r.hdr.Opcode != linux.FUSE_INIT { + if err := t.Block(conn.initializedChan); err != nil { + return err + } + } + + // This should be the only place that invokes call() with a nil task. + _, err := conn.call(nil, r) + return err +} + +// call makes a call without blocking checks. +func (conn *connection) call(t *kernel.Task, r *Request) (*Response, error) { + if !conn.connected { + return nil, syserror.ENOTCONN + } + + if conn.connInitError { + return nil, syserror.ECONNREFUSED + } + fut, err := conn.callFuture(t, r) if err != nil { return nil, err @@ -160,7 +342,7 @@ func (r *Response) UnmarshalPayload(m marshal.Marshallable) error { // callFuture makes a request to the server and returns a future response. // Call resolve() when the response needs to be fulfilled. -func (conn *Connection) callFuture(t *kernel.Task, r *Request) (*futureResponse, error) { +func (conn *connection) callFuture(t *kernel.Task, r *Request) (*futureResponse, error) { conn.fd.mu.Lock() defer conn.fd.mu.Unlock() @@ -195,7 +377,7 @@ func (conn *Connection) callFuture(t *kernel.Task, r *Request) (*futureResponse, } // callFutureLocked makes a request to the server and returns a future response. -func (conn *Connection) callFutureLocked(t *kernel.Task, r *Request) (*futureResponse, error) { +func (conn *connection) callFutureLocked(t *kernel.Task, r *Request) (*futureResponse, error) { conn.fd.queue.PushBack(r) conn.fd.numActiveRequests += 1 fut := newFutureResponse(r.hdr.Opcode) diff --git a/pkg/sentry/fsimpl/fuse/dev.go b/pkg/sentry/fsimpl/fuse/dev.go index f3443ac71..2225076bc 100644 --- a/pkg/sentry/fsimpl/fuse/dev.go +++ b/pkg/sentry/fsimpl/fuse/dev.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" @@ -98,7 +99,9 @@ type DeviceFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *DeviceFD) Release() {} +func (fd *DeviceFD) Release() { + fd.fs.conn.connected = false +} // PRead implements vfs.FileDescriptionImpl.PRead. func (fd *DeviceFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { @@ -124,7 +127,7 @@ func (fd *DeviceFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.R minBuffSize := linux.FUSE_MIN_READ_BUFFER inHdrLen := uint32((*linux.FUSEHeaderIn)(nil).SizeBytes()) writeHdrLen := uint32((*linux.FUSEWriteIn)(nil).SizeBytes()) - negotiatedMinBuffSize := inHdrLen + writeHdrLen + fd.fs.conn.MaxWrite + negotiatedMinBuffSize := inHdrLen + writeHdrLen + fd.fs.conn.maxWrite if minBuffSize < negotiatedMinBuffSize { minBuffSize = negotiatedMinBuffSize } @@ -385,9 +388,9 @@ func (fd *DeviceFD) sendError(ctx context.Context, errno int32, req *Request) er // FUSE_INIT. func (fd *DeviceFD) noReceiverAction(ctx context.Context, r *Response) error { if r.opcode == linux.FUSE_INIT { - // TODO: process init response here. - // Maybe get the creds from the context? - // creds := auth.CredentialsFromContext(ctx) + creds := auth.CredentialsFromContext(ctx) + rootUserNs := kernel.KernelFromContext(ctx).RootUserNamespace() + return fd.fs.conn.InitRecv(r, creds.HasCapabilityIn(linux.CAP_SYS_ADMIN, rootUserNs)) } return nil diff --git a/pkg/sentry/fsimpl/fuse/dev_test.go b/pkg/sentry/fsimpl/fuse/dev_test.go index fcd77832a..122415275 100644 --- a/pkg/sentry/fsimpl/fuse/dev_test.go +++ b/pkg/sentry/fsimpl/fuse/dev_test.go @@ -22,7 +22,6 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/fsimpl/testutil" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" @@ -60,25 +59,25 @@ func TestFUSECommunication(t *testing.T) { Name: "SingleClientSingleServer", NumClients: 1, NumServers: 1, - MaxActiveRequests: MaxActiveRequestsDefault, + MaxActiveRequests: maxActiveRequestsDefault, }, { Name: "SingleClientMultipleServers", NumClients: 1, NumServers: 10, - MaxActiveRequests: MaxActiveRequestsDefault, + MaxActiveRequests: maxActiveRequestsDefault, }, { Name: "MultipleClientsSingleServer", NumClients: 10, NumServers: 1, - MaxActiveRequests: MaxActiveRequestsDefault, + MaxActiveRequests: maxActiveRequestsDefault, }, { Name: "MultipleClientsMultipleServers", NumClients: 10, NumServers: 10, - MaxActiveRequests: MaxActiveRequestsDefault, + MaxActiveRequests: maxActiveRequestsDefault, }, { Name: "RequestCapacityFull", @@ -145,7 +144,7 @@ func TestFUSECommunication(t *testing.T) { // CallTest makes a request to the server and blocks the invoking // goroutine until a server responds with a response. Doesn't block // a kernel.Task. Analogous to Connection.Call but used for testing. -func CallTest(conn *Connection, t *kernel.Task, r *Request, i uint32) (*Response, error) { +func CallTest(conn *connection, t *kernel.Task, r *Request, i uint32) (*Response, error) { conn.fd.mu.Lock() // Wait until we're certain that a new request can be processed. @@ -214,7 +213,7 @@ func ReadTest(serverTask *kernel.Task, fd *vfs.FileDescription, inIOseq usermem. // fuseClientRun emulates all the actions of a normal FUSE request. It creates // a header, a payload, calls the server, waits for the response, and processes // the response. -func fuseClientRun(t *testing.T, s *testutil.System, k *kernel.Kernel, conn *Connection, creds *auth.Credentials, pid uint32, inode uint64, clientDone chan struct{}) { +func fuseClientRun(t *testing.T, s *testutil.System, k *kernel.Kernel, conn *connection, creds *auth.Credentials, pid uint32, inode uint64, clientDone chan struct{}) { defer func() { clientDone <- struct{}{} }() tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) @@ -343,7 +342,7 @@ func setup(t *testing.T) *testutil.System { AllowUserMount: true, }) - mntns, err := k.VFS().NewMountNamespace(ctx, creds, "", tmpfs.Name, &vfs.GetFilesystemOptions{}) + mntns, err := k.VFS().NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{}) if err != nil { t.Fatalf("NewMountNamespace(): %v", err) } @@ -353,7 +352,7 @@ func setup(t *testing.T) *testutil.System { // newTestConnection creates a fuse connection that the sentry can communicate with // and the FD for the server to communicate with. -func newTestConnection(system *testutil.System, k *kernel.Kernel, maxActiveRequests uint64) (*Connection, *vfs.FileDescription, error) { +func newTestConnection(system *testutil.System, k *kernel.Kernel, maxActiveRequests uint64) (*connection, *vfs.FileDescription, error) { vfsObj := &vfs.VirtualFilesystem{} fuseDev := &DeviceFD{} @@ -426,4 +425,4 @@ func (t *testPayload) CopyIn(task marshal.Task, addr usermem.Addr) (int, error) // WriteTo implements io.WriterTo.WriteTo. func (t *testPayload) WriteTo(w io.Writer) (int64, error) { panic("not implemented") -} +} \ No newline at end of file diff --git a/pkg/sentry/fsimpl/fuse/fusefs.go b/pkg/sentry/fsimpl/fuse/fusefs.go index 911b6f7cb..200a93bbf 100644 --- a/pkg/sentry/fsimpl/fuse/fusefs.go +++ b/pkg/sentry/fsimpl/fuse/fusefs.go @@ -65,7 +65,7 @@ type filesystem struct { // conn is used for communication between the FUSE server // daemon and the sentry fusefs. - conn *Connection + conn *connection // opts is the options the fusefs is initialized with. opts *filesystemOptions @@ -140,7 +140,7 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt fsopts.rootMode = rootMode // Set the maxInFlightRequests option. - fsopts.maxActiveRequests = MaxActiveRequestsDefault + fsopts.maxActiveRequests = maxActiveRequestsDefault // Check for unparsed options. if len(mopts) != 0 { @@ -157,8 +157,12 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt fs.VFSFilesystem().Init(vfsObj, &fsType, fs) - // TODO: dispatch a FUSE_INIT request to the FUSE daemon server before - // returning. Mount will not block on this dispatched request. + // Send a FUSE_INIT request to the FUSE daemon server before returning. + // This call is not blocking. + if err := fs.conn.InitSend(creds, uint32(kernelTask.ThreadID())); err != nil { + log.Warningf("%s.InitSend: failed with error: %v", fsType.Name(), err) + return nil, nil, err + } // root is the fusefs root directory. root := fs.newInode(creds, fsopts.rootMode) @@ -173,7 +177,7 @@ func NewFUSEFilesystem(ctx context.Context, devMinor uint32, opts *filesystemOpt opts: opts, } - conn, err := NewFUSEConnection(ctx, device, opts.maxActiveRequests) + conn, err := newFUSEConnection(ctx, device, opts.maxActiveRequests) if err != nil { log.Warningf("fuse.NewFUSEFilesystem: NewFUSEConnection failed with error: %v", err) return nil, syserror.EINVAL @@ -192,8 +196,8 @@ func (fs *filesystem) Release() { fs.Filesystem.Release() } -// Inode implements kernfs.Inode. -type Inode struct { +// inode implements kernfs.Inode. +type inode struct { kernfs.InodeAttrs kernfs.InodeNoDynamicLookup kernfs.InodeNotSymlink @@ -206,7 +210,7 @@ type Inode struct { } func (fs *filesystem) newInode(creds *auth.Credentials, mode linux.FileMode) *kernfs.Dentry { - i := &Inode{} + i := &inode{} i.InodeAttrs.Init(creds, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0755) i.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) i.dentry.Init(i) @@ -215,7 +219,7 @@ func (fs *filesystem) newInode(creds *auth.Credentials, mode linux.FileMode) *ke } // Open implements kernfs.Inode.Open. -func (i *Inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &i.locks, &opts) if err != nil { return nil, err diff --git a/pkg/sentry/fsimpl/fuse/init.go b/pkg/sentry/fsimpl/fuse/init.go new file mode 100644 index 000000000..779c2bd3f --- /dev/null +++ b/pkg/sentry/fsimpl/fuse/init.go @@ -0,0 +1,166 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package fuse + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" +) + +// consts used by FUSE_INIT negotiation. +const ( + // fuseMaxMaxPages is the maximum value for MaxPages received in InitOut. + // Follow the same behavior as unix fuse implementation. + fuseMaxMaxPages = 256 + + // Maximum value for the time granularity for file time stamps, 1s. + // Follow the same behavior as unix fuse implementation. + fuseMaxTimeGranNs = 1000000000 + + // Minimum value for MaxWrite. + // Follow the same behavior as unix fuse implementation. + fuseMinMaxWrite = 4096 + + // Temporary default value for max readahead, 128kb. + fuseDefaultMaxReadahead = 131072 + + // The FUSE_INIT_IN flags sent to the daemon. + // TODO(gvisor.dev/issue/3199): complete the flags. + fuseDefaultInitFlags = linux.FUSE_MAX_PAGES +) + +// Adjustable maximums for Connection's cogestion control parameters. +// Used as the upperbound of the config values. +// Currently we do not support adjustment to them. +var ( + MaxUserBackgroundRequest uint16 = fuseDefaultMaxBackground + MaxUserCongestionThreshold uint16 = fuseDefaultCongestionThreshold +) + +// InitSend sends a FUSE_INIT request. +func (conn *connection) InitSend(creds *auth.Credentials, pid uint32) error { + in := linux.FUSEInitIn{ + Major: linux.FUSE_KERNEL_VERSION, + Minor: linux.FUSE_KERNEL_MINOR_VERSION, + // TODO(gvisor.dev/issue/3196): find appropriate way to calculate this + MaxReadahead: fuseDefaultMaxReadahead, + Flags: fuseDefaultInitFlags, + } + + req, err := conn.NewRequest(creds, pid, 0, linux.FUSE_INIT, &in) + if err != nil { + return err + } + + // Since there is no task to block on and FUSE_INIT is the request + // to unblock other requests, use nil. + return conn.CallAsync(nil, req) +} + +// InitRecv receives a FUSE_INIT reply and process it. +func (conn *connection) InitRecv(res *Response, hasSysAdminCap bool) error { + if err := res.Error(); err != nil { + return err + } + + var out linux.FUSEInitOut + if err := res.UnmarshalPayload(&out); err != nil { + return err + } + + return conn.initProcessReply(&out, hasSysAdminCap) +} + +// Process the FUSE_INIT reply from the FUSE server. +func (conn *connection) initProcessReply(out *linux.FUSEInitOut, hasSysAdminCap bool) error { + // No support for old major fuse versions. + if out.Major != linux.FUSE_KERNEL_VERSION { + conn.connInitError = true + + // Set the connection as initialized and unblock the blocked requests + // (i.e. return error for them). + conn.SetInitialized() + + return nil + } + + // Start processing the reply. + conn.connInitSuccess = true + conn.minor = out.Minor + + // No support for limits before minor version 13. + if out.Minor >= 13 { + conn.bgLock.Lock() + + if out.MaxBackground > 0 { + conn.maxBackground = out.MaxBackground + + if !hasSysAdminCap && + conn.maxBackground > MaxUserBackgroundRequest { + conn.maxBackground = MaxUserBackgroundRequest + } + } + + if out.CongestionThreshold > 0 { + conn.congestionThreshold = out.CongestionThreshold + + if !hasSysAdminCap && + conn.congestionThreshold > MaxUserCongestionThreshold { + conn.congestionThreshold = MaxUserCongestionThreshold + } + } + + conn.bgLock.Unlock() + } + + // No support for the following flags before minor version 6. + if out.Minor >= 6 { + conn.asyncRead = out.Flags&linux.FUSE_ASYNC_READ != 0 + conn.bigWrites = out.Flags&linux.FUSE_BIG_WRITES != 0 + conn.dontMask = out.Flags&linux.FUSE_DONT_MASK != 0 + conn.writebackCache = out.Flags&linux.FUSE_WRITEBACK_CACHE != 0 + conn.cacheSymlinks = out.Flags&linux.FUSE_CACHE_SYMLINKS != 0 + conn.abortErr = out.Flags&linux.FUSE_ABORT_ERROR != 0 + + // TODO(gvisor.dev/issue/3195): figure out how to use TimeGran (0 < TimeGran <= fuseMaxTimeGranNs). + + if out.Flags&linux.FUSE_MAX_PAGES != 0 { + maxPages := out.MaxPages + if maxPages < 1 { + maxPages = 1 + } + if maxPages > fuseMaxMaxPages { + maxPages = fuseMaxMaxPages + } + conn.maxPages = maxPages + } + } + + // No support for negotiating MaxWrite before minor version 5. + if out.Minor >= 5 { + conn.maxWrite = out.MaxWrite + } else { + conn.maxWrite = fuseMinMaxWrite + } + if conn.maxWrite < fuseMinMaxWrite { + conn.maxWrite = fuseMinMaxWrite + } + + // Set connection as initialized and unblock the requests + // issued before init. + conn.SetInitialized() + + return nil +} diff --git a/pkg/syserror/syserror.go b/pkg/syserror/syserror.go index c73072c42..798e07b01 100644 --- a/pkg/syserror/syserror.go +++ b/pkg/syserror/syserror.go @@ -61,6 +61,7 @@ var ( ENOMEM = error(syscall.ENOMEM) ENOSPC = error(syscall.ENOSPC) ENOSYS = error(syscall.ENOSYS) + ENOTCONN = error(syscall.ENOTCONN) ENOTDIR = error(syscall.ENOTDIR) ENOTEMPTY = error(syscall.ENOTEMPTY) ENOTSOCK = error(syscall.ENOTSOCK) -- cgit v1.2.3 From cd1149eef30af3dff5f07d838cd465cc26afbf6f Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Tue, 23 Jun 2020 23:35:10 -0400 Subject: supporting sError injection step 2 on Arm64 I disabled DAIF(DEBUG, sError, IRQ, FIQ) in guest kernel mode, and enabled them in guest user mode. So, I can make sure all DAIF-s come from guest user mode, and then the case 'TestBounceStress' can passed on Arm64. Test steps: 1, cd pkg/sentry/platform/kvm 2, bazel test kvm_test --strip=never --test_output=streamed Signed-off-by: Bin Lu --- pkg/sentry/platform/kvm/bluepill_arm64.go | 4 +-- pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go | 2 +- pkg/sentry/platform/kvm/machine_arm64_unsafe.go | 6 ---- pkg/sentry/platform/ring0/aarch64.go | 35 ++++++++++++------------ pkg/sentry/platform/ring0/entry_arm64.s | 8 ------ pkg/sentry/platform/ring0/kernel_arm64.go | 2 +- 6 files changed, 22 insertions(+), 35 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/kvm/bluepill_arm64.go b/pkg/sentry/platform/kvm/bluepill_arm64.go index dba563160..ed5ae03d3 100644 --- a/pkg/sentry/platform/kvm/bluepill_arm64.go +++ b/pkg/sentry/platform/kvm/bluepill_arm64.go @@ -49,7 +49,7 @@ func bluepillArchEnter(context *arch.SignalContext64) (c *vCPU) { regs.Sp = context.Sp regs.Pc = context.Pc regs.Pstate = context.Pstate - regs.Pstate &^= uint64(ring0.KernelFlagsClear) + regs.Pstate &^= uint64(ring0.PsrFlagsClear) regs.Pstate |= ring0.KernelFlagsSet return } @@ -63,7 +63,7 @@ func bluepillArchExit(c *vCPU, context *arch.SignalContext64) { context.Sp = regs.Sp context.Pc = regs.Pc context.Pstate = regs.Pstate - context.Pstate &^= uint64(ring0.UserFlagsClear) + context.Pstate &^= uint64(ring0.PsrFlagsClear) context.Pstate |= ring0.UserFlagsSet lazyVfp := c.GetLazyVFP() diff --git a/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go b/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go index 8b64f3a1e..b35c930e2 100644 --- a/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go @@ -41,7 +41,7 @@ func fpsimdPtr(addr *byte) *arch.FpsimdContext { func dieArchSetup(c *vCPU, context *arch.SignalContext64, guestRegs *userRegs) { // If the vCPU is in user mode, we set the stack to the stored stack // value in the vCPU itself. We don't want to unwind the user stack. - if guestRegs.Regs.Pstate&ring0.PSR_MODE_MASK == ring0.PSR_MODE_EL0t { + if guestRegs.Regs.Pstate&ring0.PsrModeMask == ring0.UserFlagsSet { regs := c.CPU.Registers() context.Regs[0] = regs.Regs[0] context.Sp = regs.Sp diff --git a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go index ff8c068c0..307a7645f 100644 --- a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go @@ -151,12 +151,6 @@ func (c *vCPU) initArchState() error { // the MMIO address base. arm64HypercallMMIOBase = toLocation - data = ring0.PsrDefaultSet | ring0.KernelFlagsSet - reg.id = _KVM_ARM64_REGS_PSTATE - if err := c.setOneRegister(®); err != nil { - return err - } - // Initialize the PCID database. if hasGuestPCID { // Note that NewPCIDs may return a nil table here, in which diff --git a/pkg/sentry/platform/ring0/aarch64.go b/pkg/sentry/platform/ring0/aarch64.go index 8122ac6e2..87a573cc4 100644 --- a/pkg/sentry/platform/ring0/aarch64.go +++ b/pkg/sentry/platform/ring0/aarch64.go @@ -26,30 +26,31 @@ const ( _PMD_PGT_SIZE = 0x4000 _PTE_PGT_BASE = 0x7000 _PTE_PGT_SIZE = 0x1000 - - _PSR_D_BIT = 0x00000200 - _PSR_A_BIT = 0x00000100 - _PSR_I_BIT = 0x00000080 - _PSR_F_BIT = 0x00000040 ) const ( - // PSR bits - PSR_MODE_EL0t = 0x00000000 - PSR_MODE_EL1t = 0x00000004 - PSR_MODE_EL1h = 0x00000005 - PSR_MODE_MASK = 0x0000000f + // DAIF bits:debug, sError, IRQ, FIQ. + _PSR_D_BIT = 0x00000200 + _PSR_A_BIT = 0x00000100 + _PSR_I_BIT = 0x00000080 + _PSR_F_BIT = 0x00000040 + _PSR_DAIF_SHIFT = 6 + _PSR_DAIF_MASK = 0xf << _PSR_DAIF_SHIFT - // KernelFlagsSet should always be set in the kernel. - KernelFlagsSet = PSR_MODE_EL1h + // PSR bits. + _PSR_MODE_EL0t = 0x00000000 + _PSR_MODE_EL1t = 0x00000004 + _PSR_MODE_EL1h = 0x00000005 + _PSR_MODE_MASK = 0x0000000f - // UserFlagsSet are always set in userspace. - UserFlagsSet = PSR_MODE_EL0t + PsrFlagsClear = _PSR_MODE_MASK | _PSR_DAIF_MASK + PsrModeMask = _PSR_MODE_MASK - KernelFlagsClear = PSR_MODE_MASK - UserFlagsClear = PSR_MODE_MASK + // KernelFlagsSet should always be set in the kernel. + KernelFlagsSet = _PSR_MODE_EL1h | _PSR_D_BIT | _PSR_A_BIT | _PSR_I_BIT | _PSR_F_BIT - PsrDefaultSet = _PSR_D_BIT | _PSR_A_BIT | _PSR_I_BIT | _PSR_F_BIT + // UserFlagsSet are always set in userspace. + UserFlagsSet = _PSR_MODE_EL0t ) // Vector is an exception vector. diff --git a/pkg/sentry/platform/ring0/entry_arm64.s b/pkg/sentry/platform/ring0/entry_arm64.s index 9fd02d628..d8a7bc2f9 100644 --- a/pkg/sentry/platform/ring0/entry_arm64.s +++ b/pkg/sentry/platform/ring0/entry_arm64.s @@ -312,12 +312,6 @@ ISB $15; \ DSB $15; -#define IRQ_ENABLE \ - MSR $2, DAIFSet; - -#define IRQ_DISABLE \ - MSR $2, DAIFClr; - #define VFP_ENABLE \ MOVD $FPEN_ENABLE, R0; \ WORD $0xd5181040; \ //MSR R0, CPACR_EL1 @@ -509,8 +503,6 @@ TEXT ·kernelExitToEl1(SB),NOSPLIT,$0 // Start is the CPU entrypoint. TEXT ·Start(SB),NOSPLIT,$0 - IRQ_DISABLE - // Init. MOVD $SCTLR_EL1_DEFAULT, R1 MSR R1, SCTLR_EL1 diff --git a/pkg/sentry/platform/ring0/kernel_arm64.go b/pkg/sentry/platform/ring0/kernel_arm64.go index fca3a5478..5b98659b2 100644 --- a/pkg/sentry/platform/ring0/kernel_arm64.go +++ b/pkg/sentry/platform/ring0/kernel_arm64.go @@ -56,7 +56,7 @@ func (c *CPU) SwitchToUser(switchOpts SwitchOpts) (vector Vector) { // Sanitize registers. regs := switchOpts.Registers - regs.Pstate &= ^uint64(UserFlagsClear) + regs.Pstate &= ^uint64(PsrFlagsClear) regs.Pstate |= UserFlagsSet SetTLS(regs.TPIDR_EL0) -- cgit v1.2.3 From bb25c9611ba9cb12ab42b08a3a91a095b6f7c824 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Thu, 30 Jul 2020 03:42:46 -0400 Subject: add usr-tls test cases for Arm64 Signed-off-by: Bin Lu --- pkg/sentry/platform/kvm/testutil/testutil_arm64.go | 4 ++++ pkg/sentry/platform/kvm/testutil/testutil_arm64.s | 14 ++++++++++---- 2 files changed, 14 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/platform/kvm/testutil/testutil_arm64.go b/pkg/sentry/platform/kvm/testutil/testutil_arm64.go index ca902c8c1..4dad877ba 100644 --- a/pkg/sentry/platform/kvm/testutil/testutil_arm64.go +++ b/pkg/sentry/platform/kvm/testutil/testutil_arm64.go @@ -56,5 +56,9 @@ func CheckTestRegs(regs *arch.Registers, full bool) (err error) { err = addRegisterMismatch(err, fmt.Sprintf("R%d", i), regs.Regs[i], need) } } + // Check tls. + if need := ^uint64(11); regs.TPIDR_EL0 != need { + err = addRegisterMismatch(err, "tpdir_el0", regs.TPIDR_EL0, need) + } return } diff --git a/pkg/sentry/platform/kvm/testutil/testutil_arm64.s b/pkg/sentry/platform/kvm/testutil/testutil_arm64.s index 07658144e..6caf7282d 100644 --- a/pkg/sentry/platform/kvm/testutil/testutil_arm64.s +++ b/pkg/sentry/platform/kvm/testutil/testutil_arm64.s @@ -52,6 +52,8 @@ start: TEXT ·FloatingPointWorks(SB),NOSPLIT,$0-8 NO_LOCAL_POINTERS + // gc will touch fpsimd, so we should test it. + // such as in . FMOVD $(9.9), F0 MOVD $SYS_GETPID, R8 // getpid SVC @@ -102,11 +104,15 @@ isNaN: TEXT ·TwiddleRegsSyscall(SB),NOSPLIT,$0 TWIDDLE_REGS() + MSR R10, TPIDR_EL0 + // Trapped in el0_svc. SVC RET // never reached TEXT ·TwiddleRegsFault(SB),NOSPLIT,$0 - TWIDDLE_REGS() - // Branch to Register branches unconditionally to an address in . - JMP (R4) // <=> br x4, must fault - RET // never reached + TWIDDLE_REGS() + MSR R10, TPIDR_EL0 + // Trapped in el0_ia. + // Branch to Register branches unconditionally to an address in . + JMP (R6) // <=> br x6, must fault + RET // never reached -- cgit v1.2.3 From c43305731e04ce8d4f2c9b2949c326abc9c2b77e Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Thu, 30 Jul 2020 10:21:12 -0700 Subject: Fix SETOWN_EX return value. Return on success should be 0, not size of the struct copied out. PiperOrigin-RevId: 324029193 --- pkg/sentry/syscalls/linux/sys_file.go | 8 ++++---- pkg/sentry/syscalls/linux/vfs2/fd.go | 4 ++-- test/syscalls/linux/fcntl.cc | 38 ++++++++++++++++++++--------------- 3 files changed, 28 insertions(+), 22 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go index 2797c6a72..8cf6401e7 100644 --- a/pkg/sentry/syscalls/linux/sys_file.go +++ b/pkg/sentry/syscalls/linux/sys_file.go @@ -1057,7 +1057,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall case linux.F_SETOWN_EX: addr := args[2].Pointer() var owner linux.FOwnerEx - n, err := t.CopyIn(addr, &owner) + _, err := t.CopyIn(addr, &owner) if err != nil { return 0, nil, err } @@ -1069,21 +1069,21 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return 0, nil, syserror.ESRCH } a.SetOwnerTask(t, task) - return uintptr(n), nil, nil + return 0, nil, nil case linux.F_OWNER_PID: tg := t.PIDNamespace().ThreadGroupWithID(kernel.ThreadID(owner.PID)) if tg == nil { return 0, nil, syserror.ESRCH } a.SetOwnerThreadGroup(t, tg) - return uintptr(n), nil, nil + return 0, nil, nil case linux.F_OWNER_PGRP: pg := t.PIDNamespace().ProcessGroupWithID(kernel.ProcessGroupID(owner.PID)) if pg == nil { return 0, nil, syserror.ESRCH } a.SetOwnerProcessGroup(t, pg) - return uintptr(n), nil, nil + return 0, nil, nil default: return 0, nil, syserror.EINVAL } diff --git a/pkg/sentry/syscalls/linux/vfs2/fd.go b/pkg/sentry/syscalls/linux/vfs2/fd.go index 517394ba9..67f191551 100644 --- a/pkg/sentry/syscalls/linux/vfs2/fd.go +++ b/pkg/sentry/syscalls/linux/vfs2/fd.go @@ -185,11 +185,11 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return 0, nil, err case linux.F_SETOWN_EX: var owner linux.FOwnerEx - n, err := t.CopyIn(args[2].Pointer(), &owner) + _, err := t.CopyIn(args[2].Pointer(), &owner) if err != nil { return 0, nil, err } - return uintptr(n), nil, setAsyncOwner(t, file, owner.Type, owner.PID) + return 0, nil, setAsyncOwner(t, file, owner.Type, owner.PID) case linux.F_GETPIPE_SZ: pipefile, ok := file.Impl().(*pipe.VFSPipeFD) if !ok { diff --git a/test/syscalls/linux/fcntl.cc b/test/syscalls/linux/fcntl.cc index 5467fa2c8..34016d4bd 100644 --- a/test/syscalls/linux/fcntl.cc +++ b/test/syscalls/linux/fcntl.cc @@ -1004,7 +1004,8 @@ TEST(FcntlTest, SetOwnPid) { pid_t pid; EXPECT_THAT(pid = getpid(), SyscallSucceeds()); - ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, pid), SyscallSucceeds()); + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, pid), + SyscallSucceedsWithValue(0)); EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN), SyscallSucceedsWithValue(pid)); @@ -1018,7 +1019,8 @@ TEST(FcntlTest, SetOwnPgrp) { pid_t pgid; EXPECT_THAT(pgid = getpgrp(), SyscallSucceeds()); - ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, -pgid), SyscallSucceeds()); + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, -pgid), + SyscallSucceedsWithValue(0)); // Verify with F_GETOWN_EX; using F_GETOWN on Linux may incorrectly treat the // negative return value as an error, converting the return value to -1 and @@ -1038,8 +1040,10 @@ TEST(FcntlTest, SetOwnUnset) { // Set and unset pid. pid_t pid; EXPECT_THAT(pid = getpid(), SyscallSucceeds()); - ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, pid), SyscallSucceeds()); - ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, 0), SyscallSucceeds()); + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, pid), + SyscallSucceedsWithValue(0)); + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, 0), + SyscallSucceedsWithValue(0)); EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN), SyscallSucceedsWithValue(0)); @@ -1047,8 +1051,10 @@ TEST(FcntlTest, SetOwnUnset) { // Set and unset pgid. pid_t pgid; EXPECT_THAT(pgid = getpgrp(), SyscallSucceeds()); - ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, -pgid), SyscallSucceeds()); - ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, 0), SyscallSucceeds()); + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, -pgid), + SyscallSucceedsWithValue(0)); + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, 0), + SyscallSucceedsWithValue(0)); EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN), SyscallSucceedsWithValue(0)); @@ -1120,7 +1126,7 @@ TEST(FcntlTest, SetOwnExTid) { EXPECT_THAT(owner.pid = syscall(__NR_gettid), SyscallSucceeds()); ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), - SyscallSucceeds()); + SyscallSucceedsWithValue(0)); EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN), SyscallSucceedsWithValue(owner.pid)); @@ -1136,7 +1142,7 @@ TEST(FcntlTest, SetOwnExPid) { EXPECT_THAT(owner.pid = getpid(), SyscallSucceeds()); ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), - SyscallSucceeds()); + SyscallSucceedsWithValue(0)); EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN), SyscallSucceedsWithValue(owner.pid)); @@ -1152,7 +1158,7 @@ TEST(FcntlTest, SetOwnExPgrp) { EXPECT_THAT(set_owner.pid = getpgrp(), SyscallSucceeds()); ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &set_owner), - SyscallSucceeds()); + SyscallSucceedsWithValue(0)); // Verify with F_GETOWN_EX; using F_GETOWN on Linux may incorrectly treat the // negative return value as an error, converting the return value to -1 and @@ -1176,10 +1182,10 @@ TEST(FcntlTest, SetOwnExUnset) { owner.type = F_OWNER_PID; EXPECT_THAT(owner.pid = getpid(), SyscallSucceeds()); ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), - SyscallSucceeds()); + SyscallSucceedsWithValue(0)); owner.pid = 0; ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), - SyscallSucceeds()); + SyscallSucceedsWithValue(0)); EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN), SyscallSucceedsWithValue(0)); @@ -1188,10 +1194,10 @@ TEST(FcntlTest, SetOwnExUnset) { owner.type = F_OWNER_PGRP; EXPECT_THAT(owner.pid = getpgrp(), SyscallSucceeds()); ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), - SyscallSucceeds()); + SyscallSucceedsWithValue(0)); owner.pid = 0; ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), - SyscallSucceeds()); + SyscallSucceedsWithValue(0)); EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN), SyscallSucceedsWithValue(0)); @@ -1207,7 +1213,7 @@ TEST(FcntlTest, GetOwnExTid) { EXPECT_THAT(set_owner.pid = syscall(__NR_gettid), SyscallSucceeds()); ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &set_owner), - SyscallSucceeds()); + SyscallSucceedsWithValue(0)); f_owner_ex got_owner = {}; ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN_EX, &got_owner), @@ -1225,7 +1231,7 @@ TEST(FcntlTest, GetOwnExPid) { EXPECT_THAT(set_owner.pid = getpid(), SyscallSucceeds()); ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &set_owner), - SyscallSucceeds()); + SyscallSucceedsWithValue(0)); f_owner_ex got_owner = {}; ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN_EX, &got_owner), @@ -1243,7 +1249,7 @@ TEST(FcntlTest, GetOwnExPgrp) { EXPECT_THAT(set_owner.pid = getpgrp(), SyscallSucceeds()); ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &set_owner), - SyscallSucceeds()); + SyscallSucceedsWithValue(0)); f_owner_ex got_owner = {}; ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN_EX, &got_owner), -- cgit v1.2.3 From 68a7da9549ac4f6290fe85d2dab550e11f16b209 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Fri, 31 Jul 2020 00:38:57 -0700 Subject: Clean up vfs2 fallocate. Move to setstat.go and add a FileDescription wrapper method. PiperOrigin-RevId: 324165277 --- pkg/sentry/syscalls/linux/vfs2/filesystem.go | 50 ---------------------------- pkg/sentry/syscalls/linux/vfs2/setstat.go | 50 ++++++++++++++++++++++++++++ pkg/sentry/vfs/file_description.go | 7 +++- 3 files changed, 56 insertions(+), 51 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/vfs2/filesystem.go b/pkg/sentry/syscalls/linux/vfs2/filesystem.go index 6b14c2bef..b6d2ddd65 100644 --- a/pkg/sentry/syscalls/linux/vfs2/filesystem.go +++ b/pkg/sentry/syscalls/linux/vfs2/filesystem.go @@ -18,7 +18,6 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -245,55 +244,6 @@ func renameat(t *kernel.Task, olddirfd int32, oldpathAddr usermem.Addr, newdirfd }) } -// Fallocate implements linux system call fallocate(2). -func Fallocate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := args[0].Int() - mode := args[1].Uint64() - offset := args[2].Int64() - length := args[3].Int64() - - file := t.GetFileVFS2(fd) - - if file == nil { - return 0, nil, syserror.EBADF - } - defer file.DecRef() - - if !file.IsWritable() { - return 0, nil, syserror.EBADF - } - - if mode != 0 { - return 0, nil, syserror.ENOTSUP - } - - if offset < 0 || length <= 0 { - return 0, nil, syserror.EINVAL - } - - size := offset + length - - if size < 0 { - return 0, nil, syserror.EFBIG - } - - limit := limits.FromContext(t).Get(limits.FileSize).Cur - - if uint64(size) >= limit { - t.SendSignal(&arch.SignalInfo{ - Signo: int32(linux.SIGXFSZ), - Code: arch.SignalInfoUser, - }) - return 0, nil, syserror.EFBIG - } - - return 0, nil, file.Impl().Allocate(t, mode, uint64(offset), uint64(length)) - - // File length modified, generate notification. - // TODO(gvisor.dev/issue/1479): Reenable when Inotify is ported. - // file.Dirent.InotifyEvent(linux.IN_MODIFY, 0) -} - // Rmdir implements Linux syscall rmdir(2). func Rmdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { pathAddr := args[0].Pointer() diff --git a/pkg/sentry/syscalls/linux/vfs2/setstat.go b/pkg/sentry/syscalls/linux/vfs2/setstat.go index 6daedd173..37fa56c19 100644 --- a/pkg/sentry/syscalls/linux/vfs2/setstat.go +++ b/pkg/sentry/syscalls/linux/vfs2/setstat.go @@ -20,6 +20,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -211,6 +212,55 @@ func Ftruncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys return 0, nil, handleSetSizeError(t, err) } +// Fallocate implements linux system call fallocate(2). +func Fallocate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + mode := args[1].Uint64() + offset := args[2].Int64() + length := args[3].Int64() + + file := t.GetFileVFS2(fd) + + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + if !file.IsWritable() { + return 0, nil, syserror.EBADF + } + + if mode != 0 { + return 0, nil, syserror.ENOTSUP + } + + if offset < 0 || length <= 0 { + return 0, nil, syserror.EINVAL + } + + size := offset + length + + if size < 0 { + return 0, nil, syserror.EFBIG + } + + limit := limits.FromContext(t).Get(limits.FileSize).Cur + + if uint64(size) >= limit { + t.SendSignal(&arch.SignalInfo{ + Signo: int32(linux.SIGXFSZ), + Code: arch.SignalInfoUser, + }) + return 0, nil, syserror.EFBIG + } + + return 0, nil, file.Allocate(t, mode, uint64(offset), uint64(length)) + + // File length modified, generate notification. + // TODO(gvisor.dev/issue/1479): Reenable when Inotify is ported. + // file.Dirent.InotifyEvent(linux.IN_MODIFY, 0) +} + // Utime implements Linux syscall utime(2). func Utime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { pathAddr := args[0].Pointer() diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 0c42574db..93861fb4a 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -354,7 +354,7 @@ type FileDescriptionImpl interface { // represented by the FileDescription. StatFS(ctx context.Context) (linux.Statfs, error) - // Allocate grows file represented by FileDescription to offset + length bytes. + // Allocate grows the file to offset + length bytes. // Only mode == 0 is supported currently. Allocate(ctx context.Context, mode, offset, length uint64) error @@ -563,6 +563,11 @@ func (fd *FileDescription) StatFS(ctx context.Context) (linux.Statfs, error) { return fd.impl.StatFS(ctx) } +// Allocate grows file represented by FileDescription to offset + length bytes. +func (fd *FileDescription) Allocate(ctx context.Context, mode, offset, length uint64) error { + return fd.impl.Allocate(ctx, mode, offset, length) +} + // Readiness implements waiter.Waitable.Readiness. // // It returns fd's I/O readiness. -- cgit v1.2.3 From 2a7b2a61e3ea32129c26eeaa6fab3d81a5d8ad6e Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Thu, 11 Jun 2020 20:33:56 -0700 Subject: iptables: support SO_ORIGINAL_DST Envoy (#170) uses this to get the original destination of redirected packets. --- pkg/abi/linux/netfilter.go | 8 +- pkg/abi/linux/socket.go | 4 +- pkg/sentry/socket/netstack/netstack.go | 17 ++++ pkg/sentry/strace/socket.go | 1 + pkg/tcpip/stack/conntrack.go | 26 ++++++ pkg/tcpip/stack/iptables.go | 11 ++- pkg/tcpip/tcpip.go | 4 + pkg/tcpip/transport/tcp/endpoint.go | 11 +++ test/iptables/BUILD | 1 + test/iptables/iptables_test.go | 8 ++ test/iptables/iptables_unsafe.go | 63 ++++++++++++++ test/iptables/iptables_util.go | 51 ++++++++++- test/iptables/nat.go | 152 ++++++++++++++++++++++++++++++++- 13 files changed, 344 insertions(+), 13 deletions(-) create mode 100644 test/iptables/iptables_unsafe.go (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/netfilter.go b/pkg/abi/linux/netfilter.go index a91f9f018..9c27f7bb2 100644 --- a/pkg/abi/linux/netfilter.go +++ b/pkg/abi/linux/netfilter.go @@ -59,7 +59,7 @@ var VerdictStrings = map[int32]string{ NF_RETURN: "RETURN", } -// Socket options. These correspond to values in +// Socket options for SOL_SOCKET. These correspond to values in // include/uapi/linux/netfilter_ipv4/ip_tables.h. const ( IPT_BASE_CTL = 64 @@ -74,6 +74,12 @@ const ( IPT_SO_GET_MAX = IPT_SO_GET_REVISION_TARGET ) +// Socket option for SOL_IP. This corresponds to the value in +// include/uapi/linux/netfilter_ipv4.h. +const ( + SO_ORIGINAL_DST = 80 +) + // Name lengths. These correspond to values in // include/uapi/linux/netfilter/x_tables.h. const ( diff --git a/pkg/abi/linux/socket.go b/pkg/abi/linux/socket.go index c24a8216e..d6946bb82 100644 --- a/pkg/abi/linux/socket.go +++ b/pkg/abi/linux/socket.go @@ -239,11 +239,13 @@ const SockAddrMax = 128 type InetAddr [4]byte // SockAddrInet is struct sockaddr_in, from uapi/linux/in.h. +// +// +marshal type SockAddrInet struct { Family uint16 Port uint16 Addr InetAddr - Zero [8]uint8 // pad to sizeof(struct sockaddr). + _ [8]uint8 // pad to sizeof(struct sockaddr). } // InetMulticastRequest is struct ip_mreq, from uapi/linux/in.h. diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index f86e6cd7a..31a168f7e 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -1490,6 +1490,10 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (marsha vP := primitive.Int32(boolToInt32(v)) return &vP, nil + case linux.SO_ORIGINAL_DST: + // TODO(gvisor.dev/issue/170): ip6tables. + return nil, syserr.ErrInvalidArgument + default: emitUnimplementedEventIPv6(t, name) } @@ -1600,6 +1604,19 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in vP := primitive.Int32(boolToInt32(v)) return &vP, nil + case linux.SO_ORIGINAL_DST: + if outLen < int(binary.Size(linux.SockAddrInet{})) { + return nil, syserr.ErrInvalidArgument + } + + var v tcpip.OriginalDestinationOption + if err := ep.GetSockOpt(&v); err != nil { + return nil, syserr.TranslateNetstackError(err) + } + + a, _ := ConvertAddress(linux.AF_INET, tcpip.FullAddress(v)) + return a.(*linux.SockAddrInet), nil + default: emitUnimplementedEventIP(t, name) } diff --git a/pkg/sentry/strace/socket.go b/pkg/sentry/strace/socket.go index c0512de89..b51c4c941 100644 --- a/pkg/sentry/strace/socket.go +++ b/pkg/sentry/strace/socket.go @@ -521,6 +521,7 @@ var sockOptNames = map[uint64]abi.ValueSet{ linux.IP_ROUTER_ALERT: "IP_ROUTER_ALERT", linux.IP_PKTOPTIONS: "IP_PKTOPTIONS", linux.IP_MTU: "IP_MTU", + linux.SO_ORIGINAL_DST: "SO_ORIGINAL_DST", }, linux.SOL_SOCKET: { linux.SO_ERROR: "SO_ERROR", diff --git a/pkg/tcpip/stack/conntrack.go b/pkg/tcpip/stack/conntrack.go index 559a1c4dd..470c265aa 100644 --- a/pkg/tcpip/stack/conntrack.go +++ b/pkg/tcpip/stack/conntrack.go @@ -240,7 +240,10 @@ func (ct *ConnTrack) connFor(pkt *PacketBuffer) (*conn, direction) { if err != nil { return nil, dirOriginal } + return ct.connForTID(tid) +} +func (ct *ConnTrack) connForTID(tid tupleID) (*conn, direction) { bucket := ct.bucket(tid) now := time.Now() @@ -604,3 +607,26 @@ func (ct *ConnTrack) reapTupleLocked(tuple *tuple, bucket int, now time.Time) bo return true } + +func (ct *ConnTrack) originalDst(epID TransportEndpointID) (tcpip.Address, uint16, *tcpip.Error) { + // Lookup the connection. The reply's original destination + // describes the original address. + tid := tupleID{ + srcAddr: epID.LocalAddress, + srcPort: epID.LocalPort, + dstAddr: epID.RemoteAddress, + dstPort: epID.RemotePort, + transProto: header.TCPProtocolNumber, + netProto: header.IPv4ProtocolNumber, + } + conn, _ := ct.connForTID(tid) + if conn == nil { + // Not a tracked connection. + return "", 0, tcpip.ErrNotConnected + } else if conn.manip == manipNone { + // Unmanipulated connection. + return "", 0, tcpip.ErrInvalidOptionValue + } + + return conn.original.dstAddr, conn.original.dstPort, nil +} diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go index cbbae4224..110ba073d 100644 --- a/pkg/tcpip/stack/iptables.go +++ b/pkg/tcpip/stack/iptables.go @@ -218,19 +218,16 @@ func (it *IPTables) Check(hook Hook, pkt *PacketBuffer, gso *GSO, r *Route, addr // Many users never configure iptables. Spare them the cost of rule // traversal if rules have never been set. it.mu.RLock() + defer it.mu.RUnlock() if !it.modified { - it.mu.RUnlock() return true } - it.mu.RUnlock() // Packets are manipulated only if connection and matching // NAT rule exists. shouldTrack := it.connections.handlePacket(pkt, hook, gso, r) // Go through each table containing the hook. - it.mu.RLock() - defer it.mu.RUnlock() priorities := it.priorities[hook] for _, tableID := range priorities { // If handlePacket already NATed the packet, we don't need to @@ -418,3 +415,9 @@ func (it *IPTables) checkRule(hook Hook, pkt *PacketBuffer, table Table, ruleIdx // All the matchers matched, so run the target. return rule.Target.Action(pkt, &it.connections, hook, gso, r, address) } + +// OriginalDst returns the original destination of redirected connections. It +// returns an error if the connection doesn't exist or isn't redirected. +func (it *IPTables) OriginalDst(epID TransportEndpointID) (tcpip.Address, uint16, *tcpip.Error) { + return it.connections.originalDst(epID) +} diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index a634b9b60..45f59b60f 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -954,6 +954,10 @@ type DefaultTTLOption uint8 // classic BPF filter on a given endpoint. type SocketDetachFilterOption int +// OriginalDestinationOption is used to get the original destination address +// and port of a redirected packet. +type OriginalDestinationOption FullAddress + // IPPacketInfo is the message structure for IP_PKTINFO. // // +stateify savable diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 0f7487963..682687ebe 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -2017,6 +2017,17 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { *o = tcpip.TCPDeferAcceptOption(e.deferAccept) e.UnlockUser() + case *tcpip.OriginalDestinationOption: + ipt := e.stack.IPTables() + addr, port, err := ipt.OriginalDst(e.ID) + if err != nil { + return err + } + *o = tcpip.OriginalDestinationOption{ + Addr: addr, + Port: port, + } + default: return tcpip.ErrUnknownProtocolOption } diff --git a/test/iptables/BUILD b/test/iptables/BUILD index 40b63ebbe..66453772a 100644 --- a/test/iptables/BUILD +++ b/test/iptables/BUILD @@ -9,6 +9,7 @@ go_library( "filter_input.go", "filter_output.go", "iptables.go", + "iptables_unsafe.go", "iptables_util.go", "nat.go", ], diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go index 550b6198a..fda5f694f 100644 --- a/test/iptables/iptables_test.go +++ b/test/iptables/iptables_test.go @@ -371,3 +371,11 @@ func TestFilterAddrs(t *testing.T) { } } } + +func TestNATPreOriginalDst(t *testing.T) { + singleTest(t, NATPreOriginalDst{}) +} + +func TestNATOutOriginalDst(t *testing.T) { + singleTest(t, NATOutOriginalDst{}) +} diff --git a/test/iptables/iptables_unsafe.go b/test/iptables/iptables_unsafe.go new file mode 100644 index 000000000..bd85a8fea --- /dev/null +++ b/test/iptables/iptables_unsafe.go @@ -0,0 +1,63 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package iptables + +import ( + "fmt" + "syscall" + "unsafe" +) + +type originalDstError struct { + errno syscall.Errno +} + +func (e originalDstError) Error() string { + return fmt.Sprintf("errno (%d) when calling getsockopt(SO_ORIGINAL_DST): %v", int(e.errno), e.errno.Error()) +} + +// SO_ORIGINAL_DST gets the original destination of a redirected packet via +// getsockopt. +const SO_ORIGINAL_DST = 80 + +func originalDestination4(connfd int) (syscall.RawSockaddrInet4, error) { + var addr syscall.RawSockaddrInet4 + var addrLen uint32 = syscall.SizeofSockaddrInet4 + if errno := originalDestination(connfd, syscall.SOL_IP, unsafe.Pointer(&addr), &addrLen); errno != 0 { + return syscall.RawSockaddrInet4{}, originalDstError{errno} + } + return addr, nil +} + +func originalDestination6(connfd int) (syscall.RawSockaddrInet6, error) { + var addr syscall.RawSockaddrInet6 + var addrLen uint32 = syscall.SizeofSockaddrInet6 + if errno := originalDestination(connfd, syscall.SOL_IPV6, unsafe.Pointer(&addr), &addrLen); errno != 0 { + return syscall.RawSockaddrInet6{}, originalDstError{errno} + } + return addr, nil +} + +func originalDestination(connfd int, level uintptr, optval unsafe.Pointer, optlen *uint32) syscall.Errno { + _, _, errno := syscall.Syscall6( + syscall.SYS_GETSOCKOPT, + uintptr(connfd), + level, + SO_ORIGINAL_DST, + uintptr(optval), + uintptr(unsafe.Pointer(optlen)), + 0) + return errno +} diff --git a/test/iptables/iptables_util.go b/test/iptables/iptables_util.go index ca80a4b5f..5125fe47b 100644 --- a/test/iptables/iptables_util.go +++ b/test/iptables/iptables_util.go @@ -15,6 +15,8 @@ package iptables import ( + "encoding/binary" + "errors" "fmt" "net" "os/exec" @@ -218,17 +220,58 @@ func filterAddrs(addrs []string, ipv6 bool) []string { // getInterfaceName returns the name of the interface other than loopback. func getInterfaceName() (string, bool) { - var ifname string + iface, ok := getNonLoopbackInterface() + if !ok { + return "", false + } + return iface.Name, true +} + +func getInterfaceAddrs(ipv6 bool) ([]net.IP, error) { + iface, ok := getNonLoopbackInterface() + if !ok { + return nil, errors.New("no non-loopback interface found") + } + addrs, err := iface.Addrs() + if err != nil { + return nil, err + } + + // Get only IPv4 or IPv6 addresses. + ips := make([]net.IP, 0, len(addrs)) + for _, addr := range addrs { + parts := strings.Split(addr.String(), "/") + var ip net.IP + // To16() returns IPv4 addresses as IPv4-mapped IPv6 addresses. + // So we check whether To4() returns nil to test whether the + // address is v4 or v6. + if v4 := net.ParseIP(parts[0]).To4(); ipv6 && v4 == nil { + ip = net.ParseIP(parts[0]).To16() + } else { + ip = v4 + } + if ip != nil { + ips = append(ips, ip) + } + } + return ips, nil +} + +func getNonLoopbackInterface() (net.Interface, bool) { if interfaces, err := net.Interfaces(); err == nil { for _, intf := range interfaces { if intf.Name != "lo" { - ifname = intf.Name - break + return intf, true } } } + return net.Interface{}, false +} - return ifname, ifname != "" +func htons(x uint16) uint16 { + buf := make([]byte, 2) + binary.BigEndian.PutUint16(buf, x) + return binary.LittleEndian.Uint16(buf) } func localIP(ipv6 bool) string { diff --git a/test/iptables/nat.go b/test/iptables/nat.go index ac0d91bb2..b7fea2527 100644 --- a/test/iptables/nat.go +++ b/test/iptables/nat.go @@ -18,12 +18,11 @@ import ( "errors" "fmt" "net" + "syscall" "time" ) -const ( - redirectPort = 42 -) +const redirectPort = 42 func init() { RegisterTestCase(NATPreRedirectUDPPort{}) @@ -42,6 +41,8 @@ func init() { RegisterTestCase(NATOutRedirectInvert{}) RegisterTestCase(NATRedirectRequiresProtocol{}) RegisterTestCase(NATLoopbackSkipsPrerouting{}) + RegisterTestCase(NATPreOriginalDst{}) + RegisterTestCase(NATOutOriginalDst{}) } // NATPreRedirectUDPPort tests that packets are redirected to different port. @@ -471,6 +472,151 @@ func (NATLoopbackSkipsPrerouting) LocalAction(ip net.IP, ipv6 bool) error { return nil } +// NATPreOriginalDst tests that SO_ORIGINAL_DST returns the pre-NAT destination +// of PREROUTING NATted packets. +type NATPreOriginalDst struct{} + +// Name implements TestCase.Name. +func (NATPreOriginalDst) Name() string { + return "NATPreOriginalDst" +} + +// ContainerAction implements TestCase.ContainerAction. +func (NATPreOriginalDst) ContainerAction(ip net.IP, ipv6 bool) error { + // Redirect incoming TCP connections to acceptPort. + if err := natTable(ipv6, "-A", "PREROUTING", + "-p", "tcp", + "--destination-port", fmt.Sprintf("%d", dropPort), + "-j", "REDIRECT", "--to-port", fmt.Sprintf("%d", acceptPort)); err != nil { + return err + } + + addrs, err := getInterfaceAddrs(ipv6) + if err != nil { + return err + } + return listenForRedirectedConn(ipv6, addrs) +} + +// LocalAction implements TestCase.LocalAction. +func (NATPreOriginalDst) LocalAction(ip net.IP, ipv6 bool) error { + return connectTCP(ip, dropPort, sendloopDuration) +} + +// NATOutOriginalDst tests that SO_ORIGINAL_DST returns the pre-NAT destination +// of OUTBOUND NATted packets. +type NATOutOriginalDst struct{} + +// Name implements TestCase.Name. +func (NATOutOriginalDst) Name() string { + return "NATOutOriginalDst" +} + +// ContainerAction implements TestCase.ContainerAction. +func (NATOutOriginalDst) ContainerAction(ip net.IP, ipv6 bool) error { + // Redirect incoming TCP connections to acceptPort. + if err := natTable(ipv6, "-A", "OUTPUT", "-p", "tcp", "-j", "REDIRECT", "--to-port", fmt.Sprintf("%d", acceptPort)); err != nil { + return err + } + + connCh := make(chan error) + go func() { + connCh <- connectTCP(ip, dropPort, sendloopDuration) + }() + + if err := listenForRedirectedConn(ipv6, []net.IP{ip}); err != nil { + return err + } + return <-connCh +} + +// LocalAction implements TestCase.LocalAction. +func (NATOutOriginalDst) LocalAction(ip net.IP, ipv6 bool) error { + // No-op. + return nil +} + +func listenForRedirectedConn(ipv6 bool, originalDsts []net.IP) error { + // The net package doesn't give guarantee access to the connection's + // underlying FD, and thus we cannot call getsockopt. We have to use + // traditional syscalls for SO_ORIGINAL_DST. + + // Create the listening socket, bind, listen, and accept. + family := syscall.AF_INET + if ipv6 { + family = syscall.AF_INET6 + } + sockfd, err := syscall.Socket(family, syscall.SOCK_STREAM, 0) + if err != nil { + return err + } + defer syscall.Close(sockfd) + + var bindAddr syscall.Sockaddr + if ipv6 { + bindAddr = &syscall.SockaddrInet6{ + Port: acceptPort, + Addr: [16]byte{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}, // in6addr_any + } + } else { + bindAddr = &syscall.SockaddrInet4{ + Port: acceptPort, + Addr: [4]byte{0, 0, 0, 0}, // INADDR_ANY + } + } + if err := syscall.Bind(sockfd, bindAddr); err != nil { + return err + } + + if err := syscall.Listen(sockfd, 1); err != nil { + return err + } + + connfd, _, err := syscall.Accept(sockfd) + if err != nil { + return err + } + defer syscall.Close(connfd) + + // Verify that, despite listening on acceptPort, SO_ORIGINAL_DST + // indicates the packet was sent to originalDst:dropPort. + if ipv6 { + got, err := originalDestination6(connfd) + if err != nil { + return err + } + // The original destination could be any of our IPs. + for _, dst := range originalDsts { + want := syscall.RawSockaddrInet6{ + Family: syscall.AF_INET6, + Port: htons(dropPort), + } + copy(want.Addr[:], dst.To16()) + if got == want { + return nil + } + } + return fmt.Errorf("SO_ORIGINAL_DST returned %+v, but wanted one of %+v (note: port numbers are in network byte order)", got, originalDsts) + } else { + got, err := originalDestination4(connfd) + if err != nil { + return err + } + // The original destination could be any of our IPs. + for _, dst := range originalDsts { + want := syscall.RawSockaddrInet4{ + Family: syscall.AF_INET, + Port: htons(dropPort), + } + copy(want.Addr[:], dst.To4()) + if got == want { + return nil + } + } + return fmt.Errorf("SO_ORIGINAL_DST returned %+v, but wanted one of %+v (note: port numbers are in network byte order)", got, originalDsts) + } +} + // loopbackTests runs an iptables rule and ensures that packets sent to // dest:dropPort are received by localhost:acceptPort. func loopbackTest(ipv6 bool, dest net.IP, args ...string) error { -- cgit v1.2.3 From 8908baaf79fdd137241596fd2444828d1a33fe27 Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Fri, 31 Jul 2020 12:23:01 -0700 Subject: Internal change. PiperOrigin-RevId: 324259991 --- pkg/sentry/fdimport/fdimport.go | 5 +++++ 1 file changed, 5 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fdimport/fdimport.go b/pkg/sentry/fdimport/fdimport.go index a4199f9e9..b8686adb4 100644 --- a/pkg/sentry/fdimport/fdimport.go +++ b/pkg/sentry/fdimport/fdimport.go @@ -15,6 +15,8 @@ package fdimport import ( + "fmt" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/host" @@ -84,6 +86,9 @@ func importFS(ctx context.Context, fdTable *kernel.FDTable, console bool, fds [] func importVFS2(ctx context.Context, fdTable *kernel.FDTable, console bool, stdioFDs []int) (*hostvfs2.TTYFileDescription, error) { k := kernel.KernelFromContext(ctx) + if k == nil { + return nil, fmt.Errorf("cannot find kernel from context") + } var ttyFile *vfs.FileDescription for appFD, hostFD := range stdioFDs { -- cgit v1.2.3 From 1fbbc795ef8918006214032d96fa88b8a21e5e0a Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Mon, 3 Aug 2020 12:30:23 -0700 Subject: Add inotify events for fallocate and tests for fallocate/sendfile. Updates #1479, #2923. PiperOrigin-RevId: 324658826 --- pkg/sentry/syscalls/linux/vfs2/setstat.go | 9 ++--- test/syscalls/linux/inotify.cc | 55 +++++++++++++++++++++++++++++++ 2 files changed, 60 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/vfs2/setstat.go b/pkg/sentry/syscalls/linux/vfs2/setstat.go index 37fa56c19..25cdb7a55 100644 --- a/pkg/sentry/syscalls/linux/vfs2/setstat.go +++ b/pkg/sentry/syscalls/linux/vfs2/setstat.go @@ -254,11 +254,12 @@ func Fallocate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys return 0, nil, syserror.EFBIG } - return 0, nil, file.Allocate(t, mode, uint64(offset), uint64(length)) + if err := file.Allocate(t, mode, uint64(offset), uint64(length)); err != nil { + return 0, nil, err + } - // File length modified, generate notification. - // TODO(gvisor.dev/issue/1479): Reenable when Inotify is ported. - // file.Dirent.InotifyEvent(linux.IN_MODIFY, 0) + file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) + return 0, nil, nil } // Utime implements Linux syscall utime(2). diff --git a/test/syscalls/linux/inotify.cc b/test/syscalls/linux/inotify.cc index 220874aeb..6cf398097 100644 --- a/test/syscalls/linux/inotify.cc +++ b/test/syscalls/linux/inotify.cc @@ -18,6 +18,7 @@ #include #include #include +#include #include #include @@ -1681,6 +1682,60 @@ TEST(Inotify, EpollNoDeadlock) { } } +TEST(Inotify, Fallocate) { + const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(file.path(), O_RDWR)); + + const FileDescriptor inotify_fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + const int wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(inotify_fd.get(), file.path(), IN_ALL_EVENTS)); + + // Do an arbitrary modification with fallocate. + ASSERT_THAT(fallocate(fd.get(), 0, 0, 123), SyscallSucceeds()); + std::vector events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({Event(IN_MODIFY, wd)})); +} + +TEST(Inotify, Sendfile) { + SKIP_IF(IsRunningWithVFS1()); + + const TempPath root = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const TempPath in_file = ASSERT_NO_ERRNO_AND_VALUE( + TempPath::CreateFileWith(root.path(), "x", 0644)); + const TempPath out_file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); + const FileDescriptor in = + ASSERT_NO_ERRNO_AND_VALUE(Open(in_file.path(), O_RDONLY)); + const FileDescriptor out = + ASSERT_NO_ERRNO_AND_VALUE(Open(out_file.path(), O_WRONLY)); + + // Create separate inotify instances for the in and out fds. If both watches + // were on the same instance, we would have discrepancies between Linux and + // gVisor (order of events, duplicate events), which is not that important + // since inotify is asynchronous anyway. + const FileDescriptor in_inotify = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + const FileDescriptor out_inotify = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + const int in_wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(in_inotify.get(), in_file.path(), IN_ALL_EVENTS)); + const int out_wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(out_inotify.get(), out_file.path(), IN_ALL_EVENTS)); + + ASSERT_THAT(sendfile(out.get(), in.get(), /*offset=*/nullptr, 1), + SyscallSucceeds()); + + // Expect a single access event and a single modify event. + std::vector in_events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(in_inotify.get())); + std::vector out_events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(out_inotify.get())); + EXPECT_THAT(in_events, Are({Event(IN_ACCESS, in_wd)})); + EXPECT_THAT(out_events, Are({Event(IN_MODIFY, out_wd)})); +} + // On Linux, inotify behavior is not very consistent with splice(2). We try our // best to emulate Linux for very basic calls to splice. TEST(Inotify, SpliceOnWatchTarget) { -- cgit v1.2.3 From b2ae7ea1bb207eddadd7962080e7bd0b8634db96 Mon Sep 17 00:00:00 2001 From: Nayana Bidari Date: Mon, 3 Aug 2020 13:33:47 -0700 Subject: Plumbing context.Context to DecRef() and Release(). context is passed to DecRef() and Release() which is needed for SO_LINGER implementation. PiperOrigin-RevId: 324672584 --- pkg/refs/BUILD | 6 +- pkg/refs/refcounter.go | 19 +-- pkg/refs/refcounter_test.go | 38 ++--- pkg/sentry/control/proc.go | 2 +- pkg/sentry/devices/memdev/full.go | 2 +- pkg/sentry/devices/memdev/null.go | 2 +- pkg/sentry/devices/memdev/random.go | 2 +- pkg/sentry/devices/memdev/zero.go | 2 +- pkg/sentry/devices/ttydev/ttydev.go | 2 +- pkg/sentry/devices/tundev/tundev.go | 4 +- pkg/sentry/fdimport/fdimport.go | 8 +- pkg/sentry/fs/copy_up.go | 12 +- pkg/sentry/fs/copy_up_test.go | 4 +- pkg/sentry/fs/dev/net_tun.go | 4 +- pkg/sentry/fs/dirent.go | 110 +++++++------- pkg/sentry/fs/dirent_cache.go | 3 +- pkg/sentry/fs/dirent_refs_test.go | 16 +-- pkg/sentry/fs/dirent_state.go | 3 +- pkg/sentry/fs/fdpipe/pipe.go | 2 +- pkg/sentry/fs/fdpipe/pipe_opener_test.go | 16 +-- pkg/sentry/fs/fdpipe/pipe_test.go | 18 +-- pkg/sentry/fs/file.go | 10 +- pkg/sentry/fs/file_operations.go | 2 +- pkg/sentry/fs/file_overlay.go | 22 +-- pkg/sentry/fs/fsutil/file.go | 4 +- pkg/sentry/fs/gofer/file.go | 4 +- pkg/sentry/fs/gofer/gofer_test.go | 8 +- pkg/sentry/fs/gofer/handles.go | 5 +- pkg/sentry/fs/gofer/inode.go | 5 +- pkg/sentry/fs/gofer/path.go | 6 +- pkg/sentry/fs/gofer/session.go | 16 +-- pkg/sentry/fs/gofer/session_state.go | 3 +- pkg/sentry/fs/gofer/socket.go | 6 +- pkg/sentry/fs/host/control.go | 2 +- pkg/sentry/fs/host/file.go | 4 +- pkg/sentry/fs/host/inode_test.go | 2 +- pkg/sentry/fs/host/socket.go | 10 +- pkg/sentry/fs/host/socket_test.go | 38 ++--- pkg/sentry/fs/host/tty.go | 4 +- pkg/sentry/fs/host/wait_test.go | 2 +- pkg/sentry/fs/inode.go | 11 +- pkg/sentry/fs/inode_inotify.go | 5 +- pkg/sentry/fs/inode_overlay.go | 30 ++-- pkg/sentry/fs/inode_overlay_test.go | 8 +- pkg/sentry/fs/inotify.go | 8 +- pkg/sentry/fs/inotify_watch.go | 9 +- pkg/sentry/fs/mount.go | 12 +- pkg/sentry/fs/mount_overlay.go | 6 +- pkg/sentry/fs/mount_test.go | 29 ++-- pkg/sentry/fs/mounts.go | 30 ++-- pkg/sentry/fs/mounts_test.go | 2 +- pkg/sentry/fs/overlay.go | 10 +- pkg/sentry/fs/proc/fds.go | 18 +-- pkg/sentry/fs/proc/mounts.go | 8 +- pkg/sentry/fs/proc/net.go | 12 +- pkg/sentry/fs/proc/proc.go | 2 +- pkg/sentry/fs/proc/task.go | 4 +- pkg/sentry/fs/ramfs/dir.go | 18 +-- pkg/sentry/fs/ramfs/tree_test.go | 2 +- pkg/sentry/fs/timerfd/timerfd.go | 4 +- pkg/sentry/fs/tmpfs/file_test.go | 2 +- pkg/sentry/fs/tty/dir.go | 8 +- pkg/sentry/fs/tty/fs.go | 2 +- pkg/sentry/fs/tty/master.go | 8 +- pkg/sentry/fs/tty/slave.go | 4 +- pkg/sentry/fs/user/path.go | 8 +- pkg/sentry/fs/user/user.go | 8 +- pkg/sentry/fs/user/user_test.go | 8 +- pkg/sentry/fsbridge/bridge.go | 2 +- pkg/sentry/fsbridge/fs.go | 8 +- pkg/sentry/fsbridge/vfs.go | 6 +- pkg/sentry/fsimpl/devpts/devpts.go | 4 +- pkg/sentry/fsimpl/devpts/master.go | 6 +- pkg/sentry/fsimpl/devpts/slave.go | 6 +- pkg/sentry/fsimpl/devtmpfs/devtmpfs.go | 6 +- pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go | 8 +- pkg/sentry/fsimpl/eventfd/eventfd.go | 6 +- pkg/sentry/fsimpl/eventfd/eventfd_test.go | 12 +- pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go | 6 +- pkg/sentry/fsimpl/ext/dentry.go | 7 +- pkg/sentry/fsimpl/ext/directory.go | 2 +- pkg/sentry/fsimpl/ext/ext.go | 10 +- pkg/sentry/fsimpl/ext/ext_test.go | 4 +- pkg/sentry/fsimpl/ext/filesystem.go | 68 ++++----- pkg/sentry/fsimpl/ext/regular_file.go | 2 +- pkg/sentry/fsimpl/ext/symlink.go | 2 +- pkg/sentry/fsimpl/fuse/dev.go | 2 +- pkg/sentry/fsimpl/fuse/dev_test.go | 4 +- pkg/sentry/fsimpl/fuse/fusefs.go | 4 +- pkg/sentry/fsimpl/gofer/directory.go | 4 +- pkg/sentry/fsimpl/gofer/filesystem.go | 90 ++++++------ pkg/sentry/fsimpl/gofer/gofer.go | 40 +++--- pkg/sentry/fsimpl/gofer/gofer_test.go | 6 +- pkg/sentry/fsimpl/gofer/regular_file.go | 2 +- pkg/sentry/fsimpl/gofer/socket.go | 6 +- pkg/sentry/fsimpl/gofer/special_file.go | 4 +- pkg/sentry/fsimpl/host/control.go | 2 +- pkg/sentry/fsimpl/host/host.go | 14 +- pkg/sentry/fsimpl/host/socket.go | 12 +- pkg/sentry/fsimpl/host/tty.go | 4 +- pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go | 2 +- pkg/sentry/fsimpl/kernfs/fd_impl_util.go | 2 +- pkg/sentry/fsimpl/kernfs/filesystem.go | 68 ++++----- pkg/sentry/fsimpl/kernfs/inode_impl_util.go | 10 +- pkg/sentry/fsimpl/kernfs/kernfs.go | 26 ++-- pkg/sentry/fsimpl/kernfs/kernfs_test.go | 18 +-- pkg/sentry/fsimpl/overlay/copy_up.go | 8 +- pkg/sentry/fsimpl/overlay/directory.go | 8 +- pkg/sentry/fsimpl/overlay/filesystem.go | 64 ++++----- pkg/sentry/fsimpl/overlay/non_directory.go | 18 +-- pkg/sentry/fsimpl/overlay/overlay.go | 48 +++---- pkg/sentry/fsimpl/pipefs/pipefs.go | 6 +- pkg/sentry/fsimpl/proc/filesystem.go | 4 +- pkg/sentry/fsimpl/proc/task_fds.go | 18 +-- pkg/sentry/fsimpl/proc/task_files.go | 14 +- pkg/sentry/fsimpl/proc/task_net.go | 12 +- pkg/sentry/fsimpl/proc/tasks_test.go | 10 +- pkg/sentry/fsimpl/signalfd/signalfd.go | 4 +- pkg/sentry/fsimpl/sockfs/sockfs.go | 4 +- pkg/sentry/fsimpl/sys/sys.go | 4 +- pkg/sentry/fsimpl/sys/sys_test.go | 2 +- pkg/sentry/fsimpl/testutil/kernel.go | 2 +- pkg/sentry/fsimpl/testutil/testutil.go | 6 +- pkg/sentry/fsimpl/timerfd/timerfd.go | 6 +- pkg/sentry/fsimpl/tmpfs/benchmark_test.go | 50 +++---- pkg/sentry/fsimpl/tmpfs/directory.go | 4 +- pkg/sentry/fsimpl/tmpfs/filesystem.go | 106 +++++++------- pkg/sentry/fsimpl/tmpfs/pipe_test.go | 20 +-- pkg/sentry/fsimpl/tmpfs/regular_file.go | 2 +- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 34 ++--- pkg/sentry/fsimpl/tmpfs/tmpfs_test.go | 6 +- pkg/sentry/kernel/abstract_socket_namespace.go | 13 +- pkg/sentry/kernel/epoll/epoll.go | 14 +- pkg/sentry/kernel/epoll/epoll_test.go | 5 +- pkg/sentry/kernel/eventfd/eventfd.go | 4 +- pkg/sentry/kernel/fd_table.go | 55 +++---- pkg/sentry/kernel/fd_table_test.go | 6 +- pkg/sentry/kernel/fs_context.go | 31 ++-- pkg/sentry/kernel/futex/BUILD | 1 + pkg/sentry/kernel/futex/futex.go | 35 ++--- pkg/sentry/kernel/futex/futex_test.go | 66 +++++---- pkg/sentry/kernel/kernel.go | 57 ++++---- pkg/sentry/kernel/pipe/node.go | 6 +- pkg/sentry/kernel/pipe/node_test.go | 2 +- pkg/sentry/kernel/pipe/pipe.go | 2 +- pkg/sentry/kernel/pipe/pipe_test.go | 16 +-- pkg/sentry/kernel/pipe/pipe_util.go | 2 +- pkg/sentry/kernel/pipe/reader.go | 3 +- pkg/sentry/kernel/pipe/vfs.go | 8 +- pkg/sentry/kernel/pipe/writer.go | 3 +- pkg/sentry/kernel/sessions.go | 5 +- pkg/sentry/kernel/shm/shm.go | 10 +- pkg/sentry/kernel/signalfd/signalfd.go | 2 +- pkg/sentry/kernel/task.go | 8 +- pkg/sentry/kernel/task_clone.go | 10 +- pkg/sentry/kernel/task_exec.go | 6 +- pkg/sentry/kernel/task_exit.go | 8 +- pkg/sentry/kernel/task_log.go | 2 +- pkg/sentry/kernel/task_start.go | 6 +- pkg/sentry/kernel/thread_group.go | 4 +- pkg/sentry/loader/elf.go | 4 +- pkg/sentry/loader/loader.go | 6 +- pkg/sentry/memmap/memmap.go | 2 +- pkg/sentry/mm/aio_context.go | 6 +- pkg/sentry/mm/lifecycle.go | 2 +- pkg/sentry/mm/metadata.go | 5 +- pkg/sentry/mm/special_mappable.go | 4 +- pkg/sentry/mm/syscalls.go | 4 +- pkg/sentry/mm/vma.go | 4 +- pkg/sentry/socket/control/control.go | 8 +- pkg/sentry/socket/control/control_vfs2.go | 8 +- pkg/sentry/socket/hostinet/socket.go | 8 +- pkg/sentry/socket/netlink/provider.go | 2 +- pkg/sentry/socket/netlink/socket.go | 14 +- pkg/sentry/socket/netlink/socket_vfs2.go | 4 +- pkg/sentry/socket/netstack/netstack.go | 6 +- pkg/sentry/socket/netstack/netstack_vfs2.go | 2 +- pkg/sentry/socket/socket.go | 4 +- pkg/sentry/socket/unix/transport/connectioned.go | 10 +- pkg/sentry/socket/unix/transport/connectionless.go | 12 +- pkg/sentry/socket/unix/transport/queue.go | 13 +- pkg/sentry/socket/unix/transport/unix.go | 48 +++---- pkg/sentry/socket/unix/unix.go | 38 ++--- pkg/sentry/socket/unix/unix_vfs2.go | 18 +-- pkg/sentry/strace/strace.go | 12 +- pkg/sentry/syscalls/epoll.go | 18 +-- pkg/sentry/syscalls/linux/sys_aio.go | 8 +- pkg/sentry/syscalls/linux/sys_eventfd.go | 2 +- pkg/sentry/syscalls/linux/sys_file.go | 78 +++++----- pkg/sentry/syscalls/linux/sys_futex.go | 6 +- pkg/sentry/syscalls/linux/sys_getdents.go | 2 +- pkg/sentry/syscalls/linux/sys_inotify.go | 10 +- pkg/sentry/syscalls/linux/sys_lseek.go | 2 +- pkg/sentry/syscalls/linux/sys_mmap.go | 4 +- pkg/sentry/syscalls/linux/sys_mount.go | 2 +- pkg/sentry/syscalls/linux/sys_pipe.go | 6 +- pkg/sentry/syscalls/linux/sys_poll.go | 10 +- pkg/sentry/syscalls/linux/sys_prctl.go | 4 +- pkg/sentry/syscalls/linux/sys_read.go | 12 +- pkg/sentry/syscalls/linux/sys_shm.go | 10 +- pkg/sentry/syscalls/linux/sys_signal.go | 4 +- pkg/sentry/syscalls/linux/sys_socket.go | 46 +++--- pkg/sentry/syscalls/linux/sys_splice.go | 12 +- pkg/sentry/syscalls/linux/sys_stat.go | 8 +- pkg/sentry/syscalls/linux/sys_sync.go | 8 +- pkg/sentry/syscalls/linux/sys_thread.go | 6 +- pkg/sentry/syscalls/linux/sys_timerfd.go | 6 +- pkg/sentry/syscalls/linux/sys_write.go | 10 +- pkg/sentry/syscalls/linux/sys_xattr.go | 8 +- pkg/sentry/syscalls/linux/vfs2/aio.go | 8 +- pkg/sentry/syscalls/linux/vfs2/epoll.go | 14 +- pkg/sentry/syscalls/linux/vfs2/eventfd.go | 4 +- pkg/sentry/syscalls/linux/vfs2/execve.go | 12 +- pkg/sentry/syscalls/linux/vfs2/fd.go | 12 +- pkg/sentry/syscalls/linux/vfs2/filesystem.go | 22 +-- pkg/sentry/syscalls/linux/vfs2/fscontext.go | 22 +-- pkg/sentry/syscalls/linux/vfs2/getdents.go | 2 +- pkg/sentry/syscalls/linux/vfs2/inotify.go | 14 +- pkg/sentry/syscalls/linux/vfs2/ioctl.go | 2 +- pkg/sentry/syscalls/linux/vfs2/lock.go | 2 +- pkg/sentry/syscalls/linux/vfs2/memfd.go | 2 +- pkg/sentry/syscalls/linux/vfs2/mmap.go | 4 +- pkg/sentry/syscalls/linux/vfs2/mount.go | 4 +- pkg/sentry/syscalls/linux/vfs2/path.go | 12 +- pkg/sentry/syscalls/linux/vfs2/pipe.go | 6 +- pkg/sentry/syscalls/linux/vfs2/poll.go | 10 +- pkg/sentry/syscalls/linux/vfs2/read_write.go | 48 +++---- pkg/sentry/syscalls/linux/vfs2/setstat.go | 20 +-- pkg/sentry/syscalls/linux/vfs2/signal.go | 4 +- pkg/sentry/syscalls/linux/vfs2/socket.go | 46 +++--- pkg/sentry/syscalls/linux/vfs2/splice.go | 20 +-- pkg/sentry/syscalls/linux/vfs2/stat.go | 30 ++-- pkg/sentry/syscalls/linux/vfs2/sync.go | 6 +- pkg/sentry/syscalls/linux/vfs2/timerfd.go | 8 +- pkg/sentry/syscalls/linux/vfs2/xattr.go | 16 +-- pkg/sentry/vfs/anonfs.go | 8 +- pkg/sentry/vfs/dentry.go | 37 ++--- pkg/sentry/vfs/epoll.go | 6 +- pkg/sentry/vfs/file_description.go | 24 ++-- pkg/sentry/vfs/file_description_impl_util_test.go | 18 +-- pkg/sentry/vfs/filesystem.go | 6 +- pkg/sentry/vfs/inotify.go | 34 ++--- pkg/sentry/vfs/mount.go | 54 +++---- pkg/sentry/vfs/pathname.go | 18 +-- pkg/sentry/vfs/resolving_path.go | 43 +++--- pkg/sentry/vfs/vfs.go | 160 ++++++++++----------- pkg/tcpip/link/tun/BUILD | 1 + pkg/tcpip/link/tun/device.go | 9 +- runsc/boot/fs.go | 12 +- runsc/boot/loader.go | 16 +-- runsc/boot/loader_test.go | 8 +- runsc/boot/vfs.go | 10 +- 252 files changed, 1711 insertions(+), 1668 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/refs/BUILD b/pkg/refs/BUILD index 74affc887..9888cce9c 100644 --- a/pkg/refs/BUILD +++ b/pkg/refs/BUILD @@ -24,6 +24,7 @@ go_library( ], visibility = ["//:sandbox"], deps = [ + "//pkg/context", "//pkg/log", "//pkg/sync", ], @@ -34,5 +35,8 @@ go_test( size = "small", srcs = ["refcounter_test.go"], library = ":refs", - deps = ["//pkg/sync"], + deps = [ + "//pkg/context", + "//pkg/sync", + ], ) diff --git a/pkg/refs/refcounter.go b/pkg/refs/refcounter.go index c45ba8200..61790221b 100644 --- a/pkg/refs/refcounter.go +++ b/pkg/refs/refcounter.go @@ -23,6 +23,7 @@ import ( "runtime" "sync/atomic" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sync" ) @@ -38,7 +39,7 @@ type RefCounter interface { // Note that AtomicRefCounter.DecRef() does not support destructors. // If a type has a destructor, it must implement its own DecRef() // method and call AtomicRefCounter.DecRefWithDestructor(destructor). - DecRef() + DecRef(ctx context.Context) // TryIncRef attempts to increase the reference counter on the object, // but may fail if all references have already been dropped. This @@ -57,7 +58,7 @@ type RefCounter interface { // A WeakRefUser is notified when the last non-weak reference is dropped. type WeakRefUser interface { // WeakRefGone is called when the last non-weak reference is dropped. - WeakRefGone() + WeakRefGone(ctx context.Context) } // WeakRef is a weak reference. @@ -123,7 +124,7 @@ func (w *WeakRef) Get() RefCounter { // Drop drops this weak reference. You should always call drop when you are // finished with the weak reference. You may not use this object after calling // drop. -func (w *WeakRef) Drop() { +func (w *WeakRef) Drop(ctx context.Context) { rc, ok := w.get() if !ok { // We've been zapped already. When the refcounter has called @@ -145,7 +146,7 @@ func (w *WeakRef) Drop() { // And now aren't on the object's list of weak references. So it won't // zap us if this causes the reference count to drop to zero. - rc.DecRef() + rc.DecRef(ctx) // Return to the pool. weakRefPool.Put(w) @@ -427,7 +428,7 @@ func (r *AtomicRefCount) dropWeakRef(w *WeakRef) { // A: TryIncRef [transform speculative to real] // //go:nosplit -func (r *AtomicRefCount) DecRefWithDestructor(destroy func()) { +func (r *AtomicRefCount) DecRefWithDestructor(ctx context.Context, destroy func(context.Context)) { switch v := atomic.AddInt64(&r.refCount, -1); { case v < -1: panic("Decrementing non-positive ref count") @@ -448,7 +449,7 @@ func (r *AtomicRefCount) DecRefWithDestructor(destroy func()) { if user != nil { r.mu.Unlock() - user.WeakRefGone() + user.WeakRefGone(ctx) r.mu.Lock() } } @@ -456,7 +457,7 @@ func (r *AtomicRefCount) DecRefWithDestructor(destroy func()) { // Call the destructor. if destroy != nil { - destroy() + destroy(ctx) } } } @@ -464,6 +465,6 @@ func (r *AtomicRefCount) DecRefWithDestructor(destroy func()) { // DecRef decrements this object's reference count. // //go:nosplit -func (r *AtomicRefCount) DecRef() { - r.DecRefWithDestructor(nil) +func (r *AtomicRefCount) DecRef(ctx context.Context) { + r.DecRefWithDestructor(ctx, nil) } diff --git a/pkg/refs/refcounter_test.go b/pkg/refs/refcounter_test.go index 1ab4a4440..6d0dd1018 100644 --- a/pkg/refs/refcounter_test.go +++ b/pkg/refs/refcounter_test.go @@ -18,6 +18,7 @@ import ( "reflect" "testing" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sync" ) @@ -31,11 +32,11 @@ type testCounter struct { destroyed bool } -func (t *testCounter) DecRef() { - t.AtomicRefCount.DecRefWithDestructor(t.destroy) +func (t *testCounter) DecRef(ctx context.Context) { + t.AtomicRefCount.DecRefWithDestructor(ctx, t.destroy) } -func (t *testCounter) destroy() { +func (t *testCounter) destroy(context.Context) { t.mu.Lock() defer t.mu.Unlock() t.destroyed = true @@ -53,7 +54,7 @@ func newTestCounter() *testCounter { func TestOneRef(t *testing.T) { tc := newTestCounter() - tc.DecRef() + tc.DecRef(context.Background()) if !tc.IsDestroyed() { t.Errorf("object should have been destroyed") @@ -63,8 +64,9 @@ func TestOneRef(t *testing.T) { func TestTwoRefs(t *testing.T) { tc := newTestCounter() tc.IncRef() - tc.DecRef() - tc.DecRef() + ctx := context.Background() + tc.DecRef(ctx) + tc.DecRef(ctx) if !tc.IsDestroyed() { t.Errorf("object should have been destroyed") @@ -74,12 +76,13 @@ func TestTwoRefs(t *testing.T) { func TestMultiRefs(t *testing.T) { tc := newTestCounter() tc.IncRef() - tc.DecRef() + ctx := context.Background() + tc.DecRef(ctx) tc.IncRef() - tc.DecRef() + tc.DecRef(ctx) - tc.DecRef() + tc.DecRef(ctx) if !tc.IsDestroyed() { t.Errorf("object should have been destroyed") @@ -89,19 +92,20 @@ func TestMultiRefs(t *testing.T) { func TestWeakRef(t *testing.T) { tc := newTestCounter() w := NewWeakRef(tc, nil) + ctx := context.Background() // Try resolving. if x := w.Get(); x == nil { t.Errorf("weak reference didn't resolve: expected %v, got nil", tc) } else { - x.DecRef() + x.DecRef(ctx) } // Try resolving again. if x := w.Get(); x == nil { t.Errorf("weak reference didn't resolve: expected %v, got nil", tc) } else { - x.DecRef() + x.DecRef(ctx) } // Shouldn't be destroyed yet. (Can't continue if this fails.) @@ -110,7 +114,7 @@ func TestWeakRef(t *testing.T) { } // Drop the original reference. - tc.DecRef() + tc.DecRef(ctx) // Assert destroyed. if !tc.IsDestroyed() { @@ -126,7 +130,8 @@ func TestWeakRef(t *testing.T) { func TestWeakRefDrop(t *testing.T) { tc := newTestCounter() w := NewWeakRef(tc, nil) - w.Drop() + ctx := context.Background() + w.Drop(ctx) // Just assert the list is empty. if !tc.weakRefs.Empty() { @@ -134,14 +139,14 @@ func TestWeakRefDrop(t *testing.T) { } // Drop the original reference. - tc.DecRef() + tc.DecRef(ctx) } type testWeakRefUser struct { weakRefGone func() } -func (u *testWeakRefUser) WeakRefGone() { +func (u *testWeakRefUser) WeakRefGone(ctx context.Context) { u.weakRefGone() } @@ -165,7 +170,8 @@ func TestCallback(t *testing.T) { }}) // Drop the original reference, this must trigger the callback. - tc.DecRef() + ctx := context.Background() + tc.DecRef(ctx) if !called { t.Fatalf("Callback not called") diff --git a/pkg/sentry/control/proc.go b/pkg/sentry/control/proc.go index 1bae7cfaf..dfa936563 100644 --- a/pkg/sentry/control/proc.go +++ b/pkg/sentry/control/proc.go @@ -139,7 +139,6 @@ func ExecAsync(proc *Proc, args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadID func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadID, *host.TTYFileOperations, *hostvfs2.TTYFileDescription, error) { // Import file descriptors. fdTable := proc.Kernel.NewFDTable() - defer fdTable.DecRef() creds := auth.NewUserCredentials( args.KUID, @@ -177,6 +176,7 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI initArgs.MountNamespaceVFS2.IncRef() } ctx := initArgs.NewContext(proc.Kernel) + defer fdTable.DecRef(ctx) if kernel.VFS2Enabled { // Get the full path to the filename from the PATH env variable. diff --git a/pkg/sentry/devices/memdev/full.go b/pkg/sentry/devices/memdev/full.go index af66fe4dc..511179e31 100644 --- a/pkg/sentry/devices/memdev/full.go +++ b/pkg/sentry/devices/memdev/full.go @@ -46,7 +46,7 @@ type fullFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *fullFD) Release() { +func (fd *fullFD) Release(context.Context) { // noop } diff --git a/pkg/sentry/devices/memdev/null.go b/pkg/sentry/devices/memdev/null.go index 92d3d71be..4918dbeeb 100644 --- a/pkg/sentry/devices/memdev/null.go +++ b/pkg/sentry/devices/memdev/null.go @@ -47,7 +47,7 @@ type nullFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *nullFD) Release() { +func (fd *nullFD) Release(context.Context) { // noop } diff --git a/pkg/sentry/devices/memdev/random.go b/pkg/sentry/devices/memdev/random.go index 6b81da5ef..5e7fe0280 100644 --- a/pkg/sentry/devices/memdev/random.go +++ b/pkg/sentry/devices/memdev/random.go @@ -56,7 +56,7 @@ type randomFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *randomFD) Release() { +func (fd *randomFD) Release(context.Context) { // noop } diff --git a/pkg/sentry/devices/memdev/zero.go b/pkg/sentry/devices/memdev/zero.go index c6f15054d..2e631a252 100644 --- a/pkg/sentry/devices/memdev/zero.go +++ b/pkg/sentry/devices/memdev/zero.go @@ -48,7 +48,7 @@ type zeroFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *zeroFD) Release() { +func (fd *zeroFD) Release(context.Context) { // noop } diff --git a/pkg/sentry/devices/ttydev/ttydev.go b/pkg/sentry/devices/ttydev/ttydev.go index fbb7fd92c..fd4b79c46 100644 --- a/pkg/sentry/devices/ttydev/ttydev.go +++ b/pkg/sentry/devices/ttydev/ttydev.go @@ -55,7 +55,7 @@ type ttyFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *ttyFD) Release() {} +func (fd *ttyFD) Release(context.Context) {} // PRead implements vfs.FileDescriptionImpl.PRead. func (fd *ttyFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { diff --git a/pkg/sentry/devices/tundev/tundev.go b/pkg/sentry/devices/tundev/tundev.go index dfbd069af..852ec3c5c 100644 --- a/pkg/sentry/devices/tundev/tundev.go +++ b/pkg/sentry/devices/tundev/tundev.go @@ -108,8 +108,8 @@ func (fd *tunFD) Ioctl(ctx context.Context, uio usermem.IO, args arch.SyscallArg } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *tunFD) Release() { - fd.device.Release() +func (fd *tunFD) Release(ctx context.Context) { + fd.device.Release(ctx) } // PRead implements vfs.FileDescriptionImpl.PRead. diff --git a/pkg/sentry/fdimport/fdimport.go b/pkg/sentry/fdimport/fdimport.go index b8686adb4..1b7cb94c0 100644 --- a/pkg/sentry/fdimport/fdimport.go +++ b/pkg/sentry/fdimport/fdimport.go @@ -50,7 +50,7 @@ func importFS(ctx context.Context, fdTable *kernel.FDTable, console bool, fds [] if err != nil { return nil, err } - defer appFile.DecRef() + defer appFile.DecRef(ctx) // Remember this in the TTY file, as we will // use it for the other stdio FDs. @@ -69,7 +69,7 @@ func importFS(ctx context.Context, fdTable *kernel.FDTable, console bool, fds [] if err != nil { return nil, err } - defer appFile.DecRef() + defer appFile.DecRef(ctx) } // Add the file to the FD map. @@ -102,7 +102,7 @@ func importVFS2(ctx context.Context, fdTable *kernel.FDTable, console bool, stdi if err != nil { return nil, err } - defer appFile.DecRef() + defer appFile.DecRef(ctx) // Remember this in the TTY file, as we will use it for the other stdio // FDs. @@ -119,7 +119,7 @@ func importVFS2(ctx context.Context, fdTable *kernel.FDTable, console bool, stdi if err != nil { return nil, err } - defer appFile.DecRef() + defer appFile.DecRef(ctx) } if err := fdTable.NewFDAtVFS2(ctx, int32(appFD), appFile, kernel.FDFlags{}); err != nil { diff --git a/pkg/sentry/fs/copy_up.go b/pkg/sentry/fs/copy_up.go index ab1424c95..735452b07 100644 --- a/pkg/sentry/fs/copy_up.go +++ b/pkg/sentry/fs/copy_up.go @@ -201,7 +201,7 @@ func copyUpLocked(ctx context.Context, parent *Dirent, next *Dirent) error { parentUpper := parent.Inode.overlay.upper root := RootFromContext(ctx) if root != nil { - defer root.DecRef() + defer root.DecRef(ctx) } // Create the file in the upper filesystem and get an Inode for it. @@ -212,7 +212,7 @@ func copyUpLocked(ctx context.Context, parent *Dirent, next *Dirent) error { log.Warningf("copy up failed to create file: %v", err) return syserror.EIO } - defer childFile.DecRef() + defer childFile.DecRef(ctx) childUpperInode = childFile.Dirent.Inode case Directory: @@ -226,7 +226,7 @@ func copyUpLocked(ctx context.Context, parent *Dirent, next *Dirent) error { cleanupUpper(ctx, parentUpper, next.name, werr) return syserror.EIO } - defer childUpper.DecRef() + defer childUpper.DecRef(ctx) childUpperInode = childUpper.Inode case Symlink: @@ -246,7 +246,7 @@ func copyUpLocked(ctx context.Context, parent *Dirent, next *Dirent) error { cleanupUpper(ctx, parentUpper, next.name, werr) return syserror.EIO } - defer childUpper.DecRef() + defer childUpper.DecRef(ctx) childUpperInode = childUpper.Inode default: @@ -352,14 +352,14 @@ func copyContentsLocked(ctx context.Context, upper *Inode, lower *Inode, size in if err != nil { return err } - defer upperFile.DecRef() + defer upperFile.DecRef(ctx) // Get a handle to the lower filesystem, which we will read from. lowerFile, err := overlayFile(ctx, lower, FileFlags{Read: true}) if err != nil { return err } - defer lowerFile.DecRef() + defer lowerFile.DecRef(ctx) // Use a buffer pool to minimize allocations. buf := copyUpBuffers.Get().([]byte) diff --git a/pkg/sentry/fs/copy_up_test.go b/pkg/sentry/fs/copy_up_test.go index 91792d9fe..c7a11eec1 100644 --- a/pkg/sentry/fs/copy_up_test.go +++ b/pkg/sentry/fs/copy_up_test.go @@ -126,7 +126,7 @@ func makeOverlayTestFiles(t *testing.T) []*overlayTestFile { if err != nil { t.Fatalf("failed to create file %q: %v", name, err) } - defer f.DecRef() + defer f.DecRef(ctx) relname, _ := f.Dirent.FullName(lowerRoot) @@ -171,7 +171,7 @@ func makeOverlayTestFiles(t *testing.T) []*overlayTestFile { if err != nil { t.Fatalf("failed to find %q: %v", f.name, err) } - defer d.DecRef() + defer d.DecRef(ctx) f.File, err = d.Inode.GetFile(ctx, d, fs.FileFlags{Read: true}) if err != nil { diff --git a/pkg/sentry/fs/dev/net_tun.go b/pkg/sentry/fs/dev/net_tun.go index dc7ad075a..ec474e554 100644 --- a/pkg/sentry/fs/dev/net_tun.go +++ b/pkg/sentry/fs/dev/net_tun.go @@ -80,8 +80,8 @@ type netTunFileOperations struct { var _ fs.FileOperations = (*netTunFileOperations)(nil) // Release implements fs.FileOperations.Release. -func (fops *netTunFileOperations) Release() { - fops.device.Release() +func (fops *netTunFileOperations) Release(ctx context.Context) { + fops.device.Release(ctx) } // Ioctl implements fs.FileOperations.Ioctl. diff --git a/pkg/sentry/fs/dirent.go b/pkg/sentry/fs/dirent.go index 65be12175..a2f751068 100644 --- a/pkg/sentry/fs/dirent.go +++ b/pkg/sentry/fs/dirent.go @@ -325,7 +325,7 @@ func (d *Dirent) SyncAll(ctx context.Context) { for _, w := range d.children { if child := w.Get(); child != nil { child.(*Dirent).SyncAll(ctx) - child.DecRef() + child.DecRef(ctx) } } } @@ -451,7 +451,7 @@ func (d *Dirent) walk(ctx context.Context, root *Dirent, name string, walkMayUnl // which don't hold a hard reference on their parent (their parent holds a // hard reference on them, and they contain virtually no state). But this is // good house-keeping. - child.DecRef() + child.DecRef(ctx) return nil, syscall.ENOENT } @@ -468,20 +468,20 @@ func (d *Dirent) walk(ctx context.Context, root *Dirent, name string, walkMayUnl // their pins on the child. Inotify doesn't properly support filesystems that // revalidate dirents (since watches are lost on revalidation), but if we fail // to unpin the watches child will never be GCed. - cd.Inode.Watches.Unpin(cd) + cd.Inode.Watches.Unpin(ctx, cd) // This child needs to be revalidated, fallthrough to unhash it. Make sure // to not leak a reference from Get(). // // Note that previous lookups may still have a reference to this stale child; // this can't be helped, but we can ensure that *new* lookups are up-to-date. - child.DecRef() + child.DecRef(ctx) } // Either our weak reference expired or we need to revalidate it. Unhash child first, we're // about to replace it. delete(d.children, name) - w.Drop() + w.Drop(ctx) } // Slow path: load the InodeOperations into memory. Since this is a hot path and the lookup may be @@ -512,12 +512,12 @@ func (d *Dirent) walk(ctx context.Context, root *Dirent, name string, walkMayUnl // There are active references to the existing child, prefer it to the one we // retrieved from Lookup. Likely the Lookup happened very close to the insertion // of child, so considering one stale over the other is fairly arbitrary. - c.DecRef() + c.DecRef(ctx) // The child that was installed could be negative. if cd.IsNegative() { // If so, don't leak a reference and short circuit. - child.DecRef() + child.DecRef(ctx) return nil, syscall.ENOENT } @@ -531,7 +531,7 @@ func (d *Dirent) walk(ctx context.Context, root *Dirent, name string, walkMayUnl // we did the Inode.Lookup. Fully drop the weak reference and fallback to using the child // we looked up. delete(d.children, name) - w.Drop() + w.Drop(ctx) } // Give the looked up child a parent. We cannot kick out entries, since we just checked above @@ -587,7 +587,7 @@ func (d *Dirent) exists(ctx context.Context, root *Dirent, name string) bool { return false } // Child exists. - child.DecRef() + child.DecRef(ctx) return true } @@ -622,7 +622,7 @@ func (d *Dirent) Create(ctx context.Context, root *Dirent, name string, flags Fi } child := file.Dirent - d.finishCreate(child, name) + d.finishCreate(ctx, child, name) // Return the reference and the new file. When the last reference to // the file is dropped, file.Dirent may no longer be cached. @@ -631,7 +631,7 @@ func (d *Dirent) Create(ctx context.Context, root *Dirent, name string, flags Fi // finishCreate validates the created file, adds it as a child of this dirent, // and notifies any watchers. -func (d *Dirent) finishCreate(child *Dirent, name string) { +func (d *Dirent) finishCreate(ctx context.Context, child *Dirent, name string) { // Sanity check c, its name must be consistent. if child.name != name { panic(fmt.Sprintf("create from %q to %q returned unexpected name %q", d.name, name, child.name)) @@ -650,14 +650,14 @@ func (d *Dirent) finishCreate(child *Dirent, name string) { panic(fmt.Sprintf("hashed child %q over a positive child", child.name)) } // Don't leak a reference. - old.DecRef() + old.DecRef(ctx) // Drop d's reference. - old.DecRef() + old.DecRef(ctx) } // Finally drop the useless weak reference on the floor. - w.Drop() + w.Drop(ctx) } d.Inode.Watches.Notify(name, linux.IN_CREATE, 0) @@ -686,17 +686,17 @@ func (d *Dirent) genericCreate(ctx context.Context, root *Dirent, name string, c panic(fmt.Sprintf("hashed over a positive child %q", old.(*Dirent).name)) } // Don't leak a reference. - old.DecRef() + old.DecRef(ctx) // Drop d's reference. - old.DecRef() + old.DecRef(ctx) } // Unhash the negative Dirent, name needs to exist now. delete(d.children, name) // Finally drop the useless weak reference on the floor. - w.Drop() + w.Drop(ctx) } // Execute the create operation. @@ -756,7 +756,7 @@ func (d *Dirent) Bind(ctx context.Context, root *Dirent, name string, data trans if e != nil { return e } - d.finishCreate(childDir, name) + d.finishCreate(ctx, childDir, name) return nil }) if err == syscall.EEXIST { @@ -901,7 +901,7 @@ func direntReaddir(ctx context.Context, d *Dirent, it DirIterator, root *Dirent, // references to children. // // Preconditions: d.mu must be held. -func (d *Dirent) flush() { +func (d *Dirent) flush(ctx context.Context) { expired := make(map[string]*refs.WeakRef) for n, w := range d.children { // Call flush recursively on each child before removing our @@ -912,7 +912,7 @@ func (d *Dirent) flush() { if !cd.IsNegative() { // Flush the child. cd.mu.Lock() - cd.flush() + cd.flush(ctx) cd.mu.Unlock() // Allow the file system to drop extra references on child. @@ -920,13 +920,13 @@ func (d *Dirent) flush() { } // Don't leak a reference. - child.DecRef() + child.DecRef(ctx) } // Check if the child dirent is closed, and mark it as expired if it is. // We must call w.Get() again here, since the child could have been closed // by the calls to flush() and cache.Remove() in the above if-block. if child := w.Get(); child != nil { - child.DecRef() + child.DecRef(ctx) } else { expired[n] = w } @@ -935,7 +935,7 @@ func (d *Dirent) flush() { // Remove expired entries. for n, w := range expired { delete(d.children, n) - w.Drop() + w.Drop(ctx) } } @@ -977,7 +977,7 @@ func (d *Dirent) mount(ctx context.Context, inode *Inode) (newChild *Dirent, err if !ok { panic("mount must mount over an existing dirent") } - weakRef.Drop() + weakRef.Drop(ctx) // Note that even though `d` is now hidden, it still holds a reference // to its parent. @@ -1002,13 +1002,13 @@ func (d *Dirent) unmount(ctx context.Context, replacement *Dirent) error { if !ok { panic("mount must mount over an existing dirent") } - weakRef.Drop() + weakRef.Drop(ctx) // d is not reachable anymore, and hence not mounted anymore. d.mounted = false // Drop mount reference. - d.DecRef() + d.DecRef(ctx) return nil } @@ -1029,7 +1029,7 @@ func (d *Dirent) Remove(ctx context.Context, root *Dirent, name string, dirPath // Child does not exist. return err } - defer child.DecRef() + defer child.DecRef(ctx) // Remove cannot remove directories. if IsDir(child.Inode.StableAttr) { @@ -1055,7 +1055,7 @@ func (d *Dirent) Remove(ctx context.Context, root *Dirent, name string, dirPath atomic.StoreInt32(&child.deleted, 1) if w, ok := d.children[name]; ok { delete(d.children, name) - w.Drop() + w.Drop(ctx) } // Allow the file system to drop extra references on child. @@ -1067,7 +1067,7 @@ func (d *Dirent) Remove(ctx context.Context, root *Dirent, name string, dirPath // inode may have other links. If this was the last link, the events for the // watch removal will be queued by the inode destructor. child.Inode.Watches.MarkUnlinked() - child.Inode.Watches.Unpin(child) + child.Inode.Watches.Unpin(ctx, child) d.Inode.Watches.Notify(name, linux.IN_DELETE, 0) return nil @@ -1100,7 +1100,7 @@ func (d *Dirent) RemoveDirectory(ctx context.Context, root *Dirent, name string) // Child does not exist. return err } - defer child.DecRef() + defer child.DecRef(ctx) // RemoveDirectory can only remove directories. if !IsDir(child.Inode.StableAttr) { @@ -1121,7 +1121,7 @@ func (d *Dirent) RemoveDirectory(ctx context.Context, root *Dirent, name string) atomic.StoreInt32(&child.deleted, 1) if w, ok := d.children[name]; ok { delete(d.children, name) - w.Drop() + w.Drop(ctx) } // Allow the file system to drop extra references on child. @@ -1130,14 +1130,14 @@ func (d *Dirent) RemoveDirectory(ctx context.Context, root *Dirent, name string) // Finally, let inotify know the child is being unlinked. Drop any extra // refs from inotify to this child dirent. child.Inode.Watches.MarkUnlinked() - child.Inode.Watches.Unpin(child) + child.Inode.Watches.Unpin(ctx, child) d.Inode.Watches.Notify(name, linux.IN_ISDIR|linux.IN_DELETE, 0) return nil } // destroy closes this node and all children. -func (d *Dirent) destroy() { +func (d *Dirent) destroy(ctx context.Context) { if d.IsNegative() { // Nothing to tear-down and no parent references to drop, since a negative // Dirent does not take a references on its parent, has no Inode and no children. @@ -1153,19 +1153,19 @@ func (d *Dirent) destroy() { if c.(*Dirent).IsNegative() { // The parent holds both weak and strong refs in the case of // negative dirents. - c.DecRef() + c.DecRef(ctx) } // Drop the reference we just acquired in WeakRef.Get. - c.DecRef() + c.DecRef(ctx) } - w.Drop() + w.Drop(ctx) } d.children = nil allDirents.remove(d) // Drop our reference to the Inode. - d.Inode.DecRef() + d.Inode.DecRef(ctx) // Allow the Dirent to be GC'ed after this point, since the Inode may still // be referenced after the Dirent is destroyed (for instance by filesystem @@ -1175,7 +1175,7 @@ func (d *Dirent) destroy() { // Drop the reference we have on our parent if we took one. renameMu doesn't need to be // held because d can't be reparented without any references to it left. if d.parent != nil { - d.parent.DecRef() + d.parent.DecRef(ctx) } } @@ -1201,14 +1201,14 @@ func (d *Dirent) TryIncRef() bool { // DecRef decreases the Dirent's refcount and drops its reference on its mount. // // DecRef implements RefCounter.DecRef with destructor d.destroy. -func (d *Dirent) DecRef() { +func (d *Dirent) DecRef(ctx context.Context) { if d.Inode != nil { // Keep mount around, since DecRef may destroy d.Inode. msrc := d.Inode.MountSource - d.DecRefWithDestructor(d.destroy) + d.DecRefWithDestructor(ctx, d.destroy) msrc.DecDirentRefs() } else { - d.DecRefWithDestructor(d.destroy) + d.DecRefWithDestructor(ctx, d.destroy) } } @@ -1359,7 +1359,7 @@ func (d *Dirent) MayDelete(ctx context.Context, root *Dirent, name string) error if err != nil { return err } - defer victim.DecRef() + defer victim.DecRef(ctx) return d.mayDelete(ctx, victim) } @@ -1411,7 +1411,7 @@ func Rename(ctx context.Context, root *Dirent, oldParent *Dirent, oldName string if err != nil { return err } - defer renamed.DecRef() + defer renamed.DecRef(ctx) // Check that the renamed dirent is deletable. if err := oldParent.mayDelete(ctx, renamed); err != nil { @@ -1453,13 +1453,13 @@ func Rename(ctx context.Context, root *Dirent, oldParent *Dirent, oldName string // Check that we can delete replaced. if err := newParent.mayDelete(ctx, replaced); err != nil { - replaced.DecRef() + replaced.DecRef(ctx) return err } // Target should not be an ancestor of source. if oldParent.descendantOf(replaced) { - replaced.DecRef() + replaced.DecRef(ctx) // Note that Linux returns EINVAL if the source is an // ancestor of target, but ENOTEMPTY if the target is @@ -1470,7 +1470,7 @@ func Rename(ctx context.Context, root *Dirent, oldParent *Dirent, oldName string // Check that replaced is not a mount point. if replaced.isMountPointLocked() { - replaced.DecRef() + replaced.DecRef(ctx) return syscall.EBUSY } @@ -1478,11 +1478,11 @@ func Rename(ctx context.Context, root *Dirent, oldParent *Dirent, oldName string oldIsDir := IsDir(renamed.Inode.StableAttr) newIsDir := IsDir(replaced.Inode.StableAttr) if !newIsDir && oldIsDir { - replaced.DecRef() + replaced.DecRef(ctx) return syscall.ENOTDIR } if !oldIsDir && newIsDir { - replaced.DecRef() + replaced.DecRef(ctx) return syscall.EISDIR } @@ -1493,13 +1493,13 @@ func Rename(ctx context.Context, root *Dirent, oldParent *Dirent, oldName string // open across renames is currently broken for multiple // reasons, so we flush all references on the replaced node and // its children. - replaced.Inode.Watches.Unpin(replaced) + replaced.Inode.Watches.Unpin(ctx, replaced) replaced.mu.Lock() - replaced.flush() + replaced.flush(ctx) replaced.mu.Unlock() // Done with replaced. - replaced.DecRef() + replaced.DecRef(ctx) } if err := renamed.Inode.Rename(ctx, oldParent, renamed, newParent, newName, replaced != nil); err != nil { @@ -1513,14 +1513,14 @@ func Rename(ctx context.Context, root *Dirent, oldParent *Dirent, oldName string // can't destroy oldParent (and try to retake its lock) because // Rename's caller must be holding a reference. newParent.IncRef() - oldParent.DecRef() + oldParent.DecRef(ctx) } if w, ok := newParent.children[newName]; ok { - w.Drop() + w.Drop(ctx) delete(newParent.children, newName) } if w, ok := oldParent.children[oldName]; ok { - w.Drop() + w.Drop(ctx) delete(oldParent.children, oldName) } @@ -1551,7 +1551,7 @@ func Rename(ctx context.Context, root *Dirent, oldParent *Dirent, oldName string // Same as replaced.flush above. renamed.mu.Lock() - renamed.flush() + renamed.flush(ctx) renamed.mu.Unlock() return nil diff --git a/pkg/sentry/fs/dirent_cache.go b/pkg/sentry/fs/dirent_cache.go index 33de32c69..7d9dd717e 100644 --- a/pkg/sentry/fs/dirent_cache.go +++ b/pkg/sentry/fs/dirent_cache.go @@ -17,6 +17,7 @@ package fs import ( "fmt" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sync" ) @@ -101,7 +102,7 @@ func (c *DirentCache) remove(d *Dirent) { panic(fmt.Sprintf("trying to remove %v, which is not in the dirent cache", d)) } c.list.Remove(d) - d.DecRef() + d.DecRef(context.Background()) c.currentSize-- if c.limit != nil { c.limit.dec() diff --git a/pkg/sentry/fs/dirent_refs_test.go b/pkg/sentry/fs/dirent_refs_test.go index 98d69c6f2..176b894ba 100644 --- a/pkg/sentry/fs/dirent_refs_test.go +++ b/pkg/sentry/fs/dirent_refs_test.go @@ -51,7 +51,7 @@ func TestWalkPositive(t *testing.T) { t.Fatalf("child name = %q has a ref count of %d, want %d", d.name, got, 1) } - d.DecRef() + d.DecRef(ctx) if got := root.ReadRefs(); got != 1 { t.Fatalf("root has a ref count of %d, want %d", got, 1) @@ -61,7 +61,7 @@ func TestWalkPositive(t *testing.T) { t.Fatalf("child name = %q has a ref count of %d, want %d", d.name, got, 0) } - root.flush() + root.flush(ctx) if got := len(root.children); got != 0 { t.Fatalf("root has %d children, want %d", got, 0) @@ -114,7 +114,7 @@ func TestWalkNegative(t *testing.T) { t.Fatalf("child has a ref count of %d, want %d", got, 2) } - child.DecRef() + child.DecRef(ctx) if got := child.(*Dirent).ReadRefs(); got != 1 { t.Fatalf("child has a ref count of %d, want %d", got, 1) @@ -124,7 +124,7 @@ func TestWalkNegative(t *testing.T) { t.Fatalf("root has %d children, want %d", got, 1) } - root.DecRef() + root.DecRef(ctx) if got := root.ReadRefs(); got != 0 { t.Fatalf("root has a ref count of %d, want %d", got, 0) @@ -351,9 +351,9 @@ func TestRemoveExtraRefs(t *testing.T) { t.Fatalf("dirent has a ref count of %d, want %d", got, 1) } - d.DecRef() + d.DecRef(ctx) - test.root.flush() + test.root.flush(ctx) if got := len(test.root.children); got != 0 { t.Errorf("root has %d children, want %d", got, 0) @@ -403,8 +403,8 @@ func TestRenameExtraRefs(t *testing.T) { t.Fatalf("Rename got error %v, want nil", err) } - oldParent.flush() - newParent.flush() + oldParent.flush(ctx) + newParent.flush(ctx) // Expect to have only active references. if got := renamed.ReadRefs(); got != 1 { diff --git a/pkg/sentry/fs/dirent_state.go b/pkg/sentry/fs/dirent_state.go index f623d6c0e..67a35f0b2 100644 --- a/pkg/sentry/fs/dirent_state.go +++ b/pkg/sentry/fs/dirent_state.go @@ -18,6 +18,7 @@ import ( "fmt" "sync/atomic" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" ) @@ -48,7 +49,7 @@ func (d *Dirent) saveChildren() map[string]*Dirent { for name, w := range d.children { if rc := w.Get(); rc != nil { // Drop the reference count obtain in w.Get() - rc.DecRef() + rc.DecRef(context.Background()) cd := rc.(*Dirent) if cd.IsNegative() { diff --git a/pkg/sentry/fs/fdpipe/pipe.go b/pkg/sentry/fs/fdpipe/pipe.go index 9fce177ad..b99199798 100644 --- a/pkg/sentry/fs/fdpipe/pipe.go +++ b/pkg/sentry/fs/fdpipe/pipe.go @@ -115,7 +115,7 @@ func (p *pipeOperations) Readiness(mask waiter.EventMask) (eventMask waiter.Even } // Release implements fs.FileOperations.Release. -func (p *pipeOperations) Release() { +func (p *pipeOperations) Release(context.Context) { fdnotifier.RemoveFD(int32(p.file.FD())) p.file.Close() p.file = nil diff --git a/pkg/sentry/fs/fdpipe/pipe_opener_test.go b/pkg/sentry/fs/fdpipe/pipe_opener_test.go index e556da48a..b9cec4b13 100644 --- a/pkg/sentry/fs/fdpipe/pipe_opener_test.go +++ b/pkg/sentry/fs/fdpipe/pipe_opener_test.go @@ -182,7 +182,7 @@ func TestTryOpen(t *testing.T) { // Cleanup the state of the pipe, and remove the fd from the // fdnotifier. Sadly this needed to maintain the correctness // of other tests because the fdnotifier is global. - pipeOps.Release() + pipeOps.Release(ctx) } continue } @@ -191,7 +191,7 @@ func TestTryOpen(t *testing.T) { } if pipeOps != nil { // Same as above. - pipeOps.Release() + pipeOps.Release(ctx) } } } @@ -279,7 +279,7 @@ func TestPipeOpenUnblocksEventually(t *testing.T) { pipeOps, err := Open(ctx, opener, flags) if pipeOps != nil { // Same as TestTryOpen. - pipeOps.Release() + pipeOps.Release(ctx) } // Check that the partner opened the file successfully. @@ -325,7 +325,7 @@ func TestCopiedReadAheadBuffer(t *testing.T) { ctx := contexttest.Context(t) pipeOps, err := pipeOpenState.TryOpen(ctx, opener, fs.FileFlags{Read: true}) if pipeOps != nil { - pipeOps.Release() + pipeOps.Release(ctx) t.Fatalf("open(%s, %o) got file, want nil", name, syscall.O_RDONLY) } if err != syserror.ErrWouldBlock { @@ -351,7 +351,7 @@ func TestCopiedReadAheadBuffer(t *testing.T) { if pipeOps == nil { t.Fatalf("open(%s, %o) got nil file, want not nil", name, syscall.O_RDONLY) } - defer pipeOps.Release() + defer pipeOps.Release(ctx) if err != nil { t.Fatalf("open(%s, %o) got error %v, want nil", name, syscall.O_RDONLY, err) @@ -471,14 +471,14 @@ func TestPipeHangup(t *testing.T) { f := <-fdchan if f < 0 { t.Errorf("%s: partner routine got fd %d, want > 0", test.desc, f) - pipeOps.Release() + pipeOps.Release(ctx) continue } if test.hangupSelf { // Hangup self and assert that our partner got the expected hangup // error. - pipeOps.Release() + pipeOps.Release(ctx) if test.flags.Read { // Partner is writer. @@ -490,7 +490,7 @@ func TestPipeHangup(t *testing.T) { } else { // Hangup our partner and expect us to get the hangup error. syscall.Close(f) - defer pipeOps.Release() + defer pipeOps.Release(ctx) if test.flags.Read { assertReaderHungup(t, test.desc, pipeOps.(*pipeOperations).file) diff --git a/pkg/sentry/fs/fdpipe/pipe_test.go b/pkg/sentry/fs/fdpipe/pipe_test.go index a0082ecca..1c9e82562 100644 --- a/pkg/sentry/fs/fdpipe/pipe_test.go +++ b/pkg/sentry/fs/fdpipe/pipe_test.go @@ -98,10 +98,11 @@ func TestNewPipe(t *testing.T) { } f := fd.New(gfd) - p, err := newPipeOperations(contexttest.Context(t), nil, test.flags, f, test.readAheadBuffer) + ctx := contexttest.Context(t) + p, err := newPipeOperations(ctx, nil, test.flags, f, test.readAheadBuffer) if p != nil { // This is necessary to remove the fd from the global fd notifier. - defer p.Release() + defer p.Release(ctx) } else { // If there is no p to DecRef on, because newPipeOperations failed, then the // file still needs to be closed. @@ -153,13 +154,14 @@ func TestPipeDestruction(t *testing.T) { syscall.Close(fds[1]) // Test the read end, but it doesn't really matter which. - p, err := newPipeOperations(contexttest.Context(t), nil, fs.FileFlags{Read: true}, f, nil) + ctx := contexttest.Context(t) + p, err := newPipeOperations(ctx, nil, fs.FileFlags{Read: true}, f, nil) if err != nil { f.Close() t.Fatalf("newPipeOperations got error %v, want nil", err) } // Drop our only reference, which should trigger the destructor. - p.Release() + p.Release(ctx) if fdnotifier.HasFD(int32(fds[0])) { t.Fatalf("after DecRef fdnotifier has fd %d, want no longer registered", fds[0]) @@ -282,7 +284,7 @@ func TestPipeRequest(t *testing.T) { if err != nil { t.Fatalf("%s: newPipeOperations got error %v, want nil", test.desc, err) } - defer p.Release() + defer p.Release(ctx) inode := fs.NewMockInode(ctx, fs.NewMockMountSource(nil), fs.StableAttr{Type: fs.Pipe}) file := fs.NewFile(ctx, fs.NewDirent(ctx, inode, "pipe"), fs.FileFlags{Read: true}, p) @@ -334,7 +336,7 @@ func TestPipeReadAheadBuffer(t *testing.T) { rfile.Close() t.Fatalf("newPipeOperations got error %v, want nil", err) } - defer p.Release() + defer p.Release(ctx) inode := fs.NewMockInode(ctx, fs.NewMockMountSource(nil), fs.StableAttr{ Type: fs.Pipe, @@ -380,7 +382,7 @@ func TestPipeReadsAccumulate(t *testing.T) { } // Don't forget to remove the fd from the fd notifier. Otherwise other tests will // likely be borked, because it's global :( - defer p.Release() + defer p.Release(ctx) inode := fs.NewMockInode(ctx, fs.NewMockMountSource(nil), fs.StableAttr{ Type: fs.Pipe, @@ -448,7 +450,7 @@ func TestPipeWritesAccumulate(t *testing.T) { } // Don't forget to remove the fd from the fd notifier. Otherwise other tests // will likely be borked, because it's global :( - defer p.Release() + defer p.Release(ctx) inode := fs.NewMockInode(ctx, fs.NewMockMountSource(nil), fs.StableAttr{ Type: fs.Pipe, diff --git a/pkg/sentry/fs/file.go b/pkg/sentry/fs/file.go index ca41520b4..72ea70fcf 100644 --- a/pkg/sentry/fs/file.go +++ b/pkg/sentry/fs/file.go @@ -142,17 +142,17 @@ func NewFile(ctx context.Context, dirent *Dirent, flags FileFlags, fops FileOper } // DecRef destroys the File when it is no longer referenced. -func (f *File) DecRef() { - f.DecRefWithDestructor(func() { +func (f *File) DecRef(ctx context.Context) { + f.DecRefWithDestructor(ctx, func(context.Context) { // Drop BSD style locks. lockRng := lock.LockRange{Start: 0, End: lock.LockEOF} f.Dirent.Inode.LockCtx.BSD.UnlockRegion(f, lockRng) // Release resources held by the FileOperations. - f.FileOperations.Release() + f.FileOperations.Release(ctx) // Release a reference on the Dirent. - f.Dirent.DecRef() + f.Dirent.DecRef(ctx) // Only unregister if we are currently registered. There is nothing // to register if f.async is nil (this happens when async mode is @@ -460,7 +460,7 @@ func (f *File) UnstableAttr(ctx context.Context) (UnstableAttr, error) { func (f *File) MappedName(ctx context.Context) string { root := RootFromContext(ctx) if root != nil { - defer root.DecRef() + defer root.DecRef(ctx) } name, _ := f.Dirent.FullName(root) return name diff --git a/pkg/sentry/fs/file_operations.go b/pkg/sentry/fs/file_operations.go index f5537411e..305c0f840 100644 --- a/pkg/sentry/fs/file_operations.go +++ b/pkg/sentry/fs/file_operations.go @@ -67,7 +67,7 @@ type SpliceOpts struct { // - File.Flags(): This value may change during the operation. type FileOperations interface { // Release release resources held by FileOperations. - Release() + Release(ctx context.Context) // Waitable defines how this File can be waited on for read and // write readiness. diff --git a/pkg/sentry/fs/file_overlay.go b/pkg/sentry/fs/file_overlay.go index dcc1df38f..9dc58d5ff 100644 --- a/pkg/sentry/fs/file_overlay.go +++ b/pkg/sentry/fs/file_overlay.go @@ -54,7 +54,7 @@ func overlayFile(ctx context.Context, inode *Inode, flags FileFlags) (*File, err // Drop the extra reference on the Dirent. Now there's only one reference // on the dirent, either owned by f (if non-nil), or the Dirent is about // to be destroyed (if GetFile failed). - dirent.DecRef() + dirent.DecRef(ctx) return f, err } @@ -89,12 +89,12 @@ type overlayFileOperations struct { } // Release implements FileOperations.Release. -func (f *overlayFileOperations) Release() { +func (f *overlayFileOperations) Release(ctx context.Context) { if f.upper != nil { - f.upper.DecRef() + f.upper.DecRef(ctx) } if f.lower != nil { - f.lower.DecRef() + f.lower.DecRef(ctx) } } @@ -164,7 +164,7 @@ func (f *overlayFileOperations) Seek(ctx context.Context, file *File, whence See func (f *overlayFileOperations) Readdir(ctx context.Context, file *File, serializer DentrySerializer) (int64, error) { root := RootFromContext(ctx) if root != nil { - defer root.DecRef() + defer root.DecRef(ctx) } dirCtx := &DirCtx{ @@ -497,7 +497,7 @@ func readdirOne(ctx context.Context, d *Dirent) (map[string]DentAttr, error) { if err != nil { return nil, err } - defer dir.DecRef() + defer dir.DecRef(ctx) // Use a stub serializer to read the entries into memory. stubSerializer := &CollectEntriesSerializer{} @@ -521,10 +521,10 @@ type overlayMappingIdentity struct { } // DecRef implements AtomicRefCount.DecRef. -func (omi *overlayMappingIdentity) DecRef() { - omi.AtomicRefCount.DecRefWithDestructor(func() { - omi.overlayFile.DecRef() - omi.id.DecRef() +func (omi *overlayMappingIdentity) DecRef(ctx context.Context) { + omi.AtomicRefCount.DecRefWithDestructor(ctx, func(context.Context) { + omi.overlayFile.DecRef(ctx) + omi.id.DecRef(ctx) }) } @@ -544,7 +544,7 @@ func (omi *overlayMappingIdentity) InodeID() uint64 { func (omi *overlayMappingIdentity) MappedName(ctx context.Context) string { root := RootFromContext(ctx) if root != nil { - defer root.DecRef() + defer root.DecRef(ctx) } name, _ := omi.overlayFile.Dirent.FullName(root) return name diff --git a/pkg/sentry/fs/fsutil/file.go b/pkg/sentry/fs/fsutil/file.go index 08695391c..dc9efa5df 100644 --- a/pkg/sentry/fs/fsutil/file.go +++ b/pkg/sentry/fs/fsutil/file.go @@ -31,7 +31,7 @@ import ( type FileNoopRelease struct{} // Release is a no-op. -func (FileNoopRelease) Release() {} +func (FileNoopRelease) Release(context.Context) {} // SeekWithDirCursor is used to implement fs.FileOperations.Seek. If dirCursor // is not nil and the seek was on a directory, the cursor will be updated. @@ -296,7 +296,7 @@ func (sdfo *StaticDirFileOperations) IterateDir(ctx context.Context, d *fs.Diren func (sdfo *StaticDirFileOperations) Readdir(ctx context.Context, file *fs.File, serializer fs.DentrySerializer) (int64, error) { root := fs.RootFromContext(ctx) if root != nil { - defer root.DecRef() + defer root.DecRef(ctx) } dirCtx := &fs.DirCtx{ Serializer: serializer, diff --git a/pkg/sentry/fs/gofer/file.go b/pkg/sentry/fs/gofer/file.go index b2fcab127..c0bc63a32 100644 --- a/pkg/sentry/fs/gofer/file.go +++ b/pkg/sentry/fs/gofer/file.go @@ -114,7 +114,7 @@ func NewFile(ctx context.Context, dirent *fs.Dirent, name string, flags fs.FileF } // Release implements fs.FileOpeations.Release. -func (f *fileOperations) Release() { +func (f *fileOperations) Release(context.Context) { f.handles.DecRef() } @@ -122,7 +122,7 @@ func (f *fileOperations) Release() { func (f *fileOperations) Readdir(ctx context.Context, file *fs.File, serializer fs.DentrySerializer) (int64, error) { root := fs.RootFromContext(ctx) if root != nil { - defer root.DecRef() + defer root.DecRef(ctx) } dirCtx := &fs.DirCtx{ diff --git a/pkg/sentry/fs/gofer/gofer_test.go b/pkg/sentry/fs/gofer/gofer_test.go index 2df2fe889..326fed954 100644 --- a/pkg/sentry/fs/gofer/gofer_test.go +++ b/pkg/sentry/fs/gofer/gofer_test.go @@ -232,7 +232,7 @@ func TestRevalidation(t *testing.T) { // We must release the dirent, of the test will fail // with a reference leak. This is tracked by p9test. - defer dirent.DecRef() + defer dirent.DecRef(ctx) // Walk again. Depending on the cache policy, we may // get a new dirent. @@ -246,7 +246,7 @@ func TestRevalidation(t *testing.T) { if !test.preModificationWantReload && dirent != newDirent { t.Errorf("Lookup with cachePolicy=%s got new dirent %+v, wanted old dirent %+v", test.cachePolicy, newDirent, dirent) } - newDirent.DecRef() // See above. + newDirent.DecRef(ctx) // See above. // Modify the underlying mocked file's modification // time for the next walk that occurs. @@ -287,7 +287,7 @@ func TestRevalidation(t *testing.T) { if test.postModificationWantUpdatedAttrs && gotModTimeSeconds != nowSeconds { t.Fatalf("Lookup with cachePolicy=%s got new modification time %v, wanted %v", test.cachePolicy, gotModTimeSeconds, nowSeconds) } - newDirent.DecRef() // See above. + newDirent.DecRef(ctx) // See above. // Remove the file from the remote fs, subsequent walks // should now fail to find anything. @@ -303,7 +303,7 @@ func TestRevalidation(t *testing.T) { t.Errorf("Lookup with cachePolicy=%s got new dirent and error %v, wanted old dirent and nil error", test.cachePolicy, err) } if err == nil { - newDirent.DecRef() // See above. + newDirent.DecRef(ctx) // See above. } }) } diff --git a/pkg/sentry/fs/gofer/handles.go b/pkg/sentry/fs/gofer/handles.go index fc14249be..f324dbf26 100644 --- a/pkg/sentry/fs/gofer/handles.go +++ b/pkg/sentry/fs/gofer/handles.go @@ -47,7 +47,8 @@ type handles struct { // DecRef drops a reference on handles. func (h *handles) DecRef() { - h.DecRefWithDestructor(func() { + ctx := context.Background() + h.DecRefWithDestructor(ctx, func(context.Context) { if h.Host != nil { if h.isHostBorrowed { h.Host.Release() @@ -57,7 +58,7 @@ func (h *handles) DecRef() { } } } - if err := h.File.close(context.Background()); err != nil { + if err := h.File.close(ctx); err != nil { log.Warningf("error closing p9 file: %v", err) } }) diff --git a/pkg/sentry/fs/gofer/inode.go b/pkg/sentry/fs/gofer/inode.go index 51d7368a1..3a225fd39 100644 --- a/pkg/sentry/fs/gofer/inode.go +++ b/pkg/sentry/fs/gofer/inode.go @@ -441,8 +441,9 @@ func (i *inodeOperations) Release(ctx context.Context) { // asynchronously. // // We use AsyncWithContext to avoid needing to allocate an extra - // anonymous function on the heap. - fs.AsyncWithContext(ctx, i.fileState.Release) + // anonymous function on the heap. We must use background context + // because the async work cannot happen on the task context. + fs.AsyncWithContext(context.Background(), i.fileState.Release) } // Mappable implements fs.InodeOperations.Mappable. diff --git a/pkg/sentry/fs/gofer/path.go b/pkg/sentry/fs/gofer/path.go index cf9800100..3c66dc3c2 100644 --- a/pkg/sentry/fs/gofer/path.go +++ b/pkg/sentry/fs/gofer/path.go @@ -168,7 +168,7 @@ func (i *inodeOperations) Create(ctx context.Context, dir *fs.Inode, name string // Construct the positive Dirent. d := fs.NewDirent(ctx, fs.NewInode(ctx, iops, dir.MountSource, sattr), name) - defer d.DecRef() + defer d.DecRef(ctx) // Construct the new file, caching the handles if allowed. h := handles{ @@ -371,7 +371,7 @@ func (i *inodeOperations) Remove(ctx context.Context, dir *fs.Inode, name string // Find out if file being deleted is a socket or pipe that needs to be // removed from endpoint map. if d, err := i.Lookup(ctx, dir, name); err == nil { - defer d.DecRef() + defer d.DecRef(ctx) if fs.IsSocket(d.Inode.StableAttr) || fs.IsPipe(d.Inode.StableAttr) { switch iops := d.Inode.InodeOperations.(type) { @@ -392,7 +392,7 @@ func (i *inodeOperations) Remove(ctx context.Context, dir *fs.Inode, name string return err } if key != nil { - i.session().overrides.remove(*key) + i.session().overrides.remove(ctx, *key) } i.touchModificationAndStatusChangeTime(ctx, dir) diff --git a/pkg/sentry/fs/gofer/session.go b/pkg/sentry/fs/gofer/session.go index b5efc86f2..7cf3522ff 100644 --- a/pkg/sentry/fs/gofer/session.go +++ b/pkg/sentry/fs/gofer/session.go @@ -89,10 +89,10 @@ func (e *overrideMaps) addPipe(key device.MultiDeviceKey, d *fs.Dirent, inode *f // remove deletes the key from the maps. // // Precondition: maps must have been locked with 'lock'. -func (e *overrideMaps) remove(key device.MultiDeviceKey) { +func (e *overrideMaps) remove(ctx context.Context, key device.MultiDeviceKey) { endpoint := e.keyMap[key] delete(e.keyMap, key) - endpoint.dirent.DecRef() + endpoint.dirent.DecRef(ctx) } // lock blocks other addition and removal operations from happening while @@ -197,7 +197,7 @@ type session struct { } // Destroy tears down the session. -func (s *session) Destroy() { +func (s *session) Destroy(ctx context.Context) { s.client.Close() } @@ -329,7 +329,7 @@ func Root(ctx context.Context, dev string, filesystem fs.Filesystem, superBlockF s.client, err = p9.NewClient(conn, s.msize, s.version) if err != nil { // Drop our reference on the session, it needs to be torn down. - s.DecRef() + s.DecRef(ctx) return nil, err } @@ -340,7 +340,7 @@ func Root(ctx context.Context, dev string, filesystem fs.Filesystem, superBlockF ctx.UninterruptibleSleepFinish(false) if err != nil { // Same as above. - s.DecRef() + s.DecRef(ctx) return nil, err } @@ -348,7 +348,7 @@ func Root(ctx context.Context, dev string, filesystem fs.Filesystem, superBlockF if err != nil { s.attach.close(ctx) // Same as above, but after we execute the Close request. - s.DecRef() + s.DecRef(ctx) return nil, err } @@ -393,13 +393,13 @@ func (s *session) fillKeyMap(ctx context.Context) error { // fillPathMap populates paths for overrides from dirents in direntMap // before save. -func (s *session) fillPathMap() error { +func (s *session) fillPathMap(ctx context.Context) error { unlock := s.overrides.lock() defer unlock() for _, endpoint := range s.overrides.keyMap { mountRoot := endpoint.dirent.MountRoot() - defer mountRoot.DecRef() + defer mountRoot.DecRef(ctx) dirPath, _ := endpoint.dirent.FullName(mountRoot) if dirPath == "" { return fmt.Errorf("error getting path from dirent") diff --git a/pkg/sentry/fs/gofer/session_state.go b/pkg/sentry/fs/gofer/session_state.go index 2d398b753..48b423dd8 100644 --- a/pkg/sentry/fs/gofer/session_state.go +++ b/pkg/sentry/fs/gofer/session_state.go @@ -26,7 +26,8 @@ import ( // beforeSave is invoked by stateify. func (s *session) beforeSave() { if s.overrides != nil { - if err := s.fillPathMap(); err != nil { + ctx := &dummyClockContext{context.Background()} + if err := s.fillPathMap(ctx); err != nil { panic("failed to save paths to override map before saving" + err.Error()) } } diff --git a/pkg/sentry/fs/gofer/socket.go b/pkg/sentry/fs/gofer/socket.go index 40f2c1cad..8a1c69ac2 100644 --- a/pkg/sentry/fs/gofer/socket.go +++ b/pkg/sentry/fs/gofer/socket.go @@ -134,14 +134,14 @@ func (e *endpoint) UnidirectionalConnect(ctx context.Context) (transport.Connect // We don't need the receiver. c.CloseRecv() - c.Release() + c.Release(ctx) return c, nil } // Release implements transport.BoundEndpoint.Release. -func (e *endpoint) Release() { - e.inode.DecRef() +func (e *endpoint) Release(ctx context.Context) { + e.inode.DecRef(ctx) } // Passcred implements transport.BoundEndpoint.Passcred. diff --git a/pkg/sentry/fs/host/control.go b/pkg/sentry/fs/host/control.go index 39299b7e4..0d8d36afa 100644 --- a/pkg/sentry/fs/host/control.go +++ b/pkg/sentry/fs/host/control.go @@ -57,7 +57,7 @@ func (c *scmRights) Clone() transport.RightsControlMessage { } // Release implements transport.RightsControlMessage.Release. -func (c *scmRights) Release() { +func (c *scmRights) Release(ctx context.Context) { for _, fd := range c.fds { syscall.Close(fd) } diff --git a/pkg/sentry/fs/host/file.go b/pkg/sentry/fs/host/file.go index 3e48b8b2c..86d1a87f0 100644 --- a/pkg/sentry/fs/host/file.go +++ b/pkg/sentry/fs/host/file.go @@ -110,7 +110,7 @@ func newFileFromDonatedFD(ctx context.Context, donated int, saveable, isTTY bool name := fmt.Sprintf("host:[%d]", inode.StableAttr.InodeID) dirent := fs.NewDirent(ctx, inode, name) - defer dirent.DecRef() + defer dirent.DecRef(ctx) if isTTY { return newTTYFile(ctx, dirent, flags, iops), nil @@ -169,7 +169,7 @@ func (f *fileOperations) Readiness(mask waiter.EventMask) waiter.EventMask { func (f *fileOperations) Readdir(ctx context.Context, file *fs.File, serializer fs.DentrySerializer) (int64, error) { root := fs.RootFromContext(ctx) if root != nil { - defer root.DecRef() + defer root.DecRef(ctx) } dirCtx := &fs.DirCtx{ Serializer: serializer, diff --git a/pkg/sentry/fs/host/inode_test.go b/pkg/sentry/fs/host/inode_test.go index c507f57eb..41a23b5da 100644 --- a/pkg/sentry/fs/host/inode_test.go +++ b/pkg/sentry/fs/host/inode_test.go @@ -36,7 +36,7 @@ func TestCloseFD(t *testing.T) { if err != nil { t.Fatalf("Failed to create File: %v", err) } - file.DecRef() + file.DecRef(ctx) s := make([]byte, 10) if c, err := syscall.Read(p[0], s); c != 0 || err != nil { diff --git a/pkg/sentry/fs/host/socket.go b/pkg/sentry/fs/host/socket.go index cfb089e43..a2f3d5918 100644 --- a/pkg/sentry/fs/host/socket.go +++ b/pkg/sentry/fs/host/socket.go @@ -194,7 +194,7 @@ func newSocket(ctx context.Context, orgfd int, saveable bool) (*fs.File, error) } // Send implements transport.ConnectedEndpoint.Send. -func (c *ConnectedEndpoint) Send(data [][]byte, controlMessages transport.ControlMessages, from tcpip.FullAddress) (int64, bool, *syserr.Error) { +func (c *ConnectedEndpoint) Send(ctx context.Context, data [][]byte, controlMessages transport.ControlMessages, from tcpip.FullAddress) (int64, bool, *syserr.Error) { c.mu.RLock() defer c.mu.RUnlock() @@ -271,7 +271,7 @@ func (c *ConnectedEndpoint) EventUpdate() { } // Recv implements transport.Receiver.Recv. -func (c *ConnectedEndpoint) Recv(data [][]byte, creds bool, numRights int, peek bool) (int64, int64, transport.ControlMessages, bool, tcpip.FullAddress, bool, *syserr.Error) { +func (c *ConnectedEndpoint) Recv(ctx context.Context, data [][]byte, creds bool, numRights int, peek bool) (int64, int64, transport.ControlMessages, bool, tcpip.FullAddress, bool, *syserr.Error) { c.mu.RLock() defer c.mu.RUnlock() @@ -318,7 +318,7 @@ func (c *ConnectedEndpoint) Recv(data [][]byte, creds bool, numRights int, peek } // close releases all resources related to the endpoint. -func (c *ConnectedEndpoint) close() { +func (c *ConnectedEndpoint) close(context.Context) { fdnotifier.RemoveFD(int32(c.file.FD())) c.file.Close() c.file = nil @@ -374,8 +374,8 @@ func (c *ConnectedEndpoint) RecvMaxQueueSize() int64 { } // Release implements transport.ConnectedEndpoint.Release and transport.Receiver.Release. -func (c *ConnectedEndpoint) Release() { - c.ref.DecRefWithDestructor(c.close) +func (c *ConnectedEndpoint) Release(ctx context.Context) { + c.ref.DecRefWithDestructor(ctx, c.close) } // CloseUnread implements transport.ConnectedEndpoint.CloseUnread. diff --git a/pkg/sentry/fs/host/socket_test.go b/pkg/sentry/fs/host/socket_test.go index affdbcacb..9d58ea448 100644 --- a/pkg/sentry/fs/host/socket_test.go +++ b/pkg/sentry/fs/host/socket_test.go @@ -67,11 +67,12 @@ func TestSocketIsBlocking(t *testing.T) { if fl&syscall.O_NONBLOCK == syscall.O_NONBLOCK { t.Fatalf("Expected socket %v to be blocking", pair[1]) } - sock, err := newSocket(contexttest.Context(t), pair[0], false) + ctx := contexttest.Context(t) + sock, err := newSocket(ctx, pair[0], false) if err != nil { t.Fatalf("newSocket(%v) failed => %v", pair[0], err) } - defer sock.DecRef() + defer sock.DecRef(ctx) // Test that the socket now is non-blocking. if fl, err = getFl(pair[0]); err != nil { t.Fatalf("getFl: fcntl(%v, GETFL) => %v", pair[0], err) @@ -93,11 +94,12 @@ func TestSocketWritev(t *testing.T) { if err != nil { t.Fatalf("host socket creation failed: %v", err) } - socket, err := newSocket(contexttest.Context(t), pair[0], false) + ctx := contexttest.Context(t) + socket, err := newSocket(ctx, pair[0], false) if err != nil { t.Fatalf("newSocket(%v) => %v", pair[0], err) } - defer socket.DecRef() + defer socket.DecRef(ctx) buf := []byte("hello world\n") n, err := socket.Writev(contexttest.Context(t), usermem.BytesIOSequence(buf)) if err != nil { @@ -115,11 +117,12 @@ func TestSocketWritevLen0(t *testing.T) { if err != nil { t.Fatalf("host socket creation failed: %v", err) } - socket, err := newSocket(contexttest.Context(t), pair[0], false) + ctx := contexttest.Context(t) + socket, err := newSocket(ctx, pair[0], false) if err != nil { t.Fatalf("newSocket(%v) => %v", pair[0], err) } - defer socket.DecRef() + defer socket.DecRef(ctx) n, err := socket.Writev(contexttest.Context(t), usermem.BytesIOSequence(nil)) if err != nil { t.Fatalf("socket writev failed: %v", err) @@ -136,11 +139,12 @@ func TestSocketSendMsgLen0(t *testing.T) { if err != nil { t.Fatalf("host socket creation failed: %v", err) } - sfile, err := newSocket(contexttest.Context(t), pair[0], false) + ctx := contexttest.Context(t) + sfile, err := newSocket(ctx, pair[0], false) if err != nil { t.Fatalf("newSocket(%v) => %v", pair[0], err) } - defer sfile.DecRef() + defer sfile.DecRef(ctx) s := sfile.FileOperations.(socket.Socket) n, terr := s.SendMsg(nil, usermem.BytesIOSequence(nil), []byte{}, 0, false, ktime.Time{}, socket.ControlMessages{}) @@ -158,18 +162,19 @@ func TestListen(t *testing.T) { if err != nil { t.Fatalf("syscall.Socket(syscall.AF_UNIX, syscall.SOCK_STREAM, 0) => %v", err) } - sfile1, err := newSocket(contexttest.Context(t), pair[0], false) + ctx := contexttest.Context(t) + sfile1, err := newSocket(ctx, pair[0], false) if err != nil { t.Fatalf("newSocket(%v) => %v", pair[0], err) } - defer sfile1.DecRef() + defer sfile1.DecRef(ctx) socket1 := sfile1.FileOperations.(socket.Socket) - sfile2, err := newSocket(contexttest.Context(t), pair[1], false) + sfile2, err := newSocket(ctx, pair[1], false) if err != nil { t.Fatalf("newSocket(%v) => %v", pair[1], err) } - defer sfile2.DecRef() + defer sfile2.DecRef(ctx) socket2 := sfile2.FileOperations.(socket.Socket) // Socketpairs can not be listened to. @@ -185,11 +190,11 @@ func TestListen(t *testing.T) { if err != nil { t.Fatalf("syscall.Socket(syscall.AF_UNIX, syscall.SOCK_STREAM, 0) => %v", err) } - sfile3, err := newSocket(contexttest.Context(t), sock, false) + sfile3, err := newSocket(ctx, sock, false) if err != nil { t.Fatalf("newSocket(%v) => %v", sock, err) } - defer sfile3.DecRef() + defer sfile3.DecRef(ctx) socket3 := sfile3.FileOperations.(socket.Socket) // This socket is not bound so we can't listen on it. @@ -237,9 +242,10 @@ func TestRelease(t *testing.T) { } c := &ConnectedEndpoint{queue: &waiter.Queue{}, file: fd.New(f)} want := &ConnectedEndpoint{queue: c.queue} - want.ref.DecRef() + ctx := contexttest.Context(t) + want.ref.DecRef(ctx) fdnotifier.AddFD(int32(c.file.FD()), nil) - c.Release() + c.Release(ctx) if !reflect.DeepEqual(c, want) { t.Errorf("got = %#v, want = %#v", c, want) } diff --git a/pkg/sentry/fs/host/tty.go b/pkg/sentry/fs/host/tty.go index 82a02fcb2..b5229098c 100644 --- a/pkg/sentry/fs/host/tty.go +++ b/pkg/sentry/fs/host/tty.go @@ -113,12 +113,12 @@ func (t *TTYFileOperations) Write(ctx context.Context, file *fs.File, src userme } // Release implements fs.FileOperations.Release. -func (t *TTYFileOperations) Release() { +func (t *TTYFileOperations) Release(ctx context.Context) { t.mu.Lock() t.fgProcessGroup = nil t.mu.Unlock() - t.fileOperations.Release() + t.fileOperations.Release(ctx) } // Ioctl implements fs.FileOperations.Ioctl. diff --git a/pkg/sentry/fs/host/wait_test.go b/pkg/sentry/fs/host/wait_test.go index ce397a5e3..c143f4ce2 100644 --- a/pkg/sentry/fs/host/wait_test.go +++ b/pkg/sentry/fs/host/wait_test.go @@ -39,7 +39,7 @@ func TestWait(t *testing.T) { t.Fatalf("NewFile failed: %v", err) } - defer file.DecRef() + defer file.DecRef(ctx) r := file.Readiness(waiter.EventIn) if r != 0 { diff --git a/pkg/sentry/fs/inode.go b/pkg/sentry/fs/inode.go index a34fbc946..b79cd9877 100644 --- a/pkg/sentry/fs/inode.go +++ b/pkg/sentry/fs/inode.go @@ -96,13 +96,12 @@ func NewInode(ctx context.Context, iops InodeOperations, msrc *MountSource, satt } // DecRef drops a reference on the Inode. -func (i *Inode) DecRef() { - i.DecRefWithDestructor(i.destroy) +func (i *Inode) DecRef(ctx context.Context) { + i.DecRefWithDestructor(ctx, i.destroy) } // destroy releases the Inode and releases the msrc reference taken. -func (i *Inode) destroy() { - ctx := context.Background() +func (i *Inode) destroy(ctx context.Context) { if err := i.WriteOut(ctx); err != nil { // FIXME(b/65209558): Mark as warning again once noatime is // properly supported. @@ -122,12 +121,12 @@ func (i *Inode) destroy() { i.Watches.targetDestroyed() if i.overlay != nil { - i.overlay.release() + i.overlay.release(ctx) } else { i.InodeOperations.Release(ctx) } - i.MountSource.DecRef() + i.MountSource.DecRef(ctx) } // Mappable calls i.InodeOperations.Mappable. diff --git a/pkg/sentry/fs/inode_inotify.go b/pkg/sentry/fs/inode_inotify.go index efd3c962b..9911a00c2 100644 --- a/pkg/sentry/fs/inode_inotify.go +++ b/pkg/sentry/fs/inode_inotify.go @@ -17,6 +17,7 @@ package fs import ( "fmt" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sync" ) @@ -136,11 +137,11 @@ func (w *Watches) Notify(name string, events, cookie uint32) { } // Unpin unpins dirent from all watches in this set. -func (w *Watches) Unpin(d *Dirent) { +func (w *Watches) Unpin(ctx context.Context, d *Dirent) { w.mu.RLock() defer w.mu.RUnlock() for _, watch := range w.ws { - watch.Unpin(d) + watch.Unpin(ctx, d) } } diff --git a/pkg/sentry/fs/inode_overlay.go b/pkg/sentry/fs/inode_overlay.go index 537c8d257..dc2e353d9 100644 --- a/pkg/sentry/fs/inode_overlay.go +++ b/pkg/sentry/fs/inode_overlay.go @@ -85,7 +85,7 @@ func overlayLookup(ctx context.Context, parent *overlayEntry, inode *Inode, name upperInode = child.Inode upperInode.IncRef() } - child.DecRef() + child.DecRef(ctx) } // Are we done? @@ -108,7 +108,7 @@ func overlayLookup(ctx context.Context, parent *overlayEntry, inode *Inode, name entry, err := newOverlayEntry(ctx, upperInode, nil, false) if err != nil { // Don't leak resources. - upperInode.DecRef() + upperInode.DecRef(ctx) parent.copyMu.RUnlock() return nil, false, err } @@ -129,7 +129,7 @@ func overlayLookup(ctx context.Context, parent *overlayEntry, inode *Inode, name if err != nil && err != syserror.ENOENT { // Don't leak resources. if upperInode != nil { - upperInode.DecRef() + upperInode.DecRef(ctx) } parent.copyMu.RUnlock() return nil, false, err @@ -152,7 +152,7 @@ func overlayLookup(ctx context.Context, parent *overlayEntry, inode *Inode, name } } } - child.DecRef() + child.DecRef(ctx) } } @@ -183,7 +183,7 @@ func overlayLookup(ctx context.Context, parent *overlayEntry, inode *Inode, name // unnecessary because we don't need to copy-up and we will always // operate (e.g. read/write) on the upper Inode. if !IsDir(upperInode.StableAttr) { - lowerInode.DecRef() + lowerInode.DecRef(ctx) lowerInode = nil } } @@ -194,10 +194,10 @@ func overlayLookup(ctx context.Context, parent *overlayEntry, inode *Inode, name // Well, not quite, we failed at the last moment, how depressing. // Be sure not to leak resources. if upperInode != nil { - upperInode.DecRef() + upperInode.DecRef(ctx) } if lowerInode != nil { - lowerInode.DecRef() + lowerInode.DecRef(ctx) } parent.copyMu.RUnlock() return nil, false, err @@ -248,7 +248,7 @@ func overlayCreate(ctx context.Context, o *overlayEntry, parent *Dirent, name st // user) will clobber the real path for the underlying Inode. upperFile.Dirent.Inode.IncRef() upperDirent := NewTransientDirent(upperFile.Dirent.Inode) - upperFile.Dirent.DecRef() + upperFile.Dirent.DecRef(ctx) upperFile.Dirent = upperDirent // Create the overlay inode and dirent. We need this to construct the @@ -259,7 +259,7 @@ func overlayCreate(ctx context.Context, o *overlayEntry, parent *Dirent, name st // The overlay file created below with NewFile will take a reference on // the overlayDirent, and it should be the only thing holding a // reference at the time of creation, so we must drop this reference. - defer overlayDirent.DecRef() + defer overlayDirent.DecRef(ctx) // Create a new overlay file that wraps the upper file. flags.Pread = upperFile.Flags().Pread @@ -399,7 +399,7 @@ func overlayRename(ctx context.Context, o *overlayEntry, oldParent *Dirent, rena if !replaced.IsNegative() && IsDir(replaced.Inode.StableAttr) { children, err := readdirOne(ctx, replaced) if err != nil { - replaced.DecRef() + replaced.DecRef(ctx) return err } @@ -407,12 +407,12 @@ func overlayRename(ctx context.Context, o *overlayEntry, oldParent *Dirent, rena // included among the returned children, so we don't // need to bother checking for them. if len(children) > 0 { - replaced.DecRef() + replaced.DecRef(ctx) return syserror.ENOTEMPTY } } - replaced.DecRef() + replaced.DecRef(ctx) } } @@ -455,12 +455,12 @@ func overlayBind(ctx context.Context, o *overlayEntry, parent *Dirent, name stri // Grab the inode and drop the dirent, we don't need it. inode := d.Inode inode.IncRef() - d.DecRef() + d.DecRef(ctx) // Create a new overlay entry and dirent for the socket. entry, err := newOverlayEntry(ctx, inode, nil, false) if err != nil { - inode.DecRef() + inode.DecRef(ctx) return nil, err } // Use the parent's MountSource, since that corresponds to the overlay, @@ -672,7 +672,7 @@ func overlayGetlink(ctx context.Context, o *overlayEntry) (*Dirent, error) { // ground and claim that jumping around the filesystem like this // is not supported. name, _ := dirent.FullName(nil) - dirent.DecRef() + dirent.DecRef(ctx) // Claim that the path is not accessible. err = syserror.EACCES diff --git a/pkg/sentry/fs/inode_overlay_test.go b/pkg/sentry/fs/inode_overlay_test.go index 389c219d6..aa9851b26 100644 --- a/pkg/sentry/fs/inode_overlay_test.go +++ b/pkg/sentry/fs/inode_overlay_test.go @@ -316,7 +316,7 @@ func TestCacheFlush(t *testing.T) { t.Fatalf("NewMountNamespace failed: %v", err) } root := mns.Root() - defer root.DecRef() + defer root.DecRef(ctx) ctx = &rootContext{ Context: ctx, @@ -345,7 +345,7 @@ func TestCacheFlush(t *testing.T) { } // Drop the file reference. - file.DecRef() + file.DecRef(ctx) // Dirent should have 2 refs left. if got, want := dirent.ReadRefs(), 2; int(got) != want { @@ -361,7 +361,7 @@ func TestCacheFlush(t *testing.T) { } // Drop our ref. - dirent.DecRef() + dirent.DecRef(ctx) // We should be back to zero refs. if got, want := dirent.ReadRefs(), 0; int(got) != want { @@ -398,7 +398,7 @@ func (d *dir) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags if err != nil { return nil, err } - defer file.DecRef() + defer file.DecRef(ctx) // Wrap the file's FileOperations in a dirFile. fops := &dirFile{ FileOperations: file.FileOperations, diff --git a/pkg/sentry/fs/inotify.go b/pkg/sentry/fs/inotify.go index e3a715c1f..c5c07d564 100644 --- a/pkg/sentry/fs/inotify.go +++ b/pkg/sentry/fs/inotify.go @@ -80,7 +80,7 @@ func NewInotify(ctx context.Context) *Inotify { // Release implements FileOperations.Release. Release removes all watches and // frees all resources for an inotify instance. -func (i *Inotify) Release() { +func (i *Inotify) Release(ctx context.Context) { // We need to hold i.mu to avoid a race with concurrent calls to // Inotify.targetDestroyed from Watches. There's no risk of Watches // accessing this Inotify after the destructor ends, because we remove all @@ -93,7 +93,7 @@ func (i *Inotify) Release() { // the owner's destructor. w.target.Watches.Remove(w.ID()) // Don't leak any references to the target, held by pins in the watch. - w.destroy() + w.destroy(ctx) } } @@ -321,7 +321,7 @@ func (i *Inotify) AddWatch(target *Dirent, mask uint32) int32 { // // RmWatch looks up an inotify watch for the given 'wd' and configures the // target dirent to stop sending events to this inotify instance. -func (i *Inotify) RmWatch(wd int32) error { +func (i *Inotify) RmWatch(ctx context.Context, wd int32) error { i.mu.Lock() // Find the watch we were asked to removed. @@ -346,7 +346,7 @@ func (i *Inotify) RmWatch(wd int32) error { i.queueEvent(newEvent(watch.wd, "", linux.IN_IGNORED, 0)) // Remove all pins. - watch.destroy() + watch.destroy(ctx) return nil } diff --git a/pkg/sentry/fs/inotify_watch.go b/pkg/sentry/fs/inotify_watch.go index 900cba3ca..605423d22 100644 --- a/pkg/sentry/fs/inotify_watch.go +++ b/pkg/sentry/fs/inotify_watch.go @@ -18,6 +18,7 @@ import ( "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sync" ) @@ -105,12 +106,12 @@ func (w *Watch) Pin(d *Dirent) { // Unpin drops any extra refs held on dirent due to a previous Pin // call. Calling Unpin multiple times for the same dirent, or on a dirent // without a corresponding Pin call is a no-op. -func (w *Watch) Unpin(d *Dirent) { +func (w *Watch) Unpin(ctx context.Context, d *Dirent) { w.mu.Lock() defer w.mu.Unlock() if w.pins[d] { delete(w.pins, d) - d.DecRef() + d.DecRef(ctx) } } @@ -125,11 +126,11 @@ func (w *Watch) TargetDestroyed() { // this watch. Destroy does not cause any new events to be generated. The caller // is responsible for ensuring there are no outstanding references to this // watch. -func (w *Watch) destroy() { +func (w *Watch) destroy(ctx context.Context) { w.mu.Lock() defer w.mu.Unlock() for d := range w.pins { - d.DecRef() + d.DecRef(ctx) } w.pins = nil } diff --git a/pkg/sentry/fs/mount.go b/pkg/sentry/fs/mount.go index 37bae6810..ee69b10e8 100644 --- a/pkg/sentry/fs/mount.go +++ b/pkg/sentry/fs/mount.go @@ -51,7 +51,7 @@ type MountSourceOperations interface { DirentOperations // Destroy destroys the MountSource. - Destroy() + Destroy(ctx context.Context) // Below are MountSourceOperations that do not conform to Linux. @@ -165,16 +165,16 @@ func (msrc *MountSource) DecDirentRefs() { } } -func (msrc *MountSource) destroy() { +func (msrc *MountSource) destroy(ctx context.Context) { if c := msrc.DirentRefs(); c != 0 { panic(fmt.Sprintf("MountSource with non-zero direntRefs is being destroyed: %d", c)) } - msrc.MountSourceOperations.Destroy() + msrc.MountSourceOperations.Destroy(ctx) } // DecRef drops a reference on the MountSource. -func (msrc *MountSource) DecRef() { - msrc.DecRefWithDestructor(msrc.destroy) +func (msrc *MountSource) DecRef(ctx context.Context) { + msrc.DecRefWithDestructor(ctx, msrc.destroy) } // FlushDirentRefs drops all references held by the MountSource on Dirents. @@ -264,7 +264,7 @@ func (*SimpleMountSourceOperations) ResetInodeMappings() {} func (*SimpleMountSourceOperations) SaveInodeMapping(*Inode, string) {} // Destroy implements MountSourceOperations.Destroy. -func (*SimpleMountSourceOperations) Destroy() {} +func (*SimpleMountSourceOperations) Destroy(context.Context) {} // Info defines attributes of a filesystem. type Info struct { diff --git a/pkg/sentry/fs/mount_overlay.go b/pkg/sentry/fs/mount_overlay.go index 78e35b1e6..7badc75d6 100644 --- a/pkg/sentry/fs/mount_overlay.go +++ b/pkg/sentry/fs/mount_overlay.go @@ -115,9 +115,9 @@ func (o *overlayMountSourceOperations) SaveInodeMapping(inode *Inode, path strin } // Destroy drops references on the upper and lower MountSource. -func (o *overlayMountSourceOperations) Destroy() { - o.upper.DecRef() - o.lower.DecRef() +func (o *overlayMountSourceOperations) Destroy(ctx context.Context) { + o.upper.DecRef(ctx) + o.lower.DecRef(ctx) } // type overlayFilesystem is the filesystem for overlay mounts. diff --git a/pkg/sentry/fs/mount_test.go b/pkg/sentry/fs/mount_test.go index a3d10770b..6c296f5d0 100644 --- a/pkg/sentry/fs/mount_test.go +++ b/pkg/sentry/fs/mount_test.go @@ -18,6 +18,7 @@ import ( "fmt" "testing" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/contexttest" ) @@ -32,13 +33,13 @@ func cacheReallyContains(cache *DirentCache, d *Dirent) bool { return false } -func mountPathsAre(root *Dirent, got []*Mount, want ...string) error { +func mountPathsAre(ctx context.Context, root *Dirent, got []*Mount, want ...string) error { gotPaths := make(map[string]struct{}, len(got)) gotStr := make([]string, len(got)) for i, g := range got { if groot := g.Root(); groot != nil { name, _ := groot.FullName(root) - groot.DecRef() + groot.DecRef(ctx) gotStr[i] = name gotPaths[name] = struct{}{} } @@ -69,7 +70,7 @@ func TestMountSourceOnlyCachedOnce(t *testing.T) { t.Fatalf("NewMountNamespace failed: %v", err) } rootDirent := mm.Root() - defer rootDirent.DecRef() + defer rootDirent.DecRef(ctx) // Get a child of the root which we will mount over. Note that the // MockInodeOperations causes Walk to always succeed. @@ -125,7 +126,7 @@ func TestAllMountsUnder(t *testing.T) { t.Fatalf("NewMountNamespace failed: %v", err) } rootDirent := mm.Root() - defer rootDirent.DecRef() + defer rootDirent.DecRef(ctx) // Add mounts at the following paths: paths := []string{ @@ -150,14 +151,14 @@ func TestAllMountsUnder(t *testing.T) { if err := mm.Mount(ctx, d, submountInode); err != nil { t.Fatalf("could not mount at %q: %v", p, err) } - d.DecRef() + d.DecRef(ctx) } // mm root should contain all submounts (and does not include the root mount). rootMnt := mm.FindMount(rootDirent) submounts := mm.AllMountsUnder(rootMnt) allPaths := append(paths, "/") - if err := mountPathsAre(rootDirent, submounts, allPaths...); err != nil { + if err := mountPathsAre(ctx, rootDirent, submounts, allPaths...); err != nil { t.Error(err) } @@ -181,9 +182,9 @@ func TestAllMountsUnder(t *testing.T) { if err != nil { t.Fatalf("could not find path %q in mount manager: %v", "/foo", err) } - defer d.DecRef() + defer d.DecRef(ctx) submounts = mm.AllMountsUnder(mm.FindMount(d)) - if err := mountPathsAre(rootDirent, submounts, "/foo", "/foo/bar", "/foo/qux", "/foo/bar/baz"); err != nil { + if err := mountPathsAre(ctx, rootDirent, submounts, "/foo", "/foo/bar", "/foo/qux", "/foo/bar/baz"); err != nil { t.Error(err) } @@ -193,9 +194,9 @@ func TestAllMountsUnder(t *testing.T) { if err != nil { t.Fatalf("could not find path %q in mount manager: %v", "/waldo", err) } - defer waldo.DecRef() + defer waldo.DecRef(ctx) submounts = mm.AllMountsUnder(mm.FindMount(waldo)) - if err := mountPathsAre(rootDirent, submounts, "/waldo"); err != nil { + if err := mountPathsAre(ctx, rootDirent, submounts, "/waldo"); err != nil { t.Error(err) } } @@ -212,7 +213,7 @@ func TestUnmount(t *testing.T) { t.Fatalf("NewMountNamespace failed: %v", err) } rootDirent := mm.Root() - defer rootDirent.DecRef() + defer rootDirent.DecRef(ctx) // Add mounts at the following paths: paths := []string{ @@ -240,7 +241,7 @@ func TestUnmount(t *testing.T) { if err := mm.Mount(ctx, d, submountInode); err != nil { t.Fatalf("could not mount at %q: %v", p, err) } - d.DecRef() + d.DecRef(ctx) } allPaths := make([]string, len(paths)+1) @@ -259,13 +260,13 @@ func TestUnmount(t *testing.T) { if err := mm.Unmount(ctx, d, false); err != nil { t.Fatalf("could not unmount at %q: %v", p, err) } - d.DecRef() + d.DecRef(ctx) // Remove the path that has been unmounted and the check that the remaining // mounts are still there. allPaths = allPaths[:len(allPaths)-1] submounts := mm.AllMountsUnder(rootMnt) - if err := mountPathsAre(rootDirent, submounts, allPaths...); err != nil { + if err := mountPathsAre(ctx, rootDirent, submounts, allPaths...); err != nil { t.Error(err) } } diff --git a/pkg/sentry/fs/mounts.go b/pkg/sentry/fs/mounts.go index 3f2bd0e87..d741c4339 100644 --- a/pkg/sentry/fs/mounts.go +++ b/pkg/sentry/fs/mounts.go @@ -234,7 +234,7 @@ func (mns *MountNamespace) flushMountSourceRefsLocked() { // After destroy is called, the MountNamespace may continue to be referenced (for // example via /proc/mounts), but should free all resources and shouldn't have // Find* methods called. -func (mns *MountNamespace) destroy() { +func (mns *MountNamespace) destroy(ctx context.Context) { mns.mu.Lock() defer mns.mu.Unlock() @@ -247,13 +247,13 @@ func (mns *MountNamespace) destroy() { for _, mp := range mns.mounts { // Drop the mount reference on all mounted dirents. for ; mp != nil; mp = mp.previous { - mp.root.DecRef() + mp.root.DecRef(ctx) } } mns.mounts = nil // Drop reference on the root. - mns.root.DecRef() + mns.root.DecRef(ctx) // Ensure that root cannot be accessed via this MountNamespace any // more. @@ -265,8 +265,8 @@ func (mns *MountNamespace) destroy() { } // DecRef implements RefCounter.DecRef with destructor mns.destroy. -func (mns *MountNamespace) DecRef() { - mns.DecRefWithDestructor(mns.destroy) +func (mns *MountNamespace) DecRef(ctx context.Context) { + mns.DecRefWithDestructor(ctx, mns.destroy) } // withMountLocked prevents further walks to `node`, because `node` is about to @@ -312,7 +312,7 @@ func (mns *MountNamespace) Mount(ctx context.Context, mountPoint *Dirent, inode if err != nil { return err } - defer replacement.DecRef() + defer replacement.DecRef(ctx) // Set the mount's root dirent and id. parentMnt := mns.findMountLocked(mountPoint) @@ -394,7 +394,7 @@ func (mns *MountNamespace) Unmount(ctx context.Context, node *Dirent, detachOnly panic(fmt.Sprintf("Last mount in the chain must be a undo mount: %+v", prev)) } // Drop mount reference taken at the end of MountNamespace.Mount. - prev.root.DecRef() + prev.root.DecRef(ctx) } else { mns.mounts[prev.root] = prev } @@ -496,11 +496,11 @@ func (mns *MountNamespace) FindLink(ctx context.Context, root, wd *Dirent, path // non-directory root is hopeless. if current != root { if !IsDir(current.Inode.StableAttr) { - current.DecRef() // Drop reference from above. + current.DecRef(ctx) // Drop reference from above. return nil, syserror.ENOTDIR } if err := current.Inode.CheckPermission(ctx, PermMask{Execute: true}); err != nil { - current.DecRef() // Drop reference from above. + current.DecRef(ctx) // Drop reference from above. return nil, err } } @@ -511,12 +511,12 @@ func (mns *MountNamespace) FindLink(ctx context.Context, root, wd *Dirent, path // Allow failed walks to cache the dirent, because no // children will acquire a reference at the end. current.maybeExtendReference() - current.DecRef() + current.DecRef(ctx) return nil, err } // Drop old reference. - current.DecRef() + current.DecRef(ctx) if remainder != "" { // Ensure it's resolved, unless it's the last level. @@ -570,11 +570,11 @@ func (mns *MountNamespace) resolve(ctx context.Context, root, node *Dirent, rema case nil: // Make sure we didn't exhaust the traversal budget. if *remainingTraversals == 0 { - target.DecRef() + target.DecRef(ctx) return nil, syscall.ELOOP } - node.DecRef() // Drop the original reference. + node.DecRef(ctx) // Drop the original reference. return target, nil case syscall.ENOLINK: @@ -582,7 +582,7 @@ func (mns *MountNamespace) resolve(ctx context.Context, root, node *Dirent, rema return node, nil case ErrResolveViaReadlink: - defer node.DecRef() // See above. + defer node.DecRef(ctx) // See above. // First, check if we should traverse. if *remainingTraversals == 0 { @@ -608,7 +608,7 @@ func (mns *MountNamespace) resolve(ctx context.Context, root, node *Dirent, rema return d, err default: - node.DecRef() // Drop for err; see above. + node.DecRef(ctx) // Drop for err; see above. // Propagate the error. return nil, err diff --git a/pkg/sentry/fs/mounts_test.go b/pkg/sentry/fs/mounts_test.go index a69b41468..975d6cbc9 100644 --- a/pkg/sentry/fs/mounts_test.go +++ b/pkg/sentry/fs/mounts_test.go @@ -51,7 +51,7 @@ func TestFindLink(t *testing.T) { } root := mm.Root() - defer root.DecRef() + defer root.DecRef(ctx) foo, err := root.Walk(ctx, root, "foo") if err != nil { t.Fatalf("Error walking to foo: %v", err) diff --git a/pkg/sentry/fs/overlay.go b/pkg/sentry/fs/overlay.go index a8ae7d81d..35013a21b 100644 --- a/pkg/sentry/fs/overlay.go +++ b/pkg/sentry/fs/overlay.go @@ -107,7 +107,7 @@ func NewOverlayRoot(ctx context.Context, upper *Inode, lower *Inode, flags Mount msrc := newOverlayMountSource(ctx, upper.MountSource, lower.MountSource, flags) overlay, err := newOverlayEntry(ctx, upper, lower, true) if err != nil { - msrc.DecRef() + msrc.DecRef(ctx) return nil, err } @@ -130,7 +130,7 @@ func NewOverlayRootFile(ctx context.Context, upperMS *MountSource, lower *Inode, msrc := newOverlayMountSource(ctx, upperMS, lower.MountSource, flags) overlay, err := newOverlayEntry(ctx, nil, lower, true) if err != nil { - msrc.DecRef() + msrc.DecRef(ctx) return nil, err } return newOverlayInode(ctx, overlay, msrc), nil @@ -230,16 +230,16 @@ func newOverlayEntry(ctx context.Context, upper *Inode, lower *Inode, lowerExist }, nil } -func (o *overlayEntry) release() { +func (o *overlayEntry) release(ctx context.Context) { // We drop a reference on upper and lower file system Inodes // rather than releasing them, because in-memory filesystems // may hold an extra reference to these Inodes so that they // stay in memory. if o.upper != nil { - o.upper.DecRef() + o.upper.DecRef(ctx) } if o.lower != nil { - o.lower.DecRef() + o.lower.DecRef(ctx) } } diff --git a/pkg/sentry/fs/proc/fds.go b/pkg/sentry/fs/proc/fds.go index 35972e23c..45523adf8 100644 --- a/pkg/sentry/fs/proc/fds.go +++ b/pkg/sentry/fs/proc/fds.go @@ -56,11 +56,11 @@ func walkDescriptors(t *kernel.Task, p string, toInode func(*fs.File, kernel.FDF // readDescriptors reads fds in the task starting at offset, and calls the // toDentAttr callback for each to get a DentAttr, which it then emits. This is // a helper for implementing fs.InodeOperations.Readdir. -func readDescriptors(t *kernel.Task, c *fs.DirCtx, offset int64, toDentAttr func(int) fs.DentAttr) (int64, error) { +func readDescriptors(ctx context.Context, t *kernel.Task, c *fs.DirCtx, offset int64, toDentAttr func(int) fs.DentAttr) (int64, error) { var fds []int32 t.WithMuLocked(func(t *kernel.Task) { if fdTable := t.FDTable(); fdTable != nil { - fds = fdTable.GetFDs() + fds = fdTable.GetFDs(ctx) } }) @@ -116,7 +116,7 @@ func (f *fd) GetFile(context.Context, *fs.Dirent, fs.FileFlags) (*fs.File, error func (f *fd) Readlink(ctx context.Context, _ *fs.Inode) (string, error) { root := fs.RootFromContext(ctx) if root != nil { - defer root.DecRef() + defer root.DecRef(ctx) } n, _ := f.file.Dirent.FullName(root) return n, nil @@ -135,13 +135,7 @@ func (f *fd) Truncate(context.Context, *fs.Inode, int64) error { func (f *fd) Release(ctx context.Context) { f.Symlink.Release(ctx) - f.file.DecRef() -} - -// Close releases the reference on the file. -func (f *fd) Close() error { - f.file.DecRef() - return nil + f.file.DecRef(ctx) } // fdDir is an InodeOperations for /proc/TID/fd. @@ -227,7 +221,7 @@ func (f *fdDirFile) Readdir(ctx context.Context, file *fs.File, ser fs.DentrySer if f.isInfoFile { typ = fs.Symlink } - return readDescriptors(f.t, dirCtx, file.Offset(), func(fd int) fs.DentAttr { + return readDescriptors(ctx, f.t, dirCtx, file.Offset(), func(fd int) fs.DentAttr { return fs.GenericDentAttr(typ, device.ProcDevice) }) } @@ -261,7 +255,7 @@ func (fdid *fdInfoDir) Lookup(ctx context.Context, dir *fs.Inode, p string) (*fs // locks, and other data. For now we only have flags. // See https://www.kernel.org/doc/Documentation/filesystems/proc.txt flags := file.Flags().ToLinux() | fdFlags.ToLinuxFileFlags() - file.DecRef() + file.DecRef(ctx) contents := []byte(fmt.Sprintf("flags:\t0%o\n", flags)) return newStaticProcInode(ctx, dir.MountSource, contents) }) diff --git a/pkg/sentry/fs/proc/mounts.go b/pkg/sentry/fs/proc/mounts.go index 1fc9c703c..6a63c47b3 100644 --- a/pkg/sentry/fs/proc/mounts.go +++ b/pkg/sentry/fs/proc/mounts.go @@ -47,7 +47,7 @@ func forEachMount(t *kernel.Task, fn func(string, *fs.Mount)) { // The task has been destroyed. Nothing to show here. return } - defer rootDir.DecRef() + defer rootDir.DecRef(t) mnt := t.MountNamespace().FindMount(rootDir) if mnt == nil { @@ -64,7 +64,7 @@ func forEachMount(t *kernel.Task, fn func(string, *fs.Mount)) { continue // No longer valid. } mountPath, desc := mroot.FullName(rootDir) - mroot.DecRef() + mroot.DecRef(t) if !desc { // MountSources that are not descendants of the chroot jail are ignored. continue @@ -97,7 +97,7 @@ func (mif *mountInfoFile) ReadSeqFileData(ctx context.Context, handle seqfile.Se if mroot == nil { return // No longer valid. } - defer mroot.DecRef() + defer mroot.DecRef(ctx) // Format: // 36 35 98:0 /mnt1 /mnt2 rw,noatime master:1 - ext3 /dev/root rw,errors=continue @@ -216,7 +216,7 @@ func (mf *mountsFile) ReadSeqFileData(ctx context.Context, handle seqfile.SeqHan if root == nil { return // No longer valid. } - defer root.DecRef() + defer root.DecRef(ctx) flags := root.Inode.MountSource.Flags opts := "rw" diff --git a/pkg/sentry/fs/proc/net.go b/pkg/sentry/fs/proc/net.go index bd18177d4..83a43aa26 100644 --- a/pkg/sentry/fs/proc/net.go +++ b/pkg/sentry/fs/proc/net.go @@ -419,7 +419,7 @@ func (n *netUnix) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]s } sfile := s.(*fs.File) if family, _, _ := sfile.FileOperations.(socket.Socket).Type(); family != linux.AF_UNIX { - s.DecRef() + s.DecRef(ctx) // Not a unix socket. continue } @@ -479,7 +479,7 @@ func (n *netUnix) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]s } fmt.Fprintf(&buf, "\n") - s.DecRef() + s.DecRef(ctx) } data := []seqfile.SeqData{ @@ -574,7 +574,7 @@ func commonReadSeqFileDataTCP(ctx context.Context, n seqfile.SeqHandle, k *kerne panic(fmt.Sprintf("Found non-socket file in socket table: %+v", sfile)) } if family, stype, _ := sops.Type(); !(family == fa && stype == linux.SOCK_STREAM) { - s.DecRef() + s.DecRef(ctx) // Not tcp4 sockets. continue } @@ -664,7 +664,7 @@ func commonReadSeqFileDataTCP(ctx context.Context, n seqfile.SeqHandle, k *kerne fmt.Fprintf(&buf, "\n") - s.DecRef() + s.DecRef(ctx) } data := []seqfile.SeqData{ @@ -752,7 +752,7 @@ func (n *netUDP) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]se panic(fmt.Sprintf("Found non-socket file in socket table: %+v", sfile)) } if family, stype, _ := sops.Type(); family != linux.AF_INET || stype != linux.SOCK_DGRAM { - s.DecRef() + s.DecRef(ctx) // Not udp4 socket. continue } @@ -822,7 +822,7 @@ func (n *netUDP) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]se fmt.Fprintf(&buf, "\n") - s.DecRef() + s.DecRef(ctx) } data := []seqfile.SeqData{ diff --git a/pkg/sentry/fs/proc/proc.go b/pkg/sentry/fs/proc/proc.go index c659224a7..77e0e1d26 100644 --- a/pkg/sentry/fs/proc/proc.go +++ b/pkg/sentry/fs/proc/proc.go @@ -213,7 +213,7 @@ func (rpf *rootProcFile) Readdir(ctx context.Context, file *fs.File, ser fs.Dent // Add dot and dotdot. root := fs.RootFromContext(ctx) if root != nil { - defer root.DecRef() + defer root.DecRef(ctx) } dot, dotdot := file.Dirent.GetDotAttrs(root) names = append(names, ".", "..") diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go index 4bbe90198..9cf7f2a62 100644 --- a/pkg/sentry/fs/proc/task.go +++ b/pkg/sentry/fs/proc/task.go @@ -185,7 +185,7 @@ func (f *subtasksFile) Readdir(ctx context.Context, file *fs.File, ser fs.Dentry // Serialize "." and "..". root := fs.RootFromContext(ctx) if root != nil { - defer root.DecRef() + defer root.DecRef(ctx) } dot, dotdot := file.Dirent.GetDotAttrs(root) if err := dirCtx.DirEmit(".", dot); err != nil { @@ -295,7 +295,7 @@ func (e *exe) Readlink(ctx context.Context, inode *fs.Inode) (string, error) { if err != nil { return "", err } - defer exec.DecRef() + defer exec.DecRef(ctx) return exec.PathnameWithDeleted(ctx), nil } diff --git a/pkg/sentry/fs/ramfs/dir.go b/pkg/sentry/fs/ramfs/dir.go index bfa304552..f4fcddecb 100644 --- a/pkg/sentry/fs/ramfs/dir.go +++ b/pkg/sentry/fs/ramfs/dir.go @@ -219,7 +219,7 @@ func (d *Dir) Remove(ctx context.Context, _ *fs.Inode, name string) error { } // Remove our reference on the inode. - inode.DecRef() + inode.DecRef(ctx) return nil } @@ -250,7 +250,7 @@ func (d *Dir) RemoveDirectory(ctx context.Context, _ *fs.Inode, name string) err } // Remove our reference on the inode. - inode.DecRef() + inode.DecRef(ctx) return nil } @@ -326,7 +326,7 @@ func (d *Dir) Create(ctx context.Context, dir *fs.Inode, name string, flags fs.F // Create the Dirent and corresponding file. created := fs.NewDirent(ctx, inode, name) - defer created.DecRef() + defer created.DecRef(ctx) return created.Inode.GetFile(ctx, created, flags) } @@ -412,11 +412,11 @@ func (*Dir) Rename(ctx context.Context, inode *fs.Inode, oldParent *fs.Inode, ol } // Release implements fs.InodeOperation.Release. -func (d *Dir) Release(_ context.Context) { +func (d *Dir) Release(ctx context.Context) { // Drop references on all children. d.mu.Lock() for _, i := range d.children { - i.DecRef() + i.DecRef(ctx) } d.mu.Unlock() } @@ -456,7 +456,7 @@ func (dfo *dirFileOperations) IterateDir(ctx context.Context, d *fs.Dirent, dirC func (dfo *dirFileOperations) Readdir(ctx context.Context, file *fs.File, serializer fs.DentrySerializer) (int64, error) { root := fs.RootFromContext(ctx) if root != nil { - defer root.DecRef() + defer root.DecRef(ctx) } dirCtx := &fs.DirCtx{ Serializer: serializer, @@ -473,13 +473,13 @@ func hasChildren(ctx context.Context, inode *fs.Inode) (bool, error) { // dropped when that dirent is destroyed. inode.IncRef() d := fs.NewTransientDirent(inode) - defer d.DecRef() + defer d.DecRef(ctx) file, err := inode.GetFile(ctx, d, fs.FileFlags{Read: true}) if err != nil { return false, err } - defer file.DecRef() + defer file.DecRef(ctx) ser := &fs.CollectEntriesSerializer{} if err := file.Readdir(ctx, ser); err != nil { @@ -530,7 +530,7 @@ func Rename(ctx context.Context, oldParent fs.InodeOperations, oldName string, n if err != nil { return err } - inode.DecRef() + inode.DecRef(ctx) } // Be careful, we may have already grabbed this mutex above. diff --git a/pkg/sentry/fs/ramfs/tree_test.go b/pkg/sentry/fs/ramfs/tree_test.go index a6ed8b2c5..3e0d1e07e 100644 --- a/pkg/sentry/fs/ramfs/tree_test.go +++ b/pkg/sentry/fs/ramfs/tree_test.go @@ -67,7 +67,7 @@ func TestMakeDirectoryTree(t *testing.T) { continue } root := mm.Root() - defer mm.DecRef() + defer mm.DecRef(ctx) for _, p := range test.subdirs { maxTraversals := uint(0) diff --git a/pkg/sentry/fs/timerfd/timerfd.go b/pkg/sentry/fs/timerfd/timerfd.go index 88c344089..f362ca9b6 100644 --- a/pkg/sentry/fs/timerfd/timerfd.go +++ b/pkg/sentry/fs/timerfd/timerfd.go @@ -55,7 +55,7 @@ type TimerOperations struct { func NewFile(ctx context.Context, c ktime.Clock) *fs.File { dirent := fs.NewDirent(ctx, anon.NewInode(ctx), "anon_inode:[timerfd]") // Release the initial dirent reference after NewFile takes a reference. - defer dirent.DecRef() + defer dirent.DecRef(ctx) tops := &TimerOperations{} tops.timer = ktime.NewTimer(c, tops) // Timerfds reject writes, but the Write flag must be set in order to @@ -65,7 +65,7 @@ func NewFile(ctx context.Context, c ktime.Clock) *fs.File { } // Release implements fs.FileOperations.Release. -func (t *TimerOperations) Release() { +func (t *TimerOperations) Release(context.Context) { t.timer.Destroy() } diff --git a/pkg/sentry/fs/tmpfs/file_test.go b/pkg/sentry/fs/tmpfs/file_test.go index aaba35502..d4d613ea9 100644 --- a/pkg/sentry/fs/tmpfs/file_test.go +++ b/pkg/sentry/fs/tmpfs/file_test.go @@ -46,7 +46,7 @@ func newFile(ctx context.Context) *fs.File { func TestGrow(t *testing.T) { ctx := contexttest.Context(t) f := newFile(ctx) - defer f.DecRef() + defer f.DecRef(ctx) abuf := bytes.Repeat([]byte{'a'}, 68) n, err := f.Pwritev(ctx, usermem.BytesIOSequence(abuf), 0) diff --git a/pkg/sentry/fs/tty/dir.go b/pkg/sentry/fs/tty/dir.go index 108654827..463f6189e 100644 --- a/pkg/sentry/fs/tty/dir.go +++ b/pkg/sentry/fs/tty/dir.go @@ -132,7 +132,7 @@ func (d *dirInodeOperations) Release(ctx context.Context) { d.mu.Lock() defer d.mu.Unlock() - d.master.DecRef() + d.master.DecRef(ctx) if len(d.slaves) != 0 { panic(fmt.Sprintf("devpts directory still contains active terminals: %+v", d)) } @@ -263,7 +263,7 @@ func (d *dirInodeOperations) allocateTerminal(ctx context.Context) (*Terminal, e } // masterClose is called when the master end of t is closed. -func (d *dirInodeOperations) masterClose(t *Terminal) { +func (d *dirInodeOperations) masterClose(ctx context.Context, t *Terminal) { d.mu.Lock() defer d.mu.Unlock() @@ -277,7 +277,7 @@ func (d *dirInodeOperations) masterClose(t *Terminal) { panic(fmt.Sprintf("Terminal %+v doesn't exist in %+v?", t, d)) } - s.DecRef() + s.DecRef(ctx) delete(d.slaves, t.n) d.dentryMap.Remove(strconv.FormatUint(uint64(t.n), 10)) } @@ -322,7 +322,7 @@ func (df *dirFileOperations) IterateDir(ctx context.Context, d *fs.Dirent, dirCt func (df *dirFileOperations) Readdir(ctx context.Context, file *fs.File, serializer fs.DentrySerializer) (int64, error) { root := fs.RootFromContext(ctx) if root != nil { - defer root.DecRef() + defer root.DecRef(ctx) } dirCtx := &fs.DirCtx{ Serializer: serializer, diff --git a/pkg/sentry/fs/tty/fs.go b/pkg/sentry/fs/tty/fs.go index 8fe05ebe5..2d4d44bf3 100644 --- a/pkg/sentry/fs/tty/fs.go +++ b/pkg/sentry/fs/tty/fs.go @@ -108,4 +108,4 @@ func (superOperations) ResetInodeMappings() {} func (superOperations) SaveInodeMapping(*fs.Inode, string) {} // Destroy implements MountSourceOperations.Destroy. -func (superOperations) Destroy() {} +func (superOperations) Destroy(context.Context) {} diff --git a/pkg/sentry/fs/tty/master.go b/pkg/sentry/fs/tty/master.go index fe07fa929..e00746017 100644 --- a/pkg/sentry/fs/tty/master.go +++ b/pkg/sentry/fs/tty/master.go @@ -75,7 +75,7 @@ func newMasterInode(ctx context.Context, d *dirInodeOperations, owner fs.FileOwn } // Release implements fs.InodeOperations.Release. -func (mi *masterInodeOperations) Release(ctx context.Context) { +func (mi *masterInodeOperations) Release(context.Context) { } // Truncate implements fs.InodeOperations.Truncate. @@ -120,9 +120,9 @@ type masterFileOperations struct { var _ fs.FileOperations = (*masterFileOperations)(nil) // Release implements fs.FileOperations.Release. -func (mf *masterFileOperations) Release() { - mf.d.masterClose(mf.t) - mf.t.DecRef() +func (mf *masterFileOperations) Release(ctx context.Context) { + mf.d.masterClose(ctx, mf.t) + mf.t.DecRef(ctx) } // EventRegister implements waiter.Waitable.EventRegister. diff --git a/pkg/sentry/fs/tty/slave.go b/pkg/sentry/fs/tty/slave.go index 9871f6fc6..7c7292687 100644 --- a/pkg/sentry/fs/tty/slave.go +++ b/pkg/sentry/fs/tty/slave.go @@ -71,7 +71,7 @@ func newSlaveInode(ctx context.Context, d *dirInodeOperations, t *Terminal, owne // Release implements fs.InodeOperations.Release. func (si *slaveInodeOperations) Release(ctx context.Context) { - si.t.DecRef() + si.t.DecRef(ctx) } // Truncate implements fs.InodeOperations.Truncate. @@ -106,7 +106,7 @@ type slaveFileOperations struct { var _ fs.FileOperations = (*slaveFileOperations)(nil) // Release implements fs.FileOperations.Release. -func (sf *slaveFileOperations) Release() { +func (sf *slaveFileOperations) Release(context.Context) { } // EventRegister implements waiter.Waitable.EventRegister. diff --git a/pkg/sentry/fs/user/path.go b/pkg/sentry/fs/user/path.go index 397e96045..2f5a43b84 100644 --- a/pkg/sentry/fs/user/path.go +++ b/pkg/sentry/fs/user/path.go @@ -82,7 +82,7 @@ func resolve(ctx context.Context, mns *fs.MountNamespace, paths []string, name s // Caller has no root. Don't bother traversing anything. return "", syserror.ENOENT } - defer root.DecRef() + defer root.DecRef(ctx) for _, p := range paths { if !path.IsAbs(p) { // Relative paths aren't safe, no one should be using them. @@ -100,7 +100,7 @@ func resolve(ctx context.Context, mns *fs.MountNamespace, paths []string, name s if err != nil { return "", err } - defer d.DecRef() + defer d.DecRef(ctx) // Check that it is a regular file. if !fs.IsRegular(d.Inode.StableAttr) { @@ -121,7 +121,7 @@ func resolve(ctx context.Context, mns *fs.MountNamespace, paths []string, name s func resolveVFS2(ctx context.Context, creds *auth.Credentials, mns *vfs.MountNamespace, paths []string, name string) (string, error) { root := mns.Root() - defer root.DecRef() + defer root.DecRef(ctx) for _, p := range paths { if !path.IsAbs(p) { // Relative paths aren't safe, no one should be using them. @@ -148,7 +148,7 @@ func resolveVFS2(ctx context.Context, creds *auth.Credentials, mns *vfs.MountNam if err != nil { return "", err } - dentry.DecRef() + dentry.DecRef(ctx) return binPath, nil } diff --git a/pkg/sentry/fs/user/user.go b/pkg/sentry/fs/user/user.go index f4d525523..936fd3932 100644 --- a/pkg/sentry/fs/user/user.go +++ b/pkg/sentry/fs/user/user.go @@ -62,7 +62,7 @@ func getExecUserHome(ctx context.Context, rootMns *fs.MountNamespace, uid auth.K // doesn't exist we will return the default home directory. return defaultHome, nil } - defer dirent.DecRef() + defer dirent.DecRef(ctx) // Check read permissions on the file. if err := dirent.Inode.CheckPermission(ctx, fs.PermMask{Read: true}); err != nil { @@ -81,7 +81,7 @@ func getExecUserHome(ctx context.Context, rootMns *fs.MountNamespace, uid auth.K if err != nil { return "", err } - defer f.DecRef() + defer f.DecRef(ctx) r := &fileReader{ Ctx: ctx, @@ -105,7 +105,7 @@ func getExecUserHomeVFS2(ctx context.Context, mns *vfs.MountNamespace, uid auth. const defaultHome = "/" root := mns.Root() - defer root.DecRef() + defer root.DecRef(ctx) creds := auth.CredentialsFromContext(ctx) @@ -123,7 +123,7 @@ func getExecUserHomeVFS2(ctx context.Context, mns *vfs.MountNamespace, uid auth. if err != nil { return defaultHome, nil } - defer fd.DecRef() + defer fd.DecRef(ctx) r := &fileReaderVFS2{ ctx: ctx, diff --git a/pkg/sentry/fs/user/user_test.go b/pkg/sentry/fs/user/user_test.go index 7d8e9ac7c..12b786224 100644 --- a/pkg/sentry/fs/user/user_test.go +++ b/pkg/sentry/fs/user/user_test.go @@ -39,7 +39,7 @@ func createEtcPasswd(ctx context.Context, root *fs.Dirent, contents string, mode if err != nil { return err } - defer etc.DecRef() + defer etc.DecRef(ctx) switch mode.FileType() { case 0: // Don't create anything. @@ -49,7 +49,7 @@ func createEtcPasswd(ctx context.Context, root *fs.Dirent, contents string, mode if err != nil { return err } - defer passwd.DecRef() + defer passwd.DecRef(ctx) if _, err := passwd.Writev(ctx, usermem.BytesIOSequence([]byte(contents))); err != nil { return err } @@ -110,9 +110,9 @@ func TestGetExecUserHome(t *testing.T) { if err != nil { t.Fatalf("NewMountNamespace failed: %v", err) } - defer mns.DecRef() + defer mns.DecRef(ctx) root := mns.Root() - defer root.DecRef() + defer root.DecRef(ctx) ctx = fs.WithRoot(ctx, root) if err := createEtcPasswd(ctx, root, tc.passwdContents, tc.passwdMode); err != nil { diff --git a/pkg/sentry/fsbridge/bridge.go b/pkg/sentry/fsbridge/bridge.go index 8e7590721..7e61209ee 100644 --- a/pkg/sentry/fsbridge/bridge.go +++ b/pkg/sentry/fsbridge/bridge.go @@ -44,7 +44,7 @@ type File interface { IncRef() // DecRef decrements reference. - DecRef() + DecRef(ctx context.Context) } // Lookup provides a common interface to open files. diff --git a/pkg/sentry/fsbridge/fs.go b/pkg/sentry/fsbridge/fs.go index 093ce1fb3..9785fd62a 100644 --- a/pkg/sentry/fsbridge/fs.go +++ b/pkg/sentry/fsbridge/fs.go @@ -49,7 +49,7 @@ func (f *fsFile) PathnameWithDeleted(ctx context.Context) string { // global there. return "" } - defer root.DecRef() + defer root.DecRef(ctx) name, _ := f.file.Dirent.FullName(root) return name @@ -87,8 +87,8 @@ func (f *fsFile) IncRef() { } // DecRef implements File. -func (f *fsFile) DecRef() { - f.file.DecRef() +func (f *fsFile) DecRef(ctx context.Context) { + f.file.DecRef(ctx) } // fsLookup implements Lookup interface using fs.File. @@ -124,7 +124,7 @@ func (l *fsLookup) OpenPath(ctx context.Context, path string, opts vfs.OpenOptio if err != nil { return nil, err } - defer d.DecRef() + defer d.DecRef(ctx) if !resolveFinal && fs.IsSymlink(d.Inode.StableAttr) { return nil, syserror.ELOOP diff --git a/pkg/sentry/fsbridge/vfs.go b/pkg/sentry/fsbridge/vfs.go index 89168220a..323506d33 100644 --- a/pkg/sentry/fsbridge/vfs.go +++ b/pkg/sentry/fsbridge/vfs.go @@ -43,7 +43,7 @@ func NewVFSFile(file *vfs.FileDescription) File { // PathnameWithDeleted implements File. func (f *VFSFile) PathnameWithDeleted(ctx context.Context) string { root := vfs.RootFromContext(ctx) - defer root.DecRef() + defer root.DecRef(ctx) vfsObj := f.file.VirtualDentry().Mount().Filesystem().VirtualFilesystem() name, _ := vfsObj.PathnameWithDeleted(ctx, root, f.file.VirtualDentry()) @@ -86,8 +86,8 @@ func (f *VFSFile) IncRef() { } // DecRef implements File. -func (f *VFSFile) DecRef() { - f.file.DecRef() +func (f *VFSFile) DecRef(ctx context.Context) { + f.file.DecRef(ctx) } // FileDescription returns the FileDescription represented by f. It does not diff --git a/pkg/sentry/fsimpl/devpts/devpts.go b/pkg/sentry/fsimpl/devpts/devpts.go index e6fda2b4f..7169e91af 100644 --- a/pkg/sentry/fsimpl/devpts/devpts.go +++ b/pkg/sentry/fsimpl/devpts/devpts.go @@ -103,9 +103,9 @@ func (fstype FilesystemType) newFilesystem(vfsObj *vfs.VirtualFilesystem, creds } // Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { +func (fs *filesystem) Release(ctx context.Context) { fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) - fs.Filesystem.Release() + fs.Filesystem.Release(ctx) } // rootInode is the root directory inode for the devpts mounts. diff --git a/pkg/sentry/fsimpl/devpts/master.go b/pkg/sentry/fsimpl/devpts/master.go index 1081fff52..3bb397f71 100644 --- a/pkg/sentry/fsimpl/devpts/master.go +++ b/pkg/sentry/fsimpl/devpts/master.go @@ -60,7 +60,7 @@ func (mi *masterInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vf } fd.LockFD.Init(&mi.locks) if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}); err != nil { - mi.DecRef() + mi.DecRef(ctx) return nil, err } return &fd.vfsfd, nil @@ -98,9 +98,9 @@ type masterFileDescription struct { var _ vfs.FileDescriptionImpl = (*masterFileDescription)(nil) // Release implements vfs.FileDescriptionImpl.Release. -func (mfd *masterFileDescription) Release() { +func (mfd *masterFileDescription) Release(ctx context.Context) { mfd.inode.root.masterClose(mfd.t) - mfd.inode.DecRef() + mfd.inode.DecRef(ctx) } // EventRegister implements waiter.Waitable.EventRegister. diff --git a/pkg/sentry/fsimpl/devpts/slave.go b/pkg/sentry/fsimpl/devpts/slave.go index a91cae3ef..32e4e1908 100644 --- a/pkg/sentry/fsimpl/devpts/slave.go +++ b/pkg/sentry/fsimpl/devpts/slave.go @@ -56,7 +56,7 @@ func (si *slaveInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs } fd.LockFD.Init(&si.locks) if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}); err != nil { - si.DecRef() + si.DecRef(ctx) return nil, err } return &fd.vfsfd, nil @@ -103,8 +103,8 @@ type slaveFileDescription struct { var _ vfs.FileDescriptionImpl = (*slaveFileDescription)(nil) // Release implements fs.FileOperations.Release. -func (sfd *slaveFileDescription) Release() { - sfd.inode.DecRef() +func (sfd *slaveFileDescription) Release(ctx context.Context) { + sfd.inode.DecRef(ctx) } // EventRegister implements waiter.Waitable.EventRegister. diff --git a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go index d0e06cdc0..2ed5fa8a9 100644 --- a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go +++ b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go @@ -92,9 +92,9 @@ func NewAccessor(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth } // Release must be called when a is no longer in use. -func (a *Accessor) Release() { - a.root.DecRef() - a.mntns.DecRef() +func (a *Accessor) Release(ctx context.Context) { + a.root.DecRef(ctx) + a.mntns.DecRef(ctx) } // accessorContext implements context.Context by extending an existing diff --git a/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go b/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go index b6d52c015..747867cca 100644 --- a/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go +++ b/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go @@ -30,7 +30,7 @@ func TestDevtmpfs(t *testing.T) { creds := auth.CredentialsFromContext(ctx) vfsObj := &vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { + if err := vfsObj.Init(ctx); err != nil { t.Fatalf("VFS init: %v", err) } // Register tmpfs just so that we can have a root filesystem that isn't @@ -48,9 +48,9 @@ func TestDevtmpfs(t *testing.T) { if err != nil { t.Fatalf("failed to create tmpfs root mount: %v", err) } - defer mntns.DecRef() + defer mntns.DecRef(ctx) root := mntns.Root() - defer root.DecRef() + defer root.DecRef(ctx) devpop := vfs.PathOperation{ Root: root, Start: root, @@ -69,7 +69,7 @@ func TestDevtmpfs(t *testing.T) { if err != nil { t.Fatalf("failed to create devtmpfs.Accessor: %v", err) } - defer a.Release() + defer a.Release(ctx) // Create "userspace-initialized" files using a devtmpfs.Accessor. if err := a.UserspaceInit(ctx); err != nil { diff --git a/pkg/sentry/fsimpl/eventfd/eventfd.go b/pkg/sentry/fsimpl/eventfd/eventfd.go index d12d78b84..812171fa3 100644 --- a/pkg/sentry/fsimpl/eventfd/eventfd.go +++ b/pkg/sentry/fsimpl/eventfd/eventfd.go @@ -59,9 +59,9 @@ type EventFileDescription struct { var _ vfs.FileDescriptionImpl = (*EventFileDescription)(nil) // New creates a new event fd. -func New(vfsObj *vfs.VirtualFilesystem, initVal uint64, semMode bool, flags uint32) (*vfs.FileDescription, error) { +func New(ctx context.Context, vfsObj *vfs.VirtualFilesystem, initVal uint64, semMode bool, flags uint32) (*vfs.FileDescription, error) { vd := vfsObj.NewAnonVirtualDentry("[eventfd]") - defer vd.DecRef() + defer vd.DecRef(ctx) efd := &EventFileDescription{ val: initVal, semMode: semMode, @@ -107,7 +107,7 @@ func (efd *EventFileDescription) HostFD() (int, error) { } // Release implements FileDescriptionImpl.Release() -func (efd *EventFileDescription) Release() { +func (efd *EventFileDescription) Release(context.Context) { efd.mu.Lock() defer efd.mu.Unlock() if efd.hostfd >= 0 { diff --git a/pkg/sentry/fsimpl/eventfd/eventfd_test.go b/pkg/sentry/fsimpl/eventfd/eventfd_test.go index 20e3adffc..49916fa81 100644 --- a/pkg/sentry/fsimpl/eventfd/eventfd_test.go +++ b/pkg/sentry/fsimpl/eventfd/eventfd_test.go @@ -36,16 +36,16 @@ func TestEventFD(t *testing.T) { for _, initVal := range initVals { ctx := contexttest.Context(t) vfsObj := &vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { + if err := vfsObj.Init(ctx); err != nil { t.Fatalf("VFS init: %v", err) } // Make a new eventfd that is writable. - eventfd, err := New(vfsObj, initVal, false, linux.O_RDWR) + eventfd, err := New(ctx, vfsObj, initVal, false, linux.O_RDWR) if err != nil { t.Fatalf("New() failed: %v", err) } - defer eventfd.DecRef() + defer eventfd.DecRef(ctx) // Register a callback for a write event. w, ch := waiter.NewChannelEntry(nil) @@ -74,16 +74,16 @@ func TestEventFD(t *testing.T) { func TestEventFDStat(t *testing.T) { ctx := contexttest.Context(t) vfsObj := &vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { + if err := vfsObj.Init(ctx); err != nil { t.Fatalf("VFS init: %v", err) } // Make a new eventfd that is writable. - eventfd, err := New(vfsObj, 0, false, linux.O_RDWR) + eventfd, err := New(ctx, vfsObj, 0, false, linux.O_RDWR) if err != nil { t.Fatalf("New() failed: %v", err) } - defer eventfd.DecRef() + defer eventfd.DecRef(ctx) statx, err := eventfd.Stat(ctx, vfs.StatOptions{ Mask: linux.STATX_BASIC_STATS, diff --git a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go index 89caee3df..8f7d5a9bb 100644 --- a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go +++ b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go @@ -53,7 +53,7 @@ func setUp(b *testing.B, imagePath string) (context.Context, *vfs.VirtualFilesys // Create VFS. vfsObj := &vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { + if err := vfsObj.Init(ctx); err != nil { return nil, nil, nil, nil, err } vfsObj.MustRegisterFilesystemType("extfs", ext.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ @@ -68,7 +68,7 @@ func setUp(b *testing.B, imagePath string) (context.Context, *vfs.VirtualFilesys root := mntns.Root() tearDown := func() { - root.DecRef() + root.DecRef(ctx) if err := f.Close(); err != nil { b.Fatalf("tearDown failed: %v", err) @@ -169,7 +169,7 @@ func BenchmarkVFS2ExtfsMountStat(b *testing.B) { if err != nil { b.Fatalf("failed to walk to mount point: %v", err) } - defer mountPoint.DecRef() + defer mountPoint.DecRef(ctx) // Create extfs submount. mountTearDown := mount(b, fmt.Sprintf("/tmp/image-%d.ext4", depth), vfsfs, &pop) diff --git a/pkg/sentry/fsimpl/ext/dentry.go b/pkg/sentry/fsimpl/ext/dentry.go index 55902322a..7a1b4219f 100644 --- a/pkg/sentry/fsimpl/ext/dentry.go +++ b/pkg/sentry/fsimpl/ext/dentry.go @@ -15,6 +15,7 @@ package ext import ( + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/vfs" ) @@ -55,7 +56,7 @@ func (d *dentry) TryIncRef() bool { } // DecRef implements vfs.DentryImpl.DecRef. -func (d *dentry) DecRef() { +func (d *dentry) DecRef(ctx context.Context) { // FIXME(b/134676337): filesystem.mu may not be locked as required by // inode.decRef(). d.inode.decRef() @@ -64,7 +65,7 @@ func (d *dentry) DecRef() { // InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. // // TODO(b/134676337): Implement inotify. -func (d *dentry) InotifyWithParent(events, cookie uint32, et vfs.EventType) {} +func (d *dentry) InotifyWithParent(ctx context.Context, events, cookie uint32, et vfs.EventType) {} // Watches implements vfs.DentryImpl.Watches. // @@ -76,4 +77,4 @@ func (d *dentry) Watches() *vfs.Watches { // OnZeroWatches implements vfs.Dentry.OnZeroWatches. // // TODO(b/134676337): Implement inotify. -func (d *dentry) OnZeroWatches() {} +func (d *dentry) OnZeroWatches(context.Context) {} diff --git a/pkg/sentry/fsimpl/ext/directory.go b/pkg/sentry/fsimpl/ext/directory.go index 357512c7e..0fc01668d 100644 --- a/pkg/sentry/fsimpl/ext/directory.go +++ b/pkg/sentry/fsimpl/ext/directory.go @@ -142,7 +142,7 @@ type directoryFD struct { var _ vfs.FileDescriptionImpl = (*directoryFD)(nil) // Release implements vfs.FileDescriptionImpl.Release. -func (fd *directoryFD) Release() { +func (fd *directoryFD) Release(ctx context.Context) { if fd.iter == nil { return } diff --git a/pkg/sentry/fsimpl/ext/ext.go b/pkg/sentry/fsimpl/ext/ext.go index dac6effbf..08ffc2834 100644 --- a/pkg/sentry/fsimpl/ext/ext.go +++ b/pkg/sentry/fsimpl/ext/ext.go @@ -123,32 +123,32 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt fs.vfsfs.Init(vfsObj, &fsType, &fs) fs.sb, err = readSuperBlock(dev) if err != nil { - fs.vfsfs.DecRef() + fs.vfsfs.DecRef(ctx) return nil, nil, err } if fs.sb.Magic() != linux.EXT_SUPER_MAGIC { // mount(2) specifies that EINVAL should be returned if the superblock is // invalid. - fs.vfsfs.DecRef() + fs.vfsfs.DecRef(ctx) return nil, nil, syserror.EINVAL } // Refuse to mount if the filesystem is incompatible. if !isCompatible(fs.sb) { - fs.vfsfs.DecRef() + fs.vfsfs.DecRef(ctx) return nil, nil, syserror.EINVAL } fs.bgs, err = readBlockGroups(dev, fs.sb) if err != nil { - fs.vfsfs.DecRef() + fs.vfsfs.DecRef(ctx) return nil, nil, err } rootInode, err := fs.getOrCreateInodeLocked(disklayout.RootDirInode) if err != nil { - fs.vfsfs.DecRef() + fs.vfsfs.DecRef(ctx) return nil, nil, err } rootInode.incRef() diff --git a/pkg/sentry/fsimpl/ext/ext_test.go b/pkg/sentry/fsimpl/ext/ext_test.go index 64e9a579f..2dbaee287 100644 --- a/pkg/sentry/fsimpl/ext/ext_test.go +++ b/pkg/sentry/fsimpl/ext/ext_test.go @@ -65,7 +65,7 @@ func setUp(t *testing.T, imagePath string) (context.Context, *vfs.VirtualFilesys // Create VFS. vfsObj := &vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { + if err := vfsObj.Init(ctx); err != nil { t.Fatalf("VFS init: %v", err) } vfsObj.MustRegisterFilesystemType("extfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ @@ -80,7 +80,7 @@ func setUp(t *testing.T, imagePath string) (context.Context, *vfs.VirtualFilesys root := mntns.Root() tearDown := func() { - root.DecRef() + root.DecRef(ctx) if err := f.Close(); err != nil { t.Fatalf("tearDown failed: %v", err) diff --git a/pkg/sentry/fsimpl/ext/filesystem.go b/pkg/sentry/fsimpl/ext/filesystem.go index 557963e03..c714ddf73 100644 --- a/pkg/sentry/fsimpl/ext/filesystem.go +++ b/pkg/sentry/fsimpl/ext/filesystem.go @@ -84,7 +84,7 @@ var _ vfs.FilesystemImpl = (*filesystem)(nil) // - filesystem.mu must be locked (for writing if write param is true). // - !rp.Done(). // - inode == vfsd.Impl().(*Dentry).inode. -func stepLocked(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, inode *inode, write bool) (*vfs.Dentry, *inode, error) { +func stepLocked(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, inode *inode, write bool) (*vfs.Dentry, *inode, error) { if !inode.isDir() { return nil, nil, syserror.ENOTDIR } @@ -100,7 +100,7 @@ func stepLocked(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, inode *inode, write boo } d := vfsd.Impl().(*dentry) if name == ".." { - isRoot, err := rp.CheckRoot(vfsd) + isRoot, err := rp.CheckRoot(ctx, vfsd) if err != nil { return nil, nil, err } @@ -108,7 +108,7 @@ func stepLocked(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, inode *inode, write boo rp.Advance() return vfsd, inode, nil } - if err := rp.CheckMount(&d.parent.vfsd); err != nil { + if err := rp.CheckMount(ctx, &d.parent.vfsd); err != nil { return nil, nil, err } rp.Advance() @@ -143,7 +143,7 @@ func stepLocked(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, inode *inode, write boo child.name = name dir.childCache[name] = child } - if err := rp.CheckMount(&child.vfsd); err != nil { + if err := rp.CheckMount(ctx, &child.vfsd); err != nil { return nil, nil, err } if child.inode.isSymlink() && rp.ShouldFollowSymlink() { @@ -167,12 +167,12 @@ func stepLocked(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, inode *inode, write boo // // Preconditions: // - filesystem.mu must be locked (for writing if write param is true). -func walkLocked(rp *vfs.ResolvingPath, write bool) (*vfs.Dentry, *inode, error) { +func walkLocked(ctx context.Context, rp *vfs.ResolvingPath, write bool) (*vfs.Dentry, *inode, error) { vfsd := rp.Start() inode := vfsd.Impl().(*dentry).inode for !rp.Done() { var err error - vfsd, inode, err = stepLocked(rp, vfsd, inode, write) + vfsd, inode, err = stepLocked(ctx, rp, vfsd, inode, write) if err != nil { return nil, nil, err } @@ -196,12 +196,12 @@ func walkLocked(rp *vfs.ResolvingPath, write bool) (*vfs.Dentry, *inode, error) // Preconditions: // - filesystem.mu must be locked (for writing if write param is true). // - !rp.Done(). -func walkParentLocked(rp *vfs.ResolvingPath, write bool) (*vfs.Dentry, *inode, error) { +func walkParentLocked(ctx context.Context, rp *vfs.ResolvingPath, write bool) (*vfs.Dentry, *inode, error) { vfsd := rp.Start() inode := vfsd.Impl().(*dentry).inode for !rp.Final() { var err error - vfsd, inode, err = stepLocked(rp, vfsd, inode, write) + vfsd, inode, err = stepLocked(ctx, rp, vfsd, inode, write) if err != nil { return nil, nil, err } @@ -216,7 +216,7 @@ func walkParentLocked(rp *vfs.ResolvingPath, write bool) (*vfs.Dentry, *inode, e // the rp till the parent of the last component which should be an existing // directory. If parent is false then resolves rp entirely. Attemps to resolve // the path as far as it can with a read lock and upgrades the lock if needed. -func (fs *filesystem) walk(rp *vfs.ResolvingPath, parent bool) (*vfs.Dentry, *inode, error) { +func (fs *filesystem) walk(ctx context.Context, rp *vfs.ResolvingPath, parent bool) (*vfs.Dentry, *inode, error) { var ( vfsd *vfs.Dentry inode *inode @@ -227,9 +227,9 @@ func (fs *filesystem) walk(rp *vfs.ResolvingPath, parent bool) (*vfs.Dentry, *in // of disk. This reduces congestion (allows concurrent walks). fs.mu.RLock() if parent { - vfsd, inode, err = walkParentLocked(rp, false) + vfsd, inode, err = walkParentLocked(ctx, rp, false) } else { - vfsd, inode, err = walkLocked(rp, false) + vfsd, inode, err = walkLocked(ctx, rp, false) } fs.mu.RUnlock() @@ -238,9 +238,9 @@ func (fs *filesystem) walk(rp *vfs.ResolvingPath, parent bool) (*vfs.Dentry, *in // walk is fine as this is a read only filesystem. fs.mu.Lock() if parent { - vfsd, inode, err = walkParentLocked(rp, true) + vfsd, inode, err = walkParentLocked(ctx, rp, true) } else { - vfsd, inode, err = walkLocked(rp, true) + vfsd, inode, err = walkLocked(ctx, rp, true) } fs.mu.Unlock() } @@ -283,7 +283,7 @@ func (fs *filesystem) statTo(stat *linux.Statfs) { // AccessAt implements vfs.Filesystem.Impl.AccessAt. func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error { - _, inode, err := fs.walk(rp, false) + _, inode, err := fs.walk(ctx, rp, false) if err != nil { return err } @@ -292,7 +292,7 @@ func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds // GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { - vfsd, inode, err := fs.walk(rp, false) + vfsd, inode, err := fs.walk(ctx, rp, false) if err != nil { return nil, err } @@ -312,7 +312,7 @@ func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, op // GetParentDentryAt implements vfs.FilesystemImpl.GetParentDentryAt. func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { - vfsd, inode, err := fs.walk(rp, true) + vfsd, inode, err := fs.walk(ctx, rp, true) if err != nil { return nil, err } @@ -322,7 +322,7 @@ func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPa // OpenAt implements vfs.FilesystemImpl.OpenAt. func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - vfsd, inode, err := fs.walk(rp, false) + vfsd, inode, err := fs.walk(ctx, rp, false) if err != nil { return nil, err } @@ -336,7 +336,7 @@ func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf // ReadlinkAt implements vfs.FilesystemImpl.ReadlinkAt. func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { - _, inode, err := fs.walk(rp, false) + _, inode, err := fs.walk(ctx, rp, false) if err != nil { return "", err } @@ -349,7 +349,7 @@ func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (st // StatAt implements vfs.FilesystemImpl.StatAt. func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { - _, inode, err := fs.walk(rp, false) + _, inode, err := fs.walk(ctx, rp, false) if err != nil { return linux.Statx{}, err } @@ -360,7 +360,7 @@ func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf // StatFSAt implements vfs.FilesystemImpl.StatFSAt. func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { - if _, _, err := fs.walk(rp, false); err != nil { + if _, _, err := fs.walk(ctx, rp, false); err != nil { return linux.Statfs{}, err } @@ -370,7 +370,7 @@ func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linu } // Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { +func (fs *filesystem) Release(ctx context.Context) { fs.vfsfs.VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) } @@ -390,7 +390,7 @@ func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. return syserror.EEXIST } - if _, _, err := fs.walk(rp, true); err != nil { + if _, _, err := fs.walk(ctx, rp, true); err != nil { return err } @@ -403,7 +403,7 @@ func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts v return syserror.EEXIST } - if _, _, err := fs.walk(rp, true); err != nil { + if _, _, err := fs.walk(ctx, rp, true); err != nil { return err } @@ -416,7 +416,7 @@ func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v return syserror.EEXIST } - _, _, err := fs.walk(rp, true) + _, _, err := fs.walk(ctx, rp, true) if err != nil { return err } @@ -430,7 +430,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa return syserror.ENOENT } - _, _, err := fs.walk(rp, false) + _, _, err := fs.walk(ctx, rp, false) if err != nil { return err } @@ -440,7 +440,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa // RmdirAt implements vfs.FilesystemImpl.RmdirAt. func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error { - _, inode, err := fs.walk(rp, false) + _, inode, err := fs.walk(ctx, rp, false) if err != nil { return err } @@ -454,7 +454,7 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error // SetStatAt implements vfs.FilesystemImpl.SetStatAt. func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { - _, _, err := fs.walk(rp, false) + _, _, err := fs.walk(ctx, rp, false) if err != nil { return err } @@ -468,7 +468,7 @@ func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, targ return syserror.EEXIST } - _, _, err := fs.walk(rp, true) + _, _, err := fs.walk(ctx, rp, true) if err != nil { return err } @@ -478,7 +478,7 @@ func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, targ // UnlinkAt implements vfs.FilesystemImpl.UnlinkAt. func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error { - _, inode, err := fs.walk(rp, false) + _, inode, err := fs.walk(ctx, rp, false) if err != nil { return err } @@ -492,7 +492,7 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error // BoundEndpointAt implements FilesystemImpl.BoundEndpointAt. func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.BoundEndpointOptions) (transport.BoundEndpoint, error) { - _, inode, err := fs.walk(rp, false) + _, inode, err := fs.walk(ctx, rp, false) if err != nil { return nil, err } @@ -506,7 +506,7 @@ func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath // ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, size uint64) ([]string, error) { - _, _, err := fs.walk(rp, false) + _, _, err := fs.walk(ctx, rp, false) if err != nil { return nil, err } @@ -515,7 +515,7 @@ func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, si // GetxattrAt implements vfs.FilesystemImpl.GetxattrAt. func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetxattrOptions) (string, error) { - _, _, err := fs.walk(rp, false) + _, _, err := fs.walk(ctx, rp, false) if err != nil { return "", err } @@ -524,7 +524,7 @@ func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt // SetxattrAt implements vfs.FilesystemImpl.SetxattrAt. func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { - _, _, err := fs.walk(rp, false) + _, _, err := fs.walk(ctx, rp, false) if err != nil { return err } @@ -533,7 +533,7 @@ func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt // RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { - _, _, err := fs.walk(rp, false) + _, _, err := fs.walk(ctx, rp, false) if err != nil { return err } diff --git a/pkg/sentry/fsimpl/ext/regular_file.go b/pkg/sentry/fsimpl/ext/regular_file.go index 66d14bb95..e73e740d6 100644 --- a/pkg/sentry/fsimpl/ext/regular_file.go +++ b/pkg/sentry/fsimpl/ext/regular_file.go @@ -79,7 +79,7 @@ type regularFileFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *regularFileFD) Release() {} +func (fd *regularFileFD) Release(context.Context) {} // PRead implements vfs.FileDescriptionImpl.PRead. func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { diff --git a/pkg/sentry/fsimpl/ext/symlink.go b/pkg/sentry/fsimpl/ext/symlink.go index 62efd4095..2fd0d1fa8 100644 --- a/pkg/sentry/fsimpl/ext/symlink.go +++ b/pkg/sentry/fsimpl/ext/symlink.go @@ -73,7 +73,7 @@ type symlinkFD struct { var _ vfs.FileDescriptionImpl = (*symlinkFD)(nil) // Release implements vfs.FileDescriptionImpl.Release. -func (fd *symlinkFD) Release() {} +func (fd *symlinkFD) Release(context.Context) {} // PRead implements vfs.FileDescriptionImpl.PRead. func (fd *symlinkFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { diff --git a/pkg/sentry/fsimpl/fuse/dev.go b/pkg/sentry/fsimpl/fuse/dev.go index 2225076bc..e522ff9a0 100644 --- a/pkg/sentry/fsimpl/fuse/dev.go +++ b/pkg/sentry/fsimpl/fuse/dev.go @@ -99,7 +99,7 @@ type DeviceFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *DeviceFD) Release() { +func (fd *DeviceFD) Release(context.Context) { fd.fs.conn.connected = false } diff --git a/pkg/sentry/fsimpl/fuse/dev_test.go b/pkg/sentry/fsimpl/fuse/dev_test.go index 84c222ad6..1ffe7ccd2 100644 --- a/pkg/sentry/fsimpl/fuse/dev_test.go +++ b/pkg/sentry/fsimpl/fuse/dev_test.go @@ -356,12 +356,12 @@ func newTestConnection(system *testutil.System, k *kernel.Kernel, maxActiveReque vfsObj := &vfs.VirtualFilesystem{} fuseDev := &DeviceFD{} - if err := vfsObj.Init(); err != nil { + if err := vfsObj.Init(system.Ctx); err != nil { return nil, nil, err } vd := vfsObj.NewAnonVirtualDentry("genCountFD") - defer vd.DecRef() + defer vd.DecRef(system.Ctx) if err := fuseDev.vfsfd.Init(fuseDev, linux.O_RDWR|linux.O_CREAT, vd.Mount(), vd.Dentry(), &vfs.FileDescriptionOptions{}); err != nil { return nil, nil, err } diff --git a/pkg/sentry/fsimpl/fuse/fusefs.go b/pkg/sentry/fsimpl/fuse/fusefs.go index 200a93bbf..a1405f7c3 100644 --- a/pkg/sentry/fsimpl/fuse/fusefs.go +++ b/pkg/sentry/fsimpl/fuse/fusefs.go @@ -191,9 +191,9 @@ func NewFUSEFilesystem(ctx context.Context, devMinor uint32, opts *filesystemOpt } // Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { +func (fs *filesystem) Release(ctx context.Context) { fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) - fs.Filesystem.Release() + fs.Filesystem.Release(ctx) } // inode implements kernfs.Inode. diff --git a/pkg/sentry/fsimpl/gofer/directory.go b/pkg/sentry/fsimpl/gofer/directory.go index 8c7c8e1b3..1679066ba 100644 --- a/pkg/sentry/fsimpl/gofer/directory.go +++ b/pkg/sentry/fsimpl/gofer/directory.go @@ -122,7 +122,7 @@ type directoryFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *directoryFD) Release() { +func (fd *directoryFD) Release(context.Context) { } // IterDirents implements vfs.FileDescriptionImpl.IterDirents. @@ -139,7 +139,7 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba fd.dirents = ds } - d.InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + d.InotifyWithParent(ctx, linux.IN_ACCESS, 0, vfs.PathEvent) if d.cachedMetadataAuthoritative() { d.touchAtime(fd.vfsfd.Mount()) } diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 00e3c99cd..e6af37d0d 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -55,7 +55,7 @@ func (fs *filesystem) Sync(ctx context.Context) error { // Sync regular files. for _, d := range ds { err := d.syncSharedHandle(ctx) - d.DecRef() + d.DecRef(ctx) if err != nil && retErr == nil { retErr = err } @@ -65,7 +65,7 @@ func (fs *filesystem) Sync(ctx context.Context) error { // handles (so they won't be synced by the above). for _, sffd := range sffds { err := sffd.Sync(ctx) - sffd.vfsfd.DecRef() + sffd.vfsfd.DecRef(ctx) if err != nil && retErr == nil { retErr = err } @@ -133,7 +133,7 @@ afterSymlink: return d, nil } if name == ".." { - if isRoot, err := rp.CheckRoot(&d.vfsd); err != nil { + if isRoot, err := rp.CheckRoot(ctx, &d.vfsd); err != nil { return nil, err } else if isRoot || d.parent == nil { rp.Advance() @@ -146,7 +146,7 @@ afterSymlink: // // Call rp.CheckMount() before updating d.parent's metadata, since if // we traverse to another mount then d.parent's metadata is irrelevant. - if err := rp.CheckMount(&d.parent.vfsd); err != nil { + if err := rp.CheckMount(ctx, &d.parent.vfsd); err != nil { return nil, err } if d != d.parent && !d.cachedMetadataAuthoritative() { @@ -164,7 +164,7 @@ afterSymlink: if child == nil { return nil, syserror.ENOENT } - if err := rp.CheckMount(&child.vfsd); err != nil { + if err := rp.CheckMount(ctx, &child.vfsd); err != nil { return nil, err } if child.isSymlink() && mayFollowSymlinks && rp.ShouldFollowSymlink() { @@ -239,7 +239,7 @@ func (fs *filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.Vir // has 0 references, drop it). Wait to update parent.children until we // know what to replace the existing dentry with (i.e. one of the // returns below), to avoid a redundant map access. - vfsObj.InvalidateDentry(&child.vfsd) + vfsObj.InvalidateDentry(ctx, &child.vfsd) if child.isSynthetic() { // Normally we don't mark invalidated dentries as deleted since // they may still exist (but at a different path), and also for @@ -332,7 +332,7 @@ func (fs *filesystem) resolveLocked(ctx context.Context, rp *vfs.ResolvingPath, func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir bool, createInRemoteDir func(parent *dentry, name string) error, createInSyntheticDir func(parent *dentry, name string) error) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) + defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds) start := rp.Start().Impl().(*dentry) if !start.cachedMetadataAuthoritative() { // Get updated metadata for start as required by @@ -384,7 +384,7 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir if dir { ev |= linux.IN_ISDIR } - parent.watches.Notify(name, uint32(ev), 0, vfs.InodeEvent, false /* unlinked */) + parent.watches.Notify(ctx, name, uint32(ev), 0, vfs.InodeEvent, false /* unlinked */) return nil } if fs.opts.interop == InteropModeShared { @@ -405,7 +405,7 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir if dir { ev |= linux.IN_ISDIR } - parent.watches.Notify(name, uint32(ev), 0, vfs.InodeEvent, false /* unlinked */) + parent.watches.Notify(ctx, name, uint32(ev), 0, vfs.InodeEvent, false /* unlinked */) return nil } if child := parent.children[name]; child != nil { @@ -426,7 +426,7 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir if dir { ev |= linux.IN_ISDIR } - parent.watches.Notify(name, uint32(ev), 0, vfs.InodeEvent, false /* unlinked */) + parent.watches.Notify(ctx, name, uint32(ev), 0, vfs.InodeEvent, false /* unlinked */) return nil } @@ -434,7 +434,7 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir bool) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) + defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds) start := rp.Start().Impl().(*dentry) if !start.cachedMetadataAuthoritative() { // Get updated metadata for start as required by @@ -470,7 +470,7 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b } vfsObj := rp.VirtualFilesystem() mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() + defer mntns.DecRef(ctx) parent.dirMu.Lock() defer parent.dirMu.Unlock() @@ -600,17 +600,17 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b // Generate inotify events for rmdir or unlink. if dir { - parent.watches.Notify(name, linux.IN_DELETE|linux.IN_ISDIR, 0, vfs.InodeEvent, true /* unlinked */) + parent.watches.Notify(ctx, name, linux.IN_DELETE|linux.IN_ISDIR, 0, vfs.InodeEvent, true /* unlinked */) } else { var cw *vfs.Watches if child != nil { cw = &child.watches } - vfs.InotifyRemoveChild(cw, &parent.watches, name) + vfs.InotifyRemoveChild(ctx, cw, &parent.watches, name) } if child != nil { - vfsObj.CommitDeleteDentry(&child.vfsd) + vfsObj.CommitDeleteDentry(ctx, &child.vfsd) child.setDeleted() if child.isSynthetic() { parent.syntheticChildren-- @@ -637,7 +637,7 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b // but dentry slices are allocated lazily, and it's much easier to say "defer // fs.renameMuRUnlockAndCheckCaching(&ds)" than "defer func() { // fs.renameMuRUnlockAndCheckCaching(ds) }()" to work around this. -func (fs *filesystem) renameMuRUnlockAndCheckCaching(ds **[]*dentry) { +func (fs *filesystem) renameMuRUnlockAndCheckCaching(ctx context.Context, ds **[]*dentry) { fs.renameMu.RUnlock() if *ds == nil { return @@ -645,20 +645,20 @@ func (fs *filesystem) renameMuRUnlockAndCheckCaching(ds **[]*dentry) { if len(**ds) != 0 { fs.renameMu.Lock() for _, d := range **ds { - d.checkCachingLocked() + d.checkCachingLocked(ctx) } fs.renameMu.Unlock() } putDentrySlice(*ds) } -func (fs *filesystem) renameMuUnlockAndCheckCaching(ds **[]*dentry) { +func (fs *filesystem) renameMuUnlockAndCheckCaching(ctx context.Context, ds **[]*dentry) { if *ds == nil { fs.renameMu.Unlock() return } for _, d := range **ds { - d.checkCachingLocked() + d.checkCachingLocked(ctx) } fs.renameMu.Unlock() putDentrySlice(*ds) @@ -668,7 +668,7 @@ func (fs *filesystem) renameMuUnlockAndCheckCaching(ds **[]*dentry) { func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) + defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return err @@ -680,7 +680,7 @@ func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) + defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return nil, err @@ -701,7 +701,7 @@ func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, op func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) + defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds) start := rp.Start().Impl().(*dentry) if !start.cachedMetadataAuthoritative() { // Get updated metadata for start as required by @@ -812,7 +812,7 @@ func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) + defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds) start := rp.Start().Impl().(*dentry) if !start.cachedMetadataAuthoritative() { @@ -1126,7 +1126,7 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving } childVFSFD = &fd.vfsfd } - d.watches.Notify(name, linux.IN_CREATE, 0, vfs.PathEvent, false /* unlinked */) + d.watches.Notify(ctx, name, linux.IN_CREATE, 0, vfs.PathEvent, false /* unlinked */) return childVFSFD, nil } @@ -1134,7 +1134,7 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) + defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return "", err @@ -1154,7 +1154,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa var ds *[]*dentry fs.renameMu.Lock() - defer fs.renameMuUnlockAndCheckCaching(&ds) + defer fs.renameMuUnlockAndCheckCaching(ctx, &ds) newParent, err := fs.walkParentDirLocked(ctx, rp, rp.Start().Impl().(*dentry), &ds) if err != nil { return err @@ -1244,7 +1244,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa return nil } mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() + defer mntns.DecRef(ctx) if err := vfsObj.PrepareRenameDentry(mntns, &renamed.vfsd, replacedVFSD); err != nil { return err } @@ -1269,7 +1269,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa } // Update the dentry tree. - vfsObj.CommitRenameReplaceDentry(&renamed.vfsd, replacedVFSD) + vfsObj.CommitRenameReplaceDentry(ctx, &renamed.vfsd, replacedVFSD) if replaced != nil { replaced.setDeleted() if replaced.isSynthetic() { @@ -1331,17 +1331,17 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts fs.renameMu.RLock() d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { - fs.renameMuRUnlockAndCheckCaching(&ds) + fs.renameMuRUnlockAndCheckCaching(ctx, &ds) return err } if err := d.setStat(ctx, rp.Credentials(), &opts, rp.Mount()); err != nil { - fs.renameMuRUnlockAndCheckCaching(&ds) + fs.renameMuRUnlockAndCheckCaching(ctx, &ds) return err } - fs.renameMuRUnlockAndCheckCaching(&ds) + fs.renameMuRUnlockAndCheckCaching(ctx, &ds) if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { - d.InotifyWithParent(ev, 0, vfs.InodeEvent) + d.InotifyWithParent(ctx, ev, 0, vfs.InodeEvent) } return nil } @@ -1350,7 +1350,7 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) + defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return linux.Statx{}, err @@ -1367,7 +1367,7 @@ func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) + defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return linux.Statfs{}, err @@ -1417,7 +1417,7 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.BoundEndpointOptions) (transport.BoundEndpoint, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) + defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return nil, err @@ -1443,7 +1443,7 @@ func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, size uint64) ([]string, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) + defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return nil, err @@ -1455,7 +1455,7 @@ func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, si func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetxattrOptions) (string, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) + defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return "", err @@ -1469,16 +1469,16 @@ func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt fs.renameMu.RLock() d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { - fs.renameMuRUnlockAndCheckCaching(&ds) + fs.renameMuRUnlockAndCheckCaching(ctx, &ds) return err } if err := d.setxattr(ctx, rp.Credentials(), &opts); err != nil { - fs.renameMuRUnlockAndCheckCaching(&ds) + fs.renameMuRUnlockAndCheckCaching(ctx, &ds) return err } - fs.renameMuRUnlockAndCheckCaching(&ds) + fs.renameMuRUnlockAndCheckCaching(ctx, &ds) - d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) + d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil } @@ -1488,16 +1488,16 @@ func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, fs.renameMu.RLock() d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { - fs.renameMuRUnlockAndCheckCaching(&ds) + fs.renameMuRUnlockAndCheckCaching(ctx, &ds) return err } if err := d.removexattr(ctx, rp.Credentials(), name); err != nil { - fs.renameMuRUnlockAndCheckCaching(&ds) + fs.renameMuRUnlockAndCheckCaching(ctx, &ds) return err } - fs.renameMuRUnlockAndCheckCaching(&ds) + fs.renameMuRUnlockAndCheckCaching(ctx, &ds) - d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) + d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil } diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index e20de84b5..2e5575d8d 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -482,7 +482,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt root, err := fs.newDentry(ctx, attachFile, qid, attrMask, &attr) if err != nil { attachFile.close(ctx) - fs.vfsfs.DecRef() + fs.vfsfs.DecRef(ctx) return nil, nil, err } // Set the root's reference count to 2. One reference is returned to the @@ -495,8 +495,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt } // Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { - ctx := context.Background() +func (fs *filesystem) Release(ctx context.Context) { mf := fs.mfp.MemoryFile() fs.syncMu.Lock() @@ -1089,10 +1088,10 @@ func (d *dentry) TryIncRef() bool { } // DecRef implements vfs.DentryImpl.DecRef. -func (d *dentry) DecRef() { +func (d *dentry) DecRef(ctx context.Context) { if refs := atomic.AddInt64(&d.refs, -1); refs == 0 { d.fs.renameMu.Lock() - d.checkCachingLocked() + d.checkCachingLocked(ctx) d.fs.renameMu.Unlock() } else if refs < 0 { panic("gofer.dentry.DecRef() called without holding a reference") @@ -1109,7 +1108,7 @@ func (d *dentry) decRefLocked() { } // InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. -func (d *dentry) InotifyWithParent(events, cookie uint32, et vfs.EventType) { +func (d *dentry) InotifyWithParent(ctx context.Context, events, cookie uint32, et vfs.EventType) { if d.isDir() { events |= linux.IN_ISDIR } @@ -1117,9 +1116,9 @@ func (d *dentry) InotifyWithParent(events, cookie uint32, et vfs.EventType) { d.fs.renameMu.RLock() // The ordering below is important, Linux always notifies the parent first. if d.parent != nil { - d.parent.watches.Notify(d.name, events, cookie, et, d.isDeleted()) + d.parent.watches.Notify(ctx, d.name, events, cookie, et, d.isDeleted()) } - d.watches.Notify("", events, cookie, et, d.isDeleted()) + d.watches.Notify(ctx, "", events, cookie, et, d.isDeleted()) d.fs.renameMu.RUnlock() } @@ -1131,10 +1130,10 @@ func (d *dentry) Watches() *vfs.Watches { // OnZeroWatches implements vfs.DentryImpl.OnZeroWatches. // // If no watches are left on this dentry and it has no references, cache it. -func (d *dentry) OnZeroWatches() { +func (d *dentry) OnZeroWatches(ctx context.Context) { if atomic.LoadInt64(&d.refs) == 0 { d.fs.renameMu.Lock() - d.checkCachingLocked() + d.checkCachingLocked(ctx) d.fs.renameMu.Unlock() } } @@ -1149,7 +1148,7 @@ func (d *dentry) OnZeroWatches() { // do nothing. // // Preconditions: d.fs.renameMu must be locked for writing. -func (d *dentry) checkCachingLocked() { +func (d *dentry) checkCachingLocked(ctx context.Context) { // Dentries with a non-zero reference count must be retained. (The only way // to obtain a reference on a dentry with zero references is via path // resolution, which requires renameMu, so if d.refs is zero then it will @@ -1171,14 +1170,14 @@ func (d *dentry) checkCachingLocked() { // reachable by path resolution and should be dropped immediately. if d.vfsd.IsDead() { if d.isDeleted() { - d.watches.HandleDeletion() + d.watches.HandleDeletion(ctx) } if d.cached { d.fs.cachedDentries.Remove(d) d.fs.cachedDentriesLen-- d.cached = false } - d.destroyLocked() + d.destroyLocked(ctx) return } // If d still has inotify watches and it is not deleted or invalidated, we @@ -1213,7 +1212,7 @@ func (d *dentry) checkCachingLocked() { if !victim.vfsd.IsDead() { // Note that victim can't be a mount point (in any mount // namespace), since VFS holds references on mount points. - d.fs.vfsfs.VirtualFilesystem().InvalidateDentry(&victim.vfsd) + d.fs.vfsfs.VirtualFilesystem().InvalidateDentry(ctx, &victim.vfsd) delete(victim.parent.children, victim.name) // We're only deleting the dentry, not the file it // represents, so we don't need to update @@ -1221,7 +1220,7 @@ func (d *dentry) checkCachingLocked() { } victim.parent.dirMu.Unlock() } - victim.destroyLocked() + victim.destroyLocked(ctx) } // Whether or not victim was destroyed, we brought fs.cachedDentriesLen // back down to fs.opts.maxCachedDentries, so we don't loop. @@ -1233,7 +1232,7 @@ func (d *dentry) checkCachingLocked() { // // Preconditions: d.fs.renameMu must be locked for writing. d.refs == 0. d is // not a child dentry. -func (d *dentry) destroyLocked() { +func (d *dentry) destroyLocked(ctx context.Context) { switch atomic.LoadInt64(&d.refs) { case 0: // Mark the dentry destroyed. @@ -1244,7 +1243,6 @@ func (d *dentry) destroyLocked() { panic("dentry.destroyLocked() called with references on the dentry") } - ctx := context.Background() d.handleMu.Lock() if !d.handle.file.isNil() { mf := d.fs.mfp.MemoryFile() @@ -1276,7 +1274,7 @@ func (d *dentry) destroyLocked() { // d.fs.renameMu. if d.parent != nil { if refs := atomic.AddInt64(&d.parent.refs, -1); refs == 0 { - d.parent.checkCachingLocked() + d.parent.checkCachingLocked(ctx) } else if refs < 0 { panic("gofer.dentry.DecRef() called without holding a reference") } @@ -1514,7 +1512,7 @@ func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) return err } if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { - fd.dentry().InotifyWithParent(ev, 0, vfs.InodeEvent) + fd.dentry().InotifyWithParent(ctx, ev, 0, vfs.InodeEvent) } return nil } @@ -1535,7 +1533,7 @@ func (fd *fileDescription) Setxattr(ctx context.Context, opts vfs.SetxattrOption if err := d.setxattr(ctx, auth.CredentialsFromContext(ctx), &opts); err != nil { return err } - d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) + d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil } @@ -1545,7 +1543,7 @@ func (fd *fileDescription) Removexattr(ctx context.Context, name string) error { if err := d.removexattr(ctx, auth.CredentialsFromContext(ctx), name); err != nil { return err } - d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) + d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil } diff --git a/pkg/sentry/fsimpl/gofer/gofer_test.go b/pkg/sentry/fsimpl/gofer/gofer_test.go index adff39490..56d80bcf8 100644 --- a/pkg/sentry/fsimpl/gofer/gofer_test.go +++ b/pkg/sentry/fsimpl/gofer/gofer_test.go @@ -50,7 +50,7 @@ func TestDestroyIdempotent(t *testing.T) { } parent.cacheNewChildLocked(child, "child") - child.checkCachingLocked() + child.checkCachingLocked(ctx) if got := atomic.LoadInt64(&child.refs); got != -1 { t.Fatalf("child.refs=%d, want: -1", got) } @@ -58,6 +58,6 @@ func TestDestroyIdempotent(t *testing.T) { if got := atomic.LoadInt64(&parent.refs); got != -1 { t.Fatalf("parent.refs=%d, want: -1", got) } - child.checkCachingLocked() - child.checkCachingLocked() + child.checkCachingLocked(ctx) + child.checkCachingLocked(ctx) } diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go index 09f142cfc..420e8efe2 100644 --- a/pkg/sentry/fsimpl/gofer/regular_file.go +++ b/pkg/sentry/fsimpl/gofer/regular_file.go @@ -48,7 +48,7 @@ type regularFileFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *regularFileFD) Release() { +func (fd *regularFileFD) Release(context.Context) { } // OnClose implements vfs.FileDescriptionImpl.OnClose. diff --git a/pkg/sentry/fsimpl/gofer/socket.go b/pkg/sentry/fsimpl/gofer/socket.go index d6dbe9092..85d2bee72 100644 --- a/pkg/sentry/fsimpl/gofer/socket.go +++ b/pkg/sentry/fsimpl/gofer/socket.go @@ -108,7 +108,7 @@ func (e *endpoint) UnidirectionalConnect(ctx context.Context) (transport.Connect // We don't need the receiver. c.CloseRecv() - c.Release() + c.Release(ctx) return c, nil } @@ -136,8 +136,8 @@ func (e *endpoint) newConnectedEndpoint(ctx context.Context, flags p9.ConnectFla } // Release implements transport.BoundEndpoint.Release. -func (e *endpoint) Release() { - e.dentry.DecRef() +func (e *endpoint) Release(ctx context.Context) { + e.dentry.DecRef(ctx) } // Passcred implements transport.BoundEndpoint.Passcred. diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go index 811528982..fc269ef2b 100644 --- a/pkg/sentry/fsimpl/gofer/special_file.go +++ b/pkg/sentry/fsimpl/gofer/special_file.go @@ -80,11 +80,11 @@ func newSpecialFileFD(h handle, mnt *vfs.Mount, d *dentry, locks *vfs.FileLocks, } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *specialFileFD) Release() { +func (fd *specialFileFD) Release(ctx context.Context) { if fd.haveQueue { fdnotifier.RemoveFD(fd.handle.fd) } - fd.handle.close(context.Background()) + fd.handle.close(ctx) fs := fd.vfsfd.Mount().Filesystem().Impl().(*filesystem) fs.syncMu.Lock() delete(fs.specialFileFDs, fd) diff --git a/pkg/sentry/fsimpl/host/control.go b/pkg/sentry/fsimpl/host/control.go index b9082a20f..0135e4428 100644 --- a/pkg/sentry/fsimpl/host/control.go +++ b/pkg/sentry/fsimpl/host/control.go @@ -58,7 +58,7 @@ func (c *scmRights) Clone() transport.RightsControlMessage { } // Release implements transport.RightsControlMessage.Release. -func (c *scmRights) Release() { +func (c *scmRights) Release(ctx context.Context) { for _, fd := range c.fds { syscall.Close(fd) } diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index c894f2ca0..bf922c566 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -117,7 +117,7 @@ func NewFD(ctx context.Context, mnt *vfs.Mount, hostFD int, opts *NewFDOptions) d.Init(i) // i.open will take a reference on d. - defer d.DecRef() + defer d.DecRef(ctx) // For simplicity, fileDescription.offset is set to 0. Technically, we // should only set to 0 on files that are not seekable (sockets, pipes, @@ -168,9 +168,9 @@ type filesystem struct { devMinor uint32 } -func (fs *filesystem) Release() { +func (fs *filesystem) Release(ctx context.Context) { fs.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) - fs.Filesystem.Release() + fs.Filesystem.Release(ctx) } func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { @@ -431,12 +431,12 @@ func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Cre } // DecRef implements kernfs.Inode. -func (i *inode) DecRef() { - i.AtomicRefCount.DecRefWithDestructor(i.Destroy) +func (i *inode) DecRef(ctx context.Context) { + i.AtomicRefCount.DecRefWithDestructor(ctx, i.Destroy) } // Destroy implements kernfs.Inode. -func (i *inode) Destroy() { +func (i *inode) Destroy(context.Context) { if i.wouldBlock { fdnotifier.RemoveFD(int32(i.hostFD)) } @@ -542,7 +542,7 @@ func (f *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux } // Release implements vfs.FileDescriptionImpl. -func (f *fileDescription) Release() { +func (f *fileDescription) Release(context.Context) { // noop } diff --git a/pkg/sentry/fsimpl/host/socket.go b/pkg/sentry/fsimpl/host/socket.go index fd16bd92d..4979dd0a9 100644 --- a/pkg/sentry/fsimpl/host/socket.go +++ b/pkg/sentry/fsimpl/host/socket.go @@ -139,7 +139,7 @@ func NewConnectedEndpoint(ctx context.Context, hostFD int, addr string, saveable } // Send implements transport.ConnectedEndpoint.Send. -func (c *ConnectedEndpoint) Send(data [][]byte, controlMessages transport.ControlMessages, from tcpip.FullAddress) (int64, bool, *syserr.Error) { +func (c *ConnectedEndpoint) Send(ctx context.Context, data [][]byte, controlMessages transport.ControlMessages, from tcpip.FullAddress) (int64, bool, *syserr.Error) { c.mu.RLock() defer c.mu.RUnlock() @@ -216,7 +216,7 @@ func (c *ConnectedEndpoint) EventUpdate() { } // Recv implements transport.Receiver.Recv. -func (c *ConnectedEndpoint) Recv(data [][]byte, creds bool, numRights int, peek bool) (int64, int64, transport.ControlMessages, bool, tcpip.FullAddress, bool, *syserr.Error) { +func (c *ConnectedEndpoint) Recv(ctx context.Context, data [][]byte, creds bool, numRights int, peek bool) (int64, int64, transport.ControlMessages, bool, tcpip.FullAddress, bool, *syserr.Error) { c.mu.RLock() defer c.mu.RUnlock() @@ -317,8 +317,8 @@ func (c *ConnectedEndpoint) destroyLocked() { // Release implements transport.ConnectedEndpoint.Release and // transport.Receiver.Release. -func (c *ConnectedEndpoint) Release() { - c.ref.DecRefWithDestructor(func() { +func (c *ConnectedEndpoint) Release(ctx context.Context) { + c.ref.DecRefWithDestructor(ctx, func(context.Context) { c.mu.Lock() c.destroyLocked() c.mu.Unlock() @@ -347,8 +347,8 @@ func (e *SCMConnectedEndpoint) Init() error { // Release implements transport.ConnectedEndpoint.Release and // transport.Receiver.Release. -func (e *SCMConnectedEndpoint) Release() { - e.ref.DecRefWithDestructor(func() { +func (e *SCMConnectedEndpoint) Release(ctx context.Context) { + e.ref.DecRefWithDestructor(ctx, func(context.Context) { e.mu.Lock() if err := syscall.Close(e.fd); err != nil { log.Warningf("Failed to close host fd %d: %v", err) diff --git a/pkg/sentry/fsimpl/host/tty.go b/pkg/sentry/fsimpl/host/tty.go index 4ee9270cc..d372c60cb 100644 --- a/pkg/sentry/fsimpl/host/tty.go +++ b/pkg/sentry/fsimpl/host/tty.go @@ -67,12 +67,12 @@ func (t *TTYFileDescription) ForegroundProcessGroup() *kernel.ProcessGroup { } // Release implements fs.FileOperations.Release. -func (t *TTYFileDescription) Release() { +func (t *TTYFileDescription) Release(ctx context.Context) { t.mu.Lock() t.fgProcessGroup = nil t.mu.Unlock() - t.fileDescription.Release() + t.fileDescription.Release(ctx) } // PRead implements vfs.FileDescriptionImpl. diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go index c6c4472e7..12adf727a 100644 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -122,7 +122,7 @@ func (fd *DynamicBytesFD) PWrite(ctx context.Context, src usermem.IOSequence, of } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *DynamicBytesFD) Release() {} +func (fd *DynamicBytesFD) Release(context.Context) {} // Stat implements vfs.FileDescriptionImpl.Stat. func (fd *DynamicBytesFD) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index 1d37ccb98..fcee6200a 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -113,7 +113,7 @@ func (fd *GenericDirectoryFD) PWrite(ctx context.Context, src usermem.IOSequence } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *GenericDirectoryFD) Release() {} +func (fd *GenericDirectoryFD) Release(context.Context) {} func (fd *GenericDirectoryFD) filesystem() *vfs.Filesystem { return fd.vfsfd.VirtualDentry().Mount().Filesystem() diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 61a36cff9..d7edb6342 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -56,13 +56,13 @@ afterSymlink: return vfsd, nil } if name == ".." { - if isRoot, err := rp.CheckRoot(vfsd); err != nil { + if isRoot, err := rp.CheckRoot(ctx, vfsd); err != nil { return nil, err } else if isRoot || d.parent == nil { rp.Advance() return vfsd, nil } - if err := rp.CheckMount(&d.parent.vfsd); err != nil { + if err := rp.CheckMount(ctx, &d.parent.vfsd); err != nil { return nil, err } rp.Advance() @@ -77,7 +77,7 @@ afterSymlink: if err != nil { return nil, err } - if err := rp.CheckMount(&next.vfsd); err != nil { + if err := rp.CheckMount(ctx, &next.vfsd); err != nil { return nil, err } // Resolve any symlink at current path component. @@ -88,7 +88,7 @@ afterSymlink: } if targetVD.Ok() { err := rp.HandleJump(targetVD) - targetVD.DecRef() + targetVD.DecRef(ctx) if err != nil { return nil, err } @@ -116,7 +116,7 @@ func (fs *Filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.Vir // Cached dentry exists, revalidate. if !child.inode.Valid(ctx) { delete(parent.children, name) - vfsObj.InvalidateDentry(&child.vfsd) + vfsObj.InvalidateDentry(ctx, &child.vfsd) fs.deferDecRef(&child.vfsd) // Reference from Lookup. child = nil } @@ -234,7 +234,7 @@ func checkDeleteLocked(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Den } // Release implements vfs.FilesystemImpl.Release. -func (fs *Filesystem) Release() { +func (fs *Filesystem) Release(context.Context) { } // Sync implements vfs.FilesystemImpl.Sync. @@ -246,7 +246,7 @@ func (fs *Filesystem) Sync(ctx context.Context) error { // AccessAt implements vfs.Filesystem.Impl.AccessAt. func (fs *Filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error { fs.mu.RLock() - defer fs.processDeferredDecRefs() + defer fs.processDeferredDecRefs(ctx) defer fs.mu.RUnlock() _, inode, err := fs.walkExistingLocked(ctx, rp) @@ -259,7 +259,7 @@ func (fs *Filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds // GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. func (fs *Filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { fs.mu.RLock() - defer fs.processDeferredDecRefs() + defer fs.processDeferredDecRefs(ctx) defer fs.mu.RUnlock() vfsd, inode, err := fs.walkExistingLocked(ctx, rp) if err != nil { @@ -282,7 +282,7 @@ func (fs *Filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, op // GetParentDentryAt implements vfs.FilesystemImpl.GetParentDentryAt. func (fs *Filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { fs.mu.RLock() - defer fs.processDeferredDecRefs() + defer fs.processDeferredDecRefs(ctx) defer fs.mu.RUnlock() vfsd, _, err := fs.walkParentDirLocked(ctx, rp) if err != nil { @@ -300,7 +300,7 @@ func (fs *Filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. fs.mu.Lock() defer fs.mu.Unlock() parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) - fs.processDeferredDecRefsLocked() + fs.processDeferredDecRefsLocked(ctx) if err != nil { return err } @@ -337,7 +337,7 @@ func (fs *Filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts v fs.mu.Lock() defer fs.mu.Unlock() parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) - fs.processDeferredDecRefsLocked() + fs.processDeferredDecRefsLocked(ctx) if err != nil { return err } @@ -365,7 +365,7 @@ func (fs *Filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v fs.mu.Lock() defer fs.mu.Unlock() parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) - fs.processDeferredDecRefsLocked() + fs.processDeferredDecRefsLocked(ctx) if err != nil { return err } @@ -397,7 +397,7 @@ func (fs *Filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf // Do not create new file. if opts.Flags&linux.O_CREAT == 0 { fs.mu.RLock() - defer fs.processDeferredDecRefs() + defer fs.processDeferredDecRefs(ctx) defer fs.mu.RUnlock() vfsd, inode, err := fs.walkExistingLocked(ctx, rp) if err != nil { @@ -429,7 +429,7 @@ func (fs *Filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf } afterTrailingSymlink: parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) - fs.processDeferredDecRefsLocked() + fs.processDeferredDecRefsLocked(ctx) if err != nil { return nil, err } @@ -483,7 +483,7 @@ afterTrailingSymlink: } if targetVD.Ok() { err := rp.HandleJump(targetVD) - targetVD.DecRef() + targetVD.DecRef(ctx) if err != nil { return nil, err } @@ -507,7 +507,7 @@ func (fs *Filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (st fs.mu.RLock() d, inode, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() - fs.processDeferredDecRefs() + fs.processDeferredDecRefs(ctx) if err != nil { return "", err } @@ -526,7 +526,7 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa noReplace := opts.Flags&linux.RENAME_NOREPLACE != 0 fs.mu.Lock() - defer fs.processDeferredDecRefsLocked() + defer fs.processDeferredDecRefsLocked(ctx) defer fs.mu.Unlock() // Resolve the destination directory first to verify that it's on this @@ -584,7 +584,7 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa } mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() + defer mntns.DecRef(ctx) virtfs := rp.VirtualFilesystem() // We can't deadlock here due to lock ordering because we're protected from @@ -615,7 +615,7 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa dstDir.children = make(map[string]*Dentry) } dstDir.children[pc] = src - virtfs.CommitRenameReplaceDentry(srcVFSD, replaced) + virtfs.CommitRenameReplaceDentry(ctx, srcVFSD, replaced) return nil } @@ -624,7 +624,7 @@ func (fs *Filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error fs.mu.Lock() defer fs.mu.Unlock() vfsd, inode, err := fs.walkExistingLocked(ctx, rp) - fs.processDeferredDecRefsLocked() + fs.processDeferredDecRefsLocked(ctx) if err != nil { return err } @@ -648,7 +648,7 @@ func (fs *Filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error defer parentDentry.dirMu.Unlock() mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() + defer mntns.DecRef(ctx) if err := virtfs.PrepareDeleteDentry(mntns, vfsd); err != nil { return err } @@ -656,7 +656,7 @@ func (fs *Filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error virtfs.AbortDeleteDentry(vfsd) return err } - virtfs.CommitDeleteDentry(vfsd) + virtfs.CommitDeleteDentry(ctx, vfsd) return nil } @@ -665,7 +665,7 @@ func (fs *Filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts fs.mu.RLock() _, inode, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() - fs.processDeferredDecRefs() + fs.processDeferredDecRefs(ctx) if err != nil { return err } @@ -680,7 +680,7 @@ func (fs *Filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf fs.mu.RLock() _, inode, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() - fs.processDeferredDecRefs() + fs.processDeferredDecRefs(ctx) if err != nil { return linux.Statx{}, err } @@ -692,7 +692,7 @@ func (fs *Filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linu fs.mu.RLock() _, _, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() - fs.processDeferredDecRefs() + fs.processDeferredDecRefs(ctx) if err != nil { return linux.Statfs{}, err } @@ -708,7 +708,7 @@ func (fs *Filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, targ fs.mu.Lock() defer fs.mu.Unlock() parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) - fs.processDeferredDecRefsLocked() + fs.processDeferredDecRefsLocked(ctx) if err != nil { return err } @@ -733,7 +733,7 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error fs.mu.Lock() defer fs.mu.Unlock() vfsd, _, err := fs.walkExistingLocked(ctx, rp) - fs.processDeferredDecRefsLocked() + fs.processDeferredDecRefsLocked(ctx) if err != nil { return err } @@ -753,7 +753,7 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error parentDentry.dirMu.Lock() defer parentDentry.dirMu.Unlock() mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() + defer mntns.DecRef(ctx) if err := virtfs.PrepareDeleteDentry(mntns, vfsd); err != nil { return err } @@ -761,7 +761,7 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error virtfs.AbortDeleteDentry(vfsd) return err } - virtfs.CommitDeleteDentry(vfsd) + virtfs.CommitDeleteDentry(ctx, vfsd) return nil } @@ -770,7 +770,7 @@ func (fs *Filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath fs.mu.RLock() _, inode, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() - fs.processDeferredDecRefs() + fs.processDeferredDecRefs(ctx) if err != nil { return nil, err } @@ -785,7 +785,7 @@ func (fs *Filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, si fs.mu.RLock() _, _, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() - fs.processDeferredDecRefs() + fs.processDeferredDecRefs(ctx) if err != nil { return nil, err } @@ -798,7 +798,7 @@ func (fs *Filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt fs.mu.RLock() _, _, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() - fs.processDeferredDecRefs() + fs.processDeferredDecRefs(ctx) if err != nil { return "", err } @@ -811,7 +811,7 @@ func (fs *Filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt fs.mu.RLock() _, _, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() - fs.processDeferredDecRefs() + fs.processDeferredDecRefs(ctx) if err != nil { return err } @@ -824,7 +824,7 @@ func (fs *Filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, fs.mu.RLock() _, _, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() - fs.processDeferredDecRefs() + fs.processDeferredDecRefs(ctx) if err != nil { return err } diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 579e627f0..c3efcf3ec 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -40,7 +40,7 @@ func (InodeNoopRefCount) IncRef() { } // DecRef implements Inode.DecRef. -func (InodeNoopRefCount) DecRef() { +func (InodeNoopRefCount) DecRef(context.Context) { } // TryIncRef implements Inode.TryIncRef. @@ -49,7 +49,7 @@ func (InodeNoopRefCount) TryIncRef() bool { } // Destroy implements Inode.Destroy. -func (InodeNoopRefCount) Destroy() { +func (InodeNoopRefCount) Destroy(context.Context) { } // InodeDirectoryNoNewChildren partially implements the Inode interface. @@ -366,12 +366,12 @@ func (o *OrderedChildren) Init(opts OrderedChildrenOptions) { } // DecRef implements Inode.DecRef. -func (o *OrderedChildren) DecRef() { - o.AtomicRefCount.DecRefWithDestructor(o.Destroy) +func (o *OrderedChildren) DecRef(ctx context.Context) { + o.AtomicRefCount.DecRefWithDestructor(ctx, o.Destroy) } // Destroy cleans up resources referenced by this OrderedChildren. -func (o *OrderedChildren) Destroy() { +func (o *OrderedChildren) Destroy(context.Context) { o.mu.Lock() defer o.mu.Unlock() o.order.Reset() diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index 46f207664..080118841 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -116,17 +116,17 @@ func (fs *Filesystem) deferDecRef(d *vfs.Dentry) { // processDeferredDecRefs calls vfs.Dentry.DecRef on all dentries in the // droppedDentries list. See comment on Filesystem.mu. -func (fs *Filesystem) processDeferredDecRefs() { +func (fs *Filesystem) processDeferredDecRefs(ctx context.Context) { fs.mu.Lock() - fs.processDeferredDecRefsLocked() + fs.processDeferredDecRefsLocked(ctx) fs.mu.Unlock() } // Precondition: fs.mu must be held for writing. -func (fs *Filesystem) processDeferredDecRefsLocked() { +func (fs *Filesystem) processDeferredDecRefsLocked(ctx context.Context) { fs.droppedDentriesMu.Lock() for _, d := range fs.droppedDentries { - d.DecRef() + d.DecRef(ctx) } fs.droppedDentries = fs.droppedDentries[:0] // Keep slice memory for reuse. fs.droppedDentriesMu.Unlock() @@ -212,16 +212,16 @@ func (d *Dentry) isSymlink() bool { } // DecRef implements vfs.DentryImpl.DecRef. -func (d *Dentry) DecRef() { - d.AtomicRefCount.DecRefWithDestructor(d.destroy) +func (d *Dentry) DecRef(ctx context.Context) { + d.AtomicRefCount.DecRefWithDestructor(ctx, d.destroy) } // Precondition: Dentry must be removed from VFS' dentry cache. -func (d *Dentry) destroy() { - d.inode.DecRef() // IncRef from Init. +func (d *Dentry) destroy(ctx context.Context) { + d.inode.DecRef(ctx) // IncRef from Init. d.inode = nil if d.parent != nil { - d.parent.DecRef() // IncRef from Dentry.InsertChild. + d.parent.DecRef(ctx) // IncRef from Dentry.InsertChild. } } @@ -230,7 +230,7 @@ func (d *Dentry) destroy() { // Although Linux technically supports inotify on pseudo filesystems (inotify // is implemented at the vfs layer), it is not particularly useful. It is left // unimplemented until someone actually needs it. -func (d *Dentry) InotifyWithParent(events, cookie uint32, et vfs.EventType) {} +func (d *Dentry) InotifyWithParent(ctx context.Context, events, cookie uint32, et vfs.EventType) {} // Watches implements vfs.DentryImpl.Watches. func (d *Dentry) Watches() *vfs.Watches { @@ -238,7 +238,7 @@ func (d *Dentry) Watches() *vfs.Watches { } // OnZeroWatches implements vfs.Dentry.OnZeroWatches. -func (d *Dentry) OnZeroWatches() {} +func (d *Dentry) OnZeroWatches(context.Context) {} // InsertChild inserts child into the vfs dentry cache with the given name under // this dentry. This does not update the directory inode, so calling this on @@ -326,12 +326,12 @@ type Inode interface { type inodeRefs interface { IncRef() - DecRef() + DecRef(ctx context.Context) TryIncRef() bool // Destroy is called when the inode reaches zero references. Destroy release // all resources (references) on objects referenced by the inode, including // any child dentries. - Destroy() + Destroy(ctx context.Context) } type inodeMetadata interface { diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index dc407eb1d..c5d5afedf 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -46,7 +46,7 @@ func newTestSystem(t *testing.T, rootFn RootDentryFn) *testutil.System { ctx := contexttest.Context(t) creds := auth.CredentialsFromContext(ctx) v := &vfs.VirtualFilesystem{} - if err := v.Init(); err != nil { + if err := v.Init(ctx); err != nil { t.Fatalf("VFS init: %v", err) } v.MustRegisterFilesystemType("testfs", &fsType{rootFn: rootFn}, &vfs.RegisterFilesystemTypeOptions{ @@ -163,7 +163,7 @@ func (d *dir) NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (* dir := d.fs.newDir(creds, opts.Mode, nil) dirVFSD := dir.VFSDentry() if err := d.OrderedChildren.Insert(name, dirVFSD); err != nil { - dir.DecRef() + dir.DecRef(ctx) return nil, err } d.IncLinks(1) @@ -175,7 +175,7 @@ func (d *dir) NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (* f := d.fs.newFile(creds, "") fVFSD := f.VFSDentry() if err := d.OrderedChildren.Insert(name, fVFSD); err != nil { - f.DecRef() + f.DecRef(ctx) return nil, err } return fVFSD, nil @@ -213,7 +213,7 @@ func TestBasic(t *testing.T) { }) }) defer sys.Destroy() - sys.GetDentryOrDie(sys.PathOpAtRoot("file1")).DecRef() + sys.GetDentryOrDie(sys.PathOpAtRoot("file1")).DecRef(sys.Ctx) } func TestMkdirGetDentry(t *testing.T) { @@ -228,7 +228,7 @@ func TestMkdirGetDentry(t *testing.T) { if err := sys.VFS.MkdirAt(sys.Ctx, sys.Creds, pop, &vfs.MkdirOptions{Mode: 0755}); err != nil { t.Fatalf("MkdirAt for PathOperation %+v failed: %v", pop, err) } - sys.GetDentryOrDie(pop).DecRef() + sys.GetDentryOrDie(pop).DecRef(sys.Ctx) } func TestReadStaticFile(t *testing.T) { @@ -246,7 +246,7 @@ func TestReadStaticFile(t *testing.T) { if err != nil { t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) } - defer fd.DecRef() + defer fd.DecRef(sys.Ctx) content, err := sys.ReadToEnd(fd) if err != nil { @@ -273,7 +273,7 @@ func TestCreateNewFileInStaticDir(t *testing.T) { } // Close the file. The file should persist. - fd.DecRef() + fd.DecRef(sys.Ctx) fd, err = sys.VFS.OpenAt(sys.Ctx, sys.Creds, pop, &vfs.OpenOptions{ Flags: linux.O_RDONLY, @@ -281,7 +281,7 @@ func TestCreateNewFileInStaticDir(t *testing.T) { if err != nil { t.Fatalf("OpenAt(pop:%+v) = %+v failed: %v", pop, fd, err) } - fd.DecRef() + fd.DecRef(sys.Ctx) } func TestDirFDReadWrite(t *testing.T) { @@ -297,7 +297,7 @@ func TestDirFDReadWrite(t *testing.T) { if err != nil { t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) } - defer fd.DecRef() + defer fd.DecRef(sys.Ctx) // Read/Write should fail for directory FDs. if _, err := fd.Read(sys.Ctx, usermem.BytesIOSequence([]byte{}), vfs.ReadOptions{}); err != syserror.EISDIR { diff --git a/pkg/sentry/fsimpl/overlay/copy_up.go b/pkg/sentry/fsimpl/overlay/copy_up.go index 8f8dcfafe..b3d19ff82 100644 --- a/pkg/sentry/fsimpl/overlay/copy_up.go +++ b/pkg/sentry/fsimpl/overlay/copy_up.go @@ -98,7 +98,7 @@ func (d *dentry) copyUpLocked(ctx context.Context) error { if err != nil { return err } - defer oldFD.DecRef() + defer oldFD.DecRef(ctx) newFD, err := vfsObj.OpenAt(ctx, d.fs.creds, &newpop, &vfs.OpenOptions{ Flags: linux.O_WRONLY | linux.O_CREAT | linux.O_EXCL, Mode: linux.FileMode(d.mode &^ linux.S_IFMT), @@ -106,7 +106,7 @@ func (d *dentry) copyUpLocked(ctx context.Context) error { if err != nil { return err } - defer newFD.DecRef() + defer newFD.DecRef(ctx) bufIOSeq := usermem.BytesIOSequence(make([]byte, 32*1024)) // arbitrary buffer size for { readN, readErr := oldFD.Read(ctx, bufIOSeq, vfs.ReadOptions{}) @@ -241,13 +241,13 @@ func (d *dentry) copyUpLocked(ctx context.Context) error { Mask: linux.STATX_INO, }) if err != nil { - d.upperVD.DecRef() + d.upperVD.DecRef(ctx) d.upperVD = vfs.VirtualDentry{} cleanupUndoCopyUp() return err } if upperStat.Mask&linux.STATX_INO == 0 { - d.upperVD.DecRef() + d.upperVD.DecRef(ctx) d.upperVD = vfs.VirtualDentry{} cleanupUndoCopyUp() return syserror.EREMOTE diff --git a/pkg/sentry/fsimpl/overlay/directory.go b/pkg/sentry/fsimpl/overlay/directory.go index f5c2462a5..fccb94105 100644 --- a/pkg/sentry/fsimpl/overlay/directory.go +++ b/pkg/sentry/fsimpl/overlay/directory.go @@ -46,7 +46,7 @@ func (d *dentry) collectWhiteoutsForRmdirLocked(ctx context.Context) (map[string readdirErr = err return false } - defer layerFD.DecRef() + defer layerFD.DecRef(ctx) // Reuse slice allocated for maybeWhiteouts from a previous layer to // reduce allocations. @@ -108,7 +108,7 @@ type directoryFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *directoryFD) Release() { +func (fd *directoryFD) Release(ctx context.Context) { } // IterDirents implements vfs.FileDescriptionImpl.IterDirents. @@ -177,7 +177,7 @@ func (d *dentry) getDirents(ctx context.Context) ([]vfs.Dirent, error) { readdirErr = err return false } - defer layerFD.DecRef() + defer layerFD.DecRef(ctx) // Reuse slice allocated for maybeWhiteouts from a previous layer to // reduce allocations. @@ -282,6 +282,6 @@ func (fd *directoryFD) Sync(ctx context.Context) error { return err } err = upperFD.Sync(ctx) - upperFD.DecRef() + upperFD.DecRef(ctx) return err } diff --git a/pkg/sentry/fsimpl/overlay/filesystem.go b/pkg/sentry/fsimpl/overlay/filesystem.go index 6b705e955..986b36ead 100644 --- a/pkg/sentry/fsimpl/overlay/filesystem.go +++ b/pkg/sentry/fsimpl/overlay/filesystem.go @@ -77,7 +77,7 @@ func putDentrySlice(ds *[]*dentry) { // but dentry slices are allocated lazily, and it's much easier to say "defer // fs.renameMuRUnlockAndCheckDrop(&ds)" than "defer func() { // fs.renameMuRUnlockAndCheckDrop(ds) }()" to work around this. -func (fs *filesystem) renameMuRUnlockAndCheckDrop(ds **[]*dentry) { +func (fs *filesystem) renameMuRUnlockAndCheckDrop(ctx context.Context, ds **[]*dentry) { fs.renameMu.RUnlock() if *ds == nil { return @@ -85,20 +85,20 @@ func (fs *filesystem) renameMuRUnlockAndCheckDrop(ds **[]*dentry) { if len(**ds) != 0 { fs.renameMu.Lock() for _, d := range **ds { - d.checkDropLocked() + d.checkDropLocked(ctx) } fs.renameMu.Unlock() } putDentrySlice(*ds) } -func (fs *filesystem) renameMuUnlockAndCheckDrop(ds **[]*dentry) { +func (fs *filesystem) renameMuUnlockAndCheckDrop(ctx context.Context, ds **[]*dentry) { if *ds == nil { fs.renameMu.Unlock() return } for _, d := range **ds { - d.checkDropLocked() + d.checkDropLocked(ctx) } fs.renameMu.Unlock() putDentrySlice(*ds) @@ -126,13 +126,13 @@ afterSymlink: return d, nil } if name == ".." { - if isRoot, err := rp.CheckRoot(&d.vfsd); err != nil { + if isRoot, err := rp.CheckRoot(ctx, &d.vfsd); err != nil { return nil, err } else if isRoot || d.parent == nil { rp.Advance() return d, nil } - if err := rp.CheckMount(&d.parent.vfsd); err != nil { + if err := rp.CheckMount(ctx, &d.parent.vfsd); err != nil { return nil, err } rp.Advance() @@ -142,7 +142,7 @@ afterSymlink: if err != nil { return nil, err } - if err := rp.CheckMount(&child.vfsd); err != nil { + if err := rp.CheckMount(ctx, &child.vfsd); err != nil { return nil, err } if child.isSymlink() && mayFollowSymlinks && rp.ShouldFollowSymlink() { @@ -272,11 +272,11 @@ func (fs *filesystem) lookupLocked(ctx context.Context, parent *dentry, name str }) if lookupErr != nil { - child.destroyLocked() + child.destroyLocked(ctx) return nil, lookupErr } if !existsOnAnyLayer { - child.destroyLocked() + child.destroyLocked(ctx) return nil, syserror.ENOENT } @@ -430,7 +430,7 @@ func (fs *filesystem) resolveLocked(ctx context.Context, rp *vfs.ResolvingPath, func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir bool, create func(parent *dentry, name string, haveUpperWhiteout bool) error) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) start := rp.Start().Impl().(*dentry) parent, err := fs.walkParentDirLocked(ctx, rp, start, &ds) if err != nil { @@ -501,7 +501,7 @@ func (fs *filesystem) cleanupRecreateWhiteout(ctx context.Context, vfsObj *vfs.V func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return err @@ -513,7 +513,7 @@ func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.BoundEndpointOptions) (transport.BoundEndpoint, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return nil, err @@ -532,7 +532,7 @@ func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return nil, err @@ -553,7 +553,7 @@ func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, op func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) start := rp.Start().Impl().(*dentry) d, err := fs.walkParentDirLocked(ctx, rp, start, &ds) if err != nil { @@ -720,7 +720,7 @@ func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) start := rp.Start().Impl().(*dentry) if rp.Done() { @@ -825,7 +825,7 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vf fd.LockFD.Init(&d.locks) layerFDOpts := layerFD.Options() if err := fd.vfsfd.Init(fd, layerFlags, mnt, &d.vfsd, &layerFDOpts); err != nil { - layerFD.DecRef() + layerFD.DecRef(ctx) return nil, err } return &fd.vfsfd, nil @@ -920,7 +920,7 @@ func (fs *filesystem) createAndOpenLocked(ctx context.Context, rp *vfs.Resolving fd.LockFD.Init(&child.locks) upperFDOpts := upperFD.Options() if err := fd.vfsfd.Init(fd, upperFlags, mnt, &child.vfsd, &upperFDOpts); err != nil { - upperFD.DecRef() + upperFD.DecRef(ctx) // Don't bother with cleanup; the file was created successfully, we // just can't open it anymore for some reason. return nil, err @@ -932,7 +932,7 @@ func (fs *filesystem) createAndOpenLocked(ctx context.Context, rp *vfs.Resolving func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return "", err @@ -952,7 +952,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa var ds *[]*dentry fs.renameMu.Lock() - defer fs.renameMuUnlockAndCheckDrop(&ds) + defer fs.renameMuUnlockAndCheckDrop(ctx, &ds) newParent, err := fs.walkParentDirLocked(ctx, rp, rp.Start().Impl().(*dentry), &ds) if err != nil { return err @@ -979,7 +979,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) start := rp.Start().Impl().(*dentry) parent, err := fs.walkParentDirLocked(ctx, rp, start, &ds) if err != nil { @@ -1001,7 +1001,7 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error } vfsObj := rp.VirtualFilesystem() mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() + defer mntns.DecRef(ctx) parent.dirMu.Lock() defer parent.dirMu.Unlock() @@ -1086,7 +1086,7 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error return err } - vfsObj.CommitDeleteDentry(&child.vfsd) + vfsObj.CommitDeleteDentry(ctx, &child.vfsd) delete(parent.children, name) ds = appendDentry(ds, child) parent.dirents = nil @@ -1097,7 +1097,7 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return err @@ -1132,7 +1132,7 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return linux.Statx{}, err @@ -1160,7 +1160,7 @@ func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) _, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return linux.Statfs{}, err @@ -1211,7 +1211,7 @@ func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, targ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) start := rp.Start().Impl().(*dentry) parent, err := fs.walkParentDirLocked(ctx, rp, start, &ds) if err != nil { @@ -1233,7 +1233,7 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error } vfsObj := rp.VirtualFilesystem() mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() + defer mntns.DecRef(ctx) parent.dirMu.Lock() defer parent.dirMu.Unlock() @@ -1298,7 +1298,7 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error } if child != nil { - vfsObj.CommitDeleteDentry(&child.vfsd) + vfsObj.CommitDeleteDentry(ctx, &child.vfsd) delete(parent.children, name) ds = appendDentry(ds, child) } @@ -1310,7 +1310,7 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, size uint64) ([]string, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) _, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return nil, err @@ -1324,7 +1324,7 @@ func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, si func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetxattrOptions) (string, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) _, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return "", err @@ -1336,7 +1336,7 @@ func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) _, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return err @@ -1348,7 +1348,7 @@ func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) _, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return err diff --git a/pkg/sentry/fsimpl/overlay/non_directory.go b/pkg/sentry/fsimpl/overlay/non_directory.go index c0749e711..d3060a481 100644 --- a/pkg/sentry/fsimpl/overlay/non_directory.go +++ b/pkg/sentry/fsimpl/overlay/non_directory.go @@ -81,11 +81,11 @@ func (fd *nonDirectoryFD) currentFDLocked(ctx context.Context) (*vfs.FileDescrip oldOff, oldOffErr := fd.cachedFD.Seek(ctx, 0, linux.SEEK_CUR) if oldOffErr == nil { if _, err := upperFD.Seek(ctx, oldOff, linux.SEEK_SET); err != nil { - upperFD.DecRef() + upperFD.DecRef(ctx) return nil, err } } - fd.cachedFD.DecRef() + fd.cachedFD.DecRef(ctx) fd.copiedUp = true fd.cachedFD = upperFD fd.cachedFlags = statusFlags @@ -99,8 +99,8 @@ func (fd *nonDirectoryFD) currentFDLocked(ctx context.Context) (*vfs.FileDescrip } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *nonDirectoryFD) Release() { - fd.cachedFD.DecRef() +func (fd *nonDirectoryFD) Release(ctx context.Context) { + fd.cachedFD.DecRef(ctx) fd.cachedFD = nil } @@ -138,7 +138,7 @@ func (fd *nonDirectoryFD) Stat(ctx context.Context, opts vfs.StatOptions) (linux Mask: layerMask, Sync: opts.Sync, }) - wrappedFD.DecRef() + wrappedFD.DecRef(ctx) if err != nil { return linux.Statx{}, err } @@ -187,7 +187,7 @@ func (fd *nonDirectoryFD) PRead(ctx context.Context, dst usermem.IOSequence, off if err != nil { return 0, err } - defer wrappedFD.DecRef() + defer wrappedFD.DecRef(ctx) return wrappedFD.PRead(ctx, dst, offset, opts) } @@ -209,7 +209,7 @@ func (fd *nonDirectoryFD) PWrite(ctx context.Context, src usermem.IOSequence, of if err != nil { return 0, err } - defer wrappedFD.DecRef() + defer wrappedFD.DecRef(ctx) return wrappedFD.PWrite(ctx, src, offset, opts) } @@ -250,7 +250,7 @@ func (fd *nonDirectoryFD) Sync(ctx context.Context) error { return err } wrappedFD.IncRef() - defer wrappedFD.DecRef() + defer wrappedFD.DecRef(ctx) fd.mu.Unlock() return wrappedFD.Sync(ctx) } @@ -261,6 +261,6 @@ func (fd *nonDirectoryFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOp if err != nil { return err } - defer wrappedFD.DecRef() + defer wrappedFD.DecRef(ctx) return wrappedFD.ConfigureMMap(ctx, opts) } diff --git a/pkg/sentry/fsimpl/overlay/overlay.go b/pkg/sentry/fsimpl/overlay/overlay.go index e720d4825..75cc006bf 100644 --- a/pkg/sentry/fsimpl/overlay/overlay.go +++ b/pkg/sentry/fsimpl/overlay/overlay.go @@ -123,7 +123,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt // filesystem with any number of lower layers. } else { vfsroot := vfs.RootFromContext(ctx) - defer vfsroot.DecRef() + defer vfsroot.DecRef(ctx) upperPathname, ok := mopts["upperdir"] if ok { delete(mopts, "upperdir") @@ -147,13 +147,13 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt ctx.Warningf("overlay.FilesystemType.GetFilesystem: failed to resolve upperdir %q: %v", upperPathname, err) return nil, nil, err } - defer upperRoot.DecRef() + defer upperRoot.DecRef(ctx) privateUpperRoot, err := clonePrivateMount(vfsObj, upperRoot, false /* forceReadOnly */) if err != nil { ctx.Warningf("overlay.FilesystemType.GetFilesystem: failed to make private bind mount of upperdir %q: %v", upperPathname, err) return nil, nil, err } - defer privateUpperRoot.DecRef() + defer privateUpperRoot.DecRef(ctx) fsopts.UpperRoot = privateUpperRoot } lowerPathnamesStr, ok := mopts["lowerdir"] @@ -190,13 +190,13 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt ctx.Warningf("overlay.FilesystemType.GetFilesystem: failed to resolve lowerdir %q: %v", lowerPathname, err) return nil, nil, err } - defer lowerRoot.DecRef() + defer lowerRoot.DecRef(ctx) privateLowerRoot, err := clonePrivateMount(vfsObj, lowerRoot, true /* forceReadOnly */) if err != nil { ctx.Warningf("overlay.FilesystemType.GetFilesystem: failed to make private bind mount of lowerdir %q: %v", lowerPathname, err) return nil, nil, err } - defer privateLowerRoot.DecRef() + defer privateLowerRoot.DecRef(ctx) fsopts.LowerRoots = append(fsopts.LowerRoots, privateLowerRoot) } } @@ -264,19 +264,19 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt Mask: rootStatMask, }) if err != nil { - root.destroyLocked() - fs.vfsfs.DecRef() + root.destroyLocked(ctx) + fs.vfsfs.DecRef(ctx) return nil, nil, err } if rootStat.Mask&rootStatMask != rootStatMask { - root.destroyLocked() - fs.vfsfs.DecRef() + root.destroyLocked(ctx) + fs.vfsfs.DecRef(ctx) return nil, nil, syserror.EREMOTE } if isWhiteout(&rootStat) { ctx.Warningf("overlay.FilesystemType.GetFilesystem: filesystem root is a whiteout") - root.destroyLocked() - fs.vfsfs.DecRef() + root.destroyLocked(ctx) + fs.vfsfs.DecRef(ctx) return nil, nil, syserror.EINVAL } root.mode = uint32(rootStat.Mode) @@ -319,17 +319,17 @@ func clonePrivateMount(vfsObj *vfs.VirtualFilesystem, vd vfs.VirtualDentry, forc } // Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { +func (fs *filesystem) Release(ctx context.Context) { vfsObj := fs.vfsfs.VirtualFilesystem() vfsObj.PutAnonBlockDevMinor(fs.dirDevMinor) for _, lowerDevMinor := range fs.lowerDevMinors { vfsObj.PutAnonBlockDevMinor(lowerDevMinor) } if fs.opts.UpperRoot.Ok() { - fs.opts.UpperRoot.DecRef() + fs.opts.UpperRoot.DecRef(ctx) } for _, lowerRoot := range fs.opts.LowerRoots { - lowerRoot.DecRef() + lowerRoot.DecRef(ctx) } } @@ -452,10 +452,10 @@ func (d *dentry) TryIncRef() bool { } // DecRef implements vfs.DentryImpl.DecRef. -func (d *dentry) DecRef() { +func (d *dentry) DecRef(ctx context.Context) { if refs := atomic.AddInt64(&d.refs, -1); refs == 0 { d.fs.renameMu.Lock() - d.checkDropLocked() + d.checkDropLocked(ctx) d.fs.renameMu.Unlock() } else if refs < 0 { panic("overlay.dentry.DecRef() called without holding a reference") @@ -466,7 +466,7 @@ func (d *dentry) DecRef() { // becomes deleted. // // Preconditions: d.fs.renameMu must be locked for writing. -func (d *dentry) checkDropLocked() { +func (d *dentry) checkDropLocked(ctx context.Context) { // Dentries with a positive reference count must be retained. (The only way // to obtain a reference on a dentry with zero references is via path // resolution, which requires renameMu, so if d.refs is zero then it will @@ -476,14 +476,14 @@ func (d *dentry) checkDropLocked() { return } // Refs is still zero; destroy it. - d.destroyLocked() + d.destroyLocked(ctx) return } // destroyLocked destroys the dentry. // // Preconditions: d.fs.renameMu must be locked for writing. d.refs == 0. -func (d *dentry) destroyLocked() { +func (d *dentry) destroyLocked(ctx context.Context) { switch atomic.LoadInt64(&d.refs) { case 0: // Mark the dentry destroyed. @@ -495,10 +495,10 @@ func (d *dentry) destroyLocked() { } if d.upperVD.Ok() { - d.upperVD.DecRef() + d.upperVD.DecRef(ctx) } for _, lowerVD := range d.lowerVDs { - lowerVD.DecRef() + lowerVD.DecRef(ctx) } if d.parent != nil { @@ -510,7 +510,7 @@ func (d *dentry) destroyLocked() { // Drop the reference held by d on its parent without recursively // locking d.fs.renameMu. if refs := atomic.AddInt64(&d.parent.refs, -1); refs == 0 { - d.parent.checkDropLocked() + d.parent.checkDropLocked(ctx) } else if refs < 0 { panic("overlay.dentry.DecRef() called without holding a reference") } @@ -518,7 +518,7 @@ func (d *dentry) destroyLocked() { } // InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. -func (d *dentry) InotifyWithParent(events uint32, cookie uint32, et vfs.EventType) { +func (d *dentry) InotifyWithParent(ctx context.Context, events uint32, cookie uint32, et vfs.EventType) { // TODO(gvisor.dev/issue/1479): Implement inotify. } @@ -531,7 +531,7 @@ func (d *dentry) Watches() *vfs.Watches { // OnZeroWatches implements vfs.DentryImpl.OnZeroWatches. // // TODO(gvisor.dev/issue/1479): Implement inotify. -func (d *dentry) OnZeroWatches() {} +func (d *dentry) OnZeroWatches(context.Context) {} // iterLayers invokes yield on each layer comprising d, from top to bottom. If // any call to yield returns false, iterLayer stops iteration. diff --git a/pkg/sentry/fsimpl/pipefs/pipefs.go b/pkg/sentry/fsimpl/pipefs/pipefs.go index 811f80a5f..2ca793db9 100644 --- a/pkg/sentry/fsimpl/pipefs/pipefs.go +++ b/pkg/sentry/fsimpl/pipefs/pipefs.go @@ -63,9 +63,9 @@ func NewFilesystem(vfsObj *vfs.VirtualFilesystem) (*vfs.Filesystem, error) { } // Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { +func (fs *filesystem) Release(ctx context.Context) { fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) - fs.Filesystem.Release() + fs.Filesystem.Release(ctx) } // PrependPath implements vfs.FilesystemImpl.PrependPath. @@ -160,6 +160,6 @@ func NewConnectedPipeFDs(ctx context.Context, mnt *vfs.Mount, flags uint32) (*vf inode := newInode(ctx, fs) var d kernfs.Dentry d.Init(inode) - defer d.DecRef() + defer d.DecRef(ctx) return inode.pipe.ReaderWriterPair(mnt, d.VFSDentry(), flags) } diff --git a/pkg/sentry/fsimpl/proc/filesystem.go b/pkg/sentry/fsimpl/proc/filesystem.go index 609210253..2463d51cd 100644 --- a/pkg/sentry/fsimpl/proc/filesystem.go +++ b/pkg/sentry/fsimpl/proc/filesystem.go @@ -77,9 +77,9 @@ func (ft FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualF } // Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { +func (fs *filesystem) Release(ctx context.Context) { fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) - fs.Filesystem.Release() + fs.Filesystem.Release(ctx) } // dynamicInode is an overfitted interface for common Inodes with diff --git a/pkg/sentry/fsimpl/proc/task_fds.go b/pkg/sentry/fsimpl/proc/task_fds.go index fea29e5f0..f0d3f7f5e 100644 --- a/pkg/sentry/fsimpl/proc/task_fds.go +++ b/pkg/sentry/fsimpl/proc/task_fds.go @@ -43,12 +43,12 @@ func getTaskFD(t *kernel.Task, fd int32) (*vfs.FileDescription, kernel.FDFlags) return file, flags } -func taskFDExists(t *kernel.Task, fd int32) bool { +func taskFDExists(ctx context.Context, t *kernel.Task, fd int32) bool { file, _ := getTaskFD(t, fd) if file == nil { return false } - file.DecRef() + file.DecRef(ctx) return true } @@ -68,7 +68,7 @@ func (i *fdDir) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, off var fds []int32 i.task.WithMuLocked(func(t *kernel.Task) { if fdTable := t.FDTable(); fdTable != nil { - fds = fdTable.GetFDs() + fds = fdTable.GetFDs(ctx) } }) @@ -135,7 +135,7 @@ func (i *fdDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, erro return nil, syserror.ENOENT } fd := int32(fdInt) - if !taskFDExists(i.task, fd) { + if !taskFDExists(ctx, i.task, fd) { return nil, syserror.ENOENT } taskDentry := i.fs.newFDSymlink(i.task, fd, i.fs.NextIno()) @@ -204,9 +204,9 @@ func (s *fdSymlink) Readlink(ctx context.Context) (string, error) { if file == nil { return "", syserror.ENOENT } - defer file.DecRef() + defer file.DecRef(ctx) root := vfs.RootFromContext(ctx) - defer root.DecRef() + defer root.DecRef(ctx) return s.task.Kernel().VFS().PathnameWithDeleted(ctx, root, file.VirtualDentry()) } @@ -215,7 +215,7 @@ func (s *fdSymlink) Getlink(ctx context.Context, mnt *vfs.Mount) (vfs.VirtualDen if file == nil { return vfs.VirtualDentry{}, "", syserror.ENOENT } - defer file.DecRef() + defer file.DecRef(ctx) vd := file.VirtualDentry() vd.IncRef() return vd, "", nil @@ -258,7 +258,7 @@ func (i *fdInfoDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, return nil, syserror.ENOENT } fd := int32(fdInt) - if !taskFDExists(i.task, fd) { + if !taskFDExists(ctx, i.task, fd) { return nil, syserror.ENOENT } data := &fdInfoData{ @@ -297,7 +297,7 @@ func (d *fdInfoData) Generate(ctx context.Context, buf *bytes.Buffer) error { if file == nil { return syserror.ENOENT } - defer file.DecRef() + defer file.DecRef(ctx) // TODO(b/121266871): Include pos, locks, and other data. For now we only // have flags. // See https://www.kernel.org/doc/Documentation/filesystems/proc.txt diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index 859b7d727..830b78949 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -677,7 +677,7 @@ func (s *exeSymlink) Readlink(ctx context.Context) (string, error) { if err != nil { return "", err } - defer exec.DecRef() + defer exec.DecRef(ctx) return exec.PathnameWithDeleted(ctx), nil } @@ -692,7 +692,7 @@ func (s *exeSymlink) Getlink(ctx context.Context, _ *vfs.Mount) (vfs.VirtualDent if err != nil { return vfs.VirtualDentry{}, "", err } - defer exec.DecRef() + defer exec.DecRef(ctx) vd := exec.(*fsbridge.VFSFile).FileDescription().VirtualDentry() vd.IncRef() @@ -748,7 +748,7 @@ func (i *mountInfoData) Generate(ctx context.Context, buf *bytes.Buffer) error { // Root has been destroyed. Don't try to read mounts. return nil } - defer rootDir.DecRef() + defer rootDir.DecRef(ctx) i.task.Kernel().VFS().GenerateProcMountInfo(ctx, rootDir, buf) return nil } @@ -779,7 +779,7 @@ func (i *mountsData) Generate(ctx context.Context, buf *bytes.Buffer) error { // Root has been destroyed. Don't try to read mounts. return nil } - defer rootDir.DecRef() + defer rootDir.DecRef(ctx) i.task.Kernel().VFS().GenerateProcMounts(ctx, rootDir, buf) return nil } @@ -825,7 +825,7 @@ func (s *namespaceSymlink) Getlink(ctx context.Context, mnt *vfs.Mount) (vfs.Vir dentry.Init(&namespaceInode{}) vd := vfs.MakeVirtualDentry(mnt, dentry.VFSDentry()) vd.IncRef() - dentry.DecRef() + dentry.DecRef(ctx) return vd, "", nil } @@ -887,8 +887,8 @@ func (fd *namespaceFD) SetStat(ctx context.Context, opts vfs.SetStatOptions) err } // Release implements FileDescriptionImpl. -func (fd *namespaceFD) Release() { - fd.inode.DecRef() +func (fd *namespaceFD) Release(ctx context.Context) { + fd.inode.DecRef(ctx) } // LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. diff --git a/pkg/sentry/fsimpl/proc/task_net.go b/pkg/sentry/fsimpl/proc/task_net.go index 6bde27376..a4c884bf9 100644 --- a/pkg/sentry/fsimpl/proc/task_net.go +++ b/pkg/sentry/fsimpl/proc/task_net.go @@ -212,7 +212,7 @@ func (n *netUnixData) Generate(ctx context.Context, buf *bytes.Buffer) error { continue } if family, _, _ := s.Impl().(socket.SocketVFS2).Type(); family != linux.AF_UNIX { - s.DecRef() + s.DecRef(ctx) // Not a unix socket. continue } @@ -281,7 +281,7 @@ func (n *netUnixData) Generate(ctx context.Context, buf *bytes.Buffer) error { } fmt.Fprintf(buf, "\n") - s.DecRef() + s.DecRef(ctx) } return nil } @@ -359,7 +359,7 @@ func commonGenerateTCP(ctx context.Context, buf *bytes.Buffer, k *kernel.Kernel, panic(fmt.Sprintf("Found non-socket file in socket table: %+v", s)) } if fa, stype, _ := sops.Type(); !(family == fa && stype == linux.SOCK_STREAM) { - s.DecRef() + s.DecRef(ctx) // Not tcp4 sockets. continue } @@ -455,7 +455,7 @@ func commonGenerateTCP(ctx context.Context, buf *bytes.Buffer, k *kernel.Kernel, fmt.Fprintf(buf, "\n") - s.DecRef() + s.DecRef(ctx) } return nil @@ -524,7 +524,7 @@ func (d *netUDPData) Generate(ctx context.Context, buf *bytes.Buffer) error { panic(fmt.Sprintf("Found non-socket file in socket table: %+v", s)) } if family, stype, _ := sops.Type(); family != linux.AF_INET || stype != linux.SOCK_DGRAM { - s.DecRef() + s.DecRef(ctx) // Not udp4 socket. continue } @@ -600,7 +600,7 @@ func (d *netUDPData) Generate(ctx context.Context, buf *bytes.Buffer) error { fmt.Fprintf(buf, "\n") - s.DecRef() + s.DecRef(ctx) } return nil } diff --git a/pkg/sentry/fsimpl/proc/tasks_test.go b/pkg/sentry/fsimpl/proc/tasks_test.go index 19abb5034..3c9297dee 100644 --- a/pkg/sentry/fsimpl/proc/tasks_test.go +++ b/pkg/sentry/fsimpl/proc/tasks_test.go @@ -218,7 +218,7 @@ func TestTasks(t *testing.T) { if err != nil { t.Fatalf("vfsfs.OpenAt(%q) failed: %v", path, err) } - defer fd.DecRef() + defer fd.DecRef(s.Ctx) buf := make([]byte, 1) bufIOSeq := usermem.BytesIOSequence(buf) if _, err := fd.Read(s.Ctx, bufIOSeq, vfs.ReadOptions{}); err != syserror.EISDIR { @@ -336,7 +336,7 @@ func TestTasksOffset(t *testing.T) { if err != nil { t.Fatalf("vfsfs.OpenAt(/) failed: %v", err) } - defer fd.DecRef() + defer fd.DecRef(s.Ctx) if _, err := fd.Seek(s.Ctx, tc.offset, linux.SEEK_SET); err != nil { t.Fatalf("Seek(%d, SEEK_SET): %v", tc.offset, err) } @@ -441,7 +441,7 @@ func iterateDir(ctx context.Context, t *testing.T, s *testutil.System, fd *vfs.F t.Errorf("vfsfs.OpenAt(%v) failed: %v", absPath, err) continue } - defer child.DecRef() + defer child.DecRef(ctx) stat, err := child.Stat(ctx, vfs.StatOptions{}) if err != nil { t.Errorf("Stat(%v) failed: %v", absPath, err) @@ -476,7 +476,7 @@ func TestTree(t *testing.T) { if err != nil { t.Fatalf("failed to create test file: %v", err) } - defer file.DecRef() + defer file.DecRef(s.Ctx) var tasks []*kernel.Task for i := 0; i < 5; i++ { @@ -501,5 +501,5 @@ func TestTree(t *testing.T) { t.Fatalf("vfsfs.OpenAt(/proc) failed: %v", err) } iterateDir(ctx, t, s, fd) - fd.DecRef() + fd.DecRef(ctx) } diff --git a/pkg/sentry/fsimpl/signalfd/signalfd.go b/pkg/sentry/fsimpl/signalfd/signalfd.go index 242ba9b5d..6297e1df4 100644 --- a/pkg/sentry/fsimpl/signalfd/signalfd.go +++ b/pkg/sentry/fsimpl/signalfd/signalfd.go @@ -54,7 +54,7 @@ var _ vfs.FileDescriptionImpl = (*SignalFileDescription)(nil) // New creates a new signal fd. func New(vfsObj *vfs.VirtualFilesystem, target *kernel.Task, mask linux.SignalSet, flags uint32) (*vfs.FileDescription, error) { vd := vfsObj.NewAnonVirtualDentry("[signalfd]") - defer vd.DecRef() + defer vd.DecRef(target) sfd := &SignalFileDescription{ target: target, mask: mask, @@ -133,4 +133,4 @@ func (sfd *SignalFileDescription) EventUnregister(entry *waiter.Entry) { } // Release implements FileDescriptionImpl.Release() -func (sfd *SignalFileDescription) Release() {} +func (sfd *SignalFileDescription) Release(context.Context) {} diff --git a/pkg/sentry/fsimpl/sockfs/sockfs.go b/pkg/sentry/fsimpl/sockfs/sockfs.go index ee0828a15..c61818ff6 100644 --- a/pkg/sentry/fsimpl/sockfs/sockfs.go +++ b/pkg/sentry/fsimpl/sockfs/sockfs.go @@ -67,9 +67,9 @@ func NewFilesystem(vfsObj *vfs.VirtualFilesystem) (*vfs.Filesystem, error) { } // Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { +func (fs *filesystem) Release(ctx context.Context) { fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) - fs.Filesystem.Release() + fs.Filesystem.Release(ctx) } // PrependPath implements vfs.FilesystemImpl.PrependPath. diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go index 01ce30a4d..f81b0c38f 100644 --- a/pkg/sentry/fsimpl/sys/sys.go +++ b/pkg/sentry/fsimpl/sys/sys.go @@ -87,9 +87,9 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt } // Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { +func (fs *filesystem) Release(ctx context.Context) { fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) - fs.Filesystem.Release() + fs.Filesystem.Release(ctx) } // dir implements kernfs.Inode. diff --git a/pkg/sentry/fsimpl/sys/sys_test.go b/pkg/sentry/fsimpl/sys/sys_test.go index 242d5fd12..9fd38b295 100644 --- a/pkg/sentry/fsimpl/sys/sys_test.go +++ b/pkg/sentry/fsimpl/sys/sys_test.go @@ -59,7 +59,7 @@ func TestReadCPUFile(t *testing.T) { if err != nil { t.Fatalf("OpenAt(pop:%+v) = %+v failed: %v", pop, fd, err) } - defer fd.DecRef() + defer fd.DecRef(s.Ctx) content, err := s.ReadToEnd(fd) if err != nil { t.Fatalf("Read failed: %v", err) diff --git a/pkg/sentry/fsimpl/testutil/kernel.go b/pkg/sentry/fsimpl/testutil/kernel.go index e743e8114..1e57744e8 100644 --- a/pkg/sentry/fsimpl/testutil/kernel.go +++ b/pkg/sentry/fsimpl/testutil/kernel.go @@ -127,7 +127,7 @@ func CreateTask(ctx context.Context, name string, tc *kernel.ThreadGroup, mntns return nil, err } m := mm.NewMemoryManager(k, k, k.SleepForAddressSpaceActivation) - m.SetExecutable(fsbridge.NewVFSFile(exe)) + m.SetExecutable(ctx, fsbridge.NewVFSFile(exe)) config := &kernel.TaskConfig{ Kernel: k, diff --git a/pkg/sentry/fsimpl/testutil/testutil.go b/pkg/sentry/fsimpl/testutil/testutil.go index 0556af877..568132121 100644 --- a/pkg/sentry/fsimpl/testutil/testutil.go +++ b/pkg/sentry/fsimpl/testutil/testutil.go @@ -97,8 +97,8 @@ func (s *System) WithTemporaryContext(ctx context.Context) *System { // Destroy release resources associated with a test system. func (s *System) Destroy() { - s.Root.DecRef() - s.MntNs.DecRef() // Reference on MntNs passed to NewSystem. + s.Root.DecRef(s.Ctx) + s.MntNs.DecRef(s.Ctx) // Reference on MntNs passed to NewSystem. } // ReadToEnd reads the contents of fd until EOF to a string. @@ -149,7 +149,7 @@ func (s *System) ListDirents(pop *vfs.PathOperation) *DirentCollector { if err != nil { s.t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) } - defer fd.DecRef() + defer fd.DecRef(s.Ctx) collector := &DirentCollector{} if err := fd.IterDirents(s.Ctx, collector); err != nil { diff --git a/pkg/sentry/fsimpl/timerfd/timerfd.go b/pkg/sentry/fsimpl/timerfd/timerfd.go index 2dc90d484..86beaa0a8 100644 --- a/pkg/sentry/fsimpl/timerfd/timerfd.go +++ b/pkg/sentry/fsimpl/timerfd/timerfd.go @@ -47,9 +47,9 @@ var _ vfs.FileDescriptionImpl = (*TimerFileDescription)(nil) var _ ktime.TimerListener = (*TimerFileDescription)(nil) // New returns a new timer fd. -func New(vfsObj *vfs.VirtualFilesystem, clock ktime.Clock, flags uint32) (*vfs.FileDescription, error) { +func New(ctx context.Context, vfsObj *vfs.VirtualFilesystem, clock ktime.Clock, flags uint32) (*vfs.FileDescription, error) { vd := vfsObj.NewAnonVirtualDentry("[timerfd]") - defer vd.DecRef() + defer vd.DecRef(ctx) tfd := &TimerFileDescription{} tfd.timer = ktime.NewTimer(clock, tfd) if err := tfd.vfsfd.Init(tfd, flags, vd.Mount(), vd.Dentry(), &vfs.FileDescriptionOptions{ @@ -129,7 +129,7 @@ func (tfd *TimerFileDescription) ResumeTimer() { } // Release implements FileDescriptionImpl.Release() -func (tfd *TimerFileDescription) Release() { +func (tfd *TimerFileDescription) Release(context.Context) { tfd.timer.Destroy() } diff --git a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go index 2fb5c4d84..d263147c2 100644 --- a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go +++ b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go @@ -83,7 +83,7 @@ func fileOpOn(ctx context.Context, mntns *fs.MountNamespace, root, wd *fs.Dirent } err = fn(root, d) - d.DecRef() + d.DecRef(ctx) return err } @@ -105,17 +105,17 @@ func BenchmarkVFS1TmpfsStat(b *testing.B) { if err != nil { b.Fatalf("failed to create mount namespace: %v", err) } - defer mntns.DecRef() + defer mntns.DecRef(ctx) var filePathBuilder strings.Builder filePathBuilder.WriteByte('/') // Create nested directories with given depth. root := mntns.Root() - defer root.DecRef() + defer root.DecRef(ctx) d := root d.IncRef() - defer d.DecRef() + defer d.DecRef(ctx) for i := depth; i > 0; i-- { name := fmt.Sprintf("%d", i) if err := d.Inode.CreateDirectory(ctx, d, name, fs.FilePermsFromMode(0755)); err != nil { @@ -125,7 +125,7 @@ func BenchmarkVFS1TmpfsStat(b *testing.B) { if err != nil { b.Fatalf("failed to walk to directory %q: %v", name, err) } - d.DecRef() + d.DecRef(ctx) d = next filePathBuilder.WriteString(name) filePathBuilder.WriteByte('/') @@ -136,7 +136,7 @@ func BenchmarkVFS1TmpfsStat(b *testing.B) { if err != nil { b.Fatalf("failed to create file %q: %v", filename, err) } - file.DecRef() + file.DecRef(ctx) filePathBuilder.WriteString(filename) filePath := filePathBuilder.String() @@ -176,7 +176,7 @@ func BenchmarkVFS2TmpfsStat(b *testing.B) { // Create VFS. vfsObj := vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { + if err := vfsObj.Init(ctx); err != nil { b.Fatalf("VFS init: %v", err) } vfsObj.MustRegisterFilesystemType("tmpfs", tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ @@ -186,14 +186,14 @@ func BenchmarkVFS2TmpfsStat(b *testing.B) { if err != nil { b.Fatalf("failed to create tmpfs root mount: %v", err) } - defer mntns.DecRef() + defer mntns.DecRef(ctx) var filePathBuilder strings.Builder filePathBuilder.WriteByte('/') // Create nested directories with given depth. root := mntns.Root() - defer root.DecRef() + defer root.DecRef(ctx) vd := root vd.IncRef() for i := depth; i > 0; i-- { @@ -212,7 +212,7 @@ func BenchmarkVFS2TmpfsStat(b *testing.B) { if err != nil { b.Fatalf("failed to walk to directory %q: %v", name, err) } - vd.DecRef() + vd.DecRef(ctx) vd = nextVD filePathBuilder.WriteString(name) filePathBuilder.WriteByte('/') @@ -228,12 +228,12 @@ func BenchmarkVFS2TmpfsStat(b *testing.B) { Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL, Mode: 0644, }) - vd.DecRef() + vd.DecRef(ctx) vd = vfs.VirtualDentry{} if err != nil { b.Fatalf("failed to create file %q: %v", filename, err) } - defer fd.DecRef() + defer fd.DecRef(ctx) filePathBuilder.WriteString(filename) filePath := filePathBuilder.String() @@ -278,14 +278,14 @@ func BenchmarkVFS1TmpfsMountStat(b *testing.B) { if err != nil { b.Fatalf("failed to create mount namespace: %v", err) } - defer mntns.DecRef() + defer mntns.DecRef(ctx) var filePathBuilder strings.Builder filePathBuilder.WriteByte('/') // Create and mount the submount. root := mntns.Root() - defer root.DecRef() + defer root.DecRef(ctx) if err := root.Inode.CreateDirectory(ctx, root, mountPointName, fs.FilePermsFromMode(0755)); err != nil { b.Fatalf("failed to create mount point: %v", err) } @@ -293,7 +293,7 @@ func BenchmarkVFS1TmpfsMountStat(b *testing.B) { if err != nil { b.Fatalf("failed to walk to mount point: %v", err) } - defer mountPoint.DecRef() + defer mountPoint.DecRef(ctx) submountInode, err := tmpfsFS.Mount(ctx, "tmpfs", fs.MountSourceFlags{}, "", nil) if err != nil { b.Fatalf("failed to create tmpfs submount: %v", err) @@ -309,7 +309,7 @@ func BenchmarkVFS1TmpfsMountStat(b *testing.B) { if err != nil { b.Fatalf("failed to walk to mount root: %v", err) } - defer d.DecRef() + defer d.DecRef(ctx) for i := depth; i > 0; i-- { name := fmt.Sprintf("%d", i) if err := d.Inode.CreateDirectory(ctx, d, name, fs.FilePermsFromMode(0755)); err != nil { @@ -319,7 +319,7 @@ func BenchmarkVFS1TmpfsMountStat(b *testing.B) { if err != nil { b.Fatalf("failed to walk to directory %q: %v", name, err) } - d.DecRef() + d.DecRef(ctx) d = next filePathBuilder.WriteString(name) filePathBuilder.WriteByte('/') @@ -330,7 +330,7 @@ func BenchmarkVFS1TmpfsMountStat(b *testing.B) { if err != nil { b.Fatalf("failed to create file %q: %v", filename, err) } - file.DecRef() + file.DecRef(ctx) filePathBuilder.WriteString(filename) filePath := filePathBuilder.String() @@ -370,7 +370,7 @@ func BenchmarkVFS2TmpfsMountStat(b *testing.B) { // Create VFS. vfsObj := vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { + if err := vfsObj.Init(ctx); err != nil { b.Fatalf("VFS init: %v", err) } vfsObj.MustRegisterFilesystemType("tmpfs", tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ @@ -380,14 +380,14 @@ func BenchmarkVFS2TmpfsMountStat(b *testing.B) { if err != nil { b.Fatalf("failed to create tmpfs root mount: %v", err) } - defer mntns.DecRef() + defer mntns.DecRef(ctx) var filePathBuilder strings.Builder filePathBuilder.WriteByte('/') // Create the mount point. root := mntns.Root() - defer root.DecRef() + defer root.DecRef(ctx) pop := vfs.PathOperation{ Root: root, Start: root, @@ -403,7 +403,7 @@ func BenchmarkVFS2TmpfsMountStat(b *testing.B) { if err != nil { b.Fatalf("failed to walk to mount point: %v", err) } - defer mountPoint.DecRef() + defer mountPoint.DecRef(ctx) // Create and mount the submount. if err := vfsObj.MountAt(ctx, creds, "", &pop, "tmpfs", &vfs.MountOptions{}); err != nil { b.Fatalf("failed to mount tmpfs submount: %v", err) @@ -432,7 +432,7 @@ func BenchmarkVFS2TmpfsMountStat(b *testing.B) { if err != nil { b.Fatalf("failed to walk to directory %q: %v", name, err) } - vd.DecRef() + vd.DecRef(ctx) vd = nextVD filePathBuilder.WriteString(name) filePathBuilder.WriteByte('/') @@ -448,11 +448,11 @@ func BenchmarkVFS2TmpfsMountStat(b *testing.B) { Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL, Mode: 0644, }) - vd.DecRef() + vd.DecRef(ctx) if err != nil { b.Fatalf("failed to create file %q: %v", filename, err) } - fd.DecRef() + fd.DecRef(ctx) filePathBuilder.WriteString(filename) filePath := filePathBuilder.String() diff --git a/pkg/sentry/fsimpl/tmpfs/directory.go b/pkg/sentry/fsimpl/tmpfs/directory.go index 0a1ad4765..78b4fc5be 100644 --- a/pkg/sentry/fsimpl/tmpfs/directory.go +++ b/pkg/sentry/fsimpl/tmpfs/directory.go @@ -95,7 +95,7 @@ type directoryFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *directoryFD) Release() { +func (fd *directoryFD) Release(ctx context.Context) { if fd.iter != nil { dir := fd.inode().impl.(*directory) dir.iterMu.Lock() @@ -110,7 +110,7 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba fs := fd.filesystem() dir := fd.inode().impl.(*directory) - defer fd.dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + defer fd.dentry().InotifyWithParent(ctx, linux.IN_ACCESS, 0, vfs.PathEvent) // fs.mu is required to read d.parent and dentry.name. fs.mu.RLock() diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index ef210a69b..fb77f95cc 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -40,7 +40,7 @@ func (fs *filesystem) Sync(ctx context.Context) error { // stepLocked is loosely analogous to fs/namei.c:walk_component(). // // Preconditions: filesystem.mu must be locked. !rp.Done(). -func stepLocked(rp *vfs.ResolvingPath, d *dentry) (*dentry, error) { +func stepLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry) (*dentry, error) { dir, ok := d.inode.impl.(*directory) if !ok { return nil, syserror.ENOTDIR @@ -55,13 +55,13 @@ afterSymlink: return d, nil } if name == ".." { - if isRoot, err := rp.CheckRoot(&d.vfsd); err != nil { + if isRoot, err := rp.CheckRoot(ctx, &d.vfsd); err != nil { return nil, err } else if isRoot || d.parent == nil { rp.Advance() return d, nil } - if err := rp.CheckMount(&d.parent.vfsd); err != nil { + if err := rp.CheckMount(ctx, &d.parent.vfsd); err != nil { return nil, err } rp.Advance() @@ -74,7 +74,7 @@ afterSymlink: if !ok { return nil, syserror.ENOENT } - if err := rp.CheckMount(&child.vfsd); err != nil { + if err := rp.CheckMount(ctx, &child.vfsd); err != nil { return nil, err } if symlink, ok := child.inode.impl.(*symlink); ok && rp.ShouldFollowSymlink() { @@ -98,9 +98,9 @@ afterSymlink: // fs/namei.c:path_parentat(). // // Preconditions: filesystem.mu must be locked. !rp.Done(). -func walkParentDirLocked(rp *vfs.ResolvingPath, d *dentry) (*directory, error) { +func walkParentDirLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry) (*directory, error) { for !rp.Final() { - next, err := stepLocked(rp, d) + next, err := stepLocked(ctx, rp, d) if err != nil { return nil, err } @@ -118,10 +118,10 @@ func walkParentDirLocked(rp *vfs.ResolvingPath, d *dentry) (*directory, error) { // resolveLocked is loosely analogous to Linux's fs/namei.c:path_lookupat(). // // Preconditions: filesystem.mu must be locked. -func resolveLocked(rp *vfs.ResolvingPath) (*dentry, error) { +func resolveLocked(ctx context.Context, rp *vfs.ResolvingPath) (*dentry, error) { d := rp.Start().Impl().(*dentry) for !rp.Done() { - next, err := stepLocked(rp, d) + next, err := stepLocked(ctx, rp, d) if err != nil { return nil, err } @@ -141,10 +141,10 @@ func resolveLocked(rp *vfs.ResolvingPath) (*dentry, error) { // // Preconditions: !rp.Done(). For the final path component in rp, // !rp.ShouldFollowSymlink(). -func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(parentDir *directory, name string) error) error { +func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir bool, create func(parentDir *directory, name string) error) error { fs.mu.Lock() defer fs.mu.Unlock() - parentDir, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) + parentDir, err := walkParentDirLocked(ctx, rp, rp.Start().Impl().(*dentry)) if err != nil { return err } @@ -182,7 +182,7 @@ func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(pa if dir { ev |= linux.IN_ISDIR } - parentDir.inode.watches.Notify(name, uint32(ev), 0, vfs.InodeEvent, false /* unlinked */) + parentDir.inode.watches.Notify(ctx, name, uint32(ev), 0, vfs.InodeEvent, false /* unlinked */) parentDir.inode.touchCMtime() return nil } @@ -191,7 +191,7 @@ func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(pa func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error { fs.mu.RLock() defer fs.mu.RUnlock() - d, err := resolveLocked(rp) + d, err := resolveLocked(ctx, rp) if err != nil { return err } @@ -202,7 +202,7 @@ func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { fs.mu.RLock() defer fs.mu.RUnlock() - d, err := resolveLocked(rp) + d, err := resolveLocked(ctx, rp) if err != nil { return nil, err } @@ -222,7 +222,7 @@ func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, op func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { fs.mu.RLock() defer fs.mu.RUnlock() - dir, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) + dir, err := walkParentDirLocked(ctx, rp, rp.Start().Impl().(*dentry)) if err != nil { return nil, err } @@ -232,7 +232,7 @@ func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPa // LinkAt implements vfs.FilesystemImpl.LinkAt. func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry) error { - return fs.doCreateAt(rp, false /* dir */, func(parentDir *directory, name string) error { + return fs.doCreateAt(ctx, rp, false /* dir */, func(parentDir *directory, name string) error { if rp.Mount() != vd.Mount() { return syserror.EXDEV } @@ -251,7 +251,7 @@ func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. return syserror.EMLINK } i.incLinksLocked() - i.watches.Notify("", linux.IN_ATTRIB, 0, vfs.InodeEvent, false /* unlinked */) + i.watches.Notify(ctx, "", linux.IN_ATTRIB, 0, vfs.InodeEvent, false /* unlinked */) parentDir.insertChildLocked(fs.newDentry(i), name) return nil }) @@ -259,7 +259,7 @@ func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. // MkdirAt implements vfs.FilesystemImpl.MkdirAt. func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error { - return fs.doCreateAt(rp, true /* dir */, func(parentDir *directory, name string) error { + return fs.doCreateAt(ctx, rp, true /* dir */, func(parentDir *directory, name string) error { creds := rp.Credentials() if parentDir.inode.nlink == maxLinks { return syserror.EMLINK @@ -273,7 +273,7 @@ func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts v // MknodAt implements vfs.FilesystemImpl.MknodAt. func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MknodOptions) error { - return fs.doCreateAt(rp, false /* dir */, func(parentDir *directory, name string) error { + return fs.doCreateAt(ctx, rp, false /* dir */, func(parentDir *directory, name string) error { creds := rp.Credentials() var childInode *inode switch opts.Mode.FileType() { @@ -308,7 +308,7 @@ func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf if opts.Flags&linux.O_CREAT == 0 { fs.mu.RLock() defer fs.mu.RUnlock() - d, err := resolveLocked(rp) + d, err := resolveLocked(ctx, rp) if err != nil { return nil, err } @@ -330,7 +330,7 @@ func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf return start.open(ctx, rp, &opts, false /* afterCreate */) } afterTrailingSymlink: - parentDir, err := walkParentDirLocked(rp, start) + parentDir, err := walkParentDirLocked(ctx, rp, start) if err != nil { return nil, err } @@ -368,7 +368,7 @@ afterTrailingSymlink: if err != nil { return nil, err } - parentDir.inode.watches.Notify(name, linux.IN_CREATE, 0, vfs.PathEvent, false /* unlinked */) + parentDir.inode.watches.Notify(ctx, name, linux.IN_CREATE, 0, vfs.PathEvent, false /* unlinked */) parentDir.inode.touchCMtime() return fd, nil } @@ -376,7 +376,7 @@ afterTrailingSymlink: return nil, syserror.EEXIST } // Is the file mounted over? - if err := rp.CheckMount(&child.vfsd); err != nil { + if err := rp.CheckMount(ctx, &child.vfsd); err != nil { return nil, err } // Do we need to resolve a trailing symlink? @@ -445,7 +445,7 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.Open func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { fs.mu.RLock() defer fs.mu.RUnlock() - d, err := resolveLocked(rp) + d, err := resolveLocked(ctx, rp) if err != nil { return "", err } @@ -467,7 +467,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa // Resolve newParent first to verify that it's on this Mount. fs.mu.Lock() defer fs.mu.Unlock() - newParentDir, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) + newParentDir, err := walkParentDirLocked(ctx, rp, rp.Start().Impl().(*dentry)) if err != nil { return err } @@ -555,7 +555,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa } vfsObj := rp.VirtualFilesystem() mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() + defer mntns.DecRef(ctx) var replacedVFSD *vfs.Dentry if replaced != nil { replacedVFSD = &replaced.vfsd @@ -566,17 +566,17 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa if replaced != nil { newParentDir.removeChildLocked(replaced) if replaced.inode.isDir() { - newParentDir.inode.decLinksLocked() // from replaced's ".." + newParentDir.inode.decLinksLocked(ctx) // from replaced's ".." } - replaced.inode.decLinksLocked() + replaced.inode.decLinksLocked(ctx) } oldParentDir.removeChildLocked(renamed) newParentDir.insertChildLocked(renamed, newName) - vfsObj.CommitRenameReplaceDentry(&renamed.vfsd, replacedVFSD) + vfsObj.CommitRenameReplaceDentry(ctx, &renamed.vfsd, replacedVFSD) oldParentDir.inode.touchCMtime() if oldParentDir != newParentDir { if renamed.inode.isDir() { - oldParentDir.inode.decLinksLocked() + oldParentDir.inode.decLinksLocked(ctx) newParentDir.inode.incLinksLocked() } newParentDir.inode.touchCMtime() @@ -591,7 +591,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error { fs.mu.Lock() defer fs.mu.Unlock() - parentDir, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) + parentDir, err := walkParentDirLocked(ctx, rp, rp.Start().Impl().(*dentry)) if err != nil { return err } @@ -626,17 +626,17 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error defer mnt.EndWrite() vfsObj := rp.VirtualFilesystem() mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() + defer mntns.DecRef(ctx) if err := vfsObj.PrepareDeleteDentry(mntns, &child.vfsd); err != nil { return err } parentDir.removeChildLocked(child) - parentDir.inode.watches.Notify(name, linux.IN_DELETE|linux.IN_ISDIR, 0, vfs.InodeEvent, true /* unlinked */) + parentDir.inode.watches.Notify(ctx, name, linux.IN_DELETE|linux.IN_ISDIR, 0, vfs.InodeEvent, true /* unlinked */) // Remove links for child, child/., and child/.. - child.inode.decLinksLocked() - child.inode.decLinksLocked() - parentDir.inode.decLinksLocked() - vfsObj.CommitDeleteDentry(&child.vfsd) + child.inode.decLinksLocked(ctx) + child.inode.decLinksLocked(ctx) + parentDir.inode.decLinksLocked(ctx) + vfsObj.CommitDeleteDentry(ctx, &child.vfsd) parentDir.inode.touchCMtime() return nil } @@ -644,7 +644,7 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error // SetStatAt implements vfs.FilesystemImpl.SetStatAt. func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { fs.mu.RLock() - d, err := resolveLocked(rp) + d, err := resolveLocked(ctx, rp) if err != nil { fs.mu.RUnlock() return err @@ -656,7 +656,7 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts fs.mu.RUnlock() if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { - d.InotifyWithParent(ev, 0, vfs.InodeEvent) + d.InotifyWithParent(ctx, ev, 0, vfs.InodeEvent) } return nil } @@ -665,7 +665,7 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { fs.mu.RLock() defer fs.mu.RUnlock() - d, err := resolveLocked(rp) + d, err := resolveLocked(ctx, rp) if err != nil { return linux.Statx{}, err } @@ -678,7 +678,7 @@ func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { fs.mu.RLock() defer fs.mu.RUnlock() - if _, err := resolveLocked(rp); err != nil { + if _, err := resolveLocked(ctx, rp); err != nil { return linux.Statfs{}, err } statfs := linux.Statfs{ @@ -695,7 +695,7 @@ func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linu // SymlinkAt implements vfs.FilesystemImpl.SymlinkAt. func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, target string) error { - return fs.doCreateAt(rp, false /* dir */, func(parentDir *directory, name string) error { + return fs.doCreateAt(ctx, rp, false /* dir */, func(parentDir *directory, name string) error { creds := rp.Credentials() child := fs.newDentry(fs.newSymlink(creds.EffectiveKUID, creds.EffectiveKGID, 0777, target)) parentDir.insertChildLocked(child, name) @@ -707,7 +707,7 @@ func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, targ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error { fs.mu.Lock() defer fs.mu.Unlock() - parentDir, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) + parentDir, err := walkParentDirLocked(ctx, rp, rp.Start().Impl().(*dentry)) if err != nil { return err } @@ -738,7 +738,7 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error defer mnt.EndWrite() vfsObj := rp.VirtualFilesystem() mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() + defer mntns.DecRef(ctx) if err := vfsObj.PrepareDeleteDentry(mntns, &child.vfsd); err != nil { return err } @@ -746,11 +746,11 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error // Generate inotify events. Note that this must take place before the link // count of the child is decremented, or else the watches may be dropped // before these events are added. - vfs.InotifyRemoveChild(&child.inode.watches, &parentDir.inode.watches, name) + vfs.InotifyRemoveChild(ctx, &child.inode.watches, &parentDir.inode.watches, name) parentDir.removeChildLocked(child) - child.inode.decLinksLocked() - vfsObj.CommitDeleteDentry(&child.vfsd) + child.inode.decLinksLocked(ctx) + vfsObj.CommitDeleteDentry(ctx, &child.vfsd) parentDir.inode.touchCMtime() return nil } @@ -759,7 +759,7 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.BoundEndpointOptions) (transport.BoundEndpoint, error) { fs.mu.RLock() defer fs.mu.RUnlock() - d, err := resolveLocked(rp) + d, err := resolveLocked(ctx, rp) if err != nil { return nil, err } @@ -778,7 +778,7 @@ func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, size uint64) ([]string, error) { fs.mu.RLock() defer fs.mu.RUnlock() - d, err := resolveLocked(rp) + d, err := resolveLocked(ctx, rp) if err != nil { return nil, err } @@ -789,7 +789,7 @@ func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, si func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetxattrOptions) (string, error) { fs.mu.RLock() defer fs.mu.RUnlock() - d, err := resolveLocked(rp) + d, err := resolveLocked(ctx, rp) if err != nil { return "", err } @@ -799,7 +799,7 @@ func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt // SetxattrAt implements vfs.FilesystemImpl.SetxattrAt. func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { fs.mu.RLock() - d, err := resolveLocked(rp) + d, err := resolveLocked(ctx, rp) if err != nil { fs.mu.RUnlock() return err @@ -810,14 +810,14 @@ func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt } fs.mu.RUnlock() - d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) + d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil } // RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { fs.mu.RLock() - d, err := resolveLocked(rp) + d, err := resolveLocked(ctx, rp) if err != nil { fs.mu.RUnlock() return err @@ -828,7 +828,7 @@ func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, } fs.mu.RUnlock() - d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) + d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil } diff --git a/pkg/sentry/fsimpl/tmpfs/pipe_test.go b/pkg/sentry/fsimpl/tmpfs/pipe_test.go index 1614f2c39..ec2701d8b 100644 --- a/pkg/sentry/fsimpl/tmpfs/pipe_test.go +++ b/pkg/sentry/fsimpl/tmpfs/pipe_test.go @@ -32,7 +32,7 @@ const fileName = "mypipe" func TestSeparateFDs(t *testing.T) { ctx, creds, vfsObj, root := setup(t) - defer root.DecRef() + defer root.DecRef(ctx) // Open the read side. This is done in a concurrently because opening // One end the pipe blocks until the other end is opened. @@ -55,13 +55,13 @@ func TestSeparateFDs(t *testing.T) { if err != nil { t.Fatalf("failed to open pipe for writing %q: %v", fileName, err) } - defer wfd.DecRef() + defer wfd.DecRef(ctx) rfd, ok := <-rfdchan if !ok { t.Fatalf("failed to open pipe for reading %q", fileName) } - defer rfd.DecRef() + defer rfd.DecRef(ctx) const msg = "vamos azul" checkEmpty(ctx, t, rfd) @@ -71,7 +71,7 @@ func TestSeparateFDs(t *testing.T) { func TestNonblockingRead(t *testing.T) { ctx, creds, vfsObj, root := setup(t) - defer root.DecRef() + defer root.DecRef(ctx) // Open the read side as nonblocking. pop := vfs.PathOperation{ @@ -85,7 +85,7 @@ func TestNonblockingRead(t *testing.T) { if err != nil { t.Fatalf("failed to open pipe for reading %q: %v", fileName, err) } - defer rfd.DecRef() + defer rfd.DecRef(ctx) // Open the write side. openOpts = vfs.OpenOptions{Flags: linux.O_WRONLY} @@ -93,7 +93,7 @@ func TestNonblockingRead(t *testing.T) { if err != nil { t.Fatalf("failed to open pipe for writing %q: %v", fileName, err) } - defer wfd.DecRef() + defer wfd.DecRef(ctx) const msg = "geh blau" checkEmpty(ctx, t, rfd) @@ -103,7 +103,7 @@ func TestNonblockingRead(t *testing.T) { func TestNonblockingWriteError(t *testing.T) { ctx, creds, vfsObj, root := setup(t) - defer root.DecRef() + defer root.DecRef(ctx) // Open the write side as nonblocking, which should return ENXIO. pop := vfs.PathOperation{ @@ -121,7 +121,7 @@ func TestNonblockingWriteError(t *testing.T) { func TestSingleFD(t *testing.T) { ctx, creds, vfsObj, root := setup(t) - defer root.DecRef() + defer root.DecRef(ctx) // Open the pipe as readable and writable. pop := vfs.PathOperation{ @@ -135,7 +135,7 @@ func TestSingleFD(t *testing.T) { if err != nil { t.Fatalf("failed to open pipe for writing %q: %v", fileName, err) } - defer fd.DecRef() + defer fd.DecRef(ctx) const msg = "forza blu" checkEmpty(ctx, t, fd) @@ -152,7 +152,7 @@ func setup(t *testing.T) (context.Context, *auth.Credentials, *vfs.VirtualFilesy // Create VFS. vfsObj := &vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { + if err := vfsObj.Init(ctx); err != nil { t.Fatalf("VFS init: %v", err) } vfsObj.MustRegisterFilesystemType("tmpfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index abbaa5d60..0710b65db 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -270,7 +270,7 @@ type regularFileFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *regularFileFD) Release() { +func (fd *regularFileFD) Release(context.Context) { // noop } diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 2545d88e9..68e615e8b 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -185,7 +185,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt case linux.S_IFDIR: root = &fs.newDirectory(rootKUID, rootKGID, rootMode).dentry default: - fs.vfsfs.DecRef() + fs.vfsfs.DecRef(ctx) return nil, nil, fmt.Errorf("invalid tmpfs root file type: %#o", rootFileType) } return &fs.vfsfs, &root.vfsd, nil @@ -197,7 +197,7 @@ func NewFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *au } // Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { +func (fs *filesystem) Release(ctx context.Context) { fs.vfsfs.VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) } @@ -249,12 +249,12 @@ func (d *dentry) TryIncRef() bool { } // DecRef implements vfs.DentryImpl.DecRef. -func (d *dentry) DecRef() { - d.inode.decRef() +func (d *dentry) DecRef(ctx context.Context) { + d.inode.decRef(ctx) } // InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. -func (d *dentry) InotifyWithParent(events, cookie uint32, et vfs.EventType) { +func (d *dentry) InotifyWithParent(ctx context.Context, events, cookie uint32, et vfs.EventType) { if d.inode.isDir() { events |= linux.IN_ISDIR } @@ -266,9 +266,9 @@ func (d *dentry) InotifyWithParent(events, cookie uint32, et vfs.EventType) { d.inode.fs.mu.RLock() // The ordering below is important, Linux always notifies the parent first. if d.parent != nil { - d.parent.inode.watches.Notify(d.name, events, cookie, et, deleted) + d.parent.inode.watches.Notify(ctx, d.name, events, cookie, et, deleted) } - d.inode.watches.Notify("", events, cookie, et, deleted) + d.inode.watches.Notify(ctx, "", events, cookie, et, deleted) d.inode.fs.mu.RUnlock() } @@ -278,7 +278,7 @@ func (d *dentry) Watches() *vfs.Watches { } // OnZeroWatches implements vfs.Dentry.OnZeroWatches. -func (d *dentry) OnZeroWatches() {} +func (d *dentry) OnZeroWatches(context.Context) {} // inode represents a filesystem object. type inode struct { @@ -359,12 +359,12 @@ func (i *inode) incLinksLocked() { // remove a reference on i as well. // // Preconditions: filesystem.mu must be locked for writing. i.nlink != 0. -func (i *inode) decLinksLocked() { +func (i *inode) decLinksLocked(ctx context.Context) { if i.nlink == 0 { panic("tmpfs.inode.decLinksLocked() called with no existing links") } if atomic.AddUint32(&i.nlink, ^uint32(0)) == 0 { - i.decRef() + i.decRef(ctx) } } @@ -386,9 +386,9 @@ func (i *inode) tryIncRef() bool { } } -func (i *inode) decRef() { +func (i *inode) decRef(ctx context.Context) { if refs := atomic.AddInt64(&i.refs, -1); refs == 0 { - i.watches.HandleDeletion() + i.watches.HandleDeletion(ctx) if regFile, ok := i.impl.(*regularFile); ok { // Release memory used by regFile to store data. Since regFile is // no longer usable, we don't need to grab any locks or update any @@ -701,7 +701,7 @@ func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) } if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { - d.InotifyWithParent(ev, 0, vfs.InodeEvent) + d.InotifyWithParent(ctx, ev, 0, vfs.InodeEvent) } return nil } @@ -724,7 +724,7 @@ func (fd *fileDescription) Setxattr(ctx context.Context, opts vfs.SetxattrOption } // Generate inotify events. - d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) + d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil } @@ -736,13 +736,13 @@ func (fd *fileDescription) Removexattr(ctx context.Context, name string) error { } // Generate inotify events. - d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) + d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil } // NewMemfd creates a new tmpfs regular file and file description that can back // an anonymous fd created by memfd_create. -func NewMemfd(mount *vfs.Mount, creds *auth.Credentials, allowSeals bool, name string) (*vfs.FileDescription, error) { +func NewMemfd(ctx context.Context, creds *auth.Credentials, mount *vfs.Mount, allowSeals bool, name string) (*vfs.FileDescription, error) { fs, ok := mount.Filesystem().Impl().(*filesystem) if !ok { panic("NewMemfd() called with non-tmpfs mount") @@ -757,7 +757,7 @@ func NewMemfd(mount *vfs.Mount, creds *auth.Credentials, allowSeals bool, name s } d := fs.newDentry(inode) - defer d.DecRef() + defer d.DecRef(ctx) d.name = name // Per Linux, mm/shmem.c:__shmem_file_setup(), memfd files are set up with diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs_test.go b/pkg/sentry/fsimpl/tmpfs/tmpfs_test.go index a240fb276..6f3e3ae6f 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs_test.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs_test.go @@ -34,7 +34,7 @@ func newTmpfsRoot(ctx context.Context) (*vfs.VirtualFilesystem, vfs.VirtualDentr creds := auth.CredentialsFromContext(ctx) vfsObj := &vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { + if err := vfsObj.Init(ctx); err != nil { return nil, vfs.VirtualDentry{}, nil, fmt.Errorf("VFS init: %v", err) } @@ -47,8 +47,8 @@ func newTmpfsRoot(ctx context.Context) (*vfs.VirtualFilesystem, vfs.VirtualDentr } root := mntns.Root() return vfsObj, root, func() { - root.DecRef() - mntns.DecRef() + root.DecRef(ctx) + mntns.DecRef(ctx) }, nil } diff --git a/pkg/sentry/kernel/abstract_socket_namespace.go b/pkg/sentry/kernel/abstract_socket_namespace.go index 920fe4329..52ed5cea2 100644 --- a/pkg/sentry/kernel/abstract_socket_namespace.go +++ b/pkg/sentry/kernel/abstract_socket_namespace.go @@ -17,6 +17,7 @@ package kernel import ( "syscall" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sync" @@ -31,7 +32,7 @@ type abstractEndpoint struct { } // WeakRefGone implements refs.WeakRefUser.WeakRefGone. -func (e *abstractEndpoint) WeakRefGone() { +func (e *abstractEndpoint) WeakRefGone(context.Context) { e.ns.mu.Lock() if e.ns.endpoints[e.name].ep == e.ep { delete(e.ns.endpoints, e.name) @@ -64,9 +65,9 @@ type boundEndpoint struct { } // Release implements transport.BoundEndpoint.Release. -func (e *boundEndpoint) Release() { - e.rc.DecRef() - e.BoundEndpoint.Release() +func (e *boundEndpoint) Release(ctx context.Context) { + e.rc.DecRef(ctx) + e.BoundEndpoint.Release(ctx) } // BoundEndpoint retrieves the endpoint bound to the given name. The return @@ -93,13 +94,13 @@ func (a *AbstractSocketNamespace) BoundEndpoint(name string) transport.BoundEndp // // When the last reference managed by rc is dropped, ep may be removed from the // namespace. -func (a *AbstractSocketNamespace) Bind(name string, ep transport.BoundEndpoint, rc refs.RefCounter) error { +func (a *AbstractSocketNamespace) Bind(ctx context.Context, name string, ep transport.BoundEndpoint, rc refs.RefCounter) error { a.mu.Lock() defer a.mu.Unlock() if ep, ok := a.endpoints[name]; ok { if rc := ep.wr.Get(); rc != nil { - rc.DecRef() + rc.DecRef(ctx) return syscall.EADDRINUSE } } diff --git a/pkg/sentry/kernel/epoll/epoll.go b/pkg/sentry/kernel/epoll/epoll.go index 4c0f1e41f..15519f0df 100644 --- a/pkg/sentry/kernel/epoll/epoll.go +++ b/pkg/sentry/kernel/epoll/epoll.go @@ -76,8 +76,8 @@ type pollEntry struct { // WeakRefGone implements refs.WeakRefUser.WeakRefGone. // weakReferenceGone is called when the file in the weak reference is destroyed. // The poll entry is removed in response to this. -func (p *pollEntry) WeakRefGone() { - p.epoll.RemoveEntry(p.id) +func (p *pollEntry) WeakRefGone(ctx context.Context) { + p.epoll.RemoveEntry(ctx, p.id) } // EventPoll holds all the state associated with an event poll object, that is, @@ -144,14 +144,14 @@ func NewEventPoll(ctx context.Context) *fs.File { // name matches fs/eventpoll.c:epoll_create1. dirent := fs.NewDirent(ctx, anon.NewInode(ctx), fmt.Sprintf("anon_inode:[eventpoll]")) // Release the initial dirent reference after NewFile takes a reference. - defer dirent.DecRef() + defer dirent.DecRef(ctx) return fs.NewFile(ctx, dirent, fs.FileFlags{}, &EventPoll{ files: make(map[FileIdentifier]*pollEntry), }) } // Release implements fs.FileOperations.Release. -func (e *EventPoll) Release() { +func (e *EventPoll) Release(ctx context.Context) { // We need to take the lock now because files may be attempting to // remove entries in parallel if they get destroyed. e.mu.Lock() @@ -160,7 +160,7 @@ func (e *EventPoll) Release() { // Go through all entries and clean up. for _, entry := range e.files { entry.id.File.EventUnregister(&entry.waiter) - entry.file.Drop() + entry.file.Drop(ctx) } e.files = nil } @@ -423,7 +423,7 @@ func (e *EventPoll) UpdateEntry(id FileIdentifier, flags EntryFlags, mask waiter } // RemoveEntry a files from the collection of observed files. -func (e *EventPoll) RemoveEntry(id FileIdentifier) error { +func (e *EventPoll) RemoveEntry(ctx context.Context, id FileIdentifier) error { e.mu.Lock() defer e.mu.Unlock() @@ -445,7 +445,7 @@ func (e *EventPoll) RemoveEntry(id FileIdentifier) error { // Remove file from map, and drop weak reference. delete(e.files, id) - entry.file.Drop() + entry.file.Drop(ctx) return nil } diff --git a/pkg/sentry/kernel/epoll/epoll_test.go b/pkg/sentry/kernel/epoll/epoll_test.go index 22630e9c5..55b505593 100644 --- a/pkg/sentry/kernel/epoll/epoll_test.go +++ b/pkg/sentry/kernel/epoll/epoll_test.go @@ -26,7 +26,8 @@ func TestFileDestroyed(t *testing.T) { f := filetest.NewTestFile(t) id := FileIdentifier{f, 12} - efile := NewEventPoll(contexttest.Context(t)) + ctx := contexttest.Context(t) + efile := NewEventPoll(ctx) e := efile.FileOperations.(*EventPoll) if err := e.AddEntry(id, 0, waiter.EventIn, [2]int32{}); err != nil { t.Fatalf("addEntry failed: %v", err) @@ -44,7 +45,7 @@ func TestFileDestroyed(t *testing.T) { } // Destroy the file. Check that we get no more events. - f.DecRef() + f.DecRef(ctx) evt = e.ReadEvents(1) if len(evt) != 0 { diff --git a/pkg/sentry/kernel/eventfd/eventfd.go b/pkg/sentry/kernel/eventfd/eventfd.go index 87951adeb..bbf568dfc 100644 --- a/pkg/sentry/kernel/eventfd/eventfd.go +++ b/pkg/sentry/kernel/eventfd/eventfd.go @@ -70,7 +70,7 @@ func New(ctx context.Context, initVal uint64, semMode bool) *fs.File { // name matches fs/eventfd.c:eventfd_file_create. dirent := fs.NewDirent(ctx, anon.NewInode(ctx), "anon_inode:[eventfd]") // Release the initial dirent reference after NewFile takes a reference. - defer dirent.DecRef() + defer dirent.DecRef(ctx) return fs.NewFile(ctx, dirent, fs.FileFlags{Read: true, Write: true}, &EventOperations{ val: initVal, semMode: semMode, @@ -106,7 +106,7 @@ func (e *EventOperations) HostFD() (int, error) { } // Release implements fs.FileOperations.Release. -func (e *EventOperations) Release() { +func (e *EventOperations) Release(context.Context) { e.mu.Lock() defer e.mu.Unlock() if e.hostfd >= 0 { diff --git a/pkg/sentry/kernel/fd_table.go b/pkg/sentry/kernel/fd_table.go index 4b7d234a4..ce53af69b 100644 --- a/pkg/sentry/kernel/fd_table.go +++ b/pkg/sentry/kernel/fd_table.go @@ -98,7 +98,7 @@ type FDTable struct { func (f *FDTable) saveDescriptorTable() map[int32]descriptor { m := make(map[int32]descriptor) - f.forEach(func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { + f.forEach(context.Background(), func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { m[fd] = descriptor{ file: file, fileVFS2: fileVFS2, @@ -109,6 +109,7 @@ func (f *FDTable) saveDescriptorTable() map[int32]descriptor { } func (f *FDTable) loadDescriptorTable(m map[int32]descriptor) { + ctx := context.Background() f.init() // Initialize table. for fd, d := range m { f.setAll(fd, d.file, d.fileVFS2, d.flags) @@ -118,9 +119,9 @@ func (f *FDTable) loadDescriptorTable(m map[int32]descriptor) { // reference taken by set above. switch { case d.file != nil: - d.file.DecRef() + d.file.DecRef(ctx) case d.fileVFS2 != nil: - d.fileVFS2.DecRef() + d.fileVFS2.DecRef(ctx) } } } @@ -144,14 +145,15 @@ func (f *FDTable) drop(file *fs.File) { d.InotifyEvent(ev, 0) // Drop the table reference. - file.DecRef() + file.DecRef(context.Background()) } // dropVFS2 drops the table reference. func (f *FDTable) dropVFS2(file *vfs.FileDescription) { // Release any POSIX lock possibly held by the FDTable. Range {0, 0} means the // entire file. - err := file.UnlockPOSIX(context.Background(), f, 0, 0, linux.SEEK_SET) + ctx := context.Background() + err := file.UnlockPOSIX(ctx, f, 0, 0, linux.SEEK_SET) if err != nil && err != syserror.ENOLCK { panic(fmt.Sprintf("UnlockPOSIX failed: %v", err)) } @@ -161,10 +163,10 @@ func (f *FDTable) dropVFS2(file *vfs.FileDescription) { if file.IsWritable() { ev = linux.IN_CLOSE_WRITE } - file.Dentry().InotifyWithParent(ev, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(ctx, ev, 0, vfs.PathEvent) // Drop the table's reference. - file.DecRef() + file.DecRef(ctx) } // NewFDTable allocates a new FDTable that may be used by tasks in k. @@ -175,15 +177,15 @@ func (k *Kernel) NewFDTable() *FDTable { } // destroy removes all of the file descriptors from the map. -func (f *FDTable) destroy() { - f.RemoveIf(func(*fs.File, *vfs.FileDescription, FDFlags) bool { +func (f *FDTable) destroy(ctx context.Context) { + f.RemoveIf(ctx, func(*fs.File, *vfs.FileDescription, FDFlags) bool { return true }) } // DecRef implements RefCounter.DecRef with destructor f.destroy. -func (f *FDTable) DecRef() { - f.DecRefWithDestructor(f.destroy) +func (f *FDTable) DecRef(ctx context.Context) { + f.DecRefWithDestructor(ctx, f.destroy) } // Size returns the number of file descriptor slots currently allocated. @@ -195,7 +197,7 @@ func (f *FDTable) Size() int { // forEach iterates over all non-nil files in sorted order. // // It is the caller's responsibility to acquire an appropriate lock. -func (f *FDTable) forEach(fn func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags)) { +func (f *FDTable) forEach(ctx context.Context, fn func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags)) { // retries tracks the number of failed TryIncRef attempts for the same FD. retries := 0 fd := int32(0) @@ -214,7 +216,7 @@ func (f *FDTable) forEach(fn func(fd int32, file *fs.File, fileVFS2 *vfs.FileDes continue // Race caught. } fn(fd, file, nil, flags) - file.DecRef() + file.DecRef(ctx) case fileVFS2 != nil: if !fileVFS2.TryIncRef() { retries++ @@ -224,7 +226,7 @@ func (f *FDTable) forEach(fn func(fd int32, file *fs.File, fileVFS2 *vfs.FileDes continue // Race caught. } fn(fd, nil, fileVFS2, flags) - fileVFS2.DecRef() + fileVFS2.DecRef(ctx) } retries = 0 fd++ @@ -234,7 +236,8 @@ func (f *FDTable) forEach(fn func(fd int32, file *fs.File, fileVFS2 *vfs.FileDes // String is a stringer for FDTable. func (f *FDTable) String() string { var buf strings.Builder - f.forEach(func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { + ctx := context.Background() + f.forEach(ctx, func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { switch { case file != nil: n, _ := file.Dirent.FullName(nil /* root */) @@ -242,7 +245,7 @@ func (f *FDTable) String() string { case fileVFS2 != nil: vfsObj := fileVFS2.Mount().Filesystem().VirtualFilesystem() - name, err := vfsObj.PathnameWithDeleted(context.Background(), vfs.VirtualDentry{}, fileVFS2.VirtualDentry()) + name, err := vfsObj.PathnameWithDeleted(ctx, vfs.VirtualDentry{}, fileVFS2.VirtualDentry()) if err != nil { fmt.Fprintf(&buf, "\n", err) return @@ -541,9 +544,9 @@ func (f *FDTable) GetVFS2(fd int32) (*vfs.FileDescription, FDFlags) { // // Precondition: The caller must be running on the task goroutine, or Task.mu // must be locked. -func (f *FDTable) GetFDs() []int32 { +func (f *FDTable) GetFDs(ctx context.Context) []int32 { fds := make([]int32, 0, int(atomic.LoadInt32(&f.used))) - f.forEach(func(fd int32, _ *fs.File, _ *vfs.FileDescription, _ FDFlags) { + f.forEach(ctx, func(fd int32, _ *fs.File, _ *vfs.FileDescription, _ FDFlags) { fds = append(fds, fd) }) return fds @@ -552,9 +555,9 @@ func (f *FDTable) GetFDs() []int32 { // GetRefs returns a stable slice of references to all files and bumps the // reference count on each. The caller must use DecRef on each reference when // they're done using the slice. -func (f *FDTable) GetRefs() []*fs.File { +func (f *FDTable) GetRefs(ctx context.Context) []*fs.File { files := make([]*fs.File, 0, f.Size()) - f.forEach(func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) { + f.forEach(ctx, func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) { file.IncRef() // Acquire a reference for caller. files = append(files, file) }) @@ -564,9 +567,9 @@ func (f *FDTable) GetRefs() []*fs.File { // GetRefsVFS2 returns a stable slice of references to all files and bumps the // reference count on each. The caller must use DecRef on each reference when // they're done using the slice. -func (f *FDTable) GetRefsVFS2() []*vfs.FileDescription { +func (f *FDTable) GetRefsVFS2(ctx context.Context) []*vfs.FileDescription { files := make([]*vfs.FileDescription, 0, f.Size()) - f.forEach(func(_ int32, _ *fs.File, file *vfs.FileDescription, _ FDFlags) { + f.forEach(ctx, func(_ int32, _ *fs.File, file *vfs.FileDescription, _ FDFlags) { file.IncRef() // Acquire a reference for caller. files = append(files, file) }) @@ -574,10 +577,10 @@ func (f *FDTable) GetRefsVFS2() []*vfs.FileDescription { } // Fork returns an independent FDTable. -func (f *FDTable) Fork() *FDTable { +func (f *FDTable) Fork(ctx context.Context) *FDTable { clone := f.k.NewFDTable() - f.forEach(func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { + f.forEach(ctx, func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { // The set function here will acquire an appropriate table // reference for the clone. We don't need anything else. switch { @@ -622,11 +625,11 @@ func (f *FDTable) Remove(fd int32) (*fs.File, *vfs.FileDescription) { } // RemoveIf removes all FDs where cond is true. -func (f *FDTable) RemoveIf(cond func(*fs.File, *vfs.FileDescription, FDFlags) bool) { +func (f *FDTable) RemoveIf(ctx context.Context, cond func(*fs.File, *vfs.FileDescription, FDFlags) bool) { f.mu.Lock() defer f.mu.Unlock() - f.forEach(func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { + f.forEach(ctx, func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { if cond(file, fileVFS2, flags) { f.set(fd, nil, FDFlags{}) // Clear from table. // Update current available position. diff --git a/pkg/sentry/kernel/fd_table_test.go b/pkg/sentry/kernel/fd_table_test.go index 29f95a2c4..e3f30ba2a 100644 --- a/pkg/sentry/kernel/fd_table_test.go +++ b/pkg/sentry/kernel/fd_table_test.go @@ -154,7 +154,7 @@ func TestFDTable(t *testing.T) { if ref == nil { t.Fatalf("fdTable.Remove(1) for an existing FD: failed, want success") } - ref.DecRef() + ref.DecRef(ctx) if ref, _ := fdTable.Remove(1); ref != nil { t.Fatalf("r.Remove(1) for a removed FD: got success, want failure") @@ -191,7 +191,7 @@ func BenchmarkFDLookupAndDecRef(b *testing.B) { b.StartTimer() // Benchmark. for i := 0; i < b.N; i++ { tf, _ := fdTable.Get(fds[i%len(fds)]) - tf.DecRef() + tf.DecRef(ctx) } }) } @@ -219,7 +219,7 @@ func BenchmarkFDLookupAndDecRefConcurrent(b *testing.B) { defer wg.Done() for i := 0; i < each; i++ { tf, _ := fdTable.Get(fds[i%len(fds)]) - tf.DecRef() + tf.DecRef(ctx) } }() } diff --git a/pkg/sentry/kernel/fs_context.go b/pkg/sentry/kernel/fs_context.go index 47f78df9a..8f2d36d5a 100644 --- a/pkg/sentry/kernel/fs_context.go +++ b/pkg/sentry/kernel/fs_context.go @@ -17,6 +17,7 @@ package kernel import ( "fmt" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/vfs" @@ -89,28 +90,28 @@ func NewFSContextVFS2(root, cwd vfs.VirtualDentry, umask uint) *FSContext { // Note that there may still be calls to WorkingDirectory() or RootDirectory() // (that return nil). This is because valid references may still be held via // proc files or other mechanisms. -func (f *FSContext) destroy() { +func (f *FSContext) destroy(ctx context.Context) { // Hold f.mu so that we don't race with RootDirectory() and // WorkingDirectory(). f.mu.Lock() defer f.mu.Unlock() if VFS2Enabled { - f.rootVFS2.DecRef() + f.rootVFS2.DecRef(ctx) f.rootVFS2 = vfs.VirtualDentry{} - f.cwdVFS2.DecRef() + f.cwdVFS2.DecRef(ctx) f.cwdVFS2 = vfs.VirtualDentry{} } else { - f.root.DecRef() + f.root.DecRef(ctx) f.root = nil - f.cwd.DecRef() + f.cwd.DecRef(ctx) f.cwd = nil } } // DecRef implements RefCounter.DecRef with destructor f.destroy. -func (f *FSContext) DecRef() { - f.DecRefWithDestructor(f.destroy) +func (f *FSContext) DecRef(ctx context.Context) { + f.DecRefWithDestructor(ctx, f.destroy) } // Fork forks this FSContext. @@ -165,7 +166,7 @@ func (f *FSContext) WorkingDirectoryVFS2() vfs.VirtualDentry { // This will take an extra reference on the Dirent. // // This is not a valid call after destroy. -func (f *FSContext) SetWorkingDirectory(d *fs.Dirent) { +func (f *FSContext) SetWorkingDirectory(ctx context.Context, d *fs.Dirent) { if d == nil { panic("FSContext.SetWorkingDirectory called with nil dirent") } @@ -180,21 +181,21 @@ func (f *FSContext) SetWorkingDirectory(d *fs.Dirent) { old := f.cwd f.cwd = d d.IncRef() - old.DecRef() + old.DecRef(ctx) } // SetWorkingDirectoryVFS2 sets the current working directory. // This will take an extra reference on the VirtualDentry. // // This is not a valid call after destroy. -func (f *FSContext) SetWorkingDirectoryVFS2(d vfs.VirtualDentry) { +func (f *FSContext) SetWorkingDirectoryVFS2(ctx context.Context, d vfs.VirtualDentry) { f.mu.Lock() defer f.mu.Unlock() old := f.cwdVFS2 f.cwdVFS2 = d d.IncRef() - old.DecRef() + old.DecRef(ctx) } // RootDirectory returns the current filesystem root. @@ -226,7 +227,7 @@ func (f *FSContext) RootDirectoryVFS2() vfs.VirtualDentry { // This will take an extra reference on the Dirent. // // This is not a valid call after free. -func (f *FSContext) SetRootDirectory(d *fs.Dirent) { +func (f *FSContext) SetRootDirectory(ctx context.Context, d *fs.Dirent) { if d == nil { panic("FSContext.SetRootDirectory called with nil dirent") } @@ -241,13 +242,13 @@ func (f *FSContext) SetRootDirectory(d *fs.Dirent) { old := f.root f.root = d d.IncRef() - old.DecRef() + old.DecRef(ctx) } // SetRootDirectoryVFS2 sets the root directory. It takes a reference on vd. // // This is not a valid call after free. -func (f *FSContext) SetRootDirectoryVFS2(vd vfs.VirtualDentry) { +func (f *FSContext) SetRootDirectoryVFS2(ctx context.Context, vd vfs.VirtualDentry) { if !vd.Ok() { panic("FSContext.SetRootDirectoryVFS2 called with zero-value VirtualDentry") } @@ -263,7 +264,7 @@ func (f *FSContext) SetRootDirectoryVFS2(vd vfs.VirtualDentry) { vd.IncRef() f.rootVFS2 = vd f.mu.Unlock() - old.DecRef() + old.DecRef(ctx) } // Umask returns the current umask. diff --git a/pkg/sentry/kernel/futex/BUILD b/pkg/sentry/kernel/futex/BUILD index c5021f2db..daa2dae76 100644 --- a/pkg/sentry/kernel/futex/BUILD +++ b/pkg/sentry/kernel/futex/BUILD @@ -51,6 +51,7 @@ go_test( srcs = ["futex_test.go"], library = ":futex", deps = [ + "//pkg/context", "//pkg/sync", "//pkg/usermem", ], diff --git a/pkg/sentry/kernel/futex/futex.go b/pkg/sentry/kernel/futex/futex.go index bcc1b29a8..e4dcc4d40 100644 --- a/pkg/sentry/kernel/futex/futex.go +++ b/pkg/sentry/kernel/futex/futex.go @@ -19,6 +19,7 @@ package futex import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" @@ -66,9 +67,9 @@ type Key struct { Offset uint64 } -func (k *Key) release() { +func (k *Key) release(t Target) { if k.MappingIdentity != nil { - k.MappingIdentity.DecRef() + k.MappingIdentity.DecRef(t) } k.Mappable = nil k.MappingIdentity = nil @@ -94,6 +95,8 @@ func (k *Key) matches(k2 *Key) bool { // Target abstracts memory accesses and keys. type Target interface { + context.Context + // SwapUint32 gives access to usermem.IO.SwapUint32. SwapUint32(addr usermem.Addr, new uint32) (uint32, error) @@ -296,7 +299,7 @@ func (b *bucket) wakeWaiterLocked(w *Waiter) { // bucket "to". // // Preconditions: b and to must be locked. -func (b *bucket) requeueLocked(to *bucket, key, nkey *Key, n int) int { +func (b *bucket) requeueLocked(t Target, to *bucket, key, nkey *Key, n int) int { done := 0 for w := b.waiters.Front(); done < n && w != nil; { if !w.key.matches(key) { @@ -308,7 +311,7 @@ func (b *bucket) requeueLocked(to *bucket, key, nkey *Key, n int) int { requeued := w w = w.Next() // Next iteration. b.waiters.Remove(requeued) - requeued.key.release() + requeued.key.release(t) requeued.key = nkey.clone() to.waiters.PushBack(requeued) requeued.bucket.Store(to) @@ -456,7 +459,7 @@ func (m *Manager) Wake(t Target, addr usermem.Addr, private bool, bitmask uint32 r := b.wakeLocked(&k, bitmask, n) b.mu.Unlock() - k.release() + k.release(t) return r, nil } @@ -465,12 +468,12 @@ func (m *Manager) doRequeue(t Target, addr, naddr usermem.Addr, private bool, ch if err != nil { return 0, err } - defer k1.release() + defer k1.release(t) k2, err := getKey(t, naddr, private) if err != nil { return 0, err } - defer k2.release() + defer k2.release(t) b1, b2 := m.lockBuckets(&k1, &k2) defer b1.mu.Unlock() @@ -488,7 +491,7 @@ func (m *Manager) doRequeue(t Target, addr, naddr usermem.Addr, private bool, ch done := b1.wakeLocked(&k1, ^uint32(0), nwake) // Requeue the number required. - b1.requeueLocked(b2, &k1, &k2, nreq) + b1.requeueLocked(t, b2, &k1, &k2, nreq) return done, nil } @@ -515,12 +518,12 @@ func (m *Manager) WakeOp(t Target, addr1, addr2 usermem.Addr, private bool, nwak if err != nil { return 0, err } - defer k1.release() + defer k1.release(t) k2, err := getKey(t, addr2, private) if err != nil { return 0, err } - defer k2.release() + defer k2.release(t) b1, b2 := m.lockBuckets(&k1, &k2) defer b1.mu.Unlock() @@ -571,7 +574,7 @@ func (m *Manager) WaitPrepare(w *Waiter, t Target, addr usermem.Addr, private bo // Perform our atomic check. if err := check(t, addr, val); err != nil { b.mu.Unlock() - w.key.release() + w.key.release(t) return err } @@ -585,7 +588,7 @@ func (m *Manager) WaitPrepare(w *Waiter, t Target, addr usermem.Addr, private bo // WaitComplete must be called when a Waiter previously added by WaitPrepare is // no longer eligible to be woken. -func (m *Manager) WaitComplete(w *Waiter) { +func (m *Manager) WaitComplete(w *Waiter, t Target) { // Remove w from the bucket it's in. for { b := w.bucket.Load() @@ -617,7 +620,7 @@ func (m *Manager) WaitComplete(w *Waiter) { } // Release references held by the waiter. - w.key.release() + w.key.release(t) } // LockPI attempts to lock the futex following the Priority-inheritance futex @@ -648,13 +651,13 @@ func (m *Manager) LockPI(w *Waiter, t Target, addr usermem.Addr, tid uint32, pri success, err := m.lockPILocked(w, t, addr, tid, b, try) if err != nil { - w.key.release() + w.key.release(t) b.mu.Unlock() return false, err } if success || try { // Release waiter if it's not going to be a wait. - w.key.release() + w.key.release(t) } b.mu.Unlock() return success, nil @@ -730,7 +733,7 @@ func (m *Manager) UnlockPI(t Target, addr usermem.Addr, tid uint32, private bool err = m.unlockPILocked(t, addr, tid, b, &k) - k.release() + k.release(t) b.mu.Unlock() return err } diff --git a/pkg/sentry/kernel/futex/futex_test.go b/pkg/sentry/kernel/futex/futex_test.go index 7c5c7665b..d0128c548 100644 --- a/pkg/sentry/kernel/futex/futex_test.go +++ b/pkg/sentry/kernel/futex/futex_test.go @@ -22,6 +22,7 @@ import ( "testing" "unsafe" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/usermem" ) @@ -29,28 +30,33 @@ import ( // testData implements the Target interface, and allows us to // treat the address passed for futex operations as an index in // a byte slice for testing simplicity. -type testData []byte +type testData struct { + context.Context + data []byte +} const sizeofInt32 = 4 func newTestData(size uint) testData { - return make([]byte, size) + return testData{ + data: make([]byte, size), + } } func (t testData) SwapUint32(addr usermem.Addr, new uint32) (uint32, error) { - val := atomic.SwapUint32((*uint32)(unsafe.Pointer(&t[addr])), new) + val := atomic.SwapUint32((*uint32)(unsafe.Pointer(&t.data[addr])), new) return val, nil } func (t testData) CompareAndSwapUint32(addr usermem.Addr, old, new uint32) (uint32, error) { - if atomic.CompareAndSwapUint32((*uint32)(unsafe.Pointer(&t[addr])), old, new) { + if atomic.CompareAndSwapUint32((*uint32)(unsafe.Pointer(&t.data[addr])), old, new) { return old, nil } - return atomic.LoadUint32((*uint32)(unsafe.Pointer(&t[addr]))), nil + return atomic.LoadUint32((*uint32)(unsafe.Pointer(&t.data[addr]))), nil } func (t testData) LoadUint32(addr usermem.Addr) (uint32, error) { - return atomic.LoadUint32((*uint32)(unsafe.Pointer(&t[addr]))), nil + return atomic.LoadUint32((*uint32)(unsafe.Pointer(&t.data[addr]))), nil } func (t testData) GetSharedKey(addr usermem.Addr) (Key, error) { @@ -83,7 +89,7 @@ func TestFutexWake(t *testing.T) { // Start waiting for wakeup. w := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w) + defer m.WaitComplete(w, d) // Perform a wakeup. if n, err := m.Wake(d, 0, private, ^uint32(0), 1); err != nil || n != 1 { @@ -106,7 +112,7 @@ func TestFutexWakeBitmask(t *testing.T) { // Start waiting for wakeup. w := newPreparedTestWaiter(t, m, d, 0, private, 0, 0x0000ffff) - defer m.WaitComplete(w) + defer m.WaitComplete(w, d) // Perform a wakeup using the wrong bitmask. if n, err := m.Wake(d, 0, private, 0xffff0000, 1); err != nil || n != 0 { @@ -141,7 +147,7 @@ func TestFutexWakeTwo(t *testing.T) { var ws [3]*Waiter for i := range ws { ws[i] = newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(ws[i]) + defer m.WaitComplete(ws[i], d) } // Perform two wakeups. @@ -174,9 +180,9 @@ func TestFutexWakeUnrelated(t *testing.T) { // Start two waiters waiting for wakeup on different addresses. w1 := newPreparedTestWaiter(t, m, d, 0*sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w1) + defer m.WaitComplete(w1, d) w2 := newPreparedTestWaiter(t, m, d, 1*sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w2) + defer m.WaitComplete(w2, d) // Perform two wakeups on the second address. if n, err := m.Wake(d, 1*sizeofInt32, private, ^uint32(0), 2); err != nil || n != 1 { @@ -216,9 +222,9 @@ func TestWakeOpFirstNonEmpty(t *testing.T) { // Add two waiters on address 0. w1 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w1) + defer m.WaitComplete(w1, d) w2 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w2) + defer m.WaitComplete(w2, d) // Perform 10 wakeups on address 0. if n, err := m.WakeOp(d, 0, sizeofInt32, private, 10, 0, 0); err != nil || n != 2 { @@ -244,9 +250,9 @@ func TestWakeOpSecondNonEmpty(t *testing.T) { // Add two waiters on address sizeofInt32. w1 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w1) + defer m.WaitComplete(w1, d) w2 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w2) + defer m.WaitComplete(w2, d) // Perform 10 wakeups on address sizeofInt32 (contingent on // d.Op(0), which should succeed). @@ -273,9 +279,9 @@ func TestWakeOpSecondNonEmptyFailingOp(t *testing.T) { // Add two waiters on address sizeofInt32. w1 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w1) + defer m.WaitComplete(w1, d) w2 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w2) + defer m.WaitComplete(w2, d) // Perform 10 wakeups on address sizeofInt32 (contingent on // d.Op(1), which should fail). @@ -302,15 +308,15 @@ func TestWakeOpAllNonEmpty(t *testing.T) { // Add two waiters on address 0. w1 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w1) + defer m.WaitComplete(w1, d) w2 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w2) + defer m.WaitComplete(w2, d) // Add two waiters on address sizeofInt32. w3 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w3) + defer m.WaitComplete(w3, d) w4 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w4) + defer m.WaitComplete(w4, d) // Perform 10 wakeups on address 0 (unconditionally), and 10 // wakeups on address sizeofInt32 (contingent on d.Op(0), which @@ -344,15 +350,15 @@ func TestWakeOpAllNonEmptyFailingOp(t *testing.T) { // Add two waiters on address 0. w1 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w1) + defer m.WaitComplete(w1, d) w2 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w2) + defer m.WaitComplete(w2, d) // Add two waiters on address sizeofInt32. w3 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w3) + defer m.WaitComplete(w3, d) w4 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w4) + defer m.WaitComplete(w4, d) // Perform 10 wakeups on address 0 (unconditionally), and 10 // wakeups on address sizeofInt32 (contingent on d.Op(1), which @@ -388,7 +394,7 @@ func TestWakeOpSameAddress(t *testing.T) { var ws [4]*Waiter for i := range ws { ws[i] = newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(ws[i]) + defer m.WaitComplete(ws[i], d) } // Perform 1 wakeup on address 0 (unconditionally), and 1 wakeup @@ -422,7 +428,7 @@ func TestWakeOpSameAddressFailingOp(t *testing.T) { var ws [4]*Waiter for i := range ws { ws[i] = newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(ws[i]) + defer m.WaitComplete(ws[i], d) } // Perform 1 wakeup on address 0 (unconditionally), and 1 wakeup @@ -472,7 +478,7 @@ func (t *testMutex) Lock() { for { // Attempt to grab the lock. if atomic.CompareAndSwapUint32( - (*uint32)(unsafe.Pointer(&t.d[t.a])), + (*uint32)(unsafe.Pointer(&t.d.data[t.a])), testMutexUnlocked, testMutexLocked) { // Lock held. @@ -490,7 +496,7 @@ func (t *testMutex) Lock() { panic("WaitPrepare returned unexpected error: " + err.Error()) } <-w.C - t.m.WaitComplete(w) + t.m.WaitComplete(w, t.d) } } @@ -498,7 +504,7 @@ func (t *testMutex) Lock() { // This will notify any waiters via the futex manager. func (t *testMutex) Unlock() { // Unlock. - atomic.StoreUint32((*uint32)(unsafe.Pointer(&t.d[t.a])), testMutexUnlocked) + atomic.StoreUint32((*uint32)(unsafe.Pointer(&t.d.data[t.a])), testMutexUnlocked) // Notify all waiters. t.m.Wake(t.d, t.a, true, ^uint32(0), math.MaxInt32) diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 15dae0f5b..316df249d 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -376,7 +376,8 @@ func (k *Kernel) Init(args InitKernelArgs) error { k.netlinkPorts = port.New() if VFS2Enabled { - if err := k.vfs.Init(); err != nil { + ctx := k.SupervisorContext() + if err := k.vfs.Init(ctx); err != nil { return fmt.Errorf("failed to initialize VFS: %v", err) } @@ -384,19 +385,19 @@ func (k *Kernel) Init(args InitKernelArgs) error { if err != nil { return fmt.Errorf("failed to create pipefs filesystem: %v", err) } - defer pipeFilesystem.DecRef() + defer pipeFilesystem.DecRef(ctx) pipeMount, err := k.vfs.NewDisconnectedMount(pipeFilesystem, nil, &vfs.MountOptions{}) if err != nil { return fmt.Errorf("failed to create pipefs mount: %v", err) } k.pipeMount = pipeMount - tmpfsFilesystem, tmpfsRoot, err := tmpfs.NewFilesystem(k.SupervisorContext(), &k.vfs, auth.NewRootCredentials(k.rootUserNamespace)) + tmpfsFilesystem, tmpfsRoot, err := tmpfs.NewFilesystem(ctx, &k.vfs, auth.NewRootCredentials(k.rootUserNamespace)) if err != nil { return fmt.Errorf("failed to create tmpfs filesystem: %v", err) } - defer tmpfsFilesystem.DecRef() - defer tmpfsRoot.DecRef() + defer tmpfsFilesystem.DecRef(ctx) + defer tmpfsRoot.DecRef(ctx) shmMount, err := k.vfs.NewDisconnectedMount(tmpfsFilesystem, tmpfsRoot, &vfs.MountOptions{}) if err != nil { return fmt.Errorf("failed to create tmpfs mount: %v", err) @@ -407,7 +408,7 @@ func (k *Kernel) Init(args InitKernelArgs) error { if err != nil { return fmt.Errorf("failed to create sockfs filesystem: %v", err) } - defer socketFilesystem.DecRef() + defer socketFilesystem.DecRef(ctx) socketMount, err := k.vfs.NewDisconnectedMount(socketFilesystem, nil, &vfs.MountOptions{}) if err != nil { return fmt.Errorf("failed to create sockfs mount: %v", err) @@ -430,8 +431,8 @@ func (k *Kernel) SaveTo(w wire.Writer) error { defer k.extMu.Unlock() // Stop time. - k.pauseTimeLocked() - defer k.resumeTimeLocked() + k.pauseTimeLocked(ctx) + defer k.resumeTimeLocked(ctx) // Evict all evictable MemoryFile allocations. k.mf.StartEvictions() @@ -447,12 +448,12 @@ func (k *Kernel) SaveTo(w wire.Writer) error { // Remove all epoll waiter objects from underlying wait queues. // NOTE: for programs to resume execution in future snapshot scenarios, // we will need to re-establish these waiter objects after saving. - k.tasks.unregisterEpollWaiters() + k.tasks.unregisterEpollWaiters(ctx) // Clear the dirent cache before saving because Dirents must be Loaded in a // particular order (parents before children), and Loading dirents from a cache // breaks that order. - if err := k.flushMountSourceRefs(); err != nil { + if err := k.flushMountSourceRefs(ctx); err != nil { return err } @@ -505,7 +506,7 @@ func (k *Kernel) SaveTo(w wire.Writer) error { // flushMountSourceRefs flushes the MountSources for all mounted filesystems // and open FDs. -func (k *Kernel) flushMountSourceRefs() error { +func (k *Kernel) flushMountSourceRefs(ctx context.Context) error { // Flush all mount sources for currently mounted filesystems in each task. flushed := make(map[*fs.MountNamespace]struct{}) k.tasks.mu.RLock() @@ -521,7 +522,7 @@ func (k *Kernel) flushMountSourceRefs() error { // There may be some open FDs whose filesystems have been unmounted. We // must flush those as well. - return k.tasks.forEachFDPaused(func(file *fs.File, _ *vfs.FileDescription) error { + return k.tasks.forEachFDPaused(ctx, func(file *fs.File, _ *vfs.FileDescription) error { file.Dirent.Inode.MountSource.FlushDirentRefs() return nil }) @@ -531,7 +532,7 @@ func (k *Kernel) flushMountSourceRefs() error { // each task. // // Precondition: Must be called with the kernel paused. -func (ts *TaskSet) forEachFDPaused(f func(*fs.File, *vfs.FileDescription) error) (err error) { +func (ts *TaskSet) forEachFDPaused(ctx context.Context, f func(*fs.File, *vfs.FileDescription) error) (err error) { // TODO(gvisor.dev/issue/1663): Add save support for VFS2. if VFS2Enabled { return nil @@ -544,7 +545,7 @@ func (ts *TaskSet) forEachFDPaused(f func(*fs.File, *vfs.FileDescription) error) if t.fdTable == nil { continue } - t.fdTable.forEach(func(_ int32, file *fs.File, fileVFS2 *vfs.FileDescription, _ FDFlags) { + t.fdTable.forEach(ctx, func(_ int32, file *fs.File, fileVFS2 *vfs.FileDescription, _ FDFlags) { if lastErr := f(file, fileVFS2); lastErr != nil && err == nil { err = lastErr } @@ -555,7 +556,7 @@ func (ts *TaskSet) forEachFDPaused(f func(*fs.File, *vfs.FileDescription) error) func (ts *TaskSet) flushWritesToFiles(ctx context.Context) error { // TODO(gvisor.dev/issue/1663): Add save support for VFS2. - return ts.forEachFDPaused(func(file *fs.File, _ *vfs.FileDescription) error { + return ts.forEachFDPaused(ctx, func(file *fs.File, _ *vfs.FileDescription) error { if flags := file.Flags(); !flags.Write { return nil } @@ -602,7 +603,7 @@ func (k *Kernel) invalidateUnsavableMappings(ctx context.Context) error { return nil } -func (ts *TaskSet) unregisterEpollWaiters() { +func (ts *TaskSet) unregisterEpollWaiters(ctx context.Context) { // TODO(gvisor.dev/issue/1663): Add save support for VFS2. if VFS2Enabled { return @@ -623,7 +624,7 @@ func (ts *TaskSet) unregisterEpollWaiters() { if _, ok := processed[t.fdTable]; ok { continue } - t.fdTable.forEach(func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) { + t.fdTable.forEach(ctx, func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) { if e, ok := file.FileOperations.(*epoll.EventPoll); ok { e.UnregisterEpollWaiters() } @@ -900,7 +901,7 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, root := args.MountNamespaceVFS2.Root() // The call to newFSContext below will take a reference on root, so we // don't need to hold this one. - defer root.DecRef() + defer root.DecRef(ctx) // Grab the working directory. wd := root // Default. @@ -918,7 +919,7 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, if err != nil { return nil, 0, fmt.Errorf("failed to find initial working directory %q: %v", args.WorkingDirectory, err) } - defer wd.DecRef() + defer wd.DecRef(ctx) } opener = fsbridge.NewVFSLookup(mntnsVFS2, root, wd) fsContext = NewFSContextVFS2(root, wd, args.Umask) @@ -933,7 +934,7 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, root := mntns.Root() // The call to newFSContext below will take a reference on root, so we // don't need to hold this one. - defer root.DecRef() + defer root.DecRef(ctx) // Grab the working directory. remainingTraversals := args.MaxSymlinkTraversals @@ -944,7 +945,7 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, if err != nil { return nil, 0, fmt.Errorf("failed to find initial working directory %q: %v", args.WorkingDirectory, err) } - defer wd.DecRef() + defer wd.DecRef(ctx) } opener = fsbridge.NewFSLookup(mntns, root, wd) fsContext = newFSContext(root, wd, args.Umask) @@ -1054,7 +1055,7 @@ func (k *Kernel) Start() error { // If k was created by LoadKernelFrom, timers were stopped during // Kernel.SaveTo and need to be resumed. If k was created by NewKernel, // this is a no-op. - k.resumeTimeLocked() + k.resumeTimeLocked(k.SupervisorContext()) // Start task goroutines. k.tasks.mu.RLock() defer k.tasks.mu.RUnlock() @@ -1068,7 +1069,7 @@ func (k *Kernel) Start() error { // // Preconditions: Any task goroutines running in k must be stopped. k.extMu // must be locked. -func (k *Kernel) pauseTimeLocked() { +func (k *Kernel) pauseTimeLocked(ctx context.Context) { // k.cpuClockTicker may be nil since Kernel.SaveTo() may be called before // Kernel.Start(). if k.cpuClockTicker != nil { @@ -1090,7 +1091,7 @@ func (k *Kernel) pauseTimeLocked() { // This means we'll iterate FDTables shared by multiple tasks repeatedly, // but ktime.Timer.Pause is idempotent so this is harmless. if t.fdTable != nil { - t.fdTable.forEach(func(_ int32, file *fs.File, fd *vfs.FileDescription, _ FDFlags) { + t.fdTable.forEach(ctx, func(_ int32, file *fs.File, fd *vfs.FileDescription, _ FDFlags) { if VFS2Enabled { if tfd, ok := fd.Impl().(*timerfd.TimerFileDescription); ok { tfd.PauseTimer() @@ -1112,7 +1113,7 @@ func (k *Kernel) pauseTimeLocked() { // // Preconditions: Any task goroutines running in k must be stopped. k.extMu // must be locked. -func (k *Kernel) resumeTimeLocked() { +func (k *Kernel) resumeTimeLocked(ctx context.Context) { if k.cpuClockTicker != nil { k.cpuClockTicker.Resume() } @@ -1126,7 +1127,7 @@ func (k *Kernel) resumeTimeLocked() { } } if t.fdTable != nil { - t.fdTable.forEach(func(_ int32, file *fs.File, fd *vfs.FileDescription, _ FDFlags) { + t.fdTable.forEach(ctx, func(_ int32, file *fs.File, fd *vfs.FileDescription, _ FDFlags) { if VFS2Enabled { if tfd, ok := fd.Impl().(*timerfd.TimerFileDescription); ok { tfd.ResumeTimer() @@ -1511,7 +1512,7 @@ type SocketEntry struct { } // WeakRefGone implements refs.WeakRefUser.WeakRefGone. -func (s *SocketEntry) WeakRefGone() { +func (s *SocketEntry) WeakRefGone(context.Context) { s.k.extMu.Lock() s.k.sockets.Remove(s) s.k.extMu.Unlock() @@ -1600,7 +1601,7 @@ func (ctx supervisorContext) Value(key interface{}) interface{} { return vfs.VirtualDentry{} } mntns := ctx.k.GlobalInit().Leader().MountNamespaceVFS2() - defer mntns.DecRef() + defer mntns.DecRef(ctx) // Root() takes a reference on the root dirent for us. return mntns.Root() case vfs.CtxMountNamespace: diff --git a/pkg/sentry/kernel/pipe/node.go b/pkg/sentry/kernel/pipe/node.go index 4b688c627..6497dc4ba 100644 --- a/pkg/sentry/kernel/pipe/node.go +++ b/pkg/sentry/kernel/pipe/node.go @@ -93,7 +93,7 @@ func (i *inodeOperations) GetFile(ctx context.Context, d *fs.Dirent, flags fs.Fi if i.p.isNamed && !flags.NonBlocking && !i.p.HasWriters() { if !waitFor(&i.mu, &i.wWakeup, ctx) { - r.DecRef() + r.DecRef(ctx) return nil, syserror.ErrInterrupted } } @@ -111,12 +111,12 @@ func (i *inodeOperations) GetFile(ctx context.Context, d *fs.Dirent, flags fs.Fi // On a nonblocking, write-only open, the open fails with ENXIO if the // read side isn't open yet. if flags.NonBlocking { - w.DecRef() + w.DecRef(ctx) return nil, syserror.ENXIO } if !waitFor(&i.mu, &i.rWakeup, ctx) { - w.DecRef() + w.DecRef(ctx) return nil, syserror.ErrInterrupted } } diff --git a/pkg/sentry/kernel/pipe/node_test.go b/pkg/sentry/kernel/pipe/node_test.go index ab75a87ff..ce0db5583 100644 --- a/pkg/sentry/kernel/pipe/node_test.go +++ b/pkg/sentry/kernel/pipe/node_test.go @@ -167,7 +167,7 @@ func TestClosedReaderBlocksWriteOpen(t *testing.T) { f := NewInodeOperations(ctx, perms, newNamedPipe(t)) rFile, _ := testOpenOrDie(ctx, t, f, fs.FileFlags{Read: true, NonBlocking: true}, nil) - rFile.DecRef() + rFile.DecRef(ctx) wDone := make(chan struct{}) // This open for write should block because the reader is now gone. diff --git a/pkg/sentry/kernel/pipe/pipe.go b/pkg/sentry/kernel/pipe/pipe.go index 79645d7d2..297e8f28f 100644 --- a/pkg/sentry/kernel/pipe/pipe.go +++ b/pkg/sentry/kernel/pipe/pipe.go @@ -152,7 +152,7 @@ func NewConnectedPipe(ctx context.Context, sizeBytes, atomicIOBytes int64) (*fs. d := fs.NewDirent(ctx, fs.NewInode(ctx, iops, ms, sattr), fmt.Sprintf("pipe:[%d]", ino)) // The p.Open calls below will each take a reference on the Dirent. We // must drop the one we already have. - defer d.DecRef() + defer d.DecRef(ctx) return p.Open(ctx, d, fs.FileFlags{Read: true}), p.Open(ctx, d, fs.FileFlags{Write: true}) } diff --git a/pkg/sentry/kernel/pipe/pipe_test.go b/pkg/sentry/kernel/pipe/pipe_test.go index bda739dbe..fe97e9800 100644 --- a/pkg/sentry/kernel/pipe/pipe_test.go +++ b/pkg/sentry/kernel/pipe/pipe_test.go @@ -27,8 +27,8 @@ import ( func TestPipeRW(t *testing.T) { ctx := contexttest.Context(t) r, w := NewConnectedPipe(ctx, 65536, 4096) - defer r.DecRef() - defer w.DecRef() + defer r.DecRef(ctx) + defer w.DecRef(ctx) msg := []byte("here's some bytes") wantN := int64(len(msg)) @@ -47,8 +47,8 @@ func TestPipeRW(t *testing.T) { func TestPipeReadBlock(t *testing.T) { ctx := contexttest.Context(t) r, w := NewConnectedPipe(ctx, 65536, 4096) - defer r.DecRef() - defer w.DecRef() + defer r.DecRef(ctx) + defer w.DecRef(ctx) n, err := r.Readv(ctx, usermem.BytesIOSequence(make([]byte, 1))) if n != 0 || err != syserror.ErrWouldBlock { @@ -62,8 +62,8 @@ func TestPipeWriteBlock(t *testing.T) { ctx := contexttest.Context(t) r, w := NewConnectedPipe(ctx, capacity, atomicIOBytes) - defer r.DecRef() - defer w.DecRef() + defer r.DecRef(ctx) + defer w.DecRef(ctx) msg := make([]byte, capacity+1) n, err := w.Writev(ctx, usermem.BytesIOSequence(msg)) @@ -77,8 +77,8 @@ func TestPipeWriteUntilEnd(t *testing.T) { ctx := contexttest.Context(t) r, w := NewConnectedPipe(ctx, atomicIOBytes, atomicIOBytes) - defer r.DecRef() - defer w.DecRef() + defer r.DecRef(ctx) + defer w.DecRef(ctx) msg := []byte("here's some bytes") diff --git a/pkg/sentry/kernel/pipe/pipe_util.go b/pkg/sentry/kernel/pipe/pipe_util.go index aacf28da2..6d58b682f 100644 --- a/pkg/sentry/kernel/pipe/pipe_util.go +++ b/pkg/sentry/kernel/pipe/pipe_util.go @@ -33,7 +33,7 @@ import ( // the old fs architecture. // Release cleans up the pipe's state. -func (p *Pipe) Release() { +func (p *Pipe) Release(context.Context) { p.rClose() p.wClose() diff --git a/pkg/sentry/kernel/pipe/reader.go b/pkg/sentry/kernel/pipe/reader.go index 7724b4452..ac18785c0 100644 --- a/pkg/sentry/kernel/pipe/reader.go +++ b/pkg/sentry/kernel/pipe/reader.go @@ -15,6 +15,7 @@ package pipe import ( + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/waiter" ) @@ -29,7 +30,7 @@ type Reader struct { // Release implements fs.FileOperations.Release. // // This overrides ReaderWriter.Release. -func (r *Reader) Release() { +func (r *Reader) Release(context.Context) { r.Pipe.rClose() // Wake up writers. diff --git a/pkg/sentry/kernel/pipe/vfs.go b/pkg/sentry/kernel/pipe/vfs.go index 45d4c5fc1..28f998e45 100644 --- a/pkg/sentry/kernel/pipe/vfs.go +++ b/pkg/sentry/kernel/pipe/vfs.go @@ -101,7 +101,7 @@ func (vp *VFSPipe) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, s // If this pipe is being opened as blocking and there's no // writer, we have to wait for a writer to open the other end. if vp.pipe.isNamed && statusFlags&linux.O_NONBLOCK == 0 && !vp.pipe.HasWriters() && !waitFor(&vp.mu, &vp.wWakeup, ctx) { - fd.DecRef() + fd.DecRef(ctx) return nil, syserror.EINTR } @@ -112,12 +112,12 @@ func (vp *VFSPipe) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, s // Non-blocking, write-only opens fail with ENXIO when the read // side isn't open yet. if statusFlags&linux.O_NONBLOCK != 0 { - fd.DecRef() + fd.DecRef(ctx) return nil, syserror.ENXIO } // Wait for a reader to open the other end. if !waitFor(&vp.mu, &vp.rWakeup, ctx) { - fd.DecRef() + fd.DecRef(ctx) return nil, syserror.EINTR } } @@ -169,7 +169,7 @@ type VFSPipeFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *VFSPipeFD) Release() { +func (fd *VFSPipeFD) Release(context.Context) { var event waiter.EventMask if fd.vfsfd.IsReadable() { fd.pipe.rClose() diff --git a/pkg/sentry/kernel/pipe/writer.go b/pkg/sentry/kernel/pipe/writer.go index 5bc6aa931..ef4b70ca3 100644 --- a/pkg/sentry/kernel/pipe/writer.go +++ b/pkg/sentry/kernel/pipe/writer.go @@ -15,6 +15,7 @@ package pipe import ( + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/waiter" ) @@ -29,7 +30,7 @@ type Writer struct { // Release implements fs.FileOperations.Release. // // This overrides ReaderWriter.Release. -func (w *Writer) Release() { +func (w *Writer) Release(context.Context) { w.Pipe.wClose() // Wake up readers. diff --git a/pkg/sentry/kernel/sessions.go b/pkg/sentry/kernel/sessions.go index 0e19286de..5c4c622c2 100644 --- a/pkg/sentry/kernel/sessions.go +++ b/pkg/sentry/kernel/sessions.go @@ -16,6 +16,7 @@ package kernel import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/syserror" @@ -70,7 +71,7 @@ func (s *Session) incRef() { // // Precondition: callers must hold TaskSet.mu for writing. func (s *Session) decRef() { - s.refs.DecRefWithDestructor(func() { + s.refs.DecRefWithDestructor(nil, func(context.Context) { // Remove translations from the leader. for ns := s.leader.pidns; ns != nil; ns = ns.parent { id := ns.sids[s] @@ -162,7 +163,7 @@ func (pg *ProcessGroup) decRefWithParent(parentPG *ProcessGroup) { } alive := true - pg.refs.DecRefWithDestructor(func() { + pg.refs.DecRefWithDestructor(nil, func(context.Context) { alive = false // don't bother with handleOrphan. // Remove translations from the originator. diff --git a/pkg/sentry/kernel/shm/shm.go b/pkg/sentry/kernel/shm/shm.go index 55b4c2cdb..13ec7afe0 100644 --- a/pkg/sentry/kernel/shm/shm.go +++ b/pkg/sentry/kernel/shm/shm.go @@ -431,8 +431,8 @@ func (s *Shm) InodeID() uint64 { // DecRef overrides refs.RefCount.DecRef with a destructor. // // Precondition: Caller must not hold s.mu. -func (s *Shm) DecRef() { - s.DecRefWithDestructor(s.destroy) +func (s *Shm) DecRef(ctx context.Context) { + s.DecRefWithDestructor(ctx, s.destroy) } // Msync implements memmap.MappingIdentity.Msync. Msync is a no-op for shm @@ -642,7 +642,7 @@ func (s *Shm) Set(ctx context.Context, ds *linux.ShmidDS) error { return nil } -func (s *Shm) destroy() { +func (s *Shm) destroy(context.Context) { s.mfp.MemoryFile().DecRef(s.fr) s.registry.remove(s) } @@ -651,7 +651,7 @@ func (s *Shm) destroy() { // destroyed once it has no references. MarkDestroyed may be called multiple // times, and is safe to call after a segment has already been destroyed. See // shmctl(IPC_RMID). -func (s *Shm) MarkDestroyed() { +func (s *Shm) MarkDestroyed(ctx context.Context) { s.registry.dissociateKey(s) s.mu.Lock() @@ -663,7 +663,7 @@ func (s *Shm) MarkDestroyed() { // // N.B. This cannot be the final DecRef, as the caller also // holds a reference. - s.DecRef() + s.DecRef(ctx) return } } diff --git a/pkg/sentry/kernel/signalfd/signalfd.go b/pkg/sentry/kernel/signalfd/signalfd.go index 8243bb93e..b07e1c1bd 100644 --- a/pkg/sentry/kernel/signalfd/signalfd.go +++ b/pkg/sentry/kernel/signalfd/signalfd.go @@ -76,7 +76,7 @@ func New(ctx context.Context, mask linux.SignalSet) (*fs.File, error) { } // Release implements fs.FileOperations.Release. -func (s *SignalOperations) Release() {} +func (s *SignalOperations) Release(context.Context) {} // Mask returns the signal mask. func (s *SignalOperations) Mask() linux.SignalSet { diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index c4db05bd8..5aee699e7 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -730,17 +730,17 @@ func (t *Task) SyscallRestartBlock() SyscallRestartBlock { func (t *Task) IsChrooted() bool { if VFS2Enabled { realRoot := t.mountNamespaceVFS2.Root() - defer realRoot.DecRef() + defer realRoot.DecRef(t) root := t.fsContext.RootDirectoryVFS2() - defer root.DecRef() + defer root.DecRef(t) return root != realRoot } realRoot := t.tg.mounts.Root() - defer realRoot.DecRef() + defer realRoot.DecRef(t) root := t.fsContext.RootDirectory() if root != nil { - defer root.DecRef() + defer root.DecRef(t) } return root != realRoot } diff --git a/pkg/sentry/kernel/task_clone.go b/pkg/sentry/kernel/task_clone.go index e1ecca99e..fe6ba6041 100644 --- a/pkg/sentry/kernel/task_clone.go +++ b/pkg/sentry/kernel/task_clone.go @@ -237,7 +237,7 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { var fdTable *FDTable if opts.NewFiles { - fdTable = t.fdTable.Fork() + fdTable = t.fdTable.Fork(t) } else { fdTable = t.fdTable fdTable.IncRef() @@ -294,7 +294,7 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { nt, err := t.tg.pidns.owner.NewTask(cfg) if err != nil { if opts.NewThreadGroup { - tg.release() + tg.release(t) } return 0, nil, err } @@ -510,7 +510,7 @@ func (t *Task) Unshare(opts *SharingOptions) error { var oldFDTable *FDTable if opts.NewFiles { oldFDTable = t.fdTable - t.fdTable = oldFDTable.Fork() + t.fdTable = oldFDTable.Fork(t) } var oldFSContext *FSContext if opts.NewFSContext { @@ -519,10 +519,10 @@ func (t *Task) Unshare(opts *SharingOptions) error { } t.mu.Unlock() if oldFDTable != nil { - oldFDTable.DecRef() + oldFDTable.DecRef(t) } if oldFSContext != nil { - oldFSContext.DecRef() + oldFSContext.DecRef(t) } return nil } diff --git a/pkg/sentry/kernel/task_exec.go b/pkg/sentry/kernel/task_exec.go index 7803b98d0..47c28b8ff 100644 --- a/pkg/sentry/kernel/task_exec.go +++ b/pkg/sentry/kernel/task_exec.go @@ -199,11 +199,11 @@ func (r *runSyscallAfterExecStop) execute(t *Task) taskRunState { t.tg.pidns.owner.mu.Unlock() oldFDTable := t.fdTable - t.fdTable = t.fdTable.Fork() - oldFDTable.DecRef() + t.fdTable = t.fdTable.Fork(t) + oldFDTable.DecRef(t) // Remove FDs with the CloseOnExec flag set. - t.fdTable.RemoveIf(func(_ *fs.File, _ *vfs.FileDescription, flags FDFlags) bool { + t.fdTable.RemoveIf(t, func(_ *fs.File, _ *vfs.FileDescription, flags FDFlags) bool { return flags.CloseOnExec }) diff --git a/pkg/sentry/kernel/task_exit.go b/pkg/sentry/kernel/task_exit.go index 231ac548a..c165d6cb1 100644 --- a/pkg/sentry/kernel/task_exit.go +++ b/pkg/sentry/kernel/task_exit.go @@ -269,12 +269,12 @@ func (*runExitMain) execute(t *Task) taskRunState { // Releasing the MM unblocks a blocked CLONE_VFORK parent. t.unstopVforkParent() - t.fsContext.DecRef() - t.fdTable.DecRef() + t.fsContext.DecRef(t) + t.fdTable.DecRef(t) t.mu.Lock() if t.mountNamespaceVFS2 != nil { - t.mountNamespaceVFS2.DecRef() + t.mountNamespaceVFS2.DecRef(t) t.mountNamespaceVFS2 = nil } t.mu.Unlock() @@ -282,7 +282,7 @@ func (*runExitMain) execute(t *Task) taskRunState { // If this is the last task to exit from the thread group, release the // thread group's resources. if lastExiter { - t.tg.release() + t.tg.release(t) } // Detach tracees. diff --git a/pkg/sentry/kernel/task_log.go b/pkg/sentry/kernel/task_log.go index eeccaa197..ab86ceedc 100644 --- a/pkg/sentry/kernel/task_log.go +++ b/pkg/sentry/kernel/task_log.go @@ -203,6 +203,6 @@ func (t *Task) traceExecEvent(tc *TaskContext) { trace.Logf(t.traceContext, traceCategory, "exec: << unknown >>") return } - defer file.DecRef() + defer file.DecRef(t) trace.Logf(t.traceContext, traceCategory, "exec: %s", file.PathnameWithDeleted(t)) } diff --git a/pkg/sentry/kernel/task_start.go b/pkg/sentry/kernel/task_start.go index 8485fb4b6..64c1e120a 100644 --- a/pkg/sentry/kernel/task_start.go +++ b/pkg/sentry/kernel/task_start.go @@ -102,10 +102,10 @@ func (ts *TaskSet) NewTask(cfg *TaskConfig) (*Task, error) { t, err := ts.newTask(cfg) if err != nil { cfg.TaskContext.release() - cfg.FSContext.DecRef() - cfg.FDTable.DecRef() + cfg.FSContext.DecRef(t) + cfg.FDTable.DecRef(t) if cfg.MountNamespaceVFS2 != nil { - cfg.MountNamespaceVFS2.DecRef() + cfg.MountNamespaceVFS2.DecRef(t) } return nil, err } diff --git a/pkg/sentry/kernel/thread_group.go b/pkg/sentry/kernel/thread_group.go index 4dfd2c990..0b34c0099 100644 --- a/pkg/sentry/kernel/thread_group.go +++ b/pkg/sentry/kernel/thread_group.go @@ -308,7 +308,7 @@ func (tg *ThreadGroup) Limits() *limits.LimitSet { } // release releases the thread group's resources. -func (tg *ThreadGroup) release() { +func (tg *ThreadGroup) release(t *Task) { // Timers must be destroyed without holding the TaskSet or signal mutexes // since timers send signals with Timer.mu locked. tg.itimerRealTimer.Destroy() @@ -325,7 +325,7 @@ func (tg *ThreadGroup) release() { it.DestroyTimer() } if tg.mounts != nil { - tg.mounts.DecRef() + tg.mounts.DecRef(t) } } diff --git a/pkg/sentry/loader/elf.go b/pkg/sentry/loader/elf.go index ddeaff3db..20dd1cc21 100644 --- a/pkg/sentry/loader/elf.go +++ b/pkg/sentry/loader/elf.go @@ -281,7 +281,7 @@ func mapSegment(ctx context.Context, m *mm.MemoryManager, f fsbridge.File, phdr } defer func() { if mopts.MappingIdentity != nil { - mopts.MappingIdentity.DecRef() + mopts.MappingIdentity.DecRef(ctx) } }() if err := f.ConfigureMMap(ctx, &mopts); err != nil { @@ -663,7 +663,7 @@ func loadELF(ctx context.Context, args LoadArgs) (loadedELF, arch.Context, error ctx.Infof("Error opening interpreter %s: %v", bin.interpreter, err) return loadedELF{}, nil, err } - defer intFile.DecRef() + defer intFile.DecRef(ctx) interp, err = loadInterpreterELF(ctx, args.MemoryManager, intFile, bin) if err != nil { diff --git a/pkg/sentry/loader/loader.go b/pkg/sentry/loader/loader.go index 986c7fb4d..8d6802ea3 100644 --- a/pkg/sentry/loader/loader.go +++ b/pkg/sentry/loader/loader.go @@ -154,7 +154,7 @@ func loadExecutable(ctx context.Context, args LoadArgs) (loadedELF, arch.Context return loadedELF{}, nil, nil, nil, err } // Ensure file is release in case the code loops or errors out. - defer args.File.DecRef() + defer args.File.DecRef(ctx) } else { if err := checkIsRegularFile(ctx, args.File, args.Filename); err != nil { return loadedELF{}, nil, nil, nil, err @@ -223,7 +223,7 @@ func Load(ctx context.Context, args LoadArgs, extraAuxv []arch.AuxEntry, vdso *V if err != nil { return 0, nil, "", syserr.NewDynamic(fmt.Sprintf("failed to load %s: %v", args.Filename, err), syserr.FromError(err).ToLinux()) } - defer file.DecRef() + defer file.DecRef(ctx) // Load the VDSO. vdsoAddr, err := loadVDSO(ctx, args.MemoryManager, vdso, loaded) @@ -292,7 +292,7 @@ func Load(ctx context.Context, args LoadArgs, extraAuxv []arch.AuxEntry, vdso *V m.SetEnvvStart(sl.EnvvStart) m.SetEnvvEnd(sl.EnvvEnd) m.SetAuxv(auxv) - m.SetExecutable(file) + m.SetExecutable(ctx, file) symbolValue, err := getSymbolValueFromVDSO("rt_sigreturn") if err != nil { diff --git a/pkg/sentry/memmap/memmap.go b/pkg/sentry/memmap/memmap.go index c188f6c29..59c92c7e8 100644 --- a/pkg/sentry/memmap/memmap.go +++ b/pkg/sentry/memmap/memmap.go @@ -238,7 +238,7 @@ type MappingIdentity interface { IncRef() // DecRef decrements the MappingIdentity's reference count. - DecRef() + DecRef(ctx context.Context) // MappedName returns the application-visible name shown in // /proc/[pid]/maps. diff --git a/pkg/sentry/mm/aio_context.go b/pkg/sentry/mm/aio_context.go index 1999ec706..16fea53c4 100644 --- a/pkg/sentry/mm/aio_context.go +++ b/pkg/sentry/mm/aio_context.go @@ -258,8 +258,8 @@ func newAIOMappable(mfp pgalloc.MemoryFileProvider) (*aioMappable, error) { } // DecRef implements refs.RefCounter.DecRef. -func (m *aioMappable) DecRef() { - m.AtomicRefCount.DecRefWithDestructor(func() { +func (m *aioMappable) DecRef(ctx context.Context) { + m.AtomicRefCount.DecRefWithDestructor(ctx, func(context.Context) { m.mfp.MemoryFile().DecRef(m.fr) }) } @@ -367,7 +367,7 @@ func (mm *MemoryManager) NewAIOContext(ctx context.Context, events uint32) (uint if err != nil { return 0, err } - defer m.DecRef() + defer m.DecRef(ctx) addr, err := mm.MMap(ctx, memmap.MMapOpts{ Length: aioRingBufferSize, MappingIdentity: m, diff --git a/pkg/sentry/mm/lifecycle.go b/pkg/sentry/mm/lifecycle.go index aac56679b..4d7773f8b 100644 --- a/pkg/sentry/mm/lifecycle.go +++ b/pkg/sentry/mm/lifecycle.go @@ -258,7 +258,7 @@ func (mm *MemoryManager) DecUsers(ctx context.Context) { mm.executable = nil mm.metadataMu.Unlock() if exe != nil { - exe.DecRef() + exe.DecRef(ctx) } mm.activeMu.Lock() diff --git a/pkg/sentry/mm/metadata.go b/pkg/sentry/mm/metadata.go index 28e5057f7..0cfd60f6c 100644 --- a/pkg/sentry/mm/metadata.go +++ b/pkg/sentry/mm/metadata.go @@ -15,6 +15,7 @@ package mm import ( + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fsbridge" "gvisor.dev/gvisor/pkg/usermem" @@ -147,7 +148,7 @@ func (mm *MemoryManager) Executable() fsbridge.File { // SetExecutable sets the executable. // // This takes a reference on d. -func (mm *MemoryManager) SetExecutable(file fsbridge.File) { +func (mm *MemoryManager) SetExecutable(ctx context.Context, file fsbridge.File) { mm.metadataMu.Lock() // Grab a new reference. @@ -164,7 +165,7 @@ func (mm *MemoryManager) SetExecutable(file fsbridge.File) { // Do this without holding the lock, since it may wind up doing some // I/O to sync the dirent, etc. if orig != nil { - orig.DecRef() + orig.DecRef(ctx) } } diff --git a/pkg/sentry/mm/special_mappable.go b/pkg/sentry/mm/special_mappable.go index 0e142fb11..4cdb52eb6 100644 --- a/pkg/sentry/mm/special_mappable.go +++ b/pkg/sentry/mm/special_mappable.go @@ -50,8 +50,8 @@ func NewSpecialMappable(name string, mfp pgalloc.MemoryFileProvider, fr memmap.F } // DecRef implements refs.RefCounter.DecRef. -func (m *SpecialMappable) DecRef() { - m.AtomicRefCount.DecRefWithDestructor(func() { +func (m *SpecialMappable) DecRef(ctx context.Context) { + m.AtomicRefCount.DecRefWithDestructor(ctx, func(context.Context) { m.mfp.MemoryFile().DecRef(m.fr) }) } diff --git a/pkg/sentry/mm/syscalls.go b/pkg/sentry/mm/syscalls.go index 3f496aa9f..e74d4e1c1 100644 --- a/pkg/sentry/mm/syscalls.go +++ b/pkg/sentry/mm/syscalls.go @@ -101,7 +101,7 @@ func (mm *MemoryManager) MMap(ctx context.Context, opts memmap.MMapOpts) (userme if err != nil { return 0, err } - defer m.DecRef() + defer m.DecRef(ctx) opts.MappingIdentity = m opts.Mappable = m } @@ -1191,7 +1191,7 @@ func (mm *MemoryManager) MSync(ctx context.Context, addr usermem.Addr, length ui mr := vseg.mappableRangeOf(vseg.Range().Intersect(ar)) mm.mappingMu.RUnlock() err := id.Msync(ctx, mr) - id.DecRef() + id.DecRef(ctx) if err != nil { return err } diff --git a/pkg/sentry/mm/vma.go b/pkg/sentry/mm/vma.go index 16d8207e9..bd751d696 100644 --- a/pkg/sentry/mm/vma.go +++ b/pkg/sentry/mm/vma.go @@ -377,7 +377,7 @@ func (mm *MemoryManager) removeVMAsLocked(ctx context.Context, ar usermem.AddrRa vma.mappable.RemoveMapping(ctx, mm, vmaAR, vma.off, vma.canWriteMappableLocked()) } if vma.id != nil { - vma.id.DecRef() + vma.id.DecRef(ctx) } mm.usageAS -= uint64(vmaAR.Length()) if vma.isPrivateDataLocked() { @@ -446,7 +446,7 @@ func (vmaSetFunctions) Merge(ar1 usermem.AddrRange, vma1 vma, ar2 usermem.AddrRa } if vma2.id != nil { - vma2.id.DecRef() + vma2.id.DecRef(context.Background()) } return vma1, true } diff --git a/pkg/sentry/socket/control/control.go b/pkg/sentry/socket/control/control.go index 8b439a078..70ccf77a7 100644 --- a/pkg/sentry/socket/control/control.go +++ b/pkg/sentry/socket/control/control.go @@ -68,7 +68,7 @@ func NewSCMRights(t *kernel.Task, fds []int32) (SCMRights, error) { for _, fd := range fds { file := t.GetFile(fd) if file == nil { - files.Release() + files.Release(t) return nil, syserror.EBADF } files = append(files, file) @@ -100,9 +100,9 @@ func (fs *RightsFiles) Clone() transport.RightsControlMessage { } // Release implements transport.RightsControlMessage.Release. -func (fs *RightsFiles) Release() { +func (fs *RightsFiles) Release(ctx context.Context) { for _, f := range *fs { - f.DecRef() + f.DecRef(ctx) } *fs = nil } @@ -115,7 +115,7 @@ func rightsFDs(t *kernel.Task, rights SCMRights, cloexec bool, max int) ([]int32 fd, err := t.NewFDFrom(0, files[0], kernel.FDFlags{ CloseOnExec: cloexec, }) - files[0].DecRef() + files[0].DecRef(t) files = files[1:] if err != nil { t.Warningf("Error inserting FD: %v", err) diff --git a/pkg/sentry/socket/control/control_vfs2.go b/pkg/sentry/socket/control/control_vfs2.go index fd08179be..d9621968c 100644 --- a/pkg/sentry/socket/control/control_vfs2.go +++ b/pkg/sentry/socket/control/control_vfs2.go @@ -46,7 +46,7 @@ func NewSCMRightsVFS2(t *kernel.Task, fds []int32) (SCMRightsVFS2, error) { for _, fd := range fds { file := t.GetFileVFS2(fd) if file == nil { - files.Release() + files.Release(t) return nil, syserror.EBADF } files = append(files, file) @@ -78,9 +78,9 @@ func (fs *RightsFilesVFS2) Clone() transport.RightsControlMessage { } // Release implements transport.RightsControlMessage.Release. -func (fs *RightsFilesVFS2) Release() { +func (fs *RightsFilesVFS2) Release(ctx context.Context) { for _, f := range *fs { - f.DecRef() + f.DecRef(ctx) } *fs = nil } @@ -93,7 +93,7 @@ func rightsFDsVFS2(t *kernel.Task, rights SCMRightsVFS2, cloexec bool, max int) fd, err := t.NewFDFromVFS2(0, files[0], kernel.FDFlags{ CloseOnExec: cloexec, }) - files[0].DecRef() + files[0].DecRef(t) files = files[1:] if err != nil { t.Warningf("Error inserting FD: %v", err) diff --git a/pkg/sentry/socket/hostinet/socket.go b/pkg/sentry/socket/hostinet/socket.go index 532a1ea5d..242e6bf76 100644 --- a/pkg/sentry/socket/hostinet/socket.go +++ b/pkg/sentry/socket/hostinet/socket.go @@ -100,12 +100,12 @@ func newSocketFile(ctx context.Context, family int, stype linux.SockType, protoc return nil, syserr.FromError(err) } dirent := socket.NewDirent(ctx, socketDevice) - defer dirent.DecRef() + defer dirent.DecRef(ctx) return fs.NewFile(ctx, dirent, fs.FileFlags{NonBlocking: nonblock, Read: true, Write: true, NonSeekable: true}, s), nil } // Release implements fs.FileOperations.Release. -func (s *socketOpsCommon) Release() { +func (s *socketOpsCommon) Release(context.Context) { fdnotifier.RemoveFD(int32(s.fd)) syscall.Close(s.fd) } @@ -269,7 +269,7 @@ func (s *socketOpsCommon) Accept(t *kernel.Task, peerRequested bool, flags int, syscall.Close(fd) return 0, nil, 0, err } - defer f.DecRef() + defer f.DecRef(t) kfd, kerr = t.NewFDFromVFS2(0, f, kernel.FDFlags{ CloseOnExec: flags&syscall.SOCK_CLOEXEC != 0, @@ -281,7 +281,7 @@ func (s *socketOpsCommon) Accept(t *kernel.Task, peerRequested bool, flags int, syscall.Close(fd) return 0, nil, 0, err } - defer f.DecRef() + defer f.DecRef(t) kfd, kerr = t.NewFDFrom(0, f, kernel.FDFlags{ CloseOnExec: flags&syscall.SOCK_CLOEXEC != 0, diff --git a/pkg/sentry/socket/netlink/provider.go b/pkg/sentry/socket/netlink/provider.go index 0d45e5053..31e374833 100644 --- a/pkg/sentry/socket/netlink/provider.go +++ b/pkg/sentry/socket/netlink/provider.go @@ -97,7 +97,7 @@ func (*socketProvider) Socket(t *kernel.Task, stype linux.SockType, protocol int } d := socket.NewDirent(t, netlinkSocketDevice) - defer d.DecRef() + defer d.DecRef(t) return fs.NewFile(t, d, fs.FileFlags{Read: true, Write: true, NonSeekable: true}, s), nil } diff --git a/pkg/sentry/socket/netlink/socket.go b/pkg/sentry/socket/netlink/socket.go index 98ca7add0..68a9b9a96 100644 --- a/pkg/sentry/socket/netlink/socket.go +++ b/pkg/sentry/socket/netlink/socket.go @@ -140,14 +140,14 @@ func NewSocket(t *kernel.Task, skType linux.SockType, protocol Protocol) (*Socke // Bind the endpoint for good measure so we can connect to it. The // bound address will never be exposed. if err := ep.Bind(tcpip.FullAddress{Addr: "dummy"}, nil); err != nil { - ep.Close() + ep.Close(t) return nil, err } // Create a connection from which the kernel can write messages. connection, err := ep.(transport.BoundEndpoint).UnidirectionalConnect(t) if err != nil { - ep.Close() + ep.Close(t) return nil, err } @@ -164,9 +164,9 @@ func NewSocket(t *kernel.Task, skType linux.SockType, protocol Protocol) (*Socke } // Release implements fs.FileOperations.Release. -func (s *socketOpsCommon) Release() { - s.connection.Release() - s.ep.Close() +func (s *socketOpsCommon) Release(ctx context.Context) { + s.connection.Release(ctx) + s.ep.Close(ctx) if s.bound { s.ports.Release(s.protocol.Protocol(), s.portID) @@ -621,7 +621,7 @@ func (s *socketOpsCommon) sendResponse(ctx context.Context, ms *MessageSet) *sys if len(bufs) > 0 { // RecvMsg never receives the address, so we don't need to send // one. - _, notify, err := s.connection.Send(bufs, cms, tcpip.FullAddress{}) + _, notify, err := s.connection.Send(ctx, bufs, cms, tcpip.FullAddress{}) // If the buffer is full, we simply drop messages, just like // Linux. if err != nil && err != syserr.ErrWouldBlock { @@ -648,7 +648,7 @@ func (s *socketOpsCommon) sendResponse(ctx context.Context, ms *MessageSet) *sys // Add the dump_done_errno payload. m.Put(int64(0)) - _, notify, err := s.connection.Send([][]byte{m.Finalize()}, cms, tcpip.FullAddress{}) + _, notify, err := s.connection.Send(ctx, [][]byte{m.Finalize()}, cms, tcpip.FullAddress{}) if err != nil && err != syserr.ErrWouldBlock { return err } diff --git a/pkg/sentry/socket/netlink/socket_vfs2.go b/pkg/sentry/socket/netlink/socket_vfs2.go index dbcd8b49a..a38d25da9 100644 --- a/pkg/sentry/socket/netlink/socket_vfs2.go +++ b/pkg/sentry/socket/netlink/socket_vfs2.go @@ -57,14 +57,14 @@ func NewVFS2(t *kernel.Task, skType linux.SockType, protocol Protocol) (*SocketV // Bind the endpoint for good measure so we can connect to it. The // bound address will never be exposed. if err := ep.Bind(tcpip.FullAddress{Addr: "dummy"}, nil); err != nil { - ep.Close() + ep.Close(t) return nil, err } // Create a connection from which the kernel can write messages. connection, err := ep.(transport.BoundEndpoint).UnidirectionalConnect(t) if err != nil { - ep.Close() + ep.Close(t) return nil, err } diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 31a168f7e..e4846bc0b 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -330,7 +330,7 @@ func New(t *kernel.Task, family int, skType linux.SockType, protocol int, queue } dirent := socket.NewDirent(t, netstackDevice) - defer dirent.DecRef() + defer dirent.DecRef(t) return fs.NewFile(t, dirent, fs.FileFlags{Read: true, Write: true, NonSeekable: true}, &SocketOperations{ socketOpsCommon: socketOpsCommon{ Queue: queue, @@ -479,7 +479,7 @@ func (s *socketOpsCommon) fetchReadView() *syserr.Error { } // Release implements fs.FileOperations.Release. -func (s *socketOpsCommon) Release() { +func (s *socketOpsCommon) Release(context.Context) { s.Endpoint.Close() } @@ -854,7 +854,7 @@ func (s *SocketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, if err != nil { return 0, nil, 0, err } - defer ns.DecRef() + defer ns.DecRef(t) if flags&linux.SOCK_NONBLOCK != 0 { flags := ns.Flags() diff --git a/pkg/sentry/socket/netstack/netstack_vfs2.go b/pkg/sentry/socket/netstack/netstack_vfs2.go index a9025b0ec..3335e7430 100644 --- a/pkg/sentry/socket/netstack/netstack_vfs2.go +++ b/pkg/sentry/socket/netstack/netstack_vfs2.go @@ -169,7 +169,7 @@ func (s *SocketVFS2) Accept(t *kernel.Task, peerRequested bool, flags int, block if err != nil { return 0, nil, 0, err } - defer ns.DecRef() + defer ns.DecRef(t) if err := ns.SetStatusFlags(t, t.Credentials(), uint32(flags&linux.SOCK_NONBLOCK)); err != nil { return 0, nil, 0, syserr.FromError(err) diff --git a/pkg/sentry/socket/socket.go b/pkg/sentry/socket/socket.go index d112757fb..04b259d27 100644 --- a/pkg/sentry/socket/socket.go +++ b/pkg/sentry/socket/socket.go @@ -46,8 +46,8 @@ type ControlMessages struct { } // Release releases Unix domain socket credentials and rights. -func (c *ControlMessages) Release() { - c.Unix.Release() +func (c *ControlMessages) Release(ctx context.Context) { + c.Unix.Release(ctx) } // Socket is an interface combining fs.FileOperations and SocketOps, diff --git a/pkg/sentry/socket/unix/transport/connectioned.go b/pkg/sentry/socket/unix/transport/connectioned.go index a1e49cc57..c67b602f0 100644 --- a/pkg/sentry/socket/unix/transport/connectioned.go +++ b/pkg/sentry/socket/unix/transport/connectioned.go @@ -211,7 +211,7 @@ func (e *connectionedEndpoint) Listening() bool { // The socket will be a fresh state after a call to close and may be reused. // That is, close may be used to "unbind" or "disconnect" the socket in error // paths. -func (e *connectionedEndpoint) Close() { +func (e *connectionedEndpoint) Close(ctx context.Context) { e.Lock() var c ConnectedEndpoint var r Receiver @@ -233,7 +233,7 @@ func (e *connectionedEndpoint) Close() { case e.Listening(): close(e.acceptedChan) for n := range e.acceptedChan { - n.Close() + n.Close(ctx) } e.acceptedChan = nil e.path = "" @@ -241,11 +241,11 @@ func (e *connectionedEndpoint) Close() { e.Unlock() if c != nil { c.CloseNotify() - c.Release() + c.Release(ctx) } if r != nil { r.CloseNotify() - r.Release() + r.Release(ctx) } } @@ -340,7 +340,7 @@ func (e *connectionedEndpoint) BidirectionalConnect(ctx context.Context, ce Conn return nil default: // Busy; return ECONNREFUSED per spec. - ne.Close() + ne.Close(ctx) e.Unlock() ce.Unlock() return syserr.ErrConnectionRefused diff --git a/pkg/sentry/socket/unix/transport/connectionless.go b/pkg/sentry/socket/unix/transport/connectionless.go index 4b06d63ac..70ee8f9b8 100644 --- a/pkg/sentry/socket/unix/transport/connectionless.go +++ b/pkg/sentry/socket/unix/transport/connectionless.go @@ -54,10 +54,10 @@ func (e *connectionlessEndpoint) isBound() bool { // Close puts the endpoint in a closed state and frees all resources associated // with it. -func (e *connectionlessEndpoint) Close() { +func (e *connectionlessEndpoint) Close(ctx context.Context) { e.Lock() if e.connected != nil { - e.connected.Release() + e.connected.Release(ctx) e.connected = nil } @@ -71,7 +71,7 @@ func (e *connectionlessEndpoint) Close() { e.Unlock() r.CloseNotify() - r.Release() + r.Release(ctx) } // BidirectionalConnect implements BoundEndpoint.BidirectionalConnect. @@ -108,10 +108,10 @@ func (e *connectionlessEndpoint) SendMsg(ctx context.Context, data [][]byte, c C if err != nil { return 0, syserr.ErrInvalidEndpointState } - defer connected.Release() + defer connected.Release(ctx) e.Lock() - n, notify, err := connected.Send(data, c, tcpip.FullAddress{Addr: tcpip.Address(e.path)}) + n, notify, err := connected.Send(ctx, data, c, tcpip.FullAddress{Addr: tcpip.Address(e.path)}) e.Unlock() if notify { @@ -135,7 +135,7 @@ func (e *connectionlessEndpoint) Connect(ctx context.Context, server BoundEndpoi e.Lock() if e.connected != nil { - e.connected.Release() + e.connected.Release(ctx) } e.connected = connected e.Unlock() diff --git a/pkg/sentry/socket/unix/transport/queue.go b/pkg/sentry/socket/unix/transport/queue.go index d8f3ad63d..ef6043e19 100644 --- a/pkg/sentry/socket/unix/transport/queue.go +++ b/pkg/sentry/socket/unix/transport/queue.go @@ -15,6 +15,7 @@ package transport import ( + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserr" @@ -57,10 +58,10 @@ func (q *queue) Close() { // Both the read and write queues must be notified after resetting: // q.ReaderQueue.Notify(waiter.EventIn) // q.WriterQueue.Notify(waiter.EventOut) -func (q *queue) Reset() { +func (q *queue) Reset(ctx context.Context) { q.mu.Lock() for cur := q.dataList.Front(); cur != nil; cur = cur.Next() { - cur.Release() + cur.Release(ctx) } q.dataList.Reset() q.used = 0 @@ -68,8 +69,8 @@ func (q *queue) Reset() { } // DecRef implements RefCounter.DecRef with destructor q.Reset. -func (q *queue) DecRef() { - q.DecRefWithDestructor(q.Reset) +func (q *queue) DecRef(ctx context.Context) { + q.DecRefWithDestructor(ctx, q.Reset) // We don't need to notify after resetting because no one cares about // this queue after all references have been dropped. } @@ -111,7 +112,7 @@ func (q *queue) IsWritable() bool { // // If notify is true, ReaderQueue.Notify must be called: // q.ReaderQueue.Notify(waiter.EventIn) -func (q *queue) Enqueue(data [][]byte, c ControlMessages, from tcpip.FullAddress, discardEmpty bool, truncate bool) (l int64, notify bool, err *syserr.Error) { +func (q *queue) Enqueue(ctx context.Context, data [][]byte, c ControlMessages, from tcpip.FullAddress, discardEmpty bool, truncate bool) (l int64, notify bool, err *syserr.Error) { q.mu.Lock() if q.closed { @@ -124,7 +125,7 @@ func (q *queue) Enqueue(data [][]byte, c ControlMessages, from tcpip.FullAddress } if discardEmpty && l == 0 { q.mu.Unlock() - c.Release() + c.Release(ctx) return 0, false, nil } diff --git a/pkg/sentry/socket/unix/transport/unix.go b/pkg/sentry/socket/unix/transport/unix.go index 2f1b127df..475d7177e 100644 --- a/pkg/sentry/socket/unix/transport/unix.go +++ b/pkg/sentry/socket/unix/transport/unix.go @@ -37,7 +37,7 @@ type RightsControlMessage interface { Clone() RightsControlMessage // Release releases any resources owned by the RightsControlMessage. - Release() + Release(ctx context.Context) } // A CredentialsControlMessage is a control message containing Unix credentials. @@ -74,9 +74,9 @@ func (c *ControlMessages) Clone() ControlMessages { } // Release releases both the credentials and the rights. -func (c *ControlMessages) Release() { +func (c *ControlMessages) Release(ctx context.Context) { if c.Rights != nil { - c.Rights.Release() + c.Rights.Release(ctx) } *c = ControlMessages{} } @@ -90,7 +90,7 @@ type Endpoint interface { // Close puts the endpoint in a closed state and frees all resources // associated with it. - Close() + Close(ctx context.Context) // RecvMsg reads data and a control message from the endpoint. This method // does not block if there is no data pending. @@ -252,7 +252,7 @@ type BoundEndpoint interface { // Release releases any resources held by the BoundEndpoint. It must be // called before dropping all references to a BoundEndpoint returned by a // function. - Release() + Release(ctx context.Context) } // message represents a message passed over a Unix domain socket. @@ -281,8 +281,8 @@ func (m *message) Length() int64 { } // Release releases any resources held by the message. -func (m *message) Release() { - m.Control.Release() +func (m *message) Release(ctx context.Context) { + m.Control.Release(ctx) } // Peek returns a copy of the message. @@ -304,7 +304,7 @@ type Receiver interface { // See Endpoint.RecvMsg for documentation on shared arguments. // // notify indicates if RecvNotify should be called. - Recv(data [][]byte, creds bool, numRights int, peek bool) (recvLen, msgLen int64, cm ControlMessages, CMTruncated bool, source tcpip.FullAddress, notify bool, err *syserr.Error) + Recv(ctx context.Context, data [][]byte, creds bool, numRights int, peek bool) (recvLen, msgLen int64, cm ControlMessages, CMTruncated bool, source tcpip.FullAddress, notify bool, err *syserr.Error) // RecvNotify notifies the Receiver of a successful Recv. This must not be // called while holding any endpoint locks. @@ -333,7 +333,7 @@ type Receiver interface { // Release releases any resources owned by the Receiver. It should be // called before droping all references to a Receiver. - Release() + Release(ctx context.Context) } // queueReceiver implements Receiver for datagram sockets. @@ -344,7 +344,7 @@ type queueReceiver struct { } // Recv implements Receiver.Recv. -func (q *queueReceiver) Recv(data [][]byte, creds bool, numRights int, peek bool) (int64, int64, ControlMessages, bool, tcpip.FullAddress, bool, *syserr.Error) { +func (q *queueReceiver) Recv(ctx context.Context, data [][]byte, creds bool, numRights int, peek bool) (int64, int64, ControlMessages, bool, tcpip.FullAddress, bool, *syserr.Error) { var m *message var notify bool var err *syserr.Error @@ -398,8 +398,8 @@ func (q *queueReceiver) RecvMaxQueueSize() int64 { } // Release implements Receiver.Release. -func (q *queueReceiver) Release() { - q.readQueue.DecRef() +func (q *queueReceiver) Release(ctx context.Context) { + q.readQueue.DecRef(ctx) } // streamQueueReceiver implements Receiver for stream sockets. @@ -456,7 +456,7 @@ func (q *streamQueueReceiver) RecvMaxQueueSize() int64 { } // Recv implements Receiver.Recv. -func (q *streamQueueReceiver) Recv(data [][]byte, wantCreds bool, numRights int, peek bool) (int64, int64, ControlMessages, bool, tcpip.FullAddress, bool, *syserr.Error) { +func (q *streamQueueReceiver) Recv(ctx context.Context, data [][]byte, wantCreds bool, numRights int, peek bool) (int64, int64, ControlMessages, bool, tcpip.FullAddress, bool, *syserr.Error) { q.mu.Lock() defer q.mu.Unlock() @@ -502,7 +502,7 @@ func (q *streamQueueReceiver) Recv(data [][]byte, wantCreds bool, numRights int, var cmTruncated bool if c.Rights != nil && numRights == 0 { - c.Rights.Release() + c.Rights.Release(ctx) c.Rights = nil cmTruncated = true } @@ -557,7 +557,7 @@ func (q *streamQueueReceiver) Recv(data [][]byte, wantCreds bool, numRights int, // Consume rights. if numRights == 0 { cmTruncated = true - q.control.Rights.Release() + q.control.Rights.Release(ctx) } else { c.Rights = q.control.Rights haveRights = true @@ -582,7 +582,7 @@ type ConnectedEndpoint interface { // // syserr.ErrWouldBlock can be returned along with a partial write if // the caller should block to send the rest of the data. - Send(data [][]byte, c ControlMessages, from tcpip.FullAddress) (n int64, notify bool, err *syserr.Error) + Send(ctx context.Context, data [][]byte, c ControlMessages, from tcpip.FullAddress) (n int64, notify bool, err *syserr.Error) // SendNotify notifies the ConnectedEndpoint of a successful Send. This // must not be called while holding any endpoint locks. @@ -616,7 +616,7 @@ type ConnectedEndpoint interface { // Release releases any resources owned by the ConnectedEndpoint. It should // be called before droping all references to a ConnectedEndpoint. - Release() + Release(ctx context.Context) // CloseUnread sets the fact that this end is closed with unread data to // the peer socket. @@ -654,7 +654,7 @@ func (e *connectedEndpoint) GetLocalAddress() (tcpip.FullAddress, *tcpip.Error) } // Send implements ConnectedEndpoint.Send. -func (e *connectedEndpoint) Send(data [][]byte, c ControlMessages, from tcpip.FullAddress) (int64, bool, *syserr.Error) { +func (e *connectedEndpoint) Send(ctx context.Context, data [][]byte, c ControlMessages, from tcpip.FullAddress) (int64, bool, *syserr.Error) { discardEmpty := false truncate := false if e.endpoint.Type() == linux.SOCK_STREAM { @@ -669,7 +669,7 @@ func (e *connectedEndpoint) Send(data [][]byte, c ControlMessages, from tcpip.Fu truncate = true } - return e.writeQueue.Enqueue(data, c, from, discardEmpty, truncate) + return e.writeQueue.Enqueue(ctx, data, c, from, discardEmpty, truncate) } // SendNotify implements ConnectedEndpoint.SendNotify. @@ -707,8 +707,8 @@ func (e *connectedEndpoint) SendMaxQueueSize() int64 { } // Release implements ConnectedEndpoint.Release. -func (e *connectedEndpoint) Release() { - e.writeQueue.DecRef() +func (e *connectedEndpoint) Release(ctx context.Context) { + e.writeQueue.DecRef(ctx) } // CloseUnread implements ConnectedEndpoint.CloseUnread. @@ -798,7 +798,7 @@ func (e *baseEndpoint) RecvMsg(ctx context.Context, data [][]byte, creds bool, n return 0, 0, ControlMessages{}, false, syserr.ErrNotConnected } - recvLen, msgLen, cms, cmt, a, notify, err := e.receiver.Recv(data, creds, numRights, peek) + recvLen, msgLen, cms, cmt, a, notify, err := e.receiver.Recv(ctx, data, creds, numRights, peek) e.Unlock() if err != nil { return 0, 0, ControlMessages{}, false, err @@ -827,7 +827,7 @@ func (e *baseEndpoint) SendMsg(ctx context.Context, data [][]byte, c ControlMess return 0, syserr.ErrAlreadyConnected } - n, notify, err := e.connected.Send(data, c, tcpip.FullAddress{Addr: tcpip.Address(e.path)}) + n, notify, err := e.connected.Send(ctx, data, c, tcpip.FullAddress{Addr: tcpip.Address(e.path)}) e.Unlock() if notify { @@ -1001,6 +1001,6 @@ func (e *baseEndpoint) GetRemoteAddress() (tcpip.FullAddress, *tcpip.Error) { } // Release implements BoundEndpoint.Release. -func (*baseEndpoint) Release() { +func (*baseEndpoint) Release(context.Context) { // Binding a baseEndpoint doesn't take a reference. } diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go index 0482d33cf..2b8454edb 100644 --- a/pkg/sentry/socket/unix/unix.go +++ b/pkg/sentry/socket/unix/unix.go @@ -62,7 +62,7 @@ type SocketOperations struct { // New creates a new unix socket. func New(ctx context.Context, endpoint transport.Endpoint, stype linux.SockType) *fs.File { dirent := socket.NewDirent(ctx, unixSocketDevice) - defer dirent.DecRef() + defer dirent.DecRef(ctx) return NewWithDirent(ctx, dirent, endpoint, stype, fs.FileFlags{Read: true, Write: true, NonSeekable: true}) } @@ -97,17 +97,17 @@ type socketOpsCommon struct { } // DecRef implements RefCounter.DecRef. -func (s *socketOpsCommon) DecRef() { - s.DecRefWithDestructor(func() { - s.ep.Close() +func (s *socketOpsCommon) DecRef(ctx context.Context) { + s.DecRefWithDestructor(ctx, func(context.Context) { + s.ep.Close(ctx) }) } // Release implemements fs.FileOperations.Release. -func (s *socketOpsCommon) Release() { +func (s *socketOpsCommon) Release(ctx context.Context) { // Release only decrements a reference on s because s may be referenced in // the abstract socket namespace. - s.DecRef() + s.DecRef(ctx) } func (s *socketOpsCommon) isPacket() bool { @@ -234,7 +234,7 @@ func (s *SocketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, } ns := New(t, ep, s.stype) - defer ns.DecRef() + defer ns.DecRef(t) if flags&linux.SOCK_NONBLOCK != 0 { flags := ns.Flags() @@ -284,7 +284,7 @@ func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { if t.IsNetworkNamespaced() { return syserr.ErrInvalidEndpointState } - if err := t.AbstractSockets().Bind(p[1:], bep, s); err != nil { + if err := t.AbstractSockets().Bind(t, p[1:], bep, s); err != nil { // syserr.ErrPortInUse corresponds to EADDRINUSE. return syserr.ErrPortInUse } @@ -294,7 +294,7 @@ func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { var name string cwd := t.FSContext().WorkingDirectory() - defer cwd.DecRef() + defer cwd.DecRef(t) // Is there no slash at all? if !strings.Contains(p, "/") { @@ -302,7 +302,7 @@ func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { name = p } else { root := t.FSContext().RootDirectory() - defer root.DecRef() + defer root.DecRef(t) // Find the last path component, we know that something follows // that final slash, otherwise extractPath() would have failed. lastSlash := strings.LastIndex(p, "/") @@ -318,7 +318,7 @@ func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { // No path available. return syserr.ErrNoSuchFile } - defer d.DecRef() + defer d.DecRef(t) name = p[lastSlash+1:] } @@ -332,7 +332,7 @@ func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { if err != nil { return syserr.ErrPortInUse } - childDir.DecRef() + childDir.DecRef(t) } return nil @@ -378,9 +378,9 @@ func extractEndpoint(t *kernel.Task, sockaddr []byte) (transport.BoundEndpoint, FollowFinalSymlink: true, } ep, e := t.Kernel().VFS().BoundEndpointAt(t, t.Credentials(), &pop, &vfs.BoundEndpointOptions{path}) - root.DecRef() + root.DecRef(t) if relPath { - start.DecRef() + start.DecRef(t) } if e != nil { return nil, syserr.FromError(e) @@ -393,15 +393,15 @@ func extractEndpoint(t *kernel.Task, sockaddr []byte) (transport.BoundEndpoint, cwd := t.FSContext().WorkingDirectory() remainingTraversals := uint(fs.DefaultTraversalLimit) d, e := t.MountNamespace().FindInode(t, root, cwd, path, &remainingTraversals) - cwd.DecRef() - root.DecRef() + cwd.DecRef(t) + root.DecRef(t) if e != nil { return nil, syserr.FromError(e) } // Extract the endpoint if one is there. ep := d.Inode.BoundEndpoint(path) - d.DecRef() + d.DecRef(t) if ep == nil { // No socket! return nil, syserr.ErrConnectionRefused @@ -415,7 +415,7 @@ func (s *socketOpsCommon) Connect(t *kernel.Task, sockaddr []byte, blocking bool if err != nil { return err } - defer ep.Release() + defer ep.Release(t) // Connect the server endpoint. err = s.ep.Connect(t, ep) @@ -473,7 +473,7 @@ func (s *socketOpsCommon) SendMsg(t *kernel.Task, src usermem.IOSequence, to []b if err != nil { return 0, err } - defer ep.Release() + defer ep.Release(t) w.To = ep if ep.Passcred() && w.Control.Credentials == nil { diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go index 05c16fcfe..dfa25241a 100644 --- a/pkg/sentry/socket/unix/unix_vfs2.go +++ b/pkg/sentry/socket/unix/unix_vfs2.go @@ -136,7 +136,7 @@ func (s *SocketVFS2) Accept(t *kernel.Task, peerRequested bool, flags int, block if err != nil { return 0, nil, 0, err } - defer ns.DecRef() + defer ns.DecRef(t) if flags&linux.SOCK_NONBLOCK != 0 { ns.SetStatusFlags(t, t.Credentials(), linux.SOCK_NONBLOCK) @@ -183,19 +183,19 @@ func (s *SocketVFS2) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { if t.IsNetworkNamespaced() { return syserr.ErrInvalidEndpointState } - if err := t.AbstractSockets().Bind(p[1:], bep, s); err != nil { + if err := t.AbstractSockets().Bind(t, p[1:], bep, s); err != nil { // syserr.ErrPortInUse corresponds to EADDRINUSE. return syserr.ErrPortInUse } } else { path := fspath.Parse(p) root := t.FSContext().RootDirectoryVFS2() - defer root.DecRef() + defer root.DecRef(t) start := root relPath := !path.Absolute if relPath { start = t.FSContext().WorkingDirectoryVFS2() - defer start.DecRef() + defer start.DecRef(t) } pop := vfs.PathOperation{ Root: root, @@ -333,7 +333,7 @@ func (*providerVFS2) Socket(t *kernel.Task, stype linux.SockType, protocol int) f, err := NewSockfsFile(t, ep, stype) if err != nil { - ep.Close() + ep.Close(t) return nil, err } return f, nil @@ -357,14 +357,14 @@ func (*providerVFS2) Pair(t *kernel.Task, stype linux.SockType, protocol int) (* ep1, ep2 := transport.NewPair(t, stype, t.Kernel()) s1, err := NewSockfsFile(t, ep1, stype) if err != nil { - ep1.Close() - ep2.Close() + ep1.Close(t) + ep2.Close(t) return nil, nil, err } s2, err := NewSockfsFile(t, ep2, stype) if err != nil { - s1.DecRef() - ep2.Close() + s1.DecRef(t) + ep2.Close(t) return nil, nil, err } diff --git a/pkg/sentry/strace/strace.go b/pkg/sentry/strace/strace.go index 68ca537c8..87b239730 100644 --- a/pkg/sentry/strace/strace.go +++ b/pkg/sentry/strace/strace.go @@ -147,14 +147,14 @@ func fd(t *kernel.Task, fd int32) string { root := t.FSContext().RootDirectory() if root != nil { - defer root.DecRef() + defer root.DecRef(t) } if fd == linux.AT_FDCWD { wd := t.FSContext().WorkingDirectory() var name string if wd != nil { - defer wd.DecRef() + defer wd.DecRef(t) name, _ = wd.FullName(root) } else { name = "(unknown cwd)" @@ -167,7 +167,7 @@ func fd(t *kernel.Task, fd int32) string { // Cast FD to uint64 to avoid printing negative hex. return fmt.Sprintf("%#x (bad FD)", uint64(fd)) } - defer file.DecRef() + defer file.DecRef(t) name, _ := file.Dirent.FullName(root) return fmt.Sprintf("%#x %s", fd, name) @@ -175,12 +175,12 @@ func fd(t *kernel.Task, fd int32) string { func fdVFS2(t *kernel.Task, fd int32) string { root := t.FSContext().RootDirectoryVFS2() - defer root.DecRef() + defer root.DecRef(t) vfsObj := root.Mount().Filesystem().VirtualFilesystem() if fd == linux.AT_FDCWD { wd := t.FSContext().WorkingDirectoryVFS2() - defer wd.DecRef() + defer wd.DecRef(t) name, _ := vfsObj.PathnameWithDeleted(t, root, wd) return fmt.Sprintf("AT_FDCWD %s", name) @@ -191,7 +191,7 @@ func fdVFS2(t *kernel.Task, fd int32) string { // Cast FD to uint64 to avoid printing negative hex. return fmt.Sprintf("%#x (bad FD)", uint64(fd)) } - defer file.DecRef() + defer file.DecRef(t) name, _ := vfsObj.PathnameWithDeleted(t, root, file.VirtualDentry()) return fmt.Sprintf("%#x %s", fd, name) diff --git a/pkg/sentry/syscalls/epoll.go b/pkg/sentry/syscalls/epoll.go index d9fb808c0..d23a0068a 100644 --- a/pkg/sentry/syscalls/epoll.go +++ b/pkg/sentry/syscalls/epoll.go @@ -28,7 +28,7 @@ import ( // CreateEpoll implements the epoll_create(2) linux syscall. func CreateEpoll(t *kernel.Task, closeOnExec bool) (int32, error) { file := epoll.NewEventPoll(t) - defer file.DecRef() + defer file.DecRef(t) fd, err := t.NewFDFrom(0, file, kernel.FDFlags{ CloseOnExec: closeOnExec, @@ -47,14 +47,14 @@ func AddEpoll(t *kernel.Task, epfd int32, fd int32, flags epoll.EntryFlags, mask if epollfile == nil { return syserror.EBADF } - defer epollfile.DecRef() + defer epollfile.DecRef(t) // Get the target file id. file := t.GetFile(fd) if file == nil { return syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the epollPoll operations. e, ok := epollfile.FileOperations.(*epoll.EventPoll) @@ -73,14 +73,14 @@ func UpdateEpoll(t *kernel.Task, epfd int32, fd int32, flags epoll.EntryFlags, m if epollfile == nil { return syserror.EBADF } - defer epollfile.DecRef() + defer epollfile.DecRef(t) // Get the target file id. file := t.GetFile(fd) if file == nil { return syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the epollPoll operations. e, ok := epollfile.FileOperations.(*epoll.EventPoll) @@ -99,14 +99,14 @@ func RemoveEpoll(t *kernel.Task, epfd int32, fd int32) error { if epollfile == nil { return syserror.EBADF } - defer epollfile.DecRef() + defer epollfile.DecRef(t) // Get the target file id. file := t.GetFile(fd) if file == nil { return syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the epollPoll operations. e, ok := epollfile.FileOperations.(*epoll.EventPoll) @@ -115,7 +115,7 @@ func RemoveEpoll(t *kernel.Task, epfd int32, fd int32) error { } // Try to remove the entry. - return e.RemoveEntry(epoll.FileIdentifier{file, fd}) + return e.RemoveEntry(t, epoll.FileIdentifier{file, fd}) } // WaitEpoll implements the epoll_wait(2) linux syscall. @@ -125,7 +125,7 @@ func WaitEpoll(t *kernel.Task, fd int32, max int, timeout int) ([]linux.EpollEve if epollfile == nil { return nil, syserror.EBADF } - defer epollfile.DecRef() + defer epollfile.DecRef(t) // Extract the epollPoll operations. e, ok := epollfile.FileOperations.(*epoll.EventPoll) diff --git a/pkg/sentry/syscalls/linux/sys_aio.go b/pkg/sentry/syscalls/linux/sys_aio.go index ba2557c52..e9d64dec5 100644 --- a/pkg/sentry/syscalls/linux/sys_aio.go +++ b/pkg/sentry/syscalls/linux/sys_aio.go @@ -247,7 +247,7 @@ func getAIOCallback(t *kernel.Task, file *fs.File, cbAddr usermem.Addr, cb *linu ev.Result = -int64(kernel.ExtractErrno(err, 0)) } - file.DecRef() + file.DecRef(ctx) // Queue the result for delivery. actx.FinishRequest(ev) @@ -257,7 +257,7 @@ func getAIOCallback(t *kernel.Task, file *fs.File, cbAddr usermem.Addr, cb *linu // wake up. if eventFile != nil { eventFile.FileOperations.(*eventfd.EventOperations).Signal(1) - eventFile.DecRef() + eventFile.DecRef(ctx) } } } @@ -269,7 +269,7 @@ func submitCallback(t *kernel.Task, id uint64, cb *linux.IOCallback, cbAddr user // File not found. return syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Was there an eventFD? Extract it. var eventFile *fs.File @@ -279,7 +279,7 @@ func submitCallback(t *kernel.Task, id uint64, cb *linux.IOCallback, cbAddr user // Bad FD. return syserror.EBADF } - defer eventFile.DecRef() + defer eventFile.DecRef(t) // Check that it is an eventfd. if _, ok := eventFile.FileOperations.(*eventfd.EventOperations); !ok { diff --git a/pkg/sentry/syscalls/linux/sys_eventfd.go b/pkg/sentry/syscalls/linux/sys_eventfd.go index ed3413ca6..3b4f879e4 100644 --- a/pkg/sentry/syscalls/linux/sys_eventfd.go +++ b/pkg/sentry/syscalls/linux/sys_eventfd.go @@ -37,7 +37,7 @@ func Eventfd2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc event.SetFlags(fs.SettableFileFlags{ NonBlocking: flags&linux.EFD_NONBLOCK != 0, }) - defer event.DecRef() + defer event.DecRef(t) fd, err := t.NewFDFrom(0, event, kernel.FDFlags{ CloseOnExec: flags&linux.EFD_CLOEXEC != 0, diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go index 8cf6401e7..1bc9b184e 100644 --- a/pkg/sentry/syscalls/linux/sys_file.go +++ b/pkg/sentry/syscalls/linux/sys_file.go @@ -40,7 +40,7 @@ func fileOpAt(t *kernel.Task, dirFD int32, path string, fn func(root *fs.Dirent, // Common case: we are accessing a file in the root. root := t.FSContext().RootDirectory() err := fn(root, root, name, linux.MaxSymlinkTraversals) - root.DecRef() + root.DecRef(t) return err } else if dir == "." && dirFD == linux.AT_FDCWD { // Common case: we are accessing a file relative to the current @@ -48,8 +48,8 @@ func fileOpAt(t *kernel.Task, dirFD int32, path string, fn func(root *fs.Dirent, wd := t.FSContext().WorkingDirectory() root := t.FSContext().RootDirectory() err := fn(root, wd, name, linux.MaxSymlinkTraversals) - wd.DecRef() - root.DecRef() + wd.DecRef(t) + root.DecRef(t) return err } @@ -97,19 +97,19 @@ func fileOpOn(t *kernel.Task, dirFD int32, path string, resolve bool, fn func(ro } else { d, err = t.MountNamespace().FindLink(t, root, rel, path, &remainingTraversals) } - root.DecRef() + root.DecRef(t) if wd != nil { - wd.DecRef() + wd.DecRef(t) } if f != nil { - f.DecRef() + f.DecRef(t) } if err != nil { return err } err = fn(root, d, remainingTraversals) - d.DecRef() + d.DecRef(t) return err } @@ -186,7 +186,7 @@ func openAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint) (fd uint if err != nil { return syserror.ConvertIntr(err, kernel.ERESTARTSYS) } - defer file.DecRef() + defer file.DecRef(t) // Success. newFD, err := t.NewFDFrom(0, file, kernel.FDFlags{ @@ -242,7 +242,7 @@ func mknodAt(t *kernel.Task, dirFD int32, addr usermem.Addr, mode linux.FileMode if err != nil { return err } - file.DecRef() + file.DecRef(t) return nil case linux.ModeNamedPipe: @@ -332,7 +332,7 @@ func createAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint, mode l if err != nil { break } - defer found.DecRef() + defer found.DecRef(t) // We found something (possibly a symlink). If the // O_EXCL flag was passed, then we can immediately @@ -357,7 +357,7 @@ func createAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint, mode l resolved, err = found.Inode.Getlink(t) if err == nil { // No more resolution necessary. - defer resolved.DecRef() + defer resolved.DecRef(t) break } if err != fs.ErrResolveViaReadlink { @@ -384,7 +384,7 @@ func createAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint, mode l if err != nil { break } - defer newParent.DecRef() + defer newParent.DecRef(t) // Repeat the process with the parent and name of the // symlink target. @@ -416,7 +416,7 @@ func createAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint, mode l if err != nil { return syserror.ConvertIntr(err, kernel.ERESTARTSYS) } - defer newFile.DecRef() + defer newFile.DecRef(t) case syserror.ENOENT: // File does not exist. Proceed with creation. @@ -432,7 +432,7 @@ func createAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint, mode l // No luck, bail. return err } - defer newFile.DecRef() + defer newFile.DecRef(t) found = newFile.Dirent default: return err @@ -596,7 +596,7 @@ func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Shared flags between file and socket. switch request { @@ -671,9 +671,9 @@ func Getcwd(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal addr := args[0].Pointer() size := args[1].SizeT() cwd := t.FSContext().WorkingDirectory() - defer cwd.DecRef() + defer cwd.DecRef(t) root := t.FSContext().RootDirectory() - defer root.DecRef() + defer root.DecRef(t) // Get our fullname from the root and preprend unreachable if the root was // unreachable from our current dirent this is the same behavior as on linux. @@ -722,7 +722,7 @@ func Chroot(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal return err } - t.FSContext().SetRootDirectory(d) + t.FSContext().SetRootDirectory(t, d) return nil }) } @@ -747,7 +747,7 @@ func Chdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return err } - t.FSContext().SetWorkingDirectory(d) + t.FSContext().SetWorkingDirectory(t, d) return nil }) } @@ -760,7 +760,7 @@ func Fchdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Is it a directory? if !fs.IsDir(file.Dirent.Inode.StableAttr) { @@ -772,7 +772,7 @@ func Fchdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal return 0, nil, err } - t.FSContext().SetWorkingDirectory(file.Dirent) + t.FSContext().SetWorkingDirectory(t, file.Dirent) return 0, nil, nil } @@ -791,7 +791,7 @@ func Close(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) err := file.Flush(t) return 0, nil, handleIOError(t, false /* partial */, err, syserror.EINTR, "close", file) @@ -805,7 +805,7 @@ func Dup(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) newFD, err := t.NewFDFrom(0, file, kernel.FDFlags{}) if err != nil { @@ -826,7 +826,7 @@ func Dup2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if oldFile == nil { return 0, nil, syserror.EBADF } - defer oldFile.DecRef() + defer oldFile.DecRef(t) return uintptr(newfd), nil, nil } @@ -850,7 +850,7 @@ func Dup3(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if oldFile == nil { return 0, nil, syserror.EBADF } - defer oldFile.DecRef() + defer oldFile.DecRef(t) err := t.NewFDAt(newfd, oldFile, kernel.FDFlags{CloseOnExec: flags&linux.O_CLOEXEC != 0}) if err != nil { @@ -925,7 +925,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) switch cmd { case linux.F_DUPFD, linux.F_DUPFD_CLOEXEC: @@ -1132,7 +1132,7 @@ func Fadvise64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // If the FD refers to a pipe or FIFO, return error. if fs.IsPipe(file.Dirent.Inode.StableAttr) { @@ -1171,7 +1171,7 @@ func mkdirAt(t *kernel.Task, dirFD int32, addr usermem.Addr, mode linux.FileMode switch err { case nil: // The directory existed. - defer f.DecRef() + defer f.DecRef(t) return syserror.EEXIST case syserror.EACCES: // Permission denied while walking to the directory. @@ -1349,7 +1349,7 @@ func linkAt(t *kernel.Task, oldDirFD int32, oldAddr usermem.Addr, newDirFD int32 if target == nil { return syserror.EBADF } - defer target.DecRef() + defer target.DecRef(t) if err := mayLinkAt(t, target.Dirent.Inode); err != nil { return err } @@ -1602,7 +1602,7 @@ func Ftruncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Reject truncation if the file flags do not permit this operation. // This is different from truncate(2) above. @@ -1730,7 +1730,7 @@ func chownAt(t *kernel.Task, fd int32, addr usermem.Addr, resolve, allowEmpty bo if file == nil { return syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return chown(t, file.Dirent, uid, gid) } @@ -1768,7 +1768,7 @@ func Fchown(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return 0, nil, chown(t, file.Dirent, uid, gid) } @@ -1833,7 +1833,7 @@ func Fchmod(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return 0, nil, chmod(t, file.Dirent, mode) } @@ -1893,10 +1893,10 @@ func utimes(t *kernel.Task, dirFD int32, addr usermem.Addr, ts fs.TimeSpec, reso if f == nil { return syserror.EBADF } - defer f.DecRef() + defer f.DecRef(t) root := t.FSContext().RootDirectory() - defer root.DecRef() + defer root.DecRef(t) return setTimestamp(root, f.Dirent, linux.MaxSymlinkTraversals) } @@ -2088,7 +2088,7 @@ func Fallocate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) if offset < 0 || length <= 0 { return 0, nil, syserror.EINVAL @@ -2141,7 +2141,7 @@ func Flock(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // flock(2): EBADF fd is not an open file descriptor. return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) nonblocking := operation&linux.LOCK_NB != 0 operation &^= linux.LOCK_NB @@ -2224,8 +2224,8 @@ func MemfdCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S return 0, nil, err } - defer dirent.DecRef() - defer file.DecRef() + defer dirent.DecRef(t) + defer file.DecRef(t) newFD, err := t.NewFDFrom(0, file, kernel.FDFlags{ CloseOnExec: cloExec, diff --git a/pkg/sentry/syscalls/linux/sys_futex.go b/pkg/sentry/syscalls/linux/sys_futex.go index f04d78856..9d1b2edb1 100644 --- a/pkg/sentry/syscalls/linux/sys_futex.go +++ b/pkg/sentry/syscalls/linux/sys_futex.go @@ -73,7 +73,7 @@ func futexWaitAbsolute(t *kernel.Task, clockRealtime bool, ts linux.Timespec, fo err = t.BlockWithDeadline(w.C, true, ktime.FromTimespec(ts)) } - t.Futex().WaitComplete(w) + t.Futex().WaitComplete(w, t) return 0, syserror.ConvertIntr(err, kernel.ERESTARTSYS) } @@ -95,7 +95,7 @@ func futexWaitDuration(t *kernel.Task, duration time.Duration, forever bool, add } remaining, err := t.BlockWithTimeout(w.C, !forever, duration) - t.Futex().WaitComplete(w) + t.Futex().WaitComplete(w, t) if err == nil { return 0, nil } @@ -148,7 +148,7 @@ func futexLockPI(t *kernel.Task, ts linux.Timespec, forever bool, addr usermem.A timer.Destroy() } - t.Futex().WaitComplete(w) + t.Futex().WaitComplete(w, t) return syserror.ConvertIntr(err, kernel.ERESTARTSYS) } diff --git a/pkg/sentry/syscalls/linux/sys_getdents.go b/pkg/sentry/syscalls/linux/sys_getdents.go index b126fecc0..f5699e55d 100644 --- a/pkg/sentry/syscalls/linux/sys_getdents.go +++ b/pkg/sentry/syscalls/linux/sys_getdents.go @@ -68,7 +68,7 @@ func getdents(t *kernel.Task, fd int32, addr usermem.Addr, size int, f func(*dir if dir == nil { return 0, syserror.EBADF } - defer dir.DecRef() + defer dir.DecRef(t) w := &usermem.IOReadWriter{ Ctx: t, diff --git a/pkg/sentry/syscalls/linux/sys_inotify.go b/pkg/sentry/syscalls/linux/sys_inotify.go index b2c7b3444..cf47bb9dd 100644 --- a/pkg/sentry/syscalls/linux/sys_inotify.go +++ b/pkg/sentry/syscalls/linux/sys_inotify.go @@ -40,7 +40,7 @@ func InotifyInit1(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. NonBlocking: flags&linux.IN_NONBLOCK != 0, } n := fs.NewFile(t, dirent, fileFlags, fs.NewInotify(t)) - defer n.DecRef() + defer n.DecRef(t) fd, err := t.NewFDFrom(0, n, kernel.FDFlags{ CloseOnExec: flags&linux.IN_CLOEXEC != 0, @@ -71,7 +71,7 @@ func fdToInotify(t *kernel.Task, fd int32) (*fs.Inotify, *fs.File, error) { ino, ok := file.FileOperations.(*fs.Inotify) if !ok { // Not an inotify fd. - file.DecRef() + file.DecRef(t) return nil, nil, syserror.EINVAL } @@ -98,7 +98,7 @@ func InotifyAddWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kern if err != nil { return 0, nil, err } - defer file.DecRef() + defer file.DecRef(t) path, _, err := copyInPath(t, addr, false /* allowEmpty */) if err != nil { @@ -128,6 +128,6 @@ func InotifyRmWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne if err != nil { return 0, nil, err } - defer file.DecRef() - return 0, nil, ino.RmWatch(wd) + defer file.DecRef(t) + return 0, nil, ino.RmWatch(t, wd) } diff --git a/pkg/sentry/syscalls/linux/sys_lseek.go b/pkg/sentry/syscalls/linux/sys_lseek.go index 3f7691eae..1c38f8f4f 100644 --- a/pkg/sentry/syscalls/linux/sys_lseek.go +++ b/pkg/sentry/syscalls/linux/sys_lseek.go @@ -33,7 +33,7 @@ func Lseek(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) var sw fs.SeekWhence switch whence { diff --git a/pkg/sentry/syscalls/linux/sys_mmap.go b/pkg/sentry/syscalls/linux/sys_mmap.go index 91694d374..72786b032 100644 --- a/pkg/sentry/syscalls/linux/sys_mmap.go +++ b/pkg/sentry/syscalls/linux/sys_mmap.go @@ -75,7 +75,7 @@ func Mmap(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC } defer func() { if opts.MappingIdentity != nil { - opts.MappingIdentity.DecRef() + opts.MappingIdentity.DecRef(t) } }() @@ -85,7 +85,7 @@ func Mmap(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) flags := file.Flags() // mmap unconditionally requires that the FD is readable. diff --git a/pkg/sentry/syscalls/linux/sys_mount.go b/pkg/sentry/syscalls/linux/sys_mount.go index eb5ff48f5..bd0633564 100644 --- a/pkg/sentry/syscalls/linux/sys_mount.go +++ b/pkg/sentry/syscalls/linux/sys_mount.go @@ -115,7 +115,7 @@ func Mount(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall }); err != nil { // Something went wrong. Drop our ref on rootInode before // returning the error. - rootInode.DecRef() + rootInode.DecRef(t) return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/sys_pipe.go b/pkg/sentry/syscalls/linux/sys_pipe.go index 43c510930..3149e4aad 100644 --- a/pkg/sentry/syscalls/linux/sys_pipe.go +++ b/pkg/sentry/syscalls/linux/sys_pipe.go @@ -34,10 +34,10 @@ func pipe2(t *kernel.Task, addr usermem.Addr, flags uint) (uintptr, error) { r, w := pipe.NewConnectedPipe(t, pipe.DefaultPipeSize, usermem.PageSize) r.SetFlags(linuxToFlags(flags).Settable()) - defer r.DecRef() + defer r.DecRef(t) w.SetFlags(linuxToFlags(flags).Settable()) - defer w.DecRef() + defer w.DecRef(t) fds, err := t.NewFDs(0, []*fs.File{r, w}, kernel.FDFlags{ CloseOnExec: flags&linux.O_CLOEXEC != 0, @@ -49,7 +49,7 @@ func pipe2(t *kernel.Task, addr usermem.Addr, flags uint) (uintptr, error) { if _, err := t.CopyOut(addr, fds); err != nil { for _, fd := range fds { if file, _ := t.FDTable().Remove(fd); file != nil { - file.DecRef() + file.DecRef(t) } } return 0, err diff --git a/pkg/sentry/syscalls/linux/sys_poll.go b/pkg/sentry/syscalls/linux/sys_poll.go index f0198141c..3435bdf77 100644 --- a/pkg/sentry/syscalls/linux/sys_poll.go +++ b/pkg/sentry/syscalls/linux/sys_poll.go @@ -70,7 +70,7 @@ func initReadiness(t *kernel.Task, pfd *linux.PollFD, state *pollState, ch chan } if ch == nil { - defer file.DecRef() + defer file.DecRef(t) } else { state.file = file state.waiter, _ = waiter.NewChannelEntry(ch) @@ -82,11 +82,11 @@ func initReadiness(t *kernel.Task, pfd *linux.PollFD, state *pollState, ch chan } // releaseState releases all the pollState in "state". -func releaseState(state []pollState) { +func releaseState(t *kernel.Task, state []pollState) { for i := range state { if state[i].file != nil { state[i].file.EventUnregister(&state[i].waiter) - state[i].file.DecRef() + state[i].file.DecRef(t) } } } @@ -107,7 +107,7 @@ func pollBlock(t *kernel.Task, pfd []linux.PollFD, timeout time.Duration) (time. // result, we stop registering for events but still go through all files // to get their ready masks. state := make([]pollState, len(pfd)) - defer releaseState(state) + defer releaseState(t, state) n := uintptr(0) for i := range pfd { initReadiness(t, &pfd[i], &state[i], ch) @@ -266,7 +266,7 @@ func doSelect(t *kernel.Task, nfds int, readFDs, writeFDs, exceptFDs usermem.Add if file == nil { return 0, syserror.EBADF } - file.DecRef() + file.DecRef(t) var mask int16 if (rV & m) != 0 { diff --git a/pkg/sentry/syscalls/linux/sys_prctl.go b/pkg/sentry/syscalls/linux/sys_prctl.go index f92bf8096..64a725296 100644 --- a/pkg/sentry/syscalls/linux/sys_prctl.go +++ b/pkg/sentry/syscalls/linux/sys_prctl.go @@ -128,7 +128,7 @@ func Prctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // They trying to set exe to a non-file? if !fs.IsFile(file.Dirent.Inode.StableAttr) { @@ -136,7 +136,7 @@ func Prctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } // Set the underlying executable. - t.MemoryManager().SetExecutable(fsbridge.NewFSFile(file)) + t.MemoryManager().SetExecutable(t, fsbridge.NewFSFile(file)) case linux.PR_SET_MM_AUXV, linux.PR_SET_MM_START_CODE, diff --git a/pkg/sentry/syscalls/linux/sys_read.go b/pkg/sentry/syscalls/linux/sys_read.go index 071b4bacc..3bbc3fa4b 100644 --- a/pkg/sentry/syscalls/linux/sys_read.go +++ b/pkg/sentry/syscalls/linux/sys_read.go @@ -48,7 +48,7 @@ func Read(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the file is readable. if !file.Flags().Read { @@ -84,7 +84,7 @@ func Readahead(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the file is readable. if !file.Flags().Read { @@ -118,7 +118,7 @@ func Pread64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate and does not overflow. if offset < 0 || offset+int64(size) < 0 { @@ -164,7 +164,7 @@ func Readv(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the file is readable. if !file.Flags().Read { @@ -195,7 +195,7 @@ func Preadv(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate. if offset < 0 { @@ -244,7 +244,7 @@ func Preadv2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate. if offset < -1 { diff --git a/pkg/sentry/syscalls/linux/sys_shm.go b/pkg/sentry/syscalls/linux/sys_shm.go index 4a8bc24a2..f0ae8fa8e 100644 --- a/pkg/sentry/syscalls/linux/sys_shm.go +++ b/pkg/sentry/syscalls/linux/sys_shm.go @@ -39,7 +39,7 @@ func Shmget(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, err } - defer segment.DecRef() + defer segment.DecRef(t) return uintptr(segment.ID), nil, nil } @@ -66,7 +66,7 @@ func Shmat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if err != nil { return 0, nil, syserror.EINVAL } - defer segment.DecRef() + defer segment.DecRef(t) opts, err := segment.ConfigureAttach(t, addr, shm.AttachOpts{ Execute: flag&linux.SHM_EXEC == linux.SHM_EXEC, @@ -108,7 +108,7 @@ func Shmctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, syserror.EINVAL } - defer segment.DecRef() + defer segment.DecRef(t) stat, err := segment.IPCStat(t) if err == nil { @@ -132,7 +132,7 @@ func Shmctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, syserror.EINVAL } - defer segment.DecRef() + defer segment.DecRef(t) switch cmd { case linux.IPC_SET: @@ -145,7 +145,7 @@ func Shmctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal return 0, nil, err case linux.IPC_RMID: - segment.MarkDestroyed() + segment.MarkDestroyed(t) return 0, nil, nil case linux.SHM_LOCK, linux.SHM_UNLOCK: diff --git a/pkg/sentry/syscalls/linux/sys_signal.go b/pkg/sentry/syscalls/linux/sys_signal.go index d2b0012ae..20cb1a5cb 100644 --- a/pkg/sentry/syscalls/linux/sys_signal.go +++ b/pkg/sentry/syscalls/linux/sys_signal.go @@ -536,7 +536,7 @@ func sharedSignalfd(t *kernel.Task, fd int32, sigset usermem.Addr, sigsetsize ui if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Is this a signalfd? if s, ok := file.FileOperations.(*signalfd.SignalOperations); ok { @@ -553,7 +553,7 @@ func sharedSignalfd(t *kernel.Task, fd int32, sigset usermem.Addr, sigsetsize ui if err != nil { return 0, nil, err } - defer file.DecRef() + defer file.DecRef(t) // Set appropriate flags. file.SetFlags(fs.SettableFileFlags{ diff --git a/pkg/sentry/syscalls/linux/sys_socket.go b/pkg/sentry/syscalls/linux/sys_socket.go index 414fce8e3..fec1c1974 100644 --- a/pkg/sentry/syscalls/linux/sys_socket.go +++ b/pkg/sentry/syscalls/linux/sys_socket.go @@ -200,7 +200,7 @@ func Socket(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal s.SetFlags(fs.SettableFileFlags{ NonBlocking: stype&linux.SOCK_NONBLOCK != 0, }) - defer s.DecRef() + defer s.DecRef(t) fd, err := t.NewFDFrom(0, s, kernel.FDFlags{ CloseOnExec: stype&linux.SOCK_CLOEXEC != 0, @@ -235,8 +235,8 @@ func SocketPair(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy } s1.SetFlags(fileFlags) s2.SetFlags(fileFlags) - defer s1.DecRef() - defer s2.DecRef() + defer s1.DecRef(t) + defer s2.DecRef(t) // Create the FDs for the sockets. fds, err := t.NewFDs(0, []*fs.File{s1, s2}, kernel.FDFlags{ @@ -250,7 +250,7 @@ func SocketPair(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy if _, err := t.CopyOut(socks, fds); err != nil { for _, fd := range fds { if file, _ := t.FDTable().Remove(fd); file != nil { - file.DecRef() + file.DecRef(t) } } return 0, nil, err @@ -270,7 +270,7 @@ func Connect(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -301,7 +301,7 @@ func accept(t *kernel.Task, fd int32, addr usermem.Addr, addrLen usermem.Addr, f if file == nil { return 0, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -360,7 +360,7 @@ func Bind(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -387,7 +387,7 @@ func Listen(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -416,7 +416,7 @@ func Shutdown(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -447,7 +447,7 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -529,7 +529,7 @@ func SetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -567,7 +567,7 @@ func GetSockName(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -595,7 +595,7 @@ func GetPeerName(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -628,7 +628,7 @@ func RecvMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -681,7 +681,7 @@ func RecvMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -775,7 +775,7 @@ func recvSingleMsg(t *kernel.Task, s socket.Socket, msgPtr usermem.Addr, flags i } if !cms.Unix.Empty() { mflags |= linux.MSG_CTRUNC - cms.Release() + cms.Release(t) } if int(msg.Flags) != mflags { @@ -795,7 +795,7 @@ func recvSingleMsg(t *kernel.Task, s socket.Socket, msgPtr usermem.Addr, flags i if e != nil { return 0, syserror.ConvertIntr(e.ToError(), kernel.ERESTARTSYS) } - defer cms.Release() + defer cms.Release(t) controlData := make([]byte, 0, msg.ControlLen) controlData = control.PackControlMessages(t, cms, controlData) @@ -851,7 +851,7 @@ func recvFrom(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flag if file == nil { return 0, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -880,7 +880,7 @@ func recvFrom(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flag } n, _, sender, senderLen, cm, e := s.RecvMsg(t, dst, int(flags), haveDeadline, deadline, nameLenPtr != 0, 0) - cm.Release() + cm.Release(t) if e != nil { return 0, syserror.ConvertIntr(e.ToError(), kernel.ERESTARTSYS) } @@ -924,7 +924,7 @@ func SendMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -962,7 +962,7 @@ func SendMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -1066,7 +1066,7 @@ func sendSingleMsg(t *kernel.Task, s socket.Socket, file *fs.File, msgPtr userme n, e := s.SendMsg(t, src, to, int(flags), haveDeadline, deadline, controlMessages) err = handleIOError(t, n != 0, e.ToError(), kernel.ERESTARTSYS, "sendmsg", file) if err != nil { - controlMessages.Release() + controlMessages.Release(t) } return uintptr(n), err } @@ -1084,7 +1084,7 @@ func sendTo(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flags if file == nil { return 0, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) diff --git a/pkg/sentry/syscalls/linux/sys_splice.go b/pkg/sentry/syscalls/linux/sys_splice.go index 77c78889d..b8846a10a 100644 --- a/pkg/sentry/syscalls/linux/sys_splice.go +++ b/pkg/sentry/syscalls/linux/sys_splice.go @@ -101,7 +101,7 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if inFile == nil { return 0, nil, syserror.EBADF } - defer inFile.DecRef() + defer inFile.DecRef(t) if !inFile.Flags().Read { return 0, nil, syserror.EBADF @@ -111,7 +111,7 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if outFile == nil { return 0, nil, syserror.EBADF } - defer outFile.DecRef() + defer outFile.DecRef(t) if !outFile.Flags().Write { return 0, nil, syserror.EBADF @@ -192,13 +192,13 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if outFile == nil { return 0, nil, syserror.EBADF } - defer outFile.DecRef() + defer outFile.DecRef(t) inFile := t.GetFile(inFD) if inFile == nil { return 0, nil, syserror.EBADF } - defer inFile.DecRef() + defer inFile.DecRef(t) // The operation is non-blocking if anything is non-blocking. // @@ -300,13 +300,13 @@ func Tee(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo if outFile == nil { return 0, nil, syserror.EBADF } - defer outFile.DecRef() + defer outFile.DecRef(t) inFile := t.GetFile(inFD) if inFile == nil { return 0, nil, syserror.EBADF } - defer inFile.DecRef() + defer inFile.DecRef(t) // All files must be pipes. if !fs.IsPipe(inFile.Dirent.Inode.StableAttr) || !fs.IsPipe(outFile.Dirent.Inode.StableAttr) { diff --git a/pkg/sentry/syscalls/linux/sys_stat.go b/pkg/sentry/syscalls/linux/sys_stat.go index 46ebf27a2..a5826f2dd 100644 --- a/pkg/sentry/syscalls/linux/sys_stat.go +++ b/pkg/sentry/syscalls/linux/sys_stat.go @@ -58,7 +58,7 @@ func Fstatat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return 0, nil, fstat(t, file, statAddr) } @@ -100,7 +100,7 @@ func Fstat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return 0, nil, fstat(t, file, statAddr) } @@ -158,7 +158,7 @@ func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) uattr, err := file.UnstableAttr(t) if err != nil { return 0, nil, err @@ -249,7 +249,7 @@ func Fstatfs(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return 0, nil, statfsImpl(t, file.Dirent, statfsAddr) } diff --git a/pkg/sentry/syscalls/linux/sys_sync.go b/pkg/sentry/syscalls/linux/sys_sync.go index 5ad465ae3..f2c0e5069 100644 --- a/pkg/sentry/syscalls/linux/sys_sync.go +++ b/pkg/sentry/syscalls/linux/sys_sync.go @@ -39,7 +39,7 @@ func Syncfs(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Use "sync-the-world" for now, it's guaranteed that fd is at least // on the root filesystem. @@ -54,7 +54,7 @@ func Fsync(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) err := file.Fsync(t, 0, fs.FileMaxOffset, fs.SyncAll) return 0, nil, syserror.ConvertIntr(err, kernel.ERESTARTSYS) @@ -70,7 +70,7 @@ func Fdatasync(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) err := file.Fsync(t, 0, fs.FileMaxOffset, fs.SyncData) return 0, nil, syserror.ConvertIntr(err, kernel.ERESTARTSYS) @@ -103,7 +103,7 @@ func SyncFileRange(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // SYNC_FILE_RANGE_WAIT_BEFORE waits upon write-out of all pages in the // specified range that have already been submitted to the device diff --git a/pkg/sentry/syscalls/linux/sys_thread.go b/pkg/sentry/syscalls/linux/sys_thread.go index 00915fdde..2d16e4933 100644 --- a/pkg/sentry/syscalls/linux/sys_thread.go +++ b/pkg/sentry/syscalls/linux/sys_thread.go @@ -117,7 +117,7 @@ func execveat(t *kernel.Task, dirFD int32, pathnameAddr, argvAddr, envvAddr user resolveFinal := flags&linux.AT_SYMLINK_NOFOLLOW == 0 root := t.FSContext().RootDirectory() - defer root.DecRef() + defer root.DecRef(t) var wd *fs.Dirent var executable fsbridge.File @@ -133,7 +133,7 @@ func execveat(t *kernel.Task, dirFD int32, pathnameAddr, argvAddr, envvAddr user if f == nil { return 0, nil, syserror.EBADF } - defer f.DecRef() + defer f.DecRef(t) closeOnExec = fdFlags.CloseOnExec if atEmptyPath && len(pathname) == 0 { @@ -155,7 +155,7 @@ func execveat(t *kernel.Task, dirFD int32, pathnameAddr, argvAddr, envvAddr user } } if wd != nil { - defer wd.DecRef() + defer wd.DecRef(t) } // Load the new TaskContext. diff --git a/pkg/sentry/syscalls/linux/sys_timerfd.go b/pkg/sentry/syscalls/linux/sys_timerfd.go index cf49b43db..34b03e4ee 100644 --- a/pkg/sentry/syscalls/linux/sys_timerfd.go +++ b/pkg/sentry/syscalls/linux/sys_timerfd.go @@ -43,7 +43,7 @@ func TimerfdCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel return 0, nil, syserror.EINVAL } f := timerfd.NewFile(t, c) - defer f.DecRef() + defer f.DecRef(t) f.SetFlags(fs.SettableFileFlags{ NonBlocking: flags&linux.TFD_NONBLOCK != 0, }) @@ -73,7 +73,7 @@ func TimerfdSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne if f == nil { return 0, nil, syserror.EBADF } - defer f.DecRef() + defer f.DecRef(t) tf, ok := f.FileOperations.(*timerfd.TimerOperations) if !ok { @@ -107,7 +107,7 @@ func TimerfdGettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne if f == nil { return 0, nil, syserror.EBADF } - defer f.DecRef() + defer f.DecRef(t) tf, ok := f.FileOperations.(*timerfd.TimerOperations) if !ok { diff --git a/pkg/sentry/syscalls/linux/sys_write.go b/pkg/sentry/syscalls/linux/sys_write.go index 6ec0de96e..485526e28 100644 --- a/pkg/sentry/syscalls/linux/sys_write.go +++ b/pkg/sentry/syscalls/linux/sys_write.go @@ -48,7 +48,7 @@ func Write(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the file is writable. if !file.Flags().Write { @@ -85,7 +85,7 @@ func Pwrite64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate and does not overflow. if offset < 0 || offset+int64(size) < 0 { @@ -131,7 +131,7 @@ func Writev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the file is writable. if !file.Flags().Write { @@ -162,7 +162,7 @@ func Pwritev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate. if offset < 0 { @@ -215,7 +215,7 @@ func Pwritev2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate. if offset < -1 { diff --git a/pkg/sentry/syscalls/linux/sys_xattr.go b/pkg/sentry/syscalls/linux/sys_xattr.go index c24946160..97474fd3c 100644 --- a/pkg/sentry/syscalls/linux/sys_xattr.go +++ b/pkg/sentry/syscalls/linux/sys_xattr.go @@ -49,7 +49,7 @@ func FGetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if f == nil { return 0, nil, syserror.EBADF } - defer f.DecRef() + defer f.DecRef(t) n, err := getXattr(t, f.Dirent, nameAddr, valueAddr, size) if err != nil { @@ -153,7 +153,7 @@ func FSetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if f == nil { return 0, nil, syserror.EBADF } - defer f.DecRef() + defer f.DecRef(t) return 0, nil, setXattr(t, f.Dirent, nameAddr, valueAddr, uint64(size), flags) } @@ -270,7 +270,7 @@ func FListXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy if f == nil { return 0, nil, syserror.EBADF } - defer f.DecRef() + defer f.DecRef(t) n, err := listXattr(t, f.Dirent, listAddr, size) if err != nil { @@ -384,7 +384,7 @@ func FRemoveXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. if f == nil { return 0, nil, syserror.EBADF } - defer f.DecRef() + defer f.DecRef(t) return 0, nil, removeXattr(t, f.Dirent, nameAddr) } diff --git a/pkg/sentry/syscalls/linux/vfs2/aio.go b/pkg/sentry/syscalls/linux/vfs2/aio.go index e5cdefc50..399b4f60c 100644 --- a/pkg/sentry/syscalls/linux/vfs2/aio.go +++ b/pkg/sentry/syscalls/linux/vfs2/aio.go @@ -88,7 +88,7 @@ func submitCallback(t *kernel.Task, id uint64, cb *linux.IOCallback, cbAddr user if fd == nil { return syserror.EBADF } - defer fd.DecRef() + defer fd.DecRef(t) // Was there an eventFD? Extract it. var eventFD *vfs.FileDescription @@ -97,7 +97,7 @@ func submitCallback(t *kernel.Task, id uint64, cb *linux.IOCallback, cbAddr user if eventFD == nil { return syserror.EBADF } - defer eventFD.DecRef() + defer eventFD.DecRef(t) // Check that it is an eventfd. if _, ok := eventFD.Impl().(*eventfd.EventFileDescription); !ok { @@ -169,7 +169,7 @@ func getAIOCallback(t *kernel.Task, fd, eventFD *vfs.FileDescription, cbAddr use ev.Result = -int64(kernel.ExtractErrno(err, 0)) } - fd.DecRef() + fd.DecRef(ctx) // Queue the result for delivery. aioCtx.FinishRequest(ev) @@ -179,7 +179,7 @@ func getAIOCallback(t *kernel.Task, fd, eventFD *vfs.FileDescription, cbAddr use // wake up. if eventFD != nil { eventFD.Impl().(*eventfd.EventFileDescription).Signal(1) - eventFD.DecRef() + eventFD.DecRef(ctx) } } } diff --git a/pkg/sentry/syscalls/linux/vfs2/epoll.go b/pkg/sentry/syscalls/linux/vfs2/epoll.go index 34c90ae3e..c62f03509 100644 --- a/pkg/sentry/syscalls/linux/vfs2/epoll.go +++ b/pkg/sentry/syscalls/linux/vfs2/epoll.go @@ -37,11 +37,11 @@ func EpollCreate1(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. return 0, nil, syserror.EINVAL } - file, err := t.Kernel().VFS().NewEpollInstanceFD() + file, err := t.Kernel().VFS().NewEpollInstanceFD(t) if err != nil { return 0, nil, err } - defer file.DecRef() + defer file.DecRef(t) fd, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{ CloseOnExec: flags&linux.EPOLL_CLOEXEC != 0, @@ -62,11 +62,11 @@ func EpollCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S return 0, nil, syserror.EINVAL } - file, err := t.Kernel().VFS().NewEpollInstanceFD() + file, err := t.Kernel().VFS().NewEpollInstanceFD(t) if err != nil { return 0, nil, err } - defer file.DecRef() + defer file.DecRef(t) fd, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{}) if err != nil { @@ -86,7 +86,7 @@ func EpollCtl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if epfile == nil { return 0, nil, syserror.EBADF } - defer epfile.DecRef() + defer epfile.DecRef(t) ep, ok := epfile.Impl().(*vfs.EpollInstance) if !ok { return 0, nil, syserror.EINVAL @@ -95,7 +95,7 @@ func EpollCtl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) if epfile == file { return 0, nil, syserror.EINVAL } @@ -135,7 +135,7 @@ func EpollWait(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if epfile == nil { return 0, nil, syserror.EBADF } - defer epfile.DecRef() + defer epfile.DecRef(t) ep, ok := epfile.Impl().(*vfs.EpollInstance) if !ok { return 0, nil, syserror.EINVAL diff --git a/pkg/sentry/syscalls/linux/vfs2/eventfd.go b/pkg/sentry/syscalls/linux/vfs2/eventfd.go index aff1a2070..807f909da 100644 --- a/pkg/sentry/syscalls/linux/vfs2/eventfd.go +++ b/pkg/sentry/syscalls/linux/vfs2/eventfd.go @@ -38,11 +38,11 @@ func Eventfd2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc fileFlags |= linux.O_NONBLOCK } semMode := flags&linux.EFD_SEMAPHORE != 0 - eventfd, err := eventfd.New(vfsObj, initVal, semMode, fileFlags) + eventfd, err := eventfd.New(t, vfsObj, initVal, semMode, fileFlags) if err != nil { return 0, nil, err } - defer eventfd.DecRef() + defer eventfd.DecRef(t) fd, err := t.NewFDFromVFS2(0, eventfd, kernel.FDFlags{ CloseOnExec: flags&linux.EFD_CLOEXEC != 0, diff --git a/pkg/sentry/syscalls/linux/vfs2/execve.go b/pkg/sentry/syscalls/linux/vfs2/execve.go index aef0078a8..066ee0863 100644 --- a/pkg/sentry/syscalls/linux/vfs2/execve.go +++ b/pkg/sentry/syscalls/linux/vfs2/execve.go @@ -71,7 +71,7 @@ func execveat(t *kernel.Task, dirfd int32, pathnameAddr, argvAddr, envvAddr user } root := t.FSContext().RootDirectoryVFS2() - defer root.DecRef() + defer root.DecRef(t) var executable fsbridge.File closeOnExec := false if path := fspath.Parse(pathname); dirfd != linux.AT_FDCWD && !path.Absolute { @@ -90,7 +90,7 @@ func execveat(t *kernel.Task, dirfd int32, pathnameAddr, argvAddr, envvAddr user } start := dirfile.VirtualDentry() start.IncRef() - dirfile.DecRef() + dirfile.DecRef(t) closeOnExec = dirfileFlags.CloseOnExec file, err := t.Kernel().VFS().OpenAt(t, t.Credentials(), &vfs.PathOperation{ Root: root, @@ -101,19 +101,19 @@ func execveat(t *kernel.Task, dirfd int32, pathnameAddr, argvAddr, envvAddr user Flags: linux.O_RDONLY, FileExec: true, }) - start.DecRef() + start.DecRef(t) if err != nil { return 0, nil, err } - defer file.DecRef() + defer file.DecRef(t) executable = fsbridge.NewVFSFile(file) } // Load the new TaskContext. mntns := t.MountNamespaceVFS2() // FIXME(jamieliu): useless refcount change - defer mntns.DecRef() + defer mntns.DecRef(t) wd := t.FSContext().WorkingDirectoryVFS2() - defer wd.DecRef() + defer wd.DecRef(t) remainingTraversals := uint(linux.MaxSymlinkTraversals) loadArgs := loader.LoadArgs{ Opener: fsbridge.NewVFSLookup(mntns, root, wd), diff --git a/pkg/sentry/syscalls/linux/vfs2/fd.go b/pkg/sentry/syscalls/linux/vfs2/fd.go index 67f191551..72ca916a0 100644 --- a/pkg/sentry/syscalls/linux/vfs2/fd.go +++ b/pkg/sentry/syscalls/linux/vfs2/fd.go @@ -38,7 +38,7 @@ func Close(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) err := file.OnClose(t) return 0, nil, slinux.HandleIOErrorVFS2(t, false /* partial */, err, syserror.EINTR, "close", file) @@ -52,7 +52,7 @@ func Dup(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) newFD, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{}) if err != nil { @@ -72,7 +72,7 @@ func Dup2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if file == nil { return 0, nil, syserror.EBADF } - file.DecRef() + file.DecRef(t) return uintptr(newfd), nil, nil } @@ -101,7 +101,7 @@ func dup3(t *kernel.Task, oldfd, newfd int32, flags uint32) (uintptr, *kernel.Sy if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) err := t.NewFDAtVFS2(newfd, file, kernel.FDFlags{ CloseOnExec: flags&linux.O_CLOEXEC != 0, @@ -121,7 +121,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) switch cmd { case linux.F_DUPFD, linux.F_DUPFD_CLOEXEC: @@ -332,7 +332,7 @@ func Fadvise64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // If the FD refers to a pipe or FIFO, return error. if _, isPipe := file.Impl().(*pipe.VFSPipeFD); isPipe { diff --git a/pkg/sentry/syscalls/linux/vfs2/filesystem.go b/pkg/sentry/syscalls/linux/vfs2/filesystem.go index b6d2ddd65..01e0f9010 100644 --- a/pkg/sentry/syscalls/linux/vfs2/filesystem.go +++ b/pkg/sentry/syscalls/linux/vfs2/filesystem.go @@ -56,7 +56,7 @@ func linkat(t *kernel.Task, olddirfd int32, oldpathAddr usermem.Addr, newdirfd i if err != nil { return err } - defer oldtpop.Release() + defer oldtpop.Release(t) newpath, err := copyInPath(t, newpathAddr) if err != nil { @@ -66,7 +66,7 @@ func linkat(t *kernel.Task, olddirfd int32, oldpathAddr usermem.Addr, newdirfd i if err != nil { return err } - defer newtpop.Release() + defer newtpop.Release(t) return t.Kernel().VFS().LinkAt(t, t.Credentials(), &oldtpop.pop, &newtpop.pop) } @@ -95,7 +95,7 @@ func mkdirat(t *kernel.Task, dirfd int32, addr usermem.Addr, mode uint) error { if err != nil { return err } - defer tpop.Release() + defer tpop.Release(t) return t.Kernel().VFS().MkdirAt(t, t.Credentials(), &tpop.pop, &vfs.MkdirOptions{ Mode: linux.FileMode(mode & (0777 | linux.S_ISVTX) &^ t.FSContext().Umask()), }) @@ -127,7 +127,7 @@ func mknodat(t *kernel.Task, dirfd int32, addr usermem.Addr, mode linux.FileMode if err != nil { return err } - defer tpop.Release() + defer tpop.Release(t) // "Zero file type is equivalent to type S_IFREG." - mknod(2) if mode.FileType() == 0 { @@ -174,7 +174,7 @@ func openat(t *kernel.Task, dirfd int32, pathAddr usermem.Addr, flags uint32, mo if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) file, err := t.Kernel().VFS().OpenAt(t, t.Credentials(), &tpop.pop, &vfs.OpenOptions{ Flags: flags | linux.O_LARGEFILE, @@ -183,7 +183,7 @@ func openat(t *kernel.Task, dirfd int32, pathAddr usermem.Addr, flags uint32, mo if err != nil { return 0, nil, err } - defer file.DecRef() + defer file.DecRef(t) fd, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{ CloseOnExec: flags&linux.O_CLOEXEC != 0, @@ -227,7 +227,7 @@ func renameat(t *kernel.Task, olddirfd int32, oldpathAddr usermem.Addr, newdirfd if err != nil { return err } - defer oldtpop.Release() + defer oldtpop.Release(t) newpath, err := copyInPath(t, newpathAddr) if err != nil { @@ -237,7 +237,7 @@ func renameat(t *kernel.Task, olddirfd int32, oldpathAddr usermem.Addr, newdirfd if err != nil { return err } - defer newtpop.Release() + defer newtpop.Release(t) return t.Kernel().VFS().RenameAt(t, t.Credentials(), &oldtpop.pop, &newtpop.pop, &vfs.RenameOptions{ Flags: flags, @@ -259,7 +259,7 @@ func rmdirat(t *kernel.Task, dirfd int32, pathAddr usermem.Addr) error { if err != nil { return err } - defer tpop.Release() + defer tpop.Release(t) return t.Kernel().VFS().RmdirAt(t, t.Credentials(), &tpop.pop) } @@ -278,7 +278,7 @@ func unlinkat(t *kernel.Task, dirfd int32, pathAddr usermem.Addr) error { if err != nil { return err } - defer tpop.Release() + defer tpop.Release(t) return t.Kernel().VFS().UnlinkAt(t, t.Credentials(), &tpop.pop) } @@ -329,6 +329,6 @@ func symlinkat(t *kernel.Task, targetAddr usermem.Addr, newdirfd int32, linkpath if err != nil { return err } - defer tpop.Release() + defer tpop.Release(t) return t.Kernel().VFS().SymlinkAt(t, t.Credentials(), &tpop.pop, target) } diff --git a/pkg/sentry/syscalls/linux/vfs2/fscontext.go b/pkg/sentry/syscalls/linux/vfs2/fscontext.go index 317409a18..a7d4d2a36 100644 --- a/pkg/sentry/syscalls/linux/vfs2/fscontext.go +++ b/pkg/sentry/syscalls/linux/vfs2/fscontext.go @@ -31,8 +31,8 @@ func Getcwd(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal root := t.FSContext().RootDirectoryVFS2() wd := t.FSContext().WorkingDirectoryVFS2() s, err := t.Kernel().VFS().PathnameForGetcwd(t, root, wd) - root.DecRef() - wd.DecRef() + root.DecRef(t) + wd.DecRef(t) if err != nil { return 0, nil, err } @@ -67,7 +67,7 @@ func Chdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) vd, err := t.Kernel().VFS().GetDentryAt(t, t.Credentials(), &tpop.pop, &vfs.GetDentryOptions{ CheckSearchable: true, @@ -75,8 +75,8 @@ func Chdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if err != nil { return 0, nil, err } - t.FSContext().SetWorkingDirectoryVFS2(vd) - vd.DecRef() + t.FSContext().SetWorkingDirectoryVFS2(t, vd) + vd.DecRef(t) return 0, nil, nil } @@ -88,7 +88,7 @@ func Fchdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) vd, err := t.Kernel().VFS().GetDentryAt(t, t.Credentials(), &tpop.pop, &vfs.GetDentryOptions{ CheckSearchable: true, @@ -96,8 +96,8 @@ func Fchdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, err } - t.FSContext().SetWorkingDirectoryVFS2(vd) - vd.DecRef() + t.FSContext().SetWorkingDirectoryVFS2(t, vd) + vd.DecRef(t) return 0, nil, nil } @@ -117,7 +117,7 @@ func Chroot(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) vd, err := t.Kernel().VFS().GetDentryAt(t, t.Credentials(), &tpop.pop, &vfs.GetDentryOptions{ CheckSearchable: true, @@ -125,7 +125,7 @@ func Chroot(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, err } - t.FSContext().SetRootDirectoryVFS2(vd) - vd.DecRef() + t.FSContext().SetRootDirectoryVFS2(t, vd) + vd.DecRef(t) return 0, nil, nil } diff --git a/pkg/sentry/syscalls/linux/vfs2/getdents.go b/pkg/sentry/syscalls/linux/vfs2/getdents.go index c7c7bf7ce..5517595b5 100644 --- a/pkg/sentry/syscalls/linux/vfs2/getdents.go +++ b/pkg/sentry/syscalls/linux/vfs2/getdents.go @@ -44,7 +44,7 @@ func getdents(t *kernel.Task, args arch.SyscallArguments, isGetdents64 bool) (ui if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) cb := getGetdentsCallback(t, addr, size, isGetdents64) err := file.IterDirents(t, cb) diff --git a/pkg/sentry/syscalls/linux/vfs2/inotify.go b/pkg/sentry/syscalls/linux/vfs2/inotify.go index 5d98134a5..11753d8e5 100644 --- a/pkg/sentry/syscalls/linux/vfs2/inotify.go +++ b/pkg/sentry/syscalls/linux/vfs2/inotify.go @@ -35,7 +35,7 @@ func InotifyInit1(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. if err != nil { return 0, nil, err } - defer ino.DecRef() + defer ino.DecRef(t) fd, err := t.NewFDFromVFS2(0, ino, kernel.FDFlags{ CloseOnExec: flags&linux.IN_CLOEXEC != 0, @@ -66,7 +66,7 @@ func fdToInotify(t *kernel.Task, fd int32) (*vfs.Inotify, *vfs.FileDescription, ino, ok := f.Impl().(*vfs.Inotify) if !ok { // Not an inotify fd. - f.DecRef() + f.DecRef(t) return nil, nil, syserror.EINVAL } @@ -96,7 +96,7 @@ func InotifyAddWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kern if err != nil { return 0, nil, err } - defer f.DecRef() + defer f.DecRef(t) path, err := copyInPath(t, addr) if err != nil { @@ -109,12 +109,12 @@ func InotifyAddWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kern if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) d, err := t.Kernel().VFS().GetDentryAt(t, t.Credentials(), &tpop.pop, &vfs.GetDentryOptions{}) if err != nil { return 0, nil, err } - defer d.DecRef() + defer d.DecRef(t) fd, err = ino.AddWatch(d.Dentry(), mask) if err != nil { @@ -132,6 +132,6 @@ func InotifyRmWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne if err != nil { return 0, nil, err } - defer f.DecRef() - return 0, nil, ino.RmWatch(wd) + defer f.DecRef(t) + return 0, nil, ino.RmWatch(t, wd) } diff --git a/pkg/sentry/syscalls/linux/vfs2/ioctl.go b/pkg/sentry/syscalls/linux/vfs2/ioctl.go index fd6ab94b2..38778a388 100644 --- a/pkg/sentry/syscalls/linux/vfs2/ioctl.go +++ b/pkg/sentry/syscalls/linux/vfs2/ioctl.go @@ -29,7 +29,7 @@ func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Handle ioctls that apply to all FDs. switch args[1].Int() { diff --git a/pkg/sentry/syscalls/linux/vfs2/lock.go b/pkg/sentry/syscalls/linux/vfs2/lock.go index bf19028c4..b910b5a74 100644 --- a/pkg/sentry/syscalls/linux/vfs2/lock.go +++ b/pkg/sentry/syscalls/linux/vfs2/lock.go @@ -32,7 +32,7 @@ func Flock(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // flock(2): EBADF fd is not an open file descriptor. return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) nonblocking := operation&linux.LOCK_NB != 0 operation &^= linux.LOCK_NB diff --git a/pkg/sentry/syscalls/linux/vfs2/memfd.go b/pkg/sentry/syscalls/linux/vfs2/memfd.go index bbe248d17..519583e4e 100644 --- a/pkg/sentry/syscalls/linux/vfs2/memfd.go +++ b/pkg/sentry/syscalls/linux/vfs2/memfd.go @@ -47,7 +47,7 @@ func MemfdCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S } shmMount := t.Kernel().ShmMount() - file, err := tmpfs.NewMemfd(shmMount, t.Credentials(), allowSeals, memfdPrefix+name) + file, err := tmpfs.NewMemfd(t, t.Credentials(), shmMount, allowSeals, memfdPrefix+name) if err != nil { return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/vfs2/mmap.go b/pkg/sentry/syscalls/linux/vfs2/mmap.go index 60a43f0a0..dc05c2994 100644 --- a/pkg/sentry/syscalls/linux/vfs2/mmap.go +++ b/pkg/sentry/syscalls/linux/vfs2/mmap.go @@ -61,7 +61,7 @@ func Mmap(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC } defer func() { if opts.MappingIdentity != nil { - opts.MappingIdentity.DecRef() + opts.MappingIdentity.DecRef(t) } }() @@ -71,7 +71,7 @@ func Mmap(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // mmap unconditionally requires that the FD is readable. if !file.IsReadable() { diff --git a/pkg/sentry/syscalls/linux/vfs2/mount.go b/pkg/sentry/syscalls/linux/vfs2/mount.go index ea337de7c..4bd5c7ca2 100644 --- a/pkg/sentry/syscalls/linux/vfs2/mount.go +++ b/pkg/sentry/syscalls/linux/vfs2/mount.go @@ -108,7 +108,7 @@ func Mount(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if err != nil { return 0, nil, err } - defer target.Release() + defer target.Release(t) return 0, nil, t.Kernel().VFS().MountAt(t, creds, source, &target.pop, fsType, &opts) } @@ -140,7 +140,7 @@ func Umount2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) opts := vfs.UmountOptions{ Flags: uint32(flags), diff --git a/pkg/sentry/syscalls/linux/vfs2/path.go b/pkg/sentry/syscalls/linux/vfs2/path.go index 97da6c647..90a511d9a 100644 --- a/pkg/sentry/syscalls/linux/vfs2/path.go +++ b/pkg/sentry/syscalls/linux/vfs2/path.go @@ -42,7 +42,7 @@ func getTaskPathOperation(t *kernel.Task, dirfd int32, path fspath.Path, shouldA haveStartRef := false if !path.Absolute { if !path.HasComponents() && !bool(shouldAllowEmptyPath) { - root.DecRef() + root.DecRef(t) return taskPathOperation{}, syserror.ENOENT } if dirfd == linux.AT_FDCWD { @@ -51,13 +51,13 @@ func getTaskPathOperation(t *kernel.Task, dirfd int32, path fspath.Path, shouldA } else { dirfile := t.GetFileVFS2(dirfd) if dirfile == nil { - root.DecRef() + root.DecRef(t) return taskPathOperation{}, syserror.EBADF } start = dirfile.VirtualDentry() start.IncRef() haveStartRef = true - dirfile.DecRef() + dirfile.DecRef(t) } } return taskPathOperation{ @@ -71,10 +71,10 @@ func getTaskPathOperation(t *kernel.Task, dirfd int32, path fspath.Path, shouldA }, nil } -func (tpop *taskPathOperation) Release() { - tpop.pop.Root.DecRef() +func (tpop *taskPathOperation) Release(t *kernel.Task) { + tpop.pop.Root.DecRef(t) if tpop.haveStartRef { - tpop.pop.Start.DecRef() + tpop.pop.Start.DecRef(t) tpop.haveStartRef = false } } diff --git a/pkg/sentry/syscalls/linux/vfs2/pipe.go b/pkg/sentry/syscalls/linux/vfs2/pipe.go index 4a01e4209..9b4848d9e 100644 --- a/pkg/sentry/syscalls/linux/vfs2/pipe.go +++ b/pkg/sentry/syscalls/linux/vfs2/pipe.go @@ -42,8 +42,8 @@ func pipe2(t *kernel.Task, addr usermem.Addr, flags int32) error { return syserror.EINVAL } r, w := pipefs.NewConnectedPipeFDs(t, t.Kernel().PipeMount(), uint32(flags&linux.O_NONBLOCK)) - defer r.DecRef() - defer w.DecRef() + defer r.DecRef(t) + defer w.DecRef(t) fds, err := t.NewFDsVFS2(0, []*vfs.FileDescription{r, w}, kernel.FDFlags{ CloseOnExec: flags&linux.O_CLOEXEC != 0, @@ -54,7 +54,7 @@ func pipe2(t *kernel.Task, addr usermem.Addr, flags int32) error { if _, err := t.CopyOut(addr, fds); err != nil { for _, fd := range fds { if _, file := t.FDTable().Remove(fd); file != nil { - file.DecRef() + file.DecRef(t) } } return err diff --git a/pkg/sentry/syscalls/linux/vfs2/poll.go b/pkg/sentry/syscalls/linux/vfs2/poll.go index ff1b25d7b..7b9d5e18a 100644 --- a/pkg/sentry/syscalls/linux/vfs2/poll.go +++ b/pkg/sentry/syscalls/linux/vfs2/poll.go @@ -73,7 +73,7 @@ func initReadiness(t *kernel.Task, pfd *linux.PollFD, state *pollState, ch chan } if ch == nil { - defer file.DecRef() + defer file.DecRef(t) } else { state.file = file state.waiter, _ = waiter.NewChannelEntry(ch) @@ -85,11 +85,11 @@ func initReadiness(t *kernel.Task, pfd *linux.PollFD, state *pollState, ch chan } // releaseState releases all the pollState in "state". -func releaseState(state []pollState) { +func releaseState(t *kernel.Task, state []pollState) { for i := range state { if state[i].file != nil { state[i].file.EventUnregister(&state[i].waiter) - state[i].file.DecRef() + state[i].file.DecRef(t) } } } @@ -110,7 +110,7 @@ func pollBlock(t *kernel.Task, pfd []linux.PollFD, timeout time.Duration) (time. // result, we stop registering for events but still go through all files // to get their ready masks. state := make([]pollState, len(pfd)) - defer releaseState(state) + defer releaseState(t, state) n := uintptr(0) for i := range pfd { initReadiness(t, &pfd[i], &state[i], ch) @@ -269,7 +269,7 @@ func doSelect(t *kernel.Task, nfds int, readFDs, writeFDs, exceptFDs usermem.Add if file == nil { return 0, syserror.EBADF } - file.DecRef() + file.DecRef(t) var mask int16 if (rV & m) != 0 { diff --git a/pkg/sentry/syscalls/linux/vfs2/read_write.go b/pkg/sentry/syscalls/linux/vfs2/read_write.go index cd25597a7..a905dae0a 100644 --- a/pkg/sentry/syscalls/linux/vfs2/read_write.go +++ b/pkg/sentry/syscalls/linux/vfs2/read_write.go @@ -44,7 +44,7 @@ func Read(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the size is legitimate. si := int(size) @@ -75,7 +75,7 @@ func Readv(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Get the destination of the read. dst, err := t.IovecsIOSequence(addr, iovcnt, usermem.IOOpts{ @@ -94,7 +94,7 @@ func read(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, opt n, err := file.Read(t, dst, opts) if err != syserror.ErrWouldBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) } return n, err } @@ -102,7 +102,7 @@ func read(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, opt allowBlock, deadline, hasDeadline := blockPolicy(t, file) if !allowBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) } return n, err } @@ -135,7 +135,7 @@ func read(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, opt file.EventUnregister(&w) if total > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) } return total, err } @@ -151,7 +151,7 @@ func Pread64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate and does not overflow. if offset < 0 || offset+int64(size) < 0 { @@ -188,7 +188,7 @@ func Preadv(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate. if offset < 0 { @@ -226,7 +226,7 @@ func Preadv2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate. if offset < -1 { @@ -258,7 +258,7 @@ func pread(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, of n, err := file.PRead(t, dst, offset, opts) if err != syserror.ErrWouldBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) } return n, err } @@ -266,7 +266,7 @@ func pread(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, of allowBlock, deadline, hasDeadline := blockPolicy(t, file) if !allowBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) } return n, err } @@ -299,7 +299,7 @@ func pread(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, of file.EventUnregister(&w) if total > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) } return total, err } @@ -314,7 +314,7 @@ func Write(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the size is legitimate. si := int(size) @@ -345,7 +345,7 @@ func Writev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Get the source of the write. src, err := t.IovecsIOSequence(addr, iovcnt, usermem.IOOpts{ @@ -364,7 +364,7 @@ func write(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, op n, err := file.Write(t, src, opts) if err != syserror.ErrWouldBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) } return n, err } @@ -372,7 +372,7 @@ func write(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, op allowBlock, deadline, hasDeadline := blockPolicy(t, file) if !allowBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) } return n, err } @@ -405,7 +405,7 @@ func write(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, op file.EventUnregister(&w) if total > 0 { - file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) } return total, err } @@ -421,7 +421,7 @@ func Pwrite64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate and does not overflow. if offset < 0 || offset+int64(size) < 0 { @@ -458,7 +458,7 @@ func Pwritev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate. if offset < 0 { @@ -496,7 +496,7 @@ func Pwritev2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate. if offset < -1 { @@ -528,7 +528,7 @@ func pwrite(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, o n, err := file.PWrite(t, src, offset, opts) if err != syserror.ErrWouldBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) } return n, err } @@ -536,7 +536,7 @@ func pwrite(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, o allowBlock, deadline, hasDeadline := blockPolicy(t, file) if !allowBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) } return n, err } @@ -569,7 +569,7 @@ func pwrite(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, o file.EventUnregister(&w) if total > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) } return total, err } @@ -601,7 +601,7 @@ func Lseek(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) newoff, err := file.Seek(t, offset, whence) return uintptr(newoff), nil, err @@ -617,7 +617,7 @@ func Readahead(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the file is readable. if !file.IsReadable() { diff --git a/pkg/sentry/syscalls/linux/vfs2/setstat.go b/pkg/sentry/syscalls/linux/vfs2/setstat.go index 25cdb7a55..5e6eb13ba 100644 --- a/pkg/sentry/syscalls/linux/vfs2/setstat.go +++ b/pkg/sentry/syscalls/linux/vfs2/setstat.go @@ -66,7 +66,7 @@ func Fchmod(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return 0, nil, file.SetStat(t, vfs.SetStatOptions{ Stat: linux.Statx{ @@ -151,7 +151,7 @@ func Fchown(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) var opts vfs.SetStatOptions if err := populateSetStatOptionsForChown(t, owner, group, &opts); err != nil { @@ -197,7 +197,7 @@ func Ftruncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) if !file.IsWritable() { return 0, nil, syserror.EINVAL @@ -224,7 +224,7 @@ func Fallocate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) if !file.IsWritable() { return 0, nil, syserror.EBADF @@ -258,7 +258,7 @@ func Fallocate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys return 0, nil, err } - file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) return 0, nil, nil } @@ -438,7 +438,7 @@ func populateSetStatOptionsForUtimens(t *kernel.Task, timesAddr usermem.Addr, op func setstatat(t *kernel.Task, dirfd int32, path fspath.Path, shouldAllowEmptyPath shouldAllowEmptyPath, shouldFollowFinalSymlink shouldFollowFinalSymlink, opts *vfs.SetStatOptions) error { root := t.FSContext().RootDirectoryVFS2() - defer root.DecRef() + defer root.DecRef(t) start := root if !path.Absolute { if !path.HasComponents() && !bool(shouldAllowEmptyPath) { @@ -446,7 +446,7 @@ func setstatat(t *kernel.Task, dirfd int32, path fspath.Path, shouldAllowEmptyPa } if dirfd == linux.AT_FDCWD { start = t.FSContext().WorkingDirectoryVFS2() - defer start.DecRef() + defer start.DecRef(t) } else { dirfile := t.GetFileVFS2(dirfd) if dirfile == nil { @@ -457,13 +457,13 @@ func setstatat(t *kernel.Task, dirfd int32, path fspath.Path, shouldAllowEmptyPa // VirtualFilesystem.SetStatAt(), since the former may be able // to use opened file state to expedite the SetStat. err := dirfile.SetStat(t, *opts) - dirfile.DecRef() + dirfile.DecRef(t) return err } start = dirfile.VirtualDentry() start.IncRef() - defer start.DecRef() - dirfile.DecRef() + defer start.DecRef(t) + dirfile.DecRef(t) } } return t.Kernel().VFS().SetStatAt(t, t.Credentials(), &vfs.PathOperation{ diff --git a/pkg/sentry/syscalls/linux/vfs2/signal.go b/pkg/sentry/syscalls/linux/vfs2/signal.go index 623992f6f..b89f34cdb 100644 --- a/pkg/sentry/syscalls/linux/vfs2/signal.go +++ b/pkg/sentry/syscalls/linux/vfs2/signal.go @@ -45,7 +45,7 @@ func sharedSignalfd(t *kernel.Task, fd int32, sigset usermem.Addr, sigsetsize ui if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Is this a signalfd? if sfd, ok := file.Impl().(*signalfd.SignalFileDescription); ok { @@ -68,7 +68,7 @@ func sharedSignalfd(t *kernel.Task, fd int32, sigset usermem.Addr, sigsetsize ui if err != nil { return 0, nil, err } - defer file.DecRef() + defer file.DecRef(t) // Create a new descriptor. fd, err = t.NewFDFromVFS2(0, file, kernel.FDFlags{ diff --git a/pkg/sentry/syscalls/linux/vfs2/socket.go b/pkg/sentry/syscalls/linux/vfs2/socket.go index 8096a8f9c..4a68c64f3 100644 --- a/pkg/sentry/syscalls/linux/vfs2/socket.go +++ b/pkg/sentry/syscalls/linux/vfs2/socket.go @@ -196,7 +196,7 @@ func Socket(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if e != nil { return 0, nil, e.ToError() } - defer s.DecRef() + defer s.DecRef(t) if err := s.SetStatusFlags(t, t.Credentials(), uint32(stype&linux.SOCK_NONBLOCK)); err != nil { return 0, nil, err @@ -230,8 +230,8 @@ func SocketPair(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy return 0, nil, e.ToError() } // Adding to the FD table will cause an extra reference to be acquired. - defer s1.DecRef() - defer s2.DecRef() + defer s1.DecRef(t) + defer s2.DecRef(t) nonblocking := uint32(stype & linux.SOCK_NONBLOCK) if err := s1.SetStatusFlags(t, t.Credentials(), nonblocking); err != nil { @@ -253,7 +253,7 @@ func SocketPair(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy if _, err := t.CopyOut(addr, fds); err != nil { for _, fd := range fds { if _, file := t.FDTable().Remove(fd); file != nil { - file.DecRef() + file.DecRef(t) } } return 0, nil, err @@ -273,7 +273,7 @@ func Connect(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -304,7 +304,7 @@ func accept(t *kernel.Task, fd int32, addr usermem.Addr, addrLen usermem.Addr, f if file == nil { return 0, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -363,7 +363,7 @@ func Bind(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -390,7 +390,7 @@ func Listen(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -419,7 +419,7 @@ func Shutdown(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -450,7 +450,7 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -532,7 +532,7 @@ func SetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -570,7 +570,7 @@ func GetSockName(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -598,7 +598,7 @@ func GetPeerName(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -631,7 +631,7 @@ func RecvMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -684,7 +684,7 @@ func RecvMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -778,7 +778,7 @@ func recvSingleMsg(t *kernel.Task, s socket.SocketVFS2, msgPtr usermem.Addr, fla } if !cms.Unix.Empty() { mflags |= linux.MSG_CTRUNC - cms.Release() + cms.Release(t) } if int(msg.Flags) != mflags { @@ -798,7 +798,7 @@ func recvSingleMsg(t *kernel.Task, s socket.SocketVFS2, msgPtr usermem.Addr, fla if e != nil { return 0, syserror.ConvertIntr(e.ToError(), kernel.ERESTARTSYS) } - defer cms.Release() + defer cms.Release(t) controlData := make([]byte, 0, msg.ControlLen) controlData = control.PackControlMessages(t, cms, controlData) @@ -854,7 +854,7 @@ func recvFrom(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flag if file == nil { return 0, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -883,7 +883,7 @@ func recvFrom(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flag } n, _, sender, senderLen, cm, e := s.RecvMsg(t, dst, int(flags), haveDeadline, deadline, nameLenPtr != 0, 0) - cm.Release() + cm.Release(t) if e != nil { return 0, syserror.ConvertIntr(e.ToError(), kernel.ERESTARTSYS) } @@ -927,7 +927,7 @@ func SendMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -965,7 +965,7 @@ func SendMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -1069,7 +1069,7 @@ func sendSingleMsg(t *kernel.Task, s socket.SocketVFS2, file *vfs.FileDescriptio n, e := s.SendMsg(t, src, to, int(flags), haveDeadline, deadline, controlMessages) err = slinux.HandleIOErrorVFS2(t, n != 0, e.ToError(), kernel.ERESTARTSYS, "sendmsg", file) if err != nil { - controlMessages.Release() + controlMessages.Release(t) } return uintptr(n), err } @@ -1087,7 +1087,7 @@ func sendTo(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flags if file == nil { return 0, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) diff --git a/pkg/sentry/syscalls/linux/vfs2/splice.go b/pkg/sentry/syscalls/linux/vfs2/splice.go index 63ab11f8c..16f59fce9 100644 --- a/pkg/sentry/syscalls/linux/vfs2/splice.go +++ b/pkg/sentry/syscalls/linux/vfs2/splice.go @@ -53,12 +53,12 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if inFile == nil { return 0, nil, syserror.EBADF } - defer inFile.DecRef() + defer inFile.DecRef(t) outFile := t.GetFileVFS2(outFD) if outFile == nil { return 0, nil, syserror.EBADF } - defer outFile.DecRef() + defer outFile.DecRef(t) // Check that both files support the required directionality. if !inFile.IsReadable() || !outFile.IsWritable() { @@ -175,7 +175,7 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal // On Linux, inotify behavior is not very consistent with splice(2). We try // our best to emulate Linux for very basic calls to splice, where for some // reason, events are generated for output files, but not input files. - outFile.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) + outFile.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) return uintptr(n), nil, nil } @@ -203,12 +203,12 @@ func Tee(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo if inFile == nil { return 0, nil, syserror.EBADF } - defer inFile.DecRef() + defer inFile.DecRef(t) outFile := t.GetFileVFS2(outFD) if outFile == nil { return 0, nil, syserror.EBADF } - defer outFile.DecRef() + defer outFile.DecRef(t) // Check that both files support the required directionality. if !inFile.IsReadable() || !outFile.IsWritable() { @@ -251,7 +251,7 @@ func Tee(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo if n == 0 { return 0, nil, err } - outFile.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) + outFile.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) return uintptr(n), nil, nil } @@ -266,7 +266,7 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if inFile == nil { return 0, nil, syserror.EBADF } - defer inFile.DecRef() + defer inFile.DecRef(t) if !inFile.IsReadable() { return 0, nil, syserror.EBADF } @@ -275,7 +275,7 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if outFile == nil { return 0, nil, syserror.EBADF } - defer outFile.DecRef() + defer outFile.DecRef(t) if !outFile.IsWritable() { return 0, nil, syserror.EBADF } @@ -419,8 +419,8 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc return 0, nil, err } - inFile.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) - outFile.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) + inFile.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) + outFile.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) return uintptr(n), nil, nil } diff --git a/pkg/sentry/syscalls/linux/vfs2/stat.go b/pkg/sentry/syscalls/linux/vfs2/stat.go index bb1d5cac4..0f5d5189c 100644 --- a/pkg/sentry/syscalls/linux/vfs2/stat.go +++ b/pkg/sentry/syscalls/linux/vfs2/stat.go @@ -65,7 +65,7 @@ func fstatat(t *kernel.Task, dirfd int32, pathAddr, statAddr usermem.Addr, flags } root := t.FSContext().RootDirectoryVFS2() - defer root.DecRef() + defer root.DecRef(t) start := root if !path.Absolute { if !path.HasComponents() && flags&linux.AT_EMPTY_PATH == 0 { @@ -73,7 +73,7 @@ func fstatat(t *kernel.Task, dirfd int32, pathAddr, statAddr usermem.Addr, flags } if dirfd == linux.AT_FDCWD { start = t.FSContext().WorkingDirectoryVFS2() - defer start.DecRef() + defer start.DecRef(t) } else { dirfile := t.GetFileVFS2(dirfd) if dirfile == nil { @@ -85,7 +85,7 @@ func fstatat(t *kernel.Task, dirfd int32, pathAddr, statAddr usermem.Addr, flags // former may be able to use opened file state to expedite the // Stat. statx, err := dirfile.Stat(t, opts) - dirfile.DecRef() + dirfile.DecRef(t) if err != nil { return err } @@ -96,8 +96,8 @@ func fstatat(t *kernel.Task, dirfd int32, pathAddr, statAddr usermem.Addr, flags } start = dirfile.VirtualDentry() start.IncRef() - defer start.DecRef() - dirfile.DecRef() + defer start.DecRef(t) + dirfile.DecRef(t) } } @@ -132,7 +132,7 @@ func Fstat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) statx, err := file.Stat(t, vfs.StatOptions{ Mask: linux.STATX_BASIC_STATS, @@ -177,7 +177,7 @@ func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } root := t.FSContext().RootDirectoryVFS2() - defer root.DecRef() + defer root.DecRef(t) start := root if !path.Absolute { if !path.HasComponents() && flags&linux.AT_EMPTY_PATH == 0 { @@ -185,7 +185,7 @@ func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } if dirfd == linux.AT_FDCWD { start = t.FSContext().WorkingDirectoryVFS2() - defer start.DecRef() + defer start.DecRef(t) } else { dirfile := t.GetFileVFS2(dirfd) if dirfile == nil { @@ -197,7 +197,7 @@ func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // former may be able to use opened file state to expedite the // Stat. statx, err := dirfile.Stat(t, opts) - dirfile.DecRef() + dirfile.DecRef(t) if err != nil { return 0, nil, err } @@ -207,8 +207,8 @@ func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } start = dirfile.VirtualDentry() start.IncRef() - defer start.DecRef() - dirfile.DecRef() + defer start.DecRef(t) + dirfile.DecRef(t) } } @@ -282,7 +282,7 @@ func accessAt(t *kernel.Task, dirfd int32, pathAddr usermem.Addr, mode uint) err if err != nil { return err } - defer tpop.Release() + defer tpop.Release(t) // access(2) and faccessat(2) check permissions using real // UID/GID, not effective UID/GID. @@ -328,7 +328,7 @@ func readlinkat(t *kernel.Task, dirfd int32, pathAddr, bufAddr usermem.Addr, siz if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) target, err := t.Kernel().VFS().ReadlinkAt(t, t.Credentials(), &tpop.pop) if err != nil { @@ -358,7 +358,7 @@ func Statfs(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) statfs, err := t.Kernel().VFS().StatFSAt(t, t.Credentials(), &tpop.pop) if err != nil { @@ -377,7 +377,7 @@ func Fstatfs(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) statfs, err := t.Kernel().VFS().StatFSAt(t, t.Credentials(), &tpop.pop) if err != nil { diff --git a/pkg/sentry/syscalls/linux/vfs2/sync.go b/pkg/sentry/syscalls/linux/vfs2/sync.go index 0d0ebf46a..a6491ac37 100644 --- a/pkg/sentry/syscalls/linux/vfs2/sync.go +++ b/pkg/sentry/syscalls/linux/vfs2/sync.go @@ -34,7 +34,7 @@ func Syncfs(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return 0, nil, file.SyncFS(t) } @@ -47,7 +47,7 @@ func Fsync(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return 0, nil, file.Sync(t) } @@ -77,7 +77,7 @@ func SyncFileRange(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // TODO(gvisor.dev/issue/1897): Currently, the only file syncing we support // is a full-file sync, i.e. fsync(2). As a result, there are severe diff --git a/pkg/sentry/syscalls/linux/vfs2/timerfd.go b/pkg/sentry/syscalls/linux/vfs2/timerfd.go index 5ac79bc09..7a26890ef 100644 --- a/pkg/sentry/syscalls/linux/vfs2/timerfd.go +++ b/pkg/sentry/syscalls/linux/vfs2/timerfd.go @@ -50,11 +50,11 @@ func TimerfdCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel return 0, nil, syserror.EINVAL } vfsObj := t.Kernel().VFS() - file, err := timerfd.New(vfsObj, clock, fileFlags) + file, err := timerfd.New(t, vfsObj, clock, fileFlags) if err != nil { return 0, nil, err } - defer file.DecRef() + defer file.DecRef(t) fd, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{ CloseOnExec: flags&linux.TFD_CLOEXEC != 0, }) @@ -79,7 +79,7 @@ func TimerfdSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) tfd, ok := file.Impl().(*timerfd.TimerFileDescription) if !ok { @@ -113,7 +113,7 @@ func TimerfdGettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) tfd, ok := file.Impl().(*timerfd.TimerFileDescription) if !ok { diff --git a/pkg/sentry/syscalls/linux/vfs2/xattr.go b/pkg/sentry/syscalls/linux/vfs2/xattr.go index af455d5c1..ef99246ed 100644 --- a/pkg/sentry/syscalls/linux/vfs2/xattr.go +++ b/pkg/sentry/syscalls/linux/vfs2/xattr.go @@ -49,7 +49,7 @@ func listxattr(t *kernel.Task, args arch.SyscallArguments, shouldFollowFinalSyml if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) names, err := t.Kernel().VFS().ListxattrAt(t, t.Credentials(), &tpop.pop, uint64(size)) if err != nil { @@ -72,7 +72,7 @@ func Flistxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) names, err := file.Listxattr(t, uint64(size)) if err != nil { @@ -109,7 +109,7 @@ func getxattr(t *kernel.Task, args arch.SyscallArguments, shouldFollowFinalSymli if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) name, err := copyInXattrName(t, nameAddr) if err != nil { @@ -141,7 +141,7 @@ func Fgetxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) name, err := copyInXattrName(t, nameAddr) if err != nil { @@ -188,7 +188,7 @@ func setxattr(t *kernel.Task, args arch.SyscallArguments, shouldFollowFinalSymli if err != nil { return err } - defer tpop.Release() + defer tpop.Release(t) name, err := copyInXattrName(t, nameAddr) if err != nil { @@ -222,7 +222,7 @@ func Fsetxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) name, err := copyInXattrName(t, nameAddr) if err != nil { @@ -262,7 +262,7 @@ func removexattr(t *kernel.Task, args arch.SyscallArguments, shouldFollowFinalSy if err != nil { return err } - defer tpop.Release() + defer tpop.Release(t) name, err := copyInXattrName(t, nameAddr) if err != nil { @@ -281,7 +281,7 @@ func Fremovexattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) name, err := copyInXattrName(t, nameAddr) if err != nil { diff --git a/pkg/sentry/vfs/anonfs.go b/pkg/sentry/vfs/anonfs.go index 641e3e502..5a0e3e6b5 100644 --- a/pkg/sentry/vfs/anonfs.go +++ b/pkg/sentry/vfs/anonfs.go @@ -82,7 +82,7 @@ type anonDentry struct { } // Release implements FilesystemImpl.Release. -func (fs *anonFilesystem) Release() { +func (fs *anonFilesystem) Release(ctx context.Context) { } // Sync implements FilesystemImpl.Sync. @@ -294,7 +294,7 @@ func (d *anonDentry) TryIncRef() bool { } // DecRef implements DentryImpl.DecRef. -func (d *anonDentry) DecRef() { +func (d *anonDentry) DecRef(ctx context.Context) { // no-op } @@ -303,7 +303,7 @@ func (d *anonDentry) DecRef() { // Although Linux technically supports inotify on pseudo filesystems (inotify // is implemented at the vfs layer), it is not particularly useful. It is left // unimplemented until someone actually needs it. -func (d *anonDentry) InotifyWithParent(events, cookie uint32, et EventType) {} +func (d *anonDentry) InotifyWithParent(ctx context.Context, events, cookie uint32, et EventType) {} // Watches implements DentryImpl.Watches. func (d *anonDentry) Watches() *Watches { @@ -311,4 +311,4 @@ func (d *anonDentry) Watches() *Watches { } // OnZeroWatches implements Dentry.OnZeroWatches. -func (d *anonDentry) OnZeroWatches() {} +func (d *anonDentry) OnZeroWatches(context.Context) {} diff --git a/pkg/sentry/vfs/dentry.go b/pkg/sentry/vfs/dentry.go index cea3e6955..bc7ea93ea 100644 --- a/pkg/sentry/vfs/dentry.go +++ b/pkg/sentry/vfs/dentry.go @@ -17,6 +17,7 @@ package vfs import ( "sync/atomic" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) @@ -102,7 +103,7 @@ type DentryImpl interface { TryIncRef() bool // DecRef decrements the Dentry's reference count. - DecRef() + DecRef(ctx context.Context) // InotifyWithParent notifies all watches on the targets represented by this // dentry and its parent. The parent's watches are notified first, followed @@ -113,7 +114,7 @@ type DentryImpl interface { // // Note that the events may not actually propagate up to the user, depending // on the event masks. - InotifyWithParent(events, cookie uint32, et EventType) + InotifyWithParent(ctx context.Context, events, cookie uint32, et EventType) // Watches returns the set of inotify watches for the file corresponding to // the Dentry. Dentries that are hard links to the same underlying file @@ -135,7 +136,7 @@ type DentryImpl interface { // The caller does not need to hold a reference on the dentry. OnZeroWatches // may acquire inotify locks, so to prevent deadlock, no inotify locks should // be held by the caller. - OnZeroWatches() + OnZeroWatches(ctx context.Context) } // IncRef increments d's reference count. @@ -150,8 +151,8 @@ func (d *Dentry) TryIncRef() bool { } // DecRef decrements d's reference count. -func (d *Dentry) DecRef() { - d.impl.DecRef() +func (d *Dentry) DecRef(ctx context.Context) { + d.impl.DecRef(ctx) } // IsDead returns true if d has been deleted or invalidated by its owning @@ -168,8 +169,8 @@ func (d *Dentry) isMounted() bool { // InotifyWithParent notifies all watches on the targets represented by d and // its parent of events. -func (d *Dentry) InotifyWithParent(events, cookie uint32, et EventType) { - d.impl.InotifyWithParent(events, cookie, et) +func (d *Dentry) InotifyWithParent(ctx context.Context, events, cookie uint32, et EventType) { + d.impl.InotifyWithParent(ctx, events, cookie, et) } // Watches returns the set of inotify watches associated with d. @@ -182,8 +183,8 @@ func (d *Dentry) Watches() *Watches { // OnZeroWatches performs cleanup tasks whenever the number of watches on a // dentry drops to zero. -func (d *Dentry) OnZeroWatches() { - d.impl.OnZeroWatches() +func (d *Dentry) OnZeroWatches(ctx context.Context) { + d.impl.OnZeroWatches(ctx) } // The following functions are exported so that filesystem implementations can @@ -214,11 +215,11 @@ func (vfs *VirtualFilesystem) AbortDeleteDentry(d *Dentry) { // CommitDeleteDentry must be called after PrepareDeleteDentry if the deletion // succeeds. -func (vfs *VirtualFilesystem) CommitDeleteDentry(d *Dentry) { +func (vfs *VirtualFilesystem) CommitDeleteDentry(ctx context.Context, d *Dentry) { d.dead = true d.mu.Unlock() if d.isMounted() { - vfs.forgetDeadMountpoint(d) + vfs.forgetDeadMountpoint(ctx, d) } } @@ -226,12 +227,12 @@ func (vfs *VirtualFilesystem) CommitDeleteDentry(d *Dentry) { // did for reasons outside of VFS' control (e.g. d represents the local state // of a file on a remote filesystem on which the file has already been // deleted). -func (vfs *VirtualFilesystem) InvalidateDentry(d *Dentry) { +func (vfs *VirtualFilesystem) InvalidateDentry(ctx context.Context, d *Dentry) { d.mu.Lock() d.dead = true d.mu.Unlock() if d.isMounted() { - vfs.forgetDeadMountpoint(d) + vfs.forgetDeadMountpoint(ctx, d) } } @@ -278,13 +279,13 @@ func (vfs *VirtualFilesystem) AbortRenameDentry(from, to *Dentry) { // that was replaced by from. // // Preconditions: PrepareRenameDentry was previously called on from and to. -func (vfs *VirtualFilesystem) CommitRenameReplaceDentry(from, to *Dentry) { +func (vfs *VirtualFilesystem) CommitRenameReplaceDentry(ctx context.Context, from, to *Dentry) { from.mu.Unlock() if to != nil { to.dead = true to.mu.Unlock() if to.isMounted() { - vfs.forgetDeadMountpoint(to) + vfs.forgetDeadMountpoint(ctx, to) } } } @@ -303,7 +304,7 @@ func (vfs *VirtualFilesystem) CommitRenameExchangeDentry(from, to *Dentry) { // // forgetDeadMountpoint is analogous to Linux's // fs/namespace.c:__detach_mounts(). -func (vfs *VirtualFilesystem) forgetDeadMountpoint(d *Dentry) { +func (vfs *VirtualFilesystem) forgetDeadMountpoint(ctx context.Context, d *Dentry) { var ( vdsToDecRef []VirtualDentry mountsToDecRef []*Mount @@ -316,9 +317,9 @@ func (vfs *VirtualFilesystem) forgetDeadMountpoint(d *Dentry) { vfs.mounts.seq.EndWrite() vfs.mountMu.Unlock() for _, vd := range vdsToDecRef { - vd.DecRef() + vd.DecRef(ctx) } for _, mnt := range mountsToDecRef { - mnt.DecRef() + mnt.DecRef(ctx) } } diff --git a/pkg/sentry/vfs/epoll.go b/pkg/sentry/vfs/epoll.go index 5b009b928..1b5af9f73 100644 --- a/pkg/sentry/vfs/epoll.go +++ b/pkg/sentry/vfs/epoll.go @@ -93,9 +93,9 @@ type epollInterest struct { // NewEpollInstanceFD returns a FileDescription representing a new epoll // instance. A reference is taken on the returned FileDescription. -func (vfs *VirtualFilesystem) NewEpollInstanceFD() (*FileDescription, error) { +func (vfs *VirtualFilesystem) NewEpollInstanceFD(ctx context.Context) (*FileDescription, error) { vd := vfs.NewAnonVirtualDentry("[eventpoll]") - defer vd.DecRef() + defer vd.DecRef(ctx) ep := &EpollInstance{ interest: make(map[epollInterestKey]*epollInterest), } @@ -110,7 +110,7 @@ func (vfs *VirtualFilesystem) NewEpollInstanceFD() (*FileDescription, error) { } // Release implements FileDescriptionImpl.Release. -func (ep *EpollInstance) Release() { +func (ep *EpollInstance) Release(ctx context.Context) { // Unregister all polled fds. ep.interestMu.Lock() defer ep.interestMu.Unlock() diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 93861fb4a..576ab3920 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -171,7 +171,7 @@ func (fd *FileDescription) TryIncRef() bool { } // DecRef decrements fd's reference count. -func (fd *FileDescription) DecRef() { +func (fd *FileDescription) DecRef(ctx context.Context) { if refs := atomic.AddInt64(&fd.refs, -1); refs == 0 { // Unregister fd from all epoll instances. fd.epollMu.Lock() @@ -196,11 +196,11 @@ func (fd *FileDescription) DecRef() { } // Release implementation resources. - fd.impl.Release() + fd.impl.Release(ctx) if fd.writable { fd.vd.mount.EndWrite() } - fd.vd.DecRef() + fd.vd.DecRef(ctx) fd.flagsMu.Lock() // TODO(gvisor.dev/issue/1663): We may need to unregister during save, as we do in VFS1. if fd.statusFlags&linux.O_ASYNC != 0 && fd.asyncHandler != nil { @@ -335,7 +335,7 @@ func (fd *FileDescription) Impl() FileDescriptionImpl { type FileDescriptionImpl interface { // Release is called when the associated FileDescription reaches zero // references. - Release() + Release(ctx context.Context) // OnClose is called when a file descriptor representing the // FileDescription is closed. Note that returning a non-nil error does not @@ -526,7 +526,7 @@ func (fd *FileDescription) Stat(ctx context.Context, opts StatOptions) (linux.St Start: fd.vd, }) stat, err := fd.vd.mount.fs.impl.StatAt(ctx, rp, opts) - vfsObj.putResolvingPath(rp) + vfsObj.putResolvingPath(ctx, rp) return stat, err } return fd.impl.Stat(ctx, opts) @@ -541,7 +541,7 @@ func (fd *FileDescription) SetStat(ctx context.Context, opts SetStatOptions) err Start: fd.vd, }) err := fd.vd.mount.fs.impl.SetStatAt(ctx, rp, opts) - vfsObj.putResolvingPath(rp) + vfsObj.putResolvingPath(ctx, rp) return err } return fd.impl.SetStat(ctx, opts) @@ -557,7 +557,7 @@ func (fd *FileDescription) StatFS(ctx context.Context) (linux.Statfs, error) { Start: fd.vd, }) statfs, err := fd.vd.mount.fs.impl.StatFSAt(ctx, rp) - vfsObj.putResolvingPath(rp) + vfsObj.putResolvingPath(ctx, rp) return statfs, err } return fd.impl.StatFS(ctx) @@ -674,7 +674,7 @@ func (fd *FileDescription) Listxattr(ctx context.Context, size uint64) ([]string Start: fd.vd, }) names, err := fd.vd.mount.fs.impl.ListxattrAt(ctx, rp, size) - vfsObj.putResolvingPath(rp) + vfsObj.putResolvingPath(ctx, rp) return names, err } names, err := fd.impl.Listxattr(ctx, size) @@ -703,7 +703,7 @@ func (fd *FileDescription) Getxattr(ctx context.Context, opts *GetxattrOptions) Start: fd.vd, }) val, err := fd.vd.mount.fs.impl.GetxattrAt(ctx, rp, *opts) - vfsObj.putResolvingPath(rp) + vfsObj.putResolvingPath(ctx, rp) return val, err } return fd.impl.Getxattr(ctx, *opts) @@ -719,7 +719,7 @@ func (fd *FileDescription) Setxattr(ctx context.Context, opts *SetxattrOptions) Start: fd.vd, }) err := fd.vd.mount.fs.impl.SetxattrAt(ctx, rp, *opts) - vfsObj.putResolvingPath(rp) + vfsObj.putResolvingPath(ctx, rp) return err } return fd.impl.Setxattr(ctx, *opts) @@ -735,7 +735,7 @@ func (fd *FileDescription) Removexattr(ctx context.Context, name string) error { Start: fd.vd, }) err := fd.vd.mount.fs.impl.RemovexattrAt(ctx, rp, name) - vfsObj.putResolvingPath(rp) + vfsObj.putResolvingPath(ctx, rp) return err } return fd.impl.Removexattr(ctx, name) @@ -752,7 +752,7 @@ func (fd *FileDescription) MappedName(ctx context.Context) string { vfsroot := RootFromContext(ctx) s, _ := fd.vd.mount.vfs.PathnameWithDeleted(ctx, vfsroot, fd.vd) if vfsroot.Ok() { - vfsroot.DecRef() + vfsroot.DecRef(ctx) } return s } diff --git a/pkg/sentry/vfs/file_description_impl_util_test.go b/pkg/sentry/vfs/file_description_impl_util_test.go index 3b7e1c273..1cd607c0a 100644 --- a/pkg/sentry/vfs/file_description_impl_util_test.go +++ b/pkg/sentry/vfs/file_description_impl_util_test.go @@ -80,9 +80,9 @@ type testFD struct { data DynamicBytesSource } -func newTestFD(vfsObj *VirtualFilesystem, statusFlags uint32, data DynamicBytesSource) *FileDescription { +func newTestFD(ctx context.Context, vfsObj *VirtualFilesystem, statusFlags uint32, data DynamicBytesSource) *FileDescription { vd := vfsObj.NewAnonVirtualDentry("genCountFD") - defer vd.DecRef() + defer vd.DecRef(ctx) var fd testFD fd.vfsfd.Init(&fd, statusFlags, vd.Mount(), vd.Dentry(), &FileDescriptionOptions{}) fd.DynamicBytesFileDescriptionImpl.SetDataSource(data) @@ -90,7 +90,7 @@ func newTestFD(vfsObj *VirtualFilesystem, statusFlags uint32, data DynamicBytesS } // Release implements FileDescriptionImpl.Release. -func (fd *testFD) Release() { +func (fd *testFD) Release(context.Context) { } // SetStatusFlags implements FileDescriptionImpl.SetStatusFlags. @@ -109,11 +109,11 @@ func TestGenCountFD(t *testing.T) { ctx := contexttest.Context(t) vfsObj := &VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { + if err := vfsObj.Init(ctx); err != nil { t.Fatalf("VFS init: %v", err) } - fd := newTestFD(vfsObj, linux.O_RDWR, &genCount{}) - defer fd.DecRef() + fd := newTestFD(ctx, vfsObj, linux.O_RDWR, &genCount{}) + defer fd.DecRef(ctx) // The first read causes Generate to be called to fill the FD's buffer. buf := make([]byte, 2) @@ -167,11 +167,11 @@ func TestWritable(t *testing.T) { ctx := contexttest.Context(t) vfsObj := &VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { + if err := vfsObj.Init(ctx); err != nil { t.Fatalf("VFS init: %v", err) } - fd := newTestFD(vfsObj, linux.O_RDWR, &storeData{data: "init"}) - defer fd.DecRef() + fd := newTestFD(ctx, vfsObj, linux.O_RDWR, &storeData{data: "init"}) + defer fd.DecRef(ctx) buf := make([]byte, 10) ioseq := usermem.BytesIOSequence(buf) diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index 6bb9ca180..df3758fd1 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -100,12 +100,12 @@ func (fs *Filesystem) TryIncRef() bool { } // DecRef decrements fs' reference count. -func (fs *Filesystem) DecRef() { +func (fs *Filesystem) DecRef(ctx context.Context) { if refs := atomic.AddInt64(&fs.refs, -1); refs == 0 { fs.vfs.filesystemsMu.Lock() delete(fs.vfs.filesystems, fs) fs.vfs.filesystemsMu.Unlock() - fs.impl.Release() + fs.impl.Release(ctx) } else if refs < 0 { panic("Filesystem.decRef() called without holding a reference") } @@ -149,7 +149,7 @@ func (fs *Filesystem) DecRef() { type FilesystemImpl interface { // Release is called when the associated Filesystem reaches zero // references. - Release() + Release(ctx context.Context) // Sync "causes all pending modifications to filesystem metadata and cached // file data to be written to the underlying [filesystem]", as by syncfs(2). diff --git a/pkg/sentry/vfs/inotify.go b/pkg/sentry/vfs/inotify.go index 167b731ac..aff220a61 100644 --- a/pkg/sentry/vfs/inotify.go +++ b/pkg/sentry/vfs/inotify.go @@ -100,7 +100,7 @@ func NewInotifyFD(ctx context.Context, vfsObj *VirtualFilesystem, flags uint32) id := uniqueid.GlobalFromContext(ctx) vd := vfsObj.NewAnonVirtualDentry(fmt.Sprintf("[inotifyfd:%d]", id)) - defer vd.DecRef() + defer vd.DecRef(ctx) fd := &Inotify{ id: id, scratch: make([]byte, inotifyEventBaseSize), @@ -118,7 +118,7 @@ func NewInotifyFD(ctx context.Context, vfsObj *VirtualFilesystem, flags uint32) // Release implements FileDescriptionImpl.Release. Release removes all // watches and frees all resources for an inotify instance. -func (i *Inotify) Release() { +func (i *Inotify) Release(ctx context.Context) { var ds []*Dentry // We need to hold i.mu to avoid a race with concurrent calls to @@ -144,7 +144,7 @@ func (i *Inotify) Release() { i.mu.Unlock() for _, d := range ds { - d.OnZeroWatches() + d.OnZeroWatches(ctx) } } @@ -350,7 +350,7 @@ func (i *Inotify) AddWatch(target *Dentry, mask uint32) (int32, error) { // RmWatch looks up an inotify watch for the given 'wd' and configures the // target to stop sending events to this inotify instance. -func (i *Inotify) RmWatch(wd int32) error { +func (i *Inotify) RmWatch(ctx context.Context, wd int32) error { i.mu.Lock() // Find the watch we were asked to removed. @@ -374,7 +374,7 @@ func (i *Inotify) RmWatch(wd int32) error { i.mu.Unlock() if remaining == 0 { - w.target.OnZeroWatches() + w.target.OnZeroWatches(ctx) } // Generate the event for the removal. @@ -462,7 +462,7 @@ func (w *Watches) Remove(id uint64) { // Notify queues a new event with watches in this set. Watches with // IN_EXCL_UNLINK are skipped if the event is coming from a child that has been // unlinked. -func (w *Watches) Notify(name string, events, cookie uint32, et EventType, unlinked bool) { +func (w *Watches) Notify(ctx context.Context, name string, events, cookie uint32, et EventType, unlinked bool) { var hasExpired bool w.mu.RLock() for _, watch := range w.ws { @@ -476,13 +476,13 @@ func (w *Watches) Notify(name string, events, cookie uint32, et EventType, unlin w.mu.RUnlock() if hasExpired { - w.cleanupExpiredWatches() + w.cleanupExpiredWatches(ctx) } } // This function is relatively expensive and should only be called where there // are expired watches. -func (w *Watches) cleanupExpiredWatches() { +func (w *Watches) cleanupExpiredWatches(ctx context.Context) { // Because of lock ordering, we cannot acquire Inotify.mu for each watch // owner while holding w.mu. As a result, store expired watches locally // before removing. @@ -495,15 +495,15 @@ func (w *Watches) cleanupExpiredWatches() { } w.mu.RUnlock() for _, watch := range toRemove { - watch.owner.RmWatch(watch.wd) + watch.owner.RmWatch(ctx, watch.wd) } } // HandleDeletion is called when the watch target is destroyed. Clear the // watch set, detach watches from the inotify instances they belong to, and // generate the appropriate events. -func (w *Watches) HandleDeletion() { - w.Notify("", linux.IN_DELETE_SELF, 0, InodeEvent, true /* unlinked */) +func (w *Watches) HandleDeletion(ctx context.Context) { + w.Notify(ctx, "", linux.IN_DELETE_SELF, 0, InodeEvent, true /* unlinked */) // As in Watches.Notify, we can't hold w.mu while acquiring Inotify.mu for // the owner of each watch being deleted. Instead, atomically store the @@ -744,12 +744,12 @@ func InotifyEventFromStatMask(mask uint32) uint32 { // InotifyRemoveChild sends the appriopriate notifications to the watch sets of // the child being removed and its parent. Note that unlike most pairs of // parent/child notifications, the child is notified first in this case. -func InotifyRemoveChild(self, parent *Watches, name string) { +func InotifyRemoveChild(ctx context.Context, self, parent *Watches, name string) { if self != nil { - self.Notify("", linux.IN_ATTRIB, 0, InodeEvent, true /* unlinked */) + self.Notify(ctx, "", linux.IN_ATTRIB, 0, InodeEvent, true /* unlinked */) } if parent != nil { - parent.Notify(name, linux.IN_DELETE, 0, InodeEvent, true /* unlinked */) + parent.Notify(ctx, name, linux.IN_DELETE, 0, InodeEvent, true /* unlinked */) } } @@ -762,13 +762,13 @@ func InotifyRename(ctx context.Context, renamed, oldParent, newParent *Watches, } cookie := uniqueid.InotifyCookie(ctx) if oldParent != nil { - oldParent.Notify(oldName, dirEv|linux.IN_MOVED_FROM, cookie, InodeEvent, false /* unlinked */) + oldParent.Notify(ctx, oldName, dirEv|linux.IN_MOVED_FROM, cookie, InodeEvent, false /* unlinked */) } if newParent != nil { - newParent.Notify(newName, dirEv|linux.IN_MOVED_TO, cookie, InodeEvent, false /* unlinked */) + newParent.Notify(ctx, newName, dirEv|linux.IN_MOVED_TO, cookie, InodeEvent, false /* unlinked */) } // Somewhat surprisingly, self move events do not have a cookie. if renamed != nil { - renamed.Notify("", linux.IN_MOVE_SELF, 0, InodeEvent, false /* unlinked */) + renamed.Notify(ctx, "", linux.IN_MOVE_SELF, 0, InodeEvent, false /* unlinked */) } } diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index 32f901bd8..d1d29d0cd 100644 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go @@ -200,8 +200,8 @@ func (vfs *VirtualFilesystem) MountDisconnected(ctx context.Context, creds *auth if err != nil { return nil, err } - defer root.DecRef() - defer fs.DecRef() + defer root.DecRef(ctx) + defer fs.DecRef(ctx) return vfs.NewDisconnectedMount(fs, root, opts) } @@ -221,7 +221,7 @@ func (vfs *VirtualFilesystem) ConnectMountAt(ctx context.Context, creds *auth.Cr if vd.dentry.dead { vd.dentry.mu.Unlock() vfs.mountMu.Unlock() - vd.DecRef() + vd.DecRef(ctx) return syserror.ENOENT } // vd might have been mounted over between vfs.GetDentryAt() and @@ -243,7 +243,7 @@ func (vfs *VirtualFilesystem) ConnectMountAt(ctx context.Context, creds *auth.Cr // This can't fail since we're holding vfs.mountMu. nextmnt.root.IncRef() vd.dentry.mu.Unlock() - vd.DecRef() + vd.DecRef(ctx) vd = VirtualDentry{ mount: nextmnt, dentry: nextmnt.root, @@ -268,7 +268,7 @@ func (vfs *VirtualFilesystem) MountAt(ctx context.Context, creds *auth.Credentia if err != nil { return err } - defer mnt.DecRef() + defer mnt.DecRef(ctx) if err := vfs.ConnectMountAt(ctx, creds, mnt, target); err != nil { return err } @@ -293,13 +293,13 @@ func (vfs *VirtualFilesystem) UmountAt(ctx context.Context, creds *auth.Credenti if err != nil { return err } - defer vd.DecRef() + defer vd.DecRef(ctx) if vd.dentry != vd.mount.root { return syserror.EINVAL } vfs.mountMu.Lock() if mntns := MountNamespaceFromContext(ctx); mntns != nil { - defer mntns.DecRef() + defer mntns.DecRef(ctx) if mntns != vd.mount.ns { vfs.mountMu.Unlock() return syserror.EINVAL @@ -335,10 +335,10 @@ func (vfs *VirtualFilesystem) UmountAt(ctx context.Context, creds *auth.Credenti vfs.mounts.seq.EndWrite() vfs.mountMu.Unlock() for _, vd := range vdsToDecRef { - vd.DecRef() + vd.DecRef(ctx) } for _, mnt := range mountsToDecRef { - mnt.DecRef() + mnt.DecRef(ctx) } return nil } @@ -479,7 +479,7 @@ func (mnt *Mount) IncRef() { } // DecRef decrements mnt's reference count. -func (mnt *Mount) DecRef() { +func (mnt *Mount) DecRef(ctx context.Context) { refs := atomic.AddInt64(&mnt.refs, -1) if refs&^math.MinInt64 == 0 { // mask out MSB var vd VirtualDentry @@ -490,10 +490,10 @@ func (mnt *Mount) DecRef() { mnt.vfs.mounts.seq.EndWrite() mnt.vfs.mountMu.Unlock() } - mnt.root.DecRef() - mnt.fs.DecRef() + mnt.root.DecRef(ctx) + mnt.fs.DecRef(ctx) if vd.Ok() { - vd.DecRef() + vd.DecRef(ctx) } } } @@ -506,7 +506,7 @@ func (mntns *MountNamespace) IncRef() { } // DecRef decrements mntns' reference count. -func (mntns *MountNamespace) DecRef() { +func (mntns *MountNamespace) DecRef(ctx context.Context) { vfs := mntns.root.fs.VirtualFilesystem() if refs := atomic.AddInt64(&mntns.refs, -1); refs == 0 { vfs.mountMu.Lock() @@ -517,10 +517,10 @@ func (mntns *MountNamespace) DecRef() { vfs.mounts.seq.EndWrite() vfs.mountMu.Unlock() for _, vd := range vdsToDecRef { - vd.DecRef() + vd.DecRef(ctx) } for _, mnt := range mountsToDecRef { - mnt.DecRef() + mnt.DecRef(ctx) } } else if refs < 0 { panic("MountNamespace.DecRef() called without holding a reference") @@ -534,7 +534,7 @@ func (mntns *MountNamespace) DecRef() { // getMountAt is analogous to Linux's fs/namei.c:follow_mount(). // // Preconditions: References are held on mnt and d. -func (vfs *VirtualFilesystem) getMountAt(mnt *Mount, d *Dentry) *Mount { +func (vfs *VirtualFilesystem) getMountAt(ctx context.Context, mnt *Mount, d *Dentry) *Mount { // The first mount is special-cased: // // - The caller is assumed to have checked d.isMounted() already. (This @@ -565,7 +565,7 @@ retryFirst: // Raced with umount. continue } - mnt.DecRef() + mnt.DecRef(ctx) mnt = next d = next.root } @@ -578,7 +578,7 @@ retryFirst: // // Preconditions: References are held on mnt and root. vfsroot is not (mnt, // mnt.root). -func (vfs *VirtualFilesystem) getMountpointAt(mnt *Mount, vfsroot VirtualDentry) VirtualDentry { +func (vfs *VirtualFilesystem) getMountpointAt(ctx context.Context, mnt *Mount, vfsroot VirtualDentry) VirtualDentry { // The first mount is special-cased: // // - The caller must have already checked mnt against vfsroot. @@ -602,12 +602,12 @@ retryFirst: if !point.TryIncRef() { // Since Mount holds a reference on Mount.key.point, this can only // happen due to a racing change to Mount.key. - parent.DecRef() + parent.DecRef(ctx) goto retryFirst } if !vfs.mounts.seq.ReadOk(epoch) { - point.DecRef() - parent.DecRef() + point.DecRef(ctx) + parent.DecRef(ctx) goto retryFirst } mnt = parent @@ -635,16 +635,16 @@ retryFirst: if !point.TryIncRef() { // Since Mount holds a reference on Mount.key.point, this can // only happen due to a racing change to Mount.key. - parent.DecRef() + parent.DecRef(ctx) goto retryNotFirst } if !vfs.mounts.seq.ReadOk(epoch) { - point.DecRef() - parent.DecRef() + point.DecRef(ctx) + parent.DecRef(ctx) goto retryNotFirst } - d.DecRef() - mnt.DecRef() + d.DecRef(ctx) + mnt.DecRef(ctx) mnt = parent d = point } diff --git a/pkg/sentry/vfs/pathname.go b/pkg/sentry/vfs/pathname.go index cd78d66bc..e4da15009 100644 --- a/pkg/sentry/vfs/pathname.go +++ b/pkg/sentry/vfs/pathname.go @@ -47,7 +47,7 @@ func (vfs *VirtualFilesystem) PathnameWithDeleted(ctx context.Context, vfsroot, haveRef := false defer func() { if haveRef { - vd.DecRef() + vd.DecRef(ctx) } }() @@ -64,12 +64,12 @@ loop: // of FilesystemImpl.PrependPath() may return nil instead. break loop } - nextVD := vfs.getMountpointAt(vd.mount, vfsroot) + nextVD := vfs.getMountpointAt(ctx, vd.mount, vfsroot) if !nextVD.Ok() { break loop } if haveRef { - vd.DecRef() + vd.DecRef(ctx) } vd = nextVD haveRef = true @@ -101,7 +101,7 @@ func (vfs *VirtualFilesystem) PathnameReachable(ctx context.Context, vfsroot, vd haveRef := false defer func() { if haveRef { - vd.DecRef() + vd.DecRef(ctx) } }() loop: @@ -112,12 +112,12 @@ loop: if vd.mount == vfsroot.mount && vd.mount.root == vfsroot.dentry { break loop } - nextVD := vfs.getMountpointAt(vd.mount, vfsroot) + nextVD := vfs.getMountpointAt(ctx, vd.mount, vfsroot) if !nextVD.Ok() { return "", nil } if haveRef { - vd.DecRef() + vd.DecRef(ctx) } vd = nextVD haveRef = true @@ -145,7 +145,7 @@ func (vfs *VirtualFilesystem) PathnameForGetcwd(ctx context.Context, vfsroot, vd haveRef := false defer func() { if haveRef { - vd.DecRef() + vd.DecRef(ctx) } }() unreachable := false @@ -157,13 +157,13 @@ loop: if vd.mount == vfsroot.mount && vd.mount.root == vfsroot.dentry { break loop } - nextVD := vfs.getMountpointAt(vd.mount, vfsroot) + nextVD := vfs.getMountpointAt(ctx, vd.mount, vfsroot) if !nextVD.Ok() { unreachable = true break loop } if haveRef { - vd.DecRef() + vd.DecRef(ctx) } vd = nextVD haveRef = true diff --git a/pkg/sentry/vfs/resolving_path.go b/pkg/sentry/vfs/resolving_path.go index 9d047ff88..3304372d9 100644 --- a/pkg/sentry/vfs/resolving_path.go +++ b/pkg/sentry/vfs/resolving_path.go @@ -18,6 +18,7 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sync" @@ -136,31 +137,31 @@ func (vfs *VirtualFilesystem) getResolvingPath(creds *auth.Credentials, pop *Pat return rp } -func (vfs *VirtualFilesystem) putResolvingPath(rp *ResolvingPath) { +func (vfs *VirtualFilesystem) putResolvingPath(ctx context.Context, rp *ResolvingPath) { rp.root = VirtualDentry{} - rp.decRefStartAndMount() + rp.decRefStartAndMount(ctx) rp.mount = nil rp.start = nil - rp.releaseErrorState() + rp.releaseErrorState(ctx) resolvingPathPool.Put(rp) } -func (rp *ResolvingPath) decRefStartAndMount() { +func (rp *ResolvingPath) decRefStartAndMount(ctx context.Context) { if rp.flags&rpflagsHaveStartRef != 0 { - rp.start.DecRef() + rp.start.DecRef(ctx) } if rp.flags&rpflagsHaveMountRef != 0 { - rp.mount.DecRef() + rp.mount.DecRef(ctx) } } -func (rp *ResolvingPath) releaseErrorState() { +func (rp *ResolvingPath) releaseErrorState(ctx context.Context) { if rp.nextStart != nil { - rp.nextStart.DecRef() + rp.nextStart.DecRef(ctx) rp.nextStart = nil } if rp.nextMount != nil { - rp.nextMount.DecRef() + rp.nextMount.DecRef(ctx) rp.nextMount = nil } } @@ -236,13 +237,13 @@ func (rp *ResolvingPath) Advance() { // Restart resets the stream of path components represented by rp to its state // on entry to the current FilesystemImpl method. -func (rp *ResolvingPath) Restart() { +func (rp *ResolvingPath) Restart(ctx context.Context) { rp.pit = rp.origParts[rp.numOrigParts-1] rp.mustBeDir = rp.mustBeDirOrig rp.symlinks = rp.symlinksOrig rp.curPart = rp.numOrigParts - 1 copy(rp.parts[:], rp.origParts[:rp.numOrigParts]) - rp.releaseErrorState() + rp.releaseErrorState(ctx) } func (rp *ResolvingPath) relpathCommit() { @@ -260,13 +261,13 @@ func (rp *ResolvingPath) relpathCommit() { // Mount, CheckRoot returns (unspecified, non-nil error). Otherwise, path // resolution should resolve d's parent normally, and CheckRoot returns (false, // nil). -func (rp *ResolvingPath) CheckRoot(d *Dentry) (bool, error) { +func (rp *ResolvingPath) CheckRoot(ctx context.Context, d *Dentry) (bool, error) { if d == rp.root.dentry && rp.mount == rp.root.mount { // At contextual VFS root (due to e.g. chroot(2)). return true, nil } else if d == rp.mount.root { // At mount root ... - vd := rp.vfs.getMountpointAt(rp.mount, rp.root) + vd := rp.vfs.getMountpointAt(ctx, rp.mount, rp.root) if vd.Ok() { // ... of non-root mount. rp.nextMount = vd.mount @@ -283,11 +284,11 @@ func (rp *ResolvingPath) CheckRoot(d *Dentry) (bool, error) { // to d. If d is a mount point, such that path resolution should switch to // another Mount, CheckMount returns a non-nil error. Otherwise, CheckMount // returns nil. -func (rp *ResolvingPath) CheckMount(d *Dentry) error { +func (rp *ResolvingPath) CheckMount(ctx context.Context, d *Dentry) error { if !d.isMounted() { return nil } - if mnt := rp.vfs.getMountAt(rp.mount, d); mnt != nil { + if mnt := rp.vfs.getMountAt(ctx, rp.mount, d); mnt != nil { rp.nextMount = mnt return resolveMountPointError{} } @@ -389,11 +390,11 @@ func (rp *ResolvingPath) HandleJump(target VirtualDentry) error { return resolveMountRootOrJumpError{} } -func (rp *ResolvingPath) handleError(err error) bool { +func (rp *ResolvingPath) handleError(ctx context.Context, err error) bool { switch err.(type) { case resolveMountRootOrJumpError: // Switch to the new Mount. We hold references on the Mount and Dentry. - rp.decRefStartAndMount() + rp.decRefStartAndMount(ctx) rp.mount = rp.nextMount rp.start = rp.nextStart rp.flags |= rpflagsHaveMountRef | rpflagsHaveStartRef @@ -412,7 +413,7 @@ func (rp *ResolvingPath) handleError(err error) bool { case resolveMountPointError: // Switch to the new Mount. We hold a reference on the Mount, but // borrow the reference on the mount root from the Mount. - rp.decRefStartAndMount() + rp.decRefStartAndMount(ctx) rp.mount = rp.nextMount rp.start = rp.nextMount.root rp.flags = rp.flags&^rpflagsHaveStartRef | rpflagsHaveMountRef @@ -423,12 +424,12 @@ func (rp *ResolvingPath) handleError(err error) bool { // path. rp.relpathCommit() // Restart path resolution on the new Mount. - rp.releaseErrorState() + rp.releaseErrorState(ctx) return true case resolveAbsSymlinkError: // Switch to the new Mount. References are borrowed from rp.root. - rp.decRefStartAndMount() + rp.decRefStartAndMount(ctx) rp.mount = rp.root.mount rp.start = rp.root.dentry rp.flags &^= rpflagsHaveMountRef | rpflagsHaveStartRef @@ -440,7 +441,7 @@ func (rp *ResolvingPath) handleError(err error) bool { // path, including the symlink target we just prepended. rp.relpathCommit() // Restart path resolution on the new Mount. - rp.releaseErrorState() + rp.releaseErrorState(ctx) return true default: diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 522e27475..9c2420683 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -122,7 +122,7 @@ type VirtualFilesystem struct { } // Init initializes a new VirtualFilesystem with no mounts or FilesystemTypes. -func (vfs *VirtualFilesystem) Init() error { +func (vfs *VirtualFilesystem) Init(ctx context.Context) error { if vfs.mountpoints != nil { panic("VFS already initialized") } @@ -145,7 +145,7 @@ func (vfs *VirtualFilesystem) Init() error { devMinor: anonfsDevMinor, } anonfs.vfsfs.Init(vfs, &anonFilesystemType{}, &anonfs) - defer anonfs.vfsfs.DecRef() + defer anonfs.vfsfs.DecRef(ctx) anonMount, err := vfs.NewDisconnectedMount(&anonfs.vfsfs, nil, &MountOptions{}) if err != nil { // We should not be passing any MountOptions that would cause @@ -192,11 +192,11 @@ func (vfs *VirtualFilesystem) AccessAt(ctx context.Context, creds *auth.Credenti for { err := rp.mount.fs.impl.AccessAt(ctx, rp, creds, ats) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return nil } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return err } } @@ -214,11 +214,11 @@ func (vfs *VirtualFilesystem) GetDentryAt(ctx context.Context, creds *auth.Crede dentry: d, } rp.mount.IncRef() - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return vd, nil } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return VirtualDentry{}, err } } @@ -236,7 +236,7 @@ func (vfs *VirtualFilesystem) getParentDirAndName(ctx context.Context, creds *au } rp.mount.IncRef() name := rp.Component() - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return parentVD, name, nil } if checkInvariants { @@ -244,8 +244,8 @@ func (vfs *VirtualFilesystem) getParentDirAndName(ctx context.Context, creds *au panic(fmt.Sprintf("%T.GetParentDentryAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return VirtualDentry{}, "", err } } @@ -260,14 +260,14 @@ func (vfs *VirtualFilesystem) LinkAt(ctx context.Context, creds *auth.Credential } if !newpop.Path.Begin.Ok() { - oldVD.DecRef() + oldVD.DecRef(ctx) if newpop.Path.Absolute { return syserror.EEXIST } return syserror.ENOENT } if newpop.FollowFinalSymlink { - oldVD.DecRef() + oldVD.DecRef(ctx) ctx.Warningf("VirtualFilesystem.LinkAt: file creation paths can't follow final symlink") return syserror.EINVAL } @@ -276,8 +276,8 @@ func (vfs *VirtualFilesystem) LinkAt(ctx context.Context, creds *auth.Credential for { err := rp.mount.fs.impl.LinkAt(ctx, rp, oldVD) if err == nil { - vfs.putResolvingPath(rp) - oldVD.DecRef() + vfs.putResolvingPath(ctx, rp) + oldVD.DecRef(ctx) return nil } if checkInvariants { @@ -285,9 +285,9 @@ func (vfs *VirtualFilesystem) LinkAt(ctx context.Context, creds *auth.Credential panic(fmt.Sprintf("%T.LinkAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) - oldVD.DecRef() + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) + oldVD.DecRef(ctx) return err } } @@ -313,7 +313,7 @@ func (vfs *VirtualFilesystem) MkdirAt(ctx context.Context, creds *auth.Credentia for { err := rp.mount.fs.impl.MkdirAt(ctx, rp, *opts) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return nil } if checkInvariants { @@ -321,8 +321,8 @@ func (vfs *VirtualFilesystem) MkdirAt(ctx context.Context, creds *auth.Credentia panic(fmt.Sprintf("%T.MkdirAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return err } } @@ -346,7 +346,7 @@ func (vfs *VirtualFilesystem) MknodAt(ctx context.Context, creds *auth.Credentia for { err := rp.mount.fs.impl.MknodAt(ctx, rp, *opts) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return nil } if checkInvariants { @@ -354,8 +354,8 @@ func (vfs *VirtualFilesystem) MknodAt(ctx context.Context, creds *auth.Credentia panic(fmt.Sprintf("%T.MknodAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return err } } @@ -408,31 +408,31 @@ func (vfs *VirtualFilesystem) OpenAt(ctx context.Context, creds *auth.Credential for { fd, err := rp.mount.fs.impl.OpenAt(ctx, rp, *opts) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) if opts.FileExec { if fd.Mount().Flags.NoExec { - fd.DecRef() + fd.DecRef(ctx) return nil, syserror.EACCES } // Only a regular file can be executed. stat, err := fd.Stat(ctx, StatOptions{Mask: linux.STATX_TYPE}) if err != nil { - fd.DecRef() + fd.DecRef(ctx) return nil, err } if stat.Mask&linux.STATX_TYPE == 0 || stat.Mode&linux.S_IFMT != linux.S_IFREG { - fd.DecRef() + fd.DecRef(ctx) return nil, syserror.EACCES } } - fd.Dentry().InotifyWithParent(linux.IN_OPEN, 0, PathEvent) + fd.Dentry().InotifyWithParent(ctx, linux.IN_OPEN, 0, PathEvent) return fd, nil } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return nil, err } } @@ -444,11 +444,11 @@ func (vfs *VirtualFilesystem) ReadlinkAt(ctx context.Context, creds *auth.Creden for { target, err := rp.mount.fs.impl.ReadlinkAt(ctx, rp) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return target, nil } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return "", err } } @@ -472,19 +472,19 @@ func (vfs *VirtualFilesystem) RenameAt(ctx context.Context, creds *auth.Credenti return err } if oldName == "." || oldName == ".." { - oldParentVD.DecRef() + oldParentVD.DecRef(ctx) return syserror.EBUSY } if !newpop.Path.Begin.Ok() { - oldParentVD.DecRef() + oldParentVD.DecRef(ctx) if newpop.Path.Absolute { return syserror.EBUSY } return syserror.ENOENT } if newpop.FollowFinalSymlink { - oldParentVD.DecRef() + oldParentVD.DecRef(ctx) ctx.Warningf("VirtualFilesystem.RenameAt: destination path can't follow final symlink") return syserror.EINVAL } @@ -497,8 +497,8 @@ func (vfs *VirtualFilesystem) RenameAt(ctx context.Context, creds *auth.Credenti for { err := rp.mount.fs.impl.RenameAt(ctx, rp, oldParentVD, oldName, renameOpts) if err == nil { - vfs.putResolvingPath(rp) - oldParentVD.DecRef() + vfs.putResolvingPath(ctx, rp) + oldParentVD.DecRef(ctx) return nil } if checkInvariants { @@ -506,9 +506,9 @@ func (vfs *VirtualFilesystem) RenameAt(ctx context.Context, creds *auth.Credenti panic(fmt.Sprintf("%T.RenameAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) - oldParentVD.DecRef() + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) + oldParentVD.DecRef(ctx) return err } } @@ -531,7 +531,7 @@ func (vfs *VirtualFilesystem) RmdirAt(ctx context.Context, creds *auth.Credentia for { err := rp.mount.fs.impl.RmdirAt(ctx, rp) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return nil } if checkInvariants { @@ -539,8 +539,8 @@ func (vfs *VirtualFilesystem) RmdirAt(ctx context.Context, creds *auth.Credentia panic(fmt.Sprintf("%T.RmdirAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return err } } @@ -552,11 +552,11 @@ func (vfs *VirtualFilesystem) SetStatAt(ctx context.Context, creds *auth.Credent for { err := rp.mount.fs.impl.SetStatAt(ctx, rp, *opts) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return nil } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return err } } @@ -568,11 +568,11 @@ func (vfs *VirtualFilesystem) StatAt(ctx context.Context, creds *auth.Credential for { stat, err := rp.mount.fs.impl.StatAt(ctx, rp, *opts) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return stat, nil } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return linux.Statx{}, err } } @@ -585,11 +585,11 @@ func (vfs *VirtualFilesystem) StatFSAt(ctx context.Context, creds *auth.Credenti for { statfs, err := rp.mount.fs.impl.StatFSAt(ctx, rp) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return statfs, nil } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return linux.Statfs{}, err } } @@ -612,7 +612,7 @@ func (vfs *VirtualFilesystem) SymlinkAt(ctx context.Context, creds *auth.Credent for { err := rp.mount.fs.impl.SymlinkAt(ctx, rp, target) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return nil } if checkInvariants { @@ -620,8 +620,8 @@ func (vfs *VirtualFilesystem) SymlinkAt(ctx context.Context, creds *auth.Credent panic(fmt.Sprintf("%T.SymlinkAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return err } } @@ -644,7 +644,7 @@ func (vfs *VirtualFilesystem) UnlinkAt(ctx context.Context, creds *auth.Credenti for { err := rp.mount.fs.impl.UnlinkAt(ctx, rp) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return nil } if checkInvariants { @@ -652,8 +652,8 @@ func (vfs *VirtualFilesystem) UnlinkAt(ctx context.Context, creds *auth.Credenti panic(fmt.Sprintf("%T.UnlinkAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return err } } @@ -671,7 +671,7 @@ func (vfs *VirtualFilesystem) BoundEndpointAt(ctx context.Context, creds *auth.C for { bep, err := rp.mount.fs.impl.BoundEndpointAt(ctx, rp, *opts) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return bep, nil } if checkInvariants { @@ -679,8 +679,8 @@ func (vfs *VirtualFilesystem) BoundEndpointAt(ctx context.Context, creds *auth.C panic(fmt.Sprintf("%T.BoundEndpointAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return nil, err } } @@ -693,7 +693,7 @@ func (vfs *VirtualFilesystem) ListxattrAt(ctx context.Context, creds *auth.Crede for { names, err := rp.mount.fs.impl.ListxattrAt(ctx, rp, size) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return names, nil } if err == syserror.ENOTSUP { @@ -701,11 +701,11 @@ func (vfs *VirtualFilesystem) ListxattrAt(ctx context.Context, creds *auth.Crede // fs/xattr.c:vfs_listxattr() falls back to allowing the security // subsystem to return security extended attributes, which by // default don't exist. - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return nil, nil } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return nil, err } } @@ -718,11 +718,11 @@ func (vfs *VirtualFilesystem) GetxattrAt(ctx context.Context, creds *auth.Creden for { val, err := rp.mount.fs.impl.GetxattrAt(ctx, rp, *opts) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return val, nil } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return "", err } } @@ -735,11 +735,11 @@ func (vfs *VirtualFilesystem) SetxattrAt(ctx context.Context, creds *auth.Creden for { err := rp.mount.fs.impl.SetxattrAt(ctx, rp, *opts) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return nil } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return err } } @@ -751,11 +751,11 @@ func (vfs *VirtualFilesystem) RemovexattrAt(ctx context.Context, creds *auth.Cre for { err := rp.mount.fs.impl.RemovexattrAt(ctx, rp, name) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return nil } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return err } } @@ -777,7 +777,7 @@ func (vfs *VirtualFilesystem) SyncAllFilesystems(ctx context.Context) error { if err := fs.impl.Sync(ctx); err != nil && retErr == nil { retErr = err } - fs.DecRef() + fs.DecRef(ctx) } return retErr } @@ -831,9 +831,9 @@ func (vd VirtualDentry) IncRef() { // DecRef decrements the reference counts on the Mount and Dentry represented // by vd. -func (vd VirtualDentry) DecRef() { - vd.dentry.DecRef() - vd.mount.DecRef() +func (vd VirtualDentry) DecRef(ctx context.Context) { + vd.dentry.DecRef(ctx) + vd.mount.DecRef(ctx) } // Mount returns the Mount associated with vd. It does not take a reference on diff --git a/pkg/tcpip/link/tun/BUILD b/pkg/tcpip/link/tun/BUILD index e0db6cf54..6c137f693 100644 --- a/pkg/tcpip/link/tun/BUILD +++ b/pkg/tcpip/link/tun/BUILD @@ -12,6 +12,7 @@ go_library( visibility = ["//visibility:public"], deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/refs", "//pkg/sync", "//pkg/syserror", diff --git a/pkg/tcpip/link/tun/device.go b/pkg/tcpip/link/tun/device.go index 04ae58e59..22b0a12bd 100644 --- a/pkg/tcpip/link/tun/device.go +++ b/pkg/tcpip/link/tun/device.go @@ -18,6 +18,7 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" @@ -64,14 +65,14 @@ func (d *Device) beforeSave() { } // Release implements fs.FileOperations.Release. -func (d *Device) Release() { +func (d *Device) Release(ctx context.Context) { d.mu.Lock() defer d.mu.Unlock() // Decrease refcount if there is an endpoint associated with this file. if d.endpoint != nil { d.endpoint.RemoveNotify(d.notifyHandle) - d.endpoint.DecRef() + d.endpoint.DecRef(ctx) d.endpoint = nil } } @@ -341,8 +342,8 @@ type tunEndpoint struct { } // DecRef decrements refcount of e, removes NIC if refcount goes to 0. -func (e *tunEndpoint) DecRef() { - e.DecRefWithDestructor(func() { +func (e *tunEndpoint) DecRef(ctx context.Context) { + e.DecRefWithDestructor(ctx, func(context.Context) { e.stack.RemoveNIC(e.nicID) }) } diff --git a/runsc/boot/fs.go b/runsc/boot/fs.go index 59639ba19..9dd5b0184 100644 --- a/runsc/boot/fs.go +++ b/runsc/boot/fs.go @@ -640,7 +640,7 @@ func (c *containerMounter) createMountNamespace(ctx context.Context, conf *Confi func (c *containerMounter) mountSubmounts(ctx context.Context, conf *Config, mns *fs.MountNamespace) error { root := mns.Root() - defer root.DecRef() + defer root.DecRef(ctx) for _, m := range c.mounts { log.Debugf("Mounting %q to %q, type: %s, options: %s", m.Source, m.Destination, m.Type, m.Options) @@ -868,7 +868,7 @@ func (c *containerMounter) mountSubmount(ctx context.Context, conf *Config, mns if err != nil { return fmt.Errorf("can't find mount destination %q: %v", m.Destination, err) } - defer dirent.DecRef() + defer dirent.DecRef(ctx) if err := mns.Mount(ctx, dirent, inode); err != nil { return fmt.Errorf("mount %q error: %v", m.Destination, err) } @@ -889,12 +889,12 @@ func (c *containerMounter) mountSharedSubmount(ctx context.Context, mns *fs.Moun if err != nil { return fmt.Errorf("can't find mount destination %q: %v", mount.Destination, err) } - defer target.DecRef() + defer target.DecRef(ctx) // Take a ref on the inode that is about to be (re)-mounted. source.root.IncRef() if err := mns.Mount(ctx, target, source.root); err != nil { - source.root.DecRef() + source.root.DecRef(ctx) return fmt.Errorf("bind mount %q error: %v", mount.Destination, err) } @@ -997,12 +997,12 @@ func (c *containerMounter) mountTmp(ctx context.Context, conf *Config, mns *fs.M switch err { case nil: // Found '/tmp' in filesystem, check if it's empty. - defer tmp.DecRef() + defer tmp.DecRef(ctx) f, err := tmp.Inode.GetFile(ctx, tmp, fs.FileFlags{Read: true, Directory: true}) if err != nil { return err } - defer f.DecRef() + defer f.DecRef(ctx) serializer := &fs.CollectEntriesSerializer{} if err := f.Readdir(ctx, serializer); err != nil { return err diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index 9cd9c5909..e0d077f5a 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -346,7 +346,7 @@ func New(args Args) (*Loader, error) { if err != nil { return nil, fmt.Errorf("failed to create hostfs filesystem: %v", err) } - defer hostFilesystem.DecRef() + defer hostFilesystem.DecRef(k.SupervisorContext()) hostMount, err := k.VFS().NewDisconnectedMount(hostFilesystem, nil, &vfs.MountOptions{}) if err != nil { return nil, fmt.Errorf("failed to create hostfs mount: %v", err) @@ -755,7 +755,7 @@ func (l *Loader) createContainerProcess(root bool, cid string, info *containerIn return nil, fmt.Errorf("creating process: %v", err) } // CreateProcess takes a reference on FDTable if successful. - info.procArgs.FDTable.DecRef() + info.procArgs.FDTable.DecRef(ctx) // Set the foreground process group on the TTY to the global init process // group, since that is what we are about to start running. @@ -890,22 +890,20 @@ func (l *Loader) executeAsync(args *control.ExecArgs) (kernel.ThreadID, error) { // Add the HOME environment variable if it is not already set. if kernel.VFS2Enabled { - defer args.MountNamespaceVFS2.DecRef() - root := args.MountNamespaceVFS2.Root() - defer root.DecRef() ctx := vfs.WithRoot(l.k.SupervisorContext(), root) + defer args.MountNamespaceVFS2.DecRef(ctx) + defer root.DecRef(ctx) envv, err := user.MaybeAddExecUserHomeVFS2(ctx, args.MountNamespaceVFS2, args.KUID, args.Envv) if err != nil { return 0, err } args.Envv = envv } else { - defer args.MountNamespace.DecRef() - root := args.MountNamespace.Root() - defer root.DecRef() ctx := fs.WithRoot(l.k.SupervisorContext(), root) + defer args.MountNamespace.DecRef(ctx) + defer root.DecRef(ctx) envv, err := user.MaybeAddExecUserHome(ctx, args.MountNamespace, args.KUID, args.Envv) if err != nil { return 0, err @@ -1263,7 +1261,7 @@ func createFDTable(ctx context.Context, console bool, stdioFDs []int) (*kernel.F fdTable := k.NewFDTable() ttyFile, ttyFileVFS2, err := fdimport.Import(ctx, fdTable, console, stdioFDs) if err != nil { - fdTable.DecRef() + fdTable.DecRef(ctx) return nil, nil, nil, err } return fdTable, ttyFile, ttyFileVFS2, nil diff --git a/runsc/boot/loader_test.go b/runsc/boot/loader_test.go index 8e6fe57e1..aa3fdf96c 100644 --- a/runsc/boot/loader_test.go +++ b/runsc/boot/loader_test.go @@ -450,13 +450,13 @@ func TestCreateMountNamespace(t *testing.T) { } root := mns.Root() - defer root.DecRef() + defer root.DecRef(ctx) for _, p := range tc.expectedPaths { maxTraversals := uint(0) if d, err := mns.FindInode(ctx, root, root, p, &maxTraversals); err != nil { t.Errorf("expected path %v to exist with spec %v, but got error %v", p, tc.spec, err) } else { - d.DecRef() + d.DecRef(ctx) } } }) @@ -491,7 +491,7 @@ func TestCreateMountNamespaceVFS2(t *testing.T) { } root := mns.Root() - defer root.DecRef() + defer root.DecRef(ctx) for _, p := range tc.expectedPaths { target := &vfs.PathOperation{ Root: root, @@ -502,7 +502,7 @@ func TestCreateMountNamespaceVFS2(t *testing.T) { if d, err := l.k.VFS().GetDentryAt(ctx, l.root.procArgs.Credentials, target, &vfs.GetDentryOptions{}); err != nil { t.Errorf("expected path %v to exist with spec %v, but got error %v", p, tc.spec, err) } else { - d.DecRef() + d.DecRef(ctx) } } }) diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go index cfe2d36aa..252ca07e3 100644 --- a/runsc/boot/vfs.go +++ b/runsc/boot/vfs.go @@ -103,7 +103,7 @@ func registerFilesystems(k *kernel.Kernel) error { if err != nil { return fmt.Errorf("creating devtmpfs accessor: %w", err) } - defer a.Release() + defer a.Release(ctx) if err := a.UserspaceInit(ctx); err != nil { return fmt.Errorf("initializing userspace: %w", err) @@ -252,7 +252,7 @@ func (c *containerMounter) prepareMountsVFS2() ([]mountAndFD, error) { func (c *containerMounter) mountSubmountVFS2(ctx context.Context, conf *Config, mns *vfs.MountNamespace, creds *auth.Credentials, submount *mountAndFD) error { root := mns.Root() - defer root.DecRef() + defer root.DecRef(ctx) target := &vfs.PathOperation{ Root: root, Start: root, @@ -387,7 +387,7 @@ func (c *containerMounter) mountTmpVFS2(ctx context.Context, conf *Config, creds } root := mns.Root() - defer root.DecRef() + defer root.DecRef(ctx) pop := vfs.PathOperation{ Root: root, Start: root, @@ -481,10 +481,10 @@ func (c *containerMounter) mountSharedSubmountVFS2(ctx context.Context, conf *Co if err != nil { return err } - defer newMnt.DecRef() + defer newMnt.DecRef(ctx) root := mns.Root() - defer root.DecRef() + defer root.DecRef(ctx) if err := c.makeSyntheticMount(ctx, mount.Destination, root, creds); err != nil { return err } -- cgit v1.2.3 From 5626ccf61f718d80d46049e11d040a3a8fc72fb9 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Mon, 3 Aug 2020 15:23:05 -0700 Subject: Remove old TODO. Fixes #2920. PiperOrigin-RevId: 324695118 --- pkg/sentry/syscalls/linux/vfs2/fd.go | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/vfs2/fd.go b/pkg/sentry/syscalls/linux/vfs2/fd.go index 72ca916a0..4856554fe 100644 --- a/pkg/sentry/syscalls/linux/vfs2/fd.go +++ b/pkg/sentry/syscalls/linux/vfs2/fd.go @@ -208,7 +208,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall case linux.F_SETLK, linux.F_SETLKW: return 0, nil, posixLock(t, args, file, cmd) default: - // TODO(gvisor.dev/issue/2920): Everything else is not yet supported. + // Everything else is not yet supported. return 0, nil, syserror.EINVAL } } -- cgit v1.2.3 From fe441dd251ba19e6ba8a8ed6f0fa3c147e03360c Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Mon, 3 Aug 2020 15:25:25 -0700 Subject: Internal change. PiperOrigin-RevId: 324695672 --- pkg/sentry/pgalloc/pgalloc.go | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/pgalloc/pgalloc.go b/pkg/sentry/pgalloc/pgalloc.go index 3243d7214..46d3be58c 100644 --- a/pkg/sentry/pgalloc/pgalloc.go +++ b/pkg/sentry/pgalloc/pgalloc.go @@ -409,7 +409,6 @@ func (f *MemoryFile) Allocate(length uint64, kind usage.MemoryKind) (memmap.File f.mappingsMu.Unlock() } - // Mark selected pages as in use. if f.opts.ManualZeroing { if err := f.forEachMappingSlice(fr, func(bs []byte) { for i := range bs { @@ -419,6 +418,7 @@ func (f *MemoryFile) Allocate(length uint64, kind usage.MemoryKind) (memmap.File return memmap.FileRange{}, err } } + // Mark selected pages as in use. if !f.usage.Add(fr, usageInfo{ kind: kind, refs: 1, -- cgit v1.2.3 From ad7c9fc4c3c027ea7ed211a6bdb35c38156280e6 Mon Sep 17 00:00:00 2001 From: Ayush Ranjan Date: Mon, 3 Aug 2020 18:15:11 -0700 Subject: [vfs2] Implement /sys/devices/system/cpu/cpuX. Fixes #3364 PiperOrigin-RevId: 324724614 --- pkg/sentry/fsimpl/sys/sys.go | 24 ++++++++++++++++-------- 1 file changed, 16 insertions(+), 8 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go index f81b0c38f..0401726b6 100644 --- a/pkg/sentry/fsimpl/sys/sys.go +++ b/pkg/sentry/fsimpl/sys/sys.go @@ -30,6 +30,7 @@ import ( // Name is the default filesystem name. const Name = "sysfs" +const defaultSysDirMode = linux.FileMode(0755) // FilesystemType implements vfs.FilesystemType. type FilesystemType struct{} @@ -57,9 +58,6 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt devMinor: devMinor, } fs.VFSFilesystem().Init(vfsObj, &fsType, fs) - k := kernel.KernelFromContext(ctx) - maxCPUCores := k.ApplicationCores() - defaultSysDirMode := linux.FileMode(0755) root := fs.newDir(creds, defaultSysDirMode, map[string]*kernfs.Dentry{ "block": fs.newDir(creds, defaultSysDirMode, nil), @@ -70,11 +68,7 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt "dev": fs.newDir(creds, defaultSysDirMode, nil), "devices": fs.newDir(creds, defaultSysDirMode, map[string]*kernfs.Dentry{ "system": fs.newDir(creds, defaultSysDirMode, map[string]*kernfs.Dentry{ - "cpu": fs.newDir(creds, defaultSysDirMode, map[string]*kernfs.Dentry{ - "online": fs.newCPUFile(creds, maxCPUCores, linux.FileMode(0444)), - "possible": fs.newCPUFile(creds, maxCPUCores, linux.FileMode(0444)), - "present": fs.newCPUFile(creds, maxCPUCores, linux.FileMode(0444)), - }), + "cpu": cpuDir(ctx, fs, creds), }), }), "firmware": fs.newDir(creds, defaultSysDirMode, nil), @@ -86,6 +80,20 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt return fs.VFSFilesystem(), root.VFSDentry(), nil } +func cpuDir(ctx context.Context, fs *filesystem, creds *auth.Credentials) *kernfs.Dentry { + k := kernel.KernelFromContext(ctx) + maxCPUCores := k.ApplicationCores() + children := map[string]*kernfs.Dentry{ + "online": fs.newCPUFile(creds, maxCPUCores, linux.FileMode(0444)), + "possible": fs.newCPUFile(creds, maxCPUCores, linux.FileMode(0444)), + "present": fs.newCPUFile(creds, maxCPUCores, linux.FileMode(0444)), + } + for i := uint(0); i < maxCPUCores; i++ { + children[fmt.Sprintf("cpu%d", i)] = fs.newDir(creds, linux.FileMode(0555), nil) + } + return fs.newDir(creds, defaultSysDirMode, children) +} + // Release implements vfs.FilesystemImpl.Release. func (fs *filesystem) Release(ctx context.Context) { fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) -- cgit v1.2.3 From 25798f214c6d1991916906ea8fca9e7029a8c423 Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Mon, 3 Aug 2020 22:06:46 -0700 Subject: Add callbacks to support lazy loading/restoring thread states PiperOrigin-RevId: 324748508 --- pkg/sentry/kernel/kernel.go | 7 +++++ pkg/sentry/kernel/ptrace.go | 10 +++++++- pkg/sentry/kernel/task_clone.go | 4 +++ pkg/sentry/kernel/task_log.go | 43 +++++++++++++++++++++++++++++++ pkg/sentry/kernel/task_run.go | 2 +- pkg/sentry/kernel/task_signals.go | 10 ++++++-- pkg/sentry/kernel/task_stop.go | 16 ++++++++++++ pkg/sentry/memmap/memmap.go | 7 +++++ pkg/sentry/mm/lifecycle.go | 2 ++ pkg/sentry/mm/vma.go | 7 ++++- pkg/sentry/platform/kvm/BUILD | 1 + pkg/sentry/platform/kvm/address_space.go | 6 +++++ pkg/sentry/platform/kvm/context.go | 10 +++++++- pkg/sentry/platform/platform.go | 44 +++++++++++++++++++++++++++++++- pkg/sentry/platform/ptrace/BUILD | 1 + pkg/sentry/platform/ptrace/ptrace.go | 10 +++++++- pkg/sentry/platform/ptrace/subprocess.go | 6 +++++ pkg/sentry/state/state.go | 1 + 18 files changed, 179 insertions(+), 8 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 316df249d..1028d13c6 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -1263,6 +1263,13 @@ func (k *Kernel) Pause() { k.tasks.aioGoroutines.Wait() } +// ReceiveTaskStates receives full states for all tasks. +func (k *Kernel) ReceiveTaskStates() { + k.extMu.Lock() + k.tasks.PullFullState() + k.extMu.Unlock() +} + // Unpause ends the effect of a previous call to Pause. If Unpause is called // without a matching preceding call to Pause, Unpause may panic. func (k *Kernel) Unpause() { diff --git a/pkg/sentry/kernel/ptrace.go b/pkg/sentry/kernel/ptrace.go index e23e796ef..6c03d9041 100644 --- a/pkg/sentry/kernel/ptrace.go +++ b/pkg/sentry/kernel/ptrace.go @@ -1018,6 +1018,9 @@ func (t *Task) Ptrace(req int64, pid ThreadID, addr, data usermem.Addr) error { if err != nil { return err } + + t.p.PullFullState(t.MemoryManager().AddressSpace(), t.Arch()) + ar := ars.Head() n, err := target.Arch().PtraceGetRegSet(uintptr(addr), &usermem.IOReadWriter{ Ctx: t, @@ -1044,10 +1047,14 @@ func (t *Task) Ptrace(req int64, pid ThreadID, addr, data usermem.Addr) error { if err != nil { return err } + + mm := t.MemoryManager() + t.p.PullFullState(mm.AddressSpace(), t.Arch()) + ar := ars.Head() n, err := target.Arch().PtraceSetRegSet(uintptr(addr), &usermem.IOReadWriter{ Ctx: t, - IO: t.MemoryManager(), + IO: mm, Addr: ar.Start, Opts: usermem.IOOpts{ AddressSpaceActive: true, @@ -1056,6 +1063,7 @@ func (t *Task) Ptrace(req int64, pid ThreadID, addr, data usermem.Addr) error { if err != nil { return err } + t.p.FloatingPointStateChanged() ar.End -= usermem.Addr(n) return t.CopyOutIovecs(data, usermem.AddrRangeSeqOf(ar)) diff --git a/pkg/sentry/kernel/task_clone.go b/pkg/sentry/kernel/task_clone.go index fe6ba6041..9d7a9128f 100644 --- a/pkg/sentry/kernel/task_clone.go +++ b/pkg/sentry/kernel/task_clone.go @@ -161,6 +161,10 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { return 0, nil, syserror.EINVAL } + // Pull task registers and FPU state, a cloned task will inherit the + // state of the current task. + t.p.PullFullState(t.MemoryManager().AddressSpace(), t.Arch()) + // "If CLONE_NEWUSER is specified along with other CLONE_NEW* flags in a // single clone(2) or unshare(2) call, the user namespace is guaranteed to // be created first, giving the child (clone(2)) or caller (unshare(2)) diff --git a/pkg/sentry/kernel/task_log.go b/pkg/sentry/kernel/task_log.go index ab86ceedc..d23cea802 100644 --- a/pkg/sentry/kernel/task_log.go +++ b/pkg/sentry/kernel/task_log.go @@ -27,6 +27,9 @@ const ( // maxStackDebugBytes is the maximum number of user stack bytes that may be // printed by debugDumpStack. maxStackDebugBytes = 1024 + // maxCodeDebugBytes is the maximum number of user code bytes that may be + // printed by debugDumpCode. + maxCodeDebugBytes = 128 ) // Infof logs an formatted info message by calling log.Infof. @@ -61,6 +64,7 @@ func (t *Task) IsLogging(level log.Level) bool { func (t *Task) DebugDumpState() { t.debugDumpRegisters() t.debugDumpStack() + t.debugDumpCode() if mm := t.MemoryManager(); mm != nil { t.Debugf("Mappings:\n%s", mm) } @@ -128,6 +132,45 @@ func (t *Task) debugDumpStack() { } } +// debugDumpCode logs user code contents at log level debug. +// +// Preconditions: The caller must be running on the task goroutine. +func (t *Task) debugDumpCode() { + if !t.IsLogging(log.Debug) { + return + } + m := t.MemoryManager() + if m == nil { + t.Debugf("Memory manager for task is gone, skipping application code dump.") + return + } + t.Debugf("Code:") + // Print code on both sides of the instruction register. + start := usermem.Addr(t.Arch().IP()) - maxCodeDebugBytes/2 + // Round addr down to a 16-byte boundary. + start &= ^usermem.Addr(15) + // Print 16 bytes per line, one byte at a time. + for offset := uint64(0); offset < maxCodeDebugBytes; offset += 16 { + addr, ok := start.AddLength(offset) + if !ok { + break + } + var data [16]byte + n, err := m.CopyIn(t, addr, data[:], usermem.IOOpts{ + IgnorePermissions: true, + }) + // Print as much of the line as we can, even if an error was + // encountered. + if n > 0 { + t.Debugf("%x: % x", addr, data[:n]) + } + if err != nil { + t.Debugf("Error reading stack at address %x: %v", addr+usermem.Addr(n), err) + break + } + } +} + // trace definitions. // // Note that all region names are prefixed by ':' in order to ensure that they diff --git a/pkg/sentry/kernel/task_run.go b/pkg/sentry/kernel/task_run.go index 7d4f44caf..abaf29216 100644 --- a/pkg/sentry/kernel/task_run.go +++ b/pkg/sentry/kernel/task_run.go @@ -260,7 +260,7 @@ func (app *runApp) execute(t *Task) taskRunState { region := trace.StartRegion(t.traceContext, runRegion) t.accountTaskGoroutineEnter(TaskGoroutineRunningApp) - info, at, err := t.p.Switch(t.MemoryManager().AddressSpace(), t.Arch(), t.rseqCPU) + info, at, err := t.p.Switch(t, t.MemoryManager(), t.Arch(), t.rseqCPU) t.accountTaskGoroutineLeave(TaskGoroutineRunningApp) region.End() diff --git a/pkg/sentry/kernel/task_signals.go b/pkg/sentry/kernel/task_signals.go index 79766cafe..2180fd27d 100644 --- a/pkg/sentry/kernel/task_signals.go +++ b/pkg/sentry/kernel/task_signals.go @@ -255,10 +255,11 @@ func (t *Task) deliverSignalToHandler(info *arch.SignalInfo, act arch.SignalAct) } } + mm := t.MemoryManager() // Set up the signal handler. If we have a saved signal mask, the signal // handler should run with the current mask, but sigreturn should restore // the saved one. - st := &arch.Stack{t.Arch(), t.MemoryManager(), sp} + st := &arch.Stack{t.Arch(), mm, sp} mask := t.signalMask if t.haveSavedSignalMask { mask = t.savedSignalMask @@ -273,12 +274,13 @@ func (t *Task) deliverSignalToHandler(info *arch.SignalInfo, act arch.SignalAct) // Please see the linux code as reference: // linux/arch/arm64/kernel/signal.c:setup_return() if act.Flags&linux.SA_RESTORER == 0 { - act.Restorer = t.MemoryManager().VDSOSigReturn() + act.Restorer = mm.VDSOSigReturn() } if err := t.Arch().SignalSetup(st, &act, info, &alt, mask); err != nil { return err } + t.p.FloatingPointStateChanged() t.haveSavedSignalMask = false // Add our signal mask. @@ -310,6 +312,7 @@ func (t *Task) SignalReturn(rt bool) (*SyscallControl, error) { // Restore our signal mask. SIGKILL and SIGSTOP should not be blocked. t.SetSignalMask(sigset &^ UnblockableSignals) + t.p.FloatingPointStateChanged() return ctrlResume, nil } @@ -636,6 +639,7 @@ func (t *Task) SetSavedSignalMask(mask linux.SignalSet) { // SignalStack returns the task-private signal stack. func (t *Task) SignalStack() arch.SignalStack { + t.p.PullFullState(t.MemoryManager().AddressSpace(), t.Arch()) alt := t.signalStack if t.onSignalStack(alt) { alt.Flags |= arch.SignalStackFlagOnStack @@ -1050,6 +1054,8 @@ func (*runInterrupt) execute(t *Task) taskRunState { // Are there signals pending? if info := t.dequeueSignalLocked(t.signalMask); info != nil { + t.p.PullFullState(t.MemoryManager().AddressSpace(), t.Arch()) + if linux.SignalSetOf(linux.Signal(info.Signo))&StopSignals != 0 { // Indicate that we've dequeued a stop signal before unlocking the // signal mutex; initiateGroupStop will check for races with diff --git a/pkg/sentry/kernel/task_stop.go b/pkg/sentry/kernel/task_stop.go index 10c6e455c..296735d32 100644 --- a/pkg/sentry/kernel/task_stop.go +++ b/pkg/sentry/kernel/task_stop.go @@ -205,6 +205,22 @@ func (ts *TaskSet) BeginExternalStop() { } } +// PullFullState receives full states for all tasks. +func (ts *TaskSet) PullFullState() { + ts.mu.Lock() + defer ts.mu.Unlock() + if ts.Root == nil { + return + } + for t := range ts.Root.tids { + t.Activate() + if mm := t.MemoryManager(); mm != nil { + t.p.PullFullState(t.MemoryManager().AddressSpace(), t.Arch()) + } + t.Deactivate() + } +} + // EndExternalStop indicates the end of an external stop started by a previous // call to TaskSet.BeginExternalStop. EndExternalStop does not wait for task // goroutines to resume. diff --git a/pkg/sentry/memmap/memmap.go b/pkg/sentry/memmap/memmap.go index 59c92c7e8..65d83096f 100644 --- a/pkg/sentry/memmap/memmap.go +++ b/pkg/sentry/memmap/memmap.go @@ -360,6 +360,13 @@ type MMapOpts struct { // // TODO(jamieliu): Replace entirely with MappingIdentity? Hint string + + // Force means to skip validation checks of Addr and Length. It can be + // used to create special mappings below mm.layout.MinAddr and + // mm.layout.MaxAddr. It has to be used with caution. + // + // If Force is true, Unmap and Fixed must be true. + Force bool } // File represents a host file that may be mapped into an platform.AddressSpace. diff --git a/pkg/sentry/mm/lifecycle.go b/pkg/sentry/mm/lifecycle.go index 4d7773f8b..09dbc06a4 100644 --- a/pkg/sentry/mm/lifecycle.go +++ b/pkg/sentry/mm/lifecycle.go @@ -57,6 +57,8 @@ func (mm *MemoryManager) SetMmapLayout(ac arch.Context, r *limits.LimitSet) (arc // Fork creates a copy of mm with 1 user, as for Linux syscalls fork() or // clone() (without CLONE_VM). func (mm *MemoryManager) Fork(ctx context.Context) (*MemoryManager, error) { + mm.AddressSpace().PreFork() + defer mm.AddressSpace().PostFork() mm.metadataMu.Lock() defer mm.metadataMu.Unlock() mm.mappingMu.RLock() diff --git a/pkg/sentry/mm/vma.go b/pkg/sentry/mm/vma.go index bd751d696..c4e1989ed 100644 --- a/pkg/sentry/mm/vma.go +++ b/pkg/sentry/mm/vma.go @@ -42,7 +42,12 @@ func (mm *MemoryManager) createVMALocked(ctx context.Context, opts memmap.MMapOp Map32Bit: opts.Map32Bit, }) if err != nil { - return vmaIterator{}, usermem.AddrRange{}, err + // Can't force without opts.Unmap and opts.Fixed. + if opts.Force && opts.Unmap && opts.Fixed { + addr = opts.Addr + } else { + return vmaIterator{}, usermem.AddrRange{}, err + } } ar, _ := addr.ToRange(opts.Length) diff --git a/pkg/sentry/platform/kvm/BUILD b/pkg/sentry/platform/kvm/BUILD index b5d27a72a..3970dd81d 100644 --- a/pkg/sentry/platform/kvm/BUILD +++ b/pkg/sentry/platform/kvm/BUILD @@ -41,6 +41,7 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/atomicbitops", + "//pkg/context", "//pkg/cpuid", "//pkg/log", "//pkg/procid", diff --git a/pkg/sentry/platform/kvm/address_space.go b/pkg/sentry/platform/kvm/address_space.go index 98a3e539d..af5c5e191 100644 --- a/pkg/sentry/platform/kvm/address_space.go +++ b/pkg/sentry/platform/kvm/address_space.go @@ -248,3 +248,9 @@ func (as *addressSpace) Release() { // Drop all cached machine references. as.machine.dropPageTables(as.pageTables) } + +// PreFork implements platform.AddressSpace.PreFork. +func (as *addressSpace) PreFork() {} + +// PostFork implements platform.AddressSpace.PostFork. +func (as *addressSpace) PostFork() {} diff --git a/pkg/sentry/platform/kvm/context.go b/pkg/sentry/platform/kvm/context.go index 6507121ea..eb92721fb 100644 --- a/pkg/sentry/platform/kvm/context.go +++ b/pkg/sentry/platform/kvm/context.go @@ -15,6 +15,7 @@ package kvm import ( + pkgcontext "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/interrupt" @@ -37,7 +38,8 @@ type context struct { } // Switch runs the provided context in the given address space. -func (c *context) Switch(as platform.AddressSpace, ac arch.Context, _ int32) (*arch.SignalInfo, usermem.AccessType, error) { +func (c *context) Switch(ctx pkgcontext.Context, mm platform.MemoryManager, ac arch.Context, _ int32) (*arch.SignalInfo, usermem.AccessType, error) { + as := mm.AddressSpace() localAS := as.(*addressSpace) // Grab a vCPU. @@ -88,3 +90,9 @@ func (c *context) Interrupt() { // Release implements platform.Context.Release(). func (c *context) Release() {} + +// FloatingPointStateChanged implements platform.Context.FloatingPointStateChanged. +func (c *context) FloatingPointStateChanged() {} + +// PullFullState implements platform.Context.PullFullState. +func (c *context) PullFullState(as platform.AddressSpace, ac arch.Context) {} diff --git a/pkg/sentry/platform/platform.go b/pkg/sentry/platform/platform.go index 4b13eec30..3f99afdd1 100644 --- a/pkg/sentry/platform/platform.go +++ b/pkg/sentry/platform/platform.go @@ -22,6 +22,7 @@ import ( "os" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/seccomp" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/memmap" @@ -114,6 +115,17 @@ func (NoCPUPreemptionDetection) PreemptAllCPUs() error { panic("This platform does not support CPU preemption detection") } +// MemoryManager represents an abstraction above the platform address space +// which manages memory mappings and their contents. +type MemoryManager interface { + //usermem.IO provides access to the contents of a virtual memory space. + usermem.IO + // MMap establishes a memory mapping. + MMap(ctx context.Context, opts memmap.MMapOpts) (usermem.Addr, error) + // AddressSpace returns the AddressSpace bound to mm. + AddressSpace() AddressSpace +} + // Context represents the execution context for a single thread. type Context interface { // Switch resumes execution of the thread specified by the arch.Context @@ -143,7 +155,30 @@ type Context interface { // concurrent call to Switch(). // // - ErrContextCPUPreempted: See the definition of that error for details. - Switch(as AddressSpace, ac arch.Context, cpu int32) (*arch.SignalInfo, usermem.AccessType, error) + Switch(ctx context.Context, mm MemoryManager, ac arch.Context, cpu int32) (*arch.SignalInfo, usermem.AccessType, error) + + // PullFullState() pulls a full state of the application thread. + // + // A platform can support lazy loading/restoring of a thread state + // which includes registers and a floating point state. + // + // For example, when the Sentry handles a system call, it may have only + // syscall arguments without other registers and a floating point + // state. And in this case, if the Sentry will need to construct a + // signal frame to call a signal handler, it will need to call + // PullFullState() to load all registers and FPU state. + // + // Preconditions: The caller must be running on the task goroutine. + PullFullState(as AddressSpace, ac arch.Context) + + // FloatingPointStateChanged forces restoring a full state of the application thread. + // + // A platform can support lazy loading/restoring of a thread state. + // This means that if the Sentry has not changed a thread state, + // the platform may not restore it. + // + // Preconditions: The caller must be running on the task goroutine. + FloatingPointStateChanged() // Interrupt interrupts a concurrent call to Switch(), causing it to return // ErrContextInterrupt. @@ -218,6 +253,13 @@ type AddressSpace interface { // must be acquired via platform.NewAddressSpace(). Release() + // PreFork() is called before creating a copy of AddressSpace. This + // guarantees that this address space will be in a consistent state. + PreFork() + + // PostFork() is called after creating a copy of AddressSpace. + PostFork() + // AddressSpaceIO methods are supported iff the associated platform's // Platform.SupportsAddressSpaceIO() == true. AddressSpaces for which this // does not hold may panic if AddressSpaceIO methods are invoked. diff --git a/pkg/sentry/platform/ptrace/BUILD b/pkg/sentry/platform/ptrace/BUILD index 29fd23cc3..e04165fbf 100644 --- a/pkg/sentry/platform/ptrace/BUILD +++ b/pkg/sentry/platform/ptrace/BUILD @@ -24,6 +24,7 @@ go_library( visibility = ["//:sandbox"], deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/log", "//pkg/procid", "//pkg/safecopy", diff --git a/pkg/sentry/platform/ptrace/ptrace.go b/pkg/sentry/platform/ptrace/ptrace.go index 08d055e05..45ff2bcc3 100644 --- a/pkg/sentry/platform/ptrace/ptrace.go +++ b/pkg/sentry/platform/ptrace/ptrace.go @@ -48,6 +48,7 @@ import ( "os" "gvisor.dev/gvisor/pkg/abi/linux" + pkgcontext "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/interrupt" @@ -95,7 +96,8 @@ type context struct { } // Switch runs the provided context in the given address space. -func (c *context) Switch(as platform.AddressSpace, ac arch.Context, cpu int32) (*arch.SignalInfo, usermem.AccessType, error) { +func (c *context) Switch(ctx pkgcontext.Context, mm platform.MemoryManager, ac arch.Context, cpu int32) (*arch.SignalInfo, usermem.AccessType, error) { + as := mm.AddressSpace() s := as.(*subprocess) isSyscall := s.switchToApp(c, ac) @@ -180,6 +182,12 @@ func (c *context) Interrupt() { // Release implements platform.Context.Release(). func (c *context) Release() {} +// FloatingPointStateChanged implements platform.Context.FloatingPointStateChanged. +func (c *context) FloatingPointStateChanged() {} + +// PullFullState implements platform.Context.PullFullState. +func (c *context) PullFullState(as platform.AddressSpace, ac arch.Context) {} + // PTrace represents a collection of ptrace subprocesses. type PTrace struct { platform.MMapMinAddr diff --git a/pkg/sentry/platform/ptrace/subprocess.go b/pkg/sentry/platform/ptrace/subprocess.go index c990f3454..e1d54d8a2 100644 --- a/pkg/sentry/platform/ptrace/subprocess.go +++ b/pkg/sentry/platform/ptrace/subprocess.go @@ -662,3 +662,9 @@ func (s *subprocess) Unmap(addr usermem.Addr, length uint64) { panic(fmt.Sprintf("munmap(%x, %x)) failed: %v", addr, length, err)) } } + +// PreFork implements platform.AddressSpace.PreFork. +func (s *subprocess) PreFork() {} + +// PostFork implements platform.AddressSpace.PostFork. +func (s *subprocess) PostFork() {} diff --git a/pkg/sentry/state/state.go b/pkg/sentry/state/state.go index 9eb626b76..a06c9b8ab 100644 --- a/pkg/sentry/state/state.go +++ b/pkg/sentry/state/state.go @@ -60,6 +60,7 @@ type SaveOpts struct { func (opts SaveOpts) Save(k *kernel.Kernel, w *watchdog.Watchdog) error { log.Infof("Sandbox save started, pausing all tasks.") k.Pause() + k.ReceiveTaskStates() defer k.Unpause() defer log.Infof("Tasks resumed after save.") -- cgit v1.2.3 From d64ba89da3a5e8aaf28a9da1b40cb2231a4c6b42 Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Tue, 4 Aug 2020 09:29:11 -0700 Subject: Internal change. PiperOrigin-RevId: 324826968 --- pkg/sentry/fsimpl/testutil/kernel.go | 4 ++++ 1 file changed, 4 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/testutil/kernel.go b/pkg/sentry/fsimpl/testutil/kernel.go index 1e57744e8..1813269e0 100644 --- a/pkg/sentry/fsimpl/testutil/kernel.go +++ b/pkg/sentry/fsimpl/testutil/kernel.go @@ -122,6 +122,10 @@ func Boot() (*kernel.Kernel, error) { // CreateTask creates a new bare bones task for tests. func CreateTask(ctx context.Context, name string, tc *kernel.ThreadGroup, mntns *vfs.MountNamespace, root, cwd vfs.VirtualDentry) (*kernel.Task, error) { k := kernel.KernelFromContext(ctx) + if k == nil { + return nil, fmt.Errorf("cannot find kernel from context") + } + exe, err := newFakeExecutable(ctx, k.VFS(), auth.CredentialsFromContext(ctx), root) if err != nil { return nil, err -- cgit v1.2.3 From 0500f84b6f2e279ce953faaff20454aca69ae493 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 4 Aug 2020 15:46:33 -0700 Subject: Add reference counting utility to VFS2. The utility has several differences from the VFS1 equivalent: - There are no weak references, which have a significant overhead - In order to print useful debug messages with the type of the reference- counted object, we use a generic Refs object with the owner type as a template parameter. In vfs1, this was accomplished by storing a type name and caller stack directly in the ref count (as in vfs1), which increases the struct size by 6x. (Note that the caller stack was needed because fs types like Dirent were shared by all fs implementations; in vfs2, each impl has its own data structures, so this is no longer necessary.) As an example, the utility is added to tmpfs.inode. Updates #1486. PiperOrigin-RevId: 324906582 --- pkg/refs/refcounter.go | 7 +++ pkg/refs_vfs2/BUILD | 28 +++++++++ pkg/refs_vfs2/refs.go | 36 +++++++++++ pkg/refs_vfs2/refs_template.go | 133 +++++++++++++++++++++++++++++++++++++++ pkg/sentry/fsimpl/tmpfs/BUILD | 13 ++++ pkg/sentry/fsimpl/tmpfs/tmpfs.go | 27 ++------ 6 files changed, 223 insertions(+), 21 deletions(-) create mode 100644 pkg/refs_vfs2/BUILD create mode 100644 pkg/refs_vfs2/refs.go create mode 100644 pkg/refs_vfs2/refs_template.go (limited to 'pkg/sentry') diff --git a/pkg/refs/refcounter.go b/pkg/refs/refcounter.go index 61790221b..3f39edb66 100644 --- a/pkg/refs/refcounter.go +++ b/pkg/refs/refcounter.go @@ -215,6 +215,8 @@ type AtomicRefCount struct { // LeakMode configures the leak checker. type LeakMode uint32 +// TODO(gvisor.dev/issue/1624): Simplify down to two modes once vfs1 ref +// counting is gone. const ( // UninitializedLeakChecking indicates that the leak checker has not yet been initialized. UninitializedLeakChecking LeakMode = iota @@ -244,6 +246,11 @@ func SetLeakMode(mode LeakMode) { atomic.StoreUint32(&leakMode, uint32(mode)) } +// GetLeakMode returns the current leak mode. +func GetLeakMode() LeakMode { + return LeakMode(atomic.LoadUint32(&leakMode)) +} + const maxStackFrames = 40 type fileLine struct { diff --git a/pkg/refs_vfs2/BUILD b/pkg/refs_vfs2/BUILD new file mode 100644 index 000000000..7f180c7bd --- /dev/null +++ b/pkg/refs_vfs2/BUILD @@ -0,0 +1,28 @@ +load("//tools:defs.bzl", "go_library") +load("//tools/go_generics:defs.bzl", "go_template") + +package(licenses = ["notice"]) + +go_template( + name = "refs_template", + srcs = [ + "refs_template.go", + ], + types = [ + "T", + ], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/log", + "//pkg/refs", + ], +) + +go_library( + name = "refs", + srcs = [ + "refs.go", + ], + visibility = ["//pkg/sentry:internal"], + deps = ["//pkg/context"], +) diff --git a/pkg/refs_vfs2/refs.go b/pkg/refs_vfs2/refs.go new file mode 100644 index 000000000..ee01b17b0 --- /dev/null +++ b/pkg/refs_vfs2/refs.go @@ -0,0 +1,36 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package refs defines an interface for a reference-counted object. +package refs + +import ( + "gvisor.dev/gvisor/pkg/context" +) + +// RefCounter is the interface to be implemented by objects that are reference +// counted. +type RefCounter interface { + // IncRef increments the reference counter on the object. + IncRef() + + // DecRef decrements the object's reference count. Users of refs_template.Refs + // may specify a destructor to be called once the reference count reaches zero. + DecRef(ctx context.Context) + + // TryIncRef attempts to increment the reference count, but may fail if all + // references have already been dropped, in which case it returns false. If + // true is returned, then a valid reference is now held on the object. + TryIncRef() bool +} diff --git a/pkg/refs_vfs2/refs_template.go b/pkg/refs_vfs2/refs_template.go new file mode 100644 index 000000000..3e5b458c7 --- /dev/null +++ b/pkg/refs_vfs2/refs_template.go @@ -0,0 +1,133 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package refs_template defines a template that can be used by reference counted +// objects. +package refs_template + +import ( + "runtime" + "sync/atomic" + + "gvisor.dev/gvisor/pkg/log" + refs_vfs1 "gvisor.dev/gvisor/pkg/refs" +) + +// T is the type of the reference counted object. It is only used to customize +// debug output when leak checking. +type T interface{} + +// ownerType is used to customize logging. Note that we use a pointer to T so +// that we do not copy the entire object when passed as a format parameter. +var ownerType *T + +// Refs implements refs.RefCounter. It keeps a reference count using atomic +// operations and calls the destructor when the count reaches zero. +// +// Note that the number of references is actually refCount + 1 so that a default +// zero-value Refs object contains one reference. +// +// +stateify savable +type Refs struct { + // refCount is composed of two fields: + // + // [32-bit speculative references]:[32-bit real references] + // + // Speculative references are used for TryIncRef, to avoid a CompareAndSwap + // loop. See IncRef, DecRef and TryIncRef for details of how these fields are + // used. + refCount int64 +} + +func (r *Refs) finalize() { + var note string + switch refs_vfs1.GetLeakMode() { + case refs_vfs1.NoLeakChecking: + return + case refs_vfs1.UninitializedLeakChecking: + note = "(Leak checker uninitialized): " + } + if n := r.ReadRefs(); n != 0 { + log.Warningf("%sAtomicRefCount %p owned by %T garbage collected with ref count of %d (want 0)", note, r, ownerType, n) + } +} + +// EnableLeakCheck checks for reference leaks when Refs gets garbage collected. +func (r *Refs) EnableLeakCheck() { + if refs_vfs1.GetLeakMode() != refs_vfs1.NoLeakChecking { + runtime.SetFinalizer(r, (*Refs).finalize) + } +} + +// ReadRefs returns the current number of references. The returned count is +// inherently racy and is unsafe to use without external synchronization. +func (r *Refs) ReadRefs() int64 { + // Account for the internal -1 offset on refcounts. + return atomic.LoadInt64(&r.refCount) + 1 +} + +// IncRef implements refs.RefCounter.IncRef. +// +//go:nosplit +func (r *Refs) IncRef() { + if v := atomic.AddInt64(&r.refCount, 1); v <= 0 { + panic("Incrementing non-positive ref count") + } +} + +// TryIncRef implements refs.RefCounter.TryIncRef. +// +// To do this safely without a loop, a speculative reference is first acquired +// on the object. This allows multiple concurrent TryIncRef calls to distinguish +// other TryIncRef calls from genuine references held. +// +//go:nosplit +func (r *Refs) TryIncRef() bool { + const speculativeRef = 1 << 32 + v := atomic.AddInt64(&r.refCount, speculativeRef) + if int32(v) < 0 { + // This object has already been freed. + atomic.AddInt64(&r.refCount, -speculativeRef) + return false + } + + // Turn into a real reference. + atomic.AddInt64(&r.refCount, -speculativeRef+1) + return true +} + +// DecRef implements refs.RefCounter.DecRef. +// +// Note that speculative references are counted here. Since they were added +// prior to real references reaching zero, they will successfully convert to +// real references. In other words, we see speculative references only in the +// following case: +// +// A: TryIncRef [speculative increase => sees non-negative references] +// B: DecRef [real decrease] +// A: TryIncRef [transform speculative to real] +// +//go:nosplit +func (r *Refs) DecRef(destroy func()) { + switch v := atomic.AddInt64(&r.refCount, -1); { + case v < -1: + panic("Decrementing non-positive ref count") + + case v == -1: + // Call the destructor. + if destroy != nil { + destroy() + } + } +} diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD index e73732a6b..5cd428d64 100644 --- a/pkg/sentry/fsimpl/tmpfs/BUILD +++ b/pkg/sentry/fsimpl/tmpfs/BUILD @@ -26,6 +26,17 @@ go_template_instance( }, ) +go_template_instance( + name = "inode_refs", + out = "inode_refs.go", + package = "tmpfs", + prefix = "inode", + template = "//pkg/refs_vfs2:refs_template", + types = { + "T": "inode", + }, +) + go_library( name = "tmpfs", srcs = [ @@ -34,6 +45,7 @@ go_library( "directory.go", "filesystem.go", "fstree.go", + "inode_refs.go", "named_pipe.go", "regular_file.go", "socket_file.go", @@ -47,6 +59,7 @@ go_library( "//pkg/context", "//pkg/fspath", "//pkg/log", + "//pkg/refs", "//pkg/safemem", "//pkg/sentry/arch", "//pkg/sentry/fs", diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 68e615e8b..5640380dc 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -285,13 +285,10 @@ type inode struct { // fs is the owning filesystem. fs is immutable. fs *filesystem - // refs is a reference count. refs is accessed using atomic memory - // operations. - // // A reference is held on all inodes as long as they are reachable in the // filesystem tree, i.e. nlink is nonzero. This reference is dropped when // nlink reaches 0. - refs int64 + refs inodeRefs // xattrs implements extended attributes. // @@ -327,7 +324,6 @@ func (i *inode) init(impl interface{}, fs *filesystem, kuid auth.KUID, kgid auth panic("file type is required in FileMode") } i.fs = fs - i.refs = 1 i.mode = uint32(mode) i.uid = uint32(kuid) i.gid = uint32(kgid) @@ -339,6 +335,7 @@ func (i *inode) init(impl interface{}, fs *filesystem, kuid auth.KUID, kgid auth i.mtime = now // i.nlink initialized by caller i.impl = impl + i.refs.EnableLeakCheck() } // incLinksLocked increments i's link count. @@ -369,25 +366,15 @@ func (i *inode) decLinksLocked(ctx context.Context) { } func (i *inode) incRef() { - if atomic.AddInt64(&i.refs, 1) <= 1 { - panic("tmpfs.inode.incRef() called without holding a reference") - } + i.refs.IncRef() } func (i *inode) tryIncRef() bool { - for { - refs := atomic.LoadInt64(&i.refs) - if refs == 0 { - return false - } - if atomic.CompareAndSwapInt64(&i.refs, refs, refs+1) { - return true - } - } + return i.refs.TryIncRef() } func (i *inode) decRef(ctx context.Context) { - if refs := atomic.AddInt64(&i.refs, -1); refs == 0 { + i.refs.DecRef(func() { i.watches.HandleDeletion(ctx) if regFile, ok := i.impl.(*regularFile); ok { // Release memory used by regFile to store data. Since regFile is @@ -395,9 +382,7 @@ func (i *inode) decRef(ctx context.Context) { // metadata. regFile.data.DropAll(regFile.memFile) } - } else if refs < 0 { - panic("tmpfs.inode.decRef() called without holding a reference") - } + }) } func (i *inode) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes) error { -- cgit v1.2.3 From b44408b40e3e8762a77ccf5eeb7f2ef567235c43 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 4 Aug 2020 18:18:01 -0700 Subject: Automated rollback of changelist 324906582 PiperOrigin-RevId: 324931854 --- pkg/refs/refcounter.go | 7 --- pkg/refs_vfs2/BUILD | 28 --------- pkg/refs_vfs2/refs.go | 36 ----------- pkg/refs_vfs2/refs_template.go | 133 --------------------------------------- pkg/sentry/fsimpl/tmpfs/BUILD | 13 ---- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 27 ++++++-- 6 files changed, 21 insertions(+), 223 deletions(-) delete mode 100644 pkg/refs_vfs2/BUILD delete mode 100644 pkg/refs_vfs2/refs.go delete mode 100644 pkg/refs_vfs2/refs_template.go (limited to 'pkg/sentry') diff --git a/pkg/refs/refcounter.go b/pkg/refs/refcounter.go index 3f39edb66..61790221b 100644 --- a/pkg/refs/refcounter.go +++ b/pkg/refs/refcounter.go @@ -215,8 +215,6 @@ type AtomicRefCount struct { // LeakMode configures the leak checker. type LeakMode uint32 -// TODO(gvisor.dev/issue/1624): Simplify down to two modes once vfs1 ref -// counting is gone. const ( // UninitializedLeakChecking indicates that the leak checker has not yet been initialized. UninitializedLeakChecking LeakMode = iota @@ -246,11 +244,6 @@ func SetLeakMode(mode LeakMode) { atomic.StoreUint32(&leakMode, uint32(mode)) } -// GetLeakMode returns the current leak mode. -func GetLeakMode() LeakMode { - return LeakMode(atomic.LoadUint32(&leakMode)) -} - const maxStackFrames = 40 type fileLine struct { diff --git a/pkg/refs_vfs2/BUILD b/pkg/refs_vfs2/BUILD deleted file mode 100644 index 7f180c7bd..000000000 --- a/pkg/refs_vfs2/BUILD +++ /dev/null @@ -1,28 +0,0 @@ -load("//tools:defs.bzl", "go_library") -load("//tools/go_generics:defs.bzl", "go_template") - -package(licenses = ["notice"]) - -go_template( - name = "refs_template", - srcs = [ - "refs_template.go", - ], - types = [ - "T", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/log", - "//pkg/refs", - ], -) - -go_library( - name = "refs", - srcs = [ - "refs.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = ["//pkg/context"], -) diff --git a/pkg/refs_vfs2/refs.go b/pkg/refs_vfs2/refs.go deleted file mode 100644 index ee01b17b0..000000000 --- a/pkg/refs_vfs2/refs.go +++ /dev/null @@ -1,36 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package refs defines an interface for a reference-counted object. -package refs - -import ( - "gvisor.dev/gvisor/pkg/context" -) - -// RefCounter is the interface to be implemented by objects that are reference -// counted. -type RefCounter interface { - // IncRef increments the reference counter on the object. - IncRef() - - // DecRef decrements the object's reference count. Users of refs_template.Refs - // may specify a destructor to be called once the reference count reaches zero. - DecRef(ctx context.Context) - - // TryIncRef attempts to increment the reference count, but may fail if all - // references have already been dropped, in which case it returns false. If - // true is returned, then a valid reference is now held on the object. - TryIncRef() bool -} diff --git a/pkg/refs_vfs2/refs_template.go b/pkg/refs_vfs2/refs_template.go deleted file mode 100644 index 3e5b458c7..000000000 --- a/pkg/refs_vfs2/refs_template.go +++ /dev/null @@ -1,133 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package refs_template defines a template that can be used by reference counted -// objects. -package refs_template - -import ( - "runtime" - "sync/atomic" - - "gvisor.dev/gvisor/pkg/log" - refs_vfs1 "gvisor.dev/gvisor/pkg/refs" -) - -// T is the type of the reference counted object. It is only used to customize -// debug output when leak checking. -type T interface{} - -// ownerType is used to customize logging. Note that we use a pointer to T so -// that we do not copy the entire object when passed as a format parameter. -var ownerType *T - -// Refs implements refs.RefCounter. It keeps a reference count using atomic -// operations and calls the destructor when the count reaches zero. -// -// Note that the number of references is actually refCount + 1 so that a default -// zero-value Refs object contains one reference. -// -// +stateify savable -type Refs struct { - // refCount is composed of two fields: - // - // [32-bit speculative references]:[32-bit real references] - // - // Speculative references are used for TryIncRef, to avoid a CompareAndSwap - // loop. See IncRef, DecRef and TryIncRef for details of how these fields are - // used. - refCount int64 -} - -func (r *Refs) finalize() { - var note string - switch refs_vfs1.GetLeakMode() { - case refs_vfs1.NoLeakChecking: - return - case refs_vfs1.UninitializedLeakChecking: - note = "(Leak checker uninitialized): " - } - if n := r.ReadRefs(); n != 0 { - log.Warningf("%sAtomicRefCount %p owned by %T garbage collected with ref count of %d (want 0)", note, r, ownerType, n) - } -} - -// EnableLeakCheck checks for reference leaks when Refs gets garbage collected. -func (r *Refs) EnableLeakCheck() { - if refs_vfs1.GetLeakMode() != refs_vfs1.NoLeakChecking { - runtime.SetFinalizer(r, (*Refs).finalize) - } -} - -// ReadRefs returns the current number of references. The returned count is -// inherently racy and is unsafe to use without external synchronization. -func (r *Refs) ReadRefs() int64 { - // Account for the internal -1 offset on refcounts. - return atomic.LoadInt64(&r.refCount) + 1 -} - -// IncRef implements refs.RefCounter.IncRef. -// -//go:nosplit -func (r *Refs) IncRef() { - if v := atomic.AddInt64(&r.refCount, 1); v <= 0 { - panic("Incrementing non-positive ref count") - } -} - -// TryIncRef implements refs.RefCounter.TryIncRef. -// -// To do this safely without a loop, a speculative reference is first acquired -// on the object. This allows multiple concurrent TryIncRef calls to distinguish -// other TryIncRef calls from genuine references held. -// -//go:nosplit -func (r *Refs) TryIncRef() bool { - const speculativeRef = 1 << 32 - v := atomic.AddInt64(&r.refCount, speculativeRef) - if int32(v) < 0 { - // This object has already been freed. - atomic.AddInt64(&r.refCount, -speculativeRef) - return false - } - - // Turn into a real reference. - atomic.AddInt64(&r.refCount, -speculativeRef+1) - return true -} - -// DecRef implements refs.RefCounter.DecRef. -// -// Note that speculative references are counted here. Since they were added -// prior to real references reaching zero, they will successfully convert to -// real references. In other words, we see speculative references only in the -// following case: -// -// A: TryIncRef [speculative increase => sees non-negative references] -// B: DecRef [real decrease] -// A: TryIncRef [transform speculative to real] -// -//go:nosplit -func (r *Refs) DecRef(destroy func()) { - switch v := atomic.AddInt64(&r.refCount, -1); { - case v < -1: - panic("Decrementing non-positive ref count") - - case v == -1: - // Call the destructor. - if destroy != nil { - destroy() - } - } -} diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD index 5cd428d64..e73732a6b 100644 --- a/pkg/sentry/fsimpl/tmpfs/BUILD +++ b/pkg/sentry/fsimpl/tmpfs/BUILD @@ -26,17 +26,6 @@ go_template_instance( }, ) -go_template_instance( - name = "inode_refs", - out = "inode_refs.go", - package = "tmpfs", - prefix = "inode", - template = "//pkg/refs_vfs2:refs_template", - types = { - "T": "inode", - }, -) - go_library( name = "tmpfs", srcs = [ @@ -45,7 +34,6 @@ go_library( "directory.go", "filesystem.go", "fstree.go", - "inode_refs.go", "named_pipe.go", "regular_file.go", "socket_file.go", @@ -59,7 +47,6 @@ go_library( "//pkg/context", "//pkg/fspath", "//pkg/log", - "//pkg/refs", "//pkg/safemem", "//pkg/sentry/arch", "//pkg/sentry/fs", diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 5640380dc..68e615e8b 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -285,10 +285,13 @@ type inode struct { // fs is the owning filesystem. fs is immutable. fs *filesystem + // refs is a reference count. refs is accessed using atomic memory + // operations. + // // A reference is held on all inodes as long as they are reachable in the // filesystem tree, i.e. nlink is nonzero. This reference is dropped when // nlink reaches 0. - refs inodeRefs + refs int64 // xattrs implements extended attributes. // @@ -324,6 +327,7 @@ func (i *inode) init(impl interface{}, fs *filesystem, kuid auth.KUID, kgid auth panic("file type is required in FileMode") } i.fs = fs + i.refs = 1 i.mode = uint32(mode) i.uid = uint32(kuid) i.gid = uint32(kgid) @@ -335,7 +339,6 @@ func (i *inode) init(impl interface{}, fs *filesystem, kuid auth.KUID, kgid auth i.mtime = now // i.nlink initialized by caller i.impl = impl - i.refs.EnableLeakCheck() } // incLinksLocked increments i's link count. @@ -366,15 +369,25 @@ func (i *inode) decLinksLocked(ctx context.Context) { } func (i *inode) incRef() { - i.refs.IncRef() + if atomic.AddInt64(&i.refs, 1) <= 1 { + panic("tmpfs.inode.incRef() called without holding a reference") + } } func (i *inode) tryIncRef() bool { - return i.refs.TryIncRef() + for { + refs := atomic.LoadInt64(&i.refs) + if refs == 0 { + return false + } + if atomic.CompareAndSwapInt64(&i.refs, refs, refs+1) { + return true + } + } } func (i *inode) decRef(ctx context.Context) { - i.refs.DecRef(func() { + if refs := atomic.AddInt64(&i.refs, -1); refs == 0 { i.watches.HandleDeletion(ctx) if regFile, ok := i.impl.(*regularFile); ok { // Release memory used by regFile to store data. Since regFile is @@ -382,7 +395,9 @@ func (i *inode) decRef(ctx context.Context) { // metadata. regFile.data.DropAll(regFile.memFile) } - }) + } else if refs < 0 { + panic("tmpfs.inode.decRef() called without holding a reference") + } } func (i *inode) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes) error { -- cgit v1.2.3 From 102735bfb45820dd840df14827b42744da77c9a0 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Tue, 4 Aug 2020 19:04:12 -0700 Subject: Inline gofer.regularFileFD.pwriteLocked Go compiler barely inlines anything, so inline by hand pwriteLocked since it's called from a single place. PiperOrigin-RevId: 324937734 --- pkg/sentry/fsimpl/gofer/regular_file.go | 102 ++++++++++++++++++-------------- 1 file changed, 56 insertions(+), 46 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go index 420e8efe2..db6bed4f6 100644 --- a/pkg/sentry/fsimpl/gofer/regular_file.go +++ b/pkg/sentry/fsimpl/gofer/regular_file.go @@ -184,6 +184,7 @@ func (fd *regularFileFD) pwrite(ctx context.Context, src usermem.IOSequence, off d.metadataMu.Lock() defer d.metadataMu.Unlock() + // Set offset to file size if the fd was opened with O_APPEND. if fd.vfsfd.StatusFlags()&linux.O_APPEND != 0 { // Holding d.metadataMu is sufficient for reading d.size. @@ -194,70 +195,79 @@ func (fd *regularFileFD) pwrite(ctx context.Context, src usermem.IOSequence, off return 0, offset, err } src = src.TakeFirst64(limit) - n, err := fd.pwriteLocked(ctx, src, offset, opts) - return n, offset + n, err -} -// Preconditions: fd.dentry().metatdataMu must be locked. -func (fd *regularFileFD) pwriteLocked(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - d := fd.dentry() if d.fs.opts.interop != InteropModeShared { // Compare Linux's mm/filemap.c:__generic_file_write_iter() => // file_update_time(). This is d.touchCMtime(), but without locking // d.metadataMu (recursively). d.touchCMtimeLocked() } - if fd.vfsfd.StatusFlags()&linux.O_DIRECT != 0 { - // Write dirty cached pages that will be touched by the write back to - // the remote file. - if err := d.writeback(ctx, offset, src.NumBytes()); err != nil { - return 0, err - } - // Remove touched pages from the cache. - pgstart := usermem.PageRoundDown(uint64(offset)) - pgend, ok := usermem.PageRoundUp(uint64(offset + src.NumBytes())) - if !ok { - return 0, syserror.EINVAL - } - mr := memmap.MappableRange{pgstart, pgend} - var freed []memmap.FileRange - d.dataMu.Lock() - cseg := d.cache.LowerBoundSegment(mr.Start) - for cseg.Ok() && cseg.Start() < mr.End { - cseg = d.cache.Isolate(cseg, mr) - freed = append(freed, memmap.FileRange{cseg.Value(), cseg.Value() + cseg.Range().Length()}) - cseg = d.cache.Remove(cseg).NextSegment() - } - d.dataMu.Unlock() - // Invalidate mappings of removed pages. - d.mapsMu.Lock() - d.mappings.Invalidate(mr, memmap.InvalidateOpts{}) - d.mapsMu.Unlock() - // Finally free pages removed from the cache. - mf := d.fs.mfp.MemoryFile() - for _, freedFR := range freed { - mf.DecRef(freedFR) - } - } + rw := getDentryReadWriter(ctx, d, offset) + defer putDentryReadWriter(rw) + if fd.vfsfd.StatusFlags()&linux.O_DIRECT != 0 { + if err := fd.writeCache(ctx, d, offset, src); err != nil { + return 0, offset, err + } + // Require the write to go to the remote file. rw.direct = true } + n, err := src.CopyInTo(ctx, rw) - putDentryReadWriter(rw) - if n != 0 && fd.vfsfd.StatusFlags()&(linux.O_DSYNC|linux.O_SYNC) != 0 { - // Write dirty cached pages touched by the write back to the remote - // file. + if err != nil { + return n, offset, err + } + if n > 0 && fd.vfsfd.StatusFlags()&(linux.O_DSYNC|linux.O_SYNC) != 0 { + // Write dirty cached pages touched by the write back to the remote file. if err := d.writeback(ctx, offset, src.NumBytes()); err != nil { - return 0, err + return n, offset, err } // Request the remote filesystem to sync the remote file. - if err := d.handle.file.fsync(ctx); err != nil { - return 0, err + if err := d.handle.sync(ctx); err != nil { + return n, offset, err } } - return n, err + return n, offset + n, nil +} + +func (fd *regularFileFD) writeCache(ctx context.Context, d *dentry, offset int64, src usermem.IOSequence) error { + // Write dirty cached pages that will be touched by the write back to + // the remote file. + if err := d.writeback(ctx, offset, src.NumBytes()); err != nil { + return err + } + + // Remove touched pages from the cache. + pgstart := usermem.PageRoundDown(uint64(offset)) + pgend, ok := usermem.PageRoundUp(uint64(offset + src.NumBytes())) + if !ok { + return syserror.EINVAL + } + mr := memmap.MappableRange{pgstart, pgend} + var freed []memmap.FileRange + + d.dataMu.Lock() + cseg := d.cache.LowerBoundSegment(mr.Start) + for cseg.Ok() && cseg.Start() < mr.End { + cseg = d.cache.Isolate(cseg, mr) + freed = append(freed, memmap.FileRange{cseg.Value(), cseg.Value() + cseg.Range().Length()}) + cseg = d.cache.Remove(cseg).NextSegment() + } + d.dataMu.Unlock() + + // Invalidate mappings of removed pages. + d.mapsMu.Lock() + d.mappings.Invalidate(mr, memmap.InvalidateOpts{}) + d.mapsMu.Unlock() + + // Finally free pages removed from the cache. + mf := d.fs.mfp.MemoryFile() + for _, freedFR := range freed { + mf.DecRef(freedFR) + } + return nil } // Write implements vfs.FileDescriptionImpl.Write. -- cgit v1.2.3 From 87ee3898f78cb7d520281ead600036052d6a2049 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Tue, 4 Aug 2020 19:10:28 -0700 Subject: Handle EOF in vfs2 sendfile. Discovered by syzkaller. PiperOrigin-RevId: 324938438 --- pkg/sentry/syscalls/linux/vfs2/splice.go | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/vfs2/splice.go b/pkg/sentry/syscalls/linux/vfs2/splice.go index 16f59fce9..75bfa2c79 100644 --- a/pkg/sentry/syscalls/linux/vfs2/splice.go +++ b/pkg/sentry/syscalls/linux/vfs2/splice.go @@ -347,6 +347,11 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc } else { spliceN, err = inFile.Read(t, outPipeFD.IOSequence(count), vfs.ReadOptions{}) } + if spliceN == 0 && err == io.EOF { + // We reached the end of the file. Eat the error and exit the loop. + err = nil + break + } n += spliceN if err == syserror.ErrWouldBlock && !nonBlock { err = dw.waitForBoth(t) @@ -367,8 +372,7 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc readN, err = inFile.Read(t, usermem.BytesIOSequence(buf), vfs.ReadOptions{}) } if readN == 0 && err == io.EOF { - // We reached the end of the file. Eat the - // error and exit the loop. + // We reached the end of the file. Eat the error and exit the loop. err = nil break } -- cgit v1.2.3 From a2e129b540e530c56cb7648ae7e04325c6e763b1 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Wed, 5 Aug 2020 08:33:52 -0700 Subject: Add missing case in tmpfs.inode.direntType. This was discovered by syzkaller. PiperOrigin-RevId: 325025193 --- pkg/sentry/fsimpl/tmpfs/tmpfs.go | 2 ++ 1 file changed, 2 insertions(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 68e615e8b..4681a2f52 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -558,6 +558,8 @@ func (i *inode) direntType() uint8 { return linux.DT_LNK case *socketFile: return linux.DT_SOCK + case *namedPipe: + return linux.DT_FIFO case *deviceFile: switch impl.kind { case vfs.BlockDevice: -- cgit v1.2.3 From 190b1e6bd4aae56eff3ff846efea38629361b3a9 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Wed, 5 Aug 2020 11:28:15 -0700 Subject: Stop profiling when the sentry exits Also removes `--profile-goroutine` because it's equivalent to `debug --stacks`. PiperOrigin-RevId: 325061502 --- pkg/sentry/control/pprof.go | 6 ++--- pkg/test/dockerutil/profile.go | 37 ++++++++++++--------------- pkg/test/dockerutil/profile_test.go | 13 +++++----- runsc/boot/controller.go | 51 ++++++++++++++++++++++--------------- runsc/boot/loader.go | 3 +++ runsc/cmd/debug.go | 40 ++++++++++------------------- runsc/sandbox/sandbox.go | 20 --------------- 7 files changed, 71 insertions(+), 99 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/control/pprof.go b/pkg/sentry/control/pprof.go index 663e51989..2bf3c45e1 100644 --- a/pkg/sentry/control/pprof.go +++ b/pkg/sentry/control/pprof.go @@ -49,6 +49,9 @@ type ProfileOpts struct { // - dump out the stack trace of current go routines. // sentryctl -pid pprof-goroutine type Profile struct { + // Kernel is the kernel under profile. It's immutable. + Kernel *kernel.Kernel + // mu protects the fields below. mu sync.Mutex @@ -57,9 +60,6 @@ type Profile struct { // traceFile is the current execution trace output file. traceFile *fd.FD - - // Kernel is the kernel under profile. - Kernel *kernel.Kernel } // StartCPUProfile is an RPC stub which starts recording the CPU profile in a diff --git a/pkg/test/dockerutil/profile.go b/pkg/test/dockerutil/profile.go index 1fab33083..f0396ef24 100644 --- a/pkg/test/dockerutil/profile.go +++ b/pkg/test/dockerutil/profile.go @@ -49,17 +49,16 @@ type Profile interface { // should have --profile set as an option in /etc/docker/daemon.json in // order for profiling to work with Pprof. type Pprof struct { - BasePath string // path to put profiles - BlockProfile bool - CPUProfile bool - GoRoutineProfile bool - HeapProfile bool - MutexProfile bool - Duration time.Duration // duration to run profiler e.g. '10s' or '1m'. - shouldRun bool - cmd *exec.Cmd - stdout io.ReadCloser - stderr io.ReadCloser + BasePath string // path to put profiles + BlockProfile bool + CPUProfile bool + HeapProfile bool + MutexProfile bool + Duration time.Duration // duration to run profiler e.g. '10s' or '1m'. + shouldRun bool + cmd *exec.Cmd + stdout io.ReadCloser + stderr io.ReadCloser } // MakePprofFromFlags makes a Pprof profile from flags. @@ -68,13 +67,12 @@ func MakePprofFromFlags(c *Container) *Pprof { return nil } return &Pprof{ - BasePath: filepath.Join(*pprofBaseDir, c.runtime, c.Name), - BlockProfile: *pprofBlock, - CPUProfile: *pprofCPU, - GoRoutineProfile: *pprofGo, - HeapProfile: *pprofHeap, - MutexProfile: *pprofMutex, - Duration: *duration, + BasePath: filepath.Join(*pprofBaseDir, c.runtime, c.Name), + BlockProfile: *pprofBlock, + CPUProfile: *pprofCPU, + HeapProfile: *pprofHeap, + MutexProfile: *pprofMutex, + Duration: *duration, } } @@ -138,9 +136,6 @@ func (p *Pprof) makeProfileArgs(c *Container) []string { if p.CPUProfile { ret = append(ret, fmt.Sprintf("--profile-cpu=%s", filepath.Join(p.BasePath, "cpu.pprof"))) } - if p.GoRoutineProfile { - ret = append(ret, fmt.Sprintf("--profile-goroutine=%s", filepath.Join(p.BasePath, "go.pprof"))) - } if p.HeapProfile { ret = append(ret, fmt.Sprintf("--profile-heap=%s", filepath.Join(p.BasePath, "heap.pprof"))) } diff --git a/pkg/test/dockerutil/profile_test.go b/pkg/test/dockerutil/profile_test.go index b7b4d7618..8c4ffe483 100644 --- a/pkg/test/dockerutil/profile_test.go +++ b/pkg/test/dockerutil/profile_test.go @@ -51,13 +51,12 @@ func TestPprof(t *testing.T) { { name: "All", pprof: Pprof{ - BasePath: basePath, - BlockProfile: true, - CPUProfile: true, - GoRoutineProfile: true, - HeapProfile: true, - MutexProfile: true, - Duration: 2 * time.Second, + BasePath: basePath, + BlockProfile: true, + CPUProfile: true, + HeapProfile: true, + MutexProfile: true, + Duration: 2 * time.Second, }, expectedFiles: []string{block, cpu, goprofle, heap, mutex}, }, diff --git a/runsc/boot/controller.go b/runsc/boot/controller.go index 3e5e4c22f..626a3816e 100644 --- a/runsc/boot/controller.go +++ b/runsc/boot/controller.go @@ -101,14 +101,13 @@ const ( // Profiling related commands (see pprof.go for more details). const ( - StartCPUProfile = "Profile.StartCPUProfile" - StopCPUProfile = "Profile.StopCPUProfile" - HeapProfile = "Profile.HeapProfile" - GoroutineProfile = "Profile.GoroutineProfile" - BlockProfile = "Profile.BlockProfile" - MutexProfile = "Profile.MutexProfile" - StartTrace = "Profile.StartTrace" - StopTrace = "Profile.StopTrace" + StartCPUProfile = "Profile.StartCPUProfile" + StopCPUProfile = "Profile.StopCPUProfile" + HeapProfile = "Profile.HeapProfile" + BlockProfile = "Profile.BlockProfile" + MutexProfile = "Profile.MutexProfile" + StartTrace = "Profile.StartTrace" + StopTrace = "Profile.StopTrace" ) // Logging related commands (see logging.go for more details). @@ -129,42 +128,52 @@ type controller struct { // manager holds the containerManager methods. manager *containerManager + + // pprop holds the profile instance if enabled. It may be nil. + pprof *control.Profile } // newController creates a new controller. The caller must call // controller.srv.StartServing() to start the controller. func newController(fd int, l *Loader) (*controller, error) { - srv, err := server.CreateFromFD(fd) + ctrl := &controller{} + var err error + ctrl.srv, err = server.CreateFromFD(fd) if err != nil { return nil, err } - manager := &containerManager{ + ctrl.manager = &containerManager{ startChan: make(chan struct{}), startResultChan: make(chan error), l: l, } - srv.Register(manager) + ctrl.srv.Register(ctrl.manager) if eps, ok := l.k.RootNetworkNamespace().Stack().(*netstack.Stack); ok { net := &Network{ Stack: eps.Stack, } - srv.Register(net) + ctrl.srv.Register(net) } - srv.Register(&debug{}) - srv.Register(&control.Logging{}) + ctrl.srv.Register(&debug{}) + ctrl.srv.Register(&control.Logging{}) + if l.root.conf.ProfileEnable { - srv.Register(&control.Profile{ - Kernel: l.k, - }) + ctrl.pprof = &control.Profile{Kernel: l.k} + ctrl.srv.Register(ctrl.pprof) } - return &controller{ - srv: srv, - manager: manager, - }, nil + return ctrl, nil +} + +func (c *controller) stop() { + if c.pprof != nil { + // These are noop if there is nothing being profiled. + _ = c.pprof.StopCPUProfile(nil, nil) + _ = c.pprof.StopTrace(nil, nil) + } } // containerManager manages sandbox containers. diff --git a/runsc/boot/loader.go b/runsc/boot/loader.go index 92f0b16e1..533b9c5e7 100644 --- a/runsc/boot/loader.go +++ b/runsc/boot/loader.go @@ -1008,6 +1008,9 @@ func (l *Loader) WaitExit() kernel.ExitStatus { // Wait for container. l.k.WaitExited() + // Cleanup + l.ctrl.stop() + return l.k.GlobalInit().ExitStatus() } diff --git a/runsc/cmd/debug.go b/runsc/cmd/debug.go index b5de2588b..742f8c344 100644 --- a/runsc/cmd/debug.go +++ b/runsc/cmd/debug.go @@ -32,20 +32,19 @@ import ( // Debug implements subcommands.Command for the "debug" command. type Debug struct { - pid int - stacks bool - signal int - profileHeap string - profileCPU string - profileGoroutine string - profileBlock string - profileMutex string - trace string - strace string - logLevel string - logPackets string - duration time.Duration - ps bool + pid int + stacks bool + signal int + profileHeap string + profileCPU string + profileBlock string + profileMutex string + trace string + strace string + logLevel string + logPackets string + duration time.Duration + ps bool } // Name implements subcommands.Command. @@ -69,7 +68,6 @@ func (d *Debug) SetFlags(f *flag.FlagSet) { f.BoolVar(&d.stacks, "stacks", false, "if true, dumps all sandbox stacks to the log") f.StringVar(&d.profileHeap, "profile-heap", "", "writes heap profile to the given file.") f.StringVar(&d.profileCPU, "profile-cpu", "", "writes CPU profile to the given file.") - f.StringVar(&d.profileGoroutine, "profile-goroutine", "", "writes goroutine profile to the given file.") f.StringVar(&d.profileBlock, "profile-block", "", "writes block profile to the given file.") f.StringVar(&d.profileMutex, "profile-mutex", "", "writes mutex profile to the given file.") f.DurationVar(&d.duration, "duration", time.Second, "amount of time to wait for CPU and trace profiles") @@ -153,18 +151,6 @@ func (d *Debug) Execute(_ context.Context, f *flag.FlagSet, args ...interface{}) } log.Infof("Heap profile written to %q", d.profileHeap) } - if d.profileGoroutine != "" { - f, err := os.Create(d.profileGoroutine) - if err != nil { - return Errorf(err.Error()) - } - defer f.Close() - - if err := c.Sandbox.GoroutineProfile(f); err != nil { - return Errorf(err.Error()) - } - log.Infof("Goroutine profile written to %q", d.profileGoroutine) - } if d.profileBlock != "" { f, err := os.Create(d.profileBlock) if err != nil { diff --git a/runsc/sandbox/sandbox.go b/runsc/sandbox/sandbox.go index 2afcc27af..36bb0c9c9 100644 --- a/runsc/sandbox/sandbox.go +++ b/runsc/sandbox/sandbox.go @@ -1012,26 +1012,6 @@ func (s *Sandbox) StopCPUProfile() error { return nil } -// GoroutineProfile writes a goroutine profile to the given file. -func (s *Sandbox) GoroutineProfile(f *os.File) error { - log.Debugf("Goroutine profile %q", s.ID) - conn, err := s.sandboxConnect() - if err != nil { - return err - } - defer conn.Close() - - opts := control.ProfileOpts{ - FilePayload: urpc.FilePayload{ - Files: []*os.File{f}, - }, - } - if err := conn.Call(boot.GoroutineProfile, &opts, nil); err != nil { - return fmt.Errorf("getting sandbox %q goroutine profile: %v", s.ID, err) - } - return nil -} - // BlockProfile writes a block profile to the given file. func (s *Sandbox) BlockProfile(f *os.File) error { log.Debugf("Block profile %q", s.ID) -- cgit v1.2.3 From 1403a88c67ccc87af383f5e54fd2f77c673eebe7 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Wed, 5 Aug 2020 16:14:40 -0700 Subject: Release extra memfd reference. PiperOrigin-RevId: 325122849 --- pkg/sentry/syscalls/linux/vfs2/memfd.go | 1 + 1 file changed, 1 insertion(+) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/vfs2/memfd.go b/pkg/sentry/syscalls/linux/vfs2/memfd.go index 519583e4e..c4c0f9e0a 100644 --- a/pkg/sentry/syscalls/linux/vfs2/memfd.go +++ b/pkg/sentry/syscalls/linux/vfs2/memfd.go @@ -51,6 +51,7 @@ func MemfdCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S if err != nil { return 0, nil, err } + defer file.DecRef(t) fd, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{ CloseOnExec: cloExec, -- cgit v1.2.3 From 7ed4b2b5a6928b3a4a88d0117a764dd4795be61a Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Wed, 5 Aug 2020 18:15:01 -0700 Subject: Correctly decrement link counts in tmpfs rename operations. When a directory is replaced by a rename operation, its link count should reach zero. We were missing the link from `dir/.` PiperOrigin-RevId: 325141730 --- pkg/sentry/fsimpl/tmpfs/filesystem.go | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index fb77f95cc..065812065 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -566,7 +566,9 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa if replaced != nil { newParentDir.removeChildLocked(replaced) if replaced.inode.isDir() { - newParentDir.inode.decLinksLocked(ctx) // from replaced's ".." + // Remove links for replaced/. and replaced/.. + replaced.inode.decLinksLocked(ctx) + newParentDir.inode.decLinksLocked(ctx) } replaced.inode.decLinksLocked(ctx) } -- cgit v1.2.3 From 35312a95c4c8626365b4ece5ffb0bcab44b4bede Mon Sep 17 00:00:00 2001 From: Nayana Bidari Date: Wed, 5 Aug 2020 20:45:02 -0700 Subject: Add loss recovery option for TCP. /proc/sys/net/ipv4/tcp_recovery is used to enable RACK loss recovery in TCP. PiperOrigin-RevId: 325157807 --- pkg/sentry/fs/proc/sys_net.go | 95 +++++++++++++++++++++++++++++++++++++ pkg/sentry/fsimpl/proc/tasks_sys.go | 49 ++++++++++++++++++- pkg/sentry/inet/inet.go | 17 +++++++ pkg/sentry/inet/test_stack.go | 12 +++++ pkg/sentry/socket/hostinet/stack.go | 11 +++++ pkg/sentry/socket/netstack/stack.go | 14 ++++++ pkg/tcpip/transport/tcp/protocol.go | 33 +++++++++++++ test/syscalls/linux/proc_net.cc | 38 +++++++++++++++ 8 files changed, 268 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fs/proc/sys_net.go b/pkg/sentry/fs/proc/sys_net.go index 702fdd392..8615b60f0 100644 --- a/pkg/sentry/fs/proc/sys_net.go +++ b/pkg/sentry/fs/proc/sys_net.go @@ -272,6 +272,96 @@ func (f *tcpSackFile) Write(ctx context.Context, _ *fs.File, src usermem.IOSeque return n, f.tcpSack.stack.SetTCPSACKEnabled(*f.tcpSack.enabled) } +// +stateify savable +type tcpRecovery struct { + fsutil.SimpleFileInode + + stack inet.Stack `state:"wait"` + recovery inet.TCPLossRecovery +} + +func newTCPRecoveryInode(ctx context.Context, msrc *fs.MountSource, s inet.Stack) *fs.Inode { + ts := &tcpRecovery{ + SimpleFileInode: *fsutil.NewSimpleFileInode(ctx, fs.RootOwner, fs.FilePermsFromMode(0644), linux.PROC_SUPER_MAGIC), + stack: s, + } + sattr := fs.StableAttr{ + DeviceID: device.ProcDevice.DeviceID(), + InodeID: device.ProcDevice.NextIno(), + BlockSize: usermem.PageSize, + Type: fs.SpecialFile, + } + return fs.NewInode(ctx, ts, msrc, sattr) +} + +// Truncate implements fs.InodeOperations.Truncate. +func (*tcpRecovery) Truncate(context.Context, *fs.Inode, int64) error { + return nil +} + +// GetFile implements fs.InodeOperations.GetFile. +func (r *tcpRecovery) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + flags.Pread = true + flags.Pwrite = true + return fs.NewFile(ctx, dirent, flags, &tcpRecoveryFile{ + tcpRecovery: r, + stack: r.stack, + }), nil +} + +// +stateify savable +type tcpRecoveryFile struct { + fsutil.FileGenericSeek `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoSplice `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + fsutil.FileUseInodeUnstableAttr `state:"nosave"` + waiter.AlwaysReady `state:"nosave"` + + tcpRecovery *tcpRecovery + + stack inet.Stack `state:"wait"` +} + +// Read implements fs.FileOperations.Read. +func (f *tcpRecoveryFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { + if offset != 0 { + return 0, io.EOF + } + + recovery, err := f.stack.TCPRecovery() + if err != nil { + return 0, err + } + f.tcpRecovery.recovery = recovery + s := fmt.Sprintf("%d\n", f.tcpRecovery.recovery) + n, err := dst.CopyOut(ctx, []byte(s)) + return int64(n), err +} + +// Write implements fs.FileOperations.Write. +func (f *tcpRecoveryFile) Write(ctx context.Context, _ *fs.File, src usermem.IOSequence, offset int64) (int64, error) { + if src.NumBytes() == 0 { + return 0, nil + } + src = src.TakeFirst(usermem.PageSize - 1) + + var v int32 + n, err := usermem.CopyInt32StringInVec(ctx, src.IO, src.Addrs, &v, src.Opts) + if err != nil { + return 0, err + } + f.tcpRecovery.recovery = inet.TCPLossRecovery(v) + if err := f.tcpRecovery.stack.SetTCPRecovery(f.tcpRecovery.recovery); err != nil { + return 0, err + } + return n, nil +} + func (p *proc) newSysNetCore(ctx context.Context, msrc *fs.MountSource, s inet.Stack) *fs.Inode { // The following files are simple stubs until they are implemented in // netstack, most of these files are configuration related. We use the @@ -351,6 +441,11 @@ func (p *proc) newSysNetIPv4Dir(ctx context.Context, msrc *fs.MountSource, s ine contents["tcp_wmem"] = newTCPMemInode(ctx, msrc, s, tcpWMem) } + // Add tcp_recovery. + if _, err := s.TCPRecovery(); err == nil { + contents["tcp_recovery"] = newTCPRecoveryInode(ctx, msrc, s) + } + d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) return newProcInode(ctx, d, msrc, fs.SpecialDirectory, nil) } diff --git a/pkg/sentry/fsimpl/proc/tasks_sys.go b/pkg/sentry/fsimpl/proc/tasks_sys.go index 6dac2afa4..b71778128 100644 --- a/pkg/sentry/fsimpl/proc/tasks_sys.go +++ b/pkg/sentry/fsimpl/proc/tasks_sys.go @@ -55,7 +55,8 @@ func (fs *filesystem) newSysNetDir(root *auth.Credentials, k *kernel.Kernel) *ke if stack := k.RootNetworkNamespace().Stack(); stack != nil { contents = map[string]*kernfs.Dentry{ "ipv4": kernfs.NewStaticDir(root, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), 0555, map[string]*kernfs.Dentry{ - "tcp_sack": fs.newDentry(root, fs.NextIno(), 0644, &tcpSackData{stack: stack}), + "tcp_recovery": fs.newDentry(root, fs.NextIno(), 0644, &tcpRecoveryData{stack: stack}), + "tcp_sack": fs.newDentry(root, fs.NextIno(), 0644, &tcpSackData{stack: stack}), // The following files are simple stubs until they are implemented in // netstack, most of these files are configuration related. We use the @@ -207,3 +208,49 @@ func (d *tcpSackData) Write(ctx context.Context, src usermem.IOSequence, offset *d.enabled = v != 0 return n, d.stack.SetTCPSACKEnabled(*d.enabled) } + +// tcpRecoveryData implements vfs.WritableDynamicBytesSource for +// /proc/sys/net/ipv4/tcp_recovery. +// +// +stateify savable +type tcpRecoveryData struct { + kernfs.DynamicBytesFile + + stack inet.Stack `state:"wait"` +} + +var _ vfs.WritableDynamicBytesSource = (*tcpRecoveryData)(nil) + +// Generate implements vfs.DynamicBytesSource. +func (d *tcpRecoveryData) Generate(ctx context.Context, buf *bytes.Buffer) error { + recovery, err := d.stack.TCPRecovery() + if err != nil { + return err + } + + buf.WriteString(fmt.Sprintf("%d\n", recovery)) + return nil +} + +func (d *tcpRecoveryData) Write(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) { + if offset != 0 { + // No need to handle partial writes thus far. + return 0, syserror.EINVAL + } + if src.NumBytes() == 0 { + return 0, nil + } + + // Limit the amount of memory allocated. + src = src.TakeFirst(usermem.PageSize - 1) + + var v int32 + n, err := usermem.CopyInt32StringInVec(ctx, src.IO, src.Addrs, &v, src.Opts) + if err != nil { + return 0, err + } + if err := d.stack.SetTCPRecovery(inet.TCPLossRecovery(v)); err != nil { + return 0, err + } + return n, nil +} diff --git a/pkg/sentry/inet/inet.go b/pkg/sentry/inet/inet.go index 2916a0644..c0b4831d1 100644 --- a/pkg/sentry/inet/inet.go +++ b/pkg/sentry/inet/inet.go @@ -56,6 +56,12 @@ type Stack interface { // settings. SetTCPSACKEnabled(enabled bool) error + // TCPRecovery returns the TCP loss detection algorithm. + TCPRecovery() (TCPLossRecovery, error) + + // SetTCPRecovery attempts to change TCP loss detection algorithm. + SetTCPRecovery(recovery TCPLossRecovery) error + // Statistics reports stack statistics. Statistics(stat interface{}, arg string) error @@ -189,3 +195,14 @@ type StatSNMPUDP [8]uint64 // StatSNMPUDPLite describes UdpLite line of /proc/net/snmp. type StatSNMPUDPLite [8]uint64 + +// TCPLossRecovery indicates TCP loss detection and recovery methods to use. +type TCPLossRecovery int32 + +// Loss recovery constants from include/net/tcp.h which are used to set +// /proc/sys/net/ipv4/tcp_recovery. +const ( + TCP_RACK_LOSS_DETECTION TCPLossRecovery = 1 << iota + TCP_RACK_STATIC_REO_WND + TCP_RACK_NO_DUPTHRESH +) diff --git a/pkg/sentry/inet/test_stack.go b/pkg/sentry/inet/test_stack.go index d8961fc94..9771f01fc 100644 --- a/pkg/sentry/inet/test_stack.go +++ b/pkg/sentry/inet/test_stack.go @@ -25,6 +25,7 @@ type TestStack struct { TCPRecvBufSize TCPBufferSize TCPSendBufSize TCPBufferSize TCPSACKFlag bool + Recovery TCPLossRecovery } // NewTestStack returns a TestStack with no network interfaces. The value of @@ -91,6 +92,17 @@ func (s *TestStack) SetTCPSACKEnabled(enabled bool) error { return nil } +// TCPRecovery implements Stack.TCPRecovery. +func (s *TestStack) TCPRecovery() (TCPLossRecovery, error) { + return s.Recovery, nil +} + +// SetTCPRecovery implements Stack.SetTCPRecovery. +func (s *TestStack) SetTCPRecovery(recovery TCPLossRecovery) error { + s.Recovery = recovery + return nil +} + // Statistics implements inet.Stack.Statistics. func (s *TestStack) Statistics(stat interface{}, arg string) error { return nil diff --git a/pkg/sentry/socket/hostinet/stack.go b/pkg/sentry/socket/hostinet/stack.go index a48082631..fda3dcb35 100644 --- a/pkg/sentry/socket/hostinet/stack.go +++ b/pkg/sentry/socket/hostinet/stack.go @@ -53,6 +53,7 @@ type Stack struct { interfaceAddrs map[int32][]inet.InterfaceAddr routes []inet.Route supportsIPv6 bool + tcpRecovery inet.TCPLossRecovery tcpRecvBufSize inet.TCPBufferSize tcpSendBufSize inet.TCPBufferSize tcpSACKEnabled bool @@ -350,6 +351,16 @@ func (s *Stack) SetTCPSACKEnabled(enabled bool) error { return syserror.EACCES } +// TCPRecovery implements inet.Stack.TCPRecovery. +func (s *Stack) TCPRecovery() (inet.TCPLossRecovery, error) { + return s.tcpRecovery, nil +} + +// SetTCPRecovery implements inet.Stack.SetTCPRecovery. +func (s *Stack) SetTCPRecovery(recovery inet.TCPLossRecovery) error { + return syserror.EACCES +} + // getLine reads one line from proc file, with specified prefix. // The last argument, withHeader, specifies if it contains line header. func getLine(f *os.File, prefix string, withHeader bool) string { diff --git a/pkg/sentry/socket/netstack/stack.go b/pkg/sentry/socket/netstack/stack.go index 67737ae87..f0fe18684 100644 --- a/pkg/sentry/socket/netstack/stack.go +++ b/pkg/sentry/socket/netstack/stack.go @@ -207,6 +207,20 @@ func (s *Stack) SetTCPSACKEnabled(enabled bool) error { return syserr.TranslateNetstackError(s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SACKEnabled(enabled))).ToError() } +// TCPRecovery implements inet.Stack.TCPRecovery. +func (s *Stack) TCPRecovery() (inet.TCPLossRecovery, error) { + var recovery tcp.Recovery + if err := s.Stack.TransportProtocolOption(tcp.ProtocolNumber, &recovery); err != nil { + return 0, syserr.TranslateNetstackError(err).ToError() + } + return inet.TCPLossRecovery(recovery), nil +} + +// SetTCPRecovery implements inet.Stack.SetTCPRecovery. +func (s *Stack) SetTCPRecovery(recovery inet.TCPLossRecovery) error { + return syserr.TranslateNetstackError(s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.Recovery(recovery))).ToError() +} + // Statistics implements inet.Stack.Statistics. func (s *Stack) Statistics(stat interface{}, arg string) error { switch stats := stat.(type) { diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go index b34e47bbd..d9abb8d94 100644 --- a/pkg/tcpip/transport/tcp/protocol.go +++ b/pkg/tcpip/transport/tcp/protocol.go @@ -80,6 +80,25 @@ const ( // enable/disable SACK support in TCP. See: https://tools.ietf.org/html/rfc2018. type SACKEnabled bool +// Recovery is used by stack.(*Stack).TransportProtocolOption to +// set loss detection algorithm in TCP. +type Recovery int32 + +const ( + // RACKLossDetection indicates RACK is used for loss detection and + // recovery. + RACKLossDetection Recovery = 1 << iota + + // RACKStaticReoWnd indicates the reordering window should not be + // adjusted when DSACK is received. + RACKStaticReoWnd + + // RACKNoDupTh indicates RACK should not consider the classic three + // duplicate acknowledgements rule to mark the segments as lost. This + // is used when reordering is not detected. + RACKNoDupTh +) + // DelayEnabled is used by stack.(Stack*).TransportProtocolOption to // enable/disable Nagle's algorithm in TCP. type DelayEnabled bool @@ -161,6 +180,7 @@ func (s *synRcvdCounter) Threshold() uint64 { type protocol struct { mu sync.RWMutex sackEnabled bool + recovery Recovery delayEnabled bool sendBufferSize SendBufferSizeOption recvBufferSize ReceiveBufferSizeOption @@ -280,6 +300,12 @@ func (p *protocol) SetOption(option interface{}) *tcpip.Error { p.mu.Unlock() return nil + case Recovery: + p.mu.Lock() + p.recovery = Recovery(v) + p.mu.Unlock() + return nil + case DelayEnabled: p.mu.Lock() p.delayEnabled = bool(v) @@ -394,6 +420,12 @@ func (p *protocol) Option(option interface{}) *tcpip.Error { p.mu.RUnlock() return nil + case *Recovery: + p.mu.RLock() + *v = Recovery(p.recovery) + p.mu.RUnlock() + return nil + case *DelayEnabled: p.mu.RLock() *v = DelayEnabled(p.delayEnabled) @@ -535,6 +567,7 @@ func NewProtocol() stack.TransportProtocol { minRTO: MinRTO, maxRTO: MaxRTO, maxRetries: MaxRetries, + recovery: RACKLossDetection, } p.dispatcher.init(runtime.GOMAXPROCS(0)) return &p diff --git a/test/syscalls/linux/proc_net.cc b/test/syscalls/linux/proc_net.cc index 3377b65cf..4fab097f4 100644 --- a/test/syscalls/linux/proc_net.cc +++ b/test/syscalls/linux/proc_net.cc @@ -477,6 +477,44 @@ TEST(ProcNetSnmp, CheckSnmp) { EXPECT_EQ(value_count, 1); } +TEST(ProcSysNetIpv4Recovery, Exists) { + EXPECT_THAT(open("/proc/sys/net/ipv4/tcp_recovery", O_RDONLY), + SyscallSucceeds()); +} + +TEST(ProcSysNetIpv4Recovery, CanReadAndWrite) { + // TODO(b/162988252): Enable save/restore for this test after the bug is + // fixed. + DisableSave ds; + + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability((CAP_DAC_OVERRIDE)))); + + auto const fd = ASSERT_NO_ERRNO_AND_VALUE( + Open("/proc/sys/net/ipv4/tcp_recovery", O_RDWR)); + + char buf[10] = {'\0'}; + char to_write = '2'; + + // Check initial value is set to 1. + EXPECT_THAT(PreadFd(fd.get(), &buf, sizeof(buf), 0), + SyscallSucceedsWithValue(sizeof(to_write) + 1)); + EXPECT_EQ(strcmp(buf, "1\n"), 0); + + // Set tcp_recovery to one of the allowed constants. + EXPECT_THAT(PwriteFd(fd.get(), &to_write, sizeof(to_write), 0), + SyscallSucceedsWithValue(sizeof(to_write))); + EXPECT_THAT(PreadFd(fd.get(), &buf, sizeof(buf), 0), + SyscallSucceedsWithValue(sizeof(to_write) + 1)); + EXPECT_EQ(strcmp(buf, "2\n"), 0); + + // Set tcp_recovery to any random value. + char kMessage[] = "100"; + EXPECT_THAT(PwriteFd(fd.get(), kMessage, strlen(kMessage), 0), + SyscallSucceedsWithValue(strlen(kMessage))); + EXPECT_THAT(PreadFd(fd.get(), buf, sizeof(kMessage), 0), + SyscallSucceedsWithValue(sizeof(kMessage))); + EXPECT_EQ(strcmp(buf, "100\n"), 0); +} } // namespace } // namespace testing } // namespace gvisor -- cgit v1.2.3 From 63447e5afae62dab64940054372166dab45a057c Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Thu, 6 Aug 2020 11:01:13 -0700 Subject: Only register /dev/net/tun if supported. PiperOrigin-RevId: 325266487 --- pkg/sentry/devices/tundev/tundev.go | 4 ++-- runsc/boot/vfs.go | 16 +++++++++++----- 2 files changed, 13 insertions(+), 7 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/devices/tundev/tundev.go b/pkg/sentry/devices/tundev/tundev.go index 852ec3c5c..a40625e19 100644 --- a/pkg/sentry/devices/tundev/tundev.go +++ b/pkg/sentry/devices/tundev/tundev.go @@ -160,8 +160,8 @@ func (fd *tunFD) EventUnregister(e *waiter.Entry) { fd.device.EventUnregister(e) } -// isNetTunSupported returns whether /dev/net/tun device is supported for s. -func isNetTunSupported(s inet.Stack) bool { +// IsNetTunSupported returns whether /dev/net/tun device is supported for s. +func IsNetTunSupported(s inet.Stack) bool { _, ok := s.(*netstack.Stack) return ok } diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go index 252ca07e3..e7d6035bb 100644 --- a/runsc/boot/vfs.go +++ b/runsc/boot/vfs.go @@ -37,6 +37,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fsimpl/proc" "gvisor.dev/gvisor/pkg/sentry/fsimpl/sys" "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" + "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" @@ -89,6 +90,12 @@ func registerFilesystems(k *kernel.Kernel) error { if err := ttydev.Register(vfsObj); err != nil { return fmt.Errorf("registering ttydev: %w", err) } + tunSupported := tundev.IsNetTunSupported(inet.StackFromContext(ctx)) + if tunSupported { + if err := tundev.Register(vfsObj); err != nil { + return fmt.Errorf("registering tundev: %v", err) + } + } if kernel.FUSEEnabled { if err := fuse.Register(vfsObj); err != nil { @@ -96,9 +103,6 @@ func registerFilesystems(k *kernel.Kernel) error { } } - if err := tundev.Register(vfsObj); err != nil { - return fmt.Errorf("registering tundev: %v", err) - } a, err := devtmpfs.NewAccessor(ctx, vfsObj, creds, devtmpfs.Name) if err != nil { return fmt.Errorf("creating devtmpfs accessor: %w", err) @@ -114,8 +118,10 @@ func registerFilesystems(k *kernel.Kernel) error { if err := ttydev.CreateDevtmpfsFiles(ctx, a); err != nil { return fmt.Errorf("creating ttydev devtmpfs files: %w", err) } - if err := tundev.CreateDevtmpfsFiles(ctx, a); err != nil { - return fmt.Errorf("creating tundev devtmpfs files: %v", err) + if tunSupported { + if err := tundev.CreateDevtmpfsFiles(ctx, a); err != nil { + return fmt.Errorf("creating tundev devtmpfs files: %v", err) + } } if kernel.FUSEEnabled { -- cgit v1.2.3 From f20e63e31b56784c596897e86f03441f9d05f567 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Thu, 6 Aug 2020 18:13:55 -0700 Subject: Add LinkAt support to gofer Updates #1198 PiperOrigin-RevId: 325350818 --- images/basic/linktest/Dockerfile | 7 +++ images/basic/linktest/link_test.c | 93 +++++++++++++++++++++++++++++++++++ pkg/sentry/fsimpl/gofer/filesystem.go | 26 +++++++++- pkg/sentry/fsimpl/gofer/gofer.go | 2 - scripts/docker_tests.sh | 2 +- test/e2e/integration_test.go | 18 +++++++ 6 files changed, 143 insertions(+), 5 deletions(-) create mode 100644 images/basic/linktest/Dockerfile create mode 100644 images/basic/linktest/link_test.c (limited to 'pkg/sentry') diff --git a/images/basic/linktest/Dockerfile b/images/basic/linktest/Dockerfile new file mode 100644 index 000000000..baebc9b76 --- /dev/null +++ b/images/basic/linktest/Dockerfile @@ -0,0 +1,7 @@ +FROM ubuntu:bionic + +WORKDIR /root +COPY . . + +RUN apt-get update && apt-get install -y gcc +RUN gcc -O2 -o link_test link_test.c diff --git a/images/basic/linktest/link_test.c b/images/basic/linktest/link_test.c new file mode 100644 index 000000000..45ab00abe --- /dev/null +++ b/images/basic/linktest/link_test.c @@ -0,0 +1,93 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include +#include +#include +#include +#include +#include + +// Basic test for linkat(2). Syscall tests requires CAP_DAC_READ_SEARCH and it +// cannot use tricks like userns as root. For this reason, run a basic link test +// to ensure some coverage. +int main(int argc, char** argv) { + const char kOldPath[] = "old.txt"; + int fd = open(kOldPath, O_RDWR | O_CREAT | O_TRUNC, 0600); + if (fd < 0) { + errx(1, "open(%s) failed", kOldPath); + } + const char kData[] = "some random content"; + if (write(fd, kData, sizeof(kData)) < 0) { + err(1, "write failed"); + } + close(fd); + + struct stat old_stat; + if (stat(kOldPath, &old_stat)) { + errx(1, "stat(%s) failed", kOldPath); + } + + const char kNewPath[] = "new.txt"; + if (link(kOldPath, kNewPath)) { + errx(1, "link(%s, %s) failed", kOldPath, kNewPath); + } + + struct stat new_stat; + if (stat(kNewPath, &new_stat)) { + errx(1, "stat(%s) failed", kNewPath); + } + + // Check that files are the same. + if (old_stat.st_dev != new_stat.st_dev) { + errx(1, "files st_dev is different, want: %lu, got: %lu", old_stat.st_dev, + new_stat.st_dev); + } + if (old_stat.st_ino != new_stat.st_ino) { + errx(1, "files st_ino is different, want: %lu, got: %lu", old_stat.st_ino, + new_stat.st_ino); + } + + // Check that link count is correct. + if (new_stat.st_nlink != old_stat.st_nlink + 1) { + errx(1, "wrong nlink, want: %lu, got: %lu", old_stat.st_nlink + 1, + new_stat.st_nlink); + } + + // Check taht contents are the same. + fd = open(kNewPath, O_RDONLY); + if (fd < 0) { + errx(1, "open(%s) failed", kNewPath); + } + char buf[sizeof(kData)] = {}; + if (read(fd, buf, sizeof(buf)) < 0) { + err(1, "read failed"); + } + close(fd); + + if (strcmp(buf, kData) != 0) { + errx(1, "file content mismatch: %s", buf); + } + + // Cleanup. + if (unlink(kNewPath)) { + errx(1, "unlink(%s) failed", kNewPath); + } + if (unlink(kOldPath)) { + errx(1, "unlink(%s) failed", kOldPath); + } + + // Success! + return 0; +} diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index e6af37d0d..582b744bb 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -15,6 +15,7 @@ package gofer import ( + "math" "sync" "sync/atomic" @@ -724,8 +725,29 @@ func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. if rp.Mount() != vd.Mount() { return syserror.EXDEV } - // 9P2000.L supports hard links, but we don't. - return syserror.EPERM + d := vd.Dentry().Impl().(*dentry) + if d.isDir() { + return syserror.EPERM + } + gid := auth.KGID(atomic.LoadUint32(&d.gid)) + uid := auth.KUID(atomic.LoadUint32(&d.uid)) + mode := linux.FileMode(atomic.LoadUint32(&d.mode)) + if err := vfs.MayLink(rp.Credentials(), mode, uid, gid); err != nil { + return err + } + if d.nlink == 0 { + return syserror.ENOENT + } + if d.nlink == math.MaxUint32 { + return syserror.EMLINK + } + if err := parent.file.link(ctx, d.file, childName); err != nil { + return err + } + + // Success! + atomic.AddUint32(&d.nlink, 1) + return nil }, nil) } diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 2e5575d8d..6ae796c6d 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -558,8 +558,6 @@ type dentry struct { // filesystem.renameMu. name string - // We don't support hard links, so each dentry maps 1:1 to an inode. - // file is the unopened p9.File that backs this dentry. file is immutable. // // If file.isNil(), this dentry represents a synthetic file, i.e. a file diff --git a/scripts/docker_tests.sh b/scripts/docker_tests.sh index 07e9f3109..be0b0a3ec 100755 --- a/scripts/docker_tests.sh +++ b/scripts/docker_tests.sh @@ -23,5 +23,5 @@ test_runsc //test/image:image_test //test/e2e:integration_test install_runsc_for_test docker --vfs2 IMAGE_FILTER="Hello|Httpd|Ruby|Stdio" -INTEGRATION_FILTER="LifeCycle|Pause|Connect|JobControl|Overlay|Exec|DirCreation/root" +INTEGRATION_FILTER="LifeCycle|Pause|Connect|JobControl|Overlay|Exec|DirCreation/root|Link" test_runsc //test/e2e:integration_test //test/image:image_test --test_filter="${IMAGE_FILTER}|${INTEGRATION_FILTER}" diff --git a/test/e2e/integration_test.go b/test/e2e/integration_test.go index 6fe6d304f..71ec4791e 100644 --- a/test/e2e/integration_test.go +++ b/test/e2e/integration_test.go @@ -467,6 +467,24 @@ func TestHostOverlayfsRewindDir(t *testing.T) { } } +// Basic test for linkat(2). Syscall tests requires CAP_DAC_READ_SEARCH and it +// cannot use tricks like userns as root. For this reason, run a basic link test +// to ensure some coverage. +func TestLink(t *testing.T) { + ctx := context.Background() + d := dockerutil.MakeContainer(ctx, t) + defer d.CleanUp(ctx) + + if got, err := d.Run(ctx, dockerutil.RunOpts{ + Image: "basic/linktest", + WorkDir: "/root", + }, "./link_test"); err != nil { + t.Fatalf("docker run failed: %v", err) + } else if got != "" { + t.Errorf("test failed:\n%s", got) + } +} + func TestMain(m *testing.M) { dockerutil.EnsureSupportedDockerVersion() flag.Parse() -- cgit v1.2.3 From 4fa1c304a133297bc6895729d74aa35a015e759e Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Fri, 7 Aug 2020 00:10:35 -0700 Subject: Try to update atime and mtime on VFS2 gofer files on dentry eviction. PiperOrigin-RevId: 325388385 --- pkg/sentry/fsimpl/gofer/gofer.go | 150 +++++++++++++++++++++++---------------- pkg/sentry/fsimpl/gofer/time.go | 3 + 2 files changed, 91 insertions(+), 62 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 6ae796c6d..59323086d 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -192,10 +192,14 @@ const ( // // - File timestamps are based on client clocks. This ensures that users of // the client observe timestamps that are coherent with their own clocks - // and consistent with Linux's semantics. However, since it is not always - // possible for clients to set arbitrary atimes and mtimes, and never - // possible for clients to set arbitrary ctimes, file timestamp changes are - // stored in the client only and never sent to the remote filesystem. + // and consistent with Linux's semantics (in particular, it is not always + // possible for clients to set arbitrary atimes and mtimes depending on the + // remote filesystem implementation, and never possible for clients to set + // arbitrary ctimes.) If a dentry containing a client-defined atime or + // mtime is evicted from cache, client timestamps will be sent to the + // remote filesystem on a best-effort basis to attempt to ensure that + // timestamps will be preserved when another dentry representing the same + // file is instantiated. InteropModeExclusive InteropMode = iota // InteropModeWritethrough is appropriate when there are read-only users of @@ -621,6 +625,12 @@ type dentry struct { // File size, protected by both metadataMu and dataMu (i.e. both must be // locked to mutate it; locking either is sufficient to access it). size uint64 + // If this dentry does not represent a synthetic file, deleted is 0, and + // atimeDirty/mtimeDirty are non-zero, atime/mtime may have diverged from the + // remote file's timestamps, which should be updated when this dentry is + // evicted. + atimeDirty uint32 + mtimeDirty uint32 // nlink counts the number of hard links to this dentry. It's updated and // accessed using atomic operations. It's not protected by metadataMu like the @@ -801,10 +811,12 @@ func (d *dentry) updateFromP9AttrsLocked(mask p9.AttrMask, attr *p9.Attr) { if attr.BlockSize != 0 { atomic.StoreUint32(&d.blockSize, uint32(attr.BlockSize)) } - if mask.ATime { + // Don't override newer client-defined timestamps with old server-defined + // ones. + if mask.ATime && atomic.LoadUint32(&d.atimeDirty) == 0 { atomic.StoreInt64(&d.atime, dentryTimestampFromP9(attr.ATimeSeconds, attr.ATimeNanoSeconds)) } - if mask.MTime { + if mask.MTime && atomic.LoadUint32(&d.mtimeDirty) == 0 { atomic.StoreInt64(&d.mtime, dentryTimestampFromP9(attr.MTimeSeconds, attr.MTimeNanoSeconds)) } if mask.CTime { @@ -901,51 +913,44 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, opts *vfs return err } defer mnt.EndWrite() - setLocalAtime := false - setLocalMtime := false + + if stat.Mask&linux.STATX_SIZE != 0 { + // Reject attempts to truncate files other than regular files, since + // filesystem implementations may return the wrong errno. + switch mode.FileType() { + case linux.S_IFREG: + // ok + case linux.S_IFDIR: + return syserror.EISDIR + default: + return syserror.EINVAL + } + } + + var now int64 if d.cachedMetadataAuthoritative() { - // Timestamp updates will be handled locally. - setLocalAtime = stat.Mask&linux.STATX_ATIME != 0 - setLocalMtime = stat.Mask&linux.STATX_MTIME != 0 - stat.Mask &^= linux.STATX_ATIME | linux.STATX_MTIME - - // Prepare for truncate. - if stat.Mask&linux.STATX_SIZE != 0 { - switch mode.FileType() { - case linux.ModeRegular: - if !setLocalMtime { - // Truncate updates mtime. - setLocalMtime = true - stat.Mtime.Nsec = linux.UTIME_NOW - } - case linux.ModeDirectory: - return syserror.EISDIR - default: - return syserror.EINVAL + // Truncate updates mtime. + if stat.Mask&(linux.STATX_SIZE|linux.STATX_MTIME) == linux.STATX_SIZE { + stat.Mask |= linux.STATX_MTIME + stat.Mtime = linux.StatxTimestamp{ + Nsec: linux.UTIME_NOW, } } + + // Use client clocks for timestamps. + now = d.fs.clock.Now().Nanoseconds() + if stat.Mask&linux.STATX_ATIME != 0 && stat.Atime.Nsec == linux.UTIME_NOW { + stat.Atime = statxTimestampFromDentry(now) + } + if stat.Mask&linux.STATX_MTIME != 0 && stat.Mtime.Nsec == linux.UTIME_NOW { + stat.Mtime = statxTimestampFromDentry(now) + } } + d.metadataMu.Lock() defer d.metadataMu.Unlock() - if stat.Mask&linux.STATX_SIZE != 0 { - // The size needs to be changed even when - // !d.cachedMetadataAuthoritative() because d.mappings has to be - // updated. - d.updateFileSizeLocked(stat.Size) - } if !d.isSynthetic() { if stat.Mask != 0 { - if stat.Mask&linux.STATX_SIZE != 0 { - // Check whether to allow a truncate request to be made. - switch d.mode & linux.S_IFMT { - case linux.S_IFREG: - // Allow. - case linux.S_IFDIR: - return syserror.EISDIR - default: - return syserror.EINVAL - } - } if err := d.file.setAttr(ctx, p9.SetAttrMask{ Permissions: stat.Mask&linux.STATX_MODE != 0, UID: stat.Mask&linux.STATX_UID != 0, @@ -967,6 +972,11 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, opts *vfs }); err != nil { return err } + if stat.Mask&linux.STATX_SIZE != 0 { + // Privatized copy-on-write mappings of truncated pages need to + // be invalidated even if InteropModeShared is in effect. + d.updateFileSizeLocked(stat.Size) + } } if d.fs.opts.interop == InteropModeShared { // There's no point to updating d's metadata in this case since @@ -976,7 +986,6 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, opts *vfs return nil } } - now := d.fs.clock.Now().Nanoseconds() if stat.Mask&linux.STATX_MODE != 0 { atomic.StoreUint32(&d.mode, d.fileType()|uint32(stat.Mode)) } @@ -986,23 +995,18 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, opts *vfs if stat.Mask&linux.STATX_GID != 0 { atomic.StoreUint32(&d.gid, stat.GID) } - if setLocalAtime { - if stat.Atime.Nsec == linux.UTIME_NOW { - atomic.StoreInt64(&d.atime, now) - } else { - atomic.StoreInt64(&d.atime, dentryTimestampFromStatx(stat.Atime)) - } - // Restore mask bits that we cleared earlier. - stat.Mask |= linux.STATX_ATIME - } - if setLocalMtime { - if stat.Mtime.Nsec == linux.UTIME_NOW { - atomic.StoreInt64(&d.mtime, now) - } else { - atomic.StoreInt64(&d.mtime, dentryTimestampFromStatx(stat.Mtime)) - } - // Restore mask bits that we cleared earlier. - stat.Mask |= linux.STATX_MTIME + // Note that stat.Atime.Nsec and stat.Mtime.Nsec can't be UTIME_NOW because + // if d.cachedMetadataAuthoritative() then we converted stat.Atime and + // stat.Mtime to client-local timestamps above, and if + // !d.cachedMetadataAuthoritative() then we returned after calling + // d.file.setAttr(). For the same reason, now must have been initialized. + if stat.Mask&linux.STATX_ATIME != 0 { + atomic.StoreInt64(&d.atime, dentryTimestampFromStatx(stat.Atime)) + atomic.StoreUint32(&d.atimeDirty, 0) + } + if stat.Mask&linux.STATX_MTIME != 0 { + atomic.StoreInt64(&d.mtime, dentryTimestampFromStatx(stat.Mtime)) + atomic.StoreUint32(&d.mtimeDirty, 0) } atomic.StoreInt64(&d.ctime, now) return nil @@ -1248,7 +1252,7 @@ func (d *dentry) destroyLocked(ctx context.Context) { // Write dirty pages back to the remote filesystem. if d.handleWritable { if err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, mf, d.handle.writeFromBlocksAt); err != nil { - log.Warningf("gofer.dentry.DecRef: failed to write dirty data back: %v", err) + log.Warningf("gofer.dentry.destroyLocked: failed to write dirty data back: %v", err) } } // Discard cached data. @@ -1261,6 +1265,28 @@ func (d *dentry) destroyLocked(ctx context.Context) { d.handleMu.Unlock() if !d.file.isNil() { + if !d.isDeleted() { + // Write dirty timestamps back to the remote filesystem. + atimeDirty := atomic.LoadUint32(&d.atimeDirty) != 0 + mtimeDirty := atomic.LoadUint32(&d.mtimeDirty) != 0 + if atimeDirty || mtimeDirty { + atime := atomic.LoadInt64(&d.atime) + mtime := atomic.LoadInt64(&d.mtime) + if err := d.file.setAttr(ctx, p9.SetAttrMask{ + ATime: atimeDirty, + ATimeNotSystemTime: atimeDirty, + MTime: mtimeDirty, + MTimeNotSystemTime: mtimeDirty, + }, p9.SetAttr{ + ATimeSeconds: uint64(atime / 1e9), + ATimeNanoSeconds: uint64(atime % 1e9), + MTimeSeconds: uint64(mtime / 1e9), + MTimeNanoSeconds: uint64(mtime % 1e9), + }); err != nil { + log.Warningf("gofer.dentry.destroyLocked: failed to write dirty timestamps back: %v", err) + } + } + } d.file.close(ctx) d.file = p9file{} // Remove d from the set of syncable dentries. diff --git a/pkg/sentry/fsimpl/gofer/time.go b/pkg/sentry/fsimpl/gofer/time.go index 0eef4e16e..2cb8191b9 100644 --- a/pkg/sentry/fsimpl/gofer/time.go +++ b/pkg/sentry/fsimpl/gofer/time.go @@ -47,6 +47,7 @@ func (d *dentry) touchAtime(mnt *vfs.Mount) { now := d.fs.clock.Now().Nanoseconds() d.metadataMu.Lock() atomic.StoreInt64(&d.atime, now) + atomic.StoreUint32(&d.atimeDirty, 1) d.metadataMu.Unlock() mnt.EndWrite() } @@ -67,6 +68,7 @@ func (d *dentry) touchCMtime() { d.metadataMu.Lock() atomic.StoreInt64(&d.mtime, now) atomic.StoreInt64(&d.ctime, now) + atomic.StoreUint32(&d.mtimeDirty, 1) d.metadataMu.Unlock() } @@ -76,4 +78,5 @@ func (d *dentry) touchCMtimeLocked() { now := d.fs.clock.Now().Nanoseconds() atomic.StoreInt64(&d.mtime, now) atomic.StoreInt64(&d.ctime, now) + atomic.StoreUint32(&d.mtimeDirty, 1) } -- cgit v1.2.3 From 93cb66825bf098f8a19b3d7f34b33272ceed8cb3 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Fri, 7 Aug 2020 12:54:14 -0700 Subject: Support separate read/write handles in fsimpl/gofer.dentry. PiperOrigin-RevId: 325490674 --- pkg/sentry/fsimpl/gofer/BUILD | 1 + pkg/sentry/fsimpl/gofer/directory.go | 12 +- pkg/sentry/fsimpl/gofer/filesystem.go | 18 +- pkg/sentry/fsimpl/gofer/gofer.go | 308 +++++++++++++++++++++----------- pkg/sentry/fsimpl/gofer/gofer_test.go | 4 +- pkg/sentry/fsimpl/gofer/handle.go | 19 +- pkg/sentry/fsimpl/gofer/regular_file.go | 118 ++++++------ pkg/sentry/fsimpl/gofer/special_file.go | 11 +- pkg/sentry/vfs/file_description.go | 5 + test/syscalls/BUILD | 1 + 10 files changed, 312 insertions(+), 185 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/BUILD b/pkg/sentry/fsimpl/gofer/BUILD index 4a800dcf9..16787116f 100644 --- a/pkg/sentry/fsimpl/gofer/BUILD +++ b/pkg/sentry/fsimpl/gofer/BUILD @@ -85,5 +85,6 @@ go_test( deps = [ "//pkg/p9", "//pkg/sentry/contexttest", + "//pkg/sentry/pgalloc", ], ) diff --git a/pkg/sentry/fsimpl/gofer/directory.go b/pkg/sentry/fsimpl/gofer/directory.go index 1679066ba..2a8011eb4 100644 --- a/pkg/sentry/fsimpl/gofer/directory.go +++ b/pkg/sentry/fsimpl/gofer/directory.go @@ -90,10 +90,8 @@ func (d *dentry) createSyntheticChildLocked(opts *createSyntheticOpts) { uid: uint32(opts.kuid), gid: uint32(opts.kgid), blockSize: usermem.PageSize, // arbitrary - handle: handle{ - fd: -1, - }, - nlink: uint32(2), + hostFD: -1, + nlink: uint32(2), } switch opts.mode.FileType() { case linux.S_IFDIR: @@ -205,14 +203,14 @@ func (d *dentry) getDirents(ctx context.Context) ([]vfs.Dirent, error) { off := uint64(0) const count = 64 * 1024 // for consistency with the vfs1 client d.handleMu.RLock() - if !d.handleReadable { + if d.readFile.isNil() { // This should not be possible because a readable handle should // have been opened when the calling directoryFD was opened. d.handleMu.RUnlock() panic("gofer.dentry.getDirents called without a readable handle") } for { - p9ds, err := d.handle.file.readdir(ctx, off, count) + p9ds, err := d.readFile.readdir(ctx, off, count) if err != nil { d.handleMu.RUnlock() return nil, err @@ -304,5 +302,5 @@ func (fd *directoryFD) Seek(ctx context.Context, offset int64, whence int32) (in // Sync implements vfs.FileDescriptionImpl.Sync. func (fd *directoryFD) Sync(ctx context.Context) error { - return fd.dentry().handle.sync(ctx) + return fd.dentry().syncRemoteFile(ctx) } diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 582b744bb..eaef2594d 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -55,7 +55,7 @@ func (fs *filesystem) Sync(ctx context.Context) error { // Sync regular files. for _, d := range ds { - err := d.syncSharedHandle(ctx) + err := d.syncCachedFile(ctx) d.DecRef(ctx) if err != nil && retErr == nil { retErr = err @@ -1107,12 +1107,18 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving useRegularFileFD := child.fileType() == linux.S_IFREG && !d.fs.opts.regularFilesUseSpecialFileFD if useRegularFileFD { child.handleMu.Lock() - child.handle.file = openFile - if fdobj != nil { - child.handle.fd = int32(fdobj.Release()) + if vfs.MayReadFileWithOpenFlags(opts.Flags) { + child.readFile = openFile + if fdobj != nil { + child.hostFD = int32(fdobj.Release()) + } + } else if fdobj != nil { + // Can't use fdobj if it's not readable. + fdobj.Close() + } + if vfs.MayWriteFileWithOpenFlags(opts.Flags) { + child.writeFile = openFile } - child.handleReadable = vfs.MayReadFileWithOpenFlags(opts.Flags) - child.handleWritable = vfs.MayWriteFileWithOpenFlags(opts.Flags) child.handleMu.Unlock() } // Insert the dentry into the tree. diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 59323086d..f1d3bf911 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -506,9 +506,9 @@ func (fs *filesystem) Release(ctx context.Context) { for d := range fs.syncableDentries { d.handleMu.Lock() d.dataMu.Lock() - if d.handleWritable { + if h := d.writeHandleLocked(); h.isOpen() { // Write dirty cached data to the remote file. - if err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, fs.mfp.MemoryFile(), d.handle.writeFromBlocksAt); err != nil { + if err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, fs.mfp.MemoryFile(), h.writeFromBlocksAt); err != nil { log.Warningf("gofer.filesystem.Release: failed to flush dentry: %v", err) } // TODO(jamieliu): Do we need to flushf/fsync d? @@ -518,9 +518,9 @@ func (fs *filesystem) Release(ctx context.Context) { d.dirty.RemoveAll() d.dataMu.Unlock() // Close the host fd if one exists. - if d.handle.fd >= 0 { - syscall.Close(int(d.handle.fd)) - d.handle.fd = -1 + if d.hostFD >= 0 { + syscall.Close(int(d.hostFD)) + d.hostFD = -1 } d.handleMu.Unlock() } @@ -622,7 +622,12 @@ type dentry struct { mtime int64 ctime int64 btime int64 - // File size, protected by both metadataMu and dataMu (i.e. both must be + // File size, which differs from other metadata in two ways: + // + // - We make a best-effort attempt to keep it up to date even if + // !dentry.cachedMetadataAuthoritative() for the sake of O_APPEND writes. + // + // - size is protected by both metadataMu and dataMu (i.e. both must be // locked to mutate it; locking either is sufficient to access it). size uint64 // If this dentry does not represent a synthetic file, deleted is 0, and @@ -643,30 +648,28 @@ type dentry struct { // the file into memmap.MappingSpaces. mappings is protected by mapsMu. mappings memmap.MappingSet - // If this dentry represents a regular file or directory: - // - // - handle is the I/O handle used by all regularFileFDs/directoryFDs - // representing this dentry. - // - // - handleReadable is true if handle is readable. - // - // - handleWritable is true if handle is writable. - // - // Invariants: + // - If this dentry represents a regular file or directory, readFile is the + // p9.File used for reads by all regularFileFDs/directoryFDs representing + // this dentry. // - // - If handleReadable == handleWritable == false, then handle.file == nil - // (i.e. there is no open handle). Conversely, if handleReadable || - // handleWritable == true, then handle.file != nil (i.e. there is an open - // handle). + // - If this dentry represents a regular file, writeFile is the p9.File + // used for writes by all regularFileFDs representing this dentry. // - // - handleReadable and handleWritable cannot transition from true to false - // (i.e. handles may not be downgraded). + // - If this dentry represents a regular file, hostFD is the host FD used + // for memory mappings and I/O (when applicable) in preference to readFile + // and writeFile. hostFD is always readable; if !writeFile.isNil(), it must + // also be writable. If hostFD is -1, no such host FD is available. // // These fields are protected by handleMu. - handleMu sync.RWMutex - handle handle - handleReadable bool - handleWritable bool + // + // readFile and writeFile may or may not represent the same p9.File. Once + // either p9.File transitions from closed (isNil() == true) to open + // (isNil() == false), it may be mutated with handleMu locked, but cannot + // be closed until the dentry is destroyed. + handleMu sync.RWMutex + readFile p9file + writeFile p9file + hostFD int32 dataMu sync.RWMutex @@ -680,7 +683,7 @@ type dentry struct { // tracks dirty segments in cache. dirty is protected by dataMu. dirty fsutil.DirtySet - // pf implements platform.File for mappings of handle.fd. + // pf implements platform.File for mappings of hostFD. pf dentryPlatformFile // If this dentry represents a symbolic link, InteropModeShared is not in @@ -742,9 +745,7 @@ func (fs *filesystem) newDentry(ctx context.Context, file p9file, qid p9.QID, ma uid: uint32(fs.opts.dfltuid), gid: uint32(fs.opts.dfltgid), blockSize: usermem.PageSize, - handle: handle{ - fd: -1, - }, + hostFD: -1, } d.pf.dentry = d if mask.UID { @@ -835,9 +836,13 @@ func (d *dentry) updateFromP9AttrsLocked(mask p9.AttrMask, attr *p9.Attr) { // Preconditions: !d.isSynthetic() func (d *dentry) updateFromGetattr(ctx context.Context) error { - // Use d.handle.file, which represents a 9P fid that has been opened, in - // preference to d.file, which represents a 9P fid that has not. This may - // be significantly more efficient in some implementations. + // Use d.readFile or d.writeFile, which represent 9P fids that have been + // opened, in preference to d.file, which represents a 9P fid that has not. + // This may be significantly more efficient in some implementations. Prefer + // d.writeFile over d.readFile since some filesystem implementations may + // update a writable handle's metadata after writes to that handle, without + // making metadata updates immediately visible to read-only handles + // representing the same file. var ( file p9file handleMuRLocked bool @@ -847,8 +852,11 @@ func (d *dentry) updateFromGetattr(ctx context.Context) error { d.metadataMu.Lock() defer d.metadataMu.Unlock() d.handleMu.RLock() - if !d.handle.file.isNil() { - file = d.handle.file + if !d.writeFile.isNil() { + file = d.writeFile + handleMuRLocked = true + } else if !d.readFile.isNil() { + file = d.readFile handleMuRLocked = true } else { file = d.file @@ -973,8 +981,9 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, opts *vfs return err } if stat.Mask&linux.STATX_SIZE != 0 { - // Privatized copy-on-write mappings of truncated pages need to - // be invalidated even if InteropModeShared is in effect. + // d.size should be kept up to date, and privatized + // copy-on-write mappings of truncated pages need to be + // invalidated, even if InteropModeShared is in effect. d.updateFileSizeLocked(stat.Size) } } @@ -1245,22 +1254,31 @@ func (d *dentry) destroyLocked(ctx context.Context) { panic("dentry.destroyLocked() called with references on the dentry") } + mf := d.fs.mfp.MemoryFile() d.handleMu.Lock() - if !d.handle.file.isNil() { - mf := d.fs.mfp.MemoryFile() - d.dataMu.Lock() + d.dataMu.Lock() + if h := d.writeHandleLocked(); h.isOpen() { // Write dirty pages back to the remote filesystem. - if d.handleWritable { - if err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, mf, d.handle.writeFromBlocksAt); err != nil { - log.Warningf("gofer.dentry.destroyLocked: failed to write dirty data back: %v", err) - } + if err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, mf, h.writeFromBlocksAt); err != nil { + log.Warningf("gofer.dentry.destroyLocked: failed to write dirty data back: %v", err) } - // Discard cached data. - d.cache.DropAll(mf) - d.dirty.RemoveAll() - d.dataMu.Unlock() - // Clunk open fids and close open host FDs. - d.handle.close(ctx) + } + // Discard cached data. + d.cache.DropAll(mf) + d.dirty.RemoveAll() + d.dataMu.Unlock() + // Clunk open fids and close open host FDs. + if !d.readFile.isNil() { + d.readFile.close(ctx) + } + if !d.writeFile.isNil() && d.readFile != d.writeFile { + d.writeFile.close(ctx) + } + d.readFile = p9file{} + d.writeFile = p9file{} + if d.hostFD >= 0 { + syscall.Close(int(d.hostFD)) + d.hostFD = -1 } d.handleMu.Unlock() @@ -1393,80 +1411,120 @@ func (d *dentry) ensureSharedHandle(ctx context.Context, read, write, trunc bool // O_TRUNC). if !trunc { d.handleMu.RLock() - if (!read || d.handleReadable) && (!write || d.handleWritable) { - // The current handle is sufficient. + if (!read || !d.readFile.isNil()) && (!write || !d.writeFile.isNil()) { + // Current handles are sufficient. d.handleMu.RUnlock() return nil } d.handleMu.RUnlock() } - haveOldFD := false + fdToClose := int32(-1) + invalidateTranslations := false d.handleMu.Lock() - if (read && !d.handleReadable) || (write && !d.handleWritable) || trunc { - // Get a new handle. - wantReadable := d.handleReadable || read - wantWritable := d.handleWritable || write - h, err := openHandle(ctx, d.file, wantReadable, wantWritable, trunc) + if (read && d.readFile.isNil()) || (write && d.writeFile.isNil()) || trunc { + // Get a new handle. If this file has been opened for both reading and + // writing, try to get a single handle that is usable for both: + // + // - Writable memory mappings of a host FD require that the host FD is + // opened for both reading and writing. + // + // - NOTE(b/141991141): Some filesystems may not ensure coherence + // between multiple handles for the same file. + openReadable := !d.readFile.isNil() || read + openWritable := !d.writeFile.isNil() || write + h, err := openHandle(ctx, d.file, openReadable, openWritable, trunc) + if err == syserror.EACCES && (openReadable != read || openWritable != write) { + // It may not be possible to use a single handle for both + // reading and writing, since permissions on the file may have + // changed to e.g. disallow reading after previously being + // opened for reading. In this case, we have no choice but to + // use separate handles for reading and writing. + ctx.Debugf("gofer.dentry.ensureSharedHandle: bifurcating read/write handles for dentry %p", d) + openReadable = read + openWritable = write + h, err = openHandle(ctx, d.file, openReadable, openWritable, trunc) + } if err != nil { d.handleMu.Unlock() return err } - if !d.handle.file.isNil() { - // Check that old and new handles are compatible: If the old handle - // includes a host file descriptor but the new one does not, or - // vice versa, old and new memory mappings may be incoherent. - haveOldFD = d.handle.fd >= 0 - haveNewFD := h.fd >= 0 - if haveOldFD != haveNewFD { - d.handleMu.Unlock() - ctx.Warningf("gofer.dentry.ensureSharedHandle: can't change host FD availability from %v to %v across dentry handle upgrade", haveOldFD, haveNewFD) - h.close(ctx) - return syserror.EIO - } - if haveOldFD { - // We may have raced with callers of d.pf.FD() that are now - // using the old file descriptor, preventing us from safely - // closing it. We could handle this by invalidating existing - // memmap.Translations, but this is expensive. Instead, use - // dup3 to make the old file descriptor refer to the new file - // description, then close the new file descriptor (which is no - // longer needed). Racing callers may use the old or new file - // description, but this doesn't matter since they refer to the - // same file (unless d.fs.opts.overlayfsStaleRead is true, - // which we handle separately). - if err := syscall.Dup3(int(h.fd), int(d.handle.fd), syscall.O_CLOEXEC); err != nil { + + if d.hostFD < 0 && openReadable && h.fd >= 0 { + // We have no existing FD; use the new FD for at least reading. + d.hostFD = h.fd + } else if d.hostFD >= 0 && d.writeFile.isNil() && openWritable { + // We have an existing read-only FD, but the file has just been + // opened for writing, so we need to start supporting writable memory + // mappings. This may race with callers of d.pf.FD() using the existing + // FD, so in most cases we need to delay closing the old FD until after + // invalidating memmap.Translations that might have observed it. + if !openReadable || h.fd < 0 { + // We don't have a read/write FD, so we have no FD that can be + // used to create writable memory mappings. Switch to using the + // internal page cache. + invalidateTranslations = true + fdToClose = d.hostFD + d.hostFD = -1 + } else if d.fs.opts.overlayfsStaleRead { + // We do have a read/write FD, but it may not be coherent with + // the existing read-only FD, so we must switch to mappings of + // the new FD in both the application and sentry. + if err := d.pf.hostFileMapper.RegenerateMappings(int(h.fd)); err != nil { d.handleMu.Unlock() - ctx.Warningf("gofer.dentry.ensureSharedHandle: failed to dup fd %d to fd %d: %v", h.fd, d.handle.fd, err) + ctx.Warningf("gofer.dentry.ensureSharedHandle: failed to replace sentry mappings of old FD with mappings of new FD: %v", err) h.close(ctx) return err } - syscall.Close(int(h.fd)) - h.fd = d.handle.fd - if d.fs.opts.overlayfsStaleRead { - // Replace sentry mappings of the old FD with mappings of - // the new FD, since the two are not necessarily coherent. - if err := d.pf.hostFileMapper.RegenerateMappings(int(h.fd)); err != nil { - d.handleMu.Unlock() - ctx.Warningf("gofer.dentry.ensureSharedHandle: failed to replace sentry mappings of old FD with mappings of new FD: %v", err) - h.close(ctx) - return err - } + invalidateTranslations = true + fdToClose = d.hostFD + d.hostFD = h.fd + } else { + // We do have a read/write FD. To avoid invalidating existing + // memmap.Translations (which is expensive), use dup3 to make + // the old file descriptor refer to the new file description, + // then close the new file descriptor (which is no longer + // needed). Racing callers of d.pf.FD() may use the old or new + // file description, but this doesn't matter since they refer + // to the same file, and any racing mappings must be read-only. + if err := syscall.Dup3(int(h.fd), int(d.hostFD), syscall.O_CLOEXEC); err != nil { + oldHostFD := d.hostFD + d.handleMu.Unlock() + ctx.Warningf("gofer.dentry.ensureSharedHandle: failed to dup fd %d to fd %d: %v", h.fd, oldHostFD, err) + h.close(ctx) + return err } - // Clunk the old fid before making the new handle visible (by - // unlocking d.handleMu). - d.handle.file.close(ctx) + fdToClose = h.fd } + } else { + // h.fd is not useful. + fdToClose = h.fd + } + + // Switch to new fids. + var oldReadFile p9file + if openReadable { + oldReadFile = d.readFile + d.readFile = h.file + } + var oldWriteFile p9file + if openWritable { + oldWriteFile = d.writeFile + d.writeFile = h.file + } + // NOTE(b/141991141): Clunk old fids before making new fids visible (by + // unlocking d.handleMu). + if !oldReadFile.isNil() { + oldReadFile.close(ctx) + } + if !oldWriteFile.isNil() && oldReadFile != oldWriteFile { + oldWriteFile.close(ctx) } - // Switch to the new handle. - d.handle = h - d.handleReadable = wantReadable - d.handleWritable = wantWritable } d.handleMu.Unlock() - if d.fs.opts.overlayfsStaleRead && haveOldFD { - // Invalidate application mappings that may be using the old FD; they + if invalidateTranslations { + // Invalidate application mappings that may be using an old FD; they // will be replaced with mappings using the new FD after future calls // to d.Translate(). This requires holding d.mapsMu, which precedes // d.handleMu in the lock order. @@ -1474,7 +1532,51 @@ func (d *dentry) ensureSharedHandle(ctx context.Context, read, write, trunc bool d.mappings.InvalidateAll(memmap.InvalidateOpts{}) d.mapsMu.Unlock() } + if fdToClose >= 0 { + syscall.Close(int(fdToClose)) + } + + return nil +} +// Preconditions: d.handleMu must be locked. +func (d *dentry) readHandleLocked() handle { + return handle{ + file: d.readFile, + fd: d.hostFD, + } +} + +// Preconditions: d.handleMu must be locked. +func (d *dentry) writeHandleLocked() handle { + return handle{ + file: d.writeFile, + fd: d.hostFD, + } +} + +func (d *dentry) syncRemoteFile(ctx context.Context) error { + d.handleMu.RLock() + defer d.handleMu.RUnlock() + return d.syncRemoteFileLocked(ctx) +} + +// Preconditions: d.handleMu must be locked. +func (d *dentry) syncRemoteFileLocked(ctx context.Context) error { + // If we have a host FD, fsyncing it is likely to be faster than an fsync + // RPC. + if d.hostFD >= 0 { + ctx.UninterruptibleSleepStart(false) + err := syscall.Fsync(int(d.hostFD)) + ctx.UninterruptibleSleepFinish(false) + return err + } + if !d.writeFile.isNil() { + return d.writeFile.fsync(ctx) + } + if !d.readFile.isNil() { + return d.readFile.fsync(ctx) + } return nil } diff --git a/pkg/sentry/fsimpl/gofer/gofer_test.go b/pkg/sentry/fsimpl/gofer/gofer_test.go index 56d80bcf8..36cca3625 100644 --- a/pkg/sentry/fsimpl/gofer/gofer_test.go +++ b/pkg/sentry/fsimpl/gofer/gofer_test.go @@ -20,10 +20,13 @@ import ( "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/sentry/contexttest" + "gvisor.dev/gvisor/pkg/sentry/pgalloc" ) func TestDestroyIdempotent(t *testing.T) { + ctx := contexttest.Context(t) fs := filesystem{ + mfp: pgalloc.MemoryFileProviderFromContext(ctx), syncableDentries: make(map[*dentry]struct{}), opts: filesystemOptions{ // Test relies on no dentry being held in the cache. @@ -31,7 +34,6 @@ func TestDestroyIdempotent(t *testing.T) { }, } - ctx := contexttest.Context(t) attr := &p9.Attr{ Mode: p9.ModeRegular, } diff --git a/pkg/sentry/fsimpl/gofer/handle.go b/pkg/sentry/fsimpl/gofer/handle.go index 8792ca4f2..104157512 100644 --- a/pkg/sentry/fsimpl/gofer/handle.go +++ b/pkg/sentry/fsimpl/gofer/handle.go @@ -63,6 +63,10 @@ func openHandle(ctx context.Context, file p9file, read, write, trunc bool) (hand }, nil } +func (h *handle) isOpen() bool { + return !h.file.isNil() +} + func (h *handle) close(ctx context.Context) { h.file.close(ctx) h.file = p9file{} @@ -124,18 +128,3 @@ func (h *handle) writeFromBlocksAt(ctx context.Context, srcs safemem.BlockSeq, o } return cp, cperr } - -func (h *handle) sync(ctx context.Context) error { - // Handle most common case first. - if h.fd >= 0 { - ctx.UninterruptibleSleepStart(false) - err := syscall.Fsync(int(h.fd)) - ctx.UninterruptibleSleepFinish(false) - return err - } - if h.file.isNil() { - // File hasn't been touched, there is nothing to sync. - return nil - } - return h.file.fsync(ctx) -} diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go index db6bed4f6..7e1cbf065 100644 --- a/pkg/sentry/fsimpl/gofer/regular_file.go +++ b/pkg/sentry/fsimpl/gofer/regular_file.go @@ -64,34 +64,34 @@ func (fd *regularFileFD) OnClose(ctx context.Context) error { } d.handleMu.RLock() defer d.handleMu.RUnlock() - return d.handle.file.flush(ctx) + if d.writeFile.isNil() { + return nil + } + return d.writeFile.flush(ctx) } // Allocate implements vfs.FileDescriptionImpl.Allocate. func (fd *regularFileFD) Allocate(ctx context.Context, mode, offset, length uint64) error { - d := fd.dentry() d.metadataMu.Lock() defer d.metadataMu.Unlock() - size := offset + length - // Allocating a smaller size is a noop. - if size <= d.size { + size := offset + length + if d.cachedMetadataAuthoritative() && size <= d.size { return nil } - d.handleMu.Lock() - defer d.handleMu.Unlock() - - err := d.handle.file.allocate(ctx, p9.ToAllocateMode(mode), offset, length) + d.handleMu.RLock() + err := d.writeFile.allocate(ctx, p9.ToAllocateMode(mode), offset, length) + d.handleMu.RUnlock() if err != nil { return err } d.dataMu.Lock() atomic.StoreUint64(&d.size, size) d.dataMu.Unlock() - if !d.cachedMetadataAuthoritative() { + if d.cachedMetadataAuthoritative() { d.touchCMtimeLocked() } return nil @@ -113,7 +113,7 @@ func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offs // Check for reading at EOF before calling into MM (but not under // InteropModeShared, which makes d.size unreliable). d := fd.dentry() - if d.fs.opts.interop != InteropModeShared && uint64(offset) >= atomic.LoadUint64(&d.size) { + if d.cachedMetadataAuthoritative() && uint64(offset) >= atomic.LoadUint64(&d.size) { return 0, io.EOF } @@ -217,16 +217,23 @@ func (fd *regularFileFD) pwrite(ctx context.Context, src usermem.IOSequence, off n, err := src.CopyInTo(ctx, rw) if err != nil { - return n, offset, err + return n, offset + n, err } if n > 0 && fd.vfsfd.StatusFlags()&(linux.O_DSYNC|linux.O_SYNC) != 0 { - // Write dirty cached pages touched by the write back to the remote file. + // Note that if any of the following fail, then we can't guarantee that + // any data was actually written with the semantics of O_DSYNC or + // O_SYNC, so we return zero bytes written. Compare Linux's + // mm/filemap.c:generic_file_write_iter() => + // include/linux/fs.h:generic_write_sync(). + // + // Write dirty cached pages touched by the write back to the remote + // file. if err := d.writeback(ctx, offset, src.NumBytes()); err != nil { - return n, offset, err + return 0, offset, err } // Request the remote filesystem to sync the remote file. - if err := d.handle.sync(ctx); err != nil { - return n, offset, err + if err := d.syncRemoteFile(ctx); err != nil { + return 0, offset, err } } return n, offset + n, nil @@ -317,10 +324,11 @@ func (rw *dentryReadWriter) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) // coherence with memory-mapped I/O), or if InteropModeShared is in effect // (which prevents us from caching file contents and makes dentry.size // unreliable), or if the file was opened O_DIRECT, read directly from - // dentry.handle without locking dentry.dataMu. + // dentry.readHandleLocked() without locking dentry.dataMu. rw.d.handleMu.RLock() - if (rw.d.handle.fd >= 0 && !rw.d.fs.opts.forcePageCache) || rw.d.fs.opts.interop == InteropModeShared || rw.direct { - n, err := rw.d.handle.readToBlocksAt(rw.ctx, dsts, rw.off) + h := rw.d.readHandleLocked() + if (rw.d.hostFD >= 0 && !rw.d.fs.opts.forcePageCache) || rw.d.fs.opts.interop == InteropModeShared || rw.direct { + n, err := h.readToBlocksAt(rw.ctx, dsts, rw.off) rw.d.handleMu.RUnlock() rw.off += n return n, err @@ -388,7 +396,7 @@ func (rw *dentryReadWriter) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) End: gapEnd, } optMR := gap.Range() - err := rw.d.cache.Fill(rw.ctx, reqMR, maxFillRange(reqMR, optMR), mf, usage.PageCache, rw.d.handle.readToBlocksAt) + err := rw.d.cache.Fill(rw.ctx, reqMR, maxFillRange(reqMR, optMR), mf, usage.PageCache, h.readToBlocksAt) mf.MarkEvictable(rw.d, pgalloc.EvictableRange{optMR.Start, optMR.End}) seg, gap = rw.d.cache.Find(rw.off) if !seg.Ok() { @@ -403,7 +411,7 @@ func (rw *dentryReadWriter) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) } else { // Read directly from the file. gapDsts := dsts.TakeFirst64(gapMR.Length()) - n, err := rw.d.handle.readToBlocksAt(rw.ctx, gapDsts, gapMR.Start) + n, err := h.readToBlocksAt(rw.ctx, gapDsts, gapMR.Start) done += n rw.off += n dsts = dsts.DropFirst64(n) @@ -435,11 +443,12 @@ func (rw *dentryReadWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, erro // If we have a mmappable host FD (which must be used here to ensure // coherence with memory-mapped I/O), or if InteropModeShared is in effect // (which prevents us from caching file contents), or if the file was - // opened with O_DIRECT, write directly to dentry.handle without locking - // dentry.dataMu. + // opened with O_DIRECT, write directly to dentry.writeHandleLocked() + // without locking dentry.dataMu. rw.d.handleMu.RLock() - if (rw.d.handle.fd >= 0 && !rw.d.fs.opts.forcePageCache) || rw.d.fs.opts.interop == InteropModeShared || rw.direct { - n, err := rw.d.handle.writeFromBlocksAt(rw.ctx, srcs, rw.off) + h := rw.d.writeHandleLocked() + if (rw.d.hostFD >= 0 && !rw.d.fs.opts.forcePageCache) || rw.d.fs.opts.interop == InteropModeShared || rw.direct { + n, err := h.writeFromBlocksAt(rw.ctx, srcs, rw.off) rw.off += n rw.d.dataMu.Lock() if rw.off > rw.d.size { @@ -501,7 +510,7 @@ func (rw *dentryReadWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, erro // for detecting or avoiding this. gapMR := gap.Range().Intersect(mr) gapSrcs := srcs.TakeFirst64(gapMR.Length()) - n, err := rw.d.handle.writeFromBlocksAt(rw.ctx, gapSrcs, gapMR.Start) + n, err := h.writeFromBlocksAt(rw.ctx, gapSrcs, gapMR.Start) done += n rw.off += n srcs = srcs.DropFirst64(n) @@ -527,7 +536,7 @@ exitLoop: if err := fsutil.SyncDirty(rw.ctx, memmap.MappableRange{ Start: start, End: rw.off, - }, &rw.d.cache, &rw.d.dirty, rw.d.size, mf, rw.d.handle.writeFromBlocksAt); err != nil { + }, &rw.d.cache, &rw.d.dirty, rw.d.size, mf, h.writeFromBlocksAt); err != nil { // We have no idea how many bytes were actually flushed. rw.off = start done = 0 @@ -545,6 +554,7 @@ func (d *dentry) writeback(ctx context.Context, offset, size int64) error { } d.handleMu.RLock() defer d.handleMu.RUnlock() + h := d.writeHandleLocked() d.dataMu.Lock() defer d.dataMu.Unlock() // Compute the range of valid bytes (overflow-checked). @@ -558,7 +568,7 @@ func (d *dentry) writeback(ctx context.Context, offset, size int64) error { return fsutil.SyncDirty(ctx, memmap.MappableRange{ Start: uint64(offset), End: uint64(end), - }, &d.cache, &d.dirty, d.size, d.fs.mfp.MemoryFile(), d.handle.writeFromBlocksAt) + }, &d.cache, &d.dirty, d.size, d.fs.mfp.MemoryFile(), h.writeFromBlocksAt) } // Seek implements vfs.FileDescriptionImpl.Seek. @@ -615,24 +625,23 @@ func regularFileSeekLocked(ctx context.Context, d *dentry, fdOffset, offset int6 // Sync implements vfs.FileDescriptionImpl.Sync. func (fd *regularFileFD) Sync(ctx context.Context) error { - return fd.dentry().syncSharedHandle(ctx) + return fd.dentry().syncCachedFile(ctx) } -func (d *dentry) syncSharedHandle(ctx context.Context) error { +func (d *dentry) syncCachedFile(ctx context.Context) error { d.handleMu.RLock() defer d.handleMu.RUnlock() - if d.handleWritable { + if h := d.writeHandleLocked(); h.isOpen() { d.dataMu.Lock() // Write dirty cached data to the remote file. - err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, d.fs.mfp.MemoryFile(), d.handle.writeFromBlocksAt) + err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, d.fs.mfp.MemoryFile(), h.writeFromBlocksAt) d.dataMu.Unlock() if err != nil { return err } } - // Sync the remote file. - return d.handle.sync(ctx) + return d.syncRemoteFileLocked(ctx) } // ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. @@ -656,7 +665,7 @@ func (fd *regularFileFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpt return syserror.ENODEV } d.handleMu.RLock() - haveFD := d.handle.fd >= 0 + haveFD := d.hostFD >= 0 d.handleMu.RUnlock() if !haveFD { return syserror.ENODEV @@ -677,7 +686,7 @@ func (d *dentry) mayCachePages() bool { return true } d.handleMu.RLock() - haveFD := d.handle.fd >= 0 + haveFD := d.hostFD >= 0 d.handleMu.RUnlock() return haveFD } @@ -735,7 +744,7 @@ func (d *dentry) CopyMapping(ctx context.Context, ms memmap.MappingSpace, srcAR, // Translate implements memmap.Mappable.Translate. func (d *dentry) Translate(ctx context.Context, required, optional memmap.MappableRange, at usermem.AccessType) ([]memmap.Translation, error) { d.handleMu.RLock() - if d.handle.fd >= 0 && !d.fs.opts.forcePageCache { + if d.hostFD >= 0 && !d.fs.opts.forcePageCache { d.handleMu.RUnlock() mr := optional if d.fs.opts.limitHostFDTranslation { @@ -771,7 +780,8 @@ func (d *dentry) Translate(ctx context.Context, required, optional memmap.Mappab } mf := d.fs.mfp.MemoryFile() - cerr := d.cache.Fill(ctx, required, maxFillRange(required, optional), mf, usage.PageCache, d.handle.readToBlocksAt) + h := d.readHandleLocked() + cerr := d.cache.Fill(ctx, required, maxFillRange(required, optional), mf, usage.PageCache, h.readToBlocksAt) var ts []memmap.Translation var translatedEnd uint64 @@ -840,9 +850,12 @@ func (d *dentry) InvalidateUnsavable(ctx context.Context) error { // Write the cache's contents back to the remote file so that if we have a // host fd after restore, the remote file's contents are coherent. mf := d.fs.mfp.MemoryFile() + d.handleMu.RLock() + defer d.handleMu.RUnlock() + h := d.writeHandleLocked() d.dataMu.Lock() defer d.dataMu.Unlock() - if err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, mf, d.handle.writeFromBlocksAt); err != nil { + if err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, mf, h.writeFromBlocksAt); err != nil { return err } @@ -857,20 +870,23 @@ func (d *dentry) InvalidateUnsavable(ctx context.Context) error { // Evict implements pgalloc.EvictableMemoryUser.Evict. func (d *dentry) Evict(ctx context.Context, er pgalloc.EvictableRange) { + mr := memmap.MappableRange{er.Start, er.End} + mf := d.fs.mfp.MemoryFile() d.mapsMu.Lock() defer d.mapsMu.Unlock() + d.handleMu.RLock() + defer d.handleMu.RUnlock() + h := d.writeHandleLocked() d.dataMu.Lock() defer d.dataMu.Unlock() - mr := memmap.MappableRange{er.Start, er.End} - mf := d.fs.mfp.MemoryFile() // Only allow pages that are no longer memory-mapped to be evicted. for mgap := d.mappings.LowerBoundGap(mr.Start); mgap.Ok() && mgap.Start() < mr.End; mgap = mgap.NextGap() { mgapMR := mgap.Range().Intersect(mr) if mgapMR.Length() == 0 { continue } - if err := fsutil.SyncDirty(ctx, mgapMR, &d.cache, &d.dirty, d.size, mf, d.handle.writeFromBlocksAt); err != nil { + if err := fsutil.SyncDirty(ctx, mgapMR, &d.cache, &d.dirty, d.size, mf, h.writeFromBlocksAt); err != nil { log.Warningf("Failed to writeback cached data %v: %v", mgapMR, err) } d.cache.Drop(mgapMR, mf) @@ -882,8 +898,8 @@ func (d *dentry) Evict(ctx context.Context, er pgalloc.EvictableRange) { // cannot implement both vfs.DentryImpl.IncRef and memmap.File.IncRef. // // dentryPlatformFile is only used when a host FD representing the remote file -// is available (i.e. dentry.handle.fd >= 0), and that FD is used for -// application memory mappings (i.e. !filesystem.opts.forcePageCache). +// is available (i.e. dentry.hostFD >= 0), and that FD is used for application +// memory mappings (i.e. !filesystem.opts.forcePageCache). type dentryPlatformFile struct { *dentry @@ -891,8 +907,8 @@ type dentryPlatformFile struct { // by dentry.dataMu. fdRefs fsutil.FrameRefSet - // If this dentry represents a regular file, and handle.fd >= 0, - // hostFileMapper caches mappings of handle.fd. + // If this dentry represents a regular file, and dentry.hostFD >= 0, + // hostFileMapper caches mappings of dentry.hostFD. hostFileMapper fsutil.HostFileMapper // hostFileMapperInitOnce is used to lazily initialize hostFileMapper. @@ -916,15 +932,13 @@ func (d *dentryPlatformFile) DecRef(fr memmap.FileRange) { // MapInternal implements memmap.File.MapInternal. func (d *dentryPlatformFile) MapInternal(fr memmap.FileRange, at usermem.AccessType) (safemem.BlockSeq, error) { d.handleMu.RLock() - bs, err := d.hostFileMapper.MapInternal(fr, int(d.handle.fd), at.Write) - d.handleMu.RUnlock() - return bs, err + defer d.handleMu.RUnlock() + return d.hostFileMapper.MapInternal(fr, int(d.hostFD), at.Write) } // FD implements memmap.File.FD. func (d *dentryPlatformFile) FD() int { d.handleMu.RLock() - fd := d.handle.fd - d.handleMu.RUnlock() - return int(fd) + defer d.handleMu.RUnlock() + return int(d.hostFD) } diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go index fc269ef2b..a6368fdd0 100644 --- a/pkg/sentry/fsimpl/gofer/special_file.go +++ b/pkg/sentry/fsimpl/gofer/special_file.go @@ -17,6 +17,7 @@ package gofer import ( "sync" "sync/atomic" + "syscall" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" @@ -279,5 +280,13 @@ func (fd *specialFileFD) Seek(ctx context.Context, offset int64, whence int32) ( // Sync implements vfs.FileDescriptionImpl.Sync. func (fd *specialFileFD) Sync(ctx context.Context) error { - return fd.dentry().syncSharedHandle(ctx) + // If we have a host FD, fsyncing it is likely to be faster than an fsync + // RPC. + if fd.handle.fd >= 0 { + ctx.UninterruptibleSleepStart(false) + err := syscall.Fsync(int(fd.handle.fd)) + ctx.UninterruptibleSleepFinish(false) + return err + } + return fd.handle.file.fsync(ctx) } diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 576ab3920..d3c1197e3 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -356,6 +356,8 @@ type FileDescriptionImpl interface { // Allocate grows the file to offset + length bytes. // Only mode == 0 is supported currently. + // + // Preconditions: The FileDescription was opened for writing. Allocate(ctx context.Context, mode, offset, length uint64) error // waiter.Waitable methods may be used to poll for I/O events. @@ -565,6 +567,9 @@ func (fd *FileDescription) StatFS(ctx context.Context) (linux.Statfs, error) { // Allocate grows file represented by FileDescription to offset + length bytes. func (fd *FileDescription) Allocate(ctx context.Context, mode, offset, length uint64) error { + if !fd.IsWritable() { + return syserror.EBADF + } return fd.impl.Allocate(ctx, mode, offset, length) } diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index c19b30b4a..a31612b41 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -1023,6 +1023,7 @@ syscall_test( syscall_test( add_overlay = True, test = "//test/syscalls/linux:truncate_test", + vfs2 = "True", ) syscall_test( -- cgit v1.2.3 From 0b7cd1e751f70cf6b335754d2b173d9c12b674c7 Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Fri, 7 Aug 2020 16:33:37 -0700 Subject: Mark dropped pages unevictable in fsimpl/gofer.dentry.destroyLocked. PiperOrigin-RevId: 325531657 --- pkg/sentry/fsimpl/gofer/gofer.go | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index f1d3bf911..4ac8dd81d 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -1264,8 +1264,11 @@ func (d *dentry) destroyLocked(ctx context.Context) { } } // Discard cached data. - d.cache.DropAll(mf) - d.dirty.RemoveAll() + if !d.cache.IsEmpty() { + mf.MarkAllUnevictable(d) + d.cache.DropAll(mf) + d.dirty.RemoveAll() + } d.dataMu.Unlock() // Clunk open fids and close open host FDs. if !d.readFile.isNil() { -- cgit v1.2.3 From 343661770aa96efe3b539a82addff13df235413f Mon Sep 17 00:00:00 2001 From: Jamie Liu Date: Fri, 7 Aug 2020 18:32:03 -0700 Subject: Don't hold gofer.filesystem.renameMu during dentry destruction. PiperOrigin-RevId: 325546629 --- pkg/sentry/fsimpl/gofer/gofer.go | 20 +++++++++++++++----- pkg/sentry/fsimpl/gofer/gofer_test.go | 2 ++ 2 files changed, 17 insertions(+), 5 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 4ac8dd81d..63e589859 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -1158,7 +1158,8 @@ func (d *dentry) OnZeroWatches(ctx context.Context) { // operation. One of the calls may destroy the dentry, so subsequent calls will // do nothing. // -// Preconditions: d.fs.renameMu must be locked for writing. +// Preconditions: d.fs.renameMu must be locked for writing; it may be +// temporarily unlocked. func (d *dentry) checkCachingLocked(ctx context.Context) { // Dentries with a non-zero reference count must be retained. (The only way // to obtain a reference on a dentry with zero references is via path @@ -1238,11 +1239,13 @@ func (d *dentry) checkCachingLocked(ctx context.Context) { } } -// destroyLocked destroys the dentry. It may flushes dirty pages from cache, -// close p9 file and remove reference on parent dentry. +// destroyLocked destroys the dentry. // -// Preconditions: d.fs.renameMu must be locked for writing. d.refs == 0. d is -// not a child dentry. +// Preconditions: +// * d.fs.renameMu must be locked for writing; it may be temporarily unlocked. +// * d.refs == 0. +// * d.parent.children[d.name] != d, i.e. d is not reachable by path traversal +// from its former parent dentry. func (d *dentry) destroyLocked(ctx context.Context) { switch atomic.LoadInt64(&d.refs) { case 0: @@ -1254,6 +1257,10 @@ func (d *dentry) destroyLocked(ctx context.Context) { panic("dentry.destroyLocked() called with references on the dentry") } + // Allow the following to proceed without renameMu locked to improve + // scalability. + d.fs.renameMu.Unlock() + mf := d.fs.mfp.MemoryFile() d.handleMu.Lock() d.dataMu.Lock() @@ -1315,6 +1322,9 @@ func (d *dentry) destroyLocked(ctx context.Context) { delete(d.fs.syncableDentries, d) d.fs.syncMu.Unlock() } + + d.fs.renameMu.Lock() + // Drop the reference held by d on its parent without recursively locking // d.fs.renameMu. if d.parent != nil { diff --git a/pkg/sentry/fsimpl/gofer/gofer_test.go b/pkg/sentry/fsimpl/gofer/gofer_test.go index 36cca3625..bfe75dfe4 100644 --- a/pkg/sentry/fsimpl/gofer/gofer_test.go +++ b/pkg/sentry/fsimpl/gofer/gofer_test.go @@ -52,6 +52,8 @@ func TestDestroyIdempotent(t *testing.T) { } parent.cacheNewChildLocked(child, "child") + fs.renameMu.Lock() + defer fs.renameMu.Unlock() child.checkCachingLocked(ctx) if got := atomic.LoadInt64(&child.refs); got != -1 { t.Fatalf("child.refs=%d, want: -1", got) -- cgit v1.2.3 From 3be26a271cd0fc9618fbcb34e1a2a84c4f234c86 Mon Sep 17 00:00:00 2001 From: Ayush Ranjan Date: Fri, 7 Aug 2020 20:06:39 -0700 Subject: [vfs2] Fix tmpfs mounting. Earlier we were using NLink to decide if /tmp is empty or not. However, NLink at best tells us about the number of subdirectories (via the ".." entries). NLink = n + 2 for n subdirectories. But it does not tell us if the directory is empty. There still might be non-directory files. We could also not rely on NLink because host overlayfs always returned 1. VFS1 uses Readdir to decide if the directory is empty. Used a similar approach. We now use IterDirents to decide if the "/tmp" directory is empty. Fixes #3369 PiperOrigin-RevId: 325554234 --- pkg/sentry/fsimpl/overlay/directory.go | 10 ++++++---- runsc/boot/vfs.go | 27 +++++++++++++++++++-------- 2 files changed, 25 insertions(+), 12 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/overlay/directory.go b/pkg/sentry/fsimpl/overlay/directory.go index fccb94105..6a79f7ffe 100644 --- a/pkg/sentry/fsimpl/overlay/directory.go +++ b/pkg/sentry/fsimpl/overlay/directory.go @@ -51,7 +51,7 @@ func (d *dentry) collectWhiteoutsForRmdirLocked(ctx context.Context) (map[string // Reuse slice allocated for maybeWhiteouts from a previous layer to // reduce allocations. maybeWhiteouts = maybeWhiteouts[:0] - if err := layerFD.IterDirents(ctx, vfs.IterDirentsCallbackFunc(func(dirent vfs.Dirent) error { + err = layerFD.IterDirents(ctx, vfs.IterDirentsCallbackFunc(func(dirent vfs.Dirent) error { if dirent.Name == "." || dirent.Name == ".." { return nil } @@ -68,7 +68,8 @@ func (d *dentry) collectWhiteoutsForRmdirLocked(ctx context.Context) (map[string } // Non-whiteout file in the directory prevents rmdir. return syserror.ENOTEMPTY - })); err != nil { + })) + if err != nil { readdirErr = err return false } @@ -182,7 +183,7 @@ func (d *dentry) getDirents(ctx context.Context) ([]vfs.Dirent, error) { // Reuse slice allocated for maybeWhiteouts from a previous layer to // reduce allocations. maybeWhiteouts = maybeWhiteouts[:0] - if err := layerFD.IterDirents(ctx, vfs.IterDirentsCallbackFunc(func(dirent vfs.Dirent) error { + err = layerFD.IterDirents(ctx, vfs.IterDirentsCallbackFunc(func(dirent vfs.Dirent) error { if dirent.Name == "." || dirent.Name == ".." { return nil } @@ -201,7 +202,8 @@ func (d *dentry) getDirents(ctx context.Context) ([]vfs.Dirent, error) { dirent.NextOff = int64(len(dirents) + 1) dirents = append(dirents, dirent) return nil - })); err != nil { + })) + if err != nil { readdirErr = err return false } diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go index e7d6035bb..08dce8b6c 100644 --- a/runsc/boot/vfs.go +++ b/runsc/boot/vfs.go @@ -400,21 +400,28 @@ func (c *containerMounter) mountTmpVFS2(ctx context.Context, conf *Config, creds Path: fspath.Parse("/tmp"), } // TODO(gvisor.dev/issue/2782): Use O_PATH when available. - statx, err := c.k.VFS().StatAt(ctx, creds, &pop, &vfs.StatOptions{}) + fd, err := c.k.VFS().OpenAt(ctx, creds, &pop, &vfs.OpenOptions{Flags: linux.O_RDONLY | linux.O_DIRECTORY}) switch err { case nil: - // Found '/tmp' in filesystem, check if it's empty. - if linux.FileMode(statx.Mode).FileType() != linux.ModeDirectory { - // Not a dir?! Leave it be. + defer fd.DecRef(ctx) + + err := fd.IterDirents(ctx, vfs.IterDirentsCallbackFunc(func(dirent vfs.Dirent) error { + if dirent.Name != "." && dirent.Name != ".." { + return syserror.ENOTEMPTY + } return nil - } - if statx.Nlink > 2 { + })) + switch err { + case nil: + log.Infof(`Mounting internal tmpfs on top of empty "/tmp"`) + case syserror.ENOTEMPTY: // If more than "." and ".." is found, skip internal tmpfs to prevent // hiding existing files. log.Infof(`Skipping internal tmpfs mount for "/tmp" because it's not empty`) return nil + default: + return err } - log.Infof(`Mounting internal tmpfs on top of empty "/tmp"`) fallthrough case syserror.ENOENT: @@ -429,8 +436,12 @@ func (c *containerMounter) mountTmpVFS2(ctx context.Context, conf *Config, creds } return c.mountSubmountVFS2(ctx, conf, mns, creds, &mountAndFD{Mount: tmpMount}) + case syserror.ENOTDIR: + // Not a dir?! Let it be. + return nil + default: - return fmt.Errorf(`stating "/tmp" inside container: %w`, err) + return fmt.Errorf(`opening "/tmp" inside container: %w`, err) } } -- cgit v1.2.3 From 13a8ae81b2361cd32f8e73d14ca5b9bca9569b1a Mon Sep 17 00:00:00 2001 From: Andrei Vagin Date: Fri, 7 Aug 2020 22:47:55 -0700 Subject: Add context.FullStateChanged() It indicates that the Sentry has changed the state of the thread and next calls of PullFullState() has to do nothing. PiperOrigin-RevId: 325567415 --- pkg/sentry/kernel/ptrace.go | 2 +- pkg/sentry/kernel/task_exec.go | 1 + pkg/sentry/kernel/task_signals.go | 4 ++-- pkg/sentry/platform/kvm/context.go | 4 ++-- pkg/sentry/platform/platform.go | 14 ++++++++++---- pkg/sentry/platform/ptrace/ptrace.go | 4 ++-- 6 files changed, 18 insertions(+), 11 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/kernel/ptrace.go b/pkg/sentry/kernel/ptrace.go index 6c03d9041..619b0cb7c 100644 --- a/pkg/sentry/kernel/ptrace.go +++ b/pkg/sentry/kernel/ptrace.go @@ -1063,7 +1063,7 @@ func (t *Task) Ptrace(req int64, pid ThreadID, addr, data usermem.Addr) error { if err != nil { return err } - t.p.FloatingPointStateChanged() + t.p.FullStateChanged() ar.End -= usermem.Addr(n) return t.CopyOutIovecs(data, usermem.AddrRangeSeqOf(ar)) diff --git a/pkg/sentry/kernel/task_exec.go b/pkg/sentry/kernel/task_exec.go index 47c28b8ff..5e4fb3e3a 100644 --- a/pkg/sentry/kernel/task_exec.go +++ b/pkg/sentry/kernel/task_exec.go @@ -226,6 +226,7 @@ func (r *runSyscallAfterExecStop) execute(t *Task) taskRunState { t.tc = *r.tc t.mu.Unlock() t.unstopVforkParent() + t.p.FullStateChanged() // NOTE(b/30316266): All locks must be dropped prior to calling Activate. t.MemoryManager().Activate(t) diff --git a/pkg/sentry/kernel/task_signals.go b/pkg/sentry/kernel/task_signals.go index 2180fd27d..cff2a8365 100644 --- a/pkg/sentry/kernel/task_signals.go +++ b/pkg/sentry/kernel/task_signals.go @@ -280,7 +280,7 @@ func (t *Task) deliverSignalToHandler(info *arch.SignalInfo, act arch.SignalAct) if err := t.Arch().SignalSetup(st, &act, info, &alt, mask); err != nil { return err } - t.p.FloatingPointStateChanged() + t.p.FullStateChanged() t.haveSavedSignalMask = false // Add our signal mask. @@ -312,7 +312,7 @@ func (t *Task) SignalReturn(rt bool) (*SyscallControl, error) { // Restore our signal mask. SIGKILL and SIGSTOP should not be blocked. t.SetSignalMask(sigset &^ UnblockableSignals) - t.p.FloatingPointStateChanged() + t.p.FullStateChanged() return ctrlResume, nil } diff --git a/pkg/sentry/platform/kvm/context.go b/pkg/sentry/platform/kvm/context.go index eb92721fb..6e6b76416 100644 --- a/pkg/sentry/platform/kvm/context.go +++ b/pkg/sentry/platform/kvm/context.go @@ -91,8 +91,8 @@ func (c *context) Interrupt() { // Release implements platform.Context.Release(). func (c *context) Release() {} -// FloatingPointStateChanged implements platform.Context.FloatingPointStateChanged. -func (c *context) FloatingPointStateChanged() {} +// FullStateChanged implements platform.Context.FullStateChanged. +func (c *context) FullStateChanged() {} // PullFullState implements platform.Context.PullFullState. func (c *context) PullFullState(as platform.AddressSpace, ac arch.Context) {} diff --git a/pkg/sentry/platform/platform.go b/pkg/sentry/platform/platform.go index 3f99afdd1..ba031516a 100644 --- a/pkg/sentry/platform/platform.go +++ b/pkg/sentry/platform/platform.go @@ -171,14 +171,20 @@ type Context interface { // Preconditions: The caller must be running on the task goroutine. PullFullState(as AddressSpace, ac arch.Context) - // FloatingPointStateChanged forces restoring a full state of the application thread. + // FullStateChanged() indicates that a thread state has been changed by + // the Sentry. This happens in case of the rt_sigreturn, execve, etc. // - // A platform can support lazy loading/restoring of a thread state. - // This means that if the Sentry has not changed a thread state, + // First, it indicates that the Sentry has the full state of the thread + // and PullFullState() has to do nothing if it is called after + // FullStateChanged(). + // + // Second, it forces restoring the full state of the application + // thread. A platform can support lazy loading/restoring of a thread + // state. This means that if the Sentry has not changed a thread state, // the platform may not restore it. // // Preconditions: The caller must be running on the task goroutine. - FloatingPointStateChanged() + FullStateChanged() // Interrupt interrupts a concurrent call to Switch(), causing it to return // ErrContextInterrupt. diff --git a/pkg/sentry/platform/ptrace/ptrace.go b/pkg/sentry/platform/ptrace/ptrace.go index 45ff2bcc3..b52d0fbd8 100644 --- a/pkg/sentry/platform/ptrace/ptrace.go +++ b/pkg/sentry/platform/ptrace/ptrace.go @@ -182,8 +182,8 @@ func (c *context) Interrupt() { // Release implements platform.Context.Release(). func (c *context) Release() {} -// FloatingPointStateChanged implements platform.Context.FloatingPointStateChanged. -func (c *context) FloatingPointStateChanged() {} +// FullStateChanged implements platform.Context.FullStateChanged. +func (c *context) FullStateChanged() {} // PullFullState implements platform.Context.PullFullState. func (c *context) PullFullState(as platform.AddressSpace, ac arch.Context) {} -- cgit v1.2.3 From 7ba5e3558aace529f9a0553015cc5074705e7a79 Mon Sep 17 00:00:00 2001 From: Kevin Krakauer Date: Mon, 10 Aug 2020 13:49:51 -0700 Subject: ip6tables: move target-specific code to targets.go This is purely moving code, no changes. netfilter.go is cluttered and targets.go is a good place for this. #3549 PiperOrigin-RevId: 325879965 --- pkg/sentry/socket/netfilter/netfilter.go | 240 ------------------------------ pkg/sentry/socket/netfilter/targets.go | 247 +++++++++++++++++++++++++++++++ 2 files changed, 247 insertions(+), 240 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index a9f0604ae..e91b0624c 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -32,15 +32,6 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) -// errorTargetName is used to mark targets as error targets. Error targets -// shouldn't be reached - an error has occurred if we fall through to one. -const errorTargetName = "ERROR" - -// redirectTargetName is used to mark targets as redirect targets. Redirect -// targets should be reached for only NAT and Mangle tables. These targets will -// change the destination port/destination IP for packets. -const redirectTargetName = "REDIRECT" - // enableLogging controls whether to log the (de)serialization of netfilter // structs between userspace and netstack. These logs are useful when // developing iptables, but can pollute sentry logs otherwise. @@ -202,130 +193,6 @@ func convertNetstackToBinary(stack *stack.Stack, tablename linux.TableName) (lin return entries, info, nil } -func marshalTarget(target stack.Target) []byte { - switch tg := target.(type) { - case stack.AcceptTarget: - return marshalStandardTarget(stack.RuleAccept) - case stack.DropTarget: - return marshalStandardTarget(stack.RuleDrop) - case stack.ErrorTarget: - return marshalErrorTarget(errorTargetName) - case stack.UserChainTarget: - return marshalErrorTarget(tg.Name) - case stack.ReturnTarget: - return marshalStandardTarget(stack.RuleReturn) - case stack.RedirectTarget: - return marshalRedirectTarget(tg) - case JumpTarget: - return marshalJumpTarget(tg) - default: - panic(fmt.Errorf("unknown target of type %T", target)) - } -} - -func marshalStandardTarget(verdict stack.RuleVerdict) []byte { - nflog("convert to binary: marshalling standard target") - - // The target's name will be the empty string. - target := linux.XTStandardTarget{ - Target: linux.XTEntryTarget{ - TargetSize: linux.SizeOfXTStandardTarget, - }, - Verdict: translateFromStandardVerdict(verdict), - } - - ret := make([]byte, 0, linux.SizeOfXTStandardTarget) - return binary.Marshal(ret, usermem.ByteOrder, target) -} - -func marshalErrorTarget(errorName string) []byte { - // This is an error target named error - target := linux.XTErrorTarget{ - Target: linux.XTEntryTarget{ - TargetSize: linux.SizeOfXTErrorTarget, - }, - } - copy(target.Name[:], errorName) - copy(target.Target.Name[:], errorTargetName) - - ret := make([]byte, 0, linux.SizeOfXTErrorTarget) - return binary.Marshal(ret, usermem.ByteOrder, target) -} - -func marshalRedirectTarget(rt stack.RedirectTarget) []byte { - // This is a redirect target named redirect - target := linux.XTRedirectTarget{ - Target: linux.XTEntryTarget{ - TargetSize: linux.SizeOfXTRedirectTarget, - }, - } - copy(target.Target.Name[:], redirectTargetName) - - ret := make([]byte, 0, linux.SizeOfXTRedirectTarget) - target.NfRange.RangeSize = 1 - if rt.RangeProtoSpecified { - target.NfRange.RangeIPV4.Flags |= linux.NF_NAT_RANGE_PROTO_SPECIFIED - } - // Convert port from little endian to big endian. - port := make([]byte, 2) - binary.LittleEndian.PutUint16(port, rt.MinPort) - target.NfRange.RangeIPV4.MinPort = binary.BigEndian.Uint16(port) - binary.LittleEndian.PutUint16(port, rt.MaxPort) - target.NfRange.RangeIPV4.MaxPort = binary.BigEndian.Uint16(port) - return binary.Marshal(ret, usermem.ByteOrder, target) -} - -func marshalJumpTarget(jt JumpTarget) []byte { - nflog("convert to binary: marshalling jump target") - - // The target's name will be the empty string. - target := linux.XTStandardTarget{ - Target: linux.XTEntryTarget{ - TargetSize: linux.SizeOfXTStandardTarget, - }, - // Verdict is overloaded by the ABI. When positive, it holds - // the jump offset from the start of the table. - Verdict: int32(jt.Offset), - } - - ret := make([]byte, 0, linux.SizeOfXTStandardTarget) - return binary.Marshal(ret, usermem.ByteOrder, target) -} - -// translateFromStandardVerdict translates verdicts the same way as the iptables -// tool. -func translateFromStandardVerdict(verdict stack.RuleVerdict) int32 { - switch verdict { - case stack.RuleAccept: - return -linux.NF_ACCEPT - 1 - case stack.RuleDrop: - return -linux.NF_DROP - 1 - case stack.RuleReturn: - return linux.NF_RETURN - default: - // TODO(gvisor.dev/issue/170): Support Jump. - panic(fmt.Sprintf("unknown standard verdict: %d", verdict)) - } -} - -// translateToStandardTarget translates from the value in a -// linux.XTStandardTarget to an stack.Verdict. -func translateToStandardTarget(val int32) (stack.Target, error) { - // TODO(gvisor.dev/issue/170): Support other verdicts. - switch val { - case -linux.NF_ACCEPT - 1: - return stack.AcceptTarget{}, nil - case -linux.NF_DROP - 1: - return stack.DropTarget{}, nil - case -linux.NF_QUEUE - 1: - return nil, errors.New("unsupported iptables verdict QUEUE") - case linux.NF_RETURN: - return stack.ReturnTarget{}, nil - default: - return nil, fmt.Errorf("unknown iptables verdict %d", val) - } -} - // SetEntries sets iptables rules for a single table. See // net/ipv4/netfilter/ip_tables.c:translate_table for reference. func SetEntries(stk *stack.Stack, optVal []byte) *syserr.Error { @@ -562,113 +429,6 @@ func parseMatchers(filter stack.IPHeaderFilter, optVal []byte) ([]stack.Matcher, return matchers, nil } -// parseTarget parses a target from optVal. optVal should contain only the -// target. -func parseTarget(filter stack.IPHeaderFilter, optVal []byte) (stack.Target, error) { - nflog("set entries: parsing target of size %d", len(optVal)) - if len(optVal) < linux.SizeOfXTEntryTarget { - return nil, fmt.Errorf("optVal has insufficient size for entry target %d", len(optVal)) - } - var target linux.XTEntryTarget - buf := optVal[:linux.SizeOfXTEntryTarget] - binary.Unmarshal(buf, usermem.ByteOrder, &target) - switch target.Name.String() { - case "": - // Standard target. - if len(optVal) != linux.SizeOfXTStandardTarget { - return nil, fmt.Errorf("optVal has wrong size for standard target %d", len(optVal)) - } - var standardTarget linux.XTStandardTarget - buf = optVal[:linux.SizeOfXTStandardTarget] - binary.Unmarshal(buf, usermem.ByteOrder, &standardTarget) - - if standardTarget.Verdict < 0 { - // A Verdict < 0 indicates a non-jump verdict. - return translateToStandardTarget(standardTarget.Verdict) - } - // A verdict >= 0 indicates a jump. - return JumpTarget{Offset: uint32(standardTarget.Verdict)}, nil - - case errorTargetName: - // Error target. - if len(optVal) != linux.SizeOfXTErrorTarget { - return nil, fmt.Errorf("optVal has insufficient size for error target %d", len(optVal)) - } - var errorTarget linux.XTErrorTarget - buf = optVal[:linux.SizeOfXTErrorTarget] - binary.Unmarshal(buf, usermem.ByteOrder, &errorTarget) - - // Error targets are used in 2 cases: - // * An actual error case. These rules have an error - // named errorTargetName. The last entry of the table - // is usually an error case to catch any packets that - // somehow fall through every rule. - // * To mark the start of a user defined chain. These - // rules have an error with the name of the chain. - switch name := errorTarget.Name.String(); name { - case errorTargetName: - nflog("set entries: error target") - return stack.ErrorTarget{}, nil - default: - // User defined chain. - nflog("set entries: user-defined target %q", name) - return stack.UserChainTarget{Name: name}, nil - } - - case redirectTargetName: - // Redirect target. - if len(optVal) < linux.SizeOfXTRedirectTarget { - return nil, fmt.Errorf("netfilter.SetEntries: optVal has insufficient size for redirect target %d", len(optVal)) - } - - if filter.Protocol != header.TCPProtocolNumber && filter.Protocol != header.UDPProtocolNumber { - return nil, fmt.Errorf("netfilter.SetEntries: invalid argument") - } - - var redirectTarget linux.XTRedirectTarget - buf = optVal[:linux.SizeOfXTRedirectTarget] - binary.Unmarshal(buf, usermem.ByteOrder, &redirectTarget) - - // Copy linux.XTRedirectTarget to stack.RedirectTarget. - var target stack.RedirectTarget - nfRange := redirectTarget.NfRange - - // RangeSize should be 1. - if nfRange.RangeSize != 1 { - return nil, fmt.Errorf("netfilter.SetEntries: invalid argument") - } - - // TODO(gvisor.dev/issue/170): Check if the flags are valid. - // Also check if we need to map ports or IP. - // For now, redirect target only supports destination port change. - // Port range and IP range are not supported yet. - if nfRange.RangeIPV4.Flags&linux.NF_NAT_RANGE_PROTO_SPECIFIED == 0 { - return nil, fmt.Errorf("netfilter.SetEntries: invalid argument") - } - target.RangeProtoSpecified = true - - target.MinIP = tcpip.Address(nfRange.RangeIPV4.MinIP[:]) - target.MaxIP = tcpip.Address(nfRange.RangeIPV4.MaxIP[:]) - - // TODO(gvisor.dev/issue/170): Port range is not supported yet. - if nfRange.RangeIPV4.MinPort != nfRange.RangeIPV4.MaxPort { - return nil, fmt.Errorf("netfilter.SetEntries: invalid argument") - } - - // Convert port from big endian to little endian. - port := make([]byte, 2) - binary.BigEndian.PutUint16(port, nfRange.RangeIPV4.MinPort) - target.MinPort = binary.LittleEndian.Uint16(port) - - binary.BigEndian.PutUint16(port, nfRange.RangeIPV4.MaxPort) - target.MaxPort = binary.LittleEndian.Uint16(port) - return target, nil - } - - // Unknown target. - return nil, fmt.Errorf("unknown target %q doesn't exist or isn't supported yet.", target.Name.String()) -} - func filterFromIPTIP(iptip linux.IPTIP) (stack.IPHeaderFilter, error) { if containsUnsupportedFields(iptip) { return stack.IPHeaderFilter{}, fmt.Errorf("unsupported fields in struct iptip: %+v", iptip) diff --git a/pkg/sentry/socket/netfilter/targets.go b/pkg/sentry/socket/netfilter/targets.go index b91ba3ab3..8ebdaff18 100644 --- a/pkg/sentry/socket/netfilter/targets.go +++ b/pkg/sentry/socket/netfilter/targets.go @@ -15,10 +15,257 @@ package netfilter import ( + "errors" + "fmt" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" + "gvisor.dev/gvisor/pkg/usermem" ) +// errorTargetName is used to mark targets as error targets. Error targets +// shouldn't be reached - an error has occurred if we fall through to one. +const errorTargetName = "ERROR" + +// redirectTargetName is used to mark targets as redirect targets. Redirect +// targets should be reached for only NAT and Mangle tables. These targets will +// change the destination port/destination IP for packets. +const redirectTargetName = "REDIRECT" + +func marshalTarget(target stack.Target) []byte { + switch tg := target.(type) { + case stack.AcceptTarget: + return marshalStandardTarget(stack.RuleAccept) + case stack.DropTarget: + return marshalStandardTarget(stack.RuleDrop) + case stack.ErrorTarget: + return marshalErrorTarget(errorTargetName) + case stack.UserChainTarget: + return marshalErrorTarget(tg.Name) + case stack.ReturnTarget: + return marshalStandardTarget(stack.RuleReturn) + case stack.RedirectTarget: + return marshalRedirectTarget(tg) + case JumpTarget: + return marshalJumpTarget(tg) + default: + panic(fmt.Errorf("unknown target of type %T", target)) + } +} + +func marshalStandardTarget(verdict stack.RuleVerdict) []byte { + nflog("convert to binary: marshalling standard target") + + // The target's name will be the empty string. + target := linux.XTStandardTarget{ + Target: linux.XTEntryTarget{ + TargetSize: linux.SizeOfXTStandardTarget, + }, + Verdict: translateFromStandardVerdict(verdict), + } + + ret := make([]byte, 0, linux.SizeOfXTStandardTarget) + return binary.Marshal(ret, usermem.ByteOrder, target) +} + +func marshalErrorTarget(errorName string) []byte { + // This is an error target named error + target := linux.XTErrorTarget{ + Target: linux.XTEntryTarget{ + TargetSize: linux.SizeOfXTErrorTarget, + }, + } + copy(target.Name[:], errorName) + copy(target.Target.Name[:], errorTargetName) + + ret := make([]byte, 0, linux.SizeOfXTErrorTarget) + return binary.Marshal(ret, usermem.ByteOrder, target) +} + +func marshalRedirectTarget(rt stack.RedirectTarget) []byte { + // This is a redirect target named redirect + target := linux.XTRedirectTarget{ + Target: linux.XTEntryTarget{ + TargetSize: linux.SizeOfXTRedirectTarget, + }, + } + copy(target.Target.Name[:], redirectTargetName) + + ret := make([]byte, 0, linux.SizeOfXTRedirectTarget) + target.NfRange.RangeSize = 1 + if rt.RangeProtoSpecified { + target.NfRange.RangeIPV4.Flags |= linux.NF_NAT_RANGE_PROTO_SPECIFIED + } + // Convert port from little endian to big endian. + port := make([]byte, 2) + binary.LittleEndian.PutUint16(port, rt.MinPort) + target.NfRange.RangeIPV4.MinPort = binary.BigEndian.Uint16(port) + binary.LittleEndian.PutUint16(port, rt.MaxPort) + target.NfRange.RangeIPV4.MaxPort = binary.BigEndian.Uint16(port) + return binary.Marshal(ret, usermem.ByteOrder, target) +} + +func marshalJumpTarget(jt JumpTarget) []byte { + nflog("convert to binary: marshalling jump target") + + // The target's name will be the empty string. + target := linux.XTStandardTarget{ + Target: linux.XTEntryTarget{ + TargetSize: linux.SizeOfXTStandardTarget, + }, + // Verdict is overloaded by the ABI. When positive, it holds + // the jump offset from the start of the table. + Verdict: int32(jt.Offset), + } + + ret := make([]byte, 0, linux.SizeOfXTStandardTarget) + return binary.Marshal(ret, usermem.ByteOrder, target) +} + +// translateFromStandardVerdict translates verdicts the same way as the iptables +// tool. +func translateFromStandardVerdict(verdict stack.RuleVerdict) int32 { + switch verdict { + case stack.RuleAccept: + return -linux.NF_ACCEPT - 1 + case stack.RuleDrop: + return -linux.NF_DROP - 1 + case stack.RuleReturn: + return linux.NF_RETURN + default: + // TODO(gvisor.dev/issue/170): Support Jump. + panic(fmt.Sprintf("unknown standard verdict: %d", verdict)) + } +} + +// translateToStandardTarget translates from the value in a +// linux.XTStandardTarget to an stack.Verdict. +func translateToStandardTarget(val int32) (stack.Target, error) { + // TODO(gvisor.dev/issue/170): Support other verdicts. + switch val { + case -linux.NF_ACCEPT - 1: + return stack.AcceptTarget{}, nil + case -linux.NF_DROP - 1: + return stack.DropTarget{}, nil + case -linux.NF_QUEUE - 1: + return nil, errors.New("unsupported iptables verdict QUEUE") + case linux.NF_RETURN: + return stack.ReturnTarget{}, nil + default: + return nil, fmt.Errorf("unknown iptables verdict %d", val) + } +} + +// parseTarget parses a target from optVal. optVal should contain only the +// target. +func parseTarget(filter stack.IPHeaderFilter, optVal []byte) (stack.Target, error) { + nflog("set entries: parsing target of size %d", len(optVal)) + if len(optVal) < linux.SizeOfXTEntryTarget { + return nil, fmt.Errorf("optVal has insufficient size for entry target %d", len(optVal)) + } + var target linux.XTEntryTarget + buf := optVal[:linux.SizeOfXTEntryTarget] + binary.Unmarshal(buf, usermem.ByteOrder, &target) + switch target.Name.String() { + case "": + // Standard target. + if len(optVal) != linux.SizeOfXTStandardTarget { + return nil, fmt.Errorf("optVal has wrong size for standard target %d", len(optVal)) + } + var standardTarget linux.XTStandardTarget + buf = optVal[:linux.SizeOfXTStandardTarget] + binary.Unmarshal(buf, usermem.ByteOrder, &standardTarget) + + if standardTarget.Verdict < 0 { + // A Verdict < 0 indicates a non-jump verdict. + return translateToStandardTarget(standardTarget.Verdict) + } + // A verdict >= 0 indicates a jump. + return JumpTarget{Offset: uint32(standardTarget.Verdict)}, nil + + case errorTargetName: + // Error target. + if len(optVal) != linux.SizeOfXTErrorTarget { + return nil, fmt.Errorf("optVal has insufficient size for error target %d", len(optVal)) + } + var errorTarget linux.XTErrorTarget + buf = optVal[:linux.SizeOfXTErrorTarget] + binary.Unmarshal(buf, usermem.ByteOrder, &errorTarget) + + // Error targets are used in 2 cases: + // * An actual error case. These rules have an error + // named errorTargetName. The last entry of the table + // is usually an error case to catch any packets that + // somehow fall through every rule. + // * To mark the start of a user defined chain. These + // rules have an error with the name of the chain. + switch name := errorTarget.Name.String(); name { + case errorTargetName: + nflog("set entries: error target") + return stack.ErrorTarget{}, nil + default: + // User defined chain. + nflog("set entries: user-defined target %q", name) + return stack.UserChainTarget{Name: name}, nil + } + + case redirectTargetName: + // Redirect target. + if len(optVal) < linux.SizeOfXTRedirectTarget { + return nil, fmt.Errorf("netfilter.SetEntries: optVal has insufficient size for redirect target %d", len(optVal)) + } + + if filter.Protocol != header.TCPProtocolNumber && filter.Protocol != header.UDPProtocolNumber { + return nil, fmt.Errorf("netfilter.SetEntries: invalid argument") + } + + var redirectTarget linux.XTRedirectTarget + buf = optVal[:linux.SizeOfXTRedirectTarget] + binary.Unmarshal(buf, usermem.ByteOrder, &redirectTarget) + + // Copy linux.XTRedirectTarget to stack.RedirectTarget. + var target stack.RedirectTarget + nfRange := redirectTarget.NfRange + + // RangeSize should be 1. + if nfRange.RangeSize != 1 { + return nil, fmt.Errorf("netfilter.SetEntries: invalid argument") + } + + // TODO(gvisor.dev/issue/170): Check if the flags are valid. + // Also check if we need to map ports or IP. + // For now, redirect target only supports destination port change. + // Port range and IP range are not supported yet. + if nfRange.RangeIPV4.Flags&linux.NF_NAT_RANGE_PROTO_SPECIFIED == 0 { + return nil, fmt.Errorf("netfilter.SetEntries: invalid argument") + } + target.RangeProtoSpecified = true + + target.MinIP = tcpip.Address(nfRange.RangeIPV4.MinIP[:]) + target.MaxIP = tcpip.Address(nfRange.RangeIPV4.MaxIP[:]) + + // TODO(gvisor.dev/issue/170): Port range is not supported yet. + if nfRange.RangeIPV4.MinPort != nfRange.RangeIPV4.MaxPort { + return nil, fmt.Errorf("netfilter.SetEntries: invalid argument") + } + + // Convert port from big endian to little endian. + port := make([]byte, 2) + binary.BigEndian.PutUint16(port, nfRange.RangeIPV4.MinPort) + target.MinPort = binary.LittleEndian.Uint16(port) + + binary.BigEndian.PutUint16(port, nfRange.RangeIPV4.MaxPort) + target.MaxPort = binary.LittleEndian.Uint16(port) + return target, nil + } + + // Unknown target. + return nil, fmt.Errorf("unknown target %q doesn't exist or isn't supported yet", target.Name.String()) +} + // JumpTarget implements stack.Target. type JumpTarget struct { // Offset is the byte offset of the rule to jump to. It is used for -- cgit v1.2.3 From 51e64d2fc590b0271d4e0cbbc75882cf81ada182 Mon Sep 17 00:00:00 2001 From: Craig Chi Date: Mon, 10 Aug 2020 18:15:32 -0700 Subject: Implement FUSE_GETATTR FUSE_GETATTR is called when a stat(2), fstat(2), or lstat(2) is issued from VFS2 layer to a FUSE filesystem. Fixes #3175 --- pkg/abi/linux/fuse.go | 55 +++++++++++++ pkg/sentry/fsimpl/fuse/fusefs.go | 96 ++++++++++++++++++++++ test/fuse/BUILD | 8 ++ test/fuse/linux/BUILD | 21 +++-- test/fuse/linux/fuse_base.cc | 15 ++-- test/fuse/linux/fuse_base.h | 6 +- test/fuse/linux/stat_test.cc | 169 +++++++++++++++++++++++++++++++++++++++ test/runner/defs.bzl | 45 +++++------ 8 files changed, 375 insertions(+), 40 deletions(-) create mode 100644 test/fuse/linux/stat_test.cc (limited to 'pkg/sentry') diff --git a/pkg/abi/linux/fuse.go b/pkg/abi/linux/fuse.go index 5c6ffe4a3..7e30483ee 100644 --- a/pkg/abi/linux/fuse.go +++ b/pkg/abi/linux/fuse.go @@ -246,3 +246,58 @@ type FUSEInitOut struct { _ [8]uint32 } + +// FUSEGetAttrIn is the request sent by the kernel to the daemon, +// to get the attribute of a inode. +// +// +marshal +type FUSEGetAttrIn struct { + // GetAttrFlags specifies whether getattr request is sent with a nodeid or + // with a file handle. + GetAttrFlags uint32 + + _ uint32 + + // Fh is the file handler when GetAttrFlags has FUSE_GETATTR_FH bit. If + // used, the operation is analogous to fstat(2). + Fh uint64 +} + +// FUSEAttr is the struct used in the reponse FUSEGetAttrOut. +// +// +marshal +type FUSEAttr struct { + Ino uint64 + Size uint64 + Blocks uint64 + Atime uint64 + Mtime uint64 + Ctime uint64 + AtimeNsec uint32 + MtimeNsec uint32 + CtimeNsec uint32 + Mode uint32 + Nlink uint32 + UID uint32 + GID uint32 + Rdev uint32 + BlkSize uint32 + _ uint32 +} + +// FUSEGetAttrOut is the reply sent by the daemon to the kernel +// for FUSEGetAttrIn. +// +// +marshal +type FUSEGetAttrOut struct { + // AttrValid and AttrValidNsec describe the attribute cache duration + AttrValid uint64 + + // AttrValidNsec is the nanosecond part of the attribute cache duration + AttrValidNsec uint32 + + _ uint32 + + // Attr contains the metadata returned from the FUSE server + Attr FUSEAttr +} diff --git a/pkg/sentry/fsimpl/fuse/fusefs.go b/pkg/sentry/fsimpl/fuse/fusefs.go index a1405f7c3..83c24ec25 100644 --- a/pkg/sentry/fsimpl/fuse/fusefs.go +++ b/pkg/sentry/fsimpl/fuse/fusefs.go @@ -226,3 +226,99 @@ func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentr } return fd.VFSFileDescription(), nil } + +// statFromFUSEAttr makes attributes from linux.FUSEAttr to linux.Statx. The +// opts.Sync attribute is ignored since the synchronization is handled by the +// FUSE server. +func statFromFUSEAttr(attr linux.FUSEAttr, mask, devMinor uint32) linux.Statx { + var stat linux.Statx + stat.Blksize = attr.BlkSize + stat.DevMajor, stat.DevMinor = linux.UNNAMED_MAJOR, devMinor + + rdevMajor, rdevMinor := linux.DecodeDeviceID(attr.Rdev) + stat.RdevMajor, stat.RdevMinor = uint32(rdevMajor), rdevMinor + + if mask&linux.STATX_MODE != 0 { + stat.Mode = uint16(attr.Mode) + } + if mask&linux.STATX_NLINK != 0 { + stat.Nlink = attr.Nlink + } + if mask&linux.STATX_UID != 0 { + stat.UID = attr.UID + } + if mask&linux.STATX_GID != 0 { + stat.GID = attr.GID + } + if mask&linux.STATX_ATIME != 0 { + stat.Atime = linux.StatxTimestamp{ + Sec: int64(attr.Atime), + Nsec: attr.AtimeNsec, + } + } + if mask&linux.STATX_MTIME != 0 { + stat.Mtime = linux.StatxTimestamp{ + Sec: int64(attr.Mtime), + Nsec: attr.MtimeNsec, + } + } + if mask&linux.STATX_CTIME != 0 { + stat.Ctime = linux.StatxTimestamp{ + Sec: int64(attr.Ctime), + Nsec: attr.CtimeNsec, + } + } + if mask&linux.STATX_INO != 0 { + stat.Ino = attr.Ino + } + if mask&linux.STATX_SIZE != 0 { + stat.Size = attr.Size + } + if mask&linux.STATX_BLOCKS != 0 { + stat.Blocks = attr.Blocks + } + return stat +} + +// Stat implements kernfs.Inode.Stat. +func (i *inode) Stat(ctx context.Context, fs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { + fusefs := fs.Impl().(*filesystem) + conn := fusefs.conn + task, creds := kernel.TaskFromContext(ctx), auth.CredentialsFromContext(ctx) + if task == nil { + log.Warningf("couldn't get kernel task from context") + return linux.Statx{}, syserror.EINVAL + } + + var in linux.FUSEGetAttrIn + // We don't set any attribute in the request, because in VFS2 fstat(2) will + // finally be translated into vfs.FilesystemImpl.StatAt() (see + // pkg/sentry/syscalls/linux/vfs2/stat.go), resulting in the same flow + // as stat(2). Thus GetAttrFlags and Fh variable will never be used in VFS2. + req, err := conn.NewRequest(creds, uint32(task.ThreadID()), i.Ino(), linux.FUSE_GETATTR, &in) + if err != nil { + return linux.Statx{}, err + } + + res, err := conn.Call(task, req) + if err != nil { + return linux.Statx{}, err + } + if err := res.Error(); err != nil { + return linux.Statx{}, err + } + + var out linux.FUSEGetAttrOut + if err := res.UnmarshalPayload(&out); err != nil { + return linux.Statx{}, err + } + + // Set all metadata into kernfs.InodeAttrs. + if err := i.SetStat(ctx, fs, creds, vfs.SetStatOptions{ + Stat: statFromFUSEAttr(out.Attr, linux.STATX_ALL, fusefs.devMinor), + }); err != nil { + return linux.Statx{}, err + } + + return statFromFUSEAttr(out.Attr, opts.Mask, fusefs.devMinor), nil +} diff --git a/test/fuse/BUILD b/test/fuse/BUILD index 34b950644..56157c96b 100644 --- a/test/fuse/BUILD +++ b/test/fuse/BUILD @@ -1 +1,9 @@ +load("//test/runner:defs.bzl", "syscall_test") + package(licenses = ["notice"]) + +syscall_test( + fuse = "True", + test = "//test/fuse/linux:stat_test", + vfs2 = "True", +) diff --git a/test/fuse/linux/BUILD b/test/fuse/linux/BUILD index 49dc96c20..4871bb531 100644 --- a/test/fuse/linux/BUILD +++ b/test/fuse/linux/BUILD @@ -1,20 +1,31 @@ -load("//tools:defs.bzl", "cc_library", "gtest") +load("//tools:defs.bzl", "cc_binary", "cc_library", "gtest") package( default_visibility = ["//:sandbox"], licenses = ["notice"], ) +cc_binary( + name = "stat_test", + testonly = 1, + srcs = ["stat_test.cc"], + deps = [ + gtest, + ":fuse_base", + "//test/util:test_main", + "//test/util:test_util", + ], +) + cc_library( name = "fuse_base", testonly = 1, - srcs = [ - "fuse_base.cc", - "fuse_base.h", - ], + srcs = ["fuse_base.cc"], + hdrs = ["fuse_base.h"], deps = [ gtest, "//test/util:posix_error", + "//test/util:temp_path", "//test/util:test_util", "@com_google_absl//absl/strings:str_format", ], diff --git a/test/fuse/linux/fuse_base.cc b/test/fuse/linux/fuse_base.cc index ce69276c9..4a2c64998 100644 --- a/test/fuse/linux/fuse_base.cc +++ b/test/fuse/linux/fuse_base.cc @@ -12,23 +12,23 @@ // See the License for the specific language governing permissions and // limitations under the License. -#include "fuse_base.h" +#include "test/fuse/linux/fuse_base.h" #include #include -#include #include #include #include +#include #include #include #include -#include "gmock/gmock.h" -#include "gtest/gtest.h" #include "absl/strings/str_format.h" +#include "gtest/gtest.h" #include "test/util/posix_error.h" +#include "test/util/temp_path.h" #include "test/util/test_util.h" namespace gvisor { @@ -78,13 +78,14 @@ void FuseTest::MountFuse() { EXPECT_THAT(dev_fd_ = open("/dev/fuse", O_RDWR), SyscallSucceeds()); std::string mount_opts = absl::StrFormat("fd=%d,%s", dev_fd_, kMountOpts); - EXPECT_THAT(mount("fuse", kMountPoint, "fuse", MS_NODEV | MS_NOSUID, - mount_opts.c_str()), + mount_point_ = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + EXPECT_THAT(mount("fuse", mount_point_.path().c_str(), "fuse", + MS_NODEV | MS_NOSUID, mount_opts.c_str()), SyscallSucceedsWithValue(0)); } void FuseTest::UnmountFuse() { - EXPECT_THAT(umount(kMountPoint), SyscallSucceeds()); + EXPECT_THAT(umount(mount_point_.path().c_str()), SyscallSucceeds()); // TODO(gvisor.dev/issue/3330): ensure the process is terminated successfully. } diff --git a/test/fuse/linux/fuse_base.h b/test/fuse/linux/fuse_base.h index b008778de..3a2f255a9 100644 --- a/test/fuse/linux/fuse_base.h +++ b/test/fuse/linux/fuse_base.h @@ -20,14 +20,13 @@ #include -#include "gmock/gmock.h" #include "gtest/gtest.h" #include "test/util/posix_error.h" +#include "test/util/temp_path.h" namespace gvisor { namespace testing { -constexpr char kMountPoint[] = "/mnt"; constexpr char kMountOpts[] = "rootmode=755,user_id=0,group_id=0"; class FuseTest : public ::testing::Test { @@ -55,6 +54,9 @@ class FuseTest : public ::testing::Test { // complains if the FUSE server responds failure during tests. void WaitCompleted(); + protected: + TempPath mount_point_; + private: void MountFuse(); void UnmountFuse(); diff --git a/test/fuse/linux/stat_test.cc b/test/fuse/linux/stat_test.cc new file mode 100644 index 000000000..172e09867 --- /dev/null +++ b/test/fuse/linux/stat_test.cc @@ -0,0 +1,169 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include +#include +#include +#include +#include +#include +#include + +#include + +#include "gtest/gtest.h" +#include "test/fuse/linux/fuse_base.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +namespace { + +class StatTest : public FuseTest { + public: + bool CompareRequest(void* expected_mem, size_t expected_len, void* real_mem, + size_t real_len) override { + if (expected_len != real_len) return false; + struct fuse_in_header* real_header = + reinterpret_cast(real_mem); + + if (real_header->opcode != FUSE_GETATTR) { + std::cerr << "expect header opcode " << FUSE_GETATTR << " but got " + << real_header->opcode << std::endl; + return false; + } + return true; + } + + bool StatsAreEqual(struct stat expected, struct stat actual) { + // device number will be dynamically allocated by kernel, we cannot know + // in advance + actual.st_dev = expected.st_dev; + return memcmp(&expected, &actual, sizeof(struct stat)) == 0; + } +}; + +TEST_F(StatTest, StatNormal) { + struct iovec iov_in[2]; + struct iovec iov_out[2]; + + struct fuse_in_header in_header = { + .len = sizeof(struct fuse_in_header) + sizeof(struct fuse_getattr_in), + .opcode = FUSE_GETATTR, + .unique = 4, + .nodeid = 1, + .uid = 0, + .gid = 0, + .pid = 4, + .padding = 0, + }; + struct fuse_getattr_in in_payload = {0}; + iov_in[0].iov_len = sizeof(in_header); + iov_in[0].iov_base = &in_header; + iov_in[1].iov_len = sizeof(in_payload); + iov_in[1].iov_base = &in_payload; + + mode_t expected_mode = S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH; + struct timespec atime = {.tv_sec = 1595436289, .tv_nsec = 134150844}; + struct timespec mtime = {.tv_sec = 1595436290, .tv_nsec = 134150845}; + struct timespec ctime = {.tv_sec = 1595436291, .tv_nsec = 134150846}; + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_attr_out), + .error = 0, + .unique = 4, + }; + struct fuse_attr attr = { + .ino = 1, + .size = 512, + .blocks = 4, + .atime = static_cast(atime.tv_sec), + .mtime = static_cast(mtime.tv_sec), + .ctime = static_cast(ctime.tv_sec), + .atimensec = static_cast(atime.tv_nsec), + .mtimensec = static_cast(mtime.tv_nsec), + .ctimensec = static_cast(ctime.tv_nsec), + .mode = expected_mode, + .nlink = 2, + .uid = 1234, + .gid = 4321, + .rdev = 12, + .blksize = 4096, + }; + struct fuse_attr_out out_payload = { + .attr = attr, + }; + iov_out[0].iov_len = sizeof(out_header); + iov_out[0].iov_base = &out_header; + iov_out[1].iov_len = sizeof(out_payload); + iov_out[1].iov_base = &out_payload; + + SetExpected(iov_in, 2, iov_out, 2); + + struct stat stat_buf; + EXPECT_THAT(stat(mount_point_.path().c_str(), &stat_buf), SyscallSucceeds()); + + struct stat expected_stat = { + .st_ino = attr.ino, + .st_nlink = attr.nlink, + .st_mode = expected_mode, + .st_uid = attr.uid, + .st_gid = attr.gid, + .st_rdev = attr.rdev, + .st_size = static_cast(attr.size), + .st_blksize = attr.blksize, + .st_blocks = static_cast(attr.blocks), + .st_atim = atime, + .st_mtim = mtime, + .st_ctim = ctime, + }; + EXPECT_TRUE(StatsAreEqual(stat_buf, expected_stat)); + WaitCompleted(); +} + +TEST_F(StatTest, StatNotFound) { + struct iovec iov_in[2]; + struct iovec iov_out[2]; + + struct fuse_in_header in_header = { + .len = sizeof(struct fuse_in_header) + sizeof(struct fuse_getattr_in), + .opcode = FUSE_GETATTR, + .unique = 4, + }; + struct fuse_getattr_in in_payload = {0}; + iov_in[0].iov_len = sizeof(in_header); + iov_in[0].iov_base = &in_header; + iov_in[1].iov_len = sizeof(in_payload); + iov_in[1].iov_base = &in_payload; + + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header), + .error = -ENOENT, + .unique = 4, + }; + iov_out[0].iov_len = sizeof(out_header); + iov_out[0].iov_base = &out_header; + + SetExpected(iov_in, 2, iov_out, 1); + + struct stat stat_buf; + EXPECT_THAT(stat(mount_point_.path().c_str(), &stat_buf), + SyscallFailsWithErrno(ENOENT)); + WaitCompleted(); +} + +} // namespace + +} // namespace testing +} // namespace gvisor diff --git a/test/runner/defs.bzl b/test/runner/defs.bzl index ba4732ca6..1ae13a4a5 100644 --- a/test/runner/defs.bzl +++ b/test/runner/defs.bzl @@ -150,31 +150,12 @@ def syscall_test( if not tags: tags = [] - _syscall_test( - test = test, - shard_count = shard_count, - size = size, - platform = "native", - use_tmpfs = False, - add_uds_tree = add_uds_tree, - tags = list(tags), - ) - - for (platform, platform_tags) in platforms.items(): - _syscall_test( - test = test, - shard_count = shard_count, - size = size, - platform = platform, - use_tmpfs = use_tmpfs, - add_uds_tree = add_uds_tree, - tags = platform_tags + tags, - ) - vfs2_tags = list(tags) if vfs2: # Add tag to easily run VFS2 tests with --test_tag_filters=vfs2 vfs2_tags.append("vfs2") + if fuse: + vfs2_tags.append("fuse") else: # Don't automatically run tests tests not yet passing. @@ -191,19 +172,31 @@ def syscall_test( add_uds_tree = add_uds_tree, tags = platforms[default_platform] + vfs2_tags, vfs2 = True, + fuse = fuse, + ) + if fuse: + # Only generate *_vfs2_fuse target if fuse parameter is enabled. + return + + _syscall_test( + test = test, + shard_count = shard_count, + size = size, + platform = "native", + use_tmpfs = False, + add_uds_tree = add_uds_tree, + tags = list(tags), ) - if vfs2 and fuse: + for (platform, platform_tags) in platforms.items(): _syscall_test( test = test, shard_count = shard_count, size = size, - platform = default_platform, + platform = platform, use_tmpfs = use_tmpfs, add_uds_tree = add_uds_tree, - tags = platforms[default_platform] + vfs2_tags + ["fuse"], - vfs2 = True, - fuse = True, + tags = platform_tags + tags, ) # TODO(gvisor.dev/issue/1487): Enable VFS2 overlay tests. -- cgit v1.2.3 From 05d742ede47bf1603115fca593e2266cd6e2b309 Mon Sep 17 00:00:00 2001 From: Bin Lu Date: Wed, 12 Aug 2020 05:37:27 -0400 Subject: Running hello-world on Thunderx2 with kvm Signed-off-by: Bin Lu --- pkg/sentry/arch/arch_aarch64.go | 2 + pkg/sentry/arch/arch_arm64.go | 2 + pkg/sentry/platform/kvm/kvm_const_arm64.go | 1 + pkg/sentry/platform/kvm/machine_arm64_unsafe.go | 11 ++++- pkg/sentry/platform/ring0/entry_arm64.s | 56 +++++++++------------- pkg/sentry/platform/ring0/kernel_arm64.go | 2 - .../ring0/pagetables/pagetables_aarch64.go | 9 ++-- 7 files changed, 43 insertions(+), 40 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/arch/arch_aarch64.go b/pkg/sentry/arch/arch_aarch64.go index fd95eb2d2..0f433ee79 100644 --- a/pkg/sentry/arch/arch_aarch64.go +++ b/pkg/sentry/arch/arch_aarch64.go @@ -101,6 +101,8 @@ func NewFloatingPointData() *FloatingPointData { // State contains the common architecture bits for aarch64 (the build tag of this // file ensures it's only built on aarch64). +// +// +stateify savable type State struct { // The system registers. Regs Registers diff --git a/pkg/sentry/arch/arch_arm64.go b/pkg/sentry/arch/arch_arm64.go index cabbf60e0..550741d8c 100644 --- a/pkg/sentry/arch/arch_arm64.go +++ b/pkg/sentry/arch/arch_arm64.go @@ -73,6 +73,8 @@ const ( ) // context64 represents an ARM64 context. +// +// +stateify savable type context64 struct { State sigFPState []aarch64FPState // fpstate to be restored on sigreturn. diff --git a/pkg/sentry/platform/kvm/kvm_const_arm64.go b/pkg/sentry/platform/kvm/kvm_const_arm64.go index fdc599477..9a7be3655 100644 --- a/pkg/sentry/platform/kvm/kvm_const_arm64.go +++ b/pkg/sentry/platform/kvm/kvm_const_arm64.go @@ -72,6 +72,7 @@ const ( _TCR_T0SZ_VA48 = 64 - 48 // VA=48 _TCR_T1SZ_VA48 = 64 - 48 // VA=48 + _TCR_A1 = 1 << 22 _TCR_ASID16 = 1 << 36 _TCR_TBI0 = 1 << 37 diff --git a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go index 307a7645f..905712076 100644 --- a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go @@ -61,7 +61,6 @@ func (c *vCPU) initArchState() error { reg.addr = uint64(reflect.ValueOf(&data).Pointer()) regGet.addr = uint64(reflect.ValueOf(&dataGet).Pointer()) - vcpuInit.target = _KVM_ARM_TARGET_GENERIC_V8 vcpuInit.features[0] |= (1 << _KVM_ARM_VCPU_PSCI_0_2) if _, _, errno := syscall.RawSyscall( syscall.SYS_IOCTL, @@ -272,8 +271,16 @@ func (c *vCPU) SwitchToUser(switchOpts ring0.SwitchOpts, info *arch.SignalInfo) return c.fault(int32(syscall.SIGSEGV), info) case ring0.Vector(bounce): // ring0.VirtualizationException return usermem.NoAccess, platform.ErrContextInterrupt + case ring0.El0Sync_undef, + ring0.El1Sync_undef: + *info = arch.SignalInfo{ + Signo: int32(syscall.SIGILL), + Code: 1, // ILL_ILLOPC (illegal opcode). + } + info.SetAddr(switchOpts.Registers.Pc) // Include address. + return usermem.AccessType{}, platform.ErrContextSignal default: - return usermem.NoAccess, platform.ErrContextSignal + panic(fmt.Sprintf("unexpected vector: 0x%x", vector)) } } diff --git a/pkg/sentry/platform/ring0/entry_arm64.s b/pkg/sentry/platform/ring0/entry_arm64.s index d8a7bc2f9..9d29b7168 100644 --- a/pkg/sentry/platform/ring0/entry_arm64.s +++ b/pkg/sentry/platform/ring0/entry_arm64.s @@ -364,6 +364,9 @@ TEXT ·Halt(SB),NOSPLIT,$0 CMP RSV_REG, R9 BNE mmio_exit MOVD $0, CPU_REGISTERS+PTRACE_R9(RSV_REG) + + // Flush dcache. + WORD $0xd5087e52 // DC CISW mmio_exit: // Disable fpsimd. WORD $0xd5381041 // MRS CPACR_EL1, R1 @@ -381,6 +384,9 @@ mmio_exit: MRS VBAR_EL1, R9 MOVD R0, 0x0(R9) + // Flush dcahce. + WORD $0xd5087e52 // DC CISW + RET // HaltAndResume halts execution and point the pointer to the resume function. @@ -414,6 +420,7 @@ TEXT ·Current(SB),NOSPLIT,$0-8 // Prepare the vcpu environment for container application. TEXT ·kernelExitToEl0(SB),NOSPLIT,$0 // Step1, save sentry context into memory. + MRS TPIDR_EL1, RSV_REG REGISTERS_SAVE(RSV_REG, CPU_REGISTERS) MOVD RSV_REG_APP, CPU_REGISTERS+PTRACE_R9(RSV_REG) @@ -425,34 +432,13 @@ TEXT ·kernelExitToEl0(SB),NOSPLIT,$0 MOVD CPU_REGISTERS+PTRACE_R3(RSV_REG), R3 - // Step2, save SP_EL1, PSTATE into kernel temporary stack. - // switch to temporary stack. + // Step2, switch to temporary stack. LOAD_KERNEL_STACK(RSV_REG) - WORD $0xd538d092 //MRS TPIDR_EL1, R18 - - SUB $STACK_FRAME_SIZE, RSP, RSP - MOVD CPU_REGISTERS+PTRACE_SP(RSV_REG), R11 - MOVD CPU_REGISTERS+PTRACE_PSTATE(RSV_REG), R12 - STP (R11, R12), 16*0(RSP) - - MOVD CPU_REGISTERS+PTRACE_R11(RSV_REG), R11 - MOVD CPU_REGISTERS+PTRACE_R12(RSV_REG), R12 - // Step3, test user pagetable. - // If user pagetable is empty, trapped in el1_ia. - WORD $0xd538d092 //MRS TPIDR_EL1, R18 - SWITCH_TO_APP_PAGETABLE(RSV_REG) - WORD $0xd538d092 //MRS TPIDR_EL1, R18 - SWITCH_TO_KVM_PAGETABLE(RSV_REG) - WORD $0xd538d092 //MRS TPIDR_EL1, R18 - - // If pagetable is not empty, recovery kernel temporary stack. - ADD $STACK_FRAME_SIZE, RSP, RSP - - // Step4, load app context pointer. + // Step3, load app context pointer. MOVD CPU_APP_ADDR(RSV_REG), RSV_REG_APP - // Step5, prepare the environment for container application. + // Step4, prepare the environment for container application. // set sp_el0. MOVD PTRACE_SP(RSV_REG_APP), R1 WORD $0xd5184101 //MSR R1, SP_EL0 @@ -480,13 +466,13 @@ TEXT ·kernelExitToEl0(SB),NOSPLIT,$0 LDP 16*0(RSP), (RSV_REG, RSV_REG_APP) ADD $STACK_FRAME_SIZE, RSP, RSP + ISB $15 ERET() // kernelExitToEl1 is the entrypoint for sentry in guest_el1. // Prepare the vcpu environment for sentry. TEXT ·kernelExitToEl1(SB),NOSPLIT,$0 WORD $0xd538d092 //MRS TPIDR_EL1, R18 - MOVD CPU_REGISTERS+PTRACE_PSTATE(RSV_REG), R1 WORD $0xd5184001 //MSR R1, SPSR_EL1 @@ -503,6 +489,8 @@ TEXT ·kernelExitToEl1(SB),NOSPLIT,$0 // Start is the CPU entrypoint. TEXT ·Start(SB),NOSPLIT,$0 + // Flush dcache. + WORD $0xd5087e52 // DC CISW // Init. MOVD $SCTLR_EL1_DEFAULT, R1 MSR R1, SCTLR_EL1 @@ -558,6 +546,7 @@ TEXT ·El1_sync(SB),NOSPLIT,$0 B el1_invalid el1_da: +el1_ia: WORD $0xd538d092 //MRS TPIDR_EL1, R18 WORD $0xd538601a //MRS FAR_EL1, R26 @@ -570,9 +559,6 @@ el1_da: B ·HaltAndResume(SB) -el1_ia: - B ·HaltAndResume(SB) - el1_sp_pc: B ·Shutdown(SB) @@ -644,9 +630,10 @@ el0_svc: MOVD $Syscall, R3 MOVD R3, CPU_VECTOR_CODE(RSV_REG) - B ·HaltAndResume(SB) + B ·kernelExitToEl1(SB) el0_da: +el0_ia: WORD $0xd538d092 //MRS TPIDR_EL1, R18 WORD $0xd538601a //MRS FAR_EL1, R26 @@ -658,10 +645,10 @@ el0_da: MOVD $PageFault, R3 MOVD R3, CPU_VECTOR_CODE(RSV_REG) - B ·HaltAndResume(SB) + MRS ESR_EL1, R3 + MOVD R3, CPU_ERROR_CODE(RSV_REG) -el0_ia: - B ·Shutdown(SB) + B ·kernelExitToEl1(SB) el0_fpsimd_acc: B ·Shutdown(SB) @@ -676,7 +663,10 @@ el0_sp_pc: B ·Shutdown(SB) el0_undef: - B ·Shutdown(SB) + MOVD $El0Sync_undef, R3 + MOVD R3, CPU_VECTOR_CODE(RSV_REG) + + B ·kernelExitToEl1(SB) el0_dbg: B ·Shutdown(SB) diff --git a/pkg/sentry/platform/ring0/kernel_arm64.go b/pkg/sentry/platform/ring0/kernel_arm64.go index 42009dac0..d0afa1aaa 100644 --- a/pkg/sentry/platform/ring0/kernel_arm64.go +++ b/pkg/sentry/platform/ring0/kernel_arm64.go @@ -53,7 +53,6 @@ func IsCanonical(addr uint64) bool { //go:nosplit func (c *CPU) SwitchToUser(switchOpts SwitchOpts) (vector Vector) { - // Sanitize registers. regs := switchOpts.Registers regs.Pstate &= ^uint64(PsrFlagsClear) @@ -69,6 +68,5 @@ func (c *CPU) SwitchToUser(switchOpts SwitchOpts) (vector Vector) { vector = c.vecCode - // Perform the switch. return } diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64.go b/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64.go index 78510ebed..6409d1d91 100644 --- a/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64.go +++ b/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64.go @@ -72,13 +72,14 @@ const ( ) const ( - mtNormal = 0x4 << 2 + mtDevicenGnRE = 0x1 << 2 + mtNormal = 0x4 << 2 ) const ( executeDisable = xn optionMask = 0xfff | 0xfff<<48 - protDefault = accessed | shared | mtNormal + protDefault = accessed | shared ) // MapOpts are x86 options. @@ -184,8 +185,10 @@ func (p *PTE) Set(addr uintptr, opts MapOpts) { if opts.User { v |= user + v |= mtNormal } else { v = v &^ user + v |= mtDevicenGnRE // Strong order for the addresses with ring0.KernelStartAddress. } atomic.StoreUintptr((*uintptr)(p), v) } @@ -200,7 +203,7 @@ func (p *PTE) setPageTable(pt *PageTables, ptes *PTEs) { // This should never happen. panic("unaligned physical address!") } - v := addr | typeTable | protDefault + v := addr | typeTable | protDefault | mtNormal atomic.StoreUintptr((*uintptr)(p), v) } -- cgit v1.2.3 From 776c5f633edc993652bc8e95418729e0420c6324 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Wed, 12 Aug 2020 11:12:12 -0700 Subject: Fix race in vfs.FileDescription.statusFlag PiperOrigin-RevId: 326270643 --- pkg/sentry/vfs/file_description.go | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index d3c1197e3..dcafffe57 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -289,7 +289,7 @@ func (fd *FileDescription) SetStatusFlags(ctx context.Context, creds *auth.Crede if flags&linux.O_DIRECT != 0 && !fd.opts.AllowDirectIO { return syserror.EINVAL } - // TODO(jamieliu): FileDescriptionImpl.SetOAsync()? + // TODO(gvisor.dev/issue/1035): FileDescriptionImpl.SetOAsync()? const settableFlags = linux.O_APPEND | linux.O_ASYNC | linux.O_DIRECT | linux.O_NOATIME | linux.O_NONBLOCK fd.flagsMu.Lock() if fd.asyncHandler != nil { @@ -301,7 +301,7 @@ func (fd *FileDescription) SetStatusFlags(ctx context.Context, creds *auth.Crede fd.asyncHandler.Unregister(fd) } } - fd.statusFlags = (oldFlags &^ settableFlags) | (flags & settableFlags) + atomic.StoreUint32(&fd.statusFlags, (oldFlags&^settableFlags)|(flags&settableFlags)) fd.flagsMu.Unlock() return nil } -- cgit v1.2.3 From c5ed390162930b51719ff45956d7ac1bc963c8e7 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Wed, 12 Aug 2020 13:03:14 -0700 Subject: Release fd references on aio callback cancellation. Discovered by reference leak checker on tmpfs.inode. PiperOrigin-RevId: 326294755 --- pkg/sentry/syscalls/linux/vfs2/aio.go | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/vfs2/aio.go b/pkg/sentry/syscalls/linux/vfs2/aio.go index 399b4f60c..42559bf69 100644 --- a/pkg/sentry/syscalls/linux/vfs2/aio.go +++ b/pkg/sentry/syscalls/linux/vfs2/aio.go @@ -144,6 +144,12 @@ func submitCallback(t *kernel.Task, id uint64, cb *linux.IOCallback, cbAddr user func getAIOCallback(t *kernel.Task, fd, eventFD *vfs.FileDescription, cbAddr usermem.Addr, cb *linux.IOCallback, ioseq usermem.IOSequence, aioCtx *mm.AIOContext) kernel.AIOCallback { return func(ctx context.Context) { + // Release references after completing the callback. + defer fd.DecRef(ctx) + if eventFD != nil { + defer eventFD.DecRef(ctx) + } + if aioCtx.Dead() { aioCtx.CancelPendingRequest() return @@ -169,8 +175,6 @@ func getAIOCallback(t *kernel.Task, fd, eventFD *vfs.FileDescription, cbAddr use ev.Result = -int64(kernel.ExtractErrno(err, 0)) } - fd.DecRef(ctx) - // Queue the result for delivery. aioCtx.FinishRequest(ev) @@ -179,7 +183,6 @@ func getAIOCallback(t *kernel.Task, fd, eventFD *vfs.FileDescription, cbAddr use // wake up. if eventFD != nil { eventFD.Impl().(*eventfd.EventFileDescription).Signal(1) - eventFD.DecRef(ctx) } } } -- cgit v1.2.3 From d797f2666629737920b39a2fbac369ce5853dc35 Mon Sep 17 00:00:00 2001 From: Fabricio Voznika Date: Wed, 12 Aug 2020 13:12:05 -0700 Subject: Redirect TODO Fixes #2923 PiperOrigin-RevId: 326296589 --- pkg/sentry/fsimpl/host/host.go | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index bf922c566..bd6caba06 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -552,7 +552,7 @@ func (f *fileDescription) Allocate(ctx context.Context, mode, offset, length uin return syserror.ESPIPE } - // TODO(gvisor.dev/issue/2923): Implement Allocate for non-pipe hostfds. + // TODO(gvisor.dev/issue/3589): Implement Allocate for non-pipe hostfds. return syserror.EOPNOTSUPP } -- cgit v1.2.3 From 42b610d56750b4bb8e3d69b680e4fb538f8fb554 Mon Sep 17 00:00:00 2001 From: Ayush Ranjan Date: Wed, 12 Aug 2020 17:17:17 -0700 Subject: [vfs2][gofer] Return appropriate errors when opening and creating files. Fixes php test ext/standard/tests/file/touch_variation5.phpt on vfs2. Updates #3516 Also spotted a bug with O_EXCL, where we did not return EEXIST when we tried to open the root of the filesystem with O_EXCL | O_CREAT. Added some more tests for open() corner cases. PiperOrigin-RevId: 326346863 --- pkg/sentry/fsimpl/gofer/filesystem.go | 11 +++++++++++ pkg/sentry/fsimpl/tmpfs/filesystem.go | 2 +- test/syscalls/linux/open.cc | 32 ++++++++++++++++++++++++++++++++ 3 files changed, 44 insertions(+), 1 deletion(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index eaef2594d..40fec890a 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -844,6 +844,13 @@ func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf } } if rp.Done() { + // Reject attempts to open mount root directory with O_CREAT. + if mayCreate && rp.MustBeDir() { + return nil, syserror.EISDIR + } + if mustCreate { + return nil, syserror.EEXIST + } return start.openLocked(ctx, rp, &opts) } @@ -856,6 +863,10 @@ afterTrailingSymlink: if err := parent.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { return nil, err } + // Reject attempts to open directories with O_CREAT. + if mayCreate && rp.MustBeDir() { + return nil, syserror.EISDIR + } // Determine whether or not we need to create a file. parent.dirMu.Lock() child, err := fs.stepLocked(ctx, rp, parent, false /* mayFollowSymlinks */, &ds) diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 065812065..a4864df53 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -320,7 +320,7 @@ func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf fs.mu.Lock() defer fs.mu.Unlock() if rp.Done() { - // Reject attempts to open directories with O_CREAT. + // Reject attempts to open mount root directory with O_CREAT. if rp.MustBeDir() { return nil, syserror.EISDIR } diff --git a/test/syscalls/linux/open.cc b/test/syscalls/linux/open.cc index bf350946b..c7147c20b 100644 --- a/test/syscalls/linux/open.cc +++ b/test/syscalls/linux/open.cc @@ -95,6 +95,38 @@ TEST_F(OpenTest, OTruncAndReadOnlyFile) { Open(dirpath.c_str(), O_TRUNC | O_RDONLY, 0666)); } +TEST_F(OpenTest, OCreateDirectory) { + SKIP_IF(IsRunningWithVFS1()); + auto dirpath = GetAbsoluteTestTmpdir(); + + // Normal case: existing directory. + ASSERT_THAT(open(dirpath.c_str(), O_RDWR | O_CREAT, 0666), + SyscallFailsWithErrno(EISDIR)); + // Trailing separator on existing directory. + ASSERT_THAT(open(dirpath.append("/").c_str(), O_RDWR | O_CREAT, 0666), + SyscallFailsWithErrno(EISDIR)); + // Trailing separator on non-existing directory. + ASSERT_THAT(open(JoinPath(dirpath, "non-existent").append("/").c_str(), + O_RDWR | O_CREAT, 0666), + SyscallFailsWithErrno(EISDIR)); + // "." special case. + ASSERT_THAT(open(JoinPath(dirpath, ".").c_str(), O_RDWR | O_CREAT, 0666), + SyscallFailsWithErrno(EISDIR)); +} + +TEST_F(OpenTest, MustCreateExisting) { + auto dirPath = GetAbsoluteTestTmpdir(); + + // Existing directory. + ASSERT_THAT(open(dirPath.c_str(), O_RDWR | O_CREAT | O_EXCL, 0666), + SyscallFailsWithErrno(EEXIST)); + + // Existing file. + auto newFile = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(dirPath)); + ASSERT_THAT(open(newFile.path().c_str(), O_RDWR | O_CREAT | O_EXCL, 0666), + SyscallFailsWithErrno(EEXIST)); +} + TEST_F(OpenTest, ReadOnly) { char buf; const FileDescriptor ro_file = -- cgit v1.2.3 From 36134667b289a3e96e90abf9e18e1c2137069d6b Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Wed, 12 Aug 2020 18:09:34 -0700 Subject: Add reference leak checking to vfs2 tmpfs.inode. Updates #1486. PiperOrigin-RevId: 326354750 --- pkg/refs_vfs2/refs_template.go | 2 +- pkg/sentry/fsimpl/tmpfs/BUILD | 13 +++++++++++++ pkg/sentry/fsimpl/tmpfs/tmpfs.go | 27 ++++++--------------------- 3 files changed, 20 insertions(+), 22 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/refs_vfs2/refs_template.go b/pkg/refs_vfs2/refs_template.go index 3e5b458c7..99c43c065 100644 --- a/pkg/refs_vfs2/refs_template.go +++ b/pkg/refs_vfs2/refs_template.go @@ -59,7 +59,7 @@ func (r *Refs) finalize() { note = "(Leak checker uninitialized): " } if n := r.ReadRefs(); n != 0 { - log.Warningf("%sAtomicRefCount %p owned by %T garbage collected with ref count of %d (want 0)", note, r, ownerType, n) + log.Warningf("%sRefs %p owned by %T garbage collected with ref count of %d (want 0)", note, r, ownerType, n) } } diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD index e73732a6b..5cd428d64 100644 --- a/pkg/sentry/fsimpl/tmpfs/BUILD +++ b/pkg/sentry/fsimpl/tmpfs/BUILD @@ -26,6 +26,17 @@ go_template_instance( }, ) +go_template_instance( + name = "inode_refs", + out = "inode_refs.go", + package = "tmpfs", + prefix = "inode", + template = "//pkg/refs_vfs2:refs_template", + types = { + "T": "inode", + }, +) + go_library( name = "tmpfs", srcs = [ @@ -34,6 +45,7 @@ go_library( "directory.go", "filesystem.go", "fstree.go", + "inode_refs.go", "named_pipe.go", "regular_file.go", "socket_file.go", @@ -47,6 +59,7 @@ go_library( "//pkg/context", "//pkg/fspath", "//pkg/log", + "//pkg/refs", "//pkg/safemem", "//pkg/sentry/arch", "//pkg/sentry/fs", diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 4681a2f52..de2af6d01 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -285,13 +285,10 @@ type inode struct { // fs is the owning filesystem. fs is immutable. fs *filesystem - // refs is a reference count. refs is accessed using atomic memory - // operations. - // // A reference is held on all inodes as long as they are reachable in the // filesystem tree, i.e. nlink is nonzero. This reference is dropped when // nlink reaches 0. - refs int64 + refs inodeRefs // xattrs implements extended attributes. // @@ -327,7 +324,6 @@ func (i *inode) init(impl interface{}, fs *filesystem, kuid auth.KUID, kgid auth panic("file type is required in FileMode") } i.fs = fs - i.refs = 1 i.mode = uint32(mode) i.uid = uint32(kuid) i.gid = uint32(kgid) @@ -339,6 +335,7 @@ func (i *inode) init(impl interface{}, fs *filesystem, kuid auth.KUID, kgid auth i.mtime = now // i.nlink initialized by caller i.impl = impl + i.refs.EnableLeakCheck() } // incLinksLocked increments i's link count. @@ -369,25 +366,15 @@ func (i *inode) decLinksLocked(ctx context.Context) { } func (i *inode) incRef() { - if atomic.AddInt64(&i.refs, 1) <= 1 { - panic("tmpfs.inode.incRef() called without holding a reference") - } + i.refs.IncRef() } func (i *inode) tryIncRef() bool { - for { - refs := atomic.LoadInt64(&i.refs) - if refs == 0 { - return false - } - if atomic.CompareAndSwapInt64(&i.refs, refs, refs+1) { - return true - } - } + return i.refs.TryIncRef() } func (i *inode) decRef(ctx context.Context) { - if refs := atomic.AddInt64(&i.refs, -1); refs == 0 { + i.refs.DecRef(func() { i.watches.HandleDeletion(ctx) if regFile, ok := i.impl.(*regularFile); ok { // Release memory used by regFile to store data. Since regFile is @@ -395,9 +382,7 @@ func (i *inode) decRef(ctx context.Context) { // metadata. regFile.data.DropAll(regFile.memFile) } - } else if refs < 0 { - panic("tmpfs.inode.decRef() called without holding a reference") - } + }) } func (i *inode) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes) error { -- cgit v1.2.3 From 47515f475167ffa23267ca0b9d1b39e7907587d6 Mon Sep 17 00:00:00 2001 From: Ting-Yu Wang Date: Thu, 13 Aug 2020 13:07:03 -0700 Subject: Migrate to PacketHeader API for PacketBuffer. Formerly, when a packet is constructed or parsed, all headers are set by the client code. This almost always involved prepending to pk.Header buffer or trimming pk.Data portion. This is known to prone to bugs, due to the complexity and number of the invariants assumed across netstack to maintain. In the new PacketHeader API, client will call Push()/Consume() method to construct/parse an outgoing/incoming packet. All invariants, such as slicing and trimming, are maintained by the API itself. NewPacketBuffer() is introduced to create new PacketBuffer. Zero value is no longer valid. PacketBuffer now assumes the packet is a concatenation of following portions: * LinkHeader * NetworkHeader * TransportHeader * Data Any of them could be empty, or zero-length. PiperOrigin-RevId: 326507688 --- pkg/sentry/socket/netfilter/tcp_matcher.go | 4 +- pkg/sentry/socket/netfilter/udp_matcher.go | 4 +- pkg/tcpip/buffer/view.go | 10 + pkg/tcpip/link/channel/channel.go | 4 +- pkg/tcpip/link/fdbased/BUILD | 1 + pkg/tcpip/link/fdbased/endpoint.go | 14 +- pkg/tcpip/link/fdbased/endpoint_test.go | 134 ++++--- pkg/tcpip/link/fdbased/mmap.go | 16 +- pkg/tcpip/link/fdbased/packet_dispatchers.go | 45 ++- pkg/tcpip/link/loopback/loopback.go | 21 +- pkg/tcpip/link/muxed/injectable_test.go | 25 +- pkg/tcpip/link/nested/nested_test.go | 4 +- pkg/tcpip/link/sharedmem/sharedmem.go | 25 +- pkg/tcpip/link/sharedmem/sharedmem_test.go | 135 +++---- pkg/tcpip/link/sharedmem/tx.go | 14 +- pkg/tcpip/link/sniffer/sniffer.go | 19 +- pkg/tcpip/link/tun/device.go | 23 +- pkg/tcpip/link/waitable/waitable_test.go | 12 +- pkg/tcpip/network/arp/arp.go | 26 +- pkg/tcpip/network/arp/arp_test.go | 8 +- pkg/tcpip/network/ip_test.go | 80 +++-- pkg/tcpip/network/ipv4/icmp.go | 38 +- pkg/tcpip/network/ipv4/ipv4.go | 188 +++++----- pkg/tcpip/network/ipv4/ipv4_test.go | 72 ++-- pkg/tcpip/network/ipv6/icmp.go | 55 ++- pkg/tcpip/network/ipv6/icmp_test.go | 46 +-- pkg/tcpip/network/ipv6/ipv6.go | 47 +-- pkg/tcpip/network/ipv6/ipv6_test.go | 16 +- pkg/tcpip/network/ipv6/ndp_test.go | 29 +- pkg/tcpip/stack/BUILD | 2 + pkg/tcpip/stack/conntrack.go | 31 +- pkg/tcpip/stack/forwarder_test.go | 65 ++-- pkg/tcpip/stack/headertype_string.go | 39 ++ pkg/tcpip/stack/iptables.go | 2 +- pkg/tcpip/stack/iptables_targets.go | 9 +- pkg/tcpip/stack/ndp.go | 32 +- pkg/tcpip/stack/ndp_test.go | 22 +- pkg/tcpip/stack/nic.go | 40 +-- pkg/tcpip/stack/nic_test.go | 4 +- pkg/tcpip/stack/packet_buffer.go | 260 ++++++++++++-- pkg/tcpip/stack/packet_buffer_test.go | 397 +++++++++++++++++++++ pkg/tcpip/stack/route.go | 5 +- pkg/tcpip/stack/stack_test.go | 61 ++-- pkg/tcpip/stack/transport_demuxer_test.go | 14 +- pkg/tcpip/stack/transport_test.go | 54 ++- .../tests/integration/multicast_broadcast_test.go | 18 +- pkg/tcpip/transport/icmp/endpoint.go | 39 +- pkg/tcpip/transport/packet/endpoint.go | 35 +- pkg/tcpip/transport/raw/endpoint.go | 30 +- pkg/tcpip/transport/tcp/connect.go | 30 +- pkg/tcpip/transport/tcp/protocol.go | 17 +- pkg/tcpip/transport/tcp/segment.go | 2 +- pkg/tcpip/transport/tcp/testing/context/context.go | 28 +- pkg/tcpip/transport/udp/endpoint.go | 26 +- pkg/tcpip/transport/udp/protocol.go | 57 ++- pkg/tcpip/transport/udp/udp_test.go | 58 ++- 56 files changed, 1568 insertions(+), 924 deletions(-) create mode 100644 pkg/tcpip/stack/headertype_string.go create mode 100644 pkg/tcpip/stack/packet_buffer_test.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/socket/netfilter/tcp_matcher.go b/pkg/sentry/socket/netfilter/tcp_matcher.go index 4f98ee2d5..0bfd6c1f4 100644 --- a/pkg/sentry/socket/netfilter/tcp_matcher.go +++ b/pkg/sentry/socket/netfilter/tcp_matcher.go @@ -97,7 +97,7 @@ func (*TCPMatcher) Name() string { // Match implements Matcher.Match. func (tm *TCPMatcher) Match(hook stack.Hook, pkt *stack.PacketBuffer, interfaceName string) (bool, bool) { - netHeader := header.IPv4(pkt.NetworkHeader) + netHeader := header.IPv4(pkt.NetworkHeader().View()) if netHeader.TransportProtocol() != header.TCPProtocolNumber { return false, false @@ -111,7 +111,7 @@ func (tm *TCPMatcher) Match(hook stack.Hook, pkt *stack.PacketBuffer, interfaceN return false, false } - tcpHeader := header.TCP(pkt.TransportHeader) + tcpHeader := header.TCP(pkt.TransportHeader().View()) if len(tcpHeader) < header.TCPMinimumSize { // There's no valid TCP header here, so we drop the packet immediately. return false, true diff --git a/pkg/sentry/socket/netfilter/udp_matcher.go b/pkg/sentry/socket/netfilter/udp_matcher.go index 3f20fc891..7ed05461d 100644 --- a/pkg/sentry/socket/netfilter/udp_matcher.go +++ b/pkg/sentry/socket/netfilter/udp_matcher.go @@ -94,7 +94,7 @@ func (*UDPMatcher) Name() string { // Match implements Matcher.Match. func (um *UDPMatcher) Match(hook stack.Hook, pkt *stack.PacketBuffer, interfaceName string) (bool, bool) { - netHeader := header.IPv4(pkt.NetworkHeader) + netHeader := header.IPv4(pkt.NetworkHeader().View()) // TODO(gvisor.dev/issue/170): Proto checks should ultimately be moved // into the stack.Check codepath as matchers are added. @@ -110,7 +110,7 @@ func (um *UDPMatcher) Match(hook stack.Hook, pkt *stack.PacketBuffer, interfaceN return false, false } - udpHeader := header.UDP(pkt.TransportHeader) + udpHeader := header.UDP(pkt.TransportHeader().View()) if len(udpHeader) < header.UDPMinimumSize { // There's no valid UDP header here, so we drop the packet immediately. return false, true diff --git a/pkg/tcpip/buffer/view.go b/pkg/tcpip/buffer/view.go index 9a3c5d6c3..ea0c5413d 100644 --- a/pkg/tcpip/buffer/view.go +++ b/pkg/tcpip/buffer/view.go @@ -65,6 +65,16 @@ func (v View) ToVectorisedView() VectorisedView { return NewVectorisedView(len(v), []View{v}) } +// IsEmpty returns whether v is of length zero. +func (v View) IsEmpty() bool { + return len(v) == 0 +} + +// Size returns the length of v. +func (v View) Size() int { + return len(v) +} + // VectorisedView is a vectorised version of View using non contiguous memory. // It supports all the convenience methods supported by View. // diff --git a/pkg/tcpip/link/channel/channel.go b/pkg/tcpip/link/channel/channel.go index e12a5929b..c95aef63c 100644 --- a/pkg/tcpip/link/channel/channel.go +++ b/pkg/tcpip/link/channel/channel.go @@ -274,7 +274,9 @@ func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. func (e *Endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { p := PacketInfo{ - Pkt: &stack.PacketBuffer{Data: vv}, + Pkt: stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: vv, + }), Proto: 0, GSO: nil, } diff --git a/pkg/tcpip/link/fdbased/BUILD b/pkg/tcpip/link/fdbased/BUILD index 507b44abc..10072eac1 100644 --- a/pkg/tcpip/link/fdbased/BUILD +++ b/pkg/tcpip/link/fdbased/BUILD @@ -37,5 +37,6 @@ go_test( "//pkg/tcpip/header", "//pkg/tcpip/link/rawfile", "//pkg/tcpip/stack", + "@com_github_google_go_cmp//cmp:go_default_library", ], ) diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go index c18bb91fb..975309fc8 100644 --- a/pkg/tcpip/link/fdbased/endpoint.go +++ b/pkg/tcpip/link/fdbased/endpoint.go @@ -390,8 +390,7 @@ const ( func (e *endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { if e.hdrSize > 0 { // Add ethernet header if needed. - eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize)) - pkt.LinkHeader = buffer.View(eth) + eth := header.Ethernet(pkt.LinkHeader().Push(header.EthernetMinimumSize)) ethHdr := &header.EthernetFields{ DstAddr: remote, Type: protocol, @@ -420,7 +419,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne if e.Capabilities()&stack.CapabilityHardwareGSO != 0 { vnetHdr := virtioNetHdr{} if gso != nil { - vnetHdr.hdrLen = uint16(pkt.Header.UsedLength()) + vnetHdr.hdrLen = uint16(pkt.HeaderSize()) if gso.NeedsCsum { vnetHdr.flags = _VIRTIO_NET_HDR_F_NEEDS_CSUM vnetHdr.csumStart = header.EthernetMinimumSize + gso.L3HdrLen @@ -443,11 +442,9 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne builder.Add(vnetHdrBuf) } - builder.Add(pkt.Header.View()) - for _, v := range pkt.Data.Views() { + for _, v := range pkt.Views() { builder.Add(v) } - return rawfile.NonBlockingWriteIovec(fd, builder.Build()) } @@ -463,7 +460,7 @@ func (e *endpoint) sendBatch(batchFD int, batch []*stack.PacketBuffer) (int, *tc if e.Capabilities()&stack.CapabilityHardwareGSO != 0 { vnetHdr := virtioNetHdr{} if pkt.GSOOptions != nil { - vnetHdr.hdrLen = uint16(pkt.Header.UsedLength()) + vnetHdr.hdrLen = uint16(pkt.HeaderSize()) if pkt.GSOOptions.NeedsCsum { vnetHdr.flags = _VIRTIO_NET_HDR_F_NEEDS_CSUM vnetHdr.csumStart = header.EthernetMinimumSize + pkt.GSOOptions.L3HdrLen @@ -486,8 +483,7 @@ func (e *endpoint) sendBatch(batchFD int, batch []*stack.PacketBuffer) (int, *tc var builder iovec.Builder builder.Add(vnetHdrBuf) - builder.Add(pkt.Header.View()) - for _, v := range pkt.Data.Views() { + for _, v := range pkt.Views() { builder.Add(v) } iovecs := builder.Build() diff --git a/pkg/tcpip/link/fdbased/endpoint_test.go b/pkg/tcpip/link/fdbased/endpoint_test.go index 7b995b85a..709f829c8 100644 --- a/pkg/tcpip/link/fdbased/endpoint_test.go +++ b/pkg/tcpip/link/fdbased/endpoint_test.go @@ -26,6 +26,7 @@ import ( "time" "unsafe" + "github.com/google/go-cmp/cmp" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" @@ -43,9 +44,36 @@ const ( ) type packetInfo struct { - raddr tcpip.LinkAddress - proto tcpip.NetworkProtocolNumber - contents *stack.PacketBuffer + Raddr tcpip.LinkAddress + Proto tcpip.NetworkProtocolNumber + Contents *stack.PacketBuffer +} + +type packetContents struct { + LinkHeader buffer.View + NetworkHeader buffer.View + TransportHeader buffer.View + Data buffer.View +} + +func checkPacketInfoEqual(t *testing.T, got, want packetInfo) { + t.Helper() + if diff := cmp.Diff( + want, got, + cmp.Transformer("ExtractPacketBuffer", func(pk *stack.PacketBuffer) *packetContents { + if pk == nil { + return nil + } + return &packetContents{ + LinkHeader: pk.LinkHeader().View(), + NetworkHeader: pk.NetworkHeader().View(), + TransportHeader: pk.TransportHeader().View(), + Data: pk.Data.ToView(), + } + }), + ); diff != "" { + t.Errorf("unexpected packetInfo (-want +got):\n%s", diff) + } } type context struct { @@ -159,19 +187,28 @@ func testWritePacket(t *testing.T, plen int, eth bool, gsoMaxSize uint32, hash u RemoteLinkAddress: raddr, } - // Build header. - hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength()) + 100) - b := hdr.Prepend(100) - for i := range b { - b[i] = uint8(rand.Intn(256)) + // Build payload. + payload := buffer.NewView(plen) + if _, err := rand.Read(payload); err != nil { + t.Fatalf("rand.Read(payload): %s", err) } - // Build payload and write. - payload := make(buffer.View, plen) - for i := range payload { - payload[i] = uint8(rand.Intn(256)) + // Build packet buffer. + const netHdrLen = 100 + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(c.ep.MaxHeaderLength()) + netHdrLen, + Data: payload.ToVectorisedView(), + }) + pkt.Hash = hash + + // Build header. + b := pkt.NetworkHeader().Push(netHdrLen) + if _, err := rand.Read(b); err != nil { + t.Fatalf("rand.Read(b): %s", err) } - want := append(hdr.View(), payload...) + + // Write. + want := append(append(buffer.View(nil), b...), payload...) var gso *stack.GSO if gsoMaxSize != 0 { gso = &stack.GSO{ @@ -183,11 +220,7 @@ func testWritePacket(t *testing.T, plen int, eth bool, gsoMaxSize uint32, hash u L3HdrLen: header.IPv4MaximumHeaderSize, } } - if err := c.ep.WritePacket(r, gso, proto, &stack.PacketBuffer{ - Header: hdr, - Data: payload.ToVectorisedView(), - Hash: hash, - }); err != nil { + if err := c.ep.WritePacket(r, gso, proto, pkt); err != nil { t.Fatalf("WritePacket failed: %v", err) } @@ -296,13 +329,14 @@ func TestPreserveSrcAddress(t *testing.T) { LocalLinkAddress: baddr, } - // WritePacket panics given a prependable with anything less than - // the minimum size of the ethernet header. - hdr := buffer.NewPrependable(header.EthernetMinimumSize) - if err := c.ep.WritePacket(r, nil /* gso */, proto, &stack.PacketBuffer{ - Header: hdr, - Data: buffer.VectorisedView{}, - }); err != nil { + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + // WritePacket panics given a prependable with anything less than + // the minimum size of the ethernet header. + // TODO(b/153685824): Figure out if this should use c.ep.MaxHeaderLength(). + ReserveHeaderBytes: header.EthernetMinimumSize, + Data: buffer.VectorisedView{}, + }) + if err := c.ep.WritePacket(r, nil /* gso */, proto, pkt); err != nil { t.Fatalf("WritePacket failed: %v", err) } @@ -331,24 +365,25 @@ func TestDeliverPacket(t *testing.T) { defer c.cleanup() // Build packet. - b := make([]byte, plen) - all := b - for i := range b { - b[i] = uint8(rand.Intn(256)) + all := make([]byte, plen) + if _, err := rand.Read(all); err != nil { + t.Fatalf("rand.Read(all): %s", err) } - - var hdr header.Ethernet - if !eth { - // So that it looks like an IPv4 packet. - b[0] = 0x40 - } else { - hdr = make(header.Ethernet, header.EthernetMinimumSize) + // Make it look like an IPv4 packet. + all[0] = 0x40 + + wantPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: header.EthernetMinimumSize, + Data: buffer.NewViewFromBytes(all).ToVectorisedView(), + }) + if eth { + hdr := header.Ethernet(wantPkt.LinkHeader().Push(header.EthernetMinimumSize)) hdr.Encode(&header.EthernetFields{ SrcAddr: raddr, DstAddr: laddr, Type: proto, }) - all = append(hdr, b...) + all = append(hdr, all...) } // Write packet via the file descriptor. @@ -360,24 +395,15 @@ func TestDeliverPacket(t *testing.T) { select { case pi := <-c.ch: want := packetInfo{ - raddr: raddr, - proto: proto, - contents: &stack.PacketBuffer{ - Data: buffer.View(b).ToVectorisedView(), - LinkHeader: buffer.View(hdr), - }, + Raddr: raddr, + Proto: proto, + Contents: wantPkt, } if !eth { - want.proto = header.IPv4ProtocolNumber - want.raddr = "" - } - // want.contents.Data will be a single - // view, so make pi do the same for the - // DeepEqual check. - pi.contents.Data = pi.contents.Data.ToView().ToVectorisedView() - if !reflect.DeepEqual(want, pi) { - t.Fatalf("Unexpected received packet: %+v, want %+v", pi, want) + want.Proto = header.IPv4ProtocolNumber + want.Raddr = "" } + checkPacketInfoEqual(t, pi, want) case <-time.After(10 * time.Second): t.Fatalf("Timed out waiting for packet") } @@ -572,8 +598,8 @@ func TestDispatchPacketFormat(t *testing.T) { t.Fatalf("len(sink.pkts) = %d, want %d", got, want) } pkt := sink.pkts[0] - if got, want := len(pkt.LinkHeader), header.EthernetMinimumSize; got != want { - t.Errorf("len(pkt.LinkHeader) = %d, want %d", got, want) + if got, want := pkt.LinkHeader().View().Size(), header.EthernetMinimumSize; got != want { + t.Errorf("pkt.LinkHeader().View().Size() = %d, want %d", got, want) } if got, want := pkt.Data.Size(), 4; got != want { t.Errorf("pkt.Data.Size() = %d, want %d", got, want) diff --git a/pkg/tcpip/link/fdbased/mmap.go b/pkg/tcpip/link/fdbased/mmap.go index 2dfd29aa9..c475dda20 100644 --- a/pkg/tcpip/link/fdbased/mmap.go +++ b/pkg/tcpip/link/fdbased/mmap.go @@ -18,6 +18,7 @@ package fdbased import ( "encoding/binary" + "fmt" "syscall" "golang.org/x/sys/unix" @@ -170,10 +171,9 @@ func (d *packetMMapDispatcher) dispatch() (bool, *tcpip.Error) { var ( p tcpip.NetworkProtocolNumber remote, local tcpip.LinkAddress - eth header.Ethernet ) if d.e.hdrSize > 0 { - eth = header.Ethernet(pkt) + eth := header.Ethernet(pkt) p = eth.Type() remote = eth.SourceAddress() local = eth.DestinationAddress() @@ -190,10 +190,14 @@ func (d *packetMMapDispatcher) dispatch() (bool, *tcpip.Error) { } } - pkt = pkt[d.e.hdrSize:] - d.e.dispatcher.DeliverNetworkPacket(remote, local, p, &stack.PacketBuffer{ - Data: buffer.View(pkt).ToVectorisedView(), - LinkHeader: buffer.View(eth), + pbuf := stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: buffer.View(pkt).ToVectorisedView(), }) + if d.e.hdrSize > 0 { + if _, ok := pbuf.LinkHeader().Consume(d.e.hdrSize); !ok { + panic(fmt.Sprintf("LinkHeader().Consume(%d) must succeed", d.e.hdrSize)) + } + } + d.e.dispatcher.DeliverNetworkPacket(remote, local, p, pbuf) return true, nil } diff --git a/pkg/tcpip/link/fdbased/packet_dispatchers.go b/pkg/tcpip/link/fdbased/packet_dispatchers.go index d8f2504b3..8c3ca86d6 100644 --- a/pkg/tcpip/link/fdbased/packet_dispatchers.go +++ b/pkg/tcpip/link/fdbased/packet_dispatchers.go @@ -103,7 +103,7 @@ func (d *readVDispatcher) dispatch() (bool, *tcpip.Error) { d.allocateViews(BufConfig) n, err := rawfile.BlockingReadv(d.fd, d.iovecs) - if err != nil { + if n == 0 || err != nil { return false, err } if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 { @@ -111,17 +111,22 @@ func (d *readVDispatcher) dispatch() (bool, *tcpip.Error) { // isn't used and it isn't in a view. n -= virtioNetHdrSize } - if n <= d.e.hdrSize { - return false, nil - } + + used := d.capViews(n, BufConfig) + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: buffer.NewVectorisedView(n, append([]buffer.View(nil), d.views[:used]...)), + }) var ( p tcpip.NetworkProtocolNumber remote, local tcpip.LinkAddress - eth header.Ethernet ) if d.e.hdrSize > 0 { - eth = header.Ethernet(d.views[0][:header.EthernetMinimumSize]) + hdr, ok := pkt.LinkHeader().Consume(d.e.hdrSize) + if !ok { + return false, nil + } + eth := header.Ethernet(hdr) p = eth.Type() remote = eth.SourceAddress() local = eth.DestinationAddress() @@ -138,13 +143,6 @@ func (d *readVDispatcher) dispatch() (bool, *tcpip.Error) { } } - used := d.capViews(n, BufConfig) - pkt := &stack.PacketBuffer{ - Data: buffer.NewVectorisedView(n, append([]buffer.View(nil), d.views[:used]...)), - LinkHeader: buffer.View(eth), - } - pkt.Data.TrimFront(d.e.hdrSize) - d.e.dispatcher.DeliverNetworkPacket(remote, local, p, pkt) // Prepare e.views for another packet: release used views. @@ -268,17 +266,22 @@ func (d *recvMMsgDispatcher) dispatch() (bool, *tcpip.Error) { if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 { n -= virtioNetHdrSize } - if n <= d.e.hdrSize { - return false, nil - } + + used := d.capViews(k, int(n), BufConfig) + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: buffer.NewVectorisedView(int(n), append([]buffer.View(nil), d.views[k][:used]...)), + }) var ( p tcpip.NetworkProtocolNumber remote, local tcpip.LinkAddress - eth header.Ethernet ) if d.e.hdrSize > 0 { - eth = header.Ethernet(d.views[k][0][:header.EthernetMinimumSize]) + hdr, ok := pkt.LinkHeader().Consume(d.e.hdrSize) + if !ok { + return false, nil + } + eth := header.Ethernet(hdr) p = eth.Type() remote = eth.SourceAddress() local = eth.DestinationAddress() @@ -295,12 +298,6 @@ func (d *recvMMsgDispatcher) dispatch() (bool, *tcpip.Error) { } } - used := d.capViews(k, int(n), BufConfig) - pkt := &stack.PacketBuffer{ - Data: buffer.NewVectorisedView(int(n), append([]buffer.View(nil), d.views[k][:used]...)), - LinkHeader: buffer.View(eth), - } - pkt.Data.TrimFront(d.e.hdrSize) d.e.dispatcher.DeliverNetworkPacket(remote, local, p, pkt) // Prepare e.views for another packet: release used views. diff --git a/pkg/tcpip/link/loopback/loopback.go b/pkg/tcpip/link/loopback/loopback.go index 781cdd317..38aa694e4 100644 --- a/pkg/tcpip/link/loopback/loopback.go +++ b/pkg/tcpip/link/loopback/loopback.go @@ -77,16 +77,16 @@ func (*endpoint) Wait() {} // WritePacket implements stack.LinkEndpoint.WritePacket. It delivers outbound // packets to the network-layer dispatcher. func (e *endpoint) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { - views := make([]buffer.View, 1, 1+len(pkt.Data.Views())) - views[0] = pkt.Header.View() - views = append(views, pkt.Data.Views()...) + // Construct data as the unparsed portion for the loopback packet. + data := buffer.NewVectorisedView(pkt.Size(), pkt.Views()) // Because we're immediately turning around and writing the packet back // to the rx path, we intentionally don't preserve the remote and local // link addresses from the stack.Route we're passed. - e.dispatcher.DeliverNetworkPacket("" /* remote */, "" /* local */, protocol, &stack.PacketBuffer{ - Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views), + newPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: data, }) + e.dispatcher.DeliverNetworkPacket("" /* remote */, "" /* local */, protocol, newPkt) return nil } @@ -98,18 +98,17 @@ func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, stack.PacketBufferList // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: vv, + }) // There should be an ethernet header at the beginning of vv. - hdr, ok := vv.PullUp(header.EthernetMinimumSize) + hdr, ok := pkt.LinkHeader().Consume(header.EthernetMinimumSize) if !ok { // Reject the packet if it's shorter than an ethernet header. return tcpip.ErrBadAddress } linkHeader := header.Ethernet(hdr) - vv.TrimFront(len(linkHeader)) - e.dispatcher.DeliverNetworkPacket("" /* remote */, "" /* local */, linkHeader.Type(), &stack.PacketBuffer{ - Data: vv, - LinkHeader: buffer.View(linkHeader), - }) + e.dispatcher.DeliverNetworkPacket("" /* remote */, "" /* local */, linkHeader.Type(), pkt) return nil } diff --git a/pkg/tcpip/link/muxed/injectable_test.go b/pkg/tcpip/link/muxed/injectable_test.go index 0744f66d6..3e4afcdad 100644 --- a/pkg/tcpip/link/muxed/injectable_test.go +++ b/pkg/tcpip/link/muxed/injectable_test.go @@ -46,14 +46,14 @@ func TestInjectableEndpointRawDispatch(t *testing.T) { func TestInjectableEndpointDispatch(t *testing.T) { endpoint, sock, dstIP := makeTestInjectableEndpoint(t) - hdr := buffer.NewPrependable(1) - hdr.Prepend(1)[0] = 0xFA + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: 1, + Data: buffer.NewViewFromBytes([]byte{0xFB}).ToVectorisedView(), + }) + pkt.TransportHeader().Push(1)[0] = 0xFA packetRoute := stack.Route{RemoteAddress: dstIP} - endpoint.WritePacket(&packetRoute, nil /* gso */, ipv4.ProtocolNumber, &stack.PacketBuffer{ - Header: hdr, - Data: buffer.NewViewFromBytes([]byte{0xFB}).ToVectorisedView(), - }) + endpoint.WritePacket(&packetRoute, nil /* gso */, ipv4.ProtocolNumber, pkt) buf := make([]byte, 6500) bytesRead, err := sock.Read(buf) @@ -67,13 +67,14 @@ func TestInjectableEndpointDispatch(t *testing.T) { func TestInjectableEndpointDispatchHdrOnly(t *testing.T) { endpoint, sock, dstIP := makeTestInjectableEndpoint(t) - hdr := buffer.NewPrependable(1) - hdr.Prepend(1)[0] = 0xFA - packetRoute := stack.Route{RemoteAddress: dstIP} - endpoint.WritePacket(&packetRoute, nil /* gso */, ipv4.ProtocolNumber, &stack.PacketBuffer{ - Header: hdr, - Data: buffer.NewView(0).ToVectorisedView(), + + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: 1, + Data: buffer.NewView(0).ToVectorisedView(), }) + pkt.TransportHeader().Push(1)[0] = 0xFA + packetRoute := stack.Route{RemoteAddress: dstIP} + endpoint.WritePacket(&packetRoute, nil /* gso */, ipv4.ProtocolNumber, pkt) buf := make([]byte, 6500) bytesRead, err := sock.Read(buf) if err != nil { diff --git a/pkg/tcpip/link/nested/nested_test.go b/pkg/tcpip/link/nested/nested_test.go index 7d9249c1c..c1f9d308c 100644 --- a/pkg/tcpip/link/nested/nested_test.go +++ b/pkg/tcpip/link/nested/nested_test.go @@ -87,7 +87,7 @@ func TestNestedLinkEndpoint(t *testing.T) { t.Error("After attach, nestedEP.IsAttached() = false, want = true") } - nestedEP.DeliverNetworkPacket(emptyAddress, emptyAddress, header.IPv4ProtocolNumber, &stack.PacketBuffer{}) + nestedEP.DeliverNetworkPacket(emptyAddress, emptyAddress, header.IPv4ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{})) if disp.count != 1 { t.Errorf("After first packet with dispatcher attached, got disp.count = %d, want = 1", disp.count) } @@ -101,7 +101,7 @@ func TestNestedLinkEndpoint(t *testing.T) { } disp.count = 0 - nestedEP.DeliverNetworkPacket(emptyAddress, emptyAddress, header.IPv4ProtocolNumber, &stack.PacketBuffer{}) + nestedEP.DeliverNetworkPacket(emptyAddress, emptyAddress, header.IPv4ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{})) if disp.count != 0 { t.Errorf("After second packet with dispatcher detached, got disp.count = %d, want = 0", disp.count) } diff --git a/pkg/tcpip/link/sharedmem/sharedmem.go b/pkg/tcpip/link/sharedmem/sharedmem.go index 507c76b76..7fb8a6c49 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem.go +++ b/pkg/tcpip/link/sharedmem/sharedmem.go @@ -186,8 +186,7 @@ func (e *endpoint) LinkAddress() tcpip.LinkAddress { // AddHeader implements stack.LinkEndpoint.AddHeader. func (e *endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { // Add ethernet header if needed. - eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize)) - pkt.LinkHeader = buffer.View(eth) + eth := header.Ethernet(pkt.LinkHeader().Push(header.EthernetMinimumSize)) ethHdr := &header.EthernetFields{ DstAddr: remote, Type: protocol, @@ -207,10 +206,10 @@ func (e *endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.Net func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { e.AddHeader(r.LocalLinkAddress, r.RemoteLinkAddress, protocol, pkt) - v := pkt.Data.ToView() + views := pkt.Views() // Transmit the packet. e.mu.Lock() - ok := e.tx.transmit(pkt.Header.View(), v) + ok := e.tx.transmit(views...) e.mu.Unlock() if !ok { @@ -227,10 +226,10 @@ func (e *endpoint) WritePackets(r *stack.Route, _ *stack.GSO, pkts stack.PacketB // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { - v := vv.ToView() + views := vv.Views() // Transmit the packet. e.mu.Lock() - ok := e.tx.transmit(v, buffer.View{}) + ok := e.tx.transmit(views...) e.mu.Unlock() if !ok { @@ -276,16 +275,18 @@ func (e *endpoint) dispatchLoop(d stack.NetworkDispatcher) { rxb[i].Size = e.bufferSize } - if n < header.EthernetMinimumSize { + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: buffer.View(b).ToVectorisedView(), + }) + + hdr, ok := pkt.LinkHeader().Consume(header.EthernetMinimumSize) + if !ok { continue } + eth := header.Ethernet(hdr) // Send packet up the stack. - eth := header.Ethernet(b[:header.EthernetMinimumSize]) - d.DeliverNetworkPacket(eth.SourceAddress(), eth.DestinationAddress(), eth.Type(), &stack.PacketBuffer{ - Data: buffer.View(b[header.EthernetMinimumSize:]).ToVectorisedView(), - LinkHeader: buffer.View(eth), - }) + d.DeliverNetworkPacket(eth.SourceAddress(), eth.DestinationAddress(), eth.Type(), pkt) } // Clean state. diff --git a/pkg/tcpip/link/sharedmem/sharedmem_test.go b/pkg/tcpip/link/sharedmem/sharedmem_test.go index 8f3cd9449..22d5c97f1 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem_test.go +++ b/pkg/tcpip/link/sharedmem/sharedmem_test.go @@ -266,21 +266,23 @@ func TestSimpleSend(t *testing.T) { for iters := 1000; iters > 0; iters-- { func() { + hdrLen, dataLen := rand.Intn(10000), rand.Intn(10000) + // Prepare and send packet. - n := rand.Intn(10000) - hdr := buffer.NewPrependable(n + int(c.ep.MaxHeaderLength())) - hdrBuf := hdr.Prepend(n) + hdrBuf := buffer.NewView(hdrLen) randomFill(hdrBuf) - n = rand.Intn(10000) - buf := buffer.NewView(n) - randomFill(buf) + data := buffer.NewView(dataLen) + randomFill(data) + + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: hdrLen + int(c.ep.MaxHeaderLength()), + Data: data.ToVectorisedView(), + }) + copy(pkt.NetworkHeader().Push(hdrLen), hdrBuf) proto := tcpip.NetworkProtocolNumber(rand.Intn(0x10000)) - if err := c.ep.WritePacket(&r, nil /* gso */, proto, &stack.PacketBuffer{ - Header: hdr, - Data: buf.ToVectorisedView(), - }); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, proto, pkt); err != nil { t.Fatalf("WritePacket failed: %v", err) } @@ -317,7 +319,7 @@ func TestSimpleSend(t *testing.T) { // Compare contents skipping the ethernet header added by the // endpoint. - merged := append(hdrBuf, buf...) + merged := append(hdrBuf, data...) if uint32(len(contents)) < pi.Size { t.Fatalf("Sum of buffers is less than packet size: %v < %v", len(contents), pi.Size) } @@ -344,14 +346,14 @@ func TestPreserveSrcAddressInSend(t *testing.T) { LocalLinkAddress: newLocalLinkAddress, } - // WritePacket panics given a prependable with anything less than - // the minimum size of the ethernet header. - hdr := buffer.NewPrependable(header.EthernetMinimumSize) + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + // WritePacket panics given a prependable with anything less than + // the minimum size of the ethernet header. + ReserveHeaderBytes: header.EthernetMinimumSize, + }) proto := tcpip.NetworkProtocolNumber(rand.Intn(0x10000)) - if err := c.ep.WritePacket(&r, nil /* gso */, proto, &stack.PacketBuffer{ - Header: hdr, - }); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, proto, pkt); err != nil { t.Fatalf("WritePacket failed: %v", err) } @@ -403,12 +405,12 @@ func TestFillTxQueue(t *testing.T) { // until the tx queue if full. ids := make(map[uint64]struct{}) for i := queuePipeSize / 40; i > 0; i-- { - hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(c.ep.MaxHeaderLength()), + Data: buf.ToVectorisedView(), + }) - if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, &stack.PacketBuffer{ - Header: hdr, - Data: buf.ToVectorisedView(), - }); err != nil { + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, pkt); err != nil { t.Fatalf("WritePacket failed unexpectedly: %v", err) } @@ -422,11 +424,11 @@ func TestFillTxQueue(t *testing.T) { } // Next attempt to write must fail. - hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, &stack.PacketBuffer{ - Header: hdr, - Data: buf.ToVectorisedView(), - }); err != want { + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(c.ep.MaxHeaderLength()), + Data: buf.ToVectorisedView(), + }) + if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, pkt); err != want { t.Fatalf("WritePacket return unexpected result: got %v, want %v", err, want) } } @@ -450,11 +452,11 @@ func TestFillTxQueueAfterBadCompletion(t *testing.T) { // Send two packets so that the id slice has at least two slots. for i := 2; i > 0; i-- { - hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, &stack.PacketBuffer{ - Header: hdr, - Data: buf.ToVectorisedView(), - }); err != nil { + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(c.ep.MaxHeaderLength()), + Data: buf.ToVectorisedView(), + }) + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, pkt); err != nil { t.Fatalf("WritePacket failed unexpectedly: %v", err) } } @@ -473,11 +475,11 @@ func TestFillTxQueueAfterBadCompletion(t *testing.T) { // until the tx queue if full. ids := make(map[uint64]struct{}) for i := queuePipeSize / 40; i > 0; i-- { - hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, &stack.PacketBuffer{ - Header: hdr, - Data: buf.ToVectorisedView(), - }); err != nil { + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(c.ep.MaxHeaderLength()), + Data: buf.ToVectorisedView(), + }) + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, pkt); err != nil { t.Fatalf("WritePacket failed unexpectedly: %v", err) } @@ -491,11 +493,11 @@ func TestFillTxQueueAfterBadCompletion(t *testing.T) { } // Next attempt to write must fail. - hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, &stack.PacketBuffer{ - Header: hdr, - Data: buf.ToVectorisedView(), - }); err != want { + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(c.ep.MaxHeaderLength()), + Data: buf.ToVectorisedView(), + }) + if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, pkt); err != want { t.Fatalf("WritePacket return unexpected result: got %v, want %v", err, want) } } @@ -517,11 +519,11 @@ func TestFillTxMemory(t *testing.T) { // we fill the memory. ids := make(map[uint64]struct{}) for i := queueDataSize / bufferSize; i > 0; i-- { - hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, &stack.PacketBuffer{ - Header: hdr, - Data: buf.ToVectorisedView(), - }); err != nil { + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(c.ep.MaxHeaderLength()), + Data: buf.ToVectorisedView(), + }) + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, pkt); err != nil { t.Fatalf("WritePacket failed unexpectedly: %v", err) } @@ -536,11 +538,11 @@ func TestFillTxMemory(t *testing.T) { } // Next attempt to write must fail. - hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, &stack.PacketBuffer{ - Header: hdr, - Data: buf.ToVectorisedView(), + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(c.ep.MaxHeaderLength()), + Data: buf.ToVectorisedView(), }) + err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, pkt) if want := tcpip.ErrWouldBlock; err != want { t.Fatalf("WritePacket return unexpected result: got %v, want %v", err, want) } @@ -564,11 +566,11 @@ func TestFillTxMemoryWithMultiBuffer(t *testing.T) { // Each packet is uses up one buffer, so write as many as possible // until there is only one buffer left. for i := queueDataSize/bufferSize - 1; i > 0; i-- { - hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, &stack.PacketBuffer{ - Header: hdr, - Data: buf.ToVectorisedView(), - }); err != nil { + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(c.ep.MaxHeaderLength()), + Data: buf.ToVectorisedView(), + }) + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, pkt); err != nil { t.Fatalf("WritePacket failed unexpectedly: %v", err) } @@ -579,23 +581,22 @@ func TestFillTxMemoryWithMultiBuffer(t *testing.T) { // Attempt to write a two-buffer packet. It must fail. { - hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - uu := buffer.NewView(bufferSize).ToVectorisedView() - if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, &stack.PacketBuffer{ - Header: hdr, - Data: uu, - }); err != want { + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(c.ep.MaxHeaderLength()), + Data: buffer.NewView(bufferSize).ToVectorisedView(), + }) + if want, err := tcpip.ErrWouldBlock, c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, pkt); err != want { t.Fatalf("WritePacket return unexpected result: got %v, want %v", err, want) } } // Attempt to write the one-buffer packet again. It must succeed. { - hdr := buffer.NewPrependable(int(c.ep.MaxHeaderLength())) - if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, &stack.PacketBuffer{ - Header: hdr, - Data: buf.ToVectorisedView(), - }); err != nil { + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(c.ep.MaxHeaderLength()), + Data: buf.ToVectorisedView(), + }) + if err := c.ep.WritePacket(&r, nil /* gso */, header.IPv4ProtocolNumber, pkt); err != nil { t.Fatalf("WritePacket failed unexpectedly: %v", err) } } diff --git a/pkg/tcpip/link/sharedmem/tx.go b/pkg/tcpip/link/sharedmem/tx.go index 6b8d7859d..44f421c2d 100644 --- a/pkg/tcpip/link/sharedmem/tx.go +++ b/pkg/tcpip/link/sharedmem/tx.go @@ -18,6 +18,7 @@ import ( "math" "syscall" + "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/link/sharedmem/queue" ) @@ -76,9 +77,9 @@ func (t *tx) cleanup() { syscall.Munmap(t.data) } -// transmit sends a packet made up of up to two buffers. Returns a boolean that -// specifies whether the packet was successfully transmitted. -func (t *tx) transmit(a, b []byte) bool { +// transmit sends a packet made of bufs. Returns a boolean that specifies +// whether the packet was successfully transmitted. +func (t *tx) transmit(bufs ...buffer.View) bool { // Pull completions from the tx queue and add their buffers back to the // pool so that we can reuse them. for { @@ -93,7 +94,10 @@ func (t *tx) transmit(a, b []byte) bool { } bSize := t.bufs.entrySize - total := uint32(len(a) + len(b)) + total := uint32(0) + for _, data := range bufs { + total += uint32(len(data)) + } bufCount := (total + bSize - 1) / bSize // Allocate enough buffers to hold all the data. @@ -115,7 +119,7 @@ func (t *tx) transmit(a, b []byte) bool { // Copy data into allocated buffers. nBuf := buf var dBuf []byte - for _, data := range [][]byte{a, b} { + for _, data := range bufs { for len(data) > 0 { if len(dBuf) == 0 { dBuf = t.data[nBuf.Offset:][:nBuf.Size] diff --git a/pkg/tcpip/link/sniffer/sniffer.go b/pkg/tcpip/link/sniffer/sniffer.go index 509076643..4fb127978 100644 --- a/pkg/tcpip/link/sniffer/sniffer.go +++ b/pkg/tcpip/link/sniffer/sniffer.go @@ -134,7 +134,7 @@ func (e *endpoint) dumpPacket(prefix string, gso *stack.GSO, protocol tcpip.Netw logPacket(prefix, protocol, pkt, gso) } if writer != nil && atomic.LoadUint32(&LogPacketsToPCAP) == 1 { - totalLength := pkt.Header.UsedLength() + pkt.Data.Size() + totalLength := pkt.Size() length := totalLength if max := int(e.maxPCAPLen); length > max { length = max @@ -155,12 +155,11 @@ func (e *endpoint) dumpPacket(prefix string, gso *stack.GSO, protocol tcpip.Netw length -= n } } - write(pkt.Header.View()) - for _, view := range pkt.Data.Views() { + for _, v := range pkt.Views() { if length == 0 { break } - write(view) + write(v) } } } @@ -185,9 +184,9 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { - e.dumpPacket("send", nil, 0, &stack.PacketBuffer{ + e.dumpPacket("send", nil, 0, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: vv, - }) + })) return e.Endpoint.WriteRawPacket(vv) } @@ -201,12 +200,8 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.P var fragmentOffset uint16 var moreFragments bool - // Create a clone of pkt, including any headers if present. Avoid allocating - // backing memory for the clone. - views := [8]buffer.View{} - vv := buffer.NewVectorisedView(0, views[:0]) - vv.AppendView(pkt.Header.View()) - vv.Append(pkt.Data) + // Examine the packet using a new VV. Backing storage must not be written. + vv := buffer.NewVectorisedView(pkt.Size(), pkt.Views()) switch protocol { case header.IPv4ProtocolNumber: diff --git a/pkg/tcpip/link/tun/device.go b/pkg/tcpip/link/tun/device.go index 22b0a12bd..3b1510a33 100644 --- a/pkg/tcpip/link/tun/device.go +++ b/pkg/tcpip/link/tun/device.go @@ -215,12 +215,11 @@ func (d *Device) Write(data []byte) (int64, error) { remote = tcpip.LinkAddress(zeroMAC[:]) } - pkt := &stack.PacketBuffer{ - Data: buffer.View(data).ToVectorisedView(), - } - if ethHdr != nil { - pkt.LinkHeader = buffer.View(ethHdr) - } + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: len(ethHdr), + Data: buffer.View(data).ToVectorisedView(), + }) + copy(pkt.LinkHeader().Push(len(ethHdr)), ethHdr) endpoint.InjectLinkAddr(protocol, remote, pkt) return dataLen, nil } @@ -265,21 +264,22 @@ func (d *Device) encodePkt(info *channel.PacketInfo) (buffer.View, bool) { // If the packet does not already have link layer header, and the route // does not exist, we can't compute it. This is possibly a raw packet, tun // device doesn't support this at the moment. - if info.Pkt.LinkHeader == nil && info.Route.RemoteLinkAddress == "" { + if info.Pkt.LinkHeader().View().IsEmpty() && info.Route.RemoteLinkAddress == "" { return nil, false } // Ethernet header (TAP only). if d.hasFlags(linux.IFF_TAP) { // Add ethernet header if not provided. - if info.Pkt.LinkHeader == nil { + if info.Pkt.LinkHeader().View().IsEmpty() { d.endpoint.AddHeader(info.Route.LocalLinkAddress, info.Route.RemoteLinkAddress, info.Proto, info.Pkt) } - vv.AppendView(info.Pkt.LinkHeader) + vv.AppendView(info.Pkt.LinkHeader().View()) } // Append upper headers. - vv.AppendView(buffer.View(info.Pkt.Header.View()[len(info.Pkt.LinkHeader):])) + vv.AppendView(info.Pkt.NetworkHeader().View()) + vv.AppendView(info.Pkt.TransportHeader().View()) // Append data payload. vv.Append(info.Pkt.Data) @@ -361,8 +361,7 @@ func (e *tunEndpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip. if !e.isTap { return } - eth := header.Ethernet(pkt.Header.Prepend(header.EthernetMinimumSize)) - pkt.LinkHeader = buffer.View(eth) + eth := header.Ethernet(pkt.LinkHeader().Push(header.EthernetMinimumSize)) hdr := &header.EthernetFields{ SrcAddr: local, DstAddr: remote, diff --git a/pkg/tcpip/link/waitable/waitable_test.go b/pkg/tcpip/link/waitable/waitable_test.go index c448a888f..94827fc56 100644 --- a/pkg/tcpip/link/waitable/waitable_test.go +++ b/pkg/tcpip/link/waitable/waitable_test.go @@ -104,21 +104,21 @@ func TestWaitWrite(t *testing.T) { wep := New(ep) // Write and check that it goes through. - wep.WritePacket(nil, nil /* gso */, 0, &stack.PacketBuffer{}) + wep.WritePacket(nil, nil /* gso */, 0, stack.NewPacketBuffer(stack.PacketBufferOptions{})) if want := 1; ep.writeCount != want { t.Fatalf("Unexpected writeCount: got=%v, want=%v", ep.writeCount, want) } // Wait on dispatches, then try to write. It must go through. wep.WaitDispatch() - wep.WritePacket(nil, nil /* gso */, 0, &stack.PacketBuffer{}) + wep.WritePacket(nil, nil /* gso */, 0, stack.NewPacketBuffer(stack.PacketBufferOptions{})) if want := 2; ep.writeCount != want { t.Fatalf("Unexpected writeCount: got=%v, want=%v", ep.writeCount, want) } // Wait on writes, then try to write. It must not go through. wep.WaitWrite() - wep.WritePacket(nil, nil /* gso */, 0, &stack.PacketBuffer{}) + wep.WritePacket(nil, nil /* gso */, 0, stack.NewPacketBuffer(stack.PacketBufferOptions{})) if want := 2; ep.writeCount != want { t.Fatalf("Unexpected writeCount: got=%v, want=%v", ep.writeCount, want) } @@ -135,21 +135,21 @@ func TestWaitDispatch(t *testing.T) { } // Dispatch and check that it goes through. - ep.dispatcher.DeliverNetworkPacket("", "", 0, &stack.PacketBuffer{}) + ep.dispatcher.DeliverNetworkPacket("", "", 0, stack.NewPacketBuffer(stack.PacketBufferOptions{})) if want := 1; ep.dispatchCount != want { t.Fatalf("Unexpected dispatchCount: got=%v, want=%v", ep.dispatchCount, want) } // Wait on writes, then try to dispatch. It must go through. wep.WaitWrite() - ep.dispatcher.DeliverNetworkPacket("", "", 0, &stack.PacketBuffer{}) + ep.dispatcher.DeliverNetworkPacket("", "", 0, stack.NewPacketBuffer(stack.PacketBufferOptions{})) if want := 2; ep.dispatchCount != want { t.Fatalf("Unexpected dispatchCount: got=%v, want=%v", ep.dispatchCount, want) } // Wait on dispatches, then try to dispatch. It must not go through. wep.WaitDispatch() - ep.dispatcher.DeliverNetworkPacket("", "", 0, &stack.PacketBuffer{}) + ep.dispatcher.DeliverNetworkPacket("", "", 0, stack.NewPacketBuffer(stack.PacketBufferOptions{})) if want := 2; ep.dispatchCount != want { t.Fatalf("Unexpected dispatchCount: got=%v, want=%v", ep.dispatchCount, want) } diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index 31a242482..1ad788a17 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -99,7 +99,7 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBu } func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { - h := header.ARP(pkt.NetworkHeader) + h := header.ARP(pkt.NetworkHeader().View()) if !h.IsValid() { return } @@ -110,17 +110,17 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { if e.linkAddrCache.CheckLocalAddress(e.nicID, header.IPv4ProtocolNumber, localAddr) == 0 { return // we have no useful answer, ignore the request } - hdr := buffer.NewPrependable(int(e.linkEP.MaxHeaderLength()) + header.ARPSize) - packet := header.ARP(hdr.Prepend(header.ARPSize)) + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(e.linkEP.MaxHeaderLength()) + header.ARPSize, + }) + packet := header.ARP(pkt.NetworkHeader().Push(header.ARPSize)) packet.SetIPv4OverEthernet() packet.SetOp(header.ARPReply) copy(packet.HardwareAddressSender(), r.LocalLinkAddress[:]) copy(packet.ProtocolAddressSender(), h.ProtocolAddressTarget()) copy(packet.HardwareAddressTarget(), h.HardwareAddressSender()) copy(packet.ProtocolAddressTarget(), h.ProtocolAddressSender()) - e.linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, &stack.PacketBuffer{ - Header: hdr, - }) + _ = e.linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, pkt) fallthrough // also fill the cache from requests case header.ARPReply: addr := tcpip.Address(h.ProtocolAddressSender()) @@ -168,17 +168,17 @@ func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, remoteLinkAdd r.RemoteLinkAddress = header.EthernetBroadcastAddress } - hdr := buffer.NewPrependable(int(linkEP.MaxHeaderLength()) + header.ARPSize) - h := header.ARP(hdr.Prepend(header.ARPSize)) + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(linkEP.MaxHeaderLength()) + header.ARPSize, + }) + h := header.ARP(pkt.NetworkHeader().Push(header.ARPSize)) h.SetIPv4OverEthernet() h.SetOp(header.ARPRequest) copy(h.HardwareAddressSender(), linkEP.LinkAddress()) copy(h.ProtocolAddressSender(), localAddr) copy(h.ProtocolAddressTarget(), addr) - return linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, &stack.PacketBuffer{ - Header: hdr, - }) + return linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, pkt) } // ResolveStaticAddress implements stack.LinkAddressResolver.ResolveStaticAddress. @@ -210,12 +210,10 @@ func (*protocol) Wait() {} // Parse implements stack.NetworkProtocol.Parse. func (*protocol) Parse(pkt *stack.PacketBuffer) (proto tcpip.TransportProtocolNumber, hasTransportHdr bool, ok bool) { - hdr, ok := pkt.Data.PullUp(header.ARPSize) + _, ok = pkt.NetworkHeader().Consume(header.ARPSize) if !ok { return 0, false, false } - pkt.NetworkHeader = hdr - pkt.Data.TrimFront(header.ARPSize) return 0, false, true } diff --git a/pkg/tcpip/network/arp/arp_test.go b/pkg/tcpip/network/arp/arp_test.go index a35a64a0f..c2c3e6891 100644 --- a/pkg/tcpip/network/arp/arp_test.go +++ b/pkg/tcpip/network/arp/arp_test.go @@ -106,9 +106,9 @@ func TestDirectRequest(t *testing.T) { inject := func(addr tcpip.Address) { copy(h.ProtocolAddressTarget(), addr) - c.linkEP.InjectInbound(arp.ProtocolNumber, &stack.PacketBuffer{ + c.linkEP.InjectInbound(arp.ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: v.ToVectorisedView(), - }) + })) } for i, address := range []tcpip.Address{stackAddr1, stackAddr2} { @@ -118,9 +118,9 @@ func TestDirectRequest(t *testing.T) { if pi.Proto != arp.ProtocolNumber { t.Fatalf("expected ARP response, got network protocol number %d", pi.Proto) } - rep := header.ARP(pi.Pkt.Header.View()) + rep := header.ARP(pi.Pkt.NetworkHeader().View()) if !rep.IsValid() { - t.Fatalf("invalid ARP response pi.Pkt.Header.UsedLength()=%d", pi.Pkt.Header.UsedLength()) + t.Fatalf("invalid ARP response: len = %d; response = %x", len(rep), rep) } if got, want := tcpip.LinkAddress(rep.HardwareAddressSender()), stackLinkAddr1; got != want { t.Errorf("got HardwareAddressSender = %s, want = %s", got, want) diff --git a/pkg/tcpip/network/ip_test.go b/pkg/tcpip/network/ip_test.go index 615bae648..e6768258a 100644 --- a/pkg/tcpip/network/ip_test.go +++ b/pkg/tcpip/network/ip_test.go @@ -156,13 +156,13 @@ func (t *testObject) WritePacket(_ *stack.Route, _ *stack.GSO, protocol tcpip.Ne var dstAddr tcpip.Address if t.v4 { - h := header.IPv4(pkt.Header.View()) + h := header.IPv4(pkt.NetworkHeader().View()) prot = tcpip.TransportProtocolNumber(h.Protocol()) srcAddr = h.SourceAddress() dstAddr = h.DestinationAddress() } else { - h := header.IPv6(pkt.Header.View()) + h := header.IPv6(pkt.NetworkHeader().View()) prot = tcpip.TransportProtocolNumber(h.NextHeader()) srcAddr = h.SourceAddress() dstAddr = h.DestinationAddress() @@ -243,8 +243,11 @@ func TestIPv4Send(t *testing.T) { payload[i] = uint8(i) } - // Allocate the header buffer. - hdr := buffer.NewPrependable(int(ep.MaxHeaderLength())) + // Setup the packet buffer. + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(ep.MaxHeaderLength()), + Data: payload.ToVectorisedView(), + }) // Issue the write. o.protocol = 123 @@ -260,10 +263,7 @@ func TestIPv4Send(t *testing.T) { Protocol: 123, TTL: 123, TOS: stack.DefaultTOS, - }, &stack.PacketBuffer{ - Header: hdr, - Data: payload.ToVectorisedView(), - }); err != nil { + }, pkt); err != nil { t.Fatalf("WritePacket failed: %v", err) } } @@ -303,9 +303,13 @@ func TestIPv4Receive(t *testing.T) { if err != nil { t.Fatalf("could not find route: %v", err) } - pkt := stack.PacketBuffer{Data: view.ToVectorisedView()} - proto.Parse(&pkt) - ep.HandlePacket(&r, &pkt) + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: view.ToVectorisedView(), + }) + if _, _, ok := proto.Parse(pkt); !ok { + t.Fatalf("failed to parse packet: %x", pkt.Data.ToView()) + } + ep.HandlePacket(&r, pkt) if o.dataCalls != 1 { t.Fatalf("Bad number of data calls: got %x, want 1", o.dataCalls) } @@ -455,17 +459,25 @@ func TestIPv4FragmentationReceive(t *testing.T) { } // Send first segment. - pkt := stack.PacketBuffer{Data: frag1.ToVectorisedView()} - proto.Parse(&pkt) - ep.HandlePacket(&r, &pkt) + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: frag1.ToVectorisedView(), + }) + if _, _, ok := proto.Parse(pkt); !ok { + t.Fatalf("failed to parse packet: %x", pkt.Data.ToView()) + } + ep.HandlePacket(&r, pkt) if o.dataCalls != 0 { t.Fatalf("Bad number of data calls: got %x, want 0", o.dataCalls) } // Send second segment. - pkt = stack.PacketBuffer{Data: frag2.ToVectorisedView()} - proto.Parse(&pkt) - ep.HandlePacket(&r, &pkt) + pkt = stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: frag2.ToVectorisedView(), + }) + if _, _, ok := proto.Parse(pkt); !ok { + t.Fatalf("failed to parse packet: %x", pkt.Data.ToView()) + } + ep.HandlePacket(&r, pkt) if o.dataCalls != 1 { t.Fatalf("Bad number of data calls: got %x, want 1", o.dataCalls) } @@ -485,8 +497,11 @@ func TestIPv6Send(t *testing.T) { payload[i] = uint8(i) } - // Allocate the header buffer. - hdr := buffer.NewPrependable(int(ep.MaxHeaderLength())) + // Setup the packet buffer. + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(ep.MaxHeaderLength()), + Data: payload.ToVectorisedView(), + }) // Issue the write. o.protocol = 123 @@ -502,10 +517,7 @@ func TestIPv6Send(t *testing.T) { Protocol: 123, TTL: 123, TOS: stack.DefaultTOS, - }, &stack.PacketBuffer{ - Header: hdr, - Data: payload.ToVectorisedView(), - }); err != nil { + }, pkt); err != nil { t.Fatalf("WritePacket failed: %v", err) } } @@ -545,9 +557,13 @@ func TestIPv6Receive(t *testing.T) { t.Fatalf("could not find route: %v", err) } - pkt := stack.PacketBuffer{Data: view.ToVectorisedView()} - proto.Parse(&pkt) - ep.HandlePacket(&r, &pkt) + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: view.ToVectorisedView(), + }) + if _, _, ok := proto.Parse(pkt); !ok { + t.Fatalf("failed to parse packet: %x", pkt.Data.ToView()) + } + ep.HandlePacket(&r, pkt) if o.dataCalls != 1 { t.Fatalf("Bad number of data calls: got %x, want 1", o.dataCalls) } @@ -673,11 +689,9 @@ func TestIPv6ReceiveControl(t *testing.T) { // becomes Data. func truncatedPacket(view buffer.View, trunc, netHdrLen int) *stack.PacketBuffer { v := view[:len(view)-trunc] - if len(v) < netHdrLen { - return &stack.PacketBuffer{Data: v.ToVectorisedView()} - } - return &stack.PacketBuffer{ - NetworkHeader: v[:netHdrLen], - Data: v[netHdrLen:].ToVectorisedView(), - } + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: v.ToVectorisedView(), + }) + _, _ = pkt.NetworkHeader().Consume(netHdrLen) + return pkt } diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go index 94803a359..067d770f3 100644 --- a/pkg/tcpip/network/ipv4/icmp.go +++ b/pkg/tcpip/network/ipv4/icmp.go @@ -89,12 +89,14 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer) { return } + // Make a copy of data before pkt gets sent to raw socket. + // DeliverTransportPacket will take ownership of pkt. + replyData := pkt.Data.Clone(nil) + replyData.TrimFront(header.ICMPv4MinimumSize) + // It's possible that a raw socket expects to receive this. h.SetChecksum(wantChecksum) - e.dispatcher.DeliverTransportPacket(r, header.ICMPv4ProtocolNumber, &stack.PacketBuffer{ - Data: pkt.Data.Clone(nil), - NetworkHeader: append(buffer.View(nil), pkt.NetworkHeader...), - }) + e.dispatcher.DeliverTransportPacket(r, header.ICMPv4ProtocolNumber, pkt) remoteLinkAddr := r.RemoteLinkAddress @@ -116,24 +118,26 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer) { // Use the remote link address from the incoming packet. r.ResolveWith(remoteLinkAddr) - vv := pkt.Data.Clone(nil) - vv.TrimFront(header.ICMPv4MinimumSize) - hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv4MinimumSize) - pkt := header.ICMPv4(hdr.Prepend(header.ICMPv4MinimumSize)) - copy(pkt, h) - pkt.SetType(header.ICMPv4EchoReply) - pkt.SetChecksum(0) - pkt.SetChecksum(^header.Checksum(pkt, header.ChecksumVV(vv, 0))) + // Prepare a reply packet. + icmpHdr := make(header.ICMPv4, header.ICMPv4MinimumSize) + copy(icmpHdr, h) + icmpHdr.SetType(header.ICMPv4EchoReply) + icmpHdr.SetChecksum(0) + icmpHdr.SetChecksum(^header.Checksum(icmpHdr, header.ChecksumVV(replyData, 0))) + dataVV := buffer.View(icmpHdr).ToVectorisedView() + dataVV.Append(replyData) + replyPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(r.MaxHeaderLength()), + Data: dataVV, + }) + + // Send out the reply packet. sent := stats.ICMP.V4PacketsSent if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{ Protocol: header.ICMPv4ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS, - }, &stack.PacketBuffer{ - Header: hdr, - Data: vv, - TransportHeader: buffer.View(pkt), - }); err != nil { + }, replyPkt); err != nil { sent.Dropped.Increment() return } diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index 9ff27a363..3cd48ceb3 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -21,7 +21,6 @@ package ipv4 import ( - "fmt" "sync/atomic" "gvisor.dev/gvisor/pkg/tcpip" @@ -127,14 +126,12 @@ func (e *endpoint) NetworkProtocolNumber() tcpip.NetworkProtocolNumber { } // writePacketFragments calls e.linkEP.WritePacket with each packet fragment to -// write. It assumes that the IP header is entirely in pkt.Header but does not -// assume that only the IP header is in pkt.Header. It assumes that the input -// packet's stated length matches the length of the header+payload. mtu -// includes the IP header and options. This does not support the DontFragment -// IP flag. +// write. It assumes that the IP header is already present in pkt.NetworkHeader. +// pkt.TransportHeader may be set. mtu includes the IP header and options. This +// does not support the DontFragment IP flag. func (e *endpoint) writePacketFragments(r *stack.Route, gso *stack.GSO, mtu int, pkt *stack.PacketBuffer) *tcpip.Error { // This packet is too big, it needs to be fragmented. - ip := header.IPv4(pkt.Header.View()) + ip := header.IPv4(pkt.NetworkHeader().View()) flags := ip.Flags() // Update mtu to take into account the header, which will exist in all @@ -148,88 +145,84 @@ func (e *endpoint) writePacketFragments(r *stack.Route, gso *stack.GSO, mtu int, outerMTU := innerMTU + int(ip.HeaderLength()) offset := ip.FragmentOffset() - originalAvailableLength := pkt.Header.AvailableLength() + + // Keep the length reserved for link-layer, we need to create fragments with + // the same reserved length. + reservedForLink := pkt.AvailableHeaderBytes() + + // Destroy the packet, pull all payloads out for fragmentation. + transHeader, data := pkt.TransportHeader().View(), pkt.Data + + // Where possible, the first fragment that is sent has the same + // number of bytes reserved for header as the input packet. The link-layer + // endpoint may depend on this for looking at, eg, L4 headers. + transFitsFirst := len(transHeader) <= innerMTU + for i := 0; i < n; i++ { - // Where possible, the first fragment that is sent has the same - // pkt.Header.UsedLength() as the input packet. The link-layer - // endpoint may depend on this for looking at, eg, L4 headers. - h := ip - if i > 0 { - pkt.Header = buffer.NewPrependable(int(ip.HeaderLength()) + originalAvailableLength) - h = header.IPv4(pkt.Header.Prepend(int(ip.HeaderLength()))) - copy(h, ip[:ip.HeaderLength()]) + reserve := reservedForLink + int(ip.HeaderLength()) + if i == 0 && transFitsFirst { + // Reserve for transport header if it's going to be put in the first + // fragment. + reserve += len(transHeader) + } + fragPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: reserve, + }) + fragPkt.NetworkProtocolNumber = header.IPv4ProtocolNumber + + // Copy data for the fragment. + avail := innerMTU + + if n := len(transHeader); n > 0 { + if n > avail { + n = avail + } + if i == 0 && transFitsFirst { + copy(fragPkt.TransportHeader().Push(n), transHeader) + } else { + fragPkt.Data.AppendView(transHeader[:n:n]) + } + transHeader = transHeader[n:] + avail -= n } + + if avail > 0 { + n := data.Size() + if n > avail { + n = avail + } + data.ReadToVV(&fragPkt.Data, n) + avail -= n + } + + copied := uint16(innerMTU - avail) + + // Set lengths in header and calculate checksum. + h := header.IPv4(fragPkt.NetworkHeader().Push(len(ip))) + copy(h, ip) if i != n-1 { h.SetTotalLength(uint16(outerMTU)) h.SetFlagsFragmentOffset(flags|header.IPv4FlagMoreFragments, offset) } else { - h.SetTotalLength(uint16(h.HeaderLength()) + uint16(pkt.Data.Size())) + h.SetTotalLength(uint16(h.HeaderLength()) + copied) h.SetFlagsFragmentOffset(flags, offset) } h.SetChecksum(0) h.SetChecksum(^h.CalculateChecksum()) - offset += uint16(innerMTU) - if i > 0 { - newPayload := pkt.Data.Clone(nil) - newPayload.CapLength(innerMTU) - if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, &stack.PacketBuffer{ - Header: pkt.Header, - Data: newPayload, - NetworkHeader: buffer.View(h), - NetworkProtocolNumber: header.IPv4ProtocolNumber, - }); err != nil { - return err - } - r.Stats().IP.PacketsSent.Increment() - pkt.Data.TrimFront(newPayload.Size()) - continue - } - // Special handling for the first fragment because it comes - // from the header. - if outerMTU >= pkt.Header.UsedLength() { - // This fragment can fit all of pkt.Header and possibly - // some of pkt.Data, too. - newPayload := pkt.Data.Clone(nil) - newPayloadLength := outerMTU - pkt.Header.UsedLength() - newPayload.CapLength(newPayloadLength) - if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, &stack.PacketBuffer{ - Header: pkt.Header, - Data: newPayload, - NetworkHeader: buffer.View(h), - NetworkProtocolNumber: header.IPv4ProtocolNumber, - }); err != nil { - return err - } - r.Stats().IP.PacketsSent.Increment() - pkt.Data.TrimFront(newPayloadLength) - } else { - // The fragment is too small to fit all of pkt.Header. - startOfHdr := pkt.Header - startOfHdr.TrimBack(pkt.Header.UsedLength() - outerMTU) - emptyVV := buffer.NewVectorisedView(0, []buffer.View{}) - if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, &stack.PacketBuffer{ - Header: startOfHdr, - Data: emptyVV, - NetworkHeader: buffer.View(h), - NetworkProtocolNumber: header.IPv4ProtocolNumber, - }); err != nil { - return err - } - r.Stats().IP.PacketsSent.Increment() - // Add the unused bytes of pkt.Header into the pkt.Data - // that remains to be sent. - restOfHdr := pkt.Header.View()[outerMTU:] - tmp := buffer.NewVectorisedView(len(restOfHdr), []buffer.View{buffer.NewViewFromBytes(restOfHdr)}) - tmp.Append(pkt.Data) - pkt.Data = tmp + offset += copied + + // Send out the fragment. + if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, fragPkt); err != nil { + return err } + r.Stats().IP.PacketsSent.Increment() } return nil } -func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadSize int, params stack.NetworkHeaderParams) header.IPv4 { - ip := header.IPv4(hdr.Prepend(header.IPv4MinimumSize)) - length := uint16(hdr.UsedLength() + payloadSize) +func (e *endpoint) addIPHeader(r *stack.Route, pkt *stack.PacketBuffer, params stack.NetworkHeaderParams) { + ip := header.IPv4(pkt.NetworkHeader().Push(header.IPv4MinimumSize)) + length := uint16(pkt.Size()) // RFC 6864 section 4.3 mandates uniqueness of ID values for non-atomic // datagrams. Since the DF bit is never being set here, all datagrams // are non-atomic and need an ID. @@ -245,14 +238,12 @@ func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadS DstAddr: r.RemoteAddress, }) ip.SetChecksum(^ip.CalculateChecksum()) - return ip + pkt.NetworkProtocolNumber = header.IPv4ProtocolNumber } // WritePacket writes a packet to the given destination address and protocol. func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, pkt *stack.PacketBuffer) *tcpip.Error { - ip := e.addIPHeader(r, &pkt.Header, pkt.Data.Size(), params) - pkt.NetworkHeader = buffer.View(ip) - pkt.NetworkProtocolNumber = header.IPv4ProtocolNumber + e.addIPHeader(r, pkt, params) // iptables filtering. All packets that reach here are locally // generated. @@ -269,7 +260,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw // only NATted packets, but removing this check short circuits broadcasts // before they are sent out to other hosts. if pkt.NatDone { - netHeader := header.IPv4(pkt.NetworkHeader) + netHeader := header.IPv4(pkt.NetworkHeader().View()) ep, err := e.stack.FindNetworkEndpoint(header.IPv4ProtocolNumber, netHeader.DestinationAddress()) if err == nil { route := r.ReverseRoute(netHeader.SourceAddress(), netHeader.DestinationAddress()) @@ -286,7 +277,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw if r.Loop&stack.PacketOut == 0 { return nil } - if pkt.Header.UsedLength()+pkt.Data.Size() > int(e.linkEP.MTU()) && (gso == nil || gso.Type == stack.GSONone) { + if pkt.Size() > int(e.linkEP.MTU()) && (gso == nil || gso.Type == stack.GSONone) { return e.writePacketFragments(r, gso, int(e.linkEP.MTU()), pkt) } if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, pkt); err != nil { @@ -306,9 +297,7 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe } for pkt := pkts.Front(); pkt != nil; { - ip := e.addIPHeader(r, &pkt.Header, pkt.Data.Size(), params) - pkt.NetworkHeader = buffer.View(ip) - pkt.NetworkProtocolNumber = header.IPv4ProtocolNumber + e.addIPHeader(r, pkt, params) pkt = pkt.Next() } @@ -333,7 +322,7 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe continue } if _, ok := natPkts[pkt]; ok { - netHeader := header.IPv4(pkt.NetworkHeader) + netHeader := header.IPv4(pkt.NetworkHeader().View()) if ep, err := e.stack.FindNetworkEndpoint(header.IPv4ProtocolNumber, netHeader.DestinationAddress()); err == nil { src := netHeader.SourceAddress() dst := netHeader.DestinationAddress() @@ -402,17 +391,14 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBu r.Stats().IP.PacketsSent.Increment() - ip = ip[:ip.HeaderLength()] - pkt.Header = buffer.NewPrependableFromView(buffer.View(ip)) - pkt.Data.TrimFront(int(ip.HeaderLength())) return e.linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, pkt) } // HandlePacket is called by the link layer when new ipv4 packets arrive for // this endpoint. func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { - h := header.IPv4(pkt.NetworkHeader) - if !h.IsValid(pkt.Data.Size() + len(pkt.NetworkHeader) + len(pkt.TransportHeader)) { + h := header.IPv4(pkt.NetworkHeader().View()) + if !h.IsValid(pkt.Data.Size() + pkt.NetworkHeader().View().Size() + pkt.TransportHeader().View().Size()) { r.Stats().IP.MalformedPacketsReceived.Increment() return } @@ -426,7 +412,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { } if h.More() || h.FragmentOffset() != 0 { - if pkt.Data.Size()+len(pkt.TransportHeader) == 0 { + if pkt.Data.Size()+pkt.TransportHeader().View().Size() == 0 { // Drop the packet as it's marked as a fragment but has // no payload. r.Stats().IP.MalformedPacketsReceived.Increment() @@ -470,7 +456,6 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { } p := h.TransportProtocol() if p == header.ICMPv4ProtocolNumber { - pkt.NetworkHeader.CapLength(int(h.HeaderLength())) e.handleICMP(r, pkt) return } @@ -560,14 +545,19 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) (proto tcpip.TransportProtocolNu } ipHdr := header.IPv4(hdr) - // If there are options, pull those into hdr as well. - if headerLen := int(ipHdr.HeaderLength()); headerLen > header.IPv4MinimumSize && headerLen <= pkt.Data.Size() { - hdr, ok = pkt.Data.PullUp(headerLen) - if !ok { - panic(fmt.Sprintf("There are only %d bytes in pkt.Data, but there should be at least %d", pkt.Data.Size(), headerLen)) - } - ipHdr = header.IPv4(hdr) + // Header may have options, determine the true header length. + headerLen := int(ipHdr.HeaderLength()) + if headerLen < header.IPv4MinimumSize { + // TODO(gvisor.dev/issue/2404): Per RFC 791, IHL needs to be at least 5 in + // order for the packet to be valid. Figure out if we want to reject this + // case. + headerLen = header.IPv4MinimumSize + } + hdr, ok = pkt.NetworkHeader().Consume(headerLen) + if !ok { + return 0, false, false } + ipHdr = header.IPv4(hdr) // If this is a fragment, don't bother parsing the transport header. parseTransportHeader := true @@ -576,8 +566,6 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) (proto tcpip.TransportProtocolNu } pkt.NetworkProtocolNumber = header.IPv4ProtocolNumber - pkt.NetworkHeader = hdr - pkt.Data.TrimFront(len(hdr)) pkt.Data.CapLength(int(ipHdr.TotalLength()) - len(hdr)) return ipHdr.TransportProtocol(), parseTransportHeader, true } diff --git a/pkg/tcpip/network/ipv4/ipv4_test.go b/pkg/tcpip/network/ipv4/ipv4_test.go index 63e2c36c2..afd3ac06d 100644 --- a/pkg/tcpip/network/ipv4/ipv4_test.go +++ b/pkg/tcpip/network/ipv4/ipv4_test.go @@ -17,6 +17,7 @@ package ipv4_test import ( "bytes" "encoding/hex" + "fmt" "math/rand" "testing" @@ -91,15 +92,11 @@ func TestExcludeBroadcast(t *testing.T) { }) } -// makeHdrAndPayload generates a randomize packet. hdrLength indicates how much +// makeRandPkt generates a randomize packet. hdrLength indicates how much // data should already be in the header before WritePacket. extraLength // indicates how much extra space should be in the header. The payload is made // from many Views of the sizes listed in viewSizes. -func makeHdrAndPayload(hdrLength int, extraLength int, viewSizes []int) (buffer.Prependable, buffer.VectorisedView) { - hdr := buffer.NewPrependable(hdrLength + extraLength) - hdr.Prepend(hdrLength) - rand.Read(hdr.View()) - +func makeRandPkt(hdrLength int, extraLength int, viewSizes []int) *stack.PacketBuffer { var views []buffer.View totalLength := 0 for _, s := range viewSizes { @@ -108,8 +105,16 @@ func makeHdrAndPayload(hdrLength int, extraLength int, viewSizes []int) (buffer. views = append(views, newView) totalLength += s } - payload := buffer.NewVectorisedView(totalLength, views) - return hdr, payload + + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: hdrLength + extraLength, + Data: buffer.NewVectorisedView(totalLength, views), + }) + pkt.NetworkProtocolNumber = header.IPv4ProtocolNumber + if _, err := rand.Read(pkt.TransportHeader().Push(hdrLength)); err != nil { + panic(fmt.Sprintf("rand.Read: %s", err)) + } + return pkt } // comparePayloads compared the contents of all the packets against the contents @@ -117,9 +122,9 @@ func makeHdrAndPayload(hdrLength int, extraLength int, viewSizes []int) (buffer. func compareFragments(t *testing.T, packets []*stack.PacketBuffer, sourcePacketInfo *stack.PacketBuffer, mtu uint32) { t.Helper() // Make a complete array of the sourcePacketInfo packet. - source := header.IPv4(packets[0].Header.View()[:header.IPv4MinimumSize]) - source = append(source, sourcePacketInfo.Header.View()...) - source = append(source, sourcePacketInfo.Data.ToView()...) + source := header.IPv4(packets[0].NetworkHeader().View()[:header.IPv4MinimumSize]) + vv := buffer.NewVectorisedView(sourcePacketInfo.Size(), sourcePacketInfo.Views()) + source = append(source, vv.ToView()...) // Make a copy of the IP header, which will be modified in some fields to make // an expected header. @@ -132,8 +137,7 @@ func compareFragments(t *testing.T, packets []*stack.PacketBuffer, sourcePacketI var reassembledPayload []byte for i, packet := range packets { // Confirm that the packet is valid. - allBytes := packet.Header.View().ToVectorisedView() - allBytes.Append(packet.Data) + allBytes := buffer.NewVectorisedView(packet.Size(), packet.Views()) ip := header.IPv4(allBytes.ToView()) if !ip.IsValid(len(ip)) { t.Errorf("IP packet is invalid:\n%s", hex.Dump(ip)) @@ -144,10 +148,17 @@ func compareFragments(t *testing.T, packets []*stack.PacketBuffer, sourcePacketI if got, want := len(ip), int(mtu); got > want { t.Errorf("fragment is too large, got %d want %d", got, want) } - if got, want := packet.Header.UsedLength(), sourcePacketInfo.Header.UsedLength()+header.IPv4MinimumSize; i == 0 && want < int(mtu) && got != want { - t.Errorf("first fragment hdr parts should have unmodified length if possible: got %d, want %d", got, want) + if i == 0 { + got := packet.NetworkHeader().View().Size() + packet.TransportHeader().View().Size() + // sourcePacketInfo does not have NetworkHeader added, simulate one. + want := header.IPv4MinimumSize + sourcePacketInfo.TransportHeader().View().Size() + // Check that it kept the transport header in packet.TransportHeader if + // it fits in the first fragment. + if want < int(mtu) && got != want { + t.Errorf("first fragment hdr parts should have unmodified length if possible: got %d, want %d", got, want) + } } - if got, want := packet.Header.AvailableLength(), sourcePacketInfo.Header.AvailableLength()-header.IPv4MinimumSize; got != want { + if got, want := packet.AvailableHeaderBytes(), sourcePacketInfo.AvailableHeaderBytes()-header.IPv4MinimumSize; got != want { t.Errorf("fragment #%d should have the same available space for prepending as source: got %d, want %d", i, got, want) } if got, want := packet.NetworkProtocolNumber, sourcePacketInfo.NetworkProtocolNumber; got != want { @@ -284,22 +295,14 @@ func TestFragmentation(t *testing.T) { for _, ft := range fragTests { t.Run(ft.description, func(t *testing.T) { - hdr, payload := makeHdrAndPayload(ft.hdrLength, ft.extraLength, ft.payloadViewsSizes) - source := &stack.PacketBuffer{ - Header: hdr, - // Save the source payload because WritePacket will modify it. - Data: payload.Clone(nil), - NetworkProtocolNumber: header.IPv4ProtocolNumber, - } + pkt := makeRandPkt(ft.hdrLength, ft.extraLength, ft.payloadViewsSizes) + source := pkt.Clone() c := buildContext(t, nil, ft.mtu) err := c.Route.WritePacket(ft.gso, stack.NetworkHeaderParams{ Protocol: tcp.ProtocolNumber, TTL: 42, TOS: stack.DefaultTOS, - }, &stack.PacketBuffer{ - Header: hdr, - Data: payload, - }) + }, pkt) if err != nil { t.Errorf("err got %v, want %v", err, nil) } @@ -344,16 +347,13 @@ func TestFragmentationErrors(t *testing.T) { for _, ft := range fragTests { t.Run(ft.description, func(t *testing.T) { - hdr, payload := makeHdrAndPayload(ft.hdrLength, header.IPv4MinimumSize, ft.payloadViewsSizes) + pkt := makeRandPkt(ft.hdrLength, header.IPv4MinimumSize, ft.payloadViewsSizes) c := buildContext(t, ft.packetCollectorErrors, ft.mtu) err := c.Route.WritePacket(&stack.GSO{}, stack.NetworkHeaderParams{ Protocol: tcp.ProtocolNumber, TTL: 42, TOS: stack.DefaultTOS, - }, &stack.PacketBuffer{ - Header: hdr, - Data: payload, - }) + }, pkt) for i := 0; i < len(ft.packetCollectorErrors)-1; i++ { if got, want := ft.packetCollectorErrors[i], (*tcpip.Error)(nil); got != want { t.Errorf("ft.packetCollectorErrors[%d] got %v, want %v", i, got, want) @@ -472,9 +472,9 @@ func TestInvalidFragments(t *testing.T) { s.CreateNIC(nicID, sniffer.New(ep)) for _, pkt := range tc.packets { - ep.InjectLinkAddr(header.IPv4ProtocolNumber, remoteLinkAddr, &stack.PacketBuffer{ + ep.InjectLinkAddr(header.IPv4ProtocolNumber, remoteLinkAddr, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buffer.NewVectorisedView(len(pkt), []buffer.View{pkt}), - }) + })) } if got, want := s.Stats().IP.MalformedPacketsReceived.Value(), tc.wantMalformedIPPackets; got != want { @@ -859,9 +859,9 @@ func TestReceiveFragments(t *testing.T) { vv := hdr.View().ToVectorisedView() vv.AppendView(frag.payload) - e.InjectInbound(header.IPv4ProtocolNumber, &stack.PacketBuffer{ + e.InjectInbound(header.IPv4ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: vv, - }) + })) } if got, want := s.Stats().UDP.PacketsReceived.Value(), uint64(len(test.expectedPayloads)); got != want { diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index ded91d83a..39ae19295 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -83,7 +83,7 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme return } h := header.ICMPv6(v) - iph := header.IPv6(pkt.NetworkHeader) + iph := header.IPv6(pkt.NetworkHeader().View()) // Validate ICMPv6 checksum before processing the packet. // @@ -276,8 +276,10 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme optsSerializer := header.NDPOptionsSerializer{ header.NDPTargetLinkLayerAddressOption(r.LocalLinkAddress), } - hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv6NeighborAdvertMinimumSize + int(optsSerializer.Length())) - packet := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborAdvertSize)) + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(r.MaxHeaderLength()) + header.ICMPv6NeighborAdvertMinimumSize + int(optsSerializer.Length()), + }) + packet := header.ICMPv6(pkt.TransportHeader().Push(header.ICMPv6NeighborAdvertSize)) packet.SetType(header.ICMPv6NeighborAdvert) na := header.NDPNeighborAdvert(packet.NDPPayload()) na.SetSolicitedFlag(solicited) @@ -293,9 +295,7 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme // // The IP Hop Limit field has a value of 255, i.e., the packet // could not possibly have been forwarded by a router. - if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, TOS: stack.DefaultTOS}, &stack.PacketBuffer{ - Header: hdr, - }); err != nil { + if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, TOS: stack.DefaultTOS}, pkt); err != nil { sent.Dropped.Increment() return } @@ -384,7 +384,7 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme case header.ICMPv6EchoRequest: received.EchoRequest.Increment() - icmpHdr, ok := pkt.Data.PullUp(header.ICMPv6EchoMinimumSize) + icmpHdr, ok := pkt.TransportHeader().Consume(header.ICMPv6EchoMinimumSize) if !ok { received.Invalid.Increment() return @@ -409,16 +409,15 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme // Use the link address from the source of the original packet. r.ResolveWith(remoteLinkAddr) - pkt.Data.TrimFront(header.ICMPv6EchoMinimumSize) - hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv6EchoMinimumSize) - packet := header.ICMPv6(hdr.Prepend(header.ICMPv6EchoMinimumSize)) + replyPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(r.MaxHeaderLength()) + header.ICMPv6EchoMinimumSize, + Data: pkt.Data, + }) + packet := header.ICMPv6(replyPkt.TransportHeader().Push(header.ICMPv6EchoMinimumSize)) copy(packet, icmpHdr) packet.SetType(header.ICMPv6EchoReply) packet.SetChecksum(header.ICMPv6Checksum(packet, r.LocalAddress, r.RemoteAddress, pkt.Data)) - if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, &stack.PacketBuffer{ - Header: hdr, - Data: pkt.Data, - }); err != nil { + if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, replyPkt); err != nil { sent.Dropped.Increment() return } @@ -539,17 +538,19 @@ func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, remoteLinkAdd r.RemoteLinkAddress = header.EthernetAddressFromMulticastIPv6Address(snaddr) } - hdr := buffer.NewPrependable(int(linkEP.MaxHeaderLength()) + header.IPv6MinimumSize + header.ICMPv6NeighborAdvertSize) - pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborAdvertSize)) - pkt.SetType(header.ICMPv6NeighborSolicit) - copy(pkt[icmpV6OptOffset-len(addr):], addr) - pkt[icmpV6OptOffset] = ndpOptSrcLinkAddr - pkt[icmpV6LengthOffset] = 1 - copy(pkt[icmpV6LengthOffset+1:], linkEP.LinkAddress()) - pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) - - length := uint16(hdr.UsedLength()) - ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(linkEP.MaxHeaderLength()) + header.IPv6MinimumSize + header.ICMPv6NeighborAdvertSize, + }) + icmpHdr := header.ICMPv6(pkt.TransportHeader().Push(header.ICMPv6NeighborAdvertSize)) + icmpHdr.SetType(header.ICMPv6NeighborSolicit) + copy(icmpHdr[icmpV6OptOffset-len(addr):], addr) + icmpHdr[icmpV6OptOffset] = ndpOptSrcLinkAddr + icmpHdr[icmpV6LengthOffset] = 1 + copy(icmpHdr[icmpV6LengthOffset+1:], linkEP.LinkAddress()) + icmpHdr.SetChecksum(header.ICMPv6Checksum(icmpHdr, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) + + length := uint16(pkt.Size()) + ip := header.IPv6(pkt.NetworkHeader().Push(header.IPv6MinimumSize)) ip.Encode(&header.IPv6Fields{ PayloadLength: length, NextHeader: uint8(header.ICMPv6ProtocolNumber), @@ -559,9 +560,7 @@ func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, remoteLinkAdd }) // TODO(stijlist): count this in ICMP stats. - return linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, &stack.PacketBuffer{ - Header: hdr, - }) + return linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, pkt) } // ResolveStaticAddress implements stack.LinkAddressResolver. diff --git a/pkg/tcpip/network/ipv6/icmp_test.go b/pkg/tcpip/network/ipv6/icmp_test.go index f86aaed1d..2a2f7de01 100644 --- a/pkg/tcpip/network/ipv6/icmp_test.go +++ b/pkg/tcpip/network/ipv6/icmp_test.go @@ -183,7 +183,11 @@ func TestICMPCounts(t *testing.T) { } handleIPv6Payload := func(icmp header.ICMPv6) { - ip := header.IPv6(buffer.NewView(header.IPv6MinimumSize)) + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: header.IPv6MinimumSize, + Data: buffer.View(icmp).ToVectorisedView(), + }) + ip := header.IPv6(pkt.NetworkHeader().Push(header.IPv6MinimumSize)) ip.Encode(&header.IPv6Fields{ PayloadLength: uint16(len(icmp)), NextHeader: uint8(header.ICMPv6ProtocolNumber), @@ -191,10 +195,7 @@ func TestICMPCounts(t *testing.T) { SrcAddr: r.LocalAddress, DstAddr: r.RemoteAddress, }) - ep.HandlePacket(&r, &stack.PacketBuffer{ - NetworkHeader: buffer.View(ip), - Data: buffer.View(icmp).ToVectorisedView(), - }) + ep.HandlePacket(&r, pkt) } for _, typ := range types { @@ -323,12 +324,10 @@ func routeICMPv6Packet(t *testing.T, args routeArgs, fn func(*testing.T, header. pi, _ := args.src.ReadContext(context.Background()) { - views := []buffer.View{pi.Pkt.Header.View(), pi.Pkt.Data.ToView()} - size := pi.Pkt.Header.UsedLength() + pi.Pkt.Data.Size() - vv := buffer.NewVectorisedView(size, views) - args.dst.InjectLinkAddr(pi.Proto, args.dst.LinkAddress(), &stack.PacketBuffer{ - Data: vv, + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: buffer.NewVectorisedView(pi.Pkt.Size(), pi.Pkt.Views()), }) + args.dst.InjectLinkAddr(pi.Proto, args.dst.LinkAddress(), pkt) } if pi.Proto != ProtocolNumber { @@ -340,7 +339,9 @@ func routeICMPv6Packet(t *testing.T, args routeArgs, fn func(*testing.T, header. t.Errorf("got remote link address = %s, want = %s", pi.Route.RemoteLinkAddress, args.remoteLinkAddr) } - ipv6 := header.IPv6(pi.Pkt.Header.View()) + // Pull the full payload since network header. Needed for header.IPv6 to + // extract its payload. + ipv6 := header.IPv6(stack.PayloadSince(pi.Pkt.NetworkHeader())) transProto := tcpip.TransportProtocolNumber(ipv6.NextHeader()) if transProto != header.ICMPv6ProtocolNumber { t.Errorf("unexpected transport protocol number %d", transProto) @@ -558,9 +559,10 @@ func TestICMPChecksumValidationSimple(t *testing.T) { SrcAddr: lladdr1, DstAddr: lladdr0, }) - e.InjectInbound(ProtocolNumber, &stack.PacketBuffer{ + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buffer.NewVectorisedView(len(ip)+len(icmp), []buffer.View{buffer.View(ip), buffer.View(icmp)}), }) + e.InjectInbound(ProtocolNumber, pkt) } stats := s.Stats().ICMP.V6PacketsReceived @@ -719,12 +721,12 @@ func TestICMPChecksumValidationWithPayload(t *testing.T) { handleIPv6Payload := func(typ header.ICMPv6Type, size, payloadSize int, payloadFn func(buffer.View), checksum bool) { icmpSize := size + payloadSize hdr := buffer.NewPrependable(header.IPv6MinimumSize + icmpSize) - pkt := header.ICMPv6(hdr.Prepend(icmpSize)) - pkt.SetType(typ) - payloadFn(pkt.Payload()) + icmpHdr := header.ICMPv6(hdr.Prepend(icmpSize)) + icmpHdr.SetType(typ) + payloadFn(icmpHdr.Payload()) if checksum { - pkt.SetChecksum(header.ICMPv6Checksum(pkt, lladdr1, lladdr0, buffer.VectorisedView{})) + icmpHdr.SetChecksum(header.ICMPv6Checksum(icmpHdr, lladdr1, lladdr0, buffer.VectorisedView{})) } ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) @@ -735,9 +737,10 @@ func TestICMPChecksumValidationWithPayload(t *testing.T) { SrcAddr: lladdr1, DstAddr: lladdr0, }) - e.InjectInbound(ProtocolNumber, &stack.PacketBuffer{ + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: hdr.View().ToVectorisedView(), }) + e.InjectInbound(ProtocolNumber, pkt) } stats := s.Stats().ICMP.V6PacketsReceived @@ -895,14 +898,14 @@ func TestICMPChecksumValidationWithPayloadMultipleViews(t *testing.T) { handleIPv6Payload := func(typ header.ICMPv6Type, size, payloadSize int, payloadFn func(buffer.View), checksum bool) { hdr := buffer.NewPrependable(header.IPv6MinimumSize + size) - pkt := header.ICMPv6(hdr.Prepend(size)) - pkt.SetType(typ) + icmpHdr := header.ICMPv6(hdr.Prepend(size)) + icmpHdr.SetType(typ) payload := buffer.NewView(payloadSize) payloadFn(payload) if checksum { - pkt.SetChecksum(header.ICMPv6Checksum(pkt, lladdr1, lladdr0, payload.ToVectorisedView())) + icmpHdr.SetChecksum(header.ICMPv6Checksum(icmpHdr, lladdr1, lladdr0, payload.ToVectorisedView())) } ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) @@ -913,9 +916,10 @@ func TestICMPChecksumValidationWithPayloadMultipleViews(t *testing.T) { SrcAddr: lladdr1, DstAddr: lladdr0, }) - e.InjectInbound(ProtocolNumber, &stack.PacketBuffer{ + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buffer.NewVectorisedView(header.IPv6MinimumSize+size+payloadSize, []buffer.View{hdr.View(), payload}), }) + e.InjectInbound(ProtocolNumber, pkt) } stats := s.Stats().ICMP.V6PacketsReceived diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index d7d7fc611..0ade655b2 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -99,9 +99,9 @@ func (e *endpoint) GSOMaxSize() uint32 { return 0 } -func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadSize int, params stack.NetworkHeaderParams) header.IPv6 { - length := uint16(hdr.UsedLength() + payloadSize) - ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) +func (e *endpoint) addIPHeader(r *stack.Route, pkt *stack.PacketBuffer, params stack.NetworkHeaderParams) { + length := uint16(pkt.Size()) + ip := header.IPv6(pkt.NetworkHeader().Push(header.IPv6MinimumSize)) ip.Encode(&header.IPv6Fields{ PayloadLength: length, NextHeader: uint8(params.Protocol), @@ -110,26 +110,20 @@ func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadS SrcAddr: r.LocalAddress, DstAddr: r.RemoteAddress, }) - return ip + pkt.NetworkProtocolNumber = header.IPv6ProtocolNumber } // WritePacket writes a packet to the given destination address and protocol. func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, pkt *stack.PacketBuffer) *tcpip.Error { - ip := e.addIPHeader(r, &pkt.Header, pkt.Data.Size(), params) - pkt.NetworkHeader = buffer.View(ip) - pkt.NetworkProtocolNumber = header.IPv6ProtocolNumber + e.addIPHeader(r, pkt, params) if r.Loop&stack.PacketLoop != 0 { - // The inbound path expects the network header to still be in - // the PacketBuffer's Data field. - views := make([]buffer.View, 1, 1+len(pkt.Data.Views())) - views[0] = pkt.Header.View() - views = append(views, pkt.Data.Views()...) loopedR := r.MakeLoopedRoute() - e.HandlePacket(&loopedR, &stack.PacketBuffer{ - Data: buffer.NewVectorisedView(len(views[0])+pkt.Data.Size(), views), - }) + e.HandlePacket(&loopedR, stack.NewPacketBuffer(stack.PacketBufferOptions{ + // The inbound path expects an unparsed packet. + Data: buffer.NewVectorisedView(pkt.Size(), pkt.Views()), + })) loopedR.Release() } @@ -151,9 +145,7 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe } for pb := pkts.Front(); pb != nil; pb = pb.Next() { - ip := e.addIPHeader(r, &pb.Header, pb.Data.Size(), params) - pb.NetworkHeader = buffer.View(ip) - pb.NetworkProtocolNumber = header.IPv6ProtocolNumber + e.addIPHeader(r, pb, params) } n, err := e.linkEP.WritePackets(r, gso, pkts, ProtocolNumber) @@ -171,8 +163,8 @@ func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBuff // HandlePacket is called by the link layer when new ipv6 packets arrive for // this endpoint. func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { - h := header.IPv6(pkt.NetworkHeader) - if !h.IsValid(pkt.Data.Size() + len(pkt.NetworkHeader) + len(pkt.TransportHeader)) { + h := header.IPv6(pkt.NetworkHeader().View()) + if !h.IsValid(pkt.Data.Size() + pkt.NetworkHeader().View().Size() + pkt.TransportHeader().View().Size()) { r.Stats().IP.MalformedPacketsReceived.Increment() return } @@ -181,8 +173,8 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { // - Any IPv6 header bytes after the first 40 (i.e. extensions). // - The transport header, if present. // - Any other payload data. - vv := pkt.NetworkHeader[header.IPv6MinimumSize:].ToVectorisedView() - vv.AppendView(pkt.TransportHeader) + vv := pkt.NetworkHeader().View()[header.IPv6MinimumSize:].ToVectorisedView() + vv.AppendView(pkt.TransportHeader().View()) vv.Append(pkt.Data) it := header.MakeIPv6PayloadIterator(header.IPv6ExtensionHeaderIdentifier(h.NextHeader()), vv) hasFragmentHeader := false @@ -410,7 +402,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { // // For reassembled fragments, pkt.TransportHeader is unset, so this is a // no-op and pkt.Data begins with the transport header. - extHdr.Buf.TrimFront(len(pkt.TransportHeader)) + extHdr.Buf.TrimFront(pkt.TransportHeader().View().Size()) pkt.Data = extHdr.Buf if p := tcpip.TransportProtocolNumber(extHdr.Identifier); p == header.ICMPv6ProtocolNumber { @@ -581,17 +573,14 @@ traverseExtensions: } } - // Put the IPv6 header with extensions in pkt.NetworkHeader. - hdr, ok = pkt.Data.PullUp(header.IPv6MinimumSize + extensionsSize) + // Put the IPv6 header with extensions in pkt.NetworkHeader(). + hdr, ok = pkt.NetworkHeader().Consume(header.IPv6MinimumSize + extensionsSize) if !ok { panic(fmt.Sprintf("pkt.Data should have at least %d bytes, but only has %d.", header.IPv6MinimumSize+extensionsSize, pkt.Data.Size())) } ipHdr = header.IPv6(hdr) - - pkt.NetworkProtocolNumber = header.IPv6ProtocolNumber - pkt.NetworkHeader = hdr - pkt.Data.TrimFront(len(hdr)) pkt.Data.CapLength(int(ipHdr.PayloadLength())) + pkt.NetworkProtocolNumber = header.IPv6ProtocolNumber return nextHdr, foundNext, true } diff --git a/pkg/tcpip/network/ipv6/ipv6_test.go b/pkg/tcpip/network/ipv6/ipv6_test.go index 3d65814de..081afb051 100644 --- a/pkg/tcpip/network/ipv6/ipv6_test.go +++ b/pkg/tcpip/network/ipv6/ipv6_test.go @@ -65,9 +65,9 @@ func testReceiveICMP(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst DstAddr: dst, }) - e.InjectInbound(ProtocolNumber, &stack.PacketBuffer{ + e.InjectInbound(ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: hdr.View().ToVectorisedView(), - }) + })) stats := s.Stats().ICMP.V6PacketsReceived @@ -123,9 +123,9 @@ func testReceiveUDP(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst DstAddr: dst, }) - e.InjectInbound(ProtocolNumber, &stack.PacketBuffer{ + e.InjectInbound(ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: hdr.View().ToVectorisedView(), - }) + })) stat := s.Stats().UDP.PacketsReceived @@ -637,9 +637,9 @@ func TestReceiveIPv6ExtHdrs(t *testing.T) { DstAddr: addr2, }) - e.InjectInbound(ProtocolNumber, &stack.PacketBuffer{ + e.InjectInbound(ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: hdr.View().ToVectorisedView(), - }) + })) stats := s.Stats().UDP.PacketsReceived @@ -1469,9 +1469,9 @@ func TestReceiveIPv6Fragments(t *testing.T) { vv := hdr.View().ToVectorisedView() vv.Append(f.data) - e.InjectInbound(ProtocolNumber, &stack.PacketBuffer{ + e.InjectInbound(ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: vv, - }) + })) } if got, want := s.Stats().UDP.PacketsReceived.Value(), uint64(len(test.expectedPayloads)); got != want { diff --git a/pkg/tcpip/network/ipv6/ndp_test.go b/pkg/tcpip/network/ipv6/ndp_test.go index 64239ce9a..fe159b24f 100644 --- a/pkg/tcpip/network/ipv6/ndp_test.go +++ b/pkg/tcpip/network/ipv6/ndp_test.go @@ -136,9 +136,9 @@ func TestNeighorSolicitationWithSourceLinkLayerOption(t *testing.T) { t.Fatalf("got invalid = %d, want = 0", got) } - e.InjectInbound(ProtocolNumber, &stack.PacketBuffer{ + e.InjectInbound(ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: hdr.View().ToVectorisedView(), - }) + })) linkAddr, c, err := s.GetLinkAddress(nicID, lladdr1, lladdr0, ProtocolNumber, nil) if linkAddr != test.expectedLinkAddr { @@ -380,9 +380,9 @@ func TestNeighorSolicitationResponse(t *testing.T) { t.Fatalf("got invalid = %d, want = 0", got) } - e.InjectLinkAddr(ProtocolNumber, test.nsSrcLinkAddr, &stack.PacketBuffer{ + e.InjectLinkAddr(ProtocolNumber, test.nsSrcLinkAddr, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: hdr.View().ToVectorisedView(), - }) + })) if test.nsInvalid { if got := invalid.Value(); got != 1 { @@ -410,7 +410,7 @@ func TestNeighorSolicitationResponse(t *testing.T) { t.Errorf("got p.Route.RemoteLinkAddress = %s, want = %s", p.Route.RemoteLinkAddress, test.naDstLinkAddr) } - checker.IPv6(t, p.Pkt.Header.View(), + checker.IPv6(t, stack.PayloadSince(p.Pkt.NetworkHeader()), checker.SrcAddr(test.naSrc), checker.DstAddr(test.naDst), checker.TTL(header.NDPHopLimit), @@ -497,9 +497,9 @@ func TestNeighorAdvertisementWithTargetLinkLayerOption(t *testing.T) { t.Fatalf("got invalid = %d, want = 0", got) } - e.InjectInbound(ProtocolNumber, &stack.PacketBuffer{ + e.InjectInbound(ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: hdr.View().ToVectorisedView(), - }) + })) linkAddr, c, err := s.GetLinkAddress(nicID, lladdr1, lladdr0, ProtocolNumber, nil) if linkAddr != test.expectedLinkAddr { @@ -560,7 +560,11 @@ func TestNDPValidation(t *testing.T) { nextHdr = uint8(header.IPv6FragmentExtHdrIdentifier) } - ip := header.IPv6(buffer.NewView(header.IPv6MinimumSize + len(extensions))) + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: header.IPv6MinimumSize + len(extensions), + Data: payload.ToVectorisedView(), + }) + ip := header.IPv6(pkt.NetworkHeader().Push(header.IPv6MinimumSize + len(extensions))) ip.Encode(&header.IPv6Fields{ PayloadLength: uint16(len(payload) + len(extensions)), NextHeader: nextHdr, @@ -571,10 +575,7 @@ func TestNDPValidation(t *testing.T) { if n := copy(ip[header.IPv6MinimumSize:], extensions); n != len(extensions) { t.Fatalf("expected to write %d bytes of extensions, but wrote %d", len(extensions), n) } - ep.HandlePacket(r, &stack.PacketBuffer{ - NetworkHeader: buffer.View(ip), - Data: payload.ToVectorisedView(), - }) + ep.HandlePacket(r, pkt) } var tllData [header.NDPLinkLayerAddressSize]byte @@ -885,9 +886,9 @@ func TestRouterAdvertValidation(t *testing.T) { t.Fatalf("got rxRA = %d, want = 0", got) } - e.InjectInbound(header.IPv6ProtocolNumber, &stack.PacketBuffer{ + e.InjectInbound(header.IPv6ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: hdr.View().ToVectorisedView(), - }) + })) if got := rxRA.Value(); got != 1 { t.Fatalf("got rxRA = %d, want = 1", got) diff --git a/pkg/tcpip/stack/BUILD b/pkg/tcpip/stack/BUILD index bfc7a0c7c..900938dd1 100644 --- a/pkg/tcpip/stack/BUILD +++ b/pkg/tcpip/stack/BUILD @@ -57,6 +57,7 @@ go_library( "conntrack.go", "dhcpv6configurationfromndpra_string.go", "forwarder.go", + "headertype_string.go", "icmp_rate_limit.go", "iptables.go", "iptables_state.go", @@ -143,6 +144,7 @@ go_test( "neighbor_cache_test.go", "neighbor_entry_test.go", "nic_test.go", + "packet_buffer_test.go", ], library = ":stack", deps = [ diff --git a/pkg/tcpip/stack/conntrack.go b/pkg/tcpip/stack/conntrack.go index 470c265aa..7dd344b4f 100644 --- a/pkg/tcpip/stack/conntrack.go +++ b/pkg/tcpip/stack/conntrack.go @@ -199,12 +199,12 @@ type bucket struct { func packetToTupleID(pkt *PacketBuffer) (tupleID, *tcpip.Error) { // TODO(gvisor.dev/issue/170): Need to support for other // protocols as well. - netHeader := header.IPv4(pkt.NetworkHeader) - if netHeader == nil || netHeader.TransportProtocol() != header.TCPProtocolNumber { + netHeader := header.IPv4(pkt.NetworkHeader().View()) + if len(netHeader) < header.IPv4MinimumSize || netHeader.TransportProtocol() != header.TCPProtocolNumber { return tupleID{}, tcpip.ErrUnknownProtocol } - tcpHeader := header.TCP(pkt.TransportHeader) - if tcpHeader == nil { + tcpHeader := header.TCP(pkt.TransportHeader().View()) + if len(tcpHeader) < header.TCPMinimumSize { return tupleID{}, tcpip.ErrUnknownProtocol } @@ -344,8 +344,8 @@ func handlePacketPrerouting(pkt *PacketBuffer, conn *conn, dir direction) { return } - netHeader := header.IPv4(pkt.NetworkHeader) - tcpHeader := header.TCP(pkt.TransportHeader) + netHeader := header.IPv4(pkt.NetworkHeader().View()) + tcpHeader := header.TCP(pkt.TransportHeader().View()) // For prerouting redirection, packets going in the original direction // have their destinations modified and replies have their sources @@ -377,8 +377,8 @@ func handlePacketOutput(pkt *PacketBuffer, conn *conn, gso *GSO, r *Route, dir d return } - netHeader := header.IPv4(pkt.NetworkHeader) - tcpHeader := header.TCP(pkt.TransportHeader) + netHeader := header.IPv4(pkt.NetworkHeader().View()) + tcpHeader := header.TCP(pkt.TransportHeader().View()) // For output redirection, packets going in the original direction // have their destinations modified and replies have their sources @@ -396,8 +396,7 @@ func handlePacketOutput(pkt *PacketBuffer, conn *conn, gso *GSO, r *Route, dir d // Calculate the TCP checksum and set it. tcpHeader.SetChecksum(0) - hdr := &pkt.Header - length := uint16(pkt.Data.Size()+hdr.UsedLength()) - uint16(netHeader.HeaderLength()) + length := uint16(pkt.Size()) - uint16(netHeader.HeaderLength()) xsum := r.PseudoHeaderChecksum(header.TCPProtocolNumber, length) if gso != nil && gso.NeedsCsum { tcpHeader.SetChecksum(xsum) @@ -423,7 +422,7 @@ func (ct *ConnTrack) handlePacket(pkt *PacketBuffer, hook Hook, gso *GSO, r *Rou } // TODO(gvisor.dev/issue/170): Support other transport protocols. - if pkt.NetworkHeader == nil || header.IPv4(pkt.NetworkHeader).TransportProtocol() != header.TCPProtocolNumber { + if nh := pkt.NetworkHeader().View(); nh.IsEmpty() || header.IPv4(nh).TransportProtocol() != header.TCPProtocolNumber { return false } @@ -433,8 +432,8 @@ func (ct *ConnTrack) handlePacket(pkt *PacketBuffer, hook Hook, gso *GSO, r *Rou return true } - tcpHeader := header.TCP(pkt.TransportHeader) - if tcpHeader == nil { + tcpHeader := header.TCP(pkt.TransportHeader().View()) + if len(tcpHeader) < header.TCPMinimumSize { return false } @@ -455,7 +454,7 @@ func (ct *ConnTrack) handlePacket(pkt *PacketBuffer, hook Hook, gso *GSO, r *Rou // Mark the connection as having been used recently so it isn't reaped. conn.lastUsed = time.Now() // Update connection state. - conn.updateLocked(header.TCP(pkt.TransportHeader), hook) + conn.updateLocked(header.TCP(pkt.TransportHeader().View()), hook) return false } @@ -474,7 +473,7 @@ func (ct *ConnTrack) maybeInsertNoop(pkt *PacketBuffer, hook Hook) { } // We only track TCP connections. - if pkt.NetworkHeader == nil || header.IPv4(pkt.NetworkHeader).TransportProtocol() != header.TCPProtocolNumber { + if nh := pkt.NetworkHeader().View(); nh.IsEmpty() || header.IPv4(nh).TransportProtocol() != header.TCPProtocolNumber { return } @@ -486,7 +485,7 @@ func (ct *ConnTrack) maybeInsertNoop(pkt *PacketBuffer, hook Hook) { return } conn := newConn(tid, tid.reply(), manipNone, hook) - conn.updateLocked(header.TCP(pkt.TransportHeader), hook) + conn.updateLocked(header.TCP(pkt.TransportHeader().View()), hook) ct.insertConn(conn) } diff --git a/pkg/tcpip/stack/forwarder_test.go b/pkg/tcpip/stack/forwarder_test.go index c962693f5..944f622fd 100644 --- a/pkg/tcpip/stack/forwarder_test.go +++ b/pkg/tcpip/stack/forwarder_test.go @@ -75,7 +75,7 @@ func (f *fwdTestNetworkEndpoint) ID() *NetworkEndpointID { func (f *fwdTestNetworkEndpoint) HandlePacket(r *Route, pkt *PacketBuffer) { // Dispatch the packet to the transport protocol. - f.dispatcher.DeliverTransportPacket(r, tcpip.TransportProtocolNumber(pkt.NetworkHeader[protocolNumberOffset]), pkt) + f.dispatcher.DeliverTransportPacket(r, tcpip.TransportProtocolNumber(pkt.NetworkHeader().View()[protocolNumberOffset]), pkt) } func (f *fwdTestNetworkEndpoint) MaxHeaderLength() uint16 { @@ -97,7 +97,7 @@ func (f *fwdTestNetworkEndpoint) NetworkProtocolNumber() tcpip.NetworkProtocolNu func (f *fwdTestNetworkEndpoint) WritePacket(r *Route, gso *GSO, params NetworkHeaderParams, pkt *PacketBuffer) *tcpip.Error { // Add the protocol's header to the packet and send it to the link // endpoint. - b := pkt.Header.Prepend(fwdTestNetHeaderLen) + b := pkt.NetworkHeader().Push(fwdTestNetHeaderLen) b[dstAddrOffset] = r.RemoteAddress[0] b[srcAddrOffset] = f.id.LocalAddress[0] b[protocolNumberOffset] = byte(params.Protocol) @@ -144,13 +144,11 @@ func (*fwdTestNetworkProtocol) ParseAddresses(v buffer.View) (src, dst tcpip.Add } func (*fwdTestNetworkProtocol) Parse(pkt *PacketBuffer) (tcpip.TransportProtocolNumber, bool, bool) { - netHeader, ok := pkt.Data.PullUp(fwdTestNetHeaderLen) + netHeader, ok := pkt.NetworkHeader().Consume(fwdTestNetHeaderLen) if !ok { return 0, false, false } - pkt.NetworkHeader = netHeader - pkt.Data.TrimFront(fwdTestNetHeaderLen) - return tcpip.TransportProtocolNumber(pkt.NetworkHeader[protocolNumberOffset]), true, true + return tcpip.TransportProtocolNumber(netHeader[protocolNumberOffset]), true, true } func (f *fwdTestNetworkProtocol) NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache LinkAddressCache, dispatcher TransportDispatcher, ep LinkEndpoint, _ *Stack) (NetworkEndpoint, *tcpip.Error) { @@ -290,7 +288,7 @@ func (e *fwdTestLinkEndpoint) WritePackets(r *Route, gso *GSO, pkts PacketBuffer // WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. func (e *fwdTestLinkEndpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { p := fwdTestPacketInfo{ - Pkt: &PacketBuffer{Data: vv}, + Pkt: NewPacketBuffer(PacketBufferOptions{Data: vv}), } select { @@ -382,9 +380,9 @@ func TestForwardingWithStaticResolver(t *testing.T) { // forwarded to NIC 2. buf := buffer.NewView(30) buf[dstAddrOffset] = 3 - ep1.InjectInbound(fwdTestNetNumber, &PacketBuffer{ + ep1.InjectInbound(fwdTestNetNumber, NewPacketBuffer(PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) var p fwdTestPacketInfo @@ -419,9 +417,9 @@ func TestForwardingWithFakeResolver(t *testing.T) { // forwarded to NIC 2. buf := buffer.NewView(30) buf[dstAddrOffset] = 3 - ep1.InjectInbound(fwdTestNetNumber, &PacketBuffer{ + ep1.InjectInbound(fwdTestNetNumber, NewPacketBuffer(PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) var p fwdTestPacketInfo @@ -450,9 +448,9 @@ func TestForwardingWithNoResolver(t *testing.T) { // forwarded to NIC 2. buf := buffer.NewView(30) buf[dstAddrOffset] = 3 - ep1.InjectInbound(fwdTestNetNumber, &PacketBuffer{ + ep1.InjectInbound(fwdTestNetNumber, NewPacketBuffer(PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) select { case <-ep2.C: @@ -480,17 +478,17 @@ func TestForwardingWithFakeResolverPartialTimeout(t *testing.T) { // not be forwarded. buf := buffer.NewView(30) buf[dstAddrOffset] = 4 - ep1.InjectInbound(fwdTestNetNumber, &PacketBuffer{ + ep1.InjectInbound(fwdTestNetNumber, NewPacketBuffer(PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) // Inject an inbound packet to address 3 on NIC 1, and see if it is // forwarded to NIC 2. buf = buffer.NewView(30) buf[dstAddrOffset] = 3 - ep1.InjectInbound(fwdTestNetNumber, &PacketBuffer{ + ep1.InjectInbound(fwdTestNetNumber, NewPacketBuffer(PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) var p fwdTestPacketInfo @@ -500,8 +498,8 @@ func TestForwardingWithFakeResolverPartialTimeout(t *testing.T) { t.Fatal("packet not forwarded") } - if p.Pkt.NetworkHeader[dstAddrOffset] != 3 { - t.Fatalf("got p.Pkt.NetworkHeader[dstAddrOffset] = %d, want = 3", p.Pkt.NetworkHeader[dstAddrOffset]) + if nh := PayloadSince(p.Pkt.NetworkHeader()); nh[dstAddrOffset] != 3 { + t.Fatalf("got p.Pkt.NetworkHeader[dstAddrOffset] = %d, want = 3", nh[dstAddrOffset]) } // Test that the address resolution happened correctly. @@ -529,9 +527,9 @@ func TestForwardingWithFakeResolverTwoPackets(t *testing.T) { for i := 0; i < 2; i++ { buf := buffer.NewView(30) buf[dstAddrOffset] = 3 - ep1.InjectInbound(fwdTestNetNumber, &PacketBuffer{ + ep1.InjectInbound(fwdTestNetNumber, NewPacketBuffer(PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) } for i := 0; i < 2; i++ { @@ -543,8 +541,8 @@ func TestForwardingWithFakeResolverTwoPackets(t *testing.T) { t.Fatal("packet not forwarded") } - if p.Pkt.NetworkHeader[dstAddrOffset] != 3 { - t.Fatalf("got p.Pkt.NetworkHeader[dstAddrOffset] = %d, want = 3", p.Pkt.NetworkHeader[dstAddrOffset]) + if nh := PayloadSince(p.Pkt.NetworkHeader()); nh[dstAddrOffset] != 3 { + t.Fatalf("got p.Pkt.NetworkHeader[dstAddrOffset] = %d, want = 3", nh[dstAddrOffset]) } // Test that the address resolution happened correctly. @@ -575,9 +573,9 @@ func TestForwardingWithFakeResolverManyPackets(t *testing.T) { buf[dstAddrOffset] = 3 // Set the packet sequence number. binary.BigEndian.PutUint16(buf[fwdTestNetHeaderLen:], uint16(i)) - ep1.InjectInbound(fwdTestNetNumber, &PacketBuffer{ + ep1.InjectInbound(fwdTestNetNumber, NewPacketBuffer(PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) } for i := 0; i < maxPendingPacketsPerResolution; i++ { @@ -589,13 +587,14 @@ func TestForwardingWithFakeResolverManyPackets(t *testing.T) { t.Fatal("packet not forwarded") } - if b := p.Pkt.Header.View(); b[dstAddrOffset] != 3 { + b := PayloadSince(p.Pkt.NetworkHeader()) + if b[dstAddrOffset] != 3 { t.Fatalf("got b[dstAddrOffset] = %d, want = 3", b[dstAddrOffset]) } - seqNumBuf, ok := p.Pkt.Data.PullUp(2) // The sequence number is a uint16 (2 bytes). - if !ok { - t.Fatalf("p.Pkt.Data is too short to hold a sequence number: %d", p.Pkt.Data.Size()) + if len(b) < fwdTestNetHeaderLen+2 { + t.Fatalf("packet is too short to hold a sequence number: len(b) = %d", b) } + seqNumBuf := b[fwdTestNetHeaderLen:] // The first 5 packets should not be forwarded so the sequence number should // start with 5. @@ -632,9 +631,9 @@ func TestForwardingWithFakeResolverManyResolutions(t *testing.T) { // maxPendingResolutions + 7). buf := buffer.NewView(30) buf[dstAddrOffset] = byte(3 + i) - ep1.InjectInbound(fwdTestNetNumber, &PacketBuffer{ + ep1.InjectInbound(fwdTestNetNumber, NewPacketBuffer(PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) } for i := 0; i < maxPendingResolutions; i++ { @@ -648,8 +647,8 @@ func TestForwardingWithFakeResolverManyResolutions(t *testing.T) { // The first 5 packets (address 3 to 7) should not be forwarded // because their address resolutions are interrupted. - if p.Pkt.NetworkHeader[dstAddrOffset] < 8 { - t.Fatalf("got p.Pkt.NetworkHeader[dstAddrOffset] = %d, want p.Pkt.NetworkHeader[dstAddrOffset] >= 8", p.Pkt.NetworkHeader[dstAddrOffset]) + if nh := PayloadSince(p.Pkt.NetworkHeader()); nh[dstAddrOffset] < 8 { + t.Fatalf("got p.Pkt.NetworkHeader[dstAddrOffset] = %d, want p.Pkt.NetworkHeader[dstAddrOffset] >= 8", nh[dstAddrOffset]) } // Test that the address resolution happened correctly. diff --git a/pkg/tcpip/stack/headertype_string.go b/pkg/tcpip/stack/headertype_string.go new file mode 100644 index 000000000..5efddfaaf --- /dev/null +++ b/pkg/tcpip/stack/headertype_string.go @@ -0,0 +1,39 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at // +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Code generated by "stringer -type headerType ."; DO NOT EDIT. + +package stack + +import "strconv" + +func _() { + // An "invalid array index" compiler error signifies that the constant values have changed. + // Re-run the stringer command to generate them again. + var x [1]struct{} + _ = x[linkHeader-0] + _ = x[networkHeader-1] + _ = x[transportHeader-2] + _ = x[numHeaderType-3] +} + +const _headerType_name = "linkHeadernetworkHeadertransportHeadernumHeaderType" + +var _headerType_index = [...]uint8{0, 10, 23, 38, 51} + +func (i headerType) String() string { + if i < 0 || i >= headerType(len(_headerType_index)-1) { + return "headerType(" + strconv.FormatInt(int64(i), 10) + ")" + } + return _headerType_name[_headerType_index[i]:_headerType_index[i+1]] +} diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go index 110ba073d..c37da814f 100644 --- a/pkg/tcpip/stack/iptables.go +++ b/pkg/tcpip/stack/iptables.go @@ -394,7 +394,7 @@ func (it *IPTables) checkRule(hook Hook, pkt *PacketBuffer, table Table, ruleIdx rule := table.Rules[ruleIdx] // Check whether the packet matches the IP header filter. - if !rule.Filter.match(header.IPv4(pkt.NetworkHeader), hook, nicName) { + if !rule.Filter.match(header.IPv4(pkt.NetworkHeader().View()), hook, nicName) { // Continue on to the next rule. return RuleJump, ruleIdx + 1 } diff --git a/pkg/tcpip/stack/iptables_targets.go b/pkg/tcpip/stack/iptables_targets.go index dc88033c7..5f1b2af64 100644 --- a/pkg/tcpip/stack/iptables_targets.go +++ b/pkg/tcpip/stack/iptables_targets.go @@ -99,7 +99,7 @@ func (rt RedirectTarget) Action(pkt *PacketBuffer, ct *ConnTrack, hook Hook, gso } // Drop the packet if network and transport header are not set. - if pkt.NetworkHeader == nil || pkt.TransportHeader == nil { + if pkt.NetworkHeader().View().IsEmpty() || pkt.TransportHeader().View().IsEmpty() { return RuleDrop, 0 } @@ -118,17 +118,16 @@ func (rt RedirectTarget) Action(pkt *PacketBuffer, ct *ConnTrack, hook Hook, gso // TODO(gvisor.dev/issue/170): Check Flags in RedirectTarget if // we need to change dest address (for OUTPUT chain) or ports. - netHeader := header.IPv4(pkt.NetworkHeader) + netHeader := header.IPv4(pkt.NetworkHeader().View()) switch protocol := netHeader.TransportProtocol(); protocol { case header.UDPProtocolNumber: - udpHeader := header.UDP(pkt.TransportHeader) + udpHeader := header.UDP(pkt.TransportHeader().View()) udpHeader.SetDestinationPort(rt.MinPort) // Calculate UDP checksum and set it. if hook == Output { udpHeader.SetChecksum(0) - hdr := &pkt.Header - length := uint16(pkt.Data.Size()+hdr.UsedLength()) - uint16(netHeader.HeaderLength()) + length := uint16(pkt.Size()) - uint16(netHeader.HeaderLength()) // Only calculate the checksum if offloading isn't supported. if r.Capabilities()&CapabilityTXChecksumOffload == 0 { diff --git a/pkg/tcpip/stack/ndp.go b/pkg/tcpip/stack/ndp.go index 5174e639c..93567806b 100644 --- a/pkg/tcpip/stack/ndp.go +++ b/pkg/tcpip/stack/ndp.go @@ -746,12 +746,16 @@ func (ndp *ndpState) sendDADPacket(addr tcpip.Address, ref *referencedNetworkEnd panic(fmt.Sprintf("ndp: route resolution not immediate for route to send NDP NS for DAD (%s -> %s on NIC(%d))", header.IPv6Any, snmc, ndp.nic.ID())) } - hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv6NeighborSolicitMinimumSize) - pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborSolicitMinimumSize)) - pkt.SetType(header.ICMPv6NeighborSolicit) - ns := header.NDPNeighborSolicit(pkt.NDPPayload()) + icmpData := header.ICMPv6(buffer.NewView(header.ICMPv6NeighborSolicitMinimumSize)) + icmpData.SetType(header.ICMPv6NeighborSolicit) + ns := header.NDPNeighborSolicit(icmpData.NDPPayload()) ns.SetTargetAddress(addr) - pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) + icmpData.SetChecksum(header.ICMPv6Checksum(icmpData, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) + + pkt := NewPacketBuffer(PacketBufferOptions{ + ReserveHeaderBytes: int(r.MaxHeaderLength()), + Data: buffer.View(icmpData).ToVectorisedView(), + }) sent := r.Stats().ICMP.V6PacketsSent if err := r.WritePacket(nil, @@ -759,7 +763,7 @@ func (ndp *ndpState) sendDADPacket(addr tcpip.Address, ref *referencedNetworkEnd Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, TOS: DefaultTOS, - }, &PacketBuffer{Header: hdr}, + }, pkt, ); err != nil { sent.Dropped.Increment() return err @@ -1897,12 +1901,16 @@ func (ndp *ndpState) startSolicitingRouters() { } } payloadSize := header.ICMPv6HeaderSize + header.NDPRSMinimumSize + int(optsSerializer.Length()) - hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + payloadSize) - pkt := header.ICMPv6(hdr.Prepend(payloadSize)) - pkt.SetType(header.ICMPv6RouterSolicit) - rs := header.NDPRouterSolicit(pkt.NDPPayload()) + icmpData := header.ICMPv6(buffer.NewView(payloadSize)) + icmpData.SetType(header.ICMPv6RouterSolicit) + rs := header.NDPRouterSolicit(icmpData.NDPPayload()) rs.Options().Serialize(optsSerializer) - pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) + icmpData.SetChecksum(header.ICMPv6Checksum(icmpData, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) + + pkt := NewPacketBuffer(PacketBufferOptions{ + ReserveHeaderBytes: int(r.MaxHeaderLength()), + Data: buffer.View(icmpData).ToVectorisedView(), + }) sent := r.Stats().ICMP.V6PacketsSent if err := r.WritePacket(nil, @@ -1910,7 +1918,7 @@ func (ndp *ndpState) startSolicitingRouters() { Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, TOS: DefaultTOS, - }, &PacketBuffer{Header: hdr}, + }, pkt, ); err != nil { sent.Dropped.Increment() log.Printf("startSolicitingRouters: error writing NDP router solicit message on NIC(%d); err = %s", ndp.nic.ID(), err) diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go index 5d286ccbc..21bf53010 100644 --- a/pkg/tcpip/stack/ndp_test.go +++ b/pkg/tcpip/stack/ndp_test.go @@ -541,7 +541,7 @@ func TestDADResolve(t *testing.T) { // As per RFC 4861 section 4.3, a possible option is the Source Link // Layer option, but this option MUST NOT be included when the source // address of the packet is the unspecified address. - checker.IPv6(t, p.Pkt.Header.View(), + checker.IPv6(t, stack.PayloadSince(p.Pkt.NetworkHeader()), checker.SrcAddr(header.IPv6Any), checker.DstAddr(snmc), checker.TTL(header.NDPHopLimit), @@ -550,8 +550,8 @@ func TestDADResolve(t *testing.T) { checker.NDPNSOptions(nil), )) - if l, want := p.Pkt.Header.AvailableLength(), int(test.linkHeaderLen); l != want { - t.Errorf("got p.Pkt.Header.AvailableLength() = %d; want = %d", l, want) + if l, want := p.Pkt.AvailableHeaderBytes(), int(test.linkHeaderLen); l != want { + t.Errorf("got p.Pkt.AvailableHeaderBytes() = %d; want = %d", l, want) } } }) @@ -667,9 +667,10 @@ func TestDADFail(t *testing.T) { // Receive a packet to simulate multiple nodes owning or // attempting to own the same address. hdr := test.makeBuf(addr1) - e.InjectInbound(header.IPv6ProtocolNumber, &stack.PacketBuffer{ + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: hdr.View().ToVectorisedView(), }) + e.InjectInbound(header.IPv6ProtocolNumber, pkt) stat := test.getStat(s.Stats().ICMP.V6PacketsReceived) if got := stat.Value(); got != 1 { @@ -1024,7 +1025,9 @@ func raBufWithOptsAndDHCPv6(ip tcpip.Address, rl uint16, managedAddress, otherCo DstAddr: header.IPv6AllNodesMulticastAddress, }) - return &stack.PacketBuffer{Data: hdr.View().ToVectorisedView()} + return stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: hdr.View().ToVectorisedView(), + }) } // raBufWithOpts returns a valid NDP Router Advertisement with options. @@ -5134,16 +5137,15 @@ func TestRouterSolicitation(t *testing.T) { t.Errorf("got remote link address = %s, want = %s", p.Route.RemoteLinkAddress, want) } - checker.IPv6(t, - p.Pkt.Header.View(), + checker.IPv6(t, stack.PayloadSince(p.Pkt.NetworkHeader()), checker.SrcAddr(test.expectedSrcAddr), checker.DstAddr(header.IPv6AllRoutersMulticastAddress), checker.TTL(header.NDPHopLimit), checker.NDPRS(checker.NDPRSOptions(test.expectedNDPOpts)), ) - if l, want := p.Pkt.Header.AvailableLength(), int(test.linkHeaderLen); l != want { - t.Errorf("got p.Pkt.Header.AvailableLength() = %d; want = %d", l, want) + if l, want := p.Pkt.AvailableHeaderBytes(), int(test.linkHeaderLen); l != want { + t.Errorf("got p.Pkt.AvailableHeaderBytes() = %d; want = %d", l, want) } } waitForNothing := func(timeout time.Duration) { @@ -5288,7 +5290,7 @@ func TestStopStartSolicitingRouters(t *testing.T) { if p.Proto != header.IPv6ProtocolNumber { t.Fatalf("got Proto = %d, want = %d", p.Proto, header.IPv6ProtocolNumber) } - checker.IPv6(t, p.Pkt.Header.View(), + checker.IPv6(t, stack.PayloadSince(p.Pkt.NetworkHeader()), checker.SrcAddr(header.IPv6Any), checker.DstAddr(header.IPv6AllRoutersMulticastAddress), checker.TTL(header.NDPHopLimit), diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index eaaf756cd..2315ea5b9 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -1299,7 +1299,7 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp } } - src, dst := netProto.ParseAddresses(pkt.NetworkHeader) + src, dst := netProto.ParseAddresses(pkt.NetworkHeader().View()) if n.stack.handleLocal && !n.isLoopback() && n.getRef(protocol, src) != nil { // The source address is one of our own, so we never should have gotten a @@ -1401,24 +1401,19 @@ func (n *NIC) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tc func (n *NIC) forwardPacket(r *Route, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { // TODO(b/143425874) Decrease the TTL field in forwarded packets. - // TODO(b/151227689): Avoid copying the packet when forwarding. We can do this - // by having lower layers explicity write each header instead of just - // pkt.Header. - // pkt may have set its NetworkHeader and TransportHeader. If we're - // forwarding, we'll have to copy them into pkt.Header. - pkt.Header = buffer.NewPrependable(int(n.linkEP.MaxHeaderLength()) + len(pkt.NetworkHeader) + len(pkt.TransportHeader)) - if n := copy(pkt.Header.Prepend(len(pkt.TransportHeader)), pkt.TransportHeader); n != len(pkt.TransportHeader) { - panic(fmt.Sprintf("copied %d bytes, expected %d", n, len(pkt.TransportHeader))) - } - if n := copy(pkt.Header.Prepend(len(pkt.NetworkHeader)), pkt.NetworkHeader); n != len(pkt.NetworkHeader) { - panic(fmt.Sprintf("copied %d bytes, expected %d", n, len(pkt.NetworkHeader))) - } + // pkt may have set its header and may not have enough headroom for link-layer + // header for the other link to prepend. Here we create a new packet to + // forward. + fwdPkt := NewPacketBuffer(PacketBufferOptions{ + ReserveHeaderBytes: int(n.linkEP.MaxHeaderLength()), + Data: buffer.NewVectorisedView(pkt.Size(), pkt.Views()), + }) - // WritePacket takes ownership of pkt, calculate numBytes first. - numBytes := pkt.Header.UsedLength() + pkt.Data.Size() + // WritePacket takes ownership of fwdPkt, calculate numBytes first. + numBytes := fwdPkt.Size() - if err := n.linkEP.WritePacket(r, nil /* gso */, protocol, pkt); err != nil { + if err := n.linkEP.WritePacket(r, nil /* gso */, protocol, fwdPkt); err != nil { r.Stats().IP.OutgoingPacketErrors.Increment() return } @@ -1443,34 +1438,31 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN // validly formed. n.stack.demux.deliverRawPacket(r, protocol, pkt) - // TransportHeader is nil only when pkt is an ICMP packet or was reassembled + // TransportHeader is empty only when pkt is an ICMP packet or was reassembled // from fragments. - if pkt.TransportHeader == nil { + if pkt.TransportHeader().View().IsEmpty() { // TODO(gvisor.dev/issue/170): ICMP packets don't have their TransportHeader // fields set yet, parse it here. See icmp/protocol.go:protocol.Parse for a // full explanation. if protocol == header.ICMPv4ProtocolNumber || protocol == header.ICMPv6ProtocolNumber { // ICMP packets may be longer, but until icmp.Parse is implemented, here // we parse it using the minimum size. - transHeader, ok := pkt.Data.PullUp(transProto.MinimumPacketSize()) - if !ok { + if _, ok := pkt.TransportHeader().Consume(transProto.MinimumPacketSize()); !ok { n.stack.stats.MalformedRcvdPackets.Increment() return } - pkt.TransportHeader = transHeader - pkt.Data.TrimFront(len(pkt.TransportHeader)) } else { // This is either a bad packet or was re-assembled from fragments. transProto.Parse(pkt) } } - if len(pkt.TransportHeader) < transProto.MinimumPacketSize() { + if pkt.TransportHeader().View().Size() < transProto.MinimumPacketSize() { n.stack.stats.MalformedRcvdPackets.Increment() return } - srcPort, dstPort, err := transProto.ParsePorts(pkt.TransportHeader) + srcPort, dstPort, err := transProto.ParsePorts(pkt.TransportHeader().View()) if err != nil { n.stack.stats.MalformedRcvdPackets.Increment() return diff --git a/pkg/tcpip/stack/nic_test.go b/pkg/tcpip/stack/nic_test.go index a70792b50..0870c8d9c 100644 --- a/pkg/tcpip/stack/nic_test.go +++ b/pkg/tcpip/stack/nic_test.go @@ -311,7 +311,9 @@ func TestDisabledRxStatsWhenNICDisabled(t *testing.T) { t.FailNow() } - nic.DeliverNetworkPacket("", "", 0, &PacketBuffer{Data: buffer.View([]byte{1, 2, 3, 4}).ToVectorisedView()}) + nic.DeliverNetworkPacket("", "", 0, NewPacketBuffer(PacketBufferOptions{ + Data: buffer.View([]byte{1, 2, 3, 4}).ToVectorisedView(), + })) if got := nic.stats.DisabledRx.Packets.Value(); got != 1 { t.Errorf("got DisabledRx.Packets = %d, want = 1", got) diff --git a/pkg/tcpip/stack/packet_buffer.go b/pkg/tcpip/stack/packet_buffer.go index 9e871f968..17b8beebb 100644 --- a/pkg/tcpip/stack/packet_buffer.go +++ b/pkg/tcpip/stack/packet_buffer.go @@ -14,16 +14,43 @@ package stack import ( + "fmt" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" ) +type headerType int + +const ( + linkHeader headerType = iota + networkHeader + transportHeader + numHeaderType +) + +// PacketBufferOptions specifies options for PacketBuffer creation. +type PacketBufferOptions struct { + // ReserveHeaderBytes is the number of bytes to reserve for headers. Total + // number of bytes pushed onto the headers must not exceed this value. + ReserveHeaderBytes int + + // Data is the initial unparsed data for the new packet. If set, it will be + // owned by the new packet. + Data buffer.VectorisedView +} + // A PacketBuffer contains all the data of a network packet. // // As a PacketBuffer traverses up the stack, it may be necessary to pass it to -// multiple endpoints. Clone() should be called in such cases so that -// modifications to the Data field do not affect other copies. +// multiple endpoints. +// +// The whole packet is expected to be a series of bytes in the following order: +// LinkHeader, NetworkHeader, TransportHeader, and Data. Any of them can be +// empty. Use of PacketBuffer in any other order is unsupported. +// +// PacketBuffer must be created with NewPacketBuffer. type PacketBuffer struct { _ sync.NoCopy @@ -31,36 +58,27 @@ type PacketBuffer struct { // PacketBuffers. PacketBufferEntry - // Data holds the payload of the packet. For inbound packets, it also - // holds the headers, which are consumed as the packet moves up the - // stack. Headers are guaranteed not to be split across views. + // Data holds the payload of the packet. + // + // For inbound packets, Data is initially the whole packet. Then gets moved to + // headers via PacketHeader.Consume, when the packet is being parsed. // - // The bytes backing Data are immutable, but Data itself may be trimmed - // or otherwise modified. + // For outbound packets, Data is the innermost layer, defined by the protocol. + // Headers are pushed in front of it via PacketHeader.Push. + // + // The bytes backing Data are immutable, a.k.a. users shouldn't write to its + // backing storage. Data buffer.VectorisedView - // Header holds the headers of outbound packets. As a packet is passed - // down the stack, each layer adds to Header. Note that forwarded - // packets don't populate Headers on their way out -- their headers and - // payload are never parsed out and remain in Data. - // - // TODO(gvisor.dev/issue/170): Forwarded packets don't currently - // populate Header, but should. This will be doable once early parsing - // (https://github.com/google/gvisor/pull/1995) is supported. - Header buffer.Prependable + // headers stores metadata about each header. + headers [numHeaderType]headerInfo - // These fields are used by both inbound and outbound packets. They - // typically overlap with the Data and Header fields. + // header is the internal storage for outbound packets. Headers will be pushed + // (prepended) on this storage as the packet is being constructed. // - // The bytes backing these views are immutable. Each field may be nil - // if either it has not been set yet or no such header exists (e.g. - // packets sent via loopback may not have a link header). - // - // These fields may be Views into other slices (either Data or Header). - // SR dosen't support this, so deep copies are necessary in some cases. - LinkHeader buffer.View - NetworkHeader buffer.View - TransportHeader buffer.View + // TODO(gvisor.dev/issue/2404): Switch to an implementation that header and + // data are held in the same underlying buffer storage. + header buffer.Prependable // NetworkProtocol is only valid when NetworkHeader is set. // TODO(gvisor.dev/issue/3574): Remove the separately passed protocol @@ -89,20 +107,137 @@ type PacketBuffer struct { PktType tcpip.PacketType } -// Clone makes a copy of pk. It clones the Data field, which creates a new -// VectorisedView but does not deep copy the underlying bytes. -// -// Clone also does not deep copy any of its other fields. +// NewPacketBuffer creates a new PacketBuffer with opts. +func NewPacketBuffer(opts PacketBufferOptions) *PacketBuffer { + pk := &PacketBuffer{ + Data: opts.Data, + } + if opts.ReserveHeaderBytes != 0 { + pk.header = buffer.NewPrependable(opts.ReserveHeaderBytes) + } + return pk +} + +// ReservedHeaderBytes returns the number of bytes initially reserved for +// headers. +func (pk *PacketBuffer) ReservedHeaderBytes() int { + return pk.header.UsedLength() + pk.header.AvailableLength() +} + +// AvailableHeaderBytes returns the number of bytes currently available for +// headers. This is relevant to PacketHeader.Push method only. +func (pk *PacketBuffer) AvailableHeaderBytes() int { + return pk.header.AvailableLength() +} + +// LinkHeader returns the handle to link-layer header. +func (pk *PacketBuffer) LinkHeader() PacketHeader { + return PacketHeader{ + pk: pk, + typ: linkHeader, + } +} + +// NetworkHeader returns the handle to network-layer header. +func (pk *PacketBuffer) NetworkHeader() PacketHeader { + return PacketHeader{ + pk: pk, + typ: networkHeader, + } +} + +// TransportHeader returns the handle to transport-layer header. +func (pk *PacketBuffer) TransportHeader() PacketHeader { + return PacketHeader{ + pk: pk, + typ: transportHeader, + } +} + +// HeaderSize returns the total size of all headers in bytes. +func (pk *PacketBuffer) HeaderSize() int { + // Note for inbound packets (Consume called), headers are not stored in + // pk.header. Thus, calculation of size of each header is needed. + var size int + for i := range pk.headers { + size += len(pk.headers[i].buf) + } + return size +} + +// Size returns the size of packet in bytes. +func (pk *PacketBuffer) Size() int { + return pk.HeaderSize() + pk.Data.Size() +} + +// Views returns the underlying storage of the whole packet. +func (pk *PacketBuffer) Views() []buffer.View { + // Optimization for outbound packets that headers are in pk.header. + useHeader := true + for i := range pk.headers { + if !canUseHeader(&pk.headers[i]) { + useHeader = false + break + } + } + + dataViews := pk.Data.Views() + + var vs []buffer.View + if useHeader { + vs = make([]buffer.View, 0, 1+len(dataViews)) + vs = append(vs, pk.header.View()) + } else { + vs = make([]buffer.View, 0, len(pk.headers)+len(dataViews)) + for i := range pk.headers { + if v := pk.headers[i].buf; len(v) > 0 { + vs = append(vs, v) + } + } + } + return append(vs, dataViews...) +} + +func canUseHeader(h *headerInfo) bool { + // h.offset will be negative if the header was pushed in to prependable + // portion, or doesn't matter when it's empty. + return len(h.buf) == 0 || h.offset < 0 +} + +func (pk *PacketBuffer) push(typ headerType, size int) buffer.View { + h := &pk.headers[typ] + if h.buf != nil { + panic(fmt.Sprintf("push must not be called twice: type %s", typ)) + } + h.buf = buffer.View(pk.header.Prepend(size)) + h.offset = -pk.header.UsedLength() + return h.buf +} + +func (pk *PacketBuffer) consume(typ headerType, size int) (v buffer.View, consumed bool) { + h := &pk.headers[typ] + if h.buf != nil { + panic(fmt.Sprintf("consume must not be called twice: type %s", typ)) + } + v, ok := pk.Data.PullUp(size) + if !ok { + return + } + pk.Data.TrimFront(size) + h.buf = v + return h.buf, true +} + +// Clone makes a shallow copy of pk. // -// FIXME(b/153685824): Data gets copied but not other header references. +// Clone should be called in such cases so that no modifications is done to +// underlying packet payload. func (pk *PacketBuffer) Clone() *PacketBuffer { - return &PacketBuffer{ + newPk := &PacketBuffer{ PacketBufferEntry: pk.PacketBufferEntry, Data: pk.Data.Clone(nil), - Header: pk.Header, - LinkHeader: pk.LinkHeader, - NetworkHeader: pk.NetworkHeader, - TransportHeader: pk.TransportHeader, + headers: pk.headers, + header: pk.header, Hash: pk.Hash, Owner: pk.Owner, EgressRoute: pk.EgressRoute, @@ -110,4 +245,55 @@ func (pk *PacketBuffer) Clone() *PacketBuffer { NetworkProtocolNumber: pk.NetworkProtocolNumber, NatDone: pk.NatDone, } + return newPk +} + +// headerInfo stores metadata about a header in a packet. +type headerInfo struct { + // buf is the memorized slice for both prepended and consumed header. + // When header is prepended, buf serves as memorized value, which is a slice + // of pk.header. When header is consumed, buf is the slice pulled out from + // pk.Data, which is the only place to hold this header. + buf buffer.View + + // offset will be a negative number denoting the offset where this header is + // from the end of pk.header, if it is prepended. Otherwise, zero. + offset int +} + +// PacketHeader is a handle object to a header in the underlying packet. +type PacketHeader struct { + pk *PacketBuffer + typ headerType +} + +// View returns the underlying storage of h. +func (h PacketHeader) View() buffer.View { + return h.pk.headers[h.typ].buf +} + +// Push pushes size bytes in the front of its residing packet, and returns the +// backing storage. Callers may only call one of Push or Consume once on each +// header in the lifetime of the underlying packet. +func (h PacketHeader) Push(size int) buffer.View { + return h.pk.push(h.typ, size) +} + +// Consume moves the first size bytes of the unparsed data portion in the packet +// to h, and returns the backing storage. In the case of data is shorter than +// size, consumed will be false, and the state of h will not be affected. +// Callers may only call one of Push or Consume once on each header in the +// lifetime of the underlying packet. +func (h PacketHeader) Consume(size int) (v buffer.View, consumed bool) { + return h.pk.consume(h.typ, size) +} + +// PayloadSince returns packet payload starting from and including a particular +// header. This method isn't optimized and should be used in test only. +func PayloadSince(h PacketHeader) buffer.View { + var v buffer.View + for _, hinfo := range h.pk.headers[h.typ:] { + v = append(v, hinfo.buf...) + } + return append(v, h.pk.Data.ToView()...) } diff --git a/pkg/tcpip/stack/packet_buffer_test.go b/pkg/tcpip/stack/packet_buffer_test.go new file mode 100644 index 000000000..c6fa8da5f --- /dev/null +++ b/pkg/tcpip/stack/packet_buffer_test.go @@ -0,0 +1,397 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at // +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package stack + +import ( + "bytes" + "testing" + + "gvisor.dev/gvisor/pkg/tcpip/buffer" +) + +func TestPacketHeaderPush(t *testing.T) { + for _, test := range []struct { + name string + reserved int + link []byte + network []byte + transport []byte + data []byte + }{ + { + name: "construct empty packet", + }, + { + name: "construct link header only packet", + reserved: 60, + link: makeView(10), + }, + { + name: "construct link and network header only packet", + reserved: 60, + link: makeView(10), + network: makeView(20), + }, + { + name: "construct header only packet", + reserved: 60, + link: makeView(10), + network: makeView(20), + transport: makeView(30), + }, + { + name: "construct data only packet", + data: makeView(40), + }, + { + name: "construct L3 packet", + reserved: 60, + network: makeView(20), + transport: makeView(30), + data: makeView(40), + }, + { + name: "construct L2 packet", + reserved: 60, + link: makeView(10), + network: makeView(20), + transport: makeView(30), + data: makeView(40), + }, + } { + t.Run(test.name, func(t *testing.T) { + pk := NewPacketBuffer(PacketBufferOptions{ + ReserveHeaderBytes: test.reserved, + // Make a copy of data to make sure our truth data won't be taint by + // PacketBuffer. + Data: buffer.NewViewFromBytes(test.data).ToVectorisedView(), + }) + + allHdrSize := len(test.link) + len(test.network) + len(test.transport) + + // Check the initial values for packet. + checkInitialPacketBuffer(t, pk, PacketBufferOptions{ + ReserveHeaderBytes: test.reserved, + Data: buffer.View(test.data).ToVectorisedView(), + }) + + // Push headers. + if v := test.transport; len(v) > 0 { + copy(pk.TransportHeader().Push(len(v)), v) + } + if v := test.network; len(v) > 0 { + copy(pk.NetworkHeader().Push(len(v)), v) + } + if v := test.link; len(v) > 0 { + copy(pk.LinkHeader().Push(len(v)), v) + } + + // Check the after values for packet. + if got, want := pk.ReservedHeaderBytes(), test.reserved; got != want { + t.Errorf("After pk.ReservedHeaderBytes() = %d, want %d", got, want) + } + if got, want := pk.AvailableHeaderBytes(), test.reserved-allHdrSize; got != want { + t.Errorf("After pk.AvailableHeaderBytes() = %d, want %d", got, want) + } + if got, want := pk.HeaderSize(), allHdrSize; got != want { + t.Errorf("After pk.HeaderSize() = %d, want %d", got, want) + } + if got, want := pk.Size(), allHdrSize+len(test.data); got != want { + t.Errorf("After pk.Size() = %d, want %d", got, want) + } + checkViewEqual(t, "After pk.Data.Views()", concatViews(pk.Data.Views()...), test.data) + checkViewEqual(t, "After pk.Views()", concatViews(pk.Views()...), + concatViews(test.link, test.network, test.transport, test.data)) + // Check the after values for each header. + checkPacketHeader(t, "After pk.LinkHeader", pk.LinkHeader(), test.link) + checkPacketHeader(t, "After pk.NetworkHeader", pk.NetworkHeader(), test.network) + checkPacketHeader(t, "After pk.TransportHeader", pk.TransportHeader(), test.transport) + // Check the after values for PayloadSince. + checkViewEqual(t, "After PayloadSince(LinkHeader)", + PayloadSince(pk.LinkHeader()), + concatViews(test.link, test.network, test.transport, test.data)) + checkViewEqual(t, "After PayloadSince(NetworkHeader)", + PayloadSince(pk.NetworkHeader()), + concatViews(test.network, test.transport, test.data)) + checkViewEqual(t, "After PayloadSince(TransportHeader)", + PayloadSince(pk.TransportHeader()), + concatViews(test.transport, test.data)) + }) + } +} + +func TestPacketHeaderConsume(t *testing.T) { + for _, test := range []struct { + name string + data []byte + link int + network int + transport int + }{ + { + name: "parse L2 packet", + data: concatViews(makeView(10), makeView(20), makeView(30), makeView(40)), + link: 10, + network: 20, + transport: 30, + }, + { + name: "parse L3 packet", + data: concatViews(makeView(20), makeView(30), makeView(40)), + network: 20, + transport: 30, + }, + } { + t.Run(test.name, func(t *testing.T) { + pk := NewPacketBuffer(PacketBufferOptions{ + // Make a copy of data to make sure our truth data won't be taint by + // PacketBuffer. + Data: buffer.NewViewFromBytes(test.data).ToVectorisedView(), + }) + + // Check the initial values for packet. + checkInitialPacketBuffer(t, pk, PacketBufferOptions{ + Data: buffer.View(test.data).ToVectorisedView(), + }) + + // Consume headers. + if size := test.link; size > 0 { + if _, ok := pk.LinkHeader().Consume(size); !ok { + t.Fatalf("pk.LinkHeader().Consume() = false, want true") + } + } + if size := test.network; size > 0 { + if _, ok := pk.NetworkHeader().Consume(size); !ok { + t.Fatalf("pk.NetworkHeader().Consume() = false, want true") + } + } + if size := test.transport; size > 0 { + if _, ok := pk.TransportHeader().Consume(size); !ok { + t.Fatalf("pk.TransportHeader().Consume() = false, want true") + } + } + + allHdrSize := test.link + test.network + test.transport + + // Check the after values for packet. + if got, want := pk.ReservedHeaderBytes(), 0; got != want { + t.Errorf("After pk.ReservedHeaderBytes() = %d, want %d", got, want) + } + if got, want := pk.AvailableHeaderBytes(), 0; got != want { + t.Errorf("After pk.AvailableHeaderBytes() = %d, want %d", got, want) + } + if got, want := pk.HeaderSize(), allHdrSize; got != want { + t.Errorf("After pk.HeaderSize() = %d, want %d", got, want) + } + if got, want := pk.Size(), len(test.data); got != want { + t.Errorf("After pk.Size() = %d, want %d", got, want) + } + // After state of pk. + var ( + link = test.data[:test.link] + network = test.data[test.link:][:test.network] + transport = test.data[test.link+test.network:][:test.transport] + payload = test.data[allHdrSize:] + ) + checkViewEqual(t, "After pk.Data.Views()", concatViews(pk.Data.Views()...), payload) + checkViewEqual(t, "After pk.Views()", concatViews(pk.Views()...), test.data) + // Check the after values for each header. + checkPacketHeader(t, "After pk.LinkHeader", pk.LinkHeader(), link) + checkPacketHeader(t, "After pk.NetworkHeader", pk.NetworkHeader(), network) + checkPacketHeader(t, "After pk.TransportHeader", pk.TransportHeader(), transport) + // Check the after values for PayloadSince. + checkViewEqual(t, "After PayloadSince(LinkHeader)", + PayloadSince(pk.LinkHeader()), + concatViews(link, network, transport, payload)) + checkViewEqual(t, "After PayloadSince(NetworkHeader)", + PayloadSince(pk.NetworkHeader()), + concatViews(network, transport, payload)) + checkViewEqual(t, "After PayloadSince(TransportHeader)", + PayloadSince(pk.TransportHeader()), + concatViews(transport, payload)) + }) + } +} + +func TestPacketHeaderConsumeDataTooShort(t *testing.T) { + data := makeView(10) + + pk := NewPacketBuffer(PacketBufferOptions{ + // Make a copy of data to make sure our truth data won't be taint by + // PacketBuffer. + Data: buffer.NewViewFromBytes(data).ToVectorisedView(), + }) + + // Consume should fail if pkt.Data is too short. + if _, ok := pk.LinkHeader().Consume(11); ok { + t.Fatalf("pk.LinkHeader().Consume() = _, true; want _, false") + } + if _, ok := pk.NetworkHeader().Consume(11); ok { + t.Fatalf("pk.NetworkHeader().Consume() = _, true; want _, false") + } + if _, ok := pk.TransportHeader().Consume(11); ok { + t.Fatalf("pk.TransportHeader().Consume() = _, true; want _, false") + } + + // Check packet should look the same as initial packet. + checkInitialPacketBuffer(t, pk, PacketBufferOptions{ + Data: buffer.View(data).ToVectorisedView(), + }) +} + +func TestPacketHeaderPushCalledAtMostOnce(t *testing.T) { + const headerSize = 10 + + pk := NewPacketBuffer(PacketBufferOptions{ + ReserveHeaderBytes: headerSize * int(numHeaderType), + }) + + for _, h := range []PacketHeader{ + pk.TransportHeader(), + pk.NetworkHeader(), + pk.LinkHeader(), + } { + t.Run("PushedTwice/"+h.typ.String(), func(t *testing.T) { + h.Push(headerSize) + + defer func() { recover() }() + h.Push(headerSize) + t.Fatal("Second push should have panicked") + }) + } +} + +func TestPacketHeaderConsumeCalledAtMostOnce(t *testing.T) { + const headerSize = 10 + + pk := NewPacketBuffer(PacketBufferOptions{ + Data: makeView(headerSize * int(numHeaderType)).ToVectorisedView(), + }) + + for _, h := range []PacketHeader{ + pk.LinkHeader(), + pk.NetworkHeader(), + pk.TransportHeader(), + } { + t.Run("ConsumedTwice/"+h.typ.String(), func(t *testing.T) { + if _, ok := h.Consume(headerSize); !ok { + t.Fatal("First consume should succeed") + } + + defer func() { recover() }() + h.Consume(headerSize) + t.Fatal("Second consume should have panicked") + }) + } +} + +func TestPacketHeaderPushThenConsumePanics(t *testing.T) { + const headerSize = 10 + + pk := NewPacketBuffer(PacketBufferOptions{ + ReserveHeaderBytes: headerSize * int(numHeaderType), + }) + + for _, h := range []PacketHeader{ + pk.TransportHeader(), + pk.NetworkHeader(), + pk.LinkHeader(), + } { + t.Run(h.typ.String(), func(t *testing.T) { + h.Push(headerSize) + + defer func() { recover() }() + h.Consume(headerSize) + t.Fatal("Consume should have panicked") + }) + } +} + +func TestPacketHeaderConsumeThenPushPanics(t *testing.T) { + const headerSize = 10 + + pk := NewPacketBuffer(PacketBufferOptions{ + Data: makeView(headerSize * int(numHeaderType)).ToVectorisedView(), + }) + + for _, h := range []PacketHeader{ + pk.LinkHeader(), + pk.NetworkHeader(), + pk.TransportHeader(), + } { + t.Run(h.typ.String(), func(t *testing.T) { + h.Consume(headerSize) + + defer func() { recover() }() + h.Push(headerSize) + t.Fatal("Push should have panicked") + }) + } +} + +func checkInitialPacketBuffer(t *testing.T, pk *PacketBuffer, opts PacketBufferOptions) { + t.Helper() + reserved := opts.ReserveHeaderBytes + if got, want := pk.ReservedHeaderBytes(), reserved; got != want { + t.Errorf("Initial pk.ReservedHeaderBytes() = %d, want %d", got, want) + } + if got, want := pk.AvailableHeaderBytes(), reserved; got != want { + t.Errorf("Initial pk.AvailableHeaderBytes() = %d, want %d", got, want) + } + if got, want := pk.HeaderSize(), 0; got != want { + t.Errorf("Initial pk.HeaderSize() = %d, want %d", got, want) + } + data := opts.Data.ToView() + if got, want := pk.Size(), len(data); got != want { + t.Errorf("Initial pk.Size() = %d, want %d", got, want) + } + checkViewEqual(t, "Initial pk.Data.Views()", concatViews(pk.Data.Views()...), data) + checkViewEqual(t, "Initial pk.Views()", concatViews(pk.Views()...), data) + // Check the initial values for each header. + checkPacketHeader(t, "Initial pk.LinkHeader", pk.LinkHeader(), nil) + checkPacketHeader(t, "Initial pk.NetworkHeader", pk.NetworkHeader(), nil) + checkPacketHeader(t, "Initial pk.TransportHeader", pk.TransportHeader(), nil) + // Check the initial valies for PayloadSince. + checkViewEqual(t, "Initial PayloadSince(LinkHeader)", + PayloadSince(pk.LinkHeader()), data) + checkViewEqual(t, "Initial PayloadSince(NetworkHeader)", + PayloadSince(pk.NetworkHeader()), data) + checkViewEqual(t, "Initial PayloadSince(TransportHeader)", + PayloadSince(pk.TransportHeader()), data) +} + +func checkPacketHeader(t *testing.T, name string, h PacketHeader, want []byte) { + t.Helper() + checkViewEqual(t, name+".View()", h.View(), want) +} + +func checkViewEqual(t *testing.T, what string, got, want buffer.View) { + t.Helper() + if !bytes.Equal(got, want) { + t.Errorf("%s = %x, want %x", what, got, want) + } +} + +func makeView(size int) buffer.View { + b := byte(size) + return bytes.Repeat([]byte{b}, size) +} + +func concatViews(views ...buffer.View) buffer.View { + var all buffer.View + for _, v := range views { + all = append(all, v...) + } + return all +} diff --git a/pkg/tcpip/stack/route.go b/pkg/tcpip/stack/route.go index 9ce0a2c22..e267bebb0 100644 --- a/pkg/tcpip/stack/route.go +++ b/pkg/tcpip/stack/route.go @@ -173,7 +173,7 @@ func (r *Route) WritePacket(gso *GSO, params NetworkHeaderParams, pkt *PacketBuf } // WritePacket takes ownership of pkt, calculate numBytes first. - numBytes := pkt.Header.UsedLength() + pkt.Data.Size() + numBytes := pkt.Size() err := r.ref.ep.WritePacket(r, gso, params, pkt) if err != nil { @@ -203,8 +203,7 @@ func (r *Route) WritePackets(gso *GSO, pkts PacketBufferList, params NetworkHead writtenBytes := 0 for i, pb := 0, pkts.Front(); i < n && pb != nil; i, pb = i+1, pb.Next() { - writtenBytes += pb.Header.UsedLength() - writtenBytes += pb.Data.Size() + writtenBytes += pb.Size() } r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(writtenBytes)) diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index fe1c1b8a4..0273b3c63 100644 --- a/pkg/tcpip/stack/stack_test.go +++ b/pkg/tcpip/stack/stack_test.go @@ -102,7 +102,7 @@ func (f *fakeNetworkEndpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuff f.proto.packetCount[int(f.id.LocalAddress[0])%len(f.proto.packetCount)]++ // Handle control packets. - if pkt.NetworkHeader[protocolNumberOffset] == uint8(fakeControlProtocol) { + if pkt.NetworkHeader().View()[protocolNumberOffset] == uint8(fakeControlProtocol) { nb, ok := pkt.Data.PullUp(fakeNetHeaderLen) if !ok { return @@ -118,7 +118,7 @@ func (f *fakeNetworkEndpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuff } // Dispatch the packet to the transport protocol. - f.dispatcher.DeliverTransportPacket(r, tcpip.TransportProtocolNumber(pkt.NetworkHeader[protocolNumberOffset]), pkt) + f.dispatcher.DeliverTransportPacket(r, tcpip.TransportProtocolNumber(pkt.NetworkHeader().View()[protocolNumberOffset]), pkt) } func (f *fakeNetworkEndpoint) MaxHeaderLength() uint16 { @@ -143,10 +143,10 @@ func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, params // Add the protocol's header to the packet and send it to the link // endpoint. - pkt.NetworkHeader = pkt.Header.Prepend(fakeNetHeaderLen) - pkt.NetworkHeader[dstAddrOffset] = r.RemoteAddress[0] - pkt.NetworkHeader[srcAddrOffset] = f.id.LocalAddress[0] - pkt.NetworkHeader[protocolNumberOffset] = byte(params.Protocol) + hdr := pkt.NetworkHeader().Push(fakeNetHeaderLen) + hdr[dstAddrOffset] = r.RemoteAddress[0] + hdr[srcAddrOffset] = f.id.LocalAddress[0] + hdr[protocolNumberOffset] = byte(params.Protocol) if r.Loop&stack.PacketLoop != 0 { f.HandlePacket(r, pkt) @@ -249,12 +249,10 @@ func (*fakeNetworkProtocol) Wait() {} // Parse implements TransportProtocol.Parse. func (*fakeNetworkProtocol) Parse(pkt *stack.PacketBuffer) (tcpip.TransportProtocolNumber, bool, bool) { - hdr, ok := pkt.Data.PullUp(fakeNetHeaderLen) + hdr, ok := pkt.NetworkHeader().Consume(fakeNetHeaderLen) if !ok { return 0, false, false } - pkt.NetworkHeader = hdr - pkt.Data.TrimFront(fakeNetHeaderLen) return tcpip.TransportProtocolNumber(hdr[protocolNumberOffset]), true, true } @@ -315,9 +313,9 @@ func TestNetworkReceive(t *testing.T) { // Make sure packet with wrong address is not delivered. buf[dstAddrOffset] = 3 - ep.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ + ep.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) if fakeNet.packetCount[1] != 0 { t.Errorf("packetCount[1] = %d, want %d", fakeNet.packetCount[1], 0) } @@ -327,9 +325,9 @@ func TestNetworkReceive(t *testing.T) { // Make sure packet is delivered to first endpoint. buf[dstAddrOffset] = 1 - ep.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ + ep.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) if fakeNet.packetCount[1] != 1 { t.Errorf("packetCount[1] = %d, want %d", fakeNet.packetCount[1], 1) } @@ -339,9 +337,9 @@ func TestNetworkReceive(t *testing.T) { // Make sure packet is delivered to second endpoint. buf[dstAddrOffset] = 2 - ep.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ + ep.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) if fakeNet.packetCount[1] != 1 { t.Errorf("packetCount[1] = %d, want %d", fakeNet.packetCount[1], 1) } @@ -350,9 +348,9 @@ func TestNetworkReceive(t *testing.T) { } // Make sure packet is not delivered if protocol number is wrong. - ep.InjectInbound(fakeNetNumber-1, &stack.PacketBuffer{ + ep.InjectInbound(fakeNetNumber-1, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) if fakeNet.packetCount[1] != 1 { t.Errorf("packetCount[1] = %d, want %d", fakeNet.packetCount[1], 1) } @@ -362,9 +360,9 @@ func TestNetworkReceive(t *testing.T) { // Make sure packet that is too small is dropped. buf.CapLength(2) - ep.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ + ep.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) if fakeNet.packetCount[1] != 1 { t.Errorf("packetCount[1] = %d, want %d", fakeNet.packetCount[1], 1) } @@ -383,11 +381,10 @@ func sendTo(s *stack.Stack, addr tcpip.Address, payload buffer.View) *tcpip.Erro } func send(r stack.Route, payload buffer.View) *tcpip.Error { - hdr := buffer.NewPrependable(int(r.MaxHeaderLength())) - return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}, &stack.PacketBuffer{ - Header: hdr, - Data: payload.ToVectorisedView(), - }) + return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}, stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(r.MaxHeaderLength()), + Data: payload.ToVectorisedView(), + })) } func testSendTo(t *testing.T, s *stack.Stack, addr tcpip.Address, ep *channel.Endpoint, payload buffer.View) { @@ -442,9 +439,9 @@ func testFailingRecv(t *testing.T, fakeNet *fakeNetworkProtocol, localAddrByte b func testRecvInternal(t *testing.T, fakeNet *fakeNetworkProtocol, localAddrByte byte, ep *channel.Endpoint, buf buffer.View, want int) { t.Helper() - ep.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ + ep.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) if got := fakeNet.PacketCount(localAddrByte); got != want { t.Errorf("receive packet count: got = %d, want %d", got, want) } @@ -2285,9 +2282,9 @@ func TestNICStats(t *testing.T) { // Send a packet to address 1. buf := buffer.NewView(30) - ep1.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ + ep1.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) if got, want := s.NICInfo()[1].Stats.Rx.Packets.Value(), uint64(1); got != want { t.Errorf("got Rx.Packets.Value() = %d, want = %d", got, want) } @@ -2367,9 +2364,9 @@ func TestNICForwarding(t *testing.T) { // Send a packet to dstAddr. buf := buffer.NewView(30) buf[dstAddrOffset] = dstAddr[0] - ep1.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ + ep1.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) pkt, ok := ep2.Read() if !ok { @@ -2377,8 +2374,8 @@ func TestNICForwarding(t *testing.T) { } // Test that the link's MaxHeaderLength is honoured. - if capacity, want := pkt.Pkt.Header.AvailableLength(), int(test.headerLen); capacity != want { - t.Errorf("got Header.AvailableLength() = %d, want = %d", capacity, want) + if capacity, want := pkt.Pkt.AvailableHeaderBytes(), int(test.headerLen); capacity != want { + t.Errorf("got LinkHeader.AvailableLength() = %d, want = %d", capacity, want) } // Test that forwarding increments Tx stats correctly. diff --git a/pkg/tcpip/stack/transport_demuxer_test.go b/pkg/tcpip/stack/transport_demuxer_test.go index 73dada928..1339edc2d 100644 --- a/pkg/tcpip/stack/transport_demuxer_test.go +++ b/pkg/tcpip/stack/transport_demuxer_test.go @@ -128,11 +128,10 @@ func (c *testContext) sendV4Packet(payload []byte, h *headers, linkEpID tcpip.NI u.SetChecksum(^u.CalculateChecksum(xsum)) // Inject packet. - c.linkEps[linkEpID].InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ - Data: buf.ToVectorisedView(), - NetworkHeader: buffer.View(ip), - TransportHeader: buffer.View(u), + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: buf.ToVectorisedView(), }) + c.linkEps[linkEpID].InjectInbound(ipv4.ProtocolNumber, pkt) } func (c *testContext) sendV6Packet(payload []byte, h *headers, linkEpID tcpip.NICID) { @@ -166,11 +165,10 @@ func (c *testContext) sendV6Packet(payload []byte, h *headers, linkEpID tcpip.NI u.SetChecksum(^u.CalculateChecksum(xsum)) // Inject packet. - c.linkEps[linkEpID].InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ - Data: buf.ToVectorisedView(), - NetworkHeader: buffer.View(ip), - TransportHeader: buffer.View(u), + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: buf.ToVectorisedView(), }) + c.linkEps[linkEpID].InjectInbound(ipv6.ProtocolNumber, pkt) } func TestTransportDemuxerRegister(t *testing.T) { diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go index 7e8b84867..6c6e44468 100644 --- a/pkg/tcpip/stack/transport_test.go +++ b/pkg/tcpip/stack/transport_test.go @@ -84,16 +84,16 @@ func (f *fakeTransportEndpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions return 0, nil, tcpip.ErrNoRoute } - hdr := buffer.NewPrependable(int(f.route.MaxHeaderLength()) + fakeTransHeaderLen) - hdr.Prepend(fakeTransHeaderLen) v, err := p.FullPayload() if err != nil { return 0, nil, err } - if err := f.route.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}, &stack.PacketBuffer{ - Header: hdr, - Data: buffer.View(v).ToVectorisedView(), - }); err != nil { + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(f.route.MaxHeaderLength()) + fakeTransHeaderLen, + Data: buffer.View(v).ToVectorisedView(), + }) + _ = pkt.TransportHeader().Push(fakeTransHeaderLen) + if err := f.route.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}, pkt); err != nil { return 0, nil, err } @@ -328,13 +328,8 @@ func (*fakeTransportProtocol) Wait() {} // Parse implements TransportProtocol.Parse. func (*fakeTransportProtocol) Parse(pkt *stack.PacketBuffer) bool { - hdr, ok := pkt.Data.PullUp(fakeTransHeaderLen) - if !ok { - return false - } - pkt.TransportHeader = hdr - pkt.Data.TrimFront(fakeTransHeaderLen) - return true + _, ok := pkt.TransportHeader().Consume(fakeTransHeaderLen) + return ok } func fakeTransFactory() stack.TransportProtocol { @@ -382,9 +377,9 @@ func TestTransportReceive(t *testing.T) { // Make sure packet with wrong protocol is not delivered. buf[0] = 1 buf[2] = 0 - linkEP.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ + linkEP.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) if fakeTrans.packetCount != 0 { t.Errorf("packetCount = %d, want %d", fakeTrans.packetCount, 0) } @@ -393,9 +388,9 @@ func TestTransportReceive(t *testing.T) { buf[0] = 1 buf[1] = 3 buf[2] = byte(fakeTransNumber) - linkEP.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ + linkEP.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) if fakeTrans.packetCount != 0 { t.Errorf("packetCount = %d, want %d", fakeTrans.packetCount, 0) } @@ -404,9 +399,9 @@ func TestTransportReceive(t *testing.T) { buf[0] = 1 buf[1] = 2 buf[2] = byte(fakeTransNumber) - linkEP.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ + linkEP.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) if fakeTrans.packetCount != 1 { t.Errorf("packetCount = %d, want %d", fakeTrans.packetCount, 1) } @@ -459,9 +454,9 @@ func TestTransportControlReceive(t *testing.T) { buf[fakeNetHeaderLen+0] = 0 buf[fakeNetHeaderLen+1] = 1 buf[fakeNetHeaderLen+2] = 0 - linkEP.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ + linkEP.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) if fakeTrans.controlCount != 0 { t.Errorf("controlCount = %d, want %d", fakeTrans.controlCount, 0) } @@ -470,9 +465,9 @@ func TestTransportControlReceive(t *testing.T) { buf[fakeNetHeaderLen+0] = 3 buf[fakeNetHeaderLen+1] = 1 buf[fakeNetHeaderLen+2] = byte(fakeTransNumber) - linkEP.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ + linkEP.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) if fakeTrans.controlCount != 0 { t.Errorf("controlCount = %d, want %d", fakeTrans.controlCount, 0) } @@ -481,9 +476,9 @@ func TestTransportControlReceive(t *testing.T) { buf[fakeNetHeaderLen+0] = 2 buf[fakeNetHeaderLen+1] = 1 buf[fakeNetHeaderLen+2] = byte(fakeTransNumber) - linkEP.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ + linkEP.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) if fakeTrans.controlCount != 1 { t.Errorf("controlCount = %d, want %d", fakeTrans.controlCount, 1) } @@ -636,9 +631,9 @@ func TestTransportForwarding(t *testing.T) { req[0] = 1 req[1] = 3 req[2] = byte(fakeTransNumber) - ep2.InjectInbound(fakeNetNumber, &stack.PacketBuffer{ + ep2.InjectInbound(fakeNetNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: req.ToVectorisedView(), - }) + })) aep, _, err := ep.Accept() if err != nil || aep == nil { @@ -655,10 +650,11 @@ func TestTransportForwarding(t *testing.T) { t.Fatal("Response packet not forwarded") } - if dst := p.Pkt.NetworkHeader[0]; dst != 3 { + nh := stack.PayloadSince(p.Pkt.NetworkHeader()) + if dst := nh[0]; dst != 3 { t.Errorf("Response packet has incorrect destination addresss: got = %d, want = 3", dst) } - if src := p.Pkt.NetworkHeader[1]; src != 1 { + if src := nh[1]; src != 1 { t.Errorf("Response packet has incorrect source addresss: got = %d, want = 3", src) } } diff --git a/pkg/tcpip/tests/integration/multicast_broadcast_test.go b/pkg/tcpip/tests/integration/multicast_broadcast_test.go index 0ff3a2b89..9f0dd4d6d 100644 --- a/pkg/tcpip/tests/integration/multicast_broadcast_test.go +++ b/pkg/tcpip/tests/integration/multicast_broadcast_test.go @@ -80,9 +80,9 @@ func TestPingMulticastBroadcast(t *testing.T) { DstAddr: dst, }) - e.InjectInbound(header.IPv4ProtocolNumber, &stack.PacketBuffer{ + e.InjectInbound(header.IPv4ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: hdr.View().ToVectorisedView(), - }) + })) } rxIPv6ICMP := func(e *channel.Endpoint, dst tcpip.Address) { @@ -102,9 +102,9 @@ func TestPingMulticastBroadcast(t *testing.T) { DstAddr: dst, }) - e.InjectInbound(header.IPv6ProtocolNumber, &stack.PacketBuffer{ + e.InjectInbound(header.IPv6ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: hdr.View().ToVectorisedView(), - }) + })) } tests := []struct { @@ -204,7 +204,7 @@ func TestPingMulticastBroadcast(t *testing.T) { t.Errorf("got pkt.Route.RemoteAddress = %s, want = %s", pkt.Route.RemoteAddress, expectedDst) } - src, dst := proto.ParseAddresses(pkt.Pkt.NetworkHeader) + src, dst := proto.ParseAddresses(pkt.Pkt.NetworkHeader().View()) if src != expectedSrc { t.Errorf("got pkt source = %s, want = %s", src, expectedSrc) } @@ -252,9 +252,9 @@ func TestIncomingMulticastAndBroadcast(t *testing.T) { DstAddr: dst, }) - e.InjectInbound(header.IPv4ProtocolNumber, &stack.PacketBuffer{ + e.InjectInbound(header.IPv4ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: hdr.View().ToVectorisedView(), - }) + })) } rxIPv6UDP := func(e *channel.Endpoint, dst tcpip.Address) { @@ -280,9 +280,9 @@ func TestIncomingMulticastAndBroadcast(t *testing.T) { DstAddr: dst, }) - e.InjectInbound(header.IPv6ProtocolNumber, &stack.PacketBuffer{ + e.InjectInbound(header.IPv6ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: hdr.View().ToVectorisedView(), - }) + })) } tests := []struct { diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index 4612be4e7..bd6f49eb8 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -430,9 +430,12 @@ func send4(r *stack.Route, ident uint16, data buffer.View, ttl uint8, owner tcpi return tcpip.ErrInvalidEndpointState } - hdr := buffer.NewPrependable(header.ICMPv4MinimumSize + int(r.MaxHeaderLength())) + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: header.ICMPv4MinimumSize + int(r.MaxHeaderLength()), + }) + pkt.Owner = owner - icmpv4 := header.ICMPv4(hdr.Prepend(header.ICMPv4MinimumSize)) + icmpv4 := header.ICMPv4(pkt.TransportHeader().Push(header.ICMPv4MinimumSize)) copy(icmpv4, data) // Set the ident to the user-specified port. Sequence number should // already be set by the user. @@ -447,15 +450,12 @@ func send4(r *stack.Route, ident uint16, data buffer.View, ttl uint8, owner tcpi icmpv4.SetChecksum(0) icmpv4.SetChecksum(^header.Checksum(icmpv4, header.Checksum(data, 0))) + pkt.Data = data.ToVectorisedView() + if ttl == 0 { ttl = r.DefaultTTL() } - return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}, &stack.PacketBuffer{ - Header: hdr, - Data: data.ToVectorisedView(), - TransportHeader: buffer.View(icmpv4), - Owner: owner, - }) + return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}, pkt) } func send6(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Error { @@ -463,9 +463,11 @@ func send6(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Err return tcpip.ErrInvalidEndpointState } - hdr := buffer.NewPrependable(header.ICMPv6MinimumSize + int(r.MaxHeaderLength())) + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: header.ICMPv6MinimumSize + int(r.MaxHeaderLength()), + }) - icmpv6 := header.ICMPv6(hdr.Prepend(header.ICMPv6MinimumSize)) + icmpv6 := header.ICMPv6(pkt.TransportHeader().Push(header.ICMPv6MinimumSize)) copy(icmpv6, data) // Set the ident. Sequence number is provided by the user. icmpv6.SetIdent(ident) @@ -477,15 +479,12 @@ func send6(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Err dataVV := data.ToVectorisedView() icmpv6.SetChecksum(header.ICMPv6Checksum(icmpv6, r.LocalAddress, r.RemoteAddress, dataVV)) + pkt.Data = dataVV if ttl == 0 { ttl = r.DefaultTTL() } - return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}, &stack.PacketBuffer{ - Header: hdr, - Data: dataVV, - TransportHeader: buffer.View(icmpv6), - }) + return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: ttl, TOS: stack.DefaultTOS}, pkt) } // checkV4MappedLocked determines the effective network protocol and converts @@ -748,14 +747,18 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk // Only accept echo replies. switch e.NetProto { case header.IPv4ProtocolNumber: - h := header.ICMPv4(pkt.TransportHeader) + h := header.ICMPv4(pkt.TransportHeader().View()) + // TODO(b/129292233): Determine if len(h) check is still needed after early + // parsing. if len(h) < header.ICMPv4MinimumSize || h.Type() != header.ICMPv4EchoReply { e.stack.Stats().DroppedPackets.Increment() e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() return } case header.IPv6ProtocolNumber: - h := header.ICMPv6(pkt.TransportHeader) + h := header.ICMPv6(pkt.TransportHeader().View()) + // TODO(b/129292233): Determine if len(h) check is still needed after early + // parsing. if len(h) < header.ICMPv6MinimumSize || h.Type() != header.ICMPv6EchoReply { e.stack.Stats().DroppedPackets.Increment() e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() @@ -791,7 +794,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk } // ICMP socket's data includes ICMP header. - packet.data = pkt.TransportHeader.ToVectorisedView() + packet.data = pkt.TransportHeader().View().ToVectorisedView() packet.data.Append(pkt.Data) e.rcvList.PushBack(packet) diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go index df478115d..1b03ad6bb 100644 --- a/pkg/tcpip/transport/packet/endpoint.go +++ b/pkg/tcpip/transport/packet/endpoint.go @@ -433,9 +433,9 @@ func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress, // Push new packet into receive list and increment the buffer size. var packet packet // TODO(gvisor.dev/issue/173): Return network protocol. - if len(pkt.LinkHeader) > 0 { + if !pkt.LinkHeader().View().IsEmpty() { // Get info directly from the ethernet header. - hdr := header.Ethernet(pkt.LinkHeader) + hdr := header.Ethernet(pkt.LinkHeader().View()) packet.senderAddr = tcpip.FullAddress{ NIC: nicID, Addr: tcpip.Address(hdr.SourceAddress()), @@ -458,9 +458,14 @@ func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress, case tcpip.PacketHost: packet.data = pkt.Data case tcpip.PacketOutgoing: - // Strip Link Header from the Header. - pkt.Header = buffer.NewPrependableFromView(pkt.Header.View()[len(pkt.LinkHeader):]) - combinedVV := pkt.Header.View().ToVectorisedView() + // Strip Link Header. + var combinedVV buffer.VectorisedView + if v := pkt.NetworkHeader().View(); !v.IsEmpty() { + combinedVV.AppendView(v) + } + if v := pkt.TransportHeader().View(); !v.IsEmpty() { + combinedVV.AppendView(v) + } combinedVV.Append(pkt.Data) packet.data = combinedVV default: @@ -471,9 +476,8 @@ func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress, // Raw packets need their ethernet headers prepended before // queueing. var linkHeader buffer.View - var combinedVV buffer.VectorisedView if pkt.PktType != tcpip.PacketOutgoing { - if len(pkt.LinkHeader) == 0 { + if pkt.LinkHeader().View().IsEmpty() { // We weren't provided with an actual ethernet header, // so fake one. ethFields := header.EthernetFields{ @@ -485,19 +489,14 @@ func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress, fakeHeader.Encode(ðFields) linkHeader = buffer.View(fakeHeader) } else { - linkHeader = append(buffer.View(nil), pkt.LinkHeader...) + linkHeader = append(buffer.View(nil), pkt.LinkHeader().View()...) } - combinedVV = linkHeader.ToVectorisedView() - } - if pkt.PktType == tcpip.PacketOutgoing { - // For outgoing packets the Link, Network and Transport - // headers are in the pkt.Header fields normally unless - // a Raw socket is in use in which case pkt.Header could - // be nil. - combinedVV.AppendView(pkt.Header.View()) + combinedVV := linkHeader.ToVectorisedView() + combinedVV.Append(pkt.Data) + packet.data = combinedVV + } else { + packet.data = buffer.NewVectorisedView(pkt.Size(), pkt.Views()) } - combinedVV.Append(pkt.Data) - packet.data = combinedVV } packet.timestampNS = ep.stack.Clock().NowNanoseconds() diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index f85a68554..edc2b5b61 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -352,18 +352,23 @@ func (e *endpoint) finishWrite(payloadBytes []byte, route *stack.Route) (int64, } if e.hdrIncluded { - if err := route.WriteHeaderIncludedPacket(&stack.PacketBuffer{ + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buffer.View(payloadBytes).ToVectorisedView(), - }); err != nil { + }) + if err := route.WriteHeaderIncludedPacket(pkt); err != nil { return 0, nil, err } } else { - hdr := buffer.NewPrependable(len(payloadBytes) + int(route.MaxHeaderLength())) - if err := route.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: e.TransProto, TTL: route.DefaultTTL(), TOS: stack.DefaultTOS}, &stack.PacketBuffer{ - Header: hdr, - Data: buffer.View(payloadBytes).ToVectorisedView(), - Owner: e.owner, - }); err != nil { + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(route.MaxHeaderLength()), + Data: buffer.View(payloadBytes).ToVectorisedView(), + }) + pkt.Owner = e.owner + if err := route.WritePacket(nil /* gso */, stack.NetworkHeaderParams{ + Protocol: e.TransProto, + TTL: route.DefaultTTL(), + TOS: stack.DefaultTOS, + }, pkt); err != nil { return 0, nil, err } } @@ -691,12 +696,13 @@ func (e *endpoint) HandlePacket(route *stack.Route, pkt *stack.PacketBuffer) { // slice. Save/restore doesn't support overlapping slices and will fail. var combinedVV buffer.VectorisedView if e.TransportEndpointInfo.NetProto == header.IPv4ProtocolNumber { - headers := make(buffer.View, 0, len(pkt.NetworkHeader)+len(pkt.TransportHeader)) - headers = append(headers, pkt.NetworkHeader...) - headers = append(headers, pkt.TransportHeader...) + network, transport := pkt.NetworkHeader().View(), pkt.TransportHeader().View() + headers := make(buffer.View, 0, len(network)+len(transport)) + headers = append(headers, network...) + headers = append(headers, transport...) combinedVV = headers.ToVectorisedView() } else { - combinedVV = append(buffer.View(nil), pkt.TransportHeader...).ToVectorisedView() + combinedVV = append(buffer.View(nil), pkt.TransportHeader().View()...).ToVectorisedView() } combinedVV.Append(pkt.Data) packet.data = combinedVV diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 46702906b..290172ac9 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -746,11 +746,7 @@ func (e *endpoint) sendTCP(r *stack.Route, tf tcpFields, data buffer.VectorisedV func buildTCPHdr(r *stack.Route, tf tcpFields, pkt *stack.PacketBuffer, gso *stack.GSO) { optLen := len(tf.opts) - hdr := &pkt.Header - packetSize := pkt.Data.Size() - // Initialize the header. - tcp := header.TCP(hdr.Prepend(header.TCPMinimumSize + optLen)) - pkt.TransportHeader = buffer.View(tcp) + tcp := header.TCP(pkt.TransportHeader().Push(header.TCPMinimumSize + optLen)) tcp.Encode(&header.TCPFields{ SrcPort: tf.id.LocalPort, DstPort: tf.id.RemotePort, @@ -762,8 +758,7 @@ func buildTCPHdr(r *stack.Route, tf tcpFields, pkt *stack.PacketBuffer, gso *sta }) copy(tcp[header.TCPMinimumSize:], tf.opts) - length := uint16(hdr.UsedLength() + packetSize) - xsum := r.PseudoHeaderChecksum(ProtocolNumber, length) + xsum := r.PseudoHeaderChecksum(ProtocolNumber, uint16(pkt.Size())) // Only calculate the checksum if offloading isn't supported. if gso != nil && gso.NeedsCsum { // This is called CHECKSUM_PARTIAL in the Linux kernel. We @@ -801,17 +796,18 @@ func sendTCPBatch(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso packetSize = size } size -= packetSize - var pkt stack.PacketBuffer - pkt.Header = buffer.NewPrependable(hdrSize) + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: hdrSize, + }) pkt.Hash = tf.txHash pkt.Owner = owner pkt.EgressRoute = r pkt.GSOOptions = gso pkt.NetworkProtocolNumber = r.NetworkProtocolNumber() data.ReadToVV(&pkt.Data, packetSize) - buildTCPHdr(r, tf, &pkt, gso) + buildTCPHdr(r, tf, pkt, gso) tf.seq = tf.seq.Add(seqnum.Size(packetSize)) - pkts.PushBack(&pkt) + pkts.PushBack(pkt) } if tf.ttl == 0 { @@ -837,12 +833,12 @@ func sendTCP(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso *stac return sendTCPBatch(r, tf, data, gso, owner) } - pkt := &stack.PacketBuffer{ - Header: buffer.NewPrependable(header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen), - Data: data, - Hash: tf.txHash, - Owner: owner, - } + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen, + Data: data, + }) + pkt.Hash = tf.txHash + pkt.Owner = owner buildTCPHdr(r, tf, pkt, gso) if tf.ttl == 0 { diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go index 49a673b42..c5afa2680 100644 --- a/pkg/tcpip/transport/tcp/protocol.go +++ b/pkg/tcpip/transport/tcp/protocol.go @@ -21,7 +21,6 @@ package tcp import ( - "fmt" "runtime" "strings" "time" @@ -547,22 +546,22 @@ func (p *protocol) SynRcvdCounter() *synRcvdCounter { // Parse implements stack.TransportProtocol.Parse. func (*protocol) Parse(pkt *stack.PacketBuffer) bool { - hdr, ok := pkt.Data.PullUp(header.TCPMinimumSize) + // TCP header is variable length, peek at it first. + hdrLen := header.TCPMinimumSize + hdr, ok := pkt.Data.PullUp(hdrLen) if !ok { return false } // If the header has options, pull those up as well. if offset := int(header.TCP(hdr).DataOffset()); offset > header.TCPMinimumSize && offset <= pkt.Data.Size() { - hdr, ok = pkt.Data.PullUp(offset) - if !ok { - panic(fmt.Sprintf("There should be at least %d bytes in pkt.Data.", offset)) - } + // TODO(gvisor.dev/issue/2404): Figure out whether to reject this kind of + // packets. + hdrLen = offset } - pkt.TransportHeader = hdr - pkt.Data.TrimFront(len(hdr)) - return true + _, ok = pkt.TransportHeader().Consume(hdrLen) + return ok } // NewProtocol returns a TCP transport protocol. diff --git a/pkg/tcpip/transport/tcp/segment.go b/pkg/tcpip/transport/tcp/segment.go index bb60dc29d..94307d31a 100644 --- a/pkg/tcpip/transport/tcp/segment.go +++ b/pkg/tcpip/transport/tcp/segment.go @@ -68,7 +68,7 @@ func newSegment(r *stack.Route, id stack.TransportEndpointID, pkt *stack.PacketB route: r.Clone(), } s.data = pkt.Data.Clone(s.views[:]) - s.hdr = header.TCP(pkt.TransportHeader) + s.hdr = header.TCP(pkt.TransportHeader().View()) s.rcvdTime = time.Now() return s } diff --git a/pkg/tcpip/transport/tcp/testing/context/context.go b/pkg/tcpip/transport/tcp/testing/context/context.go index 37e7767d6..927bc71e0 100644 --- a/pkg/tcpip/transport/tcp/testing/context/context.go +++ b/pkg/tcpip/transport/tcp/testing/context/context.go @@ -257,8 +257,8 @@ func (c *Context) GetPacket() []byte { c.t.Fatalf("Bad network protocol: got %v, wanted %v", p.Proto, ipv4.ProtocolNumber) } - hdr := p.Pkt.Header.View() - b := append(hdr[:len(hdr):len(hdr)], p.Pkt.Data.ToView()...) + vv := buffer.NewVectorisedView(p.Pkt.Size(), p.Pkt.Views()) + b := vv.ToView() if p.GSO != nil && p.GSO.L3HdrLen != header.IPv4MinimumSize { c.t.Errorf("L3HdrLen %v (expected %v)", p.GSO.L3HdrLen, header.IPv4MinimumSize) @@ -284,8 +284,8 @@ func (c *Context) GetPacketNonBlocking() []byte { c.t.Fatalf("Bad network protocol: got %v, wanted %v", p.Proto, ipv4.ProtocolNumber) } - hdr := p.Pkt.Header.View() - b := append(hdr[:len(hdr):len(hdr)], p.Pkt.Data.ToView()...) + vv := buffer.NewVectorisedView(p.Pkt.Size(), p.Pkt.Views()) + b := vv.ToView() checker.IPv4(c.t, b, checker.SrcAddr(StackAddr), checker.DstAddr(TestAddr)) return b @@ -318,9 +318,10 @@ func (c *Context) SendICMPPacket(typ header.ICMPv4Type, code uint8, p1, p2 []byt copy(icmp[header.ICMPv4PayloadOffset:], p2) // Inject packet. - c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), }) + c.linkEP.InjectInbound(ipv4.ProtocolNumber, pkt) } // BuildSegment builds a TCP segment based on the given Headers and payload. @@ -374,26 +375,29 @@ func (c *Context) BuildSegmentWithAddrs(payload []byte, h *Headers, src, dst tcp // SendSegment sends a TCP segment that has already been built and written to a // buffer.VectorisedView. func (c *Context) SendSegment(s buffer.VectorisedView) { - c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: s, }) + c.linkEP.InjectInbound(ipv4.ProtocolNumber, pkt) } // SendPacket builds and sends a TCP segment(with the provided payload & TCP // headers) in an IPv4 packet via the link layer endpoint. func (c *Context) SendPacket(payload []byte, h *Headers) { - c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: c.BuildSegment(payload, h), }) + c.linkEP.InjectInbound(ipv4.ProtocolNumber, pkt) } // SendPacketWithAddrs builds and sends a TCP segment(with the provided payload // & TCPheaders) in an IPv4 packet via the link layer endpoint using the // provided source and destination IPv4 addresses. func (c *Context) SendPacketWithAddrs(payload []byte, h *Headers, src, dst tcpip.Address) { - c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: c.BuildSegmentWithAddrs(payload, h, src, dst), }) + c.linkEP.InjectInbound(ipv4.ProtocolNumber, pkt) } // SendAck sends an ACK packet. @@ -514,9 +518,8 @@ func (c *Context) GetV6Packet() []byte { if p.Proto != ipv6.ProtocolNumber { c.t.Fatalf("Bad network protocol: got %v, wanted %v", p.Proto, ipv6.ProtocolNumber) } - b := make([]byte, p.Pkt.Header.UsedLength()+p.Pkt.Data.Size()) - copy(b, p.Pkt.Header.View()) - copy(b[p.Pkt.Header.UsedLength():], p.Pkt.Data.ToView()) + vv := buffer.NewVectorisedView(p.Pkt.Size(), p.Pkt.Views()) + b := vv.ToView() checker.IPv6(c.t, b, checker.SrcAddr(StackV6Addr), checker.DstAddr(TestV6Addr)) return b @@ -566,9 +569,10 @@ func (c *Context) SendV6PacketWithAddrs(payload []byte, h *Headers, src, dst tcp t.SetChecksum(^t.CalculateChecksum(xsum)) // Inject packet. - c.linkEP.InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), }) + c.linkEP.InjectInbound(ipv6.ProtocolNumber, pkt) } // CreateConnected creates a connected TCP endpoint. diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 4a2b6c03a..73608783c 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -986,13 +986,16 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { // sendUDP sends a UDP segment via the provided network endpoint and under the // provided identity. func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort uint16, ttl uint8, useDefaultTTL bool, tos uint8, owner tcpip.PacketOwner, noChecksum bool) *tcpip.Error { - // Allocate a buffer for the UDP header. - hdr := buffer.NewPrependable(header.UDPMinimumSize + int(r.MaxHeaderLength())) + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: header.UDPMinimumSize + int(r.MaxHeaderLength()), + Data: data, + }) + pkt.Owner = owner - // Initialize the header. - udp := header.UDP(hdr.Prepend(header.UDPMinimumSize)) + // Initialize the UDP header. + udp := header.UDP(pkt.TransportHeader().Push(header.UDPMinimumSize)) - length := uint16(hdr.UsedLength() + data.Size()) + length := uint16(pkt.Size()) udp.Encode(&header.UDPFields{ SrcPort: localPort, DstPort: remotePort, @@ -1019,12 +1022,7 @@ func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort u Protocol: ProtocolNumber, TTL: ttl, TOS: tos, - }, &stack.PacketBuffer{ - Header: hdr, - Data: data, - TransportHeader: buffer.View(udp), - Owner: owner, - }); err != nil { + }, pkt); err != nil { r.Stats().UDP.PacketSendErrors.Increment() return err } @@ -1372,7 +1370,7 @@ func (e *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { // endpoint. func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt *stack.PacketBuffer) { // Get the header then trim it from the view. - hdr := header.UDP(pkt.TransportHeader) + hdr := header.UDP(pkt.TransportHeader().View()) if int(hdr.Length()) > pkt.Data.Size()+header.UDPMinimumSize { // Malformed packet. e.stack.Stats().UDP.MalformedPacketsReceived.Increment() @@ -1443,9 +1441,9 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk // Save any useful information from the network header to the packet. switch r.NetProto { case header.IPv4ProtocolNumber: - packet.tos, _ = header.IPv4(pkt.NetworkHeader).TOS() + packet.tos, _ = header.IPv4(pkt.NetworkHeader().View()).TOS() case header.IPv6ProtocolNumber: - packet.tos, _ = header.IPv6(pkt.NetworkHeader).TOS() + packet.tos, _ = header.IPv6(pkt.NetworkHeader().View()).TOS() } // TODO(gvisor.dev/issue/3556): r.LocalAddress may be a multicast or broadcast diff --git a/pkg/tcpip/transport/udp/protocol.go b/pkg/tcpip/transport/udp/protocol.go index 0e7464e3a..63d4bed7c 100644 --- a/pkg/tcpip/transport/udp/protocol.go +++ b/pkg/tcpip/transport/udp/protocol.go @@ -82,7 +82,7 @@ func (*protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) { // HandleUnknownDestinationPacket handles packets targeted at this protocol but // that don't match any existing endpoint. func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, pkt *stack.PacketBuffer) bool { - hdr := header.UDP(pkt.TransportHeader) + hdr := header.UDP(pkt.TransportHeader().View()) if int(hdr.Length()) > pkt.Data.Size()+header.UDPMinimumSize { // Malformed packet. r.Stack().Stats().UDP.MalformedPacketsReceived.Increment() @@ -130,7 +130,7 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans } headerLen := int(r.MaxHeaderLength()) + header.ICMPv4MinimumSize available := int(mtu) - headerLen - payloadLen := len(pkt.NetworkHeader) + len(pkt.TransportHeader) + pkt.Data.Size() + payloadLen := pkt.NetworkHeader().View().Size() + pkt.TransportHeader().View().Size() + pkt.Data.Size() if payloadLen > available { payloadLen = available } @@ -139,22 +139,21 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans // For example, a raw or packet socket may use what UDP // considers an unreachable destination. Thus we deep copy pkt // to prevent multiple ownership and SR errors. - newHeader := append(buffer.View(nil), pkt.NetworkHeader...) - newHeader = append(newHeader, pkt.TransportHeader...) + newHeader := append(buffer.View(nil), pkt.NetworkHeader().View()...) + newHeader = append(newHeader, pkt.TransportHeader().View()...) payload := newHeader.ToVectorisedView() payload.AppendView(pkt.Data.ToView()) payload.CapLength(payloadLen) - hdr := buffer.NewPrependable(headerLen) - pkt := header.ICMPv4(hdr.Prepend(header.ICMPv4MinimumSize)) - pkt.SetType(header.ICMPv4DstUnreachable) - pkt.SetCode(header.ICMPv4PortUnreachable) - pkt.SetChecksum(header.ICMPv4Checksum(pkt, payload)) - r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, &stack.PacketBuffer{ - Header: hdr, - TransportHeader: buffer.View(pkt), - Data: payload, + icmpPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: headerLen, + Data: payload, }) + icmpHdr := header.ICMPv4(icmpPkt.TransportHeader().Push(header.ICMPv4MinimumSize)) + icmpHdr.SetType(header.ICMPv4DstUnreachable) + icmpHdr.SetCode(header.ICMPv4PortUnreachable) + icmpHdr.SetChecksum(header.ICMPv4Checksum(icmpHdr, icmpPkt.Data)) + r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, icmpPkt) case header.IPv6AddressSize: if !r.Stack().AllowICMPMessage() { @@ -175,24 +174,24 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans } headerLen := int(r.MaxHeaderLength()) + header.ICMPv6DstUnreachableMinimumSize available := int(mtu) - headerLen - payloadLen := len(pkt.NetworkHeader) + len(pkt.TransportHeader) + pkt.Data.Size() + network, transport := pkt.NetworkHeader().View(), pkt.TransportHeader().View() + payloadLen := len(network) + len(transport) + pkt.Data.Size() if payloadLen > available { payloadLen = available } - payload := buffer.NewVectorisedView(len(pkt.NetworkHeader)+len(pkt.TransportHeader), []buffer.View{pkt.NetworkHeader, pkt.TransportHeader}) + payload := buffer.NewVectorisedView(len(network)+len(transport), []buffer.View{network, transport}) payload.Append(pkt.Data) payload.CapLength(payloadLen) - hdr := buffer.NewPrependable(headerLen) - pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6DstUnreachableMinimumSize)) - pkt.SetType(header.ICMPv6DstUnreachable) - pkt.SetCode(header.ICMPv6PortUnreachable) - pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, payload)) - r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, &stack.PacketBuffer{ - Header: hdr, - TransportHeader: buffer.View(pkt), - Data: payload, + icmpPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: headerLen, + Data: payload, }) + icmpHdr := header.ICMPv6(icmpPkt.TransportHeader().Push(header.ICMPv6DstUnreachableMinimumSize)) + icmpHdr.SetType(header.ICMPv6DstUnreachable) + icmpHdr.SetCode(header.ICMPv6PortUnreachable) + icmpHdr.SetChecksum(header.ICMPv6Checksum(icmpHdr, r.LocalAddress, r.RemoteAddress, icmpPkt.Data)) + r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, icmpPkt) } return true } @@ -215,14 +214,8 @@ func (*protocol) Wait() {} // Parse implements stack.TransportProtocol.Parse. func (*protocol) Parse(pkt *stack.PacketBuffer) bool { - h, ok := pkt.Data.PullUp(header.UDPMinimumSize) - if !ok { - // Packet is too small - return false - } - pkt.TransportHeader = h - pkt.Data.TrimFront(header.UDPMinimumSize) - return true + _, ok := pkt.TransportHeader().Consume(header.UDPMinimumSize) + return ok } // NewProtocol returns a UDP transport protocol. diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index 1a32622ca..71776d6db 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -388,8 +388,8 @@ func (c *testContext) getPacketAndVerify(flow testFlow, checkers ...checker.Netw c.t.Fatalf("Bad network protocol: got %v, wanted %v", p.Proto, flow.netProto()) } - hdr := p.Pkt.Header.View() - b := append(hdr[:len(hdr):len(hdr)], p.Pkt.Data.ToView()...) + vv := buffer.NewVectorisedView(p.Pkt.Size(), p.Pkt.Views()) + b := vv.ToView() h := flow.header4Tuple(outgoing) checkers = append( @@ -410,14 +410,14 @@ func (c *testContext) injectPacket(flow testFlow, payload []byte) { h := flow.header4Tuple(incoming) if flow.isV4() { buf := c.buildV4Packet(payload, &h) - c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ + c.linkEP.InjectInbound(ipv4.ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) } else { buf := c.buildV6Packet(payload, &h) - c.linkEP.InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ + c.linkEP.InjectInbound(ipv6.ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) } } @@ -804,9 +804,9 @@ func TestV4ReadSelfSource(t *testing.T) { h.srcAddr = h.dstAddr buf := c.buildV4Packet(payload, &h) - c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ + c.linkEP.InjectInbound(ipv4.ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) if got := c.s.Stats().IP.InvalidSourceAddressesReceived.Value(); got != tt.wantInvalidSource { t.Errorf("c.s.Stats().IP.InvalidSourceAddressesReceived got %d, want %d", got, tt.wantInvalidSource) @@ -1766,9 +1766,8 @@ func TestV4UnknownDestination(t *testing.T) { return } - var pkt []byte - pkt = append(pkt, p.Pkt.Header.View()...) - pkt = append(pkt, p.Pkt.Data.ToView()...) + vv := buffer.NewVectorisedView(p.Pkt.Size(), p.Pkt.Views()) + pkt := vv.ToView() if got, want := len(pkt), header.IPv4MinimumProcessableDatagramSize; got > want { t.Fatalf("got an ICMP packet of size: %d, want: sz <= %d", got, want) } @@ -1844,9 +1843,8 @@ func TestV6UnknownDestination(t *testing.T) { return } - var pkt []byte - pkt = append(pkt, p.Pkt.Header.View()...) - pkt = append(pkt, p.Pkt.Data.ToView()...) + vv := buffer.NewVectorisedView(p.Pkt.Size(), p.Pkt.Views()) + pkt := vv.ToView() if got, want := len(pkt), header.IPv6MinimumMTU; got > want { t.Fatalf("got an ICMP packet of size: %d, want: sz <= %d", got, want) } @@ -1897,9 +1895,9 @@ func TestIncrementMalformedPacketsReceived(t *testing.T) { u := header.UDP(buf[header.IPv6MinimumSize:]) u.SetLength(u.Length() + 1) - c.linkEP.InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ + c.linkEP.InjectInbound(ipv6.ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) const want = 1 if got := c.s.Stats().UDP.MalformedPacketsReceived.Value(); got != want { @@ -1952,9 +1950,9 @@ func TestShortHeader(t *testing.T) { copy(buf[header.IPv6MinimumSize:], udpHdr) // Inject packet. - c.linkEP.InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ + c.linkEP.InjectInbound(ipv6.ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) if got, want := c.s.Stats().MalformedRcvdPackets.Value(), uint64(1); got != want { t.Errorf("got c.s.Stats().MalformedRcvdPackets.Value() = %d, want = %d", got, want) @@ -1986,9 +1984,9 @@ func TestIncrementChecksumErrorsV4(t *testing.T) { } } - c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ + c.linkEP.InjectInbound(ipv4.ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) const want = 1 if got := c.s.Stats().UDP.ChecksumErrors.Value(); got != want { @@ -2019,9 +2017,9 @@ func TestIncrementChecksumErrorsV6(t *testing.T) { u := header.UDP(buf[header.IPv6MinimumSize:]) u.SetChecksum(u.Checksum() + 1) - c.linkEP.InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ + c.linkEP.InjectInbound(ipv6.ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) const want = 1 if got := c.s.Stats().UDP.ChecksumErrors.Value(); got != want { @@ -2049,9 +2047,9 @@ func TestPayloadModifiedV4(t *testing.T) { buf := c.buildV4Packet(payload, &h) // Modify the payload so that the checksum value in the UDP header will be incorrect. buf[len(buf)-1]++ - c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ + c.linkEP.InjectInbound(ipv4.ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) const want = 1 if got := c.s.Stats().UDP.ChecksumErrors.Value(); got != want { @@ -2079,9 +2077,9 @@ func TestPayloadModifiedV6(t *testing.T) { buf := c.buildV6Packet(payload, &h) // Modify the payload so that the checksum value in the UDP header will be incorrect. buf[len(buf)-1]++ - c.linkEP.InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ + c.linkEP.InjectInbound(ipv6.ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) const want = 1 if got := c.s.Stats().UDP.ChecksumErrors.Value(); got != want { @@ -2110,9 +2108,9 @@ func TestChecksumZeroV4(t *testing.T) { // Set the checksum field in the UDP header to zero. u := header.UDP(buf[header.IPv4MinimumSize:]) u.SetChecksum(0) - c.linkEP.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ + c.linkEP.InjectInbound(ipv4.ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) const want = 0 if got := c.s.Stats().UDP.ChecksumErrors.Value(); got != want { @@ -2141,9 +2139,9 @@ func TestChecksumZeroV6(t *testing.T) { // Set the checksum field in the UDP header to zero. u := header.UDP(buf[header.IPv6MinimumSize:]) u.SetChecksum(0) - c.linkEP.InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ + c.linkEP.InjectInbound(ipv6.ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), - }) + })) const want = 1 if got := c.s.Stats().UDP.ChecksumErrors.Value(); got != want { -- cgit v1.2.3 From d6520e1d0592f99161faedef3eba49439b140917 Mon Sep 17 00:00:00 2001 From: Ayush Ranjan Date: Thu, 13 Aug 2020 19:31:17 -0700 Subject: [vfs2][gofer] Fix file creation flags sent to gofer. Fixes php runtime test ext/standard/tests/file/readfile_basic.phpt Fixes #3516 fsgofers only want the access mode in the OpenFlags passed to Create(). If more flags are supplied (like O_APPEND in this case), read/write from that fd will fail with EBADF. See runsc/fsgofer/fsgofer.go:WriteAt() VFS2 was providing more than just access modes. So filtering the flags using p9.OpenFlagsModeMask == linux.O_ACCMODE fixes the issue. Gofer in VFS1 also only extracts the access mode flags while making the create RPC. See pkg/sentry/fs/gofer/path.go:Create() Even in VFS2, when we open a handle, we extract out only the access mode flags + O_TRUNC. See third_party/gvisor/pkg/sentry/fsimpl/gofer/handle.go:openHandle() Added a test for this. PiperOrigin-RevId: 326574829 --- pkg/sentry/fsimpl/gofer/filesystem.go | 6 ++---- test/syscalls/linux/open.cc | 20 ++++++++++++++++++++ 2 files changed, 22 insertions(+), 4 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 40fec890a..610a7ed78 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -1082,10 +1082,8 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving } creds := rp.Credentials() name := rp.Component() - // Filter file creation flags and O_LARGEFILE out; the create RPC already - // has the semantics of O_CREAT|O_EXCL, while some servers will choke on - // O_LARGEFILE. - createFlags := p9.OpenFlags(opts.Flags &^ (vfs.FileCreationFlags | linux.O_LARGEFILE)) + // We only want the access mode for creating the file. + createFlags := p9.OpenFlags(opts.Flags) & p9.OpenFlagsModeMask fdobj, openFile, createQID, _, err := dirfile.create(ctx, name, createFlags, (p9.FileMode)(opts.Mode), (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)) if err != nil { dirfile.close(ctx) diff --git a/test/syscalls/linux/open.cc b/test/syscalls/linux/open.cc index c7147c20b..8f0c9cb49 100644 --- a/test/syscalls/linux/open.cc +++ b/test/syscalls/linux/open.cc @@ -147,6 +147,26 @@ TEST_F(OpenTest, WriteOnly) { EXPECT_THAT(write(wo_file.get(), &buf, 1), SyscallSucceedsWithValue(1)); } +TEST_F(OpenTest, CreateWithAppend) { + std::string data = "text"; + std::string new_file = NewTempAbsPath(); + const FileDescriptor file = ASSERT_NO_ERRNO_AND_VALUE( + Open(new_file, O_WRONLY | O_APPEND | O_CREAT, 0666)); + EXPECT_THAT(write(file.get(), data.c_str(), data.size()), + SyscallSucceedsWithValue(data.size())); + EXPECT_THAT(lseek(file.get(), 0, SEEK_SET), SyscallSucceeds()); + EXPECT_THAT(write(file.get(), data.c_str(), data.size()), + SyscallSucceedsWithValue(data.size())); + + // Check that the size of the file is correct and that the offset has been + // incremented to that size. + struct stat s0; + EXPECT_THAT(fstat(file.get(), &s0), SyscallSucceeds()); + EXPECT_EQ(s0.st_size, 2 * data.size()); + EXPECT_THAT(lseek(file.get(), 0, SEEK_CUR), + SyscallSucceedsWithValue(2 * data.size())); +} + TEST_F(OpenTest, ReadWrite) { char buf; const FileDescriptor rw_file = -- cgit v1.2.3 From 97263e505324b7d1c58a3daaac7175241921998f Mon Sep 17 00:00:00 2001 From: Ayush Ranjan Date: Mon, 17 Aug 2020 10:03:38 -0700 Subject: [vfs] Return EIO when opening /dev/tty. This is in compliance with VFS1. See pkg/sentry/fs/dev/tty.go in the struct ttyInodeOperations. Fixes the failure of python runtime test_ioctl. Updates #3515 PiperOrigin-RevId: 327042758 --- pkg/sentry/devices/ttydev/BUILD | 2 +- pkg/sentry/devices/ttydev/ttydev.go | 46 +++---------------------------------- 2 files changed, 4 insertions(+), 44 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/devices/ttydev/BUILD b/pkg/sentry/devices/ttydev/BUILD index 12e49b58a..b4b6ca38a 100644 --- a/pkg/sentry/devices/ttydev/BUILD +++ b/pkg/sentry/devices/ttydev/BUILD @@ -11,6 +11,6 @@ go_library( "//pkg/context", "//pkg/sentry/fsimpl/devtmpfs", "//pkg/sentry/vfs", - "//pkg/usermem", + "//pkg/syserror", ], ) diff --git a/pkg/sentry/devices/ttydev/ttydev.go b/pkg/sentry/devices/ttydev/ttydev.go index fd4b79c46..664e54498 100644 --- a/pkg/sentry/devices/ttydev/ttydev.go +++ b/pkg/sentry/devices/ttydev/ttydev.go @@ -12,10 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -// Package ttydev implements devices for /dev/tty and (eventually) -// /dev/console. -// -// TODO(b/159623826): Support /dev/console. +// Package ttydev implements an unopenable vfs.Device for /dev/tty. package ttydev import ( @@ -23,7 +20,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fsimpl/devtmpfs" "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/pkg/syserror" ) const ( @@ -37,44 +34,7 @@ type ttyDevice struct{} // Open implements vfs.Device.Open. func (ttyDevice) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &ttyFD{} - if err := fd.vfsfd.Init(fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{ - UseDentryMetadata: true, - }); err != nil { - return nil, err - } - return &fd.vfsfd, nil -} - -// ttyFD implements vfs.FileDescriptionImpl for /dev/tty. -type ttyFD struct { - vfsfd vfs.FileDescription - vfs.FileDescriptionDefaultImpl - vfs.DentryMetadataFileDescriptionImpl - vfs.NoLockFD -} - -// Release implements vfs.FileDescriptionImpl.Release. -func (fd *ttyFD) Release(context.Context) {} - -// PRead implements vfs.FileDescriptionImpl.PRead. -func (fd *ttyFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - return 0, nil -} - -// Read implements vfs.FileDescriptionImpl.Read. -func (fd *ttyFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - return 0, nil -} - -// PWrite implements vfs.FileDescriptionImpl.PWrite. -func (fd *ttyFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - return src.NumBytes(), nil -} - -// Write implements vfs.FileDescriptionImpl.Write. -func (fd *ttyFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { - return src.NumBytes(), nil + return nil, syserror.EIO } // Register registers all devices implemented by this package in vfsObj. -- cgit v1.2.3 From 3bd066d5032c297e501f5c71be301ffa2fe9ed34 Mon Sep 17 00:00:00 2001 From: Dean Deng Date: Mon, 17 Aug 2020 11:40:08 -0700 Subject: Remove weak references from unix sockets. The abstract socket namespace no longer holds any references on sockets. Instead, TryIncRef() is used when a socket is being retrieved in BoundEndpoint(). Abstract sockets are now responsible for removing themselves from the namespace they are in, when they are destroyed. Updates #1486. PiperOrigin-RevId: 327064173 --- pkg/refs_vfs2/BUILD | 6 +- pkg/refs_vfs2/refs.go | 4 +- pkg/sentry/kernel/BUILD | 1 + pkg/sentry/kernel/abstract_socket_namespace.go | 77 +++++++++++++++++--------- pkg/sentry/socket/unix/BUILD | 14 +++++ pkg/sentry/socket/unix/unix.go | 22 ++++++-- pkg/sentry/socket/unix/unix_vfs2.go | 6 +- 7 files changed, 91 insertions(+), 39 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/refs_vfs2/BUILD b/pkg/refs_vfs2/BUILD index 7f180c7bd..7b3e10683 100644 --- a/pkg/refs_vfs2/BUILD +++ b/pkg/refs_vfs2/BUILD @@ -19,10 +19,8 @@ go_template( ) go_library( - name = "refs", - srcs = [ - "refs.go", - ], + name = "refs_vfs2", + srcs = ["refs.go"], visibility = ["//pkg/sentry:internal"], deps = ["//pkg/context"], ) diff --git a/pkg/refs_vfs2/refs.go b/pkg/refs_vfs2/refs.go index ee01b17b0..99a074e96 100644 --- a/pkg/refs_vfs2/refs.go +++ b/pkg/refs_vfs2/refs.go @@ -12,8 +12,8 @@ // See the License for the specific language governing permissions and // limitations under the License. -// Package refs defines an interface for a reference-counted object. -package refs +// Package refs_vfs2 defines an interface for a reference-counted object. +package refs_vfs2 import ( "gvisor.dev/gvisor/pkg/context" diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD index f6886a758..5416a310d 100644 --- a/pkg/sentry/kernel/BUILD +++ b/pkg/sentry/kernel/BUILD @@ -163,6 +163,7 @@ go_library( "//pkg/log", "//pkg/metric", "//pkg/refs", + "//pkg/refs_vfs2", "//pkg/safemem", "//pkg/secio", "//pkg/sentry/arch", diff --git a/pkg/sentry/kernel/abstract_socket_namespace.go b/pkg/sentry/kernel/abstract_socket_namespace.go index 52ed5cea2..1b9721534 100644 --- a/pkg/sentry/kernel/abstract_socket_namespace.go +++ b/pkg/sentry/kernel/abstract_socket_namespace.go @@ -15,29 +15,21 @@ package kernel import ( + "fmt" "syscall" "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/refs" + "gvisor.dev/gvisor/pkg/refs_vfs2" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sync" ) // +stateify savable type abstractEndpoint struct { - ep transport.BoundEndpoint - wr *refs.WeakRef - name string - ns *AbstractSocketNamespace -} - -// WeakRefGone implements refs.WeakRefUser.WeakRefGone. -func (e *abstractEndpoint) WeakRefGone(context.Context) { - e.ns.mu.Lock() - if e.ns.endpoints[e.name].ep == e.ep { - delete(e.ns.endpoints, e.name) - } - e.ns.mu.Unlock() + ep transport.BoundEndpoint + socket refs_vfs2.RefCounter + name string + ns *AbstractSocketNamespace } // AbstractSocketNamespace is used to implement the Linux abstract socket functionality. @@ -46,7 +38,11 @@ func (e *abstractEndpoint) WeakRefGone(context.Context) { type AbstractSocketNamespace struct { mu sync.Mutex `state:"nosave"` - // Keeps mapping from name to endpoint. + // Keeps a mapping from name to endpoint. AbstractSocketNamespace does not hold + // any references on any sockets that it contains; when retrieving a socket, + // TryIncRef() must be called in case the socket is concurrently being + // destroyed. It is the responsibility of the socket to remove itself from the + // abstract socket namespace when it is destroyed. endpoints map[string]abstractEndpoint } @@ -58,15 +54,15 @@ func NewAbstractSocketNamespace() *AbstractSocketNamespace { } // A boundEndpoint wraps a transport.BoundEndpoint to maintain a reference on -// its backing object. +// its backing socket. type boundEndpoint struct { transport.BoundEndpoint - rc refs.RefCounter + socket refs_vfs2.RefCounter } // Release implements transport.BoundEndpoint.Release. func (e *boundEndpoint) Release(ctx context.Context) { - e.rc.DecRef(ctx) + e.socket.DecRef(ctx) e.BoundEndpoint.Release(ctx) } @@ -81,32 +77,59 @@ func (a *AbstractSocketNamespace) BoundEndpoint(name string) transport.BoundEndp return nil } - rc := ep.wr.Get() - if rc == nil { - delete(a.endpoints, name) + if !ep.socket.TryIncRef() { + // The socket has reached zero references and is being destroyed. return nil } - return &boundEndpoint{ep.ep, rc} + return &boundEndpoint{ep.ep, ep.socket} } // Bind binds the given socket. // -// When the last reference managed by rc is dropped, ep may be removed from the +// When the last reference managed by socket is dropped, ep may be removed from the // namespace. -func (a *AbstractSocketNamespace) Bind(ctx context.Context, name string, ep transport.BoundEndpoint, rc refs.RefCounter) error { +func (a *AbstractSocketNamespace) Bind(ctx context.Context, name string, ep transport.BoundEndpoint, socket refs_vfs2.RefCounter) error { a.mu.Lock() defer a.mu.Unlock() + // Check if there is already a socket (which has not yet been destroyed) bound at name. if ep, ok := a.endpoints[name]; ok { - if rc := ep.wr.Get(); rc != nil { - rc.DecRef(ctx) + if ep.socket.TryIncRef() { + ep.socket.DecRef(ctx) return syscall.EADDRINUSE } } ae := abstractEndpoint{ep: ep, name: name, ns: a} - ae.wr = refs.NewWeakRef(rc, &ae) + ae.socket = socket a.endpoints[name] = ae return nil } + +// Remove removes the specified socket at name from the abstract socket +// namespace, if it has not yet been replaced. +func (a *AbstractSocketNamespace) Remove(name string, socket refs_vfs2.RefCounter) { + a.mu.Lock() + defer a.mu.Unlock() + + ep, ok := a.endpoints[name] + if !ok { + // We never delete a map entry apart from a socket's destructor (although the + // map entry may be overwritten). Therefore, a socket should exist, even if it + // may not be the one we expect. + panic(fmt.Sprintf("expected socket to exist at '%s' in abstract socket namespace", name)) + } + + // A Bind() operation may race with callers of Remove(), e.g. in the + // following case: + // socket1 reaches zero references and begins destruction + // a.Bind("foo", ep, socket2) replaces socket1 with socket2 + // socket1's destructor calls a.Remove("foo", socket1) + // + // Therefore, we need to check that the socket at name is what we expect + // before modifying the map. + if ep.socket == socket { + delete(a.endpoints, name) + } +} diff --git a/pkg/sentry/socket/unix/BUILD b/pkg/sentry/socket/unix/BUILD index 061a689a9..cb953e4dc 100644 --- a/pkg/sentry/socket/unix/BUILD +++ b/pkg/sentry/socket/unix/BUILD @@ -1,12 +1,25 @@ load("//tools:defs.bzl", "go_library") +load("//tools/go_generics:defs.bzl", "go_template_instance") package(licenses = ["notice"]) +go_template_instance( + name = "socket_refs", + out = "socket_refs.go", + package = "unix", + prefix = "socketOpsCommon", + template = "//pkg/refs_vfs2:refs_template", + types = { + "T": "socketOpsCommon", + }, +) + go_library( name = "unix", srcs = [ "device.go", "io.go", + "socket_refs.go", "unix.go", "unix_vfs2.go", ], @@ -15,6 +28,7 @@ go_library( "//pkg/abi/linux", "//pkg/context", "//pkg/fspath", + "//pkg/log", "//pkg/refs", "//pkg/safemem", "//pkg/sentry/arch", diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go index 2b8454edb..b7e8e4325 100644 --- a/pkg/sentry/socket/unix/unix.go +++ b/pkg/sentry/socket/unix/unix.go @@ -24,7 +24,6 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" @@ -80,7 +79,7 @@ func NewWithDirent(ctx context.Context, d *fs.Dirent, ep transport.Endpoint, sty stype: stype, }, } - s.EnableLeakCheck("unix.SocketOperations") + s.EnableLeakCheck() return fs.NewFile(ctx, d, flags, &s) } @@ -89,17 +88,26 @@ func NewWithDirent(ctx context.Context, d *fs.Dirent, ep transport.Endpoint, sty // // +stateify savable type socketOpsCommon struct { - refs.AtomicRefCount + socketOpsCommonRefs socket.SendReceiveTimeout ep transport.Endpoint stype linux.SockType + + // abstractName and abstractNamespace indicate the name and namespace of the + // socket if it is bound to an abstract socket namespace. Once the socket is + // bound, they cannot be modified. + abstractName string + abstractNamespace *kernel.AbstractSocketNamespace } // DecRef implements RefCounter.DecRef. func (s *socketOpsCommon) DecRef(ctx context.Context) { - s.DecRefWithDestructor(ctx, func(context.Context) { + s.socketOpsCommonRefs.DecRef(func() { s.ep.Close(ctx) + if s.abstractNamespace != nil { + s.abstractNamespace.Remove(s.abstractName, s) + } }) } @@ -284,10 +292,14 @@ func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { if t.IsNetworkNamespaced() { return syserr.ErrInvalidEndpointState } - if err := t.AbstractSockets().Bind(t, p[1:], bep, s); err != nil { + asn := t.AbstractSockets() + name := p[1:] + if err := asn.Bind(t, name, bep, s); err != nil { // syserr.ErrPortInUse corresponds to EADDRINUSE. return syserr.ErrPortInUse } + s.abstractName = name + s.abstractNamespace = asn } else { // The parent and name. var d *fs.Dirent diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go index dfa25241a..d066ef8ab 100644 --- a/pkg/sentry/socket/unix/unix_vfs2.go +++ b/pkg/sentry/socket/unix/unix_vfs2.go @@ -183,10 +183,14 @@ func (s *SocketVFS2) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { if t.IsNetworkNamespaced() { return syserr.ErrInvalidEndpointState } - if err := t.AbstractSockets().Bind(t, p[1:], bep, s); err != nil { + asn := t.AbstractSockets() + name := p[1:] + if err := asn.Bind(t, name, bep, s); err != nil { // syserr.ErrPortInUse corresponds to EADDRINUSE. return syserr.ErrPortInUse } + s.abstractName = name + s.abstractNamespace = asn } else { path := fspath.Parse(p) root := t.FSContext().RootDirectoryVFS2() -- cgit v1.2.3 From e3e1b36896750e9c4d07ecb6eaf1a738b08cd876 Mon Sep 17 00:00:00 2001 From: Ayush Ranjan Date: Mon, 17 Aug 2020 13:24:09 -0700 Subject: [vfs] Do O_DIRECTORY check after resolving symlinks. Fixes python runtime test test_glob. Updates #3515 We were checking is the to-be-opened dentry is a dir or not before resolving symlinks. We should check that after resolving symlinks. This was preventing us from opening a symlink which pointed to a directory with O_DIRECTORY. Also added this check in tmpfs and removed a duplicate check. PiperOrigin-RevId: 327085895 --- pkg/sentry/fsimpl/gofer/filesystem.go | 6 +++--- pkg/sentry/fsimpl/tmpfs/filesystem.go | 5 ++--- test/syscalls/linux/open.cc | 8 ++++++++ 3 files changed, 13 insertions(+), 6 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 610a7ed78..a3903db33 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -886,9 +886,6 @@ afterTrailingSymlink: if mustCreate { return nil, syserror.EEXIST } - if !child.isDir() && rp.MustBeDir() { - return nil, syserror.ENOTDIR - } // Open existing child or follow symlink. if child.isSymlink() && rp.ShouldFollowSymlink() { target, err := child.readlink(ctx, rp.Mount()) @@ -901,6 +898,9 @@ afterTrailingSymlink: start = parent goto afterTrailingSymlink } + if rp.MustBeDir() && !child.isDir() { + return nil, syserror.ENOTDIR + } return child.openLocked(ctx, rp, &opts) } diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index a4864df53..cb8b2d944 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -389,9 +389,8 @@ afterTrailingSymlink: start = &parentDir.dentry goto afterTrailingSymlink } - // Open existing file. - if mustCreate { - return nil, syserror.EEXIST + if rp.MustBeDir() && !child.inode.isDir() { + return nil, syserror.ENOTDIR } return child.open(ctx, rp, &opts, false) } diff --git a/test/syscalls/linux/open.cc b/test/syscalls/linux/open.cc index 8f0c9cb49..77f390f3c 100644 --- a/test/syscalls/linux/open.cc +++ b/test/syscalls/linux/open.cc @@ -27,6 +27,7 @@ #include "test/util/cleanup.h" #include "test/util/file_descriptor.h" #include "test/util/fs_util.h" +#include "test/util/posix_error.h" #include "test/util/temp_path.h" #include "test/util/test_util.h" #include "test/util/thread_util.h" @@ -408,6 +409,13 @@ TEST_F(OpenTest, FileNotDirectory) { SyscallFailsWithErrno(ENOTDIR)); } +TEST_F(OpenTest, SymlinkDirectory) { + auto dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + std::string link = NewTempAbsPath(); + ASSERT_THAT(symlink(dir.path().c_str(), link.c_str()), SyscallSucceeds()); + ASSERT_NO_ERRNO(Open(link, O_RDONLY | O_DIRECTORY)); +} + TEST_F(OpenTest, Null) { char c = '\0'; ASSERT_THAT(open(&c, O_RDONLY), SyscallFailsWithErrno(ENOENT)); -- cgit v1.2.3 From d42d01b976107b4a5db9c552e485e5be5b21a995 Mon Sep 17 00:00:00 2001 From: Nicolas Lacasse Date: Mon, 17 Aug 2020 16:28:19 -0700 Subject: Stop masking the IO error in handleIOError. PiperOrigin-RevId: 327123331 --- pkg/sentry/syscalls/linux/error.go | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) (limited to 'pkg/sentry') diff --git a/pkg/sentry/syscalls/linux/error.go b/pkg/sentry/syscalls/linux/error.go index 64de56ac5..46060f6f5 100644 --- a/pkg/sentry/syscalls/linux/error.go +++ b/pkg/sentry/syscalls/linux/error.go @@ -36,8 +36,8 @@ var ( // errors, we may consume the error and return only the partial read/write. // // op and f are used only for panics. -func HandleIOErrorVFS2(t *kernel.Task, partialResult bool, err, intr error, op string, f *vfs.FileDescription) error { - known, err := handleIOErrorImpl(t, partialResult, err, intr, op) +func HandleIOErrorVFS2(t *kernel.Task, partialResult bool, ioerr, intr error, op string, f *vfs.FileDescription) error { + known, err := handleIOErrorImpl(t, partialResult, ioerr, intr, op) if err != nil { return err } @@ -46,7 +46,7 @@ func HandleIOErrorVFS2(t *kernel.Task, partialResult bool, err, intr error, op s fs := f.Mount().Filesystem().VirtualFilesystem() root := vfs.RootFromContext(t) name, _ := fs.PathnameWithDeleted(t, root, f.VirtualDentry()) - log.Traceback("Invalid request partialResult %v and err (type %T) %v for %s operation on %q", partialResult, err, err, op, name) + log.Traceback("Invalid request partialResult %v and err (type %T) %v for %s operation on %q", partialResult, ioerr, ioerr, op, name) partialResultOnce.Do(partialResultMetric.Increment) } return nil @@ -56,15 +56,15 @@ func HandleIOErrorVFS2(t *kernel.Task, partialResult bool, err, intr error, op s // errors, we may consume the error and return only the partial read/write. // // op and f are used only for panics. -func handleIOError(t *kernel.Task, partialResult bool, err, intr error, op string, f *fs.File) error { - known, err := handleIOErrorImpl(t, partialResult, err, intr, op) +func handleIOError(t *kernel.Task, partialResult bool, ioerr, intr error, op string, f *fs.File) error { + known, err := handleIOErrorImpl(t, partialResult, ioerr, intr, op) if err != nil { return err } if !known { // An unknown error is encountered with a partial read/write. name, _ := f.Dirent.FullName(nil /* ignore chroot */) - log.Traceback("Invalid request partialResult %v and err (type %T) %v for %s operation on %q, %T", partialResult, err, err, op, name, f.FileOperations) + log.Traceback("Invalid request partialResult %v and err (type %T) %v for %s operation on %q, %T", partialResult, ioerr, ioerr, op, name, f.FileOperations) partialResultOnce.Do(partialResultMetric.Increment) } return nil -- cgit v1.2.3 From 6d0c5803d557d453f15ac6f683697eeb46dab680 Mon Sep 17 00:00:00 2001 From: gVisor bot Date: Mon, 17 Aug 2020 16:29:10 -0700 Subject: Add a skeleton for verity file system PiperOrigin-RevId: 327123477 --- pkg/sentry/fsimpl/verity/BUILD | 23 +++ pkg/sentry/fsimpl/verity/filesystem.go | 333 +++++++++++++++++++++++++++++++ pkg/sentry/fsimpl/verity/verity.go | 355 +++++++++++++++++++++++++++++++++ 3 files changed, 711 insertions(+) create mode 100644 pkg/sentry/fsimpl/verity/BUILD create mode 100644 pkg/sentry/fsimpl/verity/filesystem.go create mode 100644 pkg/sentry/fsimpl/verity/verity.go (limited to 'pkg/sentry') diff --git a/pkg/sentry/fsimpl/verity/BUILD b/pkg/sentry/fsimpl/verity/BUILD new file mode 100644 index 000000000..28d2a4bcb --- /dev/null +++ b/pkg/sentry/fsimpl/verity/BUILD @@ -0,0 +1,23 @@ +load("//tools:defs.bzl", "go_library") + +licenses(["notice"]) + +go_library( + name = "verity", + srcs = [ + "filesystem.go", + "verity.go", + ], + visibility = ["//pkg/sentry:internal"], + deps = [ + "//pkg/abi/linux", + "//pkg/context", + "//pkg/fspath", + "//pkg/sentry/fs/lock", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/socket/unix/transport", + "//pkg/sentry/vfs", + "//pkg/sync", + "//pkg/syserror", + ], +) diff --git a/pkg/sentry/fsimpl/verity/filesystem.go b/pkg/sentry/fsimpl/verity/filesystem.go new file mode 100644 index 000000000..78c6074bd --- /dev/null +++ b/pkg/sentry/fsimpl/verity/filesystem.go @@ -0,0 +1,333 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package verity + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/syserror" +) + +// Sync implements vfs.FilesystemImpl.Sync. +func (fs *filesystem) Sync(ctx context.Context) error { + // All files should be read-only. + return nil +} + +var dentrySlicePool = sync.Pool{ + New: func() interface{} { + ds := make([]*dentry, 0, 4) // arbitrary non-zero initial capacity + return &ds + }, +} + +func appendDentry(ds *[]*dentry, d *dentry) *[]*dentry { + if ds == nil { + ds = dentrySlicePool.Get().(*[]*dentry) + } + *ds = append(*ds, d) + return ds +} + +// Preconditions: ds != nil. +func putDentrySlice(ds *[]*dentry) { + // Allow dentries to be GC'd. + for i := range *ds { + (*ds)[i] = nil + } + *ds = (*ds)[:0] + dentrySlicePool.Put(ds) +} + +// renameMuRUnlockAndCheckDrop calls fs.renameMu.RUnlock(), then calls +// dentry.checkDropLocked on all dentries in *ds with fs.renameMu locked for +// writing. +// +// ds is a pointer-to-pointer since defer evaluates its arguments immediately, +// but dentry slices are allocated lazily, and it's much easier to say "defer +// fs.renameMuRUnlockAndCheckDrop(&ds)" than "defer func() { +// fs.renameMuRUnlockAndCheckDrop(ds) }()" to work around this. +func (fs *filesystem) renameMuRUnlockAndCheckDrop(ctx context.Context, ds **[]*dentry) { + fs.renameMu.RUnlock() + if *ds == nil { + return + } + if len(**ds) != 0 { + fs.renameMu.Lock() + for _, d := range **ds { + d.checkDropLocked(ctx) + } + fs.renameMu.Unlock() + } + putDentrySlice(*ds) +} + +func (fs *filesystem) renameMuUnlockAndCheckDrop(ctx context.Context, ds **[]*dentry) { + if *ds == nil { + fs.renameMu.Unlock() + return + } + for _, d := range **ds { + d.checkDropLocked(ctx) + } + fs.renameMu.Unlock() + putDentrySlice(*ds) +} + +// resolveLocked resolves rp to an existing file. +func (fs *filesystem) resolveLocked(ctx context.Context, rp *vfs.ResolvingPath, ds **[]*dentry) (*dentry, error) { + // TODO(b/159261227): Implement resolveLocked. + return nil, nil +} + +// walkParentDirLocked resolves all but the last path component of rp to an +// existing directory, starting from the given directory (which is usually +// rp.Start().Impl().(*dentry)). It does not check that the returned directory +// is searchable by the provider of rp. +// +// Preconditions: fs.renameMu must be locked. !rp.Done(). +func (fs *filesystem) walkParentDirLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry, ds **[]*dentry) (*dentry, error) { + // TODO(b/159261227): Implement walkParentDirLocked. + return nil, nil +} + +// AccessAt implements vfs.Filesystem.Impl.AccessAt. +func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error { + // Verity file system is read-only. + if ats&vfs.MayWrite != 0 { + return syserror.EROFS + } + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return err + } + return d.checkPermissions(creds, ats) +} + +// GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. +func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return nil, err + } + if opts.CheckSearchable { + if !d.isDir() { + return nil, syserror.ENOTDIR + } + if err := d.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { + return nil, err + } + } + d.IncRef() + return &d.vfsd, nil +} + +// GetParentDentryAt implements vfs.FilesystemImpl.GetParentDentryAt. +func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) + start := rp.Start().Impl().(*dentry) + d, err := fs.walkParentDirLocked(ctx, rp, start, &ds) + if err != nil { + return nil, err + } + d.IncRef() + return &d.vfsd, nil +} + +// LinkAt implements vfs.FilesystemImpl.LinkAt. +func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry) error { + // Verity file system is read-only. + return syserror.EROFS +} + +// MkdirAt implements vfs.FilesystemImpl.MkdirAt. +func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error { + // Verity file system is read-only. + return syserror.EROFS +} + +// MknodAt implements vfs.FilesystemImpl.MknodAt. +func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MknodOptions) error { + // Verity file system is read-only. + return syserror.EROFS +} + +// OpenAt implements vfs.FilesystemImpl.OpenAt. +func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + //TODO(b/159261227): Implement OpenAt. + return nil, nil +} + +// ReadlinkAt implements vfs.FilesystemImpl.ReadlinkAt. +func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return "", err + } + //TODO(b/162787271): Provide integrity check for ReadlinkAt. + return fs.vfsfs.VirtualFilesystem().ReadlinkAt(ctx, d.fs.creds, &vfs.PathOperation{ + Root: d.lowerVD, + Start: d.lowerVD, + }) +} + +// RenameAt implements vfs.FilesystemImpl.RenameAt. +func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldParentVD vfs.VirtualDentry, oldName string, opts vfs.RenameOptions) error { + // Verity file system is read-only. + return syserror.EROFS +} + +// RmdirAt implements vfs.FilesystemImpl.RmdirAt. +func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error { + // Verity file system is read-only. + return syserror.EROFS +} + +// SetStatAt implements vfs.FilesystemImpl.SetStatAt. +func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { + // Verity file system is read-only. + return syserror.EROFS +} + +// StatAt implements vfs.FilesystemImpl.StatAt. +func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return linux.Statx{}, err + } + + var stat linux.Statx + stat, err = fs.vfsfs.VirtualFilesystem().StatAt(ctx, fs.creds, &vfs.PathOperation{ + Root: d.lowerVD, + Start: d.lowerVD, + }, &opts) + if err != nil { + return linux.Statx{}, err + } + return stat, nil +} + +// StatFSAt implements vfs.FilesystemImpl.StatFSAt. +func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { + // TODO(b/159261227): Implement StatFSAt. + return linux.Statfs{}, nil +} + +// SymlinkAt implements vfs.FilesystemImpl.SymlinkAt. +func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, target string) error { + // Verity file system is read-only. + return syserror.EROFS +} + +// UnlinkAt implements vfs.FilesystemImpl.UnlinkAt. +func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error { + // Verity file system is read-only. + return syserror.EROFS +} + +// BoundEndpointAt implements FilesystemImpl.BoundEndpointAt. +func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.BoundEndpointOptions) (transport.BoundEndpoint, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) + if _, err := fs.resolveLocked(ctx, rp, &ds); err != nil { + return nil, err + } + return nil, syserror.ECONNREFUSED +} + +// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. +func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, size uint64) ([]string, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return nil, err + } + lowerVD := d.lowerVD + return fs.vfsfs.VirtualFilesystem().ListxattrAt(ctx, d.fs.creds, &vfs.PathOperation{ + Root: lowerVD, + Start: lowerVD, + }, size) +} + +// GetxattrAt implements vfs.FilesystemImpl.GetxattrAt. +func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetxattrOptions) (string, error) { + var ds *[]*dentry + fs.renameMu.RLock() + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) + d, err := fs.resolveLocked(ctx, rp, &ds) + if err != nil { + return "", err + } + lowerVD := d.lowerVD + return fs.vfsfs.VirtualFilesystem().GetxattrAt(ctx, d.fs.creds, &vfs.PathOperation{ + Root: lowerVD, + Start: lowerVD, + }, &opts) +} + +// SetxattrAt implements vfs.FilesystemImpl.SetxattrAt. +func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { + // Verity file system is read-only. + return syserror.EROFS +} + +// RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. +func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { + // Verity file system is read-only. + return syserror.EROFS +} + +// PrependPath implements vfs.FilesystemImpl.PrependPath. +func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { + fs.renameMu.RLock() + defer fs.renameMu.RUnlock() + mnt := vd.Mount() + d := vd.Dentry().Impl().(*dentry) + for { + if mnt == vfsroot.Mount() && &d.vfsd == vfsroot.Dentry() { + return vfs.PrependPathAtVFSRootError{} + } + if &d.vfsd == mnt.Root() { + return nil + } + if d.parent == nil { + return vfs.PrependPathAtNonMountRootError{} + } + b.PrependComponent(d.name) + d = d.parent + } +} diff --git a/pkg/sentry/fsimpl/verity/verity.go b/pkg/sentry/fsimpl/verity/verity.go new file mode 100644 index 000000000..cb29d33a5 --- /dev/null +++ b/pkg/sentry/fsimpl/verity/verity.go @@ -0,0 +1,355 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package verity provides a filesystem implementation that is a wrapper of +// another file system. +// The verity file system provides integrity check for the underlying file +// system by providing verification for path traversals and each read. +// The verity file system is read-only, except for one case: when +// allowRuntimeEnable is true, additional Merkle files can be generated using +// the FS_IOC_ENABLE_VERITY ioctl. +package verity + +import ( + "sync/atomic" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/syserror" +) + +// Name is the default filesystem name. +const Name = "verity" + +// testOnlyDebugging allows verity file system to return error instead of +// crashing the application when a malicious action is detected. This should +// only be set for tests. +var testOnlyDebugging bool + +// FilesystemType implements vfs.FilesystemType. +type FilesystemType struct{} + +// filesystem implements vfs.FilesystemImpl. +type filesystem struct { + vfsfs vfs.Filesystem + + // creds is a copy of the filesystem's creator's credentials, which are + // used for accesses to the underlying file system. creds is immutable. + creds *auth.Credentials + + // allowRuntimeEnable is true if using ioctl with FS_IOC_ENABLE_VERITY + // to build Merkle trees in the verity file system is allowed. If this + // is false, no new Merkle trees can be built, and only the files that + // had Merkle trees before startup (e.g. from a host filesystem mounted + // with gofer fs) can be verified. + allowRuntimeEnable bool + + // lowerMount is the underlying file system mount. + lowerMount *vfs.Mount + + // rootDentry is the mount root Dentry for this file system, which + // stores the root hash of the whole file system in bytes. + rootDentry *dentry + + // renameMu synchronizes renaming with non-renaming operations in order + // to ensure consistent lock ordering between dentry.dirMu in different + // dentries. + renameMu sync.RWMutex +} + +// InternalFilesystemOptions may be passed as +// vfs.GetFilesystemOptions.InternalData to FilesystemType.GetFilesystem. +type InternalFilesystemOptions struct { + // RootMerkleFileName is the name of the verity root Merkle tree file. + RootMerkleFileName string + + // LowerName is the name of the filesystem wrapped by verity fs. + LowerName string + + // RootHash is the root hash of the overall verity file system. + RootHash []byte + + // AllowRuntimeEnable specifies whether the verity file system allows + // enabling verification for files (i.e. building Merkle trees) during + // runtime. + AllowRuntimeEnable bool + + // LowerGetFSOptions is the file system option for the lower layer file + // system wrapped by verity file system. + LowerGetFSOptions vfs.GetFilesystemOptions + + // TestOnlyDebugging allows verity file system to return error instead + // of crashing the application when a malicious action is detected. This + // should only be set for tests. + TestOnlyDebugging bool +} + +// Name implements vfs.FilesystemType.Name. +func (FilesystemType) Name() string { + return Name +} + +// GetFilesystem implements vfs.FilesystemType.GetFilesystem. +func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { + //TODO(b/159261227): Implement GetFilesystem. + return nil, nil, nil +} + +// Release implements vfs.FilesystemImpl.Release. +func (fs *filesystem) Release(ctx context.Context) { + fs.lowerMount.DecRef(ctx) +} + +// dentry implements vfs.DentryImpl. +type dentry struct { + vfsd vfs.Dentry + + refs int64 + + // fs is the owning filesystem. fs is immutable. + fs *filesystem + + // mode, uid and gid are the file mode, owner, and group of the file in + // the underlying file system. + mode uint32 + uid uint32 + gid uint32 + + // parent is the dentry corresponding to this dentry's parent directory. + // name is this dentry's name in parent. If this dentry is a filesystem + // root, parent is nil and name is the empty string. parent and name are + // protected by fs.renameMu. + parent *dentry + name string + + // If this dentry represents a directory, children maps the names of + // children for which dentries have been instantiated to those dentries, + // and dirents (if not nil) is a cache of dirents as returned by + // directoryFDs representing this directory. children is protected by + // dirMu. + dirMu sync.Mutex + children map[string]*dentry + + // lowerVD is the VirtualDentry in the underlying file system. + lowerVD vfs.VirtualDentry + + // lowerMerkleVD is the VirtualDentry of the corresponding Merkle tree + // in the underlying file system. + lowerMerkleVD vfs.VirtualDentry + + // rootHash is the rootHash for the current file or directory. + rootHash []byte +} + +// newDentry creates a new dentry representing the given verity file. The +// dentry initially has no references; it is the caller's responsibility to set +// the dentry's reference count and/or call dentry.destroy() as appropriate. +// The dentry is initially invalid in that it contains no underlying dentry; +// the caller is responsible for setting them. +func (fs *filesystem) newDentry() *dentry { + d := &dentry{ + fs: fs, + } + d.vfsd.Init(d) + return d +} + +// IncRef implements vfs.DentryImpl.IncRef. +func (d *dentry) IncRef() { + atomic.AddInt64(&d.refs, 1) +} + +// TryIncRef implements vfs.DentryImpl.TryIncRef. +func (d *dentry) TryIncRef() bool { + for { + refs := atomic.LoadInt64(&d.refs) + if refs <= 0 { + return false + } + if atomic.CompareAndSwapInt64(&d.refs, refs, refs+1) { + return true + } + } +} + +// DecRef implements vfs.DentryImpl.DecRef. +func (d *dentry) DecRef(ctx context.Context) { + if refs := atomic.AddInt64(&d.refs, -1); refs == 0 { + d.fs.renameMu.Lock() + d.checkDropLocked(ctx) + d.fs.renameMu.Unlock() + } else if refs < 0 { + panic("verity.dentry.DecRef() called without holding a reference") + } +} + +// checkDropLocked should be called after d's reference count becomes 0 or it +// becomes deleted. +func (d *dentry) checkDropLocked(ctx context.Context) { + // Dentries with a positive reference count must be retained. Dentries + // with a negative reference count have already been destroyed. + if atomic.LoadInt64(&d.refs) != 0 { + return + } + // Refs is still zero; destroy it. + d.destroyLocked(ctx) + return +} + +// destroyLocked destroys the dentry. +// +// Preconditions: d.fs.renameMu must be locked for writing. d.refs == 0. +func (d *dentry) destroyLocked(ctx context.Context) { + switch atomic.LoadInt64(&d.refs) { + case 0: + // Mark the dentry destroyed. + atomic.StoreInt64(&d.refs, -1) + case -1: + panic("verity.dentry.destroyLocked() called on already destroyed dentry") + default: + panic("verity.dentry.destroyLocked() called with references on the dentry") + } + + if d.lowerVD.Ok() { + d.lowerVD.DecRef(ctx) + } + + if d.lowerMerkleVD.Ok() { + d.lowerMerkleVD.DecRef(ctx) + } + + if d.parent != nil { + d.parent.dirMu.Lock() + if !d.vfsd.IsDead() { + delete(d.parent.children, d.name) + } + d.parent.dirMu.Unlock() + if refs := atomic.AddInt64(&d.parent.refs, -1); refs == 0 { + d.parent.checkDropLocked(ctx) + } else if refs < 0 { + panic("verity.dentry.DecRef() called without holding a reference") + } + } +} + +// InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. +func (d *dentry) InotifyWithParent(ctx context.Context, events, cookie uint32, et vfs.EventType) { + //TODO(b/159261227): Implement InotifyWithParent. +} + +// Watches implements vfs.DentryImpl.Watches. +func (d *dentry) Watches() *vfs.Watches { + //TODO(b/159261227): Implement Watches. + return nil +} + +// OnZeroWatches implements vfs.DentryImpl.OnZeroWatches. +func (d *dentry) OnZeroWatches(context.Context) { + //TODO(b/159261227): Implement OnZeroWatches. +} + +func (d *dentry) isSymlink() bool { + return atomic.LoadUint32(&d.mode)&linux.S_IFMT == linux.S_IFLNK +} + +func (d *dentry) isDir() bool { + return atomic.LoadUint32(&d.mode)&linux.S_IFMT == linux.S_IFDIR +} + +func (d *dentry) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes) error { + return vfs.GenericCheckPermissions(creds, ats, linux.FileMode(atomic.LoadUint32(&d.mode)), auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))) +} + +func (d *dentry) readlink(ctx context.Context) (string, error) { + return d.fs.vfsfs.VirtualFilesystem().ReadlinkAt(ctx, d.fs.creds, &vfs.PathOperation{ + Root: d.lowerVD, + Start: d.lowerVD, + }) +} + +// FileDescription implements vfs.FileDescriptionImpl for verity fds. +// FileDescription is a wrapper of the underlying lowerFD, with support to build +// Merkle trees through the Linux fs-verity API to verify contents read from +// lowerFD. +type fileDescription struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl + vfs.LockFD + + // d is the corresponding dentry to the fileDescription. + d *dentry + + // isDir specifies whehter the fileDescription points to a directory. + isDir bool + + // lowerFD is the FileDescription corresponding to the file in the + // underlying file system. + lowerFD *vfs.FileDescription + + // merkleReader is the read-only FileDescription corresponding to the + // Merkle tree file in the underlying file system. + merkleReader *vfs.FileDescription + + // merkleWriter is the FileDescription corresponding to the Merkle tree + // file in the underlying file system for writing. This should only be + // used when allowRuntimeEnable is set to true. + merkleWriter *vfs.FileDescription + + // parentMerkleWriter is the FileDescription of the Merkle tree for the + // directory that contains the current file/directory. This is only used + // if allowRuntimeEnable is set to true. + parentMerkleWriter *vfs.FileDescription +} + +// Release implements vfs.FileDescriptionImpl.Release. +func (fd *fileDescription) Release(ctx context.Context) { + fd.lowerFD.DecRef(ctx) + fd.merkleReader.DecRef(ctx) + if fd.merkleWriter != nil { + fd.merkleWriter.DecRef(ctx) + } + if fd.parentMerkleWriter != nil { + fd.parentMerkleWriter.DecRef(ctx) + } +} + +// Stat implements vfs.FileDescriptionImpl.Stat. +func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { + // TODO(b/162788573): Add integrity check for metadata. + stat, err := fd.lowerFD.Stat(ctx, opts) + if err != nil { + return linux.Statx{}, err + } + return stat, nil +} + +// SetStat implements vfs.FileDescriptionImpl.SetStat. +func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { + // Verity files are read-only. + return syserror.EPERM +} + +// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. +func (fd *fileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return fd.Locks().LockPOSIX(ctx, &fd.vfsfd, uid, t, start, length, whence, block) +} + +// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. +func (fd *fileDescription) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return fd.Locks().UnlockPOSIX(ctx, &fd.vfsfd, uid, start, length, whence) +} -- cgit v1.2.3